Fields and Galois Theory

J.S. Milne
Version 4.50
March 18, 2014
These notes give a concise exposition of the theory of elds, including the Galois theory
of nite and innite extensions and the theory of transcendental extensions. The rst six
chapters form a standard course, and the nal three chapters are more advanced.
BibTeX information
@misc{milneFT,
author={Milne, James S.},
title={Fields and Galois Theory (v4.50)},
year={2014},
note={Available at www.jmilne.org/math/},
pages={138}
}
v2.01 (August 21, 1996). First version on the web.
v2.02 (May 27, 1998). Fixed about 40 minor errors; 57 pages.
v3.00 (April 3, 2002). Revised notes; minor additions to text; added 82 exercises with
solutions, an examination, and an index; 100 pages.
v3.01 (August 31, 2003). Fixed many minor errors; no change to numbering; 99 pages.
v4.00 (February 19, 2005). Minor corrections and improvements; added proofs to the
section on innite Galois theory; added material to the section on transcendental
extensions; 107 pages.
v4.10 (January 22, 2008). Minor corrections and improvements; added proofs for Kummer
theory; 111 pages.
v4.20 (February 11, 2008). Replaced Maple with PARI; 111 pages.
v4.21 (September 28, 2008). Minor corrections; xed problem with hyperlinks; 111 pages.
v4.22 (March 30, 2011). Minor changes; changed T
E
Xstyle; 126 pages.
v4.30 (April 15, 2012). Minor xes; added sections on etale algebras; 124 pages.
v4.40 (March 20, 2013). Minor xes and additions; 130 pages.
v4.50 (March 18, 2014). Added chapter explaining Grothendiecks approach to Galois
theory (Chapter 8) and made many minor improvements; numbering has changed;
138 pages.
Available at www.jmilne.org/math/
Please send comments and corrections to me at the address on my web page.
The photograph is of Sabre Peak, Moraine Creek, New Zealand.
Copyright c _1996, 1998, 2002, 2003, 2005, 2008, 2011, 2014 J.S. Milne.
Single paper copies for noncommercial personal use may be made without explicit permission
from the copyright holder.
Contents
Contents 3
Notations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1 Basic Denitions and Results 7
Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
The characteristic of a eld . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Review of polynomial rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Factoring polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Extension elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
The subring generated by a subset . . . . . . . . . . . . . . . . . . . . . . . . . 14
The subeld generated by a subset . . . . . . . . . . . . . . . . . . . . . . . . . 15
Construction of some extension elds . . . . . . . . . . . . . . . . . . . . . . . 15
Stem elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Algebraic and transcendental elements . . . . . . . . . . . . . . . . . . . . . . . 17
Transcendental numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Constructions with straight-edge and compass. . . . . . . . . . . . . . . . . . . . 20
Algebraically closed elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2 Splitting Fields; Multiple Roots 27
Maps from simple extensions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Splitting elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Multiple roots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3 The Fundamental Theorem of Galois Theory 35
Groups of automorphisms of elds . . . . . . . . . . . . . . . . . . . . . . . . . 35
Separable, normal, and Galois extensions . . . . . . . . . . . . . . . . . . . . . 37
The fundamental theorem of Galois theory . . . . . . . . . . . . . . . . . . . . . 39
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Constructible numbers revisited . . . . . . . . . . . . . . . . . . . . . . . . . . 43
The Galois group of a polynomial . . . . . . . . . . . . . . . . . . . . . . . . . 44
Solvability of equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4 Computing Galois Groups 47
When is G
(

n
? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3
When is G
(
transitive? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Polynomials of degree at most three . . . . . . . . . . . . . . . . . . . . . . . . 49
Quartic polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Examples of polynomials with S
]
as Galois group over Q . . . . . . . . . . . . . 52
Finite elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Computing Galois groups over Q . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5 Applications of Galois Theory 59
Primitive element theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Fundamental Theorem of Algebra . . . . . . . . . . . . . . . . . . . . . . . . . 61
Cyclotomic extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Dedekinds theorem on the independence of characters . . . . . . . . . . . . . . 65
The normal basis theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Hilberts Theorem 90 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Cyclic extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Kummer theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Proof of Galoiss solvability theorem . . . . . . . . . . . . . . . . . . . . . . . . 74
Symmetric polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
The general polynomial of degree n . . . . . . . . . . . . . . . . . . . . . . . . 77
Norms and traces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6 Algebraic Closures 85
Zorns lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
First proof of the existence of algebraic closures . . . . . . . . . . . . . . . . . . 86
Second proof of the existence of algebraic closures . . . . . . . . . . . . . . . . 86
Third proof of the existence of algebraic closures . . . . . . . . . . . . . . . . . 87
(Non)uniqueness of algebraic closures . . . . . . . . . . . . . . . . . . . . . . . 88
Separable closures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
7 Innite Galois Extensions 91
Topological groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
The Krull topology on the Galois group . . . . . . . . . . . . . . . . . . . . . . 92
The fundamental theorem of innite Galois theory . . . . . . . . . . . . . . . . . 95
Galois groups as inverse limits . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Nonopen subgroups of nite index . . . . . . . . . . . . . . . . . . . . . . . . . 99
8 The Galois theory of etale algebras 101
Review of commutative algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

Etale algebras over a eld . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

Classication of etale algebras over a eld . . . . . . . . . . . . . . . . . . . . . 104
Comparison with the theory of covering spaces. . . . . . . . . . . . . . . . . . . 107
9 Transcendental Extensions 109
Algebraic independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Transcendence bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
L uroths theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Separating transcendence bases . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Transcendental Galois theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
A Review Exercises 119
B Two-hour Examination 125
C Solutions to the Exercises 127
Index 137
5
Notations.
We use the standard (Bourbaki) notations:
N ={0. 1. 2. . . .].
Z =ring of integers,
R =eld of real numbers,
C =eld of complex numbers,
F
]
=Z,Z =eld with elements, a prime number.
Given an equivalence relation, +| denotes the equivalence class containing +. The cardinality
of a set S is denoted by [S[ (so [S[ is the number of elements in S when S is nite). Let 1
and be sets. A family of elements of indexed by 1, denoted by (a
i
)
iJ
, is a function
i a
i
: 1 . Throughout the notes, is a prime number: =2. 3. 5. 7. 11. . . ..
X Y X is a subset of Y (not necessarily proper).
X
def
=Y X is dened to be Y , or equals Y by denition.
X ~Y X is isomorphic to Y .
X .Y X and Y are canonically isomorphic (or there is a given or unique isomorphism).
PREREQUISITES
Group theory (for example, GT), basic linear algebra, and some elementary theory of rings.
References.
Jacobson, N., 1964, Lectures in Abstract Algebra, Volume III Theory of Fields and Galois
Theory, van Nostrand.
Also, the following of my notes (available at www.jmilne.org/math/).
GT Group Theory, v3.13, 2013.
ANT Algebraic Number Theory, v3.05, 2013.
CA A Primer of Commutative Algebra, v3.00, 2013.
A reference monnnn is to http://mathoverflow.net/questions/nnnn/
PARI is an open source computer algebra system freely available here.
ACKNOWLEDGEMENTS
I thank the following for providing corrections and comments for earlier versions of the
notes: Mike Albert, Lior Bary-Soroker, Maren Baumann, Leendert Bleijenga, Tommaso
Centeleghe, Sergio Chouhy, Demetres Christodes, Antoine Chambert-Loir, Dustin Clausen,
Keith Conrad, Hardy Falk, Jens Hansen, Albrecht Hess, Philip Horowitz, Trevor Jarvis,
Henry Kim, Martin Klazar, Jasper Loy Jiabao, Dmitry Lyubshin, Geir Arne Magnussen,
John McKay, Georges E. Melki, Courtney Mewton, Shuichi Otsuka, Dmitri Panov, Alain
Pichereau, David G. Radcliffe, Roberto La Scala, Chad Schoen, Prem L Sharma, Dror
Speiser, Bhupendra Nath Tiwari, Mathieu Vienney, Martin Ward (and class), Xiande Yang,
W Xu, and others.
6
CHAPTER 1
Basic Denitions and Results
Rings
A ring is a set 1 with two composition laws and such that
(a) (1. ) is a commutative group;
(b) is associative, and there exists
1
an element 1
T
such that a 1
T
=a =1
T
a for all
a 1:
(c) the distributive law holds: for all a. b. c 1,
(ab) c =a c b c
a (b c) =a b a c.
We usually omit and write 1 for 1
T
when this causes no confusion. If 1
T
= 0, then
1 ={0].
A subring S of a ring 1 is a subset that contains 1
T
and is closed under addition, passage
to the negative, and multiplication. It inherits the structure of a ring from that on 1.
A homomorphism of rings : 1 1
t
is a map such that
(ab) =(a) (b). (ab) =(a)(b). (1
T
) =1
T
0
for all a. b 1. A ring 1 is said to be commutative if multiplication is commutative:
ab =ba for all a. b 1.
A commutative ring is said to be an integral domain if 1
T
= 0 and the cancellation law
holds for multiplication:
ab =ac, a =0, implies b =c.
An ideal 1 in a commutative ring 1is a subgroup of (1. ) that is closed under multiplication
by elements of 1:
r 1, a 1, implies ra 1.
The ideal generated by elements a
1
. . . . . a
n
is denoted by (a
1
. . . . . a
n
). For example, (a) is
the principal ideal a1.
We assume that the reader has some familiarity with the elementary theory of rings.
For example, in Z (more generally, any Euclidean domain) an ideal 1 is generated by any
smallest nonzero element of 1.
1
We follow Bourbaki in requiring that rings have a 1, which entails that we require homomorphisms to
preserve it.
7
8 1. BASIC DEFINITIONS AND RESULTS
Fields
DEFINITION 1.1 A eld is a set J with two composition laws and such that
(a) (J. ) is a commutative group;
(b) (J

. ), where J

=J {0], is a commutative group;

(c) the distributive law holds.
Thus, a eld is a nonzero commutative ring such that every nonzero element has an inverse.
In particular, it is an integral domain. A eld contains at least two distinct elements, 0 and 1.
The smallest, and one of the most important, elds is F
2
=Z,2Z ={0. 1].
A subeld S of a eld J is a subring that is closed under passage to the inverse. It
inherits the structure of a eld from that on J.
LEMMA 1.2 A nonzero commutative ring 1 is a eld if and only if it has no ideals other
than (0) and 1.
PROOF. Suppose 1 is a eld, and let 1 be a nonzero ideal in 1. If a is a nonzero element
of 1, then 1 =a
-1
a 1, and so 1 =1. Conversely, suppose 1 is a commutative ring with
no proper nonzero ideals. If a =0, then (a) =1, and so there exists a b in 1 such that
ab =1.
2
EXAMPLE 1.3 The following are elds: Q, R, C, F
]
=Z,Z ( prime).
A homomorphism of elds : J J
t
is simply a homomorphism of rings. Such a
homomorphism is always injective, because its kernel is a proper ideal (it doesnt contain 1),
which must therefore be zero.
The characteristic of a eld
One checks easily that the map
Z J. n 1
T
1
T
1
T
(n copies).
is a homomorphism of rings. For example,
(1
T
1
T

n
) (1
T
1
T

n
) =1
T
1
T

nn
because of associativity. Therefore its kernel is an ideal in Z.
CASE 1: The kernel of the map is (0), so that
n 1
T
=0 == n =0 (in Z).
Nonzero integers map to invertible elements of J under n n 1
T
: Z J, and so this map
extends to a homomorphism
m
n
(m 1
T
)(n 1
T
)
-1
: QJ.
Thus, in this case, J contains a copy of Q, and we say that it has characteristic zero.
Review of polynomial rings 9
CASE 2: The kernel of the map is =(0), so that n 1
T
=0 for some n =0. The smallest
positive such n will be a prime (otherwise there will be two nonzero elements in J whose
product is zero), and generates the kernel. Thus, the map n n 1
T
: Z J denes an
isomorphism from Z,Z onto the subring
{m 1
T
[ m Z]
of J. In this case, J contains a copy of F
]
, and we say that it has characteristic .
The elds F
2
. F
3
. F
5
. . . . . Q are called the prime elds. Every eld contains a copy of
exactly one of them.
REMARK 1.4 The binomial theorem
(ab)
n
=a
n

_
n
1
_
a
n-1
b
_
n
2
_
a
n-2
b
2
b
n
holds in every commutative ring. If is prime, then divides
_
]
n
i
_
for all r with 1 _r _

n
1. Therefore, when J has characteristic ,
(ab)
]
n
=a
]
n
b
]
n
all n _1.
and so the map a a
]
: J J is a homomorphism. It is called the Frobenius endomor-
phism of J. When J is nite, the Frobenius endomorphism is an automorphism.
Review of polynomial rings
Let J be a eld.
1.5 The ring JX| of polynomials in the symbol (or indeterminate or variable) X
with coefcients in J is an J-vector space with basis 1, X, . . . , X
n
, . . . , and with the
multiplication dened by
_

i
a
i
X
i
__

}
b
}
X
}
_
=

k
_

i}=k
a
i
b
}
_
X
k
.
For any ring 1 containing J as a subring and element r of 1, there is a unique homomor-
phism : JX| 1 such that (X) =r and (a) =a for all a J.
1.6 Division algorithm: given (X) and g(X) JX| with g = 0, there exist q(X),
r(X) JX| with r =0 or deg(r) <deg(g) such that
=gq r:
moreover, q(X) and r(X) are uniquely determined. Thus JX| is a Euclidean domain with
deg as norm, and so is a unique factorization domain.
1.7 Let JX| and a J. Then
=(X a)q c
with q JX| and c J. Therefore, if a is a root of (that is, (a) = 0), then X a
divides . From unique factorization, it now follows that has at most deg( ) roots (see
also Exercise 1-3).
10 1. BASIC DEFINITIONS AND RESULTS
1.8 Euclids algorithm: Let and g JX| have gcd J(X). Euclids algorithm constructs
polynomials a(X) and b(X) such that
a(X) (X) b(X) g(X) =J(X). deg(a) <deg(g). deg(b) <deg( ).
Recall how it goes. We may assume that deg( ) _deg(g) since the argument is the same in
the opposite case. Using the division algorithm, we construct a sequence of quotients and
remainders
=q
0
g r
0
g =q
1
r
0
r
1
r
0
=q
2
r
1
r
2

r
n-2
=q
n
r
n-1
r
n
r
n-1
=q
n1
r
n
with r
n
the last nonzero remainder. Then, r
n
divides r
n-1
, hence r
n-2
,. . . , hence g, and
hence . Moreover,
r
n
=r
n-2
q
n
r
n-1
=r
n-2
q
n
(r
n-3
q
n-1
r
n-2
) = =a bg
and so every common divisor of and g divides r
n
: we have shown r
n
=gcd(. g).
Let a bg =J. If deg(a) _deg(g), write a =gq r with deg(r) <deg(g); then
r (b q )g =J.
and b q automatically has degree <deg( ).
PARI knows Euclidean division: typing divrem(13,5) in PARI returns 2. 3|, meaning
that 13 =253, and gcd(m,n) returns the greatest common divisor of m and n.
1.9 Let 1 be a nonzero ideal in JX|, and let be a nonzero polynomial of least degree in
1; then 1 =( ) (because JX| is a Euclidean domain). When we choose to be monic, i.e.,
to have leading coefcient one, it is uniquely determined by 1. Thus, there is a one-to-one
correspondence between the nonzero ideals of JX| and the monic polynomials in JX|.
The prime ideals correspond to the irreducible monic polynomials.
1.10 Since JX| is an integral domain, we can form its eld of fractions J(X). Its
elements are quotients ,g, and g polynomials, g =0.
Factoring polynomials
The following results help in deciding whether a polynomial is reducible, and in nding its
factors.
PROPOSITION 1.11 Suppose r Q is a root of a polynomial
a
n
X
n
a
n-1
X
n-1
a
0
. a
i
Z.
and let r =c,J, c. J Z, gcd(c. J) =1. Then c[a
0
and J[a
n
.
Factoring polynomials 11
PROOF. It is clear from the equation
a
n
c
n
a
n-1
c
n-1
J a
0
J
n
=0
that J[a
n
c
n
, and therefore, J[a
n
. Similarly, c[a
0
.
2
EXAMPLE 1.12 The polynomial (X) =X
3
3X 1 is irreducible in QX| because its
only possible roots are 1, and (1) =0 =(1).
PROPOSITION 1.13 (GAUSSS LEMMA) Let (X) ZX|. If (X) factors nontrivially
in QX|, then it factors nontrivially in ZX|.
PROOF. Let =gh in QX| with g. h Q. For suitable integers m and n, g
1
def
=mg and
h
1
def
=nh have coefcients in Z, and so we have a factorization
mn =g
1
h
1
in ZX|.
If a prime divides mn, then, looking modulo , we obtain an equation
0 =g
1
h
1
in F
]
X|.
Since F
]
X| is an integral domain, this implies that divides all the coefcients of at least
one of the polynomials g
1
. h
1
, say g
1
, so that g
1
=g
2
for some g
2
ZX|. Thus, we have
a factorization
(mn,) =g
2
h
1
in ZX|.
Continuing in this fashion, we eventually remove all the prime factors of mn, and so obtain
a nontrivial factorization of in ZX|.
2
PROPOSITION 1.14 If ZX| is monic, then every monic factor of in QX| lies in
ZX|.
PROOF. Let g be a monic factor of in QX|, so that =gh with h QX| also monic.
Let m. n be the positive integers with the fewest prime factors such that mg. nh ZX|. As
in the proof of Gausss Lemma, if a prime divides mn, then it divides all the coefcients
of at least one of the polynomials mg. nh, say mg, in which case it divides m because g is
monic. Now
n
]
g ZX|, which contradicts the denition of m.
2
ASIDE 1.15 We sketch an alternative proof of Proposition 1.14. A complex number is said to be
an algebraic integer if it is a root of a monic polynomial in ZX|. Proposition 1.11 shows that every
algebraic integer in Q lies in Z. The algebraic integers form a subring of C for an elementary
proof of this, using nothing but the symmetric polynomials theorem (5.35), see Chapter 1 of my notes
on algebraic geometry. Now let
1
. . . . .
n
be the roots of in C. By denition, they are algebraic
integers, and the coefcients of any monic factor of are polynomials in (certain of) the
i
, and
therefore are algebraic integers. If they lie in Q, then they lie in Z.
PROPOSITION 1.16 (EISENSTEINS CRITERION) Let
=a
n
X
n
a
n-1
X
n-1
a
0
. a
i
Z:
suppose that there is a prime such that:
12 1. BASIC DEFINITIONS AND RESULTS
does not divide a
n
,
divides a
n-1
. .... a
0
,

2
does not divide a
0
.
Then is irreducible in QX|.
PROOF. If (X) factors in QX|, then it factors in ZX|, say,
a
n
X
n
a
n-1
X
n-1
a
0
=(b
i
X
i
b
0
)(c
x
X
x
c
0
)
with b
i
. c
i
Z and r. s < m. Since , but not
2
, divides a
0
=b
0
c
0
, must divide exactly
one of b
0
, c
0
, say, b
0
. Now from the equation
a
1
=b
0
c
1
b
1
c
0
.
we see that [b
1
. and from the equation
a
2
=b
0
c
2
b
1
c
1
b
2
c
0
.
that [b
2
. By continuing in this way, we nd that divides b
0
. b
1
. . . . . b
i
, which contradicts
the condition that does not divide a
n
.
2
The last three propositions hold with Z replaced by any unique factorization domain.
REMARK 1.17 There is an algorithm for factoring a polynomial in QX|. To see this,
consider QX|. Multiply (X) by a rational number so that it is monic, and then
replace it by D
deg(( )
(
A
T
), with D equal to a common denominator for the coefcients
of , to obtain a monic polynomial with integer coefcients. Thus we need consider only
polynomials
(X) =X
n
a
1
X
n-1
a
n
. a
i
Z.
From the fundamental theorem of algebra (see 5.6 below), we know that splits
completely in CX|:
(X) =
n

i=1
(X
i
).
i
C.
From the equation
0 =(
i
) =
n
i
a
1

n-1
i
a
n
,
it follows that [
i
[ is less than some bound depending only on the degree and coefcients of
; in fact,
[
i
[ _max{1. mT], T =max[a
i
[.
Now if g(X) is a monic factor of (X), then its roots in C are certain of the
i
, and its
coefcients are symmetric polynomials in its roots (see p.75). Therefore, the absolute values
of the coefcients of g(X) are bounded in terms of the degree and coefcients of . Since
they are also integers (by 1.14), we see that there are only nitely many possibilities for
g(X). Thus, to nd the factors of (X) we (better PARI) have to do only a nite amount of
checking.
2
2
Of course, there are faster methods than this. The BerlekampZassenhaus algorithm factors the polynomial
over certain suitable nite elds F
]
, lifts the factorizations to rings Z,
n
Z for some m, and then searches for
factorizations in ZX| with the correct form modulo
n
.
Extension elds 13
Therefore, we need not concern ourselves with the problem of factoring polynomi-
als in the rings QX| or F
]
X| since PARI knows how to do it. For example, typing
content(6*X^2+18*X-24) in PARI returns 6, and factor(6*X^2+18*X-24) returns
X 1 and X 4, showing that
6X
2
18X 24 =6(X 1)(X 4)
in QX|. Typing factormod(X^2+3*X+3,7) returns X 4 and X 6, showing that
X
2
3X 3 =(X 4)(X 6)
in F
T
X|.
REMARK 1.18 One other observation is useful. Let ZX|. If the leading coefcient of
is not divisible by a prime , then a nontrivial factorization =gh in ZX| will give a
nontrivial factorization

= g

h in F
]
X|. Thus, if (X) is irreducible in F
]
X| for some
prime not dividing its leading coefcient, then it is irreducible in ZX|. This test is very
useful, but it is not always effective: for example, X
4
10X
2
1 is irreducible in ZX| but
it is reducible
3
modulo every prime .
Extension elds
A eld 1 containing a eld J is called an extension eld of J (or simply an extension of
J, and we speak of an extension 1,J). Such an 1 can be regarded as an J-vector space,
and we write 1: J| for the dimension, possibly innite, of 1 as an J-vector space. This
dimension is called the degree of 1 over J. We say that 1 is nite over J when it has nite
degree over J.
When 1 and 1
t
are extension elds of J, an J-homomorphism 1 1
t
is a homo-
morphism : 1 1
t
such that (c) =c for all c J.
EXAMPLE 1.19 (a) The eld of complex numbers C has degree 2 over R (basis {1. i ]).
(b) The eld of real numbers R has innite degree over Q: the eld Q is countable,
and so every nite-dimensional Q-vector space is also countable, but a famous argument of
Cantor shows that R is not countable.
(c) The eld of Gaussian numbers
Q(i )
def
={abi C [ a. b Q]
3
Here is a proof using only that the product of two nonsquares in F

]
is a square, which follows from the
fact that F

]
is cyclic (see Exercise 1-3). If 2 is a square in F
]
, then
X
4
10X
2
1 =(X
2
2
_
2X 1)(X
2
2
_
2X 1).
If 3 is a square in F
]
, then
X
4
10X
2
1 =(X
2
2
_
3X 1)(X
2
2
_
3X 1).
If neither 2 nor 3 are squares, 6 will be a square in F
]
, and
X
4
10X
2
1 =(X
2
(52
_
6))(X
2
(52
_
6)).
The general study of such polynomials requires nonelementary methods. See, for example, the paper
Brandl, R., Amer. Math. Monthly, 93 (1986), pp286288, which proves that every nonprime integer n _ 1
occurs as the degree of a polynomial in ZX| that is irreducible over Z but reducible modulo all primes.
14 1. BASIC DEFINITIONS AND RESULTS
has degree 2 over Q (basis {1. i ]).
(d) The eld J(X) has innite degree over J; in fact, even its subspace JX| has
innite dimension over J (basis 1. X. X
2
. . . .).
PROPOSITION 1.20 (MULTIPLICATIVITY OF DEGREES) Consider elds 11 J. Then
1,J is of nite degree if and only if 1,1 and 1,J are both of nite degree, in which case
1: J| =1: 1|1: J|.
PROOF. If 1 is of nite degree over J, then it is certainly of nite degree over 1. Moreover,
1, being a subspace of a nite dimensional J-vector space, is also nite dimensional.
Thus, assume that 1,1 and 1,J are of nite degree, and let (e
i
)
1_i_n
be a basis for 1
as an J-vector space and let (l
}
)
1_}_n
be a basis for 1 as an 1-vector space. To complete
the proof, it sufces to show that (e
i
l
}
)
1_i_n,1_}_n
is a basis for 1 over J, because then
1 will be nite over J of the predicted degree.
First, (e
i
l
}
)
i,}
spans 1. Let ; 1. Then, because (l
}
)
}
spans 1 as an 1-vector space,
; =

}
l
}
. some
}
1.
and because (e
i
)
i
spans 1 as an J-vector space,

}
=

i
a
i}
e
i
. some a
i}
J.
On putting these together, we nd that
; =

i,}
a
i}
e
i
l
}
.
Second, (e
i
l
}
)
i,}
is linearly independent. A linear relation

a
i}
e
i
l
}
= 0, a
i}
J,
can be rewritten

}
(

i
a
i}
e
i
)l
}
=0. The linear independence of the l
}
s now shows that

i
a
i}
e
i
=0 for each , and the linear independence of the e
i
s shows that each a
i}
=0.
2
The subring generated by a subset
An intersection of subrings of a ring is again a ring. Let J be a subeld of a eld 1, and let
S be a subset of 1. The intersection of all the subrings of 1 containing J and S is evidently
the smallest subring of 1 containing J and S. We call it the subring of 1 generated by J
and S (or generated over J by S), and we denote it JS|. When S ={
1
. ....
n
], we write
J
1
. ....
n
| for JS|. For example, C =R
_
1|.
LEMMA 1.21 The ring JS| consists of the elements of 1 that can be expressed as nite
sums of the form

a
i
1
i
n

i
1
1

i
n
n
. a
i
1
i
n
J.
i
S. (*)
PROOF. Let 1 be the set of all such elements. Evidently, 1 is a subring containing J and S
and contained in every other such subring. Therefore 1 equals JS|.
2
EXAMPLE 1.22 The ring Q|, =3.14159..., consists of the complex numbers that can
be expressed as a nite sum
a
0
a
1
a
2

2
a
n

n
. a
i
Q.
The ring Qi | consists of the complex numbers of the form abi , a. b Q.
The subeld generated by a subset 15
Note that the expression of an element in the form (*) will not be unique in general. This
is so already in Ri |.
LEMMA 1.23 Let 1 be an integral domain containing a subeld J (as a subring). If 1 is
nite dimensional when regarded as an J-vector space, then it is a eld.
PROOF. Let be a nonzero element of 1 we have to show that has an inverse in 1.
The map . .: 1 1 is an injective linear map of nite dimensional J-vector spaces,
and is therefore surjective. In particular, there is an element 1 such that =1.
2
Note that the lemma applies to subrings (containing J) of an extension eld 1 of J of
nite degree.
The subeld generated by a subset
An intersection of subelds of a eld is again a eld. Let J be a subeld of a eld 1, and let
S be a subset of 1. The intersection of all the subelds of 1 containing J and S is evidently
the smallest subeld of 1 containing J and S. We call it the subeld of 1 generated by
J and S (or generated over J by S), and we denote it J(S). It is the eld of fractions of
JS| in 1, since this is a subeld of 1 containing J and S and contained in every other
such eld. When S ={
1
. ....
n
], we write J(
1
. ....
n
) for J(S). Thus, J
1
. . . . .
n
|
consists of all elements of 1 that can be expressed as polynomials in the
i
with coefcients
in J, and J(
1
. . . . .
n
) consists of all elements of 1 that can be expressed as the quotient
of two such polynomials.
Lemma 1.23 shows that JS| is already a eld if it is nite dimensional over J, in which
case J(S) =JS|.
EXAMPLE 1.24 The eld Q(), =3.14. . ., consists of the complex numbers that can be
expressed as a quotient
g(),h(). g(X). h(X) QX|. h(X) =0.
The ring Qi | is already a eld.
An extension 1 of J is said to be simple if 1 =J() some 1. For example, Q()
and Qi | are simple extensions of Q.
Let J and J
t
be subelds of a eld 1. The intersection of the subelds of 1 containing
J and J
t
is evidently the smallest subeld of 1 containing both J and J
t
. We call it the
composite of J and J
t
in 1, and we denote it J J
t
. It can also be described as the subeld
of 1 generated over J by J
t
, or the subeld generated over J
t
by J:
J(J
t
) =J J
t
=J
t
(J).
Construction of some extension elds
Let (X) JX| be a monic polynomial of degree m, and let ( ) be the ideal gener-
ated by . Consider the quotient ring JX|,((X)), and write . for the image of X in
JX|,((X)), i.e., . is the coset X ((X)). Then:
16 1. BASIC DEFINITIONS AND RESULTS
(a) The map
1(X) 1(.): JX| J.|
is a surjective homomorphism in which (X) maps to 0. Therefore, (.) =0.
(b) From the division algorithm, we know that each element g of JX|,( ) is rep-
resented by a unique polynomial r of degree < m. Hence each element of J.| can be
expressed uniquely as a sum
a
0
a
1
. a
n-1
.
n-1
. a
i
J. (*)
(c) To add two elements, expressed in the form (*), simply add the corresponding
coefcients.
(d) To multiply two elements expressed in the form (*), multiply in the usual way, and
use the relation (.) =0 to express the monomials of degree _ m in . in terms of lower
degree monomials.
(e) Now assume (X) is irreducible. Then every nonzero J.| has an inverse,
which can be found as follows. Use (b) to write =g(.) with g(X) is a polynomial of
degree _m1, and use Euclids algorithm in JX| to obtain polynomials a(X) and b(X)
such that
a(X)(X) b(X)g(X) =J(X)
with J(X) the gcd of and g. In our case, J(X) is 1 because (X) is irreducible and
degg(X) <deg(X). When we replace X with ., the equality becomes
b(.)g(.) =1.
Hence b(.) is the inverse of g(.).
From these observations, we conclude:
1.25 For a monic irreducible polynomial (X) of degree m in JX|,
J.|
def
=JX|,((X))
is a eld of degree m over J. Moreover, computations in J.| reduce to computations in J.
EXAMPLE 1.26 Let (X) =X
2
1 RX|. Then R.| has:
elements: ab., a. b R:
addition: (ab.) (a
t
b
t
.) =(aa
t
) (b b
t
).:
multiplication: (ab.)(a
t
b
t
.) =(aa
t
bb
t
) (ab
t
a
t
b)..
We usually write i for . and C for R.|.
EXAMPLE 1.27 Let (X) = X
3
3X 1 QX|. We observed in (1.12) that this is
irreducible over Q, and so Q.| is a eld. It has basis {1. .. .
2
] as a Q-vector space. Let
=.
4
2.
3
3 Q.|.
Then using that .
3
3. 1 =0, we nd that =3.
2
7. 5. Because X
3
3X 1 is
irreducible,
gcd(X
3
3X 1. 3X
2
7X 5) =1.
In fact, Euclids algorithm gives
(X
3
3X 1)
_
-T
3T
X
29
111
_
(3X
2
7X 5)
_
T
111
X
2

26
111
X
2S
111
_
=1.
Stem elds 17
Hence
(3.
2
7. 5)
_
T
111
.
2

26
111
.
2S
111
_
=1.
and we have found the inverse of .
We can also do this in PARI: beta=Mod(X^4+2*X^3+3,X^3-3*X-1) reveals that =
3.
2
7. 5 in Q.|, and beta^(-1) reveals that
-1
=
T
111
.
2

26
111
.
2S
111
.
Stem elds
Let be a monic irreducible polynomial in JX|. A pair (1. ) consisting of an extension
1 of J and an 1 is called a stem eld
4
for if 1 =J| and () =0. For example,
the pair (1. ) with 1 =JX|,( ) =J.| and =. is a stem eld for . Let (1. ) be
a stem eld, and consider the surjective homomorphism of J-algebras
g(X) g(): JX| 1.
Its kernel is generated by a nonzero monic polynomial, which divides , and so must equal
it. Therefore the homomorphism denes an J-isomorphism
. : J.| 1. J.|
def
=JX|,( ).
In other words, the stem eld (1. ) of is J-isomorphic to the standard stem eld
(JX|,( ). .). In particular, each element of a stem eld (1. ) for has a unique
expression
a
0
a
1
a
n-1

n-1
. a
i
J. m=deg( ),
i.e., 1. . . . . .
n-1
is an J-basis for J|, and arithmetic in J| can be performed using
the same rules as in J.|. If (1.
t
) is a second stem eld for , then there is a unique
J-isomorphism 1 1
t
sending to
t
. We sometimes write stem eld J| instead of
stem eld (J|. ).
Algebraic and transcendental elements
For a eld J and an element of an extension eld 1, we have a homomorphism
(X) (): JX| 1.
There are two possibilities.
CASE 1: The kernel of the map is (0), so that, for JX|,
() =0 == =0 (in JX|).
In this case, we say that transcendental over J. The homomorphism JX| J| is an
isomorphism, and it extends to an isomorphism J(X) J().
CASE 2: The kernel is =(0), so that g() =0 for some nonzero g JX|. In this case,
we say that is algebraic over J. The polynomials g such that g() =0 form a nonzero
ideal in JX|, which is generated by the monic polynomial of least degree such () =0.
We call the minimum polynomial
5
of over J. It is irreducible, because otherwise there
4
Following A. Albert, Modern Higher Algebra, 1937, who calls the splitting eld of a polynomial its root
eld.
5
Some authors write minimal polynomial but the polynomial in question is minimum (it is the unique
minimal element of the set of monic polynomials with () =0).
18 1. BASIC DEFINITIONS AND RESULTS
would be two nonzero elements of 1 whose product is zero. The minimum polynomial is
characterized as an element of JX| by each of the following sets of conditions:
is monic; () =0 and divides every other polynomial g in JX| with g() =0.
is the monic polynomial of least degree such that () =0:
is monic, irreducible, and () =0.
Note that g(X) g() denes an isomorphism JX|,( ) J|. Since the rst is a
eld, so also is the second:
J() =J|.
Thus, J| is a stem eld for .
EXAMPLE 1.28 Let C be such that
3
3 1 = 0. Then X
3
3X 1 is monic,
irreducible, and has as a root, and so it is the minimum polynomial of over Q. The set
{1. .
2
] is a basis for Q| over Q. The calculations in Example 1.27 show that if is the
element
4
2
3
3 of Q|, then =3
2
7 5, and

-1
=
T
111

26
111

2S
111
.
REMARK 1.29 PARI knows how to compute in Q|. For example, factor(X^4+4) re-
turns the factorization
X
4
4 =(X
2
2X 2)(X
2
2X 2)
in QX|. Now type nf=nfinit(a^2+2*a+2) to dene a number eld nf generated over
Q by a root a of X
2
2X 1. Then nffactor(nf,x^4+4) returns the factorization
X
4
4 =(X a2)(X a)(X a))(X a2).
in Qa|.
A eld extension 1,J is said to be algebraic, and 1 is said to be algebraic over J, if
all elements of 1 are algebraic over J; otherwise it is said to be transcendental (or 1 is
said to be transcendental over J). Thus, 1,J is transcendental if at least one element of
1 is transcendental over J.
PROPOSITION 1.30 A eld extension 1,J is nite if and only if 1 is algebraic and nitely
generated (as a eld) over J.
PROOF. ==: To say that is transcendental over J amounts to saying that its powers
1. .
2
. . . . are linearly independent over J. Therefore, if 1 is nite over J, then it is
algebraic over J. It remains to show that 1 is nitely generated over J. If 1 =J, then it
is generated by the empty set. Otherwise, there exists an
1
1J. If 1 =J
1
|, there
exists an
2
1J
1
|, and so on. Since
J
1
|: J| <J
1
.
2
|: J| < <1: J|
this process terminates.
==: Let 1 =J(
1
. ....
n
) with
1
.
2
. . . .
n
algebraic over J. The extension J(
1
),J
is nite because
1
is algebraic over J, and the extension J(
1
.
2
),J(
1
) is nite because

2
is algebraic over J and hence over J(
1
). Thus, by ( 1.20), J(
1
.
2
) is nite over J.
Now repeat the argument.
2
Transcendental numbers 19
COROLLARY 1.31 (a) If 1 is algebraic over J, then every subring 1 of 1 containing J is
a eld.
(b) If in 11 J, 1is algebraic over 1 and 1 is algebraic over J, then 1is algebraic
over J.
PROOF. (a) We observed above (p. 18), that if is algebraic over J, then J| is a eld. If
1, then J| 1, and so has an inverse in 1.
(b) Every 1 is a root of a monic polynomial =X
n
a
n-1
X
n-1
a
0

1X|. Now each of the extensions Ja

0
. . . . . a
n-1
. | Ja
0
. . . . . a
n-1
| J is nite
(1.20), and so Ja
0
. . . . . a
n-1
. | is nite (hence algebraic) over J.
2
Transcendental numbers
A complex number is said to be algebraic or transcendental according as it is algebraic or
transcendental over Q. First some history:
1844: Liouville showed that certain numbers, now called Liouville numbers, are tran-
scendental.
1873: Hermite showed that e is transcendental.
1874: Cantor showed that the set of algebraic numbers is countable, but that R is not
countable. Thus most numbers are transcendental (but it is usually very difcult to prove
that any particular number is transcendental).
6
1882: Lindemann showed that is transcendental.
1934: Gelfond and Schneider independently showed that

is transcendental if and
are algebraic, =0. 1, and Q. (This was the seventh of Hilberts famous problems.)
2013: Eulers constant
; = lim
n-o
_
n

k=1
1,k logn
_
has not yet been proven to be transcendental or even irrational (see Lagarias, Jeffrey C.,
Eulers constant: Eulers work and modern developments. Bull. Amer. Math. Soc. 50
(2013), no. 4, 527628; arXiv:1303:1856).
2013: The numbers e and e are surely transcendental, but again they have not
even been proved to be irrational!
PROPOSITION 1.32 The set of algebraic numbers is countable.
PROOF. Dene the height h(r) of a rational number to be max([m[. [n[), where r =m,n
is the expression of r in its lowest terms. There are only nitely many rational numbers
with height less than a xed number N. Let (N) denote the set of algebraic numbers
whose minimum equation over Q has degree _N and has coefcients of height <N. Then
(N) is nite for each N. Choose a bijection from some segment 0. n(1)| of N onto (10);
extend it to a bijection from a segment 0. n(2)| onto (100), and so on.
2
6
In 1873 Cantor proved the rational numbers countable. . . . He also showed that the algebraic numbers. . .
were countable. However his attempts to decide whether the real numbers were countable proved harder. He
had proved that the real numbers were not countable by December 1873 and published this in a paper in 1874
(MacTutor).
20 1. BASIC DEFINITIONS AND RESULTS
A typical Liouville number is

o
n=0
1
10
n
in its decimal expansion there are increas-
ingly long strings of zeros. Since its decimal expansion is not periodic, the number is not
rational. We prove that the analogue of this number in base 2 is transcendental.
THEOREM 1.33 The number =

1
2
n
is transcendental.
PROOF.
7
Suppose not, and let
(X) =X
d
a
1
X
d-1
a
d
. a
i
Q.
be the minimum polynomial of over Q. Thus Q|: Q| =J. Choose a nonzero integer D
such that D (X) ZX|.
Let
1
=

1
n=0
1
2
n
, so that
1
as N o, and let .
1
=(
1
). As is not
rational, (X), being irreducible of degree >1, has no rational root. Since
1
=, it cant
be a root of (X), and so .
1
=0. Evidently, .
1
Q; in fact (2
1
)
d
D.
1
Z, and so
[(2
1
)
d
D.
1
[ _1. (*)
From the fundamental theorem of algebra (see 5.6 below), we know that splits in
CX|, say,
(X) =
d

i=1
(X
i
).
i
C.
1
=.
and so
[.
1
[ =
d

i=1
[
1

i
[ _[
1

1
[(
1
M)
d-1
. where M =max
iy1
{1. [
i
[].
But
[
1

1
[ =
o

n=11
1
2
n
_
1
2
(11)
_
o

n=0
1
2
n
_
=
2
2
(11)
.
Hence
[.
1
[ _
2
2
(11)
(
1
M)
d-1
and
[(2
1
)
d
D.
1
[ _2
2
d1
D
2
(11)
(
1
M)
d-1
which tends to 0 as N obecause
2
dN
2
.NC1/
=
_
2
d
2
NC1
_
1
0. This contradicts (*).
2
Constructions with straight-edge and compass.
The Greeks understood integers and the rational numbers. They were surprised to nd
that the length of the diagonal of a square of side 1, namely,
_
2, is not rational. They
thus realized that they needed to extend their number system. They then hoped that the
constructible numbers would sufce. Suppose we are given a length, which we call 1, a
straight-edge, and a compass (device for drawing circles). A real number (better a length) is
constructible if it can be constructed by forming successive intersections of
7
This proof, which I learnt from David Masser, also works for

1
o
n
for every integer a _2.
Constructions with straight-edge and compass. 21
lines drawn through two points already constructed, and
circles with centre a point already constructed and radius a constructed length.
This led them to three famous questions that they were unable to answer: is it possible
to duplicate the cube, trisect an angle, or square the circle by straight-edge and compass
constructions? Well see that the answer to all three is negative.
Let J be a subeld of R. For a positive a J,
_
a denotes the positive square root of a
in R. The J-plane is J J RR. We make the following denitions:
An J-line is a line in RR through two points in the J-plane. These are the
lines given by equations
a. b, c =0. a. b. c J.
An J-circle is a circle in RR with centre an J-point and radius an element
of J. These are the circles given by equations
(. a)
2
(, b)
2
=c
2
. a. b. c J.
LEMMA 1.34 Let 1 =1
t
be J-lines, and let C =C
t
be J-circles.
(a) 11
t
=0 or consists of a single J-point.
(b) 1C =0 or consists of one or two points in the J
_
e|-plane, some e J, e >0.
(c) C C
t
=0 or consists of one or two points in the J
_
e|-plane, some e J, e >0.
PROOF. The points in the intersection are found by solving the simultaneous equations, and
hence by solving (at worst) a quadratic equation with coefcients in J.
2
LEMMA 1.35 (a) If c and J are constructible, then so also are c J, c, cJ, and
c
d
(J =0).
(b) If c >0 is constructible, then so also is
_
c.
SKETCH OF PROOF. First show that it is possible to construct a line perpendicular to a given
line through a given point, and then a line parallel to a given line through a given point.
Hence it is possible to construct a triangle similar to a given one on a side with given length.
By an astute choice of the triangles, one constructs cJ and c
-1
. For (b), draw a circle of
radius
c1
2
and centre (
c1
2
. 0), and draw a vertical line through the point =(1. 0) to meet
the circle at 1. The length 1 is
_
c. (For more details, see Artin, M., 1991, Algebra,
Prentice Hall, Chapter 13, Section 4.)
2
THEOREM 1.36 (a) The set of constructible numbers is a eld.
(b) A number is constructible if and only if it is contained in a subeld of R of the form
Q
_
a
1
. . . . .
_
a
i
|. a
i
Q
_
a
1
. . . . .
_
a
i-1
|. a
i
>0.
PROOF. (a) Immediate from (a) of Lemma 1.35.
(b) It follows from Lemma 1.34 that every constructible number is contained in such
a eld Q
_
a
1
. . . . .
_
a
i
|. Conversely, if all the elements of Q
_
a
1
. . . . .
_
a
i-1
| are con-
structible, then
_
a
i
is constructible (by 1.35b), and so all the elements of Q
_
a
1
. . . . .
_
a
i
|
are constructible (by (a)). Applying this for i =0. 1. . . ., we nd that all the elements of
Q
_
a
1
. . . . .
_
a
i
| are constructible.
2
22 1. BASIC DEFINITIONS AND RESULTS
COROLLARY 1.37 If is constructible, then is algebraic over Q, and Q|: Q| is a power
of 2.
PROOF. According to Proposition 1.20, Q|: Q| divides
Q
_
a
1
|
_
a
i
|: Q|
and Q
_
a
1
. . . . .
_
a
i
|: Q| is a power of 2.
2
COROLLARY 1.38 It is impossible to duplicate the cube by straight-edge and compass
constructions.
PROOF. The problem is to construct a cube with volume 2. This requires constructing the
real root of the polynomial X
3
2. But this polynomial is irreducible (by Eisensteins
criterion 1.16 for example), and so Q
3
_
2|: Q| =3.
2
COROLLARY 1.39 In general, it is impossible to trisect an angle by straight-edge and
compass constructions.
PROOF. Knowing an angle is equivalent to knowing the cosine of the angle. Therefore, to
trisect 3, we have to construct a solution to
cos3 =4cos
3
3cos.
For example, take 3 = 60 degrees. As cos60

=
1
2
, to construct , we have to solve
8.
3
6. 1 =0, which is irreducible (apply 1.11).
2
COROLLARY 1.40 It is impossible to square the circle by straight-edge and compass con-
structions.
PROOF. A square with the same area as a circle of radius r has side
_
r. Since is
transcendental
8
, so also is
_
.
2
We next consider another problem that goes back to the ancient Greeks: list the n such
that the regular n-sided polygon can be constructed. Here we consider the question for a
prime (see 5.12 for the general case). Note that X
]
1 is not irreducible; in fact
X
]
1 =(X 1)(X
]-1
X
]-2
1).
LEMMA 1.41 If is prime, then X
]-1
1 is irreducible; hence Qe
2ti{]
| has degree
1 over Q.
PROOF. Let (X) =(X
]
1),(X 1) =X
]-1
1; then
(X 1) =
(X 1)
]
1
X
=X
]-1
a
2
X
2
a
1
X .
with a
i
=
_
]
i1
_
. Now[a
i
for i =1. .... 2, and so (X1) is irreducible by Eisensteins
criterion 1.16. This implies that (X) is irreducible.
2
8
Proofs of this can be found in many books on number theory, for example, in 11.14 of
Hardy, G. H., and Wright, E. M., An Introduction to the Theory of Numbers, Fourth Edition, Oxford, 1960.
Algebraically closed elds 23
In order to construct a regular -gon, an odd prime, we need to construct
cos
2t
]
=(e
2i
p
(e
2i
p
)
-1
),2.
But
Qe
2i
p
| Qcos
2t
]
| Q.
and the degree of Qe
2i
p
| over Qcos
2t
]
| is 2 the equation

2
2cos
2t
]
1 =0. =e
2i
p
.
shows that it is _2, and it is not 1 because Qe
2i
p
| is not contained in R. Hence
Qcos
2t
]
|: Q| =
1
2
.
Thus, if the regular -gon is constructible, then (1),2 =2
k
for some k (later (5.12),
we shall see a converse), which implies that =2
k1
1. But 2
i
1 can be a prime only
if r is a power of 2, because otherwise r has an odd factor t and for t odd,
Y
t
1 =(Y 1)(Y
t-1
Y
t-2
1):
whence
2
xt
1 =(2
x
1)((2
x
)
t-1
(2
x
)
t-2
1).
Thus primes for which the regular -gon is constructible are exactly those of the form
= 2
2
k
1 for some k. Such are called Fermat primes (because he conjectured
that all numbers of the form 2
2
k
1 are prime). For k = 0. 1. 2. 3. 4, we have 2
2
k

1 = 3. 5. 17. 257. 65537, which are indeed all prime, but Euler showed that 2
32
1 =
(641)(6700417), and we dont know whether there are any more Fermat primes. Thus,
we do not know the list of primes for which the regular -gon is constructible.
Gauss showed that
9
cos
2
17
=
1
16

1
16
_
17
1
16
_
342
_
17
1
8
_
173
_
17
_
342
_
172
_
342
_
17
when he was 18 years old. This success encouraged him to become a mathematician.
Algebraically closed elds
We say that a polynomial splits in JX| (or, more loosely, in J) if it is a product of
polynomials of degree 1 in JX|.
PROPOSITION 1.42 For a eld , the following statements are equivalent:
(a) Every nonconstant polynomial in X| splits in X|.
(b) Every nonconstant polynomial in X| has at least one root in .
(c) The irreducible polynomials in X| are those of degree 1.
9
Or perhaps that
cos
2t
1T
=
1
16

1
16
_
17
1
16
_
342
_
17
1
S
_
173
_
172
_
342
_
17
_
17026
_
17
both expressions are correct.
24 1. BASIC DEFINITIONS AND RESULTS
(d) Every eld of nite degree over equals .
PROOF. The implications (a) ==(b) ==(c) ==(a) are obvious.
(c) ==(d). Let 1 be a nite extension of , and let 1. The minimum polynomial of
has degree 1, and so .
(d) ==(c). Let be an irreducible polynomial in X|. Then X|,( ) is an extension
eld of of degree deg( ) (see 1.30), and so deg( ) =1.
2
DEFINITION 1.43 (a) A eld is said to be algebraically closed if it satises the equivalent
statements of Proposition 1.42.
(b) A eld is said to be an algebraic closure of a subeld J when it is algebraically
closed and algebraic over J.
For example, the fundamental theorem of algebra (see 5.6 below) says that C is alge-
braically closed. It is an algebraic closure of R.
PROPOSITION 1.44 If is algebraic over J and every polynomial JX| splits in
X|, then is algebraically closed (hence an algebraic closure of J).
PROOF. Let be a nonconstant polynomial in X|. We have to show that has a root in
. We know (see 1.25) that has a root in some nite extension
t
of . Set
=a
n
X
n
a
0
, a
i
.
and consider the elds
J Ja
0
. . . . . a
n
| Ja
0
. . . . . a
n
. |.
Each extension is algebraic and nitely generated, and hence nite (by 1.30). Therefore
lies in a nite extension of J, and so is algebraic over J it is a root of a polynomial g
with coefcients in J. By assumption, g splits in X|, and so the roots of g in
t
all lie in
. In particular, .
2
PROPOSITION 1.45 Let J; then
{ [ algebraic over J]
is a eld.
PROOF. If and are algebraic over J, then J. | is a eld (by 1.31) of nite degree
over J (by 1.30). Thus, every element of J. | is algebraic over J, including , ,,
.
2
The eld constructed in the proposition is called the algebraic closure of J in .
COROLLARY 1.46 Let be an algebraically closed eld. For any subeld J of , the
algebraic closure of J in is an algebraic closure of J.
PROOF. From its denition, we see that it is algebraic over J and every polynomial in JX|
splits in it. Now Proposition 1.44 shows that it is an algebraic closure of J.
2
Exercises 25
Thus, when we admit the fundamental theorem of algebra (5.6), every subeld of C has
an algebraic closure (in fact, a canonical algebraic closure). Later (Chapter 6) we shall prove
(using the axiom of choice) that every eld has an algebraic closure.
ASIDE 1.47 Although various classes of eld, for example, number elds and function elds, had
been studied earlier, the rst systematic account of the theory of abstract elds was given by Steinitz
in 1910 (Algebraische Theorie der K orper, J. Reine Angew. Math., 137:167309). Here he introduced
the notion of a prime eld, distinguished between separable and inseparable extensions, and showed
that every eld can be obtained as an algebraic extension of a purely transcendental extension. He
also proved that every eld has an algebraic closure, unique up to isomorphism. His work inuenced
later algebraists (Noether, van der Waerden, Artin, . . . ) and his article has been described by Bourbaki
as . . . a fundamental work which may be considered as the origin of todays concept of algebra.
See: Roquette, Peter, In memoriam Ernst Steinitz (18711928). J. Reine Angew. Math. 648 (2010),
111.
Exercises
1-1 Let 1 =Q|, where
3

2
2 =0. Express (
2
1)(
2
) and (1)
-1
in the form a
2
b c with a. b. c Q.
1-2 Determine Q(
_
2.
_
3): Q|.
1-3 Let J be a eld, and let (X) JX|.
(a) For every a J, show that there is a polynomial q(X) JX| such that
(X) =q(X)(X a) (a).
(b) Deduce that (a) =0 if and only if (X a)[(X).
(c) Deduce that (X) can have at most deg roots.
(d) Let G be a nite abelian group. If G has at most m elements of order dividing m for
each divisor m of (G: 1), show that G is cyclic.
(e) Deduce that a nite subgroup of J

, J a eld, is cyclic.
1-4 Show that with straight-edge, compass, and angle-trisector, it is possible to construct a
regular 7-gon.
1-5 Let (X) be an irreducible polynomial over J of degree n, and let 1 be a eld
extension of J with 1 : J| =m. If gcd(m. n) =1, show that is irreducible over 1.
CHAPTER 2
Splitting Fields; Multiple Roots
Maps from simple extensions.
Let 1 and 1
t
be elds containing J. Recall that an J-homomorphism is a homomorphism
: 1 1
t
such that (a) =a for all a J. Thus an J-homorphism maps a polynomial

a
i
1
i
m

i
1
1

i
m
n
. a
i
1
i
m
J.
to

a
i
1
i
m
(
1
)
i
1
(
n
)
i
m
.
An J-isomorphism is a bijective J-homomorphism.
An J-homomorphism 1 1
t
of elds is, in particular, an injective J-linear map
of J-vector spaces, and so, if 1 and 1
t
have the same nite degree over J, then every
J-homomorphism is an J-isomorphism.
PROPOSITION 2.1 Let J() be a simple eld extension of a eld J, and let be a second
eld containing J.
(a) Let be transcendental over J. For every J-homomorphism : J() , () is
transcendental over J, and the map () denes a one-to-one correspondence
{J-homomorphisms : J() ] -{elements of transcendental over J].
(b) Let be algebraic over J with minimumpolynomial (X). For every J-homomorphism
: J| , () is a root of (X) in , and the map () denes a one-to-
one correspondence
{J-homomorphisms : J| ] -{roots of in ].
In particular, the number of such maps is the number of distinct roots of in .
PROOF. (a) To say that is transcendental over J means that J| is isomorphic to the
polynomial ring in the symbol with coefcients in J. For every ; , there is a unique
J-homomorphism : J| sending to ; (see 1.5). This extends to the eld of
fractions J() of J| if and only if all nonzero elements of J| are sent to nonzero
elements of , which is so if and only if ; is transcendental.
27
28 2. SPLITTING FIELDS; MULTIPLE ROOTS
(b) Let (X) =

a
i
X
i
, and consider an J-homomorphism: J| . On applying
to the equation

a
i

i
=0, we obtain the equation

a
i
()
i
=0, which shows that
() is a root of (X) in . Conversely, if ; is a root of (X), then the map
JX| , g(X) g(;), factors through JX|,((X)). When composed with the inverse
of the isomorphismX(X) : JX|,((X)) J|, this becomes a homomorphism
J| sending to ;.
2
We shall need a slight generalization of this result.
PROPOSITION 2.2 Let J() be a simple eld extension of a eld J, and let
0
: J be
a homomorphism of J into a second eld .
(a) If is transcendental over J, then the map () denes a one-to-one correspon-
dence
{extensions : J() of
0
] -{elements of transcendental over
0
(J)].
(b) If is algebraic over J, with minimum polynomial (X), then the map ()
denes a one-to-one correspondence
{extensions : J| of
0
] -{roots of
0
in ].
In particular, the number of such maps is the number of distinct roots of
0
in .
By
0
we mean the polynomial obtained by applying
0
to the coefcients of :
if =

a
i
X
i
then
0
=

(a
i
)X
i
. By an extension of
0
to J() we mean a
homomorphism : J() such that [J =
0
.
The proof of the proposition is essentially the same as that of the preceding proposition.
Splitting elds
Let be a polynomial with coefcients in J. A eld 1 containing J is said to split if
splits in 1X|: (X) =

n
i=1
(X
i
) with
i
1. If, in addition, 1 is generated by the
roots of ,
1 =J
1
. . . . .
n
|.
then it is called a splitting or root eld for . Note that

i
(X)
n
i
(m
i
_1) and

i
(X)
have the same splitting elds. Also, that if has deg( ) 1 roots in 1, then it splits in
1X|.
EXAMPLE 2.3 (a) Let (X) =aX
2
bX c QX|, and let =
_
b
2
4ac. The sub-
eld Q| of C is a splitting eld for .
(b) Let (X) = X
3
aX
2
bX c QX| be irreducible, and let
1
.
2
.
3
be its
roots in C. Since the nonreal roots of occur in conjugate pairs, either 1 or 3 of the
i
are
real. Then Q
1
.
2
.
3
| =Q
1
.
2
| is a splitting eld for (X). Note that Q
1
|: Q| =3
and that Q
1
.
2
|: Q
1
|| =1 or 2, and so Q
1
.
2
|: Q| =3 or 6. Well see later (4.2) that
the degree is 3 if and only if the discriminant of (X) is a square in Q. For example, the
discriminant of X
3
bX c is 4b
3
27c
2
, and so the splitting eld of X
3
10X 1
has degree 6 over Q.
Splitting elds 29
PROPOSITION 2.4 Every polynomial JX| has a splitting eld 1
(
, and
1
(
: J| _(deg ) (factorial deg ).
PROOF. Let J
1
=J
1
| be a stem eld for some monic irreducible factor of in JX|.
Then (
1
) =0, and we let J
2
=J
1

2
| be a stem eld for some monic irreducible factor
of (X),(X
1
) in J
1
X|. Continuing in this fashion, we arrive at a splitting eld 1
(
.
Let n =deg . Then J
1
: J| =degg
1
_n, J
2
: J
1
| _n1. ..., and so 1
(
: 1| _n.
2
REMARK 2.5 Let J be a eld. For a given integer n, there may or may not exist polynomials
of degree n in JX| whose splitting eld has degree n this depends on J. For example,
there do not for n > 1 if J =C (see 5.6), nor for n > 2 if J =F
]
(see 4.22) or J =R.
However, later (4.33) we shall see how to write down innitely many polynomials of degree
n in QX| whose splitting elds have degree n.
EXAMPLE 2.6 (a) Let (X) =(X
]
1),(X 1) QX|, prime. If is one root of ,
then the remaining roots are
2
.
3
. . . . .
]-1
, and so the splitting eld of is Q|.
(b) Suppose J is of characteristic , and let =X
]
X a JX|. If is one root
of , then the remaining roots are 1. .... 1, and so any eld generated over J by
is a splitting eld for (and J| .JX|,( ) if is irreducible).
(c) If is one root of X
n
a, then the remaining roots are all of the form , where

n
=1. Therefore, if J contains all the nth roots of 1 (by which we mean that X
n
1 splits
in JX|), then J| is a splitting eld for X
n
a. Note that if is the characteristic of J,
then X
]
1 =(X 1)
]
, and so J automatically contains all the th roots of 1.
PROPOSITION 2.7 Let JX|. Let 1 be a eld generated over J by roots of , and let
be a eld containing J in which splits.
(a) There exists an J-homomorphism : 1 ; the number of such homomorphisms is
at most 1: J|, and equals 1: J| if has distinct roots in .
(b) If 1 and are both splitting elds for , then each J-homomorphism 1 is an
isomorphism. In particular, any two splitting elds for are J-isomorphic.
By splitting in , we mean that
(X) =

deg(( )
i=1
(X
i
).
i
.
in X|. By having distinct roots in , we mean that
i
=
}
if i = .
PROOF. We begin with an observation: let J, , and be as in the statement of the
proposition, let 1 be a subeld of containing J, and let g be a factor of in 1X|; then g
divides in X| and so (by unique factorization in X|), g is product of certain number
of the factors X
i
of in X|; in particular, we see that g splits in , and that its roots
are distinct if the roots of are distinct.
(a) By assumption, 1 = J
1
. ....
n
| with the
i
(some of the) roots of (X). The
minimum polynomial of
1
is an irreducible polynomial
1
dividing , and deg(
1
) =
J
1
|: J|. From the initial observation with 1 =J, we see that
1
splits in , and that
its roots are distinct if the roots of are distinct. According to Proposition 2.1, there exists
an J-homomorphism
1
: J
1
| , and the number of such homomorphisms is at most
J
1
|: J|, with equality holding when has distinct roots in .
30 2. SPLITTING FIELDS; MULTIPLE ROOTS
The minimum polynomial of
2
over J
1
| is an irreducible factor
2
of in J
1
|X|.
On applying the initial observation with 1=
1
J
1
| and g =
1

2
, we see that
1

2
splits
in , and that its roots are distinct if the roots of are distinct. According to Proposition
2.2, each
1
extends to a homomorphism
2
: J
1
.
2
| , and the number of extensions
is at most J
1
.
2
|: J
1
||, with equality holding when has distinct roots in .
On combining these statements we conclude that there exists an J-homomorphism
: J
1
.
2
| .
and that the number of such homomorphisms is at most J
1
.
2
|: J|, with equality holding
if has distinct roots in .
After repeating the argument m times, we obtain (a).
(b) Every J-homomorphism 1 is injective, and so, if there exists such a homomor-
phisms, 1: J| _: J|. If 1 and are both splitting elds for , then (a) shows that there
exist homomorphism J 1, and so 1: J| =: J|. Therefore, every J-homomorphism
1 is an isomorphism.
2
COROLLARY 2.8 Let 1 and 1 be extension elds of J, with 1 nite over J.
(a) The number of J-homomorphisms 1 1 is at most 1: J|.
(b) There exists a nite extension ,1 and an J-homomorphism 1 .
PROOF. Write 1 =J
1
. . . . .
n
|, and be the product of the minimum polynomials of the

i
. Let be a splitting eld for regarded as an element of 1X|. The proposition shows
that there exists an J-homomorphism 1 , and the number of such homomorphisms is
_ 1: J|. This proves (b), and since an J-homomorphism 1 1 can be regarded as an
J-homomorphism 1 , it also proves (a).
2
REMARK 2.9 (a) Let 1
1
. 1
2
. . . . . 1
n
be nite extensions of J, and let 1 be an extension of
J. The corollary shows that there exists a nite extension 1
1
,1 containing an isomorphic
image of 1
1
; then that there exists a nite extension 1
2
,1
1
containing an isomorphic
image of 1
2
. On continuing in this fashion, we nd that there exists a nite extension /1
containing an isomorphic copy of every 1
i
.
(b) Let JX|. If 1 and 1
t
are both splitting elds of , then we know there is an
J-isomorphism 1 1
t
, but there will in general be no preferred such isomorphism. Error
and confusion can result if the elds are simply identied. Also, it makes no sense to speak
of the eld J| generated by a root of unless is irreducible (the elds generated
by the roots of two different factors are unrelated). Even when is irreducible, it makes
no sense to speak of the eld J. | generated by two roots . of (the extensions of
J| generated by the roots of two different factors of in J|X| may be very different).
Multiple roots
Let . g JX|. Even when and g have no common factor in JX|, one might expect
that they could acquire a common factor in X| for some J. In fact, this doesnt
happen greatest common divisors dont change when the eld is extended.
PROPOSITION 2.10 Let and g be polynomials in JX|, and let be an extension of J.
If r(X) is the gcd of and g computed in JX|, then it is also the gcd of and g in X|.
In particular, distinct monic irreducible polynomials in JX| do not acquire a common root
in any extension eld of J.
Multiple roots 31
PROOF. Let r
T
(X) and r

(X) be the greatest common divisors of and g in JX| and

X| respectively. Certainly r
T
(X)[r

(X) in X|, but Euclids algorithm (1.8) shows

that there are polynomials a and b in JX| such that
a(X)(X) b(X)g(X) =r
T
(X).
and so r

(X) divides r
T
(X) in X|.
For the second statement, note that the hypotheses imply that gcd(. g) =1 (in JX|),
and so and g cant acquire a common factor in any extension eld.
2
The proposition allows us to speak of the greatest common divisor of and g without
reference to a eld.
Let JX|. Then splits into linear factors
(X) =a
i

i=1
(X
i
)
n
i
.
i
distinct, m
i
_1,
i

i=1
m
i
=deg( ). (*)
in X| for some extension eld of J (see 2.4). We say that
i
is a root of of
multiplicity m
i
in . If m
i
> 1,
i
is said to be a multiple root of , and otherwise it is a
simple root.
The unordered sequence of integers m
1
. . . . . m
i
in (*) is independent of the extension
eld chosen to split . Certainly, it is unchanged when is replaced with its subeld
J
1
. . . . .
n
|, but J
1
. . . . .
n
| is a splitting eld for , and any two splitting elds are
J-isomorphic (2.7b). We say that has a multiple root when at least one of the m
i
> 1,
and we say that has only simple roots when all m
i
=1.
We wish to determine when a polynomial has a multiple root. If has a multiple factor
in JX|, say =

i
(X)
n
i
with some m
i
>1, then obviously it will have a multiple root.
If =

i
with the
i
distinct monic irreducible polynomials, then Proposition 2.10 shows
that has a multiple root if and only if at least one of the
i
has a multiple root. Thus, it
sufces to determine when an irreducible polynomial has a multiple root.
EXAMPLE 2.11 Let J be of characteristic =0, and assume that J contains an element a
that is not a th-power, for example, a =T in the eld F
]
(T ). Then X
]
a is irreducible
in JX|, but X
]
a
1.4
= (X )
]
in its splitting eld. Thus an irreducible polynomial can
have multiple roots.
Dene the derivative
t
(X) of a polynomial (X) =

a
i
X
i
to be

i a
i
X
i-1
. When
has coefcients in R, this agrees with the denition in calculus. The usual rules for
differentiating sums and products still hold, but note that in characteristic the derivative of
X
]
is zero.
PROPOSITION 2.12 For a nonconstant irreducible polynomial in JX|, the following
statements are equivalent:
(a) has a multiple root;
(b) gcd(.
t
) =1;
(c) J has characteristic =0 and is a polynomial in X
]
;
(d) all the roots of are multiple.
32 2. SPLITTING FIELDS; MULTIPLE ROOTS
PROOF. (a) == (b). Let be a multiple root of , and write =(X )
n
g(X), m>1,
in some splitting eld. Then

t
(X) =m(X )
n-1
g(X) (X )
n
g
t
(X). (1)
Hence
t
() =0, and so gcd(.
t
) =1.
(b) == (c). Since is irreducible and deg(
t
) <deg( ),
gcd(.
t
) =1 ==
t
=0.
But, because is nonconstant,
t
can be zero only if the characteristic is =0 and is a
polynomial in X
]
.
(c) == (d). Suppose (X) =g(X
]
), and let g(X) =

i
(X a
i
)
n
i
in some splitting
eld of . Then
(X) =g(X
]
) =

i
(X
]
a
i
)
n
i
=

i
(X
i
)
]n
i
where
]
i
=a
i
. Hence every root of (X) has multiplicity at least .
(d) == (a). Obvious.
2
PROPOSITION 2.13 For a nonzero polynomial in JX|, the following statements are
equivalent:
(a) gcd(.
t
) =1;
(b) has only simple roots (in any eld splitting of ).
PROOF. Let be an extension of J such that splits in X|. A root of in is
multiple if and only if it is also a root of
t
(see (1)).
If gcd(.
t
) =1, then and
t
have no common factor X in X| (see 2.10), and
so they have no common root. Hence has only simple roots.
If has only simple roots, then J
def
=gcd(.
t
) must be constant, because otherwise it
would have a root in which would be a common root of (X) and
t
(X).
2
DEFINITION 2.14 A polynomial JX| is said to be separable if it is nonzero and
satises the equivalent conditions of (2.13)
1
According to this denition, a polynomial with multiple factors is not separable. The
preceding discussion shows that JX| without multiple factors will be separable unless
(a) the characteristic of J is =0, and
(b) at least one of the irreducible factors of is a polynomial in X
]
.
Note that, if JX| is separable, then it remains separable over every eld containing
J (condition (a) of 2.13 continues to hold see 2.10).
DEFINITION 2.15 A eld J is said to be perfect if all irreducible polynomials in JX| are
separable.
1
This is Bourbakis denition. Often (e.g., in the books of Jacobson and in earlier versions of these notes) a
polynomial is said to be separable if none of its irreducible factors has a multiple root.
Exercises 33
PROPOSITION 2.16 Every eld of characteristic zero is perfect. A eld J of characteristic
=0 is perfect if and only if every element of J is a th power.
PROOF. A eld of characteristic zero is obviously perfect, and so we may suppose J to
be of characteristic = 0. If J contains an element a that is not a th power, then the
polynomial X
]
a JX| is not separable (see 2.11). Conversely, if every element of J is
a th power, then every polynomial in X
]
with coefcients in J is a th power in JX|,

a
i
X
]
=(

b
i
X)
]
if a
i
=b
]
i
,
and so is not irreducible.
2
EXAMPLE 2.17 (a) A nite eld J is perfect, because the Frobenius endomorphism
a a
]
: J J is injective and therefore surjective (by counting).
(b) A eld that can be written as a union of perfect elds is perfect. Therefore, every eld
algebraic over F
]
is perfect.
(c) Every algebraically closed eld is perfect.
(d) If J
0
has characteristic =0, then J =J
0
(X) is not perfect, because X is not a th
power.
Exercises
2-1 Let J be a eld of characteristic =2.
(a) Let 1 be quadratic extension of J (i.e., 1: J| =2); show that
S(1) ={a J

[ a is a square in 1]
is a subgroup of J

containing J
2
.
(b) Let 1 and 1
t
be quadratic extensions of J; show that there is an J-isomorphism
: 1 1
t
if and only if S(1) =S(1
t
).
(c) Show that there is an innite sequence of elds 1
1
. 1
2
. . . . with 1
i
a quadratic
extension of Q such that 1
i
is not isomorphic to 1
}
for i = .
(d) Let be an odd prime. Show that, up to isomorphism, there is exactly one eld with

2
elements.
2-2 (a) Let J be a eld of characteristic . Show that if X
]
X a is reducible in JX|,
then it splits into distinct factors in JX|.
(b) For every prime , show that X
]
X 1 is irreducible in QX|.
2-3 Construct a splitting eld for X
5
2 over Q. What is its degree over Q?
2-4 Find a splitting eld of X
]
m
1 F
]
X|. What is its degree over F
]
?
2-5 Let JX|, where J is a eld of characteristic 0. Let J(X) =gcd(.
t
). Show
that g(X) =(X)J(X)
-1
has the same roots as (X), and these are all simple roots of
g(X).
2-6 Let (X) be an irreducible polynomial in JX|, where J has characteristic . Show
that (X) can be written (X) =g(X
]
e
) where g(X) is irreducible and separable. Deduce
that every root of (X) has the same multiplicity
e
in any splitting eld.
CHAPTER 3
The Fundamental Theorem of Galois
Theory
In this chapter, we prove the fundamental theorem of Galois theory, which gives a one-to-one
correspondence between the subelds of the splitting eld of a separable polynomial and the
subgroups of the Galois group of .
Groups of automorphisms of elds
Consider elds 1 J. An J-isomorphism 1 1 is called an J-automorphism of 1.
The J-automorphisms of 1 form a group, which we denote Aut(1,J).
EXAMPLE 3.1 (a) There are two obvious automorphisms of C, namely, the identity map
and complex conjugation. Well see later (9.18) that by using the Axiom of Choice one can
construct uncountably many more.
(b) Let 1 = C(X). An automorphism of 1 sends X to another generator of 1 over
C. It follows from (9.24) below that these are exactly the elements
oAb
cAd
, aJ bc =0.
Therefore Aut(1,C) consists of the maps (X)
_
oAb
cAd
_
, aJ bc =0, and so
Aut(1,C) .PGL
2
(C).
the group of invertible 22 matrices with complex coefcients modulo its centre. Analysts
will note that this is the same as the automorphism group of the Riemann sphere. This is not a
coincidence: the eld of meromorphic functions on the Riemann sphere P
1
C
is C(:) .C(X),
and so there is certainly a map Aut(P
1
C
) Aut(C(:),C), which one can show to be an
isomorphism.
(c) The group Aut(C(X
1
. X
2
),C) is quite complicated there is a map
PGL
3
(C) =Aut(P
2
C
) Aut(C(X
1
. X
2
),C).
but this is very far from being surjective. When there are more Xs, the group is not known.
The group Aut(C(X
1
. . . . . X
n
),C) is the group of birational automorphisms of P
n
C
, and is
called the Cremona group. Its study is part of algebraic geometry. See the Wikipedia.
In this section, we shall be concerned with the groups Aut(1,J) when 1 is a nite
extension of J.
35
36 3. THE FUNDAMENTAL THEOREM OF GALOIS THEORY
PROPOSITION 3.2 If 1 is a splitting eld of a separable polynomial JX|, then
Aut(1,J) has order 1: J|.
PROOF. As is separable, it has deg distinct roots in its splitting eld 1. NowProposition
2.7 shows that there are 1: J| distinct J-homomorphisms 1 1. Because 1 has nite
degree over J, they are automatically isomorphisms.
2
EXAMPLE 3.3 (a) Consider a simple extension 1 =J|, and let be a polynomial with
coefcients in J having as a root. If has no root in 1 other than , then Aut(1,J) =1.
For example, if
3
_
2 denotes the real cube root of 2, then Aut(Q
3
_
2|,Q) =1. Thus, in the
proposition, it is essential that 1 be a splitting eld.
(b) Let J be a eld of characteristic =0, and let a be an element of J that is not a th
power. Then =X
]
a has only one root in a splitting eld 1, and so Aut(1,J) =1.
Thus, in the proposition, it is essential that 1 be a splitting eld of a separable polynomial.
When G is a group of automorphisms of a eld 1, we set
1
G
=Inv(G) ={ 1 [ o =, all o G].
It is a subeld of 1, called the subeld of G-invariants of 1 or the xed eld of G.
In this section, we shall show that, when 1 is the splitting eld of a separable polynomial
in JX| and G =Aut(1,J), then the maps
M Aut(1,M). H Inv(H)
give a one-to-one correspondence between the set of intermediate elds M, J M 1,
and the set of subgroups H of G.
THEOREM 3.4 (E. ARTIN) Let G be a nite group of automorphisms of a eld 1, and let
J =1
G
; then 1: J| _(G: 1).
PROOF. Let G ={o
1
=1. . . . . o
n
], and let
1
. . . . .
n
be n > m elements of 1. We shall
show that the
i
are linearly dependent over J. Consider the system of linear equations
o
1
(
1
)X
1
o
1
(
n
)X
n
=0
.
.
.
o
n
(
1
)X
1
o
n
(
n
)X
n
=0
with coefcients in 1. There are m equations and n > m unknowns, and hence there are
nontrivial solutions in 1. We choose one (c
1
. . . . . c
n
) having the fewest possible nonzero
elements. After renumbering the
i
s, we may suppose that c
1
=0, and then (after multiply-
ing by a scalar) that c
1
J. With these normalizations, well show that all c
i
J. Then the
rst equation

1
c
1

n
c
n
=0
(recall that o
1
=1) will be a linear relation on the
i
.
If not all c
i
are in J, then o
k
(c
i
) =c
i
for some k and i , k =1 =i . On applying o
k
to
the equations
o
1
(
1
)c
1
o
1
(
n
)c
n
=0
.
.
.
o
n
(
1
)c
1
o
n
(
n
)c
n
=0
Separable, normal, and Galois extensions 37
and using that {o
k
o
1
. . . . . o
k
o
n
] is a permutation of {o
1
. . . . . o
n
], we nd that
(c
1
. o
k
(c
2
). . . . . o
k
(c
i
). . . .)
is also a solution to the system of equations (*). On subtracting it from the rst, we obtain a
solution (0. . . . . c
i
o
k
(c
i
). . . .), which is nonzero (look at the i th coordinate), but has more
zeros than the rst solution (look at the rst coordinate) contradiction.
2
COROLLARY 3.5 For any nite group G of automorphisms of a eld 1,
G =Aut(1,1
G
).
PROOF. As G Aut(1,1
G
), we have inequalities
1: 1
G
|
3.4
_ (G: 1) _(Aut(1,1
G
): 1)
2.8a
_ 1: 1
G
|.
These must be equalities, and so G =Aut(1,1
G
).
2
Separable, normal, and Galois extensions
DEFINITION 3.6 An algebraic extension 1,J is said to be separable if the minimum
polynomial of every element of 1 is separable; otherwise, it is inseparable.
Thus, an algebraic extension 1,J is separable if every irreducible polynomial in JX|
having a root in 1 is separable, and it is inseparable if
J is nonperfect, and in particular has characteristic =0, and
there is an element of 1 whose minimal polynomial is of the formg(X
]
), g JX|.
For example, 1 =F
]
(T ) is an inseparable extension of F
]
(T
]
).
DEFINITION 3.7 An algebraic extension 1,J is normal if the minimum polynomial of
every element of 1 splits in 1X|.
In other words, an algebraic extension 1,J is normal if every irreducible polynomial
JX| having a root in 1 splits in 1X|.
Let be an irreducible polynomial of degree m in JX|. If has a root in 1, then
1,J separable == roots of distinct
1,J normal == splits in 1
_
== has m distinct roots in 1.
Therefore, 1,J is normal and separable if and only if, for each 1, the minimum
polynomial of has J|: J| distinct roots in 1.
EXAMPLE 3.8 (a) The eld Q
3
_
2|, where
3
_
2 is the real cube root of 2, is separable but
not normal over Q (because X
3
2 doesnt split in Q|).
(b) The eld F
]
(T ) is normal but not separable over F
]
(T
]
) the minimumpolynomial
of T is the inseparable polynomial X
]
T
]
.
DEFINITION 3.9 Let J be a eld. A nite extension 1 of J is said to be Galois if J is
the xed eld of the group of J-automorphisms of 1. This group is then called the Galois
group of 1 over J, and it is denoted by Gal(1,J).
38 3. THE FUNDAMENTAL THEOREM OF GALOIS THEORY
THEOREM 3.10 For an extension 1,J, the following statements are equivalent:
(a) 1 is the splitting eld of a separable polynomial JX|.
(b) J =1
G
for some nite group G of automorphisms of 1.
(c) 1 is normal, separable, and of nite degree over J.
(d) 1 is Galois over J.
PROOF. (a) == (d). Let G = Aut(1,J), and let J
t
= 1
G
J. Then 1 is also the
splitting eld of regarded as a polynomial with coefcients in J
t
, and is still separable
when it is regarded in this way. Hence Proposition 3.2 shows that
1: J
t
| =

Aut(1,J
t
)

1: J| =[Aut(1,J)[ .
Since Aut(1,J
t
)
(3.5)
= G =Aut(1,J) , we conclude that J =J
t
, and so J =1
G
.
(d) == (b). According to (2.8a) , Gal(1,J) is nite, and so this is obvious.
(b) == (c). By Proposition 3.4, we know that 1: J| _(G: 1); in particular, it is nite.
Let 1 and let be the minimum polynomial of ; we have to prove that splits into
distinct factors in 1X|. Let {
1
=. ....
n
] be the orbit of under the action of G on 1,
and let
g(X) =

(X
i
) =X
n
a
1
X
n-1
a
n
.
Every o G merely permutes the
i
. Since the a
i
are symmetric polynomials in the
i
,
we nd that oa
i
= a
i
for all i , and so g(X) JX|. It is monic, and g() = 0, and so
(X)[g(X) (see the denition of the minimum polynomial p. 17). But also g(X)[(X),
because each
i
is a root of (X) (if
i
=o, then applying o to the equation () =0
gives (
i
) =0). We conclude that (X) =g(X), and so (X) splits into distinct factors
in 1.
(c) == (a). Because 1 has nite degree over J, it is generated over J by a nite
number of elements, say, 1 =J
1
. ....
n
|,
i
1,
i
algebraic over J. Let
i
be the
minimum polynomial of
i
over J. Because 1 is normal over J, each
i
splits in 1, and
so 1 is the splitting eld of =

i
. Because 1 is separable over J, is separable.
2
REMARK 3.11 (a) Let 1 be Galois over J with Galois group G, and let 1. The
elements
1
=,
2
. ....
n
of the orbit of under G are called the conjugates of . In
the course of the proof of (b) == (c) of the above theorem we showed that the minimum
polynomial of is

(X
i
).
(b) Let 1,J be a nite extension of elds, and let G be a nite group of J-automorphisms
of 1. According to (3.10), 1 is the splitting eld over 1
G
of a separable polynomial in
1
G
X|. Therefore, Proposition 2.7 shows that Gal(1,J) has 1: J| elements. Combined
with Artins theorem (3.4), this shows that G =Gal(1,J) and (G: 1) =1: J]. In particular,
an J-automorphism of 1 not in G does not x the elements of 1
G
.
COROLLARY 3.12 Every nite separable extension 1 of J is contained in a nite Galois
extension.
PROOF. Let 1 =J
1
. ....
n
|. Let
i
be the minimum polynomial of
i
over J, and take
1
t
to be the splitting eld of

i
over J.
2
The fundamental theorem of Galois theory 39
COROLLARY 3.13 Let 1 M J; if 1 is Galois over J, then it is Galois over M.
PROOF. We know 1 is the splitting eld of some separable JX|; it is also the splitting
eld of regarded as an element of MX|.
2
REMARK 3.14 When we drop the assumption that 1 is separable over J, we can still say
something. An element of an algebraic extension of J is said to be separable over J if
its minimum polynomial over J is separable. The proof of Corollary 3.12 shows that every
nite extension generated by separable elements is separable. Therefore, the elements of a
nite extension 1 of J that are separable over J form a subeld 1
sep
of 1 that is separable
over J; write 1: J|
sep
=1
sep
: J| (separable degree of 1 over J). If is an algebraically
closed eld containing J, then every J-homomorphism 1
sep
extends uniquely to 1,
and so the number of J-homomorphisms 1 is 1: J|
sep
. When 1 M J (nite
extensions),
1: J|
sep
=1: M|
sep
M: J|
sep
.
In particular,
1 is separable over J 1 is separable over M and M is separable over J.
See Jacobson 1964, I 10, for more details.
DEFINITION 3.15 A nite extension 1 J is called a cyclic, abelian, ..., solvable exten-
sion if it is Galois with cyclic, abelian, ..., solvable Galois group.
The fundamental theorem of Galois theory
THEOREM 3.16 (FUNDAMENTAL THEOREM OF GALOIS THEORY) Let 1 be a Galois ex-
tension of J, and let G =Gal(1,J). The maps H 1
1
and M Gal(1,M) are inverse
bijections between the set of subgroups of G and the set of intermediate elds between 1
and J:
{subgroups of G] -{intermediate elds J M 1].
Moreover,
(a) the correspondence is inclusion-reversing: H
1
H
2
1
1
1
1
1
2
:
(b) indexes equal degrees: (H
1
: H
2
) =1
1
2
: 1
1
1
|;
(c) oHo
-1
-oM, i.e., 1
c1c
1
=o(1
1
); Gal(1,oM) =o Gal(1,M)o
-1
.
(d) H is normal in G 1
1
is normal (hence Galois) over J, in which case
Gal(1
1
,J) .G,H.
PROOF. For the rst statement, we have to show that H 1
1
and M Gal(1,M) are
inverse maps.
Let H be a subgroup of G. Then, as we observed in (3.11b), Gal(1,1
1
) =H.
Let M be an intermediate eld. Then 1 is Galois over M by (3.13), which means that
1
Gal(T{)
=M .
(a) We have the obvious implications:
H
1
H
2
== 1
1
1
1
1
2
== Gal(1,1
1
1
) Gal(1,1
1
2
).
40 3. THE FUNDAMENTAL THEOREM OF GALOIS THEORY
But Gal(1,1
1
i
) =H
i
.
(b) As we observed in (3.11b), for every subgroup H of G, 1: 1
1
| =(Gal(1,1
1
): 1).
This proves (b) in the case H
2
=1, and the general case follows, using that
(H
1
: 1) =(H
1
: H
2
)(H
2
: 1) and 1: 1
1
1
| =1: 1
1
2
|1
1
2
: 1
1
1
|.
(c) For t G and 1, t = oto
-1
(o) =o. Therefore, Gal(1,oM) =
o Gal(1,M)o
-1
, and so o Gal(1,M)o
-1
-oM.
(d) Let H be a normal subgroup of G. Because oHo
-1
=H for all o G, we must
have o1
1
=1
1
for all o G, i.e., the action of G on 1 stabilizes 1
1
. We therefore have
a homomorphism
o o[1
1
: G Aut(1
1
,J)
whose kernel is H. As (1
1
)
G{1
=J, we see that 1
1
is Galois over J (by Theorem 3.10)
and that G,H .Gal(1
1
,J) (by 3.11b).
Conversely, assume that M is normal over J, and write M =J
1
. ....
n
|. For o G,
o
i
is a root of the minimum polynomial of
i
over J, and so lies in M. Hence oM =M,
and this implies that oHo
-1
=H (by (c)).
2
REMARK 3.17 The theorem shows that there is an order reversing bijection between the
intermediate elds of 1,J and the subgroups of G. Using this we can read off more results.
(a) Let M
1
. M
2
. . . . . M
i
be intermediate elds, and let H
i
be the subgroup corresponding
to M
i
(i.e., H
i
= Gal(1,M
i
)). Then (by denition) M
1
M
2
M
i
is the smallest eld
containing all M
i
; hence it must correspond to the largest subgroup contained in all H
i
,
which is
_
H
i
. Therefore
Gal(1,M
1
M
i
) =H
1
... H
i
.
(b) Let H be a subgroup of G and let M =1
1
. The largest normal subgroup contained
in H is N =
_
cG
oHo
-1
(see GT 4.10), and so 1
1
, which is the composite of the elds
oM, is the smallest normal extension of J containing M. It is called the normal, or Galois,
closure of M in 1.
PROPOSITION 3.18 Let 1 and 1 be eld extensions of J contained in some common eld.
If 1,J is Galois, then 11,1 and 1,11 are Galois, and the map
o o[1: Gal(11,1) Gal(1,11)
is an isomorphism.
PROOF. Because 1 is Galois over J, it is the splitting eld of a separable polynomial
JX|. Then 11 is the splitting eld of over 1, and 1 is the splitting
eld of over 1 1. Hence 11,1 and 1,1 1 are Galois. Every
automorphism o of 11 xing the elements of 1 maps roots of to roots
of , and so o1 =1. There is therefore a homomorphism
o o[1: Gal(11,1) Gal(1,11).
If o Gal(11,1) xes the elements of 1, then it xes the elements of
11, and hence is 1. Thus, o o[1 is injective. If 1 is xed by all
o Gal(11,1), then 11. By the fundamental theorem,
11
1 1
11
J
=
=
this implies that the image of o o[1 is Gal(1,11).
2
The fundamental theorem of Galois theory 41
COROLLARY 3.19 Suppose, in the proposition, that 1 is nite over J. Then
11: J| =
1: J|1: J|
11: J|
.
PROOF. According to Proposition 1.20,
11: J| =11: 1|1: J|.
but
11: 1|
3.1S
= 1: 11|
1.20
=
1: J|
11: J|
.
2
PROPOSITION 3.20 Let 1
1
and 1
2
be eld extensions of J contained in some common
eld. If 1
1
and 1
2
are Galois over J, then 1
1
1
2
and 1
1
1
2
are Galois over J, and
o (o[1
1
. o[1
2
): Gal(1
1
1
2
,J) Gal(1
1
,J) Gal(1
2
,J)
is an isomorphism of Gal(1
1
1
2
,J) onto the subgroup
H ={(o
1
. o
2
) [ o
1
[1
1
1
2
=o
2
[1
1
1
2
]
of Gal(1
1
,J) Gal(1
2
,J).
PROOF: Let a 1
1
1
2
, and let be its minimum polynomial over J. Then has
deg distinct roots in 1
1
and deg distinct roots in 1
2
. Since
can have at most deg roots in 1
1
1
2
, it follows that it has deg
distinct roots in 1
1
1
2
. This shows that 1
1
1
2
is normal and
separable over J, and hence Galois (3.10). As 1
1
and 1
2
are
Galois over J, they are splitting elds of separable polynomials

1
.
2
JX|. Now 1
1
1
2
is a splitting eld for
1

2
, and hence
it also is Galois over J. The map o (o[1
1
. o[1
2
) is clearly
an injective homomorphism, and its image is contained in H. We
prove that the image is the whole of H by counting.
1
1
1
2
1
1
1
2
1
1
1
2
J
From the fundamental theorem,
Gal(1
2
,J),Gal(1
2
,1
1
1
2
) .Gal(1
1
1
2
,J),
and so, for each o
1
Gal(1
1
,J), o
1
[1
1
1
2
has exactly 1
2
: 1
1
1
2
| extensions to an
element of Gal(1
2
,J). Therefore,
(H: 1) =1
1
: J|1
2
: 1
1
1
2
| =
1
1
: J| 1
2
: J|
1
1
1
2
: J|
.
which equals 1
1
1
2
: J| by (3.19).
42 3. THE FUNDAMENTAL THEOREM OF GALOIS THEORY
Examples
EXAMPLE 3.21 We analyse the extension Q|,Q, where is a primitive 7th root of 1, say
=e
2ti{T
.
Note that Q| is the splitting eld of the polyno-
mial X
T
1, and that has minimum polynomial
X
6
X
5
X
4
X
3
X
2
X 1
(see 1.41). Therefore, Q| is Galois of degree
6 over Q. For any o Gal(Q|,Q), o =
i
,
some i , 1 _i _6, and the map o i denes an
isomorphism Gal(Q|,Q) (Z,7Z)

. Let o be
the element of Gal(Q|,Q) such that o =
3
.
Then o generates Gal(Q|,Q) because the class
of 3 in (Z,7Z)

generates it (the powers of 3 mod

7 are 3. 2. 6. 4. 5. 1). We investigate the subelds
of Q| corresponding to the subgroups (o
3
) and
(o
2
).
Q|
Q

| Q
_
7|
Q
(c
3
) (c
2
)
(c){(c
3
) (c){(c
2
)
Note that o
3
=
6
=

(complex conjugate of ). The subeld of Q| corresponding to

(o
3
) is Q

|, and

=2cos
2t
T
. Since (o
3
) is a normal subgroup of (o), Q

| is
Galois over Q, with Galois group (o),(o
3
). The conjugates of
1
def
=

are
3
=
3

-3
,

2
=
2

-2
. Direct calculation shows that

3
=

6
i=1

i
=1.

3
=2.

3
=(
6
)(
2

5
)(
3

4
)
=(
3

6
)(
3

4
)
=(
4

6
1
2

5
1
3
)
=1.
Hence the minimum polynomial
1
of

is
g(X) =X
3
X
2
2X 1.
The minimum polynomial of cos
2t
T
=

1
2
is therefore
g(2X)
8
=X
3
X
2
,2X,21,8.
The subeld of Q| corresponding to (o
2
) is generated by =
2

4
. Let
t
=o.
Then (
t
)
2
=7. Hence the eld xed by (o
2
) is Q
_
7|.
EXAMPLE 3.22 We compute the Galois group of a splitting eld 1 of X
5
2 QX|.
1
More directly, on setting X =

in
(X
3
3X) (X
2
2) X 1
one obtains 1
2

6
=0.
Constructible numbers revisited 43
Recall from Exercise 2-3 that 1 =Q. | where is a primitive
5th root of 1, and is a root of X
5
2. For example, we could take
1 to be the splitting eld of X
5
2 in C, with =e
2ti{5
and
equal to the real 5th root of 2. We have the picture at right, and
Q| : Q| =4. Q| : Q| =5.
Because 4 and 5 are relatively prime,
Q. | : Q| =20.
Q. |
Q| Q|
Q
1 1
G{1
Hence G =Gal(Q. |,Q) has order 20, and the subgroups N and H xing Q| and Q|
have orders 5 and 4 respectively. Because Q| is normal over Q (it is the splitting eld
of X
5
1), N is normal in G. Because Q| Q| =Q. |, we have H N =1, and so
G =N
0
H. Moreover, H .G,N .(Z,5Z)

, which is cyclic, being generated by the

class of 2. Let t be the generator of H corresponding to 2 under this isomorphism, and let o
be a generator of N. Thus o() is another root of X
5
2, which we can take to be (after
possibly replacing o by a power). Hence:
_
t =
2
t =
_
o =
o = .
Note that tot
-1
() =to =t() =
2
and it xes ; therefore tot
-1
=o
2
. Thus G
has generators o and t and dening relations
o
5
=1. t
4
=1. tot
-1
=o
2
.
The subgroup H has ve conjugates, which correspond to the ve elds Q
i
|,
o
i
Ho
-i
-o
i
Q| =Q
i
|. 1 _i _5.
Constructible numbers revisited
Earlier, we showed (1.36) that a real number is constructible if and only if it is con-
tained in a subeld of R of the form Q
_
a
1
. . . . .
_
a
i
| with each a
i
a positive element of
Q
_
a
1
. . . . .
_
a
i-1
|. In particular
constructible == Q|: Q| =2
x
some s. (2)
Now we can prove a partial converse to this last statement.
THEOREM 3.23 If is contained in a subeld of R that is Galois of degree 2
i
over Q, then
it is constructible.
PROOF. Suppose 1 Rwhere 1 is Galois of degree 2
i
over Q, and let G =Gal(1,Q).
Because nite -groups are solvable (GT 6.7), there exists a sequence of groups
{1] =G
0
G
1
G
2
G
i
=G
with G
i
,G
i-1
of order 2. Correspondingly, there will be a sequence of elds,
1 =1
0
1
1
1
2
1
i
=Q
with 1
i-1
of degree 2 over 1
i
. The next lemma shows that 1
i
= 1
i-1

_
a
i
| for some
a
i
1
i-1
, and a
i
>0 because otherwise 1
i
would not be real. This proves the theorem.
2
44 3. THE FUNDAMENTAL THEOREM OF GALOIS THEORY
LEMMA 3.24 Let 1,J be a quadratic extension of elds of characteristic = 2. Then
1 =J
_
J| for some J J.
PROOF. Let 1, J, and let X
2
bX c be the minimum polynomial of . Then
=
-b
_
b
2
-4c
2
, and so 1 =J
_
b
2
4c|.
2
COROLLARY 3.25 If is a prime of the form 2
k
1, then cos
2t
]
is constructible.
PROOF. The eld Qe
2ti{]
| is Galois over Q with Galois group G .(Z,Z)

, which has
order 1 =2
k
. The eld Qcos
2t
]
| is contained in Qe
2ti{]
|, and therefore is Galois of
degree dividing 2
k
(fundamental theorem 3.16 and 1.20). As Qcos
2t
]
| is a subeld of R,
we can apply the theorem.
2
Thus a regular -gon, prime, is constructible if and only if is a Fermat prime,
i.e., of the form 2
2
r
1. For example, we have proved that the regular 65537-polygon is
constructible, without (happily) having to exhibit an explicit formula for cos
2t
6553T
.
REMARK 3.26 The converse to (2) is false. The polynomial (X) =X
4
4X 2 QX|
is irreducible, and well show below (4.9) that the Galois group of the splitting eld 1 over
Q of the polynomial is S
4
. Each root of (X) lies in an extension of degree 2
2
of Q . If the
four roots of (X) were constructible, then all the elements of 1 would be constructible
(1.36a). Let H be a Sylow subgroup of S
4
. Then 1
1
has odd degree over Q, and so the
elements of 1
1
Q cant be constructible.
2
The Galois group of a polynomial
If the polynomial JX| is separable, then its splitting eld J
(
is Galois over J, and we
call Gal(J
(
,J) the Galois group G
(
of .
Let =

n
i=1
(X
i
) in a splitting eld J
(
. We know that the elements of Gal(J
(
,J)
map roots of to roots of , i.e., they map the set {
1
.
2
. . . . .
n
] into itself. Being
automorphisms, they dene permutations of {
1
.
2
. . . . .
n
], and as the
i
generated J
(
,
an element of Gal(J
(
,J) is uniquely determined by the permutation it denes. Thus
G
(
can be identied with a subset of Sym({
1
.
2
. . . . .
n
]) ~S
n
(symmetric group on n
symbols). In fact, G
(
consists exactly of the permutations o of {
1
.
2
. . . . .
n
] such that,
for 1 JX
1
. . . . . X
n
|,
1(
1
. . . . .
n
) =0 == 1(o
1
. . . . . o
n
) =0.
This gives a description of G
(
without mentioning elds or abstract groups (neither of which
were available to Galois).
Note that this shows again that (G
(
: 1), hence J
(
: J|, divides deg( ).
2
As Shuichi Otsuka has pointed out to me, it is possible to prove this without appealing to the Sylow
theorems. If a root of (X) were constructible, then there would exist a tower of quadratic extensions
Q| M Q. By Galois theory, the groups Gal(1,M) Gal(1,Q|) have orders 12 and 6 respectively.
As Gal(1,Q) =S
4
, Gal(1,M) would be
4
. But
4
has no subgroup of order 6, a contradiction. Thus no
root of (X) is constructible. (Actually Gal(1,Q|) =S
3
, but that does not matter here.)
Solvability of equations 45
Solvability of equations
For a polynomial JX|, we say that (X) =0 is solvable in radicals if its solutions
can be obtained by the algebraic operations of addition, subtraction, multiplication, division,
and the extraction of mth roots, or, more precisely, if there exists a tower of elds
J =J
0
J
1
J
2
J
n
such that
(a) J
i
=J
i-1

i
|,
n
i
i
J
i-1
;
(b) J
n
contains a splitting eld for .
THEOREM 3.27 (GALOIS, 1832) Let J be a eld of characteristic zero. The equation
=0 is solvable in radicals if and only if the Galois group of is solvable.
We shall prove this later (5.33). Also we shall exhibit polynomials (X) QX| with
Galois group S
n
, which are therefore not solvable when n _5 by GT 4.37.
REMARK 3.28 If J has characteristic , then the theorem fails for two reasons:
(a) may not be separable, and so not have a Galois group;
(b) X
]
X a =0 is not solvable by radicals even though it is separable with abelian
Galois group (cf. Exercise 2-2).
If the denition of solvable is changed to allow extensions of the type in (b) in the chain, and
is required to be separable, then the theorem becomes true in characteristic .
NOTES Much of what has been written about Galois is unreliable see Tony Rothman, Genius
and Biographers: The Fictionalization of Evariste Galois, Amer. Math. Mon. 89, 84 (1982). Revised
version available online at various locations.
For a careful explanation of Galoiss Premier M emoire, see Edwards, Harold M., Galois for
21st-century readers. Notices AMS 59 (2012), no. 7, 912923.
Exercises
3-1 Let J be a eld of characteristic 0. Show that J(X
2
) J(X
2
X) =J (intersection
inside J(X)). [Hint: Find automorphisms o and t of J(X), each of order 2, xing J(X
2
)
and J(X
2
X) respectively, and show that ot has innite order.]
3-2
3
Let be an odd prime, and let be a primitive th root of 1 in C. Let 1 =Q|,
and let G =Gal(1,Q); thus G =(Z,())

. Let H be the subgroup of index 2 in G. Put

=

i1

i
and =

iG\1

i
. Show:
(a) and are xed by H;
(b) if o G\H, then o =, o =.
Thus and are roots of the polynomial X
2
X QX|. Compute and show
that the xed eld of H is Q
_
| when 1 mod 4 and Q
_
| when 3 mod 4.
3
This problem shows that every quadratic extension of Q is contained in a cyclotomic extension of Q. The
Kronecker-Weber theorem says that every abelian extension of Q is contained in a cyclotomic extension.
46 3. THE FUNDAMENTAL THEOREM OF GALOIS THEORY
3-3 Let M =Q
_
2.
_
3| and 1 =M
_
(
_
22)(
_
33)| (subelds of R).
(a) Show that M is Galois over Q with Galois group the 4-group C
2
C
2
.
(b) Show that 1 is Galois over Q with Galois group the quaternion group.
3-4 Let 1 be a Galois extension of J with Galois group G, and let 1 be the xed eld of a
subgroup H of G. Show that the automomorphism group of 1,J is N,H where N is the
normalizer of H in G.
CHAPTER 4
Computing Galois Groups
In this chapter, we investigate general methods for computing Galois groups.
When is G

n
?
Let o be a permutation of the set {1. 2. . . . . n]. The pairs (i. ) with i < but o(i ) > o( )
are called the inversions of o, and o is said to be even or odd according as the number of
inversions is even or odd. The signature of o, sign(o), is 1 or 1 according as o is even or
odd. We can dene the signature of a permutation o of any set S of n elements by choosing
a numbering of the set and identifying o with a permutation of {1. . . . . n]. Then sign is the
unique homomorphism Sym(S) {1] such that sign(o) =1 for every transposition. In
particular, it is independent of the choice of the numbering. See GT, 4.25.
Now consider a monic polynomial
(X) =X
n
a
1
X
n-1
a
n
and let (X) =

n
i=1
(X
i
) in some splitting eld. Set
z( ) =

1_i~}_n
(
i

}
). D( ) =z( )
2
=

1_i~}_n
(
i

}
)
2
.
The discriminant of is dened to be D( ). Note that D( ) is nonzero if and only if
has only simple roots, i.e., if is separable with no multiple factors. Let G
(
be the Galois
group of , and identify it with a subgroup of Sym({
1
. . . . .
n
]) (as on p. 44).
PROPOSITION 4.1 Let JX| be a separable polynomial, and let o G
(
.
(a) oz( ) =sign(o)z( ), where sign(o) is the signature of o.
(b) oD( ) =D( ).
PROOF. Each inversion of o introduces a negative sign into oz( ), and so (a) follows from
the denition of sign(o). The equation in (b) is obtained by squaring that in (a).
2
While z( ) depends on the choice of the numbering of the roots of , D( ) does not.
COROLLARY 4.2 Let (X) JX| be of degree n and have only simple roots. Let J
(
be
a splitting eld for , so that G
(
=Gal(J
(
,J).
47
48 4. COMPUTING GALOIS GROUPS
(a) The discriminant D( ) J.
(b) The subeld of J
(
corresponding to
n
G
(
is Jz( )|. Hence
G
(

n
z( ) J D( ) is a square in J.
PROOF. (a) The discriminant of is an element of J
(
xed by G
(
def
= Gal(J
(
,J), and
hence lies in J (by the fundamental theorem of Galois theory).
(b) Because has simple roots, z( ) =0, and so the formula oz( ) =sign(o)z( )
shows that an element of G
(
xes z( ) if and only if it lies in
n
. Thus, under the Galois
correspondence,
G
(

n
-Jz( )|.
Hence,
G
(

n
=G
(
Jz( )| =J.
2
The roots of X
2
bX c are
-b
_
b
2
-4c
2
and so
z(X
2
bX c) =
_
b
2
4c (or
_
b
2
4c),
D(X
2
bX c) =b
2
4c.
Although there is a not a universal formula for the roots of in terms its coefcients when
the degree of is >4 , there is for its discriminant. For example,
D(X
3
bX c) =4b
3
27c
2
.
By completing the cube, one can put any cubic polynomial in this form (in characteristic
=3).
The formulas for the discriminant rapidly become very complicated, for example, that
for X
5
aX
4
bX
3
cX
2
JX e has 59 terms. Fortunately, PARI knows them. For
example, typing poldisc(X^3+a*X^2+b*X+c,X) returns the discriminant of X
3
aX
2

bX c, namely,
4ca
3
b
2
a
2
18cba(4b
3
27c
2
).
REMARK 4.3 Suppose J R. Then D( ) will not be a square if it is negative. It is known
that the sign of D( ) is (1)
x
where 2s is the number of nonreal roots of in C (see ANT
2.40). Thus if s is odd, then G
(
is not contained in
n
. This can be proved more directly by
noting that complex conjugation acts on the roots as the product of s disjoint transpositions.
Of course the converse is not true: when s is even, G
(
is not necessarily contained in

n
.
When is G

transitive?
PROPOSITION 4.4 Let (X) JX| have only simple roots. Then (X) is irreducible if
and only if G
(
permutes the roots of transitively.
PROOF. ==: If and are two roots of (X) in a splitting eld J
(
for , then they both
have (X) as their minimum polynomial, and so J| and J| are both stem elds for .
Hence, there is an J-isomorphism
J| .J|. -.
Polynomials of degree at most three 49
Write J
(
= J
1
.
2
. ...| with
1
= and
2
.
3
. . . . the other roots of (X). Then the
J-homomorphism : J| J
(
extends (step by step) to an J-homomorphism
J
(
J
(
(use 2.2b), which is an J-isomorphism sending to .
==: Let g(X) JX| be an irreducible factor of , and let be one of its roots. If
is a second root of , then (by assumption) =o for some o G
(
. Now, because g has
coefcients in J,
g(o) =og() =0.
and so is also a root of g. Therefore, every root of is also a root of g, and so (X) =
g(X).
2
Note that when (X) is irreducible of degree n, n[(G
(
: 1) because J|: J| =n and
J|: J| divides J
(
: J| =(G
(
: 1). Thus G
(
is a transitive subgroup of S
n
whose order is
divisible by n.
Polynomials of degree at most three
EXAMPLE 4.5 Let (X) JX| be a polynomial of degree 2. Then is inseparable
J has characteristic 2 and (X) =X
2
a for some a J J
2
. If is separable, then
G
(
=1(=
2
) or S
2
according as D( ) is a square in J or not.
EXAMPLE 4.6 Let (X) JX| be a polynomial of degree 3. We can assume to be
irreducible, for otherwise we are essentially back in the previous case. Then is inseparable
if and only if J has characteristic 3 and (X) = X
3
a for some a J \ J
3
. If is
separable, then G
(
is a transitive subgroup of S
3
whose order is divisible by 3. There are
only two possibilities: G
(
=
3
or S
3
according as D( ) is a square in J or not. Note that

3
is generated by the cycle (123).
For example, X
3
3X1 QX| is irreducible (see 1.12), its discriminant is 4(3)
3

27 =81 =9
2
, and so its Galois group is
3
.
On the other hand, X
3
3X 1 QX| is also irreducible (apply 1.11), but its discrim-
inant is 135 which is not a square in Q, and so its Galois group is S
3
.
Quartic polynomials
Let (X) be a quartic polynomial without multiple roots. In order to determine G
(
we shall
exploit the fact that S
4
has
V ={1. (12)(34). (13)(24). (14)(23)]
as a normal subgroup it is normal because it contains all elements of type 22 (GT 4.29).
Let 1 be a splitting eld of , and let (X) =

(X
i
) in 1. We identify the Galois
group G
(
of with a subgroup of the symmetric group Sym({
1
.
2
.
3
.
4
]). Consider
the partially symmetric elements
=
1

4
=
1

4
; =
1

3
.
50 4. COMPUTING GALOIS GROUPS
They are distinct because the
i
are distinct; for example,
=
1
(
2

3
)
4
(
3

2
) =(
1

4
)(
2

3
).
The group Sym({
1
.
2
.
3
.
4
]) permutes {. . ;] transitively. The stabilizer of each of
. . ; must therefore be a subgroup of index 3 in S
4
, and hence has order 8. For example,
the stabilizer of is ((1234). (13)). Groups of order 8 in S
4
are Sylow 2-subgroups. There
are three of them, all isomorphic to D
4
. By the Sylow theorems, V is contained in a Sylow
2-subgroup; in fact, because the Sylow 2-subgroups are conjugate and V is normal, it is
contained in all three. It follows that V is the intersection of the three Sylow 2-subgroups.
Each Sylow 2-subgroup xes exactly one of . . or ;, and therefore their intersection V is
the subgroup of Sym({
1
.
2
.
3
.
4
]) xing , , and ;.
LEMMA 4.7 The xed eld of G
(
V is J. . ;|. Hence
J. . ;| is Galois over J with Galois group G
(
,G
(
V .
PROOF. The above discussion shows that the subgroup of G
(
of
elements xing J. . ;| is G
(
V , and so 1
G
f
V
=J. . ;|
by the fundamental theorem of Galois theory. The remaining state-
ments follow from the fundamental theorem using that V is nor-
mal.
2
1
J. . ;|
J
G
f
V
G
f
{G
f
V
Let M =J. . ;|, and let g(X) =(X )(X )(X ;) MX| it is called the
resolvent cubic of . Every permutation of the
i
(a fortiori, every element of G
(
) merely
permutes . . ;, and so xes g(X). Therefore (by the fundamental theorem) g(X) has
coefcients in J. More explicitly, we have:
LEMMA 4.8 The resolvent cubic of =X
4
bX
3
cX
2
JX e is
g =X
3
cX
2
(bJ 4e)X b
2
e 4ce J
2
.
The discriminants of and g are equal.
SKETCH OF PROOF. Expand =(X
1
)(X
2
)(X
3
)(X
4
) to express b. c. J. e
in terms of
1
.
2
.
3
.
4
. Expand g =(X )(X )(X ;) to express the coefcients
of g in terms of
1
.
2
.
3
.
4
, and substitute to express them in terms of b. c. J. e.
2
Now let be an irreducible separable quartic. Then G =G
(
is a transitive subgroup of
S
4
whose order is divisible by 4. There are the following possibilities for G:
G (GV : 1) (G: V G)
S
4
4 6

4
4 3
V 4 1
D
4
4 2
C
4
2 2
(GV : 1) =1: M|
(G: V G) =M: J|
Quartic polynomials 51
The groups of type D
4
are the Sylow 2-subgroups discussed above, and the groups of type
C
4
are those generated by cycles of length 4.
We can compute (G: V G) from the resolvent cubic g, because G,V G =Gal(M,J)
and M is the splitting eld of g. Once we know (G: V G), we can deduce G except in the
case that it is 2. If M: J| =2, then GV =V or C
2
. Only the rst group acts transitively
on the roots of , and so (from 4.4) we see that in this case G =D
4
or C
4
according as
is irreducible or not in MX|.
EXAMPLE 4.9 Consider (X) = X
4
4X 2 QX|. It is irreducible by Eisensteins
criterion (1.16), and its resolvent cubic is g(X) = X
3
8X 16, which is irreducible
because it has no roots in F
5
. The discriminant of g(X) is 4864, which is not a square, and
so the Galois group of g(X) is S
3
. From the table, we see that the Galois group of (X) is
S
4
.
EXAMPLE 4.10 Consider (X) =X
4
4X
2
2 QX|. It is irreducible by Eisensteins
criterion (1.16), and its resolvent cubic is (X 4)(X
2
8); thus M =Q
_
2|. From the
table we see that G
(
is of type D
4
or C
4
, but factors over M (even as a polynomial in
X
2
), and hence G
(
is of type C
4
.
EXAMPLE 4.11 Consider (X) =X
4
10X
2
4 QX|. It is irreducible in QX| be-
cause (by inspection) it is irreducible in ZX|. Its resolvent cubic is (X10)(X4)(X4),
and so G
(
is of type V .
EXAMPLE 4.12 Consider (X) =X
4
2 QX|. It is irreducible by Eisensteins criterion
(1.16), and its resolvent cubic is g(X) =X
3
8X. Hence M =Qi
_
2|. One can check
that is irreducible over M, and G
(
is of type D
4
.
Alternatively, analyse the equation as in (3.22).
As we explained in (1.29), PARI knows how to factor polynomials with coefcients in
Q|.
EXAMPLE 4.13 (From the web, sci.math.research, search for nal analysis.) Consider
(X) = X
4
2cX
3
JX
2
2cJX Jc
2
ZX| with a > 0, b > 0, c > 0, a > b and
J = a
2
b
2
. Let r = J,c
2
and let n be the unique positive real number such that r =
n
3
,(n
2
4). Let m be the number of roots of (X) in Z (counted with multiplicities). The
Galois group of is as follows:
If m=0 and n not rational, then G is S
4
.
If m=1 and n not rational then G is S
3
.
If n is rational and n
2
4 is not a square then G =D4.
If n is rational and n
2
4 is a square then G =V =C
2
C
2
.
This covers all possible cases. The hard part was to establish that m=2 could never happen.
ASIDE 4.14 For a discussion of whether the method of solving a quartic by reducing to a cubic
generalizes to other even degrees, see mo149099.
52 4. COMPUTING GALOIS GROUPS
Examples of polynomials with S

as Galois group over Q

The next lemma gives a criterion for a subgroup of S
]
to be the whole of S
]
.
LEMMA 4.15 For prime, the symmetric group S
]
is generated by any transposition and
any -cycle.
PROOF. After renumbering, we may assume that the transposition is t =(12), and we may
write the -cycle o so that 1 occurs in the rst position, o =(1i
2
i
]
). Now some power
of o will map 1 to 2 and will still be a -cycle (here is where we use that is prime). After
replacing o with the power, we have o =(12
3
. . .
]
), and after renumbering again, we
have o =(123. . . ). Now
(i i 1) =o
i
(12)o
-i
(see GT 4.29) and so lies in the subgroup generated by o and t. These transpositions generate
S
]
.
2
PROPOSITION 4.16 Let be an irreducible polynomial of prime degree in QX|. If
splits in C and has exactly two nonreal roots, then G
(
=S
]
.
PROOF. Let 1 be the splitting eld of in C, and let 1 be a root of . Because is
irreducible, Q|: Q| =deg =, and so [1: Q| =(G
(
: 1). Therefore G
(
contains an
element of order (Cauchys theorem, GT 4.13), but the only elements of order in S
]
are
-cycles (here we use that is prime again).
Let o be complex conjugation on C. Then o transposes the two nonreal roots of (X)
and xes the rest. Therefore G
(
S
]
and contains a transposition and a -cycle, and so is
the whole of S
]
.
2
It remains to construct polynomials satisfying the conditions of the Proposition.
EXAMPLE 4.17 Let _5 be a prime number. Choose a positive even integer m and even
integers
n
1
<n
2
< <n
]-2
.
and let
g(X) =(X
2
m)(X n
1
)...(X n
]-2
).
The graph of g crosses the .-axis exactly at the points n
1
. . . . . n
]-2
, and it doesnt have a
local maximum or minimum at any of those points (because the n
i
are simple roots). Thus
e =min

0
(x)=0
[g(.)[ >0, and we can choose an odd positive integer n such that
2
n
<e.
Consider
(X) =g(X)
2
n
.
As
2
n
<e, the graph of also crosses the .-axis at exactly 2 points, and so has exactly
two nonreal roots. On the other hand, when we write
n(X) =nX
]
a
1
X
]-1
a
]
.
the a
i
are all even and a
]
is not divisible by 2
2
, and so Eisensteins criterion implies that
is irreducible. Over R, has 2 linear factors and one quadratic factor, and so it certainly
splits over C (high school algebra). Therefore, the proposition applies to .
1
1
If m is taken sufciently large, then g(X) 2 will have exactly two nonreal roots, i.e., we can take n =1,
but the proof is longer (see Jacobson 1964, p107, who credits the example to Brauer). The shorter argument in
the text was suggested to me by Martin Ward.
Finite elds 53
EXAMPLE 4.18 The reader shouldnt think that, in order to have Galois group S
]
, a polyno-
mial must have exactly two nonreal roots. For example, the polynomial X
5
5X
3
4X 1
has Galois group S
5
but all of its roots are real.
Finite elds
Let F
]
=Z,Z, the eld of elements. As we noted in 1, every eld 1 of characteristic
contains a copy of F
]
, namely, {m1
T
[ m Z]. No harm results if we identify F
]
with
this subeld of 1.
Let 1 be a eld of degree n over F
]
. Then 1 has q =
n
elements, and so 1

is a group
of order q 1. Hence the nonzero elements of 1 are roots of X
q-1
1, and all elements of
1 (including 0) are roots of X
q
X. Hence 1 is a splitting eld for X
q
X, and so any
two elds with q elements are isomorphic.
PROPOSITION 4.19 Every extension of nite elds is simple.
PROOF. Consider 1 J. Then 1

is a nite subgroup of the multiplicative group of a

eld, and hence is cyclic (see Exercise 1-3). If generates 1

as a multiplicative group,
then certainly 1 =J|.
2
Now let 1 be the splitting eld of (X) =X
q
X, q =
n
. The derivative
t
(X) =1,
which is relatively prime to (X) (in fact, to every polynomial), and so (X) has q distinct
roots in 1. Let S be the set of its roots. Then S is obviously closed under multiplication and
the formation of inverses, but it is also closed under subtraction: if a
q
=a and b
q
=b, then
(ab)
q
=a
q
b
q
=ab.
Hence S is a eld, and so S =1. In particular, 1 has
n
elements.
PROPOSITION 4.20 For each power q =
n
there is a eld F
q
with q elements. It is
the splitting eld of X
q
X, and hence any two such elds are isomorphic. Moreover,
F
q
is Galois over F
]
with cyclic Galois group generated by the Frobenius automorphism
o(a) =a
]
.
PROOF. Only the nal statement remains to be proved. The eld F
q
is Galois over F
]
because it is the splitting eld of a separable polynomial. We noted in (1.4) that .
c
.
]
is an automorphism of F
q
. An element a of F
q
is xed by o if and only if a
]
=a, but F
]
consists exactly of such elements, and so the xed eld of (o) is F
]
. This proves that F
q
is
Galois over F
]
and that (o) =Gal(F
q
,F
]
) (see 3.11b).
2
COROLLARY 4.21 Let 1 be a eld with
n
elements. For each divisor m of n, m_0, 1
contains exactly one eld with
n
elements.
PROOF. We know that 1 is Galois over F
]
and that Gal(1,F
]
) is the cyclic group of order
n generated by o. The group (o) has one subgroup of order n,m for each m dividing n,
namely, (o
n
), and so 1 has exactly one subeld of degree m over F
]
for each m dividing n,
namely, 1
(c
m
)
. Because it has degree m over F
]
, 1
(c
m
)
has
n
elements.
2
54 4. COMPUTING GALOIS GROUPS
COROLLARY 4.22 Each monic irreducible polynomial of degree J[n in F
]
X| occurs
exactly once as a factor of X
]
n
X; hence, the degree of the splitting eld of is _J.
PROOF. First, the factors of X
]
n
X are distinct because it has no common factor with
its derivative. If (X) is irreducible of degree J, then (X) has a root in a eld of degree
J over F
]
. But the splitting eld of X
]
n
X contains a copy of every eld of degree
J over F
]
with J[n. Hence some root of X
]
n
X is also a root of (X), and therefore
(X)[X
]
n
X. In particular, divides X
]
d
X, and therefore it splits in its splitting
eld, which has degree J over F
]
.
2
PROPOSITION 4.23 Let F be an algebraic closure of F
]
. Then F contains exactly one eld
F
]
n for each integer n _1, and F
]
n consists of the roots of X
]
n
X. Moreover,
F
]
m F
]
n m[n.
The partially ordered set of nite subelds of F is isomorphic to the set of integers n _ 1
partially ordered by divisibility.
PROOF. Obvious from what we have proved.
2
PROPOSITION 4.24 The eld F
]
has an algebraic closure F.
PROOF. Choose a sequence of integers 1 =n
1
< n
2
< n
3
< such that n
i
[n
i1
for all
i , and every integer n divides some n
i
. For example, let n
i
= i . Dene the elds F
]
n
i
inductively as follows: F
]
n
1
= F
]
; F
]
n
i
is the splitting eld of X
]
n
i
X over F
]
n
i1
.
Then, F
]
n
1
F
]
n
2
F
]
n
3
, and we dene F =
_
F
]
n
i
. As a union of a chain of elds
algebraic over F
]
, it is again a eld algebraic over F
]
. Moreover, every polynomial in F
]
X|
splits in F, and so it is an algebraic closure of F (by 1.44).
2
REMARK 4.25 Since the F
]
ns are not subsets of a xed set, forming the union requires
explanation. Dene S to be the disjoint union of the F
]
n. For a. b S, set a -b if a =b in
one of the F
]
n. Then - is an equivalence relation, and we let F =S, -.
PARI factors polynomials modulo very quickly. Recall that the syntax is
factormod(f(X),p). For example, to obtain a list of all monic polynomials of degree 1. 2.
or 4 over F
5
, ask PARI to factor X
625
X modulo 5 (note that 625 =5
4
).
ASIDE 4.26 In one of the few papers published during his lifetime, Galois dened nite elds of
arbitrary prime power order and established their basic properties, for example, the existence of a
primitive element (Notices AMS, Feb. 2003, p. 198). For this reason nite elds are often called
Galois elds and the eld with q elements is often denoted by GF(q).
Computing Galois groups over Q
In the remainder of this chapter, I sketch a practical method for computing Galois groups
over Q and similar elds. Recall that for a separable polynomial JX|, J
(
denotes a
splitting eld for J, and G
(
=Gal(J
(
,J) denotes the Galois group of J. Moreover, G
(
permutes the roots
1
.
2
. . . . of in J
(
:
G Sym{
1
.
2
. . . .].
The rst result generalizes Proposition 4.4.
Computing Galois groups over Q 55
PROPOSITION 4.27 Let (X) be a polynomial in JX| with only simple roots, and suppose
that the orbits of G
(
acting on the roots of have m
1
. . . . . m
i
elements respectively. Then
factors as =
1

i
with
i
irreducible of degree m
i
.
PROOF. We may assume that is monic. Let
1
. . . . .
n
, m=deg , be the roots of (X)
in J
(
. The monic factors of (X) in J
(
X| correspond to subsets S of {
1
. . . . .
n
],
S -
S
=

S
(X ),
and
S
is xed under the action of G
(
(and hence has coefcients in J) if and only if S
is stable under G
(
. Therefore the irreducible factors of in JX| are the polynomials S
corresponding to minimal subsets S of {
1
. . . . .
n
] stable under G
(
, but these subsets S
are precisely the orbits of G
(
in {
1
. . . . .
n
].
2
REMARK 4.28 Note that the proof shows the following: let {
1
. . . . .
n
] =
_
O
i
be the
decomposition of {
1
. . . . .
n
] into a disjoint union of orbits for the group G
(
; then
=

i
.
i
=

i
O
i
(X
i
)
is the decomposition of into a product of irreducible polynomials in JX|.
Now suppose J is nite, with
n
elements say. Then G
(
is a cyclic group generated by
the Frobenius automorphism o: . .
]
. When we regard o as a permutation of the roots of
, then distinct orbits of o correspond to the factors in its cycle decomposition (GT 4.26).
Hence, if the degrees of the distinct irreducible factors of are m
1
. m
2
. . . . . m
i
, then o has
a cycle decomposition of type
m
1
m
i
=deg.
LEMMA 4.29 Let 1 be a unique factorization domain with eld of fractions J, and let
be a monic polynomial in 1X|. Let 1 be a prime ideal in 1, and let

be the image
of in (1,1)X|. Assume that neither nor

has a multiple root. Then the roots

1
. . . . .
n
of lie in some nite extension 1
t
of 1, and their reductions
i
modulo
11
t
are the roots of

. Moreover G

(
G
(
when both are identied with subgroups of
Sym{
1
. . . . .
n
] =Sym{
1
. . . . .
n
].
PROOF. Omitted see van der Waerden, Modern Algebra, I, 61 (second edition) or ANT
3.41.
2
On combining these results, we obtain the following theorem.
THEOREM 4.30 (DEDEKIND) Let (X) ZX| be a monic polynomial of degree m, and
let be a prime such that mod has simple roots (equivalently, D( ) is not divisible by
). Suppose that

=

i
with
i
irreducible of degree m
i
in F
]
X|. Then G
(
contains
an element whose cycle decomposition is of type
m=m
1
m
i
.
56 4. COMPUTING GALOIS GROUPS
EXAMPLE 4.31 Consider X
5
X 1. Modulo 2, this factors as
(X
2
X 1)(X
3
X
2
1).
and modulo 3 it is irreducible. Hence G
(
contains (i k)(lmn) and (12345), and so also
((i k)(lmn))
3
=(i k). Therefore G
(
=S
5
by (4.15).
LEMMA 4.32 A transitive subgroup of H S
n
containing a transposition and an (n1)-
cycle is equal to S
n
.
PROOF. After possibly renumbering, we may suppose the (n1)-cycle is (123. . . n1).
Because of the transitivity, the transposition can be transformed into (i n), some 1 _i _n1.
Conjugating (i n) by (123. . . n1) and its powers will transform it into (1n). (2n). . . . . (n
1n), and these elements obviously generate S
n
.
2
EXAMPLE 4.33 Select separable monic polynomials of degree n,
1
.
2
.
3
with coef-
cients in Z such that:
(a)
1
is irreducible modulo 2;
(b)
2
=(degree 1)(irreducible of degree n1) mod 3;
(c)
3
=(irreducible of degree 2)(product of 1 or 2 irreducible polynomials of odd degree)
mod 5.
Take
=15
1
10
2
6
3
.
Then
(i) G
(
is transitive (it contains an n-cycle because
1
mod 2);
(ii) G
(
contains a cycle of length n1 (because
2
mod 3);
(iii) G
(
contains a transposition (because
3
mod 5, and so it contains the product of a
transposition with a commuting element of odd order; on raising this to an appropriate
odd power, we are left with the transposition). Hence G
(
is S
n
.
The above results give the following strategy for computing the Galois group of an
irreducible polynomial QX|. Factor modulo a sequence of primes not dividing
D( ) to determine the cycle types of the elements in G
(
a difcult theorem in number
theory, the effective Chebotarev density theorem, says that if a cycle type occurs in G
(
, then
this will be seen by looking modulo a set of prime numbers of positive density, and will
occur for a prime less than some bound. Now look up a table of transitive subgroups of S
n
with order divisible by n and their cycle types. If this doesnt sufce to determine the group,
then look at its action on the set of subsets of r roots for some r.
See, Butler and McKay, The transitive groups of degree up to eleven, Comm. Algebra 11
(1983), 863911. This lists all transitive subgroups of S
n
, n _11, and gives the cycle types
of their elements and the orbit lengths of the subgroup acting on the r-sets of roots. With
few exceptions, these invariants are sufcient to determine the subgroup up to isomorphism.
PARI can compute Galois groups for polynomials of degree _11 over Q. The syntax is
polgalois(f) where is an irreducible polynomial of degree _11 (or _7 depending on
your setup), and the output is (n. s. k.name) where n is the order of the group, s is 1 or
1 according as the group is a subgroup of the alternating group or not, and name is the
name of the group. For example, polgalois(X^5-5*X^3+4*X-1) (see 4.18) returns the
Exercises 57
symmetric group S
5
, which has order 120, polgalois(X^11-5*X^3+4*X-1) returns the
symmetric group S
11
, which has order 39916800, and
polgalois(X^12-5*X^3+4*X-1) returns an apology. The reader should use PARI to
check the examples 4.94.12.
See also, Soicher and McKay, Computing Galois groups over the rationals, J. Number
Theory, 20 (1985) 273281.
Exercises
4-1 Find the splitting eld of X
n
1 F
]
X|.
4-2 Find the Galois group of X
4
2X
3
8X 3 over Q.
4-3 Find the degree of the splitting eld of X
S
2 over Q.
4-4 Give an example of a eld extension 1,J of degree 4 such that there does not exist a
eld M with J M 1, M: J| =2.
4-5 List all irreducible polynomials of degree 3 over F
T
in 10 seconds or less (there are
112).
4-6 It is a thought-provoking question that few graduate students would know how to
approach the question of determining the Galois group of, say,
X
6
2X
5
3X
4
4X
3
5X
2
6X 7.
[over Q].
(a) Can you nd it?
(b) Can you nd it without using the polgalois command in PARI?
4-7 Let (X) =X
5
aX b, a. b Q. Show that G
(
~D
5
(dihedral group) if and only
if
(a) (X) is irreducible in QX|, and
(b) the discriminant D( ) =4
4
a
5
5
5
b
4
of (X) is a square, and
(c) the equation (X) =0 is solvable by radicals.
4-8 Show that a polynomial of degree n =

k
i=1

i
i
i
is irreducible over F
q
if and only
if gcd((.). .
q
n=p
i
.) =1 for all i .
4-9 Let (X) be an irreducible polynomial in QX| with both real and nonreal roots. Show
that its Galois group is nonabelian. Can the condition that is irreducible be dropped?
4-10 Let J be a Galois extension of Q, and let be an element of J such that J
2
is not
xed by the action of Gal(J,Q) on J

,J
2
. Let =
1
. . . . .
n
be the orbit of under
Gal(J,Q). Show:
(a) J
_

1
. . . . .
_

n
|,J is Galois with commutative Galois group contained in (Z,2Z)
n
.
(b) J
_

1
. . . . .
_

n
|,Qis Galois with noncommutative Galois group contained in (Z,2Z)
n

Gal(J,Q). (Cf. mo113794.)

CHAPTER 5
Applications of Galois Theory
In this chapter, we apply the fundamental theorem of Galois theory to obtain other results
about polynomials and extensions of elds.
Primitive element theorem.
Recall that a nite extension of elds 1,J is simple if 1 =J| for some element of
1. Such an is called a primitive element of 1. We shall show that (at least) all separable
extensions have primitive elements.
Consider for example Q
_
2.
_
3|,Q. We know (see Exercise 3-3) that its Galois group
over Q is a 4-group (o. t). where
_
o
_
2 =
_
2
o
_
3 =
_
3
.
_
t
_
2 =
_
2
t
_
3 =
_
3
.
Note that
o(
_
2
_
3) =
_
2
_
3.
t(
_
2
_
3) =
_
2
_
3.
(ot)(
_
2
_
3) =
_
2
_
3.
These all differ from
_
2
_
3, and so only the identity element of Gal(Q
_
2.
_
3|,Q)
xes the elements of Q
_
2
_
3|. According to the fundamental theorem, this implies that
_
2
_
3 is a primitive element:
Q
_
2.
_
3| =Q
_
2
_
3|.
It is clear that this argument should work much more generally.
Recall that an element algebraic over a eld J is separable over J if its minimum
polynomial over J has no multiple roots.
THEOREM 5.1 Let 1 =J
1
. ....
i
| be a nite extension of J, and assume that
2
. ....
i
are separable over J (but not necessarily
1
). Then there is an element ; 1 such that
1 =J;|.
PROOF. For nite elds, we proved this in (4.19). Hence we may assume J to be innite.
It sufces to prove the statement for r =2, for then
J
1
.
2
. . . . .
i
| =J
t
1
.
3
. . . . .
i
| =J
tt
1
.
4
. . . . .
i
| = .
59
60 5. APPLICATIONS OF GALOIS THEORY
Thus let 1 =J. | with separable over J. Let and g be the minimum polynomials
of and over J, and let 1 be a splitting eld for g containing 1. Let
1
=. . . . .
x
be
the roots of in 1, and let
1
=,
2
. . . . .
t
be the roots of g. For =1,
}
=, and so
the the equation

i
X
}
= X.
has exactly one solution, namely, X =

i
-
-
j
. If we choose a c J different from any of
these solutions (using that J is innite), then

i
c
}
= c unless i =1 =.
Let ; = c. I claim that
J. | =J;|.
The polynomials g(X) and (; cX) have coefcients in J;|, and have as a root:
g() =0. (; c) =() =0.
In fact, is their only common root, because we chose c so that ; c
}
=
i
unless
i =1 = . Therefore
gcd(g(X). (; cX)) =X .
Here we computed the gcd in 1X|, but this is equal to the gcd computed in J;|X|
(Proposition 2.10). Hence J;|, and this implies that =; c also lies in J;|. This
proves the claim.
2
REMARK 5.2 When J is innite, the proof shows that ; can be chosen to be of the form
; =
1
c
2

2
c
i

i
. c
i
J.
If J
1
. . . . .
i
| is Galois over J, then an element of this form will be a primitive element
provided it is moved by every nontrivial element of the Galois group. This remark makes it
very easy to write down primitive elements.
Our hypotheses are minimal: if two of the s are not separable, then the extension need
not be simple. Before giving an example to illustrate this, we need another result.
PROPOSITION 5.3 Let 1 =J;| be a simple algebraic extension of J. Then there are only
nitely many intermediate elds M,
J M 1.
PROOF. Let M be such a eld, and let g(X) be the minimum polynomial of ; over M. Let
M
t
be the subeld of 1 generated over J by the coefcients of g(X). Clearly M
t
M,
but (equally clearly) g(X) is the minimum polynomial of ; over M
t
. Hence
1: M
t
| =deg(g) =1: M|.
and so M =M
t
; we have shown that M is generated by the coefcients of g(X).
Let (X) be the minimum polynomial of ; over J. Then g(X) divides (X) in
MX|, and hence also in 1X|. Therefore, there are only nitely many possible gs, and
consequently only nitely many possible Ms.
2
Fundamental Theorem of Algebra 61
REMARK 5.4 (a) Note that the proof in fact gives a description of all the intermediate
elds: each is generated over J by the coefcients of a factor g(X) of (X) in 1X|. The
coefcients of such a g(X) are partially symmetric polynomials in the roots of (X) (that
is, xed by some, but not necessarily all, of the permutations of the roots).
(b) The proposition has a converse: if 1 is a nite extension of J and there are only
nitely many intermediate elds M, J M 1, then 1 is a simple extension of J. This
gives another proof of Theorem 5.1 in the case that 1 is separable over J, because Galois
theory shows that there are only nitely many intermediate elds in this case (even the Galois
closure of 1 over J has only nitely many intermediate elds).
EXAMPLE 5.5 The simplest nonsimple algebraic extension is k(X. Y ) k(X
]
. Y
]
), where
k is an algebraically closed eld of characteristic . Let J =k(X
]
. Y
]
). For all c k, we
have
k(X. Y ) =JX. Y | JX cY | J
with the degree of each extension equal to . If
JX cY | =JX c
t
Y |. c =c
t
.
then JX cY | would contain both X and Y , which is impossible because k(X. Y ): J| =

2
. Hence there are innitely many distinct intermediate elds.
1
Alternatively, note that the degree of k(X. Y ) over k(X
]
. Y
]
) is
2
, but if k(X. Y ),
then
]
k(X
]
. Y
]
), and so generates a eld of degree at most over k(X
]
. Y
]
).
Fundamental Theorem of Algebra
We nally prove the misnamed
2
fundamental theorem of algebra.
THEOREM 5.6 The eld C of complex numbers is algebraically closed.
PROOF. We dene C to be the splitting eld of X
2
1 over R, and we let i denote a root
of X
2
1 in C. Thus C =Ri |. We have to show (see 1.44) that every (X) RX| has a
root in C.
The two facts we need to assume about R are:
Positive real numbers have square roots.
Every polynomial of odd degree with real coefcients has a real root.
1
Zariski showed that there is even an intermediate eld M that is not isomorphic to J(X. Y ), and Piotr
Blass showed in his thesis (University of Michigan 1977), using the methods of algebraic geometry, that there is
an innite sequence of intermediate elds, no two of which are isomorphic.
2
Because it is not strictly a theorem in algebra: it is a statement about R whose construction is part of
analysis (or maybe topology). In fact, I prefer the proof based on Liouvilles theorem in complex analysis to
the more algebraic proof given in the text: if (:) is a polynomial without a root in C, then (:)
-1
will be
bounded and holomorphic on the whole complex plane, and hence (by Liouville) constant. The Fundamental
Theorem was quite a difcult theorem to prove. Gauss gave a proof in his doctoral dissertation in 1798 in which
he used some geometric arguments which he didnt justify. He gave the rst rigorous proof in 1816. The elegant
argument given here is a simplication by Emil Artin of earlier proofs (see Artin, E., Algebraische Konstruction
reeller K orper, Hamb. Abh., Bd. 5 (1926), 85-90; translation available in Artin, Emil. Exposition by Emil Artin:
a selection. AMS; LMS 2007).
62 5. APPLICATIONS OF GALOIS THEORY
Both are immediate consequences of the Intermediate Value Theorem, which says that
a continuous function on a closed interval takes every value between its maximum and
minimum values (inclusive). (Intuitively, this says that, unlike the rationals, the real line has
no holes.)
We rst show that every element of C has a square root. Write =abi , with a. b R,
and choose c. J to be real numbers such that
c
2
=
(a
_
a
2
b
2
)
2
. J
2
=
(a
_
a
2
b
2
)
2
.
Then c
2
J
2
=a and (2cJ)
2
=b
2
. If we choose the signs of c and J so that cJ has the
same sign as b, then (c Ji )
2
= and so c Ji is a square root of .
Let (X) RX|, and let 1 be a splitting eld for (X)(X
2
1) we have to show
that 1 =C. Since R has characteristic zero, the polynomial is separable, and so 1 is Galois
over R. Let G be its Galois group, and let H be a Sylow 2-subgroup of G.
Let M = 1
1
. Then M has of degree (G: H) over R, which is odd. Therefore, the
minimum polynomial over R of any M has odd degree (by the multiplicativity of
degrees, 1.20), and so has a real root. Therefore the minimum polynomial has degree 1, and
R. It follows that M =R and G =H.
We now know that Gal(1,C) is a 2-group. If it is = 1, then it has a subgroup N of
index 2 (GT 4.17). The eld 1
1
has degree 2 over C, and so it is generated by the square
root of an element of C (see 3.24), but we have seen that such square roots lie in C. Hence
1
1
=C, which is a contradiction. Thus Gal(1,C) =1 and 1 =C.
2
COROLLARY 5.7 (a) The eld C is the algebraic closure of R.
(b) The set of all algebraic numbers is an algebraic closure of Q.
PROOF. Part (a) is obvious from the denition of algebraic closure (1.43), and (b) follows
from Corollary 1.46.
2
Cyclotomic extensions
A primitive nth root of 1 in J is an element of order n in J

. Such an element can exist

only if J has characteristic 0 or characteristic not dividing n.
PROPOSITION 5.8 Let J be a eld of characteristic 0 or characteristic not dividing n.
Let 1 be the splitting eld of X
n
1.
(a) There exists a primitive nth root of 1 in 1.
(b) If is a primitive nth root of 1 in 1, then 1 =J|.
(c) The eld 1 is Galois over J; for each o Gal(1,J), there is an i (Z,nZ)

such
that o =
i
for all with
n
=1; the map o i | is an injective homomorphism
Gal(1,J) (Z,nZ)

.
PROOF. (a) The roots of X
n
1 are distinct, because its derivative nX
n-1
has only zero
as a root (here we use the condition on the characteristic), and so 1 contains n distinct nth
roots of 1. The nth roots of 1 form a nite subgroup of 1

, and so (see Exercise 3) they

form a cyclic group. Every generator has order n, and hence will be a primitive nth root of 1.
(b) The roots of X
n
1 are the powers of , and J| contains them all.
Cyclotomic extensions 63
(c) The extension 1,J is Galois because 1 is the splitting eld of a separable polynomial.
If
0
is one primitive nth root of 1, then the remaining primitive nth roots of 1 are the elements

i
0
with i relatively prime to n. Since, for any automorphism o of 1, o
0
is again a primitive
nth root of 1, it equals
i
0
for some i relatively prime to n, and the map o i mod n is
injective because
0
generates 1 over J. It obviously is a homomorphism. Moreover, for
any other nth root of 1, =
n
0
,
o =(o
0
)
n
=
in
0
=
i
.
2
The map o i |: Gal(J|,J) (Z,nZ)

need not be surjective. For example, if

J =C, then its image is {1], and if J =R, it is either {1|] or {1|. 1|]. On the other hand,
when n = is prime, we saw in (1.41) that Q|: Q| =1, and so the map is surjective.
We now prove that the map is surjective for all n when J =Q.
The polynomial X
n
1 has some obvious factors in QX|, namely, the polynomials
X
d
1 for any J[n. The quotient of X
n
1 by all these factors for J <n is called the nth
cyclotomic polynomial
n
. Thus

n
=

(X ) (product over the primitive nth roots of 1).

It has degree (n), the order of (Z,nZ)

. Since every nth root of 1 is a primitive Jth root

of 1 for exactly one J dividing n, we see that
X
n
1 =

d[n

d
(X).
For example,
1
(X) =X 1,
2
(X) =X 1,
3
(X) =X
2
X 1, and

6
(X) =
X
6
1
(X 1)(X 1)(X
2
X 1)
=X
2
X 1.
This gives an easy inductive method of computing the cyclotomic polynomials. Alternatively
type polcyclo(n,X) in PARI.
Because X
n
1 has coefcients in Z and is monic, every monic factor of it in QX| has
coefcients in Z (see 1.14). In particular, the cyclotomic polynomials lie in ZX|.
LEMMA 5.9 Let J be a eld of characteristic 0 or not dividing n, and let be a primitive
nth root of 1 in some extension eld. The following are equivalent:
(a) the nth cyclotomic polynomial
n
is irreducible;
(b) the degree J|: J| =(n);
(c) the homomorphism
Gal(J|,J) (Z,nZ)

is an isomorphism.
PROOF. Because is a root of
n
, the minimum polynomial of divides
n
. It is equal to
it if and only if J|: J| =(n), which is true if and only if the injection Gal(J|,J)
(Z,nZ)

is onto.
2
THEOREM 5.10 The nth cyclotomic polynomial
n
is irreducible in QX|.
64 5. APPLICATIONS OF GALOIS THEORY
PROOF. Let (X) be a monic irreducible factor of
n
in QX|. Its roots will be primitive
nth roots of 1, and we have to show they include all primitive nth roots of 1. For this it
sufces to show that
a root of (X) ==
i
a root of (X) for all i such that gcd(i. n) =1.
Such an i is a product of primes not dividing n, and so it sufces to show that
a root of (X) ==
]
a root of (X) for all primes not dividing n.
Write

n
(X) =(X)g(X).
Proposition 1.14 shows that (X) and g(X) lie in ZX|. Suppose is a root of but
that, for some prime not dividing n,
]
is not a root of . Then
]
is a root of g(X),
g(
]
) =0, and so is a root of g(X
]
). As (X) and g(X
]
) have a common root, they
have a nontrivial common factor in QX| (2.10), which automatically lies in ZX| (1.14).
Write h(X)

h(X) for the quotient map ZX| F
]
X|, and note that, because (X)
and g(X
]
) have a common factor of degree _ 1 in ZX|, so also do

(X) and g(X
]
) in
F
]
X|. The mod binomial theorem shows that
g(X)
]
= g(X
]
)
(recall that a
]
=a for all a F
]
), and so

(X) and g(X) have a common factor of degree
_1 in F
]
X|. Hence X
n
1, when regarded as an element of F
]
X|, has multiple roots, but
we saw in the proof of Proposition 5.8 that it doesnt. Contradiction.
2
REMARK 5.11 This proof is very old in essence it goes back to Dedekind in 1857
but its general scheme has recently become popular: take a statement in characteristic zero,
reduce modulo (where the statement may no longer be true), and exploit the existence
of the Frobenius automorphism a a
]
to obtain a proof of the original statement. For
example, commutative algebraists use this method to prove results about commutative rings,
and there are theorems about complex manifolds that were rst proved by reducing things to
characteristic .
There are some beautiful and mysterious relations between what happens in characteristic
0 and in characteristic . For example, let (X
1
. .... X
n
) ZX
1
. .... X
n
|. We can
(a) look at the solutions of =0 in C, and so get a topological space;
(b) reduce mod , and look at the solutions of

=0 in F
]
n.
The Weil conjectures (Weil 1949; proved in part by Grothendieck in the 1960s and com-
pletely by Deligne in 1973) assert that the Betti numbers of the space in (a) control the
cardinalities of the sets in (b).
THEOREM 5.12 The regular n-gon is constructible if and only if n =2
k

1

x
where the

i
are distinct Fermat primes.
PROOF. The regular n-gon is constructible if and only if cos
2t
n
(or = e
2ti{n
) is con-
structible. We know that Q| is Galois over Q, and so (according to 1.37 and 3.23) is
constructible if and only if Q|: Q| is a power of 2. But (see GT 3.5)
(n) =

][n
(1)
n(])-1
. n =

n(])
.
and this is a power of 2 if and only if n has the required form.
2
Dedekinds theorem on the independence of characters 65
REMARK 5.13 (a) As mentioned earlier, the Fermat primes are those of the form 2
2
k
1.
It is known that these numbers are prime when k =0. 1. 2. 3. 4, but it is not known whether
or not there are more Fermat primes. Thus the problem of listing the n for which the regular
n-gon is constructible is not yet solved. See the Wikipedia (Fermat number).
(b) The nal section of Gausss, Disquisitiones Arithmeticae (1801) is titled Equations
dening sections of a Circle. In it Gauss proves that the nth roots of 1 form a cyclic group,
that X
n
1 is solvable (this was before the theory of abelian groups had been developed,
and before Galois), and that the regular n-gon is constructible when n is as in the Theorem.
He also claimed to have proved the converse statement. This leads some people to credit
him with the above proof of the irreducibility of
n
, but in the absence of further evidence,
Im sticking with Dedekind.
Dedekinds theorem on the independence of characters
THEOREM 5.14 (DEDEKIND) Let J be a eld, and let G be a group. Then every nite set
{
1
. . . . .
n
] of homomorphisms G J

is linearly independent over J, i.e.,

a
i

i
=0 (as a function G J) == a
1
=0. . . . . a
n
=0.
PROOF. We use induction on m. For m=1, the statement is obvious. Assume it for m1,
and suppose that, for some set {
1
. . . . .
n
] of homomorphisms G J

and a
i
J,
a
1

1
(.) a
2

2
(.) a
n

n
(.) =0 for all . G.
We have to show that the a
i
are zero. As
1
and
2
are distinct, they will take distinct values
on some g G. On replacing . with g. in the equation, we nd that
a
1

1
(g)
1
(.) a
2

2
(g)
2
(.) a
n

n
(g)
n
(.) =0 for all . G.
On multiplying the rst equation by
1
(g) and subtracting it from the second, we obtain the
equation
a
t
2

2
a
t
n

n
=0. a
t
i
=a
i
(
i
(g)
1
(g)).
The induction hypothesis shows that a
t
i
=0 for i =2. 3. . . .. As
2
(g)
1
(g) =0, this
implies that a
2
=0, and so
a
1

1
a
3

3
a
n

n
=0.
The induction hypothesis now shows that the remaining a
i
s are also zero.
2
COROLLARY 5.15 Let J and 1 be elds, and let o
1
. .... o
n
be distinct homomorphisms
J 1. Then o
1
. .... o
n
are linearly independent over 1.
PROOF. Apply the theorem to
i
=o
i
[J

.
2
COROLLARY 5.16 Let 1 be a nite separable extension of J of degree m. Let
1
. . . . .
n
be a basis for 1 over J, and let o
1
. . . . . o
n
be distinct J-homomorphisms from 1 into a
eld . Then the matrix whose (i. )th-entry is o
i

}
is invertible.
PROOF. If not, there exist c
i
such that

n
i=1
c
i
o
i
(
}
) = 0 for all . But the map

n
i=1
c
i
o
i
: 1 is J-linear, and so this implies that

n
i=1
c
i
o
i
() =0 for all 1,
which contradicts Corollary 5.15.
2
66 5. APPLICATIONS OF GALOIS THEORY
The normal basis theorem
DEFINITION 5.17 Let 1 be a nite Galois extension of J with Galois group G. A basis for
1 as an J-vector space is called a normal basis if it consists of the conjugates of a single
element of 1.
In other words, a normal basis is one of the form
{o [ o G]
for some 1.
THEOREM 5.18 (NORMAL BASIS THEOREM) Every Galois extension has a normal basis.
The group algebra JG of a group G is the J-vector space with basis the elements of
G endowed with the multiplication extending that of G. Thus an element of JG is a sum

cG
a
c
o, a
c
J, and
_
c
a
c
o
__
c
b
c
o
_
=

c
_
c
1
c
2
=c
a
c
1
b
c
2
_
o.
Any J-linear action of G on an J-vector space V extends uniquely to an action of JG on
V .
Let 1,J be a Galois extension with Galois group G. Then 1 is an JG-module, and
Theorem 5.18 says that there exists an element 1 such that the map

c
a
c
o

c
a
c
o: JG 1
is an isomorphism of JG-modules, i.e., that 1 is a free JG-module of rank 1.
We give three proofs of Theorem 5.18. The rst assumes that J is innite and the second
that G is cyclic. Since every Galois extension of a nite eld is cyclic (4.20), this covers all
cases. The third proof applies to both nite and innite elds, but uses the Krull-Schmidt
theorem.
PROOF FOR INFINITE FIELDS
LEMMA 5.19 Let JX
1
. . . . . X
n
|, and let S be an innite subset of J. If (a
1
. . . . . a
n
) =
0 for all a
1
. . . . . a
n
S, then is the zero polynomial (i.e., =0 in JX
1
. . . . . X
n
|).
PROOF. We prove this by induction on m. For m=1, the lemma becomes the statement
that a nonzero polynomial in one symbol has only nitely many roots (see 1.7). For m> 1,
write as a polynomial in X
n
with coefcients in JX
1
. . . . . X
n-1
|, say,
=

c
i
(X
1
. . . . . X
n-1
)X
i
n
.
For any m1-tuple a
1
. . . . . a
n-1
of elements of S,
(a
1
. . . . . a
n-1
. X
n
)
is a polynomial in X
n
having every element of S as a root. Therefore, each of its coefcients
is zero: c
i
(a
1
. . . . . a
n-1
) =0 for all i . Since this holds for all (a
1
. . . . . a
n-1
), the induction
hypothesis shows that c
i
(X
1
. . . . . X
n-1
) is the zero polynomial.
2
The normal basis theorem 67
We now prove 5.18 in the case that J is innite. Number the elements of G as o
1
. . . . . o
n
(with o
1
=1).
Let JX
1
. . . . . X
n
| have the property that
(o
1
. . . . . o
n
) =0
for all 1. For a basis
1
. . . . .
n
of 1 over J, let
g(Y
1
. . . . . Y
n
) =(

n
i=1
Y
i
o
1

i
.

n
i=1
Y
i
o
2

i
. . . .) 1Y
1
. . . . . Y
n
|.
The hypothesis on implies that g(a
1
. . . . . a
n
) =0 for all a
i
J, and so g =0 (because
J is innite). But the matrix (o
i

}
) is invertible (5.16). Since g is obtained from by an
invertible linear change of variables, can be obtained from g by the inverse linear change
of variables. Therefore it also is zero.
Write X
i
= X(o
i
), and let = (X(o
i
o
}
)), i.e., is the mm matrix having X
k
in
the (i. )th place if o
i
o
}
=o
k
. Then det() is a polynomial in X
1
. . . . . X
n
, say, det() =
h(X
1
. . . . . X
n
). Clearly, h(1. 0. . . . . 0) is the determinant of a matrix having exactly one 1 in
each row and each column and its remaining entries 0. Hence the rows of the matrix are a
permutation of the rows of the identity matrix, and so its determinant is 1. In particular, h is
not identically zero, and so there exists an 1

such that h(o

1
. . . . . o
n
) (=det(o
i
o
}
))
is nonzero. We shall show that {o
i
] is a normal basis. For this, it sufces to show that o
i

are linearly independent over J. Suppose

n
}=1
a
}
o
}
=0
for some a
}
J. On applying o
1
. . . . . o
n
successively, we obtain a system of m-equations

a
}
o
i
o
}
=0
in the m unknowns a
}
. Because this system of equations is nonsingular, the a
}
s are zero.
This completes the proof of the theorem in the case that J is innite.
PROOF WHEN G IS CYCLIC.
Assume that G is generated by an element o
0
of order n. Then 1: J| =n. The minimum
polynomial of o
0
regarded as an endomorphism of the J-vector space 1 is the monic
polynomial in JX| of least degree such that 1(o
0
) =0 (as an endomorphism of 1). It
has the property that it divides every polynomial Q(X) JX| such that Q(o
0
) =0. Since
o
n
0
=1, 1(X) divides X
n
1. On the other hand, Dedekinds theorem on the independence
of characters (5.14) implies that 1. o
0
. . . . . o
n-1
0
are linearly independent over J, and so
deg1(X) > n 1. We conclude that 1(X) = X
n
1. Therefore, as an JX|-module
with X acting as o
0
, 1 is isomorphic to JX|,(X
n
1). For any generator of 1 as an
JX|-module, . o
0
. . . . . o
0

n-1
is an J-basis for 1.
UNIFORM PROOF
The Krull-Schmidt theorem says that every module M of nite length over a ring can be
written as a direct sum of indecomposable modules and that the indecomposable modules
occurring in a decomposition are unique up to order and isomorphism. Thus M =

i
m
i
M
i
where M
i
is indecomposable and m
i
M
i
denotes the direct sum of m
i
copies of M
i
; the set
of isomorphism classes of the M
i
is uniquely determined and, when we choose the M
i
to
68 5. APPLICATIONS OF GALOIS THEORY
be pairwise nonisomorphic, each m
i
is uniquely determined. From this it follows that two
modules M and M
t
of nite length over a ring are isomorphic if mM ~mM
t
for some
m_1.
Consider the J-vector space 1
T
1. We let 1 act on the rst factor, and G act on the
second factor (so a(. ,) =a. ,, a 1, and o(. ,) =. o,, o G). We shall
prove Theorem 5.18 by showing that
JG JG

n
~1
T
1 ~1 1

n
as JG-modules (n =1: J|).
For o G, let z
c
: 1
T
1 1 denote the map . , . o,. Then z
c
is obviously
1-linear, and z
c
(t:) =z
cr
(:) for all t G and : 1
T
1. I claim that {z
c
[ o G] is
an 1-basis for Hom
T-linear
(1
T
1. 1). As this space has dimension n, it sufces to show
that the set is linearly independent. But if

c
c
c
z
c
=0, c
c
1, then
0 =

c
c
c
(z
c
(1,)) =

c
c
c
o,
for all , 1, which implies that all c
c
=0 by Dedekinds theorem 5.14.
Consider the map
: 1
T
1 1G. :

c
z
c
(:) o
-1
.
Then is 1-linear. If (:) =0, then z
c
(:) =0 for all o G, and so : =0 in 1
T
1
(because the z
c
s span the dual space). Therefore is injective, and as 1
T
1 and 1G
both have dimension n over 1, it is an isomorphism. For t G,
(t:) =

c
z
c
(t:) o
-1
=

c
z
cr
(:) t(ot)
-1
=t(:).
and so is an isomorphism of 1G-modules. Thus
1
1
1 .1G ~JG JG
as an JG-module.
On the other hand, for any basis {e
1
. . . . . e
n
] for 1 as an J-vector space,
1
T
1 =(e
1
1) (e
n
1) .1 1
as JG-modules. This completes the proof.
NOTES The normal basis theorem was stated for nite elds by Eisenstein in 1850, and proved for
nite elds by Hensel in 1888. Dedekind used normal bases in number elds in his work on the
discriminant in 1880, but he had no general proof. Noether gave a proof for some innite elds
(1932) and Deuring gave a uniform proof (also 1932). The above proof simplies that of Deuring
see Blessenohl, Dieter. On the normal basis theorem. Note Mat. 27 (2007), 510. According to
the Wikipedia, normal bases are frequently used in cryptographic applications that are based on the
discrete logarithm problem such as elliptic curve cryptography.
Hilberts Theorem 90 69
Hilberts Theorem 90
Let G be a group. A G-module is an abelian group M together with an action of G, i.e., a
map GM M such that
(a) o(mm
t
) =omom
t
for all o G, m. m
t
M;
(b) (ot)(m) =o(t m) for all o. t G, m M;
(c) 1m=m for all m M.
Thus, to give an action of G on M is the same as to give a homomorphism G Aut(M)
(automorphisms of M as an abelian group).
EXAMPLE 5.20 Let 1 be a Galois extension of J with Galois group G. Then (1. ) and
(1

. ) are G-modules.
Let M be a G-module. A crossed homomorphism is a map : G M such that
(ot) =(o) o(t) for all o. t G.
Note that the condition implies that (1) =(1 1) =(1) (1), and so (1) =0.
EXAMPLE 5.21 (a) Let : G M be a crossed homomorphism. For any o G,
(o
2
) =(o) o(o).
(o
3
) =(o o
2
) =(o) o(o) o
2
(o)

(o
n
) =(o) o(o) o
n-1
(o).
Thus, if G is a cyclic group of order n generated by o, then a crossed homomorphism
: G M is determined by its value, . say, on o, and . satises the equation
. o. o
n-1
. =0. (*)
Moreover, if . M satises (*), then the formulas (o
i
) =. o. o
i-1
. dene a
crossed homomorphism : G M. Thus, for a nite group G =(o), there is a one-to-one
correspondence
{crossed homs : G M]
( -((c)
{. M satisfying (*)].
(b) For every . M, we obtain a crossed homomorphism by putting
(o) =o. .. all o G.
A crossed homomorphism of this form is called a principal crossed homomorphism.
(c) If G acts trivially on M, i.e., om = m for all o G and m M, then a crossed
homomorphism is simply a homomorphism, and there are no nonzero principal crossed
homomorphisms.
The sum and difference of two crossed homomorphisms is again a crossed homo-
morphism, and the sum and difference of two principal crossed homomorphisms is again
principal. Thus we can dene
H
1
(G. M) =
{crossed homomorphisms]
{principal crossed homomorphisms]
70 5. APPLICATIONS OF GALOIS THEORY
(quotient abelian group). The cohomology groups H
n
(G. M) have been dened for all
n N, but since this was not done until the twentieth century, it will not be discussed in this
course. An exact sequence of G-modules
0 M
t
M M
tt
0
gives rise to an exact sequence
0 M
tG
M
G
M
ttG
d
H
1
(G. M
t
) H
1
(G. M) H
1
(G. M
tt
).
Let m
tt
M
ttG
, and let m M map to m
tt
. For all o G, omm lies in the submodule
M
t
of M, and the crossed homomorphism o omm: G M
t
represents J(m
tt
). It is
an exercise for the reader to check the exactness.
EXAMPLE 5.22 Let :

X X be the universal covering space of a topological space X,
and let 1 be the group of covering transformations. Under some fairly general hypotheses,
a 1 -module M will dene a sheaf Mon X, and H
1
(X. M) .H
1
(1. M). For example,
when M = Z with the trivial action of 1 , this becomes the isomorphism H
1
(X. Z) .
H
1
(1. Z) =Hom(1. Z).
THEOREM 5.23 Let 1 be a Galois extension of J with group G; then H
1
(G. 1

) =0, i.e.,
every crossed homomorphism G 1

is principal.
PROOF. Let be a crossed homomorphism G 1

. In multiplicative notation, this

means,
(ot) =(o) o((t)). o. t G.
and we have to nd a ; 1

such that (o) =

c;
;
for all o G. Because the (t) are
nonzero, Corollary 5.15 implies that

rG
(t)t: 1 1
is not the zero map, i.e., there exists an 1 such that

def
=

rG
(t)t =0.
But then, for o G,
o =

rG
o((t)) ot()
=

rG
(o)
-1
(ot) ot()
=(o)
-1

rG
(ot)ot().
which equals (o)
-1
because, as t runs over G, so also does ot. Therefore, (o) =

c()
and we can take =;
-1
.
2
Let 1 be a Galois extension of J with Galois group G. We dene the norm of an
element 1 to be
Nm =

cG
o.
Cyclic extensions 71
For t G,
t(Nm) =

cG
to =Nm.
and so Nm J. The map
Nm: 1

is a obviously a homomorphism.
EXAMPLE 5.24 The norm map C

is [[
2
and the norm map Q
_
J|

is ab
_
J a
2
Jb
2
.
We are interested in determining the kernel of the norm map. Clearly an element of the
form

r
has norm 1, and our next result show that, for cyclic extensions, all elements with
norm 1 are of this form.
COROLLARY 5.25 (HILBERTS THEOREM 90)
3
Let 1 be a nite cyclic extension of J
with Galois group (o): if Nm
T{T
=1, then =,o for some 1.
PROOF. Let m=1: J|. The condition on is that o o
n-1
=1, and so (see 5.21a)
there is a crossed homomorphism : (o) 1

with (o) =. Theorem 5.23 now shows

that is principal, which means that there is a with (o) =,o.
2
Cyclic extensions
Let J be a eld containing a primitive nth root of 1, some n _2, and write j
n
for the group
of nth roots of 1 in J. Then j
n
is a cyclic subgroup of J

of order n with generator . In

this section, we classify the cyclic extensions of degree n of J.
Consider a eld 1 =J| generated by an element whose nth power (but no smaller
power) is in J. Then is a root of X
n
a, and the remaining roots are the elements
i
,
1 _ i _ n1. Since these are all in 1, 1 is a Galois extension of J, with Galois group
G say. For every o G, o is also a root of X
n
a, and so o =
i
for some i . Hence
o, j
n
. The map
o o,: G j
n
doesnt change when is replaced by a conjugate, and it follows that the map is a homo-
morphism:
cr

=
c(r)
r
r

. Because generates 1 over J, the map is injective. If it

is not surjective, then G maps into a subgroup j
d
of j
n
, some J[n, J < n. In this case,
(o,)
d
=1, i.e., o
d
=
d
, for all o G, and so
d
J. Thus the map is surjective.
We have proved the rst part of the following statement.
PROPOSITION 5.26 Let J be a eld containing a primitive nth root of 1. Let 1 =J|
where
n
J and no smaller power of is in J. Then 1 is a Galois extension of J with
cyclic Galois group of order n. Conversely, if 1 is a cyclic extension of J of degree n, then
1 =J| for some with
n
J.
3
This is Satz 90 in Hilberts book, Theorie der Algebraischen Zahlk orper, 1897. The theorem was discovered
by Kummer in the special case of Q
]
|,Q, and generalized to Theorem 5.23 by E. Noether. Theorem 5.23, as
well as various vast generalizations of it, are also referred to as Hilberts Theorem 90.
For an illuminating discussion of Hilberts book, see the introduction to the English translation (Springer
1998) written by F. Lemmermeyer and N. Schappacher.
72 5. APPLICATIONS OF GALOIS THEORY
PROOF. It remains to prove the last statement. Let o generate G and let generate j
n
. It
sufces to nd an element 1

such that o =
-1
, for then
n
J, and
n
is the
smallest power of that lies in J. As 1. o. . . . . o
n-1
are distinct homomorphisms J

,
Dedekinds Theorem 5.14 shows that

n-1
i=0

i
o
i
is not the zero function, and so there exists
a ; such that
def
=

i
o
i
; =0. Now o =
-1
.
2
ASIDE 5.27 (a) It is not difcult to show that the polynomial X
n
a is irreducible in JX| if a is
not a th power for any prime dividing n. When we drop the condition that J contains a primitive
nth root of 1, this is still true except that, if 4[n, we need to add the condition that a 4J
4
. See
Lang, Algebra, Springer, 2002, VI, 9, Theorem 9.1, p.297.
(b) If J has characteristic (hence has no th roots of 1 other than 1), then X
]
X a is
irreducible in JX| unless a =b
]
b for some b J, and when it is irreducible, its Galois group is
cyclic of order (generated by 1 where is a root). Moreover, every extension of J which
is cyclic of degree is the splitting eld of such a polynomial.
PROPOSITION 5.28 Let J be a eld containing a primitive nth root of 1. Two cyclic
extensions Ja
1
n
| and Jb
1
n
| of J of degree n are equal if and only if a =b
i
c
n
for some
r Z relatively prime to n and some c J

, i.e., if and only if a and b generate the same

subgroup of J

,J
n
.
PROOF. Only the only if part requires proof. We are given that J| =J| with
n
=a
and
n
=b. Let o be the generator of the Galois group with o =, and let o =
i
,
(i. n) =1. We can write
=
n-1

}=0
c
}

}
. c
}
J.
and then
o =
n-1

}=0
c
}

}
.
On comparing this with o =
i
, we nd that
i
c
}
=
}
c
}
for all . Hence c
}
=0 for
=i , and therefore =c
i

i
.
2
Kummer theory
Throughout this section, J is a eld containing a primitive nth root of 1, . In particular, J
either has characteristic 0 or characteristic not dividing n.
The last two proposition give us a complete classication of the cyclic extensions of J
of degree n. We now extend this to a classication of all abelian extensions of exponent
n. (Recall that a group G has exponent n if o
n
= 1 for all o G and n is the smallest
positive integer for which this is true. A nite abelian group of exponent n is isomorphic to
a subgroup of (Z,nZ)
i
for some r.)
Let 1,J be a nite Galois extension with Galois group G. From the exact sequence
1 j
n
1

x|-x
n
1
n
1
we obtain a cohomology sequence
1 j
n
J

x|-x
n
J

1
n
H
1
(G. j
n
) 1.
Kummer theory 73
The 1 at the right is because of Hilberts Theorem 90. Thus we obtain an isomorphism
J

1
n
,J
n
Hom(G. j
n
).
This map can be described as follows: let a be an element of J

that becomes an nth power

in 1, say a =
n
; then a maps to the homomorphism o
c

. If G is abelian of exponent
n, then
[Hom(G. j
n
)[ =(G: 1).
THEOREM 5.29 The map
1 J

1
n
denes a one-to-one correspondence between the nite abelian extensions of J of exponent
n contained in some xed algebraic closure of J and the subgroups T of J

containing
J
n
as a subgroup of nite index. The extension corresponding to T is JT
1
n
|, the
smallest subeld of containing J and an nth root of each element of T. If 1 -T, then
1: J| =(T: J
n
).
PROOF. For any nite Galois extension 1 of J, dene T(1) = J

1
n
. Then 1
JT(1)
1
n
|, and for any group T containing J
n
as a subgroup of nite index, T(JT
1
n
|)
T. Therefore,
1: J| _JT(1)
1
n
|: J| =(T(JT(1)
1
n
|): J
n
) _(T(1): J
n
).
If 1,J is abelian of exponent n, then 1: J| =(T(1): J
n
), and so equalities hold through-
out: 1 =JT(1)
1
n
|.
Next consider a group T containing J
n
as a subgroup of nite index, and let 1 =
JT
1
n
|. Then 1 is a composite of the extensions Ja
1
n
| for a running through a set of
generators for T,J
n
, and so it is a nite abelian extension of exponent n. Therefore
a
_
o
oa
1
n
a
1
n
_
: T(1),J
n
Hom(G. j
n
). G =Gal(1,J).
is an isomorphism. This map sends T,J
n
isomorphically onto the subgroup Hom(G,H. j
n
)
of Hom(G. j
n
) where H consists of the o G such that oa
1
n
,a
1
n
=1 for all a T. But
such a o xes all a
1
n
for a T, and therefore is the identity automorphism on 1 =JT
1
n
|.
This shows that T(1) =T, and hence 1 T(1) and T JT
1
n
| are inverse bijections.
2
EXAMPLE 5.30 (a) The quadratic extensions of R are (certainly) in one-to-one correspon-
dence with the subgroups of R

,R
2
={1].
(b) The nite abelian extensions of Q of exponent 2 are in one-to-one correspondence
with the nite subgroups of Q

,Q
2
, which is a direct sum of cyclic groups of order 2
indexed by the prime numbers plus o(modulo squares, every nonzero rational number has
a unique representative of the form
1

i
with the
i
prime numbers).
REMARK 5.31 Let 1 be an abelian extension of J of exponent n, and let
T(1) ={a J

[ a becomes an nth power in 1].

There is a perfect pairing
(a. o)
oa
1
n
a
1
n
:
T(1)
J
n
Gal(1,J) j
n
.
Cf. Exercise 2-1 for the case n =2.
74 5. APPLICATIONS OF GALOIS THEORY
Proof of Galoiss solvability theorem
LEMMA 5.32 Let JX| be separable, and let J
t
be an extension eld of J. Then the
Galois group of as an element of J
t
X| is a subgroup of that of as an element of JX|.
PROOF. Let 1
t
be a splitting eld for over J
t
, and let
1
. . . . .
n
be the roots of (X) in
1
t
. Then 1 =J
1
. ....
n
| is a splitting eld of over J. Every element of Gal(1
t
,J
t
)
permutes the
i
and so maps 1 into itself. The map o o[1 is an injection Gal(1
t
,J
t
)
Gal(1,J).
2
THEOREM 5.33 Let J be a eld of characteristic 0. A polynomial in JX| is solvable if
and only if its Galois group is solvable.
PROOF. ==: Let JX| have solvable Galois group G
(
. Let J
t
=J| where is a
primitive nth root of 1 for some large n for example, n =(deg ) will do. The lemma
shows that the Galois group G of as an element of J
t
X| is a subgroup of G
(
, and hence
is also solvable (GT 6.6a). This means that there is a sequence of subgroups
G =G
0
G
1
G
n-1
G
n
={1]
such that each G
i
is normal in G
i-1
and G
i-1
,G
i
is cyclic. Let 1 be a splitting eld of
(X) over J
t
, and let J
i
=1
G
i
. We have a sequence of elds
J J| =J
t
=J
0
J
1
J
2
J
n
=1
with J
i
cyclic over J
i-1
. Theorem 5.26 shows that J
i
=J
i-1

i
| with
T
i
:T
i1
j
i
J
i-1
,
each i , and this shows that is solvable.
==: It sufces to show that G
(
is a quotient of a solvable group (GT 6.6a). Hence it
sufces to nd a solvable extension

1 of J such that (X) splits in

1X|.
We are given that there exists a tower of elds
J =J
0
J
1
J
2
J
n
such that
(a) J
i
=J
i-1

i
|,
i
i
i
J
i-1
;
(b) J
n
contains a splitting eld for .
Let n =r
1
r
n
, and let be a eld Galois over J and containing (a copy of) J
n
and a
primitive nth root of 1. For example, choose a primitive element ; for J
n
over J (see 5.1),
and take to be a splitting eld of g(X)(X
n
1) where g(X) is the minimum polynomial
of ; over J.
Let G be the Galois group of ,J, and let

1 be the Galois closure of J
n
| in .
According to (3.17a),

1 is the composite of the elds oJ
n
|, o G, and so it is generated
over J by the elements
.
1
.
2
. . . . .
n
. o
1
. . . . . o
n
. o
t

1
. . . . .
We adjoin these elements to J one by one to get a sequence of elds
J J| J.
1
| J
t
J
tt


1
in which each eld J
tt
is obtained from its predecessor J
t
by adjoining an rth root of an
element of J
t
(r =r
1
. . . . . r
n
. or n). According to (5.8) and (5.26), each of these extensions
is abelian (and even cyclic after for the rst), and so

1,J is a solvable extension.
2
Symmetric polynomials 75
ASIDE 5.34 One of Galoiss major achievements was to show that an irreducible polynomial of
prime degree in QX| is solvable by radicals if and only if its splitting eld is generated by any two
roots of the polynomial.
4
This theorem of Galois answered a question on mathoverow in 2010
(mo24081). For a partial generalization of Galoiss theorem, see mo110727.
Symmetric polynomials
Let 1 be a commutative ring (with 1). A polynomial 1(X
1
. .... X
n
) 1X
1
. . . . . X
n
| is said
to be symmetric if it is unchanged when its variables are permuted, i.e., if
1(X
c(1)
. . . . . X
c(n)
) =1(X
1
. . . . . X
n
). all o S
n
.
For example

1
=

i
X
i
=X
1
X
2
X
n
.

2
=

i~}
X
i
X
}
=X
1
X
2
X
1
X
3
X
1
X
n
X
2
X
3
X
n-1
X
n
.

3
=

i~}~k
X
i
X
}
X
k
. =X
1
X
2
X
3

i
=

i
1
~~i
r
X
i
1
...X
i
r

n
= X
1
X
2
X
n
are each symmetric because
i
is the sum of all monomials of degree r made up out of dis-
tinct X
i
s. These particular polynomials are called the elementary symmetric polynomials.
THEOREM 5.35 (SYMMETRIC POLYNOMIALS THEOREM) Every symmetric polynomial
1(X
1
. .... X
n
) in 1X
1
. .... X
n
| is equal to a polynomial in the elementary symmetric poly-
nomials with coefcients in 1, i.e., 1 1
1
. ....
n
|.
PROOF. We dene an ordering on the monomials in the X
i
by requiring that
X
i
1
1
X
i
2
2
X
i
n
n
>X
}
1
1
X
}
2
2
X
}
n
n
if either
i
1
i
2
i
n
>
1

2

n
or equality holds and, for some s,
i
1
=
1
. . . . . i
x
=
x
. but i
x1
>
x1
.
For example,
X
1
X
2
X
3
3
>X
1
X
2
2
X
3
>X
1
X
2
X
2
3
.
Let 1(X
1
. . . . . X
n
) be a symmetric polynomial, and let X
i
1
1
X
i
n
n
be the highest mono-
mial occurring in 1 with a nonzero coefcient, so
1 =cX
i
1
1
X
i
n
n
lower terms, c =0.
Because 1 is symmetric, it contains all monomials obtained from X
i
1
1
X
i
n
n
by permuting
the Xs. Hence i
1
_i
2
_ _i
n
.
4
Pour quune equation de degr e premier soit r esoluble par radicaux, il faut et il suft que deux quelconques
de ces racines etant connues, les autres sen d eduisent rationnellement (

Evariste Galois, Bulletin de M. F erussac,

XIII (avril 1830), p. 271).
76 5. APPLICATIONS OF GALOIS THEORY
The highest monomial in
i
is X
1
X
i
, and it follows that the highest monomial in

d
1
1

d
n
n
is
X
d
1
d
2
d
n
1
X
d
2
d
n
2
X
d
n
n
. (3)
Therefore the highest monomial of
1(X
1
. . . . . X
n
) c
i
1
-i
2
1

i
2
-i
3
2

i
n
n
(4)
is strictly less than the highest monomial in 1(X
1
. . . . . X
n
). We can repeat this argument
with the polynomial (4), and after a nite number of steps, we will arrive at a representation
of 1 as a polynomial in
1
. . . . .
n
.
2
REMARK 5.36 (a) The proof is algorithmic. Consider, for example,
5
1(X
1
. X
2
) =(X
1
7X
1
X
2
X
2
)
2
=X
2
1
2X
1
X
2
14X
2
1
X
2
X
2
2
14X
1
X
2
2
49X
2
1
X
2
2
.
The highest monomial is 49X
2
1
X
2
2
, and so we subtract 49
2
2
, getting
1 49
2
2
=X
2
1
2X
1
X
2
14X
2
1
X
2
X
2
2
14X
1
X
2
2
.
Continuing, we get
1 49
2
2
14
1

2
=X
2
1
2X
1
X
2
X
2
2
and nally,
1 49
2
2
14
1

2
1
=0.
(b) The expression of 1 as a polynomial in the
i
in (5.35) is unique. Otherwise, by
subtracting, we would get a nontrivial polynomial Q(
1
. . . . .
n
) in the
i
which is zero
when expressed as a polynomial in the X
i
. But the highest monomials (3) in the polynomials

d
1
1

d
n
n
are distinct (the map (J
1
. . . . . J
n
) (J
1
J
n
. . . . . J
n
) is injective), and so
they cant cancel.
Let
(X) =X
n
a
1
X
n-1
a
n
1X|.
and suppose that splits over some ring S containing 1:
(X) =

n
i=1
(X
i
).
i
S.
Then
a
1
=
1
(
1
. . . . .
n
). a
2
=
2
(
1
. . . . .
n
). . . . . a
n
=(1)
n

n
(
1
. . . . .
n
).
Thus the elementary symmetric polynomials in the roots of (X) lie in 1, and so the theorem
implies that every symmetric polynomial in the roots of (X) lies in 1. For example, the
discriminant
D( ) =

i~}
(
i

}
)
2
of lies in 1.
5
From the Wikipedia.
The general polynomial of degree n 77
THEOREM 5.37 (SYMMETRIC FUNCTIONS THEOREM) Let J be a eld. When S
n
acts on
J(X
1
. .... X
n
) by permuting the X
i
, the eld of invariants is J(
1
. ....
n
).
PROOF. Let J(X
1
. . . . . X
n
) be symmetric (i.e., xed by S
n
). Set = g,h, g. h
JX
1
. . . . . X
n
|. The polynomials H =

cS
n
oh and H are symmetric, and therefore lie
in J
1
. . . . .
n
| by (5.35). Hence their quotient =H,H lies in J(
1
. . . . .
n
).
2
COROLLARY 5.38 The eld J(X
1
. .... X
n
) is Galois over J(
1
. ....
n
) with Galois group
S
n
(acting by permuting the X
i
).
PROOF. We have shown that J(
1
. . . . .
n
) =J(X
1
. . . . . X
n
)
S
n
, and so this follows from
(3.10).
2
The eld J(X
1
. . . . . X
n
) is the splitting eld over J(
1
. . . . .
n
) of
g(T ) =(T X
1
) (T X
n
) =X
n

1
X
n-1
(1)
n

n
.
Therefore, the Galois group of g(T ) J(
1
. . . . .
n
)T | is S
n
.
ASIDE 5.39 Symmetric polynomials played an important role in the work of Galois. In his M emoire
sur les conditions de r esolubilit e des equations par radicaux, he prove the following proposition:
Let be a polynomial with coefcients o
1
. . . . . o
n
. Let .
1
. . . . . .
n
be its roots, and let
U. V. . . . be certain numbers that are rational functions in the .
i
. Then there exists a
group G of permutations of the .
i
such that the rational functions in the .
i
that are
xed under all permutations in G are exactly those that are rationally expressible in
terms of o
1
. . . . . o
n
and U. V. . . .
When we take U. V. . . . to be the elements of a eld 1 intermediate between the eld of coefcients
of and the splitting eld of , this says that the exists a group G of permutations of the .
i
whose
xed eld (when G acts on the splitting eld) is exactly 1.
The general polynomial of degree n
When we say that the roots of
aX
2
bX c
are
b
_
b
2
4ac
2a
we are thinking of a. b. c as symbols: for any particular values of a. b. c, the formula gives
the roots of the particular equation. We shall prove in this section that there is no similar
formula for the roots of the general polynomial of degree _5.
We dene the general polynomial of degree n to be
(X) =X
n
t
1
X
n-1
(1)
n
t
n
Jt
1
. .... t
n
|X|
where the t
i
are symbols. We shall show that, when we regard as a polynomial in X with
coefcients in the eld J(t
1
. . . . . t
n
), its Galois group is S
n
. Then Theorem 5.33 proves the
above remark (at least in characteristic zero).
THEOREM 5.40 The Galois group of the general polynomial of degree n is S
n
.
78 5. APPLICATIONS OF GALOIS THEORY
PROOF. Let (X) be the general polynomial of degree n,
(X) =X
n
t
1
X
n-1
(1)
n
t
n
Jt
1
. .... t
n
|X|.
If we can show that the map
t
i

i
: Jt
1
. . . . . t
n
| J
1
. . . . .
n
|
is injective (i.e., the
i
are algebraically independent over J, see p. 109), then it will extend
to an isomorphism
J(t
1
. . . . . t
n
) J(
1
. . . . .
n
)
sending (X) to
g(X) =X
n

1
X
n-1
(1)
n

n
J(
1
. . . . .
n
)X|.
Therefore the statement will follow from Corollary 5.38.
We now prove that the
i
are algebraically independent
6
. Suppose on the contrary
that there exists a 1(t
1
. . . . . t
n
) such that 1(
1
. . . . .
n
) = 0. Equation (3), p. 76, shows
that if m
1
(t
1
. . . . . t
n
) and m
2
(t
1
. . . . . t
n
) are distinct monomials, then m
1
(
1
. . . . .
n
) and
m
2
(
1
. . . . .
n
) have distinct highest monomials. Therefore, cancellation cant occur, and so
1(t
1
. . . . . t
n
) must be the zero polynomial.
2
REMARK 5.41 Since S
n
occurs as a Galois group over Q, and every nite group occurs
as a subgroup of some S
n
, it follows that every nite group occurs as a Galois group over
some nite extension of Q, but does every nite Galois group occur as a Galois group over
Q itself? This is known as the inverse Galois problem.
The Hilbert-Noether program for proving this was the following. Hilbert proved that
if G occurs as the Galois group of an extension 1 Q(t
1
. .... t
n
) (the t
i
are symbols), then
it occurs innitely often as a Galois group over Q. For the proof, realize 1 as the splitting
eld of a polynomial (X) kt
1
. . . . . t
n
|X| and prove that for innitely many values of
the t
i
, the polynomial you obtain in QX| has Galois group G. (This is quite a difcult
theorem see Serre, J.-P., Lectures on the Mordell-Weil Theorem, 1989, Chapter 9.) Noether
conjectured the following: Let G S
n
act on J(X
1
. .... X
n
) by permuting the X
i
; then
J(X
1
. . . . . X
n
)
G
~J(t
1
. .... t
n
) (for symbols t
i
). However, Swan proved in 1969 that the
conjecture is false for G the cyclic group of order 47. Hence this approach can not lead to
a proof that all nite groups occur as Galois groups over Q, but it doesnt exclude other
approaches. For more information on the problem, see Serre, ibid., Chapter 10; Serre, J.-P.,
Topics in Galois Theory, 1992; and the Wikipedia (Inverse Galois problem).
REMARK 5.42 Take J =C, and consider the subset of C
n1
dened by the equation
X
n
T
1
X
n-1
(1)
n
T
n
=0.
It is a beautiful complex manifold S of dimension n. Consider the projection
: S C
n
. (.. t
1
. . . . . t
n
) (t
1
. . . . . t
n
).
Its bre over a point (a
1
. . . . . a
n
) is the set of roots of the polynomial
X
n
a
1
X
n-1
(1)
n
a
n
.
The discriminant D( ) of (X) = X
n
T
1
X
n-1
(1)
n
T
n
is a polynomial in
CT
1
. . . . . T
n
|. Let z be the zero set of D( ) in C
n
. Then over each point of C
n
z,
there are exactly n points of S, and S
-1
(z) is a covering space over C
n
z.
6
This can also be proved by noting that, because J(X
1
. . . . . X
n
) is algebraic over J(
1
. . . . .
n
), the latter
must have transcendence degree n (see 8).
Norms and traces 79
A BRIEF HISTORY
As far back as 1500 BC, the Babylonians (at least) knew a general formula for the roots
of a quadratic polynomial. Cardan (about 1515 AD) found a general formula for the roots
of a cubic polynomial. Ferrari (about 1545 AD) found a general formula for the roots of
a quartic polynomial (he introduced the resolvent cubic, and used Cardans result). Over
the next 275 years there were many fruitless attempts to obtain similar formulas for higher
degree polynomials, until, in about 1820, Rufni and Abel proved that there are none.
Norms and traces
Recall that, for an nn matrix =(a
i}
)
Tr() =

i
a
i i
(trace of )
det() =

cS
n
sign(o)a
1c(1)
a
nc(n)
. (determinant of )
c

(X) = det(X1
n
) (characteristic polynomial of ).
Moreover,
c

(X) =X
n
Tr()X
n-1
(1)
n
det().
None of these is changed when is replaced by its conjugate UU
-1
by an invertible
matrix U. Therefore, for any endomorphism of a nite dimensional vector space V , we
can dene
7
Tr() =Tr(), det() =det(), c

(X) =c

(X)
where is the matrix of with respect to any basis of V . If is a second endomorphism of
V ,
Tr( ) =Tr() Tr():
det() =det()det().
Now let 1 be a nite eld extension of J of degree n. An element of 1 denes an
J-linear map

1
: 1 1. . ..
and we dene
Tr
T{T
() = Tr(
1
) (trace of )
Nm
T{T
() = det(
1
) (norm of )
c
,T{T
(X) = c

L
(X) (characteristic polynomial of ).
Thus, Tr
T{T
is a homomorphism (1. ) (J. ), and Nm
T{T
is a homomorphism
(1

. ) (J

. ).
7
The coefcients of the characteristic polynomial
c

(X) =X
n
c
1
X
n-1
c
n
.
of have the following description
c
i
=(1)
i
Tr([

i
V )
see Bourbaki, N., Algebra, Chapter 3, 8.11.
80 5. APPLICATIONS OF GALOIS THEORY
EXAMPLE 5.43 (a) Consider the eld extension C R. For =abi , the matrix of
1
with respect to the basis {1. i ] is
_
o -b
b o
_
, and so
Tr
C{R
() =2m(), Nm
C{R
() =[[
2
.
(b) For a J, a
1
is multiplication by the scalar a. Therefore
Tr
T{T
(a) =na, Nm
T{T
(a) =a
n
, c
o,T{T
(X) =(X a)
n
where n =1: J|.
Let 1 =Q. i | be the splitting eld of X
S
2. Then 1 has degree 16 over Q, and so to
compute the trace and norm an element of 1, the denition requires us to compute the trace
and norm of a 1616 matrix. The next proposition gives us a quicker method.
PROPOSITION 5.44 Let 1,J be a nite extension of elds, and let (X) be the minimum
polynomial of 1. Then
c
,T{T
(X) =(X)
T:Tjj
.
PROOF. Suppose rst that 1 =J|. In this case, we have to show that c

(X) =(X).
Note that
1
is an injective homomorphism from 1 into the ring of endomorphisms
of 1 as a vector space over J. The Cayley-Hamilton theorem shows that c

(
1
) =0, and
therefore c

() =0. Hence [c

, but they are monic of the same degree, and so they are
equal.
For the general case, let
1
. ....
n
be a basis for J| over J, and let ;
1
. .... ;
n
be a basis
for 1 over J|. As we saw in the proof of (1.20), {
i
;
k
] is a basis for 1 over J. Write

i
=

a
}i

}
. Then, according to the rst case proved, =
def
(a
i}
) has characteristic
polynomial (X). But
i
;
k
=

a
}i

}
;
k
, and so the matrix of
1
with respect to {
i
;
k
]
breaks up into nn blocks with s down the diagonal and zero matrices elsewhere, from
which it follows that c

L
(X) =c

(X)
n
=(X)
n
.
2
COROLLARY 5.45 Suppose that the roots of the minimum polynomial of are
1
. . . . .
n
(in some splitting eld containing 1), and that 1: J|| =m. Then
Tr() =m

n
i=1

i
. Nm
T{T
=
_
n
i=1

i
_
n
.
PROOF. Write the minimum polynomial of as
(X) =X
n
a
1
X
n-1
a
n
=

(X
i
).
so that
a
1
=

i
, and
a
n
=(1)
n

i
.
Then
c

(X) =((X))
n
=X
nn
ma
1
X
nn-1
a
n
n
.
so that
Tr
T{T
() =ma
1
=m

i
, and
Nm
T{T
() =(1)
nn
a
n
n
=(

i
)
n
.
2
Norms and traces 81
EXAMPLE 5.46 (a) Consider the extension C R. If C\R, then
c

(X) =(X) =X
2
2m()X [[
2
.
If R, then c

(X) =(X a)
2
.
(b) Let 1 be the splitting eld of X
S
2. Then 1 has degree 16 over Q and is generated
by =
8
_
2 and i =
_
1 (see Exercise 16). The minimum polynomial of is X
S
2, and
so
c
,Qj{Q
(X) = X
S
2. c
,T{Q
(X) = (X
S
2)
2
Tr
Qj{Q
= 0. Tr
T{Q
= 0
Nm
Qj{Q
= 2. Nm
T{Q
= 4
REMARK 5.47 Let 1 be a separable extension of J, and let be the set of J-homomorphisms
of 1 into an algebraic closure of J. Then
Tr
T{T
=

c
o
Nm
T{T
=

c
o.
When 1 = J|, this follows from 5.45 and the observation (cf. 2.1b) that the o are
the roots of the minimum polynomial (X) of over J. In the general case, the o are
still roots of (X) in , but now each root of (X) occurs 1: J|| times (because each
J-homomorphism J| has 1: J|| extensions to 1). For example, if 1 is Galois
over J with Galois group G, then
Tr
T{T
=

cG
o
Nm
T{T
=

cG
o.
PROPOSITION 5.48 For nite extensions 1 M J, we have
Tr
{T
Tr
T{
=Tr
T{T
.
Nm
{T
Nm
T{
=Nm
T{T
.
PROOF. If 1 is separable over J, then this can be proved fairly easily using the descriptions
in the above remark. We omit the proof in the general case.
2
PROPOSITION 5.49 Let (X) be a monic irreducible polynomial with coefcients in J,
and let be a root of in some splitting eld of . Then
disc(X) =(1)
n(n-1){2
Nm
Tj{T

t
()
where
t
is the formal derivative
d(
dA
of .
PROOF. Let (X) =

n
i=1
(X
i
) be the factorization of in the given splitting eld,
and number the roots so that =
1
. Compute that
disc(X)
def
=

i~}
(
i

}
)
2
=(1)
n(n-1){2

i
(

}yi
(
i

}
))
=(1)
n(n-1){2

t
(
i
)
=(1)
n(n-1){2
Nm
Tj{T
(
t
()) (by 5.47).
2
82 5. APPLICATIONS OF GALOIS THEORY
EXAMPLE 5.50 We compute the discriminant of
(X) =X
n
aX b. a. b J.
assumed to be irreducible and separable, by computing the norm of
;
def
=
t
() =n
n-1
a. () =0.
On multiplying the equation

n
a b =0
by n
-1
and rearranging, we obtain the equation
n
n-1
=nanb
-1
.
Hence
; =n
n-1
a =(n1)anb
-1
.
Solving for gives
=
nb
; (n1)a
.
From the last two equations, it is clear that J| =J;|, and so the minimum polynomial
of ; over J has degree n also. If we write

_
nb
X (n1)a
_
=
1(X)
Q(X)
1(X) =(X (n1)a)
n
na(X (n1)a)
n-1
(1)
n
n
n
b
n-1
Q(X) =(X (n1)a)
n
,b.
then
1(;) =() Q(;) =0.
As
Q(;) =
(; (n1)a)
n
b
=
(nb)
n

n
b
=0
and 1(X) is monic of degree n, it must be the minimum polynomial of ;. Therefore Nm;
is (1)
n
times the constant term of 1(X), namely,
Nm; =n
n
b
n-1
(1)
n-1
(n1)
n-1
a
n
.
Therefore,
disc(X
n
aX b) =(1)
n(n-1){2
(n
n
b
n-1
(1)
n-1
(n1)
n-1
a
n
).
which is something PARI doesnt know (because it doesnt understand symbols as exponents).
For example,
disc(X
5
aX b) =5
5
b
4
4
4
a
5
.
Exercises 83
Exercises
5-1 For a Q, let G
o
be the Galois group of X
4
X
3
X
2
X a. Find integers
a
1
. a
2
. a
3
. a
4
such that i = == G
o
i
is not isomorphic to G
o
j
.
5-2 Prove that the rational solutions a. b Q of Pythagorass equation a
2
b
2
=1 are of
the form
a =
s
2
t
2
s
2
t
2
. b =
2st
s
2
t
2
. s. t Q.
and deduce that every right triangle with integer sides has sides of length
J(m
2
n
2
. 2mn. m
2
n
2
)
for some integers J, m, and n (Hint: Apply Hilberts Theorem 90 to the extension Qi |,Q.)
5-3 Prove that a nite extension of Q can contain only nitely many roots of 1.
CHAPTER 6
Algebraic Closures
In this chapter, we use Zorns lemma to show that every eld J has an algebraic closure .
Recall that if J is a subeld C, then the algebraic closure of J in C is an algebraic closure
of J (1.46). If J is countable, then the existence of can be proved as in the nite eld
case (4.24), namely, the set of monic irreducible polynomials in JX| is countable, and so
we can list them
1
.
2
. . . .; dene 1
i
inductively by, 1
0
=J, 1
i
= a splitting eld of
i
over 1
i-1
; then =
_
1
i
is an algebraic closure of J.
The difculty in showing the existence of an algebraic closure of an arbitrary eld J is
in the set theory. Roughly speaking, we would like to take a union of a family of splitting
elds indexed by the monic irreducible polynomials in JX|, but we need to nd a way
of doing this that is allowed by the axioms of set theory. After reviewing the statement of
Zorns lemma, we sketch three solutions
1
to the problem.
Zorns lemma
DEFINITION 6.1 (a) A relation _ on a set S is a partial ordering if it reexive, transitive,
and anti-symmetric (a _b and b _a == a =b).
(b) A partial ordering is a total ordering if, for all s. t T , either s _t or t _s.
(c) An upper bound for a subset T of a partially ordered set (S. _) is an element s S
such that t _s for all t T .
(d) A maximal element of a partially ordered set S is an element s such that s _s
t
==
s =s
t
.
A partially ordered set need not have any maximal elements, for example, the set of nite
subsets of an innite set is partially ordered by inclusion, but it has no maximal elements.
LEMMA 6.2 (ZORN) Let (S. _) be a nonempty partially ordered set for which every totally
ordered subset has an upper bound in S. Then S has a maximal element.
Zorns lemma
2
is equivalent to the Axiom of Choice, and hence independent of the
axioms of set theory.
1
There do exist naturally occurring uncountable elds not contained in C. For example, the eld of formal
Laurent series J((T )) over a eld J is uncountable even if J is nite.
2
The following is quoted from A.J. Berrick and M.E. Keating, An Introduction to Rings and Modules,
2000: The name of the statement, although widely used (allegedly rst by Lefschetz), has attracted the attention
of historians (Campbell 1978). As a maximum principle, it was rst brought to prominence, and used for
algebraic purposes in Zorn 1935, apparently in ignorance of its previous usage in topology, most notably in
85
86 6. ALGEBRAIC CLOSURES
REMARK 6.3 The set S of nite subsets of an innite set doesnt contradict Zorns lemma,
because it contains totally ordered subsets with no upper bound in S.
The following proposition is a typical application of Zorns lemma we shall use a * to
signal results that depend on Zorns lemma (equivalently, the Axiom of Choice).
PROPOSITION 6.4 (*) Every nonzero commutative ring has a maximal ideal (meaning,
maximal among proper ideals).
PROOF. Let S be the set of all proper ideals in , partially ordered by inclusion. If T is a
totally ordered set of ideals, then J =
_
JT
1 is again an ideal, and it is proper because if
1 J then 1 1 for some 1 in T , and 1 would not be proper. Thus J is an upper bound for
T . Now Zorns lemma implies that S has a maximal element, which is a maximal ideal in
.
2
First proof of the existence of algebraic closures
(Bourbaki, Alg` ebre, Chap. V, 4.) An J-algebra is a ring containing J as a subring. Let
(
i
)
iJ
be a family of commutative J-algebras, and dene

T

i
to be the quotient of the
J-vector space with basis

iJ

i
by the subspace generated by elements of the form:
(.
i
) (,
i
) (:
i
) with .
}
,
}
=:
}
for one 1 and .
i
=,
i
=:
i
for all i = ;
(.
i
) a(,
i
) with .
}
=a,
}
for one 1 and .
i
=,
i
for all i = ,
(ibid., Chap. II, 3.9). It can be made into a commutative J-algebra in an obvious fashion,
and there are canonical homomorphisms
i

T

i
of J-algebras.
For each polynomial JX|, choose a splitting eld 1
(
, and let =(

T
1
(
),M
where M is a maximal ideal in

T
1
(
(whose existence is ensured by Zorns lemma).
Note that J

T
1
(
and MJ =0. As has no ideals other than (0) and , and it is
a eld (see 1.2). The composite of the J-homomorphisms 1
(

T
1
(
, being a
homomorphism of elds, is injective. Since splits in 1
(
, it must also split in the larger
eld . The algebraic closure of J in is therefore an algebraic closure of J (by 1.44).
ASIDE 6.5 In fact, it sufces to take =(

T
1
(
),M where runs over the monic irreducible
polynomials in JX| and 1
(
is the stem eld JX|,( ) of (apply the statement in 6.7 below).
Second proof of the existence of algebraic closures
(Jacobson 1964, p144.) After (4.24) we may assume J to be innite. This implies that the
cardinality of every eld algebraic over J is the same as that of J (ibid. p143). Choose
an uncountable set S of cardinality greater than that of J, and identify J with a subset
of S. Let S be the set triples (1. . ) with 1 S and (. ) a eld structure on 1 such
that (1. . ) contains J as a subeld and is algebraic over it. Write (1. . ) _(1
t
.
t
.
t
)
if the rst is a subeld of the second. Apply Zorns lemma to show that S has maximal
elements, and then show that a maximal element is algebraically closed. (See ibid. p144 for
the details.)
Kuratowski 1922. Zorn attributed to Artin the realization that the lemma is in fact equivalent to the Axiom of
Choice (see Jech 1973). Zorns contribution was to observe that it is more suited to algebraic applications like
ours.
Third proof of the existence of algebraic closures 87
Third proof of the existence of algebraic closures
(Emil Artin.) Consider the polynomial ring J. . . . .
(
. . . .| in a family of symbols .
(
indexed
by the nonconstant monic polynomials JX|. If 1 lies in the ideal 1 of J. . . . .
(
. . . .|
generated by the polynomials (.
(
), then
g
1

1
(.
(
1
) g
n

n
(.
(
n
) =1 (in J. . . . .
(
. . . .|)
for some g
i
J. . . . .
(
. . . .| and some nonconstant monic
i
JX|. Let 1 be an extension
of J such that each
i
, i = 1. . . . . n, has a root
i
in 1. Under the J-homomorphism
J. . . . .
(
. . . .| 1 sending
_
.
(
i

i
.
(
0. {
1
. . . . .
n
]
the above relation becomes 0 =1. From this contradiction, we deduce that 1 does not lie in
1, and so Proposition 6.4 applied to J. . . . .
(
. . . .|,1 shows that 1 is contained in a maximal
ideal M of J. . . . .
(
. . . .|. Let =J. . . . .
(
. . . .|,M. Then is a eld containing (a copy
of) J in which every nonconstant polynomial in JX| has at least one root. Repeat the
process starting with 1
1
instead of J to obtain a eld 1
2
. Continue in this fashion to obtain
a sequence of elds
J =1
0
1
1
1
2
.
and let 1 =
_
i
1
i
. Then 1 is algebraically closed because the coefcients of any noncon-
stant polynomial g in 1X| lie in 1
i
for some i , and so g has a root in 1
i1
. Therefore, the
algebraic closure of J in 1 is an algebraic closure of J (1.46).
ASIDE 6.6 In fact, 1 is algebraic over J. To see this, note that 1
1
is generated by algebraic elements
over J, and so is algebraic over J (apply 1.45). Similarly, 1
2
is algebraic over 1
1
, and hence over J
(apply 1.31b). Continuing in this fashion, we nd that every element of every 1
i
is algebraic over J.
ASIDE 6.7 In fact, 1
1
is already algebraically closed (hence the algebraic closure of J). This
follows from the statement:
Let be a eld. If is algebraic over a subeld J and every nonconstant polynomial
in JX| has a root in , then is algebraically closed.
In order to prove this, it sufces to show that every irreducible polynomial in JX| splits in X|
(see 1.44). Suppose rst that is separable, and let 1 be a splitting eld for . According to
Theorem 5.1, 1 =J;| for some ; 1. Let g(X) be the minimum polynomial of ; over J. Then
g(X) has coefcients in J, and so it has a root in . Both of J;| and J| are stem elds for g,
and so there is an J-isomorphism J;| J| . As splits over J;|, it must split over .
This completes the proof when J is perfect. Otherwise, J has characteristic =0, and we
let J
t
be the set of elements . of such that .
]
m
J for some m. It is easy to see that J
t
is a
eld, and we shall complete the proof of the lemma by showing that (a) J
t
is perfect, and (b) every
polynomial in J
t
X| has a root in .
PROOF OF (a). Let a J
t
, so that b
def
= a
]
m
J for some m. The polynomial X
]
mC1
b has
coefcients in J, and so it has a root , which automatically lies in J
t
. Now
]
mC1
=a
]
m
,
which implies that
]
=a, because the th power map is injective on elds of characteristic .
Before continuing, we note that, because is algebraic over a perfect eld J
t
, it is itself perfect:
let a , and let g be the minimum polynomial of a over J
t
; if X
]
a is irreducible in X|, then
g(X
]
) is irreducible in J
t
X|, but it is not separable, which is a contradiction.
88 6. ALGEBRAIC CLOSURES
PROOF OF (b). Let (X) J
t
X|, say, (X) =

i
a
i
X
i
, a
i
J
t
. For some m, the polynomial

i
a
]
m
i
X
i
has coefcients in J, and therefore has a root . As is perfect, we can write
=
]
m
with . Now
(())
]
m
=
_

i
a
i

i
_
]
m
=

i
a
]
m
i

i
=0.
and so is a root of .
(Non)uniqueness of algebraic closures
THEOREM 6.8 (*) Let be an algebraic closure of J, and let 1 be an algebraic extension
of J. There exists an J-homomorphism 1 , and, if 1 is also an algebraic closure of
J, then every such homomorphism is an isomorphism.
PROOF. Suppose rst that 1 is countably generated over J, i.e., 1 = J
1
. ....
n
. . . .|.
Then we can extend the inclusion map J to J
1
| (map
1
to any root of its minimal
polynomial in ), then to J
1
.
2
|. and so on (see 2.2).
In the uncountable case, we use Zorns lemma. Let S be the set of pairs (M.

) with
M a eld J M 1 and

an J-homomorphism M . Write (M.

) _(N.
1
)
if M N and
1
[M =

. This makes S into a partially ordered set. Let T be a

totally ordered subset of S. Then M
t
=
_
T
M is a subeld of 1, and we can dene a
homomorphism
t
: M
t
by requiring that
t
(.) =

(.) if . M. The pair (M

t
.
t
)
is an upper bound for T in S. Hence Zorns lemma gives us a maximal element (M. ) in
S. Suppose that M =1. Then there exists an element 1, M. Since is algebraic
over M, we can apply (2.2) to extend to M|, contradicting the maximality of M. Hence
M =1, and the proof of the rst statement is complete.
If 1 is algebraically closed, then every polynomial JX| splits in 1X| and hence
in (1)X|. Let , and let (X) be the minimum polynomial of . Then X is
a factor of (X) in X|, but, as we just observed, (X) splits in (1)X|. Because of
unique factorization, this implies that (1).
2
The above proof is a typical application of Zorns lemma: once we know how to do
something in a nite (or countable) situation, Zorns lemma allows us to do it in general.
REMARK 6.9 Even for a nite eld J, there will exist uncountably many isomorphisms
from one algebraic closure to a second, none of which is to be preferred over any other. Thus
it is (uncountably) sloppy to say that the algebraic closure of J is unique. All one can say is
that, given two algebraic closures ,
t
of J, then, thanks to Zorns lemma, there exists an
J-isomorphism
t
.
Separable closures
Let be a eld containing J, and let E be a set of intermediate elds J 1 with the
following property:
(*) for all 1
1
. 1
2
E, there exists an 1 E such that 1
1
. 1
2
1.
Then 1(E) =
_
TE
1 is a subeld of (and we call
_
TE
1 a directed union), because
(*) implies that every nite set of elements of 1(E) is contained in a common 1 E, and
therefore their product, sum, etc., also lie in 1(E).
Separable closures 89
We apply this remark to the set of subelds 1 of that are nite and separable over J.
As the composite of any two such subelds is again nite and separable over J (cf. 3.14),
we see that the union 1 of all such 1 is a subeld of . We call 1 the separable closure of
J in clearly, it is separable over J and every element of separable over J lies in 1.
Moreover, because a separable extension of a separable extension is separable, is purely
inseparable over 1.
DEFINITION 6.10 (a) A eld is said to be separably closed if every nonconstant separa-
ble polynomial in X| splits in .
(b) A eld is said to be a separable closure of a subeld J if it is separable and
algebraic over J and it is separably closed.
THEOREM 6.11 (*) (a) Every eld has a separable closure.
(b) Let 1 be a separable algebraic extension of J, and let be a separable algebraic
closure of J. There exists an J-homomorphism 1 , and, if 1 is also a separable
closure of J, then every such homomorphism is an isomorphism.
PROOF. Replace polynomial with separable polynomial in the proofs of the correspond-
ing theorems for algebraic closures. Alternatively, dene to be the separable closure of J
in an algebraic closure, and apply the preceding theorems.
2
ASIDE 6.12 It is not necessary to assume the full axiom of choice to prove the existence of algebraic
closures and their uniqueness up to isomorphism, but only a weaker axiom. See Banaschewski,
Bernhard. Algebraic closure without choice. Z. Math. Logik Grundlag. Math. 38 (1992), no. 4,
383385.
CHAPTER 7
Innite Galois Extensions
In this chapter, we make free use of the axiom of choice.
1
Topological groups
DEFINITION 7.1 A set G together with a group structure and a topology is a topological
group if the maps
(g. h) gh: GG G.
g g
-1
: G G
are both continuous.
Let a be an element of a topological group G. Then a
1
: G
|-o
G is continuous
because it is the composite of
G
|-(o,)
GG
(,h)|-h
G.
In fact, it is a homeomorphism with inverse (a
-1
)
1
. Similarly a
T
: g ga and g g
-1
are
both homeomorphisms. In particular, for any subgroup H of G, the coset aH of H is open
or closed if H is open or closed. As the complement of H in G is a union of such cosets,
this shows that H is closed if it is open, and it is open if it is closed and of nite index.
Recall that a neighbourhood base for a point . of a topological space X is a set of
neighbourhoods N such that every open subset U of X containing . contains an N from N.
PROPOSITION 7.2 Let G be a topological group, and let N be a neighbourhood base for
the identity element e of G. Then
2
(a) for all N
1
. N
2
N, there exists an N
t
N such that e N
t
N
1
N
2
;
(b) for all N N, there exists an N
t
N such that N
t
N
t
N;
(c) for all N N, there exists an N
t
N such that N
t
N
-1
;
(d) for all N N and all g G, there exists an N
t
N such that N
t
gNg
-1
:
1
It is necessary to assume the axiom of choice in order to have a sensible Galois theory of innite extensions.
For example, it is consistent with Zermelo-Fraenkel set theory that there exist an algebraic closure 1 of the
Q with no nontrivial automorphisms. See: Hodges, Wilfrid, L auchlis algebraic closure of Q. Math. Proc.
Cambridge Philos. Soc. 79 (1976), no. 2, 289297.
2
For subsets S and S
t
of G, we let SS
t
={ss
t
[ s S, s
t
S
t
] and S
-1
={s
-1
[ s S].
91
92 7. INFINITE GALOIS EXTENSIONS
(e) for all g G, {gN [ N N] is a neighbourhood base for g.
Conversely, if G is a group and N is a nonempty set of subsets of G satisfying (a,b,c,d), then
there is a (unique) topology on G for which (e) holds.
PROOF. If N is a neighbourhood base at e in a topological group G, then (b), (c), and
(d) are consequences of the continuity of (g. h) gh, g g
-1
, and h ghg
-1
respec-
tively. Moreover, (a) is a consequence of the denitions and (e) of the fact that g
1
is a
homeomorphism.
Conversely, let N be a nonempty collection of subsets of a group G satisfying the
conditions (a)(d). Note that (a) implies that e lies in all the N in N. Dene U to be the
collection of subsets U of G such that, for every g U, there exists an N N with gN U.
Clearly, the empty set and G are in U, and unions of sets in U are in U. Let U
1
. U
2
U, and
let g U
1
U
2
; by denition there exist N
1
. N
2
N with gN
1
. gN
2
U; on applying (a)
we obtain an N
t
N such that gN
t
U
1
U
2
, which shows that U
1
U
2
U. It follows
that the elements of U are the open sets of a topology on G. In fact, one sees easily that it is
the unique topology for which (e) holds.
We next use (b) and (d) to show that (g. g
t
) gg
t
is continuous. Note that the sets
g
1
N
1
g
2
N
2
form a neighbourhood base for (g
1
. g
2
) in G G. Therefore, given an
open U G and a pair (g
1
. g
2
) such that g
1
g
2
U, we have to nd N
1
. N
2
N such
that g
1
N
1
g
2
N
2
U. As U is open, there exists an N N such that g
1
g
2
N U. Ap-
ply (b) to obtain an N
t
such that N
t
N
t
N; then g
1
g
2
N
t
N
t
U. But g
1
g
2
N
t
N
t
=
g
1
(g
2
N
t
g
-1
2
)g
2
N
t
, and it remains to apply (d) to obtain an N
1
N such that N
1

g
2
N
t
g
-1
2
.
Finally, we use (c) and (d) to show that g g
-1
is continuous. Given an open U G
and a g G such that g
-1
U, we have to nd an N N such that gN U
-1
. By
denition, there exists an N N such that g
-1
N U. Now N
-1
g U
-1
, and we use
(c) to obtain an N
t
N such that N
t
g U
-1
, and (d) to obtain an N
tt
N such that
gN
tt
g(g
-1
N
t
g) U
-1
.
2
The Krull topology on the Galois group
Recall (3.9) that a nite extension of J is Galois over J if it is normal and separable, i.e.,
if every irreducible polynomial JX| having a root in has deg distinct roots in .
Similarly, we dene an algebraic extension of J to be Galois over J if it is normal and
separable. For example, J
sep
is a Galois extension of J. Clearly, is Galois over J if and
only if it is a union of nite Galois extensions.
PROPOSITION 7.3 If is Galois over J, then it is Galois over every intermediate eld M.
PROOF. Let (X) be an irreducible polynomial in MX| having a root a in . The
minimum polynomial g(X) of a over J splits into distinct degree-one factors in X|. As
divides g (in MX|), it also must split into distinct degree-one factors in X|.
2
PROPOSITION 7.4 Let be a Galois extension of J and let 1 be a subeld of containing
J. Then every J-homomorphism 1 extends to an J-isomorphism .
PROOF. The same Zorns lemma argument as in the proof of Theorem 6.8 shows that every
J-homomorphism 1 extends to an J-homomorphism : . Let a , and let
The Krull topology on the Galois group 93
be its minimum polynomial over J. Then contains exactly deg( ) roots of , and so
therefore does (). Hence a (), which shows that is surjective.
2
COROLLARY 7.5 Let 1 J be as in the proposition. If 1 is stable under Aut(,J),
then 1 is Galois over J.
PROOF. Let (X) be an irreducible polynomial in JX| having a root a in 1. Because
is Galois over J, (X) has n = deg( ) distinct roots a
1
. . . . . a
n
in . There is an
J-isomorphism Ja| Ja
i
| sending a to a
i
(they are both stem elds for ), which
extends to an J-isomorphism . As 1 is stable under Aut(,J), this shows that
a
i
1.
2
Let be a Galois extension of J, and let G =Aut(,J). For any nite subset S of
, let
G(S) ={o G [ os =s for all s S].
PROPOSITION 7.6 There is a unique structure of a topological group on G for which the
sets G(S) form an open neighbourhood base of 1. For this topology, the sets G(S) with S
G-stable form a neighbourhood base of 1 consisting of open normal subgroups.
PROOF. We show that the collection of sets G(S) satises (a,b,c,d) of (7.2). It satises (a)
because G(S
1
) G(S
2
) =G(S
1
LS
2
). It satises (b) and (c) because each set G(S) is a
group. Let S be a nite subset of . Then J(S) is a nite extension of J, and so there are
only nitely many J-homomorphisms J(S) . Since oS =tS if o[J(S) =t[J(S),
this shows that

S =
_
cG
oS is nite. Now o

S =

S for all o G, and it follows that G(

S)
is normal in G. Therefore, oG(

S)o
-1
=G(

S) G(S), which proves (d). It also proves

the second statement.
2
The topology on Aut(,J) dened in the proposition is called the Krull topology. We
write Gal(,J) for Aut(,J) endowed with the Krull topology, and call it the Galois
group of ,J. The Galois group of J
sep
over J is called the absolute Galois group
3
of J.
If S is a nite set stable under G, then J(S) is a nite extension of J stable under G,
and hence Galois over J (7.5). Therefore,
{Gal(,1) [ 1 nite and Galois over J]
is a neighbourhood base of 1 consisting of open normal subgroups.
PROPOSITION 7.7 Let be Galois over J. For every intermediate eld 1 nite and Galois
over J, the map
o o[1: Gal(,J) Gal(1,J)
is a continuous surjection (discrete topology on Gal(1,J)).
3
But note that the absolute Galois group of J is only dened up to an inner automorphism: let J
t
be a
second separable algebraic closure of J; the choice of an isomorphism J
t
J
sep
determines an isomorphism
Gal(J
t
,J) Gal(J
sep
,J); a second isomorphism J
t
J
sep
will differ from the rst by an element o of
Gal(J
sep
,J), and the isomorphism Gal(J
t
,J) Gal(J
sep
,J) it denes differs from the rst by inn(o).
94 7. INFINITE GALOIS EXTENSIONS
PROOF. Let o Gal(1,J), and regard it as an J-homomorphism 1 . Then o extends
to an J-isomorphism (see 7.4), which shows that the map is surjective. For every
nite set S of generators of 1 over J, Gal(,1) =G(S), which shows that the inverse
image of 1
Gal(T{T)
is open in G. By homogeneity, the same is true for every element of
Gal(1,J).
2
PROPOSITION 7.8 All Galois groups are compact and totally disconnected.
4
PROOF. Let G =Gal(,J). We rst show that G is Hausdorff. If o =t, then o
-1
t =1
G
,
and so it moves some element of , i.e., there exists an a such that o(a) =t(a). For
any S containing a, oG(S) and tG(S) are disjoint because their elements act differently on
a. Hence they are disjoint open subsets of G containing o and t respectively.
We next show that G is compact. As we noted above, if S is a nite set stable under G,
then G(S) is a normal subgroup of G, and it has nite index because it is the kernel of
G Sym(S).
Since every nite set is contained in a stable nite set, the argument in the last paragraph
shows that the map
G

S nite stable under G

G,G(S)
is injective. When we endow

G,G(S) with the product topology, the induced topology

on G is that for which the G(S) form an open neighbourhood base of e, i.e., it is the
Krull topology. According to the Tychonoff theorem,

G,G(S) is compact, and so it
remains to show that G is closed in the product. For each S
1
S
2
, there are two continuous
maps

G,G(S) G,G(S
1
), namely, the projection onto G,G(S
1
) and the projection
onto G,G(S
2
) followed by the quotient map G,G(S
2
) G,G(S
1
). Let 1(S
1
. S
2
) be
the closed subset of

G,G(S) on which the two maps agree. Then

_
S
1
S
2
1(S
1
. S
2
) is
closed, and equals the image of G.
Finally, for each nite set S stable under G, G(S) is a subgroup that is open and hence
closed. Since
_
G(S) ={1
G
], this shows that the connected component of G containing
1
G
is just {1
G
]. By homogeneity, a similar statement is true for every element of G.
2
PROPOSITION 7.9 For every Galois extension ,J,
Gal({T)
=J.
PROOF. Every element of J lies in a nite Galois extension of J, and so this follows
from the surjectivity in Proposition 7.7.
2
ASIDE 7.10 There is a converse to Proposition 7.8: every compact totally disconnected group
arises as the Galois group of some Galois extension of elds of characteristic zero (Douady, A.,
Cohomologie des groupes compact totalement discontinus (dapr` es J. Tate), S eminaire Bourbaki
1959/60, no. 189). However, not all such groups arise as the absolute Galois group of a eld of
characteristic zero. For example, the absolute Galois group of a eld of characteristic zero, if nite,
must have order 1 or 2.
5
4
Following Bourbaki, we require compact spaces to be Hausdorff. A topological space is totally discon-
nected if its connected components are the one-point sets.
5
Theorem (Artin-Schreier, 1927): Let 1 be an algebraically closed eld and let J be a proper subeld of 1
with 1: J| <o. Then F is real-closed and 1 =J
_
1|. See, for example, Jacobson 1964, Chapter VI.
The fundamental theorem of innite Galois theory 95
The fundamental theorem of innite Galois theory
PROPOSITION 7.11 Let be Galois over J, with Galois group G.
(a) Let M be a subeld of containing J. Then is Galois over M, the Galois group
Gal(,M) is closed in G, and
Gal({)
=M.
(b) For every subgroup H of G, Gal(,
1
) is the closure of H.
PROOF. (a) The rst assertion was proved in (7.3). For each nite subset S M, G(S) is
an open subgroup of G, and hence it is closed. But Gal(,M) =
_
S
G(S), and so it
also is closed. The nal statement now follows from (7.9).
(b) Since Gal(,
1
) contains H and is closed, it certainly contains the closure

H of
H. On the other hand, let o G

H; we have to show that o moves some element of
1
.
Because o is not in the closure of H,
o Gal(,1) H =0
for some nite Galois extension 1 of J in (because the sets Gal(,1) form a neigh-
bourhood base of 1; see above). Let denote the surjective map Gal(,J) Gal(1,J).
Then o[1 H, and so o moves some element of 1
1

1
(apply 3.11b).
2
THEOREM 7.12 Let be Galois over J with Galois group G. The maps
H
1
. M Gal(,M)
are inverse bijections between the set of closed subgroups of G and the set of intermediate
elds between and J:
{closed subgroups of G] -{intermediate elds J M ].
Moreover,
(a) the correspondence is inclusion-reversing: H
1
H
2

1
1

1
2
;
(b) a closed subgroup H of G is open if and only if
1
has nite degree over J, in
which case (G: H) =
1
: J|;
(c) oHo
-1
-oM, i.e.,
c1c
1
=o(
1
); Gal(,oM) =o Gal(,M)o
-1
;
(d) a closed subgroup H of G is normal if and only if
1
is Galois over J, in which
case Gal(
1
,J) .G,H.
PROOF. For the rst statement, we have to show that H
1
and M Gal(,M) are
inverse maps.
Let H be a closed subgroup of G. Then is Galois over
1
and Gal(,
1
) =H
(see 7.11).
Let M be an intermediate eld. Then Gal(,M) is a closed subgroup of G and

Gal({)
=M (see 7.11).
(a) We have the obvious implications:
H
1
H
2
==
1
1

1
2
== Gal(,
1
1
) Gal(,
1
2
).
But Gal(,
1
i
) =H
i
(see 7.11).
96 7. INFINITE GALOIS EXTENSIONS
(b) As we noted earlier, a closed subgroup of nite index in a topological group is always
open. Because G is compact, conversely an open subgroup of G is always of nite index.
Let H be such a subgroup. The map o o[
1
denes a bijection
G,H Hom
T
(
1
. )
(apply 7.4) from which the statement follows.
(c) For t G and , t = oto
-1
(o) =o. Therefore, Gal(,oM) =
o Gal(,M)o
-1
, and so o Gal(,M)o
-1
-oM.
(d) Let H -M. It follows from (c) that H is normal if and only if M is stable under
the action of G. But M is stable under the action of G if and only it is a union of nite
extensions of J stable under G, i.e., of nite Galois extensions of G. We have already
observed that an extension is Galois if and only if it is a union of nite Galois extensions.
2
REMARK 7.13 As in the nite case (3.17), we can deduce the following statements.
(a) Let (M
i
)
iJ
be a (possibly innite) family of intermediate elds, and let H
i
-M
i
.
Let

M
i
be the smallest eld containing all the M
i
; then because
_
iJ
H
i
is the largest
(closed) subgroup contained in all the H
i
,
Gal(,

M
i
) =
_
iJ
H
i
.
(b) Let M - H. The largest (closed) normal subgroup contained in H is N =
_
c
oHo
-1
(cf. GT 4.10), and so
1
, which is the composite of the elds oM, is the
smallest normal extension of J containing M.
PROPOSITION 7.14 Let 1 and 1 be eld extensions of J con-
tained in some common eld. If 1,J is Galois, then 11,1 and
1,11 are Galois, and the map
o o[1: Gal(11,1) Gal(1,11)
is an isomorphism of topological groups.
11
1 1
11
J
=
=
PROOF. We rst prove that the map is continuous. Let G
1
= Gal(11,1) and let G
2
=
Gal(1,11). For any nite set S of elements of 1, the inverse image of G
2
(S) in G
1
is
G
1
(S).
We next show that the map is an isomorphism of groups (neglecting the topology). As
in the nite case, it is an injective homomorphism (3.18). Let H be the image of the map.
Then the xed eld of H is 11, which implies that H is dense in Gal(1,11). But H
is closed because it is the continuous image of a compact space in a Hausdorff space, and so
H =Gal(1,11).
Finally, we prove that it is open. An open subgroup of Gal(11,1) is closed (hence
compact) of nite index; therefore its image in Gal(1,11) is compact (hence closed) of
nite index, and hence open.
2
COROLLARY 7.15 Let be an algebraically closed eld containing J, and let 1 and 1
be as in the proposition. If j: 1 and o: 1 are J-homomorphisms such that
j[11 =o[11, then there exists an J-homomorphism t: 11 such that t[1 =j
and t[1 =o.
The fundamental theorem of innite Galois theory 97
PROOF. According to (7.4), o extends to an J-homomorphism s: 11 . As s[11 =
j[11, we can write s[1 =jc for some c Gal(1,11). According to the proposition,
there exists a unique e Gal(11,1) such that e[1 =c. Dene t =s e
-1
.
2
EXAMPLE 7.16 Let be an algebraic closure of the nite eld F
]
. Then G =Gal(,F
]
)
contains a canonical Frobenius element, o = (a a
]
), and it is generated by it as a
topological group, i.e., G is the closure of (o). We now determine the structure of G.
Endow Z with the topology for which the groups nZ, n _1, form a fundamental system
of neighbourhoods of 0. Thus two integers are close if their difference is divisible by a large
integer.
As for any topological group, we can complete Z for this topology. A Cauchy sequence
in Z is a sequence (a
i
)
i_1
, a
i
Z, satisfying the following condition: for all n _ 1, there
exists an N such that a
i
a
}
mod n for i. > N. Call a Cauchy sequence in Z trivial if
a
i
0 as i o, i.e., if for all n _ 1, there exists an N such that a
i
0 mod n for all
i >N. The Cauchy sequences form a commutative group, and the trivial Cauchy sequences
form a subgroup. We dene

Z to be the quotient of the rst group by the second. It has a
ring structure, and the map sending m Z to the constant sequence m. m. m. . . . identies Z
with a subgroup of

Z.
Let

Z be represented by the Cauchy sequence (a

i
). The restriction of the Frobenius
element o to F
]
n has order n. Therefore (o[F
]
n)
o
i
is independent of i provided it is
sufciently large, and we can dene o

Gal(,F
]
) to be such that, for each n, o

[F
]
n =
(o[F
]
n)
o
i
for all i sufciently large (depending on n). The map o

Z Gal(,F
]
)
is an isomorphism.
The group

Z is uncountable. To most analysts, it is a little weirdits connected
components are one-point sets. To number theorists it will seem quite natural the
Chinese remainder theorem implies that it is isomorphic to

] prime
Z
]
where Z
]
is the ring
of -adic integers.
EXAMPLE 7.17 Let Q
al
be the algebraic closure of Q in C. Then Gal(Q
al
,Q) is one of the
most basic, and intractable, objects in mathematics. It is expected that every nite group
occurs as a quotient of it. This is known, for example, for S
n
and for every sporadic simple
group except possibly M
23
. See (5.41) and mo80359.
On the other hand, we do understand Gal(J
ab
,J) where J Q
al
is a nite extension
of Q and J
ab
is the union of all nite abelian extensions of J contained in Q
al
. For example,
Gal(Q
ab
,Q) .

. This is abelian class eld theory see my notes Class Field Theory.
ASIDE 7.18 A simple Galois correspondence is a system consisting of two partially ordered sets
1 and Q and order reversing maps : 1 Q and g: Q1 such that g() _ for all 1
and g(q) _ q for all q Q. Then g = , because g() _ and g() _ implies
(g) _() for all 1. Similarly, gg =g, and it follows that and g dene a one-to-one
correspondence between the sets g(Q) and (1).
From a Galois extension of J we get a simple Galois correspondence by taking 1 to be the
set of subgroups of Gal(,J) and Q to be the set of subsets of , and by setting (H) =
1
and
g(S) =G(S). Thus, to prove the one-to-one correspondence in the fundamental theorem, it sufces
to identify the closed subgroups as exactly those in the image of g and the intermediate elds as
exactly those in the image of . This is accomplished by (7.11).
98 7. INFINITE GALOIS EXTENSIONS
Galois groups as inverse limits
DEFINITION 7.19 A partial ordering _ on a set 1 is said to be directed, and the pair (1. _)
is called a directed set, if for all i. 1 there exists a k 1 such that i. _k.
DEFINITION 7.20 Let (1. _) be a directed set, and let C be a category (for example, the cat-
egory of groups and homomorphisms, or the category of topological groups and continuous
homomorphisms).
(a) An inverse system in C indexed by (1. _) is a family (
i
)
iJ
of objects of C together
with a family (
}
i
:
}

i
)
i_}
of morphisms such that
i
i
=id

i
and
}
i

k
}
=
k
i
all i _ _k.
(b) An object of C together with a family (
}
:
}
)
}J
of morphisms satisfying

}
i

}
=
i
all i _ is said to be an inverse limit of the system in (a) if it has the
following universal property: for any other object T and family (q
}
: T
}
) of
morphisms such
}
i
q
}
=q
i
all i _ , there exists a unique morphism r: T
such that
}
r =q
}
for ,
T

i
q
j
q
i
]
i
]
j
]
j
i
i
Clearly, the inverse limit (if it exists), is uniquely determined by this condition up to a unique
isomorphism. We denote it lim

(
i
.
}
i
), or just lim

i
.
EXAMPLE 7.21 Let (G
i
.
}
i
: G
}
G
i
) be an inverse system of groups. Let
G ={(g
i
)

G
i
[
}
i
(g
}
) =g
i
all i _ ].
and let
i
: G G
i
be the projection map. Then
}
i

}
=
i
is just the equation
}
i
(g
}
) =
g
i
. Let (H. q
i
) be a second family such that
}
i
q
}
=q
i
. The image of the homomorphism
h (q
i
(h)): H

G
i
is contained in G, and this is the unique homomorphism H G carrying q
i
to
i
. Hence
(G.
i
) =lim

(G
i
.
}
i
).
EXAMPLE 7.22 Let (G
i
.
}
i
: G
}
G
i
) be an inverse system of topological groups and
continuous homomorphisms. When endowed with the product topology,

G
i
becomes a
topological group
G ={(g
i
)

G
i
[
}
i
(g
}
) =g
i
all i _ ].
Nonopen subgroups of nite index 99
and G becomes a topological subgroup with the subspace topology. The projection maps

i
are continuous. Let H be (H. q
i
) be a second family such that
}
i
q
}
= q
i
. The
homomorphism
h (q
i
(h)): H

G
i
is continuous because its composites with projection maps are continuous (universal property
of the product). Therefore H G is continuous, and this shows that (G.
i
) =lim

(G
i
.
}
i
).
EXAMPLE 7.23 Let (G
i
.
}
i
: G
}
G
i
) be an inverse system of nite groups, and regard
it as an inverse system of topological groups by giving each G
i
the discrete topology. A
topological group G arising as an inverse limit of such a system is said to be pronite
6
.
If (.
i
) G, say
}
0
i
0
(.
}
0
) =.
i
0
, then
G{(g
}
) [ g
}
0
=.
}
0
. g
i
0
=.
i
0
] =0.
As the second set is an open neighbourhood of (.
i
), this shows that G is closed in

G
i
. By
Tychonoffs theorem,

G
i
is compact, and so G is also compact. The map
i
: G G
i
is
continuous, and its kernel U
i
is an open subgroup of nite index in G (hence also closed).
As
_
U
i
={e], the connected component of G containing e is just {e]. By homogeneity,
the same is true for every point of G: the connected components of G are the one-point sets
G is totally disconnected.
We have shown that a pronite group is compact and totally disconnected, and it is an
exercise to prove the converse.
7
EXAMPLE 7.24 Let be a Galois extension of J. The composite of two nite Galois
extensions of in is again a nite Galois extension, and so the nite Galois subextensions
of form a directed set 1. For each 1 in 1 we have a nite group Gal(1,J), and for each
1 1
t
we have a restriction homomorphism
T
0
T
: Gal(1
t
,J) Gal(1,J). In this way,
we get an inverse system of nite groups (Gal(1,J).
T
0
T
) indexed by 1.
For each 1, there is a restriction homomorphism
T
: Gal(,J) Gal(1,J) and,
because of the universal property of inverse limits, these maps dene a homomorphism
Gal(,J) lim

Gal(1,J).
This map is an isomorphism of topological groups. This is a restatement of what we showed
in the proof of (7.8).
Nonopen subgroups of nite index
We apply Zorns lemma
8
to construct a nonopen subgroup of nite index in Gal(Q
al
,Q).
9
LEMMA 7.25 Let V be an innite dimensional vector space. For all n _ 1, there exists a
subspace V
n
of V such that V,V
n
has dimension n.
6
An inverse limit is also called a projective limit. Thus a pronite group is a projective limit of nite groups.
7
More precisely, it is Exercise 3 of 7 of Chapter 3 of Bourbakis General Topology.
8
This is really needed see mo106216.
9
Contrast: . . . it is not known, even when G = Gal(

Q,Q), whether every subgroup of nite index in

G is open; this is one of a number of related unsolved problems, all of which appear to be very difcult.
Swinnerton-Dyer, H. P. F., A brief guide to algebraic number theory. Cambridge, 2001, p133.
100 7. INFINITE GALOIS EXTENSIONS
PROOF. Zorns lemma shows that V contains maximal linearly independent subsets, and
then the usual argument shows that such a subset spans V , i.e., is a basis. Choose a basis,
and take V
n
to be the subspace spanned by the set obtained by omitting n elements from the
basis.
2
PROPOSITION 7.26 The group Gal(Q
al
,Q) has nonopen normal subgroups of index 2
n
for
all n >1.
PROOF. Let 1 be the subeld Q
_
1.
_
2. . . . .
_
. . . .|, prime, of C. For each ,
Gal(Q
_
1.
_
2. . . . .
_
|,Q)
is a product of copies of Z,2Z indexed by the set {primes _] L{o] (apply 5.31; see also
5.30b). As
Gal(1,Q) =lim

Gal(Q
_
1.
_
2. . . . .
_
|,Q).
it is a direct product of copies of Z,2Z indexed by the primes l of Q (including l =o)
endowed with the product topology. Let G =Gal(1,Q), and let
H ={(a
I
) G [ a
I
=0 for all but nitely many l].
This is a subgroup of G (in fact, it is a direct sum of copies of Z,2Z indexed by the primes
of Q), and it is dense in G because
10
clearly every open subset of G contains an element of
H. We can regard G,H as vector space over F
2
and apply the lemma to obtain subgroups
G
n
of index 2
n
in G containing H. If G
n
is open in G, then it is closed, which contradicts
the fact that H is dense. Therefore, G
n
is not open, and its inverse image in Gal(Q
al
,Q) is
the desired subgroup.
11
2
ASIDE 7.27 Let G = Gal(Q
al
,Q). We showed in the above proof that there is a closed normal
subgroup N =Gal(Q
al
,1) of G such that G,N is an uncountable vector space over F
2
. Let (G,N)

be the dual of this vector space (also uncountable). Every nonzero (G,N)

denes a surjective
map G F
2
whose kernel is a subgroup of index 2 in G. These subgroups are distinct, and so G
has uncountably many subgroups of index 2. Only countably many of them are open because Q has
only countably many quadratic extensions in a xed algebraic closure.
ASIDE 7.28 Let G be a pronite group that is nitely generated as a topological group. It is a
difcult theorem, only recently proved, that every subgroup of nite index in G is open (Nikolov,
Nikolay; Segal, Dan. On nitely generated pronite groups. I. Strong completeness and uniform
bounds. Ann. of Math. (2) 165 (2007), no. 1, 171238.)
10
Alternatively, let (a
I
) G; then the sequence
(a
o
. 0. 0. 0. . . .), (a
o
. a
2
. 0. 0. . . .), (a
o
. a
2
. a
3
. 0. . . .). . . .
in H converges to (a
I
).
11
The inverse image is not open because every continuous homomorphism from a compact group to a
separated group is open. Alternatively, if the inverse image were open, its xed eld would be a nontrivial
extension 1 of Q contained in Q
_
1.
_
2. . . . .
_
. . . .|; but then 1 would be xed by G
n
, which is dense.
CHAPTER 8
The Galois theory of etale algebras
For Grothendieck, the classication of eld extensions by Galois groups, and the classi-
cation of covering spaces by fundamental groups, are two aspects of the same theory. In
this chapter, we re-interprete classical Galois theory from Grothendiecks point of view. We
assume the reader is familiar with the language of category theory (Wikipedia: Category
theory; Equivalence of categories).
Throughout, J is a eld, all rings and J-algebras are commutative, and unadorned tensor
products are over J. An J-algebra is nite if it is nitely generated as an J-module.
Review of commutative algebra
We shall need the following standard results from commutative algebra.
Two ideals 1 and J in a ring are said to be relatively prime if 1 J =. For example,
any two distinct maximal ideals in are relatively prime.
THEOREM 8.1 (CHINESE REMAINDER THEOREM) Let 1
1
. . . . . 1
n
be ideals in a ring . If
1
i
is relatively prime to 1
}
whenever i = , then the map
a (. . . . a1
i
. . . .): ,1
1
,1
n
(5)
is surjective with kernel

1
i
(so

1
i
=
_
1
i
).
PROOF. CA 2.12.
2
THEOREM 8.2 (STRONG NULLSTELLENSATZ) Let 1 be an ideal in the polynomial ring
JX
1
. . . . . X
n
| and let 7(1) denote the set of zeros of 1 in (J
al
)
n
. If a polynomial h
JX
1
. . . . . X
n
| vanishes on 7(1), then some power of it lies in 1.
PROOF. CA 12.8.
2
The radical of an ideal 1 in a ring is the set of such that
n
1 for some
n N. It is again an ideal, and it is equal to its own radical.
The nilradical N of is the radical of the ideal (0). It consists of the nilpotents in . If
N =0, then is said to be reduced.
101
102 8. THE GALOIS THEORY OF ETALE ALGEBRAS
PROPOSITION 8.3 Let be a nitely generated J-algebra, and let 1 be an ideal in . The
radical of 1 is equal to the intersection of the maximal ideals containing it:
rad(1) =
_
{M [ M 1, M maximal].
In particular, is reduced if and only if
_
{M [ M maximal] =0.
PROOF. Because of the correspondence between ideals in a ring and in a quotient of the
ring, it sufces to prove this for =JX
1
. . . . . X
n
|.
The inclusion rad(1)
_
{M [ M 1, M maximal] holds in any ring (because maximal
ideals are radical and rad(1) is the smallest radical ideal containing 1).
For the reverse inclusion, let h lie in all maximal ideals containing 1, and let (a
1
. . . . . a
n
)
7(1). The image of the evaluation map
(a
1
. . . . . a
n
): JX
1
. . . . . X
n
| J
al
is a subring of J
al
which is algebraic over J, and hence is a eld (see 1.31a). Therefore,
the kernel of the map is a maximal ideal, which contains 1, and therefore also contains h.
This shows that h(a
1
. . . . . a
n
) = 0, and we conclude from the strong Nullstellensatz that
h rad(1).
2

Etale algebras over a eld

DEFINITION 8.4 An J-algebra is diagonalizable if it is isomorphic to the product algebra
J
n
for some n, and it is etale if 1 is diagonalizable for some eld 1 containing J.
1
The degree : J| of a nite J-algebra is its dimension as an J-vector space.
Let be a nite J-algebra. For any nite set S of maximal ideals in , the Chinese
remainder theorem (8.1) shows that the map

S
,M is surjective with kernel
_
S
M. In particular, [S[ _: J|, and so has only nitely many maximal ideals. If S
is the set of all maximal ideals in , then
_
S
M is the nilradical N of (8.3), and so
,N is a nite product of elds.
PROPOSITION 8.5 The following conditions on a nite J-algebra are equivalent:
(a) is etale;
(b) 1 is reduced for all elds 1 containing J:
(c) is a product of separable eld extensions of J.
PROOF. (a)=(b). Let 1 be a eld containing J. By hypothesis, there exists a eld 1
t
containing J such that 1
t
is diagonalizable. Let 1
tt
be a eld containing (copies
of) both 1 and 1
t
(e.g., take 1
tt
to be a quotient of 11
t
by a maximal ideal). Then
1
tt
=1
tt

1
0 1
t
is diagonalizable, and the map 1 1
tt
dened by the
inclusion 1 1
tt
is injective, and so 1 is reduced.
(b)=(c). The map a 1a: 1 is injective, and so if 1 is reduced, then
so also is . The discussion above shows that it is a nite product of elds. Let J
t
be one of
the factors of . If J
t
is not separable over J, then J has characteristic =0 and there
exists an element u of J
t
whose minimum polynomial is of the form (X
]
) with JX|
1
This is Bourbakis terminology

Etale algebras over a eld 103

(see 3.6, et seq.). Let 1 be a eld containing J such that all the coefcients of are th
powers in 1. Then
1Ju| .1(JX|,((X
]
)) .1X|,((X
]
)).
which is not reduced because (X
]
) is a th power in 1X|. Hence 1 is not reduced.
(c)=(a). We may suppose that itself is a separable eld extension of J. From the
primitive element theorem (5.1), we know that = Ju| for some u. Because Ju| is
separable over J, the minimum polynomial (X) of u is separable, which means that
(X) =

(X u
i
). u
i
=u
}
for i =.
in a splitting eld 1 for . Now
1 .1JX|,( ) .1X|,( ),
and, according to the Chinese remainder theorem (8.1),
1X|,( ) .

i
1X|,(X u
i
) .1 1.
2
COROLLARY 8.6 An J-algebra is etale if and only if J
sep
is diagonalizable.
PROOF. The proof that (c) implies (a) in (8.5) shows that 1 is diagonalizable if certain
separable polynomials split in 1. By denition, all separable polynomials split in J
sep
.
2
EXAMPLE 8.7 Let JX|, and let =JX|,( ). Let =

n
i
i
with the
i
irre-
ducible and distinct. According to the Chinese remainder theorem (CA 2.12)
.

i
JX|,(
n
i
i
).
The J-algebra JX|,(
n
i
i
) is a eld if and only if m
i
=1, in which case it is a separable
extension of J if and only if
i
is separable. Therefore is an etale J-algebra if and only if
is a separable polynomial.
PROPOSITION 8.8 Finite products, tensor products, and quotients of diagonalizable (resp.
etale) J-algebras are diagonalizable (resp. etale).
PROOF. This is obvious for diagonalizable algebras, and it follows for etale algebras.
2
COROLLARY 8.9 The composite of any nite set of etale subalgebras of a J-algebra is
etale.
PROOF. Let
i
be etale subalgebras of T. Then
1

n
is the image of the map
a
1
a
n
a
1
a
n
:
1

n
T.
and so is a quotient of
1

n
.
2
PROPOSITION 8.10 Let be etale over J, and let J
t
be a eld containing J. Then J
t

is etale over J
t
.
104 8. THE GALOIS THEORY OF ETALE ALGEBRAS
PROOF. Let 1 be such that 1~1
n
, and let 1
t
be a eld containing (copies of) both 1
and J
t
. Then
1
t

T
0
_
J
t

_
.1
t
.1
t

1
(1) ~1
t

1
1
n
.
_
1
t
_
n
.
2
REMARK 8.11 Let be an etale algebra over J, and write as a product of elds, =

i

i
. A generator for as an J-algebra is a tuple (
i
) with each
i
a generator for
i
as an J-algebra. Because each
i
is separable over J, such an exists (primitive element
theorem 5.1). Choose an , and let =

i

i
be the product of the minimum polynomials
of the
i
. Then is a monic polynomial whose irreducible factors are separable.
Conversely, let be a monic polynomial whose irreducible factors (
i
)
i
are separable.
Then
def
=

i
JX|,(
i
) is an etale algebra over J with a canonical generator.
In this way, we get a one-to-one correspondence between the set of isomorphism classes
of pairs (. ) consisting of an etale J-algebra and a generator and the set of monic
polynomials whose irreducible factors are separable.
Classication of etale algebras over a eld
Fix a separable closure of J, and let G be the Galois group of over J. Recall (Chapter
7) that this is the group of J-automorphisms of equipped with the Krull topology. Let 1
be a subeld of , nite and Galois over J. An argument using Zorns lemma shows that
o o[1: G Gal(1,J)
is surjective. The open normal subgroups of G are exactly the kernels of such homomor-
phisms, and G =lim

Gal(1,J).
Let X be a nite set with an action of G,
GX X.
We say that the action is continuous if the map is continuous for the discrete topology on X
and the Krull topology on G. Because X is nite, this is equivalent to saying that the action
factors through G Gal(1,J) for some subeld 1 of nite and Galois over J.
For an etale J-algebra , let F() denote the set of J-algebra homomorphisms .
Then G acts on F() through its action on :
(o )(a) =o((a)). o G, F(), a .
i.e., o =o . There exists a nite Galois extension 1 of J containing the image of every
homomorphism , and the action of G on F() factors through Gal(1,J); therefore
it is continuous.
Now F() is a contravariant functor from the category of etale J-algebras to the
category of nite continuous G-sets.
EXAMPLE 8.12 Let =JX|,( ) where is a separable polynomial in JX|. Then
F() .{roots of (X) in ].
Classication of etale algebras over a eld 105
Suppose is a product of etale J-algebras, =
1

n
. Because has no
nonzero zero divisors, every homomorphism : is zero on all but one
i
, and so, to
give a homomorphism amounts to giving a homomorphism
i
for some i . In
other words,
F(

i

i
) .

i
F(
i
).
In particular, for an etale J-algebra .

i
J
i
,
F() .
_
i
Hom
T-algebra
(J
i
. ).
From Proposition 2.7, we deduce that F() is nite of order : J|.
THEOREM 8.13 The functor F() is a contravariant equivalence from the category
of etale J-algebras to the category of nite continuous G-sets.
PROOF. We have to prove the following two statements.
(a) The functor F is fully faithful, i.e., for all etale J-algebras and T, the map
Hom
T-algebras
(. T) Hom
G-sets
(F(T). F())
is bijective.
(b) The functor F is essentially surjective, i.e., every nite continuous G-set is isomorphic
to F() for some etale J-algebra .
Let V be a vector space over J, and let V

=
T
V . Then G acts on V

through its
action on , and
V .(V

)
G
def
={ V

[ o = for all o G].

To see this, choose an J-basis e ={e
1
. . . . . e
n
] for V . Then e is an -basis forV

, and
o(a
1
e
1
a
n
e
n
) =(oa
1
)e
1
(oa
n
)e
n
. a
i
.
Therefore a
1
e
1
a
n
e
n
is xed by all o G if and only if a
1
. . . . . a
n
J.
Similarly, if W is a second vector space over J, then G acts on Hom
-linear
(V

. W

)
by o =o o
-1
, and
Hom
T-linear
(V. W) .Hom
-linear
(V

. W

)
G
. (6)
Indeed, a choice of bases for V and W determines isomorphisms Hom
T-linear
(V. W) .
M
n,n
(J) (mn matrices with entries from J) and Hom
-linear
(V

. W

) . M
n,n
(),
and G acts on M
n,n
() in the obvious way. Now (6) follows from the obvious statement:
M
n,n
(J) =M
n,n
()
G
.
Let and T be etale J-algebras. Under the isomorphism
Hom
T-linear
(. T) .Hom
-linear
(

. T

)
G
.
J-algebra homomorphisms correspond to -algebra homomorphisms, and so
Hom
T-algebra
(. T) .Hom
-algebra
(

. T

)
G
.
From (8.6), we know that

(resp. T

) is a product of copies of indexed by the elements

of F() (resp. F(T)). Let t be a map of sets F(T) F(). Then
(a
i
)
iF()
(b
}
)
}F(B)
. b
}
=a
t(})
.
106 8. THE GALOIS THEORY OF ETALE ALGEBRAS
is a homomorphism of -algebras

, and every homomorphism of -algebras

is of this form for a unique t . Thus

Hom
-algebra
(

. T

) .Hom
Sets
(F(T). F()).
This isomorphism is compatible with the actions of G, and so
Hom
-algebra
(

. T

)
G
.Hom
Sets
(F(T). F())
G
.
In other words,
Hom
T-algebra
(. T) .Hom
G-sets
(F(T). F()).
This proves (a). For (b), let S be a nite G-set, and let S =

iJ
S
i
be the decomposition
of S into a union of G-orbits. For each i , choose an s
i
S
i
, and let J
i
be the subeld of
xed by the stabilizer of s
i
. Then
F
_
iJ
J
i
_
.S.
2
SECOND PROOF OF THEOREM 8.13
We sketch a second proof of the theorem. For a nite set S with a continuous action of G,
we let

S
=Hom(S. ).
In other words,
S
is a product of copies of indexed by the elements of S. We let G act
on
S
through its actions on both and S:
(; )(o) =;((;
-1
o)). ; G. : S . o S.
For every etale J-algebra , there is a canonical isomorphism
ac (oa c)
cF()
:
F()
. (7)
When we let G act on through its action on , the map (7) becomes equivariant.
Now:
(a) for every etale J-algebra ,
.()
G
:
(b) for every nite set S with a continuous action of G, (
S
)
G
is an etale J-subalgebra
of
S
, and
F((
S
)
G
) .S.
Therefore, F() is an equivalence of categories with quasi-inverse S (
S
)
G
.
IMPROVEMENT OF THEOREM 8.13
Fix a Galois extension of J (nite or innite), and let G = Gal(,J). An etale J-
algebra is split by if is isomorphic to a product of copies of . For such an
J-algebra, let F() =Hom
k-algebra
(. ).
THEOREM 8.14 The functor F() is a contravariant equivalence from the category
of etale J-algebras split by to the category of nite continuous G-sets.
The proof is the same as that of Theorem 8.13. When is a nite extension of J,
continuous may be omitted.
Comparison with the theory of covering spaces. 107
GEOMETRIC RE-STATEMENT OF THEOREM 8.13
In this subsection, we assume that the reader is familiar with the notion of an algebraic
variety over a eld J (geometrically-reduced separated scheme of nite type over J). The
functor Spec() is a contravariant equivalence from the category of etale algebras
over J to the category of zero-dimensional algebraic varieties over J. In particular, all
zero-dimensional algebraic varieties are afne. If V =Spec(), then
Hom
T-algebra
(. ) .Hom
Spec(T)
(Spec(). V )
def
=V()
(set of points of V with coordinates in ).
THEOREM 8.15 The functor V V() is an equivalence from the category of zero-
dimensional algebraic varieties over J to the category of nite continuous G-sets. Under
this equivalence, connected varieties correspond to sets with a transitive action.
PROOF. Combine Theorem 8.13 with the equivalence Spec().
2
Comparison with the theory of covering spaces.
The reader should note the similarity of (8.13) and (8.15) with the following statement:
Let J be a connected and locally simply connected topological space, and let
: J be a universal covering space of J. Let G denote the group of
covering transformations of ,J (the choice of a point e determines an
isomorphism of G with the fundamental group
1
(J. e)). For a covering space
1 of J, let F(1) denote the set of covering maps 1. Then 1 F(1)
is an equivalence from the category of covering spaces of J to the category of
(right) G-sets.
For more on this, see the section on the etale fundamental group in my Lectures on

Etale
Cohomology and Szamuely, Tam as, Galois groups and fundamental groups. CUP, 2009.
CHAPTER 9
Transcendental Extensions
In this chapter we consider elds J with much bigger than J. For example, we
could have C Q.
Algebraic independence
Elements
1
. ....
n
of give rise to an J-homomorphism
(
1
. ....
n
): JX
1
. . . . . X
n
| .
If the kernel of this homomorphism is zero, then the
i
are said to be algebraically inde-
pendent over J, and otherwise, they are algebraically dependent over J. Thus, the
i
are algebraically dependent over J if there exists a nonzero polynomial (X
1
. .... X
n
)
JX
1
. .... X
n
| such that (
1
. ....
n
) =0, and they are algebraically independent if
a
i
1
,...,i
n
J.

a
i
1
,...,i
n

i
1
1
...
i
n
n
=0 == a
i
1
,...,i
n
=0 all i
1
. .... i
n
.
Note the similarity with linear independence. In fact, if is required to be homogeneous of
degree 1, then the denition becomes that of linear independence.
EXAMPLE 9.1 (a) A single element is algebraically independent over J if and only if it
is transcendental over J.
(b) The complex numbers and e are almost certainly algebraically independent over
Q, but this has not been proved.
An innite set is algebraically independent over J if every nite subset of is
algebraically independent; otherwise, it is algebraically dependent over J.
REMARK 9.2 If
1
. ....
n
are algebraically independent over J, then the map
(X
1
. .... X
n
) (
1
. ....
n
): JX
1
. .... X
n
| J
1
. ....
n
|
is an injection, and hence an isomorphism. This isomorphism then extends to the elds of
fractions,
X
i

i
: J(X
1
. .... X
n
) J(
1
. ....
n
)
In this case, J(
1
. ....
n
) is called a pure transcendental extension of J. The polynomial
(X) =X
n

1
X
n-1
(1)
n

n
has Galois group S
n
over J(
1
. ....
n
) (see 5.40).
109
110 9. TRANSCENDENTAL EXTENSIONS
LEMMA 9.3 Let ; and let . The following conditions are equivalent:
(a) ; is algebraic over J();
(b) there exist
1
. . . . .
n
J() such that ;
n

1
;
n-1

n
=0;
(c) there exist
0
.
1
. . . . .
n
J|, not all 0, such that
0
;
n

1
;
n-1

n
=0;
(d) there exists an (X
1
. . . . . X
n
. Y ) JX
1
. . . . X
n
. Y | and
1
. . . . .
n
such that
(
1
. . . . .
n
. Y ) =0 but (
1
. . . . .
n
. ;) =0.
PROOF. (a) ==(b) ==(c) ==(a) are obvious.
(d) ==(c). Write (X
1
. . . . . X
n
. Y ) as a polynomial in Y with coefcients in the ring
JX
1
. . . . . X
n
|,
(X
1
. . . . . X
n
. Y ) =

i
(X
1
. . . . . X
n
)Y
n-i
.
Then (c) holds with
i
=
i
(
1
. . . . .
n
).
(c) ==(d). The
i
in (c) can be expressed as polynomials in a nite number of elements

1
. . . . .
n
of , say,
i
=
i
(
1
. . . . .
n
) with
i
JX
1
. . . . . X
n
|. Then (d) holds with
=

i
(X
1
. . . . . X
n
)Y
n-i
.
2
DEFINITION 9.4 When ; satises the equivalent conditions of Lemma 9.3, it is said to be
algebraically dependent on (over J). A set T is algebraically dependent on if each
element of T is algebraically dependent on .
The theory in the remainder of this chapter is logically very similar to a part of linear
algebra. It is useful to keep the following correspondences in mind:
Linear algebra Transcendence
linearly independent algebraically independent
span(T) algebraically dependent on T
basis transcendence basis
dimension transcendence degree
Transcendence bases
THEOREM 9.5 (FUNDAMENTAL RESULT) Let = {
1
. ....
n
] and T = {
1
. ....
n
] be
two subsets of . Assume
(a) is algebraically independent (over J);
(b) is algebraically dependent on T (over J).
Then m_n.
We rst prove two lemmas.
LEMMA 9.6 (THE EXCHANGE PROPERTY) Let {
1
. ....
n
] be a subset of ; if is al-
gebraically dependent on {
1
. ....
n
] but not on {
1
. ....
n-1
], then
n
is algebraically
dependent on {
1
. ....
n-1
. ].
Transcendence bases 111
PROOF. Because is algebraically dependent on {
1
. . . . .
n
], there exists a polynomial
(X
1
. .... X
n
. Y ) with coefcients in J such that
(
1
. ....
n
. Y ) =0. (
1
. ....
n
. ) =0.
Write as a polynomial in X
n
,
(X
1
. .... X
n
. Y ) =

i
a
i
(X
1
. .... X
n-1
. Y )X
n-i
n
.
and observe that, because (
1
. . . . .
n
. Y ) =0, at least one of the polynomials
a
i
(
1
. ....
n-1
. Y ).
say a
i
0
, is not the zero polynomial. Because is not algebraically dependent on
{
1
. ....
n-1
].
a
i
0
(
1
. ....
n-1
. ) =0. Therefore, (
1
. ....
n-1
. X
n
. ) =0. Since (
1
. ....
n
. ) =
0, this shows that
n
is algebraically dependent on {
1
. ....
n-1
. ].
2
LEMMA 9.7 (TRANSITIVITY OF ALGEBRAIC DEPENDENCE) If C is algebraically depen-
dent on T, and T is algebraically dependent on , then C is algebraically dependent on
.
PROOF. The argument in the proof of Proposition 1.44 shows that if ; is algebraic over a
eld 1 which is algebraic over a eld J, then ; is algebraic over J (if a
1
. . . . . a
n
are the
coefcients of the minimum polynomial of ; over 1, then the eld Ja
1
. . . . . a
n
. ;| has
nite degree over J). Apply this with 1 =J(LT) and J =J().
2
PROOF (OF THEOREM 9.5) Let k be the number of elements that and T have in com-
mon. If k = m, then T, and certainly m _ n. Suppose that k < m, and write T =
{
1
. ....
k
.
k1
. ....
n
]. Since
k1
is algebraically dependent on {
1
. ....
k
.
k1
. ....
n
]
but not on {
1
. ....
k
], there will be a
}
, k 1 _ _ n, such that
k1
is algebraically
dependent on {
1
. ....
k
.
k1
. ....
}
] but not
{
1
. ....
k
.
k1
. ....
}-1
].
The exchange lemma then shows that
}
is algebraically dependent on
T
1
def
=T L{
k1
] {
}
].
Therefore T is algebraically dependent on T
1
, and so is algebraically dependent on T
1
(by 9.7). If k 1 <m, repeat the argument with and T
1
. Eventually well achieve k =m,
and m_n.
2
DEFINITION 9.8 A transcendence basis for over J is an algebraically independent set
such that is algebraic over J().
LEMMA 9.9 If is algebraic over J(), and is minimal among subsets of with this
property, then it is a transcendence basis for over J.
112 9. TRANSCENDENTAL EXTENSIONS
PROOF. If is not algebraically independent, then there is an that is algebraically
dependent on {]. It follows from Lemma 9.7 that is algebraic over J({]).
2
THEOREM 9.10 If there is a nite subset such that is algebraic over J(), then
has a nite transcendence basis over J. Moreover, every transcendence basis is nite,
and they all have the same number of elements.
PROOF. In fact, every minimal subset
t
of such that is algebraic over J(
t
) will be a
transcendence basis. The second statement follows from Theorem 9.5.
2
LEMMA 9.11 Suppose that is algebraically independent, but that L{] is algebraically
dependent. Then is algebraic over J().
PROOF. The hypothesis is that there exists a nonzero polynomial
(X
1
. .... X
n
. Y ) JX
1
. .... X
n
. Y |
such that (
1
. ....
n
. ) = 0, some distinct
1
. ....
n
. Because is algebraically
independent, Y does occur in . Therefore
=g
0
Y
n
g
1
Y
n-1
g
n
. g
i
JX
1
. .... X
n
|. g
0
=0. m_1.
As g
0
=0 and the
i
are algebraically independent, g
0
(
1
. ....
n
) =0. Because is a root
of
=g
0
(
1
. ....
n
)X
n
g
1
(
1
. ....
n
)X
n-1
g
n
(
1
. ....
n
).
it is algebraic over J(
1
. ....
n
) J().
2
PROPOSITION 9.12 Every maximal algebraically independent subset of is a transcen-
dence basis for over J.
PROOF. We have to prove that is algebraic over J() if is maximal among algebraically
independent subsets. But the maximality implies that, for every , L{] is
algebraically dependent, and so the lemma shows that is algebraic over J().
2
Recall that (except in 7), we use an asterisk to signal a result depending on Zorns
lemma.
THEOREM 9.13 Every algebraically independent subset of is contained in a transcen-
dence basis for over J; in particular, transcendence bases exist.
PROOF. Let S be the set of algebraically independent subsets of containing the given
set. We can partially order it by inclusion. Let T be a totally ordered subset of S, and let
T =
_
{ [ T ]. I claim that T S, i.e., that T is algebraically independent. If not,
there exists a nite subset T
t
of T that is not algebraically independent. But such a subset
will be contained in one of the sets in T , which is a contradiction. Now Zorns lemma shows
that there exists a maximal algebraically independent containing S, which Proposition 9.12
shows to be a transcendence basis for over J.
2
Transcendence bases 113
It is possible to show that any two (possibly innite) transcendence bases for over J
have the same cardinality. The cardinality of a transcendence basis for over J is called
the transcendence degree of over J. For example, the pure transcendental extension
J(X
1
. . . . . X
n
) has transcendence degree n over J.
EXAMPLE 9.14 Let
1
. . . . .
n
be the elementary symmetric polynomials in X
1
. . . . . X
n
.
The eld J(X
1
. . . . . X
n
) is algebraic over J(
1
. . . . .
n
), and so {
1
.
2
. . . . .
n
] contains a
transcendence basis for J(X
1
. . . . . X
n
). Because J(X
1
. . . . . X
n
) has transcendence degree
n, the
i
s must themselves be a transcendence basis.
EXAMPLE 9.15 Let be the eld of meromorphic functions on a compact complex mani-
fold M.
(a) The only meromorphic functions on the Riemann sphere are the rational functions in
:. Hence, in this case, is a pure transcendental extension of C of transcendence degree 1.
(b) If M is a Riemann surface, then the transcendence degree of over C is 1, and is
a pure transcendental extension of C M is isomorphic to the Riemann sphere
(c) If M has complex dimension n, then the transcendence degree is _n, with equality
holding if M is embeddable in some projective space.
PROPOSITION 9.16 Any two algebraically closed elds with the same transcendence
degree over J are J-isomorphic.
PROOF. Choose transcendence bases and
t
for the two elds. By assumption, there
exists a bijection
t
, which extends uniquely to an J-isomorphism J| J
t
|, and
hence to an J-isomorphism of the elds of fractions J() J(
t
). Use this isomorphism
to identify J() with J(
t
). Then the two elds in question are algebraic closures of the
same eld, and hence are isomorphic (Theorem 6.8).
2
REMARK 9.17 Any two algebraically closed elds with the same uncountable cardinality
and the same characteristic are isomorphic. The idea of the proof is as follows. Let J and
J
t
be the prime subelds of and
t
; we can identify J with J
t
. Then show that when
is uncountable, the cardinality of is the same as the cardinality of a transcendence basis
over J. Finally, apply the proposition.
REMARK 9.18 What are the automorphisms of C? There are only two continuous auto-
morphisms (cf. Exercise A-8 and solution). If we assume Zorns lemma, then it is easy to
construct many: choose any transcendence basis for C over Q, and choose any permu-
tation of ; then denes an isomorphism Q() Q() that can be extended to an
automorphism of C. Without Zorns lemma, there are only two, because the noncontinuous
automorphisms are nonmeasurable,
1
and it is known that the Zorns lemma is required to
construct nonmeasurable functions.
2
1
A fairly elementary theorem of G. Mackey says that measurable homomorphisms of Lie groups are
continuous (see Theorem B.3, p. 198 of Zimmer, Robert J., Ergodic theory and semisimple groups. Birkh auser,
1984.)
2
We show that the existence of a non-Lebesgue measurable set cannot be proved in Zermelo-Frankel set
theory (ZF) if use of the axiom of choice is disallowed... R. Solovay, Ann. of Math., 92 (1970), 156.
114 9. TRANSCENDENTAL EXTENSIONS
L uroths theorem
THEOREM 9.19 (L UROTH) Let 1 =J(X) with X transcendental over J. Every subeld
1 of 1 properly containing J is of the form 1 =J(u) for some u 1 transcendental over
J.
We rst sketch a geometric proof of L uroths theorem. The inclusion of 1 into 1
corresponds to a map from the projective line P
1
onto a complete regular curve C. Now the
Riemann-Hurwitz formula shows that C has genus 0. Since it has an J-rational point (the
image of any J-rational point of P
1
), it is isomorphic to P
1
. Therefore 1 =J(u) for some
u 1 transcendental over J.
Before giving the elementary proof, we review Gausss lemma and its consequences.
GAUSSS LEMMA
Let 1 be a unique factorization domain, and let Q be its eld of fractions, for example,
1 =JX| and Q=J(X). A polynomial (T ) =

a
i
T
i
in 1T | is said to be primitive
if its coefcients a
i
have no common factor other than units. Every polynomial in QX|
can be written =c( )
1
with c( ) Q and
1
primitive (write =a,a with a a
common denominator for the coefcients of , and then write = (b,a)
1
with b the
greatest common divisor of the coefcients of a ). The element c( ) is uniquely determined
up to a unit, and 1X| if and only if c( ) 1.
9.20 If . g 1T | are primitive, so also is g.
Let =

a
i
T
i
and g =

b
i
T
i
, and let be a prime element of 1. Because is
primitive, there exists a coefcient a
i
not divisible by let a
i
1
be the rst such coefcient.
Similarly, let b
i
2
be the rst coefcient of g not divisible by . Then the coefcient of
T
i
1
i
2
in g is not divisible by . This shows that g is primitive.
9.21 For any . g 1T |, c(g) =c( )c(g) and (g)
1
=
1
g
1
.
Let =c( )
1
and g =c(g)g
1
with
1
and g
1
primitive. Then g =c( )c(g)
1
g
1
with

1
g
1
primitive, and so c(g) =c( )c(g) and (g)
1
=
1
g
1
.
9.22 Let be a polynomial in 1T |. If factors into the product of two nonconstant
polynomials in QT |, then it factors into the product of two nonconstant polynomials in
1T |.
Suppose =gh in QT |. Then
1
=g
1
h
1
in 1T |, and so =c( )
1
=(c( ) g
1
)h
1
with c( ) g
1
and h
1
in 1T |.
9.23 Let . g 1T |. If divides g in QT | and is primitive, then it divides g in 1T |.
Let q =g with q QT |. Then c(q) =c(g) 1, and so q 1T |.
L uroths theorem 115
PROOF OF L

UROTHS THEOREM
We dene the degree deg(u) of an element u of J(X) to be the larger of the degrees of the
numerator and denominator of u when it is expressed in its simplest form.
LEMMA 9.24 Let u J(X) J. Then u is transcendental over J, X is algebraic over
J(u), and J(X): J(u)| =deg(u).
PROOF. Let u(X) =a(X),b(X) with a(X) and b(X) relatively prime polynomials. Now
a(T ) b(T )u J(u)T |, and it has X as a root, and so X is algebraic over J(u). It follows
that u is transcendental over J (otherwise X would be algebraic over J; 1.31b).
The polynomial a(T ) b(T )7 J7. T | is clearly irreducible. As u is transcendental
over J,
J7. T | .Ju. T |. 7 -u. T -T.
and so a(T ) b(T )u is irreducible in Ju. T |, and hence also in J(u)T | by Gausss lemma
(9.22). It has X as a root, and so, up to a constant, it is the minimum polynomial of X over
J(u), and its degree is deg(u), which proves the lemma.
2
EXAMPLE 9.25 We have J(X) =J(u) if and if
u =
aX b
cX J
with ac = 0 and neither aX b nor cX J a constant multiple of the other. These
conditions are equivalent to aJ bc =0.
We now prove Theorem 9.19. Let u be an element of 1 not in J. Then
J(X): 1| _J(X): J(u)| =deg(u).
and so X is algebraic over 1. Let
(T ) =T
n
a
1
T
n-1
a
n
. a
i
1.
be its minimum polynomial. As X is transcendental over J, some a
}
J, and we shall
show that, for any such a
}
, 1 =J(a
}
).
Let J(X) JX| be the polynomial of least degree such that J(X)a
i
(X) JX| for
all i , and let

1
(X. T ) =J(T ) =JT
n
Ja
1
T
n-1
Ja
n
JX. T |.
Then
1
is primitive as a polynomial in T , i.e., gcd(J. Ja
1
. . . . . Ja
n
) = 1 in JX|. The
degree m of
1
in X is the largest degree of one of the polynomials Ja
1
. Ja
2
. . . ., say
m = deg(Ja
i
). Write a
i
= b,c with b. c relatively prime polynomials in JX|. Now
b(T ) c(T )a
i
(X) is a polynomial in 1T | having X as a root, and so it is divisible by ,
say
(T ) q(T ) =b(T ) c(T ) a
i
(X). q(T ) 1T |.
On multiplying through by c(X), we nd that
c(X) (T ) q(T ) =c(X) b(T ) c(T ) b(X).
116 9. TRANSCENDENTAL EXTENSIONS
Therefore
1
divides c(X) b(T ) c(T ) b(X) in J(X)T |. As
1
is primitive, it divides
c(X) b(T ) c(T ) b(X) in JX. T | (by 9.23), i.e., there exists a polynomial h(X. T )
JX. T | such that

1
(X. T ) h(X. T ) =c(X) b(T ) c(T ) b(X).
The polynomial c(X) b(T ) c(T ) b(X) has degree at most m in X, and m is the degree of

1
(X. T ) in X. Therefore, c(X) b(T )c(T ) b(X) has degree exactly min X, and h(X. T )
is not divisible by a nonconstant polynomial in JX|. By symmetry, c(X) b(T ) c(T )
b(X) has degree m in T , and h(X. T ) not divisible by a nonconstant polynomial in JT |.
Hence h(X. T ) J

, and so
1
(X. T ) is a constant multiple of c(X) b(T ) c(T ) b(X).
On comparing degrees in T , we see see that n =m. Thus
J(X): J(a
i
)| =deg(a
i
) _deg(Ja
i
) =m=n =J(X): 1| _J(X): J(a
i
)|.
and so 1 =Ja
i
|. Finally, if a
}
J, then
J(X): 1| _J(X): J(a
}
)| =deg(a
}
) _deg(Ja
}
) _deg(Ja
i
) =m=J(X): 1|.
and so 1 =J(a
}
) as claimed.
REMARK 9.26 L uroths theorem fails when there is more than one variable see Zariskis
example (footnote to Remark 5.5) and Swans example (Remark 5.41). However, the
following is true: if J(X. Y ): 1| <oand J is algebraically closed of characteristic zero,
then 1 is a pure transcendental extension of J (Theorem of Zariski, 1958).
NOTES L uroth proved his theorem over C in 1876. For general elds, it was proved by Steinitz in
1910, by the above argument.
Separating transcendence bases
Let 1 J be elds with 1 nitely generated over J. A subset {.
1
. . . . . .
d
] of 1 is a
separating transcendence basis for 1,J if it is algebraically independent over J and 1 is
a nite separable extension of J(.
1
. . . . . .
d
).
THEOREM 9.27 If J is perfect, then every nitely generated extension 1 of J admits a
separating transcendence basis over J.
PROOF. If J has characteristic zero, then every transcendence basis is separating, and so
the statement becomes that of (9.10). Thus, we may assume J has characteristic =0.
Because J is perfect, every polynomial in X
]
1
. . . . . X
]
n
with coefcients in J is a th power
in JX
1
. . . . . X
n
|:

a
i
1
i
n
X
i
1
]
1
. . . X
i
n
]
n
=
_

a
1
p
i
1
i
n
X
i
1
1
. . . X
i
n
n
_
]
.
Let 1 =J(.
1
. . . . . .
n
), and assume n >J 1 where J is the transcendence degree of
1 over J. After renumbering, we may suppose that .
1
. . . . . .
d
are algebraically independent
(9.9). Then (.
1
. . . . . .
d1
) =0 for some nonzero irreducible polynomial (X
1
. . . . . X
d1
)
with coefcients in J. Not all d,dX
i
are zero, for otherwise would be a polynomial in
X
]
1
. . . . . X
]
d1
, which implies that it is a th power. After renumbering, we may suppose that
Transcendental Galois theory 117
d,dX
d1
=0. Then J(.
1
. . . . . .
d1
. .
d2
) is algebraic over J(.
1
. . . . . .
d
) and .
d1
is
separable over J(.
1
. . . . . .
d
), and so, by the primitive element theorem (5.1), there is an
element , such that J(.
1
. . . . . .
d2
) = J(.
1
. . . . . .
d
. ,). Thus 1 is generated by n1
elements (as a eld containing J). After repeating the process, possibly several times, we
will have 1 =J(:
1
. . . . . :
d1
) with :
d1
separable over J(:
1
. . . . . :
d
).
2
ASIDE 9.28 In fact, we showed that 1 admits a separating transcendence basis with J 1 elements
where J is the transcendence degree. This has the following geometric interpretation: every irre-
ducible algebraic variety of dimension J over a perfect eld J is birationally equivalent with a
hypersurface H in A
d1
for which the projection (a
1
. . . . . a
d1
) (a
1
. . . . . a
d
) realizes J(H) as a
nite separable extension of J(A
d
) (see my notes on Algebraic Geometry).
Transcendental Galois theory
THEOREM 9.29 Let be an algebraically closed eld and let J be a perfect subeld of .
If is xed by all J-automorphisms of , then J, i.e.,
Aut({T)
=J.
PROOF. Let J. If is algebraic over J, then there is an J-homomorphism
J| sending to a conjugate of in different from . This homomorphism
extends to a homomorphism from the algebraic closure J
al
of J in to (by 6.8). Now
choose a transcendence basis for over J
al
. We can extend our homomorphism to a
homomorphism J() by mapping each element of to itself. Finally, we can extend
this homomorphism to a homomorphism from the algebraic closure of J() to . The
J-homomorphism we obtain is automatically an isomorphism (cf. 6.8).
If is transcendental over J, then it is part of a transcendence basis for over
J (see 9.13). If has at least two elements, then there exists an automorphism o of
such that o() =. Now o denes an J-homomorphism J() , which extends to an
isomorphism as before. If ={], then we let J() be the J-homomorphism
sending to 1. Again, this extends to an isomorphism .
2
REMARK 9.30 Theorem 9.29 holds with only separably closed. To see this, let
al
be
an algebraic closure of . Then every automorphism o of ,J extends uniquely to an
automorphism o of
al
,J: let
al
and let
]
n
; then o() is the unique root of
X
]
n
o(
]
n
) in
al
. Thus, if is xed by all J-automorphisms of , then it is xed
by all J-automorphisms of
al
, and so it lies in J.
Let J be elds and let G =Aut(,J). For any nite subset S of , let
G(S) ={o G [ os =s for all s S].
Then, as in 7, the subgroups G(S) of G form a neighbourhood base for a unique topology
on G, which we again call the Krull topology. The same argument as in 7 shows that this
topology is Hausdorff (but it is not necessarily compact).
THEOREM 9.31 Let J be elds such that
G
=J, G =Aut(,J).
(a) For every nite extension 1 of J in ,
Aut({T)
=1.
(b) The maps
H
1
. M Aut(,M) (8)
118 9. TRANSCENDENTAL EXTENSIONS
are inverse bijections between the set of compact subgroups of G and the set of intermediate
elds over which is Galois (possibly innite):
{compact subgroups of G] -{elds M such that J M
Galois
].
(c) If there exists an M nitely generated over J such that is Galois over M, then G
is locally compact, and under (8):
{open compact subgroups of G]
1:1
-{elds M such that J
nitely generated
M
Galois
].
(d) Let H be a subgroup of G, and let M =
1
. Then the algebraic closure M
1
of M
is Galois over M. If moreover H =Aut(,M), then Aut(,M
1
) is a normal subgroup
of H, and o o[M
1
maps H,Aut(,M
1
) isomorphically onto a dense subgroup of
Aut(M
1
,M).
PROOF. See 6.3 of Shimura, Goro., Introduction to the arithmetic theory of automorphic
functions. Princeton, 1971.
2
Exercises
9-1 Find the centralizer of complex conjugation in Aut(C,Q).
APPENDIX A
Review Exercises
A-1 Let be a prime number, and let m and n be positive integers.
(a) Give necessary and sufcient conditions on m and n for F
]
n to have a subeld
isomorphic with F
]
m. Prove your answer.
(b) If there is such a subeld, how many subelds isomorphic with F
]
m are there, and
why?
A-2 Show that the Galois group of the splitting eld J of X
3
7 over Q is isomorphic
to S
3
, and exhibit the elds between Q and J. Which of the elds between Q and J are
normal over Q?
A-3 Prove that the two elds Q
_
7| and Q
_
11| are not isomorphic.
A-4 (a) Prove that the multiplicative group of all nonzero elements in a nite eld is
cyclic.
(b) Construct explicitly a eld of order 9, and exhibit a generator for its multiplicative
group.
A-5 Let X be transcendental over a eld J, and let 1 be a subeld of J(X) properly
containing J. Prove that X is algebraic over 1.
A-6 Prove as directly as you can that if is a primitive th root of 1, prime, then the
Galois group of Q| over Q is cyclic of order 1.
A-7 Let G be the Galois group of the polynomial X
5
2 over Q.
(a) Determine the order of G.
(b) Determine whether G is abelian.
(c) Determine whether G is solvable.
A-8 (a) Show that every eld homomorphism from R to R is bijective.
(b) Prove that C is isomorphic to innitely many different subelds of itself.
A-9 Let J be a eld with 16 elements. How many roots in J does each of the following
polynomials have? X
3
1; X
4
1; X
15
1; X
1T
1.
119
120 A. REVIEW EXERCISES
A-10 Find the degree of a splitting eld of the polynomial (X
3
5)(X
3
7) over Q.
A-11 Find the Galois group of the polynomial X
6
5 over each of the elds Q and R.
A-12 The coefcients of a polynomial (X) are algebraic over a eld J. Show that (X)
divides some nonzero polynomial g(X) with coefcients in J.
A-13 Let (X) be a polynomial in JX| of degree n, and let 1 be a splitting eld of .
Show that 1: J| divides n.
A-14 Find a primitive element for the eld Q
_
3.
_
7| over Q, i.e., an element such that
Q
_
3.
_
7| =Q|.
A-15 Let G be the Galois group of (X
4
2)(X
3
5) over Q.
(a) Give a set of generators for G, as well as a set of dening relations.
(b) What is the structure of G as an abstract group (is it cyclic, dihedral, alternating,
symmetric, etc.)?
A-16 Let J be a nite eld of characteristic =2. Prove that X
2
=1 has a solution in J
if and only if [J[ 1 mod 4.
A-17 Let 1 be the splitting eld over Q of (X
2
2)(X
2
5)(X
2
7). Find an element
in 1 such that 1 =Q|. (You must prove that 1 =Q|.)
A-18 Let 1 be a Galois extension of J with Galois group S
n
, n > 1 not prime. Let H
1
be
the subgroup of S
n
of elements xing 1, and let H
2
be the subgroup generated by the cycle
(123. . . n). Let 1
i
=1
1
i
, i =1. 2. Find the degrees of 1
1
, 1
2
, 1
1
1
2
, and 1
1
1
2
over
J. Show that there exists a eld M such that J M 1
2
, M =J, M =1
2
, but that no
such eld exists for 1
1
.
A-19 Let be a primitive 12th root of 1 over Q. How many elds are there strictly between
Q
3
| and Q|.
A-20 For the polynomial X
3
3, nd explicitly its splitting eld over Q and elements that
generate its Galois group.
A-21 Let 1 =Q|,
5
=1, =1. Show that i 1, and that if 1 =1i |, then 1 is a
norm from 1 to 1. Here i =
_
1.
A-22 Let 1 be an extension eld of J, and let be an algebraic closure of 1. Let
o
1
. . . . . o
n
be distinct J-isomorphisms 1 .
(a) Show that o
1
. . . . . o
n
are linearly dependent over .
(b) Show that 1: J| _m.
(c) Let J have characteristic > 0, and let 1 be a subeld of containing 1 and
such that a
]
1 for all a 1. Show that each o
i
has a unique extension to a
homomorphism o
t
i
: 1 .
121
A-23 Identify the Galois group of the splitting eld J of X
4
3 over Q. Determine the
number of quadratic subelds.
A-24 Let J be a subeld of a nite eld 1. Prove that the trace map T =Tr
T{T
and the
norm map N =Nm
T{T
of 1 over J both map 1 onto J. (You may quote basic properties
of nite elds and the trace and norm.)
A-25 Prove or disprove by counterexample.
(a) If 1,J is an extension of elds of degree 2, then there is an automorphism o of 1
such that J is the xed eld of o.
(b) The same as (a) except that 1 is also given to be nite.
A-26 A nite Galois extension 1 of a eld 1 has degree 8100. Show that there is a eld
J with 1 J 1 such that J: 1| =100.
A-27 An algebraic extension 1 of a eld 1 of characteristic 0 is generated by an element
0 that is a root of both of the polynomials X
3
1 and X
4
X
2
1. Given that 1=1, nd
the minimum polynomial of 0.
A-28 Let J,Q be a Galois extension of degree 3
n
, n _ 1. Prove that there is a chain of
elds
Q=J
0
J
1
J
n
=J
such that for every i , 0 _i _n1, J
i1
: J
i
| =3.
A-29 Let 1 be the splitting eld over Q of an equation of degree 5 with distinct roots.
Suppose that 1 has an automorphism that xes three of these roots while interchanging the
other two and also an automorphism =1 of order 5.
(a) Prove that the group of automorphisms of 1 is the symmetric group on 5 elements.
(b) How many proper subelds of 1 are normal extensions of Q? For each such eld J,
what is J: Q|?
A-30 If 1,1 is a separable algebraic eld extension of nite degree J, show that the
number of elds between 1 and 1 is at most 2
d
.
A-31 Let 1 be the splitting eld over Q of X
5
1. Describe the Galois group Gal(1,Q)
of 1 over Q, and show that 1 has exactly one subeld of degree 2 over Q, namely, Q
4
|,
=1 a root of X
5
1. Find the minimum polynomial of
4
over Q. Find Gal(1,Q)
when 1 is the splitting eld over Q of
(a) (X
2
5)(X
5
1);
(b) (X
2
3)(X
5
1).
A-32 Let
1
and
2
be algebraically closed elds of transcendence degree 5 over Q, and
let :
1

2
be a homomorphism (in particular, (1) =1). Show that is a bijection.
(State carefully all theorems you use.)
A-33 Find the group of Q-automorphisms of the eld k =Q
_
3.
_
2|.
122 A. REVIEW EXERCISES
A-34 Prove that the polynomial (X) =X
3
5 is irreducible over the eld Q
_
7|. If
1 is the splitting eld of (X) over Q
_
7|, prove that the Galois group of 1,Q
_
7| is
isomorphic to S
3
. Prove that there must exist a subeld 1 of 1 such that the Galois group
of 1,1 is cyclic of order 3.
A-35 Identify the Galois group G of the polynomial (X) =X
5
6X
4
3 over J, when
(a) J =Q and when (b) J =F
2
. In each case, if 1 is the splitting eld of (X) over J,
determine how many elds 1 there are such that 1 1 J with 1: J| =2.
A-36 Let 1 be a eld of characteristic , say with
n
elements, and let 0 be the au-
tomorphism of 1 that maps every element to its th power. Show that there exists an
automorphism of 1 such that 0
2
=1 if and only if n is odd.
A-37 Describe the splitting eld and Galois group, over Q, of the polynomial X
5
9.
A-38 Suppose that 1 is a Galois eld extension of a eld J such that 1: J| =5
3
(43)
2
.
Prove that there exist elds 1
1
and 1
2
lying strictly between J and 1 with the following
properties: (i) each 1
i
is a Galois extension of J; (ii) 1
1
1
2
=J; and (iii) 1
1
1
2
=1.
A-39 Let J =F
]
for some prime . Let m be a positive integer not divisible by , and let
1 be the splitting eld of X
n
1. Find 1: J| and prove that your answer is correct.
A-40 Let J be a eld of 81 elements. For each of the following polynomials g(X),
determine the number of roots of g(X) that lie in J: X
S0
1, X
S1
1, X
SS
1.
A-41 Describe the Galois group of the polynomial X
6
7 over Q.
A-42 Let 1 be a eld of characteristic >0 and let J =1(u. ) be a eld extension of
degree
2
such that u
]
1 and
]
1. Prove that 1 is not nite, that J is not a simple
extension of 1, and that there exist innitely many intermediate elds J 1 1.
A-43 Find the splitting eld and Galois group of the polynomial X
3
5 over the eld
Q
_
2|.
A-44 For every prime , nd the Galois group over Q of the polynomial X
5
5
4
X .
A-45 Factorize X
4
1 over each of the nite elds (a) F
5
; (b) F
25
; and (c) F
125
. Find its
splitting eld in each case.
A-46 Let Q| be a eld of nite degree over Q. Assume that there is a q Q, q = 0,
such that [j()[ = q for all homomorphisms j: Q| C. Show that the set of roots of
the minimum polynomial of is the same as that of q
2
,. Deduce that there exists an
automorphism o of Q| such that
(a) o
2
=1 and
(b) j(o;) =j(;) for all ; Q| and j: Q| C.
123
A-47 Let J be a eld of characteristic zero, and let be a prime number. Suppose that
J has the property that all irreducible polynomials (X) JX| have degree a power
of (1 =
0
is allowed). Show that every equation g(X) =0, g JX|, is solvable by
extracting radicals.
A-48 Let 1 =Q
_
5.
_
7| and let 1 be the splitting eld over Q of (X) =X
3
10.
(a) Determine the Galois groups of 1 and 1 over Q.
(b) Decide whether 1 contains a root of .
(c) Determine the degree of the eld 11 over Q.
[Assume all elds are subelds of C.]
A-49 Find the splitting eld (over F
]
) of X
]
r
X F
]
X|, and deduce that X
]
r
X
has an irreducible factor F
]
X| of degree r. Let g(X) ZX| be a monic polynomial
that becomes equal to (X) when its coefcients are read modulo . Show that g(X) is
irreducible in QX|.
A-50 Let 1 be the splitting eld of X
3
51 over Q. List all the subelds of 1, and nd
an element ; of 1 such that 1 =Q;|.
A-51 Let k =F
1024
be the eld with 1024 elements, and let 1 be an extension of k of
degree 2. Prove that there is a unique automorphism o of 1 of order 2 which leaves k
elementwise xed and determine the number of elements of 1

such that o(.) =.

-1
.
A-52 Let J and 1 be nite elds of the same characteristic. Prove or disprove these
statements:
(a) There is a ring homomorphism of J into 1 if and only if [1[ is a power of [J[.
(b) There is an injective group homomorphism of the multiplicative group of J into the
multiplicative group of 1 if and only if [1[ is a power of [J[.
A-53 Let 1,1 be an algebraic extension of elds. Prove that 1 is algebraically closed if
every polynomial over 1 factors completely over 1.
A-54 Let 1 be a eld, and let M =1(X), X an indeterminate. Let 1 be an intermediate
eld different from 1. Prove that M is nite-dimensional over 1.
A-55 Let 0
1
. 0
2
. 0
3
be the roots of the polynomial (X) =X
3
X
2
9X 1.
(a) Show that the 0
i
are real, nonrational, and distinct.
(b) Explain why the Galois group of (X) over Q must be either
3
or S
3
. Without
carrying it out, give a brief description of a method for deciding which it is.
(c) Show that the rows of the matrix
_
_
_
_
3 9 9 9
3 0
1
0
2
0
3
3 0
2
0
3
0
1
3 0
3
0
1
0
2
_
_
_
_
are pairwise orthogonal; compute their lengths, and compute the determinant of the
matrix.
124 A. REVIEW EXERCISES
A-56 Let 1,1 be a Galois extension of degree
2
q where and q are primes, q < and
q not dividing
2
1. Prove that:
(a) there exist intermediate elds 1 and M such that 1: 1| =
2
and M: 1| =q;
(b) such elds 1 and M must be Galois over 1; and
(c) the Galois group of 1,1 must be abelian.
A-57 Let be a primitive 7th root of 1 (in C).
(a) Prove that 1X X
2
X
3
X
4
X
5
X
6
is the minimum polynomial of over
Q.
(b) Find the minimum polynomial of
1
(
over Q.
A-58 Find the degree over Q of the Galois closure 1 of Q2
1
4
| and determine the isomor-
phism class of Gal(1,Q).
A-59 Let . q be distinct positive prime numbers, and consider the extension 1=Q
_
.
_
q|
Q.
(a) Prove that the Galois group is isomorphic to C
2
C
2
.
(b) Prove that every subeld of 1 of degree 2 over Q is of the form Q
_
m| where
m {. q. q].
(c) Show that there is an element ; 1 such that 1 =Q;|.
APPENDIX B
Two-hour Examination
1. (a) Let o be an automorphism of a eld 1. If o
4
=1 and
o() o
3
() = o
2
() all 1.
show that o
2
=1.
(b) Let be a prime number and let a. b be rational numbers such that a
2
b
2
=1. Show
that there exist rational numbers c. J such that a =
c
2
-]d
2
c
2
]d
2
and b =
2cd
c
2
]d
2
. !!Check!!
2. Let (X) be an irreducible polynomial of degree 4 in QX|, and let g(X) be the resolvent
cubic of . What is the relation between the Galois group of and that of g? Find the
Galois group of if
(a) g(X) =X
3
3X 1;
(b) g(X) =X
3
3X 1.
3. (a) How many monic irreducible factors does X
255
1 F
2
X| have, and what are their
degrees.
(b) How many monic irreducible factors does X
255
1 QX| have, and what are their
degrees?
4. Let 1 be the splitting eld of (X
5
3)(X
5
7) QX|. What is the degree of 1 over
Q? How many proper subelds of 1 are there that are not contained in the splitting elds of
both X
5
3 and X
5
7?
[You may assume that 7 is not a 5th power in the splitting eld of X
5
3.]
5. Consider an extension J of elds. Dene a to be J-constructible if it is
contained in a eld of the form
J
_
a
1
. . . . .
_
a
n
|. a
i
J
_
a
1
. . . . .
_
a
i-1
|.
Assume is a nite Galois extension of J and construct a eld 1, J 1 , such that
every a is 1-constructible and 1 is minimal with this property.
6. Let be an extension eld of a eld J. Show that every J-homomorphism is
an isomorphism provided:
(a) is algebraically closed, and
(b) has nite transcendence degree over J.
125
126 B. TWO-HOUR EXAMINATION
Can either of the conditions (i) or (ii) be dropped? (Either prove, or give a counterexam-
ple.)
You should prove all answers. You may use results proved in class or in the notes, but you
should indicate clearly what you are using.
Possibly useful facts: The discriminant of X
3
aXb is 4a
3
27b
2
and 2
S
1 =255 =
3517.
APPENDIX C
Solutions to the Exercises
These solutions fall somewhere between hints and complete solutions. Students were expected
to write out complete solutions.
1-1. Similar to Example 1.28.
1-2. Verify that 3 is not a square in Q
_
2|, and so Q
_
2.
_
3|: Q| =4.
1-3. (a) Apply the division algorithm, to get (X) = q(X)(X a) r(X) with r(X)
constant, and put X =a to nd r =(a).
(c) Use that factorization in JX| is unique (or use induction on the degree of ).
(d) If G had two cyclic factors C and C
t
whose orders were divisible by a prime , then G
would have (at least)
2
elements of order dividing . This doesnt happen, and it follows
that G is cyclic.
(e) The elements of order m in J

are the roots of the polynomial X

n
1, and so there are
at most m of them. Hence every nite subgroup G of J

satises the condition in (d).

1-4. Note that it sufces to construct =cos
2t
T
, and that Q|: Q| =
T-1
2
=3, and so its
minimum polynomial has degree 3 (see Example 3.21). There is a standard method (once
taught in high schools) for solving cubics using the equation
cos30 =4cos
3
0 3cos0.
By completing the cube, reduce the cubic to the form X
3
X q. Then construct a
square root a of
4]
3
, so that a
2
=
4]
3
. Let 30 be the angle such that cos30 =
4q
o
3
, and use
the angle trisector to construct cos0. From the displayed equation, we nd that =acos0
is a root of X
3
X q.
1-5. Let
1
be an irreducible factor of in 1X|, and let (1. ) be a stem eld for
1
over
1. Then m[1: J| because 1 1 (1.20). But () =0, and so (J|. ) is a stem eld for
over J, which implies that J|: J| =n. Now n[1: J| because 1 J|. We deduce
that 1: J| =mn and 1: 1| =n. But 1: 1| =deg(
1
), and so
1
= .
2-1. (a) is obvious, as is the only if in (b). For the if note that for any a S(1), a J
2
,
1 ~JX|,(X
2
a).
(c) Take 1
i
= Q
_

i
| with
i
the i th prime. Check that
i
is the only prime that
becomes a square in 1
i
. For this use that (ab
_
)
2
Q == 2ab =0.
(d) Every eld of characteristic contains (an isomorphic copy of) F
]
, and so we are
looking at the quadratic extensions of F
]
. The homomorphism a a
2
: F

]
F

]
has kernel
{1], and so its image has index 2 in F

]
. Thus the only possibility for S(1) is F

]
, and
127
128 C. SOLUTIONS TO THE EXERCISES
so there is at most one 1 (up to F
]
-isomorphism). To get one, take 1 =JX|,(X
2
a),
a F
2
]
.
2-2. (a) If is a root of (X) =X
]
X a (in some splitting eld), then the remaining
roots are 1. . . . . 1, which obviously lie in whichever eld contains . Moreover,
they are distinct. Suppose that, in JX|,
(X) =(X
i
a
1
X
i-1
a
i
)(X
]-i
). 0 < r <.
Then a
1
is a sum of r of the roots of , a
1
=r J some J Z 1
T
, and it follows
that J.
(b) As 0 and 1 are not roots of X
]
X 1 in F
]
it cant have distinct roots in F
]
, and
so (a) implies that X
]
X1 is irreducible in F
]
X| and hence also in ZX| and QX| (see
1.18, 1.13).
2-3. Let be the real 5th root of 2. Eisensteins criterion shows that X
5
2 is irre-
ducible in QX|, and so Q
5
_
2| has degree 5 over Q. The remaining roots of X
5
2 are
.
2
.
3
.
4
, where is a primitive 5th root of 1. It follows that the subeld of C
generated by the roots of X
5
2 is Q. |. The degree of Q. | is 20, since it must be
divisible by Q|: Q| =4 and Q|: Q| =5.
2-4. Its F
]
because X
]
m
1 =(X 1)
]
m
. (Perhaps I meant X
]
m
X that would have
been more interesting.)
2-5. If (X) =

(X
i
)
n
i
,
i
=
}
, then

t
(X) =

m
i
(X)
X
i
and so J(X) =

n
i
>1
(X
i
)
n
i
-1
. Therefore g(X) =

(X
i
).
2-6. From (2.12) we know that either is separable or (X) =
1
(X
]
) for some polynomial

1
. Clearly
1
is also irreducible. If
1
is not separable, it can be written
1
(X) =
2
(X
]
).
Continue in the way until you arrive at a separable polynomial. For the nal statement, note
that g(X) =

(X a
i
), a
i
=a
}
, and so (X) =g(X
]
e
) =

(X
i
)
]
e
with
]
e
i
=a
i
.
3-1. Let o and t be automorphisms of J(X) given by o(X) = X and t(X) = 1 X.
Then o and t x X
2
and X
2
X respectively, and so ot xes 1
def
=J(X) J(X
2
X).
But tX = 1 X, and so (ot)
n
(X) = mX. Thus Aut(J(X),1) is innite, which
implies that J(X): 1| is innite (otherwise J(X) =1
1
. . . . .
n
|; an 1-automorphism of
J(X) is determined by its values on the
i
, and its value on
i
is a root of the minimum
polynomial of
i
). If 1 contains a polynomial (X) of degree m > 0, then J(X): 1| _
J(X): J((X))| =m contradiction.
3-2. Since 1
]-1
= 0, we have = 1. If i H, then iH = H and
i(GH) =GH, and so and are xed by H. If GH, then H =GH and
(GH) =H, and so = and =. Hence Q, and and are the roots of
X
2
X . Note that
=

i,}

i}
. i H. GH.
How many times do we have i =0? If i =0, then 1 =i
-1
, which is a nonsquare;
conversely, if 1 is a nonsquare, take i =1 and =1 to get i =0. Hence
i =0 some i H. GH 1 is a square mod 1 mod 4.
129
If we do have a solution to i = 0, we get all solutions by multiplying it through
by the
]-1
2
squares. So in the sum for we see 1 a total of
]-1
2
times when 3
mod 4 and not at all if 1 mod 4. In either case, the remaining terms add to a rational
number, which implies that each power of occurs the same number of times. Thus for
1 mod 4, = (
]-1
2
)
2
,( 1) =
]-1
4
; the polynomial satised by and is
X
2
X
]-1
4
, whose roots are (1
_
11),2; the xed eld of H is Q
_
|. For
1 mod 4, =
]-1
2
(1)
_
(
]-1
2
)
2

]-1
2
_
,( 1) =
]-1
2

]-3
4
=
]1
4
; the
polynomial is X
2
X
]-1
4
, with roots (1
_
11),2; the xed eld of H is
Q
_
|.
3-3. (a) It is easy to see that M is Galois over Q with Galois group (o. t):
_
o
_
2 =
_
2
o
_
3 =
_
3
_
t
_
2 =
_
2
t
_
3 =
_
3
.
(b) We have
o
2

2
=
2
_
2
2
_
2
=
(2
_
2)
2
42
=
_
2
_
2
_
2
_
2
=(
_
21)
2
.
i.e., o
2
=((
_
21))
2
. Thus, if M, then o =(
_
21), and
o
2
=(
_
21)(
_
21) =:
as o
2
= =0, this is impossible. Hence M, and so 1: Q| =8.
Extend o to an automorphism (also denoted o) of 1. Again o =(
_
21) and o
2
=
, and so o
2
=1. Now o
4
=, o
4
[M =1, and so we can conclude that o has order 4.
After possibly replacing o with its inverse, we may suppose that o =(
_
21).
Repeat the above argument with t:
r
2

2
=
3-
_
3
3
_
3
=
_
3-
_
3
_
6
_
2
, and so we can extend t to an
automorphism of 1 (also denoted t) with t =
3-
_
3
_
6
. The order of t is 4.
Finally compute that
ot =
3
_
3

_
6
(
_
21): to =(
_
21)
3
_
3
_
6
.
Hence ot =to, and Gal(1,Q) has two noncommuting elements of order 4. Since it has
order 8, it must be the quaternion group.
4-1. The splitting eld is the smallest eld containing all mth roots of 1. Hence it is F
]
n
where n is the smallest positive integer such that m
0
[
n
1, m=m
0

i
, where is prime
and does not divide m
0
.
4-2. We have X
4
2X
3
8X3 =(X
3
X
2
3X1)(X3), and g(X) =X
3
X
2

3X 1 is irreducible over Q (use 1.11), and so its Galois group is either

3
or S
3
. Either
check that its discriminant is not a square or, more simply, show by examining its graph that
g(X) has only one real root, and hence its Galois group contains a transposition (cf. the
proof of 4.16).
4-3. Eisensteins criterion shows that X
S
2 is irreducible over Q, and so Q|: Q| =8
where is a positive 8th root of 2. As usual for polynomials of this type, the splitting eld
is Q. | where is any primitive 8th root of 1. For example, can be taken to be
1i
_
2
,
130 C. SOLUTIONS TO THE EXERCISES
which lies in Q. i |. It follows that the splitting eld is Q. i |. Clearly Q. i | =Q|,
because Q|, unlike i , is contained in R, and so Q. i |: Q|| =2. Therefore the degree is
28 =16.
4-4. Find an extension 1,J with Galois group S
4
, and let 1 be the xed eld of S
3
S
4
.
There is no subgroup strictly between S
n
and S
n-1
, because such a subgroup would be
transitive and contain an (n1)-cycle and a transposition, and so would equal S
n
. We can
take 1 =1
S
3
. More specically, we can take 1 to be the splitting eld of X
4
X 2 over
Q and 1 to be the subeld generated by a root of the polynomial (see 3.26).
4-5. Type: Factor(X
343
X) mod 7; and discard the 7 factors of degree 1.
4-6. Type galois(X
6
2X
5
3X
4
4X
3
5X
2
6X 7);. It is the group PGL
2
(F
5
)
(group of invertible 2 2 matrices over F
5
modulo scalar matrices) which has order 120.
Alternatively, note that there are the following factorizations: mod 3, irreducible; mod 5 (deg
3)(deg 3); mod 13 (deg 1)(deg 5); mod 19, (deg 1)
2
(deg 4); mod 61 (deg 1)
2
(deg 2)
2
; mod
79, (deg 2)
3
. Thus the Galois group has elements of type:
6. 33. 15. 114. 1122. 222.
No element of type 2, 3, 32, or 42 turns up by factoring modulo any of the rst 400
primes (or, so I have been told). This suggests it is the group T14 in the tables in Butler and
McKay, which is indeed PGL
2
(F
5
).
4-7. == : Condition (a) implies that G
(
contains a 5-cycle, condition (b) implies that
G
(

5
, and condition (c) excludes
5
. That leaves D
5
and C
5
as the only possibilities
(see, for example, Jacobson, Basic Algebra I, p305, Ex 6). The derivative of is 5X
4
a,
which has at most 2 real zeros, and so (from its graph) we see that can have at most 3 real
zeros. Thus complex conjugation acts as an element of order 2 on the splitting eld of ,
and this shows that we must have G
(
=D
5
.
==: Regard D
5
as a subgroup of S
5
by letting it act on the vertices of a regular pentagon
all subgroups of S
5
isomorphic to D
5
look like this one. If G
(
=D
5
, then (a) holds because
D
5
is transitive, (b) holds because D
5

5
, and (c) holds because D
5
is solvable.
4-8. Omitted.
4-9. Let a
1
. a
2
be conjugate nonreal roots, and let a
3
be a real root. Complex conjugation
denes an element o of the Galois group of switching a
1
and a
2
and xing a
3
. On the
other hand, because is irreducible, its Galois group acts transitively on its roots, and so
there is a t such that t(a
3
) =a
1
. Now
a
3
r
a
1
c
a
2
a
3
c
a
3
r
a
1.
This statement is false for reducible polynomials consider for example (X) =(X
2

1)(X 1).
5-1. For a =1, this is the polynomial
5
(X), whose Galois group is cyclic of order 4.
For a =0, it is X(X
3
X
2
X 1) =X(X 1)(X
2
1), whose Galois group is cyclic
of order 2.
For a =4, it is (X 1)(X
3
2X
2
3X 4). The cubic does not have 1. 2. or 4
as roots, and so it is irreducible in QX|. Hence its Galois group is S
3
or
3
. But looking
modulo 2, we see it contains a 2-cycle, so it must be S
3
.
131
For any a, the resolvent cubic is
g(X) =X
3
X
2
(14a)X 3a1.
Take a = 1. Then = X
4
X
3
X
2
X 1 is irreducible modulo 2, and so it is
irreducible in QX|. We have g = X
3
X
2
5X 4, which is irreducible. Moreover
g
t
= 3X
2
2X 5 = 3(X
1
3
)
2
4
2
3
> 0 always, and so g has exactly one real root.
Hence the Galois group of g is S
3
, and therefore the Galois group of is S
4
. [In fact, 4 is
the maximum number of integers giving distinct Galois groups: checking mod 2, we see
there is a 2-cycle or a 4-cycle, and so 1.
3
.
4
. V
4
are not possible. For D
S
, a cant be an
integer.]
5-2. We have Nm(a i b) =a
2
b
2
. Hence a
2
b
2
=1 if and only a i b =
xit
x-it
for
some s. t Q (Hilberts Theorem 90). The rest is easy.
5-3. The degree Q
n
|: Q| =(n),
n
a primitive nth root of 1, and (n) oas n o.
9-1. If some element centralizes the complex conjugation, then it must preserve the real
numbers as a set. Now, since any automorphism of the real numbers preserves the set of
squares, it must preserve the order; and hence be continuous. Since Q is xed, this implies
that the real numbers are xed pointwise. It follows that any element which centralized the
complex conjugation must be the identity or the complex conjugation itself. See mo121083,
Andreas Thom.
A-1. (a) Need that m[n, because
n =F
]
n: F
]
| =F
]
n: F
]
m| F
]
m: F
]
| =F
]
n: F
]
m| m.
Use Galois theory to show there exists one, for example. (b) Only one; it consists of all the
solutions of X
]
m
X =0.
A-2. The polynomial is irreducible by Eisensteins criterion. The polynomial has only one
real root, and therefore complex conjugation is a transposition in G
(
. This proves that
G
(
~S
3
. The discriminant is 1323 =3
3
7
2
. Only the subeld Q
_
3| is normal over
Q. The subelds Q
3
_
7|, Q
3
_
7| Q
2
3
_
7| are not normal over Q. [The discriminant of
X
3
a is 27a
2
=3(3a)
2
.]
A-3. The prime 7 becomes a square in the rst eld, but 11 does not: (a b
_
7)
2
=
a
2
7b
2
2ab
_
7, which lies in Qonly if ab =0. Hence the rational numbers that become
squares in Q
_
7| are those that are already squares or lie in 7Q
2
.
A-4.(a) See Exercise 3.
(b) Let J =F
3
X|,(X
2
1). Modulo 3
X
S
1 =(X 1)(X 1)(X
2
1)(X
2
X 2)(X
2
2X 2).
Take to be a root of X
2
X 2.
A-5. Since 1 =J, 1 contains an element
(

with the degree of or g >0. Now

(T )
(X)
g(X)
g(T )
is a nonzero polynomial having X as a root.
A-6. Use Eisenstein to show that X
]-1
1 is irreducible, etc. Done in class.
132 C. SOLUTIONS TO THE EXERCISES
A-7. The splitting eld is Q. | where
5
=1 and
5
=2. It is generated by o =(12345)
and t =(2354), where o = and t =
2
. The group has order 20. It is not abelian
(because Q| is not Galois over Q), but it is solvable (its order is <60).
A-8. (a) A homomorphism : RR acts as the identity map on Z, hence on Q, and it maps
positive real numbers to positive real numbers, and therefore preserves the order. Hence, for
each real number a,
{r Q[ a <r] ={r Q[ (a) <r].
which implies that (a) =a.
(b) Choose a transcendence basis for Cover Q. Because it is innite, there is a bijection
:
t
from onto a proper subset. Extend to an isomorphism Q() Q(
t
), and
then extend it to an isomorphism C C
t
where C
t
is the algebraic closure of Q(
t
) in C.
A-9. The group J

is cyclic of order 15. It has 3 elements of order dividing 3, 1 element of

order dividing 4, 15 elements of order dividing 15, and 1 element of order dividing 17.
A-10. If 1
1
and 1
2
are Galois extensions of J, then 1
1
1
2
and 1
1
1
2
are Galois over J,
and there is an exact sequence
1 Gal(1
1
1
2
,J) Gal(1
1
,J) Gal(1
2
,J) Gal(1
1
1
2
,J) 1.
In this case, 1
1
1
2
=Q| where is a primitive cube root of 1. The degree is 18.
A-11. Over Q, the splitting eld is Q. | where
6
=5 and
3
=1 (because is then a
primitive 6th root of 1). The degree is 12, and the Galois group is D
6
(generators (26)(35)
and (123456)).
Over R, the Galois group is C
2
.
A-12. Let the coefcients of be a
1
. . . . . a
n
they lie in the algebraic closure of J. Let
g(X) be the product of the minimum polynomials over J of the roots of in .
Alternatively, the coefcients will lie in some nite extension 1 of J, and we can take
the norm of (X) from 1X| to JX|.
A-13. If is separable, 1: J| =(G
(
: 1), which is a subgroup of S
n
. Etc..
A-14.
_
3
_
7 will do.
A-15. The splitting eld of X
4
2 is 1
1
=Qi. | where
4
=2; it has degree 8, and Galois
group D
4
. The splitting eld of X
3
5 is 1
2
=Q. |; it has degree 6, and Galois group
D
3
. The Galois group is the product (they could only intersect in Q
_
3|, but
_
3 does not
become a square in 1
1
).
A-16. The multiplicative group of J is cyclic of order q 1. Hence it contains an element
of order 4 if and only if 4[q 1.
A-17. Take =
_
2
_
5
_
7.
A-18. We have 1
1
= 1
1
1
, which has degree n over J, and 1
2
= 1
~1n>
, which has
degree (n1) over J, etc.. This is really a problem in group theory posing as a problem in
eld theory.
A-19. We have Q| =Qi.
t
| where
t
is a primitive cube root of 1 and i =
3
etc..
A-20. The splitting eld is Q.
3
_
3|, and the Galois group is S
3
.
A-21. Use that
(
4
)(1
2
) =
4

133
A-22. (a) is Dedekinds theorem. (b) is Artins theorem 3.4. (c) is O.K. because X
]
a
]
has a unique root in .
A-23. The splitting eld is Qi. | where
4
=3, and the Galois group is D
4
with generators
(1234) and (13) etc..
A-24. From Hilberts theorem 90, we know that the kernel of the map N: 1

consists
of elements of the form
c

. The map 1

,
c

, has kernel J

. Therefore the
kernel of N has order
q
m
-1
q-1
, and hence its image has order q 1. There is a similar proof
for the trace I dont know how the examiners expected you to prove it.
A-25. (a) is falsecould be inseparable. (b) is truecouldnt be inseparable.
A-26. Apply the Sylow theorem to see that the Galois group has a subgroup of order 81.
Now the Fundamental Theorem of Galois theory shows that J exists.
A-27. The greatest common divisor of the two polynomials over Q is X
2
X 1, which
must therefore be the minimum polynomial for 0.
A-28. Theorem on -groups plus the Fundamental Theorem of Galois Theory.
A-29. It was proved in class that S
]
is generated by an element of order and a transposition
(4.15). There is only one J, and it is quadratic over Q.
A-30. Let 1=1|. The splitting eld of the minimum polynomial of has degree at most
J, and a set with J elements has at most 2
d
subsets. [Of course, this bound is much too
high: the subgroups are very special subsets. For example, they all contain 1 and they are
invariant under a a
-1
.]
A-31. The Galois group is (Z,5Z)

, which cyclic of order 4, generated by 2.

(
4
) (
2

3
) =1. (
4
)(
2

3
) =1.
(a) Omit.
(b) Certainly, the Galois group is a product C
2
C
4
.
A-32. Let a
1
. . . . . a
5
be a transcendence basis for
1
,Q. Their images are algebraically
independent, therefore they are a maximal algebraically independent subset of
2
, and
therefore they form a transcendence basis, etc..
A-33. C
2
C
2
.
A-34. If (X) were reducible over Q
_
7|, it would have a root in it, but it is irreducible
over Q by Eisensteins criterion. The discriminant is 675, which is not a square in R, much
less Q
_
7|.
A-35. (a) Should be X
5
6X
4
3. The Galois group is S
5
, with generators (12) and
(12345) it is irreducible (Eisenstein) and (presumably) has exactly 2 nonreal roots. (b) It
factors as (X 1)(X
4
X
3
X
2
X 1). Hence the splitting eld has degree 4 over F
2
,
and the Galois group is cyclic.
A-36. This is really a theorem in group theory, since the Galois group is a cyclic group of
order n generated by 0. If n is odd, say n =2m1, then =0
n
does.
A-37. It has order 20, generators (12345) and (2354).
A-38. Take 1
1
and 1
2
to be the elds corresponding to the Sylow5 and Sylow43 subgroups.
Note that of the possible numbers 1. 6. 11. 16. 21. ... of Sylow 5-subgroups, only 1 divides
43. There are 1, 44, 87, ... subgroups of ....
134 C. SOLUTIONS TO THE EXERCISES
A-39. See Exercise 14.
A-40. The group J

is cyclic of order 80; hence 80, 1, 8.

A-41. Its D
6
, with generators (26)(35) and (123456). The polynomial is irreducible by
Eisensteins criterion, and its splitting eld is Q. | where =1 is a cube root of 1.
A-42. Example 5.5.
A-43. Omit.
A-44. Its irreducible by Eisenstein. Its derivative is 5X
4
5
4
, which has the roots X =.
These are the max and mins, X = gives negative; X = gives positive. Hence the graph
crosses the .-axis 3 times and so there are 2 imaginary roots. Hence the Galois group is S
5
.
A-45. Its roots are primitive 8th roots of 1. It splits completely in F
25
. (a) (X
2
2)(X
2
3).
A-46. j()j() =q
2
, and j()j(
q
2

) =q
2
. Hence j(
q
2

) is the complex conjugate of j().

Hence the automorphism induced by complex conjugation is independent of the embedding
of Q| into C.
A-47. The argument that proves the Fundamental Theorem of Algebra, shows that its Galois
group is a -group. Let 1 be the splitting eld of g(X), and let H be the Sylow -subgroup.
Then 1
1
=J, and so the Galois group is a -group.
A-48. (a) C
2
C
2
and S
3
. (b) No. (c). 1
A-49. Omit.
A-50. Omit.
A-51. 1024 =2
10
. Want o. . =1, i.e., N. =1. They are the elements of the form
cx
x
;
have
1 k

x|-
x
x
1

.
Hence the number is 2
11
,2
10
=2.
A-52. Pretty standard. False; true.
A-53. Omit.
A-54. Similar to a previous problem.
A-55. Omit.
A-56. This is really a group theory problem disguised as a eld theory problem.
A-57. (a) Prove its irreducible by apply Eisenstein to (X 1). (b) See example worked
out in class.
A-58. Its D
4
, with generators (1234) and (12).
A-59. Omit.
SOLUTIONS FOR THE EXAM.
1. (a) Let o be an automorphism of a eld 1. If o
4
=1 and
o() o
3
() = o
2
() all 1.
show that o
2
=1.
135
If o
2
= 1, then 1. o. o
2
. o
3
are distinct automorphisms of 1, and hence are linearly
independent (Dedekind 5.14) contradiction. [If o
2
= 1, then the condition becomes
2o =2, so either o =1 or the characteristic is 2 (or both).]
(b) Let be a prime number and let a. b be rational numbers such that a
2
b
2
=1. Show
that there exist rational numbers c. J such that a =
c
2
]d
2
c
2
-]d
2
and b =
2cd
c
2
-]d
2
.
Apply Hilberts Theorem 90 to Q
_
| (or Q
_
|, depending how you wish to correct
the sign).
2. Let (X) be an irreducible polynomial of degree 4 in QX|, and let g(X) be the resolvent
cubic of . What is the relation between the Galois group of and that of g? Find the
Galois group of if
(a) g(X) =X
3
3X 1;
(b) g(X) =X
3
3X 1.
We have G

=G
(
,G
(
V , where V ={1. (12)(34). . . .]. The two cubic polynomials
are irreducible, because their only possible roots are 1. From their discriminants, one nds
that the rst has Galois group
3
and the second S
3
. Because (X) is irreducible, 4[(G
(
: 1)
and it follows that G
(
=
4
and S
4
in the two cases.
3. (a) How many monic irreducible factors does X
255
1 F
2
X| have, and what are their
degrees?
Its roots are the nonzero elements of F
2
8, which has subelds F
2
4F
2
2F
2
. There
are 25616 elements not in F
16
, and their minimum polynomials all have degree 8. Hence
there are 30 factors of degree 8, 3 of degree 4, and 1 each of degrees 2 and 1.
(b) How many monic irreducible factors does X
255
1 QX| have, and what are their
degrees?
Obviously, X
255
1 =

d[255

d
=
1

15

255
, and we showed in class that
the
d
are irreducible. They have degrees 1. 2. 4. 8. 16. 32. 64. 128.
4. Let 1 be the splitting eld of (X
5
3)(X
5
7) QX|. What is the degree of 1 over
Q? How many proper subelds of 1 are there that are not contained in the splitting elds of
both X
5
3 and X
5
7?
The splitting eld of X
5
3 is Q. |, which has degree 5 over Q| and 20 over Q.
The Galois group of X
5
7 over Q. | is (by ...) a subgroup of a cyclic group of order
5, and hence has order 1 or 5. Since 7 is not a 5th power in Q. |, it must be 5. Thus
1: Q| =100, and
G =Gal(1,Q) =(C
5
C
5
) C
4
.
We want the nontrivial subgroups of G not containing C
5
C
5
. The subgroups of order
5 of C
5
C
5
are lines in (F
5
)
2
, and hence C
5
C
5
has 6 1 =7 proper subgroups. All
are normal in G. Each subgroup of C
5
C
5
is of the form H (C
5
C
5
) for exactly 3
subgroups H of G corresponding to the three possible images in G,(C
5
C
5
) =C
4
. Hence
we have 21 subgroups of G not containing C
5
C
5
, and 20 nontrivial ones. Typical elds:
Q|, Q. cos
2t
5
|, Q. |.
[You may assume that 7 is not a 5th power in the splitting eld of X
5
3.]
5. Consider an extension J of elds. Dene to be J-constructible if it is
contained in a eld of the form
J
_
a
1
. . . . .
_
a
n
|. a
i
J
_
a
1
. . . . .
_
a
i-1
|.
136 C. SOLUTIONS TO THE EXERCISES
Assume is a nite Galois extension of J and construct a eld 1, J 1 , such that
every a is 1-constructible and 1 is minimal with this property.
Suppose 1 has the required property. From the primitive element theorem, we know
=1a| for some a. Now a 1-constructible == : 1| is a power of 2. Take 1 =
1
,
where H is the Sylow 2-subgroup of Gal(,J).
6. Let be an extension eld of a eld J. Show that every J-homomorphism is
an isomorphism provided:
(a) is algebraically closed, and
(b) has nite transcendence degree over J.
Can either of the conditions (i) or (ii) be dropped? (Either prove, or give a counterexam-
ple.)
Let be a transcendence basis for ,J. Because o: is injective, o() is
algebraically independent over J, and hence (because it has the right number of elements)
is a transcendence basis for ,J. Now Jo| o . Because is algebraic over
Jo| and o is algebraically closed, the two are equal. Neither condition can be dropped.
E.g., C(X)C(X), X X
2
. E.g., = the algebraic closure of C(X
1
. X
2
. X
3
. . . .), and
consider an extension of the map X
1
X
2
, X
2
X
3
, . . ..
Index
algebra
diagonalizable, 102
etale, 102
algebraic, 1719
algebraic closure, 24
in a larger eld, 24
algebraic integer, 11
algebraically closed, 24
algebraically dependent, 109
algebraically independent, 109
algorithm
division, 9
Euclids, 10
factoring a polynomial, 12
automorphism, 35
base
neighbourhood, 91
basis
separating transcendence, 116
transcendence, 111
bound
upper, 85
characteristic
p, 9
zero, 8
closure
separable, 89
cohomology group, 70
commutative, 7
composite of elds, 15
conjugates, 38
constructible, 20, 43
cubic
resolvent, 50
cyclotomic polynomial, 63
degree, 13
of an algebra, 102
separable, 39
directed, 98
discriminant, 47
Eisensteins criterion, 11
element
maximal, 85
exponent, 72
extension
abelian, 39
cyclic, 39
nite, 13
Galois, 37
inseparable, 37
normal, 37
separable, 37
simple, 15
solvable, 39
extension eld, 13
J-homomorphism, 13
eld, 8
prime, 9
stem, 17
xed eld, 36
Frobenius
endomorphism, 9, 33
fundamental theorem
of algebra, 12, 20, 24, 25, 61
of Galois theory, 39
Galois, 92
Galois closure, 40
Galois correspondence, 97
Galois eld, 54
Galois group, 37
absolute, 93
innite, 93
of a polynomial, 44
Gaussian numbers, 13
general polynomial, 77
group
Cremona, 35
pronite, 99
topological, 91
group algebra, 66
homomorphism
crossed, 69
of elds, 8
of rings, 7
principal crossed, 69
ideal, 7
integral domain, 7
invariants, 36
inverse limit, 98
inverse system, 98
137
138 INDEX
Lemma
Gausss, 11
module
G-, 69
multiplicity, 31
norm, 70, 79
normal basis, 66
normal closure, 40
ordering
partial, 85
total, 85
PARI, 6, 10, 12, 13, 17, 18, 48, 51, 54, 56, 57, 63,
82
perfect eld, 32
(n), 63
polynomial
minimum, 17
monic, 10
primitive, 114
separable, 32
prime
Fermat, 23
primitive element, 59
primitive root of 1, 62
regular n-gon, 64
relatively prime, 101
ring, 7
root
multiple, 31
of a polynomial, 9
simple, 31
separable, 59
separable element, 39
separably closed, 89
S
n
, 44
solvable in radicals, 45
split, 28
splits, 23
splitting eld, 28
subeld, 8
generated by subset, 15
subring, 7
generated by subset, 14
symmetric polynomial, 75
elementary, 75
theorem
Artins, 36
binomial in characteristic , 9
Chinese remainder, 101
constructibility of n-gons, 64
constructible numbers, 21, 43
cyclotomic polynomials, 63
Dedekind, 55
Galois 1832, 45
Galois extensions, 38
independence of characters, 65
Liouville, 20
normal basis, 66
primitive element, 59
strong Nullstellensatz, 101
topology
Krull, 93, 117
trace, 79
transcendence degree, 113
transcendental, 1719