Groups, Rings and Fields

Karl-Heinz Fieseler
Uppsala 2010
1
Preface
These notes give an introduction to the basic notions of abstract algebra,
groups, rings (so far as they are necessary for the construction of eld exten-
sions) and Galois theory. Each section is followed by a series of problems,
partly to check understanding (marked with the letter R: Recommended
problem), partly to present further examples or to extend theory.
For useful hints and remarks I am indebted to my colleague Ernst Dieterich.
Uppsala, September 2010 Karl-Heinz Fieseler
2
Contents
1 Introduction 4
2 Groups 12
2.1 Denitions and Examples . . . . . . . . . . . . . . . . . . . . 12
2.2 Homomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.1 Digression: Quaternions . . . . . . . . . . . . . . . . . 32
2.4 Order and Cyclic Groups . . . . . . . . . . . . . . . . . . . . . 37
2.5 Factor Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.5.1 Digression: Free Groups . . . . . . . . . . . . . . . . . 44
2.6 Simple Groups and Composition Series . . . . . . . . . . . . . 52
2.7 Abelian Groups . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.7.1 Digression: Free Abelian Groups . . . . . . . . . . . . 62
2.8 Sylow Subgroups . . . . . . . . . . . . . . . . . . . . . . . . . 63
3 Rings 70
3.1 Denitions and Examples . . . . . . . . . . . . . . . . . . . . 70
3.2 Homomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.3 Ideals and Factor Rings . . . . . . . . . . . . . . . . . . . . . 82
3.3.1 Digression: p-adic number elds . . . . . . . . . . . . . 90
3.4 Irreducibility Criteria . . . . . . . . . . . . . . . . . . . . . . . 95
4 Field Extensions and Galois Theory 100
4.1 Basic Denitions . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.2 Automorphism Groups . . . . . . . . . . . . . . . . . . . . . . 106
4.3 Formal Derivatives and Multiplicities . . . . . . . . . . . . . . 112
4.4 Splitting Fields . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.5 Finite Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4.5.1 Digression 1: Quadratic reciprocity . . . . . . . . . . . 126
4.5.2 Digression 2: Further Simple Groups . . . . . . . . . . 130
4.6 Galois Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.7 The Fundamental Theorem of Algebra . . . . . . . . . . . . . 142
4.8 Cyclotomic Extensions . . . . . . . . . . . . . . . . . . . . . . 143
4.9 Solvability by Radicals . . . . . . . . . . . . . . . . . . . . . . 147
5 Annex: Zorns Lemma 150
3
1 Introduction
Assume we want to solve an equation
0 = f(x) := x
n
+ a
n1
x
n1
+ ... + a
1
x + a
0
,
where the coecients a
n1
, ..., a
1
, a
0
Q of the polynomial f(x) are rational
numbers. The fundamental theorem of algebra
1
tells us that for any n > 0
and arbitrary complex coecients a
n1
, ..., a
0
C there is a complex solu-
tion x = C, and an iterated application of that fact then leads to a
factorization
f(x) = (x
1
) ... (x
n
)
of the polynomial f(x) with (not necessarily pairwise dierent) complex num-
bers
1
, ...,
n
C. In particular f(x) = 0 if and only if x =
j
for some
j 1, ..., n.
But is it possible to describe these solutions explicitly? For n = 2 we
have the solutions x =
1,2
C given by the well known formula

1,2
=
a
1
2

_
a
2
1
4
a
0
,
and for n = 3, 4 there are similar, but more complicated formulae due
to Gerolamo Cardano (1501-1576) involving the four arithmetic operations
+, , , : as well as taking square and cubic roots.
On the other hand for n 5 there was no further progress during the
next 300 years.. So one started to suspect:
Theorem 1.1. For n 5 there is no formula giving the solutions
1
, ...,
n

C of the equation
0 = f(x) := x
n
+ a
n1
x
n1
+ ... + a
1
x + a
0
,
as a function of the coecients a
n1
, ..., a
0
using only the four arithmetic
operations +, , , : as well as taking r-th roots z
r

z (with arbitrary
r N).
1
There are several proofs of that result, all of them use completeness arguments. We
shall present in section 4.7 an algebraic version, where the only result from analysis needed
is the fact that a real polynomial of odd degree has a real zero.
4
Having thus no hope anymore in the general case we are led to the fol-
lowing
Question: Which polynomial equations f(x) = 0 are solvable by radicals,
i.e. when can we obtain the solutions
1
, ...,
n
C of the equation
f(x) = 0
starting with rational numbers and using only the four arithmetic operations
+, , , : as well as taking r-th roots z
r

z (with arbitrary r N)?

So here we do not require that we take the coecients a
0
, ..., a
n1
Q as
starting point and perform the operations independently from the polynomial
f(x).
Example 1.2. For the polynomial f(x) = x
n
+ a
0
it is obviously possible:
The numbers
n

a
0
are the solutions of the equation f(x) = 0. On the other hand the solutions
of the equation
x
5
4x + 2 = 0
do not admit such a representation as we shall see later on.
Strategy: First of all it can happen that some solutions
i
can be obtained
from rational numbers using only the four arithmetic operations +, , , : as
well as taking roots and others can not, e.g. if
f(x) = (x
n
+ a
0
)(x
5
4x + 2).
So we can expect a reasonable answer depending only on f(x) only if we
assume the polynomial f(x) to be irreducible, i.e. there should be no factor-
ization
f(x) = g(x)h(x)
of f(x) as a product of nonconstant polynomials g(x) and h(x) with rational
coecients.
5
Example 1.3. The polynomial
f(x) = x
2
1 = (x 1)(x + 1)
is not irreducible (or rather reducible), while
f(x) = x
2
+ 1
is: Otherwise it would be the product of two linear polynomials each of which
would give rise to a rational zero of f(x).
The zeros
1
, ...,
n
C of an irreducible polynomial are pairwise dier-
ent, and so the set
N(f) := x C; f(x) = 0 =
1
, ...,
n

of all complex zeros of our polynomial f(x) satises

[N(f)[ = n.
Denote
S(N(f)) := : N(f) N(f) bijective .
the set of its permutations.
The Galois group of the polynomial f(x) is a subset
Gal(f) S(N(f))
closed with respect to the composition and inversion of maps, hence it forms
a group in the sense of Def.2.1. And from the properties of Gal(f) as a group
we can read o whether the equation f(x) = 0 is solvable by radicals or not.
In order to describe which permutations of N(f) belong to the Galois
group Gal(f) we rst determine the splitting eld
E = E(f) C
of the polynomial f(x), a subset of the complex plane containing N(f).
This splitting eld E C admits natural symmetries, i.e. distinguished
permutations
: E E
6
called automorphisms. Each automorphism preserves N(f) E, so we
obtain a diagram
: E E

[
N(f)
: N(f) N(f)
,
and an automorphism is uniquely determined by its restriction [
N(f)
. The
elements of the Galois group then are the restrictions of such automorphisms.
Thus, if we denote
Aut(E) S(E)
the set of all automorphisms we can simply write
Gal(f) = Aut(E)[
N(f)
using the suggestive notation
Aut(E)[
N(f)
:= [
N(f)
, Aut(E).
So nally what is a splitting eld? And an automorphism?
First of all we give a restricted denition of a eld, indeed an embedded
version of the abstract notion of a eld:
Denition 1.4. A eld (kropp) is any subset E C of the set of complex
numbers containing the numbers 0, 1 and being closed with respect to the
four arithmetic operations.
Remark 1.5. 1. A subset E C is a eld i 0, 1 E and E is closed
with respect to addition, multiplication and inversion of nonzero num-
bers.
2. Any eld E C contains Q.
3. An arbitrary intersection of elds is again a eld.
Example 1.6. 1. The subsets E = Q, R, C are elds.
2. The subset
E = Q+Qi := a + bi; a, b Q
is a eld.
7
3. The subset
E = Q+Q

d := a + b

d; a, b Q
is a eld, where d Q
>0
is not a square..
Denition 1.7. The splitting eld E = E(f) C of the polynomial
f(x) = (x
1
) ... (x
n
)
is dened as the intersection
E :=

N(f)F eld
F
of all elds F N(f). With other words E = E(f) is the smallest eld
containing N(f).
Remark 1.8. Here is an explicit description of the splitting eld of f(x) =
(x
1
) ... (x
n
), namely
E =
_

1
,...,
n
N
q

1
,...,
n

1
1
...

n
n
;
1
, ...,
n
N : q

1
,...,
n
Q
_
.
Obviously the right hand side contains 0, 1 and is closed with respect to
addition and multiplication. Only the fact that with a nonzero element its
reciprocal is again a sum of the given type is nontrivial: This follows from
the fact that each
i
is an algebraic number, i.e. a zero of some polynomial
g
i
(x) with rational coecients (namely g
i
(x) = f(x)).
Example 1.9. 1. The splitting eld of f(x) = x
2
+ 1 is E = Q+Qi.
2. The splitting eld of f(x) = x
2
d is E = Q+Q

d.
The symmetries of the splitting eld E are called automorphisms:
Denition 1.10. An automorphism of the eld E C is a bijective map
: E E
compatible with the four arithmetic operations, i.e.
(x y) = (x) (y), (xy) = (x)(y),
_
x
y
_
=
(x)
(y)
.
8
We denote
Aut(E) := : E E automorphism
the set of all automorphisms of the eld E.
Remark 1.11. For an automorphism Aut(E) we have (0) = 0 and
(1) = 1, since for example (a) = (a+0) = (a) +(0). As a consequence
(n) = n for n N and nally
[
Q
= id
Q
.
In particular
(f()) = f(())
for any polynomial f(x) with rational coecients and E.
Example 1.12. 1. For E = Q+Qi we have
Aut(E) = id
E
, ,
where (z) = z is complex conjugation. Indeed, if Aut(E), we
have
(a + bi) = (a) + (b)(i) = a + b(i),
while (i)
2
= (i
2
) = (1) = 1 implies (i) = i. Hence = id
E
or = .
2. For E = E(f) the equality (f()) = f(()) together with (0) =
0 implies that (N(f)) = N(f), and the explicit description of the
splitting eld E(f) in Rem.1.8 gives that can be reconstructed from
its restriction [
N(f)
. From the point of view of abstract group theory
we need thus not distinguish between the Galois group
Gal(f) = Aut(E)[
N(f)
S(N(f))
of the polynomial f(x) and the automorphism group
Aut(E) S(E)
of its splitting eld E = E(f).
9
In the general approach one considers not only polynomials f(x) with
rational coecients. Instead they are taken from a xed base eld K, where
now in contrast to the restricted denition Def.1.4 we mean an abstract,
not an embedded eld, i.e. a set K with two distinguished elements 0, 1
and four arithmetic operations satisfying the usual rules. Then given a
polynomial one has rst to construct the splitting eld E = E(f) K, since
in general there is a priori no eld at hand taking the r ole of the complex
numbers in the case K = Q. For that construction one has to study the
basics of commutative ring theory. Finally the automorphism group Aut(E)
is replaced with
Aut
K
(E) := : E E automorphism, [
K
= id
K
.
Here is a short survey of the material presented in these notes:
1. Chapter I: Groups. Here we discuss the basic notions of group the-
ory: Groups play an important r ole nearly in every part of mathematics
and can be used to study the symmetries of a mathematical object.
2. Chapter II: Rings. Commutative rings R are sets with three arith-
metic operations: Addition, subtraction and multiplication , as for
example the set Z of all integers, while division in general is not always
possible. We need rings, that are not elds, mainly in order to con-
struct extensions of a given eld K, but they play also an important
r ole in algebraic number theory (Number theory deals with the ring
Z Q of integers, but for a deeper understanding of Z one has to
extend the notion of an integer and to study rings of algebraic integers
R E, where E C is the splitting eld of some polynomial with
rational coecients) and in algebraic geometry, where one investigates
the set of solutions of polynomial equations in several variables.
3. Chapter III: Field Extensions and Galois Theory. The main
result relates subgroups of Aut
K
(E) for the splitting eld E K of
some polynomial with coecients in K to intermediate elds of the
extension E K. As an easy application we prove the fundamental
theorem of algebra and discuss cyclotomic elds, i.e. the splitting elds
over Qof the polynomials f(x) = x
n
1, before we eventually attack our
10
original problem, whether the zeros of an irreducible equation f(x) =
0 can be obtained from elements in the base eld K with our ve
operations.
4. Appendix: Zorns lemma. Here we prove Zorns lemma. That lemma
is usually needed if one wants to show the existence of certain objects in
case one deals with innite sets, e.g. the existence of bases of innite
dimensional vector spaces. An other example is this: There are no
explicit automorphisms Aut(C) except the identity and complex
conjugation, but with Zorns lemma we see that any automorphism of
a splitting eld E C of a polynomial f(x) with rational coecients
can be extended to an automorphism of C.
11
2 Groups
2.1 Denitions and Examples
Denition 2.1. A group is a pair (G, ) with a non-empty set G and a
binary operation, i.e., a map
: GG G, (a, b) ab := (a, b),
called the group multiplication or group law, satisfying the following con-
ditions
G
1
: Group multiplication is associative, i.e. for all a, b, c G we have
(ab)c = a(bc) .
G
2
: Existence of a neutral element: There is an element e G such that
ea = a = ae
for all elements a G.
G
3
: Existence of inverse elements: For all a G there is an element
a
1
G, such that
aa
1
= e = a
1
a .
Notation: Often one writes G instead of (G, ). And the number [G[ is
called the order of the group G. Here we denote [M[ N the
number of elements in the set M. We dene powers a
n
for a G and n N
inductively by
a
0
:= e, a
n+1
:= a
n
a,
and the associative law yields
a
n+m
= a
n
a
m
.
Remark 2.2. 1. There is only one neutral element in a group G: If e G
is a further neutral element, we obtain e = ee = e.
12
2. There is only one inverse element a
1
for a given element a G: If a
is a further inverse element, we have
a = ae = a(aa
1
) = ( aa)a
1
= ea
1
= a
1
.
In particular we can also dene negative powers
a
n
:= (a
1
)
n
.
3. In many books for a group only the existence of a left neutral element
e, i.e. such that ea = a holds for all a G, and left inverse elements
a
1
(with e = a
1
a) is required. So the group axioms are a priori
weaker, but it turns out that they are equivalent to ours, though group
multiplication need not be commutative:
Denition 2.3. A group is called commutative or abelian (Niels Henrik
Abel, 1802-1829) i ab = ba holds for all a, b G.
Notation: If a group G is commutative, one often writes the group law in
additive notation:
a + b := (a, b) ,
the symbol 0 denotes the neutral element, and a the inverse element of
a G, while for n 0 powers look as follows
na := a + ... + a
. .
n times
, (n)a := n(a) = (a) + ... + (a)
. .
n times
.
In that case we say that G is an additively written group.
Example 2.4. 1. A set G := e with only one element e and the obvious
group multiplication constitutes a group, the trivial group.
2. With the ordinary addition of numbers as group law the set Z of all
integers forms a (commutative) group, and so do the rational, real and
complex numbers:
Q := all rational numbers , R := all real numbers and
C := all complex numbers.
13
3. A real or complex vector space V endowed with the addition of vectors
is a (commutative) group.
4. Let K = Q, R or C. With the ordinary multiplication of numbers
the set K

:= K 0 becomes a (commutative) group. (We have to

exclude 0, since it does not have an inverse element.)
5. For K = Q, R, C we denote K
n,n
the set of all square matrices of size
n with entries in K. Then the set
GL
n
(K) := A K
n,n
, det A ,= 0
of all invertible matrices in K
n,n
with matrix multiplication as group
law is a group, which for n 2 is not commutative. It is called the
(n-dimensional) general linear group over K.
6. For a set M let
S(M) := f : M M bijective map
be the set of all bijective maps from M to itself (permutations of
M). It constitutes a group together with the composition of maps as
group law , i.e., fg = (f, g) := f g. (The neutral element is the
identity id
M
.)
7. Let V be a nite dimensional vector space over K with K = Q, R, C.
Then we dene the general linear group GL(V ) of V as
GL(V ) := f S(V ); f K-linear,
endowed with the group multiplication induced by S(V ).
8. For M
n
:= 1, ..., n, the group
S
n
:= S(M
n
)
is called the symmetric group on n letters. For m n we can
understand S
n
as subset of S
m
by extending f S
n
to

f S
m
with

f(k) = k for k > n.

There are two dierent ways to denote a permutation f S
n
, either as
a 2 n-matrix:
_
1 2 n
f(1) f(2) f(n)
_
14
or as product of cycles: A permutation f S
n
is called a cycle of
length r 2, if there are pairwise distinct numbers a
1
, . . . , a
r
M
n
,
such that
f(k) =
_
_
_
a
i+1
, if k = a
i
, i < r
a
1
, if k = a
r
k , otherwise
.
In that case we write also f = (a
1
, . . . , a
r
). Obviously (a
1
, ..., a
r
) =
(b
1
, ..., b
s
), i s = r and there is a number N, 0 < r, with
b
j
=
_
a
j+
, , if j + r
a
j+r
, , if j + > r
.
Two cycles f = (a
1
, ..., a
r
), g = (b
1
, ..., b
s
) are called disjoint i a
i
,= b
j
for all indices i = 1, ..., r, j = 1, ..., s. In that case the cycles commute:
fg = gf, but otherwise, that need not be true, e.g.:
(1, 2, 3)(1, 2) = (1, 3) ,= (2, 3) = (1, 2)(1, 2, 3) .
An arbitrary permutation can be factorized as product of pairwise dis-
joint cycles, the factors being unique up to reordering (the identity
permutation being the empty product: In a group a product with-
out factors (a contradictio in se?) is dened to be the neutral element.)
For example the permutation
_
1 2 3 4 5 6 7 8
2 7 5 4 6 3 8 1
_
becomes (1, 2, 7, 8)(3, 5, 6), while (1, 2, 3, 7)(4, 8) has the matrix
_
1 2 3 4 5 6 7 8
2 3 7 8 5 6 1 4
_
.
We remark that S
n
is abelian i n 2, cf. the above counterexample
for n 3.
9. If G
1
, ..., G
r
are groups, then their cartesian product G
1
... G
r
endowed with the componentwise multiplication
(g
1
, ..., g
r
)(h
1
, ..., h
r
) := (g
1
h
1
, ..., g
r
h
r
)
is again a group, the direct product of the groups G
1
, ..., G
r
. The
group S
2
S
2
is also called four-group or Kleins four-group
(Felix Klein, 1849-1925).
15
Remark 2.5. If the group G = e, a, b, c, ... is nite, the group multiplica-
tion can be given in a multiplication table
e a b c ...
e e a b c ...
a a a
2
ab ac ...
b b ba b
2
bc ...
c c ca cb c
2
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
,
where an element in G occurs in every row and column exactly once, since
the equation
ax = b resp. xa = b
has in G a unique solution, namely x = a
1
b resp. x = ba
1
.
Two such tables are equivalent from the point of view of algebra, if one of
them is obtained from the other by exchange of letters. The corresponding
groups then are called isomorphic, cf. Def. 2.7.
Problems 2.6. 1. R: For a set M denote T(M) its power set. Consider the following
binary operations T(M) T(M) T(M) on it:
(A, B) A B, A B, AB := (A B) (B A) .
Which one does provide on T(M) a group law?
2. Which of the following subsets of R
n,n
endowed with either matrix addition or
matrix multiplication becomes a group?
(a) R: The diagonal matrices.
(b) R: The diagonal matrices, where all entries in the diagonal are non-zero.
(c) R: The symmetric matrices.
(d) R: The invertible symmetric matrices.
(e) The diagonalizable matrices.
(f) The invertible diagonalizable matrices.
3. R: For K = Q, R, C we denote A
n
(K) S(K
n
) the subset of all ane linear
transformations:
A
n
(K) := f S(K
n
); A GL
n
(K), b K
n
: f(x) = Ax +b, x K
n
.
Show: A
n
(K) is a group together with the composition of maps as group multi-
plication. Is A(K) := A
1
(K) commutative?
16
4. R: Let G := e, a, b, c be a set with four elements. Show that there are exactly two
dierent multiplication tables for a group law on G (up to exchange of letters).
5. R: Let f := (1, 2, 3), g := (1, 2) S
3
. Show: S
3
= id, f, f
2
, g, gf, gf
2
. Using that
notation write a multiplication table for S
3
!
6. R: Show: [S
n
[ = n!.
7. R: A cycle (i, j) S
n
of length 2 is called a transposition. Show: Every permu-
tation f S
n
can be written as a product of transpositions. Hint: Induction on
n.
8. R: Show: A group, where a
2
= e for all a G, is abelian.
9. A group is called nitely generated i there are elements a
1
, ..., a
r
G such that
every g G can be written g = a
k
1
i
1
... a
k
s
i
s
with integers k
1
, ..., k
s
, 1 i
1
, .., i
s
r.
In that case the elements a
1
, ..., a
r
are said to genererate the group G, or to be
generators of G (but as such they are of course not uniquely determined - there
are many dierent systems of generators!) If G is abelian, the dening condition
for generators a
1
, ..., a
r
can be simplied: It is enough to require that every g G
can be written g = a
k
1
1
... a
k
r
r
resp., with additive notation, g = k
1
a
1
+... +k
r
a
r
with integers k
1
, ..., k
r
Z. Show: The group Z is generated by the element 1 as
well as by 2, 3. The group (Q, +) is not nitely generated!
10. Show: The symmetric group S
n
is generated by the transpositions (1, 2), ..., (1, n)
resp. by (1, 2), (2, 3), ..., (n 1, n) resp. by (1, 2), (1, 2, 3, ..., n).
2.2 Homomorphisms
Denition 2.7. A map
: G H
between two groups G and H is called a (group) homomorphism, i
(ab) = (a)(b) a, b G.
A bijective homomorphism is called a (group) isomorphism, and an iso-
morphism : G G is called a (group) automorphism.
Two groups G, H are called isomorphic, i there is a (group) isomorphism
: G H. In that case one writes G

= H.
17
Remark 2.8. For a group homomorphism we have always
(e
G
) = e
H
, (a
n
) = (a)
n
with the neutral elements e
G
and e
H
of G resp. H, and arbitrary elements
a G, n Z.
Example 2.9. 1. For all groups G, H the constant map : G H
with (a) = e := e
H
, a G, is a homomorphism.
2. For every group the identity map := id
G
: G G is an automor-
phism.
3. Let G be a group and a G. Then the exponential map
a
: Z
G, n a
n
, is a homomorphism, since
a
(n + m) = a
n+m
= a
n
a
m
=

a
(n)
a
(m).
4. Let G be a group and n Z. We regard the n-th power map p
n
:
G G, a a
n
, and nd: p
0
and p
1
are always homomorphisms, cf.
the previous points, while p
1
is a so called anti-homomorphism, i.e.,
we have
(ab)
1
= b
1
a
1
.
Only if G is abelian, all the power maps p
n
, n Z, are homomorphisms.
5. The complex exponential map : C C

, z e
2iz
, is a group
homomorphism from an additively written group into a multiplicatively
written one, since (z + w) = (z)(w).
6. Let K = Q, R, C and V be a K-vector space, dimV = n < . If we
choose a basis e
1
, ..., e
n
of V , then
GL(V )

=
GL
n
(K), f A
f
,
where A
f
= (
ij
) K
n,n
denotes the matrix of f with respect to the
basis e
1
, ..., e
n
, i.e. f(e
j
) =

n
i=1

ij
e
i
, is a group isomorphism.
7. The determinant det : GL
n
(K) K

, A det(A), is a group homo-

morphism. (K = Q, R or C).
18
8. The map sign : S
n
Q

, where
sign(f) :=

1i<jn
f(i) f(j)
i j
,
is a group homomorphism, and sign(f) = 1 for all permutations f
S
n
. The latter follows immediately from the fact that any factor i j
equals a factor f() f(k) up to sign. In order to see that sign indeed
is a group homomorphism, we consider for any set
A = i, j M
n
consisting of of two dierent natural numbers i, j between 1 and n the
rational number

A
(f) :=
f(i) f(j)
i j
.
It is well dened, since the RHS remains unchanged after exchange of
i and j. Denote T
2
(n) the set of all subsets A M
n
of order [A[ = 2.
Since

A
(g f) =
f(A)
(g)
A
(f),
we obtain
sign(g f) =

AP
2
(n)

A
(g f) =

AP
2
(n)

f(A)
(g)

AP
2
(n)

A
(f)
_
_

AP
2
(n)

A
(g)
_
_
sign(f) = sign(g) sign(f).
9. For each element g G in a group G the conjugation with g, i.e., the
map

g
: G G, a gag
1
,
is a group homomorphism. We remark, that
g
even is an automor-
phism of the group G with inverse
g
1. Such an automorphism is also
called an inner automorphism, since it can be described using the
group multiplication of the group G. Indeed any automorphism is the
restriction to G of an inner automorphism of a suitable bigger group

G G. Note that
g
= id
G
for all g G i G is abelian.
19
10. The map : G S(G), g
g
, associating to g G the conjugation
with g, is a group homomorphism for every group G. Remember that
S(G) is the group of all permutations of G; so we have to show that

g

h
=
gh
for all g, h G. Do that!
The homomorphisms : G S(M) from a group G to the permutation
group S(M) of a set M correspond to G-actions on M:
Denition 2.10. Let G be a group and M a set. A G-action on M is a
map
GM M, (g, x) gx ,
satisfying the following conditions:
1. ex = x, x M ,
2. g(hx) = (gh)x, g, h G, x M.
A group action is called eective if gx = x for all x M implies g = e.
Remark 2.11. The G-actions on a set M correspond bijectively to group
homomorphisms : G S(M). Namely:
1. Given a G-action on M and an element g G, the map

g
: M M, x gx,
is bijective with inverse
g
1. Hence we obtain a homomorphism
: G S(M), g
g
.
2. On the other hand to a homomorphism : G S(M), g
g
, we
associate the G-action (g, x) gx :=
g
(x).
In particular a group action is eective i the homomorphism : G S(M)
is injective.
Example 2.12. 1. The symmetric group S
n
acts naturally on M
n
:=
1, ..., n by (f, k) f(k).
20
2. The general linear group GL
n
(K) acts on K
n
by (A, x) Ax, where
Ax denotes the product of the matrix A with the column vector x K
n
.
3. The general linear group GL
n
(K) acts on K
n,n
by (A, X) AXA
1
.
4. The general linear group GL
n
(K) acts on the subspace Sym
n
(K)
K
n,n
of all symmetric matrices by (A, X) AXA
T
, where A
T
denotes
the transposed matrix.
5. For M = G there are three dierent natural actions of G on itself (The
products in the below formulae denote the group multiplication in G!):
(a) Left translation: GG (g, x) gx G.
(b) Right translation with the inverse element: G G (g, x)
xg
1
G.
(c) Conjugation, i.e., the rst two (commuting!) actions simultane-
ously: GG (g, x)
g
(x) = gxg
1
G.
Denition 2.13. Let the group G act on the set M. The orbit Gx M of
an element x M is dened as
Gx := gx; g G.
Example 2.14. With respect to the natural action of GL
n
(K) on K
n
there
are exactly two dierent orbits in K
n
, namely GL
n
(K)0 = 0 and GL
n
(K)x =
K
n
0, where x K
n
is an arbitrary vector ,= 0.
In general we have
Proposition 2.15. Two orbits in a set M with G-action are either equal or
disjoint. In particular M is the disjoint union of all orbits.
21
Proof. If y Gx, we have Gy = Gx: Let y = hx, h G. Gy Gx follows
from gy = g(hx) = (gh)x. But on the other hand also x = h
1
y Gy, so
Gx Gy resp. Gx = Gy.
Assume now that Gx Gy ,= , say z Gx Gy: According to what we
already know, that implies Gx = Gz = Gy.
Remark 2.16. If there is only one orbit (which then coincides with M), one
says that G acts transitively on M. For example, that is the case, if G acts
on itself by left translation or by right translation with the inverse element.
But for conjugation the situation is dierent: The orbits

G
(x) = gxg
1
; g G G
then are called conjugacy classes and two elements in the same orbit are
called conjugate. Since
G
(e) = e, the action of G on itself by conjugation
is never transitive for a nontrivial group G. Note that a cunjugacy class
G
(x)
is trivial, i.e.
G
(x) = x, i x commutes with all elements g G.
Denition 2.17. Let G be a group. The center Z(G) G is the subset
Z(G) := x G;
G
(x) = x .
consisting of the elements x G with trivial conjugacy class. Equivalently
Z(G) = x G; xg = gx g G .
In particular, a group G is abelian i Z(G) = G.
Problems 2.18. 1. R: Determine all automorphisms of Kleins four group S
2
S
2
!
What about the other group (cf. Problem 2.6.4) of order 4?
2. R: Compute sign(f) for a cycle f S
n
of given length! Hint: Induction on the
length starting with 2-cycles (transpositions) resp. = (1, 2).
3. R: Assume that f S
n
admits two dierent factorizations as product of transposi-
tions (2-cycles). Show that the numbers of factors have the same parity!
4. R: Show: Two permutations f, g S
n
are conjugate i in their respective factor-
ization as product of disjoint cycles, for any given length r 2, there are as many
cycles of length r in the factorization of g as in the factorization of f. How many
conjugacy classes are there in S
5
?
22
5. Show: An automorphism : S
n
S
n
mapping transpositions to transpositions
is an inner automorphism, i.e. has the form
h
with some h S
n
. Hint: Consider
the transpositions
i
:= (i, i + 1) for i < n.
6. Show: An automorphism : S
n
S
n
maps transpositions to permutations which
are the product of mutually disjoint 2-cycles.
7. Show that for n ,= 6 every automorphism of S
n
is an inner automorphism. Hint:
Compute the number [
S
n
(f)[ of elements in the conjugacy class of a permutation
f S
n
, which is the product of mutually disjoint 2-cycles.
8. R: Let R

act on the plane R

2
by t (x, y) := (t
a
x, t
b
y) with integers a, b Z. Sketch
the orbits!
9. Classify the orbits of the GL
2
(C)-action of Example 2.12.3.
10. Classify the orbits of the GL
2
(R)-action of Example 2.12.4.
11. R: Show Z(S
n
) = id
M
n
for n 3.
12. R: Show Z(GL
n
(K)) = K

E with the unit matrix E of size n.

13. The group GL
n+1
(K) acts in a natural way on the set
P
n
(K) := L K
n+1
a one dimensional subspace
of all lines in K
n+1
through the origin via:
GL
n+1
(K) P
n
(K) P
n
(K), (A, L) A(L).
The set P
n
(K) is also called the n-dimensional projective space over K. Determine
the kernel of the corresponding group homomorphism : GL
n+1
(K) S(P
n
(K))!
14. A continuation of the previous problem: Using the bijection

K := K P
1
(K), x K(x, 1), K(1, 0)
and writing A =
_
a b
c d
_
compute the corresponding action
GL
2
(K)

K

K, (A, x) Ax = ? .
Show that given two triples (L
1
, L
2
, L
3
) and (

L
1
,

L
2
,

L
3
) of pairwise dierent lines
there is a matrix A GL
2
(K), unique up to a nonzero scalar factor, satisfying

L
i
= A(L
i
) for i = 1, 2, 3.
For K = C the set

C is nothing but the Riemann sphere from complex analysis,
and the transformations z Az are called Mobius transformations (August
Ferdinand Mobius, 1790-1868).
23
2.3 Subgroups
Denition 2.19. Let G be a group. A non-empty subset H G is called
a subgroup i H is closed with respect to both, group multiplication and
inversion
a, b H =ab H , a H =a
1
H .
In order to emphasize that a subset H G actually is a subgroup, we also
write
H G.
Remark 2.20. 1. Let H G be a subgroup, a H. Since a
1
H,
we get e = aa
1
H. In particular a subgroup is itself a group when
endowed with the restriction of the group multiplication of G.
2. The intersection H
1
H
2
of two subgroups H
1
, H
2
G is again a
subgroup: H
1
H
2
G. But the union H
1
H
2
in general is not.
Example 2.21. 1. For any group G the subsets H = e G as well as
H = G provide subgroups.
2. Let : G F be a group homomorphism between the groups G and
F. If G
0
G, F
0
F are subgroups, the image (G
0
) F of G
0
as
well as the inverse image
1
(F
0
) G of F
0
are subgroups of F resp.
of G. If G
0
= G, we obtain the image
(G) = (g); g G F
of , and if F
0
= e the corresponding subgroup is called the kernel
of the homomorphism :
Denition 2.22. Let : G F be a group homomorphism between groups
G and F. Its kernel is the subgroup
ker() := g G; (g) = e G ,
where e := e
F
denotes the neutral element in the group F.
24
Proposition 2.23. A group homomorphism : G F is injective i
ker() = e.
Proof. In any case we have e ker(); so if is injective, necessarily ker() =
e. Assume now ker() = e and (a) = (b). Then we have (ab
1
) =
(a)(b)
1
= e F, i.e., ab
1
ker() = e resp. ab
1
= e resp. a = b.
Example 2.24. 1. The center Z(G) G of a group G, cf. Def. 2.17,
satises
Z(G) = ker( : G S(G))
with the homomorphism : G S(G), g
g
, in particular it is a
subgroup of G.
2. The unit circle
S
1
:= z C; [z[ = 1
is a subgroup of the multiplicative group C

of nonzero complex num-

bers, in fact the kernel of the homomorphism C

, z [z[.
3. The kernel of the determinant homomorphism det : GL
n
(K) K

,
the (sub)group SL
n
(K) := ker(det) GL
n
(K) is called the special
linear group. Remember that the neutral element of K

is 1 K

=
K 0; so
SL
n
(K) = A GL
n
(K); det(A) = 1 = A K
n,n
; det(A) = 1.
4. Let : R
n
R
n
R be the the standard inner product, i.e., (x, y) :=
x
T
y. Then
O(n) := A GL
n
(R); (Ax, Ay) = (x, y) x, y R
n

= A GL
n
(R); A
T
A = E
(E R
n,n
denotes the unit matrix) is a subgroup of GL
n
(R), the
orthogonal group.
5. The intersection SO(n) := SL
n
(R) O(n) is also a subgroup, the
special orthogonal group. For n = 2, 3 the matrices A SO(n)
correspond to rotations around the origin resp. some axis through
25
the origin. In particular, if we identify R
2
with the complex plane C
and denote R

the counterclockwise rotation with angle R, i.e.

R

(z) := e
i
z, we obtain
SO(2) = R

; R

= S
1
.
The orthogonal group O(2) itself is the union
O(2) = SO(2) SO(2) S,
of its subgroup SO(2) O(2) and the set
SO(2) S = R

S; R
of all reections. Here S denotes the reection at the real axis, i.e.
complex conjugation S(z) := z.
With C instead of R there is an analogous construction:
6. Let : C
n
C
n
C be the standard inner product, i.e. (x, y) :=
x
T
y. Then
U(n) := A GL
n
(C); (Ax, Ay) = (x, y) x, y C
n

= A GL
n
(C); A
T
A = E
is a subgroup of GL
n
(C), the unitary group. And
SU(n) := SL
n
(C) U(n)
is called the special unitary group. We remark that
SU(2) = A C
2,2
; A =
_
z w
w z
_
; z, w C, zz + ww = 1;
so as a topological space, it is nothing but the three dimensional unit
sphere S
3
C
4

= C
2,2
.
7. A
n
:= ker(sign) S
n
is called the alternating group on n letters.
The permutations in A
n
are called even, the other ones odd. As a
consequence of problems 2.6.7 and 2.18.3 we see that if an even resp.
odd permutation is a product of transpositions, then the number of
factors is even resp. odd; this explains the etymology.
26
8. For n 1 we consider the power map p
n
: C

, z z
n
. Its
kernel C
n
:= ker(p
n
) is called the group of all n-th roots of unity,
i.e.
C
n
= z C; z
n
= 1 = 1, , ...,
n1

with := e
2i
n
.
9. Given a subset P K
n
with K = R or K = C we dene the sym-
metry group of P as the subgroup Sym(P) GL
n
(K) of the general
linear group GL
n
(K) consisting of all matrices transforming P into
itself:
Sym(P) := A GL
n
(K); A(P) = P GL
n
(K).
10. The dihedral group D
n
, n 3 : In the complex plane C we consider
the regular n-gon P
n
with the n-th roots of unity 1, , ...,
n1
as vertices
(where := e
2i
n
). We identify C with R
2
as usual and dene the
dihedral group D
n
as the (real) symmetry group of P
n
, i.e.
D
n
:= Sym(P
n
) GL
n
(R).
Then, with the rotation R := R
2/n
and the reection S on the x-axis,
we have
(1) D
n
= E, R, ..., R
n1
, S, RS, ..., R
n1
S,
where E denotes the identity map (or the unit matrix). We note the
relations
R
n
= E = S
2
, SRS
1
= R
1
.
Of course SRS
1
= SRS because of S
2
= E, but for systematic reasons
we prefer the expression SRS
1
, since in order to classify groups up
to isomorphy, it is important to understand how elements in the group
act on others by conjugation.
Proof of Equality 1. Obviously the given 2n linear maps transform P
n
into itself, and we have to show the opposite inclusion. A linear map
A D
n
maps the vertices of P
n
onto themselves, i.e. A(C
n
) = C
n
. In
particular A(1) =

for some , 0 n1. Then B := R

A D
n
satises B(1) = 1; and since edges are mapped to edges we obtain
27
B() = or B() =
n1
= . But 1, is a basis of the real vector
space C, and thus B = E or B = S. For the matrix A that means
A = R

or A = R

S.
11. Let I := [1, 1] R. The complex symmetry group of the 4-cube
I
4
R
4

= C
2
is the group
Sym
C
(I
4
) := A GL
2
(C); A(I
4
) = I
4

=
_
A GL
2
(C); A =
_
0
0
_
or A =
_
0
0
_
with , C
4
_
.
In particular [Sym
C
(I
4
)[ = 32. As motivation we can say that a matrix
A Sym
C
(I
4
) maps midpoints of the facets of the cube again to such
midpoints (they are the arithmetic means of the vertices). But these
are the points (, 0) or (0, ) with C
4
.
12. The intersection
Q := Sym
C
(I
4
) SL
2
(C)
of the complex symmetry group Sym
C
(I
4
) of the 4-cube with SL
2
(C)
is called the quaternion group, it has 8 elements:
Q = E, I, J, K
with the unit matrix E and
I :=
_
i 0
0 i
_
, J :=
_
0 1
1 0
_
, K := IJ =
_
0 i
i 0
_
.
We note that E commutes with all elements in Q and the relations
IJ = K = JI, JK = I = KJ, KI = J = IK.
13. For each group G the set
Aut(G) := S(G); group automorphism
is a subgroup of the group S(G) of all permutations of the set G, it is
called the automorphism group of G.
The subgroups of the additive group Z of all integers are determined in the
next proposition:
28
Proposition 2.25. The subgroups H Z are the sets Zn := kn; k Z,
where n N.
Proof. The subsets H := Zn Z obviously are subgroups. Consider now
an arbitrary subgroup H Z. For H = 0 we have H = Z0, and for non-
trivial H there is a least positive integer n := min(HN
1
) in H, the subset
H being symmetric with respect to the origin: a H = a H. Let us
show H = Zn: On the one hand n H =Zn H, since H is a subgroup.
Now take any a H and divide a by n with remainder: a = qn + r with
q, r Z, 0 r < n. But r = a qn H, such that because of the choice of
n H only r = 0 is possible. Hence a = qn Zn.
Let H G be a subgroup of the group G. We restrict the left translation
resp. right translation with the inverse to the subgroup H G and obtain
H-actions
H G G.
These actions are not any longer transitive; their orbits
aH := ah; h H = ah
1
; h H resp. Ha := ha; h H ,
are called left resp. right cosets:
Denition 2.26. Let H G be a subgroup of the group G. A left coset
resp. right coset or residue class mod(ulo) H is a set aH G resp.
Ha G, where a G. A representantive of the coset aH resp. Ha is any
element b aH resp. b Ha, i.e. any element b = ah with some h H
resp. b = ha with an h H. Then aH = bH resp. Ha = Hb.
If G is abelian, the left and right coset given by a representative a G agree,
and in additive notation it is written a + H or H + a.
The set of all left resp. right cosets is denoted G/H resp. HG, i.e.
G/H := aH; a G ( resp. HG := Ha; a G) .
Example 2.27. 1. Let G := S
n
and H := f S
n
; f(n) = n

= S
n1
.
Then there are n left resp. right cosets mod H, namely
(i, n)H = f S
n
; f(n) = i
29
resp.
H(i, n) = f S
n
; f(i) = n,
where 1 i n.
2. If G = Z and H = Zn, n 1, there are again n cosets, namely the sets
r +Zn, 0 r < n.
So a coset consists of all integers which give after division with n the
same remainder r, 0 r < n.
3. A geometric interpretation: Let G = R
2
and H R
2
a line through
the origin. The cosets mod H are then nothing but the lines parallel
to H.
4. The cosets of S
1
C

are the circles centered at the origin.

Proposition 2.28. Let H G be a subgroup of the group G.
1. Two left (resp. right) cosets mod H are disjoint or coincide; we have
aH = bH (resp. Ha = Hb), i a
1
b H (resp. ab
1
H).
2. If H is nite, a coset mod H contains as many elements as H.
3. Let G be nite. Then
[G[ = [G/H[ [H[ = [HG[ [H[;
in particular the order [H[ of the subgroup H divides the order [G[ of
the group G.
4. If G/H is nite, so is HG, more precisely [G/H[ = [HG[.
Denition 2.29. Let H G be a subgroup of the group G. If G/H is nite,
the number
(G : H) := [G/H[ = [HG[
is called the index of the subgroup H in G.
30
Proof. i) The rst part is a consequence of Prop. 2.15, the second is left to
the reader as an exercise.
ii) The left translation
a
: G G, x ax, is bijective, and
a
(H) = aH,
whence [aH[ = [H[ for all a G. (An analogous argument works for right
cosets).
iii) follows immediately from i) and ii).
iv) The inversion p
1
: G G, a a
1
is bijective and satises p
1
(aH) =
Ha
1
, hence induces a bijection G/H HG.
Given an action of a group G on a set M we associate to each element
x M a subgroup G
x
G as follows:
Denition 2.30. Let G be a group acting on the set M. For an element
x M we dene its isotropy group or stabilizer G
x
by
G
x
:= g G; gx = x .
The left coset space G/G
x
can be identied with the orbit Gx M:
Proposition 2.31 (Class formula). Assume the group G acts on the set M.
Then:
1. For any x M the map G/G
x
Gx, gG
x
gx, is bijective; in
particular we have [Gx[ = (G : G
x
).
2. If M is nite and Gx
1
, .., Gx
r
are the (pairwise dierent) G-orbits, the
class formula says
[M[ = (G : G
x
1
) + ... + (G : G
x
r
) .
Proof. The given map is obviously well dened and surjective, and it is injec-
tive as well, since gx = hx implies g
1
h G
x
and thus hG
x
= g(g
1
h)G
x
=
gG
x
. (Its inverse is given by y
1
(y) with the orbit map : G
Gx, g gx.) Finally M is the disjoint union of the orbits Gx
1
, ..., Gx
r
.
If G acts on itself by conjugation, then the center Z(G) G consists
exactly of the elements with one point conjugacy class. In that case the class
formula reads as follows:
31
Corollary 2.32. Let G be a nite group group and x
1
, ..., x
r
G a system of
representatives for the non-trivial conjugacy classes in G, i.e. every conju-
gacy class with more than one element is of the form
G
(x
i
) and the
G
(x
i
)
are pairwise distinct. Then we have
[G[ = [Z(G)[ + (G : G
x
1
) + ... + (G : G
x
r
)
with the proper subgroups G
x
i
= a G; ax
i
a
1
= x
i
G. In particular if
[G[ = p
r
> 1 with a prime number p, its center is nontrivial: Z(G) ,= e.
Proof. The order [G[ and the indices (G : G
x
i
), i = 1, ..., r, are divisible with
p, hence so is the order [Z(G)[ of the center, in particular [Z(G)[ > 1.
2.3.1 Digression: Quaternions
Motivated by the quaternion group, cf. Example 2.24.12, we conclude this
section with a digression about quaternions in general:
The elements in the four-dimensional real vector space
H :=
__
z w
w z
_
; z, w C
_
= R
0
SU(2) C
2,2
are called quaternions; they have been found by William Rowan Hamil-
ton (1805-1865) - of course not in this form, but instead using a basis like
E, I, J, K below. For the second equality we refer to Example 2.24.6; as a
consequence we see that H is even closed with respect to matrix multiplica-
tion and that all non-zero quaternions A H are invertible, but note that
H C
2,2
is not a complex vector subspace! In fact
H = R E +R I +R J +R K

= R
4
with the unit matrix E C
2,2
and the matrices I, J, K of Example 2.24.12.
There is more structure on the vector space H: We have a conjugation:
H H, A A

:= A
T
satisfying
(A + B)

= A

+ B

, (AB)

= B

, (A

= A
32
and the inner product
: HH R, (A, B) (A, B) :=
1
2
trace(AB

).
It satises
(AB, C) = (A, CB

),
and the corresponding norm
[A[ =
_
(A, A) =
_
det(A)
is multiplicative, i.e.
[AB[ = [A[ [B[ as well as [A

[ = [A[
and has unit sphere
S
3
= A H; [A[ = 1 = SU(2).
We can write H as an orthogonal sum
H = Re(H) Im(H)
with the real subspace
Re(H) := A H; A

= A = R E
and the imaginary subspace
Im(H) := A H; A

= A = R I +R J +R K

= R
3
.
Note that AA

= [A[
2
E, whence
A
2
= [A[
2
E, A Im(H),
this motivates the name imaginary subspace.
Realizations of the quaternion group: Let us now show that, given two
orthogonal purely imaginary unit quaternions A, B Im(H), we obtain an
ON-basis
A, B, AB = BA
33
of Im(H). From that and A
2
= E = B
2
we easily derive that E, A, B, C
with C := AB is a group isomorphic to the quaternion group Q.
First of all AB, BA Im(H) = E

, since for example (AB, E) =

(A, B

) = (A, B) = 0. Hence
AB = (AB)

= B

= (B)(A) = BA.
Furthermore, AB is orthogonal to both A and B:
(AB, B) = (A, BB

) = (A, E) = 0,
while
(AB, A) = (A, AB

) = (AB

, A) = (AB, A) = (AB, A),

whence (AB, A) = 0 as well. As a consequence of this discussion we see
that the product of quaternions is closely related to vector geometry in three
space: The map
Im(H) Im(H) Im(H), (A, B) Im(AB) =
1
2
(AB BA)
is nothing but the vector product of the vectors A, B Im(H)

= R
3
: It
is bilinear, alternating and associates to two orthogonal unit vectors a unit
vector orthogonal to both factors.
Orthogonal and unitary groups: Finally let us use quaternions in order
to nd an interesting relationship between the special unitary group
SU(2) = S
3
= A H; [A[ = 1
and the special orthogonal groups SO(3) and SO(4).
We have an action by conjugation
SU(2) H H, (A, X) AXA

= AXA
1
.
Since [AXA

[ = [X[, that action is isometric. Furthermore since Im(H) =

E

and AEA

= E, the imaginary subspace Im(H)

= R
3
is invariant under
that action; hence we obtain a group homomorphism
SU(2) SO(3).
34
In fact, it is onto and has the kernel E: Write A = cos()E+sin()B with
a matrix B Im(H) and 0 . Take C Im(H) orthogonal to B and
D := BC. Then E, B, C, D is an ON-basis for H and E, B, C, D
S
3
is isomorphic to the quaternion group Q. An explicit calculation now
shows that
ACA

= cos(2)C + sin(2)D, ADA

= sin(2)C + cos(2)D
while AAA

= A implies ABA

= B. With other words X AXA

is a
rotation around the axis R B Im(H). From this it follows immediately
that any transformation in SO(3) can be written in the above form and that
X AXA

is the identity i = 0, resp. A = E.

There is also a similar description for SO(4): We consider the action
SU(2)
2
H H, ((A, B), X) AXB

.
It induces a surjective homomorphism
SU(2)
2
SO(4)
with kernel (E, E), namely: If F : H H is an isometry, consider
G : H H with G(X) := F(E)
1
F(X). Then G(E) = E implies
G(Im(H)) = Im(H), and we nd as above a matrix B SU(2) with G(X) =
BXB

. Finally take A := F(E)B S

3
= SU(2). On the other hand if
AXB

= X for all X, then X = E gives us AB

= E resp. B = A and we
may proceed as above.
Problems 2.33. 1. R: Show: If a, b Z are integers and gcd(a, b) = 1, one has
Za + Zb = Z, in particular 1 can be written in the form 1 = ra + sb with integers
r, s Z.
2. R: Determine all subgroups of S
3
, Q, C
2
C
2
, D
4
!
3. R: Show that the subgroups of C
n
are the groups C
m
with m[n. Hint: Consider
the homomorphism
a
: Z C
n
, n a
n
with a := e
2i
n
.
4. R: Let G be a group and a G, := e
2i
n
. Show: There is a homomorphism
: C
n
G with () = a, i a
n
= 1.
5. R: Show: Aut(C
2
C
2
)

= S
3
.
6. Compute Aut(Q)!
35
7. Show that the Mobius-action
GL
2
(C)

C

C,
see Problem 2.18.12, restricts to an action
SL
2
(R) H H
on the upper half plane H := z = x +iy C; y > 0.
8. Show that for the Mobius action on triplets
GL
2
(K) P
1
(K)
3
P
1
(K)
3
the stabilizer of any triplet (L
1
, L
2
, L
3
) P
1
(K)
3
with pairwise dierent lines L
i

K
2
is K

E GL
2
(K), see also Problem 2.18.11.
9. Show that SL
2
(Z) := SL
2
(Q) Z
2,2
is a subgroup of SL
2
(Q).
10. Show that O(n) = Sym(B
n
), i.e. O(n) is the symmetry group of the closed unit
ball B
n
:= x R
n
; [[x[[ 1.
11. R: Show: Every group is isomorphic to a subgroup of a permutation group S(M)
with a suitable set M.
12. R: Let G be a group. Show that a non-empty set H G is a subgroup i ab
1
H
for all a, b H.
13. R: Let H
1
, H
2
G be subgroups of the group G. Show: The union H
1
H
2
G
is again a subgroup i H
1
H
2
or H
2
H
1
.
14. Show: A nitely generated subgroup H of Q can be generated by one element only,
i.e. H = Za with some a Q.
15. Show that the following statements for a group G are equivalent:
(a) Every subgroup H G is nitely generated.
(b) Each increasing sequence H
1
H
2
... of subgroups H
n
G becomes
constant, i.e. there is n
0
N with H
n
= H
n
0
for n n
0
.
(c) Every set A, whose elements are (certain) subgroups of G, has (at least) one
maximal element H
0
A, i.e. A H H
0
=H = H
0
.
16. Let P R
3
be a regular tetrahedron or a cube with the origin as barycenter. Is
Sym(P) isomorphic to one of the groups you have encountered up to now?
36
2.4 Order and Cyclic Groups
In this section we study the most basic nontrivial groups:
Denition 2.34. For an element a G of a group G the subgroup
a
Z
:= a
n
; n Z G
is called the cyclic (sub)group generated by a G. The group G is
called cyclic i
G = a
Z
for some a G.
In order to understand cyclic groups we need the notion of the order of an
element a G:
Denition 2.35. Let G be a group. The order ord(a) N of the
element a G is dened as
ord(a) := minn N
1
; a
n
= e ,
where we use the convention min := .
Example 2.36. 1. In the multiplicative group C

we have: ord(z) <

i z = exp(2i
p
q
) with a rational number
p
q
Q; in that case
ord(exp(2i
p
q
)) = q, if p, q Z are relatively prime and q > 0.
2. For the elements R, S in the dihedral group D
n
we have ord(R) = n
and ord(S) = 2.
Proposition 2.37. Let G be a group, a G and
a
: Z G the homo-
morphism n a
n
.
1. If ord(a) = , the homomorphism
a
is injective, hence induces an
isomorphism Z

= a
Z
.
2. If ord(a) = d < , then ker(
a
) = Zd, and exp(2i
k
d
) a
k
denes
an isomorphism C
d

= a
Z
. In particular the elements e, a, .., a
d1
G
are pairwise dierent and constitute a
Z
, i.e., a
Z
= e, a, ..., a
d1
, and
a
k
= e i d[k.
3. ord(a) = [ a
Z
[.
37
Proof. 1. According to 2.23 it suces to show ker(
a
) = 0. Therefore let
n ker(
a
), i.e. a
n
= e. Since a
n
= e a
n
= e, we may assume n 0.
But if ord(a) = this is possible only with n = 0.
2. If ord(a) = d < , we even have ker(
a
) = Zd: The inclusion is
obvious; if on the other hand a
n
= e, write n = qd + r, 0 r < d, whence
e = a
n
= (a
d
)
q
a
r
= a
r
. But by the choice of d that implies r = 0 resp.
n Zd. Since exp(2i
k
d
) = exp(2i

d
) i d[( k) i k ker(
a
) i
a
k
= a

, the map exp(2i

k
d
) a
k
is both well dened and injective, while
the surjectivity is immediate.
3. follows immediately from 1. and 2.
Corollary 2.38. Theorem of Lagrange (Joseph Louis Lagrange, 1736 -
1813) Let G be a nite group. Then the order ord(a) of any element a G
divides the group order [G[, or with other words
a
|G|
= e
holds for all a G.
Proof. Since G is nite, every element a G has nite order:
ord(a) = [a
Z
[ = d < . On the other hand according to 2.28.3. the order of
the subgroup a
Z
G divides [G[.
Problems 2.39. 1. R: Let G be a group and a, b G elements of order m, n N
respectively. Show that ord(a
k
) = m/ gcd(k, m) and ord(ab) = mn, if ab = ba and
the numbers m, n are relatively prime.
2. R: Let f S
n
be a permutation. Given a factorization of f as product of pairwise
disjoint cycles determine the order of f!
3. Let G :=

n=1
C
2
n. Show: The group G is not nitely generated, cf. Problem
2.6.9, but every proper subgroup is cyclic (and thus nitely generated), namely
coincides with some C
2
n.
4. R: Show: The direct product GH of two nontrivial cyclic groups is again cyclic,
i both G and H are nite and their orders are relatively prime.
5. Let p be a prime number. Show that S
p
is generated by an arbitrary p-cycle and
any transposition. Hint: Use problem 2.6.10.
6. R: Show: Every automorphism C
n
C
n
is of the form p
k
(a) = a
k
with some
k Z, gcd(k, n) = 1. When does p
k
= p

hold? Hint: 2.39.1.

7. R: Let p be a prime number. Show that a group of order p
2
is isomorphic either to
C
p
2 or to C
p
C
p
! Hint: Corollary 2.32.
38
2.5 Factor Groups
Having discussed cyclic groups as the most basic non-trivial groups one can
attack the problem of classication of (nite) groups by trying to decompose
a given group into smaller pieces. If for example H G is a subgroup of
the group G, one could look at H and the set G/H of all left cosets mod
H. But is G/H again a group? If so, one would expect the quotient map
: G G/H, a aH, to be a group homomorphism. That means nothing
but:
aH bH = abH,
and it looks like we have succeeded in making G/H a group, taking the RHS
as a denition of the LHS. But unfortunately, there is the following problem:
The elements a and b are by no means distinguished representatives of the
cosets aH and bH, and we have to show that the right hand side depends
only on aH and bH as sets (and not on the way to write them), i.e. does not
change if we replace a and b with other representatives a = ah
1
and

b = bh
2
,
where h
1
, h
2
H.
On the other hand if in some given situation the above denition works,
then the subgroup H G is the kernel of the group homomorphism : G
G/H, and subgroups realized as the kernel of a suitable group homomorphism
turn out to be normal:
Denition 2.40. A subgroup H of a group G is called normal if it is in-
variant under conjugation, i.e. if for all a G we have

a
(H) = aHa
1
= aha
1
; h H H .
In that case we also write
H G.
Remark 2.41. 1. For a normal subgroup H G, the above condition is
satised for all a G, in particular also for a
1
, i.e.
a
1
H(a
1
)
1
= a
1
Ha H H aHa
1
.
Hence we could also have required aHa
1
= H for all a G, but the
given condition is a priori easier to check.
2. Since aHa
1
= H is equivalent to aH = Ha, for a normal subgroup
left and right cosets coincide. In particular G/H = HG for a normal
subgroup H G.
39
3. Let us emphasize that for normality we do not require aha
1
= h, h
H, a G. If H satises that condition, it is of course normal, but not
vice versa.
Example 2.42. 1. Let h := (1, 2, 3), g := (1, 2) S
3
. Then h
Z
S
3
is
normal, but g
Z
S
3
is not.
2. In an abelian group every subgroup is normal.
3. The center Z(G) G of a group G is a normal subgroup.
4. A subgroup H G of index (G : H) = 2 is normal, since for a GH
we have aH = G eH = G He = Ha.
5. The kernel ker() of a group homomorphism : G F is a nor-
mal subgroup, since h ker() = (aha
1
) = (a)(h)(a
1
) =
(a)e(a)
1
= e, i.e. aha
1
ker().
6. Note that F H G does not imply F G. For example consider
G = D
4
. We have H := E, R
2
= E, S, SR
2
= S G, but
F = S
Z
= E, S H is not a normal subgroup of D
4
, since RSR
1
=
R
2
S = S.
Hence H has to be a normal subgroup, in order to have on G/H a natural
group structure. Luckily that is the only condition needed:
Proposition 2.43. If H is a normal subgroup of the group G, then
aH bH := abH
denes a group multiplication on G/H, and the coset map : G G/H, a
aH becomes a group homomorphism.
Proof. It is sucient to show that the group multiplication is well dened,
i.e. does not depend on choices of representatives. So let a := ah
1
,

b := bh
2
with elements h
i
H. Then we have
a

b = ah
1
bh
2
= ab(b
1
h
1
b)h
2
= abh
with h = (b
1
h
1
b)h
2
H, since H is normal. With other words a

bH =
abH.
40
Denition 2.44. If H is a normal subgroup of the group G, then the (left)
coset space G/H with the multiplication
aH bH := abH
is called the factor group of G with respect to (or mod(ulo)) H.
Proposition 2.45. Let : G F be a group homomorphism and H
ker() a normal subgroup of G. Then there is a unique homomorphism :
G/H F with = , where : G G/H is the coset map.
We have ker() = ker()/H. In particular, if : G F is surjective,
: G/ ker() F is a group isomorphism: G/ ker()

= F.
We leave the proof as an exercise to the reader.
Notation: If it is clear from the context which groups G and H are involved,
we usually write a instead of aH.
Example 2.46. 1. For G := Z and H := Zn we write
Z
n
:= Z/Zn
additively, i.e.
Z
n
=
_
0 = Zn = 0 +Zn, 1 = 1 +Zn, .., n 1 = (n 1) +Zn
_
.
2. Let G be a group, a G an element of order ord(a) = d < . Then the
homomorphism
a
: Z G, n a
n
, has the kernel ker(
a
) = Zd and
we obtain an isomorphy Z
d

= a
Z
with the isomorphism Z
d
k a
k
.
In particular with G = C
d
, a = exp(
2i
d
) we nd Z
d

= C
d
.
3. The homomorphism exp : R S
1
, e
2i
, induces an isomorphism
R/Z

= S
1
.
4. The homomorphism sign : S
n
Q

induces an isomorphism S
n
/A
n

=
sign(S
n
) = C
2

= Z
2
.
Corollary 2.47. Let G be a group and E, HG normal subgroups of G and
E H. Then H/E is a normal subgroup of G/E, and there is a natural
isomorphism
G/E
H/E

=
G/H, aE(H/E) aH.
41
Proof. The kernel of the coset map G G/H contains E G, hence,
according to 2.45, factors through a unique surjective group homomorphism
G/E G/H, the kernel of which is H/E G/E. Now apply once again
2.45.
Returning to our motivation in the beginning of this section we ask whether
a group G can be reconstructed from a normal subgroup H G and the
factor group G/H. Unfortunately the answer is no: For H = C
2
G = C
4
and

H = C
2
1

G = C
2
C
2
we have

H

= H and

G/

H

= G/H, but

G ,

= G. In order to avoid such counterexamples let us assume that H G

admits a complementary subgroup F G, i.e. such that the composition
F G G/H
of the inclusion F G and the coset map G G/H is an isomorphism.
Then the map
H F G, (h, f) hf
is bijective, but unfortunately only a group homomorphism if fh = hf holds
for all f F, h H.
For example take G := S
3
and H := h
Z

= Z
3
, F := g
Z

= Z
2
with the 3-
cycle h := (1, 2, 3) and the transposition g := (1, 2). Then G ,

= H F, since
HF

= C
3
C
2

= C
6
is abelian. In fact, the information lost when passing
from G to H, G/H is the way how the complementary subgroup F G acts
on the normal subgroup H by conjugation, i.e. the homomorphism
: F Aut(H), f
f
[
H
.
If id
H
, then G

= H F. Otherwise we can reconstruct G from H, F
and as follows:
Denition 2.48. Let F, H be groups and : F Aut(H), f
f
:= (f)
a group homomorphism. Then the group
H

F := (H F, :=

), where

((h, f), (h

, f

)) := (h
f
(h

), ff

)
is called the semidirect product of H and F with respect to the homomor-
phism .
42
So the underlying set of a semidirect product is in any case the cartesian
product, but the group multiplication is the componentwise multiplication
only if : F Aut(H) is the trivial homomorphism:
f
= id
H
for all
f F.
The proof that really is a group multiplication is left to the reader.
Remark 2.49. 1. The sets He and eF are subgroups of H

F
isomorphic to H resp. F, and H e is normal; furthermore
F

= e F H

F (H

F)/(H e)
is an isomorphism.
2. Let G be a group and F, H G subgroups, H normal. If the restriction
[
F
: F G/H of the coset map : G G/H is an isomorphism
and we take : F Aut(H) with
f
:=
f
[
H
: H H, h fhf
1
,
then
H

=
G, (h, f) hf
is a group isomorphism.
3. Let us now consider the case that G/H is cyclic: G/H = a
Z
. Take
F := a
Z

= Z
n
, i.e. n = 0, if ord(a) = and n = ord(a) otherwise
(where Z
0

= Z). Dene d N by H F = (a
d
)
Z
, where we may
assume that d[n. The homomorphism : F Aut(H) is dened by

a
:= (
a
[
H
)

with the restriction

a
[
H
: H H of the conjugation.
We remark that
n
a
=
a
n = id
H
as well as
a
(a
d
) = a
d
. Finally we
obtain an isomorphism
(H

F)/(a
d
, a
d
)
Z

=
G, (h, a

) ha

.
The corresponding abstract construction starts with a group H, natural
numbers d, n N with d[n and a group isomorphism : H H
(playing the r ole of
a
[
H
), such that
n
= id
H
, furthermore an element
h
0
H (corresponding to a
d
H) with (h
0
) = h
0
. The role of F
is played by the (additive) group Z
n
. Let now : Z
n
Aut(H) be
the homomorphism with
k
=
k
. The cyclic subgroup (h
1
0
, d)
Z

(H

Z
n
) is normal, and the group we associate to these data is
(H

Z
n
)/(h
1
0
, d)
Z
.
43
Example 2.50. 1. Let H := C
3
and F := Z
2
. The automorphism group
Aut(C
3
) consists of two maps, the power maps p
1
= id
C
3
and p
1
(remember that p
n
(a) := a
n
), hence there are two possibilities for :
Z
2
Aut(C
3
): Either
1
= id
C
3
= p
1
or
1
= p
1
: C
3
C
3
. In
the rst case we obtain the direct product C
3
Z
2

= C
3
C
2

= C
6
,
while in the second case we get C
3

Z
2

= S
3
.
2. We consider the quaternion group Q, with H := I
Z

= C
4
, F := J
Z

=
C
4

= Z
4
. We nd F H = (E)
Z
= (J
2
)
Z
and thus n = 4, d = 2 with
= p
1
and a = J, h
0
= a
2
= E. So the abstract construction is
Q

= (C
4

Z
4
)/(1, 2)
Z
,
where : Z
4
Aut(C
4
) is determined by
1
= = p
1
and h
0
=
1 C
4
.
2.5.1 Digression: Free Groups
In this section we present a method how to describe innite groups in a
concise way. Indeed, it is even useful for nite groups, since in general it is
not very economic to write down a complete multiplication table. First of
all we need the notion of a set (system) of generators of a group G:
Denition 2.51. Let G be a group, M G a subset. Then the intersection
of all subgroups H G containing M, i.e.
M :=

MHG
H,
is called the subgroup of G generated by the subset M. If G = M, the set
M G is also called a system of generators for the group G.
Remark 2.52. The subgroup M G generated by a subset M G
admits also an explicit description:
M = a
n
1
1
... a
n
r
r
; a
1
, ..., a
r
M, r N
>0
, n
1
, ..., n
r
Z,
the right hand side being a subgroup contained in any subgroup H G
containing M.
44
Notation: If M = c
1
, ..., c
s
we often simply write
c
1
, ..., c
s
:= c
1
, ..., c
s
.
Example 2.53. 1. A group admits a generator system with only one el-
ement i it is cyclic.
2. Z = 1 = 2, 3, so distinct minimal generator systems (a generator
system which can not be shrinked) may have dierent cardinalities!
3. The cyclic group C
n
of all n-th roots of unity is generated by e
2i/n
, i.e.
C
n
= e
2i/n
.
4. The dihedral group D
n
satises
D
n
= R, S
with the counterclockwise rotation R(z) = e
2i/n
z and the reection
S(z) = z at the real axis.
5. The quaternion group Q satises
Q = I, J.
6. The symmetric group S
n
satises
S
n
= T,
where T := (i, j); 1 i < j n is the set of all transpositions.
7. Commutator subgroup C(G) G: If G is any group and
M := aba
1
b
1
; a, b G,
the subgroup C(G) := M is called the commutator subgroup. It is
a characteristic subgroup of G, i.e. invariant under every automor-
phism : G G: (C(G)) = C(G). So in particular it is a normal
subgroup: C(G) G.
45
In order to describe a group G one often nds the following: A system
M G of generators is specied together with a set of relations, i.e. a
number of equalities
L
i
= R
i
, i = 1, ..., s,
where both L
i
and R
i
are products of elements and inverses of elements of
the system M of generators.
Example 2.54. 1. A nite group G = a, b with a G can be charac-
terized by the relations
a
n
= e,
where n denotes the order of a G,
b
m
= a
r
,
where m is the order of b G/a, and the conjugation relation
bab
1
= a
s
.
In particular:
2. For the dihedral group D
n
= R, S we have the relations
R
n
= E, S
2
= E, SRS
1
= R
1
.
3. For the quaternion group Q = I, J a possible set of relations is
I
4
= E, J
2
= I
2
, JIJ
1
= I
3
.
We want to explain what people mean with such a description: First
we should understand the case, where there are no relations at all: Such a
system of generators is called free:
Denition 2.55. A system of generators M F of a group F is called free,
if the pair (F, M) satises the following universal mapping property: Given
any map : M G into a group G, there is a unique group homomorphism
: F G extending , i.e. [
M
= . A group F is called free if it admits
a free system of generators.
Example 2.56. 1. The trivial group F = e is free: Take M = .
46
2. The additive group Z of integers is free: Take M = 1.
3. A free generator system M of a commutative group F is either empty
or contains one element: If M = a, b, ... is a free generator system
of an abelian group F with a ,= b and g, h G arbitrary elements
in some group G, there is a group homomorphism : F G with
(a) = g, (b) = h. But then ab = ba implies gh = hg, so we could
conclude that any group is abelian!
4. The elements of a free system of generators M have innite order (hence
the trivial group is the only free nite group!). If a M, there is a
group homomorphism : F R

with (a) = 2. Since 2 R

has
innite order, the element a F has as well.
Hence given a set M with more than one element, it is not at all clear
whether there is a group F M freely generated by M, but if it exists, it is
unique up to isomorphy:
Remark 2.57. Let F
i
M be groups freely generated by M for i = 1, 2
and
i
: M F
i
the inclusion. Then
2
: F
1
F
2
is an isomorphism. In
order to see that we apply the universal mapping property for (F
2
, M) to the
inclusion
1
: M F
1
and obtain the extension
1
: F
2
F
1
. Since both

1

2
and id
F
1
extend the identity id
M
, we get
1

2
= id
F
1
, and in the
same way,
2

1
= id
F
2
.
Fortunately we have:
Proposition 2.58. For every set M there is a group F(M) M freely
generated by M. Indeed any element g F e has a unique representation
g = a
n
1
1
... a
n
r
r
,
where r N
>0
, a
1
, ..., a
r
M, the exponents are nonzero: n
1
, ...., n
r
Z0
and immediate neighbours are dierent: a
i+1
,= a
i
for 1 i < r.
Before we prove the proposition we come back to our original problem.
Consider any group G with a system of generators M G. Denote : M
G the inclusion. Then the group homomorphism
: F(M) G
47
is onto, hence G

= F(M)/H with the normal subgroup H := ker(). Thus
in order to describe G up to isomorphy it suces to determine a system of
generators for the subgroup H. But since H G is a normal subgroup it is
enough to give a system of generators of H as a normal subgroup: Given a
subset R F(M) we denote N(R) F(M) the smallest normal subgroup
of F(M) containing R; in fact
N(R) =

RNF(M)
N .
Or equivalently, N(R) is the subgroup generated by

F(M)
(R) :=
_
gF(M)

g
(R),
where
g
: F(M) F(M), x gxg
1
, denotes the conjugation with g.
Hence in order to describe a group G completely one gives a set M G
of generators and a set R F(M), such that H = N(R) respectively
G

= F(M)/N(R) .
The elements in R then correspond to relations L
i
= R
i
with the right hand
side R
i
= e.
On the other hand, if it is said that a group G is generated by M with
relations L
i
= R
i
, i = 1, ..., s, one means that the natural map F(M) G
has kernel N(R) with
R = L
i
R
1
i
, i = 1, ..., s,
where L
i
, R
i
are understood as elements in F(M). Of course R = R
1
i
L
i
, i =
1, ..., s or R = L
1
i
R
i
, i = 1, ..., s etc. are possible choices as well.
Example 2.59. 1. For the generator system M = R, S of the dihedral
group D
n
we may take
R := R
n
, S
2
, SRSR F(M).
In order to see that, we consider the surjective group homomorphism
F(M) D
n
; it factors through F(M)/N(R). Hence it is sucient
to show that F(M)/N(R) has at most order 2n = [D
n
[; and this we
leave as an exercise for the reader.
48
2. For the generator system M = I, J of the quaternion group Q we
may take
R = I
4
, I
2
J
2
, JIJ
1
I F(M).
Proof of Proposition 2.58. We take a second (with M disjoint) copy M
1
of
the set M, the elements being denoted b
1
, b M (this is nothing but a
notation). Thus there is a bijection
M M
1
, b b
1
with the inverse
M
1
M, c = b
1
c
1
:= b.
Hence, if A := M M
1
, then A A, a a
1
is a permutation of A
interchanging M and M
1
. Set
W(A) := e, (a
1
, ..., a
r
); a
1
, ..., a
r
A, r N
>0
,
so the elements in W(A) are nite sequences, whose elements belong to A
(they are also called words in the alphabet A), and e denotes the empty
sequence (word). Words can be composed by concatenation and be simplied:
A simplication step looks as follows
u = (a
1
, ..., a
i1
, a, a
1
, a
i+2
, ..., a
r
) u

= (a
1
, ..., a
i1
, a
i+2
, ..., a
r
)
with u

= e for u = (a, a
1
). Call a word u = (a
1
, ..., a
r
) reduced if it can not
be simplied to a shorter word, i.e. if a
i+1
,= a
1
i
for i = 1, ..., r 1. Denote
RW(A) W(A) the subset of all reduced words.
It is clear that any word u W(A) can be transformed into a reduced
word u
0
RW(A) by a nite number of simplications. Indeed, the resulting
reduced word depends only on u and not on the simplication steps applied:
Lemma 2.60. Let u W(A) be a word. Then there is a unique reduced
word u
0
RW(A), such that u
0
is the outcome of any iterated simplication
procedure leading from u to a reduced word.
Let us now derive our claim from Lemma 2.60. Set F(M) := RW(A)
with the group law:
F(M) F(M) F(M), (u, v) (uv)
0
,
49
where uv denotes the concatenation of the words u, v. The neutral element
is the empty word e, the element u = (a
1
, ..., a
r
) has the inverse (a
1
r
, ..., a
1
1
).
Finally associativity is obtained as follows:
((uv)
0
w)
0
= (uvw)
0
= (u(vw)
0
)
0
.
Namely: The left hand side is obtained from uvw by simplifying rst to
(uv)
0
w and then to the reduced word ((wv)
0
u)
0
. Now Lemma 2.60 gives
the rst equality; the second one follows with an analogous argument.
Identify A = M M
1
with the one letter words in F(M) = RW(A).
Now given a map : M G extend it rst to : A G by setting
(a
1
) := (a)
1
and then to

: W(A) G with

((a
1
, ..., a
r
)) := (a
1
) ... (a
r
), (e) = e
G
where (a
1
, ..., a
r
) is any word. Obviously

(uv) =

(u)

(v) and

(w
0
) =

(w). It follows immediately that :=

[
RW(A)
: F(M) = RW(A) G is
a group homomorphism.
Proof of Lemma 2.60. We do induction on the length of the word u. Assume
the word u may be reduced to the reduced words u
1
and u
2
. If u itself is
reduced, we obviously have u
1
= u = u
2
. Otherwise denote v
i
the result of
the rst simplication on the way from u to u
i
. If v
1
= v
2
=: v, we may
apply the induction hypothesis to v and obtain u
1
= u
2
. Otherwise u =
(a
1
, ..., a
r
) and v
1
= (a
1
, ..., a
i1
, a
i+2
, ..., a
r
), v
2
= (a
1
, ..., a
j1
, a
j+2
, ..., a
r
),
where we may assume i j. Since j = i, i + 1 gives v
1
= v
2
, we have
i + 2 j, and then v := (a
1
, ..., a
i1
, a
i+2
, ..., a
j1
, a
j+2
, ..., a
r
) for j > i + 2
resp. v := (a
1
, ..., a
i1
, a
j+2
, ..., a
r
) for j = i +2 is both a simplication of v
1
and v
2
.
But v
i
being shorter than u has according to the induction hypothesis u
i
as its unique reduction. Since we can obtain it via an iterated simplication
procedure through v, we have u
1
= u
2
.
Remark 2.61. It is not dicult to see that F(M)

= F(N) implies [M[ =
[N[. A more surprising fact is, that any subgroup H F(M) of a free group
again is free, i.e. H

= F(N) with some set N, but that not necessarily
[N[ [M[, if [M[ 2.
Problems 2.62. 1. R: Show: C/Z

= C

= C.
50
2. R: Determine all groups of order 6.
3. R: Determine all non-commutative groups of order 8.
4. R: Let H S
4
be the subgroup consisting of the identity and the products of two
disjoint 2-cycles. Show: H S
4
is a normal subgroup and S
4
/H

= S
3
. Hint:
S
3
H = id.
5. Let A
n
(R) := f S(R
n
); f(x) = Ax +b with A GL
n
(R), b R
n
be the ane
linear group, cf. problem 2.6.3. Show: The subgroup T :=
b
; b R
n
(where

b
(x) = x + b is the translation with the vector b R
n
) is normal. Determine a
homomorphism : GL
n
(R) Aut(R
n
), such that A
n
(R)

= R
n

GL
n
(R)! Is
GL
n
(R) A
n
(R) a normal subgroup?
6. R: We dene on Z

n
:= a Z
n
; gcd(a, n) = 1 a group structure: a b := ab. Show
that this denes a group multiplication, and that Z

n
Aut(C
n
), k p
k
: C
n

C
n
with p
k
() =
k
is a group isomorphism. Hint: Problem 2.39.6. If we replace
C
n
with the isomorphic group Z
n
, what does the corresponding automorphism look
like?
7. Let p be a prime number. Give an example of a non-abelian group of order p
3
.
Hint: Consider semidirect products Z
p
2

Z
p
and use the previous problem.
8. Let p be a prime, n N
>0
. Show that
U(p
n
) := 1 +kp Z

p
n; k Z
is a subgroup of Z

p
n of order p
n1
, and that U(p
n
) is cyclic for p > 2, more precisely
U(p
n
) = 1 +rp
Z
for any r , Zp.
9. Show: The permutations f : Z
n
Z
n
of the form f(x) = kx + b, where k Z
and gcd(k, n) = 1 as well as b Z
n
constitute a subgroup of S(Z
n
). Show that it is
a semidirect product Z
n

n
! Cf. with problem 2.62.6 and the group A
1
(R) in
problem 2.62.5.
10. Let , : F Aut(H) be group homomorphisms. Assume there are automor-
phisms : F F and : H H, such that
f
=
(f)

1
for all f F.
Show: H

F

= H

F.
11. Assume G = a with the relations a
n
= e, a
m
= e, where n, m N
>0
. Compute
[G[!
12. Let n, m, r N
>0
and G = a, b be the group generated by the two elements a, b
with the relations a
n
= e = b
m
, ba = a
r
b, where n, m, r N
>0
. Show [G[ nm
with equality i r
m
1 mod (n). Compute the order of a and b as well as [G[ in
the general case!
Hint: a G and b Aut(a), g
g
is a group homomorphism.
13. Write C
2
:=

n=1
C
2
n as a factor group F(M)/N(R)!
14. Let M be a nite set, say [M[ = n. Show F(M)/C(F(M))

= Z
n
. Use that in order
to conclude: F(M)

= F(N) =[M[ = [N[, where M, N denote nite sets.

51
15. Let M := a, b be a set with two elements. Set
M
0
:= b
k
ab
k
; k N.
For the subgroup M
0
F(M) show M
0

= F(M
0
).
2.6 Simple Groups and Composition Series
We motivated factor groups with the idea to break down a given group into
smaller pieces. The nal idea behind is that a group should somehow be
composed of smallest pieces, atomic groups:
Denition 2.63. A group G is called simple if there are no normal sub-
groups of G except the trivial subgroup e and the entire group G itself.
Example 2.64. A non-trivial abelian group is simple i it is cyclic of prime
order, i.e. isomorphic to C
p
(or Z
p
) for some prime number p.
Here is a series of non-commutative simple groups:
Proposition 2.65. The alternating groups A
n
with n 5 are simple.
Proof. The proof is divided into three steps:
1.) The conjugacy class
A
n
((1, 2, 3)) consists of all 3-cycles: Let (i, j, k) be
any three cycle. For the permutation g S
n
with g(1) = i, g(2) = j, g(3) = k
and g() = for > 3 we have (i, j, k) =
g
((1, 2, 3)). If g has sign 1 and
thus g A
n
, we are done, otherwise replace g with g := g A
n
, where
= (r, s) with two dierent numbers ,= i, j, k.
2.) The group A
n
is generated by 3-cycles: According to Problem 2.6.7 any
f S
n
is a product of transpositions; if even f A
n
, there is an even
number of such factors (transpositions having sign -1); so it will be sucient
to represent any product ,= id of two transpositions as a product of 3-cycles.
Consider rst f = (i, j)(k, ) with four pairwise dierent numbers i, j, k, .
Then f = (i, j)(j, k)(j, k)(k, ) = (i, j, k)(j, k, ). The second non-trivial
case is f = (i, j)(j, k) with three pairwise dierent numbers i, j, k. Then
f = (i, j, k) is itself a 3-cycle!
3.) Finally we show that a non-trivial normal subgroup N A
n
contains a
3-cycle. We take any f N id. If it already is a 3-cycle, we are done.
Otherwise we consider the elements h := f
g
(f
1
) N for g A
n
. We
rewrite f
g
(f
1
) =
f
(g)g
1
and choose g as a 3-cycle g = (a, b, c), such
that
f
(g) = (f(a), f(b), f(c)). We distinguish three cases:
52
a) If f contains a cycle (i, j, k, , ...)...(...) of length at least 4, we take
g = (i, j, k) and obtain
f
(g) = (j, k, ), g
1
= (k, j, i) and h = (i, , j).
b) If f = (i, j, k)(, m, ...)...(...) contains a 3-cycle, we take g = (i, j, )
and obtain
f
(g) = (j, k, m), g
1
= (, j, i) and h = (i, , k, m, j), hence can
apply the case a) with h instead of f.
c) If f = (i, j)(k, )...(...) contains two 2-cycles, we choose m dierent
from i, j, k, and take g = (i, k, m) and obtain
f
(g) = (j, , f(m)), g
1
=
(m, k, i) and then have the following possibilities for h: If f(m) = m, then
h = (i, j, , m, k), so we may apply the case a) with h instead of f. Otherwise
h = (j, , f(m))(m, k, i) and we may again apply the case b) with h instead
of f.
Remark 2.66. The classication of all nite simple groups was a great
challenge; in fact, it has been completed only so late as in the early 1980-
ies: There are 17 series of nite simple groups, the rst one consisting of
the alternating groups A
n
, n 5. The remaining 16 series contain the simple
groups of Lie type: Given a nite eld F (cf. section 4.5), their construction
is analogous to that of simple real or complex Lie groups (Marius Sophus Lie,
1842 - 1899) with F replacing R resp. C: They are realized as factor groups
of subgroups of the general linear group GL
n
(F) over the nite eld F. The
rst of these 16 series is discussed in Theorem 4.57. Eventually, there are 26
sporadic simple groups, which do not t in one of the above 17 series. For
more detailed information see [3].
Now let us consider an arbitrary nite group G and study, how we can
nd the simple groups it is composed of. That is done using the notion of
a composition series:
Denition 2.67. Let G be a group. A normal series is a nite increasing
sequence of subgroups
G
0
= e G
1
... G
r
= G ,
i.e. G
i
is a normal subgroup of G
i+1
for i < r. The successive factor groups
G
i+1
/G
i
for i = 0, .., r 1 are called the factors of the normal series. A
composition series is a strictly increasing normal series with simple fac-
tors.
53
Every nite group G admits a composition series, as one proves by in-
duction on the group order: Choose a maximal proper normal subgroup H.
Then H has by induction hypothesis such a composition series and we can
extend it with G, since G/H is simple.
In fact the multiplicity of a simple group as a factor of a composition
series of a given group G depends only on G itself, i.e. is independent of the
actual composition series:
Proposition 2.68. Theorem of Jordan-Holder (Camille Jordan, 1838-
1922, and Otto Holder, 1859-1937) Let G be a nite group and e = G
0

G
1
... G
r
= G as well as e = H
0
H
1
... H
s
= G composition
series. Then s = r and there is a permutation f S
r
, such that for j = f(i)
we have
G
i
/G
i1

= H
j
/H
j1
.
Proof. We forget about the condition that a composition series be strictly
increasing and prove the statement for normal series with simple factors, but
of given length - we may extend the shorter series by adding terms e or G.
This gives our theorem, since the trivial group then has the same multiplicity
in both sequences.
Assume rst r = s = 2. Then our composition series are of the form
e E G and e F G with simple normal subgroups E, F G
and simple G/E, G/F. If G itself is simple or F = E, nothing remains to be
shown. Otherwise we have w.l.o.g. E F F. But then, F being simple,
the proper normal subgroup E F is trivial. Now consider the injective
homomorphism E G G/F: Its image is a normal non-trivial subgroup
of G/F, since E G is normal, hence the entire group, i.e. E

= G/F. By
symmetry we have F

= G/E as well.
In the general case we consider the groups
G
ij
:= G
i
H
j
G
and composition series of length r +s starting with G
00
and ending with G
rs
and inclusion steps
G
ij
G
i+1,j
or G
ij
G
i,j+1
.
Then the composition series
G
00
G
10
... G
r0
G
r1
.... G
rs
54
resp.
G
00
G
01
... G
0s
G
1s
.... G
rs
has the same non-trivial(!) factors as H
0
H
1
... H
s
resp. G
0
G
1

... G
r
, and the rst one can be connected to the second one by a chain of
composition series of length r +s, such that two successive series dier only
in two successive inclusions:
G
ij
G
i+1,j
G
i+1,j+1
is replaced with
G
ij
G
i,j+1
G
i+1,j+1
.
But then it is clear from our initial argument that the two successive com-
position series have the same simple factors taken with multiplicities.
If all the simple factors of a nite group are cyclic, it can successively
be obtained from cyclic groups using the extension procedure described in
2.49.3. Such groups are called solvable:
Denition 2.69. A group G is called solvable if it has a normal series with
abelian factors.
In fact we have:
Lemma 2.70. A nite group G is solvable if and only if all its simple factors
are cyclic of prime order.
Proof. The non-trivial implication is as follows: Take a normal series with
abelian factors G
i+1
/G
i
, i = 1, ..., r 1. For each factor of that normal series
take a series with simple factors
eG
i
= H
i+1
0
... H
i+1
s
i+1
= G
i+1
/G
i
and then rene the original normal series by inserting the subgroups
1
i+1
(H
i+1
j
), j =
0, ..., s
i+1
for i = 1, ..., r 1. Here
i+1
: G
i+1
G
i+1
/G
i
denotes the coset
map. As a consequence of 2.47 the simple factors of G are, counted with
multiplicities, exactly the simple factors of the G
i+1
/G
i
, i = 1, ..., r 1.
Remark 2.71. 1. Abelian groups are solvable.
55
2. A subgroup H G of a solvable group is solvable: A normal series
G
0
... G
r
with abelian factors for G induces a normal series H
i
:=
G
i
H H for H with factors H
i+1
/H
i
G
i+1
/G
i
, i.e. the restricted
sequence has abelian factors as well.
3. If H is a normal subgroup of a group G, we have: G is solvable if
both, H and G/H are. That follows immediately from the fact that
the simple factors of G are obtained as the union of the simple factors
of H and those of G/H.
4. A group G of order [G[ < 60 is solvable. This can be seen using the
results of the next section.
We shall prove here that p-groups are solvable:
Denition 2.72. Let p be a prime number. A group G is called a p-group
if [G[ = p
r
is a power of p.
Proposition 2.73. A p-group is solvable
Proof. We use induction on [G[. The center Z(G) G is, according to
Corollary 2.32 and Example 2.42.3 a non-trivial normal subgroup. Hence
we may apply the induction hypothesis to G/Z(G) and conclude that with
G/Z(G) and Z(G) (the latter being abelian) also G is solvable.
In fact, all nite groups of odd order are solvable, as conjectured 1902
by Burnside (William Burnside, 1852-1927) and proved 1963 by Feit and
Thompson (Walter Feit, 1930- , John Griggs Thompson, 1932- ).
Problems 2.74. 1. R: For G = C
pq
with dierent primes p, q determine all composi-
tion series!
2.7 Abelian Groups
In this section we present a complete classication of nite (or more generally:
nitely generated) abelian groups: They are all direct products of cyclic
groups. In order to have a systematic notation we shall write all groups
additively, such that for example ab, a
n
, e, aH is replaced with a + b, na, 0
and a + H.
An abelian group contains as a characteristic subgroup, i.e. a subgroup
invariant under all automorphisms, its torsion subgroup:
56
Denition 2.75. Let G be a group. We denote T(G) G the subset
T(G) := a G; ord(a) <
consisting of all torsion elements, i.e. elements of nite order. For an abelian
group T(G) G is a subgroup, invariant under all automorphisms. We call
G torsion free if T(G) = 0.
We remark that for nonabelian groups T(G) G need not be a subgroup.
As one easily sees the factor group G/T(G) is torsion free. So one could start
the classication of abelian groups by looking for a complementary subgroup
F G, i.e. such that F G G/T(G) is an isomorphism, and hence,
G being abelian, even G

= T(G) F. Such a complementary subgroup F

exists always for a nitely generated group (though not in a natural way):
Denition 2.76. An abelian group G is called nitely generated, if there
is a surjective homomorphism Z
n
G for some natural number n N,
i.e. if there are elements a
1
, ..., a
n
G, such that
G = Za
1
+ ... +Za
n
:= k
1
a
1
+ ... + k
n
a
n
; k
1
, ..., k
n
Z .
A nitely generated abelian group G is called free, if G

= Z
n
for some n N.
Here we have to insert a warning about the use of the word free: A
nitely generated free abelian group is not a group which is free (cf. 2.55),
nitely generated and abelian! In fact, a free group F(M) is abelian if and
only if [M[ 1 (then F(M) is trivial or isomorphic to Z); and vice versa,
the nitely generated free abelian group Z
n
is a free group only for n = 1.
For such a group G, the number n, such that G

= Z
n
, is called the rank
of G. It is well dened because of
Lemma 2.77. If Z
r

= Z
s
, then r = s.
Proof. If Z
r

= Z
s
, then also
(Z
2
)
r

= Z
r
/2Z
r

= Z
s
/2Z
s

= (Z
2
)
s
,
whence 2
r
= [Z
r
2
[ = [Z
s
2
[ = 2
s
resp. r = s.
Example 2.78. 1. A nite (abelian) group is nitely generated.
57
2. The additive group Q is not nitely generated: If Q = Za
1
+ ... +Za
n
with a
i
=
p
i
q
i
, p
i
Z, q
i
Z 0, we would have Q q
1
Z with
q := q
1
... q
n
. Contradiction!
Eventually we shall not go on with a direct proof of the existence of
a subgroup complementary to the torsion subgroup; instead it will be a
byproduct of a more general theorem.
First note that an abelian group G is nitely generated i it is isomorphic
to a factor group F/H with a subgroup H F of a nitely generated free
abelian group F

= Z
n
. The next proposition is a generalization of Proposi-
tion 2.25: It describes the possible subgroups H F up to an automorphism
of F (Note that F = Z admits only the automorphisms Z Z, x x):
Theorem 2.79. Let H F be a subgroup of the nitely generated free
abelian group F

= Z
n
. Then there are natural numbers q
1
, ..., q
n
N with
q
i
[q
i+1
and an isomorphism
: F

=
Z
n
, such that (H) = q
1
Z ... q
n
Z Z
n
.
We remark that the numbers q
1
, ..., q
n
are uniquely determined by H F
as a consequence of Th. 2.84. The proof of the above theorem is divided
into several steps, formulated as lemmata. Let us rst show that H itself is
again a nitely generated free abelian group:
Lemma 2.80. Every subgroup H Z
n
is isomorphic to a group Z
r
, r n.
Proof. We prove the lemma by induction on n. The case n = 1 is nothing
but Prop. 2.25. For n > 1 take H
0
:= H (Z
n1
0). According to the
induction hypothesis there is an isomorphism : Z
s
H
0
, s n 1. Let
now :=
n
: Z
n
Z be the projection onto the last component. Its image
(H) Z is a subgroup and has therefore, again because of Prop. 2.25, the
form (H) = Zq with some q N. If q = 0, we have H = H
0
and we are
done. Otherwise choose v H with (v) = q. Then Z
s
Z H, (u, k)
(u) + kv is an isomorphism.
Furthermore we need the following easy, but useful fact:
58
Lemma 2.81. Let F be a nitely generated free abelian group and e F a
primitive element (or vector), i.e.
e = w, Z, w F = = 1.
Then there is a group homomorphism : F Z with (e) = 1.
Proof. We may assume F = Z
n
and e = (k
1
, ..., k
n
) with gcd(k
1
, ..., k
n
) = 1.
But then there are integers r
1
, ..., r
n
Z with r
1
k
1
+ ... + r
n
k
n
= 1 and we
dene : F Z by (x
1
, ..., x
n
) = r
1
x
1
+ ... + r
n
x
n
.
Remark 2.82. As we see from the proof of Lemma 2.80, a nonzero ele-
ment e F is primitive, i the subgroup Ze F admits a complementary
subgroup F
0
F, i.e. such that the inclusions Ze, F
0
F induce an iso-
morphism F
0
Ze

= F. Equivalently, an element e F is primitive, i
it can serve as the rst element of a basis of F, i.e. i there are elements
e
2
, ..., e
n
F, such that e
1
:= e, e
2
, ..., e
n
is a basis of F, i.e. such that
Z
n
F, (k
1
, ..., k
n
) k
1
e
1
+ ... + k
n
e
n
, is a group isomorphism.
The essential argument in the proof of Th. 2.79 is the following:
Lemma 2.83. Let H F be a subgroup of the nitely generated free abelian
group F. If H is not contained in kF for any k > 1, then H contains a
primitive vector.
Proof. We consider the set Pr(F) of all projections, i.e. surjective group
homomorphisms : F Z. Choose some Pr(F) with maximal (H)
Z. Writing (H) = Zq, q 0, we are done if we succeed in showing q = 1,
since an element v F with (v) = 1 is primitive. Assume the contrary:
q > 1. Take now v H with (v) = q. Write v = e with 1 and
a primitive vector e F. Obviously q = (e). We show = q and
thus (e) = 1. By Lemma 2.81 there is Pr(F) with (e) = 1 resp.
(v) = . Thus (H) Z Zq = (H) resp. (H) = (H) because
of the maximality of (H). With other words q = and (e) = 1. Let
F
0
:= ker(), H
0
:= H F
0
. Then the map
F
0
Z F, (u, k) u + ke
is an isomorphism restricting to an isomorphism
H
0
Zq H.
59
Since H , qF, there is a vector v
0
= e
0
H
0
with , Zq and primitive
e
0
F
0
. Now apply Lemma 2.80 once again and obtain a projection
0
:
F
0
Z with
0
(e
0
) = 1. Then :=
0
+id
Z
: F
0
Z Z is a projection
with , q (H), in particular (H) = Zq (H), a contradiction.
Proof of 2.79. We use induction on n. Take q 1 maximal with H qF.
We apply Lemma 2.83 to H qF and nd a primitive vector e F with
qe H (the primitive vectors in qF are of the form qe with primitive e F).
Choose a projection : F Z with (e) = 1 and dene F
0
H
0
as in the
proof of 2.83. By the induction hypothesis H
0
F
0
satises 2.79, so we nd
numbers q
2
, ..., q
n
with the given properties. Set q
1
:= q and note that q[q
2
because of H
0
qF
0
.
As a corollary we obtain the classication of all nitely generated abelian
groups:
Theorem 2.84. Fundamental Theorem on nitely generated abelian
groups: A nitely generated abelian group G is isomorphic to a nite direct
product of cyclic groups:
G

= Z
q
1
... Z
q
n
,
where Z
0
:= Z and the natural numbers q
1
, ..., q
n
N 1 satisfy one of the
following two conditions:
1. The number q
i
divides q
i+1
for i = 1, ..., n 1.
2. All q
i
> 0 are prime powers.
The numbers q
1
, ..., q
n
are unique in case 1) and unique up to order in
case 2). (PS: The number n need not be the same in both cases!).
If we apply the above theorem to G := F/H in Th. 2.79, we see that the
numbers q
1
, ..., q
n
there are uniquely determined by H F.
Proof. Existence: 1.) Let us start with the rst case: We have G

= Z
n
/H
with a subgroup H Z
n
as in Th.2.79; hence
G

=
Z ... Z
q
1
Z ... q
n
Z

= Z
q
1
... Z
q
n
,
where we of course may assume q
i
,= 1 for i = 1, ..., n.
2.) We use the rst part and apply the below Chinese remainder theorem
to all q = q
i
> 1.
60
Proposition 2.85 (Chinese Remainder Theorem). Let q = p
k
1
1
... p
k
r
r
be
the prime factorization of q Z (p
1
, .., p
r
pairwise distinct). Then
Z
q
Z
p
k
1
1
... Z
p
k
r
r
, = +Zq ( +Zp
k
1
1
, ..., +Zp
k
r
r
)
denes an isomorphism of groups.
Proof. If + Zp
k
i
i
= 0 for all i = 1, ..., r, then all p
k
i
i
divide the number ,
hence q[ resp. +Zq = 0. So our group homomorphism is injective, but then
also surjective since the start and the target group have the same order.
Uniqueness: The number r 0 of zeroes in q
1
, ..., q
n
is nothing but the rank
of the free abelian group G/T(G). The numbers q
i
> 0 can be read o from
T(G) G as follows:
2.) For m N
>0
the m-torsion subgroup
T
m
(G) := a G; ma = 0
behaves as follows:
Remark 2.86. 1. T
m
(GH) = T
m
(G) T
m
(H).
2. T
m
(Z) = 0.
3. For gcd(m, n) = 1 we have
T
m
(Z
n
) = 0,
since with r, s Z, rm + sn = 1, and a T
m
(Z
n
) we have
a = (rm + sn)a = r(ma) + s(na) = r 0 + s 0 = 0.
4. Let p N
>1
be a prime number. Then
T
p
(Z
p
k) =
_
p
k
Z
p
k , if k
Z
p
k , if k
.
Hence
T
p
+1(Z
p
k)
T
p
(Z
p
k)

=
_
Z
p
, if < k
0 , if k
.
61
As a consequence we obtain
[T
p
+1(G)/T
p
(G)[ = p
s
,
where s = s(p, ) is the number of the q
i
= p
k
with k > . Obviously the
numbers s(p, ) determine the prime powers q
i
> 0
1.) Given the numbers q
i
in a decomposition of type 1.) and a prime
number p, denote (p, i) 0 the multiplicity of p as divisor of q
i
. Now a
decomposition of type 2.) is determined by the group G itself and provides
for every prime number p the supply of possible (p, i). Since on the other
hand (p, i) (p, i + 1), we can reconstruct the numbers q
i
.
2.7.1 Digression: Free Abelian Groups
In many applications one even needs not necessarily nitely generated free
abelian groups. We give here a short comment only: Let M be a set. We
consider the group
Z
M
:= f : M Z
of all maps from M to Z with the argument-wise addition (f + g)(x) =
f(x) + g(x). Its subgroup
Z[M] := f Z
M
, [f
1
(Z 0)[ <
containing the maps f : M Z, which are non-zero only on a nite subset
of M is called the free abelian group generated by M. If for a M, we
denote
a
the map
a
(x) =
ax
, any f Z[M] can uniquely be written
f =

aM
n
a

a
with n
a
:= f(a). Formally, the above sum is innite, and as such not well
dened in the framework of algebra, but since n
a
= 0 for only nitely many
a M, one can dene

aM
n
a

a
:=

a,n
a
=0
n
a

a
.
Furthermore one usually writes simply a instead of
a
and thinks of the
elements in Z[M] as nite formal sums

aM
n
a
a
62
with integral coecients in the elements of M. Now a free abelian group is
dened to be a group isomorphic to a group Z[M]; so an abelian group is
free i there is a subset M (a basis), such that any element has a unique
representation as a nite linear combination

aM
n
a
a. Finally note that
the abelian group Z[M] satises a similar universal property as F(M): Any
map : M G to an abelian group G has a unique extension to a group
homomorphism : Z[M] G, or, more down to earth, the values of a
group homomorphism can be arbitrarily prescribed on the elements of M,
and these values determine the entire homomorphism. We leave it to the
reader to check that
Z[M]

= F(M)/N(K)
with the set
K := aba
1
b
1
; a, b M
of all commutators of elements a, b M. Note that N(K) F(M) is
the subgroup generated (in the ordinary sense) by all commutators aba
1
b
1
with a, b F(M), the conjugate of a commutator being again a commutator.
Problems 2.87. 1. R: Give an example of a (non-commutative) group G, for which
the elements of nite order do not constitute a subgroup!
2. R: Let H Z
2
be the subgroup generated by (a, b), (c, d) Z. Write the factor
group Z
2
/H as in Th.2.84!
3. Show that Aut(Z
n
)

= GL
n
(Z) := A Z
n,n
, det(A) = 1. Furthermore for
A Z
n,n
, det A ,= 0, that the index of the subgroup A(Z
n
) Z
n
is [ det A[.
4. R: Write the groups Z

n
, cf. 2.62.6, for n = 13, 16, 25, 72, 624 as a direct product
of cyclic groups as in Th.2.84! Hint: The Chinese remainder isomorphism 2.85
induces an isomorphism Z

q
Z

p
k
1
1
... Z

p
k
r
r
for q = p
k
1
1
... p
k
r
r
.
5. Show: Z

= Z
2
and Z

2
n

= Z
2
Z
2
n2 for n > 2. More precisely
Z

2
n

=< 1 > < 1 + 4r > with any odd number r Z.
6. Show that T(G) G in general does not have a complementary subgroup F G ,
i.e. such that F G G/T(G) is an isomorphism.
2.8 Sylow Subgroups
According to Proposition 2.73 a p-group is solvable, hence, at least theoreti-
cally, understandable. That explains, why one in the study of a general nite
group G looks for subgroups which are p-groups.
63
Denition 2.88. A subgroup H G of a nite group G is called a p-Sylow-
subgroup (Peter Ludvig Mejdell Sylow, 1832-1918), if [H[ is the maximal
p-power dividing [G[.
First of all we are not talking about the empty set:
Theorem 2.89. Let G be a nite group. For every prime p dividing the
group order [G[, there is a p-Sylow-subgroup.
Proof. We do induction on the group order [G[: Choose a system of represen-
tatives x
1
, ..., x
r
G of the non-trivial conjugacy classes, i.e. [
G
(x
i
)[ > 1.
If an index (G : G
x
i
) of an isotropy group is not divisible with p, then a
p-Sylow subgroup of G
x
i
is also a p-Sylow subgroup of G, and the induction
hypothesis can be applied to the proper subgroup G
x
i
G. Otherwise as a
consequence of Proposition 2.32 also [Z(G)[ is divisible with p, and there is
an element a Z(G) of order p: This follows immediately from 2.84. But a
commutes with all elements in G; so < a > G is a normal subgroup and we
may apply the induction hypothesis to the factor group G/ < a > and nd
a p-Sylow subgroup H G/ < a >. Then the inverse image
1
(H) G
with respect to the coset map : G G/ < a > is a p-Sylow subgroup of
G.
Theorem 2.90. Let G be a nite group and p a prime number. Then
1. Any p-subgroup F of G (i.e., F G is a subgroup and [F[ is a p-power)
is contained in a p-Sylow-subgroup.
2. Any two p-Sylow-subgroups are conjugate.
3. The number of p-Sylow-subgroups divides the group order [G[ and has
the form 1 + kp with a natural number k N.
Proof. The rst two parts are an immediate consequence of the following
fact: Given a p-Sylow subgroup H G and a p-subgroup F G, there is
an a G with F
a
(H): Denote Syl
p
(G) the set of all p-Sylow subgroups
of G. The group G acts by conjugation on Syl
p
(G):
GSyl
p
(G) (g, H)
g
(H) = gHg
1
Syl
p
(G) .
Choose some p-Sylow-subgroup H Syl
p
(G). Its stabilizer G
H
:= g
G;
g
(H) = H is usually called the normalizer of the subgroup H in G
64
and denoted N
G
(H). Obviously N
G
(H) = G
H
contains H - indeed N
G
(H)
is the largest subgroup of G containing H as normal subgroup - , hence
its index (G : G
H
) is not divisible with p. We consider now the induced
F-action F Syl
p
(G) Syl
p
(G) (i.e., obtained by restriction): The G-
orbit
G
(H) Syl
p
(G) of H Syl
p
(G) then is a disjoint union of F-orbits

F
(H
i
), i = 1, ..., r with H
i
=
a
i
(H), and the number of elements in such
an F-orbit is some p-power = p
r
i
= (F : F
H
i
). If r
i
> 0 for all i, then
[
G
(H)[ = (G : G
H
) is divisible with p. So there is an F-orbit consisting
only of one element, say H
1
. With other words, F normalizes H
1
, and thus
FH
1
:= fh
1
; f F, h
1
H
1
G
is a subgroup satisfying
(FH
1
)/H
1

= F/(F H
1
)
with an isomorphism induced by the homomorphismF FH
1
(FH
1
)/H
1
,
the composite of the inclusion and the factor map.
As a consequence it is as a factor group of the p-group F again a p-group,
but on the other side also H
1
is p-group, and thus FH
1
as well because of
[FH
1
[ = [H
1
[ [(FH
1
)/H
1
[ = [H
1
[ [F/(F H
1
)[. Since [H
1
[ is the maximal
p-power dividing [G[, it follows that H
1
= FH
1
resp. F H
1
with the
p-Sylow-subgroup H
1
=
a
1
(H).
In particular G acts transitively on Syl
p
(G) and therefore [Syl
p
(G)[ = (G :
G
H
) is a divisor of [G[, where H Syl
p
(G) is arbitrary. But a p-Sylow-
subgroup F instead of G acts no longer transitively because of
F
(F) = F,
while according to the above reasoning the remaining F-orbits contain more
than one element: Since the number of elements in such an orbit is a p-power
p
r
, r > 0, we have shown 3).
As an application we prove:
Theorem 2.91. A non-abelian simple group G of order [G[ 60 is isomor-
phic to the alternating group on 5 letters: G

= A
5
. In particular, a group G
of order [G[ 60 is either solvable or isomorphic to A
5
.
Proof. First, given a group G of non-prime order [G[ < 60 we hunt for a
non-trivial normal proper subgroup H G, then consider the case [G[ = 60:
If [G[ = p
r
, r 2, we may choose H := Z(G) ,= e, cf. Corollary 2.32.
The next case is:
65
Proposition 2.92. Let p, q be dierent prime numbers. Then any group of
order pq or pq
2
has a normal Sylow subgroup.
Proof. Let us rst consider the case p < q. Then the number of q-Sylow
subgroups is of the form 1 + nq, and we want to show n = 0. Since on the
other hand 1 + nq divides [G[ = pq
2
or = pq, we obtain (1 + nq)[p, but
that is obviously possible only for n = 0. Secondly, if p > q, we denote
1 + np the number of p-Sylow subgroups of G. Now we get (1 + np)[q
2
, i.e.,
1 + np = 1, so n = 0, or 1 + np = q (but that is absurd!) or 1 + np = q
2
.
Thus p[(q
2
1) = (q + 1)(q 1), whence p[(q + 1) or rather p = q + 1,
i.e., q = 2, p = 3, in particular [G[ = 12. We now assume that a 2-Sylow
subgroup H G is not normal and show that then there is exactly one (and
hence normal) 3-Sylow subgroup: Take a conjugate H

. Since [H H

[ 2,
we have at least 5 elements of order 2 or 4 in G, hence at most 6 elements
of order 3. As a consequence, there are not more than 3 dierent 3-Sylow
subgroups. On the other hand, there are 1 + 3n such subgroups; so n = 0 is
the only remaining possibility.
Hence the cases still to be considered are [G[ = 24, 30, 36, 40, 42, 48, 54, 56.
We remark rst, that if p, but not p
2
, divides G, and H G is a non-normal
p-Sylow subgroup, then G contains at least (p + 1)(p 1) = p
2
1 elements
of order p.
Proposition 2.93. A group G of order 30 or 56 has a normal Sylow sub-
group.
Proof. Assume [G[ = 30 = 2 3 5. If no Sylow subgroup is normal we
get, counting the elements of order 1, 2, 3, 5, following the above reasoning
30 1 + 3(2 1) + 4(3 1) + 6(5 1) = 36, a contradiction. Now assume
[G[ = 56 = 7 8. If there is no normal 7-Sylow subgroup, we obtain at least
48 = 8(7 1) elements of order 7, so there are at most 7 elements of even
order. But that means, that there is only one 2-Sylow-subgroup.
Proposition 2.94. A group of order 40, 42 or 54 admits a normal p-Sylow
subgroup with p the biggest prime dividing the group order.
Proof. If 1 +np divides [G[, then necessarily n = 0.
66
When looking at the cases [G[ = 24, 36, 48, 60 we shall encounter the
following situation: Let F G be a subgroup. Denote
X :=
G
(F) =
g
(F); g G
the set of all subgroups of G conjugate to F. It satises
m := [X[ = [G : N
G
(F)] [G : F]
with the normalizer N
G
(F) F of F in G. We consider the homomorphism
(2) : G S(X)

= S
m
, g
g
with the permutation

g
: X X, H
g
(H).
Proposition 2.95. Let G be a group of order [G[ = 2
r
3, r 2. Then
G admits a non-trivial normal 2-subgroup (but not necessarily a normal 2-
Sylow-subgroup!).
Proof. We take as subgroup F G a 2-Sylow subgroup. If it is normal,
choose H := F. Otherwise the set X of its conjugates contains m = 3
elements. The image (G) of the homomorphism : G S(X)

= S
3
contains a 3-cycle, since it acts transitively on X, hence H := ker() is a
nontrivial (r 2) normal 2-subgroup of G.
Finally:
Proposition 2.96. A group G of order [G[ = 36 admits a non-trivial normal
3-group H (but not necessarily a normal 3-Sylow-subgroup!).
Proof. Take F G as a 3-Sylow subgroup. If it is normal, choose H := F.
Otherwise we have m = 4. Since [S
4
[ = 24, the kernel K := ker() contains
a 3-group. It has index [G : K] > 2, since a group of order 2 can not act
transitively on a set X with 4 elements. If [K[ = 9, it is itself a 9-Sylow
subgroup, hence, according to our assumption, not normal, a contradiction.
If [K[ = 3, we choose H := K, and if [K[ = 6, we have K

= C
2
C
3
or
K

= S
3
. In both cases the elements in K of order 3 constitute a 3-subgroup
H of K, invariant with respect to every automorphism of K (why?), in
particular normal in G.
67
Eventually we come to the case [G[ = 60:
Proposition 2.97. A simple group G of order [G[ = 60 is isomorphic to the
alternating group on 5 letters: G

= A
5
.
Proof. We show that the index [G : F] of any non-trivial proper subgroup
F G of a non-abelian simple group G is at least 5, and that in our case
[G[ = 60 there really is a subgroup F G of index [G : F] = 5.
Since the kernel of any homomorphism is a normal subgroup, our homo-
morphism
: G S(X)

= S
m
is injective (F being not normal and G being simple). Hence G

= (G) S
m
with m = [X[. But S
m
being solvable for m 4, we have 5 m [G : F].
Now let us look for a subgroup F of index [G : F] = 5: We consider a
2-Sylow subgroup H G. Since it is not normal in G, there is a conjugate
H

=
g
(H) ,= H. Consider the subgroup E G generated by H and
H

. It contains H H

as a normal subgroup (since either H H

= e
or H H

H, H

is normal as a subgroup of index 2) and has an index

[G : E] < [G : H] = 15, so either [G : E] = 5, 1 - the possibility [G : E] = 3
having already being excluded by the above argument applied to F = E. If
[G : E] = 5, we choose F := E. Otherwise we have E = G and it follows
that H H

= e as a proper normal subgroup of G. So we may assume

that H H

= e for dierent 2-Sylow subgroups H, H

.
Now consider the normalizer N
G
(H) H. It has the possible indices 5
or 15, (3 being again impossible) and thus we are done, if we can see that
15 is not possible either: In that case there are 15 pairwise dierent 2-Sylow
subgroups H
1
, ..., H
15
, with the H

i
= H
i
e being pairwise disjoint. So
there are 45 elements in G of order 2 or 4, on the other hand there are at
least 24 elements of order 5 (Take a 5-Sylow subgroup U G: Since it is not
normal, it has at least 6 pairwise dierent conjugates which (pairwise) only
have the neutral element in common), but 1 + 45 + 24 > 60: Contradiction!
Finally, identifying G with its image (G) S
5
, we show G = A
5
: If not,
A
5
G G is a proper normal subgroup of G, hence trivial: GA
5
= id.
But then G necessarily contains a 4-cycle f, implying id ,= f
2
G A
5
, a
contradiction! So A
5
G resp. A
5
= G because of [G[ = 60.
This nishes the proof of Theorem 2.91.
68
Problems 2.98. 1. R: Determine all Sylow-subgroups of S
3
, D
n
, S
4
, A
4
!
2. R: Show: A group G of order [G[ = 15 is cyclic.
3. Let G be a group of order [G[ = 12. Show: There is a normal p-Sylow-subgroup in
G (p = 2 or p = 3). Then G is isomorphic with a semidirect product. Classify now
all groups of order 12.
4. R: Show: If all Sylow-subgroups of a nite group G are normal, then G is isomorphic
with the direct product of its Sylow-subgroups. Hint: If H, F G are dierent
Sylow-subgroups, we have H F = e. Conclude ab = ba for all a H, b F
because of F (aba
1
)b
1
= a(ba
1
b
1
) H.
69
3 Rings
3.1 Denitions and Examples
Denition 3.1. A ring is a triple (R, , ), with a set R together with two
maps,
: R R R, (a, b) a + b := (a, b) ,
the addition, and
: R R R, (a, b) ab := (a, b) ,
the multiplication, such that
R
1
: The pair (R, ) is an (additively written) abelian group.
R
2
: The multiplication is associative:
(ab)c = a(bc), a, b, c R .
R
3
: The multiplication is distributive over the addition:
a(b + c) = ab + ac , (a + b)c = ac + bc , a, b, c R .
R
4
: There is an element 1 R 0, such that
1a = a = a1 , a R .
R
5
: The multiplication is commutative
ab = ba , a, b R .
Remark 3.2. 1. Usually only the conditions R
1
- R
3
are required for a
ring in the literature; if even R
4
, R
5
hold, it is called a commutative
ring with unity. Since we shall exclusively deal with commutative
rings with unity, we have chosen to follow the convention, that the
word ring should mean a commutative ring with unity.
70
2. The above axioms ensure that the arithmetic in a ring R is more
or less the familiar one: To be on the safe side let us mention the
following rules:
0 a = 0 = a 0
holds, since a = 1 a = (1 + 0)a = a + 0 a and
(1)a = a(1) = a
follows from 0 = (1 +(1))a = a +(1)a. But there is no cancellation
rule for the multiplication, since there may be nontrivial zero divisors,
i.e. elements a R 0, such that
ab = 0
for some b ,= 0, and it can happen that
1 + ... + 1 = 0 .
So we should actually derive all computation rules we use from the ring
axioms!
Example 3.3. 1. The sets R = Z, Q, R, C with the usual addition and
multiplication of integers resp. rational, real or complex numbers are
rings.
2. The factor groups Z
n
, cf. 2.26, constitute rings with their group law
as addition a + b = a + b and the multiplication a

b := ab. We have
to check that the multiplication is well dened, since then the axioms
R
1
R
5
carry over from Z to Z
n
. So let a
1
= a, b
1
= b, i.e. a
1
=
a + kn, b
1
= b + n with integers k, Z. Then we obtain a
1
b
1
=
(a + kn)(b + n) = ab + (a + bk + kn)n resp. a
1
b
1
= ab.
3. If R
1
, ..., R
n
are rings, their direct product
n

i=1
R
i
:= R
1
... R
n
is the cartesian product of the sets R
1
, ..., R
n
with the componentwise
ring operations. In particular for a ring R the n-fold cartesian product
R
n
is again a ring.
71
4. If M is any set and R a ring, so is the set
R
M
:= f : M R
of all R-valued maps on M with the argumentwise addition and multi-
plication of functions:
(f + g)(x) := f(x) + g(x), (fg)(x) := f(x)g(x).
5. Formal power series over a ring R : For a ring R we dene
R[[T]] := R
N
as additive group.
Hence, the elements in R[[T]] can be thought of as sequences (a

)
N
,
where a

R, N, by identifying a function f : N R with the

sequence (f())
N
, and the addition is componentwise. But we dene
a new multiplication, sometimes also called Cauchy multiplication,
on R[[T]]:
(a

) (b

) := (c

), where c

:=

k=0
a
k
b
k
=

k+=
a
k
b

.
Unity then is the sequence (1, 0, 0, ...), and, if we abbreviate T :=
(0, 1, 0, 0, ...), we nd
T
n
= (0, ..., 0
. .
n times
, 1, 0, 0, ...) .
So, if we identify an element a R with the sequence (a, 0, 0, ...), we
can write a sequence (a

), where almost all (i.e. with only nitely many

exceptions) elements a

= 0:
(a
0
, a
1
, ...., a
n
, 0, 0, ...) = a
n
T
n
+a
n1
T
n1
+... +a
1
T +a
0
=
n

=0
a

.
The ring R[[T]] is called the power series ring in one variable
over R, and the elements are usually, in analogy to the above equality
written as formal series

=0
a

corresponding to the sequences

(a

). But since we do not have the notion of an innite sum, this is a

priori nothing but a notational convention, and only in the nite case
it can be interpreted as a sum in a ring.
72
6. The polynomial ring R[T] in one variable over a ring R is the
subset:
R[T] :=
_
n

=0
a

; n N, a
0
, ..., a
n
R
_
R[[T]] ,
which is obviously closed with respect to the ring operations of R[[T]]
and itself a ring. If one wants to avoid the above abstract denition,
one can introduce polynomials over a ring R in a more naive way: We
dene them as nite formal sums
f =
n

=0
a

; n N, a
0
, ..., a
n
R
meaning that

if and only if a

= b

for all N.
The addition and multiplication are then as follows

(a

+ b

)T

,
(

)(

) =

k=0
a
k
b
k
)T

.
A polynomial f R[T] is called monic if it has the form f = T
n
+

<n
a

.
Denition 3.4. Let R ba a ring. The degree function
deg : R[T] N
is dened for f R[T] by
deg(f) :=
_
n , if f =

n
=0
a

, a
n
,= 0
, if f = 0
.
Remark 3.5. We have
deg(f + g) max(deg(f), deg(g)) , deg(fg) deg(f) + deg(g)
and
deg(fg) = deg(f) + deg(g)
if one of the polynomials f, g is monic.
73
The division algorithm for polynomials plays a central role in the arithmetics
of a polynomial ring:
Theorem 3.6 (Division algorithm for polynomials). Let g R[T] be a
monic polynomial. Then every polynomial f R[T] can be written as
f = qg + r
with uniquely determined polynomials q, r R[T], deg(r) < n = deg(g).
Proof. Uniqueness: Let f = qg + r = qg + r. Then we have:
(q q)g = ( r r) .
Now the polynomial on the right hand side has a degree < n, while for
q q ,= 0 the left hand side has at least degree n (since g is monic). Hence
q = q and then of course also r = r.
Existence: We do induction on deg(f): For deg(f) < n we take q = 0 and
r = f.
If deg(f) =: m n, say f = b
m
T
m
+ ... + b
0
, we consider the polynomial

f := f b
m
T
mn
g. Being of a degree < deg(f), the induction hypothesis
provides q, r with

f = qg + r
and deg( r) < n. Finally choose q := q + b
m
T
mn
, r := r.
Denition 3.7. Let R be a ring.
1. An element a R is called a nonzero divisor, i ab = 0 = b = 0.
If all elements in R 0 are nonzero divisors and 1 ,= 0, the ring R is
called an integral domain (integritetsomrade).
2. An element a R0 is called a unit i there is an element a
1
R,
such that aa
1
= 1(= a
1
a). We denote R

the set of all units in R:

R

:= a R; a
1
R : aa
1
= 1 ,
and call R

the group of units of the ring R.

3. A ring R is called a eld (kropp) i R

= R 0.
74
Note that R

is a (multiplicatively written) abelian group.

Example 3.8. 1. Units are are nonzero divisors, in particular elds are
integral domains: If a R

and ab = 0, we nd 0 = a
1
(ab) =
(a
1
a)b = 1b = b.
2. The rings Q, R, C are actually elds.
3. The ring Z of integers is an integral domain, but not a eld: We have
Z

= C
2
= 1.
4. In a nite ring R nonzero divisors are even units: For a nonzero divisor
a R the multiplication with a, the map
a
: R R, x ax, is
injective, hence also surjective, R being nite. Therefore there is an
element b R with ab = 1, i.e. a R

.
5. A residue class a Z
n
is a nonzero divisor i gcd(a, n) = 1. Hence
according to the previous point, we obtain the equality
2
:
Z

n
= a Z
n
; gcd(a, n) = 1 .
The function : N
>0
N dened as
(n) :=
_
1 , if n = 1
[Z

n
[ , if n 2
is called Eulers -function (Leonhard Euler, 1707-1783). For a prime
power n = p
k
we get: Zero divisors in Z
p
k are exactly the elements in
pZ
p
k, and thus [Z

p
k
[ = [Z
p
k[ [pZ
p
k[ = p
k
p
k1
= p
k1
(p 1). Hence
we have found
(p
k
) = p
k1
(p 1) for k 1.
6. The residue class ring Z
n
is a eld i it is an integral domain i n = p
is prime.
7. Let R be an integral domain. Then also the polynomial ring R[T] is
an integral domain with the same group of units as the original ring R,
i.e. R[T]

= R

, where we identify R with the constant polynomials,

i.e. of degree 0.
2
In Problem 2.62.6 we took its RHS as a denition for the LHS.
75
Proof. The ring R being an integral domain, we have
deg(fg) = deg(f) + deg(g) .
Since only the zero polynomial has degree , this implies that R[T]
is an integral domain. The inclusion R

R[T]

is obvious. On the
other hand the constant polynomial 1 has degree = 0, so the above
degree equality gives, that units have degree 0 and thus are units in R,
i.e. R[T]

.
8. In the same way as one obtains from Z the rationals one can associate
to any integral domain R a eld Q(R) containing R: More generally, a
subset S R 0 is called multiplicative, if 1 S and s, t S =
st S. On the cartesian product R S we dene an equivalence
relation as follows
(a, s) (b, t) :at = bs .
The set S
1
R := (R S)/ of its equivalence classes can be made
a ring: Denote
a
s
the equivalence class of the pair (a, s). Then the
addition and multiplication
a
s
+
b
t
:=
at + bs
st
,
a
s

b
t
:=
ab
st
provide well dened (check that!) ring operations; the resulting ring
S
1
R is called the localization of R with respect to the multi-
plicative subset S. If S = R 0, the localization S
1
R is actually
a eld, called the eld of fractions
Q(R) := (R 0)
1
R
of the integral domain R. The most important examples of this con-
struction are the rationals Q := Q(Z) and, with a eld K, the eld of
fractions
K(T) := Q(K[T])
of the polynomial ring K[T], also called the eld of rational func-
tions in one variable over K.
76
9. If R is an integral domain, so is the formal power series ring R[[T]] over
R, but in contrast to the polynomial ring R[T] the group of units is
quite big:
R[[T]]

= f =

=0
a

; a
0
R

.
Proof. We replace the degree function with the order function :
R[[T]] N dened as
(f) :=
_
n , if f =

=n
a

, a
n
,= 0
, if f = 0
.
So (f) is the order of f at 0. We have (fg) (f) + (g), and
even (fg) = (f) + (g), if R is an integral domain. So, since only
f = 0 has order , it follows, that there are no zero divisors in R[[T]]
if there are none in R.
In order to investigate the group of units R[[T]]

, we have to dene
certain innite sums of power series: Let (f
k
)
kN
R[[T]] be a sequence
of formal series, say f
k
=

=0
a
k

, such that lim

k
(f
k
) = .
Then we can dene

k=0
f
k
:=

=0
a

with a

k=0
a
k

,
where for each index only nitely many a
k

are ,= 0. So the sum

dening the coecient a

is in fact nite!
We show now that a series f =

=0
a

is a unit i the coecient

a
0
is a unit in R. The condition is necessary, since (a
0
+...)(b
0
+...) =
(a
0
b
0
+ ....) = 1 implies a
0
b
0
= 1. On the other hand, if a
0
R

it of
course suces to consider the series a
1
0
f, i.e. we may assume a
0
= 1
and write then f = 1g, where the series g has order (g) 1. Then it
follows (g
k
) k, such that the geometric series h :=

k=0
g
k
denes
a formal series, satisfying h(1 g) = 1.
Problems 3.9. 1. R: An element x R in a ring is called nilpotent if there is a
natural number n N with x
n
= 0. The set n =
_
0 of all nilpotent element is
called the nilradical of R. Show: 1 + n R

. Hint: Geometric series!

2. R: Let K be a eld. Show: R := K + T
2
K[T] := f = 1 + T
2
g ; g K[T] with
the from K[T] induced ring operations is an integral domain with Q(R) = K(T).
77
3. R: For a subset P
0
P of the set P N of all primes denote S(P
0
) the multiplicative
subset consisting of 1 and all natural numbers, which are a product of primes in
P
0
. Show: All rings R Q (endowed with the induced ring operations) have the
form R = S(P
0
)
1
Z with a suitable subset P
0
P.
4. For a ring R we dene its ane linear group A(R) S(R) by A(R) := f
S(R); a R

, b R : f(x) = ax + b x R. Show: A(R) is a semidirect

product A(R)

= R
+

with some homomorphism : R

Aut(R
+
), where
R
+
denotes the ring R considered as additive group.
5. Let d N be not a square. Show that R := Z + Z

d R is a ring. Furthermore
that the norm N : R Z, x = a + b

d N(x) := a
2
b
2
d satises N(xy) =
N(x)N(y) and conclude x R

N(x) = 1. Assuming that there is x R Z

with N(x) = 1 (that is always true, but non-trivial!) show R

= ZZ
2
as groups.
Hint: The units x > 1 are of the form x = a + b

d with a, b > 0. Conclude that

there is a smallest unit x R

, x > 1.
6. R: Let R C be a ring with z R = [z[
2
N, invariant under complex
conjugation, i.e. z R =z R. Determine R

.
7. R: Let C R be a complex number with
2
Z+Z := a +b; a, b Z C.
Show, that Z+Z endowed with the addition and multiplication of complex numbers
is a ring, satisfying the conditions of the ring R C in the previous problem. (Hint:
If
2
= a + b, then (T )(T ) = T
2
aT b.) Show that R

is nite! For
= i, = := e
2i
3
=
1
2
(1 + i

3) determine the group of units explicitly! What

does hold for the remaining rings R = Z +Z?
8. Let K be a eld. Interpret K
N
as a subset of K
Z
by extending a function N K,
such that it assigns the value 0 to negative numbers in Z N. The set K((T)) of
all formal Laurent series with nite principal part consists of all functions
K
Z
which vanish for almost all n < 0. Show that the Cauchy multiplication for
K[[T]] = K
N
can be extended to K((T)) K
Z
and that K((T)) is a eld, the eld
of fractions of K[[T]]. Indeed
K((T)) = Q(K[[T]]) = T
N
K[[T]] = K[[T]]

n=1
KT
n
,
where T
N
K[[T]] := S
1
K[[T]] denotes the localization of K[[T]] with respect to
the multiplicative set S = T
N
of all T-powers.
9. Formal Laurent Series from a topological point of view: Let K be a eld. We
extend the order function : K[[T]] Z to K((T)) by setting (T
n
h) =
(h)n for g K[[T]] and dene the absolute value [f[ R
0
of a series f K((T))
by [f[ := 2
(f)
(with the convention 2

= 0). Show: The absolute value satises

[f +g[ max[f[, [g[ , [fg[ = [f[ [g[
for all f, g K((T)) (The rst inequality is sometimes called the strong triangle
inequality). Then d(f, g) := [f g[ is a metric (distance function) on the set K((T))
78
and the resulting metric space is complete, i.e. every Cauchy sequence has a limit.
Indeed K[[T]] = f K((T)); [f[ 1 then is nothing but the closed ball of radius
1 around 0, and any series

=
a

= lim
n

n
=
a

is the limit of its

partial sums - but note that this convergence is not a convergence of functions! In
K there is no notion of convergence!
3.2 Homomorphisms
Denition 3.10. Let R, S be rings. A map : R S is called a ring
homomorphism i
(a + b) = (a) + (b) , (ab) = (a)(b) ,
for all a, b R and
(1) = 1 ,
where 1 denotes the unity in the respective ring R or S. It is called a ring
isomorphism i it is in addition bijective, and R is isomorphic to S, in
symbols: R

= S, i there is a ring isomorphism : R S.

Remark 3.11. The condition (1) = 1 guarantees that 0 is not
admitted as a ring homomorphism. As an other consequence, the map
R R
2
, a (a, 0), is not a ring homomorphism either, though it is
compatible with the ring operations.
Example 3.12. For every ring R there is exactly one ring homomorphism
: Z R, mapping 0 Z to 0 R, n Z
>0
to the n-fold sum 1 + ... + 1
and n Z
<0
to (1) + ... + (1) (n times). Indeed, this is nothing but
the group homomorphism
a
: Z G of Example 2.9.3, with the additive
group R instead of a multiplicatively written G, and a = 1. Often one writes
simply n instead of (n); but note that need not be injective: n = 0 can
hold in R, though n > 0 i Z. For example take R = Z
n
!
Denition 3.13. Let R be a ring and : Z R the natural ring homo-
morphism.
1. The characteristic char(R) of the ring R is dened as the natural
number n such that
ker() =
1
(0) = Zn .
With other words: Either char(R) = 0 or
char(R) = mink N
>0
; k = 0 in R > 1.
79
2. The prime eld P(K) K of a eld K is dened as
P(K) := (Z),
if char(K) > 0 and as
P(K) := (Q),
if char(K) = 0 and : Q = Q(Z) K is the unique extension of the
natural ring homomorphism : Z K.
Remark 3.14. 1. char(Z
n
) = n.
2. For an integral domain R, its characteristic satises char(R) = 0 or
char(R) = p is a prime. This is a consequence of the fact, that
Z
char(R)

= (Z) R is an integral domain as well.
3. For a eld K of characteristic 0 we have P(K)

= Q, while for char(K) =

p > 0 we nd P(K)

= Z
p
.
4. Let n = p
k
1
1
... p
k
r
r
be the prime factorization of the natural number
n N. The Chinese remainder theorem 2.85 provides even a ring
isomorphism
Z
n

=
Z
p
k
1
1
... Z
p
k
r
r
, = +Zn ( +Zp
k
1
1
, ..., +Zp
k
r
r
) ,
hence in particular a group isomorphism of the corresponding groups
of units
Z

=
Z

p
k
1
1
... Z

p
k
r
r
.
As a consequence, Eulers -function satises
(n) = (p
k
1
1
) ... (p
k
r
r
) .
The polynomial ring is, similarly as for example free groups, characterized
by a universal mapping property:
Proposition 3.15. Let : R S be a ring homomorphism and a S.
Then there is a unique ring homomorphism
a
: R[T] S with
a
[
R
=
and
a
(T) = a.
Proof. For f =

set
a
(f) =

(a

)a

. Since f = 0 i all coe-

cients a

= 0, the homomorphism
a
is well dened and obviously unique.
80
Remark 3.16. If R S and : R S is the inclusion, we also write f(a)
instead of
a
(f), i.e.
f(a) =

R , if f =

,
and call the ring homomorphism
a
: R[T] R, f f(a), the evaluation
(homomorphism) at a S.
Thus a polynomial f R[T] induces a function

f : R R , a f(a) ,
and the map R[T] R
R
, f

f is a not necessarily injective ring ho-
momorphism: For example consider R = Z
p
and f = T
p
T: Obviously
f(0) = 0, while all elements a Z

p
have order p 1, whence a
p
= a resp.
f(a) = 0. Indeed this can happen only for nite integral domains:
Calling an element a R with f(a) = 0 a zero of f, we have:
Proposition 3.17. A polynomial f R[T] 0 over an integral domain R
has at most deg(f) distinct zeros in R.
Proof. We do induction on deg(f). We may assume that deg(f) > 0. If
a R is a zero of f, the division algorithm 3.6 yields f = q (T a) + r,
where deg(r) < 1, i.e. r = b R. Hence 0 = f(a) = q(a)(a a) + b = b
resp. f = q (T a). Since deg(q) < deg(f), the induction hypothesis tells
us, that q has at most deg(q) distinct zeros in R. But R being an integral
domain, a zero of f is either a zero of q or of T a, i.e. equals a. Hence we
are done.
Corollary 3.18. For an innite integral domain R the homomorphism R[T]
R
R
, f

f is injective, i.e. for a polynomial f R[T] we have
f = 0 f(a) = 0 , a R .
Problems 3.19. 1. R: An element e R in a ring R is called idempotent i e
2
= e.
Show: If R is an integral domain, then 0, 1 are the only idempotent elements.
Furthermore: If 1 = e
1
+ ... + e
s
with elements e
i
,= 0 and e
i
e
j
= 0 for i ,= j, the
elements e
i
are idempotent and
R

=
s

i=1
R
i
with the rings R
i
:= Re
i
.
81
2. R: Show: There is a (unique) ring homomorphism Z
n
Z
m
, i m[n.
3. R: Show for q := p
n
with a prime number p that Z
q
[T]

= Z

q
+pTZ
q
[T].
4. R: A universal mapping property: Let R be an integral domain and S R 0 a
multiplicative subset. Show: Every ring homomorphism : R P to a ring P
with (S) P

can uniquely be extended to a homomorphism

: S
1
R P.
5. R: Let H R

be a nite subgroup of the group of units R

of an integral domain
R. Show: H is a cyclic group. Hint: Show rst: If all elements a H have an order
< [H[, there is an exponent q < [H[ with a
q
= 1 for all a H. Then consider the
zeros of the polynomial T
q
1 R[T]! Show as well H = C
n
(R) := a R; a
n
= 1
with n := [H[.
6. R: Every ring homomorphism : R[T] R[T] with [
R
= id
R
has the form
=
g
with a polynomial g R[T], i.e. it is a substitution homomorphism, where
(f) is obtained by substituting T with g, i.e.
g
(f) = f(g). Furthermore: If R is
an integral domain:
g
is an isomorphism (or automorphism) i g = aT + b with
a R

, b R. Determine an isomorphism A(R)

=
Aut
R
(R[T]) between the
ane linear group A(R), and the group Aut
R
(R[T]) of all automorphisms of R[T]
xing the elements in the ring R.
7. A continuation of the previous problem: Show: Every ring homomorphism
g
:
R[T] R[[T]] with a power series g TR[[T]] extends uniquely to a ring ho-
momorphism

g
: R[[T]] R[[T]] (such that we may dene substitutions even
for formal power series: f(g) :=

g
(f) in case g TR[[T]]). It is an isomor-
phism i g R

T + R[[T]]T
2
. Hint: An equality in R[[T]] holds i it does in
R[[T]]/(T
n
)

= R[T]/(T
n
) for all n N.
8. R: Let K be a eld, a
1
, ..., a
s
, b
1
, ..., b
s
K, with the elements a
1
, ..., a
s
pairwise
distinct. Show: There is a polynomial f K[T] with f(a
i
) = b
i
for i = 1, ..., s.
9. Let K be a eld and A K
n,n
. For f =

K[T] let f(A) := a

0
E +

>0
a

K
n,n
. Show:
K[A] := B K
n,n
; B = f(A) for some f K[T]
is a ring, in particular commutative. Furthermore: If A is diagonalizable, then
K[A]

= K
r
with some r N.
3.3 Ideals and Factor Rings
Let K be a eld and f K[T] a polynomial without zeros in K. In this
section we explain how we can nd a larger eld E K, where f has a zero.
To begin with, let us rst assume that E K together with a zero a E
of f are already given. Then the evaluation homomorphism

a
: K[T] E, p p(a),
82
induces an isomorphism
K[T]/ ker
a

=
K[a] := p(a) E; p K[T]
of abelian groups. Indeed it is an isomorphism of rings as well: First of all,
given any ring homomorphism
: R S
its kernel ker R is
1. an additive subgroup
ker R
and
2. satises
R ker ker .
Thus we are led to the notion of an ideal of a ring R.
Denition 3.20. Let R be a ring. An ideal a R is an additive subgroup
satisfying
a a, b R =ab a.
An ideal a R is called a proper ideal i a ,= R or equivalently, i 1 , a.
Proposition 3.21. Let a R be a proper ideal. Then the (additive) factor
group R/a endowed with the multiplication
a b := ab
is a ring.
Proof. We have to show that the multiplication is well dened. So let a
1
=
a, b
1
= b, i.e. a
1
= a + c, b
1
= b + d with elements c, d a. Then we have
a
1
b
1
= (a + c)(b + d) = ab + (ad + bc + cd), where the expression in the
parenthesis belongs to a. Hence a
1
b
1
= ab.
Example 3.22. 1. The most basic ideals are the entire ring R itself, also
called the unit ideal, and the zero ideal 0.
83
2. Given an element a R the set (a) := Ra := ba, b R of all
multiples of a is an ideal, the principal ideal generated by a.
3. A ring R is a eld i the unit and the zero ideal are the only ideals in
R.
Proof. =: If R is a eld and a R a non-zero ideal, there is an
element a a, a ,= 0. But then it follows 1 = a
1
a a and thus a = R.
=: We have to show R

= R 0 or rather : Let a R 0.
Then Ra ,= 0, hence Ra = R. So there is b R with 1 = ba, i.e.
a R

.
4. In the ring Z additive subgroups and ideals coincide; indeed in 2.25
we have seen that they are all principal ideals a = Zn with a (unique)
n N.
5. Let a, b R be ideals. Then also their intersection a b, their sum
a + b := a + b; a a, b b
as well as their product
a b := a
1
b
1
+ ... + a
r
b
r
; r N, a
i
a, b
i
b, i = 1, ..., r
are ideals. In fact
a b a b a, b a + b.
6. The elements a
1
, ..., a
r
R are said to generate the ideal a R i
a = Ra
1
+ ... + Ra
r
.
7. Let : R S be a ring homomorphism between the rings R and S,
a R, b S ideals. Then the inverse image
1
(b) R is an ideal in
R, while the image (a) S is an ideal in the ring S, if is surjective.
In particular its kernel
ker() =
1
(0)
is an ideal in R.
84
8. A ring homomorphism : K S from a eld to some ring S ,= 0
is injective (since 1 , ker() we have ker() = 0) and therefore often
treated as an inclusion.
Let us return to our starting point: The ring
K[a]

= K[T]/a
with a = ker
a
is an integral domain. In general we dene:
Denition 3.23. A proper ideal a R of a ring R is called
1. prime or a prime ideal if the factor ring R/a is an integral domain
or equivalently if
ab a =a a b a .
2. maximal i there is no proper ideal b R in R containing a but a
itself:
a b =b = a.
3. a principal ideal, if it has the form a = Ra with some element a R,
it is then called the principal ideal generated by a. A dierent notation
is (a) := Ra.
An integral domain R is called a principal ideal domain, a PID for short,
if every ideal in R is a principal ideal.
Example 3.24. 1. Since the ideals in Z are nothing but the additive
subgroups, Proposition 2.25 yields that Z is a principal ideal domain;
the maximal ideals are the ideals (p) = Zp with a prime number p N
and beside the maximal ideals there is only one prime ideal, the zero
ideal (0) = Z0.
2. If p R is a prime ideal in the integral domain R, its complement
S := R p is a multiplicative set. The ring R
p
:= S
1
R is called the
localization of R with respect to the prime ideal p.
Remark 3.25. One can show that every proper ideal a R is contained in
a maximal ideal m R, cf. 5.7.
85
As with prime ideals, maximal ideals can be dened in terms of the cor-
responding factor ring:
Proposition 3.26. A proper ideal a R is a maximal ideal i the factor
ring R/a is a eld. In particular a maximal ideal is a prime ideal.
Proof. Use Example 3.22.3 together with the fact that given an ideal a R
in the ring R, there is a bijective correspondence between the set of all ideals
in the factor ring R/a and the set o all ideals in R, which contain a: To
an ideal b a we associate the ideal (b) R/a, where : R R/a
denotes the quotient projection. On the other hand to c R/a corresponds
the inverse image
1
(c) R.
Back to our original situation: Since
a
(f) = f(a) = 0, the prime ideal
a = ker
a
K[T] contains f. But, as we shall see later on, a nontrivial
prime ideal of K[T] is maximal. Consequently the minimal solutions of our
problem are given by
E := K[T]/m, a := T,
where m K[T] is a maximal ideal containing f (we remark that the com-
position K K[T] E is injective according to 3.22.8). By this we mean
that an arbitrary solution contains one of the above type.
On the other hand as a consequence of the next result, nontrivial ideals
in K[T] are in one-to-one correspondence with monic polynomials:
Proposition 3.27. The polynomial ring K[T] over a eld K is a principal
ideal domain.
Proof. Let a K[T] be an ideal. If a ,= 0, we can choose a polynomial
f a 0 of minimal degree, and since K is a eld, we may assume that
f is monic. Then we have a = (f). The inclusion is obvious. :
Consider a polynomial h a. We apply the division algorithm 3.6 and write
h = qf + r, where deg(r) < deg(f). But then, since even r = h qf a,
necessarily r = 0 by the choice of f. So h = qf (f).
Example 3.28. For the ring K[[T]] of formal power series over a eld K the
situation is strikingly dierent from that one for the polynomial ring: There
are only the following ideals: (T
n
), n N and the zero ideal; (T) is the only
maximal ideal, and beside the zero ideal the only prime ideal. In particular
K[[T]] is a principal ideal domain. The proof relies on the fact that its group
of units K[[T]]

consists of all series with a nonzero constant term.

86
In particular we obtain that a maximal ideal m f is of the form
m = K[T]g
with an irreducible polynomial g K[T] dividing f (a polynomial is irre-
ducible if it does not admit a factorization as a product of polynomials of
lower degree): For a reducible polynomial g the factor ring K[T]/m would
not be an integral domain.
We discuss the relevant notions in general for an arbitrary integral do-
main:
Denition 3.29. Let R be an integral domain.
1. An element u R(R

0) is called irreducible, i u = ab =a
R

b R

, i.e. u can not be written as the product of two non-units.

2. Two elements u, u

R 0 are called associated, i (u) = (u

) i
u

= eu with a unit e R

.
3. An element p R (R

0) is called prime, i (p) = Rp is a prime

ideal i
p[ab =p[a p[b .
Obviously, (u) = (u

) holds for associated elements u, u

R. If on the
other hand (u) = (u

), we can write u

= eu and u = e

with elements
e, e

R, whence u = e

eu resp. 0 = (1 e

e)u. The ring R being an integral

domain, we conclude 1 e

e = 0 resp. e R

.
Remark 3.30. 1. A prime element p R is irreducible. Show that!
2. Since K[T]

= K

, two polynomials f, g K[T] are associated i their

dier by a nonzero constant: g = f with some K

.
3. A polynomial f K[T] is irreducible, if it can not be written f = gh
with polynomials g, h of lower degree.
Proposition 3.31. An irreducible element u R in a principal ideal domain
R is prime.
87
Proof. We assume u[ab, and show that if u does not divide a, then it divides
b. Consider the ideal Ru +Ra := ru +sa; r, s R, consisting of all linear
combinations of u and a with coecients in R. Since R is a principal ideal
domain, there is an element d R with Ru + Ra = Rd. In particular d[u,
say u = cd. But u being irreducible either c or d is a unit. If c R

, we
obtain, since d[a, that also u[a, a contradiction to our assumption. So d is
a unit, and therefore 1 Rd = Ru + Ra: As a consequence we may write
1 = ru + sa with elements r, s R. Now we multiply with b and nd that
b = rub + s(ab) is divisible by u.
As an immediate consequence we obtain:
Corollary 3.32. i) Let R be a principal ideal domain which is not a eld.
An ideal (a) := Ra is
prime, i a = 0 or a is irreducible.
maximal, i a is irreducible.
ii) Let K be a eld and f K[T] an irreducible polynomial. Then
K[T]/(f) is a eld.
Proof. The zero ideal is prime, since R is an integral domain, but not max-
imal, R

= R/0 being not a eld. On the other hand, let (a) R be a
prime ideal, a ,= 0. Then a is irreducible: Assume a = bc with non-units b, c.
So one of the factors, say b, is contained in (a), and thus (b) (a) (b), i.e.
b is associated to a respectively c R

.
It remains to show, that (a) is maximal, if a is irreducible: Consider an ideal
b containing (a). Since R is a principal ideal domain, it is of the form b = (b),
hence a (b) or a = bc with some element c R. But a being irreducible,
either b or c is a unit, with other words either (b) = R or (b) = (a).
The second part follows now from the fact that K[T] is a principal ideal
domain and the rst part.
Irreducible polynomials being prime, we obtain a factorization of a poly-
nomial analogous to the prime factorization of a natural number:
88
Proposition 3.33. Every monic polynomial f K[T] can be written uniquely
(up to order) as a product
f = f
k
1
1
... f
k
r
r
with pairwise distinct irreducible monic polynomials f
i
K[T] and exponents
k
i
1.
In particular, for every polynomial f K[T] there exists a eld E containing
K, such that f has a zero in E, namely E := E
i
:= K[T]/(f
i
) with some
i, 1 i r.
Proof. The existence of such a factorization follows by induction on deg(f):
If f is irreducible, nothing has to be shown; if not, we write f = gh with
polynomials of strictly lower degree, which according to the induction hy-
pothesis are products of irreducible polynomials, hence f itself as well. The
uniqueness follows from the fact that irreducible polynomials are prime: We
use induction on the number k := k
1
+ ... + k
r
of factors in such a represen-
tation. For k = 1 the statement is clear, since then f = f
1
is irreducible and
does not admit a nontrivial factorization. Assume now there is an other fac-
torization f = g
1
... g

with irreducible monic polynomials g

j
K[T]. Since
f
1
is prime, we nd that for some index j we have f
1
[g
j
or rather f
1
= g
j
,
the polynomial g
j
being irreducible and both f
1
, g
j
monic. We may assume
j = 1 and then obtain f
k
1
1
1
f
k
2
2
.... f
k
r
r
= g
2
... g

and can apply the

induction hypothesis.
Integral domains admitting unique prime factorization get a name:
Denition 3.34. An integral domain R is called factorial (or a UFD=
Unique Factorization Domain), i
1. every irreducible element u R is prime,
2. every non-unit is a nite product of irreducible elements.
Example 3.35. The ring Z as well as the ring K[T] are UFDs.
For a factorial ring an analogue of Prop. 3.33 holds: Every non-unit can
be written as a product of irreducible elements, and the factors are unique
up to order and multiplication with units.
Here are more factorial rings:
89
Proposition 3.36. A PID is a UFD.
Proof. Call a non-unit u R 0 nonfactorizable, if it is not a product
of (nitely many) irreducible elements. Because of Prop. 3.31 it suces to
show that there are no nonfactorizable elements in a PID. Otherwise we can
construct a strictly increasing sequence of principal ideals a
i
= Ru
i
generated
by nonfactorizable elements u
i
R, i.e. such that
a
0
a
1
...... .
Start with any nonfactorizable element u
0
R. The nonfactorizable element
u
i
being found, write it as a product of two non-units. At least one of the
two factors is again nonfactorizable, choose it as the element u
i+1
. Now
a :=

_
i=0
a
i
is an ideal. But R is a PID; thus a = Ru. Then u a
n
for some n N and
hence
a
i
= a = a
n
holds for i n, a contradiction.
Remark 3.37. 1. An example of an integral domain with nonfactorizable
elements is discussed in Problem 3.38.15.
2. The polynomial ring over a UFD is again a UFD, see Problem 3.46.8.
On the other hand: The polynomial ring over a PID, which is not a
eld, is never a PID.
3.3.1 Digression: p-adic number elds
Let p be a prime number. The ring

Z
p
of all p-adic integers is dened as a
subring

Z
p

n=1
Z
p
n
of the direct product of the residue class rings Z
p
n, n N
>0
, namely

Z
p
:=
_
(
n
)
n1

n=1
Z
p
n; n 2 :
n
(
n
) =
n1
_
,
90
where
n
: Z
p
n Z
p
n1 is the natural ring homomorphism. Since the
sequence (
n
:= 1+(p
n
))
n1
has innite order, it has characteristic char(

Z
p
) =
0, in particular Z

Z
p
, and its group of units is

p
=
_
e = (
n
)

Z
p
;
1
,= 0
_
.
Indeed any x = (
n
)

Z
p
0 can uniquely be written as a product
x = p
r
e, r N
0
, e

Z

p
.
This is seen as follows: If
n
= 0 for n r and
r+1
,= 0, then, writing

n
= a
n
+ (p
n
), we have a
n
= p
r
b
n
for n r, and may take
n
= b
n+r
+ (p
n
).
As a consequence

Z
p
is a PID with the (p
r
), r N, as the nonzero ideals.
The ring

Z
p
Z can be understood as a completion of Z: For a p-adic
integer x

Z
p
its absolute value [x[ R
0
is dened as
[x[ :=
_
p
r
, if x p
r

p
0 , if x = 0
.
The absolute value satises
[x + y[ max[x[, [y[ , [xy[ = [x[ [y[ , [x[ 1
for all x, y

Z
p
(The rst inequality is sometimes called the strong triangle
inequality). Then d(x, y) := [xy[ is a metric (distance function) on the set

Z
p
and the resulting metric space is complete, i.e. every Cauchy sequence
has a limit, with Z

Z
p
as dense subset: For x = (
n
= a
n
+ (p
n
)), we have
x = lim
n
a
n
. Assuming 0 a
n
< p
n
, the coecients c
k
of the nite p-adic
expansions a
n
=

k<n
c
k
p
k
, 0 c
k
< p, do not depend on n and provide an
innite unique p-adic expansion
x =

k=0
c
k
p
k
, 0 c
k
< p.
This suggests that there should be a relation between p-adic integers and
formal power series with integer coecients: Since [p[ < 1 and [c[ 1 for all
c

Z
p
, the series

c
k
p
k
converges for any choice of the coecients c
k
Z;
in particular we may dene an evaluation homomorphism
Z[[T]]

Z
p
, f =

k=0
c
k
T
k
f(p) :=

k=0
c
k
p
k
;
91
it is surjective and has kernel (T p) Z[[T]]: If f(p) = 0, necessarily
c
0
= pb
0
with some b
0
Z, since otherwise [c
0
[ = 1 and [f(p)[ = [c
0
[ = 1
(using [x + y[ = max[x[, [y[ for [x[ ,= [y[). Now replace f
0
:= f with
f
1
= T
1
(f (T p)b
0
) and repeat the same argument to nd b
1
etc.; the
resulting series g =

b
k
T
k
satises f = (T p)g. Altogether we have found
an alternative description of the ring of p-adic integers:

Z
p

= Z[[T]]/(T p) .
The eld of fractions Q
p
:= Q(

Z
p
) can be realized as
Q
p
= p
N

Z
p
with the multiplicative subset S = p
N
(writing (p
N
)
1
= p
N
), and the abso-
lute value extends in an obvious way, indeed
Q
p
= 0

_
r=
p
r

p
,
where p
r

p
is the sphere of radius p
r
and center 0. Furthermore the
equality

p
Q =
_
a
b
; a, b Z, gcd(a, p) = 1 = gcd(b, p)
_
explains how to compute [x[ for x Q Q
p
.
The elements in Q
p
are called p-adic numbers and the eld Q
p
the
p-adic number eld, introduced by Kurt Hensel (1861-1941).
Since Q is dense in Q
p
, the p-adic numbers can, as the reals, be thought
of as a completion of the rationals, but note that Z R is discrete:
R Z = Z,
and unbounded, while Z Q
p
is bounded, indeed its closure Z Q
p
is the
closed unit ball:
Q
p
Z = x Q
p
; [x[ 1 =

Z
p
.
There are more strange features from the topological point of view: The
strong triangle inequality implies, that the open balls
B

(0) := x

Z
p
; [x[ < , 1,
92
form ideals in

Z
p
; in particular two balls B

(x) = x+B

(0) and B

(y) = y +
B

(0) being ideal residue classes either coincide or are disjoint. Hence a ball
B

(x) is both open and closed its complement is the union of all B

(y), y ,
B

(x). As a consequence the metric space

Z
p
is totally disconnected, i.e.
there are no non-empty connected open sets in

Z
p
. Furthermore the natural
order relation on Z can not be extended continuously to the ring of p-adic
integers: E.g. the negative number 1 p is a unit with the multiplicative
inverse
(1 p)
1
=

n=0
p
n
,
an innite sum of positive numbers!
Problems 3.38. 1. R: Show that a ring R has exactly one maximal ideal i the non-
units in R, i.e. the set R R

, provide an ideal. In that case R is called a local

ring. Which local rings do you know?
2. R: The nilradical of a ring R is dened as
n :=
_
0 := x R; n N : x
n
= 0 .
Show: The nilradical is an ideal and the factor ring R/n is reduced, i.e., does not
contain non-zero nilpotent elements, or equivalently, its nilradical is the zero ideal.
3. R: Let K be a eld and R := f K[T]; f = a
0
+ T
2
g, g K[T], cf. Problem
3.9.2. Show: The elements T
2
, T
3
are irreducible in R, but not prime.
4. R: A ring is called euclidean if it is an integral domain and there is a function
R 0 N, x [[x[[ satisfying:
(a) If b[a, then [[b[[ [[a[[
(b) If a, b R, b ,= 0, we can write a = qb + r, where the remainder r satises
either r = 0 or [[r[[ < [[b[[.
Show: A euclidean ring is a principal ideal domain.
(In fact, the rst condition is not needed in the proof, we have only added it
following the tradition!)
5. R: Show that a subring R C is euclidean if z R = [z[
2
N and for every
w C there is an element z R with [z w[ < 1. Hint: In order to check the
division algorithm for a, b R, regard w :=
a
b
C!
6. R: Show that the ring Z[i] is euclidean and hence a principal ideal domain! (Its
elements are called gaussian integers.) Determine the units in Z[i]! Is the number
5 Z[i] irreducible? If not, factorize it!
93
7. R: Let := e
2i
3
. Regard the rings Z[] Z[i

3]. Show: In the smaller ring the

elements 2, 2, 2
2
are non-associated irreducible elements and 2 2 = 4 = 2 2
2
,
while the bigger ring Z[] is even euclidean.
8. R: Let K be a eld and h
1
, ..., h
r
K[T] polynomials without a common divisor.
Show: There are polynomials g
1
, ..., g
r
K[T] with g
1
h
1
+... +g
r
h
r
= 1.
9. Some linear algebra: Let K be a eld, A K
n,n
and f K[T] the minimal
polynomial of A. Let f = f
k
1
1
... f
k
r
r
be the factorization of f in irreducible monic
polynomials f
1
, ..., f
r
. Set
h
i
:= f
k
1
1
... f
k
i1
i1
f
k
i+1
i+1
... f
k
r
r
and choose g
i
K[T] as in the previous problem with g
1
h
1
+... +g
r
h
r
= 1. Show:
The matrices P
i
:= g
i
(A)h
i
(A) satisfy E = P
1
+ ... + P
r
with the unit matrix
E K
n,n
and P
i
P
j
=
ij
P
i
, and
K[A]

=
r

i=1
K[A
i
]
with A
i
:= AP
i
. Furthermore that A
i
has the minimal polynomial f
k
i
i
. And that the
vector space V := K
n
is the direct sum of the A-invariant subspaces V
i
:= Im(P
i
)
(Hint: AP
i
= P
i
A and Im(P
i
) is the eigenspace of P
i
for the eigenvalue 1!). What
does the polynomial f
i
look like for K = C? Show: C[A
i
]

= C[T]/[T
k
i
]. Is there a
relationship to the Jordan normal form? (Camille Jordan, 1838-1922)
10. If one replaces in the denition of the p-adic integers the rings Z
p
n with R[T]/(T
n
),
where R is an arbitrary ring, what does one obtain?
11. Show that the following conditions for a ring R are equivalent:
(a) Every ideal a R is nitely generated, i.e., there are elements f
1
, ..., f
r
R,
such that
a = Rf
1
+... +Rf
r
=
r

i=1
g
i
f
i
; g
1
, ..., g
r
R
(b) Every increasing sequence (chain) of ideals a
1
a
2
... in the ring R becomes
stationary, i.e. there is n N such that a
m
= a
n
m n.
(c) Every subset A Ideal(R) of the set Ideal(R) of all ideals in R, has a maximal
element b A, i.e. such that there is no ideal in A containing b as a proper
subset, i.e., a A : b a =b = a.
A ring R is called noetherian (Emmy Noether, 1882-1935) if one (and thus all) of
the above conditions are satised.
12. Show Hilberts Basissatz (David Hilbert, 1862-1943): The polynomial ring R[T]
over a noetherian ring is again noetherian. Hint: For an ideal b R[T] regard the
chain a
n
:= a R; f = aT
n
+... b R of ideals in the ring R.
94
13. Show: In a noetherian integral domain every element can be written as a product of
(nitely many) irreducible elements. Hint: Assuming the contrary construct with
the previous problem a strictly increasing innite chain of ideals.
14. Show: The ideal a C
N
consisting of all sequences (a

)
N
with only nitely many
a

,= 0 is not nitely generated.

15. Let O(C) be the ring of all entire functions (i.e. holomorphic everywhere in the
complex plane C) and
a := f O(C); n(f) N : n N, n n(f) : f(n) = 0.
Show, that the ideal a is not nitely generated. (Do you see a connection with the
situation of the preceding theorem?) Show: An irreducible function is prime, but
not all functions can be written as a product of prime functions. (Instead there is
an innite factorization according to Weierstra theorem! (Karl Theodor Wilhelm
Weierstra, 1815-1897).
16. Let C(R) be the ring of all continuous real valued functions on R. Show: The ideal
m
0
:= f C(R); f(0) = 0 is not nitely generated.
17. Let C

(R) be the ring of all innitely often dierentiable real valued functions on
R. Show: p := f C

(R); f
(n)
(0) = 0 n N is a prime ideal. Here f
(n)
denotes the n-th derivative of the function f. (Indeed C

(R)/p

= C[[T]].)
3.4 Irreducibility Criteria
If a polynomial f K[T] of degree deg(f) > 1 is irreducible, it has no zero
a K: Otherwise we could factorize f = (T a)g with some polynomial
g K[T] K. On the other hand a polynomial f K[T] of degree deg(f)
3 without a zero in K is also irreducible: If we can write f = gh with
polynomials g, h of lower degree, one factor is linear, and thus provides a
zero of f.
But how to check whether f has zeros or not? If K is nite, then, at least
theoretically, we could simply check by computing all possible values. On
the other hand, if K = Q and f Z[T] is monic, every rational zero a Q
already is an integer: a Z (Show that or cf. Corollary 3.41.2), dividing
a
0
= f(0) (using the factorization f = (T a)g with g Z[T]). So there
are only nitely many candidates for possible zeros, if a
0
,= 0 - and that can
always be assumed. But what does hold for deg(f) > 3? Again there is at
least theoretically no problem, if the eld K is nite, since then there are
only nitely many candidates for the polynomials g, h.
If K = Q, we may assume that f Q[T] even has integer coecients:
f Z[T] if not, multiply f with some natural number. Indeed, we shall see,
95
that if f Z[T] is not irreducible in Q[T], then there is even a factorization
in Z[T]. Eventually, in order to exclude that possibility, we pass to the
polynomial ring Z
m
[T] over some factor ring Z
m
: Fix a natural number
m N
>1
and consider the following ring homomorphism
Z[T] Z
m
[T], f =
n

=0
a

f :=
n

=0
a

.
The polynomial

f Z
m
[T] then is called the reduction of f mod m.
First we show that a factorization f = gh Z[T] with polynomials g, h
Q[T] of lower degree always can be realized with polynomials g, h Z[T].
Denition 3.39. The content cont(f) Q
>0
of a polynomial f Q[T]0
is dened as the positive rational number satisfying
f = cont(f)

f with a polynomial

f =
n

=0
a

Z[T] ,
whose coecients have greatest common divisor gcd(a
0
, ..., a
n
) = 1.
Proposition 3.40 (Gau lemma). (Carl-Friedrich Gau, 1777-1855) The
content is a multiplicative function, i.e. for two polynomials f, g Q[T] 0
we have
cont(fg) = cont(f)cont(g) .
Proof. We may assume that cont(f) = 1 = cont(g), so in particular f, g
Z[T], and have to show cont(fg) = 1, or equivalently that in the ring Z[T]
we have
p[fg =p[f or p[g
for all primes p Z, i.e. prime numbers p Z Z[T] are even prime in Z[T]!
In order to see that we consider the corresponding reduced polynomials

f, g Z
p
[T]. Since that ring is an integral domain, we know that 0 =

fg =

f g
implies

f = 0 or g = 0, and that is exactly what we need.
Corollary 3.41. 1. If f Z[T] can not be written as a product f = gh of
polynomials g, h Z[T] of lower degree, the polynomial f is irreducible
in Q[T].
96
2. If f = gh is a factorization of a monic polynomial f Z[T] with monic
factors g, h Q[T], then we have even g, h Z[T]. In particular the
rational zeroes of such a polynomial f Z[T] are integers.
Proof. 1) Assume f Z[T] is reducible (not irreducible), i.e., can be written
f = pq with polynomials p, q Q[T] of lower degree. Write p = cont(p)g, q =
cont(q)h with polynomials g, h Z[T]. Then cont(p)cont(q) = cont(f) Z
and f = (cont(p)cont(q)g)h is a factorization of f in Z[T] into polynomials
of lower degree. Contradiction!
2) The content of a monic polynomial in Q[T] is of the form 1/m with a
natural number m 1. Hence, f having content 1 = cont(g) cont(h),
so do g and h and thus g, h Z[T]. Finally apply this to a factorization
f = (T a)h, if a Q is a zero of f.
So it suces to discuss irreducibility questions within the ring Z[T]. Here
is a sucient criterion:
Proposition 3.42. Let f Z[T] 0 be a polynomial of degree 1 and
m N
>1
, such that deg(

f) = deg(f) for

f Z
m
[T]. If then

f does not
admit a factorization into polynomials of lower degree in Z
m
[T], then f is
irreducible in Q[T].
Proof. Assume f = gh in Q[T]. According to 3.41 we may assume g, h
Z[T]. But then we have

f = g

h as well with polynomials g,

h Z
m
[T] of
lower degree. Contradiction.
Example 3.43. Let f = T
5
T
2
+ 1. Take m = 2. We obtain

f =
T
5
+ T
2
+ 1 Z
2
[T]. Assume

f is not irreducible. Since

f has no zeroes in
Z
2
, there is an irreducible (monic) polynomial of degree 2 dividing

f. But the
quadratic polynomials in Z
2
are T
2
, T
2
+T = T(T +1), T
2
+1 = (T +1)
2
and
T
2
+T + 1. The last one has no zero in Z
2
and thus is irreducible, jfr. 2.29,
while the others are reducible. Now the division algorithm for polynomials
3.6 gives
T
5
+ T
2
+ 1 = (T
3
+ T
2
)(T
2
+ T + 1) + 1 ;
so

f is not divisible with T
2
+ T + 1 and hence irreducible. It follows that
f Q[T] indeed was irreducible.
Proposition 3.44 (Eisensteins criterion). (Ferdinand Gotthold Max Eisen-
stein, 1823-1852)): Let f = a
n
T
n
+ ... + a
1
T + a
0
Z[T] and p be a prime
number such that
p ,[a
n
, p[a

< n , p
2
,[a
0
.
97
Then f Q[T] is irreducible.
Proof. Assume f = gh with polynomials g, h Q[T] of lower degree, g =

k
=0
b

, h =

=0
c

with k + = n. According to 3.41 we may again

assume g, h Z[T]. We reduce mod p and obtain then
a
n
T
n
=

f = g

h .
in the polynomial ring Z
p
[T]. Hence g = b
k
T
k
and

h = c

. But k, < n
resp. k, > 0, hence b
0
= 0 = c
0
resp. p divides both b
0
and c
0
. Consequently
a
0
= b
0
c
0
is divisible with p
2
. Contradiction.
Example 3.45. Let p be a prime number. We consider the polynomial
f = T
p1
+ T
p2
+ ... + T + 1 Z[T] .
It looks like that Eisensteins criterion is not of much use here. But we can
transform the polynomial by substituting T + 1 for T: Every ring automor-
phism : Q[T] Q[T] maps irreducible polynomials onto irreducible poly-
nomials, and according to 3.15 there is a ring homomorphism : Q[T]
Q[T], which is the identity on Q Q[T] and satises (T) = T + 1. In-
deed, it is an isomorphism with inverse determined by
1
[
Q
= id
Q
and

1
(T) = T 1. Hence it is sucient to check that
(f) = f(T + 1) = (T + 1)
p1
+ ... + (T + 1) + 1
is irreducible. But
T
p
1 = (T 1)f =(T + 1)
p
1 = ((T + 1) 1)f(T + 1) = Tf(T + 1) ,
whence we obtain with the binomial formula
f(T + 1) = T
p1
+ pT
p2
+ ... +
_
p
i
_
T
pi1
+ ... +
_
p
p 2
_
T + p
and Eisensteins criterion assures that f(T + 1) - and thus also f itself - is
irreducible.
Problems 3.46. 1. R: Check whether the following polynomials are irreducible: i)
T
4
+T + 1 Z
2
[T], ii) T
3
T 1 Z
3
[T], iii) 4T
3
+ 81T
2
+ 8T + 32 Q[T], iv)
T
5
4T + 2.
98
2. R: Show that the polynomial f := T
4
10T
2
+ 1 is irreducible. Hint: Show that
f has no zeroes in Z resp. Q, and that it is not the product of two quadratic
polynomials. (We shall see in Problem 4.58.6 that

f Z
p
[T] is reducible for all
prime numbers p.)
3. R: Factorize the polynomials T
4
+ 1, T
4
4, T
4
+ 4 over Q, R, C as a product of
irreducible polynomials.
4. R: The cyclotomic polynomials f
d
Z[T], where d N
>0
, are dened by the
formula T
n
1 =

d|n
f
d
for all n N
>0
. Compute f
d
for d 9 and a prime
number d and check whether they are irreducible.
5. R: Show that Z[T] is not a principal ideal domain!
6. R: Show that Z[T] is a factorial ring.
7. Show that Gau lemma 3.40 holds for a factorial ring R instead of Z with an
appropriate(!) denition of the content. Hint: For an irreducible element u R in
a factorial ring R the factor ring R/(u) is an integral domain.
8. Show: The polynomial ring over a factorial ring R is again factorial. In particular
the polynomial ring K[Y, T] := (K[Y ])[T] is factorial.
9. Let K(Y ) := Q(K[Y ]) be the eld of all rational functions in the variable Y and
g, h K[T] two relatively prime polynomials. Show that g Y h K(Y )[T] is
irreducible. Hint: Since K[Y ] is factorial, it suces to show, that it is irreducible
in (K[Y ])[T]

= (K[T])[Y ].
99
4 Field Extensions and Galois Theory
4.1 Basic Denitions
Denition 4.1. A eld extension of the eld K is a pair (E, ) with a
eld E and a ring homomorphism : K E.
Remark 4.2. 1. According to Example 3.22.8 is automatically injec-
tive, and therefore we may identify K with (K) E, such that we
usually write K E or E K in order to denote a eld extension.
2. The notion of a real or complex vector space can easily be generalized
to that of a vector space over a eld K: In the axioms we have only
to replace R or C with K. In particular the following notions apply:
linearly dependent resp. independent, basis, dimension, linear map,
determinant of an endomorphism etc.
3. If E K is a eld extension, then E is a K-vector space - the scalar
multiplication is taken to be the eld multiplication of elements in
K E with elements in E.
As a consequence of 4.2.3 we obtain
Corollary 4.3. For any nite eld F its order q := [F[ is of the form q = p
n
with p := char(F) > 0.
Proof. Since F is nite, it has positive characteristic p > 0, and thus we
obtain the eld extension F P(F)

= Z
p
. In particular F is a nite dimen-
sional Z
p
-vector space and hence, as vector spaces, F

= (Z
p
)
n
for n = dimF.
Consequently [F[ = p
n
.
Indeed, in section 4.5 we shall see that for any q = p
n
there is, up to
isomorphy, exactly one nite eld F
q
of order [F
q
[ = q.
Let us now study some explicit examples:
Example 4.4. 1. Let f K[T] be an irreducible polynomial. Then
(E := K[T]/(f), ) is a eld extension, where the ring homomorphism
: K E is the composite of the inclusion K K[T] and the
quotient map K[T] K[T]/(f) = E.
In particular let us mention:
100
(a) The complex numbers as an extension of the real numbers: C R.
Indeed C

= R[T ]/(T
2
+ 1).
(b) Let d N
>0
be not a square. Then Q[

d] := Q+ Q

d Q is a
eld extension, and Q[

d ]

= Q[T]/(T
2
d).
(c) A nite counterpart to C: Let p be an odd prime number p ,
1 mod(4). Then the polynomial f = T
2
+ 1 has no zeros in
K := F
p
:= Z
p
, since a zero would be an element of order 4 in F

p
,
but the order of that group is not divisible with 4.
So we obtain a new eld F
p
2 := F
p
[T]/(f), where the element
i := T satises i
2
= 1 and every element can be written uniquely
in the form a+bi; a, b F
p
. The arithmetic in F
p
2 thus is the same
as that for complex numbers with the reals R replaced by F
p
.
2. The real numbers as an extension of the rational numbers: R Q.
3. The p-adic number eld as an extension of the rationals: Q
p
Q.
4. For a given eld K we mention the extensions:
(a) K(T) K, where
K(T) := Q(K[T])
denotes the eld of fractions of the polynomial ring K[T], usually
called the eld of rational functions in one variable over K;
(b) K((T)) K, where
K((T)) = T
N
K[[T]] = Q(K[[T]])
denotes the eld of formal Laurent series with nite principal part
over K, see Problem 3.9.8, and
(c) K((T)) K(T): The inclusion K[T] K[[T]] extends to an
injective homomorphism K[T]
(T)
K[[T]], since K[T] (T)
K[[T]]

. Now localization with respect to the multiplicative set

S = T
N
yields a homomorphism
K(T) = T
N
K[T]
(T)
K((T)) = T
N
K[[T]].
101
Denition 4.5. Let E K be a eld extension. The degree [E : K]
N
>0
is dened as the dimension of E as K-vector space, i.e.
[E : K] := dim
K
E .
The eld extension E K is called nite i [E : K] < .
Example 4.6. In Example 4.4.1 we have [E : K] = n := dim(f), since
according to ?? a basis of the K-vector space E is given by 1, := T, ...,
n1
.
In particular, the extensions C R and F
p
2 F
p
have degree 2, while the
remaining extensions in 4.4.2-4 are innite.
Remark 4.7. Let E K be a eld extension. We take an element a E
and consider the ring homomorphism
a
: K[T] E[T] E, f f(a),
which is the restriction to K[T] E[T] of the evaluation homomorphism
E[T] E, f f(a), cf. 3.15 with R = E. We denote
K[a] :=
a
(K[T]) = f(a); f K[T] E,
its image and
m
a
:= ker(
a
) = f K[T]; f(a) = 0
its kernel. As kernel of a ring homomorphism m
a
is an ideal, such that
K[a]

= K[T]/m
a
.
In fact, Prop.2.45 provides an isomorphism of the underlying additive groups,
which even is a ring isomorphism.
If m
a
= 0, then K[a]

= K[T] is an innite dimensional K-vector space.
Otherwise there is according to 3.27 a polynomial p
a
K[T] 0 such
that m
a
= (p
a
), and we then have dim
K
K[a] = deg(p
a
) according to ??.
The polynomial p
a
K[T] is determined up to a constant non-zero factor;
requiring it to be monic, it becomes unique. It is irreducible, since the factor
ring K[T]/m
a

= K[a] E is an integral domain, indeed, even a eld: Non-
trivial prime ideals in the principal ideal domain K[T] are maximal. So we
have a eld extension K[a] K with [K[a] : K] = deg(p
a
).
Denition 4.8. Let E K be a eld extension. An element a E is called
1. transcendent over K, i
a
is injective, i.e., i m
a
= 0.
102
2. algebraic over K, i m
a
,= 0. In that case m
a
= (p
a
) with a unique
monic (irreducible) polynomial p
a
K[T], which is called the minimal
polynomial of a E over K.
Remark 4.9. 1. If f K[T] is an irreducible monic polynomial and
a E K a zero of f, then p
a
= f.
2. In order to compute the minimal polynomial of an element a E one
considers the powers a
0
= 1, a, a
2
, ..., a
n
for n N. If they are linearly
independent for all n N, the element a E is transcendent over
K, otherwise choose n N minimal with a
0
= 1, a, a
2
, ..., a
n
linearly
dependent. So there is a linear combination

n
a
n
+ ... +
1
a +
0
= 0
with
i
K not all zero. If
n
= 0, already 1, a, ..., a
n1
are linearly
dependent, so necessarily
n
,= 0 according to the choice of n. Then
p
a
= T
n
+
n1

i=0

1
n

i
T
i
is the minimal polynomial of a E over K. Hence if a E K is
the zero of a quadratic monic polynomial f K[T], it is the minimal
polynomial: p
a
= f, e.g. the numbers

2, i C Q have the minimal

polynomials p

2
= T
2
2 resp. p
i
= T
2
+ 1. In Example 4.14. we
present an explicit calculation leading to n = 4.
3. Transcendent numbers: A complex number is called algebraic resp.
transcendent, if it is algebraic resp. transcendent over Q. The set Q
a

C of all algebraic numbers is countable, since Q[T] is countable, while
C itself is uncountable: So the vast majority of all complex numbers
numbers is transcendent, but nevertheless it is not easy to show that
a specic number is transcendent. For example one knows that each
number e
a
with an algebraic number a C is transcendent - conclude
that both e (Charles Hermite, 1822-1901) and (cf. Problem 4.15.2)
(Carl Louis Ferdinand von Lindemann, 1852-1939) are transcendent!
- or that irrational numbers a R Q are transcendent, if they can
be approximated very well by rational numbers (if one compares the
error in relation to the size of the denominators of the approximating
rational numbers). It is even known that e

is transcendent, but what

about e + , e,
e
?
103
But here we are essentially interested in algebraic or even nite extensions:
Denition 4.10. A eld extension E K is called algebraic, if every ele-
ment a E is algebraic over K.
Proposition 4.11. A nite eld extension is algebraic.
Proof. Let E K be nite and a E. Since K[a] E is a vector subspace,
we have dim
K
K[a] dim
K
E < . Now use 4.7!
Proposition 4.12. Let L E and E K be eld extensions. If they are
algebraic resp. nite, the composite eld extension L K is as well.
Furthermore, in the nite case, the degree is multiplicative, i.e.
[L : K] = [L : E] [E : K] .
Before we prove 4.12, we extend the notation K[a]:
Notation: Let E K be a eld extension, a
1
, ..., a
r
E. Then we dene
K[a
1
, ..., a
r
] :=
_
_
_

(
1
,...,
r
)N
r
b

1
,...,
r
a

1
1
... a

r
r
; b

1
,...,
r
K almost all = 0
_
_
_
E ,
where almost all means all except nitely many. With other words: All
our sums are nite.
For r = 1 this is our old denition, and furthermore for r > 1:
K[a
1
, ..., a
r
] = (K[a
1
, ..., a
r1
])[a
r
] .
Proof of 4.12. The case of nite extensions together with the degree formula
follows from the following observation: If u
1
, ..., u
n
is a basis of the K-vector
space E and v
1
, ..., v
m
a basis of the E-vector space L, then the products
u
i
v
j
, 1 i n, 1 j m, form a basis of L as K-vector space - the details
are left to the reader.
Let us now consider the algebraic case. Given an element a L we construct
a nite extension L
0
K containing it, thus proving that it is algebraic
over K. First of all: If a
1
, ..., a
r
L are algebraic, then K[a
1
, ..., a
r
] K
is a nite eld extension. To see that use remark 4.7 and induction on r as
well as what we just have noticed. Consider now the minimal polynomial
p
a
= T
m
+ b
m1
T
m1
+ ... + b
1
T + b
0
E[T] of a over E and take E
0
:=
K[b
0
, ..., b
m1
], L
0
:= E
0
[a]. Then E
0
K (since b
0
, ..., b
m1
are algebraic
over K) as well as L
0
E
0
are nite extensions; so L
0
K is as well.
104
Remark 4.13. If E K is a eld extension, the elements in E, which are
algebraic over K, constitute a subeld E
a
E, i.e. E
a
endowed with E:s
eld operations is itself a eld. For that we have to show b, c E
a
=
b + c, bc, b
1
E
a
, where b ,= 0 in the last case. But that is evident since
K[b, c] K is a nite eld extension and thus K[b, c] E
a
.
Example 4.14. We know already that

2, i C are algebraic over Q and

thus also a :=

2 +i. We compute its minimal polynomial p

a
Q[T]: First
we consider the eld E := Q[

2, i] a. Since

2 has minimal polynomial

T
2
2 over Q and i has minimal polynomial T
2
+ 1 over Q[

2], it folllows
that [E : Q] = [Q[

2][i] : Q[

2]] [Q[

2] : Q] = 2 2 = 4 and that a basis of

the Q-vector space E is given by 1,

2, i, i

2.
Since deg(p
a
) = [Q[a] : Q] divides [E : Q] = 4, the minimal polynomial has
either degree 1, 2 or 4. But 1, a =

2 + i, a
2
= 1 + 2 i

2 are obviously
linearly independent, so we obtain the degree 4, in particular E = Q[

2+i].
Furthermore a
3
=

2+5i and a
4
= 7+4i

2 = 2(1+2i

2)9 = 2a
2
9.
Hence p
a
= T
4
2T
2
+ 9.
Problems 4.15. 1. R: Let E K be a nite eld extension and a E. Show: The
K-vector space endomorphism
a
: E E, x ax, has characteristic polynomial
(p
a
)
s
with s := [E : K[a]].
2. R: Show: The number R is transcendent, using the fact that e
a
is transcendent
for any algebraic number a C.
3. Let K(X) = Q(K[X]) be the eld of all rational functions in one variable X over
K (the letter X replacing our usual T in order to reserve the latter for polynomials
over K(X)). Show that the extension K(X) K is purely transcendent, i.e.
K(X)
a
= K. Hint: Take f = g/h K(X)
a
with relatively prime polynomials
g, h K[X] and consider the equation p
f
(gh
1
)h
n
= 0, where n := deg(p
f
) with
the minimal polynomial p
f
K[T] of f over K.
4. Let K(Y ) = Q(K[Y ]) and K(X) = Q(K[X]) be two copies of the eld of all
rational functions in one variable Y resp. X over K. If f := g/h K(X) K,
where g, h K[X] are relatively prime, the ring homomorphism : K[Y ] K(X)
with [
K
= id
K
, (Y ) = f is injective and can therefore uniquely be extended to
a homomorphism : K(Y ) K(X). Show that the eld extension (K(Y ), )
has degree max(deg(g), deg(h))! Hint: The minimal polynomial of X over K(Y )

=
(K(Y )) is up to a constant factor K

Y K(Y )

the polynomial g(T)

Y h(T) K(Y )[T]. Here g(T), h(T) K[T] denote the polynomials obtained from
g, h K[X] by substituting T for X!
5. R: Let K be a eld and f = f
k
1
1
...f
k
r
r
the factorization of the polynomial f K[T]
as product of irreducible polynomials. Show an analogue of the Chinese Remainder
105
Theorem:
K[T]/(f)

=
r

i=1
K[T]/(f
k
i
i
) .
Show that the summands are local rings, with their nilradical as the maximal ideal.
Hint: Compare the dimensions of the underlying K-vector spaces!
6. Let K[] = K[T]/(f) with := T. Determine the matrix of the K-linear map

: K[] K[] given by

(x) := x with respect to the basis 1, , ...,

n1
,
where n = deg(f). Show: K[A]

= K[].
7. Algebraic integers: A complex number C is called an algebraic integer if
it is the zero of a monic polynomial f Z[T]. Show:
(a) A rational algebraic integer (sometimes simply called a rational integer) is
a usual integer.
(b) A complex number C is an algebraic integer i there is an n N and a
square matrix A Z
n,n
, i.e. with integer entries, having as an eigenvalue.
(c) The algebraic integers form a subring of C. (For this part you need to know
the notion of the tensor product V W of two K-vector spaces V and W.)
(PS: We shall see in Problam 4.46.6 that we also could have required for Q
a
to be an algebraic integer that p

Z[T] holds for the minimal polynomial p

of
over Q.)
4.2 Automorphism Groups
The modern formulation of Galois theory has been created in the 1920-ies by
Emil Artin (1898-1962). The central notion is that of a eld automorphism.
Denition 4.16. Let E K and L K be eld extensions of the same
eld K. A K-morphism (or simply morphism, if it is clear, which eld
K has to be taken) between the eld extensions E K and L K is a ring
homomorphism
: E L ,
such that [
K
= id
K
.
Two extensions E K and L K of a given eld are called isomorphic,
if there is a K-morphism : E L, which is a (ring) isomorphism.
If L = E such a is called a K-automorphism (or simply an automor-
phism) of the eld extension E K.
The set
Aut
K
(E) := : E E Kautomorphism
106
constitutes a subgroup of the permutation group S(E) and is called the au-
tomorphism group of the extension E K.
Example 4.17. 1. The identity id
E
is in any case an automorphism of
the eld extension E K.
2. Complex conjugation : C C, z z, is an automorphism of the
extension C R.
3. Let d N
>0
not be a square. Then Q[

d ] Q is a eld, and beside

id
Q
we have the automorphism : a+b

d ab

d, where a, b Q.
A non-algebraic observation: The automorphism is not continuous
(with respect to the topology on Q[

d ] R as subset of the reals):

Take a sequence (x

) Q converging to

d R. Then x

d tends
to 0, but (x

d) = x

d does not converge to (0) = 0, but to

2

d.
4. We have Aut
Q
(R) = id
R
according to 4.20.3.
5. In striking contrast the automorphism group Aut
Q
(C) is quite large:
Every automorphism : K K of some subeld K C can be
extended to an automorphism : C C, cf. Problem 5.11.3. But
the only continuous ones are the identity or complex conjugation, cf.
Problem 4.20.4.
6. Let E be a eld with char(E) = p > 0. Then E P(E)

= Z
p
is a
eld extension, and : E E, x x
p
, is a Z
p
-morphism: Obviously
(xy) = (x)(y), while
(x+y) = (x+y)
p
= x
p
+
p1

=1
_
p

_
x

y
p
+y
p
= x
p
+y
p
= (x)+(y) ,
since for 1 p 1 the binomial coecients
_
p

_
=
p!
!(p)!
are
divisible by p (p being a prime) and therefore = 0 in the eld E. Fur-
thermore x
p
= x
p1
x = x for x Z
p
according to Lagranges theorem
2.38 applied to Z

p
. It is even surjective and thus an automorphism,
if E is a nite eld, an injective map from a nite set to itself being
surjective as well. It is called Frobenius homomorphism resp. - if
it is even surjective - Frobenius automorphism (Georg Frobenius,
1849-1917).
107
The next proposition describes Aut
K
(E) for E = K[a] as a set. But
for technical reasons the formulation is slightly more general: We consider a
(ring) isomorphism : K K

between two elds K and K

. It induces
an isomorphism of the corresponding polynomial rings
K[T] K

[T], f =

:=

(a

)T

.
Then we have:
Proposition 4.18. 1. Let : K K

be a ring isomorphism between

the elds K and K

, E = K[a] K and E

eld extensions,
[E : K] < and p
a
K[T] the minimal polynomial of a over K. Then
there is a bijective correspondence between the ring homomorphisms
: E = K[a] E

extending , and the zeroes of the polynomial p

a
in the eld E

, given
by
(a) .
2. If E

= E = K[a], then
Aut
K
(E) N
E
(p
a
), (a)
is a bijection. Here for a polynomial f K[T] we denote
N
E
(f) := b E; f(b) = 0
the set of all zeroes of f in E.
Proof. i) Let : E E

be an extension of : K K

. Then for any

polynomial f K[T] we have
(f(a)) = f

((a)) .
Now every element in K[a] is of the form f(a), hence the above formula
shows that is uniquely determined by (a).
On the other hand, taking f = p
a
we see that p

a
((a)) = (p
a
(a)) = (0) =
0. It remains to show that every zero of p

a
can be realized as (a): The ring
isomorphism
K[T] E, f f(a)
108
induces an isomorphism K[T]/(p
a
)

= K[a] = E, while according to 3.15,
there is for every b E

a ring homomorphism

b
: K[T] E

with
b
[
K
= and
b
(T) = b here we regard as a ring homomorphism
K E

. If now b is a zero of p

a
, we have
b
(p
a
) = p

a
(b) = 0. Hence
(p
a
) ker(
b
), and therefore
b
factors through K[T]/(p
a
)

= K[a] = E, the
second factor being the looked for map with (a) = b.
ii) Apply i) with K

= K and = id
K
, using the fact that a K-morphism of
a nite eld extension E K to itself automatically is an automorphism, as
an injective endomorphism of the nite dimensional K-vector space E.
Let us now compute some automorphism groups:
Example 4.19. 1. Aut
R
(C) = id
C
,

= Z
2
with the complex conjuga-
tion : z z. Apply 4.18.ii) to C = R[i], p
i
= T
2
+ 1.
2. Let d N
>0
be not a square. Then Aut
Q
(Q[

d]) = id
C
,

= Z
2
with
: a + b

d a b

d. Apply 4.18.ii) to Q[

d ] Q, p

d
= T
2
d.
3. Consider a monic polynomial f Q[T] of degree deg(f) = 3 without
zero in Q. (So it is irreducible over Q). Then f has at least one real zero
a R, since the polynomial function R R, x f(x), is continuous
and lim
x
f(x) = , lim
x
f(x) = (apply the theorem on
intermediate values). We investigate the extension E := Q[a] Q:
First of all write f = (T a)g with a quadratic polynomial g E[T]
R[T]. Since any quadratic polynomial g R[T] has complex zeroes,
we may write g = (T b)(T c) with complex numbers b, c C. Now
there are two possibilities:
1.) Either b , E (and then c , E as well), e.g. if b, c C R as in the
case f = T
3
2, where we have a =
3

2, b = a, c = a
2
with the third
root of unity =
1
2
(1 + i

3). Then obviously Aut

Q
(E) = id
E
. Or
2.) b E (and then c E as well). The real zeroes a, b, c are pairwise
dierent: Regard f as function f : R R. Since f is the minimal
polynomial of its zeroes, its derivative f

as a polynomial of degree
2 < 3 = deg(f) has none of them as a zero, i.e. f has only simple and
hence 3 pairwise dierent zeroes.
109
Then there are unique automorphisms , Aut
Q
(E) with (a) =
b, (a) = c, and Aut
Q
(E) = id
E
, , . Hence [Aut
Q
(E)[ = 3, and
since a group of prime order is cyclic, it is generated by or and
=
2
, =
2
.
As an example of such a polynomial f we can take f = T
3
3T + 1.
Indeed, the identity
f(T
2
2) = T
6
6T
4
+9T
2
1 = (T
3
3T1)(T
3
3T+1) = (T
3
3T1)f
implies that b := a
2
2 ,= a (the elements 1, a, a
2
constitute a basis of
the Q-vector space E) is another zero of f. In fact, we can give a zero of
f explicitly, namely the real number a := 2 cos(2/9) = + = +
1
with := e
2i/9
: Since
3
is a third root of unity, we get (
3
)
2
+
3
+1 =
0, whence:
f(a) = ( +
1
)
3
3( +
1
)+1 =
3
+3 +3
1
+
3
3( +
1
)+1
=
3
+ 3 + 3
1
+
3
3( +
1
) + 1 =
3
+
6
+ 1 = 0 .
4. Finally we consider the extension E = Q[

2, i] Q. We know already
that E = Q[

2 + i] and
p

2+i
= T
4
2T
2
+9 = (T(

2+i))(T(

2i))(T+(

2+i))(T+(

2i)) .
In particular p

2+i
has 4 pairwise distinct zeroes in E, so there are 4
automorphisms as well. It remains to determine the group structure:
There are exactly two non-isomorphic groups of order 4, namely Z
4
and
Z
2
Z
2
, cf. Problem 2.6.4.
Now for every automorphism Aut
Q
(E) we have (

2)
2
= (

2
2
) =
(2) = 2 and (i)
2
= (i
2
) = (1) = 1. Hence (

2) =

2, (i) =
i, and any distribution of signs can be realized, since there really are
4 automorphisms and each of them is uniquely determined by its values
(

2), (i). In particular

2
= id
E
for all automorphisms, i.e. there is
no element of order 4, whence Aut
Q
(E)

= Z
2
Z
2
.
Problems 4.20. 1. R: For a eld K and n N set C
n
(K) := a K; a
n
= 1 K

,
the subgroup of K

of all n-th roots of unity in K (so C

n
= C
n
(C)). An element
C
n
(K) of order n is called a primitive n-th root of unity. Let F := K[] K
110
with a primitive n-th root of unity C
n
(F). Show: Aut
K
(F) is isomorphic with
a subgroup of the group of units Z

n
of the ring Z
n
. Hint: Every automorphism
Aut
K
(F) induces a group automorphism [
C
n
(F)
: C
n
(F) C
n
(F), and
Problem 2.39.6.
2. Let K be a eld containing all n-th roots of unity, i.e. the polynomial T
n
1 K[T]
can be factorized as a product of linear polynomials, and E := K[b] K a eld
extension, where b
n
= a K. Show that Aut
K
(E) is isomorphic with a subgroup
of the group C
n
(K) K

of all n-th roots of unity in K, hence in particular a

cyclic group. Furthermore: If n = p is a prime number, either E = K or T
p
a is
the minimal polynomial p
b
K[T] of b E over K.
3. R: Show: Aut
Q
(R) = id
R
. Hint: R
0
= x
2
; x R.
4. R: Show: Any continuous automorphism : C C is the identity or complex
conjugation.
5. Let E K be a eld extension. Show: A family of pairwise distinct automorphisms

1
, ...,
n
Aut
K
(E) End
K
(E) is linearly independent over K, where End
K
(E)
denotes the K-vector space of all endomorphisms of the K-vector space E. Hint:
Induction on n; if
1

1
+ ... +
n

n
= 0 is a non-trivial relation, we may assume

1
,= 0 and choose y E with
1
(y) ,=
n
(y). Let x E be arbitrary. Replace x
with yx in the above relation and get a new relation
1

1
(y)
1
+....+
n

n
(y)
n
= 0,
multiply the old one with
n
(y) and subtract them. One obtains a non-trivial
relation for
1
, ...,
n1
.
6. Show: Aut
K
(K(X))

= GL
2
(K)/K

E, where K

E denotes the subgroup of all

matrices E, K

, i.e. being scalar multiples of the unit matrix E K

2,2
.
Hint: Every K-morphism : K(X) K(X) is uniquely determined by its value
f := (X) K(X). It is surjective and thus an automorphism i f = g/h with
non-proportional linear polynomials g, h K[X] to see that use Problem 4.15.4.
Then one denes a homomorphism
GL
2
(K) Aut
K
(K(X)), A
A
1 ,
where the automorphism
A
: K(X) K(X) for a matrix
A =
_
a b
c d
_
GL
2
(K)
is determined by

A
(X) =
aX +b
cX +d
.
Check rst that
AB
=
B

A
! (The group PGL
n
(K) := GL
n
(K)/K

E is called
the projective linear group (of size n) over K.). Cf. Problem 3.19.6, where we
saw that Aut
K
(K[X])

= A(K).
111
4.3 Formal Derivatives and Multiplicities
In Prop. 4.18 we have seen that in order to determine the order [Aut
K
(E)[
of the automorphism group of an extension E K, it is important to know
how many zeros an irreducible polynomial f K[T] may have. An upper
bound is its degree deg(f). But is there always an extension E K with
[N
E
(f)[ = deg(f)? In that case every zero of f in E has multiplicity 1:
Denition 4.21. The multiplicity of a zero a K of a polynomial f
K[T] 0 is the unique number N, such that
f = (T a)

h with some h K[T], h(a) ,= 0 .

Remark 4.22. Let f K[T] be an irreducible polynomial and a E
K and b L K be zeros of f. Then the multiplicities of f E[T]
at a and f L[T] at b L coincide: The factorizations f = (T a)

h
resp. f = (T b)
k
g as in Def. 4.21 are in fact already over the isomorphic
subelds K[a ] E resp. K[b ] L: We have m
a
= (f) = m
b
and thus
K[a ]

= K[T]/(f)

= K[b ]. Consequently k = .
For a more detailed investigation the notion of the formal derivative of a
polynomial f K[T] plays an important r ole:
Denition 4.23. The formal derivative of a polynomial f K[T], f =

n
=0
a

, is dened as the polynomial

f

:=
n

=1
a

T
1
K[T] .
Proposition 4.24. The formal derivative
K[T] K[T], f f

,
is a K-linear map, satisfying the Leibniz rule, (Gottfried Wilhelm Leibniz,
1646-1716):
(fg)

= f

g + fg

,
as well as the chain rule:
g(f)

= g

(f) f

.
112
Furthermore for char(K) = 0:
f

= 0 f = a
0
K ,
while for char(K) = p > 0:
f

= 0 f K[T
p
] .
Proof. We comment on the Leibniz rule and the chain rule: Both the left
and the right hand side of the Leibniz rule dene bilinear maps K[T]
K[T] K[T], so it suces to check it for the polynomials 1, T, T
2
, ....
(which constitute a base of the K-vector space K[T]). But for f = T
m
, g =
T
n
it holds obviously. The chain rule is linear in g, hence we may assume
g = T
n
- and then it follows from a repeated application of the Leibniz
rule.
Corollary 4.25. A zero a K of the polynomial f K[T] is simple i
f

(a) ,= 0.
Proof. We may assume f ,= 0 and write f = (T a)

h with a polynomial
h K[T], h(a) ,= 0. Then we have f

= (T a)
1
h + (T a)

and hence
f

(a) = 0 i > 1.
Now we are able to show:
Proposition 4.26. Let f K[T] be an irreducible polynomial. Then
1. If f

,= 0, so in particular if char(K) = 0, every zero of f in some

extension E K is simple.
2. If char(K) = p > 0, we may write
f = g(T
p
n
)
with an irreducible polynomial g K[T] with g

,= 0 and hence only

simple zeros in any extension E K. In particular all zeros of f in
any extension E K have multiplicity p
n
.
Proof. We may assume that f is monic.
i) We show, that f(a) = 0 implies f

(a) ,= 0. Otherwise f

m
a
= (f)
since f as a monic irreducible polynomial is the minimal polynomial p
a
of
113
its zero a E hence f[f

. Since deg(f) > deg(f

) that implies f

= 0.
Contradiction!
ii) Choose n N maximal, such that all the exponents of the monomials
in the polynomial f are divisible with p
n
. Then f = g(T
p
n
), where g
K[T], g

,= 0. Since f is irreducible, g is as well. Let now a E be a

zero of f. Then b = a
p
n
is a simple zero of g according to the rst part, i.e.
g = (T b)h, where h(b) ,= 0. Finally f = (T
p
n
b)h(T
p
n
) = (T a)
p
n
h(T
p
n
)
with h(a
p
n
) ,= 0, i.e., a has multiplicity p
n
.
The elds K where all irreducible polynomials have automatically only
simple zeros get a name:
Denition 4.27. A eld K is called perfect if either char(K) = 0 or
char(K) = p > 0 and the Frobenius homomorphism : K K, x x
p
, is
surjective.
Example 4.28. 1. A nite eld F is perfect: The Frobenius homomor-
phism F F being an injective map from a nite set to itself is also
surjective.
2. Let char(K) = p > 0. Then K(T) = Q(K[T]) is not perfect, since the
indeterminate T is not a p-th power.
Remark 4.29. 1. An irreducible polynomial over a perfect eld K has
only simple zeros: Write f = g(T
p
n
) as in 4.26, with g =

.
Since K is perfect there are (unique) p
n
-th roots c

K of the coe-
cients a

, i.e. (c

)
p
n
= a

. Then for h =

we obtain f = h
p
n
,
and thus, f being irreducible, we nd n = 0 and f = g.
2. On the other hand, if K is not perfect, there is an irreducible polynomial
with multiple roots: Choose an element a K, which is not a p-th
power: Then f := T
p
a K[T] is an irreducible polynomial. Assume
h[f with an irreducible polynomial h K[T]. Take an extension E
K, such that f(b) = 0 for some b E. Then f = (T b)
p
and
h = (T b)

with 2 p, the ring E[T] being factorial. But

according to Prop. 4.26 the multiplicity is a p-power. Since on the
other hand b , K, we have = p, and f = h is irreducible.
For later use we mention the following consequence:
114
Corollary 4.30. Let char(K) = p > 0 and E K be a nite purely insep-
arable extension, i.e., such that for every x E there is some s N with
x
p
s
K . Then [E : K] = p

for some N.
Proof. We do induction on the extension degree [E : K]. Take some b EK
with a = b
p
K. Then according to remark 4.29 we have p
b
= T
p
a and
thus [K[b ] : K] = p. Since on the other hand [E : K[b ]] is a p-power by
induction hypothesis, we obtain that [E : K] is as well.
Problems 4.31. 1. Show Hensels lemma: Let f

Z
p
[T] be a monic polynomial
over the ring

Z
p
of p-adic integers, denote

f Z
p
[T]

= (

Z
p
/(p))[T] the induced
polynomial. Show: A simple zero Z
p
of

f has a unique lift a = (,
2
,
3
, ...)

Z
p
to a zero of f in

Z
p
. Hint: Construct inductively the components
n
Z
p
n! If

n
= c + (p
n
), then
n+1
= c + tp
n
+ (p
n+1
), where 0 t < p. Consider the
expansion f(T +c) = f(c) +f

(c)T +....
2. Let r N
>0
, relatively prime to both p and p 1. Show for the sphere

p
= x Q
p
; [x[ = 1 Q
p
the following characterization

p
= x Q

p
; x is an r
n
-th power in Q
p
for all n N.
Hint: For the inclusion it is sucient to see that any x

Z

p
is an r-th power
(why?), then use the previous problem with the polynomial f = T
r
x.
3. Show Aut
Q
(Q
p
) = id
Q
p
. Hint: Use the previous problem in order to see that
(

p
)

Z

p
for any automorphism : Q
p
Q
p
. As a consequence, is an
isometry, and thus, Q being dense in Q
p
, the identity.
4. For the ring K[[X]] of formal power series over a eld K and the factor ring
K[[T]]/(T)

= K formulate and show Hensels lemma, cf. the rst problem of this
section 4.31 as well as Problem 3.38.10.
5. Let K be a eld and
K((X)) = Q(K[[X]]) = K[[X]]

n=1
KX
n
the eld of fractions of the ring K[[X]] of formal power series over K, furthermore
r ,= char(K) a prime number. Show:
K[[X]]

= K

+XK[[X]]
= K

f K((X))

; f is an r
n
-th power in K((X)) for all n N.
115
6. Show that for Aut
K
(K((X))) we have (K[[X]]) K[[X]]. (Hint: Use the
previous problem in order to see (K[[X]]

) K[[X]]

.) Hence : K((X))
K((X)) is the unique extension of a substitution automorphism

g
: K[[X]]
K[[X]], f f(g), with a series g K

X +K[[X]]X
2
, cf. Problem 3.19.7.
7. Show Aut
K(X)
(K((X))) = id
K((X))
.
4.4 Splitting Fields
The example 4.19.3 suggests that a eld extension E K has a big auto-
morphism group, if it has the following form:
Denition 4.32. 1. A nite eld extension E K is called normal i
there is a polynomial f K[T], such that
f = (T a
1
) ... (T a
r
)
with elements a
1
, ..., a
r
E and
E = K[a
1
, ..., a
r
] .
In that case E is called a splitting eld (rotkropp) of the polynomial
f K[T].
2. A normal extension E K is called a Galois extension or galois
(

Evariste Galois, 1811-1832) if we can choose the elements a

1
, ..., a
r
E
pairwise distinct, i.e. such that f only has simple zeroes (or roots)
in E, and the automorphism group Aut
K
(E) then also is called the
Galois group of the extension E K.
Example 4.33. 1. All eld extensions E K with [E : K] = 2 are
normal. Why?
2. E = Q[

2, i] Q is normal, in fact even a Galois extension, since E is

the splitting eld of the polynomial T
4
2T
2
+ 9 or (T
2
2)(T
2
+ 1),
both having 4 distinct zeros.
Proposition 4.34. For every polynomial f K[T] there is an extension
E K, such that E is a splitting eld of f. If F K is another splitting
eld of f K[T], then the extensions E K and F K are isomorphic,
but note that there are in general several isomorphisms E

=
F.
116
Proof. Existence: Induction on deg(f). For deg(f) = 1 take E := K. Now
we assume the existence of a splitting eld E K for any polynomial f
K[T] with deg(f) < n and any eld K. Consider then a polynomial f K[T]
with deg(f) = n. According to 3.33 there is an extension E K, such that
f has a zero a E. We may even assume E = K[a] - the element a E is
algebraic over K and K[a] thus a eld.
Now write f = (T a)g with a polynomial g E[T]. Since deg(g) = n 1
there is an extension L E = K[a], such that L is a splitting eld of
g E[T]. But then L is also a splitting eld of f = (T a)g K[T].
Uniqueness: Apply the below proposition.
Proposition 4.35. Let : K F be a ring homomorphism between the
elds K and F, and let E K be a splitting eld of f K[T]. Assume that
f

F[T] is split, i.e., can be factorized as a product of linear polynomials.

1. There is an extension of : K F to a ring homomorphism :
E F. The number of such extensions equals [E : K], if f has only
simple zeros or if K is perfect, while for a non-perfect eld K it is of
the form p

[E : K] with p := char(K) and some N.

2. If g K[T] is an irreducible polynomial dividing f and a N
E
(g), b
N
F
(g

), we can even require (a) = b.

3. If F L := (K) is a splitting eld of f

all extensions : E F
are isomorphisms.
Proof. 1.) We do induction on [E : K]: For [E : K] = 1 we have E = K and
= .
Now assume that for every extension E K as above with [E : K] < n and
any : K F the statement 1.) holds.
Let now [E : K] = n. Take an element a N
E
(f)K and let g := p
a
K[T].
According to 4.18.ii) with E

= F, K

:= (K) there is for every b N

F
(g

)
precisely one extension : K[a] F with (a) = b, and N
F
(g

) ,= , since
g

divides f

. Because of [E : K[a]] < n we know according to the induction

hypothesis that every can be extended to some : E F. If f has only
simple zeros, so does g as well as g

and thus there are deg(g

) = deg(g) =
[K[a] : K] dierent choices for . On the other hand, by induction hypothesis
every admits [E : K[a ]] dierent extensions : E F. Altogether
117
: K F can be extended exactly in [E : K[a ]][K[a ] : K] = [E : K]
dierent ways.
If K is perfect we may replace f with a polynomial

f with only simple zeros:
Denote f
1
, ..., f
r
the (pairwise distinct) monic irreducible divisors of f, then
set

f = f
1
... f
r
.
Finally if K is not perfect and char(K) = p > 0, write f
i
= g
i
(T
p
s
i
) with
an irreducible polynomial g
i
, K[T
p
] as in Prop. 4.26 and denote E
0
E
the splitting eld of g
1
... g
r
. Then we have x E
0
= x
p
s
E
0
with
s = max(s
1
, ..., s
r
), since that holds for all x N
E
(f). Now there are [E
0
: K]
dierent extensions
0
: E
0
F of , while for every
0
there is only one
: E F extending
0
, the Frobenius homomorphism being injective.
Since on the other hand [E : E
0
] = p

with some N according to Cor.

4.30, we are done.
2.) is an immediate consequence of the above argument, if a , K, while the
case a K is trivial.
3.) Applying i) with the ring homomorphism
1
: L := (K) K
we obtain an extension : F E. The homomorphisms : E
E and : F F then are necessarily automorphisms as injective
endomorphisms of nite dimensional K- resp. L-vector spaces. In particular
is an isomorphism.
Remark 4.36. 1. For every nite extension E K there is an extension
L E such that L K is normal: Write E = K[a
1
, ..., a
r
] and set
f := p
a
1
... p
a
r
K[T], where p
a
i
K[T] is the minimal polynomial of
a
i
over K. Let L E be the splitting eld of the polynomial f E[T].
Obviously it is also the splitting eld of f K[T].
2. If L E K is a eld extension and L K is normal, so is L E.
Furthermore if both L E and E K are normal, then L K is
normal, if L E is the splitting eld of a polynomial K[T] E[T],
but otherwise it may happen that L K is not normal any longer, cf.
Example 4.41.2.
A natural question now arises: Is it possible to nd a (not necessarily nite)
algebraic extension E K, such that every polynomial f K[T] can be
written as a product of linear polynomials in E[T]? A possible construction
could be like this:
118
We assume that the irreducible monic polynomials in K[T] of degree > 1 can
be arranged in a sequence (f
n
)
nN
+ for K = Q this applies since Q[T] =

n=1
Q[T]
n
itself is countable (where K[T]
n
:= K KT ... KT
n
).
Then we dene inductively an increasing sequence of extensions (E
n
)
nN
:
Set E
0
:= K and for n 1 dene E
n
E
n1
to be a splitting eld of
f
n
K[T] E
n1
[T] and take E :=

n=1
E
n
as their union.
In the general case we have to apply Zorns lemma, cf. 5.4 and 5.8. In
any case the eld E deserves a name:
Denition 4.37. 1. Let E K be an algebraic extension, such that every
polynomial f K[T] can be written as a product of linear polynomials
in E[T]. Then E is called an algebraic closure of the eld K.
2. A eld K is called algebraically closed, i it is its own algebraic
closure, i.e., i every polynomial in K[T] can be written in K[T] as a
product of linear polynomials.
In fact two algebraic closures E K and L K of a given eld K are
isomorphic, cf. 5.8. Algebraically closed elds are characterized in:
Proposition 4.38. For a eld K the following statements are equivalent:
1. K is algebraically closed.
2. There are no non-trivial nite extensions E K (i.e., such that E ,=
K).
3. Every irreducible polynomial K[T] is a linear polynomial.
We leave the easy proof as an exercise to the reader.
Example 4.39. 1. The eld C of all complex numbers is algebraically
closed, cf. 4.66.
2. The algebraic closure E K of a eld K is algebraically closed: Oth-
erwise there is a non-trivial nite extension L E. The composite
extension L K is algebraic; take an element a L E and consider
its minimal polynomial p
a
K[T] over K. It can be factorized into
linear polynomials over E and hence a E. Contradiction!
119
3. The set
Q
a
:= z C; z algebraic over Q C ,
of all algebraic complex numbers constitutes a eld, cf. 4.13, the alge-
braic closure of Q. Note that it is much smaller than C, more precisely:
Q
a
is countable, but C is not.
Let us return to nite extensions! The next proposition characterizes
splitting elds:
Proposition 4.40. For a nite eld extension E K the following condi-
tions are equivalent:
1. The extension E K is normal.
2. If an irreducible polynomial g K[T] has a zero in E, then it can
already be factorized into linear polynomials in E[T].
3. If L E is another nite eld extension, then (E) = E for all
automorphisms Aut
K
(L).
Before we prove 4.40, let us discuss some examples:
Example 4.41. 1. The extension Q[
3

2] Q is not normal, since the

irreducible polynomial f = T
3
2 has a zero in Q[
3

2], but can not be

written as a product of linear polynomials in (Q[
3

2])[T].
2. A warning: If L E and E K are normal extensions, the composite
extension L K need not be normal: Consider L = Q[
4

2], E =
Q[

2], K = Q: The extensions L E and E K have degree 2

and hence are normal, but L K is not: The irreducible polynomial
T
4
2 Q[T] has the zeros
4

2 E, but is not the product of linear

polynomials L[T], since L R, while T
4
2 has the non-real zeros
i
4

2 C.
Proof. i) = iii): Assume that E is the splitting eld of the polynomial
f K[T] and : L L is a K-automorphism. Let N(f) = N
L
(f) =
a
1
, ..., a
r
. Since f((a
i
)) = (f(a
i
)) = (0) = 0, we have (N(f)) N(f)
resp. (N(f)) = N(f) (N(f) is nite and injective) and thus (E) = E
for E = K[a
1
, ..., a
r
].
iii) = ii): Assume that the irreducible polynomial g K[T] has the zero
a E. According to 4.35 there is an extension F E such that F is the
120
splitting eld of some polynomial h K[T]. Now let L F be the splitting
eld of g K[T]; then L K is the splitting eld of f = gh K[T].
We have to show that N
L
(g) E. Take b N
L
(g). According to 4.18 with
the inclusion K E as there is an automorphism Aut
K
(L) with
b = (a). But then b (E) = E.
ii) = i) : Write E = K[a
1
, ..., a
r
]. Then E is the splitting eld of the
polynomial p
a
1
... p
a
r
K[T].
Let us briey recall the facts we know about the automorphism group Aut
K
(E)
of a normal extension E K:
Theorem 4.42. Let E K be a normal extension, the splitting eld of the
polynomial f K[T]. Then
1. If g K[T] is irreducible, the automorphism group Aut
K
(E) acts tran-
sitively on the (possibly empty) set N
E
(g) := a E; g(a) = 0 of
zeros of g, i.e., for arbitrary a, b N
E
(g) there is an automorphism
Aut
K
(E) with (a) = b.
2. We have
[Aut
K
(E)[ = [E : K],
if E K is galois, e.g. if K is perfect. In the general case we have
[E : K] = p

[Aut
K
(E)[
for p = char(K) and some N.
3. The map [
N
E
(f)
denes an injective group homomorphism
Aut
K
(E) S(N
E
(f))
from the automorphism group of the extension E K to the group of
permutations of the set N
E
(f) of zeros of the polynomial f in E. If
f
1
, ..., f
r
K[T] are the pairwise distinct irreducible monic divisors of
f, then the above homomorphism can be factorized:
Aut
K
(E) S(N
E
(f
1
)) ... S(N
E
(f
r
)) S(N
E
(f)).
121
Proof. i) According to 4.40 we have either N
E
(g) = or g can be written
as a product of linear polynomials in E[T]. In the rst case there is nothing
to be shown, while in the second case we may assume that g divides the
polynomial giving rise to E as its splitting eld - if not, we may multiply it
with g - the splitting eld remains the same. Now apply 4.35.2.
ii) is nothing but 4.35.i) with F = E and the inclusion : K E.
iii) follows from the fact that E = K[a
1
, ..., a
r
], where N
E
(f) = a
1
, ..., a
r
,
and that a K-automorphism is uniquely determined by its values (a
i
)
N
E
(f), 1 i r.
Remark 4.43. Let us briey recall the explicit construction of automor-
phisms : E = K[a
1
, ..., a
r
] E for a normal extension E K: Ev-
ery automorphism Aut
K
(E) is obtained in the following way: Let
E
i
:= K[a
1
, ..., a
i
]. We have in any case [
E
0
= id
E
0
. If [
E
i
already is
given, we may choose (a
i+1
) freely in the (non-empty) zero set N
E
(g

),
where g E
i
[T] is the minimal polynomial of a
i+1
E over E
i
. Indeed, the
minimal polynomial h := p
a
i+1
K[T] of a
i+1
over K is split over E, and
g[h implies g

[h

= h.
Example 4.44. 1. Take the polynomial f = T
3
2 Q[T], cf. 4.19 3.1.
Its splitting eld is E := Q[
3

2,
3

2,
3

2
2
] = Q[
3

2, ] with the third

root of unity :=
1
2
(1 + i

3). We have E
1
= Q[
3

2] and there are

three dierent Q-morphisms

: E
1
E; 0 2, with

(
3

2) =
3

. Each

: E
1
E in turn may be extended in two ways: The
minimal polynomial g E
1
[T] of is g = (T )(T
1
) = T
2
+T +1,
lying even in Q[T]; hence g

= g for = 0, 1, 2, and

extends to
the automorphisms

: E E with

() =
1
. Note that any
permutation of N
E
(f) is realized by some automorphism.
2. The situation is more interesting for the polynomial f = T
4
2
Q[T]. The splitting eld is E := Q[
4

2, i] with the fourth root of unity

i, and we could argue as in the rst case. Instead we consider the
representation E = Q[

2, i,
4

2] and discuss the step from E

2
=: L to
E
3
= E. The intermediate eld L = Q[

2, i] is the splitting eld of

(T
2
2)(T
2
+1), cf. 4.19 iv). Since L Q is normal, we have (L) = L
for every Aut
Q
(E). The minimal polynomial g L[T] of
4

2 is
T
2

2 and g

= T
2

2. Depending on the sign 1 necessarily

(
4

2) =
4

2 or (
4

2) =
4

2i. In contrast to the situation in i) not

122
every permutation of N
E
(f) is the restriction of some automorphism,
cf. Problem 4.46.2.
The case where the splitting eld E K is of maximal size is character-
ized by:
Proposition 4.45. Let E K be the splitting eld of f K[T], n :=
deg(f). Then we have [E : K] n!. Furthermore for n 3, equality
[E : K] = n! holds if and only if [N
E
(f)[ = n and the homomorphism
Aut
K
(E) S(N
E
(f)), [
N
E
(f)
, is an isomorphism.
Proof. The degree estimate is shown by induction on n.
= : If r := [N
E
(f)[ < n, there is a multiple root a of f, i.e. such that
f = (T a)
2
g. But then we obtain E L := K[a] as splitting eld of
g L[T] and [E : K] = [E : L][L : K] (n 2)! n < n!. Contradiction!.
So f has n pairwise distinct zeros and the extension E K is galois, hence
[Aut
K
(E)[ = [E : K] = n! and the injective homomorphism Aut
K
(E)
S(N
E
(f)) is even an isomorphism.
= : Because of [N
E
(f)[ = n the polynomial f has no multiple roots;
thus E K is galois. We have then [E : K] = [Aut
K
(E)[ = n!.
Problems 4.46. 1. R: Let K be a eld of characteristic p > 0 and E K a splitting
eld of T
p
T c K[T]. Show, that either E = K or E K is Galois with
[E : K] = p and cyclic Galois group Aut
K
(E)! What can be said, if K = Z
p
? Hint:
If a E K is a zero of f, compute f(a +n)!
2. R: Let E Q be the splitting eld of the polynomial T
4
2. Show that Aut
Q
(E)

=
D
4
, where D
4
denotes the dihedral group. Hint: Consider the square with vertices

2,
4

2i!
3. R: Compute the splitting eld E Q of f := T
4
10T
2
+ 1. Determine Aut
Q
(E)!
Hint: Consider the splitting eld F of the polynomial T
2
10T + 1? Then try
to compute the square roots of its zeros in F - that is not possible since there is
missing what in F?
4. Let f Q[T] be an irreducible polynomial whose degree is a prime p and which has
exactly two (simple) non-real roots. Show that its splitting eld E Q satises
Aut
Q
(E)

= S
p
. Hint: Apply Problem 2.39.5.
5. Show that the polynomial f := T
5
4T +2 satises the conditions in the preceding
problem. Hint: Apply real analysis: Determine the real zeros of its derivative f

and the sign of the corresponding value of f. Then one can apply which theorem?
6. Show: An algebraic number Q
a
is an algebraic integer i its minimal polynomial
p

Q[T] has integer coecients: p

Z[T]. Hint: Consider a splitting eld E Q

containing ; if is an algebraic integers its conjugates () E, Aut
Q
(E),
are algebraic integers as well. Cf. also Problem 4.15.7.
123
4.5 Finite Fields
The aim of this section is the complete classication of all nite elds. We
start our investigations with the following general result about nite sub-
groups of the multiplicative group K

of any eld K:
Proposition 4.47. Any nite subgroup of the multiplicative group K

of a
eld K is cyclic. In particular, the multiplicative group F

of a nite eld F
is cyclic
Proof. Denote G K

a nite subgroup of the multiplicative group K

of
our eld K. Since G is abelian, there is according to 2.84 an isomorphism
G

= Z
n
1
q
1
... Z
n
r
q
r
- (with the right hand side additively written!) - where
q
1
, ..., q
r
are pairwise distinct prime powers and n
i
N
>0
. Then the least
common multiple q := lcm(q
1
, ..., q
r
) satises a
q
= 1 for all a G, i.e., the
zero set of the polynomial T
q
1 K[T] contains the entire group G. A
polynomial of degree q over an integral domain has at most q zeros, cf. 3.17,
hence q
n
1
1
... q
n
r
r
= [G[ q.
But that is possible only if the numbers q
1
, ..., q
r
are pairwise relatively prime
and n
1
= ... = n
r
= 1. In that case (

1, ...,

1) has order q
1
... q
r
= q =
[Z
q
1
... Z
q
r
[, so G is a cyclic group.
To every prime power q = p
n
we associate a nite eld F
q
:
Denition 4.48. Let q = p
n
with a prime number p and n N
>0
. We
denote F
q
Z
p
a splitting eld of the polynomial T
q
T Z
p
[T].
Remark 4.49. 1. For n = 1 we have F
p
= Z
p
, while F
p
n ,

= Z
p
n for n > 1,
since the ring Z
p
n, containing non-zero nilpotent elements, neither is a
eld nor an integral domain.
2. There is a further dierence between the denition of Z
q
and that of F
q
:
While the rst one is well dened even as a set, the latter has as a set no
natural realization, though all constructions lead to isomorphic elds,
cf. 4.34. Hence from the point of view of algebra, the choices entering
in a concrete realization are not really interesting, and one usually
refers to the eld F
q
. For example given a generator a F

q
of its
multiplicative group, we nd F
q

= Z
p
[T]/(p
a
), where p
a
is the minimal
polynomial of the generator a, but unfortunately the polynomial p
a
in
general really depends on the choice of that generator.
124
The main result of this section is
Theorem 4.50. 1. We have [F
p
n[ = p
n
resp. [F
p
n : F
p
] = n.
2. Every nite eld is isomorphic to a eld F
p
n.
3. The eld F
p
m is isomorphic to a subeld of F
p
n, i m is a divisor of n.
(But note that for m > 1 there are then several ring homomorphisms
F
p
m F
p
n, two such homomorphisms diering by an automorphism
Aut
F
p
(F
p
m)!)
4. The extension F
p
n F
p
is a Galois extension, and Aut
F
p
(F
p
n) is a
cyclic group of order n, generated by the Frobenius automorphism :
F
p
n F
p
n, x x
p
.
Remark 4.51. The eld F
p
n is also called the eld with p
n
elements. A
more old fashioned notation for it is GF(p
n
), where GF is the abbreviation
for Galois eld. This is also the reason why in English the word eld is
used in general, while in most other languages the corresponding translation
of kropp applies, as introduced by Bourbaki, a group of french mathe-
maticians which beginning in the 1930ies tried to modernize and systematize
mathematics.
Proof. i) The polynomial f := T
p
n
T has derivative f

= 1 and hence
only simple zeros, and thus [N
F
p
n
(f)[ = p
n
. But on the other side we may
interpret that zero set as the xed point set of the n-th iterate
n
of the
Frobenius automorphism : F
p
n F
p
n, x x
p
, i.e.:
N
F
p
n
(f) = Fix(
n
) := x F
p
n; x =
n
(x) = x
p
n

and hence is in particular itself a eld! Since F

p
n is the smallest extension of
F
p
, over which f is split, it follows F
p
n = Fix(
n
).
ii) According to Corollary 4.3 we already know [F[ = p
n
with some n N
>0
.
On the other hand Corollary 2.38 tells us that x
p
n
1
= x
|F

|
= 1 holds for all
x F

resp. x
p
n
= x for all x F. So the polynomial f := T
p
n
T has the
entire eld F as its zero set, in particular
T
p
n
T =

aF
(T a),
both sides being monic polynomials with the same (simple) zeroes. With
other words, the eld F is a splitting eld of f Z
p
[T] = F
p
[T].
125
iii) =: Assume : F
p
m F
p
n is a ring homomorphism (it is not unique
for m > 1). Then F
p
n is a F
p
m-vector space, whence F
p
n

= (F
p
m)
r
with
r = dim
F
p
m
F
p
n resp. p
n
= (p
m
)
r
= p
mr
resp. n = mr.
=: Let n = mr. Since F
p
m is the splitting eld of T
p
m
1
1, it is sucient
to show:
(T
p
m
1
1) divides (T
p
n
1
1) .
In any case T 1 divides T
r
1, and after substitution of T by T
m
we see
that T
m
1 divides T
n
1. In particular the number p
m
1 divides p
n
1,
and nally, with p
m
1 and p
n
1 instead of m and n we arrive at our claim.
iv) The extension F
p
n F
p
= Z
p
is galois, since T
p
n
T only has simple
zeros. If we can show, that the Frobenius automorphism : F
p
n F
p
n has
order n, we are done, since the automorphism group Aut
F
p
(F
p
n) has order n
according to 4.42. But
k
= id
F
p
n
is equivalent to x
p
k
1
= 1 for all x F

p
n.
Since F

p
n is cyclic and thus there is an element x F

p
n of order p
n
1, we
see that
k
,= id
F
p
n
f or k < n.
Remark 4.52. Let us give here an explicit description of an algebraic closure
F
p
F
p
of the nite eld F
p
. Take E
n
:= F
p
n! and choose ring homomor-
phisms
n
: E
n
E
n+1
. Then, interpreting
n
as an inclusion E
n
E
n+1
we may dene
F
p
:=

_
n=1
E
n
.
4.5.1 Digression 1: Quadratic reciprocity
As an application of nite elds we shall give a proof of the law of quadratic
reciprocity. In order to formulate it we need the Legendre symbol (Andre
Marie Legendre, 1752 - 1833):
Denition 4.53. Denote P
>2
the set of all odd primes. The Legendre symbol
is the map
_ _
: Z P
>2
0, 1, (a, p)
_
a
p
_
,
where
_
a
p
_
:=
_
_
_
1 , if a F

p
is a square
1 , if a F

p
is not a square
0 , if a = 0 F
p
.
126
Remark 4.54. 1. The Legendre symbol depends only on a F
p
, so for
convenience we dene
_
a
p
_
:=
_
a
p
_
for a F
p
.
2. If c F

p
is a generator of the (cyclic) multiplicative group F

p
, i.e.,
F

p
= c
Z
, we have
_
c

p
_
= (1)

.
3. 0, 1 K for any eld K with char(K) ,= 2.
4. With that convention we have
_
a
p
_
= a
p1
2
F
p
for all a F
p
.
5. The Legendre symbol is multiplicative in the upper variable:
_
ab
p
_
=
_
a
p
__
b
p
_
for a, b Z as well as a, b Z
p
. In particular it is sucient to compute
the Legendre symbol for a being a prime as well.
Theorem 4.55. Let p P
>2
be an odd prime.
1. We have
_
2
p
_
=
_
1 , if p 1 mod (8)
1 , if p 3 mod (8)
,
or, more briey
_
2
p
_
= (1)
p
2
1
8
.
2. The law of quadratic reciprocity: For a prime q P
>2
dierent from p
we have
_
p
q
_
= (1)
p1
2
q1
2
_
q
p
_
.
127
Proof. 1.) The eld F
p
2 F
p
contains an element of order 8, since the
order p
2
1 of the cyclic group (F
p
2)

is divisible by 8.
Now for := +
1
we have

2
=
2
+ 2 +
2
=
2
+ 2
2
= 2,
since
4
= 1. As a consequence 2 is a square in F
p
i F
p
F
p
2 i
=
p
. Now

p
=
p
+
p
=
s
+
s
,
where p = 8d + r. Thus for r = 1 we nd
p
= , while r = 3 yields

p
= .
2.) We do computations in the splitting eld E F
p
of the polynomial
T
q
1 F
p
[T], denote E 1 a root of it and use the notation
(a) :=
_
a
q
_
0, 1 E.
We need the following auxiliary lemma:
Lemma 4.56. The square of the Gau sum
:=
q1

i=0
(i)
i
satises

2
= (1)q F

p
E.
Let us rst nish the proof of the theorem: We apply the Frobenius map
E E, x x
p
, to our Gau sum:

p
=
q1

i=0
(i)
p

ip
=
q1

i=0
(i)
ip
=
q1

i=0
(p
2
i)
ip
= (p)
q1

i=0
(ip)
ip
= (p),
128
using (i)
p
= (i) = (p
2
i) and the fact that Z
q
Z
q
, i ip, is a bijection.
Since ,= 0, we may conclude
_
p
q
_
=
p1
= (
2
)
p1
2
= ((1)q)
p1
2
= ((1)
q1
2
)
p1
2
q
p1
2
= (1)
q1
2
p1
2
_
q
p
_
,
where we have used Remark 4.54.4 with respect to a = 1 and the prime q
as well as a = q and the prime p.
Proof of the lemma. First of all we may understand the exponents of as
well as the argument of as elements in Z
q
. Using that convention we
remark the following identities:

iZ
q
(i) = 0,

iZ
q

i
= 0.
The rst one follows from the fact that (0) = 0 and there are
q1
2
quadratic
residues as well as
q1
2
quadratic non-residues in Z
q
, for the second one notes
that the sum not changing when multiplied by has to be = 0.
Now

2
=

(i,j)(Z
q
)
2
(i)(j)
i+j
=

Z
q
_

i+j=
(i)(j)
_

.
Let us now compute the inner sums. The case = 0 yields

i+j=0
(i)(j) =

iZ
q
(i)(i) =

iZ
q
(i
2
) = (1)(q 1),
since (i
2
) = (i
2
)(1) = (1) for i ,= 0, while (0
2
) = 0.
129
Finally we treat the case ,= 0 and get

i+j=
(i)(j) =

iZ

q
(i)( i)
=

iZ

q
(i
1
)( i)
=

iZ

q
(i
1
1)
=

iZ
q
\{1}
(i) = (1),
since (0) = 0, (i
1
) = (i) and i
1
1; i Z

q
= Z
q
1. Hence

2
= (1)(q 1) (1)

= (1)(q 1) (1) (1) = (1)q.

4.5.2 Digression 2: Further Simple Groups
Finite elds may be used to give further examples of simple groups. First of
all note that we may dene the general linear group GL
n
(K) and the special
linear group SL
n
(K) of 2.4.5 and 2.24.2 for any eld K, in particular for
nite elds K = F.
The general linear group acts in a natural way on the projective space
P
n1
(K), the set of all lines (:= one dimensional subspaces) in K
n
, cf. Prob-
lems 2.18.11 and 4.20.6. The kernel of the corresponding group homomor-
phism GL
n
(K) S(P
n1
(K)) is K

E, the subgroup of all non-zero multi-

ples of the unit matrix E = (
ij
). So the projective (general) linear group
PGL
n
(K) := GL
n
(K)/K

E
can be understood as a group of projective transformations, i.e. permutations
of P
n1
(K), usually called projective linear transformations.
Restricting everything to SL
n
(K) GL
n
(K) we obtain the projective
special linear group
PSL
n
(K) := SL
n
(K)/C
n
(K)E
130
with the group
C
n
(K) := a K

; a
n
= 1
of n-th roots of unity in the eld K. We note that K

E GL
n
(K) and
C
n
(K) E SL
n
(K) are nothing but the centers of the general resp. special
linear group of size n. Obviously
PSL
n
(K) PGL
n
(K),
with factor group
PGL
n
(K)/PSL
n
(K)

= GL
n
(K)/K

SL
n
(K)

= K

/ det(K

E) = K

/p
n
(K

)
with the n-th power map p
n
: K

, x x
n
.
The central result of this digression is:
Theorem 4.57. Let K be a eld. The group PSL
n
(K) is simple for n > 2
and for n = 2, [K[ > 3.
Proof. Since the inverse image of a normal subgroup with respect to a group
homomorphism itself is normal, it suces to show the following: Any normal
subgroup N SL
n
(K) containing the center C
n
(K)E as a proper subgroup
coincides with SL
n
(K).
The strategy is as follows: We show
1. The group G := SL
n
(K) is generated by elementary matrices.
2. If NG is a normal subgroup containing the center C
n
(K)E as proper
subgroup, then the factor group G/N is abelian.
3. If n > 2 or n = 2 and [K[ > 3 any elementary matrix can be written
as a commutator ABA
1
B
1
with matrices A, B G.
By 2) every commutator ABA
1
B
1
belongs to the subgroup N G, and
then 1) and 3) tell us that N = G.
Generators for SL
n
(K): We identify a matrix A GL
n
(K) freely with
the corresponding linear map K
n
K
n
, x Ax, and denote e
1
, ..., e
n
the
131
standard base of the vector space K
n
. Denote E = (
ij
) the unit matrix and
E
k
= (
ik

j
), i.e.
E
k
e
i
=
_
e
k
, if i =
0 , otherwise
,
whence E
k
E
rs
=
r
E
ks
. An elementary matrix now is a matrix
Q
ij
() := E + E
ij
, i ,= j, K

.
Note that K SL
n
(K), Q
ij
() is a group homomorphism:
Q
ij
( + ) = Q
ij
()Q
ij
(),
and that for n > 2 all Q
ij
() for xed K

belong to the same conjugacy

class in SL
n
(K): There is some P SL
n
(K) with Pe
1
= e
i
, Pe
2
= e
j
,
whence
PQ
12
()P
1
= Q
ij
().
For n = 2 we take P with Pe
1
= e
2
, Pe
2
= e
1
and nd
PQ
12
()P
1
= Q
21
().
From linear algebra it is well known that GL
n
(K) is generated by the
following three groups of matrices:
The elementary matrices Q
ij
(), i ,= j, K

,
the transposition matrices T
k
, k < , satisfying
T
k
e
i
=
_
_
_
e
k
, if i =
e

, if i = k
e
i
, otherwise
,
and
the diagonal matrices.
More precisely, any matrix A GL
n
(K) is of the form A = A
0
TD, where
A
0
is a product of elementary matrices, T a product of transposition matrices
and D a diagonal matrix (the matrix TD having in every row and column
exactly one nonzero entry). In order to get generators of SL
n
(K) we have
to modify the matrices of the second and third kind: We claim that
132
the elementary matrices Q
ij
(), i ,= j, K

,
the special transposition matrices P
k
, k < , with
P
k
e
i
=
_
_
_
e
k
, if i =
e

, if i = k
e
i
, otherwise
,
and
the special diagonal matrices, i.e. the diagonal matrices
D
1
(
1
)D
2
(
2
) ... D
n1
(
n1
)
with (unique)
1
, ...,
n1
K

, where D
k
(), k < n, is dened by
D
k
() e
i
=
_
_
_
e
k
, if i = k

1
e
k+1
, if i = k + 1
e
i
, otherwise
,
together generate SL
n
(K): Let A SL
n
(K), A = A
0
TD as above. If we
replace T, a product of matrices T
k
with P, the corresponding product of the
matrices P
k
, we have A = A
0
P

D with a diagonal matrix

D, whose entries
coincide with those of D up to sign. Since A
0
, P are special matrices,

D is
special as well.
1.) The elementary matrices generate SL
n
(K): We show that the ma-
trices P
k
, k < n, and D
k
(), k = 1, ..., n 1, are products of elementary
matrices. In fact
P
k
= Q
k
(1)Q
k
(1)Q
k
(1),
and for the D
k
() we may assume n = 2 and nd for D() := D
1
() that
D() = Q
12
()Q
21
(
1
)Q
12
()P
12
,
which yields the desired result, since we already know that P
12
is a product
of elementary matrices.
2.) Abelian Factor group: The stabilizer
U := SL
n
(K)
L
= A SL
n
(K); A(L) = L
133
of the line L := Ke
1
P
n1
(K) satises
A U A =
_
b
0 C
_
with K

, b
T
, 0 K
n1
, C K
n1.n1
, det C = 1.
Consider the kernel U
0
U of the natural group homomorphism
U GL(L) GL(K
n
/L), A (A[
L
, A),
where A : K
n
/L K
n
/L is the linear map x + L Ax + L. In fact,
A U
0
A =
_
1 b
0 E
_
with b
T
, 0 K
n1
, E = (
ij
) K
n1.n1
,
and
K
n1
U
0
, y
_
1 y
T
0 E
_
is a group isomorphism; in particular, U
0
is abelian.
We have
SL
n
(K) = NU,
since U = SL
n
(K)
L
is the stabilizer of L P
n1
(K) in SL
n
(K) and the
subgroup N SL
n
(K) acts transitively on P
n1
(K): Take a line Ky ,= L.
Since N C
n
(K)E, there is some B NK

E. Hence we can nd x K
n
,
such that x, Bx K
n
are linearly independent and choose C SL
n
(K) with
Cx = e
1
, CBx = y for some K

. Then A := CBC
1
N satises
Ae
1
= y, in particular Ky = A(L).
Now we show that even
SL
n
(K) = NU
0
.
First of all, NU
0
NU = SL
n
(K) is a normal subgroup. To see that we
have to show CNU
0
= NU
0
C for all C SL
n
(K) = NU. We may assume
C N or C U. For C U that is obvious since both U
0
U and N G
are normal subgroups. For C N we nd CN = N = NC and NU
0
= U
0
N,
hence CNU
0
= NU
0
= U
0
N = U
0
NC.
Now Q
12
() U
0
NU
0
SL
n
(K) for any K

. Since Q
ij
() is
conjugate to Q
12
() or Q
12
(), all the generators Q
ij
() of SL
n
(K) are
contained in NU
0
, hence NU
0
coincides with SL
n
(K).
134
Finally we see that
SL
n
(K)/N = (NU
0
)/N = (U
0
N)/N

= U
0
/(U
0
N)
is isomorphic to a factor group of the abelian group U
0
and hence itself
abelian.
3.) Commutators: For n 3, given two distinct indices i, j choose a third
index ,= i, j. Then
Q
ij
() = Q
i
()Q
j
(1)Q
i
()Q
j
(1)
is a commutator because of Q
rs
()
1
= Q
rs
(). For n = 2 and [K[ > 3
choose K 0, 1 , = . Then with the diagonal matrix
D() :=
_
0
0
1
_
we get
D()Q
12
()D(
1
)Q
12
() = Q
12
((
2
1)).
Hence because of
2
1 ,= 0, any elementary matrix is again a commutator.
Problems 4.58. 1. R: Show that Z

p
n is a cyclic group for any prime p > 2. Fur-
thermore the residue class of an integer k Z Zp generates Z

p
2
if and only if it
generates Z

p
n for all n N
>0
. Hint: According to Problem 2.62.7 the subgroup
U(p
n
) Z

p
n is cyclic.
2. R: Let char(K) = p > 0. Assume a E K with b := a
p
r
K, a
p
r1
, K. Show:
T
p
r
b is the minimal polynomial p
a
of a over K. Hint: Consider rst the case
r = 1 and show then by induction [K[a] : K] = p
r
.
3. R: Determine all generators of the cyclic group F

9
and their minimal polynomials
over F
3
! Hint: Example 4.4 c).
4. R: Find a concrete realization of the eld F
p
p, i.e. an isomorphism F
p
p

= Z
p
[T]/(f),
where f = .... ? Same question for F
16
, cf. Problems 4.46.1 and 3.46.1.
5. R: Let K be an algebraically closed eld. Determine the order [C
n
(K)[ of the group
C
n
(K) K

of n-th roots of unity in K.

6. Show the converse of 4.47: If K is a eld and K

cyclic, then K is nite. Hint:

Assume K

=< a > and [K[ = . Then K has characteristic char(K) = 2 - why?

- and a is transcendent over F
2

= P(K) K resp. K = Q(F
2
[a])

= F
2
(T).
135
7. Let F be a nite eld. Show for a, b F: If ab is not a square in F, one of the
factors is a square in F and the other is not. Assume now that p is an odd prime
and 6 not a square in F
p
. Then F
p
2 = F
p
[] with an element F
p
2,
2
= 6. Show
that the number 5 + 2 F
p
2 is a square in F
p
2.
8. Let f := T
4
10T
2
+1 Z[T]. Show that the reduced polynomial

f Z
p
[T] = F
p
[T]
is reducible for all primes p. Hint: The case p = 2 is easy. Otherwise write
f = (T
2
5)
2
24. If 6 is a square in F
p
, we are done. Otherwise F
p
2 = F
p
[] with
an element F
p
2,
2
= 6. Then f = ((T
2
5) 2)((T
2
5) +2). Use now the
preceding problem 4.58.5 in order to factorize f as product of linear polynomials
over F
p
2. They can be paired together to quadratic polynomials F
p
[T].
9. R: Let F := F
q
be the nite eld with q = p
r
element. Show: The ring homomor-
phism F[T] F
F
, f

f, cf. Remark 3.16, is surjective with kernel (T
q
T).
10. Compute the orders of the groups GL
n
(F
q
), SL
n
(F
q
), PGL
n
(F
q
), PSL
n
(F
q
)!
11. Show PSL
2
(F
2
) = PGL
2
(F
2
)

= S
3
. Hint: [P
2
(F
2
)[ = 3, Problem 2.18.12.
12. Show PGL
2
(F
3
)

= S
4
and PSL
2
(F
3
)

= A
4
. Hint: [P
2
(F
3
)[ = 4.
4.6 Galois Theory
The fundamental theorem of Galois theory explains the structure of a
Galois extension E K in terms of its automorphism group (Galois group)
Aut
K
(E).
Denition 4.59. An intermediate eld L of an extension E K is a
subeld L E containing K, i.e. K L E.
Theorem 4.60 (Fundamental Theorem of Galois Theory). Let E K
be a Galois extension and let G := Aut
K
(E) be its Galois group. Then there
is a bijection
L intermediate elds of the extension E K H G subgroup
between the set of all intermediate elds of the extension E K and the set
of all subgroups of G = Aut
K
(E), dened as follows
E L H := Aut
L
(E) G
resp. in the reverse direction:
G H L := Fix(H) := a E; (a) = a , H .
136
It satises
[Aut
L
(E)[ = [E : L] , [H[ = [E : Fix(H)] ,
and H = Aut
L
(E) is a normal subgroup of G = Aut
K
(E), i the extension
L K is normal. In that case the restriction
G = Aut
K
(E) Aut
K
(L) , [
L
,
induces an isomorphism
G/H

= Aut
K
(L) .
Example 4.61. The splitting eld of f := T
3
2 Q[T] is E = Q[
3

2, ]
with the third root of unity :=
1
2
(1+i

3), cf. Example 4.44.1. We already

know the automorphisms , G := Aut
Q
(E) with (
3

2) =
3

2, () =
, (
3

2) =
3

2, () =
2
. In fact Aut
Q
(E)

= S(N
E
(f))

= S
3

= D
3
.
The automorphism has order 3 and order 2 and id
E
, ,
2
, , ,
2
constitute the entire Galois group; they have order 1,3,3,2,2,2.
Let us now determine the non-trivial subgroups: The possible orders being
2 and 3, such a subgroup is cyclic. Hence we nd < >=<
2
>, < >,
< >, <
2
>. If we use that always
[H[ = [E : Fix(H)] = 6/[Fix(H) : Q]
we obtain the following table of subgroups and corresponding xed elds:
H Fix(H)
G Q[
3

2, ]
< > Q[]
< > Q[
3

2]
< > Q[
3

2]
<
2
> Q[
3

2
2
]
< id
E
> E
Proof. We have to show that the given maps are inverse one to another, i.e.:
Fix(Aut
L
(E)) = L , Aut
Fix(H)
(E) = H
for every intermediate eld L of E K and every subgroup H G. Since
E L resp. E Fix(H) are again Galois extensions, we may assume
L = K resp. Fix(H) = K and show then for a Galois extension E K the
propositions 4.62 and 4.64.
137
Proposition 4.62. For a Galois extension E K we have
Fix(Aut
K
(E)) = K .
Proof. The inclusion K F := Fix(Aut
K
(E)) is obvious. On the other side
we have Aut
F
(E) = Aut
K
(E) and therefore (see 4.42):
[E : F] = [Aut
F
(E)[ = [Aut
K
(E)[ = [E : K] ,
whence F = K.
Before we come to the second proposition, we need an auxiliary result
telling us when an extension is of the form K[a] K:
Theorem 4.63. (Primitive Element Theorem) A nite eld extension
E K admits a primitive element a E, i.e., such that E = K[a], i
there are only nitely many intermediate elds for the extension E K.
That condition is satised, if E K can be extended to a Galois extension,
in particular if char(K) = 0 or more generally, if K is perfect.
Proof. Since both conditions are satised if K (and with K also E) are nite
(according to 4.47 we have E

=< a >, whence E = K[a]), we may assume

that K is innite.
=: We do induction on n := [E : K], the case n = 1 being trivial.
Let now n > 1. Choose an element b , K. Then E K[b ] satises
[E : K[b ]] < n and has as well only nitely many intermediate elds, hence
admits according to the induction hypothesis a primitive element c E,
i.e., E = (K[b ])[c ] = K[b, c ]. Now we consider the intermediate elds
K[b +c ], K. Since there are only nitely many intermediate elds and
[K[ = , we nd two elements
1
,
2
K with K[b+
1
c] = K[b+
2
c] =: L.
But then we have even b, c L resp. E = K[b, c] L resp. E = L = K[a]
with a := b +
1
c.
=: Assume now E = K[a]. To every intermediate eld L we can as-
sociate the minimal polynomial p
L
L[T] of our primitive element a over
L. The map L p
L
L[T] E[T] is injective: If n = [E : L] and
p
L
= T
n
+

n1
=0

, we have L = K[
0
, ...,
n1
], i.e. we may reconstruct
L from p
L
. Let F := K[
0
, ...,
n1
]. In any case we have F L, but on the
other hand [F : K] = [L : K] because of
n[F : K] = [E : F][F : K] = [E : K] = [E : L][L : K] = n[L : K] .
138
Consequently L = F.
Now every polynomial p
L
is a divisor of the minimal polynomial p
K
in the
ring E[T]. Write then p
K
as a product of (nitely many) monic irreducible
polynomials E[T], cf. 3.33. Any polynomial p
L
is then a product of certain
of these polynomials (here we use again the unique factorization property
3.33!), and there are of course only nitely many possibilities.
It remains to show that a Galois extension E K satises the given con-
dition, but Prop. 4.62 with a subeld L instead of K means that L
Aut
L
(E) G is injective, and G has of course only nitely many sub-
groups.
Proposition 4.64. Let E K be a Galois extension and H Aut
K
(E) a
subgroup with Fix(H) = K. Then we have
H = Aut
K
(E) = Aut
Fix(H)
(E).
Proof. Let G := Aut
K
(E). It suces to show [G[ = [E : K] [H[. Ac-
cording to 4.63 we may write E = K[a] with a primitive element a E. We
consider the polynomial
f :=

bHa
(T b) E[T] .
Here Ha denotes the H-orbit of a E, i.e.
Ha = (a); H E.
Every automorphism H induces a permutation [
Ha
S(Ha), therefore
we obtain f

= f for all automorphisms H, with other words f

Fix(H)[T] = K[T] and thus, since the minimal polynomial p
a
K[T] of a
over K is a divisor of f because of f(a) = 0, we have
[E : K] = [K[a] : K] = deg(p
a
) deg(f) = [H[.
We continue the proof of 4.60: The Galois group acts on the set of inter-
mediate eld of [E : K] by (, L) (L), and on the set of subgroups of G
by conjugation (, H) H
1
, such that
Aut
(L)
(E) = Aut
L
(E)
1
.
Hence we can conclude that H = Aut
L
(E) G is a normal subgroup i
(L) = L for all automorphisms G. But that is equivalent to L K
139
being normal: For = we may refer to 4.40, while = is not dicult
either: Again according to 4.40 it is sucient to show that every irreducible
polynomial g K[T] with a zero a L is split over L. Since E K is
normal, that is true over E and we have to show N
E
(g) N
L
(g). So let
b N
E
(g). From 4.42.i) we know that there is an automorphism G =
Aut
K
(E) with (a) = b. Consequently b (L) = L.
In particular we see that in this situation the group homomorphism
G = Aut
K
(E) Aut
K
(L) , [
L
,
is well dened, since (L) = L for all automorphisms Aut
K
(E). Its
kernel is H := Aut
L
(E), and being surjective according to 4.35, it induces
an isomorphism G/H

= Aut
K
(L).
This nishes the proof of Theorem 4.60.
Problems 4.65. 1. R: Determine all intermediate elds of the extension Q[
4

2, i]
Q. Which of them are normal? Cf. Problem 4.46.2.
2. Let char(F) = p and F[X, Y ] := (F[X])[Y ] be the polynomial ring over F in the
variables X, Y and F(X, Y ) := Q(F[X, Y ]). Compute the degree [E : K] of the eld
extension E := F(X, Y ) K := F(X
p
, Y
p
) and show that there is no primitive
element a E, i.e. such that E = K[a] holds.
3. R: Let E K be a nite eld extension and p := char(K) > 0. An element
a E is called separable over K, if p
a
K[T] has only simple zeros and purely
inseparable over K, if there is an r N with a
p
r
K. We denote E
s
E resp.
E
in
E the set of all separable resp. purely inseparable elements. The extension
E K is called separable resp. purely inseparable i E = E
s
resp. E = E
in
.
(a) Show: A nite extension E K is galois, if [Aut
K
(E)[ = [E : K]. Hint:
If the latter is satised, we have E
in
= K (why?) and E can be written
E = K[a].
(b) Show: The sets E
s
, E
in
E are intermediate elds. (Hint: Assume rst
that E K is normal and use the fact that E
in
= K for a Galois extension
E K). The intermediate eld E
s
K is also called the separable hull of K
in E. Show that E E
s
is purely inseparable, and that E E
in
separable
for a normal extension E K.
4. Let E K be a Galois extension with Galois group G := Aut
K
(E). Show that the
trace Tr : E K, Tr(x) :=

G
(x) and the norm N : E

, N(x) :=

G
(x) dene homomorphisms between the additive resp. multiplicative groups
of E and K. Let a E and n := [E : K], s := [E : K[a]]. Show that (p
a
)
s
=
T
n
+Tr(a)T
n1
+... +(1)
n
N(a). Indeed (p
a
)
s
is the characteristic polynomial of
the multiplication
a
End
K
(E).
140
(a) Let p be a prime and K a eld with char(K) ,= p. Show: If a , K
p
, then the
polynomial f = T
p
a K[T] is irreducible. Hint: We may assume p > 2.
Let E K be the splitting eld of f. It is a Galois extension because of
char(K) ,= p. If f would be reducible, then n := [E : K] is not divisible with
p. Take b E with b
p
= a. Then N(b)
p
= N(b
p
) = N(a) = a
n
. Consequently
a
n
K
p
resp. a K
p
(why?), a contradiction.
(b) Let p be an odd prime and f
r
:= T
p
r
a K[T] with a eld K of characteristic
char(K) ,= p. Show: f
r
is irreducible i a , K
p
. Hint for the non-trivial
implication: According to a) the polynomial f := T
p
a K[T] is irreducible,
hence [K[b] : K] = p, where b
p
= a. If b is not a p-th power in K[b] we may use
the induction hypothesis and obtain that T
p
r1
b K[b][T] is irreducible,
resp. that [K[c] : K[b]] = p
r1
, if b = c
p
r1
. Altogether [K[c] : K] = p
r
, i.e.
T
p
r
a K[T] is irreducible. Otherwise take c K[b] with c
p
= b and let
E K[b] be the splitting eld of f with corresponding norm N : E

.
As in a) we get the Galois extension E K, where s := [E : K[b]] is not
divisible with p. Then with n := [E : K] we nd (1)
s
a
s
= (1)
n
N(b) =
(1)
n
N(c
p
) = (1)
n
N(c)
p
, and since p is odd and relatively prime to s, that
implies a K
p
, a contradiction!
(c) Take now p = 2 in b). Show: f
r
:= T
2
r
a K[T] is irreducible i a , K
2
, ,
4K
4
. Hint: Reason as before and exclude the possibility b = c
2
with some
c K[b].
(d) Show that the polynomial f := T
n
a K[T], where the exponent n is not
divisible with char(K), is irreducible, i a , K
p
for all primes p dividing n,
and if a , 4K
4
in case 4[n.
5. In the two last problems we investigate Galois extensions E K, whose Galois
group Aut
K
(E) =< > is cyclic of prime order and look for a primitive element
a E, i.e. E = K[a], with a minimal polynomial of standard form. The element
a E is characterized by the fact that (a) should be of a special form, either
(a) = a with a primitive p-th root of unity or (a) = a + 1.
(a) Let E K be a Galois extension, with its degree [E : K] = p being a prime
number. Show: If K contains a primitive p-th root of unity K (this
implies char(K) ,= p), we can write E = K[a] with an element a E, such
that b := a
p
K, or, with other words, the minimal polynomial p
a
K[T]
of a over K is p
a
= T
p
b. In that case E K is also called a simple
radical extension. Hint: The Galois group is cyclic and generated by any
automorphism ,= id
E
. The element a E can be found as an eigenvector of
the K-linear map : E E belonging to the eigenvalue K: In fact, it
has characteristic polynomial

= p

= f := T
p
1 K[T], since f() = 0
because of
p
=id
E
, while id
E
, , ...,
p1
are linearly independent according
to Problem 4.20.5.
(b) Let E K be a Galois extension with cyclic Galois group Aut
K
(E) =< >
and [E : K] = p = char(K). Show:There is an element a E with minimal
141
polynomial p
a
= T
p
T c K[T], cf. problems 4.46.1 and 4.58.2. Hint:
Consider := id
E
End
K
(E) ( is not an automorphism!). Show:

p
= 0 ,=
p1
- if already
p1
= 0 choose integers r, s with rp+s(p1) = 1
and conclude = 0. Hence dimker() = 1 resp. ker() = K E. But
being nilpotent, we have ker() (E) ,= 0 resp. K (E). Take now
a E with (a) = 1 resp. (a) = a + 1. Finally (a
p
a) = a
p
a and thus
c := a
p
a K.
4.7 The Fundamental Theorem of Algebra
As an application of the Fundamental Theorem of Galois Theory 4.60 we
show
Theorem 4.66. (Fundamental Theorem of Algebra) The eld C of all
complex numbers is algebraically closed.
Proof. According to Proposition 4.38 it suces to prove that there are no
non-trivial nite extensions E C. In any case the degree [E : C] of such
an extension is a 2-power: It is a divisor of [E : R] and there we have:
Proposition 4.67. The degree of a nite extension E R is a power of 2,
i.e., [E : R] = 2
r
with some r N.
Proof. We may assume that E R is normal: Otherwise there is an exten-
sion L E, such that L R is normal, see 4.36.1. But [E : R] is a divisor
of [L : R] by 4.12. So let us consider a normal extension E R. It is then
automatically galois, and we choose a 2-Sylow subgroup H G := Aut
R
(E).
Now it suces to show: F := Fix(H) = R, since that implies according to
4.60 H = G and [E : R] = [G[ = [H[ = 2
r
with some r N.
Because of [E : F] = [H[ and [G[ = [E : R] = [E : F][F : R], the extension
degree [F : R] is odd; in particular every element a F has a minimal
polynomial p
a
R[T] of odd degree, but on the other hand every polynomial
R[T] of odd degree has a real zero.
Thus the irreducible polynomial p
a
has degree 1 and a R. So we have seen
F R resp. F = R and are done.
Let us now go on with the proof of 4.66: The extension E C has degree [E :
C] = 2
s
with some s N. We show that for s 1 there is a subgroup H
G := Aut
C
(E) of index (G : H) = 2. Taking that for granted, Fix(H) C is
an extension of degree 2 and thus Fix(H) = C[a] with some element a whose
142
minimal polynomial p
a
C[T] has degree 2. But every quadratic polynomial
C[T] has a zero in C, in particular it is reducible. Contradiction!
Existence of the subgroup H G: We show that every p-group G has a
subgroup H of index (G : H) = p. If [G[ = p, take H := e. Let : G
G/Z(G) be the quotient projection. According to 2.32 we have [G/Z(G)[ <
[G[, and now may assume, that we already have found a subgroup H
0

G/Z(G) of index p. Finally take H :=
1
(H
0
).
Problems 4.68. 1. Show: If K has characteristic char(K) = 0 and E K is a nite
extension with an algebraically closed eld E, then either E = K or E = K[i] with
an element i E, i
2
= 1. Hint: Since i E, we may replace K with K[i] resp.
assume i K and have to show E = K. In any case E K is galois. Take a
prime number p dividing [E : K] = [Aut
K
(E)[ and an automorphism Aut
K
(E)
of order p and let F E be its xed eld. We then have [E : F] = p and all
non-linear irreducible polynomials F[T] have degree p. In particular the p-th
roots of unity belong to F - their minimal polynomial being of degree < p, while
all element F E have a minimal polynomial of degree p. According to Problem
4.65.5 c) we may write E = F[a], where b := a
p
F. In particular b , F
p
and,
for p = 2, nor b 4F
4
- otherwise a would be a square in F because of i F.
So the polynomial T
p
2
b F[T] is irreducible according to Problem 4.65.4 c).
Contradiction!
4.8 Cyclotomic Extensions
The nite eld F
p
n is the splitting eld of the polynomial T
p
n
1
1 F
p
[T]
and has cyclic Galois group. In this section we investigate the splitting eld
E
n
Q of f = T
n
1 Q[T]. Indeed, E
n
= Q[
n
] with
n
:= e
2i
n
, it is called
the n-th cyclotomic eld, since the n-th roots of unity 1,
n
, ...,
n1
n
divide
the unit circle into n sectors of the same size. (o = circle, =
to cut).
We determine rst the minimal polynomial p

n
Q[T] of
n
. Recall that

k
n
, k Z, only depends on the residue class

k Z
n
of k modulo n.
Denition 4.69. The n-th cyclotomic polynomial f
n
C[T] is the poly-
nomial
f
n
:=

kZ

n
(T
k
n
) , n 2,
while f
1
:= T 1.
Remark 4.70. 1. We have deg(f
n
) = (n) with Eulers -function, cf.
Example 3.8.5.
143
2. Let S
1
:= z C; [z[ = 1 be the unit circle. Then
f
n
=

aS
1
,ord(a)=n
(T a)
and thus:
T
n
1 =

aC
n
(T a) =

d|n
_

aS
1
,ord(a)=d
(T a)
_
=

d|n
f
d
.
Using that formula we can compute the cyclotomic polynomials in-
ductively, cf. Problem 3.46.4. For example for a prime p we obtain
T
p
1 = (T 1)f
p
, and the division algorithm for polynomials yields
f
p
= T
p1
+ T
p2
+ ... + T + 1 ,
while T
4
1 = f
1
f
2
f
4
= (T 1)(T + 1)f
4
, whence f
4
= T
2
+ 1 (of
course f
4
= (T i)(T + i) as well), and
T
6
1 = f
1
f
2
f
3
f
6
= (T 1)(T + 1)(T
2
+ T + 1)f
6
leads to f
6
= T
2
T +1. Eventually T
8
1 = (T 1)(T +1)(T
2
+1)f
8
,
such that f
8
= T
4
+ 1 etc..
In any case we see that the cyclotomic polynomials have integral coef-
cients:
f
n
Z[T] .
The n-th root of unity
n
being a zero of f
n
, its minimal polynomial p

n
is a divisor of f
n
. In fact
Proposition 4.71. The n-th cyclotomic polynomial f
n
Z[T] is irreducible.
In particular, it agrees with the minimal polynomial p

n
Q[T] of
n
= e
2i/n
,
i.e.
p

n
= f
n
.
Proof. Denote f := p

n
Q[T] the minimal polynomial of
n
over Q. Since
f[f
n
and f
n
only has simple zeros, it is sucient to show that f := p

n
Q[T]
and f
n
have the same zeros, i.e., we have to see that f(
k
n
) = 0 for all k N
relatively prime to n.
144
Writing a given k as a product of primes, we see that we are done if for
all a C
n
and primes p not dividing n, we can prove the implication:
(3) f(a) = 0 =f(a
p
) = 0 .
The polynomial f is a divisor of T
n
1, say T
n
1 = fh. Gau lemma
3.40 tells us 1 = cont(f)cont(h), but with f also h is a monic polynomial, and
the content of a monic polynomial is a number of the form 1/m, m N
>0
.
Hence necessarily cont(f) = cont(h) = 1, in particular f, h Z[T].
So let us x a prime p with p ,[n. Assume f(a) = 0 and f(a
p
) ,= 0.
Since a C
n
, we see 0 = (a
n
)
p
1 = (a
p
)
n
1 = f(a
p
)h(a
p
). Thus, if not
f(a
p
) = 0, we must have h(a
p
) = 0. That can also be formulated in a more
sophisticated way, by saying: The polynomial h(T
p
) has a as a zero. But then
f, being the minimal polynomial of a, divides h(T
p
) in Q[T] resp. Z[T]. For
the modulo p reduced polynomials

f,

h(T
p
) =

h
p
Z
p
[T] we have the same
divisibility relation

f[

h(T
p
) =

h
p
. But that implies that every zero b N
F
(

f)
of

f Z
p
[T] in some extension F Z
p
is also a zero of

h
p
resp. of

h itself.
We choose F as splitting eld of the polynomial g := T
n
1 =

f

h Z
p
[T].
In F there is of course such a zero b, which then is at least a double zero of
g. But since g

= nT
n1
and p ,[n, we have g

(b) ,= 0. Contradiction.
Let us now determine the Galois group of Q[
n
] Q:
Theorem 4.72. For the cyclotomic extension Q[
n
] Q we have
1. Its degree satises [Q[
n
] : Q] = (n) with Eulers -function :
N
>0
N.
2. Let C
n
Q[
n
] denote the group of all n-th roots of unity. Then the
restriction map
Aut
Q
(Q[
n
]) Aut(C
n
), [
C
n
is an isomorphism. Here Aut(C
n
)

= Z

n
is the automorphism group of
the group C
n
. So altogether
Aut
Q
(Q[
n
])

= Z

n
.
145
Proof. The degree equality is an immediate consequence of 4.71, while the
restriction map in the second part is injective, an automorphism being
determined by its value (
n
), and surjective, since the Galois group acts
transitively on the roots of the irreducible polynomial f
n
. Finally remember
the isomorphism
Z

=
Aut(C
n
), k p
k
with the k-th power map p
k
: C
n
C
n
, a a
k
.
Remark 4.73. Since the automorphism group of a cyclotomic extension is
abelian, every intermediate eld L of Q[
n
] Q provides a Galois extension
L Q with abelian Galois group. A deep result of Leopold Kronecker
(1823-1891) assures that (up to isomorphy) every Galois extension E Q
with abelian Galois group Aut
Q
(E) is obtained in that way. We only discuss
an example:
Example 4.74. We consider the real part of the n-th cyclotomic eld:
L
n
:= Q[
n
] R = Fix() = Q[a]
with the complex conjugation : z z and a :=
n
+
n
= 2 cos(
2
n
). (Note
that [
C
n
= p
1
.)
We nd [Q[
n
] : L
n
] = 2, since
n
is a root of the quadratic polynomial
T
2
aT +1 L
n
[T], and thus [L
n
: Q] = (n)/2 for n > 2. Obviously L
n

Q[a], while equality follows from the fact, that even T
2
aT +1 (Q[a])[T].
The case n = 9 has already been discussed, see Example 4.19.2).
Problems 4.75. 1. Let E Q be the splitting eld of the polynomial T
n
a, where
a Q

. Show that Q[
n
] E and Aut(E) := Aut
Q
(E) is isomorphic to a sub-
group of the semidirect product C
n

n
with the homomorphism : Z

n

Aut(C
n
), k p
k
. Let now n = p be a prime number and a Z not a p-th power.
Show that in this case Aut(E)

= C
n

n
. Hint: The polynomial T
p
a is irre-
ducible over Q, cf. Problem 4.65.4. Conclude that p divides [E : Q] = (n)k and
thus p[k because of gcd(p, (p)) = 1 resp. even p = k.
2. Show that, with the notation of the previous problem: C
n

= A(Z
n
), cf.
Problem 3.9.8.
146
4.9 Solvability by Radicals
In the nal section we explain how Galois theory leads to a solution of our
problem to decide when a polynomial equation over a eld K of characteristic
char(K) = 0 can be solved by radicals. First of all we have to make precise
that notion. We begin with
Denition 4.76. An extension E K is called a simple radical exten-
sion, if it can be written E = K[a] with a primitive element a E, such
that a
m
K for some exponent m > 0.
Remark 4.77. Given a simple radical extension, the polynomial T
m
b
K[T] with b = a
m
is not necessarily the minimal polynomial p
a
K[T] of a
over K (even if m > 0 is minimal with a
m
K). But in any case p
a
is of
course a divisor of T
m
b.
Example 4.78. Let
m
C be the m-th root of unity
m
:= exp(
2i
m
). Then
Q[
m
] Q is a simple radical extension. Another example is Q[
3

2] Q.
Let us now discuss the automorphism group Aut
K
(E) for a simple radical
extension E = K[a] K with b = a
m
K.
Proposition 4.79. The automorphism group Aut
K
(E) of a simple radical
extension E = K[a] K (with, say, a
m
K) is solvable.
Proof. Consider the group C
m
(E) E

of m-th roots of unity in E, a cyclic

group (cf. Prop. 4.47) whose order := [C
m
(E)[ divides m (with = m
for char(K) = 0) and the intermediate eld L = K[C
m
(E)], obtained from
K by adjoining the elements in C
m
(E), the splitting eld of the polynomial
T
m
1. Then the normal series
id
E
Aut
L
(E) Aut
K
(E)
has abelian factors: The automorphism group Aut
L
(E) is isomorphic to a
subgroup of C
m
(E), the group homomorphism
Aut
L
(E) C
m
(E), (a)a
1
being injective. On the other hand
Aut
K
(E)/Aut
L
(E)

= Aut
K
(L) Aut(C
m
(E)), [
C
m
(E)
,
with the abelian group Aut(C
m
(E))

= Z

, where a residue class k corre-

sponds to the k-th power map p
k
: C
m
(E) C
m
(E), a a
k
.
147
Denition 4.80. An extension E K is called a radical extension, if
it is the composite of nitely many simple radical extensions E
i
E
i1
, i =
1, ..., r, (often also called a tower):
K = E
0
E
1
... E
r
= E .
Eventually we can dene solvability by radicals:
Denition 4.81. Let f K[T] be an irreducible polynomial. We say that
the equation f(x) = 0 is solvable with radicals, if there is a radical extension
L K, such that f has a zero in L.
Example 4.82. As we shall see later on the radical extension L K can
always be taken as an extension L E of the splitting eld E K of
f K[T], but it may happen that we can not actually choose L = E. For
example consider f = T
3
3T + 1 Q[T]. Its splitting eld is E := Q[a]
with a :=
9
+
1
9
= 2 cos(
2
9
) and we choose L := Q[
9
] E; so the
equation f(x) = 0 is solvable with radicals. Now the extension E Q has
degree [E : Q] = 3 and hence no proper intermediate elds, and it is itself
not a simple radical extension: We have E R and thus [C
m
(E)[ 2 and
Aut(C
m
(E)) = id; so, if E Q would be a simple radical extension, its
automorphism group had at most two elements. But we have already seen
that it is cyclic of order 3.
Here is the central result characterizing polynomials solvable by radicals:
Theorem 4.83. Let char(K) = 0 and f K[T] be an irreducible polynomial.
Then the equation f(x) = 0 is solvable with radicals, i its Galois group, i.e.
the automorphism group Aut
K
(E) of its splitting eld E K, is solvable.
Proof. =: Let n := [E : K] and L E be the splitting eld of the
polynomial T
n!
1 E[T]. Then L K is the splitting eld of (T
n!
1)f
K[T] and thus a Galois extension because of char(K) = 0. Furthermore
its Galois group Aut
K
(L) is solvable, since both, Aut
E
(L) Aut
K
(L) and
Aut
K
(L)/Aut
E
(L)

= Aut
K
(E) are solvable: The extension L E is a
simple radical extension: L = E[] with a primitive n!-th root of unity ,
and Aut
K
(E) is solvable by assumption. We take now L
0
:= K, L
1
:= K[].
The Galois group Aut
L
1
(L) Aut
K
(L) is also solvable. We have now
148
[L : L
1
] [E : K], since we may write E = K[a] with some primitive
element a E. Then L = L
1
[a], and the minimal polynomial p
a
K[T] of
a over K is divisible with the minimal polynomial q
a
L
1
[T] of a over L
1
.
And [E : K] = deg(p
a
) as well as [L : L
1
] = deg(q
a
).
It remains to show that L L
1
is a radical extension. In order to simplify
notation we replace L
1
with K and L with E, where we may assume that
C
n
(K) with n := [E : K] has order n, i.e. all n-th roots of unity lie already
in the base eld K. Take now a normal series
G
0
= G := Aut
K
(E) G
1
... G
r
:= Aut
E
(E) = id
E

with cyclic factors G

i
/G
i+1
of prime order and consider the corresponding
tower of intermediate elds
K = L
0
L
1
.... L
r
= E .
Then L
i+1
L
i
is a Galois extension with cyclic Galois group Aut
L
i
(L
i+1
)

=
G
i
/G
i+1
of prime order. The extension degree p := [L
i+1
: L
i
] divides n =
[E : K] and hence C
p
(L
i
) C
n
(L
i
) = C
n
(K) has order p, i.e., all p-th roots
of unity belong to L
i
. Problem 4.65.5 a) with K = L
i
and E = L
i+1
gives
us, that we really have a simple radical extension.
=: We take a radical extension L K, such that f has a zero in L and
construct an extension F L, such that F K is a Galois extension with
solvable Galois group Aut
K
(F). This yields the result, since the splitting
eld E K of f then is isomorphic to an intermediate eld of F K. Take
a tower
L
0
:= K L
1
... L
r
= L
of simple radical extensions L
i+1
L
i
, say L
i
= L
i1
[a
i
], where b
i
:= a
m
i
i

L
i1
. Let m := m
1
... m
r
. Then we take F K as splitting eld of
(T
m
1)g K[T], where g = p
a
is the minimal polynomial over K of some
primitive element a L for the eld extension L K, i.e., L = K[a]. Hence
we may regard L as intermediate eld of the extension F K. On the other
hand
F = K[, (a); Aut
K
(F)] = K[, (a
i
); i = 1, ...., r, Aut
K
(F)]
with a primitive m-th root of unity F. Let
Aut
K
(F) =
1
:= id
F
,
2
, ...,
n
.
149
Now consider the tower
F
0
:= K F
1
:= K[] F
2
:= F
1
[a
2
] ... F
1+nr
with
F
1+(j1)r+i
:= K[, a
1
, ..., a
r
,
2
(a
1
), ....,
j1
(a
r
),
j
(a
1
), ...,
j
(a
i
)] ,
where j = 1, ..., n and i = 1, ..., r. Then it is sucient to show that F

F
1
is a Galois extension with abelian Galois group: That extension is the
splitting eld of T
m
1 F
0
[T] = K[T] for = 1 and of T
m
i

j
(b
i
) F
1
[T]
for = 1 +(j 1)r +i, since F
1
contains all m
i
-th roots of unity. As above
we see that Aut
F
1
(F

) C
m
i
(F
1
) is cyclic.
5 Annex: Zorns Lemma
If in algebra innite or even uncountable sets are involved, it can be useful to
know about the existence of certain objects even if there is no constructive
method to create them: A generally accepted tool in this context is Zorns
lemma, which we shall discuss in this annex.
Denition 5.1. A partial order on a set M is a relation _, which is
reexive, antisymmetric and transitive, i.e.
1. x M : x _ x ,
2. x, y M : x _ y y _ x =x = y and
3. x, y, z M : x _ y y _ z =x _ z.
Such a relation is sometimes simply called an order (relation) on M. A
total or linear order is a partial order, where any two elements x, y M
are related:
x, y M : x _ y y _ x .
A well ordering on M is a linear order, such that every non-empty subset
M
0
M has a rst element (with respect to _), i.e.,
M
0
M a M
0
: x M
0
: a _ x .
An element a M is called maximal (w.r.t. the order _), i
x M : a _ x =a = x,
i.e., there are no elements bigger than a.
150
Example 5.2. In many applications the set M is realized as a subset M
T(U) of the power set of some set U (the universe) with the inclusion as
order relation
A _ B A B.
Remark 5.3. 1. If x _ a for all x M, the element a is obviously
maximal, but in general that need not hold for a maximal element: It
is allowed for a maximal element a M, that there are elements in M
not related to a.
2. The set N = 0, 1, 2, ... of all natural numbers, endowed with the
natural order, is well ordered, but Z, Q and R are not. The set N
2
,
endowed with the lexicographic order
(x, y) _ (x

, y

) x < x

(x = x

and y y

) .
is well ordered. A subset of a well ordered set has by denition a unique
rst element, but in general no last element, and every element has an
immediate successor - the rst element of the set of all elements after
the given one, but not necessarily an immediate predecessor. An initial
segment M
0
of a linearly ordered set M is a subset M
0
M satisfying
M y _ x M
0
=y M
0
, i.e., with an element x M
0
all elements
y _ x before x belong to M
0
. If M is well ordered, such an initial
segment satises either M
0
= M or M
0
= M
a
:= x M; x a.
Namely, given an initial segment M
0
,= M, choose a as the rst element
in the complement M M
0
.
Theorem 5.4. (Zorns lemma) (Max August Zorn, 1906-1993): Let M be
a set with the partial order _. If for every (w.r.t. _) linearly ordered subset
T M there is an upper bound b M, i.e. such that t _ b for all t T
(written briey as T _ b), then there are maximal elements in M.
Example 5.5. If M T(U) as in Example 5.2, the upper bound B of
a linearly ordered subset T M T(U) usually is taken as the union
B :=

AT
A of all sets A T, and it remains to check that in fact B M.
Before we prove Zorns lemma we present the most important applications.
The rst one is basic for Linear Algebra:
Theorem 5.6. Every K-vector space V has a basis.
151
Proof. Take M T(V ) as the set of all linearly independent subsets of V .
We can apply Zorns lemma as in Example 5.5. So there is a maximal linearly
independent set B M, indeed B is a basis: We have to show that any vector
v V is a nite linear combination of vectors in B. So let v V . If v B,
we are done, otherwise B v , M the set B M being maximal in M
and hence there is a non-trivial relation
0 = v +
1
v
1
+ ... +
r
v
r
with v
1
, ..., v
r
B and ,
1
, ...,
r
K. But ,= 0, since the vectors v
1
, ..., v
r
are linearly independent, i.e., we may solve for v V .
Theorem 5.7. Every proper ideal a R in a (commutative) ring (with 1)
is contained in a maximal ideal m R.
Proof. Take M T(R) as the subset of all proper ideals in R containing a.
Since an ideal a is proper i 1 , a, it is obvious that the union of a linearly
ordered set of proper ideals again is a proper ideal.
As a corollary of Theorem 5.7 we obtain:
Theorem 5.8. Every eld K has an algebraic closure E K, and any two
algebraic closures are isomorphic as K-extensions.
Proof. Existence: First we make the eld K perfect, if char(K) = p > 0:
The pair (E := K, ) with the Frobenius homomorphism : K E = K
denes a eld extension
p

K K, where every element x K has a p-

th root. We may iterate that procedure and obtain after n steps the eld
p
n

K K. Since for m n there is a unique (injective) morphism

p
n

K
p
m

K of K-extensions, we may treat it is an inclusion and dene

K

:=

_
n=0
p
n

K ,
which obviously is a perfect eld.
A more explicit construction of K

is as follows:
K

:=
_

_
n=0
K n
_
/ ,
152
where for m n, we have (x, n) (y, m) i y =
mn
(x), and K K

is given by x [(x, 0)] :=the equivalence class of (x, 0). We leave it to the
reader to dene the addition and multiplication of equivalence classes. Thus
it remains to nd an algebraic closure of K

.
So, from now on we may assume that K is perfect. Then we construct a
eld E K, such that every irreducible polynomial f K[T] has a zero in
E and use
Proposition 5.9. Let E K be an algebraic extension of the perfect eld
K, such that every irreducible polynomial f K[T] has a zero in E. Then
E K is an algebraic closure of K.
Proof. We have to show that every irreducible polynomial f K[T] is split
over E. Consider a splitting eld L K of f. Since K is perfect, we can
write L = K[a] according to the Primitive Element Theorem 4.63. But the
minimal polynomial p
a
K[T] is irreducible and thus has a zero b E.
Therefore f is split over E K[b ]

= L.
The extension E K satisfying the assumptions of Proposition 5.9 is
obtained as follows: Index the irreducible monic polynomials K[T] as
f

, A, with some index set A, and consider the polynomial ring K[T
A
] of
all polynomials in the variables T

, A, every individual polynomial de-

pending only on nitely many variables. To be more precise: The polynomial
ring K[T
1
, ..., T
n
] may be dened inductively:
K[T
1
, ..., T
n+1
] := (K[T
1
, ..., T
n
])[T
n+1
].
Now for a nite subset A
0
A, say A
0
=
1
, ...,
n
, we set
K[T
A
0
] := K[T

1
, ..., T

n
] ,
and nally
K[T
A
] :=
_
A
0
A, |A
0
|<
K[T
A
0
] .
Now let
g

:= f

(T

) K[T

] K[T
A
],
i.e. every polynomial f

gets its own variable T

! Let now a K[T

A
] be the
ideal generated by the g

, i.e.,
a =
_
r

i=1
h

i
g

i
; h

i
K[T
A
], r N,
1
, ...,
r
A
_
.
153
Indeed a is a proper ideal: Otherwise we can write
1 =
n

i=1
h

i
g

i
.
Now take a nite set A
0
A with h

i
, g

i
K[T
A
0
] for i = 1, ..., n. In order
to simplify notation write h
i
, g
i
instead of h

i
, g

i
and A
0
= 1, ..., m with
some m n. So we have the equality
1 =
n

i=1
h
i
g
i
in the ring K[T
1
, ..., T
m
]. Now consider a splitting eld F K of f := f
1
...f
n
and substitute x = (x
1
, ..., x
n
, 0, ..., 0) F
m
, where x
i
F is a zero of f
i
:
Since g
i
(x) = f
i
(x
i
) = 0 for i = 1, ..., n, we obtain that 1 = 0 holds in F.
Contradiction!
Eventually Theorem 5.7 provides a maximal ideal m a, and we may
set E := K[T
A
]/m. Obviously K K[T
A
]/m is a eld extension, where
f

K[T] has the zero x

:= T

+ m. In particular E K is algebraic.
Uniqueness: Let E K and F K be two algebraic closures. Consider
the set M of all pairs (L, ), where L is an intermediate eld of E K and
: L F a morphism of K-extensions; furthermore we dene
(L, ) _ (L

) L L

and =

[
L
.
Let now (L
i
,
i
), i I be a linearly ordered subset. The upper bound we
are looking for can be taken as (L

, ), where
L

:=
_
iI
L
i
, [
L
i
:=
i
.
According to Zorns lemma there is a maximal element (L, ) and it remains
to show L = E and (E) = F. If L ,= E, take an element a E L, denote
p
a
L[T] its minimal polynomial over L. Since F is algebraically closed, the
polynomial p

a
(L)[T] has a zero b F. Then we have (L[a], ) ~ (L, ),
if [
L
= , (a) = b. So necessarily L = E. But E being algebraically closed,
(E) is algebraically closed as well, in particular (E) has no non-trivial
nite or algebraic extensions, i.e. F = (E).
154
Proof of Th.5.4. We assume that there is no maximal element in M, but
that every linearly ordered subset T M has an upper bound (T) M.
So there is a function
: Lin(M) M
from the set Lin(M) T(M) of all subsets linearly ordered with respect to
_, such that T _ (T). We may even assume that (T) is a strict upper
bound: T (T) or, equivalently (T) , T. If only (T) T is possible,
the element (T) would be a maximal element for the entire set M.
Then we use the function in order to produce recursively a linearly
ordered, indeed even well ordered, subset not admitting an upper bound,
contrary to our hypothesis. We take x
1
:= () as its rst element. If
x
1
, ..., x
n
are found one denes x
n+1
:= (x
1
, ..., x
n
). In this way we obtain
a sequence (x
n
)
nN
with x
1
x
2
..., but the chain x
n
; n N can be
extended further: Take y
1
:= (x
n
; n N), y
2
:= (y
1
, x
n
; n N).
In order to make sure that this idea really works, we introduce the concept
of a -chain: We shall call a subset K M a -chain, if (K, _) is well
ordered and for any y K the initial segment K
y
:= x K; x y
satises
y = (K
y
) .
We shall see that given two -chains K, L one of them is an initial segment
of the other. Taking this for granted the set
T :=
_
K -chain
K
is obviously a maximal -chain. On the other hand

T := T (T) is
-chain as well, so

T T resp. (T) T a contradiction!
It remains to show that of two -chains K, L one is an initial segment of
the other: Denote K
0
= L
0
the union of all sets which are initial segments
of both K and L. Obviously it is an initial segment of both K and L. If
K
0
= K or L
0
= L, we are done; otherwise K
0
= K
a
and L
0
= L
b
. In that
case we have
a = (K
a
) = (L
b
) = b L ,
i.e., K
a
= L
b
is an initial segment of both K and L, a contradiction!
155
Remark 5.10. The above proof of Zorns lemma is a naive one. The most
problematic part is the existence of the function
: Lin(M) M,
since in general there is no recipe for an explicit construction, the set Lin(M)
being quite big. Instead one has to derive it from the
Axiom of Choice: Given a family (A
i
)
iI
of pairwise disjoint subsets A
i

M of a set M, there is a set A M containing precisely one element out of
each set A
i
, i I, i.e., it has the form A = x
i
; i I with x
i
A
i
for all
i I.
The axiom of choice, though looking quite harmless, has striking conse-
quences, as for example the fact, that every set admits a well ordering, cf.
Problem 5.11.4. Indeed, no human being has up to now succeeded in well
ordering the set of all real numbers.
Problems 5.11. 1. Let S R 0 be a multiplicative subset in the ring R. Show:
There is a maximal ideal a R in the set of all ideals not intersecting S. If S = 1,
it is a maximal ideal in the sense of Def. 3.23, but otherwise not necessarily. But
in any case it is a prime ideal!
2. Show that

pR prime ideal
p =
_
0 .
Hint: Use Problem 4.75.1! Here we denote
_
0 := x R; n N : x
n
= 0 the
nilradical of the ring R.
3. Let K C be a subeld. Show: Every automorphism : K K can be extended
to an automorphism : C C! Note that the corresponding statement for R
instead of C is wrong! (Cf. Problem 4.20.3)
4. Show: Every set M admits a well ordering. Hint: Choose a map : T(M)M
M with (A) M A for all proper subsets A M. Then argue as in the proof of
Zorns lemma!
References
[1] Cohn, Paul Moritz: Classic Algebra, John Wiley & Sons, 2000.
156
[2] Ebbinghaus, Hans Dieter; Hermes, Hans; Hirzebruch, Friedrich;
Koecher, Max; Mainzer, Klaus; Prestel, Alexander and Rem-
mert, Reinhold: Zahlen, Springer 1983.
[3] Gorenstein, Daniel; Lyons, Richard and Solomon, Ronald: The
Classication of the Finite Simple Groups, Math. Surveys and Monographs,
AMS, Providence, Rhode Island.
[4] Grillet, Pierre Antoine: Abstract Algebra, GTM 242, Springer, 2006.
[5] Hasse, Helmut: Hohere Algebra I/II, Sammlung Goschen Bd.931/932,
1969/1967.
[6] Lang, Serge: Algebra, Addison Wesley 1965.
[7] Reiffen, Hans-J org; Scheja, G unter and Vetter, Udo: Algebra, BI
Hochschultaschenb ucher 110/110a.
[8] Ringel, Claus-Michael: Leitfaden zur Algebra I, SS 2001, Homepage Uni-
versitat Bielefeld.
[9] Rotman, Joseph: Galois Theory, Springer 1998.
[10] Scharlau, Winfried, and Opolka, Hans: Von Fermat bis Minkowski,
Springer 1980.
[11] van der Waerden, Bartel Ludwig: Algebra I/II; Springer, Heidelberger
Taschenb ucher 12/13 (8th edition of his classical book Moderne Algebra).
[12] Willems, Wolfgang: Codierungstheorie; Walter de Gruyter 1999.
[13] Zariski, Oscar and Samuel, Pierre: Commutative Algebra I/II, GTM
28/29, Springer.
157