Algebra

by

Stephen S. Shatz∗ and Jean Gallier∗∗

∗
Department of Mathematics
∗∗
Department of Computer and Information Science
University of Pennsylvania
Philadelphia, PA 19104, USA

©Stephen S. Shatz & Jean Gallier

August 5, 2023
ii
 iii

To the memories of Peter A. Cassileth, Stanley M.K. Chung, and Ralph C. Marcove,
physicians and friends all. Being mortal and denied the gift of life, he gives and they gave
the next best thing: Time.

To Anne, Mia, Philippe and Sylvie

iv
Contents

Preface vii

For the Student ix

Problems 1

1 Group Theory 57
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
1.2 Group Actions and First Applications; The Three Sylow Theorems . . . . . . . . . . . . . . . 57
1.3 Elementary Theory of p-Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
1.4 Group Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
1.5 Solvable and Nilpotent Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
1.6 Ω-Groups and the Jordan-Hölder-Schreier Theorem . . . . . . . . . . . . . . . . . . . . . . . . 92
1.7 Categories, Functors and Free Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
1.8 Further Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

2 Rings and Modules 109

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
2.2 Polynomial Rings, Commutative and Noncommutative . . . . . . . . . . . . . . . . . . . . . . 109
2.3 Operations on Modules; Finiteness Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
2.4 Projective and Injective Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
2.5 The Five Lemma and the Snake Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
2.6 Tensor Products and Flat Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
2.7 Limit Processes in Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
2.8 Flat Modules (Again) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
2.9 Further Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172

3 Commutative Rings 173

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
3.2 Classical Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
3.3 Prime and Maximal Ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
3.4 First Applications of Fraction Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
3.5 Integral Dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
3.6 Primary Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
3.7 Theorems of Krull and Artin–Rees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
3.8 Further Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

v
vi CONTENTS

4 Fields and Galois Theory 243

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
4.2 Algebraic Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
4.3 Separable Extensions, Kähler Differentials, Mac Lane’s Criterion . . . . . . . . . . . . . . . . 248
4.4 The Extension Lemma and Splitting Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
4.5 The Theorems of Dedekind and Artin; Galois Groups . . . . . . . . . . . . . . . . . . . . . . 260
4.6 Primitive Elements, Normal Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
4.7 Galois Cohomology, Norms and Traces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
4.8 Krull’s Galois Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
4.9 Kummer Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
4.10 An Amazing Theorem of Galois Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
4.11 Algebraic Closures; Steinitz’s Theory of Fields . . . . . . . . . . . . . . . . . . . . . . . . . . 298
4.12 Further Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308

5 Homological Algebra 309

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
5.2 Complexes, Resolutions, Derived Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
5.3 Various (Co)homological Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
5.4 Spectral Sequences; First Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
5.5 The Koszul Complex and Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
5.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
5.7 Supplementary Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397

Index 401
Preface

A book on “Abstract” or “Modern” Algebra is a commonplace thing in today’s mathematical milieu. Even a
book for well-prepared, serious beginning graduate students who intend to become research mathematicians
is not so strange any longer. But, the genesis of this book, which is intended for serious, well-prepared
graduate students, is somewhat strange.
To begin with, it is a reworking of notes for a year long graduate course I gave several years ago–not
in itself a strange thing. But, I possess no such notes nor did I ever make any and I never lecture with a
written aide memoir of any sort. Rather, my method is to work out fully during lecture (at the board) each
proof and example. Students will thus see what are the “inner workings” of the subject. Of course, this is
pedagogically to their advantage and, furthermore, it slows me down.
Then where did the notes (to be reworked) come from? They were provided by my friend and colleague
Jean H. Gallier (of the Computer Science Department at Penn). Determined to augment his mathematical
knowledge, he began several years ago to audit some of my graduate courses. “Audit” for him means
faithfully attending lectures, doing all the problem assignments, participating in each bi-weekly problem
session (where he takes his turn presenting problems), writing excellent notes from my oral presentation and
rendering these notes in LATEX form.1 That this book will appear is, in large measure, his doing. While I
have been responsible for its writing, he has on occasion introduced results and/or alternate proofs that have
rendered some material more perspicacious from a student’s point of view–these have improved the text. He
is in every sense a joint author, save that errors are solely my responsibility. There is no way I can thank
him adequately here in plain words and I won’t try except to say, Je te remercie vivement, mon ami Jean,
pour tout ton travail .
Others should be thanked as well–in particular the members of the class that attended the course from
which the book is formed.2 By their interest and attention to detail, they kept me on my toes. One
particular member of that class deserves special mention: Mathew Cross.3 Mathew started the index and
set the original 115 problems in LATEX. He lightened our burden by a considerable amount.
The content of the book follows rather closely the oral lectures–with just a few exceptions. These are: In
Chapter 3, the section on Integral Dependence is now augmented by proofs of all results, the original lectures
had statements only of some of these (due to exigencies of time) and Gallier insisted on a full treatment.
In Chapter 4, the sections on Norms and Traces as well as Kummer Theory and Transcendental Extensions
are likewise augmented by full proofs. In Chapter 5, there is now more to the section on (co)homological
functors and there are full proofs in the last section on the Koszul Complex. Otherwise, the material is just
(a smoothed out version of) what was presented. One will have to move fast to present it to students in one
year, at least I did.
But the heart of the book is the Problem section. Here, I’ve attempted to simulate at the beginning
graduate level some of the features of real mathematical work. There is a jumbling of the problems vis a
1 One must realize he maintains a full research and teaching schedule, directs Ph.D. students, attends to administrative duties

and has a family life in addition to this “auditing”!

2 The members of the class were: A. Bak, D. Boyarchenko, S. Brooks, M. Campbell, S. Corry, M. Cross, C. Daenzer, C.

Devena, J. Gallier, S. Guerra, C. Hoelscher, T. Jaeger, J. Long, S. Mason, T. Zhu.

3 Mathew spells his name with but one “t”; there is no misprint.

vii
viii Preface

vis subject matter just as in real research one never knows what kind of mathematics will be needed in the
solution of a problem. There is no hint of the level of difficulty of a problem (save for the few problems
where suggestions are offered), and anyway the notion of difficulty is ill-defined. And, the problems refer to
each other, just as in real work one is constantly reminded of past efforts (successful or not). In effect, as
suggested in the preface for students, one should begin with the problems and use the text as a means to
fill in knowledge as required to do them (as well as to do other problems assigned by an instructor in this
course or another course).
This brings me to the text material itself. There is no attempt to be encyclopedic. After all, the material
is a faithful copy of what was actually covered in a year and any competent instructor can add material
that has been omitted. I regret not covering the Wederburn-Artin Theory of DCC rings, the Brauer Group,
and some basic material on group representations. What is covered, however, is to my mind central to
the education of any prospective mathematician who aspires to contribute to what is now the mainstream
of mathematical endeavor. Also, while there are over 150 problems filling some 55 pages of text (some of
the problems are rather long being multi-part), other problems of an instructor’s choosing can certainly be
assigned. As to the attribution of the origins of these problems, I have assigned names when they are known
to me. If no name is assigned, the problem comes from some source in my past (perhaps one of my own
teachers or their teachers) and in no way do I claim it as my own. Good problems from all sources are the
treasure hoard of practicing mathematicians in their role as passers on of our common heritage.
I refer to the special symbols (DX) and the “curves ahead” road sign (appearing at odd places in the
text) in the student preface; no repeat of the explanations I offer there is necessary. If you as instructor are
lucky enough to have a class as interested and tough to satisfy as I did, you are lucky indeed and need no
further assurance that mathematics will be in good hands in the future. I intend this book to be of service
to such individuals as they begin their long climb to mathematical independence and maturity.

Tolda Santa Cotogna

Summer, 2006
For the Student

It may be surprising but the most important part of the book you now hold before you is the very first
section–the one labeled “Problems.” To learn mathematics one must do mathematics. Indeed, the best
way to read this book is to turn immediately to the problem section and begin to do the problems. Of
course, you will soon reach some unknown terminology or not have enough knowledge to meet the technical
demands of a problem and this is where you turn to the text to fill in gaps, see ideas explained and techniques
demonstrated. Then you plunge once more back into the problems and repeat the whole process.

The book is designed for serious, well-prepared students who plan on becoming research mathematicians.
It presumes you have had previous acquaintance with algebra; in particular you have met the concepts of
group, ring, field, vector space, homomorphism, isomorphism, and the elementary theorems about these
things. No book on mathematics can be simply read, rather you must recreate the text yourself line by line
checking at each stage all details including those omitted. This is slow work and, as you know, mathematics
has very high density on the page.

In the text, you will find two special symbols: (DX) and a sign such as one sees on the road warning
of dangerous curves ahead. The symbol (DX) stands for “diagnostic exercise”, it means some elementary
details have been omitted and that supplying them should be easy. However, if supplying them is not easy,
then you should go back a page or two as something fundamental has skipped you by. In this way, the sign
(DX) is like a medical test: failing it is sure to tell you if something is wrong (no false positives), however,
if you pass it (supply the details), something still might be wrong. Just read on and anything wrong will
surface later. As for the dangerous curves sign, it precedes counter-examples to naively made conjectures, it
warns when things could go wrong if hypotheses are omitted, and generally forces you to slow down in the
reading and recreating.

If you use this book in a course or even for self study, I recommend that you tackle the problems in a
small group (two to four persons, total). This is because no person has a monopoly on ideas, a good idea
or half-idea can germ in any head, and the working out of a problem by a committed group is akin to the
actual way much research mathematics is accomplished. In your group, you want constant give and take,
and there must be time to think alone so that a real contribution to the group’s effort can be made.

The problems are all jumbled up by area and there is no signal given as to a problem’s difficulty (exceptions
are the few cases where hints or suggestions are given). In real mathematical life, no signs are given that a
question being attacked involves a certain small area of mathematical knowledge or is hard or easy; any such
sign is gleaned by virtue of experience and that is what you are obtaining by doing mathematics in these
problems. Moreover, hard and easy are in the eyes of the beholder; they are not universal characteristics
of a problem. About all one can say is that if a large number of people find a problem difficult, we may
classify it so. However, we shouldn’t be surprised when an individual solves it and claims that, “it was not
that hard”. In any case, guard against confusing mathematical talent either with overall intelligence or with
mathematical speed. Some quick people are in fact talented, many are just quick. Don’t be discouraged if
you find yourself slower than another, the things that really count in doing mathematics (assuming talent)
are persistence and courage.

ix
x For the Student

I can think of no better lines to close with than these which come from B. Pasternak’s poem entitled
“Night”4

“And maybe in an attic

And under ancient slates
A man sits wakeful working
He thinks and broods and waits.”

“He looks upon the planet,

As if the heavenly spheres
Were part of his entrusted
Nocturnal private cares.”

“Fight off your sleep: be wakeful,

Work on, keep up your pace,
Keep vigil like the pilot,
Like all the stars in space.”

“Work on, work on, creator–

To sleep would be a crime–
Eternity’s own hostage,
And prisoner of Time.”

Tolda Santa Cotogna

Summer, 2006

4 From the collection entitled “When It Clears Up”, 1956. Translated by Lydia Pasternak Slater (the poet’s sister).
Problems

“... den Samen in den Wind streuend; fasse, wer es fassen kann”.
—Hermann Weyl

Problem 1

1. Suppose G is a finite group and that AutGr (G) = {1}. (Here, AutGr (G) is the group of all bijections,
G → G, which are also group homomorphisms.) Find all such groups G.
Q Q
2. Write Z/2Z
for the cyclic group of order 2. If G = Z/2Z ··· Z/2Z, t-times, compute
# AutGr (G) . When t = 2, determine the group AutGr (G). When t = 3, determine the struc-
ture of the odd prime Sylows. Can you decide whether AutGr (G) has any normal subgroups in the
case t = 3?

Problem 2

1. (Poincaré). In an infinite group, prove that the intersection of two subgroups of finite index has finite
index itself.

2. Show that if a group, G, has a subgroup of finite index, then it possesses a normal subgroup of finite
index. Hence, an infinite simple group has no subgroups of finite index.

3. Sharpen (2) by proving: if (G : H) = r, then G possesses a normal subgroup, N , with (G : N ) ≤ r!.

Conclude immediately that a group of order 36 cannot be simple.

Problem 3 Let G = GL(n, C) and ∆n be the subgroup of matrices with entries only along the diagonal.
Describe precisely NG (∆n ) in terms of what the matrices look like.

Problem 4 Say G is a group and #(G) = pr g0 , where p is a prime and (p, g0 ) = 1. Assume

0 −1 X
gX
r> [j/pk ]
j=1 k>0

[x] = largest integer ≤ x . Prove that G is not simple. Show that this governs all groups of order < 60,
except for #(G) = 30, 40, 56. We know that #(G) = 30 =⇒ G not simple. Show by explicit argument that
groups of orders 40, 56 are not simple. (Here, of course, by simple we mean non-abelian and simple.)

Problem 5 In a p-group, G, we cannot have

(G : Z(G)) = p.
Q
Show that for non-abelian groups of order p3 , Z(G) ∼
= Z/pZ and G/Z(G) ∼
= Z/pZ Z/pZ.

1
2 PROBLEMS

Problem 6 Let G be the group of automorphisms of a regular polyhedron with v vertices, e edges, and f
faces. Show that G has order g = f s = vr = 2e, where s is the number of sides to a face and r is the number
of edges emanating from a vertex. From topology, one knows Euler’s formula

v − e + f = 2.

Find the only possible values for v, e, f, r, s, g. Make a table.

Problem 7 Let p be a prime number. Find all non-abelian groups of order p3 . Get started with the
Burnside basis theorem, but be careful to check that the groups on your list are non-isomorphic. Also make
sure your list is exhaustive. Your list should be a description of the generators of your groups and the
relations they satisfy.

Problem 8 Let G be a finite group and write c(G) for the number of distinct conjugacy classes in G. This
number will increase (in general) as #(G) → ∞; so, look at
c(G)
c(G) = .
#(G)
The number c(G) measures the “average number of conjugacy classes per element of G” and is 1 if G is
abelian. Assume G is non-abelian from now on. Then 0 < c(G) < 1.
1. Prove that for all such G, we have c(G) ≤ 5/8.
2. Suppose p is the smallest prime with p |#(G). Prove that
1 1 1
c(G) ≤ + − 3.
p p2 p

Is the bound of (1) sharp; that is, does there exist a G with c(G) = 5/8? How about the bound of (2)?

Problem 9 If G is a finite group and H a normal subgroup of G, write P for a p-Sylow subgroup of H.
1. Show that the natural injection
NG (P )/NH (P ) → G/H
(why does it exist, why injective?) is actually an isomorphism.
2. Prove that the Frattini subgroup, Φ(G), of ANY finite group, G, has property N (cf. Section 1.3,
Chapter 1).

Problem 10 We’ve remarked that Φ(G) is a kind of “radical” in the group-theoretic setting. In this problem
we study various types of radicals.
A normal subgroup, H, of G is called small iff for every X C G, the equality H · X = G implies that X = G.
(Note: {1} is small, Φ(G) is small; so they exist.) Check that if H and L are small, so is HL, and if H is
small and K C G, then K ⊆ H =⇒ K is small.
1. The small radical of G, denoted J ∗∗ (G), is

J ∗∗ (G) = x ∈ G Gp{Cl(x)} is small .

(Here, Cl(x) is the conjugacy class of x in G, and Gp{S} is the group generated by S.) Prove that
J ∗∗ (G) is a subgroup of G.
2. The Jacobson radical of G, denoted J ∗ (G), is the intersection of all maximal, normal subgroups of G;
while the Baer radical of G, denoted J (G), is the product (inside G) of all the small subgroups of G.
Prove
J ∗∗ (G) ⊆ J (G) ⊆ J ∗ (G).
PROBLEMS 3

3. Prove Baer’s Theorem: J ∗∗ (G) = J (G) = J ∗ (G). (Suggestion: if x 6∈ J ∗∗ (G), find N C G (6= G) so
that Gp{Cl(x)}N = G. Now construct an appropriate maximal normal subgroup not containing x.)

Problem 11 Recall that a characteristic subgroup is one taken into itself by all automorphisms of the
group.

1. Prove that a group possessing no proper characteristic subgroups is isomorphic to a product of iso-
morphic simple groups. (Hints: Choose G e of smallest possible order (> 1) normal in G. Consider all
∼ Q Q
subgroups, H, for which H = G1 ··· Gt , where each Gj C G and each Gj ∼ = G.e Pick t so that
#(H) is maximal. Prove that H is characteristic. Show K C G1 (say) =⇒ K C G.)

2. Prove: In every finite group, G, a minimal normal subgroup, H, is either an elementary abelian p-group
or is isomorphic to a product of mutually isomorphic, non-abelian, simple groups.

3. Show that in a solvable group, G, only the first case in (2) occurs.


Problem 12 Let G be a finite p-group and suppose ϕ ∈ Aut(G) has order n (i.e., ϕ ϕ(· · · (ϕ(x)) · · · ) = Id,
all x ∈ G: we do ϕ n-times in succession and n is minimal). Suppose (n, p) = 1. Now ϕ induces an

it ϕ, as Φ(G) is characteristic. Remember that G/Φ(G) is a vector space
automorphism of G/Φ(G), call
over Fp ; so, ϕ ∈ GL G/Φ(G) .

1. Prove ϕ = identity ⇐⇒ ϕ = identity.

2. Show that if d is the Burnside dimension of G, then

d

d(d−1) Y
# GL(G/Φ(G)) = p 2 (pk − 1),
k=1



and that if P is a p-Sylow subgroup of GL G/Φ(G) , then P ⊆ SL G/Φ(G) ; i.e., σ ∈ P =⇒
det(σ) = 1.

3. Let P = {ϕ ∈ Aut(G) | ϕ ∈ P, no restriction on the order of ϕ}. Show that P is a p-subgroup of

Aut(G).


4. Call an element σ ∈ GL G/Φ(G) liftable iff it is ϕ for some ϕ ∈ Aut(G). Examine all G of order
p, p2 , p3 to help answer the following: Is every σ liftable? If not, how can you tell (given σ) if σ is
liftable?

Problem 13 Let p be a prime number and consider a set, S, of p objects: S = {α1 , . . . , αp }. Assume G
is a transitive group of permutations of S (i.e., the elements of S form an orbit under G); further assume
(α1 α2 ) ∈ G (here (α1 α2 ) is the transposition). Prove: G = Sp . (Suggestion: let M = {αj |(α1 αj ) ∈ G},
show if σ ∈ Sp and σ = 1 outside M then σ ∈ G. Now prove #(M )| p.)

Problem 14 A Fermat prime, p, is a prime number of the form 2α + 1. E.g., 2, 3, 5, 17, 257, . . ..

1. Show if 2α + 1 is prime then α = 2β .

2. Say p is a Fermat prime (they are quite big) and Q g0 is an odd number with g0 < p. Prove that any
group of order g0 p is isomorphic to a product G0 (Z/pZ), where #(G0 ) = g0 . Hence, for example,
the groups of orders 51(= 3 · 17), 85(= 5 · 17), 119(= 7 · 17), 153(= 9 · 17), 187(= 11 · 17), 221(= 13 · 17),
255(= 3 · 5 · 17) are all abelian. Most we knew already, but 153 = 32 · 17 and 255 = 3 · 5 · 17 are new.

3. Generalize to any prime, p, and g0 < p, with p 6≡ 1 mod g0 . For example, find all groups of order 130.

Problem 15 Recall that a group, G, is finitely generated (f.g.) iff (∃ σ1 , . . . , σn ∈ G)(G = Gp{σ1 , . . . , σn }).
4 PROBLEMS

1. If G is an abelian f.g. group, prove each of its subgroups is f.g.

2. In an arbitrary group, G, an element σ ∈ G is called n-torsion (n ∈ N) ⇐⇒ σ n = 1; σ is torsion iff

it is n-torsion for some n ∈ N. The element σ ∈ G is torsion free ⇐⇒ it is not torsion. Show that in
an abelian group, the set
t(G) = {σ ∈ G | σ is torsion}
is a subgroup and that G/t(G) is torsion free (i.e., all its non-identity elements are torsion free).

3. In the solvable group 0 → Z → G → Z/2Z → 0 (split extension, non-trivial action) find two elements
e possessing elements
x, y satisfying: x2 = y 2 = 1 and xy is torsion free. Can you construct a group, G,
x, y of order 2, so that xy has order n, where n is predetermined in N? Can you construct G e solvable
with these properties?

4. Back to the abelian case. If G is abelian and finitely generated show that t(G) is a finite group.

5. Say G is abelian, f.g., and torsion-free. Write d for the minimal number of generators of G. Prove that
G is isomorphic to a product of d copies of Z.

6. If G is abelian and f.g., prove that

Y

G∼
= t(G) G/t(G) .

Problem 16 Let (P) be a property of groups. We say a group, G, is locally (P) ⇐⇒ each f.g. subgroup
of G has (P). Usually, one says a locally cyclic group is a rank one group.

1. Prove that a rank one group is abelian.

2. Show that the additive group of rational numbers, Q+ , is a rank one group.

3. Show that every torsion-free, rank one group is isomorphic to a subgroup of Q+ .

Problem 17 Fix a group, G, and consider the set, Mn (G), of n × n matrices with entries from G or
so that αij = 0 (i.e., entries are 0 or from G). Assume for each row and each column there is one and
only one non-zero entry. These matrices form a group under ordinary “matrix multiplication” if we define
0 · group element = group element · 0 = 0. Establish an isomorphism of this group with the wreath product
Gn o Sn . As an application, for the subgroup of GL(n, C) consisting of diagonal matrices, call it ∆n , show
that
NG (∆n ) ∼
= Cn o Sn , here G = GL(n, C).

Problem 18

1. Say G is a simple group of order n and say p is a prime number dividing n. If σ1 , . . . , σt is a listing of
the elements of G of exact order p, prove that G = Gp{σ1 , . . . , σt }.

2. Suppose G is any finite group of order n and that d is a positive integer relatively prime to n. Show
that every element of G is a dth power.

Problem 19 We know that when G is a (finite) cyclic group, and A is any G-module, we have an isomor-
phism
∼
AG /N (A) −→ H 2 (G, A).
This problem is designed to lead to a proof. There are other proofs which you might dig out of books (after
some effort), but do this proof.
PROBLEMS 5

1. Suppose
Q G is any group and A, B, C are G-modules. Suppose further, we are given a G-pairing of
A B → C i.e., a map Y
θ:A B→C
which is bi-additive and “G-linear”:
σθ(a, b) = θ(σa, σb).
If f , g are r-, s-cochains of G with values in A, B (respectively), we can define an (r + s)-cochain of
G with values in C via the formula:


(f `θ g)(σ1 , . . . , σr , σr+1 , . . . , σr+s ) = θ f (σ1 , . . . , σr ), σ1 . . . σr g(σr+1 , . . . , σr+s ) .
Prove that δ(f `θ g) = δf `θ g + (−1)r f `θ δg. Show how you conclude from this that we have a
pairing of abelian groups Y
`θ : H r (G, A) H s (G, B) → H r+s (G, C).
(Notation and nomenclature: α `θ β, cup-product.)
2. Again G is any group, this time finite. Let Z and Q/Z be G-modules with trivial action. Consider
e where addition in G
the abelian group Homgr (G, Q/Z) = G, e is by pointwise operation on functions. If
e
χ ∈ G, then χ(σ) ∈ Q/Z, all σ ∈ G. Show that the function
fχ (σ, τ ) = δχ(σ, τ ) = σχ(τ ) − χ(στ ) + χ(σ)
has values in Z and actually is a 2-cocycle with values in Z. (This is an example of the principle: If it
looks like a coboundary, it is certainly a cocycle.) The map
e 7→ cohomology class of fχ (σ, τ )
χ∈G (†)

gives a homomorphism G e → H 2 (G, Z).

Now any 2-cocycle g(σ, τ ) with values in Z can be regarded as a 2-cocycle with values in Q (corre-
sponding to the injection Z ,→ Q). Show that as a 2-cocycle in Q it is a coboundary (of some h(σ),
values in Q). So, g(σ, τ ) = δh(σ, τ ), some h. Use this construction to prove:
e with H 2 (G, Z).
For any finite group, G, the map (†) above gives an isomorphism of G

Q let G be finite, A be any G-module, and Z have the trivial G-action. We have an obvious G-pairing
3. Now
Z A → A, namely (n, a) 7→ na, hence by (1) and (2) we obtain a pairing
Y
e H 2 (G, Z))
G(= AG → H 2 (G, A).

Show that if ξ = N α, for α ∈ A, then (χ, ξ) goes to 0 in H 2 (G, A); hence, we obtain a pairing:
Y
e (AG /N A) → H 2 (G, A).
G
P
(Hint: If f (σ, τ ) is a 2-cocycle of G in A, consider the 1-cochain uf (τ ) = σ∈G f (σ, τ ). Using the
cocycle condition and suitable choices
P of the variables, show the values of uf are in AG and that uf is
related to N f , i.e., N f (τ, ρ) = σ σf (τ, ρ) can be expressed by uf .)
e
4. Finally, when G is cyclic, we pick a generator σ0 . There exists a distinguished element, χ0 , of G
corresponding to σ0 , namely χ0 is that homomorphism G → Q/Z whose value at σ0 is n mod Z, where
1

n = #(G). Show that the map

AG /N A → H 2 (G, A)
via
α 7→ (χ0 , α) 7→ δχ0 ` α ∈ H 2 (G, A)
is the required isomorphism. For surjectivity, I suggest you consider the construction of uf in part (3)
above.
6 PROBLEMS

Problem 20 Let G = SL(2, Z) be the group of all 2 × 2 integral matrices of determinant 1; pick a prime,
p, and write U for the set of 2 × 2 integral matrices having determinant p. G acts on U via u(∈ U ) 7→ σu,
where σ ∈ G.

1. Show that the orbit space has p + 1 elements: 0, 1, . . . , p − 1, ∞, where j corresponds to the matrix
 
1 j
wj =
0 p
 
p 0
and ∞ corresponds to the matrix w∞ = .
0 1

2. If τ ∈ G and r ∈ S = {0, 1, . . . , p − 1, ∞} = G\U , show there exists a unique r0 ∈ S with wr τ −1 in

the orbit of wr0 . Write τ · r = r0 and prove this gives an action of G on S. Hence, we have a group
homomorphism P : G → Aut(S) = Sp+1 .

3. If N = ker P , prove that G/N is isomorphic to the group PSL(2, Fp ) consisting of all “fractional linear
transformations”
ax + b
x 7→ x0 = , a, b, c, d ∈ Fp , ad − bc = 1.
cx + d
Show further that
(

p(p + 1)(p − 1)
i. # PSL(2, Fp ) = if p 6= 2
2
6 if p = 2
and

ii. PSL(2, Fp ) acts transitively on S under the action of (2).

4. Now prove: PSL(2, Fp ) is simple if p ≥ 5. (Note: PSL(2, F3 ) is A4 , PSL(2, F5 ) is A5 , but PSL(2, Fp ) is

not An if p ≥ 7. So, you now have a second infinite collection of simple finite groups—these are finite
group analogs of the Lie groups PSL(2, C)).

Problem 21 We write PSL(2, Z) for the group SL(2, Z)/(±I).

(1) Let ξ be a chosen generator for Z/3Z and η the generator of Z/2Z. Map Z/3Z and Z/2Z to PSL(2, Z)
via  
0 −1
ϕ(ξ) = x = (mod ± I)
1 1
and  
0 −1
ψ(η) = y = (mod ± I)
1 0
Then we obtain a map
ϕ q ψ : Z/3Z q Z/2Z −→ PSL(2, Z)
(here, the coproduct is in the category Grp). What is the image of ϕ q ψ? What is the kernel?
(2) If
   
1 1 1 0
a= and b = in PSL(2, Z)
0 1 1 1
express x and y above (in SL(2, Z)) in terms of a and b and show that SL(2, Z) = Grp{a, b}. Can you express
a and b in terms of x and y?
PROBLEMS 7

(3) For any odd prime number, p, the element

 p−1

1
σ(p) = 2
0 1

is equal to a(p−1)/2 . For any σ ∈ SL(2, Z), we define the weight of σ with respect to a and b by

wt(σ) = inf(length of all words in a, b, a−1 , b−1 , which words equal σ)

By deep theorems of Selberg, Margulis and others (in geometry and analysis) one knows that

wt(σ(p)) = O(log p) as p → ∞.

(Our expression for σ(p) as a power of a shows that we have a word of size O(p) for σ(p), yet no explicit word
of size O(log p) is known as of now (Fall, 2005) and the role of b in this is very mysterious.) Now the Cayley
graph of a group, G, generated by the elements g1 , . . . , gt is that graph whose vertices are the elements of G
and whose edges emanating from a vertex τ ∈ G are the ones connecting τ and τ g1 , . . . , τ gt . Show that the
diameter of the Cayley graph of the group SL(2, Z/pZ) with respect to the generators a and b is O(log p).

Problem 22 Let G be a finite group in this problem.

1. Classify all group extensions
0 → Q → G → G → 0. (E)
Your answer should be in terms of the collection of all subgroups of G, say H, with (G : H) ≤ 2, plus,
perhaps, other data.
2. Same question as (1) for group extensions

0 → Z → G → G → 0, (E)

same kind of answer.

Q
3. Write V for the “four-group” Z/2Z Z/2Z. There are two actions of Z/2Z on V : Flip the factors,
take each element to its inverse. Are these the only actions? Find all group extensions

0 → V → G → Z/pZ → 0. (E)

The group G is a group of order 8; compare your results with what you know from Problems 1–6.
4. Say H is any other group, G need no longer be finite and A, B are abelian groups. Suppose ϕ : H → G
is a homomorphism and we are given a group extension

0 → A → G → G → 0. (E)

Show that, in a canonical way, we can make a group extension

0 → A → Ge → H → 0. (ϕ∗ E)

(Note: your answer has to be in terms of G, H, G and any homomorphisms between them as these are
the only “variables” present. You’ll get the idea if you view an extension as a fibre space.)

Now say ψ : A → B is a group homomorphism and we are given an extension

0 → A → G → G → 0. (E)

Construct, in a canonical way, an extension

0 → B → Ge → G → 0. (ψ∗ E)
8 PROBLEMS

5. Explain, carefully, the relevance of these two constructions to parts (1) and (2) of this problem.

Problem 23 Say A is any abelian group, and write G for the wreath product An o Sn . Show:
1. [G, G] 6= G


2. G : [G, G] = ∞ ⇐⇒ A is infinite
3. If n ≥ 2, then [G, G] 6= {1}.
4. Give a restriction on n which prevents G from being solvable.
a
Problem 24 If {Gα }α∈Λ is a family of abelian groups, write Gα for
α

a  Y 
Gα = (ξα ) ∈ Gα | for all but finitely many α, we have ξα = 0 .
α α
`
Then α Gα is the coproduct of the Gα in Ab. Write as well

(Q/Z)p = {ξ ∈ Q/Z | pr ξ = 0, some r > 0};

here, p is a prime. Further, call an abelian group, A, divisible iff

n
(∀ n)(A −→ A → 0 is exact).

Prove: Theorem Every divisible (abelian) group is a coproduct of copies of Q and (Q/Z)p for various primes
p. The group is torsion iff no copies of Q appear, it is torsion-free iff no copies of (Q/Z)p appear (any p).
Every torsion-free, divisible, abelian group is naturally a vector space over Q.

Problem 25
1. If G is a group of order n, show that G o Aut(G) is isomorphic to a subgroup of Sn .
2. Consider the cycle (1, 2, . . . , n) ∈ Sn ; let H be the subgroup (of Sn ) generated by the cycle. Prove
that
NSn (H) ∼ = (Z/nZ) o Aut(Z/nZ).

Problem 26 Let TOP denote the category of topological spaces.

1. Show that TOP possesses finite fibred products and finite fibred coproducts.
2. Is (1) true without the word “finite”?
3. Write T2TOP for the full subcategory of TOP consisting of Hausdorff topological spaces. Are (1) and
(2) true in T2TOP? If you decide the answer is “no”, give reasonable conditions under which a positive
result holds. What relation is there between the product (coproduct) you constructed in (1) (or (2))
and the corresponding objects in this part of the problem?

Problem 27 Let R be a ring (not necessarily commutative) and write Mod(R) for the category of (left)
R-modules; i.e., the action of R on a module, M , is on the left. We know Mod(R) has finite products and
finite fibred products.
1. What is the situation for infinite products and infinite fibred products?
2. What is the situation for coproducts (finite or infinite) and for fibred coproducts (both finite and
infinite)?
PROBLEMS 9

Problem 28 As usual, write Gr for the category of groups. Say G and G0 are groups and ϕ : G → G0 is
a homomorphism. Then (G, ϕ) ∈ GrG0 , the comma category of “groups

over G0 ”. The group {1} possess a
0
canonical morphism to G , namely the inclusion,
Q i. Thus, {1}, i ∈ GrG0 , as well. We form their product in
GrG0 , i.e., we form the fibred product G {1}. Prove that there exists a canonical monomorphism
G0
Y
G {1} → G.
G0

Identify its image in G.

0
Now consider the “dual” situation: G0 maps to G, so G ∈ GrG (via ϕ) the “groups co-over G0 ”. We also have
0 0
the canonical map G → {1}, killing all the elements of G ; so, as above, we can form the fibred coproduct
G0
of G and {1}: G q {1}. Prove that there exists a canonical epimorphism

G0
G → G q {1},

identify its kernel in G.

Problem 29 Write CR for the category of commutative rings with unity and RNG for the category of rings
with unity.

1. Consider the following two functors from CR to Sets:

(a) |Mpq | : A underlying set of p × q matrices with entries from A

(b) |GLn | : A underlying set of all invertible n × n matrices with entries from A.

Prove the these two functors are representable.

2. A slight modification of (b) above yields a functor from CR to Gr: namely,

GLn : A group of all invertible n × n matrices with entries from A.

When n = 1, we can extend this to a functor from RNG to Gr. That is we get the functor

Gm : A group of all invertible elements of A.

Prove that the functor Gm has a left adjoint, let’s temporarily call it (†); that is: There is a functor
(†) from Gr to RNG, so that

(∀ G ∈ Gr)(∀ R ∈ RNG)(HomRNG ((†)(G), R) ∼

= HomGr (G, Gm (R))),

via a functorial isomorphism.

3. Show that without knowing what ring (†)(G) is, namely that
it exists and that (†) is left adjoint to Gm ,
we can prove: the category of (†)(G)-modules, Mod (†)(G) , is equivalent—in fact isomorphic—to the
category of G-modules.

4. There is a functor from Gr to Ab, namely send G to Gab = G/[G, G]. Show this functor has a right
adjoint, call it I. Namely, there exists a functor I : Ab → Gr, so that

(∀ G ∈ Gr)(∀ H ∈ Ab)(HomGr (G, I(H)) ∼

= HomAb (Gab , H)).

Does G Gab have a left adjoint?

10 PROBLEMS

Problem 30 (Kaplansky) If A and B are 2 × 2-matrices with entries in Z, we embed A and B into the 4 × 4
matrices as follows:
 
aug 0 I
A =
A 0
 
aug 0 I
B = .
B 0

Is it true that if Aaug and B aug are similar over Z, then A and B are similar over Z? Proof or counter-
example. What about the case where the entries lie in Q?

Problem 31 We fix a commutative ring with unity, A, and write M for Mpq (A), the p × q matrices with
entries in A. Choose a q × p matrix, Γ, and make M a ring via:
Addition: as usual among p × q matrices
Multiplication: if R, S ∈ M, set R ∗ S = RΓS, where RΓS is the ordinary product of matrices.
Write M(Γ) for M with these operations, then M(Γ) is an A-algebra (a ring which is an A-module).
1. Suppose that A is a field. Prove that the isomorphism classes of M(Γ)’s are finite in number (here
p and q are fixed while Γ varies); in fact, are in natural one-to-one correspondence with the integers
0, 1, 2, . . . , B where B is to be determined by you.
2. Given two q × p matrices Γ and Γ e we call them equivalent iff Γ
e = W ΓZ, where W ∈ GL(q, A) and
Z ∈ GL(p, A). Prove: each Γ is equivalent to a matrix
 
Ir 0
0 H

where Ir = r × r identity matrix and the entries of H are non-units of A. Is r uniquely determined by
Γ? How about the matrix H?
3. Call the commutative ring, A, a local ring provided it possesses exactly one maximal ideal, mA . For
example, any field is a local ring; the ring Z/pn Z is local if p is a prime; other examples of this large,
important class of rings will appear below. We have the descending chain of ideals

A ⊇ mA ⊇ m2A ⊇ · · · .
\
For some local rings one knows that mtA = (0); let’s call such local rings “good local rings” for
t≥0
temporary nomenclature. If A is a good local ring, we can define a function on A to Z ∪ {∞}, call it
ord, as follows:
ord(ξ) = 0 if ξ 6∈ mA
n+1
ord(ξ) = n if ξ ∈ mnA but ξ 6∈ mA
ord(0) = ∞.
The following properties are simple to prove:
ord(ξ ± η) ≥ min{ord(ξ), ord(η)}
ord(ξη) ≥ ord(ξ) + ord(η).
Consider the q × p matrices under equivalence and look at the following three conditions:
 
Ir 0
(i) Γ is equivalent to , with H = (0)
0 H
PROBLEMS 11
 
Ir 0
(ii) Γ is equivalent to with H having non-unit entries and r ≥ 1
0 H
(iii) (∃ Q ∈ M)(ΓQΓ = Γ).

Of course, i. =⇒ ii. if Γ 6= (0), A any ring. Prove: if A is any (commutative) ring then i. =⇒ iii., and
if A is good local i. and iii. are equivalent. Show further that if A is good local then M(Γ) possesses
a non-trivial idempotent, P , (an element such that P ∗ P = P , P 6= 0, 6= 1) if and only if Γ has ii.

4. Write I = {U ∈ M(Γ) | ΓU Γ = 0} and given P ∈ M(Γ), set

B(P ) = {V ∈ M(Γ) | (∃ Z ∈ M(Γ))(V = P ∗ Z ∗ P )}.

If iii. above holds, show there exists P ∈ M(Γ) so that P ∗ P = P and ΓP Γ = Γ. For such a P , prove
that B(P ) is a subring of M(Γ), that M(Γ) ∼ = B(P ) q I in the category of A-modules, and that I
is a two-sided ideal of M(Γ) (by exhibiting I as the kernel of a surjective ring homomorphism whose
image you should find). Further show if i. holds, then B(P ) is isomorphic to the ring of r × r matrices
with entries from A. When A is a field show I is a maximal 2-sided ideal of M(Γ), here Γ 6= (0). Is I
the unique maximal (2-sided) ideal in this case?

5. Call an idempotent, P , of a ring maximal (also called principal ) iff when L is another idempotent,
then P L = 0 =⇒ L = 0. Suppose Γ satisfies condition iii. above, prove that an idempotent, P , of
M(Γ) is maximal iff ΓP Γ = Γ.

Problem 32 Let A be the field of real numbers R and conserve the notations of Problem 31. Write X for
a p × q matrix of functions of one variable, t, and consider the Γ-Riccati Equation

dX
= XΓX. ((∗)Γ )
dt
1. If q = p and Γ is invertible, show that either the solution, X(t), blows up at some finite t, or else X(t)
is equivalent to a matrix  
0 O(1) O(t) . . . O(tp−1 )
0 0 O(1) . . . O(tp−2 )
e
X(t) = ,
 ........... 
0 0 0 ... 0
where O(ts ) means a polynomial of degree ≤ s. Hence, in this case, X(t) must be nilpotent.

2. Suppose q 6= p and Γ has rank r. Let P be an idempotent of M(Γ) with ΓP Γ = Γ. If Z ∈ M(Γ), write
Z [ for Z − P ∗ Z ∗ P ; so Z [ ∈ I. Observe that I has dimension pq − r2 as an R-vector space. Now
assume that for a solution, X(t), of (∗)Γ , we have X(0) ∈ I. Prove that X(t) exists for all t. Can you
give necessary and sufficient conditions for X(t) to exist for all t?

3. Apply the methods of (2) to the case p = q but r = rank Γ < p. Give a similar discussion.

Problem 33 A module, M , over a ring, R, is called indecomposable iff we cannot find two submodules M1
∼
and M2 of M so that M −→ M1 q M2 in the category of R-modules.

1. Every ring is a module over itself. Show that if R is a local ring, then R is indecomposable as an
R-module.

2. Every ring, R, with unity admits a homomorphism Z → R (i.e., Z is an initial object in the category
RNG). The kernel of Z → R is the principal ideal nZ for some n ≥ 0; this n is the characteristic of R.
Show that the characteristic of a local ring must be 0 or a prime power. Show by example that every
possibility occurs as a characteristic of some local ring.
12 PROBLEMS

3. Pick a point in R or C; without loss of generality, we may assume this point is 0. If f is a function we
say f is locally defined at 0 iff f has a domain containing some (small) open set, U , about 0 (in either
R or C). Here, f is R- or C-valued, independent of where its domain is. When f and g are locally
defined at 0, say f makes sense on U and g on V , we’ll call f and g equivalent at 0 ⇐⇒ there exists
open W , 0 ∈ W , W ⊆ U ∩ V and f  W = g  W . A germ of a function at 0 is an equivalence class of
a function. If we consider germs of functions that are at least continuous near 0, then when they form
a ring they form a local ring.
Consider the case C and complex valued germs of holomorphic functions at 0. This is a local ring.
Show it is a good local ring.
In the case R, consider the germs of real valued C k functions at 0, for some k with 0 ≤ k ≤ ∞. Again,
this is a local ring; however, show it is NOT a good local ring.
Back to the case C and the good local ring of germs of complex valued holomorphic functions at 0.
Show that this local ring is also a principal ideal domain.

In the case of real valued C ∞ germs at 0 ∈ R, exhibit an infinite set of germs, each in the maximal
ideal, no finite subset of which generates the maximal ideal (in the sense of ideals). These germs are
NOT to belong to m2 .

Problem 34 Recall that for every integral domain, A, there is a field, Frac(A), containing A minimal among
all fields containing A. If B is an A-algebra, an element b ∈ B is integral over A ⇐⇒ there exists a monic
polynomial, f (X) ∈ A[X], so that f (b) = 0. The domain, A, is integrally closed in B iff every b ∈ B which
is integral over A actually comes from A (via the map A → B). The domain, A, is integrally closed (also
called normal ) iff it is integrally closed in Frac(A). Prove:


1. A is integrally closed ⇐⇒ A[X]/ f (X) is an integral domain for every MONIC irreducible polyno-
mial, f (X).


2. A is a UFD ⇐⇒ A possesses the ACC on principal ideals and A[X]/ f (X) is an integral domain for
every irreducible polynomial f (X). (It follows that every UFD is a normal domain.)

3. If k is a field and the characteristic of k is not 2, show that A = k[X, Y, Z, W ]/(XY − ZW ) is a normal
domain. What happens if char(k) = 2?

Problem 35 Suppose that R is an integral domain and F is its fraction field, Frac(R). Prove that, as
R-module, the field F is “the” injective hull of R. A sufficient condition that F/R be injective is that R be
a PID. Is this condition necessary? Proof or counter-example.

Problem 36 If A is a ring, write End∗ (A) for the collection of surjective ring endomorphisms of A. Suppose
A is commutative and noetherian, prove End∗ (A) = Aut(A).

Problem 37 Write M (n, A) for the ring of all n × n matrices with entries from A (A is a ring). Suppose
K and k are fields and K ⊇ k.

1. Show that if M, N ∈ M (n, k) and if there is a P ∈ GL(n, K) so that P M P −1 = N , then there is a

Q ∈ GL(n, k) so that QM Q−1 = N .

2. Prove that (1) is false for rings B ⊇ A via the following counterexample:
A = R[X, Y ]/(X 2 + Y 2 − 1), B = C[X, Y ]/(X 2 + Y 2 − 1). Find two matrices similar in M (2, B) but
NOT similar in M (2, A).
Pn
3. Let S n be the n-sphere and represent S n ⊆ Rn+1 as {(z0 , . . . , zn ) ∈ Rn+1 | 2
j=0 zj = 1}. Show
Pn
that there is a natural injection of R[X0 , . . . , Xn ]/( j=0 Xj − 1) into C(S ), the ring of (real valued)
2 n

continuous functions on S n . Prove further that the former ring is an integral domain but C(S n ) is not.
Find the group of units in the former ring.
PROBLEMS 13

Problem 38 (Rudakov) Say A is a ring and M is a rank 3 free A-module. Write Q for the bilinear form
whose matrix (choose some basis for M ) is
 
1 a b
0 1 c .
0 0 1

Thus, if v = (x, y, z) and w = (ξ, η, ζ), we have

  
1 a b ξ
Q(v, w) = (x, y, x) 0 1 c  η  .
0 0 1 ζ

Prove that Q(w, v) = Q(v, Bw) with B = I + nilpotent ⇐⇒ a2 + b2 + c2 = abc.

Problem 39 Let M be a Λ-module (Λ is not necessarily commutative) and say N and N 0 are submodules
of M .

1. Suppose N + N 0 and N ∩ N 0 are f.g. Λ-modules. Prove that both N and N 0 are then f.g. Λ-modules.

2. Give a generalization to finitely many submodules, N1 , . . . , Nt of M .

3. Can you push part (2) to an infinite number of Nj ?

4. If M is noetherian as a Λ-module, is Λ necessarily noetherian as a ring (left noetherian as M is a left

module)? What about Λ = Λ/Ann(M )?

Problem 40 Suppose that V is a not necessarily finite dimensional vector space over a field, k. We assume
given a map from subsets, S, of V to subspaces, [S], of V which map satisfies:

(a) For every S, we have S ⊆ [S]

(b) [ ] is monotone; that is, S ⊆ T implies [S] ⊆ [T ].

(c) For every S, we have [S] = [[S]]

(d) If W is a subspace of V and W 6= V , then [W ] 6= V .

(1) Under conditions (a)—(d), prove that [S] = Span S.

(2) Give counter-examples to show that the result is false if we remove either (a) or (d). What about (b)
or (c)?
(3) What happens if we replace k by a ring R, consider subsets and submodules and replace Span S by
the R-module generated by S?

Problem 41 (Continuation of Problem 34)

1. Consider the ring A(n) = C[X1 , . . . , Xn ]/(X12 + · · · + Xn2 ). There is a condition on n, call it C(n), so
that A(n) is a UFD iff C(n) holds. Find explicitly C(n) and prove the theorem.

2. Consider the ring B(n) = C[X1 , . . . , Xn ]/(X12 + X22 + X33 + · · · + Xn3 ). There is a condition on n, call
it D(n), so that B(n) is a UFD iff D(n) holds. Find explicitly D(n) and prove the theorem.

3. Investigate exactly what you can say if C(n) (respectively D(n)) does not hold.

4. Replace C by R and answer (1) and (2).

14 PROBLEMS



5. Can you formulate a theorem about the ring A[X, Y ]/ f (X, Y ) of the form A[X, Y ]/ f (X, Y ) is a
UFD provided f (X, Y ) · · · ? Here, A is a given UFD and f is a polynomial in A[X, Y ]. Your theorem
must be general enough to yield (1) and (2) as easy consequences. (You must prove it too.)

Problem 42 (Exercise on projective modules) In this problem, A ∈ Ob(CR).

1. Suppose P and P 0 are projective A-modules, and M is an A-module. If

0 → K →P → M → 0 and
0 0
0 → K →P → M → 0

are exact, prove that K 0 q P ∼

= K q P 0.
2. If P is a f.g. projective A-module, write P D for the A-module HomA (P, A). We have a canonical map
P → P DD . Prove this is an isomorphism.
3. Again, P is f.g. projective; suppose we’re given an A-linear map µ : EndA (P ) → A. Prove: there
exists a unique element f ∈ EndA (P ) so that (∀ h ∈ EndA (P ))(µ(h) = tr(hf )). Here, you must define
the trace, tr, for f.g. projectives, P , as a well-defined map, then prove the result.
4. Again, P is f.g. projective; µ is as in (3). Show that µ(gh) = µ(hg) ⇐⇒ µ = a tr for some a ∈ A.
5. Situation as in (2), then each f ∈ EndA (P ) gives rise to f D ∈ EndA (P D ). Show that tr(f ) = tr(f D ).

6. Using categorical principles, reformulate (1) for injective modules and prove your reformulation.

Problem 43 Suppose K is a commutative ring and a, b ∈ K. Write A = K[T ]/(T 2 − a); there is an
automorphism of A (the identity on K) which sends t to −t, where t is the image of T in A. If ξ ∈ A, we
write ξ for the image of ξ under this automorphism. Let H(K; a, b) denote the set
(  )
ξ bη
H(K; a, b) = ξ, η ∈ A ,
η ξ

this is a subring of the 2 × 2 matrices over A. Observe that q ∈ H(K; a, b) is a unit there iff q is a unit of
the 2 × 2 matrices over A.
1. Consider the non-commutative polynomial ring KhX, Y i. There is a 2-sided ideal, I, in KhX, Y i
so that I is symmetrically generated vis a vis a and b and KhX, Y i/I is naturally isomorphic to
H(K; a, b). Find the generators of I and establish the explicit isomorphism.
2. For pairs (a, b) and (α, β) decide exactly when H(K; a, b) is isomorphic to H(K; α, β) as objects of the
comma category RNGK .
3. Find all isomorphism classes of H(K; a, b) when K = R and when K = C. If K = Fp , p 6= 2 answer
the same question and then so do for F2 .
4. When K is just some field, show H(K; a, b) is a “division ring” (all non-zero elements are units) ⇐⇒
the equation X 2 − aY 2 = b has no solution in K (here we assume a is not a square in K). What is
the case if a is a square in K?
5. What is the center of H(K; a, b)?
6. For the field K = Q, prove that H(Q; a, b) is a division ring ⇐⇒ the surface aX 2 + bY 2 = Z 2 has no
points whose coordinates are integers except 0.
PROBLEMS 15

Problem 44

1. If A is a commutative ring and f (X) ∈ A[X], suppose (∃ g(X) 6= 0)(g(X) ∈ A[X] and g(X)f (X) = 0).
Show: (∃ α ∈ A)(α 6= 0 and αf (X) = 0). Caution: A may possess non-trivial nilpotent elements.

2. Say K is a field and A = K[Xij , 1 ≤ i, j ≤ n]. The matrix

 
X11 . . . X1n
M = . . . . . 
Xn1 . . . Xnn

has entries in A and det(M ) ∈ A. Prove that det(M ) is an irreducible polynomial of A.

Problem 45 Let A be a commutative noetherian ring and suppose B is a commutative A-algebra which is
f.g. as an A-algebra. If G ⊆ AutA−alg (B) is a finite subgroup, write

B G = {b ∈ B | σ(b) = b, all σ ∈ G}.

Prove that B G is also f.g. as an A-algebra; hence B G is noetherian.

Problem 46 Again, A is a commutative ring. Write RCF(A) for the ring of ∞ × ∞ matrices all of whose
rows and all of whose columns possess but finitely many (not bounded) non-zero entries. This is a ring
under ordinary matrix multiplication (as you see easily).

1. Specialize to the case A = C; find a maximal two-sided ideal, E, of RCF(C). Prove it is such and is
the only such. You are to find E explicitly. Write E(C) for the ring RCF(C)/E.

2. Show that there exists a natural injection of rings Mn (= n × n complex matrices) ,→ RCF(C) so
that the composition Mn → E(C) is still injective. Show further that if p | q we have a commutative
diagram

Mp   q / Mq
FF x m
FF
FF xxx M
FF xx
" |xx
E(C)

Problem 47 (Left and right noetherian) For parts (1) and (2), let A = ZhX, Y i/(Y X, Y 2 )—a
non-commutative ring.

1. Prove that
Z[X] ,→ ZhX, Y i → A


is an injection and that A = Z[X] q Z[X]y as a left Z[X]-module (y is the image of Y in A); hence
A is a left noetherian ring.

2. However, the right ideal generated by {X n y | n ≥ 0} is NOT f.g. (prove!); so, A is not right noetherian.

3. Another example. Let ( )


a b
C= a ∈ Z; b, c, ∈ Q .
0 c

Then C is right noetherian but NOT left noetherian (prove!).

Problem 48 If {Bα , ϕβα } is a right mapping system of Artinian rings and if B = limα Bα and B is noetherian,
−→
prove that B is Artinian.
16 PROBLEMS

Problem 49 Suppose that A is a commutative noetherian ring and B is a given A-algebra which is flat and
finite as an A-module. Define a functor Idem B/A (−) which associates to each A-algebra, T the set
Idem B/A (T ) = Idem(B ⊗A T ) consisting of all idempotent elements of the ring B ⊗A T .
(1) Prove the functor Idem B/A is representable.
(2) Show the representing ring, C, is a noetherian A-algebra and that it is étale over A.

Problem 50 (Vector bundles) As usual, TOP is the category of topological spaces and k will be either the
real or complex numbers. All vector spaces are to be finite dimensional. A vector space family over X is an
object, V , of TOPX (call p the map V → X) so that
i. (∀ x ∈ X)(p−1 (x) (denoted Vx ) is a k-vector space)
ii. The induced topology on Vx is the usual topology it has as a vector space over k.
Example: The trivial family X Π k n (fixed n).
Vector space families over X form a category, VF(X), if we define the morphisms to be those morphisms,
ϕ, from TOPX which satisfy:

(∀ x ∈ X)(ϕx : Vx → Wx is a linear map.)

θ
1. Say Y −→ X is a continuous map. Define a functor θ∗ : VF(X) VF(Y ), called pullback. When Y
is a subspace of X, the pullback, θ∗ (V ), is called the restriction of V to Y , written V  Y .
A vector space family is a vector bundle ⇐⇒ it is locally trivial, that is:
(∀ x ∈ X)(∃ open U )(x ∈ U ) (so that V  U is isomorphic (in VF(U )) to U Π k n , some n). Let
Vect(X) denote the full subcategory of VF(X) formed by the objects that are vector bundles.
2. Say X is an r-dimensional vector space considered in TOP. Write P(X) for the collection of all
hyperplanes through 0 ∈ X, then P(X) is a topological space and is covered by opens

each isomorphic
to an (r − 1)-dimensional vector space. On P(X) we make an element of VF P(X) : W is the set of
pairs (ξ, ν) ∈ P(X) Π X D so that ξ ⊂ ker ν. Here, X D is the dual space of X. Show that W is a line
bundle on P(X).
3. If V ∈ Vect(X) and X is connected, then dim(Vx ) is constant on X. This number is the rank of V .
4. A section of V over U is a map σ : U → V  U so that p ◦ σ = idU . Write Γ(U, V ) for the collection of
sections of V over U . Show: If V ∈ Vect(X), each section of V over U is just a compatible family of
locally defined vector valued functions on U . Show further that Γ(U, V ) is a vector space in a natural
way.
5. Say V and W are in Vect(X), with ranks p and q respectively. Show: Hom(V, W ) is isomorphic to the
collection of locally defined “compatible” families of continuous functions U → Hom(k p , k q ), via the
local description
ϕ ∈ Hom(V, W ) ϕe : U → Hom(k p , k q ),


e
where ϕ(u, v) = u, ϕ(u)(v) . Here, V  U is trivial and v ∈ k p .

Now Iso(k p , k q ) = {ψ ∈ Hom(k p , k q ) | ψ is invertible} is an open of Hom(k p , k q ).

6. Show: ϕ ∈ Hom(V, W ) is an isomorphism ⇐⇒ for a covering family of opens, U (⊆ X), we have
e ) ⊆ Iso(k p , k q ) ⇐⇒ (∀ x ∈ X)(ϕx : Vx → Wx is an isomorphism).
ϕ(U
7. Show {x | ϕx is an isomorphism (here, ϕ ∈ Hom(U, V ))} is open in X.
8. Show all of (1) to (6) go over when X ∈ C k −MAN (0 ≤ k ≤ ∞) with appropriate modifications; C k
replacing continuity where it appears.
PROBLEMS 17

Problem 51 (Linear algebra for vector bundles). First just look at finite dimensional vector spaces over k
(remember k is R or C) and say F is some functor from vector spaces to vector spaces (F might
even be a
several variable functor). Call F continuous ⇐⇒ the map Hom(V, W ) → Hom F (V ), F (W ) is continuous.
(Same definition for C k , 1 ≤ k ≤ ∞, ω). If we have such an F , extend it to bundles via the following steps:

1. Suppose V is the trivial bundle: X Π k p . As sets, F (X Π k p ), is to be just X Π F (k p ), so we

give F (X Π k p ) the product

topology. Prove: If ϕ ∈ Hom(V, W ), then F (ϕ) is continuous, therefore
F (ϕ) ∈ Hom F (V ), F (U ) . Show, further, ϕ is an isomorphism =⇒ F (ϕ) is an isomorphism.
S
2. Set F (V ) = x∈X (x, Vx ), then the topology on F (V ), when V is trivial, appears to depend on the
specific trivialization. Show this is not true—it is actually independent of same.

3. If V is any bundle, then V  U is trivial for small open U , so by (1)
and (2), F (V  U ) is a trivial
bundle. Topologize F (V ) by calling a set, Z, open iff Z ∩ F (V  U ) is open in F (V  U ) for all U
where V  U is trivial. Show that if Y ⊆ X, then the topology on F (V  Y ) is just that on F (V )  Y ,
that ϕ : V → W continuous =⇒ F (ϕ)
is continuous

and extend all these things to C k . Finally prove:
∗ ∼ ∗
If f : Y → X in TOP then f F (V ) = F f (V ) and similarly in C −MAN.k

4. If we apply (3) , we get for vector bundles:

(a) V q W , more generally finite coproducts

(b) V D , the dual bundle
(c) V ⊗ W
(d) Hom(V, W ), the vector bundle of (locally defined) homomorphisms.


Prove: Γ U, Hom(V, W ) ∼ = Hom(V  U, W  U ) for every open, U , of X. Is this true for the bundles
of (a), (b), (c)?

Problem 52 Recall that if R ∈ RNG, J(R)—the Jacobson radical of R— is just the intersection of all (left)
maximal ideals of R. The ideal, J(R), is actually 2-sided.

1. Say J(R) = (0) (e.g., R = Z). Show that no non-projective R-module has a projective cover.
`
2. Suppose Mi , i = 1,`. . . , t are R-modules with projective covers P1 , . . . , Pt . Prove that i Pi is a
projective cover of i Mi .

3. Say M and N are R-modules and assume M and M q N have projective covers. Show that N has one.

4. In M is an R-module, write (as usual) M D = HomR (M, R). Then M D is an Rop -module. Prove that
if M is finitely generated and projective as an R-module, then M D has the same properties as an
Rop -module.

Problem 53 Let {Mα } be a given family of Rop -modules. Define, for R-modules, two functors:
 Y  
U :N M α ⊗R N
α
Y
V :N (Mα ⊗R N ).
α

1. Show that V is right-exact and is exact iff each Mα is flat over R.

2. Show there exists a morphism of functors θ : U → V . Prove that θN : U (N ) → V (N ) is surjective if

N is finitely generated, while θN is an isomorphism if N is finitely presented.
18 PROBLEMS

Problem 54 (Continuation of Problems 50 and 51). Let V and W be vector bundles and ϕ : V → W a
homomorphism. Call ϕ a monomorphism (respectively epimorphism) iff
(∀ x ∈ X)(ϕx : Vx → Wx is a monomorphism (respectively epimorphism)). Note: ϕ is a monomorphism iff
ϕD : W D → V D is an epimorphism. A sub-bundle of V is a subset which is a vector bundle in the induced
structure.

1. Prove: If ϕ : V → W is a monomorphism, then ϕ(V ) is a sub-bundle of W . Moreover, locally on X,

there exists a vector bundle, G, say on the open U ⊆ X, so that (V  U ) q G ∼= W  U (i.e., every sub-
bundle is locally part of a coproduct decomposition of W ). Prove also: {x | ϕx is a monomorphism} is
open in X. (Suggestion: Say x ∈ X, pick a subspace of Wx complementary to ϕ(Vx ), call it Z. Form
G = X Π Z. Then there exists a homomorphism V q G → W , look at this homomorphism near the
point x.)
S
2. Say V is a sub-bundle of W , show that x∈X (x, Wx /Vx ) (with the quotient topology) is actually a
vector bundle (not just a vector space family) over X.

3. Now note we took a full subcategory of VF(X), so for ϕ ∈ Hom(V, W ) with V, W ∈ Vect(X), the
dimension of ker ϕx need not be locally constant on X. When it is locally constant, call ϕ a bundle
homomorphism. Prove that if ϕ is a bundle homomorphism from V to W , then
S
(i) x (x, ker ϕx ) is a sub-bundle of V
S
(ii) x (x, Im ϕx ) is a sub-bundle of W , hence
S
(iii) x (x, coker ϕx ) is a vector bundle (quotient topology).

We refer to these bundles as ker ϕ, Im ϕ and coker ϕ, respectively. Deduce from your argument for (i)
that

(iv) Given x ∈ X, there exists an open U , with x ∈ U , so that (∀ y ∈ U )(rank ϕy ≥ rank ϕx ). Of

course, this ϕ is not necessarily a bundle homomorphism.

Problem 55 (Continuation of Problem 54) In this problem, X is compact Hasudorff . We use two results
from analysis:

A) (Tietze extension theorem). If X is a normal space and Y a closed subspace while V is a real vector
space, then every continuous map Y → V admits an extension to a continuous map X → V . Same
result for X ∈ C k −MAN and C k maps.

B) (Partitions of unity). Say X is compact Hausdorff and {Uα } is a finite open cover of X. There exist
continuous maps, fα , taking X to R such that

(i) fα ≥ 0, (all α)
(ii) supp(fα ) ⊆ Uα (so fα ∈ C00 (Uα ))
P
(iii) (∀ x ∈ X)( α fα (x) = 1).

The same is true for C k −MAN (X compact!) and C k functions (1 ≤ k ≤ ∞).

1. Extend Tietze to vector bundles: If X is compact Hausdorff, Y ⊆ X closed and V ∈ Vect(X), then every
section σ ∈ Γ(Y, V Y ) extends to a section in Γ(X, V ). (Therefore, there exist plenty of continuous
or C ∞ global sections of V . FALSE for holomorphic sections). Apply this to the bundle Hom(V, W )
and prove: If Y is closed in X with X (as usual) compact Hausdorff or compact C k -manifold and if
ϕ : V  Y → W  Y is an isomorphism of vector bundles, then there exists an open, U , with Y ⊆ U , so
that ϕ extends to an isomorphism V  U → W  U .
PROBLEMS 19

2. Every vector space possesses a metric (take any of the p-norms, or take the 2-norm for simplicity). It’s
easy to see that metrics then exist on trivial bundles. In fact, use the 2-norm, so we can “bundleize”
the notion of Hermitian form (Problem 51) and get the bundle Herm(V ). Then an Hermitian metric
on V is a global section of Herm(V ) which is positive definite, at each x ∈ X. Show every bundle
possesses an Hermitian metric.
3. If we are given vector bundles and bundle homomorphisms, we say the sequence
· · · → Vj → Vj+1 → Vj+2 → · · ·
of such is exact iff for each x ∈ X, the sequence of vector spaces
· · · → Vj,x → Vj+1,x → Vj+2,x → · · ·
is exact. Prove: If 0 → V 0 → V → V 00 → 0 is an exact sequence of vector bundles and bundle
homomorphisms, then V ∼= V 0 q V 00 . (This is not true for holomorphic bundles.)
4. Consider a vector bundle, V , and a subspace, Σ, of the vector space Γ(X, V ). We get the trivial bundle
X Π Σ and a natural homomorphism X Π Σ → V , via
(x, σ) → σ(x).
Prove: If X is compact Hausdorff (or compact C k −MAN), there exists a finite dimensional subspace,
Σ, of Γ(X, V ) so that the map X Π Σ → V is surjective. Thus there exists a finite dimensional
surjective family of C-(respectively C k -) sections of V . Use (3) to deduce: Under the usual assumption
on X, for each vector bundle, V , on X, there exists a vector bundle, W , on X, so that V q W is a
trivial bundle.
5. Write C(X) (respectively C k (X), 1 ≤ k ≤ ∞) for the ring of continuous (respectively C k ) functions
(values in our field) on X, where X is compact Hausdorff (respectively a compact manifold). In a
natural way (pointwise multiplication), Γ(X, V ) is an A-module (A = C(X), C k (X)), and Γ gives a
functor from vector bundles, V , to Mod(A). Trivial bundles go to free modules of finite rank over A
(why?) Use the results above to prove:
Γ gives an equivalence of categories: Vect(X) (as full subcaregory of V F (X)) and the full
subcategory of A-modules whose objects are f.g. projective modules.
Problem 56
1. Say M is a f.g. Z-module, 6= (0). Prove there exists a prime p so that M ⊗Z Z/pZ 6= (0). Deduce: No
divisible abelian group [cf. Problem 24] can be f.g.
2. Say M , M 00 are Z-modules and M is f.g. while M 00 is torsion free. Given ϕ ∈ Hom(M, M 00 ) suppose
(∀ primes p)(the induced map M ⊗Z Z/pZ → M 00 ⊗Z Z/pZ is a monomorphism). Show that ϕ is a
monomorphism and M is free.
3. If M is a divisible abelian group, prove that M possesses no maximal subgroup. Why does Zorn’s
Lemma fail?
Problem 57 Given Λ, Γ ∈ RNG and a ring homomorphism Λ → Γ (thus, Γ is a Λ-algebra), if M is a
Λ-module, then M ⊗Λ Γ has the natural structure of a Γop -module. Similarly, if Z is both a Λop -module and
a Γ-module, then Z ⊗Λ M is still a Γ-module. Now let N be a Γ-module,
1. Prove there is a natural isomorphism
∼
HomΓ (Z ⊗Λ M, N ) −→ HomΛ (M, HomΓ (Z, N )). (∗)
Prove, in fact, the functors M M ⊗Λ Z and N HomΓ (Z, N ) are adjoint functors, i.e., (∗) is
functorial.
20 PROBLEMS

2. Establish an analog of (∗):

HomΓ (M, HomΛ (Z, N )) ∼
= HomΛ (Z ⊗Γ M, N ) (∗∗)
under appropriate conditions on Z, M and N (what are they?)
3. Show: M projective as a Λop -module, Z projective as a Γop -module =⇒ M ⊗Λ Z is projective as a
Γop -module. In particular, M projective as a Λop -module =⇒ M ⊗Λ Γ is projective as a Γop -module
and of course, the same statement (without the op) for Z ⊗Λ M and Γ ⊗Λ M . Show further that, if N
is Λ-injective, then HomΛ (Γ, N ) is Γ-injective.
4. For abelian groups, M , write M D = HomZ (M, Q/Z). Then, if M is free, M D is injective as a Z-module
(why?). From this deduce: Every abelian group is a subgroup of an injective abelian group.
5. (Eckmann) Use (3) and (4) to prove the Baer Embedding Theorem: For every ring, Γ, each Γ-module
is a submodule of an injective Γ-module.
Problem 58 Here, A and B are commutative rings and ϕ : A → B a ring homomorphism so that B is an
A-algebra. Assume B is flat (i.e., as an A-module, it’s flat). Define a homomorphism
θ : HomA (M, N ) ⊗A B → HomB (M ⊗A B, N ⊗A B)
(functorial in M and N )—how?
1. If M is f.g. as an A-module, θ is injective.
2. If M is f.p. as an A-module, θ is an isomorphism.
3. Assume M is f.p. as an A-module, write a for the annihilator of M (= (M → (0))). Prove that a ⊗A B
is the annihilator of M ⊗A B in B.
Problem 59 Let k be a field and f be a monic polynomial of even degree in k[X].
1
1. Prove there exist g, r ∈ k[X] such that f = g 2 + r and deg r < 2 deg f . Moreover, g and r are unique.

Now specialize to the case k = Q, and suppose f has integer coefficients. Assume f (X) is not the
square of a polynomial with rational coefficients.
2. Prove there exist only finitely many integers, x, such that the value f (x) is a square, say y 2 , where
y ∈ Z. In which ways can you get the square of an integer, y, by adding 1 to third and fourth powers
of an integer, x?
3. Show there exists a constant, KN , depending ONLY on the degree, N , of f so that:
If all coefficients of f are bounded in absolute value by C (≥ 1) then whenever hx, yi is a
solution of y 2 = f (x) (with x, y ∈ Z) we have |x| ≤ KN C N .
4. What can you say about the number of points hx, yi with rational coordinates which lie on the (hyper-
elliptic) curve Y 2 = f (X)?
Q
Problem 60 Consider Mod(Z) and copies of Z indexed by N = {1, 2, . . .}. Form the module Z. It is a
Q N
product of ℵ0 projective modules. Show M = Z is not projective as a Z-module. (Suggestions: Establish
N
that each submodule of a free module over a PID is again free, therefore we need to show M is not free.
Look at
K = {ξ = (ξj ) ∈ M | (∀ n)(∃ k = k(n))(2n | ξj if j > k(n))}.5
This is a submodule of M ; show K/2K is a vector space over Z/2Z of the same dimension as K and finish
up. Of course, 2 could be replaced by any prime). So, products of projectives need not be projective.
5 The condition means that limj7→∞ ξj is zero in the “2-adic numbers” Q2 .
PROBLEMS 21

Problem 61
θ
1. Say A −→ B is a homomorphism of commutative rings and suppose it makes B a faithfully flat
A-module. Show that θ is injective.
2. Hypotheses as in (1), but also assume B is finitely presented as an A-algebra (e.g., B is finitely
generated and A is noetherian). Show that there exists an A-module, M , so that B ∼ = A q M , as
A-modules.
3. Assume A and B are local rings, θ : A → B is a ring map (N.B. so that we assume
θ(mA ) ⊆ mB ) and B, as an A-module, is flat. Write N (A), respectively N (B), for the nilradicals of
A, respectively B. [That is,

N (A) = ξ ∈ A | (∃n ∈ N)(ξ n = 0) , etc.]

Prove:
(a) If N (B) = (0), then N (A) = (0).
(b) If B is an integral domain, so is A.
Are the converses of (a), (b) true? Proof or counter-example.

Problem 62 Here, I is an index set and S(I) is the set of all finite subsets of I. Partially order S(I) by
inclusion, then it is directed6 Also, let C be a category having finite products or finite coproducts as the
case may be below (e.g., groups, Ω-groups,` modules). ∗Say Qfor each α ∈ I we are given an object Mα ∈ C.
For ease of notation below, write MS = Mα and MS = Mα , where S ∈ S(I) is given. Prove:
α∈S α∈S
If C has right limits and finite coproducts, then C has arbitrary coproducts; indeed,
a
lim MS = Mα .
−→
S∈S(I) α∈I

Prove a similar statement for left limits and products.

Problem 63 Recall that a ring, Λ, is semi-simple 7 iff every Λ-module, M , has the property:
(∀ submodules, M 0 , of M )(∃ another submodule, M 00 , of M )(M ∼
= M 0 q M 00 ).
There is a condition on the positive integer, n, so that n has this condition ⇐⇒ Z/nZ is semi-simple. Find
the condition and prove the theorem.


Problem 64 In this problem, A ∈ CR. If α1 , . . . , αm are in A, write α1 , . . . , αn for the ideal generated by
α1 , . . . , αn in A. Recall that K0 (A), the Grothendieck group of A, is the quotient of the free abelian group
on the (isomorphism classes of) finitely generated A-modules (as generators) by the subgroup generated by
the relations: if 0 → M 0 → M → M 00 → 0 is exact in Mod(A), then [M ] − [M 0 ] − [M 00 ] is a relation.
1. If α ∈ A, show that in K0 (A) we have
   
((α) → 0) = A/(α)

2. If A is a PID and M is a finite length A-module, show that [M ] = 0 in K0 (A).

3. Prove: If A is a PID, then for all finitely generated A-modules, M , there exists a unique integer
r = r(M ), so that [M ] = r[A] in K0 (A); hence, K0 (A) is Z. Prove further that
r(M ) = dim(M ⊗A Frac(A)).
6 One also says S(I) has the Moore–Smith property.
7 Cf. also, Problem 145.
22 PROBLEMS

Problem 65 Write LCAb for the category of locally compact abelian topological groups, the morphisms
being continuous homomorphisms. Examples include: Every abelian group with the discrete topology; R;
C; R/Z = T, etc. If G ∈ LCAb, write
GD = Homcts (G, T),
make GD a group via pointwise operations and topologize GD via the compact-open topology; that is, take
the sets 
U (C, ) = f ∈ GD | Im(f  C) ⊆ − < arg z < 
—where C runs over the compact subsets of G containing 0,  is positive and we identify T with the unit
circle in C—as a fundamental system of neighborhoods at 0 in GD .
1. Suppose G is actually compact. Prove GD is discrete in this topology. Likewise, prove if G is discrete,
then GD is compact in this topology. Finally prove GD is locally compact in this topology.
If {Gα , ϕβα } is
2. n a right (respectively left) mapping family of finite abelian groups, then
D

o
β D
Gα , ϕα becomes a left (respectively right) mapping family, again of finite abelian groups (how,
why?). Prove that
 D
lim G ∼
= lim GD
−→ α ←− α
α α
and  D
lim G ∼
= lim GD
←− α −→ α
α α
as topological groups. We call a group profinite ⇐⇒ it is isomorphic, as a topological group, to a left
limit of finite groups.
3. Prove the following three conditions are equivalent for an abelian topological group, G:
(a) G is profinite
(b) G is a compact, Hausdorff, totally disconnected group
(c) GD is a discrete torsion group.

4. For this part, Gα is a family of compact groups, not necessarily abelian, and the index set has
Moore–Smith. Assume we are given, for each α, a closed, normal
 subgroup of Gα , call it Sα , and that
β ≥ α =⇒ Gβ ⊆ Gα and Sβ ⊆ Sα . Show that the family Hα = Gα /Sα α can be made into a left
mapping family, in a natural way, and that
\ \
lim Hα ∼
= Gα / Sα (as topological groups.)
←−
α α α


5. If G is a compact topological group, write Uα |α ∈ I for the family of all open, normal subgroups of
G. Continue (3) by proving:
\
G is profinite ⇐⇒ G is compact and Uα = {1}.
α

6. Here, G need not be abelian. We define Zp as lim Z/pn Z and Ẑ as lim Z/nZ (Artin ordering for the
←− ←−
n
n’s). Quickly use (2) to compute ZD
p and (Ẑ) . Now consider the following mathematical statements:
D

Q
(a) Ẑ ∼
= p Zp
Q
(b) Q∗ ∼= Z/2Z Π p Z
PROBLEMS 23

X∞ Y
1 1
(c) s
= , if Re s > 1
n=1
n p
1 − 1/ps
(d) A statement you know well and are to fill in here concerning arithmetic in Z.
Show (a)-(d) are mutually equivalent.

Problem 66 Fix an abelian group, A, for what follows. Write An = A,

m

all n ∈ N andngive N the Artin
ordering. If
n  m (i.e. n|m) define ϕm
n : A n → A m by ϕm
n (ξ) = n ξ, and define ψm : Am → An by
n
ψm (ξ) = mn ξ, too. Let
 
à = lim An , ϕm n
and T (A) = lim Am , ψm .
−→ n ←−
(T (A) = full Tate group of A).
1. Prove that both à and T (A) are divisible groups.
ϕ
2. Show that if A = A1 −→ Ã is the canonical map into the direct limit, then ker(ϕ) = t(A), the torsion
subgroup of A. Hence, every torsion free abelian group is a subgroup of a divisible group. Given any
abelian group , A, write
0 → K → F → A → 0,
for some free abelian group F . Show that A may be embedded in F̃ /K; hence deduce anew that every
abelian group embeds in a divisible abelian group.
3. If A is a free Z-module, what is T (A)?
4. If A → B → 0 is exact, need T (A) → T (B) → 0 also be exact? Proof or counterexample.
5. Show that if T (A) 6= (0), then A is not finitely generated.

Problem 67 Again, as in Problem 61, let θ : A → B be a homomorphism of commutative rings and assume
B is faithfully flat over A via θ. If M is an A-module, write MB for M ⊗A B.
1. Prove: M is finitely generated as an A-module iff MB is finitely generated as a B-module.
2. Same as (1) but for finite presentation instead of finite generation.
3. Show: M is locally free over A iff MB is locally free over B.
4. When, if ever, is S −1 A faithfully flat over A?
Note, of course, that these are results on faithfully flat descent.

Problem 68 Here, Λ ∈ RNG and assume

0 → M 0 → M → M 00 → 0

is an exact sequence of Λ-modules.

1. Assume further, M 00 is a flat Λ-module. Prove: For all Λop -modules, N , the sequence

0 → N ⊗Λ M 0 → N ⊗Λ M → N ⊗Λ M 00 → 0

is again exact. (You might look at the special case when M is free first.)
2. Again assume M 00 is flat; prove M and M 0 are flat ⇐⇒ either is flat. Give an example of Λ, M 0 , M ,
M 00 in which both M and M 0 are flat but M 00 is not flat.
24 PROBLEMS

Problem 69 (Topologies, Sheaves and Presheaves). Let X be a topological space. We can make a category,
TX , which is specified by and specifies the topology as follows: Ob TX consists of the open sets in X. If
U, V ∈ Ob TX , we let 
∅ if U 6⊆ V ,
Hom(U, V ) =
{incl} if U ⊆ V ,
here {incl} is the one element set consisting of the inclusion map incl : U → V .
1. Show that U Π V —the fibred product of U and V (over X) in TX —is just U ∩ V . Therefore TX has
X
finite fibred products.
2. If C is a given category (think of C as Sets, Ab, or more generally Λ-Modules) a presheaf on X with
values in C is a cofunctor from TX to C. So, F is a presheaf iff (∀ open U ⊆ X)(F (U ) ∈ C) and when
U ,→ V , we have a map ρU V : F (V ) → F (U ) (in C) usually called restriction from V to U . Of course,
we have ρW V = ρ W
U ◦ ρ U
V . The basic example, from which all the terminology comes, is this:
C = R-modules (= vector spaces over R)
F (U ) = {continuous real valued functions on the open set U }.
ϕ
Now recall that a category is an abelian category iff for each morphism A −→ B in C, there are two
pairs: (ker ϕ, i) and (coker ϕ, j) with ker ϕ and coker ϕ objects of C and i : ker ϕ → A, j : B → coker ϕ
so that:
(a) HomC (A, B) is an abelian group, operation denoted +
(b) ker ϕ → A → B is zero in HomC (ker ϕ, B)
u
(c) If C −→ A → B is zero, there is a unique morphism C → ker ϕ so that u is the composition
i
C → ker ϕ −→ A
(d) Similar to (c) for coker, with appropriate changes.
j
Define Im ϕ as ker(B −→ coker ϕ). Now exact sequences make sense in C (easy, as you see). Write
P(X, C) for the category of presheaves on X with values in C. If C is abelian show that P(X, C) is an
abelian category, too, in a natural way.
3. If A ∈ Ob C, we can make a presheaf A by: A(U ) = A, all open U and if V ,→ U then ρU V = idA . This
is the constant presheaf with values in A. Generalize it as follows: Fix open U of X, define AU by:
a 
(0) if W 6⊆ U
AU (W ) = A=
A if W ⊆ U .
Hom(W,U )

Show AU is a presheaf and A is one of these AU ; which one? Generalize further: Say F is a presheaf
of sets on X, define AF by:
a 
AF (W ) = A = functions : F(W ) → A | these functions have finite support .
F (W )

Make AF into a presheaf on X; it is a clear generalization of AU and this, in turn, generalizes A.

4. Just as with the defining example in (2), which is called the presheaf of germs of continuous functions on
X, so we can define the presheaf of germs of C k -functions, real-analytic functions, complex holomorphic
functions, meromorphic functions when X is a real (resp. complex) manifold. Namely:
C k (U ) = {f : U → R | f is C k on U } 0≤k≤∞
C (U ) = {f : U → R | f is real analytic on U }
ω

Hol(U ) = {f : U → C | f is holomorphic on U }
Mer(U ) = {f : U → C | f is meromorphic on U }.
PROBLEMS 25

Prove: The collection {ZU | U open in X} is a set of generators for P(X, Ab); that is: For all presheaves
F, there
 is a subcollection
 of the U ’s, say {Uα | α ∈ Λ}, so that there is a surjection
` `
ZU  F, for some set I. (Then it turns out that every presheaf embeds in an injective
I α∈Λ
presheaf.)

5. Now sheaves are special kinds of presheaves. Say U ∈ TX and we have a family of morphisms of
TX : {Uα → US}α∈Λ (we’ll suppress mention of Λ in what follows). We call this family a covering
family ⇐⇒ U = U , i.e. the Uα form an open covering of U . Of course, if ξ ∈ F (U ), then

α α
ρU
U
α
(ξ) ∈ F U α , each α; here, F is a presheaf. Hence we get a map
Y

θ : F (U ) → F Uα .
α

U ∩U


Now if ξα ∈ F Uα , for each α, then ρUαα β ξα lies in F Uα ∩ Uβ therefore we get a map

Y

p1,α : F Uα → F Uα ∩ Uβ .
β

Take the product of these over α and get a map

Y
Y

p1 : F Uα → F Uα ∩ Uβ .
α α,β

U ∩U


If ξβ ∈ F Uβ then ρUαβ β ξβ ∈ F Uα ∩ Uβ therefore we get a map

Y

p2,β : F Uβ → F Uα ∩ Uβ .
α

Again the product over β gives:

Y
Y

p2 : F Uβ → F Uα ∩ Uβ ,
β α,β

hence we get two maps:

Y
p1 Y

F Uγ −→
−→ F Uα ∩ Uβ .
γ p2 α,β

Here
 is the definition of a sheaf : A sheaf, F , of sets is a presheaf, F , of sets so that (∀ open U )

∀ covers Uα → U α , the sequence

θ
Y
p1 Y

F (U ) → F Uγ −→
−→ F Uα ∩ Uβ (S)
γ p2 α,β

Q

is exact in the sense that θ maps F (U ) bijectively to the set ξγ ∈ F Uγ for which
γ





p1 ξγ = p2 ξγ .

Show that the presheaves of germs of continuous, k-fold continuous, differentiable, analytic, holomorphic
and meromorphic functions are all sheaves. In so doing understand what exactness of sequence (S) means.
Prove, however, that A is NOT generally a sheaf. (Note: a sheaf with values in Ab or RNG or Ω-groups is
just a presheaf with these values which forms a sheaf of sets.) For which presheaves, F, is AF a sheaf?
26 PROBLEMS

Problem 70 Consider P(X) and S(X) the categories of presheaves and sheaves of sets on X (our results
will also work for other image categories based on sets, e.g., Ab, RNG, TOP, etc.) We have the definition
of a sheaf so that
Y p1 Y
F (Uα ) −→
θ
F (U ) −→ −→ F (Uβ ∩ Uγ ) (S)
α p2
β,γ

is exact for all open covers, {Uα −→ U }α of any open U .

(1) There are two parts to the exactness of (S): θ is injective and the image of θ is the equalizer of p1
and p2 . Write that F satisfies (+) if θ is injective. Suppose that F is any presheaf, define
Y Y
F (+) = lim Ker ( F (Uα ) −→ −→ F (Uβ ∩ Uγ ))
−→
{Uα −→U } α β,γ

(the limit taken over all open covers, {Uα −→ U }, of the open U ). Show that F (+) satisfies (+).
ϕ
(2) If 0 −→ F 0 −→ F is exact in P(X, Ab), set (Cok ϕ)(U ) = Coker ϕ(U ) = Coker(F 0 (U ) −→ F (U )).
Prove that Cok ϕ satisfies (+).
(3) Suppose that F satisfies (+) show that F (+) satisfies (S), i.e., F (+) is a sheaf. Show further that, if
F satisfies (+), then Ker (F (U ) −→ F (+) (U )) = (0), i.e., F −→ F (+) is an injective map of presheaves. Set
F # = (F (+) )(+) , for any presheaf F .
(4) We know # is exact and i : S(X) → P(X) is left-exact. Prove that # is the left adjoint of i, that is

HomS(X) (F # , G) ∼
= HomP(X) (F, i(G)).

(5) For the derived functor Hq (F ) (= (Rq i)(F )) of i : S(X, Ab) P(X, Ab), prove that

(Hq (F ))# = (0).

Problem 71 (Grothendieck) In Problem 69, you proved the collection {ZU | U open in X} is a set of
generators for
P(X, Ab).
(1) Show that the collection {ZU } has the following property:
(G): For each presheaf, F , and for each monomorphism 0 −→ F 0 −→ F (in P(X, Ab)) with F 0 6= F , there
ϕ
is an open U ⊆ X and a morphism ZU −→ F , so that ϕ does not factor through a morphism ZU −→ F 0 .
Prove moreover that property (G) is equivalent to the fact that {ZU | U open in X} is a family of
generators for P(X, Ab).
a
(2) Write Z for the coproduct ZU in P(X, Ab), then Z is a generator for P(X, Ab). Show that
all U
a presheaf, Q, on X is injective if and only if for each monomorphism 0 −→ W −→ Z, every morphism
θ : W → Q extends to a morphism Z −→ Q.
(3) Imitate the construction for rings R, ideals A ⊆ R and R-modules M , of an injective hull for M (with
the correspondence R ←→ Z; A ←→ W ; M ←→ a presheaf F ) to show:
There exists a functor Q : F Q(F ) and a morphism of functors ψ : id → Q so that
(a) (∀F ∈ P(X, Ab))(ψF : F → Q(F ) is a monomorphism)
and
(b) Each Q(F ) is an injective presheaf.
PROBLEMS 27

This gives the proof that P(X, Ab) has enough injective objects.
(4) The ZU in S(X, Ab) defined as Z#
U form a set of generators for S(X, Ab). The same argument as in
(3) goes through and we obtain another proof (but similar to the text’s proof) that S(X, Ab) has enough
injectives.

Problem 72 (Grothendieck) Let P stand for the category of abelian presheaves, P(X, Ab), on the space
X.
(1) If U is an open in X and {Uα −→ U }α is an open covering of U , we have induced a diagram of
families of maps

←− ←−
U ←− {Uα } ←− ←− · · ·
←− {Uβ ∩ Uγ }β,γ ←−
←− {Uδ ∩ U ∩ Uη }δ,,η ←−
←−
Q Q Q
coming from the various projections (note that Uβ ∩ Uγ = Uβ Uγ ; Uδ ∩ U ∩ Uη = Uδ U Uη ; etc.).
When F is a presheaf, we get a simplicial diagram
Y Y −→ Y −→
F (U ) −→ F (Uα ) −→ F (Uβ ∩ Uγ ) −→ F (Uδ ∩ U ∩ Uη ) −→ · · ·
−→ −→ −→
α β,γ δ,,η
−→

and, by taking the alternating sum of these maps, we make a sequence

Y δ0
Y δ1
Y δ2
F (U ) −→ F (Uα ) −→ F (Uβ ∩ Uγ ) −→ F (Uδ ∩ U ∩ Uη ) −→ · · · . (∗)
α β,γ δ,,η
Q
For notation, write C r ({Uα −→ U }, F ) = α0 ,...,αr F (Uα0 ∩ · · · ∩ Uαr ), so that (∗) becomes

δ0 δ1 δ2
F (U ) −→ C 0 ({Uα −→ U }, F ) −→ C 1 ({Uα −→ U }, F ) −→ C 2 ({Uα −→ U }, F ) −→ · · · . (∗∗)

Show that (∗∗) is an augmented complex (of abelian groups). We’ll call (∗∗) the explicit Čech cochain
q
complex of the cover {Uα −→ U } with coefficients in F . Denote by Hxpl ({Uα −→ U }, F ) its q th cohomology
q q−1
group (= Ker δ /Im δ ).
(2) We know that HomP (ZV , F ) = F (V ) for all open V of X, show that
a
ZV = Z.
Hom(U,V )

(3) Now let F be an injective presheaf from P, show that

δ0 δ1 δ2
C 0 ({Uα −→ U }, F ) −→ C 1 ({Uα −→ U }, F ) −→ C 2 ({Uα −→ U }, F ) −→ · · · (∗∗∗)

is an exact sequence. (Suggestions. Show that the exactness of (∗∗∗) is equivalent to the exactness of
a a a
ZUα ←− ZUβ ∩Uγ ←− ZUδ ∩U ∩Uη ←− · · · (†)
α β,γ δ,,η

in the category P and check the latter exactness by evaluation on any open Y of X. For this, show that the
last sequence is induced by the simplicial diagram of indexing sets
a a ←− a ←−
Hom(Y, Uα ) ←− Hom(Y, Uβ ∩ Uγ ) ←− Hom(Y, Uδ ∩ U ∩ Uη ) ←− · · ·
←− ←− ←−
α β,γ δ,,η
←−
28 PROBLEMS

` Q `
and we can identify β,γ Hom(Y, Uβ ∩ Uγ ) with M M , where M = α Hom(Y, Uα ), etc. Thus, that (†) is
exact comes down to the exactness of the diagram
a a ←− a ←−
Z ←−
←− Z ←−
←− Z ←−
←− · · · .
M M
Q
M M
Q
M
Q
M
←−

But, construct a contracting homotopy for this last diagram and so complete proving (∗∗∗) is exact.)
•
(4) Prove that the δ-functor F Hxpl ({Uα −→ U }, F ) is universal and show that we have an isomorphism
H • ({Uα −→ U }, F ) ∼ •
= Hxpl ({Uα −→ U }, F )
(functorial in F ). Thus, the complex (∗∗∗) gives an explicit method for computing the cohomology groups,
H • ({Uα −→ U }, −), of the covering {Uα −→ U }α .
(5) Pass to the limit over all coverings of X and give an explicit complex to compute the Čech cohomology
groups Ȟ • (X, −).
Problem 73 If F is a sheaf of abelian groups on the space X, let’s agree to write F again when we consider
F as a presheaf.
(1) Show that there is an exact sequence
0 −→ Ȟ 2 (X, F ) −→ H 2 (X, F ) −→ Ȟ 1 (X, H1 (F ))
and that if Ȟ 3 (X, F ) = (0), then
0 −→ Ȟ 2 (X, F ) −→ H 2 (X, F ) −→ Ȟ 1 (X, H1 (F )) −→ 0
is exact.
(2) Let {Uα −→ X}α be an open cover of X and assume that
(∀α, β)(H 1 (Uα ∩ Uβ , F ) = (0)).
Deduce that the natural map
Ȟ 2 (X, F ) −→ H 2 (X, F )
is an isomorphism. If you assume only that
(∀α)(H 1 (Uα , F ) = (0))
can you still deduce that Ȟ 2 (X, F ) ∼
= H 2 (X, F )? Proof or counter-example.
(3) Can you continue the line of argument of (2) applied to groups such as H ? (Uα ∩ Uβ ∩ Uγ , F ),
etc. and deduce further isomorphisms between Čech and derived functor cohomology? For example, try
Ȟ 3 (X, F ) ∼
= H 3 (X, F ).
(4) In a similar vein to (2) and (3) above, prove the following (known as Cartan’s Isomorphism Theorem):
For the space X, let U be a family of open sets covering X so that
(a) If U, V ∈ U, then U ∩ V ∈ U
(b) U contains arbitrarily small opens of X
(c) If U ∈ U and q > 0, then Ȟ q (U, F ) = (0).

Then, the natural maps

Ȟ q (X, F ) −→ H q (X, F )
are isomorphisms for all q ≥ 0.
PROBLEMS 29

(Suggestions: Use induction on q, but replace X by any of the U of U. Use a spectral sequence at the
induction step to get Ȟ q (X, F ) ∼
= H q (X, F ). Now how do you further deduce Ȟ q (U, F ) ∼
= H q (U, F ) all
U ∈ U to complete the induction?)

Remark: Two main uses of Cartan’s Theorem are when X is a manifold and U is the family of all finite
intersections of all sufficiently small open balls around each point of X and when X is an algebraic variety
(over a field) and U is the collection of its affine open subvarieties.

Problem 74 Let k be a field, X an indeterminate (or transcendental) over k. Write A = k[X] and consider
an ideal, a, of A. The ideal, a, determines a topology on k[X]—called the a-adic topology—defined by taking
as a fundamental system of neighborhoods of 0 the powers {an | n ≥ 0} of a. Then a fundamental system of
neighborhoods at ξ ∈ A is just the collection {ξ + an | n ≥ 0}.
1. Show A becomes a topological ring (i.e. addition and multiplication are continuous) in this topology.
When is A Hausdorff in this topology?
2. The rings A/an = An form a left mapping system. Write
b = lim A/an
A
←−
n

b the a-adic completion of A. There is a map A → A;

and call A b when is it injective?

3. Consider a = (X) = all polynomials with no constant term. The ring A b in this case has special
notation: k[[X]]. Establish an isomorphism of k[[X]]

with the ring of formal power series over k (in
X) i.e. with the ring consisting of sequences cn , n ≥ 0, cn ∈ k with addition and multiplication
defined by:



cn + dn = cn + dn



X
cn · dn = en , en = ci dj
i+j=n

∞
X 
cn ↔ cn X n explains the name .
n=0

4. Show k[X] ,→ k[[X]], that k[[X]] is an integral domain and a local ring. What is its maximal ideal?
Now (X) = a is a prime ideal of k[X], so we can form k[X](X) . Prove that

k[X] ⊆ k[X](X) ⊆ k[[X]].

We have the (prime) ideal (X)e of k[X](X) . Form the completion of k[X](X) with respect to the
(X)e -adic topology. What ring do you get?

Problem 75 If k is any field, write A = k[[T1 , . . . , Tn ]] for the ring of formal power series over k in the
indeterminates T1 , . . . , Tn . Denote by Autk (A) the group of all k-automorphisms of A.
(1) Give necessary and sufficient conditions on the n power series S1 (T1 , . . . , Tn ), . . . , Sn (T1 , . . . , Tn ) in
order that the map
σ : Tj 7→ Sj (T1 , . . . , Tn )
be an element of Autk (A). In so doing, describe the group Autk (A).
(2) If now k is no longer necessarily a field but merely a commutative ring with unity, answer question
(1) for this case.
(3) Fix k, a commutative ring with unity, and consider the category, Alg(k), of k-algebras (say commu-
tative). Define a functor Aut(k[[T1 , . . . , Tn ]]/k)(−) by sending B ∈ Alg(k) to AutB (B[[T1 , . . . , Tn ]]) ∈ Grp.
Is this functor representable? How?
30 PROBLEMS

Problem 76 Prove that in the category of commutative A-algebras, the tensor product is the coproduct:

B ⊗A C ∼
= B q C.
A
Q
Which A-algebra is the product B C (in commutative A-algebras)?
A

Problem 77 Suppose A is a (commutative) semi-local ring obtained by localizing a f.g. C-algebra with
b for the J-adic
respect to a suitable multiplicative subset. Let J be the Jacobson radical of A and write A
b
completion of A. Is it true that every finitely generated A-module, M , has the form M = M0 ⊗A Ab for some
finitely generated A-module, M0 ? Proof or counter-example.

Problem 78 Here A is a commutative ring and we write Mn (A) for the ring of n × n matrices over A.
1. Prove: The following are equivalent
(a) A is noetherian
(b) For some n, Mn (A) has the ACC on 2-sided ideals
(c) For all n, Mn (A) has the ACC on 2-sided ideals.
2. Is this still valid if “noetherian” is replaced by “artinian” and “ACC” by “DCC”? Proof or counterex-
ample.
3. Can you make this quantitative? For example, suppose all ideals of A are generated by less than or
equal to N elements. What can you say about an upper bound for the number of generators of the
ideals of Mn (A)? How about the converse?

Problem 79 Refer to Problem 74. Write k((X)) for Frac(k[[X]]).

1. Show that ( )
∞
X
j
k((X)) = aj X | aj ∈ k and (∃ N )(aj = 0 if j < N )
j=−∞

where on the right hand side we use the obvious addition and multiplication for such expressions. If
ξ ∈ k((X)), write ord(ξ) = N ⇐⇒ N = largest integer so that a = 0 when j < N ; here, ξ 6= 0. If
ξ = 0, set ord(ξ) = ∞. One sees immediately that k[[X]] = ξ ∈ k((X)) | ord(ξ) ≥ 0 .


2. Write U for Gm k[[X]] and M for {ξ | ord(ξ) > 0}. Prove that k((X)) = M−1 ∪ U ∪ M (disjointly),
where
M−1 = {ξ | 1/ξ ∈ M}.
Now fix a real number, c, with 0 < c < 1. Define for ξ, η ∈ k((X)),

d(ξ, η) = cord(ξ−η) ,

then it should be clear that k((X)) becomes a metric space and that addition and multiplication
are continuous in the metric topology. Prove that k((X)) is complete in this topology (i.e., Cauchy
sequences converge), and that the topology is independent of which number c is chosen (with 0 < c < 1).
P∞
3. Suppose u ∈ k[[X]], u = j=0 aj X j , and a0 = 1. Pick an integer n ∈ Z and assume
(n, char(k)) = 1. Prove: There exists w ∈ k[[X]] such that wn = u. There is a condition on k so that
k((X)) is locally compact. What is it? Give the proof. As an example of limiting operations, prove

X ∞
1
= X j = lim (1 + X + · · · + X N ).
1 − x j=0 N →∞
PROBLEMS 31

∞
X
4. Given aj X j ∈ k((X)), its derivative is defined formally as
j=−∞

∞
X
jaj X j−1 ∈ k((X)).
j=−∞


Assume ch(k) = 0. Check mentally that α0 = 0 α ∈ k((X)) =⇒ α ∈ k. Is the map α 7→ α0 a
X∞
1 j
continuous linear transformation k((X)) → k((X))? Set η = X , so η ∈ k((X)). Prove that X
j=0
j!
and η are independent transcendentals over k.

5. Assume ch(k) = 0. A topological ring is one where addition and multiplication are continuous and
we have a Hausdorff topology. Topological k-algebras (k has the discrete topology) form a category
in which the morphisms are continuous k-algebra homomorphisms. An element λ in such a ring
is topologically nilpotent iff limn→∞ λn = 0. Let Ntop denote the functor which associates to each
topological k-algebra the set of its topological nilpotent elements. Prove that Ntop is representable.
As an application, let
∞
X ∞
X
X 2j+1 X 2j
s(X) = (−1)j , c(X) = (−1)j .
j=0
(2j + 1)! j=0
(2j)!

Then s0 (X) = c(X) and c0 (X) = −s(X), so c2 (X)+s2 (X) lies in k (the constants). Without computing
c2 (X) + s2 (X), show it is 1. (You’ll need Ntop , so be careful.)

6. Show that even though k(X) is dense in k((X)), the field k((X)) possesses infinitely many algebraically
independent transcendental elements over k(X). (Suggestion: Look in a number theory book under
“Liouville Numbers”; mimic what you find there.)



7. Assume ch(k) = 0. Let Ck k((X)) = α ∈ k((X)) | α is algebraic over k . Show that Ck k((X)) =
k.  
m m(m − 1) · · · (m − j + 1)
If R ⊆ k, write = for m ∈ R. If R 6⊆ k, do this only for m ∈ Q. Set
j j(j − 1) · · · 3 · 2 · 1
∞  
X m
ym = X j ∈ k[[X]].
j
j=0

If m ∈ Q and m = r/s, prove that ym

s
= (1 + x)r .
X∞

X j+1
Note that ym = 1 + O(X) and that O(X) ∈ Ntop k[[X]] . Let L(1 + X) = (−1)j , and set
j=0
(j + 1)


f (X)m = η m · L(f (X)) , where

X∞
1 j
η(X) = X and f (X) = 1 + O(X), some O(X)
j=0
j!

and m ∈ R (here, R ⊆ k). Show that

(1 + X)m = ym .

Problem 80 Say K is a field, A is a subring of K. Write k = Frac A.

1. If K is a finitely generated A-module, prove that k = A.

32 PROBLEMS

2. Suppose there exist finitely many elements α1 , . . . , αm ∈ K algebraic over k such that

K = A[α1 , . . . , αm ].




Prove ∃ b ∈ A b 6= 0 so that k = A[1/b] . Prove, moreover, that b belongs to every maximal ideal of A.

Problem 81 Refer to Problem 69. Look at P(X, Ab). We know that the functor F F (U ) taking
P(X, Ab) to Ab is representable.
1. Grothendieck realized that when computing algebraic invariants of a “space” (say homology, cohomol-
ogy, homotopy, K-groups, . . . ) the sheaf theory one needs to use could be done far more generally and
with far more richness if one abstracted the notion of “topology”. Here is the generalization:
(a) Replace TX by any category T .
To do sheaves, we need a notion of “covering”:
(b) We isolate for each U ∈ Ob T some families of morphisms {Uα → U }α and call each of these
a “covering” of U . So we get a whole collection of families of morphisms called Cov T and we
require
(i) Any isomorphism {V → U } is in Cov T
n o
(α) (α)
(ii) If {Uα → U }α is in Cov T and for all α, {Vβ → Uα }β is in Cov T , then Vβ → U is
α,β
in Cov T (i.e., a covering of a covering is a covering).
(iii) If {Uα → U }α is in Cov T and V → U is arbitrary then Uα Π V exists in T and
U
n o
Uα Π V → V
U α

is in Cov T (i.e., the restriction of a covering to V is a covering of V ; this allows the relative
topology—it is the axiom with teeth).
Intuition: A morphism V → U in T is an “open subset of U ”. N.B. The same V and U can give
more than one “open subset” (vary the morphism) so the theory isSvery rich. In our original example:
T = TX ; the family {Uα → U }α is in Cov T when and only when α Uα = U . Check the axioms (i),
(ii) and (iii).
Now a presheaf is just a cofunctor T → Sets or Ab, etc. and a sheaf is a presheaf for which
Y
p1 Y  
F (U ) → F Uγ −→
−→ F U α Π Uβ (S)
p2 U
γ α,β

is exact for every U ∈ T and every {Uγ → U }γ in Cov T . One calls the category T and its distinguished
families Cov T a site (topology used to be called “analysis situs”).

Given a category, say T , assume T has finite fibred products. A family of morphisms {Uα → U }α in
T is called a family of universal, effective epimorphisms iff
(a) ∀Z ∈ Ob T Y Y



Hom(U, Z) → Hom Uγ , Z −→
−→ Hom Uα Π Uβ , Z
U
γ α,β

is exact (in Sets) AND

n o
(b) The same for Uα Π V → V vis a vis all Z as in (a). (Condition (b) expresses universality,
U α
and (a) expresses effectivity of epimorphisms.)
PROBLEMS 33

Decree that Cov T is to consist of families of universal, effective epimorphisms. Show that T with this
Cov T is a site—it is called the canonical site on T , denoted Tcan .
2. For Tcan , every representable cofunctor on T is a sheaf (give the easy proof). Note that if T ⊆ Te
where Te is a bigger category, and if Cov T lies in the universal, effective epimorphisms for Te , then any
cofunctor on T , representable in Te , is a sheaf on Tcan . For example, prove that if Te is all topological
spaces and TX is our beginning category of Problem 69, then TX ⊆ Te and prove that open coverings in
TX (as in Problem 69) are universal, effective epimorphisms in Te . Hence, for ANY topological space,
Y, U Homtop.spaces (U, Y ) is a sheaf on TX .
S
3. Let T = Sets and let {Uα → U }α be in Cov T when α (Images of Uα ) = U . Prove that the sheaves
on T with values in Sets are exactly the representable cofunctors on T .
4. Generalize (3): If G is a given group, let TG be the category of sets with a G-action.
S Make (TG )can
the canonical site on TG . Prove: Coverings are families {Uα → U }α so that α (Im Uα ) = U (all
are G-sets, morphisms are G-morphisms). Once again, prove: The sheaves on (TG )can are exactly
the representable cofunctors on TG . Prove further: The sheaves on (TG )can with values in Ab form a
category equivalent to the category of G-modules; namely the equivalence is given by taking a sheaf
to its representing object, a G-module.

Problem 82 Consider the two rings A = R[T ] and B = C[T ]. Show that Max(B) is in one-to-one cor-
respondence with the points of the complex plane while Max(A) is in one-to-one correspondence with the
closed upper half plane: {ξ ∈ C | Im(ξ) ≥ 0}. Since A is a PID (so is B) we can characterize an ideal by
its generator. In these terms, which ideals of Max(A) correspond to points in Im(ξ) > 0, which to points on
the real line? What about Spec B and Spec A?

Problem 83 Suppose that f (X, Y ) and g(X, Y ) are two irreducible polynomials with complex coefficients.
Assume neither is a scalar multiple of the other. Show that the set

S = {(α, β) ∈ C2 | f (α, β) = g(α, β) = 0}

is finite. (There are many ways of doing this; try to pick a way that is as elementary as possible.)

Problem 84 When X is compact Hausdorff and A = C(X), we identified X and Max(A) in the text via
x 7→ mx . Now Max(A) has the induced topology from Spec A.
1. Show the induced topology on Max(A) is compact Hausdorff by proving x 7→ mx is a homeomorphism.
2. Prove all finitely generated ideals of A are principal but that no maximal ideal is finitely generated.
Note that some extra condition on X is needed. For example, X should not be finite.

Problem 85
1. Given A → B a homomorphism prove that B is faithfully flat over A iff B is flat over A and the map
Spec B → Spec A is surjective.
2. Say A → B is a homomorphism and B is faithfully flat over A. Assume A is noetherian. Show that
the topology on Spec A is the quotient topology from Spec B.
Q
Problem 86 Here A is a commutative ring, but not necessarily with unity. Let A# denote A Z (category
of sets) with addition componentwise and multiplication given by

ha, nihb, qi = hab + nb + qa, nqi.

1. Clearly, A# is a commutative ring with unity h0, 1i. A is a subring of A# , even an ideal. Suppose A
has the ACC on ideals, prove that A# does, too. Can you make this quantitative as in Problem 78
part (3)?
34 PROBLEMS

2. If you know all the prime ideals of A, can you find all the prime ideals of A# ?

Problem 87 Let B, C be commutative A-algebras, where A is also commutative. Write D for the A-algebra
B ⊗A C.
1. Give an example to show that Spec D is not Spec B × Spec C (category of sets over Spec A).
Spec A

2. We have A-algebra maps B → D and C → D and so we get maps Spec D → Spec B and
Spec D → Spec C (even maps over Spec A), and these are maps of topological spaces (over Spec A).
Hence, we do get a map

θ : Spec D → Spec B Π Spec C (top. spaces).

Spec A

Show there are closed sets in Spec D not of the form θ−1 (Q), where Q is a closed set in the product
topology of Spec B Π Spec C.
Spec A

Problem 88 Let A = Z[T ], we are interested in Spec A.

1. If p ∈ Spec A, prove that ht(p) ≤ 2.
2. If {p} is closed in Spec A, show that ht(p) = 2. Is the converse true?
π
3. We have the map Z ,→ Z[T ] = A, hence the continuous map Spec A −→ Spec Z. Pick a prime number,
say p, of Z. Describe π −1 (p), is it closed?
4. When exactly is a p ∈ Spec A the generic point (point whose closure is everything) of π −1 (p) for some
prime number p?
5. Describe exactly those p ∈ Spec A whose image, π(p), is dense in Spec Z. What is ht(p) in these cases?
6. Is there a p ∈ Spec A so that the closure of {p} is all of Spec A? What is ht(p)?
7. For a general commutative ring, B, if p and q are elements of Spec B and if q ∈ {p} show that
ht(q) ≥ ht(p) (assuming finite height). If p, q are as just given and ht(q) = ht(p) is q necessarily p?
Prove that the following are equivalent:
(a) Spec B is irreducible (that is, it is not the union of two properly contained closed subsets)
(b) (∃ p ∈ Spec B)(closure of {p} = Spec B)
(c) (∃ unique p ∈ Spec B)(closure of {p} = Spec B)
(d) N (B) ∈ Spec B. (Here, N (B) is the nilradical of B)
8. Draw a picture of Spec Z[T ] as a kind of plane over the “line” Spec Z and exhibit in your picture all
the different kinds of p ∈ Spec Z[T ].

Problem 89 If A is a commutative ring, we can view f ∈ A as a “function” on the topological space Spec A
as follows: for each p in Spec A, as usual write κ(p) for Frac(A/p) [note that[
κ(p) = Ap /its max. ideal]
and set f (p) = image of f in A/p considered in κ(p). Thus, f : Spec A → κ(p). Observe that if
p∈Spec A
f ∈ N (A), then f (p) = 0 all p, yet f need not be zero as an element of A.
1. Let A = k[X1 , . . . , Xn ]. There are fields, Ω, containing k so that
(a) Ω has infinitely many transcendental elements independent of each other and of the Xj over k
and
PROBLEMS 35

(b) Ω is algebraically closed, i.e., all polynomials with coefficients in Ω have a root in Ω.
An example of this is when k = Q or some finite extension of Q and we take Ω = C. In any case, fix
such an Ω. Establish a set-theoretic map Ωn → Spec A so that f ∈ A = k[X1 , . . . , Xn ] viewed in the
usual way as a function on Ωn agrees with f viewed as a function on Spec A. We can topologize Ωn as
follows: Call a subset of Ωn k-closed iff there are finitely many polynomials f1 , . . . , fp from A so that
the subset is exactly the set of common zeros of f1 , . . . , fp . This gives Ωn the k-topology (an honest
topology, as one checks). Show that your map Ωn → Spec A is continuous between these topological
spaces. Prove, further, that Ωn maps onto Spec A.
2. Show that Ωn is irreducible in the k-topology. (Definition in 7(a) of Problem 88)
3. Define an equivalence relation on Ωn : ξ ∼ η ⇐⇒ each point lies in the closure (k-topological) of the
other. Prove that Ωn / ∼ is homeomorphic to Spec A under your map.

Problem 90 (Continuation of Problem 89) Let A be an integral domain and write K for Frac(A). For each
ξ ∈ K, we set

dom(ξ) = {p ∈ Spec A | ξ can be written ξ = a/b, with a, b ∈ A and b(p) 6= 0}.

1. Show dom(ξ) is open in Spec A.

2. If A = R[X, Y ]/(X 2 + Y 2 − 1), set ξ = (1 − y)/x (where x = X and y = Y ). What is dom(ξ)?
3. Set A = C[X, Y ]/(Y 2 − X 2 − X 3 ) and let ξ = y/x. What is dom(ξ)?
4. Note that as ideals of A (any commutative ring) are A-modules, we can ask if they are free

or locally
free. Check that the non-zero ideal, a, of A is free ⇐⇒ it is principal and a → (0) = (0). The
second condition is automatic in a domain. Now look again at A = R[X, Y ]/(X 2 + Y 2 − 1), you should
see easily that this is a domain. Characterize as precisely as you can the elements m ∈ Max(A) which
are free as A-modules. If there are other elements of Max(A), are these locally free? What is the
complement of Max(A) in Spec A? Prove that A ⊗R C is a PID.
5. Consider the descent question for PIDs: Given rings S and T with S → T a homomorphism, suppose
A is an S-algebra and T is faithfully flat over S. If A ⊗S T is a PID, is A necessarily a PID?
6. Do part (5) where PID is replaced by UFD.
√
Problem 91 Let p be an odd prime number, set m = 2p − 1 and write A = Z[ −m] ∼ = Z[T ]/(T 2 + m).
Assume m is square free.
√
1. Let a be the ideal (p, 1 + −m) of A. Prove that a is not principal, yet that a, as a module, is locally
free (necessarily of rank one). Prove further that A is not a UFD.
2. For p = 3 and 7, find all the ideals, a, which are not free, yet are locally free.
N.B. By results of the text you have non-free projectives here.

Problem 92 In this problem A is an integral domain and K = Frac(A).

1. Is it true that if p ∈ Spec A[X] and if p∩A = (0), then p is a principal ideal? Proof or counterexample.
2. Say A is a UFD and η ∈ K, with η 6= 0. Write η = a/b, where a and b are relatively prime. Prove that
A[η] ∼
= A[X]/(bX − a). When is A[η] a flat A-module?
3. If k is a field and ξ ∈ k(X) is a non-constant rational function, write ξ = f (X)/g(X) where f and g
are relatively prime polynomials. Of course, k(ξ)

is a subfield of k(X), so k(X) is a k(ξ) vector space
(and a k(ξ)-algebra). Prove that dimk(ξ) k(X) < ∞ and compute this dimension in terms of f and g.
36 PROBLEMS

Problem 93 If A is a commutative ring and B = A[[X1 , . . . , Xn ]] denotes the ring of formal power series
in the variables X1 , . . . , Xn (the case n = 1 was discussed in Problem 79) over A:
1. Prove:
A is noetherian ⇐⇒ B is noetherian
A is an integral domain ⇐⇒ B is an integral domain
A is a local ring ⇐⇒ B is a local ring.

2. Write K((X1 , . . . , Xn )) for Frac B, where K = Frac A and A is a domain. Say A = K = C, n = 2. Is

C((X, Y )) equal to C((X))((Y ))? If not, does one contain the other; which?

Problem 94 If A is a noetherian ring, write X = Spec A with the Zariski topology. Prove the following are
equivalent:
1. X is T1
2. X is T2
3. X is discrete
4. X is finite and T1 .

Problem 95 Call a commutative ring semi-local iff it possesses just finitely many maximal ideals.
St
1. If p1 , . . . , pt ∈ Spec A and S = A − j=1 pj , then S −1 A is semi-local.
2. Say A is semi-local and m1 , . . . , mt are its maximal ideals. Show that the natural map of rings
t
Y
A/J (A) → A/mi
i=1

is an isomorphism. (Here, J (A) is the Jacobson radical of A)

3. If A is semi-local, show Pic(A) = (0).

Problem 96 Let A be a domain. An element a ∈ A, not a unit, is called irreducible iff it is not the product
a = bc in which neither b nor c is a unit. The element a is a prime iff the principal ideal, Aa, is a prime
ideal. Of course, prime =⇒ irreducible.
1. Assume A is noetherian, show each non-unit of A is a finite product of irreducible elements. (A need
not be a domain for this.)
2. Prove that the factorization of (1) is unique (when it exists) iff every irreducible element of A is prime.
3. Say A is a UFD and S a multiplicative subset of A. Show that S −1 A is a UFD. If A is locally a UFD
is A a UFD?
4. Prove: If A is noetherian then A is a PID ⇐⇒ A is a UFD and dim A = 1.
5. Assume A is just a domain. A weight function, w, on A is a function A − {0} → Z≥0 so that
(a) a | b =⇒ w(a) ≤ w(b), with equality ⇐⇒ b | a, too
(b) If a and b ∈A and say a - b and b - a, then ∃ p, q, r ∈ A so that r = pa + qb and
w(r) < min w(a), w(b) .
Prove: A domain is a PID ⇐⇒ it possesses a weight function. Can you characterize the fields among
the PIDs by their weight functions?
PROBLEMS 37

Problem 97 Prove: A noetherian domain is a UFD iff each height 1 prime is principal.

Problem 98 Examples and Counterexamples:

1. Let A = k[X, Y ] with k a field; write m = (X, Y ). Show that q = (X, Y 2 ) is m-primary, but q is not a
power of any prime ideal of A. Therefore, primary ideals need not be powers of prime ideals.
2. Let A = k[X, Y, Z]/(XY − Z 2 ) = k[x, y, z]. Write p for the ideal (x, z) of A. Prove that p ∈ Spec A,
but p2 is not primary. Hence, powers of non-maximal prime ideals need not be primary. What is the
primary decomposition of p2 ?
√
3. Say A = k[X, Y ] as in part (1) and write a = (X 2 , XY ). Show that a is not primary yet a is
a prime ideal—which one? So, here a non-primary ideal has a prime radical. What is the primary
decomposition of a?
4. If A is a UFD and p is a prime element of A, then q = Apn is always primary. Conversely, show if q is
√
primary and q = Ap, then (∃ n ≥ 1)(q = Apn ). Compare with (3) above.

Problem 99 Assume A is a noetherian integral domain. The argument at the end of Theorem 3.56 shows
that height one primes of A are elements of Pic(A) if A is normal .
(1) Use this remark to prove that in a normal (noetherian) domain, each isolated prime of a principal
ideal has height one (special case of Krull’s principal ideal theorem).
(2) Say A is a noetherian normal domain. Show that A is a UFD iff Pic(A) = (0).

Problem 100 A Little Number Theory.

Let Q be the rational numbers, and consider fields k = Q[X]/ f (X) where f (X) is an irreducible polynomial
over Q. (Each finite extension of Q has this form, by Chapter 4, Section 4.6.) Such a k will be called a
“number field” and we write Ok for Intk (Z).
1. Show Ok is a noetherian normal domain with dim Ok = 1.
2. If p ∈ Spec Ok , then (Ok )p is a PID and Ok is a UFD iff Pic(Ok ) = (0) iff Ok is a PID.
√ √ √ √ √
3. Let k be the fields: Q(i), Q( 2), Q( 3), Q( 5), Q( −3), Q( −5), Q(ζ), where ζ is a primitive 7th
root of 1. In each case, find Ok and compute Pic(Ok ). Make a table.
√ √ √ √ √

4. In Q( −3), look at Z[ −3] = {a + b −3 | a, b ∈ Z}. Is Z[ −3] = Ok ? If not, what is Pic Z[ −3] ?
√
Same question for Z[ −5].
5. Let A be a noetherian, normal domain of dimension 1, write k = Frac A (e.g., Ok = A by (1)). We
examine submodules (for A) of k. Call one of these, M , a fractional ideal iff
(∃ b ∈ A)(b 6= 0)(bM ⊆ A). Prove that the following are equivalent for A-submodules of k:
(a) M is a fractional ideal
(b) M is a finitely generated A-module
(c) M is a rank one projective A-module.
6. Under multiplication, M N , the fractional ideals form a group, denote it I(A). (M N goes over to
M ⊗A N in Pic(A)). Let CA be the (localizing) category of finite length modules over A and write
e
K(A) for the Grothendieck group, K0 (CA ), of CA . By the theory of associated primes, each M in CA
has a composition series
M = M0 ⊃ M1 ⊃ M2 ⊃ · · · ⊃ Mn+1 = (0)
and
Mi /Mi+1 ∼
= A/pi for some pi ∈ Max(A).
38 PROBLEMS

These pi are unique up to order and we set

n
Y
χA (M ) = pi ∈ I(A).
i=0

e
Prove that χA is an isomorphism (first prove homomorphism) of the abelian groups K(A) −
g→ I(A).
e
What is the kernel of the map K(A) → Pic(A)?
7. Lastly, assume A is actually a PID. Say M = An is a free A-module of rank n and choose u ∈ EndA M .
Assume det(u) 6= 0 and show
det(u) · A = χA (coker u).

Problem 101 More examples.

1. Let A = k[X, Y, Z, W ]/(XY − ZW ), where k is a field and char(k) 6= 2. By Problem 34; A is a normal
domain. Compute Pic(A).
2. If A = C[t3 , t7 , t8 ] ⊆ C[t], compute Pic(A). If A = {f ∈ C[T ] | f 0 (0) = f 00 (0) and f (1) = f (−1)}
compute Pic(A).
3. If A = C[X, Y, Z]/(X 2 + Y 2 + Z 2 − 1), show Pic(A) 6= (0).

Problem 102
1. Write A = K[X, Y, Z], with K a field. Set a = (X, Y )(X, Z). Find a primary decomposition of a.
2. Let A = K[X, XY, Y 2 , Y 3 ] ⊆ K[X, Y ] = B, here K is a field. Write p = Y B ∩ A = (XY, Y 2 , Y 3 ).
Prove that p2 = (X 2 Y 2 , XY 3 , Y 4 , Y 5 ) and is not primary. Find a primary decomposition of p2 involving
(Y 2 , Y 3 ). All ideals are ideals of A.

Problem 103
1. Say A is an integral domain. Prove
\ \
A= Ap = Am .
p∈Spec A m∈Max(A)

2. Now let A
be a commutative ring and let f (T ) be a polynomial of degree d in A[T ]. Prove that
A[T ]/ f (T ) is an A-projective module of rank d iff the coefficient of T d in f (T ) is a unit of A.

Problem 104 Write A for the polynomial ring k[T1 , . . . , TN ] in which k is a field and B = A/p for some
prime ideal, p, of A. Let the transcendence degree of B over k be d and assume d ≥ 1. Now let S0 , S1 , . . . , Sm
be further indeterminates independent of the T1 , . . . , TN , write K for the rational function field k(S0 , . . . , Sm )
and L for k(S1 , . . . , Sm ).
(1) For a polynomial f ∈ L ⊗k A, write P for the ideal of K ⊗k A generated by p and the element f − S0
and prove that tr.d.K (K ⊗k A)/P ≤ d − 1.
(2) Assume further m ≤ N and consider the composed map

k[T1 , . . . , Tm ] ,→ A −→ B.

We assume the composed map is injective and further that the polynomial f ∈ L ⊗k A has the form
m
X
f= Sj Tj + g(Tm+1 , . . . , TN ).
j=1
PROBLEMS 39

Prove that tr.d.K (K ⊗k A)/P = d − 1.

(3) Under the hypotheses of (2), assume for each prime ideal, B, of B, the local ring, BB , is regular.
Write C = (K ⊗k A)/P, and let q be any element of Spec C. Show that Cq is regular.
(4) Revisit Problem 83 and give a quick proof.
Problem 105 Suppose k is a field (if necessary, assume ch(k) = 0) and A and C are the following n × n
matrices with entries from k:
   
a0 · ··· · an−1 0 1 0 0 ··· 0
an−1 a0 · · · · a  0 0 1 0 ··· 0
 n−2   
A=  · · ··· · ·  
 ; C = · · · · ··· ·

 · · ··· · ·  0 0 0 0 ··· 1
a1 a2 · · · an−1 a0 1 0 0 0 ··· 0
Of course, C n = I.
(1) In k find all the eigenvalues and eigenvectors of C.
(2) Find a polynomial, f (X) ∈ k[X], so that A = f (C).
(3) Compute the eigenvalues of A in k and show that the corresponding eigenvectors are those of C.
(4) Give a criterion for A to be invertible. Can you give a criterion (in the same spirit) for A to be
diagonalizable?
Problem 106 A discrete valuation, ν, on a (commutative) ring A, is a function ν : A → Z ∪ {∞} satisfying
(a) ν(xy) = ν(x) + ν(y)

(b) ν(x + y) ≥ min ν(x), ν(y) , with equality if ν(x) 6= ν(y)
(c) ν(x) = ∞ ⇐⇒ x = 0.
A pair (A, ν) where A a commutative ring and ν is a discrete valuation is called a discrete valuation ring
(DVR). Prove the following are equivalent:
(1) A is a DVR
(2) A is a local PID
(3) A is a local, noetherian, normal domain of Krull dimension 1


(4) A is a local, noetherian, normal domain and (mA → A) = {ξ ∈ Frac A | ξmA ⊆ A} 6= A. Here, mA
is the maximal ideal of A.
Problem 107 Let A be a commutative ring with unity and assume A is semi-local (it possesses just finitely
many maximal ideals). Write J for the Jacobson radical of A and give A its J -adic topology.
1. Prove that A is noetherian iff each maximal ideal of A is finitely generated and each ideal is closed in
the J -adic topology.
2. Assume A is noetherian, then the map A → Ared gives Ared its J -adic topology. If Ared is complete
prove that A is complete.

Problem 108
40 PROBLEMS

1. Let A be a local ring, give A its m-adic topology (m = mA is the maximal ieal of A) and assume A
b
is complete. Given an A-algebra, B, suppose B is finitely generated as an A-module. Prove that B
is a finite product of A-algebras each of which is a local ring. Give an example to show that some
hypothesis like completness is necessary for the conclusion to be valid.
2. (Hensel) Again A is complete and local, assume f (X) ∈ A[X] is a monic polynomial. Write f (X)
for the image of f in (A/m)[X]. If f (X) factors as g(X)h(X) where g and h are relatively prime in
(A/m)[X], show that f factors as G(X)H(X) where G(X) = g(X); H(X) = h(X). What can you say
about deg G, deg H and uniqueness of this factorization? Compare parts (1) and (2).
Problem 109 In this problem, A is an integral domain and k = Frac A. If ν and ω are two discrete
valuations of k (cf. Problem 106), the functions ν and ω are defined on A and extended to k via ν(a/b) =
ν(a) − ν(b), etc.), let’s call ν, ω inequivalent iff one is not a constant multiple of the other. Write S for a set
of inequivalent discrete valuations of k and say that A is adapted to S provided

A = x ∈ k | (∀ν ∈ S)(ν(x) ≥ 0) .
1. Prove the following are equivalent:
(a) A is a Dedekind domain
(b) (∀ ideals, a, of A)(∀x, x 6= 0, x ∈ a)(∃ y ∈ a)(a = (x, y)).
(c) There is a family of discrete valuations of k, say S, for which A is adapted to S and so that the
following holds:


(∀ν, ω ∈ S) ν 6= ω =⇒ (∃ a ∈ A)(ν(a) ≥ 1 and ω(a − 1) ≥ 1) .

2. Vis a vis part (1), describe a one-to-one correspondence S ↔ Max(A).

3. Take k = Q, consider all prime numbers p with p ≡ 1 (mod 4), write ordp(n) for the highest exponent,
e, so that pe | n. Then ordp is a discrete valuation of Q, and we set S = ordp | p ≡ 1 (mod 4) .
Illustrate (c) in part (1) above with this S. What is A, in concrete terms? It is pretty clear now how
to make many Dedekind domains.
4. Say A is a Dedekind domain and a, b are two non-zero ideals of A. Show ∃ x ∈ k(= Frac A), so that
a + xb = A.
5. Again,Tlet A be a Dedekind domain and let L be a finite subset of Max(A). Write
AL = {Ap | p 6∈ L}, then A ⊆ AL and so Gm (A) ⊆ Gm (AL ). Recall, Gm (B) is the group of units
of the ring B. Prove that Pic(A) is a torsion group ⇐⇒ Gm (AL )/Gm (A) is a free abelian group of
rank #(L) for every finite set, L, of Max(A).
Problem 110 (Suggested by A. Auel) Suppose that R is a P.I.D. and consider the functor
t : R-mod R-mod
that assigns to each M its torsion submodule. Of course, t is left-exact; what are its right derived functors?
If instead, R is just a domain but we assume the Rp t are given as in your answer for the case of a P.I.D.,
must R be a P.I.D.? Proof or counter-example.
Problem 111 Here, k is a field and A = k[Xα ]α∈I . The index set, I, may possibly be infinite. Write m
for the ideal generated by all the Xα , α ∈ I. Set Ai = A/mi+1 , so A0 = k. These Ai form a left mapping
system and we set
Ab = lim Ai
←−
and, as usual, call Ab the completion of A in the m-adic topology. Note that the kernel of A
b → Aj is the
closure of m j+1 b
in A.
PROBLEMS 41

1. Show that A b is canonically isomorphic to the ring of formal power series in the Xα in which only
finitely many monomials of each degree occur.

2. Now let I = N (the counting numbers) and write m b By adapting Cantor’s

b for the closure of m in A.
b
b is not Am. Which is bigger?
diagonal argument, prove that m

3. (Bourbaki) Again, I as in (2). Let k be a finite field, prove the

Lemma. If k is a finite field and λ > 0, (∃ nλ )(∀n ≥ nλ ), there is a homogeneous polynomial,
Fn ∈ k[n2 variables], so that deg Fn = n and Fn cannot be written as the sum of terms of degree n of
any polynomial P1 Q1 + · · · + Pλ Qλ , where Pj , Qj are in k[n2 variables] and have no constant term.
Use the lemma to prove (m) d
b 2 6= (m2 ).

b is not complete in the m-adic

4. Use (2) and (3) to prove that A b topology.

5. All the pathology exhibited in (2), (3) and (4) arises as I is not finite; indeed, when I is finite, prove:
b
b is Am;
(a) m
(b) m d
b 2 = (m2 );

b is complete in the m-adic

(c) A b topology.

Problem 112 Consider the category TOP (topological spaces and continuous maps) and T2TOP the full
subcategory of Hausdorff topological spaces.

1. At first, use the ordinary Cartesian product in TOP, with the product topology. Denote this Y × Z.
Show that Y ∈ T2TOP ⇐⇒ the diagonal map ∆ : Y → Y × Y is closed.
f
2. For X, Y ∈ T2TOP, recall that X −→ Y is called a proper map ⇐⇒ f −1 (compact) is compact. (Of
course, any map f : X → Y will be proper if X is compact.) Show that f : X → Y is proper iff
(∀T ∈ T2TOP)(fT : X × T → Y × T is a closed map.)
Y Y

3. With (1) and (2) as background, look at another subcategory, AFF, of TOP: here A is a commutative
ring, AFF consists of the topological spaces Spec B, where B is an A-algebra. Maps in AFF are those
coming from homomorphisms of A-algebras, viz: B → C gives Spec C → Spec B. Define

(Spec B) Π (Spec C) = Spec (B ⊗A C)

and prove that AFF possesses products.

NB:

(a) The topology on Spec B Π Spec C is not the product topology—it is stronger (more opens and
closeds)
(b) Spec B Π Spec C 6= Spec B × Spec C as sets.
(Cf. Problem 87)

Prove: The diagonal map ∆Y : Y → Y Π Y is closed (Y = Spec B). This recaptures (1) in the
Spec A
non-Hausdorff setting of AFF.

4. Given f : Spec C → Spec B (arising from an A-algebra map B → C) call f proper ⇐⇒

(i) C is a finitely generated B-algebra and

(ii) (∀T = Spec D)(fT : Spec C Π Spec D → Spec B Π Spec D is a closed map.)
Spec A Spec A
42 PROBLEMS


Prove: If C is integral over B, then f is proper. However, prove also, Spec B[T ] → Spec B is never
proper.
5. Say A = C. For which A-algebras, B, is the map Spec B → Spec A proper?

Problem 113 Assume A is noetherian local, mA is its maximal ideal, and

b = lim A/mn+1 = completion of A in the m-adic topology.
A
←−
n

Let B, mB be another noetherian local ring and its maximal ideal. Assume f : A → B is a ring homomor-
phism and we always assume f (mA ) ⊆ mB .
f
b −→ b (and m b → m b ).
b
1. Prove: f gives rise to a homomorphism A B A B

2. Prove: fb is an isomorphism ⇐⇒
(a) B is flat over A
(b) f (mA ) · B = mB
(c) A/mA → B/mB is an isomorphism.
b and B
3. Use (2) to give examples of B’s that are finite A-modules, non-isomorphic to A, yet A b are
isomorphic.

Problem 114 Suppose that f ∈ Z[X] is a non-constant polynomial.

(1) Show there exists an n ∈ Z so that f (n) is not a prime number.
(2) Consider the sequence {f (n)}∞
n=1 and write P for the set of primes dividing at least one term of this
sequence. Show P is infinite.

Problem 115 If k is a field and f ∈ k[T ], suppose f has degree n and has n distinct roots α1 , . . . , αn in
some extension of k. Write Ω = k(α1 , . . . , αn ) for the splitting field of f and further take n + 1 independent
indeterminates X, u1 , . . . , un over Ω. Let e e for e
k = k(u1 , . . . , un ), write Ω k(α1 , . . . , αn ) and let
e
ω = α1 u1 + · · · + αn un ∈ Ω. If σ is an arbitrary permutation of α1 , . . . , αn set

σω = σ(α1 )u1 + · · · + σ(αn )un ,

and finally set Y

h(X) = (X − σω).
σ∈Sn

1. Show that h(X) has coefficients in k[u1 , . . . , un ].

2. Split h(X) into irreducible factors in e

k[X]; show all the factors have the same degree, r. (Hint: Natural
Irrationalities). Moreover, prove if σω is a root of a given factor,
the other roots of this factor are
exactly the τ σω, with τ ∈ g(Ω/k). Hence, prove that r = # g(Ω/k) .
3. Using (2), give a procedure for explicitly determining those permutations, σ ∈ Sn , which belong to
g(Ω/k). Illustrate your procedure with the examples: k = Q, f = T 3 − 2 and
f = T 4 + T 3 + T 2 + T + 1.

Problem 116 Here k is a field and Ω is a finite normal extension of k. Prove that there exists a normal
tower of fields
k = k0 ⊂ k1 ⊂ k2 ⊂ · · · ⊂ kn = Ω
so that
PROBLEMS 43

(a) the first r of these extensions are separable and the set {g(ki /ki−1 ) | 1 ≤ i ≤ r} is exactly the set of
composition factors of g(Ω/k), and
(b) The last n − r are each purely inseparable over the previous and kj arises from kj−1 by adjunction of
a root of X p − aj , with aj ∈ kj−1 . (Here, p = char(k).)

Problem 117 Let g1 , . . . , gn be polynomials (one variable) with coefficients in k = k0 , . . . , kn−1 respectively,
and with kj the splitting field for gj . In this case, we say kn arises from the successive solution of a chain
of equations g1 = 0, g2 = 0, . . . , gn = 0. If f is a polynomial, we say f = 0 can be solved by means of an
auxiliary chain, gi = 0, of equations ⇐⇒ kn contains a splitting field for f . When the gi (X) have the
special form gi (X) = X mi − ai , we say f = 0 may be solved by radicals.
1. Suppose f = 0 may be solved by means of the auxiliary chain g1 = 0, . . . , gn = 0. Let s(G) denote the

Sconstituents
(composition factors) of a given finite group, G. Prove that
set of simple
s gk (f ) ⊆ s gkj−1 (gj ) .
2. Prove “Galois’
Theorem”: If k is a field, f ∈ k[X], and Ω is a splitting field for f over k, assume
char(k), [Ω : k] = 1; then f = 0 is solvable by radicals ⇐⇒ gk (f ) is a solvable group.

Problem 118 Here k is a field, α is a root of an irreducible polynomial, f ∈ k[X].

1. Prove: α lies in a field extension, L, of k obtained by successive solution of a chain of quadratic
equations g1 = 0, . . . , gn = 0 ⇐⇒ the degree of a splitting field for f over k is a power of 2.
2. Given a line in the plane, we conceive of the line as the real line and the plane as C. But, no numbers
are represented on the line. However, two points are indicated on the line; we take these as 0 and
1 and label them so. We are given a straight edge (no markings on it) and a pair of dividers (no
scale on it either) which we can set to any length and which will hold that length. But, if we reset
the dividers, the original setting cannot be recaptured if not marked on our plane as a pair of points
“already constructed.” We can use our implements to make any finite number of the following moves:
(a) Set the dividers to a position corresponding to two points already constructed, make any arc or
circle with the dividers where one leg is at a point already constructed. (A point is constructed
iff it is the intersection of an arc and a line, an arc and an arc, a line and a line.)
(b) Given any pair of previously constructed points use the straight edge to draw a line or segment
of a line through these points.
You should be able to see that from 0 and 1 we can construct p/q ∈ Q (all p, q) therefore it is
legitimate to label Q on our real axis. Call a point (x, y) ∈ C constructible iff its real and imaginary
parts are constructible; that is these numbers, constructed as lengths, can be obtained from Q by a
finite number of moves (a) and (b). Show that α ∈ C is constructible iff Q(α) may be obtained from
Q by the successive solution of a chain of quadratic equations.
3. Prove
(a) The duplication of a cube by straight edge and dividers is impossible.
(b) The trisection of an angle by straight edge and dividers is impossible (try π/3).
4. (Gauss) Prove that a regular n-gon is constructible by straight edge and dividers iff n = 2r p1 p2 · · · pt ,
where r is non-negative and the pj are distinct Fermat primes (cf. Problem 14).

Problem 119 What is wrong with the following argument?

Let k be a field, write f (X) ∈ k[X], deg(f ) = n, and suppose f has n distinct roots α1 , . . . , αn , in a suitable
extension field L/k. Write Ω for the normal extension k(α1 , . . . , αn ). An element, ω, of Ω has the form
ω = g(α1 , . . . , αn ), where g is a polynomial in n variables with coefficients in k. Let σ be an arbitrary
44 PROBLEMS

permutation of the αi , then σ maps g(α1 , . . . , αn ) to g(α10 , . . . , αn0 ) where αj0 = σ(αj ). If h(α1 , . . . , αn ) is
another polynomial with coefficients in k, then h(α1 , . . . , αn ) 7→ h(α10 , . . . , αn0 ) by σ and we have

g(α1 , . . . , αn ) + h(α1 , . . . , αn ) → g(α10 , . . . , αn0 ) + h(α10 , . . . , αn0 )

g(α1 , . . . , αn )h(α1 , . . . , αn ) → g(α10 , . . . , αn0 )h(α10 , . . . , αn0 ).

Thus, we have an automorphism of Ω and the elements of k remain fixed. So, the arbitrary permutation, σ,
belongs to the group of k-automorphisms of Ω; hence, the latter group has order greater than or equal to
n!. By Artin’s Theorem, [Ω : k] ≥ n!. (Theorem 4.32)

Problem 120 If k is a field, f ∈ k[X] a separable polynomial and Ω is a splitting field for f over k, write
g = g(Ω/k) and consider g as a subgroup of the permutation group on the roots of f . Show that g is a
transitive permutation group ⇐⇒ f is an irreducible polynomial over k. Use this to give a necessary
condition that σ ∈ Sn actually belongs to gk (f ), for f an arbitrary separable polynomial of degree n over
k. Illustrate your condition by finding the Galois groups over Q of the polynomials: X 5 − 1, X 5 + X + 1.

Problem 121 Here, K is a finite field of q elements and q is odd.

1. Let sq : K ∗ → K ∗ be the homomorphism given by sq(x) = x2 . Show that
# ker sq = # coker sq = 2 and # Im sq = (q − 1)/2.
2. Prove: !
n
∗ (q−1)/2 1 if x is a square in K
(∀x ∈ K ) x =
−1 otherwise

3. If K = Fp , then K contains a square root of −1 iff p ≡ 1 mod 4.

4. For any finite field, K, every element of K is a sum of squares. Is it true that each element of K is a
sum of (at most) two squares?

Problem 122 If k is a field of characteristic zero and f ∈ k[X] is a monic polynomial, factor f into monic
irreducible polynomials in k[X] and set
f = g1 g22 · · · grr
where gj is the product of the distinct irreducible factors of f which divide f with exact exponent j. Prove
that the g.c.d. of f and its, derivative, f 0 , is

g2 g32 · · · grr−1 .

Assume Euclid’s algorithm for finding the g.c.d. of two polynomials. Show that g1 , . . . , gr may be determined
constructively. If n is an integer, illustrate with

f (X) = X n − 1 ∈ Q[X].

Problem 123 If k is a field and f , g are non-constant

polynomials in k[X], with f irreducible, prove that
the degree of every irreducible factor of f g(X) in k[X] is divisible by deg f .

Problem 124 If k is a field, X is transcendental over k, and f (X) ∈ k[X] is irreducible in k[X], write
α1 , . . . , αn for a full set of roots of f in a suitable extension field of k. If char(k) = 0, prove that none of the
differences αi − αj (i 6= j) can lie in k. Give a counterexample for char(k) = p > 0 (any prime p).

Problem 125 Let k ⊆ K be two fields of characteristic zero. Assume the following two statements:
(a) Every f (X) ∈ k[X] of odd degree has a root in k
(b) (∀α ∈ k)(X 2 − α has a root in K)
PROBLEMS 45

1. Prove: Each non-constant polynomial g ∈ k[X] has a root in K.

2. Assume as well that K/k is normal of finite degree. Prove that K is algebraically closed. (Suggestion:
(r)
Use induction on ν where deg g = 2ν n0 (n0 odd). If r ∈ Z, set γij = αi + αj + rαi αj , where α1 , . . . , αn
(r)
are the roots of g in some Ω ⊇ K. Fix r, show there is a polynomial h(X) ∈ k[X], so that the γij are
(r) (r ) (r )
roots of h; for all i, j. Show some γij ∈ K; now vary r and find r1 6= r2 so that γij 1 ∈ K, γij 2 ∈ K.)
3. Take k = R and K = C. By elementary analysis, (a) and (b) hold. Deduce C is algebraically closed
(Gauss’ first proof).

Problem 126 Let Q be the rational numbers, R the real numbers , X a transcendental over R and suppose
f ∈ Q[X] is a polynomial of degree 3 irreducible in Q[X] having three real roots α, β, γ. Show that if

k0 = Q ⊆ k1 ⊆ k2 ⊆ · · · ⊆ km
√
n
is a finite chain of fields each obtained from the preceding one by adjunction of a real radical ρj = j cj
(nj ∈ Z, nj > 0, cj ∈ kj−1 ), the field km cannot contain ANY of the roots, α, β, γ of f . (Suggestion: If
wrong, show we may assume each nj is prime, let kj be the field with maximal j where f is still irreducible.
If α ∈ kj+1 show ρj+1 ∈ kj (α).) This is the famous “casus irreducibilis” of the cubic equation f = 0: if the
three roots are real, the equation cannot be solved by real radicals.

Problem 127 Here, f is an irreducible quartic polynomial with coefficients in k; assume f has four dis-
tinct roots α1 , α2 , α3 , α4 in some extension field of k. Write β = α1 α2 + α3 α4 , L = k(β), and let Ω be
k(α1 , α2 , α3 , α4 ).
1. Assume g(Ω/k) has full size, i.e., 24, find g(Ω/L).
2. Show that, in any case, β is the root of a cubic polynomial, h, with coefficients in k (Lagrange’s “cubic
resolvent” for f ).

Problem 128 Let k be a field, char(k) 6= 2, write K/k for an extension of degree 2 and L/K for an extension
also of degree 2.
1. Show ∃ α, β with α ∈ K, in fact K = k(α), and α2 = a ∈ k and β ∈ L, β 2 = u + vα; u, v ∈ k and
L = K(β). (All this is very easy).
2. Let Ω be a normal closure of k containing L. Show that [Ω : k] is 4 or 8. In the case v = 0 (part (1)),
show Ω = k(α, β) = L and that ∃ σ, τ ∈ g(Ω/k) so that σ(α) = −α, σ(β) = β, τ (α) = α, τ (β) = −β.
Determine precisely the group g(Ω/k).
3. When v 6= 0, let β1 be a conjugate, not equal to ±β, of β. Prove Ω = k(β, β1 ) and that ∃ σ ∈ g(Ω/k)
such that σ(β) = β1 and σ(β1 ) is one of β or −β.
4. Show if [Ω : k] = 8 we may assume in (3) that σ maps β1 to −β. Prove σ is an element of order 4 and
that ∃ τ ∈ g(Ω/k), of order 2, with τ −1 στ = σ −1 . Deduce that g(Ω/k) = Gp{σ, τ }; which of the two
non-abelian groups of order 8 is it?
5. Illustrate (1)-(4) with a discussion of X 4 − a over Q.
6. With the above notation, show that the normal closure of K is cyclic of degree 4 iff a can be written as
the sum of two squares, b2 + c2 , in k. (Hints: if Ω is the field above, show g(Ω/k) is cyclic, order 4, iff
Ω contains exactly one subfield of degree 2 over k. Then u2 − av 2 must equal aw2 for some w ∈ k. Now
show a is the sum of two squares. You may need to prove that if −1 is a square then every element of
k is a sum of two squares in k; cf. Problem 121.) Investigate, from the above, which primes, p ∈ Z,
are the sum of two squares in Z.
46 PROBLEMS

Problem 129 Suppose p is a prime number, let Sp denote the symmetric group on p letters and write G
for a transitive subgroup of Sp (i.e., the p letters form an orbit for G).
(1) If G contains a transposition, we know (Problem 13) that G = Sp . Use this to show there exist
extensions, K, of Q whose Galois group is Sp .
(2) Hilbert proved the following theorem:
Hilbert Irreducibility Theorem. If f ∈ Q[T1 , . . . , Tr , Z1 , . . . , Zs ], where the T ’s and Z’s are all algebraically
independent, and if f is irreducible, then there exist integers a1 , . . . , ar so that substituting aj for Tj (j =
1, . . . , r), the resulting polynomial fe ∈ Q[Z1 , . . . , Zs ] is still irreducible. (Actually, there are infinitely many
choices for the aj s.)
Use Hilbert’s theorem to exhibit Sn as a Galois group over Q.
(3) Now An is a subgroup of Sn ; can you exhibit An as a Galois group over Q? (There is an old open
question: Is every finite group, G, the Galois group of some finite normal extension of Q? If G is solvable,
this is known (due to Shafarevich) and hard to prove. Many simple groups are known to be Galois groups
over Q.)
(4) Write f (X) = X 5 + aX + 1 with a ∈ Z and let Ω be the splitting field of f over Q. Determine
G(Ω/Q).

Problem 130 (Bourbaki)

1. Say k is a field, char(k) = p > 2; let K = k(X, Y ) where X and Y are independent transcendentals
over k. Write L = K(θ), where θ is a root of

f (Z) = Z 2p + XZ p + Y ∈ K[Z].

Show that L/K is inseparable yet does not contain any purely inseparable elements over K. (Sugges-
tion: First show f is irreducible and say ∃ β ∈ L, β p ∈ K, β 6∈ K. Then prove f becomes reducible in
K(β)[Z] and that then X 1/p and Y 1/p would lie in L. Prove then that [L : K] ≥ p2 .)
2. Find the Galois group g(Ω/K) where Ω is a normal closure of L/K.
3. Now just assume char(k) 6= 2, write K = k(X) in this case.
Let σ, τ be the 2-torsion k-automorphisms
of K given by σ(X) = −X; τ (X) = 1 − X (i.e., σ f (X) = f (−X), etc.). Show the fixed field of σ is
k(X 2 ); that of τ is k(X 2 − X). If char(k) = 0, show that Gp{σ, τ } is an infinite group and prove that
k = k(X 2 ) ∩ k(X 2 − X).
4. Now assume again char(k) = p > 2. Show in this case k(X 2 ) ∩ k(X 2 − X) is strictly bigger than
k—determine it explicitly and find the degree
 
k(X) : (k(X 2 ) ∩ k(X 2 − X)) .

5. What is the situation in (3) and (4) if char(k) = 2?

Problem 131 (Various Galois groups). Determine the Galois groups of the following polynomials over the
given fields:
1. (X 2 − p1 ) · · · (X 2 − pt ) over Q, where p1 , . . . , pt are distinct prime numbers.
2. X 4 − t over R(t).
3. X p − m over Q, where p is a prime number and m is a square free integer. (Hint: Here, g fits into a
split exact sequence of groups

0 / Z/pZ / g o_ _ _/ ? / 0.)
PROBLEMS 47
√
4. X 8 − 2 over Q( 2), over Q(i), over Q. (Cf. Problem 128)

Problem 132 Show that x7 − 7x + 3 has a simple group of order 168 as its Galois group over Q. Can you
be more precise as to which group this is?

Problem 133

1. Here K/k is a finite extension of fields. Show the following are equivalent:

(a) K/k is separable

(b) K ⊗k L is a product of fields (product in the category of rings) for any field L over k
(c) K ⊗k k is a product of fields
(d) K ⊗k K is a product of fields.

2. Now assume K/k is also a normal extension, and let

Kpi = {α ∈ K | α is purely inseparable over k}.

For the map

θ : Kpi ⊗k Kpi → Kpi via θ(ξ ⊗ η) = ξη,
show that the kernel of θ is exactly the nilradical of Kpi ⊗k Kpi .

3. Prove: If K/k is a finite normal extension, then K ⊗k K is an Artin ring with exactly [K : k]s prime
ideals. The residue fields of all its localizations at these prime ideals are each the same field, K. A
necessary and sufficient condition that K/k be purely inseparable is that K ⊗k K be a local ring.
(Hints: K = Ks ⊗k Kpi and the normal basis theorem.)

Problem 134 Throughout this problem, G is a finite group, k is a field, and R = k[G]. We further assume
that (#(G), char(k)) = 1.
(1) If S is a k-algebra (not necessarily commutative) write Fcn(G, S) for the k-module of all functions
from G to S under pointwise addition and k-multiplication.
For f ∈ Fcn(G, S), we set Z
1 X
f (σ)dσ = f (σ).
G #(G)
σ∈G

Further, write fτ (σ) = f (τ σ) and show that

Z Z Z
fτ (σ)dσ = f (τ σ)dσ = f (σ)dσ
G G G

as well as Z
1dσ = 1.
G

(We can write this as d(τ σ) = dσ and refer to the above as the “left invariance of the integral”. Of course,
the integral is also right invariant as well as “inverse invariant” (i.e., d(σ −1 ) = dσ.) The integral is also
called a “mean” on G as it averages the values of the function f .
(2) If M is an R-module (i.e., a G-module which is also a k-vector space) and N is a sub-R-module of M ,
write π for any k-projection of M onto N . (So then, M = Ker π q N as k-spaces.) Now π ∈ Endk (M ) (= S),
so we can form Z
T = (σ −1 πσ)dσ.
G
48 PROBLEMS

Prove that T is a G-invariant projection from M onto N and that

M = Ker T q N, as R-modules.

Deduce
Maschke’s Theorem (1898) If G, R and k are as above with (#(G), char(k)) = 1, then R is semi-simple
as k-algebra.
(3) If M is a simple R-module, prove that M is finite-dimensional as a k-vector space. (R-modules
are called (linear) representation spaces for G and the map G −→ Aut(M ), making M a G-module, is
called a representation of G with space M . The dimension of M (as k-space) is called the degree of the
representation.) It is a known theorem of Wederburn that a simple k-algebra with the D.C.C. (on left ideals)
is isomorphic (as k-algebra) to the r × r matrices over a division ring, D. If k is algebraically closed, prove
that D is k itself. Now prove that
(a) For each finite group, G, and algebraically closed field, k, with (#(G), char(k)) = 1, the number of
non-isomorphic simple k[G]-modules is finite,
and
(b) We have g = f12 + · · · + ft2 , where fj is the degree of the j th simple R-module and g = #(G).

Problem 135 Say R is a not necessarily commutative ring but that R is noetherian (on the left).
(1) Given a f.g. R-module, M , show that projdimR (M ) ≤ d if and only if for all finitely generated
R-modules, N , we have
Extd+1
R (M, N ) = (0).

(2) Does the same criterion work for non f.g. R-modules M ?

Problem 136 (Yoneda) Here, R is a ring and M 0 , M 00 are R-modules.

(1) Consider exact sequences of the form

0 −→ M 0 −→ X1 −→ X2 −→ M 00 −→ 0 (E2 )

where the Xi are R-modules. Call such “2-fold extensions of M 00 by M 0 ” and, on the model of ordinary
extensions, define an equivalence relation on the 2-fold extensions. Prove that the equivalence classes so
defined are in 1-1 correspondence with Ext2R (M 00 , M 0 ).
(2) Generalize part (1) to “n-fold extensions”:

0 −→ M 0 −→ X1 −→ · · · −→ Xn −→ M 00 −→ 0 (En )

including the 1-1 correspondence of the equivalence classes with ExtnR (M 00 , M 0 ).

f0 and if
(3) We know ExtnR (A, B) is a co-functor in A and a functor in the variable B. If M 0 −→ M
ξ ∈ ExtnR (M 00 , M 0 ) is represented by

0 −→ M 0 −→ X1 −→ · · · −→ Xn −→ M 00 −→ 0,
f0 ). Same question but for
describe explicitly an n-fold extension representing the image of ξ in ExtnR (M 00 , M
a morphism M 00 −→ M f00 and an element ξe ∈ ExtR (M
n f00 , M 0 ).

(4) ExtnR (−, −) is an abelian group, as we know. Start with n = 1 and describe, in terms of representing
extensions,
0 −→ M 0 −→ X −→ M 00 −→ 0,
PROBLEMS 49

the abelian group structure on ExtnR (M 00 , M 0 ). (Of course, you must show your explicit construction of the
equivalence class of a sum of two extensions

0 −→ M 0 −→ X −→ M 00 −→ 0 (a)
0 00
0 −→ M −→ Y −→ M −→ 0 (b)

is independent of the choice of the representatives (a) and (b).) Continue with the general case of n-fold
extensions.
(5) Say
0 −→ M 0 −→ X1 −→ · · · −→ Xr −→ Z −→ 0
and
0 −→ Z −→ Y1 −→ · · · −→ Ys −→ M 00 −→ 0
are an r-fold (resp. s-fold) extension of Z by M 0 (resp. of M 00 by Z). We can splice these to obtain an
r + s-fold extension of M 00 by M 0 :

0 −→ M 0 −→ X1 −→ · · · −→ Xr −→ Y1 −→ · · · −→ Ys −→ M 00 −→ 0.

Prove that this process respects the equivalence relation on extensions and therefore yields a map
Y
θ : ExtsR (M 00 , Z) ExtrR (Z, M 0 ) −→ Extr+s 00 0
R (M , M ).

Show that from an r-fold extension

0 −→ M 0 −→ X1 −→ · · · −→ Xr −→ Z −→ 0 (Er )

we obtain an “iterated connecting homomorphism”

δr : HomR (M 0 , A) −→ ExtrR (Z, A)

for any R-module, A. If we take A = M 0 and compute δr (idM 0 ), we get an element χ(Er ) in ExtrR (Z, M 0 ).
Prove that χ(Er ) depends only on the equivalence class of Er and gives the 1-1 correspondence of part (2).
Discuss the pairing θ in terms of these “characteristic classes”, χ(Er ), of extensions.
(6) Show that θ is actually bi-additive, hence it is Z-bilinear and therefore we get a map

ExtsR (M 00 , Z) ⊗Z ExtrR (Z, M 0 ) −→ Extr+s 00 0

R (M , M ).

Take M = Z = M 00 , call the common value M . Then we can compute θ(α, β) and θ(β, α) for
α ∈ ExtrR (M, M ) and β ∈ ExtsR (M, M ). Is θ commutative? Is θ graded commutative
(θ(α, β) = (−1)rs θ(β, α))? Neither?

Problem 137 We take G to be a group and write R for Z[G].

(1) Recall from Chapter 5, Section 5.3, that there is an isomorphism

H p (G, M ) ∼ p
= ExtR (Z, M )

for every p ≥ 0. Here, M is a G-module (so, an R-module). When p = 2, the left hand group classifies group
extensions
0 −→ M −→ G −→ G −→ 1 (E)
up to equivalence, while the right hand side classifies 2-extensions (of R-modules)

0 −→ M −→ X1 −→ X2 −→ Z −→ 0, (E)
50 PROBLEMS

again up to equivalence.
In terms of exact sequences and natural operations with them describe the 1-1 correspondence between
sequences (E) and (E).
(2) Again, with the G-action on M fixed, extensions (E) can be classified by equivalence classes of 2-
cocycles of G with values in M . Given such a 2-cocycle, show how to construct, explicitly, a 2-extension (E).
Carry through the verification that cohomologous 2-cocycles yield equivalent 2-extensions.
(3) Transfer the Yoneda addition of 2-extensions from Problem 136 to the addition of group extensions—
the so called Baer addition.

Problem 138


1. Let A = k[X1 , . . . , Xn ]/ f (X1 , . . . , Xn ) , where k is a field. Assume, for each maximal ideal, p, of
A, we have (grad f )(p) 6= 0 (i.e., (∀ p)(∃ component of grad f not in p)). Show that Derk (A, A) is a
projective A-module.
2. Suppose now A = k[X, Y ]/(Y 2 − X 3 ), char(k) 6= 2, 6= 3. Consider the linear map A q A → A given by
the matrix (X 2 , Y ); find generators for the kernel of this map.
3. In the situation of (2), show that Derk (A, A) is not projective over A.

Problem 139 Suppose in a ring R (assumed commutative for simplicity) we have elements f1 , . . . , fr . We
−
→
let f = (f1 , . . . , fr ); prove that
−
→ −→ −→
K• ( f ) ∼
= K• ( f1 ) ⊗R · · · ⊗R K• ( fr ),

where on the right hand side we mean the total complex.

Problem 140 For G a group and M a right G-module, let M be considered as a “trivial” (left) Z[G]-module
and consider the bar complex as in Section 5.3, Chapter 5 of the text with boundary map
n−1
X
∂n (m ⊗ σ1 ⊗ · · · ⊗ σn ) = mσ1 ⊗ σ2 ⊗ · · · ⊗ σn + (−1)i m ⊗ σ1 ⊗ · · · ⊗ σi σi+1 ⊗ · · · ⊗ σn
i=1
+ (−1)n+1 m ⊗ σ1 ⊗ · · · ⊗ σn−1 .

Define
e n (G, M ) = Ker ∂n /Im ∂n+1
H
and prove that M e • (G, M )} is a universal ∂-functor as stated in the text. Thus, complete, by elemen-
{H
tary methods, the identification of group homology for (right) G-modules, M , and Hochschild homology for
the ring Z[G] and the modules ∗ M (definition on page 339, top).

Problem 141 Suppose that G is a profinite group and that H is a closed subgroup of G.
(1) Show that c.d(H) ≤ c.d(G).
(2) If H is open in G (and hence automatically closed in G), can you strengthen the inequality of (1)?
(3) Suppose G is a finite group. Prove that
n
0
c.d(G) =
∞
and c.d(G) = 0 when and only when G = {1}.
PROBLEMS 51

−→
Problem 142 For simplicity, assume in this problem that A is a commutative ring. If f = (f1 , . . . , fr )
−
→
and −
→
g = (g1 , . . . , gr ) are two ordered sequences of elements of A, write f g for the sequence (f1 g1 , . . . , fr gr ).
Now, we have a map
−
→ −
→
→
ϕ−g : K• (f g) −→ K• ( f )

induced by
→
g (ξ1 , . . . , ξr ) = (g1 ξ1 , . . . , gr ξr ).
ϕ−

(1) Show that this map is a chain map.

−
→
(2) Write f p = (f1p , . . . , frp ), then, for 0 < s < t, we get a map
−→ −→
ϕ−−→ : K• ( f t ) −→ K• ( f s )
t−s
f

and hence
• −→s −→t
ϕ•−− •
→ (M ) : K ( f , M ) −→ K ( f , M ).
t−s
f

We set
−
→ −→
C • (( f ), M ) = lim K • ( f t , M )
−→
(with respect to these maps) and further set
−
→ −
→
H • (( f ), M ) = H • (C • (( f ), M )).

Prove that
−
→ −→
H • (( f ), M ) = lim H • ( f t , M ).
−→
−
→
(3) Now, fix f and for the given −
→
g , define
−
→ −
→
Eg : K• ( f ) −→ K• ( f )

by the equation  
r
X
(Eg )• (z) =  gj ej  ∧ z; the ej are a base for Ar .
j=1

Prove that !
r
X −
→
d ◦ Eg + Eg ◦ d = gi fi id on Kt ( f ), all t ≥ 0.
i=1

Deduce the
Proposition Suppose f1 , . . . , fr generate the unit ideal of A, then for all A-modules, M , the complexes
−→ −→ −→ −→
K• ( f t ); K• ( f t , M ); K • ( f t , M ); C • (( f t ), M )

have trivial (co)homology in all dimensions.

−
→ −→ −
→ −→
(4) The homology and cohomology modules H0 ( f , M ), Hr ( f , M ), H 0 ( f , M ), H r ( f , M ) depend only
−
→
on the ideal, A, generated by f1 , . . . , fr . Is it true that H • (( f ), M ) depends only on A as (3) suggests?
52 PROBLEMS

Problem 143 Give the proof of “Lemma C” (= Lemma 5.51 of the text) following the methods used for
“Lemmas A & B”.
Problem 144 If A is a P.I.D., prove that gldim(A) ≤ 1. Under what conditions does the strict inequality
hold? You may wish to investigate first the relations between gldim(A) and gldim(Ap ) for a commutative
(noetherian?) ring, A, and all its prime ideals, p. Is the inequality gldim(A) ≤ 1 still valid if A is just a
principal ideal ring (not a domain)? If A is a Dedekind ring, what is gldim(A)?
Problem 145
1. Prove the six conditions of Proposition 5.70 are indeed equivalent.
2. Prove that the ten conditions listed in Proposition 5.71 are equivalent.
Problem 146 Here, A is a commutative ring, A is an ideal of A and M is an A-module.
1. Prove that the number of elements in a maximal M -regular sequence from A is independent of the
choice of these elements (from A). Thus, depthA M is well-defined.
2. Reformulate Koszul’s Proposition (our 5.66) in terms of A-depth.
−
→
3. If A and M are graded and (f1 , . . . , ft ) = f is an M -regular sequence of homogeneous elements then
any permutation of (f1 , . . . , ft ) is still an M -regular sequence.
Problem 147 (R. Brauer) Here, G is a group and T is a finite subgroup of order m. For σ, τ ∈ G, we define
σ ∼ τ ⇐⇒ (∃t ∈ T )(σ −i tτ i ∈ T, all i ∈ Z).
1. Show that ∼ is an equivalence relation and that each equivalence class has m elements.
2. Say σ ∼ τ , prove there is an x ∈ T so that τ m = x−1 σ m x.
3. Let S be a subset of Z(G); pick a suitable T as above and show: Given n ∈ Z, either
#({z ∈ G | z n ∈ S}) = ∞
or this cardinality is divisible by g.c.d(n, m).
4. When #(G) = g < ∞, show that the cardinality of the set in (3) is divisible by g.c.d(g, n).
Problem 148 If F (r) is the free group of rank r, and if Γn (F (r)) is the nth term in the lower central series
for F (r), prove that the group G = F (r)/Γn (F (r)) is torsion-free.
S∞
Problem 149 Suppose A is a commutative ring, write GL(A) for the group n=1 GL(n, A) in which
GL(n, A) is a subgroup of GL(n + 1, A) by the map
 
ξ 0
ξ 7→
0 1
1. When A = Z, consider elements of GL(n + 1, Z) of the form
 
∗ 
 .. 
 I .  n
 
 ∗ 
0 ··· 0 ∗
| {z }
n
and their transposes. Show these matrices generate GL(n + 1, Z) (as a group).
PROBLEMS 53

2. Prove that for any α ∈ GL(n, A), there exist elements x, β ∈ GL(A) with β of the form
!
I 0 n
β =
0 ∗ r
|{z} |{z}
n r

and α = xβx−1 .

Problem 150 Let k be a field, ch(k) 6= 2 and write F for any overfield of k. Denote by Vn (F ) the set of all
symmetric, nilpotent n × n matrices, A, with entries in F and rank(A) = n − 1.
1. In the ring of all n×n matrices over F , show that if a matrix commutes with A it must be a polynomial
(coefficients in F ) in A.
2. When n = 2 and F = Fp , prove that V2 (F ) is non-empty when and only when p ≡ 1 (mod 4).
3. If n = 3 and p ≡ 1 (mod 4) then V3 (Fp ) 6= ∅. Show, moreover, that V3 (F3 ) 6= ∅.
4. Let Zp denote the ring of p-adic integers with p 6= 2. Prove there is an n × n symmetric matrix, B,
with entries in Zp so that B n = pC iff Vn (Fp ) 6= ∅. (Here, C is an invertible n × n matrix with entries
in Zp .)
5. As usual, write F for the algebraic closure of F and On (F ) for the group of orthogonal matrices for
the standard diagonal form (entries in F ). If D ∈ GL(n, F ), write Cay(D) = D> D (this is the Cayley
transform of D) and show the map
D 7→ Cay(D)
is an isomorphism of the coset space On (F )\GL(n, F ) with the set, Sn (F ), consisting of symmetric,
invertible n × n matrices from F . Is this true when F replaces F ?
6. Write N for the nilpotent matrix (n × n)
 
0 0 0 ··· 0 0
1 0 0 ··· 0 0
 
 ··· 0
N = 0 1 0 0 .
 .. .. .. .. .. 
. . . . .
0 0 0 ··· 1 0

If S is a symmetric n × n matrix prove that SN = N > S iff S has the form

 
sn sn−1 · · · s2 s1
sn−1 sn−2 · · · s1 0 
 
 .. 
S =  ... ..
.
..
. .
 
 s2 s1 ··· 0 0
s1 0 ··· 0 0

and show further that S is invertible iff s1 is a unit.

7. Say p 6= 2, prove that Vn (Fp ) 6= ∅. Using only (5) and (6) above, determine how big an extension, K,
of Fp you need to guarantee Vn (K) 6= ∅.

Problem 151 (Continuation of Problem 150) Here, ch(F ) 6= 2.

1. Prove On (F ) acts transitively on Vn (F ).
54 PROBLEMS

2. Show Vn (F ) is a principal homogeneous space (= a torsor ) for the group P On (F ), which, by definition,
is On (F )/(±I).
3. If n is odd, show Vn (F ) is a torsor for SOn (F ); while if n is even, prove Vn (F ) has two components.
Problem 152 (Sierpinski) Write π(x) for the nunmber of prime integers. lessthan or equal to the positive
x
real number x. The Prime Number Theorem asserts that limx7→∞ π(x) log x = 1. Call a rational number
p
special if it has the form q where p and q are prime integers. Prove that the special rational numbers are
dense in the positive reals.
Problem 153 Suppose (Bα , ϕβα ) is a right mapping system of Artinian rings. Write B for lim Bα , and
−→
assume B is noetherian. Prove that B is Artinian. That is, B is Artinian iff it is noetherian.
Problem 154 Fix a commutative ring, R, and an R-module, E. Suppose A and B are submodules of E so
that B is free (of rank r) and is a direct summand of E. Prove that for an integer q ≥ 0, the following are
equivalent:
Vq Vq
(a) The map A −→ (E/B) is zero.
Vq Vq
(b) The map ((A + B)/B) −→ (E/B) is zero.
Vq+r Vq+r
(c) The map (A + B) −→ E is zero.
Problem 155 Throughout this problem A, B, C are three subgroups of a group, G, and we assume
AB = BA, AC = CA and C ⊆ B.
1. Prove that (B : C) = (AB : AC)/(A ∩ B : A ∩ C).
2. Suppose ϕ maps B onto a group B ∗ and write C ∗ for the image of C under ϕ. Prove that
(B : C) = (B ∗ : C ∗ )(Ker ϕ : Ker (ϕ  C)).

3. Here, let ϕ and ψ be in End(G); assume ϕψ and ψϕ are each the trivial homomorphism. Let H be
any subgroup of G stable under both ϕ and ψ. Show that
(G : H)(Ker (ϕ  H) : Im(ψ  H)) = (ϕ(G) : ϕ(H))(ψ(G) : ψ(H))(Ker ϕ : Im ψ).

4. Under the hypotheses of (3), if (G : H) < ∞, deduce Herbrand’s Lemma:

(Ker ϕ : Im ψ)(Ker (ψ  H) : ϕ(H)) = (Ker ψ : Im ϕ)(Ker (ϕ  H) : ψ(H)).

Problem 156 Suppose A is a (commutative) local or semi-local ring. Recall that the (strict) Henselization
of A, denoted Ah , is the right limit, lim C, in which C runs over the family of finitely presented étale
−→
A-algebras.
1. If B is a semi-local A-algebra (A also being semi-local) and if B is integral over A, prove that B ⊗A Ah
is both semi-local and isomorphic to B h .
2. Suppose A is local and Henselian (i.e. A = Ah ), show that for every p ∈ Spec A the integral closure of
A/p in Frac(A/p) is again a local ring.
Problem 157 (Eilenberg) Let R be the non-commutative polynomial ring in n variables, T1 , . . . , Tn , over
the field k; so, R = khT1 , . . . , Tn i. If M is a two-sided R-module, then a crossed homomorphism from R to
M is an R-module map R −→ M so that
f (ξη) = ξf (η) + f (ξ)η.
(Also called a derivation).
PROBLEMS 55

1. Given elements m1 , . . . , mn from M , show that the assigment Tj 7→ mj gives rise to a unique crossed
homomorphism R −→ M . Here, there is no restriction on the mj .
2. As in Section 5.3 of the text, consider the augmentation ideal, J, for the map ∂0 : Re → R. Prove that
J is a free Re -module on the base Tj ⊗ 1 − 1 ⊗ Tjop , j = 1, 2, . . . , n.
3. Deduce from (2) that dimRe (R) = 1 (n > 0) in contradistinction to the commutative case.

Problem 158 (Serre) Here, G is a group and it acts on a set, S.

1. Suppose G is finite and S is finite. Write χ for the function on G to C given by

χ(σ) = # of fixed points of σ on S.

R
Prove Burnside’s Lemma: The number of orbits of G acting on S equals χ(σ)dσ (cf. Problem 134
for notation). (Suggestions. Show it suffices to give the proof when S is an orbit. In this case write
Z Z X  XZ
χ(σ)dσ = 1 dσ = 1dσ,
x∈S σ s∈S Gx

where Gx = {σ ∈ G | σx = x}.)
Q 2
Q part (1)Rto 2the set S S with its G-action to see that χ (σ) counts the fixed points of σ on
2. Apply
S S. Prove: χ (σ)dσ ≥ 2.
3. Write G0 = {σ ∈ G | χ(σ) = 0} = the σ’s of G having no fixed points. Set n = #(S) and prove
Z
(χ(σ) − 1)(χ(σ) − n)dσ ≤ 0.
G−G0

Next assume n ≥ 2 and G acts transitively on S. Prove that

Z
(χ(σ) − 1)(χ(σ) − n)dσ ≥ 1
G
R
and evaluate G0
(χ(σ) − 1)(χ(σ) − n)dσ. Put all together to prove the

Cameron-Cohen Inequality: If n ≥ 2 and S is a G-orbit then

#(G0 ) 1
≥ .
#(G) n

Deduce Jordan’s Theorem: If G acts on S transitively and #(S) ≥ 2, then there is a σ ∈ G having no
fixed point on S.

Problem 159 (Kaplansky) R is a ring and we are interested in “big” R-modules, i.e., those generated by
more than ℵ0 generators. For this reason, modules finitely or countably generated will be called “atoms”
and we use the locution “finite atom” for a f.g. module.
`
1. Suppose M is an R-module that is a coproduct of (an arbitrary number of) atoms, say M = Mi .
Suppose further P is a direct summand of M ; that is,

M =P qQ (some Q)

Prove there exists a well-ordered increasing family {Sα }α an ordinal of submodules of M having the
following properties:
56 PROBLEMS

(a) Each Sα is a coproduct of some of the Mi

`
(b) Each Sα splits as (Sα ∩ P ) (Sα ∩ Q)
S
(c) If α is a limit ordinal, then Sα = β<α Sβ
(d) Sα+1 /Sα is an atom.
(Hints: We use transfinite induction. By (c) we know how to proceed at a limit ordinal, check properties
(a) and (b). The only point is to construct Sα+1 from Sα . One of the Mi is not contained in Sα , call
it M ∗ . Write the generators of the atom M ∗ as

x11 x12 x13 x14 · · ·

Begin with x11 and split it into its P and Q components giving us two new elements of M . Show only
finitely
` many Mi ’s appear in the coproduct decomposition of these new elements; so, if we take
{Mi | Mi appears} we get an atom. Write its generators as a second row of the infinite matrix being
constructed. Repeat for x12 and get the third row x31 x32 · · · . Now just as in the counting of Q take
the elements in “diagonal order”: x11 , x12 , x21 , x13 , x22 , x31 , · · · and keep repeating. Show that

Sα+1 = module generated by Sα and all xij

has (a) and (b) ((d) is obvious).)

S
2. Write Pα = P ∩ Sα , show Pα is a direct summand of Pα+1 , that Pα = β<α Pβ (when α is a limit
ordinal) and that Pα+1 /Pα is an atom. Finally, deduce P is a coproduct of atoms and so prove

Kaplansky’s Theorem. Every direct summand of a module which is a coproduct of atoms is itself a
coproduct of atoms. Every projective R-module is a coproduct of atoms.
Chapter 1

Group Theory

1.1 Introduction
Groups are probably the most useful of the structures of algebra; they appear throughout mathematics,
physics1 and chemistry. They almost always occur as “groups of transformations” and that is the way we
will use them at first. This allows of tremendous freedom, constrained only by the imagination in finding
objects on which to let groups act, or, what is the same, in finding homomorphisms from the group to the
“automorphisms” of some object or structure. Then we will look into groups qua groups, and here there
is a sharp distinction between the finite case and the infinite case. In the finite case, there is a subtle
interplay (not yet fully understood) between the order of a group and its structure, whereas in the infinite
case“geometric” arguments and applications are more the norm.

1.2 Group Actions and First Applications; The Three Sylow The-
orems
We begin by reviewing the notion of group action.

Definition 1.1 Let G be a group and S be a set. We say that G acts on S (on the left) (or that there is a
(left) G-action on S) iff there is a map
Y
G S −→ S
(σ, s) 7→ σ·s

called the action, satisfying the two rules:

(1) (∀s ∈ S)(1 · s = s)

(2) (∀σ, τ ∈ G)(∀s ∈ S)(σ · (τ · s) = (στ ) · s).

Remarks:

(1) For every σ ∈ G, the map s 7→ σ · s is a bijection of S to itself. Its inverse is the map s 7→ σ −1 · s. We
let Aut(S) denote the set of all set theoretic bijections of S.
1 The word group even occurs in Einstein’s first paper [12] on special relativity; it is the only place to my knowledge where

that word appears in Einstein’s corpus of scientific work.

57
58 CHAPTER 1. GROUP THEORY

(2) Write θ(σ) for the element of Aut(S) given by remark (1), i.e.,

θ(σ)(s) = σ · s.

Then, the map θ : G → Aut(S) is a homomorphism of groups (where Aut(S) is a group under compo-
sition).

(3) Conversely, a n.a.s.c. that G act on S is that there is a homomorphism θ : G → Aut(S). (The action
gives θ by remarks (1) and (2), and given θ, define the corresponding action by σ · s = θ(σ)(s). Check
that this is an action (DX).)

Say G acts on S, and for any given s consider

St(s) = {σ ∈ G | σ · s = s},

the stabilizer of s. It is always a subgroup of G. The set

{t ∈ S | (∃σ ∈ G)(σ · s = t)}

is the orbit of s under the action, and it is denoted OG (s).

(4) There is a one-to-one correspondence between the elements of the orbit of s and the left cosets of St(s)
in G. Namely, if H = St(s), there are maps

σH 7→ σ · s
σ · s 7→ σH,

for any left coset, σH. The first map is well-defined because if σH = τ H, then τ = σh for some h ∈ H,
and
τ · s = (σh) · s = σ · (h · s) = σ · s
as h ∈ St(s). The reader should check that the second map is well-defined (DX).

If G is finite or (G : St(s)) is finite (here, (G : H) denotes the index of the subgroup H in G, i.e., the
number of (left) cosets of H in G), then OG (s) is a finite set and when G is finite, #(OG (s)) divides
#(G).

(5) Say t ∈ OG (s) and H = St(s). Write t = σ · s. What is St(t)?

We have τ ∈ St(t) iff τ · t = t iff τ · (σ · s) = σ · s iff (σ −1 τ σ) · s = s iff σ −1 τ σ ∈ H iff τ ∈ σHσ −1 . In

conclusion, we see that St(σ · s) = σSt(s)σ −1 , a conjugate subgroup of St(s).

(6) The reader can check that the relation ∼ on the set S defined by

s ∼ t iff t = σ · s for some σ ∈ G

is an equivalence relation on S, and that the equivalence classes of this relation are exactly the distinct
orbits OG (s). Thus, given two orbits, OG (s) and OG (t), either OG (s) ∩ OG (t) = ∅ or OG (s) = OG (t).
As a conclusion, [
S= · OG (s).
distinct orbits

The orbit space, G \ S, is the quotient set S/ ∼, i.e., the collection of orbits, each considered as a
distinct entity.
1.2. GROUP ACTIONS AND FIRST APPLICATIONS; THE THREE SYLOW THEOREMS 59

Q
Obviously, we can define the notion of right action using a map S G −→ G. It is obvious how to modify
conditions (1) and (2) in Definition 1.1.
We now give some examples of group actions.

Example 1.1

(1) Trivial action. Let G be any group and S be any set. The action is

σ · s = s,

that is, it leaves every element of S fixed.

(2) LetQG be a group and H be a subgroup of G. Consider G as a set, H as a group, and the action
H G −→ G given by
(τ, s) 7→ τ · s = τ s ∈ G.
This action is called translation. Observe that

St(s) = {τ ∈ H | τ s = s} = {1},

and

OH (s) = {t ∈ G | (∃σ ∈ H)(σ · s = t)}

= {t ∈ G | (∃σ ∈ H)(σs = t)}
= Hs = a right coset of s.

(3) LetQG be a group and H be a subgroup of G. Consider G as a set, H as a group, and the action
H G −→ G given by
(τ, s) 7→ τ · s = τ sτ −1 ∈ G.
This action is called conjugation. Note that

St(s) = {τ ∈ H | τ sτ −1 = s}
= {τ ∈ H | τ s = sτ },

the collection of τ ’s in H which commute with s. When H = G, we see that St(s) is the centralizer of
s in G, denoted ZG (s). For an arbitrary subgroup H of G, we get St(s) = ZG (s) ∩ H. We also have

OH (s) = {t ∈ G | (∃σ ∈ H)(σsσ −1 = t)},

the H-conjugacy class of s, denoted ClH (s). When H = G, we get the conjugacy class of s, denoted
Cl(s).

(4) Suppose the set S has some structure. Two very important special cases are:

(a) The set S is a vector space over a field. Then, we require θ : G → Aut(S) to land in the linear
automorphisms of S, i.e., in the invertible linear maps. In this case, our action is called a (linear)
representation of G.
(b) The set S is an abelian group under addition, +. Then, we require θ : G → Aut(S) to land in
the group of group automorphisms of S. Our action makes S into a G-module. Observe that in
addition to the axioms (1) and (2) of Definition 1.1, a G-module action also satisfies the axiom

σ · (a + b) = (σ · a) + (σ · b), for all σ ∈ G and all a, b ∈ S.

60 CHAPTER 1. GROUP THEORY

Now, assume that G is finite. Observe that the converse of Lagrange’s theorem is false; namely, if G has
order n and h divides n, then there isn’t necessarily a subgroup of order h. Indeed, the group, A4 , of even
permutations on four elements, has order 12 and 6 | 12, yet A4 has no subgroup of order 6. In 1872, Sylow
(pronounce “Zŏloff”) discovered the Sylow existence theorem and the classification theorem, known now as
Sylow theorems I & II.

Theorem 1.1 (Sylow, I) If G is a finite group of order g and p is a given prime number, then whenever
pα | g (with α ≥ 0), there exists a subgroup, H, of G of exact order pα .

To prove Theorem 1.1, we need an easy counting lemma. If m is an integer, write ordp (m) for the
maximal exponent to which p divides m (i.e., ordp (m) = β for the largest β such that pβ | m). The following
simple properties hold (DX):
(1) ordp (mn) = ordp (m) + ordp (n).
(2) ordp (m ± n) ≥ min{ordp (m), ordp (n)},
with equality if ordp (m) = ordp (n).
(3) By convention, ordp (0) = ∞.

Lemma 1.2 (Counting lemma) Let p be a prime, α, m positive integers. Then,

 α 
p m
ordp = ordp (m).
pα

Proof . We know that  α 

p m pα m(pα m − 1) · · · (pα m − (pα − 1))
= .
pα pα (pα − 1) · · · 2 · 1
Observe that for 0 < i < pα , we have (DX)

ordp (pα m − i) = ordp (pα − i).

Thus,  
pα m
= mK, where K is prime to p.
pα
Therefore,  
pα m
ordp = ordp (m),
pα
as contended.
Proof of Sylow I . (Wielandt, 1959) If S is any subset of G, let

σ · S = {σt | t ∈ S},

and note that σ · S is a subset of the same cardinality of that of S. Let

S = {S ⊆ G | #(S) = pα }.

Note that in the above definition, S is any subset of G, and not necessarily a subgroup of G. Of course,
 α 
p m
#(S) = .
pα

The group G acts on S by translation, i.e., via, S 7→ σ · S.

1.2. GROUP ACTIONS AND FIRST APPLICATIONS; THE THREE SYLOW THEOREMS 61

Claim. There is some S ∈ S so that

ordp (#(OG (S))) ≤ ordp (m).

If not, then for all S ∈ S, we have ordp (#(OG (S))) > ordp (m). But we know that S can be written as a
disjoint union of G-orbits, [
S= · OG (S).
distinct orbits

So, X
#(S) = #(OG (S)).
distinct orbits

Consequently,
ordp (#(S)) ≥ min{ordp (#(OG (S)))} > ordp (m).
But  α 
p m
ordp (#(S)) = ordp ,
pα
contradicting Lemma 1.2. This proves the claim.
Now, pick some S ∈ S so that ordp (#(OG (S))) ≤ ordp (m). Let H be the stabilizer of S. We know that
(a) #(OG (S)) = (G : St(S)) = (G : H).
(b) pα m = #(G) = #(H)#(OG (S)).
From (b), applying the ord function, we get

α + ordp (m) = ordp (#(H)) + ordp (#(OG (S))) ≤ ordp (#(H)) + ordp (m).

So, α ≤ ordp (#(H)) and then, pα divides #(H), and thus, #(H) ≥ pα . Now, H takes S elementwise to
itself by translation, and for every s ∈ S,

St(s) = {σ ∈ H | σs = s} = {1}.

Therefore, #(H) = #(OH (s)) for every s ∈ S, and yet every orbit is contained in S. Thus,

#(OH (s)) ≤ #(S) = pα ,

from which we deduce that #(H) ≤ pα . We conclude that #(H) = pα , and H is the required subgroup.

Corollary 1.3 (Original Sylow I) If pβ is the maximal power of p to divide #(G) and p is a prime number,
then G possesses a subgroup of order pβ .

The subgroups of maximal p-power order arising in Corollary 1.3 are called the p-Sylow subgroups of G
(there can be more than one).

Corollary 1.4 (Cauchy, 1840) Say G is a finite group and p | #(G), where p is a prime number. Then,
there is some σ of order p in G.

Nomenclature: A p-group is a finite group whose order is a power of the prime number p.

Corollary 1.5 Say G is a p-group, with #(G) = pr . Then G possesses a descending chain

G = G0 > G1 · · · > Gr−1 > Gr = {1},

so that (Gi : Gi+1 ) = p for all i with 0 ≤ i ≤ r − 1. Hence, #(Gi ) = pr−i .

62 CHAPTER 1. GROUP THEORY

Proof . By Sylow I, a subgroup G1 of order pr−1 exists. An induction finishes the proof.

Remark: It is not clear that Gi+1 is normal in Gi . In fact, this is true, but it takes more work (see
Proposition 1.10).
To prove Sylow II, we need the local embedding lemma. In order to state this lemma, we need to recall
the concept of a normalizer. If S denotes the collection of all subsets of G, then G acts on S by conjugation:
S 7→ σSσ −1 . This action preserves cardinality. For every S ∈ S, we have

St(S) = {σ ∈ G | σSσ −1 = S}.

The group St(S) is called the normalizer of S in G, and it is denoted NG (S). If S is a subgroup of G, then
S is normal in NG (S) (denoted S C NG (S)), and NG (S) is the biggest subgroup in which S is normal (DX).
The “philosophy” behind the local embedding lemma is that if P is any subgroup of a group G, then
NG (P ) is a “local neighborhood” of P in which P perhaps behaves nicely. We recall the following proposition
which is used for proving Lemma 1.7.

Proposition 1.6 Given a group G, for any two subgroups S and P , if S ⊆ NG (P ), then P S = SP is the subgroup
of NG (P ) generated by S ∪ P , the subgroup P is normal in SP and (SP )/P ∼
= S/(S ∩ P ).

Proof . Since S ⊆ NG (P ), we have σP σ −1 = P for all σ ∈ S, and thus, it clear that SP = P S. We have στ σ −1 ∈ P
for all σ ∈ S and all τ ∈ P , and thus, for all a, c ∈ S and all b, d ∈ P , we have

(ab)(cd) = (ac)(c−1 bc)d

−1 −1
b a = a−1 (ab−1 a−1 ).

The above identities prove that SP is a group. Since S and P contain the identity, this group contains S and P , and
clearly any subgroup containing S and P contains SP . Therefore, SP is indeed the subgroup of NG (P ) generated
by S ∪ P and it is clear that P is normal in SP . Now, look at the composition ϕ of the injection S −→ SP with the
quotient map SP −→ (SP )/P . It is surjective, and ϕ(σ) = σP for every σ ∈ S. Thus, σ ∈ Ker ϕ iff σ ∈ S ∩ P , and
so Ker ϕ = S ∩ P , and the first isomorphism theorem yields
(SP )/P ∼= S/(S ∩ P ).
After this short digression, we return to the main stream of the lecture.

Lemma 1.7 (Local embedding lemma) Suppose that P is a p-Sylow subgroup of G. Then for every σ ∈
NG (P ), if σ has p-power order then σ ∈ P . In particular, if H is a p-subgroup of NG (P ), then H ⊆ P and
P is unique in NG (P ).

Proof . Let S be any p-subgroup of NG (P ). Look at the group, H, generated by S and P in NG (P ),

denoted Gp{S, P }. Since P is normal in NG (P ), from Proposition 1.6, we have H = SP = P S, and
H/P = (SP )/P ∼ = S/(S ∩ P ). Thus,
(H : P ) = (S : S ∩ P ),
and (S : S ∩ P ) is a p-power, since S is a p-group. On the other hand, (S : S ∩ P ) is prime to p, as (G : P ) =
(G : H)(H : P ) and (G : P ) is prime to p by definition of P . So, we must have (H : P ) = (S : S ∩ P ) = 1,
which implies that H = P . Thus, S = S ∩ P , and S ⊆ P . We finish the proof by letting S be the cyclic
p-group generated by σ.

Theorem 1.8 (Sylow II) If G is a finite group, write Sylp (G) for the collection of all p-Sylow subgroups of
G, and P for the collection of all the p-subgroups of G, where p is a prime number. Then, the following
hold:

(1) sylp (G) = #(Sylp (G)) ≡ 1 (mod p).

1.2. GROUP ACTIONS AND FIRST APPLICATIONS; THE THREE SYLOW THEOREMS 63

(2) For all S ∈ P(G) and all P ∈ Sylp (G), there is some σ ∈ G so that S ⊆ σP σ −1 . In particular, any
two p-Sylow subgroups of G are conjugate in G.
(3) sylp (G) divides #(G); in fact, sylp (G) divides the prime to p part of #(G).

Proof . (1) The group G acts by conjugation on Syl(G) (drop the subscript p in the course of this proof). So
[
Syl(G) = · OG (P ).
distinct orbits

Any S ∈ P(G) also acts by conjugation on Syl(G), and so

[
Syl(G) = · OS (P ).
distinct orbits

What is St(P )? We have

St(P ) = {σ ∈ S | σP σ −1 = P } = S ∩ NG (P ).
But S has p-power order, so S ∩ NG (P ) is a p-subgroup of NG (P ). The embedding lemma implies that
S ∩ NG (P ) ⊆ P , from which we deduce that S ∩ NG (P ) = S ∩ P . So,

#(OS (P )) = (S : S ∩ P ).

Now, take for S one of the p-Sylow subgroups, say P . Then, #(OP (Q)) = (P : P ∩ Q). If Q 6= P , then
P ∩ Q < P , and so, (P : P ∩ Q) is a nontrivial p-power (i.e, not equal to 1). If P = Q, then (P : P ∩ Q) = 1.
Therefore, in the orbit decomposition
[
Syl(G) = · OP (Q),
distinct orbits
Q∈Syl(G)

one orbit has cardinality 1, the rest having nontrivial p-power cardinalities. We conclude that
X
#(Syl(G)) = 1 + p-powers,

and sylp (G) = #(Sylp (G)) ≡ 1 (mod p), as claimed.

(2) Let S ∈ P(G) and look at OG (P ) where P ∈ Syl(G). The subgroup S acts by conjugation on OG (P ).
So, we have
[
OG (P ) = · OS (Q). (∗)
distinct orbits
Q∈OG (P )

If Q ∈ OG (P ), then consider the stabilizer of Q in S,

St(Q) = {σ ∈ S | σQσ −1 = Q} = S ∩ NG (Q).

As before, by the embedding lemma, S ∩ NG (Q) = S ∩ Q. Then, #(OS (Q)) = (S : S ∩ Q). Take S = P
itself. If Q = P , then (P : P ∩ P ) = 1 and #(OP (P )) = 1. On the other hand, if P 6= Q, then (P : P ∩ Q)
is a nontrivial p-power. Thus, as before, using (∗), we deduce that

#(OG (P )) ≡ 1 (mod p).

Assume that (2) is false. Then, there exist some S and some P such that S 6⊆ σP σ −1 for any σ ∈ G. Let
this S act on OG (P ), for this P . But we have
X
#(OG (P )) = #(OS (Q)), (∗∗)
distinct orbits
Q∈OG (P )
64 CHAPTER 1. GROUP THEORY

and #(OS (Q)) = (S : S ∩ Q) where Q is a conjugate of P , so that S 6⊆ Q, and therefore (S : S ∩ Q) is a

nontrivial p-power. Then, (∗∗) implies

#(OG (P )) ≡ 0 (mod p),

a contradiction. Thus, neither S nor P exist and (2) holds.

(3) By (2), Syl(G) = OG (P ), for some fixed P . But the size of an orbit divides the order of the group.
The rest is clear.

Theorem 1.9 (Sylow III) If G is a finite group and P is a p-Sylow subgroup of G, then NG (NG (P )) =
NG (P ).

Proof . Let T = NG (NG (P )) and S = NG (P ), so that T = NG (S) and S C T .

Claim. For every σ ∈ T , if σ has p-power order then σ ∈ P .
The order of T /S is (T : S). But

(G : P ) = (G : T )(T : S)(S : P )

and (G : P ) is prime to p by definition of P . So, (T : S) is prime to p. Consider σ, the image of σ in

T /S. The element σ has p-power order, yet #(T /S) is prime to p. Thus, σ = 1, and so, σ ∈ S. The local
embedding lemma yields σ ∈ P . Therefore, if H is a p-subgroup of T , we have H ⊆ P . Thus, any p-Sylow
subgroup, H, of T is contained in P ; but since H has maximal p-size, H = P . This implies that T has a
single p-Sylow subgroup, namely P . By Sylow II, the group P is normal in T and so T ⊆ NG (P ) = S. Yet,
S ⊆ T , trivially, and S = T .

Remark: A p-Sylow subgroup is unique iff it is normal in G.

Definition 1.2 A group, G, is simple if and only if it possesses no nontrivial normal subgroups ({1} and G
itself are the two trivial normal subgroups).

Example 1.2
(1) Assume that G is a group of order pq, with p and q prime and p < q. Look at the q-Sylow subgroups.
Write syl(q) for the number of q-Sylow subgroups of G. We know that

syl(q) ≡ 1 (mod q) and syl(q) | p.

This implies that syl(q) = 1, p. But p < q, so that p ≡ p (mod q), and the only possibility is syl(q) = 1.
Therefore, the unique q-Sylow subgroup is normal, and G is not simple.
(2) Assume that G is a group of order pqr, with p, q, r prime and p < q < r. Look at the r-Sylow
subgroups. We must have
syl(r) ≡ 1 (mod r) and syl(r) | pq.
This implies that syl(r) = 1, p, q, pq. Since p < r and q < r, as above, p and q are ruled out, and syl(r) = 1, pq.
Suppose that syl(r) = pq. We see immediately that r < pq. Now, each r-Sylow subgroup is isomorphic to
Z/rZ (cyclic of prime order), and any two distinct such subgroups intersect in the identity (since, otherwise,
they would coincide). Hence, there are pq(r − 1) elements of order r. We shall now show that if syl(r) = pq,
then syl(q) = 1. Assume that syl(r) = pq and look at the q-Sylow subgroups of G. We have

syl(q) ≡ 1 (mod q) and syl(q) | pr.

This implies that syl(q) = 1, p, r, pr and, as before, p is ruled out since p < q. So, syl(q) = 1, r, pr. Suppose
that syl(q) = r or syl(q) = pr, and call it x. Reasoning as before but now on the q-Sylow subgroups, we see
1.2. GROUP ACTIONS AND FIRST APPLICATIONS; THE THREE SYLOW THEOREMS 65

that there are x(q − 1) elements of order q. Now, q − 1 ≥ p and x ≥ r. Thus, there are at least rp elements
of order q. But r > q, so there are more than pq elements of order q. Now, since there are pq(r − 1) elements
of order r and more than pq elements of order q, there are more than

pq(r − 1) + pq = pqr − pq + pq = pqr

elements in G, a contradiction. So, either the r-Sylow subgroup is normal in G (which is the case when
r > pq) or the q-Sylow subgroup is normal in G. In either case, G is not simple.

Cases (1) and (2) have the following generalizations:

(a) Frobenius (1890’s) showed that if #(G) = p1 p2 · · · pt , a product of distinct primes, then G is not simple.
The proof uses group representations and characters.

(b) Burnside (1901) proved the “pa q b -theorem”: If #(G) = pa q b , where p, q are distinct primes and a, b ∈ N,
then G is not simple. There are three known proofs, all hard, and all but one use group representations.

Obvious generalizations of (a) and (b) are false. The easiest case is #(G) = 22 · 3 · 5 = 60. Indeed, the
alternating group, A5 , is simple. After proving (b), Burnside conjectured (circa 1902) that every nonabelian
group of odd order is not simple. This conjecture was proved in 1961 by W. Feit and J. Thompson. The
proof is very hard, and very long (over 200 pages).
A piece of the proof of (a) and (b) is the following proposition:

Proposition 1.10 If G is a finite group and p is the smallest prime number which divides the order of G,
then any subgroup, H, of index p is automatically normal in G.

Proof . Take H so that (G : H) = p. Consider the set S = {H1 = H, H2 , . . . , Hp } of cosets of H in G. The

group G acts on S by translation,

σ · Hj = σHj = Hl , for some l, with 1 ≤ l ≤ p.

This action is nontrivial, that is, we get a nontrivial homomorphism θ : G → Sp (where Sp ∼

= Aut(S) is the
group of permutations on p elements), and Im θ 6= {1}. We shall prove that H = Ker θ, which yields H C G.
Observe that #(G) = #(Ker θ) · #(Im θ). We must have

(1) #(Im θ) | p!

(2) #(Im θ) | #(G).

But #(G) = pα K, where K contains primes greater than p. Therefore, #(Im θ) = pa J, where J = 1 or
J contains primes greater than p. If J 6= 1, then J contains some prime q > p, and since pα J divides
p! = p(p − 1) · · · 2 · 1, the prime q must divide p!. Since q is prime, q must divide one of the terms in p!,
which is impossible, since q > p. We conclude that J = 1. Now, a ≥ 1 since Im θ is nontrivial. If a ≥ 2,
since pa−1 | (p − 1) · · · 2 · 1, the prime p should divide p − j, for some j with 1 ≤ j ≤ p − 1. However, this
is impossible, and so, a = 1. Therefore, #(Im θ) = p and (G : Ker θ) = p. Note that σ ∈ Ker θ iff σ acts
trivially on S iff στ H = τ H iff τ −1 Hτ = H iff τ −1 στ ∈ H for all τ iff σ ∈ τ Hτ −1 for all τ ∈
/ H iff
\
σ∈ τ Hτ −1 .
τ ∈G

We deduce that \
Ker θ = τ Hτ −1 ⊆ H.
τ ∈G
 66 CHAPTER 1. GROUP THEORY

As (G : Ker θ) = p = (G : H) and Ker θ ⊆ H, we get H = Ker θ, and H is indeed normal in G.

Note that we can now improve Corollary 1.5 as follows: If G is a p-group with #(G) = pr , then there is
a descending chain of subgroups
G = G0 > G1 > · · · > Gr = {1},
where each Gj+1 is normal in Gj and each quotient Gj+1 /Gj is simple; so, Gj+1 /Gj = Z/pZ, a cyclic group
of order p.

Definition 1.3 A composition series for a group G is a chain of subgroups

G = G0 > G1 > · · · > Gt = {1}

in which each subgroup Gj+1 is maximal, normal in Gj . The factor groups G/G1 , G1 /G2 , . . ., Gt−1 /Gt =
Gt−1 are called the composition factors of the given composition series and each one is a simple group.

Remark: Every finite group possesses a composition series (DX).

 Not every group possesses maximal subgroups, even maximal normal subgroups (such groups must be
infinite).

However, finitely generated groups do possess maximal subgroups, but because such groups can be infinite,
the proof requires a form of transfinite induction known as Zorn’s lemma. Since this lemma is an important
tool, we briefly digress to state the lemma and illustrate how it is used.
Recall that a partially ordered set or poset is a pair, (S, ≤), where S is a set and ≤ is a partial order on
S, which means that ≤ is a binary relation on S satisfying the properties: For all a, b, c ∈ S, we have:
(1) a ≤ a (reflexivity)
(2) If a ≤ b and b ≤ c, then a ≤ c (transitivity)
(3) If a ≤ b and b ≤ a, then a = b. (antisymmetry)
Observe that given a, b ∈ S, it may happen that neither a ≤ b nor b ≤ a. A chain, C, in S is a linearly
ordered subset of S (which means that for all a, b ∈ C, either a ≤ b or b ≤ a). The empty set is considered
a chain. An element, b ∈ S, is an upper bound of C (resp. a lower bound of C) if a ≤ b for all a ∈ C (resp.
b ≤ a for all a ∈ C). Note that an upper bound of C (resp. a lower bound of C) need not belong to C. We
say that C ⊆ S is bounded above if it possesses some upper bound (in S) (resp. bounded below if it possesses
some lower bound (in S)). The notion of least upper bound (resp. greatest lower bound) of a chain is clear
as is the notion of least or greatest element of a chain. These need not exist. A set, S, which is a chain, is
well ordered iff every nonempty subset of S has a least element.

Remark: Obviously, the notions of upper bound (resp. lower bound), maximal (resp. minimal) element, greatest
(resp. smallest) element, all make sense for arbitrary subsets of a poset, and not just for chains. Some books define
a well ordered set to be a poset so that every nonempty subset of S has a least element. Thus, it is not required that
S be a chain, but it is required that every nonempty subset have a least element, not just chains. It follows that a
well ordered set (under this new definition) is necessarily a chain. Indeed, for any two elements a, b ∈ S, the subset
{a, b} must have a smallest element, so, either a ≤ b or b ≤ a.
Hausdorff maximal principle: Every nonempty poset possesses a maximal chain.
From set theory, it is known that Hausdorff’s maximal principle is equivalent to the axiom of choice,
which is also equivalent to Zermelo’s well ordering principle (every nonempty subset can be well ordered).
We say that a poset is inductive iff every nonempty chain possesses a least upper bound.
1.2. GROUP ACTIONS AND FIRST APPLICATIONS; THE THREE SYLOW THEOREMS 67

Zorn’s lemma: Each inductive poset possesses a maximal element.

Proof . By Hausdorff.

Remark: Some books define a poset to be inductive iff every nonempty chain is bounded above. Zorn’s lemma still
holds under this slightly weaker assumption. In practice, this makes little difference, because when proving that a
chain is bounded above, one usually shows that this chain has a least upper bound.
Here are two illustrations of the use of Zorn’s lemma.

Theorem 1.11 Every finitely generated group, G, possesses a maximal subgroup.

Proof . Consider the set, S, of all proper subgroups, S H, of G. Partially order S by inclusion (ı.e., H ≤ K
iff H ⊆ K). Let {Hα } be a chain in S. If H = α Hα , we see that H is a group and that it is the
least upper bound of {Hα }. We must show that H 6= G. If H = G, then as G is finitely generated,
H = G = Gp{σ1 , . . . , σt }, with σi ∈ H for i = 1, . . . , t. This means that, for each i, there is some αi so that
σi ∈ Hαi . Since {Hα } is a chain, there is some s so that Hαj ⊆ Hαs for j = 1, . . . , t. Thus, σ1 , . . . , σt ∈ Hαs ,
and so, Hαs = G, contradicting the fact that Hαs 6= G. Therefore, S is inductive, and consequently, by
Zorn’s lemma, it possesses a maximal element. Such an element is a maximal subgroup of G.
As a second illustration of Zorn’s lemma, we prove that every vector space has a Hamel basis. Given a
vector space, V , over a field, k, a Hamel basis of V is a family, {eα }α∈Λ , so that:
(1) For every v ∈ V , there exists a finite subset of Λ, say I, and some elements of k for these α’s in I, say
cα , so that X
v= cα eα .
α∈I
P
(2) The eα ’s are linearly independent, i.e., given any finite subset I of Λ, if α∈I cα eα = 0, then cα = 0,
for all α ∈ I.

Theorem 1.12 Every vector space, V , possesses a Hamel basis.

Proof . Let S ∗ be the collection of all subspaces, W , of V which possess a Hamel basis, together with a choice
of a basis. Write (W, {eα }) for any element of S ∗ . The collection, S ∗ , is nonempty, since finitely dimensional
vector spaces have bases. Partially order S ∗ by (W, {eα }) ≤ (W f , {fβ }) iff

f and
(a) W ⊆ W
(b) {eα } ⊆ {fβ }, which means that the basis {fβ } extends the basis {eα }.
We claim that S ∗ is inductive.
(λ)
Given a chain, {W (λ) , {eα }), in S ∗ , take
[ [
W = W (λ) and {eγ } = {e(λ)
α } ⊆ W.
λ λ

The reader should check that {eγ } is a basis for W (DX); therefore, (W, {eγ }) is the least upper bound of
our chain. By Zorn’s lemma, there exists a maximal element of S ∗ , call it (W0 , {eγ }). We need to show that
W0 = V . If not, there is some v ∈ V with v ∈
/ W0 . Consider the subspace

Z = W0 q kv = {w + ξv | w ∈ W0 , ξ ∈ k}.

The subspace, Z, strictly contains W0 and {eγ } ∪ {v} is a Hamel basis for Z (DX). However, this contradicts
the maximality of W0 . Therefore, W0 = V .
68 CHAPTER 1. GROUP THEORY

Corollary 1.13 If W is a subspace of V and {eα } is a Hamel basis for W , then there exists a Hamel basis
of V extending {eα }.

Application: The field, R, is a vector space over Q, and 1 ∈ Q is a Hamel basis for Q. We can extend
this basis of Q to a Hamel basis for R (over Q), call it {eα }α∈Λ , and say, e0 = 1; then, R/Q is a vector space
(over Q) spanned by the eα other than e0 . So, we have
a
R/Q ∼= Q.
α∈Λ,α6=0
1.3. ELEMENTARY THEORY OF P -GROUPS 69

1.3 Elementary Theory of p-Groups

Recall that for a group G, the center of G, denoted Z(G), is given by

Z(G) = {σ ∈ G | (∀τ ∈ G)(στ = τ σ)}.

We write [σ, τ ] for the element στ σ −1 τ −1 , called the commutator of σ and τ . Observe that [τ, σ] = [σ, τ ]−1 .
Also,
Z(G) = {σ ∈ G | (∀τ ∈ G)([σ, τ ] = 1)}
and Z(G) is the centralizer of G under conjugation.
Let G act on itself by conjugation. When do we have OG (σ) = {σ}? This happens when

(∀τ ∈ G)(τ στ −1 = σ) i.e. (∀τ ∈ G)(τ στ −1 σ −1 = [τ, σ] = 1).

Thus, σ ∈ Z(G) iff OG (σ) = {σ}.

Remark: Obviously, \
Z(G) = ZG (σ).
σ∈G

Moreover, it is obvious that σ ∈ ZG (σ) for every σ ∈ G. Thus, for every σ ∈

/ Z(G), we have Z(G) < ZG (σ) (obviously,
ZG (σ) = G if σ ∈ Z(G).) Therefore, if G is nonabelian, then Z(G) < ZG (σ) for all σ ∈ G.

Proposition 1.14 The center, Z(G), of a p-group, G, is nontrivial.

Proof . If we let G act on itself by conjugation, we know that G is the disjoint union of distinct orbits, and
since OG (σ) is the conjugacy class of σ and σ ∈ Z(G) iff OG (σ) = {σ}, we get
[
G = Z(G) ∪· · OG (τ ).
distinct orbits
τ ∈Z(G)
/

Consequently, using the fact that #(OG (τ )) = (G : St(τ )), we get

X
#(G) = #(Z(G)) + (G : St(τ )). (∗)
distinct orbits
τ ∈Z(G)
/

But #(G) = pr , so that each term (G : St(τ )) for τ ∈

/ Z(G) is a nontrivial p-power. So, in (∗), all terms
must be divisible by p. Therefore, p | #(Z(G)).
Note that Z(G) is normal in G. Thus, G/Z(G) is a p-group of strictly smaller order, providing a basis
for induction proofs.
We make the following provisional definition (due to E. Galois, 1832). A finite group, G, is solvable iff it
possesses a composition series all of whose factors are abelian, or equivalently iff it possesses a composition
series all of whose factors are cyclic of prime order.
We have shown that a p-group is solvable.

Remark: The above definition is provisional because it only works for finite group (c.f. Definition 1.7), but
the concept of a solvable group can be defined for an arbitrary group.

Corollary 1.15 Every p-group of order less than or equal to p2 is abelian.

70 CHAPTER 1. GROUP THEORY

Proof . Since #(G) = 1, p, p2 and G is obviously abelian in the first two cases, we may assume that #(G) = p2 .
We know that Z(G) is non-trivial and we must prove that Z(G) = G. If Z(G) < G, then there is some
σ ∈ G so that σ ∈/ Z(G). Clearly, Z(G) ⊆ ZG (σ) (where ZG (σ) denotes the centralizer of σ in G). But
σ ∈ ZG (σ) implies that (ZG (σ) : Z(G)) ≥ p and since Z(G) is nontrivial, we must have ZG (σ) = G. So,
σ ∈ Z(G), a contradiction.
We now consider a nice property possessed by p-groups called property (N). If G is any group, G has
property (N) iff for every proper subgroup, H, of G, the group H is a proper subgroup of NG (H).

Remark: An abelian group has (N). Indeed, every subgroup of an abelian group is normal, and so, NG (H) =
G.

Proposition 1.16 Every p-group has (N).

Proof . We proceed by induction on #(G) = pr . Corollary 1.15 takes care of the base case of the induction.
Next, let #(G) = pr+1 and assume that the induction hypothesis holds up to r. We know that Z(G) is
nontrivial, and so #(G/Z(G)) ≤ pr . Thus, G/Z(G) has (N). Pick H, any proper subgroup of G. Of course,
Z(G) ⊆ NG (H), and we may assume that Z(G) ⊆ H (since, otherwise, it is clear that H < NG (H)). By
the second homomorphism theorem, the question: H < NG (H)? is reduced to the question: H < NG (H)?,
where the bar means pass to G/Z(G). But in this case, as Z(G) ⊆ H, we see that (DX)

NG (H) = NG (H),

and we just remarked that G = G/Z(G) has (N). Therefore, NG (H) > H, and so, NG (H) > H, as desired.

Groups that have property (N) tend to have good properties. Here are a few of them.

Proposition 1.17 Say G is a finite group having (N), then each of its p-Sylow subgroups is unique and
normal in G. Every maximal subgroup of G is also normal and has prime index.

Proof . Look at P , a p-Sylow subgroup of G. Now, if NG (P ) 6= G, then by (N), we have NG (NG (P )) >
NG (P ), a contradiction to Sylow III. Thus, NG (P ) = G and so, P C G. Next, let H be a maximal subgroup.
By (N), we have NG (H) > H, yet H is maximal, so NG (H) = G, and H C G. It follows that G/H is a group
with no nontrivial subgroup. But then, G/H is cyclic of prime order.

Proposition 1.18 Say G is a finite group and suppose that

(a) g = #(G) = pa1 1 · · · pat t (where the pi ’s are distinct primes)

(b) G has (N).

Write Pj for the pj -Sylow subgroup of G. Then, the map

Y Y ϕ
P1 ··· Pt −→ G

via ϕ(σ1 , . . . , σt ) = σ1 · · · σt is an isomorphism of groups. Hence, G is isomorphic to a product of p-groups.

The proof depends on the following lemma:

Lemma 1.19 Let G be a group and let H and K be normal subgroups of G. If H ∩ K = {1}, then every
element of H commutes with every element of K. Suppose that σ and τ are commuting elements in G, with
orders r and s respectively. If r and s are relatively prime then the order of στ is rs.
1.3. ELEMENTARY THEORY OF P -GROUPS 71

Proof . Look at [σ, τ ], where σ ∈ H and τ ∈ K. We have

[σ, τ ] = στ σ −1 τ −1 = (στ σ −1 )τ −1 = σ(τ σ −1 τ −1 ).

Now, στ σ −1 ∈ K, since K C G. Thus, (στ σ −1 )τ −1 ∈ K. Similarly, σ(τ σ −1 τ −1 ) ∈ H. But H ∩ K = {1},

and since we just proved that [σ, τ ] ∈ H ∩ K, we have [σ, τ ] = 1. The second part of the lemma is left to the
reader (DX).
Proof of Proposition 1.18. By Proposition 1.17, each pj -Sylow subgroup Pj is normal in G. First, we claim
Q Q ϕ
that the map P1 · · · Pt −→ G is a group homomorphism. Now, because the orders of Pi and Pj are
relatively prime if i 6= j, we have Pi ∩ Pj = {1}. Since

ϕ((σ1 , . . . , σt )(τ1 , . . . , τt )) = σ1 τ1 · · · σt τt ,

using Lemma 1.19, we can push each τj past σj+1 · · · σt , and we get

ϕ((σ1 , . . . , σt )(τ1 , . . . , τt )) = σ1 · · · σt τ1 · · · τt = ϕ(σ1 , . . . , σt )ϕ(τ1 , . . . , τt ),

proving that ϕ is a homomorphism. The kernel of ϕ consists of those σ = (σ1 , . . . , σt ) so that σ1 · · · σt = 1, or

equivalently, σt−1 = σ1 · · · σt−1 . Using Lemma 1.19 and an obvious induction, the order on the righthand side
lt−1
is pl11 · · · pt−1 and the order on the left hand side in pltt , which implies that l1 = · · · = lt , and Q all σj = 1.
thus, Q
Therefore, Ker ϕ = {1} and ϕ is injective. One more application of Lemma 1.19 yields #(P1 · · · Pt ) = g.
Since ϕ is injective, it is an isomorphism.

Remark: The proof of Proposition 1.18 only uses the fact that every p-Sylow subgroup is normal in G.

Definition 1.4 Let G be any group, then the Frattini subgroup of G, denoted Φ(G), is the intersection of
all the maximal proper subgroups of G. In case G has no maximal proper subgroup, we set Φ(G) = G.

Remark: The additive abelian group (Q, +) has no maximal proper subgroup.

Definition 1.5 In a group, G, an element σ is a non-generator iff for every subset, A, if G = Gp{A, σ},
then G = Gp{A} (where Gp{A} denotes the subgroup of G generated by A).

As an example, assume that G is a cyclic group of order pr . Then, Φ(G) is the cyclic subgroup of order
r−1
p .

Proposition 1.20 The Frattini subgroup of G is a characteristic subgroup of G, i.e., for every automor-
phism, ϕ ∈ Aut(G), we have ϕ(Φ(G)) = Φ(G). In particular, Φ(G) is normal in G. Furthermore, if G is
finite, then
Φ(G) = {σ ∈ G | σ is a non-generator}.

Proof . Every automorphism permutes the collection of maximal subgroups of G. Therefore, Φ(G) is char-
acteristic. Now assume G is finite, or, at least, that every proper subgroup is contained in a maximal
subgroup.
Claim: If Gp{A, Φ(G)} = G, then Gp{A} = G.
If not, Gp{A} 6= G, and so, there exists a maximal subgroup, M , containing Gp{A}. Now, Φ(G) ⊆ M ,
therefore, Gp{A, Φ(G)} ⊆ M 6= G, a contradiction. This proves that Φ(G) is contained in the set of
non-generators.
Conversely, assume that σ is a non-generator. Were σ ∈ / Φ(G), we would have a maximal subgroup,
M , with σ ∈ / M . Take M = A in the definition of a non-generator. Look at Gp{M, σ}. Of course,
M ⊆ Gp{M, σ} and σ ∈ Gp{M, σ}, so M < Gp{M, σ}. But M is maximal, and so, Gp{M, σ} = G. By
definition (since σ is a non-generator), G = Gp{M }, and thus, G = M , a contradiction.
72 CHAPTER 1. GROUP THEORY

Definition 1.6 A group G is an elementary abelian p-group iff

(1) It is abelian, and

(2) For every σ ∈ G, we have σ p = 1.

Remark: Any elementary abelian p-group is, in a natural way, a vector space over Fp . Conversely, for
any vector space over the finite field Fp , its additive group is an elementary abelian p-group. Under this
correspondence, an endomorphism of G goes over to a linear map and an automorphism of G goes to an
invertible linear map. The group G is finite iff the corresponding vector space is finite dimensional.
(Given G, write the group operation additively. Thus, we have

p · σ = σ + · · · + σ = 0.
| {z }
p

The finite field Fp acts on G as follows: If λ ∈ Fp = Z/pZ, i.e., λ ≡ 0, 1, . . . , p − 1 (mod p), we set

λ · σ = σ + ··· + σ .
| {z }
λ (mod p) times

The reader should check that scalar multiplication is indeed well defined and that the facts asserted in the
previous remark are true (DX).)

Proposition 1.21 For any p-group, G, the quotient group, G/Φ(G), is an elementary abelian p-group.

Proof . Say H is a maximal subgroup of G. Since G has (N), the group, H, is normal in G and (G : H) = p.
Therefore, G/H is cyclic of order p. Write σ for the image of σ in G/H. We know that (σ)p = 1. So, σ p = 1,
i.e., σ p ∈ H. But H is arbitrary, and so,
\
σp ∈ H = Φ(G).
H maximal

Now, G/H is abelian since G/H = Z/pZ. This implies that [G, G] ⊆ H (here [G, G] is the subgroup of G
generated by the commutators, called the commutator group of G; it is the smallest normal subgroup, K, of
G such that G/K is abelian). Since H is arbitrary, we get
\
[G, G] ⊆ H = Φ(G).
H maximal

This shows that G/Φ(G) is abelian. As σ p ∈ Φ(G), we get (σ)p = 1 in G/Φ(G), where σ is the image of σ
in G/Φ(G).
We now come to a famous theorem of Burnside.

Theorem 1.22 (Burnside Basis Theorem) Say G is a p-group and let d be the minimal number of elements
found among all minimal generating sets for G. The following properties hold:

(1) Given any set of d elements in G, say σ1 , . . . , σd , they generate G iff σ1 , . . . , σd are a basis of G/Φ(G).

(2) More generally, any set of t elements σ1 , . . . , σt in G generates G iff {σ1 , . . . , σt } spans G/Φ(G). Hence,
any set of generators of G possesses a subset of exactly d elements which generates G. The number d
is the dimension of G/Φ(G) over Fp .
1.3. ELEMENTARY THEORY OF P -GROUPS 73

Proof . Everything follows from the statement: σ1 , . . . , σt generate G iff σ1 , . . . , σt generate G = G/Φ(G)
(DX).
The implication (=⇒) is trivial and always true. Conversely, if σ1 , . . . , σt generate G, then

G = Gp{σ1 , . . . , σt , Φ(G)}.

But then, as Φ(G) is the set of nongenerators, we have

G = Gp{σ1 , . . . , σt , Φ(G)} = Gp{σ1 , . . . , σt },

as desired.
Let G be a group (possibly infinite). We set ∆(0) (G) = G, and ∆(1) (G) = [G, G] and, more generally

∆(j+1) (G) = [∆(j) (G), ∆(j) (G)] = ∆(1) (∆(j) (G)).

Observe that ∆(1) (G) = [G, G] is the commutator group of G, and recall that for any normal subgroup, H,
of G, we have ∆(1) (G) ⊆ H iff G/H is abelian. Moreover, for a simple nonabelian group, [G, G] = G.

Proposition 1.23 Suppose G is a group, then each ∆(j) (G) is a characteristic subgroup of G and each
group ∆(j) (G)/∆(j+1) (G) is abelian (j ≥ 0). If G has property (N), then ∆(1) (G) ⊆ Φ(G) < G (provided
maximal subgroups exist). If G is a p-group, then the chain

G ⊇ ∆(1) (G) ⊇ ∆(2) (G) ⊇ · · · ⊇ ∆(t) (G) ⊇ · · ·

is strictly descending and reaches {1} after finitely many steps.

Proof . The group ∆(1) (G) consists of products of the form

[σ1 , τ1 ] · · · [σl , τl ], l ≥ 1.

If ϕ ∈ Aut(G), then
ϕ([σ1 , τ1 ] · · · [σl , τl ]) = ϕ([σ1 , τ1 ]) · · · ϕ([σl , τl ]),
and ϕ([σ, τ ]) = [ϕ(σ), ϕ(τ )], so ∆(1) (G) is characteristic. We prove that ∆(j) (G) is characteristic by induction
on j. The base case j = 1 has just been established. Look at ∆(j+1) (G). By the induction hypothesis, we
have ϕ(∆(j) (G)) = ∆(j) (G). Therefore, ϕ is an automorphism of ∆(j) (G). Yet, ∆(j+1) (G) = ∆(1) (∆(j) (G)),
and we proved that ∆(1) (H) is characteristic for any group H (case j = 1). Now, G/∆(1) (G) is abelian for
any group G, so ∆(j) (G)/∆(j+1) (G) = ∆(j) (G)/∆(1) (∆(j) (G)) is abelian.
Say G has (N) and possesses maximal subgroups. If H is a maximal subgroup of G we know that H C G
and H has prime index. So, G/H is abelian, and thus, ∆(1) (G) ⊆ H. Since H is arbitrary, we deduce that
\
∆(1) (G) ⊆ H = Φ(G).
H maximal

Now, assume that G is a p-group. Then, G has (N), and thus, ∆(1) (G) ⊆ Φ(G) < G. But ∆(1) (G) in
turn is a p-group, so we can apply the argument to ∆(1) (G) and we get ∆(2) (G) < ∆(1) (G), etc.
Nomenclature.

(1) The group ∆(1) (G) is called the first derived group of G (or commutator group of G).

(2) The group ∆(j) (G) is the j-th derived group of G.

74 CHAPTER 1. GROUP THEORY

(3) The sequence

G = ∆(0) (G) ⊇ ∆(1) (G) ⊇ ∆(2) (G) ⊇ · · · ⊇ ∆(t) (G) ⊇ · · ·
is the derived series of G.

(4) The smallest t ≥ 0 for which ∆(t) (G) = {1} is the derived length of G and if ∆(t) (G) is never {1} (e.g.,
in a nonabelian simple group) then the derived length is infinite. Write δ(G) for the derived length of
G.

Look at the derived series of G:

G = ∆(0) (G) ⊇ ∆(1) (G) ⊇ ∆(2) (G) ⊇ · · · ⊇ ∆(t) (G) ⊇ · · · .

Each quotient ∆(j) (G)/∆(j+1) (G) is abelian. Suppose G is finite, then ∆(j) (G)/∆(j+1) (G) is finite abelian.
Interpolate between ∆(j) (G) and ∆(j+1) (G) a sequence of subgroups, necessarily normal, each maximal in
the previous one. If δ(G) < ∞, we get a composition series all of whose factors are cyclic of prime order.
This proves half of the

Proposition 1.24 A necessary and sufficient condition that a finite group be solvable (in the sense of Galois)
is that δ(G) < ∞.

Proof . We need only prove: If G is (Galois) solvable, then δ(G) < ∞. Say

G = G0 > G1 > G2 > · · · > Gt = {1}

is a composition series with abelian factors. We have G1 < G and G/G1 is abelian. Therefore, by a previous
remark, ∆(1) (G) ⊆ G1 . Each quotient Gj /Gj+1 is abelian, so ∆(1) (Gj ) ⊆ Gj+1 for all j. Now, ∆(1) (G) ⊆ G1
implies that ∆(1) (∆(1) (G)) ⊆ ∆(1) (G1 ), and so,

∆(2) (G) ⊆ ∆(1) (G1 ) ⊆ G2 .

An easy induction yields ∆(r) (G) ⊆ Gr (DX). Therefore, ∆(t) (G) ⊆ {1}, i.e., δ(G) ≤ t.
Observe that we actually proved more: The derived length, δ(G), of a solvable finite group is less than
or equal to the length of any composition series for G.

Definition 1.7 An arbitrary group, G, is solvable iff δ(G) < ∞.

Proposition 1.25 Say G is a p-group of order at least p2 . Then, (G : ∆(1) (G)) ≥ p2 .

Proof . We may assume that G is nonabelian, else ∆(1) (G) = {1} and so, (G : ∆(1) (G)) = #(G) ≥ p2 . As
G is a p-group, if (G : ∆(1) (G)) < p2 , then (G : ∆(1) (G)) = p. We know that ∆(1) (G) ⊆ Φ(G). Therefore,
(G : Φ(G)) = p and the Burnside dimension of G (i.e. dimFp G/Φ(G)) is equal to 1. By the Burnside basis
theorem, G is cyclic, so abelian, a contradiction.
1.4. GROUP EXTENSIONS 75

1.4 Group Extensions

Let G be a finite group and let
G = G0 > G1 > G2 > · · · > Gt = {1}
be a composition series. We have the groups Gj /Gj+1 = Gj , the composition factors of the composition
series.
Problem: Given the (ordered) sequence G0 , G1 , G2 , . . . , Gt−1 , try to reconstruct G.
Say H and K are two groups, G is a “big” group and H C G with G/H f→ K. Note, this is exactly the
case at the small end of a composition series. We have
Gt−1 = Gt−1 = Gt−1 /Gt . We also have Gt−1 C Gt−2 , and the quotient is Gt−2 , so we are in the above
situation with H = Gt−1 = Gt−1 , K = Gt−2 , G = Gt−2 , and G/H f
→ K.
The above situation is a special case of an exact sequence. A diagram of groups and homomorphisms
ϕ ψ
0 −→ H −→ G −→ K −→ 0,
where the map 0 −→ H is the inclusion of {1} into H and the map K −→ 0 is the surjection sending
every element of K to 1 in the trivial group {1}, is called a short exact sequence iff the kernel of every
homomorphism is equal to the image of the previous homomorphism on its left. This means that
(1) Ker ϕ = {1}, so ϕ is injective, and we identify H with a subgroup of G.
(2) H = Im ϕ = Ker ψ, so H is normal in G.
(3) Im ψ = K, so ψ is surjective. By the first homomorphism theorem, G/H f
→ K.
(4) Properties (1), (2), (3) are equivalent to 0 −→ H −→ G −→ K −→ 0 is exact.

Going back to composition series, we have Gj+1 C Gj and Gj = Gj /Gj+1 . So, a composition series is
equivalent with a collection of short exact sequences
0 −→ Gt−1 −→ Gt−2 −→ Gt−2 −→ 0
0 −→ Gt−2 −→ Gt−3 −→ Gt−3 −→ 0
····································
0 −→ G1 −→ G −→ G0 −→ 0.
So our problem reduces to the problem of group extensions: Given H and K, groups, find (classify) all
groups, G, which can possibly fit into an exact sequence
0 −→ H −→ G −→ K −→ 0.

The problem is very hard when H is nonabelian.

Definition 1.8 If A, G are groups, a group, G, is an extension of G by A iff G fits into an exact sequence
(E) 0 −→ A −→ G −→ G −→ 0.
Two such extensions (E), (E 0 ) are equivalent iff there exists a commutative diagram

(E) 0 /A /G /G /0

ψ

(E 0 ) 0 /A / G0 /G / 0.
76 CHAPTER 1. GROUP THEORY

Remarks:
(1) The homomorphism, ψ, in the above diagram is an isomorphism of groups. So, the notion of equivalence
is indeed an equivalence relation (DX).
(2) Equivalence of group extensions is stronger than isomorphism of G with G 0 .
(3) The group G in (E) should be considered a “fibre space” whose base is G and whose “fibre” is A.

As we remarked before, the theory is good only when A is abelian. From now on, we assume A is an
abelian group.

Proposition 1.26 Say

(E) 0 −→ A −→ G −→ G −→ 0

is a group extension and A is abelian. Then, there exists a natural action of G on A; so, A is a G-module.
Equivalent extensions give rise to the same action.

Proof . Denote the surjective homomorphism G −→ G in (E) by bar ( ). Pick ξ ∈ G and any a ∈ A. There
exists x ∈ G with x = ξ. Consider xax−1 . Since A C G, we have xax−1 ∈ A. If y ∈ G and if y = x = ξ, then
x = yα for some α ∈ A. Then,
xax−1 = yαaα−1 y −1 = yay −1 ,
as A is abelian. Therefore, if we set
ξ · a = xax−1 ,
this is a well-defined map. The reader should check that it is an action (DX). Assume we have an equivalence
of extensions between (E) and (E 0 ):

(E) 0 /A /G /G /0

ψ

(E 0 ) 0 /A / G0 /G / 0.

Pick ξ ∈ G and any a ∈ A. Denote the E-action by · and the E 0 -action by · ·. Observe that

ξ · a = ψ(ξ · a) = ψ(xax−1 ) = ψ(x)ψ(a)ψ(x)−1 = ψ(x)aψ(x)−1 ,

since the left vertical arrow is the identity in the diagram, yet ψ(x) lifts ξ in G 0 , as the right vertical arrow
is the identity in the diagram. However, by definition,

ξ · · a = ψ(x)aψ(x)−1 ,

so, ξ · · a = ξ · a for all a ∈ A.

The type of (E) is the structure of A as G-module, i.e., the action of G on A. We get a first invariant of
a group extension, its action (of G on A).
Fix the action of (E). Can we classify the extensions up to equivalence? Say we are given an extension
π
(E) 0 −→ A −→ G −→ G −→ 0.

There is always a set-theoretic section s : G → G, i.e., a set map, s, so that π(s(σ)) = σ for all σ ∈ G. Write
uσ for the s-lift of σ, i.e., s(σ) = uσ . So, π(uσ ) = uσ = σ. As s is not necessarily a group homomorphism,
what is the obstruction? Consider

uσ uτ (uστ )−1 = f (σ, τ ). (∗)

1.4. GROUP EXTENSIONS 77

Note that f (σ, τ ) = 1 iff s : σ 7→ uσ is a group homomorphism. If we apply the homomorphism bar to (∗),
Qf (σ, τ ) = 1, and so, f (σ, τ ) ∈ A. Observe that f is a function
we get
f : G G → A. Given x ∈ G, look at x. We know that x = σ ∈ G. If we apply bar to xu−1 σ , we get 1,
−1 −1 −1
because uσ = σ and x = σ. So, we have xuσ ∈ A, which yields x = auσ , for some a ∈ A.
Observe that:
(1) Each x determines uniquely a representation x = auσ , with a ∈ A and σ ∈ G.
Q Q
(2) The map A G −→ G (where A G is the product of A and G as sets) via

(a, σ) 7→ auσ

is a bijection of sets (an isomorphism in the category of sets).

Q
(3) G (as a set) is just A G (product in the category of sets).2

Can we recover the group multiplication of G? We have

(auσ )(buτ ) = a(uσ b)uτ

= a(uσ bu−1
σ )uσ uτ
= a(σ · b)uσ uτ
= a(σ · b)f (σ, τ )uστ
= cuστ ,

where c = a(σ · b)f (σ, τ ), and c ∈ A. Therefore, knowledge of the action and f (σ, τ ) gives us knowledge of
the group multiplication.
Thus, it is natural to try to go backwards and make G from the groups A and G, the action of G on A,
and f . It is customary to use an additive notation for the group operation in A, since A is abelian. The
underlying set of the group G is Y
A G = {ha, σi | a ∈ A, σ ∈ G}.
Multiplication is given by

ha, σihb, τ i = ha + σ · b + f (σ, τ ), στ i. (†)

However, the multiplication defined by (†) is supposed to make G into a group, and this imposes certain
conditions on f . First, we deal with associativity. For this, we go back to the original G where we have the
associative law:
(auσ )((buτ )(cuρ )) = ((auσ )(buτ ))(cuρ ).
Expanding the left hand side, we get

(auσ )((buτ )(cuρ )) = (auσ )(b(τ · c)f (τ, ρ)uτ ρ )

= (aσ · (b(τ · c)f (τ, ρ))f (σ, τ ρ)uσ(τ ρ)
= a(σ · b)(στ · c)(σ · f (τ, ρ))f (σ, τ ρ)uσ(τ ρ) .

Expanding the righthand side, we get

((auσ )(buτ ))(cuρ ) = a(σ · b)f (σ, τ )uστ )(cuρ )

= a(σ · b)f (σ, τ )(στ · c)f (στ, ρ)u(στ )ρ .
2 In (2) and (3) we give a foretaste of the language of categories to be introduced in Section 1.7.
78 CHAPTER 1. GROUP THEORY

Thus, the associative law becomes (writing RHS = LHS)

f (σ, τ )(στ · c)f (στ, ρ) = (στ · c)(σ · f (τ, ρ))f (σ, τ ρ).
Now, all the above terms are in A, and since A is abelian, we can permute terms and perform cancellations,
and we get
f (σ, τ )f (στ, ρ) = (σ · f (τ, ρ))f (σ, τ ρ). (††)
This identity is equivalent to the associativity law in G.
Q
Nomenclature: A function from G G to A is called a 2-cochain on G with values in A. Any 2-cochain
satisfying (††) is called a 2-cocycle with coefficients in A.
Q
Therefore, (†) is an associative multiplication in A G iff f is a 2-cocycle with values in A.
Q
Does A G with multiplication (†) have an identity?
The original group, G, has identity 1 and we have 1 = u−1 −1
1 u1 , where u1 ∈ A, and so, u1 ∈ A. For all
b ∈ A and all τ ∈ G, we have
(u−1
1 u1 )(buτ ) = buτ ,
which yields
u−1 −1
1 (1 · b)f (1, τ )uτ = u1 bf (1, τ )uτ = buτ .
Since A is abelian, we get
f (1, τ ) = u1 ,
which shows that f (1, τ ) is independent of τ . In particular, u1 = f (1, 1).
Question: Is (††) sufficient to imply that f (1, τ ) = f (1, 1) for all τ ∈ G?
In (††), take σ = 1. We get
f (1, τ )f (τ, ρ) = f (τ, ρ)f (1, τ ρ).
Again, since A is abelian, we deduce that f (1, τ ) = f (1, τ ρ). If we take τ = 1, we get f (1, 1) = f (1, ρ), for
all ρ.
Q
Therefore, (††) is sufficient and A G has an identity 1 = hf (1, 1)−1 , 1i, or in additive notation (since
A is abelian),
1 = h−f (1, 1), 1i. (∗)

Finally, what about inverses? Once again, go back to our original G.

We have (auσ )−1 = u−1
σ a
−1
. Now,

u−1
σ = (uσ )
−1
= σ −1 = uσ−1 .
Therefore, there is some α ∈ A so that u−1
σ = αuσ −1 . By multiplying on the right by uσ , we get

1 = αuσ−1 uσ = αf (σ −1 , σ)uσσ−1 = αf (σ −1 , σ)u1 = αf (σ −1 , σ)f (1, 1),

since u1 = f (1, 1). So, α = f (1, 1)−1 f (σ −1 , σ)−1 . Consequently, we get
(auσ )−1 = u−1
σ a
−1

= αuσ−1 a−1
= α(uσ−1 a−1 u−1
σ −1 )uσ
−1

= α(σ −1 · a−1 )uσ−1

= f (1, 1)−1 f (σ −1 , σ)−1 (σ −1 · a−1 )uσ−1
= f (1, 1)−1 f (σ −1 , σ)−1 (σ −1 · a)−1 uσ−1
= ((σ −1 · a)f (σ −1 , σ)f (1, 1))−1 uσ−1 .
1.4. GROUP EXTENSIONS 79

Q
Therefore, in A G (switching to additive notation since A is abelian), inverses are given by

ha, σi−1 = h−σ −1 · a − f (σ −1 , σ) − f (1, 1), σ −1 i. (∗∗)

Q
We find that A G can be made into a group via (†), provided f (σ, τ ) satisfies (††). The formulae
Q (∗)
and (∗∗) give the unit element and inverses, respectively. For temporary notation, let us write (A G; f )
for this group. Also, since A is abelian, let us rewrite (††) in additive notation, since this will be more
convenient later on:

σ · f (τ, ρ) + f (σ, τ ρ) = f (στ, ρ) + f (σ, τ ). (††)

Go back to the original group, G, and its set-theoretic section s : G → G (with s(σ) = uσ ). We might have
chosen another set-theoretic section, t : G → G, namely, t(σ) = vσ . We get a 2-cocycle g(σ, τ ) = vσ vτ (vστ )−1 ,
i.e., vσ vτ = g(σ, τ )vστ .
What is the relation between f and g?
We know that vσ = σ = uσ , which implies that there is some k(σ) ∈ A with vσ = k(σ)uσ . Then, we have

vσ vτ = g(σ, τ )vστ = g(σ, τ )k(στ )uστ ,

and also
vσ vτ = k(σ)uσ k(τ )uτ = k(σ)(σ · k(τ ))uσ uτ = k(σ)(σ · k(τ ))f (σ, τ )uστ .
By equating these expressions, we get

g(σ, τ )k(στ ) = k(σ)(σ · k(τ ))f (σ, τ ).

But A is abelian, so we can write the above

g(σ, τ ) − f (σ, τ ) = σ · k(τ ) − k(στ ) + k(σ). (∗)

Observe that k : G → A is a function of one variable on G. We call k a 1-cochain on G with values in A.

For a 1-cochain, define a corresponding 2-cochain, called its coboundary, δk, by

(δk)(σ, τ ) = σ · k(τ ) − k(στ ) + k(σ).

Remarks:

(1) Every coboundary of a 1-cochain is automatically a 2-cocycle (DX).

(2) Cocycles form a group under addition of functions denoted by Z 2 (G, A). The special 2-cocycles which
are coboundaries (of 1-cochains) form a group (DX) denoted by B 2 (G, A). Item (1) says that B 2 (G, A)
is a subgroup of Z 2 (G, A).

(3) The quotient group, Z 2 (G, A)/B 2 (G, A), denoted H 2 (G, A), is the second cohomology group of G with
coefficients in A.

(4) Equation (∗) above says: If we change the choice of section from s to t, the corresponding cocycles, f
and g, are cohomologous, i.e., g − f = δk, i.e., the image of f in H 2 (G, A) is the same as the image of
g in H 2 (G, A). Thus, it is the cohomology class of f which is determined by (E).
80 CHAPTER 1. GROUP THEORY

Q Q
Now, make (A G; f ). Then, we can map A into (A G; f ) via

a 7→ ha − f (1, 1), 1i.

Q
Claim. The set {ha − f (1, 1), 1i | a ∈ A} is a subgroup of (A G; f ). In fact, it is a normal subgroup and
the quotient is G.
Proof . We have

ha − f (1, 1), 1ihb − f (1, 1), 1i = ha − f (1, 1) + b − f (1, 1) + f (1, 1), 1i = ha + b − f (1, 1), 1i,

and thus, the map λ : a 7→ ha − f (1, 1), 1i is a group homomorphism. We leave the rest as a (DX).
Q Q
Qf − g = δk, i.e.,
Say Q f and g are cohomologous, and make (A G; f ) and (A G; g). Consider the map
θ : (A G; f ) → (A G; g) given by
θ : ha, σi 7→ ha + k(σ), σi.
We claim that θ is a homomorphism. Since

ha, σihb, τ i = ha + σ · b + f (σ, τ ), στ i,

we have
θ(ha, σihb, τ i) = ha + σ · b + f (σ, τ ) + k(στ ), στ i.
We also have

θ(ha, σi)θ(hb, τ i) = ha + k(σ), σihb + k(τ ), τ i)

= ha + k(σ) + σ · b + σ · k(τ ) + g(σ, τ ), στ i).

In order for θ to be a homomorphism, we need

k(σ) + σ · k(τ ) + g(σ, τ ) = f (σ, τ ) + k(στ ),

Q
Qis, f − g = δk. Consequently, θ is a homomorphism, in fact, an isomorphism. Moreover, (A
that G; f ) and
(A G; g) fit into two extensions and we have the following diagram:

(E)f 0 /A / (A Q G; f ) /G /0

θ
Q
(E 0 )g 0 /A / (A G; g) /G / 0.

The rightmost rectangle commutes,

Q but we need to check that the leftmost rectangle commutes. Going over
horizontally and down from (A G; f ), for any a ∈ A, we have

a 7→ ha − f (1, 1), 1i 7→ ha − f (1, 1) + k(1), 1i,

and going horizontally from the lower A, we have

a 7→ ha − g(1, 1), 1i.

For the rectangle to commute, we need: g(1, 1) = f (1, 1) − k(1). However,

f (σ, τ ) = g(σ, τ ) + δk(σ, τ ) and δk(σ, τ ) = σ · k(τ ) − k(στ ) + k(σ). If we set σ = τ = 1, we get

δk(1, 1) = k(1) − k(1) + k(1) = k(1),

and it follows that g(1, 1) = f (1, 1) − k(1), as desired.

1.4. GROUP EXTENSIONS 81

Hence, cohomologous 2-cocycles give rise to equivalent group extensions (the action is fixed). Conversely,
we now show that equivalent group extensions give rise to cohomologous 2-cocycles. Say

(E) 0 /A /G /G /0

ψ

(E 0 ) 0 /A / G0 /G / 0.

is an equivalence of extensions (i.e., the diagram commutes). We know, up to the notion of being cohomol-
ogous, that we may adjust both cocycles f and g associated with (E) and (E 0 ) by choice of sections. In
both cases, take u1 = 0 (since we are using additive notation). Therefore, f (1, 1) = g(1, 1) = 0. From the
commutativity of the diagram, ψ must be of the form

ψha, σi = hϕ(a, σ), σi

Q
for some function ϕ : A G → A. By the above choice, the maps A −→ G and A −→ G 0 are given by
a 7→ ha, 1i in both cases. Therefore,
ψ(a, 1) = hϕ(a, 1), 1) = (a, 1), and so,

ϕ(a, 1) = a, for all a ∈ A.

Since ψ is a homomorphism, we have

ψ(ha, σihb, τ i) = ψ(ha, σi)ψ(hb, τ i),

and this yields an identity relating f , g and ϕ. The left hand side of the above equation is equal to

ψ(ha + σ · b + f (σ, τ ), στ i) = hϕ(a + σ · b + f (σ, τ ), στ i), στ i,

and the righthand side is equal to

hϕ(a, σ), σihϕ(b, τ ), τ i = hϕ(a, σ) + σ · ϕ(b, τ ) + g(σ, τ ), στ i,

and by equating them, we get

ϕ(a + σ · b + f (σ, τ ), στ i) = ϕ(a, σ) + σ · ϕ(b, τ ) + g(σ, τ ). (†††)

By taking τ = 1 (using the fact that ϕ(b, 1) = b), we get

ϕ(a + σ · b + f (σ, 1), σi) = ϕ(a, σ) + σ · b + g(σ, 1). (∗∗∗)

Now, (††) can be written as

σ · f (τ, ρ) − f (στ, ρ) + f (σ, τ ρ) − f (σ, τ ) = 0.

If we take ρ = 1, we get
σ · f (τ, 1) − f (στ, 1) + f (σ, τ ) − f (σ, τ ) = 0.
which yields
σ · f (τ, 1) = f (στ, 1).
If we take τ = 1, we get σ · f (1, 1) = f (σ, 1), but f (1, 1) = 0, and so,

f (σ, 1) = 0.

Consequently, (∗∗∗) yields

ϕ(a + σ · b, σ) = ϕ(a, σ) + σ · b.
82 CHAPTER 1. GROUP THEORY

Writing b = σ −1 · c, we get
ϕ(a + c, σ) = ϕ(a, σ) + c, for all a, c ∈ A.
In particular, when a = 0, we get ϕ(c, σ) = ϕ(0, σ)+c. Let ϕ(0, σ) = k(σ). Now, if we use ϕ(a, σ) = ϕ(0, σ)+a
in (†††), we get
a + σ · b + f (σ, τ ) + k(στ ) = a + k(σ) + σ · (b + k(τ )) + g(σ, τ ),
which yields
f (σ, τ ) + k(στ ) = g(σ, τ ) + k(σ) + σ · k(τ ),
that is, f − g = δk. Hence, we have proved almost all of the following fundamental theorem:

Theorem 1.27 If G and A are groups and A is abelian, then each group extension
π
(E) 0 −→ A −→ G −→ G −→ 0

makes A into a G-module; the G-module structure is the type of (E) and equivalent extensions have the same
type. For a given type, the equivalence classes of extensions of G by A are in one-to-one correspondence
with H 2 (G, A), the second cohomology group of G with coefficients in A. Hence, the distinct extensions of
G by A (up to equivalence) are classified by the pairs (type(E), χ(E)), where χ(E) is the cohomology class
in H 2 (G, A) corresponding to (E). In this correspondence, central extensions correspond to G-modules, A,
with trivial action ((E) is central iff A ⊆ Z(G)). An extension of any type splits iff χ(E) = 0 in H 2 (G, A).
((E) is split iff there is a group homomorphism s : G → G so that π ◦ s = id).

Proof . We just have to prove the last two facts. Note that the type of extension is trivial iff

(∀σ ∈ G)(∀a ∈ A)(σ · a = a)

iff
(∀x ∈ G)(∀a ∈ A)(x−1 ax = a)
iff
(∀x ∈ G)(∀a ∈ A)([x, a] = 1)
iff A ⊆ Z(G).
Finally, the cohomology is trivial iff every cocycle is a coboundary iff every cocycle is cohomologous to
0 iff in (E) there is a map σ 7→ uσ with f (σ, τ ) = 0. Such a map is a homomorphism. Thus, χ(E) = 0 in
H 2 (G, A) iff (E) has a splitting.

Examples. (I) Find all extensions

0 −→ Z −→ G −→ Z/2Z −→ 0.

There are several cases to consider depending on the type and the cohomology class of the extension.
(a) Trivial type (the action of Z/2Z on Z is trivial).
Q
(a1) Split extension. We get G f→ Z (Z/2Z).
(a2) Nonsplit extensions. In this case, we have to compute H 2 (Z/2Z, Z) (trivial action). We know from
previous work that (up to cohomology) we can restrict ourselves to normalized cochains, f (σ, τ ), i.e., cochains
such that
f (σ, 1) = f (1, σ) = 0.
Elements in Z/2Z are ±1. We need to know what f (−1, −1) is. The reader should check that the co-
cycle condition, δf = 0, gives no condition on the integer f (−1, −1), and thus, we have an isomorphism
Z 2 (Z/2Z, Z) ∼
= Z.
1.4. GROUP EXTENSIONS 83

What about coboundaries: f = δk? Such k’s are also normalized, and so, k(1) = 0. We have k(−1) = b,
for any b ∈ Z. Since
δk(σ, τ ) = σ · k(τ ) − k(στ ) + k(σ),
using the fact that the action is trivial and that k(1) = 0, we get

δk(−1, −1) = (−1) · k(−1) − k(1) + k(−1) = k(−1) + k(−1) = 2b.

So, we can adjust f , up to parity by coboundaries, and H 2 (Z/2Z, Z) ∼

= Z/2Z. Consequently, we have exactly
one nonsplit, trivial-type extension
G = {(n, ±1) | n ∈ Z}.
The group operation is given by

(n, ±1)(m, 1) = (n + m, ±1)

(n, 1)(m, ±1) = (n + m, ±1)
(n, −1)(m, −1) = (n + m + 1, 1),

where in this last equation, we assumed without loss of generality that f (−1, −1) = 1.
(b) Nontrivial type. We need a nontrivial map Z/2Z −→ Aut(Z). Since Z is generated by 1 and −1,
there is only one nontrivial action:
(−1) · n = −n.
(Recall that 1 · n = n, always).
(b1) The split, nontrivial type extension. In this case

G = {(n, σ) | n ∈ Z, σ ∈ Z/2Z},

with multiplication given by

(n, σ)(m, τ ) = (n + σ · m, στ ).
Now, consider the map  
σ n
(n, σ) 7→ .
0 1
Observe that matrix multiplication yields
    
σ n τ m στ n+σ·m
= .
0 1 0 1 0 1

Therefore, G is isomorphic to the group of matrices

 
σ n
0 1

under matrix product. This is a nonabelian group, it is infinite and we claim that G is solvable with δ(G) = 2.
Indeed, we have G/Z ∼
= Z/2Z, an abelian group, and so ∆(1) (G) ⊆ Z. So,
(2) (1)
∆ (G) ⊆ ∆ (Z) = {0}, and we conclude that δ(G) = 2.
(b2) Nonsplit, nontrivial type extension. We need to figure out what the cocycles are in order to compute
H 2 (Z/2Z, Z). By the same reasoning as before, we need to know what is f (−1, −1). We know that δf (σ, τ ) =
0. So, we have
σ · f (τ, ρ) − f (στ, ρ) + f (σ, τ ρ) − f (σ, τ ) = 0.
84 CHAPTER 1. GROUP THEORY

Let τ = ρ = −1 in the above equation. We get

σ · f (−1, −1) − f (−σ, −1) + f (σ, 1) − f (σ, −1) = σ · f (−1, −1) − f (−σ, −1) − f (σ, −1) = 0,

since f (σ, 1) = 0. If we let σ = −1, since f (1, −1) = 0, we get

−f (−1, −1) − f (−1, −1) = 0,

and so, 2f (−1, −1) = 0. Since f (−1, −1) ∈ Z, we get f (−1, −1) = 0. Therefore, f ≡ 0 and the cohomology
is trivial: H 2 (Z/2Z, Z) = (0) (for nontrivial action).
As a conclusion, there exist three extension classes and three distinct groups, two of them abelian, the
third solvable and faithfully representable by matrices.
(II) Let V be a finite dimensional vector space and consider V + as additive group. Let G = GL(V ) and
let the action of G on V be the natural one (i.e, for any ϕ ∈ GL(V ) and any v ∈ V , ϕ · v = ϕ(v)). We have
the split extension
0 → V → G  GL(V ) → 0.
The group, G, in the above exact sequence is the affine group of V .
(III) Again, we restrict ourselves to split extensions. Let A be any abelian group and let n ∈ N. The
group Y Y Y
A A ··· A
| {z }
n

is acted on by the symmetric group, Sn , simply by permuting the factors. We have a split extension
Y Y Y
0→A A ··· A → G  Sn → 0.
| {z }
n

The group, G, is called the wreath product of A by Sn and is denoted A o Sn . We denote the split extension
of a given type of G by A by A \\
// G (note that this notation does not refer to the action).
Here are some useful facts on cohomology:

(1) If G is arbitrary and A is n-torsion, which means that nA = 0, then H 2 (G, A) is n-torsion.

(2) If G is a finite group, say #(G) = g and A is arbitrary, then H 2 (G, A) is g-torsion (this is not trivial
to prove!).

(3) Suppose that A is n-torsion and G is finite, with #(G) = g, and suppose that (g, n) = 1. Then,
H 2 (G, A) = (0). (This is a clear consequence of (1) and (2).)

(4) Suppose that G is finite. We can define a homomorphism (of G-modules) A −→ A, called the G-norm
and denoted NG (we will usually drop the subscript G), defined by
X
NG (a) = σ · a.
σ∈G

Moreover, assume that G is a finite cyclic group. Then, for any A, there is an isomorphism

AG /N A ∼
= H 2 (G, A),

where
AG = {a ∈ A | σ · a = a, for all σ ∈ G}.
1.4. GROUP EXTENSIONS 85

Here is an example of how to use the above facts.

(IV) Find all the groups of order pq (with p, q prime and 0 < p < q).
We know that the q-Sylow subgroup is normal, namely, it is Z/qZ = A C G, and
G = G/A = Z/pZ. Therefore, whatever G is, it fits in the group extension

0 −→ Z/qZ −→ G −→ Z/pZ −→ 0.

By (3), since (p, q) = 1, we have H 2 (G, A) = (0). So, we only have split extensions. What is Aut(Z/qZ)?
Clearly, it is Z/(q − 1)Z. So, we have to consider the homomorphisms

Z/pZ −→ Aut(Z/qZ) = Z/(q − 1)Z. (∗)

If (∗) is non-trivial, then p | (q − 1), i.e., q ≡ 1 (mod p). So, if q 6≡ 1 (mod p), then we have trivial action and
we find that Y
G∼ = (Z/qZ) (Z/qZ) ∼ = Z/pqZ.

If q ≡ 1 (mod p), we also can have trivial action, and we get Z/pqZ, again. So, we now consider nontrivial
actions. The unique cyclic group of order p in Z/(q − 1)Z is generated by λ q−1
p , where λ = 1, 2, . . . , p − 1.
If we send 1 ∈ Z/pZ to λ q−1
p , the corresponding action is

q−1
n 7→ nλ (mod q).
p

Thus, there are p − 1 nontrivial (split) group extensions, (Eλ ), with central groups

Gλ = {(n, ζ m ) | 0 ≤ m ≤ p − 1}

(here the elements of Z/pZ are 1, ζ, ζ 2 , . . . , ζ p−1 ) and multiplication given by

 
m s q − 1 m+s
(n, ζ )(r, ζ ) = (n + rmλ ,ζ .
p

Consider the map Gλ −→ G1 given by

(n, ζ m ) 7→ (m, ζ λm ).
This is a group isomorphism. So, here we have all inequivalent extensions, (Eλ ), with p − 1 different actions,
yet the groups Gλ are mutually isomorphic. Thus, G1 and Z/pqZ are the two groups of order pq when
q ≡ 1 (mod p).
The second cohomology group, H 2 (G, A), has appeared naturally in the solution to the group extension
problem. Consequently, it is natural at this stage to define cohomology groups in general.
Q
The set up is: We have a group, G, and a G-action, G A −→ A, where A is an abelian group. For
every n ∈ N, we define
Y Y
C n (G, A) = {f : G ··· G → A},
| {z }
n
Y Y
where G ··· G is the product of G with itself n times (in the category of sets). By convention, when
| {z }
n
n = 0, this set product is the one point set, {∗}. The set C n (G, A) is an abelian group under addition
of functions (e.g, f + g is the function defined by (f + g)(x) = f (x) + g(x) for all x ∈ G). The group
86 CHAPTER 1. GROUP THEORY

C n (G, A) is called the group of n-cochains of G with coefficients in A. We define the coboundary map,
δn : C n (G, A) → C n+1 (G, A), for every n ≥ 0, by the formula:
n
X
(δn f )(σ1 , . . . , σn+1 ) = σ1 · f (σ2 , . . . , σn+1 ) + (−1)j f (σ1 , . . . , σj−1 , σj σj+1 , σj+2 , . . . , σn+1 )
j=1
n+1
+ (−1) f (σ1 , . . . , σn ),

for all f ∈ C n (G, A) and all σ1 , . . . , σn+1 ∈ G.

(1) Check (DX): For all n ≥ 0,
δn (δn−1 f ) ≡ 0.
(By convention, δ−1 = 0).
(2) Set Z n (G, A) = Ker δn , a subgroup of C n (G, A), the group of n-cocycles of G with coefficients in
A. We also let B n (G, A) = Im δn−1 , a subgroup of C n (G, A), the group of n-coboundaries of G with
coefficients in A. Observe that since δ−1 = 0, we have B 0 (G, A) = (0). Furthermore, (1) implies that
B n (G, A) ⊆ Z n (G, A), for all n ≥ 0.
(3) Set H n (G, A) = Z n (G, A)/B n (G, A); this is the nth cohomology group of G with coefficients in A.

Examples. (i) Case n = 0: Then, B 0 = (0). The functions, f , in C 0 (G, A) are in one-to-one correspondence
with the elements f (∗) of A, and so, C 0 (G, A) = A. Note that for any σ ∈ G, if f ∈ C 0 (G, A) corresponds
to the element a in A, we have

(δ0 f )(σ) = σ · f (∗) − f (∗) = σ · a − a.

Thus,
Z 0 (G, A) = {a ∈ A | δ0 (a) = 0} = {a ∈ A | σ · a = a, for all σ ∈ G} = AG .
So, we also have H 0 (G, A) = AG .
(ii) Case n = 1: Then, C 1 (G, A) is the set of all functions f : G → A. For any
f ∈ C 1 (G, A), we have
(δ1 f )(σ, τ ) = σ · f (τ ) − f (στ ) + f (σ).
It follows that

Z 1 (G, A) = {f ∈ C 1 (G, A) | δ1 f = 0} = {f ∈ C 1 (G, A) | f (στ ) = σ · f (τ ) + f (σ)}.

This is the set of crossed (or twisted ) homomorphisms from G to A.

Remark: If A has trivial G-action, then Z 1 (G, A) = HomGr (G, A).

We have B 1 (G, A) = Im δ0 = all functions, g, so that g(σ) = (δ0 (a))(σ) = σ · a − a, for some a ∈ A.
Such objects are twisted homomorphisms, called principal (or inner ) twisted homomorphisms.

Remark: If A has trivial G-action, then B 1 (G, A) = (0). So, H 1 (G, A) is the quotient of the twisted
homomorphisms modulo the principal twisted homomorphisms if the action is nontrivial, and H 1 (G, A) =
HomGr (G, A) if the action is trivial.
(iii) Case n = 2: We have already encountered thisQcase in dealing with group extensions. We content
ourselves with computing δ2 . Since C 2 (G, A) = {f : G G → A}, we have

(δ2 f )(σ, τ, ρ) = σ · f (τ, ρ) − f (στ, ρ) + f (σ, τ ρ) − f (σ, τ ).

1.4. GROUP EXTENSIONS 87

We note that Z 2 (G, A) gives us back the group of “old” 2-cocycles, B 2 (G, A) gives us back the group of
“old” 2-coboundaries, and H 2 (G, A) is in one-to-one correspondence with the equivalence classes of group
extensions of a fixed type.

Remark: Given a group, G, Eilenberg and Mac Lane (1940’s) constructed a topological space, K(G, 1),
unique up to homotopy type, with the following properties:

G if n = 1
πn (K(G, 1)) =
(0) if n 6= 1.

Fact: If we compute the integral cohomology of K(G, 1), denoted H n (K(G, 1), Z), we get

H n (K(G, 1), Z) ∼
= H n (G, Z).

Here, the G-action on Z is trivial.

88 CHAPTER 1. GROUP THEORY

1.5 Solvable and Nilpotent Groups

Given a group, G, its derived series,

G = ∆(0) (G) ⊇ ∆(1) (G) ⊇ ∆(2) (G) ⊇ · · · ⊇ ∆(t) (G) ⊇ · · · ,

may decrease very quickly, and even though the solvable groups (those for which the derived series reaches
{1} after finitely many steps, i.e., those for which δ(G) is finite) are not as “wild” as groups for which
δ(G) = ∞, it desirable to delineate families of groups with an even “nicer” behavior. One way of doing
so is to define descending (or ascending) chains that do not decrease (or increase) too quickly and then to
investigate groups whose chains are finite. The collection of nilpotent groups is such a family of groups, and,
moreover, nilpotent groups tend to show up as fundamental groups of spaces arising naturally in geometry.
Every nilpotent group is solvable and solvability is inherited by subgroups and quotient groups, as shown in
the following proposition:

Proposition 1.28 If G is a group and G is solvable, then for every subgroup, H, of G, the group, H, is
solvable. Moreover, if H is normal in G, then G/H is solvable. In fact, for both groups, δ(either) ≤ δ(G).
Conversely, say G possesses a normal subgroup, H, so that both H and G/H are solvable. Then, G is
solvable. In fact, δ(G) ≤ δ(H) + δ(G/H).

Proof . Let G be solvable. Then, H ⊆ G implies ∆(1) (H) ⊆ ∆(1) (G); therefore (by induction),

∆(j) (H) ⊆ ∆(j) (G),

and we deduce that δ(H) ≤ δ(G). Consider G = G/H when H C G. Then, [x, y] = [x, y] and this implies
∆(1) (G) = ∆(1) (G). Hence (by induction),

∆(j) (G) = ∆(j) (G).

Therefore, δ(G) ≤ δ(G).

Conversely, assume that H and G/H are solvable (with H C G). We have ∆(j) (G) = ∆(j) (G) and if
j ≥ δ(G), then ∆(j) (G) = {1}, which implies that ∆(j) (G) ⊆ H. So, ∆(k+j) (G) ⊆ ∆(k) (H), and the latter
is {1} if k = δ(H). Therefore,
∆(δ(G)+δ(H)) (G) = {1},
and so, δ(G) ≤ δ(H) + δ(G/H).

Proposition 1.29 Let (P) be some property of finite groups. Assume that (P) satisfies:
(a) The trivial group has (P), every cyclic group of prime order has (P).
(b) Suppose G has (P), then H C G implies H and G/H have (P).
(c) If G has (P) (with G 6= {1}), then G is not simple unless G is cyclic of prime order.
Then, when G has (P), the group G is solvable.

Proof . We proceed by induction on #(G). The case G = {1} is trivial, by (a) (nothing to check). Assume
that the proposition holds for all G with #(G) ≤ n, and assume #(G) = n + 1. If n + 1 is prime, then G is
cyclic of prime order, which implies that it is solvable. Thus, we may assume that n + 1 is not prime and
that G has (P). By (c), the group G has some nontrivial normal subgroup, H. By (b), both H and G/H
have (P), and the induction hypothesis implies that both H and G/H are solvable. Proposition 1.28 implies
that G is solvable.

Corollary 1.30 (Burnside, Feit & Thompson) Every group G, of order pa q b or odd order is solvable.
1.5. SOLVABLE AND NILPOTENT GROUPS 89

Remark: Corollary 1.30 is not really proved. It depends on establishing (c) for the two properties: pa q b ,
odd order. As remarked just before Proposition 1.10, this is not easy.
Definition 1.9 Let G be any group. The lower central series (LCS) of G is the descending chain of
subgroups
G = Γ0 ⊇ Γ1 ⊇ · · · ⊇ Γd ⊇ · · · ,
where Γj+1 = [G, Γj ]. The upper central series (UCS) of G is the ascending chain of subgroups
{1} = Z0 ⊆ Z1 ⊆ Z2 ⊆ · · · ⊆ Zd ⊆ · · · ,
where Zj = the inverse image in G of Z(G/Zj−1 ).

Remarks:
(1) Γ1 (G) = [G, Γ0 ] = [G, G] = ∆(1) (G), and
Γ2 (G) = [G, Γ1 ] = [G, ∆(1) (G)] ⊇ [∆(1) (G), ∆(1) (G)] = ∆(2) (G),
and so, Γ2 (G) ⊇ ∆(2) (G). The reader should check (DX) that Γd (G) ⊇ ∆(d) (G), for all d ≥ 0.
(2) Z1 (G) = inverse image in G of Z(G/Z0 ) = inverse image of Z(G), so Z1 (G) = Z(G).
(3) If for some j, the equality Γj (G) = Γj+1 (G) holds, then Γj (G) = Γd (G), for all d ≥ j. The lower
central series strictly descends until the first repetition.
(4) Similarly, if for some j, the equality Zj (G) = Zj+1 (G) holds, then Zj (G) = Zd (G), for all d ≥ j. The
upper central series strictly ascends until the first repetition.
Proposition 1.31 Suppose the lower central series of G reaches {1} after r steps. Then, for every j ≤ r,
we have Γr−j ⊆ Zj . Consequently, the upper central series reaches G after r steps. Conversely, suppose that
the upper central series reaches G after r steps. Then, for every j ≤ r, we have Γj ⊆ Zr−j . Consequently,
the lower central series reaches {1} after r steps.
Proof . By induction on j. For j = 0, we have Γr = Γr−0 , and by hypothesis, Γr = {1} and Z0 = {1}, so the
basis of the induction holds. Before we do the induction step, let us also consider the case j = 1. We need
to show that Γr−1 ⊆ Z1 = Z(G). But Γr = {1}, yet Γr = [G, Γr−1 ]. This means that for all σ ∈ G and all
τ ∈ Γn−1 , we have [σ, τ ] ∈ Γr = {1}. Thus, τ commutes with all σ ∈ G, and so, τ ∈ Z(G) = Z1 . Let us now
assume our statement, Γr−j ⊆ Zj , for some j, and look at the case j + 1. Now, Γr−j = [G, Γr−j−1 ]. By the
induction hypothesis,
[G, Γr−j−1 ] ⊆ Zj .
Consider the map G −→ G/Zj = G. Then,
[G, Γr−j−1 ] = {1} in G.
Therefore, Γr−j−1 is contained in the inverse image of Z(G) = Z(G/Zj ) = Zj+1 , concluding the induction
step.
For the converse, again, use induction on j. When j = 0, we have Γ0 = G and Zr = Zr−0 = G, by
hypothesis, and the basis of the induction holds. Assume that Γj ⊆ Zr−j for some j, and consider the case
j + 1. We have
Γj+1 = [G, Γj ] ⊆ [G, Zr−j ],
by the induction hypothesis. Look at the map G −→ G/Zr−j−1 = G. We have
Γj+1 ⊆ [G, Z r−j ].
But, by definition, Z r−j = Z(G). Thus, [G, Z r−j ] = {1} in G. Therefore,
Γj+1 ⊆ Ker (G −→ G) = Zr−j−1 .
90 CHAPTER 1. GROUP THEORY

Definition 1.10 A group, G, is nilpotent if and only if the lower central series reaches {1} after finitely
many steps. The smallest number of steps, say c, is the nilpotence class of G. We write G ∈ N ilp(c). (We
let c = ∞ if the LCS does not reach {1} in finitely many steps.)

Remarks:

(1) N ilp(0) = the class consisting only of the trivial group.

N ilp(1) = the collection of abelian, nontrivial groups. If we let N ilp(c) denote the union of the
collections N ilp(k) for k = 0, . . . , c, then it turns out that we have a strictly ascending chain

Ab = N ilp(1) < N ilp(2) < N ilp(3) < · · ·

of “worse and worse behaved” groups.

(2) We have G ∈ N ilp(c) iff the UCS reaches G after c steps and c is minimal with this property.

(3) Each nilpotent group is automatically solvable, but the converse is false, even for finite groups, even for
small finite groups. Indeed, we observed earlier that ∆(r) (G) ⊆ Γr (G). Therefore, δ(G) ≤ nilpotence
class of G. For a counter-example, take G = S3 . This group has order 6, its center is trivial, and so
Z1 = Z0 and G is not nilpotent. Yet, we have an exact sequence

0 −→ Z/3Z −→ S3 −→ Z/2Z −→ 0,

and the extremes are solvable (even nilpotent, even abelian), so the middle is solvable.

(4) Every p-group is nilpotent. This is because the center of a p-group is nontrivial, so the UCS is strictly
ascending and our group is finite; so, this implies that our group is nilpotent.

Remark: The fundamental groups of many spaces arising in geometry tend to be nilpotent groups.

Proposition 1.32 (Modified Sylow III) Say G is a finite group, P is a p-Sylow subgroup of G and H is
some subgroup of G. If H ⊇ NG (P ), then NG (H) = H.

Proof . (Frattini Argument). Pick σ ∈ NG (H). Then, σHσ −1 = H and σP σ −1 ⊆ σHσ −1 (since H ⊇
NG (P )). So, P and σP σ −1 are two p-Sylow subgroups of H, and by Sylow II, there is some τ ∈ H so that
τ P τ −1 = σP σ −1 . Thus, τ −1 σP (τ −1 σ)−1 = P , and so, τ −1 σ ∈ NG (P ) ⊆ H, by hypothesis. So, σ ∈ τ H = H
(since τ ∈ H).

Theorem 1.33 Let G be a finite group. Then, the following statements are equivalent:

(1) G is nilpotent.

(2) G has property (N).

(3) Every maximal subgroup of G is normal.

(4) ∆(1) (G) ⊆ Φ(G).

(5) Every p-Sylow subgroup of G is normal in G.

Q
(6) G is isomorphic to the product of its p-Sylow subgroups. (We write G ∼
= p Gp ).
1.5. SOLVABLE AND NILPOTENT GROUPS 91

Proof . (1) ⇒ (2). Let H be a proper subgroup of G, we must prove that NG (H) > H. Now, there is some
c with Γc = {1}. Obviously, Γc ⊆ H, so pick a smallest d for which Γd ⊆ H, so that Γd−1 6⊆ H.
Claim: Γd−1 ⊆ NG (H).
If the claim holds, then H < NG (H), i.e., G has property (N). Pick ξ ∈ Γd−1 ; so,

[H, ξ] ⊆ [H, Γd−1 ] ⊆ [G, Γd−1 ] = Γd .

Pick h ∈ H and look at [h−1 , ξ]. The element [h−1 , ξ] is in Γd , and so, in H (since Γd ⊆ H). Consequently,
h−1 ξhξ −1 ∈ H, from which we deduce ξhξ −1 ∈ H, and since this is true for all h ∈ H, we have ξ ∈ NG (H),
as desired.
(2) ⇒ (3). This has already been proved (c.f. Proposition 1.17).
(3) ⇒ (4). This has already been proved (c.f. Proposition 1.23).
(4) ⇒ (5). Let P be a p-Sylow subgroup of G. Look at NG (P ). If NG (P ) 6= G, then NG (P ) is contained
is some maximal subgroup, M . By modified Sylow III, we get NG (M ) = M . Now, ∆(1) (G) ⊆ Φ(G) ⊆ M , by
hypothesis, and the second homomorphism theorem implies that M corresponds to a subgroup of G/∆(1) (G)
and normal subgroups correspond to normal subgroups. Yet, G/∆(1) (G) is abelian, so all its subgroups are
normal, which implies that M is normal, a contradiction.
(5) ⇒ (6). This has already been proved (c.f. Proposition 1.18).
(6) ⇒ (1). Since every p-group is nilpotent, the implication (6) ⇒ (1) follows from the following

Proposition 1.34 Say Gj ∈ N ilp(cj ), for j = 1, . . . , t. Then,

t
Y
Gj ∈ N ilp( max {cj }).
1≤j≤t
j=1

Proof . An obvious induction reduces us to the case t = 2. In this case, we use an induction on max{c1 , c2 }.
The cases c1 ≤ 1 and c2 ≤ 1 are trivial. Now, we have (DX)
Y Y
Z(G1 G2 ) ∼
= Z(G1 ) Z(G2 ).
Q Q Q
But then, (G1 G2 )/Z(G1 G2 ) ∼ = (G1 /Z(G1 )) (G2 /Z(G2 )); on the left hand side, the purported nilpo-
tence class is down by 1 and on the righthand side, both are down by 1. We conclude by applying the
induction hypothesis.
This concludes the proof of Theorem 1.33.
92 CHAPTER 1. GROUP THEORY

1.6 Ω-Groups and the Jordan-Hölder-Schreier Theorem

Let Ω be some set. If M is a group, we denote the monoid of group endomorphisms of M (under composition)
by EndGr (M ) and the group of (group) automorphisms of M by AutGr (M ).

Definition 1.11 A group, M , is an Ω-group iff there exists a set map Ω −→ EndGr (M ). If Ω is itself a
group, we demand that our map be a homomorphism (so, the image lies in AutGr (M )). If Ω is a ring, we
demand that M be an abelian group and that our map be a ring homomorphism taking 1 ∈ Ω to the identity
endomorphism of M .

Examples.
(1) When Ω is a group, we get an Ω-action on M (at first, as a set) and further, we obtain:

1 · m = m,
ξ · (η · m) = (ξη) · m,
ξ · (mn) = (ξ · m)(η · n).

In particular, (ξ · m)−1 = ξ · m−1 .

(2) When Ω is a group and M is abelian, we just get an Ω-module.
(3) If Ω is a ring, then the nomenclature is Ω-module instead of Ω-group.
(4) When Ω is a field, then an Ω-module is a vector space over Ω.
(5) Being an Z-module is equivalent to being an abelian group.
An Ω-subgroup of M (resp. Ω-normal subgroup of M ) is just a subgroup (resp. a normal subgroup), N ,
of M stable under Ω, i.e., for all ξ ∈ Ω, for all n ∈ N , we have ξ · n ∈ N .
Blanket Assertion (DX). The three isomorphism theorems of ordinary group theory are true for Ω-
groups provided everywhere “subgroup” appears we substitute “Ω-subgroup”, mutatis–mutandis for “normal
subgroups.”

Definition 1.12 A normal flag (normal series, normal chain) is a descending chain of Ω-subgroups of M :

M = M0 ⊇ M1 ⊇ M2 ⊇ · · · ⊇ Mr = {1}, (∗)

each Mj being normal in the preceding Mj−1 . A normal flag is nonrepetitious if for no j do we have
Mj = Mj−1 . Given a second normal flag:

M = M00 ⊇ M10 ⊇ M20 ⊇ · · · ⊇ Ms0 = {1}, (∗∗)

the flag (∗∗) refines (∗) iff for every i the Ω-group Mi occurs as some Mj0 . Two normal flags (∗) and (∗∗)
0
are isomorphic iff the collection of their successive quotients, Mi−1 /Mi and Mj−1 /Mj0 may be rearranged so
that, after rearrangement, they become pairwise isomorphic (in their new order). When this happens, the
lengths r and s are equal.

Theorem 1.35 (Schreier refinement theorem, 1928) For an Ω-group, any two normal flags possess isomor-
phic refinements. If both normal flags are nonrepetitious, so are their isomorphic refinements.

The main corollary of the Schreier refinement theorem is:

Corollary 1.36 (Jordan–Hölder theorem) Any two composition series for an Ω-group are isomorphic.
1.6. Ω-GROUPS AND THE JORDAN-HÖLDER-SCHREIER THEOREM 93

Proof . A composition series has no refinements except itself—apply Schreier’s theorem.

Zassenhaus proved a lemma specifically designed to give the smoothest proof of Schreier’s theorem—this
is
Lemma 1.37 (Zassenhaus’ butterfly lemma) Say G is an Ω-group and A and C are subgroups. Suppose
B C A and D C C are further Ω-subgroups. Then,
(A ∩ C)B/(A ∩ D)B ∼
= (C ∩ A)D/(C ∩ B)D.

A C

A∩C

B D
B∩C A∩D

Figure 1.1: The butterfly lemma

Proof . Let T = A ∩ C = C ∩ A, M = B ∩ C and N = A ∩ D. The conclusion of the lemma is

T B/N B ∼
= T D/M D.
First of all, there is right-left symmetry in the statement of the lemma and its conclusion (A ↔ C, B ↔ D;
under these substitutions, T ↔ T and M ↔ N ). We must prove that N B C T B. Pick t ∈ G and look at
tN Bt−1 = tN t−1 tBt−1 . If t ∈ A, then tBt−1 = B, since B C A. Thus, if t ∈ A then tN Bt−1 = tN t−1 B. If
t ∈ T ⊆ C, then as N = D ∩ C ∩ A = D ∩ T and D C C, we get
tN t−1 = tDt−1 ∩ tT t−1 = tDt−1 ∩ T = D ∩ T = N.
Thus, if t ∈ T then tN Bt−1 = N B.
Say ξ = tb ∈ T B. Since B C A and N ⊆ A, we have BN = N B. Then, we find
ξN Bξ −1 = tbN Bb−1 t−1
= tbN Bt−1
= tbBN t−1
= tBN t−1
= tN Bt−1
= N B.
Therefore, N B C T B. By symmetry, we get M D C T D. Look at T B/N B = T N B/BN (since N ⊆ T ). By
the third isomorphism theorem, we have
T B/N B ∼
= T /T ∩ N B.
94 CHAPTER 1. GROUP THEORY

By symmetry,
T D/N D ∼
= T /T ∩ M D.
If we prove that T ∩ N B = T ∩ N M (and so, T ∩ M D = T ∩ N M , by symmetry), we will be done.
Pick ξ ∈ T ∩ N B. We can write ξ = nb ∈ N B, so b = n−1 ξ ∈ N T = T (since N ⊆ T ). Thus,
b ∈ B ∩ T = B ∩ C ∩ A ⊆ M , and so, b ∈ M . Consequently, ξ = nb ∈ N M and since we also have ξ ∈ T ,
then ξ ∈ T ∩ N M . This proves that T ∩ N B ⊆ T ∩ N M . The reverse inclusion is trivial, since M ⊆ B.
Therefore, T ∩ N B = T ∩ N M , as claimed.
Proof of Theorem 1.35. Let

M = M0 ⊇ M1 ⊇ M2 ⊇ · · · Mi−1 ⊇ Mi ⊇ · · · ⊇ Mr = {1}, (∗)

and

M = M00 ⊇ M10 ⊇ M20 ⊇ · · · Mj−1

0
⊇ Mj0 ⊇ · · · ⊇ Ms0 = {1}, (∗∗)

be two normal nonrepetitious chains. Consider the groups

(j)
Mi−1 = (Mi−1 ∩ Mj0 )Mi .

(0) (s)
As j varies, these groups start at Mi−1 (= Mi−1 ) and end at Mi (= Mi−1 ) and we get a refinement of (∗) if
we do this between any pair in (∗). Also consider the groups
0
(i)
0
Mj−1 = (Mj−1 ∩ Mi )Mj0 ,
0
and let i vary. These groups interpolate between Mj−1 and Mj0 , just as above. Look at the successive
quotients
0 0
(j−1) (j) (i−1) (i)
Mi−1 /Mi−1 ; Mj−1 /Mj−1 . (†)
0
If we let A = Mi−1 , B = Mi (C A), C = Mj−1 and D = Mj0 (C C), we can write the first quotient group of
(†) as
(j−1) (j) 0
Mi−1 /Mi−1 = (Mi−1 ∩ Mj−1 )Mi /(Mi−1 ∩ Mj0 )Mi = (A ∩ C)B/(A ∩ D)B,
the left hand side of Zassenhaus’ lemma. By symmetry, the second quotient group of (†) is the righthand
side of Zassenhaus’ lemma and we are done.
1.7. CATEGORIES, FUNCTORS AND FREE GROUPS 95

1.7 Categories, Functors and Free Groups

Definition 1.13 A category, C, is a pair: hOb(C), Fl(C)i, in which Ob(C) and Fl(C) are classes, called the
objects of C and the morphisms (or arrows) of C, respectively. We require the following conditions:

(1) For all A, B ∈ Ob(C), there is a unique set, HomC (A, B), called the collection of morphisms from A
to B, and any two such are either disjoint or equal. Further
[
Fl(C) = HomC (A, B).
A,B

For the morphisms, we also require:

(2) For every u ∈ HomC (A, B) and v ∈ HomC (B, C), there exists a unique morphism
w = v ◦ u ∈ HomC (A, C), called the composition of v and u.

(3) For every A ∈ Ob(C), there is some arrow, 1A ∈ HomC (A, A), so that for every B ∈ Ob(C) and
u ∈ HomC (A, B), we have
1
A u u
A −→ A −→ B = A −→ B
u 1B u
A −→ B −→ B = A −→ B.

Note: This shows that 1A is unique for each A (DX).

(4) We have the associativity law

u ◦ (v ◦ w) = (u ◦ v) ◦ w,
whenever the compositions all make sense.

Examples of Categories:

(1) Sets, the category of sets; Ob(Sets) = all sets, Fl(Sets) = all maps of sets.

(2) Gr, the category of groups; Ob(Gr) = all groups, Fl(Gr) = all homomorphisms of groups. A special
case is Ab, the category of abelian groups.

(3) Ω-Gr, the category of Ω-groups. Special cases are: The category of G-modules, Mod(G); the category
of R-modules, Mod(R) (where R is a ring); and the category of vector spaces, Vect(k) (where k is a
field). Also, Ab = Mod(Z).

(4) TOP, the category of topological spaces; Ob(TOP) = all topological spaces,
Fl(TOP) = all continuous maps.

(5) C k -MAN, the category of C k -manifolds; Ob(C k -MAN) = all (real) C k -manifolds (0 ≤ k ≤ ∞ or ω),
Fl(C k -MAN) = all C k -maps of C k -manifolds.

(6) HOL, the category of complex analytic manifolds; Ob(HOL) = all complex analytic manifolds,
Fl(HOL) = all complex analytic maps of holomorphic manifolds.

(7) RNG, the category of all rings; Ob(RNG) = all rings (with unity), Fl(RNG) = all homomorphisms of
rings. A special case is CR, the category of commutative rings.

A subcategory, D, of C is a category, hOb(D), Fl(D)i, so that

(a) Ob(D) ⊆ Ob(C).

96 CHAPTER 1. GROUP THEORY

(b) Fl(D) ⊆ Fl(C), in such a way that for all A, B ∈ Ob(D), we have
HomD (A, B) ⊆ HomC (A, B).

We say that D is a full subcategory of C iff for all A, B ∈ Ob(D), we have

HomD (A, B) = HomC (A, B).
Examples of Subcategories:

(1) The category, Ab, is a full subcategory of Gr; the category, CR, is a full subcategory of RNG.

(2) Recall that u ∈ HomC (A, B) is an isomorphism (in C) iff there is some v ∈ HomC (B, A) so that
u v A 1
A −→ B −→ A = A −→ A
v u B 1
B −→ A −→ B = B −→ B.

Take D so that Ob(D) = Ob(C), and morphisms, set S

HomD (A, B) = {u ∈ HomC (A, B) | u is an isomorphism} and Fl(D) = A,B HomD (A, B). (Note that
HomD (A, B) may be empty.) The category, D, is generally a nonfull subcategory of C, for example
when C = Sets.

Say C is a category, we can make a new category, C D , the dual or opposite category, as follows: Ob(C D ) =
Ob(C) and reverse the arrows, i.e., for all A, B ∈ Ob(C),

HomC D (A, B) = HomC (B, A).

Definition 1.14 Let C and C 0 be categories. A functor (respectively, a cofunctor ), F , from C to C 0 is a rule
which associates to each object A ∈ Ob(C) an object F (A) ∈ Ob(C 0 ) and to each arrow u ∈ HomC (A, B) an
arrow F (u) ∈ HomC 0 (F (A), F (B)) (resp. F (u) ∈ HomC 0 (F (B), F (A))) so that,

F (1A ) = 1F (A)
F (u ◦ v) = F (u) ◦ F (v)
(resp. F (u ◦ v) = F (v) ◦ F (u), for cofunctors.)

Remark: Obviously, Definition 1.14 can be made more formal by defining a functor, F , from C to C 0 as a
pair, hF ob , F fl i, where F ob : Ob(C) → Ob(C 0 ) and F fl : Fl(C) → Fl(C 0 ), so that, for every u ∈ HomC (A, B),
we have F fl (u) ∈ HomC 0 (F ob (A), F ob (B)), and the conditions of Definition 1.14 hold (and similarly for
cofunctors).
We use the notation A F (A) (or u F (u)) to indicate that F : C → C 0 is a functor from C to C 0 , and
not just an ordinary function.
Examples of Functors

(1) For the categories in Examples (2)–(7), consider the rule:

A ∈ Ob(C) |A| = the underlying set of A, and
u ∈ Fl(C) |u| = the morphism, u, as a map of sets.
The functor, | |, is a functor from C to Sets, called the forgetful functor or stripping functor .
0
(2) A cofunctor, F : C → C 0 , is just a functor, F : C D → C 0 (equivalently, F : C → C D
).

(3) We have the functor, Ga : RNG → Ab, given by taking Ga (R) = R as an additive group, for every
ring, R. The functor, Ga , is called the additive group functor .
1.7. CATEGORIES, FUNCTORS AND FREE GROUPS 97

(4) For every integer, n ≥ 0, we have the functor, GLn : CR → Gr, where GLn (A) is the group of invertible
n × n matrices with entries in A. When n = 1, the group GL1 is denoted Gm . This is the multiplicative
group functor , it takes CR to Ab. The functor Gm can be promoted to a functor, RNG −→ Gr, taking
the ring, A, to its group, A∗ , of units.
(5) Let (TOP, ∗) be the category of topological spaces together with a base point. We have the subcategory
(C-TOP, ∗) consisting of connected and locally connected topological spaces with a base point. The
morphisms of (C-TOP, ∗) preserve base points. We have the functors (fundamental group)

π1 : (C-TOP, ∗) → Gr,

and for n > 1 (nth homotopy group),

πn : (C-TOP, ∗) → Ab.

(6) For every integer, n ≥ 0, we have a functor (integral homology), TOP −→ Ab, given by X Hn (X, Z)
and a cofunctor (integral cohomology), TOP −→ Ab, given by X H n (X, Z).
(7) math.upenn.edu/ Given a group, G, for any integer, n ≥ 0, we have a functor, Mod(G) −→ Ab, given
by A H n (G, A).

Definition 1.15 Say F and F 0 are two functors C −→ C 0 . A morphism, θ, from F to F 0 is a collection
{θA | A ∈ Ob(C)}, where:
(1) θA : F (A) → F 0 (A) in C 0 , so that (consistency)
(2) For every v : A → B in C, the diagram

F (A)
θA
/ F 0 (A)

F (v) F 0 (v)
 
F (B)
θB
/ F 0 (B)

commutes, for all A, B ∈ Ob(C).

A morphism of functors is also called a natural transformation of functors.

Examples of Morphisms of Functors:

(1) In the category (C-TOP, ∗), we have the functors π1 and H1 (−, Z). The Hurewicz map
u
X
π1 (X) −→ H1 (X, Z)

defines a morphism of functors.

(2) If G is a group and K is a subgroup of G, we have the obvious restriction functor
res : Mod(G) → Mod(K), and it induces a morphism of functors res : H n (G, −) → H n (K, −).
(3) The determinant, det : GLn → Gm , is a morphism of functors (from CR to Ab).
(4) Check (DX) that with the above notion of morphisms, the functors from C to C 0 form a category
themselves. This category is denoted Fun(C, C 0 ).

Proposition 1.38 Given a category, C, each object, A, of C gives rise to both a functor, hA , and a cofunctor,
hD
A , from C to Sets.
98 CHAPTER 1. GROUP THEORY

Proof . For any given A ∈ Ob(C), let

hA (B) = HomC (A, B)
hD
A (B) = HomC (B, A).
Moreover, for every v ∈ HomC (B, C), define hA (v) : HomC (A, B) → HomC (A, C) by composition, so that for
every u ∈ HomC (A, B),
hA (v)(u) = v ◦ u,
and, for every v ∈ HomC (B, C), define hD
A (v) : HomC (C, A) → HomC (B, A), again by composition, so that
for every u ∈ HomC (C, A),
hDA (v)(u) = u ◦ v.
The reader should check that hA and hD
A are indeed functors (DX).
The following proposition is half of the Yoneda embedding lemma:
Proposition 1.39 Let A and A e be two objects of C and suppose that the corresponding functors hA and
hAe are isomorphic, say by θ : hA → hAe. Then, A and A e are isomorphic via a canonically determined
isomorphism (dependent on θ).
Proof . For every B ∈ Ob(C), we have an isomorphism

θB : HomC (A, B) −
g e B),
→ HomC (A,

and this is functorial. Let B = A, then θA : HomC (A, A) −

g→HomC (A, e A), and we set ψ = θA (1A ), a morphism
e e e
in HomC (A, A). Now, if we let B = A, we get θAe : HomC (A, A) −
g→ HomC (A, e A),
e and we set ϕ = θ−1 (1 e), a
A
e A
morphism in HomC (A, A). e Pick any z in HomC (A, B). We would like to understand what θB (z) is. We have
the commutative diagram
z ∈ HomC (A, B)
θB
/ HomC (A,e B)
O O
z◦− z◦−

1A ∈ HomC (A, A) / HomC (A,

e A).
θA

Following the above commutative diagram clockwise, we get θB (z), and following it counterclockwise, we get
z ◦ ψ. We conclude that
θB (z) = z ◦ ψ.
e B), by considering the commutative diagram involving θ−1 and θ−1 , we get
Similarly, for any ze ∈ HomC (A, A
e B

−1
θB z ) = ze ◦ ϕ.
(e
But then, we have
−1
1Ae = θAe(ϕ) = ϕ ◦ ψ and 1A = θA (ψ) = ψ ◦ ϕ,
which shows that ϕ and ψ are inverse isomorphisms. Furthermore, ϕ (resp. ψ) determine θ, just as θ
determines ϕ and ψ.
Example. Recall that Vect(k) is the category of vector spaces over a field, k. There exists a cofunctor,
D : Vect(k) −→ Vect(k), given by: V V D = HomVect(k) (V, k) = the dual space of V ; and for any linear
map, θ : V → W , the map θ : W → V D is the adjoint of θ. By applying D again, we get a functor,
D D

DD : Vect(k) −→ Vect(k). However, it is well-known that there exists a morphism of functors,

ηV
η : id → DD, where id(V ) = V −→ DD(V ) = V DD , and this is functorial.
Two categories, C and C 0 , are equivalent (resp. isomorphic) iff there exist functors F : C → C 0 and
F : C 0 → C so that F 0 ◦ F ∼
0
= 1C and F ◦ F 0 ∼
= 1C 0 (resp. F 0 ◦ F = 1C and F ◦ F 0 = 1C 0 ). Here 1C denotes the
identity functor from C to itself.
1.7. CATEGORIES, FUNCTORS AND FREE GROUPS 99

Proposition 1.40 (Yoneda’s Embedding Lemma) The functor A hD A establishes an equivalence of the
D D
category, C, with a full subcategory of Fun (C, Sets) (where Fun (C, C 0 ) denotes the category of cofunctors
from C to C 0 ).

Proof . We already know from Proposition 1.39 that if we have an isomorphism θ : hD A → hA

D
e , then θ de-
e e
termines uniquely two mutually inverse isomorphisms ψ : A → A and ϕ : A → A. So, two objects A and
Ae in Ob(C) give isomorphic cofunctors iff they themselves are isomorphic. Given any v ∈ HomF (hD , hD ),
A A
e
where F = FunD (C, Sets), we know (again) that there exists a morphism ψ : A → A, e so that v is given by
composing with ψ, i.e., given a consistent family of morphisms, vB : hD
A (B) → hA
D
e (B), that is,
e
vB : HomC (B, A) → HomC (B, A), we have vB (z) = ψ ◦ z, and our ψ is given by ψ = vA (1A ) (all this from
the proof of Proposition 1.39). Hence, from v, we get a morphism ψ : A → A, e thus

e ∼
HomC (A, A) = HomF (hD D
A , hA
e ).

So, we indeed have an equivalence with a full subcategory of F, namely the image consists of those cofunctors
of the form hD
A (easy details are left to the reader (DX)).

Remark: What does Yoneda’s lemma say? It says that any object A ∈ Ob(C) is determined by its corre-
sponding cofunctor hD D
A . The cofunctor, hA , is a “collection of interconnected sets”, HomC (B, A) being the
set associated with B.

Definition 1.16 Given a functor, F , from C to Sets (resp. a cofunctor, G, from C D to Sets), it is repre-
sentable iff there exists a pair, (A, ξ), where A ∈ Ob(C) and ξ ∈ F (A), so that F is isomorphic to hA via the
morphism of functors, ξe: hA → F , given by the consistent family of morphisms ξeB : HomC (A, B) → F (B)
defined via
ξeB (u) = F (u)(ξ),

(resp. G is isomorphic to hD e D e
A via the morphism of functors, ξ : hA → G, given by ξB : HomC (B, A) → G(B).
Here, ξeB is defined via ξeB (u) = G(u)(ξ)).

The notion of representable functor is a key concept of modern mathematics. The underlying idea is to
“lift” as much as possible of the knowledge we have about the category of sets to other categories. More
specifically, we are interested in those functors from a category C to Sets that are of the form hA for some
object A ∈ Ob(C).

Remark: If (A, ξ) and (A0 , ξ 0 ) represent the same functor, then there exists one and only one isomorphism
A− → A0 so that ξ ∈ F (A) maps to ξ 0 ∈ F (A0 ). This is because we have the isomorphisms ξe: hA −
g → F and
g
−1
e 0
ξ : hA0 −
g→ F ; and so, we have an isomorphism ξ e0 e
◦ ξ : hA −
g→ hA0 . By Yoneda’s lemma, Ag−→ A0 via the
e e0 0
isomorphism determined by ξ and ξ and this maps ξ to ξ . Uniqueness follows as everything is determined
by ξ and ξ 0 .

Examples of Representable Functors:

(1) Let C = Sets; consider the functor F : SetsD → Sets given by: F (S) = the collection of all subsets of
S, and if θ : S → T is a map of sets, the morphism F (θ) : F (T ) → F (S) is the map that sends every
subset, V , of T to its inverse image, θ−1 (V ), a subset of S. Is this a representable functor?

We need a set, Q, and an element, ξ ∈ F (Q), i.e., some subset of Q, so that

hD
Q (B) = HomSets (B, Q) −
g→ F (B), via ξeB (u) = F (u)(ξ).
100 CHAPTER 1. GROUP THEORY

Now, we know that F (u) : F (Q) → F (B) is the map that sends a subset, S, of Q to its inverse image,
u−1 (S), a subset of B. So, F (u)(ξ) is the inverse image of our chosen ξ.

Take Q = {0, 1} and ξ = {1} ⊆ Q. Then, subsets of B are exactly of the form, u−1 (1), for the various
u ∈ HomSets (B, Q), which are thus characteristic functions.
(2) Let C = RNG, and let F : RNG → Sets be the stripping functor. Is it representable?

We need a ring, P , and an element, ξ ∈ P , so that for all rings, B,

HomRNG (P, B) −
g→ |B|,

via
u ∈ HomRNG (P, B) 7→ u(ξ) ∈ |B|.
Take P = Z[T ], the polynomial ring in one variable with integral coefficients, and ξ = T . Then, any
ring homomorphism u ∈ HomRNG (Z[T ], B) is uniquely determined by u(T ) = b ∈ |B|, and any b can
be used.

Definition 1.17 Let F : C → C 0 and G : C 0 → C be two functors. The functor F is the left (resp. right)
adjoint of G iff for every A ∈ Ob(C) and B ∈ Ob(C 0 ), we have functorial isomorphisms (in both A and B)

HomC 0 (F (A), B) −
g→ HomC (A, G(B)).
(resp. HomC 0 (B, F (A)) −
g→ HomC (G(B), A)).

Observe that F is left-adjoint to G iff G is right-adjoint to F . Many so-called “universal constructions”

arise from the existence of adjoint functors; this is a key concept in modern mathematics.

Remark: The concept of adjointness is related to the notion of representability of a functor, as shown by the
following proposition whose simple proof is left to the reader:

Proposition 1.41 A functor, G : C 0 → C, has a left-adjoint if and only if, for every A ∈ C, the functor B
HomC (A, G(B)) from C 0 to Sets is representable. If (F (A), ξ) represents this functor (so that ξeB : HomC 0 (F (A), B) ∼
=
HomC (A, G(B)) is an isomorphism for every B ∈ C 0 ), then F is the object part of a left-adjoint of G for which the
isomorphism ξeB is functorial in B and yields the adjointness.

A functor may have a right adjoint, but no left adjoint, and conversely (or no adjoint at all). For example,
the functor, G G/[G, G] = Gab , from Gr to Ab, is the left adjoint of the inclusion functor from Ab to Gr.
The inclusion views an abelian group just as a group. So, G Gab has a right adjoint. However, we now
prove that it has no left adjoint.
Suppose such a left adjoint, F , exists.
Claim 1: For any abelian group, H, the group F (H) can never be simple unless F (H) = {1}, in which
case, H = {1}.
The adjointness property states that for every group, G, we have a functorial isomorphism

HomGr (F (H), G) ∼
= HomAb (H, Gab ). (∗)

If we take G = F (H) in (∗), we have

HomGr (F (H), F (H)) ∼

= HomAb (H, F (H)ab ).

If F (H) 6= {1} and F (H) is non-abelian simple, then, on the left hand side there are at least two maps
(id and the constant map that sends all elements to 1), even though on the righthand side there is a single
1.7. CATEGORIES, FUNCTORS AND FREE GROUPS 101

map, since F (H) is non-abelian simple, a contradiction. If F (H) is Z/pZ for some prime p, take G in (∗)
to be A3p . Again, there are at least two maps in HomGr (F (H), A3p ), namely: the constant map and an
embedding. But A3p is simple; so, the righthand side has only one element, again a contradiction. Now, if
F (H) = {1}, take G = H in (∗). In this case, the left hand side has a single map but the righthand side has
at least two maps if H 6= {1}.
Claim 2: F (H) has no maximal normal subgroups. If M C F (H) and M is maximal, then F (H)/M
is simple. Let G = F (H)/M in (∗). If F (H)/M is non-abelian, there are at least two maps on the left
hand side, but only one on the righthand side, a contradiction. If F (H)/M is abelian, say Z/pZ, again
take G = A3p in (∗). There are two maps (at least) on the left hand side (stemming from the two maps
F (H) −→ F (H)/M ) and only one on the righthand side. So, if F (H) exists, it is not finitely generated.
Take H = G = Z/2Z. Then, we have

HomGr (F (Z/2Z), Z/2Z) ∼

= HomAb (Z/2Z, Z/2Z).

Clearly, the righthand side has exactly two maps, and thus, so does the left hand side. But one of these
maps is the constant map sending all elements to 1, so the other map must be surjective. If so, its kernel,
K, is a subgroup of index 2, hence normal, and so, it must be maximal normal, a contradiction.
Therefore, the functor G G/[G, G] = Gab , from Gr to Ab, has no left adjoint.
One often encounters situations (for example in topology, differential geometry and algebraic geometry)
where the objects of interest are arrows “over” a given object (or the dual notion of arrows “co-over” a given
object), for example, vector bundles, fibre bundles, algebras over a ring, etc. Such situations are captured
by the abstract notion of “comma categories.”

Definition 1.18 Let C be a category and fix some object, A, in Ob(C). We let CA , the category over A
(or comma category), be the category whose objects are pairs (B, πB ), where B is some object in Ob(C)
and πB is a morphism in HomC (B, A), and whose morphisms from (B, πB ) to (C, πC ) are the morphisms
u ∈ HomC (B, C) making the following diagram commute:

B@
u / C
@@ }}
@@ }}
πB @@@ }}} πC
 }~
A

Dually, we let C A , the category co- over A (also called comma category), be the category whose objects are
pairs (B, iB ), where B is some object in Ob(C) and iB is a morphism in HomC (A, B), and whose morphisms
from (B, iB ) to (C, iC ) are the morphisms u ∈ HomC (B, C) making the following diagram commute:

B _@
u / C
@@ }}>
@@ }}
iB @@
@ }}
}} iC
A

The notion of representable functor allows us to define products and coproducts in arbitrary categories.
Let C be any category. Say {Aα }α∈Λ is a set of objects in Ob(C).
(1) We get a cofunctor, F , from C D to Sets via
Y
B HomC (B, Aα ) = F (B),
α

where the above product is just the cartesian product of sets, and
102 CHAPTER 1. GROUP THEORY

(2) We get a functor, G, from C to Sets via

Y
B HomC (Aα , B) = G(B).
α

Are these (or either) representable?

Q
First, consider (1). We need an object, P ∈ Ob(C) and some ξ ∈ F (P ), i.e. ξ ∈ α HomC (P, Aα ), which
means that ξ = {prα }α , where the prα are morphisms prα : P → Aα .

Definition 1.19 When (P, {prα }) exists, i.e. for every B ∈ Ob(C), there is a functorial isomorphism
Y
HomC (B, P ) − g→ HomC (B, Aα ),
α
Q
via u 7→ (prα ◦ u)α , the pair (P, {prα }) is the product of the Aα ’s in C. This product is denoted α Aα (one
usually drops the prα ’s). We have the (functorial) isomorphism
Y Y
HomC (B, Aα ) −
g→ HomC (B, Aα ). (∗)
α α

Remark: Definition
Q 1.19 implies that for every family of morphisms, Q
{fα : B → Aα } ∈ α HomC (B, Aα ), there is a unique morphism, u : B → α Aα , so that

fα = prα ◦ u, for all α.

This is called the universal mapping property of products. In general, universal mapping properties are another name
for representing a functor. The latter is a more general and supple notion and we will mainly stick to it.
Q
Now, consider (2). We need an object, Q ∈ Ob(C), and some ξ ∈ G(Q), i.e. ξ ∈ α HomC (Aα , Q), which
means that ξ = {iα }α , where the iα are morphisms iα : Aα → Q.

Definition 1.20 When (Q, {iα }) exists, i.e. for every B ∈ Ob(C), there is a functorial isomorphism
Y
HomC (Q, B) − g→ HomC (Aα , B),
α
`
via u 7→ (u ◦ iα )α , the pair (Q, {iα }) is the coproduct of the Aα ’s in C. This coproduct is denoted α Aα
(one usually drops the iα ’s). We have the (functorial) isomorphism
a Y
HomC ( Aα , B) − g→ HomC (Aα , B). (∗∗)
α α

Of course, as above, there is a universal mapping property here, also.

Definition 1.21 The product in CA is called the fibred product over A in C. The coproduct in C A is called
the fibred coproduct over A in C.

Remark: Given any

Q family, {(Aα , πQ
α )}α , of objects in CA (with πα : Aα → A), the fibred product of the Aα ’s over
Q
A in C is a pair, ( Aα , ξ), where Aα is some object in C (together with a morphism, π : Aα → A), and ξ
A QA A
consists of a family of morphisms, prα : Aα → Aα , with
A

πα ◦ prα = πβ ◦ prβ (= π), for all α, β;

1.7. CATEGORIES, FUNCTORS AND FREE GROUPS 103

moreover, for any object, B ∈ C, and any family of morphisms, {fα : B → Aα }α , with

πα ◦ fα = πβ ◦ fβ , for all α, β,
Q
there is a unique morphism, u : B → Aα , so that fα = prα ◦ u, for all α.
A

We leave it to the reader to unwind the definition of fibred coproducts over A in C.

Examples of Products, Coproducts, Fibred Products and Fibred Coproducts:

Q `
(1) C = Sets. Given a family of sets, {Aα }α∈Λ , does α Aα or α Aα exist? If so, what are they?
Q
For α Aα , we seek a set, P , and an element, ξ, in F (P ), where F is the cofunctor
Y
T F (T ) = HomSets (T, Aα ).
α

This means that ξ ∈ F (P ) is just a tuple of maps, prα : P → Aα . Take P to be the ordinary cartesian
product of the Aα ’s and prα : P → Aα , the αth projection. Check that this works (DX).
`
For α Aα , we seek a set, Q, and an element, ξ, in G(Q), where G is our functor
Y
T G(T ) = HomSets (Aα , T ).
α

So, we need a family of maps iα : Aα → Q. Now, if Q is to work, then for every T , we need an isomorphism
Y
θT : HomSets (Q, T ) −
g→ HomSets (Aα , T )
α
S
given by θT (ϕ) = (ϕ ◦ iα )α . Take Q = · α Aα (the disjoint union of the Aα ’s). The rest of the construction
is easy (DX).
(2) C = Ab, more generally, C = Mod(R) (R a ring) or C = Mod(G) (G a group).
We begin with products. Given a family, {Aα }α∈Λ , with each Aα in Mod(R), we seek P ∈ Mod(R) and
maps prα : P → Aα in Mod(R), so that for every T ∈ Mod(R), there is an isomorphism
Y
θT : HomR (T, P ) −
g→ HomR (T, Aα ),
α

where θT (ϕ) = {prα ◦ ϕ}α (the notation HomR (A, B) is usually used,Qinstead of the more accurate but more
cumbersome notation HomMod(R) (A, B)). WeQsee that P must be α Aα , the product in the category of
sets, if this can be made an R-module.Q Now, α Aα is an R-module via coordinatewise addition, with the
R-action given by r(ξα ) = (rξα ). So, α Aα is the product of the Aα ’s in Mod(R).
Next, we consider coproducts. We seek Q ∈ Mod(R) and maps iα : Aα → Q in Mod(R), so that for
every T ∈ Mod(R), there is an isomorphism
Y
θT : HomR (Q, T ) −
g→ HomR (Aα , T ),
α
S
where θT (ϕ) = {ϕ ◦ iα }α . The disjoint union · α Aα may be a first approximation to Q, but it is not good
enough. Instead, we let
n Y o
Q= ξ∈ Aα | prα (ξ) = 0 for all but finitely many α .
α
104 CHAPTER 1. GROUP THEORY

Q
This is an R-submodule of α Aα . The isomorphism
Y
θT : HomR (Q, T ) −
g→ HomR (Aα , T )
α

can now be established. First, let iα (u) = (δβu )β , where δαu = u and u
Pδβ = 0 for all β 6= α. Given a family,
(ϕα )α , of maps ϕα : Aα → T , for any ξ = (ξα )α ∈ Q,`set ϕ(ξ) = α ϕα (ξα ) ∈ T . If ϕ ∈ HomR (Q, T ) is
given, define ϕα = ϕ ◦ iα . This shows that if we set α Aα to be our R-module Q and the iα to be our
maps, iα : Aα → Q, as above, we have proved the proposition:

Proposition 1.42 The categories: Sets, Ab, Mod(R), Mod(G) all possess arbitrary products and coprod-
ucts.

How about fibred products and coproducts?

C = Sets, and first consider fibred products over A. A first approximation to the
(3) Let us go back to Q
product, P , in SetsA , is α Aα . However, this is not good enough because there is no “structure map”,
π : P → A, so that
P@
prα
/ Aα
@@ {{
@@ {{
π @@@ {{ π
 }{{ α
A
commutes for all α. We let
n Y o
PA = ξ ∈ Aα | πα (ξα ) = πβ (ξβ ), for all α, β .
α

Q π : PA → A by π(ξ) = πα (ξα ), for any

This is a set (possibly empty), and it lies over A; indeed, we can define
chosen α, since this is well-defined by definition of PA . We write Aα for PA and, for every α, we define
Q Q A Q
the map, prα : Aβ → Aα , as the restriction of prα : Aβ → Aα to Aα . The reader should check that
A A
this yields products in SetsA .
S
Coproducts are a bit harder. It is natural to try · α Aα as a first approximation,
S but this is not good
enough: this does not tell us what i : A → Q is. The difficulty is that · α S Aα is too big, and we need to
identify some of its elements. To do so, we define an equivalence relation on · α Aα , in two steps. First, we
define immediate equivalence. Given ξ ∈ Aα and η ∈ Aβ , we say that ξ and η are immediately equivalent,
denoted ξ ≈ η, iff there is some a ∈ A, so that ξ = iα (a) and η = iβ (a). The relation ≈ is clearly reflexive
and symmetric but it is not necessarily transitive. So, we S define ∼ to be the equivalence relation generated
by ≈. This means that ξ ∼ η iff there exist x0 , . . . , xt ∈ · α Aα , so that

ξ = x0 , x0 ≈ x1 , x1 ≈ x2 , . . . , xt−1 ≈ xt , xt = η.

(For example, if ξ ≈ x`and x ≈S η, then ξ = iα (a), x =`iβ (a), x = iβ (b) and η = iγ (b). Note that
iβ (a) = iβ (b).) We let Aα = ( · α Aα ) / ∼, and i : A → Aα is given by i(a) = class of iα (a), for any
A A `
fixed α (this is well-defined, by definition of ∼). The verification that Aα works is left as an exercise
A
(DX). Therefore, the category of sets has arbitrary fibred coproducts as well .
(4) C = Ab, Mod(R), Mod(G).
Q
For fibred products, we use Aα , as constructed for Sets, but made into an R-module (resp. G-module),
A
in the usual way.
1.7. CATEGORIES, FUNCTORS AND FREE GROUPS 105

`
For fibred coproducts, begin with α Aα (in C), and define N to be the submodule generated by the
elements iα (a) − iβ (a) with a ∈ A and α, β arbitrary. Take
!
a a
Aα = Aα /N.
A α
`
Again, the reader should check that Aα works (DX). Therefore, Ab, Mod(R), Mod(G), all have arbitrary
A
fibred products and coproducts.
We now consider
Q products and coproducts in the category of groups, Gr. Q There is no difficulty for
products: Use α Aα , the usual cartesian product of the Aα ’s, as sets, and make α Aα into a group under
coordinatewise multiplication. The same idea works for fibred products. However, coproducts require a new
idea.
Given the family of groups, {Aα }α∈Λ , write A0α = Aα − {1}. Let
[
S = · A0α ,
α

and consider, S n , the n-fold cartesian product of S. We can view S n as the set of words of length n over
the alphabet S; each word is an n-tuple, (σα1 , . . . , σαn ), with σβ ∈ Aβ . We call such a word admissible iff
Aαj 6= Aαj+1 , for j = 1, 2, . . . , n − 1. Let S n ∗ denote the set of admissible words of length n, and let
!
[
n∗
Q= S ∪ {∅}.
n≥1

(The special word, ∅, is the “empty word”.) Multiplication in Q is defined as follows:

Given (σ) = (σα1 , . . . , σαr ) and (τ ) = (τβ1 , . . . , τβs ) in Q, set

(σ)(τ ) = (σα1 , . . . , σαr , τβ1 , . . . , τβs ),

the result of concatenating the r-tuple, (σ), with the s-tuple, (τ ). In case one of (σ) or (τ ) is ∅, the
concatenation is just the non-empty word and ∅∅ is ∅. The word (σ)(τ ) is admissible of length r + s, except
if αr = β1 , in which case we need to perform a reduction process to obtain an admissible word:

(1) Form σαr τβ1 in Aαr = Aβ1 . There are two cases:

(a) σαr τβ1 6= 1αr (= 1β1 ); then

(σα1 , . . . , σαr−1 , σr τβ1 , τβ2 , . . . , τβs )

is an admissible word of length r + s − 1, and the reduction process ends with this word as output.

(b) σαr τβ1 = 1αr (= 1β1 ); then, omit σαr and τβ1 , form

(σα1 , . . . , σαr−1 , τβ2 , . . . , τβs ),

a word of length r + s − 2, and if necessary, go back to (1) above.

Since both step (a) and (b) decrease the length of the current word, the reduction process must end with
some admissible word of length l ≤ r + s, or the empty word.
The set Q with the above multiplication is indeed a group with identity element, ∅ (DX). (The map
iα : Aα → Q sends σ ∈ Aα to the length-one word (σ) if σ 6= 1 or to ∅ if σ = 1.) In summary, we get
106 CHAPTER 1. GROUP THEORY

Theorem 1.43 The category of groups, Gr, possesses arbitrary coproducts (old fashioned name: “free prod-
uct of the Aα .”)
`
Definition 1.22 Given any set, S, define the the free group on S to be the group Fr(S) = Z.
S

We have just shown that coproducts exist in the category Gr. What about coproducts in the category
GrA , where A is any group?
`
Given a family {(Gα , iα )}α∈Λ in GrA , form G = α Gα , in the category Gr. In G, consider the collection
of elements
{iα (a)i−1
β (a) | a ∈ A, iα : A → Gα , α and β ∈ Λ};

let N be the normal subgroup of G generated by the above elements. Then, G/N ∈ Ob(GrA ), because
the map i : A → G/N given by i(a) = image of iα (a) in G/N (for any fixed α) is well-defined (since image
of iα (a) = image of iβ (a) in G/N ). Check that, (DX), (G/N, i) is the fibred coproduct of the Gα ’s. (Old
terminology: amalgamated product of the Gα over A.)

Examples of fibred coproducts: (1) Let U and V be two sets. Form the intersection U ∩ V ; we have
inclusion maps iU : U ∩ V → U and iV : U ∩ V → V . We know that U q V = U ∪· V , the disjoint union of U
and V , and then, the set-theoretic union of U and V is given by
a
U ∪V =U V.
U ∩V

(2) Consider the category (TOP, ∗) of (“nice”, i.e., connected, locally connected) topological spaces with
a base point. Given two spaces (U, ∗) and (V, ∗) in (TOP, ∗), consider (U ∩ V, ∗). Then, again,
a
(U ∪ V, ∗) = (U, ∗) (V, ∗), in (TOP, ∗).
(U ∩V,∗)

Van Kampen’s theorem says that

a
π1 (U ∪ V, ∗) = π1 (U, ∗) π1 (V, ∗),
π1 (U ∩V,∗)

which may also be written as

!
a a
π1 (U, ∗) (V, ∗) = π1 (U, ∗) π1 (V, ∗).
(U ∩V,∗) π1 (U ∩V,∗)

In other words, van Kampen’s theorem says that π1 commutes with fibred coproducts.
Go back to the free group, Fr(S). We have
a
HomGr (Fr(S), G) = HomGr ( Z, G)
S
Y
∼
= HomGr (Z, G)
S
Y
∼
= |G| = HomSets (S, |G|).
S

Corollary 1.44 The functor, S Fr(S), from Sets to Gr is the left adjoint to the stripping functor,
G |G|, from Gr to Sets.
1.7. CATEGORIES, FUNCTORS AND FREE GROUPS 107

Corollary 1.45 If S −→ T is surjective, then Fr(S) −→ Fr(T ) is a surjection of groups. Also,

Fr(S) ∼
= Fr(T ) iff #(S) = #(T ) (i.e., S and T have the same cardinality).

Proof . If u : S → T is a surjection in Sets, then there is a map v : T → S so that u ◦ v = 1T . Since Fr is a

functor, we get homomorphisms Fr(u) : Fr(S) → Fr(T ) and Fr(v) : Fr(T ) → Fr(S); also, Fr(u)◦Fr(v) = 1Fr(T ) ,
which shows that Fr(u) is surjective.
If #(S) = #(T ), it is obvious that Fr(S) ∼ = Fr(T ). Conversely, assume that Fr(S) ∼ = Fr(T ). We know
that
HomGr (Fr(S), G) ∼= HomGr (Fr(T ), G)
for all G. Take G = Z/2Z. Then, the left hand side is isomorphic to HomSets (S, |Z/2Z|) = P(S) (where
P(S) = power set of S) and the righthand side is isomorphic to P(T ). Therefore, #(P(S)) = #(P(T )); and
so, #(S) = #(T ).
Given a group, G, consider its underlying set, |G|, and then the group Fr(|G|). Since

HomGr (Fr(|G|), G) ∼
= HomSets (|G|, |G|),

the image of the identity map, idG ∈ HomSets (|G|, |G|), yields a canonical surjection, Fr(|G|) −→ G. If S is
a subset of |G|, then, the inclusion map, S ,→ |G|, yields a morphism of groups, Fr(S) −→ G.

Definition 1.23 A set, S ⊆ |G|, generates a group, G, iff the canonical map Fr(S) −→ G is surjective.

This definition agrees with our old use of generation of a group in previous sections. Say S generates G.
Then, we have the exact sequence

0 −→ K −→ Fr(S) −→ G −→ 0,

where K is the kernel of the surjective morphism, Fr(S) −→ G (so, K is normal in Fr(S)). There is also a
set, T , so that
Fr(T ) −→ K −→ 0 is exact.
By splicing the two exact sequences, we get an exact sequence

Fr(T ) −→ Fr(S) −→ G −→ 0,

called a presentation of G. Sometimes, a presentation is defined as a sequence

Fr(T ) −→ Fr(S) −→ G −→ 0,

where the smallest normal subgroup containing Im (Fr(T )) is equal to the kernel of Fr(S) −→ G. (Note that
such a sequence is not necessarily exact at the group Fr(S).)
The following fundamental theorem about free groups was proved independently by J. Nielson and O.
Schreier:

Theorem 1.46 (Nielson-Schreier (1929)) Every subgroup of a free group is a free group.

The original proof is quite messy. The theory of group actions on trees yields a more direct and more
transparent proof.
We conclude this section on categories with one more interesting example of adjoint functors from ho-
motopy theory.

Example: Consider the category, h-TOP, whose objects are the same as those of TOP, but whose mor-
phisms, Homh-TOP (X, Y ), are the homotopy classes of maps X −→ Y . Given any space, X, in h-TOP, we
108 CHAPTER 1. GROUP THEORY

can form ΣX, the suspension of X: This is the space obtained by taking two new points, say 0 and 1, and
forming the double cone obtained by joining 0 and 1 to every point of X, as illustrated in Figure 1.2.
We also have, ΩY , the loop space on Y , where ΩY consists of all continuous maps, S 1 −→ Y , from the
unit circle to Y (say, mapping (1, 0) to the base point of Y ). Then, we have the isomorphism

Homh-TOP (ΣX, Y ) ∼
= Homh-TOP (X, ΩY )

i.e., suspension is left-adjoint to loops. For instance, given any θ ∈ Homh-TOP (ΣX, Y ), for any p ∈ X, send
p to the image by θ of the loop l(p) (= (∗, 0, p, 1, ∗) in ΣX), in Y .

X p
∗ ΣX

Figure 1.2: A suspension of X

1.8 Further Readings

El que anda mucho y lee mucho,
Vee mucho y sabe mucho.
—Miguel Cervantes

Some group theory is covered in every algebra text. Among them, we mention Michael Artin [2], Lang [34],
Hungerford, [27], Jacobson [29], Mac Lane and Birkhoff [37], Dummit and Foote [11], Van Der Waerden [47]
and Bourbaki [4]. More specialized books include Rotman [43], Hall [22], Zassenhaus [52], Rose [42] and
Gorenstein [19]. For group cohomology, see also Cartan and Eilenberg [9], Rotman [44], Mac Lane [36] and
Serre [45]. Mac Lane [35] is a good reference for category theory.
Chapter 2

Rings and Modules

2.1 Introduction
Linear algebra—meaning vector space theory over a field—is the part of algebra used most often in analysis,
in geometry and in various applied fields. The natural generalization to the case when the base object is
a ring rather than a field is the notion of “module.” The theory of modules both delineates in sharp relief
the elementary and deeper structure of vector spaces (and their linear transformations) and provides the
essential “linear springboard” to areas such as number theory, algebraic geometry and functional analysis.
It turns out to be surprisingly deep because the collection of “all” modules over a fixed ring has a profound
influence on the structure of that ring. For a commutative ring, it even specifies the ring! Just as in analysis,
where the first thing to consider in analyzing the local behavior of a given smooth function is its linear
approximation, so in geometric applications the first idea is to pass to an appropriate linear approximation
and this is generally a module.

2.2 Polynomial Rings, Commutative and Noncommutative

Consider the categories RNG and CR, and pick some ring, A, from each. We also have the category, RNGA ,
called the category of rings over A (or category of A-algebras), and similarly, CRA , and we have the stripping
functors RNGA Sets and CRA Sets.
Is there an adjoint functor to each? We seek a functor, P : Sets C, where C = RNGA or CRA , so that

HomC (P (S), B) ∼
= HomSets (S, |B|)

for every B ∈ C.
Case 1: CRA .

Theorem 2.1 There exists a left-adjoint functor to the stripping functor, CRA Sets.
e denote the set of non-negative integers and write N
Proof . Given a set, S, let N e S for

e S = {ξ : S → N
N e | ξ(s) = 0, except for finitely many s ∈ S}.

e with compact support (where S and N

Note that NS consists of the functions S −→ N e are given the discrete
topology).

Remark: We may think of the elements, ξ, of N

e S as finite multisets of elements of S, i.e., finite sets with multiple
occurrences of elements: For any s ∈ S, the number ξ(s) is the number of occurrences of s in ξ. If we think of each

109
110 CHAPTER 2. RINGS AND MODULES

member, s, of S as an “indeterminate,” for any ξ ∈ N

e S , if ξ(si ) = ni > 0 for i = 1, . . . , t, then ξ corresponds to the
monomial sn1
1
· · · snt
t .
e S as follows: For ξ, η ∈ N
We define a multiplication operation on N eS,

(ξη)(s) = ξ(s) + η(s).

e S is associative and has the identity element, ξ0 , with ξ0 (s) = 0 for
(This multiplication operation on N
e
all s ∈ S. Thus, NS is a monoid . Under the interpretation of elements of Ne S as multisets, multiplication
corresponds to union and under the interpretation as monomials, it corresponds to the intuitive idea of
multiplication of monomials. See below for precise ways of making these intuitions correct.)
Define A[S] by
e S → A | f (ξ) = 0,
A[S] = {f : N e S }.
except for finitely many ξ ∈ N

Remark: We should think of each f ∈ A[S] as a polynomial in the indeterminates, s (s ∈ S), with coefficients from
A; each f (ξ) is the coefficient of the monomial ξ. See below where Xs is defined.
In order to make A[S] into a ring, we define addition and multiplication as follows:

(f + g)(ξ) = f (ξ) + g(ξ)

X
(f g)(ξ) = f (η)g(η 0 ).
η,η 0 ,
ηη 0 =ξ

Multiplication in A[S] is also called the convolution product. The function with constant value, 0 ∈ A, is the
zero element for addition and the function denoted 1, given by

0 if ξ 6= ξ0
1(ξ) =
1 if ξ = ξ0 ,

is the identity element for multiplication. The reader should check that under our operations, A[S] is a
commutative ring with identity (DX). For example, we check that 1 is an identity for multiplication. We
have X X
(f · 1)(ξ) = f (η)1(η 0 ) = f (η).
ηη 0 =ξ ηξ0 =ξ

However, for all s ∈ S, we have ηξ0 (s) = η(s) + ξ0 (s) = η(s), and so, η = ξ. Consequently, (f · 1)(ξ) = f (ξ),
for all ξ.
We have an injection A −→ A[S] via α ∈ A 7→ α · 1. Here, α · 1 is given by

0 if ξ 6= ξ0
α · 1(ξ) = α(1(ξ)) =
α if ξ = ξ0 .

Therefore, A[S] ∈ CRA . It remains to check the “universal mapping property.”

e S and S ,→ A[S] (a map of sets) as
Say θ ∈ HomCRA (A[S], B). Now, we can define two injections S ,→ N
e S by
follows: Given any s ∈ S, define ∆s ∈ N

0 if t 6= s
∆s (t) =
1 if t = s,

and define Xs ∈ A[S] by 

0 if ξ 6= ∆s
Xs (ξ) =
1 if ξ = ∆s .
2.2. POLYNOMIAL RINGS, COMMUTATIVE AND NONCOMMUTATIVE 111

Then, if we set θ[ (s) = θ(Xs ), we get a set map θ[ ∈ HomSets (S, |B|).
e S → B via
e: N
Conversely, let ϕ ∈ HomSets (S, |B|). Define ϕ
Y
e
ϕ(ξ) = ϕ(s)ξ(s) ∈ B.
s∈S

Now, set ϕ] (f ), for f ∈ A[S], to be X

ϕ] (f ) = e
f (ξ)ϕ(ξ).
ξ

(Of course, since B ∈ CRA , we view f (ξ) as an element of B via the corresponding morphism A −→ B.)
The reader should check (DX) that:
(a) ϕ] is a homomorphism and
(b) The operations ] and [ are mutual inverses.

The definition of A[S] has the advantage of being perfectly rigorous, but it is quite abstract. We can give
a more intuitive description of A[S]. For this, for any ξ ∈ Ne S , set
Y
X (ξ) = Xsξ(s) , in A[S],
s∈S

and call it a monomial . The reader should check (DX) that

X (ξ) (η) = δξη , eS.

for all ξ, η ∈ N
e S to the monomials (c.f. the remark on monomials made earlier).
Hence, the map ξ 7→ X (ξ) is a bijection of N
Moreover, we claim that every f ∈ A[S] can be written as
X
f= f (ξ)X (ξ) .
ξ

This is because !
X X
f (ξ)X (ξ) (η) = f (ξ)δξη = f (η).
ξ ξ
Q
The usual notation for ξ(s) is ξs , and then, X (ξ) = s∈S Xsξs , and our f ’s in A[S] are just polynomials in
the usual sense, as hinted at already. However, since S may be infinite, our formalism allows us to deal with
polynomials in infinitely many indeterminates. Note that any polynomial involves just a finite number of
the variables.
What happened to |A| in all this? After all, in CRA , we have rings, B, and maps iA : A → B. So, the
commutative diagram
B _@ / C
@@ >
@@ }}}
@@ }}
@ }}}
A
would give
|B| / |C|
aBB { =
BB {{
BB {
BB {{
{{
|A|.
112 CHAPTER 2. RINGS AND MODULES

Consider the category of |A|-sets, Sets|A| . Given any set, S, make an |A|-set:

|A| q S = |A| ∪· S.

This is an |A|-set, since we have the canonical injection, |A| −→ |A| q S. Let T be any |A|-set and look at
HomSets|A| (|A| q S, T ), i.e., maps |A| q S −→ T such that the diagram

|A| q S / T
cGG
GG ~~?
GG ~~
GG
G ~~~
~
|A|

commutes. We know that

Y
HomSets|A| (|A| q S, T ) ⊆ HomSets (|A|, T ) HomSets (S, T )
Q
and the image is HomSets|A| (|A|, T ) HomSets (S, T ). But HomSets|A| (|A|, T ) consists of a single element,
and so,
HomSets|A| (|A| q S, T ) ∼
= HomSets (S, T ).
Thus, we have the functorial isomorphism

HomCRA (A[S], B) ∼
= HomSets|A| (|A| q S, |B|).

Corollary 2.2 A necessary and sufficient condition that Z[S] ∼

= Z[T ] (in CR) is that #(S) = #(T ).

Proof . If #(S) = #(T ), then there exist mutually inverse bijections, ϕ : S → T and ψ : T → S. Hence, by
functoriality, Z[S] is isomorphic to Z[T ] (via Z[S](ϕ) and Z[T ](ψ)). Now, take B = Z/2Z, and assume that
Z[S] ∼
= Z[T ]. Then, we know that

HomCR (Z[S], B) ∼
= HomCR (Z[T ], B),

= HomSets (S, {0, 1}) and HomCR (Z[T ], B) ∼

and since HomCR (Z[S], B) ∼ = HomSets (T, {0, 1}), we have

HomSets (S, {0, 1}) ∼

= HomSets (T, {0, 1}).

This implies that 2#(S) = 2#(T ) , and thus, #(S) = #(T ).

Case 2: RNGR , where R is a given ring (not necessarily commutative). For every set, S, and every
R-algebra, B ∈ RNGR , let
(c)
HomSets (S, |B|) = {ϕ ∈ HomSets (S, |B|) | (∀s ∈ S)(∀ξ ∈ Im (|R|))(ϕ(s)ξ = ξϕ(s))}.

Theorem 2.3 There exists a functor, RhSi, from Sets to RNGR , so that

HomRNGR (RhSi, B) ∼
(c)
= HomSets (S, |B|), functorially.

Sketch of proof . (A better proof via tensor algebras will be given later.) Given S, pick a “symbol”, Xs , for
each s ∈ S, and map N to the “positive powers of Xs ,” via n 7→ Xsn , and define Xsm · Xsn = Xsm+n . Let
Ns = {Xsn | n ≥ 1} ∼= N (as monoid), and let
a
S= Ns .
s∈S
2.2. POLYNOMIAL RINGS, COMMUTATIVE AND NONCOMMUTATIVE 113

Consider S (p) , the cartesian product of p copies of S, with p ≥ 1. An element of S (p) is a tuple of the form
a
(Xra11 , . . . , Xrpp ), and is called a monomial . Call a monomial admissible iff ri 6= ri+1 , for i = 1, . . . , p − 1.
Multiplication of admissible monomials is concatenation, with possible one-step reduction, if necessary. Call
S ∗ the union of all the admissible monomials from the various S (p) , with p ≥ 1, together with the “empty
monomial”, ∅. Set
RhSi = {f : S ∗ → R | f (ξ) = 0, except for finitely many ξ ∈ S ∗ }.
There is a map R −→ RhSi (α 7→ α∅). We make RhSi into a ring by defining addition and multiplication
as in the commutative case:
(f + g)(ξ) = f (ξ) + g(ξ)
X
(f g)(ξ) = f (η)g(η 0 ),
η,η 0 ,
ηη 0 =ξ

where ξ, η and η 0 are admissible monomials. Then, RhSi is an R-algebra, and it satisfies Theorem 2.3 (DX).

Theorem 2.4 Say T is a subset of S. Then, there exists a canonical injection i : A[T ] → A[S], and A[S]
becomes an A[T ]-algebra. In the category of A[T ]-algebras, we have the isomorphism
A[S] ∼
= A[T ][S − T ]
(Here S − T denotes the complement of T in S, and A is in CR.)
Proof . We have an inclusion, T ,→ S, and for every B ∈ CRA , restriction to T gives a surjection
res : HomSets (S, |B|) −→ HomSets (T, |B|).
Because we are in the category of sets, there is a map, θ, so that res ◦ θ = id. Now, the maps θ and res
induce maps Θ and Res so that Res ◦ Θ = id, as shown below:
∼
HomCRA (A[S], B)
= / HomSets (S, |B|)

Res Θ res θ
  ∼  
HomCRA (A[T ], B)
= / HomSets (T, |B|).

If we let B = A[S], we get a map i = Res(idA[S] ) : A[T ] −→ A[S]. If we let B = A[T ], then, since Res is
onto, there is a map π : A[S] → A[T ] so that Res(π) = idA[T ] . It follows that i is an injection, and thus, A[S]
is an A[T ]-algebra.
We have
HomCRA[T ] (A[T ][S − T ], B) ∼
= HomSets (S − T, |B|).
The given map, |A[T ]| −→ |B|, yields a fixed map, T −→ |B|. For any given map, S − T −→ |B|, therefore,
we get a canonical map, T q (S − T ) −→ |B|, i.e., S −→ |B|, depending only on the map S − T −→ |B|.
Therefore, there is an injection
HomCRA[T ] (A[T ][S − T ], B) ,→ HomCRA (A[S], B),
and the image is just HomCRA[T ] (A[S], B). By Yoneda’s lemma, A[S] ∼
= A[T ][S − T ], as an A[T ]-algebra.
From now on, we will write HomA (B, C) instead of HomCRA (B, C) and similarly for RNGR . If X (ξ) is
a monomial, then we set X
deg(X (ξ) ) = ξ(s) ∈ Z≥0 .
s∈S
114 CHAPTER 2. RINGS AND MODULES

P
If f ∈ A[S], say f = (ξ) a(ξ) X (ξ) , then

deg(f ) = sup{deg(X (ξ) ) | a(ξ) 6= 0}.

In particular, note that deg(0) = −∞.

Proposition 2.5 The canonical map, A −→ A[S], establishes an isomorphism of A with the polynomials of
degree ≤ 0 in A[S]. Any α 6= 0 in A goes to a polynomial of degree 0, only 0 ∈ A goes to a polynomial of
degree < 0. If f, g ∈ A[S], then
(a) deg(f + g) ≤ max{deg(f ), deg(g)}.
(b) deg(f g) ≤ deg(f ) + deg(g).
If A is without zero divisors then we have equality in (b) and
(c) The units of A[S] are exactly the units of A.
(d) The ring A[S] has no zero divisors.

Proof . Since we deal with degrees and each of the two polynomials f, g involves finitely many monomials,
we may assume that S is a finite set. The map A −→ A[S] is given by α 7→ α · 1 and 1 has degree 0, so it is
trivial that we have an isomorphism of A with the polynomials of degree ≤ 0.
Say S = {1, . . . , n} and label the Xs as X1 , . . . , Xn . The monomials are lexicographically ordered:

X1a1 · · · Xnan < X1b1 · · · Xnbn

iff a1 = b1 , . . . , aj = bj and aj+1 < bj+1 (j = 0, . . . , n − 1).

P P P
(a) If f = (ξ) a(ξ) X (ξ) and g = (ξ) b(ξ) X (ξ) , then f + g = (ξ) (a(ξ) + b(ξ) )X (ξ) .
If deg(f + g) > max{deg(f ), deg(g)}, then there is some η so that

deg(X (η) ) > deg(X (ξ) ), for all ξ occurring in f and g, and a(η) + b(η) 6= 0,

a contradiction.
(b) With f and g as in (a), we have
!
X X
fg = a(η) b(η0 ) X (ξ) . (∗)
ξ 0
η,η ,
ηη 0 =ξ

Now, X X
0
deg(X (η) ) + deg(X (η ) ) = (ηη 0 )(s) = ξ(s) = deg(X (ξ) ).
s s
0
However, a(η) 6= 0 implies that deg(X ) ≤ deg(f ) and b(η0 ) 6= 0 implies that deg(X (η ) ) ≤ deg(g), and this
(η)

shows that deg(X (ξ) ) ≤ deg(f ) + deg(g), for any X (ξ) with nonzero coefficient in (∗).
When A is a domain, pick η to be the first monomial in the lexicographic ordering with X (η) of degree
0
equal to deg(f ), and similarly, pick η 0 to be the first monomial in the lexicographic ordering with X (η ) of
(η) (η 0 )
degree equal to deg(g). Then (DX), X X is the monomial occurring first in the lexicographic ordering
and of degree equal to deg(f ) + deg(g) in f g. Its coefficient is a(η) b(η0 ) 6= 0, as A has no nonzero divisors;
so, we have equality in (b).
(c) Say u ∈ A[S] is a unit. Then, there is some v ∈ A[S], so that uv = vu = 1. Consequently, deg(uv) = 0,
but deg(uv) = deg(u) + deg(v). Thus, deg(u) = deg(v) = 0 (as deg(u), deg(v) ≥ 0), i.e., u, v are units of A.
(d) If f, g 6= 0, then deg(f g) = deg(f ) + deg(g) ≥ 0, so f g 6= 0.
2.3. OPERATIONS ON MODULES; FINITENESS CONDITIONS 115

Definition 2.1 Suppose A is a commutative ring and B is a commutative A-algebra. Pick a subset, S ⊆ |B|.
The set, S, is called algebraically independent over A (or the elements of S are independent transcendentals
over A) iff the canonical map, A[S] −→ B, is a monomorphism. The set, S, is algebraically dependent over A
iff the map, A[S] −→ B, is not a monomorphism. When S = {X}, then X is transcendental , resp. algebraic
over A iff S is algebraically independent (resp. algebraically dependent) over A. The algebra, B, is a finitely
generated A-algebra iff there is a finite subset, S ⊆ |B|, so that the canonical map A[S] −→ B is surjective.

2.3 Operations on Modules; Finiteness Conditions for Rings and

Modules
Let R ∈ RNG, then by an R-module, we always mean a left R-module. Observe that a right R-module is
exactly a left Rop -module. (Here, Rop is the opposite ring, whose multiplication ·op is given by x·op y = y ·x.)
Every ring, R, is a module over itself and over Rop . By ideal, we always mean a left ideal. This is just an
R-submodule of R. If an ideal, I, is both a left and a right ideal, then we call I a two-sided ideal .
Let M be an R-module and {Mα }α∈Λ be a collection of R-submodules of M .
T
(0) α Mα is an R-submodule of M .
Q
(1) Note that we have a family of inclusion maps, Mα ,→ M ; so, we get an element of α HomR (Mα , M ).
But then, we have a map
a
Mα −→ M. (∗)
α∈Λ
P
We define α Mα , a new submodule of M called the sum of the Mα , via any of the following three
equivalent (DX) ways:
`
(a) Image of ( α∈Λ Mα −→ M ).
T
(b) {N | (1) N ⊆ M, as R-submodule; (2) Mα ⊆ N, for all α ∈ Λ. }
P
(c) { finite mα | mα ∈ Mα }.
P
Clearly, α Mα is the smallest submodule of M containing all the Mα .

(2) Let S be a subset of M . For any s ∈ S, the map P ρ 7→ ρs, from R to Rs, is a`surjection, where
Rs = {ρs | ρ ∈ R}. Thus, we get the submodule s∈S Rs (equal to the image of S R −→ M ) and
called the submodule generated by S; this module is denoted mod(S) or RS. We say that S generates
M iff RS = M and that M is` a finitely generated R-module (for short, a f.g. R-module) iff there is a
finite set, S, and a surjection S R −→ M .
`
(3) The free module ` on a set, S, is just S R. Observe that (DX) the functor from Sets to Mod(R)
given by S S R is the left adjoint of the stripping functor from Mod(R) to Sets; i.e., for every
R-module, M , we have the functorial isomorphism
a
HomR ( R, M ) ∼ = HomSets (S, |M |).
S

`
Remark: An R-module, M , is free over R (i.e., M ∼ = S R for some set S) iff M possesses a Hamel
basis over R (DX). The basis is indexed by S. To give a homomorphism of a free module to a module,
M , is the same as giving the images of a Hamel basis in M , and these images may be chosen arbitrarily.
116 CHAPTER 2. RINGS AND MODULES

(4) The transporter of S to N . Let M be an R-module, S be a subset of M and N an R-submodule of

M . The transporter of S to N , denoted (S → N ), is given by

(S → N ) = {ρ ∈ R | ρS ⊆ N }.

(Old notation: (N : S). Old terminology: residual quotient of N by S.)

When N = (0), then (S → (0)) has a special name: the annihilator of S, denoted Ann(S). Observe:
(a) (S → N ) is always an ideal of R.
(b) So, Ann(S) is an ideal of R. But if S is a submodule of M , then Ann(S) is a two-sided ideal of
R. For if ρ ∈ Ann(S) and ξ ∈ R, we have (ρξ)(s) = ρ(ξs) ⊆ ρS = (0). Thus, ρξ ∈ Ann(S).
(c) Similarly, if S is a submodule of M , then (S → N ) is a two-sided ideal of R.

An R-module, M , is finitely presented (for short, f.p.) iff there are some finite sets, S and T , and an
exact sequence a a
R −→ R −→ M −→ 0.
T S
`
This means that M is finitely generated and that the kernel, K, of the surjection, S R −→ M , is also
finitely generated. Note that f.p. implies f.g.

Definition 2.2 An R-module, M , has the ascending chain condition (ACC) (resp. the descending chain
condition (DCC)) iff every ascending chain of submodules

M 1 ⊆ M 2 ⊆ M 3 ⊆ · · · ⊆ Mn ⊆ · · · ,

eventually stabilizes (resp. every descending chain of submodules

M 1 ⊇ M 2 ⊇ M 3 ⊇ · · · ⊇ Mn ⊇ · · · ,

eventually stabilizes.) If M has the ACC it is called noetherian and if it has the DCC it is called artinian. The
module, M , has the maximal condition (resp. minimal condition) iff every nonempty family of submodules
of M has a maximal (resp. minimal) element.

Proposition 2.6 Given a module, M , over R consider all the statements

(1) M is noetherian (has the ACC).
(2) M has the maximal property.
(3) Every submodule of M is finitely generated.
(4) M is artinian (has the DCC).
(5) M has the minimal property.
Then, (1)–(3) are equivalent and (4) and (5) are equivalent.

Proof . (1) =⇒ (2). Let F be a given nonempty family of submodules of M . If there is no maximal element
of F, given M1 ∈ F, there is some M2 in F so that M1 < M2 . Repeating the argument, we find there is some
M3 ∈ F so that M2 < M3 , and by induction, for every n ≥ 1, we find some Mn+1 ∈ F so that Mn < Mn+1 .
So, we find an infinite strictly ascending chain

M1 < M2 < M3 < · · · < Mt < · · · ,

contradicting (1).
2.3. OPERATIONS ON MODULES; FINITENESS CONDITIONS 117

(2) =⇒ (3). Observe that the maximal property for M is inherited by every submodule.
Claim: The maximal property for a module implies that it is finitely generated. If so, we are done. Pick
M with the maximal property and let

F = {N ⊆ M | N is a submodule of N and N is f.g.}

The family, F, is nonempty since for every m ∈ M , the module Rm ⊆ M is a submodule of M generated
by {m}, and so, Rm ∈ F. Now, F has a maximal element, say T . If T 6= M , then there is some m ∈ M
with m ∈/ T . But now, T + Rm > T and T + Rm is finitely generated by the generators of T plus the new
generator m, a contradiction. Therefore, M = T ∈ F; and so, M is f.g.
(3) =⇒ (1). Take an ascending chain,

M1 ⊆ M2 ⊆ · · · ⊆ M r ⊆ · · · ,
S∞
and look at N = i=1 Mi . Note that N is a submodule of M . So, by (3), the module N is finitely
generated. Consequently, there is some t so that Mt contains all the generators of N , and then we have
N ⊆ Mt ⊆ Mr ⊆ N , for all r ≥ t. Therefore, Mt = Mr = N for all r ≥ t.
(4) =⇒ (5). The proof is obtained from the proof of (1) =⇒ (2) mutatis mutandis.
(5) =⇒ (4). Say
M1 ⊇ M2 ⊇ M3 ⊇ · · · ⊇ Mr ⊇ · · ·
is a descending chain in M . Let F = {Mi | i ≥ 1}. By (5), the family F has a minimal element, say Mr .
Then, it is clear that the chain stabilizes at r.
Proposition 2.7 Let M be a module and write (α), (β) and (γ) for the finiteness properties
(α) finite generation
(β) ACC
(γ) DCC
Then,
(A) If M has any of (α), (β), (γ), so does every factor module of M .
(B) If M has (β) or (γ), so does every submodule of M .
(C) Say N ⊆ M is a submodule and N and M/N have any one of (α), (β), (γ). Then, M also has the
same property.
Proof . (A) If M is f.g., then there is a surjection
a
R −→ M, with #(S) finite.
S

Let M be a factor module of M ; there is a surjection M −→ M . By composition, we get a surjection

a
R −→ M −→ M ,
S

and so, M is f.g. Any ascending (resp. descending) chain in M lifts to a similar chain of M . The rest is
clear.
(B) Any ascending (resp. descending) chain in N ⊆ M is a similar chain of M ; the rest is clear.
118 CHAPTER 2. RINGS AND MODULES

(C) Say N and M/N have (α). Then, there are two finite (disjoint) sets, S and T , and surjections
a a
R −→ N −→ 0 and R −→ M/N −→ 0.
S T

Consider the diagram:

a
R

T
θ


 
M / M/N /0


0 .
` `
As T R is free, there exists an arrow, θ : T R −→ M , shown above, and the diagram commutes. Now,
consider the diagram:
a a a
0 / R / R / R /0
S S∪T T

  
0 /N /M / M/N /0

 
0 0 .
We obtain the middle vertical arrow by the map θ and the set map S −→ M (via S −→ N ,→ M ). By
construction, our diagram commutes. We claim that the middle arrow is surjective.
` For this, we chase the
diagram: Choose m in M and map m to m ∈ M/N . There is some ξ ∈ T R so that ξ 7→ m. However,
`
ξ comes from η ∈ S∪T R. Let ηe be the image in M of η. Since the diagram is commutative, ηe = m,
and so, ηe − m maps
` to 0 in` M/N . Consequently, there is some n ∈`N so that ηe − m = n. Yet, n comes
from some ρ in S R ,→ S∪T R (i.e., ρe = n). Consider η − ρ ∈ S∪T R. The image of η − ρ in M is
ηe − ρe = m + n − n = m, proving surjectivity. As S ∪ T is finite, the module, M , has (α).
Next, assume N and M/N have (β). Let

M1 ⊆ M2 ⊆ M3 ⊆ · · · ⊆ M r ⊆ · · ·

be an ascending chain in M . Write M j for the image of Mj in M/N . By the ACC in M/N , there is some
t ≥ 1 so that M j = M t for all j ≥ t. If we let Nj = Mj ∩ N , we get an ascending chain in N . By the ACC
in N , this chains stabilizes, i.e., there is some s ≥ 1 so that Nj = Ns for all j ≥ s. Let r = max{s, t}. We
claim that Mj = Mr for all j ≥ r. We have the diagram

0 / Nr / Mr / Mr /0

  
0 / Nj / Mj / Mj / 0,

where the rows are exact and the vertical arrows on the left and on the right are surjections. A diagram
chase yields the fact that the middle vertical arrow is also surjective.
Finally, assume N and M/N have (γ). The same argument works with the arrows and inclusions reversed.
2.3. OPERATIONS ON MODULES; FINITENESS CONDITIONS 119

`
Corollary 2.8 Say {Mλ }λ∈Λ is a family of R-modules. Then, λ Mλ has one of (α), (β), (γ) iff each Mλ
has the corresponding property and Λ is finite.
Proof . We have a surjection a
Mλ −→ Mµ −→ 0, for any fixed µ.
λ
`
Consequently, (α), (β), (γ) for λ Mλ implies (α), (β), (γ) for Mµ , by the previous proposition. It remains
to prove that Λ is finite.
`
First, assume
` that ` λ Mλ has (α), and further assume that Λ is infinite. There is some finite set, S, and
a surjection S R −→ λ Mλ . We may assume ` that S = {1, . . . , n}, for some positive integer,
` n. Then,
we have the canonical basis ` vectors, e1 , . . . , en , of S R, and their images e1 , . . . , en generate λ Mλ . Each
image ei is a finite tuple in λ Mλ . Yet, the union ` of the finite index sets so chosen is again finite and for
any µ not`in this finite set, the image of Mµ in λ Mλ is not covered. This contradicts the fact that the ei ’s
generate λ Mλ , and so, Λ must be finite.
We treat (β) and (γ) together. Again, assume that Λ is infinite. Then, there is a countably infinite
subset of Λ, denote it {λ1 , λ2 , . . .}, and the chains
Mλ1 < Mλ1 q Mλ2 < Mλ1 q Mλ2 q Mλ3 < · · ·
and
∞
a a a
Mλ j > Mλj > Mλ j > · · ·
j=1 j6=1 j6=1,2
`
are infinite ascending (resp. descending) chains of λ Mλ , a contradiction.
Finally, assume that each Mλ has (α) or (β) or (γ) and that Λ is finite. We use induction on #(Λ).
Consider the exact sequence a a
0 −→ Mj −→ Mj −→ M1 −→ 0.
j6=1 j∈Λ

Then, (α) (resp. (β), (γ)) holds for the right end by hypothesis, and it also holds for the left end, by
induction; so, (α) (resp. (β), (γ)) holds in the middle.
Corollary 2.9 Say R is noetherian (has the ACC on ideals) or artinian (has the DCC on ideals). Then,
`
(1) Every f.g. free module, S R, is noetherian, resp. artinian, as R-module (remember, #(S) < ∞).
(2) Every f.g. R-module is noetherian, resp. artinian.
(3) When R is noetherian, every f.g. R-module is f.p. Finitely presented modules are always f.g.
Proof . (1) and (2) are trivial from Corollary 2.8.
As for (3), that f.p. implies f.g. is clear by the definition. Say M is f.g. Then, we have an exact sequence
a
0 −→ K −→ R −→ M −→ 0,
S
`
with #(S) < ∞. By (1), the module S R is noetherian; by Proposition 2.6, the module K is f.g. Thus,
there is some finite set, T , so that a
R −→ K −→ 0 is exact.
T
By splicing the two sequences, we get the exact sequence
a a
R −→ R −→ M −→ 0,
T S

which shows that M is f.p.

 120 CHAPTER 2. RINGS AND MODULES

 Counter-examples.

(1) A subring of a noetherian ring need not be a noetherian ring. Take A = C[X1 , X2 , . . . , Xn , . . .] the
polynomial ring in countably many variables, and let K = Frac(A). Every field is noetherian as a ring
(a field only has two ideals, (0) and itself). We have A ⊆ K, yet A is not noetherian, for we claim that
we have the ascending chain of ideals

(X1 ) < (X1 , X2 ) < (X1 , X2 , X3 ) < · · ·

Would this chain stabilize, then we would have (X1 , . . . , Xn ) = (X1 , . . . , Xn , Xn+1 ), for some n ≥ 1.
Now, there would be some polynomials f1 , . . . , fn in A so that

Xn+1 = f1 X1 + · · · + fn Xn .

Map A to C via the unique homomorphism sending Xj to 0 for j = 1, . . . , n, and sending Xj to 1 for
j > n. We get 1 = 0, a contradiction. Therefore, the chain is strictly ascending.
(2) A module which is finitely generated need not be finitely presented. Let A = C[X1 , . . . , Xn , . . .], the
polynomial algebra over C in countably many variables. Then, C is an A-module because of the exact
sequence
0 −→ I = (X1 , . . . , Xn , . . .) −→ A −→ C −→ 0,
in which the map A −→ C is given by f 7→ f (0); the ring A acts on C via f · z = f (0)z, where f ∈ A
and z ∈ C. Assume that C is finitely presented. Then, there are some finite sets, S and T , and an
exact sequence a a
A −→ A −→ C −→ 0.
T S

We get the diagram

a a
A / A /C /0
T S

ϕ Θ
 
0 /I /A /C /0
`
To construct the vertical arrows, let e1 , . . . , es be the usual generators of S A. If z1 , . . . , zs ∈ C are
their images, there exist λ1 , . . . , λs ∈ A so that
s
X s
X
λj ej 7→ λj (0)zj = 1.
j=1 j=1
Ps
We have the (C-linear) map, C −→ A, so our zj lie in `A. Then, we have j=1 λj (0)zj `
= 1, in A.
If we send ej 7→ zj ∈ A, we get an A-linear map, Θ : S A → A, and there is some ξ ∈ S A with
Θ(ξ) = 1 ∈ A. Namely, take
Xs
ξ= λj (0)ej .
j=1

But then, Θ is onto,

` because its image is an ideal which contains 1. A diagram chase implies that there
exists some ϕ : T A → I rendering the diagram commutative. Another diagram chase gives the fact
that ϕ is surjective. But then, I is finitely generated, a contradiction. Therefore, C is not f.p. (over
A).

Remark: The difficulty is that A is much “bigger” than C, and thus, the surjection A −→ C has to
“kill” an infinite number of independent elements.
2.3. OPERATIONS ON MODULES; FINITENESS CONDITIONS 121

Consider the category, Mod(R). We can also look at subcategories of Mod(R) having some additional
properties. For example, a subcategory, C, of Mod(R) is a localizing subcategory iff
(a) Whenever M and N ∈ Ob(C) and θ : M → N is a morphism of C, then Ker θ and Coker θ = (N/Im θ)
lie in Ob(C) and the morphisms Ker θ −→ M and N −→ Coker θ are arrows of C.
(b) Whenever
0 −→ M 0 −→ M −→ M 00 −→ 0 is exact (in Mod(R))
0 00
and M , M ∈ Ob(C), then M ∈ Ob(C) and the sequence is exact in C.

Example: Let C = Modfg (R) be the full subcategory of finitely generated R-modules, where R is noetherian.
The reader should check that C is a localizing subcategory.
Recall that an R-module is a simple iff it has no nontrivial submodules; a composition series is a finite
descending chain
M = M0 > M1 > M2 > · · · > Mt = (0)
in which all the factors Mj /Mj+1 are simple. We know from the Jordan–Hölder theorem that the number
of composition factors, t, is an invariant and the composition factors are unique (up to isomorphism and
rearrangement). We set λR (M ) = t, and call it the length of M ; if M does not have a composition series,
set λR (M ) = ∞.
Say C is a localizing subcategory of Mod(R) and ϕ is a function on Ob(C) to some fixed abelian group,
A.

Definition 2.3 The function, ϕ, is an Euler function iff whenever

0 −→ M 0 −→ M −→ M 00 −→ 0 is exact in C,

we have ϕ(M ) = ϕ(M 0 ) + ϕ(M 00 ).

Proposition 2.10 A necessary and sufficient condition that a module, M , have finite length is that M has
both ACC and DCC on submodules. The function λR on the full subcategory of finite-length modules (which
is a localizing subcategory), is an Euler function. If ϕ is an Euler function on some localizing subcategory of
Mod(R) and if
(E) 0 −→ M1 −→ M2 −→ · · · −→ Mt −→ 0
is an exact sequence in this subcategory, then
t
X
χϕ ((E)) = (−1)j ϕ(Mj ) = 0.
j=1

Proof . First, assume that M has finite length. We prove the ACC and the DCC by induction on λR (M ).
If λR (M ) = 1, then M is simple, so the ACC and the DCC hold trivially. Assume that this is true for
λR (M ) = t, and take λR (M ) = t + 1. We have a composition series

M = M0 > M1 > M2 > · · · > Mt+1 = (0),

and so, λR (M1 ) = t and λR (M/M1 ) = 1. But the sequence

0 −→ M1 −→ M −→ M/M1 −→ 0 is exact,

and the ACC and DCC hold on the ends, by induction. Therefore, they hold in the middle.
Now, assume that the DCC and the ACC hold for M . Let

F = {N ⊆ M | N 6= M, N is a submodule of M .}
122 CHAPTER 2. RINGS AND MODULES

The family F is nonempty (the trivial module (0) is in F) and by the ACC, it has a maximal element, M1 ;
so, M/M1 is simple. Apply the same argument to M1 : We get M2 < M1 with M1 /M2 simple. By induction,
we get a strictly descending chain

M = M0 > M1 > M2 > · · · > Mt > · · ·

However, by the DCC, this chain must stabilize. Now, if it stabilizes at Mt , we must have Mt = (0), since
otherwise we could repeat the first step in the argument for Mt . This proves that λR (M ) = t < ∞.
Say 0 −→ M 0 −→ M −→ M 00 −→ 0 is exact in Modfl (R). Pick a composition series for M 00 . We get a
strictly descending chain
M 00 = M000 > M100 > M200 > · · · > Mt00 = (0).
By the second homomorphism theorem, we get a lifted sequence

M = M0 > M1 > M2 > · · · > Mt = M 0 ,

and if we pick a composition series for M 0 , we get the following composition series with
s + t = λR (M 0 ) + λR (M 00 ) factors, as required:

M = M0 > M1 > M2 > · · · > Mt = M 0 > M10 > M20 > · · · > Ms0 = (0).

Say
θ
(E) 0 −→ M1 −→ M2 −→ · · · −→ Mt−2 −→ Mt−1 −→ Mt −→ 0
is an exact sequence. Then, we have the two exact sequences

(E 0 ) 0 −→ M1 −→ M2 −→ · · · −→ Mt−2 −→ Ker θ −→ 0 and

(E 00 ) 0 −→ Ker θ −→ Mt−1 −→ Mt −→ 0.

The cases t = 1, 2, 3 are trivial (DX). By using induction on t, we see that the proposition is true for (E 0 )
and (E 00 ). Thus, we get

t−2
X
(−1)j ϕ(Mj ) + (−1)t−1 ϕ(Ker θ) = 0 and
j=1
ϕ(Ker θ) = ϕ(Mt−1 ) − ϕ(Mt ).

If we add the first equation to (−1)t times the second equation we get

t−2
X
(−1)j ϕ(Mj ) = (−1)t ϕ(Mt−1 ) − (−1)t ϕ(Mt ),
j=1

and so,
t−2
X
χϕ ((E)) = (−1)j ϕ(Mj ) + (−1)t−1 ϕ(Mt−1 ) + (−1)t ϕ(Mt ) = 0,
j=1

as claimed.

Theorem 2.11 (Hilbert Basis Theorem (1890)) If A is a commutative noetherian ring, then so is the
polynomial ring A[X].
2.3. OPERATIONS ON MODULES; FINITENESS CONDITIONS 123

Proof . Let An be the submodule of A[X] consisting of the polynomials of degree at most n. The module,
An , is a free module over A (for example, 1, X, X 2 , . . . , X n is a basis of An ). If A is an ideal of A[X], then
A ∩ An is a submodule of An . As An (being finitely generated over A) is a noetherian module, A ∩ An is
also finitely generated, say by α1 , α2 , . . . , ακ(n) (∈ A[X]). If f ∈ A and deg(f ) ≤ n, then f ∈ An ; so,

f = a1 α1 + · · · + aκ(n) ακ(n) , with aj ∈ A.

Now, let A∗ be the subset of A consisting of all a ∈ A so that either a = 0 or there is some polynomial f in
A having a as its leading coefficient, i.e., f = aX r + O(X r−1 ). We claim that A∗ is an ideal of A.
Say a and b are in A∗ . Then, there are some polynomials f, g ∈ A so that f = aX r + O(X r−1 ) and
g = bX s + O(X s−1 ). Take t = max{r, s}. Then, X t−r f ∈ A and X t−s g ∈ A, since A is an ideal. But,

X t−r f = aX t + O(X t−1 ) and X t−s g = bX t + O(X t−1 ),

and this implies that a±b ∈ A∗ , as a±b is the leading coefficient of X t−r f ±X t−s g ∈ A. If λ ∈ A and a ∈ A∗ ,
then it is clear that λa ∈ A∗ . Therefore, A∗ is indeed an ideal in A. Now, A is a noetherian ring, therefore
A∗ isPfinitely generated as an ideal. So, there exist β1 , . . . , βt ∈ A∗ ⊆ A, such that for any β ∈ A∗ , we have
t
β = i=1 λi βi , for some λi ∈ A. Now, by definition of A∗ , for every βi ∈ A∗ , there is some fi (X) ∈ A
so that fi (X) = βi X ni + O(X ni −1 ). Let n = max{n1 , . . . , nt } and consider the generators α1 , . . . , ακ(n) of
An = An ∩ A.
Claim: The set {α1 , . . . , ακ(n) , f1 , . . . , ft } generates A.
Pick some g ∈ A. Then, g(X) = βX r + O(X r−1 ), for some r. If r ≤ n, then g(X) ∈ An , and thus,
g = λ1 α1 + · · · + λκ(n) ακ(n) , with λi ∈ A. Say r > n. Now, β ∈ A∗ , so there are elements λ1 , . . . , λt ∈ A
such that β = λ1 β1 + · · · + λt βt . Consider the polynomial
t
X
P (X) = λi X r−ni fi (X),
i=1

and examine g(X) − P (X). We have

t
X
g(X) − P (X) = βX r − λi X r−ni fi (X) + O(X r−1 ) = O(X r−1 ),
i=1

and thus there is a P (X) ∈ (f1 , . . . , ft ) so that deg(g(X) − P (X)) ≤ r − 1. By repeating this process, after
finitely many steps, we get
Xt
g(X) − hi (X)fi (X) = O(X ≤n ).
i=1

Since this polynomial belongs to A, we deduce that it belongs to An . However, An is generated by

α1 , . . . , ακ(n) , and so, g(X) is an A[X]-linear combination of the fi (X)’s and the αj (X)’s, as desired.

Remark: The reader should reprove Hilbert’s theorem using the same argument but involving ascending
chains. This is Noether’s argument (DX).

Corollary 2.12 Say R ∈ RNG. If R is noetherian, so is RhXi.

Proof . We have RhXi = R[X], and the same proof works.

Corollary 2.13 If A (in CR) is noetherian, then so is A[X1 , . . . , Xn ].

Corollary 2.14 (Hilbert’s original theorem) The polynomial ring Z[X1 , . . . , Xn ] is noetherian. If k is a field
(Hilbert chose C) then k[X1 , . . . , Xn ] is noetherian.
 124 CHAPTER 2. RINGS AND MODULES

Corollary 2.15 (of the proof–(DX)) If k is a field, then k[X] is a PID.

Corollary 2.16 Say A is a noetherian ring (A ∈ CR) and B is a finitely generated A-algebra. Then, B is
a noetherian ring.

Proof . The hypothesis means that B is a homomorphic image of a polynomial ring C = A[X1 , . . . , Xn ] in
such a way that the diagram
C _@
θ /B
@@ ~~?
@@ ~
@@ ~~
@ ~~~
A
commutes, where A −→ C is the natural injection of A into C = A[X1 , . . . , Xn ]. The ring A[X1 , . . . , Xn ] is
noetherian, by Corollary 2.13. The map θ makes B into a C-module and B is finitely generated as C-module.
Now, C-submodules are exactly the ideals of B (DX). Since B is finitely generated as C-module and C is
noetherian, this implies that B is a noetherian C-module. Therefore, the ACC on C-submodules holds, and
since these are ideals of B, the ring B is noetherian.

 To be finitely generated as A-algebra is very different from being finitely generated as A-module.

Given an exact sequence of modules,

0 −→ M 0 −→ M −→ M 00 −→ 0,

there are situations where it is useful to know that M 0 is f.g, given that M and M 00 satisfy certain finiteness
conditions. We will give below a proposition to this effect. The proof makes use of Schanuel’s lemma. First,
introduce the following terminology: Given a module M , call an exact sequence

0 −→ K −→ F −→ M −→ 0,

a presentation of M if F is free. Note that M is f.p. iff there is a presentation of M in which both F and
K are f.g.

Proposition 2.17 If M is a Λ-module, then M is f.p. iff every presentation

0 −→ K −→ F −→ M −→ 0, (∗)

in which F is f.g. has K f.g. and at least one such exists.

Proof . The direction (⇐) is clear.

(⇒). Say M is f.p.; we have an exact sequence

0 −→ K 0 −→ F 0 −→ M −→ 0,

where both K 0 and F 0 are f.g. and F 0 is free. Pick any presentation, (∗), with F f.g. If we apply Schanuel’s
lemma, we get a a
F0 K∼ =F K 0,
But, the righthand side is f.g. and K is a quotient of the left hand side, so it must be f.g.

Remark: The forward implication of Proposition 2.17 also holds even if F is not free. A simple proof using the
snake lemma will be given at the end of Section 2.5.
2.4. PROJECTIVE AND INJECTIVE MODULES 125

2.4 Projective and Injective Modules

Let F : Mod(R) → Mod(S) be a functor (where R, S ∈ RNG). Say

0 / M0 /M / M 00 /0 (∗)

is exact in Mod(R). What about

0 / F (M 0 ) / F (M ) / F (M 00 ) / 0? (∗∗)

(1) The sequence (∗∗) is a complex if F is an additive functor. (Observe that HomR (M, N ) is an abelian
F (·)
group, so is HomS (F (M ), F (N )). We say F is additive iff HomR (M, N ) −→ HomS (F (M ), F (N )) is a
homomorphism, i.e., preserves addition.)
(2) The functor, F , is exact iff when (∗) is exact then (∗∗) is exact (the definition for cofunctors is identical).
(3) The functor, F , is a left-exact if whenever the sequence

0 / M0 /M / M 00

is exact, then the sequence

0 / F (M 0 ) / F (M ) / F (M 00 )

is exact, right exact if whenever the sequence

M0 /M / M 00 /0

is exact, then the sequence

F (M 0 ) / F (M ) / F (M 00 ) /0

is exact.
(4) The functor, F , is half-exact (same definition for cofunctors) iff when (∗) is exact

F (M 0 ) −→ F (M ) −→ F (M 00 )

is still exact.
(5) The cofunctor, G, is left exact if whenever the sequence

M0 /M / M 00 /0

is exact, then the sequence

0 / G(M 00 ) / G(M ) / G(M 0 )

is exact, right exact if if whenever the sequence

0 / M0 /M / M 00

is exact, then the sequence

G(M 00 ) / G(M ) / G(M 0 ) /0

is exact.
126 CHAPTER 2. RINGS AND MODULES

Remark: The chirality of a functor is determined by the image category.

Examples of exact (left-exact, right-exact, etc.) functors:

(1) Let F : Mod(R) → Mod(Z) be given by: F (M ) = underlying abelian group of M . The functor F is
exact.
(2) Take a set, Λ, and look at

Mod(R)Λ = {{Mα }α∈Λ | each Mα ∈ Mod(R)},

together with obvious morphisms. We have two functors from Mod(R)Λ to Mod(R). They are:
Y a
{Mα } Mα and {Mα } Mα .
α α

Both are exact functors (this is special to modules). The next proposition is a most important example of
left-exact functors:

Proposition 2.18 Fix an R-module, N . The functor from Mod(R) to Ab (resp. cofunctor from Mod(R)
to Ab) given by M HomR (N, M ) (resp. M HomR (M, N )) is left-exact (N.B.: both are left-exact).

Proof . Consider the case of a cofunctor (the case of a functor is left to the reader (DX)). Assume that
ϕ ψ
0 −→ M 0 −→ M −→ M 00 −→ 0

is exact. Look at the sequence obtained by applying HomR (−, N ) to the above exact sequence:
Ψ Φ
0 −→ HomR (M 00 , N ) −→ HomR (M, N ) −→ HomR (M 0 , N ) −→ 0,

where Φ = − ◦ ϕ and Ψ = − ◦ ψ. Pick α ∈ HomR (M 00 , N ) and assume that Ψ(α) = 0. We have the
commutative diagram
MD
ψ
/ M 00
DD
DD
D
Ψ(α) DD
α
" 
N
ψ
and since M −→ M 00 is surjective, we deduce that α = 0. Now, pick β ∈ HomR (M, N ) and assume that
Φ(β) = 0. We have the commutative diagram (see argument below)

M 0D
ϕ
/ M ψ / M 00
DD y
DD yy
DD β yyyy
Φ(β) D!  |yy β
N .

Since Φ(β) = 0, we have Im ϕ ⊆ Ker β; so, by the first homomorphism theorem, there is a homomorphism
β : M/M 0 = M 00 → N , as shown, making the above diagram commute. Thus, Ψ(β) = β ◦ ψ = β, and so,
β ∈ Im Ψ.
There may be some modules, N , so that our Hom functors become exact as functors of M . This is the
case for the class of R-modules introduced in the next definition:

Definition 2.4 A module, P , is projective (over R) iff the functor M HomR (P, M ) is exact. A module,
Q, is injective (over R) iff the cofunctor M HomR (M, Q) is exact.
2.4. PROJECTIVE AND INJECTIVE MODULES 127

Remarks:
(1) Any free R-module is projective over R.
` `
Proof
Q . Say F = SQ R. Consider the functor M HomR ( S R,
QM ). The righthand side is equal to
S Hom R (R, M ) = S M , but we know that the functor M S M is exact.

(2) A functor is left-exact iff it preserves the left-exactness of a short left-exact sequence (resp. a cofunctor
is left-exact iff it transforms a short right-exact sequence into a left-exact sequence), and mutatis
mutandis for right exact functors or cofunctors.
(3) Compositions of left (resp. right) exact functors are left (resp. right) exact. Similarly, compositions of
exact functors are exact.

We say that an exact sequence

i p
0 −→ M 0 −→ M −→ M 00 −→ 0

splits iff there is a map σ : M 00 → M so that p ◦ σ = idM 00 . Such a map, σ, is called a splitting of the
sequence. The following properties are equivalent (DX):

Proposition 2.19 (1) The sequence

i p
0 −→ M 0 −→ M −→ M 00 −→ 0

splits.
(2) Given our sequence as in (1),
i p
0 −→ M 0 −→ M −→ M 00 −→ 0
there is a map π : M → M 0 so that π ◦ i = idM 0 .
(3) There is an isomorphism M 0 q M 00 ∼
= M.

Proposition 2.20 Let P be an R-module, then the following are equivalent:

(1) P is projective over R.
(2) Given a diagram
P
ξ

M / M 00 / 0,

there exists a map, θ : P → M , lifting ξ, rendering the diagram commutative (lifting property).
(3) Any exact sequence 0 −→ M 0 −→ M −→ P −→ 0 splits.

(4) There exists a free module, F , and another module, Pe, so that P q Pe ∼
= F.

Proof . (1) ⇒ (2). Given the projective module, P and the diagram

P
ξ

M / M 00 / 0,
128 CHAPTER 2. RINGS AND MODULES

the exact sequence gives the map

HomR (P, M ) −→ HomR (P, M 00 ) (†)

and the diagram gives an element, ξ, of HomR (P, M 00 ). But P is projective, and so, (†) is surjective.
Consequently, ξ comes from some η ∈ HomR (P, M ), proving the lifting property.
(2) ⇒ (3). Given an exact sequence

0 −→ M 0 −→ M −→ P −→ 0,

we get the diagram

P

M /P / 0.
The lifting property gives the backwards map P −→ M , as required.
(3) ⇒ (4). Given P , there is a free module, F . and a surjection, F −→ P . We get the exact sequence

0 −→ Pe −→ F −→ P −→ 0,

where Pe = Ker (F −→ P ). By hypothesis, this sequence splits. Therefore, by property (3) of Proposition
2.19, we have F ∼
= P q Pe.
`
(4) ⇒ (1). We have F ∼= P q Pe, for some free R-module, F . Now, F = S R, for some set, S, and so,
for any N , Y Y
HomR (F, N ) = HomR (R, N ) = N.
S S
Q
The functor N HomR (F, N ) is exact; yet, this functor is N HomR (P, N ) HomR (Pe, N ). Assume that
the sequence
0 −→ M 0 −→ M −→ M 00 −→ 0 is exact,
we need to show that HomR (P, M ) −→ HomR (P, M 00 ) is surjective. This follows by chasing the diagram
(DX):
/ Hom(P, M ) Q Hom(Pe, M )
∼
=
Hom(F, M )

 
/ Hom(P, M 00 ) Q Hom(Pe, M 00 )
∼
=
Hom(F, M 00 )


0 .

Given an R-module, M , a projective resolution (resp. a free resolution) of M is an exact (possibly infinite)
sequence (= acyclic resolution) of modules

· · · −→ Pn −→ · · · −→ P2 −→ P1 −→ P0 −→ M,

with all the Pi0 s projective (resp. free)

Corollary 2.21 Every R-module possesses a projective resolution (even a free resolution).
 2.4. PROJECTIVE AND INJECTIVE MODULES 129

Proof . Since free modules are projective, it is enough to show that free resolutions exist. Find a free module,
F0 , so that there is a surjection, F0 −→ M . Let M1 = Ker (F0 −→ M ), and repeat the process. We get a
free module, F1 , and a surjection, F1 −→ M1 . By splicing the two exact sequences
0 −→ M1 −→ F0 −→ M −→ 0 and F1 −→ M1 −→ 0, we get the exact sequence F1 −→ F0 −→ M −→ 0.
We obtain a free resolution by repeating the above process.
`
Proposition 2.22 Given a family, {Pα }α∈Λ , of modules, the coproduct α Pα is projective iff each Pα is
projective.

Proof . Assume that each Pα is projective. This means that for every α, the functor M HomR (Pα , M ) is
exact. Q
As the product functor is exact and composition of exact functors is exact, the functor
M α HomR (Pα , M ) is exact. But
Y a
HomR (Pα , M ) = HomR ( Pα , M ).
α α
`
Therefore, Pα is projective.
α
`
Conversely, assume that α Pα is projective. By Proposition 2.20, there is a free module, F , and another
(projective) module, Pe, with a  a
Pα Pe ∼
= F.
α

Pick any β, then  

a a a
Pβ  Pα Pe ∼
= F.
α6=β

Again, by Proposition 2.20, the module Pβ is projective.

 The product of projectives need not be projective. (See, HW Problem V.B.VI.)

Remark: Projective modules can be viewed as a natural generalization of free modules. The following characteri-
zation of projective modules in terms of linear forms is an another illustration of this fact. Moreover, this proposition
can used to prove that invertible ideals of an integral domain are precisely the projective ideals, a fact that plays an
important role in the theory of Dedekind rings (see Chapter 3, Section 3.6).

Proposition 2.23 An R-module, M , is projective iff there exists a family, {ei }i∈I , of elements of M and a family,
{ϕi : M → R}i∈I , of R-linear maps such that
(i) For all m ∈ M , we have ϕi (m) = 0, for all but finitely i ∈ I.
P
(ii) For all m ∈ M , we have m = i ϕi (m)ei .
In particular, M is generated by the family {ei }i∈I .

Proof . First, assume that M is projective and let ψ : F → M be a surjection from a free R-module, F . The map,
ψ, splits, we let ϕ : M → F be its splitting. If {fi }i∈I is a basis of F , we set ei = ψ(fi ). Now, for each m ∈ M , the
element ϕ(m) can be written uniquely as X
ϕ(m) = rk f k ,
k

where rk ∈ R and rk = 0 for all but finitely many k. Define ϕi : M → R by ϕi (m) = ri ; it is clear that ϕi is R-linear
and that (i) holds. For every m ∈ M , we have
X  X
m = (ψ ◦ ϕ)(m) = ψ rk f k = ϕk (m)ek ,
k k
130 CHAPTER 2. RINGS AND MODULES

which is (ii). Of course, this also shows the ek generate M .

`
Conversely, assume (i) and (ii). Consider the free module F = i∈I R and let {fi }i∈I be a basis of F . Define
the map ψ : F → M via fi 7→ ei . To prove that M is projective, by Proposition 2.20 (4), it is enough to find a map
ϕ : M → F with ψ ◦ ϕ = 1M . Define ϕ via
X
ϕ(m) = ϕk (m)fk .
k

The sum on the righthand side is well-defined because of (i), and by (ii),
X
(ψ ◦ ϕ)(m) = ϕk (m)ek = m.
k

Therefore, M is a cofactor of a free module, so it is projective.

We would like to test submodules, L, of M as to whether L = M by testing via surjections M −→ N .

That is, suppose we know that for every N and every surjection M −→ N we have L ,→ M −→ N is also
surjective. How restrictive can we be with the N ’s, yet get a viable test?

There may be some superfluous N , e.g., those N for which M −→ N −→ 0 automatically implies that
L −→ M −→ N is surjective. There may even be some such N ’s that work for all L. Thus, it is preferable
to fix attention on N and seek small enough M so that N matters in the testing of all L. This yields a piece
of the following definition:

Definition 2.5 A surjection, M −→ N , is a minimal (essential , or covering) surjection iff for all L ⊆ M ,
whenever L −→ M −→ N is surjective, we can conclude L = M . A submodule, K, is small (superfluous) iff
for every submodule, L ⊆ M , when L + K = M , then L = M . A submodule, K, is large (essential ) iff for
all submodules, L ⊆ M , when L ∩ K = (0), then L = (0). The injection K −→ M is essential (minimal ) iff
K is large.

Proposition 2.24 The following are equivalent for surjections θ : M → N :

θ
(1) M −→ N is a minimal surjection.

(2) Ker θ is small.

(3) Coker (L −→ M −→ N ) = (0) implies Coker (L −→ M ) = (0), for any submodule, L ⊆ M .

Proof . (1) ⇒ (2). Pick L, and assume L + Ker θ = M . So, θ(L) = θ(M ) = N . Thus, L = M , by (1), which
shows that Ker θ is small.

(2) ⇒ (3). Say L ⊆ M and assume that Coker (L −→ M −→ N ) = (0). Therefore, N = Im (L −→ N ),

and we deduce that
L + Ker θ = M,

by the second homomorphism theorem. By (2), we get L = M ; so, Coker (L −→ M ) = (0).

(3) ⇒ (1). This is just the definition.

Definition 2.6 A surjection P −→ N is a projective cover iff

(1) The module P is projective

(2) It is a minimal surjection.

 2.4. PROJECTIVE AND INJECTIVE MODULES 131

 Projective covers may not exist. For example, Z −→ Z/2Z is a surjection and Z is projective. If
P −→ Z/2Z is a projective cover, then the lemma below implies that P is torsion-free. Hence, we
can replace P −→ Z/2Z by Z −→ Z/2Z. However, the following argument now shows that Z/2Z has no
projective cover. We have the surjection θ : Z −→ Z/2Z. This is not a minimal surjection because 2Z is not
small. (Clearly, Ker θ = 2Z; so, say L = dZ and dZ + 2Z = Z. Then, (d, 2) = 1, so d is odd. Yet, dZ = Z
θ
only when d = 1. Thus, the module 2Z is not small.) Now, suppose dZ −→ Z/2Z is surjective, then d must
be odd. If kZ ⊆ dZ maps onto Z/2Z, then, as Ker θ = 2dZ, we get (k, 2d) = d. Let b = k/d; the integer b
must be odd. Then, the diagram

0 / 2Z / Z / Z/2Z / 0

d d d
  
0 / 2dZ / dZ / Z/2Z / 0,

(in which the vertical arrows are isomorphisms: multiply by d) shows that the inclusion kZ ⊆ dZ corresponds
to the inclusion bZ ⊆ Z. Our previous argument implies b = 1; so, k = d, and dZ −→ Z/2Z is not minimal.

Lemma 2.25 If R has no zero divisors and P is a projective R-module then P is torsion-free.

Proof . Since the torsion-free property is inherited by submodules, we may assume that P is a free module.
Moreover, coproducts of torsion-free modules are torsion-free, so we may assume that P = R. But, R has
no zero-divisors; so, it is torsion-free.

Proposition 2.26 Say R is a ring and J (R) is its Jacobson radical (i.e., J (R) is equal to the intersection
of all maximal ideals of R). Then, the surjection R −→ R/J (R) is a projective cover. In particular, when
R is commutative local, then R −→ R/mR is a projective cover.

Proof . Pick L ⊆ R, a submodule of R, i.e., an ideal of R, such that L + J (R) = R. If L 6= R, then L ⊆ M,

where M is some maximal ideal. But, J (R) ⊆ M, and so L + J (R) ⊆ M. The latter inclusion shows that
L + J (R) 6= R, a contradiction; so, J (R) is small.
For injective modules, the situation is nearly dual to the projective case. It is exactly dual as far as
categorical properties are concerned. However, the notion of free module is not categorical, and so, results
about projective modules involving free modules have no counterpart for injective modules. On the other
hand, the situation for injectives is a bit better than for projectives.

Proposition 2.27 The following are equivalent for a module, Q:

(1) The module, Q, is injective.
(2) Given a diagram
0 / M0 /M
ξ

Q
there exists an extension, θ : M → Q, of ξ, making the diagram commute (extension property).
(3) Every exact sequence 0 −→ Q −→ M −→ M 00 −→ 0 splits.

Proof . (DX)
Q
Proposition 2.28 Given a family, {Qα }α∈Λ , of modules, the product α Qα is injective iff each Qα is
injective.
132 CHAPTER 2. RINGS AND MODULES

Proof . (DX)

Theorem 2.29 (Baer Representation Theorem) An R-module, Q, is injective iff it has the extension prop-
erty w.r.t. the sequence
0 /A / R, (∗)
where A is an ideal of R.

Proof . If Q is injective, it is clear that Q has the extension property w.r.t. (∗).
Conversely, assume that the extension property holds for (∗). What does this mean? We have the
diagram
0 /A /R


 ψ
ϕ
 
Q
in which ψ extends ϕ; so, for all ξ ∈ A, we have ϕ(ξ) = (ψ  A)(ξ). In particular, ψ(1) ∈ Q exists, say
q = ψ(1). Since ξ · 1 = ξ for all ξ ∈ A, we have

ϕ(ξ) = ψ(ξ) = ξψ(1) = ξq.

Given the diagram

0 / M0 /M
ϕ

Q
define S by  
(1) N is a submodule of M , (2) M 0 ⊆ N,
S= (N, ψ)
(3) ψ : N → Q extends ϕ to N .
Partially order S by inclusion and agreement of extensions. Then, S is inductive (DX). By Zorn’s lemma,
there is a maximal element, (N0 , ψ0 ), in S. We claim that N0 = M . If N0 6= M , there is some m ∈ M − N0 .
Let A be the transporter of m into N0 , i.e.,

(m −→ N0 ) = {ρ ∈ R | ρm ∈ N0 }.

Define the R-module map, θ : A → Q, by θ(ρ) = ψ0 (ρm). Look at the module N0 + Rm, which strictly
contains N0 . If z ∈ N0 + Rm, then z = z0 + ρm, for some z0 ∈ N0 and some ρ ∈ R. Set

ψ(z) = ψ0 (z0 ) + ρq,

where q = Θ(1) and Θ is an extension of θ (guaranteed to exist, by the hypothesis). We must prove that ψ
is a well-defined map, i.e., if z = z0 + ρm = ze0 + ρem, then

ψ0 (z0 ) + ρq = ψ0 (ze0 ) + ρeq.

Now, if ψ : N0 + Rm → Q is indeed well-defined, then it is an extension of ψ0 to the new module N0 + Rm >

N0 , contradicting the maximality of N0 . Therefore, N0 = M , and we are done.
If z = z0 + ρm = ze0 + ρem, then z0 − ze0 = (e
ρ − ρ)m; so ρe − ρ ∈ A. Consequently,

ρ − ρ) = ψ0 ((e
θ(e ρ − ρ)m).

Yet,
ρ − ρ) = Θ(e
θ(e ρ − ρ) = (e
ρ − ρ)Θ(1) = (e
ρ − ρ)q,
2.4. PROJECTIVE AND INJECTIVE MODULES 133

and so, we get

ψ0 (z0 − ze0 ) = ψ0 ((e
ρ − ρ)m) = θ(e
ρ − ρ) = (e
ρ − ρ)q.
Therefore, we deduce that
ψ0 (z0 ) + ρq = ψ0 (ze0 ) + ρeq,
establishing that ψ is well-defined.
λ
Recall that an R-module, M , is divisible iff for every λ ∈ R with λ 6= 0, the map M −→ M (multiplication
by λ), is surjective.

Corollary 2.30 If R ∈ CR has no zero-divisors, then an injective R-module is automatically divisible.

Moreover, if R is a P.I.D., a necessary and sufficient condition that Q be injective is that Q be divisible.
Therefore, over P.I.D.’s, every factor module of an injective is injective.
λ
Proof . Let λ ∈ R, with λ 6= 0. Since R has no zero divisors, the map R −→ R is a monomorphism. Thus,
the image of this map is an ideal, A, and the exact sequence

0 −→ A −→ R

is just the exact sequence

λ
0 −→ R −→ R.
Apply the cofunctor HomR (−, Q). If Q is injective, this cofunctor is exact, and we get the exact sequence
λ
HomR (R, Q) −→ HomR (R, Q) −→ 0.
λ
So, the sequence Q −→ Q −→ 0 is exact, which proves that Q is divisible.
If R is a P.I.D., then every ideal is principal, so, every exact sequence 0 −→ A −→ R, where A is an
λ λ
ideal, is of the form 0 −→ R −→ R, for some λ ∈ R. If Q is divisible, the sequence Q −→ Q −→ 0 is exact,
and we get that
λ
HomR (R, Q) −→ HomR (R, Q) −→ 0 is exact;
this means that HomR (−, Q) is exact on sequences

0 −→ A −→ R −→ R/A −→ 0,

where A is an ideal, i.e., the extension property holds for ideals, A, of R. By applying Baer’s theorem we
conclude that Q is injective.
The reader will easily verify that factor modules of divisible modules are divisible (DX). Consequently,
the last statement of the corollary holds.

Theorem 2.31 (Baer Embedding Theorem) Every R-module is a submodule of an injective module.

Proof . The proof assigned for homework (Problem 57) is based on Eckmann’s proof. Here is Godement’s proof [18]
(probably the shortest proof). The first step is to show that any Z-module, M , can be embedded into M DD , where
M D = HomZ (M, Q/Z). Given a Z-module, M , we define a natural Z-linear map, m 7→ m, b from M to M DD , in the
usual way: For every m ∈ M and every f ∈ HomZ (M, Q/Z),

m(f
b ) = f (m).

It is clear that such a map is Z-linear.

Proposition 2.32 For every Z-module, M , the natural map M −→ M DD is injective.

134 CHAPTER 2. RINGS AND MODULES

b 6= 0, i.e., there is some f ∈ M D = HomZ (M, Q/Z) so that

Proof . It is enough to show that m 6= 0 implies that m
f (m) 6= 0.
Consider the cyclic subgroup Zm of M generated by m. If m has finite order n ≥ 1, then Zm ∼ = Z/nZ. The
Z-linear map f : Z/nZ → Q/Z given by f (1) = 1/n (mod Z) is obviously an injection. Since 0 −→ Z/nZ −→ M is
exact and Q/Z is injective, the map f : Z/nZ → Q/Z extends to a Z-linear map fb: M → Q/Z with fb(m) 6= 0, as
claimed.
If Zm is infinite (m has infinite order), then we have the Z-linear surjection g : Zm → Z/2Z given by g(m) =
1 (mod 2). We also have the injective Z-linear map f2 : Z/2Z → Q/Z given by f2 (1) = 1/2 (mod Z), and since Q/Z
is injective, the Z-linear map f2 : Z/2Z → Q/Z extends to a Z-linear map fb2 : M → Q/Z, with fb2 (1) 6= 0. Then the
composition fb = fb2 ◦ g is a Z-linear map fb: M → Q/Z such that fb(m) = fb2 (g(m)) = fb2 (1) 6= 0.
Recall that given a ring R, the ring Rop is the ring with the same underlying set R, the same adddition operation
+, and the multiplication operation ∗op given by λ ∗op µ = µ ∗ λ for all λ, µ ∈ R. If M is an R-module and N is
any Z-module, then we can define a map from R × HomZ (M, N ) to HomZ (M, N ) as follows: for all α ∈ R and all
f ∈ HomZ (M, N ),
(αf )(m) = f (αm), for all m ∈ M . (∗R )
Since α ∗op β = β ∗ α, we have

(α(βf ))(m) = (βf )(αm) = f (β(αm)) = f ((β ∗ α)m) = ((β ∗ α)f )(m) = ((α ∗op β)f )(m).

The equation
(α(βf ))(m) = f (β(αm)) = ((α ∗op β)f )(m)
shows that (∗R ) defines a left action of Rop on HomZ (M, N ) which makes HomZ (M, N ) into a Rop -module.
Similarly, if M is an Rop -module and N is any Z-module, then (∗R ) defines a left action of R on HomZ (M, N )
which makes HomZ (M, N ) into an R-module, since

(α(βf ))(m) = (βf )(αm) = f (β(αm)) = f ((β ∗op α)m) = f ((α ∗ β)m) = ((α ∗ β)f )(m).

Then M D = HomZ (M, Q/Z) is an Rop -module if M is an R-module (resp. an R-module if M is an Rop -module).
Furthermore, the Z-injection, M −→ M DD , is an R-injection. The crux of Godement’s proof is the following
proposition.

Proposition 2.33 If M is a projective Rop -module, then M D is an injective R-module.

Proof . Consider the diagram

0 /X / X0
ϕ
ϕ0
 }
MD
where the row is exact. To prove that M D is injective, we need to prove that ϕ extends to a map ϕ0 : X 0 → M D .
The map ϕ yields the map M DD −→ X D , and since we have an injection M −→ M DD , we get a map θ : M → X D .
Now, since Q/Z is injective, HomZ (−, Q/Z) maps the exact sequence

0 −→ X −→ X 0

to the exact sequence

HomZ (X 0 , Q/Z) −→ HomZ (X, Q/Z) −→ 0,
0
D
i.e., X −→ X D −→ 0. So, we have the diagram

M
θ0
θ
| 
X
0
D / XD / 0,
2.4. PROJECTIVE AND INJECTIVE MODULES 135

0
where the row is exact, and since M is projective, the map θ lifts to a map θ0 : M → X D . Consequently, we get
0 0
a map X DD −→ M D , and since we have an injection X 0 −→ X DD , we get a map X 0 −→ M D extending ϕ, as
D
desired. Therefore, M is injective.
We can now prove Theorem 2.31. Consider the Rop -module M D . We know that there is a free Rop -module, F ,
so that
F −→ M D −→ 0 is exact.
But, F being free is projective. We get the exact sequence

0 −→ M DD −→ F D .

By Proposition 2.33, the module F D is injective. Composing the natural injection M −→ M DD with the injection
M DD −→ F D , we obtain our injection, M −→ F D , of M into an injective.

Corollary 2.34 Every R-module, M , has an injective resolution

0 −→ M −→ Q0 −→ Q1 −→ Q2 −→ · · · ,

where the Qi ’s are injective and the sequence is exact.

How about minimal injections? Recall that N −→ M is a minimal (essential) injection iff N is large in
M , which means that for any L ⊆ M , if N ∩ L = (0), then L = (0).
We have the following characterization of essential injections, analogous to the characterization of minimal
surjections:

Proposition 2.35 The following are equivalent for injections θ : N → M :

θ
(1) N −→ M is essential.
ϕ ϕ
(2) Given any module, Z, and any map, M −→ Z, if N −→ M −→ Z is injective, then ϕ is injective.
(3) Ker (N −→ M −→ Z) = (0) implies Ker (M −→ Z) = (0), for any module, Z.

Proof . (DX)
In contradistinction to the case of covering surjections, essential injections always exist.

Proposition 2.36 Given an injection, N −→ M , there exists a submodule, K, of M so that

(1) The sequence 0 −→ N −→ M/K is exact, and
(2) It is an essential injection.

Proof . Let
S = {K ⊆ M | K ∩ N = (0)}.
S
Since (0) ∈ S, the set S is nonempty. Partially order S by inclusion. If {Zα }α is a chain in S, let Z = α Zα ,
a submodule of M . We have
!
[ [
Z ∩N = Zα ∩ N = (Zα ∩ N ) = (0),
α α

since Zα ∩ N ) = (0), for all α. Therefore, S is inductive, and by Zorn’s lemma, it has a maximal element, say
K. Since K ∩ N = (0), property (1) is satisfied. For (2), take L ⊆ M/K so that L ∩ Im (N ) = (0). We must
show that L = (0). By the second homomorphism theorem, L corresponds to L e in M , with K ⊆ Le ⊆ M,
e
and we are reduced to proving that L = K.
136 CHAPTER 2. RINGS AND MODULES

e if η ∈
Claim: For every η ∈ L, / K, then η ∈
/ N.
e and η ∈
If η ∈ L / K and η ∈ N , then η ∈ L ∩ Im (N ), and so, η = 0, since L ∩ Im (N ) = (0). (As usual,
η 7→ η, denotes the canonical map M −→ M/K.) Yet η ∈ / K, a contradiction; the claim holds.
e and ξ ∈
Assume that ξ ∈ L / K. Consider K + Rξ, a submodule of L e strictly containing K. Since K is a
maximal module with K ∩ N = (0), there is some η ∈ (K + Rξ) ∩ N , with η 6= 0. Consequently, we have
η∈Le and η ∈ N . Now, if η ∈ K, then η ∈ N ∩ K = (0), contradicting the fact that η 6= 0; so, we must have
η∈
/ K. However, this contradicts the claim. Therefore, ξ cannot exist, and Le = K.

Terminology: The module Q is an injective hull of M iff

(1) M −→ Q is an essential injection, and

(2) The module Q is injective.

Theorem 2.37 (Baer–Eckmann–Schopf ) Every R-module has an injective hull.

Proof . By Baer’s embedding theorem (Theorem 2.31), there is an injective module, Q, so that
0 −→ M −→ Q is exact. Set

S = {L | M ⊆ L ⊆ Q and 0 −→ M −→ L is essential}.

Since M ∈ S, the set S is nonempty. The set S is partially ordered by inclusion, and it is inductive (DX).
By Zorn’s lemma, S has a maximal element, say L. I claim that L is injective. Look at the exact sequence
0 −→ L −→ Q. By the argument in the previous proposition on essential extensions, there is a maximal
K ⊆ Q, so that K ∩ L = (0) and 0 −→ L −→ Q/K is essential. Look at the diagram

0 /L / Q/K
ϕ

Q .

Since Q is injective, there is a map, ψ : Q/K → Q, extending ϕ; let T = Im ψ. The map ψ is injective,
because ψ  L is injective and the row is essential. Thus, ψ : Q/K → T is an isomorphism; moreover, L ⊆ T .
We contend that T = L. To see this, we will prove that 0 −→ M −→ T is essential. Now, being essential is
a transitive property (DX); since T is essential over L (because Q/K ∼= T and Q/K is essential over L) and
L is essential over M , we see that T is essential over M . But, L is maximal essential over M (in Q) and
L ⊆ T ; so, we conclude that T = L. Therefore, L ∼ = Q/K and we have the maps

Q −→ Q/K ∼
= L and L −→ Q.

It follows that the sequence

0 −→ K −→ Q −→ L −→ 0
splits. Consequently, L is also injective; so, L is the required injective hull.

Proposition 2.38 (Uniqueness of projective covers and injective hulls.) Say P −→ M is a projective cover
and Pe −→ M is another surjection with Pe projective. Then, there exist Pe0 , Pe00 ⊆ Pe, both projective so that

(a) Pe = Pe0 q Pe00 .

(b) P ∼
= Pe0 .
2.5. THE FIVE LEMMA AND THE SNAKE LEMMA 137

(c) In the diagram

Pe
p
e

P
p
/M /0


0

there are maps π : Pe → P and i : P → Pe in which π is surjective and i is injective, Pe00 = Ker π,
Pe0 = Im i and pe  Pe0 : Pe0 → M is a projective cover.

If M and M f are isomorphic modules, then every isomorphism, θ : M → M f, extends to an isomorphism of

projective covers, P −→ Pe. The same statements hold for injective hulls and injections, M −→ Q,
e where Q
e
is injective, mutatis mutandis.

Proof . As Pe is projective, there is a map π : Pe → P , making the diagram commute. We claim that the
map π is surjective. To see this, observe that p(Im π) = Im pe = M . Hence, Im π = P , as P is a covering
surjection. As P is projective and π is a surjection, π splits, i.e., there is a map i : P → Pe and π ◦ i = idP ;
it easily follows that i is injective. Define Pe00 = Ker π and Pe0 = Im i. We know that i : P → Pe0 is an
isomorphism, and
0 −→ Ker π (= Pe00 ) −→ Pe −→ P (∼ = Pe0 ) −→ 0 is split exact;
so, we deduce that Pe = Pe0 q Pe00 . The rest is clear.
For injectives, turn the arrows around, replace coproducts by products, etc. (DX).

2.5 The Five Lemma and the Snake Lemma

Proposition 2.39 (The five lemma.) Given a commutative diagram with exact rows

M1 / M2 / M3 / M4 / M5
ϕ1 ϕ2 ϕ3 ϕ4 ϕ5
    
N1 / N2 / N3 / N4 / N5 ,

then

(a) If ϕ2 and ϕ4 are injective and ϕ1 is surjective, then ϕ3 is injective.

(b) If ϕ2 and ϕ4 are surjective and ϕ5 is injective, then ϕ3 is surjective.

(c) If ϕ1 , ϕ2 , ϕ4 , ϕ5 are isomorphisms, then so is ϕ3 .

Proof . Obviously, (a) and (b) imply (c). Both (a) and (b) are proved by chasing the diagram (DX).

Proposition 2.40 (The snake lemma.) Given a commutative diagram with exact rows

M1 / M2 / M3 /0

δ1 δ2 δ3
  
0 / N1 / N2 / N3 ,
138 CHAPTER 2. RINGS AND MODULES

then there exists a six term exact sequence

δ
Ker δ1 −→ Ker δ2 −→ Ker δ3 −→ Coker δ1 −→ Coker δ2 −→ Coker δ3 ,

(where δ is called the connecting homomorphism) and if M1 −→ M2 is injective, so is Ker δ1 −→ Ker δ2 ,

while if N2 −→ N3 is surjective, so is Coker δ2 −→ Coker δ3 .

Proof . Simple diagram chasing shows Ker δ1 −→ Ker δ2 −→ Ker δ3 is exact and
Coker δ1 −→ Coker δ2 −→ Coker δ3 is also exact (DX). Moreover, it also shows the very last assertions of
the proposition.

We have to construct the connecting homomorphism, δ. Consider the commutative diagram:

Ker δ1 / Ker δ2 / Ker δ3

  
M1 / M2 p
/ M3 /0

δ1 δ2 δ3
  
0 / N1 i / N2 / N3

  
Coker δ1 / Coker δ2 / Coker δ3 .

Pick ξ ∈ Ker δ3 , and consider ξ as an element of M3 . There is some η ∈ M2 so that p(η) = ξ. So, we have
δ2 (η) ∈ N2 , and Im δ2 (η) in N3 is δ3 (ξ) = 0. As the lower row is exact and i is injective, η gives a unique
x ∈ N1 , with i(x) = δ2 (η). We define our δ(ξ) as the projection of x on Coker δ1 . However, we need to check
that this map is well-defined.

If we chose a different element, say ηe, from η, where p(η) = p(e η ) = ξ, then the construction is canonical
from there on. Take δ2 (η) and δ2 (e η ). Since η − ηe goes to zero under p, there is some y ∈ M1 , so that
η − ηe = Im (y) in M2 . Consequently η = ηe +Im (y); so, δ2 (η) = δ2 (e
η )+δ2 (Im (y)). But, δ2 (Im (y)) = i(δ1 (y)),
and so,

δ2 (η) = δ2 (e
η ) + i(δ1 (y)). (∗)

As before, we have some unique elements x and x e in N1 , so that i(x) = δ2 (η) and i(e
x) = δ2 (e
η ); so, by (∗),
we get i(x) = i(e
x) + i(δ1 (y)). As i is injective, we conclude that

e + δ1 (y);
x=x

e have equal projections in Coker δ1 , and our definition of δ(ξ) is independent of the lift, η, of ξ
so, x and x
to M2 . The rest is tedious diagram chasing (DX).

Remark: As we said in Section 2.3, Proposition 2.17 also holds under slightly more general assumptions and its
proof is a very nice illustration of the snake lemma. Here it is:

Proposition 2.41 Let

ϕ ψ
0 −→ M 0 −→ M −→ M 00 −→ 0
be an exact sequence of Λ-modules. If M is f.g. and M 00 is f.p., then, M 0 is f.g.
2.5. THE FIVE LEMMA AND THE SNAKE LEMMA 139

Proof . Let
F1 −→ F0 −→ M 00 −→ 0
be a finite presentation of M 00 (so, F0 , F1 are free and f.g.) Consider the diagram

F1 / F0 / M 00 /0

0 / M0 /M / M 00 / 0.

Now, F0 is free, so there exists a map F0 −→ M lifting the surjection F0 −→ M 00 . Call this map θ. From the
commutative diagram which results when θ is added, we deduce a map γ : F1 → M 0 . Hence, we find the bigger
commutative diagram
0


F1 / F0 / M 00 /0
γ θ
 
0 /M 0 /M / M 00 /0
ϕ ψ

  
Coker γ / Coker θ /0
But, by the snake lemma, Coker γ ∼= Coker θ. However, Coker θ is f.g. as M is f.g. The image of γ is f.g. as F1 is
f.g. And now, M 0 is caught between the f.g. modules Im γ and Coker γ; so, M 0 is f.g.
140 CHAPTER 2. RINGS AND MODULES

2.6 Tensor Products and Flat Modules

Let R be a ring (not necessarily commutative). In this section, to simplify the
Q notation, the op
product of
R-modules, M and N , viewed as sets, will be denoted M × N , instead of M N . For any R -module,
Sets
M , any R-module, N , and any abelian group, Z, we set
 
 (1) (∀m, m0 ∈ M )(∀n ∈ N )(ϕ(m + m0 , n) = ϕ(m, n) + ϕ(m0 , n)) 
BiR (M, N ; Z) = ϕ : M × N −→ Z (2) (∀m ∈ M )(∀n, n0 ∈ N )(ϕ(m, n + n0 ) = ϕ(m, n) + ϕ(m, n0 )) .
 
(3) (∀m ∈ M )(∀n ∈ N )(∀r ∈ R)(ϕ(mr, n) = ϕ(m, rn))
Observe that
(1) The set BiR (M, N ; Z) is an abelian group under addition; i.e., if ϕ, ψ ∈ BiR (M, N ; Z), then
ϕ + ψ ∈ BiR (M, N ; Z).
(2) The map Z BiR (M, N ; Z) is a functor from Ab to Sets. Is this functor representable? To be more
explicit, does there exist an abelian group, T (M, N ), and an element, Φ ∈ BiR (M, N ; T (M, N )), so
that the pair (T (M, N ), Φ) represents BiR (M, N ; −), i.e., the map
HomZ (T (M, N ), Z) −
g→ BiR (M, N ; Z)
via ϕ 7→ ϕ ◦ Φ, is a functorial isomorphism?
Theorem 2.42 The functor Z BiR (M, N ; Z) from Ab to Sets is representable.
Proof . Write F for the free abelian group on the set M × N . Recall that F consists of formal sums
X
ξα (mα , nα ),
α

where ξα ∈ Z, with ξα = 0 for all but finitely many α’s, and with mα ∈ M and nα ∈ N . Consider the
subgroup, N , of F generated by the elements
(m1 + m2 , n) − (m1 , n) − (m2 , n)
(m1 , n1 + n2 ) − (m, n1 ) − (m, n2 )
(mr, n) − (m, rn).
Form F/N and write m ⊗R n for the image of (m, n) in F/N . We have
(α) (m1 + m2 ) ⊗R n = m1 ⊗R n + m2 ⊗R n.
(β) m ⊗R (n1 + n2 ) = m ⊗R n1 + m ⊗R n2 .
(γ) (mr) ⊗R n = m ⊗R (rn).

Let T (M, N ) = F/N and let Φ be given by Φ(m, n) = m ⊗R n. Then, (α), (β), (γ) imply that Φ belongs
to BiR (M, N ; T (M, N )), and the assignment, ϕ 7→ ϕ ◦ Φ, gives the functorial map
HomZ (T (M, N ), Z) −→ BiR (M, N ; Z).
We need to prove that this map is an isomorphism. Pick θ ∈ BiR (M, N ; Z); we claim that θ yields a
homomorphism, T (M, N ) −→ Z. Such a homomorphism is merely a homomorphism, F −→ Z, that
vanishes on N . But, F is free; so we just need to know the images of the basis elements, (m, n), in Z. For
this, map (m, n) to θ(m, n). The induced homomorphism vanishes on the generators of N , as θ is bilinear;
thus, θ yields a map
Ξ(θ) : F/N −→ Z,
and we get our inverse map BiR (M, N ; Z) −→ HomZ (T (M, N ), Z). Routine checking shows that the maps
ϕ 7→ ϕ ◦ Φ and θ 7→ Ξ(θ) are functorial and mutual inverses.
 2.6. TENSOR PRODUCTS AND FLAT MODULES 141

Definition 2.7 The group, T (M, N ) = F/N , constructed in Theorem 2.42, is called the tensor product of
M and N over R and is denoted M ⊗R N .

Remark: Note that Theorem 2.42 says two things:

(1) For every Z-linear map, f : M ⊗R N → Z, the map, ϕ, given by ϕ(m, n) = f (m ⊗ n), for all m ∈ M and n ∈ N ,
is bilinear (i.e., ϕ ∈ BiR (M, N ; Z)), and
(2) For every bilinear map, ϕ ∈ BiR (M, N ; Z), there is a unique Z-linear map, f : M ⊗R N → Z, with ϕ(m, n) =
f (m ⊗ n), for all m ∈ M and n ∈ N . In most situations, this is the property to use in order to define a map
from a tensor product to another module.

 One should avoid “looking inside” a tensor product, especially when defining maps. Indeed, given some
0 0
element w ∈ M ⊗R N , there may be different pairs, (m, Pn) ∈ M × N and (m , n ) ∈ M × N , with
0 0
w = m ⊗R n = m ⊗R n . Worse, one can have m ⊗R n = α mα ⊗R nα . Thus, defining a function as
f (m ⊗R n) for all m ∈ M and n ∈ N usually does not make sense; there is no guarantee that f (m ⊗R n) and
f (m0 ⊗R n0 ) should agree when m ⊗R n = m0 ⊗R n0 . The “right way” to define a function on M ⊗R N is to
first define a function, ϕ, on M × N , and then to check that ϕ is bilinear (i.e., ϕ ∈ BiR (M, N ; Z)). Then,
there is a unique homomorphism, f : M ⊗R N → Z, so that f (m ⊗R n) = ϕ(m, n). Having shown that f
exists, we now may safely use its description in terms of elements, m ⊗ n, since they generate M ⊗R N . We
will have many occasions to use this procedure in what follows.

Basic properties of the tensor product:

Proposition 2.43 The tensor product, M ⊗R N , is a functor of each variable (from Rop -modules to Ab or
from R-modules to Ab). Moreover, as a functor, it is right-exact.

f is an Rop -morphism. Consider

Proof . Just argue for M , the argument for N being similar. Say f : M → M
e
M × N and the map: f (m, n) = f (m) ⊗ n. This is clearly a bilinear map M × N −→ M f ⊗R N . By the
defining property of M ⊗R N , we obtain our map (in Ab) M ⊗R N −→ M f ⊗R N . Consequently, now that
we know the map is defined, we see that it is given by

m ⊗ n 7→ f (m) ⊗ n.

For right-exactness, again vary M (the proof for N being similar). Consider the exact sequence
i
M 0 −→ M −→ M 00 −→ 0. (†)

We must prove that

M 0 ⊗R N −→ M ⊗R N −→ M 00 ⊗R N −→ 0 is exact. (††)

Pick a test abelian group, Z, and write C for Coker (M 0 ⊗R N −→ M ⊗R N ). We have the exact sequence

M 0 ⊗R N −→ M ⊗R N −→ C −→ 0. (∗)

Now, HomAb (−, Z) is left-exact, so we get the exact sequence

i∗
0 −→ HomAb (C, Z) −→ HomAb (M ⊗R N, Z) −→ HomAb (M 0 ⊗R N, Z). (∗∗)

The two terms on the righthand side are isomorphic to BiR (M, N ; Z) and BiR (M 0 , N ; Z), and the map, i∗ ,
is
ϕ ∈ BiR (M, N ; Z) 7→ i∗ ϕ ∈ BiR (M 0 , N ; Z), where i∗ ϕ(m0 , n) = ϕ(i(m0 ), n).
142 CHAPTER 2. RINGS AND MODULES

When is i∗ ϕ = 0? Observe that i∗ ϕ = 0 iff ϕ(i(m0 ), n) = 0 for all m0 ∈ M 0 and all n ∈ N . So,
HomAb (C, Z) is the subgroup of BiR (M, N ; Z) given by

{ϕ ∈ BiR (M, N ; Z) | (∀m0 ∈ M 0 )(∀n ∈ N )(ϕ(i(m0 ), n) = 0)},

and denoted Bi∗R (M, N ; Z).

Claim: There is a canonical (functorial in Z) isomorphism

Bi∗R (M, N ; Z) ∼
= BiR (M 00 , N ; Z).

Say ϕ ∈ Bi∗R (M, N ; Z). Pick m ∈ M 00 and n ∈ N , choose any m ∈ M lifting m and set

ψ(m, n) = ϕ(m, n).

e − m = i(m0 ) for some m0 ∈ M 0 . So, ϕ(m

e is another lift, then, as (†) is exact, m
If m e − m, n) = 0, as
ϕ ∈ Bi∗R (M, N ; Z). But, ϕ(m e − m, n) = ϕ(m,
e n) − ϕ(m, n), and so, ϕ(m,
e n) = ϕ(m, n), which proves
that ψ is well-defined. Consequently, we have the map ϕ 7→ ψ from Bi∗R (M, N ; Z) to BiR (M 00 , N ; Z). If
ψ ∈ BiR (M 00 , N ; Z), pick any m ∈ M and n ∈ N and set ϕ(m, n) = ψ(m, n) (where m is the image of m in
M 00 ). These are inverse maps. Therefore, we obtain the isomorphism

Bi∗R (M, N ; Z) ∼
= BiR (M 00 , N ; Z),

functorial in Z, as claimed. However, the righthand side is isomorphic to HomAb (M 00 ⊗R N, Z), and so, by
Yoneda’s lemma, we see that C ∼ = M 00 ⊗R N , and (††) is exact.

Proposition 2.44 Consider R as Rop -module.

`t Then, R ⊗R M −
g→ M . Similarly, if R is considered as
R-module, then M ⊗R R −
g→ M . Say M = i=1 Mi , then
t
a
M ⊗R N ∼
= (Mi ⊗R N ).
i=1

(Similarly for N .)

Proof . We treat the first case R⊗R M − g→M , the second one being analogous. Pick a test group, Z, and look
at HomAb (R ⊗R M, Z) ∼ = BiR (R, M ; Z). Any ϕ ∈ BiR (R, M ; Z) satisfies ϕ(r, m) = ϕ(1, rm), by bilinearity.
Now, set ϕ0 (m) = ϕ(1, m). Then, as ϕ is bilinear, we deduce that ϕ0 : M → Z is a group homomorphism.
The map ϕ 7→ ϕ0 is clearly an isomorphism from BiR (R, M ; Z) to HomR (M, Z), functorial in Z, and so, we
obtain an isomorphism
HomAb (R ⊗R M, Z) − g→ HomAb (M, Z)
functorial in Z. By Yoneda’s lemma, we get the isomorphism R ⊗R M −
g→ M.
For coproducts, we use an induction on t. The base case, t = 1, is trivial. For the induction step, look at
the exact sequence
a t
0 −→ M1 −→ M −→ Mj −→ 0.
j=2

This sequence is not only exact, but split exact. Now, from this, tensoring with N on the right and using
the induction hypothesis, we get another split exact sequence (DX)
t
a
0 −→ M1 ⊗R N −→ M ⊗R N −→ (Mj ⊗R N ) −→ 0;
j=2
2.6. TENSOR PRODUCTS AND FLAT MODULES 143

so,

t
a
M ⊗R N ∼
= (Mi ⊗R N ).
i=1

In the next section we will prove that tensor product commutes with arbitrary coproducts.

Computation of some tensor products:

`
(1) Say F = S R, as Rop -module (with S finite). Then,

a a a
F ⊗R N = ( R) ⊗R N ∼
= (R ⊗R N ) ∼
= N.
S S S

` `
Similarly, M ⊗R F ∼
= S M , if F = S R, as R-module (with S finite).

`
(1a) Assume G is also free, say G = T R (with T finite), as an R-module. Then,

a aa a
F ⊗R G ∼
= G= R= R.
S S T S×T

(2) Say A is an Rop -ideal of R. Then (R/A) ⊗R M ∼= M/AM. Similarly, if A is an R-ideal of R, then for
any Rop -module, M , we have M ⊗R (R/A) ∼ = M/M A. (These are basic results.)
Proof . We have the exact sequence

0 −→ A −→ R −→ R/A −→ 0,

where A is an Rop -ideal. By tensoring on the right with M , we get the right-exact sequence

A ⊗R M −→ R ⊗R M −→ (R/A) ⊗R M −→ 0.

Consider the diagram:

A ⊗R M / R ⊗R M / (R/A) ⊗R M /0


0 / AM /M / M/AM / 0.

The middle vertical arrow is an isomorphism; we claim that there is a map A ⊗R M −→ AM . Such a map
corresponds to a bilinear map in BiR (A, M ; AM ). But, (α, m) 7→ αm is just such a bilinear map. So, we
get our map A ⊗R M −→ AM . Now, of course, it is given by α ⊗ m 7→ αm. But then, there is induced a
144 CHAPTER 2. RINGS AND MODULES

righthand vertical arrow and we get the commutative diagram:

0 0 0

  
Ker ρ / Ker w / Ker y

  
A ⊗R M / R ⊗R M / (R/A) ⊗R M /0

ρ w y
  
0 / AM /M / M/AM /0

  
Coker ρ / Coker w / Coker y

  
0 0 0 .

The snake lemma yields an exact sequence

δ
Ker w −→ Ker y −→ Coker ρ −→ Coker w −→ Coker y −→ 0.

Since ρ is onto (DX), we have Coker ρ = 0, and since w is an isomorphism, we have Ker w = Coker w = 0.
Thus, Ker y = 0. As Coker w −→ Coker y −→ 0 is exact and Coker w = 0, we deduce that Coker y = 0.
Therefore, y is an isomorphism, as claimed. (One can also use the five lemma in the proof.)
(3) Compute Z/rZ ⊗Z Z/sZ.
We claim that the answer is Z/tZ, where t = g.c.d.(r, s).
We know (DX) that ⊗R is an additive functor. From the exact sequence
r
0 −→ Z −→ Z −→ Z/rZ −→ 0,

we get the exact sequence

r
Z ⊗Z (Z/sZ) −→ Z ⊗Z (Z/sZ) −→ (Z/rZ) ⊗Z (Z/sZ) −→ 0.

Write X for (Z/rZ) ⊗Z (Z/sZ). Hence,

r
Z/sZ −→ Z/sZ −→ X −→ 0 is exact.

Pick z ∈ Z/sZ, and say rz = 0, i.e., rz ≡ 0 (mod s). We have r = ρt and s = σt, with g.c.d.(ρ, σ) = 1.
Now, rz ≡ 0 (mod s) means that rz = sk, for some k; so, we have ρtz = σtk, for some k, and so, ρz = σk,
for some k. We see that σ | ρz, and since g.c.d.(ρ, σ) = 1, we conclude that σ | z. As a consequence, σt | tz;
so, s (= σt) | tz and we conclude that tz = 0 in Z/sZ. Conversely, if tz = 0, we get ρtz = 0, i.e., rz = 0 in
Z/sZ. Therefore, we have shown that

Ker (mult. by r) = Ker (mult. by t) in Z/sZ;

consequently (as this holds for no further divisor of t)

Im (mult. by r) = Im (mult. by t) in Z/sZ.

 2.6. TENSOR PRODUCTS AND FLAT MODULES 145

Thus,
X∼
= (Z/sZ)/(tZ/sZ) ∼
= Z/tZ.

(4) Say M is an S-module and an Rop -module. If

(sm)r = s(mr), for all s ∈ S and all r ∈ R,

then M is called an (S, Rop )-bimodule, or simply a bimodule when reference to S and R are clear. We will
always assume that if M is an S-module and an Rop -module, then it is a bimodule.
If M is a (S, Rop )-bimodule and N is an R-module, we claim that M ⊗R N has a natural structure of
S-module.

 Illegal procedure: s(m ⊗R n) = (sm) ⊗R n.

The correct way to proceed is to pick any s ∈ S and to consider the map, ϕs , from M × N to M ⊗R N
defined by
ϕs (m, n) = (sm) ⊗ n.
It is obvious that this map is bilinear (in m and n).

Remark: (The reader should realize that the bimodule structure of M is used here to check property (3) of
bilinearity. We have
ϕs (mr, n) = (s(mr)) ⊗ n = ((sm)r ⊗ n = (sm) ⊗ rn = ϕs (m, rn).)
So, we get a map M ⊗R N −→ M ⊗R N , corresponding to s. Check that this gives the (left) action of S on M ⊗R N .
Of course, it is
s(m ⊗R n) = (sm) ⊗R n.
Similarly, if M is an Rop -module and N is a (R, S op )-bimodule, then M ⊗R N is an S op -module; the (right) action
of S is
(m ⊗R n)s = m ⊗R (ns).

Remark: If M is an R-module, N is an (R, S op )-bimodule, and Z is an S op -module, then any S op -linear map
f : M ⊗R N −→ Z satisfies the property:

f (m ⊗R (ns)) = f (m ⊗R n)s, for all s ∈ S,

since f (m ⊗R (ns)) = f ((m ⊗R n)s) = f (m ⊗R n)s. Thus, the corresponding bilinear map ϕ : M × N −→ Z defined
by
ϕ(m, n) = f (m ⊗R n)
satisfies the property:
ϕ(m, ns) = ϕ(m, n)s, for all s ∈ S.
This suggests defining a set, S op -BiR (M, N ; Z), by

(1) (∀m, m0 ∈ M )(∀n ∈ N )

 
 
(ϕ(m + m0 , n) = ϕ(m, n) + ϕ(m0 , n))

 


 

(2) (∀m ∈ M )(∀n, n0 ∈ N )

 

S op -BiR (M, N ; Z) = ϕ : M × N −→ Z .

 (ϕ(m, n + n0 ) = ϕ(m, n) + ϕ(m, n0 )) 

(3) (∀m ∈ M )(∀n ∈ N )(∀r ∈ R)(ϕ(mr, n) = ϕ(m, rn))

 


 

(4) (∀m ∈ M )(∀n ∈ N )(∀s ∈ S)(ϕ(m, ns) = ϕ(m, n)s)
 

Then, we have

Theorem 2.45 Let M be an R-module and N be an (R, S op )-bimodule. The functor Z S op -BiR (M, N ; Z) from
op
Mod(S ) to Sets is representable by (M ⊗R N, Φ), where Φ is given by Φ(m, n) = m ⊗R n.
146 CHAPTER 2. RINGS AND MODULES

Note that the above statement includes the fact that M ⊗R N is an S op -module.
Similarly, if M is an (S, Rop )-bimodule, N is an R-module and Z is an S-module, then we can define the set,
S-BiR (M, N ; Z), in an analogous way (replace (4) by ϕ(sm, n) = sϕ(m, n)), and we find

Theorem 2.46 Let M be an (S, Rop )-bimodule and N be an R-module. The functor Z S-BiR (M, N ; Z) from
Mod(S) to Sets is representable by (M ⊗R N, Φ), where Φ is given by Φ(m, n) = m ⊗R n.

Associativity of tensor : Let M be an Rop -module, N an (R, S op )-bimodule, and Z an S-module. Then,

(M ⊗R N ) ⊗S Z ∼
= M ⊗R (N ⊗S Z).

For any test group, T , the left hand side represents the functor

T BiS (M ⊗R N, Z; T )

and the righthand side represents the functor

T BiR (M, N ⊗S Z; T ).

We easily check that both these are just the trilinear maps, “TriR,S (M, N, Z; T );” so, by the uniqueness of
objects representing functors, we get our isomorphism. In particular,
(A) (M ⊗R S) ⊗S Z ∼
= M ⊗R (S ⊗S Z) ∼
= M ⊗R Z.
(B) Say S −→ R is a given surjective ring map and say M is an Rop -module and N is an R-module. Then,
M is an S op -module, N is an S-module and

M ⊗S N ∼
= M ⊗R N.

To see this, look at F/N and see that the same elements are identified.
(C) Say S −→ R is a ring map. Then, M ⊗R N is a homomorphic image of M ⊗S N .

Remark: Adjointness Properties of tensor : We observed that when M is an (S, Rop )-bimodule and N is an R-
module, then M ⊗R N is an S-module (resp. when M is an Rop -module and N is an (R, S op )-bimodule, then
M ⊗R N is an S op -module.) The abelian group Hom(M, N ) also acquires various module structures depending on
the bimodule structures of M and N . There are four possible module structures:
(a) The module M is an (R, S op )-bimodule and N is an R-module. Define an S-action on
HomR (M, N ) as follows: For every f ∈ HomR (M, N ) and every s ∈ S,

(sf )(m) = f (ms), for all m ∈ M .

(b) The module M is an (R, S op )-bimodule and N is an S op -module. Define an Rop -action on
HomS op (M, N ) as follows: For every f ∈ HomS op (M, N ) and every r ∈ R,

(f r)(m) = f (rm), for all m ∈ M .

(c) The module M is an Rop -module and N is an (S, Rop )-bimodule. Define an S-action on
HomRop (M, N ) as follows: For every f ∈ HomRop (M, N ) and every s ∈ S,

(sf )(m) = s(f (m)), for all m ∈ M .

(d) The module M is an S-module and N is an (S, Rop )-bimodule. Define an Rop -action on
HomS (M, N ) as follows: For every f ∈ HomS (M, N ) and every r ∈ R,

(f r)(m) = (f (m))r, for all m ∈ M .

2.6. TENSOR PRODUCTS AND FLAT MODULES 147

The reader should check that the actions defined in (a), (b), (c), (d) actually give corresponding module structures.
Note how the contravariance in the left argument, M , of Hom(M, N ) flips a left action into a right action, and
conversely. As an example, let us check (a). For all r, t ∈ S,

((st)f )(m) = f (m(st)) = f ((ms)t) = (tf )(ms) = (s(tf ))(m).

We also need to check that sf is R-linear. This is where we use the bimodule structure of M . We have

(sf )(rm) = f ((rm)s) = f (r(ms)) = rf (ms) = r((sf )(m)).

We are now ready to state an important adjointness relationship between Hom and ⊗.

Proposition 2.47 If M is an Rop -module, N is an (R, S op )-bimodule, and Z is an S op -module, then there is a
natural functorial isomorphism

HomS op (M ⊗R N, Z) ∼
= HomRop (M, HomS op (N, Z)).

When M is an R-module, N is an (S, Rop )-bimodule, and Z is an S-module, then there is a natural functorial
isomorphism
HomS (N ⊗R M, Z) ∼ = HomR (M, HomS (N, Z)).

Proof . Using Theorem 2.45, it is enough to prove that

S op -BiR (M, N ; Z) ∼
= HomRop (M, HomS op (N, Z))

and using Theorem 2.46, to prove that

S-BiR (N, M ; Z) ∼
= HomR (M, HomS (N, Z)).

We leave this as a (DX).

Proposition 2.47 states that the functor − ⊗R N is left adjoint to the functor HomS op (N, −) when N is an
(R, S op )-bimodule (resp. N ⊗R − is left adjoint to HomS (N, −) when N is an (S, Rop )-bimodule).

Commutativity of tensor : If R is commutative, then M ⊗R N ∼ = N ⊗R M . The easy proof is just to

consider (m, n) 7→ n ⊗ m. It is bilinear; so, we get a map M ⊗R N −→ N ⊗R M . Interchange M and N ,
then check the maps are mutually inverse.
(5) Let G be a torsion abelian group and Q a divisible abelian group. Then,

Q ⊗Z G = (0).

Look at HomZ (Q ⊗Z G, T ) ∼ = BiZ (Q, G; T ), for any test group, T . Take ϕ ∈ BiZ (Q, G; T ) and look at ϕ(q, σ).
Since G is torsion, there is some n so that nσ = 0. But, Q is divisible, so q = ne q , for some qe ∈ Q. Thus,

ϕ(q, σ) = ϕ(ne
q , σ) = ϕ(e
q n, σ) = ϕ(e
q , nσ) = 0.

As this holds for all q and σ, we have ϕ ≡ 0, and so, Q ⊗Z G = (0).

` `
(6) Free modules (again). Let F = S R, an Rop -module and G = T R, an R-module (with both S
and T finite). We know that a
F ⊗R G = R.
S×T

We want to look at this tensor product more closely. Pick a basis, e1 , . . . , es , in F and a basis, f1 , . . . , ft , in
G, so that
as at
F = ej R and G = Rfl .
j=1 l=1
 148 CHAPTER 2. RINGS AND MODULES

Then, we get
s,t
a
F ⊗R G = (ej R) ⊗R (Rfl ).
j=1,l=1

Thus, we get copies of R indexed by elements ej ⊗ fl . Suppose that F is also an R-module. This means
that ρej ∈ F makes sense. We assume ρej ∈ ej R, that is the left action of R commutes with the coproduct
decomposition. Then F ⊗R G is an R-module and it is free of rank st if the left action, ρej , has obvious
properties (and similarly if G is also an Rop -module).

 It is not true in general that ρej = ej ρ. Call a free module a good free module iff it possesses a basis
e1 , . . . , es so that ρej = ej ρ, for all ρ ∈ R. (This is not standard terminology.)

 It is not generally true even here, that

ρm = mρ (m ∈ F ).
Ps
Say m = j=1 ej λj . Then, we have
s
X s
X s
X
ρm = ρ(ej λj ) = ρ(λj ej ) = (ρλj )ej ,
j=1 j=1 j=1

and
s
X s
X s
X
mρ = (ej λj )ρ = ej (λj ρ) = (λj ρ)ej .
j=1 j=1 j=1

In general, ρλj 6= λj ρ, and so, ρm 6= mρ.

Consider the special example in which R = k = a field. Then, all modules are free and good. Let V be
a k-vector space of dimension d, and let e1 , . . . , ed be some basis for V . We know that the dual space, V D ,
has the dual basis, f1 , . . . , fd , characterized by

fi (ej ) = δij .
P P
Every v ∈ V can be uniquely written as v = λi ei , and every f ∈ V D can be uniquely written as f = µi fi .
Consider the space
V ⊗k · · · ⊗k V ⊗k V D ⊗k · · · ⊗k V D .
| {z } | {z }
a b

Elements of this space, called (a, b)-tensors, have the unique form
i1X
,...,ia
cij11,...,i
,...,jb ei1 ⊗k · · · ⊗k eia ⊗k fj1 ⊗k · · · ⊗k fjb .
a

j1 ,...,jb

So, V ⊗k · · · ⊗k V ⊗k V D ⊗k · · · ⊗k V D may be identified with tuples (cij11,...,i

,...,jb ), of elements of k, doubly-
a

multiply indexed. They transform as ... (change of basis). A tensor in V ⊗k · · · ⊗k V ⊗k V D ⊗k · · · ⊗k V D

is cogredient of rank (or degree) a and contragredient of rank (or degree) b. A tensor field on a space, X,
is a function (of some class, C ∞ , C k , holomorphic, etc.) from X to a tensor vector space, as above. More
generally, it is a section of a tensor bundle over X. Also, we can apply fjm to eik and reduce the cogredient
⊗b ⊗(b−1)
and contragredient ranks by one each. This gives a map V ⊗a ⊗R V D −→ V ⊗(a−1) ⊗R V D , called
contraction.
2.6. TENSOR PRODUCTS AND FLAT MODULES 149

Remark: Let M be an R-module, N be an S-module, and Z be an (R, S op )-bimodule. Then, we know that
HomR (M, Z) is an S op -module and that Z ⊗S N is an R-module. We can define a canonical homomorphism of
Z-modules,
θ : HomR (M, Z) ⊗S N −→ HomR (M, Z ⊗S N ).
For this, for every n ∈ N and u ∈ HomR (M, Z), consider the map from M to Z ⊗S N given by

θ0 (u, n) : m 7→ u(m) ⊗ n.

The reader will check (DX) that θ0 (u, n) is R-linear and that θ0 ∈ BiS (HomR (M, Z), N ; HomR (M, Z⊗S N )). Therefore,
we get the desired homomorphism, θ, such that θ(u ⊗ n) is the R-linear map θ0 (u, n). The following proposition holds:

Proposition 2.48
(i) If N is a projective S-module (resp. a f.g. projective S-module), then the Z-homomorphism,
θ : HomR (M, Z) ⊗S N −→ HomR (M, Z ⊗S N ), is injective (resp. bijective).
(ii) If M is a f.g. projective R-module, then the Z-homomorphism, θ, is bijective.

Proof . In both cases, the proof reduces to the case where M (resp. N ) is a free module, and it proceeds by induction
on the number of basis vectors in the case where the free module is f.g. (DX).
The following special case is of special interest: R = S and Z = R. In this case, HomR (M, R) = M D , the dual
of M , and the Z-homomorphism, θ, becomes

θ : M D ⊗R N −→ HomR (M, N ),

where θ(u ⊗ n) is the R-linear map, m 7→ u(m)n.

Corollary 2.49 Assume that M and N are R-modules.

(i) If N is a projective R-module (resp. a f.g. projective R-module), then the Z-homomorphism,
θ : M D ⊗R N −→ HomR (M, N ), is injective (resp. bijective).
(ii) If M is a f.g. projective R-module, then the Z-homomorphism, θ, is bijective.

If the R-module, N , is also an S op -module, then θ is S op -linear. Similarly, if the R-module, M , is also an S op -
module, then θ is S-linear. Furthermore, if M is an Rop -module (and N is an R-module), then we obtain a canonical
Z-homomorphism,
θ : M DD ⊗R N −→ HomR (M D , N ).
Using the canonical homomorphism, M −→ M DD , we get a canonical homomorphism

θ0 : M ⊗R N −→ HomR (M D , N ).

Again, if M is a f.g. projective Rop -module, then the map θ0 is bijective (DX).

Some (very) important algebras:

Suppose that M is both an R and an Rop -module, and that R ∈ RNG. We also assume, as usual, that
M is a bimodule, i.e., (ρm)σ = ρ(mσ). Then, M ⊗R M is again a bimodule, so we can form M ⊗R M ⊗R M ,
etc. Define Tj (M ) (also denoted M ⊗j ) by T0 (M ) = R, T1 (M ) = M , and

Tj (M ) = M ⊗R · · · ⊗R M , if j ≥ 2.
| {z }
j

Then, form a a
T (M ) = Tj (M ) = M ⊗j .
j≥0 j≥0

We can make T (M ) into a ring, by concatenation. Define the map M r × M s −→ Tr+s (M ), by

h(m1 , . . . , mr ), (n1 , . . . , ns )i 7→ m1 ⊗ · · · ⊗ mr ⊗ n1 ⊗ · · · ⊗ ns .
150 CHAPTER 2. RINGS AND MODULES

This map is bilinear in the pair h(r−tuple), (s−tuple)i and so, it is multilinear in all the variables. Thus, we
get a map Tr (M ) ⊗R Ts (M ) −→ Tr+s (M ). Therefore, T (M ) is an R, Rop -algebra called the tensor algebra
of M .
If Z is an R-algebra, denote by (Z) the object Z considered just as an R-module (i.e., Z (Z) is the
partial stripping functor from R-alg to Mod(R).)

Proposition 2.50 There is a natural, functorial isomorphism

HomR-alg (T (M ), Z) ∼
= HomMod(R) (M, (Z)),

for every R-algebra, Z. That is, the functor M T (M ) is the left-adjoint of Z (Z).

Proof . Given ϕ ∈ HomR-alg (T (M ), Z), look at ϕ  T1 (M ) = ϕ  M . Observe that

ϕ  M ∈ HomMod(R) (M, (Z)), and clearly, as M generates T (M ), the map ϕ is determined by ϕ  M . We
get a functorial and injective map HomR-alg (T (M ), Z) −→ HomMod(R) (M, (Z)). Say ψ : M → (Z), pick
(m1 , . . . , md ) ∈ M d and form
e 1 , . . . , md ) = ψ(m1 ) · · · ψ(md ).
ψ(m
This map is R-multilinear in the mj ’s and has values in Z; it gives a map

Ξd (ψ) : M ⊗R · · · ⊗R M −→ Z,
| {z }
d

and so, we get a map Ξ(ψ) : T (M ) −→ Z. It is easy to check that ϕ 7→ ϕ  M and ψ 7→ Ξ(ψ) are inverse
functorial maps.
In T (M ), look at the two-sided ideal generated by elements

(m ⊗R n) − (n ⊗R m),
`
call it I. Now, T is a graded ring, i.e., it is a coproduct, j≥0 Tj (M ), of R-modules and multiplication
obeys:
Tj (M ) ⊗R Tl (M ) ⊆ Tj+l (M ).
The ideal, I, is a homogeneous ideal , which means that
a
I= I ∩ Tj (M ).
j≥0

To see this, we will in fact prove more:

`
Proposition 2.51 Suppose R = n≥0 Rn is a graded ring and I is a two-sided ideal generated by homoge-
neous elements {rα }α∈Λ (i.e., rα ∈ Rdα , for some dα ). Then, I is a homogeneous ideal. Moreover, the ring,
R/I, is again graded and R −→ R/I preserves degrees.
P
Proof . Pick ξ ∈ I, then ξ = α ρα rα and each ρα is of the form
∞
X
ρα = ρα,n , where ρα,n ∈ Rn ,
n=0

all the sums involved being, of course, finite. So, we have

∞
XX
ξ= ρα,n rα ;
α n=0
 2.6. TENSOR PRODUCTS AND FLAT MODULES 151

P
moreover, ρα,n rα ∈ Rn+dα and ρα,n rα ∈ I. As I is a 2-sided ideal, the same argument works for ξ = α rα ρα .
It follows that a
I= I ∩ Rn ,
n≥0

and I is homogeneous.
Write R for R/I, and let Rn be the image of Rn under the homomorphism ρ 7→ ρ. Then,
a  a  a
R= Rn / I ∩ Rn ∼= Rn /(I ∩ Rn ).
n n

But, Rn = Rn /(I ∩ Rn ), so we are done.

In T (M ), which is graded by the Tn (M ), we have the two 2-sided ideals: I, the 2-sided ideal generated
by the homogeneous elements (of degree 2)

m ⊗ n − n ⊗ m,

and K, the 2-sided ideal generated by the homogeneous elements

m⊗m and m ⊗ n + n ⊗ m.

Both I and K are homogeneous ideals, and by the proposition, T (M )/I and T (M )/K are graded rings.

Remark: For K, look at

(m + n) ⊗ (m + n) = m ⊗ m + n ⊗ n + m ⊗ n + n ⊗ m.

We deduce that if m ⊗ m ∈ K for all m, then m ⊗ n + n ⊗ m ∈ K for all m and n. The converse is true if 2
is invertible.
We define Sym(M ), the symmetric algebra of M to be T /I and set m · n =Vimage of m ⊗ n in Sym(M ).
The module Symj (M ) is called the j-th symmetric power of M . Similarly, (M ) = T /K is the exterior
V Vj
algebra of M , and we set m ∧ n = image of m ⊗ n in (M ). The module (M ) is called the j-th exterior
power of M .
V
Observe that m · n = n · m in Sym(M ) and m ∧ n = −n ∧ m in (M ), for all m, n ∈ M . Of course,
V ∧ m = 0, for all m ∈ M . Further, Sym(M ) is a commutative ring. However, we can have ω ∧ ω 6= 0 in
m
M ; for this, see the remark before Definition 2.8.
V
 The algebras Sym(M ) and (M ) are Z-algebras only, even if M is an R-bimodule, unless R is commu-
tative, and then they are R-algebras.

Why?
We know that r(m ⊗ n) = (rm ⊗ n) in T (M ). But in Sym(M ), we would have (writing = for equivalence
mod I)

r(m ⊗ n) = (rm) ⊗ n
= n ⊗ (rm)
= (nr) ⊗ m
= m ⊗ (nr)
= (m ⊗ n)r.
152 CHAPTER 2. RINGS AND MODULES

Then, for any r, s ∈ R, we would have

(rs)(m ⊗ n) = r(s(m ⊗ n))

= r((m ⊗ n)s)
= r(m ⊗ (ns))
= (m ⊗ (ns))r
= (m ⊗ n)(sr).

But, (sr)(m ⊗ n) = (m ⊗ n)(sr), and so, we would get

(rs)(m ⊗ n) = (sr)(m ⊗ n), for all r, s ∈ R.

V
So, if we insist that Sym(M ) and (M ) be R-algebras, then R must act as if it were commutative, i.e., the
2-sided ideal, M, generated by the elements rs − sr (= [r, s]) annihilates both our algebras. Yet R/M might
be the 0-ring. However, in the commutative case, no problem arises.

Proposition
Vs 2.52 Suppose M is an R-bimodule and as R-module it is finitely generated by e1 , . . . , er . Then,
M = (0) if s > r.

Proof . Note that for any ρ ∈ M and any ej , we have ej ρ ∈ M , and so,
X
ej ρ = λi ei , for some λi ’s,
i
V2
in other words, ej ρ is some linear combination of the ei ’s. Elements of M are sums
X 
X X (β)   X (γ) 
mβ ∧ mγ = λ i ei ∧ µj ej
β,γ β,γ i j
XX (β) (γ)
= λi (ei ∧ µj ej )
β,γ i,j
XX (β) (γ)
= λi (ei µj ∧ ej )
β,γ i,j
X
= ρlm (el ∧ em ),
l,m
Vs
for some ρlm . An obvious induction shows that M is generated by elements of the form ei1 ∧ · · · ∧ eis .
There are only r distinct ei ’s and there are s of the ei ’s in our wedge generators; thus, some ei occurs twice,
that is, we have
e i1 ∧ · · · ∧ e is = e i1 ∧ · · · ∧ e i ∧ · · · ∧ e i ∧ · · · ∧ e is .
However, we can repeatedly permute the second occurrence of ei with the term on its left (switching sign
each time), until we get two consecutive occurrences of ei :

ei1 ∧ · · · ∧ eis = ±ei1 ∧ · · · ∧ ei ∧ ei ∧ · · · ∧ eis .

Vs
As ei ∧ ei = 0, we get ei1 ∧ · · · ∧ eis = 0, and this for every generator. Therefore, M = (0).

V Let us now assume that M is a free R-module with basis e1 , . . . , en . What are T (M ), Sym(M ) and
(M )?
The elements
P of Tr (M ) are sums of terms of the form m1 ⊗ · · · ⊗ mr . Now, each mi is expressed uniquely
as mi = j λj ej . Therefore, in Tr (M ), elements are unique sums of terms of the form

(µ1 ei1 ) ⊗ (µ2 ei2 ) ⊗ · · · ⊗ (µr eir ),

2.6. TENSOR PRODUCTS AND FLAT MODULES 153

where eil might be equal to eik with il 6= ik . Let Xj be the image of ej in T (M ). Then, we see that the
elements of T (M ) are sums of “funny monomials”
µ1 Xi1 µ2 Xi2 · · · µd Xid ,
and in these monomials, we do not have Xµ = µX (in general). In conclusion, the general polynomial
`n
ring over R in n variables is equal to T j=1 R . If our free module is good (i.e., there exists a basis
e1 , . . . , en and λei = ei λ for all λ ∈ R and all ei ), then we get our simplified noncommutative polynomial
ring RhX1 , . . . , Xn i, as in Section 2.2.
`  `r
r
For Sym j=1 R , where j=1 R is good, we just get our polynomial ring R[X1 , . . . , Xr ].

All this presumed that the rank of a free finitely-generated R-module made sense. There are rings where
this is false. However, if a ring possesses a homomorphism into `ar field, then ranks do make sense (DX).
Under this assumption and assuming that the free module M = j=1 R has a good basis, we can determine
Vd
the ranks of Td (M ), Symd (M ) and (M ). Since elements of the form
ei1 ⊗ · · · ⊗ eid , where {i1 , . . . , id } is any subset of {1, . . . , r}
form a basis of Td (M ), we get rk(T (M )) = rd . Linear independence is reduced to the case where R is a field
in virtue of our assumption. Here, it is not very difficult linear algebra to prove linear independence. For
example, M ⊗k N is isomorphic to Homk (M D , N ), say by Corollary 2.49.
Elements of the form
e i1 ⊗ · · · ⊗ e id ,
where i1 ≤ i2 ≤ . . . ≤ id


form a basis of Symd (M ), so we get rk(Symd (M )) = r+d−1
d (DX–The linear algebra is the same as before,
only the counting is different). Let us check this formula in some simple cases. For r = d = 2, the formula
predicts dimension 3; indeed, we have the basis of 3 monomials: X12 , X22 , X1 X2 . For r = d = 3, the formula
predicts dimension 10; we have the basis of 10 monomials:
X13 , X23 , X33 , X12 X2 , X12 X3 , X22 X1 , X22 X3 , X32 X1 , X32 X2 , X1 X2 X3 .

Finally, elements of the form

ei1 ∧ · · · ∧ eid , where i1 < i2 < . . . < id
Vd Vd

form a basis of (M ), so we get dim( (M )) = dr . Again, linear independence follows from the field case.
Vd f, where Mf
Here, it will be instructive to make a filtration of M in terms of lower wedges of M and M
has rank r − 1. Then, induction can be used. All this will be left to the reader.
And now, an application to a bit of geometry. Let M be a (smooth) manifold of dimension r. For every
x ∈ M , we have the tangent space to M at x, denoted T (M )x , a rank r vector space. A basis of this vector
space is
∂ ∂
,..., ,
∂X1 ∂Xr
where X1 , . . . , Xr are local coordinates at x ∈ M . A tangent vector is just
r
X ∂
aj ,
j=1
∂Xj

the directional derivative w.r.t. the vector −

→
v = (a1 , . . . , ar ). The dual space, T (M )D
x , is called the cotangent
space at x or the space of 1-forms at x, and has the dual basis: dX1 , . . . , dXr , where
 
∂
(dXi ) = δij .
∂Xj
154 CHAPTER 2. RINGS AND MODULES

Pr
Every element of T (M )Dx is a 1-form at x, i.e., an expression j=1 bj dXj . We have the two vector space
S S D
families x∈M T (M )x and x∈M T (M )x . These vector space families are in fact vector bundles (DX), called
the tangent bundle, T (M ), and the cotangent bundle, T (M )D , respectively.
Say ϕ : M → N is a map of manifolds, then we get a vector space map,

Dϕx : T (M )x −→ T (N )ϕ(x) .

This map can be defined as follows: For any tangent vector, ξ ∈ T (M )x , at x, pick a curve through x (defined
near x), say z : I → M , and having our chosen ξ as tangent vector at t = 0 (with x = z(0)). Here, I is a
small open interval about 0. Then,
z ϕ
I −→ M −→ N
is a curve in N through ϕ(x), and we take the derivative of ϕ(z(t)) at t = 0 to be our tangent vector
(Dϕx )(ξ).
By duality, there is a corresponding map (Dϕx )∗ : T (N )D D
ϕ(x) −→ T (M )x called pull-back of differential
Vd
forms. Given any open subset, V , of N , for any section, ω ∈ Γ(V, T (N )D ), by pullback we get the section
∗ −1
Vd D
ϕ ω ∈ Γ(ϕ (V ), T (M ) ). The reader should explicate this map in terms of the local coordinates on V
and ϕ−1 (V ).
Vd
Now, consider some section, ω ∈ Γ(U, T (M )D ), where U is an open in M . In local coordinates, ω
looks like X
a(x)dxi1 ∧ · · · ∧ dxid ; x ∈ U.
i1 <···<id

Here, U is a piece of a chart, i.e., there is a diffeomorphism ϕ : V (⊆ Rr ) −

g→ U . If z : I (⊆ Rd ) −→ V is a
map of a D-disk to V , the composition ϕ ◦ z is called an elementary d-chain in U ⊆ M , and a d-chain is a
formal Z-combination of elementary d-chains. Then, we have (ϕ ◦ z)∗ ω, a d-form on I. Hence, by elementary
real calculus in several variables, Z
(ϕ ◦ z)∗ ω
I

makes sense. ((DX), compute (ϕ ◦ z)∗ ω in local coordinates.) We define the integral of ω over the elementary
d-chain ϕ(z(I)) by Z Z
ω = (ϕ ◦ z)∗ ω,
ϕ(z(I)) I

and for d-chains, let Z XZ

ω= ω.
d−chain elem. pieces

An elaboration of these simple ideas gives the theory of integration of forms on manifolds.
We also have the theory of determinants. Suppose R is a commutative ring and M is a free module of
rank d over R with basis e1 , . . . , ed . So,
d
a
M∼
= Rej .
j=1

Let N be another free module of the same rank with basis f1 , . . . , fd . Then, a linear map ϕ ∈ HomR (M, N )
gives a matrix in the usual way (ϕ(ej ) as linear combination of the fi ’s is the j-th column). By functoriality,
Vd Vd Vd Vd Vd
we get a linear map ϕ: M → N . Now, each of M and N is free of rank 1, and their bases
are e1 ∧ · · · ∧ ed and f1 ∧ · · · ∧ fd , respectively. Therefore,
^
d 
ϕ (e1 ∧ · · · ∧ ed ) = λ(f1 ∧ · · · ∧ fd ),
2.6. TENSOR PRODUCTS AND FLAT MODULES 155

for some
 unique λ ∈ R. This unique λ is the determinant of ϕ, by definition. Now,
Vd 
ϕ (e1 ∧ · · · ∧ ed ) = ϕ(e1 ) ∧ · · · ∧ ϕ(ed ), and so det(ϕ) is an alternating multilinear map on the columns
of the matrix of ϕ. If Q is yet a third free module of rank dVand if ψ : N → Q is an R-linear map and
d
g1 , . . . , gd a chosen basis for the module Q, then we find that ψ takes f1 ∧ · · · ∧ fd to µ(g1 ∧ · · · ∧ gd ),
Vd
where µ = det(ψ). Since ψ is R-linear, it takes λ(f1 ∧ · · · ∧ fd ) to λµ(g1 ∧ · · · ∧ gd ), and it follows that

det(ψ ◦ ϕ) = µλ = det(ψ) det(ϕ).

It might appear that det(ϕ) depends upon our choice of basis, but this is not entirely so. If one has two
choices of bases in each of M and N , say {ei } and {eei }; {fj } and {fej }, and if the matrices of the identity
transformations M −→ M and N −→ N in the basis pairs are the same, then det(ϕ) is the same whether
computed with e’s and f ’s or with ee’s and fe’s. This situation holds when we identify M and N as same
rank free modules, then we have just one pair of bases: The {ei } and the {eei }. The determinant of the
endomomorphism ϕ : M → M is then independent of the choice of basis.
If M and N have different ranks, say M has rank r with chosen basis e1 , . . . , er while N has rank s with
chosen basis f1 , . . . , fs , then for any R-linear ϕ : M → N , we have the induced map
d
^ d
^ d
^
ϕ: M −→ N.
Vd Vd
Consider ej1 ∧ · · · ∧ ejd , an element of the induced basis for M . We apply the map ϕ and find

^
d  X
ϕ (ej1 ∧ · · · ∧ ejd ) = λji11...i
...jd
f ∧ · · · ∧ fid .
d i1
1≤i1 <···<id ≤s

The element λji11...i

...jd
∈ R is exactly the d × d minor from the rows i1 , . . . , id and columns j1 , . . . , jd of the
d
Vd
matrix of ϕ in the given bases. So, the d × d minors form the entries for ϕ. Projectives being cofactors of
free
Vd modules

allow the definition of determinants of their endomorphisms as well. For this, one must study
P q Pe . (DX)
For the next two remarks, assume that R ∈ CR.

Remarks:
(1) Let Z be a commutative R-algebra. Then, the functor, Z (Z) (= Z as R-module), has as left-adjoint
in CR the functor M SymR (M ):

HomR-alg (SymR (M ), Z) −
g→ HomR (M, (Z))

is a functorial isomorphism (in M and Z).

(2) An alternating R-algebra is a Z/2Z-graded R-algebra (which means that Z = Zeven q Zodd = Z0 q Z1 ,
with Zi Zj ⊆ Zi+j (mod 2) ), together with the commutativity rule

ξη = (−1)deg ξ·deg η ηξ.

V
The left-adjoint property for V M is this: The functor Z (Z1 ) (= Z1 as R-module, where Z is an
alternating R-algebra) has M M as left adjoint, i.e.,
^
Homalt. R-alg ( M, Z) − g→ HomR (M, (Z1 ))

is a functorial isomorphism (in M and Z).

156 CHAPTER 2. RINGS AND MODULES

V Vp Vq
Remark: If ω ∈ ( M )even , then ω ∧ ω need not be zero. In fact, if ξ ∈ M and η ∈ M , then (DX)

ξ ∧ η = (−1)pq η ∧ ξ.

Example: M = R4 , ω = dx1 ∧ dx2 + dx3 ∧ dx4 (ω is the standard symplectic form on R4 ). We have

ω ∧ ω = (dx1 ∧ dx2 + dx3 ∧ dx4 ) ∧ (dx1 ∧ dx2 + dx3 ∧ dx4 ) = 2dx1 ∧ dx2 ∧ dx3 ∧ dx4 6= 0.

Flat Modules. As with the functor Hom, we single out those modules rendering ⊗ an exact functor.
Actually, before we study right limits, little of consequence can be done. So, here is an introduction and
some first properties; we’ll return to flatness in Section 2.8.

Definition 2.8 An Rop -module, M , is flat (over R) iff the functor N M ⊗R N is exact. If M is an
R-module then M is flat (over R) iff the functor (on Rop -modules) N N ⊗R M is exact. The module,
M , is faithfully flat iff M is flat and M ⊗R N = (0) (resp. N ⊗R M = (0)) implies N = (0).

Proposition 2.53 Say M is an R-module (resp. Rop -module) and there is another R-module (resp. Rop -
module), Mf, so that M q M f is flat. Then M is flat. Finitely generated free modules are faithfully flat.
Finitely generated projective modules are flat. Finite coproducts of flat modules are flat. (The finiteness
hypotheses will be removed in Section 2.8, but the proofs require the notion of right limit.)

Proof . Let 0 −→ N 0 −→ N −→ N 00 −→ 0 be an exact sequence; we treat the case where M is an Rop -module.
Let F = M q M f. As F is flat, the sequence

0 −→ F ⊗R N 0 −→ F ⊗R N is exact.

We have the diagram

M ⊗R N 0
θ / M ⊗R N

 
∼ ∼
F ⊗R N 0
= / M ⊗R N 0 q Mf ⊗R N 0 / M ⊗R N q Mf ⊗R N o =
F ⊗R N.

The bottom horizontal arrow is injective and the vertical arrows are injective too, as we see by tensoring the
split exact sequence
0 −→ M −→ F −→ M f −→ 0
on the right with N and N 0 . A trivial diagram chase shows that θ is injective, as contended.
` `
Assume F is free and f.g., that is, F = S R, where S 6= ∅ and S is finite. Since F ⊗R N ∼ = S N , we
have F ⊗R N = (0) iff N = (0). If we knew that finite coproducts of flats were flat, all we would need to
show is that R itself is flat. But, R ⊗R N ∼
= N , and so, R ⊗R − is exact.
f be flat and consider their coproduct, F = M q M
Let M and M f. Then, for any exact sequence

0 −→ N 0 −→ N −→ N 00 −→ 0

the maps f : M ⊗R N 0 → M ⊗R N and g : M f ⊗R N 0 → M f ⊗R N are injective, as M and M

f are flat. Since
the coproduct functor is exact, f q g is injective and so
f ⊗R N 0 ) ∼
(M ⊗R N 0 ) q (M = F ⊗R N 0 −→ F ⊗R N ∼ f ⊗R N )
= (M ⊗R N ) q (M

is injective as well, which proves that F is flat.

If P is projective and f.g., then P q Pe ∼
= F , for some module Pe and some f.g. free module, F . The first
part of the proof shows that P is flat.
2.7. LIMIT PROCESSES IN ALGEBRA 157

Proposition 2.54 If R ∈ CR is an integral domain (or R ∈ RNG has no zero divisors) then every flat
module is torsion-free. The converse is true if R is a P.I.D. (the proof will be given in Section 2.8).

ξ
Proof . If ξ ∈ R, then 0 −→ R −→ R is an injective Rop -homomorphism ((ξm)ρ = ξ(mρ)). The diagram

0 / R ⊗R M ξ
/ R ⊗R M

 
M
ξ
/M

commutes, the vertical arrows are isomorphisms, and the upper row is exact, since M is flat. This shows
that m 7→ ξm is injective; so, if ξm = 0, then m = 0.

Remark: The module Q is a flat Z-module. However, Q is not free, not projective (DX) and not faithfully
flat (Q ⊗Z Z/2Z = (0)).

2.7 Limit Processes in Algebra

Let Λ be a partially ordered set (with partial order ≤) and assume Λ has the Moore–Smith property (Λ is a
directed set), which means that for all α, β ∈ Λ, there is some γ ∈ Λ so that α ≤ γ and β ≤ γ.
Examples of Directed Sets: (1) Let X be a topological space, and pick x ∈ X; take
Λ = {U | (1) U open in X; (2) x ∈ U }, with U ≤ V iff V ⊆ U .
(2) Λ = N, and n ≤ m iff n | m (Artin ordering).
To introduce right and left limits, we consider the following set-up: We have a category, C, a collection
of objects of C indexed by Λ, say Cα . Consider the two conditions (R) and (L) stated below:

(R) For all α ≤ β, there is a morphism, ϕβα : Cα → Cβ , and there is compatibility: For all α ≤ β ≤ γ, the
diagram
C
> γ `AA γ
ϕγα }
}} AA ϕβ
} AA
}}} AA
}
Cα / Cβ
β
ϕα

commutes and ϕα
α = idCα .

(L) For all α ≤ β, there is a morphism, ψβα : Cβ → Cα , and there is compatibility: For all α ≤ β ≤ γ, the
diagram
Cγ
AA β
ψγα }}} AA ψγ
} AA
}}} AA
~}
Cα o Cβ
α
ψβ

commutes and ψαα = idCα .

Definition 2.9 A right (direct, inductive) mapping family, (Cα , ϕβα ), of C is a family of objects, Cα , and
morphisms, ϕβα , satisfying axiom (R). Mutatis mutandis for a left (inverse, projective) mapping family,
(Cα , ψβα ) and axiom (L).
158 CHAPTER 2. RINGS AND MODULES

Examples of Right and Left Mapping Families:

(1L) Let Λ = N with the usual ordering, C = Ab and Cn = Z. Pick a prime, p; for m ≤ n, define
ψnm : Z → Z as multiplication by pn−m .
(1R) Same Λ, same C, same Cn , and ϕnm : Z → Z is multiplication by pn−m .
(2L) Same Λ, Artin ordering, same C, same Cn . If n ≤ m, then n | m, so mZ ⊆ nZ, define
n
ψm : Z/mZ → Z/nZ as the projection map.
(2R) Same Λ, Artin ordering, same C, Cn = Z/nZ. If n ≤ m, then r = m/n ∈ Z, define ϕm
n : Z/nZ →
Z/mZ as multiplication by r.
Look at the functor (from C to Sets)
 
 T 


 fα ~~
~> `@@@ f 


 @@ β 
T (fα : Cα −→ T )α ~~~ @@ commutes whenever α ≤ β ,
 ~~ @ 

 / Cβ 


 Cα 

ϕβ α

denoted Lim (Cα , ϕβα )(T ), and the cofunctor (from C to Sets)
−→
α
 
 T 

 ~ @@@ c 

cα ~~

 @@ β 

~ @@
T (gα : T −→ Cα )α ~~~ @ commutes whenever α ≤ β ,

 ~~ 



 Cα o α
Cβ 


ψβ

denoted Lim (Cβ , ψβα )(T ).

←−
β

Question: Are either (or both) of these representable?

Definition 2.10 The right (direct, inductive) limit of a right mapping family, (Cα , ϕβα ), is the pair, (C, {cα }),
representing the functor Lim (Cα , ϕβα ) and is denoted lim (Cα , ϕβα ). The left (inverse, projective) limit of
−→ −→
α α
a left mapping family, (Cβ , ψβ ), is the pair, (C, {cβ }), representing the functor Lim (Cβ , ψβα ), denoted
α
←−
β
lim (C , ψ α ).
←− β β
β

Let us explicate this definition. First, consider right mapping families. The tuple {cα }α is to lie in
Lim (Cα , ϕβα )(C), the set of tuples of morphisms, cα : Cα → C, so that the diagram
−→
α

> T `@
~~~ @@@ cβ
cα
~ @@
~~ @@
~~
Cα / Cβ
β
ϕα

commutes whenever α ≤ β. We seek an object, C ∈ C, and a family of morphisms, cα : Cα → C, so that

HomC (C, T ) ∼
= Lim (Cα , ϕβα )(T ),
−→
α
2.7. LIMIT PROCESSES IN ALGEBRA 159

for every T ∈ C, via the isomorphism u 7→ {u ◦ cα }α . Thus, the above functorial isomorphism says that for
every family of morphisms, {fα : Cα → T }α ∈ Lim (Cα , ϕβα )(T ), there is a unique morphism, u : C → T , so
−→
α
that
fα = u ◦ cα , for all α ∈ Λ.
This is the universal mapping property of lim Cα .
−→
α

Next, consider left mapping families. This time, the tuple {cβ }β is to lie in Lim (Cβ , ψβα )(C), the set of
←−
β
tuples of morphisms, cβ : C → Cβ , so that the diagram

CA
}} AA c
cα
}} AA β
}} AA
}~ } A
Cα o Cβ
α
ψβ

commutes whenever α ≤ β. We seek an object, C ∈ C, and a family of morphisms, cβ : C → Cβ , so that

HomC (T, C) ∼
= Lim (Cβ , ψβα )(T ),
←−
β

for every T ∈ C, via the isomorphism u 7→ {cβ ◦ u}β . The universal mapping property of lim Cα is that for
←−
α
α
every family of morphisms, {gα : T → Cα }α ∈ Lim (Cβ , ψβ )(T ), there is a unique morphism, u : T → C, so
←−
β
that
gα = cα ◦ u, for all α ∈ Λ.

Remark: A right (resp. left) mapping family in C is the same as a left (resp. right) mapping family in the
dual category C D . Thus, lim (Cα ) exists in C iff lim (Cα ) exists in C D .
−→ ←−
α α

Let us examine Example (1L). If we assume that its inverse limit exists, then we can find out what this
is. By definition, whenever n ≤ m, the map ψm n
: Z → Z is multiplication by pm−n . Pick ξ ∈ C, hold n fixed
and look at cn (ξ) ∈ Z. For all m ≥ n, the commutativity of the diagram

C?
 ??? cm
cn
 ??
 ??
 
Zo n
Z
ψm

shows that pm−n cm (ξ) = cn (ξ), and so, pm−n divides cn (ξ) for all m ≥ n. This can only be true if cn ≡ 0.
Therefore, all the maps, cn , are the zero map. As there is a unique homomorphism from any abelian group,
T , to (0) and as the tuple of maps, {cα }α , is the tuple of zero maps, the group (0) with the zero maps is
lim C . In fact, this argument with T replacing C proves the existence of the left limit for the family (1L)
←− α
α
and exhibits it as (0).

Theorem 2.55 (Existence Theorem) If C is any one of the categories: Sets, Ω-groups (includes R-modules,
vector spaces, Ab, Gr), topological spaces, topological groups, CR, RNG, then both Lim and Lim are
−→ ←−
α α
representable (we say that C possesses arbitrary right and left limits).
160 CHAPTER 2. RINGS AND MODULES

Proof . We give a complete proof for Sets and indicate the necessary modifications for the other categories.
Let Λ be a directed index set.
(1) Right limits:S For every α ∈ Λ, we have a set, Sα , and we have set maps, ϕβα : Sα → Sβ , whenever
α ≤ β. Let S = · Sα , the coproduct of the Sα ’s in Sets (their disjoint union). Define an equivalence
relation on S as follows: For all x, y ∈ S,

if x ∈ Sα and y ∈ Sβ then x∼y iff (∃γ ∈ Λ)(α ≤ γ, β ≤ γ)(ϕγα (x) = ϕγβ (y)).

We need to check that ∼ is an equivalence relation. It is obvious that ∼ is reflexive and symmetric.
Say x ∼ y and y ∼ z. This means that x ∈ Sα , y ∈ Sβ , z ∈ Sγ and there exist δ1 , δ2 ∈ Λ so that α ≤ δ1 ;
β ≤ δ1 ; β ≤ δ2 ; γ ≤ δ2 , and
ϕδα1 (x) = ϕδβ1 (y); ϕδβ2 (y) = ϕδγ2 (z).
As Λ is directed, there is some δ ∈ Λ, with δ1 ≤ δ and δ2 ≤ δ; so, we may replace δ1 and δ2 by δ. Therefore,
ϕδα (x) = ϕδγ (z), and transitivity holds. Let S = S/ ∼. We have the maps
[ pr
sα : Sα −→ · Sλ = S −→ S/ ∼ = S,
λ

and the pair (S, {sα }) represents Lim Sα , as is easily checked.

−→
α

(2) Left Limits: We have sets, Sα , for every α ∈ Λ, and maps, ψβα : Sβ → Sα . Let
n Y o
P = (ξα ) ∈ Sα (∀α ≤ β)(ψβα (ξβ ) = ξα ) ,
α

be the collection of consistent tuples from the product. The set P might be empty.

We have the maps Y prα

pα : P ,→ Sα −→ Sα .
α

The pair (P, {pα }) represents the cofunctor Lim Sα (DX).

←−
α

Modifications: Look first at the category of groups (this also works for Ω-groups and rings).
(10 ) Right limits. Write Gα for each group (α ∈ Λ). We claim that G = lim Gα (in Sets) is already a
−→
α
group (etc., in a natural way) and as a group, it represents our functor. All we need to do is to define the
group operation on lim Gα . If x, y ∈ G = lim Gα , then x = cα (ξ) and y = cβ (η), for some ξ ∈ Gα and
−→ −→
α α
some η ∈ Gβ . Since Λ is directed, there is some γ ∈ Λ with α, β ≤ γ; consider ξ 0 = ϕγα (ξ) and η 0 = ϕγβ (η).
(Obviously, cγ (ξ 0 ) = x and cγ (η 0 ) = y.) So, we have ξ 0 , η 0 ∈ Gγ , and we set

xy = cγ (ξ 0 η 0 ).

Check (DX) that such a product is well-defined and that G is a group. Also, the maps cα are group
homomorphisms.
The existence of right limits now holds for all the algebraic categories.
Now, consider the category, TOP, S of topological spaces. Observe that when each Sα is a topological
space, then the disjoint union, S = · Sα , is also a topological space (using the disjoint union topology); in
fact, it is the coproduct in TOP. Give S = S/ ∼ the quotient topology, and then check that the maps sα
are continuous and that (S, {sα }) represents Lim Sα in TOP.
−→
α
2.7. LIMIT PROCESSES IN ALGEBRA 161

For the category of topological groups, TOPGR, check that G = lim Gα is also a topological space as
−→
α
above and (DX) that the group operations are continuous. Thus, (G, {sα }α ) represents Lim Gα in TOPGR.
−→
α
0
(2 ) Left Limits. Again, first assume each Gα is a group and the ψβα
are homomorphisms. Check that
P (= consistent tuples) is a group (in particular, note that (1, 1, . . . , 1, . . .) is consistent so that P 6= ∅) and
that the pα ’s are homomorphisms (DX); hence, (P, {pα }) represents Lim Gα . Now, similar reasoning shows
←−
α
left limits exist for all the algebraic categories.
Q
For TOP, we make α Sα into a topological Q space with the product topology. Check (DX) that the
continuity of the ψβα ’s implies that P is closed in α Sα . Then, the pα ’s are also continuous and (P, {pα })
represents Lim Sα in TOP.
←−
α

For TOPGR, similar remarks, as above for TOP and as in the discussion for groups, imply that (P, {pα })
represents Lim Gα in TOPGR.
←−
α

Remark: Say Λ is a directed index set. We can make Λ a category as follows: Ob(Λ) = Λ, and

∅ if α 6≤ β;
Hom(α, β) =
{·} if α ≤ β.

(Here, {·} denotes a one-point set.) Given a right mapping family, (Cα , ϕβα ), where ϕβα ∈ HomC (Cα , Cβ ), we
define the functor, RF, by

RF(α) = Cα
RF(· : α → β) = ϕβα .

Similarly, there is a one-to-one correspondence between left-mapping families, (Cβ , ψβα ), and cofunctors, LF,
defined by

LF(α) = Cα
LF(· : α → β) = ψβα .

If we now think of RF and LF as “functions” on Λ and view the Moore–Smith property as saying that the α’s
“grow without bound”, then we can interpret lim Cα and lim Cα as: “limits, as α → ∞, of our ‘functions’
−→ ←−
α α
RF and LF”,
lim C = lim RF(α) and lim Cα = lim LF(α).
−→ α α→∞ ←− α→∞
α α

Indeed, there is a closer analogy. Namely, we are taking the limit of RF(α) and LF(α) as nets in the sense
of general topology.
Say Γ ⊆ Λ is a subset of our index set, Λ. We say that Γ is final in Λ (old terminology, cofinal ) iff for
every α ∈ Λ, there is some β ∈ Γ with α ≤ β. Check (DX),

lim lim lim lim

−→ Cα = −→ Cα ; ←− Cα = ←− Cα .
α∈Γ α∈Λ α∈Γ α∈Λ

Examples of Right and Left Limits:

162 CHAPTER 2. RINGS AND MODULES

(1R) Recall that Λ = N with the ordinary ordering, Cn = Z and for m ≥ n, ϕm n is multiplication by
pm−n
. Consider the isomorphism, θn : Z → (1/pn )Z ⊆ Q, defined by θn (1) = 1/pn . The diagram

Cn = Z
θn
/ 1 Z /Q
pn
pm−n incl


Cm = Z / 1 Z /Q
θm pm

commutes, and so, the direct limit on the left is equal to the direct limit in the middle. There, the direct
limit is  
k
lim C = k ∈ Z, p 6 | k ⊆ Q.
−→ m pt
m
1
This subgroup, lim Cm , of Q is usually denoted ∞ Z.
−→ p
m

Generalization: Λ = N, Artin ordering (n ≤ m iff n | m), Cn = Z, and for n ≤ m, define, ϕm

n =
multiplication by m/n. We get

lim C = Q. (∗)
−→ n
n

(2R) What is lim Z/nZ? If we observe that Z/nZ ∼

= n Z/Z,
1
by (∗), we get
−→
n|m

lim Z/nZ = Q/Z.

−→
n|m

Say X and Y are topological spaces and pick x ∈ X. Let

Λx = {U | U open in X and x ∈ U };

Partially order Λx so that U ≤ V iff V ⊆ U (usual ordering on Λx ). Clearly, Λx has Moore–Smith. Let
 
(1) f : U → Y
C(U ) = f
(2) f is continuous on U (or perhaps has better properties)

Look at lim C(U ), denoted temporarily Cx . We have ξ ∈ Cx iff there is some open subset, U , of X, with
−→
Λx
x ∈ U , some continuous function, f : U → Y , and ξ is the class of f .
Two functions, f : U → Y and g : V → Y , where U, V ⊆ X are open and contain x, give the same ξ iff
there is some open, W ⊆ U ∩ V , with x ∈ W , so that f  W = g  W . Therefore, Cx is the set of germs of
continuous functions on X at x. (The usual notation for Cx is OX,x .)
(2L) Consider the left limit, lim Z/nZ, where ψm
n
: Z/mZ → Z/nZ is projection. The elements of
←−
n|m
lim Z/nZ are tuples, (ξn ), with ξn ∈ Z, such that
←−
n|m

(1) (ξn ) = (ηn ) iff (∀n)(ξn ≡ ηn (mod n)) and

(2) (consistency): If n | m, then ξm ≡ ξn (mod n).
2.7. LIMIT PROCESSES IN ALGEBRA 163

We obtain a new object, denoted Z.b We have an injective map, Z −→ Z,

b given by n 7→ (n, n, . . . , n, . . .).
You should check that the following two statements are equivalent:

(1) Chinese Remainder Theorem.

b
(2) Z is dense in Z.

Proposition 2.56 Say C = lim Cα and let x ∈ Cα and y ∈ Cβ , with cα (x) = cβ (y). Then, there is some
−→
α
γ
γ ≥ α, β, so that ϕα (x) = ϕβ (y). In particular, if all the ϕβα are injections, so are the canonical maps, cα .
γ

Proof . Clear.

Corollary 2.57 Say C = Ω-modules and each Cα is Ω-torsion-free. Then, lim Cα is torsion-free.
−→
α

Proof . Pick x ∈ C = lim Cα ; λ ∈ Ω, with λ 6= 0. Then, λx = λcα (xα ), for some α and some xα ∈ Cα . So,
−→
α
0 = λx = cα (λxα ) implies that there is some γ ≥ α, with ϕγα (λxα ) = 0. Consequently, λϕγα (xα ) = λxγ = 0.
But Cγ is torsion-free, so xγ = 0. Therefore, x = cα (xα ) = cγ (xγ ) = 0. This proves that C is torsion-free.

Corollary 2.58 Say C = Ω-modules and each Cα is Ω-torsion. Then, lim Cα is torsion.
−→
α

Proof . If x ∈ C, then there is some α and some xα ∈ Cα , with cα (xα ) = x. But, there is some λ ∈ Ω, with
λ 6= 0, so that λxα = 0, since Cα is torsion. So, λx = λcα (xα ) = cα (λxα ) = 0.

Proposition 2.59 Let Λ be an index set and C = Sets. Then, every set is the right-limit of its finite subsets
(under inclusion). The same conclusion holds if C = Gr, Ω-groups, RNG, then each object of C is equal to
the right limit of its finitely generated subobjects.

Proof . Let Λ = {T ⊆ S | T finite}. Order Λ, via T ≤ W iff T ⊆ W . Clearly, Λ has Moore–Smith. Let
Σ = lim T .
−→
T ∈Λ

For a given T ∈ Λ, we have an injective map, iT : T ,→ S. Hence, by the universal mapping property,
these maps factor through the canonical maps, γT : T → Σ, via a fixed map, ϕ : Σ → S:

Σ _?
ϕ
/S
?? ?
?? 
γT ??? 
 iT
T

Pick some ξ ∈ S. Then, {ξ} ∈ Λ; so we get a map, γ{ξ} : {ξ} → Σ. Let ψ(ξ) = γ{ξ} (ξ) ∈ Σ. This gives a
map, ψ : S → Σ. Check (DX), ϕ and ψ are inverse maps.
Modifications: Λ = {T ⊆ S | T is a finitely generated subobject of S} and proceed analogously.

Corollary 2.60 An abelian group is torsion iff it is a right-limit of finite abelian groups.

Corollary 2.61 Say C is a category with finite coproducts (or finite products). If C has right limits (resp.
left limits) then C has arbitrary coproducts (resp. arbitrary products).

Proof . Cf. Problem 62.

164 CHAPTER 2. RINGS AND MODULES

Proposition 2.62 Say {Gα }α is a left-mapping family of finite groups (not necessarily abelian). Then, the
left limit, lim Gα = G, is a compact topological group. (Such a G is called a profinite group.) Similarly,
←−
α
if the Gα are compact topological groups and form a left-mapping family with continuous homomorphisms,
then lim Gα = G is a compact topological group.
←−
α

Proof
Q . Observe that the second statement
Q implies the first. Now, G is the group of consistent tuples in
G
α α . By Tychonov’s theorem, G
α α is compact. As the ψβα are continuous, the subgroup of consistent
tuples is closed ; therefore, this subgroup is compact.
b is compact.
It follows from Proposition 2.62 that Z

2.8 Flat Modules (Again)

Proposition 2.63 Say {Ωα }α is a right-mapping family of rings, {Mα }α , {Nα }α are “right-mapping fam-
ilies” of Ωop
α (resp. Ωα )-modules, then {Mα ⊗Ωα Nα }α forms a right-mapping family (in Ab) and
   
lim (Mα ⊗Ωα Nα ) = lim Mα ⊗ lim Nα .
−→ −→ lim Ω −→
α α −→ α α
α

Proof . The hypothesis (within quotes) means that for all α ≤ β, we have

ψαβ (λα nα ) = θαβ (λα )ψαβ (nα ), for all λα ∈ Ωα and all nα ∈ Nα ,

where ψαβ : Nα → Nβ and θαβ : Ωα → Ωβ , and similarly with the Mα ’s.

Let M = lim Mα ; N = lim Nα ; Ω = lim Ωα and G = lim (Mα ⊗Ωα Nα ). Write cα : Mα → M ;
−→ −→ −→ −→
α α α α
dα : Nα → N and tα : Ωα → Ω, for the canonical maps. We have the maps

cα ⊗ dα : Mα ⊗Ωα Nα −→ M ⊗Ω N,

hence, by the universal mapping property of right limits, there is a unique map, Φ : G → M ⊗Ω N , so that
the following diagram commutes for every α:

G eJJ
Φ / M ⊗Ω N
JJJ oo o7
JJJ ooo
canα JJ oo
J ooo cα ⊗dα
Mα ⊗Ωα Nα

We also need a map, M ⊗Ω N −→ G. Pick m ∈ M and n ∈ N , since the index set is directed we may assume
that there is some α so that m = cα (mα ) and n = dα (nα ). Thus, we have mα ⊗Ωα nα ∈ Mα ⊗Ωα Nα and so,
canα (mα ⊗Ωα nα ) ∈ G. Define Ψ by

Ψ(m, n) = canα (mα ⊗Ωα nα ).

Check (DX) that

(1) Ψ is well-defined,

(2) Ψ is bilinear; thus, by the universal mapping property of tensor, there is a map, Ψ : M ⊗Ω N → G,

(3) Φ and Ψ are inverse homomorphisms.

2.8. FLAT MODULES (AGAIN) 165

Proposition 2.64 Suppose C = Mod(Ω) and Nα0 , Nα , Nα00 , are all right-mapping families of Ω-modules. If
for every α, the sequence
0 −→ Nα0 −→ Nα −→ Nα00 −→ 0 is exact,
then the sequence
0 −→ lim Nα0 −→ lim Nα −→ lim Nα00 −→ 0 is again exact.
−→ −→ −→
α α α

Proof . (DX)

Corollary 2.65 The right-limit of flat modules is flat.

Proof . The operation lim commutes with tensor and preserves exactness, as shown above.
−→
α

Corollary 2.66 Tensor product commutes with arbitrary coproducts. An arbitrary coproduct of flat modules
is flat.
` `
Proof . Look at α∈S Mα . We know from the Problems that α∈S Mα = lim MT , where T ⊆ S, with T
−→
` T
finite and MT = β∈T Mβ . So, given N , we have
a 
N ⊗Ω Mα = N ⊗Ω lim MT
−→
S T
= lim (N ⊗Ω MT )
−→
T
a
= lim (N ⊗Ω Mβ )
−→
T β∈T
a
= (N ⊗Ω Mβ ).
β∈S

The second statement follows from Corollary 2.65 and the fact that finite coproducts of flat modules are flat
(Proposition 2.53).

Remark: Corollary 2.66 extends the last part of Proposition 2.44 that only asserts that tensor commutes with
finite coproducts. It also proves that Proposition 2.53 holds for arbitrary modules, not just f.g. modules. Thus, free
modules are flat and so, projective modules are flat, too.

Proposition 2.67 Say Ω is a ring and M is an Ωop -module (resp. Ω-module). Then, M is flat iff for every
exact sequence
0 −→ N 0 −→ N −→ N 00 −→ 0
of Ω (resp. Ωop )-modules in which all three modules are f.g., the induced sequence

0 −→ M ⊗Ω N 0 −→ M ⊗Ω N −→ M ⊗Ω N 00 −→ 0
(resp. 0 −→ N 0 ⊗Ω M −→ N ⊗Ω M −→ N 00 ⊗Ω M −→ 0)

remains exact.

Proof . Given
0 −→ N 0 −→ N −→ N 00 −→ 0,
an arbitrary exact sequence of Ω-modules, write N = lim Nα , where the Nα ’s are f.g. submodules of N .
−→
α
00 00 00
Let Nα be the image of Nα in N . So, Nα is f.g., too. We get the exact sequence

0 −→ N 0 ∩ Nα −→ Nα −→ Nα00 −→ 0. (∗)
166 CHAPTER 2. RINGS AND MODULES

(α) (α)
Now, N 0 ∩ Nα = lim Nβ , where Nβ ranges over the f.g. submodules of N 0 ∩ Nα . We get the exact
−→
β
sequence
(α) 00
0 −→ Nβ −→ Nα −→ Nα,β −→ 0, (†)

00 (α)
where Nα,β = Nα /Nβ , and all the modules in (†) are f.g. The right limit of (†) is (∗). By hypothesis,
M ⊗Ω (†) is still exact, and the right limit of an exact sequence is exact; so

0 −→ M ⊗Ω (N 0 ∩ Nα ) −→ M ⊗Ω Nα −→ M ⊗Ω Nα00 −→ 0 is exact.

Now, if we pass to the right limit, this time over α, we get

0 −→ M ⊗Ω N 0 −→ M ⊗Ω N −→ M ⊗Ω N 00 −→ 0 is exact.

Theorem 2.68 (FGI-Test)1 An Ω-module, M , is flat iff for all sequences

0 −→ A −→ Ωop −→ Ωop /A −→ 0

in which A is a finitely generated Ωop -ideal, the sequence

0 −→ A ⊗Ω M −→ Ωop ⊗Ω M −→ (Ωop /A) ⊗Ω M −→ 0 is still exact.

Proof . (⇒) is trivial.

(⇐). We proceed in two steps.
Step 1. I claim: For every exact sequence of Ωop -modules of the form
a
0 −→ K −→ Ωop −→ N −→ 0, (∗)
S

in which #(S) is finite, we have an exact sequence

a 
0 −→ K ⊗Ω M −→ Ωop ⊗Ω M −→ N ⊗Ω M −→ 0.
S

We prove this by induction on the minimal number, r, of generators of N . (Note that #(S) ≥ r.) The case
r = 1 has all the ingredients of the general proof as we will see. When r = 1, look first at the base case:
#(S) = 1, too. Sequence (∗) is then:

0 −→ K −→ Ωop −→ N −→ 0. (∗)1

This means that K is an ideal of Ωop and we know K = lim Kα , where the Kα ’s are f.g. Ωop -ideals. Then,
−→
α
(∗)1 is the right limit of

0 −→ Kα −→ Ωop −→ Nα −→ 0, (∗)α

where Nα = Ωop /Kα . Our hypothesis shows that

0 −→ Kα ⊗Ω M −→ Ωop ⊗Ω M −→ Nα ⊗Ω M −→ 0 is exact.

Pass the latter sequence to the limit over α and obtain

0 −→ K ⊗Ω M −→ Ωop ⊗Ω M −→ N ⊗Ω M −→ 0 is exact.
1 FGI stands for finitely generated ideal.
2.8. FLAT MODULES (AGAIN) 167

Thus, the base case #(S) = r = 1 is proved.

We now use induction on #(S) to establish the case #(S) > r = 1. (So, our claim involves an induction
inside an induction.) The induction hypothesis is: For all exact sequences
a
0 −→ K −→ Ωop −→ N −→ 0,
S

in which #(S) = s and r (= minimal number of generators of N ) = 1, tensoring with M leaves the sequence
exact. Say it is true for all sequences with #(S) < s. Given
a
0 −→ K −→ Ωop −→ N −→ 0, #(S) = s,
S
`
pick some σ ∈ S and let Σ = S − {σ}. We have the map Ωop = Ωop σ ,→ S Ωop −→ N , and we let Nσ be
the image of this map in N . This gives the commutative diagram
0 0 0

  
0 / Kσ / Ω = Ωσ / Nα /0

  
0 /K /` Ω /N /0
S

 ` 
0 / K 00 / / N 00 /0
ΣΩ

  
0 0 0
(where N 00 = N/Nσ ) with exact rows and columns and the middle column split-exact. Note that N 00 and
Nσ have r ≤ 1 and when r = 0 the above argument is trivial. Tensor the diagram on the right with M . So,
the top and bottom rows remain exact (by the induction hypothesis and the base case), the middle column
remains exact (in fact, split) and all other rows and columns are exact:
0


0 / Kσ ⊗Ω M / Ω ⊗Ω M / Nα ⊗ Ω M /0

ν
 `  
K ⊗Ω M
α / / N ⊗Ω M /0
S Ω ⊗Ω M

π θ
 `  
0 / K 00 ⊗Ω M / Ω ⊗Ω M / N 00 ⊗Ω M /0
Σ

  
0 0 0 .
We must show that α is an injection. Take x ∈ K ⊗ΩM . Ifα(x) = 0, then θ(α(x)) = 0, which implies that
`
π(x) goes to zero under the injection (K 00 ⊗Ω M −→ Σ Ω ⊗Ω M ), and so, π(x) = 0. Then, there is some
168 CHAPTER 2. RINGS AND MODULES
` 
y ∈ Kσ ⊗Ω M with ν(y) = x. But the map Kσ ⊗Ω M −→ Ω ⊗Ω M −→ S Ω ⊗Ω M is injective and y goes
to zero under it. So, we must have y = 0, and thus, x = 0. This proves that α is injective, and completes
the interior induction (case: r = 1). By the way, α is injective by the five lemma with the two left vertical
sequences considered horizontal and read backwards!

There remains the induction on r. The case r = 1 is proved. If the statement is true for modules N with
< r minimal generators, we take an N with exactly r as its number of minimal generators. Then, for any
finite S, and any sequence
a
0 −→ K −→ Ωop −→ N −→ 0,
S

we choose, as above, σ ∈ S and set Σ = S − {σ} and let Nσ , N 00 be as before. Now redo the argument
involving the 9 term diagram; it shows α is, once again, injective and the claim is proved.

Step 2. I claim that for every sequence

0 −→ N 0 −→ N −→ N 00 −→ 0

of Ωop -modules, all of which are f.g., the sequence

0 −→ N 0 ⊗Ω M −→ N ⊗Ω M −→ N 00 ⊗Ω M −→ 0

remains exact. By the previous proposition, this will finish the proof.

Since N 0 , N and N 00 are all f.g., we have the commutative diagram

0 0 0

 ` 
0 / K0 / Ω / N0 /0
S

  
0 /K /` Ω /N /0
S∪T

  
0 / K 00 /` Ω / N 00 /0
T

  
0 0 0 ,

in which the middle column is split-exact. By tensoring this diagram with M (on the right), we get the
2.8. FLAT MODULES (AGAIN) 169

following commutative diagram with all exact rows (by Step 1) and columns:
0 Ker α

`  
0 / K 0 ⊗Ω M / Ω ⊗Ω M / N 0 ⊗Ω M /0
S

   
0 / K ⊗Ω M / ` Ω ⊗Ω M / N ⊗Ω M /0
S∪T

   
0 / K 00 ⊗Ω M β
/ ` Ω ⊗Ω M / N 00 ⊗Ω M /0
T

  
0 0 0 .

We must show that α is injective. Apply the snake lemma to the first two rows: We get
β
a 
δ
0 −→ Ker α −→ K 00 ⊗Ω M −→ Ω ⊗Ω M is exact.
T

But, Ker β = (0) implies that Im δ = (0), and so, Ker α = (0).
The second (unproven) assertion of Proposition 2.54 now follows from Theorem 2.68.
Corollary 2.69 If Ω is a P.I.D., more generally, a nonzero-divisor ring all of whose f.g. Ωop -ideals are
principal, then M is flat over Ω iff M is Ω-torsion-free.
Proof . The implication (⇒) is always true when Ω has no zero divisors.
(⇐). By the previous theorem, we only need to test against exact sequences of the form
0 −→ A −→ Ωop −→ Ωop /A −→ 0,
where A is a f.g. (hence, principal ) Ωop -ideal. So, there is some λ ∈ Ω with A = λΩ. We have the
commutative diagram
0 /Ω λ /Ω / Ω/λΩ /0

θ

0 /A /Ω / Ω/A /0

(with A considered as right ideal and where θ(µ) = λµ) and all the vertical maps are isomorphisms. Conse-
quently, we may assume that our exact sequence is
λ
0 −→ Ω −→ Ω −→ Ω/λΩ −→ 0.
By tensoring with M , we get the exact sequence
λ
Ω ⊗Ω M −→ Ω ⊗Ω M −→ (Ω/λΩ) ⊗Ω M −→ 0,
which, in view of the isomorphisms Ω ⊗Ω M ∼
= M and (Ω/λΩ) ⊗Ω M ∼
= M/λM , is equivalent to
λ
M −→ M −→ M/λM −→ 0.
Since M has no torsion, multiplication by λ is injective and the sequence is exact.
 170 CHAPTER 2. RINGS AND MODULES

 The corollary is false if Ω is not a P.I.D. Here is an example:

Consider the ring, A = C[X, Y ] (A ∈ CR). The ring A is a domain; so, it is torsion-free. (It’s even a
UFD.) Let M be the ideal of A generated by X and Y . We can write
M = {f ∈ C[X, Y ] | f (X, Y ) = g(X, Y )X + h(X, Y )Y, with g(X, Y ), h(X, Y ) ∈ C[X, Y ]}
= {f ∈ C[X, Y ] | f (0, 0) = 0, i.e., f has no constant term}.
Since M ⊆ A, we see that M is torsion-free.
Claim: M is not flat.
Now, A/M ∼= C, so C is an A-module; how?
The A-module structure on C is as follows: For any f (X, Y ) ∈ A and any λ ∈ C,
f (X, Y ) · λ = f (0, 0)λ.
Note that X · λ = Y · λ = 0. When we consider M as an A-module, write its generators as e1 and e2 . Under
the map M −→ A, we have e1 7→ X and e2 7→ Y . There is a unique nontrivial relation:
Y · e1 − X · e2 = 0.
We claim that e1 ⊗ e2 6= e2 ⊗ e1 in M ⊗A M. To see this, define a map, B : M × M → C.
(a) First, define B on the generators e1 , e2 , by setting
B(e1 , e1 ) = B(e2 , e2 ) = 0, B(e1 , e2 ) = 1, B(e2 , e1 ) = −1.

(b) We need to check that B is well-defined. Let’s check it for the left hand side argument:
        
e1 e1 e1
B Y · e1 − X · e2 , = Y · B e1 , − X · B e2 , .
e2 e2 e2
In the case of e1 , we get X · 1 = 0, and in the case of e2 , we get Y · 1 = 0. The reader should check
similarly that there is no problem for the righthand side argument.

Consequently, we get a linear map, θ : M ⊗ M −→ C. For this linear map,

θ(e1 ⊗ e1 ) = θ(e2 ⊗ e2 ) = 0, θ(e1 ⊗ e2 ) = 1, θ(e2 ⊗ e1 ) = −1.
So, e1 ⊗ e2 6= e2 ⊗ e1 , as contended. Now we will see that M is not flat as A-module. Look at the exact
sequence
0 −→ M −→ A −→ C −→ 0
and tensor it with M. We get
M ⊗A M −→ A ⊗A M −→ C ⊗A M −→ 0 is exact.
However, M ⊗A M −→ A ⊗A M is not injective. To see this, use the isomorphism µ : A ⊗A M ∼
= M, via
α ⊗ m 7→ α · m and examine the composed homomorphism
µ
ϕ : M ⊗A M −→ A ⊗A M −→ M.
Since µ is an isomorphism, all we must prove is that ϕ is not injective. But,
ϕ(e1 ⊗ e2 ) = µ(X ⊗ e2 ) = X · e2
ϕ(e2 ⊗ e1 ) = µ(Y ⊗ e1 ) = Y · e1 .
Yet, X · e2 = Y · e1 and e1 ⊗ e2 6= e2 ⊗ e1 , so ϕ is not injective and M is not flat.
Say Ω is a Λ-algebra and M is a Λop -module, then M ⊗Λ Ω is an Ωop -module. The module M ⊗Λ Ω is
called the base extension of M to Ω.
2.8. FLAT MODULES (AGAIN) 171

Proposition 2.70 Say M is a flat Λ-module, then its base extension, Ω ⊗Λ M , is again a flat Ω-module. If
N is a flat Ω-module and Ω is a flat Λ-algebra, then N considered as Λ-module (via Λ −→ Ω), is again flat
over Λ.

Proof . Assume M is flat as Λ-module. Then, we know that for any exact sequence of Λop -modules,

0 −→ N 0 −→ N −→ N 00 −→ 0,

the sequence
0 −→ N 0 ⊗Λ M −→ N ⊗Λ M is exact.
Now, take any exact sequence of Ω-modules, say

0 −→ N 0 −→ N −→ N 00 −→ 0, (†)

it is still exact as a sequence of Λ-modules. Hence,

0 −→ N 0 ⊗Λ M −→ N ⊗Λ M is exact.

Tensoring (†) with Ω ⊗Λ M over Ω, we get

N 0 ⊗Ω (Ω ⊗Λ M ) −→ N ⊗Ω (Ω ⊗Λ M ) −→ · · · . (††)

We want to show that (††) is exact on the left. But Z ⊗Ω (Ω ⊗Λ M ) ∼

= Z ⊗Λ M , for any Ωop -module, Z.
Hence, (††) becomes
N 0 ⊗Λ M −→ N ⊗Λ M −→ · · · ,
and we already observed that this sequence is exact on the left.
For the second part, take an exact sequence of Λop -modules,

0 −→ M 0 −→ M −→ M 00 −→ 0. (∗)

We need to show that

0 −→ M 0 ⊗Λ N −→ M ⊗Λ N is exact.
Tensor (∗) over Λ with Ω. The resulting sequence

0 −→ M 0 ⊗Λ Ω −→ M ⊗Λ Ω −→ · · · (∗∗)

is still exact as Ω is flat. Tensor (∗∗) with N over Ω; again, as N is flat over Ω, we get

0 −→ (M 0 ⊗Λ Ω) ⊗Ω N −→ (M ⊗Λ Ω) ⊗Ω N −→ · · · is exact.

But the latter exact sequence is just

0 −→ M 0 ⊗Λ N −→ M ⊗Λ N −→ · · · ,

as required.
Harder question: Let P (Λ) be a property of Λ-modules. Say Ω is a Λ-algebra and M is a Λ-module.
Then, we get the Ω-module, Ω ⊗Λ M , the base extension of M to Ω. Suppose, Ω ⊗Λ M has P (Ω). Does M
have P (Λ)?
If so, one says that P descends in the extension Ω over Λ. This matter is a question of descent.
A more realistic question is: Given P , or a collection of interesting P ’s, for which Λ-algebras, Ω, does
(do) P (Ω) descend?
Examples:
172 CHAPTER 2. RINGS AND MODULES

1. P1 (Λ): M is a torsion-free Λ-module.

2. P2 (Λ): M is a flat Λ-module.

3. P3 (Λ): M is a free Λ-module.
4. P4 (Λ): M is an injective Λ-module.
5. P5 (Λ): M is a torsion Λ-module.

Take Λ = Z (a very good ring: commutative, P.I.D), Ω = Q (a field, a great ring), Q is flat over Z (and
Z ,→ Q). Let M = Z q (Z/2Z). (The module M is f.p.) The module M , has, none of Pj (Z) for j = 1, 2, 3, 4.
On the other hand, Q ⊗Z M ∼ = Q, and Q has all of Pj (Q) for j = 1, 2, 3, 4. However, P5 descends in the
extension Q over Z. This follows from

Proposition 2.71 The module, M , is a torsion Z-module iff Q ⊗Z M = (0).

Proof . (⇒). This has already been proved.

(⇐). First, let M be f.g. We know that there is an exact sequence

0 −→ t(M ) −→ M −→ M/t(M ) −→ 0 (†)

where t(M ) is the torsion submodule of M and M/t(M ) is torsion-free; hence (since M is f.g.), M/t(M ) is
free. If we tensor (†) with Q, we get

Q ⊗Z M −→ Q ⊗Z (M/t(M )) −→ 0.

Since Q ⊗Z M = (0), by hypothesis, we get Q ⊗Z (M/t(M )) = (0). Yet, M/t(M ) = qS Z where S is finite;
consequently, S = ∅ and so, M/t(M ) = (0), i.e., M = t(M ). Therefore, M is torsion.
For an arbitrary M , we can write M = lim Mα , where Mα ranges over the f.g. submodules of M . We
−→
α
have an exact sequence
0 −→ Mα −→ M, for all α,
and Q is flat; so,
0 −→ Q ⊗Z Mα −→ Q ⊗Z M is still exact.
But, Q ⊗Z M = (0) implies Q ⊗Z Mα = (0). As the Mα ’s are f.g., the previous argument shows that Mα is
torsion. Then, M = lim Mα is torsion as the right limit of torsion modules is torsion.
−→
α
We now go back to the question. Given the Z-module M , we assume that Q ⊗Z M is torsion. Since Q is
a field, Q ⊗Z M = (0). Proposition 2.71 implies that M is torsion and P5 descends in the extension Q over
Z.

2.9 Further Readings

Rings and modules are covered in most algebra texts, so we shall nor repeat the references given in Section
1.8. Other references include Atiyah MacDonald [3], Lafon [32, 33], Eisenbud [13], Matsumura [39], Malliavin
[38] and Bourbaki [8].
Chapter 3

Commutative Rings

3.1 Introduction
The ordinary arithmetic of the integers and simple generalizations (such as the Gaussian Integers) as well
as of analogues like the polynomial ring in one variable over a field gave rise to the study of number theory
and then to the study of commutative rings. The assumption of commutativity in multiplication makes
possible a much deeper theory with more satisfying applications. Nowadays, a thorough knowledge of this
Chapter is essential in order to do Algebraic Geometry and Algebraic Number Theory (and their mixture:
Arithmetic Algebraic Geometry); one also needs to know the material here for Algebraic Topology. Many
of the results are direct consequences of prodding from geometry, physics and number theory. A modern
problem is to use our physical knowledge (quantum theory), our knowledge of modules and representation
theory, and the hints from the forefront of number theory to augment these results to a new and better
theory of not necessarily commutative rings. This endeavor will probably be a big part of the twenty-first
century in mathematics.

3.2 Classical Localization

All rings in this chapter are commutative with unity.

Definition 3.1 Let A ∈ CR and S ⊆ A be a subset of A. We say that S is a multiplicative subset in A iff

(1) 1 ∈ S

(2) If x, y ∈ S, then xy ∈ S

(3) 0 ∈
/ S.

Examples:

(1) S = Gm (A) = the units of A; the idea is to abstract this case.

(2) S = {α ∈ A | α is not a zero divisor in A}.

(3) S = {x ∈ R | x > 0} ⊆ Gm (R).

(3a) S has property (1) and (2) and is contained in Gm (A).

(4) Given f ∈ A, let S = {f n | n ∈ Z, n ≥ 0} and assume that f ∈

/ N (A) (i.e., f n 6= 0 for all n ≥ 0).

173
174 CHAPTER 3. COMMUTATIVE RINGS

Fix a base ring, C, and look at C-algebras in CR (we get CR when C = Z). Let A and B be C-algebras,
where B varies, and let S be a multiplicative subset in A. Look at

HomC−alg (A, B; S) = {ϕ ∈ HomC−alg (A, B) | ϕ(S) ⊆ Gm (B)}.

Check that B HomC−alg (A, B; S) is a functor from C-algebras to Sets. Is it representable? This means,
is there a C-algebra, S −1 A, and a map (of C-algebras), h : A → S −1 A, so that

θB : HomC−alg (S −1 A, B) ∼
= HomC−alg (A, B; S)

functorially, where θB (ψ) = ψ ◦ h ∈ HomC−alg (A, B; S), as illustrated below:

S −1O A
ψ
/ B
<
xxx
xx
xx ψ◦h
h
xx
A

Proposition 3.1 The functor B HomC−alg (A, B; S) is representable. The representing object, S −1 A, is
called the fraction ring of A w.r.t. S (or the localization of A w.r.t. S). The C-algebra map, h : A → S −1 A,
is the canonical map.

Proof . Look at A × S (in Sets) and form the equivalence relation, ∼, given by:

(a, s) ∼ (b, t) iff (∃u ∈ S)(u(at − sb) = 0 in A).

a
Write for the equivalence class of (a, s). So,
s
a b
= iff (∃u ∈ S)(u(at − sb) = 0).
s t
Define addition and multiplication by:
a b at + sb a b ab
+ = and · = .
s t st s t st
 
−1 a f (c)a 1
Check that these operations are well defined and that S A is a C-algebra c· = ; the C-algebra
s s
a
map, h : A → S −1 A, is given by h(a) = .
1
Functorial part. Given ψ ∈ HomC−alg (S −1 A, B), form ψ ◦ h taking A to B. Now, elements of S become
units in S −1 A, because
s 1 1
· = , the unit element of S −1 A.
1 s 1
But, ψ maps units of S −1 A to units of B, so ψ ◦h ∈ HomC−alg (A, B; S). Next, given ϕ ∈ HomC−alg (A, B; S),
define a
[ϕ] = ϕ(s)−1 ϕ(a) ∈ B.
s
Check
(a) The homomorphism [ϕ] : S −1 A → B is well defined.
(b) θB and ϕ 7→ [ϕ] are inverse maps.
1 Here, f : C → A is the ring homomorphism making A into a C-algebra.
3.2. CLASSICAL LOCALIZATION 175

We can do the same thing with modules. Let M be an A-module and S a multiplicative set in A. Make
(M × S)/ ∼, where ∼ is given by
(m, s) ∼ (n, t) iff (∃u ∈ S)(u(tm − sn) = 0 in M ).
m
Write for the equivalence class of (m, s). Define addition and the action of A by
s
m m0 s0 m + sm0 m am
+ 0 = 0
and a · = .
s s ss s s
This gives the A-module, S −1 M . We have the canonical map, h : M → S −1 M , given by h(m) = m/1.
To discuss what this means, look at the general case of a ring homomorphism, ψ : A → B. We have
two functors: ψ • : Mod(B) Mod(A) (the backward image functor ) and ψ• : Mod(A) Mod(B) (the
forward image functor ). Here, ψ • (M ) = M as an A-module via ψ; that means a · m = ψ(a) · m. The functor
ψ • is an exact functor. Also, the functor ψ• is given by: ψ• (M ) = B ⊗A M . The forward image functor is
only right-exact, in general. These functors form a pair of adjoint functors:
HomB (ψ• (M ), N ) ∼
= HomA (M, ψ • (N )).
Proposition 3.2 The module S −1 M is, in a natural way, an S −1 A-module. The map M S −1 M is a
functor from Mod(A) to Mod(S −1 A) and is left-adjoint to h• . That is,
HomS −1 A (S −1 M, N ) ∼
= HomA (M, h• (N )).
Consequently,
S −1 M ∼
= S −1 A ⊗A M ∼
= M ⊗A S −1 A = h• (M ).
a m am f in
Proof . Let · = , this is well-defined and makes S −1 M into an S −1 A-module. If ϕ : M → M
t s ts
m ϕ(m) f. Check this makes M
Mod(A), the assignment 7→ yields S −1 ϕ : S −1 M → S −1 M S −1 M a
s s
functor.
Say θ ∈ HomS −1 A (S −1 M, N ), set
m
Θ(m) = θ ∈ h• (N ).
1
Now,  am  a m m  m 
a
Θ(am) = θ =θ ·θ
= = a·θ in h• (N ) = a · Θ(m).
1 1 1 1 1 1
So, we have a map from HomS −1 A (S −1 M, N ) to HomA (M, h• (N )) given by θ 7→ Θ. Now, say
ϕ ∈ HomA (M, h• (N )); then, S −1 ϕ ∈ HomS −1 A (S −1 M, S −1 h• (N )). But, if N ∈ Mod(S −1 A), then
S −1 h• (N ) = N , and we get the map in the opposite direction, ϕ 7→ S −1 ϕ. These maps are mutually
inverse. Each of S −1 −; S −1 A ⊗A −; − ⊗A S −1 A, are left adjoint to h• ; so, they are all isomorphic.
Proposition 3.3 The functor M S −1 M is exact, hence, S −1 A is a flat A-algebra.
ϕ ψ S −1 ϕ S −1 ψ
Proof . Given any exact sequence M1 −→ M2 −→ M3 , we will show that S −1 M1 −→ S −1 M2 −→ S −1 M3
ϕ ψ
is again exact. Clearly, as M1 −→ M2 −→ M3 is exact, we have ψ ◦ϕ = 0; and so, (S −1 ψ)◦(S −1 ϕ) = 0. This
shows that Im (S −1 ϕ) ⊆ Ker (S −1 ψ). Say ξ ∈ S −1 M2 and S −1 ψ(ξ) = 0. As ξ = m/s, for some m ∈ M2 and
some s ∈ S, and as S −1 ψ(ξ) = ψ(m)/s = 0 in S −1 M3 , there is some u ∈ S with uψ(m) = 0, i.e., ψ(um) = 0.
By exactness, there is some m0 ∈ M1 so that um = ϕ(m0 ). Consider the element m0 /(su); we have
 0
−1 m ϕ(m0 ) um m
S ϕ = = = = ξ.
su su su s
Therefore, ξ ∈ Im (S −1 ϕ), as required.
Examples:
176 CHAPTER 3. COMMUTATIVE RINGS

(1) S = Gm (A) or more generally, S ⊆ Gm (A). Then, S −1 A = A.

(2) S = all nonzero divisors of A. Here, S −1 A is a bigger ring if we are not in case (1). The ring S −1 A
is called the total fraction ring of A and it is denoted Frac(A). If A is a domain, then Frac(A) is
a field, the fraction field of A. For example, Frac(Z) = Q. The field, Frac(k[X1 , . . . , Xn ]), denoted
k(X1 , . . . , Xn ), is the rational function field in n variables (where k is a field). If A is the ring of entire
(holomorphic) functions, then Frac(A) is the field of meromorphic functions on C. If A = Hol(U ), the
ring of holomorphic functions on an open, U ⊆ C, then Frac(A) = Mer(U ) = the field of meromorphic
functions on U .
/ N (A)}. The ring S −1 A has the special notation Af . Observe that
(3) S = {f n | f ∈ A (f fixed); f ∈
 
α
Af = α ∈ A, n ≥ 0 ,
fn
while, in general,
n α o
Ker (h : A −→ S −1 A) = α ∈ A = 0 = {α ∈ A | (∃u ∈ S)(uα = 0)}.
1

In cases (1) and (2), the map, h, is injective. In case (3), Ker h = {α ∈ A | (∃n ≥ 0)(f n α = 0)}. Consider
the map A[X] −→ Af , via X 7→ 1/f (a 7→ h(a), for a ∈ A). Since aX n 7→ a/f n , our map is surjective.
What is its kernel?
Consider the diagram
X7→1/f
Af [X] / Af
O
h
X7→1/f
A[X] / Af .

The kernel of the top arrow is: (X − 1/f ). The answer to our question is now easily seen to be

{P (X) ∈ A[X] | (∃r ≥ 0)(f r P (X) ∈ (Xf − 1)} = (Xf − 1)ec .

Here, (Xf − 1)ec is, for the moment, just a notation for the left hand side. So,

A[X]/(Xf − 1)ec ∼
= Af .

Generalities on extension (e) and contraction (c).

Let ψ : A → B be a map of rings. Say A is an ideal in A. Let Ae = (the extended ideal ) be the ideal of
B generated by ψ(A). If B is an ideal in B, then let Bc = (the contracted ideal ) be the ideal of A given by

Bc = ψ −1 (B) = {x ∈ A | ψ(x) ∈ B}.

Take B = S −1 A. If A ⊆ A, what is Ae ?
n o
Claim: Ae = α/s | α ∈ A, s ∈ S . Indeed, we have
( n
)
X bi ai
e
A = ai ∈ A, bi ∈ A, si ∈ S .
s 1
i=1 i

1 Pn
Such a sum is of the form ci ai , where σ = s1 · · · sn ; ci ∈ A and ai ∈ A. Since A is an ideal, this sum
σ i=1
is of the form α/σ, where α ∈ A. We have proved part of
3.2. CLASSICAL LOCALIZATION 177

Proposition 3.4 For any commutative ring, A, and any multiplicative subset, S, of A we have:
(1) Gm (S −1 A) = { α/s | (∃b ∈ A)(bα ∈ S)}.
(2) If A ⊆ A then Ae = { α/s | α ∈ A, s ∈ S}.
(3) Ae = (1) = S −1 A iff A ∩ S 6= ∅.

βα 1
Proof . (1) We have α/s ∈ Gm (S −1 A) iff there is some β/t with = 1 = iff (∃u ∈ S)((uβ)α = ust).
ts 1
But, ust ∈ S; so, if we set b = uβ, we get bα ∈ S. The converse is clear.
(2) Already done.
(3) We have Ae = (1) iff some element of Ae is a unit iff α/s is a unit for some α ∈ A iff there is some
b ∈ A with bα ∈ S. But, α ∈ A, so bα ∈ A, yet bα ∈ S; so, A ∩ S 6= ∅. Conversely, if A ∩ S 6= ∅, then
{s/1 | s ∈ S} ∩ Ae 6= ∅. Consequently, Ae has a unit in it, and so, Ae = (1).
Say A ⊆ A, when is A contracted? First an easier question: What is Aec ?
Note: for all v ∈ A, we have A ⊆ (v −→ A) (this only uses the fact that A is a two-sided ideal).
Claim: (v −→ A) = A iff v is not a zero divisor mod A, i.e., v ∈ A/A is not a zero divisor. (Terminology:
v is regular w.r.t, A).
We have (v −→ A) = A iff (v −→ A) ⊆ A iff for every ξ ∈ A, when ξv ∈ A, then ξ ∈ A. Reading this
mod A, we find the above statement is equivalent to

(∀ξ ∈ A/A)(ξv = 0 =⇒ ξ = 0),

which holds iff v is not a zero divisor in A/A.

Going back to the question: What is Aec ?, we have ξ ∈ Aec iff h(ξ) ∈ Ae iff h(ξ) = α/s, for some α ∈ A
and some s ∈ S, iff ξ/1 = α/s iff there is some u ∈ S so that u(ξs − α) = 0, i.e. uξs = uα ∈ A. As us ∈ S,
this implies that there is some v ∈ S with vξ ∈ A. Conversely, if vξ ∈ A for some v ∈ S, then
vξ 1vξ ξ
∈ Ae =⇒ ∈ Ae =⇒ ∈ Ae =⇒ h(ξ) ∈ Ae ,
11 v11 1
and so, ξ ∈ Aec . Therefore,

Aec = {ξ | (∃v ∈ S)(vξ ∈ A)}

= {ξ | (∃v ∈ S)(ξ ∈ (v −→ A))}
[
= (v −→ A).
v∈S
S
Now, A = (1 −→ A) ⊆ s∈S (s −→ A) = Aec .
When is A contracted, i.e., when is it of the form A = Bc , for some B ⊆ S −1 A?
Of course, if A = Aec , then B = Ae will do. In fact, we shall prove that A = Bc for some B ⊆ S −1 A iff
A = Aec . First, we claim that B = Bce for every B ⊆ S −1 A; that is, every ideal, B, of S −1 A is an extended
ideal. For, any ξ in B is of the form ξ = α/s, for some α ∈ A and some s ∈ S. But, sξ ∈ B, too, and so,
α/1 ∈ B, which implies that α ∈ Bc . Consequently, ξ = α/s ∈ Bce . Conversely, if ξ ∈ Bce , then ξ = β/t,
with β ∈ Bc ; it follows that ξ = (1/t)(β/1) ∈ B, and so, B = Bce .
But now, A = Bc implies that Ae = Bce = B; so, Aec = Bc = A. These remarks prove most of the

Proposition 3.5 If A ∈ CR and S is a multiplicative system in A, then

178 CHAPTER 3. COMMUTATIVE RINGS

(1) An ideal, A, of A is contracted iff A = Aec iff every element of S is regular for A.

(2) Every ideal, B ⊆ S −1 A, is extended.

(3) The map, A 7→ Ae , is a one-to-one inclusion-preserving correspondence between all the contracted ideals
of A and all ideals of S −1 A.
(4) If A is noetherian, then S −1 A is noetherian.
S
Proof . (1) We proved earlier that Aec = v∈S (v −→ A) and we know that (v −→ A) = A iff v is regular for
A. So, (1) is now clear.
(2) This has already been proved.
(3) Assume that A and A e have the same extension and both are contracted. Then, by (1) A = Aec and
e=A
A e ec , and since, by hypothesis Ae = A
e e , we get A = A.
e It is also clear that if A ⊆ A,
e then Ae ⊆ A
e e.

(4) (DX) from (1), (2), (3).

The same argument shows the corresponding proposition for modules.

Proposition 3.6 If A ∈ CR and S is a multiplicative system in A, for any module, M ∈ Mod(A),

(1) A submodule, N , of M is contracted iff it is equal to its S-saturation. The S-saturation of N is the
submodule given by [
{ξ ∈ M | (∃v ∈ S)(vξ ∈ N )} = (v −→ N ),
v∈S

where (v −→ N ) = {ξ ∈ M | vξ ∈ N }.
(2) Every submodule of S −1 M is extended, i.e., has the form S −1 N , for some submodule, N , of M .
(3) The map, N 7→ S −1 N , is a one-to-one inclusion-preserving correspondence between all the S-saturated
submodules of M and all submodules of S −1 M .
(4) If M is a noetherian module, then S −1 M is a noetherian module.

Proposition 3.7 Say A ∈ CR and S is a multiplicative system in A. For any ideal, A ⊆ A, we have
(a) The image, S, of S in A/A, is a multiplicative subset provided that S ∩ A = ∅.
−1
(b) S −1 A/Ae −
g→ S (A/A).

Proof . (a) This is trivial.

−1 −1
(b) We have A −→ A/A −→ S (A/A). The elements of S become units in S (A/A). By the universal
−1
mapping property, we have the map S −1 A −→ S (A/A). This map is a/s 7→ a/s; so, it is surjective. We
−1
have a/s = 0 in S (A/A) iff there is some u ∈ S so that u a = 0 iff a/1 ∈ Ae iff a/s ∈ Ae . Therefore, the
−1
kernel of our map is Ae , and so, S −1 A/Ae −
g→ S (A/A).
3.3. PRIME AND MAXIMAL IDEALS 179

3.3 Prime and Maximal Ideals

Recall that an ideal, p, of A ∈ CR is a prime ideal iff p 6= (1) and for all a, b ∈ A, if ab ∈ p, then one of a ∈ p
or b ∈ p holds.

Proposition 3.8 Given a commutative ring, A, for any ideal, A ⊆ A, the following are equivalent:
(1) The ideal, A, is a prime ideal.
(2) The ring A/A is an integral domain.
(3) The set S = A − A = the complement of A is a multiplicative subset of A.
e are two ideals of A and if BB
(4) If B and B e ⊆ A, then one of B ⊆ A or B
e ⊆ A holds.

(5) There is a ring, B, a homomorphism, ϕ : A → B and a maximal ideal, m, of B, so that ϕ−1 (m) = A.
(6) There is a multiplicative set, S ⊆ A, so that
(i) A ∩ S = ∅ and
(ii) A is maximal among the ideals having (i).

Proof . Equivalence of (1)–(4) is known and clear. Now, the inverse image of a prime ideal is always a prime
ideal (DX). Every maximal ideal is prime, so it follows that (5) ⇒ (1). Moreover, (1) implies (6) because
take S = A − p. This is a multiplicative set by (3) and (6) follows tautologically.
(1) ⇒ (5). Given a prime, A, let S = A − A, a multiplicative set by (3) and let B = S −1 A and ϕ = h.
−1
We claim that Ae is a maximal ideal of S −1 A. This is because S −1 A/Ae − g → S (A/A), but A/A is an
−1
integral domain and S = nonzero elements ofSA/A. Consequently, S (A/A) = Frac(A/A) is a field; so, Ae
is a maximal ideal. Now, h−1 (Ae ) = Aec = v∈S (v −→ A). Now, ξ ∈ (v −→ A) iff vξ ∈ A, where v ∈ / A.
But, A is prime, so ξ ∈ A. Therefore, (v −→ A) = A, for all v ∈ S; and so, Aec = h−1 (Ae ) = A and (5)
follows.
(6) ⇒ (1). Given any a, b ∈
/ A, we must show that ab ∈ / A. The hypotheses imply that A + (a) > A and
A + (b) > A, and by (6) (i) and (ii), we have (A + (a)) ∩ S 6= ∅ and (A + (b)) ∩ S 6= ∅. So, there are some
s, t ∈ S, where s = α + ρa, t = β + σb, with α, β ∈ A, ρ, σ ∈ A. Since st ∈ S, it follows that

αβ + ρaβ + σbα + ρσ(ab) ∈ S.

If ab ∈ A, then st ∈ A ∩ S, a contradiction. Therefore, ab ∈

/ A.

Corollary 3.9 Given any multiplicative set, S, in A, there exists a prime ideal, p, so that p ∩ S = ∅.

Proof . Look at S = {A | A an ideal and A ∩ S = ∅}. We have (0) ∈ S, partially order S by inclusion and
check that S is inductive. By Zorn’s lemma, S has some maximal element, p. By (6), the ideal p is prime.

Notation: If S = A − p, where p is a prime ideal, write Ap instead of S −1 A; the ring Ap is called the
localization of A at p. Recall that a local ring is a ring that has a unique maximal ideal.

Corollary 3.10 For any prime ideal, p, in A, the ring Ap is always a local ring and its maximal ideal is
just pe .

Proof . Say A is an ideal of A. Ideals of Ap = S −1 A are extended ideals, i.e., they are of the form Ae . We
have Ae = (1) iff A ∩ S 6= ∅ iff A 6⊆ p. Thus, Ae is a proper ideal iff A ⊆ p; the latter implies that Ae ⊆ pe .
So, pe is the maximal ideal of Ap , as contended.

Remark: We have pec = p. We saw this above in the proof that (1) ⇒ (5).
180 CHAPTER 3. COMMUTATIVE RINGS

Proposition 3.11 Let A ∈ CR be a commutative ring, S be a multiplicative set in A and let P be a prime
ideal of A. Then,
(1) The ideal Pe is a prime ideal of S −1 A iff Pe 6= (1) iff P ∩ S = ∅.
(2) Every prime ideal of S −1 A has the form Pe , for some prime ideal, P, of A.
(3) There is a one-to-one, inclusion-preserving, correspondence between the prime ideals of S −1 A and the
prime ideals, P, of A for which P ∩ S = ∅.

When S = A − p for some prime, p, of A, we have

(10 ) The ideal Pe is a prime of Ap iff P is a prime in A and P ⊆ p.
(20 ) Every prime ideal of Ap is Pe , for some prime, P, of A with P ⊆ p.
(30 ) There is a one-to-one, inclusion-preserving, correspondence between all primes of Ap and the
primes of A contained in p.

Proof . (1) We know that Pe 6= (1) iff P ∩ S = ∅. By definition, a prime ideal is never equal to (1), so, all
−1
we must show is: If P is prime in A, then Pe is prime e
S in S A (of course, P 6= (1)). Say (α/s)(β/t) ∈ P .
e
e ec ec
Then, (αβ)/1 ∈ P , and so, αβ ∈ P . But, P = v∈S (v −→ P) and ξ ∈ (v −→ P) iff vξ ∈ P; moreover,
v∈/ P since P ∩ S = ∅, so, ξ ∈ P. Therefore, Pec = P, and so, αβ ∈ P. Since P is prime, either α ∈ P or
β ∈ P; it follows that either α/s ∈ Pe or β/t ∈ Pe .
(2) If q is a prime in S −1 A, then q = qce and qc is a prime, as qc = h−1 (q). Take P = qc to satisfy (2).
Conversely, Pe is prime iff P ∩ S = ∅.
(3) follows from (1) and (2) and previous work.
Finally, (10 ), (20 ) and (30 ) are special cases of (1), (2) and (3), respectively.

Definition 3.2 If p is a prime ideal of A ∈ CR, look at chains of prime ideals

p = p0 > p1 > · · · > pn ,

where each pj is prime ideal of A. Call n the length of this chain and define the height of p by

ht(p) = sup{length of all chains p = p0 > p1 > · · · > pn }.

Observe that ht(p) might be infinite. Since there is a one-to-one inclusion-preserving correspondence
between the set of all primes, P, contained in p and the set of all prime ideals of Ap , we get

ht(p) = ht(maximal ideal of Ap ).

Definition 3.3 The Krull dimension of a commutative ring, A, denoted dim(A), is the supremum of the
set {ht(m) | m is a maximal ideal of A}.

Hence, we see that ht(p) = dim(Ap ), and

dim(A) = sup{dim(Am ) | m is a maximal ideal of A}.

Examples.
(1) Say dim(A) = 0. This holds iff every prime ideal is maximal iff every maximal ideal is a minimal
prime ideal. An example is a field, or Z/nZ.
3.3. PRIME AND MAXIMAL IDEALS 181

(2) dim(A) =√1. Here, A = a P.I.D. will do. For example, Z, Z[i], Q[T ], more generally, k[T ], for any
field, k. Also, Z[ −5], a non-P.I.D., has dimension 1.
(3) C[T1 , . . . , Tn ] has dimension n (this is not obvious, try it!) Given a commutative ring, A, for appli-
cations to algebraic geometry and number theory, it is useful to introduce two important sets, Spec A and
Max A, and to make these sets into topological spaces. Let

Spec A = {p | p is a prime ideal of A}

Max A = {m | m is a maximal ideal of A}.

The set, X = Spec A, is given a topology (the Zariski topology or spectral topology) for which a basis of
open sets consists of the sets
Xf = {p ∈ Spec A | f ∈/ p} (f ∈ A),
and Max A ⊆ Spec A is given the relative topology.

Remarks:

(1) Xf n = Xf , for all n ≥ 1. This is because f n ∈

/ p iff f ∈
/ p, as p is prime.

(2) Xf g = Xf ∩ Xg . This is because p ∈ Xf g iff f g ∈

/ p iff (f ∈
/ p) and (g ∈
/ p).

/ p, for every prime p iff f ∈ Gm (A) iff Xf = X1 .

(3) Xf = Spec A = X iff f ∈

(4) Xf = ∅ iff f ∈ p, for all primes, p.

S
The open sets in X = Spec A are just T the sets of the form f ∈T Xf , for any subset, T , of A. So, a set,
C, is closed in X iff it is of the form C = T Xfc , where

Xfc = {p ∈ Spec A | p ∈
/ Xf } = {p ∈ Spec A | f ∈ p} = {p ∈ Spec A | (f ) ⊆ p}.

Thus, p ∈ C iff the ideal generated by the set T is contained in p. This suggests the following definition: For
any ideal, A, in A, let
V (A) = {p ∈ Spec A | p ⊇ A}
be the variety defined by A. Then, we have
\ \
V (A) = Xfc = {Xfc | f is part of a generating set for A}.
f ∈A

The dual properties to (1)–(4) are:

(10 ) V (A ∩ B) = V (AB) = V (A) ∪ V (B)

P T P
(20 ) V ( α Aα ) = α V (Aα ) ( α Aα = the ideal generated by the Aα ’s).

(30 ) V (A) = ∅ iff A = (1).

(40 ) V (A) = X = Spec A iff (∀p ∈ Spec A)(A ⊆ p).

From now on, when we refer to Spec A and Max A, we mean these as topological spaces.
To give a more informative criterion for (4) and (40 ), we need to study N (A) = the nilradical of A,
defined by
N (A) = {x ∈ A | xn = 0, for some integer n > 0}.
182 CHAPTER 3. COMMUTATIVE RINGS

This is an ideal of A. Indeed, if x ∈ N (A) and y ∈ A, since A is commutative, we have (yx)n = y n xn = 0.

Also, if x, y ∈ N (A), then there is some integer n ≥ 0 so that xn = y n = 0, and by the binomial formula,

X2n  
2n j
(x ± y)2n = x (±1)2n−j y 2n−j = 0,
j=0
j

since y 2n−j = 0 if j ≤ n and xj = 0 if j ≥ n. Therefore, x ± y ∈ N (A) and N (A) is an ideal.

√
More generally, if A is an ideal, the radical of A, denoted A, is
√
A = {x ∈ A | (∃n ≥ 0)(xn ∈ A)}.
√ √ p
It is easy to check that A is an ideal and that A ⊆ A. Note: (0) = N (A).
√ bar
That A is an√ideal can also be seen as follows: Consider the projection√ map, A −→ A/A, and look at
N (A/A). Then, A is the inverse image of N (A/A) under bar, and so, A is an ideal. Furthermore, by
the first homomorphism theorem, √
A/ A ∼ = (A/A)/N (A/A).
Observe that A/N (A) is a ring without nonzero nilpotent elements. Such
√ a ring is called a reduced ring and
A/N (A) is reduced. We write Ared for A/N (A). Note: (A/A)red = A/ A. For example,
(Z/pn Z)red = Z/pZ, for any prime p.
The following facts are easy to prove (DX):
p√ √
(a) A = A.
√ √ √
(b) A ∩ B = A ∩ B.
√ √
(c) If Ak ⊆ B, for some k ≥ 1, then A ⊆ B.

There is another radical, the Jacobson radical , J (A), given by

\
J (A) = m.
m∈Max(A)

Proposition 3.12 For any ring, A ∈ CR, we have

/ Gm (A) iff there is some maximal ideal, m, so that x ∈ m.
(1) x ∈
(2) If x ∈ J (A), then 1 + x ∈ Gm (A).
T
(3) N (A) = p∈Spec A p; hence N (A) ⊆ J (A).

Proof . (1) is clear (use Zorn’s lemma).

(2) Assume (1 + x) ∈ / Gm (A). By (1), there is some m ∈ Max A, so that 1 + x ∈ m. So, x ∈
/ m (else,
1 ∈ m, a contradiction). As J (A) is contained in every maximal ideal, we get x ∈
/ J (A).
(3) Suppose x ∈ N (A); then, xn = 0, for T some n ≥ 0. Consequently, xn ∈ p, for every prime p; so,
x ∈ p, as p is prime. Conversely, assume x ∈ p∈Spec(A) p. Look at the set S = {xn | n ≥ 0}. Were S a
multiplicative set, then there would be some prime ideal, p, with p ∩ S = ∅. As x ∈ p, this is impossible.
Therefore, S is not a multiplicative set, which happens iff x is nilpotent.
Now, we can give the criteria for (4) and (40 ).
(4) Xf = ∅ iff f ∈ N (A).
3.3. PRIME AND MAXIMAL IDEALS 183

(40 ) V (A) = X = Spec A iff A ⊆ N (A).

Corollary 3.13 Given any ideal, A,

√ \ \
A = {p ∈ Spec A | p ⊇ A} = {p ∈ Spec A | p ∈ V (A)}.

Proof . There is a one-to-oneTcorrespondence between the set of prime ideals, p, containing A and the√set of
prime ideals, p, in A/A. So, {p | p ⊇ A} is the inverse image of N (A/A), but this inverse image is A.
The minimal elements among primes, p, such that p ⊇ A are called the isolated primes of A. Therefore,
√ \
A = {p ∈ Spec A | p is an isolated prime of A}.

Proposition 3.14 The space X = Spec A is always quasi-compact (i.e., compact but not necessarily Haus-
dorff ).
S S
Proof . Say α Uα = X is an open cover of X. Each open Uα has the form Uα = β Xf (α) . Therefore, we
S β
get an open cover α,β Xf (α) = X. If we prove that this cover has a finite subcover, we are done (DX). The
Tβ (α) (α)
hypothesis implies that α,β X c (α) = ∅. However the left hand side is V ((fβ )) and so (fβ ) = (1), by
fβ
previous work. We find
(α ) (α )
1 = cα1 ,β1 fβ1 1 + · · · + cαs ,βs fβs s , for some cαj ,βj ∈ A.
(α ) Ts Ss
Thus, already, (fβj j )sj=1 = (1), and so, j=1 X c (αj ) = ∅. Thus, j=1 X (αj ) = X, a finite cover.
fβ fβ
j j

Remark: The space, Spec A, is almost never Hausdorff. For example,

Spec(Z) = {(0), (2), (3), (5), (7), (11), . . .}, and {(0)} is dense in Spec(Z), i.e., every open set contains (0).
Another geometric example of Spec A and Max A is this:

Proposition 3.15 Let X be a compact, Hausdorff space and write A = C(X) (the ring of real-valued (or
complex-valued) continuous functions on X). For each x ∈ X, write mx = {f ∈ A | f (x) = 0}. Then
(1) Each mx is a maximal ideal of A and
(2) The map x 7→ mx is a bijection of X with Max A. (In fact, x 7→ mx is a homeomorphism).

Proof . Note that the map f 7→ f (x) is a homomorphism of C(X) onto R (resp. C). Its kernel is mx , and so,
mx is maximal. By Urysohn’s lemma, if x 6= y, there is some continuous function, f ∈ A, so that f (x) = 0
and f (y) = 1. Thus, f ∈ mx and f ∈ / my ; it follows that mx 6= my ; so, our map is an injection (of sets).
Take any m in Max A. Say, m 6= mx for all x ∈ X. Given x ∈ X, since m 6= mx , there is some fx ∈ m and
fx ∈/ mx , Therefore, fx (x) 6= 0. Since f is continuous, there is some open subset, Ux , with x ∈ Ux , and
f  Ux 6= 0. Then, the family {Ux } is an open cover of X, and by compactness, it contains a finite subcover,
say {Uxj }tj=1 . We have a function, fxj ∈ m, for each j = 1, . . . , t. Let
t
X  t
X 
F = fx2j F = |fxj |2 , in the complex case .
j=1 j=1

Clearly, F ≥ 0. Pick any ξ ∈ X. Then, there is some j, with 1 ≤ j ≤ t, so that ξ ∈ Uxj , and so, fxj (ξ) 6= 0.
It follows that F (ξ) > 0. Thus, F is never zero on X; consequently, 1/F ∈ A. But now, F is a unit
and yet, F ∈ m, a contradiction. Therefore, the map x 7→ mx is surjective. We leave the fact that it is a
homeomorphism as a (DX).
Here are some useful lemmas on primes.
184 CHAPTER 3. COMMUTATIVE RINGS

Tt
Lemma 3.16 If p is a prime of A and A1 , . . . , At are some given ideals, then p ⊇ j=1 Aj iff p ⊇ Aj , for
some j.
Proof . (⇐). This is a tautology.
Tt Qt
(⇒). Observe that p ⊇ j=1 Aj ⊇ j=1 Aj , and since p is prime, we must have p ⊇ Aj , for some j.
LemmaSt 3.17 (Prime avoidance lemma) Let A be an ideal and let p1 , . . . , pt be some prime ideals. If
A ⊆ j=1 pj , then A ⊆ pj , for some j. (The lemma says that if A avoids all the pj , in the sense that A 6⊆ pj ,
St
then it avoids j=1 pj ).
Proof . We proceed by induction on t. The case t = 1 is obvious. Assume the induction hypothesis S for
t < n. Given n prime ideals, p1 , . . . , pn , by the induction hypothesis, we may assume that A 6⊆ j6=i pj , for
Sn
i = 1, . . . , n. Since, by hypothesis, A ⊆ j=1 pj , for every i = 1, . . . , n, there is some xi ∈ A with
xi ∈ pi and xi ∈
/ pj , for all j 6= i. (†)
Let k be given and form
yk = x1 · · · xk−1 x
ck xk+1 · · · xn ,
where, as usual, the hat over xk means that xk is omitted. Then, yk ∈ pi , for all i 6= k. We claim that
yk ∈
/ pk . Indeed, were it not the case, then we would have yk = x1 · · · x
ck · · · xn ∈ pk ; since pk is prime, there
would be some xj ∈ pk for some j 6= k, a contradiction of (†).
Of course, yk ∈ A, for all k. Now, take a = y1 + · · · + yn .
Sn
Claim. a ∈/ j=1 pj .
Suppose that a ∈ pk , for some k. We can write
X
a = yk + yj ∈ pk , (∗)
j6=k

and since we proved that yj ∈ pk for all j 6= k, the fact that a ∈ pk implies that yk ∈ pk , a contradiction.
Sn
Lemma 3.18 Say p1 , . . . , pn are prime ideals in A, then S = A − j=1 pj is a multiplicative subset of A.
Sn
Proof . We have 0 ∈ / S and 1 ∈ S. Suppose that s, t ∈ S and st ∈ / S. Then, st ∈ j=1 pj , and so, st ∈ pj for
some j; as pj is prime, either s ∈ pj or t ∈ pj , a contradiction.
Now, I.S. Cohen (1950) showed that noetherian-ness of a ring is controlled by its prime ideals.
Lemma 3.19 (Cohen, 1950) If A is an ideal in a commutative ring, A, and if b is an element of A for
which A + (b) is f.g. and (b −→ A) is also f.g., then A is f.g.
Proof . Say A + (b) is generated by β1 , . . . , βt . Each βj is of the form aj + ρj b, for some aj ∈ A and some
ρj ∈ A. So, the elements a1 , . . . , at and b generate A + (b). Let c1 , . . . , cs generate (b −→ A). Then, cj b ∈ A,
for j = 1, . . . , s.
We claim that the elements a1 , . . . , at , c1 b, . . . , cs b generate A.
Pt
Pick α ∈ A, then α ∈ A + (b), and so, α = j=1 vj aj + ρb, with aj as above, for j = 1, . . . , t. But,
t
X
ρb = α − vj aj ∈ A,
j=1
Ps
and so, ρ ∈ (b −→ A). Consequently, we can write ρ = j=1 uj cj , as the cj ’s generate (b −→ A). It follows
that
Xt Xs
α= vj aj + uj (cj b),
j=1 j=1
as contended.
3.3. PRIME AND MAXIMAL IDEALS 185

Proposition 3.20 Let A be a commutative ring, then the following are equivalent:
(1) A is noetherian (A has the ACC).
(2) Every ideal of A is f.g.
(3) A has the maximal condition on ideals.
(4) A has the ACC on f.g. ideals.
(5) (I.S. Cohen, 1950) Every prime ideal of A is f.g.
Proof . We already proved the equivalence (1)–(3) (c.f. Proposition 2.9). Obviously, (1) implies (4) and (2)
implies (5).
(4) ⇒ (1). Suppose
A1 < A2 < A3 < · · ·
is a strictly ascending chain of ideals of A. By the axiom of choice, we can find a tuple, (aj )∞
j=1 , of elements
in A so that aj ∈ Aj and aj ∈ / Aj−1 . Look at the ascending chain
(a1 ) ⊆ (a1 , a2 ) ⊆ (a1 , a2 , a3 ) ⊆ · · · ⊆ (a1 , . . . , an ) ⊆ · · · .
This is a strictly ascending sequence, by the choice of the aj ’s, a contradiction.
(5) ⇒ (2). Take F = {A an ideal of A | A is not f.g.} and partially order F by inclusion. If F is not
empty, it is inductive (DX). By Zorn’s lemma, F has a some maximal element, A. Since A ∈ F, it is not f.g.
and by (5), the ideal A is not prime. So, there exist a, b ∈ A with a, b ∈
/ A and yet, ab ∈ A. Since b ∈
/ A, we
have A + (b) > A. Now, a ∈ (b −→ A) (since ab ∈ A), yet, a ∈ / A, and so, (b −→ A) > A. As A is maximal in
F, it follows that both A + (b) and (b −→ A) are f.g. By Cohen’s lemma, the ideal A is f.g., a contradiction.
Therefore, F = ∅, and (2) holds.
We now move back to modules. Given an A-module, M , we make the definition
Definition 3.4 The support of an A-module, M , denoted Supp(M ) is that subset of Spec A given by
Supp(M ) = {p ∈ Spec A | Mp 6= (0)}.
Proposition 3.21 If M is an A-module, then
Supp(M ) ⊆ V ((M −→ (0))) = V (Ann(M )).
If M is f.g., then
Supp(M ) = V ((M −→ (0))).
So, the support of a f.g. module is closed in Spec A.
Proof . Pick p in Supp(M ), i.e., Mp 6= (0). We need to show that p ∈ V ((M −→ (0))), i.e., p ⊇ (M −→ (0)).
We will show that if p 6⊇ (M −→ (0)) then Mp = (0). But, p 6⊇ (M −→ (0)) implies that there is some s ∈/p
with s ∈ (M −→ (0)). In Mp ,
s m sm
= = 0, as s kills M
1 t t
But, s/1 is a unit in Ap , and so, m/t = 0 already, and Mp = (0).
Now, say M is f.g. with m1 , . . . , mt as generators. Pick p ∈ V ((M −→ (0))), we need to show that
p ∈ Supp(M ). This means, if p ⊇ Ann(M ), then Mp 6= (0). We will prove that if Mp = (0), then
p 6⊇ Ann(M ).
If Mp = (0), then m/1 = 0. So, there is some s = s(m) ∈ S with sm = 0 in M . If we repeat this process
for each of the m1 , . . . , mt that generate M , we get s1 , . . . , st ∈ S such that sj mj = 0, for j = 1, . . . , t. Write
σ = s1 · · · st ∈ S. We get σmj = 0 for all j = 1, . . . , t; so, σ ∈ Ann(M ). But, σ ∈ S implies that σ ∈ / p;
consequently, p 6⊇ Ann(M ).
 186 CHAPTER 3. COMMUTATIVE RINGS

Proposition 3.22 Say M is an A-module (where A ∈ CR). Then, the following are equivalent:

(1) M = (0).

(2) Supp(M ) = ∅.

(2a) Mp = (0), for all p ∈ Spec A.

(3) Supp(M ) ∩ Max A = ∅.

(3a) Mm = (0), for all m ∈ Max A.

Proof . The implications (2) ⇔ (2a) and (3) ⇔ (3a) are obvious. Similarly, (1) ⇒ (2) and (2) ⇒ (3)
are trivial. So, we need to show (3) ⇒ (1). Let us first assume that M is f.g., Then, we know that
Supp(M ) = V ((M −→ (0))). The hypothesis (3) implies that m ⊇ (M −→ (0)) for no maximal ideal, m.
This implies that (M −→ (0)) = (1), the unit ideal. Consequently, 1 ∈ (M −→ (0)), and so, M = (0).
Let us now consider the case where M is not f.g. We can write M = lim Mα , where the Mα ’s range over
−→
α
the f.g. submodules of M . Now, Mα ⊆ M and localization being exact, (Mα )m ⊆ Mm ; so, (Mα )m = (0) for
all m ∈ Max A. By the f.g. case, we get Mα = (0) for all α, and thus, M = (0).

Remark: The implication (3) ⇒ (1) can also be proved without using right limits. Here is the proof. Assume
M 6= (0). Then, there is some m ∈ M with m 6= 0, and let Ann(m) = {a ∈ A | am = 0}; we have Ann(m) 6= (1);
so, Ann(m) ⊆ m, for some maximal ideal, m. Consider m/1 ∈ Mm . Since Mm = (0), we have λm = 0, for some
λ ∈ A − m; thus, λ ∈ Ann(m), and yet λ ∈
/ m ⊇ Ann(m), a contradiction. Therefore, M = (0).

ϕ ψ
Corollary 3.23 If M 0 −→ M −→ M 00 is a given sequence of modules and maps, then it is exact iff for all
p ∈ Spec A, the sequence Mp0 −→ Mp −→ Mp00 is exact iff for all m ∈ Max A, the sequence
0 00
Mm −→ Mm −→ Mm is exact.

Proof . (⇒). This direction is trivial as localization is an exact functor.

0 00
Observe that we need only assume that the sequence Mm −→ Mm −→ Mm is exact for all m ∈ Max A.
Then, (ψ ◦ ϕ)m = ψm ◦ ϕm = 0; so if N is the image of the map ψ ◦ ϕ, we find Nm = (0), for all m ∈ Max A.
By Proposition 3.22, we get N = (0), and thus ψ ◦ ϕ = 0.
Let H = Ker ψ/Im ϕ. The same argument (using exactness of localization) shows that
Hm ∼= (Ker ψ)m /(Im ϕ)m = (0). Again, Proposition 3.22 implies that H = (0) and Ker ψ = Im ϕ, as
contended.

 The statement is not that a whole family of local morphisms comes from a global morphism, rather we
must have the global morphisms and then exactness is a local property.

Local Terminology: If P is property of A-modules (or morphisms), then a module (or morphism) is
locally P iff for every p ∈ Spec A, the module Mp has P as Ap -module.2
Examples: Locally f.g., locally f.p., locally flat, locally exact, locally free, locally zero. etc.
Sometimes, you get a global result from an everywhere local result.

Proposition 3.24 (Local flatness criterion) Say M is an A-module (where A ∈ CR). Then, the following
are equivalent:

(1) M is flat over A.

2 In reality, this ought to be called “pointwise P ”.
3.3. PRIME AND MAXIMAL IDEALS 187

(2) M is locally flat.

(2a) For every p ∈ Spec A, the module Mp is flat over A.

(3) For every m ∈ Max A, the module Mm is flat over Am .

(3a) For every m ∈ Max A, the module Mm is flat over A.

Proof . The implications (1) ⇒ (2) and (2) ⇒ (3) hold, the first by base extension and the second because
it is a tautology. We shall prove that (3) ⇒ (1) (and along the way, (3) ⇐⇒ (3a) and hence, (2) ⇐⇒ (2a)).
Assume 0 −→ N 0 −→ N is exact. Tensoring with M , we get N 0 ⊗A M −→ N ⊗A M . Consider the exact
sequence
0 −→ K −→ N 0 ⊗A M −→ N ⊗A M,
where K = Ker (N 0 ⊗A M −→ N ⊗A M ). By localizing at m, we get the exact sequence

0 −→ K ⊗A Am −→ (N 0 ⊗A M ) ⊗A Am −→ (N ⊗A M ) ⊗A Am . (∗)

It follows that the sequence

0 −→ Km −→ N 0 ⊗A Mm −→ N ⊗A Mm is exact. (∗∗)

Now, for any module, L,

(L ⊗A M ) ⊗A Am ∼
= (L ⊗A Am ) ⊗Am (M ⊗ Am ) ∼
= Lm ⊗Am Mm ,

and so, the sequence

0
0 −→ Km −→ Nm ⊗Am Mm −→ Nm ⊗Am Mm is also exact. (†)
0
Since, the sequence 0 −→ Nm −→ Nm is exact and

(a) Mm is Am -flat; we find Km = (0).

(b) Mm is A-flat; we find Km = (0), again.

But, the above holds for all m ∈ Max A, and thus, K = (0), as required.
This method amounts to studying modules over the Ap ’s and the latter are local rings, where matters
are usually easier. The basic fact is Nakayama’s lemma.

Lemma 3.25 (Nakayama’s lemma) Say A is a commutative ring and J (A) is its Jacobson radical. Suppose
that M is a f.g. A-module and that J (A)M = M . Then, M = (0). That is, if M ⊗A (A/J (A)) = (0), then
M = (0) (recall that M ⊗A (A/J (A)) ∼= M/(J (A)M )).

Proof . Pick a generating set for M of least cardinality. If M 6= (0), this set P
is nonempty. Write m1 , . . . , mt
t
for these generators. As M = J (A)M , we can express mt ∈ M as mt = j=1 αj mj , where αj ∈ J (A).
Consequently,
t−1
X
(1 − αt )mt = αj mj .
j=1
Pt−1
Now, 1−αt ∈ Gm (A), since αt ∈ J (A). Therefore, mt = j=1 αj (1−αt )−1 mj , contradicting the minimality
of t.

Corollary 3.26 (Classical Nakayama) Say A is a local ring and mA is its maximal ideal. Suppose that M
is a f.g. A-module and that mA M = M . Then, M = (0).
188 CHAPTER 3. COMMUTATIVE RINGS

Corollary 3.27 On the category of f.g. modules, A/J (A) is a faithful module. This means if
M ⊗A (A/J (A)) = (0), then M = (0). (In the local ring case, if M ⊗A κ(A) = (0), then M = (0), with
κ(A) = A/mA .)

Corollary 3.28 Let M be an f.g. A-module and say m1 , . . . , mt ∈ M have residues m1 , . . . , mt in

M = M ⊗A (A/J (A)) ∼= M/(J (A)M ) which generate M . Then, m1 , . . . , mt generate M .

Proof . Let N be the submodule of M generated by m1 , . . . , mt . Look at M /N = M/N . Since M is f.g.,

M/N is f.g. and M/N = M /N = (0). By Corollary 3.27, we get M/N = (0), i.e., M = N .

Corollary 3.29 Let M be an f.g. A-module and let N be a submodule for which N + J (A)M = M . Then,
N = M.

Proof . The hypothesis means M = N ; so, M/N = (0). We conclude using Corollary 3.27, again.

Corollary 3.30 Let A be a local ring and M be a f.g. A-module. Write t for the minimal cardinality of a
set of generators for M . Then
(1) A set of elements m1 , . . . , mr generate M iff m1 , . . . , mr span the vector space M ⊗A κ(A).
(2) Every set of generators of M contains a subset generating M with exactly t elements.
The integer t is equal to dimκ(A) (M ⊗A κ(A)).

Proof . (1) The implication (⇒) is clear and the implication (⇐) follows from Corollary 3.28.
(2) For vector spaces, each spanning set contains a basis; this implies that each generating set of M
contains elements which pass to a basis. So, t ≥ d = dimκ(A) (M ⊗A κ(A)). As any basis of a vector space
spans the vector space, Corollary 3.28 shows that M has a generating set of d elements, and so, t ≤ d.
Therefore, t = d.

Proposition 3.31 Let A be a local ring and M be an A-module. Assume one of

(a) A is noetherian and M is f.g.
(b) M is f.p.
Then, the following are equivalent:
(1) M is free over A.
(2) M is projective over A.
(3) M is faithfully flat over A.
(4) M is flat over A.

Proof . The implications (1) ⇒ (2), (2) ⇒ (4) and (1) ⇒ (3), are already known (c.f. Remark (1) after
Definition 2.4 for (1) ⇒ (2) and c.f. Proposition 2.53 and Proposition 2.66 for (2) ⇒ (4) and (1) ⇒ (3)). We
need only prove (4) ⇒ (1). Hypothesis (b) follows from hypothesis (a), so, we assume that M is f.p. and
flat. Pick a minimal set of generators for M , having say, having t generators. We have the exact sequence

0 −→ K −→ At −→ M −→ 0.

As M is f.p. and At is f.g., by Proposition 2.41 (or Proposition 2.17), we know that K is also f.g. Since M
is flat, when we tensor with κ(A), the sequence
Θ
0 −→ K −→ κ(A)t −→ M −→ 0
3.3. PRIME AND MAXIMAL IDEALS 189

remains exact (a Homework problem, but c.f. below). Since the vector spaces M and κ(A)t have the same
dimension, Θ is an isomorphism. So, K = (0). Since K is f.g., by Nakayama’s lemma, K = (0). Therefore,
M∼ = At and M is free over A.

Remark: For the sake of completeness, here is a proof of the fact referred to during the proof of the previous
proposition.

Proposition 3.32 Let Λ be a ring and consider the exact sequence of Λ-modules

0 −→ M 0 −→ M −→ M 00 −→ 0. (∗)

If M 00 is flat, then for any Λop -module, N , the sequence

0 −→ N ⊗Λ M 0 −→ N ⊗Λ M −→ N ⊗Λ M 00 −→ 0 is still exact.

Proof . We can write N as a factor of some free Λop -module, F :

0 −→ K −→ F −→ N −→ 0. (∗∗)

Then, by tensoring (∗) with K, F and N and by tensoring (∗∗) with M 0 , M and M 00 we obtain the following
commutative diagram:
0


K ⊗Λ M 0 / K ⊗Λ M / K ⊗Λ M 00 /0
δ1 δ2 δ3
  
0 / F ⊗Λ M 0 / F ⊗Λ M / F ⊗Λ M 00 /0

  
N ⊗Λ M 0
θ
/ N ⊗Λ M / N ⊗Λ M 00 /0

  
0 0 0
The second row is exact because F is free, and thus flat; the third column is exact because M 00 is flat, and the other
rows and columns are exact because tensor is right-exact. We need to prove that θ : N ⊗Λ M 0 → N ⊗Λ M is injective.
However, this follows from the snake lemma applied to the first two rows.

Theorem 3.33 Let A be a commutative ring and M be an A-module. Assume one of

(a) A is noetherian and M is f.g.
(b) M is f.p.
Then, the following are equivalent:
(1) M is projective over A.
(2) M is flat over A.
(3) M is locally free over A.

Proof . The implication (1) ⇒ (2) is known (c.f. Proposition 2.53 and Proposition 2.66) and (2) ⇒ (3)
follows from Proposition 3.31. We need to prove (3) ⇒ (1). Consider the functor T : N HomA (M, N ); we
must show it is exact. Say
0 −→ N 0 −→ N −→ N 00 −→ 0 is exact
190 CHAPTER 3. COMMUTATIVE RINGS

and apply T . We get

0 −→ HomA (M, N 0 ) −→ HomA (M, N ) −→ HomA (M, N 00 ) −→ C −→ 0, (†)
00
where C is the cokernel of the map HomA (M, N ) −→ HomA (M, N ). We have the lemma (proved in the
Problems):
Lemma 3.34 If B is a flat A-algebra and M is a f.p. A-module, then the canonical map
HomA (M, N ) ⊗A B −→ HomB (M ⊗A B, N ⊗A B)
is an isomorphism.

Let B = Ap , for any p ∈ Spec A. If we localize (†) at p, we get

0 −→ HomA (M, N 0 )p −→ HomA (M, N )p −→ HomA (M, N 00 )p −→ Cp −→ 0,
and Lemma 3.34 implies, this is
0 −→ HomAp (Mp , Np0 ) −→ HomAp (Mp , Np ) −→ HomAp (Mp , Np00 ) −→ Cp −→ 0.
Yet, by (3), M is locally free, i.e., Mp is free over Ap . So, Cp = (0) (since Hom(F, −) is exact for F free).
As p is arbitrary, C = (0).
Proof of Lemma 3.34. Define the map θ : HomA (M, N ) × B −→ HomB (M ⊗A B, N ⊗A B) by
θ(f, b) = b(f ⊗ idB ), for all f ∈ HomA (M, N ) and all b ∈ B.
The map θ is clearly bilinear, so, it induces a canonical linear map
Θ : HomA (M, N ) ⊗A B −→ HomB (M ⊗A B, N ⊗A B).
Since M is an f.p. A-module, there is an exact sequence
a a
A −→ A −→ M −→ 0,
q p

for some integers p, q ≥ 0. Since HomA (−, N ) is a left-exact cofunctor, we get

Y Y
0 −→ HomA (M, N ) −→ HomA (A, N ) −→ HomA (A, N ) is exact.
p q

Tensoring with B, since B is a flat A-algebra, we get

Y Y
0 −→ HomA (M, N ) ⊗A B −→ HomA (A, N ) ⊗A B −→ HomA (A, N ) ⊗A B is exact.
p q

Similarly, the sequence

a  a 
A ⊗A B −→ A ⊗A B −→ M ⊗A B −→ 0 is exact,
q p

i.e., the sequence a a

B −→ B −→ M ⊗A B −→ 0 is exact,
q p

and since HomB (−, N ⊗A B) is a left-exact cofunctor, we get

Y Y
0 −→ HomB (M ⊗A B, N ⊗A B) −→ HomB (B, N ⊗A B) −→ HomB (B, N ⊗A B) is exact.
p q

Thus, we have the commutative diagram

0 / HomA (M, N ) ⊗A B / Qp HomA (A, N ) ⊗A B / Qq HomA (A, N ) ⊗A B

Θ Θp Θq
  
0 / HomB (M ⊗A B, N ⊗A B) / Q
p HomB (B, N ⊗A B) / Q
q HomB (B, N ⊗A B).

But, clearly Θp and Θq are isomorphisms; so, the five lemma shows that Θ is an isomorphism.
 3.3. PRIME AND MAXIMAL IDEALS 191

 These results are wrong if M has no finiteness properties.

nr o
Take A = Z(p) = (s, p) = 1 d
(= Z(p) ∩ Q); this is a local ring, in fact, a local P.I.D. Take M = Q
s nr o
as Z(p) -module. What is κ(p) = Z(p) /mp , where mp = (p)e = r ≡ 0 (mod p), (s, p) = 1 ? We have
s
Z(p) /mp is equal to the localization of Z/pZ, i.e., κ(p) = Z/pZ. How about Q⊗Z(p) κ(p)? We have a surjection
Q ⊗Z κ(p) −→ Q ⊗Z(p) κ(p). But, Q ⊗Z κ(p) = (0), so Q ⊗Z(p) κ(p) = (0). Therefore, κ(p) is not faithful
on Q. Now, were Q free, then Q ⊗Z(p) κ(p) would be a vector space of rank equal to rk(Q) over κ(p). So,
Q is not free overZ(p) . But Q is flat over Z(p) as Q is (Z(p) )(0) (the localization of Z(p) at (0)). Note:
1
Q = lim Z(p) n .
−→ p
n
Remarks on Mp , for any A module, M .
Let S = A − p, for a given p ∈ Spec A. We can partially order S:

f ≤g iff f | gn for some n > 0,

i.e. iff there is some ξ ∈ A with f ξ = g n . (Note, ξ ∈ S, automatically). Check: This partial order has the
Moore–Smith property. So, we can form lim Mf .
−→
f ∈p
/

Claim: lim Mf = Mp .
−→
f ∈p
/

We have maps Mf −→ Mp , for all f , and the commutative diagram

Mp
{= aDD
DD
{{{ DD
{{ DD
{{ ϕg
Mf
f
/ Mg

for all f ≤ g. (Since f ≤ g iff f ξ = g n for some ξ ∈ S and some n > 0, the map ϕgf is given by
  r
ϕgf fmr = mξ g
lim Mf −→ Mp . To go
g nr .) Check that ϕf is well-defined (DX). Hence, there exists a map −→
f ∈p
/
backwards, pick ξ ∈ Mp . The element ξ is the class of some m/s, with s ∈
/ p. Now, m/s ∈ Ms ; hence,
cans (m/s) ∈ lim Mf . Check that
−→
f ∈p
/

(1) ξ 7→ cans (m/s) is well defined. It maps Mp −→ lim Mf .

−→
f ∈p
/

(2) The map (1) and lim Mf −→ Mp from above are mutually inverse.
−→
f ∈p
/

Geometric Interpretation: We claim that f ≤ g iff Xg ⊆ Xf .

T p
Indeed, Xg ⊆ Xf iff V ((f )) ⊆ V ((g)) iff p ⊇ (f ) implies p ⊇ (g) iff p⊇(f ) p ⊇ (g) iff (f ) ⊇ (g) iff
p p p p p
(f ) ⊇ (g). Now, (f ) ⊇ (g) iff g ∈ (f ) iff g n ∈ (f ) for some n > 0 iff f | g n iff f ≤ g. This shows
that lim Mf = Mp and so, Mp represents germs of some kind. We will come back and elucidate this point
−→
Xf 3 p
later. However, we want to note that for ideals, A and B, the reasoning above shows that
√ √
V (A) ⊆ V (B) iff A ⊇ B.
192 CHAPTER 3. COMMUTATIVE RINGS

Remark: The following proposition involving comaximal ideals will be needed in the next Chapter and is often
handy.
Two ideals a and b of a ring A are comaximal iff a + b = A. The following simple fact holds (DX): If a, b1 , . . . , bn
are ideals so that a and bi are comaximal for i = 1, . . . , n, then a and b1 · · · bn are comaximal.

Proposition 3.35 (Chinese Remainder Theorem) Let a1 , . . . , an be ideals of a ring A. If for all i 6= j, the ideals ai
and aj are comaximal, then
(1) The canonical map ϕ : A → n
Q
i=1 A/ai is surjective.
Tn Qn
(2) Ker ϕ = i=1 ai = i=1 ai .
Consequently, we have a canonical isomorphism
n
.Y  n
Y
ψ: A ai → (A/ai ).
i=1 i=1
Qn
Moreover, the converse of (1) holds: If the canonical map ϕ : A → i=1 A/ai is surjective, then for all i 6= j, the
ideals ai and aj are comaximal.

Proof . We prove (1) and (2) together

Q by induction on n. If n = 2, there exist e1 ∈ a1 and e2 ∈ a2 with e1 + e2 = 1.
For any element (a1 , a2 ) ∈ A/a1 A/a2 , let a = e2 a1 + e1 a2 . Then,

πi (a) = πi (e2 a1 ) + πi (e1 a2 ) = ai , i = 1, 2

(where πi : A → A/ai is the canonical projection onto A/ai ). Thus, ϕ is surjective.

Since a1 a2 ⊆ a1 ∩ a2 , it is enough to prove that a1 ∩ a2 ⊆ a1 a2 . Now, as 1 = e1 + e2 , for every a ∈ a1 ∩ a2 , we
have a = ae1 + ae2 ; however, ae1 ∈ a1 a2 and ae2 ∈ a1 a2 , so a ∈ a1 a2 . As Ker ϕ = a1 ∩ a2 , we find Ker ϕ = a1 a2 .
For the induction step, observe that (by the fact stated just before Proposition 3.35), b = a1 · · · an−1 and
Tn−1an are
comaximal.
Qn−1 Then, by the
Tn case n
Qn= 2, we have b ∩ an = ban ; moreover, by the induction hypothesis, b = i=1 ai =
i=1 ai , so we have i=1 ai = i=1 ai .

By the case n = 2, we have an isomorphism

A/ban ∼
Y
= (A/b) (A/an )

and by the induction hypothesis, we have an isomorphism

n−1
A/b ∼
Y
= (A/ai ).
i=1

Therefore, we get an isomorphism

n
.Y  Yn
A ai ∼
= A/ai .
i=1 i=1
Qn
Finally, assume that the canonical map ϕ : A → i=1 A/ai is surjective. Pick i, j with i 6= j. By surjectivity,
there is some a ∈ A so that πi (a) = 0 and πj (a) = 1, i.e., πj (1 − a) = 0. Therefore, a ∈ ai and b = 1 − a ∈ aj with
a + b = 1, which proves ai + aj = A.
The classical version of the Chinese Remainder Theorem is the case where A = Z and ai = mi Z, where the
m1 , . . . , mn are pairwise relatively prime natural numbers. The theorem says that given any natural numbers
k1 , . . . , kn , there is some natural number, q, so that

q ≡ ki (mod mi ), i = 1, . . . , n,

and the solution, q, is unique modulo m1 m2 · · · mn .

Proposition 3.35 can be promoted to modules.
3.3. PRIME AND MAXIMAL IDEALS 193

Proposition 3.36 Let M1 , . . . , Mn be submodules of the A-module, M . Suppose the Mi are pairwise comaximal
(Mi + Mj = M ), then the natural map
n
. \  n
Y
M Mi −→ (M/Mi )
i=1 i=1

is an isomorphism. (Observe that, Mi = ai M with the ai comaximal ideals, is a special case.)

194 CHAPTER 3. COMMUTATIVE RINGS

3.4 First Applications of Fraction Rings

A) Rings with the DCC
In this subsection, every ring is a commutative ring with unity.

Lemma 3.37 If the ring A has the DCC, then Max(A) = Spec(A) and #(Max(A)) is finite. Thus,
dim(A) = 0.

Proof . Note, Max(A) = Spec(A) iff dim(A) = 0, in any commutative ring A. Pick p ∈ Spec(A) and look at
A/p; the ring A/p is a domain and it has the DCC. But, every integral domain with the DCC is a field and
conversely. This is proved as follows: Say D is a domain with the DCC, and pick x 6= 0 in D. Look at the
decreasing chain
(x) ⊇ (x2 ) ⊇ (x3 ) ⊇ · · · ⊇ (xn ) ⊇ · · · .
By the DCC, there is some n so that (xn ) = (xn+1 ). Thus, xn ∈ (xn+1 ), and so, there is some u ∈ D with
xn = uxn+1 . It follows that xn (1 − ux) = 0; as x 6= 0 and D is a domain, we get 1 − ux = 0, so, x−1 = u
and D is a field. Therefore, p is maximal since A/p is a field.
Let S be the set of finite intersections of distinct maximal ideals of A. Of course, S = 6 ∅, so, by the DCC,
S has a minimal element, say m1 ∩ m2 ∩ · · · ∩ mn . We claim that m1 , m2 , . . . , mn are all the maximal ideals
of A.
Take another maximal ideal, m, and look at m ∩ m1 ∩ m2 ∩ · · · ∩ mn . This ideal is in S and

m ∩ m1 ∩ m2 ∩ · · · ∩ mn ⊆ m1 ∩ m2 ∩ · · · ∩ mn .

By minimality, we have
m ⊇ m1 ∩ m2 ∩ · · · ∩ mn ⊇ m1 m2 · · · mn .
As m is prime, m ⊇ mj , for some j; but both m and mj are maximal, so m = mj .

Lemma 3.38 If A is a noetherian ring, then every ideal, A, contains a product of prime ideals. In particular,
(0) is a product of prime ideals.

Proof . (Noetherian induction) Say the conclusion of the lemma is false and let S denote the collection of all
ideals not containing a finite product of prime ideals. By assumption, S = 6 ∅. Since A is noetherian, S has a
maximal element, A. The ideal A can’t be prime; so, there exist a, b ∈ / A and yet, ab ∈ A. As A + (a) > A,
we have A + (a) ⊇ p1 · · · pr , for some primes pi . Similarly, A + (b) ⊇ q1 · · · qs , for some primes qj . Now, we
have A = A + (ab), since ab ∈ A; consequently, we get

A = A + (ab) ⊇ (A + (a))(A + (b)) ⊇ p1 · · · pr q1 · · · qs ,

a contradiction. Therefore, S = ∅ and the lemma holds.

Proposition 3.39 (Akizuki, 1935) Say A is a local ring with the DCC. Then, the maximal ideal, m, of A
is nilpotent (i.e., mn = (0) for some n ≥ 1) and A is noetherian. The converse is also true.

Proof . (Nagata) Consider the chain

m ⊇ m2 ⊇ m3 ⊇ · · · ⊇ mn ⊇ · · · ,

it must stop, by the DCC. Thus, there is some n > 0 so that mn = mn+1 . Were mn 6= (0), the set
S = {A | Amn 6= (0)} would not be empty as m ∈ S. By the DCC, the set S has a minimal element, call it
A. Let p = Ann(Amn ). We claim that p is a prime ideal. Pick a, b ∈
/ p. Then, by definition of p, we have
aAmn 6= (0) and bAmn 6= (0). Yet, aA ⊆ A and bA ⊆ A and A is minimal in S. Therefore,

aA = bA = A.
3.4. FIRST APPLICATIONS OF FRACTION RINGS 195

Now,
abAmn = a(bA)mn = (aA)mn = Amn 6= (0),
and so, ab ∈/ p. Consequently, p is indeed prime. By Lemma 3.37, the prime ideal, p, is maximal; as A is a
local ring, we get m = p. As m = p = Ann(Amn ), we have mAmn = (0), so, Amn+1 = (0), i.e., Amn = (0)
(remember, mn = mn+1 ), a contradiction. Therefore, the maximal ideal, m, of A is nilpotent.
To prove A has the ACC, argue by induction on the least n so that mn = (0). When n = 1, we have
m = (0) and A = κ(A) is a field. Since every field has the ACC, we are done. Assume that the induction
hypothesis holds for all r < n. Consider the exact sequence

0 −→ mn−1 /mn (= mn−1 ) −→ A/mn (= A) −→ A/mn−1 −→ 0.

The left hand term has the DCC and is a module over A/m = κ(A); so, it is vector space over κ(A) and
it is finite dimensional. Consequently, it has the ACC, The righthand term has the ACC, by the induction
hypothesis. It follows that the middle term, A, has the ACC.
Now, for the converse, assume that A is noetherian, local and that mn = (0) for some n ≥ 1. We prove
that A has the DCC by induction on the index of nilpotence of m. When n = 1, the ring A = A/m is a field
and so, it has the DCC. Assume that the induction hypothesis holds for all r < n. Say mn = (0). Then, we
have the exact sequence

0 −→ mn−1 /mn (= mn−1 ) −→ A/mn (= A) −→ A/mn−1 −→ 0,

where the righthand side has the DCC by the induction hypothesis. But, the left hand side is a module over
A/m = κ(A); so, it is vector space over κ(A) and it has the ACC because A does. Thus, mn−1 is a finite
dimensional vector space, and so, it has the DCC. Therefore, A is caught between two DCC modules, and
A is artinian.

Theorem 3.40 (Akizuki’s structure theorem, 1935) If A is a commutative ring with unity, then A has the
DCC iff A has the ACC and Max(A) = Spec(A) (i.e., dim(A) = 0). When A has the DCC, the map
Y
θ: A → Ap (∗)
p∈Spec(A)

is an isomorphism and each Ap is an Artin local ring. Moreover, each map hp : A → Ap is a surjection.

Proof . (⇒) By Lemma 3.37, we have Max(A) = Spec(A) and Max(A) only has finitely many elements.
Therefore, the product in (∗) is a finite product. Each Ap is local with the DCC, so, it has the ACC (and
its maximal ideal is nilpotent), by Proposition 3.39. If θ is an isomorphism, we are done with this part.
(1) The map θ is injective (this is true in general). Pick a ∈ A and look at the principal ideal (a) = Aa.
If θ(a) = 0, then (Aa)p = (0) for every prime, p ∈ Spec(A). Therefore, Aa = (0), so, a = 0.
(2) The map θ is surjective. The ideal pe in Ap is nilpotent. So, (pe )n = (0) in Ap , yet (pe )n = (pn )e ,
and thus,
Ap = Ap /(pe )n = Ap /(pn )e = (A/pn )p ,
where p is the image of p in A/pn . Now, p is the unique prime ideal of A which contains pn (since Spec(A) =
Max(A)). Therefore, A/pn is a local ring and p is its maximal ideal. It follows that (A/pn )p = A/pn , and so
Ap ∼= A/pn . Each hp is thereby a surjection. Since pnp and qnq are pairwise comaximal, which means that
(1) = pnp + qnq (because Spec(A) = Max(A)), the Chinese Remainder Theorem implies that θ is surjective.
(⇐) This time, A has Q the ACC and Max(A) = Spec(A). By Lemma 3.38, the ideal (0) is a product of
t
maximal ideals, say (0) = j=1 mj . Let m be any maximal ideal. Now 0 ∈ m implies that m ⊇ mj , for some
196 CHAPTER 3. COMMUTATIVE RINGS

j. Since both m and mj are maximal, m = mj . Thus, m1 , . . . , mt are all the maximal ideals of A. Consider
the descending chain
A ⊇ m1 ⊇ m1 m2 ⊇ · · · ⊇ m1 · · · mt = (0).
In this chain, we have m1 · · · ms−1 ⊇ m1 · · · ms . The module m1 · · · ms−1 /m1 · · · ms is an A/ms -module,
hence, a vector space, since A/ms is a field. By hypothesis, this vector space has the ACC. Thus, it is
finite-dimensional and it has the DCC. But then, m1 · · · ms−1 /m1 · · · ms has a composition series. If we do
this for each s, we obtain a composition series for A. Consequently, A has finite length as A-module, so, it
has the DCC.

Remark: This is false for noncommutative rings. Take the ring R of n × n lower triangular matrices over
C. The “primes of R” are n inQnnumber and the localization at the j-th one, Mj , is the full ring of j × j
matrices over C. But, θ : R → j=1 Mj (C) is only injective, not surjective.

B) Locally Free f.g. A-Modules.

We begin by restating and reproving that Supp(M ) is closed when M is f.g.

Lemma 3.41 If M is a f.g. A-module and if Mp = (0) for some p ∈ Spec A, then there exists some σ ∈
/p
so that σM = (0) and Mσ = (0).

Proof . Write m1 , . . . .mt for generators of M . Then, mj /1 = 0 in Mp = (0). So, there is some sj ∈ / p with
sj mj = 0 for j = 1, . . . , t. Let σ = s1 · · · st , then σmj = 0 for j = 1, . . . , t. Consequently, σM = (0) and
mj /1 = 0 in Mσ for j = 1, . . . , t, so, Mσ = (0).
Geometric Interpretation. If ϕ : A → B is a ring map we get a map, ϕa : Spec B → Spec A, namely,
q 7→ ϕ−1 (q). This is a continuous map (because (ϕa )−1 (V (A)) = V (B · ϕ(A)), for every ideal A ⊆ A). Since
there is a map A −→ As , we get a map Spec(As ) −→ Spec(A). For this map we have

Proposition 3.42 The map Spec(As ) −→ Spec(A) takes Spec(As ) homeomorphically onto the open set,
Xs , of Spec A.

Proof . We make a map Xs −→ Spec(As ). For this, observe that p ∈ Xs iff s ∈ / p iff pe ∈ Spec(As ). Thus,
the desired map is p 7→ pe . Now, q = pe iff p = qc = inverse image of q; therefore, our maps are inverse to
one-another and the image of the contraction is Xs (an open set in Spec A). We must now show that the
map Xs −→ Spec(As ) via p 7→ pe is continuous. The open Xs has as basis of opens the Xs ∩ Xt = Xst ,
where t ∈ A. The topology in Spec(As ) has as basis the opens Yτ , where τ ∈ As and q ∈ Yτ iff τ ∈ / q. We
have τ = t/sn , for some t and some n. Moreover, q = pe ; so τ ∈ / q iff t ∈
/ p and it follows that Xs ∩ Xt
corresponds to Yτ .
To continue with the ‘geometric interpretation, let M be an A-module. We make a presheaf over Spec A
f. For every open subset, U , in X = Spec A,
from M , denote it by M
 
 [ (1) f (p) ∈ Mp 
f(U ) = f : U −→
M Mp (2) (∀p ∈ U )(∃m ∈ M, ∃s ∈ A)(s ∈ / p,
i.e., p
∈ Xs )
 
p∈U (3) (∀q ∈ Xs ∩ U ) f (q) = image ms in Mq .

The intuition is that Mf(U ) consists of kinds of functions (“sections”) such that for every “point” p ∈ U ,
each function is locally defined in a consistent manner on a neighborhood (Xs ∩ U ) of p (in terms of some
element m ∈ M ).
f, is in fact a sheaf on Spec A (where Spec A has the Zariski
The reader should prove that the presheaf, M
topology) (DX).
f (DX):
Here are two important properties of the sheaf M
3.4. FIRST APPLICATIONS OF FRACTION RINGS 197

f is an exact functor of M . This means, if

(1) M
0 −→ M 0 −→ M −→ M 00 −→ 0
is an exact sequence of A-modules, then
f0 −→ M
0 −→ M f −→ M
g00 −→ 0

is an exact sequence of sheaves. (Recall that if F −→ G is a morphism of sheaves, it is surjective iff

for every open, U , and every ξ ∈ G(U ), there is a covering {Uα −→ U }α so that ξα = ρU
U (ξ) ∈ G(Uα )
α

comes from some ηα ∈ F(Uα ) for all α.)

(2) The functor M f commutes with arbitrary coproducts, i.e., if M = ` Mα , then M

M f=` M fα .
α α

The easiest way to see (1) and (2) is via the following ideas: Say F is a presheaf on some space X. If
x ∈ X is a point, let Fx = lim F(U ). We call Fx the stalk of the presheaf , F, at x.
−→
U3 x

f at p. This is immediate from the definition (DX).

Remark: The module, Mp , is the stalk of M
Proposition 3.43 Say θ : F → G is a map of sheaves (with values in a category based on sets, e.g., sets,
groups, rings, ...) and suppose for all x ∈ X, the map θx : Fx → Gx is injective (resp. surjective, bijective).
Then θ is injective (resp. surjective, bijective). If Fx = (0) for all x ∈ X, then F = (0). (Here, F has values
in groups or modules.)
Proof . One checks that F Fx is an exact functor of F (for each x ∈ X). Then the last statement implies
all the others. For example,
θ
0 −→ Ker θ −→ F −→ G −→ Coker θ −→ 0 is exact;
so, take stalks at x. We get
x θ
0 −→ (Ker θ)x −→ Fx −→ Gx −→ (Coker θ)x −→ 0 is exact.
If θx is injective, then (Ker θ)x = (0). By the last statement of the proposition, Ker θ = 0, etc. So, we need
to prove that Fx = (0) for all x ∈ X implies that F = (0).
Pick an open, U , of X and pick any x ∈ U . We have Fx = lim F(V ) (with V ⊆ U ). If ξ ∈ F(U ), then
−→
V3x
ξx = image of ξ in Fx = 0. This means that there is some open subset, V = Vx , with ρVU (ξ) = 0 in F(V ).
Then, as x ranges over U , we have a cover, {Vx −→ U }, of U and ρVUx (ξ) = 0, for all Vx in the cover. By the
uniqueness sheaf axiom, we must have ξ = 0. Since ξ is arbitrary in F(U ), we get F(U ) = (0).
It is clear that the remark and this proposition imply (1) and (2) above.
As a special case of the tilde construction, if we view A has a module over itself, we can make the sheaf
e on X, usually denoted OX . More explicitly, for every open subset, U , in X = Spec A,
A
 

 (1) f (p) ∈ Ap 

[
OX (U ) = f : U −→ Ap (2) (∀p ∈ U )(∃a, g ∈
 A)(g ∈
/ p, i.e.,
 p ∈ Xg ) 

 
p∈U (3) (∀q ∈ Xg ∩ U ) f (q) = image ag in Aq . 

Observe that OX is a sheaf of local rings, which means that OX (U ) is a ring for all U and OX,p (= Ap )
f is a sheaf of modules over OX .
is a local ring, for every p. The sheaf M
Given a module M and an element s ∈ A, we have the sheaves M f  Xs and Mfs . Note that Mfs is a sheaf
f
on Spec(As ) and M  Xs is a sheaf on Xs , but the map Spec(As ) −→ Spec A gives a homeomorphism of
Spec(As ) −g → Xs .
198 CHAPTER 3. COMMUTATIVE RINGS

Proposition 3.44 Under the homeomorphism, ϕ : Spec(As ) −

g fs and M
→Xs , the sheaves M f  Xs correspond.

Proof . Say ϕ : X → Y is a continuous map of spaces and F is a sheaf on X. We can make ϕ∗ F, a new sheaf
on Y , called the direct image of F. For any open, V , in Y , set

ϕ∗ F(V ) = F(ϕ−1 (V )).

fs ) and M
The sense of our proposition is that ϕ∗ (M f  Xs are isomorphic as sheaves on Xs . Now, ϕ∗ (Mfs )(U )
fs (ϕ (U )), where U is an open in Xs ⊆ Spec A. The map ϕ : Y = Spec As → Xs is just
is just M −1

q ∈ Spec(As ) 7→ qc ∈ Spec A. We have

 
 [ (1) f (p) ∈ (Ms )p 
fs (ϕ−1 (U )) = f : ϕ−1 (U ) −→
M (Ms )p (2) (∀p ∈ ϕ−1 (U ))(∃µ ∈ Ms , ∃τ ∈ As )(p ∈
Yτ )


p∈ϕ−1 (U ) (3) (∀q ∈ Yτ ∩ ϕ−1 (U )) f (q) = image µτ in (Ms )q .

Now, q ∈ ϕ−1 (U ) iff q = pe and p ∈ U ⊆ Xs . We also have µ = m/sn , for some m ∈ M ; τ = t/sn , for some
f  Xs (U ) −→ ϕ∗ (M
t ∈ A, and so, µ/τ = m/t. It follows that there exists a natural map, M fs )(U ), via
f [given by m/t] 7→ f [given by (m/sn )/(t/sn )] = µ/τ .
f  Xs −→ ϕ∗ (M
This gives a map of sheaves, M fs ). We check that on stalks the map is an isomorphism:
f f
(M  Xs )p = Mp and ϕ∗ (Ms )q = (Ms )q = (Ms )pe = Mp . Therefore, our global map, being a stalkwise
isomorphism, is an isomorphism.
f)p is just Mp . So,
Recall that the stalk (M

f(Xf ).
Mp = lim Mf = lim Mf = lim M
−→ −→ −→
f ∈p
/ p∈Xf p∈Xf

f. Thus, Ap =
Consequently, Mp consists indeed of “germs”; these are the germs of “sections” of the sheaf M
germs of functions in OX (U ), for any p ∈ U .
Say X is an open ball in Rn or Cn . Equip X with the sheaf of germs of C k -functions on it, where
0 ≤ k ≤ ∞ or k = ω:
 
 [ (1) f (u) ∈ OX,u (germs of C k -functions at u) 
OX (U ) = f : U −→ OX,u (2) (∀u ∈ U )(∃ small open X ⊆ U )(∃ C k -function, g, on X )
 
u∈U (3) (∀u ∈ X ) (f (u) = image(g) in OX,u ) .

For Cn and k = ω, we can take g to be a power series converging on X . Observe that OX is a sheaf of local
rings (i.e., OX,u (= germs at u) is a local ring).
The concept of a sheaf help us give a reasonable answer to the question, “what is geometry?”
A local ringed space (LRS) is a pair, (X, OX ), so that
(1) X is a topological space.
(2) OX is a sheaf of local rings on X.

Examples.
(1) Open balls in Rn or Cn , with the sheaf of germs of C k functions, for a given k, are local ringed spaces.
e is an LRS.
(2) (Spec A, A)

The LRS’s form a category, LRS. A map (X, OX ) −→ (Y, OY ) is a pair of maps, (ϕ, Φ), such that:
3.4. FIRST APPLICATIONS OF FRACTION RINGS 199

(a) ϕ : X → Y is a continuous map.

(b) Φ : OY → ϕ∗ OX is a homomorphism of sheaves of rings.

Now, geometry is the study of local ringed spaces that are locally standard , i.e., each point x ∈ X has a
neighborhood, U , and the LRS (U, OX  U ) is isomorphic to a standard model.
Some standard models:

(a) C k , real geometry (C k -manifolds): The standards are open balls, U , in Rn and OX (U ) is the sheaf of
germs of real C k -functions on U . (Here, 1 ≤ k ≤ ∞, and k = ω is also allowed).
(b) Holomorphic geometry: k = ω. The standards are open balls, U , in Cn and OX (U ) is the sheaf of
germs of complex C ω -functions on U (complex holomorphic manifolds).
e
(c) Algebraic geometry: The standard model is (Spec A, A).

Notice that we can “glue together” standard models to make the geometric objects that are locally
standard. Namely, given a family {(Uα , OUα )}, of standard models of fixed kind, suppose for all α, β, there
exist some opens Uαβ ⊆ Uα and Uβα ⊆ Uβ and isomorphisms ϕβα : (Uαβ , OUα  Uαβ ) → (Uβα , OUβ  Uβα ), and
suppose we also have the gluing conditions: ϕβα = (ϕαβ)
−1
and ϕγα = ϕγβ ◦ ϕβα on Uα ∩ Uβ , then we can glue all
the (Uα , OUα ) together. That is, there is an LRS, (X, OX ), and it is locally isomorphic to each (Uα , OUα ).
What about a geometric interpretation of some of our previous results?
Consider Lemma 3.41: Given a f.p. module, M , if Mp = (0) for some p ∈ Spec A, then there is some
s∈
/ p so that Ms = (0) and sM = (0).
f)p = (0) iff the stalk of M
Observe that Mp = (0) iff (M f at p is (0). Moreover, Ms = (0) iff M
fs = (0) iff
f  Xs vanishes. So, Lemma 3.41 says that if the stalk of M
M f vanishes punctually at p ∈ Spec A, then M f
vanishes on some open subset, containing p, of Spec A.

Proposition 3.45 If A is a commutative ring and M is a f.g. A-module, assume one of

(i) M is projective, or
(ii) A is noetherian and Mp is free over Ap for some p ∈ Spec A.
Then
St
(a) There exist σ1 , . . . , σt ∈ A so that Mσj is free over Aσj and X = Spec A = j=1 Xσj , or
(b) There is some σ ∈ A with p ∈ Xσ so that Mσ is free over Aσ .

Proof . We can write

0 −→ K −→ F −→ M −→ 0,
for any f.g. module, M , with F f.g. and free. If M is projective, then the sequence splits. Therefore, K
(being an image of F ) is f.g., and so, M is f.p.
In (ii), the ring A is noetherian and M is f.g, which implies that M is f.p., here, too. Thus, we will
assume that M is f.p. If we prove the (b) statement, then as a f.p. projective is locally free everywhere, the
(b) conclusion holds everywhere on Spec A. As X = Spec A is quasi-compact, we only need finitely many
opens to cover X. Therefore, we only need prove (b).
There exists a free module and a map, θ : F → M , so that at p, we have Fp ∼
= Mp . The sequence

0 −→ Ker θ −→ F −→ M −→ Coker θ −→ 0 is exact.

200 CHAPTER 3. COMMUTATIVE RINGS

Now, Coker θ is f.g. and (Coker θ)p = (0). So, there is some s ∈ A with (Coker θ)s = (0). If we restrict to
Xs ∼
= Spec As , we get

0 −→ Ker θ −→ F −→ M −→ Coker θ −→ 0 is exact on Xs .

By Proposition 2.41, as M is f.p. and F is f.g., we see that Ker θ is f.g. But, (Ker θ)p = (0), and by the
lemma, again, (Ker θ)t = (0), for some t ∈ A. If we let σ = st, then Xσ = Xs ∩ Xt , and on Xσ , we have an
isomorphism Fσ −g→ Mσ .
Given an A-module, M , we can make the OX -module, M f. This is a sheaf of OX -modules. There exist
index sets, I and J, so that
A(J) −→ A(I) −→ M −→ 0, is exact.
`
(Here, A(I) is an abbreviation for the coproduct I A.) So, we get
(J) (I) f −→ 0,
OX −→ OX −→ M

an exact sequence of sheaves. Now, M is free iff M f∼ (I)

= OX , for some I. We say that an OX -module, F, is
f
locally-free iff for every p ∈ Spec A, the module Fp is a free OX,p -module. Our proposition says: If F = M
and F is f.p. then F is projective3 iff F is locally-free. One can characterize the OX -modules, F, that are
of the form M f for some module, M ; these are called quasi-coherent OX -modules.

We proved that if Fp is a free module of finite rank and if A is noetherian and F is quasi-coherent, then
n
there is some open set, Xσ , with p ∈ Xσ , so that F  Xσ = OX  Xσ . Actually, we only used f.p., so the
statement also holds if F is projective (A not necessarily noetherian) and then it holds everywhere on small
opens, U , so that
n(U )
F  U = OX  U.

f) = rk(F), a function from Spec A to Z,

Let’s assume that M is projective and f.g. over A. Define rk(M
by
(rk F)(p) = rk(Fp ).
We showed that this function is locally constant on Spec A, i.e., rk F is a continuous function from Spec A
to Z, where Z has the discrete topology. Hence, if Spec A is connected, then the rank is a constant.

Proposition 3.46 Suppose M is a f.g. projective A-module (so, M is f.p.), and let F = M f on X = Spec A.
Then, the function rk(F) takes on only finitely many values, n1 , . . . , nt (in Z) and there exist ideals A1 , . . . , At
of A, each a commutative ring with unity, so that
Qt
(a) A = j=1 Aj ; so 1 = e1 +· · ·+et , with the ej ’s being orthogonal idempotents (which means that e2i = ei
and ei ej = 0 for i 6= j) and Aj = Aej .
St
(b) If Xej is the usual open corresponding to the element ej , then X = · j=1 Xej .
`t
(c) If Mj = Aj M , then M = j=1 Mj and each Mj is A and Aj -projective.

(d) Supp Mj = Xej , and rk(Mj ) on Xej is the constant nj .

The following lemma is needed:

S
Lemma 3.47 If X = Spec A and X = X1 · X2 is a disconnection, then there exist e1 , e2 ∈ A so that
Xj = Xej and 1 = e1 + e2 ; e21 = e1 ; e22 = e2 ; e1 e2 = 0.
3 In f.
the full subcategory of the OX -modules consisting of those of the form M
3.4. FIRST APPLICATIONS OF FRACTION RINGS 201

Proof . (DX)
Proof of Proposition 3.46. Let Xn = rk(F)−1 ({n}) for every n ≥ 0. Each Xn is an open and closed subset
of X, by continuity. The Xn cover X and by quasi-compactness only finitely many are necessary. Yet, they
are mutually disjoint. It follows that rk(F) = n1 , . . . , nt and rk(F)  Xj = nj . (Here, Xj = Xnj .) By
Lemma 3.47, there exist e1 , . . . , et , orthogonal idempotents with sum 1 and Xj = Xej , for j = 1, . . . , t. Let
Aj = Aej , this is an ideal, a ring and ej ∈ Aj is its unit element. Thus, parts (a) and (b) are proved.
`t
Write Mj = Aj M ; then, M = j=1 Mj , each Mj is a cofactor of M and, as M is A-projective, each Mj
is A-projective. The ring A acts on M via Aj ; therefore, Mj is Aj -projective.
`
Pick any q ∈ Spec Aj and write p = q i6=j Ai . This ideal, p, is a prime ideal of A. Note, ei with i 6= j
lies in p, but ej ∈
/ p, so p ∈ Xj . Since ei ej = 0, we also have ei ej = 0 in Ap . Yet, ej ∈
/ p, so ej is a unit in
Ap ; it follows that ei = 0 in Ap for all i 6= j. Then, we have
a a a
Mp = (Mi )p = (Ai M )p = (Aei M )p = (Mj )p .
i i i

The reader should check that (Mj )p = (Mj )q . Since p ∈ Xj , we deduce that (rk Mj )(q) = (rk M )(p) = nj ,
so, (rk Mj )(q) = nj . As ei = 0 iff i 6= j in Ap , we get Supp(Mj ) = Xej = Xj .
The simplest case, therefore, is: the A-module M is f.g., projective and rk M ≡ 1 on X = Spec A. We
say that M is an invertible module or a line bundle if we wish to view it geometrically.
Note: If M and M 0 are invertible, then M ⊗A M 0 is again a rank 1 projective A-module because
(M ⊗A M 0 )p = Mp ⊗Ap Mp0 . Thus, these modules form a semigroup under ⊗A and A (the free module) is
the unit element. Do they form a group?

Proposition 3.48 If A is a commutative ring and M is a f.g. A-module, then M is rank 1 projective
iff there is another module, M 0 , so that M ⊗A M 0 ∼
= A. When the latter condition holds, we can take
M 0 = M D = HomA (M, A).

Proof . (=⇒) The module M is rank 1 projective and as it is projective, it is f.p. Look at M ⊗A M D . There
exists a module map,
M ⊗A M D −→ A,
namely, the linear map induced by the bilinear map (m, f ) 7→ f (m). Localize at each p. We get

Mp ⊗Ap MpD −→ Ap ,

and MpD = HomA (M, A)p − g ∼ Ap , by

→ HomAp (Mp , Ap ), as M is f.p. and Ap is flat over A. But, Mp =
D
hypothesis and the reader should check that Mp ⊗Ap Mp −→ Ap is an isomorphism. As this holds for every
p ∈ Spec A, the map M ⊗A M D −→ A is an isomorphism.
(⇐=) Now, we have some A-module, M 0 , and M ⊗A M 0 ∼
= A. We can write

0 −→ K −→ F −→ M −→ 0,
`
for some f.g. free module, F . Look at the last three terms in this sequence, and write F = finite A:
a
A −→ M −→ 0.
finite

If we tensor with M 0 , we get a

M 0 −→ M ⊗A M 0 ∼
= A −→ 0.
finite
202 CHAPTER 3. COMMUTATIVE RINGS

`
But A is free, so the sequence splits and there is a map A −→ finite M 0 . Now, tensor with M . We get
a
A −→ M −→ 0,
finite
`
and there is a splitting map M −→ finite A. Thus, M is a cofactor of a free and f.g. module, so, M is f.g.
and projective, and hence, f.p. Now look at

M ⊗A M 0 ∼
=A

and localize at p. We get

Mp ⊗Ap Mp0 ∼
= Ap ,
and if we reduce mod pe , we get

Mp /pe Mp ⊗κ(Ap ) Mp0 /pe Mp0 ∼

= κ(Ap ). (†)

All the modules in (†) are vector spaces and, by counting dimensions, we get

dimκ(Ap ) Mp /pe Mp = 1.

Since Mp is a free Ap -module, by Nakayama, we get rk(Mp ) = 1. Lastly,

M0 ∼
= A ⊗A M 0 ∼
= (M D ⊗A M ) ⊗A M 0 ∼
= M D ⊗A (M ⊗A M 0 ) ∼
= M D ⊗A A ∼
= M D.

Therefore, M 0 ∼
= M D.
The group of (isomorphism classes) of the rank 1 projectives, M , is called the Picard group of A, denoted
Pic(A).

Corollary 3.49 If k is a field or a PID, then Pic(A) = (0).

The group Pic(A) is a subtle invariant of a ring (generally hard to compute).

3.5. INTEGRAL DEPENDENCE 203

3.5 Integral Dependence

The notion of integral dependence first arose in number theory; later, thanks to Zariski, it found application
in algebraic geometry. Throughout this section as throughout this chapter, all rings are commutative with
unity.

Definition 3.5 Suppose ϕ : A → B is a ring homomorphism and b ∈ B. The element, b, is integral over A
iff there is a non-trivial monic polynomial, f (X) ∈ A[X], so that f (b) = 0. (Here, f (b) is bn + ϕ(a1 )bn−1 +
· · · + ϕ(an−1 )b + ϕ(an ) if f (X) is X n + a1 X n−1 + · · · + an−1 X + an .) The A-algebra B is integral over A iff
all its elements are integral over A and, in this case, ϕ is an integral morphism.

Clearly, each ring surjection is an integral morphism, but this is not what is really intended. Each
homomorphism, ϕ, as above factors into a surjection whose image, A, e is a subring of B followed by the
inclusion Ae ,→ B. It is for inclusions that integrality is a real question and is decisive for certain situations.
As usual, there are a number of equivalent ways to say integrality and their equivalence is quite useful
technically.

Proposition 3.50 Suppose ϕ : A → B is a ring homomorphism and b ∈ B. Then the following are equiva-
lent conditions:
(1) b is integral over A
(2) The A-algebra A[b] (a sub-A-algebra of B) is finitely generated as A-module.
e of B which is a finitely generated A-module and b ∈ B.
(3) There exists a sub-A-algebra, B, e

e of B so that α) bB
(4) There exists a finitely generated sub-A-module, B, e⊆B
e and β) A[b]∩Ann(
e e = (0).
B)

Proof . (1) =⇒ (2). We have the equation of integral dependence

bn + a1 bn−1 + · · · + an−1 b + an = 0

(here, we drop ϕ(aj ) and just denote it by aj ). Hence, bn ∈ A-module generated by 1, b, . . . , bn−1 . But
then, bn+1 is also in this A-module, etc. Thus, A[b] is the finitely generated A-module given by generators
1, b, . . . , bn−1 .
(2) =⇒ (3). We take B e = A[b].
e of (3) for the module of (4). Of course, α) holds as B
(3) =⇒ (4). We use our subalgebra, B, e is a ring
e
by (3) and β) is clear as a ∈ B.
e as A-module. Since bB
(4) =⇒ (1). Let ξ1 , . . . , ξt be generators for B e ⊆ B,
e we see that for each i, the
element bξi is an A-linear combination of the ξ’s:
t
X
bξi = zij ξj .
j=1

That is,
t
X
(δij b − zij )ξj = 0, for i = 1, 2, . . . , t. (∗)
j=1

e Upon expanding
Write ∆ for det(δij b−zij ), then by linear algebra we get ∆ξj = 0 for all j, i.e., ∆ ∈ Ann(B).
e
∆ by minors, we find that ∆ ∈ A[b]; so, β) implies ∆ = 0. But the expansion by minors shows ∆ has the
form bt + lower powers of b and this gives (1).
204 CHAPTER 3. COMMUTATIVE RINGS

There are many corollaries, but first notice that if A is noetherian, we may replace (3) by the weaker
condition
0
e of B and A[b] ⊆ B.
(3 ) There is a finitely generated sub-A-module, B, e
Let’s write
IntA (B) = {b ∈ B | b is integral over A}
and refer to IntA (B) as the integral closure of A in B (we assume ϕ is given a priori ).
Corollary 3.51 Say A and B are given as above and b1 , . . . , bt are elements of B. Then, b1 , . . . bt ∈ IntA (B)
iff the A-algebra A[b1 , . . . , bt ] is a finitely generated A-module. In particular, IntA (B) is a A-algebra.
Proof . (⇐=). Here, A[bj ] ⊆ A[b1 , . . . , bt ] and we apply (3) of Proposition 3.50 to get bj ∈ IntA (B).
(=⇒). We have the chain of A-algebras
e
A[b1 , . . . , bt ] ⊇ · · · ⊇ A[b1 ] ⊇ A
each a finite module over its predecessor by (2) of Proposition 3.50. Then, it is clear that A[b1 , . . . , bt ] is a
finite A-module. Lastly, if x, y ∈ IntA (B), we see that x ± y and xy lie in A[x, y]. By the above, the latter
is a finite A-module and (3) of Proposition 3.50 completes the proof.
Corollary 3.52 (Transitivity of Integral Dependence) Suppose that B is an A-algebra and C is a B-algebra.
Then,
IntIntA (B) (C) = IntA (C).
In particular, if C is integral over B and B is integral over A, then C is integral over A.
Proof . If ξ ∈ C and ξ is integral over A, then ξ is a fortiori integral over the “bigger’ ring IntA (B), and so
IntA (C) ⊆ IntIntA (B) (C).
Now, if ξ is integral over IntA (B), then ξ is integral over A[b1 , . . . , bt ] where the bi are coefficients in the
polynomial of integral dependence for ξ. Each bi is in IntA (B), so Corollary 3.51 shows A[b1 , . . . , bt ] is a
finite A-module. Yet A[b1 , . . . , bt ][ξ] is a finite A[b1 , . . . , bt ]-module by integrality of ξ. Therefore ξ is in the
finitely generated A-module A[b1 , . . . , bt , ξ] which is an A-algebra and we apply (3) of Proposition 3.50. The
element ξ is then in IntA (C), as required.
When C is integral over B and B is integral over A, we get C = IntB (C) and B = IntA (B); so C = IntA (C)
by the above.
When IntA (B) is A e (image of A in B) itself, we say A is integrally closed in B. (Usually, for this
terminology, one assume ϕ is an inclusion A ,→ B.) If S is the set of non-zero divisors of A, then S is a
multiplicative set and S −1 A is the total fraction ring of A. We denote it by Frac(A). When A in integrally
closed in Frac(A), we call A a normal ring or an integrally closed ring. For example
Proposition 3.53 Every unique factorization domain is a normal ring.
Proof . We suppose A is a UFD, write K = Frac(A) (in this case K is a field as A is a domain). Let ξ = α/β
be integral over A, and put α/β in lowest terms. Then,
ξ n + a1 ξ n−1 + · · · + an−1 ξ + an = 0, the aj ∈ A.
Insert the value of ξ (= α/β) and clear denominators. We get
αn + a1 αn−1 β + · · · + an−1 αβ n−1 + an β n = 0.
If p is a prime element of A and p divides β, our equation shows p | αn ; i.e., p | α. This is a contradiction
on lowest terms and so no p divides β. This means β is a unit; so, ξ ∈ A.
 3.5. INTEGRAL DEPENDENCE 205

Proposition 3.54 If A is a normal domain and S is any multiplicative subset of A, then S −1 A is also a
normal domain.
Proof . We know Frac(A) = Frac(S −1 A). So, choose ξ ∈ Frac(A) integral over S −1 A. Then,
a1 an−1 an
ξ n + ξ n−1 + · · · + ξ+ = 0.
s1 sn−1 sn
Q
We can write this with common denominator s = sj , then
a1 an−1 an
ξ n + ξ n−1 + · · · + ξ+ = 0.
s s s
Upon multiplication by sn , we find (sξ) is integral over A. By hypothesis, sξ ∈ A; so, ξ ∈ S −1 A.
Two easy facts are useful to know. Their proof are easy and will be left to the reader (DX):
Fact A. If B is integral over A and I is any ideal of B, then B/I is integral over A/ϕ−1 (I).
Fact B . If B is integral over A and S is a multiplicative set in A with S ∩ Ker ϕ = ∅, then S −1 B is
integral over S −1 A.

 However, observe that if A is a normal ring and A is one of its ideals, then A/A need not be normal .
A standard example is a “singular curve”.

Here, we take C[X, Y ] which is a normal ring as it is a UFD. Let A = (Y 2 − X 3 ), then C[X, Y ]/A is not
normal (though it is a domain (DX)). For, the element Y /X (in FracA/A) is integral over A/A as its square
is X, yet it is not itself in A/A (DX). The interpretation is this: Y 2 − X 3 = 0 describes a curve in the plane
over C and Y /X defines by restriction a function holomorphic on the curve except at (0, 0). But, Y /X is
bounded near (0, 0) on the curve, so it ought to be extendable to a holomorphic (and algebraic) function.
Yet, the set of such (near (0, 0)) is just (AA)p , where p = {f ∈ A/A | f (0, 0) = 0}. Of course, Y /X ∈/ (AA)p .
The trouble is that Y 2 = X 3 has a “singular point” at (0, 0), it is not a complex manifold there (but it is
everywhere else). This shows up in the fact that (AA)p is not normal.
When A is a noetherian ring, we can be more precise, but we need some of the material (on primary
decomposition from Sections 3.6 and 3.7. The two main things necessary are the statement
If V is a submodule of the A-module, M , then V = (0) iff Vp = (0) for all p ∈ Ass(M ) (see Section 3.6,
Corollary 3.102 of Theorem 3.99); and Krull’s Principal Ideal Theorem (Section 3.7, Theorem 3.120).
You should skip the proof of Lemma 3.55, Theorem 3.56 and Corollary 3.57 until you read this later
material; pick up the thread in Theorem 3.58, below.
Write, for a ring A,
Pass(A) = {p | p ∈ Ass(A/(a)), for some non-zero divisor , a, of A}.

Lemma 3.55 If A is a reduced Noetherian ring, then an element ξ ∈ Frac(A) is actually in A if and only
if for every, p ∈ Pass(A), the image of ξ ∈ Frac(A)p is in Ap .
Proof . If ξ ∈ A, then of course its image in Frac(A)p lies in Ap for all p. So, assume
\
ξ ∈ {Ap | p ∈ Pass(A)}

(here, of course, we mean the images of ξ in Frac(A) are in Ap ). We write ξ = α/β, where β is a non-zero
divisor and suppose that ξ ∈ / A. Then, α is not in (β), so V = Aα ⊆ A/(β) is non-zero. By the statement
italicized above, there is a p ∈ Ass(A/(β)) with (Aα)p 6= (0). This means α/1 ∈
/ (β)p ; that is, ξ = α/β ∈
/ Ap .
Yet, p ∈ Pass(A), a contradiction.
Here is a characterization of normality for Noetherian domains:
206 CHAPTER 3. COMMUTATIVE RINGS

Theorem 3.56 Suppose that A is a noetherian domain, then the following conditions are equivalent:

(1) A is normal

(2) For every p ∈ Pass(A), the ideal pe is a principal ideal of Ap

(3) (a) Every p ∈ Pass(A) has height 1 and

(b) For all height one primes, p, of A, the ring Ap is a PID.

Proof . We first prove (2) ⇐⇒ (3). Suppose p is any prime ideal of A. If p is a principal ideal of Ap . it is an
isolated prime of itself and Krull’s Principal Ideal Theorem shows that ht(p) = 1. So by (2),

Pass(A) ⊆ {p | ht(p) = 1}.

But, ht(p) = 1 implies p is an isolated prime ideal of any of its non-zero elements and, since A is a domain,
this shows p ∈ Pass(A). We’ve proved that (2) implies that

Pass(A) = {p | ht(p) = 1}. (∗)

This shows that (2) implies (3a) and for all height one primes, p, the maximal ideal, p, of Ap is principal.
We’ll now show that Ap is a PID. Pick an ideal, A, of Ap and write m = pe . As m is the maximal ideal of Ap ,
we have A ⊆ m, and as m is principal we may assume (0) < A < m. Now mn is principal for all n ≥ 0 with
generator π n , where π generates m; we’ll show A = mn for some n. Now, were A ⊆ mn for all n, the Krull
Intersection Theorem (Theorem 3.113) would show A = (0), contrary to assumption. So, pick n minimal so
that A ⊆ mn . Then, every ξ ∈ A has the form aπ n , and for at least one ξ, the element a is a unit (else a ∈ m
implies a = bπ and all ξ have shape bπ n+1 ). But then,

A ⊇ (ξ) = (π n ) = mn ⊇ A

and A is indeed principal. Therefore, (2) implies (3a) and (3b). It is clear that (3a) and (3b) imply (3).
We come then to the main point of our theorem, that (1) is equivalent to both parts of (3). Observe that
the argument in the very early part of the proof shows that we always have

{p | ht(p) = 1} ⊆ Pass(A).

(3) =⇒ (1). By (3a), Pass(A) = {p | ht(p) = 1}; so

\ \
{Ap | ht(p) = 1} = {Ap | p ∈ Pass(A)} (∗∗)

By Lemma 3.55, the right hand side of (∗∗) is A and by (3b) each Ap is a normal domain (Proposition 3.53).
Hence, A, as an intersection of noremal domains in Frac(A), is itself normal.
(1) =⇒ (3). Here, we will actually show (1) ⇐⇒ (2), then we will be done. Pick p ∈ Pass(A), say
p ∈ Ass(A/(a)). Then, there exists an element ξ ∈ A so that p is the annihilator of ξ (mod (a)). We need to
prove pe is principal, so we may replace A by Ap and p by pe . Thus, our situation is that A is local and p is
its maximal ideal. Write

A = {η ∈ Frac(A) | ηp ⊆ A} = (p −→ A) (in Frac(A)).

Of course, Ap is an ideal of A and A ⊆ A shows that p = Ap ⊆ Ap. Hence, there are only two possibilities:
Ap = p or Ap = A. I claim that the first cannot hold. If it did, condition (4) of Proposition 3.50 applied
to each η of A (with Be = p and B = Frac(A)) would show that all these η are integral over A. By (1), the
η lies in A; so A = A. Now p annihilates the element ξ (mod (a)) and ξ ∈ / (a); that is, ξp = pξ ⊆ (a); so
3.5. INTEGRAL DEPENDENCE 207

(ξ/a)p ⊆ A. But then ξ/a ∈ A, i.e., ξ/a ∈ A. The last assertion is that ξ ∈ (a), contrary to the choice of ξ.
We deduce, therefore, that Ap = A. Now, the map

A ⊗A p −→ Ap
P
is an isomorphism Pbecause if i qi ⊗ pi goes to zero in A, then using a common denominator, say d, for the
qi , we find (1/d) i αi ⊗ pi is 0, too. Clearly, A ⊗A p −→ Ap is surjective. Proposition 3.48 now shows p is
a free rank one A-module (remember A is local), i.e., a principal ideal.

Corollary 3.57 If A is a Noetherian normal domain, then

\
A = {Ap | ht(p) = 1}.

Proof . Theorem 3.56, condition (3a) shows

Pass(A) = {p | ht(p) = 1}

and we then apply Lemma 3.55.

There are relations between the prime ideals of A and B when B is integral over A. These are expressed
in the three Cohen-Seidenberg Theorems. Here is the first of them:

Theorem 3.58 (Lying over Theorem; Cohen-Seidenberg, I) If B is integral over A and p is any prime
ideal of A, then there is a prime ideal, Q, of B lying over p (that is, ϕ−1 (Q) = p, where ϕ : A → B).

Proof . Of course, we may and do assume A ⊆ B. Let S be the collection of all ideals, B, of B with
B ∩ A ⊆ p; partially order S by inclusion. As S =
6 ∅ ((0) ∈ S) and clealry inductive, Zorn’s Lemma furnishes
a maximal element, say Q, in S. We must show both Q ∩ A = p and Q is a prime ideal.
Were Q ∩ A < p, we could find ξ ∈ p with ξ ∈ / Q ∩ A. Write Q e for the ideal Q + Bξ; as ξ ∈
/ Q, we get
e > Q. So, Q
Q e ∈
/ S and thus Q e ∩ A 6⊆ p. Therefore, there is some η ∈ Q
e ∩ A (thus η ∈ A) yet η ∈/ p. Now η
e
is in Q, so looks like q + bξ, for some b ∈ B. Note that η − bξ = q ∈ Q.
The element b is integral over A:

bn + a1 bn−1 + · · · + an−1 b + an = 0, all aj ∈ A.

If we multiply by ξ n , we find

(bξ)n + a1 ξ(bξ)n−1 + · · · + an−1 ξ n−1 (bξ) + an ξ n = 0. (∗)

View (∗) in B/Q; there η = bξ, and so,

(η)n + a1 ξ(η)n−1 + · · · + an−1 ξ n−1 η + an ξ n = 0 in A/Q. (∗∗)

But now, all elements on the left hand side of (∗∗) when read in B actually lie in A; so the left hand side of
(∗∗) is in Q ∩ A. We get
η n + a1 ξη n−1 + · · · + an−1 ξ n−1 η + an ξ n ∈ p.
Remembering that ξ ∈ p, we find η ∈ p, a contradiction. This shows Q ∩ A = p.
To show Q is a prime ideal, write S for the multiplicatice set A − p; S is a multiplicative subset of B.
Of course, Q ∩ S = ∅. Suppose Q were not maximal among ideals of B whose intersection with S is empty.
We’d find Qe > Q and Q ∩ S = ∅. But then Q ∩ A = p and so Q e lies in S where Q is maximal contradicting
e
Q > Q. Therefore, Q is maximal among ideals of B with Q ∩ S = ∅. Now, Proposition 3.8 (the implication
(6) =⇒ (1)) shows Q is prime.
208 CHAPTER 3. COMMUTATIVE RINGS

Theorem 3.59 (Going-up Theorem; Cohen-Seidenberg, II) Suppose B is integral over A and p ⊆ q are
two prime ideals of A. If P is a prime ideal of B lying over p, there exists a prime ideal, Q, of B lying over
q with P ⊆ Q.

Proof . This is just a corollary of the lying over theorem. For once again, we may assume A ⊆ B and we
consider A/p and B/P. As P ∩ A = p and B is integral over A, we find B/P is integral over A/p and apply
Cohen-Seidenberg I to A/p and B/P, using q as our ideal of A/p. There is Q, a prime of B/P, over q and
the pull-back of Q in B is what we want.

Corollary 3.60 If A and B are integral domains and B is integral over A, then A is a field iff B is a field.

Proof . Suppose A is a field and ξ 6= 0 is in B. The element ξ is integral over A; so

ξ n + a1 ξ n−1 + · · · + an−1 ξ + an = 0

for some a1 , . . . , an ∈ A. Of course, we may assume that an 6= 0. Then

ξ(ξ n−1 + a1 ξ n−2 + · · · + an−1 ) = −an ;

and, as A is field, the element

1 n−1
− (ξ + a1 ξ n−2 + · · · + an−1 )
an
lies in B and is the inverse of ξ.
If B is a field and A is not, there are prime ideals (0) < q of A. The going-up theorem gives us prime
ideals (0) and Q of B lying over (0) and A—but, B is a field; contradiction.
(We may also argue directly as in the first implication of the proof: Given ξ ∈ A, the element ξ is in B
and B is a field. So, 1/ξ ∈B; thus 1/ξ is integral over A. We have
 n  n−1  
1 1 1
+ a1 + · · · + an−1 + an = 0.
ξ ξ ξ

Multiply through by ξ n ; we find

1 = −ξ(a1 + · · · + an−1 ξ n−2 + an ξ n−1 );

so ξ has an inverse in A.)

Corollary 3.61 If B is integral over A and P ∈ Spec B lies over p ∈ Spec A, then p is maximal iff P is
maximal.

This is merely a restatement of Corollary 3.60. A more important remark is the incomparability of two
primes lying over a fixed prime:

Proposition 3.62 Say B is integral over A and P, Q are two primes of B lying over the same prime, p,
of A. Then P and Q are incomparable; that is we cannot have either P ⊆ Q or Q ⊆ P without P = Q.

Proof . Assume P < Q and reduce A mod p and B mod P. Then we may assume A and B are domains and
we have to prove no non-zero prime contracts to the zero ideal of A. In fact, we prove: If A, B are domains
with B integral over A and if B is a non-zero ideal of B, then B contracts to a non-zero ideal of A .
Choose b ∈ B with b 6= 0. Then we find

bn + a1 bn−1 + · · · + an−1 b + an = 0.
3.5. INTEGRAL DEPENDENCE 209

and we may assume an 6= 0 (else we could divide out b and lower the degree, n; etc.) But then an ∈ B ∩ A;
so B ∩ A 6= (0), as required.
Now we come to the circle of ideas around the third (and deepest) of the Cohen-Seidenberg Theorems, the
so-called “Going–Down Theorem”. This is a study of prime ideals in integral extensions where the bottom
ring is a normal ring. For the proof of the theorem, we need some simple ideas from Galois theory) most
of which are already familiar) which are covered in full in Chapter 4, sections one through four. Readers
are urged to skip the proofs of Propositions 3.63 and 3.64 and Theorem 3.65, and come back to these after
having read Sections 4.2–4.5 of Chapter 4. Once again, one can pick up the thread of our discussion in
Proposition 3.66. Nonetherless the statements of all results below are clear.
Recall that if k is a field and B is a k-algebra, and element ξ, of B is algebraic over B iff it satisfies a
(non-zero) polynomial f (X) ∈ k[X]. Of course, the set of all polynomials, g(X), with g(ξ) = 0 is a principal
ideal of k[X] and the monic polynomial generating this ideal is the minimal polynomial of ξ over k. If B
has zero divisors, the minimal polynomial of ξ over k will not, in general, be irreducible in k[X]. Even if no
non-zero element of k becomes a zero divisor in B, still the minimal polynomial might be reducible.4 But
when B is at least a domain the minimal polyomial will be irreducible. We also want to consider in k an
integral domain, A, with k = Frac(A).
So, let ξ ∈ B be integral over A, assume B is a domain. Then we can factor the minimal polynomial
f (X), for ξ over k = Frac(A) in some big field over B (Section 4.4 of Chapter 4) and it will have exactly
n roots where n = deg(f ). Write these as ξ = ξ1 , ξ2 , . . . , ξn . By Section 4.3, Chapter 4, each ni is repated
pe times where p = char(k) and e ≥ 0; pe is the degree of inseparability of ξ over k. Moreover, there is
an automorphism fixing the elements of k taking each ξ to ξ; so each ξi satisfies the equation of integral
dependence which ξ satisfies (Section 4.4, Chapter 4 again). Now when we write f (X) as a product of the
linear factors (X − ξi ) we get
n
Y n
X
f (X) = (X − ξi ) = σj (ξ1 , . . . , ξn )(−1)j X n−j ,
i=1 j=0

here the σj are the elementary symmetric functions of the ξi , given as

σ0 (ξ1 , . . . , ξn ) = 1
σ1 (ξ1 , . . . , ξn ) = ξ1 + · · · + ξn
X
σ2 (ξ1 , . . . , ξn ) = ξi ξj
i<j
..
.
X
σr (ξ1 , . . . , ξn ) = ξi1 ξi2 · · · ξir
i1 <i2 <···<ir
..
.
σn (ξ1 , . . . , ξn ) = ξ1 ξ2 · · · ξn .

Thus, when ξ is integral over A, so are all the ξi and all the elements σj (ξ1 , . . . , ξn ), for j = 1, 2, . . . , n.
But each σj (ξ1 , . . . , ξn ) is in k, therefore each σj is in Autk (A). The symmetric functions σ1 and σn have
special designation—they are the trace and norm of ξ over k, respectively. This argument gives the first
two statements of

Proposition 3.63 If A is a domain and k = Frac(A), write B for an overing of A and K for Frac(B).
Then,
4A standard example is the “ring of dual numbers over k”, namely, k[X]/(X 2 ). The minimal polynomial of X is X 2 .
 210 CHAPTER 3. COMMUTATIVE RINGS

(1) When K is a field and ξ ∈ K is integral over A, all the coefficients of the minimal polynomial for ξ over
k are integral over A (so the norm and trace of ξ are integral over A).
(2) If A is a normal domain and K is a field, the minimal k-polynomial for an element ξ ∈ K which is
integral over A already lies in A[X] and is an equation of integral dependence for ξ.
(3) If A is a normal domain and f (X), g(X) are two monic polynomials in k[X] so that f (X)g(X) is in
A[X], then each of f (X) and g(X) is already in A[X].
(4) If A is a normal domain and B is an overring of Frac(A), and if ξ ∈ B is integral over A, then the
minimal k-polynomial of ξ is already in A[X] and is an integral dependence relation for ξ. That is, (2) holds
even K is not a field (B is an integral domain), provided K ⊆ k.
(5) If A is a normal domain and B is an overring of A, with ξ ∈ B integral over A, and if non non-zero
element of A becomes a zero divisor in B, then again the minimal k-polynomial for ξ is already in
A[X] and is an integral dependence.
Q Q
Proof . (1) and (2) are already proved; consider (3). Write f (X) = i (X − ξ) and g(X) = j (X − ηj ) in
some big overfield. Now f (X)g(X) is a monic polynomial in A[X] all xii and ηj satisfy it. But such a monic
polynomial is an integral dependence relation; so, all ξi are integral over A and all the ηj are integral over
A. By the argument for (1) each of the σi (ξ1 , . . . , ξt ) and σj (η1 , . . . , ηr ) are integral over A; hence they are
in A by the normality of A. But, these are (up to sign) the coefficients of f (X) and g(X) and (3) is proved.
(4) B is a k-algebra, so ξ has a minimal polynomial, f (X) ∈ k[X]. Now ξ is also integral over A,
therefore there is a monic polynomial, h(X) ∈ A[X], with h(ξ) = 0. As f generates the principal k[X] ideal
of polynomials vanishing at ξ, there is a g(X) ∈ k[X] with f (X)g(X) = h(X) and clearly g(X) is monic.
Then, (3) shows f (X) ∈ A[X] and is an equation of integral dependence.
(5) Here, if S is the multiplicative set of nonzero elements of A, then each s ∈ S is a non-zero divisor of
B and so k = Frac(A) ⊆ S −1 B ⊆ Frac(B). We can then apply (4) to S −1 B and conclude (5).

Remark: Notice that the statement of (3) contains the essential ideal of Gauss’ classical proof that if A is
a UFD so is A[X].
The hypothesis of (5) follows from a perhaps more easily checked condition:

Proposition 3.64 If B is an A-algebra and B is flat over A, then no non-zero divisor of A becomes a
non-trivial zero divisor in B.

Proof . To say ξ is a non-zero divisor is to say

ξ
0 −→ A −→ A −→ A/Aξ −→ 0

is exact. Now, tensor this exact sequence with B over A and use flatness to get
ξ
0 −→ B −→ B −→ B/Bξ −→ 0

is exact.

Theorem 3.65 (Going-down Theorem; Cohen-Seidenberg, III) Suppose A is a normal domain and B is
an overring of A. Assume either
(1) B is integral over A and
(2) No non-zero element of A becomes a zero divisor of B
or
3.5. INTEGRAL DEPENDENCE 211

(10 ) B is integral over A and

(20 ) B is flat over A.
Then, given prime ideals p ⊆ q of A and a prime ideal Q of B over q, there is a prime ideal, P, of B, over
p so that P ⊆ Q.
e is the image of A in B and (20 ) holds, then B is flat over A
Proof . If A e and so, by Proposition 3.64, (2)
holds. Therefore, we will assume (1) and (2).
The key to the proof is to find an apt multiplicative set, S, of B and to consider S −1 B. Take S to be
the collection of products, aα, where a ∈ A − p and α ∈ B − Q. Of course, S is closed under multiplication
and 1 ∈ S; further 0 ∈/ S else a, an element of A, would be a zero divisor of B contrary to (2). Observe, by
taking a = 1 or α = 1, we find A − p ⊆ S and B − Q ⊆ S.
I claim the extended ideal, pe , of p in S −1 B is not the unit ideal. Suppose, for the momemt, the claim is
proved; we finish the proof as follows: The ideal pe is contained in some maximal ideal, M, of S −1 B, and so
Mc is a prime ideal of B. (As each ideal of S −1 B is extended, M is Ae and so Mce = Aece = Ae = M 6= S −1 B;
therefore, Me 6= B.) Since M 6= S −1 B, the ideal Mc cannot intersect S and B − Q ⊆ S shows that Mc ⊆ Q.
Now consider Mc ∩ A, it is a prime ideal of A and cannot intersect S. Again, A − p ⊆ S implies Mc ∩ A ⊆ p.
Yet
p ⊆ pB ∩ A ⊆ pec ∩ A ⊆ Mc ∩ A,
therefore Mc ∩ A = p and we can set P = Mc .
P
We are therefore down to proving our claim, that is that pB ∩ S = ∅. Pick ξ ∈ pB, write ξ = bi pi with
e = A[b1 , . . . , bt ]; it is a f.g. A-module (as well as A-algebra) by the integrality of B
pi ∈ p and bi ∈ B. Let B
over A. We have ξ Be ⊆ pBe and if ξ1 , . . . , ξr form a set of A-module generators for B,e we find from ξξj ∈ pB e
the linear equations:
X r
ξξj = pij ξi , pij ∈ p.
i=1

Just as in the argument (4) ⇐⇒ (1) of Proposition 3.50, this leads to ∆ξi = 0 for i = 1, . . . , r, where
e = 0, yet 1 ∈ B;
∆ = det(δij ξ − pij ). Thus, ∆B e so ∆ = 0. By the minor expansion of ∆, we deduce the
integral dependence
h(ξ) = ξ r + π1 ξ n−1 + · · · + πr−1 ξ + πr = 0
and here all the πi ∈ p.
Say ξ is in S, then it has the form aα, with a ∈ A − p and α ∈ B − Q. By part (5) of Proposition 3.63,
the minimal polynomial, f (X) ∈ k[X], for ξ is already in A[X] and is an integral dependence for ξ. But,
also f (X) divides h(X) in k[X] as h(ξ) = 0; so

f (X)g(X) = h(X) in k[X]

and g(X) is monic. Apply part (3) of Proposition 3.63 and get that g(X) ∈ A[X], too. This means we can
reduce the coefficients of f, g, h mod p. The polynomial h(X) becomes h(X) = X r . But A/p is a domain
and h = f g; so f (X) = X p , that is

f (X) = X s + δ1 X s−1 + · · · + δs−1 X + δs ,

and all the δi lie in p.

Now ξ = aα and by (5) of Proposition 3.63 once again, we see that the k-minimal polynomial for α is
actually in A[X] and is an integral dependence for α. Write this polynomial, m(X), as

m(X) = X v + u1 X v−1 + · · · + uv−1 X + uv

212 CHAPTER 3. COMMUTATIVE RINGS

with each ui ∈ A. Now multiply m(X) by av , we get

av m(X) = (aX)v + au1 (aX)v−1 + · · · + av−1 uv−1 (aX) + av uv .

So, for the polynomial

fe(X) = X v + au1 X v−1 + · · · + av−1 uv−1 X + av uv

we find fe(ξ) = av m(α) = 0 and therefore f (X) divides fe(X) in k[X]: fe(X) = z(X)f (X). By (3) of
Proposition 3.63, we see z(X) is monic and in A[X], and v = deg(fe) ≥ deg(f ) = s. However, by the same
token if we divide f (X) by as , we get
 s  s−1  
X δ1 X δs−1 X δs
+ + ··· + +
a a a as−1 a as

giving us the k-polynomial

δ1 s−1 δs−1 δs
F (X) = X s + X + · · · + s−1 X + s .
a a a
We have F (α) = (1/as )f (ξ) = 0; so m | F in k[X]. Therefore,

s = deg(F ) ≥ deg(m) = v;

coupled with the above this shows s = v and Z(X) = 1. Therefore, f (X) = fe(X) so that

δj = aj uj , j = 1, 2, . . . , s.

Now δj ∈ p and, by choice of S, a ∈

/ p. Therefore, all the uj belong to p.

Finally, m(α) = 0; so,

αs + u1 αs−1 + · · · + us−1 α + us = 0.

This shows αs ∈ pB ⊆ qB ⊆ Q; whence α ∈ Q—a contradiction.

The Cohen-Seidenberg Theorems have geometric content. It turns out that for a commutative ring A
(over the complex numbers), Spec A can be made into a (generalized) complex space (perhaps of infinite
dimension); that is into a complex manifold with some singularities (perhaps). For us, the important point
is that Spec A is a topological space (see Section 3.3) and we’ll only draw topological content from the
Cohen-Seidenberg Theorems.

So, first say B is integral over A. The ring map ϕ : A → B gives a continuous map Spec B −→ Spec A,
namely: P 7→ ϕ−1 (P). The lying over theorem can now be expressed as:

If B is integral over A, the continuous map Spec B −→ Spec A is surjective.

Remark: We’ve used a Cohen-Seidenberg Theorem; so, we’ve assumed A −→ B is an injection in the
above.

The question of A −→ B being an injection and the “real” content of integrality can be teased apart as
follows:

Proposition 3.66 Say A −→ B is an injection. Then the continuous map Spec B −→ Spec A has dense
image. If A −→ B is surjective, then the continuous map Spec B −→ Spec A is a homeomorphism onto a
closed subset of Spec A.
3.5. INTEGRAL DEPENDENCE 213

Proof . Write ϕ for the homomorphism A −→ B and |ϕ| for the continuous map Spec B −→ Spec A. Pick
any p ∈ Spec A and any f ∈ / p (so that p ∈ Xf in Spec A). We must find q ∈ Spec B so that |ϕ|(q) ∈ Xf .
Now f is not nilpotent; so, as ϕ is injective, neither is ϕ(f ). But then there is a prime ideal, q, of B, and
ϕ(f ) ∈
/ q (cf. either Proposition 3.8 #(6) or remark #(4) after Proposition 3.11); that is f ∈ / |ϕ|(q), which
is what we needed.
Recall, from the discussion on the Zariski topology following Proposition 3.11, that the closed sets in
Spec A are all of the form V (A) for some ideal A, of A. Now there is the usual one-to-one correspondence
of ideals, B, of A which contain A and all ideals of A/A. If we take for A the kernel of ϕ, then the first
consequence is that p 7→ |ϕ|(p) is a continuous bijection of Spec B (= Spec A/A) and the closed set, V (A),
of Spec A. But, this is also a closed map, because for B, an ideal of B, the map |ϕ| takes V (B) onto
V (ϕ−1 (B)) ⊆ Spec A.
Proposition 3.67 If B is integral over A, where ϕ : A → B need not be injective, then the map |ϕ| from
Spec B to Spec A is a closed map. In fact, it is universally closed; that is, the map
|ϕC | : Spec(B ⊗A C) → Spec C is a closed map for every A-algebra, C.
Proof . Note that if B is integral over A, then B ⊗A C is integral over C. To see this, observe that a general
element of B ⊗A C is a sum of terms b ⊗ c with b ∈ B and c ∈ C. If b ⊗ c is integral over C so is any sum
of such terms. But, b ⊗ c = (b ⊗ 1)(1 ⊗ c) and 1 ⊗ c is in C (= A ⊗A C) so all we need check is that b ⊗ 1
is integral over C. Write the integral dependence for b over A, then tensor with 1 (as in b ⊗ 1) and get the
integral dependence of b ⊗ 1 over C.
This remark reduces us to proving the first statement. Now the map A −→ B factors as
e = A/A ,→ B,
A −→ A
so for the spaces Spec A, etc., we get
e −→ Spec A.
Spec B −→ Spec A
By Proposition 3.66, the second of thesex maps is closed, therefore we are reduced to the case where A −→ B
is injective. A closed set of Spec B is V (B) and we know by Fact A following Proposition 3.54 that B/B is
integral over A/(B ∩ A). The interpretation of Cohen–Seidenberg II shows that
Spec(B/B) −→ Spec(A/(B ∩ A)) is surjective. Coupled with the homeomorphisms
Spec(B/B) ∼
= V (B); Spec(A/(B ∩ A)) ∼
= V (B ∩ A),
this finishes the proof.
Let’s continue with these topological considerations a bit further. Take p ∈ Spec A, one wants to consider
{p} as Spec(?) for some A-algebra “?”. At first Ap seems reasonable, but Spec Ap consist of all the primes
contained in p. We can get rid of all these extraneous primes by factoring out by pe and forming
κ(p) = Ap /pe .
The A-algebra, κ(p), is a field; so, Spec κ(p) is one-point—it corresponds to p. Indeed, in the map
κ(p) −→ Spec A coming from the ring map
A −→ Ap −→ Ap /pe = κ(p),
the one point of Spec κ(p) goes to p in Spec A. If B is an A-algebra, then B ⊗A κ(p) is a κ(p)-algebra
isomorphic to Bp /pBp . The commutative diagram

BO / B ⊗A κ(p)
O

A / κ(p)
214 CHAPTER 3. COMMUTATIVE RINGS

shows that the elements of Spec(B ⊗A κ(p)) all go to p under the map Spec(B ⊗A κ(p) −→ Spec A. Therefore,
Spec(B ⊗A κ(p) is the fibre of the map Spec B −→ Spec A over {p}.

Proposition 3.68 Suppose B is a finitely generated A-algebra and is also integral over A. Then, each fibre
of the map Spec B −→ Spec A is finite.

Proof . The algebra B has the form A[b1 , . . . , bt ] and each bj is integral over A. Thus, B is a finitely-generated
A-module. So, each B ⊗A κ(p) is a finitely generated κ(p)-vector space and therefore has the D.C.C. By
Lemma 3.37, Spec(B ⊗A κ(p)) is a finite set.
We have more than stated: B is not only a finitely generated A-algebra it is a f.g. A-module. This
is stronger than the condition that all the fibres of |ϕ| : Spec B → Spec A be finite. Indeed, consider the
inclusion Z ,→ Q. The points of Spec Z are {0}, {2}, {3}, . . ., {p}, . . ., and the fibres of Spec Q over Spec Z are
respectively {0}, ∅, ∅, . . ., ∅, . . .. Of course, Q is not integral over Z nor is it finitely–generated as Z-algebra.
A more germane example is C[X] as included in C[X, Y ]/(XY − 1). The primes of C[X] are {0} and the
principal ideals (X − λ), where λ ranges over C. The fibre over {0} is {0}, that over (X − λ) for λ 6= 0, is the
principal ideal which is the kernel of X 7→ λ; Y 7→ 1/λ. But, over (X), the fibre is ∅. So, all fibres are finite,
B = C[X, Y ]/(XY − 1) is a finitely generated C[X]-algebra yet B is not a finitely generated C[X]-module;
hence B is not integral over A = C[X] under the standard inclusion. Observe also that Spec B −→ Spec A is
not a closed map in this case—this turns out to be the key. For, we have the following fact due to Chevalley:

Fact. If B is a finitely-generated A-algebra under a map ϕ and if |ϕ| is both universally closed and has
finite fibres, then B is a finite A-module (in particular, B is integral over A).
The proof of this is very far from obvious and is not part of our purview. However, the discussion does
suggest the following question: Say A is a domain and write k for Frac A. if K/k is a finite degree field
extension, is IntA (K) a finitely generated A-algebra (hence, a f.g. A-module)? The answer is “no”, which
perhaps is to be expected. But, even if A is noetherian, the answer is still “no”. This is somewhat surprising
and suggests that the finite generation of IntA (K) is a delicate and deep matter. If we are willing to assume
a bit more about K/k we get a very satisfying answer. We’ll need some material from Chapter 4, Section
4.2 and 4.3 for this.

Theorem 3.69 Suppose A is a normal domain with fraction field k and say K/k is a finite separable
extension. Then, IntA (K) is contained in a f.g. A-module in K. In fact, a basis for K/k can be found which
generates the latter A-module. If A is, in addition, noetherian, then IntA (K) is itself a finite A-module;
hence is noetherian.

Proof . We use the trace from K to k (see Chapter 4, Section 4.7), this is a k-linear map, tr : K → k. We
set for x, y ∈ K
hx, yi = trK/k (xy).
The fact we need is that the separability of K/k entails the non-degeneracy of the pairing hx, yi. (Actually,
this is not proved in Section 4.7 of Chapter 4 but is an easy consequence of Newton’s Identities connecting
sums of powers of elements x1 , . . . , xt with elementary symmetric functions in x1 , . . . , xt .) This being said,
we see that K is self-dual as vector space over k, via our pairing hx, yi.
Let B = IntA (K), then in fact Frac(B) = K. To see this, choose x ∈ K, then x has a minimal
k-polynomial m(T ) ∈ k[T ], say

m(x) = xr + α1 xr−1 + · · · + αr−1 x + αr = 0,

αi ∈ k. (†)
Q
As k = Frac(A), for each i, there is si ∈ A with si αi ∈ A. We take s = si , then sαi ∈ A for all i; so
multiply (†) by sr , we get

(sx)r + sα1 (sx)r−1 + · · · + sr−1 αr−1 (sx) + sr αr = 0.

3.5. INTEGRAL DEPENDENCE 215

This shows that xs ∈ IntA (K) = B, so x ∈ Frac(B). (It shows more. Namely, K = (A − {0})−1 B.) It
follows that we may choose a k-basis for K from B; say this is b1 , . . . , bt . By the non-degeneracy of hx, yi,
the dual basis consists of elements of K, say they are c1 , . . . , ct . Thus,

hbi , cj i = δij .
P
Now, choose x ∈ B and write x in terms of the basis c1 , . . . , ct . We have x = γi ci , with the γi ∈ k.
As x and the bi lie in B, we see that xbi ∈ B and statement (1) of Proposition 3.63 shows that hx, bi i ∈ A
because A was assumed normal. But
X X
hx, bi i = h γj cj , bi i = γj δji = γi
j j

so all γi ∈ A. Therefore B ⊆ Ac1 + · · · + Act , as required in the first two conclusions of our theorem. Of
course, if A is noetherian, then B, as a sub-module of a f.g. A-module, is itself finitely generated.

Remark: We cannot expect B to be generated by just t elements as its containing module Ac1 + · · · + Act
is so generated. On the other hand, it can never be generated by fewer than t elements. For if it were, say
B = Ad1 + · · · + Adr , with r < t, then

(A − {0})−1 B = k ⊗A B = k-span of d1 , . . . , dr .

Yet the left hand side is just K and so

t = dim K ≤ r,
a contradiction. When B is generated
` `by t elements, this shows they must be a basis for K/k. If A is a
P.I.D., one knows from B ≤ Ac1 · · · Act that B is generated by t or fewer elements, and so we’ve proved

Corollary 3.70 If A is a P.I.D. and K is a finite separable extension of k = Frac A, then there exist
elements β1 , . . . , βt of B = IntA (K) so that
(1) B is the free A-module on β1 , . . . , βt and
(2) β1 , . . . , βt are a k-basis for K.

A set of elements β1 , . . . , βt having properties (1) and (2) above is called an integral basis for K/k. An
integral basis might exist for a given normal, noetherian A and an extension K/k, but it is guaranteed if A
is a P.I.D.
Theorem 3.69 shows that the difficulty of the finite generation of IntA (K) resides in the possible insepa-
rability of the layer K/k. It can happen that we must continue to add more and more elements without end
in a tower
A ⊆ B1 ⊆ B2 ⊆ · · · ⊆ Bn ⊆ · · · ⊆ B = IntA (K)
and examples (due to Nagata) exist of just this phenomenon. Fortunately, for a big class of integral domains
of interest in both number theory and algebraic geometry, this does not happen—they are well-behaved.
These are the integral domains, A, that are finitely generated k-algebras, where k is a field . We’ll refer to
them as finitely generated domains over k. We will also need some material from Chapter 4 Section 4.11,
namely the notion of transcendence basis. This is just a subset of our domain, algebraically independent
over k (i.e. satisfying no non-trivial polynomial in finitely many variables over k) and maximal with respect
to this property. Every set of generators contains a transcendence basis and all transcendence bases have
the same cardinality—called the transcendence degree of A over k. You should skip the proofs of Theorem
3.71 and 3.72 and come back to read them after Chapter 4.
A main step in proving that the finitely generated domains over k are well-behaved is the following
important theorem due to E. Noether:
216 CHAPTER 3. COMMUTATIVE RINGS

Theorem 3.71 (Noether Normalization Lemma.) If A is a finitely generated domain over the field k, say
A = k[t1 , . . . , tn ], and if d is the transcendence degree of A over k, then there exists a change of coordinates

yj = fj (t1 , . . . , tn ),

in A so that

(1) y1 , . . . , yd are a transcendence basis for A over k and

(2) the injection k[y1 , . . . , yd ] ,→ A = k[y1 , . . . , yn ] makes A integral over k[y1 , . . . , yd ].

If k is infinite, then fj may be taken to be linear. If Frac A is separably generated over k, then the yj may
be chosen to be a separating transcendence basis for Frac A over k.

Proof . (Nagata). We prove the theorem by induction on n; the cases n = 0 or n = 1 are trivial. So, assume
the theorem holds up to n − 1. If d = n, the remarks about transcendence bases just before our proof show
that A is already the polynomial ring in n variables; so, again, nothing need be proved. Therefore, we may
assume d < n. We’ll show there exists y2 , . . . , yn so that k[y2 , . . . , yn ] ,→ k[t1 , . . . , tn ] = A is an integral
morphism (separable in the separating transcendence basis case). If so, then the induction hypothesis applies
to k[y2 , . . . , yn ] and this, together with transitivity of integral dependence and separability, will complete the
proof.
Now d < n, so relabel the t1 , . . . , tn to make t1 algebraically dependent on t2 , . . . , tn . We have a non-trivial
polynomial relation X
c(α) t(α) = 0,
(α)

where (α) = (α1 , . . . , αn ) is a multi-index and t(α) = tα αn

1 · · · tn . Set
1

m
yj = tj − t1 j , j = 2, . . . , n,
mj
where the mj are as yet undetermined integers (≥ 0). Then tj = yj + t1 and so
X
c(α) tα m2 α2
1 (y2 + t1 )
1
· · · (yn + tmn αn
1 ) = 0.
(α)

Expand the latter equation by the binomial theorem to obtain the relation
X (α)·(m)
c(α) t1 + G(t1 , y2 , . . . , yn ) = 0, (†)
(α)

where (m) = (1, m2 , . . . , mn ) and (α) · (m) stands for the dot product α1 + α2 m2 + · · · + αn mn . The
polynomial G has degree in t1 less than the maximum of the exponents (α) · (m). If we can choose the
integers m2 , . . . , mn so that the products (α) · (m) are all distinct, then (†) is an integral dependence of t1
m
over k[y2 , . . . , yn ] as k is a field. But each tj is expressed as yj + t1 j for j = 2, . . . , n; so each tj is integral
over k[y2 , . . . , yn ] and therefore k[t1 , . . . , tn ] is integral over k[y2 , . . . , yn ]. When k[t1 , . . . , tn ] is separably
generated over k, Mac Lane’s Theorem P (Theorem 4.90) shows we may choose t1 separable algebraic over
k[t2 , . . . , tn ]. Then the relation (α) c(α) t(α) = 0 may be chosen to be a separable polynomial in t1 and the
m
way we will choose the m’s (below) will show t1 is separable over k[y2 , . . . , yn ]. As tj = yj + t1 j , we get the
separability of k[t1 , . . . , tn ] over k[y2 , . . . , yn ].
Now we must choose the integers m2 , . . . , mn . For this, consider the differences

(δ)αα0 = (δ1 , . . . , δn )αα0 = (α) − (α0 )

3.5. INTEGRAL DEPENDENCE 217

for all possible choices of our distinct multi-indices (α), except that we do not include (α0 ) − (α) if we have
included (α) − (α0 ). Say there are N such differences, label them δ1 , . . . , δN . Form the polynomial
N
Y
H(T2 , . . . , Tn ) = (δ1j + δ2j T2 + · · · + δnj Tn )
j=1

here, δj = (δ1j , . . . , δnj ) and T2 , . . . , Tn are indeterminates. None of the δj are zero, so H is a non-zero
polynomial and it has integer coefficients. It is well-known that there are non-negative integers m2 , . . . , mn
so that H(m2 , . . . , mn ) 6= 0. Indeed, if b is a non-negative integer larger than any component of any of
our (α)’s, then b, b2 , . . . , bm−1 is such a choice. It is also a choice which gives separability. The fact that
H(m2 , . . . , mn ) 6= 0 means that the (α) · (m) are distinct.
Finally, assume k is infinite. Just as before, arrange matters so that t1 depends algebraically (and sepa-
rably in the separably generated case) on t2 , . . . , tn . Write the minimal polynomial for t1 over k(t2 , . . . , tn )
as
P (U, t2 , . . . , tn ) = 0.
We may asume the coefficients of P (U, t2 , . . . , tn ) are in k[t2 , . . . , tn ] so that the polynomial P (U, t2 , . . . , tn )
is the result of substituting U, t2 , . . . , tn for T1 , . . . , Tn in some non-zero polynomial, P (T1 , . . . , Tn ), having
coefficients in k. Now perform the linear change of variables

yj = tj − aj t1 , j = 2, . . . , n,

where a2 , . . . , an are elements of k to be determined later. As before, each tj is yj + aj t1 ; so it suffices to

prove that t1 is integral (and separable in the separably generated case) over k[y2 , . . . , yn ].
We have
P (t1 , y2 + a2 t1 , . . . , yn + an t1 ) = 0
which gives us

tq1 f (1, a2 , . . . , an ) + Q(t1 , y2 , . . . , yn ) = 0, (∗)

where f (T1 , . . . , Tn ) is the highest degree form of P (T1 , . . . , Tn ) and q is its degree The polynomial, Q,
contains just terms of degree lower than q in t1 . If we produce elements aj in k (j = 2, 3, . . . , n) so that
f (1, a2 , . . . , an ) 6= 0, then (∗) is the required integral dependence of t1 on the y’s. In the separable case, we
also need t1 to be a simple root of its minimal polynomial, i.e.,

dP
(t1 , y2 , . . . , yn ) 6= 0
dt1

(c.f. Theorem 4.5 of Chapter 4). By the chain rule, the latter condition is

dP ∂P ∂P ∂P
(t1 , y) = + a2 + · · · + an 6= 0. (∗∗)
dt1 ∂t1 ∂t2 ∂tn

Now the middle term of (∗∗) is a linear form in a2 , . . . , an and it is not identically zero since on a2 =
a3 = · · · = an = 0 it takes the value ∂P/∂t1 and the latter is not zero because t1 is separable over
k(t2 , . . . , tn ) (Theorem 4.5, again). Thus, the vanishing of the middle term of (∗∗) defines a translate of
a (linear) hyperplane in n − 1 space over k, and on the complement of this hyperplane translate we have
dP/dt1 (t1 , y) 6= 0. The latter complement is an infinite set because k is an infinite field. But from an
infinite set we can always choose a2 , . . . , an so that f (1, a2 , . . . , an ) 6= 0; therefore both our conditions
dP/dt1 (t1 , y) 6= 0 and f (1, a2 , . . . , an ) 6= 0 will hold, and the proof is finished.
218 CHAPTER 3. COMMUTATIVE RINGS

The example discussed previously of C[X] embedded (in the standard way) in C[X, Y ]/(XY − 1) is an
extremely simple instance of the normalization lemma. Namely, rotate the coordinates
X 7→ X + Y ; Y 7→ X − Y
and let T = 1/2(X + Y ); W = 1/2(X − Y ). Then our situation becomes CT embedded in
C[T, W ]/(T 2 − W 2 − 1), an integral extension. See Figures 3.1 and 3.2 below:

Figure 3.1: Before Normalization: A non-integral morphism

becomes after π/4 rotation

Figure 3.2: After Normalization: An integral morphism

Theorem 3.71 is not the sharpest form of the normalization lemma. Here’s an improvement due to
Eisenbud based on a previous improvement of Nagata’s. We offer no proof as we won’t use this sharper
version.
3.5. INTEGRAL DEPENDENCE 219

Theorem 3.72 If A = k[t1 , . . . , tn ] is a finitely generated integral domain over a field k with tr.dk A = d
and if we are given a maximal length descending chain of prime ideals of A
p0 > p1 > · · · > pd−1 > (0),
then there exists a change of coordinates
yj = fj (t1 , . . . , tn )
so that
(1) y1 , . . . , yd are a transcendence basis for A over k,
(2) the injection k[y1 , . . . , yd ] ,→ A makes A integral over k[y1 , . . . , yd ], and
(3) pj ∩ k[y1 , . . . , yd ] = (y1 , . . . , yd−j ).

Here is the promised application of Theorem 3.71 to the well behavedness of finitely generated integral
domains over fields.
Theorem 3.73 When A is a finitely generated integral domain over k and K is a finite extension field over
Frac(A), then IntA (K) is both a finitely generated integral domain over k and a finite A-module.
Proof . We first make two reductions and then treat the main case:
(1) We may assume K = Frac A. For if it is known that the integral closure of A in its own fraction
field satisfies the conclusions of the theorem, then choose a basis y1 , . . . , ys forK over Frac A which basis
consists of elements from IntA (K). This can be done by the argument in the middle of the proof of Theorem
3.69, which argument made no use of any separability hypothesis. Of course, A[y1 , . . . , ys ]is both a finite
A-module and a finitely generated integral domain over k and its fraction field is K. So by our assumption
IntA[y1 ,...,ys ] (K) satisfies the conclusions of the theorem. But, clearly, IntA (K) = IntA[y1 ,...,ys ] (K), which
achieves our first reduction.
(2) We may assume both that k is infinite and that Frac A is separably generated over k. (Here, we are
already using reduction (1) having replaced K by Frac A.) To see this, write Ω for the algebraic closure of
Frac A (see Theorem 4.77) and note that Ω contains k, the algebraic closure of k. Now k is both infinite and
perfect, so by Corollary 4.91, the field k(t1 , . . . , tn ) is separably generated over k; here, A = k[t1 , . . . , tn ]. By
our assumption, Intk[t1 ,...,tn ] (k(t1 , . . . , tn )) is a finite k[t1 , . . . , tn ]-module and a finitely generated k-algebra,
say k[w1 , . . . , wq ].
Now by the normalization lemma (in the infinite, separable case) there are z1 , . . . , zd , algebraically inde-
pendent, which are linear combinations
X n
zj = αij ti
i=1

of the t1 , . . . , tn so that k[t1 , . . . , tn ] is integral and separable over k[z1 , . . . , zd ]. Each wj satisfies a separable,
integral dependence
gj (wj , z1 , . . . , zd ) = 0, j = 1, 2, . . . , q,
over the polynomial ring k[z1 , . . . , zd ]. Also,
Intk[z1 ,...,zd ] (k(t1 , . . . , tn )) = k[w1 , . . . , wq ].

Adjoin to k all the coefficients of these q polynomials and all the αij to get a field, e
k, of finite degree over
k. The entire situation involving k[z’s] and k[w’s] comes from the same situation involving e k[z’s] and e
k[w’s];
so, by the algebraic independence of the z’s, we find
Intek[z1 ,...,zd ] (e
k(t1 , . . . , tn )) = e
k[w1 , . . . , wq ]
220 CHAPTER 3. COMMUTATIVE RINGS

and we know
Intek[z1 ,...,zd ] (e
k(t1 , . . . , tn )) = Intk[t1 ,...,tn ] (e
k(t1 , . . . , tn )).

Of course, ek[w1 , . . . , wq ] is a finite e

k[t1 , . . . , tn ]-module and e
k[t1 , . . . , tn ] is a finite k[t1 , . . . , tn ] = A-module
e e
as k has finite degree over k. Thus, k[w1 , . . . , wq ] is a finite A-module as are all of its submodules, A being
noetherian. But IntA (Frac A)) is the submodule Frac(A) ∩ e k[w1 , . . . , wq ] and as A is a finitely generated
k-algebra so is any A-algebra which is a finite A-module. This achieves reduction (2).
Finally we have the case K = Frac A, k is infinite and Frac A is separably generated over k. By the
normalization lemma, there are linear combinations
n
X
zj = βij ti , j = 1, . . . , d
i=1

so that z1 , . . . , zd are algebraically independent and A is integral and separable over k[z1 , . . . , zd ]. By Theorem
3.69, Intk[z1 ,...,zd ] (Frac A) is a finite k[z1 , . . . , zd ]-module; hence, a finite A-module. Yet, by transitivity of
integral dependence,
Intk[z1 ,...,zd ] (Frac A) = IntA (Frac A).
So, IntA (Frac A) is a finite A-module; thereby a finitely generated k-algebra, as required.
The somewhat involved nature of the two finiteness Theorems (Theorems 3.69 and 3.73) indicates the
delicate nature of the finiteness of IntA (K) as A-module. If the Krull dimension of A is 3 or larger, it can even
happen that IntA (K) is not noetherian (even if A is so). The Japanese school around Nagata studied these
questions and Grothendieck in his algebraic geometry treatise (EGA, IV, part 1, [21]) called attention to the
class of domains having the finiteness property together with all their finitely generated algebra extensions.
He used the terminology universally Japanese rings, but it seems that Nagata rings is the one used most
often now. The formal definition is this

Definition 3.6 An integral domain, A, is a Nagata ring if and only if for every finitely generated A-algebra,
B, which is a domain and any finite extension, K, of Frac B, the ring IntB (K) is a finite B-module.

As a corollary of Theorem 3.69, we see immediately the following

Proposition 3.74 If A is the ring of integers in a number field (i.e., A = InZ (K), where K is a finite
extension of Q), then A is a Nagata ring as is A[t1 , . . . , tn ].

A main theorem, proved by Nagata, concerning these matters is the following:

Theorem 3.75 (Nagata) Say A is a complete, noetherian local domain, and K is a finite degree extension
field of Frac(A), then IntA (K) is a finitely generated A-algebra and a finite A-module.

This theorem is not part of our purview, nor will we use it; so, its proof is omitted.
There is another finiteness result involving integrality which has many uses.

Proposition 3.76 (E. Noether) If B is a finitely generated A-algebra, A being noetherian, and if C is a
sub A-algebra of B so that B is integral over C, then C is a finitely generated A-algebra.

Proof . Write B = A[t1 , . . . , tn ]; each tj satisfies an integral dependence over C

gj (tj ) = 0, j = 1, . . . , n.
3.5. INTEGRAL DEPENDENCE 221

If α1 , . . . , αq are the coefficients (∈ C) of all these equations, form A[α1 , . . . , αq ] ⊆ C. The ti are integral
over A[α1 , . . . , αq ] and they generate B; so, B is a finite A[α1 , . . . , αq ]-module. But C is a sub A[α1 , . . . , αq ]-
module of B and A[α1 , . . . , αq ] is noetherian. Therefore, C is a finitely generated A[α1 , . . . , αq ]-module, say
C = A[α1 , . . . , αq ][z1 , . . . , zs ]; we are done.
e then
What happens if A ⊆ Frac(A) is not a normal domain? Of course we’ll form IntA (Frac(A)) = A,
e
we want to study the relations between A and A. For example look at

A = Z[ni], n∈Z and n > 0

and
e = IntA (Q(i)) = Z[i].
A
The main invariant controlling the relations between A and Ae is the transporter (A
e −→ A) in A. That is,
we examine
f = (A e −→ A) = {ξ ∈ A | ξ A
e ⊆ A}.
e or just the conductor of the integral
The set f is, of course, an ideal of A; it is called the conductor of A in A
closure of A. The symbol f comes from the German word for conductor: Führer. But, clearly, f is also an
e In the example above,
ideal of A.
f = {ξ ∈ Z[ni] | n|<(ξ)}.

Remark: The domain, A, is normal if and only if f is the unit ideal. An ideal, A, of A which is also an
e must necessarily be contained in the conductor, f . That is, f is the unique largest ideal of A
ideal of A
e
which is simultaneously an ideal of A.
e ⊆ A as A is an A-ideal
The first of these statetements is obvious; for the second, we have AA e and A ⊆ A
as A is an A-ideal. Thus,
AAe⊆A⊆A
e −→ A) = f .
and this says A ⊆ (A
e vis a vis localization and prime ideals is this:
The connection between A and A

e and the conductor, f , of A in A

Proposition 3.77 For a domain, A, its integral closure A e we have

e = IntS −1 A (Frac(A))
(1) If S is a multiplicative set in A, then S −1 A
e that is S −1 A is normal.
(2) If f ∩ S 6= ∅, then S −1 A = S −1 A,
e is a finite A-module then the conductor of S −1 A in S −1 A
(3) If A e is f · S −1 A = f e .

e is a finite A-module, them S −1 A is normal if and only if f ∩ S 6= ∅.

(4) If A
e is a finite A-module, then
(5) If A

{p ∈ Spec A | Ap is not normal}

is closed in Spec A; indeed it is V (f ). Hence, in this case, Ap is a normal ring on an open dense set
of Spec A.

Proof . (1) This is clear from Proposition 3.54 and Fact B following it.
e We know sα = a ∈ A; so, α = a/s ∈ S −1 A. We find A
(2) Write s ∈ f ∩ S and choose α ∈ A. e ⊆ S −1 A,
−1 e −1
hence S A ⊆ S A. The other inclusion is clear.
222 CHAPTER 3. COMMUTATIVE RINGS

e in this part and in part (4). To check

(3) Write α1 , . . . , αt for a finite set of A-module generators for A
−1 −1 e −1 e −→ S −1 A). For the
that an element x ∈ S A lies in (S A −→ S A), it suffices to see that it is in (A
−1 e
latter, all we need is xαj ∈ S A for j = 1, . . . , t. Conversely, if x ∈ (S A −→ S −1 A), then certainly
−1

xαj ∈ S −1 A, all j.
Now xαj ∈ S −1 A implies there is some sj ∈ S with sj xαj ∈ A. If s = s1 · · · st , then sxαj ∈ A therefore
sx ∈ f , i.e., x ∈ f e . The converse is clear.
e −→ S −1 A)
(4) The “if” part of our conclusion is (2), so say S −1 A is normal. Then the conductor (S −1 A
e
is the unit ideal; so, (3) shows f = unit ideal. This implies f ∩ S 6= ∅.
(5) Write S(p) for A − p, then Ap is not normal iff f ∩ S(p) = ∅ which holds iff f ⊆ p; that is iff
p ∈ V (f ). To finish the proof we need only show that any non-empty open set of Spec A is dense when A is a
domain. But, this will hold if we show Xf ∩ Xg 6= ∅ (provided neither Xf nor Xg is empty) (DX). However,
Xf ∩ Xg = Xf g , and, as neither f nor g is zero, their product is non-zero (and not nilpotent). Now apply
Proposition 3.12 part (3).

Corollary 3.78 For a domain A and its integral closure, A, e assume A e is a finite A-module. Then, for a
prime p of Spec A not in V (f ), there exists one and only one prime ideal, e e lying over A. This prime
p, of A
e
ideal is pAp ∩ A.

Proof . Existence is clear either by the Lying Over Theorem or by the fact that pAp is prime and Ap ⊇ A. e
(The latter holds as Ap is normal since p ∈/ V (f ).) To see uniqueness, observe as p 6⊇ f there is δ ∈ f with
δ∈/ p. Then for any ideal, A, of Ae
δA ⊆ δ Ae⊆A

and δA ⊆ A, too. Therefore δA ⊆ A ∩ A; so if A is an ideal contracting to p we get δA ⊆ p. Now, δ ∈

/ p,
therefore A ⊆ pAp , so A ⊆ pAp ∩ A e = e
p. Suppose, in fact, A is prime yet A < e p, then we’d have a
contradiction to non-comparability (Proposition 3.62).
e is not a prime ideal of A;
Note: Generally, pA e but, of course, pA
e is always contained in e
p.
3.6. PRIMARY DECOMPOSITION 223

3.6 Primary Decomposition

In Z, we have unique factorization and we know this is not valid in an arbitrary (even Noetherian) commuta-
tive ring. Can one generalize so as to obtain a “decomposition” of ideals (or submodules) into special ideals
(resp. modules) which resemble prime powers? Surprisingly, the answer is connected with a generalization
of Fitting’s lemma from linear algebra.

Lemma 3.79 (Fitting’s lemma) If V is a finite dimensional vector space over a field, k, and θ : V → V is
an endomorphism, then there exist subspaces W and Z of V so that
`
(1) V = W Z.
(2) θ  W is an isomorphism.
(3) θ  Z is nilpotent.

Proof . See any introductory algebra text.

Look at Z. Pick n, then we have the ideal A = nZ. Factor n as n = pe11 pe22 · · · pet t , where p1 , . . . , pt
are distinct prime numbers. We get A = Pe11 · · · Pet t , where Pj is the prime ideal pj Z. Now, we also have
Tt e
A = j=1 Pj j , since the Pj are pairwise comaximal.
This last equality is still wrong, say in C[X, Y ], and the fault is the Pei i . They are not general enough.
Let A be a commutative ring, M an A-module and N ⊆ M a submodule. Set
p p
RadM (N ) = (M −→ N ) = {x ∈ A | xM ⊆ N } = {x ∈ A | (∃k > 0)(xk M ⊆ N )}.

This is the relative radical of N in M . The following properties are easily checked:
(1) RadM/N ((0)) = RadM (N ).
p
(2) RadM ((0)) = Ann(M ).
√
(3) RadA (q) = q.
√
(3a) RadA/q ((0)) = q.
(4) RadM (N ∩ P ) = RadM (N ) ∩ RadM (P ).
√
(5) RadM (AN ) ⊇ A ∩ RadM (N ).
Here, A is an ideal of A; M is an A-module; N is a submodule of M .

Definition 3.7 A module, M , is coprimary iff for every a ∈ A, the map σa : M → M via σa (m) = am is
either injective or nilpotent. (The map σa is called a homothety.) An ideal, q, of A is a primary ideal iff the
module, A/q, is coprimary.

Notice the clear connection of this idea with Fitting’s lemma.

Proposition 3.80 For any commutative ring, A, and any ideal, q, the following are equivalent:
/ q, then xk ∈ q, for some k ≥ 1.
(α) For all x, y ∈ A if xy ∈ q but y ∈
√
(β) For all y ∈
/ q, we have (y −→ q) ⊆ q.
S √
(γ) y6∈q (y −→ q) = q.
(δ) Every zero divisor of the ring A/q is nilpotent.
 224 CHAPTER 3. COMMUTATIVE RINGS

() The ideal q is primary.

Proof . The equivalence (α) ⇐⇒ (β) is clear and the implication (γ) =⇒ (β) is a tautology. If (β), then pick
√
ξ ∈ q. If ξ ∈ q, then ξ ∈ (y −→ q) for all y. Thus, we may assume that ξ ∈ / q and so, there is a minimum
k ≥ 2 so that ξ k ∈ q. Let y = ξ k−1 ∈
/ q. We have ξy = ξ k ∈ q, so, ξ ∈ (ξ k−1 −→ q) and (γ) holds.
(α) =⇒ (δ). Pick x ∈ A/q, a zero divisor, which means that there is some y 6= 0 with x y = 0. It follows
/ q and xy ∈ q; by (α), we get xk ∈ q, for some k, and so, xk = 0.
that y ∈
(δ) =⇒ (). Pick a ∈ A. We need to show that σa is injective or nilpotent in A/q. Say σa is not injective.
Then, there is some y 6= 0 in A/q and ay = 0 in A/q, i.e. a y = 0. But, y 6= 0, so, by (δ), a is nilpotent.
Consequently, ak = 0, and so, (σa )k = 0 in A/q.
() =⇒ (α). Pick x, y with xy ∈ q and y ∈
/ q. Look at σx on A/q. We have

σx (y) = x y = xy = 0, as xy ∈ q.

As y 6= 0, the map σx is not injective on A/q. By (), the map σx is nilpotent. This means that
(σx )k = σxk = 0 in A/q. In particular, σxk (1) = xk 1 = 0, i.e., xk = 0. Therefore, xk ∈ q.
√
Corollary 3.81 If q is maximal, then q is primary. In particular, if m is a maximal ideal, then mn is
primary for all n > 0.
√ √
Proof . The image of q in A/q is the nilradical of A/q. Since q is maximal in A, the ring A/q has a
unique maximal ideal. It follows that every element of A/q is either a unit√or nilpotent, so Proposition 3.80
(3) applies. The second part of the statement follows from the first since pn = p for every prime ideal, p.

 There exist prime ideals, p, such that pn is not primary. There exist primary ideals, q, not of the form
pn , where p ∈ Spec A.

√
Corollary 3.82 Say q is a primary ideal of A, then q is a prime ideal.
√ √
Proof . Pick x, y ∈ A with xy ∈ q and y ∈ / q. Then, xk y k = (xy)k ∈ q, for some k > 0. So, σxk (y k ) = 0
in A/q and y k 6= 0 in A/q. Therefore, our homothety, σxk , is nilpotent, so, σx is nilpotent, i.e., (σx )l ≡ 0 on
√
A/q. Then, (σx )l (1) = xl · 1 = xl = 0 in A/q, and so, xl ∈ q, i.e., x ∈ q.
√
 There exist non-primary ideals, A, yet A is prime.

Definition 3.8 A submodule, N , of a module, M , is primary in M iff M/N is co-primary. Then, RadM (N )
is prime (same argument), say p. In this case, we say N is p-primary when M/N is p-coprimary, i.e., M/N
is coprimary and RadM (N ) = p.

Say M is an A-module, N is a submodule of M and S is a multiplicative set in A. Look at

N ec = {m ∈ M | (∃s ∈ S)(sm ∈ N )} = S(N ),

and call it the S-component of N or S-saturation of N .

Further Properties:
(6) S((0)) = Ker (M −→ S −1 M ).
Tt Tt
(7) S( i=1 Ni ) = i=1 S(Ni ).
(8) S(V −→ N ) = (V −→ S(N )).
3.6. PRIMARY DECOMPOSITION 225

Proposition 3.83 If A is a commutative ring, M is a f.g. A-module and N a submodule of M , then the
following are equivalent:

(1) N is primary in M .

(2) For all multiplicative sets, S ⊆ A, we have

(
N
S(N ) = or
M.

(3) For all multiplicative sets, S ⊆ A, the map M/N −→ S −1 (M/N ) is either injective or zero.

Proof . Note that SM/N (0) = S(N ) = Ker (M/N −→ S −1 (M/N )). Therefore, (2) and (3) are equivalent.
p
(1) =⇒ (2). Take any S and examine (M −→ N ) = RadM/N ((0)). There are two cases:
(1) S ∩ RadM/N ((0)) = ∅ or (2) S ∩ RadM/N ((0)) 6= ∅.
Case 2. There is some s ∈ S with s ∈ RadM/N ((0)). So, sk ∈ (M −→ N ) and then sk ∈ S implies that
M ⊆ S(N ); thus, M = S(N ).
k
Case 1. Pick s ∈ S and look pat σs . If σs is nilpotent on M/N , then (σs ) = σsk ≡ 0 on M/N , which
k
implies that s M ⊆ N . So, s ∈ (M −→ N ) ∩ S, a contradiction. Therefore, σs must be injective on M/N ,
by (1). This means given any m ∈ M , we have σs (m) = sm = 0 in M/N iff m ∈ N , already, i.e., sm ∈ N
iff m ∈ N . As this holds for all s ∈ S, we have S(N ) = N .
(2) =⇒ (1). Pick s ∈ A and look at S = {sk | k ≥ 0}. If s ∈ N (A), then (σs )k = σsk ≡ 0 on any
module. So, we may assume s ∈ / N (A) and then, S is a multiplicative set. Thus, (2) holds for S. We have
to show that M/N is coprimary, i.e., σs is either nilpotent or injective. Say, σs is not injective on M/N ,
i.e., S(N ) 6= N . By (2), we have S(N ) = M . Pick generators, m1 , . . . , mt for M . As S(N ) = M , each
mj ∈ S(N ); so, there is some kj with skj mj ∈ N , for j = 1, . . . , t. Let k = max{k1 , . . . , kt }, then sk mj ∈ N ,
for j = 1, . . . , t. It follows that sk M ⊆ N and so, sk kills M/N , i.e. σs is nilpotent on M/N .

Proposition 3.84 (E. Noether, 1921) If M is a noetherian module, then any non-primary submodule, N ,
of M is reducible, i.e., N is the intersection, N = Q1 ∩ Q2 , of proper submodules of M properly containing
N.

Proof . (Adapted from Fitting’s lemma.) Since N is non-primary, M/N is not coprimary. So, there is some
a ∈ A so that σa is not injective and not nilpotent on M/N . Write Mj = Ker (σa )j = Ker (σaj ) on M/N .
We have an ascending chain
M 1 ⊆ M 2 ⊆ M3 ⊆ · · · .
By the ACC, the chain stops, say at r. We have Mr = Mr+1 = · · · = M2r . Let ϕ = σar ∈ EndA (M/N ).
We have Ker ϕ 6= M/N , else (σa )r ≡ 0, contradicting the non-nilpotence of σa . So, Im ϕ 6= (0). Also,
Ker ϕ ⊇ Ker σa 6= (0), as σa is not injective. I claim that Ker ϕ ∩ Im ϕ = (0).
Pick ξ ∈ Ker ϕ ∩ Im ϕ. So, ξ = ϕ(η) = ar η. As ϕ(ξ) = 0, we have ϕ(ar η) = 0; thus ar ϕ(η) = 0, and so,
a η = 0, i.e., η ∈ M2r = Mr . Consequently, ar η = 0, i.e., ξ = 0, as desired. But, now, Ker ϕ ∩ Im ϕ = (0)
2r

implies that
N = π −1 (Ker ϕ) ∩ π −1 (Im ϕ),
where π : M → M/N is the natural projection.
We need a restatement of a Proposition 3.83 for the reduction process:

Proposition 3.85 Say N is a submodule of M , and p is a given prime ideal.

 226 CHAPTER 3. COMMUTATIVE RINGS

(a) N is p-primary in M iff for all multiplicative sets, S, of A, we have


N iff p ∩ N = ∅
S(N ) =
M iff p ∩ N 6= ∅.

(b) If N1 , . . . , Nt are all p-primary, then N1 ∩ · · · ∩ Nt is again p-primary.

(c) If V is any submodule of M , then when N is p-primary, we have

A iff V ⊆ N
S(V −→ N ) =
p-primary ideal iff V 6⊆ N .

Proof . (a) The module N is primary iff M/N is coprimary iff S((0)) = (0) or S((0)) = M/N , for any
multiplicative subset, S (where (0) ⊆ M/N ) iff S(N ) = N or S(N ) = M , for any such S. (Recall,
S((0)) = S(N ).) But, the dichotomy: S(N ) = N or M , depends on S ∩ RadM (N ) = S ∩ RadM/N ((0)) =
p
S ∩ (M −→ N ). Namely, S(N ) = N iff S ∩ RadM (N ) = ∅ and S(N ) = M iff S ∩ RadM (N ) 6= ∅. But
here, p = RadM (N ), so (a) is proved.
Tt Tt
(b) Now, S( i=1 Ni ) = i=1 S(Ni ), so (a) implies (b).
(c) If V ⊆ N , then (V −→ N ) = A, so, S(V −→ N ) = A. So, we may assume V 6⊆ N . Recall that
S(V −→ N ) = (V −→ S(N )). We will test S(V −→ N ) by part (a) (here, M = A). But,

M iff S ∩ p 6= ∅
S(N ) =
N iff S ∩ p = ∅.

In the case S ∩ p 6= ∅, we have S(V −→ N ) = (V −→ M ) = A. If S ∩ p = ∅, then S(V −→ N ) = (V −→ N ),

and the test of (a) shows (c).
Reduction Process for Primary Decomposition
Say N = Q1 ∩ Q2 ∩ · · · ∩ Qt is a decomposition of N as a finite intersection of pi -primary modules, Qi .
 No assertion pi 6= pj is made.

Tt
(1) Remove all Qj from the intersection i=1 Qi , whose removal does not affect the intersection.
(2) Lump together as an intersection all the Qi ’s for which the pi ’s agree. By (b), the “new” intersection
satisfies:
(α) No Qe j , still primary, can be removed without changing the intersection.
 q 
(β) All the pj = (M −→ Q e j ) are distinct.

Such a primary decomposition is called reduced .

Theorem 3.86 (Lasker-Noether Decomposition Theorem, 1921) Every submodule, N , of a noetherian mod-
ule, M , can be represented as a reduced primary decomposition:

N = Q1 ∩ Q2 ∩ · · · ∩ Qt .

Proof . (Noetherian induction—invented for this theorem.) Let

S = {N ⊆ M | N is not a finite intersection of primary submodules}.

If S =
6 ∅, by the ACC, the set S has maximal element. Call it N . Of course, N is not primary. By Noether’s
proposition (Proposition 3.84), there exist Q1 , Q2 > N , so that N = Q1 ∩ Q2 . But N is maximal in S, so
3.6. PRIMARY DECOMPOSITION 227

Tt1 (1) T t2 (2) (i)

Qj ∈/ S for j = 1, 2. Thus, we can write Q1 = j=1 Qj and Q2 = k=1 Qj , where the Qj are primary
(i = 1, 2, finitely many j’s). Consequently, we get
(1) (1) (2) (2)
N = Q1 ∩ · · · ∩ Qt1 ∩ Q1 ∩ · · · ∩ Qt2 ,

contradicting N ∈ S. Therefore, S = ∅. Now apply the reduction process to a primary decomposition of N

and we get the conclusion.

Corollary 3.87 (Lasker’s Decomposition Theorem, original form, 1905) If A = C[X1 , . . . , Xn ], then every
ideal, A, admits a reduced primary decomposition: A = q1 ∩ · · · ∩ qt .

Corollary 3.88 (Noether’s statement) If A is any noetherian ring and A is any ideal of A, then A admits
a reduced primary decomposition: A = q1 ∩ · · · ∩ qt .

Now, what about uniqueness?

Proposition 3.89 Say that N is an A-submodule of M , and N = Q1 ∩ · · · ∩ Qt is a reduced primary

decomposition for N . Let I = {1, . . . , t} and given any multiplicative subset, S, of A, write

S(I) = {i ∈ I | S ∩ pi = ∅}.

Here, pi = RadM (Qi ) is the prime associated to Qi . Then,

\
(a) S(N ) = Qj .
j∈S(I)
 −1
S M if i ∈
/ S(I)
(b) S −1 Qi =
pei -primary submodule of S −1 M if i ∈ S(I).
\
(c) S −1 N = S −1 Qj ,
j∈S(I)

and this is a reduced primary decomposition for S −1 N as submodule of S −1 M .

Tt
Proof . (a) We know that S(N ) = j=1 S(Qj ) and S(Qj ) = M when j ∈ / S(I) and S(Qj ) = Qj for j ∈ S(I)
(previous proposition). Thus, it is clear that (a) holds.
(b) Now, Qi is pi -primary, so S(Qi ) = M if pi ∩ S 6= ∅ else S(Qi ) = Qi or equivalently, S((0)) = M/Qi
if i ∈
/ S(I) else S((0)) = (0) (where (0) is the zero ideal in M/Qi ). Say, i ∈ / S(I), then S(Qi ) = M , and
so, for every m ∈ M , there is some s = s(m) ∈ S with sm ∈ Qi . Hence, m/1 ∈ S −1 Qi and it follows that
S −1 M ⊆ S −1 Qi ; yet, of course, S −1 Qi ⊆ S −1 M , so S −1 Qi = S −1 M , as required. Now, say i ∈ S(I), so
pi ∩ S = ∅. Observe, every multiplicative set, say T , of S −1 A, has the form S −1 T0 , for some multiplicative
set, T0 , of A. But, M/Qi is coprimary which means that M/Qi −→ T0−1 (M/Qi ) is either injective (case:
T0 ((0)) = (0)) or zero (case: T0 ((0)) = M/Qi )). Therefore, as S −1 A is flat over A, we get

S −1 (M/Qi ) −→ S −1 T0−1 (M/Qi ) is injective or zero, (∗)

the first if T0 ∩ pi = ∅, i.e., T (= S −1 T0 ) ∩ pei = ∅, the second if T0 ∩ pi 6= ∅, i.e., T (= S −1 T0 ) ∩ pei 6= ∅. But,

S −1 T0−1 (M/Qi ) = T −1 (S −1 M/S −1 Qi ) and S −1 M/S −1 Qi = S −1 (M/Qi ), so

S −1 M/S −1 Qi −→ T −1 (S −1 M/S −1 Qi ) is injective or zero

depending on T ∩ pei being empty or not. Therefore, S −1 Qi satisfies our test for pei -primariness.
(c) We know from (b) that S −1 Qj is pej -primary and pei 6= pej if i 6= j, as i, j ∈ S(I) and there is a
one-to-one correspondence between the p’s so that p ∩ S = ∅ and the pe of S −1 A. The rest should be obvious
(DX).
228 CHAPTER 3. COMMUTATIVE RINGS

Theorem 3.90 (First Uniqueness Theorem) If N , an A-submodule p of M , has a reduced primary decompo-
sition N = Q1 ∩ · · · ∩ Qt , then the prime ideals p1 , . . . , pt (pi = (M −→ Qi )) are uniquely determined by
N and M , up to the order of their listing.

Proof . Assume that N = Q1 ∩ · · · ∩ Qs = Q01 ∩ · · · ∩ Q0t are two reduced decompositions for N in M . We
use induction on s + t. When s + t = 2, we have s = t = 1 and Q1 = Q01 and uniqueness is obvious. Assume
that uniqueness holds for all submodules, N , for which s + t ≤ r − 1. Consider N and two decompositions
with s + t = r and let
s−1
[ [
S =A− pi − p0j .
i=1 p0j 6=ps

Now, S ∩ pi = ∅ for i = 1, . . . , s − 1 and S ∩ p0j = ∅ for all j with p0j 6= ps . So,

s
\ s−1
\
S(N ) = S(Qi ) = Qi
i=1 i=1

as S(Qi ) = Qi whenever S ∩ pi = ∅. Also,

\ \
S(N ) = S(Q0j ) = Q0j .
p0j 6=ps p0j 6=ps

For S(N ), the sum of the number of components is at most s − 1 + t < r; so, the induction hypothesis implies
S(N ) has the uniqueness property. However, can it be that p0j 6= ps for j = 1, . . . , t? Were that true, the
Tt
second intersection would give S(N ) = j=1 Q0j = N . Thus, we would have
s
\ s−1
\
Qi = N = S(N ) = Qj ,
i=1 j=1

contradicting the fact that the first decomposition is reduced. Therefore, there is some j with p0j = ps , and
now the induction hypothesis implies

{p1 , . . . , ps−1 } = {p0j | p0j 6= ps },

and the proof is complete.

Definition 3.9 If N is a submodule of M and N has a primary decomposition, then the primes p1 , . . . , ps
corresponding to the Qj ’s which appear in the decomposition are called the essential primes of N in M .
The set of such is denoted EssM (N ). When N = (0), the primes appearing are called associated primes of
M and this set is denoted Ass(M ). Of course, Ass(M/N ) = EssM (N ). The minimal elements of EssM (N )
or Ass(M ) are called isolated essential primes of N in M (resp. isolated associated primes of M ). The Qi
corresponding to isolated primes of either type are called isolated primary components of N in M or isolated
primary components of M .

Theorem 3.91 (Second Uniqueness Theorem) The isolated primary components of N in M are uniquely
determined by M and N .

Proof . Let Q be such an isolated component of N in M and let p be the corresponding minimal prime. Look
at S = A − p. If P ∈ EssM (N ), then P > p implies that P ∩ S 6= ∅ and as p is minimal, all other P touch
S. It follows from Proposition 3.89 that S(N ) = Q.
The Lasker-Noether theorem has an immediate application to number theory. This concerns factorization
and it shows clearly how Lasker-Noether provides a generalization to Noetherian rings of unique factorization
in UFD’s.
3.6. PRIMARY DECOMPOSITION 229

Definition 3.10 A Dedekind domain is a noetherian, normal domain of Krull dimension 1.

Examples of Dedekind domains.

(1) Every P.I.D. is a Dedekind domain.
(2) If K is a finite extension of Q (that is, K is a number field ) and OK = IntZ (K) (the integral closure
of Z in K), then OK is a Dedekind domain. The ring OK is called the ring of integers in K.
(3) Let X be a compact Riemann surface and x ∈ X, any point in X. Let

A = {f ∈ Mer(X) | poles of f are only at x}.

Then, A is a Dedekind domain.

(3a) Let X be an open Riemann surface of finite character, which means that X = X ∪ finite set of points
is a compact Riemann surface. Then, Hol(X)(= the ring of all holomorphic functions on X) is a Dedekind
domain.
Say A is a Dedekind domain. If p ∈ Spec A but p 6= (0), then dimension 1 implies that p ∈ Max(A). From
Theorem 3.56, Ap is a PID. Take any non-zero ideal, A, then by Lasker-Noether, we can write A = q1 ∩· · ·∩qt ,
a reduced primary decomposition. Now,
√
pj = qj ⊇ A > (0),

so each of the pj ’s is a maximal ideal. It follows that each pj is isolated and, by the second uniqueness
theorem, the qj ’s are unique. Moreover, whenever i 6= j,
p p
qi + qj = pi + pj = A,

so that 1 ∈ qi + qj . We deduce that the qj are pairwise comaximal and the Chinese Remainder Theorem
says
\t Yt
A= qi = qi .
i=1 i=1

The ring A/qi is noetherian and any p ∈ Spec(A/qi ) corresponds to a prime of A containing pi ; that is,
p must be pi . Consequently, A/qi is a local ring with the DCC and by Nagata’s Theorem pi (= image pi in
A/qi ) is nilpotent. Let ei be its index of nilpotence so that

pei i ⊆ qi < piei −1 .

But, Api is a PID, and Proposition 3.5 shows that qi = pei i . In summary, we get the following theorem of
Dedekind:

Theorem 3.92 (Dedekind, 1878) In a Dedekind domain, every nonzero ideal, A, is a unique product of
powers of prime ideals: A = pe11 pe22 · · · pet t .

Corollary 3.93 (Kummer, 1833) In the ring of integers of a number field, every nonzero ideal is a unique
product of powers of prime ideals.

After this little excursion into number theory and the connection of primary decomposition to questions
of factorization, we resume our study of primary decomposition for modules—especially its applications to
the structure of modules.

Lemma 3.94 Say M is a p-coprimary module and N (6= (0)) is a submodule of M . Then, N is also
p-coprimary.
230 CHAPTER 3. COMMUTATIVE RINGS

Proof . Pick a ∈ A with σa not injective on N . Then, σa is not injective on M , so, σa is nilpotent on M
(as M is coprimary). Therefore, σa  N is also nilpotent; so,
p N is coprimary. Letpe
p be the prime associated
with N while p is the prime for M . We know that p = Ann(M ), while e p = Ann(N ). If x ∈ p, then
xk ∈ Ann(M ); so, xk ∈ Ann(N ), i.e, x ∈ e
p. Thus, p ⊆ e
p.
Now, pick x with σx not injective on N . This implies that (σx )k ≡ 0 on N , that is, xk ∈ Ann(N ), i.e.,
x∈e p. Thus, x ∈ e
p implies σx is not injective on N , hence σx is not injective on M , and so (σx )k ≡ 0 on M
as M is coprimary which implies that x ∈ p. Therefore, we also have e p ⊆ p, and e
p = p.

Proposition 3.95 A necessary and sufficient condition that M be p-coprimary is that Ass(M ) = {p}. Let
N ⊆ M , for arbitrary M and N , then Ass(N ) ⊆ Ass(M ).

Proof . Assume M is p-coprimary. Then (0) is p-primary in M . By the first uniqueness theorem, Ass(M ) =
{p}. Conversely, if Ass(M ) = {p}, then (0) has just one primary component, whose prime is p. So, (0) is
p-primary and it follows that M is p-coprimary.
Assume N ⊆ M . Write (0) = Q1 ∩ · · · ∩ Qt , a reduced primary decomposition of (0) in M . Then,
Ass(M ) = {p1 , . . . , pt }. By intersecting (0) = Q1 ∩ · · · ∩ Qt with N , we get

(0) = (Q1 ∩ N ) ∩ · · · ∩ (Qt ∩ N ).

Observe that we have the composite map

N ,→ M −→ M/Qi

and its kernel is N ∩ Qi . Hence, N/(N ∩ Qi ) ,→ M/Qi . But, M/Qi is pi -coprimary; so, from the argument
above, N/(N ∩ Qi ) is also pi -coprimary, provided that N/(N ∩ Qi ) 6= (0). Now, we have N/(N ∩ Qi ) = (0)
iff Qi ⊇ N . Consequently, we have

(0) = (Qi1 ∩ N ) ∩ · · · ∩ (Qis ∩ N ) in N ,

where Qil 6⊇ N , for each il , and Qil ∩ N is pil -primary in N . By the first uniqueness theorem, we deduce
that
Ass(N ) = {p ∈ Ass(M ) | Q 6⊇ N, where Q corresponds to p}.

Corollary 3.96 (of the proof ) If N ⊆ M , then

Ass(N ) = {p ∈ Ass(M ) | Q 6⊇ N, where Q corresponds to p}.

Tt
Proposition 3.97 Say (0) = i=1 Qi is a reduced primary decomposition of (0) in M and let N be a
submodule of M . Then, N is pi -coprimary
T if and only if N ∩ Qi = (0). In particular, there exist pi -
coprimary submodules of M , namely, j6=i Qj . In fact, p ∈ Ass(M ) iff M contains a submodule which is
p-coprimary. Lastly, if N ∩ (Qi + Qj ) = (0), then N = (0). Therefore, M is an essential extension of
Qi + Qj .

Proof . Say N ∩ Qi = (0), then

N = N/(N ∩ Qi ) ,→ M/Qi .
Therefore, N is a submodule of the pi -coprimary module M/Q Ti . But then, N is pi -coprimary (as a submodule
of a pi -coprimary is pi -coprimary). Conversely, since (0) = i Qi , we get
\
(0) = (Qi ∩ N ), (∗)
i
3.6. PRIMARY DECOMPOSITION 231

and we know that Qi ∩ N is pi -primary if Qi ∩ N 6= N , and that (∗) is a reduced decomposition. Since
N is pi -coprimary, by the first uniqueness theorem, there can be only one term in (∗), i.e., Qj ⊇ N for all
j 6= i; then (0) = Qi ∩ N . The second statement is now obvious. If N ∩ (Qi + Qj ) = (0), then, of course,
N ∩ Qi = N ∩ Qj = (0). Then, pi and pj would be primes of N , yet, N is coprimary by the first statement.
Consequently, pi = pj , a contradiction. So, N = (0).
To finish this chain of ideas, we need the “power lemma”:

p (Power lemma) Say A is a commutative ring with unity and M, F are A-module with F ⊆ M .
Lemma 3.98
Write A = (M −→ F ) and assume A is f.g. as ideal. Then, there is some ρ >> 0 so that Aρ M ⊆ F .
Proof . Let α1 , . . . , αt be generators for A. For l = 1, . . . , t, there is some kl > 0 so that αlkl M ⊆ F . Let
ρ = k1 + · · · + kt . Every element of A has the form r1 α1 + · · · + rt αt , where ri ∈ A. Every element of
Aρ is a sum of terms s(a1 a2 · · · aρ ); s ∈ A; a1 , . . . , aρ ∈ A. Then, a1 · · · aρ is a sum of monomials of the
form cα1i1 · · · αtit , where c ∈ A and i1 + · · · + it = ρ. Now, at least one il ≥ kl in the last sum, and then,
α1i1 · · · αtit M ⊆ F . Therefore, Aρ M ⊆ F .
Theorem 3.99 If A is a noetherian ring and M is a f.g. A-module, then for all submodules, N , of M , all
the prime ideals of Ann(N ) are in Ass(M ). A prime ideal, p, is in Ass(M ) iff there is some x ∈ M so that
p = Ann(x) iff A/p is isomorphic to a submodule of M .
T
Proof . In M , we have (0) = i Qi , a reduced primary decomposition, and we let pi correspond to Qi . The
first uniqueness theorem implies Ass(M ) = {p1 , . . . , pt }. Also,
\
Ann(N ) = (N −→ (0)) = (N −→ Qi ).
i

But, we know that 

A if N ⊆ Qi
(N −→ Qi ) =
pi -primary ideal otherwise.
We get a reduced primary decomposition of Ann(N ):
\
Ann(N ) = (N −→ Qj ).
j|Qj 6⊇N

By the first uniqueness theorem, the primes of Ann(N ) are the pj ’s for which Qj 6⊇ N , so, they are contained
in Ass(M ).
We have p = Ann(A/p) and A/p = Aξ, for some ξ (where ξ is the image of 1 modulo p). Given x with
Ann(x) = p, the map ξ 7→ x gives A/p ∼
= Ax ⊆ M and conversely. Say p kills some x exactly, then, as A/p
is p-coprimary, Ass(Ax) = {p}. Yet Ax ⊆ M , so, p ∈ Ass(M ).
Conversely, say p ∈ Ass(M ). We must find x ∈ M with Ann(x) = p.
BypProposition 3.97, ifp p ∈ Ass(M ), then there is a submodule, P , so that P is p-coprimary. Thus,
p = Ann(P ), i.e., p = (P −→ (0)). In the power lemma, set p = A, P = M , (0) = F . As A is
noetherian, p is f.g. and by the power lemma, there is some ρ >> 0 with pρ P = (0). If we choose ρ minimal
with the above property, we have pρ P = (0) and pρ−1 P 6= (0). Pick any x 6= 0 in pρ−1 P . Then,
px ⊆ ppρ−1 P = pρ P = (0),
so, p ⊆ Ann(x). But x ∈ P implies Ax ⊆ P and P is p-coprimary; consequently, Ax is also p-coprimary. It
follows that p p
Ann(Ax) = Ann(x) = p.
p
So, we get p ⊆ Ann(x) ⊆ Ann(x) = p.
In all of the following corollaries, A is a noetherian ring and M is a f.g. A-module. By taking N = M in
Theorem 3.99, we get:
232 CHAPTER 3. COMMUTATIVE RINGS

Corollary 3.100 The primes of Ann(M ) are in Ass(M ).

Corollary 3.101 Say 0 −→ N −→ M −→ M/N −→ 0 is exact. Then,

Ass(M ) ⊆ Ass(N ) ∪ Ass(M/N ).

Proof . Pick p ∈ Ass(M ) and say p ∈ / Ass(N ). By Theorem 3.99, there is some x ∈ M so that p = Ann(x).
Look at (Ax) ∩ N . We claim that (Ax) ∩ N = (0). If not, (Ax) ∩ N ⊆ Ax and Ax is p-coprimary. Thus,
(Ax) ∩ N is also p-coprimary and Ass((Ax) ∩ N ) = {p}, But, (Ax) ∩ N ⊆ N ; so, Ass((Ax) ∩ N ) ⊆ Ass(N ).
It follows that p ∈ Ass(N ), a contradiction.
Therefore, (Ax) ∩ N = (0). Thus, we have

Ax −
g→ Ax/((Ax) ∩ N ) ,→ M/N,

which means that Ax is a submodule of M/N . By Theorem 3.99, we have p ∈ Ass(M/N ).

Corollary 3.102 We have Ass(M ) ⊆ Supp(M ).

Proof . If p ∈ Ass(M ), then p = Ass(Ax), for some x ∈ M , i.e., we have the inclusion A/p −→ M . By
localizing, we get (A/p)p ⊆ Mp , yet
(A/p)p = Frac(A/p) 6= (0).
Thus, Mp 6= (0), i.e., p ∈ Supp(M ).

Corollary 3.103 Each of our M ’s possesses a chain (of submodules)

(0) = M0 < M1 < M2 < · · · < Mn = M (†)

for which Mj /Mj−1 ∼

= A/pj , for some pj ∈ Spec A. Every p ∈ Ass(M ) appears as at least one of these pj in
each such chain.

Proof . If p ∈ Ass(M ), there is some x ∈ M so that A/p = ∼ Ax ⊆ M . If we let M1 = Ax, it follows that
M1 /M0 ∼ = A/p. Look at M/M1 . If e p ∈ Ass(M/M1 ), repeat the argument to get M2 ⊆ M/M1 with M2 = Ay,
for some y ∈ M/M1 , and A/e p∼= Ay. By the second homomorphism theorem, M2 = M2 /M1 . Then, we have
(0) < M1 < M2 ; M2 /M1 ∼ = A/ep; M1 /M0 ∼= A/p. If we continue this process, we obtain an ascending chain
of the desired type
(0) = M0 < M1 < M2 < · · · < Mn < · · · .
As M is noetherian, this chain stops. This proves the first statement.
We prove the last statement by induction on the length of a given chain.
Hypothesis: If M has a chain, (†), of length n, each p ∈ Ass(M ) appears among the primes from (†).
If n = 1, then M ∼
= A/e
p. As A/e
p is e
p-coprimary, we have Ass(M ) = {e
p}; yet p ∈ Ass(M ), so, p = e
p.
Assume the induction hypothesis holds up to n − 1. Given a chain, (†), of length n and p ∈ Ass(M ), we
know there exists some x ∈ M with p = Ann(x), i.e., we have an inclusion A/p ,→ M . There is some j such
that x ∈ Mj and x ∈ / Mj−1 , where the Mj ’s are in (†). If j < n, then apply the induction hypothesis to
Mn−1 to conclude that p is among p1 , . . . , pn−1 .
So, we may assume x ∈ Mn and x ∈
/ Mn−1 . Look at (Ax) ∩ Mn−1 . There are two cases.
(a) (Ax) ∩ Mn−1 6= (0). Then, (Ax) ∩ Mn−1 ⊆ Ax, where the latter is p-coprimary; it follows that
(Ax) ∩ Mn−1 is p-coprimary and Ass((Ax) ∩ Mn−1 ) = {p}. Yet, (Ax) ∩ Mn−1 ⊆ Mn−1 , so,

Ass((Ax) ∩ Mn−1 ) ⊆ Ass(Mn−1 ).

3.6. PRIMARY DECOMPOSITION 233

Therefore, p is among p1 , . . . , pn−1 , by the induction hypothesis.

(b) (Ax) ∩ Mn−1 = (0). In this case, Ax ∼
= Ax/((Ax) ∩ Mn−1 ) ,→ M/Mn−1 ∼
= A/pn , so,
Ass(Ax) = {p} ⊆ Ass(A/pn ) = {pn }. Therefore, p = pn .
The chain, (†), shows that M is a multiple extension of the “easy” modules A/aj . That is, we have exact
sequences

0 −→ M1 = A/p1 −→ M2 −→ M2 /M1 = A/p2 −→ 0

0 −→ M2 −→ M3 −→ A/p3 −→ 0
..
.
0 −→ Mn−1 −→ M −→ A/pn −→ 0

We define Ext(M/N, N ) as the set

{M | 0 −→ N −→ M −→ M/N −→ 0}/ ∼,

where the equivalence relation ∼ is defined as in the case of group extensions. It turns out that not only is
Ext(M/N, N ) an abelian group, it is an A-module. If the A-modules Ext(A/pj , Mj−1 ) can be successively
computed, we can classify all f.g. A-modules, M .
To attempt such a task, one should note the following:

Remarks:
(1) Say 0 −→ N −→ M −→ M/N −→ 0 is exact, then

Supp(M ) = Supp(N ) ∪ Supp(M/N ).

Proof . Localize at any prime p. We get

0 −→ Np −→ Mp −→ (M/N )p −→ 0 is exact.

From this, (1) is clear.

(2) If M and N are two f.g. modules, then

Supp(M ⊗A N ) = Supp(M ) ∩ Supp(N ).

Proof . We always have

(M ⊗A N )p ∼
= Mp ⊗Ap Np .
So, if p ∈ Supp(M ⊗A N ), the left hand side is nonzero which implies that Mp 6= (0) and Np 6= (0).
Consequently,
Supp(M ⊗A N ) ⊆ Supp(M ) ∩ Supp(N ).
Now, assume p ∈ Supp(M ) ∩ Supp(N ), then Mp 6= (0) and Np 6= (0). As Mp and Np are f.g. Ap -modules
(since M and N are f.g. A-modules), Nakayama’s lemma implies

Mp /mMp 6= (0) and Np /mNp 6= (0).

As these are vector spaces over κ(Ap ), we deduce that

Mp /mMp ⊗Ap Np /mNp 6= (0).

But, this is just (Mp ⊗Ap Np )/m(Mp ⊗Ap Np ); so, Mp ⊗Ap Np 6= (0).
234 CHAPTER 3. COMMUTATIVE RINGS

(3) If M is a f.g. A-module, then p ∈ Supp(M ) iff there exists a chain

(0) = M0 < M1 < M2 · · · < Mn = M (†)

with Mj /Mj−1 ∼
= A/pj and p is one of these pj .
Proof . If we have a chain (†) and p = pj for some j, then A/pj = A/p and (A/p)p = Frac(A/p). Therefore,
(Mj /Mj−1 )p 6= (0). By exactness, (Mj )p 6= (0). As Mj ,→ M and localization is exact, Mp 6= (0) and
p ∈ Supp(M ).
Conversely, if p ∈ Supp(M ), then there is some q ∈ Ass(M ) and p ⊇ q. So, A/q is in a chain and
A/p = (A/q)/(p/q) implies (DX) p is in a chain.

Corollary 3.104 The following are equivalent conditions:

(1) p ∈ Ass(M/N ), for some submodule, N , of M .

(2) p ∈ Supp(M ).

(3) p ⊇ Ann(M ) (p ∈ V (Ann(M ))).

(4) p contains some associated prime of M .

Proof . (1) ⇒ (2). We have p ∈ Ass(M/N ) ⊆ Supp(M/N ) and remark (1) shows that p ∈ Supp(M ).
(2) ⇒ (3). This has already been proved in Section 3.3, Proposition 3.21.
p p Tt
(3) ⇒ (4). If p ⊇ Ann(M ), then p ⊇ Ann(M ). However, Ann(M ) = j=1 pj , where the pj ’s are the
primes associated with Ann(M ). So,
\t Yt
p⊇ pj ⊇ pj ,
j=1 j=1

and it follows that p ⊇ pj , for some j. By Corollary 3.100, we have pj ∈ Ass(M ) and p ⊇ pj , proving (4).
(4) ⇒ (1). Say p ⊇ q and q ∈ Ass(M ). By our theorem, we know that there is some x ∈ M so that
q = Ann(x), i.e., A/q ,→ M . But, p/q ,→ A/q ,→ M . Let N = p/q, then,

A/p ∼
= (A/q)/(p/q) ,→ M/N,

so {p} = Ass(A/p) ⊆ Ass(M/N ).

Corollary 3.105 The minimal elements of Supp(M ) and the minimal elements of Ass(M ) are the same
set.

Proof . Let p ∈ Supp(M ) be minimal. By Corollary 3.104 (4), we have p ⊇ q, for some q ∈ Ass(M ). But
Ass(M ) ⊆ Supp(M ), so, q ∈ Supp(M ). Since p is minimal, we get p = q ∈ Ass(M ). Now, p is minimal in
Supp(M ), so it is also minimal in Ass(M ).
Now, let p ∈ Ass(M ) be minimal. As Ass(M ) ⊆ Supp(M ), we have p ∈ Supp(M ). If p ⊇ q for some
q ∈ Supp(M ), then, by Corollary 3.104 (4), we have q ⊇ e
q, for some e
q ∈ Ass(M ). So, p ⊇ e
q; since p is
minimal, we get p = e
q.

Remark: We saw in Section 3.3 that Supp(M ) is closed in Spec A. In fact, Supp(M ) is a finite (irredundant)
union of irreducible subsets (recall, a set is irreducible iff it is not the union of two proper closed subsets).
In this decomposition, the irreducible components are V (p), for p an isolated prime in Ass(M ) (= a minimal
element of Supp(M )). Thus, the minimal elements of Ass(M ) are exactly the generic points of Supp(M ).
3.6. PRIMARY DECOMPOSITION 235

Corollary 3.106 If A is a noetherian ring, then

[
{x ∈ A | x is a zero divisor in A} = p.
p∈Ass(A)
S
Proof . Say ξ ∈ p∈Ass(A) p, so ξ ∈ p for some p ∈ Ass(A). By Theorem 3.99, we have p = Ann(y), for some
y ∈ A. Clearly y 6= 0 and yξ ∈ yp = (0), so ξ is a zero divisor.
S S
Conversely, pick x ∈
/ p∈Ass(A) p and let S = A − p∈Ass(A) p. We know from previous work that S is
T √
a multiplicative
T set. Now, we have a primary decomposition (0) = q, where q = p ∈ Ass(A). We get
S((0)) = q S(q) and we know that S(q) = q iff S ∩ p = ∅. By definition of S, we conclude that S((0)) = (0).
If xy = 0, as x ∈ S, we get y ∈ S((0)) = (0). Therefore, y = 0 and x is not a zero divisor.
S
Corollary 3.107 Say M = α Mα , for some submodules, Mα , of M . Then,
[
Ass(M ) = Ass(Mα ).
α
S
Proof . Since Mα ⊆ M , we get Ass(Mα ) ⊆ Ass(M ), so, α Ass(Mα ) ⊆ Ass(M ). If p ∈ Ass(M ), then there
is some m ∈ M so that p = Ann(m). But, m ∈ Mα for some α; Theorem 3.99 implies that p ∈ Ass(Mα ).
Corollary 3.108 Given an A-module, M , and any nonempty subset, Φ ⊆ Ass(M ), then there is some
submodule, N , of M so that Ass(N ) = Φ.
Proof . Let Φ = {p1 , . . . , pt }. By proposition 3.97, there are some submodules, Pj , of M so that Ass(Pj ) =
`t `t
{pj }. I claim, the map
` j=1 Pj −→ M is injective and Ass( j=1 Pj ) = Φ. First, consider the case t = 2.
Look at the map P1 P2 −→ P1 + P2 ⊆ M . This is an isomorphism iff P1 ∩ P2 = (0). But, P1 ∩ P2 ⊆ Pj
for j = 1, 2, so, Ass(P1 ∩ P2 ) ⊆ {p1 } and Ass(P1 ∩ P2 ) ⊆ {p2 }; as p1 6= p2 , we conclude that P1 ∩ P2 = (0).
Then, the sequence a
0 −→ P1 −→ P1 P2 −→ P2 −→ 0
`
is exact and split. Consequently, Ass(P1 P2 ) = {p1 , p2 }. For t > 2, we proceed by induction (DX).
Proposition 3.109 If N ⊆ M and N possesses a primary decomposition in M , then
\
RadM (N ) = p.
p∈EssN (M )
p isolated

In fact, the isolated primes of RadM (N ) are just the isolated essential primes of N in M (The hypothesis
holds if A is noetherian and M is f.g.).
p
Proof . As RadM (N ) = RadM/N ((0)) = Ann(M/N ) and EssM (N ) = Ass(M/N ), we may assume that
N = (0). We must show that p \
Ann(M ) = p.
p∈Ass(M )
p isolated
Tt
Now, we have a reduced primary decomposition (0) = j=1 Qj , so
t
\
Ann(M ) = (M −→ (0) = (M −→ Qj ).
j=1

But, (M −→ Qj ) is pj -primary, by previous work, so,

p p t q
\ \ \
Ann(M ) = (M −→ (0) = (M −→ Qj ) = p= p.
j=1 p∈Ass(M ) p∈Ass(M )
p isolated

The rest should be clear.

236 CHAPTER 3. COMMUTATIVE RINGS

3.7 Theorems of Krull and Artin–Rees

We begin with a generalization of the power lemma.

Lemma 3.110 (Herstein’s Lemma) If A is a noetherian ring, A is some ideal, M is a f.g. A-module and
N is a submodule of M , then there is some n >> 0 (depending on A, A, M , N ) so that

An M ∩ N ⊆ AN.

Proof . By reducing modulo AN , we may assume AN = (0) and we must prove An M ∩ N = (0). Let
S = {F ⊆ M | F ∩ N = (0)}. Clearly, S is nonempty and since M is f.g. and A is noetherian, S has a
maximal element, call it F , again. Let m1 , . . . , mt be generators of M and pick a ∈ A. Given mj , for any
n ≥ 0, consider
Fn(j) (a) = (an mj −→ F ) = {x ∈ A | xan mj ∈ F }.
(j)
The Fn (a)’s are ideals of A and we have
(j) (j) (j)
F1 (a) ⊆ F2 (a) ⊆ F3 (a) ⊆ · · · .

By the ACC in A, there is some Nj (a) so that

(j) (j)
FNj (a) (a) = FNj (a)+1 (a), for j = 1, . . . , t.

Let N (a) = max1≤j≤t {Nj (a)}. I claim that aN (a) M ⊆ F .

Of course, if we show that aN (a) mj ∈ F for j = 1, . . . , t, we will have proved the claim.
If the claim is false, there is some j so that aN (a) mj ∈/ F . Then, F + AaN (a) mj > F , and by maximality
N (a)
of F , we must have (F + Aa mj ) ∩ N 6= (0). So, there is some f ∈ F and some α ∈ A so that

0 6= f + αaN (a) mj ∈ N. (†)

If we multiply (†) by a, we get

af + αaN (a)+1 mj ∈ aN = (0),
since AN = (0) and a ∈ A. Thus, αaN (a)+1 mj = −af ∈ F , and so,
(j) (j)
α ∈ (aN (a)+1 mj −→ F ) = FN (a)+1 (a) = FN (a) (a).

αaN (a) mj ∈ F ; so, f + αaN (a) mj ∈ F , which means that F ∩ N 6= (0), a contradiction.
It follows that p
Therefore, a ∈ (M −→ F ); as A is f.g., by the power lemma, we get Aρ M ⊆ F . Thus, finally,
Aρ M ∩ N ⊆ F ∩ N = (0).

Theorem 3.111 (Krull Intersection Theorem) Say A is a noetherian ring, M is a f.g. A-module and A is
an ideal of A. Write S = 1 − A (= {1 − α | α ∈ A}). Then,
\
An M = S(0) = Ker (M −→ S −1 M ).
n≥0
T
Proof . Write N = An M . By Herstein’s lemma there exists ρ > 0 so that Aρ M ∩N ⊆ AN . But, N ⊆ Aρ M ,
so Aρ M ∩ N = N and it follows that N ⊆ AN . Of course, we get AN = N . Now, N is f.g., say n1 , . . . , nt
are some generators. As AN = N , there exist some αij ∈ A so that
t
X
nj = αij ni , for j = 1, . . . , t.
i=1
3.7. THEOREMS OF KRULL AND ARTIN–REES 237

Pt
Therefore, 0 = i=1 (αij − δij )ni , for j = 1, . . . , t; so, the matrix (δij − αij ) kills the vector (n1 , . . . , nt ). By
linear algebra, if ∆ = det(δij − αij ) ∈ A, then

∆nj = 0, for j = 1, . . . , t.

(This can be seen as follows: If T is the linear map given by the matrix (δij − αij ), then by the Cayley–
Hamilton theorem, χ(T ) = T t + β1 T t−1 + · · · + βt−1 T + βt I = 0. But, βt = ±∆ and if we apply χ(T ) to
(n1 , . . . , nt ), then χ(T ) and all the nonnegative powers of T kill it. Consequently, βt I(n1 , . . . , nt ) = 0.) Now,
∆ = 1 − d, for some d ∈ A. Thus, ∆ ∈ S. For all j, we have nj ∈ S(0), so N ⊆ S(0). On the other hand, if
ξ ∈ S(0), then there is some s ∈ S with sξ = 0. Yet, s = 1 − α, for some α ∈ A. Thus, (1 − α)ξ = 0, i.e.,
ξ =T αξ. An immediate induction yields ξ = αn ξ, for all n ≥ 0. However, αn ξ ∈ An M , for every n ≥ 0, so
ξ ∈ An M ; this proves that S(0) ⊆ N .
T
Corollary 3.112 Under the hypotheses of Theorem 3.111, if A ⊆ J (A), then An M = (0).

Proof . Since S = 1 − A ⊆ 1 − J (A) ⊆ units of A, we get S(0) = (0).

Corollary
T n 3.113 (Original Krull theorem) If A is local noetherian and m is its maximal ideal, then
m = (0).

Proof . As A is local, m = J (A); the result follows from Corollary 3.112 applied to M = A.

Corollary 3.114 Say X is a real or complex manifold and x ∈ X. Write OX,x for the local ring of germs
of C ∞ -functions at x. Then, OX,x is never noetherian.

Proof . As the question is local on X, we may assume X is an open ball in Rn and x = 0 in this ball (with
n even in case of a complex manifold). Let

f (x) = e
−1/(x,x)
for x ∈ Rn , x 6= 0
0 if x = 0.

(Here, (x, y) is the usual euclidean inner product on Rn .) We have f (x) ∈ C ∞ (ball). Moreover f (n) (0) = 0,
for all n ≥ 0. But, in OX,x , observe that mn consists of the classes of functions
T defined near zero so that the
n-th derivative and all previous derivatives are 0 at the origin. So, germ(f ) ∈ mn ; by the Krull intersection
theorem, our ring OX,x is not noetherian.

A-adic Topologies.
Let A be a ring, A be an ideal in A and M be an A-module. At the origin in M , take as basis of
opens (= fundamental system of neighborhoods at 0) the subsets An M , for n = 0, 1, 2, . . .. Topologise M
by translating these so that {m + An M }n≥0 is a neighborhood basis around m. When M = A, the ring A
receives a topology and A is a topological ring in this topology which is called the A-adic topology. Similarly,
the module M is a topological module in this topology also called the A-adic topology. The A-adic topology
is pseudo-metric, i.e., set  n
n if m ∈ A T M , yet m∈/ An+1 M
ordA (m) = n
∞ if m ∈ n≥0 A M ,
and define
d(m1 , m2 ) = e−ordA (m1 −m2 ) .
Then, we have

(1) d(m1 , m2 ) ≥ 0.

(2) d(m1 , m2 ) = d(m2 , m1 ).

238 CHAPTER 3. COMMUTATIVE RINGS

(3) d(m1 , m3 ) ≤ max(d(m1 , m2 ), d(m2 , m3 )) (ultrametric property).

Yet, it can happen that d(m1 , m2 )T= 0 and m1 6= m2 . The A-adic topology is Hausdorff iff d is a metric
(i.e., d(m1 , m2 ) = 0 iff m1 = m2 )) iff n≥0 An M = (0).
If the A-adic topology is Hausdorff, then we have Cauchy sequences, completeness and completions. The
reader should check: The completion of M in the A-adic topology (Hausdorff case) is equal to
def c
lim M/An M = M . The first person to make use of these ideas was Kurt Hensel (1898) in the case A = Z,
←−
n
M = Q, p = (p), where p is a prime. But here, Hensel used ordp ( rs ) = ordp (r) − ordp (s).
Corollary 3.115 The A-adic topology on a f.g. module M over a noetherian ring is Hausdorff if A ⊆ J (A).
In particular, this holds if A is local and A = mA .
Corollary 3.116 Say A is a noetherian domain and A is any proper ideal (i.e., A 6= A). Then, the A-adic
topology on A is Hausdorff.
Proof . We have S = 1 − A ⊆ nonzero elements of A. Thus, S consists of nonzero divisors. If ξ ∈ S(0), then
sξ = 0, for some s ∈ S, so, ξ = 0. Therefore, S(0) = (0) and the topology is Hausdorff.

Theorem 3.117 (Artin–Rees) Let A be a noetherian ring, A be some ideal, M be a f.g. A-module and N
a submodule of M . Then, there is some k (depending on A, A, M and N ) so that for all n ≥ k,
An M ∩ N = An−k (Ak M ∩ N ).
Proof . Define the graded ring PowA (A) ⊆ A[X], where X is an indeterminate by
a
PowA (A) = An X n
n≥0

= {z0 + z1 X + · · · + zr X r | r ≥ 0, zj ∈ Aj }.
Now, M gives rise to a graded module, M 0 , over PowA (A), namely
a
M0 = An M X n
n≥0

= {z0 + z1 X + · · · + zr X r | r ≥ 0, zj ∈ Aj M }.
Observe that PowA (A) is a noetherian ring. For, if α1 , . . . , αq generate A in A, then the elements of An are
sums of degree n monomials in the αj ’s, i.e., if Y1 , . . . , Yq are independent indeterminates the map
A[Y1 , . . . , Yq ] −→ PowA (A)
via Yj 7→ αj X is surjective, and as A[Y1 , . . . , Yq ] is noetherian, so is PowA (A).
Let m1 , . . . , mt generate M over A. Then, m1 , . . . , mt generate M 0 over PowA (A). Therefore, M 0 is a
noetherian module. Set a
N0 = (An M ∩ N )X n ⊆ M 0 ,
n≥0

a submodule of M . Moreover, N is a homogeneous submodule of M and it is f.g. as M 0 is noetherian.

0 0

Consequently, N 0 possesses a finite number of homogeneous generators: u1 X n1 , . . . , us X ns , where uj ∈

Anj M ∩ N . Let k = max{n1 , . . . , ns }. Given any n ≥ k and any z ∈ An M ∩ N , look at zX n ∈ Nn0 . We have
s
X
zX n = al X n−nl ul X nl ,
l=1
3.7. THEOREMS OF KRULL AND ARTIN–REES 239


where al X n−nl ∈ PowA (A) n−nl
. Thus,

al ∈ An−nl = An−k Ak−nl

and
al ul ∈ An−k (Ak−nl ul ) ⊆ An−k (Ak−nl (Anl M ∩ N )) ⊆ An−k (Ak M ∩ N ).
Ps
It follows that z = l=1 al ul ∈ An−k (Ak M ∩ N ), so

An M ∩ N ⊆ An−k (Ak M ∩ N ).

Now, it is clear that the righthand side is contained in An M ∩ N , as An−k N ⊆ N .

Remark: If we choose n = k + 1 in the Artin–Rees theorem, we get An M ∩ N = A(Ak M ∩ N ) ⊆ AN , hence

a new proof of Herstein’s lemma.

Corollary 3.118 If A is a noetherian ring, A is an ideal, M is a f.g. module and N is a submodule, then
the topology on N induced by the A-adic topology on M is just the A-adic topology on N .

Proof . The induced topology has as neighborhood basis at 0 the sets An M ∩N . By the Artin–Rees theorem,

An M ∩ N = An−k (Ak M ∩ N ) ⊆ An−k N,

for all n ≥ k, for some fixed k. It follows that the induced topology is finer. But, Aρ N ⊆ Aρ M ∩ N , for all
ρ; so, the A-adic topology on N is in its turn finer than the induced topology.
We turn now to two very famous theorems of Wolfgang Krull. Recall that a power of a prime ideal need
not be primary. In the proof of the first of the Krull theorems, the principal ideal theorem, we need to
remedy this situation. We are led to the notion of the symbolic powers, p(n) , of a prime ideal, p.
Let A be a ring and let p ∈ Spec A. Look at Ap = S −1 A, where S = A − p. Take the powers of p, extend
and contract them to and from Ap , to get
def
p(n) = (pn )ec = S(pn ).

Lemma 3.119 The ideal p(n) is always a p-primary ideal.

Proof . The ideal pe is maximal in Ap . Hence, (pe )n is pe -primary, by previous work. But, (pe )n = (pn )e .
Therefore, (pn )e is pe -primary. Now, S ∩ p = ∅, so (pn )ec is p-primary.
Further, we have the descending chain

p ⊇ p(2) ⊇ p(3) ⊇ · · · .

Theorem 3.120 (Krull Principal Ideal Theorem (1928)) If A is a noetherian domain and p ∈ Spec A, then
ht(p) ≤ 1 iff p is an isolated prime of a principal ideal.

Proof . (⇒) (easy part). By hypothesis, ht(p) ≤ 1 and p ⊇ (0); hence, if ht(p) = 0, then p = (0), an isolated
prime of (0). If ht(p) = 1, pick a 6= 0 in p. As p ⊇ (a), the ideal p must contain one of the isolated primes
of (a), say P. So, p ⊇ P > (0), and as ht(p) = 1, we must have p = P.
(⇐) (hard part). Here, we may assume p is an isolated prime of (a), where a 6= 0 (else, if a = 0, then
p = (0) and ht(p) = 0). We must show ht(p) = 1. Hence, we must prove that

if P ∈ Spec A and p > P, then P = (0). (†)

240 CHAPTER 3. COMMUTATIVE RINGS

Step 1. If we localize at p, there is a one-to-one correspondence between primes contained in p and all
primes in Ap . Therefore, we may assume A = Ap , i.e., A is local, p is maximal and p is an isolated prime
of (a), with a 6= 0. We must prove (†). Now, given P ∈ Spec A with p > P, could a ∈ P? If so, we would
have p > P ⊇ (a). As p is isolated, p = P, a contradiction; so, a ∈ / P. It follows that the ring A/(a) has
precisely one prime ideal and it is maximal. Since A is noetherian, by Akizuki’s theorem, A/(a) is artinian
(i.e., it has the DCC).
/ P. Examine the symbolic powers P(n) . We have
Step 2. Pick P ∈ Spec A with P < p. Of course, a ∈
P ⊇ P(2) ⊇ P(3) ⊇ · · · .
I claim this chain stops. To see this, consider the descending chain
P + (a) ⊇ P(2) + (a) ⊇ P(3) + (a) ⊇ · · · .
This chain is in one-to-one correspondence with a chain in A/(a). By step 1, the ring A/(a) has the DCC,
so, there is some n0 so that for all n ≥ n0 ,
P(n) ⊆ P(n+1) + aA.
Given x ∈ P(n) , there is some
p y∈P
(n+1)
and some z ∈ A so that x = y + za. As x − y ∈ P(n) , we have
(n)
za ∈ P ; since a ∈ / P = P , we get z ∈ P(n) . Hence,
(n)

P(n) ⊆ P(n+1) + P(n) a ⊆ P(n+1) + P(n) p.

Read this in the local ring A = A/P(n+1) whose maximal ideal is p. We get

P(n) = P(n) p. (††)

As P(n) is a f.g. A-module, by Nakayama’s lemma, P(n) = (0). Therefore,

P(n) = P(n+1) , for all n ≥ n0 . (∗)
T T e T
(n)
Step 3. By (∗), we get n≥1 P = P(n0 ) . But, (P(n0 ) )e = n≥1 P
(n)
⊆ n≥1 (P
(n) e
) . Consequently,
\ \
(P(n0 ) )e ⊆ (Pn )e = (Pe )n .
n≥1 n≥1

e
However, P is the maximal ideal of A, so by the Krull intersection theorem, the righthand side is (0).
Therefore,
(Pe )n0 = (Pn0 )e = (0).
But, A is an integral domain, therefore, Pe = (0); so, P = (0), as contended.
Now, consider the case where A is just a ring (not necessarily an integral domain).
Corollary 3.121 If A is a noetherian ring and p is an isolated prime of some (a) ⊆ A, then ht(p) ≤ 1.
Proof . Now, p is an isolated prime of some (a) ⊆ A. If a = 0, then p is a minimal prime, i.e., ht(p) = 0.
Therefore, we may assume a 6= 0. Suppose ht(p) ≥ 2, then we must have a chain
p > q > q0 .
Look in A = A/q0 , a noetherian domain. Here, we have
p > q > (0) = q0 , (∗∗)
yet, p is an isolated prime of (a), so the theorem in the domain case implies that ht(p) = 1, contradicting
(∗∗).
To prove the next and last Krull theorem, we need the chain detour lemma:
3.7. THEOREMS OF KRULL AND ARTIN–REES 241

Lemma 3.122 (Chain detour lemma) Say A is a noetherian ring and

p0 > p1 > · · · pm−1 > pm

is a given chain in Spec A. Given a finite set of primes S = {q1 , . . . , qt }, suppose p0 6⊆ qi , for i = 1, . . . , t.
Then, there exists an alternate chain (the detour)

p0 > pe1 > · · · e

pm−1 > pm

so that no pei is contained in any qj .

Proof . Say the lemma is known when m = 2, i.e., given a chain p0 > p1 > p2 , we can change p1 . Given our
chain
p0 > p1 > · · · pm−1 > pm
and the set S, we can replace p1 by pe1 with pe1 6⊆ qi for i = 1, . . . , t. But, then, we have the chain

pe1 > p2 > · · · pm−1 > pm

and we can use induction to obtain the desired chain.

Thus, we are reduced to the main case: p0 > p1 > p2 . Now, p0 > p2 and p0 6⊆ qj for j = 1, . . . , t. By the
prime avoidance lemma,
[t
p0 6⊆ p2 ∪ qj .
j=1

Hence, there is some x ∈ p0 so that x ∈

/ p2 and x ∈
/ qj for j = 1, . . . , t. Look in A = A/p2 , a noetherian
domain. In A, we have
p0 > p1 > p2 = (0)
and so, ht(p1 ) ≥ 2. Now, x ∈ p0 and it follows that some isolated prime of x, say B, is contained in p0 .
As x ∈ / p2 , we have x 6= 0 and B is an isolated prime of x; by the principal ideal theorem, ht(B) = 1. As
ht(p0 ) ≥ 2, we have p0 > B > (0) and x ∈ B. Let pe1 be the inverse image of B in A. We get:
(1) p0 > pe1 > p2 .
(2) x ∈ pe1 ; x ∈
/ qj , for j = 1, . . . , t.
(3) pe1 6⊆ qj , for j = 1, . . . , t.

Theorem 3.123 (Krull Height Theorem (1928)) If A is an ideal of the noetherian ring, A, suppose A is
generated by r elements and p is an isolated prime of A. Then ht(p) ≤ r.

Proof . We proceed by induction on r. Hypothesis: The theorem holds for all isolated primes, p, of A and
all A generated by at most r elements.
The principal ideal theorem yields the cases r = 0, 1. Next, let A = (x1 , . . . , xr ) and B = (x1 , . . . , xr−1 ).
If A = B, there is nothing to prove. Thus, we may assume that xr ∈ / B. If p (some isolated prime of A)
is an isolated prime of B, the induction hypothesis implies ht(p) ≤ r − 1. So, we may assume that p is an
isolated prime of A, not an isolated prime of B and xr ∈ / B (obviously, A 6= B). Let S = {q1 , . . . , qt } be the
finite set of isolated primes of B, let p = p0 and look at some chain

p = p0 > p1 > · · · pm−1 > pm

of Spec A, so that ht(p0 ) ≥ m. If p0 ⊆ qj , then

B ⊆ A ⊆ p0 ⊆ qj ,
242 CHAPTER 3. COMMUTATIVE RINGS

contradicting the fact that p0 is not an isolated prime of B. Therefore, p0 6⊆ qj , for j = 1, . . . , t, and by the
detour lemma, there is a chain of the same length
p0 > pe1 > · · · e
pm−1 > pm
so that no pei is contained in any qj . Our goal is to show that m ≤ r. let A = A/B. Then A becomes
principal (Axr ) in A and as p0 is an isolated prime of A, the principal ideal theorem in A implies ht(p0 ) = 1.
(ht(p0 ) > 0 because p0 is not an isolated prime of B).
Now, p0 ⊇ e
pm−1 and p0 ⊇ B, so,
p0 ⊇ e
pm−1 + B.
Then, observe that p0 ⊇ epm−1 + B and as B ⊆ qi , for all i and e pm−1 6⊆ qi , for all i, we have e
pm−1 + B 6⊆ qi ,
for all i; thus, e
pm−1 + B 6⊆ qi , for all i (here, the qi are the isolated primes of (0) in A, i.e., those of height
0 in A).
Claim. The ideal p0 is an isolated prime of e
pm−1 + B.
As p0 ⊇ epm−1 + B, we find p0 ⊇ m, where m is some isolated prime of e pm−1 + B. If p0 6= m, then as
ht(m) ≥ 1 (because epm−1 + B 6⊆ qi , for all i) we’d see that ht(p0 ) ≥ 2. But, ht(p0 ) = 1, a contradiction.
Therefore, p0 = m, as claimed.
Now, let A = A/e
pm−1 . As p0 ⊇ e
pm−1 + B, we get

p0 ⊇ e
pm−1 + B = B.

Moreover, as p0 is an isolated prime of epm−1 + B, we see that p0 is an isolated prime of e

pm−1 + B = B.
But, the number of generators of B is at most r − 1. If we apply the induction hypothesis to A, we get
ht(p0 ) ≤ r − 1. Finally, by applying double bar to our detored chain, we get

p0 > pe1 > · · · > e

pm−2 > (0),
a chain of length m − 1. Therefore, m − 1 ≤ r − 1, that is, m ≤ r.
Corollary 3.124 In a noetherian ring, the prime ideals satisfy the descending chain condition. In particular,
every prime ideal contains a minimal prime.
Proof . Given a prime, p, it is finitely generated, say by r elements. Therefore, ht(p) ≤ r and any descending
chain starting at p must stop.
Corollary 3.125 If A is a noetherian ring, then for every p ∈ Spec A, the Krull dimension, dim(Ap ), is
finite.
Corollary 3.126 Say A is noetherian, a 6= 0 is any given element in A and p is an isolated prime of Aa.
Then, every prime ideal, q, strictly contained in p is an isolated prime of (0), i.e., consists of zero-divisors.
Proof . By the principal ideal theorem, ht(p) ≤ 1, and ht(p) = 1, as q < p. It follows that ht(q) = 0, which
means that q is an isolated prime of (0).
Proposition 3.127 (Converse of the height theorem) Let A be a noetherian ring. For every p ∈ Spec A, if
ht(p) ≤ r, then there is some ideal, A, of A generated by at most r elements and p is an isolated prime of A.
Proof . (DX).

3.8 Further Readings

There is a vast literature on commutative rings and commutative algebra. Besides some of the references
already given in Section 2.9, such as Atiyah MacDonald [3], Lafon [32, 33], Eisenbud [13], Matsumura [39],
Malliavin [38], let us mention Bourbaki [6, 8, 7] Zariski and Samuel [50, 51], Jacobson [28] and Serre [46].
Chapter 4

Fields and Galois Theory

4.1 Introduction
The rational, real, complex and, much later, the finite fields were the basic inspiration for the study of fields
in general. Their ideal theory and the module theory (vector spaces) over them are very simple; so, it was
natural to look more deeply inside them. In particular, one can consider solutions of polynomial equations
in a field, the automorphisms of a field, the relation of one field to another. We owe to E. Galois the capital
idea of applying symmetry in the form of group theory to the study of polynomial equations (coefficients
in a field) and their solutions in a (perhaps bigger) field. He was preceded in partial results by such figures
as Lagrange, Abel and Gauss and the impetus he provided has sustained the subject until the current day.
What concerns one now is not so much the “classical theory” (all of which in smooth modern form is treated
below), but questions of basically geometric origin that use an admixture of group theory, ring theory and
fields to try to settle vexing questions of apparently “simple” nature. For example, if we adjoin to the
rationals all the roots of unity and call the resulting field K, is it true that every homogeneous form of
degree d > 0 in more than d variables has a non-zero solution in K? This is a conjecture of E. Artin–still
open at present.

4.2 Algebraic Extensions

Recall that if A is a commutative ring and B is an over-ring of A (i.e., an A-algebra), an element β ∈ B is
algebraic over A iff the map A[X] −→ A[β] ⊆ B is not injective; the element β is transcendental over A iff the
map A[X] −→ A[β] is injective. Moreover, β1 , . . . , βn are independent transcendentals over A (algebraically
independent over A) iff A[X1 , . . . , Xn ] −→ A[β1 , . . . , βn ] is injective. The case of interest here is: A = k, a
field, and B a subring of a field.
Algebraic elements admit of many characterizations:

Proposition 4.1 Say B is an integral domain containing a field k and α ∈ B. Then, the following are
equivalent:

(1) α is algebraic over k.

(2) k[α] (⊆ B) is a field.

(3) k(α) = k[α].

(4) 1/α ∈ k[α].

(5) k[α] (⊆ B) is a finite dimensional k-vector space.

243
244 CHAPTER 4. FIELDS AND GALOIS THEORY

(6) k[α] ⊆ L, where L (⊆ B) is a subring of B and L is a finite dimensional k-vector space.

Proof . (1) ⇒ (2). By definition there is some polynomial f ∈ k[X] so that f (α) = 0. QBy unique factorization
r
in k[X], we know that f = f1 · · · fr , where each fj is irreducible. So, 0 = f (α) = j=1 fj (α) and as B is
a domain, fj (α) = 0, for some j; so, we may assume that f is irreducible. Look at k[X]/(f (X)). Now, as
k[X] is a P.I.D and f is irreducible, it follows that (f (X)) is a maximal ideal. Thus, k[X]/(f (X)) is a field;
moreover, k[α] ∼= k[X]/(f (X)) and (2) holds.
(2) ⇒ (3) and (3) ⇒ (4) are clear.
(4) ⇒ (5). By (4),
X N
1
= aj αj
α j=0
PN
(with αN 6= 0) and this yields j=0 aj αj+1 = 1; we deduce

N
X −1
1 aj j+1
αN +1 = − α ,
aN j=0 N
a

i.e., αN +1 depends linearly on 1, α, . . . , αN . By an obvious induction, αN +i depends linearly on 1, α, . . . , αN

for all i ≥ 1 and so, 1, α, . . . , αN span k[α].
(5) ⇒ (6) is a tautology.
(6) ⇒ (1). Since k[α] is a subspace of a finite dimensional vector space, k[α] is finite dimensional over k
(i.e., (5)). Look at 1, α, . . . , αN , αN +1 , . . . There must be a linear dependence, so

aN α N + · · · + a1 α + a0 = 0

and α is a root of f (X) = aN X N + · · · + a1 X + a0 .

Proposition 4.2 Write Balg = {α ∈ B | α is algebraic over k}. Then, Balg is a ring (a domain).

Proof . Say α, β ∈ Balg . Then, k[α] is finite dimensional over k and k[α, β] = k[α][β] is finite dimensional
over k[α], which implies that k[α, β] is finite dimensional over k. As α ± β and αβ belong to k[α, β], by (6),
they are algebraic over k.

Proposition 4.3 Say α, β ∈ Balg (with β 6= 0), then α/β ∈ Balg . Therefore, Balg is actually a field.

Proof . As before, k[α, β] is finite dimensional over k[α]. But, k(α) = k[α] and k[α, β] = k[α][β], so k[α, β] =
k(α)[β]. Yet, β is algebraic over k(α); thus, k(α)[β] = k(α)(β) = k(α, β). Consequently, k[α, β] = k(α, β)
and it is finite dimensional over k. As α/β ∈ k(α, β), it is algebraic over k.

Proposition 4.4 Being algebraic is transitive.

Given an extension, K/k, the degree, deg(K/k) = [K : k], of K/k is the dimension of K as a vector
space over k. Observe that if [K : k] is finite, then K is algebraic over k (for every α ∈ K, there is a linear
dependence among 1, α, . . . , αn , . . ., so, α is the root of some polynomial in k[X]). However, an algebraic
extension K/k need not be finite.

Definition 4.1 Let K/k be a field extension (i.e., k ⊆ K where both are fields and K is a k-algebra). Say
α ∈ K is a root of f (X) ∈ k[X]. Then, α is a root of multiplicity, m, iff f (X) = (X − α)m g(X) in K[X] and
g(α) 6= 0.
4.2. ALGEBRAIC EXTENSIONS 245

Let A be a commutative ring, B be an A-algebra and C be a B-algebra.

Definition 4.2 An additive map δ : B → C is an A-derivation of B with values in C iff

(1) δ(ξη) = ξδ(η) + δ(ξ)η (Leibnitz)
(2) δ(α) = 0 whenever α ∈ A.

Notice that (1) and (2) imply the A-linearity of an A-derivation. The A-derivations of B with values in
C form a B-module denoted DerA (B, C).

Examples of Derivations.
(1) Let A be a commutative ring, let B = A[X] and let C = B.
X
N  XN
δf = δ aj X j = jaj X j−1 = f 0 (X)
j=0 j=0

is an A-derivation.
(2) Let A be a commutative ring, B = A[{Xα }α∈I ], C = B and
∂
δα = .
∂Xα

Remark: For Example 1, if h is an independent transcendental from X, we have (DX)

f (X + h) = f (X) + f 0 (X)h + O(h2 ).

Theorem 4.5 (Jacobian criterion for multiplicity) Given f (X) ∈ k[X] and K/k a field extension, for any
root α of f (X), we have:
(1) If the multiplicity of α as a root is ≥ m, then

f (α) = f 0 (α) = · · · = f (m−1) (α) = 0.

(2) If char(k) = 0 and if f (α) = f 0 (α) = · · · = f (m−1) (α) = 0 but f (m) (α) 6= 0, then α is a root of f of
exact multiplicity m.

Proof . We proceed by induction on m. Consider a root, α, of multiplicity 1. This means f (X) = (X −α)g(X)
in K[X] and g(α) 6= 0. Thus,
f 0 (X) = (X − α)g 0 (X) + g(X),
so, f 0 (α) = g(α) and f 0 (α) 6= 0. Therefore, (2) holds independently of the characteristic of k in this one case
and (1) is trivial.
Now, assume α is a root of multiplicity at least m. As f (X) = (X − α)m g(X) in K[X], we get

f 0 (X) = (X − α)m−1 ((X − α)g 0 (X) + mg(X)),

which shows that the multiplicity of α in f 0 is at least m − 1. By the induction hypothesis applied to f 0 (X),
we have f 0 (α) = f 00 (α) = · · · = f (m−1) (α) = 0. Also, f (α) = 0, so (1) holds.
(2) Again, we proceed by induction. Assume that f (α) = f 0 (α) = · · · = f (m−1) (α) = 0 but f (m) (α) 6= 0,
with char(k) = 0. Let q be the exact multiplicity of α. Then, f (X) = (X − α)q h(X) in K[X], with h(α) 6= 0.
Now, f 0 (α) = (f 0 )0 (α) = · · · = (f 0 )(m−2) (α) = 0 and the induction hypothesis applied to f 0 (X) shows that
246 CHAPTER 4. FIELDS AND GALOIS THEORY

α is a root of exact multiplicity m − 1 of f 0 . So, f 0 (X) = (X − α)m−1 g(X), with g(α) 6= 0. We know that α
is a root of multiplicity q of f , so by (1), f (α) = f 0 (α) = · · · = f (q−1) (α) = 0. If q > m, then q − 1 ≥ m, so
f (m) (α) = 0, a contradiction. Thus, q ≤ m. As

f 0 (X) = (X − α)q−1 ((X − α)h0 (X) + qh(X)),

we have
(X − α)m−1 g(X) = (X − α)q−1 ((X − α)h0 (X) + qh(X)),
and since q ≤ m, we get
(X − α)m−q g(X) = (X − α)h0 (X) + qh(X).
If we let X = α, we have qh(α) 6= 0, as h(α) 6= 0 and char(k) = 0; but then, the left hand side must not be
zero, and this implies m = q.

Proposition 4.6 Say f ∈ k[X] (k = a field), then there is an extension K/k of finite degree and an element
θ ∈ K so that f (θ) = 0. If e
k isanother field and µ : k → e
k is an isomorphism of fields, write fe ∈ e k[X] for
P P
the image of f under µ (i.e., µ gj Xj = µ(gj )X ), then f is irreducible over k[X] iff fe is irreducible
j

over e
k[X]. Let θ be a root of an irreducible polynomial, f (X), in some extension K/k and let θe be a root of
f in some extension Ω/e
e k. Then, there exists a unique extension of µ to a field isomorphism k(θ) −→ e e
k(θ),
e
so that µ(θ) = θ.

Proof . Factor f into irreducible factors in k[X], then a root of an irreducible factor is a root of f , so we
may assume that f is irreducible. Now, the ideal (f (X)) is maximal in k[X]. Therefore, K = k[X]/(f (X))
is a field and X = the image of X in K is θ, a root, and [K : k] = deg(f ) < ∞.

Next, we have µ : k → e
k and f ∈ k[X]. Of course,

k[X] ∼
= k ⊗Z Z[X] ∼
=ek ⊗Z Z[X] ∼
=ek[X],

so f is irreducible iff fe is irreducible. Now, θ ∈ K is a root of an irreducible polynomial, f , and θe ∈ Ω is a

µ
root of an irreducible polynomial fe. But, k(θ) ∼ = k[X]/(f (X)) −→ e k[X]/(fe(X)) ∼=e e As θ generates k(θ)
k(θ).
over k, the element µ(θ) determines the extension of µ to k(θ).

Proposition 4.7 Say k is a field, f ∈ k[X] and K/k is a field extension. Then, f possesses at most deg(f )
roots in K counted with multiplicity and there exists an algebraic extension L/k (in fact, [L : k] < ∞) where
f has exactly deg(f ) roots counted with multiplicity.

Proof . We use induction on deg(f ). If α ∈ K/k is a root of f , then in K[X], we have

f (X) = (X − α)g(X), where g(X) ∈ K[X]. (∗)

But, deg(g) = deg(f ) − 1, so there exist at most deg(f ) − 1 roots of g in the field, K, containing k. If β
is a root of f , either β = α or g(β) = 0 as K is a domain. Then, the first statement is proved. The last
statement is again proved by induction. In the above, we can take K = k(α), of finite degree over k. Then,
induction and (∗) imply our counting statement.

Corollary 4.8 (of the proof ) The degree [K : k] of a minimum field containing all deg(f ) roots of f always
satisfies [K : k] ≤ deg(f )!.

Remarks:
4.2. ALGEBRAIC EXTENSIONS 247

(1) Proposition 4.7 is false if K is a ring but not a domain. For example, take
Y Y Y
K=k k ··· k.
| {z }
n

Then, if ej = (0, . . . , 0, 1, 0, . . . , 0) with 1 in the j-th place, each ej solves X 2 = X.

(2) Let K = k[T ]/(T 2 ). The elements α = λT ∈ K all satisfy X 2 = 0. If k is infinite, there are infinitely
many solutions.
(3) Let k = R and K = H (the quaternions). We know that H is a division ring, i.e., every nonzero element
has a multiplicative inverse. Consider the equation X 2 + 1 = 0. Then, every α = ai + bj + ck with
a2 + b2 + c2 = 1 satisfies our equation!
(4) Given a field, k, there exists a field extension K/k having two properties:
(a) K/k is algebraic (but in general, [K : k] = ∞).
(b) For every f ∈ K[X], there exists θ ∈ K so that f (θ) = 0.
We’ll prove these facts at the end of the Chapter in Section 4.11
Such a field, K, is called an algebraic closure of k and if only (2) holds, K is called algebraically closed .
The field K is unique up to noncanonical isomorphism. The usual notation for an algebraic closure of k is k.
248 CHAPTER 4. FIELDS AND GALOIS THEORY

4.3 Separable Extensions, Kähler Differentials,

Mac Lane’s Criterion
Definition 4.3 An algebraic element α over a field k (i.e., α ∈ K is algebraic over k for some field extension
K/k) is separable over k iff α is a simple root of its minimal k-polynomial.1 A polynomial, f , is separable iff
all its irreducible factors are distinct and separable, and an irreducible polynomial is separable if it has one
(hence all) separable roots. The field extension K/k is separable iff all α ∈ K are separable over k. We use
the adjective inseparable to mean not separable.

Proposition 4.9 Suppose α is inseparable over k. Then, char(k) = p > 0. If f is the minimal polynomial
n
for α, then there is some n ≥ 1 and some irreducible polynomial g(X) ∈ k[X] so that f (X) = g(X p ). If we
choose n maximal then

(1) g(X) is a separable polynomial and

n n
(2) αp is separable over k. Any root β of f has the property that β p is separable over k.

Proof . The element α is inseparable iff f 0 (α) = 0 by the n = 1 case of the Jacobian criterion. Thus, f
Pd Pd−1
divides f 0 , yet deg(f 0 ) < deg(f ). Therefore, f 0 ≡ 0. If f (X) = j=0 aj X j , then f 0 (X) = j=0 jaj X j−1
and it follows that jaj = 0, for all j. If char(k) = 0, then aj = 0 for all j 6= 0 and f ≡ 0, as α is a root.
Thus, we must have char(k) = p > 0 and if p does not divide j, then aj = 0. We deduce that
e
X
f (X) = apr X pr = h1 (X p ),
r=0
Pe
where h1 (X) = r=0 apr X r . Note that h1 must be irreducible and repeat the above procedure if necessary.
As deg(h1 ) < deg(f ), this process must stop after finitely many steps. Thus, there is a maximum n with
n
f (X) = g(X p ) and g(X) is irreducible in k[X]. Were g(X) inseparable, the first part of the argument would
n+1
imply that g(X) = h(X p ) and so, f (X) = h(X p ), contradicting the maximality of n. Therefore, g(X) is
pn pn
separable. Yet, g(α ) = f (α) = 0, so α is a root of an irreducible separable polynomial and (2) holds.
n
Given β, we have β p again a root of g.

Definition 4.4 A field k of characteristic p > 0 is perfect iff k = k p , i.e., for every λ ∈ k, the element λ has
a p-th root in k.

Examples of Perfect and Imperfect Fields.

(1) Fp = Z/pZ, where p is prime, is perfect.
(2) Any finite field is perfect.
(3) The field k(T ), where char(k) = p > 0 is always imperfect.

/ k p (with c ∈ k), then for

Proposition 4.10 If k is a field with characteristic char(k) = p > 0 and if c ∈
pn
every n ≥ 0, the polynomial f (X) = X − c is irreducible in k[X]. Conversely, if for some n > 0 the
n
polynomial X p − c is irreducible, then c ∈/ kp .
n n
Proof . Look at f (X) = X p − c and pick a field, K, with a root, α ∈ K, of f . Then, αp − c = 0, so
n n n
f (X) = X p − αp = (X − α)p , since char(k) = p > 0. Say ϕ(X) ∈ k[X] is an irreducible factor of f (X),
then ϕ(X) | f (X) in k[X], and similarly in K[X]. By unique factorization in K[X], we have ϕ(X) = (X −α)r ,
n
for some r > 0, where αp − c = 0 and α ∈ K.
1 Recall that the minimal k-polynomial of α is the monic polynomial of minimal degree generating the principal ideal

consisting of the polynomials in k[X] that vanish on α.

 4.3. SEPARABLE EXTENSIONS, KÄHLER DIFFERENTIALS, MAC LANE’S CRITERION 249

n
Claim: X p − c is a power of ϕ(X).
n
If not, there is some irreducible polynomial, ψ(X), relatively prime to ϕ(X) and ψ(X) | X p − c in
k[X] (DX). Then, there exist s(X), t(X) with s(X)ϕ(X) + t(X)ψ(X) = 1 in k[X]. However, ψ(X) divides
n
X p − c, so ψ(α) = 0. If we let X = α, we get 1 = s(α)ϕ(α) + t(α)ψ(α) = 0, a contradiction.
n
Therefore, ϕ(X)l = X p − c. It follows that rl = pn , so r = pa and l = pb with a + b = n. Then,
a a a
ϕ(X) = (X − α)r = (X − α)p = X p − αp ,
a a b b
which implies αp ∈ k. But then, c = (αp )p ∈ k p , a contradiction if b ≥ 1. Thus, b = 0 and consequently,
a = n and f (X) = ϕ(X) is irreducible.
n
Conversely, if for some n > 0 the polynomial X p − c is irreducible and if c ∈ k p , then c = bp , for some
b ∈ k. It follows that
n n n−1
X p − c = X p − bp = (X p − b)p
n
contradicting the irreducibility of X p − c.

Definition 4.5 An element α ∈ K/k is purely inseparable over k (char(k) = p > 0) iff there is some n ≥ 0
n
so that αp ∈ k. Equivalently, α is purely inseparable over k iff the minimal k-polynomial for α is of the
pn
form X − c, for some c ∈ k.

Remark: We have α ∈ k iff α is separable and purely inseparable over k.

Proposition 4.11 If k is a field, then k is perfect iff every algebraic extension of k is separable.

Proof . (⇒). Say k is perfect and pick α ∈ K/k, with α algebraic. We know that α has a minimal k-
n
polynomial f (X) and that f (X) = g(X p ), for some irreducible polynomial, g(X), and some n ≥ 0. We
PN j
PN pn j p p2 n
have g(X) = j=0 bj X , so f (X) = j=0 bj (X ) . As k is perfect, k = k = k = · · · = k p . So,
n
bj = cpj , for some cj ∈ k and we have

N
X n n
X
N  pn
f (X) = cpj (X p )j = cj X j .
j=0 j=0

0
This contradicts the irreducibility of f (X) unless n = 0, and we know that αp = α is separable over k.
(⇐). In this case, all algebraic extensions of k are separable and say k is not perfect. Then, there is some
/ k p . Hence, by Proposition 4.10, the polynomial X p − c is irreducible over k. Let K = k(α)
c ∈ k, with c ∈
where α is some root of X p − c. Then, αp = c ∈ k and it follows that α is purely inseparable over k. But, α
is separable over k, a contradiction, as α ∈/ k.

Corollary 4.12 For a field, k, the following are equivalent:

(1) k is imperfect.

(2) k possesses nontrivial inseparable extensions.

(3) k possesses purely inseparable extensions.

 Say K/k is algebraic and inseparable. It can happen that there does not exist α ∈ K with α purely
inseparable over k.
250 CHAPTER 4. FIELDS AND GALOIS THEORY

To go further, we need derivations and Kähler differentials. Consider the situation where A, B are
commutative rings and B is an A-algebra. On B-modules, we have an endofunctor:
M DerA (B, M ).
Is the above functor representable? This means, does there exist a B-module, ΩB/A , and an element,
d ∈ DerA (B, ΩB/A ), so that functorially in M
θM : HomB (ΩB/A , M ) −
g→ DerA (B, M )?
(Note: For every ϕ ∈ HomB (ΩB/A , M ), we have θM (ϕ) = ϕ ◦ d, see below).

BE
d / ΩB/A
EE
EE
EE
θM (ϕ) EE
ϕ
" 
M
Theorem 4.13 The functor M DerA (B, M ) is representable by a pair (ΩB/A , d), as above.
µ
Proof . Consider B ⊗A B and the algebra map B ⊗A B −→ B, where µ is multiplication, i.e., µ(b ⊗ eb) = beb.
Let I = Ker µ and write I/I 2 = ΩB/A . We let B act on B ⊗A B via the left action b(ξ ⊗ η) = bξ ⊗ η. Then,
ΩB/A is a B-module. Given b ∈ B, set

db = d(b) = (1 ⊗ b − b ⊗ 1) mod I 2 .

Now, for b, eb ∈ B, we have

(1 ⊗ b − b ⊗ 1)(1 ⊗ eb − eb ⊗ 1) ∈ I 2 ,
and we get
1 ⊗ beb + beb ⊗ 1 − (b ⊗ eb + eb ⊗ b) ∈ I 2 .
So, modulo I 2 , the above is zero and
1 ⊗ beb − eb ⊗ b = b ⊗ eb − beb ⊗ 1 in ΩB/A .
Obviously, d is additive and zero on A, so we only need to check the Leibnitz rule. We have
bd(eb) = b(1 ⊗ eb − eb ⊗ 1) mod I 2
= b ⊗ eb − beb ⊗ 1 in ΩB/A
= 1 ⊗ beb − eb ⊗ b in ΩB/A
= 1 ⊗ beb − beb ⊗ 1 + beb ⊗ 1 − eb ⊗ b in ΩB/A
= d(beb) − eb(1 ⊗ b − b ⊗ 1) in ΩB/A .

So, bd(eb) = d(beb) − ebdb in ΩB/A , namely d(beb) = ebdb + bd(eb). The rest of the proof is routine.
Definition 4.6 The B-module ΩB/A (together with the derivation d) is called the module of relative Kähler
differentials of B over A.

Examples of Relative Kähler Differentials.

(1) Let B = A[T1 , . . . , Tn ]. Say D is a derivation of B −→ M trivial on A. So, we know D(T1 ), . . . , D(Tn );
these are some elements in M . Say we are given Tlr ∈ B. Then,
∂
D(Tl · · · Tl ) = rTlr−1 D(Tl ) = (T r )D(Tl ).
| {z } ∂Tl l
r
4.3. SEPARABLE EXTENSIONS, KÄHLER DIFFERENTIALS, MAC LANE’S CRITERION 251

Now,
∂ ∂
D(Tkr Tls ) = Tkr D(Tls ) + Tls D(Tkr ) = Tkr (T s )D(Tl ) + Tls (T r )D(Tk ).
∂Tl l ∂Tk k
In general
n
X
d al ∂
D(T1a1 · · · Tnan ) = T1a1 · · · T an
l · · · Tn (T al )D(Tl ), (†)
j=1
∂Tl l

and
X  X
D α(a) T (a) = α(a) D(T (a) )D(Tl ), (††)
(a)

as D  A ≡ 0. Conversely, (†) on the linear base of monomials in the T1 , . . . , Tn of B gives a derivation.

Therefore,
n
a
DerA (B, M ) −→ M,
i=1

via D 7→ (D(T1 ), . . . , D(Tn )) is a functorial isomorphism. Consequently,

n
a
ΩB/A ∼
= BdTj ,
j=1

where the dTj are A-linearly independent elements of ΩB/A (case M = ΩB/A ).
(2) Let B be a f.g. algebra over A, i.e., B = A[T1 , . . . , Tn ]/(f1 , . . . , fp ). We have

DerA (B, M ) = {ϕ ∈ DerA (A[T1 , . . . , Tn ], M ) | ϕ(fi ) = 0, i = 1, . . . , p}.

But,
Xn Xn
∂fi ∂fi
ϕ(fi ) = ϕ(Tj ) = ϕ(dTj ),
j=1
∂Tj j=1
∂Tj

where ϕ : ΩB/A −→ M (and ϕ = ϕ ◦ d). We let M = ΩB/A to determine it, and we see that

ϕ must kill dfi .

It follows that !
n
a .
ΩB/A = BdTj (submodule df1 = · · · = dfn = 0).
j=1

(3) Let B = C[X, Y ]/(Y 2 − X 3 ) and A = C. From (2) we get

ΩB/A = (BdX q BdY )/(2Y dY − 3X 2 dX).

The module ΩB/A is not a free B-module (due to the singularity at the origin of the curve Y 2 = X 3 ).
(4) Let A = R or C and B = the ring of functions on a small neighborhood of a smooth r-dimensional
manifold (over A). Derivations on B over A have values in B. Let ξ1 , . . . , ξr be coordinates on this neigh-
borhood. Then, ∂/∂ξj is a derivation defined so that

∂f f (. . . , ξj + h, . . .) − f (. . . , ξj , . . .)
= lim .
∂ξj h−→0 h
252 CHAPTER 4. FIELDS AND GALOIS THEORY

Look near a point, we may assume ξ1 = · · · = ξr = 0, there. By Taylor,

Xr
∂f
f (ξ1 + h1 , . . . , ξr + hr ) = f (ξ1 , . . . , ξr ) + hj + O(khk2 ).
j=1
∂ξj

Hence, ΩB/A is generated by dξ1 , . . . , dξr and they are linearly independent over B because the implicit
function theorem would otherwise imply that some ξj is a function of the other ξi ’s near our point, a
contradiction.

Definition 4.7 Given an A-algebra, B, the algebra B is étale over A iff

(1) The algebra B is flat over A.

(2) The algebra B is f.p. as an A-algebra.

(3) ΩB/A = (0).

The algebra B is smooth over A iff (1), (2) and (30 ): ΩB/A is a locally-free B-module, hold.

Remark: Putting aside (2), we see that checking that an algebra is étale or smooth is local on A, i.e., it is
enough to check it for Bp over Ap for every p ∈ Spec A. This is because (DX)

(ΩB/A )p = ΩBp /Ap .

It turns out that smooth means: Locally on A, the algebra B looks like

A ,→ A[T1 , . . . , Tr ] −→ B

where B/A[T1 , . . . , Tr ] is étale.

We can apply the concepts of relative Kähler differentials and étale homomorphisms to field theory. For
this, given a field, write p = char(k) and if p > 0, let k 1/p be the field

k 1/p = {x ∈ k | xp ∈ k}.

Theorem 4.14 (Main theorem on separability (alg. case).) Let K/k be an algebraic extension. Then, in
the following statements: (1) implies any of the others; (2), (2a) and (3) are equivalent; (1) and (4) are
equivalent; all are equivalent if K/k is finite.

(1) The extension K/k is separable.

(2) For all K-modules, M , we have Derk (K, M ) = (0).

(2a) ΩK/k = (0), i.e., when K/k is finite, it is étale.

(3) Every derivation of k to M (where M is a K-vector space) which admits an extension to K (i.e.,
becomes a derivation K −→ M ) admits a unique extension.

When char(k) = p > 0,

(4) Mac Lane I: The natural map k 1/p ⊗k K −→ K 1/p is injective.

(5) Mac Lane II: kK p = K.

In order to prove Theorem 4.14, we first need the following subsidiary statement:
4.3. SEPARABLE EXTENSIONS, KÄHLER DIFFERENTIALS, MAC LANE’S CRITERION 253

Proposition 4.15 If K/k is separable and if M is a K-vector space, then every derivation D : k → M
admits an extension to a derivation of K with values in M .

Proof . We use Zorn’s lemma. Let S be the set of all pairs, (L, DL ), where
(1) L is a subextension of K/k (i.e., k ⊆ L ⊆ K).
(2) DL is an extension of D to L with values in M .
0 0
As (k, D) ∈ S, the set S is nonempty. Define a partial order on S by: (L, D
SL ) ≤ (L , DL ) iff L ⊆ L
0

and DL0  L = DL . The set S is inductive. (If {Lα }α is a chain, then L = α Lα is a field, and define
DL (ξ) = DLα (ξ), where ξ ∈ Lα ; this is well-defined (DX).) By Zorn’s lemma, there exists a maximal
extension, say (L, DL ).
If K 6= L, then there is some β ∈ K with β ∈ / L. Let g(X) ∈ L[X] be the minimum L-polynomial for
β. We try to extend DL to L(β). For this, we must define DL(β) (β) and the only requirement it needs to
satisfy is
0 = DL(β) (g(β)) = g 0 (β)DL(β) (β) + DL (g)(β).
Pr Pr
Here, if g(X) = j=1 aj X j , then DL (g)(α) is j=1 αj DL (aj ) ∈ M . Since β is separable, g 0 (β) 6= 0, so we
can find the value of DL(β) (β), contradicting the maximality of our extension. Therefore, L = K.

Proof of Theorem 4.14. (1) ⇒ (2). Pick D ∈ Derk (K, M ) and α ∈ K; by (1), the element α is separable
over k, i.e., α has a minimal k-polynomial, g(X), so that g(α) = 0 and g 0 (α) 6= 0. As D is a derivation, the
argument of Proposition 4.15 implies that

0 = D(g(α)) = g 0 (α)D(α) + D(g)(α).

But, D(g) = 0, because the coefficients of g are in k and D  k ≡ 0. Since g 0 (α) 6= 0, we get D(α) = 0, i.e.,
(2) holds.
(2) ⇒ (2a). We have the functor M Derk (K, M ) and Derk (K, M ) = (0). By Yoneda’s lemma, the
representing object, ΩK/k , must vanish.
(2a) ⇒ (2). We have Homk (ΩK/k , M ) ∼
= Derk (K, M ) and ΩK/k = (0), so (2) holds.
(2) ⇒ (3). Say D and De are two extensions of the same derivation on k. Then, D − D
e is a derivation
e e e
and (D − D)  k ≡ 0. By (2), (D − D) ∈ Derk (K, M ) = (0), so D − D = 0.
(3) ⇒ (2). Choose D ∈ Derk (K, M ), so D  k ≡ 0. But then, D extends 0 and 0 extends 0; by (3),
D ≡ 0.
(1) ⇒ (5). If α ∈ K, then α is separable over k, so α is separable over kK p (as kK p ⊇ k). Yet, αp ∈ K p ,
so αp ∈ kK p ; thus, α is purely inseparable over kK p . As α is both separable and purely inseparable over
kK p , by a previous remark, α ∈ kK p . This shows K ⊆ kK p . On the other hand, kK p ⊆ K, always.
Therefore, K = kK p , i.e., (5) holds.
Before discussing the equivalence of (4) with (1), we need to elucidate the meaning of the Mac Lane
conditions.
For (5), say {ξλ }λ spans K as a k-vector space. Then, {ξλp }λ spans K p as a k p -space. As k p ⊆ k, {ξλp }λ
spans kK p as a k-space. Hence, Mac Lane II means: If {ξλ }λ spans K as a k-space, so does {ξλp }λ .
For (4), say {ξλ }λ is a linearly independent family (for short, an l.i. family) over k in K. Then, we
know that the elements 1 ⊗ ξλ are linearly independent in k 1/p ⊗k K as k 1/p -vectors (k 1/p acts on the left
on k 1/p ⊗k K). The map k 1/p ⊗k K −→ K 1/p is just
X X
aλ ⊗ ξλ 7→ aλ ξλ .
λ λ
254 CHAPTER 4. FIELDS AND GALOIS THEORY

1/p 1/p
P
P is a linear dependence of the ξλ (in K ) over k , we get
If the map is injective and if there
1/p
λ aλ ξλ = 0,
for some aλ ∈ k . But then, λ aλ ⊗ ξλ would go to zero and by injectivity
X X
aλ ⊗ ξλ = (aλ ⊗ 1)(1 ⊗ ξλ ) = 0
λ λ

in k 1/p ⊗k K. But, {1 ⊗ ξλ }λ is linearly independent in k 1/p ⊗k K, so aλ = 0, for all λ. Consequently, the

family {ξλ }λ is still linearly independent over k 1/p . Conversely (DX), if any l.i. family {ξλ }λ (with ξλ ∈ K)
over k remains l.i. over k 1/p , then our map k 1/p ⊗k K −→ K 1/p is injective. By using the isomorphism
x 7→ xp , we get: Mac Lane I says that any l.i. family {ξλ }λ over k, has the property that {ξλp }λ is still l.i.
over k.
Now, say K/k is finite, with [K : k] = n. Then, ξ1 , . . . , ξn is l.i. over k iff ξ1 , . . . , ξn span K. Condition
(4) implies ξ1p , . . . , ξnp are l.i. and since there are n of them, they span K, i.e. (5) holds. Conversely, if (5)
holds then ξ1p , . . . , ξnp span K and there are n of them, so they are l.i., i.e., (4) holds. Therefore, (4) and (5)
are equivalent if K/k is finite. We can show that (1) and (4) are equivalent (when char(k) = p > 0).
n
(4) ⇒ (1). Pick α ∈ K. We know that αp is separable over k for some n ≥ 0. Further, the minimal
n n
polynomial for β = αp is h(X), where f (X) = h(X p ) and f is the minimal k-polynomial for α. Say,
deg(f ) = d. So, d = pn d0 , with d0 = deg(h). Now, 1, α, . . . , αd−1 are l.i. over k. By (4), repeatedly,
n n n
1, αp , (α2 )p , . . . , (αd−1 )p are l.i., i.e., 1, β, . . . , β d−1 are l.i. Yet, 1, β, . . . , β d0 is the maximum l.i. family
for the powers of β, so d ≤ d0 . This can only happen if n = 0 and α is separable over k.
(1) ⇒ (4). Say {ξλ }λ is l.i. in K/k. As linear independence is checked by examining finite subfamilies, we
may assume that our family is ξ1 , . . . , ξt . We must prove, ξ1p , . . . , ξtp are still l.i. over k. Let L = k(ξ1 , . . . , ξt ),
then L/k is a finite extension. For such an extension, (4) and (5) are equivalent. But, we just proved that
(1) implies (5), so (1) implies (4).
Finally, in the case K/k is finite there remains the proof of (2) ⇒ (1). For this, it is simplest to prove a
statement we’ll record as Corollary 4.16 below. This is:
Corollary 4.16 If α1 , . . . , αt are each separable over k, then the field k(α1 , . . . , αt ) is separable over k. In
particular, if K/k is algebraic and Ksep denotes the set of all elements of K that are separable over k, then
Ksep is a field.

To prove these statements, we will apply Mac Lane II; this will suffice as L = k(α1 , . . . , αt ) is finite over
k. Now kLp = k(α1p , . . . , αtp ) and each αj is therefore purely inseparable over kLp . However, each αj is
separable over k and therefore over kLp . It follows that each αj ∈ kLp so that L = kLp and Mac Lane II
applies. For the proof, proper, that (2) ⇒ (1), assume (2) and that (1) is false. Then Ksep 6= K, so we can
find α1 , . . . , αs ∈ K, each purely inseparable over Ksep , and so that

K = Ksep (α1 , . . . , αs ) > Ksep (α1 , . . . , αs−1 ) > · · · > Ksep (α1 ) > Ksep .
r
Consider the zero derivation on Ksep (α1 , . . . , αs−1 ). Now, β = αsp ∈ Ksep (α1 , . . . , αs−1 ) for some minimal
r > 0, thus to extend the zero derivation to K we need only assign a value to D(αs ) so that
r r−1
D(αsp ) = pr αsp D(αs ) = 0. Any nonzero element of M will do, contradicting (2).

Corollary 4.17 Every algebraic extension of a perfect field is perfect. In particular, every finite field is
perfect and every absolutely algebraic field (i.e., algebraic over a prime field) is perfect.

Proof . If K/k is algebraic and k is perfect, then K/k is separable. By Mac Lane II, we have K = kK p . But,
k = k p (k perfect), so K = k p K p = (kK)p = K p . A finite field is algebraic over Fp and by little Fermat,
Fpp = Fp , i.e., perfect. (Second proof by counting: The map ξ 7→ ξ p is injective, taking Fq to itself. But, the
image has cardinality q; by finiteness, the image is all of Fq .) By the first part of the proof, an absolutely
algebraic field is perfect.
 4.3. SEPARABLE EXTENSIONS, KÄHLER DIFFERENTIALS, MAC LANE’S CRITERION 255

Corollary 4.18 Say α1 , . . . , αt are each separable over k. Then, the field k(α1 , . . . , αt ) is a separable ex-
tension of k. In particular, if K/k is algebraic and we set

Ksep = {α ∈ K | α is separable over k}

then Ksep is a subfield of K/k called the separable closure of k in K.

Corollary 4.19 Say K/k is an algebraic extension and α1 , . . . , αt ∈ K. If each αj is separable over
k(α1 , . . . , αj−1 ), then k(α1 , . . . , αt ) is separable over k. In particular, separability is transitive.

Proof . We use induction on t. When t = 1, this is Corollary 4.18. Assume that the induction hypothesis
holds for t − 1. So, L = k(α1 , . . . , αt−1 ) is separable over k and it is a finite extension, therefore Mac Lane
II yields kLp = L. Let M = k(α1 , . . . , αt ), then M = L(αt ). So, M is separable over L, by the case t = 1.
Therefore, M = LM p , by Mac Lane II. Now,

M = LM p = kLp M p = k(LM )p = kM p .

By Mac Lane II, again, M is separable over k.

Corollary 4.20 If K/k is an algebraic extension, then K is purely inseparable over Ksep .

Corollary 4.21 Pure inseparability is transitive.

 The implication (2) ⇒ (1) does not hold if K/k is not finite. Here is an example: Set k = Fp (T ), where
T is an indeterminate. Define, inductively, the chain of fields

k = k0 < k1 < · · · < kn < · · ·

via the rule

1/p
α0 = T ; αj = αj−1 ; kj = kj−1 (αj ).
S∞
Let K = k∞ = j=0 kj . Then a derivation on K, trivial on k is determined by its values on the αj . Yet, we
p
have αj+1 = αj , therefore D(αj ) = 0 for every j; hence, Derk (K, −) = 0. But, K/k is not separable; indeed
it is purely inseparable.

def
Notation: For a field, k, of characteristic p > 0, set [K : k]s = [Ksep : k], the separable degree of K/k
def
and [K : k]i = [K : Ksep ], the purely inseparable degree of K/k (if K/k is finite, [K : k]i is a power of p).
Clearly,
[K : k] = [K : k]i [K : k]s .
256 CHAPTER 4. FIELDS AND GALOIS THEORY

4.4 The Extension Lemma and Splitting Fields

We begin with a seemingly “funny” notion: Two fields K, L are related , denoted K f
rel L, iff there is some
larger field, W , so that K ⊆ W and L ⊆ W (as sets, not isomorphic copies). This notion is reflexive and
symmetric, but not transitive.

Theorem 4.22 (Extension Lemma) Let K/k be a finite extension and say e k is another field isomorphic to
e e
k via θ : k → k. Suppose Γ is another field related to k, but otherwise arbitrary. Then, there exists a finite
e e e e e
extension, K/ k, with K rel Γ and an extension of θ to an isomorphism θ : K → K.
f

KO
θe / K
e
rel Γ
O f
finite finite

k / e
k f
θ rel Γ

Proof . We proceed by induction on the number, n, of adjunctions needed to obtain K from k.

Case n = 1: K = k(α). Let g(X) ∈ k[X] be the minimum k-polynomial for α. Write ge(X) ∈ e k[X] for
the image, θ(g)(X), of g(X). Of course, ge(X) is k-irreducible. Now, there exists a field, W , with W ⊇ e
e k
and W ⊇ Γ. Thus, ge(X) ∈ W [X]; moreover, there exists an extension W 0 /W of W and some α e ∈ W 0 , so
α) = 0. It follows that e
that ge(e α) ⊆ W 0 and Γ ⊆ W ⊆ W 0 , so e
k(e k(e
α) f
rel Γ. But we know by Proposition 4.6
that θ extends to an isomorphism θe: k(α) → e k(e
α). This proves case 1.
Induction step. Assume that the induction hypothesis holds for all t ≤ n − 1. We have K = k(α1 , . . . , αn )
e and an isomorphism,
and let L = k(α1 , . . . , αn−1 ). By the induction hypothesis, there is a finite extension, L,
0 e
θ : L → L, extending θ;

L(αn ) = K
θe / K
e
O rel Γ
O f

θ0 / L
e
LO rel Γ
O f

k
θ / e
k f
rel Γ

We complete the proof using the argument in case 1 (a single generator), as illustrated in the above diagram.

e e
rel Γ, then there is a k-isomorphism K/k −→ K/k and
Corollary 4.23 If K/k is a finite extension and k f
e
K rel Γ.
f

Proof . This is the case k = e

k; θ = id.

Definition 4.8 A field extension L/k is a splitting field for the polynomial f (X) ∈ k[X] iff L = k(α1 , . . . , αn )
and α1 , . . . , αn are all the roots of f (X) in some larger field (n = deg(f )).

Remarks:
4.4. THE EXTENSION LEMMA AND SPLITTING FIELDS 257

(1) When we view f (X) ∈ L[X], then f (X) splits into linear factors

f (X) = c(X − α1 ) · · · (X − αn )

in L[X], hence the name. Conversely, if M/k is a field extension and in M [X], the polynomial f (X)
splits into linear factors, then M contains some splitting field for f . (Here, f (X) ∈ k[X].)
(2) Suppose L/k and L0 /k are two splitting fields for the same polynomial f (X) ∈ k[X]. Then L = L0 iff
0 0
Lfrel L (L and L are identical , not just isomorphic).

0
Proof . The implication (⇒) is obvious. Conversely, assume L f rel L . Say Ω is a common extension
0
of L and L in which f (X) splits. In Ω, the polynomial f has just n roots, say β1 , . . . , βn . Yet,
L = k(β1 , . . . , βn ) and L0 = k(β1 , . . . , βn ), too. Therefore, L = L0 .

(3) Suppose L/k is a splitting field for f (X) ∈ k[X] and k ∼

=ek via some isomorphism, θ. If fe(X) is the
image of f (X) by θ, and if θ extends to an isomorphism L ∼ =L e e
e for some extension L/ e is a
k, then L
e
splitting field for f (X).

Proposition 4.24 Say f (X) ∈ k[X] and θ : k → e k is an isomorphism. Write fe(X) for the image of f (X)
by θ. Then, θ extends to an isomorphism from any splitting field of f to any splitting field of fe. In particular,
any two splitting fields of f (X) are k-isomorphic (case k = ek; f = fe).

Proof . Apply the extension lemma to the case where K is any chosen splitting field for f and Γ is any
chosen splitting field for fe. The extension lemma yields an extension K/e e
k and an extension θe: K → K e with
e
K f e e e
rel Γ. By Remark (3), the field K is a splitting field for f . By Remark (2), as K and Γ are both splitting
fields and Ke
rel Γ, they are equal.
f
Definition 4.9 An algebraic field extension, M/k, is normal iff for all irreducible k-polynomials, g(X),
whenever some root of g is in M , all the roots of g are in M .

Proposition 4.25 Say M/k is a finite extension and write M = k(β1 , . . . , βt ). Then, the following are
equivalent:
(1) M/k is normal.
(2) M is the splitting field of a family, {gα }α , of k-polynomials (the family might be infinite).
(3) M is the splitting field of a single k-polynomial (not necessarily irreducible).
(4) M is identical to all its k-conjugates; here two fields are k-conjugate iff they are both related and
k-isomorphic.

Proof . (1) ⇒ (2). For each βi , there is an irreducible k-polynomial, say gi with gi (βi ) = 0. By (1), all the
other roots of gi are in M . Therefore, M contains the splitting fields of each gi . But, clearly, M is contained
in the field generated by all these splitting fields. It follows that M is equal to the splitting field of the
(finite) family of k-polynomials g1 , . . . , gt .
(2) ⇒ (3). Say {gα } is the family of k-polynomials for which M is the splitting field. (Note that we may
assume that deg(gα ) > 1 for all gα .) Pick a countable (at most) subset {g1 , g2 , . . . , } of our family. Then,
M contains the splitting field of g1 , call it M1 . We have M ⊇ M1 ⊇ k and [M : M1 ] < [M : k]. If M 6= M1 ,
then M contains the splitting field, M2 , of g1 and g2 , where we may assume that the splitting field of g2 is
distinct from M1 . Thus, we have M ⊇ M2 ⊇ M1 ⊇ k. Since M is finite over k, the above process Qt stops and
we deduce that M is the splitting field of a finite subfamily {g1 , . . . , gt }. Then, take g = i=1 gi , and (3)
holds.
f is a k-conjugate of M , then
(3) ⇒ (4). If M
 258 CHAPTER 4. FIELDS AND GALOIS THEORY

f is a splitting field (k-isomorphic to M )

(a) M

f
(b) M rel M .
f

f = M.
But, we know that (a) and (b) imply that M

(4) ⇒ (1). Pick an irreducible k-polynomial, g, and α ∈ M with g(α) = 0. Consider the extension lemma
in the situation where k = ek and Γ = M . Pick in an algebraic closure, M , of M , any root β of g. We get
the diagram

MO
θe / M
f
rel M
O f

k(α)
θ / e
k(β) f M
O O rel

k k rel M.
f

By the extension lemma applied to the upper portion of the above diagram, there exists M f with Mf
rel M
f
and an extension θe: M → M
f. But, θe  k = θ  k = id, so θe is a k-isomorphism and M
f
f
rel M . By (4), we get
f f
M = M . Since β ∈ M , we have β ∈ M .

Corollary 4.26 Say M ⊇ K ⊇ k and M is normal over k. Then, M is normal over K.

Proof . Use (3), i.e., M is the splitting field of some g ∈ k[X]. Yet, g ∈ K[X], and use (3) again.

 M normal over K and K normal over k does not imply M normal over k.

√
Here is a counter-example
√ to the√transitivity of normality. Let k = Q; K = Q( 2); the extension K/k
is normal. Let α = 2 and L = K( α); again, L/K is normal of degree 2. Observe that L is the splitting
√ over K of √
field X 2 − α ∈ K[X]. But, L/Q is not normal. This is because the polynomial X 4 − 2 has a root,
α, in L, yet i α is not in L because L ⊆ R.

 M normal over k and M ⊇ K ⊇ k does not imply K normal over k.

Corollary 4.27 (SMA, I2 ) Say M is normal over k and g is any irreducible k-polynomial with a root α ∈ M .
Then, a n.a.s.c. that an element β ∈ M be a root of g is that there exists σ, a k-automorphism of M (i.e.,
σ  k = id) so that σ(α) = β.

Proof . (⇐). If α is a root and g ∈ k[X], then

0 = σ(0) = σ(g(α)) = g(σ(α)) = g(β).

So, β is a root.
2 SMA = sufficiently many automorphisms.
4.4. THE EXTENSION LEMMA AND SPLITTING FIELDS 259

(⇒). Say β ∈ M is a root, then there is a k-isomorphism k(α) −→ k(β). Now, k(β) f
rel M ; so, in the
extension lemma, take Γ = M :
MO
θe / M
f
rel M
O f

k(α)
θ / e
k(β) f M
O O rel

k k rel M.
f

We get θe: M → M f, a k-isomorphism and M f f e

rel M . By (4), M = M . So, θ = σ is our required automorphism
f
(it takes α to β).

Corollary 4.28 (SMA, II) Let M be normal over k and say K, K 0 are subextensions of the layer M/k (i.e,
M ⊇ K ⊇ k and M ⊇ K 0 ⊇ k). If θ : K → K 0 is a k-isomorphism, then there is a k-automorphism, σ, of
M so that σ  K = θ.

Proof . Apply the extension lemma with Γ = M to the situation

MO
θe / M
f
rel M
O f

K
θ / K0 f
rel M.

There exist θe and M

f with θe a k-isomorphism and M f f e
rel M . By (4), M = M . Therefore, σ = θ is our
f
automorphism.

Corollary 4.29 Say K/k is a finite extension of degree [K : k] = n, then there exists M ⊇ K with
(1) M is normal over k and
(2) Whenever W is normal over k, W ⊇ K and W f
rel M , then automatically W ⊇ M .

(3) [M : k] ≤ n!.

The field, M , is called a normal closure of K/k.

Proof . (DX).
260 CHAPTER 4. FIELDS AND GALOIS THEORY

4.5 The Theorems of Dedekind and Artin; Galois Groups & the
Fundamental Theorem
Recollect that a K-representation of a group, G, is just a K[G]-module. So, a K-representation of a group,
G, is just a K-vector space plus a (linear) G-action on it (by K-automorphisms); that is, a homomorphism
G −→ Aut(V ). If dimK V < ∞, we have a finite dimensional representation. In this case, Aut(V ) =
GL(V ) ∼ = GLn (K), where n = dimK (V ) is the degree of the representation. Say ρ : G → GLn (K) is
our representation. Then, χρ (σ) = tr(ρ(σ)), the trace of ρ(σ), is a function G −→ K independent of
the basis chosen, called the character of our representation. The case n = 1 is very important. In this
case, the characters are the representations, χρ = ρ. Therefore, we have functions χ : G → K ∗ , with
χ(στ ) = χ(σ)χ(τ ). From now on, we use only one-dimensional characters.

Definition 4.10 Suppose {χα }α is a given family of characters, χα : G → K ∗ , of the group G. Call the
family independent iff the relation
n
X
aj χj (σ) = 0, for all σ ∈ G
j=1

implies aj = 0, for j = 1, . . . , n (all applicable n).

Theorem 4.30 (R. Dedekind, about 1890) If G is a group and {χα }α is a family of mutually distinct
characters of G (with values in K ∗ ), then they are independent.

Proof . We may assume our family is finite and we use induction on the number of elements, n, in this family.
The case n = 1 holds trivially. Assume that the result holds for all t ≤ n − 1 characters. Say χ1 , . . . , χn are
distinct characters of G and suppose
n
X
aj χj (σ) = 0, for all σ ∈ G. (∗)
j=1

The induction hypothesis implies that if the conclusion of the theorem is false, then aj 6= 0, for all j = 1, . . . , n.
Since the χj are distinct, there is some α 6= 1 with χ1 (α) 6= χn (α). Divide (∗) by an 6= 0, to obtain
n−1
X
bj χj (σ) + χn (σ) = 0, for all σ ∈ G. (∗∗)
j=1

Consider the group element ασ, then (∗∗) is true for it and we have
n−1
X
bj χj (α)χj (σ) + χn (α)χn (σ) = 0, for all σ ∈ G.
j=1

If we multiply by χn (α)−1 , we get

n−1
X
(bj χn (α)−1 χj (α))χj (σ) + χn (σ) = 0, for all σ ∈ G. (†)
j=1

If we subtract (†) from (∗∗), we get

n−1
X
bj (1 − χn (α)−1 χj (α))χj (σ) = 0, for all σ ∈ G.
j=1
4.5. THE THEOREMS OF DEDEKIND AND ARTIN; GALOIS GROUPS 261

By the induction hypothesis, bj (1 − χn (α)−1 χj (α)) = 0, for j = 1, . . . , n − 1. If we take j = 1 and we

remember that b1 = a1 /an 6= 0, we get

1 − χn (α)−1 χ1 (α) = 0,

i.e., χn (α) = χ1 (α), a contradiction.

e
Corollary 4.31 Say {χα }α is a family of mutually distinct isomorphisms of a field L with another, L.
Then, the χα are independent.
e in Dedekind’s theorem.
Proof . Take G = L∗ and K = L
e then set
Definition 4.11 If {χα }α is a family of isomorphisms K −→ K,

Fix({χα }) = {ξ ∈ K | (∀α, β)(χα (ξ) = χβ (ξ))}.

Observe that Fix({χα }) is always a subfield of K, so we call it the fixed field of {χα }α .

Note that Fix({χα }) contains the prime field of K.

e and if
Theorem 4.32 (E. Artin, 1940) If {χα }α is a family of pairwise distinct isomorphisms K −→ K
k = Fix({χα }), then
(1) [K : k] ≥ min(ℵ0 , #({χα })).

(2) Say {χα } forms a group under composition (so, K = K e and all χα ’s are automorphisms of K), then
if #({χα }) = n < ∞, we have [K : k] = n and if n = ∞ then [K : k] = ∞.

Proof . (1) First, we consider the case where we have a finite set, {χ1 , . . . , χn }, of isomorphisms K −→ K. e
Let k = Fix({χj }nj=1 ) and assume that [K : k] < n. Then, there exists a basis, ω1 , . . . , ωr , of K/k and r < n.
Consider the r equations in n unknowns (yj ’s)
n
X
yj χj (ωi ) = 0, 1 ≤ i ≤ r.
j=1

e So, we have
As r < n, this system has a nontrivial solution, call it (α1 , . . . , αn ) (with αi ∈ K).
n
X
αj χj (ωi ) = 0, 1 ≤ i ≤ r.
j=1
Pr
Pick any ξ ∈ K, as the ωi ’s form a basis, we can write ξ = i=1 ai ωi , for some (unique) ai ∈ k. We have
n
X n
X X
r  Xn X
r
αj χj (ξ) = α j χj ai ωi = αj χj (ai )χj (ωi ).
j=1 j=1 i=1 j=1 i=1

But, χj (ai ) = χl (ai ), for all j, l, as ai ∈ k and k = Fix({χj }). Write bi = χj (ai ) (independent of j). So, we
have
Xn r
X X
n 
αj χj (ξ) = bi αj χj (ωi ) .
j=1 i=1 j=1
Pn
But, j=1 αj χj (ωi ) = 0, by the choice of α1 , . . . , αn , so
n
X
αj χj (ξ) = 0, for all ξ.
j=1
262 CHAPTER 4. FIELDS AND GALOIS THEORY

This contradicts Dedekind’s theorem and thus, [K : k] ≥ n.

Now, consider the case where #({χα }) is infinite. If [K : k] were finite, then pick any n > [K : k] and
repeat the above argument with the subset {χ1 , . . . , χn }. We deduce that [K : k] must be infinite.
(2) Now, suppose {χ1 , . . . , χn } forms a group under composition (i.e., they are a group of automorphisms
of K). Then, one of the χj ’s is the identity, say χ1 = id. It follows that for every a ∈ k, we have
χj (a) = χ1 (a) = a, so
k = Fix({χj }) = {a ∈ K | χj (a) = a, j = 1, . . . , n}.
By part (1), we know [K : k] ≥ n; so, assume [K : k] > n. In this case, there exist r > n elements,
ω1 , . . . , ωr ∈ K, linearly independent over k. Consider the n equations in r unknowns (xi ’s)
r
X
xi χj (ωi ) = 0, j = 1, . . . , n.
i=1

Again, there is a nontrivial solution, say a1 , . . . , ar , with aj ∈ K. So, we have

r
X
ai χj (ωi ) = 0, j = 1, . . . , n. (†)
i=1

Note that for any nontrivial solution, the ai ’s can’t all be in k. If they were, then (†) with j = 1 gives
P r
i=1 ai ωi = 0, contradicting the linear independence of the ωi ’s.

Pick a solution containing a minimal number of nonzero ai ’s, say a1 , . . . , as 6= 0 and as+1 = · · · = ar = 0.
If we divide (†) by as , we get
s−1
X
bi χj (ωi ) + χj (ωs ) = 0, j = 1, . . . , n. (††)
i=1

By the remark above, there is some i, with 1 ≤ i ≤ s − 1, so that bi ∈ / k. By relabelling, we may assume
that b1 ∈
/ k. As b1 ∈
/ k, there is some ρ (1 ≤ ρ ≤ n) with χρ (b1 ) 6= b1 . Apply χρ to (††); we get
s−1
X
χρ (bi )(χρ ◦ χj )(ωi ) + (χρ ◦ χj )(ωs ) = 0, j = 1, . . . , n.
i=1

As χj ranges over {χ1 , . . . , χn }, so does χρ ◦ χj ; consequently, we have

s−1
X
χρ (bi )χξ (ωi ) + χξ (ωs ) = 0, ξ = 1, . . . , n. (∗)
i=1

If we subtract (∗) from (††), we obtain

s−1
X
(bi − χρ (bi ))χξ (ωi ) = 0, ξ = 1, . . . , n.
i=1

But, we know that b1 6= χρ (b1 ). For this ρ, not all the coefficients are zero, so we get a solution with strictly
fewer nonzero components, a contradiction to the minimality of (a1 , . . . , as ).

Definition 4.12 If Ω is a finite, normal extension of k, the Galois group of Ω/k, denoted G(Ω/k), is the
group of all k-automorphisms of Ω (i.e., the automorphisms, σ, of Ω so that σ  k = id). Say f ∈ k[X] and
let Ω be a splitting field for f (X) over k. The Galois group of the polynomial, f (X), over k, denoted Gk (f ),
is just G(Ω/k).
4.5. THE THEOREMS OF DEDEKIND AND ARTIN; GALOIS GROUPS 263

Lemma 4.33 Suppose Ω is finite, normal over k and G = G(Ω/k) is its Galois group. Then, a n.a.s.c. that
ξ ∈ Ω lie in Fix(G) is that ξ be purely inseparable over k.
s
Proof . If ξ is purely

inseparable over
k, then there is some s ≥ 0 so that ξ p ∈ k. Then, for every σ ∈ G,
ps ps s s s s
we have σ ξ = ξ . But, σ ξ p = (σ(ξ))p , so (σ(ξ))p = ξ p ; since char(k) = p, it follows that
s
(σ(ξ) − ξ)p = 0. Therefore, σ(ξ) − ξ = 0, i.e., ξ is fixed by σ and ξ ∈ Fix(G). Conversely, assume that
ξ ∈ Fix(G). First, pick an element α ∈ Ω, with α separable over k and α ∈ / k, if such an element exists.
Then, α is a simple root of some irreducible k-polynomial g. But, Ω is normal, so all the roots of g lie in
Ω and as α ∈ / k, we have deg(g) > 1. Consequently, there is another root, β ∈ Ω, of g with β 6= α and by
SMA, I, there exists σ ∈ G so that σ(α) = β. Now, consider our ξ ∈ Fix(G). As ξ ∈ Ω, there is some power,
r r r r
ξ p , of ξ that is separable over k. Since ξ is fixed by all σ ∈ G, so is ξ p . If ξ p were not in k, then ξ p could
r
play the role of α above, so it could be moved to some β 6= α, a contradiction. This implies that ξ p ∈ k,
i.e., ξ is purely inseparable over k.

Nomenclature & Notation.

Say Ω/k is a normal (not necessarily finite) extension. Pick an extension, K, in the layer Ω/k, i.e,
k ⊆ K ⊆ Ω. Define n o
r
K (∗) = ξ ∈ Ω | ξ p ∈ K, for some r ≥ 0 .
−∞ −∞
(Obviously, p = char(k).) Note that K (∗) = Ω ∩ K p in some algebraic closure (where K p is defined as
r
{ξ ∈ K | (∃r ≥ 0)(ξ p ∈ K)}). Also define

K(∗) = {ξ ∈ K | ξ is separable over k}.

Note: K (∗) and K(∗) are subfields of Ω/k and we have K(∗) ⊆ K ⊆ K (∗) ⊆ Ω.
0
We say that K is Galois equivalent to K 0 (where k ⊆ K ⊆ Ω and k ⊆ K 0 ⊆ Ω) iff K (∗) = K (∗) ; write
K gal K 0 . This equivalence relation fibers the subextensions of Ω/k into Galois equivalence classes.

Corollary 4.34 If Ω/k is finite, normal, then Fix(G(Ω/k)) = k (∗) . In particular, if k ⊆ L ⊆ Ω, then
Fix(G(Ω/L)) = L(∗) .

Corollary 4.35 If Ω/k is finite, normal, then #(G(Ω/k)) divides [Ω : k]; in particular,
#(G(Ω/k)) ≤ [Ω : k] < ∞.

Proof . By Artin’s theorem (Theorem 4.32) #(G(Ω/k)) = [Ω : Fix(G(Ω/k))]. By Lemma 4.33, we have
Fix(G(Ω/k)) = k (∗) . Therefore, #(G(Ω/k)) = [Ω : k (∗) ] which divides [Ω : k].

Corollary 4.36 If Ω/k is finite, normal and k is perfect, e.g. char(k) = 0, then #(G(Ω/k)) = [Ω : k].

Corollary 4.37 Say f is a separable, irreducible k-polynomial with degree deg(f ) = n. Then, there is an
injection Gk (f ) ,→ Sn (where Sn denotes the symmetric group on n elements) and this injection is unique
up to inner automorphisms in Sn . In particular, #(Gk (f )) | n!.

Proof . Write α1 , . . . , αn for all the roots of f (they are all distinct) in some order. Given σ ∈ Gk (f ), the
element σ(αi ) is some other root of f , call it αpσ (i) . Then, pσ is a permutation of the n roots, i.e., pσ ∈ Sn .
Clearly, the map σ 7→ pσ is a homomorphism Gk (f ) −→ Sn . If pσ = id, then σ(αi ) = αi for all i, so
σ  Ω = id, as Ω, the splitting field of f , is generated over k by the αi ’s. So, σ = id in Gk (f ) = G(Ω/k), and
the our map Gk (f ) −→ Sn is an injection. We can reorder (relabel) the α1 , . . . , αn ; to do so introduces an
inner automorphism of Sn .
264 CHAPTER 4. FIELDS AND GALOIS THEORY

Remarks: (On Galois equivalence)

0 r 0
(1) If K ⊆ K 0 , then K (∗) ⊆ K (∗)
. Indeed, if ξ ∈ K (∗) , then ξ p ∈ K ⊆ K 0 (for some r ≥ 0), so ξ ∈ K (∗)
.
(2) For all K in the layer Ω/k (of course, Ω/k is a finite normal extension), we have K gal K (∗) . Hence,
the Galois equivalence class of any field possesses a unique least upper bound, namely K (∗) for any K
r
in the class. For, K ⊆ K (∗) , so K (∗) ⊆ (K (∗) )(∗) . Also, if ξ ∈ (K (∗) )(∗) , then ξ p ∈ K (∗) , for some
r s r+s
r; but then, (ξ p )p ∈ K, for some s, i.e., ξ p ∈ K, which means that ξ ∈ K (∗) . Consequently,
(K ) ⊆ K and so (K ) = K , i.e. K gal K (∗) . If K gal L then K (∗) = L(∗) ; K ⊆ K (∗) and
(∗) (∗) (∗) (∗) (∗) (∗)

L ⊆ L(∗) , so K (∗) = L(∗) is indeed the least upper bound of the equivalence class of K and L.
(3) If K belongs to the layer Ω/k (where Ω/k is normal), then K(∗) gal K and K(∗) is the unique greatest
lower bound for the Galois equivalence class of K.
Proof . If we prove that (K(∗) )(∗) = K (∗) and (K (∗) )(∗) = K(∗) , we are done. The first equation will
r
prove that K(∗) gal K. As K(∗) ⊆ K, we get (K(∗) )(∗) ⊆ K (∗) . Pick ξ ∈ K (∗) , then ξ p ∈ K, for
r s r+s
some r and (ξ p )p = ξ p ∈ K(∗) , for some s, so ξ ∈ (K(∗) )(∗) ; hence, (K(∗) )(∗) = K (∗) . Now, pick
(∗)
ξ ∈ K(∗) , then ξ ∈ K (as K(∗) ⊆ K ⊆ K (∗) ) and since ξ is separable over k, we have ξ ∈ (K (∗) )(∗) .
Conversely, if ξ ∈ (K (∗) )(∗) , then ξ ∈ K (∗) , which means that ξ is inpurely separable over K. Yet, ξ is
separable over k, so ξ is separable over K. As ξ is purely inseparable over K and separable over K,
we get ξ ∈ K; moreover, as ξ is separable over k, we get ξ ∈ K(∗) .
(4) We have K gal L iff K(∗) = L(∗) , hence in each Galois equivalence class, there is a unique greatest
lower bound, it is the common K(∗) . If K gal L, then K (∗) = L(∗) , so

K(∗) = (K (∗) )(∗) = (L(∗) )(∗) = L(∗) ,

by (3). Conversely, if K(∗) = L(∗) , then

K (∗) = (K(∗) )(∗) = (L(∗) )(∗) = L(∗) ,

again, by (3), i.e., K gal L.

(5) Suppose K gal L and K, L ⊆ Ω/k, Then, G(Ω/K) = G(Ω/L), hence the maps

G(Ω/L(∗) ) ,→ G(Ω/L) ,→ G(Ω/L(∗) )

are equalities. All we need show is G(Ω/L) = G(Ω/L(∗) ). We already know G(Ω/L(∗) ) ⊆ G(Ω/L), as
r
(∗)
L⊆L . Say σ ∈ G(Ω/L) and pick any ξ ∈ L(∗) . Then, ξ p ∈ L, for some r ≥ 0. Consequently,
r
r r r r
σ ξ p = ξ p , as σ  L = id. As σ is an automorphism, we get (σ(ξ))p = ξ p , i.e., (σ(ξ) − ξ)p = 0,
and so, σ(ξ) = ξ. As ξ is arbitrary in L(∗) , we have σ  L(∗) = id; since σ is arbitrary, the proof is
complete.

Theorem 4.38 (Fundamental Theorem of Galois Theory) Suppose Ω/k is a finite, normal extension. Write
G for G(Ω/k) and write [K] for the Galois class of K ⊆ Ω/k. Then, the maps

H 7→ [Fix(H)] and [L] 7→ G(Ω/L)

establish a one-to-one order-reversing correspondence between all subgroups of G and all the Galois classes of
subextensions L/k ⊆ Ω/k. Here, [K] ⊆ [L] means K (∗) ⊆ L(∗) as fields. In this correspondence, G(Ω/L) C G
iff L(∗) is a normal extension of k iff L(∗) is a normal extension of k. When the latter is the case, then there
is a canonical exact sequence

0 −→ G(Ω/L) −→ G(Ω/k) −→ G(L(∗) /k) −→ 0.

4.5. THE THEOREMS OF DEDEKIND AND ARTIN; GALOIS GROUPS 265

Claim 1. If L = Fix(H), then L = L(∗) .

r r
r
Pick ξ ∈ L(∗) , so ξ p ∈ L, for some r ≥ 0. Then, for all σ ∈ H, we have σ ξ p = ξ p , and by a standard
argument, ξ ∈ Fix(H) = L. Consequently, L(∗) ⊆ L, yet L ⊆ L(∗) , so L = L(∗) .

Proof of Theorem 4.38. Now say H ⊆ H e and look at Fix(H). e If ξ ∈ Fix(H),

e then for every τ ∈ H,
e we have
e ⊆ Fix(H) and so,
τ (ξ) = ξ and so, for every σ ∈ H, we have σ(ξ) = ξ, i.e., ξ ∈ Fix(H). Consequently, Fix(H)
e e (∗) e (∗) e
[Fix(H)] ⊆ [Fix(H)], by Claim (1). Now, if [L] ⊆ [L], then L ⊆ L . If σ ∈ G(Ω/L), then σ ∈ G(Ω/L e (∗) )
(∗) e
(by Remark (5), above); so, σ ∈ G(Ω/L ) = G(Ω/L) (again, by Remark (5)). Thus, G(Ω/L) ⊆ G(Ω/L).

e say we know Fix(H) = Fix(H).

Given H ⊆ H, e By Artin’s theorem, we have

e = #(H).
#(H) = [Ω : Fix(H)] = [Ω : Fix(H)] e

e and #(H) = #(H),

As H ⊆ H e we get H = H.
e

Choose a subgroup, H, of G and let L = Fix(H); write H e for G(Ω/L) = G(Ω/Fix(H)). If σ ∈ H,

e e e
then σ fixes L, so σ ∈ H and H ⊆ H. But, Fix(H) = Fix(G(Ω/L)) = L(∗) , by Corollary 4.34. Thus,
e = (Fix(H))(∗) . Claim 1 implies that (Fix(H))(∗) = Fix(H), so Fix(H)
Fix(H) e = Fix(H) and, by the above,
e Therefore, H = G(Ω/Fix(H)).
we get H = H.

Consider L, make G(Ω/L) and form Fix(G(Ω/L)). By Corollary 4.34, we have Fix(G(Ω/L)) = L(∗) and
L gal L(∗) , so [L] = [Fix(G(Ω/L))].

Having proved all the statements about the order inverting correspondence, we see that only normality
statements remain.

Claim 2. If L ⊆ Ω/k, then L is normal over k iff for every σ ∈ G(Ω/k), we have σ(L) = L.

(⇒). For every σ ∈ G(Ω/k), the field σ(L) is k-conjugate to L. As L is normal over k, we find σ(L) = L.

(⇐). Assume σ(L) = L, for every σ ∈ G(Ω/k). Let g be any irreducible k-polynomial and assume that
α ∈ L is a root of g. But, α ∈ Ω and Ω is normal; consequently, all the roots of g lie in Ω. Say β ∈ Ω is any
other root of g. By SMA, I, there is some σ ∈ G so that σ(α) = β. So, β ∈ σ(L), and as σ(L) = L, we get
β ∈ L. Thus, L contains all the roots of g which means that L is normal over k.

Assume G(Ω/L) C G. Look at L(∗) and choose any σ ∈ G and any η ∈ σ(L(∗) ). Then, σ −1 (η) ∈ L(∗) and
for all τ ∈ G(Ω/L) = G(Ω/L(∗) ), we have

(στ σ −1 )(η) = σ(τ (σ −1 (η))) = (σσ −1 )(η) = η,

because σ −1 (η) ∈ L(∗) . Thus, (σG(Ω/L)σ −1 )(η) = η, and as G(Ω/L) C G, we get G(Ω/L)(η) = η, so
η ∈ Fix(G(Ω/L)) = L(∗) , as we know. In summary, if η ∈ σ(L(∗) ), then η ∈ L(∗) , i.e., σ(L(∗) ) ⊆ L(∗) . If we
apply this to σ −1 , we get σ −1 (L(∗) ) ⊆ L(∗) , i.e. L(∗) ⊆ σ(L(∗) ). Therefore, L(∗) = σ(L(∗) ) and by Claim 2,
the extension L(∗) /k is normal.

Now, say L(∗) is normal over k. Then, we know σ(L(∗) ) = L(∗) , for all σ ∈ G(Ω/k). For any ξ ∈ L(∗) and
any τ ∈ G(Ω/L), we have
(στ σ −1 )(ξ) = σ(τ (σ −1 (ξ))) = (σσ −1 )(ξ) = ξ,

because σ −1 (ξ) ∈ σ −1 (L(∗) ) = L(∗) , by hypothesis. So, στ σ −1 ∈ G(Ω/L(∗) ) = G(Ω/L) and thus, G(Ω/L) C G.
266 CHAPTER 4. FIELDS AND GALOIS THEORY

Suppose L(∗) is normal. We have a map G(Ω/k) −→ G(L(∗) /k) via σ 7→ σ  L(∗) (σ  L(∗) ∈ G(L(∗) /k),
by normality). This map is onto because, given any σ ∈ G(L(∗) /k), we have the diagram

ΩO
σ / Ω
O
e

L(∗)
σ / L(∗)
O O

k k,

e, of Ω. The kernel of our map is clearly

and by SMA, II, the automorphism σ lifts to an automorphism, σ
G(Ω/L).
Lastly, we need to show that L(∗) is normal iff L(∗) is normal. Say L(∗) is normal and σ ∈ G. If ξ ∈ L(∗) ,
then ξ ∈ L(∗) and σ(ξ) ∈ L(∗) (as L(∗) is normal). But, σ(ξ) is separable over k as ξ is. It follows that
σ(ξ) ∈ (L(∗) )(∗) = L(∗) and so, σ(L(∗) ) ⊆ L(∗) . By the usual argument, σ(L(∗) ) = L(∗) and L(∗) is normal.
r r

If L(∗) is normal and ξ ∈ L(∗) , then ξ p ∈ L(∗) , for some r ≥ 0. It follows that σ ξ p ∈ σ(L(∗) ) = L(∗) , so
r
(σ(ξ))p ∈ L(∗) , i.e., σ(ξ) ∈ (L(∗) )(∗) = L(∗) ; thus, σ(L(∗) ) ⊆ L(∗) and, by the usual argument, we conclude
that L(∗) is normal.

Proposition 4.39 Suppose Ω is normal over k and L/k ⊆ Ω/k. Then L = L(∗) iff Ω is separable over L.

Proof . (⇒). Say Ω is separable over L, then as L(∗) ⊆ Ω, we find L(∗) is separable over L. Yet, L(∗) is
purely inseparable over L. It follows that L = L(∗) .
(⇐). We must prove that Ω is separable over L(∗) . Pick α ∈ Ω and consider G(Ω/L(∗) ). Choose
σ1 , . . . , σn ∈ G(Ω/L(∗) ) so that

(1) σ1 = id and α = σ1 (α), σ2 (α), . . . , σn (α) are mutually distinct,

(2) n is maximal, i.e., no further σ ∈ G(Ω/L(∗) ) can be added while preserving (1).
Qn
Consider g(X) = i=1 (X − σj (α)). If σ ∈ G(Ω/L(∗) ), the elements σσ1 (α), . . . , σσn (α) are a per-
mutation of α, σ2 (α), . . . , σn (α), so σg(X) = g(X). This implies that the coefficients of g(X) belong to
Fix(G(Ω/L(∗) )) = L(∗) . Thus, g(X) ∈ L(∗) [X], but the roots of g(X) are distinct and α is among them.
Therefore, α is separable over L(∗) .

Corollary 4.40 Assume Ω/k is a finite normal extension. Then, the following are equivalent:

(1) Ω is separable over k.

(2) k (∗) = k.

(3) For all subextensions, L, of Ω/k, we have L(∗) = L(∗) .

(3a) For all subextensions, L, of Ω/k, the equivalence class [L] has but one element.

(4) Same as (3) but for some extension L/k ⊆ Ω/k.

(4a) Same as (3a) but for some extension L/k ⊆ Ω/k.

(5) Ω = Ω(∗) .
4.5. THE THEOREMS OF DEDEKIND AND ARTIN; GALOIS GROUPS 267

Proof . First, observe that the equivalences (3) ⇐⇒ (3a) and (4) ⇐⇒ (4a) are obvious.
(1) ⇒ (2). This is Proposition 4.39 when L = k.
(2) ⇒ (3). Given L ⊆ Ω/k, then L(∗) ⊆ Ω. By Proposition 4.39, Ω is separable over k. Thus, L(∗) is
separable over k and so L(∗) is separable over L(∗) ; yet, L(∗) is purely inseparable over L(∗) , so L(∗) = L(∗) .
(3) ⇒ (4) is a tautology.
(4) ⇒ (5). We have L(∗) = L(∗) , for some L ⊆ Ω/k. Proposition 4.39 implies that Ω is separable over L(∗) .
But, L(∗) is always separable over k and separability is transitive, so Ω is separable over k, i.e., Ω = Ω(∗) .
(5) ⇒ (1). By definition, Ω(∗) is separable over k and Ω = Ω(∗) , so Ω is separable over k.

Proposition 4.41 Say Ω/k is a finite normal extension. Then, Ω = Ω(∗) k (∗) (= the smallest field containing
Ω(∗) and k (∗) ). The natural map
Ω(∗) ⊗k k (∗) −→ Ω
is an isomorphism. Indeed, for all L/k ⊆ Ω/k, we have

(1) L(∗) = Lk (∗) = L(∗) k (∗) .

(2) L(∗) = L ∩ Ω(∗) .

(3) The natural map

L(∗) ⊗k k (∗) −→ L(∗)
is an isomorphism.

Proof . We just have to prove (1)–(3) for L/k ⊆ Ω/k and then set L = Ω to get the rest.
(1) Since L(∗) ⊇ k (∗) and L(∗) ⊇ L ⊇ L(∗) , we deduce that L(∗) ⊇ L(∗) k (∗) and L(∗) ⊇ Lk (∗) . If ξ ∈ L(∗) ,
then ξ is purely inseparable over L(∗) , so ξ is purely inseparable over L(∗) k (∗) . If ξ ∈ L(∗) , then ξ is separable
over k (∗) (by Proposition 4.39), so ξ is separable over L(∗) k (∗) . Thus, L(∗) is both separable and purely
inseparable over L(∗) k (∗) , which means that L(∗) = L(∗) k (∗)
(2) This is the definition of L(∗) , as L ⊆ Ω.

(3) The (illegal definition of the) map is α ⊗ β 7→ αβ. The image is L(∗) k (∗) = L(∗) . So, we need to prove
−∞ −∞ r
our map is injective. Now, k (∗) ⊆ k p (where k p = {ξ ∈ k | ξ p ∈ k, for some r ≥ 0}). By Mac Lane I
and right limits, we get
−∞ −∞
L(∗) ⊗k k p −→ L(∗) k p
−∞
is injective (because L(∗) is separable over k). But, 0 −→ k (∗) −→ k p is exact and vector spaces over a
field are flat, so
−∞
0 −→ L(∗) ⊗k k (∗) −→ L(∗) ⊗k k p
is still exact. Then, the diagram

0 / L(∗) ⊗k k (∗) / L(∗) ⊗k k p−∞

 

L(∗) k (∗)  / L(∗) k p−∞

commutes, and this shows that L(∗) ⊗k k (∗) −→ L(∗) k (∗) = L(∗) is injective.
268 CHAPTER 4. FIELDS AND GALOIS THEORY

Proposition 4.42 Suppose Ω/k is a finite normal extension and G = G(Ω/k). Let L/k ⊆ Ω/k and H =
G(Ω/L). Then,

(1) [Ω : L(∗) ] = #(H).

(2) [L(∗) : k] = (G : H).

Moreover, we have [Ω : Ω(∗) ] = [L(∗) : L(∗) ] = [k (∗) : k] = a p-power (the degree of inseparability of Ω/k).

Proof . We know Fix(H) = L(∗) . So, (1) is just Artin’s theorem (Theorem 4.32).

Claim: The map σ 7→ σ  Ω(∗) is an isomorphism G −

g→ G(Ω(∗) /k).

We know Ω(∗) is normal over k, so σ  Ω(∗) takes Ω(∗) to itself. Therefore, the map G −→ G(Ω(∗) /k)
given by σ 7→ σ  Ω(∗) is well defined. If σ 7→ id ∈ G(Ω(∗) /k), then σ  Ω(∗) leaves Ω(∗) element-wise fixed.
r r
r
If ξ ∈ Ω, then ξ p ∈ Ω(∗) , for some r. Therefore, σ ξ p = ξ p . By the usual argument, we conclude that
σ(ξ) = ξ. Therefore, σ = id on Ω and our map is injective. Pick σ e ∈ G(Ω(∗) /k). We have the diagram

ΩO
σ / Ω
O

Ω(∗)
σ
e / Ω(∗)
O O

k k.

e comes from a σ : Ω → Ω; so, our map is onto.

By SMA, II, our automorphism σ

We have Fix(G(Ω(∗) /k)) = k (as k (∗) = k in Ω(∗) ). By Artin’s theorem, [Ω(∗) : k] = #(G). Now,

[Ω : L(∗) ] = [Ω : L(∗) ][L(∗) : L(∗) ] = [Ω : Ω(∗) ][Ω(∗) : L(∗) ],

and
H = G(Ω/L) = G(Ω/L(∗) ) = G(Ω/L(∗) ) = G(Ω(∗) /L(∗) ),

by what’s just been proved. By Artin’s theorem, [Ω : L(∗) ] = #(H), so

[Ω : L(∗) ] = #(H)[L(∗) : L(∗) ] = [Ω : Ω(∗) ]#(H);

it follows that [L(∗) : L(∗) ] = [Ω : Ω(∗) ], for all L. As remarked above,

#(G) = [Ω(∗) : k] = [Ω(∗) : L(∗) ][L(∗) : k] = #(H)[L(∗) : k].

Consequently, [L(∗) : k] = (G : H).

A picture of the situation is shown in Figure 4.1.

4.5. THE THEOREMS OF DEDEKIND AND ARTIN; GALOIS GROUPS 269

[Ω] Ω

Ω(∗) #(H)

#(H) [L] L(∗)

L(∗) (G : H)

(G : H) [k] k (∗)

Figure 4.1: Structure of Normal Extensions

270 CHAPTER 4. FIELDS AND GALOIS THEORY

4.6 Primitive Elements, Natural Irrationalities, Normal Bases

Proposition 4.43 If G is a finite subgroup of K ∗ = Gm (K), where K is a field, then G is cyclic.

Proof . An abelian finite group is cyclic iff its p-Sylow subgroups are cyclic (DX). So, we may assume that
#(G) = pr , for some r > 0 and some prime p. Let x ∈ G be an element of maximal order, q = pt , with
t ≤ r. Pick any y ∈ G; the order of y is equal to ps for some s. But order(y) ≤ order(x), so s ≤ t. As
order(y) | order(x), we must have y q = 1. So, for every y ∈ G, the element y is a root of T q − 1. As K is
a field, this polynomial has at most q roots. But, there exist q roots in G: 1, x, . . . , xq−1 . Therefore, G is
generated by x.

Corollary 4.44 In any field, the n-th roots of unity in the field form a cyclic group. It is a finite subgroup
of Gm (K).

Corollary 4.45 The multiplicative group of a finite field is always cyclic. Every nonzero element of a finite
field is a root of unity.

Theorem 4.46 (Artin’s Theorem of the Primitive Element) Suppose K/k is a finite extension of fields, then
there is some α ∈ K so that K = k(α) iff there are only finitely many fields, L, with k ⊆ L ⊆ K. (Such an
α is called a primitive element).

Proof . (⇒). Assume K = k(α). Let L be any subfield of K, write f (X) for the minimal k-polynomial of α.
We know that f (X) is irreducible in k[X]. Let g(X) be the minimum L-polynomial for α. As k(α) = L(α),
we have [k(α) : L] = [L(α) : L] = deg(g). Take L0 to be the field obtained by adjoining the coefficients of g to
k; we have L0 ⊆ L. Thus, g(X) ∈ L0 [X] and g(X) is irreducible over L0 . Consequently, [L0 (α) : L0 ] = deg(g).
But, L0 (α) = k(α), so
deg(g) = [k(α) : L0 ] = [k(α) : L][L : L0 ] = deg(g)[L : L0 ],
and we deduce that L = L0 . This means that L is uniquely determined by g. However, every g(X) is a
factor of f (X) ∈ K[X] and since there are only finitely many factors of f (X), there are only finitely many
subfields L.
(⇐). Say K/k possesses just finitely many subfields.
Claim: Given α, β ∈ K, there is some γ ∈ K with k(α, β) ⊆ k(γ).
If the claim holds, we can finish the proof by induction on the number of generators, n, for K/k. The cases
n = 1, 2, are clear. Assume that the induction hypothesis holds for n − 1 ≥ 1, and let K = k(α1 , . . . , αn ) =
k(α1 , . . . , αn−2 )(αn−1 , αn ). The claim implies that K = k(α1 , . . . , αn−2 )(γ), and the induction hypothesis
finishes the proof. So, we just have to prove the claim.
If k is finite, so is K. Consequently, K ∗ = Gm (K) is cyclic, which means that K ∗ = Gp{α} and
K = k(α). Thus, we may assume k is infinite. Make a map from k to the subfields of k(α, β) via

λ 7→ k(α + λβ) (⊆ k(α, β)).

e with λ 6= λ,
Since k is infinite and since there are only finitely many subfields, there is a pair (λ, λ), e and

e = L.
k(α + λβ) = k(α + λβ)

e ∈ L, so (λ − λ)β
Thus, both α + λβ, α + λβ e ∈ L. But λ − λe 6= 0, so β ∈ L, and then, α ∈ L. It follows that
k(α, β) ⊆ L = k(α + λβ), and γ = α + λβ does the job.

Corollary 4.47 (Kronecker’s Theorem of the Primitive Element) Suppose K/k is a finite separable field
extension, then there is some α ∈ K so that K = k(α).
4.6. PRIMITIVE ELEMENTS, NORMAL BASES 271

Proof . If Ω is the normal closure of K, then it is normal and separable over k. By the main theorem of
Galois theory, there is a one-to-one correspondence between subfields of Ω/k and subgroups of G(Ω/k). As
G(Ω/k) is finite, there are only finitely many subfields of Ω/k. But, any subfield of K/k is a subfield of Ω/k,
which means that there are only finitely many subfields of K/k. Then, Theorem 4.46 (Artin) implies that α
exists.

Corollary 4.48 Say K/k is a finite degree field extension and Ω is some field with k ⊆ Ω. Then, the
number of k-monomorphisms K −→ Ω is at most [K : k]s . If Ke is a field k-isomorphic to K and Ke
rel Ω,
f
e
then the number of k-monomorphisms K −→ Ω is equal to [K : k]s iff Ω contains the normal closure of K.
In particular, if K ⊆ Ω, then the number of k-monomorphisms K −→ Ω is equal to [K : k]s iff Ω contains
the normal closure of K.

Proof . Look at K(∗) , then we know that [K(∗) : k] = [K : k]s . By Kronecker’s theorem of the primitive
element, there is some α ∈ K(∗) so that K(∗) = k(α). To give a k-monomorphism K −→ Ω implies that
we have a k-monomorphism K(∗) −→ Ω and the latter is determined by its value on α. Furthermore, two
k-monomorphisms of K to Ω which agree on K(∗) necessarily agree on K. Hence, the choice of an image of α
in Ω determines a k-monomorphism of K −→ Ω. The image of α, say β, satisfies the minimal k-polynomial,
g(X), for α. Consequently, the number of k-monomorphisms K −→ Ω is at most equal to the number of
roots of g(X) in Ω, which is at most deg(g) = [K : k]s .

Take K e with K
e e e ∼ e
rel Ω and say K is k-isomorphic to K. Since K = K, we are reduced to the case K = K,
f
i.e., Ω f
rel K. We obtain the maximum number of k-monomorphisms iff Ω contains all the roots of any
irreducible k-polynomial one root of which lies in k. For then all the conjugates of α are there and their pr th
roots for suitable r.

Theorem 4.49 (Natural Irrationalities) Say Ω/k is finite normal and e k ⊇ k is some field with ekf
rel Ω.
Write Ω for the compositum of k and Ω, denoted Ωk (the smallest field containing Ω and e
e e e k). Then,

e e
(1) Ω/k is a normal extension (finite degree).

e e
(2) The map σ 7→ σ  Ω gives a canonical injection G(Ω/ k) ,→ G(Ω/k). The image of this injection is
e
G(Ω/D), where D = Ω ∩ k.

Proof . (1) We know Ω = k(α1 , . . . , αt ), where α1 , . . . , αt are all the roots of a k-polynomial, f . Now,
e =e
Ω k(α1 , . . . , αt ) = a splitting field of the same f , but now viewed as a e
k-polynomial. So (1) holds.

(2) Given σ ∈ G(G/ e e e As Ω is normal,

k), look at σ  Ω. We know σ(Ω) is a k-conjugate to Ω (inside Ω).
e e
σ(Ω) = Ω, and so, σ  Ω is an automomorphism of Ω. As σ fixes k, it fixes k ⊆ k. Thus, σ  Ω ∈ G(Ω/k).
If σ  Ω were the identity, we would have σ(αj ) = αj , for all j. Also, σ  e k = id and thus, σ fixes all of
e e e e
k(α1 , . . . , αt ) = Ω. Therefore, σ = id in G(G/k), i.e., our map is injective.

Let D = Ω∩ e e e
k and let H be the image of G(Ω/ k) in G(Ω/k). We have H ∼ e e
= G(Ω/ k). As D ⊆ e k, we see that
H fixes D, so H ⊆ G(Ω/D). Let L = Fix(H). We know that L = L . As D is fixed, D ⊆ L = L(∗) ⊆ Ω.
(∗)

Now, all elements of H come from G(Ω/ e e

k), which implies that Fix(H) ⊆ Fix(G(Ω/ e e k)) = e k (∗) , by Corollary
4.34. So, D ⊆ L = L(∗) ⊆ e k (∗) and D ⊆ L = L(∗) ⊆ Ω. Pick ξ ∈ L. Then, ξ ∈ e k (∗) , so ξ p ∈ e
r
k, for some r.
But, ξ ∈ L ⊆ Ω, so ξ p ∈ Ω, and thus, ξ p ∈ e
r r
k ∩ Ω = D. It follows that L ⊆ D(∗) . As L = L(∗) , we have
L(∗) ⊆ D(∗) . Yet, D ⊆ L, so D(∗) ⊆ L(∗) and therefore L(∗) = D(∗) . It follows that

G(Ω/D) = G(Ω/L) = G(Ω/Fix(H)) = H,

by the fundamental theorem of Galois theory.

272 CHAPTER 4. FIELDS AND GALOIS THEORY

Corollary 4.50 (Original Form of Natural Irrationalities) Say f is a k-polynomial and k ⊆ e k. Then, Gek (f )
e
is a subgroup of Gk (f ) in a natural way and in fact, Gek (f ) = GD (f ), where D = Ω ∩ k and Ω f e
rel k is a
splitting field of f .

Explanation: Let Ω be a given splitting field of f . The elements of Ω were termed the natural irrationalities
of f . The reduction in Gk (f ) effected by considering f over ek is the same as that achieved by considering f
over the field of those natural irrationalities of f contained in e
k.

Theorem 4.51 (Normal Basis Theorem) Suppose K/k is a finite normal and separable extension and let
G(K/k) be its Galois group. Then, there is some θ ∈ K so that {σθ | σ ∈ G(K/k)} is a k-basis for K. (This
is called a normal basis for K/k).

Proof . By Kronecker’s theorem, K = k(α), for some α ∈ K; let f (X) be the minimum k-polynomial for α.
We know K = k[X]/(f (X)). Examine two rings: K[X] and A = K[X]/(f (X)). Note,

K ⊗k K = K ⊗k (k[X]/(f (X))) ∼
= K[X]/(f (X)) = A.

For σ ∈ G = G(K/k), write ασ for σ(α). Consider the K-polynomials

f (X)
gσ (X) = .
f 0 (ασ )(X − ασ )

Note that g1 (X) = f (X)/(f 0 (α)(X − α)), so σg1 (X) = gσ (X). The gσ ’s satisfy the following properties:
(1) Each gσ (X) has degree deg(f ) − 1.
(2) If σ 6= τ , then gσ (ατ ) = 0.
Also, by Taylor’s theorem,

f (X) = f (ασ + (X − ασ )) = f (ασ ) + f 0 (ασ )(X − ασ ) + O((X − ασ )2 ),

so, gσ (X) = 1 + O(X − ασ ) and therefore,

(3) gσ (ασ ) = 1.
P
Consider the polynomial σ∈G gσ (X) − 1 (∈ K[X]). By (2) and (3), we see that this polynomial vanishes on
the n elements α, ασ2 , . . . , ασn , where G = {1, σ2 , . . . , σn }. By (1), this polynomial has degree n − 1. Hence,
the polynomial is identically zero and we have
X
gσ (X) = 1. (partition of unity) (∗)
σ∈G

In A, we get
X
gσ (X) = 1. (∗)
σ∈G

, with σ 6= τ , and look at gσ (X)gτ (X). For all ρ ∈ G, we have gσ (αρ )gτ (αρ ) = 0. But,
Pick σ, τQ
f (X) = ρ∈G (X − αρ ), so f (X) | gσ (X)gτ (X) if σ 6= τ . If we read this in A, we get

gσ (X) gτ (X) = 0 in A, if σ 6= τ . (orthogonality) (∗∗)

If we multiply (∗) by gτ (X), we get X

gτ (X)gσ (X) = gτ (X),
σ∈G
4.6. PRIMITIVE ELEMENTS, NORMAL BASES 273

and if we read this in A and use (∗∗), we get

(gσ (X))2 = gσ (X) in A. (idempotence) (∗∗∗)
Write eσ = gσ (X), so eσ ∈ A = K ⊗k K. Then, (∗), (∗∗) and (∗∗∗) say:
X
eσ = 1; eσ eτ = δστ eσ .
σ∈G

Therefore, the eσ ’s are an orthogonal decomposition of 1 by idempotents, and so,

Y Y
K ⊗k K ∼
= Keσ ∼= K.3
σ∈G σ∈G

Order the elements of G in some fashion as we did above: 1, σ2 , . . . , σn , and consider the matrix
(gστ (X)) ∈ Mn (K[X]).
Let D(X) = det(gστ (X)). In order to compute D(X) in A, consider D(X)2 . Since
det(gστ (X)) = det(gστ (X))> , we can compute D(X)2 by multiplying columns by columns and summing.
We get X X X
gστ (X)gσρ (X) = σ(gτ (X))σ(gρ (X)) = σ(gτ (X)gρ (X)).
σ∈G σ∈G σ∈G
If we read this in A, we get
X X
gστ (X) gσρ (X) = σ(gτ (X) gρ (X)) = 0, if τ 6= ρ; and
σ∈G σ∈G
X
= σ(gρ (X)), if τ = ρ
σ∈G
X
= gσρ (X)
σ∈G
X
= gτ (X) = 1, if τ = ρ.
τ ∈G

Therefore, we find that in A, the matrix (gστ (X))(gστ (X))> is the identity matrix and so, D(X)2 = 1.
Consequently, D(X)2 ≡ 1 (mod f (X)), which shows that D(X) 6= 0.
If k is infinite, then there is some ξ ∈ k with D(ξ) 6= 0. Let θ = g1 (ξ). Then, στ θ = στ g1 (ξ) = gστ (ξ).
Consequently, det(στ (θ)) = det(gστ (ξ)) = D(ξ) 6= 0. If {σθ}σ∈G were linearly dependent, we would have
X
aτ τ θ = 0,
τ ∈G

for some aτ ∈ k, not all zero. If we apply σ, we get

X
aτ στ θ = 0.
τ ∈G

So, (aτ ) would be a nontrivial simultaneous solution to the linear system of equations
X
Xτ στ θ = 0, for σ ∈ G,
τ ∈G

a contradiction to the fact that det(στ (θ)) 6= 0. Therefore, {σθ}σ∈G is linearly independent and the case
where k is infinite is proved.
If k is finite, we don’t need the gσ (X) and D(X). We do need the following facts to be proved below:
3 At this stage, we are essentially done. However, we’ve not kept track of the G action; so, a little more argument is needed.
274 CHAPTER 4. FIELDS AND GALOIS THEORY

(1) The Galois group G(K/k) is cyclic.

(2) The Galois group G(K/k) has a canonical generator, F, where F(ξ) = ξ #(k) , for all ξ ∈ K.

Recall that for a linear transformation, T , on a finite dimensional vector space, V , if m(X) is the minimal
polynomial for T then there exists a vector, v ∈ V , so that m(T )v = 0 but no polynomial of smaller degree
than m(X) kills v. Now, our K plays the role of V and the automorphism F plays the role of T . If we can
show that the minimum polynomial of F is exactly X n − 1, where n = [K : k], then we take a Θ in K so
that no polynomial of smaller degree than Fn − 1 kills Θ. This means that

Θ, F(Θ), · · · , Fn−1 (Θ)

are linearly independent; so by (1) and (2) we have our normal basis.
Of course, by (1) and (2), Fn−1 − 1 ≡ 0 on K; therefore, whatever is the minimal polynomial for F, it
divides X n − 1 and its degree is at most n. Were m(X) = a0 X d + a1 X d−1 + · · · + ad the minimal polynomial
for F and d < n, then
0 = a0 Fd (ξ) + a1 Fd−1 (ξ) + · · · + ad−1 F(ξ) + a0 F0 (ξ) (†)
for all ξ ∈ K. But this is a contradiction of Dedekind’s Theorem as (†) is a linear dependence among I,
F, · · · , Fd , and we are done.

Remark: The argumeent actually proves (independently of previous arguments) that every cyclic extension
possesses a normal basis.
The facts concerning finite fields were proved by E.H. Moore. Here is his theorem:

Theorem 4.52 (E.H. Moore, 1892) If k is a finite field then char(k) = p > 0 and #(k) = pl , for some
prime p and some l ≥ 1. If Fp is the prime field of characteristic p, then for each integer l ≥ 1, there
exists one and only one finite field of cardinality pl , up to Fp -isomorphism. If K/k is a finite extension of
degree n and k is a finite field, then K/k is always normal and separable; the Galois group G(K/k) is cyclic
of order n and has a canonical generator, F. This F is the Frobenius automorphism, and it is given by
ξ 7→ F(ξ) = ξ #(k) , for all ξ ∈ K. Each finite field has exactly one extension of degree n for each n ≥ 1.

Proof . The statement in the first sentence is well-known. Pick l ≥ 1 and look at the splitting field of the
l
polynomial X p − X ∈ Fp [X]. Note, if ξ and η are roots of this polynomial, then ξ ± η, ξη, ξ/η are also
roots of the polynomial. Thus, the set of roots is a field and it contains Fp , because for all ξ ∈ Fp , we have
l
ξ p = ξ. It follows that the splitting field is exactly the entire set of roots and as the derivative of X p − X is
−1, the roots are distinct. Therefore, we get a field with pl elements. Conversely, any field with pl elements
has multiplicative group of order pl − 1. So, this group has a generator of order pl − 1 and for this generator,
l l
θ, we get θp = θ. Consequently, any power of θ satisfies X p − X = 0 and so, our field is a splitting field of
l
X p − X; such fields are unique up to Fp -isomorphism.
Suppose K/k has degree n, then K is a splitting field, so K/k is normal. Moreover, finite fields are
perfect, so K/k is separable.
Consider Fk ∈ G(K/k) where F = Fk is defined by F(ξ) = ξ #(k) . Look at 1 = F0 , F1 , F2 , . . . , Fn−1 .
r−s
These are distinct, as Fr (θ) = Fs (θ) implies Fr−s (θ) = θ; that is, θq −1 = 1. Yet, q r−s < #(K), a
qn n
contradiction. Now, F (ξ) = ξ . It follows from linear algebra that q = #(K) and by the above, ξ q = ξ
n n
n
implies F = 1. Observe, F(ξ) = ξ when ξ ∈ k, which implies that F is a k-automorphism and F has the
proper order.

Interpretations of the Normal Basis Theorem

(1) Algebraic Interpretation
4.6. PRIMITIVE ELEMENTS, NORMAL BASES 275

Assume K/k is normal and separable, let G = G(K/k) with #(G) = n. We claim that there is a natural
ring homomorphism Y
K ⊗k K −→ K.
σ∈G
Q
(Here σ∈G K consists of n factors of K under coordinatewise multiplication.) Take α, β ∈ K, and send
(α, β) to the n-tuple
hαβ, ασ2 β, . . . , ασn βi,
where G = {σ1 = 1, σ2 , . . . , σn }. This is a bilinear map, so we get a map
Y
K ⊗k K −→ K.
σ∈G

On the left hand side, we have a K-vector space via α ∈ K acts as α ⊗ 1. The righthand side is a K-vector
space via the action of K on each factor; thus, the above map is K-linear. We also have

(α ⊗ β)(γ ⊗ δ) = (αγ ⊗ βδ)

(ασβ)σ (γσδ)σ = (αγσ(βδ))σ .

The normal basis theorem says that this ring map is an isomorphism. Say θ is our normal basis element,
then
1 ⊗ θ, 1 ⊗ σ2 θ, . . . , 1 ⊗ σn θ
is a basis for K ⊗k K over K. Now, as
1 ⊗ τ θ 7→ hστ θiσ∈G ,
a basis on the left hand side goes to a basis on the right hand side; so, the map is an isomorphism. Check
the converse.
(2) Geometric Interpretation
Say X is a space; G is a group, and suppose G acts on X: There is a map G × X −→ X denoted
(σ, x) 7→ σx.

Definition 4.13 A space X is a principal homogeneous space for G (PHS for G) if

(1) X is a homogeneous space, i.e., for all x, y ∈ G, there is some σ ∈ G with σx = y (G acts transitively),
i.e., X is equal to an orbit of G under the action.

(2) The group element σ ∈ G in (1) is uniquely determined by x and y.

Proposition 4.53 The following statements are equivalent:

(A) X is a PHS for G.

Q Q
(B) The map G X −→ X X via (σ, x) 7→ (σx, x) is an isomorphism.
Q
Proof . (A) ⇒ (B). Given (ξ, η) ∈ X X, there is a σ ∈ G with σξ = η. Thus, (σ, ξ) 7→ (η, ξ) under our
map, which shows its surjectivity. The map is injective by property (2) of the definition.
(B) ⇒ (A). This is a tautology.
Let G be a group and let k be a field. Write A(G) for the k-algebra of all functions f : G → k under
pointwise operations (e.g., (f g)(σ) = f (σ)g(σ), etc.). The k-algebra A(G) has a basis,
{eσ }, where eσ (τ ) = δστ .
276 CHAPTER 4. FIELDS AND GALOIS THEORY

Suppose now G is a finite group, then there is a k-algebra map ∆ : A(G) → A(G) ⊗k A(G) given by
(convolution) X
∆(eτ ) = eσ ⊗ eσ−1 τ .
σ∈G

I claim: For all k-algebras, R,

A(G)(R) = Homk (A(G), R)
is a group. Given ϕ, ψ ∈ A(G)(R), we define ϕψ as the composition
∆ ϕ⊗ψ mult
A(G) −→ A(G) ⊗k A(G) −→ R ⊗k R −→ R.

Let us see what (ϕψ)(eρ ) is. We have

X X
∆(eρ ) = eσ ⊗ eσ−1 ρ and (ϕ ⊗ ψ)(∆(eρ )) = ϕ(eσ ) ⊗ ψ(eσ−1 ρ ),
σ∈G σ∈G

so X
(ϕψ)(eρ ) = ϕ(eσ )ψ(eσ−1 ρ ).
σ∈G

(Note: We can form k[G] = the group algebra and the reader should check that:

(1) As linear spaces, A(G) and k[G] are naturally dual.

(2) Multiplication in A(G) goes over to ∆ for k[G] and ∆ for A(G) goes over to ordinary multiplication in
k[G].)

The space Spec A(G) = G is a geometric

P object (at least it’s a topological space). Indeed, it is described
by the equations Xσ Xτ = δστ Xσ and σ∈G Xσ = 1 (the eσ have been replaced by the Xσ for convenience
of more usual notation). To find solutions in a ring R is to give a homomorphism A(G) −→ R, as above. If
Spec R is connected (i.e., e2 = e implies e = 0 or e = 1) then solutions correspond just to the set G and we
recover the multiplication in G from our funny multiplication using ∆.
We know that Y
Spec(B ⊗A C) = Spec B Spec C.
Spec A

The meaning of this is exactly that

Y
HomA-alg (B ⊗A C, R) = HomA-alg (B, R) HomA-alg (C, R),

where on the right we have the ordinary cartesian product of sets.

Look at A(G)⊗k K, where G = G(K/k). Remember, A(G) has the eσ ’s and K ⊗k K has the gσ (X) = eσ ’s,
too. So, there is an isomorphism of rings

A(G) ⊗k K ∼
= K ⊗k K.

Upon taking Spec’s we see that Y Y

G Spec K ∼
= Spec K Spec K.
Therefore, the fact Spec K is a PHS for G is exactly the normal basis theorem.
4.7. GALOIS COHOMOLOGY, NORMS AND TRACES 277

4.7 Galois Cohomology, Norms and Traces

Recall that in Chapter 1, Section 1.4, we introduced the notion of cohomology of a group, G, with coefficients
in a G-module, M . I urge you to review the appropriate parts of Section 1.4 now.
If 0 −→ M 0 −→ M −→ M 00 −→ 0 is an exact sequence of G-modules, then, for each r ≥ 0, the sequence

0 −→ C r (G, M 0 ) −→ C r (G, M ) −→ C r (G, M 00 ) −→ 0

is again exact and a commutative diagram of G-modules

0 / M0 / M / M 00 / 0

  
0 / N0 / N / N 00 / 0

yields a similar commmutative diagram

0 / C r (G, M 0 ) / C r (G, M ) / C r (G, M 00 ) / 0

  
0 / C r (G, N ) / C r (G, N ) / C r (G, N 00 ) / 0

for all r ≥ 0. We’ll see in the next chapter (Chapter 5, Lemma 5.7 and Corollary 5.8) that these statements
imply the following facts:
Fact I. If 0 −→ M 0 −→ M −→ M 00 −→ 0 is an exact sequence of G-modules, then we have the long
exact sequence of cohomology

BC
ED
0 / H 0 (G, M 0 ) / H 0 (G, M ) / H 0 (G, M 00 )

GF
@A
(0)
δ

BC
ED
/ H 1 (G, M 0 ) / H 1 (G, M ) / H 1 (G, M 00 )

GF
@A
δ (1)

ED
BC
/ H 2 (G, M 0 ) / ··· ···

GF
@A
δ (r−1)

ED
BC
/ H r (G, M 0 ) / H r (G, M ) / H r (G, M 00 )

GF
@A
δ (r)

/ H r+1 (G, M 0 ) / ···

(The maps δ (r) are the connecting homomorphisms of the long exact sequence.)
Fact II. A small commutative diagram of G-modules

0 / M0 / M / M 00 / 0

  
0 / N0 / N / N 00 / 0
278 CHAPTER 4. FIELDS AND GALOIS THEORY

yields a large (long) commutative diagram of cohomology:

0 / H 0 (G, M 0 ) / ··· / H r (G, M ) / H r (G, M 00 ) / H r+1 (G, M 0 ) / ···

   
0 / H 0 (G, N 0 ) / ··· / H r (G, N ) / H r (G, N 00 ) / H r+1 (G, N 0 ) / ···

The proofs of these facts do not use any of the material below, so we will assume them now without
circularity in our reasoning.
Suppose B is an abelian group. We can make, from B, a G-module, Map(G, B), as follows:

Map(G, B) = {f | f : G → B, i.e., f is a function from G to B}.

The module structure is

(σf )(τ ) = f (τ σ)
and one checks that if B is actually a G-module, there is a G-module injection

B : B → Map(G, B)

given by
B (b)(σ) = σb. (DX)

The module Map(G, B) is special in that it is “cohomologically trivial.” This is

Proposition 4.54 For every abelian group, B and every n > 0, we have

H n (G, Map(G, B)) = (0).

Proof . Choose f ∈ Z n (G, Map(G, B)) and assume n > 0. Then f is a function of n variables chosen from
G and has values in Map(G, B). We define a function, g, of n − 1 variables chosen from G with values in
Map(G, B) as follows:
g(σ1 , . . . , σn−1 )(τ ) = f (τ, σ1 , . . . , σn−1 )(1).
Let us prove that δg = f , which will finish the argument.
n−1
X
(δg)(σ1 , . . . , σn ) = σ1 g(σ2 , . . . , σn ) + (−1)r g(σ1 , . . . , σr σr+1 , . . . , σn ) + (−1)n g(σ1 , . . . , σn−1 ).
r=1

So, upon evaluating δg on an arbitrary element, τ , we get

n−1
X
(δg)(σ1 , . . . , σn )(τ ) = g(σ2 , . . . , σn )(τ σ1 ) + (−1)r g(σ1 , . . . , σr σr+1 , . . . , σn )(τ ) + (−1)n g(σ1 , . . . , σn−1 )(τ )
r=1
n−1
X
= f (τ σ1 , σ2 , . . . , σn )(1) + (−1)r f (τ, σ1 , . . . , σr σr+1 , . . . , σn )(1) + (−1)n f (τ, σ1 , . . . , σn−1 )(1).
r=1

Now, f (σ1 , . . . , σn )(τ ) = (τ f )(σ1 , . . . , σn )(1), and

n
X
0 = δf (τ, σ1 , . . . , σn ) = (τ f )(σ1 , . . . , σn ) − f (τ σ1 , σ2 , . . . , σn ) + (−1)s f (τ, σ1 , . . . , σs−1 σs , . . . , σn )
s=2
+(−1)n+1 f (τ, σ1 , . . . , σn−1 ).
4.7. GALOIS COHOMOLOGY, NORMS AND TRACES 279

Therefore,
n
X
(τ f )(σ1 , . . . , σn ) = f (τ σ1 , σ2 , . . . , σn ) + (−1)s−1 f (τ, σ1 , . . . , σs−1 σs , . . . , σn ) + (−1)n f (τ, σ1 , . . . , σn−1 ).
s=2

Let n = s − 1 in the sum above and evaluate both sides at 1. We get immediately

f (σ1 , . . . , σn )(τ ) = δg(σ1 , . . . , σn )(τ ).

Proposition 4.54 is extremely useful and very powerful. Rather than explain this in abstract terms, let’s
begin to use Proposition 4.54 and, in so doing, show how to use it and why it is powerful. One of the facts
left unproved in Chapter 1 was the fact that H r (G, M ) is #(G)-torsion if r > 0 (any module, M ). Based
on Proposition 4.54, we can now prove this and, while our proof is not the most elegant known, it certainly
requires the least machinery:

Proposition 4.55 If G is a finite group and M is any G-module, then H r (G, M ) is #(G)-torsion if r > 0.

Proof . Take the case r = 1, first. If f ∈ Z 1 (G, M ), we know

f (σρ) = σf (ρ) + f (σ).

P
Write α for the element − ρ∈G f (ρ) of M . We compute σα:
X X
σα = − σf (ρ) = − (f (σρ) − f (σ))
ρ∈G ρ∈G
X
= − f (σρ) + #(G)f (σ)
ρ∈G
= α + #(G)f (σ).

Therefore, (#(G)f )(σ) = (δα)(σ), and the case r = 1 is done.

Now, use induction on r—here is where Proposition 4.54 enters. Assume as induction hypothesis that
given r (r ≥ 1), for all modules, N , we have H r (G, N ) is #(G)-torsion. The step from r to r + 1 goes like
this:
Choose M , embed M in Map(G, M ), to get
M 
0 −→ M −→ Map(G, M ) −→ coker −→ 0.

Apply cohomology (i.e., use the long exact sequence of Fact I), we get

. . . −→ H r (G, Map(G, M )) −→ H r (G, coker) −→ H r+1 (G, M ) −→ H r+1 (G, Map(G, M )) −→ . . . (∗)

The ends of (∗) vanish by Proposition 4.54 and we obtain the isomorphism

H r (G, coker) −
g→ H r+1 (G, M ), for all r ≥ 1. (∗∗)

But, the left side of (∗∗) is #(G)-torsion by our induction hypothesis, therefore H r+1 (G, M ) is also #(G)-
torsion.
The special case when G is cyclic is both instructive and important for some material to follow. For
arbitrary (finite) G, and any G-module, M , we define the norm map, NG , taking M to itself by
X
NG (m) = σm.
σ∈G
280 CHAPTER 4. FIELDS AND GALOIS THEORY

Note that the image of NG lies in M G . Further, NG is actually a G-module map, for
X
NG (τ m) = στ m = NG (m) = τ NG (m).
σ∈G

(In cases of interest below, the map NG is usually called trace and when M is written multiplicatively then
NG is called the norm.) Now, the equation NG (τ m) = τ NG (m) shows that the elements τ m − m all lie in
Ker NG . The submodule generated by all τ m − m, as τ runs over G and m over M , is denoted IM ; so,
IM ⊆ Ker NG (M ).

Proposition 4.56 If G is a (finite) cyclic group and σ is one of its generators, then for any module, M :
(a) The map f 7→ f (σ) ∈ M is a G-isomorphism of Z 1 (G, M ) with Ker NG (M ),
(b) The submodule IM is generated by σm − m for this fixed σ and m varying over M ,
(c) There is an isomorphism H 1 (G, M ) −
g→ Ker NG /IM .

Proof . The elements of G are 1, σ, . . . , σ n−1 . Let f ∈ Z 1 (G, M ), so f (ρτ ) = ρf (τ ) + f (ρ) for all ρ and τ of
G. Apply this successively to the powers of σ:

f (σ 2 ) = f (σσ) = σf (σ) + f (σ); f (σ 3 ) = f (σσ 2 ) = σf (σ 2 ) + f (σ) = σ 2 f (σ) + σf (σ) + f (σ), etc. (∗)

We find that
f (1) = f (σ n ) = σ n−1 f (σ) + σ n−2 f (σ) + · · · + f (σ) = NG (f (σ)).
But, f (1) = f (1 · 1) = f (1) + f (1); so, f (1) = 0. Thus, when f ∈ Z 1 (G, M ), we get f (σ) ∈ Ker NG (M ).
From (∗) above, we see that f (σ) determines f when f is a cocycle, conversely an easy argument using
the inductive definition of f (σ i ) given by (∗) (namely, σf (σ i−1 ) + f (σ)) shows that if f (σ) ∈ Ker NG our
definition makes f a 1-cocycle (DX). This gives an abelian group isomorphism Z 1 (G, M ) − g → Ker NG . Since
Z 1 (G, M ) is a G-module via M , the map is a G-module isomorphism, and (a) is proved.
To prove (b), all we need to show is that τ m − m is in the submodule generated by σ m
e −m
e as m
e ranges
over M , where τ is a fixed arbitrary element of G. But, τ = σ i ; so,

τ m − m = σ i m − m = σ i m − σ i−1 m + σ i−1 m − m = σ i−1 (σm − m) + σ i−1 m − m.

A clear induction finishes the argument.

(c) The group B 1 (G, M ) consists exactly of those f for which f (τ ) = τ m − m for some m ∈ M . Hence,
f (σ) = σm − m ∈ IM and part (a) now shows that in the isomorphism Z 1 (G, M ) − g → Ker NG the subgroup
B 1 (G, M ) corresponds to IM ; (c) is thereby proved.

Given a finite normal (field) extension K/k, we can consider the cohomology groups of the Galois group
G = G(K/k). These cohomology groups give a sequence of very interesting invariants of the layer K/k. As
nomenclature, the groups H r (G(K/k), M ) are called the Galois cohomology groups of K/k with values in
M , and as notation we write H r (K/k, M ) for H r (G(K/k), M ). Probably, the most useful facts about Galois
cohomology are the two forming the statement of the next proposition.

Proposition 4.57 (Hilbert Theorem 904 .) If K/k is a finite normal extension, then
(1) H r (K/k, K + ) = (0), all r > 0 and
(2) H 1 (K/k, K ∗ ) = (0).
4 When K/k is a cyclic extension, statement (2) is the essential content of Theorem 90 (§54) of Hilbert’s magnificient paper

[23]. The general case of a normal extension is due to E. Noether.

4.7. GALOIS COHOMOLOGY, NORMS AND TRACES 281

Proof . For (1), we examine the layer K/k (∗) and apply the normal basis theorem to it. I claim that, as
G = G(K/k)-modules, Map(G, k (∗) ) and K are isomorphic. If we show this, then Proposition 4.54 and our
isomorphism establish (1).
P
If f ∈ Map(G, k (∗) ), we send f to σ∈G f (σ)σ −1 θ, where θ is a normal basis element for K/k (∗) . The
linear independence of the elements {σθ}σ∈G shows our map is injective; that it is surjective is obvious. As
for the G-action, call our map Θ then,
X X
Θ(τ f ) = (τ f )(σ)σ −1 θ = f (στ )σ −1 θ
σ∈G σ∈G
X
= f (ρ)τ ρ−1 θ
ρ∈G
X
= τ· f (ρ)ρ−1 θ
ρ∈G
= τ Θ(f ),

as contended.
The proof of (2) has a similar flavor but depends on Dedekind’s theorem (our Theorem 4.30). We take
as family of characters of K ∗ the elements of G = G(K/k). By Dedekind’s theorem, they are independent;
that is, any relation (with xσ ∈ K ∗ )
X
xσ σ(λ) = 0, all λ ∈ K ∗
σ∈G

necessarily implies that all the xσ = 0. Given f ∈ Z 1 (K/k, K ∗ ), take as the xσ the elements f (σ) ∈ K ∗ .
None of the xσ are zero, so there must be a λ ∈ K ∗ with
X
β= f (σ)σ(λ) 6= 0.
σ∈G
P
Now, τ β = σ∈G τ f (σ)τ σ(λ), and as f is a 1-cocycle, we have τ f (σ) · f (τ ) = f (τ σ). Thus,
X X
β= f (τ σ)(τ σ)(λ) = (τ f (σ) · f (τ ))(τ σ)(λ)
σ∈G σ∈G
X
= f (τ ) τ f (σ)(τ σ)(λ)
σ∈G
= f (τ ) · (τ β).

Let α = 1/β, then (τ α)/α = f (τ ), as required.

Remark: Proposition 4.57 gives yet another interpretation of the normal basis theorem. It shows that for
K normal over k, the G(K/k)-module K is of the form Map(G, −); namely, it is Map(G, k (∗) ).

Norms and Traces.

If K/k is a field extension and α ∈ K, then the k-vector space map

Tα : K → K via Tα (λ) = αλ

has a trace and a determinant.

Definition 4.14 The trace, trK/k (α), of α from K to k is the trace of Tα ; the norm, NK/k (α), of α from
K to k is det Tα .
282 CHAPTER 4. FIELDS AND GALOIS THEORY

The following three facts are extremely simple to prove and are left as (DX):
Fact I. TrK/k (α) is additive; NK/k (α) is multiplicative.
Fact II. If α ∈ k, then

TrK/k (α) = [K : k]α and NK/k (α) = α[K : k] .

Fact III. If L ⊇ K ⊇ k, then

TrL/k (α) = TrK/k (TrL/K (α)) and NL/k (α) = NK/k (NL/K (α)).

Of course, from Facts II and III , we find

TrK/k (α) = [K : k(α)]Trk(α)/k (α) and NK/k (α) = (Nk(α)/k (α))[K : k(α)] .

When K/k is normal, more can be said. First, assume K/k is both normal and separable, then
X Y
TrK/k (α) = (σα) and NK/k (α) = (σα).
α∈G(K/k) α∈G(K/k)

Both of these statements are very easy corresponding to the fact that the roots of the characteristic polyno-
mial of Tα are exactly the various σα as σ ranges over G(K/k).
Now allow inseparability. We have

NK/k (α) = NK(∗) /k (NK/K(∗) (α)) = Nk(∗) /k (NK/k(∗) (α)).

Hence, we must first investigate NK/k (α) when K/k is purely inseparable. I claim the value of this norm is
α[K : k] . To see this, observe that

NK/k (α) = Nk(α)/k (NK/k(α) (α)) = Nk(α)/k (α)[K : k(α)] . (†)

Now, the minimal and characteristic polynomials for Tα on the vector space k(α) are X q − c, where
q = [k(α) : k] = pr , and c = αq . Here, p = char(k). Therefore, the norm of α is det(Tα ) = c if p is odd and
−c = c if p is 2. Hence, Nk(α)/k (α) = αq = α[k(α) : k] . Put this together with (†) above and obtain our claim.
The general case now is

NK/k (α) = NK(∗) /k (α[K : k]i ) = (NK/k(∗) (α))[K : k]i .

Proposition 4.58 (Original Form of Hilbert Theorem 905 .) Suppose that K/k is normal and that K(∗) /k
is a cyclic extension. Then, a necessary and sufficient condition that NK/k (α) = 1 is that there exists a
β ∈ K(∗) so that
σβ
α[K(∗) (α) : K(∗) ] = .
β
Here, σ is an a priori chosen generator of G(K(∗) /k).

Proof . This is merely the confluence of Propositions 4.56 and 4.57. If α ∈ K(∗) , statements (b) and (c)
of Proposition 4.56 and (2) of Proposition 4.57 give the statement that NK(∗) /k (α) = 1 iff α = σβ/β for
some β ∈ K(∗) . But, NK/k (α) = (NK(∗) /k (α))[K : K(∗) ] in this case, and [K : K(∗) ] is a p-power. Therefore,
NK/k (α) = 1 iff NK(∗) /k (α) = 1.
5 Of course, Hilbert dealt only with the separable case.
 4.7. GALOIS COHOMOLOGY, NORMS AND TRACES 283

/ K(∗) . Then, α[K(∗) (α) : K(∗) ] is in K(∗) . But,

Suppose now that α ∈ K yet α ∈

NK/k (α) = NK(∗) /k (NK(∗) (α)/K(∗) (α))[K : K(∗) ] .

As [K : K(∗) ] is a p-power, the left hand side is 1 iff NK(∗) /k (NK(∗) (α)/K(∗) (α)) is 1. By our remarks above,
this last quantity is exactly NK(∗) /k (α[K(∗) (α) : K(∗) ] ); so, we can apply the first part of the proof to the
element α[K(∗) (α) : K(∗) ] , and we are done.

 It is not clear that β in Proposition 4.58 is of the form γ q (where q = [K(∗) (α) : K(∗) ]), because in the
proof of Proposition 4.57 part (2), the element λ may not be a qth power. If it proves to be so, then α
would be σγ/γ.
284 CHAPTER 4. FIELDS AND GALOIS THEORY

4.8 Krull’s Galois Theory

In our treatment of Galois theory, the extensions were assumed finite. W. Krull discovered a natural way to
treat (possibly) infinite algebraic extensions, His method leads to a non-trivial topology on the Galois group.
We begin with the generalization of the extension lemma.

Theorem 4.59 (General Extension Lemma) Suppose K/k is an algebraic extension and e k is another field
isomorphic to k via θ : k → ek. Let Γ be a field related to e
k, but otherwise arbitrary. Then, there exists an
e e e e e
algebraic extension, K/ k, with K rel Γ and an extension of θ to an isomorphism θ : K → K.
f

KO
θe / K
e
rel Γ
O f
alg alg

k / e
k f
θ rel Γ

Proof . This is a standard use of Zorn’s lemma. We let

e | L/k is algebraic, ϕ extends θ and is an isomorphism L −→ L
S = {(L, ϕ, L) e and L
e
rel Γ}.
f
e in S are automatically algebraic over e
Notice that the L k. We partially order S via the usual:
e ≤ (M, ψ, M
(L, ϕ, L) f) iff e⊆M
L ⊆ M; L f; ψ  L = ϕ.

e 0 ) be a maximal element of S. Were L0 6= K, there would be some

Of course, S is inductive; so, let (L0 , ϕ0 , L
α ∈ K with α ∈ / L0 . Then, the extension lemma for the finite extension L0 (α)/L0 would yield (L0 (α), ϕ e0 )
e0 , L
an element in S bigger that our maximal element—a contradiction. Therefore, L0 = K.
The material on splitting fields, etc. of Section 4.4 carries over provided no statement involving finiteness
is used (e.g., statement (3) of Proposition 4.25 would be omitted in the general case that M/k was algebraic,
not necessarily finite). The corollaries SMA, I and SMA II (Corollary 4.27 and Corollary 4.28 ) go over as
does the existence of a normal closure.

Proposition 4.60 Suppose K/k is an algebraic extension and write {Kα /k | α ∈ Λ} for the family of
sub-extensions of K/k of finite degree. Then, our family is a right mapping family in a natural way and we
have
K = lim Kα .
−→
α

If K/k is normal, we may restrict the Kα /k to the finite normal extensions. Conversely, if K = lim Kα
−→
α
and each Kα is normal over k, then so is K.

Proof . Of course, we define α ≤ β (in Λ) when and only when Kα ⊆ Kβ (everything takes place inside
K). The map Kα −→ Kβ is the inclusion. Since we have the inclusions Kα ,→ K, consistent with the
Kα −→ Kβ , we get the canonical homomorphism

lim K −→ K.
−→ α
α

Choose ξ ∈ K. Then k(ξ) is some Kα , and it is clear that ξ 7→ canα (ξ) ∈ lim Kα is well-defined and provides
−→
α
an inverse map to that above.
Of course, the family of finite normal extensions Mα /k is final in the family of all finite extensions
provided K/k is itself normal. So, all we need prove is the last statement. We have K = lim Kα and each
−→
α
4.8. KRULL’S GALOIS THEORY 285

Kα is normal over k. If ξ ∈ K, there is an α so that ξ ∈ Kα . Then all the k-conjugates of ξ lie in Kα ; hence,
they are in K.
If Ω is an algebraic normal extension of K, then we consider the group Autk (Ω). We topologize Autk (Ω)
by taking as a fundamental set of neighborhoods about 1 the subgroups of finite index in Autk (Ω). Of course,
it is the same to take the normal subgroups of finite index as our basic neighborhoods of {1} in Autk (Ω).
(Remember: To get the neighborhoods about σ ∈ Autk (Ω), we take the cosets σH, where the H are our
neighborhoods about 1.) This renders Autk (Ω) a Hausdorff topological group (use ordinary Galois Theory
to see this) and it is this group together with its topology that we call the Galois group of Ω over k and
denote by G(Ω/k). The topology itself is the Krull topology.
Theorem 4.61 The group G = G(Ω/k) is compact and totally disconnected in its Krull topology. In fact,
we have G(Ω/k) = lim G/H, where the left limit is taken over all open subgroups, H, of G. Thus, G(Ω/k) is
←−
H
a profinite group. Moreover, if we write Ω = lim Ωα , where each Ωα is a finite normal extension of k, then
−→
α
G(Ω/k) = lim G(Ωα /k).
←−
α

Proof . If σ ∈ G(Ω/k) and Ωα is one of the finite normal subextensions of Ω/k, then σ  Ωα is in G(Ωα /k).
The maps πα : G(Ω/k) → G(Ωα /k) are consistent and hence we obtain the commutative diagram

G(Ω/k)
ϕ
/ lim G(Ωβ /k)
II ←−
II
II r
β
II rr
II rr
xrrr canα
πα
I$
G(Ωα /k)

If ξ ∈ lim G(Ωβ /k), then ξ consists in a collection (ξβ ) where ξβ ∈ G(Ωβ /k) and when k ⊆ Ωβ ⊆ Ωγ , we
←−
β
have ξγ  Ωβ = ξβ . Since each x ∈ Ω lies in some finite normal extension of k, we have x ∈ Ωβ for various
β. Then ξβ (x) is well-defined and our collection (ξβ ) = ξ gives rise to an element of G(Ω/k). Therefore we
have a map
ψ
lim G(Ωβ /k) −→ G(Ω/k)
←−
β

plainly inverse to ϕ. Now, a neighborhood of 1 in lim G(Ωβ /k) consists of those tuples (ξβ ) for which finitely
←−
β
many β are the identity and otherwise arbitrary (though consistent). Such tuples when restricted to the
compositum of the Ωβ for which ξβ = 1, are the identity on the compositum which is a field of finite degree
over k, call it L. I claim G(Ω/L) has finite index in G(Ω/k). For L is normal and the usual argument shows
G(Ω/L) C G(Ω/k). Moreover, SMA II (in its extended form) implies that G(Ω/k) −→ G(Ω/L) is surjective.
Hence, the exact sequence
0 −→ G(Ω/L) −→ G(Ω/k) −→ G(L/k) −→ 0
gives the finite index assertion immediately. But then, we see that open neighborhoods of 1 in the Krull
topology on G(Ω/k) correspond to open neighnorhoods of 1 in the natural (product) topology on lim G(Ωβ /k).
←−
β
Consequently, our maps ϕ and ψ are homeomorphisms.
Since G(Ωβ /k) is compact, so is G(Ω/k) in the Krull topology, and of course G(Ω/k) is a profinite group.
Every profinite group is totally disconnected (DX); so, G(Ω/k) is totally disconnected.
That G is lim G/H as H ranges over all open normal subgroups of G is the same kind of argument
←−
H
(remember that H will be closed and of finite index). Or, it follows immediately from the next lemma whose
proof is easy.
286 CHAPTER 4. FIELDS AND GALOIS THEORY

Proposition 4.62 Suppose G is a compact (Hausdorff ) group and Gα , Hα are two families of closed subgroups
with Hα C Gα for every α. Assume that the indices α, β, . . . form a directed set and that for every β ≥ α we
have Gβ ⊆ Gα and Hβ ⊆ Hα . Then, the groups Gα /Hα form an inverse mapping family in a natural way and
\ .\
lim Gα /Hα = Gα Hα .
←−
α α α

Using the same notations as in our treatment of standard Galois Theory, we can now extend the fun-
damental theorem to the general case. First of all, Lemma 4.33 and the material on Galois equivalence
(between Lemma 4.33 and Theorem 4.38) go over word for word (together with no change in their proofs).
So, here is the theorem.

Theorem 4.63 (Fundamental Theorem of Galois Theory, General Case) If Ω/k is a normal (not necessarily
finite) algebraic extension, then the mappings

[L] 7→ G(Ω/L)
H 7→ [Fix(H)]

establish a one-to-one order-inverting correspondence between Galois classes of extension fields of k and
closed subgroups of G(Ω/k). In this correspondence:
(a) L(∗) is normal over k iff L(∗) is normal over k iff G(Ω/L) is a normal subgroup of G(Ω/k).
(b) Under the conditions of (a), we have a natural exact sequence

0 −→ G(Ω/L) −→ G(Ω/k) −→ G(L(∗) /k) −→ 0

of compact topological groups.

(c) A necessary and sufficient condition that L(∗) be a finite extension of k is that G(Ω/L) be an open
subgroup of G(Ω/k). In this case,

(G(Ω/k) : G(Ω/L)) = [L(∗) : k].

Proof . If α ∈ Ω, I claim {σ | σ(α) = α} is an open (hence closed) subgroup of G(Ω/k). Notice that if this
claim is proved, then
\
G(Ω/L) = {σ | (∀α ∈ L)(σ(α) = α)} = {σ | σ(α) = α}
α∈L

is a closed subgroup. Now, k(α) has finite degree over k, so its normal closure, L, also has finite degree. In
the proof of Theorem 4.61, we showed G(Ω/L) has finite index in G(Ω/k). But,

G(Ω/L) ⊆ G(Ω/k(α)) ⊆ G(Ω/k)

and therefore (G(Ω/k) : G(Ω/k(α))) < ∞. By definition of the Krull topology, the subgroup G(Ω/k(α)) is
open, as contended.
Next, just as in the usual (finite) case we see that G(Ω/L(∗) ) = G(Ω/L) and if L = Fix(G), then L = L(∗) .
So, if we start with [L], then we get G(Ω/L) which is G(Ω/L(∗) ). However, as mentioned, Fix(G(Ω/L(∗) )) is
(L(∗) )(∗) = L(∗) and so the correspondence inverts if we start from the field side.
Now take a closed subgroup H and form [Fix(H)]. If L = Fix(H), consider G(Ω/L). Now L = L(∗) , so in
what follows we consider only those subfields, M , of Ω/k with M = M (∗) . The Galois group don’t change
and all fields are now separable over the base field, k (∗) . For notation, drop all mention of “upper stars.”’
4.8. KRULL’S GALOIS THEORY 287

We must show H = G(Ω/L). We know H ⊆ G(Ω/L) by definition of L. Observe that Ω = lim Kα for
−→
α
fields, Kα , finite and normal over k. We find as well that L = lim Kα ∩ L. The Galois group G(Ω/Kα )
−→
α
is then an open, normal subgroup of G(Ω/k) by definition of the Krull topology. Consider the subgroups
HG(Ω/Kα ), which contains H. I claim: Fix(HG(Ω/Kα )) is just LKα . For, the elements of HG(Ω/Kα ) are
products στ , where σ ∈ H and τ ∈ G(Ω/Kα ). A ξ in Fix(HG(Ω/Kα )) satisfies στ (ξ) = ξ, for all such σ and
(∗)
τ . In particular, when σ = 1, we find ξ ∈ Fix(G(Ω/Kα ) = Kα (remember: Kα = Kα ), and when τ = 1,
we find ξ ∈ Fix(H) = L; hence, ξ ∈ Kα ∩ L. Conversely, if ξ ∈ Kα ∩ L it is fixed by both H and G(Ω/Kα );
therefore, our cliam is proved. Then, we have the commutative diagram

0 / G(Ω/Kα ) / G(Ω/k) / G(Kα /k) / 0

O O

0 / G(Ω/Kα ) / G(Ω/Kα ∩ L) / G(Kα /Kα ∩ L) / 0,

and from it we see that HG(Ω/Kα ) corresponds to a subgroup of G(Kα /k), via
HG(Ω/Kα ) 7→ HG(Ω/Kα )/G(Ω/Kα ). But then, the lower line of our diagram and the finite case of ordinary
Galois theory show that
HG(Ω/Kα )/G(Ω/Kα ) ∼ = G(Kα /Kα ∩ L).
We pass these isomorphisms to the projective limit over α; on the left hand side, Lemma 4.62 implies that
we get \ \ \
HG(Ω/Kα )/ G(Ω/Kα ) = HG(Ω/Kα )
α α α
T
while on the right hand side we get G(Ω/L). But, α HG(Ω/Kα ) is the closure of H and H is already closed.
Therefore, H = G(Ω/L).
The proofs of assertions (a) and (b) are now just as they were in the finite case. As for (c), we know that
G(Ω/L(∗) ) = G(Ω/L) and that this subgroup is of finite index in G(Ω(∗) /k). We take, as in the proof above,
a family of fields, Kα , of finite degree and normal over k so that lim Kα = Ω. Then
−→
α

(G(Kα /k) : G(Kα /Kα ∩ L(∗) )) = [Kα ∩ L(∗) : k]

by usual Galois theory. Pass to the limit over α, observe that the left side tends to (G(Ω/k) : G(Ω/L)) and
in fact is constant as soon as Kα ⊇ L(∗) , and we get (c).
We can now extend the notions and results of the previous section on Galois cohomology to the general
(not necessarily finite) case. All that is necessary is to sprinkle the word “continuous” in the appropriate
places and use some case. The group G will be a profinite group, for example G = G(Ω/k). All modules
will be given the discrete topology unless otherwise noted and our action G × M −→ M will be assumed
continuous. This means that for m ∈ M , there is an open subgroup, U , of G so that U m = m. We define

C n (G, M ) = {f : Gr → M | f is continuous},

and use the usual formula for δ, thus δ is a continuous function (C r (G, M ) inherits the discrete topology from
M as G is compact). The continuity of the cochains shows up as follows: If N C G (N is, of course, closed),
then M N is a G/N -module. If N e ⊆ N and N e C N , then there are maps G/N e −→ G/N and M N −→ N Ne .
The latter two combine to give a map

e, MN )
C r (G/N, M N ) −→ C r (G/N
e

e ; it is injective. We have
called inflation from G/N to G/N
288 CHAPTER 4. FIELDS AND GALOIS THEORY

Proposition 4.64 For a continuous G-module, M , for the profinite group, G, the modules C r (G/U, M U )
form a right mapping system as U runs over the open normal subgroups of G—the map being inflation. We
have
C r (G, M ) = lim C r (G/U, M U ),
−→
U
and passing to cohomology, we also have

H r (G, M ) = lim H r (G/U, M U ).

−→
U

The proof of this is now routine and may be safely left to the reader (DX). The cohomological triviality
of Map(G, B) (continuous functions, of course) expressed by Proposition 4.54 carries over and so does
Proposition 4.57 (Hilbert Theorem 90).
4.9. KUMMER THEORY 289

4.9 Kummer Theory

In this section we consider base fields containing prescribed roots of unity. At first, we assume our base field
contains a primitive mth root of unity. Notice that something mildly subtle is happening. We are not merely
assuming all mth roots of 1 lie in k, for that would be true is m = char(k) > 0, yet there is no primitive mth
root of unity in this case because 1 is the only mth root of unity when m = char(k) > 0. When a primitive
mth root of 1 lies in k, then necessarily (char(k), m) = 1. (Else, p = char(k) | m and X m − 1 = (X q − 1)p ,
when m = pq. So each mth root of 1 is already a qth root of 1 with q < m, contradicting primitivity.)

Proposition 4.65 Suppose the field k contains a primitive mth root of 1. A necessary and sufficient con-
dition that K/k be a normal, separable extension whose Galois group is cyclic of order m is that K = k(β)
where the minimal k-polynomial for β is X m − b.

Proof . (⇐). We assume K = k(β) and β is a root of the irreducible k-polynomial X m − b. Write µ for a
primitive mth root of 1 in k, then 1, µ, . . . , µm−1 are all the mth roots of 1, they are all distinct and lie in
k. Of course, (m, char(k)) = 1 shows K/k is separable and [K : k] = m. Now, the elements

β, µβ, µ2 β, . . . , µm−1 β

are all distinct and all are roots of X m − b, therefore K is a splitting field of X m − b; so, K/k is indeed
normal. If we consider the k-isomorphism k(β) −→ k(µβ), we see (even without SMA, I) that it gives an
element, σ, of G(K/k). The powers of σ operate on β via

σ r (β) = µr β

and so 1, σ, . . . , σ m−1 are m distinct elements of G(K/k). They thereby exhaust G(K/k) and (⇐) is proved.
(⇒). We suppose here that K/k is normal, separable and G(K/k) is cyclic of order m = [K : k]. Now
we know NK/k (µ) = µm = 1, so we can apply the original form of Hilbert Theorem 90. We find there exists
a β ∈ K so that σβ = µβ, where σ is a generator of G(K/k). Then, of course, σ r β = µr β, and so β is left
fixed only by the trivial subgroup of G(K/k). By the fundamental theorem of Galois Theory, k(β) = K.
The minimal k-polynomial of β is then
m−1
Y m−1
Y
(X − σ r β) = (X − µr β).
r=0 r=0

On the other hand, if b = β , then each σ fixes b and each µr β is a root of X m −b. The minimal polynomial
m r

for β and X m − b both have degree m; so it is clear the latter polynomial is the minimal polynomial for β.

Corollary 4.66 Suppose the field k contains a primitive mth root of 1. If n is any divisor of m, then
(1) A n.a.s.c. that an extension K/k of degree n be normal with cyclic Galois group is that K = k(α)
where the minimal polynomial of α is X n − a.
(2) The k-polynomial X m − a is irreducible in k[X] if and only if for all divisors, d, of m with d > 1, we
/ k∗ d .
have a ∈

Proof . First, as n | m, we can write m = nd. Then for our primitive mth root of 1 in k, µ, the element µd
is a primitive nth root of 1 in k and (1) is simply a restatement of Proposition 4.65 with n replacing m.
For statement (2), first consider X m − a and let α be a root in an overfield, Ω, of k. Write d for the
smallest power of α lying in k. Then, the usual division algorithm argument shows that if αq ∈ k, we have
d | q; in particular, d | m. I claim the polynomial X d −b is the minimal k-polynomial for α (here, b = αd ∈ k),
in particular it is irreducible. To see this, let f (X) be the minimal k-polynomial for α and have degree t.
290 CHAPTER 4. FIELDS AND GALOIS THEORY

Thus, f (X) | (X d − b) and t ≤ d. Yet, the roots of X d − b are α, ζα, . . . , ζ d−1 α, where ζ = µn is a primitive
dth root of 1. Thus,
f (X) = (X − ζ i1 α) · · · (X − ζ it α)
Q 
t il
and its constant term is therefore ± l=1 η αt . But then, αt ∈ k; so, d | t and t ≤ d. We find t = d and
f (X) = X d − b.
Now if a ∈/ k ∗ q for any q | m with q > 1, then the smallest power of α to lies in k is the mth. Else,
α ∈ k implies dq = m (as above) and a = αm = (αd )q ∈ k ∗ q and q > 1 if d < m. By our claim, X m − a is
d

irreducible in k[X].
Finally, assume X m − a is irreducible. Were a ∈ k ∗ d where d | m and d > 1, then as αm = a, we have
(α ) = β d for some β ∈ k ∗ . Therefore, αq = zβ for some z a dth root of 1 (hence, in k). It follows that the
q d

smallest power of α in k is, say, δ where δ ≤ q < m. Just as before, δ | m and X δ − b is k-irreducible, where
b = αδ . Write δr = m and look at α, µα, . . . , µr−1 α (as usual µ is our primitive mth root of 1). Each of
these elements has δth power in k and δ is minimal. Set ζ = µδ , then X δ − ζ i b is the minimal k-polynomial
for µi α by our claim above. But,

X m − a = (X δ − b)(X δ − ζb) · · · (X δ − ζ r−1 b),

contradicting the irreducibility of X m − a.

An important part of the proof above should be isolated and recorded:

Corollary 4.67 If k contains a primitive mth root of 1 and K is an overfield of k, then given α ∈ K with
αm ∈ k ∗ , the minimal k-polynomial for α is X d − αd , where d is the smallest positive integer so that αd ∈ k.
In fact, d | m.

Now, we can make an obvious attempt to “classify” the cyclic overfields of degree n (n | m) of k when k
possesses a primitive mth root of 1. Namely, such a K is k(α) and we could send α to αn where αn is the
image of αn in k ∗ /k ∗ n . But, α is not unique and its choice depends on µ and σ (a generator of G(K/k)).
There is a better way:

Theorem 4.68 (Kummer) Suppose k is a field possessing a primitive mth root of 1. Write Ω for the
maximal, abelian, m-torsion extension of k and denote by G its (Krull topologized) Galois group. Then there
is a natural continuous pairing
Y
G k ∗ /k ∗ m −→ µm (= mth root of 1),

and it makes G the Pontrjagin dual of k ∗ /k ∗ m .

Proof . Choose σ ∈ G and a ∈ k ∗ /k ∗ m . Lift a to some a ∈ k ∗ and take an mth root of a in an overfield, call
it α. We know that K = k(α) is cyclic of degree d and d | m by our above propositions. So, K ⊆ Ω (we fix
an algebraic closure of k and work inside it) and σα makes sense. We set
σα
(σ, a) = .
α
Note that  σα m σ(αm ) αm
= m
= m = 1,
α α α
therefore (σ, a) ∈ µm . Let’s check that (σ, a) is well-defined. First, if we change the mth root of a we get ζα
where ζ is some mth root of 1 (hence, ζ ∈ k ∗ ). Then
σ(ζα) σα
= ,
ζα α
4.9. KUMMER THEORY 291

so there is no problem with the choice of α. If we lift a to some b ∈ k ∗ , then b = λm a for some λ ∈ k ∗ . Thus,
β, an mth root of b is ζλα for some ζ as above. Once again,

σβ σ(ζλα) σα
= = ,
β ζλα α

and so (σ, a) is a well -defined mth root of 1.

It is easy to see that (σ, a) is bi-multiplicative (DX), so assume (σ, a) = 1 for all σ ∈ G. If a lifts a and
a∈/ k ∗ m (i.e., a 6= 1) then K = k(α) is a non-trivial cyclic degree d extension of k and d | m. But then, a
generator, τ , of G(K/k) comes from some σ ∈ G and σα = τ α = ζα (some dth root of 1, say ζ 6= 1). Hence,
(σ, a) = ζ 6= 1, a contradiction. Therefore, (σ, a) is non-degenerate on the right.
If (σ, a) = 1 for all a ∈ k ∗ /k ∗ m , then I claim σ must be 1. For, notice that when K/k is finite normal
(K ⊆ Ω), then G(K/k) is an abelian m-torsion group. Hence, G(K/k) is a product of various Z/dZ, where
each d | m. This means that K is generated as a field by elements, α, for which k(α) is a cyclic extension
of k. As K is arbitrary, it follows immediately that Ω is a field generated by such elements α. However,
Proposition 4.65 and our assumption (σ, a) = 1 (all a), now yield σα = α for all the α’s generating Ω. Thus,
σ = 1, as claimed.
Lastly continuity of hσ, ai 7→ (σ, a) follows because on the entire open subgroup G(Ω/k(α)) Q the pairing
h−, ai 7→ (−, a) is identically 1. Here,
Q α is, of course, an mth root of a. The product G(Ω/k(σ)) {a} is an
open neighborhood of 1 in G(Ω/k) k ∗ /k ∗ m .

Corollary 4.69 Under the assumptions and notations of Theorem 4.68, there is a one-to-one correspondence
between subgroups, S, of k ∗ /k ∗ m and sub-extensions K/k of Ω/k. It is given by

S ←→ K = k(S 1/m ).

In all the foregoing, m was relatively prime to char(k) = p > 0. What happens if p | m? Of course, we
can factor m as pr m e p) = 1. It’s not hard to see that the case for this breaks up into the pr case
e with (m,
and the previous case. So, we’ll assume m = pr . Here, we will use the additive part of Hilbert 90 and the
isomorphism (Ker TrK/k )/IK/k K + ∼ = H 1 (K/k, K + ) in case G(K/k) is cyclic.
So, assume K/k is a cyclic extension of degree pr ; choose a generator, σ, of G(K/k). For the element
1 ∈ k + , we have TrK/k (1) = [K : k] · 1 = 0, so there exists θ ∈ K + with

σ(θ) − θ = 1, i.e., σ(θ) = θ + 1.

The action of the Galois group on θ is given by

σ i (θ) = θ + 1 0 ≤ i ≤ pr − 1.

Observe that only θ, θ + 1, . . . , θ + (p − 1) are distinct, after that we repeat these in order. Thus, the
polynomial
g(X) = (X − θ)(X − (θ + 1)) · · · (X − (θ + (p − 1)))
is the minimal k-polynomial for θ and L = k(θ) is a cyclic p-extension. The only case when K = L is when r =
1; so, from now on we’ll assume K/k is cyclic of degree p. Hence, K = k(θ) and σ(θ) = θ+1. We can compute
the minimal polynomial g(X) as follows: Write Y = X − θ, then g(X) = Y (Y − 1)(Y − 2) · · · (Y − (p − 1)).
But, the elements 1, 2, . . . , p − 1 are the (p − 1)st roots of unity (and lie in Fp , the prime field), therefore

g(X) = Y (Y p−1 − 1) = Y p − Y = X p − X − (θp − θ).

If we write ℘(θ) = θp − θ, then we’ve proved the first part of

292 CHAPTER 4. FIELDS AND GALOIS THEORY

Theorem 4.70 (E. Artin & O. Schreier, 1929) If k is a field of characteristic p > 0, then every cyclic
p-extension, K/k, has the form K = k(θ) where ℘(θ) = θp − θ lies in k and the Galois group, G, acts by a
(prechosen) generator, σ, taking θ to θ + 1. The minimal k-polynomial for θ is X p − X − ℘(θ). Conversely,
the polynomial X p − X − a is k-irreducible when and only when a ∈/ ℘(k + ). If it is irreducible and θ is a
root, then k(θ) is a normal, separable, cyclic p-extension of k.

Proof . If a ∈ ℘(k) so that a = bp − b for some b ∈ k, then

(X − b)(X − (b + 1)) · · · (X − (b + (p − 1))

is exactly X p − X − ℘(b) = X p − X − a; so, our polynomial splits in k[X].

/ ℘(k + ), the polynomial X p − X − a has no root in k. Adjoin a root to k, we get an extension
If a ∈
K = k(θ). Now, θp − θ = a, so (θ + i)p − (θ + i) = a, too, where 0 ≤ i ≤ p − 1. Therefore, all the roots of
X p − X − a lie in K and K is a normal extension. But the roots of X p − X − a are all distinct, therefore
X p − X − a is separable and we find that K/k is a normal, separable extension.
If d is the degree of θ over k, then θ, θ + i2 , . . . , θ + id are the roots of its minimal k-polynomial. Were
d 6= p, there would be an integer, j, so that θ + j is not a root of the minimal k-polynomial for θ. Yet,
k(θ + j) = k(θ), so θ + j also has degree d over k and θ + j, θ + j2 , . . . , θ + jd are all the conjugates of θ + j
and all distinct from each θ + il . Continue in this way, we find the p roots θ, θ + 1, . . . , θ + (p − 1) partition
themselves into t blocks of d elements each. But then, dt = p and p is prime. As θ ∈ / k, we have d > 1
therefore d = p and so X p − X − a is indeed irreducible.
The analog of Kummer’s theorem is

Theorem 4.71 (E. Artin & O. Schreier) Suppose k is a field of characteristic p > 0 and write Ω for the
maximal, abelian, p-torsion extension of k. If G = G(Ω/k) is the Galois group of Ω/k (with Krull topology),
then there is a natural continuous pairing
Y
G k + /℘(k + ) −→ Z/pZ (⊆ R/Z)

and it makes G the Pontrjagin dual of k + /℘(k + ).

Proof . Pick σ ∈ G and a ∈ k + /℘(k + ). Lift a to some a ∈ k + and let θ ∈ k be a root of X p − X − a. Define

(σ, a) = σθ − θ.

Note that unless a = 0, the field k(θ) has degree p over k and is normal, separable cyclic. If a = 0, then
θ ∈ k. Therfore, k(θ) is contained in Ω and σθ makes sense. Now σθ is a root of X p − X − a and so σθ = θ + j
for some j ∈ Z/pZ; therefore, (σ, a) is indeed in Z/pZ.
As in the proof of Kummer’s theorem, hσ, ai 7→ (σ, a) is a pairing of the groups G(Ω/k) and k + /℘(k + ).
Just as in the proof of that theorem, the field Ω is generated by the various θ’s as above; so, if (σ, a) = 0
for all a, we find σ fixes all the θ’s and thereby σ = 1. If (σ, a) = 0 for all σ ∈ G(Ω/k), then a must be 0
else the polynomial X p − X − a would be irreducible (Theorem 4.70) and k(θ), where θ is one of its roots,
would be a cyclic p-extension. Then, σθ = θ + 1 for some σ ∈ G(k(θ)/k) and upon lifting σ to G(Ω/k) we’d
get (σ, a) 6= 0, a contradiction.
proved exactly as in Kummer’s theorem, the open neighborhood on which (σ, a) vanishes
Continuity is Q
being G(Ω/k(θ)) {a}.

Corollary 4.72 If char(k) = p > 0 there is a one-to-one correspondence between subgroups, T , of k + /℘(k + )
and p-torsion, abelian overfields of k. It is given by

T ←→ K = k(℘−1 (T )).
4.9. KUMMER THEORY 293

What happens for pr , r > 1? Here, the situation is sufficiently complicated that the solution had to wait
until 1937. Then E. Witt introduced a ring, W (k), called the ring of Witt vectors over k and he proved
that even if char(k) = p > 0, the ring W (k) is an integral domain of characteristic 0. Now, it turns out that

W (k) = lim Wn (k),

←−
n

where the Wn (k) are “truncated” Witt vector rings. There is a map F : Wnn (k) → Wn (k) playing the role of
ξ 7→ ξ p and one gets ℘ = F − id. When n = 1, the ring W1 (k) is just k, and the exact sequence
℘
0 −→ Z/pZ −→ k + −→ ℘(k + ) −→ 0

becomes an exact sequence

℘
0 −→ Z/pr Z −→ Wr (k)+ −→ ℘(Wr (k)+ ) −→ 0,

in the general case. It then turns out that if Ω is the maximal, abelian pr -torsion extension of k, the Galois
group, G(Ω/k), is naturally Pontrjagin dual to Wn (k)/℘(Wn (k)) by a pairing similar to the Artin-Schreier
pairing. See Witt [49] for the details.
294 CHAPTER 4. FIELDS AND GALOIS THEORY

4.10 An Amazing Theorem of Galois Theory

Question: If k is a field with k 6= k, when is [k : k] finite?
An example: k = R; K = C = R(i).
The answer of our question depends on an irreducibility criterion:

Theorem 4.73 (Artin’s Irreducibility Criterion) Given a field, k, consider the polynomial X n − a, where
a ∈ k ∗ . If p is prime and p divides n, assume a ∈
/ (k ∗ )p . If 4 | n, then assume as well that a ∈
/ (−4(k ∗ )4 ).
n
Under these conditions, X − a is irreducible in k[X].

We will assume this theorem for the moment, and based on it we can prove

Theorem 4.74 (Artin) Say k is a field, k is an algebraic closure of k and 1 < [k : k] < ∞. Then we have:
(1) k = k(i) (i2 = −1).
(2) char(k) = 0.

Proof . We claim that k/k is separable.

If not, let k (∗) = L, then L/L is purely inseparable and k 6= L. So, L 6= Lp implies that there is a ∈
/ Lp
pn
(where p = char(k)). We know X − a is irreducible in L[X] implies that L has extensions of degree pn ,
for all n; yet, all these extensions are contained in k, a contradiction.
Look at k(i) ⊆ k; as k/k is separable, k/k(i) is normal, separable. Let G = G(k/k(i)). We need to show
that #(G) = 1.
Pick a prime, p, with p | #(G); let H be the subgroup of G of order p and write L = Fix(H).
Step 1. p 6= char(k).
If p = char(k), then, by separability, there is some β ∈ k so that trk/L (β) = 1. We know that M/L is
separable iff the bilinear form
(u, v) 7→ trM/L (uv)

is non-degenerate on the vector space M over the field L. Now, there is βe so that trk/L (1 · β)
e 6= 0. Let
e ∈ L and form β = (1/λ)β.
λ = tr(β) e Then, we have

e = λ/λ = 1.
trk/L (β) = (1/λ)trk/L (β)

As the trace is a sum and

tr(β p ) = tr(β)p (p = char(k)),
p
we get trk/L (β − β) = 0. Our extension k/L is cyclic of degree p, say σ is a generator of H. Note that for
every ξ ∈ k, we have
trk/L (ξ) = trk/L (σξ),

so trk/L (σξ − ξ) = 0. By “Additive Hilbert 90”, every element of zero trace in k has the form (σγ − γ),
for some γ ∈ k. As trk/L (σξ − ξ) = 0, there is some γ ∈ k so that β p − β = σγ − γ. Now, the polynomial
X p − X − γ ∈ k[X] has a root in k, as k is algebraically closed. Say α ∈ k is such a root, then γ = αp − α.
We have
β p − β = σ(αp − α) − (αp − α) = σ(α)p − αp − (σ(α) − α),
and so
σ(α) − α − β = σ(α)p − αp − β p = (σ(α) − α − β)p .
4.10. AN AMAZING THEOREM OF GALOIS THEORY 295

Consequently, σ(α) − α − β ∈ Fp , call it ν. It follows that σ(α) − α = β + ν. Taking trk/L on both sides, we
get
0 = trk/L (β) + trk/L (ν) = 1 + trk/L (ν).
As ν ∈ Fp ⊆ L, we have
trk/L (ν) = [k : L]ν = pν = 0,
which implies that 0 = 1 + 0 = 1, a contradiction. Therefore, char(k) 6= p, as claimed.
Step 2. L does not exist, i.e., as no p divides #(G), we have #(G) = 1; thus, k = k(i).
We claim that [k : L] = p. Adjoin to L a p-th root of unity, say ζ is such a primitive root. Then,
[k : L(ζ)] | p; so, [L(ζ) : L] also divides p. But, L(ζ)/L has degree at most p − 1. Indeed,

X p − 1 = (X − 1)(X p−1 + X p−2 + · · · + X + 1),

so L(ζ) = L already, i.e., ζ ∈ L and it follows that [k : L] = p. As L has the p-th roots of unity, by Kummer’s
theorem (Theorem 4.65), we know k = l(α), where α is a root of X p − a, with a ∈ / (L∗ )p . But, if p is odd,
pl
the Artin irreducibility criterion implies that X − a is also irreducible for all l ≥ 1, so [k : L] ≥ pl , for all
l ≥ 1, a contradiction. Therefore, we must have p = 2. Now, our situation is

/ (L∗ )2 .
(a) k = L(α), where α is a root of X 2 − a, with a ∈

(b) i ∈ k ⊆ L.
r
Since X 2 − a cannot be irreducible for all r ≥ 0, since otherwise we would have [k : K] ≥ 2r for all r ≥ 0,
∗ 4 4 ∗
√ be that2a ∈ (−4(L ) ) (by Artin’s irreducibility). Thus, a = −4b , for some b ∈ L ; it follows that
it must
α = a = ±2ib . As 2, i, b ∈ L, we deduce that α ∈ L, a contradiction. Therefore, #(G) = 1.
Step 3. char(k) = 0.
If not, then say q = char(k) and write Fq for the prime field of k. Pick r >> 0, adjoin to Fq a primitive
2r -th root of unity, call it ζ. Apply natural irrationalities to the picture show in Figure 4.2:

k = k(i) = k(ζ)

deg 2
k

Fq (ζ)

D = k ∩ Fq (ζ)

Fq

Figure 4.2: The Extension k/Fq

If σ is the generator of G(k/k) = Z/2Z, then σ  Fq (ζ) yields an automorphism and we know

G(k/k) ∼
= G(Fq (ζ)/D) ,→ G(Fq (ζ)/Fq ).
296 CHAPTER 4. FIELDS AND GALOIS THEORY

The extension Fq (ζ)/Fq is cyclic of degree 2s , where s is the order of the image of q in (Z/2r Z)∗ . As a
cyclic group has a unique subgroup of every possible order, there is a unique subfield of degree 2 in the
extension Fq (ζ)/Fq . If Fq < D, then D contains this unique extension. But, Fq (i) has degree 1 or 2 over Fq ,
so Fq (i) ⊆ D, which yields i ∈ D, and finally, i ∈ k (D = k ∩ Fq (ζ)). Thus, k = k, a contradiction. (Note: i
is a fourth root of unity; so, as k 6= k, we have char(k) = q 6= 2). Therefore, D = Fq .
Now,
Z/2Z = G(k/k) = G(Fq (ζ)/Fq ),
s
a group of order 2 . If we let r tend to ∞, then s tends to ∞, a contradiction. Therefore, char(k) = 0.
Finally, here the proof of Theorem 4.73:
Proof of Artin’s Irreducibility Criterion. Assume at first that we know the result for n a prime power–here
is how to prove the general case: Use induction on the number of primes dividing n. If n = pr m with
(p, m) = 1, we may assume p is odd. Now, X m − a is irreducible by our induction hypothesis; let α1 , . . . , αm
be its roots. Then,
Ym
Xm − a = (X − αj )
j=1

and
m
Y
r r
X n − a = (X p )m − a = (X p − αj ).
j=1

Suppose for some j that αj is a pth power in k(αj ). Now X m − a is irreducible so its Galois group acts
transitively on its roots. Therefore, each αi is σ(αj ) for some σ and so each of the αi is a pth power in k(αi ).
There exist βi ∈ k(αi ) with βip = αi for i = 1, . . . , m. We find that
Nk(αi )/k (αi ) = Nk(αi )/k (βi )p .
But,
m
Y
αj = (−1)m+1 a = Nk(αi )/k (αi ) = Nk(αi )/k (βi )p .
j=1

If m is odd, this gives a ∈ k ∗ p , contrary our assumptions. If m is even, then a = −(Nk(αi )/k (βi )p ). But, p
is odd so
a = (−Nk(αi )/k (βi ))p ,
again contrary to hypothesis. We conclude that none of the αi are pth powers in the field k(αi ). By the one
r
prime case, the polynomials X p − αi are irreducible for i = 1, . . . , m and all r.
r
Let ξ be a root of X n − a. Then, ξ satisfies X p − αj = 0 for at least one j. According to the
r
irreducibility of X p − αj , we find that [k(ξ) : k(αj )] = pr (of course k(αj ) ⊆ k(ξ)). However, [k(αj ) : k] = m
by the induction hypothesis and so [k(ξ) : k] = n. But this means the minimal polynomial for ξ has degree
n and ξ is a root of X n − a; so, X n − a is irreducible.
We’ve achieved a reduction to the heart of the matter, the one prime case. Here, n = pr and when
p = char(k) we already know the result. Therefore, we may and do assume p 6= char(k). Now use induction
on r. Say r = 1, adjoin the pth roots of 1 to k—call ζ a primitive pth root of 1:

k(α, ζ)
GG
vvv GG
GG
v
vvv GG
G
vv
k(α) k(ζ)
HH v
HH vv
HH vv
HH v
HH vv
vv
k
4.10. AN AMAZING THEOREM OF GALOIS THEORY 297

Here, α is a root of X p −a. Were [k(α) : k] 6= p, i.e., were X p −a reducible, we would have [k(α, ζ) : k(ζ)] < p.
Now over k(ζ), the Galois group of X p − a is cyclic of order p or trivial according as a ∈ / k(ζ)p or a ∈ k(ζ)p .
p
We then would have a ∈ k(ζ) ; hence α ∈ k(ζ). We know that [k(ζ) : k] = r ≤ p − 1 and so, (r, p) = 1. Write
1 = sr + tp, for some s, t. Now,
ar = Nk(ζ)/k (a) = Nk(ζ)/k (α)p .
But,
a = asr+tp = Nk(ζ)/k (α)ps apt = (Nk(ζ)/k (α)p · at )p ∈ k ∗ p ,
a contradiction. We conclude X p − a is irreducible.
r
p p
Qp Induction Step. pConsider X − a and assume p is odd . Write α for a root of X − a, andp
further write
j=1 (X − αj ) = X − a, with α = α1 . Now α is not a pth power in k(α). For, if it were β , then

a = (−1)p+1 a = Nk(ζ)/k (α) = Nk(ζ)/k (β)p (p is odd)

contrary to the hypothesis that a is not a pth power. Again by transitivity of the Galois group on the αj ,
no αj is a pth power in k(αj ), and therefore, by induction all the polynomials
r−1
Xp − αj , j = 1, 2, . . . , p
r r−1
are irreducible (over k(αj )). If ξ is a root of X p − a, then ξ is a root of X p − αj for some j, and as before
[k(ξ) : k(αj )] = pr−1 and [k(αj ) : k] = p;
r
so, [k(ξ) : k] = pr . We conclude X p − a is indeed irreducible.
r
Finally, we have the case p = 2. We know X 2 − a is irreducible, we must prove X 2 − a is irreducible.
As usual and with familiar notation, we have
2
Y
r r−1
X2 − a = (X 2 − αj ); α = α1 ; α2 = a.
j=1
r−1 r
So, if X 2 − αj is irreducible for j = 1, 2, the usual degree argument will show X 2 − a is irreducible. The
2r−1 2 4
only way X − αj will be reducible, by the induction hypothesis, is if αj ∈ k(αj )∗ or αj ∈ −4k(αj )∗ .
We will show each of these is untenable.
(1) Say α = α1 = β 2 with β ∈ k(α)∗ ; so, the same is true of α2 . Now,
−a = Nk(α)/k (α) = Nk(α)/k (β)2 = b2
r
/ k ∗ 2 , so (−1) ∈
yet a ∈ / k ∗ 2 . Hence, i ∈
/ k and we factor X 2 − a over k(i):
r r r−1 r−1
X 2 − a = X 2 + b2 = (X 2 + ib)(X 2 − ib).
2
If, on the right hand side, one of the factors is reducible, the induction hypothesis shows ib (or −ib) ∈ k(i)∗
4 4 4
or ib (or −ib) ∈ −4k(i)∗ . Since −4 is square in k(i), the cases ib ∈ −4k(i)∗ or −ib ∈ −4k(i)∗ reduce
∗2 ∗2
respectively to ib ∈ k(i) or -ib ∈ k(i) . But −1 is also a square, so these two cases are just the one case:
2
ib ∈ k(i)∗ .
Write ib = (γ + iδ)2 , with γ, δ ∈ k. Then γ 2 = δ 2 and b = 2γδ. However, γδ = ±γ 2 , and so b2 = 4γ 4 .
But then, a = −b2 ∈ −4k ∗ 4 , a contradiction. We are left with
(2) α = −4β 4 with β ∈ k(α)∗ . Again,
−a = Nk(α)/k (α) = Nk(α)/k (−4)Nk(α)/k (β)4 .
Since Nk(α)/k (−4) = (−4)2 = 16, we deduce −a is a square and we’ve assumed a is not a square. As above,
i∈/ k and we now repeat the argument of (1) to finish the proof.
298 CHAPTER 4. FIELDS AND GALOIS THEORY

4.11 Algebraic Closures; Steinitz’s Theory of Fields

In the twentieth century, E. Steinitz examined the theory of fields, especially transcendental extensions. He
had at his disposal the then new technique of transfinite induction which he used in the form of Zermelo’s
well-ordering principle. Of course, the latter is equivalent to Zorn’s Lemma or the Axiom of Choice. Here,
we’ll examine Steinitz’s results both in the purely algebraic case (existence of an algebraic closure) and in
the general case (transcendence bases).
Recall that in Remark (4) at the close of Section 4.2 we made the following definition (but informally):
Definition 4.15 A field, K, is algebraically closed (AC ) iff for every f ∈ K[X], there exists a θ ∈ K, so
that f (θ) = 0.

We also defined an algebraic closure of the field k as a field, K, which was itself AC and moreover was
algebraic over k. Here, we’ll prove the existence of an algebraic closure for each field, k, and its (essential)
uniqueness. First, for technical agility we’ll need equivalent forms of the condition (AC):
Proposition 4.75 For a field, K, the following conditions are equivalent:
(1) K has AC
(2) For every f ∈ K[X], all the roots of f (in any extension of K) are already in K
(3) Every polynomial f ∈ K[X] factors into linear factors in K[X]
(4) The only irreducible K-polynomials are the linear ones
(5) If L/K is algebraic, then L = K (so, K is algebraically closed in any of its overfields)
(6) If k is a subfield of K for which K/k is algebraic, then for any algebraic extension, L, of k, there exists
a k-monomorphism L −→ K.
(7) If k is a subfield of K for which K/k is algebraic and if e
k is a field isomorphic to k, via an isomorphism,
e e
ϕ, then for any algebraic extension, L or k, there exists a monomorphism Φ : L e → K extending ϕ.

The proofs of the equivalences (1)–(7) are trivial (DX); in (6) and (7) one makes use of the extension
lemma.

Remark: An algebraicaly Q closed field is always infinite. For, were it finite and θ1 , . . . , θn a listing of its
n
elements, then f (T ) = 1 + j=1 (T − θj ) would be a polynomial with no root in our field.
Now for the proof of the existence of algebraic closures, we need a very basic existence theorem.
Theorem 4.76 (Basic Existence Theorem) Suppose k is a field and Kλ (λ ∈ Λ) is a family of overfields of
k. Then, there exists a field extension K/k so that for every λ ∈ Λ we have a k-monomorphism ϕλ : Kλ → K.
That is, K contains a k-isomorphic copy of each field Kλ . Moreover, we may even choose K so that it is
generated by all the subfields ϕλ (Kλ ).
N
Proof . The proof is very simple using our techniques. We form the commutative ring A = λ∈Λ Kλ . Of
course, O 
A = lim Kλ ,
−→
S∈L(Λ) λ∈Λ

where L(Λ) is the family of finite subsets of Λ. The ring A is a k-algebra (the tensor products are taken over
k) and we embed k in A as usual via α 7→ α · (1 ⊗ 1 ⊗ · · · ⊗ 1 ⊗ · · · ). Choose any maximal ideal, M, of A
and write K = NA/M. Of course, K is a field extension of k and as each Kλ has a k-algebra homomorphism
to K (Kα −→ µ Kµ = A −→ A/M = K) taking 1 to 1, we see that each Kλ is embedded in K via this
homomorphism. Now the images in A of the Kλ generate A; so, their images in K generate K.
4.11. ALGEBRAIC CLOSURES; STEINITZ’S THEORY OF FIELDS 299

Theorem 4.77 (Steinitz) If k is a field, then k possesses an algebraic closure, Ω. If Ω and Ω e are two
algebraic closures of k, then there exists a (non-canonical) k-isomorphism Ω −
g e e is
→ Ω. The set Isomk (Ω, Ω)
in one-to-one correspondence with G(Ω/k).

Proof . We wish to use the Basic Existence Theorem, so the only problem is to find a good way of parametriz-
ing all finite extensions of k. Here, a better idea is to parametrize all the finitely generated extensions of
k. Take A = k[Xj ]∞ j=0 , the polynomial algebra on ℵ0 independent transcendentals over k. View, for each
n ≥ 0, the finitely generated polynomial rings k[X0 , . . . , Xn ] as a subring of A. In each ring k[X0 , . . . , Xn ]
we have the family of its maximal ideals, M. Write K(n, M) for the field k[X0 , . . . , Xn ]/M and consider
the collection of all these K(n, M).6 By the Basic Existence Theorem, there is field, L, over k containing
a k-isomorphic copy of each K(n, M). But each finite degree extension, M , of k is k-isomorphic to at least
one K(n, M); and so, each finite degree extension is “contained” in L. Write

Ω = Lalg = {ξ ∈ L | ξ is algebraic over k}

= algebraic closure of k in L.

By construction, Ω is algebraic over k; by choice of L, each finite degree extension, M , of k is k-isomorphic to

a K(n, M) so the latter is algebraic over k; hence, in Ω. And now, (the obvious modification of) Proposition
4.75 #(6) shows Ω is algebraically closed.
e is another algebraic closure of k. Now
Having proved existence, we now investigate uniqueness. Say Ω
e we have
for Ω and Ω

Ω= lim K (†)
−→
K/k finite,K⊆Ω

e=
Ω lim e
K. (††)
−→
K/k
e finite,K⊆
e Ωe

e
Since Ω is algebraically closed, for each such K/k e −→ Ω. We may assume each such
we get a k-injection K
e e
K is normal over k and choose a maximal chain of such K. Then, twisting if necessary by the G(K/k), e we
e e −→ Ω.
obtain a consistent family of k-injections of these K into Ω. By (††), there results the k-injection Ω
But the image of Ωe is algebraically closed and Ω is algebraic over it. We deduce from Proposition 4.75 (5)
that Ω = image of Ω.e
e the map ψ −1 ◦ ϕ is in G(Ω/k). Hence, the
Lastly, if ϕ and ψ are two k-isomorphisms from Ω to Ω,
e whenever ϕ is one such and σ runs over G(Ω/k).
k-isomorphisms ϕ ◦ σ run over all k-isomorphisms Ω −→ Ω

There remains the general case of a field extension k/k. The important concept here is the notion of
transcendence basis.

Definition 4.16 A subset, S, of a field extension, K/k, is a transcendence basis for K/k iff

(1) S is algebraically independent over k and

(2) K is algebraic over k(S).

We need some technique in handling algebraically independent elements. The most useful technical
observation is the following:
Q
6 For readers with a foundational mind, note: In the first place, the pairs (n, M) are elements of the set N P(A), where
P(A) is the power set of A; so, our indexing is done by a set. Next, each field, K(n, M), is itself in P(A); so, the whole collection
is perfectly valid from the point of view of set theory.
300 CHAPTER 4. FIELDS AND GALOIS THEORY

Proposition 4.78 Suppose that K/k is a field extension and A and B are subsets of K. Then the three
conditions below are mutually equivalent:
(1) A ∩ B = ∅ and A ∪ B is algebraically independent over k
(2) A is algebraically independent over k and B is algebraically independent over k(A)
(3) Same statement as (2) with A and B interchanged.
Proof . By symmetry, (2) ⇐⇒ (3); all that remains is to prove (1) ⇐⇒ (2).
(1) =⇒ (2). As A ⊆ A ∪ B and the latter is algebraically independent over k, we find that A is
algebraically independent over k. If B is algebraically dependent over k(A), there are elements b1 , . . . , bt
and a nonzero polynomial f (T1 , . . . , Tt ) ∈ k(A)[T1 , . . . , Tt ] with f (b1 , . . . , bt ) = 0. But, the coefficients may
be chosen from k[A] and involve only finitely many elements a1 , . . . , as from A. Then, f (T1 , . . . , Tt ) is
actually a nonzero polynomial of the form fe(a1 , . . . , as , T1 , . . . , Tt ), and fe is a polynomial over k in variables
U1 , . . . , Us , T1 , . . . , Tt . It is satisfied by {a1 , . . . , as , b1 , . . . , bt } ⊆ A ∪ B contradicting (1).
(2) =⇒ (1). No element, ξ, can be in A ∩ B, else the polynomial T − ξ ∈ k(A)[T ] is satisfied by ξ ∈ B
contradicting (2). We need only show each finite subset of A ∪ B is algebraically independent and, of course,
this is immediate if that finite subset is in A or B. So, we may assume that our subset is a1 , . . . , as , b1 , . . . , bt .
Any polynomial f (U1 , . . . , Us , T1 , . . . , Tt ) ∈ k[U1 , . . . , Us , T1 , . . . , Tt ] which vanishes on a1 , . . . , as , b1 , . . . , bt
gives a polynomial
f (a1 , . . . , as , T1 , . . . , Tt ) ∈ k(A)[T1 , . . . , Tt ]
which vanishes on b1 , . . . , bt . By (2), all the coefficients of f (a1 , . . . , as , T1 , . . . , Tt ) have to vanish. By (2),
again, these coefficients which are just different polynomials gj (U1 , . . . , Us ) (coeffs in k) must be zero as
polynomials. Therefore, our original polynomial f (U1 , . . . , Us , T1 , . . . , Tt ) is identically zero. This proves (1).

We derive many corollaries from Proposition 4.78.

Corollary 4.79 Let K/k be a field extension and A be a subset of K. Then, A is algebraically independent
over k iff for all ξ ∈ A, the element ξ is transcendental over k(A − {ξ}).
Proof . By taking A − {ξ} and {ξ} as the two subsets of Proposition 4.78, we see that (=⇒) is proved.
To prove (⇐=), take a finite subset of A, say a1 , . . . , at , and suppose it is algebraically dependent over
k. We may assume no smaller subset of a1 , . . . , at is dependent by passing to that smaller subset. Apply
Proposition 4.78 to the sets {a1 } and {a2 , . . . , at }. Since a1 , . . . , at is not independent, it follows that a1
is not independent over k(a2 , . . . , at ). Hence, a1 is not independent over the bigger field k(A − {a1 }). This
contradicts our hypothesis when ξ = a1 .
Corollary 4.80 If K/k be a field extension and A is an algebraically independent subset (over k) of K, and
if ξ ∈ K has the property that ξ is transcendental over k(A), then A ∪ {ξ} is again algebraically independent
over k.
Proof . This is immediate either from Proposition 4.78 or Corollary 4.79.
Corollary 4.81 Suppose K/k is a field extension and A is an algebraically independent subset of K. A
necessary and sufficient condition that A be a transcendence basis for K/k is that A be a maximal element
(under partial ordering by set inclusion) among the algebraically independent subsets of K.
Proof . If A is a transcendence basis for L/k yet is not maximal, there is an independent set, B, of K and
B > A. In Proposition 4.78, let B − a and A be the two sets, the K ⊆ k(A)(B − A) and K/K(A) is algebraic.
So, B − A is not algebraically independent over k(A) contradicting Proposition 4.78.
Conversely, if A is maximal among algebraically independent sets and ξ ∈ K but not in K(A), then ξ
cannot be transcendental over k(A) by Proposition 4.78 (set B = {ξ}, A = A). So, ξ is algebraic over K(A);
that is, K/k(A) is algebraic.
4.11. ALGEBRAIC CLOSURES; STEINITZ’S THEORY OF FIELDS 301

Theorem 4.82 (Steinitz) If K/k is a field extension and if S ⊆ T are two subsets of K so that
(a) K is algebraic over k(S) and
(b) T is algebraically independent over k, then there exists a transcendence basis, B, for K/k with T ⊆
B ⊆ S. In particular, every field extension possesses a transcendence basis.
Proof . We let S denote the collection of subsets of S which both contain T and are algebraically independent
over k. Of course, as T ∈ S, we have S = 6 ∅. Partially order S by set-theoretic inclusion and note that S is
inductive. Let B be a maximal element of S, it exists by Zorn’s Lemma. Consider the extension k(S)/k(B).
We know if each element of S is algebraic over k(S), then k(S) will be algebraic over k(B). But by the
maximality of B and Proposition 4.78 (or Corollary 4.80), we see that every element of S is indeed algebraic
over k(B). Thus, from the facts that K is algebraic over k(S) and k(S) is algebraic over k(B), we find K is
algebraic over k(B).
Upon taking S = K and T = ∅, we deduce each field extension has a transcendence basis.
Doubtless you will have noticed an analogy between the familiar theory of linear dependence and inde-
pendence for vector spaces and our theory of algebraic independence and indepnedence for field extensions.
For example, Proposition 4.78 can be translated into the linear case. Steinitz noticed this explicitly and
transformed the analogy into an axiomatic treatment of both cases simulataneously. In the linear case, the
notion of Span is a crucial ingredient and Steinitz generalized this by setting
Σ(A) = {ξ ∈ K | ξ is algebraic over k(A) (∗)
for A a subset of K and K/k a field extension. Of course we can then write: A is a transcendence basis
for K/k iff A is algebraically independent over k and K = Σ(A). The axioms for the Σ operation are the
dictated by the linear case:
(1) A ⊆ Σ(A),
(2) If A ⊆ B, then Σ(A) ⊆ Σ(B).
(3) Σ(Σ(A)) = Σ(A).
e of A so that ξ ∈ Σ(A).
(4) If ξ ∈ Σ(A), then there is a finite subset, A, e

(5) If η ∈ Σ(A ∪ {ξ}) but η ∈

/ Σ(A), then ξ ∈ Σ(A ∪ {η}).

Conditions (1)–(3) are obvious both in the linear case (when Σ(A) = Span(A)) and in the algebraic case
(when Σ(A) is as above). However, (4) and (5) deserve some comment. Property (4) makes the formation
of Σ(A) a property “of finite character”, and allows the application of Zorn’s Lemma in proofs of statements
about Σ(A) or independence. Property (5) is called the Steinitz Exchange Lemma—it is well-known in the
linear case. Here it is in the algebraic case:
Proposition 4.83 (Steinitz Exchange Lemma) For a field extension, K/k, a subset A ⊆ K and element
ξ, η of K we have
If η ∈ Σ(A ∪ {ξ}) but η ∈
/ Σ(A), then ξ ∈ Σ(A ∪ {η}).
Here, Σ(A) is as above in (∗).
Proof . In k(A) we can choose a transcendence basis (over k) contained in A by Theorem 4.83. As k(A) is
algebraic over k(B), it is algebraic over k(B ∪ {ξ}). Now, ξ is algebraic over k(B ∪ {ξ}); so, k(A ∪ {ξ})
is algebraic over k(B ∪ {ξ}) and therefore η ∈ k(B ∪ {ξ}). If the exchange lemma were valid when A was
independent, we would deduce ξ ∈ Σ(B ∪ {η}) ⊆ Σ(A ∪ {η}).
This achieves a reduction to the case where A is algbraically independent. The silly case ξ = η is a
tautology and so we have ξ 6= η and A ∪ {ξ, η} is algebraically dependent. But then, Proposition 4.78 applied
to the sets A ∪ {η}, {ξ}| shows that ξ ∈ Σ(A ∪ {η}), as desired.
 302 CHAPTER 4. FIELDS AND GALOIS THEORY

 Clearly, the Exchange Lemma is susceptible of generalizations. But one must be careful. “Obvious”
generalizations may be false. For example, the statement: If A, B, C are subsets of K (an extension of
k) and if C ⊆ Σ(A ∪ B) but C 6⊆ Σ(A), then B ⊆ Σ(A ∪ C) is false. Indeed, even the weaker statememnt
(because the hypotheses are stronger): If C ⊆ Σ(A∪B) but no element√of C is in Σ(A), then B ⊆ Σ(A∪C) is
√ ∅ and K = k(X, Y, X), where X and Y are algebraically
false. To see why the latter is false, just take A =
independent over k. Set B = {X, Y } and C = { X}.

Proposition 4.84 (General Steinitz Exchange Lemma) Suppose K/k is a field extension and A, B, C ⊆ K.
Assume that C ⊆ Σ(A ∪ B) but C 6⊆ Σ(A). Then, there exists a subset, B 0 , of B with properties

(1) B ⊆ Σ(A ∪ C ∪ B 0 ).

(2) B 6= B 0 .

(3) B 0 ∩ C = ∅.

Before proving this form of the exchange lemma, we should remark that:

(a) The hypotheses are those of the previous strong

√ (but false) statement—the conclusion is weaker: we
need B 0 . In the example where A = ∅, C = { X}, B = {X, Y }, it is clear that B 0 = {Y }.

(b) The name come from the fact that B 00 (= B − B 0 ) and C have been exchanged. That is, we conclude
B 00 ⊆ Σ(A ∪ B 0 ∪ C) from the hypotheses C ⊆ Σ(A ∪ B 0 ∪ B 00 ) and C 6⊆ Σ(A).

Proof . Here, the notation Σ refers to algebraic dependence over k. Let A e be a maximal algebraically
e e
independent subset of A, so that Σ(A) = Σ(A). Write C for a subset of C maximal with respect to algebraic
e Because C 6⊆ Σ(A),
independence over k(A). e we see that C
e 6= ∅ and that Ae∪ C
e is algebraically independent
over k by Proposition 4.78. Now C ⊆ Σ(A e ∪ C)
e and A ⊆ Σ(A) e ⊆ Σ(A e ∪ C).
e We find that

e ∪ C).
Σ(A ∪ B) = Σ(A e

Write T = A e∪ Ce ∪ B. Now, Σ(A ∪ B) = Σ(A e ∪ B) and by hypothesis we find that C ⊆ Σ(Ae ∪ B). Therefore,
e In it, we have T ⊇ A
Σ(T ) = Σ(A ∪ B); call this field K. e ∪ C,
e the former set generates and the latter is
algebraically independent. By the existence of transcendence bases (Theorem 4.83), there is a transcendence
e
basis for K/k, call it S, so that
T ⊇S⊇A e ∪ C.
e
e ∪ C)
We set B 0 = S − (A e ⊆ B. Of course, B 0 ∩ C
e = ∅. We know

e∪C
Σ(T ) = Σ(S) = Σ(A e ∪ B0)

and B ⊆ Σ(T ); so, conclusion (1) is proved. Were B 0 = B, we’d have S = A e∪Ce ∪ B. Now C e ⊆ C and
e
by hypothesis C ⊆ Σ(A ∪ B) = Σ(A ∪ B). As S is algebraically independent, we have a contradiction of
e ∪ C)
Proposition 4.78; this proves (2). Finally, if ξ ∈ B 0 ∩ C, then ξ ∈ C ⊆ Σ(A e implies that the subset of
e∪C
B0 ∪ A e = S consisting of {ξ} ∪ Ae∪C e is dependent; contradiction on how we chose S.

The main use of the standard exchange lemma is to prove that transcendence bases have the same
cardinality. Here’s how the finite case goes.

Theorem 4.85 Suppose K/k is a field and S is a finite subset of K while T is any subset of K. Assume
that #(T ) > #(S) and T ⊆ Σ(S). Then T is algebraically dependent. In particular, if K/k has a finite
transcendence basis, then all transcendence bases of K/k are finite with the same cardinality.
4.11. ALGEBRAIC CLOSURES; STEINITZ’S THEORY OF FIELDS 303

Proof . First, replace K by Σ(S), second replace S by a transcendence basis (for K = Σ(S)) which is a subset
of S. Therefore, we may assume S is a transcendence basis for K/k. If #(T ) = ∞, then then replace T by
any finite subset with #(T ) > #(S); so we may assume T is finite, too. Now suppose the result is false and
choose a counter-example pair S, T so that #(S)+#(T ) is minimal. Of course, in this case #(T ) = #(T )+1,
else we could reduce the sum by choosing a subset of T with #(T ) = #(S) + 1.
Our situation is now that
(a) S and T are finite algebraically independent sets
(b) T ⊆ Σ(S)
(c) #(S) = n, #(T ) = n + 1
(d) n is minimal among integers having (a), (b), (c).
Label the elements of S as s1 , . . . , sn but refrain from labelling T as yet. Consider S − {s1 }. There must
be some t ∈ T so that t ∈ / Σ(S − {s1 }), else T ⊆ Σ(S − {s1 }) and (d) would show T dependent contradicting
(a). Call this element t1 . Note that {s2 , . . . , sn , t1 } is an independent set at t1 ∈ / Σ(s2 , . . . , sn ). Since
t1 ∈ Σ((S − {s1 }) ∪ {s1 }), the standard exchange lemma (Proposition 4.83) shows that s1 ∈ Σ(s2 , . . . , sn , t1 ).
All the other elements of S lie in Σ(s2 , . . . , sn , t1 ) so T − {t1 } is certainly in Σ(s2 , . . . , sn , t1 ). However,
T − {t1 } cannot be contained in Σ(s3 , . . . , sn , t1 ). For if it were, the sets {s3 , . . . , sn , t1 }, T − {t1 } would
satisfy (a) and (b), their cardinalities would be n − 1 and n respectively and (d) would be contradicted.
Since T − {t1 } 6⊆ Σ(s3 , . . . , sn , t1 ), there is an element t2 ∈ T − {t1 } with t2 ∈ / Σ(s3 , . . . , sn , t1 ). This
means {s3 , s4 , . . . , sn , t1 , t2 } is an independent set and allows the exchange lemma to be applied once more
to η = t2 , ξ = s2 , and {s3 , . . . , sn , t1 }. We conclude that s2 ∈ Σ(s3 , . . . , sn , t1 , t2 ) and so all of T (thus
also T − {t1 , t2 }) is in Σ(s3 , . . . , sn , t1 , t2 ). It is clear how to continue the process and equally clear what
is happening: We are systematically replacing the elements s1 , s2 , . . . of S by elements t1 , t2 , . . . from T .
In the end, we find T − {t1 , . . . , tn } ⊆ Σ(t1 , . . . , tn ); but, #(T ) = n + 1, so tn+1 ∈ Σ(t1 , . . . , tn )–our final
contradiction (on (a)). As (a)–(d) are untenable, no counter-example exists.
To prove if K/k has finite transcendence basis, all transcendence bases have the same cardinality, we
choose a transcendence basis, S, of minimal (so, finite) cardinality and any other transcendence basis, T .
Of course, #(T ) ≥ #(S). If #(T ) > #(S), then T ⊆ Σ(S) immediately implies from the above that T is
dependent, which is not true. Thus, #(T ) ≤ #(S), and we are done.
If the reader will go through the argument, he will see we have used only Steinitz’s rules (1)—(5) on Σ.
Thus, the argument works in the linear case—though a direct argument is simpler. In carrying this out, one
sees that Corollary 4.79 gives a way of defining algebraic independence solely in terms of the Σ operation.
Namely, A is algebraically independent iff for every ξ ∈ A, the element ξ is not in Σ(A − {ξ}).
We can now handle the infinite case.

Theorem 4.86 For every field extension K/k, any two transcendence bases have the same cardinality.

Proof . Is S and T are given transcendence bases for K/k, then, by Theorem 4.85, the sets S and T are
simultaneously finite or infinite. Of course, the only case of concern is when S and T are infinite.
I claim two statements, which taken together will quickly finish the proof.

(I) For each ξ ∈ K, there exists a unique finite subset of S, call it S(ξ), characterized by
(a) ξ ∈ Σ(S(ξ)) and
(b) If Se ⊆ S and ξ ∈ Σ(S),
e then S(ξ) ⊆ S.
e

(II) For ξ and η in T , if ξ 6= η, then S(ξ) 6= S(η).

304 CHAPTER 4. FIELDS AND GALOIS THEORY

Suppose we assume (I) and (II) and write FP(S) for the collection of all finite subsets of S. Then the
map ξ 7→ S(ξ) is, by (I) and (II), a well defined injection of T to FP(S). We find that #(T ) ≤ #(FP(S))
and we know #(S) = #(FP(S)) because #(S) is infinite. Thus, #(T ) ≤ #(S); by symmetry, #(S) ≤ #(T ).
Then, the Cantor-Schröder-Bernstein Theorem yields #(S) = #(T ).
Both (I) and (II) are consequences of the exchange lemma. For (I), choose ξ ∈ K. If ξ is algebraic over
k, the set S(ξ) = ∅ satisfies (a) and (b); we may assume ξ is transcendental over k. There is a finite subset,
{s1 , . . . , sn }, of S so that ξ ∈ Σ(s1 , . . . , sn ). We may assume no smaller subset of {s1 , . . . , sn } has ξ in the
Σ formed from it. Suppose {σ1 , . . . , σq } is another subset of S and ξ ∈ Σ(σ1 , . . . , σq ). Choose any sj , apply
the exchange lemma to ξ, sj and S − {sj }. We find that sj ∈ Σ(s1 , . . . sbj , . . . , sn , ξ). Now, ξ ∈ Σ(σ1 , . . . , σq ),
therefore
sj ∈ Σ(s1 , . . . sbj , . . . , sn , σ1 , . . . , σq ).
The elements sj and σl are in the independent set S therefore sj must be one of the σ1 , . . . , σq . Since sj is
arbitrary in {s1 , . . . , sn } we get {s1 , . . . , sn } ⊆ {σ1 , . . . , σq } and so S(ξ) = {s1 , . . . , sn } has (a) and (b).
To prove (II), first note that if ξ ∈ S, then S = {ξ}. Pick ξ, η ∈ T and assume S(ξ) = S(η). Write
{s1 , . . . , sn } for the listing of the elements of S(ξ). A standard application of the exchange lemma shows
s1 ∈ Σ(s2 , . . . , sn , ξ). Therefore, S(ξ) ⊆ Σ(s2 , . . . , sn , ξ). It follows, as S(ξ) = S(η), that η ∈ Σ(s2 , . . . , sn , ξ).
By Claim (I) property (b), we find

{s1 , . . . , sn } = S(η) ⊆ {s2 , . . . , sn , ξ}.

Hence, ξ = s1 . By symmetry, η = s1 , too; and so, ξ = η (or ξ ∈ S and hence so is η; therefore {ξ} = S(ξ) =
S(η) = {η}). We are done.

Definition 4.17 The common cardinal number of all the transcendence bases for K/k is the transcendence
degree of K/k. It is denoted tr.d.k (K). The field K is purely transcendental over k iff K = k(S) where S is
a transcendence base for K/k.

To finish up this section, we have only to discuss the notion of separability for general field extensions
(i.e., not necessarily algebraic). For this, we essentially make Mac Lane I into a definition:

Definition 4.18 A field extension, K/k, is separable iff either char(k) = 0 or when char(k) = p > 0, then
the natural map
k 1/p ⊗k K −→ K 1/p
is injective.

There is a related (but stronger) concept, namely the notion of separable generation:

Definition 4.19 A field extension, K/k, is separably generated iff there exists a transcendence base, B,
for K/k so that K is separable (algebraic) over k(B). Such a transcendence bases is called a separating
transcendence base for K/k.

Separable non-algebraic field extensions exist:

Proposition 4.87 If K = k(B) and B is an algebraically independent set, then K/k is a separable extension.

Proof . By the argument of Section 4.3, the definition of separability is that when u1 , u2 , . . . are element of
K linearly independent over k, then up1 , . . . , upn , . . . are again linearly independent over k. If we apply this
to all the monomials formed from the elements of B, we see that we must prove: The elements up , where u
ranges over B, are algebraically independent over k. But, any non-trivial polynomial relation

f (upi1 , . . . , upit ) = 0
 4.11. ALGEBRAIC CLOSURES; STEINITZ’S THEORY OF FIELDS 305

is, a fortiori , a polynomial relation for ui1 , . . . , uit ; hence, cannot be non-trivial.
We can make two remarks that will be helpful for what follows:

Remarks:
(I) Separability is transitive. To see this, say L is separable over K and K is separable over k. Then, the
two maps

k 1/p ⊗k K −→ K 1/p ; K 1/p ⊗K L −→ L1/p (∗)

are injective. But then, we get the injection

(k 1/p ⊗k K) ⊗K L −→ K 1/p ⊗K L (∗∗)

(as L is flat over K). The left hand side of (∗∗) is k 1/p ⊗k L and the right hand side injects into L1/p
by (∗); so, we are done.
(II) Any field extension of a perfect field is separable. For, if k is perfect, then k = k p ; that is, k 1/p = k.
But then, k 1/p ⊗k K ∼
= K and K ⊆ K 1/p , as required.
(III) If K ⊇ L ⊇ k and K/k is separable, then L/k is separable. For consider the map

k 1/p ⊗k L −→ L1/p .

Let its kernel be A. By the flatness of K over L, we see that

0 −→ A ⊗L K −→ (k 1/p ⊗k L) ⊗L K = k 1/p ⊗k K −→ L1/p ⊗L K

is exact. Now, the composed map k 1/p ⊗k K −→ L1/p ⊗L K −→ K 1/p is injective by hypothesis; so,
A ⊗L K = (0). But, K is faithfully flat over L, therefore A = (0).

Corollary 4.88 If K/k is separably generated, then K/k is separable.

Proof . Write B for a separating transcendence base for K/k. Then, K is separable over k(B) and the latter
is separable over k by Proposition 4.87. Now Remark (I) applies.

 Separable generation is, in general, a strictly stronger concept than separability. Here is a standard
2
example: Let k be a perfect field (i.e., k = Fp ) and write k = k(T, T 1/p , T 1/p , . . .). Thus, K = lim Kn ,
−→
n
1/pn
where Kn = k(T ) and each Kn , being pure transcendental over k, is separable over k. Of course,
K 1/p = K and k 1/p = k by choice of k; so K/k is separable. We will now see it is not separably generated.
Let’s write STB for the phrase separating transcendence basis. We know tr.dk K = 1 as each Kn is algebraic
over K1 . Were an element z ∈ K an STB, we’d have z ∈ Kn for some n. Now, we may ignore K1 , . . . , Kn−1
and still have K = lim Km (m ≥ n), so we may assume z ∈ K1 . i.e., z ∈ k(T ). but then, the diagram of
−→
m
algebraic extensions
K

k(T ) = K1

k(Z)
and the fact that K is separable over k(Z) would show that K/K1 is a separable algebraic extension and
this is nonsense.
 306 CHAPTER 4. FIELDS AND GALOIS THEORY

 Remark . In the general case when K/k is separable and L is a subextension of the layer K/k it does
not follow that K/L is separable. For example, L = K1 in the above example shows that K/K1 is not
separable even though K/k is separable.

This remark indicates that separability is not a good notion in the general case; separable generation is a
much better notion. We are going to show now that the two concepts coalesce when the big field is a finitely
generated extension; so, the cause of most difficulties is infinite generation in the general case (as should be
clear from the counter-example above). Still, even in the finitely generated case, there are problems: K2 /k is
separably generated yet it is not separably generated over K1 (or separable–notations as above). The moral
is: be careful with separability (or separable generation) in the non-algebraic case, especially with infinitely
generated extensions.

Theorem 4.89 If K/k is a finitely generated field extension, then K/k is separable if and only if it is
separably generated.

Proof . One direction is Corollary 4.88; so, assume K/k is separable and finitely generated, say K =
k(T1 , . . . , Tn ). We let r = tr.dk K and use induction on n − r. If the latter is zero, T1 , . . . , Tr are already
an STB; so, assume n = r + 1 (this turns out to be the essential case). Now T1 , . . . , Tr+1 are algebraically
dependent and, by rearranging their order, we may assume T1 , . . . , Tr are a transcendence base. Then there is
a polynomial of smallest degree in Xn+1 coefficients in k[X1 , . . . , Xr ] having content 1, say f (X1 , . . . , Xr+1 ),
so that f (T1 , . . . , Tr , Tr+1 ) = 0. The degree of this polynomial in X r+1 must be positive and if its lead-
ing coefficient is a0 (X1 , . . . , Xr ), we can localize k[X1 , . . . , Xr ] with respect to a0 and make f monic in
k[X1 , . . . , Xr ]a0 [Xr+1 ]. The division algorithm for monic polynomials shows then that if g ∈ k[X1 , . . . , Xr+1 ]
vanishes on T1 , . . . , Tr+1 , we have

as0 g = f · g in k[X1 , . . . , Xr+1 ]

for some s ≥ 0. by unique factorization in k[X1 , . . . , Xr+1 ] it shows further that f is irreducible.
/ k[X1p , . . . , Xr+1
Suppose we could show that f (X1 , . . . , Xr+1 ) ∈ p
]. If so, at least one variable occurs in f
with exponent indivisible by p, call this variable Xi . Then Ti is dependent on T1 , . . . , Ti−1 , Ti+1 , . . . , Tr+1
and the latter must be algebraically independent by Theorem 4.82. Moreover, as the exponent of Ti is not
divisible by p, the element Ti is separable over k(T1 , . . . , Ti−1 , Ti+1 , . . . , Tr+1 ) and so, T1 , . . . , Ti−1 , Ti+1 , . . .,
Tr+1 form a separating transcendence basis, as required.
/ k[X1p , . . . , Xr+1
We use the separability of K/k to prove that f (X1 , . . . , Xr+1 ) ∈ p
]. Were the contrary
true, there would be a polynomial

f (X1 , . . . , Xr+1 ) = g(X1p , . . . , Xr+1

p
).

The monomials m1 , . . . , mt comprising g all have degree less than of f , so the elements m1 (T1 , . . . , Tr+1 ), . . .,
mt (T1 , . . . , Tr+1 ) are linearly independent over k. By separability, the elements

m1 (T1 , . . . , Tr+1 )p = m1 (T1p , . . . , Tr+1

p
), . . . , mt (T1 , . . . , Tr+1 )p = mt (T1p , . . . , Tr+1
p
)

are still linearly independent over k. Yet the relation

g(T1p , . . . , Tr+1
p
) = f (T1 , . . . , Tr+1 ) = 0

is a non-trivial linear relation among mp1 , . . . , mpt , a contradiction.

For use below, we record what we have just proved:
If K/k is separable, of transcendence degree r, then any set of r+1 elements of K, say T1 , . . . , Tr+1 , which
contains a transcendence basis for K/k, already contains a separating transcendence basis for k(T1 , . . . , Tr+1 )
over k. (All wee need note is that, by Remark III above, the field k(T1 , . . . , Tr+1 ) is separable over k.)
4.11. ALGEBRAIC CLOSURES; STEINITZ’S THEORY OF FIELDS 307

Now, let’s continue with our induction and finish the proof. We have n − r > 1 and we assume separable
generation for all separable field extensions, K/k, generated by less than n − r elements (tr.dk K = r).
By Remark III, k(T1 , . . . , Tn−1 ) is separably generated; so we can take an STB U1 , . . . , Ut for it. There
are only two possibilities for t: either t = r − 1 or t = r. Since K is then separable (algebraic) over
k(U1 , . . . , Ut , Tn ), we need only show the latter field is separably generated over k. But, the transcen-
dence degree of k(U1 , . . . , Ut , Tn ) over k is r and t + 1 ≤ r + 1 by the above. Again, Remark III shows
k(U1 , . . . , Ut , Tn ) is separable over k, and so our argument above (summarized in italics above), implies the
required separable generation of k(U1 , . . . , Ut , Tn ) over k.
We can augment the reasoning in the proof of Theorem 4.89 to obtain a useful theorem of Mac Lane:

Theorem 4.90 (Mac Lane) Suppose K/k is a finitely generated, separable field extension. Then, any set
of generators for K/k already contains a separating transcendence basis for K/k.

Proof . Write r = tr.d K and say K = k(T1 , . . . , Tn ). We use, as usual, induction on n − r, the case n − r = 0
is trivial and the case n − r = 1 is covered by the italicized statement in the middle of the proof of Theorem
4.89. For the induction step, use the notation of the last part of Theorem 4.89 and note that, by the induction
hypothesis, STB U1 , . . . , Ut may be chosen from among T1 , . . . , Tn−1 . Then the r + 1 = t + 1 generators
U1 , . . . , Ut , Tn for k(U1 , . . . , Ut , Tn ) are among T1 , . . . , Tn−1 , Tn and so the case n − r = 1 now applies and
finishes the proof.
An important corollary of our theorems is this result:

Corollary 4.91 (F.K. Schmidt) If k is a perfect field, every finitely generated field extension of k is sepa-
rably generated over k.

Proof . We apply Remark II and Theorem 4.89 (or 4.90) to our finitely generated extension of k.
308 CHAPTER 4. FIELDS AND GALOIS THEORY

4.12 Further Readings

Some basics of Galois theory is covered in most algebra texts (see Section 2.9). Emil Artin’s classic [1] is a
must. Other references include Kaplanski [31], Zariski and Samuel [50], Bourbaki (Algebra, Chapter IV) [6],
Lafon [33], Morandi [41], Escofier [14] and Van Der Waerden [47].
Chapter 5

Homological Algebra

5.1 Introduction
Homological Algebra has now reached into almost every corner of modern mathematics. It started with the
invasion of algebra into topology at the hands of Emmy Noether. She pointed out that the ranks and “torsion
coefficients” computed for various spaces were just the descriptions of finitely generated abelian groups as
coproducts of cyclic groups; so, one should instead study these “homology invariants” as homology groups.
Algebraic topology was born.
In the late 30’s through the decade of the 40’s, the invasion was reversed and topology invaded algebra.
Among the principal names here were Eilenberg, MacLane, Hochschild, Chevalley and Koszul. This created
“homological algebra” and the first deeply influential book was in fact called “Homological Algebra” and
authored by H. Cartan and S. Eilenberg (1956) [9].
Our study below is necessarily abbreviated, but it will allow the reader access to the major applications
as well as forming a good foundation for deeper study in more modern topics and applications.

5.2 Complexes, Resolutions, Derived Functors

From now on, let A denote an abelian category; think of Mod(R), where R is a ring, not necessarily
commutative. This is not so restrictive an example. The Freyd-Mitchell embedding theorem [15, 40], says
that each “reasonable” abelian category admits a full embedding into Mod(R) for a suitable ring R.
We make a new category, Kom(A), its objects are sequences of objects and morphisms from A:

d−n d−n+1 d−1 d0 dn

· · · −→ A−n −→ A−n+1 −→ · · · −→ A−1 −→ A0 −→ A1 −→ · · · −→ An −→ An+1 −→ · · · ,

in which di+1 ◦ di = 0, for all i. That is, its objects are complexes from A.
Such a complex is usually denoted by A• (sometimes, (A• , d• )). The morphisms of Kom(A) are more
ϕ•
complicated. However, we have the notion of “premorphism”: (A• , d• ) −→ (B • , δ • ). This is a sequence, ϕ• ,
of morphisms from A, where ϕn : An → B n , and we require that for all n, the diagram

An
dn / An+1

ϕn ϕn+1
 
Bn / B n+1
δn

309
310 CHAPTER 5. HOMOLOGICAL ALGEBRA

commutes. Such ϕ’s are called chain maps, or cochain maps. The collection of complexes and their chain
maps forms the category PreKom(A).

Remarks:

(1) Write An = A−n . This notation is usually used when A• stops at A0 (correspondingly, write dn for
d−n ).

(2) A complex is bounded below (resp. bounded above) iff there is some N ≥ 0 so that Ak = (0) if k < −N
(resp. Ak = (0) if k > N ). It is bounded iff it is bounded above and below. The sub (pre)category of
the bounded complexes is denoted PreKomb (A).

(3) If Ak = (0) for all k < 0, we have a cohomological complex (right complex or co-complex ).

(4) If Ak = (0) for all k > 0, then we use lower indices and get a homological complex (left complex , or
just complex ).

(5) The category A has a full embedding in PreKom(A) via A 7→ A• , where Ak = (0) if k 6= 0 and A0 = A
and all dk ≡ 0.

(6) Given a sequence, {An }∞

n=−∞ from A, we get an object of PreKom(A), namely:

0 0 0 0 0
· · · −→ A−n −→ A−n+1 −→ · · · −→ A−1 −→ A0 −→ A1 −→ A2 −→ · · · ,

where all maps are the zero map. Since Kom(A) and PreKom(A) will have the same objects, we will
drop references to PreKom(A) when objects only are discussed.

(7) Given (A• , d• ) in Ob(Kom(A)), we make a new object of Kom(A): H • (A• ), with

H n (A• ) = Ker dn /Im dn−1 ∈ Ob(A),

and with all maps equal to the zero map. The object H • (A• ) is the homology of (A• , d• ).

Nomenclature. A complex (A• , d• ) ∈ Kom(A) is acylic iff H • (A• ) ≡ (0). That is, the complex (A• , d• ) is
an exact sequence.
Given A ∈ Ob(A), a left (acyclic) resolution of A is a left complex, P• = {Pn }∞
n=0 , in Kom(A) and a
map P0 −→ A so that the new complex

· · · Pn −→ Pn−1 −→ · · · −→ P0 −→ A −→ 0

is acyclic. A right (acyclic) resolution of A ∈ Ob(A) is the dual of a left acyclic resolution of A considered
as an object of AD .
We shall assume of the category A that:

(I) A has enough projectives (or enough injectives, or enough of both). That is, given any A ∈ Ob(A)
there exists some projective object, P0 , (resp. injective object Q0 ) and a surjection P0 −→ A (resp.
an injection A −→ Q0 ).

Observe that (I) implies that each A ∈ Ob(A) has an acyclic resolution P• −→ A −→ 0, with all Pn
projective, or an acyclic resolution 0 −→ A −→ Q• , with all Qn injective. These are called projective
(resp. injective) resolutions. For Mod(R), both exist. (For Sh(X), the category of sheaves of abelian
groups on the topological space, X, injective resolutions exist.)

(II) A possesses finite coproducts (resp. finite products, or both). This holds for Mod(R) and Sh(X).
5.2. COMPLEXES, RESOLUTIONS, DERIVED FUNCTORS 311

Remark: The following simple fact about projectives will be used in several of the subsequent proofs: If
we have a diagram
P
θ
f
~ 
A / B / C
ϕ ψ

in which
(1) P is projective.
(2) The lower sequence is exact (i.e., Im ϕ = Ker ψ).
(3) ψ ◦ f = 0,
then there is a map θ : P → A lifting f (as shown by the dotted arrow above). Indeed, ψ ◦ f = 0 implies that
Im f ⊆ Ker ψ; so, we have Im f ⊆ Im ϕ, and we are reduced to the usual situation where ϕ is surjective.
Of course, the dual property holds for injectives.

Proposition 5.1 Suppose we are given an exact sequence

ψ ϕ
0 −→ A0 −→ A −→ A00 −→ 0

and both A0 and A00 possess projective resolutions P•0 −→ A0 −→ 0 and P•00 −→ A00 −→ 0. Then, there exists
a projective resolution of A, denote it P• , and maps of complexes P•0 −→ P• and P• −→ P•00 , so that the
diagram
0 / P•0 ψ•
/ P• ϕ•
/ P•00 / 0

  
0 / A0 ψ
/ A ϕ
/ A00 / 0

  
0 0 0
commutes and has exact rows and columns. A similar result holds for injective resolutions.

Proof . We have 0 −→ Pn0 −→ Pn −→ Pn00 −→ 0 if Pn exists and Pn00 is projective. So, the sequence would
split and Pn = Pn0 q Pn00 . Look at
i00
ψn n
0 −→ Pn0 −→ Pn0 q Pn00 ←− Pn00 −→ 0.
| {z } −→ ϕn
Pn

We have a map Pn −→ Pn via i00n ◦ ϕn ; we also have the map id − i00n ◦ ϕn and

ϕn ◦ (id − i00n ◦ ϕn ) = ϕn − ϕn ◦ i00n ◦ ϕn = ϕn − id00n ◦ ϕn = ϕn − ϕn ≡ 0.

It follows that id − i00n ◦ ϕn factors through ψn , i.e.,

ψn
id − i00n ◦ ϕn : Pn −→ Pn0 −→ Pn .

So, we may speak of “elements of Pn ” as pairs xn = (x0n , x00n ), where x00n = ϕn (xn ) and
(id − i00n ◦ ϕn )(xn ) = xn − i00n (x00n ) = x0n . Therefore,

xn = “x0n + i00n (x00n )” = (x0n , x00n ).

312 CHAPTER 5. HOMOLOGICAL ALGEBRA

This shows that for every n, we should define Pn as Pn0 q Pn00 . We need d• on P• . The map dn takes Pn to
Pn−1 . These dn should make the diagram

0 / Pn+1
0 / Pn+1 / Pn+1
00 / 0

d0n+1 dn+1 d00

n+1
  
0 / Pn0 ψn
/ Pn ϕn
/ Pn00 / 0

d0n dn d00
n
 ψn−1  ϕn−1 
0 / Pn−1
0 / Pn−1 / Pn−1
00 / 0

commute and dn ◦ dn+1 = 0. In terms of pairs, xn = (x0n , x00n ), where ψn (x0n ) = (x0n , 0) and ϕn (xn ) = x00n , the
commutativity of the lower left square requires

dn (x0n , 0) = (d0n x0n , 0).

How about (0, x00n )? Observe that we have ϕn−1 dn (0, x00n ) = d00n (x00n ). Write dn (0, x00n ) = (αn−1 , βn−1 ); we
know that ϕn−1 (αn−1 , βn−1 ) = βn−1 , thus,

dn (0, x00n ) = (αn−1 , d00n x00n ).

0
So, we need a map θn : Pn00 → Pn−1 ; namely θn (x00n ) = αn−1 , the first component of dn (0, x00n ). If we know
θn , then

dn (xn ) = dn (x0n , x00n ) = dn ((x0n , 0) + (0, x00n ))

= (d0n (x0n ), 0) + dn (0, x00n )
= (d0n (x0n ), 0) + (θn (x00n ), d00n (x00n ))
= (d0n (x0n ) + θn (x00n ), d00n (x00n )).

Everything would be OK in one layer from Pn to Pn−1 , but we need dn ◦ dn+1 = 0. Since

dn+1 (xn+1 ) = dn+1 (x0n+1 , x00n+1 ) = (d0n+1 (x0n+1 ) + θn+1 (x00n+1 ), d00n+1 (x00n+1 )),

we must have

dn ◦ dn+1 (xn+1 ) = (d0n ◦ d0n+1 (x0n+1 ) + d0n ◦ θn+1 (x00n+1 ) + θn ◦ d00n+1 (x00n+1 ), d00n ◦ d00n+1 (x00n+1 ))
= (d0n ◦ θn+1 (x00n+1 ) + θn ◦ d00n+1 (x00n+1 ), 0) = 0.

Therefore, we need
d0n ◦ θn+1 + θn ◦ d00n+1 = 0, for all n ≥ 1. (†n )
The case n = 0 requires commutativity in the diagram

0 / P10 / P1 / P100 / 0

d01 d1 d00
1
  
0 / P00 ψ0
/ P0 ϕ0
/ P000 /0

0  00
  
0 / A0 ψ
/ A ϕ
/ A00 / 0

  
0 0 0
5.2. COMPLEXES, RESOLUTIONS, DERIVED FUNCTORS 313

Since P000 is projective, there is a map σ : P000 → A so that

ϕ ◦ σ = 00 .
We can now define . We have (x0 ) = ((x00 , x000 )) = ((x00 , 0)) + ((0, x000 )) and ((x00 , 0)) = ψ0 (x00 ), as the
lower left square commutes. We also have
ϕ((0, x000 )) = 00 (ϕ0 (0, x000 )) = 00 (x000 ) = ϕσ(x000 ).
Consequently, ((0, x000 )) − σ(x000 ) is killed by ϕ and it follows that
((x00 , x000 )) = ψ0 (x00 ) + σ(x000 ).

We construct the map θn by induction on n and begin with n = 1. Note that

0 = d1 (x01 , x001 ) = (d01 (x01 ) + θ1 (x001 ), d001 (x001 )) = ψ0 (d01 (x01 ) + θ1 (x001 )) + σd001 (x001 ).
Therefore, we need to have
ψ0 θ1 + σd001 = 0. (††)

Construction of θ1 : In the diagram

P100
θ1
−σd00
1
~ 
P00 / A / A00 / 0
0
ψ

as P100 is projective, the map −σd001 lifts to a map θ1 : P100 → P00 ; thus (††) holds.
Next, we construct θ2 : Consider the diagram
P200
θ2
−θ1 d00
2
~ 
P10 / P00 / A0 / 0.
d01 0

If we know that 0 (−θ1 d002 ) = 0, we can lift our map and get θ2 , as shown. But, apply ψ, then by (††), we get
ψ0 θ1 d002 = σd001 d002 = 0.
Yet, ψ is an injection, so 0 θ1 d002 = 0. Thus, the map θ2 exists and we have d01 θ2 = −θ1 d002 , i.e. (†1 ) holds.
Finally, consider the case n > 1 and assume the θr are constructed for r ≤ n and (†k ) holds for all
k ≤ n − 1. By the induction hypothesis,
−d0n−1 θn d00n+1 = θn−1 d00n d00n+1 = 0.
We have the diagram
00
Pn+1
θn+1
−θn d00
n+1
} 
Pn0 / Pn−1
0 / Pn−2
0
d0n d0n−1

00
in which Pn+1 is projective, −d0n−1 θn d00n+1 = 0 and the lower sequence is exact. Therefore, −θn d00n+1 lifts to
θn+1 so that
d0n θn+1 = −θn d00n+1 ,
which is (†n ). The case of injectives follows from the dual category.
314 CHAPTER 5. HOMOLOGICAL ALGEBRA

Definition 5.1 Say

d−n d−n+1 d−1 d0 dn

· · · −→ X −n −→
X
X −n+1 −→
X
· · · −→ X −1 −→
X
X 0 −→
X
X 1 −→ · · · −→ X n −→
X
X n+1 −→ · · ·

and

d−n d−n+1 d−1 d0 dn

· · · −→ Y −n −→
Y
Y −n+1 −→
Y
· · · −→ Y −1 −→
Y
Y 0 −→
Y
Y 1 −→ · · · −→ Y n −→
Y
Y n+1 −→ · · ·

are objects of Kom(A). A homotopy between two maps f • , g • : X • −→ Y • is a sequence, {sn }, of maps
sn : X n → Y n−1 so that
f n − g n = sn+1 ◦ dnX + dYn−1 ◦ sn , for all n,

as illustrated in the diagram below:

dn−1 dX n

··· / X n−1 X
/ Xn / X n+1 / ···
x x xx
x n+1 xx
x
sn xxx
x ∆n
s
xx
∆n−1
xxx xxx ∆n+1
 |xx x
 |xx 
··· / Y n−1 / Yn / Y n+1 / ···
n−1 n
dY dY

where ∆n = f n − g n .

Remark: From f • and g • we get two maps on homology:

H • (f • ) : H • (X • ) −→ H • (Y • )
• • • •
H (g ) : H (X ) −→ H • (Y • ).

But, when f • and g • are homotopic, these maps on homology are equal. Indeed,

H • (f • − g • ) = H • (s•+1 d• ) + H • (d•−1 s• )
= H • (s•+1 )H • (d• ) + H • (d•−1 )H • (s• ).

As H • (d• ) = 0 and H • (d•−1 ) = 0, we get

H • (f • ) − H • (g • ) = H • (f • − g • ) = 0,

as claimed.

Now, based on this, we define the category Kom(A) by changing the morphisms in PreKom(A).

Definition 5.2 Kom(A) is the category whose objects are the chain complexes from A and whose morphisms
are the homotopy classes of chain maps of the complexes.

Theorem 5.2 Under the usual assumptions on A, suppose P • (A) −→ A −→ 0 is a projective resolution of
A and X • (A0 ) −→ A0 −→ 0 is an acyclic resolution of A0 . If ξ : A → A0 is a map in A, it lifts uniquely
to a morphism P • (A) −→ X • (A0 ) in Kom(A). [ If 0 −→ A −→ Q• (A) is an injective resolution of A and
0 −→ A0 −→ Y • (A0 ) is an acyclic resolution of A0 , then any map ξ : A0 → A lifts uniquely to a morphism
Y • (A0 ) −→ Q• (A) in Kom(A).]
5.2. COMPLEXES, RESOLUTIONS, DERIVED FUNCTORS 315

Proof . We begin by proving the existence of the lift, stepwise, by induction. Since we have morphisms
 : P0 (A) → A and ξ : A → A0 , we get a morphism ξ ◦  : P0 (A) → A0 and we have the diagram

P0 (A)
f0
ξ◦
z 
X0 (A0 ) / A0 / 0.

As P0 (A) is projective, the map f0 : P0 (A) → X0 (A0 ) exists and makes the diagram commute. Assume the
lift exists up to level n. We have the diagram

dP dP
Pn+1 (A)
n+1
/ Pn (A) n
/ Pn−1 (A) / ···

fn fn−1 (†)
 
Xn+1 (A0 ) / Xn (A0 ) / Xn−1 (A0 ) / ··· ,
dX
n+1 dX
n

0
so we get a map fn ◦ dP
n+1 : Pn+1 (A) → Xn (A ) and a diagram

Pn+1 (A)
fn+1
fn ◦dP
n+1
x 
Xn+1 (A0 ) / Xn (A0 ) / Xn−1 (A0 ).
dX
n

But, by commutativity in (†), we get

dX P P P
n ◦ fn ◦ dn+1 = fn−1 ◦ dn ◦ dn+1 = 0.

Now, Pn+1 (A) is projective and the lower row in the above diagram is exact, so there is a lifting
fn+1 : Pn+1 (A) → Xn+1 (A0 ), as required.
Now, we prove uniqueness (in Kom(A)). Say we have two lifts {fn } and {gn }. Construct the homotopy
{sn }, by induction on n.
For the base case, we have the diagram

P0 (A)
 / A / 0
s0
f0 g0 ξ
y   
X1 (A0 ) / X0 (A0 ) / A0 / 0.
dX
1
0

As 0 (f0 − g0 ) = (ξ − ξ) = 0, the lower row is exact and P0 (A) is projective, we get our lifting
s0 : P0 (A) → X1 (A0 ) with f0 − g0 = dX1 ◦ s0 .

Assume, for the induction step, that we already have s0 , . . . , sn−1 . Write ∆n = fn − gn , then we get the
diagram
dP
Pn (A) / Pn−1 (A)
n
/ Pn−2 (A) / ···
r
rrrr
rsn−1
∆n r ∆n−1 ∆n−2 (††)
 xrrr  
Xn+1 (A0 ) / Xn (A0 ) / Xn−1 (A0 ) / Xn−2 / ···
X
dn
316 CHAPTER 5. HOMOLOGICAL ALGEBRA

0
There results a map ∆n − sn−1 ◦ dP
n : Pn (A) −→ Xn (A ) and a diagram

Pn (A)
∆n −sn−1 ◦dP
n

Xn+1 (A0 ) / Xn (A0 ) / Xn−1 (A0 ).
dX
n+1 dX
n

0
As usual, if we show that dX P
n ◦ (∆n − sn−1 ◦ dn ) = 0, then there will be a lift sn : Pn (A) → Xn+1 (A ) making
X P
the diagram commute. Now, by the commutativity of (††), we have dn ◦ ∆n = ∆n−1 ◦ dn ; so

dX P P X P
n ◦ (∆n − sn−1 ◦ dn ) = ∆n−1 ◦ dn − dn ◦ sn−1 ◦ dn .

By the induction hypothesis, ∆n−1 = fn−1 − gn−1 = sn−2 ◦ dP X

n−1 + dn ◦ sn−1 , and therefore

∆n−1 ◦ dP X P P P X P X P
n − dn ◦ sn−1 ◦ dn = sn−2 ◦ dn−1 ◦ dn + dn ◦ sn−1 ◦ dn − dn ◦ sn−1 ◦ dn = 0.

Hence, sn exists and we are done. The case of injective resolutions follows by duality.

Corollary 5.3 Say ξ : A → A0 is a morphism in A and P, P 0 are respective projective resolutions of A and
A0 . Then, ξ extends uniquely to a morphism P −→ P 0 of Kom(A). (A similar result holds for injective
resolutions.)

Corollary 5.4 If P and P 0 are two projective resolutions of the same object, A, of A, then in Kom(A), P
is uniquely isomorphic to P 0 . (Similarly for injective resolutions.)

Proof . We have the identity morphism, id : A → A, so we get unique lifts, f and g in Kom(A), where
f : P → P 0 and g : P 0 → P (each lifting the identity). But then, f ◦ g and g ◦ f lift the identity to
endomorphisms of P 0 and P respectively. Yet, the identity on each is also a lift; by the theorem we must
have f ◦ g = id and g ◦ f = id in Kom(A).
Using the same methods and no new ideas, we can prove the following important proposition. The proof
will be omitted–it provides nothing new and has many messy details.

Proposition 5.5 Suppose we have a commutative diagram

0 / A0 / A / A00 / 0

f0 f f 00
  
0 / B0 / B / B 00 / 0.
0
(We call such a diagram a “small commutative diagram.”) Given objects, X • , X • , etc. of Kom(A) as
below, an exact sequence
0 00
0 −→ X • −→ X • −→ X • −→ 0
over the A-sequence and an exact sequence
0 00
•
0 −→ Y −→ Y • −→ Y •
−→ 0
00 00 0 0
over the B-sequence, assume X • and Y • are projective resolutions, while X • and Y • are acyclic resolu-
0 0 00 00
tions. Suppose further we have maps Φ0 : X • → Y • and Φ00 : X • → Y • over f 0 and f 00 . Then, there exists
a unique Φ : X • → Y • (over f ) in Kom(A) so that the “big diagram” of augmented complexes commutes
and X • and Y • are acyclic.
5.2. COMPLEXES, RESOLUTIONS, DERIVED FUNCTORS 317

Definition 5.3 If T is a functor (resp. cofunctor) on A to another abelian category B, the left derived
functors of T are the functors, Ln T , given by

(Ln T )(A) = Hn (T (P• (A))),

where P• (A) is any projective resolution of A (resp., when T is a cofunctor, the right derived functors of T
are the functors, Rn T , given by (Rn T )(A) = H n (T (P• (A)))).

If T is a functor, its right derived functors are the functors, Rn T , given by

(Rn T )(A) = H n (T (Q• (A))),

where Q• (A) is any injective resolution of A (when T is a cofunctor, the left derived functors of T , written
(Ln T )(A), are given by (Ln T )(A) = Hn (T (Q• (A)))).

The definition of derived functors is somewhat complicated and certainly unmotivated. Much of the
complication disappears when one observes that the values of either right or left derived functors are just
the homology objects of a complex; that, no matter whether T is a functor or a cofunctor, right (resp. left)
derived functors are the homology of a right (resp. left) complex (homology of a right complex is usually
called cohomology). Thus, for a functor, T , an injective resolution will yield a right complex and so is used to
compute right derived functors of T . Mutatis mutandis for projective resolutions; for cofunctors, T , simply
reverse all arrows. Of course, what we are investigating here is the effect of T on a resolution. We always
get a complex , but acyclicity is in general not preserved and the deviation from acyclicity is measured by the
derived functors.

As for motivation, the concept arose from experience first from algebraic topology later from homological
methods applied to pure algebra. Indeed the notion of derived functor took a long time to crystallize from
all the gathered examples and results of years of work. Consider, for example, a group G and the abelian
category of G-modules. On this category, we have already met the left exact functor M M G with values
0
in Ab. Our notation for this functor was H (G, M ). Now, in Chapters 1 and 4, we constructed a sequence
of functors of M , namely H n (G, M ). An obvious question is: Are the functors H n (G, −) the right derived
functors of H 0 (G, −)? We will answer this question below by characterizing the derived functors of a given
functor, T .

Further remarks:

(1) The definition makes sense, i.e., derived functors are independent of the resolution chosen. Use Corol-
lary 5.4 to see this.

(2) Suppose T is a functor and A is a projective object of A (resp. an injective object of A), then
(Ln T )(A) = (0) for n > 0 (resp. (Rn T )(A) = (0) for n > 0). If T is a cofunctor, interchange
conclusions. (A is its own resolution in either case; so, remark (1) provides the proof.)

(3) If T is exact, then Ln T and Rn T are (0) for n > 0 (the homology of an acyclic complex is zero).

Proposition 5.6 If T is any functor, there are always maps of functors T −→ R0 T and L0 T −→ T . If Q
is injective and P projective, then T (Q) −→ (R0 T )(Q) and (L0 T )(P ) −→ T (P ) are isomorphisms. When T
is a cofunctor interchange P and Q. For either a functor or a cofunctor, T , the zeroth derived functor R0 T
is always left-exact while L0 T is always right-exact. A necessary and sufficient condition that T be left-exact
(resp. right-exact) is that T −→ R0 T be an isomorphism of functors (resp. L0 T −→ T be an isomorphism
of functors). Finally, the functor map T −→ R0 T induces an isomorphism of functors Rn T −→ Rn R0 T for
all n ≥ 0 and similarly there is an isomorphism of functors Ln L0 T −→ Ln T .
318 CHAPTER 5. HOMOLOGICAL ALGEBRA

Proof . Most of this is quite trivial. The existence of the maps T −→ R0 T and L0 T −→ T follows immediately
from the definition (and the strong uniqueness of Corollary 5.4 as applied in Remark (1) above). That R0 T
is left exact is clear because it is a kernel and because the exact sequence of resolutions lifting a given exact
sequence can always be chosen as split exact at each level. Similarly, L0 T is right exact as a cokernel. Of
course, if T is isomorphic to R0 T it must be left exact, while if T is left exact, the terms in the augmented
complex outlined by the braces form an exact sequence:

T (d0 )
0 −→ T (A) −→ T Q0 (A) −→ T Q1 (A) −→ · · · .
| {z }

Thus, the canonical map T (A) −→ (R0 T )(A) = Ker T (d0 ) is an isomorphism. Similarly for right exactness
and L0 .
Should Q be injective, the sequence
id
0 −→ Q −→ Q −→ 0 −→ 0 −→ · · ·

is an injective resolution of Q and it shows that T (Q) is equal to (R0 T )(Q). Similarly for P and for cofunctors.
But now if A is arbitrary and Q• (A) is an injective resolution of A, the diagram

0 / T (A) / T (Q0 (A)) / T (Q1 (A)) / ···

  
0 / (R0 T )(A) / (R0 T )Q0 (A) / (R0 T )Q1 (A) / ···

in which the vertical arrows except the leftmost are isomorphisms shows immediately that Rn T −→ Rn (R0 T )
is an isomorphism for all n ≥ 0. Similarly for Ln (L0 T ) −→ Ln T .
The point of the above is that right derived functors belong with left exact functors and similarly if we
interchange left and right.
There are two extremely important examples of derived functors—they appear over and over in many
applications.

Definition 5.4 If A is any abelian category and B = Ab (abelian groups), write TB (A) = HomA (A, B), for
fixed B. (This is a left-exact cofunctor, so we want its right derived functors Rn TB ). Set

ExtnA (A, B) = (Rn TB )(A). (∗)

If A = Mod(Rop ) and B = Ab, set SB (A) = A ⊗R B, for fixed B. (This is a right-exact functor, so we
want its left-derived functors Ln SB ). Set

TorR
n (A, B) = (Ln SB )(A). (∗∗)

To be more explicit, in order to compute Ext•A (A, B), we take a projective resolution of A

P • −→ A −→ 0

apply HomA (−, B) and compute the cohomology of the (right) complex HomA (P • , B). For the tensor
product, we similarly take a projective resolution of the Rop module, A,

P • −→ A −→ 0
5.2. COMPLEXES, RESOLUTIONS, DERIVED FUNCTORS 319

apply − ⊗R B and compute the homology of the complex P • ⊗R B. Because HomA (−, B) is left exact and
− ⊗R B is right exact, we have
HomA (A, B) = Ext0A (A, B)
A ⊗R B = TorR
0 (A, B).

The following proposition, which we will call the basic lemma, will give us a chief property of derived
functors and help us characterize the sequence of derived functors of a given functor.
Proposition 5.7 (Long (co)homology sequence) Suppose X • , Y • , Z • are complexes and
0 −→ X • −→ Y • −→ Z • −→ 0
is exact in pre-Kom(A) (also OK in Kom(A)). Then, there exists a long exact sequence of homology (or
cohomology)

ED
BC
··· / H n−1 (Z • )

GF
@A
BC
ED
/ H n (X • ) / H n (Y • ) / H n (Z • )

GF
@A

BC
ED
/ H n+1 (X • ) / H n+1 (Y • ) / H n+1 (Z • )

GF
@A
/ H n+2 (X • ) / ···

(for all n). The maps δ n : H n (Z • ) → H n+1 (X • ) are called connecting homomorphisms.
Proof . Look at the diagram

0 / X n−1 / Y n−1 / Z n−1 / 0

  
0 / Xn / Yn / Zn / 0

dn
X dn
Y dn
Z
  
0 / X n+1 / Y n+1 / Z n+1 / 0

  
0 / X n+2 / Y n+2 / Z n+2 / 0
and apply the snake lemma to the rows n and n + 1. We get
δ
0 −→ Ker dnX −→ Ker dnY −→ Ker dnZ −→ Coker dnX −→ Coker dnY −→ Coker dnZ −→ 0.
If we look at Im dn−1
X and apply the map X • −→ Y • , we land in Im dn−1 Y , etc. Thus, at every level we get
that
H n (X • ) −→ H n (Y • ) −→ H n (Z • ) is exact.
n−1
Now, the connecting map, δ, of the snake lemma maps Ker dnZ to H n+1 (X • ). But, clearly, Im dZ
n−1 n−1
goes to zero under δ (because every element of Im dZ comes from some element in Y ). So, we get the
connecting homomorphism
δ n : H n (Z • ) −→ H n+1 (X • ).
A diagram chase proves exactness (DX).
320 CHAPTER 5. HOMOLOGICAL ALGEBRA

Corollary 5.8 Given a commutative diagram of complexes

0 / X• / Y• / Z• / 0

  
0 / X
e• / Ye • / Z
e• / 0

we have the big diagram of long exact sequences

··· / H n (X • ) / H n (Y • ) / H n (Z • ) / H n+1 (X • ) / ···

(∗∗)
   
··· / H n (X e •) / H n (Ye • ) / H n (Ze• ) / H n+1 (X
e •) / ···

which commutes.
Proof . Chase the diagram in the usual way.
Suppose T is a right-exact functor on A and
0 −→ A −→ B −→ C −→ 0
is an exact sequence in A. Resolve this exact sequence (as we have shown is possible, cf. Proposition 5.1) to
get
0 / P • (A) / P • (B) / P • (C) / 0

A B C
  
0 / A / B / C / 0

  
0 0 0
Then, as Ln T is the homology of the T P • complexes (still horizontaly exact on the complex level, as our
objects are projectives and the horizontal complex sequences split!), from the basic lemma, we get the long
exact sequence (of derived functors)

ED
BC
··· / Ln T (A) / Ln T (B) / Ln T (C)

@A
GF
ED
BC
/ Ln−1 T (A) / ··· ···

@A
GF
BC
ED
/ ··· ··· / L1 T (C)

@A
GF
/ T (A) / T (B) / T (C) / 0

Moreover, we have a commutative diagram corresponding to (∗∗):

··· / (Ln T )(A) / (Ln T )(B) / (Ln T )(C) / (Ln−1 T )(A) / ···

   
··· / (Ln T )(A)
e / (Ln T )(B)
e / (Ln T )(C)
e / (Ln−1 T )(A)
e / ···
5.2. COMPLEXES, RESOLUTIONS, DERIVED FUNCTORS 321

stemming from Proposition 5.5.

We can abstract this behavior of the sequences {Ln T }∞ n ∞
n=0 and {R T }n=0 according to the following
definition:

Definition 5.5 A δ-functor (resp. ∂-functor ) is a sequence, {T n }, (resp. a sequence, {Tn }) of functors so
that for every exact sequence
0 −→ A −→ B −→ C −→ 0
in A, we have a long exact sequence

BC
ED
0 / T 0 (A) / T 0 (B) / T 0 (C)

GF
@A
BC
ED
/ T 1 (A) / T 1 (B) / T 1 (C)

GF
@A
/ T 2 (A) / ···

and it is functorial in morphisms of exact sequences (similarly for ∂-functors, but reverse the arrows). [This
means that for every commutative diagram

0 / A / B / C / 0

  
0 / A0 / B0 / C0 / 0,

where the rows are exact, the induced diagram

0 / T 0 (A) / T 0 (B) / T 0 (C) / T 1 (A) · · · / T n−1 (C) / T n (A) / ···

     
0 / T 0 (A0 ) / T 0 (B 0 ) / T 0 (C 0 ) / T 1 (A0 ) · · · / T n−1 (C 0 ) / T n (A0 ) / ···

is also commutative.]

Of course, our results on derived functors give us

Proposition 5.9 The sequence of derived functors {Rn T } (resp. {Ln T }) for a left-exact (resp. right-exact)
functor, T , forms a δ-functor (resp. ∂-functor).

Another δ-functor is the sequence {H n (G, −)}∞

n=0 defined on the category of G-modules. To get at the
characterization of derived functors, we need

Definition 5.6 A δ-functor {T n } is universal iff for all δ-functors, {S n }, given a map (of functors),
f 0 : T 0 → S 0 , there exists a unique extension of f 0 to a map {f n : T n → S n } of δ-functors. Similarly for
∂-functors, but reverse the directions of the arrows.

Remark: Say {T n } and {S n } are universal δ-functors and f 0 : T 0 ∼

= S 0 is an isomorphism of functors.
Then, there is a unique isomorphism {f : T → S } of δ-functors lifting f 0 . That is, universal δ-functors
n n n

are determined by their components in dimension 0.

322 CHAPTER 5. HOMOLOGICAL ALGEBRA

Proof . Since f 0 : T 0 ∼ = S 0 is an isomorphism of functors, there is a map of functors, g 0 : S 0 → T 0 so that

f0 g0 = id and g0 f0 = id. Universality implies that there exist unique f n : T n → S n and g n : S n → T n lifting
f 0 and g 0 . But, f n g n and g n f n lift f 0 g 0 and g 0 f 0 , i.e., lift id. Yet, id lifts id in both cases. By uniqueness,
f n g n = id and g n f n = id.

Theorem 5.10 (Uniqueness I; Weak effaceability criterion) Say {T n } is a δ-functor on A and suppose for
every n > 0 there is some functor, En : A → A, which is exact and for which there is a monomorphism of
functors id −→ En [ i.e., for every object A in Ob(A) and all n > 0, we have an injection A −→ En (A)
functorially in A and En is exact ] so that the map T n (A) −→ T n (En (A)) is the zero map for every n > 0.
Then, {T n } is a universal δ-functor. Hence, {T n } is uniquely determined by T 0 .

Proof . Construct the liftings by induction on n. The case n = 0 is trivial since the map f 0 : T 0 → S 0 is
given. Assume the lifting exists for all r < n. We have the exact sequence

0 −→ A −→ En (A) −→ cokA −→ 0

and so, we have a piece of the long exact diagram

T n−1 (En (A)) / T n−1 (cokA ) δ / T n (A) 0 / T n (En (A))

fn−1 fn−1 (††)

 
S n−1 (En (A)) / S n−1 (cokA ) δ / S n (A)

where the left square commutes and the rows are exact. Hence, by a simple argument, there is a unique
fn : T n (A) → S n (A) that makes the diagram commute. This construction is functorial since En is an exact
functor; when we are done, all the diagrams commute.
Now, we need to prove uniqueness. Say we have two extensions {fn } and {gn } of f0 . We use induction
to prove that fn = gn for all n. This is obviously true for n = 0. Assume that uniqueness holds for all r < n.
Write (††) again:

T n−1 (En (A)) / T n−1 (cokA ) / T n (A) 0 / T n (En (A)) .

fn−1 gn−1 fn−1 gn−1 fn gn

     
S n−1 (En (A)) / S n−1 (cokA ) / S n (A)

As fn−1 = gn−1 on all arguments, the above diagram implies fn = gn on A. As A is arbitrary, fn = gn and
the proof is complete.

Corollary 5.11 Say En = E for all n (E functorial and exact) and E satisfies the hypotheses of Theorem
5.10. (For example, this happens when E(A) is {T n }-acyclic for all A (i.e., T n (E(A)) = (0) for all A and
all n > 0).) Then, {T n } is universal.

We can apply Corollary 5.11 to the sequence {H n (G, −)}, because E(A) = Map(G, A) satisfies all the
hypotheses of that Corollary according to Proposition 4.54. Hence, we obtain the important

Corollary 5.12 The sequence of functors {H n (G, −)} is a universal δ-functor from the category G-mod to
Ab.

Corollary 5.13 If En (A) is functorial and exact for every n > 0, and En (Q) is T n -acyclic for each n and
for every injective Q, then every injective object of A is {T n }-acyclic.
5.2. COMPLEXES, RESOLUTIONS, DERIVED FUNCTORS 323

Proof . Pick Q injective, then we have an exact sequence

0 −→ Q −→ En (Q) −→ cokQ −→ 0.

Since Q is injective, the sequence splits and so,

T n (En (Q)) = T n (Q) q T n (cokQ ).

By assumption, the left hand side is zero; thus, T n (Q) = (0).

Theorem 5.14 (Uniqueness II) Say {T n } and {S n } are δ-functors on A and {fn : T n → S n } is a map of
δ-functors. If for all injectives, Q, the map fn (Q) : T n (Q) → S n (Q) is an isomorphism (all n), then {fn } is
an isomorphism of δ-functors. The same statement holds for ∂-functors and projectives.

Proof . (Eilenberg) Of course, we use induction on n. First, we consider the case n = 0.

Step 1. I claim that f0 : T 0 (A) → S 0 (A) is a monomorphism for all A.
Since A has enough injectives, we have an exact sequence

0 −→ A −→ Q −→ cokA −→ 0,

for some injective, Q. We have the commutative diagram

0 / T 0 (A) / T 0 (Q)

f0 θQ,0
 
0 / S 0 (A) / S 0 (Q)

where θQ,0 : T 0 (Q) → S 0 (Q) is an isomorphism, by hypothesis. It follows that f0 is injective.

Step 2. The map f0 is an isomorphism, for all A.
We have the commutative diagram

0 / 0 / T 0 (A) / T 0 (Q) / T 0 (cokA )

θQ,0
  
0 / 0 / S 0 (A) / S 0 (Q) / S 0 (cokA ),

where the rightmost vertical arrow is injective by step 1 and θQ,0 is an isomorphism. By the five lemma, the
middle arrow is surjective, and thus bijective.
Next, consider the induction step.
Step 3. The map fn is injective for all A.
Consider the commutative diagram

T n−1 (Q) / T n−1 (cokA ) / T n (A) / T n (Q) / T n (cokA )

θQ,n−1 fn−1 fn θQ,n

    
S n−1 (Q) / S n−1 (cokA ) / S n (A) / S n (Q) / S n (cokA ).

By the induction hypothesis, fn−1 is injective; moreover, θQ,n−1 and θQ,n are bijective, by assumption, so
the five lemma implies that fn : T n (A) → S n (A) is injective.
324 CHAPTER 5. HOMOLOGICAL ALGEBRA

Step 4. The map fn is an isomorphism for all n.

By step 3, the righthand vertical arrow is an injection and by the induction hypothesis, fn−1 is an
isomorphism. As θQ,n and θQ,n−1 are isomorphisms, by the five lemma, again, fn is surjective and thus
bijective.

Theorem 5.15 (Uniqueness III) Given a δ-functor {T n } on A, suppose that for any A ∈ A, any injective
Q and any exact sequence

0 −→ A −→ Q −→ cokA −→ 0,

the sequence

T n−1 (Q) −→ T n−1 (cokA ) −→ T n (A) −→ 0 is exact, if n > 0.

Under these conditions, {T n } is a universal δ-functor. (Similarly for ∂-functors and projectives).

Proof . We proceed by induction. Given another δ-functor, {S n }, and a morphism of functors f0 : T 0 → S 0 ,

suppose f0 is already extended to a morphism fr : T r → S r , for all r ≤ n − 1. Since A has enough injectives,
we have the exact sequence

0 −→ A −→ Q −→ cokA −→ 0

and we get the diagram

T n−1 (Q) / T n−1 (cokA ) / T n (A) / 0

fn−1 fn−1 ϕQ
  
S n−1 (Q) / S n−1 (cokA ) / S n (A) .

By a familiar argument, there exists only one map, ϕQ , making the diagram commute. Note that ϕQ might
depend on Q. To handle dependence on Q and functoriality, take some A e and its own exact sequence

e −→ Q
0 −→ A e −→ cok e −→ 0
A

e Since Q
and say we have a map g : A → A. e is injective, there exist θ and θ making the following diagram
commute:

0 / A / Q / cokA / 0

g θ θ
  
0 / A
e / Q
e / cok e / 0.
A

We have the diagram:

5.2. COMPLEXES, RESOLUTIONS, DERIVED FUNCTORS 325

T n−1 (Q) T n−1 (cokA ) T n (A) 0

T n (g)

!
T n−1 (Q) T n−1 (cokA!) !
T n (A) 0
fn−1 fn−1 ϕQ

S n−1 (Q) S n−1 (cokA ) S n (A)

fn−1 fn−1 ϕQ!
S n (g)
!
S n−1 (Q) S n−1 (cokA!) !
S n (A)

All squares at top and bottom commute and the two left hand vertical squares also commute by the
induction hypothesis. It follows that the righthand vertical square commutes (DX), i.e.:

ϕQe ◦ T n (g) = S n (g) ◦ ϕQ .

e we see that
If we set g = id (perhaps for different Q and Q),

ϕQe = ϕQ ,

so ϕ is independent of Q. Moreover, for any g, the righthand vertical diagram gives functoriality.

It remains to show commutativity with the connecting homomorphisms. Given an exact sequence

0 −→ A0 −→ A −→ A00 −→ 0

begin the resolution of A0 by injectives, i.e., consider an exact sequence

0 −→ A0 −→ Q0 −→ cok0 −→ 0.

We obtain the diagram below in which θ and θ exist making the diagram commute:

0 / A0 / A / A00 / 0

θ θ
 
0 / A 0 / Q0 / cok0 / 0.

Consequently, we get the diagram below in which all top and bottom diagrams commute and the left
vertical cube commutes:
326 CHAPTER 5. HOMOLOGICAL ALGEBRA

δT
T n−1 (A) T n−1 (A"" ) T n (A" )

T n−1 (Q" ) T n−1 (cok" ) T n (A" )

fn−1 fn−1 fn

δS
S n−1 (A) S n−1 (A"" ) S n (A" )
fn−1 fn−1 fn

S n−1 (Q" ) S n−1 (cok" ) S n (A" )

If we use the rightmost horizontal equalities, a diagram chase shows

T n−1 (A00 )
δT
/ T n (A0 )

fn−1 fn
 
S n−1 (A00 )
δS
/ S n (A0 )

commutes (DX).

Corollary 5.16 The right derived (resp. left derived) functors of T are universal δ-functors (resp. universal
∂-functors). A necessary and sufficient condition that the δ-functor {T n } be isomorphic to the δ-functor
{Rn T 0 } is that {T n } be universal. Similarly for ∂-functors and the sequence {Ln T0 }.

Corollary 5.17 For any group, G, the δ-functor {H n (G, −)} is isomorphic to the δ-functor {Rn H 0 (G, −)}.
5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 327

5.3 Various (Co)homological Functors

There are many homological and cohomological functors all over mathematics. Here, we’ll give a sample from
various areas and some simple applications. By these samples, some idea of the ubiquity of (co)homological
functors may be gleaned.

First of all, the functors Ext•A (A, B) and TorR

• (A, B) have been defined in an asymmetric manner: We
resolved A, not B. We’ll investigate now what happens if we resolve B.

Pick any B ∈ A and write

TB (−) = HomA (−, B).

[Remember, (Rn TB )(A) = ExtnA (A, B).]

If 0 −→ B 0 −→ B −→ B 00 −→ 0 is exact and P is projective, we get the exact sequence

0 −→ TB 0 (P ) −→ TB (P ) −→ TB 00 (P ) −→ 0.

[Recall, P is projective iff HomA (P, −) is exact.]


Resolve A: P• −→ A −→ 0. We get the commutative diagram

O O O

0 / TB 0 (Pn ) / TB (Pn ) / TB 00 (Pn ) / 0

O O O

.. .. ..
.O .O .O

0 / TB 0 (P0 ) / TB (P0 ) / TB 00 (P0 ) / 0

O O O

0 / TB 0 (A) / TB (A) / TB 00 (A)

O O O

0 0 0

Applying cohomology,1 we get the long exact sequence of (co)homology:

ED
BC
··· / Rn−1 TB 00 (A)

GF
@A
BC
ED
/ Rn TB 0 (A) / Rn TB (A) / Rn TB 00 (A)

GF
@A
/ Rn+1 TB 0 (A) / ···

1 The locution “apply (co)homology” always means make the long exact sequence arising from the given short one.
328 CHAPTER 5. HOMOLOGICAL ALGEBRA

Therefore, we have the exact sequence

BC
ED
··· / Extn−1 (A, B 00 )
A

GF
@A
ED
BC
/ ExtnA (A, B 0 ) / ExtnA (A, B) / ExtnA (A, B 00 )

GF
@A
/ Extn+1 (A, B 0 ) / ···
A

Consequently, we find that:

(1) Ext•A (A, B) is a functor of A and B, actually a co-functor of A and a functor of B.
(2) {Ext•A (A, −)}∞
n=0 is a δ-functor (functorial in A).

(3) {Ext•A (−, B)}∞

n=0 is a universal δ-functor (functorial in B).

•
g (A, B) for what we get by resolving the righthand variable B (using injective resolutions).
Now, write Ext A
We obtain analogs of (1), (2), (3); call them (e
1), (e
2) and (e
3). Note that
0
g A (A, B) = HomA (A, B) = Ext0A (A, B).
Ext
•
g A (A, −) is a universal δ-functor and Ext•A (A, −) is a δ-functor. Thus, there is a unique extension
Now, Ext
n ϕn
g A (A, B) −→
Ext ExtnA (A, B),
which is a map of δ-functors. When B is injective, the left hand side is (0) (as derived functors vanish on
injectives). Moreover, in this case, HomA (−, B) is exact, and so,
Rn HomA (A, B) = (0), for all n > 0 and all A.
By Uniqueness II (Theorem 5.14), we conclude
Theorem 5.18 The derived functor Ext•A can be computed by resolving either variable. The same result
holds for TorR
• (in Mod(R)).

There is a technique by which the value of Rn T (A) can be computed from Rn−1 T (A)e for a suitable A.e
2
This is known as décalage or dimension shifting. Here is how it goes for a left exact functor, T , or left
exact cofunctor, S.
For T , consider A and embed it in an acyclic object for Rn T , e.g., an injective
0 −→ A −→ Q −→ cokA −→ 0.
Now apply cohomology:

BC
ED
0 / T (A) / T (Q) / T (cokA ) / R1 T (A) /0 / R1 T (cokA )

GF
@A
BC
ED
/ R2 T (A) /0 / R2 T (cokA ) / ··· /0 / Rn−1 T (cokA )

GF
@A
/ Rn T (A) /0 / ···
2 The French word means a shift in space and is also used for time.
5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 329

We find that

Rn−1 T (cokA ) −
g→ Rn T (A), n≥2
1
cok(T (Q) −→ T (cokA )) −
g→ R T (A),

e is just cokA .
so the suitable A
For the cofunctor, S, project an acyclic object (for Rn S), e.g., a projective, onto A:

0 −→ Ker A −→ P −→ A −→ 0.

Just as above, we find

Rn−1 S(Ker A ) −
g→ Rn S(A)
cok(S(P ) −→ S(Ker A )) −
g→ R1 S(A).

Similar statements hold for right exact functors or cofunctors and their left derived functors.
There is a very important interpretation of Ext1A (A, B); indeed this interpretation is the origin of the
word “Ext” for the derived functor of Hom. To keep notation similar to that used earlier for modules in
Chapter 2, we’ll replace A by M 00 and B by M 0 and consider Ext1A (M 00 , M 0 ).
Say
0 −→ M 0 −→ M −→ M 00 −→ 0 (E)
is an extension of M 00 by M 0 . Equivalence is defined as usual: In the diagram below, the middle arrow, g,
is an isomorphism that makes the diagram commute:

0 / M0 / M / M 00 / 0

g

0 / M0 / M
f / M 00 / 0

Apply to (E) the functor HomA (M 00 , −). We get

δ(E)
0 −→ HomA (M 00 , M 0 ) −→ HomA (M 00 , M ) −→ HomA (M 00 , M 00 ) −→ Ext1A (M 00 , M 0 ).

So, δ(E) (id) is a canonical element in Ext1A (M 00 , M 0 ); it is called the characteristic class of the extension (E),
denoted χ(E). Note: χ(E) = 0 iff (E) splits.
Now, given ξ ∈ Ext1A (M 00 , M 0 ), resolve M 0 by injectives:

0 −→ M 0 −→ Q0 −→ Q1 −→ Q2 −→ · · · .

If we apply HomA (M 00 , −), we get

d d
0 −→ HomA (M 00 , M 0 ) −→ HomA (M 00 , Q0 ) −→
0
HomA (M 00 , Q1 ) −→
1
HomA (M 00 , Q2 ) −→ · · · ,

and we have Ext1 (M 00 , M 0 ) = Ker d1 /Im d0 . Consequently, ξ comes from some f ∈ HomA (M 00 , Q1 ) and
d1 (f ) = 0.
0 / M0 / Q0 d0
/ Q1 d1
/ Q2
O z<
z
zz
f
zzzd1 (f )=0
z
M 00
330 CHAPTER 5. HOMOLOGICAL ALGEBRA

Thus,
Im f ⊆ Ker d1 = Im d0 = X,
and so, f is a map M 00 −→ X ⊆ Q1 . We get

0 / M0 / Q0 d0
/ X / 0
O
f (∗)

M 00

Taking the pullback of (∗) by f , we find

0 / M0 / M / M 00 / 0 (E)
g f
 
0 / M0 / Q0 / X / 0,

i.e., we get an extension, (E). One checks that (E) is independent of f , but depends only on ξ. This involves
two steps (DX): (E) does not change if f is replaced by f + d0 (h); (E) does not change if we use another
injective resolution. Hence, we’ve proved

Theorem 5.19 There is a one-to-one correspondence

(E) 7→ χ(E)

between equivalence classes of extensions of modules of M 00 by M 0 and elements of Ext1A (M 00 , M 0 ).

An interpretation of ExtnA (M 00 , M 0 ) for n ≥ 2 will be left for the exercises. The cohomological functor
Ext•A (A, B) is the most important of the various cohomological functors because many cohomological functors
are special cases of it. The same holds for TorR • (A, B) with respect to homological functors. Here are several
examples of these considerations:
We begin with groups. Recall that we proved the δ-functor {H n (G, A)} coincided with the right-derived
functors of the functor A AG . (Of course, here G is a group and A is a G-module.) We form the group
ring R = Z[G] ; every G-module is an R-module and conversely—in particular, every abelian group is an
3

R-module with trivial action by G. Consider Z as R-module with trivial G-action and for any G-module
introduce the functor
A HomR (Z, A).
It is left exact and its derived functors are Ext•R (Z, A). But, a homomorphism f ∈ HomR (Z, A) is just an
element of A, namely f (1). And, as Z has trivial G-action, our element, f (1), is fixed by G. Therefore

HomR (Z, A) −
g→ AG ,

and so we find

Proposition 5.20 If G is any group and A is any G-module, there is a canonical isomorphism

ExtnZ[G] (Z, A) −
g→ H n (G, A), all n ≥ 0.
3 Recall that Z[G] is the free Z-module on the elements of G. Multiplication is defined by σ ⊗ τ 7→ στ , where σ, τ ∈ G and

we extend by linearity.
5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 331

As for group homology, first consider the exact sequence of G-modules


0 −→ I −→ Z[G] −→ Z −→ 0, (∗)
P P
in which  takes the element aσ · σ to aσ ; that is,  sends each group element to 1. The ideal I is by
definition Ker ; one sees easily that I is freely generated by the elements σ − 1 (for σ ∈ G, σ 6= 1) as a
Z-module. A little less obvious is the following:

Proposition 5.21 The mapping log(σ) = (σ − 1)(mod I 2 ) is an isomorphism of abelian groups

log : G/[G, G] −
g→ I/I 2 .

Proof . The operations on the two sides of the claimed isomorphism, log, are the group multiplication abelian-
ized and addition respectively. Clearly, log(σ) = (σ − 1)(mod I 2 ) is well-defined and

(στ − 1) = (σ − 1) + (τ − 1) + (σ − 1)(τ − 1)

shows it’s a homomorphism. Of course we then have log(σ −1 ) = −(σ − 1), but this is easy to see directly. It
follows immediately that [G, G] lies in the kernel of log; so we do get a map

log : G/[G, G] −→ I/I 2 .

As I is the free Z-module on the elements (σ − 1), as σ ranges over G (σ 6= 1), we can define

exp : I −→ G/[G, G],

via  
X Y
exp  nσ (σ − 1) = σ nσ mod [G, G]
σ6=1 σ6=1

and considerations entirely simililar to those above for log show that exp is a homomorphism from I to
G/[G, G] and that I 2 is killed by exp. It should be obvious that log and exp are mutually inverse, so we’re
done.
If A is a G-module, we can tensor exact sequence (∗) over Z[G] with A; this gives

I ⊗Z[G] A −→ A −→ Z ⊗Z[G] A −→ 0.

Of course, this shows

A/(IA) −
g→ Z ⊗Z[G] A.
The functor A A/IA is a right-exact functor from G-modules to Ab and its left derived functors, Hn (G, A),
are the homology groups of G with coefficients in A. The isomorphism we’ve just observed (together with
the usual arguments on universal ∂-functors) allows us to conclude

Proposition 5.22 If G is a group and A is any G-module, there is a canonical isomorphism (of ∂-functors)

Hn (G, A) −
g→ TornZ[G] (Z, A), all n ≥ 0.

We first introduced and computed group cohomology via an explicit chain complex, is there a similar
description for group homology? There is indeed, and while we can be quite direct and give it, perhaps it is
better to make a slight detour which is necessary anyway if one is to define (co)homology of algebras in a
direct manner.
 332 CHAPTER 5. HOMOLOGICAL ALGEBRA

Write K for a commutative ring and R for a (possibly non-commutative) K-algebra. In the case of
groups, K will be Z and R will be Z[G], while for other purposes K will be a field and R the polynomial
ring K[T1 , . . . , Tn ]; there will be still other purposes.
For an integer, n ≥ −1, write Cn (R) for the (n + 2)-fold tensor product of R with itself over K:

Cn (R) = R ⊗K R ⊗K · · · ⊗K R
| {z }
n+2
C−1 (R) = R
Cn+1 (R) = R ⊗K Cn (R).

Next, introduce the module R⊗K Rop . We want to make R⊗K Rop into a K-algebra by the multiplication

(ρ ⊗ σ op )(r ⊗ sop ) = ρr ⊗ σ op sop = ρr ⊗ (sσ)op

and for this we must have K in the center of R. To see this, pick λ ∈ K, set ρ = s = 1, set σ = λ, and
compute

rλ ⊗K 1op = (r ⊗K λop 1op ) = r ⊗K λop

= (1 ⊗K λop )(r ⊗K 1op )
= (λ ⊗K 1op )(r ⊗K 1op ) = λr ⊗K 1op .

As r is arbitrary, we are done. So, from now on, we shall assume K is in the center of R. The
algebra R ⊗K Rop is called the enveloping algebra of R over K; it is usually denoted Re . Now, there is a
map Re −→ R via
r ⊗ sop 7→ rs.
 This map is not a map of K-algebras, only a map of Re -modules. (Re acts on R via (r ⊗sop )(m) = rms;
in general, two-sided R-modules are just Re -modules (as well as (Re )op -modules).) It will be a K-algebra
map if R is commutative.

We should also note that the map

r ⊗ sop 7→ sop ⊗ r
is a K-isomorphism of K-algebras Re −
g→ (Re )op . (DX)
It will be best to use “homogeneous notation” for elements of Cn (R): r0 ⊗ r1 ⊗ · · · ⊗ rn+1 . Then Cn (R)
is a left Re -module under the rule

(s ⊗ top )(r0 ⊗ r1 ⊗ · · · ⊗ rn ⊗ rn+1 ) = (sr0 ) ⊗ r1 ⊗ · · · ⊗ rn ⊗ (rn+1 t).

Now we’ll make {Cn (R)}∞

n=0 into an acyclic left complex. The boundary map is

n
X
∂n (r0 ⊗ r1 ⊗ · · · ⊗ rn+1 ) = (−1)i r0 ⊗ · · · ⊗ (ri ri+1 ) ⊗ · · · ⊗ rn+1 ,
i=0

it is an Re -homomorphism Cn (R) −→ Cn−1 (R). In particular, ∂0 is the Re -module map discussed above,

∂0 (r0 ⊗ r1 ) = r0 r1

and
∂1 (r0 ⊗ r1 ⊗ r2 ) = (r0 r1 ) ⊗ r2 − r0 ⊗ (r1 r2 ).
5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 333

From these, we see ∂0 ∂1 = 0 precisely because R is associative. To prove {Cn (R)} is a complex and acyclic,
introduce the map
σn : Cn (R) −→ Cn+1 (R) via σn (ξ) = 1 ⊗K ξ.
The map σn is only an Rop -module map but it is injective because there is a map τn : Cn+1 (R) → Cn (R) given
by τn (r0 ⊗ (rest)) = r0 (rest) and we have τn σn = id. Moreover, Im (σn ) generates Cn+1 (R) as R-module! It
is easy to check the relation

∂n+1 σn + σn−1 ∂n = id on Cn (R), for n ≥ 0. (†)

Now use induction to show ∂n−1 ∂n = 0 as follows: Above we showed it for n = 1, assume it up to n and
apply ∂n (on the left) to (†), we get

∂n ∂n+1 σn + ∂n σn−1 ∂n = ∂n .

However, ∂n σn−1 = idn−1 − σn−2 ∂n−1 , by (†) at n − 1. So,

∂n ∂n+1 σn + ∂n − σn−2 ∂n−1 ∂n = ∂n ,

that is, ∂n ∂n+1 σn = 0 (because ∂n−1 ∂n = 0). But, the image of σn generates Cn+1 as R-module; so
∂n ∂n+1 = 0, as needed. Now, notice that ∂0 takes C0 (R) = Re onto C−1 (R) = R, and so

n∂ 0 ∂
· · · −→ Cn (R) −→ Cn−1 (R) −→ · · · −→ C0 (R) −→ R −→ 0

is an acyclic resolution of R as Re -module.

Since Cn (R) = R ⊗K (R ⊗K · · · ⊗K R) ⊗K R, we find
| {z }
n

Cn (R) = Re ⊗K Cn [R],

where
Cn [R] = R ⊗K · · · ⊗K R, n-times
and
C0 [R] = K.
Several things follow from this description of Cn (R): First, we see exactly how Cn (R) is an Re -module
and also see that it is simply the base extension of Cn [R] from K to Re . Next, we want a projective resolution,
so we want to insure that Cn (R) is indeed projective even over Re . For this we prove

Proposition 5.23 Suppose R is a K-algebra and R is projective as a K-module (in particular this holds if
R is K-free, for example when K is a field). Then

(1) Cn [R] is K-projective for n ≥ 0,

(2) R ⊗K Cn [R] is R-projective for n ≥ 0,

(3) Cn (R) is Re -projective for n ≥ 0.

Proof . This is a simple application of the ideas in Chapter 2, Section 2.6. Observe that (2) and (3) follow
from (1) because we have
HomR (R ⊗K Cn [R], T ) −g→ HomK (Cn [R], T ) (†)
and
HomRe (Re ⊗K Cn [R], T ) −
g→ HomK (Cn [R], T ) (††)
334 CHAPTER 5. HOMOLOGICAL ALGEBRA

where T is an R-module in (†) and an Re -module in (††). An exact sequence of R-modules (resp. Re -modules)
is exact as sequence of K-modules and (1) shows that the right sides of (†) and (††) are exact as functors of
T . Such exactness characterizes projectivity; so, (2) and (3) do indeed follow from (1).
To prove (1), use induction on n and Proposition 2.47 which states in this case

HomK (R ⊗K Cn−1 [R], T ) −

g→ HomK (Cn−1 [R], HomK (R, T )). (∗)

Now, T HomK (R, T ) is exact by hypothesis; so, the right hand side of (∗) is an exact functor of T by
induction hypothesis. Consequently, (∗) completes the proof.

Corollary 5.24 If the K-algebra, R, is K-projective, then

n∂ 0 ∂
· · · −→ Cn (R) −→ Cn−1 (R) −→ · · · −→ C0 (R) −→ R −→ 0

is an Re -projective resolution of the Re -module R.

The resolution of Corollary 5.24 is called the standard (or bar ) resolution of R. We can define the
homology and cohomology groups of the K-algebra R with coefficients in the two-sided R-module, M , as
follows:
Define the functors
H0 (R, −) : M M/M J
and
H 0 (R, −) : M {m ∈ M | rm = mr, all r ∈ R} = M R
to the category of K-modules. Here, the (left) ideal, J, of Re is defined by the exact sequence
∂
0 −→ J −→ Re −→
0
R −→ 0, (∗∗)

and is called the augmentation ideal of Re . It’s easy to check that M M/M J is right exact and M MR
is left exact. We make the definition

Definition 5.7 The n-th homology group of R with coefficients in the two-sided R-module, M , is

Hn (R, M ) = (Ln H0 )(M )

and the nth cohomology group with coefficients in M is

H n (R, M ) = (Rn H 0 )(M ).

We’ll refer to these groups as the Hochschild homology and cohomology groups of R even though our
definition is more general than Hochschild’s–he assumed K is a field and gave an explicit (co)cycle description.
We’ll recover this below and for this purpose notice that
The augmententation ideal, J, is generated (as left Re -ideal) by the elements r ⊗ 1 − 1 ⊗ rop for r ∈ R.
P P
To see this, observe that i ri ⊗ sop
i ∈ J iff we have i ri si = 0. But then
X X X X
ri ⊗ sop
i = ri ⊗ sop
i − ri si ⊗ 1 = (ri ⊗ 1)(1 ⊗ sop
i − si ⊗ 1).
i i i i

Now, to apply this, tensor our exact sequence (∗∗) with M :

1⊗∂
M ⊗Re J −→ M −→0 M ⊗Re R −→ 0,
5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 335

so we find
H0 (R, M ) = M/M J −
g→ M ⊗Re R.
It follows immediately that we have an isomorphism
e
Hn (R, M ) −
g→ TorR
n (M, R).

Similarly, we take HomRe (−, M ) of (∗∗) and get

θ
0 −→ HomRe (R, M ) −→ M −→ HomRe (J, M ).
The isomorphism
HomRe (Re , M ) −
g→ M
is just
f 7→ f (1),
e
thus if f ∈ HomRe (R , M ) and m = f (1), we find for ξ ∈ J that
(θ(f ))(ξ) = f (ξ) = ξm.
Therefore, f is in Ker θ iff ξm = 0 for all ξ ∈ J, where m = f (1). But, by the above, such ξ are generated by
r ⊗ 1 − 1 ⊗ rop , and so m ∈ Ker θ when and only when (r ⊗ 1)m = (1 ⊗ rop )m; i.e., exactly when rm = mr,
for all r ∈ R. We have proved that there is an isomorphism (of K-modules)
HomRe (R, M ) −
g→ M R = H 0 (R, M ).
Once again we obtain an isomorphism
ExtnRe (R, M ) −
g→ H n (R, M ).

Our discussion above proves the first two statements of

Theorem 5.25 If R is a K-algebra (with K contained in the center of R), then for any two-sided R-module,
M , we have canonical, functorial isomorphisms
e
Hn (R, M ) −
g→ TorR
n (M, R)

and
H n (R, M ) −
g→ ExtnRe (R, M ).
If R is K-projective, then homology can be computed from the complex
M ⊗K Cn [R]
with boundary operator
n−1
X
∂n (m ⊗ r1 ⊗ · · · ⊗ rn ) = mr1 ⊗ r2 ⊗ · · · ⊗ rn + (−1)i m ⊗ r1 ⊗ · · · ⊗ ri ri+1 ⊗ · · · ⊗ rn
i=1
+ (−1)n rn m ⊗ r1 ⊗ · · · ⊗ rn−1 ;
while cohomology can be computed from the complex
HomK (Cn [R], M )
with coboundary operator
n
X
(δn f )(r1 ⊗ · · · ⊗ rn ⊗ rn+1 ) = r1 f (r2 ⊗ · · · ⊗ rn+1 ) + (−1)i f (r1 ⊗ · · · ⊗ ri ri+1 ⊗ · · · ⊗ rn+1 )
i=1
(−1)n+1 f (r1 ⊗ · · · ⊗ rn )rn+1 .
336 CHAPTER 5. HOMOLOGICAL ALGEBRA

Proof . Only the statements about the explicit complex require proof. Since homology and cohomology are
given by specific Tor’s and Ext’s, and since the standard resolution is an Re -projective resolution of R, we
can use the latter to compute these Tor’s and Ext’s. Here, it will be important to know the Re -module
structure of Cn (R) and the fact that the map

r ⊗ sop 7→ sop ⊗ r

establishes a K-algebra isomorphism of Re and (Re )op .

Now, consider the map

Θ : M ⊗Re Cn (R) = M ⊗Re (Re ⊗K Cn [R]) −

g→ M ⊗K Cn [R].

Observe that M is treated as an (Re )op -module, the action being

m(r ⊗ sop ) = smr.

Thus,
op
Θ : m ⊗Re (r0 ⊗ · · · ⊗ rn+1 ) = m ⊗Re (r0 ⊗ rn+1 ) ⊗K (r1 ⊗ · · · ⊗ rn )
op
7→ [m · (r0 ⊗ rn+1 )] ⊗K (r1 ⊗ · · · ⊗ rn )
= (rn+1 mr0 ) ⊗K (r1 ⊗ · · · ⊗ rn ) ∈ M ⊗K Cn [R].

We now just have to see the explicit form of the boundary map induced on M ⊗K Cn [R] by the diagram

Θ−1
M ⊗Re Cn (R) o M ⊗K Cn [R]
1⊗∂n

M ⊗R e Cn−1 (R)
Θ / M ⊗K Cn−1 [R]

This goes as follows:

Θ−1
m ⊗K (r1 ⊗ · · · ⊗ rn ) −→ m ⊗Re (1 ⊗ 1) ⊗K (r1 ⊗ · · · ⊗ rn )
= m ⊗Re 1 ⊗ r1 ⊗ · · · ⊗ rn ⊗ 1
1⊗∂n
−→ m ⊗Re r1 ⊗ · · · ⊗ rn ⊗ 1
n−1
X
+ (−1)i m ⊗Re (1 ⊗ r1 ⊗ · · · ⊗ ri ri+1 ⊗ · · · ⊗ rn ⊗ 1)
i=1
+ (−1)n m ⊗Re (1 ⊗ r1 ⊗ · · · ⊗ rn )
n−1
X
Θ
−→ mr1 ⊗ r2 ⊗ · · · ⊗ rn + (−1)i m ⊗ r1 ⊗ · · · ⊗ ri ri+1 ⊗ · · · ⊗ rn
i=1
+ (−1)n rn m ⊗ r1 ⊗ · · · ⊗ rn−1 ,

exactly the formula of the theorem. For cohomology we proceed precisely the same way, but remember that
here M is treated as an Re -module. Details are left as a (DX).
When K is a field, the explicit (co)chain descriptions of Hn (R, M ) and H n (R, M ) apply; these are
Hochschild’s original descriptions for the (co)homology of K-algebras, Hochschild [25, 26].
By now, it should be clear that there is more than an analogy between the (co)homology of algebras and
that for groups. This is particularly evident from comparison of the original formula (Chapter 1, Section 1.4)
for cohomology of groups and Hochschild’s formula for the cohomology of the K-algebra, R. If we use just
5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 337

analogy then R will be replaced by Z[G] and K by Z; R is then free (rank = #(G)) over K. But, M is just
a left G-module (for cohomology) and for K-algebras, R, we assumed M was a two-sided R-module. This
is easily fixed: Make Z[G] act trivially on the right. Then, H 0 (Z[G], M ) is our old M G and the coboundary
formula becomes (it is necessary only to compute on σ1 ⊗ · · · ⊗ σn+1 as such tensors generate):

(δn f )(σ1 ⊗ · · · ⊗ σn+1 ) = σ1 f (σ2 ⊗ · · · ⊗ σn+1 )

X n
+ (−1)i f (σ1 ⊗ · · · ⊗ σi σi+1 ⊗ · · · ⊗ σn+1 )
i=1
+ (−1)n+1 f (σ1 ⊗ · · · ⊗ σn ),

as in Chapter 1. Therefore, keeping the analogy, for homology, where we have a right G-module, we should
make Z[G] act trivially on the left, and get the explicit formula:

∂n (m ⊗ σ1 ⊗ · · · ⊗ σn ) = mσ1 ⊗ σ2 ⊗ · · · ⊗ σn
n−1
X
+ (−1)i m ⊗ σ1 ⊗ · · · ⊗ σi σi+1 ⊗ · · · ⊗ σn (∗)
i=1
+ (−1)n m ⊗ σ1 ⊗ · · · ⊗ σn−1 ,

which formula we had in mind at the beginning of this discussion several pages ago.
The ideal J is generated by σ ⊗ 1 − 1 ⊗ σ op as σ ranges over G (σ 6= 1). Thus M J is the submodule
generated by {m − mσ | σ 6= 1}. Now the formula

σ −1 m = mσ (special for groups)

turns M into a left Z[G]-module and shows that M J is exactly our old IM and therefore proves the Hochschild
H0 (Z[G], M ) is our old H0 (G, M ).
However, all this is heuristic, it does not prove the Hochschild groups for Z[G] on our one-sided modules
are the (co)homology groups for G. For one thing, we are operating on a subcategory: The modules
e
with trivial action on one of their sides. For another, the Hochschild groups are Tor•Z[G] (−, Z[G]) and
Ext•Z[G]e (Z[G], −) not TorZ[G]
• (−, Z) and Ext•Z[G] (Z, −). We do know that everything is correct for cohomology
because of a previous argument made about universal δ-functors. Of course, it is perfectly possible to prove
that the groups
e n (G, M ) = Ker ∂n /Im ∂n+1
H
for ∂n given by (∗) above form a universal ∂-functor—they clearly form a ∂-functor and universality will
follow from the effaceability criterion (Theorem 5.10). The effaceing module will be M ⊗Z Z[G] in analogy
with Map(G, M ) (which is HomZ (Z[G], M )). Here, details are best left as an exercise.
Instead, there is a more systematic method that furthermore illustrates a basic principle handy in many
situations. We begin again with our K-algebra, R, and we assume there is a K-algebra homomorphism
 : R → K. Note that this is the same as saying all of (DX)

(i) K is an R-module (and R contains K in its center),


(ii) There is an R-module map R −→ K,

(iii) The composition K −→ R −→ K is the identity.

Examples to keep in mind are: K = Z, R = Z[G] and (σ) = 1, all σ ∈ G; K arbitrary (commutative),
R = K[T1 , . . . , Tn ] or KhT1 , . . . , Tn i and (Tj ) = 0, all j.
 338 CHAPTER 5. HOMOLOGICAL ALGEBRA

In general, there will be no such homomorphism. However for commutative K-algebras, R, we can
arrange a “section”4 , , after base extension. Namely, we pass to R ⊗K R and set (r ⊗ s) = rs ∈ R; so now
R plays the role of K and R ⊗K R the role of R. (The map  : R → K is also called an augmentation of
R as K-algebra.) Hence, the basic principle is that after base extension (at least for commutative R) our
K-algebra has a section; we operate assuming a section and then try to use descent (cf. Chapter 2, Section
2.8).
 This technique doesn’t quite work with non-commutative R, where we base extend to get Re = R⊗K Rop
and try ∂0 : Re → R for our . We certainly find that R is an Re -module, that ∂0 is an Re -module map,
i
that the composition R −→ Re −→ R (i(r) = r ⊗ 1) is the identity; but, R is not in the center of Re
and Re (with the multiplication we’ve given it) is not an R-algebra.

Notwithstanding this cautionary remark, we can do a descent-like comparison in the non-commutative

case provided R possesses a section  : R → K. In the first place, the section gives K a special position
as R-module. We write I = Ker , this is a two-sided ideal of R called the augmentation ideal . Further,
consider the augmentation sequence

0 −→ I −→ R −→ K −→ 0; (†)

by using condition (iii) above, we see that, as K-modules, R ∼

= I q K. The special position of K as R-module
leads to the consideration of the ∂-functor and δ-functor:

{H n (R, M ) = TorR
n (M, K)} (M an Rop -module)
n
{H (R, M ) = ExtnR (K, M )} (M an R-module)

which, as usual, are the derived functors of

M M/M I

and
M {m ∈ M | (∀ξ ∈ I)(ξm = 0)},
respectively. (Here, M is an Rop -module for the first functor and an R-module for the second.) You should
keep in mind the case: K = Z, R = Z[G], (σ) = 1 (all σ) throughout what follows. The idea is to compare
the Hochschild groups Hn (R, M ) and H n (R, M ) with their “bar” counterparts.
Secondly, we make precise the notion of giving a two-sided R-module, M , “trivial action” on one of its
sides5 . Given M , a two-sided R-module, we make ∗ M and ∗op M which are respectively an Rop -module
(“trivial action” on the left) and an R-module (“trivial action” on the right) as follows:

For m ∈ ∗ M and λop ∈ Rop , λop · m = mλ and for λ ∈ R, λ · m = (λ)m

and

For m ∈ ∗op M and λ ∈ R, λ · m = λm and for λop ∈ Rop , λop · m = m(λ).

Clearly, these ideas can be used to promote one-sided R-modules to two-sided ones (i.e., to Re -modules),
viz :
4 The term “section” is geometric: We have the “structure map” Spec R −→ Spec K (corresponding to K −→ R) and  gives

a continuous map: Spec K −→ Spec R so that Spec K −→ Spec R −→ Spec K is the identity.
5 Our earlier, heuristic, discussion was sloppy. For example, in the group ring case and for trivial action on the right, we

stated that Z[G] acts trivially on the right. But, n · 1 = n ∈ Z[G] and m · n 6= m if n 6= 1; so, our naive idea must be fixed.
5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 339

Any R, Rop , or Re -module is automatically a K-module and, as K is commutative so that K op = K, we

see that λ · m = m · λ for λ ∈ K in any of these cases. Now if we have an R-module, M , we make Re operate
by
(r ⊗ sop ) · m = rm(s) = r(s)m = (s)rm,
and similarly for Rop -modules, M , we use the action

(r ⊗ sop ) · m = (r)ms = m(r)s = ms(r).

When we use the former action and pass from an R-module to an Re -module, we denote that Re -module by
op e
∗ (M ); similarly for the latter action, we get the R -module ∗ (M ). And so we have pairs of functors
 ∗
 : Re -mod Rop -mod
op
∗ : R -mod Re -mod

and 
∗op : Re -mod R-mod
op
∗ : R-mod R e
-mod
As should be expected, each pair above is a pair of adjoint functors, the upper star is left adjoint to the
lower star and we get the following (proof is (DX)):
Proposition 5.26 If R is a K-algebra with a section  : R → K, then ∗ is left-adjoint to ∗ and similarly
for ∗op and op e 0
∗ . That is, if M is any R -module and T and T are respectively arbitrary R
op
and R-modules,
we have

HomRop (∗ M, T ) ∼
= HomRe (M, ∗ T )
∗op 0 ∼
HomR ( M, T ) = HomRe (M, op 0
∗ T ).

Lastly, we come to the comparison of the Hochschild groups with their “bar” counterparts. At first, it
will be simpler conceptually and notationally (fewer tensor product signs) to pass to a slightly more general
case: R and Re are merely rings and K and K e are chosen modules over R and R e respectively. In addition we
 
e −→ K.e By a map of the pair (R,e K)
e to (R, K), we understand
are given module surjections R −→ K and R
e

e
a ring homomorphism ϕ : R → R so that ϕ(Ker e ) ⊆ Ker . Of course, Ker  and Ker e  are just left ideals
and we obtain a map of groups, ϕ : Ke → K and a commutative diagram

e
R ϕ
/ R


e 

e 
K ϕ
/ K.

Now the ring map ϕ : Re → R makes every R-module an R-module

e (same for Rop -modules). So, K is an
e
R-module, e
and the diagram shows ϕ is an R-module map.
Suppose Pe• −→ K e −→ 0 is an R-projective
e e and P• −→ K −→ 0 is an R-projective
resolution of K
e
resolution of K. We form R ⊗Re K, then we get an R-module map

e −→ K
θ : R ⊗Re K

via
θ(r ⊗Re e
k) = rϕ(e
k).
e
(Note that as ϕ is an R-module, this makes sense.) Now the complex R ⊗Re Pe• is R-projective and surjects
e
to R ⊗Re K.
340 CHAPTER 5. HOMOLOGICAL ALGEBRA

By a slight generalization of Theorem 5.2, our R-module map lifts uniquely in Kom(R-mod) to a map

Θ : R ⊗Re Pe• −→ P•

(over θ, of course). Thus, if M is an Rop -module, we get the map on homology

H• (M ⊗Re Pe• ) = H• (M ⊗R (R ⊗Re Pe• ) −→ H• (M ⊗R P• ),

while if M is an R-module, we get the map on cohomology

H • (HomR (P• , M )) −→ H • (HomR (R ⊗Re Pe• , M )) = H • (HomRe (Pe• , M )).

But, H• (M ⊗Re Pe• ) computes TorR e eop

• (M, K) (where, M is an R -module through ϕ) and H• (M ⊗R P• )
e

R
computes Tor• (M, K). This gives the map of ∂-functors

TorR e R
• (M, K) −→ Tor• (M, K).
e

Similarly, in cohomology we get the map of δ-functors

e M ).
Ext•R (K, M ) −→ Ext•Re (K,

Our arguments give the first statement of

e K)
Theorem 5.27 If ϕ : (R, e → (R, K) is a map of pairs, then there are induced maps of ∂ and δ–functors

H• (M, ϕ) : TorR e R
• (M, K) −→ Tor• (M, K)
e

(for M ∈ Rop -mod), and

e M)
H • (M, ϕ) : Ext•R (K, M ) −→ Ext•Re (K,
(for M ∈ R-mod).
Moreover, the following three statements are equivalent:
(
a) θ : R ⊗Re Ke → K is an isomorphism, and
(1) R e
b) Torn (R, K) = (0) for n > 0,
e

(2) Both maps H• (M, ϕ) and H • (M, ϕ) are isomorphisms for all M ,
(3) The map H• (M, ϕ) is an isomorphism for all M .

Proof . (1) =⇒ (2). Write Pe• −→ K e Then R ⊗ e Pe• −→ R ⊗ e K

e −→ 0 for a projective resolution of K.
R R
e −→ 0
e
is an R-projective complex over R ⊗Re K. By (1)b), it is acyclic and by (1)a) we obtain an R-projective
resolution of K. Thus, we may choose as R-projective resolution of K the acyclic complex R ⊗Re Pe• . But
then, Θ is the identity and (2) follows.
(2) =⇒ (3). This is a tautology.
(3) =⇒ (1). We apply the isomorphism H• (M, ϕ) for M = R. This gives us the isomorphism

TorR e g → TorR
• (R, K) − • (R, K).
e

We get (1)a) from the case 0 and (1)b) from n > 0.

e K)
Corollary 5.28 If ϕ : (R, e → (R, K) is a map of pairs and conditions (1)a) and b) of Theorem 5.27 hold,
e e say Pe• −→ K
then for any R-projective resolution of K, e −→ 0, the complex R ⊗ e Pe• is an R-projective
R
resolution of K.
5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 341

Proof . This is exactly what we showed in (1) =⇒ (2).

We apply these considerations to the comparison of the Hochschild groups and their bar counterparts.
e (and, since (Re )op is K-isomorphic to Re by the map τ : sop ⊗r 7→ r⊗sop ,
The idea is to cast Re in the role of R
e op
cast (R ) as R, e too). The role of K e is played by R for Re and by Rop for (Re )op . Then R and K are just
themselves and, in the op-case, we use Rop and K.
Now ∂0 : Re → R, resp. ∂0 : (Re )op → Rop , by ∂0 (r ⊗ sop ) = rs, resp. ∂0 (sop ⊗ r) = sop rop = (rs)op , is
an Re -module map, resp. an (Re )op -module map. Moreover, the diagram

(Re )op
∼ / Re
τ

∂0 ∂0
 
Rop
op
/ R
=

commutes for our formulae for ∂0 . So, we cast ∂0 as e . But we need the map of pairs and this is where our
section, , is essential . Define ϕ : Re → R (resp. (Re )op −→ Rop ) by

ϕ(r ⊗ sop ) = r(s) (resp. ϕ(sop ⊗ r) = sop (r)).

Clearly, ϕ is a ring homomorphism and as Ker e is generated by r ⊗ 1 − 1 ⊗ rop (resp. rop ⊗ 1 − 1 ⊗ r), we
find ϕ(Ker e) ⊆ Ker . There results the commutative diagram of the map of pairs:

(Re )op
ϕ
/ Rop
II
II ∼
I=I
τ II
I$
Re
ϕ
/R 
∂0 =e

u DD
uu DD 
uuu DD
u DD
 zuu ∂0 =e ! 
op
R =R
 /K

Now consider an R-module, M (resp. an Rop -module, M ), how does ϕ make M an Re (resp. (Re )op )-
module? This way:

(r ⊗ sop ) · m = ϕ(r ⊗ sop ) · m = r(s)m

(resp. (sop ⊗ r) · m = ϕ(sop ⊗ r) · m = sop (r) · m = ms(r)).

That is, the R-module, M , goes over to the Re -module op op

∗ (M ) and the R -module, M , goes over to the
e op
(R ) -module ∗ (M ). Therefore, the map of pairs yields the comparison maps
e
H• (M, ϕ) : H• (R, ∗ (M )) = TorR R
• (∗ (M ), R) −→ Tor• (M, K) = H • (R, M )
•
H • (M, ϕ) : H (R, M ) = Ext•R (K, M ) −→ Ext•Re (R, op • op
∗ (M )) = H (R, ∗ (M )).

Theorem 5.29 If R is K-projective, then the comparison maps

H• (M, ϕ) : H• (R, ∗ (M )) −→ H • (R, M )

and
•
H • (M, ϕ) : H (R, M ) −→ H • (R, op
∗ (M ))

are isomorphisms of ∂ (resp. δ)-functors. Moreover, if Pe• −→ R −→ 0 is an Re -projective resolution of R,

then Pe• ⊗R K is an R-projective resolution of K.
342 CHAPTER 5. HOMOLOGICAL ALGEBRA

Proof . Everything will follow from Theorem 5.27 once we verify conditions (1)a) and b) of that theorem.
Here, there is the non-commutativity of R that might cause some confusion. Recall that Re operates on the
right on a module, N , via
n · (r ⊗ sop ) = snr;
so, Re operates on ∗ N via
n · (r ⊗ sop ) = (s)nr.
We apply this when N is R–itself and M is any two-sided R-module. For ρ ⊗Re m ∈ ∗ R ⊗Re M , we observe
that
(s)ρr ⊗Re m = ρ · (r ⊗ sop ) ⊗Re m = ρ ⊗Re rms; (∗)
hence, the map
α : ∗ R ⊗Re M −→ M ⊗R K
via
α(ρ ⊗Re m) = ρm ⊗R 1
is well–defined. The only (mildly) tricky thing to check is that α preserves relation (∗). But, α of the left
side of (∗) is (s)ρrm ⊗R 1 and α of the right side of (∗) is ρrms ⊗R 1. Now,

zs ⊗R 1 = z ⊗R (s) = z(s) ⊗R 1;

so, α agrees on the left and right sides of (∗). And now we see that α is an isomorphism of K-modules
because the map
β : M ⊗R K −→ ∗ R ⊗Re M
via
β(m ⊗R κ) = κ ⊗Re m
is its inverse. (Note that β is well-defined for:

mρ ⊗R κ = m ⊗R (ρ)κ

and
(ρ)κ ⊗Re m = κ · (1 ⊗ ρop ) ⊗Re m = κ ⊗Re mρ = β(mρ ⊗R κ),
while
(ρ)κ ⊗Re m = β(m ⊗R (ρ)κ), as required.)
However,
αβ(m ⊗R κ) = α(κ ⊗Re m) = κm ⊗R 1 = m ⊗R κ
βα(ρ ⊗Re m) = β(ρm ⊗R 1) = 1 ⊗Re ρm = ρ ⊗Re m.
e We find that
We can now apply the K-module isomorphism α. First, take M = R (= K).

α :  ∗ R ⊗R e R −
g→ R ⊗R K = K

and ∗ R is just R as Re (= Re -module). This gives (1)a). To see (1)b), take Pe• −→ R −→ 0 an Re (= R)-
e
e e
projective resolution. We choose M = P• an R -module (i.e., a complex of same). Now apply α:

e Re e g
TorR
• (R, K) = Tor• (∗ R, R) = H• (∗ R ⊗Re P• ) −→ H• (Pe• ⊗R K).
e
α

But, R is K-projective and so (by the usual arguments (DX)) Pe• is Rop -projective which means the last
homology complex computes TorR • (R, K). We’ve shown

TorR e g → TorR
n (R, K) − n (R, K).
e
α
5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 343

Yet, R is free (so flat) over R and so TorR

n (R, K) = (0) when n > 0; we are done.

Of course, we should apply all this to the standard resolution, C• (R), when R is K-projective. Here,

Cn (R) ⊗R K = R ⊗K Cn [R] ⊗K R ⊗R K −
g→ R ⊗K Cn [R]

via the map

Θn (r0 ⊗ · · · ⊗ rn+1 ⊗R κ) = (rn+1 )κ(r0 ⊗ · · · ⊗ rn ).
As in the proof of Theorem 5.25, the standard boundary map induces a boundary map, ∂ n , on R ⊗K Cn [R],
by the formula ∂ n = Θn−1 ◦ ∂n ◦ Θ−1
n , and we find

n−1
X
∂ n (r0 ⊗ r1 ⊗ · · · ⊗ rn ) = (−1)i r0 ⊗ · · · ⊗ ri ri+1 ⊗ · · · ⊗ rn + (−1)n (rn )r0 ⊗ · · · ⊗ rn−1 .
i=0

This gives us our R-projective resolution R ⊗K C• [R] −→ K −→ 0 with which we can compute. The case
when r0 = 1 is most important:
n−1
X
∂ n (1 ⊗ r1 ⊗ · · · ⊗ rn ) = r1 ⊗ · · · ⊗ rn + (−1)i r1 ⊗ · · · ⊗ ri ri+1 ⊗ · · · ⊗ rn + (−1)n (rn )(1 ⊗ r1 ⊗ · · · ⊗ rn−1 ).
i=1

Now, for a right R-module, M , the groups TorR

• (M, K) are the homology of

M ⊗R R ⊗K C• [R] = M ⊗K C• [R]

under 1 ⊗R ∂. We find

∂ n (m ⊗K r1 ⊗K · · · ⊗K rn ) = mr1 ⊗K r2 ⊗K · · · ⊗K rn
n−1
X
+ (−1)i m ⊗K r1 ⊗K · · · ⊗K ri ri+1 ⊗K · · · ⊗K rn
i=1
+ (−1)n (rn )m ⊗K r1 ⊗K · · · ⊗K rn−1 .

Therefore, we recover Hochschild’s homology formula for ∗ (M ), and when R = Z[G] and K = Z (with
(σ) = 1, all σ ∈ G) we also recover the explicit boundary formula for H• (G, M ).
For a left R-module, M , the groups Ext•R (K, M ) are the cohomology of

HomR (R ⊗R C• [R], M ) = HomK (C• [R], M ).

If, as usual, we write f (r1 , . . . , rn ) for f (r1 ⊗K r2 ⊗K · · · ⊗K rn ), then

(δ n f )(r1 , . . . , rn+1 ) = r1 f (r2 , . . . , rn+1 )

X n
+ (−1)i f (r1 , . . . , ri ri+1 , . . . , rn )
i=1
+ (−1)n+1 (rn+1 )f (r1 , . . . , rn ).

Here, f ∈ HomK (Cn [R], M ). Once again, we recover Hochschild’s cohomology formula for op
∗ (M ), and when
R = Z[G], etc., we get our explicit coboundary formula for H • (G, M ).
But, we’ve done more; all this applies to any K-algebra, R, with a section (especially for K-projective
algebras). In particular, we might apply it to R = K[T1 , . . . , Tn ] or R = KhT1 , . . . , Tn i, with (Tj ) = 0
for j = 1, 2, . . . , n. The standard resolution though is very inefficient for we must know m ⊗ r1 ⊗ · · · ⊗ rl
344 CHAPTER 5. HOMOLOGICAL ALGEBRA

or f (r1 , . . . , rl ) on all monomials rj of whatever degree. Instead we will find a better resolution, but we
postpone this until Section 5.5 where it fits better.
Let us turn to the cohomology of sheaves and presheaves. These objects have been introduced already
and we assume that Problem 69 has been mastered. Here, we’ll be content to examine ordinary topological
spaces (as in part (a) of that exercise) and (pre)sheaves on them. The most important fact is that the
categories of presheaves and sheaves of R-modules on the space X have enough injectives. Let us denote by
P(X, R-mod) and S(X, R-mod) these two abelian categories. Remember that

0 −→ F 0 −→ F −→ F 00 −→ 0

is exact in P(X, R-mod) iff the sequence of R-modules

0 −→ F 0 (U ) −→ F (U ) −→ F 00 (U ) −→ 0

is exact for every open U of X. But for sheaves, the situation is more complicated:

0 −→ F 0 −→ F −→ F 00 −→ 0

is exact in S(X, R-mod) iff

(a) 0 −→ F 0 (U ) −→ F (U ) −→ F 00 (U ) is exact for every open U or X and
(b) For each open U and each ξ ∈ F 00 (U ), there is an open cover {Uα −→ U }α so that each ξα (= ρU
U (ξ))
α

00
is the image of some ηα ∈ F (Uα ) under the map F (Uα ) −→ F (Uα ).

A more perspicacious way of saying this is the following: Write i : S(X, R-mod) P(X, R-mod) for the
full embedding which regards a sheaf as a presheaf. There is a functor, # : P(X, R-mod) S(X, R-mod)
which is left adjoint to i. That is, for F ∈ S and G ∈ P, we have

HomS (G# , F ) −
g→ HomP (G, i(F )).6

We can now say (b) this way: If cok(F −→ F 00 ) is the presheaf cokernel

cok(F −→ F 00 )(U ) = cok(F (U ) −→ F 00 (U )),

then cok(F −→ F 00 )# = (0).

Given x ∈ X, and a (pre)sheaf, F , we define the stalk of F at x, denoted Fx , by

Fx = lim
−→ F (U ).
{U 3 x}

It’s easy to see that (F # )x = Fx for any presheaf, F . Stalks are important because of the following simple
fact:
ϕ
Proposition 5.30 If F −→ G is a map of sheaves, then ϕ is injective (surjective, bijective) if and only if
the induced map ϕx : Fx → Gx on stalks is injective (surjective, bijective) for every x ∈ X.

We leave the proof as a (DX).

6 One constructs # by two successive limits. Given U , open in X, write G(+) (U ) for
Y Y
G(+) (U ) = lim Ker ( G(Uα ) −→
−→ G(Uβ ∩ Uγ ))
−→
{Uα −→U } α β,γ

(the limit taken over all open covers of U ) and set G# (U ) = G(+)(+) (U ).
 5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 345

 This result is false for presheaves, they are not local enough.

Property (a) above shows that i is left-exact and the proposition shows # is exact. To get at the existence
of enough injectives, we investigate what happens to P(X, R-mod) and S(X, R-mod) if we have a map of
spaces f : X → Y . In the first place, if F is a (pre)sheaf on X, we can define a (pre)sheaf f∗ F , called the
direct image of F by f via
(f∗ F )(V ) = F (f −1 (V )), V open in Y .
A simple check shows that the direct image of a sheaf is again a sheaf. Now, in the second place, we want a
functor f ∗ : P(Y ) P(X) (resp. S(Y ) S(X)) which will be left adjoint to f∗ . If we knew the (classical)
way to get a sheaf from its stalks, we could set (f ∗ G)x = Gf (x) for G ∈ S(Y ) and x ∈ X any point. But
from our present point of view this can’t be done. However, our aim is for an adjoint functor, so we can use
the method of D. Kan [30].
We start we a presheaf, G, on Y and take an open set, U , of X. We set

(f ∗ G)(U ) = lim
−→ G(V ),
{f −1 (V )⊇U }

here, as noted, V ranges over all opens of Y with f −1 (V ) ⊇ U . Then, f ∗ G is a presheaf (of R-modules) on
X. If G is a sheaf on Y , we form f ∗ G, as above, and then take (f ∗ G)# . We’ll continue to denote the latter
sheaf by f ∗ G if no confusion results. Once the idea of defining f ∗ G by a direct limit is in hand, it is easy to
prove (and the proof will be left as a (DX)):

Proposition 5.31 If f : X → Y is a map of topological spaces, then the functors f ∗ from P(Y ) to P(X)
(resp. from S(Y ) to S(X)) are left adjoint to the direct image functors. That is, for G ∈ P(Y ) and F ∈ P(X)
(resp. G ∈ S(Y ) and F ∈ S(X)), we have functorial isomorphisms

HomP(X) (f ∗ G, F ) −
g→ HomP(Y ) (G, f∗ F )

(resp.
HomS(X) (f ∗ G, F ) −
g→ HomS(Y ) (G, f∗ F )).
Moreover, we have (f ∗ G)x = Gf (x) , for all x ∈ X.

∗ ∗
Since −→ is an exact functor on R-mod, our definition of the presheaf f G shows that f is an exact∗
lim
functor P(Y ) P(X). The statement in the proposition about stalks shows (by Proposition 5.30) that f
is also an exact functor S(Y ) S(X). Of course, f∗ is a left-exact functor on sheaves and an exact functor
on presheaves.
There is a useful lemma that connects pairs of adjoint functors and injectives—it is what we’ll use to get
enough injectives in P and S.

Lemma 5.32 Say A and B are abelian categories and α : A B and β : B A are functors with β left
adjoint to α. If β is exact, then α carries injectives of A to injectives of B.

Proof . Take an injective, Q, of A and consider the co-functor (on B)

T HomB (T, α(Q)).

By adjointness, this is exactly

T HomA (β(T ), Q).
Now, HomA (β(−), Q) is the composition of the exact functor β with the exact functor HomA (−, Q) (the
latter being exact as Q is injective). But then, HomB (−, α(Q)) is exact, i.e., α(Q) is injective in B.
If we apply the lemma to the cases α = i, β = #; α = f∗ , β = f ∗ , we get
346 CHAPTER 5. HOMOLOGICAL ALGEBRA

Corollary 5.33 Let f : X → Y be a map of topological spaces and write P(X), etc. for the categories of
R-module presheaves on X, etc. Further consider the functors i : S(X) P(X) and # : P(X) S(X).
Then,

(1) If Q is an injective in P(X), the presheaf f∗ (Q) is injective on Y .

(2) If Q is an injective in S(X), the sheaf f∗ (Q) is injective on Y .

(3) If Q is an injective sheaf on X, then i(Q) is an injective presheaf on X.

Theorem 5.34 If X is a topological space, then the category S(X, R-mod) possesses enough injectives.

Proof . Pick any point, ξ, of X and consider the map of spaces iξ : {ξ} ,→ X. The categories P({ξ}) and
S({ξ}) are each just R-mod, and for any module, M , we have

M if ξ ∈ U
iξ∗ (M )(U ) =
(0) if ξ ∈
/ U.

For any sheaf F on X, look at its stalk, Fξ , at ξ and embed Fξ into an injective R-module Qξ (say jξ : Fξ ,→ Qξ
is theQembedding). We form iξ∗ (Qξ ) which is an injective sheaf on X by Corollary 5.33 and then form
Q = ξ∈X iξ∗ (Qξ ), again an injective sheaf on X. Note that
Y
Q(U ) = Qξ .
ξ∈U

Now, I claim that the map θ : F → Q via

for z ∈ F (U ) : θ(z) = (jξ (zξ ))ξ∈U ,

where zξ is the image of z in Fξ , is the desired embedding. If θ(z) = 0, then for each ξ ∈ U , the elements
jξ (zξ ) = 0; as jξ is an embedding, we get zξ = 0. By the definition of stalk, there is a neighborhood, Uξ , of
U
ξ in U where ρUξ (z) = 0. These neighborhoods give a covering of U , so we see that z goes to zero under the
map Y
F (U ) −→ F (Uξ ). (+)
ξ∈U

But, this map is injective by the sheaf axiom; so, z = 0.

Remark: The theorem is also true for presheaves and our proof above works for “good” presheaves; that is,
those for which the maps (+) are indeed injective. (For general presheaves, G, the presheaf G(+) will satisfy
(+) is injective). We can modify the argument to get the result for P(X) or use a different argument; this
will be explored in the exercises.
To define cohomology with coefficients in a sheaf, F , on X, we consider the functor

Γ: F F (X).

We already know this is left exact and we define the cohomology of X with coefficients in F by

H • (X, F ) = (R• Γ)(F ).

A little more generally, if U is open in X, we can set ΓU (F ) = F (U ) and then

H • (U, F ) = (R• ΓU )(F ).

 5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 347

If we assume proved the existence of enough injectives in P(X), then for a presheaf, G, we set
Y Y
Ȟ 0 (X, G) = G(+) (X) = lim Ker ( G(U ) −→ G(Uβ ∩ Uγ ))
−→ α −→
{Uα −→U } α β,γ

and define
Ȟ • (X, G) = (R• Ȟ 0 )(G).
There is an explicit complex that computes Ȟ • (X, G), see the exercises. The R-modules Ȟ • (X, G) are called
the Čech cohomology groups of X with coefficients in the presheaf G. Again, as above, we can generalize to
cohomology over an open, U .
Pick open U ⊆ X, and write RU for the presheaf

R if V ⊆ U
RU (V ) =
(0) if V ⊆
6 U

(so, RX is the constant presheaf, R). Also write RU for the sheaf (RU )# . It turns our that the RU form a
set of generators for P(X), while the same is true for the RU in S(X). Moreover, we have
Proposition 5.35 If X is a topological space and U is a given open set, then we have an isomorphism of
δ-functors
H • (U, −) ∼
= Ext•S(X) (RU , −)
on the category S(X) to R-mod.
Proof . All we have to check is that they agree in dimension 0. Now,
HomS(X) (RU , F ) ∼
= HomP(X) (RU , i(F )).
Notice that ρVU : R(U ) → R(V ) is just the identity if V ⊆ U and is the zero map otherwise. It follows that
HomP(X) (RU , i(F )) ∼
= HomR-mod (R, F (U )) = F (U ),
and we are done.
 We don’t compute Ext•S(X) (RU , F ) by projectively resolving RU —such a resolution doesn’t exist in
S(X). Rather, we injectively resolve F .

Recall that we have the left exact functor i : S(X) P(X), so we can inquire as to its right derived
functors, R• i. The usual notation for (R• i)(F ) is H• (F )—these are presheaves. We compute them as follows:
Proposition 5.36 The right derived functors H• (F ) are given by
H• (F )(U ) = H • (U, F ).
Proof . It should be clear that for fixed F , each H p (U, F ) is functorial in U ; that is, U H p (U, F ) is a
presheaf. Moreover, it is again clear that
F H • (U, F )
is a δ-functor from S(X) to P(X). If Q is injective in S(X), we have H p (U, Q) = (0) when p > 0; so, our
δ-functor is effaceable. But, for p = 0, the R-module H 0 (U, F ) is just F (U ); i.e., it is just H0 (F )(U ). By
the uniqueness of universal δ-functors, we are done.
For the computation of the cohomology of sheaves, manageable injective resolutions turn out to be too
hard to find. Sometimes one can prove cohomology can be computed by the Čech method applied to i(F ),
and then the explicit complex of the exercises works quite well. This will depend on subtle properties of the
space, X. More generally, Godement [18] showed that a weaker property than injectivity was all that was
needed in a resolution of F to compute the R-module H • (X, F ). This is the notion of flasqueness.7
7 The French word “flasque” can be loosely translated as “flabby”.
348 CHAPTER 5. HOMOLOGICAL ALGEBRA

Definition 5.8 A sheaf, F , on the space X is flasque if and only if for each pair of opens V ⊆ U of X, the
map
ρVU : F (U ) → F (V )
is surjective. Of course, this is the same as F (X) −→ F (U ) being surjective for each open, U .

Here are two useful lemmas that begin to tell us how flasque sheaves intervene in cohomology.:

Lemma 5.37 The following are equivalent statements about a sheaf, F 0 , on the space X:

(1) Every short exact sequence in S(X)

0 −→ F 0 −→ F −→ F 00 −→ 0

is exact in P(X).

(2) For every open U of X, the R-module H 1 (U, F 0 ) is zero.

Proof . (1) =⇒ (2). Embed F 0 in an injective and pick open U ⊆ X. From 0 −→ F 0 −→ Q −→ cok −→ 0,
we get
0 −→ F 0 (U ) −→ Q(U ) −→ cok(U ) −→ H 1 (U, F 0 ) −→ (0);
By (1), 0 −→ F 0 (U ) −→ Q(U ) −→ cok(U ) −→ 0 is exact; so, H 1 (U, F 0 ) = (0).
(2) =⇒ (1). Just apply cohomology over U to the short exact sequence 0 −→ F 0 −→ F −→ F 00 −→ 0.
We get
0 −→ F 0 (U ) −→ F (U ) −→ F 00 (U ) −→ H 1 (U, F 0 ).
By (2), we’re done as U is an arbitrary open.

Lemma 5.38 Say F 0 is a flasque sheaf and 0 −→ F 0 −→ F −→ F 00 −→ 0 is exact in S(X). Then it is

exact in P(X). Moreover, if 0 −→ F 0 −→ F −→ F 00 −→ 0 is exact in S(X), then F is flasque if and only if
F 00 is flasque (of course, F 0 is always assumed to be flasque).

Proof . Pick any open U ⊆ X; all we must prove is that F (U ) −→ F 00 (U ) is surjective. Write Σ for the
collection of pairs (V, σ) where V is open, V ⊆ U and σ is a lifting to F (V ) of ρVU (s) ∈ F 00 (V ) for some
s ∈ F 00 (U ) fixed once and for all. As s admits liftings to F locally on U , our set Σ is non-empty. Now
partially order Σ in the standard way: (V, σ) ≤ (Ve , σ
e) iff V ⊆ Ve and ρVVe (e
σ ) = σ. Of course, Σ is inductive
and Zorn’s Lemma yields a maximal lifting, σ0 , of s defined on V0 ⊆ U . We must prove V0 = U .
Were it not, there would exist ξ ∈ U with ξ ∈ / V0 . Now the stalk map Fξ −→ Fξ00 is surjective, so the
image of s in some small neighborhood, U (ξ), of ξ in U lifts to an element τ ∈ F (U (ξ)). We will get an
e = U (ξ) ∪ V0 has two opens as a disjoint cover and
immediate contradiction if U (ξ) ∩ V0 = ∅, for then U
e Q
F (U ) = F (U (ξ)) F (V0 ) by the sheaf axiom. The pair hτ, σ0 i is a lifting of s to a bigger open than V0 —a
contradiction.
Therefore, we may assume U (ξ) ∩ V0 6= ∅—it is here that the flasqueness of F 0 enters. For on the
U (ξ)∩V U (ξ)∩V0 U (ξ)∩V0
intersection both ρU (ξ) 0 (τ ) and ρV0 (σ0 ) lift ρU (s). Thus, there is an “error”  ∈ F 0 (U (ξ) ∩ V0 ),
so that
U (ξ)∩V0 U (ξ)∩V
ρV0 (σ0 ) − ρU (ξ) 0 (τ ) = .
U (ξ)
As F 0 is flasque,  lifts to F 0 (U (ξ)); call it  again on this bigger open. Then τ +  also lifts ρU (s) and τ + 
and σ0 now agree on U (ξ) ∩ V0 ; so, the sheaf axiom shows we get a lifting to the bigger open U (ξ) ∪ V0 —our
last contradiction. Thus, U = V0 .
5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 349

For the second statement, in which F 0 is given as a flasque sheaf, pick open V ⊆ U in X. We have the
commutative diagam
0 / F 0 (U ) / F (U ) / F 00 (U ) / 0

ρ0 ρ ρ00
  
0 / F 0 (V ) / F (V ) / F 00 (V ) / 0,

in which Coker (ρ0 ) = (0), By the snake lemma

Coker (ρ) −
g→ Coker (ρ00 ),

and we are done.

Remark: There is an important addendum to Lemma 5.37. We mention this as the method of argument
is fundamental in many applications. This addendum is: The statement
(3) Ȟ 1 (U, i(F 0 )) = (0) for all U open in X, is equivalent to either properties (1) or (2) of Lemma 5.37.
S
Let us show (3) ⇐⇒ (1). So say (3) holds. This means given any open cover of U , say U = α Uα , and
any elements zαβ ∈ F 0 (Uα ∩ Uβ ) so that

zαβ = −zβα and zαγ = zαβ + zβγ on Uα ∩ Uβ ∩ Uγ (∗)

we can find elements zα ∈ F 0 (Uα ) so that zαβ = zα − zβ on Uα ∩ Uβ . Now suppose we have s ∈ F 00 (Uα ),
we can cover U by opens Uα so that the sα = ρU 00
U (s) ∈ F (Uα ) lift to σα ∈ F (Uα ) for all α. The elements
α

σα − σβ ∈ F (Uα ∩ Uβ ) are not necessarily 0 but go to zero in F 00 (Uα ∩ Uβ ). That is, if we set zαβ = σα − σβ ,
the zαβ belong to F 0 (Uα ∩ Uβ ). These zαβ satisfy (∗) and so by (3) we get zαβ = zα − zβ for various
zα ∈ F 0 (Uα ). Thus
zα − zβ = σα − σβ on Uα ∩ Uβ ,
that is
σα − zα = σβ − zβ on Uα ∩ Uβ .
This equality and the sheaf axiom for F give us an element σ ∈ F (U ) with ρU
U (σ) = σα − zα . The zα go to
α

00 00
zero in F , thus σ lifts s and this shows F (U ) −→ F (U ) is surjective.
To show (1) =⇒ (3), we simply embed F 0 in an injective again to get 0 −→ F 0 −→ Q −→ cok −→ 0 in
S(X). By (1), the sequence
0 −→ i(F 0 ) −→ i(Q) −→ i(cok) −→ 0
is exact in P(X) and i(Q) is an injective of P(X). Apply Čech cohomology (a δ-functor on P(X)):

0 −→ F 0 (U ) −→ Q(U ) −→ cok(U ) −→ Ȟ 1 (U, i(F 0 )) −→ 0

is exact. Since Q(U ) −→ cok(U ) is surjective, by (1), we get (3).

Proposition 5.39 Every injective sheaf is a flasque sheaf. For every flasque sheaf, F , and every open U ,
we have H n (U, F ) = (0) for n > 0.

Proof . Pick open V ⊆ U , call our injective sheaf Q. Since V ⊆ U , we have the exact sequence

0 −→ RV −→ RU −→ cok −→ 0

in S(X). Now HomS(X) (−, Q) is an exact functor; so

0 −→ HomS(X) (cok, Q) −→ HomS(X) (RU , Q) −→ HomS(X) (RV , Q) −→ 0

350 CHAPTER 5. HOMOLOGICAL ALGEBRA

is exact. The middle and right-hand groups are Q(U ) and Q(V ) respectively; so Q is flasque.
Let’s prove the second statement by induction on n. Take a flasque sheaf, F . By Lemmas 5.37 and 5.38,
we have H 1 (U, F ) = (0) for all open U . Now embed F in an injective sheaf

0 −→ F −→ Q −→ cok −→ 0.

By Lemma 5.38 and the above, the sheaf cok is also flasque. So by our induction hypothesis all the groups
H p (U, cok) = (0) for 1 ≤ p ≤ n − 1. Now apply cohomology to our exact sequence; this gives

H p (U, cok) −
g→ H p+1 (U, F ), for p ≥ 1.

When p = n − 1 this gives the induction step for F .

The important part of this proposition is that flasque sheaves are acyclic objects over every open U of
X; that is, H n (U, F ) = (0) for all n > 0 and every open U . And now we have the following general

Proposition 5.40 Suppose A is an abelian category, T is a left exact functor from A to R-mod and

0 −→ F −→ L0 −→ L1 −→ · · ·

is an acyclic resolution of F by R• T -acyclic objects L0 , L1 , . . . (That is, (Rp T )(Lj ) = (0), all p > 0, all
j ≥ 0.) Then the cohomology of the complex

T (L0 ) −→ T (L1 ) −→ T (L2 ) −→ · · ·

computes the derived functors of T on the object F (i.e., the (Rn T )(F ), n ≥ 0).

Proof . We use induction on n and décalage. (There are other methods of proof, e.g., by double complexes.)
Of course, the assertion for n = 0 is true and trivial, being independent of resolutions. So assume for an
object G resolved by a sequence of the L’s, we have (Rp T )(G) = H p (T (L• )) for 0 ≤ p ≤ n − 1. Then the
sequence,
0 −→ F −→ L0 −→ cok −→ 0
yields the resolution
0 −→ cok −→ L1 −→ L2 −→ · · · (†)
and the cohomology sequence

ED
BC
0 / T (F ) / T (L0 ) / T (cok) / (R1 T )(F ) / ···

GF
@A
/ (Rp T )(L0 ) / (Rp T )(cok) / (Rp+1 T )(F ) / (Rp+1 T )(L0 ) / ··· .

If p ≥ 1, we get
(Rp T )(cok) −
g→ (Rp+1 T )(F )
while for p = 0, we get
T (L0 ) −→ T (cok) −→ (R1 T )(F ) −→ 0. (††)
Now by induction hypothesis on n, (Rp T )(cok) is the pth cohomology of T (L1 ) −→ T (L2 ) −→ · · · , for
p ≤ n − 1. This is exactly the p + 1 cohomology of T (L0 ) −→ T (L1 ) −→ · · · , that is H p+1 (T (L• )). So,

H p+1 (T (L• )) = (Rp T )(cok) −

g→ (Rp+1 T )(F )

if 1 ≤ p ≤ n − 1. As long as n ≥ 2, we can set p = n − 1 and get the induction step.

5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 351

So, all that remains is the step from n = 0 to n = 1. From (†), we see

T (cok) −
g→ Ker (T (L1 ) −→ T (L2 )).

By the short exact sequence for F, L0 , cok, we find that Im (T (L0 ) −→ T (cok)) is exactly the image
(T (L0 ) −→ T (L1 )); that means
T (cok)/T (L0 ) = H 1 (T (L• )).
But, we know
T (cok)/T (L0 ) −
g→ (R1 T )(F )
by (††), and we are done.
Of course, we apply this to resolving a sheaf, F , by flasque sheaves. If we do this, we get a complex
(upon applying ΓU ) and so from its cohomology we compute the H p (U, F ). It remains to give a canonical
procedure for resolving each F by flasque sheaves. This is Godement’s method of “discontinuous sections”.

Definition 5.9 For a sheaf, F , write G(F ) for the presheaf

Y
G(F )(U ) = Fx ,
x∈U

and call G(F ) the sheaf of discontinuous sections of F .

Remarks:
(1) G(F ) is always a sheaf.
S
(2) G(F ) is flasque. For, a section over V of G(F ) is merely a function on V to x∈V Fx so that its value
at x lies in Fx . We merely extend by zero outside V and get our lifting to a section of U (with U ⊇ V ).
(3) There is a canonical embedding F −→ G(F ). To see this, if s ∈ F (U ), we have s(x) ∈ Fx , its image
in Fx = − → F (V ). We send s to the function x 7→ s(x) which lies in G(F )(U ). Now, if s and t go to
lim
V3x
the same element of G(F )(U ), we know for each x ∈ U , there is a small open U (x) ⊆ U where s = t
U (x) U (x)
on U (x) (i.e., ρU (s) = ρU (t)). But these U (x) cover U , and the sheaf axiom says s = t in F (U ).

Therefore, S(X, R-mod) has enough flasques; so every sheaf, F , possesses a canonical flasque resolution
(the Godement resolution) : Namely

0 −→ F −→ G(F ) −→ cok1 −→ 0
0 −→ cok1 −→ G(cok1 ) −→ cok2 −→ 0
......................
0 −→ cokn −→ G(cokn ) −→ cokn+1 −→ 0 · · ·

This gives
0 −→ F −→ G0 (F ) −→ G1 (F ) −→ · · · −→ Gn (F ) −→ · · · ,
where we have set
G0 (F ) = G(F ) and Gn (F ) = G(cokn ) when n ≥ 1.

It’s not hard to extend all our results on sheaves of R-modules to sheaves of OX -modules, where OX
is a sheaf of rings on X. To replace maps of spaces, we need the notion of a map of ringed spaces (i.e.,
of pairs (X, OX ) in which OX is a sheaf of rings on X): By a map (X, OX ) −→ (Y, OY ) of ringed spaces,
we understand a pair (f, ϕ) in which f is a map X −→ Y and ϕ is a map of sheaves OY −→ f∗ OX (over
352 CHAPTER 5. HOMOLOGICAL ALGEBRA

Y ). For intuition think of OX as the sheaf of germs of continuous functions on X. If F is an OX -module,

then f∗ F will be an OY -module thanks to the map OY −→ f∗ OX . But if G is an OY -module, f ∗ G is not
an OX -module. We must augment the notion of inverse image. Our map ϕ : OY → f∗ OX corresponds by
adjunction to a map ϕ e : f ∗ OY → OX . Now f ∗ G is an f ∗ OY -module, so we form

(f, ϕ)∗ G = OX ⊗f ∗ OY f ∗ G

and get our improved notion of inverse image—an OX -module.

Finally, to end this long section we give some results (of a very elementary character) concerning
TorR
• (−, −) and properties of special rings, R. The first of these works for every ring:

Proposition 5.41 Say M is an R-module (resp. Rop -module), then the following are equivalent:

(1) M is R-flat.

(2) For all Z, we have TorR

n (Z, M ) = (0), for all n > 0.

(3) For all Z, we have TorR

1 (Z, M ) = (0).

Proof . (1) =⇒ (2). Since the functor Z Z ⊗R M is exact, its derived functors are zero for n > 0, i.e., (2)
holds.
(2) =⇒ (3). This is a tautology.
(3) =⇒ (1). Given an exact sequence

0 −→ Z 0 −→ Z −→ Z 00 −→ 0

tensor with M and take cohomology. We get the following piece of the long exact sequence
00 0 00
· · · −→ TorR
1 (Z , M ) −→ Z ⊗R M −→ Z ⊗R M −→ Z ⊗R M −→ 0.

By (3), we have TorR 00

1 (Z , M ) = (0), so the tensored sequence is exact.

For the rest, we’ll assume R is a domain. Now for a P.I.D. we know divisibility of a module is the same
as injectivity. That’s not true in general, but we have

Proposition 5.42 If R is an integral domain, every injective R-module is divisible. Conversely, if a module
is divisible and torsion free it is injective.

Proof . We use the exact sequence

r
0 −→ R −→ R
(R is a domain) for a given element (6= 0) of R. The functor HomR (−, Q) is exact as Q is injective. Then
we get
r
Q = HomR (R, Q) −→ HomR (R, Q) = Q −→ 0
is exact. As r is arbitrary, Q is divisible.
Next, assume M is a torsion-free, divisible module. For an exact sequence

0 −→ A −→ R,

suppose we have an R-module map ϕ : A → M . We need only prove ϕ extends to a map R −→ M . Of

course, this means we need to find m ∈ M , the image of 1 under our extension of ϕ, so that

(∀r ∈ A)(ϕ(r) = rm).

5.3. VARIOUS (CO)HOMOLOGICAL FUNCTORS 353

Now for each fixed r ∈ A, the divisibility of M shows there is an element, m(r) ∈ M , so that

ϕ(r) = r · m(r).

This element of M is uniquely determined by r because M is torsion-free. Now pick s ∈ A, s 6= 0, consider

sr. We have
ϕ(sr) = sϕ(r) = srm(r).
But, sr = rs; so
ϕ(sr) = ϕ(rs) = rϕ(s) = rsm(s).
By torsion freeness, again, we find m(r) = m(s). So, all the elements m(r) are the same, m; and we’re done.

Write F = Frac(R). The field F is a torsion-free divisible, R-module; it is therefore an injective R-module
(in fact, it is the injective hull of R). The R-module, F/R, is an R-module of some importance. For example,
HomR (F/R, M ) = (0) provided M is torsion-free. In terms of F/R we have the

Corollary 5.43 If M is a torsion-free module, then M is injective iff Ext1R (F/R, M ) = (0). In particular,
for torsion-free modules, M , the following are equivalent

(1) M is injective

(2) ExtnR (F/R, M ) = (0) all n > 0

(3) Ext1R (F/R, M ) = (0).

Proof . Everything follows from the implication (3) =⇒ (1). For this, we have the exact sequence

0 −→ R −→ F −→ F/R −→ 0

and so (using HomR (F/R, M ) = (0)) we find

θ
0 −→ HomR (F, M ) −→ M −→ Ext1R (F/R, M )

is exact. The map, θ, takes f to f (1). By (3), θ is an isomorphism. Given m ∈ M and r 6= 0 in R, there is
some f : F → M , with f (1) = m. Let q = f (1/r), then

rq = rf (1/r) = f (1) = m;

so, M is divisible and Proposition 5.42 applies.

1
−→ ( λ R), where we use the Artin ordering on R: λ ≤ µ iff λ | µ.
The field F is easily seen to be lim
λ
Consequently, F is a right limit of projective (indeed, free of rank one) modules. Now tensor commutes with
right limits, therefore so does TorR• (DX). This gives us
 
1
TorRn (F, M ) = lim
−→ TorR
n R, M = (0), if n > 0.
λ
λ

That is, F is a flat R-module. Moreover, we have

Proposition 5.44 If R is an integral domain and M is any R-module, then TorR

1 (F/R, M ) = t(M ), the
torsion submodule of M . The R-modules TorR
p (F/R, M ) vanish if p ≥ 2.
354 CHAPTER 5. HOMOLOGICAL ALGEBRA

Proof . Use the exact sequence

0 −→ R −→ F −→ F/R −→ 0
and tensor with M . We get
0 −→ TorR
1 (F/R, M ) −→ R ⊗R M (= M ) −→ F ⊗R M −→ F/R ⊗R M −→ 0

and, further back along the homology sequence

(0) = TorR R R
p+1 (F, M ) −→ Torp+1 (F/R, M ) −→ Torp (R, M ) = (0)

for all p ≥ 1. Thus, all will be proved when we show

t(M ) = Ker (M −→ F ⊗R M ).
1
Since F ⊗R M = −→ ( λ R ⊗R M ), we see ξ ∈ Ker (M −→ F ⊗R M ) iff there is some λ (6= 0) with
lim
λ
ξ ∈ Ker (M −→ λ1 R ⊗R M ). But, R is a domain, so multiplication by λ is an isomorphism of 1
λR and R.
This gives us the commutative diagram


M / 1 R ⊗R M
λ

mult. by λ
 
M R ⊗R M
and we see immediately that the left vectical arrow is also multiplication by λ. Hence
ξ ∈ Ker (M −→ ( λ1 R) ⊗R M ) when and only when λξ = 0, and we are done.
The name and symbol for TorR
• arose from this proposition.

When R is a P.I.D., the module, F/R, being divisible is injective. Consequently,

Proposition 5.45 If R is a P.I.D., the sequence
0 −→ R −→ F −→ F/R −→ 0
is an injective resolution of R. Hence, ExtpR (M, R) = (0) if p ≥ 2, while
Ext1R (M, R) = Coker (HomR (M, F ) −→ HomR (M, F/R)).
When M is a finitely generated R-module, we find
Ext1R (M, R) = HomR (t(M ), F/R).
Proof . We know the exact sequence is an injective resolution of R and we use it to compute the Ext’s. This
gives all but the last statement. For that, observe that
0 −→ t(M ) −→ M −→ M/t(M ) −→ 0
is split exact because M/t(M ) is free when R is a P.I.D. and M is f.g. Now F is torsion free, so
HomR (M, F ) = F α , α = rank M/t(M )
and
HomR (M, F/R) = HomR (t(M ), F/R) q (F/R)α .
Therefore, Ext1R (M, R) computed as the cokernel has the value claimed above.
For torsion modules, M , the R-module HomR (M, F/R) is usually called the dual of M and its elements
are characters of M . The notation for the dual of M is M D . With this terminology, we obtain
Corollary 5.46 Suppose R is a P.I.D. and M is a f.g. R-module. Then the equivalence classes of extensions
of M by R are in 1–1 correspondence with the characters of the torsion submodule of M .
5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 355

5.4 Spectral Sequences; First Applications

The invariants provided by homological algebra are obtained from the computation of the (co)homology of a
given complex. In general, this is not an easy task—we need all the help we can get. Experience shows that
many complexes come with a natural filtration (for example, the complex of differential forms on a complex
manifold with its Hodge filtration). In this case, if the filtration satisfies a few simple properties, we can go
a long way toward computing (co)homology provided there is a suitable beginning provided for us.
So let C • be a complex (say computing cohomology) and suppose C • is filtered. This means there is a
family of subobjects, {F p C • }p∈Z , of C • such that

· · · ⊇ F p C • ⊇ F p+1 C • ⊇ · · · .
S T
We also assume that p F p C • = C • and p F p C • = (0). Moreover, if d is the coboundary map of the
complex C • (also called differentiation), we assume that
(1) The filtration {F p C • } and d are compatible, which means that d(F p C • ) ⊆ F p C • , for all p.
(2) The filtration {F p C • } is compatible with the grading on C • , i.e.,
a a
F pC • = F p C • ∩ C p+q = C p,q ,
q q

where C p,q = F p C • ∩ C p+q . Then, each F p C • is itself a filtered graded complex as are the F p C • /F p+r C • ,
for all r > 0.

Remarks:
`
(1) We have F p C • = q C p,q , the elements in C p,q have degree p + q.

(2) The C p,q ’s are subobjects of C p+q .

(3) The C p,q ’s filter C p+q , and p is the index of filtration.

Now, C • possesses cohomology; H • (C • ). Also, F p C • possesses cohomology, H • (F p C • ). There is a map

of complexes F p C • ,→ C • , so we have a map H • (F p C • ) −→ H • (C • ), the image is denoted H • (C • )p and
the H • (C • )p ’s filter H • (C • ). So, H • (C • ) is graded and filtered. Thus, we can make

H(C)p,q = H • (C • )p ∩ H p+q (C • ).

There is a graded complex, gr(C • ), induced by F on C • , defined as

gr(C • )n = F n C • /F n+1 C • .
`
So, we have gr(C • ) = n gr(C • )n and it follows that
a
gr(C • ) = (F p C • /F p+1 C • )
p
" ! !#
a a . a
= F p C • ∩ C p+q F p+1 C • ∩ C p+q
p q q
aa
= (F p C • ∩ C p+q )/(F p+1 C • ∩ C p+q )
p q
a
= C p,q /C p+1,q−1 .
p,q
356 CHAPTER 5. HOMOLOGICAL ALGEBRA

So, we get a a
gr(C • ) = C p,q /C p+1,q−1 = gr(C)p,q ,
p,q p,q
p,q p,q p+1,q−1 • •
with gr(C) =C /C . Similarly, H (gr(C )) is also bigraded; we have
a
H • (gr(C • )) = H(gr(C))p,q ,
p,q

where H(gr(C))p,q = H p+q (F p C • /F p+1 C • ).

Finally, we also have the graded pieces of H p+q (C • ) in its filtration,
gr(H(C))p,q = H(C)p,q /H(C)p+1,q−1 = H p+q (C • ) ∩ H • (C • )p /H p+q (C • ) ∩ H • (C • )p+1 .
`
As a naive example of a filtration, we have F p C • = n≥p C n.
The rest of this section is replete with indices—a veritable orgy of indices. The definitions to remember
are four : C p,q , gr(C)p,q , H(gr(C))p,q and gr(H(C))p,q . Now C • is filtered and it leads to the graded object
gr(C • ). One always considers gr(C • ) as a “simpler” object than C • . Here’s an example to keep in mind
which demonstrates this idea of “simpler”. Let C be the ring of power series in one variable, x, over some
field, k. Convergence is irrelevant here, just use formal power series. Let F p C be the collection of power
series beginning with terms involving xp+1 or higher. We feel that in F p C the term of a series involving xp+1
is the “dominating term”, but there are all the rest of the terms. How to get rid of them? Simply pass to
F p C/F p+1 C, in this object only the term involving xp+1 survives. So gr(C) is the coproduct of the simplest
objects: the single terms ap+1 xp+1 . It is manifestly simpler than C. Ideally, we would like to compute the
cohomology, H • (C • ), of C • . However, experience shows that this is usually not feasible, but instead we can
begin by computing H • (gr(C • )) because gr(C • ) is simpler than C • . Then, a spectral sequence is just the
passage from H • (gr(C • )) to gr(H • (C • )); this is not quite H • (C • ) but is usually good enough.
The following assumption makes life easier in dealing with the convergence of spectral sequences: A
filtration is regular iff for every n ≥ 0, there is some µ(n) ≥ 0, so that for all p > µ(n), we have
F p C • ∩ C n = (0).
Definition 5.10 A cohomological spectral sequence is a quintuple,
E = hErp,q , dp,q p,q
r , αr , E, β
p,q
i,
where
(1) Erp,q is some object in Ob(A), with p, q ≥ 0 and 2 ≤ r ≤ ∞ (the subscript r is called the level ).
(2) dp,q p,q
r : Er → Erp+r,q−r+1 is a morphism such that dp,q p−r,q+r−1
r ◦ dr = 0, for all p, q ≥ 0 and all r < ∞.
p,q
(3) αrp,q : Ker dp,q p−r,q+r−1
r /Im dr → Er+1 is an isomorphism, for all p, q, all r < ∞.
(4) E is a graded, filtered object from A, so that each F p E is graded by the E p,q = F p E ∩ E p+q .
(5) β p,q : E∞
p,q
→ gr(E)p,q is an isomorphism, for all p, q (where gr(E)p,q = E p,q /E p+1,q−1 ).

Remarks:
(1) The whole definition is written in the compact form
E2p,q =⇒
p
E

and E is called the end of the spectral sequence. The index p is called the filtration index , p + q is
called the total or grading index and q the complementary index .
5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 357

(2) If r > q + 1, then Im dp,q

r = (0) and if r > p, then Im dr
p−r,q+r−1
= (0). So, if r > max{p, q + 1}, then
p,q p,q p,q
(3) implies that Er = Er+1 , i.e., the sequence of Er stabilizes for r >> 0.
(3) In general, when Erp,q stabilizes, Erp,q 6= E∞
p,q
. Further assumptions must be made to get Erp,q = E∞
p,q

for r >> 0.
(4) It is customary to define spectral sequence beginning from r = 2, even though the terms Eip,q are often
defined and meaningful for r = 1, and even for r = 0. However, in the case of double complexes, the
natural starting point is indeed r = 2, as pointed out in Cartan and Eilenberg [9] (Chapter XV, page
332).
(5) One can instead make the definition of a homological spectral sequence by passing to the “third quad-
rant” (p ≤ 0 and q ≤ 0) and changing arrows around after lowering indices in the usual way, viz :
H −n becomes Hn . Further, one can make 2nd or 4th quadrant spectral sequences or those creeping
beyond the quadrant boundaries. All this will be left to the reader—the cohomological case will be
quite enough for us.

Spectral sequences can be introduced in many ways. The one chosen here leads immediately into appli-
cations involving double complexes but is weaker if one passes to triangulated and derived categories. No
mastery is possible except by learning the various methods together with their strengths and weaknesses. In
the existence proof given below there are many complicated diagrams and indices. I urge you to read as far
as the definition of Zrp,q and Brp,q (one–half page) and skip the rest of the proof on a first reading.

Theorem 5.47 Say C • is a filtered right complex whose filtration is compatible with its grading and differ-
entiation. Then, H • (C • ) possesses a filtration (and is graded) and there exists a spectral sequence

E2p,q =⇒
p
H • (C • ),

in which E2p,q is the cohomology of H • (gr(C • ))—so that E1p,q = H(gr(C))p,q = H p+q (F p C • /F p+1 C • )). If
p,q
the filtration is regular, the objects E∞ (= gr(H • (C • ))p,q = H(C)p,q /H(C)p+1,q−1 = composition factors in
p+q •
the filtration of H (C )) are exactly the Erp,q when r >> 0.

In the course of the proof of Theorem 5.47, we shall make heavy use of the following simple lemma whose
proof will be left as an exercise:

Lemma 5.48 (Lemma (L)) Let

B
}> CC
} }} ϕ CCCψ
}} CC
}}  C!
A 0 / A η / A00
0
ϕ

be a commutative diagram with exact bottom row. Then, η induces an isomorphism

Im ϕ/Im ϕ0 −
g → Im ψ.

Proof of Theorem 5.47. First, we need to make Zrp,q and Brp,q and set Erp,q = Zrp,q /Brp,q .
Consider the exact sequence (we will drop the notation C • in favor of C for clarity)

0 −→ F p C −→ F p−r+1 C −→ F p−r+1 C/F p C −→ 0.

Upon applying cohomology, we obtain

δ∗
· · · −→ H p+q−1 (F p−r+1 C) −→ H p+q−1 (F p−r+1 C/F p C) −→ H p+q (F p C) −→ · · ·
358 CHAPTER 5. HOMOLOGICAL ALGEBRA

There is also the natural map H p+q (F p C) −→ H p+q (F p C/F p+1 C) induced by the projection
F p C −→ F p C/F p+1 C. Moreover, we have the projection F p C/F p+r C −→ F p C/F p+1 C, which induces a
map on cohomology
H p+q (F p C/F p+r C) −→ H p+q (F p C/F p+1 C).
Set

Zrp,q = Im (H p+q (F p C/F p+r C) −→ H p+q (F p C/F p+1 C))

Brp,q = Im (H p+q−1 (F p−r+1 C/F p C) −→ H p+q (F p C/F p+1 C)),

the latter map being the composition of δ ∗ and the projection (where r ≥ 1).
The inclusion F p−r+1 C ⊆ F p−r C yields a map F p−r+1 C/F p C −→ F p−r C/F p C; hence we obtain the
p,q
inclusion relation Brp,q ⊆ Br+1 . In a similar way, the projection F p C/F p+r+1 C −→ F p C/F p+r C yields the
p,q p,q p,q p,q
inclusion Zr+1 ⊆ Zr . When r = ∞, the coboundary map yields the inclusion B∞ ⊆ Z∞ (remember,
∞
F C = (0)). Consequently, we can write
p,q p,q
· · · ⊆ Brp,q ⊆ Br+1 p,q
⊆ · · · ⊆ B∞ p,q
⊆ Z∞ ⊆ · · · ⊆ Zr+1 ⊆ Zrp,q ⊆ · · · .

Set
Erp,q = Zrp,q /Brp,q , where 1 ≤ r ≤ ∞, and En = H n (C).
`
Then, E = n En = H(C), filtered by the H(C)p ’s, as explained earlier. When r = 1, B1p,q = (0) and

Z1p,q = H p+q (F p C/F p+1 C);

We obtain E1p,q = H p+q (F p C/F p+1 C) = H(gr(C))p,q . On the other hand, when r = ∞ (remember,
F −∞ C = C), we get
p,q
Z∞ = Im (H p+q (F p C) −→ H p+q (F p C/F p+1 C))
p,q
B∞ = Im (H p+q−1 (C/F p C) −→ H p+q (F p C/F p+1 C)).

Now the exact sequence 0 −→ F p C/F p+1 C −→ C/F p+1 C −→ C/F p C −→ 0 yields the cohomology sequence
δ∗
· · · −→ H p+q−1 (C/F p C) −→ H p+q (F p C/F p+1 C) −→ H p+q (C/F p+1 C) −→ · · ·

and the exact sequence 0 −→ F p C −→ C −→ C/F p C −→ 0 gives rise to the connecting homomorphism
H p+q−1 (C/F p C) −→ H p+q (F p C). Consequently, we obtain the commutative diagram (with exact bottom
row)
H p+q (F p C)
kk5 TTTT
kk kkk TTTT
kkkk TTTT
kkkk  TTT)
H p+q−1 p
(C/F C) / H p+q p
(F C/F p+1
C) / H p+q (C/F p+1 C)
and Lemma (L) yields an isomorphism

ξ p,q : E∞
p,q p,q
= Z∞ p,q
/B∞ −→ Im (H p+q (F p C) −→ H p+q (C/F p+1 C)).

But another application of Lemma (L) to the diagram

H p+q (F p C)
n6 QQQ
nnnnn QQQ
QQQ
nnn QQQ
nnn  Q(
H p+q (F p+1 C) / H p+q (C) / H p+q (C/F p+1 C)
5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 359

gives us the isomorphism

η p,q : gr(H(C))p,q −→ Im (H p+q (F p C) −→ H p+q (C/F p+1 C)).

Thus, (η p,q )−1 ◦ ξ p,q is the isomorphism β p,q required by part (5) of Definition 5.10.
Only two things remain to be proven to complete the proof of Theorem 5.47. They are the verification
p,q
of (2) and (3) of Definition 5.10, and the observation that E∞ , as defined above, is the common value of
p,q
the Er for r >> 0. The verification of (2) and (3) depends upon Lemma (L). Specifically, we have the two
commutative diagrams (with obvious origins)

H p+q (F p C/F p+r C)

j4 UUUU
jjj jjjj UUUU θ
UUUU
jj jjj UUUU
jjj  U*
H p+q p
(F C/F p+r+1
C) / H p+q p
(F C/F p+1
C) ∗ / H p+q+1 (F p+1 C/F p+r+1 C)
δ

and

H p+q (F p C/F p+r C)

hh4 VVVV
hh hhhhh VVVV θ
VVVV
hhhhh VVVV
VVV+
hhh 
H p+q
(F p+1
C/F p+r
C) ∗ H / p+q+1
(F p+r
C/F p+r+1
C) / H p+q+1
(F p+1 C/F p+r+1 C).
δ

Here, the map θ is the composition

H p+q (F p C/F p+r C) −→ H p+q+1 (F p+r C) −→ H p+q+1 (F p+1 C/F p+r+1 C).

Now, Lemma (L) yields the following facts:

p,q
Zrp,q /Zr+1 −
g→ Im θ,
p+r,q−r+1
Br+1 /Brp+r,q−r+1 −
g→ Im θ,

that is,
p,q p+r,q−r+1
δrp,q : Zrp,q /Zr+1 −
g→ Br+1 /Brp+r,q−r+1 .
As Brp,q ⊆ Zsp,q for every r and s, there is a surjection
p,q
πrp,q : Erp,q −→ Zrp,q /Zr+1
p,q
with kernel Zr+1 /Brp,q ; and there exists an injection

p+r,q−r+1 p+r,q−r+1
σr+1 : Br+1 /Brp+r,q−r+1 −→ Erp+r,q−r+1 .

p+r,q−r+1
The composition σr+1 ◦ δrp,q ◦ πrp,q is the map dp,qr from Erp,q to Erp+r,q−r+1 required by (2). Observe
that,
p,q p,q
Im dp−r,q+r−1
r = Br+1 /Brp,q ⊆ Zr+1 /Brp,q = Ker dp,q
r ;

hence
p−r,q+r−1 ∼ p,q p,q p,q
H(Erp,q ) = Ker dp,q
r /Im dr = Zr+1 /Br+1 = Er+1 ,
as required by (3).
 360 CHAPTER 5. HOMOLOGICAL ALGEBRA

p,q
To prove that E∞ as defined above is the common value of Erp,q for large enough r, we must make use
of the regularity of our filtration. Consider then the commutative diagram

H p+q (F p C/F p+r C)

5 SSS
lll SSS λ
ll lll SSS
SSS
lll SS)
ll 
H p+q (F p C) / H p+q (F p C/F p+1 C) / H p+q+1 (F p+1 C)

where λ is the composition

δ∗
H p+q (F p C/F p+r C) −→ H p+q+1 (F p+r C) −→ H p+q+1 (F p+1 C).

By Lemma (L), we have Zrp,q /Z∞

p,q
−
g→ Im λ. However, if r > µ(p + q + 1) − p, then δ ∗ is the zero map. This
shows Im λ = (0); hence, we have proven

Zrp,q = Z∞
p,q
for r > µ(p + q + 1) − p.
p,q
By our assumptions, the filtration begins with C = F 0 C, therefore if r > p we find Brp,q = B∞ . Hence, for

r > max{p, µ(p + q + 1) − p}

the Erp,q equal E∞

p,q
.

p,q
Remark: Even if our filtration does not start at 0, we can still understand E∞ from the Erp,q when the
filtration is regular. To see this, note that since cohomology commutes with right limits, we have

lim Brp,q = B∞
p,q
,
−→
r
S
and this implies r Brp,q = B∞
p,q
. Hence, we obtain maps

Erp,q = Zrp,q /Brp,q −→ Zsp,q /Bsp,q = Esp,q

for s ≥ r > µ(p + q + 1) − p, and these maps are surjective. (The maps are in fact induced by the drp−r,q+r−1 ’s
because of the equality
p,q p,q
Erp,q /Im dp−r,q+r−1
r = (Zrp,q /Brp,q )/(Br+1 /Brp,q ) = Er+1

for r > µ(p + q + 1) − p.) Obviously, the right limit of the surjective mapping family
p,q
Erp,q −→ Er+1 −→ · · · −→ Esp,q −→ · · ·
p,q
S
is the group Z∞ /( Brp,q ) = E∞p,q p,q
; so, each element of E∞ arises from Erp,q if r >> 0 (for fixed p, q).
Regularity is therefore still an important condition for spectral sequences that are first and second quadrant
or first and fourth quadrant.

p,q
T
 It is not true in general that Z∞ = r Zrp,q or that limr Zrp,q = Z∞
p,q
. In the first case, we have a weakly
←−
convergent spectral sequence. In the second case, we have a strongly convergent spectral sequence.

Remark: Let us keep up the convention of the above proof in which the complex C appears without the
“dot”. Then, by (5) of our theorem we find
p,q
E∞ = (H p+q (C) ∩ H(C)p )/(H p+q (C) ∩ H(C)p+1 ),
5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 361

p,q p,n−p
so that, for p + q = n, the E∞ = E∞ are the composition factors in the filtration

H n (C) ⊇ H n (C)1 ⊇ H n (C)2 ⊇ · · · ⊇ H n (C)ν ⊇ · · · .

To understand a spectral sequence, it is important to have in mind a pictorial representation of it in its

entirety. We are to imagine an “apartment house”; on the rth floor the apartments are labelled Erp,q and a
plan of the rth floor is exactly the points of the pq-plane. The roof of the apartment building is the ∞-floor.
In addition, there is the map dp,qr on the rth floor; it goes “over r and down r − 1”. Hence, a picture of the
th
r floor is shown in Figure 5.1:

Erp,q

dp,q
r

Erp+r,q−r+1

Figure 5.1: The Erp,q terms of a spectral sequence (“rth floor”)

The entire edifice is depicted in Figure 5.2. One passes vertically directly to the apartment above by
forming cohomology (with respect to dr ); so, one gets to the roof by repeated formings of cohomology at
each higher level.
0,n 1,n−1 n,0
Once on the roof—at the ∞-level—the points on the line p+q = n, i.e., the groups E∞ , E∞ , . . . , E∞ ,
n
are the composition factors for the filtration of H (C):

H n (C) ⊇ H n (C)1 ⊇ H n (C)2 ⊇ · · · ⊇ H n (C)n ⊇ (0);

See Figure 5.3.

To draw further conclusions in situations that occur in practice, we need three technical lemmas. Their
proofs should be skipped on a first reading and they are only used to isolate and formalize conditions
frequently met in the spectral sequences of applications. We’ll label them Lemmas A, B, C as their conclusions
are only used to get useful theorems on the sequences.

First, observe that if for some r, there are integers n and p1 > p0 so that Erν,n−ν = (0) whenever
ν 6= p0 , ν 6= p1 , then certainly Esν,n−ν = (0) for every s with r ≤ s ≤ ∞. If the filtration is regular, then
p0 ,n−p0 p1 ,n−p1
E∞ and E∞ are the only possible non-zero composition factors for H n (C) and therefore we obtain
the exact sequence
p1 ,n−p1
0 −→ E∞ −→ H n (C) −→ E∞ p0 ,n−p0
−→ 0. (†)
362 CHAPTER 5. HOMOLOGICAL ALGEBRA

E
∞
(ro
of
)
E p,
∞ q

E
E p, r
r q
ds E
s

E
4
E
3
E
2

Figure 5.2: The entire spectral sequence (regular filtration)

p + q = n; points = composition factors in H n (C)

p,q
Figure 5.3: The E∞ terms of a spectral sequence (“roof level”)

Lemma 5.49 (Lemma A) Let E2p,q =⇒ H • (C) be a spectral sequence with a regular filtration. Assume there
are integers r; p1 > p0 ; n so that

 u + v = n, u 6= p0 , p1
Eru,v = (0) for u + v = n + 1, u ≥ p1 + r

u + v = n − 1, u ≤ p0 − r.

Then, there is an exact sequence

Erp1 ,n−p1 −→ H n (C) −→ Erp0 ,n−p0 . (A)

5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 363

Proof . The remarks above and the first hypothesis yield sequence (†). In the proof of Theorem 5.47, we saw
that
Im dtp0 −t,n−p0 +t−1 = Bt+1
p0 ,n−p0
/Btp0 ,n−p0 .
We take ∞ > t ≥ r, let u = p0 − t and v = n − p0 + t − 1. Using these u and v and the third hypothesis, we
deduce Btp0 ,n−p0 is constant for t ≥ r. Therefore, B∞
p0 ,n−p0
= Brp0 ,n−p0 . This gives an injection
p0 ,n−p0 p0 ,n−p0
E∞ ,→ Er .
Next, with u = p1 + t; v = n − p1 − t + 1 and ∞ > t ≥ r, the second hypothesis shows that
Ker dpt 1 +t,n−p1 −t+1 = (0) and the latter is Zt+1
p1 +t,n−p1 −t+1
/Btp1 +t,n−p1 −t+1 . But,

Br•,• ⊆ Bt•,• ⊆ B∞
•,• •,•
⊆ Z∞ •,•
⊆ Zt+1 ,

and so we get
p1 +t,n−p1 −t+1
Bt+1 = Btp1 +t,n−p1 −t+1 , ∞ ≥ t ≥ r.
However, from the proof of Theorem 5.47, we find
p1 +t,n−p1 −t+1
Ztp1 ,n−p1 /Zt+1
p1 ,n−p1
' Bt+1 /Btp1 +t,n−p1 −t+1 ;

and therefore Ztp1 ,n−p1 is constant for ∞ > t ≥ r. By the regularity of the filtration, we find
Zrp1 ,n−p1 = Z∞p1 ,n−p1 p1 ,n−p1
. This gives a surjection Erp1 ,n−p1 −→ E∞ , and if we combine (†), our injection for
p0 , n − p0 and the surjection for p1 , n − p1 we get sequence (A).

Lemma 5.50 (Lemma B) Suppose that E2p,q =⇒ H • (C) is a spectral sequence with a regular filtration.
Assume that there are integers s ≥ r; p, n so that

 u + v = n − 1, u ≤ p − r
Eru,v = (0) for u + v = n, u 6= p and u ≤ p + s − r

u + v = n + 1, p + r ≤ u and u 6= p + s.

Then, there is an exact sequence

H n (C) −→ Erp,n−p −→ Erp+s,(n+1)−(p+s) . (B)

Proof . We apply dp,n−p

r to Erp,n−p and land in Erp+r,n−p−r+1 which is (0) by hypothesis three. Also,
p−r,n−p+r−1 p,n−p
Er is (0) by the first hypothesis, so the image of drp−r,n−p+r−1 is (0). This shows Erp,n−p = Er+1 .
Repeat, but with dr+1 ; as long as r + 1 < s we can continue using hypotheses one and three. Thus we obtain
Erp,n−p = Esp,n−p . Now apply dp,n−pt to Etp,n−p where t ≥ s + 1. Hypothesis three shows our map is zero
p−t,n−p+t−1
and similarly the map dt is zero by hypothesis one. So, for all t, with ∞ > t ≥ s + 1, we get
Etp,n−p = Et+1
p,n−t p,n−p
. As the filtration is regular, we obtain Es+1 p,n−p
= E∞ .
Next, by hypothesis two with u = p + (s − r) (provided s > r, otherwise there is nothing to prove), we
p+s−r,n−(p+s−r)
see that Im dr is (0). Thus,
p+s,(n+1)−(p+s)
Br+1 = Brp+s,(n+1)−(p+s) .

Should s > r + 1, we continue because

p+s−(r+1),n−(p+s−(r+1))
(0) = Im dr+1 .

This gives
p+s,(n+1)−(p+s) p+s,(n+1)−(p+s)
Br+2 = Br+1 .
Hence, we get
Bsp+s,(n+1)−(p+s) = Brp+s,(n+1)−(p+s)
364 CHAPTER 5. HOMOLOGICAL ALGEBRA

by repetition. Of course, this gives the inclusion

Esp+s,(n+1)−(p+s) ⊆ Erp+s,(n+1)−(p+s) .

p−s,(n−1)−(p−s) p−s,(n−1)−(p−s)
Lastly, by hypothesis one, Er = (0); so, Et = (0) for every t ≥ r. Take t = s,
p−s,(n−1)−(p−s)
then ds vanishes, and in the usual way we get
p,n−p
Bs+1 = Bsp,n−p .

But then, we obtain an inclusion

p,n−p
Es+1 ,→ Esp,n−p .
p,n−p p,n−p
However, the kernel of dp,n−p
s is Zs+1 /Bsp,n−p = Es+1 ; therefore we get the exact sequence

p,n−p dp,n−p
0 −→ Es+1 −→ Esp,n−p −→
s
Esp+s,(n+1)−(p+s) .

And now we have a surjection H n (C) −→ E∞ p,n−p

because E∞u,n−u
= (0) when u ≤ p + s − r (r 6= p) by
hypothesis two. If we put all this together, we get sequence (B).
In a similar manner (see the exercises) one proves

Lemma 5.51 (Lemma C) If E2p,q =⇒ H • (C) is a spectral sequence with a regular filtration and if there
exist integers s ≥ r; p, n so that

 u + v = n + 1, u ≥ p + r
Eru,v = (0) for u + v = n, p + r − s ≤ u 6= p

u + v = n − 1, p − s 6= u ≤ p − r,

then, there is an exact sequence

Erp−s,(n−1)−(p−s) −→ Erp,n−p −→ H n (C). (C)

Although Lemmas A, B, C are (dull and) technical, they do emphasize one important point: For any
level r, if Erp,q lies on the line p + q = n, then dr takes it to a group on the line p + q = n + 1 and it receives
a dr from a group on the line p + q = n − 1. From this we obtain immediately

Corollary 5.52 (Corollary D) Say E2p,q =⇒ H • (C) is a regularly filtered spectral sequence and there are
integers r, n so that 
p,q p+q =n−1
Er = (0) for
p + q = n + 1.
Then, Erp,n−p = E∞
p,n−p
and the Erp,n−p are the composition factors for H n (C) in its filtration.

Now we wish to apply Lemmas A, B, C and we begin with the simplest case—a case for which we do not
need these Lemmas. A spectral sequence E2p,q =⇒ H • (C) degenerates at (level) r when and only when for
each n there is a q(n) so that
Ern−q,q = (0) if q 6= q(n).
Of course then Esn−q,q = (0) when q 6= q(n) for all s ≥ r; so that, in the regular case, we have E∞ n−q,q
= (0)
if q 6= q(n). If we have q(n + 1) > q(n) − (r − 1) for all n (e.g., if q(n) is constant), then Ern−q,q = E∞n−q,q

for every n and q and we deduce that

H n (C) = E∞
n−q(n),q(n)
= Ern−q(n),q(n)

for all n. This proves

5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 365

Proposition 5.53 When the filtration of C is regular and the spectral sequence
E2p,q =⇒ H • (C)
n−q(n),q(n)
degenerates at r, then H n (C) = E∞ . If in addition, q(n + 1) > q(n) − (r − 1) for all n, then
H n (C) ∼
= Ern−q(n),q(n)
for every n.
Theorem 5.54 (Zipper Sequence) Suppose E2p,q =⇒ H • (C) is a regularly convergent spectral sequence and
there exist integers p0 , p1 , r with p1 − p0 ≥ r ≥ 1 so that Eru,v = (0) for all u 6= p0 or p1 . Then we have the
exact zipper sequence
· · · −→ Erp1 ,n−p1 −→ H n (C) −→ Erp0 ,n−p0 −→ Erp1 ,n+1−p1 −→ H n+1 (C) −→ · · ·
Dually, if there are integers q0 , q1 , r with q1 − q0 ≥ r − 1 ≥ 1 so that Eru,v = (0) for v 6= q0 or q1 , then the
zipper sequence is
· · · −→ Ern−q0 ,q0 −→ H n (C) −→ Ern−q1 ,q1 −→ Ern+1−q0 ,q0 −→ H n+1 (C) −→ · · ·
Proof . Write s = p1 − p0 ≥ r and apply Lemmas A, B and C (check the hypotheses using u + v = n). By
splicing the exact sequences of those lemmas, we obtain the zipper sequence. Dually, write s = 1+q1 −q0 ≥ r,
set p0 = n − q1 and p1 = n − q0 . Then Lemmas A, B and C again apply and their exact sequences splice to
give the zipper sequence.
The name “zipper sequence” comes from the following picture. In it, the dark arrows are the maps
Erp0 ,n−p0 −→ Erp1 ,n+1−p1 and the dotted arrows are the compositions Erp1 ,n+1−p1 −→ H n+1 −→ Erp0 ,n+1−p0
(one is to imagine these arrows passing through the H n+1 somewhere behind the plane of the page). As you
see, the arrows zip together the vertical lines p = p0 and p = p1 .

(Level r in the spectral sequence)

p
p = p0 p = p1

Figure 5.4: Zipper Sequence

366 CHAPTER 5. HOMOLOGICAL ALGEBRA

Theorem 5.55 (Edge Sequence) Suppose that E2p,q =⇒ H • (C) is a regularly convergent spectral sequence
and assume there is an integer n ≥ 1 so that E2p,q = (0) for every q with 0 < q < n and all p (no hypothesis
if n = 1). Then E2r,0 ∼
= H r (C) for r = 0, 1, 2, . . . , n − 1 and

0 −→ E2n,0 −→ H n (C) −→ E20,n −→ E2n+1,0 −→ H n+1 (C)

is exact (edge sequence). In particular, with no hypotheses on the vanishing of E2p,q , we have the exact
sequence
d0,1
0 −→ E21,0 −→ H 1 (C) −→ E20,1 −→
2
E22,0 −→ H 2 (C).

Proof . Since we have a cohomological (first quadrant) spectral sequence all the differentials dr,0 l vanish for
all l and if l ≥ n no differential dp,q
l hits El
r,0
if p ≥ 0 and r ≤ n − 1. All the differentials d p,q
l are 0 if q < n
and so we find E2r,0 ∼= E∞ r,0
for 0 ≤ r ≤ n − 1. But, only one non-zero term E∞ p,q
exists on the line r = p + q
for r < n by our hypothesis on the vanishing; so, indeed E2r,0 ∼ = E∞r,0
= H r (C) when 0 ≤ r ≤ n − 1.
For Eln,0 , since dp,n−p−1
n−p
p,n−p−1
: En−p n,0
→ En−p , and since p ≥ 0 implies q ≤ n − 1, we see that no non-zero
n,0 n,0 ∼
differential hits El for any l. Thus, E2 = E∞ n,0
and we get the injection E2n,0 −→ H n (C). Apply Lemma
A with p0 = 0, p1 = n, r = 2 to find the sequence

0 −→ E2n,0 −→ H n (C) −→ E20,n . (∗)

Next, in Lemma B, take r = 2, s = n + 1 ≥ 2, and p = 0. Sequence (B) splices to (∗) to yield

0 −→ E2n,0 −→ H n (C) −→ E20,n −→ E2n+1,0 . (∗∗)

And, lastly, use Lemma C with r = 2, s = n + 1 ≥ 2, the n of Lemma C to be our n + 1 = s and p = n + 1.

Upon splicing Lemma C onto (∗∗) we find the edge sequence

0 −→ E2n,0 −→ H n (C) −→ E20,n −→ E2n+1,0 −→ H n+1 (C).

Obviously, the edge sequence gets its name from the fact that the E2p,q which appear in it lie on the
edge of the quadrant in the picture of E2 as points (of the first quadrant) in the pq-plane. Equally obvious
is the notion of a morphism of spectral sequences. Whenever C and C e are graded, filtered complexes and
g: C → C e is a morphism of such complexes, we find an induced morphism

e p,q =⇒ H • (C)
ss(g) : E2p,q =⇒ H • (C) 7→ E e
2

of spectral sequences.

Theorem 5.56 Suppose C and C e are graded filtered complexes and write E •,• (C) and E •,• (C)
e for their
• •,• •,• e
associated spectral sequences. Assume both filtrations are regular and g : E (C) → E (C) is a spectral
sequence morphism. If, for some r ≥ 2, the level r map gr• : Er•,• → E er•,• is an isomorphism, then for every
•
s ≥ r the level s map, gs , is also an isomorphism (also for s = ∞) and we have an induced isomorphism on
the graded cohomology
grH(g • ) : grH • (C) −
g e
→ grH • (C).
p,q
Proof . The proof of this is obvious because by regularity E∞ = Esp,q for s >> 0. But for H n , its graded
p,n−p
pieces are the E∞ , and p ≥ 0. Thus, p ≤ n and q ≤ n; so, our choice s = s(n) >> 0 will do to get

Esp,q = E∞
p,q
(all p, q with p + q = n).
`
These groups are exactly the graded pieces of H n as we’ve remarked and p,q
p+q=n gs is our isomorphism.
5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 367

Our technical results on spectral sequences are over, now we actually need some
` spectral sequences to use
them on. Big sources of spectral sequences are double complexes. So, let C = p,q C p,q be a doubly-graded
complex (we assume that p, q ≥ 0). We have two differentiations:
dp,q
I : C
p,q
−→ C p+1,q , (horizontal)
dp,q
II : C p,q
−→ C p,q+1
(vertical)

such that
dI ◦ dI = dII ◦ dII = 0.
We will require
dp+1,q
II ◦ dp,q p,q+1
I + dI ◦ dp,q
II = 0, for all p, q.

Then we get the (singly graded) total complex

!
a a
p,q
C= C
n p+q=n

with total differential dT = dI + dII . We immediately check that dT ◦ dT = 0. There are two filtrations
a a
FIp C = C r,q and FIIq C = C p,s .
r≥p,q p,s≥q

Both have every compatibility necessary and give filtrations on the total complex and are regular. Therefore,
we find two spectral sequences

Ip,q
2 =⇒ H • (C)
p
and IIp,q
2 =⇒ H • (C).
q

Observe that
a
grI (C) = grpI (C)
a p
= (FI C/FIp+1 C)
!
a a
p,q
= C
p q

and EIp,q = H p,q (grpI (C)), which is just HII

p,q
(C). Now, we need to compute dp,q
1 in spectral sequence (I). It
is induced by the connecting homomorphism arising from the short exact sequence

0 −→ FIp+1 C/FIp+2 C −→ FIp C/FIp+2 C −→ FIp C/FIp+1 C −→ 0.

p,q
Pick ξ ∈ HII (C), represented by a cocycle with respect to dII in C p+q , call it x. The connecting homomor-
phism (= d1 ) is given by “dT x”. But, dT x = dI x + dII x = dI x, as dII x = 0. Therefore, d1 is exactly the map
p,q
induced on HII (C) by dI . It follows that
Ip,q p,q p,q p q
2 = Z2 /B2 = HI (HII (C)).

We have therefore proved

`
Theorem 5.57 Given a double complex C = p,q C p,q , we have two regular spectral sequences converging
to the cohomology of the associated total complex:

HIp (HII
q
(C)) =⇒
p
H • (C)

and
q
HII (HIp (C)) =⇒
q
H • (C).
368 CHAPTER 5. HOMOLOGICAL ALGEBRA

It still is not apparent where we’ll find an ample supply of double complexes so as to use the above
theorem. A very common source appears as the answer to the following
Problem. Given two left-exact functors F : A → B and G : B → C between abelian categories (with
enough injectives, etc.), we have GF : A → C (left-exact); how can we compute Rn (GF ) if we know Rp F
and Rq G?
In order to answer this question, we need to introduce special kinds of injective resolutions of complexes.

Definition 5.11 A Cartan–Eilenberg injective resolution of a complex, C, (with C k = (0) if k < 0) is a

resolution
0 −→ C • −→ Q• 0 −→ Q• 1 −→ Q• 2 −→ · · · ,
`
in which each Q• j = i Qi,j is a complex (differential dij ) and every Qi,j injective and so that if we write
Z i,j = Ker di,j ; B i,j = Im di−1,j and H i,j = Z i,j /B i,j , we have the injective resolutions

(1) 0 / Ci / Qi,0 / Qi,1 / ···

(2) 0 / Z i (C) / Z i,0 / Z i,1 / ···

(3) 0 / B i (C) / B i,0 / B i,1 / ···

(4) 0 / H i (C) / H i,0 / H i,1 / ··· .

The way to remember this complicated definition is through the following diagram:

O O O

0 / C i+1 / Qi+1,0 / Qi+1,1 / ···

O O O
δi di,0 di,1

0 / Ci / Qi,0 / Qi,1 / ···

O O O

0 / Zi / Z i,0 / Z i,1 / ···

O O O

0 0 0

Proposition 5.58 Every complex, C, has a Cartan–Eilenberg resolution, 0 −→ C −→ Q• , where the {Qi,j }
form a double complex. Here, we have suppressed the grading indices of C and the Qj .

Proof . We begin with injective resolutions 0 −→ B 0 (C) −→ B 0,• ; 0 −→ B 1 (C) −→ B 1,• and
0 −→ H 0 (C) −→ H 0,• of B 0 (C); B 1 (C); H 0 (C). Now, we have exact sequences

0 −→ B 0 (C) −→ Z 0 (C) −→ H 0 (C) −→ 0

and
δ0
0 −→ Z 0 (C) −→ C 0 −→ B 1 (C) −→ 0;
so, by Proposition 5.1, we get injective resolutions 0 −→ Z 0 (C) −→ Z 0,• and 0 −→ C 0 −→ Q0,• , so that

0 −→ B 0,• −→ Z 0,• −→ H 0,• −→ 0

5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 369

and
0 −→ Z 0,• −→ Q0,• −→ B 1,• −→ 0
are exact.
For the induction step, assume that the complexes B i−1,• , Z i−1,• , H i−1,• , Qi−1,• and B i,• are determined
and satisfy the required exactness properties (i ≥ 1). Pick any injective resolution H i,• of H i (C), then using
the exact sequence
0 −→ B i (C) −→ Z i (C) −→ H i (C) −→ 0
and Proposition 5.1, we get an injective resolution 0 −→ Z i (C) −→ Z i,• so that

0 −→ B i,• −→ Z i,• −→ H i,• −→ 0

is exact. Next, pick an injective resolution, 0 −→ B i+1 (C) −→ B i+1,• , of B i+1 (C) and use the exact
sequence
δi
0 −→ Z i (C) −→ C i −→ B i+1 (C) −→ 0
and Proposition 5.1 to get an injective resolution 0 −→ C i −→ Qi,• so that

0 −→ Z i,• −→ Qi,• −→ B i+1,• −→ 0

is exact. The differential di,j i,j

II of the double complex {Q } is the composition

Qi,j −→ B i+1,j −→ Z i+1,j −→ Qi+1,j

and the differential di,j

I is given by
di,j i i,j
I = (−1)  ,

where, i,• is the differential of Qi,• . The reader should check that {Qi,j } is indeed a Cartan–Eilenberg
resolution and a double complex (DX).

Note that, due to the exigencies of notation (we resolved our complex C • horizontally) the usual conven-
tions of horizontal and vertical were interchanged in the proof of Proposition 5.58 at least as far as Cartesian
coordinate notation is concerned. This will be rectified during the proof of the next theorem, which is the
result about spectral sequences having the greatest number of obvious applications and forms the solution
to the problem posed before.

Theorem 5.1 (Grothendieck) Let F : A → B and G : B → C be two left-exact functors between abelian
categories (with enough injectives, etc.) and suppose that F (Q) is G-acyclic whenever Q is injective, which
means that Rp G(F Q) = (0), if p > 0. Then, we have the spectral sequence of composed functors

Rq G((Rp F )(A)) =⇒
q
(R• (GF ))(A).

Proof . Pick some object A ∈ A and resolve it by injectives to obtain the resolution 0 −→ A −→ Q• (A):

0 −→ A −→ Q0 −→ Q1 −→ Q2 −→ · · · .

If we apply GF to Q• (A) and compute cohomology, we get Rn (GF )(A). If we just apply F to Q• (A), we
get the complex:
F (Q0 ) −→ F (Q1 ) −→ F (Q2 ) −→ · · · , (F Q• (A))
whose cohomology is Rq F (A).
370 CHAPTER 5. HOMOLOGICAL ALGEBRA

Now resolve the complex F Q• (A) in the vertical direction by a Cartan-Eilenberg resolution. There results
a double complex of injectives (with exact columns)

.. .. ..
.O .O .O

Q0,1 / Q1,1 / ··· / Qn,1 / ···

O O O

Q0,0 / Q1,0 / ··· / Qn,0 / ···

O O O

F (Q0 ) / F (Q1 ) / ··· / F (Qn ) / ···

O O O

0 0 0

in the category B. Apply the functor G to this double complex to obtain a new double complex we will label
C:
.. .. ..
.O .O .O

G(Q0,1 ) / G(Q1,1 ) / ··· / G(Qn,1 ) / ···

O O O

/ G(Q1,0 ) / ··· / G(Qn,0 ) / ··· (C)

G(Q0,0 )
O O O

GF (Q0 ) / GF (Q1 ) / ··· / GF (Qn ) / ···

O O O

0 0 0 ,
in which, by hypothesis, all the columns are still exact. Therefore, using the notations for the two spectral
sequences converging to H • (C), we have HII
•
(C) = (0) so that (by our first remarks)

H • (C) ∼
= R• (GF )(A).

From the second spectral sequence, we get

IIl,m
2
l
= HII (HIm (C)) =⇒ R• (GF )(A).
l

Since we used a Cartan-Eilenberg resolution of F Q• (A), we have the following injective resolutions

0 −→ Z p (F Q• (A)) −→ Z p,0 −→ Z p,1 −→ · · ·

0 −→ B p (F Q• (A)) −→ B p,0 −→ B p,1 −→ · · ·
0 −→ H p (F Q• (A)) −→ H p,0 −→ H p,1 −→ · · · ,
5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 371

for all p ≥ 0. Moreover, the exact sequences

0 −→ Z p,• −→ Qp,• −→ B p+1,• −→ 0

and
0 −→ B p,• −→ Z p,• −→ H p,• −→ 0
are split because the terms are injectives of B. Therefore, the sequences

0 −→ G(Z p,• ) −→ G(Qp,• ) −→ G(B p+1,• ) −→ 0

and
0 −→ G(B p,• ) −→ G(Z p,• ) −→ G(H p,• ) −→ 0
are still exact and we find
HIp (C •,q ) = G(H p,q ).
But, the H p,• form an injective resolution of H p (F Q• (A)) and the latter is just Rp F (A). So, G(H p,• ) is the
q
complex whose cohomology is exactly Rq G(Rp F (A)). Now, this cohomology is HII (G(H p,• )) and HIp (C •,• )
p,•
is G(H ) by the above. We obtain
q
Rq G(Rp F (A)) = HII (HIp (C •,• )) = IIq,p •
2 =⇒ H (C).
q

Since H • (C) ∼
= R• (GF )(A), we are done.

There are many applications of the Spectral Sequence of Composed Functors. We give just a few of these.
(I) The Hochschild-Serre Spectral Sequence for the Cohomology of Groups
Write G for a (topological) group, N for a (closed) normal subgroup and A for a (continuous) G-module.
(Our main interest for non-finite or non-discrete groups is in the case of profinite groups because of their
connection with Galois cohomology in the non-finite
Q case. For a profinite group, the G-module is always
given the discrete topology and the action G A −→ A is assumed continuous.)
We have three categories: G-mod, G/N -mod and Ab. And we have the two functors

A H 0 (N, A) = AN (G-mod G/N -mod),

and
B H 0 (G/N, B) = B G/N (G/N -mod Ab).
G
Of course, their composition is exactly A A . To apply Grothendieck’s Theorem, we have to show
that if Q is an injective G-module, then QN is G/N -cohomologically trivial. But, I claim QN is, in fact,
G/N -injective. To see this, take 0 −→ M 0 −→ M exact in G/N -mod and look at the diagram (in G-mod)

Q
O X

?
QON

0 / M0 / M

Every G/N -module is a G-module (via the map G −→ G/N ) and Q is G-injective; so, the dotted arrow
exists as a G-homomorphism rendering the diagram commutative. Let θ be the dotted arrow; look at Im θ.
372 CHAPTER 5. HOMOLOGICAL ALGEBRA

If q = θ(m) and σ ∈ N ⊆ G, then σq = θ(σm) = θ(m) = q, because M is a G/N -module so N acts trivially
on it. Therefore q ∈ QN and so θ factors through QN , as required.
We obtain the Hochschild-Serre SS

H p (G/N, H q (N, A)) =⇒ H • (G, A). (HS)

p

Here is an application of importance for profinite groups (and Galois cohomology). If G is profinite,
write c.d.(G) ≤ r (resp. c.d.p (G) ≤ r) provided H s (G, M ) = (0) whenever M is a Z-torsion G-module
(resp. p-torsion G-module) and s > r. This notion is uninteresting for finite groups (see the exercises for
the reason).

Theorem 5.2 (Tower Theorem) If G is a profinite group and N is a closed normal subgroup, then

c.d.(G) ≤ c.d.(N ) + c.d.(G/N ).

(also true for c.d.p ).

Proof . We may assume c.d.(N ) ≤ a < ∞ and c.d.(G/N ) ≤ b < ∞, otherwise the result is trivial. Let M
be a torsion G-module and suppose n > a + b. All we need show is H n (G, M ) = (0). Write n = p + q with
p ≥ 0, q ≥ 0. In the Hochschild-Serre SS, the terms

E2p,q = H p (G/N, H q (N, M ))

must vanish. For if p ≤ b, then q > a and H q (N, M ) is zero by hypothesis. Now M is torsion therefore M N
is torsion and we saw in Chapter 4 that H q (N, M ) is always torsion if q > 0 as it is a right limit of torsion
groups. So, if q ≤ a, then p > b and E2p,q = (0) by hypothesis on G/N . Therefore, Esp,q = (0) for all s with
2 ≤ s ≤ ∞, when p + q = n > a + b. Hence, the terms in the composition series for H n (G, M ) all vanish
and we’re done.

(II) The Leray Spectral Sequence

The set-up here is a morphism
π : (X, OX ) −→ (Y, OY )
of ringed spaces (c.f. Section 5.3) and the three categories are: S(X), S(Y ), Ab. The functors are

π∗ : S(X) S(Y )

and
H 0 (Y, −) : S(Y ) Ab.

Of course, H 0 (X, −) : S(X) Ab is the composition H 0 (Y, −) ◦ π∗ . We must show that if Q is an

injective sheaf on X, then π∗ Q is cohomologically trivial on Y . Now every injective is flasque and flasque
sheaves are cohomologically trivial; so, it will suffice to prove π∗ takes flasque sheaves on X to flasque sheaves
on Y .
But this is trivial, for if U and V are open on Y and V ⊆ U , then π −1 (V ) ⊆ π −1 (U ) and

π∗ F (U ) F (π −1 (U ))

 
π∗ F (V ) F (π −1 (V ))
5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 373

shows that surjectivity of the left vertical arrow follows from surjectivity on the right. We therefore obtain
the Leray Spectral Sequence
H p (Y, Rq π∗ F ) =⇒ H • (X, F ). (LSS)
p

Unfortunately, full use of this spectral sequence demands considerable control of the sheaves Rq π∗ F and
this is vitally affected by the map π; that is, by the “relative geometry and topology of X vis a vis Y ”. We
must leave matters as they stand here.

(III) The Čech Cohomology Spectral Sequence

Once again, let (X, OX ) be a ringed space and write S(X) and P(X) for the categories of sheaves of OX -
modules and presheaves of OX -modules. We also have two left exact functors from P(X) to Ab. Namely,
if {Uα −→ X}α is an open cover of X and G ∈ P(X), then H 0 ({Uα −→ X}α , G) is in Ab and we have
Ȟ 0 (X, G), where the latter abelian group is what we called G(+) (X) in footnote 6 of Section 5.3. For the
three abelian categories: S(X), P(X), Ab we now have the two composed functors


S(X) 
0
i /o /o α →X},−)
/ P(X) H /o ({U /o /o /o /o / Ab


0

S(X)
i / P(X) /o Ȟo/ /o (X,−)
/o /o /o /o / Ab.

Observe that both composed functors are the same functor:

F ∈ S(X) H 0 (X, F ) ∈ Ab.

We need to show that if Q is an injective sheaf, then i(Q) is acyclic for either H 0 ({Uα → X}, −) or
0
Ȟ (X, −). However, part (3) of Corollary 5.33 says that i(Q) is injective as presheaf and is therefore acyclic.
From Grothendieck’s Theorem, we obtain the two Čech Cohomology Spectral Sequences:

H p ({Uα −→ X}α , Hq (F )) =⇒ H • (X, F ) (CCI)

p

Ȟ p (X, Hq (F )) =⇒ H • (X, F ) (CCII)

p

Now it turns out that Hq (F )# = (0) for every q > 0 and every sheaf, F . (See the exercises.) Also,
H (F )(+) ⊆ Hq (F )# ; so, we find
q

E20,q = Ȟ 0 (X, Hq (F )) = Hq (F )(+) = (0), when q > 0.

If we apply the edge sequence to (CCII), we deduce

Proposition 5.59 If (X, OX ) is a ringed space and F is a sheaf of OX -modules and if we continue to write
F when F is considered as a presheaf (instead of i(F )), then
(1) Ȟ 1 (X, F ) −→ H 1 (X, F ) is an isomorphism and
(2) Ȟ 2 (X, F ) −→ H 2 (X, F ) is injective.

(IV) The Local to Global Ext Spectral Sequence

Again, let (X, OX ) be a ringed space and fix a sheaf, A, of OX -modules on X. Write S(X) for the
(abelian) category of OX -modules. We can make a functor from S(X) to itself, denoted HomOX (A, −) via

HomOX (A, B)(U ) = HomOX U (A  U, B  U ).

374 CHAPTER 5. HOMOLOGICAL ALGEBRA

Here, U is open in X, the functor HomOX (A, −) is usually called the sheaf Hom, it is (of course) left exact
and its right derived functors (called sheaf Ext) are denoted Ext•OX (A, −).
Therefore, we have the situation of three categories S(X), S(X), Ab and the two functors

HomOX (A, −) : S(X) S(X)

0
H (X, −) = Γ(X, −) : S(X) Ab

whose composition is the functor HomOX (A, −). In order to apply Grothendieck’s Theorem, we must show
that if Q is injective in S(X), then HomOX (A, Q) is an acyclic sheaf . This, in turn, follows from
Proposition 5.60 Suppose that Q is an injective sheaf of OX -modules. Then HomOX (A, Q) is a flasque
OX -module.
Proof . If U is open in X, recall we have the presheaf AU defined by

A(V ) if V ⊆ U
AU (V ) =
(0) if V 6⊆ U

and this gives rise to the associated sheaf (AU )] . Now by adjointness,

HomOX ((AU )] , B) ∼
= HomOX -presheaves (AU , i(B)).

On the right hand side, if V is open and V ⊆ U , then an element of HomOX (AU , i(B)) gives the map
A(V ) −→ B(V ) (consistent with restrictions). But, if V 6⊆ U , we just get 0. However, this is exactly what
we get from HomOU (A  U, B  U ); therefore

HomOX ((AU )] , B) = HomOX U (A  U, B  U ).

Now take Q to be an injective sheaf, we have to show that

HomOX (A, Q) −→ HomOX U (A  U, Q  U )

is surjective for each open U of X. This means we must show that

HomOX (A, Q) −→ HomOX ((AU )] , Q)

is surjective. But, 0 −→ (AU )] −→ A is exact and Q is injective; so, we are done.

We obtain the local to global Ext spectral sequence

H p (X, ExtqOX (A, B)) =⇒ Ext•OX (A, B). (LGExt)

p

Remark: If j : U ,→ X is the inclusion of the open set U in X, then the sheaf we have denoted (AU )] above
is usually denoted j! A. The functor, j! , is left-exact and so we have a basic sequence of sheaf invariants R• j! .
Of course, we also have R• π∗ (for a morphism π : Y → X) as well as π ∗ , j ! (adjoint to j! ). The six operations

R• π∗ , R• j! , π ∗ , j ! , R• Hom, ⊗

were singled out by A. Grothendieck as the important test cases for the permanence of sheaf properties under
morphisms.

(V) “Associativity” Spectral Sequences for Ext and Tor

5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 375

In the proof of Grothendieck’s Theorem on the spectral sequence for composed functors, there were two
parts. In the first part, we used the essential hypothesis that F (Q) was G-acyclic to compute the cohomology
of the total complex (of our double complex) as R• (GF (A))—this is the ending of the spectral sequence. In
the second part, which depends only on using a Cartan-Eilenberg resolution and did not use the G-acyclicity
of F (Q), we computed the spectral sequence IIp,q 2 =⇒ H • (C) and found Rp G(Rq F (A)) =⇒ H • (C). This
second part is always available to us by Proposition 5.58 and we’ll make use of it below.
We consider modules over various rings. In order that we have enough flexibility to specialize to varying
cases of interest, we begin with three K-algebras, R, S, T and modules A, B, C as follows:

 A is a right R and a right S-module
(†) B is a left R-module and a right T -module

C is a right S and a right T -module.
Then
A ⊗R B is a right S ⊗K T -module
and
HomT (B, C) is a right R ⊗K S-module.
Observe that A is then a right R ⊗K S-module via
a(r ⊗ s) = (ar)s
because to say A is a right R and a right S-module is to imply
(ar)s = (as)r (all a ∈ A, r ∈ R, s ∈ S).
Also, C is a right S ⊗K T -module. We know in this situation there is an “associativity” isomorphism
HomR⊗K S (A, HomT (B, C)) ∼
= HomS⊗K T (A ⊗R B, C). (∗)

If S is K-projective and P• −→ A −→ 0 is an R ⊗K S-projective resolution of A, then P• −→ A −→ 0

is still an R-projective resolution of A and similarly if 0 −→ C −→ Q• is an S ⊗K T -injective resolution, it
still is a T -injective resolution of C. Our spectral sequences IIp,q
2 then give us two spectral sequences with
the same ending (by (∗)):
ExtpR⊗K S (A, ExtqT (B, C)) =⇒ Ending•
ExtpS⊗K T (TorR
q (A, B), C) =⇒ Ending• .

In a similar way, but this time if C is a (left) S and T -module, we get the “associativity” isomorphism
A ⊗R⊗K S (B ⊗T C) ∼
= (A ⊗R B) ⊗S⊗K T C. (∗∗)
Again, we assume S is K-projective and we get two spectral sequences with the same ending (by (∗∗)):
•
TorR⊗
p
KS
(A, TorTq (B, C)) ^
=⇒ Ending
•
TorS⊗
p
KT
(TorR
q (A, B), C)
^ .
=⇒ Ending

However, it is not clear how to compute the endings in these general cases. If we assume more, this can
be done. For example, say TorR q (A, B) = (0) if q > 0—this will be true when either A or B is flat over
R—then the second Ext sequence and second Tor sequence collapse and we find
ExtpR⊗K S (A, ExtqT (B, C)) =⇒ Ext•S⊗K T (A ⊗R B, C)
TorR⊗
p
KS
(A, TorTq (B, C)) =⇒ TorS⊗
•
KT
(A ⊗R B, C).
We have proved all but the last statement of
376 CHAPTER 5. HOMOLOGICAL ALGEBRA

Proposition 5.61 Suppose R, S, T are K-algebras with S projective over K and say A is an R and S right
module, C is an S and T right (resp. left) module and B is a left R and right T -module. Then there are
spectral sequences with the same ending

ExtpR⊗K S (A, ExtqT (B, C)) =⇒ Ending•

ExtpS⊗K T (TorR
q (A, B), C) =⇒ Ending•

(resp.
•
TorR⊗
p
KS
(A, TorTq (B, C)) ^
=⇒ Ending
•
TorS⊗
p
KT
(TorR
q (A, B), C)
^ )
=⇒ Ending

If TorR
q (A, B) = (0) when q > 0 (e.g. if A or B is R-flat) then

ExtpR⊗K S (A, ExtqT (B, C)) =⇒ Ext•S⊗K T (A ⊗R B, C) (Ext)

and
TorR⊗
p
KS
(A, TorTq (B, C)) =⇒ TorS⊗
•
KT
(A ⊗R B, C)). (Tor)
ExtqT (B, C)
Lastly, if B is T -projective (more generally vanishes if q > 0 and TorTq (B, C) vanishes if q > 0),
then we have the Ext and Tor associativity formulae

ExtpR⊗K S (A, HomT (B, C)) ∼ p

= ExtS⊗K T (A ⊗R B, C)

and
TorR⊗
p
KS
(A, B ⊗T C) ∼
= TorS⊗
p
KT
(A ⊗R B, C).

Proof . The last statement is trivial as our spectral sequences (Ext), (Tor) collapse.

Upon specializing the K-algebras R, S, T and the modules A, B, C, we can obtain several corollaries of
interest. For example, let S = Rop and A = R. Then ExtpR⊗Rop (R, −) = H p (R, −) in Hochschild’s sense (by
Section 5.3) and if R is K-projective the spectral sequences involving Ext yield

Corollary 5.62 If R is K-projective then there is a spectral sequence

H p (R, ExtqT (B, C)) =⇒ Ext•Rop ⊗K T (B, C)

provided B is a left R and right T -module and C is also a left R and right T -module.

Note that this is reminiscent of the local-global Ext spectral sequence. Note further that if B is also
T -projective, we deduce an isomorphism

H p (R, HomT (B, C)) ∼ p

= ExtRop ⊗T (B, C).

Next, let A = B = R = K in the Ext-sequences. If S is K-projective the second Ext sequence collapses
and gives ExtpS⊗K T (K, C) ∼
= Endingp . The first spectral sequence then yields

Corollary 5.63 Say S is K-projective and S and T possess augmentations to K, then we have the spectral
sequence
ExtpS (K, ExtqT (K, C)) =⇒ Ext•S⊗K T (K, C),
where C is a right S and right T -module.
5.4. SPECTRAL SEQUENCES; FIRST APPLICATIONS 377

Here is another corollary:

Corollary 5.64 Say S and T are K-algebras with S being K-projective. Assume C is a two-sided S ⊗K T -
module, then there is a spectral sequence

H p (S, H q (T, C)) =⇒ H • (S ⊗K T, C).

Proof . For this use K, S e , T e in place of R, S, T . Now S e is K-projective as S is so. Further replace A, B, C
by (S, T, C)—this is O.K. because C is indeed both a right S e and right T e -module by hypothesis. The
second Ext sequence collapses; so,

Extp(S⊗K T )e (S ⊗K T, C) ∼
= Endingp .

But, the left-side is just H p (S ⊗K T, C) by definition. Now the E2p,q term of our first Ext sequence is

ExtpS e (S, ExtqT e (T, C))

that is, it equals H p (S, H q (T, C)); so our proposition concludes the proof.

Clearly, there are analogous results for homology. Here are the conclusions, the exact hypotheses and
the proofs will be left as (DX).
op
Hp (T, TorR
q (A, B)) =⇒ Tor•R⊗K T (A, B)
TorSp (TorR
q (A, K), K) =⇒ Tor•R⊗K S (A, K)
Hp (S, Hq (R, A)) =⇒ H• (R ⊗K S, A).
378 CHAPTER 5. HOMOLOGICAL ALGEBRA

5.5 The Koszul Complex and Applications

In our previous work on the Hochschild cohomology of algebras, we studied the standard or bar complex,
but we saw that it was inefficient in several cases of interest. As mentioned there, we have another, much
better complex—the Koszul complex —which will serve for varied applications and which we turn to now.
Let A be a ring and M a module over this ring. For simplicity, we’ll assume A is commutative as the
main applications occur in this case. But, all can be done wih appropriate care in the general case. The
Koszul complex is defined with respect to any given sequence (f1 , . . . , fr ) of elements of A. We write
−
→
f = (f1 , . . . , fr ).
V• r V• r
Form the graded exterior power A . We make A into a complex according to the following prescrip-
tion: Since
^• a r ^ k
Ar = Ar ,
k=0

it is a graded module, and we just have to define differentiation. Let (e1 , . . . , er ) be the canonical basis of
Ar , and set
^0
dej = fj ∈ Ar = A,
then extend d to be an antiderivation. That is, extend d via
d(α ∧ β) = dα ∧ β + (−1)degα α ∧ dβ.
For example,
d(ei ∧ ej ) = fi ej − fj ei ,
and
d(ei ∧ ej ∧ ek ) = d(ei ∧ ej ) ∧ ek + (ei ∧ ej ) ∧ dek
= (fi ej − fj ei ) ∧ ek + fk (ei ∧ ej )
= fi ebi ∧ ej ∧ ek − fj ei ∧ ebj ∧ ek + fk ei ∧ ej ∧ ebk ,
where, as usual, the hat above a symbol means that this symbol is omitted. By an easy induction, we get
the formula:
t
X
d(ei1 ∧ · · · ∧ eit ) = (−1)j−1 fij ei1 ∧ · · · ∧ ec
ij ∧ · · · ∧ eit .
j=1
−→ V• r
We denote this complex by K• ( f ), i.e., it is the graded module A with the antiderivation d that we
just defined. This is the Koszul complex .
Given an A-module M , we can make two Koszul complexes for the module M , namely:
−
→ −
→
K• ( f , M ) = K• ( f ) ⊗A M,
−
→ −
→
K •( f , M ) = HomA (K• ( f ), M ).
We can take the homology and the cohomology respectively of these complexes, and we get the modules
−
→ −
→
H• ( f , M ) and H • ( f , M ).
For the cohomology complex, we need the explicit form of δ. Now,
^ t
−
→
K t ( f , M ) = HomA ( Ar , M ),
5.5. THE KOSZUL COMPLEX AND APPLICATIONS 379

and the family of elements of the form

ei1 ∧ · · · ∧ eit with 1 ≤ i1 < i2 < . . . < it ≤ r,

Vt r Vt
is a basis of A ; thus, HomA ( Ar , M ) is isomorphic to the set of alternating functions, g, from the
Vt (i1 , . . . , it ) of length t in {1, . . . , r} to M . Hence, the coboundary δ is given (on elements
set of sequences
g ∈ HomA ( Ar , M )) by
t+1
X
(δg)(i1 , . . . it+1 ) = (−1)j−1 fij g(i1 , . . . , ibj , . . . , it+1 ).
j=1

−
→ −
→ −
→
We have H 0 ( f , M ) = Z 0 ( f , M ) = Ker δ. (Note that K 0 ( f , M ) = M , via the map g 7→ g(1).) Then,

δg(ei ) = fi g(1) = fi m,

so δf = 0 implies that fi m = 0 for all i. We find that

−
→
H 0 ( f , M ) = {m ∈ M | Am = 0}, (5.1)

where A is the ideal generated by {f1 , . . . , fr }. Also, it is clear that

−
→
H t( f , M ) = 0 if t < 0 or t > r. (5.2)
−
→
Let us compute the top cohomology group H r ( f , M ). We have
r
^
−
→ −
→
Z r ( f , M ) = K r ( f , M ) = HomA ( Ar , M ) = M,

−
→ −
→ −
→
via the map g 7→ g(e1 ∧ · · · ∧ er ). Now, Im δr−1 = B r ( f , M ), but what is B r ( f , M )? If g ∈ K r−1 ( f , M )
is an alternating function on i1 , . . . , ir−1 , then
r
X
δr−1 g(1, . . . , r) = (δr−1 g)(e1 ∧ · · · ∧ er ) = (−1)j−1 fj g(1, . . . , b
j, . . . , r).
j=1

Therefore,
B r = f1 M + · · · + fr M,
and we find that
−
→
H r ( f , M ) = M/(f1 M + · · · + fr M ) = M/AM.

It is important to connect the Koszul homology (whose boundary map is

t
X
∂(ei1 ∧ · · · ∧ eit ⊗ m) = (−1)j−1 ei1 ∧ · · · ∧ ec
ij ∧ · · · eit ⊗ fij m)
j=1

−
→
and cohomology via the notion of Koszul duality. This is the following: Consider Kt ( f , M ); an element of
−
→
Kt ( f , M ) has the form
X
h= ei1 ∧ · · · ∧ eit ⊗ zi1 ...it , where 1 ≤ i1 < i2 < . . . < it ≤ r.
380 CHAPTER 5. HOMOLOGICAL ALGEBRA

We define a map (the duality map)

−
→ −
→
Θ : Kt ( f , M ) −→ K r−t ( f , M )

as follows: Pick j1 < j2 < · · · < jr−t , and set

Θ(h)(j1 , . . . , jr−t ) = zi1 ...it ,

where
(α) i1 , . . . , it is the set of complementary indices of j1 , . . . , jr−t taken in ascending order,
(β)  is the sign of the permutation

(1, 2, . . . , r) 7→ (i1 , . . . , it , j1 , . . . , jr−t ),

where both i1 , . . . , it and j1 , . . . , jr−t are in ascending order.

We find (DX) that

Θ(∂h) = δΘ(h),
where ∂ is the homology boundary map described above. So, the isomorphism, Θ, induces an isomorphism
−
→ −
→
Ht ( f , M ) ∼
= H r−t ( f , M ) for all t ≥ 0,

which is called Koszul duality. This notion of Koszul duality does not look like a duality, but we can make
it look so. For this, write Q(A) for “the” injective hull of A as A-module and set M D = HomA (M, Q(A)).
The cofunctor M M D is exact; we’ll refer to M D as the dual of M . Now the associativity isomorphism

HomA (M ⊗A N, Z) ∼
= HomA (M, HomA (N, Z))
−
→ −
→
shows that (Kt ( f , M ))D is isomorphic to K t ( f , M D ). Moreover, it is easy to see that

−
→ ∼
= / K t (−
→
(Kt ( f , M ))D f , M D)
O O
∂tD δ t−1

−
→ ∼
= / K t−1 (−
→
(Kt−1 ( f , M ))D f , M D)

is a commutative diagram. So, it follows (by the exactness of M M D ) (DX) that our isomorphisms yield
isomorphisms
−→ −
→
Ht ( f , M )D ∼
= H t ( f , M D ), for all t ≥ 0. (5.3)
Put these together with the above notion of Koszul duality and obtain the duality isomorphisms
−
→ ∼ −
→
H t( f , M D ) = H r−t ( f , M )D
−
→ −
→
Ht ( f , M D ) ∼
= Hr−t ( f , M )D , for all t ≥ 0.

Gathering together what we have proved above, we find the following

−
→
Proposition 5.65 If A is a (commutative) ring, M is an A-module, and f = (f1 , . . . , fr ) an ordered set
of r elements from A, then for the Koszul homology and cohomology of M we have
5.5. THE KOSZUL COMPLEX AND APPLICATIONS 381

−
→ −
→
(0) Ht ( f , M ) = H t ( f , M ) = (0) if t < 0 or t > r,

(1) (Koszul duality) There is an isomorphism

−
→ −
→
Ht ( f , M ) ∼
= H r−t ( f , M ), all t ≥ 0,

−
→ −
→
H0 ( f , M ) = H r ( f , M ) = M/AM,
(2) −
→ −
→
H 0 ( f , M ) = Hr ( f , M ) = {m | Am = 0},
where A is the ideal generated by f1 , . . . , fr .

Write M D = HomA (M, Q(A)) with Q(A) the injective hull of A, then
−
→ −
→
(3) Ht ( f , M )D ∼
= H t( f , M D )

and Koszul duality becomes

−
→ ∼ −
→
H t( f , M D ) = H r−t ( f , M )D ,
−
→ ∼ −
→
Ht ( f , M D ) = Hr−t ( f , M )D , for all t ≥ 0.

We need one more definition to exhibit the main algebraic property of the Koszul complex.

−→
Definition 5.12 The sequence f = (f1 , . . . , fr ) is regular for M or M -regular if for every i, with 1 ≤ i ≤ r,
the map
z 7→ fi z
is an injection of M/(f1 M + · · · + fi−1 M ) to itself.

By its very definition, the notion of M -regularity appears to depend on the order of the elements f1 , . . . , fr .
This is indeed the case as the following classical example [39] shows: Let A be C[X, Y, Z] and f1 = X(Y − 1);
f2 = Y ; f3 = Z(Y − 1). Then unique factorization in A shows that f1 , f2 , f3 is A-regular, but f1 , f3 , f2
is certainly not A-regular as f3 X is zero in A/f1 A but X is not zero there. In the special case that A is
graded, M is a graded module and the fj are homogeneous elements of A, the order of an M -sequence does
not matter.
If A is a given ideal of A and f1 , . . . , fr ∈ A (the fj are not necessarily generators of A), and if f1 , . . . , fr is
an M -regular sequence but no for other element g ∈ A is f1 , . . . , fr , g an M -regular sequence, then f1 , . . . , fr
is a maximal M -regular sequence from A. It turns out that the number of elements in a maximal M -regular
sequence from A is independent of the choice of such a sequence; this number is called the A-depth of M
and denoted depthA M . (When A is a local ring and A = M is its maximal ideal, one writes depth M and
omits any reference to M.)
Here is the main property of the Koszul complex vis a vis M -regularity (and, hence, depth):

−
→
Proposition 5.66 (Koszul) Suppose M is an A-module and f is an M -regular sequence of length r. Then
−
→ −
→
the Koszul complexes K• ( f , M ) and K • ( f , M ) are acyclic and consequently

−
→ −
→
Hi ( f , M ) = (0) if i 6= 0 and H i ( f , M ) = (0) if i 6= r.
382 CHAPTER 5. HOMOLOGICAL ALGEBRA

Proof . The two Koszul complexes

r−1
^
−
→ ∂r ∂ ∂ ∂1
K• ( f , M ) : M −→ Ar ⊗A M −→ · · · −→ Ar ⊗A M −→ M
^ r−1
−
→ δ0 δ δ δ r−1
K • ( f , M ) : M −→ HomA (Ar , M ) −→ · · · −→ HomA ( Ar , M ) −→ M

−
→ −
→
will be exact sequences when H1 ( f , M ) = · · · = Hr−1 ( f , M ) = (0) and when
−
→ −
→
H 1 ( f , M ) = · · · = H r−1 ( f , M ) = (0); so the vanishing statement of the conclusion appears stronger than
acyclicity. But, under our hypothesis the modules
−
→ −
→
Hr ( f , M ) = H 0 ( f , M ) = {m | Am = (0)}

automatically vanish since f1 is a non-zero divisor on M .

We will prove the vanishing statements and, of course, by Koszul duality all we need prove is that
−
→
Ht ( f , M ) = (0) for all t > 0. There are several ways of proving this; all use induction on r, the length of
the M -sequence. We choose a method involving the tensor product of complexes.
If C• and D• are left complexes, we make their tensor product C• ⊗ D• by setting
a
(C• ⊗ D• )t = Ci ⊗ Dj
i+j=t

and defining differentiation by

d(α ⊗ β) = dC (α) ⊗ β + (−1)deg α α ⊗ dD (β).

Then, (C• ⊗ D• )• is a complex. Consider for example the Koszul complex for the single element f ∈ A.
Namely, 
A if t = 0 or 1
K• (f )t =
(0) if t > 1
a two term complex. Its differentiation is given by d(e) = f , where e (= 1) is a base for A as A-module; in
other words, d is just multiplication by f . With this notation, we have
−
→
K• ( f ) = K• (f1 ) ⊗ · · · ⊗ K• (fr ).

−
→
Now the vanishing statements are true and trivial for r = 0 or 1. So, write f 0 = (f1 , . . . , fr−1 ) and set
−→ −
→
L• = K• ( f 0 , M ). Since f 0 is M -regular we see that
−→
Ht ( f 0 , M ) = Ht (L• ) = (0), for all t > 0,
−
→
by the induction hypothesis. Further, set M• = K• (fr , M ). Then K• ( f , M ) = (L• ⊗ M• )• , and this will
enable our induction.
I claim that we have the exact sequence

· · · −→ H0 (Ht (L• ) ⊗ M• ) −→ Ht (L• ⊗ M• ) −→ H1 (Ht−1 (L• ) ⊗ M• ) −→ · · · (5.4)

for every t ≥ 0. Suppose this claim is proved, take t ≥ 2 (so that t − 1 ≥ 1) and get

Ht (L• ) = Ht−1 (L• ) = (0)

5.5. THE KOSZUL COMPLEX AND APPLICATIONS 383

−
→
by the induction hypothesis. The exact sequence (5.4) tells us that Ht ( f , M ) = Ht (L• ⊗ M• ) = (0) when
t ≥ 2. If t = 1, we know that H1 (L• ) vanishes, so (5.4) gives us the exact sequence
−
→
0 −→ H1 ( f , M ) −→ H1 (H0 (L• ) ⊗ M• ).

But H1 (−) = H 0 (−) by Koszul duality for M• and the latter is the kernel of multiplication by fr on (−).
However, in this case (−) is H0 (L• ) = M/(f1 M + · · · + fr−1 M ); the kernel of multiplication by fr on this
last module is zero because f1 , . . . , fr is M -regular. We conclude H1 (H0 (L• ) ⊗ M• ) is zero, finishing our
induction.
There remains only the proof of exact sequence (5.4). It, in turn, follows from a general homological
lemma:

Lemma 5.67 Suppose M is a two-term complex of A-modules, zero in degree 6= 0, 1 and for which M0 and
M1 are free A-modules. If L• is any complex of A-modules, we have the exact sequence

· · · −→ H0 (Ht (L• ) ⊗ M• ) −→ Ht (L• ⊗ M• ) −→ H1 (Ht−1 (L• ) ⊗ M• ) −→ · · · (5.4)

for all t ≥ 0.

Proof . Once again, we have more than one proof available. We’ll sketch the first and give the second in
detail. The modules comprising M• are A-free, so there is a “Künneth Formula” spectral sequence
2
Ep,q = Hp (Hq (L• ) ⊗A M• ) =⇒ H• (L• ⊗A M• ).
2
(For example, see Corollary 5.64 and its homology analog.) But, as M• is a two-term complex, Ep,q = (0) if
p 6= 0, 1 and we obtain the zipper sequence (??) of our lemma.
More explicitly (our second proof), we make two one-term complexes, Mi , in which Mi has its one term
in degree i (i = 0, 1). Each differentiation in these complexes is to be the trivial map. We form the tensor
product complexes L• ⊗ Mi and recall that

 (L• ⊗ M0 )p = Lp ⊗ M0
d(α ⊗ β) = dL (α) ⊗ β

Hp (L• ⊗ M0 ) = Hp (L• ) ⊗ M0

and 
 (L• ⊗ M1 )p = Lp−1 ⊗ M1
d(α ⊗ β) = dL (α) ⊗ β

Hp (L• ⊗ M1 ) = Hp−1 (L• ) ⊗ M1 .

Then, we obtain an exact sequence of complexes

0 −→ L• ⊗ M0 −→ L• ⊗ M• −→ L• ⊗ M1 −→ 0

and its corresponding long exact homology sequence

∂
· · · Hp+1 (L• ⊗ M1 ) −→ Hp (L• ⊗ M0 ) −→ Hp (L• ⊗ M• ) −→ Hp (L• ⊗ M1 ) −→ · · · .

But, ∂ is just 1 ⊗ dM• and so the above homology sequence is exactly (??).
The main applications we shall make of the Koszul complex concern the notion of “dimension”—and even
here most applications will be to commutative rings. We begin by defining the various notions of dimension.
Suppose R is a ring (not necessarily commutative) and M is an R-module.
384 CHAPTER 5. HOMOLOGICAL ALGEBRA

Definition 5.13 The module M has projective dimension (resp. injective dimension) ≤ n if and only if it
possesses a projective (resp. injective) resolution P• −→ M −→ 0 (resp. 0 −→ M −→ Q• ) for which Pt = 0
(resp. Qt = 0) when t > n. The infimum of the integers n for which M has projective (resp. injective)
dimension ≤ n is called the projective dimension (resp. injective dimension) of M .

Remark: Of course, if no n exists so that proj dim M ≤ n, then we write proj dim M = ∞ and similarly
for injective dimension. A module is projective (resp. injective) iff it has proj (resp. inj) dim = 0. It is
convenient to set proj (or inj) dim (0) equal to −∞. If M is a right R-module, it is an Rop -module and
so it has proj (and inj) dimension as Rop -module. Therefore, it makes sense to include R in the notation
and we’ll write dimR M for the projective or injective dimension of M (as R-module) when no confusion can
arise.
By this time, the following propositions, characterizing the various dimensions, are all routine to prove.
So, we’ll omit all the proofs leaving them as (DX’s).

Proposition 5.68 If R is a ring and M is an R-module, then the following are equivalent conditions:

(1) M has projective dimension ≤ n (here, n ≥ 0)

(2) Extn+1
R (M, −) = (0)

(3) ExtnR (M, −) is a right exact functor

(4) If 0 −→ Xn −→ Pn−1 −→ · · · −→ P0 −→ M −→ 0 is an acyclic resolution of M and if P0 , . . . , Pn−1

are R-projective, then Xn is also R-projective
Also, the following four conditions are mutually equivalent:

(10 ) M has injective dimension ≤ n (here, n ≥ 0)

(20 ) Extn+1
R (−, M ) = (0)

(30 ) ExtnR (−, M ) is a right exact functor

(40 ) If 0 −→ M −→ Q0 −→ · · · −→ Qn−1 −→ X n −→ 0 is an acyclic resolution of M and if Q0 , . . . , Qn−1

are R-injective, then X n is also R-injective.

If we use the long exact sequence of (co)homology, we get a corollary of the above:

Corollary 5.69 Say 0 −→ M 0 −→ M −→ M 00 −→ 0 is an exact sequence of R-modules.

(1) If dimR M 0 and dimR M 00 ≤ n (either both projective or both injective dimension), then dimR M ≤ n

(2) Suppose M is projective, then either

(a) dimR M 00 = 0 (i.e., M 00 is projective), in which case M 0 is also projective; or

(b) dimR M 00 ≥ 1, in which case dimR M 0 = dimR M 00 − 1.

To get an invariant of the underlying ring, R, we ask for those n for which projdimR M ≤ n (resp.
injdimR M ≤ n) for all R-modules M . For such an n, we write gldim R ≤ n and say the global dimension
of R is less than or equal to n. (It will turn out that we can check this using either projdim for all M or
injdim for all M ; so, no confusion can arise.) Of course, the infimum of all n so that gldim R ≤ n is called
the global dimension of R. When we use right R-modules, we are using Rop -modules and so are computing
gldim Rop .
5.5. THE KOSZUL COMPLEX AND APPLICATIONS 385

Notice that gldim R is an invariant computed from the category of R-modules. So, if R and S are
rings and if there is an equivalence of categories R-mod ≈ S-mod, then gldim R = gldim S. Rings for
which R-mod ≈ S-mod are called Morita equivalent rings. For commutative rings, it turns out that Morita
equivalence is just isomorphism; this is not true for non-commutative rings. Indeed, if R is a ring and Mn (R)
denotes, as usual, the ring of n × n matrices over R, then R ≈ Mn (R). Moreover, this is almost the full
story. Also, if R ≈ S, then Rop ≈ S op . Now for a field, K, we clearly have gldim K = 0; so, we find
gldim Mn (K) = 0, as well. If A is a commutative ring and G is a group, then the map σ 7→ σ −1 gives an
isomorphism of A[G] onto A[G]op . Hence, gldim A[G] = gldim A[G]op .

Proposition 5.70 Let R be a ring and let n be a non-negative integer. Then the following statements are
equivalent:

(1) Every R-module, M , has projdimR ≤ n.

(2) Every R-module, M , has injdimR ≤ n.

(3) gldim R ≤ n.

(4) ExttR (−, −) = (0) for all t > n.

(5) Extn+1
R (−, −) = (0).

(6) ExtnR (−, −) is right-exact.

A ring R is called semi-simple if and only if every submodule, N , of each R-module, M , possesses an
R-complement. (We say M is completely reducible.) That is, iff given N ⊆ M , there is a submodule N e ⊆M
` e
so that the natural map N N −→ M is an isomorphism (of R-modules). Of course each field, K, or
division ring, D, is semi-simple. But, again, semi-simplicity is a property of the category R-mod; so Mn (K)
and Mn (D) are also semi-simple. It is a theorem of Maschke that if K is a field, G is a finite group, and
(ch(K), #(G)) = 1, then the group algebra, K[G], is semi-simple. See Problem 134 for this result. Again,
there are many equivalent ways to characterize semi-simplicity:

Proposition 5.71 For any ring, R, the following statements are equivalent:

(1) R is semi-simple.

(2) Rop is semi-simple.

(3) R, as R-module, is a coproduct of simple R-modules.

(4) R, as R-module, is completely reducible.

(5) Each (left) ideal of R is an injective module.

(6) Every R-module is completely reducible.

(7) In R-mod, every exact sequence splits.

(8) Every R-module is projective.

(9) Every R-module is injective.

(10) gldim R = 0.
386 CHAPTER 5. HOMOLOGICAL ALGEBRA

The proofs of these equivalences will be left as the material of Problem 145. Note that dimension
is defined using Ext•R (−, −) and TorR • is not mentioned. There are two reasons for this. First, while
Hom, Ext, projective and injective are properties of abelian categories, tensor and Tor are generally not.
Second, the vanishing of Tor characterizes flatness which is a weaker property than projectivity. However,
for commutative rings, the notions of dimension and global dimension are frequently reduced by localization
to the case of local rings. For noetherian local rings, we already know flatness and freeness are equivalent
for f.g. modules; so over noetherian local rings the vanishing of Tor is connected with dimension (at least
on the category of f.g. modules). In the general case, when we use Tor, we call the resulting invariant the
Tor-dimension. It’s easy to see that when R is a PID we have gldim R ≤ 1.
For our main applications of the Koszul complex, we return to the situation of a pair (R, Q) in which
R is a ring and  : R → Q is a surjective R-module map. Such a pair is an augmented ring, the map  is
the augmentation (as discussed in Section 5.3) and Q is the augmentation module. As usual, write I for the
augmentation ideal (just a left ideal, in general): I = Ker . Then the exact sequence

0 −→ I −→ R −→ Q −→ 0

and Corollary 5.69 above show:

Either Q is projective (so that I is projective) or 1 + dimR I = dimR Q.
Note that if R is commutative then I is a 2-sided ideal and Q becomes a ring if we set (r) · (p) = r · (p);
i.e., if we make  a ring homomorphism. The map  is then a section in case R is a Q-algebra. Here is the
main result on which our computations will be based.

Theorem 5.72 Assume (R, Q) is an augmented ring and suppose I is finitely generated (as R-ideal) by
elements f1 , . . . , fr which commute with each other. If f1 , . . . , fr form an R-regular sequence, then
dimR Q = r (if Q 6= (0)). In particular, gldim R ≥ r.

Proof . Write A for the commutative ring Z[T1 , . . . , Tr ], then as the f1 , . . . , fr commute with each other, R
becomes an A-module if we make Tj operate via ρTj = ρfj for all ρ ∈ R. We form the Koszul complex
−
→ −
→ −
→
K• ( T ) for A and then form R ⊗A K• ( T ). The latter is clearly the Koszul complex K• ( f , R) and as
−
→
(f1 , . . . , fr ) is an R-regular sequence, K• ( f , R) is acyclic. Thus, we obtain the exact sequence
r−1
^ 1
^
∂ ∂r−1 ∂ ∂1
r
0 −→ R −→ (Rr ) −→ · · · −→
2
(Rr ) −→ R −→ Q −→ 0 (∗)

because we know the image of ∂1 is the (left) ideal generated by f1 , . . . , fr ; that is, Im ∂1 = I. Now (∗) is
visibly an R-projective resolution of Q and so dimR Q ≤ r.
Since (∗) is an R-projective resolution of Q, we can use it to compute Ext•R (Q, −). In particular, we can
compute Ext•R (Q, Q)—this is the cohomology of the complex HomR ((∗), Q). But, the latter complex is just
−
→
K • ( f , Q). We find
−
→
ExtrR (Q, Q) = H r ( f , Q) = Q
and so dimR Q = r provided Q 6= (0).

Corollary 5.73 If K is a ring (not necessarily commutative) and R is the graded ring K[T1 , . . . , Tr ], then
dimR K = r. If K is a field or division ring and R is the local ring of formal power series K[[T1 , . . . , Tr ]],
then dimR K = r. (This is also true if K is any ring though R may not be local.) Lastly, if K is a field
complete with respect to a valuation and R is the local ring of converging power series K{T1 , . . . , Tr }, then
dimR K = r. In all these cases, gldim R ≥ r.
5.5. THE KOSZUL COMPLEX AND APPLICATIONS 387

Proof . In each case, the variables T1 , . . . , Tr play the role of the f1 , . . . , fr of our theorem; all hypotheses
are satisfied.
Notice that for A (= Z[T1 , . . . , Tr ]), the Koszul resolution
r−1
^ 1
^
0 −→ A −→ (Ar ) −→ · · · (Ar ) −→ A −→ Z −→ 0 (∗∗)

•
can be used to compute TorA
• (−, Z) as well as ExtA (Z, −). So for M , any A-module,

−→ −→
TorA
p (M, Z) = Hp ( T , M ) and ExtpA (Z, M ) = H p ( T , M ).

By Koszul duality,
TorA ∼ r−p
p (M, Z) = ExtA (Z, M ).

−→
Further, the acyclicity of M ⊗A K• ( T ) is equivalent with TorA
p (M, Z) = (0) when p > 0.

Now, recall that, for a ring R possessing a section R −→ K (here, R is a K-algebra), we defined the
homology and cohomology “bar” groups by

H n (R, M ) = TorR
n (M, K) (M an Rop -module)
n
H (R, M ) = ExtnR (K, M ) (M an R-module).

In the cases
(1) R = K[T1 , . . . , Tr ]
(2) R = K[[T1 , . . . , Tr ]]
(3) R = K{T1 , . . . , Tr } (K has a topology),
our discussion above shows that
−→ n −→
H n (R, M ) = Hn ( T , M ) and H (R, M ) = H n ( T , M ).

So, by the Hochschild (co)homology comparison theorem (Theorem 5.29), we see that the Hochschild groups
−→ • −
→
Hn (R, ∗ (M )) and H n (R, op
∗ (M )) can be computed by the Koszul complexes K• ( T , M ) and K ( T , M ) in
cases (1)–(3) above. This is what we alluded to at the end of the discussion following Theorem 5.29.
We now face the problem of the global dimension of a ring R. We assume R is not only an augmented
ring (with augmentation module, Q, and ideal, I) but in fact that I is a two-sided ideal so that Q is a ring
and  : R → Q is a ring homomorphism. Experience shows that for certain types of rings some subclasses
of modules have more importance than others. For example, if R is a graded ring, the graded modules are
the important ones for these are the ones giving rise to sheaves over the geometric object corresponding
to R (a generalized projective algebraic variety) and the cohomology groups of these sheaves are geometric
invariants of the object in question. Again, if R is a (noetherian) local ring, the finitely generated modules
are the important ones as we saw in Chapter 3. It makes sense therefore to compute the global dimension
of R with respect to the class of “important” R-modules, that is to define

I-gldim R = inf{m | I-gldimR ≤ m},

where I-gldimR ≤ m iff for every important module, M , we have dimR M ≤ m; (here, I–stands for “impor-
tant”).
Eilenberg ([9]) abstracted the essential properties of the graded and finitely generated modules to give
an axiomatic treatment of the notion of the class of “important” modules. As may be expected, the factor
388 CHAPTER 5. HOMOLOGICAL ALGEBRA

ring, Q, plays a decisive role. Here is the abstract treatment together with the verification that for graded
(resp. local) rings, the graded (resp. finitely generated) modules satisfy the axioms.
(A) Call an Rop -module, M , pertinent provided M ⊗R Q 6= (0) when M 6= (0); also (0) is to be pertinent.
` `
If R is a graded ring, say R = j≥0 Rj , then we set I = R(+) = j>0 Rj and Q = R0 . ` When M is a
graded `Rop -module with grading bounded below then M is pertinent. For we have M = n≥B Mn ; so,
M I = n≥B+1 Mn 6= M . But M ⊗R Q = M/M I. When R is a local ring, we set I = MR (its maximal
ideal) and then Q = κ(R)–the residue field. Of course, all f.g. Rop -modules are pertinent by Nakayama’s
Lemma.
If S is a subset of a module, M , write F (S) for the free R (or Rop )-module generated by S. Of course,
there’s a natural map F (S) −→ M and we get an exact sequence
0 −→ Ker (S) −→ F (S) −→ M −→ cok(S) −→ 0.
(B) The subset, S, of M is good provided 0 ∈ S and for each T ⊆ S, in the exact sequence
0 −→ Ker (T ) −→ F (T ) −→ M −→ cok(T ) −→ 0,
the terms Ker (T ) and cok(T ) are pertinent.
Notice right away that free modules are pertinent; so, if S is good and we take {0} = T , then, as the map
F ({0}) −→ M is the zero map, we find M = cok({0}) and therefore M is pertinent. That is, any module
possessing a good subset is automatically pertinent. Conversely, if M is pertinent, then clearly S = {0} is a
good subset; so, we’ve proved
Proposition 5.74 If R is an augmented ring and I is two-sided, the following are equivalent conditions on
an Rop -module, M :
(a) M possesses a good subset
(b) M is pertinent
(c) {0} is a good subset of M .
In the case that R is a graded ring, we shall restrict all attention to modules whose homogeneous elements
(if any) have degrees bounded below. In this case, any set S ⊆ M consisting of 0 and homogeneous elements
is good. For suppose T ⊆ S, then we grade F (T ) by the requirement that F (T ) −→ M be a map of degree
zero (remember; M is graded and, further, observe if 0 ∈ T it goes to 0 in M and causes no trouble). But
then, Ker (T ) and cok(T ) are automatically graded (with grading bounded below) and so are pertinent. If
Rop is a noetherian local ring and M is a finitely generated Rop -module, then any finite set containing 0 is
good . For if S is finite, then any T ⊆ S is also finite and so all of F (T ), M , cok(T ) are f.g. But since Rop is
noetherian, Ker (T ) is also f.g.
(C) A family, F, of Rop -modules is a class of important modules provided
(1) If M ∈ F it has a good set S which generates M , and
(2) In the exact sequence
0 −→ Ker (S) −→ F (S) −→ M −→ 0
resulting from (1), we have Ker (S) ∈ F.
For graded rings, R, the graded modules (whose degrees are bounded below) form an important family.
This is easy since such modules are always generated by their homogeneous (= good) elements and Ker (S)
is clearly graded and has degrees bounded below. In the case that R is noetherian local, the family of all f.g.
modules is important. Again, this is easy as such modules are generated by finite (= good) sets and Ker (S)
is again f.g. because R is noetherian.
In what follows, one should keep in mind the two motivating examples and the specific translations of
the abstract concepts: pertinent modules, good sets of elements, the class of important modules.
5.5. THE KOSZUL COMPLEX AND APPLICATIONS 389

Abstract R R graded R noetherian local

pertinent module graded module with finitely generated module
degrees bounded below
good subset of subset of homogeneous finite subset of module
a module elements of a module
class of important class of graded modules class of finitely
modules with degrees bounded below generated modules

Since we have abstracted the local case, it is no surprise that we have a “generalized Nakayama’s Lemma”:
Proposition 5.75 (Generalized Nakayama’s Lemma) If (R, Q) is an augmented ring with I a two-sided
ideal, then for any good subset, S, of an Rop -module, M , whenever the image of S −{0} in M ⊗R Q generates
M ⊗R Q as Qop -module the set S − {0} generates M . Moreover, if TorR 1 (M, Q) = (0), and the image of
S − {0} freely generates M ⊗R Q as Qop -module, then S − {0} freely generates M as Rop -module.
Proof . The proof is practically identical to the usual case (write S instead of S − {0}): We have the exact
sequence
ϕ
0 −→ Ker (S) −→ F (S) −→ M −→ cok(S) −→ 0,
and we tensor with Q. We obtain the exact sequence
ϕ
F (S) ⊗R Q −→ M ⊗R Q −→ cok(S) ⊗R Q −→ 0
and we’ve assumed ϕ is surjective. Thus cok(S) ⊗R Q = (0), yet cok(S) is pertinent; so cok(S) = (0). Next,
our original sequence has become
0 −→ Ker (S) −→ F (S) −→ M −→ 0,
so we can tensor with Q again to obtain
ϕ
TorR
1 (M, Q) −→ Ker (S) ⊗R Q −→ F (S) ⊗R Q −→ M ⊗R Q −→ 0.

Since, in the second part, we’ve assume ϕ is an isomorphism and TorR

1 (M, Q) = (0), we get Ker (S) ⊗R Q =
(0). But, Ker (S) is also pertinent and so Ker (S) = (0).

If we specialize Q, we get the following:

Corollary 5.76 With (R, Q) as in the generalized Nakayama’s Lemma and assuming Q is a (skew) field,8
we have the following equivalent conditions for an Rop -module, M , which is generated by a good set:
(1) M is free over Rop
(2) M is Rop -flat
(3) TorR
n (M, Q) = (0) if n > 0

(4) TorR
1 (M, Q) = (0).

Moreover, under these equivalent conditions, every good generating set for M contains an Rop -basis for M .
Proof . (1) =⇒ (2) =⇒ (3) =⇒ (4) are trivial or are tautologies.
(4) =⇒ (1). The image, S, of our good generating set generates M ⊗R Q. But, Q is a (skew) field; so S
contains a basis and this has the form T for some T ⊆ S. Then T ∪ {0} is good and (4) with generalized
Nakayama shows T is an Rop -basis for M . This gives (1) and even proves the last asssertion.

Finally, we have the abstract I-gldim theorem, in which I is a class of important modules.
8A skew field is a division ring.
390 CHAPTER 5. HOMOLOGICAL ALGEBRA


Theorem 5.77 (I-Global Dimension Theorem) If (R, Q) is an augmented ring in which Ker (R −→ Q) is
a two-sided ideal, if Q is a (skew) field and if I is an important class of Rop -modules, then
I-gldim Rop ≤ dimR Q (proj. dim).
Proof . Of course, w.m.a. dimR Q < ∞ else there is nothing to prove; so, write n = dimR Q. The proof
is now practically forced, namely pick an important Rop -module, M , then by assumption there is a good
generating set, S0 , in M and an exact sequence
0 −→ Ker (S0 ) −→ F (S0 ) −→ M −→ 0.
Also, Ker (S0 ) is important so, there’s a good generating set, S1 , in Ker (S0 ) and an exact sequence
0 −→ Ker (S1 ) −→ F (S1 ) −→ Ker (S0 ) −→ 0
in which Ker (S1 ) is again important. We repeat and obtain the chain of exact sequences

0 −→ Ker (S0 ) −→ F (S0 ) −→ M −→ 0 


0 −→ Ker (S1 ) −→ F (S1 ) −→ Ker (S0 ) −→ 0
(†)
...................... 


0 −→ Ker (St ) −→ F (St ) −→ Ker (St−1 ) −→ 0
for all t. Upon splicing these sequences, we get the exact sequence
0 −→ Ker (St ) −→ F (St ) −→ F (St−1 ) −→ · · · −→ F (S1 ) −→ F (S0 ) −→ M −→ 0. (††)

Now in the sequence

0 −→ Ker (St ) −→ F (St ) −→ Ker (St−1 ) −→ 0
(t ≥ 0 and Ker (S−1 ) = M ), we compute Tor and find
· · · −→ TorR R R R
r+1 (F (St ), Q) −→ Torr+1 (Ker (St−1 ), Q) −→ Torr (Ker (St ), Q) −→ Torr (F (St ), Q) −→ · · ·

for all r ≥ 1 and t ≥ 0; hence the isomorphisms

TorR ∼ R
r+1 (Ker (St−1 ), Q) = Torr (Ker (St ), Q).

Take r = 1 and t = n − 1, then

TorR ∼ R
2 (Ker (Sn−2 ), Q) = Tor1 (Ker (Sn−1 ), Q)

and similarly
TorR ∼ R
3 (Ker (Sn−3 ), Q) = Tor2 (Ker (Sn−2 ), Q),

etc. We find
TorR ∼ R
n+1 (Ker (S−1 ), Q) = Tor1 (Ker (Sn−1 ), Q). (†††)
But, dimR Q = n and so TorR
n+1 (−, Q) = (0); thus,

TorR
1 (Ker (Sn−1 ), Q) = (0).

Now Ker (Sn−1 ) is important and the corollary to Generalized Nakayama shows that Ker (Sn−1 ) is Rop -free.
Thus,
0 −→ Ker (Sn−1 ) −→ F (Sn−1 ) −→ · · · −→ Ker (S1 ) −→ F (S0 ) −→ M −→ 0
is a free resolution (length n) of M and this proves dimRop (M ) ≤ n. But, M is arbitrary and we’re done.
Notice that the above argument is completely formal except at the very last stage where we used the
vanishing of TorR R
n+1 (−, Q). But, this vanishing holds if Tor -dim(Q) ≤ n and therefore we’ve actually proved
the following stronger version of the I-global dimension theorem:
5.5. THE KOSZUL COMPLEX AND APPLICATIONS 391

Corollary 5.78 Under the hypotheses of the I-global dimension theorem, we have the strong I-global di-
mension inequality
I-gldim(Rop ) ≤ TorR -dim(Q).

To recapitulate and set these ideas firmly in mind, here are the two special, motivating cases:

Theorem 5.79 (Syzygy9 Theorem) Assume of the ring R that either

(I) R is graded; R = Q q R1 q R2 q · · · , and Q is a (skew) field,
or
(II) R is local with Rop noetherian and Q = κ(R) is a (skew) field.
Then, when I is, in case (I), the class of graded Rop -modules with degrees bounded below, or in case (II),
the class of finitely generated Rop -modules, we have

I-gldim(Rop ) ≤ TorR -dim(Q).

Moreover, either Q is projective or else if A is any Rop -ideal (which in case (I) is homogeneous), then we
have
1 + dimRop (A) ≤ dimR (Q).

In case (I) of the Syzygy Theorem, note that Q is an Rop -module, too and that I can be taken to be the
class of graded (with degrees bounded below) R-modules. Therefore, dimRop (Q) ≤ TorR -dim(Q) ≤ dimR (Q).
Interchanging R and Rop as we may, we deduce
Corollary 5.80 In case (I) of the Syzygy Theorem, we have
op
(a) dimR (Q) = dimRop (Q) = TorR -dim(Q) = TorR -dim(Q),
(b) I-gldim(R) ≤ TorR -dim(Q),
(c) 1 + dimR (A) ≤ dimR (Q) if Q is not projective.

Similarly, in case (II) of the Syzygy Theorem, provided we assume R noetherian, I is again the family
of f.g. R-modules when we interchange R and Rop . There results
Corollary 5.81 If R is local with both R and Rop noetherian, then
op
(a) dimR (κ(R)) = dimRop (κ(R)) = TorR -dim(κ(R)) = TorR -dim(κ(R)),
(b) I-gldim(R) ≤ TorR -dim(κ(R)),
(c) 1 + dimR (A) ≤ TorR -dim(κ(R)).
Finally, there are the cases that appeared first in the literature:
Corollary 5.82 If K is a (skew) field and
(I) (Hilbert Syzygy Theorem) R = K[T1 , . . . , Tn ] and M is a graded R-module with degrees bounded from
below or A is a homogeneous R-ideal then

dimR M ≤ n and dimR A ≤ n − 1.

9 TheGreek (later Latin) derived word “syzygy” means a coupling, pairing, relationship. Thus, for the exact sequence
0 −→ Ker (S) −→ F (S) −→ M −→ 0, the generators of Ker (S) are relations among the generators of M and generate all such
relations. For 0 −→ Ker (S1 ) −→ F (S1 ) −→ Ker (S) −→ 0, generators of Ker (S1 ) are relations among the relations and so on.
Each of Ker (Sj ) is a syzygy module.
 392 CHAPTER 5. HOMOLOGICAL ALGEBRA

or
(II) R = K[[T1 , . . . , Tn ]] (resp. R = K{T1 , . . . , Tn } when K has a non-discrete (valuation) topology) and
M is a f.g. R-module while A is an R-ideal, then

dimR M ≤ n and dimR A ≤ n − 1.

To prove these, we use the above and Theorem 5.72.

Remarks:
(1) Note that Q appears as the “worst” module, i.e., the one with the largest homological dimension. In
the case of a commutative local ring, R, if M is generated by an R-regular sequence (R is then called
a regular local ring) of length n, we see that I-gldim(R) = n and Q = κ(R) achieves this maximum
dimension. The finiteness of global dimension turns out to be characteristic of regular local rings (Serre
[46]).


(2) One might think of interchanging R and Rop in the general global dimension theorem. But this is
not generally possible because the class of important modules usually does not behave well under this
interchange. The trouble comes from the self-referential nature of I. The R-module Q is an Rop -
module, pertinence will cause no difficulty, nor will good subsets cause difficulty (in general). But, we
need Ker (S) to be important in the sequence

0 −→ Ker (S) −→ F (S) −→ M −→ 0

if M is to be important, so we cannot get our hands on how to characterize importance “externally”

in the general case.

For the global dimension of R (that is, when I = R-mod itself) we must restrict attention to more special
rings than arbitrary augmented rings. Fix a commutative ring K and assume R is a K-algebra as in the
Hochschild Theory of Section 5.3. An obvious kind of cohomological dimension is then the smallest n so that
H n+1 (R, M ) = (0) for all Re (= R ⊗K Rop )-modules, M ; where the cohomology is Hochschild cohomology.
But, this is not a new notion because, by definition,

H r (R, M ) = ExtrRe (R, M ).

Hence, the Hochschild cohomological dimension is exactly projdimRe (R). Let us agree to write dimRe (R)
instead of projdimRe (R). It’s important to know the behavior of dimRe (R) under base extension of K as
well as under various natural operations on the K-algebra R. Here are the relevant results.

Proposition 5.83 Suppose R is projective over K and let L be a commutative base extension of K. Then

dim(L⊗K R)e (L ⊗K R) ≤ dimRe (R).

If L is faithfully flat over K, equality holds.

Remark: To explain the (perhaps) puzzling inequality of our proposition, notice that the dimension of R
is as a K-algebra while that of L ⊗K R is as an L-algebra as befits base extension. So we might have written
dimRe (R; K), etc. and then the inequality might not have been so puzzling–but, one must try to rein in
excess notation.
Proof . A proof can be based on the method of maps of pairs as given in Section 5.3, but it is just as simple
and somewhat instructive to use the associativity spectral sequence and associativity formula for Ext (cf.
5.5. THE KOSZUL COMPLEX AND APPLICATIONS 393

Proposition 5.61). To this end, we make the following substitutions for the objects, K, R, S, T, A, B, C of
that Proposition:

K −→ K, R −→ K, S −→ Re , T −→ L,
A −→ R, B −→ L, C −→ M (an arbitrary (L ⊗K R)e -module).

Since R is projective over K, the abstract hypothesis: A (our R) is R (our K)-flat is valid and moreover B
(our L) is T (again our L)-projective. Hence, the Ext associativity gives

ExtpRe (R, M ) ∼ p
= Ext(L⊗K R)e (L ⊗K R, M )

because S ⊗K T is equal to (L ⊗K R)e . Now M is an L and an Re -module; so, if p > dimRe (R) the left side
vanishes and therefore so does the right side. But, M is arbitrary and the inequality follows. (One could
also use Corollary 5.64).
We have an inequality simply because we cannot say that an arbitrary Re -module is also an L-module.
Now suppose L is faithfully flat as K-algebra, then L splits as K-module into K (= K · 1) q V so that we
i π
have a K-morphism π : L → K. The composition K ,→ L −→ K is the identity. If M is any Re -module,
then L ⊗K M is an Re and an L-module and we may apply our above Ext associativity to L ⊗K M . We find
the isomorphism
ExtpRe (R, L ⊗K M ) ∼ p
= Ext(L⊗K R)e (L ⊗K R, L ⊗K M ). (∗)
However, the composition

 i⊗1
/ L ⊗K M π⊗1
/ K ⊗K M = M
M = K ⊗K M

is the identity; so, applied to (∗) it gives

ExtpRe (R, M ) −→ ExtpRe (R, L ⊗K M ) −→ ExtpRe (R, M ) (∗∗)

whose composition is again the identity. If p > dim(L⊗K R)e (L ⊗K R) the middle group is (0) and so (∗∗)
shows ExtpRe (R, M ) = (0). M is arbitrary, therefore dimRe (R) ≤ dim(L⊗K R)e (L ⊗K R).
Of course, faithful flatness is always true if K is a field; so, we find

Corollary 5.84 If K is a field and R is a K-algebra, then for any commutative K-algebra, L, we have

dimRe (R) = dim(L⊗K R)e (L ⊗K R).

In particular, the notion of dimension is “geometric”, i.e., it is independent of the field extension.

Q
a pair of K-algebras, say R and S, then we get two new K-algebras R S and R ⊗K S.
If we’re given Q
Now, consider R S. It has the two projections Q pr1 and pr2 to R andQ S and so we get the two functors
pr1∗ and pr2∗ from R-mod (resp. S-mod) to R S – mod. If M is an R S-module, then we get two further
functors
Y
pr1 ∗ : R S – mod R-mod
Y
pr2 ∗ : R S – mod S-mod
Q
via M (1, 0)M (resp. (0, 1)M ). Observe that pri∗ (pri ∗ M ) is naturally an R S-submodule of M , therefore
we have two functors Y Y
F : R-mod S-mod R S – mod
394 CHAPTER 5. HOMOLOGICAL ALGEBRA

via
f (in R Q S – mod)
f) = pr1∗ M q pr2∗ M
F (M, M

and Y Y
G: R S – mod R-mod S-mod

via
G(M ) = (pr1 ∗ M, pr2 ∗ M );

the above shows that F and G establish an equivalence of categories

Y Y
R S – mod ≈ R-mod S-mod.

Q Q
If T = R S, then T e = Re S e ; so, applying the above, we get the category equivalence
Y Y
T e -mod = Re S e – mod ≈ Re -mod S e -mod.

Then, obvious arguments show that

Y
H p (R S, M ) ∼
= H p (R, pr1 ∗ M ) q H p (S, pr2 ∗ M )
Q
(where, M is an (R S)e -module). This proves the first statement of

Proposition 5.85 Suppose R and S are K-algebras and R is K-projective then

Y
dim(R Q S)e (R S) = max(dimRe (R), dimS e (S))

and
dim(R⊗S)e (R ⊗ S) ≤ dimRe (R) + dimS e (S).

Proof . For the second statement, we have the spectral sequence (of Corollary 5.64)

H p (R, H q (S, M )) =⇒ H • (R ⊗K S, M ).

Exactly the same arguments as used in the Tower Theorem (Theorem 5.2) for the Hochschild-Serre spectral
sequence finish the proof.

Remark: The K-projectivity of R is only used to prove the inequality for R ⊗K S.

We can go further using our spectral sequences.

Theorem 5.86 Suppose R is a K-projective K-algebra, then

gldim Re ≤ dimRe (R) + gldim(R).

Further,
gldim(R) ≤ dimRe (R) + gldim K

and
gldim(Rop ) ≤ dimRe (R) + gldim K.
5.5. THE KOSZUL COMPLEX AND APPLICATIONS 395

Proof . We apply the spectral sequence

E2p,q = H p (R, ExtqT (B, C)) =⇒ Ext•Rop ⊗K T (B, C)

of Corollary 5.62. Here, B and C are left R and right T -modules (and, in the Ext• of the ending, they are
viewed as right Rop ⊗K T -modules, or left R ⊗K T op -modules). We set T = R and see that

ExtqR (B, C) = (0) when q > gldim(R).

But,
H p (R, −) = ExtpRe (R, −)
and so
H p (R, −) = (0) when p > dimRe (R).
Therefore, E2p,q
= (0) when p + q > dimRe (R) + gldim(R). Once again, exactly as in the Tower Theorem,
we conclude ExtnR⊗K Rop (B, C) vanishes for n > dimRe (R) + gldim(R). This proves the first inequality.
For the second and third inequalities, we merely set T = K. Then, ExtqK (B, C) vanishes for all
q > gldim(K) and H p (R, −) vanishes for all p > dimRe (R) = dimRe (Rop ). Our spectral sequence argument
now yields the two desired inequalities.

Corollary 5.87 If R is a projective K-algebra and R is semi-simple as K-algebra, then

gldim(Re ) = dimRe (R).

If R is arbitrary but K is a semi-simple ring, then

gldim(R) ≤ dimRe (R)

and
gldim(Rop ) ≤ dimRe (R).

Proof . In the first inequality, gldim(R) = 0, so

gldim(Re ) ≤ dimRe (R).

The opposite inequality is always true by definition.

If now K is semi-simple, R is automatically K-projective; so, our other inequalities (of the theorem)
finish the proof as gldim(K) = 0.

Corollary 5.88 If K is semi-simple and R is a K-algebra, then Re is semi-simple if and only if R is a

projective Re -module (i.e., dimRe (R) = 0).

Proof . Suppose dimRe (R) = (0). Then, by Corollary 5.87 above, gldim(R) = 0, i.e., R is itself a semi-
simple ring. But then we apply the corollary one more time and deduce gldim(Re ) = 0. Conversely, if
gldim(Re ) = 0, then dimRe (R) = 0.

Corollary 5.89 If K is semi-simple and if Re is semi-simple, then R is semi-simple (as K-algebra).

Proof . As gldim(Re ) = 0 and K is semi-simple, we get dimRe (R) = 0. But, gldim(R) ≤ dimRe (R); so, we
are done.
We can now put together the Koszul complex and the material above to prove
396 CHAPTER 5. HOMOLOGICAL ALGEBRA

Theorem 5.90 (Global Dimension Theorem) Suppose K is a commutative ring and write
R = K[T1 , . . . , Tn ]. Then,
dimRe (R) = dimR (K) = n.
We have the inequality
n ≤ gldim R ≤ n + gldim K,
and so if K is a semi-simple ring (e.g., a field), then

gldim R = n.

Proof . By the main application of the Koszul complex to dimension (Theorem 5.72) we have dimR (K) = n
and so gldim R ≥ n. Here, K is an R-module via sending all Tj to 0. But if e is any K-algebra map R −→ K,
we can perform the automorphism Tj 7→ Tj − e (Tj ) and this takes e
 to the usual augmentation in which all
Tj −→ 0. Therefore, we still have dimR K = n (and gldim R ≥ n) when viewing K as R-module via e .
Now Rop = R; so, Re = R ⊗K R, and thus

Re = K[T1 , . . . , Tn , Z1 , . . . Zn ] = R[Z1 , . . . , Zn ].

(Remember that Tj stands for Tj ⊗ 1 and Zj for 1 ⊗ Tj ). The standard augmentation η : Re → R is given
by ρ ⊗ ρe 7→ ρe
ρ and it gives a map
Re = R[Z1 , . . . , Zn ] −→ R,
in which Zj goes to Tj ∈ R. The Zj ’s commute and we can apply Theorem 5.72 again to get

dimRe (R) = n.

Finally, Theorem 5.86 shows that

gldim(R) ≤ n + gldim K.

Remark: The global dimension theorem is a substantial improvement of Hilbert’s Syzygy Theorem. For
one thing we need not have K a field (but, in the semi-simple case this is inessential) and, more importantly,
we need not restrict to graded modules. Also, the role of the global dimension of K becomes clear.
5.6. CONCLUDING REMARKS 397

5.6 Concluding Remarks

The apparatus of (co)homological methods and constructions and, more importantly, their manifold appli-
cations to questions and situations in algebra and geometry has been the constant theme of this chapter.
Indeed, upon looking back to all earlier chapters from the first appearance of group cohomology as a compu-
tational tool to help with group extensions through the use of sequences of modules and Galois cohomology
in field theory to the theory of derived functors and spectral sequences to obtain new, subtle invariants in
algebra and geometry, we see a unified ever deepening pattern in this theme. The theme and pattern are a
major development of the last sixty years of the twentieth century—a century in which mathematics flowered
as never before. Neither theme nor pattern gives a hint of stopping and we have penetrated just to middling
ground. So read on and work on.

5.7 Supplementary Readings

The classic reference on homological algebra is Cartan and Eilenberg [9]. One may also consult Mac Lane
[36], Rotman [44], Weibel [48], Hilton and Stammback [24], Bourbaki [5], Godement [18] and Grothendieck
[20]. For recent developments and many more references, see Gelfand and Manin’s excellent books [16, 17].
For a global perspective on the role of homological algebra in mathematics, see Dieudonné [10].
398 CHAPTER 5. HOMOLOGICAL ALGEBRA
Bibliography

[1] Emil Artin. Galois Theory. Dover, second edition, 1964.

[2] Michael Artin. Algebra. Prentice Hall, first edition, 1991.
[3] M. F. Atiyah and I. G. Macdonald. Introduction to Commutative Algebra. Addison Wesley, third edition,
1969.
[4] Nicolas Bourbaki. Algèbre, Chapitres 1-3. Eléments de Mathématiques. Hermann, 1970.
[5] Nicolas Bourbaki. Algèbre, Chapitre 10. Eléments de Mathématiques. Masson, 1980.
[6] Nicolas Bourbaki. Algèbre, Chapitres 4-7. Eléments de Mathématiques. Masson, 1981.
[7] Nicolas Bourbaki. Algèbre Commutative, Chapitres 8-9. Eléments de Mathématiques. Masson, 1983.
[8] Nicolas Bourbaki. Elements of Mathematics. Commutative Algebra, Chapters 1-7. Springer–Verlag,
1989.
[9] Henri Cartan and Samuel Eilenberg. Homological Algebra. Princeton Math. Series, No. 19. Princeton
University Press, 1956.
[10] Jean Dieudonné. Panorama des matématiques pures. Le choix bourbachique. Gauthiers-Villars, second
edition, 1979.
[11] David S. Dummit and Richard M. Foote. Abstract Algebra. Wiley, second edition, 1999.
[12] Albert Einstein. Zur Elektrodynamik bewegter Körper. Annalen der Physik, 17:891–921, 1905.
[13] David Eisenbud. Commutative Algebra With A View Toward Algebraic Geometry. GTM No. 150.
Springer–Verlag, first edition, 1995.
[14] Jean-Pierre Escofier. Galois Theory. GTM No. 204. Springer Verlag, first edition, 2001.
[15] Peter Freyd. Abelian Categories. An Introduction to the theory of functors. Harper and Row, first
edition, 1964.
[16] Sergei I. Gelfand and Yuri I. Manin. Homological Algebra. Springer, first edition, 1999.
[17] Sergei I. Gelfand and Yuri I. Manin. Methods of Homological Algebra. Springer, second edition, 2003.
[18] Roger Godement. Topologie Algébrique et Théorie des Faisceaux. Hermann, first edition, 1958. Second
Printing, 1998.
[19] Daniel Gorenstein. Finite Groups. Harper and Row, first edition, 1968.
[20] Alexander Grothendieck. Sur quelques points d’algèbre homologique. Tôhoku Mathematical Journal,
9:119–221, 1957.

399
400 BIBLIOGRAPHY

[21] Alexander Grothendieck and Jean Dieudonné. Eléments de Géométrie Algébrique, IV: Etude Locale
des Schémas et des Morphismes de Schémas (Première Partie). Inst. Hautes Etudes Sci. Publ. Math.,
20:1–259, 1964.

[22] Marshall Hall. Theory of Groups. AMS Chelsea, second edition, 1976.

[23] David Hilbert. Die Theorie der algebraischen Zahlkörper. Jahresberich der DMV, 4:175–546, 1897.

[24] Peter J. Hilton and Urs Stammbach. A Course in Homological Algebra. GTM No. 4. Springer, second
edition, 1996.

[25] G. Hochschild. On the cohomology groups of an associative algebra. Ann. of Math. (2), 46:58–67, 1945.

[26] G. Hochschild. On the cohomology theory for associative algebras. Ann. of Math. (2), 47:568–579, 1946.

[27] Thomas W. Hungerford. Algebra. GTM No. 73. Springer Verlag, first edition, 1980.

[28] Nathan Jacobson. Basic Algebra II. Freeman, first edition, 1980.

[29] Nathan Jacobson. Basic Algebra I. Freeman, second edition, 1985.

[30] Daniel M. Kan. Adjoint functors. Trans. Amer. Math. Soc., 87:294–329, 1958.

[31] Irving Kaplansky. Fields and Rings. The University of Chicago Press, second edition, 1972.

[32] Jean Pierre Lafon. Les Formalismes Fondamentaux de l’ Algèbre Commutative. Hermann, first edition,
1974.

[33] Jean Pierre Lafon. Algèbre Commutative. Languages Géométrique et Algébrique. Hermann, first edition,
1977.

[34] Serge Lang. Algebra. Addison Wesley, third edition, 1993.

[35] Saunders Mac Lane. Categories for the Working Mathematician. GTM No. 5. Springer–Verlag, first
edition, 1971.

[36] Saunders Mac Lane. Homology. Grundlehren der Math. Wiss., Vol. 114. Springer–Verlag, third edition,
1975.

[37] Saunders Mac Lane and Garrett Birkhoff. Algebra. Macmillan, first edition, 1967.

[38] M.-P. Malliavin. Algèbre Commutative. Applications en Géométrie et Théorie des Nombres. Masson,
first edition, 1985.

[39] H. Matsumura. Commutative Ring Theory. Cambridge University Press, first edition, 1989.

[40] Barry Mitchell. The full embedding theorem. Amer. J. Math., 86:619–637, 1964.

[41] Patrick Morandi. Field and Galois Theory. GTM No. 167. Springer Verlag, first edition, 1997.

[42] John S. Rose. A Course on Group Theory. Cambridge University Press, first edition, 1978.

[43] Joseph J. Rotman. The Theory of Groups, An Introduction. Allyn and Bacon, second edition, 1973.

[44] Joseph J. Rotman. An Introduction to Homological Algebra. Academic Press, first edition, 1979.

[45] Jean-Pierre Serre. Corps Locaux. Hermann, third edition, 1980.

[46] Jean-Pierre Serre. Local Algebra. Springer Monographs in Mathematics. Springer, first edition, 2000.
BIBLIOGRAPHY 401

[47] B.L. Van Der Waerden. Algebra, Vol. 1. Ungar, seventh edition, 1973.

[48] Charles A. Weibel. Introduction to Homological Algebra. Studies in Advanced Mathematics, Vol. 38.
Cambridge University Press, first edition, 1994.
[49] Ernst Witt. Zyklisch Körper und Algebren der Charakteristik p vom Grade pm . J. Reine Angew. Math.,
176:126–140, 1937.

[50] Oscar Zariski and Pierre Samuel. Commutative Algebra, Vol I. GTM No. 28. Springer Verlag, first
edition, 1975.
[51] Oscar Zariski and Pierre Samuel. Commutative Algebra, Vol II. GTM No. 29. Springer Verlag, first
edition, 1975.
[52] Hans J. Zassenhaus. The Theory of Groups. Dover, second edition, 1958.
Index

M -regular sequence, 381 Artin Schreier

I-Global Dimension Theorem, 390 Theorem I, 292, 293
(Co)homological Functors, 327 Theorem II, 292
(N), 2 Artin’s irreducibility criterion, 294–296
(co)homology of algebras, 331 Artin’s theorem, 44, 261, 263, 265, 268, 271
Artin’s theorem of the primitive element, 270
enveloping algebra, 332 Artin, E., 294
Artin–Rees theorem, 238, 239
a-adic completion of A, 29 artinian module, 116
a-adic topology, 29 ascending chain condition for module, 116
abelian category, 24 associated primes of module, 228
AC, 298 atoms, 55
ACC for module, see ascending chain condition for augmentation, 338
module augmentation ideal, 334, 338
acyclic complex, 310 augmentation module, 386
acyclic resolution of modules, 128 augmented ring, 386
additive functor, 125 k-automorphism of field extension, 262
additive group functor, 96 axiom of choice, 66, 185
additive Hilbert 90, 294
A-adic topology, 237 backward image functor, 175
adjoint functors, 339 Baer embedding theorem, 20, 133, 135
admissible monomial, 113 Godement’s proof, 133
admissible word, 105 Baer radical of group, 2
affine group, 84 Baer representation theorem, 131, 133
Akizuki’s structure theorem, 195, 240 Baer’s theorem, 3
false for noncomm. rings, 196 Baer, R., 135
Akizuki, Y., 194 bar resolution, 334
algebraic closure, 298 base extension of module to algebra, 169
algebraic closure of field, 247 Basic Existence Theorem, 298
algebraic element over commutative ring, 115 bounded complex, 310
algebraic geometry, 199 above, 310
algebraically closed, 298 below, 310
algebraically closed field, 247 Brauer, R., 52
algebraically dependent set over commutative ring, bundle homomorphism of vector bundles, 18
115 Burnside basis theorem, 2, 72, 74
algebraically independent set over commutative ring, Burnside dimension, 74
115 Burnside’s Lemma, 55
alternating R-algebra, 154 Burnside, W., 65, 88
amalgamated product of groups, 106 butterfly lemma, see Zassenhaus’ butterfly lemma
annihilator of submodule, 116
arrow of category, see morphism of category Cameron-Cohen Inequality, 55
Artin ordering, 156 canonical flasque resolution, 351

402
INDEX 403

canonical map to localization, 174 cohomological spectral sequence, 356

canonical site, 32 cohomologous 2-cocycles on group, 79
Cartan–Eilenberg injective resolution of complex, 368 cohomology
casus irreducibilis of cubic, 45 long exact sequence, 277
category, 95 of a cyclic group, 280
examples, 95 of presheaves, 344
category co-over object, 101 of sheaves, 344
category over object, 101 cohomology cofunctor, 97
Cauchy, A., 61 cohomology group of R with coefficient in M , 334
Cayley graph, 7 0-th cohomology group of group
Cayley transform, 53 calculation, 86
Cayley–Hamilton theorem, 237 n-th cohomology group of group, 86
center of group, 69 cohomology of groups, 371
central group extension, 82 comaximal ideals, 192
centralizer of group element, 59 comma category, see category over or co-over object
chain, 66 commutator group of group, 72
chain detour lemma, 241, 242 commutator of group elements, 69
chain map, 310 compatible filtration and coboundary map, 355
d-chain on manifold, 153 compatible filtration and grading, 355
character, 354 complementary index of spectral sequence, 356
character of representation, 260 completely reducible module, 385
characteristic class of extension of module by module, S-component of submodule, 235
329 composition factors of composition series, 66
characteristic of ring, 11 composition of morphisms between objects of cate-
characteristic subgroup of group, 71 gory, 95
Chevalley, 214 composition series for group, 66
Chinese Remainder Theorem, 192, 195 compositum of two fields, 271
classical, 192 conductor, 221
classification of groups of order pq, 85 of the integral closure, 221
closed set in Spec, 181 conjugacy class of group element, 59
n-th coboundary group of group, 86 H-conjugacy class of group element, 59
n-th coboundary map, group cohomology, 86 k-conjugate fields, 257
coboundary of 1-cochain on group, 79 conjugation group action, 59
n-th cochain group of group, 86 connecting homomorphism, 277
cochain map, 310 connecting homomorphism in LES for (co)homology,
1-cochain on group, 79 319
2-cochain on group, 78 connecting homomorphism in snake lemma, 137
n-th cocycle group of group, 86 constant presheaf with values in object, 24
2-cocycle on group, 78 continuous functor vector spaces → vector spaces, 17
cofinal subset of index set, see final subset of index contracted ideal, 176
set contraction of tensors, 147
cofunctor, 96 contragredient tensor of rank b, 147
cogredient tensor of rank a, 147 convolution product in polynomial ring, 110
Cohen’s lemma, 184, 185 coprimary module, 223
Cohen, I. S., 184 coproduct in category, 102
Cohen-Seidenberg, 207 examples, 103–106
I: Lying over Theorem, 207 Gr, 105
II: Going-up Theorem, 208 Mod, 103
III: Going-Down Theorem, 209 Sets, 103
III: Going-down Theorem, 210 coproduct of family of groups, 8
cohomological complex, 310 cotangent bundle of manifold, 153
404 INDEX

cotangent space to manifold at point, 152 Eisenbud, 218

counting lemma, 60 elementary d-chain in manifold, 153
covering family, 25 elementary abelian p-group, 72
covering of object of category, 32 elementary symmetric functions, 209
covering surjection, see minimal surjection empty word, 105
crossed homomorphism, 54, see twisted homomor- end of spectral sequence, 356
phism epimorphism of vector bundles, 18
cup-product, group cohomology, 5 equivalent categories, 98
equivalent group extensions, 75
décalage, 328 essential injection, 130
DCC for module, see descending chain condition for essential primes of submodule in module, 228
module essential submodule, see large submodule
Dedekind domain, 229 essential surjection, see minimal surjection
examples, 229 étale algebra over comm. ring, 252
Dedekind ring, 129 Euler function, 121
Dedekind’s theorem, 260–262 exact functor, 125
Dedekind, R., 229 examples, 125
degenerate spectral sequence, 364 exact sequence of groups, 75
degree of field extension, 244 exact sequence of vector bundles, 19
degree of filtration, 355 ExtnA (·, B), 318
∂-functor, 321 extended ideal, 176
δ-functor, 321 extension lemma, 256–259
derivation of algebra with values in algebra, 245 General, 284
examples, 245 extension property for injective modules, 131
derived functor, 317 exterior algebra of module, 150
j-th derived group of group, 73 j-th exterior power of module, 150
derived length of group, 74
derived series of group, 74 f.f., see faithfully flat module
descending chain condition for module, 116 f.g., see finitely generated module
descent, 170, 338 f.p., see finitely presented module
descent of property in extension, 170 Fact I
determinant of linear map, 154 cohomology of groups, 277
dimension shifting, 328 Fact II
direct image of sheaf, 198 cohomology of groups, 277
direct limit, see right limit faithful module, 188
direct mapping family, see right mapping family faithfully flat module, 155
directed set, 156 Feit, W., 65, 88
examples, 156 Fermat prime, 3
discrete valuation on ring, 39 Fermat’s little theorem, 254
discrete valuation ring, 39 fibre, 214
divisible abelian group, 8 fibred coproduct over object, 102
divisible module, 132 examples, 103–106
dual category, 96 Gr, 106
dual module, 354 Mod, 104
DVR, see discrete valuation ring Sets, 104
fibred product over object, 102
E. Noether, 216 examples, 103–106
E. Noether: finiteness Theorem, 220 Mod, 104
Eckmann, B., 135 Sets, 104
Eilenberg, 54 filtered complex, 355
Eilenberg, S., 87, 323 filtration index of spectral sequence, 356
INDEX 405

final subset of index set, 160 generation of submodule by set, see submodule gen-
finite atom, 55 erated by set
finitely generated (f.g) module geometry, 199
need not be f.p., 120 germ of function, 12
finitely generated algebra, 115 global dimension (of a ring), 384
finitely generated group, 3 Global Dimension Theorem, 396
finitely generated ideal (FGI) test, 165 Godement, 133
finitely generated module, 115 Godement resolution, 351
finitely presented module, 116 good free module, 147
finiteness Theorems, 220 good subset, 388
first homomorphism theorem, 182 graded ring, 149
first uniqueness theorem for primary decomp., 228, Grothendieck, 220, 369
230, 231 Spectral sequence of composed functors, 369
Fitting’s lemma, 223 Ω-group, 92
five lemma, 136, 143, 167, 190, 323 examples, 92
fixed field of family of characters, 261 p-group, 61
flat module over ring, 155 group action (left), 57
flat op-module over ring, 155 examples, 59
forgetful functor, 96 group cohomology, 277
forward image functor, 175 group extension, 75
fraction field of comm. ring, 176 examples, 82–85
fraction ring of comm. ring, see localization of comm. group ring, 330
ring
fractional ideal, 37 half-exact functor, 125
fractional linear transformation, 6 Hamel basis, 67, 68
Frattini argument, 90 application to R/Q, 68
Hausdorff maximal principle, 66
Frattini subgroup, 71
height of prime ideal, 180
free group on set, 106
Hensel, K., 238
free module on set, 115
Henselization, 54
free product of groups, 106
Herbrand’s Lemma, 54
free resolution of module, 128
Herstein’s lemma, 236
Frobenius, G., 65
proof using Artin–Rees, 239
full subcategory, 96
Hilbert basis theorem
functor, 96
Noether’s argument, 122
examples, 96–97
original, 123
may have right adjoint, no left adjoint, 100
Hilbert Syzygy Theorem, 391
fundamental theorem of Galois Theory, 264, 271
Hilbert Theorem 90
general case, 286 cohomological version, 280
original form, 282
Galois cohomology groups, 280 Hochschild, 334
Galois equivalent field extensions, 263 cohomology groups, 334
Galois group homology groups, 334
with Krull topology, 285 Hochschild groups, 387
Galois group of field extension, 262 holomorphic geometry, 199
Galois group of polynomial, 262 homogeneous ideal, 149
Galois, E., 69 homogeneous space for group, 275
Γ-Riccati equation, 11 homological complex, 310
Gauss, K., 45 homology functor, 97
Generalized Nakayama’s Lemma, 389 homology group of R with coefficients in M , 334
generation of group by set, 107 homology groups of G with coefficients in A, 331
406 INDEX

homothety Jacobian criterion for multiplicity, 245, 248

example, 223 Jacobson radical, 131, 182
homotopy, 314 Jacobson radical of group, 2
homotopy functor, 97 Jordan’s Theorem, 55
homotopy invariance, 314 Jordan–Hölder theorem, 92, 121
Hurewicz map, 97
Kaplansky, 55
immediate equivalence, 104 Kaplansky’s Theorem, 56
important module, 387 Koszul complex, 378
indecomposable module, 11 duality isomorphisms, 380
independent characters, 260 Koszul duality, 379, 380
independent transcendentals over commutative ring, Kronecker’s theorem of the primitive element, 270–
see algebraically independent set over com- 272
mutative ring Krull
index of filtration, 355 Galois theory, 284
inductive limit, see right limit Krull dimension of comm. ring, 180
inductive mapping family, see right mapping family examples, 180–181
inductive poset, 66 dim = 0, 180
injective dimension, 384 dim = 1, 180
injective hull of module, 135 dim = n, 181
injective module, 126 Krull height theorem, 241
inner twisted homomorphism, see principal twisted converse, 242
homomorphism Krull intersection theorem, 236, 237, 240
inseparable field extension, 248 original, 237
integral, 203 Krull principal ideal theorem, 239, 241, 242
basis, 215 Krull topology, 285
closure, 204 Krull, W., 239
dependence, 203 Kummer, 289
dependence, transitivity of, 204 Pairing theorem, 290
morphism, 203 Kummer theory, see prime cyclic Kummer theory
over (Definition), 203 Kummer’s theorem, 295, see prime cyclic Kummer
integral element over domain, 12 theory
integrally closed, 204 Kummer, E., 229
integrally closed domain, 12
integrally closed domain in algebra, 12 Lagrange’s cubic resolvent, 45
integrally closed ring, 204 Lagrange’s theorem
inverse limit, see left limit converse false, 60
inverse mapping family, see left mapping family large submodule, 130
invertible module, 201 Lasker-Noether decomposition theorem, 226, 228
irreducible element of ring, 36 Lasker’s original, 227
irreducible set, 234 Noether’s original, 227
isolated associated prime ideals of module, 228 LCS of group, see lower central series of group
isolated essential prime ideals of submodule in mod- left acyclic resolution, 310
ule, 228 left adjoint functor, 100
isolated primary components of module, 228 left derived functors of cofunctor, 317
isolated primary components of submodule in mod- left derived functors of functor, 317
ule, 228 left limit, 157
isolated prime ideal of ideal, 183 examples, 160–162
isomorphic categories, 98 left mapping family, 156
isomorphic normal flags, 92 examples, 157
isomorphism between objects of category, 96 left-exact cofunctor, 125
INDEX 407

left-exact functor, 125 minimal polynomial of algebraic elt., 248

lemma (L), 357–360 minimal surjection, 130
lemma(L), 359 module
length of chain of prime ideals, 180 Map(B, G), 278
length of module, 121 divisible, 352
level of spectral sequence, 356 G-module, 59
lifting property for projective modules, 127 Ω-module, 92
line bundle, see invertible module module of relative Kähler differentials, 250
linear representation of group, 59 examples, 250–252
local embedding lemma, 62, 63 monomial, 111, 113
local flatness criterion, 186 monomorphism of vector bundles, 18
local property, 186 Moore’s theorem, 274
local ring, 10, 179 Moore, E.H., 274
local ringed space, 198 Moore–Smith property, see directed set
examples, 198 Morita equivalence, 385
localization of comm. ring, 174 morphism of category, 95
examples, 175 morphism of functors, see natural transformation of
localization of comm. ring at prime ideal, 179 functors
represents germs of ‘some kind’, 191–192 morphisms between objects of category, 95
localizing subcategory, 121 multiplicative group functor, 97
locally P , see local property multiplicative subset in comm. ring, 173
locally (P) group, 4 examples, 173
locally standard LRS, 199 multiplicity of root of polyn., 244
locally trivial vector space family, 16
locally-free OX -module, 200 (N), 70, 72, 73, 90
long (co)homology sequence lemma, 319 Nagata, 215, 218, 220
long exact (co)homology sequence lemma, 320 rings, 220
loop space of space, 108 rings (Definition), 220
functor is right-adjoint to suspension functor, 108 Nagata, M., 194
lower central series of group, 89 Nakayama’s lemma, 187, 189, 202, 233, 240
lower star, 339 classical, 187
LRS, see local ringed space natural irrationalities, 272
natural irrationalities theorem, 42, 271, 295
Mac Lane, 216 interpretation, 272
Mac Lane I, 252, 267, 304 original, 272
interpretation, 253 natural transformation of functors, 97
Mac Lane II, 252, 254 examples, 97
interpretation, 253 on Vect(k), 98
Mac Lane’s theorem, 307 Newton’s Identities, 214
Mac Lane, S., 87 Nielson, J., 107
Main theorem on separability, 252 nilpotence class of group, 90
counter example when K/k is not finite, 255 nilpotent group, 90
map of pairs, 339 nilradical of comm. ring, 181
Maschke’s Theorem, 48, 385 Noether Normalization Lemma, 216
maximal M -regular sequence, 381 Noether’s proposition, 225, 226
maximal condition for module, 116 Noether, E., 225
maximal idempotent of ring, 11 Noetherian induction, 194, 226
maximal spectrum of comm. ring, 181 noetherian module, 116
minimal condition for module, 116 subring of noetherian ring need not be noethe-
minimal injection, see essential injection rian, 120
minimal polynomial, 209 non-generator in group, 71
408 INDEX

nonrepetitious normal flag, 92 presheaf of germs, 24

norm, 209, 280 presheaf on topological space with values in category,
of an extension, 281 24
G-norm, 84 primary ideal, 223
norm map p-primary submodule, 224
definition, 279 primary submodule in module, 224
normal basis theorem, 47, 272, 281 prime avoidance lemma, 184, 241
algebraic interpretation, 274–275 prime element of ring, 36
geometric interpretation, 275–276 prime ideal, 179
normal chain, see normal flag Prime Number Theorem, 54
normal closure of field extension, 259 prime spectrum of comm. ring, 181
normal domain, see integrally closed domain almost never Hausdorff, 183
normal field extension, 257 primitive element, 270
counter-example to transitivity, 258 primitive element theorem, see Artin’s such or Kro-
non sequiturs, 258 necker’s such
normal flag, 92 primitive root of unity, 289
normal ring, 204 principal homogeneous space for group, 275
normal series, see normal flag principal idempotent of ring, see maximal idempo-
Ω-normal subgroup, 92 tent of ring
normality, 204 principal twisted homomorphism, 86
Noetherian domain, 206 product in category, 102
normalized 2-cochain on group, 82 examples, 103–106
normalizer of set in group, 62 Gr, 105
number field, 37, 229 Mod, 103
Sets, 103
object of category, 95 profinite group, 163
open set in Spec, 181 profinite groups, 372
opposite category, see dual category projective cover, 130
orbit space, 58 counter-example, 130
orbit under group action, 58 projective dimension, 384
orthogonal idempotents, 200 projective limit, see left limit
projective mapping family, see left mapping family
partial order on set, 66 projective module, 126
partially ordered set, see poset product of such need not be such, 20
Pass, 205 projective resolution of module, 128
perfect field, 248, 305, 307 proper map, 41
examples, 248 property (N), see (N)
pertinent module, 388 pseudo-metric topology, 237
PHS for group, see principal homogeneous space for pull-back of forms, 153
group pullback functor for vector space families, 16
Picard group of comm. ring, 202 purely inseparable degree of field extension, 255
polynomial solvable by auxiliary chain of equations, purely inseparable elt. over field, 249
43 purely transcendental, 304
Pontrjagin dual, 290, 292, 293
poset, 66 quasi-coherent OX -module, 200
power lemma, 231, 236 quasi-compact space, 183
generalization, see Herstein’s lemma
premorphism, 309 radical of ideal, 182
presentation of group, 107 rank of quasi-coherent OX -module, 200
presentation of module, 124 rank of vector bundle, 16
presheaf, 344 rank one group, 4
INDEX 409

rational function field, 176 separable element over field

C k real geometry, 199 provisional defn., 248
reduced primary decomposition, 226 separable extension, 214
reduced ring, 182 separable field extension, 304, 306
refinement of normal flag by normal flag, 92 provisional defn., 248
regular element of comm. ring, 177 separable irreducible polynomial
regular filtration, 356 provisional defn., 248
regular local ring, 392 separable polynomial
regular sequence, 381 provisional defn., 248
related fields, 256 separably generated field extension, 304, 306
relative radical of submodule in module, 223 separating transcendence base, 304
representable functor, 99 Serre, 55
examples, 99 sheaf, 344
uniqueness of representing pair, 99 Čech cohomology, 347
K-representation of group, 260 cohomology, 346
residual quotient of submodule by set, see transporter direct image, 345
of set to submodule flasque, 347
resolution, 310 sheaf Hom, 373
injective, 310 sheaf of discontinuous sections, 351
projective, 310 sheaf of local rings, 197
restriction of vector space family, 16 sheaf of sets, 25
right acyclic resolution, 310 sheafification #, 344
right adjoint functor, 100 short exact sequence of groups, 75
right derived functors of cofunctor, 317 Sierpinski, 54
right derived functors of functor, 317 simple group, 64
right limit, 157 simple module, 121
examples, 160–162 site, 32
right mapping family, 156 skew field, 389
examples, 157 SMA, I, see sufficiently many automorphisms, I
right-exact cofunctor, 125 SMA, II, see sufficiently many automorphisms, II
right-exact functor, 125 small commutative diagram, 316
ring of formal power series over field, 29 small normal subgroup, 2
ring of integers in number field, 229 small radical of group, 2
ringed space, 372 small submodule, 130
Rudakov, A., 13 smooth algebra over comm. ring, 252
snake lemma, 137, 143, 168, 189, 319
S-saturation of submodule, 178 solvable group, 74
Schanuel’s lemma, 14, 124 Galois’ sense, 69
Schmidt, F.K., 307 space of 1-forms on manifold at point, see cotangent
Schopf, A., 135 space to manifold at point
Schreier refinement theorem, 92, 93 spectral sequence
Schreier, O., 107 Čech cohomology, 373
second cohomology group of group, 79 Associativity for EXt and Tor, 374
second homomorphism theorem, 70, 91, 122, 135, 232 Ext, 376
second uniqueness theorem for primary decomp., 228 Hochschild-Serre, 371
section, 338 Leray, 372
section of vector bundle over vector bundle, 16 Local to global Ext, 373
semi-local ring, 36 Tor, 376
semi-simple ring, 385 Spectral sequences, 355
separable closure of field in extension, 255 spectral topology, see Zariski topology
separable degree of field extension, 255 split exact sequence, 126
410 INDEX

split group extension, 82 Taylor’s theorem, 272

splitting field arising from successive solution of chain tensor algebra of module, 149
of equations, 43 tensor field, 147
splitting field for polyn., 256 tensor product of two modules over ring, 140
splitting of exact sequence, 126 computations, 142–147
stabilizer under group action, 58 Thompson, J., 65, 88
stalk of a (pre)sheaf, 344 tilde construction from module, 196
stalk of presheaf at point, 197 topological ring, 31
standard model topologically nilpotent element in topological ring, 31
examples, 199 TorAn (·, B), 318
standard resolution, 334, 343 Tor dimension, 386
Steinitz, E., 298 n-torsion element of group, 4
Axioms for Σ, 301 torsion free group element, 4
Exchange Lemma, 301 total complex, 367
Existence of a transcendence basis, 301 total differential, 367
Existence of an algebraic closure, 299 total fraction ring, 204
General Exchange Lemma, 302 total fraction ring of comm. ring, 176
stripping functor, see forgetful functor total grading index of spectral sequence, 356
strongly convergent spectral sequence, 360 Tower Theorem, 372
sub-bundle of vector bundle, 18 trace, 209, 280
subcategory, 95 of an extension, 281
examples, 96 transcendence basis, 299
subextension of field extension, 259 transcendence degree, 215, 304
Ω-subgroup, 92 transcendental element over commutative ring, 115
submodule generated by set, 115 transcendental extensions, 298
sufficiently many automorphisms, I, 258, 263, 265 translation group action, 59
sufficiently many automorphisms, II, 259, 266, 268 transporter of set to submodule, 116
sum of submodules, 115 trivial group action, 59
superfluous submodule, see small submodule twisted homomorphism, 86
support of module, 185 two-sided ideal, 115
suspension of space, 108 Tychonov’s theorem, 163
functor is left-adjoint to loop space functor, 108 type of group extension, 76
Sylow classification theorem, see Sylow theorem II
Sylow existence theorem, see Sylow theorem I UCS of group, see upper central series of group
p-Sylow subgroup, 61 uniqueness I for δ-functors, 322
of group of order pq, 64 uniqueness II for δ-functors, 323, 328
of group of order pqr, 64 uniqueness III for δ-functors, 324
Sylow theorem I, 60, 62 universal δ-functor, 321
original, 61 isomorphism at n = 0 lifts, 321
Sylow theorem II, 62, 64, 90 universal ∂-functor, 321
Sylow theorem III, 64, 70 universal mapping property
modified, 90, 91 left limits, 158
Sylow, L., 60 localization, 178
symbolic power of prime ideal, 239 polynomial rings, 110
symmetric algebra of module, 150 products, 102
j-th symmetric power of module, 150 right limits, 158, 162, 163
Syzygy, 391 tensor products, 163
Theorem, 391 universal, effective epimorphism, 32
universally Japanese rings, 220
tangent bundle of manifold, 153 upper central series of group, 89
tangent space to manifold at point, 152 upper star, 339
INDEX 411

Urysohn’s lemma, 183

van Kampen’s theorem, 106

variety defined by ideal, 181
vector bundle, 16
vector space family over topological space, 16

weakly convergent spectral sequence, 360

Wederburn, 48
weight function on domain, 36
well ordered chain, 66
Wielandt, H., 60
Witt, 293
ring of Witt vectors, 293
words over alphabet, 105
wreath product, 84

Yoneda, 48
Yoneda’s embedding lemma, 99, 113, 141, 253
interpretation, 99

Zariski topology, 181, 213

Zassenhaus’ butterfly lemma, 93, 94
Zassenhaus, H., 93
Zermelo well ordering principle, 66
Zorn’s lemma, 66, 67, 132, 135, 179, 182, 185, 253