Theory of Commutative Fields - Masayoshi Nagata

Translations of Mathematical Monographs 125
Theory of
Commutative Fields
Translations of
MATHEMATICAL
MONOGRAPHS
Volume 125
Theory of
Commutative Fields
Masayoshi Nagata
Translated by
Masayoshi Nagata
American Mathematical Society

Y
Providence, Rhode Island
"J J7C p
KAKANTAI RON (Theory of Commutative Fields)
by Masayoshi Nagata
Copyright 1967, 1985 by Masayoshi Nagata and Shokabo Publishing Co., Ltd.
Originally published in Japanese by Shokabo Publishing Co., Ltd.,
Tokyo in 1967 and 1985.
Translated from the Japanese by Masayoshi Nagata
1991 Mathematics Subject Classification. Primary 12-01;
Secondary 12D 15, 12F 10, 12F20.
ABSTRACT. This book presents various topics in the theory of commutative fields. After some
preliminaries on set theory, we study, in Chapter I, basic properties of groups, rings, and fields.
Chapter II contains the theory of algebraic extensions of finite degree, including Galois theory.
Transcendental extensions are investigated in Chapter III. The theory of valuations is explained in
Chapter IV. The theory of ordered fields and topics related to Hilbert's 17th Problem are introduced
in Chapter V. Additional topics including Galois theory of algebraic extensions of infinite degree
are stated in the final chapter.
Library of Congress Cataloging-in-Publication Data

Nagata, Masayoshi, 1927-
[Kakantairon. English]
Theory of commutative fields/ Masayoshi Nagata: translated by Masayoshi Nagata.
p. cm.-(Translations of mathematical monographs; v. 125)
Includes bibliographical references and index.
ISBN 0-8218-4572-1
1. Fields, Algebraic 2. Commutative law (Mathematics) I. Title. II. Series.
QA247.N2613 1993 93-6503
512'.74-dc2O CIP
Copyright 1993 by the American Mathematical Society. All rights reserved.

Translation authorized by Shokabo Publishing Co., Ltd.
The American Mathematical Society retains all rights
except those granted to the United States Government.
Printed in the United States of America
The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Information on Copying and Reprinting can be found at the back of this volume.
This publication was typeset using AJ S-TEX,
the American Mathematical Society's TEX macro system.
1098765432 1 989796959493
Contents
Preface to the English Edition xi
Preface to the New Japanese Edition xiii

Preface to the Original Japanese Edition xv
Chapter 0. Prerequisites from Set Theory 1
0. Basic symbols 1
1. Mappings 1
2. Ordered sets 3
3. Partitions and equivalence relations 4
Chapter I. Groups, Rings, and Fields 7

1. Groups 7
2. Normal subgroups and homomorphisms 10
3. Rings and fields 16
4. Integral domains and prime ideals 20
5. Polynomial rings 23
6. Unique factorization 25
7. Modules 29
8. Symmetric forms and alternating forms 34
Exercises 37
Chapter II. Algebraic Extensions of Finite Degrees 43

1. Basic notions 43
2. Splitting fields 47
3. Separability and inseparability 48
4. Multiplicative groups of finite fields 51
5. Simple extensions 51
6. Normal extensions 54
7. Invariants of a finite group 55
8. The fundamental theorem of Galois 58
9. Roots of unity and cyclic extensions 59
10. Solvability of algebraic equations 63
11. Construction problems 67
vii
viii CONTENTS
12. Algebraically closed fields 71
Appendix 1 73
Appendix 2 74
Exercises 75
Chapter III. Transcendental Extensions 81

1. Transcendence bases 81
2. Tensor products over a field 83
3. Derivations 87
4. Separable extensions 91
5. Regular extensions 94
6. Noetherian rings 96
7. Integral extensions and prime ideals 102
8. The normalization theorem for polynomial rings 107
9. Integral closures 111
10. Algebraic varieties 113
11. The Ci-conditions 118
12. The theorem of Liiroth 122
Appendix. A theorem on valuation rings and its applications 124
Exercises 126
Chapter IV. Valuations 135

1. Multiplicative valuations 135
2. Valuations of the rational number field 138
3. Topology 140
4. Topological groups and topological fields 145
5. Completions 149
6. Archimedean valuations and absolute values 152
7. Additive valuations and valuation rings 154
8. Approximation theorems 159
9. Prolongations of a valuation 162
10. The product formula 168
11. Hensel's lemma 170
Exercises 182
Chapter V. Formally Real Fields 191

1. Ordered fields, formally real fields, and real closed fields 191
2. Real closures 196
3. Hilbert's 17th Problem 202
4. A valuation corresponding to an order 206
Exercises 211
CONTENTS ix
Chapter VI. Galois Theory of Algebraic Extensions of Infinite Degree 215

1. Topology on a Galois group 215
2. The fundamental theorem of Galois 217
3. Splitting fields, inertia fields, and ramification fields 219
4. Algebraic equaitons of high degrees 221
Exercises 224
Answers and Hints 227
Index of Symbols 243
Subject Index 245

Preface to the English Edition
The theory of commutative fields is one of the basic areas in mathemat-

ics, particularly in algebraic theories including number theory, algebra, and
algebraic geometry. Many books relating to algebraic theories contain some
exposition on commutative fields, but very few books contain sufficient ma-
terial on this area.
The author wrote the first edition of this book in 1966 (in Japanese), with
the aim of producing a useful book on commutative fields containing many
topics. In view of the progress made in the theory of commutative fields, the
author added several new topics and reformulated some results for the new
Japanese edition that appeared in 1985.
The author wishes to express his thanks to the American Mathematical
Society for publishing this English edition, which closely follows the 1985
edition mentioned above.
Masayoshi Nagata
September 1992
Xi
Preface to the New Japanese Edition
It has been 18 years since the manuscript of the original Japanese edition
was completed. After its publication, the author noticed several points that
should be improved. Because of this and in view of the length of time since
the publication of the original edition, the author proposed to write a revision.
The author wishes to express his thanks to the publisher for accepting this
proposal.
The main goal of this new Japanese edition is the same as that of the
original and can be stated in three parts.
(1) The prerequisites should be as few as possible. (The prerequisite results
on set theory are stated in Chapter 0 without proofs.)
(2) All results considered by the author to be important and fundamental
in the theory of commutative fields are included.
(3) Chapter I consists of the basic results on group theory and the theory
of commutative rings that are needed to achieve the purpose stated in (2).
In this new Japanese edition, the author has improved several points in
the first edition and added some new topics.
Masayoshi Nagata
March 1985
Xlll
Preface to the Original Japanese Edition
The theory of commutative fields is fundamental in modern algebra. Due

to the fact that algebraic methods are employed not only in algebra but also
in a wide variety of areas, the theory of commutative fields has become one
of the basic areas in modern mathematics.
However, because of the lack of sufficient time in courses for mathematics
students in universities, teachers cannot devote enough time to the theory of
commutative fields, and often they must end with a brief introduction to the
theory of algebraic extensions of fields.
Thus, the author aimed in writing this book to provide a treatise for those
who wish to study the theory of commutative fields on their own and a ref-
erence for those attending lectures on the theory of commutative fields.
To achieve these aims, the author tried especially to have the prerequisites
be as few as possible; the reader is required to have a fundamental knowledge
of set theory and some knowledge of determinants. (These prerequisites are
stated in Chapter 0 without proofs.) To make this book self-contained, the
author included fundamental results on groups and commutative rings.
Thus, the main part of this book consists of what the author judges to be
fundamental in the theory of commutative fields, preceded by a preparatory
part on groups and commutative rings.
For this reason, the presentation does not go more deeply into applications
of commutative rings. In some cases where it seemed better, from the view
point of commutative rings, to treat material under more general circum-
stances, the author chose to present the results under stronger conditions in
order to simplify the presentation.
The author wishes to express his heartfelt thanks to several people for their
help related to the writing and publishing of this book.
Masayoshi Nagata
December 1966
XV
CHAPTER 0
Prerequisites from Set Theory
0. Basic symbols
We assume the readers are familiar with basic notions such as sets, ele-
ments of a set, subsets, empty set, intersection and union of subsets.
The following symbols related to sets are used freely.
E) If a is an element of M, then a c M or M D a.
0 If a is not an element of M, then a V M or M a.
C D If N is a subset of M, then N C M or M D N.
C D If N C M, N# M, then N C M or M D N. (Note that in some
literature, our C is expressed by c ; our c is expressed by C.)
If N is not a subset of M, then N % M or M N.

n , u Intersection and union; the union and the intersection of M1 , ... ,
Mn are M1UM2u...UMn (or Un 1M!) and
(or n" M1) ; the union and the intersection of a family of subsets
1
M2 (1 E A) are UREA MA and n,A MA .

{ } The set of elements a of M satisfying condition P is denoted by
{a E MAP} or {aIP}.
If B is a subset of A , then the complement of B in A is denoted
by A-B. Namely, A-B= {aEAja V B}.
X, f j The direct product of sets M1 , ... , M11 is M1 x ... x M,1 ; the direct
product of M. (L E A) is FLEA MA .
1. Mappings
A mapping f of a set M to a set N is a correspondence which associates
with each m E M a single element n E N. If f is such a mapping, then n
is called the image of m, and there are two types of expression of the image.
One is fm (or f (m)) and the other is mf If M1 is a subset of M, then
.
the image of M1 under f is denoted by either f (M,) or Mif according to

the type of expression employed. In this book, we mainly use the expression
fm.
If f M (or Mf) = N, we say that f is a mapping of M onto N. If,
furthermore, a, b E M, a b implies f a 54 f b (or a.f : bf) , then we say
that f is a one-to-one mapping.
1
2 0. PREREQUISITES FROM SET THEORY
If we associate with each element (a, , ... , an) of the direct product MI x
x Mn of sets Mi its ith component al , then it is a mapping of the direct
product onto Mi , and this mapping is called the projection of H M to Mi.
jf
The projection, in the case of the product of infinitely many sets, is defined
similarly.
If f is a mapping of a set MI to a set M2 and if g is a mapping of M2
to a set M3, then we obtain a mapping which associates with each m E MI
the image of fm (or mf) under g. This mapping is called the product or
the composite of f and g, and is denoted by g f (in the case of fm-type
expression) or f g (in the case of mf -type expression), because the image of
m is expressed by g f m or mfg , respectively.
To each set, the notion of cardinality, which expresses something like the
number of elements, is well defined. The cardinality of a set M is denoted
by #(M). For a finite set (namely, a set consisting of a finite number of
elements), its cardinality is the number of elements. (The cardinality of the
empty set is 0) . The cardinality of the set of natural numbers is said to be
countably infinite. Countable means finite or countably infinite. The cardinal-
ity of the set of real numbers is called the cardinality of the continuum . The
product #(M) x #(N) of cardinalities is defined to be the cardinality of the
direct product M x N. If M, N are finite sets, then the product coincides
with the product as numbers. As for the product of infinite cardinalities, see
Theorem 0.1.3 below.
If there is a one-to-one mapping of M onto N, then the cardinality of
M is defined to be the same as that of N. If there is a one-to-one mapping
of M onto a subset of N, then we have that #(N) > #(M). Then we have
the
COMPARABILITY THEOREM FOR CARDINALITIES. For any two sets M, N :
(i) Either #(M) > #(N) or #(N) > #(M).
(ii) If both of these inequalities hold, then #(M) = #(N).
We have some other results as follows.
THEOREM 0.1.1. Let S(M) be the set of all subsets of a given set M. Then
#(S(M)) > #(M).
THEOREM 0.1.2. If #(M) is countably infinite, then #(S(M)) is the car-
dinality of continuum. Therefore, the cardinality of the set of real numbers is
not countable.
THEOREM 0.1.3. If M is an infinite set (namely, #(M) is not finite), then
the cardinality of the direct product M x M coincides with that of M.
One can find proofs of these results in standard text books on set theory.
EXERCISE. In the case where #(M) is countably infinite, give a direct
proof of Theorem 0.1.3 using the following idea. Assigning a number to each
element of M , we arrange all elements of M as m , 5 ... , mn , ... , and then
2. ORDERED SETS 3
arrange all elements of M x M as (m 1 , MI), (m 1 , m2) , (m2 , MI), ... ,

(ml , mr) (m2, mr-1), ... , (mr, m1), ...
1
.
2. Ordered sets
If a relation > is defined on a set M (namely, for each pair (a, b) of
elements a, b of M, it is well defined whether a > b or not), and if >
satisfies the following three conditions (1), (2), (3), then we call > an order,
and we say that M is an ordered set.
(1) a > a for any aEM.
(2) If a > b and b > a , then a = b .
(3) If a > b and b > c , then a > c .

a > b may be written as b < a. a > b , a 54 b can be expressed by a > b
orb<a.
An ordered set M is called a linearly ordered set if, for any two elements
a, b of M, either a > b or b > a. Every subset of an ordered set is again
an ordered set (with the same order), and every subset of a linearly ordered
set is again a linearly ordered set.
An element a of an ordered set M is said to be maximal if b > a for
b E M implies b = a. An element a of M is said to be the greatest element
if a > b for all b E M. Considering < instead of >, we define minimal
elements and the smallest element.
We say that the maximum condition holds in an ordered set M, if every
nonempty subset N of M has a maximal element of N (which is an ordered
set). The minimum condition is defined similarly.
We say that the ascending chain condition holds in an ordered set M if
a1 < a2 < < an < (a1 E M for all i) implies that there is an index
N such that aN = aN+1 = ' ' ' , or equivalently, there is no infinite chain
a1 < a2 < < an < with elements ai of M. The descending chain
condition is defined similarly.
THEOREM 0.2.1. The maximum condition is equivalent to the ascending
chain condition. The minimum condition is equivalent to the descending chain
condition.
PROOF. If there is an infinite ascending chain a1 < a2 < . . < an < .
, then the subset {ai I i E N} does not have a maximal element. Thus,
the maximum condition implies the ascending chain condition. Conversely,

assume that the ascending chain condition holds in M, and let N be a
nonempty subset of M. Take an ascending chain a1 < a2 < . < an . .
of elements of N. If an is not maximal in N, then there is an element

an+1 of N which is greater than an. If we proceed in this way, then by the
ascending chain condition, such a chain must stop at a finite step, say, at aN .
Then aN is a maximal element in N. The latter half is proved similarly,
considering the dual order. Q.E.D.
4 0. PREREQUISITES FROM SET THEORY
A well-ordered set is a linearly ordered set satisfying the minimum condi-

tion.
EXAMPLE 1. Let M be a set containing at least two elements, and consider
the set S(M) of all subsets of M. Then S(M) becomes an ordered set by
the inclusion relation D, and it is not linearly ordered.
EXAMPLE 2. The set of real numbers is a linearly ordered set under the
order given by the largeness of numbers.
An element a of an ordered set M is called the supremum of a subset
S if a is the least element in the set {b E Mob > s for all s E S} The .
infimum of S is defined similarly by interchanging > and <.

A nonempty ordered set M is called an inductive set, if every nonempty
well-ordered subset has a supremum in M. The following theorem on in-
ductive sets is well known.
ZORN'S LEMMA. Every inductive set has a maximal element.
As for the proof, see any standard text book on set theory.
If M1 , ... , Mn are ordered sets, we can define an order on the direct
product M1 x ... x Mn by: (a, ... , an) > (b , ... , bn) if there is an i
1
1
such that al = bi for every j < i and ai > bi . Namely, we look for the
least i such that ai # bi , and then (a , ... , an) > (bt , ... , bn) is defined
1
by ai > bi . The order defined in this way is called the lexicographical order
on the direct product M1 x ... X Mn M.
EXERCISE. Prove under the above circumstances that:
(i) If M1 , ... , Mn are all linearly ordered, then M1 x x Mn is linearly
ordered.
(ii) If M1 , ... , Mn are well-ordered, then M1 x . x Mn is well-ordered.
3. Partitions and equivalence relations

A partition of a set M means to express M as a disjoint union of non-
empty subsets S. (L runs through a set A), namely,
(I) M = U.IEA SA
(ii) if S. n S, is not empty then S. = Su
(iii) each S2 is not empty.
Each S. is called a class. If a E S2 , then a is called a representative of
A relation - defined on a set M is called an equivalence relation, if it

satisfies the following three conditions.
(i) a-a for all a.
(ii) If a - b, then b - a.
(iii) If a b and b a, then a - c.
THEOREM 0.3.1. Assume that a set M is the union of subsets S. (A runs
through a set A) such that whenever S. S. then S2 and S,, have no
common element. If we define - by a - b iff S1 D a, b for some A, then
3. EQUIVALENCE RELATIONS 5
is an equivalence relation on M. Conversely, if a relation - on M is an

equivalence relation, then the subsets Sa = {b E MIb - a} (a E M) gives a
partition of M.
The proof is easy and the readers are advised to try to prove this.
REMARK. In the conditions for an equivalence relation, (i) can be replaced
by: Given a E M there is some b E M with a- b. (Because a- b implies
b - a by (ii) and then a - a by (iii).) We cannot, however, omit (i).
CHAPTER I
Groups, Rings, and Fields
1. Groups
Assume that M is a set and that a mapping 0 of the product set M x M
into M is given. Then we can write 0(a, b) = c to indicate that an ordered
pair (a, b) of elements a, b of M defines an element c. If we regard
such a mapping 0 as a rule that to each ordered pair of elements of M
associates an element of M, then we say that M is a binary operation, or
simply an operation, on M. If such an operation is called multiplication,
then the element c = 0(a, b) defined by a, b is called the product of a, b
and is denoted by ab. If the operation is called addition, then c = 0(a, b)
is called the sum of a, b and is denoted by a + b. If 0(a, b) = 0(b, a)
holds for any a, b in M, then we say that the operation 0 is commutative.
Assume that a binary operation, say, a multiplication, is defined on a set
G, and the operation satisfies the following three conditions, then we say that
G is a group. (See the example at the end of this section.)
(1) (ab)c = a(bc) for all a, b, c E G (associative law).
(2) There is an element e in G such that, for all a E G, ea = ae = a.
This element e is called the identity and is usually denoted by 1. If we wish
to make it clear that this is the identity of G, then we write G'
1
(3) For each a E G, there is an element a' E G such that aa' = a'a = 1 .
The element a' is called the inverse of a and is denoted by a-'

If the operation is commutative, then we say that the group is a commu-
tative group or an Abelian group.
If a set M forms a commutative group with an operation +, then we call
M a module or an additive group. In this case, the identity is called zero
and is denoted by 0; the inverse of an element a is called minus a and is
denoted by -a.
THEOREM 1.1.1. In a group G, the identity is unique, and for each element
a of G, the inverse a-' is unique. Moreover,
(a-')-' = a and (ab)-' = b-' a-'
for all elements a and b in G G.
7
8 I. GROUPS, RINGS, AND FIELDS
PROOF. Assume that 1 and 1' are identities. Then 1 = 1 1' = 1', so
1 = 1' . Assume now that a' and a" are inverses of a. Then a' = a'(aa") _
(a'a)a" = all, so a' = a" . From the uniqueness of the inverse, we see that
a is the inverse of a-' . abb-' a-' = 1 = b-' a-lab , and we see that
(ab)-I = b-Ia-I . Q.E.D.
If a subset H of a group G forms a group with the operation on G, then
we say that H is a subgroup of G. Submodules are defined similarly.
If H. (with . running through a set A) are subgroups of a group G, then
the intersection n2EA HA is also a subgroup of G. Therefore, if a subset S
of G is given, the intersection of all subgroups of G which contain S is the
smallest subgroup of G containing S. This subgroup is called the subgroup
generated by S. The group generated by a single element is called a cyclic
group, and if its generator is a, then the cyclic group is denoted by (a) .
If H, and H2 are subgroups, then the subgroup generated by HI U H2 is
denoted by Hl V H2. This notation is applied also to the subgroup generated
by many subgroups (HI V H2 V V Hn , VAEA H2 , etc.).
THEOREM 1.1.2. The following three conditions on a nonempty subset H

of a group G are equivalent:
(1) H is a subgroup.
(2) ab-' EH for all a,bEH.
(3) a-'bEH for all a,bEH.
PROOF. It is obvious that (1) implies (2) and (3). Assume that (2) holds,
and choose an arbitrary a E H. Then a, a E H implies 1 = as-' E H.
Then 1, a E H and a- I= la-1 E H. Thus, given a, b E H, we have
a, b-I E H, so ab = a(b-')-' E H. Thus, (2) implies (1). Similarly, (3)
implies (1). Q.E.D.
If K , L are subsets of a group G, then we denote by KL the set {abl a E
K, b E L} . (If the operation is denoted by + , then we write K+L to denote
the set {a + b la E K, b E L}.) In particular, if H is a subgroup and a E G,
then aH = {ahl h E H} and Ha = {hal h E H1. aH (or Ha) is called a
left (or right) coset of H or a left (or right) residue class modulo H. We
denote by G/H and H\G the set of all left cosets aH and the set of all
right cosets Ha, respectively. (In G/H, even if a b, if aH = bH as
subsets of G, then we regard aH and bH to be the same. Similar remarks
hold for H\G).
THEOREM 1.1.3. Assume H is a subgroup of a group G and a, b E G .
Then
(i) aH = bH iff b-'a E H.
(ii) Ha = Hb iff ab-' E H.
(iii) aH bH iff aH n bH is the empty set.
(iv) Ha Hb iff Ha n Hb is the empty set.
1. GROUPS 9
PROOF. If aH = W, then a E bH, so a = bh for some h E H.

Thus, b-' a= h E H. Conversely, if b-' a= h E H, then a E bH and
aH c bHH = bH, and from b = ah-' E H, we have bH c aH. Thus,
aH = W, and (i) is proved. (ii) is proved similarly. As for (iii), if c c
aH n W, then, by (i), cH = aH and cH = bH. Thus, aH = bH in this
case. Namely, we see that if aH n bH is not empty, then aH = bH. The
converse of this is obvious, so we have (iii). (iv) is proved similarly. Q.E.D.
This theorem shows that left cosets (or, right cosets) of a fixed subgroup
H in a group G give a partition of G.
By the order of a group G, we mean the number of elements (or the
cardinality) #(G) of G. G is called a finite group if #(G) is finite. If a is
an element of a group, then the order of the cyclic group (a) is called the
order of a.
If H is a subgroup of a group G, then #(G/H) is called the index of H
in G. As we see by the next theorem, #(H\G) coincides with #(G/H).
THEOREM 1.1.4. If G is a finite group of order g and if H is a subgroup of
order h, then #(G/H) = g/h = #(H\G). The equality #(G/H) = #(H\G)
holds even if #(G) is infinite.
PROOF. Each coset aH consists of h elements. Therefore #(G) = h x
#(G/H) , and it suffices to see the last assertion. The set of inverses of ele-
ments in aH coincides with Ha-' , and we see that the map G/HE) aH -+
Ha-' E H\G is a one-to-one mapping. Thus #(G/H) = #(H\G). Q.E.D.
COROLLARY 1.1.5. Under the assumption in the first half of the theorem, h
and #(G/H) are divisors of g = #(G).
THEOREM 1.1.6. Assume that the order n of an element a of a group G is
finite. If m is a rational integer which is coprime to n, then the cyclic group
(a') generated by a'n coincides with (a).
PROOF. Since am E (a), we have (a m) c (a). By the coprimeness, there
are rational integers c, d such that me + nd = 1 . Therefore, (a'n )` _
amcand = amc+nd
= a. Thus a E (a m) , and (a m) _ (a). Q.E.D.
EXAMPLE. Among the important examples of groups, there are permu-
tation groups. Consider a set M, and let S(M) be the set of one-to-one
mappings of M onto itself. Then S(M) is a group with composition of
mappings as the operation on S(M). S(M) is called the symmetric group
on M. Since the group structure of S(M) is determined only by #M,
S(M) is called the symmetric group of degree n if #(M) is a finite number
n. Subgroups of S(M) are called permutation groups on M. (If n = #(M)
is finite, we call them permutation groups of degree n.)
If #(M) = n (finite), then the order of S(M) is n! (because there is a
one-to-one mapping of S(M) onto the set of permutations of elements of
M) . If M = {m1 , ... , mn} , then an element a of S(M) may be expressed
in the form
where Qm = m for each i.

m1.. mn
In S(M) we fix our notation so that at denotes the mapping m -* a(zm) .
EXERCISE. Find the following products in S(M) with M c 11, 2, ... , 5} .
1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4
(1) 2 1 4 3 4 3 2 1 (2) 2 3 4 1 2 1 4 3
1 2 3 4 5 2 (4) 1 2 3 4 5 2
(3) 2 3 4 5 1 2 1 4 5 3
2. Normal subgroups and homomorphisms

A mapping q of a group G to a group G' is called a homomorphism
if it satisfies the condition: q5(ab) = (0a) (0b) for all a, b E G. In this
case, {x E Glq5x = 11 is called the kernel of 0. If OG = G', then we say
that qS is a homomorphism onto G' If the kernel of consists only of the
.
identity, then we say that is an injection, or an isomorphism, into G'. If,

furthermore, q is an onto mapping, then we say that q is an isomorphism
to (or, onto) G'. If there is an isomorphism from G to G', then we say that
G' is isomorphic to G 'and we write G ^' G'. The set of all homomorphisms
from G to G' is denoted by Hom(G, G').
If we fix an element a of a group G, then we can define a mapping a of
G to G by ax = a-'xa. Then we have a(xy) = a-'xya = a-'xaa-'ya =
(ax)(ay). Furthermore, Q(aba-1)=b for all b E G, and if ax = 1 then
a-'xa = 1 , so x = aa- = 1 . Thus, this a is an isomorphism from
G to G. In general, an isomorphism from a group G to itself is called
an automorphism of G, and the a above is called the inner-automorphism
defined by a. Note that if we write xa instead of ax , then (xa)b = x' .
An automorphism of G which cannot be obtained as an inner-automorphism
is called an outer-automorphism. If S is a subset, or an element of G, then
its image by an inner-automorphism (a-' Sa) is called a conjugate of S.
A subgroup N of a group G is called a normal subgroup, if N is mapped
into N by all inner-automorphisms, namely, if N satisfies the condition
g-' ng E N for all g E G, n E N. This condition is equivalent to g-' Ng =
N for all g E G. (Readers are advised to prove this. Note that g-' Ng C N,
gNg-' c N imply that g-'Ng = N.) In this case, we have gN = Ng for
all g E G, namely, left cosets coincide with right cosecs. So, a coset aN
may be written as a mod N.
THEOREM 1.2.1. If N is a normal subgroup of a group G, then the set
GIN of cosets forms a group by the operation defined by (aN)(bN) = abN.
2. NORMAL SUBGROUPS AND HOMOMORPHISMS II
The mapping yr of G to GIN defined by yra = aN is a homomorphism

from G onto GIN.
GIN is called the residue class group of G modulo N, or the factor group
of G by N. The mapping yr is called the natural homomorphism from G
to GIN. (If G is a module and N is a submodule, then G/N is called a
residue class module.)
PROOF. If b E G, then b-' Nb = N and bN = Nb. So we have
(aN)(bN) = abNN = abN as subsets of G. Therefore the operation
(aN)(bN) = abN is well defined and the product does not depend on the
expression of aN, bN (note that if a' E aN then aN = a'N). Associativity
follows from ((aN)(bN))(cN) = (ab)cN = a(bc)N = (aN)((bN)(cN)). N
is the identity of GIN, and a-' N is the inverse of aN. Thus G/N is a
group. Since yr(ab) = abN = (aN)(bN) = (yra)(yrb), and we see that w is
a homomorphism from G onto GIN. Q.E.D.
THEOREM 1.2.2. If 0 is a homomorphism from a group G to a group G',
then the kernel N of 0 (N = {x E Glq5x = 'G' }) is a normal subgroup
of G, and aN -* Oa defines an isomorphism from G/N to OG. To each
subgroup H' of qG, we associate 0-' (H') = {g E Glog E H'}. Then we
have a one-to-one correspondence between the set of subgroups of OG and the
set of subgroups of G containing N. H' is a normal subgroup of OG iff
0-' (H) is a normal subgroup of G.
PROOF. First we see that 0-' (H') is a subgroup as follows. For a, b E
0- (Y), we have 0(a-' b) = (0a)_' (0b) E H', so a-' b E 0-'(H'). Thus,
c-' (H') is a subgroup by Theorem 1.1.2. That 0-' (H') contains N is obvi-
ous. If H' is a normal subgroup of qSG, we see that for g E G, a E 0-'(H'),
we have 0(g-'ag) = (0g)-'(0a)(0g) E (fig)-'H'(ag) = H'. Therefore,
c-' (H') is a normal subgroup of G. This is applied to the case where
H' = {1 G'1, and we see that N is a normal subgroup of G. The follow-
ing are equivalent: b E 0_1(0a), qb = qa, 0(a-' b) = 10' , a-' b c N,
b E aN. Thus, we see that /-' (oa) = aN. From this fact, we see easily
that the map aN -* Oa defines an isomorphism from GIN to OG. If H
is a subgroup of G containing N, we see that b E q-' (OH) if b E h N for
some h E H, which is equivalent to b E HN = H. Thus, we see that the
map H' -+ 0-' (H) defines a one-to-one correspondence as asserted, because
OH is obviously a subgroup of OG. As for the last assertion, readers are
advised to try to prove it. Q.E.D.
Note that the two theorems above show that normal subgroups are char-
acterized by the property that they are kernels of homomorphisms.
REMARK. For a given group G, the set of all automorphisms of G forms
a group (with composition of mappings as the operation). This group is
called the automorphism group of G and is denoted by Aut G. In Aut G,
the set of all inner-automorphisms is a normal subgroup and is called the

inner-automorphism group of G. (We omit the proof.)
Next, we give an important remark on homomorphisms. Let 0 be a ho-
momorphism from a group G to a group G' with kernel N, and let H be
a subgroup of G. Then the restriction 0IH of 0 to the subgroup H has
N n H as its kernel. Therefore, we have the following theorem.
THEOREM 1.2.3. Under the circumstances in the previous paragraph, H/
(H n N) can be regarded as a subgroup of GIN (hN E GIN (h E H) is
identified with h(H n N) E H/(H n N)). In particular, if G = HN, then
H/(H n N) is identified with GIN. (This last assertion is called the isomor-
phism theorem.)
Several important examples of normal subgroups are given by so-called
commutator groups. For two elements a, b of a group G, a-1 b-1 ab is
called the commutator of a, b and is denoted by [a, b]. Note that ab =
ba[a, b]. If H, K are subgroups of G, then the subgroup generated by
{[h, k]Ih E H, k E K} is called the commutator group of H, K and is
denoted by [H, K]. [G, G] is called the commutator group, or the derived
group, of G.
THEOREM 1.2.4. If N1 , N2 are normal subgroups of a group G, then
[N1 , N2] is a normal subgroup of G and is contained in N1 n N2 .
PROOF. If g E G, then
g-1[a, b]g = [g-Iag, g-1bg]
and
-11
9 [ n ,, n2]Ini E N1}g = {[n1 , n2]I nl E N }.
From this, it follows that [N1 , N2] is a normal subgroup of G. [n1 , n2] _
ni 1(n21 n1 n2) and n1 E N1 implies [n1, n2] E N1 . Thus [N1 , N2] c N1 .
Similarly, [N1 , N2] c N2 , so [N1 , N2] c N1 n N2 . Q.E.D.
COROLLARY 1.2.5. For a given group G, let Do = G, Di = [D1_ I, D._ ] . 1
Then Do D DI D ... D Dr J . Let Zo = G , Z1 = [Zj_ , G] , then1
Zo D Z1 D D Zr D . These D. , Z1 are normal subgroups of G.
The sequence Do D D1 D D Dr D ... is called the derived series

of G. G is called a solvable group, if {1 } appears in this series. (As for
the reason of the name "solvable", see 10 in Chapter II.) The sequence
Zo ZI D D Zr D is called the descending central sequence of G. G
is called a nilpotent group, if { 1 } appears in this sequence. Every nilpotent
group is solvable. Important examples of nilpotent groups are given by the
following theorem.
THEOREM 1.2.6. Every finite group whose order is a power of a prime num-
ber p (such a group is called a p-group) is nilpotent.
2. NORMAL SUBGROUPS AND HOMOMORPHISMS 13
Before proving this, we define the center of a group. In a group G, {x E

GIxz = zx for all z E G} is called the center of G and is denoted by Z(G).
As is easily seen, this is a normal subgroup of G. (Cf. Exercise 1.2.5.)
LEMMA 1.2.7. For a group G, we consider a sequence of groups Gi as

follows: G = G, G1 = G/Z(G), ... , G. = Gi_1/Z(Gi_1), .... Then G is
nilpotent iff Gn = { 1 } for some n.
PROOF. Assume first that G = Z D Z1 D ... D Zn = { 1 } is the descending

central sequence of G. We prove the only if part by induction on the length
n. If n = 1 , then the assertion is obvious and we assume that n > 1 . Since
[G, Zn_ ] = Zn = 111, we have Zn_ C Z (G) . Considering the natural
1 1
homomorphism of G onto G1 = G/Z (G) , we see that the descending

central sequence G1 = W D W_ I of G1 = OG is given by
W. = q[G, Zt_1] (cf. Exercise 1.2.8). Then Wn-1 = c[G, Zn-2] = OZn-1 =-
{ 1 } , and by our induction hypothesis, Gni = { 1 } for some n'. (It is not
hard to show Gn = { 1 } simultaneously.) Conversely, assume that Gn =
{ 1 } . We prove the converse part by induction on n . If n = 1 , then G is
commutative and is nilpotent. Assume that n > 1 . Our induction is applied
to G1 = G/Z (G) = qG, and we see that there is an r for which cZr = { 1 } .
(It is not hard to show qZn_ I = { 11 simultaneously.) It follows now that
Zr C Z (G) and Zr+1 = [G, Zr] = { 11 . Q.E.D.
PROOF OF THE THEOREM. By Lemma 1.2.7 and by the fact that homo-
morphic images of a p-group are p-groups (cf. Corollary 1.1.5), it suffices to
show that if G is a p-group not equal to { 1 } , then Z (G) { 1 } . For each
element a of G, let Ca be the set of conjugates of a. These Ca (a E G)
define a partition of G, because if c E Ca n C b then c = d -1 ad = f b f - I for
some d, f E G. Thus, b = (df )- I a(df), so Cb C Ca. Similarly, Ca C Cb.
Consider Za = {c E GIca = ac}. This is a subgroup of G. Furthermore,
#(Ca) = #(G/Za), since b-tab = d-tad if bd-1 E Za which is the same
as b E Zad. Thus, each #(Ca) is a power of p. The sum of all #(Ca) is
#(G), which is a power of p. On the other hand, #(C1) = 1 = p, and
the number of the classes Ca such that #(Ca) = 1 must be a multiple of
p. Since #(Ca) = I precisely when a E Z (G) , we see that Z (G) { 1 } .
Q.E.D.
The following is a very fundamental theorem on finite groups.
THEOREM OF SYLOW. Assume that G is a finite group and its order is

factorized as pii . . . per (p1 < < pr are prime numbers). Then
(i) G has a subgroup Si of order peg .
(ii) For each pi, the subgroups Si are conjugate to each other.
(iii) For each pi, let Al be the number of conjugates of S, . Then A, - 1 is
a multiple of pi.
(iv) If a subgroup H of G is a pi-group, then for some a E G, H C_

a -1 Sia.
Si is called a pi-Sylow subgroup of G.
Before proving this theorem, we prove the following lemma.
LEMMA 1.2.8. Assume that P is a p-group (p being a prime number),
E is a finite set, and that an action of P on E is defined (namely, (i) for
aEP and a E E, as is defined and aa E E, (ii) for o,rEP and a E E,
a(za) = (a'r)a, and (iii) la = a). We write EP = {x E Elax = x for all
a E P} . Then #(E) - #(EP) is a multiple of p .
PROOF. For each a E E , we consider Pa = {gala E P} and Sa = {o E
Pl as = al. Then, as = a'a if a ' a' E Sa. Since Sa is a subgroup of P,
we have #(Pa) = #(P/Sa) and #(Pa) is a power of p (cf. Corollary 1.1.5).
Therefore, a EP if #(Pa) is a multiple of p. Thus, #(E) - #(EP) is a
multiple of p. Q.E.D.
PROOF OF THE THEOREM OF SYLOW. pi and ei will be written as p and
e respectively. Then #(G) = peq (q is not a multiple of p). Consider
E = {X C GI #(X) = pe}.
We first show that #(E) - q is a multiple of p. #(E) = (Pe which is
e
the coefficient of X" in (X + 1)pq
e . (X + 1)pq is congruent to (X" + 1) q
modulo p. Here, we say that two polynomials f (x) , g(x) are congruent
modulo n if every coefficient of f (x) - g(x) is divisible by n, and we write
f (x) - g(x) (mod n). The same terminology is applied to the case where
f , g are integers (f - g is a multiple of n if f - g (mod n)) .
ADDITIONAL REMARK. (a + b)p - ap + by (mod p), as is easily seen by
e
the binomial theorem. By induction on e, we see (a + b)p - ap + by for
every natural number e.
Now, we go back to the proof. The coefficient of XP' in (X" + 1)q is q,
and therefore #(E) - q is a multiple of p. Note that #(E) is not a multiple
of p because q is not a multiple of p.
If X E E and a E G, then aX E E. Therefore, for each X E E, we can
consider GX = {QX IQ E G}. Since GX (X E E) give a partition of E (cf.
Exercise 1.1.6), we see that for some X E E, #(GX) is not a multiple of p .
We fix such an X and consider P = {a E GI aX = X} . Then P is a subgroup
of G and #(GX) = #(G/P). Since #(GX) = #(G)/ #(P) is not a multiple
of p, we see that #(P) is a multiple of pe (Theorem 1.1.4). On the other
hand, a E X implies Pa C X, and therefore, #(P) = #(Pa) < #(X) = pe.
Therefore P is a Sylow subgroup. Thus, (i) is proved.
Consider H in (iv). H acts on G/P, for given h E H, aP E G/P, we
have h(aP) = haP. We apply the above lemma with G/P and H for E
and P, respectively. Since #(G/P) is not a multiple of p, it follows that
for some aP E G/P, we have HaP = aP. Then a-1 Ha C P, which proves
2. NORMAL SUBGROUPS AND HOMOMORPHISMS 15
(iv). If H is a Sylow subgroup, then #(a- I Ha) _ #(P) and a- I Ha = P,

which proves (ii).
Furthermore, by the lemma above, #(G/P) #({aPJPaP = aP}) =
#({aPla-IPa = P}) = #(N(P)/P) (mod p), where N(P) = {a E G1a-IPa =
P} (the normalizer of P ; cf. Exercise 1.2.5). The number of conjugates of P
is #(G/N(P)) = #(G/P)/ #(N(P)/P), and (iii) follows from the congruence
#(G/P) - #(N(P)/P) (mod p). Q.E.D.
We say that a group G is the direct product of of the subgroups N1, ... , Nr,
if each H is a normal subgroup of G and every element g of G can be
expressed as a product n nr of elements n 1 , ... , nr of N1 , ... , Nr and
1
such an expression is unique. In this case, we write Ni x x Nr. If G is a

module, then we may say direct sum and use instead of x . We omit here
the direct product of infinitely many subgroups.
THEOREM 1.2.9. Let H1, ... , Hr be subgroups of a group G. Then G =
H1 x x Hr if and only if the following conditions are satisfied.
(I) G = H1 .. Hr ,
(ii) ab=ba for all aEH,, bEH, i j,
(iii)foreach i=2,...,r,
H1 Hr. If H1 . . H_ 1 n H contains an
.
element a o 1, then a = h1 , ht_1 (hi E Hj), which shows there are two
ways toexpressa (1 EH(j i), aEH, and h1 EH (j<i), 1EHn
(m > i)), contradicting the definition. Thus H1 . . H_ I n H, = { 1}. In
.
particular, if i j , then H, n H = { 1 } . Thus, [H , H ] c H n Hj = { 1 }

(Theorem 1.2.4), and a E Hi, b E H implies [a, b] = 1, that is, ab = ba.
Conversely, assume that the conditions (i)-(iii) hold. By (i), each element
g of G is expressed as g = h1 hr (h, E Hi). For an arbitrary b E H,
g-1 bg = hr . hi bh1 hr and this element is equal to hJ -.1 bhp because
1 1
. .
of commutativity of elements of H and Hj. Thus g- I H g c Hj, which

shows that each Hj is a normal subgroup. Assume now that an element
g of G is expressed in two ways: g = h1 hr = hi h; . Let c be the
largest member of { i I h, 0 hi } . Then h I . h e = h i
. . h' and 1 O hC
I
(h1 hi) (hC 11 hC_ 1) E HI H _1 n Hc, a contradiction. Q.E.D.
If GI , ... , Gr are groups, an operation is defined on the direct product
G1 x . . x Gr (as sets) by
(g'1,...,5'r)(S'I,...,gr)=(glg1,...,grgr
and G1 x x Gr is a group. This group is also called the direct product
of G1 , ... , Gr . If we identify each element g, of G, with (1 , ... , , g, , 1
1, ... , 1), then each G. is a normal subgroup of the direct product, and
GI x x Gr is the direct product of these normal subgroups in the sense of
the previous definition. (As for modules, we may say direct sum, as explained
previously.)
We may say that the previous definition is from the view point of the
structure of a group and that this definition is from the view point of the
construction of a group. These two definitions are substantially the same.
3. Rings and fields
A set R with two operations, say, multiplication and addition, is called
a ring if these operations satisfy the following three conditions. (As will be
explained at the end of this section, after this section, a ring will mean a ring
under stronger conditions (namely, commutative and with identity).)
(1) R is a commutative group under the addition +. (Associativity,
existence of 0, and minus of each element, and commutativity.)
(2) The multiplication satisfies the associative law ((ab)c = a(bc)) .
(3) (Distributive law) For all a, b, c E R, we have
(a + b)c = ac + bc, c(a + b) = ca + cb.
If the multiplication is commutative in a ring R (namely, ab = ba for all
a, b E R) , then we say that R is a commutative ring. On the other hand, if a
ring R has a nonzero element and also an element e satisfying ae = ea = a
for every a e R, then e is called the identity of R and is usually denoted
by 1 or by 1 R as in the case of a group.
THEOREM 1.3.1. If R is a ring, then
(i) aER=a0=0a=0,
(ii) a, b E R = (-a)b = a(-b) _ -ab (-a) (-b) = ab
, .
PROOF. 0 + 0 = 0 and a(0 + 0) = aO. By the distributive law, a0 + a0 =

aO. Adding -(a0) to both sides, we have a0 = 0. Oa = 0 is seen similarly.
(a + (-a))b = Ob = 0, and ab + (-a)b = 0. This means that (-a)b =
-(ab). Similarly, a(-b) = -(ab). (-a)(-b) = -(-ab) = ab. Q.E.D.
COROLLARY 1.3.2. If a ring R has an identity 1, then 1 0.
PROOF. Take a nonzero element a of R. Then a 1 = a 0, and 1
0. Q.E.D.
Let R be a ring with identity. If an element a has an inverse a-' under
multiplication (namely, a-' a = as-' = 1) , then we say that a is an invert-
ible element or a unit. (The set of all units in R forms a group. Readers are
advised to prove this as an exercise.)
If a is an element of a ring R and if n is a natural number, by na we
mean the sum a + + a (n times). If the set N = {natural numbers
nI for all x E R, nx = 0} is not empty, then the smallest number in N
is called the characteristic of R. If N is empty, then the characteristic of
R is defined to be 0. Note that if R has 1, then N = {natural numbers
nlnl =0}.
Next, we define fields. A ring R with identity is called a field, if every
nonzero element of R has an inverse. In this case, the set of nonzero el-
ements of R forms a group, which is sometimes called the multiplicative
3. RINGS AND FIELDS 17
group of the field R. If the multiplication is commutative, then R is called

a commutative field. A field whose multiplication is not commutative is called
a noncommutative field or a skew field. Since we will discuss commutative
fields only, we do not handle skew fields in this book. Therefore, from now
on, by a field we mean a commutative field. As examples of fields, we have
(i) the set of rational numbers,
(ii) the set of real numbers,
(iii) the set of complex numbers.
These are called the rational number field, the real number field, and the
complex number field, respectively.
Among important examples of rings, we have
(i) the set of rational integers. (Note that among integers, there are alge-
braic integers (cf. 12 in Chapter II) besides the usual integers. Therefore, we
call the usual integers "rational integers", because they are algebraic integers
which are rational numbers.)
(ii) the set of n x n matrices over a field K.
They are called the ring of rational integers and the full matrix algebra of
degree n over K, respectively.
If a subset S of a ring R forms a ring under same operations as R, then
we say that S is a subring of R. If S is, furthermore, a field, then we call
S a subfield of R .
A mapping 0 of a ring R to a ring R' is called a homomorphism, if for
all a, b R, we have
q(a + b) = Oa + q5b, q(ab) = (q5a)(gb).
{a E RJga = 0} is called the kernel of 0. As in the case of groups, we have
the following definitions. If the kernel consists only of 0, then we say that
0 is an injection, or an isomorphism, into R'. If OR = R', then we say
that 0 is a homomorphism onto R'; if, furthermore, 0 is an injection, then
we say that 0 is an isomorphism from R to (or, onto) R'. If there is an
isomorphism from R to R', then we say that R is isomorphic to R' and we
write R ^_' R'. An isomorphism from R to itself is called an automorphism
R
A homomorphism from R to itself is called an endomorphism of R.
Assume that rings R, R' contain a common subring K. A homomor-
phism (or an isomorphism) 0 from R to R' is called a K-homomorphism
(or, a K-isomorphism) if the restriction of 0 to K is the identity mapping.
(If R' = R, then 0 is called an endomorphism (or, an automorphism) of R
over K.) If there is a K-isomorphism of R to R', then we say that R is
K-isomorphic to R' and we write R NK R' .
The set of homomorphisms from a ring R to a ring R' is denoted by
Hom(R, R'). (Note that Hom for groups is different from Hom for rings,
and we must be careful if there may be misunderstandings). For the set
of K-homomorphisms, we use the symbol HomK(R, R') and for the set of
K-automorphisms, we use AutKR .
Let 0 be a homomorphism from a ring R to a ring R'. Since 0 is

a homomorphism of modules, the kernel I = {a E Rkba = 0} of 0 is a
submodule of R (by the case of homomorphisms of groups). But, I satisfies
a further condition: for x E R and a E I, we have ax E I and xa E I.
In general, if a submodule I of a ring satisfies this condition, then I is
called a two-sided ideal, or simply, an ideal, of R. A left (or, right) ideal is
a submodule of R satisfying xa E I (or, ax E I) . (The word "two-sided"
comes from its being a left and a right ideal simultaneously.) If R is a
commutative ring, then there is no necessity to distinguish between left and
right ideals; we simply call them ideals.
If we use the notion of two-sided ideal, we have results similar to those in
Theorems 1.2.1, 1.2.2.
THEOREM 1.3.3. If I is a two-sided ideal of a ring R, then the residue class
module R/I is a ring if we define multiplication by (I + a)(I + b) = I + ab
(for a, b E R). Then the natural homomorphism of R onto R/I as modules
is a homomorphism of rings.
The ring R/I is called the residue class ring of R modulo 1.
PROOF. aI C I , I b C I by the property of two-sided ideal. This means
that {cd lc E (I + a), d E (I + b)} is contained in I + ab. Therefore, the
equality (I + a)(I + b) = I + ab is well defined (independent of the choice
of representatives a, b) . Thus, multiplication is well defined, and by this
definition, we see easily the validity of associative law and the distributive
law in R/I and we see that the natural homomorphism as modules is a
homomorphism of rings. Q.E.D.
THEOREM 1.3.4. If 0 is a homomorphism from a ring R to a ring R',
then the kernel I of 0 is a two-sided ideal of R, and q5a -p I + a gives
an isomorphism from OR to R/I. To each subset H' of OR, we associate
q-1(H') = {b E RjOb E H'}. Then, (i) this gives a one-to-one correspondence
between two-sided ideals of OR and two-sided ideals of R containing I, and
(ii) similar facts hold for left ideals, right ideals and subrings, respectively.
PROOF. We saw already that the kernel is a two-sided ideal. That OR
R/I and the other assertions can be seen by arguments similar to those used
in the case of groups. Q.E.D.
REMARK. If a ring R has an identity, then for a two-sided ideal I, 10 R
if there are no invertible elements in I. Readers are advised to prove this.
For groups, we defined the notion of the subgroup generated by a subset.
If a subset S of a ring is given, then we can define the subring generated by
S, and the left (or right, or two-sided) ideal generated by S. Namely, the
smallest subring containing S (i.e., the intersection of all subrings containing
S) is the subring generated by S, and the smallest left (or right, or two-
sided) ideal containing S (i.e., the intersection of all left (or right, or two-
3. RINGS AND FIELDS 19
sided) ideals containing S) is the left (or right, or two-sided, respectively)

ideal generated S by S. The existence of these follows from the fact that
the intersection of subrings, left ideals, right ideals, or two-sided ideals is a
subring, a left ideal, a right ideal, or a two-sided ideal, respectively.
If II , I2 are left (or right, or two-sided) ideals of a ring R, then their
sum Ii + I2 = {a1 + all a1 E Ii j is a left (or right, or two-sided) ideal of
R. If I (A runs through a set A) are left (or right, or two-sided) ideals of
R, then their sum EAEA I is defined as the set of all elements of the form
al + + an with a, E I, (A,! E A) . This is precisely the left (or right, or
two-sided) ideal generated by the union of I (,. E A).
When left (or right, or two-sided) ideals II , ... , I, of a ring R are given,
the set {a1 an Jai E Ii} may not be a submodule of R and the submod-
ule generated by this set {a1 ' ' an I a, E I, } is defined to be the product of
I1 , ... , In and is denoted by II . . In . This product is a left (or right, or
.
two-sided) ideal. The n-times product of a left (or right, or two-sided) ideal
I is denoted by In .
Since the main objects of this book are commutative fields and rings, we
will be working with commutative rings with identity. Therefore, hereafter
in this book, by a ring, we mean a commutative ring with identity.
EXERCISE. Let R be a ring (commutative and with 1). Prove that the
ideal generated by an element a of R is aR = {ar1r E R} and that the ideal
generated by elements a1 , ... , an of R is a1 R+ . +a,,R = {>i a1ri I rri E R1.
A principal ideal is defined to be an ideal generated by a single element.
Here, we shall discuss direct sums of rings. When rings R , ... , Rn are
1
given, we consider the direct product 1(a, , ... , an) 1 a! E R } of R. , and we

define operations (addition and multiplication) by
(a1,...,an)+(b1,...,bn)=(a1+b1,...,an+bn),
then we obtain a new ring ((1 , ... , 1) is the identity and (0, ... , 0) is zero).
This new ring is called the direct sum, or the direct product, of R , ... , Rn .
1
(This is the definition from the view point of construction of a ring.) Assume
that I, , ... , In are ideals such that 11 +...+In = R, (It + ..+Is)nIs+1 = {0}
for each s = 1, 2, ... , n - , and Is {0} for each s = 1 , 2, ... , n , then
1
R is the direct sum of II , ... , in as modules (cf. 2). If we write the identity
1 of R in the form e1 +. - .+e,, (ei E I) , then (i) e,ej = 0 for i : j, because
I,I c I n I = 101, and (ii) for x c I we have x l = xet + . + xe,, . Then,
. -
by the property of direct sum, it follows that x = xej (xe1 = 0 for i 0 j).
Therefore, ei is the identity of and we say that R is the direct sum (or
ii,
direct product) of rings I, , ... , In (the definition from view point of the
structure of rings; cf. the two definitions of direct product of groups in 2).
The symbols x , are used similarly as in the case of groups and modules.
THEOREM 1.3.5 (Chinese Remainder Theorem). Assume that II , ... , I

(s > 2) are ideals of a ring R such that I + I = R for i 0 j. Then we have
(1) If i j , then I1 n I = I1Ij .
(2) If c1,...,cm,d1,...,d, E {1,...,s} (there may be the same num-

bers among c, 's or dj's) and if {c1 , ... , cm } n Id, , ... , d,,} is empty, then
IC1
Ic,,,
nr
Idr Id= R. In particular, if i 54 j, then I"' + In = R.
o J
(3) R/(I1 n . n IS) is isomorphic to the direct sum of R/I1 , ... , R/IS
. .
PROOF. (i) I1Ij cIi nIj=(InIj)R=(InIj)(I;+I)cI,I.

(ii) 1] (' + Idk) = R'n = R. Therefore, I I + Idk = R. Thus,
[Ik (I, in
I
+ Idk) = R" = R, and we see the assertion.
(iii) Consider the mapping Oj such that R/ (I1 n . . n Is) D (a mod I1 n . n
Is) -+ (a mod I) E R/I . This is a homomorphism from R/(I1 n n IS)
onto R/I Then we have a homomorphism 0 from R/(I1 n ... n IS) to
.
the direct sum S of R/I1 , ... , R/I defined by q5a = (q1a, ... , qsa) (a =
(a mod 11 n .. n IS) with a E R) . a is in the kernel of 0 iff a E I for all j
that is, a = 0. Thus, 0 is an injection. Now, it suffices to show that 0 is an
onto mapping. Let (a1 , ... , as) be an arbitrary element of the direct sum
S, where aj _ (aj mod I) . Since I1 + I2 Is = R by (ii), we may assume
that a 1 E I2 Is I. Similarly, we may assume that ai E I ... Ii-1 Ii+
1
Is
1 .
Set a = a1 + . + as. Then, since i # j implies ai E

. we see that
(a mod I) = (aj mod I) for every j. Therefore, qa = (a1 , ... , as), which
shows that 0 is an onto mapping. Q.E.D.
4. Integral domains and prime ideals

An element a of a ring R is called a zero-divisor, if there is a nonzero
element b of R such that ab = 0. An element a is said to be nilpotent if
there is a natural number n such that a" = 0.
If there is no zero-divisor except 0 in a ring R (with 1), then we say that
R is an integral domain. An ideal P of a ring R is called a prime ideal,
if R/P is an integral domain. Subrings of a field are integral domains. As
will be shown later (Theorem 1.4.4), every integral domain is a subring of a
certain field.
THEOREM 1.4.1. If an ideal P of a ring R is different from R itself, then

the following four conditions are equivalent to each other:
(i) P is a prime ideal of R.
(ii)Ifa,bER and abeP, then aEP orbeP.
(iii) If I, J are ideals of R and I J C P, then I C P or J C P.
(iv) If I, J are ideals of R, P c I, and P c J, then If P.
In (iii) and (iv), we can interpret IJ as either the product of ideals I, J
or as the set {abla E I, b E J} .
4. INTEGRAL DOMAINS AND PRIME IDEALS 21
PROOF.(i) = (ii) is obvious, and we shall show (ii) = (iii). Assume

that I P. Then for some a E I, we have a P. Since I J C P, we
have ab E P for any b E J, and therefore, b E P for all b E J. Thus,
.I C P. The implication (iii) =:> (iv) is obvious, and it remains only to prove
that (iv) = (i). Let 0 be the natural homomorphism of R onto RIP,
and assume that qa (a E R) is a zero-divisor, namely, that for some b E R
with q5b 0, we have (ga)(gb) = 0. Let I = P + aR and J = P + bR.
Since ab E P, we have I J C P. Since P C J, if a P, then P C I,
contradicting IJ P. Thus, a E P and q5a = 0. Therefore, RIP has no
nonzero zero-divisor. 01 is the identity of RIP, and RIP is an integral
domain. Q.E.D.
THEOREM 1.4.2. Assume that a nonempty subset S of a ring R is multi-
plicatively closed (that is, ab E S for all a, b E S) and that an ideal I of
R has no common element with S. Let F be the set of ideals J of R such
that J D I and J fl S is empty. Then F has a maximal member, say P,
(under the inclusion relation) and P is a prime ideal of R.
PROOF. We easily can see that F is an inductive set, and F has a maximal
member by Zorn's Lemma. If J1 , J2 are ideals such that P C J1 , P C J2,
then by the maximality of P, both J1 fl S and J2 fl S are nonempty. Let
s E J1 fl S, s' E J2 n S. Then ss' E S, and ss' E Jl J2 P. Therefore, by
(iv) in Theorem 1.4.1, we see that P is a prime ideal. Q.E.D.
REMARK. Under the assumptions of Theorem 1.4.2, maximal members
of F are called maximal ideals with respect to S. If S consists only of
invertible elements, then a maximal ideal with respect to S is simply called
a maximal ideal of R. The theorem above shows that if I is an ideal of
R and if I R, then there is a maximal ideal containing I and that every
maximal ideal is a prime ideal.
THEOREM 1.4.3. For an ideal M of a ring R, the following are equivalent.
(i) R/M is afield.
(ii) M is a maximal ideal.
In particular, the following are equivalent
(iii) R is afield.
(iv) {0} is a maximal ideal.
(v) R has no ideal except {0} and R.
PROOF. We start with (iii)-(v). Assume that R is a field and I is an ideal
different from 101. Let a be a nonzero element of I. By the definition of
a field, a has an inverse a-1 . Therefore, I D aR = R, which shows (iii)
implies (v). It is obvious that (iv) is equivalent to (v). Assume that (v) holds.
If 0 54 a E R , then aR = R , and for some b E R , we have ab = I. By the
commutativity of R, we see that b is a-1 . Thus, (v) implies (iii).
That (i) is equivalent to (ii) is seen by Theorem 1.3.4 and the equivalence
between (iii) and (iv) proved above. Q.E.D.
THEOREM 1.4.4. If R is an integral domain, then there is afield K satis-

fying the following two conditions:
(i) R C K.
(ii) Every element of K can be expressed in the form ab-1 with elements
a,b (b 0)ofR.
Such afield K is unique up to isomorphism, namely, if K is another field
satisfying these conditions, then there is an isomorphism 0 from K to K
such that the restriction SIR of 0 to R is the identity mapping.
This field K is called the field of fractions of R .
PROOF. We consider the set Q = {(a, b) Ia E R, 0 34 b E R1, and we

introduce a relation on Q defined by (al , b1) ' (a2 , b2) a al b2 = a2b1 .
Then we easily see that this relation is an equivalence relation. Consider

the partition of Q defined by this equivalence relation, and denote by alb
the class containing (a, b) . On the set K of these classes, we introduce
operations (addition and multiplication) by
(al /bl) + (a2/b2) = (a, b2 + a2b1)/(b1 b2) ,
(al /bl)(a2/b2) = (a1 a2)/(b1 b2).
Then we can see easily that
(1) K is a ring, where the identity is 1 / 1 and zero is 0/ 1.
(2) If we associate all with the element a of R, then we have an iso-
morphism of R into K. Therefore, we may identify a with a l l. Then
alb coincides with ab-1 .
(3) If a, b are nonzero elements of R, then alb, b/a are in K and
(a/b) (b/a) = 1/1 . Therefore, K is a field.
Thus, we saw the existence of a field of fractions. Assume that K is
also a field of fractions of R. Consider a mapping f which associates
ab-1 E K with alb E K. Since alb = a'/b' implies ab' = a'b, and hence
ab-1 = a'b'-1 , we see that f is really a mapping. By the property of the field
of fractions, we see that fK = K. Since ab-1 + cd -1 = (ad + bc)b-1 d- 1 ,
(ab-1) (cd -1) = acb-1 d -1 , we see that f is a homomorphism of rings. The
kernel of f is an ideal of K, which is a field. Thus f is an injection and
fK = K shows that f is an isomorphism from K to K. Q.E.D.
Given a ring R, let S be the set of non-zero-divisors in R. Then we can
apply the construction of K above using the set Q = { (a, b) I a c R, b E S1 .
Then we obtain a ring K with the following three properties:
(i) R C K.
(ii) Every non-zero-divisor in R is a non-zero-divisor in K, and every
non-zero-divisor in K is invertible.
(iii) Every element of K can be written in the form ab-1 with a c R
and b a non-zero-divisor in R.
The ring K is called the ring of total fractions of R.
We state here an important lemma about prime ideals.
5. POLYNOMIAL RINGS 23
LEMMA 1.4.5. Let PI , ... , P, be prime ideals of a ring R and let I be

an ideal of R.
If I is not contained in any Pj, then there is an element a E I which is
not in any Pi .
I f Pi 9 Pj f o r all i $ j, then there is an element b o f PI n . n Pr
(1 <r<n-1) that is not in any of Pr+I,...,Pn
PROOF. For the first half, we use an induction argument on n. The asser-
tion is obvious if n = 1 , and we assume that n > 1 . If for some i, Pi C Pt ,
then we may omit Pi, and therefore, we may assume that any Pi (i < n) is
not contained in Pn . By the induction hypothesis, there is an element aI
of I which is not in any of PI , ... , Pn_I . If aI is not in P, , then aI
is taken as a. Assume now that aI E Pn Since P,, is a prime ideal not
.
containing any of I, PI , ... , Pn _ I , we see that I PI P1_1 (Z P ,, and there

is an element c E IPI .
P,, with c P,, . Then a = aI + c is the required
element.
For the latter half, since PI . P g P for any j > r, we apply the first
half to PI Pr and Pr+I , ... , Pn in place of I and PI , ... , P,,, respec-
tively. Q.E.D.
5. Polynomial rings
Let R be a ring and consider n letters XI , ... , X,l . By a monomial in
XI , ... , Xn , we mean a form Xi' Xe"" (each e. is a nonnegative rational
integer), and E; e, is called the degree of the monomial. The product of two
a' .. x:)(Xe' .. e9 Xa'+e' ,a,,+e,+
monomials is defined by (XI , I X1,1 ,z
A lin-
ear combination E;= I aim, of monomials m I , ... , ms in X , ... , Xn withI
coefficients a, in R, is called a polynomial in XI , ... , Xn , with coefficients

in R, or over R. Terms whose coefficients are 0 are disregarded. X = 1
so, for instance, Xi XX3 is usually written as Xi X3 . If a polynomial is
rearranged in a form E a! m, (0 a, E R, mi monomial) so that m, mi
for i j, then the maximum of the degrees of these rn is called the degree
of f and is denoted by deg f. The maximum of deg m, before rearrange-
ment is called the apparent degree of f. The degree of 0 (the polynomial
whose coefficients are all 0) is defined to be indefinite. If f = > a,mi with
monomials mJ such that deg mi = deg f for all i, then we say that f is a
homogeneous form (or a homogeneous polynomial).
A polynomial coXn + cI Xn- I + + c,, (c, E R) in one variable X is
called monic if co = 1 .
Addition and multiplication of polynomials are defined by
(E aimi) + (E bjmj) _ E(a; + b,)m1 ,
(Ea1ml) (Eb1m1) = E
m,
T,
mI mA=n:+
ajbk m1.
We see easily that the set of polynomials in XI , ... , Xn over R forms

a ring, which is called the polynomial ring in XI , ... , Xn over R and is
denoted by R[XI , ... , Xn]. Note that R[XI , ... , Xn] coincides with the
polynomial ring R[XI , ... , XX_I ][Xn] in Xn over R[XI , ... , X,_1].
Assume that a ring R is a subring of another ring R' and that the identity
of R coincides with that of R'. Then the subring of R' generated by R
and elements aI , ... , an of R' is denoted by R[a1 , ... , an] similar to the
case of polynomial rings.
THEOREM 1.5.1. We define a mapping 0 of the polynomial ring R[XI , ... ,
Xn] to the ring R[al , ... , an] by
ci,...iXIi
' ... Xni) = cil...iaI ... an
where c. ER. Then 0 is a homomorphism from the ring R[XI, ... , Xn]
onto the ring R[a1 , ... , an] .
The proof is easy and the reader is advised to prove this theorem.
Under these circumstances, we say that aI , ... , an are algebraically inde-
pendent over R if 0 is an isomorphism. Otherwise, we say that aI , ... , an
are algebraically dependent over R. If aI , ... , an are algebraically inde-
pendent over R, we may say that R[al , ... , an] is the polynomial ring in
aI , ... , an over R. On the other hand, the kernel of 0 is called the relation
ideal of aI , ... , an over R.
THEOREM 1.5.2. If R is an integral domain, then the polynomial ring R[X]
over R is an integral domain.
PROOF. Let f = Emo ai X ` , g = E a bi X' be nonzero elements of
R[X]. We may assume that am 0 0, bn 0 0. Then the term of degree
m + n in the product fg is ambnXm+n , and this is different from 0 because
R is an integral domain. Therefore R[X] is an integral domain. Q.E.D.
The remainder theorem and the factor theorem hold also over a general
ring.
THEOREM 1.5.3. If f (x) is a polynomial in one variable x over a ring
R and if a E R, then there is a polynomial q(x) E R such that f (x) =
(x - a)q(x) + r, where r = f (a) . In particular, if f (a) = 0, then f (x) is
divisible by x - a. (The first half is the remainder theorem, and the latter
half is the factor theorem.)
PROOF. We see easily the existence of q(x)(The reader is advised to
.
prove this by induction on degf.) By the substitution of a for x, we see

that r = f (a) . Q.E.D.
COROLLARY 1.5.4. Under the circumstances of Theorem 1.5.3, assume
furthermore that R is an integral domain. If f (ai) = 0 (i = 1, ... , n)
6. UNIQUE FACTORIZATION 25
with mutually distinct elements a1 , ... , an of R, then f (x) is divisible by

(x - a 1) (x -a,). In particular, if n > (apparent degree of f), then f = 0.
6. Unique factorization
A nonzero element p of a ring R is called a prime element if the principal
ideal pR is a prime ideal of R. In the ring of rational integers, prime
numbers are prime elements. For elements a, b of a ring R, if there is an
element c such that a = be (namely, if a E bR) , then we say that a is
a multiple of b and that b is a divisor of a. If there is such an element
c which is invertible (if R is an integral domain, then this condition is
equivalent to aR = bR) , then we say that b is an associate of a.
Let a be a noninvertible element of R. It is obvious that invertible
elements and associates of a are divisors of a. If a has no other divisors,
then we say that a is irreducible. (If R is an integral domain, then this is
equivalent to the condition that if a = be (b, c E R) then one of b, c is
invertible.)
THEOREM 1.6.1. In an integral domain, every prime element is irreducible.
PROOF. Let p be a prime element in an integral domain R, and assume
that p = be with b, c E R. Since be E pR (a prime ideal), either b c pR
or c E pR. We can assume that b E pR. Then b = pd (d c R), and
p = be = pdc. It follows that p(1 - dc) = 0. Since R is an integral
domain, we have that 1 - dc = 0, namely, dc = 1 , which means that c is
invertible. Q.E.D.
A ring R is called a unique factorization domain, if R is an integral
domain and if every nonzero, noninvertible element of R is the product
of a finite number of prime elements. (The reason for the use of the word
"unique factorization" will be seen from Theorem 1.6.2.) In this case, every
irreducible element is a prime element (because it is a product of prime
elements).
THEOREM 1.6.2. An integral domain R is a unique factorization domain
if and only if the following two conditions (i), (ii) hold-
(i) Every nonzero, noninvertible element of R can be expressed as the prod-
uct of a finite number of irreducible elements.
(ii) If a = b1 b,n = c1 cn (b1, cj are all irreducible) then m = n,
and after suitable renumbering of cj, we have that each bl is an associate of
cl (a more rigorous statement might be "there is a permutation n on the set
{ 1, ... , n} so that each bi is an associate of cn1) .
Under these circumstances, every irreducible element is a prime element.
PROOF. Assume first that R is a unique factorization domain. Then (i)
obviously holds, because prime elements are irreducible. We noted just be-
fore the theorem that every irreducible element is a prime element. We shall
prove (ii) by induction on m. If m = 1 , then a is irreducible, and (ii)
holds in this case. Assume now that n > 1 . Since bI R is a prime ideal and
c, cn E b, R , we see that one of ci, say c,, is in b, R . Then c, =b , ci .
Since c, is irreducible, ci is invertible and b, is an associate of c, It fol-.
lows now that b2 bin = (cl C2) C3 . . cn . Here, c', c2 is an associate of C2 , and
.
hence, it is irreducible. Therefore by our induction hypothesis, m -1 = n -1

and, after a suitable renumbering of ci , we see that each b, (i > 1) is an
associate of ci . Thus, (ii) holds.
Conversely, assume that (i) and (ii) hold. Let p be an irreducible element
and suppose that bC E pR. Then be = pq (q E R). Then b, c, q are
products of irreducible elements; that is, b = b, . bs, c = c, ct , q =
. .
q, .. Thus, b, bsc, Ct = pq, . . qu and (ii) implies that some bi

qu . .
or ci is an associate of p. Therefore, b or c is a multiple of p, which

proves that p is a prime element. Since every irreducible element is a prime
element, R is a unique factorization domain. Q.E.D.
REMARK. An integral domain R is a unique factorization domain if (i)
above and the following (ii*) hold.
(ii*) Whenever an element a is a common multiple of elements b, c
and the elements b, c have no noninvertible common divisor, then a is a
multiple of be (in other words, if R is the only principal ideal containing
bR + cR, then bR n cR = bcR).
The reader is advised to prove this as an exercise.
A ring R is called a, Euclidean ring if there is a mapping v of R - {O} to
a well-ordered set W such that given 0 a c R, b E R there are elements
q, r in R such that
b=aq+r, with vr < va or r = 0.
Easy examples of Euclidean rings are (1) the ring of rational integers with
va = Ial and (2) the polynomial ring K[X] in one variable X over a field
K with v = deg.
We say that a ring R is a principal ideal ring if every ideal of R is prin-
cipal. If, furthermore, R is an integral domain, then we call R a principal
ideal domain.
THEOREM 1.6.3. A Euclidean ring is a principal ideal ring. A principal ideal
domain is a unique factorization domain.
If P is a nonzero prime ideal in a principal ideal domain D, then P is a
maximal ideal.
PROOF. Assume first that I is an ideal of a Euclidean ring R and we
are to prove that I is principal. If I = {0} , then I = OR, so we assume
that I {0} . Consider J = {vajO 0 a E I} . Since this is a subset of a
well-ordered set, there is a least element m of J. Take a E I such that
va=m. If bEI,then b=aq+r (q,rER, vr<va or r=0). Thus,
r E I. This and the property that m is the least element imply that r = 0.
6. UNIQUE FACTORIZATION 27
Thus, I = aR, and R is a principal ideal ring. This completes the proof of
the first assertion.
Assume next that D is a principal ideal domain. As for the second asser-
tion, it suffices to show that every nonzero, noninvertible element a can be
expressed as the product of a finite number of prime elements. There is a
maximal ideal MI containing aD. By assumption, Ml = pID (p1 (=- D).
Since MI is a prime ideal, pI is a prime element. Since aD is contained
in MI , a = plat (a, E D). If aI is not invertible, then the same argu-
ment is applied to aI , and so on. Thus, we have a = pI . . p,ta,, with prime
.
elements pi and a E D. Therefore, it suffices to show that an is invert-

ible at some step. Assume the contrary. Then we have an infinite sequence
aD C aID c . . C anD c - . Consider I = Ui aiD. This is an ideal and
. - -
I = bD for some b. Since b E I, there is an m, such that b E a,nD . Then

bD C a,nD c I = bD , a contradiction. Thus the second assertion is proved.
As for the last assertion, letting a be a generator of P , we apply the proof
above. Then we see that a is the product of pI , ... , p,1 such that each pi
generates a maximal ideal. Then PI ... P,, E P, and there is an i for which
pi E P. Therefore, P is a maximal ideal. Q.E.D.
In the rest of this section, we shall prove the following theorem.
THEOREM 1.6.4. If R is a unique factorization domain, then the polyno-
mial ring R[XI , ... , Xn] in n variables over R is a unique factorization
domain. In particular, the polynomial ring K[XI , ... , Xn] in n variables
over a field K is a unique factorization domain.
PROOF. If this is proved in the one variable case, then the general case
is proved by induction on n. Therefore, we consider the one variable case
R[X]. For our purpose, we introduce the notion of a primitive polynomial
and we use two lemmas. A polynomial f (X) in one variable X over a unique
factorization domain R is called a primitive polynomial if every common
divisor of the coefficients of f (X) is invertible.
LEMMA 1.6.5. Under the circumstances, if p is a prime element in R, then
p is a prime element in the polynomial ring R[X].
PROOF. Assume that f(X), g(X) E R[X] are not in pR[X] (namely,
some coefficients of both f(X) and g(X) are not divisible by p). Write
f (X) = >i aiX` , g(X) = >jbj.X' (ai, bi. E R) . Let ad be the one with
least i among those ai which are not in pR. Let be be the similar one
among bi. Then the coefficient cd+e of in the product f(X)g(X)
Xd+e
is E aibd+e-i , i < d implies ai E pR, and i > d implies bd+e-i E pR.

Therefore, we see that cd+e - adbe E pR. Since adbe is not in pR, we see
that cd+e is not in pR, thus f (X)g(X) pR[X], which proves that p is a
prime element in R[X]. Q.E.D.
REMARK 1. In the lemma above, the assumption that R is a unique fac-
torization domain is unnecessary. (Cf. Exercise 1.6.1.)
REMARK 2. It follows from the lemma that if f(X), g(X) are primitive
polynomials, then the product f(X)g(X) is also primitive. (The reader is
advised to prove this.)
LEMMA 1.6.6. Under the circumstances, let K be the field of fractions
of R. Assume that f(X), g(X) E R[X] are such that g(X) is primitive
and such that f(X) is divisible by g(X) as polynomials in K[X], namely,
f(X) = g(X)h(X) with h(X) E K[X]. Then f(X) is divisible by g(X) as
polynomials in R[X].
PROOF. Considering the common denominator of the coefficients of h (X) ,
we see that there is a nonzero element a of R such that ah(X) E R[X].
Considering the common divisors of the coefficients of ah (X) , we can write
ah(X) = bh'(X) with b E R and primitive h'(X). By cancellation of the
common divisors of a, b, we can assume that a and b have no common
divisor (except for invertible elements). Then af(X) = bg(X)h'(X). If a
prime element p is a divisor of a, then p is a prime element in R[X], and
therefore, p must divide one of b, g(X), h'(X). This is a contradiction.
Therefore a is invertible and h(X) E R[X]. Q.E.D.
COROLLARY 1.6.7. Under the circumstances, a primitive polynomial p(X)
is a prime element in R[X], if p(X) is a prime element in K[X].
PROOF. Assume that h(X), g(X) E R[X] and h(X)g(X) E p(X)R[X].
Since p (X) is a prime element in K [X ] , one of h (X) , g (X) , say h (X) , is
divisible by p (X) in K[X ] . Since p (X) is primitive, h (X) is divisible by
p(X) in R[X]. Q.E.D.
PROOF OF THEOREM 1.6.4. Let f (X) be a nonzero element of R[X].
Then f (X) = a fi (X) with a E R and primitive fl (X) . Lemma 1.6.5
shows that a is the product of a finite number of prime elements (in R[X]).
Since K[X] is a unique factorization domain by Theorem 1.6.3, f1(X) is
the product of a finite number of prime elements, say p (X), ... , pr (X )
1
in K[X ] . For each pi(X) (i < r), multiplying common denominators of

coefficients and dividing the common divisor of coefficients, we may assume
that pr (X) is primitive if i < r. By Lemma 1.6.6, we see that fl (X) is
divided by p1(X) , ... , pr_ 1(X) successively, and therefore, Pr(X) E R [X ] .
Since f1(X) is primitive, we see that pr(X) is also primitive. By Corollary
1.6.7, these pl(X) are prime elements. Q.E.D.
We give here an important remark which follows form the unique factor-
ization in the polynomial ring in one variable over a field.
THEOREM 1.6.8. Let f (X) be a polynomial in X over afield K. Consider
divisors of f (X) of the form {J (X-ci) (ci E K) , and choose one having the
largest r. Then the sequence c1 , ... , Cr is uniquely determined disregarding
the order but regarding the .multiplicities. If C E K satisfies f (c) = 0, then c
appears within the sequence.
7. MODULES 29
These cl are called the roots of f (X) in K , and the multiplicity of a root
c appearing in the sequence is called the multiplicity of the root c. A root
with multiplicity > 2 is called a multiple root and a root of multiplicity > m
is called an m-ple root.
PROOF. The first half follows immediately from the uniqueness of factor-
ization, and the latter part follows from the factor theorem. Q.E.D.
7. Modules
Let R be a ring. A module M is called a left R-module if an operation
(multiplication) of R on M is defined in such a way that for any r E R and
for any m E M their product rm (E M) is defined and the multiplication
satisfies the following three conditions:
(i) r(m I + m2) = rm 1 + rm2 for all r E R, m , m2 E M'
(ii) r1(r2m) = (r1 r2)m and (r1 + r2)m = rim + r2m for all r1 , r2 E R1
mEM.
(iii) 1 m = m for 1 , the identity of R, and for any m in M.
We define a right R-module similarly with multiplication of elements of
R to elements of M from the right-hand side.
When a left R-module M is given, we can define such a multiplication by
r E R, m E M = mr = rm (because of the commutativity of R), and M
becomes a right module. Therefore, from now on, we do not distinguish left
and right for R-modules. So, we simply say that M is an R-module or M
is a module over R. When M is an R-module, we mean by a submodule a
submodule which is an R-module, unless otherwise specified. (So, it may be
proper to call an R-submodule.)
Let M , N be R-modules (R being a ring). Then a homomorphism f of
additive groups M to N is called an R-homomorphism, or a homomorphism
of R-modules, if it satisfies the condition that r(f m) = f (rm) for all r E
R, m E M. If it is clear that we are speaking of R-modules, then by a
homomorphism we mean an R-homomorphism. Similarly for isomorphisms;
we consider isomorphisms of modules which are R-homomorphisms. Then,
we obtain results similar to those we proved in 2. For instance, Theorem
1.2.1-Theorem 1.2.3 are modified as follows.
THEOREM 1.7.1. (i) If N is a submodule of an R-module M, then the
natural homomorphism from M onto MIN is an R-homomorphism.
(ii) If 0 is a homomorphism from an R-module M to an R-module M',
then the kernel N of 0 is a submodule of M (as an R-module) and OM
is isomorphic to MIN. In this case, there is a one-to-one correspondence
between submodules H of M containing N and submodules H' of q5M in
such a way that each H corresponds H' = OH.
(iii) If N, H are submodules of an R-module M, then the natural homo-
morphism from M onto M/N gives an isomorphism from H/(N fl H) to
(H + N)/N IN.
(iii) is called the isomorphism theorem for modules. (The reader is advised
to prove these results.)
EXERCISE. State the modification of Theorem 1.2.9 to the case of modules
over a ring.
An R-module (R being a ring) which is generated by a finite number of
elements (namely, a module which can be written in the form a R + +anR)
I
is called a finitely generated R-module, or simply, a finite module.
Elements mI , ... , mn of an R-module M are called linearly independent
(over R) if En I rlmi = 0 with ri E R, implies rr = 0 (i = 1 , ... , n). A
subset S of M is said to be linearly independent if every finite subset of S
consists of linearly independent elements.. For convenience, we regard that
the empty set is linearly independent.
If an R-module M is generated by a linearly independent subset S, then
we say that S is a linearly independent base, or a free base, of M and that
M is a free R-module, or simply, a free module.
THEOREM 1.7.2. Every module over a field K is a free module.
PROOF. Let F be the set of all linearly independent subsets of a K-module
M. Then F is an inductive set by the inclusion relation. This is proved as
follows. (i) F is not empty because the empty set belongs to F. (ii) Let
{S2IA, E Al be a well-ordered subset of F. If bI , ... , bin are mutually
distinct elements of S* = US,,, then 1% E A, {bI, ... , b,1 } c S.. This
means that S* is linearly independent and S* E F. Obviously if S* D S2
(for all A). If S D S. (for all A), then S D US2 = S* . Therefore S* is
the supremum of {S2 JA E A}. Thus, we proved that F is an inductive set.
Therefore, by Zorn's Lemma, there is a maximal member B of F. Let N be
the submodule generated by B over K. It suffices to show that N = M. Let
m be an arbitrary element of M. We are to show that m E N. If M E B,
then m E N. Assuming that m is not in B, we consider B U fm}. By the
maximality of B, B U {m} is not linearly independent, which means that
there is a finite subset S of BU{m} which is not linearly independent. Since
B is linearly independent, m ES, and S= fm, bI, ... , bn } with b, E B.
Then there are ci E K such that co m + E cib; = 0, (c0, cI , ... , cn) 54
(0, ... , 0) . Since bI , ... , bn are linearly independent, we have co 54 0.
Multiplying -co we may assume that co = -1 . Then m = cibi and
m E N. Q.E.D.
We can see the following result by the proof above.
THEOREM 1.7.3. If M is a module over a field K and if S is a linearly
independent subset of M, then there is a linearly independent baseof M con-
taining S.
THEOREM 1.7.4. If a module M over a field K has a linearly independent
base mI , ... , mn consisting of n elements, then every linearly independent
base of M consists of n elements.
7. MODULES 31
This number n is called the length, or the dimension of M, and we

denote it by lengthKM or length M. (The word "length" came from that of
composition series (cf. Theorem 1.7.6). The word "dimension" came from
that of a vector space (note that modules over a field are vector spaces). The
Jordan-Holder-Schreier Theorem asserts that if a group has a composition
series, then its length is independent of the choice of composition series.
Though the theorem is very important, we need it only in the module case
in Chapter III, and we have stated Theorem 1.7.6 in that case. The general
case is proved similarly. (Cf. Exercise 1.7.6.)
PROOF. Though we shall later prove this in a more general form in Theo-
rem 1.7.6, we give here a familiar proof in the case of vector spaces.
First we shall show that if v1 , ... , v,. are linearly independent elements
of M, then r < n. For the purpose, we shall prove that under a suitable
renumbering of m1 , ... , Mn , it is true that v1 , ... , vi , m;+1 , ... , mn form
a linearly independent base of M (as far as i < r) , by induction on i.
If i = 0, then the assertion is obvious, and we assume that i > 0 and
that v1 , ... , Vi-1 , mi , ... , Mn form a linearly independent base of M.
v, can be expressed as v i = E`.=1 J=
cj v + EJ"= . cJ . mJ . with c E K. Since
j .
J
vl , ... , vi_1 , vi are linearly independent, at least one of ci , ... , cn is not 0.
By renumbering mi, ... , Mn suitably, we may assume that ci 0. Consider
v1 , ... , vi , mi+1 , ... , mn . Suppose E1=1 d1v1 + d1m1 = 0 (d1 E
K). Then
;-1 (i-1 ,: n
0=>d1v1+di (civi+cimi) + E djm1.
j=1 j=1 j=i j=i+1
Since v1, ... , vi_1, mi , ... , Mn are linearly independent, and since dici is
the coefficient of mi in this equality, we have d;c; = 0. Since ci 0, we
have di = 0. Thus >'-1 dj v j + En=i1 dj m1 = 0, and every d1 is 0. It now
follows that v1 , ... , vi , mi+1 , ... , Mn form a linearly independent base of
M.
Now, suppose that r > n. Then what we have proved shows that v1 , ... ,
vn form a linearly independent base. Therefore r < n. If v1 , ... , V,.
form a linearly independent base with r < n, then we have a contradic-
tion, by employing v1 , ... , yr instead of m1 , ... , Mn in our observation
made above. Q.E.D.
The set of homomorphisms from a module M to a module N is denoted
by Hom(M, N). In this set, we define the sum of two elements 01, 02
by (01 + Cat)m = q51m + q52m . Then we easily see that Hom(M, N) is a
module, and 0 in this module is the homomorphism 0 such that OM = 0.
If M, N are modules over a ring R, then the set of R-homomorphisms
from M to N is denoted by HomR (M , N), and this is a submodule of
Hom(M, N). But, if it is obvious that we are observing R-modules, we
simply write Hom(M, N) for HomR(M, N).
THEOREM 1.7.5. Under the circumstances, if N is a module over a ring R,

then Hom(M, N) becomes an R-module by defining the multiplication of an
element a of R to an element 0 of Hom(M, N) by (aq5)(m) = a(q5m).
The proof is easy and we omit it.
REMARK. In the theorem above, if M is also an R-module, then
Hom(M, N) can have an R-module structure by defining (ao) (m) = d(am) .
This multiplication coincides with the one in the theorem if 0 E Horn R(M, N).
Now, we consider a generalization of Theorem 1.7.4. Let M be a module
over a ring R. A descending sequence Mo D D Mr of submodules of M
is called a descending chain, and r is called the length of the chain. If there
is a submodule M' such that Mi D M' D Mi+I , then we can have a longer
chain by inserting M'. This is called a refinement of the chain. If a chain of
submodules starts with M, ends with 10}, and if there is no refinement of
the chain, then we call the chain a composition series of M. The following
is the module case of the Jordan-Holder-Schreier Theorem.
THEORM 1.7.6. If a module M over a ring R has a composition series
M = Mo D M = {0} of length n, then for an arbitrary descending
chain MM D ... D Mn, ofsubmodules M,, we have n' < n. Moreover, n = n'
if MM D Mn, is a composition series. In this case, there is a permutation
n on { 1, ... , n j such that M,_ I /M' ^' M(ni)- /Mni .
I
PROOF. We may assume that n is the length of the shortest composition

series, and we want to prove the assertion by induction on n . Note that if
N is a submodule of M, then there is a one-to-one correspondence between
composition series M/N = Ho D MI D ... Mr = {0} of M/N and de-
scending chains M = Mo D Mi D . . D Mr = N which start with M, end at
.
N, and without any refinement, in such a way that Mi = M1/N. (Consider

the natural homomorphism from M to M/N and the correspondence
-I ; cf. Theorem 1.2.2.)
We may assume that MM = M and that Mo M' has no refinement.
Consider N = MI n M'.
(i) The case N :A 101. MI has a composition series of length n - 1 , and
the assertion (with n - 1 in place of n) is true for MI by our induction
hypothesis. Therefore, taking a composition series MI = Mi D D Mr =
N D Mr+I D . . . D Mn = {0} which is a refinement of MI 2 N D 101,
we see that there is a permutation a on { 2, ... , n } such that M,* /Mi* ^_
I
Ma i _ I /Mai (i = 2, ..., n). Letting al = 1 , M0* = M, we see these

isomorphisms including i = I.
Next, consider MIN. M/N has a composition series Mo IN D D
Mr /N . Therefore, there is a composition series of M/N of length r (< n)
which is a refinement of M/N D M' IN D {0} , namely, there is a descending
chain M = Mo Mi = Mi' M2 Mr' = N which has no refinement
7. MODULES 33
and there is a permutation z on { 1 , ... , r} so that Ml'' /M'' ^- Mzi- l /Mzi

Letting MM' = M* and zi = i for i= r+ 1, ... , n, we see these isomor-
phisms for all i = 1 , ... , n. Therefore, Mi has a composition series of
length n-1 . Thus n' < n , and if n' = n , then it follows that Mi : M1
is a composition series of Mi and there is a permutation A on { 1 , ... , n }
(Al = 1) so that M!_ i /M! M,_1/M,. for i = 1, ... , n. Thus, the prod-
uct of a, r, A satisfies the requirement on it in our assertion.
(ii) The case N = J01. Since there is no module between M and Ml ,
we have Mi + Mi = M. Therefore, by the isomorphism theorem, we have
M/Ml M /(Ml n Mi) = Mi , M/M M1 M. Thus, there is no module
between M, and {0} and also between Mi and 10). Therefore, both
M D Ml D {0} and M D M' D {0} are composition series, and the assertion
is also true in this case. Q.E.D.
An R-module M which has a composition series is called an At-tin mod-
ule in this book. (Usually, modules satisfying the minimum condition on
submodules are called Artin modules. But, the Artin modules we discuss in
this book are those satisfying the stronger condition. So, we defined them
this way here in just for convenience in this book.)
The length of the composition series is called the length of M and is de-
noted by lengthRM or length M. (Note that the length defined at Theorem
1.7.4 is a special case of this definition.) If R is an Artin module as an
R-module, then R is called an Artin ring.
THEOREM 1.7.7. Assume that a field K is a subfield of an At-tin ring R
and that, for any maximal ideal I of R, lengthKR/I is a finite value s
(independent of I) (here, K is identified with the image of K by the natural
homomorphism of R to R/I). Then
s(lengthRR) = lengthKR.
PROOF. Consider a composition series R = Mo D Mi D . . . D 11M11 = {0} .

Each M!_ i /M1 is isomorphic to R/I as an R-module, with suitable maximal
ideal I. Since lengthKR/I = s, each Mj_ /Mi has a composition series of
I
length s as a K-module. Therefore, R has a composition series of length

sn as a K-module. Q.E.D.
THEOREM 1.7.8. Every prime ideal of an Artin ring is a maximal ideal.
PROOF. Let P be a prime ideal of an Artin ring R. If the integral domain

R = R/P has a nonzero, noninvertible element a, then in the sequence aR D
a2R D D a"R D . . . , we have a"R 54 a"" W (indeed, if anR = a"+' R ,
a"+'
then an = b (for some b E R) and a" (1 - ab) = 0, which implies
1 - ab = 0, contradicting the noninvertibility of a) , a contradiction to the
descending chain condition. Q.E.D.
8. Symmetric forms and alternating forms

We consider the polynomial ring P = R[x1 , ... , xn ] in n variables over a
ring R and the symmetric group Sn of degree n acting on the set
{ 1 , ... , n} I. We regard Sn as acting on {x1 , ... , Xn} by xi -+ xe,i (Q E
Sn) . For f (x1 ... , Xn) E P , let a f denote f (xa , ... , XQn) . Note that if
1
U, z E Sn , then (aT) f = a (z f) . A polynomial f E P is called a symmet-

ric form if a f = f for every a in Sn . f is called an alternating form if
a f = f for every a in Sn and f is not symmetric. If a1 , ... , an are
elements of a ring containing R and if f (x1 , ... , Xn) is a symmetric form,
or an alternating form, then f (a1 , ... , an) is called a symmetric form, or an
alternating form, of a1 , ... , an , respectively.
The following n polynomials are symmetric forms, and they are called
the elementary symmetric forms. (Though we write on x1 , ... , xn only, the
same is applied to elements, say, a1 , ... , an.)
SI = XI +... + Xn
S2 = E Xixj
i<j
sr = E xi xi ...xi
I 2 r
Sn = XI X2 ... Xn
THEOREM 1.8.1. Every symmetric form can be written as a polynomial in

the elementary symmetric forms. Namely, the set of all symmetric forms in
X1,...,Xn is R[s1,...,sn].
PROOF. To each monomial M = X1' . xn' , we associate the (n + 1)-
. .
tuple d (M) = (j j=I ij, i 1, ... , in) , and we introduce an order on the set of
monomials by the lexicographical order of d (M) . Namely, for monomials
M1 , M2, MI < M2 means that, letting d(M1) _ (d0, ... , do j) (j = 1, 2),
d01 < d02 , or d01 = d02 and d11 < d12 , or , namely, for some j ,
dj1 < dj2 and dkl = dk2 if k < j. We shall prove the assertion by induction
on d (M) where M is the greatest monomial appearing in a given symmetric
form f (x1 , ... , xn) .
If only the least monomial 1 appears in f , then f is an element of R , and
the assertion is obvious. Let M be the greatest monomial among the mono-
mials appearing in f . Then our induction hypothesis is that if h is a sym-
metric form such that d (M) > d (N) for every monomial N which appears
in h, then h E R[s1 , ... , sn]. Write M as xi' . xn" . Since f is symmet-
ric, xn'1 xn'n appears in f for any U E Sn . Therefore, by the greatestness
of M , we have e1 > e2 > . . > en . Consider g = Sn"sn"-1 -e" S22 -e 3s1 I -e
.
2
8. SYMMETRIC AND ALTERNATING FORMS 35
g is a symmetric form and the greatest monomial appearing in g is M.

Letting a be the coefficient of M in f , we consider f - ag . Then, by the
induction hypothesis, we have
f - ag E R[s1 , ... , sn] and f E R[s1 , ... , sn]. Q.E.D.
Now, we consider alternating forms. If the characteristic of R is 2, then
-1 = 1 , and there are no alternating forms. In the other case, the following
is an alternating form and is called the difference product of x1 , ... , xn or
the simplest alternating form in x1 , ... , xn :
A=fl(x,-xj).
i<j
Though the next theorem holds if the characteristic of R is either 0 or an
odd number, for simplicity of the proof, we prove this under the assumption
that R is a unique factorization domain of characteristic 2.
THEOREM 1.8.2. If f (x1, ... , xn) E R[x1, ... , xn] is an alternating form,
then f is the product of the difference product A and a symmetric form.
Before proving this, we mention some facts about the symmetric group
S,,. The permutation that interchanges i and j (i j) and leaves each
of other elements of {1 , ... , n} fixed is denoted by (i, j), and a permu-
tation of this type is called a transposition. We denote by (i1 , ... , i,.) the
permutation such that (i) ij (j < r) is mapped to ij+1 , (ii) it is mapped to
i1 , and (iii) any remaining elements of { 1, ... , n} are fixed. This type of
permutation is called a cyclic permutation of length r. (A transposition is a
cyclic permutation of length 2, and the identity can be regarded as a cyclic
permutation (i) of length 1.)
LEMMA 1.8.3. If a,TESn and T=(Q a2...n ) (a Ti) then QTQ-I is
the permutation obtained by mapping the letters in T by a, namely
QTQ
-I = a1 a2 ... an
Qa1 ... aa2 aan
PROOF.
-I 1 2 n Q1 a2 .. an
a1 a2 ... 2 ... n
an 1
=a
(a! a2 .. an
Q.E.D.
a1 a2 ...
an
LEMMA 1.8.4. An arbitrary permutation can be expressed as the product of

cyclic permutations such that no letter appears commonly in two of these cyclic
permutations.
PROOF. Let a be in Sn . Choose m such that am m, and take the
smallest natural number r1 such that a"Im = m. Let n be (m, a m , ... 1
ri-1
Q m) . Suppose that we have defined cyclic permutations n , ... , 71c-1 1 1
each of which is defined similarly to nl and satisfying the condition that no

letter appears commonly in two of these (Note that this is true if c = 2.)
irk.
Now, consider a letter k which is mapped by a to a letter different from

k. If k appears in some n1, then k is mapped to ak by it1 and both k
and ak are mapped by other ni to themselves, respectively. Therefore, k
is mapped to ak by nl 7Ec_1 . Therefore, if every k such that ak k
appears in some one of the n1, then a is the product of the Otherwise,
it1.
taking a letter k such that ak # k , but k does not appear in any of it1 ,
we define nc similarly. If we proceed in this way, then those k such that
ak 54 k are exhausted, and we complete the proof. Q.E.D.
LEMMA 1.8.5. Every permutation can be written as a product of transposi-

tions. Whether the number of these transpositions is even or odd is determined
by the original permutation.
We call the permutation an even permutation or an odd permutation, ac-
cording to the number of transpositions is even or odd.
PROOF. We have (a1 , a2 , ... , ar) = (a, , ar)(al , ar-1) ... (a1, a3)(a1, a2).
This equality and Lemma 1.8.4. imply the first half. Consider the difference
product A. Then, as we show below, with any transposition (i, j) (1 < i <
j < n) , we have (i, j )A = -A. Therefore, a given permutation a is even or
odd according to whether aO = 0 or aA = -A.
Proof of (i , j )A = -0 . Those factors (xs - xl) of A such that (i , j) (xs -
XI) (xs - x,) are divided into 5 classes as follows.
(1) (xs -x,), (xs -xj) with s < i,
(2) (xj - xe) , (x1- x,) with t > j ,
(3) (x,-xt) with i<t<j,
(4) (xs - xx) with i < s < j ,
(5) (x, - xj).

The product of the factors in (1) and (2) is invariant under the action of
(i, j). Factors in (3) and (4) are mapped to their minus elements. But the
numbers of factors in (3) and (4) are the same. Therefore, the product of
these factors is also invariant. The remaining factor (x1- xj) is mapped to
-(x1 - xj). Therefore, (i, j)E = -A. Q.E.D.
The set of even permutations in Sn forms a normal subgroup of Sn . This
subgroup is called the alternating group of degree n .
PROOF OF THEOREM 1.8.2. For each Or E Sn , we define c(a) (= f 1)
by a f = c(or)f. Obviously, c(a_I) = c(a)-I = c(a). Since (at) f =
a(Tf) = a(c(r)f) = c(Z)(af) = C(T)C(U)f, we have c(at) = C(U)C(T). Set
b = c((1 , 2)). Then for each transposition (i, j) , there is a a such that
all , = (i, j) (Lemma 1.8.3), and we have c((i, j)) = b. Thus, if a
2)Q-I
is an odd permutation, then c(a) = b (because b = 1) , and if a is an even

permutation, then c(a) = 1 . Since f is not symmetric, we have b = -1 and
( i, j)f = -f . We consider h = f (x1 , ... , x, 3 ... , xj-1x , , xj+1 , ... , xn)
5
EXERCISES 37
by substituting xj = xi. Then we have h = -h. Since R is an integral

domain of characteristic 2, we see that h = 0. Thus, by the factor theo-
rem, we see that f is divisible by xi - xj . Since R is a unique factorization
domain and different linear factors of 0 have no common factor, we see that
f is a multiple of A. Thus f = Og, g E R[x1 , ... , xn ] . For any a E S,,,
c(a) f = a f = (aL)(ag) = c(a)A(ag), and we have g = ag. Therefore g
is symmetric. Q.E.D.
Exercises
1. Prove that if a is an element of a group, then:

(1) If there is a natural number n such that a" = 1 , then the smallest
such n is the order of a.
(2) If there is no such natural number n , then the order of a is oo and
the cyclic group generated by a is {a =1, a, a-I, a2, a-2 , ... , a" ,
a-",...} (aà' for ij).
2. Prove that for any given natural number n, there is a cyclic group of
order n.
3. Prove that the order of an element a of a finite group G is a divisor of
#(G).
4. Give an example of a noncyclic group G of order 6. Confirm that G has
elements of order 2 and order 3, respectively. Considering a subgroup
H of order 2, confirm also that right cosets and left cosets by H (three
cosets each) give different partition of G.
5. Prove that if the order of a group is a prime number, then the group is a
cyclic group.
6. Let G be a permutation group on a set M, and let m be an element
of M. Then the set {gm1g E G} is called the locus of m by G, or the
G-locus of m. Prove that G-loci of elements of M give a partition of
M.
7. Let m, n be the respective orders of elements a, b of a commutative
group G. Prove that if m and n are coprime, then the subgroup gen-
erated by a, b is a cyclic group of order mn and is generated by ab.
8. Let Z be a finite cyclic group and let q be a divisor of #(Z). Prove
that xq = 1 has exactly q solutions in Z .
1. Prove that if NJ , N2, ... , Nr are normal subgroups of a group G, then

NI n N2 n n N is also a normal subgroup of G.
2. Assume that N is a normal subgroup of a group G and that H is a
subgroup of G. Prove that N n H is a normal subgroup of H.
3. Let N be a subgroup of a group G. Prove the following equivalence:

N is a normal subgroup of G if the left cosets by N coincide with the
right cosets by N.
4. Regard the set Z of rational integers as an additive group. Let n be a
natural number and denote by nZ the additive subgroup consisting of
multiples of n. Prove that Z/nZ is a cyclic group of order n.
5. For a subset K of a group G, the sets N(K) = {a E GIaK = Ka},
Z(K) = {a E Gjak = ka for all k c K} are called the norfnalizer and
the centralizer of K, respectively. Prove the following assertions.
(1) N(K), Z (K) are subgroups of G.
(2) Let K* be the subgroup generated by K. Then K* is a normal
subgroup in both of N(K) V K* and Z(K) V K*. Furthermore,
N(K) c N(K*) and Z(K) = Z(K*).
(3) If K is a subgroup, then N(K) = G if K is a normal subgroup.
(4) Z(G) is a normal subgroup.
6. Let G be a group. Prove that N is a normal subgroup of G and GIN
is commutative iff N is a subgroup containing [G, G].
7. Prove that if N is a normal subgroup of a group G and if H is a
subgroup of G, then [N, H] C N.
8. Assume that 0 is a homomorphism from a group G to some other group
and that H, K are subgroups of G. Prove that
q[H, K] = [OH, OK].
9. Prove that for a group G, the conditions (1)-(4) are equivalent.
(1) G is solvable.
(2) There is a normal subgroup N such that both N and GIN are
solvable.
(3) For any normal subgroup N, both N and GIN are solvable.
(4) There is a descending sequence G = Go D . D G11 = {1 } of
. .
subgroups G such that (1) G1 is a normal subgroup of G,_ , and 1
(ii) G._ I/ Gi is commutative, for i = 1, 2, ... , n.

(5) Show that a finite group G is solvable iff there is a descending se-
quence as in (4) above such that each Gj_ 1 /G1 is a cyclic group
whose order is a prime number.
10. Prove that if G1 , ... , G11 are nilpotent groups, then the direct product
G1 x . . . x G11 is nilpotent.
11. Prove that a subgroup of index 2 is always a normal subgroup, namely,
that if H is a subgroup of a group G and if #(G/H) = 2, then H is a
normal subgroup.
12. Assume that G is a finite commutative group. Prove that
(1) G is a cyclic group if every Sylow subgroup is cyclic.
(2) If G is not cyclic, then there is a divisor h of #(G) such that G has
at least two distinct cyclic subgroups of order h. (Hint. Consider
a noncyclic Sylow subgroup S an cyclic subgroups of S. Choose
EXERCISES 39
one, say (a) , with the largest order. Choose an element b of S

which is not in (a). Let h = #((b)).)
3
1. An element e of a ring R is called an idempotent if it satisfies the con-

ditions e2 = e and e 0 0. Prove the following assertions.
(1) If a ring R is the direct sum R . . Rn of rings R 1 , ... , Rn ,
1
then the identity e, of R is an idempotent and the identity of R

is
(2) Conversely, if the identity of R is the sum of idempotents el, ... , en
and if i j implies e1ej = 0, then R is the direct sum of rings
e1R,...,enR.
(3) Let e; be as above and let M be an R-module. Then e,M is a
module over the ring e.R and M is the direct sum of its submodules
elM,...,enM.
2. Let I, , ... , In be nonzero ideals of a ring R. Prove that if R is
the direct sum of Il , ... , In as modules, then R is the direct sum of
Il , ... , In as rings.
3. Let I, , ... , In be ideals of a ring R.
(1) Prove that if Il + I2 = R then
(i) I1 n I2 = I1I2 ,
(ii) Iin + IT = R (m, m' are natural numbers).

(iii) R/(II n I2) is the direct sum of ideals I,/(I, n I2),
I2/(I, n I2) , which are isomorphic to R/I2 , R/Il , respectively,
as rings. Therefore
R/(Il n I2) ^_' R/I1 R/I2
as rings.
(2) If I + I = R for j, then the ring R/(Iin' n ... n l inn) is
i
isomorphic to the direct sum R/Iin' . . R/I,1 of rings R/I;n
.
(m, are natural numbers).
1. Consider the homomorphic image Z/nZ of the ring Z of rational inte-

gers with a composite number n. Show that Z/nZ has zero-divisors, by
using a factorization of n . Show also that Z/nZ has nilpotent elements
if n has a square factor.
2. Find the field of fractions of the ring Z of rational integers.
1. Consider polynomials fl, f2 in X1 , ... , Xn over a ring R. Assume

that deg f1 =n, , deg f2 = n2 .
(1) Prove that deg f1 f2 = n1 + n2 if R is an integral domain.

(2) Prove that if R is not an integral domain, then deg f1 f2 < n + n2 ,
1
and there are fl , f2 such that deg fl f2 < n1 + n2 .

(3) What can one say about deg(f1 + f2)?
2. Assume furthermore that fl, f2 are homogeneous.
(1) Prove that fl f2 is homogeneous.
(2) Prove that if degf1 = deg f2 , then f1 + f2 is homogeneous.
6
1. Generalize Lemma 1.6.5 as follows: Let P be a prime ideal of a ring R,
and consider the polynomial ring R[X1 , ... , Xn] over R. Prove that
the ideal of the polynomial ring generated by P is a prime ideal.
2. Under the assumptions in 1., prove that if I is an ideal of R, then
R[X1 , ... , Xn]/IR[X1, ... , Xn] is isomorphic to (R/I)[X1 , ... , Xn].
3. Prove that prime ideals of the ring Z of rational integers are {0} or pZ
with prime numbers p. Prove also that if p is a prime number, then
pZ is a maximal ideal and Z/pZ is a field consisting of p elements.
4. Consider the polynomial ring R[X] in one variable over a unique factor-
ization domain R. Prove that an element f of R[X] is a prime element
if and only if f is either a prime element in R or a primitive polynomial
which is a prime element in the polynomial ring in X over the field of
fractions of R.
7
1. Show that an arbitrary additive group can be regarded naturally as a

module over the ring Z of rational integers.
2. Give a simple example of a Z-module which is not free.
3. Show that a module M over a ring R is a free module with linearly
independent base u1 , ... , us if and only if the mapping of R to M
defined by r -+ u . r is an R-isomorphism from R to u! R for each i
and M is the direct sum of u R , ... , u,, R .
1
4. Let M be an Artin module over a ring R and let N be a submodule.

Prove that
length M = length M/N + length N.
5. Let Ri (i = 1 ... , n) be Artin rings such that each Ri has a unique
maximal ideal Ji. Let R be the direct sum of these R, . Assume that R
contains an Artin ring S which has a unique maximal ideal I and that,
for each J..'=R1ED Ri_1ED JiED RI+IED ED Rnwehave JinS=I
and [R/J1': S/I] = s; < oo . Prove that
n
lengths R = s.lengthR R1.
r=I
6. By a normal chain of a group G, we mean a descending sequence G =
Go D G1 D D Gn of subgroups G such that each Gi is a normal
EXERCISES 41
subgroup of Gi _ I . If, furthermore, Gn = { 1 } and the sequence has no

refinement as a normal chain, then we call this a composition series of
G. According to this definition, prove a result similar to Theorem 1.7.6
(namely, the Jordan-Holder-Schreier Theorem).
1. Let R be a field whose characteristic is different form 2. Consider an

element f of the polynomial ring R[x1 , ... , xn] and the alternating
group An of degree n. Prove that a f = f (for all a E An) iff f is the
sum of a symmetric form and an alternating form (or 0). (Hint. Given
an f with a f = f for all a E An consider f - if with z = (1, 2).)
2. Let a, 'c be permutations on a set M. Prove that zQ = at if there is
no element m of M such that both am and im are different from m.
3. Let x"+c1x"-I+ +cn be the expansion of fI" (x-ai) as a polynomial
I
in x. Show that the elementary symmetric forms of a1 , ... , an are

(-')CI , C21 ... , (_l)rcr, ... , (-1)"Cn .
4. Suppose that a symmetric form f E R[x, , ... , xn] is written as f =
fo + - + fd with homogeneous polynomials f of degree i. Show that
each f is symmetric. Show also that if f is an alternating form, then
nonzero f are alternating forms.
5. Besides the elementary symmetric forms s1 , ... , sn of x1 , ... , xn , we
consider pj = >i x' . Prove the following inequalities:
(1) If 1 < j < n, then
pj - pj-1s1 +...+(_1)j-rprSj-r+...+(-1)j-IPisj_I +(-1)jjsj = 0.
(2) If j > n , then
pj -pj_1s1 + ... + prsj-r + ... + l)"pj_nSn = 0.
(Hint. Consider f (z) = f li(z - xi) E R[x1, .. , xn , z] and set
f'(z) _ (d/dz)f(z). Then f'(z)/f(z) = >i 1/(z - xi) = (n/z) +
(p1 /z2) + ... + (pj/zj+I) + ,
.. Multiplying by f (z)/zn , we have
(_1)n-I(Sn-1/zn-I)
n - (n - 1)(sl/z) + (n - 2)(s2/z2) + ... +
_ (1 - (sl/z) + (s2/z2) + ... + (-1)"(sn/z"))
x (n+(p/z)+'..-+(pj/zj)+...).
6. Let K be a field of characteristic different from 2 and consider the

symmetric group Sn of degree n whose action on the polynomial ring
K[x1 , ... , xn] is defined by permutations of x1 , ... , xn . Confirm that
this action is naturally extended to an action on the rational function field
K(x1 , ... , xn) . Then prove that if an irreducible fraction g(x1, ... , xn)/
f (x1 , ... , Xn) is invariant by the action of any element of Sn , then
f(x1 , ... , xn) , g(x1 , ... , Xn) are symmetric forms.
CHAPTER II
Algebraic Extensions of Finite Degrees
1. Basic notions
Let K be a subfield of a field L, and let S be a subset of L. Then the
least subfield containing K and S is denoted by K(S) and is called the field
generated by S over K or the field adjoined S to K . If S = {al , ... , an } ,
then K(S) is denoted by K(al, ... , an). If Kl K2 are subfields of L,
,
then the field KI (K2) = K2 (Kl) generated by KI and K2 may be denoted

by Kl V K2, as in the case of groups.
Since a field K is an integral domain, its subring n generated by 1 is an
integral domain, which is called the prime integral domain of K. The field
of fractions of 7r, namely, the least subfield of K, is called the prime field
of K.
THEOREM 2.1.1. The characteristic of an integral domain is either 0 or a
prime number. If the characteristic of a field K is p : 0, then its prime
integral domain coincides with the prime field, which is isomorphic to Z/pZ
(Z = the ring of rational integers). If K is of characteristic 0, then its prime
integral domain is isomorphic to Z, and the prime field is isomorphic to the
field Q of rational numbers.
PROOF. Assume that the characteristic p of an integral domain is not 0.
If p = ab with natural numbers a, b > 2, then
(1+...+1) (1+...+1) =0,
a b
and one of the factors must be 0, which contradicts the leastness of P.

Therefore p is a prime number. The mapping Z -* K given by a a1
is a homomorphism. Therefore, if p is the characteristic of K, then 7G is
isomorphic to Z/pZ. If p is a prime number, then Z/pZ is a field (cf.
Exercise 1.6.3) and 7G is a field.
We consider now the case p = 0. In this case, the mapping a E Z - a 1 E
K is an isomorphism from Z into K, and we see that 7L is isomorphic to
Z. Therefore, the prime field is isomorphic to Q. Q.E.D.
Assume that an integral domain I is a subring of another integral domain
43
44 II. ALGEBRAIC EXTENSIONS OF FINITE DEGREES
(i) We say that an element a of I' is algebraic over I if there are elements
co (34 0), c1, ...,cn of I such that
(ii) If we can choose ci so that co = 1 , then a is said to be integral over
I.
(iii) We say that I' is algebraic, or integral, over I, if every element of
I' is algebraic, or integral, respectively, over I.
(iv) An element, or I' is said to be transcendental over I, if it is not
algebraic over I.
The condition in (ii) follows if co in (i) is chosen from the invertible
elements of I. Therefore, when I is a field, "algebraic" and "integral" are
equivalent to each other, and we mainly use the term "algebraic".
If a field K' contains another field K , we say that K' is an extension
field, or simply, an extension, of K. If K' is algebraic over K, then K' is
called an algebraic extension of K. An extension field that is not algebraic
over K is called a transcendental extension of K.
In general, if a ring R and its subring S share the same identity, then the
multiplication of elements of R by elements of S is well defined. Thus, R
is an S-module.
If an extension field K' of a field K is of finite length as a K-module,
then the length is called the degree of K' over K and is denoted usually by
[K': K], (K' IK) , or lengthK K' . In this book, we use [K': K1. In this case,
K' is called a finite algebraic extension, or an algebraic extension of finite
degree, of K. On the other hand, if K' contains a transcendental element
x over K, then K[x] is isomorphic to a polynomial ring. Therefore, the
finiteness of degree implies that K' is algebraic over K.
If a field K consists of only a finite number of elements, then K is called
a finite field. A field which is not a finite field is called an infinite field.
THEOREM 2.1.2. Let K be a finite field. Then the number of elements of
K is a power pe of its characteristic p and the exponent e coincides with
[K: n] , where it is the prime field of K K. e
Elements of K are the roots of the equation xp - x = 0.

PROOF. Since K is a finite field, the prime field it consists of p elements
(Theorem 2.1.1). K has a linearly independent base a1 , ... , ae consisting
of e elements over 7r. Each element of K is expressed uniquely in the form
>eI cia, (c1 E 7r), and the number of such expressions is pe . The set K*
of nonzero elements of K forms a group, and its order is pe -1 . Therefore,
XP -1
elements of K* are the roots of e X.- 1 . Including 0 also, we see that
elements of K are the roots of xpe - Q.E.D.
REMARK. The polynomial xp e - x is of degree pe ; hence, it has only pe
roots (counting multiplicities, too) (Theorem 1.6.8). Since K consists of
pe elements, we see that xpe - x = fJaEK(x - a), and that xpe - x has no
multiple root.
1. BASIC NOTIONS 45
One important fact about a ring R of prime characteristic p is that (a +

b)' = ap+bp for all a , b E R . To see this, consider the expansion of (a+b)p
by the binomial theorem. Then, except for the terms ap , by , the coefficient
of every term is a multiple of p (cf. additional remark in the proof of the
Theorem of Sylow). This fact can be generalized as follows.
THEOREM 2.1.3. If the characteristic of a ring R is a prime number p,
then for a1 , ... , an E R and for q = pe (with e a natural number),
(a1 + ... + an)q = aq + ... + aq.
Next, we shall prove a fundamental theorem about the degree of exten-
sions.
THEOREM 2.1.4. Assume that L is a finite algebraic extension of a field
K and M is a finite algebraic extension of a field L. Then M is a finite
algebraic extension of K and [M: K] = [M: L][L: K].
PROOF. Take linearly independent bases 11 , ... , lr of L over K and
mI , ... , M. of M over L (r = [L : K] , s = [M : L]) . It suffices to show
that all of limf form a linearly independent base of M over K. Each
m E M is expressed as m = Fs=I c1m f (cf E L). Then each cf is expressed
as cf = Ei_1 difli (di f E K) . Thus, m = E i f diflimf , and we see that M
is generated by these 1i mi as a K-module.
As for linear independence, suppose 0 = > ei fli mf f E K). Set
(e1
ff = Ei eifli . Then ff E L, >f ffmf = 0. By linear independence of
ml, ... , ms, we have that f = 0 for every j. Then, by linear indepen-
dence of II , ... , lr , we see that every eif is 0. Thus, these li mf are linearly
independent over K. Q.E.D.
COROLLARY 2.1.5. Let SZ be an extension field of a field K. If elements
aI , ... , an of fl are algebraic over K, then K(a1 , ... , an) is a finite alge-
braic extension of K.
We prove here the following fact on the degree of an extension.
THEOREM 2.1.6. If an element a of a field S2 is algebraic over a subfield K,
then P = {g(x) E K[x]lg(a) = 0} is a maximal ideal of the polynomial ring
K[x] generated by a certain monic polynomial f (x) = xn + cI xn- I+_ +c n .
The degree n of f (x) coincides with [K(a) : K]. Furthermore, K(a) = K[a]
and 1, a, ... , an-I form a linearly independent base of K(a) over K.
PROOF. Consider the mapping 0 of K[x] into K[a] defined by q5(E dix' )
_ > dia' (di E K). This is a homomorphism, and its kernel is P. Since
K[a] is an integral domain, P is a nonzero prime ideal. Since K[x] is
a Euclidean ring, P is principal and maximal (Theorem 1.6.3), and we
see the existence of f(x). If En-1 dia' = 0 (di 0 for some i) , then
E " oI dix' E P, a contradiction. Therefore, , a, ... , a"- I are linearly
1
independent. Since a" = - (c1 a"-' + . . + c,), we see that , a, ... , a"-'
. 1
form a linearly independent base of K[a] over K. Since P is a maximal

ideal, K[a] = q(K[x]) is a field, and K[a] = K(a) (cf. Lemma 3.7.5).
Q.E.D.
f (x) , as above, is uniquely determined by a and K and is called the
minimal polynomial for a (over K) .
As for general finite algebraic extensions, the following theorem holds.
THEOREM 2.1.7. If L is a finite algebraic extension of a field K, then L

is K-isomorphic to K[xI , ... , x"]/I with K[x1 , ... , x"] a polynomial ring
in a certain finite number of variables xI , ... , x" over K and its maximal
ideal I. A set of generators fI , ... , f" of I can be chosen as follows.
(i) Each f E K[x1 , ... , xj] and f is a monic polynomial in xj.
f
(ii) The ideal Ij_ I of K[x1 , ... , xj_ I ] generated by Jj , ... , j_ I is a
maximal ideal and if we denote by ak the residue class of xk in Li _
K[x1, ... , xj_I]/I _j, then f (a1 , ... , ai_I , xj) is irreducible over Lj.
PROOF. There are elements bI , ... , b" such that L = K(bI , ... , b,,) . Let
fI (xi) be the minimal polynomial for bI over K. In general, we take f =
fj(x1 , ... , xj) E K[x1, ... , xj] so that f (b1 , ... , bj_I, x) is the minimal
polynomial for bj over K(b1 , ... , bj_I) . If we regard K[x1 , ... , xj]/I as
a homomorphic image of K[a1 , ... , aj_I ][xj], then the kernel is generated
by f (a1 , ... , aj_I , xj). Therefore, we see easily that K[a1 , ... , ai_I] is
a field and that K[a1 , ... , aj] is K-isomorphic to K[b1 , ... , bj] such that
ak -* bk by induction on j. Q.E.D.
The following lemma is proved easily by using Theorem 2.1.7.
LEMMA 2.1.8. Assume that L is a finite algebraic extension of a field K K.

If a is an isomorphism from K to a field K', then there is an extension field
L' of K' such that a is extended to an isomorphism from L to L'.
REMARK. This lemma is valid if L is an extension field of K. A proof

is obtained by using a polynomial ring in infinitely many variables. Another
proof is given by using Zorn's Lemma.
PROOF. Letting L = K(b1 , ... , b"), we take K[x1, ... , x"] and I as
in the theorem above. a is naturally extended to an isomorphism from
K[x1 , ... , x"] to K'[x1 , ... , x"] (axe = xj) . Then QI is a maximal ideal
of K'[x1, ... , x"] (cf. Theorem 1.3.4) and therefore L'= K'[x1,...
lx"]/u,
is a field. We associate to each f (bI , ... , b") E L (f c K[x1, ... , x"]) ,
the residue class of (a f)(x1 , ... , x") modulo QI (c f is, by definition, the
polynomial obtained from f by replacing the coefficients with their images
under a), then we see that this mapping is an isomorphism from L to L'
and is an extension of a. Q.E.D.
2. SPLITTING FIELDS 47
2. Splitting fields
Let f (x) be a polynomial in one variable x over a field K. If f (x)
is the product of linear factors as a polynomial over an algebraic extension
field L of K , then we call L a splitting field of f (x) . Then f (x) = a(x -
cl) . (x - c") (n = deg f (x) , cl E L , and a is the coefficient of x" in
f (x)) . Therefore, a subfield M of L is a splitting field of f (x) if and
only if all roots c, are in M. Among such M, K(c1 , ... , c") is the least
subfield, and we call K(c1 , ... , c") a minimal splitting field of f (x) over
K.
THEOREM 2:2.1. For any given polynomial f (x) of one variable over afield
K, there is a minimal splitting field of f (x) .
PROOF. If f (x) has no irreducible factor of degree > 2, then K is a

minimal splitting field, and we are proving the assertion by induction on the
degree of the product f1 (x) of irreducible factors of degree > 2 of f (x) . Let
p(x) be an irreducible factor of f1(x) and consider LI = K[x]lp(x)K[x].
Since p(x)K[x] is a maximal ideal, we see that LI is a field. Let c1 be the
residue class of x in LI ., Then p(c1) = 0, and c1 is a root of f (x) If .
we consider factorization of f(x) over L1 , then we can apply our induction

hypothesis and see that there is a minimal splitting field L of f (x) over
LI = K(c1). Since c1 is a root of f (x) , we see that L is a minimal splitting
field of f (x) over K. Q.E.D.
Though we saw the existence of minimal splitting fields of a polynomial
f (x) , there may be many minimal splitting fields. The following theorem
shows that they are all substantially the same.
THEOREM 2.2.2. Under the circumstances, let SI and S2 be minimal split-

ting fields of f (x) over K. Then there is a K-isomorphism 0 from S1 to
S2. More generally, if a is an isomorphism from K to a field K' and if S'
is a minimal splitting field of a f (x) (a is applied to the coefficients of f (x))
over K', then a is extended to an isomorphism of SI to S'.
PROOF. It suffices to prove the latter half, and we use an induction on

[SI : K]. If [SI : K] = 1 , then f (x) is the product of linear factors and
S' = K' . Thus, this case is proved, and we assume that [S1 : K] > 1 . Let
p(x) be a nonlinear irreducible factor of f (x) . Let aI , b1 be roots of
p(x), ap(x), respectively. Theorem 2.1.6 shows that K(a1) is K-isomorphic
to K[x]/p(x)K[x] (a1 is mapped to the residue class of x) and K'(b1) is
K'-isomorphic to K'[x]/ap(x)K'[x] (b1 is mapped to the residue class of
x). Therefore, there is an isomorphism z from K(a1) to K'(b1), which is
an extension of or and such that za1 = b1 . Then SI and S' are minimal
splitting fields of g(x) = f(x)/(x - a1) and ag(x) = a f (x)/(x - b1) , over
K(a1) and K'(b,), respectively. Therefore, by our induction hypothesis, z

is extended to an isomorphism from S, to S'. Q.E.D.
If L = K(c1 , ... , c,z) is a finite algebraic extension of a field K , then,
taking a polynomial f(x) over K which has ci as a root for each i, we
see that L is a subfield of a minimal splitting field of f (x) = f 1i f (x) . The
theorem above shows that, even if we fix such a minimal splitting field, we
can discuss properties of L without loss of generality. Hereafter, in this
book, when we consider a finite number of algebraic extension fields, we
can take the point of view that we are taking a sufficiently large algebraic
extension field, and we are considering subfields of the field. For instance,
from this point of view, we can have a well-defined meaning of the roots
of a fixed polynomial (for instance, if we consider two algebraic extensions,
each of which contains roots, then we cannot handle these two sets of roots
simultaneously). Later, we shall prove the existence of an algebraic closure
(it is a maximal algebraic extension field of a given field), and if we admit
its existence, then we can consider the algebraic closure as a sufficiently large
algebraic extension field.
Two elements a , b of an extension field L of a field K are said to be
conjugate to each other over K, if there is an irreducible polynomial g(x)
over K such that a, b are roots of g(x), i.e., g(a) = g(b) = 0. If L
is the minimal splitting field of g(x) over K, then, applying the proof of
Theorem 2.2.2 to the case S, = S2 = L , a, = a , b, = b , we see that
there is a K-automorphism 0 of L such that qa = b. Conversely, if there
is a K-automorphism 0 of L such that Oc = d (c c L) , then for any
polynomial f (x) over K, we have 0(f (c)) = f (d) . In particular, f (c) = 0
if f (d) = 0. Therefore, c and d are conjugate to each other over K. This
fact will be stated clearly in Theorem 2.6.3.
Two subfields M, , M2 of an algebraic extension of a field K are said
to be conjugate over K if M, is K-isomorphic to M2. In this case, each
element of MI is conjugate to some element of M2. This is easily seen by
considering the minimal polynomial for the element over K.
3. Separability and inseparability

Let a be an element of an algebraic extension field L of a field K,
and let fa(x) be the minimal polynomial for a over K. We say that a
is separable over K if fa (x) has no multiple-root; otherwise, we say that
a is inseparable over K. We say that a is purely inseparable over K if
fa(x) has either one root only (i.e., deg fa(x) = 1) or one multiple root only
(i.e., fa(x) = 0 has only one solution). (An element a is separable and
at the same time purely inseparable over K if and only if a is an element
of K.) One may feel oddness about these definitions since some element
can be simultaneously separable and purely inseparable. But, there are some
advantages. For instance, the set of purely inseparable elements in L over
3. SEPARABILITY AND INSEPARABILITY 49
K forms a subfield (cf. Exercise 2.3.4). Note that an element a of L is

separable over K if there is a polynomial g(x) over K such that g(x) has
no multiple root and g(a) = 0, because such g(x) is divisible by fa(x) .
We say that L is separable over K if every element of L is separable over
K ; otherwise L is said to be inseparable over K. L is purely inseparable
over K if every element of L is purely inseparable over K. We say that K
is perfect if every algebraic extension of K is separable over K.
aox"+alxn-I+
Fora polynomial f(x) = - .+an , its derivative (d/dx)f(x)
is defined to be naox"- I + (n - 1)a1 xn-2 + - - + an_ I . We have the following
lemma.
a1x"-I
LEMMA 2.3.1. Let f (x) = aox" + + - - + an be a polynomial over
-
afield k. A root c of f (x) (c is in an extension field of K) is a multiple

root of f(x) if and only if f (c) = 0, where f4(x) denotes the derivative of
f(x)
Before proving this, we note that our derivation satisfies the usual equali-
ties: (d/dx)(f(x)+g(x)) = (d/dx)f(x)+(d/dx)g(x), (d/dx)(f(x)g(x)) =
f(x)(d/dx)g(x) + g(x)(d/dx)f(x). (The proof is easy and we omit it.)
Therefore, in particular,
r
(d/dx)(f1(x) ... fr(x)) _
=1
f (x) ... f -I (x)f +1(x) ... f,.(x)(d /dx)f (x).
,
PROOF OF THE LEMMA. The derivative does not change even if we con-
sider it over a splitting field L. Therefore, we assume that f (x) = ao(x -
c1) ... (x-cn) . Then f'(x) = ao E "= ... (x-c;-1)(x-c;+I) ... (x-cn) .
Assume first that c is a multiple root of f (x) . Then we may assume that
c1 = c2 = c. Then the derivative is divisible by x - c, and f'(c) = 0. Con-
versely, assume that c is not a multiple root of f (x) . We may assume that
c1 = c, ci c (i > 2). Then f'(c) = ao(c - c2) ... (c - cn) 0. Q.E.D.
THEOREM 2.3.2. Any field of characteristic 0 is a perfect field.
PROOF. It suffices to show that if f (x) is an irreducible polynomial over a
field K of characteristic 0, then f (x) has no multiple root. The derivative
f'(x) of f (x) is a polynomial over K. If f (x) and f'(x) have a com-
mon root c, then f (x) and f4(x) must have a nontrivial common divisor,
and f'(x) must be divisible by f (x) because f (x) is irreducible. Since
deg f'(x) < deg f (x) , f'(x) must be 0. But, from f (x) = aox :' + + an
(ao 0), it follows that f4(x) = naox"- I + 0 0. Q.E.D.
Let us examine the proof above in the case of characteristic p. The proof
goes well except for the last step, and we see that if an irreducible polynomial
f (x) over a field K of characteristic p 0 0 has a multiple root, then the
derivative of f (x) must be 0. This implies, by the definition of the deriva-
tive, that f (x) is a polynomial in x' . Conversely, if f (x) is a polynomial
in x' , then its derivative f'(x) is 0 , and every root of f (x) is a multiple
root by Lemma 2.3.1.
REMARK. If Ax) = bnx"p + + brxrp + + b0 , then taking pth root ci
of b.bn 1 for each i, we have f (x) = bn (x" + C"_1 + + c0)p by Theorem
2.1.3. Therefore, every root has a multiplicity that is a multiple of p.
By the observation above, we have the following theorem.
THEOREM 2.3.3. Let f (x) be an irreducible polynomial over a field K of

characteristic p 0. Then, f (x) has a multiple root if and only if f (x) is
a polynomial in xp . Consequently, if c is an inseparable element over K,
then for a suitable power q of p, c' is separable over K.
PROOF. The first half was proved above. As for the e
latter
XPe
half, take a
natural number e such that f (x) is a polynomial F(xp) in , but is not
e+1
a polynomial in xp . Then F(x) is irreducible and is not a polynomial in
xp . cpe is a root of F(x) , and cpe is separable over K. Q.E.D.
THEOREM 2.3.4. A field K of characteristic p 0 is a perfect field if and

only if xp - a has a root in K for every a E K (namely, every element a of
K has its pth root in K).
PROOF. Assume first that there is an element a E K such that xp - a
has no root in K. Let c be one of its roots. Then (x = xp - a - c)'
(Theorem 2.1.3). Let f (x) be the minimal polynomial for c over K. Then
f (x) is a factor of (x - c)p
xr_
, and f (x) = (x - c)' with r <p. If r <p,
then the coefficient of in f (x) is -rc, which must be an element of
K, and therefore, c E K, a contradiction. Thus r = p, and xp - a is
irreducible, which shows that K is not perfect. Conversely, assume that K
is not perfect. Then there is an irreducible monic polynomial g(x) over
K that has a multiple root. Theorem 2.3.3 shows that g(x) is of the form
xp"+bn_1xp("-I)+
+b0. If a pth root ci of bi is in K for every i, then
g(x) = (x" + + ... + c0)p , contradicting the irreducibility. Q.E.D.
cn_1xn-I
REMARK. As one can see from the first part of the proof above, each el-
ement a has a unique pth root in the characteristic p # 0 case. Conse-
quently, a has a unique peth root for each natural number e.
THEOREM 2.3.5. Every finite field is a perfect field.
PROOF. If L is a finite algebraic extension of a finite field K, then L

is also a finite field and the number q of elements of L is a power of its
characteristic, say p. Elements of L are the roots of x' - x, and this
polynomial has no multiple root (Theorem 2.1.2 and the remark to it, or
Lemma 2.3.1). Therefore, L is separable over K, and K is perfect (cf.
Exercise 2.3.2). Q.E.D.
5. SIMPLE EXTENSIONS 51
4. Multiplicative groups of finite fields

In this section, we prove the following theorem.
THEOREM 2.4.1. If K is a finite field, then the multiplicative group K* of
nonzero elements of K is a cyclic group.
For the proof, we observe that, for any natural number n, the equation
xn - 1 cannot have more than n solutions (Theorem 1.6.8). Therefore, our
assertion follows from
LEMMA 2.4.2. Let G be a finite group of order g. Set Sd = {x E Gjxd =
1} Then we have that for any divisor d of g, #(Sd) < d iff G is a cyclic
.
group. Equivalently, for any natural number d, #(Sd) < d.

PROOF. Set Td = {x E GI the order of x is d j for each divisor d of
g. If x E Td, then (x) c Sd. Since #((x)) = d and #(Sd) < d, we have
that (x) = Sd and Td c_ (x) . xe E Td if e and d are coprime. Thus,
we see that if Td is not empty, then #(Td) = 0(d), where 0 is the Euler
function, namely, for a natural number n, 0(n) is the number of coprime
numbers to n among { 1 , 2, ... , n} . Now, the order of an element of G is
a divisor of g, hence, each element of G belongs to some Td. Therefore,
g = #(G) = E #(Td) = E* 0(d), where E* means the summation only
on the d such that Td is not empty. On the other hand, letting e be an
arbitrary divisor of g, we consider numbers among 1, 2, ... , g, which
have e as G.C.M. with g. Set d = g/e. Then the set of these numbers is
{melm E 11, 2, ... , d} , m is coprime to d = g/e}. Therefore, the number
of these numbers coincides with 0(d). If e runs through all divisors of g,
then d also runs through all divisors of g. Therefore, g = Ed 0(d) (d
runs through all divisors of g). This implies that E* above must be the
summation on all divisors d. Therefore, Tg is not empty, and by an element
a of Tg , we have (a) = G. The equivalence with the other statement is easy,
and we omit the proof. Q.E.D.
5. Simple extensions
We say that a field L is a simple extension of a field K if there is an
element a of L such that L = K(a). The following theorem is basic.
THEOREM 2.5.1. Let a1 , ... , an be elements of an algebraic extension field
of afield K. If a1 , ... , an_ I are separable over K, then K(a1 , ... , an) is
a simple extension of K.
We prove this theorem together with the following very important theorem
on separable extensions.
THEOREM 2.5.2. Let L be a finite algebraic extension of a field K. Then
the following 4 conditions are equivalent to each other.
(i) L is separable over K.

(ii) L = K(al , ... , an), where the elements ai are separable over K.
(iii) There is a separable element a over K such that L = K(a).
(iv) If we take a sufficiently large algebraic extension S2 of K, then the
number of injections (i.e., K-isomorphisms) of L into Q is exactly [L: K].
REMARK. If L is not separable over K, then the number of such injec-
tions is less than [L: K], as is noted in our proof below.
PROOF. First, we consider Theorem 2.5.1. If K is a finite field, then
K(al , ... , an) is also a finite field, and its multiplicative group is generated
by one element, say a. Then K(al , ... , an) = K(a). Therefore, we assume
that K is not a finite field, and we consider the case n = 2. Let f(x)
be the minimal polynomial for a. (i = 1, 2) over K, and let the roots of
these polynomials be al =all , ... , air; a2 = a21 , ... , a2s (r = degfl , s =
deg f2) . Let c be an element of K which is different from any of (a2i -
a2i)/(alk - a1,) (k 1) and set b = a2 + cal Since a2 = b - cal , we have
.
f2(b - cal) = 0, and therefore, al is a common root of the polynomials

fl (x) and g(x) = f2(b - cx) over K(b). Since fl (x) = f1; (x - all)
and g(x) = f .1 (b - cx - a21) = (-c)S f I1=l (X - al - c ' (a2 - a2i)) , we see
that x - al is the unique common factor of fl (x) and g(x). Since the
G.C.M. of two polynomials over K(b) is a polynomial over K(b), because
it is obtained by the Euclidean algorithm. Therefore, x - al is a polynomial
over K(b), and al E K(b), which implies a2 = b - cal E K(b). Therefore,
K(al , a2) c K(b), and K(al, a2) = K(b).
Thus, we have completed the proof of Theorem 2.5.1 in the case n = 2.
The case where n > 3 is proved by induction on n , provided that Theorem
2.5.2 is proved. Therefore, hereafter, we shall prove Theorem 2.5.2.
That (i) implies (ii) is obvious. We shall show first that (iii) implies (iv).
Let f (x) be the minimal polynomial for a over K. f (x) has n roots
al , ... , an in S (n = deg f (x) = [L: K]). There is a K-isomorphism al
from K(a) into S2 which maps a to a, for each i. There is no other
K-isomorphism of K(a) into 0, because, if a is a K-isomorphism from
K(a) into Q, then as must be a root of f (x) Therefore, if a is separable
.
over K, the number of such a, is exactly n = [L: K]. Our observation can
be applied even if a is inseparable, and in that case, the number of mutually
distinct roots of f (x) is less than n , and the number of these injections is
less than [L: K1. Thus, we have proved that (iii) implies (iv) and also the
converse of this implication.
Next, we shall show that (ii) implies (iii). If we prove this in the case where
r = 2, then we can prove the general case by induction on r. Therefore, we
assume r = 2. By our proof of Theorem 2.5.1 above, there is an element a
such that K(al , a2) = K(a), and it suffices to show the separability of a. By
what we proved above, we have only to show the validity of (iv). There are K-
isomorphisms of , ... , aS (s = [K(al) : K]) from K(ai) into Q. We denote
5. SIMPLE EXTENSIONS 53
by the same ai a K-isomorphism of L into n which is an extension of the

original ai . Since a a2 is separable over K(uia1) , the number of K(oial )-
isomorphism. of a L into n is [a,L: K(aia1)] = [L: K(a1)]. Let these be
Zil,...,Pr., (t= [L:K(a1)]).Then zijai (i = 1, ...,s;j=1,...,t) are
all K-isomorphisms from L into K2, which are mutually distinct. Since the
number of these injections is st = [L: K], we see that a is separable over
K. Thus we have proved that (i) implies (ii), that (ii) implies (iii), and that
(iii) is equivalent to (iv).
We shall complete the proof by showing that (iv) implies (i). Suppose
that b E L is inseparable over K. If L = K(al , ... , a,), then L =
K(b, al , ... , a,), and we consider K-isomorphisms from L into n. Ap-
plying our observation above on K(a1 , a2) to this case, we see that (1) the
number of images of b is less than [K(b) : K] , and (2) for each i-tuple of im-
ages of b, a1 , ... , ai_1 , the number of images of ai by extended injections
is at most [K(b, a1 , ... , ai) : K(b, a1, ... , ai_ 1)]. Therefore, the number
of K-isomorphisms from L into 0 is less than [L: K], contradicting the
assumption. Thus, there is no such b. Q.E.D.
COROLLARY 2.5.3. If L is an algebraic extension of a field K, then the
subset LS of L, consisting of all separable elements over K, forms a field. L
is purely inseparable over L. L.
PROOF. If a, b E LS , then K(a , b) is separable over K, and a + b, ab,
a-I (if a 0) are in Ls. Therefore, LS is a field. If K is of characteristic
0, then L = Ls, so we assume that K is of characteristic p 0. If C E L,
then for a suitable power q of p , cq E LS by Theorem 2.3.3, and L is
purely inseparable over Ls. Q.E.D.
Under these circumstances, we call LS the separable closure of K in L.
If L is of finite degree over K, then [L: Ls] and [LS : K] are called the
inseparable factor and the separable factor of the degree of the extension,
respectively.
COROLLARY 2.5.4. Let L be a finite algebraic extension of a field K.
If S2 is a sufficiently large algebraic extension of K, then the number of K-
isomorphisms from L into S2 coincides with the separable factor of the degree
of the extension L over K.
PROOF. The number of K-isomorphisms from the separable closure Ls
into 0 is [LS: K] . If the characteristic is p # 0, then peth root of an
element is unique, and the K-isomorphisms from L into 12 are determined
by the K-isomorphisms from Ls into Q. Q.E.D.
THEOREM 2.5.5. Let L be a finite algebraic extension of a field K. Then
L is a simple extension of K if and only if the number of intermediate fields
between L and K is finite.
PROOF. Assume first that L = K(a). Let f (x) = In I (x - ai) be the
minimal polynomial for a over K (with a, in a splitting field of f (x)

containing L). If M is an intermediate field between L and K, then
L = M(a). Let fM(x) be the minimal polynomial for a over M. Since
f (a) = 0, fM (x) is a factor of f (x) . If fM (x) = xr + cI xr- I +. + c,. , then
M D K(c1 , ... , Cr) and [K(a) : K(c1 , ... , c,.)] = deg ff(x) = [K(a) : M].
Therefore, M = K(c1 , ... , cr) This shows that M defines the factor fM(x)
.
of f (x) and the correspondence M fM(x) is a one-to-one correspon-

dence. Therefore, the number of M is finite because there are only a finite
number of monic factors of f (x) .
Conversely, suppose that L is not a simple extension of K. Let LS be
the separable closure of K in L. Then L is not a simple extension of
LS by Theorem 2.5.1. Take a generator aI , ... , a, of L over LS such
that r is as small as possible. Then r > 2. Let L0 = LS (aP , ... , ap) ,
where p is the characteristic of L. Then [L: L0] = pr . (Indeed, let L, be
Ls(aI , ... , a; , ap I , ... , a') for each i, and suppose L. = L,+1 for some i.
z z
Then a;+ I E L; , and therefore, ap I E Ls (a P, , ... , ai , a + , ... , ap) , which
I
z
implies that a;+I E Ls (a1 , ... , a; , a p I , a +2 , , a ) . We see similarly that
a;+ I E Ls (a I , ... , a; , ap+"j, a p2 , ... , a p) for any natural number n. But,
there is an n such that ap I E Ls , which contradicts the minimality of r.
Thus, L;+, = L;(a;+I), and L; # L;+1 . We see that [L: Ls] = pr .) Since
Ls is not a perfect field, LS is not a finite field. For each c c Ls , set y(c) =
aI + cat . Then [Lo(y(c)) : Lo] = p. If c 54 c', then Lo(y(c)) 54 Lo(y(c')).
(Indeed, if Lo(y(c)) = Lo(y(c')), then y(c) and y(c') are in Lo(y(c)) and
(c - c')a2 = y(c) - y(c') E Lo(y(c)). Then a2 E Lo(y(c)), which implies
aI E Lo(y(c)) and [Lo(y(c)) : Lo] > p2 , a contradiction.) Since LS is an
infinite field, there are infinitely many intermediate fields. Q.E.D.
6. Normal extensions
We say that an algebraic extension field L of a field K is a normal ex-
tension of K if, for any a E L, is fa(x) is the minimal polynomial for a
over K then L is a splitting field of fa(x) , namely, all conjugates of every
element of L over K (considered in a sufficiently large field containing L)
are in L. The theorem below shows that this condition is satisfied if the
same condition is satisfied by the elements of a set of generators of L over
K
We add here the definition that a normal extension L of K is called a
Galois extension if L is separable over K.
THEOREM 2.6.1. Let L = K(a1, ... , a,,) be an algebraic extension of a
field K. Then L is a normal extension of K if and only if all conjugates of
every a; (i = 1, ... , n) are in L.
7. INVARIANTS OF A FINITE GROUP 55
PROOF. The only if part is obvious and we shall show the if part. Let f(x)
be the minimal polynomial for ai over K and set f (x) = Hi fi(x). Then
L is the minimal splitting field of f (x) . Let b be an arbitrary element of
L, and consider an arbitrary conjugate b' of b in a sufficiently large field
Q containing L. Then there is a K-isomorphism a from K(b) into 92
such that ab = b'. a can be extended to a K-isomorphism from L into
92. Then aL is also a minimal splitting field of 1(x), and aL = L. Thus
b' E L. Q.E.D.
If a field L contains a field K, the set AutK L of K-automorphisms of
L forms a group (with multiplication defined by (ar)a = a(ra)). If L is a
normal extension of K, then we call this group the Galois group of L over
K and denote it by G(L/K). In this case, our proof of the theorem above
shows that for any K-isomorphism from L into a field containing L, L
is mapped to L itself. If, furthermore, L is of finite degree over K, then
Corollary 2.5.4 shows that #(G(L/K)) coincides with the separable factor
of the degree of the extension. Therefore we have
THEOREM 2.6.2. If L is a normal extension of finite degree over a field
K, then the order of the Galois group G(L/K) coincides with the separable
factor of the degree of the extension. Let LS be the separable closure of K
in L. Then LS is a Galois extension of K and G(L/K) ^' G(Ls/K) by
a natural isomorphism such that a - al L s (the restriction of a of Ls) for
a E G(L/K).
From our proof of Theorem 2.6.1, we also have
THEOREM 2.6.3. Let L be a normal extension of finite degree over a field
K. Then the set of conjugates of an element a of L over K coincides with
{aala E G(L/K)}, and the set of conjugates over K of an intermediate field
M between K and L coincides with {aMla E G(L/K)}.
7. Invariants of a finite group

As was noted above, if a field L contains a field K, then the group of
K-automorphisms is denoted by AutK L. When K is the prime field, the
group coincides with the group of automorphisms of L and is denoted by
Aut L. If there is a homomorphism 0 from a group G to AutK L, then we
define an action of G on L by as = (qa)a for a c G, a E L. Then we
say that G induces K-automorphisms of L, or G acts on L/K (or, simply,
on L) . Note that the action is really determined by the subgroup qG of
AutK L. Note also that a similar definition can be given when L is a ring or
a group.
If as = a for every a c G, then we say that a is an invariant of G or a
G-invariant. As is easily seen, the set of G-invariants forms a subfield of L
containing K. This subfield is called the field of G-invariants.
THEOREM 2.7.1. Let L be afield, and let G be a finite subgroup of Aut L L.

Let K be the field of G-invariants. Then L is a Galois extension of K,
G = G(L/K), and [L: K] = #(G).
PROOF. For each a E L, we consider H(a) _ {a E Gl as = a). H(a)
is a subgroup of G, and ca = za if aH(a) = -rH(a). Therefore, if we
take a system of representatives al , ... , ar (r = #(G/H(a)) of G/H(a),
then {a1 = a1 a, ... , ar = ara} coincides with {aal a E G} . Consider the
polynomial fQ (x) = Hr_ 1(x - a.). The coefficient ci of xr-' in fQ (x) is
(-1)1 E ae, , and therefore, cci = ci for every a in G.
I <e1 <... <e. <r ae1 ae2
Thus, fa(x) is a polynomial over K and has a as a root. We see that every
element a of L is separable over K, and the minimal polynomial for a
over K is of degree at most n = #(G).
Therefore, L is separable over K , and every finite extension of K con-
tained in L is a simple extension of degree at most n (Theorem 2.5.1.)
Therefore, L must be finite over K , L = K(b) with an element b and
[L: K] < n. Since G consists only of K-automorphisms of L, we have
#(G) < [L: K] (Theorem 2.5.2, (iv)), and we see that #(G) = [L: K]. If
our observation about fa(x) is applied to the minimal polynomial for b,
then we see that conjugates of b over K are all in L. Therefore, L is nor-
mal over K (Theorem 2.6.1); hence, L is Galois over K and G = G(L/K) .
Q.E.D.
COROLLARY 2.7.2. Let L be a normal extension of a field K of finite
degree. Consider the separable closure LS of K in L and also the set Li of
elements of L which are purely inseparable over K. Then
(i) L. is a field and L is generated by Li and LS L.
(ii) L is a Galois extension of Li and G(L/K) = G(L/Li) .
(iii) [L: L.] _ [LS : K], [L: Ls] = [Li : K1.

PROOF. Consider the field K' of invariants of the group AutK L =
G(L/K). Suppose that an element a of K' is not purely inseparable over
K. Then there is a conjugate a' of a over K such that a' # a. Then there
is an element a E G(L/K) such that as = a' (cf. the proof of Theorem
2.6.1), contradicting the assumption that a is G(L/K)-invariant. Therefore
K' CLi . Obviously, elements of Li are G(L/K)-invariants, and we see
that K' = Li . This shows that Li is a field and (ii) is valid. Consequently,
[L: Li] _ #(G(L/K)) = [LS : K] , which implies [L: Ls] = [L: K]/[Ls: K] =
[L: K]/[L: Li] = [Li : K]. Thus, (iii) is proved. Consider the field Lo gener-
ated by Li and LS Since L is purely inseparable over LS , L must be purely
.
inseparable over Lo . Since L is separable over Li (because Li = K') , L

must be separable over Lo . Therefore, L = Lo. Q.E.D.
THEOREM 2.7.3. Assume that K, L, K' are subfields of a field and that
K C LnK'. Then
(i) If L is a Galois extension of K of finite degree, then K'(L) is a Ga-
7. INVARIANTS OF A FINITE GROUP 57
lois extension of K' and G(K'(L)/K') is isomorphic to G(L/K' n L), and

therefore [K'(L) : K] = [L: (K' n L)].
(ii) If L is separable and of finite degree over K and if K' is normal
over K, then K'(L) is a separable extension of K' of finite degree and
[K'(L) : K'] = [L: (K' n L)].
(iii) If L is purely inseparable and of finite degree over K and if K' is
separable over K, then K'(L) is purely inseparable and of finite degree over
K' and [K'(L) : K'] = [L: (K' n L)].
PROOF. (i) Let f (x) be the minimal polynomial for a (E L) over K' n L
with a such that L = K(a). Let aI , ... , ar (r = deg f (x)) be the roots
of f (x) . Since L is Galois over K, conjugates of a over K are in L.
Therefore, these ai are in K'(L). Since K'(a) = K'(L), we see that K'(L)
is a Galois extension over K'. Suppose that f (x) is reducible over K' By .
a suitable renumbering of al , we may assume that rjl=I (x - a,) is a factor.

The coefficients of this factor have polynomial expressions in aI , ... , as ,
and they are in L. Since the factorization is over K' , these coefficients
must be in K'; hence, in K' n L. Therefore f (x) is irreducible over K'.
This implies that [K'(L): K'] _ [L: (K' nL)] I. By considering the restriction
of the action of G(K'(L)/K') to L, we obtain a subgroup G' of Aut L .
The natural mapping of G(K'(L)/K') to G' (by the restriction above) is a

homomorphism, and the field of G'-invariants is K'nL L. Therefore, #(G') =
[L: (K' n L)] . Thus, #(G') _ [K'(L) : K'] = #(G(K'(L)/K'), and we see that
G' is isomorphic to G(K'(L)/K') and that G' = G(L/L n K').
(ii) We take a, f (x) , a, , ... , ar as above and assume also that R= I (x -
a1) (with a = aI) is a factor of f (x) over K'. Consider the coefficients
cI , ... , cs of this factor. They are separable over K because they have
polynomial expressions in aI , ... , as. Therefore, there is a Galois extension
K" of K' n L of finite degree, contained in K' and containing cI , ... , cs .
By (i), [L(K") : L] = [K": K" n L]. s > [L(K") : K"] = [L(K") : K' n
L]/[K" : K' n L] = [L: K' n L] = r. Thus, f (x) is irreducible over K'.
Since a is separable over K, we see that a is separable over K' n L, and
(ii) is proved.
(iii) If the characteristic of K is 0, then L = K and the assertion is
obvious. Therefore, we assume that the characteristic is p 0. Take a
sequence of intermediate fields Li such that K = Lo c LI c c Lr = L
and [L,+I : L.] = p for each i = 0, ... , r - 1. Consider L' = K'(L1) . If
we show [L'+I : L'] = [L,+I : L,] for each i, then we complete the proof.
Since L' is separable over L, , it suffices to prove the assertion in the case
[L: K] = p. In this case, if a E L, a V K, then L = K(a) and a E
K. Then [K'(a) : K'] is either p or If it were 1, then a c K', which
1 .
implies that a is separable and purely inseparable over K. Then a E K, a

contradiction. Thus [K'(a): K'] = p = [L: K]. Q.E.D.
8. The fundamental theorem of Galois

THEOREM 2.8.1. (The fundamental theorem of Galois). Let L be a Galois
extension of finite degree of a field K. Then L is a Galois extension of every
intermediate field M between K and L. Furthermore
(i) There is a one-to-one correspondence between the set of intermediate
fields M and the set of subgroups H of the Galois group G(L/K) such
that M corresponds to H iff M is the field of H-invariants. Equivalently,
H = {a E GIam = m for all m E M} , which is the same as H = G(L/M).
(ii) If we denote by H(M) the subgroup corresponding to an intermediate
field M, then for two intermediate fields MI , M2, we have
(1) MI c M2 iff H(MI) D H(M2),
(2) H(MI nM2) = H(MI) V H(M2),
(3) H(MI V M2) = H(MI) n H(M2),
(4) if a c G(L/K), then H(aM) = QH(M)Q-I.
(iii) An intermediate field M is a Galois extension of K iff H(M) is a

normal subgroup of G(L/M).
In this case, G(M/K) consists of the restrictions of elements of G(L/K)
to M. and G(M/K) ^_' G(L/K)/H(M).
PROOF. In this proof, G denotes the Galois group G(L/K). The first
assertion is obvious because any conjugates of an element a of L over M
are conjugates of a over K.
(i) To each intermediate field M, we associate H(M) = G(L/M). Then
H(M) = {a E Glom = m for all m C M)} c G, #(H(M)) = [L: M],
and the field of H(M)-invariants is M (Theorem 2.7.1). Conversely, if we
associate, to each subgroup H of G, the field M' of H-invariants, then
H = G(L/M') (Theorem 2.7.1). Therefore (i) is proved.
(ii) (1) is obvious. Since MI n M2 c Mi, we have H(MI n M2) H(M1)
and H(MI n M2) D H(MI) V H(M2). Let M' be the field of invariants of
H(MI) V H(M2). Then M' CMi, and therefore, M' c MI n M2, which
implies that H(M') D H(MI n M2) H(MI) V H(M2) = H(M'). Hence,
M' = Ml n M2 , which proves (2). (3) is proved similarly. To prove (4) note
that r c H(aM) iff rua = as for all a C M. Equivalently, a- I raa = a
for all a C M. That is Q-I zQ E H(M), which is the same as the condition
I
z E aH(M)a- This completes the proof of (ii).
.
(iii) M is Galois over K if aM = M for all a E G (Theorem 2.6.3).

Equivalently, H(QM) = H(M) for all a C G). That is, QH(M)a ' =
H(M) for all a c G, which is equivalent to the condition that H(M) is
a normal subgroup of G. In this case, since aM = M, G acts on M.
Since elements of M are H(M)-invariants, the action of each element a on
M is decided by the coset aH(M) (= H(M)a) E G/H(M). If aH(M) #
rH(M), then z- I a is not an element of H(M) = G(L/M), namely, there
is an element a E M such that r - I as a. Therefore, the restriction
9. ROOTS OF UNITY AND CYCLIC EXTENSIONS 59
of G to M induces the inclusion G/H(M) c G(M/K). #(G/H(M)) =

#(G)/ #(H(M)) = [L: K]/[L: M] = [M: K] = #(G(M/K)). Hence, we
have G/H(M) = G(M/K). Q.E.D.
9. Roots of unity and cyclic extensions
We consider the following theorem in a splitting field L of x" - 1 over a
field K of characteristic p.
THEOREM 2.9.1. The set of nth roots of unity (namely, the set Z of roots
-
of x" 1) in L forms a cyclic group (with multiplication). As for the order
of the group Z we have
(i) If p = 0, then #(Z) = n,
(ii) If p 0 0 and n is not a multiple of p, then #(Z) = n,
(iii) If n = prm and m is not a multiple of p (# 0), then #(Z) = m. In
this case, Z is the set of mth roots of unity.
In the cases (i) and (ii), generators of Z are called primitive nth roots of
unity.
PROOF. In the cases (i) and (ii), x" - 1 has no multiple root, because
xn_
I has no common root with its derivative nx"- I . Therefore, #(Z) = n.
Obviously, Z forms a group. For any natural number q, the equation
xq = 1 does not have more than q solutions in Z, and we see that Z is a
cyclic group by Lemma 2.4.2.
As for the case (iii), since xpr - 1 = (x - 1)p, the equation xpr = 1 has
the unique solution x = 1 . Therefore, x" = 1 if x'" = 1 . Thus, we see the
assertion by the case (ii). Q.E.D.
Let L be a Galois extension of finite degree of a field K. (i) If the Galois
group G(L/K) is commutative, then L is called an Abelian extension of K
and (ii) if G(L/K) is a cyclic group, then L is called a cyclic extension of
K
The minimal splitting field of x" -1 over a field K is called the nth cyclo-
tomic field over K. By a cyclotomic field over K, we mean an intermediate
field between K and the nth cyclotomic field (with suitable n).
REMARK. The term "cyclotomic field" is usually used only for the case
where K is the rational number field. But, we use the term in a generalized
way.
THEOREM 2.9.2. Every cyclotomic field over afield K is an Abelian exten-
sion of K.
PROOF. Since every subextension of an Abelian extension is Abelian, it
suffices to prove that the nth cyclotomic field L over K is an Abelian
extension of K. By Theorem 2.9.1, we may assume that x" - 1 has no
multiple root. Then L is separable over K. Let C be a primitive nth root
of unity. Then L = K(g). Each element a of G(L/K) maps C to Ca with
certain natural number a, and a is uniquely determined by CC = (a. If
zC _ Cb , then a r = a,Cb = Cab

= zCa = TUC. Therefore ar = T9, which
shows that G(L/K) is abelian. Q.E.D.
If K is a finite field, then we have the following theorem.
THEOREM 2.9.3. If L is a finite algebraic extension of a finite field K, then
L is a cyclic extension of K.
PROOF. If q = #(L), then L is the minimal splitting field of xq - X. If
h is the order of a subgroup H of G(L/K), then the number of elements
of the field M of H-invariants is determined by h, namely, q = #(M) h ,
and M is the minimal splitting field of x#(M) - x. This shows that G(L/K)
has only one subgroup of order h. Therefore, G(L/K) is a cyclic group (see
Exercise 1.2.12). Q.E.D.
COROLLARY 2.9.4. In Theorem 2.9.2, if K is of characteristic p 54 0, then
L is a cyclic extension of K.
PROOF. Assume that L is a subfield of the nth cyclotomic extension of
K. Let L0 be the nth cyclotomic extension of the prime field, and set
Ko = K n Lo . Then L C K(K0) , Lo is a cyclic extension of Ko , and we see
the assertion by Theorem 2.7.3. Q.E.D.
Let us consider the nth cyclotomic field P of the rational number field
Q
THEOREM 2.9.5. The Galois group G(Pn/Q) is isomorphic to the group Un
of invertible elements in Z/nZ (where Z is the ring of rational integers).
PROOF. Our proof of Theorem 2.9.2 shows that G = G(Pn /Q) is iso-
morphic to a subgroup of Un. (Namely, consider the mapping a (a
mod n) with symbols in the proof. a-I' = Cc implies that ac - 1( mod n) ,
therefore, G is isomorphic to a subgroup of Un Note that this part can
.
be applied to the general case.) Therefore, we complete the proof by the

following theorem.
THEOREM 2.9.6. The following polynomial F (x) is irreducible over the
rational number field Q, where C is a nth primitive root of unity.
F (x) = U (x - a).
0<a<n, nZ+aE U,,
This F (x) is called the nth cyclotomic polynomial. The degree of this
polynomial is 0(n), with 0 the Euler's function.
For the proof of this theorem, we note the following lemma.
LEMMA 2.9.7. If n is the prime field of characteristic p 54 0, then f (x)p =
f (xp) for any f (X) E 7r[X]
PROOF. Let f (x) _ aix` . Then f (x)p = > apx`p . Since ai are in 7r,
we have ap = ai and f (x)p = f (xp). Q.E.D.
9. ROOTS OF UNITY AND CYCLIC EXTENSIONS 61
PROOF OF THEOREM 2.9.6. Suppose that F (x) is reducible over Q. Since

Z is a unique factorization domain, we see by Lemma 1.6.6 that F" (x) is
reducible as an element of Z[x]. Let a factorization of F (x) be h(x)k(x)
with h(x) an irreducible factor. Since F (x) has no multiple root, h(x)
and k(x) have no common root. Let G be the set of roots of h(x). Choose
an element C E G. Then a root of k(x) is written as C` with c coprime
to n . Consider the factorization of c : c = p . ps with prime numbers p1 .
I . .
Set c, = p1 . pi (i = 1 , 2, ... , s) , co = 1 . Then there is a j such that

Cì is a root of h(x) and (`J+I is a root of k(x) . By taking Cci instead of
C, we may assume that C is a root of h(x) and (p is a root of k(x) with
a prime number p that is not a factor of n. Then C is a root of k (xp) .
Thus, h(x) and k(xp) have a common root. Since h(x) is irreducible, there
is a polynomial g(x) E Z[x] such that k(xp) = h(x)g(x). We observe this
relation modulo p, and we obtain a relation of polynomials over the prime
field n = Z/pZ, i.e., (k(x))p = k(xp) = h(x)g(x) , where the residue classes
modulo p are denoted by adding (Lemma 2.9.7). This shows that h(x)
and k(x) have a common root, and Fn(x) has a multiple root, because
Fn(x) = h(x)k(x). But, since Fn(x) is a factor of x" - 1, F" (x) is a
factor of x" - 1 over 7r. Since p is not a factor of n , x" - 1 over n has
no multiple root, a contradiction. Therefore, F (x) is irreducible over Q.
Q.E.D.
THEOREM 2.9.8. Let L be a cyclic extension of degree n over a field K
such that either the characteristic p of K is 0 or n is coprime to p. If K
contains all nth roots of unity (namely, K is a splitting field of x" - 1), then
there is an element c of L such that L = K(c) and the minimal polynonmial
for c over K is x" -a (a E K) .
NOTATION. (1) When x" - a is an irreducible polynomial over a field K,

then one of its roots is denoted by " a Then other roots are of the form
.
C r a where C is an nth root of unity. (2) In the proof below, for a E Aut L
and b c L, we use b-type notation, and we mean by b 1 +'+'2+' the
bbaba2 ba,
product . . . .
PROOF. Let or be a generator of G(L/K). For two elements x, t of L,

we consider the following element which is called the Lagrange resolvent:
+a"-2 an-I
a xI+ata2 x1+a+azta3 x1+a+
u(x , t) = t + xt + + + + t
Then, as is easily seen, if x satisfies x = 1 , then we have

x(u(x, t))a = u(x, t). If we take a primitive nth root C of unity as
x, then x x and x = x = 1. Therefore, if there is a t c L
such that u((, t) 0, then setting c = u((, t) , we have c E L, ca = C-1 C)
cap "- I
= C-`c Therefore, the conjugates of c are c, Cc, ... ,
. c , and their
product is +c" E K. Thus, it suffices to show the existence of t such that
u(C, t) 54 0, which follows from the following lemma.
LEMMA 2.9.9. Let a1 , ... , an be mutually distinct isomorphisms from a

field L to a field M. If ci , ... , c,, are elements of M that are not all 0,
then there is an element u of L such that >i ciaiu # 0.
NOTATION. In this proof, >i ci ai denotes the mapping of L into M
given by u -> >i ciciu E M (where it is understood that different linear
forms may express the same mapping).
PROOF. Assume that there are some >i ci ai which map every element of
L to 0. Among such >i ciai, we choose one which has the least number of
terms, say, c, al + + crar (by changing the numbering of ai , we assume
that all the ci are not 0) . Obviously, r 1 and r > 2. Since aI 0 a2 ,
there is an element a E L such that or, a a2a. Since >i ci ai (au) = 0 for
all u, Eci(Qia)aiu = 0, and 0 = ala(Eciaiu) - >ci(aia)aiu = c2(a,a -
a2a)a2u + + cr(a1 a - ara)aru for all u, contradicting the leastness of r,
because c2(a1a - a2a) # 0. Q.E.D.
Next we consider cyclic extensions of degree p of a field K of characteris-
tic p 0, since this case is basic among cyclic extensions of degree divisible
by p.
THEOREM 2.9.10. If L is a cyclic extension of degree p of a field K of
characteristic p 0 0, then L is the minimal splitting field of xp - x - c for
a suitable c E K. Conversely, consider ,(x) = xp - x - c for an arbitrary
c E K. Let a be a roof of fc(x). Then the roots of fc(x) are a, a +
I, ... , a+ (p - 1). If a is not in K, then the minimal splitting f eld of fc(x)
over K is K(a), which is a cyclic extension of degree p over K.
PROOF. First we prove the converse part. If a is a root of ,(x) and if b
is an element of the prime field n , then f, (a + b) = (a + b)p - (a + b) - c =
ap-a-c+bp-b = 0. Thus, a, a+1 , ... , a+(p-1) are roots of fc(x). Since
deg fc is p, we see that these roots exhaust the roots of fc(x). Assume that
a is not in K. Let L be the minimal splitting field of fc(x) over K. Then
there is a a E G(L/K) such that as 0 a. Then as = a + b (0 0 b E n) .
Then a'a = a + ib; therefore, we obtain all of a, a + 1 , ... , a + (p - 1) by

a'a (i = 0, 1, ... , p - 1). This shows that fc(x) is irreducible over K and
L is of degree p over K. Therefore, L is a cyclic extension of degree p.
Next we prove the first half. Applying Lemma 2.9.9 with L = M, ci = 1
for all i , we see that there is an e E L such that >i a'e 0 0, where a'
runs through all elements of G(L/K) = (a) (cf. Exercise 2.7.3 (5)). Set
d Then d-ad
(p - 1)e = ae + ate + + ap-' e + e, which is a nonzero element of K,
say f, by our choice of e. Then d - ad = f 0 implies that d is not
in K and L = K(d) (because any cyclic group of order p has no proper
subgroup). ad - a2d = f , and in general, a'd - a'+'d = f . Thus the
conjugates of d are d, d - f, d - 2f, ... , d - (p - 1)f. Set 6 = d / f .
Then L = K(8) and the conjugates of 6 are 6, 8 - 1, ... , 8 - (p - 1).
10. SOLVABILITY OF ALGEBRAIC EQUATIONS 63
c = o(S - 1) - (p - 1)) = HCEG Q8 E K. The elements of the prime

(S
field n are the roots of x'' - x ; therefore, rjaE71 (x - a) = xp - x . Therefore,
8(8 - 1) ... (6 - (p - 1)) = 8p - 6 , which shows that o is a root of fc(x) _
x" - x - c; therefore, L is the minimal splitting field of fc(x). Q.E.D.
10. Solvability of algebraic equations
Let L be an algebraic extension of a field K. We say that L is an
extension by radicals if there is a sequence K = Ko C KI C C K, = L of
fields KI (i = 1, ... , r) such that each Ki is of the form K._ ( i f ai) (with I
ai c Ki_ I such that xni

- ai is Then the elements
irreducible over Ki_ 1) .
of L are said to be expressed by radicals over K. We say that an algebraic

equation
a1x"-I
x" + + ... + an =0 (ai E K)
is algebraically solvable if the roots of this equation are expressed by radicals
over n(a1 , ... , an), where n is the prime field of K.
It follows easily that
LEMMA 2.10.1. If L is an extension by radicals of a field K and if M is
an extension by radicals of L, then M is an extension by radicals of K.
THEOREM 2.10.2. If K is afield of characteristic 0, then roots of unity are
expressed by radicals over K. When a natural number n is given, considering
a splitting field of fI,n<n(X'n - 1), we set Fn = {CIC'" = 1 for a natural
number m < n}. Then the field K" generated by Fn over K is an extension
by radicals of K.
PROOF. K = K1 = K2 C K3 C It suffices to show that each Kn
.
(n > 3) is an extension by radicals of Kn_ . Kn is an Abelian extension

I
of Kn_1 (Theorem 2.9.2), and the degree of the extension is less than n,
because x" - 1 is reducible. In the Galois group G = G(Kn/Kn_1), we take
a sequence G = Gr D Gr_I D D G1 D Go = {1} such that each Gi/Gi_I is
a cyclic group (take a cyclic subgroup G1 first, then considering G/GI , take
G2 so that G2/G1 is cyclic, and so on). Let Mi be the intermediate field
corresponding to Gi . Then Mo = Kn, 111, = Kn_ , and each Mi (i < r) is
1
a cyclic extension of Mi+1 of degree mi less than n. Since Mi+1 contains

all mith roots of unity, Mi is of the form Mi+1( bi) (Theorem 2.9.8).
Therefore, Kn is an extension by radicals of Kn_ . Q.E.D. I
LEMMA 2.10.3. Assume that x" - a is an irreducible polynomial over a

field K (a E K). Let C be a primitive nth root of unity, and let K' be a
field containing K(g). Then the minimal polynomial for n a over K' is of
the form xs - b with b E K' and a divisor s of n .
PROOF. Conjugates of Ca over K' are of the form C'. Ca. Let these be
`' Ca, ... , C'S ra. Their product is equal to be (r a)' with c = > ii .
Since C E K', we have b = ("/a`)" E K'. Therefore, </i is a root of xS - b.

Since s is the number of conjugates, xs - b is the minimal polynomial for
" a . The conjugates of each (j r a are ("+' r a (t = 1 , ... , s) , s must
be a divisor of n. Q.E.D.
THEOREM 2.10.4. If L is a Galois extension of finite degree over a field K
of characteristic 0, then L is a subfield of an extension by radicals of K iff
G(L/K) is a solvable group.
PROOF. To prove the if part, consider the sequence of commutator sub-

groups that begins with G = G(L/K) : G = G 0 D GI = [G, G] D ... D Gi+1 =
[Gi , Gi ] D ... D Gr = [Gr_ 1 , Gr_ 11 _ {1 } . Let Li be the intermediate field
corresponding to Gi (Lo = K, Lr = L). We shall prove the assertion by
induction on the length r of the sequence. L1 is an Abelian extension of K.
Let Kn be as in Theorem 2.10.2 with n = #(G/[G, G]) = [LI : K]. Then
Kn (LI) is an extension by radicals of K,1 by Theorem 2.9.8, and therefore,
Kn (L1) is an extension by radicals by Lemma 2.10.1 and Theorem 2.10.2.
By our induction hypothesis, L is a subfield of an extension by radicals of
the field LI Therefore, we complete the proof by Lemma 2.10.1. (It is not
.
hard to show, by a proof similar to the one above that if m = #(G), then
Km(L) is an extension by radicals of the field K.)
To prove the only if part, assume that L C_ K( VaI , ... , a,.) = L,.,
where ai+1 E Li = K(' a1 , ... , ai) and xni+' - ai+1 is irreducible over
Li . Let L* be the smallest Galois extension of K containing L,.. Then it
suffices to show that G(L*/K) is solvable, since G(L/K) ^_' G(L*/K)/G(L*/L)
and any homomorphic image of a solvable group is solvable. We use an
induction argument on r. If r = 1 , then the conjugates of Ca (n =
n1 , a = a1) are " a, ... , b=n-1 " a , " a , where C is a primitive nth
root of unity. Therefore, L* K(C, " a) K(C) is an Abelian extension
.
of K, and therefore, G(K(C)/K) is a commutative group. The minimal

polynomial for r a over K(C) is of the form xS - b (Lemma 2.10.3). Since
is a cyclic group, we see that G(L*/K) is solvable. Now, we
consider the general case. Let M be the smallest Galois extension of K
containing Lr_ Then M C L*. By our induction hypothesis, G(M/K)
1 .
is solvable. With n = nr and a = ar , we see that L* is generated by the

roots of xn - ca (a runs through elements of G(M/K)) Thus C E L*.
Since K(0) is an Abelian extension of K, G(M(C)/M) is commutative.
Therefore, G(M(C)/K) is solvable. Since L* is generated by the roots of
x "-ca over M(C) and G(M((, 1ua)/M(C)) is cyclic (Lemma 2.10.3), we
see that G(L*/K) is solvable (cf. Exercise 2.10.1). Q.E.D.
COROLLARY 2.10.5. Consider an equation f (x) = x" +c1 xi-1 + +c = 0

with coefficients ci in a field containing the rational number field Q. Let L
be the minimal splitting field of the polynomial f(x) over Q(c1 , ... , cn) .
10. SOLVABILITY OF ALGEBRAIC EQUATIONS 65
Then this equation is solved algebrically if and only if G(L/Q(cl , ... , is

a solvable group.
G(L/Q(c1 , ... , cn)) is called the Galois group of the polynomial f (x) or
of the equation f (x) = 0. The similar definition applies to the general case
where the prime field replaces Q.
PROOF. Sufficiency is obvious by Theorem 2.10.4. The converse follows
from Lemma 2.10.3 (cf. Exercise 2.10.2). Q.E.D.
Let us compare these results obtained above with the known methods for
solving algebraic equations with coefficients in a field of characteristic 0.
Consider a polynomial f(x) = aoxn+alxn-I+ +an over a field K(ao
0) . The discriminant of f (x) (or of the equation f (x) = 0) is defined to be
the square fl <; (a1 - aj)2 of the difference product of the roots a , ... , an . 1
The discriminant is a symmetric form of the roots aI , ... , an (cf. p. 34-35).

Since any symmetric form is a polynomial in the elementary symmetric forms
and since (-1)àiao 1 are the elementary symmetric forms of a1 , ... , an
we have the following lemma.
LEMMA 2.10.6. Let D be the discriminant of a polynomial f (x) over a
field K, and let L be the minimal splitting field of f (x) over K. Then
D E K and x2 - D is factored into the product of two linear factors over L.
(A) Quadratic equations. For f (x) = x2 + ax + b, we have a formula

(-a f a--4b) /2 for the roots. a2 - 4b coincides with the discriminant of
-
f (x) . The field generated by a2 4b over Q(a, b) = K is the minimal
splitting field of f (x) over K and the degree of the extension is either 1 or
2 (depending on whether a --4b E K or not). These correspond to the
facts that any quadratic extension is a normal extension (cf. Exercise 2.8.1)
and that any quadratic extension of a field K, of characteristic 2, is of
the form K(im) with a c K.
(B) Cubic equations. Let a, y be the roots of a cubic equation
f (X) = x3 + axe + bx + c = 0. Let L be the minimal splitting field of
f (x) over K = Q(a, b, c), namely, L = K(a, fl, y). If a is an element of
G(L/K), then (ca, up, ay) is a permutation of (a, fl, y) and the action
of a is determined by this permutation. Denote by n(a) the permutation
(c' !l y) . Then
oa vQ ay
a /3 y a /3 y
7d(9)7r(T) =
ua up ay Ta T/3 Ty
Ta T/3 Ty a /3 y
QTa UT/l a ry Ta T/3 Ty
a /3 y
QTa QT/3 aTy
7C(ar).
Thus, n is an isomorphism from G(L/K) into the symmetric group S3 of

degree 3. Similarly, we have
LEMMA 2.10.7. Let al, ... , an be the roots of an equation f (x) = x" +
c1 xn-1 + + (,,n = 0 over an arbitrary field K. Then the Galois group
of the minimal splitting field L = K(a1 , ... , an) over K can be regarded
as a permutation group on {a1 , ... , an} . Thus, G(L/K) is regarded as a
subgroup of the symmetric group Sn of degree n.
We go back to the characteristic 0 case. The symmetric group S3 is of

order 6. Two cyclic permutations (a, fl, y), (a, y, /3) form a subgroup
N of order 3 together with the identity. Its index is 2, and therefore N
is a normal subgroup. It follows that both S3/N and N are commutative
groups, which implies that S3 is a solvable group. Therefore its subgroup
G(L/K) is solvable.
The order of G(L/K) is a divisor of 6, and so, is one of 1, 2, 3, 6.
If the order is 1 or 2, then f (x) is reducible over K.
Assume next that the order is 6. The intermediate field M corresponding
to the subgroup N is a Galois extension of K of degree 2. The difference
product 0 = (a - f)(a - y)(/3 - y) is mapped to -0 by the transposition
or = (1 , 2) and therefore 0 V K. Since 0 is invariant by elements of N,
we have 0 E M. Thus, M = K(0) . N is a cyclic group of order 3; hence,
L is a cyclic extension of M of degree 3. Therefore, the roots of f (x)
are obtained by using a cubic root co of unity and finding suitable elements
of the form ' d . Namely, we first obtain 0 by taking a square root of
the discriminant and then we obtain suitable cubic roots of some elements
of K(0, w) If the order of G(L/K) is 3, then we have 0 E K and the
.
remaining part is the same as above.

(C) Quartic equations. Let L be the minimal splitting field of a quartic
equation f (X) = x4 + a I x 3 + + a4 = 0 over the field K = Q(a1 , ... , a4) .
Then the Galois group G(L/K) is a subgroup of the symmetric group S4 of
degree 4.
As for the group S4, {1, (1, 2)(3, 4), (1, 3)(2, 4), (1, 4)(2, 3)} (note
that
(1' 2)(3' 4) = l (1' 3)(2' 4) ( 1 2 3 4)

2 1 4 3/ ' 3 4 1 2 '
and so on) forms a normal subgroup N of order 4, and S4/N ' S3 (we
omit the proof). Let us pay attention to the case G(L/K) = S4 (the existence
of this case is seen in (D) below). Then we see that we have to solve a cubic
equation in order to obtain the intermediate field corresponding to N. (Then
we solve two quadratic equations.) This is the reason why we have to solve
a cubic equation in solving a quartic equation.
11. CONSTRUCTION PROBLEMS 67
(D) Equations of degree 5 or more. It is known that

THEOREM 2.10.8. For any given natural number n (> 2), there is an
f (X)
equation = x" + a1 xr-1 + + an = 0 of degree n such that its Galois
group is the symmetric group Sn of degree n.
Though this theorem is true under the stronger condition that all the ai
are rational numbers (as will be proved in Chapter 6, 4) we are proving
here this weaker assertion by using the following lemma (whose proof will be
given in 12, Appendix 2):
LEMMA 2.10.9. There are complex numbers bl , ... , bn which are alge-
braically independent over the rational number field Q.
PROOF OF THEOREM 2.10.8. Take bl , ... , bn as above. Consider the sym-
metric group Sn on these bi and the polynomial f (x) = f I1 (x - bi) =
xn + ctxn-1 + + cn . Then (-1)`c1 (i = 1 , ... , n) are the elementary
symmetric forms of b1 , ... , bn. Each element of Sn induces an automor-
phism of Q(b1 , ... , bn), and therefore, Sn c AutQ(b1 , ... , bn) .
Let g(b1 , ... , bn) be an Sn-invariant element of Q(b, , ... , bn). Then
g(bl , ... , bn) = h(b1, ... , bn)/k(b1 , ... , bn) with h, k e Q[b, , ... , bn].
We may assume that k is a symmetric form, by changing the denomina-

tor to HOES Qk if necessary. Then, since g is Sn-invariant, we see that
h is also symmetric. Therefore the field of Sn-invariants coincides with
Q(cl , ... , cn) . Thus G(Q(b1 , ... , bn)/Q(cl , ... , cn)) = Sn and the Galois
group of f(x) = fI;(x - bi) is Sn. Q.E.D.
As is well known, Sn is not solvable if n > 5 (we omit the proof). There-
fore, among algebraic equations of degree > 5 over a field of characteristic
0, there are those which cannot be solved algebraically.
11. Construction problems
In construction problems in geometry, we are asked to obtain certain fig-
ures (circles, lines, half lines, line segments, points) by using a rule and a
pair of compasses. To obtain a line, we have only to find two points on the
line; as for a line segment, we need its two endpoints; for a circle, we need
its center and the length of its radius, and so on. Therefore, the solvability
of a construction problem can be proved if we have a characterization of the
points that we can obtain by using a rule and a pair of compasses, starting
with given points that characterize some given figures.
For this purpose, we have two methods which are substantially the same:
One is to consider the plane with real coordinates, taking one of the given
points to be the origin and another point to be (1 , 0) . The other method is
to consider the Gauss plane similarly. We shall employ the latter one.
REMARK. If an angle is given, then its vertex is a given point and points on
the edges which are of known distance from the vertex are also given points
(If only one point or no point is given, then we can addsome point(s), and
we may assume that 2 points are given and that the distance between them
defines the unit of length.) But, to take an arbitrary point on each edge will
cause some difficulty because we know nothing about the distances of these
points from the vertex. Similar comments apply to points on given lines,
given half lines, or circles.
In this section, we use Greek letters a, /3 , ... to denote complex numbers;
the symbol denotes the complex conjugate. Q denotes the rational number
field.
THEOREM 2.11.1. If 0, 1 , a I , a2 , ... , an are the given points, then a
point /3 is constructed if and only if there is a sequence Ko c KI c c Kt
of fields K. such that Ko = Q(a I , ... , an , dl , ... , an ) and such that
(i) K,D/3,
(ii) [K,:K;-I]=2 (i 1,2,...,t).
PROOF. To prove the if part, note that the point of symmetry of a given
point with respect to a given line can be constructed. Therefore, ai (i =
1) 2 , ... , n) are constructed. The product (or ratio) of two complex num-
bers is obtained by adding (or subtracting) arguments and multiplying (or
dividing) absolute values. The sum of two complex numbers is the sum
as vectors. Therefore, the product and the sum of constructible complex
numbers (i.e., points) are constructible. If a is constructible, then na,
-1
n a (with n a natural number) and -a are constructible. We see that
elements of Ko are constructible. So, we shall show that elements of K, are
constructible by induction on i. Assume that elements of K._ are con- I
structible. Since K. is a quadratic extension of Ki_ , there is /3, E Kl_

I I
such that Ki = K,_1 (J) . The argument and the absolute value of
are one half and the square root of those of /3 (note that the argument is
unique modulo 27r, and therefore, there are two choices for the argument of
modulo 27r). Therefore, is constructible. Thus, the elements of
Ki are constructible.
To prove the only if part, let us see how to construct a new point. There
are 3 types of constructions:
(1) the intersection of 2 lines.
(2) one of the intersection points of a line with a circle,
(3) one of the intersection points of two circles.
Therefore, assuming that K is a field consisting of constructible com-
plex numbers and satisfying the condition that a c K for all a c K (note
that Ko satisfies these conditions), we consider these three types of construc-
tions to obtain a new point 6. If we can show that [K(8) : K] < 2 and
[K(8 , j): K(8)] < 2, then we can take successive field extensions, satisfying
(ii), until we obtain /3 .
Case 1. Suppose two lines be lI and 12 go through two points yi , y2

and two points y3 , y4 , respectively, with yI , y2 , y3 , y4 E K such that y, 54
y2 1 y3 y4 . Points on 11 and 12 are expressed as y1 + t(y2 - y,) and
11. CONSTRUCTION PROBLEMS 69
y3 + u (y4 - y3) , respectively, with real numbers t , u . The intersection a is

determined by the solution of y1 + t(y2 - y1) = y3 + u(y4 - y3) . Since t, u
are real numbers, we have yl + t(y2 - y1) = 73 + u(74 - 73) . If (y2 - yl)(y4 -
73) - (y2 - y1)(y4 - y3) = 0, it follows that (y2 - yl)(y4 - y3) is a real number
which means that the argument 01 of y2 - yl (which coincides with the
angle between 11 and the real axis) and the argument 02 of y4 - y3 satisfy
01 + 02 = nn with rational integer n. Since the angle between 12 and the
real axis is -02 , it follows that 11 and 12 are parallel to each other, which
is not the case. Therefore (y2 - yl)(y4 - 73) - (72 - y1)(y4 - y3) 54 0 and the
solution of t is expressed as a rational form in y1 , y2 , y3 , y4 , yl 1 y2 , y3 , y4
Therefore, 6 = yl + t(y2 - y1) E K and 6, a r= K in this case.
Case 2. Let the line 1 go through yl , y2 E K (y1 # y2) , and let the circle
C have center q E K and radius r. If a is an intersection point, then
a = yl + t(y2 - y1) , 18 - qI = r E K, where t is a real number. Then
(y1 + *2 - yl) - q)(y1 + t(y2 - y1) - Y]) = r2 . This is a quadratic equation
on t, and we see that [K(6): K] < 2, [K(6, a) : K(a)] < 2.
Case 3. Let the centers and radii of the two circles be ql , rig E K and
r1 , r2 E K, respectively (ql # q2) . If 6 is an intersection point, then 6
satisfies the equations (a - q 1) (a - q 1) = ri , ((5 - q2) (a - 42) = . Namely,
1.22
as - (8ql + an;) + q;#; = r? (i = 1 , 2) . The difference of these equalities

is (q2 - ql)( + (72 - q1)(5 + glril - q2-q2 = r - r2 . Since ql q21 we have
a = (ri - r2 - q1 q1 + g2g2 - (q2 - g1)(5)/(g2 - q1) By this and the first
equation, we have a quadratic equation on 6, and the situation is similar to
Case 2. Q.E.D.
COROLLARY 2.11.2. In the theorem above, even if we add a condition that
Kt is a Galois extension of K0, we have an equivalent condition.
PROOF. Sufficiency is obvious. If we have a K, as in the theorem, then
every conjugate of K, over Ko has a similar sequence of intermediate fields;
hence, the same is true of the field generated by all conjugates of Kt . Q.E.D.
COROLLARY 2.11.3. Let fl and Ko be as in Theorem 2.11.1. If ,13 is
constructible, then the minimal polynomial for /3 over Ko must be of degree
a power of 2. (The converse is not true; cf. Exercise 2.11.3).
As applications of the above observations, we shall discuss the construction
of regular polygons and the trisection of angles.
To construct a regular n-gon is to construct the angle 27r/n or to construct
a primitive nth root C of unity. (The given points are 0, 1 only, and Ko =
Q in this case.) The minimal polynomial for C over Q is the cyclotomic
polynomial F (x) whose degree is 4(n) (= the number of those among
1 , 2, ... , n that are coprime to n) by Theorem 2.9.6. Therefore, if t; is
constructible, then 5(n) must be a power 2e of 2.

We consider first the case where n is a prime number p . In this case,

q5(p) = p -1 , and therefore p = 2e + 1 . Conversely, if p = 2e + 1 , then Q(C)
is a Galois extension of Q with Galois group G, which is commutative and
of order 2e , by Theorem 2.9.2, and we have the existence of a sequence of
fields as in Theorem 2.11.1 (cf. Exercise 1.2.9). Thus, we have the following
theorem.
THEOREM 2.11.4. If p is an odd prime number, then regular p-gon can be
-
constructed if and only if p 1 = 2e with a natural number e.
The general case can be stated as follows.
THEOREM 2.11.5. The regular n-gon (n > 3) is constructible iff n =
2ep1 pr with e a nonnegative rational integer and pI , ... , pr mutually
different odd prime numbers such that each pi = 1 + 2e" (with ej a natural
number).
PROOF. To prove the if part, note first that since the bisection of an angle
is constructible, it suffices to prove the case e = 0. Theorem 2.11.4 is the
case r = 1 , and we use an induction argument on r. Let y I and C2 be
roots of unity with arguments 2r/p1 pr_ and 2n/p,. , respectively (note
I
that these are a primitive PI pr_ i th root of unity and a primitive Prth
root of unity, respectively). If t, u are rational integers, then the argument
of C1 C2 is 2n(t/(p1 . p . ) ) + u/pr) = 2n(tpr + upI ... pr- I )l n . Since pr
1
and PI . . Pr- I are coprime, a primitive nth root of unity is obtained in the
.
form C' (2 Since (I , (2 are constructible, we see that primitive nth roots
.
of unity are constructible, which means that a regular n-gon is constructible.

To prove the only if part, note that if a regular n-gon is constructible, then
for any divisor n' > 3 of n, a regular n'-gon is constructible. Therefore n
must be of the form 2epc' .. p'r with odd prime numbers pi of the form
e,
1+2 Therefore, it suffices to show c. > 2 cannot occur. Therefore, we have
.
only to show that any regular p2-gon is not constructible with p = pi. This
-
follows from the equality q5(p2) = p(p 1) and Corollary 2.11.3. Q.E.D.
COROLLARY 2.11.6. There is an angle whose trisection is not constructible.
More generally, if p is an odd prime number, then there is an angle whose
p-section is not constructible.
PROOF. A regular p2-gon is not constructible. This implies that p-section
of either 2mr or 2n/p is not constructible. Q.E.D.
COROLLARY 2.11.7. n-section of every angle is constructible iff n is a power
of 2.
PROOF. If the n-section of an angle is constructible, then for any divisor
n' of n , the n'-section of the angle is constructible. Therefore, the assertion
follows from Corollary 2.11.6. Q.E.D.
12. ALGEBRAICALLY CLOSED FIELDS 71
12. Algebraically closed fields

A field K is algebraically closed if there is no proper algebraic extension
of K, or equivalently, if every polynomial in one variable over K factors
into the product of linear factors. It is well known that the complex number
field C is algebraically closed; we shall give a proof in Appendix 1 of this
section.
If a field K is algebraic over a field K and if K is algebraically closed,
then K is called an algebraic closure of K. The separable closure of K in
is called the separable algebraic closure of K.
When K is a subfield of a field L, then the set K* of algebraic elements
of L over K, is called the algebraic closure of K in L. We say that K is
algebraically closed in L if K* = K.
THEOREM 2.12.1. Let K, L, and K* be as above. Then K* is an alge-
braic extension field of K and is algebraically closed in L. Therefore, if L
is algebraically closed, then K* is an algebraic closure of K.
PROOF. If a, b c K* , then K(a, b) is algebraic over K (Corollary 2.1.5).
Since ab, a + b, a-1 (if a # 0) are in K(a, b) c_ K* , we see that K* is a
ixi-1
field. If x c L is algebraic over K* , then there is a relation: x" +C +
+ c" = 0 (c1 E K*). Then Theorem 2.1.4 shows that K(c1 , ... , c , x) is
algebraic over K. Therefore, K* is algebraically closed in L. Q.E.D.
EXAMPLE. Since the complex number field C is algebraically closed, the
set Q of complex numbers that are algebraic over the rational number field
Q is an algebraic closure of Q. Elements of Q are called algebraic numbers.
Those elements of Q which are integral over the ring Z of rational integers
are called algebraic integers.
THEOREM 2.12.2. For any field K, we have the following:
(i) There exists an algebraic closure of K.
(ii) If T, K are algebraic closures of K, then K and K 1
1
are
K-isomorphic. (Therefore, algebraic closures are essentially unique.)
(iii) If an algebraic extension field L of K satisfies the following condition,
then L is an algebraic closure of K :
If Ax) = x" + cl xi-1 + (c1 E K, n > 1) is an
+ c"
irreducible monic polynomial over K, then L contains at
least one root of f (x) .
PROOF. (iii). It suffices to show that if a is an element of an algebraic
extension field L' of L, then a E L. Such an a is algebraic over K as is
seen by our proof of Theorem 2.12.1. Let f (x) be the minimal polynomial
for a over K, and let S be the minimal splitting field of f (x) over K.
Since S is normal over K, S is generated by a Galois extension FI of finite
degree over K and a purely inseparable extension F2 of finite degree over
K (Corollary 2.7.2). Choose an element b such that FI = K(b) (Theorem
2.5.2). Take elements e1 , ... , em of F2 which generate F2 over K. Let

g(x) be the minimal polynomial for b over K. Then, by our assumption,
a root b' of g(x) is in L. Since K(b) is a Galois extension, we see that
b' E K(b), [K(b') : K] = [K(b) : K] (= deg g(x)) . Therefore K(b') = K(b)
and FI C L. Every root of the minimal polynomial hi (x) for ei is ei, and
therefore, every ei is in L. Namely, F2 C L. Thus, S = F1 vF2 C L. Since
aES,wehave aEL.
(i). We shall make use of a polynomial ring P with infinitely many vari-
ables. Namely, taking the set Z of irreducible monic polynomials in one
variable x, over K we consider indeterminates {X lj E .1} indexed by
the set f . Then P is the union of all polynomial rings in a finite num-
ber of these Xi over K. (Thus, we are taking an indeterminate Xh for
each irreducible monic polynomial h = h(x) over K.) Let I be the ideal
of P generated by {h (Xh) I h (x) E f'} and let M be a maximal ideal of
P containing I. (Proof of the existence of M. By the remark after The-
orem 1.4.2, it suffices to show that 1 is not in I. Assume that 1 E I.
Then 1 = f1 h1(XhI ) + + fshs(Xhs) with f E P and hi E A/. Let all of
the X appearing in this expression be Xh, , ... , Xh. Then the ideal J of
P' = K[Xh , ... , Xh generated by hi h(X11) contains . On 1
the other hand, taking a splitting field c2 of ! li hi (x) over K , we consider

a root /3i for each hi(x). Then we have a K-homomorphism from P' into
K(f , ... , /3r) such that Xh is mapped to /3i The image of the right-hand
1 .
side of the above equality is 0, a contradiction. Therefore, 1 is not in I.)

Let K be P/M, which is a field (Theorem 1.4.3). The residue class of Xh
is a root of h(x), and therefore, K is algebraic over K (Corollary 2.1.5).
Each irreducible monic polynomial over K has a root in K, which implies
that K is an algebraic closure of K by (iii) proved above.
(ii) We consider intermediate fields L between K and K such that there
are K-isomorphisms a from L into K 1 . Let S be the set of all pairs
(L, a) of such L and a. Since (K, 1) is in S, S is not empty. We
introduce an order on S by setting (Li , (71) < (L2 , a2) if L1 C L2 and
the restriction of a2 to LI coincides with a1 . It is easy to see that S is
an ordered set. S is really an inductive set. (Indeed, if {(L1, Qi) I i E I)
is a well-ordered subset of S, then we define a mapping a of L = Ui Li
into KI by as = uia with i such that a E Li By our definition of the
.
order, the fact that these Li form an ascending sequence and that L is
K.
a field, we see that a is a well-defined K-isomorphism from L into K I
Then (L, a) is the supremum of the well-ordered subset. Therefore S is an

inductive set.) By Zorn's lemma (see Chapter 0), S has a maximal member,
say (L*, a*). Suppose for a moment that L* # K. Let a be an element
of K which is not in L* and let f (x) = xd + e1 xd -1 + + ed be the
minimal polynomial for a over L*. Take a root b of a* f (x) = xd +
a*elxd-1 + - - + a'*ed . Then, since a* f(x) is irreducible over, a*L* , there
APPENDIX I 73
is an isomorphism a' of L*(a) to (u*L*)(b) such that a' is an extension

of a*. (Indeed, L*(a) L*[x]/f(x)L*[x] ((Q*L*)[x]/Q*f(x)(a*L*)[x]
(a*L*)(b).) This contradicts the maximality of (L*, u*). Therefore, L* _
K. Since K1 is algebraic over the algebraically closed field u*K, we see that
a*K = K1 . Therefore, K is K-isomorphic to K1 . Q.E.D.
From the proof of (ii) above, we have the following theorem.
THEOREM 2.12.3. Let K be the algebraic closure of a f eld K. If L1 C L2
are algebraic extensions of K and if a is an K-isomorphism from L1 into
K, then a is extended to a K-isomorphism from L2 into K.
Appendix 1
THEOREM 2.12.4. The complex number field C is algebraically closed.
A well-known proof of this theorem uses two facts: If f is a real valued
continuous function on a bounded closed set, then f takes a minimum value
at some point, and that for a polynomial f(x) in one variable x over C,
if F(x) = If(x)I has a nonzero value at a E C, then F(x) takes a smaller
value at a point on a sufficiently small circle with center a. Here, we shall
give another proof as an application of the fundamental theorem of Galois.
We use the existence theorem of Sylow subgroups (Chapter I, 2) and the
fact that if a polynomial g(x) in x over the real number field R is of odd
degree, then g(x) has a linear factor (as is obvious by the continuity of the
real numbers).
PROOF. Take an algebraic element c over C. We have only to show that
c E C. Let L be a Galois extension of finite degree over R, such that
C(c) C_ L. Let S be a 2-Sylow subgroup of G = G(L/R) and let M be
the intermediate field corresponding to S. S = G(L/M), #(S) = [L: M],
and therefore, [M: R] = #(G)/ #(S) , which is odd. By the reducibility of
polynomials of odd degree, we have M = R, and #(G) is a power of 2.
If #(G) = 2, then L = C. Assume for a moment that #(G) = 2n with
n > 1 . Letting H be the subgroup of G corresponding to C, we take a
subgroup H* of H of index 2 (the existence of H* follows from the fact
that H is nilpotent (Theorem 1.2.6), cf. Exercise 1.2.9). Let C* be the
field corresponding to H*. Then C* is a quadratic extension of C, and is
generated by a root of some quadratic equation x2 + tax + b = 0 (a, b E C).
The roots are -a f (a2 - b). Therefore, if we see that the square roots of
an arbitrary complex number r + 2s (r, s E R) are in C, then C* = C,
contradicting n > 1 , which would complete the proof. As for the square roots
of r + 2s , the existence is obvious if s = 0 , so we assume that s 0 0 . It
suffices to find real numbers x, y such that (x + y )2 = r + 2sV-_l . The
conditions are x2 - y2 = r and xy = s. Since x4 - x2y2 = rx2 , we have
x4 - rx2 - s2 = 0. Then, since x = V(r + r2 + 4s2)/2 E R and y = s/x,
we obtain square roots (x + yam) of r + 2s . Q.E.D.
Appendix 2
THEOREM 2.12.5. There is a subset S of the real number field R such

that #(S) is the cardinality of the continuum and such that S is algebraically
independent over the rational number field Q (namely, every finite subset of
S consists of algebraically independent elements over Q).
PROOF. Consider the set A of algebraically independent subsets of R over

Q. Then A is an inductive set by the order of inclusion relation. Then, by
Zorn's lemma, there is a maximal member S of A (this is a transcendence
base to be defined in Chapter III). If there is a transcendental element x E R
over Q(S) , then SU{x} is algebraically independent, contradicting the max-
imality of S. Therefore, R is algebraic over Q(S). We consider #(Q(S)).
We denote by /3 the cardinality #(S) . Let d be the cardinality of the set
Md of monomials of degree d in elements of S. Then (i) if /3 is finite,
then d is finite and (ii) if /3 is infinite, then /3 < Yd < #(S x x S)
d
and we have Yd = /3 (Theorem 0.1.3). As for the cardinality ,u of the set
M = Ud Md (the set of monomials), (i) if /3 is finite, then ,u is count-
ably infinite and (ii) if /3 is infinite, then, since u < /3 x /3 = fl, we see
that It = fl. Next we consider the cardinality of the polynomial ring Q[S]
in elements of S over Q. For each finite subset N of M, we consider
the set P(N) of polynomials that are in the vector space spanned by mem-
bers of N over Q. #(P(N)) < #(N) x #(Q) = #(Q) = countably infinite.
Consider the set Fm of subsets of M consisting of m elements. Then
#(M) < #(Fm) < (#(M))'n = #(M), and we have #(M) = It .
Q[S] = U,n F,,,, and therefore, #(Q[S]) < (countably infinite) x u, and we
see that #(Q[S]) = u (which is countably infinite if S is a finite set, oth-
erwise, it coincides with #(S)). Since elements of Q(S) are expressed as
ratios of elements of Q[S], we have #(Q(S)) = y. Therefore, we see that
#(S) is the cardinality of continuum by the lemma below.
LEMMA 2.12.6. Let L be an algebraic extension of a field K. If K is an

infinite field, then #(L) = #(K). If K is a finite field, then #(K) < #(L) <
(countably infinite).
PROOF. If K is an infinite field, then we see that #(K[x]) = #(K) as

above. Let M* be the set of irreducible monic polynomials in K[x]. Then
#(M*) < #(K[x]) = #(K). Therefore, if we take the polynomial ring P
that was used in the proof of Theorem 2.12.2, we see that #(P) = #(K) .
Since the algebraic closure K of K is a homomorphic image of P, we have

#(K) < #(P). Thus, #(K) < #(L) < #(K) < #(P) = #(K). Therefore,
#(L) = #(K). A similar argument works in the case where K is a finite
field. Q.E.D.
EXERCISES 75
Exercises
1
1. Assume that K1 , K2 are finite fields such that K1 C_ K2, n = [K2: K1].
Prove that #(K2) = (#(K1))" .
2. Prove the following equalities on the degrees of extension fields of the
rational number field Q. Here, co denotes an imaginary cubic root of
unity.
[Q(vT) : Q] = 2, [Q(w) : Q] = 2, [Q(' 2) : Q] = 3.
3. If a is an algebraic element over a field K, by the degree of a over K
we mean the degree of the minimal polynomial for a over K. Prove
that if a is an element of a finite algebraic extension field L of K, then
the degree of a over K is a divisor of [L: K].
4. Let K, L, M be fields such that L, M are algebraic extensions of K, L,
respectively. Prove that M is algebraic over K.
5. Let K be a field of characteristic p 54 0 and let q be p" with a natural
number n. Prove that Kq = {xq,X E K} forms a subfield of K.
2
1. Find the minimal splitting field for each of the following polynomials
over the rational number field Q.
(1) X4 - X2 +4, (2) x3 - 1 , (3) x3 - 2.
2. Find the degree of extension of each field obtained above.
3
1. Determine which of the following polynomials have multiple roots. We
assume that the coefficients are in some field K. Note that the answers
depend on the characteristic of K. In (4), we assume that p is the
characteristic of K and that p 54 0.
(1) X5+x, (2) x4+X+ 1, (3) x4+2X3+3x2+8x+ 1,
n n-1
(4) cOXp +C1Xp + +Cn_1XP+C,i+dx (C1) d E K, co 54 O).
2. Give a proof of Theorem 2.3.5 along the following lines. If K is a
finite field consisting of pe elements (p being the characteristic), then
K* = K - {0} is a group of order pe _1 , and therefore the order of each
element a is coprime to p. By using the fact, with some power at of
a, show that (a`)p = a.
3. Let K be a field of characteristic p 54 0. Prove that if t is a transcen-
dental element over K, then K(t) is not a perfect field.
4. Let L be an extension field of a field K. Prove that the set Li of
elements of L, which are purely inseparable over K, forms a field.
5. When a natural number n and field K of characteristic p 0 is given,
we denote by K' the set of pth " roots of elements of K (in a suffi-
ciently large field) and by Kp -00 the field U00 I Kp- .
(1) Prove that K" is purely inseparable over K.

(2) Prove that the mapping 0n given by na = ap is an isomorphism
of Kp to K.
(3) Prove that K" -00 is the smallest perfect field containing K.
6. Let K be a field of characteristic p 0 . Prove that an algebraic element
a is separable over K if and only if K(a) = K(ap).
7. Let f(x) be a polynomial in x over a field K, and let g(x) be its
derivative (d /d x) f (x) . Prove that if a is a root of f (x) of multiplicity
e > 2, then a is a root of g(x) of multiplicity > e - 1 .
1. Consider a prime number p and the ring Z of rational integers. Prove

that if p is an odd prime, then the group of units of Z/p"Z is a cyclic
group of order pn- I (p - 1), by using the fact that the group of units of
Z/pZ is cyclic.
Examine the structure of the group of units of Z/pnZ when p = 2.
2. Let G be a commutative, finite group. Prove that if, for each prime
number q, the number of solutions of xq = 1 in G is at most q , then
G is a cyclic group.
3. In the problem above, if we omit the commutativity condition on G,
then G may not be cyclic. Find such an example among the groups of
order 8.
5
1. Find a simple example of a finite algebraic extension which is not a simple

extension.
2. Let L be a simple extension of a field K. Prove that if n = [L: K] ,
then the number of intermediate fields (including K and L) is less than
2n .
1. Let L be a finite algebraic extension of a field K.

(1) Prove that if L is normal over K, then the separable closure LS
of K in L is normal over K (hence, is a Galois extension of K).
(2) Show that the converse of the assertion in (1) is not true.
2. Prove that if LI , ... , Ln are normal extensions of a field K and are
contained in a common field, then nl L. is also normal over K.
3. Let a field L be a simple algebraic extension of a field K, and let n
be [L: K]. Prove that there is the smallest extension L* among normal
extensions of K containing L (within a sufficiently large field) and that
[L* : K] < W.
EXERCISES 77
4. Let Q denote the rational number field. Show the following: Q(v ) is
a normal extension of Q and Q(" 3) is a normal extension of Q(/) ,
but Q(4 3) is not a normal extension of Q.
5. For each of the following fields, find the smallest normal extension con-
taining it over the rational number field Q. Find, also, the Galois group
of each of the normal extensions.
(1) Q(v) (2) Q(v, /) (3) Q(3 2) (4) Q(4 2)
7
1. Find the structure of the following groups, where Q denotes the rational
number field.
(1) AutQ(3 2) , (2) AutQ(, r2),
(3) AutK(X2) K(x) (where x is transcendental over a field K) ,
(4) AutK(X3) K(x) (with K, x as above).
2. Find the fields of invariants of the groups in 1 . above.
3. Let a field SZ be a finite normal extension of a field K, and let L be
an intermediate field. Let s, t be the separable and inseparable factors
of degree of L over K. Let oI , ... , ors be a set of representatives
of G(S /L)\G(S2/K) . For each element a of L, we consider a system
consisting of a,a such that each a.a appears exactly t times. We call
such a system the complete system of conjugates of a in L over K (we do
not care in what order they are arranged). Let the system be aI , ... , an
(n = st). Then Ei= a! , fJ i= I a, are called the trace and the norm of a,
I
respectively, and they are denoted by TrL/K (a) , NL/K (a) , respectively.
Prove the following.
(1) TrL/K (a) E K, NL/K (a) E K for all a c L.
(2) TrL/K(a+b)=TrL/K(a)+TrL/K(b) and NL/K(ab)=NL/K(a)NL/K(b)
for all a, b c L.
(3) Let b1 , ... , bn be a linearly independent base of L as a K-module.
For each a E L, consider the linear transformation of L given
by x -> ax. Let p(a) be the matrix which represents this linear
transformation with respect to b1 , ... , bn . Show that TrL/K (a) is
the trace (i.e., the sum of diagonal entries) of p(a) and that NL/K(a)
is the determinant of p(a) .
(4) Prove that if L' is a finite algebraic extension of L, then
TrL/K TrL,/L = TrL?/K , NL/K NL'IL = NL,/K
(5) Prove that TrL/K(a) = 0 for every a E L if and only if L is
inseparable over K.
4. Let x be a transcendental element over the complex number field C,
and consider the field C(x). For a given natural number n, let C be
(i.e., the primitive nth root of unity with argument 2n/n) .
(1) Prove that there is a unique pair a, r c Autc C(x) such that ax =
I,
x - 'Cx=Cx.
(2) Show that these a and T satisfy a2 = 1 , T" = 1 , ara- = T-I
and that the subgroup G generated by a , T has order 2n.

(3) Show that the field of G-invariants coincides with C(y) where y =
x" + x-" .
1. Prove that if a field L is a quadratic extension of a field K, then L is

a normal extension of K.
2. Find the number of intermediate fields of each of the following pairs of
fields:
(1) the rational number field Q and Q(V, vf3-, ,r5-),
(2) Q and Q(co, r2), where co is an imaginary cubic root of unity,
(3) Q and Q(/).
9
1. Prove the converse of Theorem 2.9.8, i.e., that if a field K contains all
nth roots of unity, if the characteristic p of K is either 0 or prime to
the given natural number n and if x" - a (a c K) is an irreducible
polynomial over K, then K(" a) is a cyclic extension of degree n over
K.
2. A field L is called a Kummer extension of a field K if (i) L is an Abelian
extension of K of finite degree and (ii) x" - 1 has no multiple root and
K is a splitting field of x" - 1 , where n is the maximum of the orders
of elements of G(L/K).
Prove the statement below using the fact that every finite commutative
group is the direct product of cyclic groups.
A Kummer extension L, as above, is obtained as
K(., aI , 'V a2 , ... , ,Var )
with a! E K, n is the L.C.M. of n , ... , nr and each x -ai
ai_I , n,fai+I , ... , a).
I
is irreducible over
Prove the converse of this statement assuming the same condition on
K.
3. Prove that if n is a power of an odd prime number, then the cyclotomic
field of order n over an arbitrary field K is a cyclic extension of K.
Describe the case where n is a power of 2. [Hint. Use Exercise 2.4. 1.1
4. Let L be a separable algebraic extension of degree n over a field K .
Let aI , ... , a" be the mutually distinct K-isomorphisms from L into a
sufficiently large field 0, and let wI , ... , w" be a linearly independent
EXERCISES 79
base of L as a K-module. Prove that the determinant of the (n, n)

matrix A = (ci cwt) (= the matrix whose (i, j) entry is ci w) is not 0.
5. Assume that K is an infinite field, and prove the generalization of the as-
sertion in 4 above as follows: A polynomial f (01 , ... , or = >2 ci .t a
n
Qnn in or, , ... , an with coefficients in S

denotes the mapping of L
into 92 which sends an element a in L to E c; i n (Q1 a)... (ana)'n
I
(E 0). Then the statement to be proved is that if f (X , ... 1 , Xn )

E 52[X1, ... , X J is not 0 as a polynomial, then the mapping
f(c1 , ... , an) is not 0. [Hint. Let wi be as above. If f(a1 , ... , an) _
0, then f(ala, ... , ana) = 0 for all a E L and f(>xi(Q1wi), ... ,
> X (an (vi)) = 0 as a polynomial in x , ... , xn . Then use exercise 4.]
1
6. Let L be a finite algebraic extension of a field K. If there is an element

u of L such that conjugates of u form a linearly independent base of
L over K, then such a linearly independent base is called a normal base
of L over K.
(1) Prove that if L has a normal base over K, then L is a Galois
extension of K.
(2) Prove the converse in the following form: Assume that L is a Galois
extension of finite degree over K.
(i) Let a1 , ... , an (n = [L: K]) be the elements of G(L/K) . Then
the conjugates of u E L form a normal base if the determinant
of the (n, n) matrix A = (a1aju) (= the matrix whose (i, j)
entry is ai a1 u) is not zero.
(ii) L has a normal base over K.
7. Let L be a cyclic extension of a field K, and let o be a generator
of G(L/K). Assume that [L: K] is finite. Prove the following two
assertions.
(1) (Hilbert's Theorem 90) For an element x of L, NLIK (x) = 1 if
there is a nonzero element y in L such that x = y/(ay) . In this
case, x = y1/(ay1) with y1 E L iff y1y-1 E K.
(2) For X E L, TrLIK(x) = 0 if there is an element z c L such that
x=z-az. In this case, x=z1-az1 with z1 EL if z-z1 EK.
10
1. Prove that if L1 , ... , Ln are finite Abelian extensions of a field K, then

so is the field generated by these Li .
2. Prove that if L1 , ... , Ln are Galois extensions of finite degree over a
field K with solvable Galois groups G(Li/K) (i = 1, ... , n), then so
is the field generated by these Li .
3. Let L be a finite algebraic extension of a field K.
(1) Prove that L is a Galois extension of K with solvable Galois group
G(L/K) if (i) L is a normal extension of K and (ii) there is a
sequence K = Ko C KI C C Kr = L of intermediate fields Ki

such that each KI (i = 1 , ... , r) is a cyclic extension of K,_I .
(2) Assume that condition (ii) above holds. Let L* be the smallest Ga-
lois extension of K containing L. Prove that G(L*/K) is solvable.
Show also that L* may be different from L even if r = 2.
4. Find the structure of the Galois groups of the following polynomials
whose coefficients are complex numbers:
(1) The sextic polynomial having roots v, + , + ,
(2) xn - 3 with n a natural number;

(3) x3 +X +I;
(4) x3-x+1;
(5) x4+x2+1;
(6) X4 - X2 + 1 .
11
1. Consider the construction of the angle of n with n a natural number.

Prove that the angle is constructible if and only if n is a multiple of 3.
2. Try to discuss the result of our Theorem 2.11.1 using real coordinates
and subfields of the real number field.
3. Consider the construction of a point represented by a complex number a
with given points 0, 1 only. Show that it is not necessarily constructable,
even if the degree of the minimal polynomial for a over the rational
number field Q is a power of 2, by using a root a of x4 + x + 1 .
12
1. Show that the fact that every irreducible polynomial in one variable over
the real number field R is of degree at most 2 follows from the facts that
the complex number field C is algebraically closed and that [C: R] = 2.
2. Let K be the algebraic closure of a field K. Prove that the field K* of
G(K/K)-invariants in K is the smallest perfect field containing K.
3. Let 7C be the prime field of characteristic p i4 0. Show that there is a
field K such that K is algebraic over n and #(K) = oc, but K is not
algebraically closed.
CHAPTER III
Transcendental Extensions
1. Transcendence bases
Among transcendental extensions of a field, there are special extensions
called purely transcendental extensions, which are defined as follows: A field
L is called a purely transcendental extension of a field K if L is generated
by an algebraically independent set S over K. L is also called the rational
function field of S over K. L is the field of fractions of the polynomial
ring in elements of S (which may be infinite) over K.
As we saw in the proof of Theorem 2.12.5, we have the following result.
THEOREM 3.1.1. If K is a subfield of a field L, then there is a maximal
member S among algebraically independent subsets of L over K and L is
algebraic over K(S).
This S is called a transcendence base of L over K. If L is separable
over K(S), then S is called a separating transcendence base (which may not
exist).
THEOREM 3.1.2. If K is a subfield of a field L and if both S and S' are
transcendence bases of L over K, then #(S) = #(S').
This #(S) is called the transcendence degree, or the dimension, of L over
K . We denote this by trans. degK L.
If A D I are integral domains, then the transcendence degree, denoted
by trans. deg, A , of A over I is defined to be trans. degK L with fields of
fractions L, K of A, I, respectively.
PROOF. (1) In the case where #(S) is finite, we may assume that n =
#(S) < #(S'). Let S be {x1 , ... , xn} . Take mutually different elements
yl , .. , yn of S' S. It suffices to show that, by a suitable renumbering of
xi , L is algebraic over Ki = K(y1 , ... , y; , xi+1 , ... , xn) for each i =
0, 1 , ... , n, because, then L must be algebraic over K(y.1 , ... , yn) and
S' = {y1 , ... , yn} . This assertion is true if i = 0. Therefore, assuming that
L is algebraic over K5_ , we consider the case over KS . Since ys is algebraic
1
over K,-, , there are co, ... , cl E K[y1 , ... , ys-1 , xs, ... , xn] such that
co 54 0, coys + clys-1 + + ct = 0. Since y1 , ... , ys are algebraically
independent over K, at least one xj (j > s) must be contained in some
81
82 III. TRANSCENDENTAL EXTENSIONS
cj . By renumbering if necessary, we may assume that j = s . Then xs is

algebraic over KS = K(yI , ... , YS , xs+I , .. , x,) . Since L is algebraic over
Ks_ I , we see that L is algebraic over KK .
(2) In the case where #(S) is infinite, we may assume that #(S) < #(S') .
Every element of S' is algebraic over K(S) , and therefore, for each y c S',
yin- I
there are ci E K(S) such that y'n + cI + + cm = 0. The number of
elements of S, appearing in some of these c, , is finite. Since this is true of
each element of S', we see that #(S') < #(S) x (countably infinite) = #(S)
(Theorem 0.1.3). Thus #(S') = #(S). Q.E.D.
REMARK. A field L is algebraic over its subfield K if trans. degK L = 0.
We say that a field L is finitely generated over a field K if there are ele-
ments aI , ... , an such that L = K(a1 , ... , an). The next theorem follows
immediately from the proof above.
THEOREM 3.1.3. If a field L is finitely generated over its subfield K, then
there are algebraically independent elements xI , ... , xn over K such that L
is finite algebraic over K(x1 , ... , xn) .
THEOREM 3.1.4. If a field L is finitely generated over its subfield K, then
every intermediate field M between K and L is also finitely generated over
K
PROOF. (i) In the case where M is algebraic over K, let xI , ... , xn
be a transcendence base of L over K K. Then they are algebraically inde-
pendent over M. If m,' ... , mr E M are linearly independent over K,
then they are linearly independent over K(x1 , ... , xn). (Indeed, if there
is a linear relation of m , ... , mr over K(x1 , ... , xn), then the relation
I
gives a nontrivial algebraic relation of xI , ... , xn over M.) Therefore,

the number t of such m1 is not more than [L: K(x1 , ... , xn)] Thus .
[M:K]<[L:K(xl,...,xn)].
(ii) In the general case, let xI , ... , xr be a transcendence base of M over
K, and apply the proof above, by taking a transcendence base of L contain-
ing xI , ... , xr and by considering K(x1, ... , xr) instead of K. Then we
see that [M: K(x1, ... , xr)] is finite. Q.E.D.
We add here the following lemma to be used later.
LEMMA 3.1.5. Let S2 be an algebraically closed field containing afield K.
Assume that aI , ... , ar, bI , ... , br c S2 and that there is a K-isomorphism
a from an algebraic extension L (c S2) of K(a1 , ... , ar) to an algebraic ex-
tension M (c 12) of K(b1 , ... , br), then or is extended to an automorphism
of S2.
PROOF. We can assume that aar = bi and that al , ... , ar are alge-
braically independent over K. Take a transcendence base A = {cjjj E Al of
12 containing al , ... , ar . Consider subsets S of A such that {b1 , ... , br}U
S is algebraically independent and {b1 , ... , br} n S is empty. Let So be a
2. TENSOR PRODUCTS OVER A FIELD 83
maximal subset among such S. Then b1 , ... , br form a transcendence base

of S2 over K (So) . By Theorem 3.1.2, we see that So U { aj, , ... , ai } = A for
some ji E A. We take a one-to-one mapping yr of A0 = A - {a1 , ... , a,.}
to So , and we extend a to an isomorphism of L(A0) to M(S0) so that
ac = yrc for all c c A0 (note that A0, So are algebraically independent over
L, M, respectively). SZ is the algebraic closure of K(A, L) ^' K(So, M)
and therefore, by the uniqueness of algebraic closures, a is extended to an
automorphism of Q. Q.E.D.
2. Tensor products over a field
Though the tensor products are usually defined on modules over a ring,
we consider in this book tensor products over fields only, for the simplicity
of our discussion.
Let K be a field and let M, N be K-modules. Then the tensor product of
M and N over K, which is denoted by M K N or by M N, is defined
as follows. Let B1 = { mi i E Al, B2 = { nj j E M} be linearly independent
bases of M, N, respectively. The tensor product is the K-module with
linearly independent base B1 x B2 = {(m,, nj) mi E B1 , nj E B21. These
elements (mi , nj.) are denoted by mi ni if they are regarded as elements
of M N N. If m = E cimi and n = E d,.nj. (finite sums, ci , di E K), then
m n is defined to be >2(ci di)(mi (& nj). Then we define the canonical
mapping 0 of M x N to M N, namely, 0(m, n) = m n.
THEOREM 3.2.1. The structure of M N as an K-module is characterized
by the property described below, and therefore, the structure is independent of
any particular choice of linearly independent bases of M, N.
If a mapping f of the product set M x N of M, N to a K-module M*
satisfies the conditions
f(m + m', n) = f(m, n) + f(m', n),
f (m , n + n') = f (m , n) + f (m , n'),
f(cm, n) = f(m, cn) = c- f(m, n) (m, m' E M, n,n'EN, c E K),
then there is a K-homomorphism g from M N onto the submodule of M*
generated by f (M x N) such that f = go with 0 the canonical mapping of
MxN to MN.
A map f such as above is called a bilinear mapping of M x N to M* .
is also a bilinear mapping.
PROOF. As is easily seen, the required g is defined by g(E cii (mi nj)) _
> cij f (mi , nj) . Consider the tensor product defined by another pair of lin-
early independent bases, we denote it by (M N)' and by 0' the canon-
ical mapping of M x N to (M (& N)'. Since 0' is bilinear, there is a K-
homomorphism g from M N to (M (& N)' such that 0' = go. Symmet-
rically, there is a K-homomorphism g' from (M 0 N)' to M N such that
0 = g'O'. Then g'g(m(&n) = g'gq5(m, n) = g'q'(m, n) = q(m, n) = mn.

Therefore g'g is the identity mapping on M N. Similarly, gg' is also
the identity mapping on (M N)'. Thus g is an isomorphism and M N
is unique up to isomorphism. Q.E.D.
THEOREM 3.2.2. Let M, N be modules over a field K. Then
(i) MKN ^_' NKM M. If N' is another K-module, then (M(&KN)KN' ^_'
MK (NK N').
(ii) If N is a ring, then MN becomes an N-module by the multiplication
Ei(mi (& ni)n = E mi (nin) (mi E M, ni, n c N).
(iii) If M and N are rings, then M N becomes a ring by the multipli-
cation (E mi (gni)(E m'(gn') = Ei j(mim') (nin') .
(iv) If M, N are submodules of K-modules M', N', respectively, then

M N is naturally embedded in M' N'.
PROOF. (i) is obvious. (ii) and (iii) are easy. As for (iv), if we take linearly
independent bases of M', N' which contain those of M, N, respectively,
then the assertion follows easily. Q.E.D.
If M, N are rings containing a field K, then the canonical mappings
Oil 02 of M, N into M N are defined by c, m = m 1 , 02n = 1 n
(m E M , n E N) . OP 02 are isomorphisms from M, N, respectively,
into M N, and M N is generated by 0 M and 2 N. Therefore, M
I
and N are often identified with 0IM and 02N, respectively. The following
assertion is clear (and we omit the proof).
THEOREM 3.2.3. If, furthermore, M, N are subrings of a ring R, then
there is a unique ring homomorphism 0 from M N to R such that O(m (9
n) = mn.
This 0 is called the natural mapping of M N to R (or, to the subring
M[N] generated by M, N).
If M, N are extension fields of a field K, then the ring of total fractions
of M K N is called the local tensor product of M and N over K and is
denoted by M xK N or simply by M x N. (Though the local tensor product
can be defined in more general way, we give here only this special definition,
because we do not treat the general case. As for the notation M x N, we
must be careful not to confuse this with the product set. There will be no
confusion if we say the ring M x N.) Assume that the fields M, N are
subfields of a field Q. If the natural mapping 0 of M N to the ring M[N]
generated by M, N in S2 is an isomorphism, then we say that M and N
are linearly disjoint over K. In this case 0 extends to an isomorphism
from M xK N to the field M(N). This extension is also called the natural
mapping.
THEOREM 3.2.4. Let K be a field, let M be a K-module, and let N be a
ring containing K. M is identified with M 1 c M K N by the canonical
2. TENSOR PRODUCTS OVER A FIELD 85
mapping. If M is also a ring, then N is identified similarly with 1 N C

M K N .
(i) If an element n of N is not a zero-divisor in N, and if fn = 0 for
f c MN, then f = 0. Therefore, if M is a ring, then n is not a zero-divisor
in the ring M N.
(ii) A linearly independent base of M over K is a linearly independent
base of M K N over the ring N. Therefore M N is a free N-module. In
particular, if N is afield and if lengthK M = d, then length,(M N) = d.
(iii) If N is a field and N' is an N-module, then M K N' -= (M K
N) ON N'.
PROOF. (i) Consider N' = N[n-'] . Then M N CM N' (Theorem
3.2.2, (iv)). If fn = 0, then 0 = =f. (fn)n-1
(ii) is obvious by the definition.

(iii) Take linearly independent bases {mi l i E Al, {nj j E B} of M, N,
respectively, over K, and a linearly independent base {dIs E C} of N'
over N. Then {njdl j E B, s E C} forms a linearly independent base of
N' over K. Therefore, there is an isomorphism such that each mi nods is
mapped to mi n., ds.
LEMMA 3.2.5. If a ring R is a finitely generated module over its subfield
K, then elements that are not zero-divisors of R are invertible.
PROOF. Assume that a is not a zero-divisor in R. , a, a2 , ... , a"
1
are linearly dependent if n > lengthK R. Therefore, there exists a natu-

ral number d such that , a, a2 , ... , ad -1 are linearly independent, but
1
1, a, ... , ad are linearly dependent. Take a linear relation coal + clad -1 +

+cd=0 (ciEK, (co, ... , cd) 54 (0,...,0)). Since are
l,a,...,ad-I
linearly independent, we see that co 54 0. If Cd = 0, then a (coad - I + +

-I
cd_ 1) = 0 . Since a is not a zero-divisor, we have coad + + cd_ = 0,I
which contradicts the linear independence of 1 , a, ... , ad -1 Thus cd 0.

Therefore, -acd 1(coad-1+ +cd_1) = 1 (cd'(coad-I+ +cd_1) E K[a] c
R). Thus a is invertible in R. Q.E.D.
THEOREM 3.2.6. Let M, N be extension fields of a field K. If N is
algebraic over K, then M K N = M x K N.
PROOF. It suffices to show that if f = >i=I mi ni is not a zero-divisor,
then f is invertible. Set NI = K(n , ... , nr) . Then f E M NI c M N N.
I
Since M NI is a finite M-module, f is invertible in M NI , by Lemma

3.2.5, and hence, is also invertible in M N. Q.E.D.
THEOREM 3.2.7. Let M, N be extension fields of a field K. Set L = K (S)
with a transcendence base S of N over K. Then
(i) M X K L is afield and is contained in M x K N,
(ii) if [N: L] = d, then M xK N is of length d as an (M xK L)-module,

and therefore, M xK N is an Artin ring in this case.
PROOF. (i) First we shall show that M x K L is a field. Every element of
K(S) is expressed as a rational form in a finite number of elements of S, it
suffices to prove the assertion in the case where S is a finite set {x, , ... , x1, } .
Since MKK[xi , ... , x11] is the polynomial ring M[x1, ... , x1,] in n vari-
ables over M (as is easily seen by taking the set of all monomials as a linearly
independent base of K[x, , ... , x,1]) , M K K(S) is a subring of the field
of fractions of M[x1 , ... , and therefore, M xK K(S) coincides with
the field M(xi , ... , xi,) . Now we shall prove the remaining assertion with
general S. Since M K N ^_' (M (&K L) L N and M L is an integral
domain, we see that nonzero elements of M L are not zero-divisors by
Theorem 3.2.4, (i). Therefore M xK N contains M x L.
(ii) We have (M (&K L) L N C (M xK L) L N c M xK N. By Theorem
3.2.6, we have (M x K L) L N = (Mx K L) X L N, which coincides with
M XK N. Therefore, we have (ii). Q.E.D.
The following theorem is helpful in seeing the relationship between the
field generated by two extension fields M, N of a field K and the local
tensor product M x N.
THEOREM 3.2.8. Assume that M, N are extension fields of a field K such
that the field M(N) generated by M, N is defined. Let 0 be the natural
mapping of M K N to M(N), and let P be the kernel of 0.
(i) Every prime ideal of M x K N is a maximal ideal.
(ii) The following two conditions are equivalent to each other.
(1) There is a maximal ideal I of M x K N such that I n (M N) = P.
(2) For every intermediate field M' between K and M such that M' is
finitely generated over K, we have
trans. degK M' = trans. degN N(M').
We say that M and N are free over K if the conditions in (ii) are
satisfied.
PROOF. (i) Assume that Q is a nonmaximal prime ideal of M xK N. Let
I be a maximal ideal containing Q. Choose an element m E M N that
is in I but not in Q. Then there is a subfield M' of M that is finitely
generated over K such that m E M' x K N. II = I f1 (M' XK N) and
QI = Q rl (M' XK N) are prime ideals of M xK N, and Ii # Qi because
m E II , a contradiction since M' x N is an Artin ring by Theorem 3.2.7.
(ii) It is obvious that (1) implies (2). Assume that (1) is not satisfied. Then
P contains an element f that is not a zero-divisor (cf. Theorem 3.6.12),
so we can choose a finitely generated subfield M' of M over K, such that
f E M'(9 N. Then f is not a zero-divisor and is in the kernel of the natural
mapping of M' N to N(M'). Therefore, we can assume that M is finitely
3. DERIVATIONS 87
generated over K. Take a transcendence base x1 , ... , xn of M over K.

Assume for a moment that P n K[xl , ... , xn] N = {0} . Let S be the set
of nonzero elements in K[xl, ... , xn] N = N[x1, ... , xn] and consider
the subring (M (9 N)s of M x K N that is generated by M N and inverses
of elements of S. Then P(M (9 n)s does not contain 1 and therefore there
is a prime ideal J of (M N)s containing P. But, since (M N)s
N(x1 , ... , xn), (M (9 N)s is an N(x1 , ... , xn)-module of finite length,
and hence, is an Artin ring. Thus, (M (9 N)S = M x N. Therefore, J D P
cannot contain any elements that are not zero-divisors, a contradiction to the
assumption that (1) is not satisfied. Therefore, PnK[x1 , ... , 0 10},
and x1 , ... , xn are algebraically dependent over N. Q.E.D.
3. Derivations
A mapping D of a field K to K is called a derivation of K if D satisfies
the following conditions:
If f , g c K, then (1) D(f+g) = Df+Dg, (2) D(fg) = f(Dg)+g(Df) .
If, furthermore, S is a subring of K with Ds = 0 for all s c S, then

we say that D is a derivation over S. The set Der(K/S) of derivations
of K over S is a K-module with operations: (D + Y) f = D f + D' f ,
(fD)g = f(Dg).
THEOREM 3.3.1. If D is a derivation of a field K, then
(i) D(- f) = -D f (f c K),
(ii) S = If E KID f = 0} is a subfield of K,
(iii) D(f/g) = (gDf - fDg)/g2 (f, g E K, g 0),
(iv) Dfn = nfn-1 D f (f E K, n is a rational integer),
(v) if the characteristic p of K is not 0, then S in (ii) contains K" _
{fPjfEK}.
PROOF. f = (f /g)g and D f = gD(f /g) + (f /g)Dg This implies (iii).
.
12 = 1 and D 1= D 1+ D 1. This implies D 1= 0 and 1 E S. Similarly,

O E S follows from 0+0=0. Since 1 + (-1) = 0 , DI + D(-1) = D0 = 0
and -1 E S. Thus D(- f) = -D f + f D(-1) = -D f . Therefore, f c
S implies -f c S. Now, we see easily that f g, f 1g c S for g 54 0,
f c S. Thus, (i) and (ii) are proved. Next, we shall prove (iv). We use
an induction on n (n > 0). The equality is obvious if n = 0, 1 Assume
.
that the equality is true for n = 0, 1 , ... , n - . Then Dfn = D(f"- ' f) _
1
fDfn-l+fn-1Df =
nfn-1Df . In the case n < 0, setting m = -n, we have
Dfn = D(1/f'n) = (fin D1 - Dfm)/ f2in = -(m fm-1Df )/f2n = nfn-1Df .
(v) follows from that D f" = p(fp-1 D f) = 0. Q.E.D.

THEOREM 3.3.2. Let I be an integral domain, and let K be its field of
fractions. A mapping d of I to K is the restriction of a derivation, say D,
of K if and only if d satisfies the conditions :
(1) d(f + g) = df +dg.
(2) d(fg) = fdg+gdf (for any f, gEI).

If d satisfies these conditions, then the derivation D is uniquely determined
by d.
PROOF. The only if part is obvious. Assume that d satisfies conditions

(1) and (2), then for f 1g = f'/g' (f, f', g, g' E I), we have g' f = f'g,
so fdg' + g'df = gdf' + f'dg, g'df - f'dg = gdf' - fdg' . Finally,
g'(df - (f'l g')dg) = g(df' - (f/g) dg'), so (df - (f/g) dg)l g - = (df4
(f'/g') d g')/g' , and therefore, D(fl g) = (gdf -fdg)/g2 does not depend
on the expression of f /g . Thus, we see that the D defined above is a
derivation of K by virtue of the conditions (1) and (2). This completes
the proof of the if part. In view of Theorem 3.3.1, (iii), D is uniquely
determined by d. Q.E.D.
On the polynomial ring K[xl , ... , xn] in n variables over a field K, we
can consider the partial derivation a/axi for each i, namely,
(a/axi)f(xl , ... , xn) is the derivative of f(x1 , ... , with respect to xi
only. These partial derivations satisfy the conditions (1), (2), and therefore,
each a/axi determines a derivation of K(x1 , ... , xn), which is denoted by
the same symbol a/axi On the other hand, if D is a derivation of the field
.
K, then the mapping of K[x1 , ... , xn] to itself, such that
in
Etl
ai,...i,, x1
. .
xn -* >( ) !'
ai1...i , 1 xl
. . t,i
. xn
defines a derivation of K(x, , ... , xn), which is the extension of D with

Dxi = 0 for every i. This derivation is usually denoted by the same symbol
D and is called the trivial extension of D. In general, if a field L contains
a field K, and if D is a derivation of K, then a derivation D' of L is
called an extension of D if its restriction to K coincides with D (such an
extension may not exist).
THEOREM 3.3.3. Assume that afield L is generated by elements a1 , ... , an

over a field K, that D is a derivation of K, and that v1 , ... , vn are ele-
ments of L. Let I be the relation ideal for (a1 , ... , an) in the polynomial
ring K[x, , ... , xn] and take a set of generators f1 , ... , fs of I (we can
have a finite set; see 6). For each f E K[x1 , ... , xn], we define Ff E
K(v1,...,vn)[x1,...,xn] by Ff=Df+En 1((a/axi)f)vi. Then we have
that there is an extension D' of D in L such that D'xi = vi (i = 1 , ... , n)
i f f o r every member fj of the set o f generators o f I, Ff (a , ... , an) = 0 1 .
PROOF. To prove the only if part, note that f (al , ... , an) = 0 implies
D' f (a , ... , an) = 0. On the other hand, we see that D' fj (a , ... , an) _
.
1
1
Ff (a1 , ... , an) , because D' is the prescribed derivation.

3. DERIVATIONS 89
To prove the if part, let f E I, and write f = > hj f (hj E K[x, , ... ,
x,]) , then
Ff = fjDhj + hjDfj + E fj((a/axl)hj)vi + hj((a/ax1) fj)vi ;
hence Ff(al, ... , an) = 0. Thus, if g(a,, ... , an) = g'(a,, ... , an) with
g, g'EK[x,,...,xn],then g - g'EI and Fg(a,,...,an)=Fg,(a, ,...I
an). Therefore, the mapping D' of K[x, , ... , xn] to L defined by
DI(g(al,...,an)) = Fg(a,, ... , an) for g(a,,...,an) E K[a,,...,an]
satisfies conditions (1) and (2) in Theorem 3.3.2, and therefore, the required
extension exists. Q.E.D.
REMARK. For elements fl , ... , f. of K(x, , ... , xn) , the (m, n)-matrix
whose (i, j) entry is (a/axj) f is called the Jacobian matrix of f , ... , f,n
and is denoted by J(f , ... , A similar definition is applied to infinitely
4n).
many elements of K(x, , ... , x,,) . On the other hand, if A is a matrix with
entries in K(x, , ... , xn) , then A(X =a) denotes the matrix obtained from A
setting xi = ai (i = 1 , ... , n). Then the condition in the theorem proved
above can be expressed as
(Df,)
J(f, ,... I fjn I ... (X =a, )
__ - I
I (X,=a,)
Therefore, we can conclude that
(i) If the rank of J(f, , ... , fin, ... )(X,-ai) is equal to n, then there exists
a unique extension of D to L.
(ii) Assume that the rank of J(f , ... , 1,,, , ... )(X;=aj) is n - r.
(1) If an extension D" of D to L exists, then the extensions of D to
L are in one-to-one correspondence with vectors of the form
c, b, , br,
+d, J+ +dr
cn b1 n brn
where ci = D"a, (i = 1, ... , n) and the r vectors (b j;) (j
, ... , r) are linearly independent solutions of
1
X1
J(f , ... , fm, ...)(X,_a,) =0.

Xn
An extension D* corresponds to a vector
if D*al = c .
(2) In particular, the module Der(K(a1 , ... , an)/K) of derivations of

K (a , ... , an) over K forms a K (a , ... , an) -module of length r .
1 1
THEOREM 3.3.4. Let D be a derivation of a field K, and let L = K(a) be

a simple extension of K.
(i) If a is transcendental over K, then, for each v c L , there is an exten-
sion D* of D to L such that D*a = v .
(ii) If a is separably algebraic over K, then there is a unique extension of

D to L.
(iii) Assume that K is of characteristic p 0 and ap c K, then there is
an extension of D to L iff D(ap) = 0. If this condition is satisfied, then, for
each v c L, there is an extension D* of D to L such that D*a = v.
PROOF. Let I be the relation ideal for a in the polynomial ring K[x].
(i) is obvious by Theorem 3.3.3, because I = 101. Consider (ii). In this
case I is a maximal ideal generated by an irreducible monic polynomial
f (x) = xn + CIx n-1 + cn . (a /ax) f (x) is nothing but the derivative
+
(d/dx)f(x) defined in Chapter II, 3. As we saw in the proof of Lemma
2.3.1, (dl dx) f (a) = H (a - a,) (where a, runs through roots of f (x) other
than a), which is not 0 . Since the condition on the extension D* is Ff(a) =
0 (Theorem 3.3.3), we see that v with D*a = v is unique, and therefore,
D* is uniquely determined. As for (iii), the only if part is obvious. Assume
that D(ap) = 0. I is generated by f = xp - ap , and Ff = 0. Therefore,
for any v , there is an extension D* such that D*a = v .Q.E.D.
THEOREM 3.3.5. If x1 , ... , xn are algebraically independent over a field
K, then the partial derivations alax1 , ... , alaxn form a linearly indepen-
dent base of Der(K(x1 , ... , xn)/K) .
PROOF. Let D E Der(K(x1 , ... , xn)/K) With u, = Dx1, we set D' =

.
D - > u.(a/ax;). Then D'xl = 0 and D'(K(x1 , ... , xn)) = 0, i.e., D' = 0 .
Thus D = > u!(a/ax;). If D = >v1(a/ax;) , then Dx; = vi and v. = u! .
Therefore a/ax1 , ... , a/axn are linearly independent. Q.E.D.

As we saw already, any derivation of a field K of characteristic p 54 0 is
a derivation over Kp. A subset S of K is said to be p-independent over a
subfield k of K, if [Kp (k) (s1 , ... , Sin): Kp (k) ] = p'n whenever s , ... , Sin
1
are mutually different elements of S. A maximal p-independent subset of

K over k is called a p-basis of K over k.
THEOREM 3.3.6. With K and k as above, we have
(i) If a subset S of K is p-independent over k, then there is a p-basis S*
of K over k which contains S.
(ii) We fix one S* as above. Let H be the set of mappings of S* to
K. Then there is a one-to-one mapping 0 of Der(K/k) to H such that
Ds = (qD)s for all s E S*.
4. SEPARABLE EXTENSIONS 91
(iii) In particular, if S* consists of n elements, then lengthK Der(K/k) =

n
PROOF. (i) is easy by Zorn's Lemma. To prove (iii), let S* = {s1 , ... , SO'
Set Lo = K (k) , Li = L0(s1 , ... , si-1 , si+1 , ... , s,,) (i = 1, ... , n). Since
[K: Li ] = p, there is a derivation Di of K over Li such that Di (si) =
1 (Theorem 3.3.4, (iii)). We shall show that D1 , ... , Dn form a linearly
independent base of Der(K/k). For D E Der(K/k), let ui be Dsi Then .
(D - >2 uiD1)(sj) = 0 for all j, and (D - F, uiD1)(L0(s1 , ... , 0.

Thus, D - E u1Di = 0, and D = F, u1D, If D = > viDi, then Ds! = vi
.
and vi = ui Therefore, D1 , ... , Dn are linearly independent.

.
To prove (ii) we define a mapping 0 of Der(K/k) to H by (qD)s = Ds

(s c S*). It suffices to show that q(Der(K/k)) = H. Let h E H, and define
a mapping D of K to K as follows: for each a c K, we choose elements
sl , ... , sin of S* such that a E Lo(s1 , ... , Sin) (Lo = Kp(k)). Then there
is a derivation D' of Lo(s1 , ... , s,n) over Lo such that D'si = hsi (i =
1, ... , m) (Theorem 3.3.4, (iii)). Then we define Da = D'a. We see that
this definition does not depend on the particular choice of si , because of the
uniqueness of the expression of a in a form of polynomials in elements of
S* with coefficients in Lo such that the degree in each element of S* is at
most p - I. By our definition, D is a derivation of K over k and h = OD.
Q.E.D.
4. Separable extensions
In this section and in the next section, (i) p will denote the characteristic
of the field of consideration if it is of nonzero characteristic and (ii) p will
denote 1 if we are observing a field of characteristic 0. Thus, for a field
K, K" denotes (1) {x c (an algebraic closure of K) I x" E K} if K-00 is
of characteristic p # 0, and (2) K itself if K is of characteristic 0. K"
denotes the smallest perfect field containing K (i.e., Un K" ) .
Let K be a subfield of a field L. We say that L is separable over K or

L is a separable extension of K if L K K" is an integral domain (hence,
a field). The following theorem is straightforward from this definition.
THEOREM 3.4.1. If L is a separable extension of a field K, then any in-
termediate field between K and L is separable over K. Any extension field
of a perfect field Ko is separable over Ko .
THEOREM 3.4.2. Assume that afield L is finitely generated over afield K.

Then the following five conditions are equivalent to each other.
(ii) L has a separating transcendence base over K.
(iii) lengthL Der(L/K) = trans. degK L.
(iv) lengthL Der(L/K) < trans. degK L.
(v) L K K" is an integral domain for every natural number r.
PROOF. To prove that (ii) implies (iii), let xI , ... , xn be a separating

transcendence base of L over K. a/axl , ... , a/axn form a linearly inde-
pendent base of Der(K(x1 , ... , xn)/K) (Theorem 3.3.5). Since L is sepa-
rably algebraic over K(x, , ... , xn) , each a/ax1 is uniquely extended to a
derivation Di of L over K, and these Dl form a linearly independent base
of Der(L/K) as we can see by the proofs of Theorem 3.3.5 and Theorem
3.3.6, (iii).
That (iii) implies (iv) is obvious. To prove that (iv) implies (ii), note that
if K is of characteristic 0, then the implication is obvious, so we assume that
p > 2. Let sl, ... , Sin be a p-base of L over K. Then D E Der(L/K) is
determined by Ds1 (i = 1, ... , m) and m = lengthL Der(L/K) (Theorem
3.3.6). Therefore Der(L/K(sI , ... , sm)) = {0} . If we show that L is sepa-
rably algebraic over K(sI , ... , Sin), then we see that sI , ... , Sin form a sepa-
rating transcendence base, because m < trans. degK L by (iv). Thus, we sup-
pose, for a moment, that L is not separably algebraic over K(s1 , ... , Sin) .
(1) In the case where L is algebraic over K(s1 , ... , Sin), there is a field M
such that K(sl , ... , Sm) c_ M c L and L is purely inseparable over M.
Then L'(M) 0 L and Der(L/M) 0 10}, which is a contradiction because
Der(L/M) c Der(L/K(s1 , ... , sm)) = 101. (2) In the remaining case, take
a transcendence base T of L over K(sI , ... , Sm) . If L is not separably
algebraic over K(s1 , ... , Sin, T), then we obtain a contradiction as in (1).
If L is separably algebraic over K(s1 , ... , Sin , T), then every element of
Der(K(sI , ... , Sin , T)/K(sI , ... , Sin)) {0} can be extended to an element
of Der(L/K(sI , ... , Sin)) (Theorem 3.3.4, (ii)), a contradiction.
To prove that (ii) implies (v), let xI , ... , xn be a separating transcen-
dence base of L over K. L is then a finitely generated separably alge-
-r (XI
braic extension of K(x1 , ... , xn) . Kp , ... , xn) is purely insepara-
ble over K(x1, ... , xn) ; hence, is normal over K(x1 , ... , xn) It follows
.
-r
that [L: K(xl, ... , xn)] = [L(Kp ): Kp (XI , ... , xn)] by Theorem 2.7.3,
_r -r
(ii). L K Kp ti L K(x, (K(xI , ... , xn) K Kp ) C L K(x1 ,...,x,1)
-r -r (X,,
Kp (xI , ... , xn) . This ring is of length [L: K(x1, ... , xn)] as a Kp
... , xn)-module. The image of this ring by the natural mapping to L(Kp )
coincides with L(Kp ) , whose length is equal to [L: K(xI , ... , xn)] as we
saw above. Therefore, the natural mapping is an isomorphism. This implies
-r -r
that L K(x, , ... , Kp (x 1, ... , xn) is a field and L K Kp is an integral
domain.
That (v) implies (i) is obvious because L K Kp c L K Kp .
That (i) implies (iii) is obvious if K is of characteristic 0, so we assume
that p > 2. Assume for a moment that p-independent elements s, , ... , sn
of L over K are algebraically dependent over K. Choose f (x1 , ... , xn )
of the smallest degree from the relation ideal I for s, , ... , sn . By the p-
4. SEPARABLE EXTENSIONS 93
independence of these si , we see that each term of f (xl , ... , xn) must be
of degree a multiple of p for every xi Thus, there is g(xl , ... , xn) E
.
KP [xl , ... , xn] such that f (xl , ... , xn) = g(xl , ... , xn)p. Let g =
E ci 1 n
xi' xn'l . Then the element b = E(si' S' ci ... E L Kp
I n
is
not 0, because K[sl, ... , sn] KP ^' KP [xl, ... , xn]/IKp [xl , ... , xn] ,
1
and g is of smaller degree than f . This b is mapped to 0 by the natural

mapping 0 of K[sl , ... , sn]Kp to L Kp ). Therefore, K s
I s)
- -1 -I
KP , which contains K[sl , ... , sn] K Kp , is not a field. L Kp con-
tains this ring, and therefore, L KP is not an integral domain. Q.E.D.
THEOREM 3.4.3. If L is a finitely generated extension of a field K, then
L KKp -00 is an Artin ring with only one maximal ideal. The length r of this
Artin ring coincides with [L: K(xl , ... , xn)]/[L(Kp -00 ): KP -00 (xl , ... , xn)]
provided that xl , ... , xn form a transcendence base of L over K. Conse-
quently, r is a power of p.
This length r is called the order of inseparability of L over K and is
denoted by i (L/K) .
PROOF. Assume that an element f of L K KP -00 is not invertible. f =

`n al bi (ai E L, bi E KP -00 ) , and therefore, there is a natural number
l
t such that b p E K for every i . Then I" >a' b,". Since b p E K ,

we have fpr = E (ap' bp') 1 E L 1 . Since f p, is not invertible, we see
that fp = 0. If P is a prime ideal, then f'p c P and f c P. Thus, P
contains all noninvertible elements, and-00
P is the unique maximal ideal. As
in the proof of Theorem 3.4.2, L K KP is identified with L K(xl , ... , X11)
(K(xl,...,xn)(&KKp ) and K(xl,...,xn)Kp
-00 -00
KP (xl,...,xn).
Therefore, L KKp contains the field KP (xl , ... , xn) and is of length
-00 -00
[L: K(xl , ... , xn)] as a KP (xl , ... , xn)-module. Since L K KP /P
00
L(Kp ), we have r = [L: K(xl , ... , xn)]/[L(Kp ) : Kp (xl , ... , xn)]
by Theorem 1.7.7. Q.E.D.
THEOREM 3.4.4. Let L be a finitely generated extension of a field K.
-n -n
(i) L is separable over K iff i(L/K) = 1 . In this case, i(L(K" )/K" )
1 (n = 1, 2, ...).
(ii) If i(L/K) > 1, then i(L(K" )/Kp < i(L/K).
(iii) If i(L/K) = pe , then L(K" ) is separable over KP
PROOF. (i) The first assertion follows from the fact that, as for an Artin
ring R, R is of length 1 if R is a field. The last assertion follows from
the fact that L has a separating transcendence base (Theorem 3.4.2).

(ii) If i(L/K) > 1 , then L Kp is not a field. Let P1 be the maximal
ideal of L Kp . Then L Kp /P1 -= L(K" ) and i (L(Kp ')/Kp ') =
1 1
0
length(L (9 Kp )/P1)
1 KP-1
Kp . On one hand, i(L/K) = length L K
00
Kp and L K Kp 00 = (L K Kp 1) KP_, Kp . On the other hand,
PI (L (& Kp-00) is the ideal generated by P1 KP_, Kp "O and is different from
101. Therefore, length L Kp > length(L (& Kp )/P1(L Kr-0)
00
length((L (9 K Kp-1 /PJ) KP-1 Kp .
(iii) We use induction on e. Since i(L(Kp )/K" < pe, we have

i(L(K" 1)/Kp-1) = pet with e1 < e. Then i(L(K" e)/K" e) =
1-(e-1) -I -1 -el -I -e,
i(L((Kp )p )/(K" )p i(L((Kp )p )/(K" )p ) = 1 . There-
fore L(Kp is separable over Kp e by (i). Q.E.D.
COROLLARY 3.4.5. If L is a finitely generated extension of a field K, then

there is a purely inseparable extension K' of finite degree over K such that
L(K') is separable over K'.
PROOF. We use induction on i(L/K). If i(L/K) = 1 , then we can set
K' = K. Assume that i(L/K) > 1 . Take a nonzero nilpotent element
En ciai (a E Kp , c. EL). Set K1 = K(a1 , ... , a ) . Then, as in the
i= 1 i t n
proof of Theorem 3.4.4, we see that i(L(K1)/K1) < i(L/K). Our induction
hypothesis is applied to L(K1) over K1 , and we have a purely inseparable
extension K' of finite degree over K1 such that L(K') is separable over K',
and this K' is the required extension. Q.E.D.
5. Regular extensions
We say that an extension L of a field K is a regular extension of K if
L K K, with K the algebraic closure of K, is an integral domain (hence, a
field). An integral domain R containing K is defined to be regular over K
if the field of fractions of R is a regular extension of K. The following is
immediate from our definition of a regular extension.
THEOREM 3.5.1. If L is a regular extension of a field K, then every inter-
mediate field between L and K is a regular extension of K. Every extension
of an algebraically closed field K is a regular extension of K.
THEOREM 3.5.2. For a subfield K of a field L, the following three condi-
tions are equivalent.
(i) L is a regular extension of K .
(ii) K is algebraically closed in L, and L is separable over K.

5. REGULAR EXTENSIONS 95
(iii) L K L' is an integral domain for every extension L' of K.
PROOF. That (iii) implies (i) is obvious by the definition.

To prove that (i) implies (ii), note that L is separable over K, because
L K Kp CL K K (since K is the algebraic closure of K) . Assume, for
a moment, that L n K # K . Then (L n K) K K is not a field (because, with
a E K n L, a K , a 1- 1(& a is mapped to 0 under the natural mapping
to L(K)) , a contradiction. Therefore L fl K = K .
To prove that (ii) implies (iii), assume that there is an extension L'
such that L K L' is not an integral domain. Take two nonzero elements
E"' 1
ai b,. , c d such that their product is zero. Let L" be
K(bl , ... , bm , d1 , ... , dn) . Then L L is not an integral domain. Thus
the set F of finitely generated extensions L' of K, such that L L' is not
an integral domain, form a nonempty set. Let Lo be a member of F hav-
ing the smallest transcendence degree over K. Similarly, there is a finitely
generated extension L1 of K, contained in L, such that L1 Lo is not an
integral domain. Since L satisfies the condition (ii), L1 satisfies the same
condition. In particular, L1 has a separating transcendence base z1 , ... , zt
over K. Let x1 , ... , xr be a transcendence base of Lo over K. Then
L1 XK Lo is a finite module over L1(xl , ... , x,.) = L1 XK K(x1 , ... , xr) .
Taking a maximal ideal P of L1 x Lo' we consider the natural homomor-
phism 0 from L1 x Lo to (L1 x LO)/P. P # 101, because L1 x Lo is
not an integral domain. Take a nonzero element f of P and express it in
the form Eu I ai b; such that a1 , ... , au are linearly independent over
K (this follows from the fact that, for c; E K, a, ai E L1 , b, b' E Lo'
we have (>c1ai)b=>ai clb, ab+ab'=a(& (b+b')). Among
such f , choose one fo which has the smallest number of terms and b1 1 .
Then, L1 x K(b2 , ... , bu) is not an integral domain, and therefore, we may
rechoose Lo to be K(b2 , ... , bu) . q(L1 x Lo) can be regarded as a field
generated by L1 and Lo .
First, assume that Lo = K(b2 , ... , bu) is not separably algebraic over K.
Then there is a nonzero derivation D E Der(L0/K). Since LI has a separat-
ing transcendence base z , ... , zt over K , z 1 , ... , zt form a separating
1
transcendence base of q5(L1 x Lo) over Lo. Therefore, D is extended to a

derivation of L1(Lo) so that Dzi = 0 for every i. Then DL1 = 0. Since
q(Ea.b.)=0,wehave Ea1b.=0 and 0=D(E`.` ab)=E aDb
>u 2 aiDbi. Note that D 0 implies that Db; # 0 for some i. Then
Ei=2 a. Db; E P, a contradiction to the smallestness of u in our choice of
fo. Thus, Lo = K(b2 , ... , bu) is separably algebraic over K. In particular,
r=0.
Let L* be a Galois extension of finite degree over K containing Lo .
Since K is algebraically closed in LI , we have LI n L* = K , and therefore,

[LI (L*) : L1 ] = [L*: K] by Theorem 2.7.3, (i). The length of L1 K L* as
an L1-module coincides with [L*: K] ; hence, the equality on the degrees of

extensions that we have just shown implies that the natural mapping of L1K
L* to L1(L*) is an isomorphism into L1(L*) , i.e., L1 K Lo c_ L1 K L*
L1(L*) , which contradicts that L1 Lo is not an integral domain. Q.E.D.
THEOREM 3.5.3. Let L I, L2 be extensions of a field K.
(i) If L1 is a regular extension of K, then L1 X K L2 is a regular extension
of L2 .
(ii) If both L1 and L2 are regular extensions of K, then L1 XK L2 is a
regular extension of K.
PROOF. (i) Let L2 be the algebraic closure of L2 . Then we have that
(L1 XK L2) XLz L2 = L1 XK L2 , which is a field by Theorem 3.5.2.
(ii) Let K be the algebraic closure of K. Then we see that (L1K L2) K
K ^' (L1 K K) K (L2 _K K) (cf. Exercise 3.2.4). Therefore L1 L K C
(LI XK K) xK (L2 xK K) , which is a field. Thus, (L1 XK L2) XK K is a
field. Q.E.D.
COROLLARY 3.5.4. Assume that L1 , L2 in Theorem 3.5.3 are free over K
(considering the field LI (L2)) .
(i) If L1 is a regular extension of K, then L1(L2) is a regular extension

of L2.
(ii) If both L1 and L2 are regular extensions of K, then L1(L2) is a
regular extension of K
The proof is immediate from the theorem above and Theorem 3.2.8.
6. Noetherian rings
If a module M over a ring R satisfies the maximum condition on sub-
modules (by a submodule we mean an R-submodule), then M is called a
Noetherian module. If a ring R itself is a Noetherian module as an R-
module, then R is called a Noetherian ring.
THEOREM 3.6.1. A module M over a ring R is a Noetherian module iff
every submodule of M is finitely generated. In particular, R is a Noetherian
ring iff every ideal of R is finitely generated.
PROOF. Assume first that M is Noetherian. Let N be an arbitrary sub-
module, and let F be the set of finitely generated submodules of N. By the
maximum condition, there is a maximal member No of F. If No N, then
there is a larger member N1 = No + nR with n c N, n V No, contradicting
the maximality of No. Thus, No = N, and N is finitely generated.
Conversely, assume that every submodule of M is finitely generated. Since
the maximum condition is equivalent to the ascending chain condition, we
shall show the validity of the ascending chain condition. If submodules Ni
are such that N1 c N2 c .. C N c , then Ui=1 N. = N* is a submodule
and is generated by a finite number of elements, say aI , ... , a, . Let n(i) be

6. NOETHERIAN RINGS 97
such that a. E for each i = 1 , ... , t . Then, with s such that s > n(i)
for every i , all ai are in N , which shows that Ns = N* . Thus, the validity
of the ascending chain condition is proved. Q.E.D.
THEOREM 3.6.2. If R is a Noetherian ring, then every finitely generated
module M over R is Noetherian.
PROOF. Let al , ... , a,, be a set of generators of M we use an induction
argument on r. In the case r = 1 , for a submodule N, we consider I =
{x E RIa1x E NJ. Then N = all. Since R is Noetherian, I is generated
by a finite number of elements, say, bl , ... , bt . Then N is generated by
albs, ... , alb, . (Remark. As an R-module, M = a1R R/{x E RIalx =
0} .)
In the general case, our induction hypothesis is that Ml =- i a R is
Noetherian. For a submodule N of M, set Ni = Ml n N. NI is finitely
generated because Ml is Noetherian. Consider the ideal J = {x E RIarx E
M1 + NJ. Let cl , ... , cs be a set of generators of J. For each c, , we
take an element ni of N of the form arc, + Er-1 aib1j (b1j c R). Then it
suffices to show that N is generated by n , ... , ns and Nl If n E N, then
1 .
n = aldl + . . + ardr (di c R) . Since dr E J , dr = > ciej (e; E R) , and

therefore, n - E n.e1 E Ml n N = Nl . Q.E.D.
THEOREM 3.6.3. If a ring S is finitely generated (as a ring) over a Noethe-
rian ring R, then S is a Noetherian ring.
PROOF. The general case follows easily from the case where S = R[x]
with a single element x, and we shall prove this case. Let I be an ideal of
R[x]. Set J = {c E RI there is an n such that cx" + +cEI
clx"-I
+
with ci E R} . J is an ideal of R, and J is generated by a finite number
of elements, say, bl , ... , br. We choose an element a; of I of the form
ai = bix"' + + c" i (cji c R) for each b; Let m be the
c11xn'-1
+ .
maximum of n1 , ... , nr , and set I' = I n (R + Rx + + Rxii-1) . Since

I' is a submodule of the R-module R + Rx + + , I' is finitely
Rx'"-1
generated as an R-module by Theorem 3.6.2. Let fl , ... , fs be a set of

generators. It suffices to show that fi , ... , fs , al , ... , a,. generate I. Let
a E I. Then a = c0x" +C Ixn-1
+ + c" (c, E R). We shall show that
a E E aiR[x] + E fiR[x] by induction on n. If n < m, then a c I'
and a is written as an R-linear combination of fl , ... , fs . Assume that
n > m and that the assertion is true of the case of degree < n - 1 co E J .
and co = > bid; (d1 E R). Set a* = a - E a1d;x"-"i. Then a* E I and

is of degree < n - 1 Therefore, a* E Ea1R[x] + > f R[x] Therefore,
. .
a c a,R[x] + > f R[x]. Thus, I = > E fR[x] . Q.E.D.

An ideal Q of a ring R is called a primary ideal if Q R and if every
zero-divisor of R is nilpotent, namely, if Q 0 R and ab c Q (a, b E R)
a 0 Q implies b" E Q for some n. A prime ideal is a primary ideal. For

an ideal I of R, {cc Roc" E I for some n} is called the radical of I and
is denoted by V7. This set is an ideal of R, as is seen directly or by the
following theorem.
THEOREM 3.6.4. If I is an ideal of a ring R, then the radical of I is the
intersection of all prime ideals containing I.
PROOF. If P is a prime ideal containing I and a c vff, then a" C I C P
for some n ; hence, a c P. Therefore, vU c n (prime ideal containing
I). Conversely, assume that a Of. Then { a" In = 1, 2, ... } = S is a
multiplicatively closed subset not meeting I. Therefore, there is an ideal
P containing I and maximal with respect to S (Theorem 1.4.2). Then
a V P D I, and P is a prime ideal, which implies a o n (prime ideal
containing I). Q.E.D.
THEOREM 3.6.5. If Q is a primary ideal of a ring R, then is a prime
ideal.
If P = vfQ--, we say that Q is a primary ideal belonging to P or that Q
is P-primary.
PROOF. If ab E and a V VIQ-, then b" E Q for some n ; hence,
b c VQ-. On the other hand, 34 R because 1 0 . Q.E.D.
REMARK. The condition on primary ideals in our definition can be stated
as follows. For an ideal Q # R, Q is primary iff ab E Q, a 0 imply
bEQ.
THEOREM 3.6.6. Let Q be an ideal of a Noetherian ring R such that
Q R. If there is no pair of ideals I, J such that Q = I n J, I # Q, and
J # Q, then Q is a primary ideal.
Before proving this, we introduce a notion. If J is an ideal of R and
S is a subset of R, then J : S denotes {x E R IxS C J} and is called the
ideal quotient of J by S, which is clearly an ideal of R and coincides with
J: ESESsR.
LEMMA 3.6.7. If I, J, and H are ideals of a ring R, then
(I: J): H = I: JH, (In J): H = (I: H)n(J: H).
PROOF. x c (I: J) : H if xH C I : J, or equivalently, xHJ C I. X E
(I n J) : H if xH C I n J. Note that xH C I and xH C J if X E
(I: H) n (J: 11). Q.E.D.
PROOF OF THEOREM 3.6.6. Assume, for a moment, that Q is not primary.
Then there are elements a, b such that ab c Q, a 0 Q, b' Q for all
n. Set I" = Q: b"R (n = 1, 2, ...). Since In+1 = (Q: b"R) : bR, we
have In C In+l and Il C I2 C_ . Since R is Noetherian, IN = IN+1 for
some N. Set I = Q + aR and J = Q + bNR R. Then I Q , JOQ,
and we shall obtain a contradiction by showing that I n J = Q. Obviously

Q c In J, and we shall show In J c Q. Let c E In J. Then c = q1 + art =
q2 + bNr2 (q, E Q, r1 E R). Then be = q1b + abr1 , and be E Q. Since
be = q2b + bN+1r2, we have r2 E IN+1 IN. This implies that bNr2 E Q ,
and c = q2 + bNr2 E Q. Thus, I n J = Q. Q.E.D.
An expression of an ideal I as the intersection J1 n n J, of ideals J1 is
called an irredundant expression if the omission of any one of these Ji does
not obtain I.
LEMMA 3.6.8. Assume that Q1 and Q2 are primary ideals of a ring R and
that I = Q1 n Q2 is an irredundant expression. Then
(i) if VQ-I = then I is a primary ideal belonging to
(ii) if then I is not a primary ideal.
PROOF. (i) x E VQI implies xn E Q1 and x'n E Q2 for some m , n .
Thus, X E Of. Therefore, = VQ-I . On the other hand, if be E I, b
vfI-, then be E Q1 , b 7 and therefore, c E Q, , so c c Q1 n Q2 = I .
(ii) We may assume that Q1 . Let b be an element of Q1 that
is not in Let c be an element of Q2 that is not in I. Then be E

.
Q1 Q2 c Q1 n Q2 C I. But c is not in I and b is not in v/J, because b is

not in . Q.E.D.
Assume that an ideal I has an irredundant expression Q1 n ... n Qn with a
finite number of primary ideals Q1 . The above result shows that if there are
Q1, Qj such that = A , then these primary ideals can be replaced
by the intersection of them. Thus, (making n smaller) we can have an
expression such that VQj for i j. An irredundant expression
satisfying this condition is called the shortest expression.
LEMMA 3.6.9. Let I be an ideal, and let P1 , ... , Pn be prime ideals of a
ring R. If I is not contained in any of the Pi, then there is an element a of
I that is not in any of these P1.
PROOF. We use an induction argument on the number n of prime ideals.
The assertion is obvious if n = 1 , and we consider the case n > 1 If .
P c Pn for some i < n , then we can disregard P1 , and we have the case of
n - 1 prime ideals. So, we assume that every P1, for i < n , is not contained
in Pn. By our induction hypothesis, there is an element a1 of I which is
not in any of the P , i < n. If a1 is not in Pn , then we may set a = a1 .
Assume that a E Pn P. If I P1 . . Pn-1 c Pn, then P1 c Pn for some i < n,
1
.
because Pn is a prime ideal, contradicting our assumption. Thus, there is an

element b of IP1 . . Pn_1 that is not in Pn . Then a = a1 +b is the required
.
element. Q.E.D.
THEOREM 3.6.10. Every ideal I in a Noetherian ring R can be written
as the intersection of a finite number of primary ideals. Here, we regard R
as the intersection of primary ideals of number 0. If I = Q1 n . n Qn is

an irredundant expression with primary ideals Qi, then the set { /I i =
1 , ... , n} is uniquely determined by I I.
Each of the VQ_j is called a prime divisor or an associated prime ideal of
I.
PROOF. To prove the existence part, let F be the set of ideals I of R
which are not expressible as intersections of finitely many primary ideals.
It suffices to show that F is empty. Suppose, for a moment, that F is
not empty. By the maximum condition, F has a maximal member, say J.
If there are ideals J1 , J2 such that J = J1 n J2 (Ji j4 J) , then by the
maximality of J, each Ji is the intersection of a finite number of primary
ideals. Consequently, J is also the intersection of finitely many primary
ideals, and this contradicts J E F. Thus, there are no such Ji , and J is a
primary ideal by Theorem 3.6.6. This also contradicts J E F. Thus, F is
empty.
To prove the uniqueness of { /} , assume that I = Q; n n Q;,1 is
another irredundant expression with primary ideals Q'. We prove the as-
sertion by induction on max{m, n}. Let P be a maximal member in the
set of NlQi and FQJI. (i = 1, ... , n ; j = 1 , ... , m). We can assume that
VIQI = P and that V'Qi = P = Q if i < s, j < t (respectively) (s > 1 ,
but t may be 1). By Lemma 3.6.9, there is an element a of P which is
not in any VQ_j (i > s) and Q (j > t). Then there is a power aN of
a such that aN is in all Qi (i < s) , Qj' (j < t) . By Lemma 3.6.7, we
have I: aNR = n(Qi : aNR) = n(Ql: aNR) On the other hand, for i < s,
.
N
Qi : a R = R because aN E Qi. Similarly, Q : aNR=R for j < t For .
i > s, aN is not in VQi and Qi : aNR = Qi. Similarly, Qi .: aNR = Qj for

j > t Thus I: aNR = ni>s Qi = ni>r Qj Since s > 1 I: aNR 54 1, and
. . ,
therefore, t > 1 , i.e., P appears among FQJJI Our induction hypothesis

is valid for I : a N R , and we have { /I i > s} _ { FQJ I j > t} . Thus, the
uniqueness is proved. Q.E.D.
Let I be an ideal of a Noetherian ring R (I R) Then minimal mem-
.
bers in the set of prime divisors of I are called minimal prime divisors or
minimal associated prime ideals of I.
THEOREM 3.6.11. Let I be an ideal of a Noetherian ring R (I R). If
a prime ideal P contains 1, then P contains a minimal prime divisor of I.
Consequently, a minimal prime divisor is characterized by the property that it
is minimal in the set of prime ideals containing I.
PROOF. I = Q1 n . . n Qn with primary ideals Qi. Then Q1 ... Qn C
.
I C P,which implies that for some Qi C P and c P. Therefore, P

contains a minimal prime divisor of I. Q.E.D.

We add here some results on rings of fractions. When a subset S of a
ring A consists only of non-zero-divisors, then the subring of the total ring of
fractions of A generated by {s-1 Is E S} over A is called the ring of fractions
of A with respect to S and is denoted by As . If S is the complement of
a prime ideal P, then As is often denoted by AP A.
REMARK. Rings of fractions are defined in a more general manner. But,
since we do not treat the general case in this book, we have restricted our
definition.
THEOREM 3.6.12. With A and S as above, we have the following.
(i) If I' is an ideal of As , then (I' n A)A5 = I' .
(ii) Assume that P is a prime ideal of A and Q is P-primary. Then
(1) if P n S is not empty, then QAs = PAS = As ;
(2) if P n S is empty, then PAS is a prime ideal of As and QAs
is PAS-primary. Furthermore,
QAsnA=Q, PASnA=P.
PROOF. (i) Let S* be the set of products of elements of S. Every element
of S* is invertible in As . If f E I' then there is an s E S* such that
s f c I' n A ; hence, f = (s f )s- E (I' n A)AS . Thus, I' c (I' n A)AS . The
I
converse inclusion is obvious, and we have proved (i).

(ii) (1) For s c P n S, there is a t with s' E Q, hence, 1 = s's-` E QAs
and QAs = As .
(2) First, we consider QAsnA . If f c QAsnA , then f = > crgr , c; E As,
qr c Q. Choose sr E S* such that crsr E A. Considering rj sr instead of
s! , we can assume that the sr are the same s E S* . s f = >(crs)gr E Q Q.
s is not in P because s is the product of certain elements which are not
in P. Therefore s f c Q implies f c Q. Thus QAs n A C Q; hence,
QAs n A = Q. As a special case, we see that PAS n A = P. Next, we
assume that ab c QAs , a V QAs (a, b E AS). Choose s E S* such that
sa, sb E A. (sa) (sb) E QAs n A = Q, sa V Q, and therefore, there is a
natural number n such that (sb)" E Q. Then b" E QAs and QAs is a
primary ideal. The same argument can be applied to PAS with n = 1 , and
we see that PAS is a prime ideal. If c c PAS , then c = ps-1 (p c P,
s E S*). Then p" E Q, for some n, and we have c" E QAS and QAs is
PAS-primary. Q.E.D.
COROLLARY 3.6.13. Under the circumstances, if an ideal is written as I =
Q1 n . . . n Qm with primary ideals Qr, then IAs = Q1 As n . . n Q,nAS .
. If
the expression Q1 n . . . n Q,n is irredundant and if QlAs 54 As iff i < r, then
IAs = Q1 As n n QrAS is an irredundant expression.
PROOF. I CQr implies IAs c QrAs , and therefore, IAs c nr QrAs
As for the converse, let f E nr QrAs . We can assume that QIAs 54 As if
i < r. For each i > r, we choose si c S* n Qi Let s E S* be such that

.
s f E A. Then, for each i < r, we have s f E Q,As n A = Qi. Therefore,

sr+1 srns f E Qi (i = 1, .. , m) and s,.+1 sms f E I. This shows that
f E IAs and we have proved IAs = ni QiAs Now, we consider the irre-
.
dundant case. For ij < r, we have A n (Qil As n . . n Q1,As) = ni Qij , and

.
therefore, Q1As n . n QrAs is irredundant. Q.E.D.

7. Integral extensions and prime ideals
When we defined the notion "integral" in Chapter II, 1, we did not use
the notion substantially, because it is the same as "algebraic" in the case
of over a field and because we need only such a case in Chapter II. In this
section, however, we shall study the notion further. Though, in Chapter II,
we defined the notion in the case of integral domains, we give here a more
general definition.
Let S be a subring of a ring R, having the same identity as R. Then an
element a E R is said to be integral over S if for some elements ci E S,
an + clan-1 + . + cn = 0. If every element of R is integral over S, then
. .
we say that R is integral over S.

For a better understanding of this notion, we use the following fact about
a system of linear homogeneous equations. (We assume readers are familiar
with basic facts about determinants.)
LEMMA 3.7.1. Let R be a ring. Assume that a system of linear equations
Cilal = 0 (i = 1, ... , n) holds for elements ai , ... , an of an R-
module M. Let D be the determinant of the coefficient matrix (c,3) . Then
Dai=0 (i=1,...,n).
PROOF. Let Di.i be the (i, j) cofactor of the det(c13) . Then it is well
known that En i cijD1 = D , E,' 1 c.JDik = 0 (j # k). Therefore, 0 =
Ei(> cijaj)Dik = Dak . [Another expression: Let D' be the cofactor matrix
of (ci,.). Then D' (cij) = DEn (En is the unit matrix of degree n). The
system of linear equations means that (cij)`(ai , ... , a,,) ='(0, ... , 0) and
therefore '(Da, , ... , Dan) = D'(cij)`(a, , ... , an) = `(0, ... , 0).] Q.E.D.
THEOREM 3.7.2. With S and R as above, an element a of R is integral
over S if and only if there is a ring T, containing S[a], which is a finitely
generated S-module.
PROOF. If a is integral over S, then a" + ci an-1 + + c = 0 for some
ci c S. Then S[a] = S + Sa + ... + Sa"-' , and therefore, we may take
S[a] as T. Conversely, assume that there exists a ring T = Sbi + ' ' ' + Sb
that contains S[a] . Since abi E T, we have abi = E;_ c,J bJ (c,3 E S)
-
for i = 1 , ... , n . Transposing the terms on the right-hand side, we have

a system of homogeneous linear equations in b, , ... , b with coefficient
7. INTEGRAL EXTENSIONS AND PRIME IDEALS 103
matrix (adij-c,j) , where alb is the Kronecker o function (i.e., al, = 1 , (5ij =
0 if i # j). Lemma 3.7.1 shows that Dbi = 0 (i = 1, ... , n) with D =
det(a8, - c.1) . Thus, DT = 0 and D = 0. If we expand D, then we
have D = an + e1 an-1 + + en with e1 E S. Therefore, a is integral over
S. Q.E.D.
COROLLARY 3.7.3. If S is a subring of a ring R sharing the identity, then
the set S' of elements of R which are integral over S forms a ring.
This S' is called the integral closure of S in R. If S' = S, we say that
S is integrally closed in R.
PROOF. If a, b E S', then S[a] is a finitely generated S-module and
S[a, b] is a finitely generated S[a]-module, which implies that the ring
S[a, b] is a finitely generated S-module. Therefore, every element of S[a, b]
is integral over S (Theorem 3.7.2). In particular, a + b and ab are integral
over S, hence, they are in S'. Q.E.D.
COROLLARY 3.7.4. If a ring R' is integral over a ring S and if a ring R"
contains R' and is integral over R', then R" is integral over S. Conse-
quently, S' in Corollary 3.7.3 is integrally closed in R.
PROOF. For a E R", there are ci E R' such that an + c1 an-' +... + c,t = 0.
Then the ring S[c1 , ... , cn , a] is a finitely generated S-module (cf. the
proof of Corollary 3.7.3). Therefore a is integral over S. Q.E.D.
If an integral domain R is integrally closed in its filed of fractions, we
say that R is an integrally closed integral domain or a normal ring. For a
ring R, the integral closure in its ring of total fractions is simply called the
integral closure of R.
LEMMA 3.7.5. Assume that I CK are integral domains such that K is
integral over I. If one of I, K is afield, then the other is also a field.
PROOF. Assume I is a field. If 0 54 a E K, then there c; E I such that
an-1
an +C1 + +Cn Since K is an integral domain, we may assume that
=0.
cn # 0. Then cn E I and a-1 = -cn 1(an-1+c1 an-2+ +ci_1) E I[a] C K.

1
Therefore, K is a field.
Assume K is a field. Let(a-1)n
a be a nonzero
(a-1n-1
element of I. a-' E K and
there are c, E I such that + c1 + + cn = 0. By multiplying
by an-1 , we have a-' _ -(c1 + c2a + + cnai-1) E I. Therefore, I is a
field. Q.E.D.
LEMMA 3.7.6. If a ring R is integral over its subring S and if 0 is a
homomorphism of R to some ring, then OR is integral over OS.
PROOF. If a E R, then there are ct , such that an + c1 ai-1 + + cn = 0,
and we have
+ ... +
(Oci)(0a)"-1
0. Q.E.D.
(Oa)n +
Let S be a subring of a ring R. We say that an ideal I of R lies over

an ideal J of S if I n S = J.
THEOREM 3.7.7. Let S C R be rings such that R is integral over S. Then,
for a prime ideal P of S , it holds that P' is a prime ideal of R lying over
P iff P' is a maximal ideal with respect to the complement S - P in R.
PROOF. To prove the only if part, note that since P' n S = P, we have
P' n (S - P) = the empty set. Therefore, there is a maximal ideal Q' with
respect to S - P that contains P'. Note that R = R/P' is integral over
S = SIP. Let S* be the set of nonzero elements of 9. Then Q = Q'/P'
is a maximal ideal with respect to S* . Set K = RS. and I = Ss.. Then
K is generated by W over I ; hence; is integral over I, which is the field of
fractions of 3. Therefore, K is also a field by the lemma above. On the
other hand, QK is a prime ideal of K by Theorem 3.6.12. Then QK = {0} ,
because K is a field. Thus, Q = {0} , and Q' = P', i.e., P' is maximal with
respect to S* .
To prove the if part, assume that P' is maximal with respect to S - P.
Set Q = P' n S. Since Q n (S - P) is empty, we have Q c P. Further,
R = R/P' is integral over S = S/Q, and P = P/Q is a prime ideal of 9.
Set S* _ (S - 75). Then the zero ideal in R is maximal with respect to
S* Hence, K = Ws* is a field, which is integral over I = Ss.. Thus, I is
.
also a field by the lemma above. This implies that 101, i.e., P = Q.
Therefore, P' lies over P. Q.E.D.
COROLLARY 3.7.8 (lying-over theorem). If a ring R is integral over its
subring S, then
(i) For each prime ideal P of S, there is a prime ideal of R lying over P.
(ii) If P1 and P2 are prime ideals of R such that P1 c P2, then
P1nSCP2nS.
The proof is straightforward by the theorem above.
COROLLARY 3.7.9 (going-up theorem). Assume that a ring R is integral
over its subring S. If P1 c P2 c c P, is an ascending chain of prime
ideals in S, then there is an ascending chain P1 c PP c c Pn of prime
ideals in R such that P,' n S = P1. (i = I, ... , n).
PROOF. We use an argument by induction on n . The case n = 1 is
covered in the corollary above. Assume that P1 , ... , P' satisfying the
requirement, exist. Then R = R/P'_ is integral over S = SIP,,_, and
1
P = Pn/P _ 1 is a prime ideal of S. Therefore, there is a prime ideal P' of

R that lies over P. Take the inverse image Pn of P in R (i.e., the ideal
P of R containing Psuch that Pn/Pn_1 = P') . Then Pn is a prime
ideal, and Pn n S = P because P' n S = P Q.E.D.
.
7. INTEGRAL EXTENSIONS AND PRIME IDEALS 105
Here, we shall define the notions of the height of an ideal and the Krull
dimension of a ring. For a prime ideal P of a ring R, we consider descending
chains of prime ideals that begin with P. If the chain of length n : P = P0 D
P1 D D Pn is the longest among them, then the height of P is defined to
be n ; otherwise, we say that the height of P is oo. For an ideal I (54 R),
the minimum of the heights of prime ideals containing. I is called the height
of I and is denoted by ht I. The Krull dimension of a ring R, denoted by
Krull dim R , is defined to be (i) the maximum of the heights of prime ideals
if it exists, (ii) 00 otherwise. Then we have the following corollary.
COROLLARY 3.7.10. If a ring R is integral over its subring S, then
KrulldimR = KrulldimS.
PROOF. Krull dim R > Krull dim S follows from Corollary 3.7.9. Con-
versely, if P' D Pi D . D P' is a descending chain of prime ideals in R,
then (PonS) D (Pi nS) D D (PnS) is a descending chain of prime ideals
in S (Corollary 3.7.8, (ii)). Therefore, Krull dim R < Krull dim S. Q.E.D.
Corollary 3.7.9 does not guarantee that if we first specify a prime ideal P"
that lies over Pn , then there is a chain P1 c c Pn such that Pn = P" .
However, the existence of such a chain is assured if R and S are integral

domains and if S is integrally closed. For the proof of this fact, we need the
following theorem, which is important by itself.
THEOREM 3.7.11. Let R D S be integrally closed integral domains with
fields of fractions L and K, respectively. Suppose that R is integral over S
and that L is a normal extension of K. Let G be the Galois group G(L/K).
Then
(i) G is naturally identified with the set of S-automorphisms of R ;
(ii) for some prime ideals P1 , P2 of R,
P1nS=P2nS iff P2=aP1 for some or E G.
Under the circumstances, we say that R is a normal extension of S with
Galois group G, which is denoted by G(R/S) . If furthermore L is separable
over K, then R is called a Galois extension of S.
PROOF. Note that R is the integral closure of S in L and that conjugates
of an element that is integral over S are integral over S. Thus aR c_ R for
all a c G, and aR = R, because a- I E G. Elements of L are ratios
of elements of R, and therefore, a = z if the restrictions a R , ZIR of
a, r (E G) coincide with each other, which proves (i).
(ii). The if part is obvious. To prove the only if part, first we consider
the case where #(G) is finite. Assume, for a moment, that P2 aP1 for
all Or E G. Since {aP1 I a E G} consists of a finite number of prime ideals,
there is an element a of P2 which is not in any QP1 Then as P1 for
.
all a E G. Set b = [LEG aa. Then ab = b, for all or c G, and therefore,

b is purely inseparable over K. Thus, there is a natural number e such that
be c K (e = 1 if K is of characteristic 0; e is a power of the characteristic

p in the nonzero characteristic case). Then be P1 , hence P2 n S # P1 n S ,
because be E P2 n S. This contradicts the condition P1 n S = P2 n S, and
this case is proved.
For the general case, assume that P1 n S = P2 n S. We consider the set F
of pairs (L., a,) of a normal extension LA of K and aA E G(L/K) such
that (P2 n LA) = aA(P1 n Li) . (Note that R. = R n L2 is the integral closure
of S in L2 and P n LA = Pi n RA .) We define an order in F by setting
(LA , a2) < (Lu a1,) if LA c L,, and aJLA = aU . Then we easily see that
)
F is an inductive set. Consequently, there is a maximal member (L*, a*)

of F by Zorn's Lemma. Assume, for a moment, that L* L. Take an
element a of L which is not in L* . Since there are only a finite number of
conjugates of a over K, the smallest normal extension L' of K containing
L* (a) is of finite degree over L * (a) . On the other hand, a* is extended to
an isomorphism of L' into the algebraic closure of K. Since L' is a normal
extension of K, the extension of a* is an element of G(L'/K). We denote
the extension by the same symbol a* Set P' = P' n L', P1* = P1 n L
. -
P2 = Q*P*, namely, a*Pi n L* = P2 = P2 n L* . Since G(L'/L*) is a finite

group, there is a Q' E G L' L* such that a Q*P' P' Set a = QQ*
Then a"Pi = PP and (L', a') E F, a contradiction to the maximality of
(L* , 17*). Therefore L* = L and P2 = a*P1. Q.E.D.
THEOREM 3.7.12 (going-down theorem). Assume that an integral domain
R is integral over its subring S and that S is an integrally closed integral
domain. If there are given an ascending chain P1 c P2 c c P,, of prime
ideals in S and a prime ideal P' of R lying over P then there is a chain
,
Pi C PP C c P = P' of prime ideals in R such that P' n S = P

(i=1,...,n).
PROOF. Consider an integral domain R', integral over R, and a prime
ideal P* of R' lying over P'. If we can show the existence of similar chain
of prime ideals, say P1 c P2 C ... C P,i = P , in R', then P, = P.* n R
(i = 1 , ... , n) are the required prime ideals. Therefore, we may assume
that R is a normal extension of S. On the other hand, there is a chain
Pi' C P2' c c Pn of prime ideals in R such that P," n S = P, (i =
1, ... , n) by Corollary 3.7.9. The theorem above shows that there is an
element or E G(R/S), Pn' = QPn . Then P. = aPi" (i = 1 , ... , n) are the
required prime ideals. Q.E.D.
COROLLARY 3.7.13. Assume that an integral domain R is integral over its
subring S and that S is integrally closed (in its field of fractions). Then for
every ideal I' (0 R) of R, we have
ht I' = ht(I' n S) .
PROOF. For the case where I' is prime, if I' = P' D P P is a

8. THE NORMALIZATION THEOREM FOR POLYNOMIAL RINGS 107
chain of prime ideals in R, then I' n S = P n S D Pi n S P,, n S

is a chain of prime ideals in S. Therefore, we have ht I' < ht(I' n S) .
Conversely, if I' n S = P0 P1 D Pn is a chain of prime ideals
in S, then the theorem above shows that there is a chain of prime ideals
in R such that P,"nS=P1 (i=0,...,m),and
ht I' > ht(I' n S) . Thus, ht I' = ht(I' n S) in this case.
In the general case ht I' = mine' prime D1' {ht P'} = mine, prime D1' {ht(P' n
S} > minP prime D1ns{ht P} = ht(I' n S) . On the other hand, if P is a prime
ideal of S containing I' n S, then by applying Corollary 3.7.8, (i) to R/I'
and S/(I' n S) , we see the existence of a prime ideal P' of R containing I'
and lying over P. Therefore, we see the equality ht I' = ht(I' n S) . Q.E.D.
8. The normalization theorem for polynomial rings
Here we consider a polynomial ring K[XI , ... , Xn] in n variables X.
over a field K. We denote this ring simply by K[X] unless there could be
some misunderstanding. We use similar abbreviations for other symbols.
LEMMA 3.8.1. If f E K[X] = K[X, , ... , Xn], fK, then there are
elements Y1 = f , Y2, ... , Yn E K[X] such that K[X] is integral over
K[Y] = K[Y1 , ... , Yn]. Furthermore,
(i) Yj (i = 2, ... , n) can be chosen to be of the form Yj = X, +X1ni where,
if q is a given natural number, then m1 can be taken from the multiples of
q
(ii) If #(K) is infinite, then Yj (i > 2) can be chosen to be of the form
Y1 = XI + c1X1 (c1 E K).
PROOF. (i) f can be written as E a1M, (each M, is a monomial and
0 54 al E K). Take a multiple t of q such that t > degf = d. Set
m, = ti- I and Y = X, + X,n' for i = 2, ... , n. To consider the expansion
of f as a polynomial in X1 , Y2, ... , Yn , we introduce the weight of a mono-
mial; for each monomial M = Xe, Xe'. , we define its weight w(M) to be
E1 el t'- I . Then, for two monomials M1 = X d' ... Xnd ' , M2 = X i' ... Xe'+
of degree at most d , we see that if (dn , ... , d 1) > (e,,, ... , e1) under
lexicographic order, then w (M,) > w(M) Therefore, there is a unique
.
monomial, say M, , with the largest weight among monomials appearing in

f. For each monomial M appearing in f, its expansion as a polynomial
in XI, Y2 , ... , Yn is of the form X1 '(M) + (terms in X1 , Y2, ... , Y of
lower degree in X,). Consequently, the expansion of f as a polynomial in
w(MI)
X1 , Y2, ... , Yn is of the form aX 1
+ (terms in X, , Y2, . . . , Yn of lower
degree in X1) (where a is the coefficient of Mt in f) Therefore, X1 is a
.
root of a polynomial over K[Y] of the form X"(M') - (terms of lower degree
-
in X) a-I Yj . Thus, X1 is integral over K[Y], and K[X] is integral over
K[ Y] because Xl E K[YI , ... , Y , X1] (i > 2).
(ii) Consider Y = Xi + c,X1 (i > 2). By the proof of (i), we see that it
suffices to determine c1 so that, in the expansion of f as a polynomial in
XI , Y2, ... , Yn , the term Xd (d = deg f) appears with nonzero coefficient.
This coefficient is equal to fd(1 , -c2, ... , -cn) where fd(XI , ... , Xn) is
the degree d part of f . Since fd is a nonzero homogeneous form, fd (I, X2,
... , Xn) is a nonzero polynomial, and therefore, we see the existence of
c2 , ... , cn as required. Q.E.D.
THEOREM 3.8.2 (normalization theorem for a polynomial ring). If I is an
ideal of the polynomial ring K[X] = K[XI , ... , Xn] and if htl = r, then
there are elements YI , ... , Yn of K[X] satisfying the following conditions.
Here n denotes the prime subfield.
(1) K[X] is integral over K[Y] = K[YI, ... , Y ] .
(2) I n K[Y] is generated by Y1 , ... , Y .
(3) Y+j = X +t + f with f E 7r n [XI , ... , Xr] (if the characteristic p 0,
then f En[Xf,...,Xp]) for i=1,...,n-r.
PROOF. We use an induction argument on r. If r = 0, then I = 101,
and we can set Y = Xj. Assume r > 1 Choose an ideal I' CI such that
.
ht I' = r - 1 (for instance, I' = I n P with a prime ideal P of height r - 1) .

Then, by the induction hypothesis, there are Yi , ... , Yn satisfying (1), (2),
(3) with respect to I'. Hence I' n K[Y'] Yr K[Y'] C_ In K[Y']. We
have ht(I' n K[Y']) = r - 1 and ht(I n K[Y']) = r by Corollary 3.7.13 (cf.
Exercise 3.7.7). Therefore, there is an element f (Y,', ... , Y,) of I n K[Y']
that is not in I' n K[Y']. We can take f(0, ... , 0, Yr, ... , Yn) instead
of f , because I' n K[Y'] _ E =1I Yr K[Y']. Thus, we may assume that
f E K[Yr, ... , Yn]. We apply Lemma 3.8.1 to these f and K[Y' , ... , Yn]
and we have elements f = Y , Yr+ , ... , 1' of K[Yr , ... , Yn ] such that
1
K[Yr ' Y']

n is integral over K[Y r", Yn"] and Y." = Y.' i
+ Yr"n' for
i > r ; if the characteristic is p 0 0, we choose mi to be multiples of p.
r , ... , Yn-
Set Y1 = YI' , ... , Y_1 = Y'r- I , Yr = Y" = Y"'
n
Obviously, these Yi
satisfy (3). K[Y'] is integral over K[Y], and K[X] is integral over K[Y'] ;
hence, K[X] is integral over K[Y]. I n K[Y] D >=I YK[Y], and it is
easy to see that Er1 YK[Y] is a prime ideal of height > r. Thus, we have
(2). Q.E.D.
REMARK. In the above, ht =1 Y-K[Y] = r can be shown by using the
properties of polynomial rings.
COROLLARY 3.8.3 (normalization theorem for a finitely generated ring;
normalization theorem of Noether). If a ring A is generated by elements
a1 , ... , an over a field K, then there are elements z1 , ... , zl of A such
that
(1) A is integral over K[z] = K[z1, ... , z1] and
(2) z1 , ... , z1 are algebraically independent over K.
8. THE NORMALIZATION THEOREM FOR POLYNOMIAL RINGS 109
PROOF. Let I be the relation ideal for a1 , ... , an in

K[X]=K[X1,...,Xn].
Then we obtain YI , ... , Y by applying the theorem above to these I,
K[X], and r = htl. Let q5 be the natural homomorphism of K[X] to
A and set zi = 0Y +1 (j = 1 , ... , t ; t = n - r). Since K[X] is integral
over K[Y], A = K[a] = O(K[X]) is integral over K[z] = (K[Y]). If there
is a relation > c1 z z<< = 0, then f = > ci ! Yr+1 Y`' is an element
of I , which must be 0 , because I n K[Y] _ >r-1 Y.K[Y]. Thus z1 , ... , zr
are algebraically independent over K. Q.E.D.
THEOREM 3.8.4. Assume that an integral domain A is generated by a1 ,
... , am over afield K. If {0} = P0 c P1 c ... c P. is an ascending chain
of prime ideals such that there is no distinct prime ideal between Pt and Pj+1
for every i and such that Pt is a maximal ideal. Then t = trans. degK A.
PROOF. We use an induction argument on t. There are algebraically in-
dependent elements z , ... , zn of A over K such that A is integral over
1
K[z] = K[z1 , ... , Zn] (Corollary 3.8.3). If t = 0, then A is a field and

K[z] is also a field by Lemma 3.7.5, which implies that n = 0. Thus,
the assertion is true in this case. Consider the case t > 0. By applying
Lemma 3.8.1 to K[z] and PI n [z], we can assume that zl E P1 . Then
we have t - 1 = n - 1 by applying the induction hypothesis to A/PI and
K[z]/z1K[z] ^' K[z2 , ... , zn] . Thus, we have t = n. Q.E.D.
COROLLARY 3.8.5. Assume that A is an integral domain generated by a
finite number of elements over a field K. If P is a prime ideal of A, then
trans. degK A = trans. degK (A/P) + ht P. In particular,
(i) if M is a maximal ideal of A, then AIM is algebraic over K ;
(ii) if a descending chain of prime ideals begins with P and ends at {0}
and if the chain of prime ideals has no refinement, then the length of the chain
coincides with ht P.
PROOF. Consider a chain of prime ideals, as in the theorem above, which
goes through P. Then we see the assertion by the theorem. Q.E.D.
COROLLARY 3.8.6. Let A be a ring generated by a finite number of elements
over a field K. If I is an ideal of A, then the intersection D of maximal
ideals containing I coincides with the radical of I.
PROOF. The inclusion vfI- c D follows from Theorem 3.6.4. For the proof
of D C_ V7, we can assume that I = VrI_ = {0} by considering A/V7.
(i) The case where I is a prime ideal. Let f be an arbitrary nonzero
-1
element of A. Let M' be a maximal ideal of A[ f ] . By Corollary 3.8.5,
A[ f - I JIM' is algebraic over K, which implies that M' n A is a maximal
ideal of A and does not contain f . Thus f V D and we have proved
D={0}.
(ii) The general case. We can apply (i) for each prime ideal P of A and
we see that the intersection of maximal ideals containing P coincides with
P, hence D C P. Thus, D c nP = {0} = 10}. Q.E.D.
THEOREM 3.8.7. Let A be a ring generated by a finite number of elements
over a field K. If an ideal I (0 A) of A is generated r elements and if P
is a minimal prime divisor of I, then ht P < r. Consequently, ht I < r.
REMARK. This assertion holds under a weaker condition that A is a
Noetherian ring. For the proof, we need some results on so-called local rings
and we restrict ourselves to this special case.
PROOF. We use an induction argument on r. Assume first that r = 1 .
it suffices to show that htP/Po < 1 for any prime ideal P0 (c P). Thus,
considering A/Po , we can assume that A is an integral domain. Let I be f A
and let Q1, ... , QS be the minimal prime divisors of I other than P. For
each Q i , we take qi E Qi , which is not in P. Set A 1- A [ql , ... , q-1 ] .
S
Then PAI is a prime ideal by Theorem 3.6.12. If there is a prime ideal P' of
Al such that f E P' c PAI , then f c P' n A c P. By the minimality of P,
PAI is a minimal prime divisor of fAI . Similarly, if Q' is a minimal prime
divisor of fAI , then Q' n A is a minimal prime divisor of f A . Thus PA I
is a unique minimal prime divisor of fAI . Then, considering A instead of

1
A, we may assume that P is a unique minimal prime divisor of f A . Next,

take elements z , ... , z1 of A such that they are algebraically independent
1
over K and A is integral over K[z] = K[z1 , ... , z1] (Corollary 3.8.3).
Take the least normal extension L of K(z) containing A, and let A* be
the integral closure of A in L. Let P* be a prime ideal of A* lying over
P (Theorem 3.7.7). We see that P* is a minimal prime divisor of fA* by
Corollary 3.7.8. Let q be the inseparable factor of the degree of the extension
L over K(z), and set G = G(L/K(z)) (= G(A*/K[z]), fo = (l] EC C P )"
Since K[z] is an integrally closed integral domain (cf. Exercise 3.7.7), we see
that fo c K(z) n A* = K[z]. Let Q be a minimal prime divisor of foK[z],
and let Q* be a prime ideal of A* lying over Q. Then fo E Q* implies
that c f c Q* for some a E G and f c a-1 Q* . Since the minimal prime
divisors of fA* are conjugates of P* (Theorem 3.7.11), we can assume that
P* c a_I Q* Then we have htP* < hta_IQ* = htQ* = ht(Q* n K[z]) =
.
ht Q. Now, by applying Lemma 3.8.1 to fo and K[z], we obtain K[y] with

y1 = fo. If ht Q > 1 , then Q n K[y] y1K[y] , which implies the existence
of a prime ideal Q1 of K[z] lying over y1K[y] and contained in Q. Then
fo = y1 E Q1 . But this contradicts the assumption that Q is a minimal prime
divisor of f0K[z] . Thus, ht Q = 1 . (Note that this part can be proved as is
stated in Exercise 3.8.2.) Therefore ht P = 1 .
Now, we consider the general case. Let I be Er=I JA. Take a minimal
X:r- I
prime divisor P' of I' = i=If A contained in P. By our induction
hypothesis, we have ht P' < r-1 . In the ring A/P' , P/P' is a minimal prime
9. INTEGRAL CLOSURES 1I1
divisor of the principal ideal generated by (fr mod P') and ht P/P' < 1
Therefore, ht P < r by Corollary 3.8.5, (ii). Q.E.D.
9. Integral closures
LEMMA 3.9.1. Let R be a Noetherian ring, and let Q be the ring of total
fractions of R. An R-submodule M of Q is finitely generated if there is a
non-zero divisor d of R such that d M C R.
REMARK. R-module like this M is often called a fractional ideal of R
under certain restrictions such as M contains a non-zero-divisor.
PROOF. M C Rd-1 , and Rd-' is a finitely generated R-module. Since
R is Noetherian, Rd- is a Noetherian module (Theorem 3.6.2). Q.E.D.
THEOREM 3.9.2. Let f (x) be a monic polynomial over an integrally closed
integral domain R, and let a be a root of f (x) (in some field containing
R) . Let R* be the integral closure of R[a] in its field of fractions, and let
f'(x) be the derivative (d/ dx) f (x). Then we have f (a)R* C R[a]. If d is
the discriminant of f (x) , then we also have dR* C R[a].
PROOF. Let the roots of f (x) be u1 = a, u2 , ... , ur (r = deg f (x)) ,
and we set gi(x) = f(x)/(x - u,) (i = 1, ... , r). Then f4(x) = > g,(x)
and f4(a) = g1(a) . Therefore, it suffices to prove the assertion under an
additional assumption that f (x) is irreducible. If a is inseparable, then
f'(a) = d = 0, and we assume that a is separable. Let R' be the smallest
Galois extension of R containing a, and we set G = G(R'/R) (note that
R' is integral over R and is integrally closed). Let H be the subgroup
corresponding to R* (namely, H is the set of elements of G which leave
elements of R* invariant). We take of E G such that Qi u 1 = ui for each
i = 1, ... , r (a1 = 1). Then G is the union of of H (i = 1 , ... , r)
(note that as = ui if a E a,H). o f (x) = f (x) , and therefore, gi (x) =
f(x)/(x - a1u1) = a g1(x) We write g1(x) = c,._
.
+ ... + co (ci E R[a]
1xr-1
,
cr_ = 1). Now, let b be an arbitrary element of R*. Then

1
r
b.f (a) = bg1(a) _ aibgi(a) _ (aib)(a,g1(a))
= j(oib)(oicC)a' = ((aib)(aici)) a'

i,.l J i
Since b and ci are H-invariants, E,(aib)(aicj) is G-invariant and is in R,

which shows that bf'(a) E R[a], i.e., R* f'(a) C R[a].
As for the discriminant d, we know that d = [ 1,j (ui - uj)2 . But f'(a) _
fli>1(u1- ui) and we have d c f'(a)R*. Therefore, dR* C f'(a)R* C R[a].
Q.E.D.
COROLLARY 3.9.3. Let R be an integrally closed integral domain with field

of fractions K. If L is a finite separable algebraic extension of K, then any
subring R' (D R) of L which is integral over R is a finitely generated R-
module.
PROOF. Let a be an element of L such that L = K(a). Then there are

elements c. E R with co 0 such that coa" + + cn = 0. Taking coa
instead of a, we may assume that a is integral over R. The integral closure
R* of R[a] in L coincides with the integral closure of R in L. Since a is
separable, we see the assertion by the theorem and the lemma above. Q.E.D.
LEMMA 3.9.4. In Corollary 3.8.3, if A is an integral domain that is sep-

arable over K, then we can choose z1 , ... , z1 so that A is separable over
K[z] = K[z1 , ... , z1] .
PROOF. The assertion is obvious in the characteristic 0 case, and we con-

sider the nonzero characteristic p case. We can assume that a,.+1 , ... , a,,
form a separating transcendence base of K(a) over K. Then z1 , ... , zr
(t = n - r) , obtained by our proof of Corollary 3.8.3, are such that a,.+1 - z; =
f (al , ... , ar) E n[ap , ... , ap] . On the other hand, it holds in general that
K(a) is separably algebraic over K(ar+1 , ... , an) if that D E Der(K(a)/K)
satisfies Da, = 0 for i = r + 1 , ... , n, implies D = 0. Since a,.+1 - zi E
n[ap , ... , ap] , we have that Dar+i = Dzi for all D E Der(K(a)/K. This
implies that z1 , ... , z, are the required elements. Q.E.D.
THEOREM 3.9.5. Let A be a finitely generated integral domain over afield

K and let L' be a finite algebraic extension of the field of fractions L of A.
Then the integral closure A' of A in L' is a finitely generated A-module,
and hence, is a finitely generated integral domain over K.
PROOF. We can take K[z] in Corollary 3.8.3 instead of A. Thus we can

assume that A = K[z] = K[z1, ... , z1] (polynomial ring), and L = K(z).
If L' is separable over L, then the assertion follows from Corollary 3.9.3.
We consider the inseparable case. Corollary 3.4.5 shows that there is a purely
inseparable finite extension K' of K such that L'(K') is separable over K' .
Then Corollary 3.9.4 and Theorem 3.9.3 imply that the integral closure At
of A" = K'[z1 , ... , z1] in L'(K') is finite over A". Since A" is finite over
A, A* is finite over A and A* is a Noetherian A-module. Therefore, its
submodule A' is a finitely generated A-module. Q.E.D.
REMARK. In this proof, we have to take a separating transcendence base of
L'(K') over K'. There is another way to prove this assertion. Namely, if we
take a suitable power q of the characteristic p , then L' (K') (z 11q , ... , z 1q)
is separably algebraic over K ... , zr q) Thus, we consider the integral
.
closure of K[z] in L'(K')(zl1q, ... , z,'1q) instead of A* above.

10. ALGEBRAIC VARIETIES 113
10. Algebraic varieties

One good family of algebraic varieties consists of projective varieties,
namely, algebraic varieties in projective spaces. But, as an introductory sub-
jects of algebraic varieties, we deal only with affine varieties in this book.
There are several ways to treat algebraic varieties; by coordinate systems,
by rings, or by so-called sheaves. Here, we define them by using coordinate
systems so that the readers may easily have concrete images.
Affine space of dimension n over a field K is the set { (al , ... , an) Ja, E
K} ,i.e., the n-ple product of the field K. We shall denote it by S"(K).
Each element (a1 , ... , an) of Sn (K) is called a point and these ai are
called the coordinates (the first coordinate, the second coordinate,... , the
nth coordinate) of the point. A subset A of S" (K) is called an algebraic
set if there is a subset J of the polynomial ring K[XI , ... , Xn] such that
A = {P E S"(K)I f (P) = 0 for all f c J} In this case A is called the
.
algebraic set defined by J and is denoted by V(J). Each point of V(J) is

called a zero point of J. To define the algebraic set by J is the same as
to define it by the ideal generated by J. On the other hand, if A = V (J) ,
then I(A) = f f E K[XI , ... , Xn]I f (P) = 0 for all P E Al is an ideal
containing the subset J. For each P = (a1 , ... , an) E A, we consider
MP = > 1(X_ - a!)K[X]. Then I(A) = nPEA MP M. Since each Mp is a
maximal ideal (hence is a prime ideal), we see that VI -(A) = I(A).
THEOREM 3.10.1. With the notation as above, we have
(I) V(JI U J2) = V(>fEJ, fK[X] + EgEJ2 gK[Xl) = V(J,) n V(J2);
(ii) V(JI J2) = V (>fEJ, ,gEJ2 fgK[X]) = V (JI) U V (J2) ;
(iii) I(VI u V2) = I(VI) n I(V2).
The proof is easy and we omit it.
THEOREM 3.10.2 (zero-point theorem of Hilbert). If K is an algebraically
closed field, then, for algebraic sets AI , A2 defined by ideals J1 , J2, respec-
tively, of K[X] = K[X1 , ... , Xn], we have
Al =A 2 ifs J1 = J2
REMARK. Sometimes, Corollary 3.8.6 is called the zero-point theorem of
Hilbert.
PROOF. We begin with a remark on maximal ideals. Let M be a maximal
ideal of K[X]. Since K is algebraically closed, we have K[X]/M = K by
Corollary 3.8.5, namely, there are ai E K, X, - a. E M for all i. Thus, each
maximal ideal M determines a point P = (a1 , ... , an). Now, Corollary
3.8.6 shows that J; = (l MP , where MP runs through maximal ideals
containing Ji. The set of points P determined by MP coincides with A. .
Therefore, AI =A 2 implies JJ = J2 The converse is obvious. Q.E.D.

.
The theorem above becomes false if we drop the assumption that K is

algebraically closed. For instance, if K is the real number field R, then
the algebraic set defined by X1 + XX + + Xz consists only of the ori-

gin (0, ... , 0), and coincides with the algebraic set defined by >, X1K[X] .
Therefore, we usually develop the theory first in the case where K is al-
gebraically closed and then in the other cases, we apply the results over an
algebraically closed field S2 containing K. For the purpose, we need some
notions as follows.
If K' is a subfield of K, we regard S" (K') as a subset of S" (K) and
points in S"(K') are called K'-rational points. In dealing with an algebraic
set A' in S" (K') , we first consider an algebraic set A in SN(K) , or in
S" (a) , with a big algebraically closed field 52 containing K, defined by the
same set of polynomials, and we regard A' as the set of K'-rational points in
A. In this sense, though the algebraic set in S" (R) defined by X + +X2l _
0 coincides as a point set with the one defined by X1 = . . = X" = 0 , we
.
distinguish these two as algebraic sets. More precisely, algebraic sets should
be interpreted not as point sets by themselves, but as pairs of point sets and
systems of equations (or ideals) which define them. For instance, though Xl
and Xi determine the same point set, we should understand that XI defines
a hyperplane and X2 defines a special quadratic hypersurface. However, in
this book, we intend to give some primitive notions on algebraic sets, and
therefore, we treat them mainly as point sets.
In the following, we fix an algebraically closed field S2 and consider points
in S" (a) . Namely, we restrict coordinates of points to elements of Q, and
we call this S2 the universal domain. To have enough freedom, 52 should
be big enough. Thus, we usually assume that the transcendence degree of S2
over its prime field is infinity.
An algebraic set V in S" (52) is said to be irreducible, or an algebraic vari-
ety, if it is defined by a prime ideal P of 52[X1, ... , X,1] . trans. deg, 52[X]/P
is called the dimension of V and is denoted by dim V.
THEOREM 3.10.3. Each algebraic set A in S" (52) is the union of a finite
number of algebraic varieties Vl , ... , Vt. If the union is irredundant, then
Vl , ... , V are uniquely determined by A (disregarding the order).
In this latter case, these V are called the irreducible components of A .
The maximum of dim V is called the dimension of A.

PROOF. Let J be an ideal which defines A. is the intersection of
minimal prime divisors Pl , ... , and A = V(J) = U;'t V(Pi) . Con-
1
versely, if we set Q. = I(V), then = I (A) = ni=1 i(V) = n Q! , and

each Q. is a prime ideal. If n Q. is redundant, say, Q1 is redundant in
the intersection, then Q1 nt>1 Q. and Vl c U1>, V. , a contradiction to
the irredundancy of the union. Therefore n Qi is irredundant, these Qj
must be the minimal prime divisors of J, and these V. coincide with V(Pi)
(i=1,...,m). Q.E.D.
Some special algebraic sets have their own names.
(i) A hyperplane is the one defined by a linear form E`'=1 a;Xi - b ((a1 ,
.) an) 0(0,...,0)).
(ii) A hypersurface is the one defined by a polynomial f (X1 , ... , X").
deg f is called the degree of the hypersurface.
(iii) A linear variety is the one defined by several linear forms f(X)
Ej a1jXj -b; (i = 1, ...,r).
(iv) A linear variety of dimension one is called a line and a linear variety
of dimension two is called a plane.
(v) A one-dimensional algebraic variety is called a curve and a two-
dimensional algebraic variety is called a surface.
REMARK. The terms "curve" and "surface" may be used under weaker
conditions. Namely, an algebraic set is called a curve or a surface if every
irreducible component is a curve or a surface, respectively.
REMARK. Consider the case where n is the complex number field C.
Then a line (which may be called a complex line) is two dimensional over
the real number field R ; for instance, S I (C) is the so-called Gauss-plane
(S2(R) E) (a, b) a+b E C). Therefore, it is quite hard to have
intuitive images of planes, curves, surfaces, and so on over C. Therefore,
the readers are advised to have intuitive images in S3(R) .
The following two theorems follow easily from our results in 8.
THEOREM 3.10.4. If V and V' are algebraic varieties in S" (S2) and if
V c V', then dim V < dim W. Furthermore, there are algebraic varieties U1
(i 1, ... , r - 1, with r = dim V'- dim V) such that
Ur = V' and dim Ui = dim V + i .
The proof follows from Theorem 3.8.4.

THEOREM 3.10.5. If an algebraic set A in S" (S2) is defined by f, , ... , f,. E
S2[XI , ... , X"], then the dimension of each irreducible component of A is at
least n - r.
The proof follows from Theorem 3.8.7 and Corollary 3.8.5.
We say that an algebraic variety V in S" (S2) is defined over a subfield
K of 0 or that K is a field of definition for V , if the prime ideal I (V) of
S2[X1 , ... , X"] is generated by elements of K[XI, ... , X"]. An algebraic
set A in S" (S2) is said to be K-irreducible, if A is defined by a prime
ideal P' of K[X] = K[XI, ... , X"]. In this case, K[X]/P' is called the
coordinate ring of A over K, and the field of fractions of K[X]/P' is called
the function field of A over K .
THEOREM 3.10.6. Let A be a K-irreducible algebraic set in S" (Q). Then
A is an algebraic variety defined over K if and only if the function field K(A)
of A over K is a regular extension of K.
PROOF. Let P be the prime ideal of K[X] which defines A. Then A
is defined by Pfl[X]. A is defined over K if PS2[X] is a prime ideal,
which is equivalent to (K[X]/P) K c2 being an integral domain. Since Q is

an algebraically closed field, the last condition implies that (K[X]/P) K K
(with K the algebraic closure of K) is an integral domain, namely, that
K(A) is a regular extension of K. From this, it follows that (K[X]/P) K SZ
is an integral domain. Q.E.D.
THEOREM 3.10.7. Let V be an algebraic variety in S" (0) . Then a field
K is afield of definition for V if and only if K contains all of the ci3 stated
below.
Denote by xi the residue class of Xi in Q[X]/I(V). We number all
the monomials in x1 , ... , x" : MI = 1, m2 , m3 , ... (note that the set of
monomials consists of countably infinite elements). Set Mn = >i<n m1
(n=1,2,...). SetB={ilmiCMi_1}. Then {mjIj<n, j B} isa
linearly independent base of Mn and, for each i E B, mi can be written in
the form cm (c.. E S2) .
PROOF. We denote by m' the monomial in XI, ... , X" corresponding

to mi Then ml - > ci l m' (i E B) generate I (V) , and therefore, if all the
.
ci.l
E K, then K is a field of definition for V. Conversely, assume that K
is a field of definition for V. Since S2[X]/I(V) ^_' K[X]/(I(V) flK[X]) K S2
and since the mi are elements of K[X]/(I(V) n K[X]), we can consider
M Then we have Mn Al K S2 and mi c M* , if i E B .
Therefore, if we write. mi in the form Es<i,s B ctsms , then by the linear

independence of ms , we have cis and cis C K. Q.E.D.
COROLLARY 3.10.8. Among the fields of definition for an algebraic variety
V, there is the smallest one, say k0 . Among the fields of definition containing
a field K (c 0) , K(k0) is the smallest.
This k0 is called the smallest field of definition for V and K(k0) is called
the smallest field of definition for V over K.
PROOF. k0 is the field generated by these ci1 (over the prime field).
Q.E.D.
REMARK. The smallest field of definition k0 is finitely generated over the
prime field, because the prime ideal I (V) is finitely generated and the field
generated by the coefficients of a system of generators contains k0 .
For a subset A of S" (S2) , a conjugate of A over K is aA = {(aa1 , ...

aa"Ma1 , ... , a") E Al with a K-automorphism a E AutK S If A is an .
algebraic set defined by an ideal J, then aA is the algebraic set defined by

aJ = {a f I f E J} We say that A is normally algebraic over K , if QA = A
.
for all a c AutK 0.

THEOREM 3.10.9. Let Vt , ... , V" be the irreducible components of an al-
gebraic set A in S"(0). Then the following three conditions are equivalent.
(i) A is normally algebraic over K.
(ii) A is defined by a subset of K[X1, ... , X].
(iii) These V are defined over the algebraic closure K of K, and for each
z c G(K/K), there is a c AutK S2 such that a I K = r and aA = A.
For the proof, we need two lemmas.
LEMMA 3.10.10. Let f1 , ... , ft be elements of a ring R and let J be an
ideal of R. If all of the elementary symmetric forms sl , ... , st of f1 , ... , f
are in V-T then the f are in v'j-.
PROOF. Since the f are roots of F(X) = X` - s1 X`- I + + (-1)'st , we
have ft-sl(f )t-I +- +(-1)tst = 0. Thus, ft c V7, and f c . Q.E.D.
LEMMA 3.10.11. Let ko be the smallest field of definition for an algebraic
variety V in S" (S2) . Then for a a C Aut K2,
aV=V iff as=a forallaeko.
PROOF. The elements c. j in Theorem 3.10.7 are uniquely determined by
V. Therefore, a V = V implies ac, j = ci j for all cj . Thus, as = a for
all a E ko, because ko is generated by c.
Conversely, if as = a for all
a E ko, then a certain system of generators of I (V) is invariant under a
and aI(V) = I(V) ; hence, aV = V. Q.E.D.
PROOF OF THEOREM 3.10.9. Let Ki be the smallest field of definition for
over K.
To prove that (i) implies (iii), suppose, for a moment, that some Kl, say
K1 , is not algebraic over K. Take a transcendence base z , ... , zt of K11
over K. Then for any set of algebraically independent elements u 1 , ... , ut

of 0 over K, the K-isomorphism a of K(z1, ... , z,) to K(u1, ... , u,)
is extended to a K-automorphism of S (Lemma 3.1.5). Therefore, there are
infinitely many conjugates of Vl (Lemma 3.10.11; there are infinitely many
even if we set ui = zi + ci with ci E S2) . But (i) implies that conjugates of
Vl are components of A. Thus we see (iii).
To prove that (iii) implies (ii), let f , ... , f"(t) C Ki[X] be a set of
1
generators of I(V), and let L be a finite normal extension of K containing

all K. (i = 1, ... , m). Note that each a E G(L/K) is extended to an
element of AutK S and that a fj E I (a V) = I (VS) for some s. Thus,
r1[. . I(Vi) is generated by the elements F., ...int = fI ji ) fnjm and a F.,..Jm E
Il; I (Vi) Next, for each F "Jill , we consider the elementary symmetric
.
forms s1 , ... , st of {aF. Ia E G(L/K)}. Then sq C K[X] with q =

[L: K]1. Let J be the ideal generated by all of these s7 with respect to all
of F. . Then we see that F. E JS2[X] by Lemma 3.10.10, and
rJ.I(V) c JS2[X]. Since J c r11I(V), we have VTJi = JS2[X],
which implies (ii).
That (ii) implies (i) is obvious. Q.E.D.
When subsets A , B of S'" (S2) , S"(0), respectively, are given, the set
{(al,...,am,bl,...,bn)ESm+,l(1)I(al,...,a,n)EA,
(bl,...,b,t)EB}
is called the product of A, B and is denoted by A x B (A x B C S'"+" (0))
THEOREM 3.10.12. If A
CS'" p), B C S" (0) are algebraic sets defined
respectively, by ideals JA of SZ[X1 , ... , JB of '2[Yl , ... , Yn] Then
Xm] , .
A x B is the algebraic set defined by the ideal generated by JA and JB in the

polynomial ring 52[X1 , ... , X,n , Y 1 , . . . , Yn] .
JA, JB are prime ideals iff JAS2[X , Y] + JBS2[X , Y] is a prime ideal.

Therefore, if A, B are algebraic varieties, then A x B is also an algebraic
variety.
PROOF. Set J = JAS2[X , Y] + JBS2[X , Y] , and let C be the algebraic
set in (92) defined by J. If P E A, Q E B, and f c J, then f =
S'n+n
E aigl + > bjhj (a1 E JA , bj E JB ; hi, gj c S2[X, Y]) . Thus, f (P , Q) =

>ai(P)gi(P, Q)+>bj(Q)hj(P, Q) = 0, and PxQ E C. Thus, AxB C C.
Conversely, if P x Q is not in A x B, then P 0 A or Q 0 B. Suppose
P 0 A. Then for some a c JA, a(P) = a(P, Q) 0 and P x Q 0 C.
Thus, we see that A x B = C.
S2[X, Y]/J (S2[X]/JA) o (S2[Y]/JB) . Since 0 is algebraically closed,
we see that if JA and JB are prime ideals, then J is a prime ideal by The-
orem 3.5.2 (cf. Exercise 3.5.4). The converse is obvious, because S2[X]/JA
and S2[Y]/JB are regarded as subrings of S2[X, Y]/J (cf. 2). Q.E.D.
COROLLARY 3.10.13. In the theorem above, if A and B are algebraic
varieties defined over a field K, then A x B is also defined over K.
PROOF. The ideal J is generated by JA and JB , and therefore, J is
generated by elements of K[X, Y]. Q.E.D.
11. The C,-conditions
For a given algebraic variety defined over a field K, it is usually difficult to
see if the variety has K-rational points. Here, we give some theorems which
assure the existence of K-rational points.
We say that a field K satisfies the Ci-condition or that K is a Cifield (i
is a nonnegative real number), if the following condition is satisfied.
If f(X1 , ... , Xn) is a homogeneous form of degree d in n
variables Xl , ... , X, with coefficients in K and if n > d',
then the hypersurface f (XI , ... , Xn) = 0 has a K-rational
point (a1,...,an)0(0,...,0).
Such an (al , ... , an) is called a nontrivial solution of f = 0.
We say that a field K satisfies the CI-condition or that K is a C,-field if
the following condition is satisfied.
If f(Xi , ... , XX) is a polynomial of degree d in n vari-
ables X, , ... , Xn such that f(0, ... , 0) = 0 and if n >
11. THE Ci-CONDITIONS 119
d' , then the hypersurface f = 0 has a K-rational point

(a1,...,an) (0, ...,0).
Such an (a1 , ... , an) is also called a nontrivial solution of f = 0.
It is obvious that every Cr-field is a Ci-field.
THEOREM 3.11.1. Afield K is a Ci field with i < 1 iff K is algebraically
closed.
PROOF. The if part is obvious. Assume that K is not algebraically closed.
Let L be a finite normal extension of K with linearly independent base
a1 = 1, a2, ... ,an ([L:K]=n> 1). Set G=G(L/K), e=[L:K], and
g(X1 , ... , X,,) = [LEC(aa1X1 + +
aanXn)e
. degg = n, which coincides

with the number of variables Xi , and the coefficients of g are in K. If
g(bl , ... , bn) = 0 with b, E K, then for some a E G, E aaibi = 0. Then
E aibi = 0, so bi = 0 for all i (because the ai are linearly independent).
Thus, K is not a Ci-field with i < I. Q.E.D.
REMARK. This proof shows that if we change n > d in the C1 condition
to n > d, then the latter implies the Co condition.
THEOREM 3.11.2. Let K be a Ci field. If f1 , ... , J. are homogeneous
forms of degree d in n variables X1 , ... , Xn and if n > rd', then f1 =
0, ... , fr = 0 have a common nontrivial solution in K.
THEOREM 3.11.3. Let K be a C'-field, and let f, , ... , ffr be polynomials
in n variables X1 , ... , Xn . If every f has no constant term, deg f < d,
and if n > rd', then f1 = 0, ... , fr = 0 have a common nontrivial solution
in K.
We shall prove these two theorems simultaneously. In our proof, a "form"
of "degree" d will mean (i) in the Ci condition case, a homogeneous form
of degree d and (ii) in the C, condition case, a polynomial of degree < d
having no constant term.
When 0 is a "form" in variables Xi) , ... , X(') and f1 , ... , J. are
"forms" in variables X1 , ... , Xn of "degree" d, then we define "forms"
O(i)(f1 , ... , fr) (J = 0, 1, 2, ...) as follows.
0()(fi , .. , fr) _ O(x(), ... , x o)) (the original ) .

If 0 (') (f1 , ... , fr) = gj(Xi') , ... , X))
Ni
, then letting ti be the integer part
of N / r , we introduce tin variables X ('+ 1) , ... , x' )

and define
1
U+1) -= gj (fl(XIU+1) , ... , XnU+1) ), ... , f.(X1U+1) , ... , U+1)
fI( X('+1)
n+1
X('+1)
2n )' "' ' fr( X('+1)
11+1
X('+1)
2n
f(XU+1) ,... ,X('+1)),0,...,0)
l (tj-I)n+1 ,...,X('+1)),...,
tfn f (X('+1)
r (Ui - I)n+ tin
(the number of 0's is N. - rte (< r)).

Then the following two lemmas hold.
LEMMA 3.11.4. If f1 = 0, ... , fr = 0 have no common nontrivial solution
in K and if 0 = 0 has no nontrivial solution in K, then 0(') (f1 , ... , f,.) = 0
has no nontrivial solution in K.
LEMMA 3.11.5. Assume that n > rds with s > 0, d > 1 Let Di be the .
degree of 0(1) (f1 , ... , fr) Then limb Nj/Ds = 00

. .
PROOF OF LEMMA 3.11.4. Consider the case j = 1 . If 0(1)(f1 , ... , f,.) _

1)
g1(Xi , ... , 0 has a solution (a1 , ... , aNJ) in K, then 0 = 0
has a solution (f1(a1 , ... , an), ... , f.(a1 , ... , all), f1(an+1 , ... , a2n) , ... ,
fr(an+l, ... , a2n), ... , fl (a(t0_i)n+1 , ... , aloe) , ... , f.(a(t0--1)11+1 , ... , aloe) ,
0, ... , 0), in K. Since 0 = 0 has no nontrivial solution in K, we see that
f1(a1 , ... , an) = 0, f2(al , ... , an) = 0, ... , fr(a1 , ... , an) = 0; a contra-
diction to the property that f1 = 0, ... , f,. = 0 has no nontrivial solution in
K. The general case follows from this by induction on j. Q.E.D.
PROOF OF LEMMA 3.11.5. Obviously d Di. Set a = n - rds.
Nj+1 = ntj, tar < N < (tj + 1)r, and therefore, we have Nj+1 /Di+I >
(rds + a)tj/dsDs. First, we consider the case where Di is not bounded.
Set Nj/r = tj + uj. Then ui < 1 , we have (rds + a)((N/r) -
uj)/dsD = N This last term is very small.
Since a/rds is a positive constant, we see that Nj/D increases for large j
and is not bounded in this case. Now, we consider the case where the D, are
bounded, then N. obviously tends to infinity and lim Nj/Ds = oo, in this
case too. Q.E.D.
PROOF OF THE THEOREMS ABOVE. If K is a Co field, then K is alge-
braically closed (Theorem 3.11.1) and the assertions in the theorems are ob-
vious if K is algebraically closed (Theorem 3.10.5). Therefore, we assume
that K is not algebraically closed. Then there is a "form" 0 such that q = 0
has no nontrivial solution in K. Let N be the number of variables in q . By
our proof of Theorem 3.11.1, we can assume that N > 1 . If N < r, then,
considering q(t)(0) with a suitable t (in view of Lemma 3.11.4), we may
assume that N > r, and we consider q(j)(f1 , ... , fr). Lemma 3.11.5 shows
that N./D is large for big j. Thus, there is a j such that Nj/D > 1 .
Since K is a Cl-field (or a C, -field), O(')(f1 , ... , f,.) = 0 has a nontrivial

solution in K. Therefore, there is a common nontrivial solution in K of
f1 = 0, ... , fr = 0 by Lemma 3.11.4. Q.E.D.
The following two theorems are important.
THEOREM 3.11.6. Let K be a C! field. If an extension field L of K has
transcendence degree s over K, then L is a C,+s field.
11. THE Ci-CONDITIONS 121
THEOREM 3.11.7. Let K be a C'-field. If an extension field L of K has

transcendence degree s over K, then L is a Cl+s-feld.
PROOF. We prove these two theorems simultaneously by using "forms" of
"degree" d as in the proof of the theorems above.
Let 0 be a "form" of "degree" d in n variables such that n > d1'. We
shall show that 0 = 0 has a nontrivial solution in L. By considering the
field generated by the coefficients of 0 over K , instead of L , we can assume
that L is finitely generated over K.
(i) The case s = 0. Take the smallest normal extension L* of K con-
taining L and choose gj = 0(j) (0) so that N /D > [L* : K]` . Set f =
fQEG(agj)e with G = G(L*/K) , e = [L*: K],. f is a "form" with co-
efficients in K, in Nj variables, and of "degree" [L*: K]D j . Therefore,
Nj > (degree)` and f = 0 has a nontrivial solution in K (hence, in L) .
This implies that 0 = 0 has a nontrivial solution in L, by Lemma 3.11.4.

(ii) The case s > 0. By using induction on s and the result in (i), we
have only to prove the case s = 1 and L = K(z). Then we can assume that
the coefficients of 0 are elements of K[z]. Let r be the maximum of the
degrees in z of these coefficients. Since n > d`+1 , there is a natural number
t such that n - d i+1 > (d ` (r + 1) - n) t-1 . We introduce new variables yuv
(u = 1 , ... , n ; v = 0, ... , t) and set Xu = Ev=o yuv zv . Then 0 is written
as q5 = f0 + fl z + .. I
where each fm is a "form" in the n(t + 1)
+fdt+rZdt+r
-
variables yuv of "degree" d. Since n d'+' > (d ` (r + 1) - n) t-1 , we have
n (t+ 1) > d ` (d t+r+ 1), and therefore, fo = 0, ,
fdt+r = 0 have a common
nontrivial solution (r1 uv) in K . Then u = Ev=0 quv zv (u = 1 , ... , n) give
a nontrivial solution in K(z) of 0 = 0. Q.E.D.
THEOREM 3.11.8. Every finite field is a C'-field.
For the proof of this theorem, it suffices to prove the following theorem
(because every polynomial equation having no constant term has a trivial
solution).
THEOREM 3.11.9. Let F be a finite field of characteristic p and set q =
#(F) Let f (XI , ... , Xn) be a polynomial of degree d in n variables Xi
. .
If n > d, then the number N of solutions of f = 0 is a multiple of p.

PROOF. Consider the n-dimensional affine space S = S" (F) over F .
From the fact that a E F, a 0 0, implies aq-1 = 1 , we have the following

equality in F,
f(p)q-1 = q" - N = -N.
PES
Therefore, it suffices to show that the left-hand side is 0. We write f q-1 as

> aimi (0 # ai E F, mi are monomials). Then deg mi < d (q - 1). Now, it
suffices to show EPES m, (P) = 0 for each ml . Write mi = XI Xn" with

Eej < d(q - 1) .
ynl (p) = TT cej c" ) (1: H C

R CjJ Cei JJ
PES cJEF j ccEF R1 cEF j> 1
Thus, we have APES mi (P) = flj (UCEF Ce') Each factor of this last .
product takes the following value. (i) -1 = 1 x (q - 1) if ej > 0 and ej is

a multiple of q - 1 , (ii) 0, otherwise, as is shown by the following.
Let g be the G.C.M. of ej and q - 1 , and set h = (q -1)/g Then, using .
a ggenerator co of the multiplicative group F - {} 0 EcEF Cei = i=1 9- I iej =

CO
2e. (h-1)eJ
e_ he'.
+co' +...+co) = g(1 +c0 +....+c0
e' .
) = g(1 -c0)/(l -c0) = 0.
he' e'g(c0'
Since n > d and deg mi < d (q - 1) , at least one of these factors is of the
case (ii), and we have completed the proof. Q.E.D.
12. The theorem of Luroth
LEMMA 3.12.1. Let x be a transcendental element over a field K K. We

write y E K(x) (y V K) in the form f(x)/g(x) with f (x) , g(x) E K[x]
so that f, g have no common factor. Then the polynomial g(X)y - f(X)
is an irreducible polynomial in X over K(y) (hence is an associate with the
minimal polynomial for x over K(y)). Consequently,
[K(x) : K(y)] = max{deg f , deg g} .
-
PROOF. g(X)y f(X) is linear in y, hence is irreducible as a polynomial
in y over K(X). Since K[X] a unique factorization domain, g(X)y -
f (X) is irreducible as a polynomial in X, y, by Corollary 1.6.7, hence is
irreducible as a polynomial in X over K(y). Q.E.D.
THEOREM 3.12.2 (theorem of Luroth). If x1 , ... , xr are algebraically in-
dependent over a field K and if L is a field such that K C L CK(x1 ) ... , x,.)
and trans. degK L = 1, then there is an element t such that L = K(t).
PROOF. In the case r = 1 , let h(X) = X" + +c,, (ci E L) be
+c1Xn-i
the minimal polynomial for x = x1 over L. Since these ci are in K(x), by

multiplying the least common multiple d0(x) of denominators of these ci,
we obtain
(x)Xn-i
h'(X, x) = do(x)Xn + di + ... + d,,(x)
(di E K[x] ; do, ... , d have no common factor).
On the other hand, let c be one of ci that is not in K and write c =
f (x)/g(x) so that f , g have no common factor (g is a factor of do and
f is a factor of di). By the lemma above, the minimal polynomial for x
over K(c) is (f(X) - cg(X)) x (an element of K(c)) whose degree in is
12. THE THEOREM OF LUROTH 123
equal to max{deg f , deg g} = [K(x) : K(c)]. Since h(X) is the minimal

polynomial for x over L, f(X) - cg(X) = h(X)q(X) for some q(X) E
L[X]. Multiplying by a suitable element of K[x], we have
O(x)(f (X)g(x) - f(x)g(X)) = h'(X, x)q'(X, x).
Then q5(x) divides q'(X, x) in K[X, x], and therefore, we may assume
that O(x) = 1 The degree of the left-hand side in x is m. The degree
.
of h'(X, x) in x is maxi{degd,(x)} , hence is > m. Therefore q' is a

polynomial in X only and the degree of h' in x coincides with m. Suppose
that q' is not a constant. Then, taking a root a of q' (in an algebraic
extension of K), we have f(a)g(x) - f(x)g(a) = 0. Since f(x), g(x)
have no common factor, it cannot be f(a) = g(a) = 0. If g(a) # 0, then
c = f(x)/g(x) = f (a)/g(a) and c is algebraic over K, a contradiction. If
f (a) 0 0, then we have a similar contradiction. Thus, f (X) - cg(X) = (the
minimal polynomial for x over L) x (an element of K(c)) and [K(x) : L] =
[K(x) : K(c)]. Therefore, K(c) = L, which completes the proof of the case
r= 1.
As for the case r > 1 , we prove here the case where #(K) is infinite
and for that purpose, we prove Lemma 3.12.3 below. The proof, including
the case where K is a finite field, can be given by using Lemma 3.A.2 in
Appendix.
LEMMA 3.12.3. Let K1 , K2 be finitely generated fields over an infinite field
k. If there is an element x that is transcendental over both K1 and K2 and
such that K1 (x) = K2(x) , then K1 is k-isomorphic to K2 .
PROOF. Write K1 = k(y1 , ... , Yin), K2 = k(z1 , ... , zn). There are ra-
tional expressions of y, ,zi such that y. = f (z1 , ... , z,t , x) and zi =
gj(y1,...,y,n,x). Then yj = zi =
gi (f1(z , x) , ... , fin(z , x) , x) Since #(k) is infinite, there is an element
.
a of k such that the denominators of f,., gj and also those of the new ex-
pressions of y, , zi do not vanish by the substitution of a for x. We
set yi = f (z1 , ... , zn , a) , zi = gj(y1 , ... , Yin , a). Since x is tran-
scendental over k(y), we have y, = f (g1(y , a), ... , gn (y , a), a), namely,
y; = f (z; , ... , zn , a). Similarly, zj = gj(Y1 , ... , y,,, , a). Thus, K1 =
k(y1 , ... , Yin) = k(zi , ... , z'1) and K2 = k(z1, ... , zn) = k(yi , ... , y,,,) .
Hence, trans. degk K1 =trans. degk K2. Further, if F (Y1 , ... , Y,) = 0 is
a relation satisfied by y1 , ... , Yin , then F(f1(z , x) , ... , f;,,(z , x)) = 0 .
Since x is transcendental over k (z , ... , z,) , we have F (y' , ... ,
1
0 .
Therefore, there is a k-homomorphism 0 from k[y1 , ... , y,,] to

k[yi , ... , y,n] such that qyj = y'. The coincidence of the transcendence de-
grees implies that must be an isomorphism. Thus,
Kk(y1,...,Yin) 2- k(YI,...,y;nk(z1,...,zn)=K2 Q.E.D.
The proof of Theorem 3.12.2 in r > 1 case. We use an induction on r.
We can assume that x1 is transcendental over L. Then, K(x1) c L(x1) C

K(x1 , ... , x,,). By applying the induction hypothesis to this extension, we
see that L(xl) = K(x1) (t) with a transcendental element t over K(x1).
Then, Lemma 3.12.3 shows that L is isomorphic to K(t). Q.E.D.
Appendix. A theorem on valuation rings and its applications
Although we define valuation rings in the next chapter, readers are assumed
to be familiar with them.
THEOREM 3.A.1. Let Vx be a valuation ring of the rational function field
K(x) over afield K, and set V = V n K Let Mx and M be the maximal
.
ideals of Vx and V, respectively. If kx = Vx/MX is not algebraic over k =

VIM, then there is t E kx such that kx = k'(7), where k' is the algebraic
closure of k in kx .
REMARK. We can choose x so that V[x] c Vx. In this case our proof
below shows 1 can be chosen as the residue class of the element t as follows
(in the following, - denotes the residue class modulo Mx) .
For each element h of K(x), d*(h) is defined as the maximum of the
degrees of the numerator and the denominator of the irreducible fractional
expression of h . Then t is an element of T = j h E Vx h is transcendental
over k} such that d*(t) < d*(h) (Vh E T).
PROOF. Let t be as in the remark above and let t = f(x)/g(x) be the
irreducible fractional expression of t. First, we reduce to the case where
deg f > deg g without changing k (t) . If deg f < deg g , we take t- I instead
of t. Assume that deg f = deg g . Then f = ag + r (a E K, deg r <
deg g) . (i) If a E V, we take (t - a)-1 (= g/r). (ii) Otherwise, we take
(a- I - t- I) -1(= at/(t - a) = a f /r) Thus, we have the case deg f > deg g.
.
Next, we shall show that

K[x] C K[t] + K[t]x + + K[t]xn-I (n = deg f) .
Since xn = b f + s (b c K , degs < deg f) , we have xn = bgt + s

and xn C (the right-hand side). Consequently, xn+I , xn+2 , ... are in the
right-hand side as desired.
Now, an arbitrary element h of Vx is written as h = (ao + a It + +
amt'n)/(bo + blt + ... + bete) with al , bi C K[x] such that dega, < n,
deg bi < n. By dividing both the numerator and the denominator by one of
a, or bi that generates the ideal > al Vx + E bj - Vx , we may assume that (i)
a, , bi C V , (ii) d * (a,) < n, d * (b3) < n, and that (iii) at least one of a, or
bi is . From this relation, we have
1
a, , bi are algebraic over k , because d * (a;) , d* (bj) are less than n Since
.
APPENDIX. A THEOREM ON VALUATION RINGS AND ITS APPLICATIONS 125
l is transcendental over k and at least one of ai , bi is , we see that

1
b is not 0 , and h c k'(7) . Since this is true of every h c VX we see

that kX = k'(t) . Q.E.D.
One application of this result is to the last step of the proof of the Theorem
of LUroth, and it suffices to prove
LEMMA 3.A.2. Let k be a subfield of a field K, and let x be a transcen-
dental element over K. If K(x) = k(x , t) and if x, t are algebraically
independent over k, then K ^' k(t).
PROOF. Let v be an additive valuation of K(x) such that v is trivial on
K and vx = 1 Set L = k(t) . If v is trivial on L , then the valuation ring
.
RV of v must be L[x]XL[Xl , and therefore, k(t) = L Rv/xRv K. So

we assume now that v is not trivial on L. Then the residue class field of
RV n L is algebraic over k , and therefore, we can apply Theorem 3.A.1 to
Rv (because K is not algebraic over k) , and we see that K = k'(to) (k' is
algebraic over k) . But k is algebraically closed in k(t, x) , and therefore,
algebraically closed in K , too. Thus k' = k , and K = k(to) ^_' k(t) . Q.E.D.
Another application is on antirational extensions defined below. We say
that a field K is antirational over its subfield k if it does not happen that
there are a finite algebraic extension K' of K , an intermediate field K"
between k and K', and an element t such that K' = K"(t) and t is
transcendental over K" . Namely, K is not antirational over k if there are
such K', K", and t.
THEOREM 3.A.3. Assume that afield K is antirational over k and that L
is a finitely generated field over k. If there is a transcendental element t over
L such that L(t) D K, then K is a subfield of L.
PROOF. We shall obtain a contradiction by assuming that K % L. Since
L is algebraically closed in L(t) , K(L) is transcendental over L , and L(t)
is algebraic over K(L) . Therefore, we can choose a transcendence base
x1 , ... , xr of L(t) over K from elements of L. Let vo be a discrete
valuation in the weak sense of rank r of the field K(x1 , ... , x,.) such that
its valuation ring Ro has prime ideals {0} c P1 c : c P and xl E P ,
x, 0 P _j for each i. Note that Ro is obtained as follows
(Ro)P = K(x2 , ... , xr)[x1 ]X K (x , ... , X )[X , and (R0)P is the composite of
z
i i r
i 1
(R0)pI and the similar discrete valuation ring of K(x2 , , ... , Xr) , and so on.
Let v be an extension of vo to L(t) , and let v' be the restriction of
v to L. Then v' is of rank r because the residue class field of vo is K
and since x; E L. Since trans. degk L(t) = trans. degk K + r, and since
trans. degk L > trans. degk (the residue class field of v') + rank v' (cf. Exer-
cise 4.9.6), the residue class field of v (which is a finite algebraic extension
of K ; cf. Theorem 4.9.6) is not algebraic over the residue class field of v'.
Therefore, the residue class field of v is of the form k'(?) with a field k'
containing the residue class field of v' and with t transcendental over k' ,
a contradiction. Q.E.D.
THEOREM 3.A.4. Let L be a finitely generated field over a field k. If
intermediate fields K1, K2 are antirational over k , then the field K1(K2) is
also antirational over k.
PROOF. Suppose for a moment that K3 = K1(K2) is not antirational over
k. Then there is a finite algebraic extension K3 of K3 such that K3 = k'(t)
with a field k' containing k and a transcendental element t over k'. By
the theorem above, Kl CK3 = k'(t) implies K, c k', which shows that
K3 is not algebraic over K3 = K1(K2) , a contradiction. Thus, K1(K2) is
antirational over k. Q.E.D.
COROLLARY 3.A.5. In L above, there is the largest subfield among those
which are antirational over k.
It is difficult to see if a given field extension is antirational or not. We add
here some known cases without proof.
(1) When k is algebraically closed, L is finitely generated over k and
trans. degk L = 1 , we have that L is not antirational over k if L = k (x)
for some x.
(2) When L is a regular extension of a field k and trans. degk L = 1 , we
have that L is not antirational over k if either L = k(x) or L = k(x, y)
with x, y such that axe + bye = 1 (a, b E k).
(3) When k is an algebraically closed field of characteristic 0 and
trans. degk L = 2, we have that L is not antirational over k if L = K(x)
with a field K such that trans. degk K = 1 .
Exercises
1
1. Prove that if a field L is generated by a , , ... , an over a field K , then a

subset of { a , ... , an } can be chosen as a transcendence base of L over
1
K. Consequently, trans. degK L < n.

2. Let K C L C M be fields. Prove that if either trans. degK M is finite
or both of trans. degK L and trans. degL M are finite, then these three
transcendence degrees are finite and
trans. degK M = trans . degK L + trans. degL M M.
3. Let K C- F be fields, and let L, M be intermediate fields. Prove that
if trans. degK L and trans. degK M are finite, then
trans. degK L(M) < trans. degK L + trans. degK M .
EXERCISES 1 27
1. Let M, N, M' , N' be modules over a field K, and let 0: M -> M',
v: N -+ N' be K-homomorphisms. Then we define the tensor product
of q and y/, which is denoted by q V, as a mapping of M (9 N to
M' N' determined by
V) mi n _ 0(m,) yr(n1) .
(i) Prove that 0 yr is a K-homomorphism.

(ii) Prove that if 0', yr' are K-homomorphisms from M', N' to some
modules, respectively, then
(0' (g w')(0 (D w) (0,
2. Let M be a ring such that M = K[a1 , ... , a field. Let N be

a ring containing K. Prove that if I is the relation ideal of (al , ... , an )
in the polynomial ring K[X1 , ... , Xn], then
MKN^='N[X1,...,Xn]/IN[X1,...,Xn].
Prove also that if N = K[b1 , ... , bm] and J is the relation ideal for
(b1, ... , bm) in K[Y1 , ... , Yin], then using
K[X, Y] = K[XI, ... , Xn, Y1, ... , Yn] ,
we have that M K N ^_' K[X, Y]/(IK[X, Y] + JK[X, Y]).
3. Let K c F be fields, and let M, N be intermediate fields. Show the
following implications between the five conditions below: (i), (ii), (iii)
are equivalent; they imply (iv), and they are implied by (v).
Show also that (iv) implies (iii) is false in general.
(i) L, M are linearly disjoint over K.
(ii) If a1 , ... , an (E L) are linearly independent over K, then a1, ... ,
an are linearly independent over M.
(iii) If bl , ... , b (E M) are linearly independent over K then b1 , ... ,
bn are linearly independent over L.
(iv) LnM=K.
(v) L n M = K, and M is a Galois extension of K.
4. Prove that if K is a subfield of a field L and M, N are K-modules,
then (M K N)KL(MKL)L(NKL).
1. Let L be a finitely generated field over a field K of characteristic p 0 0.

Prove that a p-base of L over K consists of a finite number of elements.
2. Prove that if a derivation D of a field K can be extended to a derivation

of a field L containing K and if M is an intermediate field, then D
can be extended to a derivation of M.
3. Let M be a module over a ring R. A mapping D of R to M is
called a derivation of R with values in M if it satisfies the conditions:
D(f+g) =Df+Dg, D(fg) = f(Dg)+g(Df) (df, g c R). If S C R
and DS = {0} , D is called a derivation over S.
Try to generalize the results on derivations of a field to this case.
1. Let K be a subfield of a field L. Prove that the following statements

(i)-(iii) are equivalent.
(ii) For any finite number of elements a1 , ... , an of L , K(a1 , ... , an)
is separable over K.
(iii) L K K' 00 is a field.
2. Let L1 , L2 be separable extensions of a field K. Prove that the field
L1(L2) is separable over K if
trans. degK L1(L2) = trans. degK L1 + trans. degK L2 < oo .
3. Let L be a finitely generated field over a field K. Set K1 = KA n L,

and let x1 , ... , xn be a transcendence base of L over K. Prove that
[K1 : K] < i(L/K) < [L: K(xl , ... , xn)] .
1. Let t, u, x, y be algebraically independent elements over a field k of

characteristic p 0, and set K = k(t, u), Lo = K(x, y). Let z be a
root of the equation V + tx'' + uy" = 0 with coefficients in LO) and set
L = K(x, y, z). Prove that K is algebraically closed in L but L is not
a regular extension of K.
2. Prove that an extension L of a field K is a regular extension if and only
if L K K' is an integral domain whenever K' is either a finite separably
algebraic extension or a finite purely inseparable extension of K.
3. Let L be a finitely generated field over a field K. Prove that there is
a finite algebraic extension K' of K satisfying the following two condi-
tions:
(i) The separable closure Ks of K in K' is contained in L.
(ii) L(K') is a regular extension of K'.
EXERCISES 129
4. Consider extension fields L = K(a1 , ... , a,,,), M = K(b1 , ... , of a

field K. Let I, J be the relation ideals for (a1 , ... , a,,,) , (b1 , ... ,
respectively, in polynomial rings K[X1, ... , X 7] , K[Y1, ... , Yt]. Prove
that the ideal generated by I, J in K[X1 , ... , X,n, Y1 , ... , Y,] is a
prime ideal if one of the following two conditions is satisfied.
(i) One of L, M is a regular extension of K.
(ii) L is separable over K, and M is purely inseparable over K.
Note that from this result we have that if K is algebraically closed, then
the ideal in K[X1 , ... , X,n, Y1 , ... , Y ] generated by prime ideals P, Q
of K[X1 , ... , X,n], K[Y1 , ... , Yt] (respectively) is a prime ideal.
1. Let M be a module over a ring R. Assume that IM = {0} with an

ideal I of R. Show that we can naturally regard M as an R/I-module
and R-submodules of M as (R/I)-submodules of M.
2. Let N be a submodule of a module M over a ring R. Prove that M
is a Noetherian module if both M/N and N are Noetherian modules.
3. Show that the following statement (*) is false, that (*) becomes true if
we add an additional assumption "there is a natural number n such that
In = {0} ", and that the converse of (*) is true.
(*) If an ideal I of a ring R is finitely generated and R/I is a Noethe-
rian ring, then R is a Noetherian ring.
4. Show that the ideal I = x2P + xyP in the polynomial ring P = K[x, y]
over a field K is not a primary ideal, but its radical is a prime ideal.
5. Prove that if M is a maximal ideal of a ring R, then any ideal I that
satisfies M' C I C M (for some t) is an M-primary ideal. (Note
that if we change "maximal" to "prime", then exercise 4 gives a counter
example.)
6. Let I and P be an ideal and a prime ideal, respectively, of a Noetherian
ring R. Prove that P is a prime divisor of I iff for some b E R,
I:bR=P.
7. If a ring R has additive subgroups R, (i = 0, 1, 2, ...) such that R
is the direct sum of these R1 (as an additive group) and R.RI c Ri+1
(for all i, j) , then R is called a graded ring. Elements of R are called
.
homogeneous elements of degree i. An ideal I is called a homogeneous

ideal if I = Ej_o(I n R.) Prove, under the circumstances, the following
.
statements.
(i) Ro is a subring of R.
(ii) M = 1:1 R, is an ideal of R. If M is generated by a finite
number of homogeneous elements f1 , ... , fN, then R is generated
by f1 , ... , fN over Ro as a ring.
(iii) R is a Noetherian ring if Ro is a Noetherian ring and M is finitely

generated as an ideal.
(iv) For homogeneous ideals P, Q, P is a prime ideal and Q is P-
primary if
(1) P is generated by all of the homogeneous elements in
(2) Whenever f , g are homogeneous elements with f9 E Q and
f 0 P, then g E Q.
(v) If R is Noetherian and if I is a homogeneous ideal, then I is
expressed as the intersection of homogeneous primary ideals and
every prime divisor of I is homogeneous.
8. Let S be a subset of R consisting only of non-zero-divisors. Let 0 be
the natural homomorphism from R to R/I with an ideal I. Show that
Rs/IRs ^_' R/I if every element of cbS has its inverse in R/I.
1. Let R, S be as in Corollary 3.7.13. Prove that if I is an ideal of S,

then ht I = ht IR .
2. Generalize the statement of Theorem 3.7.12 as follows. Drop the as-

sumption that R is an integral domain and add the condition that every
nonzero element of S is not a zero-divisor.
3. Let S be a subring of a ring R such that R is integral over S. Prove
that if T is a subset of S consisting only of non-zero-divisors in R,
then RT is integral over ST S.
4. Assume that a ring R is integrally closed in its ring of total fractions Q of
R. Show that if T is a subset of R consisting only of non-zero-divisors,
then RT is integrally closed in Q .
5. Show by an example that an irreducible monic polynomial f (x) = x" +
xn-1
cI + + c" with coefficients in an integral domain R may not be
irreducible as a polynomial over the field of fractions K of R. Prove
also that if R is integrally closed in K, then a monic polynomial with
coefficients in R is irreducible in R[x] if it is irreducible in K[x].
6. Let p be a prime element in an integral domain R. Prove first that
if a polynomial f(x) = cox" + clan-I + . + cn (c, E R) satisfies the
. .
conditions that co pR, c, E pR (i = 1 , ... , n), c" 0 p2R , and that

co , ... , cn have no common factor (except for invertible ones), then f (x)
is irreducible in R[x]. Prove next that if R is integrally closed in its
field of fractions K , then f (x) is irreducible in K[x] . This is called the
irreducibility theorem of Eisenstein.
[Hint. For the latter half, note that p is a prime element in R[co I ]
and use exercise 5.]
7. Prove that any unique factorization domain is an integrally closed integral
domain.
EXERCISES 131
1. Show that, in Theorem 3.8.2, if K is an infinite field, then the condition

(3) can be replaced by
n
Y+, = >2 c11X3 (c1j E K) . (3')
j=1
Then, apply this result to Corollary 3.8.3.

2. Let P be a prime ideal of a Noetherian integral domain R. Prove that
if aR (a E R, aR # R) contains P, then either P = aR or P = 101.
Next apply this result to the polynomial ring A = K[XI , ... , Xn], and
prove that every prime divisor P of a principal ideal aA (# 0, # A)
has height 1 (namely, the last step of the proof of Theorem 3.8.7).
3. Let P be a prime ideal of the polynomial ring K[X1 , ... , Xn ] . Prove
that if P is a maximal ideal, then P is generated by n elements. Prove
also that, if ht P = r, then in the ring K[X] p = { f / g I f , g E K [X ] , g
P1, PK[X] p is the unique maximal ideal and generated by r elements.
4. Let R, R' be finitely generated integral domains over a field K such
that R C_ R' and R' is integral over R. Prove that if I' is an ideal of
R' , then ht I' = ht(I' n R) .
1. Let R be a Noetherian integral domain with field of fractions K. Let

h be an element of K. Prove that if either (i) there is a E R such that
a 0 0, h n a E R for all n = 1, 2, ... , or (ii) there is an ideal I ( {0})
such that hI c I, then h is integral over R.
2. Let R be a Noetherian integral domain with field of fractions K. Let
R* be the integral closure of R in K and set C = {x E RlxR* C R} .
Prove the following statements.

(i) C is the largest among those which are ideals of both R and R* .
(ii) R = R* iff C = R.
(iii) R* is a finitely generated R-module if C 54 101.
This C is called the conductor of R* over R.
10
1. Prove that if 92 is an algebraically closed field and if A 1 , A2 are algebraic

sets in Sn (S2) , then A 1 n A2 is also an algebraic set and I (A I n A2) _
VII (A I) + 1(A2). On the other hand, show that if 0 is not algebraically
closed, then, though Al nA2 is an algebraic set, the equality I(A1 nA2) _
I -+I can be false.
2. Consider S" (S2) with an algebraically closed field Q. Let V be an al-
gebraic variety in S' (Q) of dimension r > 0. Prove that if H is a
hypersurface in S" (S2) and if H n V (the empty set), then the dimen-
sion of each irreducible component of H n V is r - 1 or more.
3. Prove that, in the affine space S" (K) over an arbitrary field K , the
minimum condition on algebraic sets holds good.
4. Let A be an algebraic set in the affine space S" (S2) over an algebraically
closed field K2, and let K be a subfield of 0. We say that A contains
sufficiently many K-rational points, if A is the smallest algebraic set
containing the set of K-rational points in A .
Examine whether the following quadratic hypersurfaces in S" (C) have
sufficiently many R-rational points, where C and R are the complex
number field and the real number field, respectively.
(i) c1X1+ +CrXr = Cr+1X+1 + +CSXs +1 (C1 E R, ci > 0, s > 1 ,
0<r<s<n),
(ii) 1X1
+...+CrXr2 = 2 2
Cr+1X+1+...+Cs-1Xs I+CsXs (C! E R, c > 0,
0<r<s<n).
5. Let A be an algebraic set in S" (S2) with an algebraically closed field Q.
Assume that A is normally algebraic over a subfield K of Q. Prove
that if Vl , V2 are.components of A which are conjugate over K, then
there is a Q E G(K/K) (K being the algebraic closure of K) , such
that, under an extension a' of a to an element of AutK 92, we have
a' Vl = V2. Prove also that u' Vl = V2 for any extension a' of a.
6. Let A and V be an algebraic set and an algebraic variety, respectively,
in S" (S2) . Prove that if V is contained in A , then V is contained in
an irreducible component of A.
7. Prove that if A, B are algebraic varieties in S"' (0), S"(), respec-
tively, then dim A x B = dim A + dim B.
11
1. Prove that every subfield of the real number field does not satisfy any
Cl-condition.
2. For each i = 1, 2, ... , give an example of a field K such that K is
not a Cl-field, and such that there is a finite algebraic extension which is
a C!-field but not a C1_ 1-field, for each i = 1 , 2, ... .
12
1. Consider the mapping of a simple transcendental extension K(x) of a

EXERCISES 133
field K to itself defined by x -* (cx + d)/(ax + b) with a, b, c, d E

K such that ad 54 bc. Prove that this mapping is a K-automorphism
of K(x) and that every K-automorphism of K(x) is given by such a
mapping with suitable a, b, c, d.
2. Let L = K(x, y) be an extension field of a field K of characteristic 54 2,
where x, y are defined by the relation axe + bye = c with a, b , c E K
such that abc # 0. Prove that there is a t E L such that L = K(t)
if either the curve ax2 + by2 = c on the affine plane has a K-rational
point or the equation ax2 + by2 = 0 has a nontrivial solution in K (or,
equivalently, the equation ax2 + by2 = cze has a nontrivial solution in
K) .
3. Let L = K(x, y, z) be an extension field of an algebraically closed field

K of characteristic 54 2, where x, y, z are defined by the equation
f1(z)xe + fe(z)ye = f3(z) (0 f(z) E K[z]). Prove that there is an
element w such that L = K (z , w).
1. Assume that K, K', ko are finitely generated extension fields of a

field k and that there are elements t, xI , ... , x,. such that K' = ko(t),
K(xI , ... , xr) = K'(xl , ... , xr) Prove that if t is transcendental over
.
ko and if x1 , ... , xr are algebraically independent over both K and

K' , then there is a field k1 (D k) and a transcendental element t1 over
k1 such that K = k1(t1) .
2. Prove (ii) below, assuming that (i) is known. (These results (i), (ii) are
called the Theorem of Zariski-Castelnuovo.)
(i) If x , y are algebraically independent elements over an algebraically
closed field k and if L is a field such that k c L c k(x, y) and
such that k(x, y) is separably algebraic over L, then L is a purely
transcendental extension of k.
(ii) Let x1 , ... , xn be algebraically independent elements over an alge-
braically closed field k of characteristic 0. Prove that if L is a field
such that k c_ L c k(x1 , ... , xn) and such that trans. degk L = 2,
then L is a purely transcendental extension of k.
Note that the proof of (i) is very difficult. [Hint. Prove first the
following theorem, and then apply (i).
THEOREM. Let x1, ... , xn be algebraically independent elements over an
infinite field K. If L is a field such that
K c L c K(x1 , ... , xn) and trans. degK L = r ,
then there are algebraically independent elements y1 , ... , yr over K such
that L C K(y1, ... , yr).]
CHAPTER IV
Valuations
1. Multiplicative valuations
If a mapping v of a field K to the set R of real numbers satisfies three
conditions (1)-(3) below, then v is called a multiplicative valuation, or sim-
ply, a valuation of K. {vala E K, a 0 0} forms a multiplicative group
consisting only of positive real numbers and is called the value group of v .
(1) va>0 forevery aEK,and va=O iffa=0.

(2) v(ab) _ (va)(vb) for all a, b E K.
(3) There exists a positive number C such that v(1 + a) < C for any
aEK with va< 1.
Note that if condition (2) is satisfied, then condition (3) is equivalent to
the condition that v(a + b) < C(vb) whenever va < vb.
We say that a valuation v satisfies the triangle inequality if v satisfies
(3*).
(3*) v(a + b) < va + vb for all a, b E K.
Note that if v satisfies (3*) , then (3) is satisfied with C = 2. Thus, (3*)
is stronger than (3).
If v satisfies (3**) below, which is stronger than (3*) , then we call v a
non-Archimedean valuation; otherwise, v is called an Archimedean valuation.
(3**) v(a + b) < max{va, vb} for all a, b E K.
EXAMPLE 1. If there is an isomorphism 0 of a field K into the complex
number field C, then va = Ical (I stands for the absolute value) defines a
I
typical Archimedean valuation. (In 6, we prove a theorem which shows the

converse of this example).
EXAMPLE 2. v defined by v0 = 0, va = 1 (for a 0) is a non-
Archimedean valuation and is called the trivial (multiplicative) valuation.
EXAMPLE 3. The following is a typical example of a non-Archimedean
valuation. Let R be a Noetherian integral domain with field of fractions K,
and assume that p is a prime element of R. We fix a real number e that is
greater than 1. We define a valuation v as follows. v0 = 0, for 0 54 a E K,
va = e-" , where n is such that a = p"c/b (n E Z (the ring of rational
integers), b, c E R, and b, c are not divisible by p).
PROOF. (i) To prove the uniqueness of n , suppose that a = p"'c'/b' with
135
136 IV. VALUATIONS
b' , c' similar to b, c above, then p"c/b = p'"c'/b'. Assume for a moment
that n > m. Then bc' = p cb' E pR, which contradicts pR being a
prime ideal. Therefore m = n.
(ii) Next we check the conditions for a valuation are satisfied. It is clear
that (1) and (2) are satisfied. Assume that va = e-" < vb = e-'" . Then
a = p"c/b , b = p'"c'/b' (b, c, b', c' are not divisible by p) . Since e > 1
we have n > m and a + b = p'" (p"-'" cd' + d c') /d d' . Therefore, a + b =
prC"/b" with r > m and v(a + b) = e-r < e-'" = vb. Q.E.D.
The valuation v obtained above is called the p-adic valuation.
In Example 3 above, to change the value of e is analogous to changing
the unit of length in a scale; hence, is not essential. To clarify the situation,
we define the notion of an equivalence as follows.
Two valuations v1 , v2 of a field K are equivalent to each other if there
is a positive real number t such that v2 = vi (this equality means that
v2a = (vta)' for every a E K). In Example 3, we obtain another valuation
V' , by changing e to another e' (> 1) Then e' = e' with t = loge e' and
.
V' = v` . Thus, v' is equivalent to v . It is easy to see that valuations of the

field K in Example 3, which are equivalent to v , are obtained by suitable
values of e.
If v is a valuation of a field K and if t > 0, then v' is obviously a
valuation of K (the constant C in (3) is changed to C').
LEMMA 4.1.1. If v is a multiplicative valuation of a field K, then v 1 = 1 ,
v(-a) = v(a) (for all a E K) and v(a-') = (va)-1 (for a 0 0, a E K). If

v is non-Archimedean, then va < vb implies v(a + b) = vb.
PROOF. vl = v(12) = (V1)2. Since v1 > 0, we have v1 = 1. v(-1)2 =
v l = 1 . Since v(-1) > 0, we see that v(-1) = 1 , and v(-a) = v(-l)va =
va. v(aa-1) = v1 = 1 and v(a-') = (va)-' . Assume that v is non-
Archimedean. It is obvious that va < v b implies v (a + b) < v b by (3**).
vb = v(a + b - a) < max{v(a + b), v(-a)} = max{v(a + b), val. Since
va < vb, we have vb < v(a + b) and therefore vb = v(a + b). Q.E.D.
THEOREM 4.1.2. Two multiplicative valuations v, v' of a field K are
equivalent to each other if and only if it holds that
(i) va < 1 iff v'a < 1 (for all a c K) .
If v is not trivial, then the following condition is (necessary and) sufficient:

(ii) a E K, va < 1 implies v'a < 1 .
PROOF. It is obvious that both (i) and (ii) are necessary conditions. The
sufficiency of (i) follows from that of (ii), and therefore, it is enough to show
the sufficiency of (ii). Take an element a 0 0 such that va < 1 and set
e = va, e' = v'a , t = loge, e. Then we consider v" instead of v'. Thus,
we have only to show that v = v' assuming that va = v'a < 1 . Let b be
an arbitrary nonzero element of K.
1. MULTIPLICATIVE VALUATIONS 137
(1) The case vb < 1 . Consider I = {(r, s)l(vb)s < (va)r, (vb)s > v(a)'.+1
(va)r/s.
with natural numbers r, s} . Then vb <
(va)(r+1)/s <
Therefore,
vb = lims , er/S . On the other hand, since v(ar+1) < v(bs) < v (d), we
have v(bs/ar) < 1 and v'(bs) < v'(ar) by (ii). Similarly, we have v'(ai+2) <
v'(bs) , because v(ar+2) < v(ar+1) < v(bs). Thus, < v'b < (v'a)('+2)/s
, and v'b = lims--,..

(v'a)r/s
= vb.
er/s
(2) The case vb > 1. v(b-I) < 1 and v'(b-I) = v(b-I) by (1). There-
fore, v'b = vb.
(3) The case vb = 1. v(ab) < 1 and v'(ab) = v(ab) by (1), which
implies (v'b)(v'a) = (vb)(va), and therefore, v'b = vb. Q.E.D.
Hereafter in this section, v denotes a valuation of a field K and C
denotes the constant in condition (3).
LEMMA 4.1.3. If a1 , ... , an E K, vai < e E R (the real number field),
and n < 2s with s a natural number, then v(al + + an) < Cse.
PROOF. We use an induction argument on s. If s = 1 , then the assertion
is obvious. Assume that s > 1 . Set m = 2s- I . Then v (al + + a,n) <
Cs-1 e, v (a,n+l + + an) < CS-1 e , and therefore, v (al + + an) < C
CS-le = Cse. Q.E.D.
LEMMA 4.1.4. Assume that C < 2. Then v n < n for every natural number
n (where n in vn means n-times of the identity). More generally, if vai < e
for every i = 1, ... , n, then v (a 1 + + an) < ne.
PROOF. It suffices to prove the latter half, because the former half follows
from the case where all the ai are 1. For each natural number t, we take
a natural number s(t) such that 2s(t)-1 < n' < 2S(') Then (a1 + +
is the sum of n` monomials in al , ... ) an , and each monomial is of the
form a"-..an (E ti =t). v (a;' an-) < e` and Lemma 4.1.3 implies that
v(a1 + + an)' = v((aI + + an)') < 2S(')e' . Therefore, v(aI + + an) <
2s(t)/te
and v(aI + + an) < limt,. 2S(')1'e = e = ne. Q.E.D. 2"92n
THEOREM 4.1.5. If C < 2, then v satisfies the triangle inequality. Con-

versely, if v satisfies the triangle inequality, then we can choose C to be
2.
PROOF. The latter assertion is obvious, and we shall prove the former (n.)an-r1
half. For a, b E K, we set d = va, e = vb. (a + b)n ='r

with binomial coefficients (nr) v ((nr) a
.
n -b)
r r
< (nr) dn-r er by Lemma 4.1.4,
and we have v(a + b)n = v((a + b) n) < (n + 1) . max{(n)dn-rer} < (n + 1)
(n)dn-rer
= (n+ 1) (d +e)" Therefore, v(a+b) < "fn + 1 (d +e) Thus,
. .
v(a+b)<lim,, "n+1 (d+e)=d+e. Q.E.D.

COROLLARY 4.1.6. For any given valuation v, there is a valuation that is
equivalent to v and satisfies the triangle inequality.
138 IV. VALUATIONS
PROOF. v` is the required valuation whenever the positive number t sat-

isfies C` < 2. Q.E.D.
Now we give a simple theorem which characterizes non-Archimedean val-
uations.
THEOREM 4.1.7. The following four conditions are equivalent.
(i) v is a non-Archimedean valuation.
(ii) C can be chosen so that C = 1.
(iii) For every natural number n, v n (or, more precisely, v (n x identity)) <
1
(iv) The restriction of v to the prime field is a non-A rchimedean valuation.

REMARK. Theorem 4.2.1 and Corollary 4.2.3 in 2 also distinguish non-
Archimedean valuations and Archimedean valuations.
PROOF. The equivalence between (i) and (ii) is obvious. That (i) implies
(iii) is also clear. We shall prove that (iii) implies (i). By considering v` with
a suitable real number t, we may assume that C < 2 (note that vn < 1 if
(vn)' < 1). Now we can follow the proof of Theorem 4.1.5, and we apply
our condition so that the values of all binomial coefficients are < 1 In the .
case d > e , we have v(a + b)" < (n + 1) max{d"-'e'} = (n + 1)d" Then .
v(a + b) < "fn + 1 d . Therefore, v(a + b) < d , which shows (i). Thus, (i),
(ii), and (iii) are equivalent to each other. Since condition (iii) is a condition
only on the prime field, we see that (iv) is also equivalent to the conditions
above. Q.E.D.
COROLLARY 4.1.8. Any field of characteristic p 0 has no Archimedean

valuation.
PROOF. n 1 is an element of the prime field and (n 1)p-1 is either 1
or 0. Therefore, (vn)p-1
< 1 and vn < 1 . Q.E.D.
2. Valuations of the rational number field

We have seen in Theorem 4.1.7 and Corollary 4.1.8 that if v is an
Archimedean valuation of a field K, then its restriction to the prime field
n is an Archimedean valuation and n is of characteristic 0; hence, it is
isomorphic to the rational number field Q. In view of this fact and also for a
better understanding of the notion of valuations, we shall classify valuations
of the rational number field Q.
THEOREM 4.2.1. Let v be a valuation of the rational number field Q.
(i) If v is an Archimedean valuation, then v is equivalent to the absolute
value I I
.
(ii) If v is a nontrivial non-Archimedean valuation, then v is equivalent

to the p-adic valuation defined by a prime number p in the ring of rational
integers Z (Example 3 in 1).
2. VALUATIONS OF THE RATIONAL NUMBER FIELD 139
Namely, valuations of Q are substantially covered by the absolute value

, p-adic valuations, and the trivial valuation.
Before proving the theorem, we introduce the notion of a valuation ring

(which will be discussed in detail in 7).
LEMMA 4.2.2. If v is a non-Archimedean valuation of a field K, then RV =
{a E KIva < 1} is a ring with unique maximal ideal My = {a E Kjva < 11.
This RV and My are called the valuation ring and the valuation ideal of
v , respectively.
PROOF. The conditions (1), (2), (3**) on non-Archimedean valuations
obviously imply that RV is a ring and My is an ideal of R,, . If va = 1 ,
then v(a-1) = 1 , and hence, a is invertible in RV Conversely, if a is
.
invertible in RV , then v a < 1 , v (a-') < 1 , and therefore, v a = 1 . Thus,

My is the set of noninvertible elements in RV. Every ideal different from
RV consists only of noninvertible elements, and hence is contained in M2, .
Q.E.D.
PROOF OF THEOREM 4.2.1. (ii). Let v be a nontrivial non-Archimedean
valuation of the rational number field Q. Let Rv , My be the valuation ring
and the valuation ideal, respectively, of v Then the ring of rational integers
.
Z is contained in Rv (Theorem 4.1.7). Set P = Z n My. If P = 101,

then every nonzero element of Z is invertible in Rv ; hence, Q C Rv and
v is trivial, which contradicts our assumption. Since My is a prime ideal,
P is a prime ideal of Z. Thus, P is generated by a prime number, say p.
Let e > 1 be a positive number such that vp = e-1 . On the other hand,
each element a of Q is written as p" b/c (b, c E Z, b, c are not divisible
by p, n E Z). Since b, c are not in My , we have vb = vc = 1 , and
va = vp" = e-". Thus, v is a p-adic valuation.
(i) Let v be an Archimedean valuation of Q. By Corollary 4.1.6, we
can assume that v satisfies the triangle inequality. We fix a natural number
n > 1 Set e = max{ 1 , v n } . For each natural number in > 1 , we consider
.
the n-adic expansion of in

m,nt+m,_1n`-1
m= +- +in ,n+mo (m1 E {0, 1, ,.. , n-1}, in, # 0).
Then by Lemma 4.1.4, we have
vm < (t + 1) max{v(min`)} < (t + 1)(n - l)e` (nt < in < n`+').
The same argument is applied to m" , and we have v (m") < (t (u) + 1) (n -
1)e`(") , where t(u) E Z is defined so that nr() < m" < n`(" )+' . Therefore,
vm< 1-tu)+1
Note that limn t(u)/u = log,, in. Since t(u)/u < log,, in, we have
,
11--+00
" t(u) + 1 = lim " u V in + u- = 1 Thus, we have
.
V I et(u)1" = el og" 111

t/ --- 00
140 IV. VALUATIONS
If v n < 1 , then e = 1 and v m < 1 , contradicting the Archimedean

property. Thus, v n = e > 1 If v m < 1 , then taking m instead of n, we
.
get a contradiction as above. Furthermore, changing the role of m, n in

our observation above, we see that v m = d > 1 and v n = e < d Og"t n . Since
log. n = (loge m) - I , it follows that v m = d > eI g m . Therefore, we have
vm = elog,, m . Setting w = loge n , we consider vw instead of v . Then the
equality above implies v m = m . Thus, we have v a = a I for every a E Q .
Q.E.D.
COROLLARY 4.2.3. As for a valuation v of a field, v is non-Archimedean
iff there is a natural number n > 1 such that v n < 1 .
3. Topology
We say that a topology is defined on a set S if a family of subsets called
open sets are given so that the family S2 of open sets satisfies the following
conditions.
(i) SES2.
(ii) The empty set E S2 .
(iii) If U1, U2E92,then U1nU2ES2.

(iv) If Ui E S (i E A), then UiEA Ui E n.
A set S is called a topological space if a topology is defined on S. Each
element of a topological space is called a point. A subset F of a topological
space S is called a closed set if the complement S - F is an open set.
Consequently, the family S2* of closed sets satisfies the following conditions.
(i*) The empty set E 92*
(ii*) S E 92* .
(iii*) If F1 , F2 E S2* , then F1 U F2 E S2*

(iv*) If F E S * (i E A), then ni F E S2 .
REMARK. Readers who are familiar with open sets and closed sets in a
Euclidean space are advised to compare topological spaces with Euclidean
spaces.
Let M be a subset or a point of a topological space S. By an open neigh-
borhood, or simply, a neighborhood, of M, we mean an open set containing
M. (In some literature, a neighborhood of M means a subset containing
an open neighborhood of M. But, in this book, we consider open neighbor-
hoods only.) The family v(P) of all neighborhoods of a point P is called
the system of neighborhoods of P. The following properties are obvious:
(1) S E v(P)
(2) If UEV(P),then PEU.
(3) If Ul , U2 E v(P) , then U1 n U2 E v(P) ,
(4) If Ui E v(PP) (i E A), then UEA Ui E v(F) (for all j c A).
(5) If U E v(P) and Q E U, then U E v(Q).
THEOREM 4.3.1. The conditions (1)-(5) above.form a (necessary and) suf-
ficient condition for systems of neighborhoods. That is, if a family v(P) of
3. TOPOLOGY 141
subsets is associated to each element P of a set S and if these families v(P)

satisfy the conditions (1)-(5), then a topology on S, which makes each v(P)
the system of neighborhoods of P, is uniquely determined.
PROOF. Let SZ be {the empty set} U (UPEs v(P)). We shall show that
S2 satisfies the conditions (i)-(iv) on the family of open sets. The validity
of (i), (ii) is clear. Assume that U1 , U2 E fl. If U1 n U2 is the empty
set, then U1 n U2 E Q. If P E U1 n U2 , then Ul c v(P), and we have
U1 n U2 E v(P) C SZ . Thus (iii) is verified. (iv) follows from (4). Therefore,
we can give a topology on S which makes S2 the family of open sets. The
conditions (2) and (5) show that U E v(Q) if Q E U E S2, which means that
v(Q) is the system of neighborhoods of Q under this topology. Conversely,
if a topology is given on S such that v(P) is the system of neighborhoods of
P for each P E S, then by the definition of the systems of neighborhoods,
the family of open sets must coincide with 0, which proves the uniqueness
of the topology. Q.E.D.
A topological space S is called a Tl-space (i = 0, 1 , 2, 3, 4) if S satisfies
the condition T. below.
(To) Given P, Q E S, P # Q, there is a neighborhood U of P with
Q U, or there is a neighborhood V of Q with P 0 V.
(T1) Given P, Q E S, P # Q, there is a neighborhood U of P with
Q0 U.
(T2) Given P, Q E S, P# Q, there is a neighborhood U of P and a
neighborhood V of Q with U n V empty.
(T3) Given P E S, P 0 F (a closed set), there is a neighborhood U of
P and a neighborhood V of F with U n V empty.
(T4) Given closed sets F1 , F2 with F1 n F2 empty, there are neighbor-
hoods Ul of FF (i = 1, 2) with U1 n U2 empty.
A T2-space is often called a Hausdorff space. A space satisfying both T1
and T3 is called a regular space. A space satisfying both TI and T4 is called
a normal space.
These conditions Tl (i = 0, 1 , 2, 3, 4) and the conditions for regularity
and normality are often called axioms (or, conditions) of separation. Though
there are notions of completely regular and completely normal, we do not
treat them in this book.
THEOREM 4.3.2. For a topological space S, we have that S is a TI -space
iff {P} is a closed set for every P E S .
PROOF. To prove the if part, suppose that P Q, then U(P) = S - {Q}

is a neighborhood of P not containing Q.
To prove the only if part, note that for each point Q P, there is a
neighborhood UQ of Q not containing P. The union of such UQ is S -
{P} Since each U. is an open set, the union S- {P} is an open set; hence,
.
{P} is a closed set. Q.E.D.

142 IV. VALUATIONS
If a topology on S is defined as follows, then we call S a metric space

and p is called the distance function of the metric space.
(i) p is a mapping from the product set S x S to the real number field
R and satisfies the conditions p(P, Q) = p(Q, P) > 0 (for all P, Q E S)
with p(P, Q) = 0 iff P = Q.
(ii) p(P, Q) + p(Q, R) > p(P, R) (for all P, Q, R E S) (triangle in-
equality).
(iii) For each positive number e, we set UE(P) = {Qjp(P, Q) < E} . Then
U (C S) is an open set if P E U implies that, for some E > 0,
UE(P) C U.
Under these circumstances, UE(P) is called the E-neighborhood of P .

REMARK. If there is a function p satisfying conditions (i), (ii) above is
given, then, by defining open sets by (iii), S becomes a metric space with
distance function p.
Let X be a subset of a topological space S. Then, by letting QX =
{X n U1 U is an open set of S} be the family of open sets in X, we can
define a topology on X, which is called the induced topology on X, and X
is called a subspace of S.
THEOREM 4.3.3. Let X be a subspace of a topological space S. Then
(i) F (C X) is a closed set of X iff F = (a closed set of S) n X.
(ii) If S is a metric space, then X is a metric space.
(iii) If S is a Ti-space with i < 3 , then X is a T;-space.
(iv) If S is a metric space, then S is a normal space; if S is a normal
space, then S is a regular space; if S is a regular space, then S is a Hausdorff
space; if S is a Hausdorff space, then S is a T1-space, and, if S is a Tl-
space, then S is a To-space.
REMARK. (iii) does not generalize to include either a T4-space or a normal
space.
PROOF. (i) Set U = X - F. Then F is a closed set of X if U is an
open set of X, which is equivalent to U = (an open set of S) n X, and we
see (i).
(ii) is clear from the definition of a metric space.
(iii) If i < 2, then the implication is obvious. If i = 3, then the assertion
follows from (i) and the definition.
(iv) The assertions, except for the first one, are obvious by Theorem 4.3.2.
The first assertion is proved as follows. Let p be the distance function. Let
F1 , F2 be closed sets having no common point. For each P E F1 , we set
D(P) = {p(P, Q)jQ E F2} Let d(P) be the infimum of D(P), which is
.
called the distance between P and F2. We shall show that d (P) > 0. As-
sume for a moment that d (P) = 0 . Then there is a sequence {d } such that
d,, E D(P) and 1im,. d = 0. Take Q, E F2 such that d = p(P, Q) .
3. TOPOLOGY 143
Then for any positive number e, these points Q,, are in the 8-neighborhood
UE(P) for sufficiently large n. This means that every neighborhood of P
meets F2, which contradicts the assumption that P is not in the closed set
F2. Therefore, d (P) > 0.
Set U1 = UPEF, Ud'(P)(P) with d'(P) = d(P)/2. In the same manner,
we take U2 for F2. Then U1 , U2 are neighborhoods of F1 , F2, respec-
tively. We shall show that Ul , U2 has no common point. Assume that
Q E U1 n U2. Then there are Pi E F (i = 1, 2), p(Pi, Q) < d(Pi)12,
and therefore p(P1, P2) < p(P1, Q) + p(Q, P2) < (d (P1) + d (P2))12 <
(p(P1, P2) + p(P1, P2))/2 = p(P1, P2), a contradiction. Q.E.D.
In a topological space S, a family Q0 of subsets is called a subbase of
open sets if the family of open sets of S is the smallest set containing 920
and satisfying the conditions (i)-(iv) on the family of open sets. Note that
if S is a set and if S20 is a family of subsets of S, then there is a unique
topology on S which has 00 as a subbase of open sets.
A mapping f of a topological space S1 to a topological space S2 is said
to be continuous if for any given P E S1 and any neighborhood U' of f P ,
there is a neighborhood U of P with f U C U'.
THEOREM 4.3.4. Let f be a mapping of a topological space S1 to a topo-

logical space S2. Then f is continuous iff whenever U' is an open set in S2 1
then f -' (U') = IQ E S1 I f Q E U' l is an open set in S1

.
PROOF. The if part is obvious, because, in the definition of continuity,

f ' (U') is a neighborhood of P, and we can take f- 1 (U') as the required
U To prove the only if part, let P E f -1(U') , then, since U' is a neigh-
.
borhood of fP, there is a neighborhood U(P) of P with f(U(P)) C U'

by the definition of continuity. Thus, U(P) c f -' (U') , and furthermore,
f-1(U') C UPEf-'(U') U(P) c f-'(U') . Hence, f-1(U') = U U(P) , and
therefore, f-'(U') is an open set. Q.E.D.
A mapping f of a topological space SJ to S2 is called a homeomorphism
if f gives a one-to-one correspondence between S1 and S2 and if both f
and f- are continuous. In this case, we say that S1 is homeomorphic to
I
S2. It follows that U (c Sl) is an open set of S1 if f U is an open set of

S2 .
Let Si (i E A) be topological spaces. We consider the product set S* _

[IiEA Si Let ni be the projection of S* to Si (namely, nt(jI aj) = a.) and
set S20 = {7r ' (U1) I Ul is an open set in S; , i E Al. Then S* above with
topology such that S20 is a subbase of open sets is called the product space
of these Si and is denoted by the same symbol as the product set. (If A
consists of finitely many elements, say, , ... , n , then the product space is
1
denoted by S1 x x S,t , as in the case of product set.)

144 IV. VALUATIONS
THEOREM 4.3.5. Open sets in the product space S* = S1 x x Sn are

characterized as the subsets expressed as the union of certain sets of the form
Ul x . . x Un with open sets Ui in Si.
.
PROOF. Ul x x Un = Ii (S1 x
I
x Si_ 1 x Ui X Si+1 x . . . X Sn) , and
Ul x x Un is an open set. Therefore the union of subsets of this form is
an open set. Conversely, we consider the family 0 of open sets of S* . The
subbase of open sets of the topological space S* is the family E20 consisting
of subsets of the form Sl x x Si_1 X Ui X Si+1 x . . X Sn . Hence, if
.
S2' is the family of subsets that are expressed as the union of sets of the
form U1 x x Un with open sets Ui, then S20 C Q'. On the other hand,
if U, V E Q', then U, V are expressed as U = Ui U1 i x x Uni, V =
U j V, j x x Vn j with open sets Uki , Vk j in Sk . Thus,
U n V= U(U1i x ... X Uni n Vl j x ... x Vl j)
ij
=U n(S1x...xSk-1xUkixSk+lx...XSFl)
k
n(S'1 X...XSk-1 xVkjXSk+I X...XSO
y (Usi x...xS_1 x(UkifVkj)xSk+I x...xS,t
Ulinvj)x...X(UninVnj)ES2'.
U(
Consequently, S2' contains S20 and satisfies conditions (i)-(iv) on open
sets. Therefore, S2' = K2. Q.E.D.
For a given subset X in a topological space S, the intersection X of all
closed sets containing X is the smallest closed set that contains X and is
called the closure of X. If the closure of X coincides with S, then we say
that X is dense in S.
THEOREM 4.3.6. Under these circumstances, a point P E (the closure X
of X) iff every neighborhood U(P) of P meets X.
PROOF. P X if there is a closed set F with P V F D X. This, in
turn, is equivalent to the condition that there is a neighborhood U(P) of P
such that U(P) n X is empty. Q.E.D.
Consider the case where two topologies p1 , p2 are defined on the same
set S. We denote by Si (i = 1, 2) the topological space S with topology
pi . If the identity mapping f : x -* x gives a continuous mapping of SI to
S2, (or, open sets under p2 are also open sets under pl ,) then we say that
the topology p1 is stronger than the topology p2 (or, precisely speaking, we
should say "stronger or equal") and that p2 is weaker than p1 The strongest
.
4. TOPOLOGICAL GROUPS AND TOPOLOGICAL FIELDS 145
topology on S is the one such that every subset of S is an open set. This
topology is called the discrete topology. The weakest topology is such that
the only open sets are S and the empty set. Then, obviously, S is not even
a To-space (provided that #(S) > 1) ; hence, this topology is not important.
Among further important notions on topological spaces, we mention only
those of compactness and completeness. As for completeness, we shall study
it later in 5. Here we shall define the notion of compactness.
A topological space S is said to be compact if S satisfies the conditions
in the following theorem.
THEOREM 4.3.7. In a topological space S, the following two conditions are
equivalent.
(i) If U1 (i E M) are open sets and if U1 UI = S, then there are a finite
number of elements i , ... , i,1 of M such that S = U;-1 U,
1 .
(ii) If Fl (i E M) are closed sets and if, for any finite number of elements
nn=1
i1 , ... , in of M, the intersection FJ is not empty, then nIEM F is not
empty.
REMARK. In some literature, compactness is defined with the additional
condition that the space should be a Hausdorff space.
PROOF. To prove that (i) implies (ii), set V = S - F If U; V = S, then
.
by (i), S = U;=1 VJ , that is, we have FJ is empty. Thus, U; V # S

and n; F, is not empty.
That (ii) implies (i) is proved similarly. Q.E.D.
One familiar example of a compact space is a bounded closed set in a Eu-
clidean space, where the condition (i) corresponds to the covering theorem of
Heine-Borel. Another easy example of a compact space is a space consisting
of a finite number of points with the discrete topology.
Incidentally, we add one notion. A family F of subsets of a set S is said
to have the finite intersection property if any finite number of members of F
have a common element. (ii) above can be paraphrased by saying that if a
family F consisting only of closed sets has the finite intersection property,
then there is a point common to all members of F.
4. Topological groups and topological fields
Consider the case where a topology is defined on a group G. We say
that the multiplication in G is continuous if the mapping f of G x G to
G given by f (a, b) = ab is continuous as a mapping of the product space
G x G to the topological space G. This is equivalent to the condition that for
any a, b E G and any neighborhood U(ab) of ab, there are neighborhoods
U(a), U(b) of a and b, respectively, such that U(a) U(b) = {xylx E U(a),
y E U(b)} C U(ab). This definition also applies to additive groups, addition
and multiplication in a ring. (Of course, in the case of addition, we consider
a + b, x + y instead of ab, xy.)
146 IV. VALUATIONS
A group G, on which a topology is defined, is called a topological group if

G is a To-space, the multiplication is continuous, and the mapping 0 of G
to G given by qa = a-1 is also continuous.
THEOREM 4.4.1. Let G be a group with topology.
(i) if the multiplication is continuous, then for each g E G, the mappings
rg, 1g of G to G, such that rg(a) = ag, lg(a) = ga, are homeomorphisms
of G to G.
(ii) If G is a topological group, then the map 0 given by Oa = a-I is a
homeomorphism of G to G.
PROOF. rg and 1g are continuous mappings because of the continuity of
the multiplication. This also applies to rg- , , 1g- , and we see that rg , 1g
,
are homeomorphisms. As for 0, the assertion is clear, because 0-1 = .
Q.E.D.
REMARK. The converse of (i) above is false. For instance, consider the
additive group G of rational integers with weakest topology among those
satisfing the T1-condition. Namely, a subset U is an open set if either U
is empty or the complement of U is a finite set. Then rg(= 1g) maps the set
of open sets onto itself, and therefore, rg is a homeomorphism. But, for any
two neighborhoods U1 , U2 of 0, we have that U1 + U2 = G, and therefore,
the addition is not continuous.
(Another proof. Obviously, this G is not a T2-space. If the addition is
continuous, then we see easily that G is a topological group. But, as will be
shown in the latter half of Corollary 4.4.2 below, every topological group is
a T2-space.)
COROLLARY 4.4.2. Let G be a topological group.
(i) For each a E G, we denote by o(a) the family of neighborhoods of a.
Then v(a) = v(1)a = ao(1), where v(1)a = {NaIN E v(l)} and similarlyfor
av(1).
(ii) G is a T2-space (a Hausdorff space).
REMARK. It is known that every topological group is a regular space (and,
furthermore, a completely regular space).
PROOF. (i) is obvious because ra , la are homeomorphisms.
(ii) If a, b E G with a b, then for some U E v(1) , a 0 b U or
b 0 a U (because of To-space), and hence for some V E V(1), VV' C_ U
(by continuity). Suppose that aVflbV is not empty. Then ay = bz for some
y-1 -1
y, z E V and therefore a = b z EbVV c W. Similarly, b E a U.
These contradict our choice of U. Therefore, G is a T2-space. Q.E.D.
Let G be a topological group. A family v of neighborhoods of 1 is called a
fundamental system of neighborhoods of the identity if it satisfies the condition
that if an open set U contains 1 , then V C U for some V E v V.
It is obvious that v(1) , the family of neighborhoods of 1 , is a fundamental
system of neighborhoods of the identity.
4. TOPOLOGICAL GROUPS AND TOPOLOGICAL FIELDS 147
THEOREM 4.4.3. If v is a fundamental system of neighborhoods of the

identity, then v has the following properties:
(i) l E U for all UEv .
(ii) If 1 # a E G, then a V U for some UEv.

(iii) If U1, U2 E v , then U c U1 n U2 for some UEv.
(iv) If U E V, then VV = {xyJx, y E VI c U for some VEv .
(v) If U E 0, then V-1 = E V} C U for some V E V.

{xIx-1
(vi) I f a, b E G, UEV, l E a Ub , then V c_ a U b for some VEv.

Conversely, if a family v of subsets of a group G satisfies these six condi-
tions, then we can define a topology on G so that G is a topological group in
which v is a fundamental system of neighborhoods of the identity.
PROOF. Consider the first half. (i) is obvious and (ii) follows from the
T1-condition (in view of Corollary 4.4.2, (ii)). (iii) follows since U1 n U2 is
an open set containing 1. (iv) follows from continuity, namely, considering
the case a = b = 1 in the definition, there are neighborhoods U1 , U2 of 1
such that U1 U2 C_ U , and for some V E v , we have V c U1 n U2. (v) follows
similarly from the fact that the mapping 0 (qa = a- I) is a homeomorphism.
(vi) follows from the fact that that a Ub is an open set containing 1.
To prove the latter half, let 12 be the family of subsets which are written
as the union of the members of { a U j a E G, U E V j . Similarly, we define S2'
by using { Ua I a E G, U E 01. If X E n', b E X, then for some Ua E Q',
we have b E Ua C X, and hence, 1 E b-1 Ua. Thus, there is a V E v with
V c b-1 Ua (by (iv)), so bV c Ua. Hence, if we consider the union of all
b V obtained for each b E X, then the union coincides with X, and X E 0 .
Thus, we see that S2' C 92. Similarly, we have S2 c_ S2' , and SZ = S2' If .
b E U1 a n - - n Un an (Ui E v, al , b E G), then 1 E U1 a1 b - I ; hence, there

1 -
are V E v with V C (by (vi)). Thus, for some V E v, we have

Ulalb-1
V c_ (l1 V c (U1a1 n ... n Unan)b-I , so Vb C U1a1 n n U,tan . Thus,

Ul a1 n . . n Unan E Q. In this way, we see that 0 satisfies the conditions
.
on the family of open sets. Therefore, a topology is uniquely defined on G.

Then our proof above shows that every neighborhood of a E G contains sets
of the form V a, a V with V E v.
To prove the continuity of the multiplication, let a, b E G, and let U(ab)
be a neighborhood of ab. Then there is a U E v with Uab C U(ab). Then
for some V E v, we have VV C a-I Ua (by (iv) and (vi)), and so there is a
V' E v with V' c aVa-I
. Thus, (V'a)(V'b) c (aV)(Vb) = aVVb c Uab,
and we see the continuity of multiplication.
I
The continuity of 0: a --> a-1 follows from (v), because (a V) - =
V-Ia-I . The To-condition follows from (ii). Thus, G is a topological
group. Q.E.D.
148 IV. VALUATIONS
THEOREM 4.4.4. If v' is a fundamental system of neighborhoods of the

identity of a topological group G, then v = { U n U- I I U E V'} is also a fun-
damental system of neighborhoods of the identity and v satisfies the condition
U-I
(vii) = U for every U E V
A fundamental system of neighborhoods satisfying (vii) is said to be sym-
metric.
PROOF. The validity of (vii) is obvious. That v is a fundamental system of
neighborhoods follows from (v) (or, by the continuity of 0 above). Q.E.D.
THEOREM 4.4.5. If a subgroup H of a topological group G is an open set,
then H is also a closed set.
PROOF. Each Ha (a c G) is an open set (Theorem 4.4.1). Thus,
Ua H Ha is an open set, and therefore, its complement H is a closed set.
Q.E.D.
A ring is called a topological ring if it is a topological additive group and
if the multiplication is continuous.
A field K is called a topological field if it is a topological additive group
and if K - {0} is a topological group under multiplication.
THEOREM 4.4.6. If v is a multiplicative valuation of a field K, then v
defines a topology on K so that K becomes a topological field as follows. Let
vo be a valuation of K, equivalent to v and satisfying the triangle inequality.
-
Then K is a metric space by letting p(a , b) = vo(a b) , a distance function
on K.
This topology is called the topology determined by v , because the topology
is essentially independent of the particular choice of vo .
PROOF. It is obvious that K is a metric space with distance function p.
The continuity of addition follows from vo((a + c) + (b + d) - (a + b)) =
vo (c + d) < voc + vod The continuity of the mapping a -> -a follows from
.
vo(-a) = v0a. Therefore, K is a topological additive group.

As for the multiplication, the continuity follows from
vo((a + c)(b + d) - ab) = vo(cb + ad + cd)
< (voc)(vob) + (voa)(vod) + (voc)(vod).
The continuity of the mapping a -> a- I (a 0) follows from
vo((a+c)-' -a-')= v0(a_1 (a+ c)-') - voc.
If we change vo by another one, then the distance of two points may
change. However, the order relation of distances, that is, p(a, b) > p(c, d) ,
does not change. Therefore, the topology does not change. Q.E.D.
THEOREM 4.4.7. Let K and v be as above. If K' is a subfield of K and
if v' is the restriction of v to K', then the topological field K' defined by v'
is a subspace of K.
5. COMPLETIONS 149
PROOF. For P' E K', let Ue(P') be the e-neighborhood of P' in K.

Then Ue(P') n K' is the e-neighborhood of P' in K'. On the other hand, if
Ue (P) is the e-neighborhood of P E K and if Q' E Ue (P) n K', then U. (P)
contains a 6-neighborhood of Q', and therefore, Ue(P) n K' contains the
6-neighborhood of Q' in K'. Thus, we see the assertion. Q.E.D.
5. Completions
Let S be a metric space with distance function p A sequence {P,1} _
.
P1 , P2, ... , Pn, ... of points Pn in S is called a Cauchy sequence if for

any positive number e, there is a natural number N such that m, n > N
implies p(m , P) < e. The sequence is called a regular Cauchy sequence
if m > n implies p(P,n, Pn) < 2-n , which is a special case of a Cauchy
sequence. A point P is called the limit of a sequence {Pn} = P1 , P2 1 ... ,
if for any positive number 8, there is a natural number N such that n > N
implies p(Pn, P) < e. We denote this by limn,. Pn = P, or simply by
P = lim Pn . In this case we also say that the sequence {Pn } converges to P.
It is clear that such a sequence is a Cauchy sequence. S is called complete
if every Cauchy sequence has its limit in S.
Two sequences {P } , {Qn } of points in S are called equivalent to each
other if the sequence { p(Pn , Qn)} of the distances between PJ1 , Qn con-
verges to 0 . This is an equivalence relation, and in this case if one of {Pn }
{Qn} is a Cauchy sequence, then so is the other.
LEMMA 4.5.1. (i) Let {P}, {Qn } be Cauchy sequences such that P =
lim Pn. Then {P,,} is equivalent to {Qn} iff P = lim Qn .
(ii) If {Pn} is a Cauchy sequence, then there is a regular Cauchy sequence

{Qn} that is equivalent to {Pn} .
PROOF. (i) is clear. (ii) For each natural number n , let N(n) be such that
m, m' > N(n) implies p(P,n, P,n') < 2-n , and we take natural numbers
e1 < e2 < < en < . so that en > N(n) for each n. Set Qn = Pe Then
. . .
the sequence {Qn} is the required one. Q.E.D.

THEOREM 4.5.2. If a field K has a multiplicative valuation v , then there
is an extension field K* satisfying the following three conditions:
(i) K* has a multiplicative valuation v* whose restriction to K coincides
with v .
In this case, we say that v* is a prolongation, or an extension of v .
(ii) K* is complete with respect to v* .
(iii) K is dense in K* .
Such a K* is uniquely determined up to isomorphisms and if {a,1 } is a

Cauchy sequence in K with a = lim an , then v*a = lim v(a,1) .
This K* is called the completion of K with respect to v and v* is called

the completion of v.
150 IV. VALUATIONS
PROOF. First, we prove this under the assumption that v satisfies the
triangle inequality. Let S be the set of Cauchy sequences in K and introduce
a multiplication and addition in S by
{an}{bn} = {anbn} , {a11} + {bn} = {an + bn}.
Then S is a ring in which 0, 0, ... , 0, ... and 1 , 1 , ... , , ... are 0
1
and 1 , respectively. (As is easily seen, the sum and the product of two
Cauchy sequences are again Cauchy sequences.) Let N be the set of Cauchy
sequences that converge to 0. Members of N are called null sequences. N is
an ideal of S. Furthermore, {an} is equivalent to {bn} if E N.
Therefore, S/N is the set of equivalence classes of Cauchy sequences in K.
We set K* = S/N and identify each element a of K with the class of the
sequences a, a, ... , a, .... Then the mapping a - a, a, ... , a ... is an
isomorphism from K into S, and therefore, the mapping a --> the class of
a, ... , a... is an isomorphism from K into K* . Note that if {an } is a
Cauchy sequence in K, then {v (an) } is a Cauchy sequence in the set of real
numbers, by the triangle inequality I v (an) - v (am) l < v (an - am) .
(1) To see that K* is a field, consider 0 0 a* E K* . Then a* is the
class of a Cauchy sequence {an} , which is not a null sequence. Thus,
v (an) > a for some 8 > 0 (for sufficiently large n) . Here, considering
a suitable equivalent sequence, we may assume that v (an) > 6. Then, since
p(an I , am I) = v ((am - an )an I am I) < v (am - an) a2 , we see that the sequence
{an is a Cauchy sequence, and its class is the inverse of a* . Thus, K* is
a field.
(2) We define a mapping v* of K* to be the set of real numbers as
follows. If a* is the class of {an } , then we set v * (a*) = lim v (a,1) This .
V* is well defined, because {v (an) } is a Cauchy sequence of real numbers

and if a* is also the class of {an } , then {an - a;1 } is a null sequence and
Iv(ai1) - v(aj1)l < v(a11 - an) . If a* , b* are the classes of {a11} , {bn}
respectively, then
v*(a*b*) = limv(anbn) = (liman)(limbn) = v*(a*)v*(b*)
v*(a* + b*) = lim v(an + bn) < lim(v(an) + v(bn)) = v*(a*) + v*(b*).
Obviously, v*O = 0. As we saw above, v*(a*) > 0 if a* 54 0. Thus, v* is
a multiplicative valuation of K* , which is an extension of v.
(3) To see that K is dense in K* , it suffices to prove that if a* is the
class of {an}, then a* = liman Since {an} is a Cauchy sequence, for any
.
positive number e, there is a number N = N(e) such that for all m, n > N,
v(an - am) < e. Now, v*(a* - aj) = lim v(an - aj) <e, whenever j > N.
Thus a* = lim an .
(4) To prove completeness, let fan *1 be a Cauchy sequence in K*. For

each natural number j, we take a regular Cauchy sequence {ajn } whose
class is a* (Lemma 4.5.1). Then we consider the sequence {a711 } Since .
5. COMPLETIONS 151
{an } is a Cauchy sequence, given any e > 0, there is an N = N(e) such that
-
v*(a* a*) < e for all m, n > N. This implies that if t is sufficiently large,
then v (amt - ant) < 8. Since the sequences {ajn } are regular, v (aj,n -
2-n for all m > n , and we have, for sufficiently large t,
v(amm - anti) = v(a,n,n - amt + amt - ant + ant - anti)
v(amm - ant) + v(ant - anti)
< 2-m + 2-n + v(a,nt - ant).
Therefore, the sequence {anti} is a Cauchy sequence. Let b* be its class.
Then b* = limann and v*(a*-b*) = lim n-+oo v(a .n -ann) = limn-.oo v(ajn -
aii + aii - anti) < 2-j + limn,, v(aii - anti) . Thus, if j -> oo , then v*(a* -
b*) -> 0 and b* is the limit of {an} I.
(5) To prove uniqueness; let K, v be another pair of completions of K,
v , respectively. Note first that if {an } is a Cauchy sequence in K , then
{v (an) } is a Cauchy sequence, and therefore, v (lim an) must coincide with
lim v (an) . Consider the mapping 0 from S to K such that q{ati } = (lim an
in K). Since K is dense in K, we see that q5S = K . O{an } = 0 if
0= (limati in X), that is, {an } E N, and therefore, K ^ K*. By what was
noted above, v and v* are compatible with the isomorphism.
To complete our proof, we consider the case where v does not satisfy the
triangle inequality. There is a valuation vo , which is equivalent to v that
satisfies the triangle inequality. Since vo = v' with t a positive number,
we see that the completion of K with respect to vo is the completion of K
with respect to v. Q.E.D.
THEOREM 4.5.3. Let K be a subfield of a field L with a multiplicative
valuation v', and let v be the restriction of v' to K. Assume that L is
complete with respect to v'. Then the closure K* of K in L (as a topological
space) is the completion of K.
PROOF. Each element of K* is obtained as the limit (in L) of a Cauchy
sequence in K, and we see that K* is a field by our proof of Theorem 4.5.2.
Since K is a subspace of the topological space K* (Theorem 4.4.7), we see
that K* is the completion of K. Q.E.D.
As for the value groups of the completions of valuations, there is a clear
difference between those of Archimedean valuations and non-Archimedean
valuations. Namely,
THEOREM 4.5.4. Let K, v, K*, v* be as in Theorem 4.5.2. Then
(i) If v is non-Archimedean, then, for any Cauchy sequence {an } that is
not a null sequence, limti--+00 v(an) coincides with v(at) for sufficiently large
j. Consequently, the value group of v* coincides with that of v.
(ii) If v is Archimedean, then the value group of v* coincides with the
multiplicative group of positive real numbers.
152 IV. VALUATIONS
PROOF. (i) 6 = lim v (an) is different from 0. Take r such that 8/2 >
e > 0. Then there is an N such that v (an - a,n) < E for all m , n > N N.
Then v (an) > 6 - e > e and v (an) = v (an - a,n + am) = v (an) , because
v (an - a,) < v (am) . Thus, v (an) is constant for large n , and 6 = v (an) .
(ii) K contains the rational number field Q, and the restriction of v
to Q is equivalent to the absolute value I (Theorem 4.2.1). Therefore,
I
{v*a1a E K*} contains all limits of Cauchy sequences of rational numbers.

Thus, the value group contains all positive real numbers. Q.E.D.
6. Archimedean valuations and absolute values
We prove here the following theorem.
THEOREM 4.6.1. If a field K has an Archimedean valuation v , then there
is a pair consisting of an isomorphism 0 from K into the complex number
field C and a positive number t such that (v a)' = IcaI for any a E K (where
denotes the absolute value).
If furthermore, K is complete with respect to v, then OK is either the
complex number field C or the real number field R.
PROOF. K contains the rational number field Q and the restriction of
v to Q is an Archimedean valuation of Q. By Theorem 4.2.1, there is a
positive number t such that (va)` = jal for all a E Q. Hence, considering
V t instead of v , we can assume that v a = a for all a E Q. Let K* , v* be
the completions, of K, v , respectively. Then K* contains the completion
of Q with respect to I(Theorem 4.5.3), and therefore, K* contains the
I
real number field R and v * a = a l for any a E R. Thus, considering K*

instead of K, we can assume that K is complete, and it suffices to show the
following two statements.
(i) If E K, va = jal (for all a E C), then K = C.
(ii) If V K , then v extends uniquely to a valuation v' of K(im) ,
and v' is complete.
[Reason: (1) If E K, then in R can be extended to in C.
I
Hence, by (ii), the restriction of v to C must be IThen (i) shows that

I .
K = C. (2) If 0 K, then K D R, K(im) = C, and therefore,

K=R.]
(i) The following fact is well known. If f is a real-valued continuous
function defined on a bounded closed set F in the Euclidean plane, then f
attains a smallest value in F (A more general statement, which follows from
Theorem 4.3.4, is that if f is a continuous mapping from a compact space
S to a space S', then f (S) is a compact subspace of S'.)
Assume for a moment that K C, and take a E K such that a 0 C.
Define a function f by f (z) = v (z - a) (z E C) f is a real-valued
.
continuous function defined on C, and f (z) > 0 because a C. Since

v (z - a) > v z- v a= I z j - v a , f (z) becomes large as l z j becomes large.
Therefore, if we take a sufficiently large number N, the smallest value of f
6. ARCHIMEDEAN VALUATIONS AND ABSOLUTE VALUES 153
in the closed set { z I z I < N} is the smallest value of f in C. Thus, the

smallest value of f exists by the fact stated above. Let 8 be the smallest
value of f , and take a E C such that f (a) = J. Now we shall show that
(*) If b, CE C, f(b)=8,and Icl<6,then f(b+c)=8.
Indeed, set /3 = b - a. By the definition of f , v /3 = 8 . We consider
/i" - c" for each natural number n . /3" - c" = (/3 - c) (/3 -
CC)
.. (/3 - c("- I )
where C is a primitive nth root of unity. Then v(/3" - c") = v(/3 - c)v(/3 -
v(/3-ce"-I) Since 8 is the smallest value of f , we have v(/3"-c") >
. .CC)
. .
v(/3 - c)8"-I , and therefore, v(/3 - c"//3"-1) > v(/3 _C). It follows that
v/3 + > v(/3 - c) = f(b - c). Since vc < v/3 , we have
v(c")/v(/3"-I)
(vc)"/(v/3)"-I -* 0 if n -> oo, and therefore, 6 = v/3 > f(b - c) . By the

minimality of 6, we have f (b - c) = J. Considering -c instead of c, we
also have f (b + c) = 8 . Thus (*) is proved.
Now, we go back to the proof of (i). Take z E C such that I zI is very large,
and choose a large natural number N such that Iz/NI < J. By (*) above,
we have f (a + i(z/N)) = 8 successively for i = 1, 2, ... , N. Therefore,
f (a + z) = 6 , which is a contradiction because f (a + z) > v (a + z) - va >
IzI - Ial - va.
(ii) First we define a mapping N of K(im) to K by N(a + b) _
a2 + b2 (a, b E K). Now, we shall show that if a valuation v' of K(im)
is an extension of v, then v'z = Jv(Nz) for any z in K(im). Note that
if a , b E K , then a + b and a - b are conjugate to each other over
K . If v'(a + bT) = v'(a - bT) , then v'(a + bi) = %[v + b2) _
v(N(a + b )) .
Therefore, we assume for a moment that there are a, b E K such that
v'(a + b ) < v'(a - b) Set z = (a + bv4--l-)/(a - b) . Then
.
Nz = 1 and v' z < 1 . Hence, considering elements of the form z3 for many
j, we have a sequence of elements zi = al + bi such that N(zi) = 1
and v' (zi) < 2-' (i = 1, 2, ...). Taking zi if necessary, we may
assume that v (ai) > v (bi) . Since v' (zi (ai - bi)) = 1 , we have that
2` < v'(ai - bid) < C v'(ai) , where C is the constant in condition (3)
of the definition of a valuation. Therefore, v (ai) = v' (ai) tends to oo when
i tends to oo. Set yi = zi/ai (= 1 + (bi/ai) ) . Then, since N(zi) = 1 ,
we have 1 + (bi/ai)2 = yi(1 - (bi/ai)) = al 2 , and (bi/ai)2 tends to -1

as i tends to oo. (bi+l/ai+1)2 - (bi/ai)2 = ai+ - al 2, and therefore, if
we set ci = bi/ai with suitable choice of + or - , we have a convergent
sequence {ci } . Then the square of lim ci must be -1 , which contradicts
the nonexistence of in K. This completes the first step of our proof.
Next we shall show that v'z = v(Nz) gives a valuation which extends
v . We may assume that v satisfies the triangular inequality. It is obvious
that v' is a mapping of K(im) to R and the restriction of v' to K
154 IV. VALUATIONS
coincides with v .
Furthermore, it is easy to see that N(zz') = (Nz) (Nz') . Hence, v'(zz') _
(v'z)(v'z') for all z, z' E K(im). Nz = 0 if z = 0, and consequently,
v' (z) = 0 if z = 0. Thus, it remains only to show the existence of C such
that v'(1 + z) < C for v'z < 1 . Assume for a moment that there is no such
C. Then there are z1 E K(im) , v'(z1) < 1 , v'(1 + z1) > 2' We write
.
Z (a1, b1 E K) . Then
v(N(z1)) = v(a2 + b2) < 1, v(N(1 + z1)) = v((1 + a1)2 + b2) >
221.
Since v satisfies the triangle inequality, we have v(1 + 2a1) > 22i - 1
Thus, v (a1) > - 2 > 22(1-1) Set y1 = z, /a1. Then {y, } converges to
221-1 .
0. Hence, {b1/a1} converges to -, which contradicts / K.

The completeness of v' is not hard, and we omit it. Q.E.D.
7. Additive valuations and valuation rings
The additive group of real numbers has an order (the notion of largeness),
which is compatible with addition. We define here the notion of ordered ad-
ditive group, which is a generalization of the additive group of real numbers.
An additive group G is called an ordered additive group if
(i) G is a linearly ordered set.
(ii) If a,bEG, a>b,then -a< -b.
(iii) Given a , b , c, d E G with a> b and c > d, then a + c> b+ d.
An additive valuation (or, simply, a valuation) of a field K is a mapping
w of K to G U {oo} , where G is an ordered additive group and oo is an
element that is defined to be larger than any element of G, provided that the
following three conditions are satisfied:
(1) If 00aEK,then waEG; wO=oo.
(2) For a, b E K, w(ab) = wa + wb, where it is understood that
oo+oo=oo+a=a+oo=oo for all a E G.
(3) w(a + b) > min{wa, wb} for all a, b E K.
{waJ0 0 a E K} forms an additive subgroup of G and is called the value
group of w.
LEMMA 4.7.1. Under these circumstances, we have, for 0 0 a E K, that
w(a-1) = -wa, wl = w(-l).= 0, and wa > wb implies w(a+b) = wb.
This is proved in the same manner as Lemma 4.1.1.
We obtain an easy example of an additive valuation from a
non-Archimedean valuation v of a field K. Namely, we set wa = - logs va
with a real number c bigger than 1, then w is an additive valuation. Con-
versely, if we are given an additive valuation w , whose value group is a sub-
group of the additive group of real numbers, then we get a non-Archimedean
c_u'a
valuation v by defining va = . As an ordered additive group for an
7. ADDITIVE VALUATIONS AND VALUATION RINGS 155
additive valuation, we can use a more general one than the additive group of
real numbers R ; a simple example is the n-ple product R(") = R x xR
of R with lexicographical order.
One aspect of the importance of additive valuations is its applications to
the theory of commutative rings. But, we do not go in this direction in this
book.
If an additive valuation w of a field K is given, then R. = j a E K j w a >
0} forms a ring, which we call the valuation ring of w. M. = {a E Kjwa >
0} forms an ideal of Rw . Obviously, a (E Rw) is not invertible in R. if
wa > 0. This shows that Mw is the set of noninvertible elements in R. ;
hence, Mw is the unique maximal ideal of Rw and is called the valuation
ideal of w . R. /Mw is called the residue class field of w , or, of Rw The
.
Krull dimension of Rw is called the rank of w . Two additive valuations w ,

w' of a field K are defined to be equivalent to each other if their valuation
rings coincide.
A subring R of a field K is called a valuation ring of K if there is an
additive valuation w of K such that R coincides with the valuation ring
of w. In this case, we call w the valuation of K defined by R. By the
definition of equivalence, w is unique within equivalence.
THEOREM 4.7.2. The following conditions on a subring R of a field K are
equivalent.
(i) R is a valuation ring of K.
(ii) If a E K, a R, then a-I E R.
(iii) (1) K is the field of fractions of R, and (2) for a, b E R, a bR,
we have b E aR.
(iv) (1) K is the field of fractions of R, (2) R has only one maximal
ideal, and (3) every finitely generated ideal of R is principal.
(v) (1) If S is a ring such that R c S C_ K, then S contains the inverse
of some noninvertible element of R, and (2) R has only one maximal
ideal.
(vi) If S is a ring such that R c S c K, then there is a prime ideal P of
R such that S=RP={a/bla,bER,b P}.
PROOF. To prove that (i) implies (ii), let a E K with a 0 R, then
wa < 0 (w is the valuation defined by R), so w (a-') = -wa > 0, and
hence, a- IERR.
To prove that (ii) implies (i) we shall show that a, b E. K, aR % bR
implies bR c aR. That aR Z bR implies ab-1 0 R ; hence, ba- I E R
which implies b E aR, and that bR c aR. Now, we set Go = {aRja E K}
and G = Go - {0} . G forms a multiplicative group under the multiplication
(aR)(bR) = abR. Note that abR = {xylx E aR, y E bR}, independent
of the particular choice of representatives a, b of aR, bR, and therefore,
the multiplication is well defined. We write this multiplication by addition
156 IV. VALUATIONS
(namely, [aR] + [bR] = [abR]). By what we have proved above, Go is

linearly ordered by the inclusion relation, and we consider the dual order
(namely, [aR] > [bR] if aR C bR). Then G is an ordered additive group.
Since {O} = OR C aR (for all a 0), we set co = [OR]. Then the mapping
w of K to Gu {oo} such that wa = [aR] becomes an additive valuation of
K. The zero of G is [R] = [1R]; hence, an element a is in the valuation
ring of w if [aR] > [R], i.e., aR C R, so a E R. Thus R coincides with
the valuation ring of w .
To prove that (ii) implies (iii), note that (2) can be proved as in the be-
ginning of the proof above that (ii) implies (i). (1) is clear.
To prove that (iii) implies (ii), note that a E K, a 0 R implies a = b/c
with b, c E R such that b cR which implies c E bR, and a-1 = c/b E R.
Thus, we have shown the equivalence of (i), (ii), and (iii).
That (i), (ii), (iii) imply (iv) is obvious, and we shall show that (iv) implies
(iii). For a, b E R, aR + bR = cR for some c E R. Then a' = a/c,
b' = b/c are in R and a'R + b'R = R, which implies that one of a', b',
say a', is invertible. Then aR = cR E) b.
To prove that (i) implies (vi), note that R C S C K implies that S satisfies
the condition (ii). Hence, S is a valuation ring, and has only one maximal
ideal Q. Set S' = RQnR = {a/b l a, b E R, b 0 Q n R}. b Q n R
implies that b is invertible in S ; hence, S' C_ S. Thus, R C_ S' C_ K, and
therefore, S' is also a valuation ring of K. Suppose for a moment that for
some a E S, a 0 S'. Then a-1 E S', namely, a is invertible in S and
a-1 is not invertible in S'. However, the set of noninvertible elements in S'
coincides with (QnR)S' c QnS'. Consequently, a-1 E Q. This contradicts
the fact that a is invertible in S. Thus we proved that (i) implies (vi).
To prove that (vi) implies (v), note (1) follows from (vi) and (2) is the
case S = R in (vi).
To prove that (v) implies (iii), note that (1) in (v) implies that the integral
closure of R in K must coincide with R (Corollary 3.7.8). Thus, R is
an integrally closed integral domain. Clearly, K is the field of fractions of
R. Assume that a, b E R and a 0 bR. Then (v) implies that R[a/b]
contains the inverse of some noninvertible element d of R. Thus, d-1 =
c0(a/b)" + + cn_ I (alb) + cn (c, E R) , which implies that bn = d coa" +
ab"-1
+ d cn_ I Since R has only one maximal ideal, 1 - dc,, is
+ d cn b" .
invertible in R. Therefore, multiplying by (1 - d c") -1 , we have a relation
b" = dcoan + + dc'-1 (c!
E R), which shows that b/a is integral
over R. By the integral closedness of R, we have b/a E R and b E aR.
This completes the proof that (v) implies (iii). Q.E.D.
COROLLARY 4.7.3. A valuation ring R of a field K is integrally closed in

K. Every intermediate ring between R and K is a valuation ring of K and
is of the form RP with a prime ideal P. The set of all ideals of R forms
7. ADDITIVE VALUATIONS AND VALUATION RINGS 157
a linearly ordered set under inclusion relation. If L is a subfield of K, then

R n L is a valuation ring of L.
PROOF. The first and the second assertions were already shown in the proof
above. To prove the third assertion, let I, J be ideals of R. If I Z J then
there is an a E I with a J, then a bR for all b E J, which implies
b E aR, so J c aR c I. The last assertion follows from condition (ii) in
the theorem above. Q.E.D.
COROLLARY 4.7.4. If P is a prime ideal of a valuation ring R, then P =
PRP , and R/P is a valuation ring of the f eld of fractions of RIP.
PROOF. It is obvious that P C PRP . Take a E PRP . Then a is not
invertible in RP R. If a 0 R, then a- I E R, and this contradicts the non-
invertibility of a. Thus a E R and a E PRP n R = P (Theorem 3.6.1),
namely, P = PRP . a , b E R/P implies aR c bR or bR c aR with rep-
resentatives a, b of d, 6. Then a(R/P) c b(R/P) or b(R/P) c a(R/P),
respectively, which proves that R/P is a valuation ring. Q.E.D.
We have just mentioned the relation between non-Archimedean valuations
and additive valuations when we defined them. Here we rephrase the relation
more clearly.
THEOREM 4.7.5. To each nontrivial, non-Archimedean valuation v of a
field K, we can associate an additive valuation w defined by wa =_ logs
va with a fixed real number b > 1 . This w is of rank 1 Conversely, if w is
.
an additive valuation of rank 1, then there is an order-preserving isomorphism

0 from the value group G of w into the additive group R of the real numbers,
so that a non-Archimedean valuation v is defined by va = b_o(wa). In each
of these cases, the valuation ring of v coincides with the valuation ring of
w. In other words, the equivalence classes of nontrivial non-Archimedean
valuations are in one-to-one correspondence with the equivalence classes of
additive valuations of rank 1 .
PROOF. As for the first assertion, we saw this already, except for the result
that w is of rank 1 . Let a be an element such that .0 0 va < 1 Then a
.
is a noninvertible element of the valuation ring R of v, and the maximal

ideal P of R is different from {0} . Suppose for a moment that there is a
prime ideal Q such that {0} c Q c P. Take b, c such that O :AC E Q,
b Q, E P. Then b" 0 Q for all n. On the other hand, since v b < 1 ,
there is an n with v(b") = (vb)" < vc, which means b" E cR c Q, a
contradiction. Thus, there is no such Q and w is of rank 1.
Conversely, let w be an additive valuation of rank 1 Let R, P be the
.
valuation ring and the valuation ideal of w , respectively. Fix an element a

(0 0) of P. For each natural number n and for each element b (0 0) of
K, we define a rational integer e = e(n, b) so that b" 0 aeR, b" E ae-1 R ,
or equivalently, (e - 1)(wa) < n(wb) < e(wa). Since m > n implies
158 IV. VALUATIONS
I(e(n, b)/n) - (e(m, b)/m)I < 11n, limn-,O e(n , b)/n exists and is well
determined by the value wb. We denote the limit by u(b).
We define a mapping 0 of G to R by O(wb) = u(b). We shall prove that
0 is an order-preserving isomorphism from G into R. First, let y E G.
Then there are b, c E K such that /3 = wb, y = wc, and /3 + y = w (bc) .
Now bn ae(n'b)R, bn E ae(n,b)-1R, Cn ae(n'c)R, c" E ae(n'c)-1R.
ae(n , b)+e(n , c)R b)+e(n , c)-2R
Thus (bc)n , (bC)n c ae(n , , which implies that
e(n, b) + e(n, c) - e(n, bc) is 0 or 1 . Hence, q(/3 + y) = q5/3 + y and 0

is an additive group homomorphism. Next we shall that q5/3 # 0 for /3 0.
We may assume that /3 > 0. Then b E P, which implies that e(n, b) > 1
for a sufficiently large n . Since (e (m , b) -1)/m is monotone nondecreasing
for m = n, n 2 , ... , n r , ... , we have
lim e(m, b)/m = lim
Fn-'oo r-+oo
(e(nr , b) - 1)/n' > (e(n, b) - 1)/n > 0.
Thus, q/3 > 0, and therefore, 0 is an isomorphism into R. It is easily
seen that a > /3 implies Oa > q5/3 and 0 is order-preserving. The other
assertions are clear. Q.E.D.
From the theorem above, we see that additive valuations of rank 1 are
nothing but non-Archimedean valuations. Therefore, the additive valuation
corresponding to the p-adic valuation is also called a p-adic additive valuation
or simply a p-adic valuation.
In constructing valuation rings, the following theorem is important.
THEOREM 4.7.6. Let R be a valuation ring of a field K, and let P be the
maximal ideal of R. If W is a valuation ring of the field RIP, then S = {a E
RI(a mod P) E W} forms a valuation ring of K in which P is a prime ideal
and Sp = R. Consequently, Krull dim S = (Krull dim R) + (Krull dim W).
This S is called the composite of R and W.
PROOF. Assume that a E K, a S. (i) If a R, then a-1 E P, and
a-1 ES. (ii) If a E R , then (a mod P) W and, (a- I mod P) E W,
hence a-I E S. Thus, a-I E S and S is a valuation ring of K. Since
S C R , Theorem 4.7.2 shows that R = By the definition of S, we
Spns.
have P n S = P. The last equality follows from the fact that prime ideals of
R are prime ideals of S (Corollary 4.7.4). Q.E.D.
Here we give the definition of a place of a field and state the relation
between the notion and valuation rings. A place 0 of a field K is defined
as follows. Suppose we have a field L, an element denoted by oo, and a
mapping 0 of K to L U {oo} satisfying the following conditions.
(i) a, b E K, q5a, Ob oo implies d(ab) = (qa) (qb) , q(a+b) = qa+cbb .
(ii) qa 0, qSb = oo implies q(ab) = oo.
(iii) Oa # oo, Ob = oo implies 0(a + b) = oo.
(iv) qa = oo implies 0(a-') = 0.
In other words, we define operations on oo and elements a' of L by
a' + oo = oo ; a' oo = oo for a' 0; oo-1 =0 (0. oo , oo oo are
8. APPROXIMATION THEOREMS 159
indefinite). Then, a homomorphism from K to L U {oo} is a place of K.

Two places 0, 0' are said to be equivalent to each other if q5a H q'a
gives an isomorphism of OK and q5'K.
THEOREM 4.7.7. There is a one-to-one correspondence between the set of
equivalence classes of places of a field K and the set of valuation rings of K
as follows. If a place 0 corresponds to a valuation ring R, then R = {a E
Kjqa 0 oo}. In this case, the restriction of 0 to R is equivalent to the natural
homomorphism u from R to R/P (P is the maximal ideal of R), namely,
Oa H yra gives an isomorphism between OR and yJR.
PROOF. To each valuation ring R of K, we can associate a place 0 by
using v as above (qa = oo (a E K) if a 0 R) Conversely, assume that
.
is a place of K. Set R = {a E Kkba oo}. a E K, a 0 R implies

q5a = oo. Thus, 0(a-') = 0, and a- I E R. Thus, R is a valuation ring with
maximal ideal {a E Kjoa = 01. Therefore, the kernel of the restriction of
0 to R is the maximal ideal; hence, 0 is equivalent to the place obtained
from R. Q.E.D.
8. Approximation theorems
First we give the approximation theorem for multiplicative valuations.
THEOREM 4.8.1. Assume that v1 , ... , vn are nontrivial multiplicative val-
uations of a field K, such that no vi is equivalent to any other vi Then for .
any given elements a , ... , an of K and for any positive numbers e , ... , en
1 1
there is an element b of K such that vl (b - al) < E1 for i = 1 , ... , n .

For the proof, we need the following lemma.
LEMMA 4.8.2. If a multiplicative valuation v of a field K satisfies the
triangle inequality, then for c E K, we have
(i) v(c(1 +c)-I) - 1 < v(1 +c)-1
.
(ii) vc > 2 implies 2-1 < v (c(1 + c)- I) < 2.

(iii) vc < 2-1 implies v(c(l + c)-1 < 2(vc).
C)-1
PROOF. (i) C(1 + = 1 - (1 + C)-' .
+c)-1
(ii) 1 < v(1 +c) < vc+ 1 ;hence, (vc+ 1)-I < v(1 <1 . Therefore,
2-1
< vc/(vc+ 1) < v(c(l +c)-I) < v(1 +c)-1 + 1 < 2.
2-1
(iii) v(c(l +c)-I) = vc/v(l +c), v(1 +c) > 1 -vc > , and therefore,
v(c(l + c)-I) < = 2(vc). Q.E.D.
vc/2-1
PROOF OF THEOREM 4.8.1. Replacing each v; with an equivalent valu-

ation satisfying the triangle inequality (and modifying E; accordingly), we
may assume that every vi satisfies the triangle inequality. First we show that
there are d1 , ... , do E K such that v1(dl) > 1 , vj(d,) < 1 (for all j 0 i).
We have only to show the existence of d1 because of symmetry. We use an
160 IV. VALUATIONS
induction argument on n . For the case n = 2, there are ci E K (i = 1 , 2)

with vI (CI) > 1 , v2(cI) < 1 , VI (c2) > 1 , v2(c2) < 1 (Theorem 4.1.2, (ii)).
Therefore, d, = c, c2 is the required element. Assume now the existence of
ckEK (k=1,2) such that v1(c1)>1, vj(c1)<1 for j=2,...,n-1;
vI(c2) > 1 , vj(c2) < 1 for j = 3, ... , n . If vn(c1) < 1 , then we can
take cI as dI ; if v2 (c2) < 1 , then c2 is taken as d . Hence we assume
1
that vn(c1) > 1 , v2(c2) > 1 . If vn(c1) = v2(c2) = 1 , then we can take
clc2 as dI . So, we assume that v2(c2) > 1 , v2(c2) > v11(cI) (because of
symmetry between 2 and n). Taking some powers of cI , c2 , if necessary,
we can assume that vI (c1) > 2, v(c1) < 2-' for j = 2, ... , n - 1 , and
vI (c2) > 2, v j(c2) < 2-' for j = 3, ... , n , and furthermore, v2(c2) > 2,
vn (cI c2) < 2-' Then d, = cI c2 (1 + c2) -' is the required element (Lemma
.
4.8.2).
Next, we show that there are ei E K (i = 1 , ... , n), vi(ei - 1 ) <8 i ,
v j (ei) < e j (i j) . Indeed, for each i , we take a sufficiently large real
number t so that vi (e`.) > Ei ' + 1 , v j (d,) < 2e1, then e, = d `. (1 + d')-
is the required element (Lemma 4.8.2).
Finally, we take positive numbers Ei such that Enj=1 v.(aj)e < Ei , and
I
we take ei so that the above condition is satisfied with E, instead of Ei .

Then b = > aiei is the required element, because vi (b - a1) = vi (ai (ei -
1) +>2jli ajej) < vi(ai(ei - 1)) +Ej./i vi(ajej) < vi(ai)Ei +Ej./i vi(aj)El < Ei .
Q.E.D.
Now we consider additive valuations. As for valuations of rank 1, the
theorem above can be applied. But, if some valuations of rank > 2 are
involved, then simple adaption of the theorem above becomes false. For
instance, consider two valuation rings R , R2 (RI 54 R2) of a field K,
I
which are of rank 2 and having a common prime ideal P 0 101. Then
(RI )P = (R2)P, RI /P R2/P . In this case, if we take aI E P and a2 P,
then we cannot have an element b such that the elements b-ai are "near" to
0. We shall give an approximation theorem applicable to additive valuations.
For this purpose, we give some preliminary results.
LEMMA 4.8.3. Let RI , ... R11 be valuation rings of a field K such that
,
Ri Z R j if i j. Set D= ' Ii Ri. Then, for each element a of K,

there is a natural number s > 2 such that both 1 /(1 + a + as-') and
a/(l +a+ +as-') are in D.
PROOF. We look for the condition so that these two elements are in R.
for each i. Let v = vi be the valuation determined by Ri .
(1) In the case a V Ri , v a < 0, and therefore, v (l + a + + as-')
(s - 1)(va). Thus, v(1/(1 + a + + as-')) = -(s - 1)(va) > 0, v(a/(1 +
a + + as-' )) = -(s - 2)(va) > 0, and these elements are in Ri for any s
(> 2).
8. APPROXIMATION THEOREMS 161
(2) In the case va > 0, these elements are in Ri because v (l + a + +

as- I )
= 0.
(3) In the case va = 0, v (at - 1) = 0 (t = 1, 2, ...), these elements are
in R , because v (1 + a + + as-1) = 0 follows from 1 + a + + as-' _
.
(1 - as)/(1 - a).
(4) In the case v (a - 1) > 0, these elements are in R. , if s is not a
multiple of the characteristic of the residue class field of R..
(5) In the case v (a - 1 ) = 0, v (a` - 1) > 0 (for some t > 2), let m be
the smallest among such t. Then other t are multiples of m. Therefore,
these elements are in R; if s is not a multiple of m .
Thus, the condition that these two elements are in all the R, is obtained
in the form that s (> 2) is not a multiple of certain special natural numbers.
Therefore, such s exists and the assertion is proved. Q.E.D.
THEOREM 4.8.4 (Independence theorem of additive valuations). Let
R1 , ... , Rn be valuation rings of a field K such that R. Ri if i 0 j I.
Let Pi be the maximal ideal of R, and set D= (l; R. , M1 = P. n D. Then
we have
(i) DM = RI .
(ii) The maximal ideals of D are exactly M1 , ... , M1.
PROOF. DM C Ri because D C Ri . For each a E R., there is an s (> 2)
with 1/(l + a + + as-'), a/(1 + a + + as-') E D by the lemma above.
Since a ER1, 1 ERl, and therefore, 1/(l
is an invertible element in R. , and hence, is not in M1. Thus a = (a/(1 +
a + + as-')/(1 /(1 + a + + as-') E DM , which implies (i). DMi OD ,
Mi
for i 54 j, which implies that, for i 54 j, M1 Mj in view of the condition
Rj % R.. Therefore, to complete the proof of (ii), it suffices to show that
if M is a maximal ideal of D, then M C M, for some i. Assume for a
moment that M is not contained in any Mi. Then there is a b c M with
b 0 M by Lemma 1.4.5. Since b is not in any Mi, vl (b) = 0 for every
valuation vi defined by R.. Thus b-' E R, for every i and b c D. Thus
b is invertible in D, which contradicts b E M. Q.E.D.
COROLLARY 4.8.5. Under these circumstances, if Q, is an ideal of each Ri
such that \ % Rj f o r any j i, then we have D/(n! Ql) ' R /Q1 ... ED
i
Rnl Qn .
PROOF. Set Ql = Qi n D. Assume that a maximal ideal M. contains
Ql + Q (i 54 j). Since D Q' DM, = Ri, we have D, = (R1) , .
On the other hand, since D , D DM, (Qd must be a prime ideal of

Rk and i = k. Similarly, j = k , and we obtain a contradiction to i 0 j.
Thus Q'+ Q' = D, and we have D/(nQ1) ^_' D/Q' . . . D/Q'i . On the
162 IV. VALUATIONS
other hand, we have shown above that M;/Q' is the unique maximal ideal
of D/Q'. Consequently, the class of every element of D which is not in
Mi is invertible in D/Ql. Thus, D/Q' = DM,/Q'DM, = Ri/Q'Ri = Ri/Qi.
Q.E.D.
From this result, we obtain an approximation theorem.
THEOREM 4.8.6. Let vI , ... , vn be additive valuations of a field K whose
valuation rings are RI , ... , Rn , respectively. Assume that R. Z Rj for i 34 j.
Let Qi.l
be the largest member among common prime ideals of Ri and Rj
(the existence will be shown in the proof). Let Qi be the largest among
Qil , - -- , Qi,i-1 , Qi,i+l , "' , Qin . If al , ... , an ; bI, ... ,
b,1 are elements
of K such that bi E Ri , bi Qi, and there are ti with vi(b'iai) > 0, then
there is an element c of K such that vi (c - ai) > vi (bi) for every i.
PROOF. The existence of QiJ follows from the facts that in a valuation
ring, the family of ideals forms a linearly ordered set and that if I (j E A)
form a family of prime ideals of a ring which is linearly ordered, then the
union of all I is a prime ideal.
The assertion becomes stronger if we change bi to those with bigger values
under vi . Therefore, we may assume that the bi are not invertible in Ri .
(i) In the case where all ai are in Ri , set D= n Ri , and apply the corol-
lary above to D/(nb?Ri) Then D/((l b?Ri) '" RI/biRI ... Rn/bnRn ,
.
and we have a representative c E D of > (ai mod b2Ri) (in the right-hand
side). Then c - ai E b2Ri and vi (c - ai) > vi (b?) > vi (bi) .
(ii) In the general case, apply the case (i) above to the case where both ai
and bi are b`.' , and we see that there is a d E D , vi (d - b;') > vi (b'i) . Then
d # 0, and vi (d) = vi (b;') . Set ai =aid , bl =bid , and apply (i) to ai , b; .
Then there is an e E D such that vi (e - aid) > vi (bid) . Consequently,
c = e/d is the required element. Q.E.D.
9. Prolongations of a valuation
In this section, we shall prove the following and some related theorems.
THEOREM 4.9.1. Let L be a field algebraic over its subfield K. If w is
an additive or a multiplicative valuation of K, then there is a valuation w'
of L which is a prolongation of w. All such w' are obtained as w, defined
below.
We fix a prolongation w of w to the algebraic closure K of K. For
each K-isomorphism 0 from L into K, we define a valuation wo of L by
w,(a) = w(a) (0 0' may not imply w. 0 w,,).
Consequently, the number of mutually distinct prolongations of w to L
does not exceed [L: K]S. If w is an additive valuation or a non-Archimedean
valuation, then this number coincides with the number of maximal ideals of
the integral closure of the valuation ring of w in L.
9. PROLONGATIONS OF A VALUATION 163
We begin with the proof of the following theorem.

THEOREM 4.9.2. Let S be a subring of a field K and {0} c PI C . . C P. .
an ascending chain of prime ideals of S. Then there is a valuation ring R of

K such that S C R and R has an ascending chain {0} c Q1 c . . . c Q. of
prime ideals Qi such that each Qi lies over P .
PROOF. We use an induction argument on r. If r = 0, then K is the

required ring.
In the case r = 1 , we may assume that P = PI is the unique maximal
ideal of S, by taking Sp instead of S. Consider the family F of subrings
Sj of K such that Sj D S and 1 0 PST . F is an inductive set as is easily
seen, and there is a maximal member R in F by Zorn's lemma. Then R
is integrally closed, because if a ring R' is integral over R, then we have
PR' R' by applying Corollary 3.7.8 to a prime ideal of R containing PR.
R has only one maximal ideal, because if Q is a prime ideal of R containing
PR, then RQ E F and RQ = R. Now let x E K be such that x R. Then
1 E PR[x], namely, there exist c, E PR, cox` + c1 x`- I + + c1 = 1 Since .
1 - ct is not in the unique maximal ideal Q, we see that 1 - c1 is invertible

in R, which implies that x-1 is integral over R. Thus x-I E R because R
is integrally closed, and we have proved that R is a valuation ring of K. It
is obvious that Q n S = P.
In the general case (r > 1) , by the case r = 1 , there is a valuation ring R1
(D S) of K which has a maximal ideal Q1 lying over PI . SI = S/PI is a
subring of the field K1 = R1/Q1 and has an ascending chain of prime ideals
{0} = PI /P1 c P2/PI c . c Pr/PI . By the induction hypothesis, there is a
. .
valuation ring V of K1 such that Sl C V and V has an ascending chain

of prime ideals {0} c N2 c . c Nr lying over Pi/PI (i = 2, ... , r). Then
. .
the composite R of R1 and V is the required one. Q.E.D.

THEOREM 4.9.3. Let R be a valuation ring of a field K. Let L be an
algebraic extension of K, and let D be the integral closure of R in L Then .
V is a valuation ring of L such that V n K= R iff

V = DP with P a maximal ideal of D.
PROOF. (i) In the case where L is a normal extension of finite degree,
set G = G(L/K). Theorem 4.9.2 shows that there is a valuation ring V
of L such that V n K = R (cf. Exercise 4.7.4). Set D' = ' IaEG UV . Let
VI , ... , Vn be all of mutually distinct rings among the a V (U E G) . (Note
that n = [G: H] with H = {a E GI QV = V J.) Since Vi, V are conjugate
to each other, we see that % V (i $ j) . Let d be any element of D', and
+ c, (c, E K) be the minimal polynomial for d over K.
let X ` + c1 X `- I +
Each cj is a symmetric form in {Qd l a E G} . Since Qd E D', vi(ad) > 0
with valuation v; defined by Vi, and v! (cj) > 0. Thus, cj E V n K = R,
164 IV. VALUATIONS
and d is integral over R. Since D' is integrally closed, we have D' = D.

Therefore, Theorem 4.8.4 shows that the V are D p with maximal ideals P
of D. It also follows that an arbitrary extension V is one of the DP D.
(ii) In the case where [L: K] is finite, let L* be a finite normal extension
of K containing L, and let D* be the integral closure of R in L* As for.
a valuation ring V * of L* , we have that V * n K = R if V * = D. with

maximal ideal P* Let V1* , ... , VS* be all such DP.. Then D = D* n L =
.
n (Vj* n L) = n; V , where V. runs through a finite number of valuation rings

of L such that V n K = R. For an arbitrary V in the assertion, there is
a V* such that V = W n L by case (i), and W = DP. with a maximal
ideal P*. Then P = P* n D is maximal and V = DP Conversely, it is
.
clear that Dp gives a prolongation if P is maximal (Theorem 4.8.4), and we

completed the proof of this case.
(iii) In the general case, first consider the ring DP D. For a E L, we set
D' = D n K(a). Then DPfDI is a valuation ring of K(a) and contains a
or a-1 Hence, a or a-1 E DP Thus DP is a valuation ring of L. Since
. .
P n R is the maximal ideal, D p gives a prolongation of R, which proves the

if part. As for the converse, we consider an extension V, and let M be the
maximal ideal of V. Set P = M n D, which is a maximal ideal of D. If
a E V, then a c V n K(a) = (D n K(a))PnK(a) C Dp. Thus, V C Dp, and
obviously, DP C V. Therefore, we have V = Dp. Q.E.D.
COROLLARY 4.9.4. Let w be an additive valuation of a field K, and let
w' be a prolongation of w to an algebraic extension field L. Then the rank
of w' coincides with that of w.
The proof is clear by Corollary 3.7.13.
COROLLARY 4.9.5. Let w be an additive or a non-Archimedean valuation
of a field K. Then w can be extended to a valuation of an arbitrary extension
field L of K.
In the case of an additive valuation, w can be extended to a valuation of
the same rank (cf exercise 4.7.4).
PROOF. The existence of extensions of an additive valuation follows from
Theorem 4.9.2. Since a non-Archimedean valuation is essentially an additive
valuation of rank 1, it remains to prove the latter half. Let R, P be the
valuation ring and the valuation ideal of w, and let {bi I j E Al be a tran-
scendence base of L over K. Let T be the ring generated by these bi over
R, and set V = TPT Each element a of V is written in the form a(g/ f )
.
where a E R, f, g E T, and the coefficients of each of the polynomials f ,

g in these bl have no common factor. Since R is a valuation ring, some of
the coefficients of each of f, g are invertible. Therefore, V is a valuation
ring. Obviously Krull dim V = Krull dim R. Since L is algebraic over the
field of fractions of V, we complete the proof by Corollary 4.9.4. Q.E.D.
PROOF OF THEOREM 4.9.1. (i) In the case of an additive valuation, the

existence of an extension of w is already proved. It is obvious that w. gives
an extension of w , and therefore, it suffices to show that every extension w'
of w to L is expressed as w0 Considering an extension of w' to K, we
.
can assume that L = K. Let R be the valuation ring of w and let D be the
integral closure of R in T. Then the valuation rings of w., w' are DP ,
DM with maximal ideals P, M of D. Hence, there is a U E G(K/K) such
that aP = M (Theorem 3.7.11). Thus, the valuation ring of w' coincides
with that of wQ , and we can regard that w' = w, (In fact, first, we can
.
imbed value groups of w' and w. in a common sufficiently large ordered

additive group. Then, by Theorem 4.9.6 below, for each element a of K,
there is a natural number t such that t w'a is in the value group of w , and
w'(at) = wa(at) hence w'a = wQ(a) .)
(ii) As for the case of a non-Archimedean valuation, this is substantially
the case of rank one additive valuation, and the assertion follows from the
case (i) above (cf. Corollary 4.9.4).
(iii) In the case of an Archimedean valuation, we can assume that K is a
subfield of the complex number field C and wa = Ial (for all a E K), by
Theorem 4.6.1. Since C is algebraically closed, we can assume that K C K C
C. Then an extension of w exists because Igives an extension of w. On
I
the other hand, if w' is an extension of w , then Theorem 4.6.1 shows that
there is a pair consisting of a positive real number t and a K-isomorphism
u of L into C such that (w'a)t = j oral (for all a E L). Since w'a = dual
for any a E K, we see that t = 1 . Q.E.D.
THEOREM 4.9.6. Let w be an additive valuation of a field K and let L be
a finite algebraic extension of K . Assume that mutually distinct prolongations
o f w to L are wt , ... , wn . Let R, R! ; P, P1 ; G, G. are valuation rings,
valuation ideals, and value groups of w, w,, respectively. Set e; = #(G1/G),
f = [R1lP : RIP]. Then E1 elf < [L: K1. Consequently, all et , fi are
finite.
el , f are called the ramification exponent and the degree of wl over w ,
respectively. wl is said to be unramified if e = 1 .
PROOF. (i) First, we shall show that elf < [L: K] for each i. Assume
that at , ... , as E R, are such that their residue classes modulo Pi are lin-
early independent over R/P and that at , ... , at E G. are such that their
residue classes modulo G are mutually distinct. Take b. E L such that
w (bl) = al . Assume that > Cjk ajbk = 0 with Cjk E K such that not all of
them are 0. Let I be the set of (j, k) such that wl (Cjk ajbk) is the small-
est. Then, since wl(C1) < w1(C2) implies w.(C1 + C2) = w.(c1) , we see that
wl(>(i,k)EI Cikaibk) > wl(Cikaibk) if (j, k) E I. Thus, #(I) > 2 .
wl(Cjkaj bk) = wl(Cjk) + 0 + wl(bk) = w(Cjk) + ak E (G + ak) ,

166 IV. VALUATIONS
and therefore, k is common to all (j , k) E I ; hence, w (cjk) is also common.

Then dividing by one cjk , we have a relation wl(>(j k)EI cjkaj) > 0 with
invertible cij E R, which contradicts the linear independence of the residue
classes of aj Thus, these st elements ajbk are linearly independent over
.
K and st < [L: K].

(ii) For the case where w is of rank 0, R = K in this case. Hence,
n = 1 , fl = [L: K], R 1 = L, and e1 = 1 . Therefore, the assertion is clear
in this case.
(iii) In the case where w is of rank 1, we can assume that Gi is a subad-
ditive group of the additive group of real numbers R (Theorem 4.7.5). For
each i = 1 , ... , n, we first take a linearly independent base a j I , ... , a1f
of R1/P over R/P and also a set of representatives {0 = ail , ... , aie } of
G1/G . Then we take a /3 E G such that -fl < a, j < /3 (for all i, j). By
Theorem 4.8.6, there are al j , bik E L such that
(1) a;1eR,, (a,jmodPP)=a;j and wk(alj)>/3 (k i).
(2) w.(b,j) = aik, Wk(b;j) > /3 (k 54 i).
Now assume for a moment that > cl jkal jb;k = 0 (c; jk E K, some are
C.jk # 0). Set 8. = mint j k{wl(ctjkatjbtk)} and 6 = min;{8l} . That
a = w1(ciijiki aii ji biI) , implies i1 = 1 , because, if iI 1 , then as for the
inequality w1 (c, j k a, j b. k ) < wl (c, j k ai j b. k ) , we see that this left-hand

side > w(crij.iki )+2/3 ; the right-hand side = w(cr.,jiki)+aiiki < w(cr.jjiki)+/3 ,
which contradicts /3 > 0. Set I = { (i , j, k) l w (c; jk) = 51. As we saw
l
above, (i, j, k) E I implies i = 1 , and wI (>(I j k)EI cl jkal jblk) > 6 as in

(i), which implies a contradiction.
(iv) In the case where w is of finite rank, we use an induction argument
on the rank r of w. Let Q, , Q be prime ideals of height one of R., R,
respectively. Then Q, n R = Q, by the fact that prime ideals in a valuation
ring are linearly ordered and by Theorems 3.7.12, 4.9.3. We classify the set
{ 1 , ... , n} by the coincidence of Ql , and let the classes be II , ... , 1 . Since
nt R. = D is the integral closure of R in L , we see that DS , with S = R-Q ,

is the integral closure of RQ in L. We can assume that Q1 , ... , QS are the
mutually distinct Qj (namely, that i E Ii). Then DS = n(Rf)Q+ (note that
Q' = Q, n DS (i = 1 , ... , s) are exactly the prime ideals lying over Q) .
Let e' , f' be, respectively, the ramification exponent, the degree of the
valuation defined by (Rf)Q, over the one defined by RQ. Then we have
E ei f' < [L: K], f' _ [DS/Qi : RQ/Q] . Let be the ramification exponent
of the valuation defined by R j/Qi (j E I!) over the one defined by R/Q
(note that the degree is fj) . Then, by the induction hypothesis, EjEI f. <
[DS/Q1: RQ/Q] = f' Therefore, it remains to show that e,ej" = ej (j E I.)
. .
But this follows from the following fact. Set {wj(b)lb is invertible
in (Rj)Q } and G" = {wblb is invertible in RQ}. Then ej" = #(G"J/G")
and the value groups of the valuations defined by (Rj)Qj , RQ are Gj/G",
GIG", respectively.
(v) For the general case, let gl , ... , g,,, be elements of L such that
L = K(g1 , ... , gm) . Take a subfield Ko of K such that these g, are
algebraic over K0. Let w0 , w0, be the restrictions of w to KD , of wi
to Lo = K0(g1 , ... , gm), respectively. Then w0i are prolongations of w0 .
Therefore, we take K0 above so that it is finitely generated over the prime
field and such that K0 satisfies the following 4 conditions:
(1) i 54 j implies w0i # w0i . (For each pair (i, j), take h.j E L such
that h .1 E P , h. P , and choose K0 so that h.j E L0 . )
(2) A representative of every residue class in GjIG is in the value group
of w0i (For each class, take a representative aid and an element bij such
.
that wl (b11) = a,j. Then we choose K0 so that b,1 E Lo.)

(3) For each i, K0 is chosen so that L0 contains a set of representatives
of a linearly independent base of Rj /P over RIP.
(4) [L0: KO] = [L: K].
By the restriction of w1, w to L0 , K0 , respectively, e, , f may become
bigger, but, cannot become less. By (iv), we have >2 e. f < [L: K] with new
e. , f , and we also have the inequality for the original e. , f . Q.E.D.
Here we shall give some sufficient conditions for the inequality above to
be an equality.
THEOREM 4.9.7. Let L, K, w,, w, R., R, P, P, e,, f , G, G. be as
above.
(i) If w is a discrete valuation (for the definition, see exercise 4.7.7) and
if D= n; R. is a finitely generated R-module, then >2 e, f = [L: K] .
(ii) If w is a discrete valuation and if L is separably algebraic over K,

then >2 e1 f = [L: K].
(iii) If P is not a principal ideal and if D is a finitely generated R-module,
then ei = 1 (for all i) and >2 f,. = [L: K].
(iv) If L = K(a) with a E D such that the discriminant dg of the minimal
polynomial g(x) = xd + c1 xd -1 + + cd for a does not belong to P, then
e. = 1 (for all i) and >2 f,, = [L: K].
Before proving this, we prove the following theorem, which is called the
lemma of Krull-Azumaya (some people call this Nakayama's lemma):
THEOREM 4.9.8. Let J be the intersection of all maximal ideals of a ring

R. (This J is called the Jacobson radical of R). If M is a finitely generated
R-module, N its submodule, and if M = N + JM, then M = N.
PROOF. The case N = 101. Let m1 , ... , ms be a set of generators of
M. Since M = JM, m. = E, j,1mt with j,, E J (i = 1, ... , s). These
equalities form a system of homogeneous linear equations in m , ... , ins
1
168 IV. VALUATIONS
with coefficient matrix (8;t - ,alt) , where 8;t is Kronecker 6. Let d be the
determinant of this coefficient matrix. Then d - 1 E J ; hence, d is not in
any maximal ideal, i.e., d is invertible in R. On the other hand, d mi = 0
(for all i) by Lemma 3.7.1, and ml = 0. Thus, M = {0} .
For the general case, let M = MIN. Then M = JM, and M = 101,
namely, M = N. Q.E.D.
PROOF OF THEOREM 4.9.7. (i) We take a generator a (> 0) of G. If
a' E G. , then e,a' E G. This implies that w, is also discrete. Take
a generator a.(> 0) of G.. Then e.a, = a. Choose al c D so that
wi(a;) = a , wj(a;) > a (j 0 i) (approximation theorem). Take also a
.
set of representatives b. (j = 1, ... , f) of a linearly independent base

of R./PP over R/P so that wk (b, J) > a (k 54 i) (approximation theo-
rem). By the proof of Theorem 4.9.6, we see that > eif elements a' b1j
n ; t = 0, ... , et - ; j = 1, ... , f) are linearly independent
1
over R. Let N be the R-module generated by these al b11 . Our choice of

a. , b,3 shows that N + PD = D, and therefore, N = D by Theorem 4.9.8.
Thus > e1 f = [L: K] .
(ii) Since w is discrete, R is Noetherian (Exercise 4.7.7, (i)) and integrally
closed (Corollary 4.7.3). Therefore D is finitely generated (Corollary 3.9.3).
Thus, the assertion follows from (i) above.
(iii) We take a set of representatives b,3 (j = 1 , ... , f) of a linearly
independent base of R,/P, over R/P so that wk(b,J) > 0 (k 0 i) for each
i = 1 , ... , n (approximation theorem). Then we see that > f elements
b. (i = 1 , ... , n ; j = 1 , ... , f) are linearly independent over R, as in
the proof of Theorem 4.9.6. Let N be the R-module generated by these
Then we have N+ n Pi= D. Since P is not principal, by the linearity
b.1.
of the order defined by the inclusion relation of the family of ideals, we

haver PRI = Pi, consequently, n Pl = n PR, = PD, and N = D. Thus
>f =[L:K]
(iv) dgD C R[a] C D by Theorem 3.9.2. Since dg is invertible, we have
D = R[a]. Since dg is not in P, g(x) _ (g(x) mod P) has no multiple
root. Then the factorization g(x) = gl (x) g,.(x) with irreducible g, im-
plies the direct sum decomposition D/PD = R[x]/(PR[x] + g(x)R[x]) =
(R/P)[x]/g(x)(R/P)[x] = E(R/P)[x]/g1(x)(R/P)[x] (direct sum). Thus
E f = [L: K]. Q.E.D.
10. The product formula
We say that the product formula holds in a field K with respect to a family
V of valuations of K, if V consists of infinitely many, nontrivial, mutually
nonequivalent multiplicative valuations and satisfies the condition that for
any nonzero element a of K, there are only a finite number of v E V such
that va # 1 , and furthermore, 11VE V va = 1 (though this is formally an
infinite product, this is in practice a finite product).
10. THE PRODUCT FORMULA 169
THEOREM 4.10.1. (i) The product formula holds in the rational number
field Q with respect to V = {vp Ip runs through prime numbers and oo},
where v.(x) = JxJ and vp with p a prime number is the p-adic valuation
such that vp(p) =p -1 .
(ii) The product formula holds in the rational function fields in one variable
K(x) over afield K, with respect to V = {vp Ip runs through oo and monic
irreducible polynomials in K[x]} , where vo. and vp with a monic irreducible
polynomial p are defined as follows.
We fix a positive real number c such that c < 1 Then v.0 is the x-1-adic
.
valuation with respect to K[x-1 ] such that v. (x-1) = c ; vp is the p-adic

valuation with respect to K[x] such that vp(p) = cdegp
PROOF. (i) Let 0 a E Q. vp(a) 54 1 with a prime number p implies
that p is a factor of either the denominator or the numerator in a fractional
expression of a. Therefore, the number of v in V such that va 1 is
finite. Since (11.EV va)(fJVEV vb) = 11VEV v(ab) it suffices to show that
,
l lv E V v a = 1 in the case a is a prime number or 1 . Since the equality

is clear if a = 1 , we assume that a is a prime number. In this case,
vp(a) = 1 if p is a prime number a ; va(a) = a-1 ; V. (a) = a. Therefore,
[TVEV v(a) = 1 .
(ii) By similar reasoning, it suffices to show that rivEV va = 1 in the case

where a is either a nonzero element of K or a monic irreducible polynomial
in K[x]. In the former case, va = 1 for any V E V . In the latter case,
va(a) = cdega , v.(a) = c- delta , and va = 1 for other v E V. Thus, we see
that f IvE V va = 1 . Q.E.D.
THEOREM 4.10.2. Let L be a finite algebraic extension of a field K. If the
product formula holds in K with respect to a family V of valuations, then the
product formula also holds in L with respect to a family V', which contains
a suitable equivalent valuation v" to each prolongation v' of every V E V.
PROOF. If we see the validity of the assertion in an extension of L, then
we see that assertion with respect to the family of the restrictions of these
valuations to L. Therefore, we can assume that L is a normal extension
of K. Set G = G(L/K). For each v E V, we take all prolongations
V1 , ... , vs of v to L. Theorem 4.9.1 shows that vi, vi are conjugate
to each other and for some o E G, vi = via (i.e., v = vi(aa) for any
a E L). Set Hi = {o E Gl vi = via}. Then Hi, Hj are subgroups
conjugate to each other. Set f = #(Hi). Then #(G) = fs. We choose
v f , ... , vg as the members of V' corresponding to v. Then the valid-
ity of the product formula is seen as follows. Set q = [L: K]i and take
O a E L . Then fJi of (a) =v Since (rLEG aa)' E K,
1(111EGaa).
O flvEv V (NL/K(a)) = 1 Now because

(l lv'Ev' v'a)" = [L Ev V (faEG aa)q .
elements of G are positive real numbers, and because 1 is the unique

170 IV. VALUATIONS
positive root of unity, we see that ft'' v'a = 1 . Q.E.D.

11. Hensel's lemma
To give a formulation of the theorem called Hensel's lemma, we need the
notion of resultant, which we explain first. However, if we state the theorem
as in Theorem 4.11.5, we do not need the notion. So, we state the notion for
the completeness of the theorem.
When two polynomials g(x) = cox'n + clxin-1 + + c,n , h(x) = dox" +
xn-1
d1 + + d,, in one variable x with coefficients ci , di in a ring R are
given, we denote by A(g, h) the following square matrix of degree m + n
co c1 ... Cm 0 ... ... ... 0
0 ... cm-1 cm 0 ... ... 0
1 co
n rows
0 ... 0 c0 c1 ... ..
A(g,h)= do d1 ... do-1 do 0 ..
Can
0
0 do ... ... do-1 d 0 0 m rows
0 ... 0 do d1
The determinant det A(g , h) of this matrix is called the resultant of g

and h .
THEOREM 4.11.1. With symbols as above, we assume that R is an integral
domain and codo 0. Then
(i) Let the (i, m+n)-cofactor of A(g, h) be ai . (1) If g(x) (or h(x)) is a
monic polynomial, then d E (gR[x]+hR[x]) nR iff Ele1 , ... , e,n+n , S, t E R,
such that t : 0, sai = tei (for all i) and d = c,ne,, + dne,n+n (2) The if
part of this holds without assuming that one of g, h is monic.
(ii) Let a , ... , am and p1 , ... , fl n be the roots of g, h, respectively (in
1
an algebraically closed field containing R). Then, det(A(g, h)) _

nm /
codo Ili,i(/3 - a.).
PROOF. First we prove (i) under an additional assumption that
det(A(g, h)) 54 0. (1) d E (gR[x] + hR[x]) n R if d E R, d = gk + hk'
(k , k' E R[x]) . k' can be replaced by the residue of k' divided by g ,
which is monic. Therefore, we may assume that deg k' < deg g . Hence
deg gk = deg hk' < deg g + deg h and
deg k < deg h.
Write k = + ... + e,,-1x + en' k'
elxn-I + ... + em+,, (ei Ee,t+lxm-I
=
R). The coefficients of xn-'g (i < n) form the ith row of A(g , h) , and
the coefficients of xm-'h form the (n + i)th row of A(g, h) . Therefore,
that d = gk + hk' E R means that if we multiply the ith row by ei and
take the sum of these, then (a) every entry except the last one is 0 and (b)
d = c,nen + If we write down this fact in the form of a system of
linear equations in these ei , then we see by the property of cofactors that the
11. HENSEL'S LEMMA 171
solution is (e1 , ... , d ' (a1 , ... , (5m+n)/(det(A(g, h))) ; hence, s, t

are such that st-1 = d/(det(A(g, h))) , i.e., td = s (det(A(g, h))). (2) is
obvious by the proof above.
(ii) Let c1 , u1 , ... , u,n , d1 , v1 , ... , vn be indeterminates and set g'(x)
= c1 fI. (x - u,) , h' (x) = d 1 f J (x -vj) , which are polynomials in x with coef-
ficients in R[c1 , d1 , u1 , ... , u,n , v1 , ... , vn] . Then we consider A(g', h') .
Obviously, det(A(g', h')) = ci'dn det(A(ci 1g', di lh')), and we observe
A(cl 'g', di h') . If the (i, j) entry of this matrix is not zero, then it is
1
homogeneous in these us or vt , depending on whether i < n or i > n,

and of the form (elementary symmetric form), whose degree is -i + j
or -(i - n) + j, depending on whether i < n or i > n. Therefore,
det(A(ci g' , di 1h) is a homogeneous form in u1 , ... , u,n , v1 , ... , vn of
1
degree m n .
Next, we shall show that det(A (c, 'g', d; h') is divisible by fJ,, l (vj -
1
u,) . Indeed, we substitute one vjo by ulo , then g' , h' are changed so
that they have a common root u . Then the system of homogeneous linear
.0
equations whose coefficient matrix is the modified A(ci g' , di 1 h) has a

1
solution (ulo+n , u o+n-1 ulo , 1). Hence det(A(c1 g' , d- h') vanishes
1 1
by Lemma 3.7.1. Therefore, by the factor theorem, det(A(c1 g' , d1 1 h) is

1
divisible by vjo - ulo as a polynomial with coefficients in the field generated

by these u,' vi other than ui0 , vj0 over the field of fractions K of R.
Consequently, det(A(c, 'g', di 1 h')) is divisible by vjo - u;o as a polynomial
over K . Since K[u1 , ... , Um , v1 , ... , vn] is a unique factorization domain,
we see that the determinant is divisible by f I (vi - u;). Both forms are of
degree mn, and therefore, we see that the quotient is an element of K.
Comparing the coefficients of f J vin , we have
det(A(c, 1g', di 1 h')) = [j(vj - u,),

det(A(g', h')) = 11(v, - u,).
Since A(g, h) is obtained by the substitution c1 = co , d 1 = do, u1 = aj ,
vi = /3i , we see the validity of (ii).
Now we consider (i) in the case that the resultant is 0. By (ii), we see
that there are i, j, such that a1 = /3j , which means that g, h have a
common factor g*; hence, (gR[x] + hR[x]) n R c g*R[x] n R = 101.
Therefore, the only if part is proved by e. = 0 (for all i) . Conversely, since
det(A(g, h)) = 0, st-1(a1 , ... , 6,n+n) . det(A(g, h)) = (0, ... , 0). Thus,
d=0. Q.E.D.
COROLLARY 4.11.2. With symbols as above, the resultant of g, h belongs
to (gR[x] + hR[x]) n R.
172 IV. VALUATIONS
THEOREM 4.11.3. Let d be the resultant of polynomials g(x), h(x) in

one variable over a ring R. If g', h' are polynomials with the same degree
as g, h, respectively, and if all the coefficients of g - g' and h - h' are
divisible by dm with m E R, then the resultant d' of g', h' is divisible by
d and d - d' E d mR In particular, if furthermore, m is in the Jacobson
.
radical of R, then d'R = dR.

PROOF. The first assertion is clear by the definition. As for the last asser-
tion, d'R c d R , d'R + mdR = d R and therefore d'R = d R by Theorem
4.9.8. Q.E.D.
Now we shall state Hensel's lemma:
THEOREM 4.11.4. Let R be a complete valuation ring with maximal ideal
P. Assume that f (x), g0(x), ho(x) are polynomials in x over R such that
g0(x) is monic, and d_2(f(x) - g0(x)h0(x)) E PR[x] with the resultant d
of g0(x), h0(x) . Then there are g(x), h(x) E R[x] such that (1) g(x) is
monic, (2) f = gh, (3) g(x) - g0(x), h(x) - h0(x) are in dPR[x].
REMARK. Usually, Hensel's lemma refers to the case in this theorem where
d is invertible in R. This case is reduced to the case g, h are monic and
gR[x] + hR[x] + PR[x] = R[x]. Then the assertion is easily proved without
using the notion of the resultant.
For the proof of the theorem above, it suffices, in view of Corollary 4.11.2,
to show the following theorem.
THEOREM 4.11.5. Let R be a complete valuation ring with maximal ideal
P. Assume that f (x) , g0(x), h0(x) E R[x] and (i) g0 is monic, (ii) there is
-
d E R such that d E g0R[x] + h0R[x] + dPR[x] and f g0h0 E d2PR, then
there are g(x), h(x) E R[x] such that (1) g is monic, (2) f = gh, and
(3) g - g0 E dPR[x], h - h0 E dPR[x].
PROOF. Let v be a non-Archimedean valuation whose valuation ring is
R. For each polynomial k(x) = c0xr + + cr , we define v(k(x)) to be
mint {v c1 } . Starting with go, h0 , and d0 = d, we shall construct sequences
of polynomials gl (x) , hi (x) and elements d; which satisfy the following
requirements.
(1) vd > vd1 > . > vd1 >
. . , limt_,0 vdi = 0, (2) every gi is
monic, (3) for each i , v(g; - gi_1) < vd, , v(hi - hi- 1) < vdl (consequently,
v(gl - g0) < vd, v(hl - h0) < vd), (4) v(f - g,hi) < v(d?) , (5) degh, <
deg f - deg go.
Indeed, if we find such sequences, with gl = xr + c,1 xr- I + + c1r ,
hl = ei0xs+eiixs-1+ +eis (r = degg0, s = deg f-degg0) , then (3) implies
that the sequences {c,3 j i = 0 , 1 , 2 , ... } , {e,, j i = 0 , , 2 , ... } are Cauchy
1
sequences, and using limits c* = lim1 cij and el = lim1 eij , we see that
xr- I
g = xr + C1 + + cr , h = eo xs + + es satisfy the conditions (1), (2),
(3). Now we show how to construct such sequences. Take e E P such that
f - goho E d 2eR[x] , and set di = de'. Then the requirement (1) is satisfied.
Assume that gr , h, are constructed for t < i. Since f - gihi E d. R[x],
f - gihi = d2e21elki(x) (e' E R, ki(x) E R[x]). It follows from (ii) that
d E g0R[x] + h0R[x] + dPR[x] = giR[x] + hiR[x] + dPR[x]. Therefore,
there are 0i , Y/., i E R[x] and e" E P such that
d=gi0i+hiy/i+de
We substitute d in the right-hand side by this equality, and we repeat

the substitution. Then e" is changed to any power of the original el' .
Thus, we may replace e" with e2 , and we have f - gihi = de2iel (gioiki +
hiy/iki + Since gi is monic, we replace wik, and ik1 with their
residues w, and ; , respectively, divided by gi. Then we get an equal-
ity f - gl.h1. = de2'e'(gli
l
O' + hi 4i + de2 with 0'i , yr'l , l E R[x] such
that deg y/' < deg gi = r , deg ' < r. Set g;+ 1 = gi + de 2ie; y/, , hi+ 1 =
hi + de2ie'c' . Then (2) and (3) are satisfied and
f- 2 4i
O// + d e eie
gi+lhi+1 = -d e ei Oiv/
2 , ,
12
2i-2 ,2 / /
2 2i , 2 ,
i
= di+1(ei i - e ei CbiVi).
Thus the proof is complete except for the case i = 0. In the case i = 0,
we can choose eo to be e, and f - g1 h 1 E di R[x]. Therefore, these gi, hi
are the required polyomials. Q.E.D.
A ring R is called a Hensel ring if R satisfies the following two conditions.
(i) R has only one maximal ideal (namely, the set of noninvertible ele-
ments forms an ideal).
(ii) Denote the maximal ideal by P. Then if, for a given monic polynomial
f (x) over R , there are monic polynomials go(x) , ho(x) such that go and
h0 have no common factor modulo P and f - goho E PR[x], then there
are monic polynomials g(x), h(x) such that f = gh and g - go E PR[x],
h - h0 E PR[x].
There are many results known on Hensel rings (see, for instance, Nagata,
Local rings, John Wiley, New York, 1962). In this book, we deal with several
results which are closely related to valuation rings. For this reason, we often
state some results under stronger conditions. For instance, Theorem 4.11.6
below is known to be true without assuming integral closedness.
THEOREM 4.11.6. Assume that an integrally closed integral domain R has
only one maximal ideal M. Then the following four conditions are equivalent.
(i) R is a Hensel ring.
(ii) Every integral domain R', which is integral over R, has only one
maximal ideal.
(iii) If f(x) = X'.+C1X'.-1+ - +cr (ci E R) satisfies c1 V M, C2,...,CrE
M, then f has a linear factor x + c' (c1 - c' E M).
174 IV. VALUATIONS
(iv) If g(x) = xr+dlxr-I +- - +dr (d1 E R) satisfies d,. E M, dr_1 M,

then g(x) has a linear factor x + d' (d' E M).
PROOF. That (i) implies (iii), (iv) is obvious. We shall show that (iii)
implies (ii), and (iv) implies (ii). Assume for a moment that an integral
domain R', which is integral over R, has maximal ideals M1 , M2 (M1
M2). Take an element a of M1 , which is not in M2. Then R[a] has at least
two maximal ideals. Therefore, we may replace R' with the integral closure
of R in the smallest normal extension L of the field of fractions K of R
containing K(a) . Since the number of maximal ideals does not change under
purely inseparable integral extensions, we may assume that L is separable
over K. Let G be the Galois group G(L/K) and let M1 , ... , MS be the
maximal ideals of R'. Then M1 , Mi are conjugate to each other (Theorem
3.7.11).
Set H = {a E GI aM1 = M1}, R" = {x E R'Iax = x (for all a E H)}
and M" = M, n R". Then H = G(R'/R") and M1 lies over M". Hence,
prime ideals of R' lying over M1' are conjugate to M1 over R". Elements
of H leave M1 invariant, and we see that M1 is the only prime ideal of R'
lying over M". Thus, M" Mi' (for i 0 1). Therefore, there are elements
c, d such that c E M2' n . . n Ms' , c M"; d c M", d 0 M,' (for i 1)
.
(Lemma 1.4.5).
Let f (x) , g(x) be the minimal polynomials for c, d over R, respec-
tively. Let al = 1 , a2, ... , ar be a set of representatives of G/H. Since
ac = c (for all a E H) , the conjugates of c are a1 C, ... , arc (or mutually
distinct members out of these; as is seen later, these are exactly the conju-
gates of c). By our choice of c, we have c c Mi' C M1 (for all i 1).
Since a^ : M1 (for j 54 1), a^ = M1 for some i. Thus, ajc E M1
(for j : 1) (and therefore, a1 c , ... , arc are the conjugates of c) This fact
.
implies that the elementary symmetric form of these conjugates of degree 1

is a unit and those of degree at least 2 are all in M1 . Therefore, f (x) is of
the form stated in (iii). Thus, (iii) implies (ii). Similarly, we see that g(x)
is of the form stated in (iv), and we see (iv) implies (ii).
Now it remains to show that (ii) implies (i). Assume for a moment that
there are monic polynomials f (x) , go(x), ho(x) E R[x] such that
f - g0h0 E MR[x], g0R[x] + h0R[x] + MR[x] = R[x]
but, there is no factorization f = gh such that g - g0 E MR[x], h - h0 E
MR[x]. If f is reducible over R, then the same happens for at least one
factor of f. Therefore, we can assume that f is irreducible. Since f is
monic and since R is integrally closed, f is irreducible over K (cf. exercise
3.7.5), and therefore, R' = R[x]/fR[x] is an integral domain, which is
integral over R. R'/MR' R[x]/(MR[x] + fR[x]) = R[x]/(g0h0R[x] +
MR[x]) = R[x]/g0R[x] R[x]/h0R[x] (where denotes the residue class
modulo M) . This shows that R' has at least two maximal ideals, which
proves that (ii) implies (i). Q.E.D.
REMARK. When R' is a Galois extension of an integral domain R, and
P' is a prime ideal of R', we call Hp, = {a E G(R'/R)IQP' = P'} and
{a E R'IQa = a (for all Q E Hp,)} , the splitting group and the splitting ring
of P', respectively. The proofs that (iii) implies (ii) and (iv) implies (ii)
above show the following facts.
Assume that G(R'/R) is a finite group. (1) Hp, # G(R'/R) if the splitting
ring R* of P' has at least two prime ideals lying over P' n R, and in this
case, (2) there are elements a, b E R* whose respective minimal polynomials
are of the forms
.f (x) = xr + C
JCr-1 + ....+ Cr, C1 o P' n R, C2 , ... , Cr E P' n R,
g(x)=xr+dlxr-1+...+dr, d,.-1 P'nR, d,EP'nR,
where r = [G(R'/R): Hp,] = the degree of the field of fractions of R* over
that of R.
Assume that a given integrally closed integral domain R with field of frac-
tions K has only one maximal ideal P. Let K* be the separable algebraic
closure of K and R* the integral closure of R in K*. Let P* be a max-
imal ideal of R* , and let k be the splitting ring of P*. Furthermore, set
P = P* n k, R = RP . Since maximal ideals of R* are conjugate to each
other, k is uniquely determined up to R-isomorphisms, and we _call R the
Henselization of R. Obviously, the maximal ideal of k is P =PR .
THEOREM 4.11.7. Under the above circumstances
(i) The Henselization R of R is a Hensel ring.
(ii) If yr is a homomorphism from R to a Hensel ring R" and if VIP
is contained in the maximal ideal M" of R", then yr is extended to a ho-
momorphism from the Hen_selization k to R" so that VP c M" Such an .
extension is unique and yiR is a Hensel ring.

PROOF. (i) Set S = R -P. Then the integral closure of R in K* is RS .
Assume that RS has two maximal ideals P*RS, PRS (P*' is a maximal
ideal of R* different from P*). We shall obtain a contradiction. P* and P*'
are conjugate to each other and lie over P. Taking an_ element a of P* that
is not in P*' , we consider a Galois extension R' of j?_ such that a E R' and
#(G(R'/R)) is finite. Let H be the splitting group of P* n R'. Then, by the
natural homomorphism 0 of G(R*/R) to G(R'/R) [restriction of action;
the kernel is G(R*/R')] , we have that a E Hp. implies Ou E H. Thus,
R c (the splitting ring of P* n R') c (the splitting ring of P*) = R and we
have a contradiction. Hence, RS has only one maximal ideal. The number
of maximal ideals does not increase under purely inseparable extensions, and
therefore, R is a Hensel ring (Theorem 4.11.6, (ii)).
176 IV. VALUATIONS
REMARK. In the case of a Galois extension of finite degree, by the corre-

spondence between subgroups intermediate fields, and we see that
"The splitting group is the Galois group over the splitting
ring."
But, here we are dealing with infinite extensions, and what we stated above
is not straightforward (cf. Theorem 6.3.2). Therefore, we need a somewhat
complicated proof.
(ii) yrR is a Hensel ring, because any homomorphic image of a_Hensel
ring is again a Hensel ring. Let K be the field of fractions of R. For
each intermediate field Kj between K and k , we take the integral closure
Ri of R in K> and consider Rj = (Rj)Pn j with only one maximal ideal
M = (P n Rj)Rj. Let F be the family of Rj above such that w is uniquely
extended to a homomorphism of Ri to R" so that yiMi c M" . Since
R E F, F is not empty. If R I c R2 c is a well-ordered subfamily of
F , then Uj F, E F Thus, F is an inductive set and has a maximal member,
.
say Ro .
It suffices to show R = Ro . Assume the contrary. Let K0 be the field
of fractions of Ro and Ro the integral closure of R in K0. Then Ro =
(RO)pnR . Set So = Ro - (P* n Ro). Then the integral closure of Ro in K*
o
coincides with Rso . From the definition of k, Rso has at least two maximal
ideals and therefore there is a finite Galois extension Ko of K0 such that, by
setting Ro = Rso n KK , PP = P* n Ro , the ring Ro has at least two maximal
ideals and the splitting ring RO of PP contains RO Set PO = P* n Ro Then,
. .
with a suitable element a of P* n R0, the minimal polynomial for a over

Ro is of the form x' + cl xr-' + + Cr with cr_ 1 V Po , c,. E PO . Since
R" is a Hensel ring, xr + (y/c1)xr-' + + (yrcr) has a linear factor x - c
(c E M"). Suppose that there is another linear factor x - c' (c' E M"). By
division we have
xr + (Wcl)xr-l +
... + (1Vcr) = (x - c)g1(x)
Then, by substituting x = c', we have (c' - c)g1(c') = 0. But, g, (c') is
invertible, because the constant term of g1 is not in M" (by the fact that
yrcr_I is not in M"). Thus, c = c and the linear factor x - c is unique.
Consequently, yr is uniquely extended to a homomorphism of RO[a] to
(yrRO)[c] so that yra = c, because a is a root of x'* + c1 x'*-' + +C Then ,.
yr is uniquely extended to a homomorphism of RO[a]PfRo[ ] to R" under
the condition yr (maximal ideal) c M". Thus, we have a contradiction if

it is shown that R1 = RO[a]PfRo[ ] is integrally closed, which is included in
the following theorem.
THEOREM 4.11.8. Let R' be a Galois extension of an integrally closed inte-

gral domain R and P' a maximal ideal of R'. Set P = P' n R. Assume that
G = G(R'/R) is a finite group. Let Hp, , k be the splitting group and the
splitting ring of P', respectively. Set P = P' n R and let Pl = P , P2 1 ... , PS
be the maximal ideals of R lying over P. If a E P and if a is not in any
other Pi, then R[a]PnRlal = RP. Furthermore, R/P = RIP, PRP = PRP ,
RP/PRP = RIP.
PROOF. We may assume that P is the unique maximal ideal of R. To

prove that R/P = RIP, set Q = n' .>I Pi. Then P + Q = R. What we
should show is that b E R implies - b E P for some b E R. Therefore,
we can assume that b E Q . If b E P, then we can take b = 0; otherwise, the
minimal polynomial for b is of the form xr + cIx r-I + +c with c1 ,
P,
C2 , ... , Cr E P by the remark after Theorem 4.11.6. Then b + c1 E P, and
we can take b = -c1 .
To prove that PRP = PRP , assume an element a E P satisfies the same
condition as a in the theorem, then the minimal polynomial for a is of the
form xr + d xr-1 + + dr with dr_ 1
1 P, d,. E P. In this case, we have
ai-1
a (ar-1 + +d,._1) E drRP . Since a E P and d,._, V P, + +a,. _ 1 is
invertible in RP . Hence, aRP = drRP c PRP . (This relation is used again
later.) If c E P n Q, then a + c satisfies the same condition as a above.
Consequently, (a + c)RP c PRP and (a + c) - a = c E PRP . On the other
hand, (P n Q)RP = PRP, and we have PRP = PRP.
To prove the first assertion, set R1 = R[a]PfR[a] . P is the unique maximal
ideal of R containing a, and therefore, the integral closure R of R in 1 1
RP coincides with RT where T = R[a] - (P n R[a]). Any maximal ideal

of R1 lies over P ; hence, over P or some P-, and contains a. Thus it
lies over P, and R1 = RP . It remains only to show that b E RP implies
bER1.
(i) In the case where b satisfies the condition on a above, b E PR[b]PnR[b]
c PR [b], and therefore, R +PR [b] = R [b] , which shows that R 1 = R [b]
1 1 1 1 1
by Theorem 4.9.8.
(ii) For any element c E P n Q and an element b as above, we have
b+cER1 by (i), and cER1. Thus we have seen that PnQcR1.
(iii) In the general case, we can assume that b E PRP because R/P =
RIP. Since P c (P n Q)RT with T as above, b can be written as E cr t
with c, E P n Q, t. E T. Then c, E R1 by (ii), and we have b E R1 Thus .
R1 = RP. Q.E.D.
In the rest of this section, we shall consider valuation rings only.
178 IV. VALUATIONS
THEOREM 4.11.9. If a valuation ring R is a Hensel ring and if P is a

prime ideal of R, then RP is a Hensel ring.
PROOF. Let L be an algebraic extension of the field of fractions K of R.
Then the integral closure R of R in L has only one maximal ideal. Hence
R is a valuation ring by Theorem 4.9.3. Set S = R - P. Then the integral
closure of RP in L coincides with Rs . Since R is a valuation ring, Rs is
also a valuation ring (Corollary 4.7.3), and Rs has only one maximal ideal.
Thus, RP is a Hensel ring by Theorem 4.11.6. Q.E.D.
THEOREM 4.11.10. Let R be the henselization of a valuation ring R and
let P , P be the maximal ideals of j?, R, respectively. Then, PR = P,
R/P = R/P and furthermore every ideal of R is generated by elements of R.
Consequently, if a valuation w defined by R is a prolongation of a valuation
w defined by R, then the value group of w coincides with that of w .
PROOF. In view of Theorem 4.11.8, we have only to show that b E R
implies bR = aR for some a E R.. We may assume that b is integral over
R. Consider K*, R*, P*, P, R as in the definition of Henselization. Let
R' be the smallest Galois extension_ of R containing b, and let R_' be the
splitting ring of P' = P* n R'. Set P' = P' n k,, R' = RP, , P' = P'R'. Then
R' C R and P = P n R' . Consider prime ideals Pj of R' contained in P',
and let H(P) be the splitting group of P . By the same reason as in the
proof of Theorem 4.11.9, we see that Pi C P implies H(P) 2 H(P) (i.e.,
if aP = Pj, then a(R'P) = R' and QPi = P) . Let G(R'/R) = Ho D HI D
J J
HS = H(P') be the subgroups of G = G(R'/R) which are obtained as
H(P), and use an induction argument on s.
If s = 0, then R =k and b E R. So, we may assume s > 1 . For each i
(< s) , we consider all Pi such that H, = H(Pj) ; they form a linearly ordered
inductive set and therefore there is the largest member, which we denote by
P
In the case s = 1 , take an element c of P such that c is not in any
other maximal ideal of k. Then the minimal polynomial for c is of the
form f(x) = x' + clxr-1 + ... + cr (cr E P, cr-1 P, r = #(G/H1))
As we saw in the proof of Theorem 4.11.8, we have that cR' = c,.R'. If
a (E G) is not in HI , then ac P*. Hence, ac - c Po and f (x)
modulo Po has no multiple root. Therefore, if f (x) modulo P' is reducible
over R/(P' n R) , then R' must have at least two prime ideals lying over
PonR, which contradicts G = H(P') . Thus, f(x) is irreducible modulo Po .
Consequently, in the extension(P,R',

'n
R,over R (PfR) , the extension degree
of the residue class fields coincides with that of the fields of fractions, and the
ramification exponent is 1. This shows that principal ideals of are
R generated by R.
Then one of a1 1 b, a1b-1 , say a2 , belongs to R . Then a2 (Po n R')R'

Prime ideals of k containing a2R' are not contained in valuation rings
obtained as R(apf1R,) (a c H1) . Now, we can apply the approximation
theorem and see that for some b' E R' , b' satisfies the condition for _c
above and b' - a2 E a2P' . Then, by what we have proved above, b'R' = dR'
for some d E R. Thus, bR is either a1 d R or a i d - which completes
the proof of this case.
For s > 1 , we apply the same argument as above to the splitting ring R1
of Pi P. What we should note here is that if a" E Pl = P* n R is not in any
1
other maximal ideal of R , then the minimal polynomial for a" may have
1
degree less than r , but is of the form x' + d1 xt-1 + + d1 with dt E P ,

d1-1 V P. Indeed, this follows from the facts that maximal ideals of R'
lying over P1 are {QP' J or E H1 } and that since H1 D HS , G H1
implies za" 0 aP' (da E H1) Thus, as in the case s = 1 , we see that
.
ideals of R Cpl nR,) are generated by elements of R. Therefore, we may

1
replace R with R1(P,nR,). Then we have the case with smaller s and the
proof is completed by induction Q.E.D.
In the case of rank one valuation, there is an interesting relation between
completion and Henselization as follows.
THEOREM 4.11.11. Let R be the valuation ring of a non-Archimedean val-
uation v of a field K and let R , R be the Henselization and the completion
of R, respectively. Then R can be identified with the set of separably algebraic
elements of R over R.
PROOF. The canonical inclusion mapping R --> R can be extended to a
homomorphism from k to R (Theorem 4.11.7), and we may regard k to
be a subring of K. Since every element of k is separably algebraic over
R, we have only to show that if a E R_ is separably algebraic over R, then
a E k. The prolongations of v to R , R are also denoted by the same
symbol v .
Let k be the field of fractions of R and set R' = R n K(a) . Then R'
is a valuation ring, j?' C R' CR and the valuations defined by F?, R', R_
share the same value group. On the other hand, the integral closure R" of R
in K(a) has only one maximal ideal. Hence, R" is a valuation ring, which
coincides with R'. Thus, a is integral over R . Let f (X) = x" +Cix n-1 +- - +
c" be the minimal polynomial for a over R. If n = 1 , then a E R. Assume
for a moment that n > 1 . Since a is separable, the discriminant d of f (x)
is not 0. Take a nonzero element e of R such that ve < vd. Since a E A,
a - b E elk n R' = eR' for some b E R. Since d is also the discriminant
of the minimal polynomial for a - b, we may replace a with a - b. Thus,
we may assume that va < ve. Let the conjugates of a over k be a1 = a,
a2 , ... , an, and let v' be the unique prolongation of v to K(a1 , ... , an) .
180 IV. VALUATIONS
Then v'(a1) = v'(aj). On the other hand, d = fjj,j(aj - aj )2 , and we have

ve < vd = 11j<; v'(a1 - aj )2 < (v'a)2 < ve, contradiction. Thus n = 1 .
Q.E.D.
We can prove the following theorem from the result above.
THEOREM 4.11.12. Let R be a Hensel valuation ring. Consider polyno-
mials f(x), go(x), ho(x) in one variable x over R and an element d
of R. Assume that go is monic. Q denotes (i) the maximal ideal P of
R if d is invertible in R , (ii) d R otherwise. If f - g0h0 E d 2QR[x] ,
,
d E g0R[x] + h0R[x] + d QR[x] , then there are g(x), h(x) E R[x] such that
-
f = gh, g is monic, g - g0 E d QR[x] , h h0 E d QR[x] .
PROOF. (i) In the case where Krull dim R = r < oo, we use an induction
argument on r. The assertion is clear if r = 0. In the case r = 1 , we can
choose such g, h with coefficients in the completion R of R (Theorem
4.11.5), and it suffices to show that these coefficients are separably algebraic
over R (Theorem 4.11.11). Let al, ... , an ; $1 , ... , /3S be the roots of g,
h respectively. Then, a , ... , a,, , 91 , ... , /3S are the roots of f . Since
, I
the coefficient of the highest degree term of f coincides with that of h , we

see that aj, /3j , and therefore the coefficients of g, h , too, are algebraic
over R. If there is no pair (i, j) such that a . = /3j , then the factorization
f = gh is obtained over the separable closure of the field of fractions of R ;
hence, the coefficients' of g, h are separable. So we assume for a moment
that there is a common root, say a (= a, = /3j with some i, j) of g, h.
Let k(x) be the greatest common factor of g, h (over W). Then,
d E g0R[x] + h0R[x] + dQR[x] = gR[x] + hR[x] + dQR[x]
c kR[x] + d QR[x]
implies that d = k(x)k'(x)+dgf'(x) (k', f' E R[x], q E Q) By substitut-
.
ing x = a, we have d = d q f' (a). Let w be the prolongation to K(a) of an

additive valuation defined by K. Since a is integral over W, w(f'(a)) > 0.
Thus, wd = wd + wq + w(f'(a)) > wd + wq > wd, a contradiction. Thus,
g and h have no common root, and this case is proved.
Now we assume that r > 1 . Let Q1 be the prime ideal of height one of
R. R/Q1 is a Hensel valuation ring, because it is a homomorphic image
of a Hensel valuation ring. Therefore, our induction hypothesis implies that
there are g1(x), h 1(x) E R[x] such that g1 is monic, f - g1 h E Q1 R[x] ,
1
g1 - g0 E d QR[x] , h1 = h0 E d QR[x] .
In the case QDQI,

d E g0R[x] + h0R[x] + dQR[x] = g1 R[x] + h R[x] + dQR[x].
1
Furthermore, (g1 mod Q1) and (h1 mod Q1) have no common root (indeed,
if there is a common root, then we have a contradiction similar to the one we
observed in the case r = 1) . Let h2 be the polynomial obtained from h by 1
1 1. HENSEL'S LEMMA 181
deleting terms with coefficients in Q1 . Then the resultant d1 of h2 , g1 is

not in Q1 and d1 E g2R[x] + h2R[x]. Since RQ, is a Hensel valuation ring
(Theorem 4.11.9), f - g1 h2 E Q1 RQ, and since d is invertible in RQ, , we
1
see the existence of g , h such that f = gh, g is monic, 9-9, E Q1 RQI [x ] ,

h - h2 E Q1 RQ, [x] (consequently, h - h1 E Q1 RQ, [x]). g, h are in R[x],
because Q1 RQ, = Q1 . Thus, we completed the proof of this case.
The case Q c Q1 is proved similarly using d instead of d above. 1
(ii) In the general case, by the assumption, there are q. E Q (i = 1 , 2)

and polynomials ii(x), q5o(x) , 110(x) , Co(x) (E R[x]) such that
f - goho=d q1
2
, d = g000 + ho yro + d g2Co.
Then for some n , qn = dq' (for some qi E R) . Let Ko be the field generated
by d, q1 , q2, ql , q2 and also coefficients of f , go , ho , q , 0o , wo Co) and
set Ro = R n Ko . Since Ko is finitely generated, Krull dim Ro is finite.
Thus, by our result in (i), the required g, h exist with coefficients in the
Henselization Ro of Ro , and Ro is contained in R. Q.E.D.
We close this section with an interesting application of the theorem above
and the approximation theorem.
THEOREM 4.11.13. Let R be a Hensel valuation ring of a field K such that
R # K. Suppose that, for any nonmaximal prime ideal Q of R, the field of
fractions of R/Q has a proper separably algebraic extension. Then if R' is a
Hensel valuation ring of K, either R' C R or R' D R.
PROOF. Assume that neither R' C R nor R' D R Set D = R' n R and let
.
P, P' be the maximal ideals of R, R' , respectively. Let Q be the largest

among common prime ideals of R and R'. Then P n P' is an ideal of D
such that Q c P n P'. By our assumption, there is a separably algebraic
element a over R/Q, which is not in the field of fractions of R/Q. We can
choose such an a so that a is integral over R/Q. Let 7(x) = x" + 1 x ' -1 +
+ c" (c, E R/Q) be the minimal polynomial for a over R/Q. Taking
representatives c1 (E R) of c, , we consider f(x) = x" + c1x"-1 + + c,7 .
f is irreducible modulo Q and therefore f itself is irreducible over R. Let

df be the discriminant of f . Since f is separable, we see that df V Q.
Let r1 , ... , r" E D be such that, for i 54 j , r, - ri 0 Q (note that #(D/Q)
is not finite, and set g(x) = rj;(x - r,) . Let dg be the discriminant of
g. Then dg Q. Write g(x) = x" + d xi-1 + ... + d" . Since d f, dg
1
are not in Q, we see that there are b1 E K with b. - c. E d fP d fR ,
bi - dl Ed P'dgk (i = 1, 2, ... , n) (Theorem 4.8.6). Set h(x) = x" +

2
blxn-1 + + b" . The discriminant of a monic polynomial is expressed as a

polynomial in its coefficients; therefore, it follows from b. - cr E d 2P d t R
182 IV. VALUATIONS
that the discriminant dh of h satisfies (1) dh - df E d fP d fR R. Since

b. - d; E dgP' dg R' , we have similarly that (2) dh - dg E dg P' dgR' .
Since g is reducible over D, we have a factorization g = g, g2 with monic
polynomials g.. The resultant of gi and g2 is a factor of dg, and therefore,
(2) and the theorem above show that h is reducible over D. Consequently,
(1) and the theorem above show that f is reducible over R, a contradiction.
Q.E.D.
Exercises
1
1. Examine under what additional conditions, each of the v below becomes

a multiplicative valuation. In each case, v O is defined to be 0.
(i) v is defined on the rational function field K(x) in one variable
x over a field K in the following way. a is a positive number
and v f = ade,f for 0 f E K(x), where degf is defined as
deg h - deg g if f = h/g with g, h E K[x].
(ii) v is defined on the complex number field C in the following way.
c is a fixed real number and v (a + c(a2 + b2) if a, b are
real numbers.
(iii) v is defined-on the rational number field Q in the following way.
c is a fixed real number and va = j al' for any nonzero element a.
2. Show that every finite field has no nontrivial valuation. Show also that
every algebraic extension of a finite field has no nontrivial valuation.
1. Show that any nontrivial valuation v of a rational function field K(x)

in one variable over a finite field K is as follows:
(i) If vx < 1 , then there is a prime element f (x) of K[x] such that
v is the f (x)-adic valuation.
(ii) If vx > 1 , then v is the x-' -adic valuation with respect to the
prime element x-' of K[x-' ] .
Show also that even if K is not a finite field, nontrivial valuations v of

K(x) , such that the restriction of v to K are trivial, are those obtained
in (i) and (ii) above.
2. Prove that if L is an algebraic extension of a field K and if v is a
valuation of L such that its restriction to K is trivial, then v is trivial
too, as follows.
(i) First show that v is non-Archimedean.
(ii) Then, show that if an element a of L is integral over the valuation
ring R of v, then a E R.
(iii) Show that L = R, by using (i) and (ii) above.
EXERCISES 183
1. Give an example of each of the following.

(i) A topological space that is not a To-space.
(ii) A To-space that is not a T1-space.
(iii) A T1-space that is not a T2-space.
(iv) A T2-space that is not a regular space.
(v) A regular space that is not a normal space.
2. Show that if we give a family S2* of subsets of a set S satisfying the
conditions (i*), (ii*), (iii*), (iv*), given in the beginning part of 3,
then we can define a topology on S so that SZ* is the family of closed
sets.
3. A family S2b of open sets of a topological space S is called a base of
open sets if every open set is expressed as the union of some members of
Ob'
(i) Let 0o be a subbase of open sets of a topological space S. Show
that the family of those expressed as the intersections of finitely
many members of Q0 forms a base of open sets.
(ii) Show that a family W of subsets of S is a base of open sets if and
only if the following two conditions are satisfied:
(1) Every member of W is an open set.
(2) If U is an open set and if P E U, then P E V C U for some
VEW.
4. Let S be a metric space. For each P C S, we denote by UE(P) the E-
neighborhood of P, where e is a positive number. Assume that So is a
dense subset of S and that n , n2 , ... , nr , ... is a sequence of positive
1
numbers converging to 0. Prove that S2b = { (P) I i = 1 , 2, ... , r , ... ;

Uni
P E So} forms a base of open sets of S.
5. Assume that a proposition P on a topological space S is given in terms
of open sets, closed sets, c , = , u , n Then the dual of P is the
.
new proposition P* obtained from P by interchanging open sets, closed

sets, c , u, n with closed sets, open sets, D , n, U, respectively. (For
instance, (ii) in Theorem 4.3.7 is the contraposition of the dual of (i).)
Prove that P is equivalent to the dual P* .
6. Let Si (i = 1 , 2, 3) be topological spaces, and suppose that the map-

pings f : S1 -> S2 and g : S2 -* S3 are continuous. Prove that the
composed mapping g f : S1 -* S3 is also continuous.
1. Show that each of the following groups is a topological group.

(i) The direct product G1 x x Gn of topological groups G1 , ... , G,: .
Namely, its group structure is the direct product of G1 , ... , G,l and
its topology is its topology as the product space of G1 , ... , G,l .
184 IV. VALUATIONS
(ii) The additive group Z of rational integers, where we define a topol-

ogy by each of the following (1) and (2).
( 1 ) The family of all U (m) = { z E Zjm - z is divisible by n},
for natural numbers n and m E Z, is defined to be a subbase
of open sets. (In this case, this family becomes a base of open
sets.)
(2) Fix a natural number N > 1. The family of all U, (m) zE
ZI m - z is divisible by N"} , for natural numbers n and m E Z,
is defined to be a subbase of open sets.
(iii) A subgroup H of a topological group G, where its topology is the
one as a subspace of G.
(iv) Assume that G is a topological group and H is a normal subgroup
that is a closed set. Then we define the residue class group G/H
with topology as follows. Let f be the natural mapping of G to
G/H. Then If UI U is an open set of G} is defined to be a subbase
of G/H.
2. Examine, in (iv) above, what happens if H is a normal subgroup that is
not a closed set.
3. Let G be a topological group and H a closed normal subgroup, and let
f be the natural mapping of G to G/H. Show that it is not necessarily
true that if F is a closed subset of G then f F is a closed subset of
G/H.
4. Prove that if U is an open set of a topological group G and if S is a
subset of G, then SU = {suIs E S, U E U} is an open set.
1. Let v be a non-Archimedean valuation of a field K, and let K* , v

be the completions of K, v. Then the valuation ring RV* of v* is
called the completion of the valuation ring RV of v We say that RV is .
complete if RV = Rv.. Show that

(i) RV* can be constructed from RV by a method similar to the con-
struction of K* from K.
(ii) If a E RV and if a 0, aRv # Rv , then Rv. /aRv. ^_' Rv /aRv .
2. Let R be a ring and let X1 , ... , X,, be indeterminates. An infinite sum
of the form > c1 Xi' X' (i runs through nonnegative integers,
1
c11 E R) is called a formal power series in X1 , ... , X,1 with coefficients

in R. We define addition and multiplication of formal power series by
...Xn
E i,iXl ...Xn + Ed11...1 X1
_ 1:(Ci ...1 +dl ...1 )X 1' ...Xn',

i , i n
EXERCISES 185
X ... Xi (j:di1'**i"XIi' ... Xi,

1 n n )
cli ...i
EE
(ii...i.) je+ke=re
C'1-41 dki ...k
X`. .
n
(i) Prove that the set R[[X1 , ... , Xn]] of such formal power series
forms a ring.
This ring R[[X1 , ... , Xn]] is called the formal power series ring
in X1 , ... , Xn over R.
(ii) Prove that if R is a field and if n = 1 , then R[[X1 ]] is a com-
plete valuation ring and is the completion of the valuation ring
R[X1]X R[xi] .
3. Assume that R is a complete valuation ring whose maximal ideal is gen-
erated by an element n and such that K = R/irR is a perfect field of
characteristic p 54 0.
(i) For each 0 54 a E K, we take a representative an E R of a'
for each natural number n. Prove that the sequence lap } is a
convergent sequence.
(ii) Define q5a to be limn an . Prove that 0 is a group isomorphism
from K-{0} to K*={$aIO#ii EK} (c R).
(iii) Prove that every element a of R is expressed uniquely in the form
00 n
1: ain' with ai E {0} U K* i.e., a = lim E ain'

i=o i=o
(iv) Assume that R is of characteristic 0 and 7r = p . Prove that R

is determined uniquely by the residue class field K . In this case,
if an element a is expressed as > aip' , we associate the infinite
dimensional vector (a0 aP , ... , apn , ...) to a, and we call a Witt
)
vector of length infinity. Each element a of R/p"R is expressed

in the form I:" of aip` , and therefore, a is represented by an n-
,1-I
dimensional vector (a0 ai , ... , ap_ 1 ), which is called a Witt vec-
)
tor of length n. Thus, R and R/p"R are called Witt rings of length
infinity and of length n, respectively. As for the operations among
Witt vectors, the following facts are known.
For each m = 1, 2, ... , there are polynomials fn (xo , x1 , ... ,
x,nIYp,Y1,...,Yin) , gn(x0,x1,... ,x,n,Yp,Y1,...,Yin) in
2m + 2 variables over the prime field of characteristic p, so that
for two Witt vectors
a=(ao,a,,...,an,.... ) and b=(bo,b1,...,bin,...)

186 IV. VALUATIONS
of the same length,

a+b = (ao+bo, f,(a0, al, b0, b1), ... ,
fm(a0,...,am,b0,...,bin),...),
ab=(a0b0,g1(ao) ai) bo,bi),...,
g,n(ao,...,am,bo,...,bm),...),
pa=(O,aO,ap,...5am,...).
P P P
It is also known that the mapping P of R or R/p"R to itself

defined by a' = (ao , aP , ... , a p,, ...) is an automorphism, which
is called the Frobenius mapping.
4. Let v be a complete valuation of a field K. Assume that a finitely
generated K-module M has a linearly independent base u , ... ) un and
I
that a topology on M satisfies the following two conditions.

(i) M is a metric space and is a topological group with respect to its
addition.
(ii) Denoting by IImII the distance between 0 and m (E M) , we have
IIamII = (va)IImMI for all a E K, M E M.
Prove that the topology on M is uniquely determined by these
two conditions, and the topology coincides with the one defined as
follows. For each Ku1 , we give a topology so that the mapping of
Ku1 to K defined by au, -> a is a homeomorphism. Then we
define a topology on M as the product space of Kui , ... , Kun.
6
1. Show that the rational function field C(t) in one variable over the com-
plex number field C has an Archimedean valuation and that its restriction
to C is not equivalent to I (absolute value).
I
2. Characterize fields which have no Archimedean valuation.

3. Give another proof of the uniqueness of prolongation in Theorem 4.6.1,
(ii), by using exercise 4.5.4.
1. Prove that any valuation ring of Krull dimension > 2 is not Noetherian.
(Cf. exercise 4.7.7.)
2. Prove that if Rj (j E M) are valuation rings of a field K, then the
intersection njEM Rj is an integrally closed integral domain.
3. Prove that if an additive group G has a subset P satisfying three con-
ditions below, then we can give an order on G so that G is an ordered
additive group and a E P if a > 0. Prove also that such an order is
uniquely determined by P.
(i) If a, b P, then a +bEP.
(ii) If 0 a E G and a P, then -a E P.
(iii) 0 0 P.
EXERCISES 187
4. Let R be a valuation ring of a field K, and let S be a subring of an

extension field L of K such that R C S. Prove that if a prime ideal
P of S lies over the maximal ideal of R, then S n K = R. Prove
that if, furthermore, S is a valuation ring, then the valuation w of K
defined by R can be extended to a valuation w' defined by S and that
(rankw) < (rankw').
5. Let G and R be the value group and the valuation ring, respectively, of
an additive valuation w of a field K. Let P be a prime ideal of R and
set GP = {wala E K, a P, a-I P1. Prove that
(i) GP is a subgroup of G.
(ii) If a E G, a > 0, a 0 GP , then for all $ E GP and all natural
numbers n, n/3 < a.
(iii) G/GP is naturally isomorphic to the value group of the valuation
defined by RP R.
6. With G as above, a subgroup of G satisfying the condition (ii) above
is called an isolated subgroup of G. Prove that isolated subgroups of G
are those obtained as GP with prime ideals P.
7. An additive valuation of rank 1, or a non-Archimedean valuation is said
to be discrete if the maximal ideal of its valuation ring is principal. In
this case, the valuation ring is called a discrete valuation ring. As for an
additive valuation w of rank n (finite), we call w a discrete valuation
in the weak sense, if RP /Pj_ is a discrete valuation ring for each i =
I
1, 2, ... , n , where R is the valuation ring of w and PO = {0} c PI c

c Pn are the prime ideals of R.
(i) Prove that if R is a valuation ring that is not a field, then R is
Noetherian if R is a discrete valuation ring.
(ii) Prove that an additive valuation w of rank n is a discrete valuation
in weak sense if the value group of w is isomorphic to the n-
ple direct product of the additive group of rational integers with
lexicographic order.
8. Let R be the valuation ring of an additive valuation w of rank n and
let Po = {0} c PI c . c Pn be the prime ideals of R.
. .
(i) Prove that the value group of w is isomorphic to a subgroup of

the n-ple direct product of the additive group of real numbers with
lexicographic order.
(ii) Let Gi be the value group of the valuation defined by RP /P'_ for 1
each i = 1 , 2, ... , n . Prove that if every Gi satisfies the condition

below, then the value group of w is isomorphic to the direct product
G1 x . . x Gn with lexicographic order.
.
Given a E G. and a natural number r n, there is an element

/3 E G; such that m$ = a.
(An additive group satisfying this condition is said to be divisible.)
188 IV. VALUATIONS
9. Let R, G and Gi be as above, without assuming the divisibility con-

dition stated in (ii). We regard each Gi as a subgroup of the additive
group of real numbers. The supremum of the numbers of a set of linearly
independent elements of Gi over the rational number field is called the
rational rank of Gi . The sum of rational ranks of Gi (i = 1 , ... , n) is
called the rational rank of w , or of G. Assume that R contains a field
k. Prove the following inequality, where P is the maximal ideal of R.
trans. degk R > (rational rank of w) + trans. degk RIP.
8
1. Let v1 , ... , vn be additive valuations of a field K. Find necessary and
sufficient conditions on the relation among the valuation rings R , ... , 1
Rn of VI , ... , vn , for the validity of the following.

Given a1 , ... , an, b1, ... , bn E K with b1 . . bn 0 0, there
.
is an element c E K such that vi(ai - c) > vibi .
2. Let R , ... , Rn be valuation rings of a field K. Assume that R has a

1 1
prime ideal P that is not contained in any other Ri . Prove that there is
an element a of P which is invertible in any Ri (i > 1).
1. Let K be a field of characteristic p 54 0. Prove that there is a discrete

non-Archimedean valuation v of a certain field of characteristic 0, such
that v is complete, the maximal ideal of the valuation ring R of v is
generated by p, and such that R/pR is isomorphic to K.
2. With notation as in Theorems 4.9.6-7, assume that R has a prime ideal
Q such that R/Q is a discrete valuation ring. Prove that E ei f. =
[L: K], if there is an element a of L, which is integral over R, such
that L = K(a) and the discriminant of the minimal polynomial for a
over K is not in Q.
10
1. Let L be a finite algebraic extension of the rational number field Q

and let VL be the family of valuations obtained in the proof of Theorem
4.10.2, so that the product formula holds in L with respect to VL . Prove
that the product formula holds in L with respect to a family V' of
valuations if there is a positive number t such that V' = {vÌv E VL}
(This fact shows that VL is substantially unique.)
2. Adapt the statement above to the case where L is a finite algebraic ex-
tension of a rational function field K(x) in one variable over a field K,
which is algebraic over a finite field.
EXERCISES 189
3. In the definition of the product formula, we assumed that V consists of

infinitely many valuations. Prove that if V consists of a finite number
of valuations (dropping the condition "infinitely many"), then V must
be an empty set. (Therefore, we may change the definition by V being
nonempty.)
11
1. Let P be a prime ideal of a valuation ring R. Prove that if both R/P

and RP are Hensel rings, then R is also a Hensel ring. (Note that the
converse is also true.)
2. Prove that every integrally closed integral domain R' integral over a
Hensel ring is again a Hensel ring.
3. Prove that the formal power series ring K[[X1 , ... , Xn]] over a field K
is a Hensel ring.
4. We denote by K((X)) the field of fractions of the formal power series
ring K[[X]] in one variable X over a field K.
(i) Prove that if V is a Hensel valuation ring of K((X)) and if Krull
dim V = 1, then V coincides with K[[X ]] .
(ii) Find the Hensel valuation rings of K((X)).

CHAPTER V
Formally Real Fields
1. Ordered fields, formally real fields, and real closed fields

A field K is called an ordered field if K is a linearly ordered set satisfying
the following three conditions, where a, b, c, d denote arbitrary elements
of K:
(i) If a > b and c > d , then a + c > b + d .
(ii) If a > b, then -a < -b.
(iii) If a > 0 and b > 0, then ab > 0.
If K is a field of characteristic p 54 0, then for a > 0, we have 0 > -a =
(p - 1)a = a + a + + a > 0, a contradiction. Thus we have the following
theorem.
THEOREM 5.1.1. Any ordered field is of characteristic 0.
Let K, K' be ordered fields. A mapping 0 of K to K' is called an
order isomorphism, or an isomorphism of ordered fields, if 0 is not only
an isomorphism of fields but also an order preserving mapping. Namely,
a > b if Oa > Ob. We say that K is order isomorphic to K' or that K is
isomorphic to K' as ordered fields if there is such a 0 with OK = K'.
A field K is called a formally real field if -1 = ai + + an fails to hold
for all natural numbers n and all elements ai E K. (Namely, a field K is
not formally real if there are elements ai E K such that -1 = ai + + an .)
Since 12 = 1 , as in the case of an ordered field, we see that any formally real
field is of characteristic 0. Furthermore we have the following theorem.
THEOREM 5.1.2. Any ordered field is formally real. Conversely, if K is a
formally real field, then we can define an order on K so that K becomes an
ordered field.
The first half of this theorem is clear, since -1 = ai + + an implies
0 > -1 = a2+.
+ an > 0. The latter half will be proved following Theorem

5.1.7 below.
A field K is said to be real closed if K is formally real and if any proper
algebraic extension of K is not formally real. When a field is given, the
family of subfields which are formally real is an inductive set, and therefore
we have
191
192 V. FORMALLY REAL FIELDS
THEOREM 5.1.3. If K is a formally real field, then there is an algebraic

extension of K which is real closed.
LEMMA 5.1.4. For a field K, we set S = {ai + + a2I ai E K, n is a
natural number}. Then (i) for a, b E S, we have a + b, ab c S, (ii) if
0 0 a E S, then a-' E S. Furthermore, (iii) if K is neither formally real nor
of characteristic 2, then S = K.
PROOF. (i) is clear.
To prove (ii), assume that 0 a E S. Then a = a2 + + a2 and
a-1 = as-2 = >2(ai/a)2 E S.
(iii) -1 E S because K is not formally real. For a E K, 4a = (1 +a) 2 +
(-1)(1 - a)2 Thus, 4a E S by (i), and therefore, a c S. Q.E.D.
.
THEOREM 5.1.5. If K is a real closed field, then given any a E K there is

an element b e K with a = b2 .
PROOF. Assume that a c K and a 54 b2 for all b E K. Then L = K(/)

is a proper algebraic extension of K. Therefore, L is not formally real.
Thus, there are elements dl E L such that -1 = d i + + d 2 Write each.
dl as bi+ciN/a` with bi, ci E K. Then -1 = Eb2+a(> c?)+2(Ebici)v/a-,

which implies that 1 + + a(E c2) =0, 2(E bici)v/a =0. Since K is
b2
formally real, l+Eb2,0 0, and -a = (1+Eb?)(Ec?)-' Since (E c2)-1 ,
.
1 + > b2 E S in the lemma above, we have -a E S. If a E S, then

a-1 E S and -1 = (-a)a-' E S, which contradicts that assumption that K
is formally real. Thus, a V S. If -a is not the square of any element of
K, then by the same reasoning as above, we obtain a E S, contradiction.
Therefore, a = -b2 for some b E K. Q.E.D.
LEMMA 5.1.6. If a subset P of a field K satisfies the following conditions,
then there is a unique order defined on K such that K is an ordered field and
P is the set of positive elements.
(i)If a, bEP, then a+b EP, and abEP.
(ii) 0 P.
(iii) If a E K with a 0, then either a E P or -a E P.
PROOF. We define an order by a > b if a-b E P. We shall prove that this
order is a linear order and that K is an ordered field. Set N = { -a j a E P1.
If a E PAN , then a, -a E P and 0 = a+(-a) E P, a contradiction. Hence,
P n N is empty. This means that for a E K, only one of a = 0, a c P,
-a E P holds. Consequently, for a, b e K, only one of a = b, a > b, b > a
holds. It is obvious that if a > b, and b > a , then a = b, where > means
> or =. If a > b and b > c , then a - b and b - c are elements in P, so
a - c = (a - b) + (b - c) E P, and hence, a > c. Thus, we have proved that >
is a linear order. It is easy to see that K is an ordered field under this order
1. ORDERED, FORMALLY REAL, AND REAL CLOSED FIELDS 193
and that P is the set of positive elements. Conversely, if K is an ordered

field such that P is the set of positive elements, then a > b if a - b E P,
and therefore, we see that the order is uniquely determined by P. Q.E.D.
THEOREM 5.1.7. Any real closed field K has an order such that K is an or-
dered field. Such an order is uniquely determined by K. Every automorphism
of a real closed field is an order automorphism.
PROOF. Set P = {a E Kja = b2 for some nonzero element b E K} P .
obviously satisfies condition (ii) in Lemma 5.1.6. Theorem 5.1.5 shows that
P also satisfies condition (iii). As for (i), given a, b c P, we have a = c2
and b = d 2 , for c, d E K with cd 0. Hence, ab E P. Assume for a
moment that a + b P. Then -(a + b) E P U 101, so for some e E K, we
have c2 + d 2+ e2 = 0, a contradiction. Thus, a, b E P implies a+ b E P,
and K is an ordered field.
If K is an ordered field, then every element of P must be a positive
element, since it is the square of a nonzero element. If a nonzero element b
of K is not in P, then -b E P by Theorem 5.1.5. Thus, P is exactly the
set of positive elements. Therefore, the uniqueness of the order follows from
Lemma 5.1.6. Any automorphism of K, maps P to P, and therefore, the
order is preserved. Q.E.D.
PROOF OF THE LATTER HALF OF THEOREM 5.1.2. There is a real closed field
K* that contains K (Theorem 5.1.3) and K* is an ordered field (Theorem
5.1.7). Therefore, restricting the order on K* to K, we see that K is an
ordered field. Q.E.D.
THEOREM 5.1.8. If an ordered field K satisfies conditions (i) given a > 0,
a = b2 for some b c K and (ii) for f (x) E K[x] with deg f (x) odd, there
is an element c e K with f (c) = 0, then K is real closed and K(im) is
algebraically closed.
PROOF. The fact that K(im) is algebraically closed can be shown by the
same way as in the proof that the complex number field is algebraically closed
(Theorem 2.12.4). Since K is an ordered field, K is formally real. Further-
more, the only algebraic extensions of K are K and K(im). Therefore,
K is real closed. Q.E.D.
THEOREM 5.1.9. The conditions (i) and (ii) in the theorem above charac-
terize real-closedness of an ordered field K.
PROOF. We have only to show that (i) and (ii) follow from the condition
that K is real closed. Since the order on K is unique (Theorem 5.1.7), (i)
is straightforward by Theorem 5.1.5. As for (ii), let f (x) = cox" + + c,t E
K[x] (c1 E K, co 0) be a polynomial of odd degree n. To prove the
existence of a root in K , we may assume that co = 1 . We use an induction
argument on n . Existence is obvious if n = 1 , and so, we assume that
n > 1 . If f is reducible, then one factor is of odd order, which has a root
in K by our induction hypothesis. So, we assume that f is irreducible. Let

a be a root of f . Then K(a) is not formally real and therefore there are
elements /31, ... , /3,n E K(a) with -1 = >2 /3? . Then each /3r can be written
as /3i = gi(a) with gi(x) a polynomial of degree less than n. Then the
polynomial 1 + >2 g. (x)2 has a as a root. Hence, 1 + >2 g2 = f (x)h(x) for
some h(x) E K[x]. Since the left-hand side of this equality is of even degree
and since deg g, < n , we see that h(x) is of odd degree < n . Thus, h has a
root b in K by the induction hypothesis, which implies that >2 gi(b) 2 = -1 .
This contradicts the assumption that K is formally real. Q.E.D.

THEOREM 5.1.10. A field K is real closed iff K does not contain and
K(im) is algebraically closed.
PROOF. The only if part is contained in Theorem 5.1.9. To prove the
if part, note that by assumption, irreducible polynomials in one variable
over K are of degree < 2. We shall show that given a, b E K, then
Va2 + b2 E K. This is obvious if ab = 0, so we assume that ab 0 0.
Set g(x) = (x2 - a)2 + b2 . Since deg g = 4 > 2, g is reducible over K.
The squares of the roots of g are a b
a, , /3, V K, because a2
Letting al , a2 , f , 92 be
the roots of g, we may assume that a? = a+ b , f? = a -
V K K. a1 a2 = -(a + b
, /3?
.
) , and therefore,
I
b
a2 is not a conjugate of a1 . Hence, we may assume that /31 is conjugate to
a1 . Then al /31 E K and (a1 /31)2 = a 2 = a2 + b2 Thus, we have shown
1131 .
that Va2 + b2 E K If there are a, E K such that -1 = ai + + an , then

.
what we have proved shows that -1 = a2 for some element a E K. This

contradicts / V K, and therefore, K is formally real. Consequently, K
is real closed, because K(im) is algebraically closed. Q.E.D.
COROLLARY 5.1.11. Let Ko be a subfield of a real closed field K . Then
the algebraic closure L of Ko in K is real closed.
PROOF. Since v/_-_1 is not in L, it suffices to show that L(am) is alge-
braically closed. By the theorem above, K(im) is algebraically closed, and
therefore, any algebraic element over L(am) can be written as a + b
with a, b E K. L(am) is mapped to itself by the K-automorphism of
K(im) such that v'-I -+ -/J. This shows that a - b V '-1 is also al-
gebraic over L(am) Thus, 2a and 2bv '-1 are algebraic over L. Since
.
a,beK,wehave a,bEL. Q.E.D.

LEMMA 5.1.12. If a field K of characteristic p 0 has a cyclic extension
of degree p, then for any given natural number n, there is a sequence of
extensions K = Ko c K1 c c Kn such that each K, is a cyclic extension
of degree p over K._ I .
NOTE. It is known that, under these circumstances, there is a cyclic exten-

sion of degree p n over K. But, we shall not prove this.
1. ORDERED, FORMALLY REAL, AND REAL CLOSED FIELDS 195
PROOF. For any element a of an extension field of K, we denote by

q5a the element a - a. Note the following facts (cf. Theorem 2.9.10): (i)
xp - x - a (a e K) is irreducible over K if a V OK, (ii) if a E K and
b E (the prime field) then b(qa) = q5(ab), (iii) a E (the prime field) if
cba = 0, and (iv) Oa + Ob = q5(a + b).
It suffices to prove that if a E K and a V K with Oa = a, then aap-1
qS(K(a)). (Namely, taking an element a of K such that a V OK, we
choose al E 0-1(a). When we have chosen al , ... , am , we set am =
am_ 1 am with am_ 1 = Dam , and we choose am+l E q)-l(am) . Then K! =
1
K(a1 , ... , al) are the required fields.) Assume that aap-1 E (K(a)). Then
for some b = c0 + c1 a + + cp_ 1 ap-1 (c, E K), aap-1 = by - b. Since
cap = a + a, we have
aap-' cp(a+a)`-E cla`.
Since 1, a , ... , ap-1 are linearly independent over K, by comparing the
ap-1
coefficients of , we have a= p_ 1 - cp_ 1 Thus, a E OK, a contradic-
.
tion. Q.E.D.
THEOREM 5.1.13. Assume that the field K is not algebraically closed and
that the algebraic closure K of K is of finite degree over K. Then K is real
closed and consequently, K = K(im).
PROOF. If K = K(im) , then the assertion follows from Theorem 5.1.10.
Assume for the moment that K(im) c K.
If K is not a perfect field, then the characteristic p of K is not 0 and
Kp' D K. Since Kp ' -= K, it follows that K -A Kp ' # Kp # , and 2
consequently [K: K] = oo. Thus, K is perfect and K is a Galois extension

of finite degree over K.
Set G = G(K/K(im)). Take a cyclic subgroup H of G whose order
is a prime number, say q. Let L be the subfield corresponding to H. If
q = p (the characteristic of K), then Lemma 5.1.12 implies that [K: L] =
cc , which is not the case. Thus q 0 p . Therefore, there is a primitive
q"th root of unity Cn in K for each natural number n. If C1 0 L, then
1 < [L(C1) : L] < q, which contradicts the property that H has no proper
subgroup. Thus C1 L and K = L(Q a) with a E L (cf. Theorem 2.9.8).
2 2
Let a be a root of xq -a. Then xq -a = fl 1(x - C2 a) Since [K: L] = q,

.
2
xq - a is reducible over L. Some roots of this polynomial are not in L.

2
Hence, xq - a has an irreducible factor h(x) of degree q. Let c be the

constant term of h (x) Then c is of form c = aq 2 . Since aq is a root of
.
xq - a, aq 0 L, and therefore, K = L(C2) .

Now we consider the field n (C2) generated by C2 over the prime field
7r . Take the least natural number r such that Cr E 7r (C2) , Cr+1 0 7r (C2)
(The existence of r follows from the fact that [n(C2): iv] < q2 and that
[n(Cr): 7L] is big for large r (cf. Theorem 2.9.6).) The minimal polynomial
r+I
g(x) for Cr+l over L is a factor of fl (x - Cr+I) and therefore, g E
I
1E(Cr+1)[x] . Hence, g E (n((r+1) n L)[x]. Since (,.+I L, degg = q, and

[n(Cr+I) : (n(Cr+1) n L)] = q. Next, we consider C,.. Since r > I , Cr 0 L.
Jt((r+I) is of degree q over n(Cr) , since Cr+l iv(C2) = V(Cr)) Since
Cr 0 L , n(Cr) Z L , and consequently, n((r) Th((r+I) nL . Therefore, G* =
G(n(Cr+1)/7l) has at least two cyclic subgroups of order q. This implies that
G* is not a cyclic group. Thus, Tc is not a finite field, and the characteristic is
0. Namely, 7l is the rational number field Q. Since G* is not a cyclic group,
we have q = 2 (cf. Exercise 2.9.3). By our choice of L, C2 = + V'-1 E L,
which contradicts the property that C2 0 L. Thus, we have proved that
K = K(im). Q.E.D.
2. Real closures
We have seen already that for any formally real field K, there is a real
closed field containing K (cf. Corollary 5.1.11, Theorem 5.1.10). We define
here the real closure of K to be the smallest real closed field preserving the
order on K (cf. Theorem 5.1.2), as follows.
A field K* is a real closure of an ordered field K if (i) K* is an algebraic
extension of K, (ii) K* is real closed, and (iii) the unique order on K*
(Theorem 5.1.7) is a prolongation of the order on K.
To show the existence and the uniqueness of K*, we prove the following
lemma.
LEMMA 5.2.1. If K is an ordered field, then the extension field K' gener-
ated by square roots of positive elements of K is formally real.
PROOF. If there is a relation of the form -1 = a2 + + an in K',
then there are positive elements bl , ... , bin of K such that all ai are in
K( , ... , Fbtn). Thus, it suffices to show that for bI , ... , bin > 0,
K(/ , ... , Fb) is formally real. That is, we have only to show that if
cia2 = 0, where ai c K( , ... , Fbtn) and 0 < ci c K, then ai = 0 for
all i). We shall prove this by induction on m, and we assume that our claim
is valid for K(, ... , J7) and that b,,, V K(,Ib-, , ... , b17_i).
We rewrite the relation > cia2 = 0 by setting ai = ai + /3i bn, with
ai , /3i E K(Vfbl , I) Then we have
cia2 + > Cb,nQ2 + 2 E bin = 0
i.e.,
2
cia +Ecib,n/32 = 0, 2Eciaifi = 0.
2. REAL CLOSURES 197
From the left-hand side equality, we have ai = /3i = 0 for all i. Thus,
ai = 0 for all i. Q.E.D.
THEOREM 5.2.2. If K is an ordered field, then there is a real closure K*.
Real closures of K are essentially unique. In particular, if K" is another real
closure of K, then there is an order isomorphism from K" onto K* over K K.
PROOF. Let K' be the field generated by square roots of positive elements
of K. Since K' is formally real, there is a real closed field K* that is al-
gebraic over K' (cf. Corollary 5.1.11). For a E K with a > 0, we have
Va- E K' , so a = (/)2 is positive in K* This shows that the set P of pos-
.
itive elements in K coincides with K n (the set of positive elements in K*) .
Since the order on K is uniquely determined by P (Lemma 5.1.6), the order

on K* is a prolongation of that on K. Thus the existence of a real closure
is proved. On the other hand, the uniqueness follows from the following
theorem.
THEOREM 5.2.3. Let K, K' be ordered fields having real closures K*,
K'* , respectively. If there is an order isomorphism 0 of K onto K', then
0 is extended to an order isomorphism of K* onto K'*. Such an extension
is unique. In particular, any order K-automorphism of K* is the identity
mapping.
We need some preliminaries for the proof of this theorem.
LEMMA 5.2.4. Let K be a real closed field (having a unique order). Given
f (x) E K[x] and a, b c K with f (a) > 0 and f (b) < 0, then there is an
element ceK with f(c) =0 and either a>c>b or a<c<b.
PROOF. We may assume that f (x) is a monic polynomial, and we con-
sider a factorization f (X) = rli=1 (x - ai) f=1 gi (x) with gi (x) irreducible
polynomials (deg gi > 1) . Theorem 5.1.8 shows that deg gi = 2, and we
can write gi = x2 + dix + ei . Then gi can be written as gi = (x + d') 2 + ei
(2d, = di). If el < 0, then there is an element ell E K, el = -(e1 )2 (Theo-
rem 5.1.4). Then e" - di is a root of gi , which contradicts the irreducibility
of gi. Thus, ej > 0 and gi (c) > 0 for any c E K. Therefore r j gi has pos-
itive value for any c E K . Hence, we may assume that f (x) = rll=1 (x - ai) ,
where a1 < a2 < < ar If there is no ai between a and b, then taking
.
t such that al < < at < a, b < at+1 < ... < ar , we see that both
f (a)(-1)r-' and f (b)(-1)r-t are positive, which is not the case. Q.E.D.
For a given polynomial f (x) E K[x] (K a field and f (x) K)) , the stan-
dard sequence f0 = f , f1 , ... , f, is defined as follows: f1 (x) = (d /d x) f (x)
(i.e., if f = E cix' , then f1 = E icixt-1) When f0 , ... , f are defined
.
(i > 0), then we consider the division f,-, = f qi + ri with qi , ri c K[x],

ri = 0 or deg ri < deg f . Then if ri = 0, let f be the last member (and
s = i) ; otherwise, we set f+1 = -ri .
When a sequence a1., ... , as of elements a. E K (an ordered field) is

given, by the number of variations of signs of the sequence we mean the
number of indices i such that a1 a;+J < 0, where j > 0 is chosen so that
if j > k > 0, then al+k = 0 (in other words, the number of al such that
its nonzero successor al+j has different sign than a1) . Then we have the
following theorem called the Sturm Theorem.
THEOREM 5.2.5. Let f (x) E K[x] (f (x) V K) with K a real closed
field, and let fo , f1, ... , fs be the standard sequence for f (x) For each .
c E K, we denote by Vc the number of variations of signs of the sequence

fo(c) , ... , fs(c). Assume that a < b, f (a) 0 0, and f (b) 0. Then
the number of mutually distinct roots c such that a < c < b coincides with
Va - Vb .
PROOF. fs is the greatest common divisor of fo and f1 . Hence fs(a) : 0

and f(b) # 0. Set gl = f,/J' Then each g, is a polynomial, and we get
the same Va, Vb by using go, ... , gs = Furthermore, each root of fs is a
1 .
multiple root of f and (its multiplicity as a root of fs) + 1 = (its multiplicity

as a root of f) . Therefore, the number of mutually distinct roots between
a and b can be counted by those of go.
Now, we shall show that go, ... , gs satisfy the following four conditions:
(i) gs does not vanish at any c between a and b.
(ii) go(a)go(b) 0.
(iii) If a < c < b, 0 < i < s, and g! (c) = 0, then g!_ 1(c)gi+1 (c) < 0.
(iv) If g0(c)=0 and a < c < b, then there are d 1 , d2 such that d 1 <
c < d2 and
(1) go(x)g1(x) < 0 for any x such that d < x < c,
1
(2) go(x)g1(x) > 0 for any x such that c < x < d2 .
Here we add a definition. A sequence go, ... , gs of polynomials gi is

called a Sturm sequence on the interval [a, b] = {x E Kia < x < b} if it
satisfies the four conditions above.
(i) and (ii) obviously hold.
(iii) Since f _ 1 = f qt - f+1 (qt E K[x]), we have g,_ 1 = gig, - g;+1
Hence, if gt (c) = 0, then gl_ 1(c) = -gi+1(c) . On the other hand, 1 is the
greatest common divisor of gi and g,_ . Therefore, g!_ 1(c) 54 0, and we
1
have (iii).
(iv) Assume that go(c) = 0, a < c < b. Write f = (x - c)eho(x)
(ho(c) 0), then f1 = e(x - c)e-I ho(x) + (x - c)e((d/dx)ho(x)) . Hence,
go(x) = (x - c)h(x) with h(x) E K[x] (h(c) 0), and g1(x) = eh(x) +
(x - c)k(x) with k(x) E K[x]. Since h(c) 0, we have g1 (c) 0, and
we can choose d1 < c < d2 so that both h(x) and g1(x) do not vanish
in the interval [d1 , d2] , because the numbers of roots of h and of g1 are
finite. Lemma 5.2.4 shows that each of h, g1 does not change its sign in
[d1 , Therefore g1h takes positive value in the same interval, because
d2] .
g1(c) = eh (c) Consequently, (iv) follows, because gog1 = (x - c)g1 h .

.
For the proof of our theorem, it suffices to show that the number of distinct
roots of go in the interval [a, b] is the difference of the numbers V, Vb of
variations of signs of go (a) , ... , gs(a) ; go (b) , ... , gs(b) .
If we take a = ao < a 1 < < an = b such that go (ai) 0, then
Va - Vb = 01(Vai - Vas+1 ) where Vas is the number of variations of signs of
go(ad) , ... , gs(a,.) . Therefore, we may count the number of roots of go in
each interval [ai , a,1], after choosing ai so that they satisfy the following
four conditions: (I) In each interval [ai , ai+1 ] , go has either no root or only
one (single or multiple) root. (II) Every root of gi in the interval [a, b] is
either some ai or a root of go. (III) If go(c) = 0 and a, < c < a,+1 , then
ai, ai+1 satisfy the condition on d1 , d2 in (iv) above. (IV) go(ai) 0 for
all i.
Since the number of roots of gi is finite for each i, we can choose such
ai . Now, it remains to show that Va - Va = 1 or 0 depending on whether
the interval [ai, ai+1 ] contains a root of go or not.
( 1 ) The case where go(c) = 0 for some c c [ai, a , ] : By condition (III),
1
the number of variations of signs of go, g1 is 1 at ai and 0 at ai+1 Thus, .
we have only to show that the number of variations of signs of g1 , ... , gs

takes the same values at ai and ai+1 However, g1 does not vanish in this
.
interval by the conditions (iii) and (II), and g1 , ... , gs is a Sturm sequence.
Therefore, our assertion follows from the remaining case (2).
(2) The case where go does not vanish in [ai, a,1]. If there is a gj
(s > j > 0) that does not vanish in this interval, then we obtain the claim
by observing the two shorter Sturm sequences go , ... , gi and gi , ... , gs .
Thus, we may assume that if s > j > 0, then gj vanishes at ai or ai+1 .
When gi vanishes at both ai and ai+1 , then by dividing the interval into
two intervals, we may assume that each gj vanishes at only one of ai , ai+1 .
(i) If s = I , then Var = Var+i = 0, and the assertion is clear.
(ii) Assume that s = 2. If g1(ai) = 0, then Q = 1 (Condition (iii)).
Vaf+, = 1 , because neither of go, g2 vanishes and gog2 < 0 in this interval.
The case g1(ai+1) = 0 is proved similarly.
(iii) If s > 2, then by a suitable division of the interval, we can reduce
to the case where s < 2, because there is no j (0 < j < s) such that
gj(ai) = g1+1(ai) = 0. Q.E.D.
LEMMA 5.2.6. Let K be an ordered field. If a E K is a root of xn+c1 xi-1 +
+cn (ci E K), then -m < a < m, where m = max{ 1, Ic I+Ic2I+ +I cn I } 1
(Ic1 I means the nonnegative one of ci) .
PROOF. If Ial < 1 , then there is nothing to prove. Assume that Ial > 1 .
c2a-1 cna-(n-1
It follows from -a = c1 + + + ) that
cna-n+l l < I + Ic2a-11 + :+1

I - al = Ic1 + c2a-1 +... + ICI ... + I c,,a I
IC I
+ Ic2I + ... + Icel. Q.E.D.
COROLLARY 5.2.7. Let f (x) , f1(x) be as in Theorem 5.2.5 and assume

that f (x) is monic and has no multiple root. Let /31 < fl2 < < /3s be the
mutually distinct roots of f1 (x) in K. Letting m be as in the lemma above,
we take /30 , fls+I so that /30 < min{-m, #,I, PS+1 > max{m, /3s} . Then
the number of roots of f (x) in K is at most s + I, and f (x) has at most
one root in the interval [/31, /3i+1 ] for each i = 0, 1 , ... , s.
PROOF. Since f (x) has no multiple root, the standard sequence

A, ... , f for f coincides with go, ... , gs in Theorem 5.2.5. fo(/31) 54 0
for i = 1 , ... , s, and we can choose al , c1 such that co = /30 < a1 < /31 <
c1 < a2 < Q2 < c2 < . < as < /fs < cs < f3s+1 = as+1 and such that f has
no root in the interval [a1, c1 ] for each i = 1, ... , s. On the other hand,
f1 does not vanish in each of the intervals [c1_ , al] . Then the number of
1
variations of signs of f1 , ... , fs at c1_ is equal to that at a1, which im-

1
plies V. Va < 1 . Thus, f has at most one root in each of the intervals
[c1_ 1, a1] . This completes the proof, since f is assumed to have no root in
each of the intervals [al , c,] . Q.E.D.
PROOF OF THEOREM 5.2.3. For f (x) = coxn
+c1xn-I
+ +cn E K[x], we
denote by Of (x) the polynomial (q5co)x" + (gc1)xn- I + + c,t . Note that
if fo, f1 , ... , fs is the standard sequence for f (x), then q5fo , Oft , , Ofs
is the standard sequence for Of (x) .
When a monic polynomial f (x) is given, taking m in Lemma 5.2.6, we
apply Theorem 5.2.5 to the case a < -m, m < b. Then we see (i) and (ii)
below.
(i) (The number of mutually distinct roots of f(x) in K*) = Va - Vb =
Va - VG = (the number of mutually distinct roots of Of(x) in K'*), where
V and V * denote the number of variations of signs of standard sequences
of f and Of, respectively.
(ii) If a1 < ... < an (a1 E K*) , then 0 is extended to an isomorphism (not
necessarily an order isomorphism) yr of K(a1 , ... , an) into K'* such that
t/ra1 < yra2 < < yran . Indeed, bl = a1 - a1_I E K* (i = 2, ... , n),
and we have a subfield K1 = K(a1 , ... , an , b2 , ... ,
Take an element c
b,1) .
such that K1 = K(c) (Theorem 2.5.2), and let f (x) be the minimal polyno-
mial for c over K. (i) above shows that there is a root c' of Of (x) in K'* .
Then we have an extension yr of 0 such that yrc = c'. Then yrbl E yrK(c)
and 0 < yrb2 = Oat - dal_ 1 .
Now, we shall define a mapping a of K* to K'*. For each a E K*, we
take the minimal polynomial fa(x) for a over K and let a1 < a2 < . . < an .
be the roots of fa (x) in K * . Let a i < a? < < an be the roots of cb fa (x)
in K'* (cf. (i) above), and we define a by Qa = aj , with j is such that
a=ai . Then
(1) as = Oa for a E K (because fa (x) = x - a).
(2) (i) above shows that a gives a one-to-one correspondence between K*

and K'*.
(3) a(a + b) = as + crb, a(ab) = (aa)(ab) for all a, b E K*.
Proof of (3). Consider minimal polynomials fa(x) , f'(x) , fa+b(x) , fab(x)
for a, b, a + b, ab, respectively, over K, and let c1 < c2 < < cn
be the mutually distinct roots of these in K* . Then by (ii) above 0 can
be extended to an isomorphism yr from K(c1 , ... , cn) into K'* so that
yrcl < yrc2 < . . < y/cn . Let d be any one of a, b, a + b, ab and let
clI < c.z < < c.s be the roots of fd(x) in K* . Then the roots of Ofd
in K'* are yrcl < . . < yrc1 , and by the definition of a, if d = clj , then
I s
ad = yrc,) Thus, we have ad = yrd, and (3) is proved, because yr is an

1
isomorphism of fields.
(4) That as > 0 for 0 < a c: K* follows from the properties that E K*
and as = (a/)2 > 0. This and (3) show that a is an order isomorphism
since for a > b, we have a - b > 0, and hence, a(a - b) = as - ab > 0, so
as > ab.
(1), (2), (3), and (4) above show that qS can be uniquely extended to an
order isomorphism a. In particular, if q is the identity mapping, then
a must be the identity mapping. Thus, we have completed the proof of
Theorem 5.2.3. Q.E.D.
Note that we obtain the following by Corollary 5.1.11.
THEOREM 5.2.8. If an ordered field K is a subfield of a real closed field L
as an ordered field, then the algebraic closure of K in L is the real closure of
K.
Using this result, we can prove.
THEOREM 5.2.9. Let K be an ordered field that is a subfield of a real closed
field L as an ordered field. Let K' be an algebraic extension of K, and let
H be the set of K-isomorphisms from K' into L.
(i) Assume that H is not empty. Then, fixing an element a of H, we
define an order on K' by a > b iff as > ub, with this order, K' becomes
an ordered field.
(ii) Conversely, assume that we can extend the order on K to K' so that
K' is an ordered field. Then the order is the one defined by some a E H as
in (i). Furthermore, this correspondence between orders on K' and a E H is
a one-to-one correspondence.
PROOF. (i) is clear.
(ii) Since K' contains K as ordered fields, the real closure K'* of K' is
also the real closure of K. By Theorem 5.2.3 and Theorem 5.2.8, we see
that the algebraic closure K* of K in L is order isomorphic to K'*. The
order on each of K* , K'* is unique (Theorem 5.1.7). Thus, with the unique
order K-isomorphism 0 from K'* to K*, its restriction 5IK' to K' is an
element of H, and the order, defined in (i) by this element of H, coincides

with the order on K' we are observing.
Assume now that two elements a, r in H give the same order on K'.
Then the mapping 0 = zQ-1 of QK' to zK' is an order isomorphism and
K* is the real closure of both aK' and TK'. Then, Theorem 5.2.3 shows
that 0 is the identity mapping, and therefore, a = r. Q.E.D.
3. Hilbert's 17th Problem
The following problem is called the Hilbert's 17th Problem, due to the fact
that Hilbert presented 23 problems at the International Congress of Mathe-
maticians held at Paris in 1900 and this is the 17th of these problems.
Let f(xl , ... , xn) be an element in Q(x, , ... , xn) the rational function
field in n variables over the rational number field. Assume f (xl , ... , xn )
satisfies the condition that if the elements a i , ... , an in Q do not annul
the denominator of f, then f (a, , ... , an) > 0. Does this imply that f
is the sum of square elements of Q(xi , ... xn) namely, are there elements
)
g1,..., gseQ(xl,...,xs) such that f =g2 + ... + gs2 ?1
In this section, we shall give proofs of a generalized form of this statement

and of related results due to E. Artin.
THEOREM 5.3.1. Let a be an element of a field K whose characteristic is
different from 2. Then we cannot make K an ordered field with a < 0 iff
there are elements bl , ... bs E K such that a = bi + + bs .
PROOF. The if part is clear and we consider only if part. Assume that a
is. not the sum of any square elements. Let F be the family of intermediate
fields L between K and its algebraic closure K such that a is not the
sum of any square elements in L. Then F is an inductive set and has a
maximal member L*. Lemma 5.1.4 shows that L* is formally real, and
therefore, we can define an order on L* so that L* is an ordered field. We
shall prove a < 0 in the ordered field L* , and it suffices to show that -a
is a square element in L*. Assume the contrary. Then L* (/) : L* .
By the maximality of L* , there are elements a , ... , an E L* (/) such

i
that a = ai + + an Write ar = ci + d; v,' a (c1, d, E L*). Then

.
a = > ci - a > d? + 2vr-a E cidi Consequently, a = > ci - a > d 2 ,

. and
all + > d 2) = > c2 Since L* is formally real, 1 + > d2 0. Then Lemma
.
5.1.4 shows that a is the sum of certain square elements, which is not the
case. Q.E.D.
LEMMA 5.3.2. Let xi , ... , xn be algebraically independent elements over
an ordered field K. Then we can extend the order on K to K(xi , ... , x,,)
so that K(x, , ... , xn) becomes an ordered field.
PROOF. Let K* be the real closure of K. Then K* (xi , ... , xn) is for-
mally real (see Exercise 5.1.6). Thus, K*(x, , ... , xn) can be made an or-
3. HILBERT'S 17TH PROBLEM 203
dered field. Since K* is the real closure of K, the order on K* is unique,

and its restriction on K is the original one. Therefore, the restriction of the
order on K* (xi , ... , xn) to K(x1 , ... , xn) is the required one. Q.E.D.
Here, we define a notion. Assume that K(x1 , ... , xn), as above, is an
ordered field. For (a) = (a, , ... , an) (ai E K) and gI (x) , ... , gs(x) E
K(x) = K(x1 , ... , xn), we say that {gI (x) , ... , gs(x)} has the same signs as
{g1 (a) , ... , gs(a)} if each gi(a) is well defined (i.e., (a) does not annul the
denominator of gi (x)) , and furthermore, gi (a) is positive, 0, or negative,
depending on whether gi (x) is positive, 0, or negative (in the order on
K(x)) , respectively.
LEMMA 5.3.3. Let K(x) be an ordered field as above, and let f (x , y) be
a monic polynomial in y over K(x). Suppose that f (x, y) has exactly r
mutually distinct roots in the real closure K(x)* of K(x). Then, there are
gI (x) , ... , gs(x) E K(x) with the following property:
If {g1 (x) , ... , gs(x)} has the same signs as {g1 (a) , ... , gs(a)} for an n-
tuple (a) = (a, , ... , an) of rational numbers ai, then f (a, y) is well defined
and has exactly r mutually distinct roots in the real closure K* of K.
Here, since K is of characteristic 0, we regard the rational number field
as a subfield of K.
PROOF. Write f(x, y) = ym + + . . + O,n(x) (q5i(x) E K(x))
01(x)y'n-' .
and set h(x) = Eoi(x) + 1 , where is chosen to be + if q (x) > 0

and to be - otherwise. Take the standard sequence of f (x , y) (as a
polynomial in y) fo = f (x , y), fl, ... , f , and choose gI, ... , gs E
K(x) as follows: All of the coefficients of fo, fI , ... , f (as polynomi-
als in y), all of fo(x, -h), ... , f(x, -h), f0(x, h), ... , f(x, h), all of
f (x , -h)/ f (x , -h) with j > i and f (x , -h) 54 0, and all of f (x , h)/
f (x , h) with j > i and fj(x, h) 54 0 (and arrange them in any order).
Then by Lemma 5.2.6 and Theorem 5.2.5, we see that r coincides with the
difference V h - Vh , where V h , Vh are the number of variations of signs of
fo, ... , f at -h, h, respectively. On the other hand, V h = V h(a), Vh =
Vh(a) , because, for each i, Oi(x) and O,(a) have the same sign and because
fi(x, -h)/ f (x , -h) and f (x , h)/ f (x , h) are among the gk . Therefore,
f (a , y) also has exactly r mutually distinct roots in K*. Q.E.D.
LEMMA 5.3.4. Let K(x) be as above. Assume that fI (x, Y), ... , f (x , y)
are (not necessarily distinct) monic polynomials in y with coefficients in
K(x) and that each f (x, y) has a root ai in the real closure K(x)* of
K(x) such that al < a2 < < at . Then by choosing suitable elements
gI (x) , ... , gs(x) o f K(x), * it f o l l o w s that i f {gl (x) , ... , gs(x)} and
{g1 (a) , ... , gs(a)} have the same signs for (a) = (aI , ... , an) with ratio-
nal numbers ai, then f (a , y) (i = 1, ... , t) are all well defined and each
f (a, y) has a root /3i in the real closure K* of K such that /31 < /32 < <
/3t .
PROOF. Let A be the subfield of K(x)* generated by ai (i = 1, ... , n)

and a1+1 - a, (i = 1, ... , n - 1) over K(x). Then A is a separable ex-
tension of finite degree over K(x) , and there is an element 0 such that
A = K(x)(0) (Theorem 2.5.1). Let g(x, y) = ym + + +
y1(x)y'n-'
ym(x) (y1 E K(x)) be the minimal polynomial for 0 over K(x). Then
there are bi(x, y), b'(x, y), cj(x, y) E K(x)[y] such that al = b; (x , 0),
aj+1 - aj = b(x, 0)cj(x, 0) = 1 Since f (x, al) = 0, we
.
see that f (x , bl(x , y)) = g(x, y) f'(x , y) with f'(x, y) E K(x)[y] (i =

1 , ... , t). From the equality aj+1 - cry = b (x, 0)2 , we obtain bj+1 (x, y) -
bj (x, y) - bj' (x, y) 2
= g(x, y)hj(x, y) with hj(x, y) E K(x)[y] (j =
1, ... t-1). Since b'(x, 0)cj(x, 0) = 1, we see that b'(x, y)cj(x, y)-l =
,
g(x, y)kj(x, y) with k(x, y) E K(x) [y] (j = 1, ... , t - 0.

Now, we take g1 , ... , gs as follows (with an arbitrary arrangement): All
of the coefficients of f (x , y), f'(x, y), g(x, y), hj(x, Y), k (X' Y),
bj(x, y), bj'(x, y), cj(x, y) (as polynomials in y) and the polynomials re-
quired by Lemma 5.3.3 when the lemma is applied to g(x, y).
Now, if (a) = (a1 , ... , an) is as in the lemma, then g(a, y) has a root,
say a*, in the real closure K* of K, because g(x, y) has a root 0 in K(x)*
(Lemma 5.3.3). f (a, bl(a, a*)) = f'(a, a*)g(a, a*) = 0, and therefore,
/1; = b;(a, a*) is a root of f (a, y). Now, fl 1 - j3j = b'(a, a*) 2 and
a*) E K* imply that fli+1 > fib . Q.E.D.
THEOREM 5.3.5. Let K(x) be an ordered field as above with the additional
assumption that K is order isomorphic to a subfield of the real number field.
((x) = (x1 , ... , xn) is a set of algebraically independent elements over K, as
before.) If f1(x), ... , fr(x) E K(x) are given, then we can choose rational
numbers (a) _ (a1 , ... , an) so that {f1 (x) , ... , fr(x)} has the same signs
as {fl(a), ... , fr(a)}.
PROOF. We use an induction argument on n. The assertion is clearly
true if n = 0. Now, assuming that the assertion is true in the case of n
variables, we shall prove the assertion true in the case of n + 1 variables.
(x) will denote (x1 , ... , xn) and xn+1 will be denoted by y. Thus, the
K(x) and f(x) in our assertion should be denoted by K(x, y), f (x , y),
respectively.
If f K(x)[y], then, by multiplying by the square of the denominator,
we may assume that f c K(x)[y]. If f factors as a polynomial in y over
K(x) into c, II; p;j (ci E K(x), then each pij is monic). Then we can
consider all of c, , p instead of fl , ... , fr. Thus, we may assume that
each f is either an element of K(x) or an irreducible monic polynomial in
y over K(x) .
Let a1 < a2 < < as be the complete set of roots of all of f in
K(x)*, the real closure of K(x). Then, allowing the duplication and the
3. HILBERT'S 17TH PROBLEM 205
change of the subscripts of f , we may assume that (i) if i < s then ai

is a root of f. , and (ii) f c K(x) if i > s (because mutually distinct
irreducible polynomials f , fj have no common root). We take g1 , ... , gr
as follows: fs+1 , ... , fr, all of the gj (x) E K(x) required by Lemma 5.3.3
when the lemma is applied to each f (i = 1 , ... , s), all of the gj required
by Lemma 5.3.4 for f1 , ... , fs and for a < < as, and discriminants of
1
fl, ... , fs
By our inductive hypothesis, there is an n-tuple (a) = (a1 , ... , an) of
rational numbers ai such that {g1(x) , ... , gt(x)} has the same signs as
{g1 (a), ... , g,(a)}. Since fs+1 , ... , fr are among these gi , { f +1(x) , ... ,
fr(x)} has the same signs as { f +1(a) , ... , f,.(a)}.
Now, consider f (i < s). Let a < <a. be the roots of f in
.
K(x)*. That is, fj = f iff j = ik for some k. By Lemma 5.3.4, each

f (a, y) has roots / 3i < < .
. . By applying Lemma 5.3.3 to f , we see
<
< / 3,
/3i171(1)
.
that the number of mutually distinct roots of f (a , y) in K* is also m (i) ,

which shows that these /ii are all the roots of f.(a, y) in K* .
f(x,Y) = fl(Y-aii)flgik(x,
Y)
j k
(qik are monic, irreducible over K(x)*).

On the other hand, since the discriminant di (x) of f is among these gj
and since di (x) 0, it follows that f (a , y) has no multiple root. Hence,
by the fact that these /iii are exactly the roots of f (a , y) in K* , we see
that, in
f(a,Y)=fJ(Y-/ii T7
llq,k(a,Y),
j k
where the gik(a, y) are irreducible over K* .
Therefore, each qik is of the form (y + c(x))2 + d(x) with d(x) > 0 in
K(x)*. Consequently, qik is a positive element in K(x, y). Similarly, by
the irreducibility of qik (a , y) over K* , we see that qik (a , an+1) is positive
for any an+1 E K. Thus, in observing signs of f (x , y), f.(a, y), we can
disregard the factors qik Hence, the signs are determined by the order
.
among y and these ai . (Note that K(x) * can be embedded in the real
closure K(x, y)* of K(x, y) by Theorem 5.2.8.) Just for convenience, we
set ao = flo = -oo , as+1 = fls+I = oo , and choose i such that ai < y < ai+I .
Then a rational number an+1 such that /3i < an+I < Qi+I is the required last
member. Q.E.D.
The next theorem, which is due to Artin, gives an affirmative answer to a
generalization of the Hilbert's 17th Problem. It is clear that this theorem is
a generalization, because the rational number field, as an ordered field, has
unique order (cf. Exercise 5.1.3).
THEOREM 5.3.6. Assume that a formally real field K has the property that
whenever K becomes an ordered field under some order, then K is order
isomorphic to a certain subfeld of the real number field. Then, for an element
f(x) of the purely transcendental extension K(x) = K(x1 , ... , xn) of K,
it follows that there are elements gi (x) E K(x) with f (X) _ E i gi (x)2 iff
f (a) > 0 for any order on K that makes K an ordered field and for any
(a) = (a, , ... , an) with rational numbers al whenever f (a) is well defined.
PROOF. The only if part is obvious. Assume that f (x) is not the sum of
any square elements. Then Theorem 5.3.1 shows that there is an order on
K(x) which makes K(x) an ordered field with f (x) < 0. Then there are
rational numbers ai such that f (a) < 0 by Theorem 5.3.5. Q.E.D.
COROLLARY 5.3.7. Let K be a subfield of the real number field R such
that K has a unique order as an ordered field. Let K(x) be as above, and
consider f (x) E K(x). Then f (x) is the sum of some square elements iff
f(aI , ... , an) > 0 for any rational numbers ai whenever f(a1 , ... , an) is
well defined.
Note that this corollary can be applied to the rational number field or real
closed subfields of R (cf. Exercise 5.1.3 and Theorem 5.1.7).
4. A valuation corresponding to an order
THEOREM 5.4.1. Let S be a subfield of an ordered field K. Then V =
{x E KI for some a c S, -a < x < a} forms a valuation ring of K with
maximal ideal M = {x E KI if a E S and a > 0, then -a < x < al.
Furthermore, letting yr be the natural homomorphism from V to VIM, we
see that VV = VIM is an ordered field in a natural way such that yra > 0
iffa>0 and a V M.
This V is called the valuation ring of the ordered field K with respect to
S
PROOF. Assume that x E K, x V. If x > 0, then x > a (for all
I
a (=- S). Hence, x- < b for any positive element b of S. Consequently,
x-I E M in this case. The case x < 0 can be proved similarly. Thus, V is
a valuation ring of K and M is the maximal ideal of V.
Set P' _ { yrx Ix > 0, x V MI. Obviously, 0 P'. If a > 0, b > 0,
and a, b M, then there is an element c E S, with c > 0, a > c, b > c,
so a + b > c and ab > c2 . This fact shows that if a', b' E P' , then a' + b' ,
a'b' E P'. Therefore, V/M is an ordered field with the set P' of positive
elements by Lemma 5.1.6. Q.E.D.
The main aim of this section is to prove a certain converse of this theorem.
However, for simplicity, we consider the case where S is the rational number
field Q. Then the result can be stated as follows. (For the general case, see
Exercise 5.4.2.)
THEOREM 5.4.2. Let V be a valuation ring of a field K. Assume that there
is a homomorphism yr from V to the real number field R whose kernel is
4. A VALUATION CORRESPONDING TO AN ORDER 207
the maximal ideal M. Then we can define an order on K to make K an

ordered field with the following properties: V is the valuation ring of K with
respect to Q, and yr gives an order isomorphism between the ordered field
VIM, as above, and the subfield yr V of R.
Let G be the value group of the additive valuation v defined by V, and
set H = {2glg E G} . Then #(G/H) denotes the number of orders on K. In
particular,
(i) such an order is unique iff' #(G/H) = 1, i.e., G = H.
(ii) If #(G/H) = n < oo, then there are exactly n ways to define such an
order on K.
For the proof, we need some facts about maximally complete valuation
rings. A valuation ring V of a field L is said to be maximally complete if,
for any prolongation V' of V in any proper extension L' of L, either (i)
the residue class field extension is proper, or (ii) the value group of V' is a
proper extension of that of V. In other words, V is maximally complete if
there p no proper valuation ring extension V" of V such that every residue
class of V" modulo the maximal ideal is represented by an element of V
and such that every principal ideal of V" is generated by an element of V.
An easy example of a maximally complete valuation ring is a complete
discrete valuation ring. (Indeed, if V is a complete discrete valuation ring
and if V" is as above, then, given a E V", there is an element ao E V with
a - ao E p V" (p being a generator of the maximal ideal of V) . By applying
the same result to (a - ao)/p , we see that there is an element al E V such
that a - ao - alp E p2V" , and so on. Thus there are elements ai c V with
a - En 0 a`pi E pn+l
V" . Hence a = 1imn,. En 0 aip' E V, since V is
complete. Consequently, V" C V and V" = V.)
On the other hand, in the case of a nondiscrete non-Archimedean val-
uation, completeness does not imply maximal completeness. Furthermore,
though we did not define completeness for valuation rings of Krull dimension
> 2, the notion of maximal completeness can be defined in such cases.
Let V be a valuation ring with maximal ideal M. A valuation ring V
containing V is called a maximal completion of V if V * satisfies the fol-
lowing three conditions:
(i) V*/MV* = V/M (i.e., MV* is the maximal ideal of V* and every
residue class of V * /MV * is represented by an element of V).
(ii) Every principal ideal of V* is generated by an element of V (i.e.,
value groups coincide with each other).
(iii) V* is maximally complete.
LEMMA 5.4.3. Let V be the valuation ring of an additive valuation v of a
field K. Let M, G be the maximal ideal of V and the value group of v,
KG
respectively, and set K = VIM. Then #(K) < #(KG), where denotes
the set of mappings of G to K.
PROOF. Let w be a mapping of G to K such that v(wg) = g and let

K* be a set of representatives of nonzero elements of 1. Let G* be the
family of well-ordered subsets of G+ = {x E Gjx > 01. When x E S E G*,
we denote by S(x) the subset {y c Sly < x} , which is called the segment
of S by x.
For f E KG, Suppf denotes {x E Gl fx 01. We consider F =
If E KG I Supp f E G* } When we write fs , gs , we mean that Supp fs =
.
Supp gs = S S. When an fs and a segment S' = S(x) of S are given, we

define the restriction fs, of fs to S' by (i) fs, (y) = fs(y) , for y E S', and
(ii) fs, (y) = 0, for y S' .
Now, we define inductively a correspondence between a certain subset of
F and elements of V as follows:
(1) For the zero mapping 0, Supp 0 is the empty set, hence, 0 E F. Let
0 E V correspond to the zero mapping.
(2) For an fs E F, we consider all segments S' of S and the restrictions
fs, of fs to S'. Then (i) a (E V) corresponds to fs only if, for each
segment S' = S(x), an element b corresponds to fs, and v (a - b) > x.
Namely, (ii) even though every fs, has its correspondent b in V, if there
is no a (common to all segments S') such that v(a - b) > x , or, if there is
some fs, which has no correspondent, then no element of V corresponds
to fs.
For the proof of our lemma, it suffices to prove the existence of a corre-
spondence satisfying (1) and (2) above and (3) and (4) below.
(3) If an element of V corresponds to an fs, then a unique element
corresponds to fs .
(4) For each a c V, there is an fs corresponding to a.
Consider an element fs c F whose restriction fs, to each segment S' of
S has its correspondent in V. Assume first that S has the largest element
in. Then if b is the element corresponding to fs(,n , we take the represen-
tative k (E K*) of fs (m) and let b + k (w m) be the correspondent of fs .
Assume next that S has no largest element. Let bX be the element corre-
sponding to the restriction fs(x) , for each x E S. If {a E VIv(a - bX) > x,
for all x E S} is not empty, then we take one element a of this set, and let
a correspond to fs ; otherwise, no element corresponds to fs .
By this choice of correspondence, (3) above holds, and it remains to show
the validity of (4). For this purpose, we prove the following.
If bs, CT correspond to A' 9T, respectively, and if
v(bs - CT) = r, then S' _ Is E Sis < r} coincides with
T'={tETlt<r} and fs'=gs,.
PROOF. Set U = {u c S n Tlfs(x) = gT(x) for x < u} . If S' = T' C U,
then there is nothing to prove. So, we assume the contrary. Let in be
4. A VALUATION CORRESPONDING TO AN ORDER 209
the least element in (S' - (U n S')) U (T' - (U n T')). By the symmetry

between S' and T', we can assume that m E S'. Then S(m) CU and
there is an element m' E T such that S(m) = T (m') , because T Z U.
Let bu be the correspondent of fs(,n) . Then, for S" = S(m) U {m} , the
element Nil = bu + k(wm) (where k((=- K*) is a representative of fs(m))
corresponds to fs . Similarly, for T" = T(m') U {m'} , CT" = bu + k'(wm')
(where k' (E K*) is a representative of 9T (M')) corresponds to gT" Since
.
v(bs - CT) = r > m, we have v(k(wm) - k'(wm')) > m, and therefore,

k = k' , m=m ' . This contradicts our choice of m. Q.E.D.
Now, we shall prove (4). Let c be an arbitrary element of V, and set
F = { fs E Fibs corresponds to fs and v(C - bs) > s (for all s c S)}.
By what we have proved above, fs , gT E F implies that either fs = gT or
one of S, T, say S, is a segment of T and fs is the restriction of gT .
This means that the union U of all S such that fs E F, is a nonempty
well-ordered set. Furthermore, there is a well-defined fu E F such that each
fs c F. is a restriction of fu , and there is an element bu corresponding
to fu If bU = c, then c corresponds to fu . Assume that bo # c. Set
.
g = v (C - bu) , k = (C - bu) /w g, and let k' (E K*) be the representative of

k. Since g is larger than any element of U, we can take W = U U {g} and
define fw by the condition that its restriction to U is fu and ff,(g) = k'.
Then fw E F , a contradiction. Thus bo = c. Q.E.D.
THEOREM 5.4.4. For any valuation ring V of a field K, there is a maximal
completion V* of V.
PROOF. Let G and K be as above, and take an algebraically closed field
L containing K with #(L) > #(KG) . Consider the family F of valuation
rings of intermediate fields between K and L whose residue class fields and
value groups coincide with those of V. Clearly, F is an inductive set, and
has a maximal member V*. Then #(V*) < #(L) by the lemma above.
Consequently, L has infinite transcendence degree over V * . Therefore, if
V* is not maximally complete, we have a contradiction to the maximality of
V* in F. Q.E.D.
COROLLARY 5.4.5. Every maximally complete valuation ring is a Hensel
ring.
PROOF. This is clear by the definition of maximal completeness and The-
orem 4.11.10. Q.E.D.
LEMMA 5.4.6. Let V be a valuation ring of a field K and let M be the
maximal ideal of V. If VIM is formally real, then K is also formally real.
PROOF. If Xi +X2 +...+X,,2 = 0 has a nontrivial solution (a1 , ... , aj in
K, then we have a nontrivial solution of the same equation in V/M through
(a, /ai , ... , an/aj) with aj such that a. E aj V for all i. Q.E.D.
THEOREM 5.4.7. Let V be the valuation ring of an additive valuation v

of a field K. Let M, G be the maximal ideal of V and the value group
of v, respectively. If V is maximally complete, if V/M is real closed and if
furthermore, G is divisible, then K is real closed.
Here, an additive group G is divisible if for any element a of G and for
any natural number n, there is an element b of G such that nb = a.
PROOF. K is formally real by Lemma 5.4.6. For a proper algebraic ex-
tension K" of K , let V" be the integral closure of V in K". Then V" is
a valuation ring by Lemma 5.4.5 and Theorem 4.9.3. Since G is divisible,
the value group of V" coincides with G. Since V is maximally complete,
we have V"/MV" 34 VIM. Hence V"/MV" = (V/M)() , because
VIM is real closed (Theorem 5.1.9). Thus X2 + 1 has a root in V"/MV".
Since V is a Hensel ring, V" is also a Hensel ring (Exercise 4.11.2) and
E V" C K". Consequently, K is real closed. Q.E.D.
PROOF OF THEOREM 5.4.2. (1) The existence of an order. To prove this, we
can replace K with an extension field and V with a prolongation. Therefore,
we may assume that (i) V/M is real closed (because there is a prolongation
of V with required extension of residue class field, without changing the
value group) and (ii) the value group G is divisible (because, for instance, if
a E G and if, for any prime number p, there is no b E G such that pb = a,
then, taking c E K such that vc = a, we adjoin a root of X" - c, then we
have an extension which enlarges G only), and furthermore, by considering
the maximal completion, (iii) V is maximally complete. Then, K is real
closed by Theorem 5.4.7. Consequently, K has a unique order (Theorem
5.1.7).
(2) The relationship between the order and V under the circumstances in
(1). The set P of positive elements in K coincides with {x2Ix E K} by
Theorem 5.1.5. (i) Assume that a c V and yra > 0. If a < 0, then a = -b 2
for some b). Since va = 0, we have vb = 0 and yra = -(yrb)2 < 0, a
contradiction. Thus a > 0 in this case. (ii) Consider the case where yra = 0.
Then a E M. If there is a positive rational number r such that a > r, then
a - r = b2 for some b and -r = -yrr = (yrb)2 , a contradiction. Similarly,
there is no positive rational number r such that a < -r. Thus, -r < a < r
for any positive rational number r. Now, let V' be the valuation ring of
the ordered field K with respect to the rational number field Q. Then, (i)
and (ii) above show that V C V' and also that M C_ M', which imply that
V'=VM,nv=VM=V and M=M'.
(3) The number of orders. Let K, V, G be as in the beginning. To each
element g E G, we associate an element bg E K such that v (bg) = g , and
if g E H, then we assume, further, that bg is a square element in K. If the
order on K satisfies the conditions in the theorem, then a > 0 for a E V
and yra > 0. By our choice, bh > 0 for h E H. Consequently, if x E K
and vx = h E H, then it follows that x > 0 if yr(x/bh) > 0 In particular,
.
if G = H, then the order is unique.

EXERCISES 211
Now, we consider the case where G 0 H. G/H is regarded as a module

over the prime field it = Z/2Z of characteristic 2. Let {gj j E T} be a
linearly independent base of the module G/H. For each gj, we choose a
representative gJ- E G and an element ai E K so that v (aj) = gj . In the
process (ii) in (1) above, we can adjoin all of aj, and we have an order
such that ai > 0. On the other hand, we may replace all or some of these
ai by -ai . Then these ai become negative elements. Therefore, by (2) and
(3) above, the order on K is determined by which of these ai are negative.
This proves the last part of the assertion. Q.E.D.
Exercises
1
1. Assume that a ring R is a linearly ordered set satisfying the following

conditions (where, a, b, c, d denote arbitrary elements of R) :
(i) a+c>b+d for a>b, c>d;
(ii) -a < -b for a > b ;
(iii) ab>0 for a>0, b>0.
Prove that R is an integral domain and the order on R can be uniquely
extended to an order on the field of fractions K of R so that K becomes
an ordered field.
2. Prove that in R above, ac > bd for a > b > 0, c > d > 0.
3. Prove that the rational number field has a unique order as an ordered
field.
4. An ordered field K is called an Archimedean ordered field if a, b E K
with a > 0 implies that na > b for some natural number n.
Prove that K is an Archimedean ordered field if there is an order
isomorphism from K into the real number field.
5. Let L be a subfield of an ordered field K and set P = {x E KI -a <
x < a for any positive a E L} and R = {x E KIxP C P1. Prove that
(i) R is a valuation ring of K, L C R, and P is the maximal ideal
of R.
(ii) R/P is an ordered field by defining an order by a + P > 0 if a > 0
and a 0 P. Then, given a , b E R/P with a < b , there is an
element c E L such that a<c<b.
6. Prove that if K is a formally real field, then any purely transcendental
extension of K is also formally real.
7. Prove that if 92 is an algebraically closed field of characteristic 0, then
there is a real closed subfield K of S2 such that S2 = K(im).
2
1. Assume that K is an ordered field and 0 < a E K, Va- K. Prove

that K(im) is an ordered field, extending the order on K, in exactly two
ways depending on which one of / is defined to be positive.

2. Give another proof of Theorem 5.2.3 under additional assumption that
K* is an Archimedean ordered field, using the following ideas:
(i) Prove that if 0 < E E K, a E K* then there are elements b, c c K
such that b < a < c and c - b < E .
(ii) Show that there is a mapping a : K* -+ K'* as follows. Choose

b, c so that the minimal polynomial f (x) for a has the unique
root a in the interval [b, c] Then as is defined to be the root
.
of Of(x) in the interval [0b, qc] (whose uniqueness is seen by

Theorem 5.2.5).
(iii) Prove that a is an order isomorphism by using that c - b in (ii)
can be sufficiently small (by (i)).
3. Prove that if K is an Archimedean ordered field, then the real closure is
also an Archimedean ordered field, according to the following two ways.
(i) Use Lemma 5.2.6.
(ii) Use Exercise 5.1.4, Corollary 5.1.11, and the uniqueness of real clo-
sures (Theorem 5.2.3).
4. When an ordered field K is a subfield of a formally real field L, then it
is not necessarily true that the order on K is extended to L to make L
an ordered field. Give one example of such a pair K and L.
Show also that the order on K can be extended to an order on L that
makes L an ordered field if it can never happen that 0 < al c K ,
O xi EL, and E'_laixi =0 (s> 1).

5. Let K' be an algebraic extension of degree n over an ordered field K.
Prove that the number of ways to make K' an ordered field extending
the order on K is at most n .
3
1. Let a be an element of an algebraic extension K of finite degree over

the rational number field Q. Prove that a is the sum of certain square
elements in K if for any isomorphism T from K into the real number
field, T a > 0.
2. Generalize Theorem 5.3.5 as follows: Omit the assumption that K is
order isomorphic with a subfield of the real number field. But, take a
subset S of the real closure K* of K satisfying the condition (*) below,
and change the condition on (a1 , ... , an) to ai c S.
(*) If a < f are elements of K* , then a < q < fl for some q E S .
3. Let a be an element of a field L containing an ordered field K. Prove
that a= > ci g2 with 0 < ci E K and gi E L if a is positive, whenever
L becomes an ordered field, extending the order on K. [Hint. For
the only if part, as in the proof of Theorem 5.3.1, consider the family
of algebraic extensions M of L such that a cannot be expressed as
a = > cig2 with 0 < ci E K, gi c M, and show that a maximal
EXERCISES 213
member is the real closure of K.]

4. Let x1 , ... , xn be algebraically independent elements over an ordered
field K and let S be as in Exercise 5.3.2. Prove that an element f (x)
of K(x1 , ... , xn) can be written in the form f (X) = > cig! (x)2 with
0 < ct E K and gl (x) E K(x1 , ... , xn) iff if a; E S and f (a) is well
defined, then f (a) > 0.
4
1. Prove that V = {x (=- KjxM C_ M} in Theorem 5.4.1.

2. Let L be a subfield with induced order of an ordered field L' and assume
that, for any x E L', there is an element y E L such that x < y .
Furthermore, let V be a valuation ring of a field K such that V contains

L and assume that there is given an L-homomorphism yr from V to
L' such that the kernel of yr is the maximal ideal M of V. Prove that
K is an ordered field in such a way that V is the valuation ring with
respect to L. Show also that yr induces an order isomorphism from
VIM onto yrV V.
CHAPTER VI
Galois Theory of Algebraic Extensions

of Infinite Degree
1. Topology on a Galois group

As in Chapter II, when L is a normal algebraic extension of a field K,
the group AutK L of K-automorphisms of L is called the Galois group of
L over K and is denoted by G(L/K).
LEMMA 6.1.1. Under the above circumstances, if an intermediate field L1
is also a normal extension of K, then G(L/L1) is a normal subgroup of
G(L/K) and G(L/K)/G(L/L1) G(L1 /K) In particular, if L1 is of finite
.
degree over K, then [G(L/K) : G(L/L1)] = [L1 : K]5 < [L1 : K].
PROOF. Obviously G(L/L1) C G(L/K) by definition. Let 0 be the map-
ping of G(L/K) to G(L1 /K) that associates a ((=- G(L/K)) to its restric-
tion on L1 Ou is well defined, because L1 is normal over K. Since
.
q(QZ) = (ga)(gr) (U, r E G(L/K)), 0 is a homomorphism. O u = 1 if

a E G(L/L1) ; hence, the kernel of 0 coincides with G(L/L1) , and there-
fore, G(L/L1) is a normal subgroup of G(L/K). On the other hand, if
Q E G(L1 /K) , then Q can be extended to a (E G(L/K)), which satisfies
o u = Q . Consequently, the image of 0 coincides with G(L1 1K). Q.E.D.
Let S21 be the family of normal subgroups H of G = G(L/K) such
that H = G(L1 /K) with normal intermediate fields L1 which are finite
over K. We shall define a topology on G by taking S21 as a fundamental
neighborhoods of the identity. This is justified by (i), (ii) of the following
lemma.
LEMMA 6.1.2. (i) If 10 a E G, then or 0 N for some N E K21
(ii) If N1, N2 E S21 , then N1 n N2 E S21
.
(iii) If Kl (i runs through certain set I) are intermediate fields between

K and L, then n, G(L/Kt) = G(L/K*) , where K* is the field generated by
all of these K.
(iv) Assume that a subset S of S21 satisfies the conditions that if N,
N' E S, then N" C N n N' for some N" E S, and that an element aj of G
is associated to each N E S so that if N c Nk , then a, N c ak Nk Then .
there is a u E G such that ajN = QN for all N E S.

215
216 VI. GALOIS THEORY
PROOF. (i) There is an a E L with ca a. Take a finite normal extension

L1 containing a. Then a 0 N = G(L/L1) .
(iii) G(L/K*) c niEl G(L/K,) , because G(L/K*) c G(L/K,). Con-

versely, if a c n; G(L/K,) , then ca = a for every a E K, (for all i) .
Thus, ca = a for all a E K* , and ni G(L/Kl) = G(L/K*) .

(ii) If L1 , L2 are finite normal extensions of K, then the field generated
by L1 , L2 is a finite normal extension of K. (ii) follows from this and (iii)
proved above.
(iv) For each N E S, let Lj be the Galois extension such that N =
G(L/Lj) (namely, Lj is the separable closure of K in an intermediate field
L' such that N = G(L/L')). Let L* be the field generated by all of these
Li Each a E L* is in a field generated by a finite number of these Li .
.
Therefore, there is an Lk containing a, by the assumption on S. Now we

define a mapping a of L* to L* by as = Qka (k is chosen so that a e Lk) ;
indeed, as is independent of the particular choice of ak , because if a is
also in another Len , then there is L, Lk (L in) and Ql Nt g ak Nk l Qm N ;
hence, Qka = ana = Qta Obviously, a E G(L*/K) , and the prolongation
.
of this a to an automorphism of L is the required element a. Q.E.D.

Let H be a closed subgroup (i.e., a subgroup which is a closed set under
the topology) of the topological group G. We define a topology on the set
G/H (or, H\G) of residue classes by taking a subbase of open sets as follows.
The family of images of open sets of G by the canonical mapping a -* aH
(or Ha) of G to G/H (or, H\G). These new topological spaces are called
factor spaces of G by H.
THEOREM 6.1.3. A topological group G is a compact group if there is a
family S of closed subgroups satisfying the following five conditions:
(i) If H E S, then the factor space G/H is compact.
(ii) Given H, H' E S there is an H" E S such that H" CH n H'.
(iii) If F is a closed set in G and He- S, then FH/H = { f HI f E F} is
a closed set in G/H.
(iv) If V is a neighborhood of 1 in G, then there is an H E S with
He V.
(v) If an element ai is associated to each H E S so that Hj c Hk
implies ujH C QkHk, then there is a a E G such that ci Hj = UHF
for all H E S S.
PROOF. It suffices to show that if a family C of closed subsets has the

finite intersection property, then nFEC F is not empty. The set of families
of closed sets with finite intersection property and containing C forms an
inductive set; hence, it has a maximal member. Thus, replacing C with its
maximal member, we may assume that C has the following properties:
(1) If F, F' E C, then FflF'EC.
2. THE FUNDAMENTAL THEOREM OF GALOIS 217
(2) If F' is a closed set not belonging C, then there is an F E C such

that F' n F is empty.
Let H be an arbitrary member of S. Then {FH /H I F E C} is a
family of closed sets in G/H with the finite intersection property. By the
compactness of G/Hj, there is JJHj E 1 IFEC FH/H . Since Q3H n F is
not empty (for all F E C), ujH E C by property (2). Note that this a H
is uniquely determined by H . Indeed, zH E C implies rH n YJH is not
empty, and hence, zH = JJH Thus, by property (v), there is a E G with
.
a1Hj = QHj (for all H E C). Assume for a moment that a F for some
F E C. Then there is a neighborhood U of 1 such that a U n F is empty.
Take H. E S such that H c U (property (iv)). Then, aH c uU and
QHj n F is empty, a contradiction. Thus a E nFEC F. Q.E.D.
THEOREM 6.1.4. Let L be a normal algebraic extension of a field K. Then
the Galois group G = G(L/K) is a compact group. If K' is an intermedi-
ate field between K and L, then G(L/K') is a closed subgroup of G. If
furthermore, [K' : K] is finite, then G(L/K') is an open set of G.
PROOF. Consider K21 in Lemma 6.1.2. If N E S21 , then G/N is a finite
group; hence, G/N is compact. Therefore, G is compact by Lemma 6.1.2
and Theorem 6.1.3. As for G(L/K') with a finite extension K', there is
a finite normal extension L' of K containing K'. Then N = G(L/L') E
K21 , and N is an open set. G(L/K') is the union of subsets of the form
aN; hence, G(L/K') is an open set. Consequently, G(L/K') is also closed
by Theorem 4.4.5. As for G(L/K') in the general case, K' is generated
by a certain number of finite extensions Kj of K, and G(L/K') is the
intersection of G(L/K) (Lemma 6.1.2). Since each G(L/Kj) is a closed
set, we see that G(L/K') is a closed set. Q.E.D.
THEOREM 6.1.5. Under the above circumstances, the topology on G(L/K')
coincides with induced topology (i.e., the topology as a subspace of G(L/K)).
The proof is clear by the relationship between the family of fields of the
form K'(L') with finite normal extensions L' of K and the family of finite
normal extensions of K'.
2. The fundamental theorem of Galois
Let L be a Galois extension, namely, a separable normal algebraic exten-
sion of a field K. In view of the topology defined in 1, we can state the
fundamental theorem of Galois as follows.
THEOREM 6.2.1. There is a one-to-one correspondence between the family
F of closed subgroups of G and the family 0 of intermediate fields between
K and L as follows.
If H E I' corresponds to M E A, then
(i) H = G(L/M),
(ii) M = {x E Lax = x (for all a E H)}.

In this case,
(iii) H is a normal subgroup of G iff M is a Galois extension of K, and
(iv) if each Hj E IF corresponds to Mj E A, then nj Hj corresponds to
the field generated by these Mj and ni M corresponds to the closure of the
group generated by these Hj .
PROOF. Set M(H) = {x c LI ax = x for all a E H} for each H E IF.
This is an intermediate field between K and L. Conversely, for M E A,
H = G(L/M) belongs to IF by Theorem 6.1.4. Obviously, M C M(H).
If M : M(H), there is a E M(H) that is not in M. Then since a is
separable over M, there is an element a E H = G(L/M) such that as a,
which contradicts a c M(H). Thus, M = M(H), and this is the required
one-to-one correspondence (satisfying (i), (ii)).
If H = G(L/M), a E G, then G(L/QM) = QHu-' , and therefore, (iii)
is proved.
(iv) The first half is clear by Lemma 6.1.2, (iii), and we consider the
latter half. Let H* be the subgroup corresponding to n M , namely, H* _
G(L/ n Mi) Obviously, H* contains the subgroup H generated by these
.
H Since H* is a closed set, H* contains the closure H of H. H is

.
a subgroup as we shall see in the lemma below; hence, 71 E IF. Let M

be the field corresponding to 710. Then M C M , and M c n M M. On
the other hand, n Mi c M , because H C H* . Thus, M = n Mi , and
therefore, 71 = H* . So it remains to prove the following.
LEMMA 6.2.2. If H is a subgroup of a topological group G, then the closure
H of H is also a subgroup of G.
PROOF. We have only to show that a, b E H implies ab-' E 77. An
arbitrary neighborhood U(ab-1) of ab-' contains UaUb-' with a suit-
able neighborhood U of 1 (continuity of multiplication). U-' is also a
neighborhood of 1; hence, b U-' is a neighborhood of b. Therefore, there
exists a b' E H n b U-' . Similarly, there is an element a' e H n Ua. Thus
a'b'-' E H n U(ab-'), which shows that ab-1 E H . Q.E.D.
We add here a related result on normal algebraic extensions.
THEOREM 6.2.3. Let L be a normal algebraic extension of a field K with
Galois group G = G(L/K). For an intermediate field M between K and L,
we set H={aeGlax=x (for all xEM)} and M'={aELIaa=a for
all a E H1. Then H = G(L/M) = G(L/M'), M' is purely inseparable over
M, and L is separable over M'.
PROOF. Obviously M C M', H = G(L/M) = G(L/M'). If a E M'
is not purely inseparable over M, then by taking a normal intermediate
3. SPLITTING, INERTIA, AND RAMIFICATION FIELDS 219
field M", which is finite over M and containing a, we see that there is an
element a' E G(M"/M) such that a'a a. A prolongation a of a' to an
automorphisms of L belongs to H, contradicting a E M'. Q.E.D.
3. Splitting fields, inertia fields, and ramification fields
Let R be an integrally closed integral domain with field of fractions K,
and let L be a normal algebraic extension of K. If we take the integral
closure RL of R in L, then AutR RL is identified with G = G(L/K) and
called the Galois group of RL over R (cf. Theorem 3.7.11). Futhermore,
the splitting group HP of a prime ideal P of RL is {a E GI QP = P} (cf.
Remark after Theorem 4.11.6). Here we shall define some others. For an
ideal I of RL , VI = {Q E GIQx - x E I for all x E RL} is called the
ramification group of I. Ramification groups of primary ideals belonging to
a prime ideal P are called ramification groups of P. If I is a prime ideal,
then VI is called the inertia group of I.
THEOREM 6.3.1. The subgroups HP , VI above are closed subgroups of the
Galois group G.
PROOF. Take an arbitrary a not in HP (a E G). Then uP 0 P, and
ab P for some b c: P. Then N = G(L/K(b)) is a neighborhood of
1 by Theorem 6.1.4; hence, QN is a neighborhood of a. T E N implies
arb = ab 0 P. Thus, uN and HP have no common element, which shows
that a is not in the closure of HP and HP is a closed set.
Next we consider VI. Take a VI (a c G). Then for some element
c E RL , ac - c I. Consider N = G(L/K(c)). TEN implies QTC - C =
cc - c 0 1. Thus, aN n VI is empty, and we see that VI is a closed
set. Q.E.D.
We defined the splitting field and the splitting ring as field and ring cor-
responding to the splitting group HP In the same manner, we define the
.
inertia field and the inertia ring as the field and the ring ({x E Lax = x
(for all a c VP)} , {x (=- RLIa, = x (for all a E VP)}) corresponding to
the inertia group VP . Similarly, we define the ramification field and the
ramification ring of an ideal I by using the ramification group of I.
THEOREM 6.3.2. Under the above circumstances, let L and K be the fields
of fractions of RLIP and R/(P n R), respectively. Then, the inertia group VP
is a normal subgroup of the splitting group HP , and L is a normal extension
of K with Galois group G(L/K) naturally identified with HPIVP .
PROOF. Since Q E HP if a c P implies as E P, we see that VP C HP.

If a E HP , T E VP , x E RL , then setting y = a-1 x , we have QTQ-1 x - x =
QTy - ay = Q(Ty - y) E uP = P, which shows that QTQ- I E V, and V, is a
normal subgroup of HP . Since any element a of HP is an automorphism
of RL such that aP = P, for any element a = a + P (a E RL) we can 5
define Qa to be ca + P, and we obtain an automorphism Q = Ou of RL/P

over R/(P n R). Thus, we have a homomorphism 0 from HP to Autk L .
Since Ou = 1 if as - a E P (for all a E RL) , the kernel of 0 coincides
with VP . (Note that this gives another proof that VP is a normal subgroup
of He.)
L is generated by elements of RL/P over K. For an element a = a + P
(a c: RL) , we take the minimal polynomial f (x) for a over R. Then
f (x) = (f mod P) is a monic polynomial over K and has a as a root.
Since f (x) is the product of linear factors over RL f (x) decomposes also
.
to the product of linear factors over L, and therefore, L is normal over K.

Thus, HP/VP c Aut7 L = G(L/K). In showing the equality, we can replace
K by the splitting field of P (cf. Theorem 4.11.8). Hence we can assume
that HP = G. Let RI be the inertia ring of P and we adapt our observation
on f , f above over RI . Take the minimal polynomial g(x) for a E RL
over RI , and set g = (g mod P). Then the conjugates a' of a are the
residue classes of conjugates a' of a. Since we are observing over RI , there
is an element a E VP C HP , as = a' and Oua = a' It follows that a = a',
.
because Ou = 1 This means that L is purely inseparable over the field of

.
fractions of RI/(P n RI) Therefore, we may take the inertia field instead of
.
L. Thus, we may assume that VP = { l} .

We start with a special case where G is a finite group. We may assume
that L is separable over K and [L : K] = m = #(G) From our observation
.
on f , f above, we see that each element of RL/P is a root of a polynomial

of degree at most m over R/(P n R) . Consequently, every simple extension
of K contained in L is of degree at most m. In particular, [L: KS < r n,
and therefore, #(G(L/K)) < m. But the order of the subgroup HP/VP = G
of G(L/K) is m. Therefore, G(L/K) = HP/VP . (Note that it follows that
[L: K]S = m. Hence, if there is an inseparable element in L, then there is
a simple extension of degree > m, which is not the case. Thus, we see that
L is separable over K.)
Now we consider the general case. Each finite normal extension of K
contained in L is generated by residue classes of elements of the integral
closure of R in certain finite normal extension of K. Thus, the topology on
G = HP/VP coincides with the one as a subspace of G(L/K). G is a compact
groups, and therefore, G is a closed subgroup of G(L/K). Hence, if G :
G(L/K), the intermediate field M between K and L, which corresponds
to V, contains a separable element a over K that is not in K. Then,
taking a finite normal extension of K that contains a representative a of
C7, we easily have a contradiction by the special case above. Thus, HP/VP =
G(L/K). Q.E.D.
By what we noted in the proof above, we have
COROLLARY 6.3.3. Let L, K, RL , P be as above and let RS , RI be the
splitting ring, the inertia ring of P , respectively. Then RI/(PnRI) is separable
4. ALGEBRAIC EQUATIONS OF HIGH DEGREES 221
over RSl(Pf1Rs). On the other hand, RLIP is purely inseparable over RI/(Pfl
RI).
LEMMA 6.3.4. Let R and RL be as before. Let I be an ideal of RL and
let T be a subset of R such that no element of T is a zero-divisor modulo
I (i.e., b E T, c E RL , be (=- I implies c c I). Then, taking RT , (RL)T'
and I' = I(RL)T instead of R, RL, and I, respectively, we can consider the
ramification group VI, of P. This VI, coincides with the ramification group
VI of I .
PROOF. I' fl RL = I by the assumption on T. Therefore, a E VI, if

ab - b E I' for all b E (RL)T, namely, iff ab - b E I for all b E RL , which
is equivalent to a E VI . Q.E.D.
This lemma shows that when we are dealing with the inertia group of a
prime ideal P of RL or the ramification group of a P-primary ideal, we can
reduce to the case where P n R is the unique maximal ideal of R.
We add here a result on ramification groups in a special case.
THEOREM 6.3.5. With R, RL as above, we assume that P is a maximal
ideal such that P = pRL for some p E P and such that R/(PfR) is a perfect
field. Denote by Vn the ramification group of P'. Then Vn+1 is a normal
subgroup of Vn and V,,I Vn+l is isomorphic to a subgroup of the additive group
RL/P .
PROOF. We may assume that R is the inertia ring of P, because any

algebraic extension of a perfect field is perfect. Thus, we assume that L is
separable over K and G = VI . Corollary 6.3.3 shows that RL/P = R/(P fl
R) . Let Rn be the ramification ring of
pn Then R = RI c R2 c
. c_ R11 c
... C RL . a E Vn implies p - up E
pn , and we can write up = p + p"a(a)
(a(a) E RL). Let 0, be the mapping of Vn to RL/P such that a is

mapped to d(a) = (a(a) mod P). Since RL/P = R/(Pf1R), we have a(a) E
R/(P n P), and we can write a(a) = b(a) + pc(a) (b(a) (=- R, c(a) E RL) .
Then zap = z(p + p"a(a)) = p +p"a(z) + (p +p"a(z))"(b(a) + (Tp)T(c(a))
p + p"(a(z) + a(u)) ( mod Pn+1). Consequently, is a homomorphism.
It remains only to show that the kernel of 0 is Vi+1 If a c n+1 , then
.
up _p E P"+1 , or equivalently, p"a(a) C Note a(a) E P if a(a) = 0,

p11+1 .
i.e., a is in the kernel of 0. On the other hand, each a E RL is written

as a = a0 + app (a0 E R, a1 E RL) Hence, a E V,,, ap - p G P'1+1
.
implies as - a = (aal )ap - app = (al +p'1(a(a))(p +p'ta(a)) - aip E pn+1

(a(a) E RL ; note that a(a) E P), so we have a E Vn+1 . Thus, the kernel of
0 coincides with Vn+1 . Q.E.D.
4. Algebraic equations of high degrees
We shall show here the existence of a polynomial over the rational number
field such that its Galois group is the symmetric group S of degree n , as
an application of Theorem 6.3.2. We begin with a preliminary result on Sn .

THEOREM 6.4.1. If a, T , ri E Sn (n > 2) are cyclic permutations of length
2, n, n - 1 , respectively, then (a, T , 7) = Sn .
PROOF. Rewriting letters suitably, we may assume that ri = (2, 3, 4, ... ,
n).
(1) First we shall show that if a 1 0 1 , then (a, ri) = Sn Indeed, .
a = ( 1, m) with certain m, and all of { (l , k) I k = 2, 3, ... , n j appear

among riàri-`. If j 54 k, k 0 1, j 0 1, then (j, k) = (1, j)(1, k)(1, j),
and these elements generate Sn .
(2) Assume now that aI = 1. T'QT-` is of the form (j, k). For a
suitable i, 1 appears among j or k. Then we can apply (1), to such a
T'QT-! and q. Q.E.D.
Incidentally, Theorem 6.3.2, in the rational coefficient case, can be stated
as follows.
THEOREM 6.4.2. Assume that the coefficients ci of f (x) = xn + cI xn-' +
c2xn-2 + + cn are all rational numbers. Let R be the ring of algebraic
integers in the minimal splitting field L of f (x) and let P, p be a maxi-
mal ideal of R and the prime number contained in P (i.e., P n Z = pZ),
respectively. Each element of G(L/Q) gives an automorphism of R and each
element of H = {a E G(L/Q)IQP = P} induces an automorphism of RIP.
Furthermore, R/P is a Galois extension of Z/pZ whose Galois group consists
of automorphisms induced by elements of H.
Now we shall achieve our aim by proving the following theorem.
THEOREM 6.4.3. For each natural number n > 2, there exists a polynomial
f (x) = xn + clxn-' + + cn with rational integral coefficients ci such that
its Galois group is the symmetric group Sn of degree n.
PROOF. If n = 2, then any irreducible f (x) over Q satisfies the require-
ment. The n = 3 case is also easy and we leave this case to exercise 6.4.1. So
we shall consider the case n > 4. We choose three mutually distinct prime
numbers p, q, r and three polynomials g(x), h(x), k(x) with rational
integral coefficients satisfying the following conditions.
g(x) = xn + a,xn-' + a2xn-2 + + an is irreducible modulo p (over
Z/pZ).
h (x) = xn + b1 xn-' + b2xn-2 + - + bn modulo q has an irreducible factor
of degree n - 1.
k(x) = xn + elxn-' + e2xn-2 + + en modulo r decomposes into the
product of an irreducible quadratic polynomial and n - 2 linear factors and
has no double root modulo r.
By Theorem 1.3.5, there are rational integers ci (i = 1 , ... , n) such that
ci = ai ( mod p) , ci - bi (mod q) , ci - ei ( mod r) . It suffices to show
4. ALGEBRAIC EQUATIONS OF HIGH DEGREES 223
that the Galois group of f (x) = x" + clxn-1 + c2xn-2 + + cn is Sn S.

First, we apply Theorem 6.4.2 to a maximal ideal P containing p. Then
by Theorem 2.9.3, the Galois group G((R/P)/(Z/pZ)) is a cyclic group,
hence it is generated by an element T T is a permutation of roots of f (x)
.
modulo p, namely, a permutation of roots of g(x) modulo p and these n

roots are conjugate to each other. Therefore, T must be a cyclic permutation
of length n . Thus the Galois group of f (x) , as a permutation group of roots
of f (x) , must contain a cyclic permutation of length n, by Theorem 6.4.2.
Similarly, by applying Theorem 6.4.2 to a maximal ideal containing q, we
see that the Galois group of f (x) contains a cyclic permutation of length
n - 1 By applying the theorem to a maximal ideal containing r, we see that
.
that Galois group contains a transposition. Therefore the Galois group is Sn

by Theorem 6.4.1. Q.E.D.
In the proof above, we can choose prime numbers p, q, r with only the
restriction that r > n - 2. Furthermore, once p, q, r are fixed, even when
g(x), h(x), k(x) are determined and f (x) is chosen, Sn is the Galois
group of f (x) + pgrfo(x) , whenever fo(x) E Z[x] and deg fo < n - 1 .
Thus we can see that there are many polynomials (E Q[x]) of degree n
whose Galois group is Sn .
We shall give a concrete example in the case n = 5, following the proof
above.
EXAMPLE. Let p = 5, q = 2, r = 3 and
g(x) = x5 - x - 1,
h(x) = (x4+x+ 1)(x+ 1) x5 +x4+x2 + 1 (mod 2),
k(x) _ (x2 + x - 1)(x + 1)(x - 1)x - x5 + x4 + x3 - x2 + x (mod 3).
Then, one example of f (x) is
f (x) = x5 + 25x4 + 10x3 + 5x2 + 4x + 9.
(i) To prove the irreducibility of x5-x-1 (mod 5) , note that elements

of Z/5Z annul x5 - x , and therefore, x5 - x - 1 has no root in Z/5Z.
Let a be a root of x5 - x - 1 in an extension of Z/5Z. Then for any
C E Z/5Z, (a + c)5 - (a + c) = a5 + c6 - a - c = a5 - a, consequently, a,
a + 1 , a + 2, a + 3, a + 4 are roots of this polynomial, hence they are
all of the roots. Therefore, there is a conjugate a + i (i # 0) of a and
there is an element a of the Galois group which maps a to a+ i. Then
a2(a) = a + 2i, a3(a) = a + 3i, U4(a) = a + 4i, a5(a) = a + 5i =a, and
therefore all of these a + j are conjugates of a. Thus x6 - x - 1 (mod 5)
is irreducible.
(ii) To prove the irreducibility of x4 + x + 1 (mod 2) ; note that any
root fi of this polynomial is not in Z/2Z. If this polynomial is reducible,
then it has a quadratic factor, and therefore, it suffices to show that 8 is not
in a quadratic extension K1 of Z/2Z. Since #(K1) = 4, elements of K1

annul x4 + x , which shows that /3 is not in KI
(iii) To prove the irreducibility of x2 + x - 1 (mod 3) ; note that this
polynomial has no root in Z/3Z.
Thus, we see that these g, h, k satisfy the requirement in the proof of
the theorem above.
Exercises
1
1. Let L, L' be normal extensions of a field K. Let K' be a field, and

assume that L D L' D K' D K. Prove that the following statements are
equivalent.
G(L'/K') = { 1 } .
L' is purely inseparable over K'.
G(L/L') = G(L/K').
2. Show that, in Lemma 6.1.1, the homomorphism from G(L/K) onto

G(L1 /K) is a homomorphism of topological groups, and consequently,
that G(L1 /K) is isomorphic to the factor space G(L/K)/G(L/L1) .
3. Let F be a finite field, and let F be its algebraic closure. Prove the
following.
(i) The additive group of the set Z of rational integers can be made a
topological group by letting N = {nZJn is a natural number} be a
fundamental system of neighborhoods of 0.
(ii) There is a homomorphism 0 from the additive group Z to G(F/F)
such that
(1) the topological group Z above is isomorphic to the subgroup
q5Z of G(F/F) with topology as a subspace of G(F/F),
(2) c(Z) is dense in G(F/F) ; hence, G(F/F) is the completion
of 0z'
(3) the element 01 = a has the property that, for any a E F ,
as = aq with q = #(F).
(iii) G(F/F) is a commutative group. Furthermore, if F' is another fi-
nite field and F' is its algebraic closure, then G(F'/F') G(F/F).
2
1. Let K be the algebraic closure of a field K and let H be the set of order-
finite elements a E G(K/K) Prove that either H = { 1 } or, for any a in
.
H such that a 54 1 , { 1 , a} forms a subgroup. In this latter case, there is

a one-to-one correspondence between real closed intermediate fields and
such a. Consequently, each conjugate of a in G(K/K) corresponds to
a way of making K an ordered field.
EXERCISES 225
2. Let L be a normal extension of a field K, and let 7t be the prime field

of K. Prove the following:
(i) Aut,r L = I' is a topological group under the following topology.
For a pair of n-tuples (a) = (al , ... , a1,) , (b) = (bl, ... , of
elements a, , bi , of L, we set v(a, b) = {a E I'laal = b; (i =
1, 2, ... , n)}. The family of such v(a, b) (n being varied) is
taken as a subbase of open sets of IF.
(ii) The topological group G(L/K) is a subspace of IF above.
(iii) If H is a compact subgroup of IF, then L is a Galois extension of
KO = {x E Llcx = x (for all a c H)} and H = G(L/Ko).
[Hint. For the proof of (iii), show first that {gala C H} is a finite
set for each a c L.]
3
1. Let L, K, RL, R, G be as in 3. When I is an ideal of RL , HI = j a c

GI aI = I} is called the splitting group of I. Prove that HI is a closed
subgroup of G. Prove also an analogue of Lemma 6.3.4.
2. Consider splitting groups Hp, Hp, of prime ideals P, P'. Show that
both of the following are false.
(i) P C P' implies Hp C Hp.
(ii) P C P' implies Hp, C_ Hp.
Prove, on the contrary, that as for the ramification groups I C I' implies
VICVI,CHI,.
3. Prove that, in exercise 2 above, (ii) holds good if R is a valuation ring.
1. Prove the case n = 3 in Theorem 6.4.3 in the following form. The Galois
group of f (x) = x3 + ax2 + bx + c (a, b, c being rational integers) is
the symmetric group of degree 3, if there are two prime numbers p, q
such that
(i) f (x) (mod p) is irreducible over Z/pZ, and
(ii) f (x) (mod q) is the product of a linear factor and an irreducible
quadratic factor over Z/qZ.
2. Give a concrete example of f (x) , as in the last part of 4, in the case
n=5,p=3,q=2,r=7.
Answers and Hints
CHAPTER I
1. Exercise. (1) (1234) (2)

(1234) (3)
(12345)
341 2 ' 321 4 ' 345 1 2 '
(4) (12345)
12534
7. Exercise. When H1, ... , Hr are submodules of a module G over a
ring R, then the following is a necessary and sufficient condition for G to be
the direct sum of H1 , ... , Hr. G = H1 + +Hr and (H1 + +Ht_ 1) nH1 _
{O} for i=2,...,r.
EXERCISES
2. Cf. exercise 1.2.4.

4. Symmetric group of degree 3.
7. Since (ab)et = b", (b'") = (b), we have (b) c (ab). Similarly, (a) c
(ab) and (a, b) c (ab).
8. If Z = (a), h = #(Z)/q, then the solutions of xq = 1 in Z are ah ,
a2h , ... , aqh
9. Use exercise 1.2.8. Show that if G is solvable and N is a normal sub-

group, then GIN is solvable. As for the solvability of N, consider
Go=G, G,=[G,_1,G,_1], No=N, N=[N,_1, N_J, then we
can show N C G, by induction on i. Note that this proof can be ap-
plied to show that if N is a subgroup of a solvable group, then N is
solvable.
10. The center Z(GI x x of G1 x . . x G coincides with Z(GI) x
.
xZ(G,,) and (G1 x xG,)/Z(GI

11. If #(G/H) = 2, then #(H\G) = 2; hence, a E G, a V H implies
G=HUHa=HUaH. Thus all=Ha.
12. As for (1), use exercise 1.1.7. As for (2) , use (1) and the hint.
227
228 ANSWERS AND HINTS
- I J C- I n J.
3. (1) (i) If I, J are ideals of a ring R, then (I + J)(I n J) C
(ii) R=(I1+I2)2cI +I2 and Ii +I2=R.
(2) Use (1) and an induction argument on n.
4
2. The rational number field.

5
1. (2) If a, b, c, d E R, a 0, b 0, and ab = 0, then, setting f =

ax + c , g = bx + d , we have deg f g < 1 < deg f + deg g .
(3) deg(f1 + f2) < max{deg f1 , deg f2 } ; if def fl deg f2, then deg(f1 +
f2) = max{deg f1 , deg f2} .
6
4. See the proof of Theorem 1.6.4.

7
2. Z/nZ with a natural number n, or a module containing this as a sub-
module.
4. Cf. the Jordan-Holder-Schreier theorem.
5. First take a composition series Ri = Rio D Ri1 D R(j) _ {0}
of R1-module Ri . Set Tj = Rij + RI+I + + Rn (i = 1, 2, ... , n ;
j = 0, 1, ... , c(i)). Then consider a refinement of R = T 0 D T I D
lengths R = sic(i)
REMARK 1. .
REMARK 2. By using Tl as above, we can see that lengthR R = > c(i) _

lengthR7 Ri .
1. The if part is obvious. As for the only if part, when T, T' are odd
permutations, TT', T- I T' are in An , and TT' f = f , T f = T' f . Thus,
g = f - T f is independent of T, and Tg = T f - f = -g. Hence, g is an
alternating form. h = f+Tf is a symmetric form, and f = (h/2)+(g/2) .
6. g1 f is expressed in the form k/h with h a symmetric form. Then

k is also a symmetric form. If there is a common irreducible factor
p, then the product of mutually distinct elements of {apla E Sn } is
a common factor and k/h can be reduced. Here is another solution.
(ag)/(af)=g/f (for all a E S,7), and we see that of=caf, ag=cag
with ca E K. For a, T E Sn , caT = cacT . If u is a transposition,
then ca = 1 , which shows that either both of f , g are symmetric or
alternating. The latter case is impossible by Theorem 1.8.2.
ANSWERS AND HINTS 229
CHAPTER II. EXERCISES
3. Cf. Theorem 2.1.4.

4. If a C M, then there are c1,...,cnEL, a"+cn_1a +cn=0; '+
hence, [K(a,c1,...,cn):K]=[K(a,c1,...,cn):K(c1,...,cn)][K(ci
, ... , cn) : K] < oo. Thus, K(a, c1 , ... , cn) is algebraic over K and
every element of M is algebraic over K.
2
1. (1) Q(V, /) [In finding the roots, the equality x4 - x2 + 4 = (x2 +

2)2 - 5x2 is useful.] (2) Q(om) (3) (Q(v , r2)
2. (1) 4 (2) 2 (3) 6
1. (1) It has a multiple root if the characteristic is 2 (1 is a 4-ple root); no

multiple root otherwise. (2) It has a multiple root if the characteristic
is 229 (-4/3 is a double root); no multiple root otherwise. (3) It has
two double roots if the characteristic is 2 (square roots of two roots of
x2 + x + 1 are the double roots); one multiple root if the characteristic is
3 (1 is a 4-ple root); one double root if the characteristic is 139 (-7 is a
double root); no multiple root otherwise. (4) No double root if d 0;
it has at least one multiple root of multiplicity > p, if d = 0.
3. Find a contradiction assuming that t'/p E K(t) .
6. If a is not separable, then the minimal polynomial for a is a polynomial
in xp , so (degree of ap) < (degree of a) , and hence, K(ap) K(a) .
Conversely, if K(ap) K(a), then a is not separable over K(ap), and
hence, a is not separable over K.
4
1. Take a E Z such that a + pZ is a generator of the cyclic group consist-

ing of nonzero elements of Z/pZ (such an a is called a primitive root
modulo p). Then the order of a + p"Z (in the multiplicative group U
of invertible elements of Z/p"Z) is a multiple of p - 1 . Hence, there is
an a E U of order p - 1 . Set /3 = (p + 1) +p"Z E U. Then the order of
/1 is pn-1 Since #(U) = pn-1(p - 1) , we see that a/3 generates U. If
.
p = 2, then the group is the direct product of the cyclic group (or order
2) consisting of the residue classes of , -1 and the cyclic group (or
1
order 2n-2) generated by the residue class of 5.

2. Use exercise 1.2.12.
3. We define multiplication on the set of 8 elements, the identity I, i, j, k,
-1, -i, -j, -k, as follows: i2 = j2 = k2 = -1, i3 = (-1)i = -i,

j3=(-1)j=-j, k3=(-1)k=-k, ij=k, jk = i, ki=j,
j i = -k, k j = - i ik = -j. Then these 8 elements form a group, in
,
which the solutions of x3 = 1 are 1 and -1 only.

5
1. Let t, u be algebraically independent elements over a field ko of char-

acteristic p 0, and consider k = ko(t, u) and pth roots tp , up of '
t, u. Then k(tp , ) is an extension of degree p2 over k, and is

up-
not a simple extension. For the proof, first show that the extension is of
degree p2 , and then adapt the last part of the proof of Theorem 2.5.5.
For the degree part, cf. the hint for exercise 2.3.3 in the cases:
(i) Apply it to ko(t, u) = ko(t)(u) and obtain [ko(t, up-') : ko(t, u)] _
p,
(ii) Apply it to ko(t, up-') = ko(up ')(t) and obtain
[ko(tp , up ko(t, up )] = p
2. Let L = K(a) , and consider the minimal polynomial f (x) = fl.' (x-ai ) 1
for a. For each intermediate field M, the minimal polynomial fM (x)

for a over M is a factor of f (x) , and the set SM of roots of fm is
a subset of {a1 , .'.. , a}. If M M', then SM # SM, , and therefore,
the number of intermediate fields < (the number of nonempty subsets
of {al,...,an})=2n-1.
6
1. (2) Let t, u be algebraically independent elements over a field ko of

characteristic p # 0, and set K = ko(t, u). For simplicity, we assume
that p 2. Set LS = K(/) and L = LS((Vt- + u)p ) Then LS .
is a Galois extension of K and L is purely inseparable over LS But .
(-Vt- + u)p is a conjugate of (/ + u)p over K and is not in L.

Thus L is not normal over K.
3. Assume that L = K(a) and that the minimal polynomial for a over K
is R X) = H", (x - al) . Then the degree of ai over K(a1 , ... , ai_,) is
at most n-i+1.
5. (1) Q(V), {1, U} (Uv=-v ).
(2) Q(V/'2-, V3) { 1, 9, T, UT} (U2 = T2 = 1 , UT = TU ; UV = V
= V, TO= V(2, T/=-V3)
UVr3_ .
(3) Q('2, v-3), {1,, a, U2/, T, TU, TU 2} (U 3 = 1, = 1, TUT = T2
U(-1 ; U V-3 , a(/ ) = w(3 2) with w = (- 1 + V

= r2, T V 3 - , ,/ - 3 )
T( 3 2) -
(4) Q(4 2, V-1) , {1 , a, U,
2 U,3 T, TU, TU2, TU3} (U4 =1 , T2 =1 ,
TQT = Q-' ; Q _,
ANSWERS AND HINTS
7
a(0) = r2 , T(4 2) =
1. (1) {1 } . (2) See exercise 2.6.5, (4) . (3) 111 in the characteristic 2
0,
231
case: { 1 , a} (a2 = 1 , ax = -x) ; otherwise. (4) If K has a root

w 1 of x3 - 1 , then { 1 , a, a2} (a3 = 1 , ax = (ox) ; otherwise { 1 } .
2. (1) Q(3 2) . (2) Q. (3) K(x) in the characteristic 2 case; otherwise
K(x2) . (4) K(x3) if K has a root w 1 of x3 - 1 ; K(x) otherwise.
3. If we consider the matrix of the transformation with base b' , ... , b;,
instead of b1 , ... , b , it is of the form A-' p(a)A with a regular ma-
trix A of degree n. Therefore, the trace and determinant of the trans-
formation matrix do not change (the invariance of the trace follows
from that the trace of p(a) is (the coefficient of x"-') x (-1)"-' of
det(p(a) - xE) (where E is the unit matrix)). Therefore, we can choose
a base b1, ... , b" as follows. Let xr + cix r-1 + . . + c,, be the mini-
.
mal polynomial for a over K and let d1 , ... , d,U be a linearly inde-
pendent base of L over K(a) (rv = n). Now, let b1 , ... , b be 1
a , ... , ar-l, d1 , d1a, ... , dear-1, ... dia', ... , d,U , da, ... , da"-'
(5) In the case where L is separable over K, if [L : K] is not a multiple of

the characteristic p of K, then, with nonzero element b of K, we have
TrL,K b = nb 0. So we consider the case where [L : K] is a multiple
of p. Take a such that L = K(a), and let f (x) = x" + cIx n-1 +... + c"
be the minimal polynomial for a over K. The conjugates of a are
a1a = a1 , ... , Q"a = an . Since a is separable, there is one i such that
c1 is not 0 and i is not a multiple of p. Let j be the smallest such i.
Set Pk = a1 + + ak for k < j. Then, by the relationship between

elementary symmetric forms s1 = -c1 , s2 = c2 , ... , s" _ (-1)"c" and
pk (see exercise 1.8.5), we see that Pk = 0 if k < j and pj # 0. (We
have another proof by using Lemma 2.9.9.)
8
1. The inseparable case is obvious. In the separable case, apply exercise

1.2.11 to the smallest Galois extension containing L.
2. (i) In the Q(' , , V3) case; the Galois group G = { 1 , a, -r, K, UT,
TK, KU, QTK } is commutative and every element is of order 2.
1
Therefore, there are 7 subgroups of order 2, and 7 subgroups of order 4.

By adding the number of { 1 I, G, total number of subgroups, namely,
the number of intermediate fields is 7 + 7 + 2 = 16.
(ii) In the Q(w' 2) case, the Galois group is the symmetric group of
degree 3, i.e., {1, u, a2, T, TQ, TQ2} (Q3 = 1, T2 = 1, TUT = Q-').
There are 3 subgroups of order 2 and there is only one subgroup of order
3. The total number of subgroups, namely, the number of intermediate

fields is 3 + 1 + 2 = 6.
(iii) In the Q(,4_2_) case, the smallest Galois extension containing this field
, a3 , i ,
is Q( 2 , ) , whose Galois group G is{ I, a, a2
TQ, to 2 , ta3} (Q4 = 1 , Z2 = 1 , tai = Q-1 ; a(v'--l a,-l
V , Q(4 2) _
4
2 , -c(am) _- , Z(4 2) = 0). The subgroup H corre-
sponding to Q(4 2) is { 1, z} The subgroups containing H are H, G,
.
and 11, T, T9, TUT}. Thus, the answer is 3.

9
3. If we can prove this in the case K = Q (the rational number field), then
the general case can be proved as in Theorem 2.9.4. The K = Q case
follows from Theorem 2.9.5 and exercise 2.4.1. If n = 2S with natural
number s > 2 and if K = Q, then the extension is not cyclic.
4. Take a such that L = K(a). Then 1, a, ... , an-1 form a linearly
independent base of L over K. On the other hand, det(ai (a')) 0,
because
xn-1
1 X 1
x21 1
... x2-1
1 x x2
.. ... ... ... 2..
= II(xi - xi).
Xn-1 i>j
1 xn X2n n
6. (2) If a1 u , ... , un u form a normal base, then a1 u , ... , an u form a

linearly independent base of L over K , and hence, det A 0 (cf. exer-
cise 2.9.4). Conversely, if detA 0, then a1 u , ... , an u form a linearly
independent base; hence, a1 u , ... , an u form a normal base. As for the
existence of a normal base, see exercise 2.9.5.
7. (1) To prove the only if part, set xi = x) (i = 1, 2, ... ,
n). Then axi = x-1xi+1 (i < n), xn = 1 , axn = 1 = x-1x1. By
Lemma 2.9.9, we see that y = > xiaù 0 for some u E L. Then
ay=yx -1 x=Y/(ay) ,
(2) To prove the only if part, set xi = x + ax + Then t = + at-1x .
> a' u o 0 for some u E L. Set z = E t-1 xi a` u. Then z - a z=

t-lxlaiu - E t-1(xi+l u =x.
E - x)ai+1
10

3. (2) Let Hi be the subgroup of G = G(L*/K) that corresponds to Ki.
Then G = Ho D H1 D H2 D H. D { 1 } is a normal chain and
aHra-1
naEG = { 1 } . The solvability of G follows from this and the fact

that H /Hi_ is a cyclic group for each i. An example of L* , which
1
is different from L, is obtained by taking K = Q (the rational number

field), L=Q(42) = K2, K1 =Q(/).
4. (1) The direct product of three cyclic groups of order 2.

(2) If n = 2, then it is a cyclic group of order 2. In the general case, let
U be the group of invertible elements in Z/nZ (Z being the ring of
rational integers), and let Z, = { l , b, ... , bn-1 } be the cyclic group of
order n. Then the Galois group is isomorphic to the group generated by
aba-'
U, and Z,1 with defining relation = bd if a = d + nZ E U,,.
(3) The symmetric group of degree 3. [The discriminant of the polyno-
mial is 31, and therefore 31 = (the difference product of the roots) E
the minimal splitting field. Hence, the Galois group cannot be the cyclic
group of order 3. This fact and the irreducibility imply the answer.]
(4) The symmetric group of degree 3 by a similar reason as above.
(5) The cyclic group of order 2. [The minimal splitting field is Q((O)
with an imaginary cubic root w of unity.]
(6) The direct product of two cyclic groups of order 2. [The minimal
splitting field is Q ( w , ) = Q( ,) .]
11
1. Use Theorem 2.11.5.

3. Consider a factor x2 + ax + b of x4 + x + 1 with a, b in a splitting field.
b-1
Then x4 +x + 1 = (x2 + ax + b) (x2 - ax + b-1) ;hence, a2 = b + ,
a-1 =b-' -b. Thus, 2b = a2 - a-' = a2 +

,
2b-1 a- I
and therefore,
a4 - a-2 = 4, a6 - 4a2 - 1 = 0. This shows that the minimal polynomial
for a2 is X 3 - 4X-1. Since a2 is in the minimal splitting field, the
order of the Galois group must be a multiple of 3.
12
3. Take a natural number n > 1 and consider finite fields Ki such that
#(K!) = p"i . Then K1 c K2 c , and K = Uj Ki is a field such that
- -
#(K) = oo. Take a natural number m > 1 which is relatively prime to

n, and let a be an element of degree m over 7r, then a is not in any
Ki, consequently a is not in K. Thus, K is not algebraically closed.
CHAPTER III. EXERCISES
3. To find a counterexample to (iv) implies (iii), take L = K(" 2) , M =

K(C r2) with the rational number field K, a primitive nth root C of
unity and an odd number n > I.
3
2. One remark: It is not a right answer that, considering a prolongation D'

of D to L, we take the restriction of D' to M, because D'M C M
may not be true.

If M has a separating transcendence base, then the existence of the
required prolongation is obvious. So, the characteristic 0 case is finished,
and we assume that the characteristic is p 0 0. In general, if we fix a
p-base B, then there is a one-to-one correspondence between derivations
of L and elements of Homset(B, L) (the set of mappings of B to L).
Now, we choose B so that L"(K) = LP (K n B) , L"(M) = LP (M n B) .
We define a prolongation D' of D by letting D'b be Db if b E K n B

and any element of M otherwise.
4
2. We can reduce the problem to the case where LI , L2 are finitely gener-
ated. Then we can use separating transcendence bases.
3. [KI : K] < i(L/K) is easy by considering L K KI . i(L/K) < [L
K(x1 , ... , x")]1 follows from Theorem 3.4.3 and the fact that [L(K" ) :
K -00 (x1 , ... , x")] > [L : K(x1 , ... , x,,)],. For this last fact, use Theo-
rem 2.7.3, (ii).
5
2. To prove sufficiency, note that L is separable because L K L' is an inte-

gral domain for any purely inseparable extension L'. K is algebraically
closed in L, because L K L' is an integral domain even when L' is
separable over K. Cf. Theorem 3.5.2, (ii).
3. Use Corollary 3.4.5. (and Theorem 3.5.2, (ii)).
6
3. (i) For a counterexample, let x, y be algebraically independent elements

over a field K, and set R = K[x, y, y/x, y/x2 , ... , y/x" , ... ] . Then,
y, y/x, ... E xR = I, and R/I ^_' K. (ii) For (*) in the case I" =
{0} , we shall show that R/IS is Noetherian by induction on s. It is
so if s = 1 , and we assume that s > 1 . Let J be an ideal of RIP
and let be the natural homomorphism of R/IS to R/IS-' Then .
by our induction hypothesis, J' = OJ is finitely generated, and there

exist bi , ... , bt E J, J' = > q(b1)(R/IS-I) . Then J = > bl(R/IS) +
(J n (IS- I /IS)) . Since IS-I /IS is finitely generated as an R/I-module,
its submodule J n (IS-' /IS) is finitely generated. Thus, J is finitely
generated.
4. I is not a primary ideal, because xy E I, x V I, y" I for all n .
5. M/I is the unique prime ideal of R/I. Hence, x M (x E R) implies

x mod I is invertible in R11.
6. Let I = Qi n . . n Q,t be a shortest expression of I as an intersection
.
of primary ideals. For the only if part, P = implies there is c c

Q2 n n Qn , c Q1 Then, I : c = QI :c and (Q1 : c) =P. Take

.
d such that d c (Qj : c) : P, but d 0 Qi : c. Then take b = cd. For

the if part, I : b = n(Qi : b) and there exists i, P = Qi : b.
7. (iii) For the only if part, take a set of generators fi , ... , fn of the
ideal M = EI Ri such that every f is homogeneous. We show
Ri C Ro[ f , ... , f,,n] by induction on j. This is obvious if j = 0.
Assume that j > 0. g E Ri implies g E M = g = E f gi with gi ho-
mogeneous. Since deg gi = j - (deg f) < j, we have gi E Ro[fi , , fn]
and gERo[f ,..., in]-(iv) +"
For the if part, we show that f E Ri , gj c Rj , (>ci=c f) (Ed d 9j)
E Q, E f P implies E gj E Q by a double induction on s and t.
Note that we may disregard those gj which are in Q. The s = 0 case
is easy, because fcgj are in Q. The t = 0 case is similar. So, we
assume that s > 0, t > 0. Since f gd E Q , Ej+d+I fCgj rj+d fcg,
(mod Q) .This implies that (>i fcgj) e Q By our induction
.
on t, we have F,j+d+l fcgj E Q, which shows that each fcgj E Q ; hence

(fitC+I i)(Ed d gr) E Q. If fc E P, then we can use this last relation
and an induction on s ; otherwise, that fcgj E Q for every j.
(v) Using the result (iv), adapt our proof of Theorem 3.6.10.
7
1. If a prime ideal P of S contains I, then there is a P', a prime ideal

of R, such that P' n S = P . Applying this to P such that ht I = ht P ,
we have ht I = ht P = ht P' > ht IR . Conversely, since I C IR n S, we
have ht IR = ht(IR n S) > ht I.
2. Take a prime ideal PP of height 0 and contained in Pn Then apply
.
Theorem 3.7.12 to R/Po .

5. For the first half, X2 - 2 is irreducible over Z[2V]. But X2 - 2 =
(X - /) (X + /) over its field of fractions. For the latter half, if
f (X) = 1I , (X - ai) with integral elements ai over R and if f =
g(X)h(X) with monic polynomials g, h over its field of fractions, then
the coefficients of g, h are integral over R, because they are expressed
as polynomials in ai .
6. For the first half, if f (x) _ (aox'n + + a,n)(box "-'n + + bn_,n)
with n > m > 0, then a, , ... , a,n , bl , ... , bn_,n are in pR, because
f (x) = cox" ( mod p) . Then cn = a,nbn_in E p2R , a contradiction.
8
1. For the last part, we can choose zi , ... , z, from linear combinations of
a, , ... , an with coefficients in K.
2. For the first half, assume that P 0 aR. x c P implies x = axe (for
some xi E R), so xI E P (because a V P). Thus, P = aP, and
x = anxn (xn_ I = aXn) (for some xn E R). xI R c x2R c and for

some m, XmR = xm+I R . Then x,n = aX,n+I = ax,n z (for some z E R)
and xm (l - a z) =0. Since aR OR, 1 - a z : 0, and therefore, x,n = 0-
Thus, x = 0. For the latter half, if a = PI pm with prime elements
pi, then there exists an i for which pi E P and P = p,K[XI , ... , Xn].
3. It is advised to start with the latter half. If P is a maximal ideal, then
consider the field K[XI , ... , Xn]/P. Let ai be the residue class of Xi
modulo P and set Ki = K(a I
, ...
Let f (X) be the mini-
, a, _ I) .
mal polynomial for ai over Ki and let gi(XI, ... , X.) be the monic
polynomial in Xi obtained from J. by replacing coefficients by their
representatives in K[XI , ... , X1_] and X by Xi. Then P is gen-
erated by gI , ... , gn . For the first half, we can assume that r > 0.
Let YI , ... , Yn be the elements obtained by applying Theorem 3.8.2
to K[X] and P. Then K[X] = K[XI, ... X,. , and , Y, Y ]
PK(Y +,, ... Yn) [XI , ... Xr] is a maximal ideal. We apply the latter
,
half to this maximal ideal.

4. If I' is a prime ideal, then trans. degK R'/I' = trans. degK R/(I' n R) ,
which implies ht I' = ht(I' n R) . In the general case, we can adapt our
proof of Corollary 3.7.3.
9
1. R[h] c a - I R and a - I R is a finitely generated R-module.

10
1. For the first half, use the zero-point theorem of Hilbert and the fact that
V(I(AI) + I(A2)) = AI n A2. For the last part, consider the polynomial
ring P = R[xI , ... , xn ] (n > 2) over the real number field R. Then,
with A1 = V (x1) , A2 = V (-xi + En 2x2 + 1) , we have II(A1)+I(A2)
(A _
xI P + (> 2x2 + I )P = I (AI) + 1(A2), but A n A2 is empty. I
2. Let W be a component of V n H, and let h be an element which

defines H. We apply Theorem 3.8.7 to S2[XI , ... , X]/I(V), and we
have htI(W)/I(V) < I, because I(W)/I(V) is a minimal prime divisor
of the principal ideal generated by h modulo I(V). Hence, ht I (W) <
ht I (V) + 1 by Corollary 3.8.5.
4. (i) It has sufficiently many points in R if r > 0. If r = 0, then there is
no R-rational point. (ii) It has sufficiently many points in R.
11
1. Consider xi + + xn
2. Set K = R(t , ... , ti) and L = C(t , ... , ti) with algebraically indepen-
I I
dent elements tt , ... , ti over the real number field R. Here C denotes
the complex number field. K is not a C,7-field for any n and L is a
Ci-field but not a Ci _ 1 -field.
12

2. For the only if part, if (p, q) is a K-rational point, then K(x, y) = K(t)
with t = (x-p)/(y-q) (Indeed, x = p+t(y-q) ; hence, a(p+t(y-q))2+
.
by=c=ap
2 2+bq 2 . From this relation, we have y E K(t).) If (p, q)
with p 0 is a nontrivial solution of ax2 + by2 = 0, then (qp-' , 0)

is a nontrivial solution of a + bY2 = cz2 (with x = z-' , y = xY), we
see the existence of similar t and K(z, Y) = K(x, y). For the if part,
assume that x = f(t)/g(t), y = h(t)/k(t) with polynomials f, g, h, k.
If for some p c K, g(p) 0, k(p) 0, then (f (p)/g(p) , h(p)/k(p))
is a K-rational point. In the other case, factor g , k as g(t) = tsgl (t) ,
k(t) = t"k1 (t) with g1 kI such that gI (0) , kI (0) are different from 0.
,
We can assume that s > u > 0 (> 0 follows from the nonexistence of
p). Then f (O) -A 0 and a(f (t)/g1 (t))2 + b(ts-"h(t)/k1 (t))2 = cts By .
setting t = 0, we have a nontrivial solution for ax2 + by2 = 0.

REMARK. The proof above shows that if #(K) is infinite, then ax2+by2 =
c has a K-rational point if for some t c L. L = K(t).
3. Use the preceeding exercise and the fact that K(z) is a CI-field. (Con-
sider the homogeneous form fI (z)X2 + f2(z)Y2 - f3(z)U2 in X, Y, U.)
A
1. Take a valuation ring V.. of K(x1 , ... , xr) containing K and such that
V has prime ideals PI P2 D D P,. D {0} with the property that
Pi = x(V) p for each i. Then adapt Lemma 3.A.2, considering V n
K' (XI, ... , xr- I) .
2. Here is a proof of the theorem stated in the hint. If n = r, then there
is nothing to prove. Assume that n > r. We use the normalization
theorem for polynomial rings using exercise 3.8.1, and we see that there
are linear combinations yI , ... , of xI , ... , x1, with coefficients in
K such that L[xI , ... , xn] is integral over L[y1 , ... , yn-r] Choose -
aI , ... , as E L such that L = K(a1 , ... , as). Then, x, , ... , x11 are in-
tegral over K[a1 , ... , as, yi , ... , yn_r]. There is c E K such that, when
we write aI , ... , as in fractional forms of xI , ... , xn , no denominator
is divisible by yI -c. Consider the ring V =K[x1 , ... , x,,]3, _
Since aI E V, V contains K[ai , ... Set P =
, as , x1 , ... , xn] .
(y1 - c)V n K[a1,...,as,xI,...,xn] and Q = (y, - c)V n

K[a1 ... , as , yI , ... , yn-r]
,
Since the field of fractions of
R = K[a1 , ... , as, xI , ... , xn]/P is V/(yi - c)V, we see that
trans. degK R = n - 1 R is integral over S = K[a , ... , as, y i , ... ,
. i
yn-r] and trans. degK S = n- 1. Now for T = K[ai, ... , as]/

(Q n K[al , ... , as]) trans. degK T > (n - 1) - (n - r - 1), because
,
yI - c E Q. But, trans. degK K[al , ... , as] = trans. degK L = r, and

we have Q n K[a1 , ... , as] = {0} Thus, L is regarded as a subfield of

.
V/(y, - c) V ^' K(x1 , ... , xn- ).

CHAPTER IV. EXERCISES
1. (i) a > 1 (ii) c = 1 . (iii) c > 0.

.
2. If a is a nonzero element, then there is a natural number n such that

an=1.
2
1. Similar to the proof of Theorem 4.2.1, (ii).

2. (ii) Assume that an + clan-I
+ + cn = 0, v(ci) < 1 . If va > 1 , then
v(an) > v(c1an-i) and 0 = v(an + +... + cn) = v(an) > 1 , which
Clan-I
is a contradiction.
3
1. (i) Let M be a set with at least two elements and let open sets on M be
only the empty set and M itself.
(ii) With M as above, let the open sets on M be M itself and all subsets
not containing a fixed element a.
(iii) Let M be a set containing infinitely many elements and define that
a subset S is an open set if either S is the empty set or the complement
of S consists of a finite number of elements.
(iv) We fix a line L and a point A which is not on L, on the Euclidean
plane P. Then, a new topology is defined on P by letting the following
family C of subsets be a subbase of open sets. C = BI U B2 U B3 , where
BI={{x}jxE P, x0 A, x V L}, B2={UIU is an open set in the
usual topology not containing Al , B3 = {P - (L U U) I U is the union of
a finite number of circular disks}. Then L is a closed set, and there is
no pair of neighborhoods of L and of A separating each other.
(v) On the plane P, we set U(a, b, 8,J) = {(x, y) E Pea < x < a +
e, b < y < b + 61 with real numbers a, b, e, 8 We define a new
.
topology on P by letting the family of all of such U(a, b, 8, 8) be a

subbase of open sets. Then, on the line L : y = -x, the set EI of
rational points (i.e., El = {(a, -a)Ia is a rational number}) and the set
E2 of irrational points (i.e., E2 = L - EI) are closed sets. But there is
no pair of neighborhoods of El and of E2 separating each other.
H,
2. The space is not a TI -space. Indeed, there is P E G, such that P
and any neighborhood U(P) of P meets H. This implies that any
neighborhood of f (P) contains the identity element.

3. Consider the two-dimensional vector space V = {(a , b)I a , b E R} over
the real number field R. This is a topological group under the usual
topology (as a Euclidean space). H = {(0, b)Ib E R} is a closed normal
subgroup. Then the mapping f defined by f (a , b) = a is the one as
stated. F = { (a, b) IO < a < 1 , b = a-1(1 - a)- I } is a closed subset of
V, but f F is not a closed set.
4. Each sU = {sul u E U} is an open set.
5
3. (i) a , b c R , a - b E nsR (s > 1) imply a" - by c mS+ 1 R , and therefore,

n n
Ts+nR
a" - by E
(ii) If we take an , bn for a , b E R/nR , as in (i), then a n bn is a
representative of (ab)"
(iii) is easy.
(iv) is difficult, and the reader is advised to see some book, for instance,
N. Jacobson, Lectures on abstract algebra, III.
REMARK. In general, Witt vectors and Witt rings are defined over an ar-
bitrary commutative ring K with fm , gm in (iv), which define the addition
and multiplication. It is known that the Witt ring of length infinity is a
valuation ring if and only if K is a perfect field.
4. Show that mi = En= I ail u j (aij E K) (i = 1, 2, ...) form a Cauchy
sequence which converges to 0 if ai j (i = 1 , 2, ...) form a Cauchy
sequence which converges to 0, for each j = 1 , ... , n. The if part
follows from II > bj u j <- 1 I I b j u j II = E v (b j) I I u j M I (note that the in-
II
equality follows from the fact that M is a metric space). For the only
if part, use an induction argument on n Assume that a1I , ... , ant ,
.
is not a Cauchy sequence converging to 0. Then there is a positive num-

ber e such that v(ail) > e for infinitely many i. Choosing a suitable
subsequence of {m,}, we may assume that v(ail) > E for all i. Then
{ai i 1 mi } is a Cauchy sequence converging to 0, where the coefficient of
uI in each term is 1. Thus, we may assume that aI I = a21 = . . = 1 . .
Let t(i) be natural numbers such that t(l) < t(2) < . Then di =
`1J=2 Ku j , and {di} is a Cauchy sequence converging to 0.
mt(i) - mi E
Therefore, by our induction hypothesis, {cij = at(i) j - ai j I i = 1 , 2, . .. }
is a Cauchy sequence converging to 0 for each j > 2. It follows now that
{ai j I i = 1, 2, ...1 is a Cauchy sequence. Set a = limi-+00 aij . Then
u1 + a2 u2 + + a*u n = mi = 0, which contradicts the linear
independence of u 1 , ... , un .
6
1. Let B be a transcendence base of C over Q. Since #(B) is infinite,

Q(B) Q(B, t). Hence, there is an injection of C(t) to C.

2. Either its cardinality is greater than the cardinality of continuum or the
characteristic is different from 0.
7
1. Let R be such a valuation ring, and let P be a nonmaximal, nonzero

prime ideal. Let 0 a E P, and let b be an element of the maximal
ideal that is not in P. Then b-na c R and aR c b-laR c . c
2. The intersection of integrally closed integral domains is an integrally
closed integral domain.
3.a-beP iff'a>b.
4. For the first half, use Theorem 4.7.2, (vi) and the fact that R c S n K C
K. For the latter half, show that if P is a prime ideal of R, then
v'rP--S = P.
6. Consider P = {alva > h for all h c H} for an isolated subgroup H.
7. (i) Use exercise 4.7.1 for the only if part.
(ii) For the only if part; take a. E Pi such that a.RP = P R, ,
For each .
element b of the field of fractions of R, let ml , ... , Mn be the integers

such that bRPi = am' am'R p (i = 1, ... n). Then the mapping
,
wb -* (m1 , ... , Mn) gives the required isomorphism.

8. (i) Use an induction argument on n . Let w1 be the valuation defined by
RP . Let B be a subset of P1 such that {w1 bl b E B} is maximal among
linearly independent subsets of w1(P1) over Q. Consider the valuation
v defined by R/P1 . Our induction hypothesis shows that there is an
order isomorphism cI from the value group of v into the (n - 1)-ple
direct sum RED . R R. Then we define 0 as follows. Let x be a nonzero
element of K. Since w 1 x is linearly dependent on { w b I b E B}, there
I
is a natural number m such wI (x'n) = w1 (bi' bss) with b. E B. Now

0I b's modulo P1))/m)
O(wx) = (WI (bi' ... bss)/m ,
9. It suffices to show that x1 , ... , x,n (E R) are algebraically independent
if (i) wx1 = = WXr = 0 and x1 mod P, ... , xr mod P are alge-
braically independent over k and if (ii) wxr+l , .. , wx,n are linearly
independent over the rational number field. Assume for a moment that
E C; ...1 xi' ... x,n' = 0 (Ci ..., E k) . Denote by E. C...., X j' ... X,n' the
partial sum on the terms such that w (c< r
xi' x;;;) is the least. Then
we have a contradiction from w(>1 ci w (one term in
x1'
ill?
this partial sum) (this inequality follows from the fact that wa < wb
implies w(a + b) = wa).
8
1. If i 0 j, then R, [Rj] = K. (In this case, we say that R, , ... , R,, (or,
v1 , ... , vn) are independent of each other.)

9
1. We fix an algebraically closed field 92 of characteristic 0 with sufficiently

large cardinality. Consider a subfield M which has a discrete valuation
VM such that p generates the maximal ideal of the valuation ring R. of
VM and such that there is a homomorphism OM from RM to K whose
kernel is pRM Let F be the set of all such (M, vM , OM) . We introduce
.
an order > on F by (M, vM 0M) > (M' , 'M1, OM,) if M D M'' VM

)
is a prolongation of vM, , and OM is a prolongation of qM, Then F is

.
an inductive set and has a maximal member (M*, vM. , 0M.) . Then the
residue class field of vM* is isomorphic to K. Hence, the completion of
vM* is the required one.
10
1. Show, on the one hand, that we have a contradiction if there is a v E VL
such that no member of V' is equivalent to v . Show, on the other
hand, if V' = {vt(v) l v E VL} contains v1 , v2 such that t(v1) 54 t(v2) ,
then since the product formula holds with respect to {vt(vl),v E VL} , we
obtain another V' with no member equivalent to v1 .
11
3. Adapt the proof of Theorem 4.11.5.

4. (i) Use Theorem 4.11.13.
(ii) Theorem 4.11.13 shows that such a valuation ring R' contains R =
K[[X]] or is contained in R. If R' contains R properly, then R' =
K((X)) R', in the latter case, is obtained as {ao+a1XI ao c Ro , a E R}
.
with a Hensel valuation ring Ro of K.
CHAPTER V. EXERCISES
1. When b > 0, d > 0, we have alb > c/d iff ad > be.
3. 1 > 0 because 12 = 1 . Consequently, every natural number > 0.
4. Cf. the proof of Theorem 4.7.5.
5. (i) xEK, aeL, a>x<0 implies xER.
(ii) (a + P) < (b + P) implies there is a d c L such that b - a > d > 0.
7. Use Exercise 5.1.6.
2

4. For the first half, let L and K be the real number field and Q(/) (Q
being the rational number field). Let K be an ordered field with an order
such that / < 0.
4
2. Adapt the proof of Theorem 5.4.2.

CHAPTER VI. EXERCISES
2. Let x, y be algebraically independent elements over a field k of charac-

teristic 3, and set K = k(x, y), R = k[x, y]. Let z be an algebraic
element over K defined by z2+xz+y = 0, and set L = K(z) = k(x, z),
RL = k[x, y, z] = k[x, z].
(i) P = (x - 1)RL , P' = (x - 1)RL + zRL .
(ii) P = (z - ax)RL with a root a of X2 + X + 1 , and we assume that
aEk, P'=xRL+zRL.
4
1. The condition (i) implies that the order of the Galois group G is a mul-
tiple of 3 and the condition (ii) implies that #(G) is an even number.
Since G is a subgroup of S3, we have G = S3 .
5
2. An example is f (x) = x5 +6X4_ 12X3 + 15x2 -17x + 14 from x5 +X-1

(mod 3), x(x4+x+1) (mod 2) and (x2-x+3)(x+1)(x-1)x( mod 7).
Index of Symbols
C the complex number field

Q the rational number field
R the real number field
Z the ring of rational integers
Aut set of automorphisms, 11
Der( / ) module of derivations, 87
deg degree, 23
dim dimension, 114
G(/) Galois group, 55
ht height, 105
Hom( set of homomorphisms, 10, 17, 31
i(/) order of inseparability, 93
J( Jacobian matrix, 89
Krull dim Krull dimension, 105
length length, 31
trans. deg transcendence degree, 81
Z() center, 13
#( ) number of elements; cardinality, 2, 9
V generation, 8, 43
[,] commutator, commutator group, 12
degree of extension, 44
[:]
[] (e.g.,- K[al, ... , an]), polynomial ring, generation of a ring, 24,107
() (e.g., K(al , ... , an)), generation of a field, 43
(
generation of a group, 8
(
segment, 208
direct sum, 15, 19
tensor product, 83
x direct product, 2, 15, 19, 118, 183
x local tensor product, 84
ideal quotient, 98
(e.g., RS) ring of fractions, 101
V radical, 98
243
Subject Index
Abelian group, 7 comparability theorem for, 2

Act, 55 of continuum, 2
Addition, 7 Cauchy sequence, 149
Additive group, 7 regular, 149
Additive valuation, 154 Center, 13
Adjoined, 43 Centralizer, 38
Affine space, 113 Chain, 32
Algebraic, 44 normal, 40
closure, 71 Chain condition
extension, 44 ascending, 3
integer, 71 descending, 3
number, 71 Characteristic, 16
Algebraic set, 113 Chinese Remainder Theorem, 20
-irreducible, 115 C;-condition, 118
irreducible, 114 Cl -field, 118
Algebraic variety, 114 Ci -condition, 118
Algebraically closed, 71 C,-field, 118
Algebraically dependent, 24 Class, 4
Algebraically independent, 24 representative of, 4
Algebraically solvable equation, 63 Closed set, 140
Alternating form, 34 Closure, 53, 144
Alternating group, 36 Commutative, 7
Antirational, 125 field, 17
Apparent degree, 23 group, 7
Approximation theorem, 159 ring, 16
Archimedean ordered field, 211 Commutator, 12
Artin group, 12
module, 33 Compact, 145
ring, 33 Complete, 149, 184
Associated, 25 Complete system of conjugates, 77
Associated prime ideal, 100 Completion, 149, 184
minimal, 100 Complex number field, 17
Associative law, 7 Composite, 158
Automorphism, 10, 17 Composite of mappings, 2
inner-, 10 Composition series, 32, 41
outer-, 10 Conductor, 131
Automorphism group, 11 Congruent, 14
Conjugate, 10, 48, 116
Base, 30 Continuous, 143, 145
Base of open sets, 183 Converge, 149
Bilinear mapping, 83 Coordinate, 113
Binary operation, 7 ring, 115
Coset, 8
Cardinality, 2 Countable, 2
245
246 SUBJECT INDEX
Countably infinite, 2 cyclotomic, 59

Curve, 115 finite, 44
Cyclic permutation, 35 noncommutative, 17
Cyclotomic polynomial, 60 of fractions, 22
skew, 17
Defined, 115 topological, 148
Degree, 9, 23, 36, 44, 75, 115, 165 Field of definition, 115
Dense, 144 smallest, 116
Derivation, 87 Finite intersection property, 145
partial, 88 Finitely generated, 82
Derivative, 49 Formal power series, 184
Derived group, 12 ring, 185
Derived series, 12 Formally real, 191
Descending central sequence, 12 Fractional ideal, 111
Difference product, 35 Free, 86
Dimension, 31, 81, 113, 114 base, 30
Direct product, 15, 19, 183 module, 30
Direct sum, 15, 19 Frobenius mapping, 186
Discrete, 145, 187 Full matrix algebra, 17
Discriminant, 65 Function filed, 115
Distance, 142 Fundamental system of neighborhoods of
function, 142 the identity, 146
Distributive law, 16 symmetric, 148
Divisible, 187, 210
Divisor, 25 Galois
Dual, 183 extension, 54, 55
fundamental theorem of, 58, 217
8-neighborhood, 142 group, 55, 65
Eisenstein Galois extension, 105
irreduciblity theorem of, 130 Galois group, 105
Elementary symmetric forms, 34 Generated, 8, 19, 43
Endomorphism, 17 Generated, 18
Equivalence relation, 4 Going-down theorem, 106
Equivalent, 136, 149, 155, 159 Going-up theorem, 104
Euclidean ring, 26 Greatest element
Euler function, 51 greatest, 3
Extension, 88, 149 Group, 7
Abelian, 59 cyclic, 8
algebraic, 44 finite, 9
by radicals, 63 nilpotent, 12
cyclic, 59 solvable, 12
Galois, 54, 105
inseparable, 48 -homomorphism, 17
integral, 102 Hausdorff space, 141
Kummer, 78 Height, 105
normal, 54, 105 Hensel ring, 173
purely transcendental, 81 Hensel's lemma, 172
regular, 94 Henselization, 175
separable, 49, 91 Hilbert
simple, 51 zero-point theorem of, 1 13
transcendental, 44 Hilbert's 17th Problem, 202
Hilbert's Theorem 90, 79
Factor, 53 Homeomorphic, 143
separable, 53 Homeomorphism, 143
Factor group, 1 1 Homogeneous form, 23
Factor spaces, 216 polynomial, 23
Factor theorem, 24 Homomorphism, 10, 17, 29
Field, 16 onto, 10
commutative, 17 Hyperplane, 115
SUBJECT INDEX 247
Hypersurface, 115 Lexicographical order, 4

Lie, 104
-isomorphism, 17 Limit, 149
Ideal, 18 Line, 115
fractional, 111 Linear variety, 115
generated by, 18 Linearly disjoint, 84
left, 18 Linearly independent, 30
maximal, 21 base, 30
primary, 97 Linearly ordered, 3
principal, 19 Locus, 37
product of, 19 Liiroth
right, 18 theorem of, 122
two-sided, 18 Lying-over theorem, 104
Ideal quotient, 98
Idempotent, 39 Mapping, 1
Identity, 7, 16 one-to-one, 1
Image, 1 onto, 1
Independence theorem of valuations, 161 Maximal completion, 207
Independent, 241 Maximal element
Index, 9 maximal, 3
Induce, 55 Maximal ideal, 21
Induced topology, 142 with respect to, 21
Inductive set, 4 Maximally complete, 207
Inertia Maximum condition, 3
field, 219 Metric space, 142
group, 219 complete, 149
ring, 219 Minimal element, 3
Infimum, 4 Minimal polynomial, 46
Infinite field, 44 Minimal prime divisor, 100
Injection, 10, 17 Minimum condition, 3
Inner-automorphism group, 12 Minus, 7
Inseparable, 48, 49 Module, 7, 29
Integral, 44, 102 finite, 30
Integral closure, 103 finitely generated, 30
Integral domain, 20 left, 29
Integrally closed, 103 right, 29
integral domain, 103 Modulo, 8, 14, 18
Invariant, 55 Monic, 23
Inverse, 7 Monomial, 23
Invertible, 16 Multiple, 25, 29
Irreducible, 25, 114, 115 Multiplication, 7
component, 114 Multiplicative group, 17
Irredundant expression, 99 Multiplicatively closed, 21
Isolated subgroup, 187 Multiplicity, 29
Isomorphic, 10, 17, 191
Isomorphism, 10, 17 Nakayama's lemma, 167
Isomorphism theorem, 12, 30 Natural homomorphism, 11
Natural mapping, 84
Jacobian matrix, 89 Neighborhood, 140
Jacobson radical, 167 open, 140
Jordan-Holder-Schreier Theorem, 32 system of, 140
Nilpotent, 12, 20
Kernel, 10, 17 Noether
Krull dimension, 105 normalization theorem of, 108
Krull-Azumaya Noetherian
lemma of, 167 module, 96
ring, 96
Lagrange resolvent, 61 Nontrivial solution, 118
Length, 31, 33, 35, 185 Norm, 77
248 SUBJECT INDEX
Normal base, 79 Prolongation, 149

Normal ring, 103 Purely inseparable, 48, 49
Normal space, 141 Purely transcendental, 81
Normalization theorem
for a finitely generated ring, 108 Radical, 98
for a polynomial ring, 108 Radicals
of Noether, 108 expressed by, 63
Normalizer, 38 extension by, 63
Normally algebraic, 116 Ramification
Null sequence, 150 field, 219
Number of variations of signs, 198 group, 219
ring, 219
Open set, 140 Ramification exponent, 165
Operation, 7 Rank, 155
Order, 3, 9 Rational function field, 81
isomorphism, 191 Rational number field, 17
Order of inseparability, 93 Rational point, 114
Ordered additive group, 154 Rational rank, 188
Ordered field, 191 Real closed, 191
Ordered set, 3 Real closure of, 196
Real number field, 17
p-adic, 136, 158 Refinement, 32
p-basis, 90 Regular, 94
p-group, 12 Regular space, 141
p-independent, 90 Relation ideal, 24
Partition Remainder theorem, 24
disjoint, 4 Residue class, 8
Perfect, 49 ring, 18
Permutation field, 155
cyclic, 35 group, 184
even, 36 module, 11
group, 9 Restriction, 208
odd, 36 Resultant, 170
Place, 158 Ring, 16, 17, 20, 25, 148
Plane, 115 of rational integers, 17
Point, 113, 140 of total fractions, 22
Polynomial, 23 Ring of fractions, 101
Polynomial ring, 24, 72 Root, 29
Primary ideal, 97, 98 -ple, 29
Prime Roots
field, 43 of unity, 59
integral domain, 43
Prime divisor, 100 Same signs, 203
minimal, 100 Segment, 208
Prime element, 25 Separable, 48, 91
Prime ideal, 20 algebraic closure, 71
Primitive, 27 closure, 53
Primitive root, 229 factor, 53
Primitive root of unity, 59 Separation, 141
Principal ideal, 19 axiom of, 141
domain, 26 Shortest extension, 99
ring, 26 Simplest alternating form, 35
Product, 7, 19 Smallest element, 3
of cardinalities, 2 Splitting
of mappings, 2 group, 175, 225
of sets, 118 ring, 175
Product formula, 168 Splitting field, 47
Product space, 143 minimal, 47
Projection, 2 Standard sequence, 197
SUBJECT INDEX 249
Stronger, 144 Transcendental, 44

Sturm extension, 44
sequence, 198 Transposition, 35
theorem, 198 Triangle inequality, 135, 142
Subfield, 17 Trivial extension, 88
Subgroup, 8
normal, 10 Unique factorization domain, 25
Submodule, 8, 29 Unit, 16
Subring, 17 Universal domain, 114
Subspace, 142 Unramified, 165
Sufficiently large algebraic extension, 48
Valuation, 135, 154
Sum, 7, 19 p-adic, 158
Supremum, 4 additive, 154
Surface, 115 Archimedean, 135
Sylow
discrete, 187
subgroup, 14 equivalent, 136, 155
theorem of, 13 ideal, 139, 155
Symmetric form, 34 multiplicative, 135
Symmetric group, 9 non-Archimedean, 135
p-adic, 136
T;-space, 141 ring, 139, 155, 206
Tensor product, 83 trivial, 135
local, 84
Theorem of Liiroth, 122 Value group, 135, 154
Topological
group, 146 Weaker, 144
ring, 148 Well-ordered, 4
Topological space, 140 Witt
compact, 145 ring, 185
Topology, 140 vector, 185
determined by, 148
discrete, 145 Zariski-Castelnuovo theorem of, 133
Trace, 77 Zero, 7
Transcendence Zero point, 113
base, 81 Zero-divisor, 20
degree, 81 Zorn's lemma, 4
COPYING AND REPRINTING. Individual readers of this publication, and non-
profit libraries acting for them, are permitted to make fair use of the material, such as
to copy a chapter for use in teaching or research. Permission is granted to quote brief
passages from this publication in reviews, provided the customary acknowledgment
of the source is given.
Republication, systematic copying, or multiple reproduction of any material in
this publication (including abstracts) is permitted only under license from the Amer-
ican Mathematical Society. Requests for such permission should be addressed to
the Manager of Editorial Services, American Mathematical Society, P.O. Box 6248,
Providence, Rhode Island 02940-6248.
The owner consents to copying beyond that permitted by Sections 107 or 108 of
the U.S. Copyright Law, provided that a fee of $1.00 plus $.25 per page for each
copy be paid directly to the Copyright Clearance Center, Inc., 27 Congress Street,
Salem, Massachusetts 01970. When paying this fee please use the code 0065-9282/93
to refer to this publication. This consent does not extend to other kinds of copying,
such as copying for general distribution, for advertising or promotional purposes, for
creating new collective works, or for resale.
ISBN 0-8218-4572-1

Theory of Commutative Fields - Masayoshi Nagata

Diunggah oleh

Informasi Dokumen

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Theory of Commutative Fields - Masayoshi Nagata

Diunggah oleh

Hak Cipta:

Format Tersedia

Translations of Mathematical Monographs 125

American Mathematical Society

Library of Congress Cataloging-in-Publication Data

Copyright 1993 by the American Mathematical Society. All rights reserved.

Preface to the English Edition xi

Preface to the New Japanese Edition xiii

Chapter 0. Prerequisites from Set Theory 1

Chapter I. Groups, Rings, and Fields 7

Chapter II. Algebraic Extensions of Finite Degrees 43

12. Algebraically closed fields 71

Chapter III. Transcendental Extensions 81

Appendix. A theorem on valuation rings and its applications 124

Chapter IV. Valuations 135

Chapter V. Formally Real Fields 191

Chapter VI. Galois Theory of Algebraic Extensions of Infinite Degree 215

Answers and Hints 227

Index of Symbols 243

Subject Index 245

The theory of commutative fields is one of the basic areas in mathemat-

The theory of commutative fields is fundamental in modern algebra. Due

Prerequisites from Set Theory

If N is not a subset of M, then N % M or M N.

M2 (1 E A) are UREA MA and n,A MA .

the image of M1 under f is denoted by either f (M,) or Mif according to

arrange all elements of M x M as (m 1 , MI), (m 1 , m2) , (m2 , MI), ... ,

(3) If a > b and b > c , then a > c .

the maximum condition implies the ascending chain condition. Conversely,

of elements of N. If an is not maximal in N, then there is an element

A well-ordered set is a linearly ordered set satisfying the minimum condi-

infimum of S is defined similarly by interchanging > and <.

3. Partitions and equivalence relations

A relation - defined on a set M is called an equivalence relation, if it

is an equivalence relation on M. Conversely, if a relation - on M is an

Groups, Rings, and Fields

The element a' is called the inverse of a and is denoted by a-'

(a-')-' = a and (ab)-' = b-' a-'

for all elements a and b in G G.

THEOREM 1.1.2. The following three conditions on a nonempty subset H

PROOF. If aH = W, then a E bH, so a = bh for some h E H.

where Qm = m for each i.

2. Normal subgroups and homomorphisms

identity, then we say that is an injection, or an isomorphism, into G'. If,

The mapping yr of G to GIN defined by yra = aN is a homomorphism

the set of all inner-automorphisms is a normal subgroup and is called the

Then Do D DI D ... D Dr J . Let Zo = G , Z1 = [Zj_ , G] , then1

Zo D Z1 D D Zr D . These D. , Z1 are normal subgroups of G.

The sequence Do D D1 D D Dr D ... is called the derived series

Before proving this, we define the center of a group. In a group G, {x E

LEMMA 1.2.7. For a group G, we consider a sequence of groups Gi as

PROOF. Assume first that G = Z D Z1 D ... D Zn = { 1 } is the descending

homomorphism of G onto G1 = G/Z (G) , we see that the descending

THEOREM OF SYLOW. Assume that G is a finite group and its order is

(iv) If a subgroup H of G is a pi-group, then for some a E G, H C_

(iv). If H is a Sylow subgroup, then #(a- I Ha) _ #(P) and a- I Ha = P,

such an expression is unique. In this case, we write Ni x x Nr. If G is a

particular, if i j , then H, n H = { 1 } . Thus, [H , H ] c H n Hj = { 1 }

of commutativity of elements of H and Hj. Thus g- I H g c Hj, which

PROOF. 0 + 0 = 0 and a(0 + 0) = aO. By the distributive law, a0 + a0 =

group of the field R. If the multiplication is commutative, then R is called

Let 0 be a homomorphism from a ring R to a ring R'. Since 0 is

sided) ideals containing S) is the left (or right, or two-sided, respectively)

given, we consider the direct product 1(a, , ... , an) 1 a! E R } of R. , and we

THEOREM 1.3.5 (Chinese Remainder Theorem). Assume that II , ... , I

(2) If c1,...,cm,d1,...,d, E {1,...,s} (there may be the same num-