You are on page 1of 228

Lectures on the

Algebraic Theory of Fields


By

K.G. Ramanathan

Tata Institute of Fundamental Research, Bombay


1956

Lectures on the
Algebraic Theory of Fields
By

K.G. Ramanathan

Tata Institute of Fundamental Research, Bombay


1954

Introduction
There are notes of course of lectures on Field theory aimed at providing the beginner with an introduction to algebraic extensions, algebraic function fields, formally real fields and valuated fields. These lectures were preceded by an elementary course on group theory, vector
spaces and ideal theory of ringsespecially of Noetherian rings. A
knowledge of these is presupposed in these notes. In addition, we assume a familiarity with the elementary topology of topological groups
and of the real and complex number fields.
Most of the material of these notes is to be found in the notes of
Artin and the books of Artin, Bourbaki, Pickert and Van-der-Waerden.
My thanks are due to Mr. S. Raghavan for his help in the writing of
these notes.

K.G. Ramanathan

Contents
1 General extension fields
1
Extensions . . . . . . . . .
2
Adjunctions . . . . . . . .
3
Algebraic extensions . . .
4
Algebraic Closure . . . . .
5
Transcendental extensions

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

1
. 1
. 3
. 5
. 9
. 12

2 Algebraic extension fields


1
Conjugate elements . .
2
Normal extensions . .
3
Isomorphisms of fields
4
Separability . . . . . .
5
Perfect fields . . . . .
6
Simple extensions . . .
7
Galois extensions . . .
8
Finite fields . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

17
17
18
21
24
31
35
38
46

3 Algebraic function fields


1
F.K. Schmidts theorem . . . . . . . . . . . . . . . . . .
2
Derivations . . . . . . . . . . . . . . . . . . . . . . . .
3
Rational function fields . . . . . . . . . . . . . . . . . .

49
49
54
67

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

4 Norm and Trace


75
1
Norm and trace . . . . . . . . . . . . . . . . . . . . . . 75
2
Discriminant . . . . . . . . . . . . . . . . . . . . . . . 82
v

vi

Contents

Composite extensions
87
1
Kronecker product of Vector spaces . . . . . . . . . . . 87
2
Composite fields . . . . . . . . . . . . . . . . . . . . . 93
3
Applications . . . . . . . . . . . . . . . . . . . . . . . . 97

Special algebraic extensions


1
Roots of unity . . . . . . . . .
2
Cyclotomic extensions . . . .
3
Cohomology . . . . . . . . .
4
Cyclic extensions . . . . . . .
5
Artin-Schreier theorem . . . .
6
Kummer extensions . . . . . .
7
Abelian extensions of exponent
8
Solvable extensions . . . . . .

.
.
.
.
.
.
.
.

103
103
105
113
119
126
128
133
134

.
.
.
.
.

149
149
152
156
166
170

.
.
.
.
.
.
.

175
175
177
180
184
194
201
205

Appendix
1
Decomposition theorem . . . . . . . . . . . . . . . . . .
2
Characters and duality . . . . . . . . . . . . . . . . . .
3
Pairing of two groups . . . . . . . . . . . . . . . . . . .

209
209
213
217

Formally real fields


1
Ordered rings . . . . . . .
2
Extensions of orders . . .
3
Real closed fields . . . . .
4
Completion under an order
5
Archimedian ordered fields

.
.
.
.
.

.
.
.
.
.

. .
. .
. .
. .
. .
. .
p.
. .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.

Valuated fields
1
Valuations . . . . . . . . . . . . . . . . . . .
2
Classification of valuations . . . . . . . . . .
3
Examples . . . . . . . . . . . . . . . . . . .
4
Complete fields . . . . . . . . . . . . . . . .
5
Extension of the valuation of a complete. . . .
6
Fields complete under archimedian valuations
7
Extension of valuation of an incomplete field

.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.

Chapter 1

General extension fields


1 Extensions
A field has characteristic either zero or a prime number p.
1
Let K and k be two fields such that K k. We shall say that K
is an extension field of k and k a subfield of K. Any field T such that
K T k is called an intermediary field, intermediate between K and
k.
If K and K are two fields, then any homomorphism of K into K is
either trivial or it is an isomorphism. This stems from the fact that only
ideals in K are (o) and K. Let K have characteristic p , o. Then the
mapping a a p of K into itself is an isomorphism. For,
(a b) p = a p b p
(ab) p = a p b p

and a p = b p = (a b) p = o = a = b. In fact for any integer e 1,


e
a a p is also an isomorphism of K into itself.
Let now Z be the ring of rational integers and K a field whose unit
element we denote by e. The mapping m me of Z into K obviously a
homomorphism of the ring Z into K. The kernel of the homomorphism
is the set of m is Z such that me = 0 in K. This is an ideal in Z and as Z is
a principal ideal domain, this ideal is generated by integer say p. Now p
is either zero or else is a prime. In the first case it means that K contains
1

1. General extension fields

a subring isomorphic to Z and K has characteristic zero. Therefore K


contains a subfield isomorphic to the field of rational numbers. In the
second case K has characteristic p and since Z/(p) is a finite field of p
elements, K contains a subfield isomorphic to Z/(p). Hence the
Theorem 1. A field of characteristic zero has a subfield isomorphic to
the field of rational numbers and a field of characteristic p > o has a
subfield isomorphic to the finite of p residue classes of Z modulo p.
The rational number filed and the finite field of p elements are called
prime fields. We shall denote them by . When necessary we shall
denote the finite field of p elements by p.
Let K/k be an extension field of k. We shall identity the elements
of K and k and denote the common unit element by 1. Similarly for the
zero element. K has over k the structure of a vector space. For, , K,
k = + K, K. Therefore K has over k a base { } in
the sense that every K can be uniquely written in the form
X
=
a a k

and a = 0 for almost all . If the base { } consists only of a finite


number of elements we say that K has a finite base over k. The extension
K/k is called a finite or infinite extension of k according as K has over
k a finite or an infinite base. The number of basis elements we call the
degree of K over k and denote it by (K : k). If (K : k) = n then there
exist n elements , . . . n in K which are linearly independent over k
and every n + 1 elements of K linearly dependent over k.
Let K be a finite field of q elements. Obviously K has characteristic
p , o. Therefore K contains a subfield isomorphic to p . Call it also
p . K is a finite dimensional vector space over p . Let (K : p ) = n
Then obviously K has pn elements. Thus
Theorem 2. The number of elements q in a finite field is a power of the
characteristic.
Let K T k be a tower of fields. K/T has a base { } and T/K
has a base { }. This means that for K
X
=
t

2. Adjunctions

t T and t = 0 for almost all . Also t being in T we have


X
t =
a

a = 0 for almost all . Thus


=

a ( )

Thus every element of K can be expressed linearly in terms of


{ }. On the other hand let
X
( ) = 0

a k and a = 0 for almost all , . Then


XX
0=
( a )

But

T and since the { } from a base of K/T we have


X
a = 0 for all .

But { } form a base of T/k so that a = 0 for all , . We have


thus proved that { } is a base of K/k. In particular if (K : k) is finite
then (K : T ) and (T : k) are finite and
(K : k) = (K : T )(T : k)
As special cases, (K : k) = (T : k) = K = T (T is an intermediary
field of K and k). (K : k) = (K : T ) = T = k.

2 Adjunctions
Let K/k be an extension filed and K a family of intermediary extension
T
S
fields. Then K is again an intermediary field but, in general, K 4

1. General extension fields

is not a field. We shall define for any subset S of K/k the field k(S ) is
called the field generated byS over k. It is trivial to see that
k(S ) =

T S

i.e., it is the intersection of all intermediary fields T containing S . k(S )


is said to be got from k by adjunction of S to k. If S contains a finite
number of elements, the adjunction is said to be finite otherwise infinite.
In the former case k(S ) is said to be finitely generated over k. If (K :
k) < then obviously K is finitely generated over k but the converse is
not true.
Obviously k(S US ) = k(S )(S ) because a rational function of S US
is a rational function of S over k(S ).
Let K/k be an extension field and K. Consider the ring k[x] of
polynomials in x over k. For any f (x) k[x], f (x) is an element of K.
Consider the set G of polynomials f (x) k[x] for which f () = o. G
is obviously a prime ideal. There are now two possibilities, G = (0),
G , (o). In the former case the infinite set of elements 1, , 2 , . . .
are all linearly independent over k. We call such an element of K,
transcendental over k. In the second case G , (o) and so G is a principal
ideal generated by an irreducible polynomial (x). Thus 1, , 2 , . . . are
linearly dependent. We call an element of this type algebraic over k.
We make therefore the

Definition. Let K/k be an extension field. K is said to be algebraic


over k if is root of a non zero polynomial in k[x]. Otherwise it is said
to be transcendental.
5

If is algebraic, the ideal G defined above is called the ideal of


over k and the irreducible polynomial (x) which is a generator of G
is called the irreducible polynomial of over k. (x) may be made by
multiplying by a suitable element of k. This monic polynomial we shall
call the minimum polynomial of .

3. Algebraic extensions

3 Algebraic extensions
Suppose K is algebraic over k and (x) its minimum polynomial
over k. Let f (x) k[x] and f () , o. f (x) and (x) are then coprime
and so there exist polynomials g(x), h(x) in k[x] such that
f (x)g(x) = 1 + (x)h(x)
which means that ( f ())1 = g() k[]. Thus k[] = k(). On the
other hand suppose K such that k[] = k() then there is a g() in
k[] such that g() = 1 or that satisfies xg(x) 1 in k[x] so that is
algebraic. Hence
1) K algebraic over k k[] is a field.
We now define an extension K/k to be algebraic over k if every of K
is algebraic over k. In the contrary case K is said to be transcendental
extension of k
We deduce immediately
2) K/k algebraic every ring R with k R K is a field

If R is a ring and in R then k[] R then k[] R. But is


algebraic so that 1 k[] R so that R is a field. The converse
follows from (1).

3) (K : k) < = K/k algebraic.


For let (K : k) = n then for any for K, the n + 1 elements
1, , 2 , . . . n are linearly dependant over k so that is algebraic.
6
The converse is not true
Let K/k be an extension field and K algebraic over k. Let (x)
be the minimum polynomial of over k and let degree of (x) be n.
Then 1, , 2 , . . . , n1 are linearly independent over k so that
(k() : k) n.
On the other hand any in k() is a polynomial in over k. Let
= bo + b1 + + bm m . Put (x) = bo + b1 x + + bm xm .

1. General extension fields

6
Then

(x) = (x)h(x) + R(x)


where R(x) = 0 or deg R(x) < n. Hence () = = R() and so
every cab be expressed linearly in terms of 1, , . . . , n1 .
Thus

We have hence


k() : k n.

4) If K is algebraic over K, k()/k is an algebraic extension of


degree equal to the degree of the minimum polynomial of over k.


We shall call k() : k the degree of over k

5) If is algebraic over k then for any L, k L K, is algebraic over


L.

For, the ideal of over k (which is , (0) since is algebraic) is


contained in the ideal of in L[X] k[x].

Therefore



k() : k (L() : L)

Note that the converse is not true. For let z be transcendental



 over k
and consider the field k(z) of rational functions of z. k(z) : k is not


finite. But k(z) : k(z2 ) is finite as z is a root of x2 z2 over k(z2 ).

6) If 1 , . . . , n in K are algebraic over k then k(1 , . . . , n ) is algebraic


over k.
For, put Ko = k, Ki = k(1 , . . . , i ),
Kn = k(1 , . . . , n )
Then Ki /Ki1 is algebraic and is a finite extension. Now
(Kn : k) = i (Ki : Ki1 )
which is also finite. Hence Kn is algebraic over k.
We deduce immediately

3. Algebraic extensions

7) K/T algebraic, T/k algebraic = K/k algebraic.


For if K, is a root of (x) = X n + a1 xn+1 + + an in T [x].
Thus is algebraic over k(a1 . . . , an ). Hence

k(a1 , a2 , . . . an , ) : k(a1 , a2 , . . . , an ) < .

k(a1 , a2 , . . . , an ) : k <

k(a1 , a2 , . . . , an , ) : k <

which proves the contention.


If follows that if K/k is any extension, then the set L of elements
of K algebraic over k is a field L which is algebraic over k. L is called
the algebraic closure of k in K
We shall now show how it is possible to construct algebraic extensions of a field.
If k is a field and (x) a polynomial in k[x], an element of an
extension field K is said to be root of (x) if () = o. It then follows
that (x), has in K at most n roots, n being degree of (x).
8
Let f (x) be an irreducible polynomial in k[x]
The ideal generated by f (x) in k[x] is a maximal ideal since f (x) is
irreducible. Therefore the residue class ring K of k[x]/( f (x)) is a field.
Let denote the natural homomorphism of k[x] onto K. then maps k
onto a subfield of K. We shall identify this subfield with k itself (note
that k[x] and ( f (x)) are vector spaces over k). Let in K be the element
into which x goes by
= x
Then K = k(). In the first place k() K. Any element in K is the
image, by , of an element say (x) in k[x]. But
(x) = h(x) f (x) + (x)
So () K and () = (). But (x) above has degree degree
of f . Thus
k() K k[] k()
This shows that K = k() and that (K : k) is equal to the degree
of f (x). Also in K satisfies f () = o. We have thus proved that for

1. General extension fields

every irreducible polynomial f (x) in k[x] there exists an extension field


in which f (x) has a root.
Let now g(x) be any polynomial in k[x] and f (x) an irreducible factor of g(x) in k[x]. Let K be an extension of k in which f (x) has a root
. Let in K
g(x) = (x ) (x).

Then (x) k[x]. We again take an irreducible factor of (x) and


construct K in which (x) has a root. After finite number of steps we
arrive at a field L which is an extension of k and in which g(x) splits
completely into linear factors. Let 1 , . . . , n be the distinct roots of
g(x) in L. We call k(1 , . . . , n ) the splitting field of g(x) in L.
Obviously (k(1 , . . . , n ) : k) n!
We have therefore the important
Theorem 3. If k is a field and f (x) k[x] then f (x) has a splitting field
K and (K : k) n!, n being degree of f (x).
It must be noted however that a polynomial might have several splitting fields. For instance if D is the quaternion algebra over the rational
number field , generated by 1, i, j, k then (i), ( j), ( f ), (k) are all
splitting fields of x2 + 1 in [x]. These splitting fields are all distinct.
Suppose k and k are two fields which are isomorphic by means of
an isomorphism . Then can be extended into an isomorphism
of

k[x] on k [x] by the following prescription


X
X
(

ai xi ) =
(ai )xi ai k ai k

Let now f (x) be a polynomial in k[x] which is irreducible. Denote


by f (x), its image in k [x] by means of the isomorphism .
Then f (x)
is again irreducible in k [x]; for if not one can by means of
1 obtain a
nontrivial factorization of f (x) in k[x].
Let now be a root of f (x) over k and a root of f (x) over k .
Then
k() k[x]/( f (x)), k () k [x]/( f (x))

Let be the natural homomorphism of k [x] on k [x]/( f (x)). Consider the mapping on k[x]. Since is an isomorphism, it follows

4. Algebraic Closure

that is a homomorphism of k[x] on k [x]/(F (x)). The kernel of the


homomorphism is the set of (x) in k[x] such that


(x) f (x) .
This set is precisely ( f (x)). Thus
k[x]/( f (x)) k [x]/( f (x))
By our identification, the above fields contain k and k respectively
as subfields so that there is an isomorphism of k() on k () and the
restriction of to k is .
In particular if k = k , then k() and k() are k isomorphic i.e., they
are isomorphic by means of an isomorphism which is identity on k. We
have therefore
Theorem 4. If f (x) k[x] is irreducible and and are two roots of
it (either in the same extension field of k or in different extension fields),
k() and k() are k isomorphic.
Note that the above theorem is false if f (x) is not irreducible in k[x].

4 Algebraic Closure
We have proved that every polynomial over k has a splitting field. For a
given polynomial this field might very well coincide with k itself. Suppose k has the property that every polynomial in k has a root in k. Then
it follows that the only irreducible polynomials over k are linear polynomials. We make now the
Definition. A field is algebraically closed if the only irreducible polynomials in [x] are linear polynomials.
We had already defined the algebraic closure of a field k contained
in a field K. Let us now make the
Definition. A field /k is said to be an algebraic closure of k if

10

1. General extension fields

10
1) is algebraically closed
2) /k is algebraic.

11

We now prove the important


Theorem 5. Every field k admits, upto k-isomorphism, one and only
one algebraic closure.
Proof. 1) Existence. Let M be the family of algebraic extensions K
of k. Partially order M by inclusion. Let {K } be a totally ordered
S
subfamily of M. Put K = K for K in this totally ordered family.

Now K is a field; for 1 K, 2 K means 1 K for some and


2 K for some . Therefore 1 , 2 in K or K whichever is larger
so that 1 + 2 K. Similarly 1 1
2 K. Now K/k is algebraic
since every K is in some K and so algebraic over k. Thus
K M and so we can apply Zorns lemma. This proves that M has a
maximal element . is algebraically closed; for if not let f (x) be an
irreducible polynomial in [x] and a root of f (x) in an extension
() of . Then since /k is algebraic. () is an element of M.
This contradicts maximality of . Thus is an algebraic closure of
k.

2) Uniqueness. Let k and k be two fields which are isomorphic by


means of an isomorphism . Consider the family M of triplets
{(K, K , ) } with the property 1) K is an algebraic extension of
k, K of k , 2) is an isomorphism of K on K extending . By
theorem 4, M is not empty. We partially order M in the following
manner

(K, K , ) < (K, K , )
If 1) K K , K K , 2) is an extension of . Let {K, K , ) }
S
S
be a simply ordered subfamily. Put K = K , K = K ,

12

These are then algebraic over k and k respectively.


Define
on K by

4. Algebraic Closure

11
x
= x

where x K . (Note that every x K is in some K in the simply


ordered subfamily). It is easy to see that
is well - defined. Suppose
x K and K K then is an extension of and so x = x.
This proves that
is an isomorphism of K on K and extends . Thus

the triplet (K, K , )


is in M and is an upper bound of the subfamily.
By Zorns lemma there exists a maximal triplet (, , ). We assert
that is algebraically closed; for if not let be a root of an irreducible
polynomial f (x) [x]. Then f (x) [x] is also irreducible. Let
be a root of f (x). Then can be extended to an isomorphism of ()
on (). Now ((), is in M and hence leads to a contradiction. Thus
is an algebraic closure of k, of k and an isomorphism of on
extending .
In particular if k = k and the identity isomorphism, then and

are two algebraic closures of k and is then a k-isomorphism.


Out theorem is completely demonstrated.
Let f (X) be a polynomial in k[x] and K = k(1 , . . . , n , a splitting
field of f (x), so that 1 , . . . , n are the distinct roots of f (x) in K. Let K
be any other splitting field and 1 , . . . m the distinct roots of f (x) in K .
Let be an algebraic closure of K and of K . Then and are two
algebraic closures of k. There exists therefore an isomorphism of
on which is identity on k. Let K = K1 . Then K1 = k(1 , . . . , n ).
Since 1 , . . . n are distinct 1 . . . , n are distinct and are roots of
f (x). Thus K1 is a splitting field of f (x) in . This proves that
13
K = K1 .
1 , . . . , m are distinct and are roots of f (x) in . We have m = n
and i = 2 in some order. Therefore the restriction of to K is an
isomorphism of K on K . We have
Theorem 6. Any two splitting fields K, K of a polynomial f (x) in k[x]
are k isomorphic.
Let K be a finite field of q elements. Then q = Pn where n is an
integer 1 and p is the characteristic of K. Also n = (K : ), being

1. General extension fields

12

the prime field. Let K denote the abelian group of non-zero elements
of K. Then K being a finite group of order q 1,
q1 = 1
for all K . Thus K is the splitting field of the polynomial
xq x
in [x]. It therefore follows
Theorem 7. Any two finite fields with the same number of elements are
isomorphic.
A finite field cannot be algebraically closed; for if K is a finite field
of q elements and a K the polynomial
Y
f (x) = x
(X b) + a
bK

is in K[x] and has no root in K.

5 Transcendental extensions

14

We had already defined a transcendental extension as one which contains at least on transcendental element.
Let K/k be a transcendental extension and Z1 , . . . , Zn any n elements
of K. Consider the ring R = k[x1 , . . . xn ] of polynomials over k in n
variables. Let Y be the subset of R consisting of those polynomials
f (x1 , . . . xn ) for which
f (Z1 , . . . Zn ) = 0.
Y is obviously an ideal of R. If Y = (o) we say that Z1 , . . . Zn are
algebraically independent over k. If Y , (o), they are said to be
algebraically dependent. Any element of K which is algebraic over
k(Z1 , . . . , Zn ) is therefore algebraically dependent on Z1 , . . . Zn .
We now define a subset S of K to be algebraically independent over
k if every finite subset of S is algebraically independent over k. If K/k
is transcendental there is at least one such non empty set S .

5. Transcendental extensions

13

Let K/k be a transcendental extension and let S , S be two subset of


K with the properties
i) S algebraically independent over k
ii) S algebraically independent over k(s)
Then S and S are disjoint subsets of K and S US are algebraically
independent over k. That S and S are disjoint is trivially seen. Let now
Z1 , . . . Zm S and Z1 , . . . Zn S be algebraically dependent. This will
mean that there is a polynomial f ,
f = f (x1 , . . . , xm+n )
in m + n variables with coefficients in k, such that
f (Z1 , . . . , Zm , Z1 , . . . Zn ) = 0.
Now f can be regarded as a polynomial in xm+1 , . . . , xm+n with coefficients in k(x1 , . . . , xm ). If all these coefficients are zero then Z1 , . . . Zm ,
Z1 , . . . Zn are algebraically independent over k. If some coefficient is
, 0, then f (Z1 , . . . Zm , xm+1 , . . . , xm+n ) is a non zero polynomial over
k(S ) which vanishes for xm+1 = Z1 , . . . xm+n = Zn which contradicts the
fact that S is algebraically independent over k(S ). Thus f = o identically and our contention is proved.
15
The converse of the above statement is easily proved.
An extension field K/k is said to be generated by a subset M of K if
K/k(M) is algebraic. Obviously K itself is a set of generators. A subset
B of K is said to be a transcendence base of K if
1) B is a set of generators of K/k
2) B algebraically independent over k.
If K/k is transcendental, then, it contains algebraically independent
elements. We shall prove that K has a transcendence base. Actually
much more can be proved as in

1. General extension fields

14

Theorem 8. Let K/k be a transcendental extension generated by S and


A a set of algebraically independent elements contained in S . Then
there is a transcendence base B of K with
ABS
Proof. Since S is a set of generators of K, K/k(S ) is algebraic. Let M
be the family of subsets A of K with
1) A A S
2) A algebraically independent over k.

The set M is not empty since A is in M. Partially order M by inS
clusion. Let {A } be a totally ordered subfamily. Put B0 = A . Then

16

Bo S . Any finite subset of Bo will be in some A for large and so


Bo satisfies 2) also. Thus using Zorns lemma there exists a maximal
element B in M. Every element x of S depends algebraically on B for
otherwise BU x will be in M and will be larger than B. Thus k(S )/k(B)
is algebraic. Since K/k(S ) is algebraic, it follows that B satisfies the
conditions of the theorem.
The importance of the theorem is two fold; firstly that every set of
elements algebraically independent can be completed into a transcendence base of K and further more every set of generators contains a
base.
We make the following simple observation. Let K/k be an extension,
Z1 , . . . Zm , m elements of K which have the property that K/k(Z1 , . . . Zm )
is algebraic, i.e., that Z1 , . . . , Zm is a set of generators. If Z K then Z
depends algebraically on Z1 , . . . , Zm i.e., k(Z, Z1 , . . . , Zm )/k(Z1 , . . . , Zm )
is algebraic. We may also remark that if in the algebraic relation connecting Z, Z1, . . . Zm , Z1 occurs then we can say that
k(Z, Z1 , . . . Zm )/k(Z, Z2 , . . . , Zm )
is algebraic which means that Z, Z2 , . . . Zm is again a set of generators.
We now prove the

5. Transcendental extensions

15

Theorem 9. If K/k has a transcendence base consisting of a finite number n of elements, every transcendence base has n elements.
Proof. Let Z1 , . . . , Zn and Z1 , . . . , Zm be two transcendence bases consisting of n and m elements respectively. If n , m let n < m. Now
K/k(Z1 , . . . , Zn ) is algebraic. Z1 is transcendental over k and depends algebraically on Z1 , . . . , Zn so that if Z1 appears in the algebraic relation,
by the remark above, Z1 , Z2 , . . . , Zn is again a set of generators, Z2 depends algebraically on Z1 , . . . , Zn . In this algebraic relation at least one
of Z2 , . . . , Zn has to appear since Z1 , Z2 , are algebraically independent.
If Z2 appears then Z2 , Z1 , Z3 , . . . , Zn is a set of generators. We repeat this
process n times, and find, that Z1 , Z2 , Z3 , . . . , Zn is a set of generators
, . . . , Z depend algebraically on Z , . . . , Z . This 17
which means that Zn+1
m
n
1
is a contradiction. So n m. We interchange n and m and repeat the
argument and get n m. This proves that n = m.
The unique integer n will be called the dimension of K/k.
n = dimk K

It is also called the transcendence degree.
A similar theorem is true even if K has infinite transcendence base
but we dont prove it.
Let k L K be a tower of extensions and let B1 be a transcendence base of L/k and B2 that over K/L. We assert that B1 U B2 is a
transcendence base of K/k. In the first place B1 U B2 is algebraically independent over k. Now k(B1 U B2 ) is a subfield of L(B2 ). Every element
in L(B2 ) is a ratio of two polynomials in B2 with coefficients in L. The
elements of L are algebraic over K(B1 ). Thus L(B2 ) is algebraic over
k(B1 U B2 ). But K/L(B2 ) is algebraic. Thus K/k(B1 U B2 ) is algebraic.
This proves our assertion. In particular it proves
Theorem 10. If k L K then
dimk K = dimk L + dim L K.

16

18

1. General extension fields

A transcendental extension K/k is said to be purely transcendental


if there exists a base B with K = k(B). Note that this does not mean that
every base has this property. For instance if k(x) is the field of rational
functions of x then x2 is also transcendental over x but k(x2 ) is a proper
subfield of k(x).
Let K = k(x1 , . . . xn ) and K = k(x1 , . . . xn ) be two purely transcendental extensions of dimension n. Consider the homomorphism defined by
f (x1 , . . . xn ) = f (x1 , . . . , xn )
Where f (x1 , . . . , xn ) k[x1 , . . . , xn ]. It is then easy to see that this is
an isomorphism of K on K . This proves
Theorem 11. Two purely transcendental extensions of the same dimension n over k k-isomorphic.
This theorem is true even if the dimension is infinite.

Chapter 2

Algebraic extension fields


1 Conjugate elements
Let be an algebraic closure of k and K an intermediary field. Let 19
be an algebraic closure of K and so of k. Then there is an isomorphism
of on which is trivial on k. The restriction of this isomorphism to
K gives a field K in which is k-isomorphic to K. Conversely suppose
K and K are two subfields of which are k-isomorphic. Since is a
common algebraic closure of K and K , there exists an automorphism
of which extends the k-isomorphism of K and K . Thus
1) Two subfields K, K of /k are k-isomorphic if and only if there exists
a k-automorphism of such that K = K .
We call two such fields K and K conjugate fields over k.
We define two elements , of /k, to be conjugate over k if there
exists a k-automorphism of such that
=
The automorphisms of which are trivial on k form a group and so
the above relation of conjugacy is an equivalence relation. We can
therefore put elements of into classes of conjugate elements over
k. We then have
17

2. Algebraic extension fields

18

2) Each class of conjugate elements over k contains only a finite number


of elements.
20

Proof. Let C be a class of conjugate elements and C. Let f (x)


be the minimum polynomial of in k[x]. Let be an automorphism
of /k. Then C. But is a root of f (x) = f (x). Also if
C then = for some automorphism of /k. In that case
= is again a root of f (x). Thus the elements in C are all roots of
the irreducible polynomial f (x). Our contention follows.

Notice that if , are any two roots, lying in , of the irreducible
polynomial f (x), then k() and k() are k-isomorphic. This isomorphism can be extended into an automorphism of . Thus
Theorem 1. To each class of conjugate elements of there is associated an irreducible polynomial in k[x] whose distinct roots are all the
elements of this class.
If we shall denote by C the class of . C is a finite set.

2 Normal extensions
Suppose K is a subfield of /k and an automorphism of /k. Let
K K. We assert that K = K. For let K and denote by C the
set
C K
Since K K we have K so C . Thus
C
C is a finite set and is an isomorphism of K into itself.
Thus
C = C
21

which means K. Thus K = K.


We shall now study a class of fields K /k which have the property
K K,

2. Normal extensions

19

for all automorphisms of /k. We shall call such fields, normal extensions of k in .
Let K/k be a normal extension of k and algebraic closure of k
containing K. Let K and C the class of . We assert that C K.
For if is an element of C , there is an automorphism of /k for
which = . Since K K, it follows that K. Now any element
in K is a root of an irreducible polynomial in k[x]. Since all the elements
of C are roots of this polynomial, it follows that if f (x) is an irreducible
polynomial with one root in K, then all roots of f (x) lie in K.
Conversely let K be a subfield of /k with this property. Let be
an automorphism of /k and /K. Let be an automorphism of /k
and K. Let C be the class of . Since C K, K. But is
arbitrary in K. Therefore
K K
and K is normal. Thus the
Theorem 2. Let k K . Then K = K for all automorphisms of
/k every irreducible polynomial f (x) k[x] which has one root
in K has all roots in K.
Let f (x) be a polynomial in k[x] and K its splitting field. Let be
an algebraic closure of K. Let 1 , . . . , n be the distinct roots of f (x) in 22
. Then
K = k(1 , . . . , n )
Let be an automorphism of /k. j = j for some j. Thus
takes the set 1 , . . . , n onto itself. Since every element of K is a rational
function of 1 , . . . , n , it follows that K K. Thus
i) The splitting field of a polynomial in k[x] is a normal extension of
k.
T
Let {K } be a family of normal subfields of /k. Then K is

S
trivially normal. Consider k( K ). This again is normal since for

any automorphism of /k.

[
[

[
k K k K k K

2. Algebraic extension fields

20

Let now { f (x)} be a set of polynomials in k[x] and K their splitting


S
fields, then K( K ) is normal. Also it is easy to see that

[
L = k K

is the intersection of all subfields of /k in which every one of the


polynomials f (x) splits completely. Thus
ii) if { f (x)} is a set of polynomials in k[x], the subfield of generated
by all the roots of { f (x)} is normal.
We also have
iii) If K/k is normal and k L K then K/L is also normal.
23

For if is an L- automorphism of , then is also a k-automorphism of and so K K.


The kautomorphisms of form a group G(/k). From what we
have seen above, it follows that a subfield K of /k is normal if and
only if K = K for every G(/k). Now a k automorphism
of K can be be extended into an automorphism of /k, because
every such automorphism is an isomorphism of K in . It therefore
follows
iv) K/k is normal if and only if every isomorphism of K in /k is an
automorphism of K over k.
As an example, let be the field of rationalnumbers and f (x) =
3
x3 2. Then f (x) is irreducible in [x]. Let = 2 be one of its roots.
() is of degree 3 over and is not normal since it does not contain
where = 1+2 3 . However the field (, ) of degree 6 over is
normal and is the splitting field of x3 2.
If K is the field of complex numbers, consider K(z) the field of rational function of 2 over K. Consider the polynomial x3 z in K(z)[x]. This
1
is irreducible. Let = z 3 be a root of this polynomial. Then K(z)() is
of degree 3 K(z) and is the splitting field of the polynomial x3 z.

3. Isomorphisms of fields

21

3 Isomorphisms of fields
Let K/k be an algebraic extension of k and W any extension of K and so
of k. A mapping of K into W is said to be k linear if for , K
24
( + ) = +
W and if k, () = . If is a klinear map of K into
W we define for in W by
() =
for K. This again is a klinear map and so the klinear maps of K
into W form a vector space V over W.
A kisomorphism of K into W is obviously a klinear map and
so V. We shall say, two isomorphisms , of K into W (trivial on
k) are distinct if there exists at least one K, , 0 such that
,
Let S be the set of mutually distinct isomorphisms of K into W. We
then have
Theorem 3. S is a set of linearly independent elements of V over W.
Proof. We have naturally to show that every finite subset of S is linearly
independent over W. Let on the contrary 1 , . . . , n be a finite subset of
S satisfying a non trivial linear relation
X
i i = 0
i

i W. We may clearly assume that no proper subset of 1 , . . . , n


is linearly dependent. Then in the above expression all i are different
from zero. Let be any element of K. Then
X
i i = 0
i

2. Algebraic extension fields

22

If we replace by we get, since i s are isomorphisms,


X
i i .i = o

25

for every K. This means that i , . . . , n satisfy another linear relation


X
i i .i = o
i

Since the isomorphisms are mutually distinct, we can choose in


K in such a way that
1 , n
We then get from the two linear relations, the expression
n1 
X
i
i=1

i i 
i = o.
n i

This relation is non trivial since the coefficient of 1 is different


from zero. This leads to a contradiction and our theorem is proved.
Suppose dim V < then it would mean that S is a finite set. But
the converse is false. We have however the
Theorem 4. If (K : k) < , then dim V = (K : k)
Proof. Let (K : k) = n and 1 , . . . , n a basis of K/k.

Consider the k-linear mappings 1 , . . . , n defined by

i, j
i ( j ) = o

=1
i, j

Then 1 , . . . , n are linearly independent elements of V over W.


P
For, let i i = o, i W. Then
i

X
i

26


i i i = o

for j = 1, . . . , n. This proves that j = o. Now let be any k-linear

3. Isomorphisms of fields

23

mapping. It is uniquely determined by its effects on 1 , . . . , n . Put


i = i and let be given by
X
=
i i
i

Then ( j ) = j i i i ( j ) = o so that = o. Our contention is


established.
From this we obtain the very important
Corollary . If (K : k) < then K has in /k at most (K : k) distinct
k-isomorphisms.
Let . Consider the field k()/k. Let (1) (= ), . . . , (n) be the
distinct conjugates of over k. An isomorphism of k()/k is determined completely by its effect on . Since every isomorphism comes
from an automorphism of /k, it follows that k((i) ) are all the distinct
isomorphic images of k(). Thus
1) Number of distinct k-isomorphisms of k() in is equal to the
number of distinct roots in of the minimum polynomial of .
Let K/k be an algebraic extension and an algebraic closure of k
containing K. Let K have the property that K/k has only finitely many
distinct k-isomorphisms in . Let K (1) (= K), K (2) , . . . , K (n) be the distinct isomorphic fields. Let and let have over K exactly m
distinct conjugates (1) (= ), . . . , (m) . This means that if f (x) is the
minimum polynomial of over K, then f (x) has in , m distinct roots.
We claim that K() has over k exactly mn distinct isomorphisms in . 27
For, let i (i = 1, . . . , n) be the k-isomorphisms defined by
i K (1) = K (i)
Let f i (x) be the image of the polynomial f (x) in K[x] by means
of the above isomorphism. Let the roots of f i (x) by (i1 ) , . . . , (in )
these being the distinct ones. There exists then an isomorphism i j ( j =
1, . . . , m) extending i of K (1) ((1) ) on K (i) ((i j) ). Since i has n values,
it follows that there are at least mn distinct isomorphisms of K() over
k.

24

28

2. Algebraic extension fields

Let now be any automorphism of /k. Let K = K (i) . Then it


takes (1) into a root (i j) of f (x) = f i (x) where i is the isomorphism which coincides with on K. Thus since every isomorphism
of K() over k comes from an automorphism of /k, our contention is
established.
Let now K = k(1 , . . . , n ) and Ki = k(1 , . . . , i ) so that Ko = k
and Kn = K. Let Ki have over Ki1 exactly Pi distinct Ki1 -isomorphisms. Then K/k has exactly p1 pn distinct k-isomorphisms in .
Hence
2) If K L k be a tower of finite extensions and K has ever L,
n distinct L-isomorphisms in and L has over k, m distinct k-isomorphisms then K has over k precisely mn distinct k-isomorphisms.
In particular let (K : k) < and let K have in exactly (K : k)
distinct isomorphisms. Let L be any intermediary field. Let a be the
number of distinct L-isomorphisms of K and b the number of distinct
k-isomorphisms of L.
Then
(K : k) = ab (K : L)(L : k) = (K : k)
But a (K : L), b (L : k). Thus a = (K : L) and b = (L : k).

4 Separability
Let be an algebraic closure of k and . Let (x) be the minimum
polynomial of in k. Suppose k()/k has exactly (k() : k) distinct
k-isomorphisms in . Then from the last article it follows that all the
roots of (x) are distinct. Conversely let the irreducible polynomial (x)
be of degree n and all its n roots distinct. Then k()/k has n distinct
k-isomorphisms being a root of (x). But it can have no more.
Let us therefore make the
Definition. An element is said to be separably algebraic or separable over k if its minimum polynomial has all roots distinct. Otherwise
it is said to be inseparable.
1) Let W/k be any extension field and W separable over k. Let L
be an intermediary field. Then is separable over L.

4. Separability

25

For, the minimum polynomial of over L divides that over k.


2) separable over k k()/k has in (k() : k) distinct k- 29
isomorphisms.
Let now K = k(1 , . . . , n ) and let 1 , . . . n be all separable over k.
Put Ki = k(1 , . . . , ) so that Ko = k and Kn = K. Now Ki1 (i ) and
i is separable over Ki1 so that Ki over Ki1 has exactly (Ki : Ki1 )
distinct Ki1 - isomorphisms. This proves that K has over k
(Kn : Kn1 ) . . . (K1 : Ko ) = (Kn : Ko ) = (K : k)
distinct k-isomorphisms. If therefore K, Then by previous article k() has exactly (k() : k) distinct isomorphisms and hence
is separable over k. Conversely if K/k is finite and every element
of K is separable over k, then K/k has exactly (K : k) distinct kisomorphisms. Hence
3) (K : k) < , K/k has (K : k) distinct k-isomorphisms every
element of K is separable over k.
Let us now make the
Definition. A subfield K of /k is said to be separable over k if every
element of K is separable over k.
From 3) and the definition, it follows that
4) K/k is separable for every subfield L of K with (L : K) < , L
has exactly (L : K) distinct isomorphisms over k.
5) K/L, L/k separably algebraic K/k separable.
For, let K. Then is separable over L. Let 1 , . . . , n be
the coefficients in the irreducible polynomial satisfied by over L. 30
Then has over K1 = k(1 , . . . , n ) exactly (K1 () : K1 ) distinct
K1 -isomorphisms. Also K1 /k is finite separable. Thus K1 () has
over k exactly (K1 () : k) distinct isomorphisms which proves that
is separable over k. The converse follows from 2).

2. Algebraic extension fields

26

6) If {K } is a family of separable subfields of then


S
T
2) K and b)k( K ) are separable.

a) follows easily because every element of K is separable over k.


S
b) follows since every element of k( K ) is a rational function of a

finite number of elements and as each of these is separable the result


follows from 3).

7) Let K/k be any extension-not necessarily algebraic. The set L of


elements of k separably algebraic over K is a field.
This is evident. We call L the separable closure of k in K.
We had already defined an algebraic element to be inseparable if
its minimum polynomial has repeated roots. Let us study the nature
of irreducible polynomials.
Let f (x) = ao + a1 x + + an xn be an irreducible polynomial in k[x].
If it has a root which is repeated, then is a root of
f 1 (x) = a1 + 2a2 x + + nan xn1 .
Thus f (x)| f 1 (x) which can happen only if
iai = o, i = 1, . . . , n.
31

Let k have characteristic zero. Then iai = o ai = 0 that is f (x) is


a constant polynomial. Thus
8) Over a field of characteristic zero, every non constant irreducible
polynomial has all roots distinct.
Let now k have characteristic p , o. if pi then iai = o ai = o.
Thus for f 1 (x) to be identically zero we must have ai = o for pi. In
this case
f (x) = ao + a p x p +
or that f (x) k[x p ]. Let e be the largest integer such that f (x)
e
e+1
k[x p ] but not in k[x p ]. Consider the polynomial (y) with (x p ) =

4. Separability

27

f (x). Then (y) is irreducible in k[y] and (y) has no repeated roots.
Let 1 , . . . , t be the roots of (y) in . Then
e

f (x) = (x p 1 ) (x p t ).
e

Thus n = t pe . The polynomial xP i has in all roots identical


to one of them say i . Then
e

pe

xP i = xP i = (x i ) p
Thus
e

f (x) = {(x 1 ) (x t )} p

Moreover since 1 . . . t are distinct, 1 , . . . , t are also distinct.


Hence
9) Over a field of characteristic p , o, the roots of an irreducible polynomial are repeated equally often, the multiplicity of a root being pe ,
e o.
It is important to note that (x 1 ) . . . (x t ) is not a polynomial in 32
k[x] and t is not necessarily prime to p.
We call t the reduced degree of f (x) (or of any of its roots ) and pe ,
its degree of inseparability. Thus
Degree of: Reduced degree X-degree of inseparability
If then we had seen earlier that k()/k has as many distinct
isomorphisms in as there are distinct roots in of the minimum
k()
polynomial of over k. If we call the reduced degree of
as the
k
reduced degree of we have
10) Reduced degree of = Number of distinct roots of the minimum
polynomial of over k.
We may now call a polynomial separable if and only if every root of
it in is separable. In particular if f (x) k[x] is irreducible then
f (x) is separable if one root of it is separable.

2. Algebraic extension fields

28

Let and f (x) the minimum polynomial of in k[x]. If t =


reduced degree of , then
e

f (x) = {(x 1 ) . . . (x t )} p
e

n = t pe . Let 1 = . Consider 1p = 1 . Then


e

f (x) = (x p 1 ) (x p t )
and 1 , . . . t are separable over k. Consider the field k(1 ) which is
a subfield of k(). being of degree t over k, (k(1 ) : k) = t. This
means that
(k() : k()) = pe .

33

But the interesting fact to note is that k() has over k() only the
identity isomorphism or that k() is fixed by every k()-automorphism of /k().
Also since every element of k() is a rational function of over k(),
it follows that
e
p k()
for every k(). Thus the integer e has the property that for every
e
k(), p k() and there is at least one (for instance ) for
e
which p < k(). e is called the exponent of , equivalently of k().
We define the exponent of an algebraic element over k to be the
e1
e
integer e o such that p is separable but not p . Hence
11) Exponent of is zero is separable over k.
We shall now extend this notion of exponent and reduced degree to
any finite extension.
Let K/k be finite so that K = k(1 , . . . , n ). Put as before Ko = k,
Ki = k(1 , . . . , i ) so that Kn = K. Let i have reduced degree di and
exponent ei over Ki1 . Then
(Ki : Ki1 ) = di pei

4. Separability

29

From the definition of di , it follows that the number of distinct kisomorphisms of K/k is d1 . . . dn . We put
d = d1 dn
and call it the reduced degree of K/k. Then
(K : k) = d. . . . p f
where f = e1 + + en . We call p f the degree of inseparability of K/k.
In order to be able to give another interpretation to the integer d we 34
make the following considerations.
Let K k and let K/k have the property that every k automorphism of /k acts like identity on K. Thus if G(/k) and K,
then
=
All elements of k have this property. Let K, < k. Then by
definition, has in only one conjugate. The irreducible polynomial
of has all roots equal. Thus the minimum polynomial of is
m

xp a
m

where a k. i.e., p k. On the other hand let K be an extension of k


in with the property that for every K
m

p k
for some integer m o. Let be an automorphism of /k. Then
m
m
p = p . But
m

o = p p = ( ) p

which shows that = . being arbitrary in K, it follows that every


element of G(/k) is identity on K.
Hence for the following three statements are equivalent
1) = for all G(/k)

2. Algebraic extension fields

30
m

2) p k for some m o depending on


m

3) The irreducible polynomial of over k is of the form x p a, a k.


We call an element which satisfies any one of the above 35
conditions, a purely inseparable algebraic element over k.
Let us make the
Definition . A subfield K/k of /k is said to be purely inseparable if
element of K is purely inseparable.
From what we have seen above, it follows that K/k is purely inseparable is equivalent to the fact that every k-automorphism of is identity
on K.
Let K/k be an algebraic extension and L the maximal separable sube
field of K/k. For every K, p is separable for some e o which
e
means that p L. Thus K/L is a purely inseparable extension field.
This means that every k-isomorphism of L/k can be extended uniquely
to a k-isomorphism of K/k.
Let, in particular, K/k be a finite extension and L the maximal separable subfield of K. Then K/L is purely inseparable and K/L has no
L-isomorphism other than the identity. Thus the number of distinct isomorphism other than the identity. Thus the number of distinct isomorphisms of K/k equals (L : k). But from what has gone before
(K : L) = p f , (L : k) = d.
For this reason we shall call d also the degree of separability of K/k
and denote it by [K : k]. We shall denote the degree of inseparability,


p f , by K : k . Then


(K : k) = [K : k] K : k .

Note. In case k has characteristic zero, every algebraic element over k is


separable.
36

If is the algebraic closure of k and L the maximal separable subfield of then /L is purely inseparable. coincides with L in case k

5. Perfect fields

31

has characteristic zero. But it can happen that L is a proper subfield of


.
Let K/k be an algebraic extension and L the maximal separable subfield. Consider the exponents of all elements in K. Let e be the maximum of these if it exists. we call e the exponent of the extension K/k. It
can happen that e is finite but K/L is infinite.
If K/k is a finite extension then K/L has degree p f so that the maximum e of the exponents of elements of K exists. If e is the exponent of
K/k then
e f
It can happen that e < f . For instance let k have characteristic p , 0
and let k be not a pth power in k. Then k(1/p ) is of degree p over
k. Let in k be not a pth power in k. Then k(1/p , 1/p ) is of degree p2
over k, < k(1/p ) and for every k(1/p , 1/p ), p k.
We may for instance take k(x, y) to be the field of rational functions
of two variables and K = k(x1/p , y1/p ). Then (K : k(x, y)) = p2 and
p k(x, y) for every K.

5 Perfect fields
Let k be a field of characteristic p > 0. Let be its algebraic closure.
Let k. Then there is only one element such that p = .
1
We can therefore write 1/p without any ambiguity. Let k p be the field
generated in /k by the pth roots of all elements of k. Similarly from 37
2
k p , . . . Let
[ n
K=
kp
n0

Obviously K is a field; for if , K, , k p for some large n.

We denote K by k p

We shall study k p in relation to k and . k p is called the root


field of k.

Let k p . Then k p for some n so that p k or is


purely inseparable. On the other hand let be purely inseparable.

n
Then p k for some n i.e., k p k p . Thus

2. Algebraic extension fields

32

1) k p

is the largest purely inseparable subfield of /k.

Therefore every automorphism of /k is identity on k p . The set


of elements of which are fixed under all the k automorphisms of
/k form a field called the fixed of G(/k). Since every such ele
ment , fixed under G(/k) is purely inseparable, k p . Hence

2) k p

is the fixed field of the group of k automorphism of /k.

Let f (x) be an irreducible polynomial in k p [x]. We assert that this

is separable. For if not f (x) k p [x p ]. Thus f (x) = ao + a1 x p +


t

p
+ an xnp . Since ai k p [X p ], it is in some k p and so ai = bi for

bi k p . Hence
f (x) = (bo + b1 x + + bn xn ) p
which is contradicts the fact that f (x) is irreducible. Hence
38

3) /k p

is a separable extension.

We now make the


Definition. A field k is said to be perfect if every algebraic extension
of k is separable.
It follows from the definition that
1) An algebraically closed field is perfect
2) A field of characteristic zero is perfect.
We shall now prove

3) A field k of characteristic p > 0 is perfect if and only if k = k p .


Let k have no inseparable extension. Then for a k, a1/p k
1
also; for, otherwise k(a1/p ) is inseparable over k. Thus k = k p =

= k p . The converse has already been proved.


We deduce immediately
4) A finite field is perfect.
For if k is a finite field of characteristic p > r then a a p is an
automorphism of k.

5. Perfect fields

33

5) Any algebraic extension of a perfect field is perfect.


For let K/k be algebraic and k be perfect. If is inseparable over
K, then it is already so over k.
An example of an imperfect field is the field of rational functions
of one variable x over a finite field k. For if k has characteristic
p, then x1/p < k(x) and k(x1/p ) is a purely inseparable extension
over k(x).
Note 1. If is inseparable over k, it is not true that it is inseparable
over every intermediary field, whereas this is true if is separable.
39

Note 2. If K/k is algebraic and K k p contains k properly then K is an


inseparable extension. But the converse of this is not true, that is, if K/k
is an inseparable extension, it can happen that there are no elements in
K which are purely inseparable over k. We give to this end the following
example due to Bourbaki.
Let k be a field of characteristic p > 2 and let f (x) by in irreducible
polynomial
f (x) = xn + a1 xn1 + + an
in k[x]. If 1 , . . . , t are the distinct roots of f (x) in then
n
o pe
f (x) = (x 1 ) . . . (x t ) e 1.

where n = t. pe . Put (x) = f (x p ). Then

If i = 1/p
then
i


e
(x) = (x p 1 ) . . . (x p t )} p

e+1
(x) = (x 1 ) . . . (x t ) p

and , . . . , t are distinct since 1 . . . t are distinct. Suppose (x) is


reducible in k[x] and let (x) be an irreducible factor of (x) in k[x].
Then
n
o p
(x) = (x 1 ) (x t )

2. Algebraic extension fields

34

for 0 and t. (This is because roots of (x) occur with the same
multiplicity). We can write

1
(x) = (x p 1 ) . . . (x p ) p

( has to be 1) since otherwise, it will mean that (x 1 ) (x t )


k[x]). Now this will mean that (x) = (x). W(x) in k[x p ] so that = t.
Hence

(x) = {(x 1 ) (x t )} p , 1.

40

Since (x) is irreducible and


1

(x) = {(x p 1 ) . . . (x p t )} p
we see that 1 = e or = e + 1. Thus
(x) = f (x p ) = {(x)} p

Thus if f (x p ) is reducible, it is the pth power of an irreducible polyp


nomial. In this case ai = bi , bi k, i = 1, . . . n.
p
Conversely if ai = bi , bi k, i = 1, . . . n. then f (x p ) is reducible.
Hence f (x p ) is reducible f (x) k p [x].
Let k now be a field of characteristic p > 2 given by k = (x, y), the
field of rational functions in two variables x, y over the prime field if
p elements. Consider the polynomial
f (z) = z2p + xz p + y,
in k[z]. Since x1/p , y1/p do not lie in k, f (z) is irreducible in k. Let be
a root of f (z). Then (k() : k) = 2p. Let be in k() and not in k such
that p k. Then k() k() k. Also (k() : k) = p. In k()[x]

the polynomial f (z) cannot be irreducible, since k() : k() = 2. It is
reducible and so k() will contain x1/p and y1/p . But then k(x1/p , y1/p )
k().
Thus



p2 = k(x1/p , y1/p ) : k k() : k ) : k = 2p

but this is impossible. Thus there is no element in k() with p k.


All the same k() is inseparable.

6. Simple extensions

35

6) If k is not perfect then k p


For, if k

is an infinite extension of k.
S pn
n

we have k p =
k
/k is finite then, since k p =
n

k p(n+1) for some n,


1
pn
Applying the mapping a aa we find k = k p . But this is false. 41
Thus

 (n+1)
n
: kp > 1
kp

which proves our contention.

6 Simple extensions
An algebraic extension K/k is said to be simple if there is an K such
that K = k(). Obviously (K : k) is finite. We call a primitive element
of K. The primitive element is not unique for, + , k also is
primitive. We now wish to find conditions when an algebraic extension
would be simple. We first prove
Lemma. Let k be an infinite field and , elements in an algebraic closure of k such that is separable over k. Then k(, ) is a simple
extension of k.
Proof. Let f (x) and (x) be the irreducible polynomials of and respectively in k[x], so that
f (x) = (x 1 ) (x n )

(x) = (x 1 ) (x m ).

Since is separable, 1 , . . . n are all distinct. We shall put = 1 ,
and = 1 . Construct the linear polynomials
i + X j (i = 1, . . . , m; j = 1, . . . , n)
These mn polynomials are in , the algebraic closure of k and since
k is infinite, there exists an element k such that
i + j , i , + j , , J

2. Algebraic extension fields

36
Put

= 1 + 1 = +
42

Obviously can be chosen so that , 0


Now k() k(, ). We shall will now show that k(). From
definition of , also will be in k() and that will mean
k() k(, ) k().
In order to do this consider the polynomial (.x)in k()[x]. Also
it vanishes for x = . Furthermore by our choice of , ( i ) , 0
for i > 1. In the algebraic closure , therefore, f (x) and ( x) have
just x as a factor. But f (x) and ( x) are both polynomials in
k()[x]. So x is the greatest common divisor of ( x) and f (x)
in k()[x]. Thus k(). Our lemma is demonstrated.
We have therefore
Corollary . If 1 , . . . , n are separably algebraic and then
k(, , . . . n ) is a simple extension.
We deduce immediately
Corollary. A finite separable extension is simple.
Let be the field of rational numbers and (, ) the splitting field
of the polynomial x3 2 in [x]. Then (, ) is simple. A primitive
element is given by + = . Then () is of degree 6 over . It is
easy to see that has over the minimum polynomial
(x3 3x 3)2 + 3x(x + 1)(x3 3x 3) + 9x2 (x + 1)2
Let now K/k be a finite extension and L the maximal separable
subfield of K/k. Then (K : L) = p f the degree of inseparability and
(L : k) = d the reduced degree. If we consider the exponents of elements of K, these have a maximum e and
e f.

43

We had given an example of e < f . We shall now prove the

6. Simple extensions

37

Theorem 5. If e is the exponent and p f the degree of inseparability of


finite extension K of k, then e = f K/k is simple.
Proof. Let K = k(). Let Ko be the maximal separable subfield of K/k.
Now (K : Ko ) = p f . But p f is degree of . Thus e = f

Let now K/k be a finite extension and e = f . There exists then a
t
e
in K such that p is separable and in Ko but for no, t < e p is in Ko .
Thus K = Ko (). Ko being a finite separable extension by our lemma,
Ko = k() for separable. Thus
K = k(, ).
Using the lemma again, our contention follows.
We now investigate the number of intermediary fields between K
and k where K is an algebraic extension of k. Let us first consider a
simple extension K = k(). Let (x) be the minimum polynomial of
in k[x]. Let L be any intermediary field. Let f (x) be the minimum polynomial of over L. Then f (x) divides (x). Let f (x) have coefficients
a0 , . . . an in L. Put L f the field k(a0 , . . . , an ). Then f (x) is minimum
polynomial of over L f . Thus
(K : L) = (K : L f )
But L f L. This proves that L = L f , and so for every intermediary
field there is a unique divisor of (x). Since (x) has in only finitely
many factors, K/k has only a finite number of intermediary fields.
We will now prove that the converse is also true. We shall assume k 44
is infinite.
Let now K/k be an algebraic extension having only a finite number
of intermediary fields. Let , K. Consider the elements + for
k. Since k is infinite, the fields k( + ) are infinite in number and
cannot be all distinct. Let for = 1 , 2 1 , 2
k( + 1 ) = k( + 2 ) = k().
Then + 1 , + 2 are in k(). Thus (1 2 ) k(). Hence
k() because 1 2 k. This means that k().

2. Algebraic extension fields

38
Therefore

k(, ) k() k(, ).


Hence every subfield of K, generated by 2 and hence by a finite number of elements is simple. Let K be a maximal subfield of K/k which
is simple. (This exists since K/k has only finitely many intermediary
fields). Let K0 = k(). Let K and < K0 . Then k() = k(, ) K
and K0 k() contradicting maximality of K0 . Thus () = K. We have
proved
Theorem 6. K/k is simple K/k has only finitely many intermediary
fields.
We deduce
Corollary. If K/k is simple, then every intermediary field is simple.
Note 1. We have the fact that if K/k is infinite there exist infinitely many
intermediary fields.

45

Note 2. Theorem 6 has been proved on the assumption that k is an infinite field. If k is finite the theorem is still true and we give a proof
later.

7 Galois extensions
Let K/k be an algebraic extension and G the group of automorphism of
K which are trivial on k. Let L be the subset of all elements of K which
are fixed by G. L is then a subfield of K and is called the fixed field of
G. We shall now consider the class of algebraic extensions K/k which
are such that the group G(K/k) of automorphisms of K which are trivial
on k, has k as the fixed field. We call such extensions galois extensions,
the group G(K/k) itself being called the galois group of K/k.
We now prove the
Theorem 7. k is the fixed field of the group of k automorphisms of
K K/k is a normal and separable extension.

7. Galois extensions

39

Proof. Let k be the fixed field of the group G(K/k) of k-automorphisms


of K. Let K. Let 1 (= ), . . . n be all the distinct conjugates of
that lie in K. Consider the polynomial
f (x) = (x 1 ) (x n )
If is an element of G(K/k), permutes 1 , . . . , n so that leaves
the polynomial f (x) unaltered. The coefficients of this polynomial are
fixed under all elements of G and hence since k is the fixed field G, f (x)
k[x]. Hence the minimum polynomial roots of the minimum polynomial of , since 1 , . . . n are conjugates. Thus f (x)/(x). Therefore
K splitting field of (x), (x) has all roots distinct. Thus K/k is normal
and separable.

Suppose now K/k is normal and separable. Consider the group 46
G(K/k) of k-automorphisms of K. Let K. Since K/k is separable,
all conjugates of are distinct. Also since K/k is normal K contains all
the conjugates. If is fixed under all G(K/k), then is a purely
inseparable element of K and hence is in k.
Our theorem is thus proved.
We thus see that galois extensions are identical with extension fields
which are both normal and separable.
Examples of Galois extensions are the splitting fields of polynomials
over perfect fields.

Let k be a field of characteristic , 2 and let K = k( ) for k


2

and < k.( ) = k. Every element of K is uniquely of the form

a + b, a, b k. If is an automorphism of K which is trivial on k,

then its effect on K is determined by its effect on . Now


n o

= ( )2 = ( ).( )

or that ( )/ = is such that 2 = 1. Since K, = 1. Thus


is either th identity or the automorphism

( ) =
Thus G(K/k) is a group of order 2. K/k is normal and separable.
We shall obtain some important properties of galois extensions.

2. Algebraic extension fields

40

1) If K/k is a galois extension and k L K, then K/L is a galois


extension also.
47

For, K/L is clearly separable. We had already seen that it is normal.


If we denote by G(K/L) the galois group of K over L, than G(K/L)
is a subgroup of G(K/k).
2) If k L1 L2 K then G(K/L2 ) is a subgroup of G(K/L1 ). This
is trivial.
T
3) If {K } is a family of galois extensions of k contained in then K

S
and k( K ) are galois.

This follows from the fact that this is already true for normal and
also for separable extensions.

4) If K/k is galois and L and L are two intermediary fields of K/k


which are conjugate over k, then G(G/L) and G(K/L ) are conjugate
subgroups of G(K/k) and conversely.
Proof. Since L and L are conjugate over k let be an automorphism of
K/k so that L = L . Let G(K/L). Then for every L
=
But = for L. Thus
1 =
Since this is true for every L, it follows that
1G(K/L) G(K/L)
In a similar manner one proves that G(K/L)1 G(K/L) which
proves our contention.


48

Conversely suppose that L and L are two subfields such that G(K/L)
and G(K/L ) are conjugate subgroups of G(K/k). Let G(K/L ) = 1
G(K/L). Let L and G(K/L). Then 1 G(K/L ) and so

7. Galois extensions

41
1 =

or () = . This being true for all , it follows that L for all


in L . Thus L L . We can similarly prove that 1 L L which
proves our statement.
In particular let L/k be a normal extension of k which is contained
in K. Then L = L for all G(K/k). This means that G(K/L) is a
normal subgroup of G(K/k). On the other hand if L is any subfield such
that G(K/L) is a normal subgroup of G(K/k) then by above L = L for
all G(K/k) which proves that L/k is normal. Thus
5) Let k L K. Then L/k is normal G(K/L) is a normal
subgroup of G(K/k)
6) If L/k is normal, then G(K/k) G(L/k)/G(K/L).
Let G(K/k) and
the restriction of to L. Then
is an automorphism of L/k so that
G(L/k). Now
is a homomorphism
fo G(K/k) into G(L/k). For
= = () =
for all L. Thus
=
The homomorphism is onto since every automorphism of L/k can be 49
extended into an automorphism of K/k. Now
is identity if and only if

=
for all L. Thus G(K/L). Also every G(K/L) has this
property so that the kernel of the homomorphism is G(K/L).
Let K/k be a finite galois extension. Every isomorphism of K/k in
is an an automorphism. Also K/k being separable, K/k has exactly
(K : k) district isomorphisms. This shows that
(K : k) = order of G.
We shall now prove the converse
7) If G is a finite group of automorphisms of K/k having k as the
fixed field then (K : k) = order of G.

2. Algebraic extension fields

42

Proof. Let 1 , . . . , n be the n elements of G and 1 , . . . n+1 any n + 1


elements of K. Denote by VK the vector space over K of n dimensions
formed by n- tuples (1 , . . . , n ). Define n + 1 vectors 1 , . . . , n+1 by
i = (i (i ), . . . , n (i ))i = 1, . . . , n + 1
Among these vectors there exists m n vectors linearly independent
over K. Let 1 , . . . , m be independent. Then
m+1 =

m
X

ai i

n=1

ai K

This equation gives, for the components of the s,


(m+1 ) =

m
X

ai (i ) = 1, . . . , n.

1=i


50

Since 1 , . . . , n form a group then n 1 , . . . , n n are again the


elements 1 , . . . , n in some order. Thus
h (m+1 ) =

m
X
i=1

h (ai |h (i )

which means
(m+1 ) =

m
X

h (ai ) (i )

i = 1, . . . , n

i=1

subtracting we have
m
X
(h (ai ) ai (i ) = 0
i=1

which means that

m
X
i=1


h (ai ) ai i = 0

7. Galois extensions

43

From linear independence, it follows that h (ai ) = ai for all i. But


h is arbitrary. Thus ai k. We therefore have by taking 1 to be the
identity element of G
m
X
m+1 =
ai i
i=1

ai are in k, not all zero. Hence


(K : k) n.
But every element of G is an isomorphism of K/k. Thus
(K : k) order of G = n.
Our assertion is established.
Suppose K/k is a galois extension. For every subfield L of K/k, the
extension K/L is galois. We denote its galois group by G(L) and this
is a subgroup of G(K/k). Suppose g is any subgroup of G(K/k) and let
F(g) be its fixed field. Then F(g) is a subfield of K. The galois group
G(F(g)) of K/F(g) contains g. In general one has only
51
g G(F(g))
Let now K/k be a finite galois extension. Let g be a subgroup of
G(K/k) = G and F(g), the fixed field of g. Then by above


K : F(g) = order of g.
and so

g = G K/F(g)

If g1 and g2 are two subgroups of G with g1 g2 then F(g1 ) and


F(g2 ) are distinct. For if F(g1 ) and F(g2 ) are identical, then by above
g1 = G(K/F(g2 )) = g2 . We thus have the
Main Theorem(of finite galois theory). Let K/k be a finite galois extension with galois group G. Let M denote the class of all subgroups
of G and N the class all subfield of K/k. Let be the mapping which

2. Algebraic extension fields

44

assigns to every subgroup g M, the fixed field F(g) of g in N. Then


is a mapping of M onto N which is biunivocal.
In order to restore this property even for infinite extensions, we develop a method due originally to Krull.
Let K/k be a galois extension with galois group G(K/k). For every
K, we denote by G the galois group G(K/k()). This then is
a subgroup of G(K/k). We make G(K/k) into a topological group by
prescribing the {G } as a fundamental system of neighbourhoods of the
T
T
identity element e G(K/k). Obviously G = (e). For G

52

= for all K so that = e. It is easy to verify that {G } satisfy


the axioms for a fundamental system of open sets containing the identity
elements e.
Any open set in G is therefore a union of sets of type {G} or a
finite intersection of such. Also since G are open subgroups they are
closed; for, G is open for all and hence
[
G
,e

is also open. Therefore G is closed. This proves that the topology on


G makes it totally disconnected. We call the topology on G the Krull
topology.
If g is a subgroup of G, g the closure of g is also a subgroup. We
now prove the
Lemma. Let g be a subgroup of G a and L its fixed field. Then
G(K/L) = g

(the closure of g).

Proof. Let be an element in K and f (x) its minimum polynomial in


k. Consider f (x) as a polynomial over L and let L be its splitting field
over L. Then L /L is a galois extension. The restriction of elements of
g to L are automorphisms of L with L as fixed field (by definition of
L). By finite galois theory these are all the elements of the galois group
of L /L. This means that every automorphism of L /L comes from an
elements of g.


7. Galois extensions

45

Let G(K/L). Let be any element in K and G the group


G(K/k()). The restriction of to L is an automorphism of L with
L as fixed field. There is thus an element g, which has on L the
same effect as . Hence 1 is identity on L and since L we get
1 =

This means that 1 G by definition of G . Hence 1 G 53


or G . But G is an open set containing . Therefore since
{G } for all G form a fundamental system of neighbourhoods of
we conclude that
g .
Thus G(K/L) g
Let now g . Then G is a neighbourhood of and so intersects
g in a non empty set. Let L. Let G g. Let in G such
that
=
By definition of , = . But = = . Therefore G .
But is already is in G . Therefore = . Also being arbitrary
L = L
which means that G(K/L). Thus
g G(K/L)
and our contention is established.
From the lemma, it follows that if L is an intermediary field, the
galois group G(K/L) is a closed subgroup of G(K/k). On the other hand
if g is a closed subgroup of G and F(g) its fixed field then

G K/F(g) = g = g.

we have hence the fundamental

Theorem 8. Let K/k be a galois extension and G(K/k) the galois group
with the Krull - topology. Let M denote the set of closed subgroups of
G and N the set of intermediary fields of K/k. Let be the mapping
which assigns to every g M, the fixed field F(g) of g in N. Then is a 54
biunivocal mapping of M on N.

2. Algebraic extension fields

46

55

Suppose now that K/k is a galois extension and L is an intermediary


field. Let G(K/k) and G(K/L) be the galois groups. On G(K/L) there
are two topologies, one that is induced by the topology on G(K/k) and
the other the topology that G(K/L) possesses as a galois extension. If
L is a subfield of K/L so that L /L is finite, then G(K/L ) is an open set
in the inherent topology on G(K/L). On the other hand L = L() since
L /L is a separable extension.
Thus
G(K/L ) G G(K/L)
which proves that the two topologies are equivalent. Here we have used
the fact that a finite separable extension of L is simple. We gave already
seen the truth of this statement if L is infinite. In case L is finite it is
proved in the next section. In a similar manner if K/k is galois and L is
a normal extension of k in K, then on G(L/k) there are two topologies,
one the inherent one and the other the topology of the quotient group
G(K/k)/G(K/L). One can prove that the two topologies are equivalent.
We call an extension K/k abelian or solvable according as G(K/k)
is abelian or a solvable group. If K/k is a galois extension and G(K/k)
its galois group with the Krull topology let H denote the closure of the
algebraic commutator subgroup of G(K/k). H is called the topological
commutator subgroup. If L is its fixed field, then since H is normal, L/k
is a galois extension. Its galois group is isomorphic to G/H which is
abelian. From the property of the commutator subgroup, it follows that
L is the maximal abelian subfield of K/k.

8 Finite fields
Let K be a finite field of q elements, q = pn where p is the characteristic
of K. Let be the prime field of p elements. Then
(K : ) = n
K the group of non-zero elements of K is an abelian group of order
q 1. For K we have
q1 = 1

8. Finite fields

47

1 being the unit element of K. The q 1 elements of K are roots of


xq1 1. Also
Y
xq1 1 =
(x ).
K

Let K . Let d be its order as an element of the finite group K .


Then d = 1. Consider the polynomial xd 1. It has in K at most d
roots. Also d/q 1. But
xq1 1 = (xd 1)(xq1d + )

Since xd 1 and xq1d + both have respectively at most d and


q 1 d roots in K and they together q 1 roots in K it follows that
for every divisor d of q 1, xd 1 has exactly d roots in K . These roots
from a group of order d. If it is cyclic then there is an element of order
d and there are exactly (d) elements of order d. Also
X
(d) = q 1
d/q1

which proves that for every divisor d of q1 there are (d) 1 elements 56
of order d. Thus
1) The multiplicative group of a finite field is cyclic.
Let k be a finite of q elements and K a finite extension of k of degree
n. Then K has qn elements. Since K is cyclic, let be a generator
of K . Then
K = k()
which proves
2) Every finite extension of a finite field is simple.
e

For a finite field of characteristic p, a a p is an automorphism of


K. Since k has q elements we have
aq = a
for every a k.

2. Algebraic extension fields

48

Now a aq is an automorphism of K/k which fixes elements of k.


Call this automorphism . is determined uniquely by its effect on
a generator of K . Consider the automorphisms
1, , 2 , . . . , n1
These are distinct. For,
i

i = i1 () = i1 (q ) = q

Hence q = 1 qi o( mod qn ) or i = o (Since i < n). But K/k


being of degree n cannot have more than n automorphisms. We have
3) The galois group of a finite extension of a finite field is cyclic.
The generator of this cyclic group, defined by
a = aq

57

is called the Frobenius automorphism. It is defined without any reference to a generator of K .


Let L be an intermediary field of K/k. Then (L : k) is a divisor of
(K : k) = n. If d = (L : k) then L has qd elements. Also since K/k
has a cyclic galois group, there is one and only one subgroup of a
given order d. Hence
4) The number of intermediary fields of K/k is equal to the number of
divisors of n.

Chapter 3

Algebraic function fields


1 F.K. Schmidts theorem
Let K/k be an extension field and x1 , . . . , xn+1 any n + 1 elements of 58
K. Let R = k[z1 , . . . , zn+1 ] be the ring of polynomials in n + 1 variables
over k. Let Y be the ideal in R of polynomials f (z1 , . . . , zn+1 ) with the
property
f (x1 , . . . , xn+1 ) = 0.
Then clearly Y is a prime ideal of R. Also since R is a Noetheeian
ring, Y is finitely generated. Note that Y = (o) if and only if x1 , . . . ,
xn+1 are algebraically independent over k. Y is called the ideal of the
set x1 , . . . , xn+1 .
We shall consider the case where the set x1 , . . . , xn+1 has dimension
n over k, that is that k(x1 , . . . , xn+1 ) is of transcendence degree n over k.
We prove
Theorem 1. Y is a principal ideal generated by an irreducible polynomial.
Proof. Without loss in generality we may assume that x1 , . . . , xn are algebraically independent over k and that xn+1 is algebraic over
k(x1 , . . . , xn ), so that Y , (o).

49

50

3. Algebraic function fields

Consider the degrees of the polynomials f (z1 , . . . , zn+1 ) (in Y ) in


the variable zn+1 . These degrees have a minimum greater than zero since
Y , (o), and x1 , . . . , xn are algebraically independent over k. Let be
a polynomial in Y of smallest degree in zn+1 . Put
+ . . . + A
= Ao zn+1 + A1 z1
n
59

where A0 , A1 , . . . , A are polynomials in z1 , . . . zn with coefficients in


k. We may assume that A0 , A1 , . . . , A have no common factor in
k[z1 , . . . , zn ]. For if A(z1 , . . . , zn ) is a common factor of A0 , . . . , A then
(z1 , . . . , zn+1 ) = A(z1 , . . . , zn ) 1 (z1 , . . . , zn+1 )
and so, since x1 , . . . , xn are algebraically independent,
1 (x1 , . . . , xn+1 ) = o
and 1 will serve our purpose. So we can take to be a primitive polynomial in zn+1 over R = k[z1 , . . . , zn ].
Clearly is irreducible in R . For, if
= g1 (z1 , . . . , zn+1 ) g2 (z1 , . . . , zn+1 )
then gi (x1 , . . . , xn+1 ) = 0 for i = 1 or 2, so that either g1 or g2 is in Y .
Both cannot have a term in zn+1 with non zero coefficient. For then the
degrees in zn+1 of g1 of g2 will both be less that of in zn+1 contradicting
the definition of . So one g1 , g2 say g1 is independent of zn+1 . But this
means that is not a primitive polynomial.
Thus we have chosen in Y a polynomial which os irreducible, of
the smallest degree in zn+1 and primitive in R [zn+1 ].
Let (z1 , . . . , zn+1 ) be any other polynomial in Y .
Since F = k(z1 , . . . , zn ) is a field, F[zn+1 ] is a Euclidean ring so that
in F[zn+1 ] we have
(z1 , . . . , zn+1 ) = A(z1 , . . . , zn+1 ) (z1 , . . . , zn+1 ) + L(z1 , . . . , zn+1 )

60

where A and L are polynomials in zn+1 over F. Here either L = o or


degree of L in zn+1 is less than that of . If L , o, then we may multiply

1. F.K. Schmidts theorem

51

both sides of the above equation by a suitable polynomial in z1 , . . . , zn


over k so that
B(z1 , . . . , zn ) = C(z1 , . . . , zn+1 )(z1 , . . . , zn+1 ) + L1 (z1 , . . . , zn+1 )
L1 having in zn+1 the same degree as L. Since and are in Y , it
follows that L1 Y . Because of degree of L1 , it follows that
L1 = L = 0.
Thus
B(z1 , . . . , zn ) = A(z1 , . . . , zn+1 )(z1 , . . . , zn+1 )
Since is a primitive polynomial, it follows that divides and our
theorem is proved.
We call the irreducible polynomial of x1 , . . . xn+1 over k.
Note that since x1 , . . . xn are algebraically independent over k, the
polynomial
1 (zn+1 ) = (x1 , . . . , xn , zn+1 )
over k(x1 , . . . , xn ) is irreducible in zn+1 .
Let x1 , . . . , xn+1 be of dimension n over k and the irreducible polynomial of x1 , . . . , xn+1 over k. Let be a polynomial in z1 , . . . , zi+1 but
not in zi+2 , . . . , zn+1 , that is it does not involve zi+2 , . . . , zn+1 in its expression. Consider the field L = k(x1 , . . . , xi+1 ). because x1 , . . . , xi+1 are
algebraically dependent ((x1 , . . . , xi+1 ) = 0),
61
dimk L i.
But k(x1 , . . . , xn+1 ) = L(xi+2 , . . . , xn+1 ) so that
dim L k(x1 , . . . , xn+1 ) n i.
Since dimensions are additive, we have
n = dimk L + dimL k(x1 , . . . , xn+1 ) i + n i = n.
Thus L has over k the dimension i. Since is a polynomial in
z1 , . . . , zi+1 every one of these variables occurring, with non zero coefficients, we get the

52

3. Algebraic function fields

Corollary. If x1 , . . . , xn+1 be n + 1 elements of K/k and have dimension


n, there exist among them i + 1 elements, i n, say x1 , . . . , xi+1 (in some
order) such that k(x1 , . . . xi+1 ) has dimension i over k and every i of them
are algebraically independent.
Let K/k be a transcendental extension with a transcendence base B
over k. Then K/k(B) is algebraic. We call K/k an algebraic function
field if
(1) B is a finite set
(2) K/k(B) is finite algebraic.

62

Let dimk K = n. There exist x1 , . . . , xn in K which form a transcendence base of K/k. If K is an algebraic function field then K/k(x1 ,
. . . , xn ) is finite algebraic. Hence K = k(x1 , . . . , xm ), m n, is finitely
generated. This shows that algebraic function fields are identical with
finitely generated extensions.
An algebraic function field K/k is said to be separably generated
if there exists a transcendence base x1 , . . . , xn of K/k such that
K/k(x1 , . . . , xn ) is a separable algebraic extension of finite degree.
x1 , . . . , xn is then said to be a separating base. Clearly every purely transcendental extension is separably generated. Also, if k has characteristic
zero and K is an algebraic function field, it is separably generated. In
this case every transcendence base is a separating base. This is no longer
true if k has characteristic p , o.
For example, let K = k(x, y) be a function field of transcendence
degree one and let
x2 y p = o.
Let k have characteristic p , 2. Obviously x and y are both transcendental over k. But K/k(x) is a simple extension generated by y which is
a root of
z p x2
over k(x)[z] and since k has characteristic p, K/k(x) is purely inseparable. On the other hand K/k(y) is separable since x satisfies over k(y) the
polynomial
x2 y p .

1. F.K. Schmidts theorem

53

Thus y is separating but not x.


An algebraic function field which is not separably generated is said
to be inseparably generated. This means that for every base B of K/k.
K/k(B) is inseparably algebraic.
In algebraic geometry and in algebraic function theory, it is of im- 63
portance to know when an algebraic function field is separably generated. An important theorem in this regard is theorem 2 due to F.K.
Schmidt. We shall first prove a
Lemma . Let k be a perfect field of characteristic p , o and K =
k(x1 , . . . , xn+1 ) an extension field of dimension n. Then K is separably
generated.
Proof. Let be the irreducible polynomial of x1 , . . . , xn+1 and let it be
a polynomial in z1 , . . . , zi+1 but in zi+2 , . . . , zn+1 . Then
(x1 , . . . , zt , . . . , xi+1 )
is irreducible over k(x1 , . . . , xt1 , xt+1 , . . . , xi+1 ) for every t, 1 t i + 1.
At least for one t, (z1 , . . . , zt , . . . , zi+1 ) is a separable polynomial in zt
over k(z1 , . . . , zt1 , . . . , zi+1 ). For, if it is inseparable in every zt , then
p

(z1 , . . . , zi+1 ) k[z1 , . . . , zi+1 ]


and since k is perfect, this will mean that (z1 , . . . , zi+1 ) is the pth power
of a polynomial in k[z1 , . . . , zi+1 ] which contradicts irreducibility of .
So, for some zt , say z1 , we have (z1 , x2 , . . . , xi+1 ) is a separable polynomial. Hence x1 is separable over k(x2 , . . . , xi+1 ) and so over k(x2 ,
. . . , xn+1 ). But x2 , . . . , xn+1 has dimension n and our lemma is proved.

Corollary . Under the conditions of the lemma, a separating base of n
elements may be chosen from among x1 , . . . , xn+1 .
We are now ready to prove the theorem of F.K.Schmidt.
Theorem 2. Every algebraic function field K over a perfect field k is
separably generated.

64

3. Algebraic function fields

54

Proof. Obviously, the theorem is true if k has characteristic zero. So


let k have characteristic p , o. Let K = k(x1 , . . . , xm ) and let n be the
dimension of K/k. Then m n. If m = n there is nothing to prove.
Let m = n + q. If q = 1 then lemma 1 proves the theorem. So let us
assume theorem proved for q 1 instead of q > 1. We may assume,
without loss in generality that x1 , . . . , xn is a transcendence base of K/k.
Consider the fields L = k(x1 , . . . , xn , xn+1 ). It satisfies the conditions
of the lemma. Hence there exist n elements among x1 , . . . , xn+1 say
x1 , . . . , xt1 , xt+1 , . . . xn+1 which from a separating base of L/k. Thus xt
is separable over k(x1 , . . . , xt1 , . . . , xn+1 ) and hence over
M = k(x1 , . . . , xt1 , xt+1 , xt+2 , . . . , xm )
M/k now satisfies the induction hypothesis and so is separably generated. Since K/M is separable, it follows that K is separably generated.

We could prove even more if we assume as induction hypothesis the
fact that among x1 , . . . , xm there exists a separating base.

2 Derivations
Let R be a commutative ring with unit element e. A mapping D of R
into itself is said to be a derivation of R if
65

(1) D(a + b) = Da + Db
(2) D(ab) = aDb + bDa
for a, b R. It is said to be a derivation over a subring R if for every
a R , Da = o. It then follows that for a R and x R
Dax = aDx
The set Ro of a R with Da = o is a subring of R and contains e.
For,
De = De2 = De. e + e. De = 2De ;

2. Derivations

55

so De = o. If Da = o, Db = o, then
D(a + b) =Da + Db = o
D(ab) = aDb + bDa = o
Thus Ro is a subring. We call Ro the ring of constants of the derivation D.
If R is a field, then Ro also is a field. For, if x Ro and x , o, then
o = De = Dx.x1 = Dx.x1 + x.Dx1
Thus
Dx1 = o
so that x1 Ro .
D is said to be a non-trivial derivation of R if there is an x R with
Dx , o. It follows from above that
Theorem 3. A prime field has no non-trivial derivations.
A Derivation D of R is said to be an extension of a derivation D of a
= Da for a R . We now prove the
subring R of R if Da
66
Theorem 4. If K is the quotient field of an integrity domain R, then a
derivation D of R can be uniquely extended to K.
a
Proof. Every element c in K can be expressed in the form c = , a,
b
b R. If an extension D of D exists, then
= Da
Da
But a = bc so that
= Dbc
= bDc
+ cDb
Da = Da
Therefore

= Da cDb = bDa aDb


Dc
b
b2

3. Algebraic function fields

56
If c is expressed in the form

a
, a , b R then ab = ba and so
b

Da.b + a.Db = Db.a + bDa


or that

Da cDb Da cDb
=
b
b

does not depend on the way c is expressed as


which proves that Dc
the ratio of two elements from R. We have therefore only to prove the
In order to prove this, put for c = a
existence of D.
b
= bDa aDb ;
Dc
b2
we should verify that it is a derivation, is independent of the way c is
expressed as ratio of elements in R and that it coincides with D on R.
These are very simple.

67

Let D1 , D2 be two deviations of R. Define D = D1 + D2 by Da =


D1 a + D2 a for a R. Then it is easy to verify that D is a derivation of
R. Furthermore if a R define aD by
(aD)x = a.Dx
By this means, the derivations of R from an R module, Suppose
D1 , . . . , Dr from a basis of the module of derivations of R.
Then every derivation D of R is of the form
X
D=
ai Di , ai R.
i

Let R be an integrity domain and D 1 , . . . , D r the unique extensions


of D1 , . . . , Dr respectively to K, the quotient, field of R. Then D 1 , . . . , D r
are linearly independent over K. For, if
X
1i D i = o,
1i K
i

2. Derivations
then write 1i =

57
ai
, ai , bi R. We get
bi
X ai
i

bi

D i = o.

Multiplying throughout by b1 , . . . , br which is not zero, we get


P
P
( D i )t = o for every t R. Therefore i Di = o which implies
i

that i = o or ai = o.
Also, since every derivation D of K is an extension of a derivation
of R, it follows that the derivations of K from an r-dimensional vector
space over K.
Let us now consider the case where R = k[x1 , . . . , xn ] is the ring of
polynomials in n variables x1 , . . . , xn . The n mappings
Di :

a
, i = 1, . . . , n
xi

are clearly derivations of R over k. They form a base of derivations of 68


R which are trivial on k. For, if
X
ai Di = o,
ai R
i

then, since Di (x j ) = i j , we get


X
o = ( ai Di )x j = a j , j = 1, . . . , n.
i

Also, if D is any derivation of R which is trivial on k, then let Dxi =


ai . Put
X
ai Di .
D = D
i

Then
X
j = Dx j ( ai Di )x j = o
Dx
i

which shows that since x1 , . . . , xn generate R, D = o.

3. Algebraic function fields

58

Suppose D is any derivation of k and let D be an extension of D to R.


i = ai . Let Do be another extension of D which has the property
Let Dx
Do xi = ai , i = 1, . . . , n.
Then D Do is a derivation of R which is trivial on k. But since
(D Do )xi = o,

i = 1, . . . , n

it follows that D = Do . This gives us the


Theorem 5. The derivations of K = k(x1 , . . . , xn ), the field of rational
functions of n variables over k which are trivial on k from a vector space
of dimension n over K. A basis of this space of derivations is given by
the n partial derivations
Di =
69

,
xi

i = 1, . . . , n.

Furthermore if D is any derivation of k, there exists only one extension


D of D to K for which
i = ai ,
Dx

i = 1, . . . , n

where a1 , . . . , an are any n quantities of K arbitrarily given.


We now consider derivations of algebraic function fields.
Let K = k(x1 , . . . , xm ) be a finitely generated extension of k. Put
T = k[x1 , . . . , xm ]. In order to determine all the derivations of K, it is
enough to determine the derivations of T since K is the quotient field of
T . Let D be a derivation of k; we wish to find extensions D of D to K.
Let R denote the ring of polynomials k[z1 , . . . , zm ] in m independent
f
the
variables. For any polynomial f (x1 , . . . , xm ) in T , denote by
xi
f
polynomial obtained by substituting zi = xi , i = 1, . . . , m in
where
zi
f = f (z1 , . . . , zm ) is in R.
If
X
f =
a1 , . . . , m x11 xmm

2. Derivations

59

a1 , . . . , m k, put
fD =

X
(Da1 , . . . , m )x11 xmm .

Obviously f D is a polynomial in T . If D is an extension of the


derivation D, then clearly
D f = f D +

m
X
f
Dxi
xi
i=1

for any f T . Also D is determined uniquely by its values on x1 , . . . , xm 70


i cannot be arbitrary elements of T . For, let
which generate T . Now Dx
Y be the ideal in R of the set x1 , . . . , xm .
Then for f (z1 , . . . , zm ) in Y ,
f (x1 , . . . , xm ) = o.
= 0, the Dx
i would have to satisfy the infinity
Therefore, since Do
of equations
m
X
f
Dxi
0 = fD +
xi
i=1
for every f in Y .
Conversely, suppose u1 , . . . um are m elements in T satisfying
m
X
f
ui = 0
f +
xi
i=1
D

for every f in Y . For any in T define D by


= D + f D +
D

m
X

Dxi
xi
i=1

i = ui . Then clearly D is a derivation of T and it coincides


where Dx
with D on k. Furthermore D does not depend on the way is expressed

3. Algebraic function fields

60

as a polynomial in x1 , . . . , xm . For, if = a(x1 , . . . , xm ) = b(x1 , . . . , xm )


then, since a b Y have
X a
b 
aD bD +
ui
ui = o
xi
xi
i
which proves our contention. Hence
71

Theorem 6. Let K = k(x1 , . . . , xm ) and D a derivation of k. Let Y


be ideal in k[z1 , . . . , zm ] of the set x1 , . . . , xm . Let u1 , . . . , um be any
elements of K. There exists a derivation D and only one satisfying
i = ui , i = 1, . . . , m
Dx
and extending the derivation D in k, if and only if, for every f Y
fD +

m
X
f
ui = o.
x
i
i=1

and then for every in K,


= D +
D

m
X

ui .
xi
i=1

The infinite number of conditions above can be reduced to a finite


number in the following manner. Since R = k[z1 , . . . , zm ] is a noetherian
ring, the ideal Y has a finite set f1 , . . . , fs of generators so that f Y
may be written
s
X
f =
Ai fi ,
Ai R
i=1

Suppose f1 , . . . , fs satisfy the above conditions, then since


X
X
f D (x) =
AiD fi +
fiD Ai
i

f
=
xi

X
j

Aj

f j X A j
+
fj
,
xi
xi
j

2. Derivations

61

we get
f D (x) +

m
X
f
ui = o.
xi
i=1

We may therefore replace the above by the finitely many conditions


fiD +

m
X
f
u j = o, i = 1, . . . , s.
x
j
j=1

We now consider a few special cases.


72
Let K = k(x) be a simple extension of k. Let D be a derivation of k.
We will study extensions D of D into K.
(1) First let x be transcendental over k. The ideal of x in k[z] is zero.
arbitrarily. Thus for every
This means that we can prescribe Dx
u K there exists one and only extension D with
=u
Dx
(2) Let now x be algebraic over k. Suppose x is inseparable over
k. Let f (z) be the minimum polynomial of x in k[z]. Then f (z)
generates the ideal of x in k[z]. But x being inseparable, f (x) = o.
This means that D has to satisfy
f D = o.

Also, D is uniquely fixed as soon as we assign a value u to Dx.


This can be done arbitrarily as can be easily seen. Thus there exist

an infinity of extensions D.
(3) Finally, let X be separable. Then f (z), the irreducible polynomial
of x over k is such that
f (x) , o.
Since f (z) generates Y we must have
= o,
f D (x) + f (x)Dx

3. Algebraic function fields

62
is uniquely fixed by D.
or that Dx
=
Dx
73

f D (x)
f (x)

(*)

There is thus only one extension of D to K and it is given by ().


In particular, K has no derivations, except the trivial one, over k.
We shall now prove
Theorem 7. In order that a finitely generated extension K = k(x1 ,
. . . , xn ) be separably algebraic over k, it is necessary and sufficient that
K have no non-trivial derivations over k.
Proof. If K/k is algebraically separable, then since K is finitely generated over k, it follows that K = k(x) for some x and the last of the considerations above shows that K has no nontrivial derivations over k. 
Suppose now K/k has no-trivial derivations. In case n = 1, our
considerations above show that K/k is separable. Let now n > 1 and
assume that theorem is proved for n 1 instead of n.
Put
K = K1 (xn )
K1 = k(x1 , . . . , xn1 ).

74

Then xn is separably algebraic over K1 . For, if not, let x1 be inseparable over K1 or transcendental over K1 . In both cases the zero
derivation in K1 can be extended into a non-trivial deri-vation of K contradictions hypothesis over K.
Thus xn is separable over K1 . This implies, since K has no derivations over k that K1 and our theorem is proved.
Note that in the theorem above, the fact that K/k is finitely generated

is essential. For instance, if k is an imperfect field and K = k p then


K/k is infinite. Also if a k then a = b p for some b K. If D is a
derivation of K, then
Da = Db p = pb p1 Db = o
This proves, in particular, that a perfect field of characteristic p , o,
has only the trivial derivation.

2. Derivations

63

If we take K to be the algebraic closure of the rational number filed


then K has only the trivial derivation.
Let K = k(x, y) be an algebraic function field of one variable. Let us
assume that x is a separating variable and (X, Y) the irreducible polynomial of x, y over k. Then if D is a derivation of K over k,

Dx +
Dy = o
x
y
so that if we assume that y is separable over k(x), then

Dy =

, o and hence
y

Dx

This shows that the ratio Dy/Dx is independent of D.


Also, for any rational function (x, y) of x, y
D =

Dx +
Dy
x
y

which gives,if Dx , o
D
=

Dx
x y

!
x

which is a well known formula in elementary calculus.


We shall now obtain a generalisation of theorem 7 to algebraic function fields.
Let K = k(x1 , . . . , xm ) be an algebraic function field of dimension
n, so that o n m. Let f1 , . . . , fs be a system of generators of the 75
ideal Y of polynomials f (z1 , . . . , zm ) in k[x1 , . . . , xm ] which vanish for
f
for f in k[z1 , . . . , zm ] have the same meaning as
x1 , . . . , xm . Let
xi
before.
Denote by M the matrix
!
f
i = 1, . . . , m
M=
xi

3. Algebraic function fields

64

j = 1, . . . , s
where i is the row index and j the column index. We denote by t the
rank of the matrix M which is a matrix over K.
Let VK (D) denote the vector space of derivations of K which are
trivial on k. This is a vector space over K. Denote by l the dimension of
VK (D) over K. We then have
Theorem 8.

l+ t =m.

Proof. For any integer p, denote by W p the vector space over K of dimension p, generated by p-tuples (1 , . . . , p ), i K.

Let denote the mapping
D = (Dx1 , . . . , Dxm )
of Vk (D) into Wm . This is clearly a homomorphism of Vk (D) into Wm .
The kernel of the homomorphism is the set of D for which Dxi = o; i =
1, . . . , m. But since K is generated by x1 , . . . , xm , this implies that D = o.
Thus Vk (D) is isomorphic to the subspace of Wm formed the vectors
(Dx1 , . . . , Dxm ).

76

Consider now the vector space W s and let be the mapping


((1 , . . . , m ) = (1 , . . . , m )M of Wm into W s . Put
j =

m
X
j=1

fi
; i = 1, . . . , s
x j

so that
(1 , . . . , s ) = (1 , . . . , m )M.
The rank of the mapping is clearly y equal to the rank t of the
matrix M. It is the dimension of the image by of Wm into W s . The
kernel of the mapping is the set of (1 , . . . , m )with
i = o; i = 1, . . . , s.

2. Derivations

65

which, by theorem 6 is clearly isomorphic to the subspace of Wm formed


by vectors (Dx1 , . . . , Dxm ), D VK (D). This, by previous considerations, proves the theorem.
We shall now prove the
Theorem 9. With the same notations as before, there exist elements,
say x1 , . . . , x1 of dimension n over k such that K/k(x1 , . . . , x1 ) is a separably algebraic extension.
Proof. Since the matrix M has rank t, there exists a submatrix of M of
t rows and which is non-singular. Choose notation in such a way, that
this matrix is
!
i = m t + 1, . . . , m
fj
,
P=
xi
j = s t + 1, . . . , s
Note that t Min (s, m). Let L = k(x1 , . . . , x1 ). Then
K = L(x1+p , . . . , xm ). Let D be a derivation of K over L. Then since
f j (x1 , . . . , xm ) = o, we must have, by theorem 6
m
X
fj
i=1

xi

Dxi = o.

But since D is zero on L,


Dxi = o, i = 1, . . . , l.
Thus

m
X
fj
Dxi = o.
xi
i=l+1

This means that


Dxl+1 o


P ... = ...


Dxm
o

But since |P| , o, it follows that Dxl+1 = o . . . , Dxm = o which


shows that D = o

77

3. Algebraic function fields

66

But K = L(xl+1 , . . . , xm ) is finitely generated over L. Using theorem 7, it follows that K/L is algebraic and separable. We get incidentally
n 1.
We now prove the important

Theorem 10. Let K = k(x1 , . . . , xm ) be of dimension n. Then K is


separably generated over k, if and only if dim VK (D) = n. In that case
there exists, among x1 , . . . , xm , a separating base of n elements.
Proof. If VK (D) has dimension n then theorem 9 shows that there exist n
elements x1 , . . . , xn among x1 , . . . , xm such K/k(x1 , . . . , xn ) is separably
algebraic.


78

Suppose now that K/k is separably generated. Let y1 , . . . , yn be


a separating base so that K/k(y1 , . . . , yn ) is separably algebraic. k(y1 ,
. . . , yn ) has n linearly independent derivations D1 , . . . , Dn over k defined
by

o, if i , j
Di y j =

1, if i = j

Since K/k(y1 , . . . , yn ) is separably algebraic and finite, it follows


that each of D1 , . . . , Dn has a unique extension D 1 , . . . , D n to K. Now
D 1 , . . . , D n are linearly independent over K. For, if
X
ai D i = o, ai K,
i

then
X

ai D i (y j ) = o for all j.

Hence a j = o for j = 1, . . . , n. Let now D be a derivation of K/k and


let Dyi = ai . Put
X
D = D
ai D i .
i

3. Rational function fields

67

i = o for all i. Therefore since K/k(y1 , . . . , yn ) is separable,


Then Dy
D = o.This proves that
dim VK (D) = n.
and our theorem is completely established.
Let K = k(x1 , . . . , xn , y) where k is of characteristic p , o and k is
an imperfect field. Let y be algebraic over k and be a root of
zp t
t k. Then K/k is an inseparably generated extension and
dim VK (D) = n + 1
where n is the dimension of K/k.

3 Rational function fields


79

Let us now consider the field K = k(x) of rational functions of one


f (x)
variable. Let y be any element of K. Hence y =
where f and g are
g(x)
polynomials in x over k. Also K is the quotient field of the ring L = k[x]
of polynomials in x.
Assume that ( f (x), g(x)) = 1, that is that they have no factor in
common. Let n be defined by
n = max(deg f (x), deg g(x)).
If n = o, then clearly f k, g k and so y k. Let us assume that
n , o so that at least at least one of f and g is a non-constant polynomial.
n is called the degree of y.
Let F = k(y) be the field generated over k by y.
Then x satisfies over F the polynomial
(z) = f (z) yg(z).

3. Algebraic function fields

68

(z) is not a constant polynomial over k(y). For, let


1
X

f (z) =
ai z

i=o
ai , bi k.

g(z) =
bi z

i=o

Then n = max(1, m). The coefficient of zn in (z) is

a1

a1 yb1

ybm

if
if
if

1>m
1=m
1<m

In every case, it follows that since y is not in k, (z) is a non-constant


polynomial. Since (z) has degree n in z, it follows that
(K : F) n.
80

We assert that (z) is irreducible over F. For, if it is reducible over


F[z], then since F = k(y), it will be reducible over k[y, z]. So let (z) =
1 (y, z) 2 (y, z) in k[y, z]. Since (z) is linear in y it follows that one of
1 or 2 has to be independent of y. But then since ( f (z), g(z)) = 1, (z)
is a primitive polynomial in y over k[z]. Therefore (z) is irreducible.
This means that
(K : F) = n
It proves, in particular that y is transcendental over k. Hence
Theorem 11. k is algebraically closed in k(x).
We can extend it to the case where K = k(x1 , . . . , xn ) is a purely
transcendental extension of dimension n. We use induction on n. Theorem 11 proves that n = 1, k is algebraically closed in k(x). Let, for n 1
instead of n, instead of n, it be proved that k is algebraically closed in the
purely transcendental extension k(x1 , . . . , xn1 ). Let K = k(x1 , . . . , xn )
be of dimension n.

3. Rational function fields

69

Let z in K be algebraic over k. Then z K = K1 (xn ), K1 =


k(x1 , . . . , xn1 ). Therefore by theorem 11 since z is algebraic over K1 ,
z K1 . By induction hypothesis, z k.
Thus
Corollary. If K = k(x1 , . . . , xn ) has dimension n over k, then k is algebraically closed in K
It is easy to extend this to the case where K is a purely transcendental
extension of any transcendence degree.
Since K = k(x), we call x a generator of K over k. Let y also be a 81
generator so that K = k(y). Then
(k(x) : k(y)) = 1
which shows by our considerations leading to theorem 11 that
y=

a(x)
b(x)

where a(x) and b(x) are coprime and have at most the degree 1 in x.
Thus
x +
y=
x +
where , , , are in k and since y is transcendental over k,
, 0.
An automorphism of K which is identity on k, is uniquely fixed by
its effect on x. If it takes x into y then x and y are related as above. If
x and y are related as above, then the mapping which assigns to x the
element y is an automorphism.
If we consider the group of two rowed non-singular matrices, with
elements in k, then each matrix gives rise to an automorphism of K/k.
Obviously two matrices and give rise to the same automorphism if
and only if = for some , o in k. Hence
Theorem 12. The group of automorphisms of K = k(x) over k is isomorphic to the factor group of the group of two rowed matrices over k
modulo the group of matrices E, , 0 k and E is the unit matrix of
order 2.

3. Algebraic function fields

70

82

We shall call this group P2 .


From theorem 11, it follows that if L is an intermediary field between K and k then L is transcendental over k. But much more is true as
in shown by the following theorem of Luroth.
Theorem 13. If K = k(x) is a simple transcendental extension of k and
k L K, L = K() for some K.
Proof. We shall assume that L , k so that L is transcendental over k
and contains an element t, transcendental over k andy by considerations
leading to theorem 11, we have K/k(t) is finite algebraic. Since L k(t),
it follows that
(K : L) <
Let x satisfy over L the irreducible polynomial
f (z) = zn + a1 zn1 + + an
where a1 , . . . , an L and n = (K : L). At least one ai is not in k since x
is transcendental over k. The ai s are rational functions of x. So we may
write
b0 (x) f (z) = f (x, z) = b0 (x)zn + b1 (x)zn1 + + bn (x) where
b0 (x), . . . , bn (x) are polynomials in x and f (x, z) is a primitive polynomial in z over k[x]. Let m be the maximum of the degrees of b0 (x),
. . . bn (x). Let ai be not in k. Then
ai =

bi (x)
b0 (x)

so that degree ai m. Since ai L and L k(ai ), it follows that


nm
Let us write w instead of ai . Then w =
83

bi (x)
.
b0 (x)

Let w = h(x)/g(x) where (h(x), g(x)) = 1. Then L k(w).


Also x satisfies over k(w) the polynomial
h(z) wg(z)

3. Rational function fields

71

so that since f (z) is irreducible, it follows that f (z) divides h(z) wg(z),
which is a polynomial of degree m in z. Let us therefore write
h(z) wg(z) =

f (x, z) (x, z)c1 (x)


bo (x)
co (x)

where (x, z) is a primitive polynomial in z over k[x]. We therefore get


h(x)
,
on substituting the w =
g(x)
h(z)g(x) g(z)h(x) =

g(x)c1 (x)
. f (x, z)(x, z).
bo (x)co (x)

The left hand side being a polynomial in x and z, f and being


primitive polynomials in z over k[x], it follows that
h(z)g(x) g(z)h(x) = f (x, z)1 (x, z)
where 1 (x, z) k[x, z]. We now compare degrees in x and z on both
sides of the above identity. On the right hand side the degree in x is m
since one of bo (x), . . . , bn (x) has degree m. Therefore the left side has
degree in x m. But the degree in x equals degree of w m. Thus
degree of w = m. Since the left side is symmetrical in z and x,it follows
that it has degree m in z. Therefore 1 has to be independent of x.
Hence h(z) wg(z) = f (z)(z), (z) being independent of x. This
can happen only if (z) is a constant. This proves that
n = m.
Now (K : k(w)) = m = (K : L) and L k(w). Thus
L = k(w)
and our theorem is proved.
The analogue of Luroths theorem for K = k(x1 , . . . , xn ) is not known
for n > 1.
Let K = k(x) and let G be a finite granite group of automorphism of
K/k. If L is the fixed field of G, then K/L is a finite extension of degree
equal to order of G. By Luroths theorem L = k(y) for some y. Thus
degree y = order of G.

84

3. Algebraic function fields

72

For instance, let G be the finite group of automorphisms of K = k(x)


defined by
1
1
x
1
,x
x x, x 1 x, x , x 1 , x
x
x
1x
x1
This is a group of order 6 and the fixed field will be k(y) where k(y)
consists of all rational functions f (x) of x with
1
1
1
x
f (x) = f (1 x) = f ( ) = f (1 ) = f (
) = f(
).
x
x
1x
1x
We have only to find a rational function of degree 6 which satisfies
the above conditions. The function
f (x) =

(x2 x + 1)3
x2 (x 1)2

satisfies the above conditions and so y = f (x).


Theorem 14. If G is any finite subgroup of P2 of linear transformations
ax + b
cx + d
a, b, c, d k, ad bc , 0 then there exists a rational function f (x) such
that every function (x) which is invariant under G is a rational function
of f (x). f (x) is uniquely determined up to a linear transformation
x

85

f (x) +
V f (x) +
, , k, , 0.
We now consider the case of a rational function field K = k(x1 ,
. . . xn ) of n variables. Let G be a finite group of automorphisms of K
which are trivial on k and let L be the fixed field of G. Clearly L has
transcendence degree n over k. It is not known, except in simple cases,
whether L is a purely transcendental extension of k or not. We shall,
however, consider the the case where G is the symmetric group on n

3. Rational function fields

73

symbols. So G S n . Let G operate on K in the following manner. If


is an element of S n , then is a permutation
!
1, 2, 3, . . . , n
=
.
1 , 2 , 3 , . . . , n
We define on K by
xi = xi ; 1, . . . , n.
We then obtain a faithful representation of S n on K and we denote
this group again by S n . An element of k which is fixed under S n and
which therefore is in L is called a symmetric function of x1 , . . . , xn . Obviously
(K : L) = n!
and the galois group of K/L is S n .
Consider the polynomial

86

f (z) = (z x1 ) . . . (z xn ).
Since every permutation in S n leaves f (z) fixed, it follows that f (z)
L[z]. Let us write
f (z) = zn s1 zn1 + s2 zn2 + (1)n sn
where
si =

1t1 <t2 <...<ti n

xt1 xti .

The quantities s1 , . . . , sn are called the elementary symmetric functions of x1 , . . . xn . Put L1 = k(s1 , . . . , sn ). Then L1 L. Also f (z) is a
polynomial in L1 {z} and is irreducible over it. f (z) is separable and K is
the splitting field of f (z) over L1 . Thus K/L1 is galois. Since f (z) is of
degree n
(K : L1 ) n!
Since L L1 , it follows that L = L1 and
L = k(s1 , . . . , sn ).
We have therefore the

74

3. Algebraic function fields

Theorem 15. Every rational symmetric function of x1 , . . . , xn is a rational function over k of the elementary symmetric functions s1 , . . . , sn .
Incidentally since L/k has dimension n, the elementary symmetric
functions s1 , . . . , sn are algebraically independent over k.

Chapter 4

Norm and Trace


1 Norm and trace
87

Let K/k be a finite extension and let 1 , . . . , n be a base of K/k so that


every K may be written
X
ai i
=
i

ai k. By means of the regular representation


A
where A = (ai j ) is an n rowed square matrix with
X
i =
ai j j
i = 1, . . . , n
j

the field K becomes isomorphic to the subalgebra formed by A in the


algebra mn (k) of n rowed matrices over k. We denote by NK/k , S K/K
the norm and trace respectively of K over k and they are defined by
NK/k = |A |

S K/k = trace A .

75

4. Norm and Trace

76
Defined as such, it follows that
NK/k = NK/k . NK/k
S K/k ( + ) = S K/k + S K/k

for , K. Obviously NK/k is a homomorphism of K into


K and similarly S K/k is a homomorphism of K + , the additive
group, into k+ .
Let 1 , . . . , n be any other basis of K/k. Then

88



1
1
.
..
. = P ..


n
n

where P is a non-singular matrix in mn (k). Since

it follows that



1
1

.
.. = A ...


n
n



1
1

.

1
.. = PA P ...


n
n

which shows that by means of the new basis the matrix associated to
is B where
B = PA P 1
and then we have
| B | = |A |

Trace B = Trace A .
This shows that NK/k and S K/k are invariantly defined and do not
depend on a basis of K/k.

1. Norm and trace

77

We write
fK/k (x) =| xE A |
and call it the characteristic polynomial of . Obviously fK/k (0) =
(1)n | A | so that
NK/k = (1)n fK/k (0) = (1)n an .

(1)

We also see easily that


S K/k = a1

(2)

where
fK/k (x) = xn + a1 xn1 + + an
a1 , . . . , an k.
Let k L K be a tower of finite extensions. Let (K : L) = m
and let 1 , . . . , m be a basis of K/L. Similarly let (L : k) = n and let 89
1 , . . . , n be a base of L/k. Then (1 1 , . . . , n m ) is a base of K/k.
Let L and consider the matrix of by the regular representation of
K/k in terms of the base (1 1 , . . . , n m ). Call it A .
Then it is trivial to see that

A 0

A =

0 A
a matrix of mn rows and columns. Therefore

NK/k =| A | (NK/k )(K:L)


S K/k = (K : L)S K/k

Also the characteristic polynomials of as belonging to L and to K


respectively are fL/k (x) and fK/k (x) and they are related by
fK/k (x) = ( fL/k (x))(K:L) ,

(3)

In particular let L = k(). Then fL/k (x) is the minimum polynomial


of . We, therefore, have the

4. Norm and Trace

78

Theorem 1. If K/k is a finite extension and K, (x) its minimum


polynomial over k and f (x) its characteristic polynomial, then
f (x) = ((x))r
where r = (K : k()).

90

From our formulae above, it follows that we can compute the norm
and trace of in K from a knowledge of its minimum polynomial.
Let now K/k be a finite extension and an element of K. Let
[K : k()] = m , [k() : k] = n be the degrees of separability of
K over k() and k() over k respectively. Then K/kn haso m n distinct
i=1,...,n
k-isomorphisms in an algebraic closure of k. Let i j , j=1,...,m
, be
these isomorphisms and let notation be so chosen that i1 , . . . , im , have
the same the same effect on k(). Then we may take 11 , 12 , . . . , n 1
as a complete system of distinct isomorphisms of k()/k in .
By our considerations above fk()/k (x) is the polynomial of as
well as the characteristic polynomial of in k()/k. If fK/k (x) is the
characteristic polynomial of in k, then
fK/k (x) = ( fk()/k (x))(K:k())

(4)

Now, because of the properties of the isomorphisms i j


m
n Y
n
Y
Y

(x i j ) =
(x i1 )m .
i=1 j=1

(5)

i=1

But since fk()/k (x) is the minimum polynomial of , we have


n
{k():k}

fk()/k (x) =
(x i1 )

i=1

where {k() : k}, as usual, denotes the degree of inseparability of k()


over k. Using (5) we get

{k:k}
m {K:k}

Y
Y

(x

()
=
.
(x

i
j
i1

i, j

i=1

1. Norm and trace

79

But {K : k} = {K : k()} {k() : k} so that

{K:k}

(x

)
= fk()/k (x) m {K:k()}

ij

i, j

which proves that

fK/k (x) =

(
Q

){K:k}
((x )

91

(6)

where runs through all the distinct isomorphisms of k() in . Using


(1) and (2) we get
{K:k}

NK/k =
(7)

is a conjugate of . Similarly

S K/k = {K : k}

(8)

If K/k is inseparable, then {K : k} = pt , t 1 so that for every K


S K/k = o.
On the other hand, suppose K/k is finite and separable. Let 1 ,
. . . , n be all the distinct isomorphisms of K/k in , an algebraic closure
of K. Then n = (K : k) and since 1 , . . . , n are independent k-linear
functions of K/k in , it follows that
= 1 + + n
is a non - trivial klinear function of K/k in . Therefore there exists a
K such that , 0. But by formula (8),
= 1 + . . . + n = S K/k
so that we have the

4. Norm and Trace

80

Theorem 2. A finite extension K/k is separable, if and only if there exist


in K an element whose trace over k in not zero.
In case k has characteristic zero, or k has characteristic p n = (K : 92
k), the unit element 1 in k has trace , o. In order to obtain an element
in K with S K/k , 0, in every case we proceed thus:
Let K/k be separable and K = k() for an element . Let (x) be its
irreducible polynomial over k and
(x) = (x 1 ) (x n ).
It follows then that
1
xn1 X n1
i
=
.
( ) x
(x)

i
i
i
Comparing coefficients of xn1 on both sides we get
X n1
i
= 1.
( )

i
i
If we put =

n1
and observe that () K, we get
()
S K/K = 1.

Using formula (6), it follows that if k L K is a tower of finite


extensions and K, then
NK/k = NL/k (NK/L )
S K/k = S L/k (S K/L )

We now give a simple application of formula (7) to finite fields.


Let k be a finite field of q = pa elements so that p is the characteristic
of k. Let K be a finite extension of k so that (K : k) = n. Then K has

2. Discriminant

81

qn elements. The galois group of K/k is cyclic of order n. Let be the


Frobenius automorphism. Then for K,
= q .
93

The norm of is
2

n1

NK/k =
qn 1
= q1 .
Since K has qn 1 elements,
n 1

=1

for all K . Since K is a cyclic group, the number of elements in K


with
qn 1
q1 =1
is precisely qn 1/q 1, since q 1 divides qn 1.
Now NK/k is a homomorphism of K into k and the kernel
of the homomorphism is the set of in K with
qn 1
1 = NK/k = q 1 .
By the first homomorphism theorem we have, since k has only q1
elements, the
Theorem 3. If K/k is finite and k is a finite field, then every non-zero
element of k is the norm of exactly (K : k ) elements of K.
It is clear that this theorem is not in general true if k is an infinite
field.

4. Norm and Trace

82

2 Discriminant
Let K/k be a finite extension and 1 , . . . , n a basis of K/k. Suppose 94
is a k-linear map of K into k, that is
() k

( + ) = +
() = ()
where , K, K. Let M denote the matrix
M = ((i j))
of n rows and columns. We denote by DK/k (1 , . . . , n ) its determinant and call it the - discriminant of the basis 1 , . . . , n of K/k. If
1 , . . . , n is another basis, then


1
1

.
.. = P ...


n
n

where P is an n rowed non-singular matrix with elements in k.


If P = (pi j ) then
X
pi j j
i =
j

so that
(a b ) =

pai pb j (i j )

i, j

which proves that


DK/k (1 , . . . , n ) = |P|2 DK/k (1 , . . . , n ).
Therefore DK/k = 0 if it is zero for some basis.
We now prove

2. Discriminant

83

Theorem 4. If 1 , . . . , n is a basis of K/k and a k-linear map of K


into k, then
DK/k (1 , . . . , n ) = 0
95

if and only if is the zero linear mapping.


Proof. If is the zero linear map, that is, one that assigns to every
element in K, the zero element, then DK/K = 0. Now let DK/k = 0.
This means that the matrix M with elements in k has determinant zero.
Therefore there exist a1 , . . . , an in k, not all zero, such that

a1 0

M ... = ... .

0
an
This means that
n
X

(i j )a j = 0; 1 = 1, . . . , n.

j=1

If we put z =

a j j . then we have

(i z) = 0, i = 1, . . . , n.


Let be any element in K. Since a1 , . . . , an are not all zero, z , 0


and so put

= b1 1 + + bn n , bi k.
z
  P
Then () =
z = bi (i z) = o. This proves that is the
z
i
trivial map.
The mapping S K/K is also a k-linear map of K into k. For a
basis 1 , . . . , n of K/k we call
DK/k (1 , . . . n ) = |(S k/K (i j ))|
the discriminant of the basis 1 , . . . n . Using theorem 2 and theorem 4,
we get

4. Norm and Trace

84

Theorem 5. Discriminant of a base of K/k is not zero if and only if K/k


is separable.
96

Let K/k be finite separable. Let 1 , . . . , n be the distinct k-isomorphisms of K over k in . Then
X
S K/k =
i .
i

Therefore
2

1 1 , 1 2 , . . . , 1 n


..
DK/k (1 , . . . , n ) =
.
.

n
1
1 , . . . . . . , 1

(9)

Since K/k is finite separable, K = k() for some . Also 1, , 2 ,


. . . , n1 form a base of K/k and we have


...,
1
1,

1 ,
...,
1
n1
.
D(1, , . . . , ) =
. . . . . . . . . . . . . . . . . . . . . . . . .
(n1 )1 , . . . , (n1 )n

Also, D(1, , . . . , n1 ) D(1 , . . . , n ) is the square of an element
of k. The determinant


...,
1
1,

1 ,
...,
n


. . . . . . . . . . . . . . . . . . . . . . . . .
(n1 )n , . . . , (n1 )n

is the called Van-der-Monde determinant. We call D(1, , . . . , n1 ) is


the discriminant of and denote it by DK/k ().
Let f (x) be the minimum polynomial of . Then f (x) = (x
1 ) . . . (x n ). Since f () , 0 , we have
f () = (1 2 )(1 3 ). . . . .(1 n )
and is an element of K. We call it the different of and denote it dK/k ().
Also the Vander - monde determinant shows that

2. Discriminant

85
DK/k () = (1)n(n1)2 NK/k (dK/k )

97

Suppose K/k is a finite galois extension. Let 1 , . . . , n be the distinct automorphisms of K/k. We shall now prove
Theorem 6. If k contains sufficiently many elements, then there exists
in K an element such that 1 , . . . , n form a basis of K over k.
Proof. It K such that
D(1 , . . . , n ) , o
then 1 , . . . , n form a base of K/k. For,if
X
ai i = o, a1 , . . . , an k
i

not all zero, then since 1 , . . . , n form a group


X
ai j i = o, j = 1, . . . , n.
i

Therefore from the expression (9) for DK/k (1 , . . . , n ), it follows


that

a1 o

i j 
S K/k ( ) ... = ...

o
an

or that D(1 , . . . , n ) = o which is a contradiction. We have therefore


to find an with this property. Put
i = x1 1 i + + xn n i ,

i = 1, 2, . . .

where x1 , . . . , xn are indeterminates and 1 , . . . , n is a basis of K/k


Then
S K/k (i j ) = 1 i 1 j , + + n i n j
X

=
S K/k (a i b j )xa xb .
a,b

86

4. Norm and Trace

Then DK/k (1 , . . . , n ) defined as the determinant of the matrix 98


(S K/k (i j )) is a polynomial in x1 , . . . , xn with coefficients in k.
In order to prove that DK/k (1 , . . . , n ) is non-zero polynomial,
notice that by definition,

1
1 , 1 , . . . , n 1 x1
1
2
.
..
. = . . . . . . . . . . . . . . . . . . . . . . ..
n
1 n , 2 n , . . . , n n xn

so that if 1 = 1 and i = o for i > 1, then x1 , . . . , xn are not all


zero and for this set of values of x1 , . . . , xn , the polynomial DK/k (1 ,
. . . , n ) has a value , o, as can be see from the fact that

2
1 , . . . , n
DK/k (1 , . . . , n ) = . . . . . . . . . . . . . . . . . . . .
n i , . . . , n n
Therefore if k has sufficiently many elements, there exist values in k
of x1 , . . . , xn , not all zero, such that DK/k (1 , . . . , n ) , o.
This proves the theorem
In particular, if k is an infinite field, there exists a base of K/k consisting of an element and its conjugates. Such a base is said to be a
normal base.
The theorem is also true if k is a finite field.

Chapter 5

Composite extensions
1 Kronecker product of Vector spaces
99

Let V1 and V2 be two vector spaces over a field k and V1 V2 their


cartesian product. If W is any vector space over k, a bilinear function
f (x, y) on V1 V2 is, by definition, a function on V1 V2 into W such
that for every x V1 the mapping x : y f (x, y) is a linear function
on V2 into W and for every y V2 the function y : x f (x, y) is a
linear function on V1 to W.
A Vector space T over k is said to be a Kronecker product or tensor
product of V1 and V2 over k, if there exists a bilinear function on V1 V2
into T such that
1) T is generated by (V1 V2 )
2) if V3 is any vector space over k, then for every bilinear function on
V1 V2 into V3 there exists a linear function on T into V3 such that
= .
This is shown by the commutative diagram
V1 V2M

MMM
MM
MM&&

We shall now prove the


87

// V
v;; 3
v
vv
vv

5. Composite extensions

88

Theorem 1. For any two vector spaces V1 and V2 over k there exists
one and upto k-isomorphism only one Kronecker product T of V1 and
V2 over k.
100

Proof. The uniqueness is easy to establish. For, let T 1 and T 2 be two


vector spaces satisfying the conditions 1) and 2). Let T 1 be generated
by 1 (V1 V2 ) and T 2 by 2 (V1 V2 ). Let be the linear map of T 1
into T 2 such that 1 = 2 and the linear map of T 2 into T 1 such that
2 = 1 . Since 1 (V1 V2 ) generates T 1 we see that is identity
on T 1 . Similarly is identity on T 2 . Thus and are isomorphisms
onto.

We now prove the existence of the space T .
Let V be the vector space formed by finite linear combinations
X
axy (x, y)

axy k and (x, y) V1 V2 . Every bilinear function f on V1 V2 into V3


can be extended into a linear function f of V into V3 by the prescription
X
 X
f
axy (x, y) =
ax,y f (x, y)
Let W be the subspace of V generated by elements of the type
(x + x1 , y) (x, y) (x1 , y)
(x, y + y1 ) (x, y) (x, y1 )
(ax, y) a(x, y)

(x, by) b(x, y)

101

where x, x1 V1 ; y, y1 V2 and a, b k. W is independent of V3 .


Also if f is a bilinear function on V1 V2 , its extension f on V vanishes
on W. Furthermore if f is any linear function on V vanishing on W,
its restriction to V1 V2 is a bilinear function. It is then clear that the
space of bilinear functions on V1 V2 is isomorphic to the space of linear
functions on V which vanish on W. If is any linear function on V/W
into V3 and the natural homomorphism of V on V/W then is a
linear function on V vanishing on W. Also is an isomorphism

1. Kronecker product of Vector spaces

89

of the space of linear functions on V/W into V3 on the space of linear


functions on V vanishing on W. is clearly a bilinear function on V1 V2
into T = V/W and furthermore, by definition of V, V/W is generated by
(V1 V2 ). Thus T is the required space.
By taking k itself as a vector space over k, we have
Theorem 2. The space of bilinear functions on V1 V2 into k is isomorphic to the dual of the Kronecker product T .
x k V2 the kronecker product space. When there is
We denote by V1
no doubt about the field over which the kronecker product is taken we
x 2.
will simply write V1 V
x 2 we denote by x y
x the element in T which correIf T = V1 V
sponds to (x, y) by the bilinear function on V1 V2 into T . Since
(V1 V2 ) generates T , every element of T is of the form
X
x
ax,y (x y)
axy k.
Clearly

x = o y
x = o o
x
x o

x
x = x y
x + x1 y
(x + x ) y

x + y1 ) = x y
x + x y
x 1

x (y

x = a(x y)
x
ax y

x
x by = b(x y)
x

with obvious notations.


102
From our considerations it follows that in order to define a linear
x 2 it is enough to define it on elements of the type
function on V1 V
x
x y. Also for every such linear function, there is a bilinear function on
V1 V2 .
x y x
x of V1 V
x 2 to V2 V
x 1
It is easy to see that the mapping x y
is an isomorphism onto.
Suppose now that V1 and V2 are duals of V1 and V2 respectively over
k. If V1 and V2 and we define for (x, y) V1 V2 the function
.(x, y) = x.y,
then . is a bilinear function on V1 V2 into k. We shall now prove the

5. Composite extensions

90

Theorem 3. If {x } is a base of V1 over k and {y } a base of V2 over k,


x } is a base of V1 V
x 2 over k.
then {x y
x } are linearly independent over k. For
Proof. We first prove that {x y
P
x
all a (x y ) = o for a k, a = o for all but a finite number of
, . By the method of constructing the tensor product, it follows that
P
a (x , y ) is an element of W. Let and be elements of V1 and V2
defined respectively by
(x ) = 1

if =

= o otherwise
(y ) = 1
=o

if

otherwise

103


Then is a bilinear function on V1 V2 and hence vanishes on
W. Thus
X

x ) = o.
.
a (x y
But the left side equals a . Thus all the coefficients a vanish.
x 2 is a linear combination of elements of
Next any element of V1 V
P
P
x x V1 , y V2 . But then x = a x and y = b y so
the type x y,
that

X
 X
x
x =
x y
b y
a x
P
x ). Our theorem is proved.
which equals a b (x y
We have incidentally the
Corollary. If V1 and V2 are finite dimensional over k then
x 2 = dim V1 dim V2 .
dim V1 V
Also since the dual of V1 is isomorphic in a natural manner with V1
when dim V1 is finite, we get

1. Kronecker product of Vector spaces

91

x k V2 is
Corollary . If V1 and V2 are finite dimensional over k then V1
isomorphic to the dual of the space of bilinear functions on V1 V2 into
k.
Let now A1 and A2 be two associative algebras over a field k. We
x k A2 of the vector spaces A1
can form the Kronecker product A = A1
and A2 over k. We shall now introduce a multiplication into A so as to
make it into an associative algebra.
In order to do this, observe that the multiplication defined has to be 104
a bilinear function on A A into A. Since A is generated by elements of
x it is enough to define this bilinear function on elements of
the type x y,
x 1 . Put
x and z1 = x1 y
the type (z, z1 ) in A A where z = x y
x y1 .
f (z, z1 ) = z z1 = x x1 y
x 1 is a bilinear function on A1 A2 into
Now since (x, y) xx1 yy
A, by our previous considerations z z z1 is a linear function on A into
A. Similarly z1 z z1 is a linear function on A. This proves that f is
bilinear and that the multiplication so defined distributes addition. That
the multiplication is associative is trivial to see.
A is called the Kronecker product algebra.
We obtain some very simple consequences from the definition.
) If e1 and e2 are respectively the unit elements of the algebras A1
x 2 is the unit element of A1 A
x 2.
and A2 then e1 e
x 2 is generated by elements x y,
x it is enough
For, since A = A1 A
x
x 2 ) = (e1 e
x 2 )(x y).
x
to verify (x y)(e
But this is trivial.
1 e
) If A1 has x1 , . . . , xm as a base over k and A2 has y2 , . . . , yn as a
x j ) is a base of A1 A
x 2 over k. Furthermore if the
base over k then (xi y
multiplication tables for the bases are
X
xi x j =
a(t)
i j xt
yi y j =

t
X

b(ti j ) yt

then
x q )(xr y
x s) =
(x p y

X
,

105
()

x )
a()
pr bqs (x y

5. Composite extensions

92

) If A1 and A2 have unit elements e1 and e2 respectively then the


mappings
x 2
x x e
x
y e1 y
are isomorphisms of A1 and A2 into A. Thus A contains subalgebras
isomorphic to A1 and A2 .
An important special case is the one where one of the algebras is a
field. Let A be an algebra over k and K an extension field of k. Let A have
unit element e1 and K the unit element e2 . Form the Kronecker product
x over k. Then A K
x contains subalgebras A1 and K1 isomorphic to
A K
x in A K
x we have
A and K respectively. For any x t
x = x e
x 2 .e1 t
x = e1 t.x
x e
x 2
x t
x is of the form
so that A1 and K1 commute. Also every element of A K
P
x ) where {x } is a base of A over k and {t } a base of K over
a (x t
k. But this expression can be written
XX

x ) (x e
x 2 ).
a (e1 t

106

x is an algebra over K1 with the base {x e


x 2 }.
This shows that A K
x can be considered as
If we identify A1 with A and K1 with K then A K
x over
an algebra over K, a basis of A over k serving as a base of A K
x is then called the algebra got from A by extending the ground
K. A K
field k to K. We shall denote it by AK .
It is clear that AK is commutative if and only if A is a commutative
algebra.
Note. Even if A is a field over k, AK need not be afield over K.
Let be the rational number field and o = ( d) the quadratic
field over . Let be the real number field and consider the Kronecker
x over . Let e1 , e2 be a basis of o over with the
product A = o
multiplication table,
e21 = e1 ,

e1 e2 = e2 e1 = e2 ,

e22 = de1 .

2. Composite fields

93

The elements of o are of the form ae1 + be2 , a, b in . The elements

of A are of the form ae1 + be2 with a, b .


x is not an integrity domain. For,
Let first d > o. Then o

( de1 + e2 )( de1 e2 ) = o.
It can however be seen that A is then the direct sum of the two fields
and where
e2
1
= (e1 + ),
2
d

e2
1
= (e1 ).
2
d

Let d < o. Then A is a field. For if ae1 + be2 , o, then a2 db2 , o.


a
b
Put f = 2
,g= 2
. Then
2
a db
a db2
(ae1 + be2 )( f e1 + ge2 ) = e1 .

2 Composite fields
Let K1 and K2 be two extension fields of k. Suppose K1 and K2 are
both contained in an extension field of k. Then the composite of K1
and K2 is the field generated over k by K1 and K2 . In general, given 107
two fields K1 and K2 which are extensions of k, there does not exist an
extension field containing both. Suppose, however, there is a field
/k which contains k-isomorphic images K1 , K2 of K1 and K2 , then
a composite of K1 and K2 is defined to be the fields k(K1 K2 ). A
composite extension of K1 and K2 is therefore given by a triplet (, , )
consisting of 1) and extension field of k and 2) isomorphisms ,
of K1 and K2 respectively in which are identity on k. The composite
extension is then k(K1 K2 ). We wish to study these various composites
of K1 and K2 .
If is another extension of k and , two k-isomorphisms of

K1 and K2 respectively in then k(K1 K2 ) is another composite


extension. We say that these two composite extensions are equivalent if

there exists a k-isomorphism of k(K1 K2 ) on k(K1 K2 ) such that


= , = .

5. Composite extensions

94

108

Obviously this is an equivalence relation and we can talk of a class


of composite extensions.
If k(K1 K2 ) is a composite extension we denote by R(K1 K2 )
the ring generated over k by K1 and K2 in . This ring is, in general,
different from k(K1 K2 ).
x 2 be the Kronecker product of K1 and K2 . K
Let now K = K1 K
contains subfields isomorphic to K1 and K2 . We shall identify these
subfields with K1 and K2 respectively. Let now k(K1 K2 ) be a composite extension and R(K1 K2 ) the ring of the composite extension.
Define the mapping of K into R(K1 K2 ) by
X
 X
x

axy (x y)
=
axy x y ,

where axy k. Then coincides with on K1 and with on K2 .


Since and are isomorphisms, it follows that is a k-homomorphism
of K on R(K1 K2 ). Since is a field, it follows that kernel of the
Thus
homomorphism is a prime ideal Y of K.

K/G
R(K1 K2 ).

If k(K1 K2 ) is another composite extension, then defined ear

lier, is an isomorphism of R(K1 K2 ) on R(K1 K2 ). Consider the


homomorphism defined above. Define on K by = . We have
X 

X
 X

x
=
(

axy x y)
axy x y =
axy x y .

Then is a homomorphism of K on R(K1 K2 ). But, since is an


isomorphism, it follows that has G as the kernel. Thus the prime ideal
G is the same for a class of composite extensions.
Conversely, if two composite extensions correspond to the same
it can be seen that they are equivalent.
prime ideal of K,
We have, now, only to prove the existence of a composite extension
Let G be a prime ideal of K and
associated with a prime ideal of K.
Since K has a unit element, a prime ideal G , K always exists.
G , K.
Let A be the integrity domain

A = K/G

2. Composite fields

95

109

Let be the natural homomorphism of K on A. Then K1 and K2


x
are subfields of A. Since K is generated by elements of the type x y,
K1 and K2 , are different from zero. Clearly A = R(K1 K2 ). Hence
the quotient field of A is a composite extension. Hence for every prime
there exists a composite extension. We have hence proved
ideal G , K,
the
Theorem 4. The classes of composite extensions of K1 and K2 stand
in (1,1) correspondence with the prime ideal G , K of the Kronecker
product K of K1 and K2 over k.
Consider now the special case where K2 /k is algebraic. Let k(K1
K2 ) be a composite extension. Then
k(K1 K2 ) R(K1 K2 ) K1 .
Since every element of K2 is algebraic over k, k(K1 K2 ) is algebraic
over K1 . This means that R(K1 K2 ) is a field and so coincides with
k(K1 K2 ). Thus
Theorem 5. If K/k is algebraic, then every prime ideal G , K of K is
a maximal ideal.
Let K/k be an algebraic extension and L/k any extension. The Krox k L is the extended algebra (K)L of K by extendnecker product K = K
ing k to L. K is thus an algebra (commutative) over L. If G is a prime
then by above, it is a maximal ideal and K/G

of K,
gives a compos
ite extension. Since K is an algebra over L we may regard K/G
as an
extension field of L.
none of them 110
Let now G1 , . . . , Gm be m distinct maximal ideals of K,

equal to K. Let Li = K/Gi be a composite extension. Form the direct


sum algebra
X
Li
i

as a commutative algebra over L. We shall now prove

5. Composite extensions

96
Theorem 6.
P
T Gi
i K/Gi K/
i

P
Proof. We shall construct a homomorphism of K on i Li and show
T

that the kernel is Gi .
i

In order to do this let us denote by i the natural homomorphism of


then i x Li .
K on Li (this is identity on L), i = 1, . . . , m. If x K,

Define on K by
X
(x) =
i x.
i

That this is a homomorphism on K is easily seen; for, if x, y K


X
X
X
(x + y) =
i (x + y) =
i x +
i y = x + y.
i

(xy) =

i (xy) =

X
i

Xi
X
(i x)(i y) = ( i x)(
i y) = xy.
i

The kernel of the homomorphism is set of x such that x = o. Thus


i x = o so that x Gi for all i. Hence x i Gi . But every y i Gi has
T
the property y = o. Thus the kernel is precisely Gi .
i

111

We have only to prove that the homomorphism is onto.


To this end, notice that for each i, i = 1, . . . , m, there is a bi K with

G j , j , i
bi

< Gi

For, since Gi and G j , j , i are distinct, there is a j G j which is


Q
not in Gi . Put bi = a j . Then bi satisfies above conditions since Gi is
i, j

maximal.
P
Let now ci be an element in the direct sum, ci Li . By definition
i

of bi

= o
j bi

, o

if j , i
if j = i

3. Applications

97

Since Li is a field, there exists xi , o in Li such that


xi i bi = oi .
i being a homomorphism of K on Li , let yi K with i yi = xi . Put
X
c=
bi yi
i

Then
(c) =

XX
i

j (bi yi ) =

ci

which proves the theorem completely.


Suppose in particular K/k is finite. Then KL has over L the degree
K/G

(K : k). Since, for a maximal ideal, G , K,


has over L at most the
degree (K : k), we get
i : L) (K : k) i = 1, . . . , m.
1 (K/G
This means that K has only finitely many maximal ideals and
X
TG

K/
K/G
G

the summations running through all maximal ideals of K.


112
Thus K and L have over k only finitely many inequivalent composite
extensions.

3 Applications
Throughout this section will denote an algebraically closed extension
of k and K and L will be two intermediary fields between and k. A
composite of K and L in will be the field generated over k by K and
L. It will be denoted by KL.
Let K be the Kronecker product over k of K and L. There is, then, a
homomorphism of K on R(K L) given by
X
 X
x

axy (x y)
=
axy x y.

5. Composite extensions

98

Suppose that this homomorphism is an isomorphism of K onto R(K


L). This means that
X
X
x = o
axy (x y)
axy x y = o

or that every set of elements of K which are linearly independent over k


are also so over L. Incidentally, this gives
K L = k.

113

Conversely, suppose K and L have the property that every set of


elements of K which are linearly independent over k are also so over
L. Then the mapping is an isomorphism of K on R(K L). For, if
P
axy xy = o we express x and y in terms of a base of L/k and a base of
K/k giving
X
b x y = o

But this means all b are zero. Therefore is an isomorphism.


It shows that every set of elements of L which are linearly independent over k are also so over K.
We call two such fields L and K linearly disjoint over k. Note that
L K = k. We deduce immediately
1) If L and K are are linearly disjoint over k then any intermediary
field of K/k and any intermediary field of L/k are also linearly disjoint.
Suppose now that K/k is algebraic. Then every prime ideals of K is
maximal. Let, in addition, K and L be linearly disjoint over k. Since K
is isomorphic to R(K L), it follows that (o) is a maximal ideal. Hence
K is a field. Thus
2) If K/k is algebraic and K and L are linearly disjoint over k, there
exists but one class of composite extensions of K and L over k.

Let K/k be a finite extension. Then, for some maximal ideal G , K/G

is isomorphic to KL. Since K/G may be considered as an extension field

of L, we get (K/G
: L) (K : k), that is
(KL : L) (K : k)
Clearly if G = (o), equality exists, and then K and L are linearly
disjoint over k. The converse is true, by above considerations. Hence,
in particular,

3. Applications

99

3) If (K : k) = m and (L : k) = n, then
(KL : k) mn;
equality occurs if and only if K and L are linearly disjoint over k.
114
We now consider the important case, K/k galois. By the considerations above, it follows that KL/k is algebraic over L. Since is a
algebraically closed, it contains the algebraic closure of KL. Let be
an isomorphism of KL in , which is identity on L. Its restriction to K
is an isomorphism of K in . But contains the algebraic closure of
K and hence K = K. Since KL is generated by K and L, it follows
that KL KL. Hence KL/L is a galois extension. (KL/L is already
separable since elements of K are separable over k).

KL
rr
r
r
rrr
K MMM
MMM
M

LL
LL
LL
L
L
r
rrr
r
r
r

KL
k
We now prove the

Theorem 7. If K/k is a galois extension and L, any extension of k then


G(KL/L) G(K/K L)
Proof. In the first place, we shall prove that K and L are linearly disjoint
over K L. For this, it is therefore enough to prove that every finite set
y1 , . . . , ym of elements of L which are linearly independent over K L,
are also so over K.

P
If possible, let y1 , . . . , ym be dependent over K so that xi yi = o,
xi K. Let us assume that y1 , . . . , ym are dependent over K but no

5. Composite extensions

100

proper subset of them is linearly dependent over K. Therefore, all the xi


are different from zero.
We may assume x1 = 1. Let be an element of G(KL/L). Then
115
X
X
0 = (
xi yi ) =
xi yi
i

By subtraction we get, since 1 = 1,


X
0=
(xi xi )yi
i,1

This means that xi = xi , i = 2, . . . , m. But, since is arbitrary in


G(KL/L), it follows that xi L. But xi K. Thus y1 , . . . , ym are linearly
dependent on K L which is a contradiction. Hence K and L are linearly
disjoint over K L.
Therefore, KL is isomorphic to the Kronecker product of K and L
over K L.
Suppose is any L-automorphism of KL. Its restriction to K is
an automorphism of K and leaves K L fixed. Consider the mapping

of G(KL/L) into G(K/K L). This is clearly a homomorphism.


If
is identity element of G(K/K L), then is identity on K. Since it is
already identity on L, it is identity on KL. Thus, G(KL/L) is isomorphic
to a subgroup of G(K/K L). To see that this isomorphism is onto
G(K/K L), let G(K/K L). Any element of KL may be written
(since K and L are linearly disjoint) in the from,
X
x y

where x are linearly independent elements of K over K L and y L.


This expression is unique. Extend to in G(KL/L) by defining

X
X

x y =
y (x ).

116

This is well defined; for, if y (x ) = 0, then, since {x } are


linearly independent over K L, {(x )} are also linearly independent

3. Applications

101

over K L and since K and L are linearly disjoint over K L, all y are
zero. Thus is an automorphism of KL/L.
Our theorem is thus proved.
In particular, if K and L are both galois extensions of k and K L = k
then, by above,
G(KL/L) G(K/k)

G(KL/K) G(L/k)

Also, since KL/k is algebraic, let be an isomorphism (trivial on


k) of KL in . Since its restrictions on K and L are auto morphisms, it
follows that KL/k is a galois extension. We now prove
Theorem 8. G(KL/k) is isomorphic to the direct product of G(K/k) and
G(L/k).
Proof. G(K/k) and G(KL/L) are isomorphic and for every element in
G(K/k), by the previous theorem, we have the extension ,
an element
of G(KL/L) determined uniquely by . Similarly, if G(L/k),
denotes its unique extension into an element of G(KL/k).
KL
K

tt
tt
tt

L
ttt
t
t
ttt


We now consider the mapping


(, )

of the direct product G(K/k) G(L/k) into G(KL/k).


Suppose is an element of G(KL/k). Its restriction to K is an 117
element of G(K/k). Consider 1 . Now 1
1 is identity on K. By the

5. Composite extensions

102

isomorphism of G(KL/k) and G(L/k), this defines a unique element 1


of G(L/k). Hence
= 1 1 .
Thus the mapping above is a mapping of the direct product G(K/k)
G(L/k) onto G(KL/k). We have only to prove that is a homomorphism
to obtain the theorem. It is clearly seen that is identity if and only if
1 and 1 are identity.
Let , be two elements of G(K/k) and , in G(L/k). Let and
be the unique elements in G(KL/k) defined by
=
,

= , .

Consider to element . Its restriction to K is and its restriction


to L is . Thus
=
which proves that the mapping is a homomorphism.
The theorem is now completely proved.

Chapter 6

Special algebraic extensions


1 Roots of unity
118

Consider the polynomial xm 1 in k[k], where k is a field. Let k have


characteristic p. If p = o, then xm 1 is a separable polynomial over
k, whereas if p , o, the derivative of xm 1 is mxm1 which is zero,
if p divides m. If p m, then xm 1 is a separable polynomial over
k. Therefore, let m > o be an arbitrary positive integer, if k has characteristic zero and m, an integer prime to p, if k has characteristic p , 0.
Then xm 1 is a separable polynomial over k and it has m roots in , an
algebraic closure of k. We call these m roots, the mth roots of unity.
If and are mth roots of unity then m = 1 = m . Therefore
()m = m m = 1, (1 )m = (m )1 = 1 which shows that the mth roots
of unity from a group. This group Gm is abelian and of order m.
Let now d/m. Any root of xd 1 in is also a root of xm 1.If is
an mth root of unity such that d = 1, then is a root of xd 1. Since
xd 1 has exactly d roots in , it follows that Gm has the property that
for every divisor d of m, the order of Gm , there exist exactly d elements
of Gm whose orders divide d. Such a group Gm is clearly a cyclic group.
Hence
The mth roots of unity form a cyclic group Gm of order m.
119
There exists, therefore, a generator of Gm . is called a primitive
mth root of unity. Clearly, the number of primitive mth roots of unity is
103

6. Special algebraic extensions

104

(m). All the primitive mth roots of unity are given by a with 1 a <
m, (a, m) = 1, a fixed primitive mth roots of unity.
Let be an algebraically closed field. Let m and n be two positive
integers which are arbitrary, if has characteristic zero and, prime to p,
if has characteristic p , o. If is an mth root of unity and an nth
root of unity, then
()mn = mn mn = 1,
so that is an mn th root of unity. This shows that the roots of unity in
form a group H(). We now determine the structure of H.
Theorem 1. If has characteristic zero, then H is isomorphic to the
additive group of rational numbers mod 1, whereas, if has characteristic p , o, H is isomorphic to the additive group of rational numbers
a
, (a, b) = 1, p b, mod 1.
b
Proof. Let R denote the group (additive) of rational numbers and 1 <
2 < 3 the sequence of natural numbers, if has characteristic
zero, whereas, if has characteristic p , o, let R denotes the rational
a
numbers , (b, a) = 1, p b and 1 < 2 < 3 the sequence of
b
natural numbers prime to p. Put
n = 1 n .
120

Denote, by Hn , the group of n the roots of unity in . Since every


integer m(p m, if has characteristic p) divides some n , if follows
that
[
H=
Hn .
n

Since Hn is cyclic, we can choose a generator n of Hn is such a way


that
n+1
n = n+1
.

2. Cyclotomic extensions

105

Any x in R may be written as


mapping as follows

a
where a is an integer. Define the
n

x = an ,
so that is a function on R with values in H. The mapping is well
b
, then m a = n b. Suppose m > n; then
defined: for if x =
m
b = an+1 m
n+1 m
= an by choice of n . We now verify that is a
so that bm = a
m
a
b
homomorphism of R on H. If x , y =
are in R and m n, then
n
m

(x + y) = (

an+1 m + b
n+1 m +b
) = a
m
m

which equals an bm = x y. Also, since any root of unity is in some


a
Hn , it is of the form an so that, for x = , x = an . We have, therefore,
n
to determine the kernel of the homomorphism. It is the set of x in R such
that
x = 1.
a
If x = , then 1 = x = an so that |n |a and so x is an integer. The 121
n
converse being trivial, it follows that the kernel is precisely the additive
group of integers and our theorem is established.

2 Cyclotomic extensions
Let k be a field and xm 1 a separable polynomial in k[x]. This implies,
in case k has characteristic p , o, that p does not divide m. Let
be a primitive mth root of unity in , an algebraic closure of k. Then
K = k() is the splitting field of xm 1 in . Therefore, K/k is a
separable, normal extension. Let G be its galois group. If G, is
determined by its effect on . Since is a primitive mth root of unity, so
is . For,
()m = (m ) = 1;

6. Special algebraic extensions

106

so is a root of unity. Also, if ()t = 1, then (t ) = 1. Since is


an automorphism, it follows that t = 1 or m/t. Thus
= , (, m) = 1.
If , are in G, let = , (, m) = 1 and = (, m) = 1.
Then
() = () = ( ) =
which shows that = or that G is abelian.
Consider now the mapping
g:

122

where = , (, m) = 1. This is clearly a homomorphism of G into


the multiplicative group prime residue classes mod m. The kernel of
the mapping g is set of with = .
But then t = t for all t K, so that by, galois theory, is the
identity.
Let us call extension K = k() a cyclotomic extension of k. We then
have proved the
Theorem 2. The Cyclotomic extension k()/k is an abelian extension
whose galois is isomorphic to a subgroup of the group of prime residue
classes mod m where m = 1 and is a primitive mth root of unity.
Let be the prime field contained in k. Then () is a subfield of
k() = K. Let G be the galois group K/k.
K = k()
k

qq
qqq
q
q
q

p
ppp
p
p
pp

()

Let be in G and
the restriction of to (). Since is identity
on k and is again a primitive root of unity, it follows that
is an

2. Cyclotomic extensions

107

automorphism of ()/. It is easy to see that the mapping


is an
isomorphism of G into the galois group of ()/.
We shall therefore confine ourselves to studying the galois group of
()/.
First, let be the rational number field and a primitive mth root of
unity. Let () be the cyclotomic extension. Let f (x) be the primitive
integral polynomial which is irreducible in [x] an which satisfies.
Then f (x) is a monic polynomial. For, since f (x) divides xm1 ,
xm 1 = f (x)(x).
a
(x) has rational coefficients and so (x) = 1 (x) where 1 (x) is a 123
b
primitive integral polynomial and a and b are integers. From the the
theorem of Gauss on primitive polynomials it follows that f (x) is monic.
Let p be a prime not dividing m. Let (x) be the minimum polynomial (which is monic and integral) of p . We assert that f (x) = (x).
For, if not, f (x) and (x) are coprime and so
xm 1 = f (x) (x) h(x).
for some monic integral polynomials h(x).
Consider the polynomial (x p ). It has as a root and so f (x) divides
(x p ). Hence
(x p ) = f (x)g(x),
g(x), again, a monic and integral polynomial. Considering the above
mod p we get
f (x)g(x) (x p ) ((x)) p (mod p)
so that f (x) divides ((x)) p ( mod p). If t(x) is a common factor of
f (x) and ((p)( mod p) then (t(x))2 divides xm 1 mod p, which is
impossible, since p m and xm 1 does not have x as a factor. Thus our
assumption f (x) , (x) is false.
This means that, for every prime p m, p is a root of f (x). If
(, m) = 1, then = p1 p2 p1 where p1 , . . . , p1 are primes not dividing

6. Special algebraic extensions

108

m. By using the above fact successively, we see that, for every , (, m) =


1, is a root of f (x). Therefore
Y
m (x) =
(x )
(1)
(,m)=1

124

divides f (x). But m (x) is fixed under all automorphisms of ()/ so


that f (x) = m (x). We have proved the
Theorem 3. If is the field of rational numbers and is a primitive mth
root of unity, then the galois group of the cyclotomic extension ()/
is isomorphic to the group of prime residue classes mod m. The irreducible polynomial m (x) of is given by (1).
m (x) is called the cyclotomic polynomial of order m. Its degree is
(m). In order to be able to obtain an expression for m (x) in terms of
polynomials over , we proceed thus.
We introduce the Mobius function defined as follows:
It is a function (n) defined for all positive integers n such that
1) (1) = 1
2) (p1 pt ) = (1)t where p1 , . . . , pt are distinct primes.
3) (m) = o if p2 /m, p being a prime.
From this, one deduces easily
4) (m) (n) = (mn) for (m, n) = 1.

For, one has to verify it only for m = p1 pt , n = q1 q1 where


p s and q s are all distinct primes. Then (m) = (1)t , (n) = (1)1
and (mn) = (1)t+1 .
We now prove the following simple formula

5)
X

(d) = o if m > 1

d|m

= 1 if m = 1
125

the summation running through all divisors d of m.

2. Cyclotomic extensions

109

If m = 1, the formula reduces to (1). So, let m > 1. Let m =


pa11 pat t be the prime factor decomposition of m. Any divisor d of m
is of the form pb11 pbt t where o bi ai , i = 1, . . . , t. In view of (3),
it is enough to consider divisors d of m for which o bi 1. In that
case,
t
X
X
(d) =
(ti )(1)i = o.
i=o

d|m

Let now f (n) be a function defined on positive integers, with values


in a multiplicative abelian group. Let g(n) also be such a function. We
then have the Mobius inversion formula,
Q
Q
n
f (d) = g(n) f (n) = (g(d))( d )
d|n

Suppose

Q
d|n

d|n

f (d) = g(n). Then


Y

( ng )

(g(d))

YY
di | dn

d|n

d|n

(d)
f (d1 )
.

Changing the order products, we get


Y
X
(d);
f (d1 )
d1 |n

d| dn

using formula (5), we obtain the inversion formula. The converse follows in the same way.
Consider now the integers mod m. Divide them into classes in the
following manner. Two integers a, b are in the same class if and only if
(a, m) = (b, m).
Let d/m and Cd , the class of integers a ( mod m) with (a, m) = d. 126
m
m
Then a is of the form d where (, ) = 1. Thus Cd has ( ) elements.
d
d
The classes Cd , for d|m, exhaust the set of integers mod m. If is a
primitive mth root of unity, then
xm 1 =

m
Y
(x t ).
t=1

6. Special algebraic extensions

110

In view of the above remarks, we can write



YY
t
m
(x ) .
x 1=
tCd

d|m

m
m
But if t Cd , t = d , (, ) = 1 and so t is a primitive th root
d
d
of unity. Using the definition of m (x), if follows that
Y
xm 1 =
d (x)
d|m

By the inversion formula we get


m (x) =

d|m

(xd 1)( d )

Comparison of degrees on both sides gives the formula


(m) =

X
d|m

X (d)
m
d( ) = m
.
d
d
d|m

We may compute m (x) for a few special values of m. Let m = p, a


prime number. Then
p (x)

xp 1
= x p 1 + + x + 1.
x1

If m = pq, the product of two distinct primes, then


pq (x) =

127

(x pq 1)(x 1)
.
(x p 1)(xq 1)

Let us now consider the case when is the prime filed of p elements.
Obviously, ()/ is a cyclic extension. If we define the cyclotomic
polynomial as before, it is no longer irreducible over [x]. For instance,
let p = 5 and m = 12. Then
12 (x) =

(x12 1)(x2 1)
= x4 x2 + 1.
(x6 1)(x4 1)

2. Cyclotomic extensions

111

Also
x4 x2 + 1 = (x2 2x 1)(x2 + 2x 1) (mod 5).
Therefore, ()/ has degree < (m). It is obvious since ()/
is cyclic, that, for m (x) to be irreducible, the group of prime residue
classes mod m should be cyclic. We shall prove
Theorem 4. If is the prime filed of characteristic p , o and p m,the
cyclotomic polynomial m (x) is irreducible, if and only if the group of
prime residue classes mod m is cyclic and p is a generator of this cyclic
group.
Proof. We already know that ()/ is cyclic. If m (x) is irreducible,
then ()/ has order (m). Let be the Frobenius automorphism of
()/. Then
= p ,
The (m) automorphisms

p being the number of elements in .


1, , 2 , . . . , (m)1 are distinct. Hence
2

(m)1

, p , p , . . . , p

are all distinct, which means 1, p, p2 , . . . , p(m)1 are distinct mod m.


Therefore, p is a generator of the multiplicative group of prime residue
classes mod m.

The converse is trivial.
The theorem is true, if, instead of being the prime filed of p ele- 128
ments, is a finite filed of q elements. Then q has to be a generator of
the group of prime residue classes mod m.
If Rm denotes the group of prime residue classes mod m, then Rm
is cyclic, if and only if
m = 2a , a = 1, 2

or

m = qb

or

2qb ,

where q is a prime. Thus m has necessarily to have one of these forms.


We now use the irreducibility of m (x) over the rational number field
to prove a theorem of Wedderburn.

6. Special algebraic extensions

112

Theorem 5. A division ring with a finite number of elements is a filed.


Proof. Let D be the division ring and k its centre. Then k is a field. D
being finite, let k have q elements. If D is of rank n over k, then D has
qn elements. We shall prove that n = 1.

Let D1 be a subalgebra of D over k. Let D1 have rank m over k.
Then D1 has qm elements. But D1 is a sub group of D so that qm 1
divides qn 1. This means that m|n. For, if n = tm + , o < m; then
qn 1 = q (qtm 1) + (q 1)

129

so that qm 1|q 1 which cannot happen unless = 0. Thus every


subalgebra of D has rank d dividing n.
Let x D. Consider the y D such that xy = yx. They form a
subalgebra over k. Therefore, the number of y in D which commute
with x form a group of order qd 1, for some d dividing n. This group
is the normaliser of x. Hence, the number of distinct conjugates (in the
sense of group theory) of x in D is
qn 1|qd 1.
For the finite group D , we have
D = k +

Dx ,

Dx

is the set of all conjugates of x. Comparing number of elewhere


ments on both sides
X
qn 1 = q 1 +
qn 1/qd 1
d

for some divisors d of n.


Since n (x)|xn 1, we see that n (q), which is an integer, divides
n
q 1. Also n (x) divides xn 1|xd 1 for any d|n, d , n. Therefore,
n (q) divides q 1. But
n (q) > (q 1)(n)
which shows that n 1.
This proof is due to Ernst Witt.

3. Cohomology

113

3 Cohomology
Let G be a finite group and A an abelian group on which G acts as a
group of left operators. Let A be a multiplicative group. We denote elements of G by , , , . . . , and elements of A by a, b, c, . . .. We denotes
by a the effect of on a. Then
(ab) = a b
(a ) = a
130

Let 1 be the unit element of A. Denote by Gn , n 1 the Cartesian


product of G with itself n times.
A function on Gn with values in A is said to be an n dimensional
Cochain or, simply, an n cochain. This function f (x1 , . . . , xn ) has values
in A. We denote by a 1 , . . . , n the element in A which is the value
taken by the n cochain for values 1 , . . . , n of its variables. We denote
the function also, by a1 , . . . , n . If a1 , . . . , n and b1 , . . . , n are
two functions, we define their product by
o1 ,...,n = a1 ,...,n b1 ,...,n .
Similarly
o1 ,...,n = (a1 ,...,n )1
is called the inverse of a1 ,...,n . With these definitions, the n cochains
form a group C n (G, A) or simply C n .
We define C o (G, A), the zero dimensional cochains, to be the group
of functions with constant values, that is functions a on G such that
a = a
is the same for all G. It is then clear that C o is isomorphic to A.
We now introduce a coboundary operator in the following manner.
(= n ) is a homomorphism of C n into C n+1 defined by
a1 ,...,n+1 = a1 ,...,n

6. Special algebraic extensions

114
= a12 ,...,n+1

n
Y
n+1
i
(a1 ,...,i1 ,i i+1 ,...,n+1 )(1) (a1 ,...,n )(1) .
n=1

For n o, we define the groups Z n (G, A) and Bn+1 (G, A) in


the following manner. Z n (G, A) is the kernel of the homomorphism
n

C n C n+1 and Bn+1 (G, A) is the homomorphic image. The elements


of Z n (G, A) are called n dimensional cocycles or n-cocycles and the
elements of Bn+1 (G, A) are n + 1-dimensional coboundaries or n + 1
coboundaries. The homomorphism has the property
= identity

(2)

which proves that Bn+1 (G, A) is a subgroup of Z n+1 (G, A) and we can
form for every n > o the factor group H n (G, A) the n-dimensional cohomology group. We verify (2) only in case n = o and 1 which are the
ones of use in our work.
The coboundary of a zero cochain is a one cochain given by
a = a =

a
.
a

(3)

Its coboundary, by definition is


a = a, =

a a
.
a

Substituting from (3), we get a, = 1 which verifies (2) for n = o.


We define the zero coboundary, that is the elements in Bo (G, A), to
be the function with value 1 on all G. Thus Bo (G, A) consists only
of the identity. An element in Z o (G, A) will be the constant function a
with
a = a
132

for all . Hence


H o (G, A) is isomorphic with the set of a A with the property a =
a for all G.
A one dimensional cocycle is a function a for which a = 1. But
a =

a a
.
a

131

3. Cohomology

115

Therefore, the elements in Z (G, A) are functions a with


a a = a .
A one coboundary is, already, a function a of the form a /a. It is,
certainly, a one cocycle.
For our purposes, we shall need also the additive cohomology, instead of the multiplicative cohomology above. We regard now A+ as
an additive, instead of a multiplicative, group. Then C n (G, A+ ) is an
additive abelian group. We define the coboundary operator as
a1 ,...,n+1 = a1 ,...,n
= a12 ,...,n+1 +

n
X

(1)i ai ,...,i1 ,i ,i+1 ,...,n+1 + (1)n+1 ai ,...,n .

i=1

As before, a zero cochain is a constant function on G and its coboundary a is


a = a = a a.
A one cochain a is a cocycle if
a = 0
which is the same thing as
a + a = a .
Exactly as before, we see that H o (G, A), the zero dimensional additive cohomology group is isomorphic to the set of elements a in A+ with 133
a = a for all .
We now consider the case when G is a cyclic group. Let be a generator of G so that 1, , 2 , . . . , n1 are all the elements of G. Taking
multiplicative cohomology, if a is a one cocycle, then
a = a a
or a = a /a . Substituting for successively 1, , . . . , n1 we get
n1

a1+++

= 1.

6. Special algebraic extensions

116
n1

We denote a1+++

a is a cocycle, then,

by Na and call it the norm of a . Thus, if


Na = 1.

Conversely, suppose a is an element in A with


n1 =1

Na = a1++..+

= 1.

We can define a cocycle a such that a = a. For let = , for


some . Put
1
a = a1+++ .

Obviously, a = a. Also, a is a cocycle. For, if = .


1

a a = a1+++ (a1+++

1 + +++ 1

= a1+++
= a .
134

In a similar manner, for additive cohomology, we have a = a + a


where a is a cocycle. If G is cyclic, on substituting 1, , . . . , n1 for
, we get
n1
S a = a + a + + a = o.
We call S a the spur or trace of a . If a is any element of A, with
n1
trace S a = a + a + + a = o, then the cocycle
1

a = a + a + + a
where = , has the property
a = a.

We now apply the considerations above, in the following situation.


K/k is a finite galois extension with galois group G. Then G acts on
K and also the additive group K + as a group of operators. We might,
therefore, consider the cohomology groups H o (G, K ), H 1 (G, K ), . . .
and H o (G, K + ), H 1 (G, K + ) . . . etc. As before, H o (G, K ) and H o (G, K + )
are isomorphic to subgroups of K and K + with a = a for all . This,
by galois theory, shows

3. Cohomology

117

Theorem 6. H o (G, K ) is isomorphic to k and H o (G, K + ) is isomorphic


to k+ .
But what we are interested in, is the following important theorem
due to Artin.
Theorem 7. The group H 1 (G, K ) and H 1 (G, K + ) are trivial.
Proof. Let us first consider multiplicative cohomology. If a is a cocycle, we have to prove that it is a coboundary. The elements , , . . . of
G are independent k-linear mappings of K into , the algebraic closure.
Hence, if (a ) are elements of K ,
X
a

is a non-trivial k-linear map of K into . Therefore, there exists a , o 135


in K such that
X
a , o.

Put b1 =

a =

a . Then

(b1 ) =

a .

Therefore

a X
=
a a .
b

Since (a ) is a cocycle, we get


X
a X

=
a

=
a = b1 .

Thus
a =
which is a coboundary.

b
= b1
b

6. Special algebraic extensions

118

Consider now the additive cohomology. K/k being finite and separable, there exists a K such that
X
= S K/k = 1.

Put now
b =

a being an additive cocycle. Then


X
b =
a .

But a = a 1 =

a b =

136

so that

(a + a ) =

a = b

which proves that a = b b is a coboundary.


Our theorem is completely proved.
We apply the theorem in the special case where G is cyclic. Let
be a generator of G. If a is a cocycle then, in multiplicative theory
n1

a1+++

=1

or NK/k a = 1. Similarly in additive cohomology,


n1

a + a + + a

=o

or S K/k a = o.
Using theorem 7, we obtain theorem 90 of Hilbert.
Theorem 8. If K/k is a finite cyclic extension, , a generator of the
galois group of K/k and a and b two elements of K with NK/k a = 1 and
S K/k b = 0 respectively, then
a = c1
b = d d
for two elements c, d in K.

4. Cyclic extensions

119

4 Cyclic extensions
Let K/k be a cyclic extension of degree m. Put m = npa , (n, p) = 1 if p
is the characteristic of k; otherwise, let m = n.
Let G be the galois group of K/k. It has only one sub-group of order
pa . Let L be its fixed field. Then K/L is cyclic of degree pa and L/k is
cyclic of degree n prime to p. Let be a primitive nth root of unity and
k() the cyclotomic extension. The composite F = Lk() is cyclic over
k() and of degree prime to p. We shall see that K over L and F over
k() can be described in a simple manner.
We shall, therefore, consider the following case, first.
K/k is a cyclic extension of degree m and p m, if k has character- 137
istic p , o; otherwise, m is an arbitrary integer. Also, k contains all the
mth roots of unity. We then have the theorem of Lagrange.
Theorem 9. K = k(w) where wm k.
Proof. Let be a primitive mth root of unity. is in k.

Hence, since K/k has degree m,


NK/k = m = 1.
By Hilberts theorem, therefore, if is a generator of the galois
group of K/k, then there is an K such that
1 = .
2

Since is a primitive mth root of unity, , , , . . . are all distinct


and are conjugates of . Hence our theorem.
satisfies a polynomial xm a, a k. If K = k( ), where also
satisfies a polynomial xm b, b k,then
= + ,
where t is a primitive mth root of unity and so (t, m) = 1.
Now
!
t

= t
=
t
t t

6. Special algebraic extensions

120
which shows that
= t c,

c k.

We shall call xm a a normed polynomial. We have, then, the


Corollary . If k() = K = k( ), where and are roots of normed
polynomials, then
= t c,
138

where (t, m) = 1 and c k.


We consider the special case, m = q, a prime number. Let K/k be a
cyclic extension of degree q. Let K = k() and let 1 = (), 2 , . . . , q
be the irreducible polynomial of over k. Suppose is a generator of
the galois group of K/k.
Let notation be so chosen that
1 = 2 , 2 = 3 , . . . , q1 = q , q = 1 .
Since k contains the qth roots of unity and every qth root of unity
, 1 is primitive, we construct the Lagrange Resolvent.
= (, ) = 1 + 2 + + q1 q .
Then
= 2 + 3 + + q2 q + q1 1

which shows that = 1 . Hence K = k(). Also,


(q ) = q

which proves that q k and satisfies a normed polynomial.


In particular, if k has characteristic , 2, and K/k has degree 2, then

K = k( d)
for d k.
The polynomial xq a for a k is, thus, either irreducible and, then,
a root of it generates a cyclic extension, or else, xq a is a product of
linear factors in k.

4. Cyclic extensions

139

121

We study the corresponding situation when K has characteristic p ,


o and K/k is a cyclic extension of degree pt , t 1.
We first consider cyclic extensions of degree p.
Let K/k be a cyclic extension of degree p and , a generator of the
galois group of K/k. Let be a generic element of the prime field
contained in k.
Introduce the operator P x = x p x. Then
P(x + ) = P x.
Also, the only in k satisfying P = o are the elements of and
these are all the roots of P x = o.
The element 1 in k has the property
S K/k 1 = o,
so that by Hilberts theorem, there is an K such that
1 = .
Therefore, since K/k has degree p,
K = k().
In order to determine the polynomial of which is a root, consider
P.
(P) = P = P( + 1) = P
which shows that P k. If we put P = , then is a root the
irreducible polynomial
xp x
in k[x]. is a root of the polynomial and = + 1. Since is a
generator of the galois group of K/k, the roots of x p x are , +
1, . . . , + p 1.
A polynomial of the type x p x , k is called a normed
polynomial.
Suppose K = k( ) where also satisfies a normed polynomial. 140

6. Special algebraic extensions

122

Then , + 1, . . . , + p 1 are the roots of this normed polynomial.


So
= + h, o < h p 1.
Now h satisfies
( h) = h = + h h h = h
which shows that h k. We have, hence, the
Theorem 10. If K/k is cyclic of degree p and , a generator of the galois group of K/k, then K = k() where satisfies a normed polynomial
x p x in k[x] and = + 1. If K = k( ) and also satisfies a
normed polynomial, then
= + a,
and a k.
In order to be able to construct , we proceed thus. Let be an
element in K with
S K/k = 1.
Let 1 (= ), . . . , p be the roots of the irreducible polynomial which
satisfies over k. Construct the resolvent
= 1 + 22 + + (p 1) p1 + p p .
Let notation be so chosen that
1 = 2 , 2 = 3 , . . . , p1 = p , p = 1 .
Then
= 2 + 23 + + (p 1) p

141

and so = 1 + 2 + + p = 1. Therefore p = t for some


t in k. This gives the normed polynomial.
It should be noticed that, here, we use additive cohomology whereas,
in case K/k has degree m prime to p, we used multiplicative cohomology. Furthermore, in the second case, when K/k is of degree p, the

4. Cyclic extensions

123

elements of serve the same purpose as the roots of unity in the first
case.
If k has characteristic 2 and K/k is a separable extension of degree
2, then
K = k()
where 2 k.
Observe, also, that any polynomial x p x , for k, is either
irreducible over k and so generates a cyclic extension over k, or splits
completely into linear factors in k. For, if is a root of x p x then,
for , + is also a root.

Just as we denote a root of xm by m , we shall denote, for k,

by
, a root of x p x . It is obvious that
is p valued and if is
P
P

, all the values are


one value of
P
, + 1, . . . , + p 1.
We now study the case of cyclic extension of degree pn , n 1.
Let K/k be a cyclic extension of degree pn and , a generator of
the galois group G of K/k. Since G is cyclic of order pn , it has only
one subgroup of order p and hence, there exists a unique subfield L of K
such that K/L is cyclic of degree p and L/k is cyclic of degree pn1 = m.
Let us assume that n 2.
m is of order p and hence is a generator of the galois group of K/L. 142
Thus K = L() where
m = + 1
and satisfies P = L.
Put = . Then m = (m ) m = which shows that
L. Also,
m1
S L/k = + + + .
Substituting the value of , we get
S L/k = ( ) + (2 ) + + (m m1 )
= m = 1.

6. Special algebraic extensions

124
Hence has the property

S L/k = 1.
It is easy to see that is not k. For,
P = (P) P =
and in k would mean P = o or . This means that S L/k = o.
We now proceed in the opposite direction. Let L be cyclic of degree
pn1 > 1 over k. We shall construct an extension K which is cyclic over
k and contains L as a proper subfield. Let be a generator of the galois
group of L/k.
Let us choose L, such that
S L/k = 1.
Now S L/k (P) = P(S L/k ) = o which shows that
P =
143

for some in L. Also, is not in k. We claim that for k,


xp x

(4)

is irreducible over L[x]. For, if it is not irreducible, it is completely


reducible in L. Let be a root of x p x in L. Then
p = + .
This means that
P( ) = (P) P = = P.
Thus
P( ) = o
or = . Since S L/k () = o and S L/k = o, it follows that
S L/k = o, which is a contradiction. Thus for k, (4) is irreducible

4. Cyclic extensions

125

in L[x]. Let be a root of this irreducible polynomial for some . Then


K = L()/L is cyclic of degree p.
Let
denote an extension of to an isomorphism of K/k in , the
algebraic closure of k. Then
is not identity on L. Since P = +
and k, we get
(P)

= + .

Now
P(
) = (P)

P = = P
and, again, we have
P(
) = o
or that
= + + which shows that
is an automorphism of K/k.
If t is any integer,
t1

t = + + + +

+ t.
144

n1
,

2, . . . ,
p

This shows that 1,


have all different effects on ,
so that they are distinct automorphisms of K/k. But, since K/k has
degree pn , it follows that K/k is cyclic of degree pn . We have, hence,
the important
Theorem 11. If k is a field of characteristic p and admits a cyclic extension K of k containing L and of degree pm , m > n, for every m.
In fact, if k is an infinite field, we may very over k and obtain
an infinity of extensions K over k with the said property. It follows
theorem 10.
Corollary. If k is a field of characteristic p and admits a cyclic extension
of degree pn for some n 1, then its algebraic closure is of infinite
degree over it.

6. Special algebraic extensions

126

5 Artin-Schreier theorem
We had obtained, in the previous section, a sufficient condition on a field
so that its algebraic closure may be of infinite degree over it. We would
like to know if there are fields K which are such that their algebraic
closures are finite over them. The complete answer to this question is
given by the following beautiful theorem due to Artin and Schreier.
Theorem 12. If it an algebraically closed field and K is a subfield
such that
1 < ( : K) <
145

then K has characteristic zero and = K(i), where i is a root of x2 + 1.

Proof. The proof is as follows:

1) K is a perfect field. For if not K p and K p is of infinite


degree over K. This shows that /K is a finite separable extension.
Since it is is trivially normal, it is galois extension. Let n be the order
of the galois group G of /K.
2) If p is the characteristic of K, then p n (if p , o). For, if p/n, then
G has a subgroup of order p generated by an element . Let L be
the fixed field of this subgroup. Then /L is a cyclic extension of
degree p. By corollary of theorem 11 it follows that /L has infinite
degree. This is a contradiction.
Therefore, the order n of G is prime to the characteristic of K, if
different from zero.
3) Let q be a prime dividing n. Then q , p. Let L now be the fixed field
of a cyclic subgroup of G, of order q.
Then /L is cyclic, of order q. Now L contains the primitive qth root
of unity. For, if not since satisfies an irreducible polynomial of
degree q 1, it follows that L()/L has degree q 1. But
(L() : L)|( : L)
which means that L() = L. By theorem 9, therefore
= L(1 )

5. Artin-Schreier theorem

127

where 1 = L.
4) Any irreducible polynomial over L is either linear or of degree q. 146
For, if t is its degree and , a root of it then t = (L() : L) divides
q. Thus, every polynomial in L[x] splits in L into product of linear
factors and of factors of degree q.
2

Consider, in particular, the polynomial xq . In , we can write


2

xq = (x

q2

(5)

where runs through all q2 th roots of unity. Since ( : L) > 1, the


2
polynomial xq has, in L[x], an irreducible factor of degree q.
Since this factor is formed by q of the linear factors on the right of
(5), this factor is of the form

xq + + ,
being a q2 th root of unity. We assert that this is a primitive q2 th
root of unity. For if not, it is either 1 or a primitive qth root of unity.

Since q L, it would then follow that q L. Therefore we get


= L().
5) Let be the prime field contained in L. Consider the field ().
Let () contain the q th but not q+1 roots of unity. This integer
certainly exists. For, if L has characteristic zero = 2. If L has
characteristic p (and this is different from q, from (2)), then ()
contains p f elements where f is the smallest positive integer with
p f 1( mod q2 ). If q is the largest power of q dividing p f 1 then
() contains the q th, but not the q+1 the roots of unity.
147
Let be a primitive qv+1 th root of unity. Then satisfies a polynomial of degree q(irreducible) over L. Thus () is of degree q over
() L. Now, being a primitive q+1 th root of unity q is a q th
root of unity and so is contained in (). But all the roots of xq q
are qv+1 th roots of unity. Thus xq q is the irreducible polynomial

6. Special algebraic extensions

128
of over (). Hence
sss
sss

() O

OOO
OOO

() L

() L

LLL
LLL
L

oo
ooo
o
o
oo

and (() : ()) = q = (() : (() L).


If is a finite field, then (()/ is cyclic and it cannot have two
distinct subfields () and ()L over which it has the same degree.
Since < L, it follows that () and () L are distinct and so,
has characteristic zero. In this case, if q is odd, ()/ is cyclic and
the same thing holds. Thus q = 2. We know, then that = 2. If i
denotes a fourth root of unity, then
= L(i).
6) Form above, it follows that n = 2t for some t and K does not contain
the 4th root of unity. Now t = 1, for if not, let t > 1. Let M be a
subfield of L such that (L : M) = 2. Then ( : M(i)) = 2 and, by
above considerations, M(i) cannot contain 1 which is a contradiction.
So
K(i) = .
148

We shall make use of this theorem in the next chapter.

6 Kummer extensions
We now study the structure of finite abelian extensions of a field k. We
use the following notation:k is a field of characteristic p, not necessarily > o.
n is a positive integer not divisible by p if p , o, otherwise arbitrary.
, the group of non-zero elements of k. A generic element of will
also be denoted, following Hasse and Witt, by .

6. Kummer extensions

129

n is the group of nth powers of elements of .


, a subgroup of containing n and such that
( : n ) < .

(6)

Let k contain the nth roots of unity. We shall establish a(1, 1) correspondence between abelian extensions of k of exponent dividing n and
subgroups of satisfying (6).
Let K be an extension field obtained from k by adjoining to k the nth
roots of all the elements of . We shall obtain some properties of K.
Since /n is a finite group, let 1 , . . . , t in , form a system of
generators of mod n . Then any is of the form
ai n
= a1
1 , . . . , t

where a1 , . . . , at are integers. Put i = 1/n


i , i = 1, . . . , t. Then i is
uniquely determined up to multiplication by an nth root of unity, which
is already in k. This means that
K = k(1 , . . . , t ).
149

Therefore
1) K/k is a finite extension.
Each i is a root of a polynomial of the form xn i . This polynomial, by the condition on n, is separable over k. Also, xn i splits
completely in K. Thus
2) K/k is a finite galois extension and is the splitting field of the polynomial
t
Y
f (x) =
(xn i )
i=1

over k[x].
Let us denote by the group generated by 1 , . . . , t and . Let G
denote the galois group of K/k. In the first place,

6. Special algebraic extensions

130

3) / / n. (These are isomorphic groups).

Consider the homomorphism n of into itself. It takes into


n . The kernels in and are both the same. Since is taken to
by this homomorphism, it follows that / /n . Incidentally,
therefore, / is a finite group.
We shall now prove the important property,

4) G is isomorphic to /n .
(This proves that K/k is a finite abelian extension.)
In order to prove this, consider the pairing, (, ) of G and , given
by
(, ) = 1 ,
(7)
150

G, . Because of definition of ,
(, )n =

n
= 1.
(n )

Thus (, ) is an nth root of unity. Also,


!


(, ) = = = = = (, ) (, ).


Furthermore,
(, ) =



=
,
= (, )(, ).
( )

Thus (, ) defined by (7) is a pairing.


Let Go be the subgroup of G consisting of all with (, ) = 1, for
all . Since is generated by 1 , 2 , . . . , t , we get
i = i .
But 1 , . . . , t generate K. Therefore, = for all K.
By galois theory, = 1. Thus Go = (1).
Let o be the subgroup of all with (, ) = 1, for all . For
the same reason as before, o = . Thus, by theorem on pairing, G is
isomorphic to / . (4) is thus proved.

6. Kummer extensions

131

Observe, now, that since n , it follows the G is an abelian group


whose exponent divides n.

We will denote K symbolically by K = k( ).


We shall now prove
Theorem 13. Let K/k be a finite extension with abelian galois group G

of exponent dividing n, then K = k( n ) for a sub group with


/ n finite.
denote the group of non-zero elements of K with the prop- 151
Proof. Let
n
Then, by considering the homomorphism
erty for .
n , it follows that
/ / n,

n . We shall prove that K = k( n ).


where =
In order to prove this,
it suffices to prove that
G / n.
(8)

n
For, in that case, construct
n the field k( ). Because of the properties

But, by the previous results, (k( n ) :


of K, it follows that K k( ).
k) = order of / n = order of G.
Hence

K = k( n ).
We shall, therefore, prove (8).
define the function
Let G denote the dual of G. For every in ,
() = 1
Since n , it follows that () is an nth root of unity.
on G into .
Also,
!

= = () ()
() = =



so that is a character of G.

6. Special algebraic extensions

132

Let be any character of G. Since the exponent of G divides n,


n () = 1 for all . Therefore, () is an nth root of unity and hence is
an element of k. Also, since is a character of G,
() = ()()
152

which equals () (()) . Thus () is a one cocycle and, by theorem 7,


() = 1
for K. But (())n = 1 so that
n = (n )
that .

for all or n . This means, by definition of

Consider the mapping of into G . This is, trivially a


homomorphism. By above, it is a homomorphism onto. The kernel of
the homomorphism is set of for which () = 1 for all . That is
=
for all . By galois theory . But every element in satisfies this
condition. Thus is precisely the kernel, and
/ G .

Since G is a finite abelian group, G G and our theorem is completely proved.


We call a finite abelian extension K/k, a Kummer extension if
1) k contains the nth roots of unity, p n, if p(, o) is characteristic of
k,
2) K/k has a galois Galois group of exponent dividing n.What we have
then proved is
Theorem 14. The Kummer extensions of k stand in (1, 1) correspondence with subgroups of with / n finite.

7. Abelian extensions of exponent p

133

7 Abelian extensions of exponent p


153

We had introduced earlier the operator P which is defined by


P x = x p x.
Let k be a field of characteristic p and denote by k+ the additive
group of k. Then P is a homomorphism of k+ into itself. The
kernel of the homomorphism is precisely the set of elements in , the
prime field of characteristic p, contained in k.

, a root of the polynomial x p x .


If k, we denote by
P

Obviously
is p valued. Also,
P

+
=
+
.
P
P P

(9)

Let us denote by Pk+ , the subgroup of k+ formed by elements P,


k+ . Let be a subgroup of k+ with the properties,
k+ Pk+
/Pk+ is finite.
Let K be the extension field of k, formed by adjoining to k all the

, fro . We denote
elements
P
K = k(

).
P

We then obtain in exactly the same way as before


1) K/k is a finite abelian extension.
2) The galois group G of K/k is isomorphic with /Pk+ .
3) G is an abelian group of exponent p. For the proofs, we have to use
additive, instead of multiplicative, pairing and the property (9).

6. Special algebraic extensions

134

Suppose, now, on the other hand, K/k is a finite abelian extension of 154
exponent p, where p is the characteristic of k.
Then

K = k( ),
P
+
where is a subgroup of k with /Pk+ finite. For the proof of this,
we have to use additive instead of multiplicative, cohomology. For, let
G be the galois group of K/k and G its character group. Denote, by
the additive subgroup of K + formed by elements with P k+ .
,
Then
Denote, by , the group P .
/k+ /Pk+

Define () by

() = .

Then
P( ()) = P (P) ).
But, P k+ by definition. Hence, P( ()) = o.
Therefore, () is an element of . The rest of the proof goes
through in the same way and we have the
Theorem 15. If k has characteristic p , o, the finite abelian extensions
K/k of exponent p stand in a(1, 1) correspondence with subgroups of

).
k+ such that /Pk+ is finite, and then K = k( P

8 Solvable extensions

155

We propose to study now the main problem of the theory of algebraic


equations, namely the solution of algebraic equations by radicals
k will denote a field of characteristic p,(p = o or p , o), will be its
algebraic closure and n, n1 , n2 , . . . integers > o which are prime to p, if k
has characteristic p , o, otherwise arbitrary. An element will be
said to be a simple radical over k if n k, for some integer n. k()/k
is then said to be a simple radical extension. k()/k is clearly separable.
If is a root of x p x a, for a k, is called a simple pseudo radical

8. Solvable extensions

135

over k.k()/k is a pseudo radical extension and is separable. In fact, it


is a cyclic extension over k. This situation occurs, only if p , o.
An extension field K/k is said to be a generalized radical extension
if it is a finite tower
k = Ko K1 K2 Km = K
where Ki /Ki1 is either a simple radical or a simple pseudo radical extension. Every element it K is called a generalized radical. A typical
element would be
s
r
a3
n1
n2
a1 +
a2 +
+ ..
P
Clearly, K/k is a separable extension. A generalized radical extension is called a radical extension if Ki /Ki1 is a simple radical extension
for every i. A typical element of K would, then be
q

n1
a1 + n2 a2 + .

Let f (x) be a polynomial in k[x] and let it be separable. Let K be


its splitting field. Then K/k is a galois extension. The galois group
G of K/k is called the group of the polynomial f (x). The polynomial
f (x) is said to be solvable by generalized radicals, if K is a subfield 156
of a generalized radical extension. It is, then, clear that the roots of
f (x) are generalized radicals. In order to prove the main theorem about
solvability of a polynomial by generalized radicals, we first prove some
lemmas.
Lemma 1. Every generalized radical extension is a subfield of a generalized radical extension which is galois, with a solvable galois group.
Proof. Let K/k be a generalized radical extension so that
k = Ko K1 Km = K,
where Ki /Ki1 is a simple radical or simple pseudo radical extension.
Let Ki = Ki1 (i ). Then either ni Ki1 for some ni or Pi Ki1 .
Let (K : k) = n. Put N = n, if k has characteristic zero; otherwise, let
n = N pa ,

6. Special algebraic extensions

136

a o, (p, N) = 1. Let be a primitive Nth root of unity.

Let L1 = Ko (). The L1 /Ko is a simple radical extension. Since


K1 = Ko (1 ), put L2 = Ko (, 1 ). Then L2 is the splitting field of the
polynomial (xN 1)(xn1 a1 ) or (xN 1)(P xa1 ) depending on whether
1 is a simple radical or a simple pseudo radical. In any case, L2 /Ko is
a galois extension. Furthermore
L2 = L1 (1 )
157

so that L2 /L1 is cyclic, since, when 1 is a simple radical, L1 contains


the requisite roots of unity. Let 1 , . . . , be the distinct automorphisms
of L2 /Ko . Put

Y
f (x) =
(xn2 ai
2 ),
if 2 is a simple radical with

i=1
n2
2 =

f (x) =

a2 , and

Y
(P x a2 i ),
i=1

if 2 is a simple pseudo radical with P2 = a2 .


Then f (x) is a polynomial in k[x]. Let L3 be its splitting field. Then
L3 is galois over k. Also, L3 is splitting field of f (x) over L3 so that
L3 /L2 is either a Kummer extension or else, an abelian extension of
exponent p. In this way, one constructs a galois extension T of k such
that
Lo = k L1 L2 L3 Lm1 T = Lm ,
where Li /Li1 is either a kummer extension or an abelian extension of
exponent p. Clearly, Li /Li1 and, hence, T/k is a generalized radical
extension. Let Gi , i = o, 1, 2, . . . , m be the galois group of T/Li . Then
G = Go G1 . . . . . . Gm = (e)
is a normal series. Further, by our construction, Gi1 /Gi is the galois
group of Li /Li1 and hence, abelian. Thus, G is a solvable group. The
lemma is thus proved.

8. Solvable extensions

137

Lemma 2. If K/k is a finite solvable extension, then K is a subfield of a


generalized extension over k.
Proof. Let G be the galois group of K/k and G solvable. Let n be the
order of G and put
n = pa N
with the same connotation, as before. Let be a primitive Nth root of 158
unity and L = k(). Then L/k is a simple radical extension. Let M be
a composite of K and L. Then M/L is a galois extension with a galois
group which is isomorphic to a subgroup of G and hence, solvable. Let
Go be the galois group of M/L and let it have a composition series
Go G1 Gm = (e).

Then Gi /Gi+1 is a cyclic group of prime degree. Let Lo = L,
L1 , . . . Lm = M be the fixed fields of Go , G1 , . . . , Gm respectively. Then
Li /Li1 is a cyclic extension of prime degree. Since Li1 contains the
requisite roots of unity, Li is a simple radical or a simple pseudo radical
extension of Li1 . Then M/k is a generalized radical extension and our
lemma is proved.
We are, now, ready to prove
Theorem 16. A separable polynomial f (x) k[x] is solvable by generalized radicals if and only if its group is solvable.
Proof. Let K be the splitting field of f (x) and G the galois group of
K/k. Suppose f (x) is solvable by generalized radicals. Then K L
where L/k is galois and by lemma 1, has solvable galois group H. Let
Go be the galois group of L/K Then H/Go is isomorphic to G and so G
is solvable.

Let, conversely, G be solvable. Then, by lemma 2, K is contained
in a generalized radical extension and so f (x) is solvable by generalized
radicals.
We can easily prove
159

6. Special algebraic extensions

138

Corollary. A separable polynomial f (x) k[x] is solvable by radicals


if and only if its splitting field has a solvable galois group of order prime
to the characteristic of k, if different from zero.
Let k be a field. The polynomial
f (x) = xm x1 xm1 + x2 xm2 + (1)m xm ,
where x1 , . . . , xm are algebraically independent over k, is said to be the
general polynomial of the mth degree over k. It is so called, because
every monic polynomial of degree m over k is obtained by specializing
the values of x1 , . . . , xm to be in k. Let L = k(x1 , . . . , xm ). Let y1 , . . . , ym
be roots of f (x) over L. Then
f (x) = (x y1 ) (x ym )
and y1 , . . . , ym are distinct. The splitting field k(y1 , . . . , ym ) of f (x) over
L is a galois extension whose galois group is isomorphic to S m . Hence,
the general polynomial of the mth degree over k has a group isomorphic
to the symmetric group on m symbols.
But,S m is not solvable, if m > 4, so that in virtue of theorem 16, we
have the theorem of Abel.
Theorem 17. The group of the general polynomial f (x) of the mth degree is isomorphic to the symmetric group S m on m symbols and hence,
for m > 4, f (x) is not solvable by generalized radicals.

160

We shall now explicitly show how to obtain the roots of a polynomial of degree 4 in terms of generalized radicals.
Let f (x) be a general polynomial of degree m over k and let K be
the splitting field. K/k has the group S m . Let y1 , . . . , ym be the roots of
f (x). Put
Y
(yi y j )2 .
D=
i< j

Then D is fixed under all permutations in S m and, hence,


D k.

D)
is
a gaIf we assume that the characteristic
of
k
is
,
2,
then
k(

lois extension of k and K/k( D) has the galois group isomorphic to

8. Solvable extensions

139

the alternating group on m symbols. D called the discriminant of the


polynomial f (x).
Let us, first, consider the general polynomial of the second degree
f (x) = (x y1 )(x y2 ) = x2 x1 x + x2 .
Then
y1 + y2 = x1 , y1 y2 = x2 .
Suppose, now, that k has characteristic , 2. Then
D = (y1 y2 )2 = (y1 + y2 )2 4y1 y2 = x21 4x2 .

Also, y1 + y2 = x1 , y1 y2 = D so that

x1 D
x1 + D
, y2 =
y1 =
2
2

and K = k( D) is the splitting field of f (x) and is a radical extension.


Let, now, k have characteristic 2. Then x1 , o, since f (x) is separay1
x2
ble. Put x1 x instead of x. Then the polynomial x2 x + 2 has roots
x1
x1
x2
y2
, then
161
and . But this is a normal polynomial so that if =
x1
x12
y1 = x1

, y2 = x1
+ x1
P
P

and thus k( ) is a pseudo radical extension and is splitting field of


P
f (x).
We shall now study cubic and biquadratic polynomials.
Let, first, k have characteristic , 2 or 3. Let m = 3 or 4 and
f (x) = xm x1 xm1 + + (1)m xm
be the polynomial of the mth degree whose roots are y1 , . . . , ym .
x1
instead of x, we get a polynomial whose roots are
If we put x +
m
x1
x1
y1 , . . . , ym
and which lacks the terms in xm1 .
m
m

6. Special algebraic extensions

140

We shall, therefore, take the polynomial f (x) in the form


f (x) = xm + x2 xm2 + . . . + (1)m xm .
If y1 , . . . , ym are the roots, then
y1 + y2 + . . . + ym = o.
Also , Dm =

i< j

(yi y j )2 . A simple computation shown that


D3 = 4x32 27x23

and
D4 = 16x42 x4 4x32 x23 128x22 x24 + 144x2 x23 x4
27x43 + 256x34 .

162

If K is the splitting field of f (x) over L = k(x1 , . . . ,xm ) then L( D)


is the fixed field of the alternating group Am and L( D)/L is a radical extension. In order to study the extension K/L( D), let us, first,
take the case m = 3. The symmetric group on 3 symbols, S 3 , has the
composition series
S 3 A3 (e).

K/L( D) is thus a cyclic extension


of degree 3. Let be a primitive
cube root of unity and let M = L( D, ). Let N = K M be a composite
of K and M. Then K M/M has degree 1 or 3, according as is in K or
not. In the first case, M = K. In the second case, K M/M is a cyclic
extension of degree 3 over K and M contains the cube roots of unity.

Thus K M = M( 3 ), for some M. In order of determine this , we


use Lagranges method.
K M is the splitting field over M of the polynomial x3 + x2 x x3 . Let
y1 , y2 , y3 be roots of this polynomial. Put
= y1 + y2 + 2 y3
= y1 + 2 y2 + y3 .

8. Solvable extensions

141

27
x3 + 3 D( 12 ). Changing into 2 we get 3 .
Then 3 =
2
Hence
r

1
a 3 27
x3 + 3 D( )
=
2
2
and we have a similar expression for . Here a o. In order to determine a, we use the fact that = 3x2 and so, choosing the root
of unity a for arbitrarily, the root of unity in the expression for is
uniquely determine. Now
y1 + y2 + y3 = o
y1 + y2 + 2 y3 =
y1 + 2 y2 + y3 =
and since the matrix

1 1 1

2
1
1 2

is non-singular, the values of y1 , y2 , y3 are uniquely determined K M is 163


a radical of L and contains K.
Consider, now, the polynomial of the fourth degree
f (x) = x4 + x2 x2 x3 x + x4
whose roots are y1 , y2 , y3 , y4 with y1 + y2 + y3 + y4 = o. The galois group
of the splitting field K/k is S 4 . This has the composition series
S 4 A4 B4 C4 (e).
Let K1 , Kb , Kc be the fixed fields of A4 , B4 and C4 respectively. Now
A4 is the alternating group, B4 the group consisting of the permutations
(1), (12)(34), (13)(24), (14)(23)
and C4 is the group of order 2 formed by
(1), (12)(34).

6. Special algebraic extensions

142

Ka = k( D) and is of degree 2 over k. Now Kb /Ka is of degree 3


and is cyclic. Hence Kb = Ka () where K is an element fixed by B4
but not by A4 . Such as elements, for instance, is
1 = (y1 + y2 )(y3 + y4 ).
1 has 3 conjugates 1 , 2 , 3 obtained from 1 by operating on 1 , by
representatives of cosets of A4 /B4 . Thus
2 = (y1 + y3 )(y2 + y4 )
3 = (y1 + y4 )(y2 + y3 ).
164

Consider the polynomial


(x) = (x 1 )(x 2 )(x 3 ).
It is fixed under A4 and so, its coefficients are in Ka . A simple computation shows that
(x) = x3 2x2 x2 + (x22 4x4 )x + x23 .
(x) is called the reducing cubic or the cubic resolvent. By the
method adopted for the solution of the cubic, if is a primitive cube
root of unity and M = Ka (), then Kb M, the composite, is a radical
extension of k in which 1 , 2 , 3 lie.
Kc /Kb is of degree 2 and so Kc = Kb () where is fixed under B4 ,
but not by C4 . such as element is
= y1 + y2 .
Now
2 = (y1 + y2 )2 = (y1 + y2 )(y3 + y4 ) = 1 . Thus Kc =

Kb ( 1 ). Hence, if Kb M = Kd , then Kd () is a radical extension of k


containing . Put now
T = Kd (, )

where = y1 +y3 = 3 . Then T is a radical extension of k containing


K. The indeterminacy signs in taking the square roots of 1 and 3
can be fixed by observing that
(y1 + y2 )(y1 + y3 )(y1 + y4 ) = x3 .

8. Solvable extensions

143

we now have
p

p
1 , y3 + y4 = 1
p
p
y1 + y3 = 2 , y2 + y4 = 2
p
p
y1 + y4 = 3 , y2 + y3 = 3 .

y1 + y2 =

for which y1 , y2 , y3 , y4 can be obtained. We have, hence, proved


Theorem 18. If K has characteristic , 2 or 3, then the cubic and biquadratic polynomials over k can be radicals.
Let us, now, assume that k has characteristic 3. Let x3 + x1 x2 + x2 x+
x3 be a cubic polynomial and K, its splitting field. K/k has the galois
group S 3 . Let L be the fixed field. K/k has the galois group S 3 . Let L
be the fixed field of A3 . Then

L = k( D)
where
D = (y1 y2 )2 (y2 y3 )2 (y3 y1 )2 .
K/L is now a cyclic extension of degree 3 and since k has characteristic 3, K = L(), where
3 = o
for some L. Thus K is a generalized radical extension. We shall
now determine and and therefrom, y1 , y2 and y3 .
In order to do this, we have to consider two cases, x1 = o, and
x1 , o.
Let, first, x1 = o. Then y1 + y2 + y3 = o. Let be a generator of the
galois group fo K/L and let notation be so chosen that
y1 = y2 , y2 = y3 , y3 = y1 .
2

Since S K/L y1 = y1 + y1 + y1 = o, by Hilberts theorem, there is a


in K such that
y1 = .

165

6. Special algebraic extensions

144
Also

166

X
2
2
x2 = y1 y2 + y2 y3 + y3 y1 =
( )( ) = ( + + )2 .

Since x2 , o (otherwise, polynomials is not separable), we see that


x2 = t2 , for some t L. The polynomial, therefore, has the form
x3 t2 x + x3 .
y1
Put now tx for x. Then the polynomial x3 x + x3 /t3 has roots ,
t
y2 y3
, . Let = x3 /t3 . Then
t t

+ t, y3 = t
+ 2t
y1 = t , y2 = t
P
P
P
and thus the roots are all obtained.
The indeterminacy in the sign of t does not cause any difficulty; for,

if we use (t) instead of t, then, observing that


=
, we see that
P
P
y1 = t

, y3 = t
t, y2 = t
+t
P
P
P

so that y2 and y3 get interchanged.


We now consider the case x1 , o.
Put x + a instead of x. Then the new polynomial is
x3 + x1 x2 + x(x2 + 2x1 a) + x3 + x2 a + a2 + a3 .
Choose a so that x2 + 2x1 a = o. For this value of a, = x3 + x2 a +
+ a3 , o; for, otherwise, the polynomial will be reducible and the
roots are o, o, x1 . The roots of this new polynomial are y1 a, y2 a,
y3 a.
1
Let now be written for x, then the polynomial is reduced to x3 +
x
1
1
x1 1
x+ . We are in the previous case. The roots, now, are
,
,

y1 a y2 a
1

x1
1
. If t2 = , t L and = 3 , then K = L( ) and the roots
y3 a

P
t

a2

167

8. Solvable extensions

145

y1 , y2 , y3 are given by
y1 = a +

tP

, y2 = a +

1
1
, y3 = a +

t + tP
2t + t P

Since the characteristic is 3, the biquadratic polynomial can be taken


in the form
x4 + x2 x2 + x3 x + x4 .
The proof is similar to the old one except
that Kb /Ka is a cyclic

extension of degree 3 and so Kb = Ka
for a suitable in Ka . To
P
find , we use the foregoing method. One finds that K is a generalized
radical extension.
We now consider the case where k has characteristic 2. In this case,
the cubic polynomial cab be taken in the form, x3 + x2 x + x3 . Let y1 , y2 ,
y3 be the roots. Then
y1 + y2 + y3 = o
and therefore y21 + y22 = y23 and so on. Put
y1 y2 y3
+
+ ,
y2 y3 y1
y2 y3 y1
+
+ .
=
y1 y2 y3
=

Then
+ =

y1 (y22 + y23 ) + y2 (y23 + y21 ) + y2 (y21 + y22 )


= 1,
y1 y2 y3

which shows that < k, because ,then, it will be symmetric and equal
to . Thus k() is a quadratic subfield of K. A simple computation
shows that and = + 1 are roots of
x2 x + x32 .
K/k() is cyclic of degree 3 and one uses the method of Lagrange 168
to obtain a generalized radical extension.

6. Special algebraic extensions

146

Suppose now that f (x) is a polynomial of degree 4. Let f (x) =


x4 + x1 x3 + x2 x2 + x3 x + x4 . We have to consider two cases. Let, first,
x1 = o. Then the roots y1 , y2 , y3 , y4 satisfy
y1 + y2 + y3 + y4 = o.
As before, put
1 = (y1 + y2 )(y3 + y4 ),

2 = . . . , 3 =

The reducing cubic is then


(x) = x3 + x22 x + x3 .
Furthermore 1 + 2 + 3 = o. Put now
=

1 2 3
+
+ .
2 3 1

Then is fixed under A4 but not under S 4 . Also Ka = k() is a


simple pseudo radical extension.
Now Kb /Ka is a cyclic extension of degree 3 and has to be solved
by Lagranges methods. Further, Kc /Kb is of degree 2. Put now
1 =

y3
,
y1 y2 y4

1 =

y4
.
y1 y2 y3

Then
1 +

y23 + y24
(y1 + y2 )(y3 + y4 ) 1
=
=
= .
y1 y2 y3 y4
y1 y2 y3 y4
x4

We assume x3 , o. Then 1 , o. If we put


2 =
169

x4 1
,
1

1 =

x4 1
,
1

then 2 + 2 = 1 and Kc = Kb (2 ).
In similar manner, K = Kc (3 ) where
3 =

x4
,
2

y1
.
y2 y3 y4

8. Solvable extensions

147

Suppose, now, that x1 , o. Then, as before, we construct the field


Kb . In order to exhibit Kc as a pseudo radical extension of Kb , observe
that y1 + y2 is fixed under C4 but not under B4 . Also
(y1 + y2 )2 = (y1 + y2 )(y3 + y4 + x1 ) = 1 + x1 (y1 + y2 )
which shows that
Kc = Kb

!
y1 + y2
.
x1

!
y1 + y3
.
x1
Our contentions are completely established.

Similarly, K = Kc

Chapter 7

Formally real fields


1 Ordered rings
170

A commutative ring R is said to be ordered if there is an ordering relation


> (greater than) such that
(1) for every a R, a > o, a = o or a > o.
(1) a, b R, a > o, b > o a + b > o, ab > o.
We may then define a > b by a b > o. If a > b, then, for any c in
R, a + c > b + c and if c > o, ac > bc.
If a > o, we shall say a is negative and if a > o, a said to be
positive. We denote a is negative by a < o.
Let us denote, by P, the set of elements a R with a o. Then,
from the definition or ordered ring, we have
A1 )P + P P

A2 )P P P

A3 )P (P) = (o)
A4 )P (P) = R,

where P + P denotes the set of elements of R of the form a + b, a, b P;


P denotes the set of elements a, a P P shall be called the set of
149

7. Formally real fields

150

171

non-negative elements of R. The only element which is both positive


and negative is zero. Clearly, if R is a any subset P of R satisfying the
four conditions above again determine an order on R.
Two elements a, b R, a , o , b are said to have the same or
opposite signs according as ab > o or ab < o.
Let now R be an ordered ring with unit element 1. Let a R, a , o.
Then a2 = a a = (a) (a), so that a2 > o for a , o in R. More
generally, every finite sum of squares of elements of R is positive. These
elements will be contained in the set of non-negative elements in every
order of R.
Since R has a unit element 1 and 12 = 1, we have 1 > o. Also,
n 1 = 1 + 1 + 1, n times so n > o. This proves
1) An ordered ring with unit element has characteristic o.
Let a , o, b , o be elements of R. Then a or a is positive.
Similarly b or b is positive. Hence ab or ab is positive which
proves that ab , o. Therefore
2) An order ring is an integrity domain.
We define an ordered field to be an ordered ring whose non-zero
elements form a multiplicative commutative group. We have
3) If k is an ordered field, its positive elements form a multiplicative
group.
For, if x k, x > o; then xx1 > o. If x1 > o, then xx1 > o
which contradicts x1 x > o. Thus x1 > o and (3) is proved.
Suppose R and R are two rings, R R . If R is ordered, clearly
R is ordered by means of the induced order. If however, R is
ordered, it may not be possible, in all cases, to extend this order
to R . However, in case, it is always possible, namely
4) If K is the quotient field of an ordered ring R then the order in R
can be uniquely extended to K.

172

Proof. Let R have an order. It is an integrity domain. Any elements


a
x K is of the form x = , a, b R, b , o. Let x , o. Then a , o.
b

1. Ordered rings

151

define x > o by ab > o. Then this defines an order in K. In the first


place, the definition does not depend on the way x is expressed in the
a
a
form . Suppose x = . Then ab = a b. Since b , o, multiplying
b
b
both sides of this equality by bb , we have
ab b2 = a b b2 .

a
Since ab > o, b2 > o, b2 > o, it follows that a b > o, that is > o.
b
a
> o and
In order to prove that A1 , . . . , A4 are satisfies, let x =
b

a
y = > o. Then ab > o and a b > o.
b
x+y =

ab + a b
bb

Now (ab + a b)bb = ab b2 + a b b2 . Since ab > o, a b > o,


> o, b2 > o, it follows that (ab + a b)bb > o or x + y > o. In similar
manner, xy > o.
Suppose x o and y o and x + y = o; then x = o, y = o. For, if
c
ad + bc
a
= o, so that
x = , y = , then ab o, cd o and x + y =
b
d
bd
ad + bc = o. Thus abd2 + cdb2 = o, which means that since all elements
are in R, ab = o, cd = o, i.e. a = o = c.
ab
If x R, than x =
. If x = o, then ba2 = o or b = o, so that the
a
order coincides on R with the given order in R.
That the extension is unique can be, trivially, seen.
Since every ordered field has characteristic zero, it contains a sub- 173
field isomorphic to , the rational number field has thus induced order.
We shall now prove
b2

(5) can be ordered in one way only.


For, if Z denotes the set of integers, then is the quotient field of Z.
On Z, there is only one order since 1 > o and hence n = 1+1+ +1 > o.
Thus, all natural integers have to be positive.

7. Formally real fields

152

2 Extensions of orders
Hereafter, we consider ordered fields k. Our main task will be the study
of extensions of orders in k to extension fields K of k. For this purpose,
we introduce the notion of a positive form on k.
m
P
Let k be an ordered field. A polynomial ai x2i , ai k, is said to be
i=1

an m-ary form over k. It is said to be positive if ai > o, i = 1, . . . m. An


m-ary form is said to represent a k, if there exist 1 , . . . , m in k such
that
X
ai 2i = a.
i

Clearly a positive form represents zero, only if 1 , . . . , m = o. Let


k be an ordered field and K/k, an extension field. We shall prove

Theorem 1. K has an order extending the order in k, if and only if,


every positive form over k is still positive in K.
174

Proof. We have only to prove the sufficient. To this end, consider the
family M of subsets {M } of K having the following properties. Denote,
by S , the set of elements in K of the form
X
ai 2i ,
i

ai k and ai > o and i K. (i can be all zero also).

Then
1)
3)

M S

M M M

2) M + M M

4) M (M ) = (o).

This family is not empty, since S satisfies this condition. In the usual
way, we make M a partially ordered set and obtain a maximal set P. We
have now to show that
P (P) = K
and then P will determine an order. Let x , o be an elements of K which
is not in P. Define
Q = P xP

2. Extensions of orders

153

as the set of elements of the form a xb, a, b P. Obviously Q satisfies


(1). To see that Q satisfies (2), observe that if a xb, c xd are in Q
then
(a xb) + (c xd) = (a + c) x(b + d);
but a, b, c, d being in P which is an element of M, a + c, b + d are
in P. In a similar way Q satisfies (3). That Q satisfies (4) can be seen
as follows: Let a xb and c xd be in Q with a, b, c, d, in P. Let
(a + c) x(b + d) = o. Then b + d = o. For, if b + d , o, then
x=

1
a+c
= (a + c)(b + d)
b+d
(b + d)2

and so is an element of P, which is a contradiction. Hence b + d = 0 175


and, therefore, a + c = 0. Since a, b, c, d are all in P, it follows that
a = b = c = d = 0. Thus Q M. But Q P so that, by maximality of
P, Q = P. This means that x P or x P. The theorem is therefore
proved.
If is the field of rational numbers and n is a positive integer, then
a
n = 1 + 1 + + 1. If r = is a positive rational number, then
b
a ab 1 + 1 + 1
=
=
,
b b2
b2
so that every positive rational number is a sum of squares. This shows
that every positive form over can be put in the form x21 + + x2n .
P
If k is an ordered field then 2i = 0, i k implies that i = 0,
i
P
i = 1, . . .. On the other hand, if k has the property that 2i = 0, i k
implies i = 0, then n = 1+1+ +1 , 0 so that k has characteristic zero.
Since every positive form over is essentially of the type x21 + + x2n ,
we have the
P
Theorem 2. k is ordered if and only if 2i = 0, i = k implies i = 0,
i

i = 1, 2, . . ..

It is obvious that, if, in a field k,

P
i

zero, then
1 =

X
i

2i = 0, with 1 , 2 , . . . not all

i 2,

7. Formally real fields

154
176

i k. A field in which 1 is not a sum of squares is called a formally real field. From theorem 2, it follows that formally real fields are
identical with ordered fields.
We shall, now, prove the following application of theorem 1.
Theorem 3. Let k be a formally real field with a given order and f (x),
an irreducible polynomial in k[x]. Let a, b be two elements in k such that
f (a) f (b) < 0. Suppose is a root of f (x) in , an algebraic closure of
k. Then K = k() is ordered with an order which is an extension of the
given order in k.
Proof. Let f (x) be of degree n. Then every element in k() is a polynomial in of degree n 1 with coefficients in k. If k() is not ordered
with an order extending that in k, then there is a positive form in k which
represents 1. That is,
X
1 =
ai {i ()}2 ,
i

ai > 0 in k. This means that, in k[x],


X
1+
ai {i (x)}2 = f (x)(x).
i

Since f (x) has degree n and left side has degree 2n 2, (x) has,
at most, the degree n 2.

We now use induction on n. If n = 1, these is nothing to prove.
Assume theorem proved for n 1 instead of n. Let g(x) be in irreducible
factor of (x). Then g(x) has degree n 2. Now
X
ai {i (a)}2 = f (a)(a)
0<1+
0<1+

i
X
i

177

ai {i (b)}2 = f (b)(b).

Since f (a) f (b) < 0, it follows that (a)(b) < 0. Therefore, at least
one irreducible factor, say g(x), of (x) has the property g(a)g(b) < 0.

2. Extensions of orders

155

If is a root of g(x) in , then g() = f () = 0. But by induction


hypothesis, k() has an order extending the given order in k. Hence
X
0=1+
ai {i ()}2 > 0,
i

which is a contradiction. Thus our theorem is completely proved.


Suppose k is an ordered field, let us denote, by |a|, the absolute value
of a k by

0
if a = 0

|a| =
a
if a > 0

a if a < 0.
It is then easy to prove that

|ab| = |a| |b|,

|a + b| |a| + |b|.
Let f (x) = xn + a1 xn1 + + an be a polynomial in k[x]. Put
M = max(1, |a1 | + + |an |). If t , 0 k and |t| > M, then
tn f (t) = 1 + a1 t1 + + an tn > 0
which shows that tn and f (t) have the same sign.
Suppose now f (x) is irreducible and of odd degree. Then, if M is
defined as above,
f (M) f (M) < 0.
Therefore, by theorem 3, if is a root of f (x) in an algebraic closure of
k, then k() has an order extending that in k.
If a k and a > 0, then the polynomial x2 a changes sign in k. 178
For, (a + 1)2 > a and a < 0. Thus
((a + 1)2 a)(0 a) < 0.

Therefore, k( a) has an order extending the given order in k.

156

7. Formally real fields

3 Real closed fields


We had seen above that, under certain circumstances, an order in a field
k can be extended to a finite algebraic extension of k. We shall consider,
now, a class of fields called real closed fields defined as follows: k is
said to be real closed, if
1) k is ordered
2) k has no proper algebraic extension K with an order extending that
in k.
Before we establish the existence of such fields, we shall obtain
some of their properties. We first prove
Theorem 4. For a formally real field k, the following properties are
equivalent:
1) k(i) is algebraically closed, i being a root of x2 + 1.
2) k is real closed.
3) Every polynomial of odd degree over k has a root in k and every
positive element of k is a square in k.

179

Proof. 1 2. k(i) being of degree 2 over k, there are no intermediary


fields, so that, k being ordered, and k(i) being algebraically closed, k has
no ordered algebraic extension.
2 3. Suppose f (x) is a polynomial of odd degree. Then it
changes sign in k. Hence an irreducible factor of f (x), also of odd degree, changes sign in k. If is a root of this irreducible factor, then k()
is ordered with an order extending that in k. Hence k. If a > 0 in k,

then x2 a changes sign in k. Thus a k.


3 1. The poof of this part consists of three steps. Firstly, every
element of K = k(i) is a square. For, let a + bi be an element of K, a,
b k. Put
a + ib = (c + id)2

3. Real closed fields

157

where c and d have to be determined in k. Since 1, i form a base of K/k


we get
c2 d2 = a, 2cd = b.
Therefore, (c2 + d2 )2 = a2 + b2 . But a2 + b2 > 0 in k and, since every
positive element is a square, there is a > 0 in k such that
c2 + d2 = .
Solving for c2 and d2 , we have
c2 =

+a
,
2

d2 =

a
.
2

But, since 2 = a2 + b2 , it follows that |a|, |b|. Therefore


a
+a
0,
0. Therefore
2
2
r
r
+a
a
c=
, d=
2
2
exist in k. The arbitrariness in the signs of c and d can be fixed from the
fact that 2 cd = b.

This proves that every quadratic polynomial over k has a root in 180
2
K. For,
if ax + bx + c k[x], then, in an algebraic closure of K,

b b2 4ac
are its roots (a , 0). But b2 4ac K.
2a
The second step consists in showing that every polynomial in k[x]
has a root in K. Let f (x) be in k[x] and let N be its degree N = 2n q, q
odd. We shall use induction on n. If n = 0, then N = q, and so whatever
odd number q be, f (x) has a root already in k. Let us, therefore, assume
proved that every polynomial of degree 2n1 q , where q is odd, with
coefficients in k has a root in K. Let f (x) be of degree N = 2n q, q odd.
Let 1 , . . . , t be the distinct roots of f (x) in an algebraic closure of K.
Let k be an element to be suitably chosen later. Put
i j () = i j = i + j + i j i, j = 1, . . . , t, i , j.

7. Formally real fields

158
Consider now the polynomial
(x) =

Y
(x i j ).
i, j

181

N(N 1)
This has degree
= 2n1 q , q odd. Also, by every per2
mutation of the symbols 1, . . . , t, the polynomial goes over into itself.
Thus (x) k[x]. Since its degree satisfies the induction hypothesis,
for every k, there is an i and a j such that i j () K. Since k is
an infinite field, there exist , , , and both in k such that i j ()
and i j ( ) for two integers i and j are in K. This means that i j and,
hence, i + j are in K. The polynomial x2 x(i + j ) + i j is a
polynomial in K[x]. By what we proved above, both its roots are in K.
Thus our contention is proved.
The third step consists in proving that every polynomial in K[x],
has a root in K. For, let f (x) be a polynomial in K[x]. Let be the
generating automorphism of K/k. It is of order 2. Denote by f (x)
the polynomial obtained from f by applying on the coefficients of f .
Then = f (x) f (x) is a polynomial in k[x]. The second step shows that
has a root in K. Furthermore if is a root of , is also a root of
, so that either is a root of f (x) or is a root of f (x).
We have thus proved theorem 4 completely.
We deduce from this an important corollary due to Artin and
Schreier.
Corollary 1. If is an algebraically closed field and K a subfield such
that 1 < ( : K) < , then K is real closed.
Proof. We had already proved that K(i) = and that K has characteristic zero. By virtue of theorem 4, it is enough to prove that K is formally
real. Every element of is of the form a + ib, a, b K. Also
a + ib = (c + id)2 ,
for c, d in K, since is algebraically closed. Thus
a2 + b2 = (c2 + d2 )2 .

3. Real closed fields

159

Hence every sum of two squares and, hence, of any number of


squares is a square. Therefore
X
1 =
a2i
i

is impossible in K. By theorem 2, therefore, K is formally real.

 182

This proves that the real closed fields are those and only those which
are such that their algebraic closures are finite over them.
We have again
Corollary 2. A real closed field has only one order.
For, the set of positive elements coincides with the set of squares of
the elements of the field.
This shows that, if on ordered field has two distinct orders, it has
algebraic extensions which are ordered. It must be remembered that if
a field has only one order, it is not necessarily real closed. The rational
number field, for example, has only one order.
Suppose k is a real closed field. Then every irreducible polynomial
in k[x] is of degree one or two. Suppose f (x) is a polynomial in k[x] and
a, b in k such that
f (a) f (b) < 0.
Then one of the irreducible factors (x) of f (x) must have the property
that (a)(b) < 0. If is a root of (x), then k() is ordered. But,
k being real closed, k. (x) must be a linear polynomial. Thus
(x) = x and
(a )(b ) < 0
which means that lies between a and b. Hence the
Theorem 5. If k is a real closed field, f (x) a polynomial in k[x], a, b in
k such that f (a) f (b) < 0, then there is a root of f (x) in k between a
and b.
Furthermore, we had seen that there is an M in k depending only 183
on the coefficients of f (x) such that f (a) has the same sign as an , n =
deg f (x), if |a| > M. This shows

7. Formally real fields

160

All the roots of f (x) that lie in k lie between M.


Let k be a real closed field and f (x) a polynomial in k[x]. Let f (x)
be its derivative. Put 0 = f , 1 = f . Since k[x] is a Euclidean ring,
define, by the Euclidean algorithm, the polynomials
0 = A 1

1 2

1 = A 2
...

2 3

...

r1 = Ar

r .

It is then well-known that r (x) is the greatest common divisor of 0


and 1 . The sequence 0 , 1 , . . . , r of polynomials in k[x], is known as
the Sturmian polynomial sequence.
Let a k be such that 0 (a) , 0. Then r (a) , 0. Consider the
set of elements 0 (a), 1 (a), . . . , r (a) in k. The non-zero ones among
them have a sign. Denote, by (a), the number of changes of sign in the
sequence of elements,
0 (a), 1 (a), . . . , r (a),
in this order, taking only the non-zero elements. A very important theorem due to Sturm is

184

Theorem 6. Let b and c be two elements of k, b < c and 0 (b) , 0,


0 (c) , 0. Let (b) and (c) denote the number of changes of sign in
the Sturmian sequence for the values b and c. Then f (x) has precisely
(b) (c) distinct roots in k between b and c.
Proof. Since r (b) , 0, r (c) , 0, we may divide all the Sturmian
polynomials by r (x) and obtain the sequence 0 (x), 1 (x), . . . r1 (x),
r (x)(= 1). Now 0 (x) has no multiple roots. For, if is a root of f (x)
of multiplicity t, then
0 (x) = (x )t 1 (x),
t1

1 (x) = t(x )

1 () , 0

1 (x) + (x )t 1 (x)

3. Real closed fields

161

so that (x )t1 is the highest power of of x , that divides 1 (x).


Hence
0 (x) = (x )2 (x)

1 (x) = t2 (x) + (x )3 (x),

2 (x) and 3 (x) being polynomials over k. Note that 1 (x) is not the
derivative of 0 (x). We shall drop the bars on the s and write them
as 0 , 1 , . . . , r1 , r = 1 and 0 having no multiple roots. Note that
(b) or (c) is not altered by doing the above.

The finite number of polynomials 0 , 1 , . . . , r1 have only finitely
many roots between b and c. By means of these roots, we shall split the
interval (b, c) into finitely many subintervals, the end points of which
are these roots. We shall study how the function (a) changes as a runs
from b to c.
1) No two consecutive functions of the Sturmian series 0 (x), 185
1 (x), . . . , r1 (x) can vanish at one and the same point, inside the
interval (b, c). For, suppose b < a < c and i (a) = 0 = i+1 (a),
0 < i + 1 < r. Then
i (x) = Ai i+1 (x) i+2 (x)
so that i+2 (a) = 0, and, so on, finally r (a) = 0. But r (a) = 1.
2) Inside any one of the intervals, each function keeps a constant sign;
for, if any function changed sign then, by theorem 5, there would be
a zero inside this interval.
Let d denote an end point of an interval and L and R the intervals to
the left of d and to the right of d, having d as a common end point.
3) Suppose d is a zero of 1 for 0 < 1 < r. Then
l1 = Al l l+1 ,
so that l1 (d) = l+1 (d) and none of them is zero by (1). Because
of (2), l1 has in L the same sign as at d. Similarly in R. The same

7. Formally real fields

162

is true of l+1 . Thus, whatever sign l might have in L and R, the


function (a) remains constant when a goes from L to R crossing a
zero of l , 0 < 1 < r.
4) Let now d be a zero of 0 . Then d is not a zero of 1 . We have
0 (x) = (x d)(x)

1 (x) = m(x) + (x d)1 (x),


where m is an integer > 0, (x) and 1 (x) are polynomials over k,
and (d) , 0. At d, 1 (d) has the same sign as (d).
186

In L, 0 has the sign of 0 (a) = (a d)(a). But a d < 0, so that


0 has the sign of (a). In L, 1 (a) has the same sign as (d). Also
(x) has no zero in L. Hence 1 (a) has the same sign as (a). Therefore
in L, 0 and 1 have opposite signs. In R, exactly the opposite happens,
namely 0 (a) = (a d). (d), a d > 0. Hence 0 and 1 have the
same sign in R. Thus (a) is lessened by 1, whenever a crosses a zero
of 0 (x) and remains constant in all other cases.
Our theorem is, thus, completely proved.
We make the following remark.
Remark. Suppose k is a formally real field and
f (x) = xn + a1 xn1 + + an ,
a polynomial in k[x]. Let, as before, M = max(1, |a1 | + + |an |).
Suppose there exists a real closed algebraic extension K of k with an
order which is an extension of the given order in k. f (x) can, then, be
considered as a polynomial in K[x] and, as seen earlier, f (x) has no
roots in K outside the interval (M, M). The number of these distinct
roots is thus independent of K.
We now prove
Theorem 7. Let k be an ordered field, its algebraic closure. Suppose
there exist two real closed subfields K, K of /k with orders extending
the given order in k. Then K and K are k isomorphic.

3. Real closed fields

187

163

Proof. 1) Let f (x) be a polynomial in k[x] and 1 , . . . , t the distinct


roots of f (x) in K. Let 1 , . . . , t be the distinct roots of f (x) in
K . Put L = k(1 , . . . , t ) and L = k(1 , . . . , t ). Since L/k is
finite, L = k() for some in K. Suppose (x) is the minimum
polynomial of in k[x]. Let be a root of (x) in K . Then k() and
k( ) are k-isomorphic. If is this isomorphism, then = . But
k() = L = k(1 , . . . , t ). Hence 1 , . . . , t will be distinct roots
of f (x) in K . Thus L = k( ). Hence
L L .
2) Suppose (x) is any polynomial in k[x], 1 , . . . , s its distinct roots
in K. Let 1 , . . . , s be the corresponding roots in K . Consider all
the positive quantities among i j , i , j. Their square roots exist
in K. Let (x) be a polynomial in k[x], among whose roots are these
square roots and let 1 , . . . , g be the roots of (x) in K, 1 , . . . , g
the corresponding roots in K . Then, from above,
F = k(1 , . . . , s ,

k(1 , . . . , s ,

1 , . . . , g )

1 , . . . , g ) = F .

Let be this isomorphism. Let notation be such that


(i ) = i
(i ) = i .
Suppose i > j . Then i j > 0 so that i j = 2t , for some t.
Also, (i j ) = i j . But
(i j ) = (2t ) = 2
t .
2

Hence i j = t > 0, which proves that


i > j .
The isomorphism between F and F preserves order between the 188
roots of (x) in K.

7. Formally real fields

164

3) In order, now, to construct the isomorphism between K and K , we


remark that any such isomorphism has to preserve order. For, if is
this isomorphism and > in K, then = 2 so that
( ) = (2 ) = (())2 > 0
so that > . We shall, therefore, construct an order preserving
map which we shall show to be an isomorphism.
4) Let be an element in K, h(x) its minimum polynomial over k. Let
1 , . . . , t be the distinct roots of h(x) in K and let notation be so
chosen that , < 2 < . . . < t . Let = i . Let 1 , . . . , t be the
distinct roots of h(x) in K and, again, let the notation be such that
1 < 2 < . . . < t .
Define, now, on K by
= i .
Let and be two elements of K and let f (x) be a polynomial in
k[x]n whose roots in K are , , + , , . . .. Construct the fields
F and F and the k-isomorphism of F on F . Since preserves
order of roots of f (x), it preserves order of roots of the factor of f (x)
which has as its root. Similarly of the factor having as a root.
Hence
() = (),

() = (), ( + ) = ( + ), () = ().

189

Hence ( + ) = + , () = . Thus is an isomorphism of K into K . We have similarly an isomorphism of K into


K. Thus is identity on K . Hence K and K are k-isomorphic.

Corollary. The only automorphism of K over k is the identity.
We shall now prove the theorem regarding the existence of real
closed fields, namely

3. Real closed fields

165

Theorem 8. If k is an ordered field with a given order and , its algebraic closure, there exists in , upto k-isomorphism, only one real
closed field K with an order extending the given order in k.
Proof. Let V be the family of formally real subfields of /k which have
an order extending that in k. V is not empty, since k V. We partially
order V by inclusion. Let {k } be a totally ordered subfamily of V. Let
S
K0 = k . Then K0 is a field which is contained in V. This can be

easily seen. By Zorns lemma, there exists a maximal element K in V.


K has an order extending the order in k. To prove that K is real closed, let
f (x) be a polynomial of odd degree over K. It changes sign. Therefore
there is an irreducible factor, also of odd degree, which changes sign
in K. This factor must have a root in K. Else, by theorem 3 there
exists an algebraic extension with an order extending that in k. This will
contradict maximality of K. In a similar way, every positive element of
K is a square. By theorem 4, K is real closed. If K and K are two real
closed subfields with orders extending that in k, then, by theorem 7, they 190
are k-isomorphic.

Suppose now that k is a perfect field and , its algebraic closure.
Let G be the galois group of /K. If G has elements of finite order (not
equal to identity), let K be the fixed field of the cyclic group generated
by one of them. Then ( : K) is finite and by Artin-Schreier theorem
this order has to be two. Thus any element of finite order in G has to
have the order two. Furthermore, in this case, k is an ordered field.
On the other hand, if k is an ordered field and , its algebraic closure, there exists, then by theorem 8, a real closed subfield of /k, say
K. This means that G has an element of order 2. Moreover no two elements of order 2 commute. For if , are of order 2 and commute, then
1, , , is a group order 4 which must have a fixed field L such that
( : L) = 4. This is impossible, by Artin-Schreier theorem. Hence the
Theorem 9. If k is a perfect field, its algebraic closure and G the
galois group of /k then G has elements of finite order if and only if k
is formally real. Also, then, all these elements have order 2 and no two
of them commute.

7. Formally real fields

166

4 Completion under an order


Let k be a formally real field with a given order. We had defined a
function | | on k with values in k such that
|ab| = |a| |b|,

|a + b| |a| + |b|.
This implies that
|a| |b| |a b|.

191

Also a |a| is a homomorphism of k into the set of positive elements of k. The function | | defines a metric on the field k. We define
a Cauchy sequence in k to be a sequence (a1 , . . . , an , .) of elements of k
such that for every > 0 in k, there exists n0 , an integer such that
| an am |< , n, m > n0 .
Obviously, if m > n0 + 1,
|am | = |am ano ano | < + |an0 |,
so that all elements of the sequence from n0 onwards have a value less
than a certain positive element of k.
A Cauchy sequence is said to be a null sequence if, for every > 0
in k, there is an integer n0 = n0 () such that
| an |< ,

n > n0 .

The sum and product of two Cauchy sequence is defined as follows:(a1 , a2 , . . .) + (b1 , b2 , . . .) = (a1 + b1 , a2 + b2 , . . .)
(a1 , a2 , . . .) (b1 , b2 , . . .) = (a1 + b1 , a2 + b2 , . . .)
and it is easy to verify that the Cauchy sequences in k form a ring R and
the null sequences, an ideal Y of R. We assert that Y is a maximal

4. Completion under an order

167

ideal. For, let (a1 , . . .) be a Cauchy sequence in k which is not a null


sequence. Then there exists a > 0 in k and an integer n such that
| am |> , m > n.

(1)

For, if not, for every > 0 and integer n, there exist an infinity of 192
m > n for which | am |< . Since (a1 , . . .) is a Cauchy sequence, there
exists n0 = n0 () such that
|am1 am2 | < ,

m1 , m2 > n0 .

Let m0 > n0 such that |am0 | < . Then, for all m > m0 ,
|am | |am am0 | + | am0 < 2
which proves that (a1 , . . .) is a null sequence, contradicting our assumption.
Let 1 be the unit element of k. Then the sequence (1, 1, . . .) is a
Cauchy sequence and is the unit element in R. Let c = (a1 , . . .) be in
R but not in Y . Let m be defined as in (1). Then c1 = (0, 0, . . . 0, a1
m ,
a1
,
.
.
.)
is
also
a
Cauchy
sequence.
For,
let

>
0
and
n,
an
integer
so
m+1
that
n1 , n2 > n0 .
| an1 an2 |< ,
Then





1
1
1

an2 an2 < 2 ,

=

an1 an2
|an1 | |an2 |

if n1 , n2 > max(m, n0 ). Also cc1 = (0, 0, 0, . . . , 1, 1, . . .) which proves


that cc1 (1, 1, 1, . . . (modY ). This proves that Y is a maximal ideal
and, therefore,

1) R/Y is a field k.
k is called the completion of k under the given order. For every
a k, consider the Cauchy sequence a = (a, a, . . .). Then a a is a

non-trivial homomorphism of k into k.

Since k is a field, this is an isomorphism. k thus contains a subfield 193


isomorphic to k. We shall identify it with k itself.

7. Formally real fields

168
(2) k is an extension field of k.

We shall now make k an ordered field with an order which is the


extension of the order in k. To this end, define a sequence c = (a1 , . . . , )
of R to be positive if there is an > 0 and an n0 = n0 () such that
an > ,

n > n0 .

If b = (b1 , . . .) is a null sequence, then


|bn | < /2 ,

n > m = m().

If p > max(m, n0 ), then


a p + b p > /2
which shows that a + b is also a positive sequence. The definition of
positive sequence, therefore, depends only on the residue class mod Y .
Let P denote the set of residue classes containing the positive sequences and the null sequence. We shall show that P determines an
order in k
First, let c1 and c2 be two positive sequences. There exist 1 > 0,
2 > 0 and two integers n1 and n2 such that
a p > E1 ,

p > n1 ,

b p > E2 ,

p > n2 ;

if p > max(n1 , n2 ), then


a p + b p > 1 + 2 ,
194

so that c1 + c2 is a positive sequence or P + P P. In a similar manner,


PP P.
Let now c = (a1 , . . .) be not a null sequence. Then c and c cannot
both be positive. For then, there exist , both positive and integers n,
n such that

p>n
a p > ,

a p > , p > n.

4. Completion under an order

169

If p > max(n, n1 ), then


0 = a p a p > + > 0,
which is absurd. Thus P (P) = (0).
Suppose now that c = (a1 , . . .) is not a null sequence. Suppose it
is not positive. Then c = (a1 , a2 , . . .) is a positive sequence. For,
otherwise, for every and every n,
a p < ,

p > n.

But, c being not positive, we have


a p < ,

p > n1 .

c being a Cauchy sequence, there exists n0 with


|a p aq | < ,

p, q > n0 .

Hence, if p > max(n0 , n1 ),


|a p | |a p aq | + |aq | < 2
which means that c is a null-sequence. This contradiction proves that

P (P) = k.
We therefore see that

195

(3) k is an ordered field with an order extending the order in k.


We shall denote the element (a1 , . . .) in k by a and write
a = lim an .
n

Then, clearly, given any > 0, there exists n0 such that the element
a (an0 ) in k has all its elements, from some index on, less than in
absolute value. This justifies our notation. One clearly has
lim an + lim bn = lim an + bn .

7. Formally real fields

170

lim an . lim bn = lim an + bn .


lim an
an
= lim ,
lim bn
bn
if (b1 , b2 , . . .) is not a null sequence.
If is the rational number field, it is ordered and the completion
under this unique order is called the real number field.
This method of construction of the real number field goes back to
Cantor.

5 Archimedian ordered fields


Ordered fields can be put into two classes.
A field k is said to be archimedian ordered if, for every two elements
a, b in k, a > 0, b > 0 there exists an integer n such that
nb > a
196

and, similarly, there is an integer m with ma > b.


We may state equivalently that, for every a > 0, there is an integer n
with
n > a.
A field k is said to be non-archimedian ordered if there exists a > 0
such that
a>n
for every integer n.
, the rational number field is archimedian ordered. Consider, now,
the ring [x] of polynomials. For
f (x) = a0 xn + a1 xn1 + + an , a0 , 0
define f (x) > 0, if a0 > 0 as a rational number. That this is an order can
be verified easily. Also the order is non-archimedian because
x2 + 1 > n,

5. Archimedian ordered fields

171

for every integer n. This order can be extended to (x).


This examples shows that an archimedian order in k can be extended
into a non-archimedian order in an extension field which is transcendental over k. That this is not possible in algebraic extensions is shown by
Theorem 10. If k is archimedian ordered and K/k is algebraic with
an order extending the order in k, the extended order in K is again
archimedian.
Proof. Let be > 0 in the extended order in K. Let f (x) = xn + a1 xn1 +
an be the minimum polynomial of in k[x]. Consider the quantities 197
1ai , i = 1, . . . , n. They are in k and since k and is archimedian ordered,
there exists an integer t > 0 such that
t > 1 ai , i = 1, . . . , n.
We now assert that < t. For if t, then
0 = f () = n + a1 n1 + . . . + an
n + (n1 + . . . +)(1 t)

n + (n1 + . . . + 1)(1 ) = 1,
which is absurd. Our theorem is proved.
We had already introduced the real number field. It is clearly archimedian ordered. In fact, the completion of an archimedian ordered field
is archimedian. We can prove even more, as shown by

Theorem 11. Every complete archimedian ordered field is isomorphic
to the field of real numbers.
Proof. Let K be an archimedian ordered field. It has a subfield isomorphic to , the field of rational numbers. We shall identify it with it
self. Let k be the completion of k. Then

k .


7. Formally real fields

172

198

In order to prove the inequality the other way round, let R be the
ring of Cauchy sequences in k and Y , the maximal ideal formed by
null sequences. We shall show that every residue class of R/Y can be
represented by a Cauchy sequence of rational numbers. Therefore, let
c = (a1 , a2 , . . .) be a positive Cauchy sequence in k. Since k is archimedian ordered, there exists, for every n greater than a certain m, an integer
n such that
n < nan < 1 + n
which means that



a n < 1 .
n n n
m m+1
,
, . . .) be a sequence of rational numLet d = (0, 0, 0, . . .
m m+1
bers. Let > 0 be any positive quantity in k. There exists, then, an
integer t such that t > 1, since k is archimedian ordered. Then, for
n > t and m,


an n <
n

which shows that c d Y , which proves that k .


We shall now prove the important
Theorem 12. The real number field is real closed.
Proof. We shall show that every polynomial which changes sign in

has a root in .


Let b < c be two elements of and f (x) a polynomial in [x]


with
f (b) > 0 and f (c) < 0. We define two sequences of rational numbers
b0 , b1 , b2 , . . . and c0 , c1 , c2 , . . . in the following way. Define the integers
0 , 1 , 2 , . . . inductively in the following manner. 0 = 0, and having
defined 0 , 1 , . . . , n1 , n is defined as the largest integer such that
(c b)n
) 0.
2n
(c b)n
We shall put bn = b +
and since we want bn to be an
2n
increasing sequence, we shall, in addition, require that
n
n1 + 1.
2
f (b +

5. Archimedian ordered fields

173
199

That such a sequence can be found is easily seen. Put now


cn = bn +

cb
.
2n

Then c0 , c1 , . . . is a decreasing sequence


b bn bn+1 cn+1 cn c.
Also (bn ) and (cn ) are Cauchy sequences; for,
| bn+1 bn |<

cb
.
2n+1

Furthermore, the sequence (cn bn ) is a null sequence. For,


cn = bn =

cb
2n

There is, therefore, an between b and c such that


= lim bn = lim cn .
By definition of n , f (cn ) < 0 and f (bn ) o.
Therefore
f () = lim f (bn ) 0, f () = lim f (cn ) < 0.
This shows that f () = 0. Hence is real closed.
We have the
is algebraically closed.
Corollary. (i)
is the complex number field. We have thus proved the funda(i)
mental theorem of algebra.
The algebraic closure of in is called the field of real algebraic
numbers and is real closed.

Chapter 8

Valuated fields
1 Valuations
200

Let k be a field. A valuation on k is a function | | on k with values


in the real number field satisfying
(1) | 0 |= 0
(2) | a |> 0, if a , 0
(3) | ab |=| a | | b |
(4) | a + b || a | + | b |
where a and b are elements in k. It follows that a | a | is a homomorphisms of k into the multiplicative group of positive real numbers. If
we denote by 1 the unit element of k, then
| 1 |=| 12 |=| 1 |2 =| 1 | | 1 |
so that | 1 |= 1. If is a root of unity, say an n say an n th root of unity,
then
1 =| n |=| |n
and so | |= 1. Thus means that | 1 |= 1 and so, for a k,
| a |=| a | .
175

8. Valuated fields

176
Also, since a = a b + b, we get
| a | | b | a b | .

A valuation is said to be trivial if | a |= 1 for all a , 0.


Two valuations | |1 and | |2 are said to be equivalent if for every a , 0
in k,
| a |1 < 1 | a |2 < 1,

| a |1 = 1 | a |2 = 1.
201

It is obvious that the above relation between valuations is an equivalence relation. All valuations equivalent to a given valuation form an
equivalence class of valuations.
If | | is a valuation then, for 0 c 1, | 1 |c is also a valuation. We
shall now prove
Theorem 1. If ||1 and ||2 are equivalent valuations, there exists a real
number c > 0 such that
| a |1 =| a |c2
for all a k.
Proof. Let us assume that ||1 is non-trivial. Then ||2 is also non-trivial.
Also, there exists a b k such that | b |1 > 1, | b |2 > 1. Let a k, a , 0.
Then | a |1 and | b |1 being positive real numbers.
| a |1 =| b |1
where =

log | a |1
.
log | b |1

We approximate to the real number from below and from above


m
by means of rational numbers. Let < . Then
n
| a |1 >| b |m/n
1

2. Classification of valuations

177

an
which means that | m |1 > 1. Since ||1 and ||2 are equivalent, this means
b
that
| a |2 >| b |m/n
.
2
In a similar manner, if p/q > , then
p/q

| a |2 <| b |2
This means that if

.
202

m
p
and ,then
n
q
| a |2 =| b |2 .

Therefore =
Putting c =

log | a |2
log | a |1
log | b |2
. This shows that
=
.
log | b |2
log | a |2
log | b |2

log | b |1
and observing that c > 0, our theorem follows.
log | b |2

2 Classification of valuations
A valuation is said to be archimedian if for every a k, there exists an
integer n = n(a) (that is ne, if e is the unit element of k) such that
| a |<| n | .
(Compare this with archimedian axiom in ordered fields).
A valuation of k which is not archimedian is said to be non-archimedian. We shall deduce a few simple consequences of these definitions.
1) || is archimedian there exists an integer n in k such that | n |> 1.
Proof. If || is archimedian, there exists a k with | a |> 1 and an
integer n with | n |>| a |. This means that
| n |> 1.


8. Valuated fields

178

Let || be a valuation and n, a rational integer so that | n |> 1. Let a any


element of k. If | a | 1, then clearly | a |< |n |. Let | a |> 1. Since
archimedian axiom holds in real number fields, we have an integer
m with
| a |<| n | m.
203

If m = 1, there is nothing to prove. So let m > 1. Then m =| n |


log m
( =
> 0). Let be a positive integer greater than . Then
log | n |

m <| n | =| n |. Therefore
| a |<| n | m <| n |+1
and our assertion is proved.
We deduce
2) || non-archimedian | n | 1, for every integer n in k.
This shows at once that
3) All the valuations of a field of characteristic p are non-archimedian.
We now prove the important property
4) || is an non-archimedian valuation if and only if for every a, b in k
| a + b | Max(| a |, | b |).
Proof. If || is a non-archimedian valuation, then for every integer
n, | n | 1. Let m be any positive integer. Then
m1
(a + b)m = am + (m
b + . . . + bm
1 )a

so that
| a + b |m | a |m + | a |m1 | b | + . . . + | b |m
(m + 1)Max(| a |m , | b |m ).

2. Classification of valuations

179

Taking m th roots and making m we get


| a + b | Max(| a |, | b |).
The converse is trivial, since | n |=| 1 + + 1 | 1.

We deduce easily

204

5) If || is non-archimedian and | a |,| b |, then


| a + b |= Max(| a |, | b |).
Proof. Let, for instance, | a |>| b |. Then
| a + b | Max(| a |, | b |) =| a |
Also a = a + b b, so that
| a | Max(| a + b |, | b |).
But, since | a |>| b |, | a + b || b |. Thus | a || a + b | and our
contention is proved.

More generally, we have, if | a1 |>| a j |, j , 1, then
| a1 + a2 + . . . an |=| a1 | .
In the case of non-archimedian valuations, many times, the so-called
exponential valuation is used. It is defined thus: If ||0 is an non-archimedian valuation, define the function || by
| a |= log | a |0 ,

a , 0.

This has a meaning since | a |0 > 0 for a , 0. We introduce a quantity


which has the property
+=

+a=

8. Valuated fields

180

for any real number a and

a
=0

for any real number a. Then || satisfies


1) | 0 |=
2) | a | is a real number
3) | ab |=| a | + | b |
4) | a + b | Min (| a |, | b |).
205

Then |1| = 0, || = 0 for a root of unity and the valuation is trivial


if |a| = 0 for a , 0. For two valuations ||1 and ||2 which are equivalent
|a|1 = c|a|2 ,
ci0 being a real number.

3 Examples
First, let be the finite field of q elements. Every element of satisfies
the polynomial xq1 1. Therefore, any valuation | | on is trivial.
Let now be the field of rational numbers.
Let | | be an archimedian valuation on . It is enough to determine its
effect on the set of integers in . There is an integer n such that |n| > 1.
Let m be any positive integer. Then
n = a0 + a1 m + + at mt ,
"

#
log n
where t
, 0 ai < m. Therefore |ai | ai < m so that
log m
|n| m(t + 1) max(1, |m|t ).

3. Examples

181

Replace n by nr , where r is a positive integer. Then again, we have


1
1
log n
+ 1) r max(1, |m|t ).
|n| m r (r
log m
Making r , we get
log n

|n| max(1, |m| log m ).


This proves that, since |n| > 1,
log n
|n| |m| log m
and |m| > 1.
206
Now we can repeat the argument with m and n interchanged and
thus obtain
log n
log
m
|m| |n|
Combining the two inequalities we get
log |n| log |m|
=
.
log n
log m
Since m is arbitrary, it follows that
|m| = mC
where c > 0 is a constant. Obviously, the valuation is determined by
its effect on positive integers. From the definition of equivalence, || is
equivalent to the ordinary absolute value induced by the unique order in
.
Let now || be a non-trivial non-archimedian valuation. It is enough
to determine its effect on Z, the ring of integers.
Since | | is non-trivial, consider the set Y of a Z with |a| < 1. Y
is an ideal. For,
|a| < 1, |b| < 1 |a + b| Max(|a|, |b|) < 1.

8. Valuated fields

182
Also, if |a| < 1 and b Z, then
|ab| = |a| |b| < 1.

Furthermore, if ab Y , then |ab| < 1. But |ab| = |a||b| and |a| 1,


|b| 1, since valuation is non-archimedian. Hence |a| < 1 which means
that Y is a prime ideal. Thus g = (p) generated by a prime p. Since, by
definition of Y , M < 1 p/n, we have, if n = p n1 , (n1 , p) = 1,
|n| = |p| .
207

a
If we denote |p| by c, 0 < c < 1, then for any rational number , the
b
value is

a = c
b

a
a
where = p where (a , p) = 1 = (b , p) and a rational integer.
b
b
This valuations is called the p-adic valuation.
Thus with every non-archimedian valuation, there is associated a
prime number. Conversely, let p be any prime number and let n be any
integer, n = p n1 , 0, where (n1 , p) = 1. Put
|n| = |p|

, 0 < |p| < 1.

Then | | determines a non-archimedian valuation on . Further, if p


and q are distinct primes, then the associated valuations are inequivalent.
For, if | | p and | |q are the valuations, then
|q| p = 1,

|q|q < 1.

We shall denote the valuation associated with a prime p, by || p . Then


we have the
Theorem 2. The ordinary absolute valuation and the p-adic valuation
by means of primes p form a complete system of in-equivalent valuations
of the rational number field.

3. Examples

208

183

We shall denote the ordinary absolute valuation by || .


Let us consider the case of a function field K over a ground field
k and let L be the algebraic closure of k in K. L is called the field of
constants of the function field K. The valuations on K that we shall
consider shall always be such that
|a| = o
for a k. (We consider exponential valuation). Hence valuations of
function fields are always non-archimedian, since the prime field is contained in k.
Let, now, be in L. Then satisfies the equation
n + a1 n1 + + an = o
where a1 , . . . , an k. If || is a valuation of K then
||n min(||n1 , . . . , |1|).
From this, it follows that || 0. Also, since 1/ is algebraic,
|| 0. Hence for all L
|| = 0.
Thus the valuation is trivial on the field of constants.
We shall consider the simple case where K = k(x), x transcendental
over k. Here L = k. Let || be a valuation and let |x| < 0. Let f (x) =
a0 xn + + an be in k[x]. Then
f (x) min(|x|n , |x|n1 , . . . , |1|).
Therefore
f (x) = n|x|.
This means that, if R(x) =

g(x)
is an element of K, then
h(x)

|R(x)| = (deg h(x) deg g(x))(|x|).


We denote this valuation by || .

8. Valuated fields

184

Suppose now that |x| 0. As in the case of rational integers, consider the subset Y of k[x] consisting of polynomials f (x) with | f (x)| >
0. Then, since for every (x) in k[x], |(x)| 0, it follows that Y is a
maximal ideal generated by an irreducible polynomial p(x). As in the 209
case of the rational number field
|R(x)| = |p(x)|
A(x)
where A(x) and B(x) are prime to p(x) and
B(x)
is a rational integer. If we denote, by || p(x) , this valuation and put
= ord p(x) R(x), then

where R(x) = {p(x)}

|R(x)| = c ord p(x) R(x)


where c = |p(x)| p(x) > 0. Every irreducible polynomial also gives rise to
a valuation of this type. Hence
Theorem 3. A complete system of inequivalent valuations of k(x) is
given by the valuations induced by irreducible polynomials in k[x] and
the valuation given by the difference of degrees of numerator and denominator of f (x) in k(x).

4 Complete fields
Let k be a field and || a valuation of it. The valuation function defines
on k a metric and one can complete k under this metric. The method is
the same as in the previous chapter and we give here the results without
proofs.
A sequence (a1 , a2 , . . .) of elements of k is said to be a Cauchy sequence if for every > 0, there exists an integer n = n() such that
|an1 an2 | <

, n1 , n2 > n.

It is a null sequence if for every > 0, there is an integer n such that


|am | < ,

m > n.

4. Complete fields

185

210

The Cauchy sequences form a commutative ring R and the null sequences a maximal ideal Y , therein. The quotient k = R/Y is called
the completion of k under ||. The mapping a (a, a, a, . . .) is an isomorphism of k in k and we identify this isomorphic image with k itself.
We extend to k the valuation || in k, in the following manner. The real
number field is complete under the valuation induced by the unique
then a = (a1 , a2 , . . .), a Cauchy sequence of
order on it. So, if a k,
elements of k. Put
|a| = lim |an |
n

That this is a valuation follows from the properties of limits in .


Also, the extend valuation is archimedian or non-archimedian, according as the valuation in k is archimedian or non-archimedian.
For instance, if || on k is non-archimedian and
a = (a1 , a2 , . . .)
b = (b1 , b2 , . . .)
then
are two elements of k,
a + b = (a1 + b1 , a2 , +b2 , . . .)
and
|a + b| Max(|a|, |b|) = lim (|an + bn | Max(|an |, |bn |))
n

which is certainly 0,
k is thus a complete valuated field.
It may also be seen that the elements of k are dense in k in the topol- 211
ogy induced by the metric.
We denote by S n the partial
Let c1 + c2 + c3 + be a series in k.
sum
S n = c1 + c2 + + cn .
We say that c1 + c2 + + cn + is convergent if and only if the
sequence of partial sums S 1 , S 2 , . . . , S n , . . . converges. This means that,
for every > 0, there exists an integer n = n() such that
|S m S m | = |cm+1 + + cm | < , m, m > n.

8. Valuated fields

186

Obviously c1 , c2 , . . . is a null sequence in k.


In case the valuation is non-archimedian, we have the following
property:1) The series c1 + c2 + + cn + is convergent if and only if
c1 , c2 , . . . is a null sequence.
Proof. We have only to prove the sufficiency of the condition. Suppose
that cn 0; then, for large n and m,
|S n S m | = |cm+1 + cm+2 + + cn |
max(|cm+1 |, . . . , |cn |)
and so tends to zero. This proves the contention.

212

Note that this theorem is false, in case the valuation is archimedian.


Let k be a field and || a valuation on it. a |a| is a homomorphism
of k into multiplicative group of positive real numbers. Let G(k) denote
this homomorphic image. This is a group which we call the value group
of k for the valuation. If k is the completion of k by the valuation in k,
is value group of k.

then G(k)
Suppose || is an archimedian valuation. Then k has characteristic
zero and contains , the rational number field as a subfield. On , || is
the field of real numbers,
the ordinary absolute value. Since k contains ,
it follows that
is the multiplicative group of all positive real numbers.
1) G(k)
Also, because of the definition of the extended valuation, it follows

that G(k) is dense in the group G(k).


We shall now assume that || is a non-archimedian valuation. We consider the exponential valuation. Then G(k) is a subgroup of the additive group of all real numbers. We shall now prove
= G(k).
2) G(k)
Then a = (a1 , . . .) is a Cauchy sequence,
Proof. For, let 0 , a k.
not a null sequence in k. By definition,
|a| = lim |an |
n

4. Complete fields

187

Now an = an a + a so that
|an | min(|an a|, |a|).
But, for n large, |an a| > |a| so that |an | = |a| and our contention is
established.

G(k) being an additive subgroup of the real number field is either
dense or discrete. The valuation is then called dense or discrete accordingly. In the second case, there exists in k with smallest pos- 213
itive value || || is then the generator of the infinite cyclic group
G(k) is called a uniformising parameter. It is clear that is not
unique. For, if u k with |u| = 0, then |u| = || and u is also a
uniformising parameter.
Consider in k the set O of elements a with |a| o.O is then an
integrity domain. For,
|a| 0, |b| 0 |a + b| Min(|a|, |b|) 0.
Also |ab| = |a| + |b| 0. We call O the ring of integers of the
valuation. Consider the set Y of elements a k with |a| > 0. Then
Y is a subset of O and is a maximal ideal in O. For, if a r and
b Y , then |ab| = |a| + |b| > 0. Also, if a r but not in Y , then
|a| = 0 and |a1 | = 0 so that if U is an ideal in O containing Y , then
U = Y or U = r. We call Y , the prime divisor of the valuation.
Since Y is maximal, r/Y is a field. We call /Y the residue class
field.
We denote by O the
Exactly the same notions can be defined for k.

ring of integers of the valuation so that O is the set of a k with


hence the set of a k with
|a| 0.Y is the maximal ideal in O,

|a| > 0. Also O/Y is the residue class field. Clearly


Y = Y r
We now have

8. Valuated fields

188

has the property; a r(r) or a1 Y (Y ).


2) Every a k(k)

|a| < 0 so that |a1 | > 0 and hence 214


This is evident since if a < O(O),
1

a Y (Y ).

is integrally closed in k(k).

3) O(O)
which is a root of a
Proof. We should prove that every in k(k)
n
n1
is
polynomial of the type x + a1 x + + an , a1 , . . . , an O(O),

already in O(O). For, suppose


n + a1 n1 + + an = o
Then 1 Y (Y ). Therefore
and < O(O).
1 = (a1 1 + a2 2 + + an n )
and so
o = |1| min(|a1 1 |, . . . |an n ) > o
which is absurd

4) Every element in O is the limit of a sequence of elements in O and


conversely.
a1 , . . . , an , . . . in k. Then, as
Proof. Let a = (a1 , . . . , an , . . .) be in O,
we saw earlier, for sufficiently large n
|an | = |a|.
But |a| 0 so that |an | 0. Thus, for sufficiently large n, all an s are
in O. Converse is trivial.
We now prove
Y .
5) There is a natural isomorphism of O/Y on O/

4. Complete fields

215

189

Proof. The elements of O/Y are residue classes a + Y , a r. We


now make correspond to a + Y the residue class a + Y . Then a + Y
Y . The mapping a + Y a + Y is a homomoris an element of O/
Y . Since both are fields, this
phism (non-trivial) of O/Y into /
is an isomorphism into. In order to see that it is an isomorphism of
Y , let a + Y be any residue class in O/
Y . By (4)
O/Y onto O/
therefore, given any N > 0, there exists a with
|a a| > N > 0
or a a Y . If we then take a + Y , then a + Y is the image of
a + Y . This proves that the mapping is an isomorphism onto.

Y ,
(5) enables us to choose, in k itself, a set of representatives of /
the residue class field. We shall denote this set by R and assume that
it contains the zero element of O. Also it has to be observed that, in
general, R is not a field. It is not even an additive group.
We now assume that | | is a discrete valuation. Let in O be a uniformising parameter. Then clearly
Y = ()
Then |a|/|| is a rational integer, since
is a principal ideal. Let a k.
G(k) is an infinite cyclic group. Put |a|/|| = t. Then |at | = 0. If
we call elements u in k with |u| = 0, units, then
a = t u
where u is a unit.
Then
Let R be the set defined above and a .
a a0 (

mod Y )

with a0 R. This means that (a a0 )1 is an integer (in ).


(a a0 )1 a1 (modY ),

216

8. Valuated fields

190
where a1 R. Then
a a0 + a1 (modY 2 )
In this way, one proves by induction that

a a0 + a1 + + am m (mod)Y m+1 )
where a0 , a1 , . . . , am R. Put bm = a0 + a1 + + am m .
Then a bm (modY m+1 ) which means that
|a bm | m + 1
Consider the series
b0 + (b1 b0 ) + (b2 b1 ) +
Then, since bm+1 bm = am+1 m+1 , we see that |bm+1 bm | increases
indefinitely. Hence the above series converges. Also, since its elements are integers,
b = b0 + (b1 b0 ) + (b2 b1 ) +

is an element of O.
Since b0 + (b1 b0 ) + + (bm bm1 ) = bm , it follows that
b = lim bm .
m

Thus we have
a = lim bm = a0 + a1 + a2 2 +
m

217

By the very method of construction this expression for a is unique,


once we have chosen R and .
at O for some rational integer t. Hence we have
If a k,

4. Complete fields

191

6) Every element a in k has the unique expression


a=

an n

n=t

an R and in O is a generator of Y . t is a rational integer.


If t > 0, we shall call
h(a) =

1
X

an n

n=t

the principal part of a. If t 0 we put h(a) = 0. Clearly, h(a) is an

element in k. Also a h(a) O.


We now study the two important examples of the rational number
field and the rational function field of one variable.
Let be the field of rational numbers. We shall denote by || the
ordinary absolute value and by || p the p-adic value for p, a prime. If
a is an integer, a = p n1 , (p, n1 ) = 1.
We put
|a| p = log p
and
|a| = absolute value of a.
It is then clear that each of the non-archimedian valuations of is
discrete. Let us denote by the completion of by the archimedian
valuation and by p the completion of by the p-adic valuation.
is clearly the real number field.
If O p denotes the set of integers of p and Y the prime ideal of the
valuation then
Y = (p)
A set of representatives of O p mod Y is given by the integers 0, 1, 2, 218
. . . , p 1 as can be easily seen. Hence, by (6),

8. Valuated fields

192
7) Every a p is expressed uniquely in the form
a=

an pn

n=t

where ai = 0, 1, 2, . . . , p 1.
The elements of p are called the p-adic numbers of Hensel.
As before, we denote, by h p (a), the principal part of a at p. Clearly
a h p (a) is a p-adic integer.

b
Let a be a rational number, a p |a| p 0, that is a = ,
c
(b, c) = 1 and c is prime to p. Since only finitely many primes divide
b and c, it follows that a p for almost all p, that is except for a
finite number of p. Hence h p (a) = 0 for all except a finite number of
primes. Hence for any a
X
h p (a)
p

has a meaning. Also, a h p (a) is a rational number whose denomP


inator is prime to p. Hence a h p (a) is a rational number whose
p

denominator is prime to every rational integer. Hence

8) For every rational number a


X
a
h p (a) 0(mod1).
p

This is the so-called partial fraction decomposition of a rational


number.
219

Let now k be an algebraically closed field and K = k(x) the field


of rational functions of one variable x. All the valuations are nonarchimedian. Every irreducible polynomial of k[x] is linear and of
the form x a. With every a k there is the valuation ||a associated,
which is defined by
| f (x)|a = ,

4. Complete fields

193

where f (x) k[x], f (x) = (x a) (x), (a) , 0. If we denote by


|| the valuation by degree of f (x), then, for f (x) k[x],
| f (x)| = deg f (x)
Let Ka and K denote the completions, respectively, of K at ||a and
|| . If Oa and O are the set of integers of Ka and K , Ya and Y
the respective prime divisors, then
1
Ya = {x a}, Y = { }.
x
It is clear, then, that a /Ya and /Y are both isomorphic to k and
since K contains k, we may take k itself as a set of representatives of
the residue class field. Any element f in Ka is uniquely of the form
f =

n=t

an (x a)n ,

an k. Similarly, if K ,
=

bn xn .

n=t

As before, if we denote by ha ( f ) and h ( f ) the principal parts of


f K for the two valuations, then
X
ha ( f ) + h ( f )
a

has a meaning since ha ( f ) = 0 for all but a finite number of a


If we define K to be regular ar a() if a ( ), then for 220
f K,
X
f
ha ( f )
a

where a may be infinity also, is regular at all, a k and also for the
valuation || . Such an element, clearly, is a constant. Hence

8. Valuated fields

194
9) If f K then

ha ( f ) = constant

Conversely, it is easy to see that there exists, up to an additive constant, only one f K which is regular for all a k except a1 , . . . , an
(one of which may be also) and with prescribed principal parts at
these ai . 9) gives the partial fraction decomposition of the rational
function f .

5 Extension of the valuation of a complete


non-archimedian valuated field
We shall study the following problem. Suppose k is compute under a
valuation ||. Can this valuation be extended, and if so in how many
ways, to a finite algebraic extension K of k ?
We prove, first
Lemma 1. Let k be a field complete under a valuation ||, and K a finite
algebraic extension of k. Let 1 , . . . , n be a basis of K/k. Let || have
an extension to K and
n
X
=
aiv i , aiv k,
i=1

221

v = 1, 2, 3, . . . be a Cauchy sequence in K. Then aiv i = 1, 2, . . . , n


are Cauchy sequence in K.
Proof. We consider the Cauchy sequence {v },
v =

m
X
i=1

aiv i , aiv , k

1 m n and we shall prove that the {aiv } are Cauchy sequences in k.


We use induction on m. Clearly, if m = 1,
v = aiv 1
and {v } is a Cauchy sequence in K if and only if {aiv } is a sequence in
k.


5. Extension of the valuation of a complete. . .

195

Suppose we have proved our statement for m 1 1, instead of m.


Write
m1
X
v =
aiv i + amv m .
i=1

If {amv } is a Cauchy in k, then {v amv m } is a Cauchy sequences


in K and induction hypothesis works. Let us assume that amv is not a
Cauchy sequence in k. This means that there exists a > 0 and for every
v, an integer v such that
> v
and
|am v amv | > .
Consider now the sequence {v } in K with
v =

K.

v v
.
amv amv

Because of the above property, we see that {v } is a null sequence in


Now
v m =

m1
X
i=1

!
ai v aiv
i
am v amv

and {v m } is a Cauchy sequence in K. Induction hypothesis now 222


works and so, if
!
aiv aiv
= bi ,
lim
v amv amv
then
m =

X
i=1

bi i ,

bi k.

This is impossible because 1 , . . . , m are linearly independent over


k. Therefore {amv } is a Cauchy sequence and our Lemma is thereby
proved.
We shall now prove the following theorem concerning extension of 223
valuation.

8. Valuated fields

196

Theorem 4. If the valuation || of a complete field k can be extended to a


finite extension K, then this extension is unique and K is complete under
the extended valuation.
Proof. That K is complete under the extended valuation is easy to see.
For, if {v } is a Cauchy sequence in K and
X
v =
aiv i , aiv k.,
i

by the lemma, the {aiv }s are Cauchy sequences. So, if


lim aiv = bi k,

then
lim =

X
i

1 lim aiv =
v

bi i

which is again in K.

We shall now prove that the extended valuation is unique.


From the lemma, it follows that if {v } is a null sequence in K and
P
v = aiv i , aiv k, then the aiv s are null sequences in k.
i

In particular, if K and || < 1 in some extension the valuation


P
|| in k, then , 2 , 3 , . . . is a null sequence in K. If m = a(m)
1 i ,
1

a(m)
k, then ai(m) , i = 1, . . . , n are null sequence in k.
i
P
If = xi i is a general element of x1 , . . . , xn . Put
X
t =
x(t)
i i ;
i=1

t
then NK/k t is the same polynomial in x(t)
t , i = 1, . . . , n as N K/k is in
x1 , . . . , xn . If now || < 1, is an extended valuation, then the {x(i t)} are
null sequences. Hence N, N2 is a null sequences in k, But Nt =
(N)t so that (N), (N)2 , . . ., is s a null sequence in k. This means that
|N| < 1. We have, thus, proved that if in K is such that || < 1 is an
extended valuation, then |N| < 1 in k.

5. Extension of the valuation of a complete. . .

197

In a similar manner, if || > 1, then |N| > 1. Thus we get |N| =


1 || = 1.
n
where n = (K : k)
Let now be in K and write =
N
(N)n
Then N =
. Thus
(N)n
|N| = 1.
By the above, it means that || = 1 in the valuation. Hence
p
n
|| = |N|

showing that the value of in the extended valuation is unique fixed.


224
Our theorem is thus completely proved.
In order to prove that an extension of the valuation is possible, we
shall consider the case where k is complete under a discrete
non-archimedian valuation. Let O be the ring of integers Y , the prime
divisor of the valuation and O/Y , the residue class field. We shall now
prove the celebrated lemma due Hensel
Lemma 2. Let f (x) be a polynomial of degree m in O[x], g0 (x), a monic
polynomial of degree r 1 and h0 (x), a polynomial of degree m r
both with coefficients in O such that
1) f (x) go (x)ho (x) (modY )
2) go (x) and ho (x) are coprime mod Y
Then there exists polynomials g(x) and h(x) in O[x] such that

g(x) go (x)

(mod Y ),

h(x) ho (x)

g(x) has the same degree as go (x) and f (x) = g(x) h(x).

Proof. We shall now construct two sequences of polynomials


go (x), g1 , (x), . . . and ho (x), h1 , (x), . . . satisfying
gn (x) gn1 (x)(

mod Y n )

8. Valuated fields

198
hn (x) hn1 (x)(

mod Y n )

fn (x) gn (x)hn (x)(

mod Y n+1 )

gn (x) is a monic polynomial of degree r and hn (x) of degree m r. All


the polynomials have coefficients in O.

225

The polynomials are constructed inductively. For n = 0, go (x) and


ho (x) are already given and satisfy the conditions. Assume now that
go (x), . . . , gn1 (x) and ho (x), . . . , hn1 (x) have been constructed so as
satisfy the requisite conditions.
Since gn1 (x) go (x)(mod Y ) and hn1 (x) ho (x)(mod Y ) and
g0 (x) and ho (x) are coprime mod Y , there exists, for any polynomial
fn (x) in O[x], two polynomials L(x) and M(x) with
fn (x) L(x)gn1 (x) + M(x)hn1 (x) (mod Y ).
L(x) and M(x) are clearly not uniquely determined. We can replace L(x)
by L(x) + (x)hn1 (x) and M(x) by M(x) + (x)gn1 (x).
Let be a generator of the principal ideal Y . By induction hypothesis,
fn (x) = n ( f (x) gn1 (x)hn1 (x))
is an integral polynomial, so in O[x]. Since gn1 (x) has degree r and is
monic and hn1 (x) degree m r, it is possible to choose M(x) and L(x)
so that M(x) has degree < r and L(x) degree m r. Put now
gn (x) = gn1 (x) + n M(x),
hn (x) = hn1 (x) + n L(x).
Then gn (x) is monic and of degree r, since M(x) has degree < r.hn (x)
has degree m r. Now f (x) gn (x)hn (x) = f (x) gn1 (x)hn1 (x)
n (gn1 (x)L(x) + hn1 (x)M(x))( mod Y n+1 ) By choice of L(x) and
M(x), it follows that
f (x) gn (x)hn (x)(mod Y n+1 ).

226

We have thus constructed the two sequences of functions. Put now

5. Extension of the valuation of a complete. . .

199

g(x) = go (x) + (g1 (x) go (x)) + + (gn (x) gn1 (x)) +

h(x) = ho (x) + (h1 (x) ho (x)) + + (hn (x) hn1 (x)) +

Since gn (x) gn1 (x) = 0(modn ), it follows that the corresponding coefficients of the sequence of polynomials g0 (x), g1 (x), . . . form
Cauchy sequences in O. Since k is complete and
g(x) = lim gn (x),
n

it follows that g(x) [x]. It is monic and is of degree r. In a similar


way, b(x) [x] and has degree m r.
Also since f (x) gn (x)hn (x) 0(modyn+1 ), it follows that the coefficients of ( f (x) gn (x)hn (x)) form null sequences. Hence
f (x) = lim gn (x)hn (x) = hn (x) = g(x)h(x)
n

and our lemma is proved.


We now deduce the following important .
Lemma 3. Let f (x) = xn +a1 xn1 + +an be an irreducible polynomial
k[x] , k satisfying hypothesis of lemma 2. Then f (x) O[x] if and only
if an O.
Proof. It is clearly enough to prove the sufficiency of the condition. Let
an , and if a1 , . . . , an1 (some or all of them) are not in , then there
is a smallest power a , a > 0, of such that
a f (x) = bo xn + b1 xn1 + + bn
is a primitive polynomial in O[x]. Also, now bn 0( mod Y ). and 227
at least one of b0 , . . . , bn1 is not divisible by Y . Let br be the first
coefficient from the right not divisible by Y . Then
a (b0 xr + br )xnr (mod Y )
Since br . 0( mod Y ), Hensels lemma can be applied and we see
that a f (x) is reducible in O[x]. Thus f (x) is reducible in k[x] which
contradicts the hypothesis. The lemma is therefore established.


8. Valuated fields

200

We are now ready to prove the important theorem concerning extension of discrete non-archimedian valuations, namely
Theorem 5. Let k be complete under a discrete non-archimedian valuation || and K a finite algebraic extension over k. Then || can be extended
uniquely to K and then for any in K,
|| =

1
|NK/k |.
(K : k)

Proof. Because of Theorem 4, it is enough to prove that the function


defined on K by
1
|NK/k |
|| =
(K : k)
is a valuation function. Clearly , |0| = ; || is a real number for , 0.
Also,
|| = || + ||.
We shall now prove that
| + | min(||, ||).
228

If or is zero, then the above is trivial. So let , 0, , 0. Since


or is 0, it is enough to prove that if || 0, |1 + | 0.


Let f (x) = xm + a1 xm1 + + am be the minimum polynomial of


in K over k, Then
N = ((1)m am )(K:k()) .
Also, N(1 + ) = (1)n (1 a1 + am )(K:k()) . If |N 0|,
then |am | 0 which , by lemma 3, means that |a1 |, . . . , |am |. Hence
|N(1 + )| 0. Our theorem is proved.
Incidentally it shows that the extended valuation is discrete also.

6. Fields complete under archimedian valuations

201

6 Fields complete under archimedian valuations


Suppose k is complete under an archimedian valuation. Then k has characteristic zero and contains, as a subfield, the completion of the rational number field. bar(i) is, then, the complex number field. Every
On ,
we have the
complex number is to the form a + ib, a, b .
the function
ordinary absolute value. Define in (i)
1

|z| = (a2 + b2 ) 2

where z = a + ib. It is, then, easy to verify that || is a valuation on (i)

which extends the valuation in . Also, by theorem 4, this is the only

extension of the ordinary absolute value. We consider the case k (i)


and prove the theorem of A. Ostrowski.
be the complex number field and let k be a
Theorem 6. Let k (i)

field with archimedian valuation and containing (i).


If the valuation in

k is an extension of the valuation in (i),


then k = (i).

Proof. If k , (i),
let a k but not in (i).
Denote by || the valuation in 229

k. Consider |a z| for all k = (i). Since |a z|1 0, we have


=g

1
b|a z| 0.
2

There exists, therefore, a sequences z1 , . . . , zn , .. of complex numbers such that


lim |a zn | = .
n


But zn = zn a + a and so |zn | |zn a| + |a| which shows that the
|zn |, for large n, are bounded. We may therefore, choose a subsequence
zi1 , zi2 , . . . converging to a limit point zo such that
= lim |a zin | = |a zo |
n

such that b = a z0
We have thus proved the existence of a z0 in (i)

has |b| = . Since, by assumption, a < (i) we have


|b| = > 0.

8. Valuated fields

202
, by definition, being g.l.b., it follows that
|b z|

for z (i).
Consider the set of complex numbers z with |z| < . Let n > 0 be an
arbitrary rational integer and , a primitive nth root of unity. Then
bn zn = (b z)(b z) . . . (b n1 z).
230

Therefore
|b z|n1 |b z||b z| . . . |b n1 z| = |bn zn |
|z|
|b|n + |z|n = n (1 + ( )n ).

Hence

!
|z|n
|b z| 1 + n .

But |z| < and as n is arbitrary, it follows that |b z| .


We therefore have
|z| < |b z| = .
We now prove that for every integer m > 0, |b mz| = if |z| < .
For, suppose we have proved this for m 1 instead of m, then we can
carry through the above analysis with b (m 1)z instead of b, then we
can carry through the above analysis with b (m 1)z instead of b and
then we obtain |b mz| = .
Suppose now that z is any complex number. Then there is an integer
z
m > 0 such that | | < . Therefore
m


b m z = |b z | = .

m

Now z = z b + b and so

|z | |b z | + |b| 2

6. Fields complete under archimedian valuations

203

which shows that all complex numbers are bounded in absolute value.
is false .
This is a contradiction. Hence our assumption that a < (i)
The theorem is thereby proved,
Before proving theorem 7 which gives a complete characterization
of all complete fields with archimedian valuation, we shall prove a couple of lemmas.
Lemma 4. Let k be complete under an archimedian valuation || and 231
in k such that x2 + is irreducible in k[x]. Then |1 + | 1.
Proof. If possible, let |1 + | < 1. We construct, by recurrence, the
sequence c0 , c1 , c2 , . . . , in k, defined as follows: -

cn+1

c0 = 1
1+
n = 0, 1, 2, . . .
= 2
cn

It, then, follows that |cn | 1. For, if we have proved it upto cn1 ,
then
|1 + |
|cn | 2
1.
|cn1 |
Thus cn does not vanish for any n. Also,
|cn+1 cn | =

|1 + ||cn cn1 |
|cn cn1 |
|cn ||cn1 |

where = |1 + | < 1. This means that the series


c0 + (c1 c0 ) + (c2 c1 ) +
converges in k. Let it converge to c in k. Then
c = lim co + + (cn on1 ) = lim cn .
n

1+
. But this means
c
that = c2 + 2c + 1 = (c + 1)2 which contradicts the fact x2 + is
irreducible in k[x].
We now prove the

Therefore, by definition of cn , we get c = 2

8. Valuated fields

204

Lemma 5. If k is complete under an archimedian valuation, ||, then this


valuation can be extended to k(i).

232

Proof. If i k, there is nothing to prove. Let i < k. Then every element


of k(i) is of the form a + ib, a, b k.
The norm from k(i) to k of = a + ib is
N = a2 + b2 .
By theorem 4, therefore, it is enough to prove that
1

|| = |(a2 + b2 ) 2 |
is a valuation on k(i). By putting =
|a2 + b2 | a2 . Therefore

b2
in the lemma 4, we see that
a2

|(1 + a)2 + b2 | 1 + |a2 + b2 | + 2|a|

This shows that

p
1 + |a2 + b2 | + 2 |a2 + b2 |
p
= (1 + |a2 + b2 |)2 .
|1 + | 1 + ||

and our lemma is proved.

We now obtain a complete characterization of complete archimedian


fields, namely,
Theorem 7. The only fields complete under an archimedian valuation
are the real and complex numbers fields.
Proof. k has characteristic zero and since it is complete, it contains the
field of real numbers. If k contains properly, then we assert that k
and,
contains i. For, k(i), by lemma 5, is complete and k(i) contains (i)
by theorem 6,

k(i) = (i).


7. Extension of valuation of an incomplete field

205

Therefore
= k(i) k .

(i)
: )
= 2 so that k = k(i) = (i).

But ((i)
We have thus found all complete fields with archimedian valuation.

7 Extension of valuation of an incomplete field


Suppose k is a complete field under a valuation || and let be its alge- 233
braic closure. Then || can be extended to by the prescription
1

|| = |N| n ,
where Norm is takes form k() over k and n = (k() : k). It is clear that
it defines a valuation function. For, of K is a subfields of and K/k is
finite and K contains then, by properties of norms,
1

|| = |NK/k | m ,
where m = (K : k). So, if and are in , we may take for K a field
containing and and with (K : k) finite.
Furthermore, defined as such, the valuation on is dense because,
if || > 1, then ||1/n has value as near 1 as one wishes, by increasing n
sufficiently. Also, for every n, 1/n is in .
Also, let be an automorphism of /k, and in . Then, by definition of norm,
N = N()
so that || = ||. Thus all conjugates of an element have the same
value.
We shall now study how one can extend a valuation of an incomplete
field to an algebraic extension.
Let k be a field and K a finite algebraic extension of it. Let || be a
234
valuation of k and k the completion of k under this valuation. Let k , k.

8. Valuated fields

206

uK
uu
u
uu

uu k
uu
u
u
uu

Suppose it is possible to extend || to K. Let K be the completion of


K under this extended valuation. Since K K k, it follows that K
Thus
contains K and k and therefore the composite K k.

K K k.
k is a finite extension, since K/k is finite.
On the other hand, K k/
Since k is complete, K k is complete also. K k contains K and hence its
completion K under this extended valuation. Thus

K = K k.
Thus if the valuation can be extended, then the completion of K by

this extended valuation is a composite extension of K and k.


, then, contains
Suppose now that is an algebraic closure of k.
an algebraic closure of k. We have seen above that the given valuation
of k can be extended to . Let be an isomorphism of K/k into .
The valuation in k can be extended to K k which is a subfield of .
Therefore, there is a valuation on K. Define now, for in K,
||o = ||

235

which extension is unique.


where || is the extension of || on k to K k,
It is now trivial to see that ||o is a valuation on K and extends || on k.
Hence every isomorphism of K into which is trivial on k, gives
rise to a valuation of K.
We now inverstigate when two isomorphisms give rise to the same
valuation on K. Let and be two isomorphisms of K/k into giving
the same valuation on K. K and K are subfields of and they have

7. Extension of valuation of an incomplete field

207

the same valuation. Thus = 1 is an isomorphism of K onto K

which preserves the valuation on K. Now is identity on k and so on k.


Since K k is the completion of K, it follows that is an isomorphism
Hence, if and give rise
of the composite extensions K k and K k.
to the same valuation on K, the corresponding composite extensions are
equivalent.
Suppose now that and are isomorphisms of K into such that
the composite extensions K k and K k are equivalent. There exists then
a mapping of K k on K k which is identity on k and such that
=
induces a valuation ||1 on K such that ||1 = || and induces a
valuation ||2 on K such that ||2 = ||. But is such that = or
and are conjugates over k in . Hence
|| = ||
or ||1 = ||2 which shows that , give the same valuation on K.
We have hence the
Theorem 8. A valuation || of k can be extended to a finite extension K
of k only in a finite number of ways. The number of these extensions of 236
|| to K stand in a (1, 1) correspondence with the classes of composite
extensions of K and k
From what we have already seen, the number of distinct composite
extensions of K and k is at most (K : k).
We apply these to the case where k = , rational number field and
K/ finite so that K is an algebraic number filed. From theorem 2,
it therefore follows that K has at most (K : ) distinct archimedian
valuations and that all the non-archimedian valuations of K, which are
countable in number, are discrete.
In a similar manner, if K is an algebraic function field of one variable
over a constant filed k, then all the valuations of K are non-archimedian
and discrete. This can be seen from the fact that if x K is transcendental over k, then K/k(x) is algebraic and one has only to apply theorems
3 and 8.

Appendix
Abelian groups

1 Decomposition theorem
237

All the groups that we deal with here are abelian. Before proving the
main decomposition theorem for finite abelian groups we shall prove
some lemmas.
Lemma 1. If a, b are elements in G and have orders m and n respectively and (m, n) = 1, then ab has order mn.
Proof. Clearly if t is the order of ab, t|mn, since
(ab)mn = (am )n (bn )m = e,
e being unit element of G. Also at = bt . Thus
e = atm = btm
so that n/t. Similarly m|t. Hence t = mn.

Lemma 2. Let p be a prime number dividing the order n of the group


G. Then there is, in G, an element of order p.
Proof. We use induction on n. Let a be an element in G of order m. If
p|m, then am/p has order p and we are through. Suppose p m. Let H
be the cyclic group generated by a G/H has then order n/m which is
n
divisible by p. Since |n, induction hypothesis applies, so that there is
m
a coset Hb of order p. If b has order t then bt = e and so (Hb)t = H
which means that p|t and so bt/p has order p.

209

8. Appendix

210

Lemma 3. Let G be a finite group and the maximum of the orders of


elements of G. Then
a = e
238

for all a G.
Proof. Let b be an element of order . Let a be any element in G and
let be its order. To prove the lemma, it is enough to prove that |. If
not, there is a prime p which divides to a higher power than it does
. Let pr be the highest power of p dividing and ps the highest power
dividing . Then r > s. We will see that this leads to a contradiction. 
Since
order

r are coprime, a pr has the order pr . Similarly b ps has


and
p
pr

. By lemma 1,
ps

c = a pr b p

> , which contradicts the definition of . Hence |.


ps
Lemmas 2 and 3 show that |n where n is the order of the group and
that and n have the same prime factors. is called the exponent of the
finite group G.
A set a1 , . . . , an of elements of a finite group G are said to be independent if
a1x1 anxn = e
has order pr .

239

implies aixi = e, i = 1, . . . , n.
If G is a finite group and is a direct product of cyclic groups G1 ,
. . . , Gn and if ai , i = 1, . . . , n is a generator of Gi , then a1 , . . . , an are
independent elements of G. They are said to form a base of G.
We shall now prove
Theorem 1. Let G be a finite group of order n. Then G is the direct product of cyclic groups G1 , . . . , Gl of orders 1 , . . . , l such that
i |i1 , i = 2, . . . , l, > 1.

Proof. We prove the theorem by induction on the order n of the group


G. Let us therefore assume theorem proved for groups of order < n. Let

1. Decomposition theorem

211

G have order n. Suppose 1 is the exponent of G. If 1 = n, then G


is cyclic and there and there is nothing to prove. Let therefore 1 < n.
There is an element a1 of order 1 . Let G1 be the cyclic group generated
n
by a1 . G/G1 has order
< n. Hence induction hypothesis works on
1
G/G1 .

G/G1 is thus the direct product of cyclic groups W2 , . . . , Wl of order
2 , . . . , l respectively and l |l1 | . . . |2 . Let Hi be a generator of Wi .
Then Hi = Gl bi for some bi in the coset Hi . Let the element bi in G have
order ti . Then ti |1 by lemma 3. But
Hiti = btii Gt1i = G1
which proves that i |1 . Thus l |l1 | . . . |1 .
Let now bi i = a1xi . Put xi = yi zi where (yi , 1 ) = 1 and all the prime
factors of zi divide 1 . Choose ui prime to 1 such that
ui yi 1(mod 1 ).
Then bui i i = az1i . Since 1 is the exponent of G,
e = bui i 1 = (az11 )1 /i .
Since a1 has order 1 this means that i |zi , i = 2, 3, . . . , l.
Put now
i /i
.
ai = bui i az
1
Since ui is prime to 1 and so to i , the coset Fi = Gl ai is also a
generator of Wi . Thus G1 a2 , . . . , G1 al is a base of G/G1 .
Let ai have order fi . Then
f

u f

fi zi /i

e = ai i = bi i i ai

This means that


u f

f zi /i

bi i i = ai i

Therefore by definition of bi , i |ui fi . But (ui i ) = 1. Hence i | fi .

240

8. Appendix

212
On the other hand
ai i = bui i i aizi = e.

Hence fi = i . We have thus elements a1 , . . . , al in G which have


orders 1 , . . . , l satisfying l |l1 | . . . |1 .
We maintain that a1 , . . . , a1 are independent elements of G. For, if
av11 . . . avl l = e,
241

v2
vl
1
then av22 avl l = av
1 which means that F 2 . . . F l = G 1 . But since
F2 , . . . , Fl are independent, i |vi , i = 2, . . . , l. But this will mean that
av11 = e or 1 |v1 .
Since 1 . . . l = n, if follows that a1 , . . . , al form a base of G and
the theorem is proved.
Let G be a group of group of order n and let G be direct product of
cyclic groups G1 , G2 , . . . , Gl of orders 1 , . . . , l . We now prove

Lemma 4. Let be a divisor of n. The number N() of elements a G


with
a = e
is given by
N() =

(, i ).

i=1

Proof. Let a1 , . . . , al be a base of G so that ai is of order i . Any a G


has the form
a = a1x1 alxl .
x

If a = e, then e = a11 al l . Since a1 , . . . , al is a base, this means that


xi 0(mod i ), i = 1, , l. Hence xi has precisely (, i ) possibilities
and our lemma is proved.
We can now prove the


242

Theorem 2. If G is the direct product of cyclic groups G1 , . . . Gl , of


orders 1 , . . . , l respectively with l |l1 | |1 and G is also the direct
product of cyclic groups H1 , . . . , Hm of orders 1 , . . . , m respectively
with m |m1 | |1 , then m = 1 and
i = i , i = 1, . . . , l.

2. Characters and duality

213

Proof. Without loss in generality let l m. Let a1 , . . . , al be a base of G


in the decomposition G1 G2 G1 . Since the number of elements
a with a = e is independent of the decomposition
N() =

(, i ) =

1=1

m
Y
(, j )
j=1

Put now = 1 . Then N() = 11 . But since (, j ) , it follows


that m
, so that l m. This proves
l = m.

Also it follows that each factor (1 , j ) = or | . Inverting the
roles of and we get
= .
Suppose now it is proved that q+1 = q+1 , . . . , 1 = 1 . Then by
putting = q , we have
N() =

q
q .q+1 1

q
Y
=
(q , j )q+1 .
j=1

By the same reasoning as before, it follows that q = q and we are,


therefore, through.
For this reason, the integers 1 , . . . , l are called the canonical invariants of G. From theorems 1 and 2 we have the Corollary Two finite
groups G and G are isomorphic if and only if they have the same canonical invariants.

2 Characters and duality


243

Let G be a group, not necessarily abelian and Z a cyclic group. A homomorphism of G into Z is called a character of G. Thus
(a)(b) = (ab).

8. Appendix

214

If we denote the unit element of G by e and that of Z by 1, then


(e) = 1.
The character o defined by o (a) = 1 for all a G is called the
Principal character.
If 1 and 2 are two characters, we define their product = 1 2 by
(a) = 1 (a)2 (a)
and the inverse of 1 by
1
1
1 (a) = (1 (a)) .

Under this definition, the characters form a multiplicative abelian


group G called the character group of G.
Since is a homomorphism, denote by G the kernel of the homomorphism of G into Z. Then G/G is abelian. Denote by H the
subgroup of G given by
\
H=
G ,
G .

244

Then clearly G/H is abelian.


We call Z an admissible group for G if H consists only of the identity
element. This means first that G is abelian and furthermore that given
any two elements a, b in G there exists a character of G such that, if
a , b,
(a) , (b).
If is a character of G, then can be considered as a character of
G/Go where is trivial on Go , by defining (Go a) = (a), Go a being a
coset of G modulo Go . In particular, the elements of G can be considered as characters of G/H. Moreover Z is now an admissible group for
G/H.
We now prove the
Theorem 3. If G is a finite abelian group and Z is an admissible group
for G, then G is finite and G is isomorphic to G .

2. Characters and duality

215

Proof. In the first place Z is a finite group. For, if a G, a , e, there is


a character such that
(a) , 1.
If G is of order n, then (an ) = ((a))n = 1 so that (a) in an element
of Z of finite order. Since Z is cyclic it follows that Z is finite.

From this it follows that G is finite.
Since ((a))n = 1 for every and every a, there is no loss in generality if we assume that Z is a cyclic group of order n.
In order to prove the theorem let us first assume that G is a cyclic
group of order n. Let a be a generator of G and b a generator of the
245
cyclic group Z of order n. Define the character 1 of G by
1 (a) = b.
Since a generates G any character is determined uniquely by its effect on a. 1 is an element of order n in G . Let be any character of G.
Let
(a) = b

for some integer . Consider the character = 1 .


(a)

= (1 (a)) (a) = b b = 1
which shows that = o is the principal character. Hence G is a cyclic
group of order n and the mapping
a 1
establishes an isomorphism of G on G .
Let now G be finite non-cyclic abelian of order n. G is then a direct
product of cyclic groups G1 , . . . , Gl of orders 1 , . . . , l respectively. Let
ai be a generator of Gi so that a1 , . . . , al is a base of G. Since 1 , . . . , l
divide n, we define l characters 1 , 2 , . . . , l of G by
i (a j ) = 1

j,i

i (ai ) = bi

i = 1, . . . , l,

8. Appendix

216

where bi is an element in Z of order i . These characters are then independent elements of the abelian group G . For, if t11 tll = 0 , then,
for any a G,
t11 (a) . . . tl (a) = 1.
246

Taking for a successively a1 , . . . , al we see that i |ti and so 1 , . . . , l


are independent.
Let be any character of G. Then is determined by its effect on
a1 , . . . , al . Let (ai ) = si . Since ai i = e, ((ai ))i = 1. But ((ai ))i =
si i . Thus si i = 1. Z being cyclic, there exists only one subgroup of
order i . Thus

(ai ) = si = bi i ,
for some integer i (mod i ). Consider the character = 11 . It
us clear that (a
i ) = 1 for all i so that = o or

= 11 .
Thus G is the product of cyclic group generated by 1 , . . . , 1 . By
Corollary to theorem 2, it follows that G and G are isomorphic.
Corollary. If G is a finite group and G its character group, then whatever may be Z,
OrderG OrderG.
Proof. For, if H is the subgroup of G defined earlier, then G/H is abelian
and finite, since G is finite. Also Z is admissible for G/H. Furthermore
every character of G can be considered as a character of G/H, by definition of H. Hence by theorem 3


247

Order G Order G/H Order G.


Let us now go back to the situation where G is finite abelian and Z
ia admissible for G. Then G G . Let us define on G G the function
(a, ) = (a).
For a fixed , the mapping a (a, ) is a character of G and so an
element of G . By definition of product of character, it follows that, for
fixed a, the mapping
a : (a, )

3. Pairing of two groups

217

is a homomorphism of G into Z and hence a character of G , Let G


denote the character group of G . By Corollary above,
OrderG OrderG = OrderG.
Consider now the mapping
: a a
of G into G . This is clearly a homomorphism. If a is identity, then
(a, ) = 1 for all . But since Z is admissible for G, it follows that a = e.
Hence is an isomorphism of G into G . Therefore we have
Theorem 4. The mapping a a is a natural isomorphism of G on G
Note that the isomorphism of G on G is not natural.
Under the conditions of theorem 3, we call G the dual of G. Then

G is the dual of G and theorem 4 shows that the dual of G is naturally


isomorphic to G. Theorem 4 is called the duality theorem for finite
abelian groups.

3 Pairing of two groups


Let G and G be two groups, , , . . ., elements of G and , , . . . ,
elements of G . Let Z be a cyclic group. Suppose there is a function 248
(, ) on G G into Z such that for every , the mapping
: (, )
is a homomorphism of G into Z and for every , the mapping
: (, )
is a homomorphism of G into Z, Thus and are characters of G
and G respectively. We then say that G and G are paired to Z and that
(, ) is a pairing.
For every , let G denote the kernel in G of the homomorphism
. Put
\
H =
G

8. Appendix

218

for all G. Then G /H is abelian. Also, H is the set of in G such


that
(, ) = 1
for all G. Define in a similar way the subgroup H of G.
We are going to prove
Theorem 5. If G/H is finite, then so is G /H and both are then isomorphic to each other.
Proof. From fixed , consider the function
() = (, ).

This is clearly a character of G . By definition of H , it follows that
for each , is a character of G /H .
Consider now the mapping

of G into (G /H ) . This is again a homomorphism of G into (G /H ) .
The kernel of the homomorphism is the set of for which is the
principal character of G /H . By definition of H, it follows that H is the
kernel. Hence
G/H a subgroup of (G /H ) .
(1)
249

In a similar way
(G /H ) a subgroup of (G/H) .

(2)

Let now G/H be finite. Then by Corollary to theorem 3,


ord(G/H) ordG/H.
By (2), this means that
order(G /H ) order G/H.
Reversing the roles of G and G we see that G/H and G /H have
the same order. (1) and (2) together with theorem 3 prove the theorem.
If, in particular, we take G for G , then H = (o ) and so we have

3. Pairing of two groups

219

Corollary. If G is any group, G its character group and if G/H is finite


then
G/H G .

Bibliography
[1] A. A Albert Modern Higher Algebra, Chicago, (1937)
[2] E.Artin Galois Theory, Notre Dame, (1944)
[3] E.Artin Algebraic numbers and Algebraic functions, Princeton
(1951)
[4] E. Artin and 0.Schreier Eine Kennzeichnung der reell
abgeschlossene Korper Hamb, Abhand, Bd 5 (1927) p. 225231
[5] N. Bourbaki Algebra, Chapters 3-6, paris, 1949
[6] C. Chevalley Introduction to the theory of Algebraic functions,
New York, 1951
[7] C. Chevalley Sur la theorie du corps de classes dans les corps finis
et les corps locaux Jour. Faculty of Science, Tokyo, Vol, 2, 1933,P.
365-476
[8] M. Deuring Algebren, Ergebn der Math, Vol.4, 1935
[9] S. Eilenberg and S. Maclane Cohomology theory in abstract
groups, Annals of Mathematics, Vol.48 (1947), P. 51-78
[10] H. Hasse Zahlentheorie, Berlin, 1949
[11] W. Krull Galoissche theorie der unendlicher erweiterungen Math.
Annalen Bd 100 (1928), P.678-698
221

250

222

BIBLIOGRAPHY

[12] N. H McCoy Rings and ideals,Carus Mathematical Monographs,


No.8 (1948)
251

[13] A. Ostrowski Untersuchugen zur arithmetischen Theorie der Korper Math. Zeit, Bd 39 (1935), P.269-404
[14] G. Pickert Einfuhrung in die Hohere Algebra,Gottingen 1951
[15] B.Steinite Algebraische Theorie der Korper Crelles Jour. Bd 137
(1910), P. 167-309
[16] B.L Van-der-Waerden ModerneAlgebra, Leipzing, 1931
[17] A. Weil Foundations of Algebraic gemetry New York, 1946
[18] E.Witt Uber die commutativitat endlicher Schiefkorper Hamb. Abhand, Vol 8, (1931), P.413
[19] E. Witt Der existenzsatz fur Abelsche Funktionekorper Crelles
jour. Bd 173 (1935), P 43-51