Jean-Pierre Escofier Auth. Galois Theory PDF

Graduate Texts in Mathematics 204
Editorial Board
S. Axler F.W. Gehring K.A. Ribet
Springer
New York
Berlin
Heidelberg
Barcelona
Hong Kong
London
Milan
Paris
Singapore
Tokyo
Graduate Texts in Mathematics
TAKEUTIIZARING. Introduction to 34 SPITZER. Principles of Random Walk.
Axiomatic Set Theory. 2nd ed. 2nded.
2 OXTOBY. Measure and Category. 2nd ed. 35 ALEXANDERIWERMER. Several Complex
3 SCHAEFER. Topological Vector Spaces. Variables and Banach Algebras. 3rd ed.
2nd ed. 36 KELLEy/NAMIOKA et al. Linear Topological
4 HILTON/STAMMBACH. A Course in Spaces.
Homological Algebra. 2nd ed. 37 MONK. Mathematical Logic.
5 MAC LANE. Categories for the Working 38 GRAUERTIFRITZSCHE. Several Complex
Mathematician. 2nd ed. Variables.
6 HUGHES/PIPER. Projective Planes. 39 ARVESON. An Invitation to C*-Algebras.
7 SERRE. A Course in Arithmetic. 40 KEMENY/SNELLIKNAPP. Denumerable
8 TAKEUTIIZARING. Axiomatic Set Theory. Markov Chains. 2nd ed.
9 HUMPHREYs. Introduction to Lie Algebras 41 APOSTOL. Modular Functions and
and Representation Theory. Dirichlet Series in Number Theory.
10 COHEN. A Course in Simple Homotopy 2nd ed.
Theory. 42 SERRE. Linear Representations of Finite
11 CONWAY. Functions of One Complex Groups.
Variable I. 2nd ed. 43 GILLMAN/JERISON. Rings of Continuous
12 BEALS. Advanced Mathematical Analysis. Functions.
13 ANDERSONIFULLER. Rings and Categories 44 KENDIG. Elementary Algebraic Geometry.
of Modules. 2nd ed. 45 LOEVE. Probability Theory I. 4th ed.
14 GoLUBITSKy/GUlLLEMIN. Stable Mappings 46 LoEVE. Probability Theory II. 4th ed.
and Their Singularities. 47 MOISE. Geometric Topology in
15 BERBERIAN. Lectures in Functional Dimensions 2 and 3.
Analysis and Operator Theory. 48 SACHsiWu. General Relativity for
16 WINTER. The Structure of Fields. Mathematicians.
17 ROSENBLATT. Random Processes. 2nd ed. 49 GRUENBERGlWEIR. Linear Geometry.
18 BALMOS. Measure Theory. 2nd ed.
19 BALMOS. A Hilbert Space Problem Book. 50 EDWARDS. Fermat's Last Theorem.
2nd ed. 51 KLINGENBERG. A Course in Differential
20 HUSEMOLLER. Fibre Bundles. 3rd ed. Geometry.
21 HUMPHREYS. Linear Algebraic Groups. 52 HARTSHORNE. Algebraic Geometry.
22 BARNESiMACK. An Algebraic Introduction 53 MANIN. A Course in Mathematical Logic.
to Mathematical Logic. 54 GRAVERIWATKINS. Combinatorics with
23 GREUB. Linear Algebra. 4th ed. Emphasis on the Theory of Graphs.
24 HOLMES. Geometric Functional Analysis 55 BROWNIPEARCY. Introduction to Operator
and Its Applications. Theory I: Elements of Functional Analysis.
25 HEWITT/STROMBERG. Real and Abstract 56 MASSEY. Algebraic Topology: An
Analysis. Introduction.
26 MANEs. Algebraic Theories. 57 CROWELLIFox. Introduction to Knot
27 KELLEY. General Topology. Theory.
28 ZARISKIISAMUEL. Commutative Algebra. 58 KOBLITZ. p-adic Numbers, p-adic
Vol.I. Analysis, and Zeta-Functions. 2nd ed.
29 ZARISKIISAMUEL. Commutative Algebra. 59 LANG. Cyclotomic Fields.
Vol.II. 60 ARNOLD. Mathematical Methods in
30 JACOBSON. Lectures in Abstract Algebra I. Classical Mechanics. 2nd ed.
Basic Concepts. 61 WHITEHEAD. Elements of Homotopy
31 JACOBSON. Lectures in Abstract Algebra II. Theory.
Linear Algebra. 62 KARGAPOLovlMERLDAKov.Fundamentals
32 JACOBSON. Lectures in Abstract Algebra of the Theory of Groups.
ill. Theory of Fields and Galois Theory. 63 BOLLOBAS. Graph Theory.
33 HIRSCH. Differential Topology. 64 EDWARDS. Fourier Series. Vol. I. 2nd ed.
(continued after index)

lean-Pierre Escofier
Galois Theory
Translated by Leila Schneps
With 48 Illustrations
t Springer
Jean-Pierre Escofier Translator
Institute Mathematiques de Rennes Leila Schneps
Campus de Beaulieu 36 rue de I'Orillon
Universite de Rennes I 75011 Paris
35042 Rennes Cedex France
France [email protected]
[email protected]
Editorial Board
S. Axler F.W. Gehring K.A. Ribet
Mathematics Department Mathematics Department Mathematics Department
San Francisco State East Hall University of California
University University of Michigan at Berkeley
San Francisco, CA 94132 Ann Arbor, MI 48109 Berkeley, CA 94720-3840
USA USA USA
Mathematics Subject Classification (2000): l1R32, I1S20, 12FIO, 13B05
Library of Congress Cataloging-in-Publication Data

Escofier, Jean-Pierre.
Galois theory / Jean-Pierre Escofier.
p. cm. - (Graduate texts in mathematics; 204)
Includes bibliographical references and index.
ISBN 0-387-98765-7 (alk. paper)
I. Galois theory. l. Title. II. Series.
QA174.2 .E73 2000
5l2'.3--dc2l 00-041906
Printed on acid-free paper.
Translated from the French Thiorie de Galois, by Jean-Pierre Escofier, first edition published by
Masson, Paris, © 1997, and second edition published by Dunod, Paris, © 2000, 5, rue Laromiguiere,
75005 Paris, France.
© 2001 Springer-Verlag New York, Inc.

All rights reserved. This work may not be translated or copied in whole or in part without the
written permission of the publisher (Springer-Verlag New York, Inc., 175 Fifth Avenue, New York,
NY 10010, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use
in connection with any form of information storage and retrieval, electronic adaptation, computer
software, or by similar or dissimilar methodology now known or hereafter developed is forbidden.
The use of general descriptive names, trade names, trademarks, etc., in this publication, even if the
former are not especially identified, is not to be taken as a sign that such names, as understood by
the Trade Marks and Merchandise Marks Act, may accordingly be used freely by anyone.
Production managed by Francine McNeill; manufacturing supervised by Joe Quatela.

Photocomposed copy prepared from the translator's TeX files.
Printed and bound by R.R. Donnelley and Sons, Harrisonburg, VA.
9 8 7 654 3 2 I
ISBN 0-387-98765-7 SPIN 10711904
Springer-Verlag New York Berlin Heidelberg

A member of BertelsmannSpringer Science+Business Media GmbH
Preface
This book begins with a sketch, in Chapters 1 and 2, of the study of alge-
braic equations in ancient times (before the year 1600). After introducing
symmetric polynomials in Chapter 3, we consider algebraic extensions of fi-
nite degree contained in the field C of complex numbers (to remain within
a familiar framework) and develop the Galois theory for these fields in
Chapters 4 to 8. The fundamental theorem of Galois theory, that is, the
Galois correspondence between groups and field extensions, is contained in
Chapter 8. In order to give a rounded aspect to this basic introduction of
Galois theory, we also provide
• a digression on constructions with ruler and compass (Chapter 5),
• beautiful applications (Chapters 9 and 10), and

• a criterion for solvability of equations by radicals (Chapters 11 and
12).
Many of the results presented here generalize easily to arbitrary fields (at
least in characteristic 0), or they can be adapted to extensions of infinite
degree.
I could not write a book on Galois theory without some mention of the
exceptional life of Evariste Galois (Chapter 13). The bibliography provides
details on where to obtain further information about his life, as well as
information on the moving story of Niels Abel.
After these chapters, we introduce finite fields (Chapter 14) and separable
extensions (Chapter 15). Chapter 16 presents two topics of current research:
vi Preface
firstly, the inverse Galois problem, which asks whether all finite groups
occur as Galois groups of finite extensions of Q and which we treat explicitly
in one very simple case, and secondly, a method for computing Galois
groups that can be programmed on a computer.
Most of the chapters contain exercises and problems. Some of the state-
ments are for practice, or are taken from past examinations; others suggest
interesting results beyond the scope of the text. Some solutions are given
completely, others are sketchy, and certain solutions that would involve
mathematics beyond the scope of the text are omitted completely.
Finally, this book contains a brief sketch of the history of Galois theory.
I would like to thank the municipal library in Rennes for having allowed
me to reproduce some fragments of its numerous treasures.
The entire book was written with its student readers in mind, and with
constant, careful consideration of the question of what these students will
remember of it several years from now.
lowe tremendous thanks to Annette Houdebine-Paugam, who helped
me many times, and to Bernard Le Sturn and Masson, who read the later
versions of the text and suggested many corrections and alterations.
Jean-Pierre Escofier
May 1997
Contents
Preface v
1 Historical Aspects of the Resolution of Algebraic

Equations 1
1.1 Approximating the Roots of an Equation . . . . . 1
1.2 Construction of Solutions by Intersections of Curves 2
1.3 Relations with Trigonometry . . . . . . . 2
1.4 Problems of Notation and Terminology. . 3
1.5 The Problem of Localization of the Roots 4
1.6 The Problem of the Existence of Roots. . 5
1. 7 The Problem of Algebraic Solutions of Equations 6
Toward Chapter 2 . . . . . . . . . . . . . . . . . 7
2 Resolution of Quadratic, Cubic, and Quartic Equations 9

2.1 Second-Degree Equations 9
2.1.1 The Babylonians 9
2.1.2 The Greeks . . . . 11
2.1.3 The Arabs. . . . . 11
2.1.4 Use of Negative Numbers 12
2.2 Cubic Equations . . . . . . . . . 13
2.2.1 The Greeks . . . . . . . . 13
2.2.2 Omar Khayyam and Sharaf ad Din at Tusi 13
2.2.3 Scipio del Ferro, Tartaglia, Cardan . . . . . 14
2.2.4 Algebraic Solution of the Cubic Equation . 15
2.2.5 First Computations with Complex Numbers. 16
2.2.6 Raffaele Bombelli . . . . . . . . . . . . . . . . 17
viii Contents
2.2.7 Fran<;ois Viete . 18

2.3 Quartic Equations . . . 18
Exercises for Chapter 2 19
Solutions to Some of the Exercises 22
3 Symmetric Polynomials 25
3.1 Symmetric Polynomials 25
3.1.1 Background... 25
3.1.2 Definitions . . . 26
3.2 Elementary Symmetric Polynomials 27
3.2.1 Definition........... 27
3.2.2 The Product of the X - Xi; Relations Between Co-
efficients and Roots . . . . . . . . . . . . . . . . . . 27
3.3 Symmetric Polynomials and Elementary Symmetric Polyno-
mials . . . . . . . . 29
3.3.1 Theorem 29
3.3.2 Proposition 31
3.4 Newton's Formulas 32
3.5 Resultant of Two Polynomials. 35
3.5.1 Definition....... 35
3.5.2 Proposition...... 35
3.6 Discriminant of a Polynomial 37
3.6.1 Definition. 37
3.6.2 Proposition...... 37
3.6.3 Formulas . . . . . . . 38
3.6.4 Polynomials with Real Coefficients: Real Roots and
Sign of the Discriminant . . 38
Exercises for Chapter 3 . . . . . . 39
4 Field Extensions 51
4.1 Field Extensions 51
4.1.1 Definition 51
4.1.3 The Degree of an Extension 52
4.1.4 Towers of Fields 52
4.2 The Tower Rule 53
4.2.1 Proposition .. 53
4.3 Generated Extensions 54
4.3.2 Definition.. 55
4.3.3 Proposition. 55
4.4 Algebraic Elements . 55
4.4.1 Definition.. 55
Contents ix
4.4.2 Transcendental Numbers . . . . . . . . . . . . 55

4.4.3 Minimal Polynomial of an Algebraic Element 56
4.4.4 Definition.................... 56
4.4.5 Properties of the Minimal Polynomial . . . . 57
4.4.6 Proving the Irreducibility of a Polynomial in Z[X] 57
4.5 Algebraic Extensions . . . . . . . . . . . . . . . . . . . 59
4.5.1 Extensions Generated by an Algebraic Element 59
4.5.2 Properties of K[a] ..... 59
4.5.3 Definition.................. 60
4.5.4 Extensions of Finite Degree . . . . . . . . 60
4.5.5 Corollary: Towers of Algebraic Extensions 61
4.6 Algebraic Extensions Generated by n Elements 61
4.6.1 Notation.. 61
4.6.2 Proposition................ 61
4.6.3 Corollary . . . . . . . . . . . . . . . . . 62
4.7 Construction of an Extension by Adjoining a Root 62
4.7.1 Definition. 62
4.7.2 Proposition . . . . . . . . . . . . 62
4.7.3 Corollary . . . . . . . . . . . . . 63
4.7.4 Universal Property of K[Xl/(P) 63
Toward Chapters 5 and 6 .. . . . 64
5 Constructions with Straightedge and Compass 79

5.1 Constructible Points . . . . . . . . . . . 79
5.2 Examples of Classical Constructions . . . . . . . 80
5.2.1 Projection of a Point onto a Line 80
5.2.2 Construction of an Orthonormal Basis from Two Points 80
5.2.3 Construction of a Line Parallel to a Given Line Pass-
ing Through a Point . . . . . . . . . . . . 81
5.3 Lemma . . . . . . . . . . . . . . . . . . . . . . . 82
5.4 Coordinates of Points Constructible in One Step 82
5.5 A Necessary Condition for Constructibility . . . 83
5.6 Two Problems More Than Two Thousand Years Old 84
5.6.1 Duplication of the Cube . . . . . . . 85
5.6.2 Trisection of the Angle . . . . . . . 85
5.7 A Sufficient Condition for Constructibility 85
6 K-HomoIllorphisms 93
6.1 Conjugate Numbers 93
6.2 K -Homomorphisms. 94
6.2.1 Definitions . 94
x Contents
6.2.2 Properties . 94
6.3 Algebraic Elements and K-Homomorphisms 95
6.3.2 Example. 96
6.4 Extensions of Embeddings into C 97
6.4.1 Definition 97
6.5 The Primitive Element Theorem 99
6.5.1 Theorem and Definition 99
6.5.2 Example. 100
6.6 Linear Independence of K-Homomorphisms 101
6.6.1 Characters 101
6.6.2 Emil Artin's Theorem 101
6.6.3 Corollary: Dedekind's Theorem 102
7 Normal Extensions 107

7.1 Splitting Fields 107
7.1.1 Definition 107
7.1.2 Splitting Field of a Cubic Polynomial 108
7.2 Normal Extensions 108
7.3 Normal Extensions and K-Homomorphisms 109
7.4 Splitting Fields and Normal Extensions 109
7.4.2 Converse 110
7.5 Normal Extensions and Intermediate Extensions 110
7.6 Normal Closure. 111
7.6.1 Definition 111
7.7 Splitting Fields: General Case . 112
Toward Chapter 8 113
8 Galois Groups 119

8.1 Galois Groups 119
8.1.1 The Galois Group of an Extension 119
8.1.2 The Order of the Galois Group of a Normal Exten-
sion of Finite Degree . 120
8.1.3 The Galois Group of a Polynomial 120
8.1.4 The Galois Group as a Subgroup of a Permutation
Group 120
Contents xi
8.1.5 A Short History of Groups 121

8.2 Fields of Invariants . . . . . . . . . 122
8.2.1 Definition and Proposition. 122
8.2.2 Emil Artin's Theorem . . . 122
8.3 The Example of Q [~,j]: First Part 124
8.4 Galois Groups and Intermediate Extensions 126
8.5 The Galois Correspondence . . . . . . . 126
8.6 The Example of Q [ ~,j]: Second Part 128
8.7 The Example X4 + 2 . . . . . . 128
8.7.1 Dihedral Groups . . . . . . . 128
8.7.2 The Special Case of D4 .. . 129
8.7.3 The Galois Group of X4 + 2 130
8.7.4 The Galois Correspondence . 130
8.7.5 Search for Minimal Polynomials. 132
Toward Chapters 9, 10, and 12 .. 133
9 Roots of Unity 149

9.1 The Group U(n) of Units of the Ring 7l/n71 149
9.1.1 Definition and Background 149
9.1.2 The Structure of U(n) . 150
9.2 The Mobius Function . . . . . 151
9.2.1 Multiplicative Functions 151
9.2.2 The Mobius Function . 151
9.2.3 Proposition....... 151
9.2.4 The Mobius Inversion Formula 152
9.3 Roots of Unity . . . . . . . 153
9.3.1 n-th Roots of Unity 153
9.3.2 Proposition..... 153
9.3.3 Primitive Roots. . . 153
9.3.4 Properties of Primitive Roots 153
9.4 Cyclotomic Polynomials . . . . . . . 153
9.4.1 Definition........... 153
9.4.2 Properties of the Cyclotomic Polynomial 153
9.5 The Galois Group over Q of an Extension of Q by a Root
of Unity . . . . . . . . . . . . . . . 156
10 Cyclic Extensions 179

10.1 Cyclic and Abelian Extensions . . . . . . . . 179
10.2 Extensions by a Root and Cyclic Extensions. 179
10.3 Irreducibility of XP - a 180
10.4 Hilbert's Theorem 90 . . . . . . . . . . . . . . 181
xii Contents
10.4.1 The Norm . . . . . . . . . . . . . . . . . . . . . 181

10.4.2 Hilbert's Theorem 90 182
10.5 Extensions by a Root and Cyclic Extensions: Converse 182
10.6 Lagrange Resolvents 183
10.6.1 Definition . . . . . . . . . . 183
10.6.2 Properties . . . . . . . . . 183
10.7 Resolution of the Cubic Equation 184
10.8 Solution of the Quartic Equation 186
10.9 Historical Commentary . . . . . . 188
11 Solvable Groups 195

11.1 First Definition 195
11.2 Derived or Commutator Subgroup 196
11.3 Second Definition of Solvability 196
11.4 Examples of Solvable Groups .. . 197
11.5 Third Definition . . . . . . . . . . 197
11.6 The Group An Is Simple for n ?: 5 198
11.6.1 Theorem . . . . . . . . . . 198
11.6.2 An Is Not Solvable for n ?: 5, Direct Proof. 199
11.7 Recent Results . . . . . . . . . . . 199
12 Solvability of Equations by Radicals 207

12.1 Radical Extensions and Polynomials Solvable by Radicals 207
12.1.1 Radical Extensions . . . . . . . . 207
12.1.2 Polynomials Solvable by Radicals. 208
12.1.3 First Construction . . . . . . . . . 208
12.1.4 Second Construction . . . . . . . . 208
12.2 If a Polynomial Is Solvable by Radicals, Its Galois Group Is
Solvable . . . . . . . . . . . . . . . . . . . . . . . . 209
12.3 Example of a Polynomial Not Solvable by Radicals 209
12.4 The Converse of the Fundamental Criterion 210
12.5 The General Equation of Degree n . . . . . . . . . 210
12.5.1 Algebraically Independent Elements . . . . 210
12.5.2 Existence of Algebraically Independent Elements 211
12.5.3 The General Equation of Degree n . . . . . . . . 211
12.5.4 Galois Group of the General Equation of Degree n 211
13 The Life of Evariste Galois 219

Contents xiii
14 Finite Fields 227

14.1 Algebraically Closed Fields 227
14.1.1 Definition . . . . . . 227
14.1.2 Algebraic Closures . 228
14.1.3 Theorem (Steinitz, 1910) 228
14.2 Examples of Finite Fields .. 229
14.3 The Characteristic of a Field 229
14.3.1 Definition . . . . . 229
14.3.2 Properties . . . . . 229
14.4 Properties of Finite Fields 230
14.4.1 Proposition . . . . 230
14.4.2 The Frobenius Homomorphism 231
14.5 Existence and Uniqueness of a Finite Field with pr Elements 231
14.5.1 Proposition . . . . . 231
14.5.2 Corollary . . . . . . . . . . . . . . . . . . 232
14.6 Extensions of Finite Fields. . . . . . . . . . . . . 233
14.7 Normality of a Finite Extension of Finite Fields. 233
14.8 The Galois Group of a Finite Extension of a Finite Field. 233
14.8.1 Proposition . . . . . . . . . 233
14.8.2 The Galois Correspondence 234
14.8.3 Example. . . . . . . . . . . 234
15 Separable Extensions 257

15.1 Separability . . . . . . . . . . . . . . 257
15.2 Example of an Inseparable Element. 258
15.3 A Criterion for Separability . . . . . 258
15.4 Perfect Fields . . . . . . . . . . . . . 259
15.5 Perfect Fields and Separable Extensions 259
15.6 Galois Extensions. 260
15.6.1 Definition . . . . . . . . . . 260
15.6.2 Proposition . . . . . . . . . 260
15.6.3 The Galois Correspondence 260
Toward Chapter 16 . . . . . . . . . 260
16 Recent Developments 261

16.1 The Inverse Problem of Galois Theory 261
16.1.1 The Problem . . . 261
16.1.2 The Abelian Case . . . . . . . 262
16.1.3 Example. . . . . . . . . . . . . 262
16.2 Computation of Galois Groups over Q for Small-Degree Poly-
nomials . . . . . . . . . . . . . . . . 262
16.2.1 Simplification of the Problem 263
16.2.2 The Irreducibility Problem . 263
xiv Contents
16.2.3 Embedding of G into Sn . . . . . . . . . . . . . . . . 263

16.2.4 Looking for G Among the Transitive Subgroups of Sn 264
16.2.5 Transitive Subgroups of S4 264
16.2.6 Study of (G) cAn . . 265
16.2.7 Study of (G) c D4 . . 266
16.2.8 Study of (G) c 7L/47L 267
16.2.9 An Algorithm for n = 4 268
Bibliography 271
Index 277
1
Historical Aspects of the Resolution of
Algebraic Equations
In this chapter, we briefly recall the many different aspects of the study
of algebraic equations, and give a few of the main features of each aspect.
One must always remember that notions and techniques which we take
for granted often cost mathematicians of past centuries great efforts; to
feel this, one must try to imagine oneself possessing only the knowledge
and methods which they had at their disposal. The bibliography contains
references to some very important ancient texts as well as some recent texts
on the history of these subjects (see, in particular, the books by J.-P. Tignol
and H. Edwards and the articles by C. Houzel).
1.1 Approximating the Roots of an Equation

Around the year 1600 B.c., the Babylonians are known to have been able
to give extremely precise approximate values for square roots. For instance,
they computed a value approximating v'2 with an error of just 10- 6 . In
sexagesimal notation, this number is written 1.24.51.10, which means
24 51 10
1 + 60 + 60 2 + 603 = 1,41421296 ....
Later (around the year 200 A.D.), Heron of Alexandria sketched the well-
known method of approximating square roots by using the sequence
Un+! = ~ ( Un + :n) .
J.-P.Escofier, Galois Theory
© Springer-Verlag New York, Inc. 2001
2 1. Historical Aspects of the Resolution of Algebraic Equations
It is not possible to give here the full history of approximations as devel-

oped by Chinese (who computed cube roots as far back as 50 B.C.) and Arab
mathematicians. Note, however, that the linearization method developed
by Isaac Newton using the sequence
J(Un)
Un+l = Un - J'(U n )
was already known to the Arab mathematician Sharaf ad Din at Tusi, born
in 120l.
In 1225, Leonard of Pisa gave the approximate value 1.22.7.42.33.40 (in
base 60) for the positive root of the equation x 3 + 2x2 + lOx = 20. It is an
excellent approximation, with an error on the order of just 10- 10 ; we do
not know how he obtained it.
1.2 Construction of Solutions by Intersections of

Curves
The Greeks were able to geometrically construct every positive solution of
a quadratic equation, using intersections of lines and circles, but they did
not formulate this problem in an algebraic manner. We will return to their
procedures in Chapter 5. To solve cubic equations, they used conics, as did
Omar Khayyam around 1100 (see §2.2.2); perhaps this method was already
understood by Archimedes (287-212 B.C.).
In his book Geometry, one of three treatises attached to his grand work
Discours de la Methode, Rene Descartes related solutions of algebraic equa-
tions to intersections of algebraic curves. This theme is one of the sources
of algebraic geometry.
1.3 Relations with Trigonometry

The division of the circle into a certain number of equal parts, or cyclotomy
(coming from a Greek word), was the object of a great deal of study. By
studying the construction of the regular nine-sided polygon, which leads to
a cubic equation, mathematicians of the Arab world revealed the relation,
subsequently described also by Fran«ois Viete (1540-1603), between the
trisection of an angle and the solution of a cubic equation (see Exercise
2.5). Viete also gave formulas expressing sin nO and cos nO as functions of
sinO and cosO. Laurent Wantzel showed in 1837 that the problem posed
by the Greeks, of trisecting an arbitrary angle using only a ruler and a
compass, was impossible (see §5.6).
Probably inspired by work of Alexandre Vandermonde dating back to
1770, Carl Friedrich Gauss showed how to given an algebraic solution for
1.4 Problems of Notation and Terminology 3
the division of the circle into p equal parts whenever p is a Fermat prime
(p = 17,257,65537); his results are presented in the seventh part of his
Disquisitiones arithmeticae published in 1801, which prepared the way for
Abel and Galois.
1.4 Problems of Notation and Terminology

Before the 17th century, mathematicians usually did not use any particular
notation; it is easy to conceive of the difficulty of developing algebraic meth-
ods under these conditions! Modern notation was more or less developed
by Descartes, who used it in his book Geometry.
Let us give an idea of the notation used by Viete. In his Zetetiques (1591,
from the Greek C,'f)TctV, meaning "search"), the expression
P·H+P·B =E
---=D=---+-:P=---
is written
{+;P +ll:nn BHp} requabitur E.
Viete's notation for powers of the unknown is very heavy: he writes "A
quadratum" for A 2 , "A cubus" for A 3 , "A quadrato-quadratum" for A 4 ,
etc., and "A potestas," "A gradum" for ATn,An . To indicate the dimen-
sion of the parameter P, he writes "P planum" for P of dimension 2, "P
solidum" for P of dimension 3, etc.
For example, for the general equation of the second degree in A, Viete,
who always assumes homogeneity of dimension between the variables and
the parameters B, D, Z, writes:
B in A quadratum plus D plano in A requari Z solido,

i.e. BA2 + DA = Z.
This condition of homogeneity was definitively abandoned only around
the time of Descartes (see §5.7). The great contribution of Viete was the
creation of a system of computation with letters used to represent known or
unknown quantities (logistice speciosa, as opposed to logistice numerosa).
This idea produced a deep transformation in the methods and conception
of algebra; instead of working only on numerical examples, one could con-
sider the general case. The economy of thought produced by this approach,
and the new understanding it gave rise to, made further progress possible.
Certainly, letters had been used before Viete, but not in actual computa-
tions; one letter would be used for a certain quantity, another for its square,
and so forth.
Viete was known in his time as a counselor of Henri III, and that he was
a counselor in the Parliament of Bretagne in Rennes from 1573 to 1580.
Let us give some of the main turning points in the history of algebraic
notation.
Decimals were introduced by Al Uqlidisi, the Euclidean (around 950),
as well as by Al Kashi (1427), Viete (1579), Simon Stevin (1585). The use
of a point to separate the integer and fractional parts of a number was
made popular by John Neper (in France, a comma is used instead of a
point). But even long after the introduction of the point, people continued
to write a number as an integer followed by its fractional part in the form
224176
of a fraction: 111 000000 .
The signs + and - were already in use around 1480 (+ was apparently
a deformation of the symbol &), but by the beginning of the 17th cen-
tury, they were used generally. Multiplication was written as M by Michael
Stifel (1545), and as in by Viete (1591); our current notation dates back to
William Oughtred (1637) for the symbol x, and to Wilhelm Leibniz (1698)
for the dot.
For powers of the unknown, 1, 225 + 148 x 2 was written as 1, 225 P148 2
by Nicolas Chuquet (1484), 3x 2 was written as 3~ by Raffaele Bombelli
(1572), whereas Stevin wrote 3@+ 5@- 4Q)for 3x 3 + 5x 2 - 4x. The ex-
ponential notation x 2 , x 3 , etc., came with Descartes, whose formulas are
actually written in a notation very close to our own. In the 18th century,
one sees bb for b2 , but b3 , b4 , etc.
Only after methods of explicit computation and exponential notation had
been perfected did it become possible to think clearly about computing
with polynomials. Descartes showed that a polynomial vanished at the
value a if and only if it was divisible by X-a. The history of the manner
of referring to the unknown is extremely complicated, and we will not
describe it here. The symbol = used by Michel Recorde (1557) came to
replace the symbol used by Descartes, an a written backward, toward the
end of the 17th century, thanks to Leibniz. Albert Girard (1595-1632)
introduced the notation ~, which he substituted for CD; he also introduced
the abbreviations for sine and tangent, and used the symbols <, > like
Harriot. Indices were introduced by Gabriel Cramer (1750) to write his
famous formulas (the use of primes " ", '" followed by iv, v etc. became
widespread around the same time); indices of indices were introduced by
Galois. The symbol L: was introduced by Leonhard Euler (1707-1783).
These notations passed into general usage only during the 20th century.
1.5 The Problem of Localization of the Roots

This problem concerns polynomials with real coefficients. The results of
Descartes based on the number of sign changes in the sequence of coeffi-
1.6 The Problem of the Existence of Roots 5
cients (see Exercise 3.7) were perfected in the 19th century by Jean-Baptiste
Fourier and Franc,;ois Budan, and then by Charles Sturm, who in 1830 gave
an algorithm to determine the number of real roots in a given interval.
1.6 The Problem of the Existence of Roots

Al Khwarizmi appears to have been the first, around the year 830, to have
pointed out the existence of quadratic equations having two strictly positive
roots (see, however, §2.1.1). Negative roots were taken into consideration
only around the end of the 16th century (see §2.1.4).
Girard was the first to assert that an equation of degree (or denomi-
nation, as he said) n has n roots (Figure 1.1). He did not give any proof
and his ideas about the exact nature of the solutions seem rather vague;
he thought of them as complex numbers or other similar numbers. This
vagueness did not prevent him from innovating the use of computations
with roots as though they were numbers (see §3.4). Every mathematician
will appreciate his wonderful formulation
"pour la certitude de la reigle generale"
(for the certitude of the general rule).
I I. Thcomnt.
Toutesles eqllatiQlls a'algchrerecoivent autant ae {olutions! que h
denomination de la plus haute quantite Ie demonll:re, excepte Ies in corn-
plettes
Expliciltion.
Soit une equation Gomplette I (1) efgalc 4 CD+70-340
--: 24 : alorsle denominatc;ur dela plus baute quantite ell: (1), quill-
gOlfie qu'il y a quatre certaines rolutions, & non plus ny moins, com~
me r ,2,-3,4
Done i1 [c faut refouven·ir· d'ub[crvcr tousjours ecla : on pourroit dire 1I

quay [ert ces [olutions qui font impoffib~es,jerefpond pour trois chofes,
pour la certitude de la reigle g~nerale, & qu'il ny 3'point d'autre rclu-
tions, & pour fan utilite
FIGURE 1.1. Excerpt from Girard's Invention nouvelle en l'algebre ... , 1629
Descartes was less precise about the number of roots, simply bounding
it by the degree of the equation: "Aut ant que la quantite inconnue a de
dimensions, aut ant peut-il y avoir de diverses racines." ("As many as the di-
mensions of the unknown quantity, as many there may be different roots.")
The nature of the roots also escaped Leibniz, who did not see that A J
is a complex number (1702). But the methods of integration of rational
functions, which were developed by Leibniz and Jean Bernoulli around this
time, led Leonhard Euler to the problem of showing that an algebraic equa-
tion P(x) = 0, where P is a polynomial of degree n with real coefficients,
has n real or complex roots (1749: Researches on the imaginary roots of

equations) .
This theorem is usually known as the "fundamental theorem of algebra" .
In France, it is known as d'Alembert's theorem, because Jean d'Alembert
proposed an interesting but incomplete proof of it in 1746. In his course at
the Ecole Normale in the year III of the French Revolution, Pierre Simon
de Laplace gave an elegant proof, admitting only the existence of roots
somewhere. Gauss gave an entirely satisfying proof of the theorem at least
four times (in 1797-1799, twice in 1816, and in 1849), as did Jean Argand
(1814) and Louis Augustin Cauchy (1820). The fundamental theorem of
algebra can also be obtained as an immediate corollary of the theorem
known as Liouville's theorem (actually due to Cauchy, 1844), which states
that "every holomorphic function bounded on C is constant" .
1.7 The Problem of Algebraic Solutions of

Equations
This problem is the central subject of this book. Algebraically solving an
algebraic equation (or solving it by radicals) means expressing its solutions
by means of n-th roots, i.e. reducing its solution to the solution of equations
of the form xn = a.
Around 1700 B.C., the Babylonians were already in possession of a general
method for solving quadratic equations whose coefficients were given num-
bers. Solutions to cubic equations came only with Scipio del Ferro (1515),
and quartic equations were solved by Lodovico Ferrari (1540).
Ehrenfried Tschirnhaus (1683), followed by Michel Rolle (1699), Etienne
Bezout, and Leonhard Euler (1762) attempted to go further, but Euler still
believed that all algebraic equations were solvable by radicals "... one will
grant me that expressions for the roots do not contain any other operations
than extraction of roots, apart from the four vulgar operations, and one
could hardly support the position that transcendental operations meddle
in the situation" (§77 of the 1749 article cited above).
Around 1770, Joseph Louis Lagrange and Alexandre Vandermonde (as
well as Edward Waring) independently discovered the role played by sym-
metry properties in the solution of equations. We will detail their discoveries
in Chapter 10. As for the contribution of Gauss, we mentioned it in §1.3
above.
These ideas were exploited by Paolo Ruffini (1802-1813) to prove the
impossibility of solving the general equation of the fifth degree by radicals,
and then by Niels Abel (1823-1826) to prove the impossibility of solving
the general equation of degree 2: 5 by radicals (see Chapter 12). However,
the analysis of their texts would occupy too much of this book; we refer the
reader to the books and articles cited in the introduction to this chapter.
Toward Chapter 2 7
Finally, in 1830, Galois, who knew nothing of Abel's results, created the
notions of a group (limited to permutation groups), a normal subgroup,
and a solvable group, which allowed him - at least theoretically - to re-
late the solvability of an equation by radicals to the properties of a group
associated to the equation, opening new horizons that are far from having
been completely explored even today.
Toward Chapter 2
Before giving a complete exposition of Galois theory in Chapter 4, we
devote the following chapter to the history of the solution of algebraic
equations through the year 1640.
2
History of the Resolution of
Quadratic, Cubic, and Quartic
Equations Before 1640
In this chapter, we give only a brief sketch of the rich history of low-
degree equations; in particular, we have omitted the Indian and Chinese
contributions. Readers interested in the subject can find excellent sources in
the bibliography (see, in particular, the books by Tignol, Van der Waerden,
and Yushkevich).
2.1 Second-Degree Equations

2.1.1 The Babylonians
The earliest form of writing was invented by the Sumerians in Mesopotamia
around 3300 B.C., although some people believe that Egyptian writing was
invented earlier. Archaeologists have excavated texts that were written on
humid clay tablets later dried in the sun. The earliest known texts are very
short and mostly concern accounting: sacks of grain, domestic animals,
slaves. They use a numeral system in base 60, which is at the origin of
our division - still in use after 5000 years! - of the hour into minutes and
seconds and the circle into degrees.
After various historical events, this extraordinary civilization gave way,
during the period 1900 to 1600 B.C., to an empire whose capital was Baby-
lon, on the Euphrates, just south of Baghdad today. Quantities of interest-
ing information are preserved in the tablets of this period; in particular,
they reveal that Babylonians possessed a well-developed algebra and mas-
tered the solution of second-degree equations.

10 2. Resolution of Quadratic, Cubic, and Quartic Equations
EXAMPLE. - "I added 7 times the side of my square and 11 times the
surface: 6.15" (tablet nO 13901 from the British Museum).
This problem discusses the quadratic equation llx 2 + 7x = 6.15; the

notation 6.15 in base 60 is ambiguous because the Babylonians gave no
indication of the scale: 6.15 could be 6 x (60)2 + 15 x 60 or 6 x 60 + 15, or
6/60 + 15/602, or even 6/3600 + 15, etc. (A kind of zero, serving to denote
the intermediate positions, was introduced by the Babylonians only around
300 B.C. Before that, they sometimes left a space, but more usually it was
just necessary to guess. Here, 6.15 = 6 + 15/60 = 6 + 1/4.)
To follow the solution described in the tablet, set a = 11, b = 7, and c =
-6i. The two left-hand columns of Table 2.1 are translated directly from
the tablet. The table also shows the numbers written in base 10 and the
corresponding literal computation. Note that in order to facilitate division,
the Babylonians had established tables of inverses. But 1/11 was not in the
tables, as it does not have a finite expansion in base 60.
Base 60 Base 10 Computation of
3
You will multiply 11 by 6.15 1.8.45 68+ - -ac
4
You will multiply 3.30 by 12.15

1
12+ - -
b2
4 4
3.30
You will add it to 1.8.45 1.21 81

b2
--ac
4
It is the square of 9 9 Vb: - ac

You will subtract 3.30 5.30
1
5+ -
2
-~ + Vb2 - ac
2 4
The inverse of 11 cannot be

computed
1 _~2 + Vb24 -ac

What, multiplied by 11, 30
gives 5.30?
2 a
The side of the square is 30.
TABLE 2.1. Method for solving a quadratic equation

2.1 Second-Degree Equations 11
OTHER EXAMPLES. - Here are the equations corresponding to other prob-

lems from the same tablet. The numbers in parentheses are the values to
be given to the Babylonian numbers:
x 2 +x = 45 (~)
x2 = x + 14.30 (870)
x2 - 20x 2 +X = 4.46.40 (~ and 286 + ~) .
COMMENTARY. - In these problems, the solutions are always positive num-

bers having simple finite expansions in base 60: the discriminant is the
square of a simple number, and the division by a works. Apart from these
restrictions, we see that the Babylonians mastered the algorithm for the
algebraic solution of quadratic equations. Even the case of second-degree
equations having two distinct positive roots seems to be considered in prob-
lems in which the length and width of a rectangle appear, which makes it
possible to distinguish numbers that cannot be distinguished algebraically
by using an order relation. However, they only wrote on their tablets
straightforward recipes to be followed; we have no idea how they actu-
ally thought of them. The deductive method in mathematics was invented
later, by the Greeks.
2.1.2 The Greeks

The irrationality of y'2 was proved around 430 B.C., probably by a geo-
metric argument. (The discovery is attributed to Hippasos of Metapont,
who supposedly was unable to endure the intellectual consequences of his
discovery and drowned himself in the Aegean Sea. At the very least, this
anecdote bears witness to the deep trouble provoked by the discovery.)
In Euclid's Elements (dating from about 300 B.C.), the methods are ge-
ometric; algebraic computations cannot be developed, because a product
of two lengths is considered to be a surface. Later, in the 3rd century A.D.,
Diophantus discovered an algebraic approach.
There is one important difference between the documentation at our
disposal on Babylonian and on Greek mathematics: the tablets preserve
the original state of Babylonian mathematics, whereas the work of the
Greeks is known to us only through manuscripts written a good thousand
years after the authors made their discoveries, which reworked the originals
in all kinds of ways. Some works are known only from their translations
into Arabic.
2.1.3 The Arabs

It is more correct to speak of mathematicians coming from the various
provinces of the Arab world, from Spain to the Middle East, than it is to
speak directly of "Arab mathematicians" . In the 8th century, these mathe-
maticians began to procure Greek texts from Constantinople; they also re-
ceived Indian books of computations that explained the use of zero. Around
820 to 830, al Khwarizmi (from Uzbekistan; he later became known through
Latin translations of his works, called Algorismus, origin of the word algo-
rithm), a member of the scientific community around the caliph al Mamoun,
described algebraic transformations in his treatise on algebra, which can
be expressed as the following equations in our notation:
6x 2 - 6x + 4 = 4x 2 - 2x + 8
6x 2 + 4 + 2x = 4x 2 + 8 + 6x by al jabr
3x 2 + 2 + x = 2x2 + 4 + 3x by al hatt
x2 = 2x+2 by al muqqabala.
The word al jabr, which expressed completion or setting of a fracture, is at
the origin of the appearance of the word "algebra" in the 14th century.
al Khwarizmi distinguishes six types of equations of degree less than or
equal to 2, because the coefficients a, b, and c of his equations are always
positive:
ax 2 = bx, ax 2 = b, ax = b,
ax 2 + bx = c, ax 2 + c = bx, ax 2 = bx + c.
For the equation x 2 = 40x - 4x 2, or x 2 = 8x, he gives only the root 8.
However, for the equation x 2 + 21 = lOx, he gives the two solutions 3 and
7 and asserts that the procedure is the same for all equations of the fifth
type. Geometric justifications are given, but unlike the Greeks, the spirit
of the method is algebraic.
2.1.4 Use of Negative Numbers

Negative numbers became widely used only around the end of the 16th cen-
tury. However, they actually appeared 1,000 years earlier in Indian math-
ematics and even earlier than that in Chinese mathematics.
In 1629, following ideas developed by Stevin in 1585, Girard did not
scruple to give examples of equations with negative roots: "The negative
in geometry indicates a regression, and the positive an advancement" (nor
was he bothered by complex non-real roots).
However, one must not believe that negative roots were accepted by
everyone: in 1768, Bezout still wrote that equations have negative roots
only when they are "vicious", and Lazare Carnot, the famous "organizer
of the victory" of the Republican armies, wrote in his treatise on geometry
in the year XI of the Revolution: "To obtain an isolated negative quantity,
one must remove an effective quantity from zero, but removing something
from nothing is an impossible operation."
2.2 Cubic Equations 13
2.2 Cubic Equations

2.2.1 The Greeks
On the rare occasions in which they encountered cubic equations, the
Greeks solved them by means of intersections of conics: ellipses, parabolas,
and hyperbolas. The oldest such solution goes back to Menechme (375-325
B.C.), who, to obtain an x such that x 3 = a 2 b, considered the intersection of
x 2 = ay and xy = ab (others expressed the same problem as the search for
numbers x and y such that a/x = x/y = y/b). The most famous solution,
which led to numerous further developments, goes back to Archimedes. He
sought to cut a sphere of radius R by a plane in such a way that the ratio
of the volumes of the two pieces had a given value k: we easily see that the
height h of one of the parts satisfies h 3 + (4k / (k + 1)) R3 = 3Rh2.
But the Greeks did not solve the problem of the duplication of the cube
with ruler and compass (equation x 3 = 2a 3 ), nor the trisection of the angle;
we will discuss these questions in Chapter 5.
2.2.2 Omar Khayyam and Sham! ad Din at Tusi

Omar Khayyam was a mathematician and an astronomer, but he was also
a poet, the author of many famous verses. He lived in central Asia and in
Iran (1048-1131). In his treatise on algebra (from around 1074), he studied
cubic equations in detail. He only considered equations with strictly positive
coefficients, and distinguished 25 different cases, some of which had already
been studied by al Khwarizmi. For example, the equations with three terms
not having zero as a root are of one of the following six forms (Omar
Khayyam expresses them in words, without notation, with homogeneity
conditions similar to those of §1.4):
x3 = ax + b, x3+ b = ax, x 3 + ax = b.
For x 3 + ax = b, he set a = c 2 , b = e2 h and obtained the solution as the

intersection of the parabola y = x 2/e and the circle y2 = x(h - x).
For x 3 + b = ax, he again set a = e 2, b = e 2 h and obtained the solution as
the intersection of the parabola y = x 2/ e and the hyperbola y2 = x( X - h).
One hundred years later, in a treatise that has just been reedited (see
the bibliography), Sharaf ad Din at Tusi classified equations, not according
to the sign of the coefficients like Khayyam, but according to the existence
of strictly positive roots. He solved the homogeneity problems in a manner
that appears to foreshadow Descartes (see §5.7): every number x can be
identified with a length or with a rectangular surface of sides 1 and x, or
even with the volume of a parallelepiped with sides 1,1 and x. Finally,
he inaugurated the study of polynomials via analysis, introducing their

derivative, seeking for their maxima, etc.
The solutions given by Omar Khayyam are geometric, obtained by taking
intersections of conics. As for algebraic solutions, he writes that "they are
impossible for us and even for those who are experts in this science. Perhaps
one of those who will come after us will find them." Similar remarks were
made by Luca Pacioli in 1494 but times were changing, because ...
2.2.3 Scipio del Ferro, Tartaglia, Cardan

... the work of Italian mathematicians since Leonard of Pisa finally reached
a conclusion in 1515. Scipio del Ferro, a professor in Bologna who died in
1526, discovered the algebraic solutions of the equations
q, (2.1)
px+q, (2.2)
px, (2.3)
probably with p, q > 0, Le. of type (2.1) only. The rest of the story is a
novel in episodes which is impossible to reconstruct completely, as many
of the details are known only because they were recounted by one of the
protagonists, in a manner that may lack objectivity.
In the year 1535, Fiore, a Venitian student of Scipio del Ferro, publicly
challenged Niccoli!. Tartaglia (roughly 1500-1559) to solve about 30 prob-
lems, all based on equations of type (2.1). At that time, winning a challenge
of this kind led to prestige and money, sometimes even allowing the winner
to obtain a position as a professor. Tartaglia's childhood was very dramatic:
a fatherless child, very poor, he was seriously wounded during the looting
of Brescia by troops led by Gaston de Foix in 1512. He had already at-
tempted to solve equations of this type some years earlier, and this time he
succeeded, during the night of February 12 to 13, 1535 (just in time to win
the challenge). But he kept his solution secret. He wrote it in a poem, in
which he used the word "thing", like his contemporaries, for the unknown.
Quando che'l cuba con Ie cose appresso
Se agguaglia a qualche numero discreto ... ,
(When the cube with the things is equal to a number. ... )
In 1539, Jerome Cardan, a doctor and mathematician, and a very com-
plex personality whose tumultuous life also makes a highly interesting story,
invited Tartaglia to his house in Milan to find out his secret. He flattered
him so well that he succeeded - Tartaglia showed him his poem - but swore
not to reveal it (March 25, 1539). Shortly after, Cardan succeeded in ex-
tending Tartaglia's method to equations of types (2.2) and (2.3) (unless
it was actually Tartaglia who succeeded), and one of his disciples, Ferrari
(1522-1560), solved the quartic equation in 1540.
In 1545, Cardan published all of these solutions in his book Ars Magna
(which literally means: Grand Work), taking care to thank Tartaglia three
times. But Tartaglia was furious, denounced him for lying, and the follow-
ing year published a text containing Cardan's promise, their conversations
together, and his own research. Ferrari defended his professor, saying that
he had been present at the meeting in 1539 and that there was never any
question of a secret. He then took up a new challenge proposed by Tartaglia
on August 10, 1548, which he appears to have won. And the story contin-
ued.
Cardan's Ars Magna is a very important book. In it, he gave the complete
solution of the cubic equation, finally (see, however, §2.2.5), as well as the
first computations using roots of negative numbers.
2.2.4 Algebraic Solution of the Cubic Equation

In 1545, Cardan explained on the basis of numerous numerical examples,
which he considered as clearly illustrating the general case, how to find
a root of the cubic equation. The problem of finding the three roots was
solved by Euler, in a Latin article from 1732.
Let us explain Cardan's method, using today's notation and without
distinguishing the different cases due to signs of the coefficients, as Cardan
did. We know that by translation, we can always reduce to the case of an
equation of the form x 3 + px + q = o.
Set x = u + v (for Cardan, this is either u + v or u - v according to the
signs of p and q), and require the numbers u and v to satisfy the condition
3uv = -po The equation can be written as
(U+V)3 +p(u+v) +q = 0; or as
so setting 3uv = -p, this gives
Setting U = u3 and V = v3 , this then gives

p3
U + V = -q, UV=--
27'
so that U and V are solutions of the quadratic equation X2 + qX - p3 /27 =
o. The discriminant of this quadratic equation is given by
If d is a number whose square is equal to this discriminant, then setting

U = -(q/2) + d and V = -(q/2) - d gives a solution.
Cardan concludes his procedure by giving the unique solution x =W+

W,i.e.
x = 3 _ CJ.
2
+ J
q2
4
+ p3 +
27
3 _ CJ. _
2
J
q2
4
+ p3 .
27
This formula requires the extraction of two cube roots (really just one since
v = -p/3u).
For us, this formula contains an ambiguity: each of the cube roots can
be chosen in three different ways, and their sum could have nine different
values. Let us now redo the method, considering the cube roots as Euler
did.
If u satisfies u 3 = U, then the condition 3uv = -p implies that v =
-p/3u, giving the solution
x=u+v
of the equation. The other cube roots of U are ju and j 2u, corresponding
to -p/3ju = j 2v and _p/3j2u = jv respectively; here j is a cube root of
unity, i.e. j = exp(2n /3). This gives the other solutions of the equation
If we reverse the choices of U and V, a cube root of -q/2 - d is one of the

three numbers above v, jv, j 2v, and fortunately, we find the same three
roots.
2.2.5 First Computations with Complex Numbers

The spark occurs near the end of the Ars Magna, in 1545 (Figure 2.1). The
idea was undoubtedly suggested to Cardan by the problems he studied in
dealing with cube roots as above.
um cfi minus,ideo imaginaberis ~ m: r 5 ,id eltdiffcrcntiz AD, &.

quadrupli A B,quam addc & minuc ex A e,& babcbis qlJ~lirum,fciU.
CC[5 P:~V:15 m: 40,& 5 m :~v: 15 m: 4O,fcu 5P: ~m:r ~, & 5'
m:~m: r 5,due 5 p:~ m: r 5 in 5' m: ~m: 15, dimifsis incruciarioni~
bus.fi[ 15 m:m: I 5,quod cit p: r 5,igimr hoc producrum cit 4O,nam
ra tame A D,non cit eadcm cii namra 4o,n(c A B, quia fupcrficies cfi
rcmotaanatl1ranumcri,&lin~,proximius 1 5P : RZm : 15
tame ~uie quanritari,qu~ u.(rc dlfophiruc~, 5 m:RZ m: r 5'
quomam pe~ cam, non ulm 'p uro m: nee l~ 115 m:m: r 5 d . elt 4 0 9
alijs , operatloncs cxcreerc lIeer , nee ucnan
qUid atclt,lIt addas q~ad~mm mcdictati~ ~u=:ncri ~ll~crO 'prod~
a
ctndo,& ~ aggrcgau mInuas ae addas dlmldlUm dlUldcndl.
FIGURE 2.1. Excerpt from the book Ars Magna by Cardan, 1545
This excerpt refers to the search for two numbers whose sum is 10 and
whose product is 40, leading to the equation x 2 - lOx + 40 = O. Cardan
recognized that no two numbers could satisfy this equation, but proposed
a sophisticated solution in which he imagined the number J -15; he then
checked the validity of this number by computing
(5 + yCl5) (5 - yCl5) = 25 - (-15) = 40,
writing this operation as
5p: :& m: 15,
5m: :& m: 15,
25 m : m 15 qd. est 40,
where p denotes +, m denotes -, and :& denotes the square root. One
passage provoked a great deal of commentary: dimissis incruciationibus,
which means setting aside the products in crosses, or, according to certain
translators who think Cardan is making a word play, setting aside the
mental torture.
In the case of the cubic equation, complex numbers enter in the case when
q2 /4 + p3/27 < 0, known as the irreducible case, in which the three roots
are real (see §3.6) and d is purely imaginary. Cardan did not understand
this case well; he simply showed how to obtain all three roots if one of them
is known (see Exercise 2.4).
2.2.6 Raffaele Bombelli

Born in 1530, Bombelli published a treatise on algebra in 1572 which im-
proved understanding of computations with complex numbers by showing
how Cardan's formulas can be applied in the irreducible case. He gave nu-
merous examples; one of the simplest is that of the equation which we
write as x 3 - 15x - 4 = 0, which has an obvious solution 4, knowing
which Cardan's formulas produce the quantities {h ± J-12l. Now, this
is the irreducible case since d2 = q2 /4 + p3/27 = 4 - 125 = -121 and
u 3 = U = -qj2 + d = 2 + v' -12l.
Bombelli explained this difficulty by showing that {h + v' -121 can ac-
tually be written in the form a + ib; identifying the real parts of (a + ib)3
and 2 + 11i, he found a 3 - 3ab2 = 2. The equality of the modules then gave
(a 2 +b2 )3 = (22+ 112) = 125, so a 2 +b2 = 5. He then substituted b2 = 5-a 2
into the previous equation, obtaining a 3 - 3a(5 - a 2 ) = 4a 3 -15a = 2 (this
is the original equation with x = 2a). Bombelli noticed that a = 2 is a root,
and deduced that b = 1, giving u = 2+i,v = 2-i, and u+v = 4 (with na-
tation as in §2.2.5 above). Abraham de Moivre (1667-1754) later observed
that this procedure requires having already solved the equation to sim-
plify the expression of the roots. Nonetheless, Bombelli's work is extremely
important: it opened the way to computations with complex numbers.
Bombelli's notation is Rc L2p dim 11 J: the cube root of the quantity
between the signs Land J, which is the abbreviation of of "2 pi di meno
11", where "pi di meno n" means +in. Bombelli gave rules such that:
pi di menD via pi di menD fa meno,

pi di menD via menD di menD fa pi, etc.
corresponding to (+i)( +i) = -1, (+i)( -i) = 1, etc.
2.2. 7 Frant;ois Viete

In a text published after his death, in 1615, Viete gave solutions of equations
of degree 3 and 4. For the cubic equation
A 3 +3BA = 2Z,
which we write here with our notation, but using his original letters, with
A as the unknown, he introduced a new unknown E such that EB =
E(A + E), which comes down to solving the equation x 3 + px + q = 0 with
the variable change x = (p/3y) - y, giving
A3 + 3AE(A + E) = 2Z,
a quadratic equation in E3. This makes it possible to compute E, then A,
by means of a single extraction of a cube root; the method is essentially
Cardan's.
2.3 Quartic Equations

Cardan gave a method for these equations in Chapter XXXIX of the Ars
Magna; he says that it was discovered by his student Lodovico Ferrari. It
consists in using a translation to bring the equation to the form
X4 + px 2 + qx + r =0
(Cardan, who rejected negative numbers, only gives a few cases of this).
Set z = x 2 + y, obtaining
Z2 =x4+2x2y+y2 = -px 2 -qx-r+2x 2y+y2 = (2y-p)x 2 _qX+y2_ r .

Choose y so that the right-hand term is ofthe form (Ax+B)2, by ensuring
that its discriminant vanishes, i.e.
q2 _ 4(y2 - r)(2y - p) = o.
This gives a cubic equation (which later came to be called a resolvent); one
of its roots can be found by the method of §2.2.4, giving
x 2 = -t ± (Ax + B),
and four values for x.
In the case where the right-hand term is not of degree 2, it is because

y = p/2, and then (*) shows that q = 0; the equation is biquadratic, which
we know how to solve.
In his 1615 text, Franc;ois Viete gave a clear exposition of Ferrari's
method.
Cardan detested introducing equations of degree higher than 3, because
equations of degrees 1, 2, and 3 concerned segments, areas, and volumes
and he asserted that "nature does not allow us to consider others" .
Here is another method, using indeterminate coefficients, which dates
back at least to Descartes (1637). If a, b, e, d are such that
X4 + px 2 + qx + r = (x 2 + ax + b)(x2 + ex + d),
we check (see Exercise 2.7) that a 2 is the root of a cubic equation and that
b, c, d depend rationally on a.
Exercises for Chapter 2

Exercise 2.1. Irrationality of roots of rational numbers
Let k > 1 be an integer, and let a and b be positive relatively prime

integers with no factors of the form d k for integers d > 1. Show that
{If is not a rational number.
Exercise 2.2. Cubic equations and Cardan's formulas
1) Solve the equations x 3 + 3x = 10, x 3 + 21x = 9x 2 + 5, x 3 = 7x + 7 by

Cardan's method or Viete's method.
2) Simplify the following expressions, where the roots are taken in JR,
and compare them with Cardan's formulas.
Exercise 2.3. Simplification of radicals in Cardan's formulas
If a cubic equation has an integral root, it often happens that Car-

dan's formula gives an expression with cube roots whose simplifica-
tion is not at all obvious. Tartaglia already noticed this problem in
1540, and we showed earlier how Bombelli worked on one example
(see §2.6). Let us consider what happens in the case of equations
with rational coefficients.
1) Show that if we have p, q, r, SEQ such that q, S > 0 and q is not a

square in Q, then the equality p + y'q = r + v's implies that p = r
and q = s.
2) Let a and b be rational numbers such that b > 0 is not a square in Q.
z
Suppose there exist rational numbers y and such that Va
+ Vb =
Y + vIz·
a) Show that Va - Vb = y - vIz.

b) Show that c = .ija 2 - b is rational.
e) Show that the equation x 3 - 3cx - 2a = 0 has a unique rational
root (use §3. 6. 4 below); compute y and z in terms of this root
and c.
3) Conversely, if the equation x 3 -3cx-2a = 0, with rational coefficients,

has a rational root and two non-real roots, show that there exist
rationals y and z such that Va + Vb = y+y'z, where b = a 2 _c3 > O.
4) Does this result make it possible to simplify the expression given by
Cardan's formulas for the roots of x 3 + px + q = 0 (with p and q
rational), when one of the roots is rational and the others are non-
real?
5) Simplify the following expressions, using the above; all roots are taken
in lR.:
\j10 - v'108,
Ff1f
2ff «1- ~ff.
1+ -
3
-
3'
Exercise 2.4. Cubic equations, irreducible ease, Cardan's method

This problem concerns the solution of the equation x 3 + px + q = 0
in the case where p and q are real and the discriminant is ~ O. The
equation has three real roots but Cardan's formulas lead to roots of
non-real numbers.
1) Let a be a solution of the equation. Compute the other two solutions
as functions of a and p.
2) Check the following text by Cardan for the solution of x 3 + 60 = 46x:
"A solution is 6. To find the others, raise 3, half of the first solution,
to the square; this gives 9 which, multiplied by 3, gives 27. Subtract
27 from 46, leaving 19. Subtract 3, half of the first solution, from the
square root of this number: you obtain the second solution Kg - 3.
By the same method, if you found Kg - 3 as a first solution, the
other solution will be 6."
Exercise 2.5. Cubic equation, irreducible case, Viete's method

This problem concerns the solution of the equation x 3 + px + q =0
in the case where p and q are real and the discriminant is ~ O.
1) Show that we can reduce to an equation of the form y3 - 3y = 2u,

with u E ffi. and lui::; l.
2) Solve this equation by setting v = arccosu.
3) Solve X 3 - 6X - 4 = 0 by this method.
COMMENTARY. - Viete's method shows the relation between the irreducible

case and the trisection of the angle (there is an analogy with the method of
Charles Hermite for equations of degrees 5 and 6, based on the division of
elliptic functions). In the example in 3), Cardan's formulas lead to radicals
of non-real numbers.
Exercise 2.6. Seventh roots of unity
Set (; = e2i7r /7 and a = 2 cos 27r .

7
1) Give a quadratic equation satisfied by (; over Q[a].
2) Find an irreducible cubic polynomial in Q[X] which admits a as a
root.
Exercise 2.7. Quartic equation and Descartes' method

By translation, we first reduce to the case of a quartic equation with
no third-degree term
X4 + px2 + qx + r = O.
Then, if this equation has a linear term, we look for a factorization
of the form
X4 + px 2 + qx + r = (x 2 + ax + b)(x 2 + ex + d).
1) Show that a 2 is a root of a cubic equation and that b, c, d are rational
functions of a.
2) Deduce the algebraic solution of the quartic equation from this.
3) Solve
X4 - 4x 2 - 8x + 35 0,
X4 - 17x2 - 20x - 6 O.
4) Show that if p, q, r are real, we can choose a, b, c, d real.
COMMENTARY. - Let us quote Descartes: "Au reste, j'ai omis ici les
demonstrations de la plupart de ce que j'ai dit, acause qu'elles m'ont semble
si faciles que, pourvu que vous preniez la peine d'examiner methodiquement
si j'ai failli, elles se presenteront a vous d'elles-memes; et il sera plus utile
de les apprendre en cette fa<.;on qu'en les lisant."l
The examples in 3) are those of Descartes. Question 4) is a result of
Euler (1749) in his work on the decomposition of polynomials in R.[X] into
products of linear or quadratic factors.
Solutions to Some of the Exercises

Solution to Exercise 2.1.
If there exist positive and relatively prime integers x and y such that
x/ y = {jO]b, then we have bx k = ayk. As x is prime to y, it must divide
a, so x = 1. Similarly, y = 1 and we are done.
1) To solve the equation x 3 +px+q = 0, we know that we need to determine

u and v such that u 3 + v 3 = -q and uv = -p/3, and then set x = u + v,
ju + pv, j 2 u + jv.
For x 3 + 3x - 10 = 0, we obtain u = \15 + VW, v = \15 - VW (the
roots are taken in R.).
For x 3 + 21x = 9x 2 + 5, we set y = x - 3, which leads to solving the
equation y3 + 4 = 6y; we obtain the root y = 2 and then divide by y - 2,
avoiding the extraction of cube roots. Like Cardan, we find three real roots,
namely x = 5 and x = 2 ± V3.
For x 3 -7x - 7 = 0, we obtain
u = \/7/2 + 7i/18V3 and v = \/7/2 - 7i/18V3,

where the arguments of the cube roots are chosen with opposite signs, since
we must have uv = 7/3.
2) To find the equation having a as a root, we can compute a 3 and compare
it with a. We can also compare the form of a with the general solution of
1 Besides, I left out the proofs of most of what I said here, because they appeared so
easy to me that if you just take the trouble to check methodically whether I erred, they
will present themselves to you naturally, and it will be more useful to you to learn them
this way than by reading them.
the cubic equation, which leads us to set q = -20; then 108 = q2 /4 + p3 /27
gives p = 6. The equation x 3 + 6x - 20 = 0 has 2 as a root, so when we
divide it by (x - 2), we obtain the other roots -1 ± 3i. The only real root
is 2, so we find that 0: = 2.
Similarly, we find f3 = 1.
1) We have x 3 +px+q = (x - a)(x 2 +ax +p+a2). The roots of the second

factor are real; they are given by -~ ± J-3(~)2 - p.
2) Cardan uses his formula on his example with the sign + for the root. To
check the last sentence, we set a = vi9 - 3 and note that -3(a/2)2 - p =
(9 + vi9)/2) 2 .
COMMENTARY. - Cardan gave no general method for this type of equation;
he did not use his formula and could only guess at one root in order to find
solutions for the remaining quadratic equation.
1) Setting x = o:y, we are led to take 0: = J -p/3, so that 2u = _q/0: 3 ;

we then check that lui::; 1.
2) The formula cos 38 = 4 cos3 8 - 3 cos 8 gives
2cosv/3, 2 cos «v/3) + (27r/3)), 2 cos «v/3) + (47r/3))

as roots of the equation y3 - 3y = 2 cos v.
3) We find 0: = V2, U = V2/2, v = 7r/4 and the roots are
V2 37r =-2, V2 177r .

b=22cos 4 c=22cos 12
Since -2 is a root, we can also write x 3 - 6x - 4 = (x + 2)(x 2 - 2x - 2),
which gives a = 1 + V3, c = 1 - V3.
1) ( is a root of the quadratic equation x 2 - O:X + 1 = o.

2) We have the equation 0: 3 + 0: 2 - 20: - 1 = 0, of which 1 and -1 are not
roots.
1) By identification, we successively find
a+e o (2.4)
ae+b+d p (2.5)
ad + be q (2.6)
bd r. (2.7)
We deduce that e = -a, b + d = p + a 2 and a(d - b) = q. Thus a =1= 0 since

q =1= 0, so we obtain b + d and b - d, which gives band d; plugging them
into (2.7) gives
(ap - q + a 3 )(ap + q + a 3 ) 4ra 2, (2.8)

a6 + 2pa 4 + (p2 _ 4r)a2 _ q2 O. (2.9)
The last equation is a cubic in a 2 (it is a resolvent, corresponding to the

choice of -u, -v, -w in §10.8 below). We obtain six values of a, each of
which gives a factorization. This is normal since a is the sum of two of the
four roots of the equation and (~) = 6.
2) Once the factorization is obtained, it remains only to solve quadratic

equations.
3) With a = 4, we have
X4 - 4X2 - 8X + 35 (X2 - 4X + 5)(X2 + 4X + 7),
X4 -17X2 - 20X - 6 (X2 - 4X - 3)(X2 + 4X + 2);
the rest is easy.
4) The resolvent must have a positive real root since the value of the left-
hand side of (2.9) is < 0 for a = 0 and> 0 for a sufficiently large.
3
Symmetric Polynomials
In this chapter, we first give the basics on symmetric polynomials, and then
present the notions of resultant and discriminant.
3.1 Symmetric Polynomials

3.1.1 Background
Let A be a commutative ring with unit. The A-algebra A[X1 , ... , Xnl of
polynomials in n indeterminates and coefficients in A has the following
universal property: for every A-algebra B given by a ring homomorphism
f : A -> B and every map h : {I, ... , n} -> B, there exists a unique
homomorphism of A-algebras cP : A[X1 , .•. , Xnl -> B such that cp(Xi ) =
h(i) for all i in {I, ... , n}, and cp(a) = f(a) for all a in A. In other words,
the universal property asserts that in order to construct a homomorphism
cp of A-algebras from A[X1 , .•. , Xnl to another A-algebra, it suffices to give
the images of the indeterminates, and there is nothing further to check.
In the case where n = 1 (we denote the indeterminate by X) and the
map h is defined by h(l) = b, the homomorphism cp : A[X] -> B is defined
by
cp (:LakXk) = :Lf(ak)bk .
For every element a of the group Sn of permutations of the set {I, ... ,n},
the above remarks prove that there exists a unique homomorphism of A-
algebras CPa : A[X1, ... ,Xnl -> A[X1, ... ,Xnl (often simply denoted by a)

26 3. Symmetric Polynomials
making the diagram in Figure 3.1 commutative (the notation "can" means
that the arrows are canonical).
In other words, !Pa(Xi ) = Xa(i) for i = 1, ... ,n, and more generally,
If A is an integral domain with fraction field K, the homomorphism !PO'

extends to the field K(X I , ... , Xn) ofrational functions in Xl,"" Xn with
coefficients in K. Recall that an element of this field is represented by the
quotient of two polynomials in A[X I , ... ,Xn ], with denominator not equal
to zero.
~
can
1,an
A A[Xl' .... '~] {l, ..... ,n}
can
""1 /
{1, ..... ,n}
FIGURE 3.1.
3.1.2 Definitions
A polynomial in A[XI, ... ,Xnl is said to be symmetric if for all a in Sn,
we have !Pa(P) = P.
If A is an integral domain with fraction field K, a rational function P /Q
in the field K(X I , ... , X n ), with P, Q E A[X1 , •.. , Xnl and Q #- 0, is said
to be symmetric if for all a in Sn, we have !PO' (P/Q) = P/Q.
EXAMPLES. - The following polynomials are symmetric in A[XI' X 2, X31:
Xl +X2 +X3,
X I X 2 X 3,
XrX2 + X~X3 + xlXI + X~Xl + X1X2 + XrX3,
but XrX2 + X~X3 + X1X I is not.
REMARKS. - The symmetric polynomials generate an A-subalgebra of the

algebra A[X1, ... , Xnl·
If a polynomial P in A[Xl' . " ,Xnl is symmetric and if a(Xl)kl ... (Xnln
is a monomial in P, then for all a in Sn, a(Xa(l))kl ... (Xa(n))k n is a mono-
mial of P.
3.2 Elementary Symmetric Polynomials 27
3.2 Elementary Symmetric Polynomials

3.2.1 Definition
Let n be an integer 2: O. For every integer k :S n, the elementary symmetric
polynomial of degree k in A[Xl , ... , Xn], which we denote by Sk, is given
by the formula
Sk = L
HC{l, ... ,n},IHI=k
(II
iEH
Xi)'
In other words, H runs over the set of subsets of k elements of {I, ... ,n}
and Sk is the sum of the products of the Xi for i in H. For k > n, we set
Sk = O.
EXAMPLES. - For n = 0 : So = 1 and Sk = 0 for k 2: l.

For n = 1 : So = 1, Sl = Xl, Sk = 0 for k 2: 2.
For n 2: 2 : So = 1; Sl = LXi; S2 = L XiX j . Thus, S2 is just the
lSisn lSi<jSn
sum of the products of pairs of Xi' More generally, Sk is the sum of the
products of k of the Xi for k :S n, and in particular,
Sn = II
lSisn
Xi'
REMARKS. - In general, the integer n is implied by the context and we will

not give it explicitly.
The Sk are homogeneous polynomials, i.e. polynomials all of whose mono-
mials are of the same degree, namely k.
3.2.2 The Product of the X - Xi; Relations Between

Coefficients and Roots
PROPOSITION. -
1) For all n 2: 0, we have
in the ring I£[Xl' ... ,Xn][X],
2) If P(X) = L akX k is a monic polynomial of degree n with roots

OSksn
XI, ... ,Xn belonging to a field K, then an-k = (-l)k sk (Xl,." ,xn)
for 0 :S k :S n.
PROOF. -
1) Let us show it by induction on n. The formula holds for n = 0, 1.

Suppose it holds for an integer n; let Sk denote the elementary sym-
metric polynomials in Z[Xl"'" Xn] and tk those in Z[X 1 , ... , X n+1].
We have
(X - Xi) = [ II (X - Xi)] (X - Xn+d

l::;i::;n
L (-l)kskxn-k] (X -
[O::;k::;n Xn+l)
Xn+l + L (_l)k[sk + Sk_IXn+l]X n+1 - k

I::; k::; n
On the one hand, SnXn+1 = t n+l . On the other hand, for 1 ::; k ::; n,
separating the parts with k elements of {I, ... , n + I} into those that
contain n + 1 and those that do not, we see that Sk + Sk-IXn+1 = tk'
This gives the desired result.
2) By the universal property of Z[Xl, ... , Xn], there exists a homo-

morphism K[X] such that <p(X) = X and
<p(Xi ) = Xi for 1 ::; i ::; n. Since
EXAMPLES. - Let Xl, X2, X3 denote the roots of the polynomial X 3 + aX2 +
bX + c. We have
Xl + X2 + X3 -a,
XIX2 + X2X3 + X3XI b,
XIX2X3 -c.
The analogous formulas for second-degree polynomials are well known.

3.3 Symmetric Polynomials and Elementary Symmetric Polynomials 29
3.3 Symmetric Polynomials and Elementary

Symmetric Polynomials
3.3.1 Theorem
Let A be a commutative ring with unit (in particular, our result will hold for
fields), and let P be a symmetric polynomial in A[Xl, ... ,Xn ]. Then there
exists a unique polynomial T of A[X I , ... , Xn] such that T(SI,"" sn) = P.
EXAMPLES. - 1) X; + xi = (Xl + X 2)2 - 2X I X2 = T(SI' S2) in Z[XI' X 2]

with T(SI' S2) = s'i - 2S2·
2) xl X 2 + XIX~ = X IX 2(X; + Xi) = si S2 - 2s~.
COMMENTARY. - After the remarks of Girard and Newton (see §3.4), this
result was used freely throughout the 18th century; Lagrange called it "ob-
vious in itself". It appears to have been stated and proved independently
by Waring and Vandermonde in 1770. This theorem can be considered as
a small part of the results of §8.5.
PROOF. -
1) We prove the existence of the polynomial T by giving an algorithm

to compute it.
To begin with, note that it suffices to show the result for symmetric
polynomials SCM) defined by a monomial M = (XI)kl .,. (Xn)kn, by
setting
SCM) = L U with A(M) = {ifa(M),a E Sn},

UEA(M)
since every symmetric polynomial is a linear combination of such

polynomials.
So let P be a symmetric polynomial defined by a monomial
as above. Choose a total ordering n on the set of indeterminates
{Xl,'" ,Xn }, for example Xl > ... > X n , and order the set of
monomials of P according to the lexicographical order on the n-tuples
of powers, i.e.
if there exists r < n such that for i S r, k i = li and kr+l > lr+l.
For example, (Xt)2(X2)2 X3 > (Xt)2 X 2 (X3 )2 > X I X 2(X3)3 >
(X3)5.
For monomials M, M', N, N', we easily see that M ;::: M' for
n implies that M N ;::: M'N, and this implies the following property:
if M ~ M' and N ~ N', then MN ~ M'N', since MN ~ M'N ~

M'N'.
Let M = (X1)k1 ... (Xn)k n be the largest monomial of P for the
ordering n. We have k1 ~ '" ~ kn, since otherwise the transposition
exchanging two indices i and j such that i < j and k i < k j would
transform M into another monomial of P larger than M.
Consider a polynomial of the form (Sl)11 '" (Sn)ln. In this product,
according to the above property, the largest monomial for n is the
product of the largest monomials of Sl, ... ,Sn, raised to the powers
h, ... , In, i.e. if we have
(X 1 )1,(X1X2 )12 (X 1··· X n )In -- XI,+···+lnX l2+··+ ln Xln
'" 1 2 '" n'
This monomial will be equal to M if k1 = h + .. ·+In, k2 = b+·· ·+In,

etc., i.e. if
In kn,
In-1 kn- 1 - kn,
II k1 - k2.
Set PI = P - (sd ' '" (sn)ln, where the li have the above values;
PI is a symmetric polynomial that is either zero or whose largest
monomial for n is strictly less than M.
If PI is zero, we have written P in the desired form
P(Xl, ... , Xn) = T(Sl, ... , Sn),

where T is defined by T(Xl, ... , Xn) = (Xd ' ... (Xn)ln. Otherwise,
PI is a linear combination of polynomials of the form SCM) and we
start the same procedure for each of them, until we obtain the zero
polynomial and the desired expression of P (see the example below).
2) Let us now prove the uniqueness ofT. Suppose there exists a non-zero
polynomial Tin E such that T(Sl, ... , sn) = O.
There exist two monomials in T, say
M = a(Xd ' ... (Xn)ln and N = b(X1)m1 ... (Xn)mn,
such that the n-tuples (lih::;i::;n and (mih::;i::;n are different.
Consider the polynomials M(Sl,"" sn) and N(Sl,"" sn). Their
largest monomials in Xl, ... ,Xn are
a(X1)ll+··+ ln '" (Xn)ln and b(x 1)m 1+··+mn ... (Xn)mn;
they have n-tuples with different exponents.
It follows that there exists a unique monomial M of T giving the
largest monomial in Xl,"" Xn of T(Sl' ... ,sn), which contradicts
the hypothesis T(Sl,"" sn) = O. 0
3.3 Symmetric Polynomials and Elementary Symmetric Polynomials 31
EXAMPLE. - Let us express

P(X, Y,Z) = y 3Z + YZ 3 + Z 3X + ZX 3 +X3y +Xy3
in terms of the elementary symmetric polynomials of Z[X, Y, Z]. First, we
have
SI=X+Y+Z, S2=XY+YZ+ZX, S3=XYZ,
If we select the order X > Y > Z, then the largest monomial of Pis X 3 Y,
which leads to setting PI = P - si
S2. After doing the computations, we
obtain
PI (X, Y, Z) = -5(X2y Z + Xy2 Z + XY Z2) - 2(X2y2 + y2 Z2 + Z2 X2).
The largest monomial of PI is X2Y2, which is indeed less than X 3y' Be-
cause PI is a linear combination of two polynomials of the type defined
above, we can work on each of them separately. The first part is clearly
equal to -5S1S3 (which is given immediately by the algorithm); for the
second part, we form X2Y2 + y2 Z2 + Z2 X2 - (S2)2, etc. After all compu-
tations, we obtain
The extension of Theorem 3.3.1 to rational symmetric functions is easy;

it is the object of the following proposition.
3.3.2 Proposition
Let A be an integral domain, and let P, Q be polynomials in A[X 1, ... , Xn],
with Q #- 0, such that P / Q is a symmetric rational function. Then there
exist polynomials 8 and T in A[XI, ... ,Xn ] such that
8(SI,' .. ,sn) P
T(SI, ... ,Sn) = Q'
PROOF. - If Q is symmetric, then P is symmetric and the proposition is a

consequence of Theorem 3.3.1. If Q is not symmetric, form the set
The product II q is a symmetric polynomial of which Q is a factor. Because

we have qEE
P II q
P qEE-{Q}
Q II q
qEE
and we are again in the case where both numerator and denominator are
symmetric. <>
3.3.3 Proposition
Let P be a symmetric polynomial in n variables. If S is a polynomial of
degree n having roots al, ... , an, the numbers P(al, ... , an) belong to the
ring generated by the coefficients of s.
PROOF. - This is an immediate and important consequence of Theorem

3.3.1 and Proposition 3.2.2. <>
3.4 Newton's Formulas

PROPOSITION. - For every integer d 2: 1, set
Pd = L (Xi)d
l~i~n
in A[Xl , ... , Xn].

For d 2: 1, this gives
Pd = L (-I)k-l SkPd _k + (-I)d- ldsd.

l~k~d-l
EXAMPLES. -
d=I:Pl=Sl;
d = 2 : P2 = SIPI - 2S2; for example X2 + y2 = (X + Y)(X + Y) - 2XY;
d = 3 : P3 = SlP2 - S2Pl + 3S3; for example
X 3 + y3 + Z3 = (X + Y + Z)(X2 + y2 + Z2)
-(XY + YZ + ZX)(X + Y + Z) + 3XYZ.
COMMENTARY. - This formula, which was stated by Newton around 1666

and published in 1707, makes it possible to successively compute the poly-
nomials Pd given by the sums of the d-th powers of the Xi. Recall that
Sk = 0 for k > n, which truncates the formula when d > n.
In 1629, Girard gave the expression of the Pd as a function of the Sk
for d ~ 4 (Figure 3.2); in his charming terminology, the first "meslE~" and
second "mesle" are Sl, S2. In his example, X4 + 35X 2 + 24 = 10X 3 + 50X;
the terms are placed on both sides of the equal sign in order to avoid
negative signs. This is the first time that roots, including impossible ones,
appear within formulas exactly like ordinary numbers.
3.4 Newton's Formulas 33
Exempk.
f A premier mefie.
Soit lC B fecond.
troifiefme.
~ quatriefme.
&c.'
fA ~~ folutions
A q _ B 2.
a10rtm toll- j ~ tuarez
te fone d'e~
'iuadon. '. A cub - AB3 C 3 +
0;
.!! ubes
lAq q _ AqBf+AC++Bqz-Df~. quare..quarez
Et pourmjeux: expliquerlc tout, foit 1(3)+ 3S'@+2f ef~alea
10 (3)+ S'o Q): r
l'ordrcdesmeflezefl: 1 0 ,35. o. 2 4 pour A, ~.l,;.D!
cy..detIos : tellement que 10 efl:voiremeot la fomme des [olutIons qUI
font (1,2.3, .... J Or Aq - B 2., c'eft~direlequarrede lo-deux
fois 3 f .c'cfl: la fomme des quarrez , & alnfi du refte;
FIGURE 3.2. Excerpt from the book Invention nouvelle ... by Girard, 1629
Setting X = Xi in formula (3.1) from Proposition 3.2.2 gives
Xr = L (_I)kskXr-k.
l~k~n
This gives the xt

for d 2': n and the Pd by summation. It would remain to
give a proof for d < n, which is possible.
The following method gives a proof valid for all d 2': 0, in the
framework of the A-algebra F = A [[Xl , ... , Xnll of formal power series in
n indeterminates with coefficients in A.
PROOF. - Consider the map D : F ---+ F defined on the homogeneous parts
of degree k of F by D(u) = ku, and more generally, for a formal power
L
series u = Uk written as the sum of its homogeneous parts of degree k,
k?,:O
by D(u) = L kUk·
k?,:O
The operation D satisfies D(u+v) = D(u) +D(v) and D(uv) = D(u)v+
uD(v). This D is an example of what we call a derivation in a ring; here,
Indeed, if u = L Uk and v = L vk are the expressions for u and v as

k?,:O k?,:O
sums of their homogeneous parts, we have
D(uv) D[ (~ Uk) (~ Vk) ]

D[~ (fm UkVl)]
2: m( 2: UkVl)
m2:0 k+l=m
2: 2: (kUkVl + IUkVl)
m2:0k+l=m
m2:0k+l=m m2:0k+l=m
D(u)v + uD(v).
It follows that the derivation of the product of r formal power series
UI, ... ,Ur is given by
D(UI ... u r ) = D(UI)(U2'" u r ) + ... + (UI ... ur-I)D(u r );
so for invertible power series, we have
D(UI ... u r ) D(ud D(ur )
+ ... + -
-'--=----'--'- = - - -.
UI'" Ur UI Ur
Let us apply this formula to the n power series Ui = I + Xi, I ~ i ~ n,
noting that D(I + Xi) = Xi' We obtain
D( II (I+Xd)
lS;iS;n
= -Xl
- - + ... + -Xr
-.
II
---==-----
(1 +Xi ) 1 +XI I +Xr
Recall that the inverse of I - X is the series 2: Xk. Thus,

k2:0
Setting X = -1, formula (3.1) gives IT (1 + Xi) = L Sk, and since

lS;iS;n OS;kS;n
Sk = 0 for k > n, this gives IT (1 + Xi) = 2: Sk·
lS;iS;n OS;k
As D (2: Sk) = 2: kSk, we can now finish the proof:

OS;k lS;k
3.5 Resultant of Two Polynomials 35
[ II
l::;i::;n
(1 + Xi)] [~+
1 + Xl
... +~]
1 + Xr
l~n Sk] l~n~(-l)k(Xi)k+1

l~n Sk] ~(_l)k-lpk
L L (-l)I-I SkPI
l::;dk+l=d
L[(-l)d-Ipd + L (-l)d-k-ISkPd_k] ,
l::;d l::;k::;d-l
given that So = 1. We conclude by identifying the homogeneous components

of this last equality. 0
3.5 Resultant of Two Polynomials

The resultant of two polynomials was introduced by Newton in certain
special cases (1707), and then by Euler (1748, 1764) and by Bezout (1764).
3.5.1 Definition
Let K be a field contained in an algebraically closed field C (see §16.1).
Let F and G be two polynomials in K[X], of degree m and n respectively.
Write
F(X) = a II (X - Xi) and G(X) = b II (X - Yi)
l::;i::;n
inC[X].
The resultant of the two non-zero polynomials F and G is defined to be
the product
II
I::; i::;m,l::;j::;n
If F = 0 or G = 0, we set Res(F,G) = O.
REMARK. - The result is independent of the choice of C, as we will see. We

will write Resx(F, G) whenever it is necessary to distinguish the variable
with respect to which the resultant is taken.
3.5.2 Proposition
1) The resultant of two polynomials in K[X] is zero if and only if the
two polynomials have a common root in C.
2) The resultant of two polynomials in K[X] is zero if and only if the

two polynomials have a greatest common divisor which is neither a
constant nor zero in K [X].
3) The resultant of two non-zero polynomials in K[X] is an element of

K, and we have
Formula (1): Res(F, G) = an II G(Xi),
l::;i::;m
Formula (2): Res(G, F) = (-l)mnRes(F, G),
Formula (3): Res(F, G) = (_l)mnbm-deg(R)Res(G,R)

if the Euclidean division of F by G gives F = GQ + R with deg(R) <
deg(G). If R = 0, then Res(F, G) = 0.
Formula (4): Res(F, b) = bm for a constant polynomial b.
PROOF. -
1) If F and G are non-zero and if Res (F, G) = 0, then there exist i and
j such that Xi = yj. The converse is obvious.
2) Suppose that F and G are non-zero, as the result is obvious if one

or both of them is zero. Because F and G have coefficients in K, we
know that we can compute their greatest common divisor S in K[X],
and that S is also the greatest common divisor of F and Gin C. If
F and G have a common root Xi in C, then X - Xi divides F and G,
so it divides S; this means that S is not a constant. Conversely, if S
is not a constant, then every root of S in C is a common root of F
and G. Now part 1) suffices to conclude.
3) Formula (1):
Res(F, G) = an II [b II (Xi - Yj)] = an II G(Xi)'

l::;i::;m l::;j::;n l::;i::;m
Formula (2):
II
l::;i::;m,l::;j::;n
II
l::;i::;m,l::;j::;n
Formula (3): formulas (1) and (2) give
Res(F, G) (-l)mnRes(G, F)
(-l)mnRes(G, GQ + R)
3.6 Discriminant of a Polynomial 37
(_l) mn bm II [G(Yi)Q(Yi) + R(Yi)]

l:5i:5n
(_l)mnbm-deg(R)Res(G, R).
Formula (4): this follows from formula (1).

The process of Euclidean division produces only elements of K,
so the resultant Res(F, G) lies in K by definition. <>
REMARKS. - Formulas (1)-(4) give an algorithm for computing the resul-

tant.
Historically, the resultant was introduced as a determinant (see Exer-
cise 3.3). This approach is very natural and actually renders the result of
Proposition 3.5.2 c) more precise. Indeed, if
O:5k:5m O:5k:5n
then
Res(F,G) E A[ao,.·.,am,bo, ... ,bn ].
But to compute this determinant quickly, in general, one needs to revert
to the method presented above.
3.6 Discriminant of a Polynomial

3.6.1 Definition
Let K be a field contained in an algebraically closed field C. The discrim-
inant D(P) of a non-zero polynomial P in K[X] of degree n, with leading
coefficient a, is defined by
(_1)n(n-l)/2Res(P PI)
D(P) = ' .
a
3.6.2 Proposition
The discriminant D(P) of a non-constant polynomial P is an element of
K, which is equal to zero if and only if P is a root of multiplicity greater
than or equal to 2 in C.
PROOF. - The fact that D(P) E K follows from Proposition 3.5.2 c). By
Proposition 3.5.2 a), the discriminant of P, which is the resultant of P and
pI, is zero if and only if P has a common root with p' in C, i.e. if P has
a root of multiplicity greater than or equal to 2 in C. <>
3.6.3 Formulas
1) D(aX2 + bX + c) = b2 - 4ac.
3) If F(X) =a II (X - Xi), then D(F) = a2n - 2 11 I::;i<j::;n(xi - Xj)2.

l::;i::;n
PROOF. - The proofs are contained in Exercise 3.5. <>
3.6.4 Polynomials with Real Coefficients: Real Roots and Sign

of the Discriminant
Second-degree polynomials
Let a, b, c be real numbers. We have D(aX2 + bX + c) = b2 - 4ac =

a 2(xl - X2)2, where Xl and X2 denote the roots ofaX 2 + bX + c in Co If
Xl and X2 are distinct real numbers, we have b2 - 4ac > o. If they are not
real, then they are conjugate in C, Xl - X2 is a purely imaginary number,
and b2 - 4ac < o.
Cubic polynomials
Let p, q be real numbers, and let a, b, c denote the roots of the polynomial
X 3 + pX + q in Co We have
If the three roots are distinct real numbers, then _4p 3 - 27 q2 > 0; this is
the irreducible case (see §2.2.6). The second case is when two of the roots,
say a and b, are not real numbers, but complex conjugates in C; then a - b
is purely imaginary and its square is negative: (a - C)2 and (b - C)2 are
conjugate and their product is strictly positive, so _4p 3 - 27q2 < O.
These results are summarized in Table 3.1.
D>O D<O
Degree 2 2 real roots 2 non-real conjugate roots
Degree 3 3 real roots 1 real root,
2 non-real conjugate roots
TABLE 3.1. Real roots and sign of the discriminant

Exercise 3.1. Elementary symmetric polynomials
1) Write the following symmetric polynomials as polynomials in the el-

ementary symmetric polynomials:
a) X2y2 + y2Z2 + Z2X2,

b) X3Y + Xy3 + X 3 Z + X Z3 + y3 Z + Y Z3.
2) Use Newton's formulas to compute the sum of the fourth powers of

the roots of the polynomial (X -l)(X -2)(X -3)(X -4); check your
result directly.
3) Compute the sum of the seventh powers of the roots of X 3 + pX + q.
4) Supposing that q =F 0, determine the monic polynomial whose roots

are the inverses of the squares of the roots of the polynomial X 3 +
pX+q.
5) Let (XI, ... ,Xn ) and (YI, ... ,Yn) be two n-tuples in K n such that
there exists no permutation a in Sn such that (Yo-(l), ... , Yo-(n)) =
(Xl, ... , xn). Show that there exists an elementary symmetric polyno-
mial Sk for some 1 :::; k :::; n such that Sk(XI, ... , xn) =F Sk(YI, ... , Yn).
Exercise 3.2. Tschirnhaus' method
Consider the polynomials P(X) = X 3 + pX + q with pq =F 0 and

Q(X) = X2 + alX + ao, where all coefficients lie in a subfield K of
C.
Suppose that P has a root X and set Y = Q(x).
The reader who quails before the rather lengthy computations in-
volved in this exercise may employ the following shortcut: plug the
results given in the solution to this exercise back into the exercise
and check that they work.
1) Write the linear polynomial R which vanishes at X in terms of Y and

the coefficients of P and Q.
2) a) Compute Res(P, Q - y).

b) Determine elements ao and al such that Y satisfies a relation of
the form y3 = a 3 .
3) Using the above, find the roots of P.

COMMENTARY. - In 1683, Tschirnhaus proposed his method for solving an

equation of degree n by using a change of variables of the form Y = Q(X),
where Q is a polynomial of degree n - 1, in order to reduce to an equation
of the form yn = an whose solutions are known; it then remains to solve
the equation Y = Q(X) which is of degree n -1. By induction, this method
was supposed to give solutions of polynomial equations of all degrees. But
it came to be understood latS!r that the determination of the coefficients of
Q led to equations of degree> n unless n = 2 or 3, and that for n = 4, one
obtains an equation of degree 6 which factors into two equations of degree
3.
Exercise 3.3. The resultant as a determinant
Let A be an integral domain. Consider the polynomials
F(X)
G(X)
in the ring A[X1 , ••. , X m , Yb ... , Yn]. Let us define an (m + n) x

(m + n) matrix ~ = (djk) (here j and k denote the row and column
indices respectively) with coefficients in A[X1 , ••. , X m , Y1 , .. ·, Yn ], as
in Figure 3.3.
~"
D= 'b. 'b. n
FIGURE 3.3. The matrix ~
a) In the n first columns (i.e. 1 ::; k ::; n), we set djk = am-j+k for
k ::; j ::; k + m, otherwise djk = O.
b) In the m last columns (i.e. n+1 ::; k ::; m+n), we set dj,k = bk-j
for k - n ::; j ::; k, otherwise djk = O.
Define D to be the determinant of the matrix ~. Our goal is to check

that D = Res(F, G) by the following steps.
1) Let N be an integer, and let V (TI' ... , TN) denote the determi-
nant (known as the Vandermonde determinant) of the square ma-
trix M(TI , ... , TN) of dimension n with coefficients in Z[TI ,···, TNJ
whose k-th row is given by (Tk)N-1, ... ,(Tk)2,Tk,1. Recall how to
compute the Vandermonde determinant V(TI , ... , TN)'
2) Set M = M(YI , ... , Y n , Xl"'" Xm).By computing the determinant
det(M~) in two different ways, show that D = Res(F, C).
3) Show that Res(F, C) E A[ao, . .. ,am, bo, ... ,bnJ.
COMMENTARY. - In his memoir "A new way of eliminating unknown quan-

tities from equations" , presented to the Berlin Academy of Sciences in 1764,
Euler introduced the resultant by writing that if two polynomials
F(X) = L ak Xk and C(X) = L bkX k
O~k~m O~k~n
have a common root x, then there exist polynomials
O~k~m-l
such that
F(X) = (X - X)FI(X) and C(X) = (X - x)CI(X).
Thus FC I = FI C, and when this is expanded we obtain a system of linear
equations in the Xi and Yj with m+n equations and m+n unknowns, whose
determinant must be zero in order for there to exist a non-zero solution.
This determinant is exactly the one which we introduced above; of course,
Euler did not call it a determinant (this terminology was introduced by
Sylvester in 1840), and he only wrote the explicit formula for polynomials
of small degree.
Exercise 3.4. Computing the resultant

1) Let F, G I and C 2 be polynomials with coefficients in a field K. Check
that Res(F, C I C2) = Res(F, CI)Res(F, C 2 ).
2) Compute Res(aX2 +bX +c, dX +e) in three different ways: using the
formula using the roots of one of the polynomials, using the method
of Euclidean division, and computing a determinant.
3) Show that
Res(aX2 + bX +c, a'X2 + b'X + c')
(a'c - ac')2 - (ab' - a'b)(bc' - b'c)
( ac, + a , c - 2bb') 2 - 41 (b2 - 4ac ) (b'2 - 4a, c') .

Exercise 3.5. The discriminant of a polynomial
1) Compute D(aX2 + bX + c), D(X3 + pX + q).

2) Show that if F(X) = a II (X - Xi); then its discriminant satisfies
lSiSn
D(F) = a2n - 2 II (Xi - Xj)2.

lSi<jSn
3) Show that D(F) is real if F has real coefficients.
4) Let P be a polynomial with real coefficients and with roots that may
or may not be real but which are pairwise distinct. What does the
sign of D(P) imply about the number of real roots of the equation
P(X) = O?
5) Let us give an application of the notion of discriminant which is
beyond the actual scope of this book.
Equip the set E of polynomials of degree ::; n in qXl with the
topology of C n+ 1 by identifying the element (ao, ... , an) E C n+ 1 with
the polynomial P = L akXk.
OSkSn
a) Show that the polynomials having only simple roots form an
open set of E.
b) Deduce from this that the n x n matrices with distinct eigen-
values form an open set of the set Mn (C) of n x n matrices with
entries in C.
Exercise 3.6. Computating D(X n - 1 + X n - 2 + ... + 1) for n ~ 2
1) Compute D(xn - 1) for n ~ 1.
2) Compute D(X + I), D(X2 + X + I), D(X3 + X2 + X + 1).

3) Show that if P(X) = (X - Xl)P1(X), then D(P) = P1(Xl)2 D(P1).
4) Compute D(X n- 1 + X n - 2 + ... + 1) for n ~ 2.
5) Obtain the result of part 4) by using the formulas
D(P) =
(_1)n(n-l)/2Res(P PI)
a
' and Res(F, G) = an II G(Xi)'
lSiSm
Exercise 3.7. Descartes' Lemma
Let n ;:::: 1 be an integer, and let (xo, ... , xn) be a family of real
numbers.
°
We say that the family has a sign change at index i for 1 :::: i :::: n
if and only if there exists k E {O, ... , i-I} such that XkXi < and
Xl = ° for aU I such that k < I < i.
If P is a polynomial in JR[X], write c(P) for the number of sign
changes in the sequences of coefficients of P, and let r(P) denote the
number of strictly positive real roots of P.
°
1) Show that r(P) :::: c(P) for every polynomial in JR[X]. To do this,
reduce to the case where P(O) =I- and use induction: if
is a polynomial of degree n, set k

study the variations of P.
= inf{i Ii> °and ai =I- O} and
2) Give an upper bound for the number of strictly negatIve roots of P.
3) Determine the number of strictly positive roots, strictly negative

roots, and non-real roots of the polynomial X 8 + X4 + X - 1.
4) Let us study another situation. Let
P(X) = xn + L akXk
k-S;n-p-l
°
with P 2: and an-p-l =I- 0, and write s(P) for the number of real
roots of P. Show that
a) if P is even, then s(P) :::: n - p;

b) if P is odd and an-p-l > 0, then s(P) :::: n - p - 1; and
c) if p is odd and an-p-l < 0, then s(P) :::: n - p + 1.
COMMENTARY. - The subject of Exercise 3.7 is not directly related to

the subject of this book; it was stated without proof by Descartes, and
subsequently became the object of a great deal of discussion and many
more or less incomplete proofs, until Gauss gave a complete proof in 1828.
Any classical algebra text contains results of Sturm generalizing this lemma.

1) a) Set P(X, Y, Z) = X2y2 + y2 Z2 + Z2 X2. The polynomial P is sym-

metric in Z[X, Y, Z); its largest term for the ordering X > Y > Z is X2y2.
So we compute P - 8~, i.e.
-2(X 2yZ +Xy2Z +XYZ2) = - 28183.
This gives P = 8~ - 28183.
b) See §3.3.
2) We have
(X - I)(X - 2)(X - 3)(X - 4) = X4 - lOX 3 + 35X 2 - 50X + 24,
and the sum 14 + 24 + 34 + 44 = 1 + 16 + 81 + 256 = 354 confirms the
computation
PI 81 = 10,
P2 81P1 - 282 = 30,
P3 81P2 - 82P1 + 383 = 100,
P4 81P3 - 82P2 + 83P1 - 484 = 354.
3) Let Xl, X2, X3 denote the roots of X 3 + pX + q in C. By Euclidean
division, we obtain
X7 = (X 3 + pX + q)(X4 _ pX 2 _ qX + p2) + 2pqX 2 + (q2 _ p3)X _ p2q,
which gives
2pqP2 + (q2 - p3)81 _ 3p2q = _7p2q,
taking into account the identities 81 = °

and P2 = 81P1 - 282 = -2p.
We can also successively compute the Pi: we find 0, -2 p, -3q, 2p2, 5pq,
_2p3 + 3q2, _7p2q.
4) Let a, b, c denote the roots of X 3 + pX + q and 8 1 , 8 2 , 8 3 the values of

the elementary symmetric polynomials in 1/a2, 1/b2 , 1/c2 • Then
1 1 1 a 2b2 + b2c 2 + c 2a 2 p2
a2 + b2 + c2 = a2b2c2 - q2'
1 1 1 ~+~+~ ~
--+-+--=---:::-:--::-:::-
a 2b2 b2c 2 c 2a 2 a 2b2c 2
1
2".
q
The desired polynomial is X 3 - (p2/q2)X2 - (2p/q2)X - (1/q2).

5) The polynomials II (X - Xi) and II (X - Yi) are distinct since
l::;i::;n l::;i::;n
they do not have the same roots; thus their coefficients are not all equal.
The result follows from formula (3.1) in §3.2.2.
1) It suffices to divide P(X) by Q(X) - Y via Euclidean division. We find

the remainder R(X) = (p - ao + a l + y)X + q + aOal - alY·
2
2) a) The computation was begun in 1). We find that Res(P, Q - y) is

equal to
_y3 + (3ao - 2p)y2 + (-3qal - p(al)2 - p2 - 3(ao)2 + 4pao)Y + C,
with C = q2 + 3qaOal - pqal + p2ao - 2p(ao)2 + (ao)3 - q(al)3 + pao(alf.

b) We see that ao = (2p/3); we can then choose al = (3/p) ( -(q/2)+A)
with A = J(q2/4) + (p3/27), so that y3 = (6A/p)3 (-(q/2) + A).
3) These choices give a polynomial R that is, up to a non-zero constant, of

the form X - B + (p/3B) where B is a cube root of -(q/2) + A, so what
we have here are Cardan's formulas.
Solution to Exer.cise 3.3.
1) First, subtract from each column the following column multiplied by T l ,

beginning with the first one. This gives
V(Tl, ... ,TN)= II (Tl -Ti )V(T2 , ... ,TN ).

2::;i::;N
By induction, we obtain
V(Tl, ... ,TN )= II (Ti-Tj ).

l::;i<j::;N
2) Note that F(Yj) = L ak(Yj)k for 1::; j::; n, F(Xj) = 0 for 1::; j::;
0::; k::; m
m, C(Xj) = L bk(Xj)k for 1 ::; j ::; m, and C(Yj) = 0 for 1 ::; j ::; n.
O::;k::;n
Take the product M ~ and consider its determinant. For 1 ::; j ::; n, F(Yj)
is a factor throughout the j-th row; for n + 1 ::; j ::; n + m, C(Xj _ n ) is a
factor throughout the j-th row. Once we have factored out these quantities,
we are left with a determinant which can be computed as a product of the
determinants of two blocks, each of which is a Vandermonde determinant.

We find
Comparing this with the product det(M) det(Ll), we find
We have
Consequently,
II F(Yj) II G(Xi ) = D II
The result is then obtained by multiplying the two terms by an and sim-
plifying by II
F(Yj).
lSjSn
3) The computation of the determinant gives an element of the ring in the

statement.
REMARK. - Here, in computing the determinant, we can make zeros appear

simultaneously in the places of the coefficients of the highest degree terms
in the columns corresponding to the highest degree polynomial, and then
continue to repeat this procedure: it is exactly the algorithm of Euclidean
division.
1) Use the formulas from the text.
2) a) The root of dX + e is equal to -e/d, so we have
e2 e )
Res(dX+e,aX 2 +bX+c)=d2 ( a d2 -b d +c =cd2 -bde+ae 2.
Furthermore, we have
2 (a db - ae) cd2 - bde + ae 2

aX + bX + c = (dX + e) d X + d2 + d2 '
hence
Res(aX2 + bX + e, dX + e) d2Res (dX + e, cd2 - b:: + ae 2 )
ed2 - bde + ae 2.
adO
The same result can be obtained by computing the determinant bed
e 0 e
3) We find
Res(aX2 + bX + e, a' X2 + b' X + e')
r_ r]
ab' - a'b ae' - a'e)
= aRes ( aX2 + bX + e, a X + a
=a [e (ab ~ a'b b ( ab' : a'b) (ae' : a' c) + a ( ae' : a'e

= (a'e - ae'f - (ab' - a'b)(be' - b'e).
The second formula is obtained by a simple computation.
1) It is possible to use several different formulas to compute the resultant.

For example, if Xl and X2 are the roots ofaX 2 + bX + e in <C, we have
D(aX2 + bX + c) -(2aXI + b) (2ax2 + b)

-4a 2 (~) - 2ab (-~) - b2
b2 - 4ae.
Similarly, because ...j( -p/3), -...j( -p/3) are the roots of 3X 2 + p, we have
D(X3 + pX + q) = -ReS(X3 + pX + q, 3X 2 + p)
~-27[(Rr +pR+q] [-(Rr -pRH]

=-27[~R+q] [-2:R+q] =-4p3_27q2.
Now, if a, b, e denote the roots of X 3 + pX + q in <C, then using a formula

from Exercise 3.1, we obtain
D(X3 + pX + q) = -(3a 2 + p)(3b2 + p)(3e2 + p)

-27a 2b2e2 - 9p(a 2b2 + b2e2 +

e2a 2) _ 3p(a 2 + b2 + e2) _ p3
-27q2 - 9p[(ab + be + ea)2 - (a + b + e)abe]
-3p2[(a + b + e)2 - 2(ab + be + ea)] _ p3
-27q2 - 9p[p2] _ 3p2[-2p] _ p3
_4p 3 _ 27q2.
These computations can also be done using the method of Euclidean divi-
sion or by computing a determinant.
2) Taking into account the (n 2 - n)j2 sign changes to be made in the

product, we have
(_1)n(n-l)/2Res(F, F')
D(F)
a
(-1 )n(n-l)/2 a n-l II F' (Xi)
l::;i::;n
a
(_1)n(n-l)/2 an-2 II (a II (Xi - Xj))

l::;i::;n l::;j::;n,j#i
a2n - 2 II (Xi-Xj)2.
l::;i::;j::;n
3) We know that the resultant belongs to the field generated by the coef-
ficients of the polynomials.
4) If P has only real roots, then of course we have D(P) > O.

If P does not have only real roots, then it has an even number of non-real
roots, say k pairs of complex conjugates. Let us reason by regrouping the
factors occurring in the formula of part 2).
For every non-real root X and every real root y, the terms X - Y and x - y
are conjugate, so (x - y)2 (x - y)2 > O.
For every non-real root x and every non-real root y, the terms x - y and
x - yare conjugate, as are the terms x - y and x - y, so we have
For every non-real root x of P, we have (x - x)2 < O.
°
Thus, D (P) has the same sign as (-1) k. The final conclusion is that if
D(P) > 0, 2k = mod 4 and if D(P) < 0, then 2k = 2 mod 4.
5) a) The map which associates D(P) to P is continuous (take the de-

terminant to be the expression of the discriminant), and the set of
polynomials having only simple roots corresponds to the complement

of the inverse image of 0, so it is open.
b) An n x n matrix has distinct eigenvalues if and only if its characteristic

polynomial has no double roots. The rest follows immediately.
1) Applying the formulas for computing resultants, and using the fact that
(_1)n(n-l) = 1, we find
D(xn -1) (_1)n(n-l)/2Res(X n -1,nX n - 1)

(_1)n(n-l)/2Res(nX n - \ xn - 1)
(_1)n(n-l)/2 n n( _l)n-l.
Finally, D(xn -1) = (_1)(n-l)(n+2)/2 n n. In other words, we have
D(xn - 1) = nn if n = 1,2 mod 4,

D(xn -1) = -nn if n = 0,3 mod 4.
2) Using the formula D(F) = a 2n - 2 II

l:5i<j:5n
( X't - .)2' we find
xJ
D(X + 1) 1 (empty product),

D(X2 +X + 1) .
(] - ].2)2 - -3
- ,
D(X3 + X2 + X + 1) [(i + 1)2i( -1 + i)]2 = -16.
3) If P(X) = a II (X - Xi) = (X - Xl)P1(X), we have

l:5i:5n
l:5i<j:5n
l<i:5n 2:5i<j:5n
4) We have
(-1 )(n-l)(n+2)/2 n n D(xn -1)
D[(X _1)(X n - 1 + X n - 2 + ... + 1)]
n 2D(X n - 1 + X n - 2 + ... + 1),
which gives
D(X n - 1 + X n- 2 + ... + 1) = (_1)(n-l)(n+2)/2 n n-2.
5) Set P(X) = X n - 1 + X n - 2 + ... + 1. If (= e2irr / n , we have
D(P) = (_1)n(n-l)/2Res(P, PI) = (_1)n(n-l)/2 II pI(i).

l:5i:5n-l
As
P(X) = xn -1
X-I'
we have
r(n-l)i
pI(ri) = _n".::..,._,---
" (t -1 '
and because
II (i - 1) = (_l)n-l P(l) = (_1)n- 1 n,

l:5i:5n-l
we find the preceding result.

4
Field Extensions
In this chapter, we come to the basic notions of Galois theory. Abel and
Galois defined the elements of a generated extension, but they did not envi-
sion these elements as forming a set. The concept of a field (and the word)
did not appear until the work of Dedekind between 1857 and 1871. The
abstract definition of a field was given about 20 years later by Weber and
Moore. One hundred years ago, the language of linear algebra did not exist
and results were formulated very differently from the way they are today,
as can be seen, for example, in Weber's book, listed in the bibliography.
4.1 Field Extensions

4.1.1 Definition
An extension of a field K is a field containing K as a subfield. If M is
an extension of K, then an intermediate extension (between K and M)
is an extension L of K contained in M. We will usually represent field
extensions in one of the forms shown in Figure 4.1, where the upper field
is an extension of the lower field.
L L L'
I
K
or
\/ K
FIGURE 4.1.

52 4. Field Extensions
More generally, an extension of a field K is a pair (L, i) where L is a

field and i : K ~ L a ring homomorphism between two fields; such a
homomorphism is necessarily injective (if x -=I- 0, then x has an inverse x- 1
and i(x)i(x- 1) = i(XX-l) = i(l) = 1 shows that i(x) -=I- 0).
4.1.2 Proposition
An extension L of a field K is naturally endowed with the structure of a
K -vector space and even that of a K -algebra.
PROOF. - The K-algebra structure is defined by the addition and multi-
plication operations of L, and the K-action is simply the restriction of this
multiplication to K x L. <)
4.1.3 The Degree of an Extension

Let L be an extension of a field K. The dimension of L as a K-vector space
is called the degree of the extension Lover K; it is written [L : KJ. A field
L is said to be a finite degree of K if [L : KJ is finite. An extension of
degree 2 is called a quadratic extension.
COMMENTARY. - The rest of this book is devoted to the study of finite de-
gree extensions, with just a few exceptions. The general study of extensions
of infinite degree needs topology.
EXAMPLES. -
1) The fields Rand <C are not countable, so they are extensions of Q of
infinite degree.
2) R(X) is an extension of infinite degree of R
3) <C is a quadratic extension of R, with basis {I, i} for example.
4) Q[ V2J = {a + bV2; a, b E Q} is a quadratic extension of Q, with basis

{I, V2} for example.
4.1.4 Towers of Fields

In certain problems, for example the solvability by radicals considered in
Chapter 12, one considers finite sequences of successive extensions K =
Ko c ... C Kr; such a sequence is called a tower of fields.
4.2 The Tower Rule 53
4.2 The Tower Rule

4.2.1 Proposition
Let L be a finite-degree extension of a field K, and let M be a finite-degree
extension of L. Then M is a finite-degree extension of K, and we have
[M : K] = [M : L][L : K].
EXAMPLE. -
The fields in this example will be described more precisely after this chap-
ter; they are algebraic extensions of Q. The formula gives the degree of
Q[ -v2, j] over Q, using the intermediate extension Q[ -v2] and Proposition
4.5.2 below; we can also use the intermediate extension Q[j].
REMARK. - The above formula is also useful in the form "[L : K] divides
[M:K]."
PROOF. - Set n = [L : K] and p = [M : LJ, and let {ll, ... , In} denote a
basis of Lover K and {ml, ... , mp} a basis of Mover L. The np numbers
limj for 1 ::; i ::; n, 1 ::; j ::; p, form a basis of Mover K. Indeed,
1) They form a linearly independent system because if
L xijlimj = 0,
l::'Oi::'On,I::'Oj::'Op
then
L (L xijIi)mj = O.
l::'Oj::'Op l::'Oi::'On
But the mj are linearly independent in the L-vector space M, so for

j= 1, ... ,p, we must have
L XijIi = O.
l::'Oi::'On
Thus all the Xij are zero because the Ii are linearly independent in
the K -vector space L.
2) They form a generating system because if x is in M, then there exist

Xl, ... , xp in L such that
and for each of the Xj there exist Xlj, ... , Xnj in K such that
Xj = L xijlj,
I~i~n
so
<>
COROLLARY. - Let M be a finite-degree extension of a field K and L an
intermediate extension. Then L is a finite-degree extension of K and M is
a finite-degree extension of L.
PROOF. - If {lI, ... , In} are elements of L that are linearly independent over
K, and if {mI, ... , mp} are elements of M that are linearly independent
over L, then by Proposition 4.2.1, the np numbers limj for 1 ::; i ::; n, 1 ::;
j ::; p, are linearly independent over K. Thus np::; [M : KJ, which bounds
nand p. <>
4.3 Generated Extensions

4.3.1 Proposition
Let L be an extension of a field K, and let A be a subset of L.
1) There exists an extension of K, denoted by K(A), which contains A
and has the property that it is minimal, i. e. it is contained in every
extension of K contained in L and containing A.
2) As a set, K(A) is equal to the set E of elements of L of the form

S(aI, ... ,ak)IT(aI, ... ,ak), where k is an integer running through
N, aI,"" ak are elements of A and SIT is a rational function in
K(X I , ... , Xk) of rational functions, such that T(aI, ... , ak) i- O.
PROOF. -
1) Consider the family of extensions of K contained in L and containing
A. This family is non-empty since L belongs to it, and the intersection
of all fields in the family satisfies the defining property of K(A).
2) It is easy to see that every element of E lies in every extension of K

contained in L and containing A. Furthermore, E is obviously a field
containing K and A, so we obtain E = K(A).
<>
4.4 Algebraic Elements 55
4.3.2 Definition
The extension K(A) is called the extension of K generated by A. If A =
{all"" an}, then the extension K(A) is usually written K(al,"" an).
4.3.3 Proposition
Let L be an extension of a field K, and let A, B be two subsets of L. Then
K(A u B) = K(A)(B).
PROOF. - By Proposition 4.3.1 1), we know the following facts .
• Because A U B c K(A) u B c K(A)(B), we have K(A U B) c
K(A)(B) .
• Because K(A) U B c K(A U B), we have K(A)(B) C K(A U B). 0
4.4 Algebraic Elements

4.4.1 Definition
Let K be a field and L an arbitrary extension of K. An element a of L is
said to be algebraic over K if there exists a non-zero polynomial in K[X]
having a as a root.
EXAMPLES. - .J2, -v2, and e2i7f In are all complex numbers that are alge-
braic over 'O1.
REMARK. - It is not obvious that the sum and product of two algebraic
numbers over K is also algebraic; this is proved in §6.3.
4.4.2 Transcendental Numbers

A number that is not algebraic over K is called transcendental over K. A
complex number transcendental over '01 (such a number is simply called a
transcendental number) is thus a number that is not a root of any polyno-
mial with coefficients in Z.
Joseph Liouville was the first to prove, in 1844, that certain real num-
bers are transcendental over '01, for instance L
lO-n! (see Exercise 4.1).
n>O
Hermite proved the transcendence of e in 1873 (see Exercise 4.2), and Carl
Lindemann proved the transcendence of 1[" in 1882, finally providing a neg-
ative answer to the ancient problem of squaring the circle (i.e. using only
ruler and compass to construct a square having area equal to that of a
given circle).
Alexandre Gelfond and Theodor Schneider showed in 1934 that a b is

transcendental whenever a is algebraic, a i= 0 and a i= 1, and b is an irra-
tional algebraic number. For example, 2../2 is transcendental. Alan Baker,
a 1970 Fields Medalist, extended these results considerably at the end of
the 1960s. It is still not known if numbers such as e + 7r are transcendental
or not. Transcendental number theory is at present a rapidly developing
subject.
4.4·3 Minimal Polynomial of an Algebraic Element

PROPOSITION. - Let K be a field, L an extension of K and a an element
of L which is algebraic over K. There exists a unique monic polynomial P
in K[X] having a as a root and of minimal degree among all the non-zero
polynomials in K[X] having a as a root.
PROOF. - Let us first show that the ring K[X] is a principal ideal domain,
i.e. every ideal of K[X] is generated by a single element. This is true for
the ideal I = (0), and if I is a non-zero ideal of K[X], then it contains a
polynomial P i= 0 of minimal degree, such that every other polynomial S
of I is a multiple of P. Indeed, the Euclidean division of S by P gives an
equality S = PQ + R where the degree of R is strictly less than that of P,
but R = S - PQ is an element of I, contradicting minimality of the degree
of P unless R = O.
Now, consider the set I of polynomials of K[X] which vanish at a. This
forms a non-zero ideal of K[X], so there exists a polynomial T of K[X]
such that I = (T). Dividing T by its leading coefficient, we obtain a monic
polynomial P which generates I.
The uniqueness of P is a consequence of the fact that if T is a monic
polynomial in K[X] satisfying the same conditions as P, then P - T is an
element of I which vanishes at a, so because its degree is < deg(P), it must
~zmo. ¢
4.4.4 Definition
Let K be a field and L an extension of K; let a be an element of L algebraic
over K. By the preceding section, there exists a unique monic polynomial
of minimal degree in K[X] which vanishes at a; it is called the minimal
polynomial of a over K. If deg(P) = n, then a is said to be algebraic of
degree n over K; we also say that n is the degree of a over K.
EXAMPLE. - X2 - 2 is the minimal polynomial of V2 over Q; V2 is of degree

2 over Q.
REMARKS. - The minimal polynomial of an element a of K is X-a.

The minimal polynomial depends on the field K; thus -v2 has minimal
polynomial X 6 - 2 over Q, but X 3 - V2 over Q[V2].
4.4 Algebraic Elements 57
4.4.5 Properties of the Minimal Polynomial

PROPOSITION. - Let K be a field, L an extension of K and a an element
of L which is algebraic over K of minimal polynomial Paver K.
1) Every polynomial of K[X] which vanishes at a is divisible by P.
2) P is irreducible over K, i. e. it is not the product of two non-constant

polynomials in K[X].
3) Every root of P lying in L has minimal polynomial P.
4) If L is a field of characteristic 0, in particular if Lee, a is a simple

root of P (we will say that P is separable; for a counterexample in an
infinite field of non-zero characteristic, see §15.2).
PROOF. -
1) This is an immediate consequence of the preceding proof.
2) If P = ST E K[X], the equality pea) = 0 implies that Sea) = 0 or

T(a) = o. If neither S nor T is a constant, then deg(S) < deg(P) and
deg(T) < deg(P), which is impossible.
3) Let b be a root of Pin L; thus b is an algebraic number over K and so

it has a minimal polynomial S over K. As PCb) = 0, P is a multiple
of S by part 1), but as P is irreducible, this means that P = S.
4) Suppose that a is a root of order k > 1 of P, i.e. P(X) = (X -a)kS(X)

in qX], with Sea) -=f. O.
We have P'(X) = k(X -a)k-1S(X)+(X -a)kS'(X), so P'(a) = 0,
which contradicts the definition of P as the minimal polynomial of a
since deg(P') = deg(P) - 1 and P' is non-zero in K[X].
4·4· 6 Proving the Irreducibility of a Polynomial in Z[X]

An algorithm for factoring polynomials in Z[X] or in (Z/pZ)[X] into irre-
ducible factors, developed by Berlekamp (1967), is now available as part of
computer packages in formal computation. We discuss methods for factor-
ing by hand. Let us first recall a classical result.
PROPOSITION. - A polynomial in Z[X], whose content, which is the greatest

common divisor of its coefficients, is equal to 1, is irreducible in Q[X] if
and only if it is irreducible in Z[X].
Thus, the irreducibility of a given polynomial in Q[X] can always be

expressed as the irreducibility of a polynomial in Z[X], and it is enough to
consider methods over Z.
METHODS. - Let us now recall different practical methods for studying the
irreducibility of a polynomial P(X) = L akXk in Z[X] by hand.
O$k$n
1) Eisenstein's criterion proves that a polynomial P in IQ[X] is irre-
ducible if it satisfies the following condition: there exists a prime
number p which does not divide an but divides all the other coef-
ficients of P, whereas p2 does not divide ao. Sometimes, when the
Eisenstein criterion is not directly applicable to a polynomial P(X),
it is applicable to P(X + a) for some value of a.
This criterion remains valid if we replace Z by any factorial ring
A and the field Q by the fraction field of A, and if we assume the
existence of a prime (i.e. irreducible) element p in A satisfying the
above conditions.
2) To see if P does not have a rational root, assume that pjq is a rational
root of P, written as a totally reduced fraction. Then qn P(pjq) is an
integer, but it is equal to zero, so it follows that q divides an and
p divides ao, which makes it possible to obtain the set of rational
candidates for roots of P. If this set does not have too many elements,
we can test the candidates one by one.
3) If P is of degree 3 and has no rational root, then it is irreducible.
4) If P is of degree 4, has no rational root and is not decomposable as a

product of two quadratic factors (to see this, one can use a method
of indeterminate coefficients as in Exercise 2.7), then it is irreducible.
5) It is often useful to consider the reduction of a polynomial in Z[X]

to the ring (ZjpZ)[X].
Indeed, if there exists a prime number p such that the image of P in
(ZjpZ)[X] is an irreducible polynomial of the same degree, then P is itself
irreducible (assuming that its content is equal to 1). The converse is false;
the classical example is the polynomial X4 + 1. We show later (in Exercise
14.8) that this polynomial is reducible modulo p for all primes p, although
it is clearly irreducible in Z[X].
If the image of P in (ZjpZ) [Xl is a polynomial of strictly smaller degree,
we cannot conclude anything; for example, one can consider the polynomial
(pX + l)(X + 1).
The irreducibility of Pin (ZjpZ)[X] can be proved, for example:
1) for polynomials of degree 2 or 3, by systematically testing if the ele-
ments of ZjpZ are roots of P;
2) for polynomials of degree 4, by systematically testing if the elements

of ZjpZ are roots of P and showing that a decomposition into a
product of two quadratic polynomials is impossible;
4.5 Algebraic Extensions 59
3) for polynomials of degree n, by listing the irreducible polynomials of

degree :S n/2 and testing if they divide P.
Note, finally, that it is sometimes interesting to compare the de-

grees of the irreducible factors of the images of P in (Z/pZ)[X] and
(Z/qZ)[X].
4.5 Algebraic Extensions

4.5.1 Extensions Generated by an Algebraic Element
Let L be an extension of a field K, and let a be an element of L which is
algebraic over K. Let K[a] denote the smallest subring containing K and a,
i.e. the image of the homomorphism f: K[X] ---> L defined by f(X) = a,
where flK denotes the inclusion of K into L.
Thus, the expressions K[a] and K(a) refer respectively to images of the
ring of polynomials K[X] and of the field of rational functions K(X). If a
is algebraic over K, these two images coincide.
4·5.2 Properties of K[a]

PROPOSITION. - Let L be an extension of a field K, and let a be an element
of L which is algebraic of degree n over K, with minimal polynomial P over
K. Then
1) K[a] is an extension of K and K[a] = K(a),
2) [K[a] : K] = n and the set {a\ O:S k :S n - I} forms a basis of K[a]
as a K -vector space, and
3) The homomorphism f : K[X] ---> K[a] defined by f(X) = a and

f(k) = k E K[a] for k E K induces a K-algebra isomorphism c.p :
K[Xl/(P) ---> K[a] which leaves the elements of K invariant and
makes the diagram in Figure 4.2 commute (n denotes the canonical
projection).
1t
FIGURE 4.2.
PROOF. -
1) We have K[aJ C K(a), so let us show the inverse inclusion. By Propo-
sition 4.3.1, every element of K(a) is of the form S(a)jT(a) with S
and T in K[X] and T(a) =f. 0; thus T is not divisible by P, and as
P is irreducible over K, T is relatively prime to P. Bezout's the-
orem ensures the existence of polynomials U and V of K[X] such
that UT + VP = 1, which proves that S(a)jT(a) = S(a)U(a). Thus
S(a)jT(a) belongs to K[a].
2) Let us first show that the family {a k ; 0 :::; k :::; n - I} is free over K. If
there exists a family {Ak; 0 :::; k :::; n - I} of elements of K such that
L )..k ak = 0, the polynomial S(X) = L AkXk in K[X]
O~k~n-l O~k~n-l
vanishes at a. As deg(S) < deg(P), we have S = 0, which proves that
Ak = 0 for 0 :::; k :::; n - l.
The family {a\ 0:::; k :::; n - I} generates K[a], because if S is a
polynomial of K[X] and if S = PQ + R by Euclidean division, we
have S(a) = R(a). As deg(R) :::; n - 1, R(a) belongs to the K-space
generated by {ak;O:::; k:::; n -I}.
3) The map f defined above has image K[a], and kernel the set of poly-
nomials vanishing at a, i.e. the ideal (P). This gives the desired fac-
torization. <>
4.5.3 Definition
An extension L of a field K is said to be algebraic if every element of L is
algebraic over K.
EXAMPLE. - C is an algebraic extension of JR., because any complex number

a + ib with a and b real is a root of the polynomial X2 - 2aX + a 2 + b2 = 0
in JR[X].
4.5.4 Extensions of Finite Degree

PROPOSITION. - Let K be a field. Every extension L of K of finite degree
n is algebraic over K, and every element of L is algebraic of degree :::; n
over K.
PROOF. - Let x E L. The family {xk; 0 :::; k :::; n} has more than n ele-
ments, so it is not linearly independent over K. This means that there
exists a family {Ak; 0 :::; k :::; n} of elements of K, not all zero, such that
L Ak Xk = O. In other words, the polynomial S(X) = L
AkXk of
O~k~n-l O~k~n
K[X] vanishes at x. Because S =f. 0, x is algebraic of degree :::; n over K. <>
4.6 Algebraic Extensions Generated by n Elements 61
4.5.5 Corollary: Towers of Algebraic Extensions

Let K be a field, and L an extension of K. Let r be an integer and (Ki)Os,is,r
a tower of extensions of K contained in L such that Ko = K and there
exist elements al, . .. , ar of L such that for i = 1, ... , r, ai is algebraic of
degree ni over K i - l and Ki = Ki-dai]. Then Kr is an algebraic extension
of K of degree n = nl ... n r .
PROOF. - By §4.5.2, Ki is of degree ni over Ki-li the tower rule shows
that Kn is an extension of finite degree n of K. Proposition 3.5.4 concludes
the proof. <)
4.6 Algebraic Extensions Generated by n Elements

4.6.1 Notation
Let L be an extension of a field K, and let al, ... , an be elements of L
which are algebraic over K. Write K[al, ... , an] for the image of the ho-
momorphism f: K[XI, ... ,Xn ]--- L defined by f(P) = P(al, ... ,an ). It
is the K-algebra generated by al, ... , an.
4.6.2 Proposition
With notation as above, K[al, ... , an] is an algebraic extension of finite
degree of K, equal to K(al, ... , an).
PROOF. - By induction, we will construct the tower of extensions (Ki)o<i<n

of K contained in L such that Ko = K, and such that for each i = 1, ... , n,
there exists an element ai, algebraic over K (so also over K i - l ), such that
Ki = Ki-dai]. Corollary 4.5.5 shows that Kn is an algebraic extension of
finite degree of K.
Let us show by induction on i that Ki = K[ab ... , ad for i = 1, ... , n.
The case n = 1 was considered in §4.5.2. Suppose that
K i- l = K[al, ... , ai-d·
Because every polynomial P(X I , ... , Xi) in K[XI' ... ' Xi] can be written
in the form L Pk(Xb ·.· ,Xi-d(Xi)k, we have
k
P(ab ... ,ai) = LPk(al, ... ,ai-l)(ai)k,

k
so K[al, . .. , ail = K[al, ... , ai-l][ai] = Ki-dai] = K i . As Ki is a field, we

have Ki = K(ab ... , ai) for i = 1, ... , n. <)
REMARK. - The degree of K[al, ... , an] over K is less than or equal to the
product of the degrees of the ak over K.
4.6.3 Corollary
Let L be an extension of a field K and let a, b be elements of L that are
algebraic over K. Then a + b, ab, alb are all algebraic over K.
PROOF. - These elements belong to K[a, b], which is an algebraic extension
of finite degree of K by §4.6.2. <>
COMMENTARY. - This corollary settles the point raised in §4.4.1: it proves
for example that -v2 + {17 is algebraic of degree ~ 15 over Q. A more
general statement is that the complex numbers which are algebraic over a
subfield K of C form a field (called the algebraic closure of K in C).
A polynomial having a + b, ab etc. as a root can be computed using
resultants (see Exercise 4.8).
4.7 Construction of an Extension by Adjoining a

Root
So far in this chapter, we have considered a very particular situation,
namely, the case where K is a subfield of a field L, and we have defined
what it means for an element of L to be algebraic over K. Fields that
are subfields of C correspond to this situation; it is the situation which
Dedekind considered in 1871, when he gave the first definition of a field.
Almost all the exercises in Chapters 4-12 concern this situation. However,
it is also possible to consider the following more general situation: K is an
arbitrary field and P is a polynomial K[X] that has no roots in K. Then,
even without knowing any extension of K beforehand (see, for example,
Chapter 14), we can construct an extension L of K in which P has at least
one root. This more general situation was studied by Leopold Kronecker
and Henri Weber in the years 1880 to 1900.
4.7.1 Definition
Let K be a field, and let P be a polynomial in K[X]. An extension L of K
is called a rupture field of P over K if there exists a root x of Pin Land
if L = K[x].
EXAMPLE. - Q[ -v2] and Q[j -v2] are rupture fields of X 3 - 2 over Q.
4. 7.2 Proposition
Let P be an irreducible polynomial of degree n in K[X]. The quotient ring
L = K[X]/(P) is a rupture field of P, i.e. an extension of degree n of K
containing the class x of X as a root of P. Moreover, P is the minimal
polynomial of x over K.
4.7 Construction of an Extension by Adjoining a Root 63
PROOF. - It suffices to show that the ideal (P) is maximal; we will prove
this so as to give a method making it possible to compute the inverse of a
non-zero element of the quotient. Let 7r: K[X] ---> L denote the canonical
projection. Every non-zero element of Lis of the form 7r(S) with S in K[X]
not a multiple of P. Consequently, because P is irreducible, P and S are
relatively prime, and by Bezout's theorem, there exist U and V in K[X]
such that US + VP = 1. Thus we have 7r(U)7r(S) = 1 in L, which shows
that 7r(S) is invertible, so L is a field.
Let i: K ---> L denote the composition of the injection K ---> K[X] and
7r. Since i is a homomorphism of rings with unit, it is injective, so L is an
extension of K. Set x = 7r(X) and P = LakXk. The polynomial P
O:S;k:S;n
has coefficients in K, so we have P(x) = L akxk = 7r(P) = 0. Thus,
O:S;k:S;n
X is algebraic over K. If a polynomial S in K[X] vanishes at x, we have
Sex) = S(7r(X)) = 7r(S) = 0, so S is a multiple of Pin K[X]. Consequently,

P is the minimal polynomial of x over K. 0
By §4.5.3, every element of L can be written uniquely as L akxk

O:S;k<n
In other words, the set {I, x, ... , x n - 1} forms a basis of L as a K -vector
space.
4.7.3 Corollary
Let K be a field and P a non-constant polynomial of K[X]. Then there
exists a rupture field L of P over K such that [L : K] :::; deg(P).
PROOF. - Consider an irreducible factor p of P, and set L = K[XJ!(p) ,

7r : K[X] ---> L, x = 7r(X). We have p(x) = 0. The extension L of K
contains a root of p, so it contains a root of P. 0
4· 7.4 Universal Property of K[X]j(P)

Let P be an irreducible polynomial of degree n of K[X]. The extension K --->
K[XJ!(P) has a universal property which is an immediate consequence of
the universal property of quotients.
PROPOSITION. - For every extension M of K and every root a of P in

M, there exists a unique field homomorphism 'P: K[X]/(P) ---> M such
that 'P(x) = a and 'PIK = id (in Chapter 6, we will say that 'P is a K-
homomorphism) .
°
PROOF. - Consider the ring homomorphism f: K[X] ---> M defined by
flK : K M and f(X) = a. We have f(P) = pea) = so fl(P) = 0,
--->
which gives the factorization of f by K[X]/(P) by the universal property

of quotients. <>
EXAMPLES. - Set P = X2 - 2, K = Q, M = R Because P has two roots
in lR, there exist two Q-homomorphisms 'P, 'P' : K = Q[Xl/(X2 - 2) -+ lR;
one sends x to h, and the other to -h.
For P = X 3 - 2, K = Q, M = lR, there exists a unique Q-homomorphism
'P : Q[X]/(X 3 - 2) -+ lR, which sends x to -v2. However, if M = <C,
there exist three Q-homomorphisms 'P : Q[X]/(X 3 - 2) -+ <C, defined by
'P(x) = -v2, j-v2, j 2 -v2.
Toward Chapters 5 and 6

In order to prove the impossibility of certain geometric constructions that
were sought for over 2000 years, it suffices to use just a part of the results of
this chapter. These problems and the proofs of their impossibility form the
subject of Chapter 5, which should be considered as a joyous digression.
We will return to Galois theory proper in Chapter 6.

We found it natural, in a chapter concerning algebraic numbers, to present
some of the famous and little taught examples of transcendental numbers.
We follow the proofs given in Alan Baker's book listed in the bibliography.
Exercise 4.1. An example ofa transcendental number: L a n 10- n !

n>O
1) Liouville's Theorem. Let a be a real number that is algebraic over
Q of degree n > 1, and let P be an irreducible polynomial of degree
n in Z[X] having a as a root. Show that there exists a real number
c > 0 such that for every rational number p/q (with q > 0) we have
la - (p/q)1 > (c/qn). We distin[;uish the two cases la - (p/q) I > 1 and
la - (p/q) I ~ 1; in the second case, apply the mean value theorem and
give an upper bound for IP'I on the interval [a - 1, a + 1].
2) Deduce the transcendence of a = L anlO- n! for every sequence (an)

n>O
of natural numbers between 1 and 9.
Exercise 4.2. The transcendence of e

As in the previous exercise, we will assume that e is algebraic; we will
obtain a double inequality (see 3) and 4) below) which can never be
simultaneously satisfied.
Suppose that e is algebraic over Q, a root of the irreducible poly-
nomial L qk Xk in Z[X].
O:Sk:Sn
If P(X) = L akXk is a polynomial of Z[X], set
O:Sk:SN
PI(X) = L lakl Xk ,
O:Sk:SN
and for every real t, set
J = L qkJ(k).
O:Sk:Sn
1) Show that J(t) = et L p(k)(O) - L p(k)(t).
Show that IJ(t)1 ::; ItleltlPl(1tl).
2) From now on, let P(X) = Xp-I(X - 1)p ... (X - n)p, where p is a
prime number > n. Show that
p(k)(O)= ... = p(k)(n) = 0 for k < p - 1,
p(p-I)(l) = ... = p(p-l)(n) = 0,
p(k)(O), ... , p(k)(n) are divisible by p! for k"2 p.
3) A lower bound for J:

Show that J = -qo L p(k)(O) - ... - qn L
p(k) (n).
O:Sk:SN O:Sk:SN
Deduce that IJI "2 (p - I)! for p > qo, by checking that J = 0 mod
(p - I)! and J -::f:. 0 mod p!.
4) An upper bound for J:
a) Show that PI(k) < (2n)N for 0::; k ::; n.

b) Deduce that IJI ::; A . BP, where A and B are numbers not
depending on p.
5) Conclude that e is transcendental.

Exercise 4.3. Determination of the rational roots of a polynomial

with integral coefficients
1) Consider a polynomial P(X) = L akXk E Z[X]. Show that the
O~k~n
rational roots of P are of the form !!. such that p and q are relatively
q
prime, p divides ao and q divides an.
2) Study the irreducibility of X 3 - 4X2 - ~X - ~ in Q[X].
3) Study the irreducibility of 30X 3 + 277X2 - 31X - 28 in Q[X].
Exercise 4.4. Factorization of polynomials

1) For a prime p, show the irreducibility over Q of the cyclotomic poly-
nomial
cI>p(X) = LXk
O~k~p-l
by applying Eisenstein's criterion to cI>p(X + 1).
2) Study the irreducibility of X 6 + 3 in the rings Q[X], Q[j][X], Q[i][X]

(you will need at least a little background knowledge on factorial rings
to solve the problem quickly over the third ring).
3) Study the irreducibility of X4 + 1 in Q[X] (use the variable change

Y = X + 1).
Factorize X4 + 1 into a product of irreducible factors in Q[(][X],
where ( is a root of this polynomial.
4) Study the irreducibility of P(X) = X 5 - X + 1 in the ring Q[X]

(show, for example, that P has no linear factor in Z[X], and then
determine the quadratic polynomials S which could divide P in Z[X]
by studying the values these candidates can take at -1,0, 1).
5) Study the irreducibility of P(X) = X4 - 15X 3 + 7 in the ring Q[X].

6) Let p be a prime number and n 2: 2 an integer. Show that P(X) =
xn + pX + p2 is irreducible in Z[X].
7) Give an example of an irreducible quadratic polynomial P in Q[X]

such that P(X2) is reducible in Q[X].
Exercise 4.5. The degree of an algebraic extension

Determine the degrees of the following extensions by finding bases for
them, and answer the questions.
1) Q[V2], Q[{/2] over Q.
2) Q[{/2,V2] over Q[{/2], over Q[V2] , over Q.
3) Q[v3, V2] over Q[v3], over Q[V2], over Q.

Compare Q[v3, V2] with Q[v3 + V2]. Determine the minimal poly-
nomial of v3 + V2 over Q.
4) Q[j] over Q.
Does v3 lie in Q[j]? Does i lie in Q[j]? Does j lie in Q[i]?
Q[v3,j], Q[v3, i,j], Q[v3, i], Q[v3 + i] over Q.
21f 21f 21f 21f
5) Q[cos 3], Q[sin 3], Q[cos 5], Q[sin 5 t] over Q.
6) Q[V2, v3, v5] over Q.
Exercise 4.6. Computing in algebraic extensions
Consider the polynomial P(X) = X 3 + 2X + 2 in Q[X]; let a denote

one of its roots.
1) Show that P is irreducible.
2) Express the elements 1ja, 1j(a 2 + a + 1) and u = a 6 + 3a 4 + 2a 3 + 6a

as functions of 1, a, a2 (for the quotients, use Euclid's algorithm and
the method of indeterminate coefficients).
3) Determine the minimal polynomial of u over Q.
Exercise 4.7. Algebraic extensions
1) Let L be an extension of finite degree n of a field K.
a) What happens in the case n = I? What happens if L' is another

extension of K, contained in L and of degree n over K?
b) Show that if n is prime, there exists no field strictly containing

K and strictly contained in L.
2) Let K be a field, and let a be an algebraic element of degree nand

of minimal polynomial P over K . Let b be an algebraic element over
K whose degree m is relatively prime to n.
Determine the degree of K[a, b] over K and show that P is
irreducible over K[b]. What is the intersection K[a] n K[b]?
3) Let x be an algebraic element of odd degree over a field K. Show that

x 2 is algebraic over K and that K[x] = K[X2].
4) Let K be a field, L an extension of K of finite degree n and a an

element of L which is algebraic over K of minimal polynomial P over
K. Show that deg(P) divides n.
5) Show that every quadratic extension ofQ is of the form Q[v'a), where
a is a squarefree relative integer.
6) What are the algebraic extensions of C? Give an example of an infinite

degree extension of Co
7) Let K be a field contained in C, and let L be an algebraic (but not

necessarily finite) extension of K. Let a be an element of C algebraic
over L. Show that a is an algebraic element over K.
8) Show that e and 7r are transcendental over every algebraic extension

ofQ.
Exercise 4.8. Elimination

1) Let K be a field contained in C, and let a and b be two algebraic
numbers over K, roots of two polynomials F and G respectively in
K[X]. Set deg(G) = n, and assume that there exists an algebraically
closed field containing K(Z).
a) Show that R(Z) = Resx(F(X), G(Z - X)) is a polynomial of

K[Z] having a + b as a root.
b) More generally, show that if T(X, Y) E A[X, YJ, then
R(Z) = Resx(F(X), Resy(Z - T(X, Y), G(Y)))
is a polynomial in K[Z] having T(a, b) as a root.
e) Show that R(Z) = Res x (F(X),XnG(Z/X)) is a polynomial

of K[Z] having ab as a root.
d) If T(Y) E K[YJ, give a non-zero polynomial which vanishes at

T(a).
2) Either directly or using results from the preceding problem, determine

polynomials which are minimal polynomials over Q for the following
numbers: (2/3) + V5/7, J2 + ~, J2 + ~, a + J2 where a is a root
of X 3 + X + I, and b2 + b where b is a root of X 3 + 3X + 3.

1) If p/q is a root of P, we have qn P(p/q) = 0, i.e. L akpkqn-k = o.
O~k~n
This produces the two inequalities
p L akpk-lqn-k
l~k~n
q L akpkqn-k-l
O~k~n-l
which give the result.
2) The only possible rational roots of 2X 3 - 8X 2 - 9X - 5 are ±5, ±1,

±(1/2), ±(5/2). We see that 5 is a root, divide by it and complete the
factorization.
3) This question needs more work than does the preceding one. First de-
termine that -(28/3) is a root, and then check that the factorization is
given by (3X + 28) (lOX 2 - X-I).
1) By Eisenstein's criterion, the polynomial
is irreducible, since the formula k!(p - k)!(~) = p! shows that the (~)
are divisible by p for 1 ~ k ~ p - 1 and that (:) = 1, (~) = p. It follows
that p is irreducible. This classical result is generalizable to all cyclotomic
polynomials (see Chapter 9).
2) Applying Eisenstein's criterion with the prime number p = 3, we see

that the polynomial X 6 + 3 is irreducible in Q[X].
As X 6 + 3 = (X 3 + iV3)(X3 - iV3) and iV3 = 2j + 1, the polynomial
X + 3 is not irreducible in Q[j] [X]; we have X 6 + 3 = (X 3 + 2j + 1 )(X 3 -
6
2j - 1).
The polynomial X 6 + 3 has coefficients in the factorial ring Z[i], whose
fraction field is Q [i].
Let us show that the prime p = 3 is irreducible in Z[i]. If we write
3 = (a + ib)(c + id) with a, b, c, d E Z, then taking moduli, we find that
9 = (a 2 + b2 )(C2 + d 2 ). It is impossible to realize a 2 + b2 = 3 or c2 + d 2 = 3
in Z, so a + ib or e + id has modulus 1, and it is invertible in Z[i]. As 3 is

irreducible in the factorial ring Z[i], it is prime in Z[i].
Applying Eisenstein's criterion, we then see that the polynomial X 6 + 3
is irreducible in lQ[i][X].
Because we know the factorization of X 6 + 3 into linear factors in C, we
can also show that X 6 + 3 has no factors of degree 1, 2, or 3 in lQ[i][X], by
trying to regroup two or three factors (and taking care not to forget any
cases).
3) Eisenstein's criterion applies to (X + 1)4 + 1.

Solving X4 = -1 = e i7r leads to setting ( = ei7r / 4 = (1 + i)jv'2. The
roots of X4 + 1 are e i7r / 4, e3i7r / 4 = (3, e5i7r /4 = (5, e7i7r / 4 = C, so in the ring
IQ[(][X] we have
X4 + 1 = (X - ()(X - (3)(X - (5)(X - C).
Every other choice of ( leads to the same result.
4) The method we propose below is Kronecker's algorithm (1882); gener-

ally, it is too long to be useful in practice.
The only possible rational roots of the polynomial are ±1, which obvi-
ously do not work. If X 5 - X + 1 is reducible in IQ[X], it has a quadratic
factor T(X) that we can choose in Z[X]. As T(O), T(l), and T( -1) respec-
tively divide P(O), P(l), and P( -1), they are equal to ±1. The possible
polynomials Tare ±(2X 2 - 1), ±(X2 + X - 1), ±(X2 - X - 1), and we
check that none of them divides P(X).
5) If P is reducible in IQ[X], it factors into a product of two non-constant

monic polynomials in Z. Consider the situation mod 2. The image PI of
Pin (Zj2Z)[X] is equal to X4 + X 3 + 1. As PI has no root in Zj2Z (the
only possible roots, 0 and 1, do not work), PI has no linear factor. Suppose
that PI(X) = (X2 + aX + b)(X2 + eX + d); the equality bd = 1 leads to
b = d = 1, then to a + e = 1 and a + e = 0, which are contradictory.
6) We denote the image of a polynomial in Z[X] in (ZjpZ)[X] by adding

a subscript 1.
If P = ST in Z[X], where Sand T are non-invertible monic polynomials
in Z[X] of degrees sand t respectively, we have S l T 1 = xn so S1 (X) = X s ,
T1 (X) = X t , SeX) = xs + L
pakXk, and T(X) = X t + pbkXk.L
O~k<s O~k<t
The value s = 1 (and similarly, t = 1) leads to a contradiction because
the equality of the constant terms would give aobo = 1, so ao = ±1 and
-pao = ±p would be a root of P, which is impossible.
Finally, s, t > 1 is impossible, since the coefficient of X in the product is
divisible by p2.
7) If P(X) = X2 -7X + 1, we have P(X2) = (X2 +3X + 1)(X2 -3X + 1).

1) X2 - 2 and X 3 - 2 are irreducible over IQ by Eisenstein's criterion with

p = 2.
As the minimal polynomial of v2 over IQ is X2 - 2, we have
[1Q[v2] : IQ) = 2
and a basis of 1Q[v2] over IQ is given by {I, v2}.

As the minimal polynomial of .v2 over IQ is X 3 - 2, we have [1Q[.v2] : IQ] =
3, and a basis of 1Q[.v2] over IQ is given by {I, .v2, «4}.
2) 1Q[.v2, v2] is an extension of 1Q[.v2], of 1Q[v2], and of IQ (Figure 4.3).

The tower rule gives
[1Q[.v2,v2]: 1Q[v2]) [1Q[v2]: IQ).

[1Q[.v2, v2] : 1Q[.v2]) [1Q[.v2]' IQ) ,
which proves that [1Q[.v2, v2] : IQ) is a multiple of 2 and 3, so it is a multiple

of 6.
As .v2 is of degree::; 3 over every field containing IQ, we have
[1Q[.v2, v2] : IQ) = [1Q[.v2, v2] : 1Q[v2]) [1Q[v2, IQ]) ::; 6.
Finally, [1Q[.v2, v2] : IQ) = 6. Consequently, [1Q[.v2, v2] : 1Q[v2]) = 3 and

[1Q[.v2, v2] : Q[.v2J] = 2.
FIGURE 4.3.
A basis of IQ[ .v2, v2] over IQ can be obtained from the bases {I, ij2, 01}
of Q[.v2J over Q and {I, v'2} of 1Q[.v2, v2J over 1Q[.v2} We obtain
{ I, .v2, «4, v2, 25 / 6 ,2 7 / 6 } .
As 27 / 6 = 2 X 2 1/ 6 , a simpler basis is given by { 2 k / 6 , 0 ::; k ::; 5} . This last

remark leads to the equality Q[.v2, v2] = IQ[ V'2]. We could have proved
this equality directly by showing two easy converse inclusions.
3) In this case, we cannot use the same kind of reasoning as in the previous
case, because the remark that [Q[V3, V2] : Q] is a multiple of [Q[V3] : Q]
and of [Q[V2] : Q] only shows that [Q[V3, V2] : Q] is a multiple of 2.
To see, for example, that [Q[V3, V2], Q[V2]] = 2, one must check that
V3 is not in Q[V2]. If V3 = a+bV2, a and bin Q, 3 = a2 +2b2 +2abV2. As
{I, V2} is a basis of Q[V2] over Q, ab = 0; if b = 0, 3 = a 2 is impossible.
If a = 0, then 3 = 2b 2 is impossible (see Exercise 2.1).
Finally, [Q[ V3, V2] : Q] = 4. A basis of the extension is given by
{ 1, V2, V3, V6}.
The inclusion Q[V3 + V2] c Q[V3, V2] is obvious.

Conversely, V3 and V2 are in Q[V3+ V2] because they are the half-sum
and half-difference of a = V3 + V2 and its inverse V3 - V2.
The minimal polynomial of V3 + V2 over Q (which by the above is a
degree 4 polynomial) is obtained by setting a = V3 + V2, giving a 2 =
5 + 2V6, a4 - 10a 2 + 1 = 0. The desired polynomial is X4 - lOX 2 + 1.
4) As j is not real, its degree over Q is .2: 2. We know that j2 + j + 1 = 0,

so the minimal polynomial of j over Q is X2 + X + 1. A basis of Q[j] over
Q is {1,j}.
In what follows, note that two quadratic extensions Q[a] and Q[b] are
equal whenever a E Q[b].
If V3 E Q[j], we would have Q[J] = Q[V3] c lR, which is false. Similarly,
i E Q[j] and j E Q[i] are impossible, since 2j + 1 = iV3 would imply that
Q[i] = Q[j] = Q[V3].
As j is not real, we are sure that it is of degree 2 over Q[ V3] (of minimal
polynomial X2 + X + lover Q[V3]). Hence [Q[V3,j] : Q[V3]] = 2, then
[Q[V3,j] : Q] = 4, and finally [Q[V3,j] : Q[J]] = 2, which means that V3
is of degree 2 over Q[j].
A basis of Q[V3,j] over Q is {I, V3,j,jV3}.
It is clear that Q[V3,j] C Q[V3, i,j]; furthermore, i = (2j +1)/V3 shows
that i E Q[V3,j], so that Q[V3,j] = Q[V3, i,j].
We check similarly that Q[V3,j] = Q[V3, i] = Q[i,j].
Finally, if a = i + V3, we have (a - i)2 = 3, (a - V3)2 = -1, which gives
i and V3 in terms of a, and Q[i + V3] = Q[i, V3] of degree 4 over Q.
Two other bases of Q[V3,j] over Q are thus given by
5) We have
2n 1
lQ[cos 3] = 1Q[-2] = IQ
and
2n r;;
lQ[sin 3 ] = lQ[v3].
Let ( = e 2i7r / 5 . As ( is a root of the polynomial X4 + X 3 + X2 + X + 1,

which is irreducible over IQ, ( is of degree 4 over IQ.
Set a = 2cos(2n/5) = (+ (4. We have a 2 = (2 + (3 + 2 = -a + 1; a
is a root of X2 + X - I , so a = (-1 ± V5)/2. As cos(2n /5) > 0, we see
that cos(2n/5) = (-1+V5)/4, so lQ[cos(2n/5)] = 1Q[V5]. As 2 cos(2n/5) =
(+ (4, we have lQ[cos(2n/5)] C IQ[(] and [IQ[(] : lQ[cos(2n/5)]] = 2 (Figure
4.4).
It follows that sin(2n/5) = Jl-cos 2 (2n/5) = V(10+2V5)/4, since
sin(2n /5) is> O. It is not obvious that sin(2n /5) is of degree 4 over IQ, but
if we set b = 2 sin(2n /5), we have b2 = 10 + 2V5, so b4 - 5b 2 + 5 = 0 and
Eisenstein's criterion works with p = 5. As V5 E lQ[sin(2n /50]' lQ[sin(2n /5)]
is a quadratic extension of lQ[cos(2n /5)]. Note that IQ[(, i] = IQ[(, sin(2n /5)].
<Q[sin(21t/5)]
)
Q[cos(2n:/5)] = Q[V3]
12
FIGURE 4.4.
6) To show that 1Q[J2, V3, V5] is of degree 8 over IQ, it suffices to show
that V5 is of degree 2 over 1Q[J2, V3].
If V5 = a + bJ2 + cV3 + dv'6 for rational numbers a, b, c, d, squaring
gives an equality in 1Q[J2, V3] which, using the fact that 1, J2, V3, v'6
forms a basis, implies that
ba + 3ed 0,
ea + 2bd 0,
ad + be 0,
a2 + 2b2 + 3e2 + 6d 2 5.
The first two equations give a linear system in a and d. Its determinant
is 2b 2 - 3e2 , which is zero only if b = e = O. In this case, a or d is zero
by the third equation, and 5 = a 2 + 6d2 is impossible. If 2b 2 - 3e 2 of- 0,
then a = d = 0, bore is zero by the third equation, and 5 = 2b2 + 3e2 is
impossible.
Later (Exercise 8.10), we will see a generalization of this result to the

degree of an extension of Ql by a finite set of square roots of distinct prime
numbers.
1) Use Eisenstein's criterion or check that P has no rational root: the only
possible roots would be ±1, ±2, and they do not work.
2) The equality a 3 + 2a + 2 = 0 immediately gives the inverse of a since it

implies that a (_(a 2/2) - 1) = 1, so that (1/a) = -(a2/2) - l.
The inverse of a 2 + a + 1 can be found by the method of indeterminate
coefficients or the method of successive Euclidean division (see §4.5.2):
X 3 +2X +2 (X2 + X + 1)(X - 1) + 2X + 3,

X 1 7
X2 +X +1 (2X + 3)( 2 - 4) + 4'
gives Bezout's identity
7
- =-
4
(X
-
2
1) (X
- -
4
3
+ 2X + 2) + (X 2 + X + 1) 2X
2 - 3X
4
+ 5.
It follows that 1/(a2 + a + 1) = (2a 2 - 3a + 5)/7.
This last computation can be done by Euclidean division:
X6 + 3X 4 + 2X 3 + 6X = (X 3 + 2X + 2)(X 3 + X) - 2X2 + 4X,

so u = -2a 2 + 4a = 2a(2 - a).
Alternatively, we can compute using a table of the an expressed in terms
of basis vectors (Table 4.1 below), which is a good way of computing in an
extension when one does not have a computer package for formal compu-
tation at one's disposal.
1 a a2
1 1
a 1
a2 1
a3 -2 -2
a4 -2 -2
a5 4 4 -2
a6 4 8 4
TABLE 4.1.
3) As u is not in Q, it is of degree 3 over Q, and we obtain its minimal

polynomial over Q by computing the powers of u in the basis 1, a, a 2 , using
Table 4.2, and then looking for a linear combination of 1, u, u 2 , u 3 equal
to zero:
1 a a2
1 1
u 4 -2
u2 32 24 8
u3 32 192 64
TABLE 4.2.
We find
0: + 32')' 32,
4(3 + 24')' 192,
- 2(3 + 8')' 64.
This gives 0: = -224, (3 = 0, ')' = 8, and the minimal polynomial of u is
X 3 - 8X 2 + 224.

1) The first exercise is trivial, but its results are useful.
a) If n = 1, then L = K. If L' eLand [L' : K] = n, we have [L' : L] = 1

so L' = L.
b) If n = [L : K] is prime and K C L' c L, then [L' : K] divides n so it
is equal to 1 or n, and either L' = K or L' = L by a).
2) We have already seen some examples of this reasoning; both [K[a] : K]

and [K[b] : K] divide [K[a, b] : K] which is thus a multiple of mn, and more-
over, [K[a, b] : K[all :S [K[b] : K] implies that [K[a, b] : K] :S mn. Finally,
[K[a, b] : K] = mn, so [K[a, b] : K[bll = m, and P is irreducible over K[b].
As [K[a] n K[b] : K] divides both [K[a] : K] and [K[b] : K], it is equal to
1 and we have K[a] n K[b] = K.
3) As x 2 E K[x], x 2 is algebraic over K and K[x 2] C K[x]. As x is a root of

the polynomial X2 - x 2 with coefficients in K[x 2], x is algebraic of degree
at most 2 over K[X2]. Moreover, [K[x] : K[x 2J] divides [K[x] : K], which is
odd, so it is equal to 1. Hence K[x] = K[x2]. The minimal polynomial of x
over K gives an expression for x in terms of x 2 by regrouping the terms of
even order and the terms of odd order.
4) It suffices to note that deg(P) = [K[a] : K], so deg(P) divides n.
5) A quadratic extension of IQ is an extension by a number of the form

(x ± y'y)/2 with x, y rational; thus it is an extension of the form IQ[JYl. If
y = m/n with m and n rational, then IQ[JYl = lQ[vmn/n] = 1Q[y'mn].
If mn = d2 a for a squarefree integer a, then
1Q[y'mn] = lQ[dv'al = 1Q[v'al.
6) By d'Alembert's theorem, C has no algebraic extensions of finite degree
except for itself. The field C(X) of rational functions with complex coef-
ficients is an example of an extension of C that is not algebraic, and has
infinite degree.
7) Let P(X) = L akXk be the minimal polynomial of a over L. In

O$k$n
fact, a lies in the extension K[ao, . .. ,an][a], which is an extension of finite
degree of K since it is an extension of finite degree of K[ao, . .. ,an] and
K[ao, . .. ,an] is an extension of finite degree of K (Figure 4.5). This solves
the problem.
FIGURE 4.5.
8) Otherwise, e and 7r would be of finite degree over IQ, so they would be

algebraic over IQ.

COMMENTARY. - To find a non-zero polynomial which vanishes at an al-
gebraic number given as a polynomial function of other algebraic numbers,
we can use resultants, but thr polynomial we obtain may not be minimal.
1) a) To compute Resx(F(X), G(Z -X)), we work with polynomials with
coefficients in K[Z], so that the resultant lies in the same ring. To show
that R(a + b) = 0, it suffices to check that F(X) and G(a + b - X) have a
common root, but a itself is such a root.
Note that even if F and G are the minimal polynomials of a and b, we
do not necessarily find the minimal polynomial of a + b, as the degree of
the resultant is the product of the degrees of a and b over K.
b) As Resy(Z - T(X, Y), C(Y)) lies in K[Z][X], R(Z) lies in K[Z].

For Z = T(a, b), F(X) and Resy(T(a, b) - T(X, Y), C(Y)) are two poly-
nomials in X having a a as a common root, since the two polynomials
T(a, b) - T(a, Y) and C(Y) in Y have b as a common root.
c) ab is a root of R( Z) since a is a common root of the polynomials F( X)

and XnC(ab/ X).
d) We can take R(Z) = Resx(F(X), Z - T(X)) since F(X) and T(a)-

T(X) have a as a common root.
2) These numbers lie in extensions of Q of degree n = 2,6,4,6,3, respec-

tively. To find a polynomial that vanishes at these numbers, we can
a) look for a relation of linear dependence between the k-th powers of

the numbers for 0 :::; k :::; n, in a suitable extension of Q;
b) use resultants as described above;
c) proceed more directly if the expression of the number allows this.

As x = (2/3) + V577satisfies (x - (2/3))2 = (5/7), x is a root of X2 -
(4/3)X - (17/63). As x is of degree 2 over Q, because Q[x] = Q[~] =
Q[J35], this polynomial is the minimal polynomial of x over Q.
If y = h + «3, (Y - h) 3 = 3, y3 + 6y - 3 = h(3y2 + 2) and squaring,
we get y6 - 6y4 - 6y3 + 12y2 - 36y + 1 = o. We can also obtain this result
by computing Resx(X3 - 3, (Z - X)2 - 2).
The polynomial P(X) = X 6 - 6X 4 - 6X 3 + 12X2 - 36X + 1 is the
minimal polynomial of y over Q, which is of degree 6. There are several
ways to prove that it is of degree 6. For example, note that y generates
Q[h, «3] since h = (y3 + 6y - 3)/(3y2 + 2).
If z = h + ~, we have (z - h) 2 = h, so that z4 - 4z2 - 8z + 2 = o.
The corresponding polynomial is the minimal polynomial of z over Q, which
is irreducible by Eisenstein's criterion.
If t = a + h, then (t - V2)3 + t - V2 + 1 = 0, so that if we separate the
terms containing h and square, we obtain t 6 -4t 4+2t3+13t 2+14t-17 = o.
This is the minimal polynomial of t over Q, by a reasoning analogous to
the one used for y.
Ifu = b2+b, u 2 = -6-9b-2b2, u 3 = 33+33b-9b2 and u 3 = a+,Bu+,u 2
leads to a linear system whose solution gives u 3 + 6u 2 + 21u + 3 = O. This
is a minimal polynomial for u since b lies in a cubic extension of Q and not
in Q (one can also use Eisenstein's criterion).
5
Constructions with Straightedge and
Compass
For the ancient Greeks, a geometric construction was a construction done

using only straightedge and compass (a "straightedge" is a ruler not marked
with any measurements). In this chapter, we consider planar problems, in
the sense of elementary geometry. The verb "construct" means construct
with straightedge and compass, according to the procedures described more
precisely below.
5.1 Constructible Points

Let E be a set of points in the plane. Write V E for the set of lines in the
plane passing through two distinct points of E, and let CE denote the set
of circles in the plane whose center is a point of E and whose radius is a
distance between two distinct points of E.
DEFINITIONS. - A point of the plane is said to be constructible in one step

from E if it is either
1) an intersection of two lines in VE,
2) an intersection of a line in V E and a circle in CE, or
3) an intersection of two circles in CEo

A point P in the plane is said to be constructible in n steps from E if there
exists a finite sequence PI, ... , Pn of points in the plane such that Pn = P
and for i = 1, ... , n, Pi is constructible in one step from E U {Pj;j < i}.

80 5. Constructions with Straightedge and Compass
A point P in the plane is said to be constructible from E if there exists

an integer n such that P is constructible in n steps from E.
REMARK. - If E contains only one element, then no new points can be

constructed from it.
The number of steps depends on the construction procedure, though

there exists a minimal such number.
5.2 Examples of Classical Constructions

5.2.1 Projection of a Point onto a Line
Let (AB) be a line passing through two points A and B, and let M be a
point not on (AB). We construct the projection H of M onto (AB) from
{ A, B, M} by the following steps:
1) draw the circle of center A and radius AM,
2) draw the circle of center B and radius BM that intersects the previous
circle at N, and
3) draw the line (MN), that intersects the line (AB) at H (Figure 5.1).
B A
FIGURE 5.1. Construction of a projection
5.2.2 Construction of an Orthonormal Basis from Two

Points
Let 0 and A be two distinct points in the plane which we assume to be at a

distance of 1 from each other. We construct an orthonormal basis (0, A, B)
from {O, A} by the following steps:
1) draw the circle C of center 0 and radius OA, which intersects (OA)
at A';
5.2 Examples of Classical Constructions 81
2) draw the circle of center A and radius AA';
3) draw the circle of center A' and radius AA', which intersects the
previous circle at M and N; and
4) draw the line M N which intersects the circle e at B and B' (Figure
5.2).
M
A' A
N
FIGURE 5.2. Construction of an orthonormal basis
REMARK. - Using the two above constructions together, we can construct

the line perpendicular to a given line, passing through a given point which
mayor may not lie on the line.
5.2.3 Construction of a Line Parallel to a Given Line

Passing Through a Point
Let (AB) be a line and M a point not on it. We construct the line parallel to
(AB) passing through M from the set of points {A, B, M} by the following
steps:
1) draw the circle of center A and radius R = AM, which intersects
(AB) at C and e';
2) draw the circle of center e and radius R; and
3) draw the circle of center M and radius R, which intersects the previ-
ous circle at A and N (Figure 5.3).
~N
FIGURE 5.3. Construction of a line parallel to a given line
The desired line is then (M N).

5.3 Lemma
Let E be a set of points of the plane containing at least two elements 0 and
A. Let B denote a point such that n = (0, A, B) is an orthonormal basis
(which by §5.2.2 is constructible from 0 and A), and let K = Q(F) be the
extension of Q generated by the set F of the (real) abscissas and ordinates
of the points of E in this basis. Then
1) every line in DE has an equation in n of the form
ax+by+c=O with a,b,c E K;
2) every circle in CE has an equation in n of the form
x 2 + y2 + ax + by + c = 0 with a, b, c E K.
PROOF. - If a line in DE passes through two distinct points of E, with

coordinates respectively given by (XI,YI) and (X2,Y2) in the basis n, then
its equation is given by
which has the desired form.

Moreover, if a circle of CE has center (xo, Yo) in n and radius equal to
the distance between two points of E with coordinates (Xl, YI) and (X2' Y2)
in n, then its equation is given by
which has the desired form. <)
5.4 Coordinates of Points Constructible in One

Step
PROPOSITION. - Keep the notation of Lemma 5.3. If P is a point in the

plane with coordinates (p, q) in n, constructible in one step from E, then
K(p, q) is equal to K or to a quadratic extension of K.
PROOF. - If P is the intersection of two lines in DE given by equations
ax + by + c 0,
a'x+b'y+c' 0,
with a, a', b, b', c, c' in K, then ab' - a'b =t 0 since the two lines are not
parallel; Cramer's rule then shows that p and q lie in K, so that K(p,q) =
K.
5.5 A Necessary Condition for Constructibility 83
If P is the intersection of a line in 'DE and a circle in CE given by

equations
ax+ by+c 0,
x 2 + y2 + a' x + b' y + c' 0,
with a, a', b, b', c, c' in K, we see that if a -I- 0, p = -(bq + c)/a so p and
q lie in the same extension of K. Then
bq +C)2
(-
a
- +q 2 - a - - + b' q+c, =
,bq+c
a
°
shows that q is a root of a quadratic polynomial P E K[X]. The element
q lies in K or in a quadratic extension of K according to whether P is not
°
or is irreducible over K. If a = 0, then b -I- and a similar reasoning holds.
If P is the intersection of two circles in CE given by equations
x 2 + y2 + ax + by + c 0,
x + y2 + a' x + b' y + c'
2 0,
with a, a', b, b', c, c' in K, we reduce to the preceding case by noting that
P is the intersection of the first circle with the line given by the equation
(a - a')x + (b - b')y + c - c' = 0.
5.5 A Necessary Condition for Constructibility

PROPOSITION. - Keep the notation of Lemma 5.3. For every point P =
(p, q) constructible from E,
1) there exists a finite sequence of fields (Ki )O:5i:5m, each of which is a
quadratic extension of the preceding one, with Ko = K, Km C JR, and
p,q E Km; and
2) p and q are algebraic over K; their degrees over K are powers of 2.
PROOF. -
1) We use induction on the number n of steps needed to construct P.

If n = 0, the result is obvious.
Suppose that for every point Q constructible in n steps from E, via
the sequence of points (Qjh5,i:5n, there exists a finite increasing se-
quence (Ks)o:s;s:5r such that Ko = K, Kr C JR, each field is quadratic
over the preceding one and the coordinates of the Qj, 1 :S j :S n, lie
in K r .
If P is constructible in n + 1 steps from E, there exists a sequence

of points (Pi h<i<n+1 such that Pn +1 = P, and for i = 0, ... , n, PHI
is constructible in one step from E U {Pj ; j ::; i}. By the induction
hypothesis, there exists a finite increasing sequence (Ks)o:s;s:s;r such
that Ko = K, Kr C 1R, each one is quadratic over the preceding one
and the coordinates of the Pi, 1 ::; i ::; n, lie in K r . As the point P is
constructible in one step from E U {Pi; i ::; n}, Proposition 5.4 applies
(we add a term to the sequence if and only if p and q do not lie in
Kr).
2) As [Ki : K i - I ] = 2 for i ::; m, the tower rule gives [Km : K] = 2m.
This formula also shows that the degrees of p and q over K must
divide 2m , so they are powers of 2. <>
5.6 Two Problems More Than Two Thousand

Years Old
The second part of Proposition 5.5 provides a negative answer to problems
set by the Greeks over two thousand years ago. This was first proved by
Wantzel, who published his result in 1837 (although Gauss may have known
it as early as 1796).
Pu M. L. WANTZEL,
Eleve-Jngenicur des Ponts-eL-Chaussees.
Supposons qu'un probleme de Geometrie puisse etre resolu par des

intersections de lignes droites et de circonferences de cerde : si ron
joint les points ainsi ohtenus avec les centres des ces-des et avec les
point; qui determinent les droibls on formera un encbainement de
triangles rectilignes dont les ~Hements pouITont etre calcules par les
formules de la Trigonometrie; d'ailleul's ces formules sont des equa-
tions algehriques qui ne renfermcnt les cotes et les lignes trigonome-
triques des angles qu'au premier et au second degre; aiosi }'inconnue
pcincipale du probleme s'ohtiendra par la resolution d'une serie d'e-
quations du ~cond degre dont les coefli"cients seront fonctions ration-
nelles des donnees de la question et des raeines des equations prece-
dentes. D'apres eela, pourreconnattre si 1a construction d'un probleme
de Geometrie peut s'efi'ectuer avec la regIe et Ie com pas, il faut chercher
s'ilest possible de faire d.'ependre les racines de l'equation a laquellc il
conduit de celles d'unsysteme d'equations du second degre composees
COmme on vient de l'indiql1er. Nous traiterons seulement ici Ie cas ou
requation du probleme est a1gebrique.
FIGURE 5.4. The beginning of Wantzel's proof
5.7 A Sufficient Condition for Constructibility 85
5.6.1 Duplication of the Cube

A legend recounted by Eratosthene (around 276-196 B.C.) recounts that
while the plague was raging in Delos, a small island of the Cyclades,
Apollo's oracle declared that Apollo desired a cubic altar whose size was
exactly the double of his former altar, and that he would stop the epidemic
only when he determined that such a cubic altar had been built. Thus, the
length of a side of the new altar was to be the length of a side of the old
one multiplied by ~.
It turns out to be impossible to construct a segment whose length is
~ times a given length using only straightedge and compass. Indeed, this
would be equivalent to constructing a point with coordinates (~, 0) in an
orthogonal basis constructed from 0 and A. But ~ is of degree 3 over Q.
Of course, 3 does not divide any power of 2, so Proposition 5.5 shows that
such a construction is impossible.
5.6.2 Trisection of the Angle

The trisection of the angle is the problem of dividing an arbitrary angle
into three equal parts using only straightedge and· compass. The ancient
Greeks actually possessed other methods for trisecting angles (see Exercise
5.5).
Constructing an angle 8 when its triple is known is equivalent to con-
structing the point (cos 8, sin 8) of the unit circle in the plane, with respect
to a basis R = (O,A,B), when the point (cos38,sin38) is known. The
formula cos 38 = 4 cos3 8 - 3 cos 8 (due to Viete) shows that cos 8 is a root
of the polynomial 4X 3 - 3X - cos 38.
In general, this polynomial is irreducible over Q[cos38]; for example,
when 8 = 40°, it is equivalent to 8X 3 - 6X + 1, which is irreducible over
Q because it has no rational roots. The same reasoning as in the preceding
section then shows that the construction is impossible. It follows that angles
of (120/n)0 for integers n 2: 1 are never "trisect able" (otherwise, the 120°
angle would be).
5.7 A Sufficient Condition for Constructibility

PROPOSITION. - Let us keep the notation of Lemma 5.3.
1) Every point with coordinates in K = Q(F) is constructible from E.
2) Every point whose coordinates lie in a quadratic extension of K is
constructible from E.
3) Consider a point P with coordinates (p, q) such that there exists a

finite increasing sequence of fields (Ki)o::;i::;m, each one quadratic over
the preceding one, with Ko = K, Km C lR and p, q E Km. Then P is

constructible from E.
PROOF. -
1) Let p and q lie in K. To show that the point P = (p, q) is constructible

from E, it suffices, by §5.2.2, to show that the points (p,O) and (0, q)
are constructible from E. Let us show that (p,O) is constructible.
By §4.3.1, p is of the form S(al,"" ak)/T(al, ... , ak), where k is an
integer, Sand T are polynomials in Q[Xl , ... ,Xkj, and al,'" ,ak are
elements of F. We construct (p,O) one step at a time by noting that if
(x, 0) and (y,O) are constructible from E, then (-x, 0) and (x + y, 0)
are constructible from E (this is trivial), and also (l/x, 0) and (xy, 0)
are constructible from E. Indeed, to construct (1/ x, 0) , we construct
the line that passes through the point B = (0,1) and is parallel to the
line passing through A = (1,0) and (0, x), and to construct (xy, 0),
we construct the line passing through (0, y) and parallel to the line
passing through the points B and (x,O) (Figure 5.5).
B=(O,l) B=(O,l)
(O,x) (O,y)
o A=(1,O) '----- (l/x,0) o (xy,O) (x, 0)
FIGURE 5.5. Construction of (l/x, 0) and (xy, 0)
2) Let p be of degree 2 over K. Then p is a root of a quadratic polynomial

T with coefficients in K. If T(X) = X2 + aX + b, then
-a ± Ja 2 - 4b
p= 2
if we set c = a 2 - 4b, the construction of (p,O) uses the construction

of (1, vIC) knowing (c,O).
We first construct the point C = (c + 1,0), the midpoint M of
DC; to do this, construct the mediatrix of DC. Then we construct
the intersection D of the circle of center M and radius MD with the
line perpendicular to DA passing through A. Then D = (1, v'c), so
we conclude with the last step (Figure 5.6).
(0, Vc) I---."r D
o A=(l;O) C=(c+l,O)
FIGURE 5.6. Constructions of (D,ye)
3) This result is easily proved by induction, using the two previous results.
This concludes the proof. <>
COMMENTARY. - The constructions described above are given in the first
pages of Descartes' Geometry. They enabled him to consider the product of
two lengths x and y as a length, rather than considering it as the rectangle
with sides x and y, and this led him to suppress the homogeneity condi-
tions that had made the work of his predecessors so heavy: " ... OU il est
a. remarquer que, par a2 , ou b3 , ou semblables, je ne cOll(,;ois ordinairement
que des lignes toutes simples, encore que pour me servir des noms usites
en l'algebre, je les nomme des carres ou des cubes, etc." (" ... it should be
noted that, by a 2 , or b3 , or others, I do not ordinarily conceive anything
but perfectly simple lines, although in order to employ the names usual in
algebra, I call them squares, cubes, etc.")
Exercise 5.1. Roots of quadratic equations

Let 0 AB denote an orthogonal basis of the plane.
1) Given two strictly positive real numbers sand p, construct the roots
of the equation X2 - sX + p = 0, starting by constructing a segment
of length ,;p. Recover the usual algebraic condition geometrically.
2) Construct a segment of length n.

Exercise 5.2. Construction of the regular pentagon
1) Check the classical construction of the regular polygon with five sides,
with one vertex labeled A, inscribed in a circle C of center 0 and
radius OA.
• Draw a diameter BB' perpendicular to ~A.

• Let I denote the midpoint of OB'; draw the circle of center I

and radius IA which intersects OB at D.
• The length of AD is equal to the length of the side of the regular

pentagon to be constructed.
2) Construct a regular polygon with 30 sides using straightedge and

compass.
COMMENTARY. - A different construction of the regular pentagon is given

in book 4, proposition 11, of Euclid's Elements (around 300 B.C.). It is the
climax of this part of the Elements; the propositions of books 2, 3, and 4
build methodically toward this end. The solution is entirely geometric.
Exercise 5.3. Constructible elements of degree 4

1) Let x be a complex number of degree 4 over Q, and let P(X) =
X4 + pX 2 + qX + r be its minimal polynomial over Q. Show that
there exists a quadratic extension L of Q, contained in Q[x], if and
only if the polynomial R(X) = X 3 + 2pX 2 + (p2 - 4r)X -l has a
rational root.
2) Show that a real root of the polynomial P(X) = X4 + 2X - 2 is not

constructible.
Exercise 5.4. al Biruni's third-degree equation for the 20° angle

Let A and B denote two adjacent vertices of a regular I8-sided poly-
gon inscribed in a circle of center 0, as in Figure 5.7. The 10th century
mathematician al Biruni constructed the isosceles triangles ABC with
Con OB and AB = AC, ACD with D on OA and AC = DC, CDE
with Eon OA and CD = DE.
A
x~
B H C E o
FIGURE 5.7. al Biruni's figure
1) Show that OE = AB.
2) Set OA = 1 and AB = x. Show that x 3 + 1 = 3x.

Exercise 5.5. Trisection of angles

1) Archimedes' method (Figure 5.8):
c A
FIGURE 5.8. Archimedes' method
Consider a circle of center 0 and radius T, and an angle AOB where

A and B are two points on the circumference of the circle. Suppose
we are in possession of a straightedge marked with two points at a
distance T from each other, and that with this straightedge, we know
how to construct a line passing through B, intersecting the extension
of the segment [~Al toward 0 at C, and intersecting the circle at D
in such a way that [CD] is of length T.
Show that the angle DCO is one-third of the angle AOE.
2) Pappus' method.
A C
o H
FIGURE 5.9. Pappus' method
Suppose we are given a right triangle AOH at H whose side [~Al

is of length a. Draw the line D parallel to (OH) passing through
A. Suppose that we are in possession of a straightedge marked with
two points at a distance of 2a from each other, and that with this
straightedge, we know how to construct a line passing through 0,
intersecting the segment [AH] at B and the line D at C in such a
way that the length of the segment [BC] is 2a (Figure 5.9).
Show that the angle BOH is equal to one-third of the angle AOH.
3) Origami (the Japanese art of paper-folding, called kami in Japanese).

, - _ - - - , - = -_ _ _ _ _ _--, C
r-~~-------~l
'r-+-"';---'-------~ G
L----------~D
FIGURE 5.10. Trisection by origami
All straightedge and compass constructions can by done by origami,

and many others as well. The origami method for trisecting an angle is
due to the Japanese Abe. Take a rectangular piece of paper ABCD
and form the angle (J = DAE by folding along AE (Figure 5.10).
By two successive foldings of an equal, arbitrary width, we obtain
segments FG and HI parallel to AD such that AF = F H. Then fold
in such a way as two simultaneously bring A to a point on FG and
H to a a point of AE.
Show that the angle between AB and the axis of this last fold is
equal to (J /3.

(O,YP) 1----71" D
c
o a H b C=(s,O)
FIGURE 5.11. Construction of the roots of a quadratic equation
1) We first construct a segment of length vp (see Proposition 5.7. 2», and

then a point C with coordinates (8,0) and a circle C of diameter DC. We
construct a point D with ordinate vp of C, which projects to H on DC.
The point H divides DC into segments of lengths a and b, which are the
roots of the given equation.
The construction is possible on the one hand if p ::::: 0, on the other

hand if there exist points with ordinate vP on the circle C, i.e. vP ::; 8/2;
we recover the condition 82 - 4p ::::: O.
2) We use the construction of Proposition 5.7 2) twice, first to construct a
segment of length v'2, and then to construct a segment whose length is a
square root of v'2.
1) Using Pythagorus' theorem, we easily compute AD; if OA = 1, it is

J 10 - 2,j5/2 and we check that this is equal to 2 sin (7r / 5) by the identity
27r -1 + ,j5 . 2 7r
cos - = = 1 - 2sm -
5 4 5
(see part 5) of Exercise 4.5).
2) Let D' denote an intersection of the circle of center A and radius AD
with C. The equality 7r /15 = (27r /5) - (7r /3) shows that an arc ofthe circle
of center A and radius OA cuts the circle C at a point E between A and
D' such that D' and E are adjacent vertices of a regular 30-sided polygon.
1) In the special case of a biquadratic equation, q = 0, R(O) = 0, and the

intermediate extension always exists.
Assume the existence of L. The number x is of degree 2 over L; write
X2 + aX + b, with a, b in L for its minimal polynomial over L. This polyno-
mial divides P in L[X], so we have P(X) = (X2 + aX + b)(X2 + eX + d),
with e and d in L. By Descartes' method, we know that a 2 is a root of R.
If a2 is not rational, Q[a 2] = L so a 2 is of degree 2 over Q, its minimal
polynomial over Q divides R in Q[X], and thus R has a linear factor in
Q[X], i.e. a rational root.
Conversely, if R has a rational root, there exists a decomposition of P
in Q[X] into a product of the form (X 2+aX +b)(X2+eX +d), where a 2 is a
rational root of R. Set L = Q[a]. We know that L contains b,e,d, therefore
x, a root of one of the two factors, is of degree ::; 2 over L. As [L: Q] ::; 2,
we have [L[x] : Q] ::; 4, and as Q[x] C L[x], we have Q[x] = L[x]. We
conclude that L gives the desired result.
2) P is irreducible by Eisenstein's criterion, and it has a real root since

P(O) = -2, P(l) = 1. The corresponding polynomial R, equal to X 3 +
8X - 4, has no rational root, since the possible candidates ±l, ±2, ±4 fail.
We conclude by using 1) and the constructibility theorem.
6
K -Homomorphisms
In this chapter and the coming ones, we continue to restrict our attention to
the situation of fields that can be realized as subfields of the field of complex
numbers C. However, the definitions and results all generalize directly to
arbitrary fields contained in an algebraically closed field C of characteristic
o (for fields of characteristic p i- 0, see Chapters 14 and 15).
One surprising aspect of the theory is the very minor role played by
polynomials, which appeared in previous chapters as the main subject of
Galois theory. This is due to the efforts of Dedekind at the end of the 19th
century, and Emil Artin in the 1920s and 1930s, to clarify the linear aspects
of Galois theory - in particular, the notion of K-homomorphisms, which
extends the original idea of permutations of roots of a polynomial.
6.1 Conjugate Numbers
DEFINITION. - Let K be a field, which as before we take to be an inter-

mediate extension between Q and C. Let a be a complex number that is
algebraic over K, with minimal polynomial P over K. Each root of Pin C
is called a conjugate of a over K. Two roots of an irreducible polynomial
of K[X] are said to be conjugate over K.
EXAMPLES. -
1) The numbers i and -i are conjugate over Q and over JR., but not over
C.

94 6. K-Homomorphisms
2) The numbers v2 and -v2 are conjugate over Q, but not over Q[v2].
3) The numbers ij2, j ij2 and j2 ij2 are conjugate over Q.
6.2 K - Homomorphisms
6.2.1 Definitions
Let K be a field, and let Land L' be two extensions of K contained in
<C. A K -homomorphism from L to L' is a homomorphism of rings with
unit from L to L', which leaves the elements of K invariant (i.e. whose
restriction to K is the identity on K). In other words, a K-homomorphism
is a homomorphism from the K-algebra L to the K-algebra L'.
We will often use the letter a to denote a K-homomorphism. We will
use the term K-isomorphism, respectively K-automorphism when a is an
isomorphism, respectively an automorphism.
EXAMPLES. -
1) Complex conjugation, which sends a complex number a + ib, with a
and b real, to a - ib, is an lR-automorphism of C; we will often consider
its restriction to subfields.
2) The map a : Q [v2] -> C defined by a (a + bv2) = a - bv2 with a
and b rational, is a Q-automorphism.
3) If Land L' are two extensions of K, and if K is an extension of K',
then every K-homomorphism a : L -> L' is a K'-homomorphism,
and in particular a Q-homomorphism.
6.2.2 Properties
Let a : L -> L' be a K-homomorphism. It has the following properties:
1) it is a K-linear map between L and L' considered as K-vector spaces;
2) it is an injective map, like every ring homomorphism between two
fields;
3) its image is a field a(L) which is K-isomorphic to L;
4) if L is of finite degree over K, a K-homomorphism a : L -> L is a
K-automorphism of L, because since a is a linear injective endomor-
phism of a finite-dimensional K-vector space, it must be surjective;
5) if L = K[all.'" an], a K-homomorphism is defined over L by its
values at the generators of L as an K-algebra, i.e. by its values at
all." ,an (which can be related; think of the case of Q j, [i, J3] for
example). If L = K[a], it suffices to specify a(a) in order to completely
6.3 Algebraic Elements and K-Homomorphisms 95
determine 0-. For an element of the form P(a) with P E K[X], we

have o-(P(a)) = P(o-(a)).
6.3 Algebraic Elements and K - Homomorphisms

6.3.1 Proposition
Let K be a field, and let a be an algebraic element of finite degree nand
minimal polynomial P over K.
1) If L is an extension of K, if a ELand if 0- : L -+ <C is a K-

homomorphism, then o-(a) is a conjugate of a over K. If L contains
all the roots of a polynomial S in K[X], then 0- induces a permutation
of the set of roots of S.
2) If b is a conjugate of a over K, there exists a unique K -isomorphism

0- : K[a] -+ K[b] such that o-(a) = b.
3) The number of distinct K-homomorphisms from K[a] to <C is equal

to n.
PROOF. -
1) Let P(X) L akXk be the minimal polynomial of a over K.

O::;k::;n
Because 0- is a K-homomorphism and the coefficients of P lie in K,
we have o-(ak) = ak for k = 0, ... ,n, so
Thus o-(a) is a conjugate of a over K.

Because 0- is injective, the rest of the argument works by decom-
posing S into a product of irreducible factors.
2) Consider the diagram in Figure 6.1, where 7r denotes the canonical

surjection, f and 9 are the homomorphisms defined by f(X) = a and
g(X) = b, and 'P and 'IjJ are their factorizations through 7r.
K[X]
f 1 g
K[X]/(P)
K[a]
7 cr
~ • K[b]
FIGURE 6.1.
Set IJ = '¢¢-l. Like 'P and ,¢, it is a ring homomorphism (see §4.5.2);
we have IJ f = '¢'P- 1f = '¢7r = g, so
lJ(a) = IJf(X) = g(X) = b.
Thus, for every element x of K, we have lJ(x) = IJf(x) = g(x) = x;

this gives the result.
Let us show the uniqueness of IJ. If IJ' satisfies the same properties,
then the restrictions of IJ f and IJ' f to K are the identity and IJ f(X) =
b = IJ' f(X); these two conditions imply that IJ f = IJ' f· As f is
surjective, it follows that IJ = IJ'. An element of K[a] of the form
L ak ak has image L akb k .
l~k~n l~k~n
3) Because P is irreducible of finite degree n, it has n distinct roots in Co

By part 2), each of these roots gives rise to a unique K-isomorphism
from K[a] to a subfield of Co Then part 1) implies that there are no
others. \;
6.3.2 Example
The minimal polynomial X 3 - 2 of -V2 over IQ has roots -V2, j -V2, and j 2 -V2
in Co Thus there exist three IQ-homomorphisms from IQ[ -V2] to C defined
by the possible images of -V2 (Table 6.1).
-V2
IJI ~
1J2 j~
1J3 j2~
TABLE 6.1.
6.4 Extensions of Embeddings into C 97
The first one, aI, is the inclusion map; the other two, a2 and a3, can be
expressed in the basis { 1, <12, W} of Q [<12] over Q by
a2 (a + b<12 + cW) a + bj<12 + cj2W,

a3 (a + b<12 + cW) a + bl<12 + cjW,
with a, b, c rational.
6.4 Extensions of Embeddings into C
6.4.1 Definition
An embedding of a field L (contained in q into C is a homomorphism
a : L ---+ C of rings with unit.
If L is an extension of a field K, a K-homomorphism of L to C is an
embedding. This generalization is useful for proving the corollary to Propo-
sition 6.4.3 below.
6.4.2 Proposition
Let L be a field, and let a : L ---+ <C be a field embedding. Let a be an algebraic
number of finite degree n over L. Then there are exactly n embeddings of
L[a] into C extending a.
PROOF. - Let P denote the minimal polynomial of a over L. Consider the
diagram in Figure 6.2, where
• i and j are the canonical inclusions;
• a' is the homomorphism defined by a'(X) = X and a'iL = a;

• 7r and p are the canonical projections;
• a" is defined by passing to the quotient of a';

• b is one of the n roots of a' (P) in C;
• f and 9 are the homomorphisms defined by f(X) = a and g(X) = b;

fi and gj are the canonical inclusions;
• r.p and 'IjJ are the isomorphisms defined as in §4.5.2 (to see that 'IjJ
is an isomorphism, note that if a'(P) is reducible over a(L), i.e. if
a'(P) = ST with Sand T non-invertible, then P = a,-l(ST) =
a,-l(S)a,-l(T), which shows that a,-l(S) or a'-l(T) is invertible
and consequently S or T is also invertible, which is a contradiction);
L .. cr(L)
t
L[X] • cr(L)[X]
L~vJ<Scr(L)[J]I~\
L[a] .. cr(L)[b]
FIGURE 6.2.
Firstly, T extends (7 because T fi = 'Ij;(7" rp -1 fi = 'Ij;(7" 7ri = gj(7. If T' is a

second extension of (7 to L[a] such that T' (a) = b, we have T' fi = gj(7 = T fi
and T' f(X) = T'(a) = b = T f(X), which shows that T' f = T f, by the
universal property of L[X]. Hence T' = T, since f is surjective. As T(a) is
necessarily one of the n roots of (7'(P), since T(P(a)) = T(f(P)) = ... =
(7'(P)(T(a)), the proposition is proved. <>
The following formula can also be deduced from the proposition:
6.4.3 Proposition
Let L be a field, and let M c <C be an extension of finite degree n of L.
Let (7 : L -> <C be an embedding. The number of embeddings of Minto <C
extending (7 is equal to n.
PROOF. - We use induction on n. If n = 1, M = L and the result follows.
Assume n > 1. Figure 6.3 gives a diagram of the situation.
M~
degreesi ~
L[a] Or ~ C
degreerL~
FIGURE 6.3.
6.5 The Primitive Element Theorem 99
Suppose that the property holds for every extension of finite degree
strictly less than n, and let M be an extension (contained in C) of fi-
nite degree n of L. Let a be an element of M - L; then a is algebraic over
L of finite degree r > 1. The preceding proposition shows that there exist
r extensions of (j to L[a]. If M = L[a], we obtain the desired result; other-
wise, by the induction hypothesis, since s = [M : L[a]] < n, we can assert
that for each of the r extensions of (j to L[a], there exist s extensions to
M. The result then follows from the tower rule. <)
COROLLARY. ~ Let K be a field, and let L c C be an extension of finite

degree n of K. There exist n K -homomorphisms from L to C.
PROOF. ~ It suffices to apply Proposition 6.4.3 to the embedding given by
the inclusion of K into C (Figure 6.4). <)
degree n
inclusion
K
FIGURE 6.4.
6.5 The Primitive Element Theorem

6.5.1 Theorem and Definition
Let K be a field contained in C, and let L be an extension of finite degree
n of K. There exists an element a of L such that L = K[a]. Every element
generating L is called a primitive element for the extension L of K.
COMMENTARY. ~ This result is due to Galois, who gave it without proof and
deduced from it that the roots of a polynomial can be expressed rationally
in terms of one particular element, using the following lemma:
"Lemma II. Given an arbitrary equation, which has no equal roots, whose
roots are a, b, c, ... , we can always form a function V of the roots, such
that none of the values we obtain by permuting the roots in this function
in every possible way are ever the same.
For example, we can take:
V = Aa + Bb + Cc + ... ,
where A, B, C, ... are suitably chosen integers."

A primitive element used to be called a Galois resolvent. The same term

was also used for the minimal polynomial of such an element.
We prove the result here for subfields of C; it generalizes not only to
fields of characteristic 0 but even to all separable algebraic extensions of
finite degree (defined in Chapter 15).
PROOF. - By §6.4.4, there exist n distinct K-homomorphisms from L to
C; let us denote them by 0"1, .•• , O"n. By the above lemma applied to the
hyperplanes ker(O"i - O"j), 1 $ i < j $ n, there exists an element a in L
whose images under the 0"1, ..• , O"n are all distinct. For i = 1, ... , n, O"i(a)
is a conjugate of a over K, a is of degree 2:: n over K, so [K[a] : K] = n
and L = K[a]. <)
LEMMA. - Let V be a vector space over an infinite field k, and let HI, ... ,Hr
be a finite family of strict subspaces V. Then V i= U
Hi·
l:::;i:::;r
PROOF. - We use induction on r.
If r = 1, the result is clear. If r > 1, suppose the result holds for every
family of r - 1 strict subspaces of V. Suppose that V = U
Hi; then by
l:::;i:::;r
the induction hypothesis, there exists an element x in V which does not
belong to U Hi. So x E Hr. On the other hand, let y be such that
y ~ H r , so Y E U Hi· The set E = {X+Ay;A E k} is infinite; thus

1:::;i:::;r-1
there exists an integer i such that 1 $ i $ r, and distinct elements A and
11 in k, such that x + Ay E Hi and x + I1Y E Hi. We successively obtain
(A -11)Y E Hi, Y E Hi, X E Hi, which contradicts the choices of x and y. <)
6.5.2 Example
Consider the extension L = Q [J2, V3] of Q. It is an extension of degree
4 of Q. By §6.4.4, there are exactly four Q-homomorphisms from L to <C.
To construct them, we need to first construct the two Q-homomorphisms
71,72 : Q[J2] -+ C like in §6.3.1 2), associating to J2 one of its conjugates
J2 or -J2, which leads to 71 ( J2) = J2,72 (J2) = -J2. Each of these
two Q-homomorphisms extends in two different ways, still by §6.3.1 2), to a
Q-homomorphism O"i : Q [J2] [V3] -+ C, 1 $ i $ 4, by associating to V3
one of its conjugates ±V3. The primitive elements of L are the elements
whose images under these Q-homomorphisms are all distinct; J2 + V3,
J2 + V6, mJ2 + nV3 with m and n non-zero rational numbers, etc., are
examples of such elements (Table 6.2).
6.6 Linear Independence of K-Homomorphisms 101
J2 J3 J2+J3 J2+V6
0"1 J2 J3 J2+J3 J2+V6
0"2 J2 -J3 J2-J3 J2-V6
0"3 -J2 J3 -J2+J3 -J2- V6
0"4 -J2 -J3 -J2 - J3 -J2+V6
TABLE 6.2.
6.6 Linear Independence of K-Homomorphisms

The goal of this section is to state Dedekind's theorem (see §6.6.3). This
theorem is a direct consequence of a theorem on characters.
6.6.1 Characters
DEFINITION. - A character of a group G into an arbitrary field K is a group
homomorphism from G to the multiplicative group K*.
Let X : G ---> K* be a character. If e is the identity element of G, we have
x(e) = 1; if g is of order n in G, then X(g)n = X(gn) = 1 shows that X(g)
is an n-th root of unity in K.
The characters of G are elements of the set F of set maps from G to
K; we will say that characters are linearly independent if they are linearly
independent in the K-vector space F.
6.6.2 Emil Arlin's Theorem

Distinct characters of a group G into a field K are linearly independent.
PROOF. - Suppose there exist families of characters of G into K that
are not linearly independent. Among all such families, choose a family
Xl,· .. ,Xn such that n is minimal. Because n is minimal, there exist ele-
ments AI, ... , An of K such that L
AiXi = 0 and such that none of the
l:'Oi:Sn
Ai is zero. Of course, we have n > 1.
Because Xl i= X2, there exists x in G such that Xl(X) - X2(X) -10. The
linear combination L
AiXi is zero, so for all y of G, we have
l:Si:Sn
L AiXi(Y) 0, (6.1)
l:Si:Sn
L AiXi(XY) L AiXi(X)xi(Y) = o. (6.2)

l:Si:Sn l:Si:Sn
Multiplying (1) by Xl(X), we obtain
L AiXl(X)xi(Y) = o. (6.3)
l~i~n
Subtracting (2) from (3), we have
Because at least one of the coefficients is not zero, this equality contradicts
the minimality of n. <>
6.6.3 Corollary: Dedekind's Theorem

Let K be a field and L c C an extension of K. The K-homomorphisms
from L to C form a set of linearly independent vectors in the C-vector space
of linear maps from L to C.
PROOF. - The C-vector space structure of the set of linear maps from L
to C is induced by that of the C-vector space of setwise maps from L to
C. Every K-homomorphism a from L to C induces a character from the
multiplicative group L* to C, since a(xy) = a(x)a(y); a is a K-linear map
since a(ax + by) = aa(x) + ba(y) for all x, yin L and all a, bin K. Thus,
Theorem 6.2 applies. <>

Exercise 6.1. K-homomorphisms
1) Determine IQ-automorphisms of the rings 1Q[7 + 3i] and lQ[a] where a
is a root of X2 + 52.
2) Determine all the subfields of C that are Q-isomorphic to 1Q[V3],

1Q[05], to IQ[W]' to IQ[W].
3) a) Determine the IQ-automorphisms of the rings 1Q[v'2] , IQ[~] and
IQ[~].
b) Show that U 1Q[21/n] is an extension of IQ (of infinite degree).

l~n
Determine its IQ-automorphisms.
Exercise 6.2. Primitive elements

1) Let K c <C be a field, and let a and b be two elements of K. Show that
if s = Va +..jb (where Va and ..jb denote elements of <C with squares
a and b) is non-zero, then it is a primitive element of K [Va, ..jb] over
K.
2) Show that j-/5 is a primitive element of IQ [j, -/5] .

3) Show that «2+V2 is a primitive element of the extension IQ [«2, V2]
of IQ. Give other primitive elements for this extension.
4) Determine a primitive element of lQ[a, j], where a is a root of X 3 -

X+l.
5) Let K be a field, and let a and b be algebraic elements over K. Show

that there exists an integer r in Z such that a + rb is a primitive
element of L = K[a, b] over K. Generalize.
Exercise 6.3. K -endomorphisms of an algebraic extension

Let L be an algebraic extension of a field K of finite or infinite degree,
and let a : L --; L be a K-homomorphism. Let a be an element of L,
P its minimal polynomial over K, and A the set of roots of Pin L.
1) Show that a(A) = A.

2) Deduce that a is a K-isomorphism.

1) As 1Q[7 +3i] = lQ[i] , there are two IQ-automorphisms of 1Q[7 +3i], namely
the identity and the one induced by complex conjugation.
As lQ[a] = lQ[iV13], there are two IQ-automorphisms of lQ[a], defined
by a(a) = ±iV13; namely the identity and the one induced by complex
conjugation.
2) The only field that is IQ-isomorphic to 1Q[v'3] is itself, although v'3 has
two conjugates over IQ.
Over IQ, {'ll5 is a conjugate of itself, of j {'ll5, and of P {'ll5. Thus the
field 1Q[{'ll5] is IQ-isomorphic to 1Q[{'ll5]' lQ[j{'ll5]' and lQ[j2{'ll5]. These
three fields are distinct; the first one is contained in lR, unlike the other
two, and if lQ[j {'ll5] = lQ[j2 {'ll5], then the element j, which is the quotient
of the two generators and is quadratic over Q, would belong to a cubic

extension.
Because the polynomial X 5 + 3 is irreducible over Q, ~ has five conju-
gates over Q and Q[~] is Q-isomorphic to the five fields Q[~], Q[(~],
Q[(2~], Q[(3~], Q[(4~], where ( = e 2i7r / 5 . These five fields are all
distinct, since the equality of two of them would imply that one of the
(k, 1 ::; k ::; 4, which are of degree 4 over Q, would belong to it, even
though the extensions are of degree 5.
As -&'3 has six conjugates over Q : ± W, ±j W, ±j2 W, the field Q[ W]
is Q-isomorphic to one of the three fields Q[ W], Q[j~] or Q[P ~].
Let us show that these three fields are all distinct. The first one is con-
tained in IR, and the other two are not. The two last ones are of degree 6 over
Q. If they were equal, then j would belong to Q[jW], and consequently
Q[jW] = Q[j, W]. But the last field is of degree 12 over Q.
3) a) The only Q-automorphisms of Q[J2]' of Q[0], and of Q[~] are
defined by 0'(J2) = ±J2, 0'(0) = ±0, O'(~) = ±~.
b) Let x and y lie in U1<nQ[2 1/ n ]. There exist integers rand s such that
x E Q[21/r] and y E Q[21/S]; we see that x and y lie in the field Q[2 1 / rs ],
and we easily deduce that U1<nQ[2 1/ n ] is a field.
Let 0' be a Q-automorphis~ of U1<nQ[2 1/ n]. We have 0'(21/n) = ±21/n.
As 0'(21/n) = 0"[(21/2n)2] > 0 for eve~y integer n, we have 0'(21/n) = 21/n
and 0' =id. This result does not tell us anything about the extension, but
we are in the case of an extension from which the conjugates of 0, V'2,
etc., are absent.
1) This follows from fa = (s/2)+(a-b)/(2s) and Vb = (s/2)-(a-b)/(2s).
2) Set a = jv's. We have v's = a3 /5 and j = a4 /25. The result follows.
3) We know that [Q[.v2,J2] : Q] = 6. An element of Q[.v2,J2] is

primitive if its images under the six Q-homomorphisms from Q [.v2, J2] to
C are distinct. The image of .v2 under one of these six Q-homomorphisms
is .v2, j.v2, j 2 .v2 and the image of h is ±h. The element .v2 + h is
thus primitive, as are the elements of the form x.v2 + yh for x and y non-
zero rational numbers (two of the images are real and obviously distinct,
and the other four are non-real with distinct real or imaginary parts. To
visualize them, draw a regular hexagon, place each of the six numbers at a
vertex and draw a line directly between two vertices if we have shown that
they correspond to distinct numbers. To conclude, every pair of vertices
must be connected by a line.)
A simpler primitive element is 2 1/ 6 .
4) Set b = a + j. The condition (b - j)3 - b + j +1 = 0 shows that

Q[J] c Q[b], so Q[b] = Q[a,j].
5) Set n = [L : K]. There exist distinct n K -homomorphisms from L to
<C; we denote them by al, ... , an. If there did not exist r E Z such that
a+rb has distinct images under these n K-homomorphisms, then for every
r there would exist distinct integers i and j with 1 ::::: i, j ::::: n such that
ai(a + rb) = aj(a + rb). As r can take an infinity of values, and the pairs
(i, j) only have a finite number of values, there would exist distinct integers
rand s, and distinct integers i and j, such that ai(a+rb) = aj(a+rb) and
ai(a + sb) = aj(a + sb). Taking the difference and simplifying by r - s, we
obtain ai(b) = aj(b) then ai(a) = aj(a), which would imply that ai = aj,
a contradiction.
It follows that if L = K[ar, ... ,an], there exist integers rl, ... , rn such
that L riai is a primitive element of Lover K.
l:$i:$n
1) We know that a is injective. As a(A) c A and A is finite, we have

a(A) = A.
2) By 1), a lies in the image of a; consequently, a is surjective, which

gives the result.
7
Normal Extensions
7.1 Splitting Fields

7.1.1 Definition
Let K be a subfield of C, and let P be a polynomial of degree n in K[X].
Let Xl, ... ,Xn be the (not necessarily distinct) roots of P in C. The field
K[XI' ... , x n ], which by §4.6.2 is an extension of K of finite degree, is called
the splitting field of P over K.
Let P be a polynomiaL Note that the polynomial PI = P/gcd(P,P')
has the same roots as P, but all as simple roots. Thus, in characteristic 0,
the splitting field of P is the same as that of PI, since the splitting field
is obtained by adjoining the roots. Thus, we can restrict our attention to
polynomials with distinct roots.
REMARK - We saw in §4.7 how to construct an extension of a field K

containing a root of a polynomial P irreducible over K; we called such an
extension a rupture field for P.
EXAMPLE. - The splitting field of X 3 - 2 over Q is Q[j, -v2] since the roots
of X - 2 are -v2, j -Y2, and j2 -Y2. It is an extension of degree 6 of Q. The
3
rupture fields of X 3 - 2 over Q contained in Care Q[ -v2], Q[j -Y2], and
Q[j 2 -Y2].

108 7. Normal Extensions
7.1.2 Splitting Field of a Cubic Polynomial

PROPOSITION. - Let K be a field contained in C and P(X) = X 3 + pX + q
an irreducible cubic polynomial in K[X]. Let a, b, and c denote the roots
of P in C and d = (a - b)(b - c)(c - a).
The splitting field of P over K is K[a,d]. It is of degree 3 or 6 over K
according to whether the discriminant D(P) = _4p 3 - 27q2 of P is or is
not a square in K.
PROOF. - The splitting field of P over K is K[a, b, c], and the inclusion
K[a, d] c K[a, b, c] is clear.
Conversely, we know that d2 = D(P) i- 0 and a i- O. The equality
(a - b)(c - a) = a(b + c) - a 2 - bc = -2a 2 + ~

a
shows that b-c = dj ((a - b)(c - a)) lies in K[a,d]. As b+c = -a also lies
in K[a, d], b, c both lie in K[a, d] and we have K[a, d] = K[a, b, c].
Now, K[a] and K[d] are intermediate extensions between K[a, d] and K,
but [K[a] : K] = 3 and [K[d] : K] is equal to either 1 or 2 according to
whether d E K or d 'f. K; this suffices for the proof. <>
REMARK. - If a cubic polynomial is not irreducible over K, its splitting
field can be of degree 1 or 2 over K.
7.2 Normal Extensions

DEFINITION. - A normal extension of a field K (either contained in C or
not) is an algebraic extension N of K (of finite or infinite degree) such that
every irreducible polynomial in K[X] having a root in N has all its roots
in N. In other words, all conjugates of elements of N must lie in N.
We will try as much as possible to reserve the letter K for the base field,
the letters L, L', ... fo~ arbitrary extensions of K, and N, N', ... for normal
extensions K.
EXAMPLES. -
a) By Proposition 7.4.1 below, a splitting field is a normal extension of
finite degree.
b) The intersection of normal extensions is normal.
c) A normal extension N of a field K is also a normal extension of

every intermediate extension L between K and N (see Proposition
7.5 below).
COMMENTARY. - In the case of infinite fields of non-zero characteristic, it is

necessary to distinguish between "normal extensions" and Galois extensions
7.3 Normal Extensions and K-Homomorphisms 109
(see Chapter 15). In characteristic 0, the two notions are equivalent; we use
the the expression "normal extension" .
7.3 Normal Extensions and K - Homomorphisms

PROPOSITION. - An algebraic extension Nee of finite degree of a field K
is normal if and only if the image of every K -homomorphism u : N -> C
is contained in N.
PROOF. - If N is a normal extension of K, then for every x in Nand
every K-homomorphism u : N -> C, x is algebraic over K, and by §4.3.1,
u(x) is a conjugate of x. Consequently, u(x) lies in N, so that u(N) C N.
Moreover, u is a K-automorphism of N by §6.2.2.
Conversely (Figure 7.1), if an element x of N has minimal polynomial
P over K, and if y is a root of P in C, then by §6.3.1, there exists a
K-homomorphism u : K[x] -> C such that u(x) = y. By §6.4.3, this homo-
morphism extends to a K-homomorphism u' : N -> C. But as u'(N) eN
and u'(x) = y, y is an element of N.
I~ <r
K[x] (J ~
1~
FIGURE 7.1.
7.4 Splitting Fields and Normal Extensions

In this section, we give the precise relation between a splitting field and a
normal extension of finite degree.
7.4.1 Proposition
Let K C C be a field. Let P be a polynomial in K[X] of degree n, and let
N be the splitting field of P over K. Then
1) N is a normal extension of finite degree of K.
2) [N: K] divides nL
PROOF. -
1) Let Xl, ... ,Xn be the (not necessarily distinct) roots of P in C. We

know that N = K[XI, ... ,xn ] is an algebraic extension of K.
Let a- : N -+ C be a K-homomorphism. For i = 1, ... , n, we
have 0 = a-(P(Xi)) = P(a-(Xi)), so a-(Xi) E {Xl, ... ,Xn} C N. As
{Xl, ... , Xn} generates N, we have a-(N) C N and Proposition 7.3
concludes the proof.
2) We use induction on n. For n = I, the result is obvious. Suppose now

that n > 1. If P is irreducible and X is a root of P in N, we have
[K[x] : K] = n, and by the induction hypothesis, [N : K[x]] divides
(n - I)!, where N is the splitting field of PI(X - x) over K[x]. This
gives the result. If P has an irreducible factor S of degree k, then
n > k ~ I, and if N' is the splitting field of S over K, then [N' : K]
divides k! and [N : N'] divides (n - k)!, where N is the splitting field
of PIS over N. Consequently, [N: K] divides k!(n-k)!, so it divides
n!. <)
7.4.2 Converse
Let N C C be a normal extension of finite degree of a field K. Then N is
the splitting field over K of an irreducible polynomial P in K[X] of degree
[N:K].
PROOF. - Because N is an extension of finite degree of K, the Primitive
Element Theorem (§6.5) shows that there exists an element X in N such
that N = K[x]. The minimal polynomial P of X over K is of degree [N : K].
It is irreducible, so all of its roots are in N since N is a normal extension
of K. The splitting field of P over K is thus N. <)
7.5 Normal Extensions and Intermediate

Extensions
PROPOSITION. - Let N be a normal extension of finite degree of a field K.
Then N is a normal extension of every intermediate extension L between
K and N.
PROOF. - As [N : K] is finite, [N : L] is finite. By §7.4.2, N is the splitting
field over K of a polynomial P in K[X]. If Xl, ... , Xn are the roots of Pin
C, we have N = K[XI"" ,Xn]; as K C LeN, we have N = L[XI".' ,xn],
which shows that N is the splitting field of P over L, where P is considered
as a polynomial in L[X]. By §7.4.1, this gives the result. <)
7.6 Normal Closure III
REMARK. - The above property can be recalled by the diagram in Figure

7.2. The example of Figure 7.3 shows that conversely, L mayor may not
be a normal extension of K.
N
\nonnal
nonna! L
/
K
FIGURE 7.2. FIGURE 7.3.
7.6 Normal Closure

7.6.1 Definition
Let K be a field, and let Lee be an algebraic extension of K. The normal
closure of Lover K in C is the smallest normal extension N of K containing
L. This extension always exists; it is the intersection of the set of normal
extensions of K containing L, a set which is non-empty since C is a normal
extension of all its subfields.
7. 6.2 Proposition
Let K c <C be a field. The normal closure in <C of a finite degree extension
L C <C of K is a finite degree extension of K. It is the splitting field of the
minimal polynomial of a primitive element of Lover K.
PROOF. - By the Primitive Element Theorem, there exists an element a
of L such that L = K[aJ. Let P be the minimal polynomial of a over K
and let N be the splitting field of P over K. N is a normal extension of
finite degree of K, which contains L; moreover, every normal extension of
K containing L contains all the roots P, so it contains N. The normal
closure of K in <C is thus N. 0
7.6.3 Proposition
Let K C C be a field, and let L = K[al, ... , an] C C be a finite degree
extension of K. The normal closure N of Lover K inside <C is the extension
of K by the set of conjugates over K of the elements all"" an.
PROOF. - Let A be the set of conjugates over K of the elements aI, ... ,an'
Because N is a normal extension of K, we have N J A, so N J K[A].
As K[A] is the splitting field of the product of the minimal polynomials of

al, .. " an over K, it is a normal extension of K. Thus N = K[A]. <>
7.7 Splitting Fields: General Case

Let us now consider arbitrary fields instead of only sub fields of C. Let K
be an arbitrary field, and P an irreducible polynomial of degree n in K[X].
We know that P has no roots in K. In §4.7, we saw how to construct an
extension of K containing a root of P. Let us now indicate how to construct
an extension N of K which is a splitting field of P over K, i.e. a field which
contains all the roots of P and is generated by them.
PROPOSITION. (Kronecker) - Let K be an arbitrary field, and let P be a

polynomial of degree n in K[X].
1) There exists an extension N of K which is a splitting field for P over
K.
2) [N: K] divides n!.
3) Two splitting fields Nand N' of P over K are K-isomorphic. If the

roots of P are simple in Nand N' (we say that P is separable), then
the number of K-isomorphisms from N to N' is equal to [N : K].
PROOF. -
1) We use induction on n. If n :::; 1, then N = K works.

If n > 1, factor P as a product of irreducible factors over K. If all
these factors are of degree 1, then again N = K works. Otherwise,
let S be an irreducible factor of degree> 1 of P. Set L = K[Xl/(S)
and let x denote the class of X in L. L is a field and in L[X], we
can factor P in the form (X - x)T with T E L[X]. By the induction
hypothesis, there exists a field N which is a splitting field of T over
L. This field is a splitting field of P over K.
2) See the proof of §7.4.1 2).
3) LEMMA. Let (j : K ---> K' be a field isomorphism and N' a splitting

field of (j(P) over K'. We can extend (j to a homomorphism T : N --->
N'. If the roots of P are simple, the number of extensions is equal to
[N:K].
To prove the lemma, we use induction on [N: K]. If [N : K] = 1,
then P factors as a product of linear factors in K[X] and everything
is clear. If [N : K] > 1, let S be an irreducible factor of degree> 1
of P. Let x be a root of S in N, and set L = K[x]. Let x' be a
Toward Chapter 8 113
root of a(S) in N', and set L' = K'[x']. As in §6.4.2, we construct

the isomorphisms L ~ K[X]j(S) ~ K'[X]j(a(S)) ~ L' which define
the isomorphism al : L ~ L' with al(x) = x'. By the induction
hypothesis, this al extends to a homomorphism T : N ~ N'. If the
roots of P are simple, the number of such extensions al is equal to
the number of roots of a(S) in N', i.e. to deg(a(S)) = [L : K]. The
induction hypothesis then gives the desired number.
To finish the proof, we use the lemma to extend id(K), obtaining
a (necessarily injective) homomorphism T : N ~ N', proving that
[N : K] ~ [N' : K]. Inverting the roles of Nand N', we have [N' :
K] ~ [N: KJ, so T is an isomorphism. <>
Toward Chapter 8
Chapters 4, 6, and 7 suffice to open the doors of the "paradise" created
(discovered?) by Galois; Chapter 8 will describe the heart of his theory, the
Galois correspondence.

Exercise 7.1. Splitting fields
1) Determine the splitting field ofaX 2 + bX + c with a i=- 0 over K =
Q(a, b, c).
2) Determine the splitting fields of the following polynomials by deter-

mining the degrees of the corresponding extensions of Q:
a) X 3 -1;
b) X 6 -1;
c) X4-7;
d) X 3 - X2 - X - 2;
e) X 6 -10X 4 + 31X 2 - 30;
f) X 5 + X4 + X 3 + X2 + X + 1;
g) X4 + 1.
Exercise 7.2. Normal extensions

1) Show that a quadratic extension of a field K is necessarily normal.
2) Does the fact that ij2 is an element of Q[ ij2] whereas j ij2 and j2 ij2
are not imply that Q[ ij2] is not a normal extension of itself?
3) Let N be a normal extension of a field L. Is it a normal extension

of every sub field K of L? Consider in particular the example Q C
Q[vi2] C Q[ 0]; give the normal closure of the extension Q[ 0] of Q.
4) Determine the normal extensions of Q generated by V3 + vi2, y3 + i,

ij2 + y'2, sin 2;, ( = e 2i7r / n for integers n.
5) Let N C C be a normal extension of finite degree of a field K, and

let L be an extension of K and N' C C the extension of K generated
by L U N. Show that N' is a normal extension of L.
Exercise 7.3. Cubic normal extensions

1) Let K be a field contained in C, and let P be an irreducible cubic
polynomial in K[X] of discriminant D = d2 , dEC Let a be a root
of Pin C Show that K[a] is a normal extension of K if and only if
d lies in K.
2) Let a, b, c be the roots of X 3 - 3X + 1 in C
a) Show that Q[a] is a normal extension of Q.
b) Express band c in the basis {I, a, a 2 } .
Exercise 7.4. Proof of d' Alembert's theorem (fundamental theo-

rem of algebra)
In this problem, we will show that every non-constant polynomial P
in qX] has at least one root in C, following an idea due to Pierre
Samuel (see the bibliography).
1) Show that it is enough to consider polynomials with real coefficients

(consider PP).
Now assume that P has real coefficients; we will proceed by induc-
tion on r = V2 (deg( P)), where V2 (n) is the largest power of 2 dividing
n.
2) Show the result in the case r = 0, then in the case deg(P) = 2.

3) Now suppose r ~ 1. Take a splitting field of P. Note that we can-
not assume that this splitting field is contained in C (since that is
a consequence of the theorem we are now proving), so we need to

construct the splitting field by successive quotients as in §7.7. For
1 ::; i ::; n = deg(P), let ai denote the roots of P in this splitting
field.
For every real number c, set
Qc(X) = II (X - ai - aj - Caiaj)'
l:'Si<j::on
a) Determine v2(deg(Qc».
b) Show that the coefficients of Q c are real.
c) Finish the proof.

1) According to whether the discriminant b2 - 4ac is or is not a square in

K, the splitting field ofaX 2 + bX + c is either K or the quadratic extension
of K generated by a root of b2 - 4ac.
2 ) a), b), d), and f). Q[j] is the splitting field in these four cases, since we
have
X 3 -1 = (X - 1)(X2 + X + 1),
X 6 - 1 = (X _1)(X2 + X + 1)(X + 1)(X2 - X + 1),
X 3 - X2 - X - 2 = (X - 2)(X2 + X + 1),
X 6 -1
X5+X4+X3+X2+X +1 = X = (X2+X +1)(X +1)(X2_X +1).
-1
c) The polynomial X4 -7 is irreducible over Q by Eisenstein's criterion,
and its splitting field is Q[±ifl, ±iifl] = Q[i, ifl]. This is an extension of
degree 8 over Q.
e) The polynomial factors as (X2 - 2) (X2 - 3) (X2 - 5); its splitting field
is Q[h, V3, V5], which is of degree 8 over Q by Exercise 4.56).
g) Set ( = eirr / 4 = (1 + i)/h). The splitting field N of X4 + 1 over
Q is Q[(,(3,(5,(7] = Q[(]; it contains i = (2, so it contains h, so that
N = Q[i,h].
1) Let L be a quadratic extension of K. Then we must have L = K[x]

for every x in L which is not in K, since they are all of degree 2 over K.
Let y be the second root of the minimal polynomial of x over K. Because

x + y E K, we have L = K[x, y], so L is a splitting field.
2) No: .v2 is a root of the polynomial X - .v2 of Q[.v2][X].
3) This result is false in general, since a field is always a normal extension

of itself, but is not always a normal extension of its subfields.
A false argument: let P be an irreducible polynomial of K[X] having a
root in N. As a polynomial in L[X], it has a root in N, so all its roots must
lie in N. Thus N is a normal extension of K.
The error in this reasoning is that in fact, P may no longer be irreducible
over L. For example, take K = Q c Q[ h] = L c Q[~] = Nand
P(X) = X4 - 2. Then P is irreducible over Q, but not over Q[h] since
X4 - 2 = (X2 - h)(X2 + h) and the second factor has no roots in N.
The normal closure of the extension Q[~] of Q is Q[i, ~].
4) We already saw that Q[ v'3 + h] = Q[ v'3, h], Q[ v'3 + i] = Q[ v'3, i] and

Q[.v2 + h] = Q[.v2, h] (Exercise 4.2). Thus Q[v'3 + h] and Q[v'3 + i]
are normal extensions, since they are the splitting fields of (X2 - 2) (X2 - 3)
and (X2 + 1)(X2 - 3), and .v2 + h generates the normal extension N =
Q[.v2,j, h], which is the splitting field of (X2 - 2)(X 3 - 2), of degree 12
over Q.
We saw in Exercise 4.5 that a = 4sin 2; = VlO + 2J5 and that the
minimal polynomial of a over Q is X4 - 20X 2+ 80. The conjugates of a over
Q are thus ±VlO ± 2J5. Now, VlO - 2J5 = 4J5/VlO + 2J5 belongs to
Q[sin(27r/5}] since.J5 = (a 2 -10}/2 belongs to it. Thus, Q[sin(27r/5)] is a
normal extension of Q.
Finally, in the case ( = e2i1r / n for an integer n, Q[(] is the splitting field
of Xn - 1 over Q, so it is a normal extension of Q.
5) By the Primitive Element Theorem, we know that N is the splitting field

of a polynomial P in K[X]; then Nt is the splitting field of the polynomial
P over L.
1) K[a] is a normal extension of K if and only if K[a] = K[a, dj, since

K[a, d] is the splitting field of P over K (§7.1.2); this condition is equivalent
to d E K[a], so dE K, since d is of degree 1 or 2 over K.
2) a) As X 3 - 3X + 1 has no rational root, it is an irreducible polynomial;

its discriminant is 81, which is a square in Q. Thus [Q[a] : Q] = 3 and Q[a]
is a normal extension of K.
b) We use the computations of §7.1.2. On the one hand, we have
b+c = -a.
On the other, we have
d d
b- c = ..,.--..,.-,-----:- 1.
(a - b)(c - a)
-2a 2 +-a
Take d = 9 (the other possible choice, -9, simply ends up exchanging
the roles of b and c). Using the techniques of Chapter 4 (the method of
indeterminate coefficients or Bezout's identity), we find b - c = 4 - a - 2a 2 ,
which gives b = 2 - a - a 2 and c = a 2 - 2.
8
Galois Groups
In this chapter, we reach the very heart of Galois theory. To every poly-
nomial with coefficients in a field K, with splitting field N over K, we
associate a group G called its Galois group. We show that the subgroups
of G are in bijective correspondence with the intermediate extensions be-
tween Nand K. This correspondence makes it possible to solve problems
about polynomials and their splitting fields algebraically, by computing
groups. Over the following chapters, we sketch out this dictionary between
the properties of an equation and the algebraic properties of its associated
group.
8.1 Galois Groups

8.1.1 The Galois Group of an Extension
DEFINITION. - Let K be a field, and let L be an extension of K. The set
of K-automorphisms of L is equipped with a group structure whose group
law is the composition of K -automorphisms. We will denote this group by
Gal(LIK), and call it the Galois group of the extension Lover K.
Throughout most of this text, we will consider Galois groups of normal

extensions of finite degree; non-normal extensions L of K do not possess
enough K -automorphisms to make the fundamental theorem of the Galois
correspondence (see §8.5 below) work.

120 8. Galois Groups
8.1.2 The Order of the Galois Group of a Normal Extension

of Finite Degree
PROPOSITION. -Let N be a normal extension of finite degree of a field K

contained in C. Then the order of the group Gal(NIK) is equal to [N: K].
PROOF. - By §6.4.4, there exist [N : K] K-homomorphisms from N to
C. By §7.3, their image lies in N; thus these homomorphisms are actually
K-automorphisms (§6.2.2). ~
8.1.3 The Galois Group of a Polynomial

Let K be a field, and let P be a polynomial of degree n in K[X] such
that P has distinct roots. If K c <C, recall that we can always replace P
by a polynomial having the same roots as P, but as simple roots. More
generally, if K is of characteristic 0, we can replace P by the greatest
common divisor of P and its derivative P'; this method was first suggested
by Johann Hudde in 1657 (see the book by J.-P. Tignol). Let E be the
set of roots of P in <C, and let N = K[E] be the splitting field of P over
K. We know that N is a finite degree extension of K, which is normal
by Proposition 7.4.1. The Galois group G = Gal(NIK) is also called the
Galois group of the polynomial P over K.
8.1.4 The Galois Group as a Subgroup of a Permutation

Group
In Exercise 8.1, we recall the definition and some properties of group actions
on sets.
PROPOSITION. - 1) With the notation of§8.1.3, G acts on E. This action

makes it possible to identify G with a subgroup of the group SE of per-
mutations of E, and with mutually conjugate subgroups of the group Sn of
permutations of {I, ... , n}.
2) The order of G divides nL
3) If P is irreducible, the action of G on E is transitive.
PROOF. - 1) Let a E G. For every x E E, a(x) lies in E, since it is a
conjugate of x (§6.3.1). As a is injective, a induces an injection of E into
E, so a bijection of E; the map A : G ...... SE defined by A(a) = alE is
an injective homomorphism of groups since E generates N. This gives the
identification of G with a subgroup A(G) of SE.
Let Xl, ... , xn be the roots of P; then we can define a bijection 'P :
{I, ... ,n} ...... E with Xi = 'P( i). Such a bijection makes it possible to define
an injective group homomorphism ~ : G ...... Sn such that s = ~(a) is the
permutation given by a(xi) = Xs(i)' This construction makes it possible to
identify G with a subgroup ~(G) of Sn.
8.1 Galois Groups 121
'I' (cr)

{ I, .... ,nlE E '" {I, .... ,n}
'1'-1 -1
 E be two bijections. They define injective homo-
morphisms <P, \lI : G -> Sn by <p(a) = 'P- 1A(a)'P and \lI(a) = 'IjJ-l A(a)'IjJ
(Figure 8.1). Thus, we associate to G the groups <p(G) and \lI(G); these
groups are mutually conjugate in Sn since 'IjJ-1'P<P(G)'P-1'IjJ = \lI(G).
2) This is an immediate consequence of 1) and Lagrange's theorem.
3) For every i and every j E {I, ... , n}, there exists a K - homomorphism
from K[XiJ to K[xjJ by §6.3.1. This extends to an embedding of N in C
which, as N is normal, induces an element a of G such that a(xi) = Xj' <>
8.1.5 A Short History of Groups

The first proof which actually uses the idea of a group is due to Lagrange
(1770).
In attempting to determine the degree of the equation (whose coefficients
were rational functions in the elementary symmetric polynomials) satisfied
by a rational function f = f(X 1 , •.. , Xn), he formed the product
B(t) = II (t-f(Xa(l),Oo"Xa(n))
aESn
and tried to write it in the form Bl (t) k. For every a in Sn, set a . f
f(Xa(l), ... , Xa(n))' and take the natural action of the group Sn on E =
{a· f,a E Sn}; then the product of the cardinal of the orbit O(J) by the
cardinal k of the stabilizer S(J) is n!, and we have B(t) = Bl (t)k.
Obviously, this was not the language used by Lagrange; he reasoned on
. I case. If f( x I ,X" ,XIII , x IV , . .. ) = f("
a speCla X, x III ,X,I x IV , . .. ),t h en every
other value of f is also taken twice, as he showed by the example
x". . . ) = f( x"
f(X IV , XIII , x' " )
III x' xIV , x" , ....
Then, he asserted, the same holds in the other cases. Who would dare
to write this way nowadays? And yet, his argument is the basis of the
argument used to prove that every equivalence class of elements in a group
modulo a subgroup contains the same number of elements.
Galois was the first to actually use the word "group", but he meant it as
a subgroup of the permutation group of the set of roots of a polynomial,
as in Proposition 8.1.4. Galois did not use the notation for permutations
that we use now; he wrote successive arrangements of the roots which he

called permutations:
abcde
bcdea
cdeab
emphasizing the fundamental notion of what he calls substitution, which

makes it possible to pass from one permutation to another. Galois did not
use associativity, identity elements, or inverses, only the internal group law,
writing: "Thus, if in such a group we have the substitutions Sand T, we
are sure to have the substitution ST." Indeed, for a subset of Sn to be a
subgroup, it is necessary and sufficient for it to be stable under composition.
The abstract form of the definition of a group, which we use today,
was built up slowly over the course of the 19th century, with suggested
definitions by Cayley (1854), Kronecker (1870), Weber (1882), Burnside
(1897), and Pierpont (1900). The axioms of associativity, identity element
and inverses were first stated in their present form by Pierpont.
8.2 Fields of Invariants

8.2.1 Definition and Proposition
Let K be a field, and let L be an extension of K and H a subgroup of
Gal(LIK). The set I(H) of elements x E L invariant under H, i.e. such
that for every a E H we have a(x) = x, is a subfield of L called the field
of invariants of H (Figure 8.2).
L cr ,. L
\ /
I(H)
I
<Q
FIGURE 8.2.
8.2.2 Emil Arlin's Theorem

Let K be a field, and let L e e be an extension of K and H a finite
subgroup of orderr ofGal(LIK). Then L is a normal extension of I(H) of
degree r, and Gal(LII(H)) = H.
8.2 Fields of Invariants 123
PROOF. - Let x E L, and let us show that x is of degree less than or equal
to rover I(H), and that its minimal polynomial P over I(H) has all its
roots in L.
The set E = {h( x), h E H} is finite, say of order t with (t :S r). For every
i, 1 :S i :S t, there exists hi in H such that {h i (x),l :S i :S t} = E. We set
SeX) = II (X - hi(x)); it is a polynomial in L[X]. Let us show that the
l::;i::;t
coefficients of S lie in I (H) .
Let h E H. For every i with 1 :S i :S t, we have h(hi(x)) E E; as h is
injective and E is finite, we have h(E) = E.
Let h' : L[X] ...... L[X] be the algebra endomorphism defined by h'lL = h
and h'(X) = X, i.e. by h' (L:akX k ) = L:h(ak)X k. By the above, we
have
h'(S)(X) II h'(X - hi (x))

l::;i::;t
II (X - h(hi(x)))
l::;i::;t
SeX).
The coefficients of S are thus invariant under the elements of H and belong
to I(H).
Furthermore, E contains x since the identity map of L belongs to H,
so S(x) = 0; thus x is algebraic over I(H) and its minimal polynomial P
divides S. It follows that x is of degree less than or equal to t = deg(S), so
less than or equal to rover I(H). As S factors into linear factors in L, the
same holds for P: all the roots of P lie in L. Thus L is a normal extension
of I(H).
Let us now show that L is an extension of finite degree r of I(H). As every
element x E L is of degree :S rover I(H), we can set s to be the maximum
of the degrees of the elements of L over I (H). Let y be an element of degree
s. If I(H)[y] is strictly contained in L, then there exists z in L - I(H)[y];
z is algebraic over I(H), so it is algebraic over I(H)[y] and [I(H)[y][z] :
I(H)] > [I(H)[y] : I(H)] = s. By the Primitive Element Theorem (§6.5),
there exists an element w of L such that I(H)[y][z] = I(H)[w] (Figure 8.3).
L
I
I(H)[y][z] =I(H)[ w]
1
I(H)[y]
I(H)
I
FIGURE 8.3.
We thus have [I(H)[w] : I(H)] > s, which contradicts the maximality

of s. Hence L = I(H)[y]' so L is an extension of finite degree of I(H)
and [L : I(H)] = s :::; r. Finally, IGal(LII(H)1 = [L : I(H)], by §8.1.2;
as H c Gal(LII(H), we have r :::; [L : I(H)] so that r = [L : I(H)] and
H = Gal(LII(H). 0
8.3 The Example of Q [ij2, j]: First Part

We know that the splitting field of x 3 - 2 over Q is the normal extension
N = Q [~,j]. As [N : Q] = 6, the Galois group G = Gal(NIQ) has six
elements; let us call them 0"1, ••• ,0"6. For 1:::; i :::; 6, O"i is determined by the
images of j and ~, which are conjugates of these elements over Q. The
image of j is j or j2; the image of ~ is ~,j~, or i~. This gives the
six possibilities listed in Table 8.1.
0"1 0"2 0"3 0"4 0"5 0"6
~ .y2" .y2" j.y2" j.y2" P.y2" j2.y2"
j j j2 j j2 j P
TABLE 8.1.
Note that 0"1 is the identity on N and the identity element of G, and that
0"2 is induced by complex conjugation.
To determine the structure of G, we note that it has three elements
of order two: 0"2,0"4,0"6, and two elements of order 3: 0"3,0"5. Thus it is
isomorphic to 8 3 (the only other group with six elements is the cyclic
group (Zj6Z, +), which contains only one element of order two). We can
also note that G is a group with six elements, and since it can be identified
with a subgroup of 8 3 , by §8.1.4, it must be isomorphic to 8 3 .
8.3 The Example of Q [ ~,j]: First Part 125
If we number the three roots of X 3 - 2, for example if we number ~

as 1, j ~ as 2 and i ~ as 3, we construct an isomorphism from G to 8 3
(Table 8.2).
0"1 0"2 0"3 0"4 0"5 0"6
id (23) (123) (12) (132) (13)

TABLE 8.2.
Let us determine the field of invariants of a subgroup H of G.

If H = {id} , it is clear that J(H) = N.
Recall that every element x of N can be written uniquely in the form
If H = (0"3), then x belongs to J(H) if x = 0"3(X); i.e.
x = a + bj + cj ~ + d( -1 - j) ~ + e( -1 - j) W + 1 W,
which implies that c = -d, d = c- d, e = -e + I, 1 = -e, i.e. c = d = e =
1=0.
Thus J(H) = Q[j]. As shown by Artin's theorem, we indeed have
[N : Q[j]] = 3 = IHI and Gal(NIQ[j]) = H.
Similarly, we find
If H = G, all of the preceding conditions must be satisfied, so we find that

J(H) = Q.
The correspondence between the set of subgroups of 8 3 (or of G), ordered
by inclusion, and the set of the intermediate extensions between Q and
N, also ordered by inclusion, is just one example of the central tenet of
Galois theory, the Galois correspondence, which we will study in §8.5. It is
summarized in Figure 8.4.
FIGURE 8.4.
8.4 Galois Groups and Intermediate Extensions

PROPOSITION. - Let K be a field. Let N be a normal extension of finite
degree of K and L an intermediate extension between K and N. We know
(see §7.5) that N is a normal extension of L. We also assume that L is a
normal extension of K (Figure 8.5).
N •N
~mru I
L ------- L
!7
normru
FIGURE 8.5.
1) The map Gal(LIK) obtained by taking the restriction
to L of a K -automorphism of N is a surjective group homomorphism with
kernel equal to Gal(NIL).
2) The group Gal(NIL) is a normal subgroup of Gal(NIK), and
Gal(NIK)jGal(NIL) ~ Gal(LIK).
COMMENTARY. - We thus have the following exact sequence of (not nec-

essarily abelian) groups:
1 --> Gal(NIL) --> Gal(NIK) ~ Gal(LIK) --> 1.
In other words, Gal(NIK) is a (group) extension of Gal(NIL) by Gal(LIK).

PROOF. - 1) <p is a group homomorphism. By §6.4.3, it is surjective. More-
over, if a E ker(<p), then aiL = id(L), so a E Gal(NIL) and ker(<p) c
Gal(NIL). The converse is obvious.
2) We know that the kernel of a group homomorphism is a normal subgroup;

the isomorphism is a consequence of a classical theorem on quotients. <)
8.5 The Galois Correspondence

As promised at the beginning of this chapter, we now come to the very
heart of Galois theory.
FUNDAMENTAL THEOREM OF GALOIS THEORY. - Let K be a field, and

let N be a normal extension of finite degree of K. Let £ be the set of
8.5 The Galois Correspondence 127
intermediate extensions between K and N, and let 9 be the set of subgroups

of Gal(NIK).
Let I : 9 ---+ £ denote the map which associates to a subgroup H of
Gal(NIK) the field of invariants I(H), and let G : £ ---+ 9 be the map
which associates to an extension L of £ the group Gal(NIL).
1) I and G define inverse bijections which are decreasing for the inclusion
relation.
2) By restriction, I and G define inverse bijections of the set £' of nor-

mal extensions of K contained in N to the set 9' of normal subgroups of
Gal(NIK).
3) If Land L' are intermediate extensions between K and N, then L' is
a normal extension of L if and only if Gal(NIL') is a normal subgroup of
Gal(NIL), in which case Gal(L'IL) = Gal(NIL)jGal(NIL').
4) If Land L' are intermediate extensions between K and N with L' :) L,

then
[L' : L] = IGal(NIL)1 j IGal(NIL')I.
COMMENTS. -
1) I and G define a trellis isomorphism. Let us explain this.
Recall that the expression z = sup (x, y) in an ordered set means that z
is the smallest element satisfying z ::::: x and z ::::: y, i.e. for every t such that
t ::::: x, t ::::: y, we have t ::::: z. The definition of z = inf(x, y) is analogous. A
trellis is an ordered set in which the sup and the inf of any two elements
exists.
£ is a trellis for inclusion, since inf(L, L') = L n L' and sup(L, L') =
K(L U L').
9 is a trellis for inclusion: indeed, if Hand H' are subgroups of Gal(NIK),
then inf(H, H') = H n H' and sup(H, H') is the subgroup generated by H
and H'.
2) This theorem asserts in particular that an element of N lies in K if it
is invariant under all of the elements of Gal(NIK), a result that is often
useful.
PROOF. - 1) The fact that I and G are decreasing is clear.

Let us show that 10 G = idee). Let L E £ and L' = I(G(L)). N is a
normal extension of L and L'. We have L' :) L since every a E G(L) =
Gal(NIL) satisfies aiL = id(L). As G is decreasing, we have G(L') C G(L);
moreover, if a E G(L), we have aiL' = id(L') by definition of L'. Thus, a E
G(L'), so that G(L) = G(L'). As IG(L)I = [N : L] and IG(L')I = [N : L'],
we have [L' : L] = 1, so L' = L.
The equality Go J = id(Q) was the object of Artin's theorem (Theorem

8.2.2).
2) Proposition 8.4 shows that if L lies in £', then G(L) is a normal
subgroup of Gal(NIK). Conversely, if H lies in 9', let us show that J(H) lies
in £' by using §7.3. Let (T : J(H) ~ LeN be a K-homomorphism. As N is
a normal extension of finite degree of J(H), there exists a K-automorphism
T of N such that TIJ(H) = (T. For every p of Gal(NIJ(H» = G(I(H» = H,
we have T-Ipr E H since H lies in 9'. Consequently, for every x in J(H),
we have T-Ipr(X) = x, so that T(X) = p(T(X», which shows that (T(x) =
T(X) E J(H) by 1); J(H) is indeed a normal extension of K.
3) It suffices to apply 2), replacing K By L, and Proposition 8.4.
4) Indeed, we have
IGal(NIL)1 / IGal(NIL')1 = [NIL]j[NIL'] = [L' : L]. 0
8.6 The Example of Q [?'2, j]: Second Part

In this example, A3 = ((123)} is the only non-trivial normal subgroup of
8 3 , and the normal extension Q[j] of Q corresponds to it.
Let us give the computation of Proposition 8.4 in detail. Gal(NIQ) has
six elements which we denoted by (TI, .•• ,(T6. Gal(NIQ[j]) is the subgroup
consisting of the elements (TI, (T3, (T5' We have
Gal(Q[j]IQ) ~ Gal(NIQ)/Gal(NIQ[j]) ~ 7/.,/27/.,;
the elements of Gal(Q[j]IQ) are the restrictions of the elements of Gal(NIQ)

to Q[j]. Restricted to Q[j], (TI, (T3, (T5 act like the identity and (T2, (T4, (T6 act
like complex conjugation.
8.7 The Example X4 +2

In the following chapters, we will see classes of polynomials for which one
can actually compute the Galois group. We refer to Chapter 16 for a sketch
of algorithmic methods used for polynomials of small degree; in this section,
we give one rather long, complete example. We need to begin by recalling
dihedral groups.
8.7.1 Dihedral Groups

The group dihedral Dn is the group of isometries of a regular polygon with
n ~ 3 sides. As an isometry preserves barycenters, it preserves the center
o of the polygon, the isobarycenter of the set of vertices, and induces a
permutation of the vertices of the polygon.
Let A o, AI, ... , A n - l be the n vertices of the polygon. An isometry J is
determined by the images of Ao and AI' If J(Ao) = A k , J(A I ) is A k - l or
8.7 The Example X4 +2 129
Ak+l (we consider the indices modulo n) since f conserve the distances.
If f(Ad = Ak+1, then f is the rotation of center 0 and angle 2k1r/n. If
f(Al) = A k - 1 , f is the symmetry with respect to line D passing through
o such that (OA o, D) = k7r /n.
Let r denote the rotation of angle 27r /n and s the symmetry with respect
to ~Ao. Thus, the dihedral group Dn has 2n elements:
1) the rotations of angles 2k7r/n, equal to rk for 0 ::; k ::; n - 1; and
2) the symmetries with respect to the lines D passing through 0 such

that (OAo,D) = k7r/n, with 0 ::; k ::; n -1; they are equal to rks
since rks(Ao) = Ak and rks(Al) = A k - 1.
The group Dn is thus generated by rand s, and in order to perform

all computations in the group D n , it suffices to use the three relations:
rn = id, s2 = id, sr = r- 1 s. The two first relations are obvious, and the
third is a consequence of the fact that sr(Ao) = A n - 1 = r-1s(Ao) and
sr(Al) = An-2 = r-1s(A 1). It follows that sr k = r-ks. We prove that if a
group is generated by two elements satisfying these three properties, it is
a quotient of D n , and these relations characterize Dn among all groups of
order 2n.
8.7.2 The Special Case of D4

We will see in §8.7.3 below that the structure of the Galois group of the
polynomial X4 + 2 over IQ is that of the group D 4 . Let us recall some
properties of this group, its subgroups, and its embeddings in 8 4 , so as to
highlight the Galois correspondence.
The group D4 is the group of isometries of a square ABCD. Let .6, and
.6,' denote the mediatrices of (AB) and (BC) (Figure 8.6).
c A
FIGURE 8.6.
The group D4 contains eight elements: the identity, two elements of order
4, r = rot(O, 'IT /2) and r3 = rot(O, -'IT /2) = r-l, and five elements of order
2, r2 = rot(O,'IT) = -id = sym/O, s = sym/(CA), rs = sym/!::.. = sr 3,
r 2s = sym/(BD) = sr2, and r3 s = sym/!::..' = sr.
The subgroups of D4 have 1, 2, 4, or 8 elements. The subgroup with
one element is {id}; the subgroups with two elements are (r2) = {id, r2},
(s), (rs), (r 2s), (r3 s ); the subgroups with four elements are (r), (s,r 2s) =
{id, s, r 2s, r2}, (rs, r3 s ) = {id, rs, r3s, r2}; the subgroup with eight elements
is D4 itself. The notation (Xl, ... , Xn) means the subgroup generated by
the elements Xl, ... ,Xn- One of the subgroups with four elements is cyclic
of order 4, and the other two are generated by symmetries defined by two
orthogonal lines: (AC) and (BD) or !::.. and !::..'; showing that these are the
only subgroups is a simple matter of patience.
Every bijection {A,B,C,D} ----> {1,2,3,4} defines an embedding of D4
into S4, but there are only three distinct subgroups of S4 isomorphic to D4
(a quick proof is given by the Sylow theorems; see Exercise 8.2).
8.7.3 The Galois Group of X4 +2

The polynomial X4 + 2 is irreducible over Q. Its roots in <C are ±( 0 and
±i( 0, where ( = e i7r / 4 = (1 + i) / J2. The splitting field of X4 + 2 is thus
N = Q[(0,i(0]. As N c Q[0,i] and (0-i(0 = 2 x TI/4, ~ lies
in N, so N = Q[0,i].
We see that [N : Q] = 8 since 0 is of degree 4 over Q and i is of degree
2 over Q[ 0]. Thus, the Galois group G of X4 + 2 is of order 8. We also
know that it is a subgroup of S4. Using the Sylow theorems, we can assert
that G ~ D4 without even giving an explicit isomorphism.
This is not sufficient for our purposes, because we want to use the group
D 4 as a model group, in the sense that we know how to determine its
subgroups, normal subgroups, etc., and can transport this information into
G by explicit isomorphism.
To define the elements of G, it suffices to give their value at 0 and i,
as we saw in Chapter 6. Defining p and u by p( 0) = i 0, p( i) = i and
u( 0) = 0, u(i) = -i, we can check that p is of order 4 and u of order
2, and that p and u generate G and the eight elements of G are defined as
in Table 8.3.
In this table, we added columns giving the images of the remarkable
elements J2 and (, although these can of course be deduced from the
other columns. An isomorphism cp : D4 ----> G is strongly suggested by the
notation, namely the one defined by cp(r) = p, cp(s) = u, etc.
8.7.4 The Galois Correspondence

Using the isomorphism cp : D4 ----> G defined in §7.3, we obtain the subgroups
of G as the images of those of D 4. They are given by {id} , (p2), (u), (pu),
8.7 The Example X4 +2 131
~ i J2 (
id ~ i J2 (
P i~ i -J2 -(
p2 -~ i J2 (
p3 -i~ i -J2 -(
(J ~ -i J2 -i(
P(J iV12 -i -J2 i(
p2 (J -V12 -i J2 -i(
p3 (J -iV12 -i -J2 i(
TABLE 8.3.
(p2(J), (p3(J), ((J, p2(J) , (P(J, p3(J) , (p), and G itself. We can then proceed to
completely determine the trellis of intermediate extensions between Q and
N, and its Galois correspondence with the trellis of subgroups of G (Figure
8.7).
FIGURE 8.7. 'Trellis of subgroups and the corresponding intermediate extensions

Do not forget that some intermediate extensions are obvious once we

know that N = Q[{/2,i] : Q[{/2],Q[i],Q[J2], etc. Using the table of val-
ues of p and a and the degrees of these extensions, we see to which sub-
groups they correspond. For example, Q[iJ2] is of degree 2 over Q, so
it corresponds to a subgroup with 8/2 = 4 elements. As pa(iJ2) = iJ2
and p2(iJ2) = iJ2, this subgroup is necessarily given by (pa, p3a ). Thus,
Q[iJ2] = J( (pa, p3a )).
To identify J( (pa)), in the absence of a better idea, we need to have
recourse to linear algebra. A basis of N over Q is given by 1, {/2, V2, ~,
i, i~, iV2, and i~. For
x = a+b{/2+cJ2+d~+ei + ji~+ giJ2 + hi~,
with coefficients in Q, the equality x = pa(x) implies that b = j, c =

e = 0, h = -d; we are tempted to set 0: = (1 + i) ~ and note that
x = a+bo:+g'0:2+d'0:3 with coefficients in Q, giving J((pa)) = Q[o:]. Note
that Q[o:] = Q[(3{/2]. Similarly, we find that J((la)) = Q[(~].
Looking for normal extensions in the trellis of extensions between Q and
N is equivalent to looking for normal subgroups in the trellis of subgroups
between {id} and G. The trivial normal subgroups G and {id} correspond
to Q and N. The subgroups of index 2 are all normal, and the corresponding
extensions of degree 2 are obviously normal. The only non-trivial case is
that of the subgroup (p2), corresponding to the normal subgroup (r2) of
D 4 ; it corresponds to the normal extension Q[i, J2] of Q. Even if we already
knew, before reading this chapter, that these extensions of Q are normal,
we can now assert that they are the only normal ones in the trellis of
extensions between Q and N.
8.7.5 Search for Minimal Polynomials

Knowing the Galois group gives a new way of finding the minimal polyno-
mial of an element of a normal extension.
For example, to find the minimal polynomial P of 0: = i + ( over Q, we
look for the set of conjugates of 0: over Q; this is the set of images of 0:
under the elements of G, and we note that this set contains exactly four
elements: 0:, i - ( = p(o:), -i - i( = a(o:), -i + i( = pa(o:). After some
longish computations, we obtain
P(X) (X - i - ()(X - i + ()(X + i + i()(X + i - i()

X4 + 2X2 + 4X + 2.
Toward Chapters 9, 10, and 12 133
Toward Chapters 9, 10, and 12

It is now possible to give some important applications of Galois theory: to
roots of unity, first of all, then to extensions with cyclic Galois group, and
finally to the problem of solvability by radicals.

In the two first exercises, we collect the basic facts about group actions on
sets, and the Sylow theorems necessary to understand some parts of the
main text. The solutions are contained in all classic algebra books.
Exercise 8.1. Groups acting on sets

We say that a group G acts on a set E if there exists a group ho-
momorphism r.p : G -> SE where SE is the permutation group of E,
i.e. the set of bijections of E to itself. For every 9 E G and every
x E E, the action of 9 on x gives an element which we denote by
g. x = r.p(g) (x). In this exercise, we show the following properties:
1) e· x = x for every x E E, where e is the identity element of G;
2) (gh)· x = g. (h· x) for every x E E and every g, hE G.
Some frequently encountered examples are the action of a subgroup

of Sn acting on the set {I, ... , n}, the actions of geometric groups
on a set of points in the plane (rotations of the points on a circle,
etc.), the action of a group on itself or on one of its quotients by left
translation, the action of a group acting on the set of its subgroups
by conjugation, and the action of the Galois group of a polynomial
acting on its roots.
We define the orbit of a point x E E under the action of G
by Oc(x) = {g. x,g E G}, and the stabilizer Sc(x) by Sc(x) =
{g E G, g. x = x}. We simply write O(x) and S(x) when there is no
danger of confusion.
We say that G acts transitively if the whole set forms a single orbit,
i.e. if for every x, y E E, there exists 9 in G such that y = 9 . x. A
transitive subgroup of Sn is a subgroup of Sn which acts transitively
on {l, ... ,n}.
In what follows, we assume that G and E are finite.
1) Show that S(x) is a subgroup of G and that IS(x)IIO(x)1 = IGI.
,·X
2) a) Let x and y be two elements belonging to the same orbit, so that
y= for some, E G. Show that S(x) and S(y) are conjugate
subgroups; more precisely, show that S(y) = ,S(xh- 1 .
b) If G acts transitively on E, show that the stabilizers of the

elements of E all have the same number of elements.
3) Let p be a prime. We say that a group is a p-group if its order is a

power of p.
a) Burnside's theorem: Show that a non-trivial p-group has a non-

trivial normal subgroup, namely its center, by letting G act
on the set of its conjugacy classes (the conjugacy class of an
element x EGis the set {gxg-l,g E G}), and counting the
elements modulo p.
b) Show that a group of order p2 is necessarily commutative.
Exercise 8.2. The Sylow theorems

The Norwegian mathematician Ludwig Sylow (1832-1918) was ex-
tremely interested in the work of Abel and Galois; he even wrote
their biographies. His results on finite groups, published in 1872, are
extremely useful.
If a prime number p divides the order of a finite group G, we define
a p-subgroup of G to be any subgroup of G whose order is a power of
p, and a Sylow p-subgroup of G to be any p-subgroup of G whose order
is the maximal power of p dividing IGI. Let mp denote the number of
Sylow p-subgroups. The Sylow theorems assert that
1) every p-subgroup of G is contained in a Sylow p-subgroup of G;

2) mp = 1 mod p (so mp -=f. 0; one easy consequence of this result
is Cauchy's theorem proving the existence of elements of order
p in G);
3) the group G acts transitively by conjugation on the set of its Sy-

low p-subgroups, which implies that any two Sylow p-subgroups
of G are conjugate, so that in particular mp divides IGI and that
if mp = 1, the unique Sylow p-subgroup of G is normal in G.
The Sylow theorems are particularly interesting for groups whose

order is not a prime power. Using the relations mp = 1 mod p and
the fact that mp divides IGI, we can list the possible values of mp for
a given group G; if IGI is small, there are not many possibilities.
Exercise 8.3. Non-normal extensions

Set L = Q[~, <13]. Determine G = Gal(LIQ). Does the Galois cor-
respondence hold here?
Exercise 8.4. Computation of Galois groups

For each of the following polynomials P of degree n, determine:
• the Galois group G of the splitting field N of P over Q;
• the correspondence between subgroups of G and subextensions of N;
• the factorization of P over each of its intermediate extensions;
• a subgroup of Sn isomorphic to G, by numbering the roots.

1) X2 - 2; 2) (X2 - 2)(X2 - 3);
3) X 3 - 1; 4) X 3 + 2;
5) X4 + 1; 6) X4 - 1.
Exercise 8.5. Galois groups

1) Let N be a normal extension of finite degree of a field K. Let H be a
subgroup of Gal(NIK), and let J(H) be the field of invariants under
H. Show that [J(H) : K] = IGal(NIK)jHI.
2) Let L c <C be an extension of finite degree of a field K, and let G be

the Galois group of Lover K. Show that if IGI = [L : KJ, then L is
a normal extension of K.
3) For every even integer n ~ 2, find a polynomial of degree n with

distinct non-rational roots, whose Galois group over Q is Zj2Z.
4) Let C(X) be the field of rational functions in one indeterminate, and

let G be the group of <C-automorphisms of C(X) consisting of the set
of maps eTa defined for every a E <C by
( P(X)) P(X+a)
eTa Q(X) = Q(X + a)"
Show that J (G) = C.
Exercise 8.6. Galois groups that are direct products

1) Let K be a field, and let N be a normal extension of K. Let L' and
L" be extensions of K contained in N, and set G = Gal(NIK), G' =
Gal(NIL') and G" = Gal(NIL"). Show that if L' and L" are normal
extensions of K such that L' U L" generates Nand L' n L" = K, then
G is the direct product of its subgroups G' and G", i.e. G ~ G' x G".
2) Show the converse: if K is a field and N a normal extension of K, if

G = Gal(NIK) is a direct product G ~ G' x G", and if L' = J(G')
and L" = I(G"), then L'UL" generates Nand L'nL" = K. (Recall

that G is a direct product of two of its subgroups Hand K if
a) Hand K are normal subgroups of G;
b) HnK={e};
c) HK = {hk,h E H,k E K} = G.)
3) Show that, under the conditions of 1), we have
Gal(NIK) ~ Gal(L'IK) x Gal(L"IK).
Exercise 8.7. Cubic equations

Consider the polynomial P(X) = X 3 + pX + q. Write a, b, c for its
roots and D for its discriminant, and let d be a complex number such
that d 2 = D, K = Q(p,q), G = Gal(K[a,b,c]IK).
1) Assume that P is reducible over K. What are the possible structures

of G? Give them in terms of d.
2) Assume that P is irreducible over K. What are the possible group

structures on G? (Distinguish between the case d E K and the case
d f/. K.) Give the intermediate extensions between K and K[a, b, c] in
each case.
3) Determine the Galois groups of

a) X 3 - 2X - lover Q and over Q[v5];
b) X 3 - 3X + lover Q;
c) X 3 - 4X + lover Q;
d) X 3 - 2X + 2 over Q and over Q[iv'19];
e) X 3 - 5 over Q, over Q[ij5] and over Q[iV3];
f) X 5 + X2 - 9X + 3 over Q.
Exercise 8.8. Biquadratic extensions

The fields we consider here are all intermediate extensions between
Q and <C.
A biquadratic extension of a field K is an algebraic extension L of
degree 4 of K such that there exist a and b E K with L = K[Ja, Jb].
1) a) Show that every biquadratic extension N of a field K is normal;

give its Galois group G and the Galois correspondence between
subgroups of G and intermediate extensions between K and N.
b) Let N be a normal extension of degree 4 of a field K such

that Gal(NIK) ::: '"£/2'"£ x '"£/2'"£. Show that N is a biquadratic
extension of K.
2) Set a = V6 + Jil, (3 = V6 - Jil, and N = Q[al·

a) Determine the minimal polynomial P of a over Q.
b) Compare Q[JITl and N. Show that (3 lies in N, and give its

expression in the basis {1, a, a 2 , a 3 } .
c) Show that N is a normal extension of degree 4 of Q.
d) Set G = Gal(NIQ). What is the structure of G?
e) What are the degrees of a + (3 and a - (3 over Q? Deduce that

a and (3 can be written in the form .;a ± Vb for elements a and
b E Q; compute a and b.
f) Give the trellis of intermediate extensions between Q and N.
Exercise 8.9. The normal closure of Q [{II +)2]

Set P(X) = X 3 + 3X - 2, u = {II +)2 and v = {II -)2, where
the cube roots lie in R Let D be the discriminant of P, and let d be
the number, with positive imaginary part, such that d 2 = D and N
is the normal closure of Q(u) over Q.
1) Determine the number of real roots of P.

Now, let c, a, and b denote the roots of P with negative, zero, or
positive imaginary parts respectively, and let r be the Galois group
Gal(Q[a, b, clIQ).
2) Determine a, b, c and the structure of r.
3) What is the degree of u over Q?
4) Show that N = Q[u,jl.
5) Give examples of distinct non-trivial intermediate extensions between

Q and N.
Now, let G be the Galois group Gal(NIQ).

6) Determine the order of G.
For what follows, one can proceed by constructing a table giving

the values g(x) for 9 in G and certain x in N.
7) Let a,p : N -+ N denote the elements of G such that a(u) = u,
a(j) = l, p(u) = jv, and p(j) = j2. What are the orders of a and
p? Show that p and a generate G. Compute p(V2),a(d), and p(d).
8) What is the structure of G?
9) Which subgroup of Gleaves Q[u] invariant? Is it a normal subgroup

ofG?
10) Determine the extensions invariant under the groups (a, p2), (pa, p2),
(p).
11) Give the precise Galois correspondence between subgroups of G and

intermediate extensions between Q and N.
Exercise 8.10. Linear independence of v'Pk for prime numbers Pk
Set K = Q[V2, V3].

1) Show that v's is of degree 2 over K, by comparing Q[v's] to the
intermediate extensions between Q and K.
Set L = Q[V2, V3, v's].
2) Show that L is a normal extension of Q.
3) Determine the structure of the Galois group G' = Gal(LIQ) (recall

that a group all of whose non-trivial elements are of order 2 is isomor-
phic to a product of groups isomorphic to 7l,/27l,j this result follows if
one considers the group as a 7l,/2Z-vector space).
4) Let us admit the fact that G' has seven subgroups of index 2 (they
are hyperplanes for the 7l,/2Z-vector space structure, and there are
as many of them as there are independent non-zero vectors in the
dual of this vector space). Find the set of quadratic extensions of Q
contained in L.
5) Generalize the above to extensions of Q by {vPjP E E}, where E is

a finite subset of the set of prime numbers.
COMMENTARY. - By results of Besicovitch and J. Richard (see the book by

Gaal), this result can be generalized to a finite set of n-th roots of distinct
prime numbers.

Solution to Exercise 8.3. G = {id} since the IQ-algebra L is generated
by ?'2 and {13, and a IQ-automorphism of L leaves them fixed since their
other conjugates are not real, so do not lie in L. The Galois correspondence
does not hold in this situation since L has non-trivial subextensions, namely
1Q[?'2], IQ[ {13], whereas G has no non-trivial subgroups.

1) X2 - 2 has splitting field 1Q[v'2] over IQ. As this extension is of degree 2
over IQ, the group Gal(IQ[V2]IIQ) has two elements, the identity on 1Q[v'2]
and the element a : 1Q[v'2] -> 1Q[v'2] defined by a(v'2) = -v'2. This group
is thus isomorphic to Zj2Z and has only trivial subgroups. Thus, there
is no intermediate extension between IQ and IQ[ v'2], which we could have
noted directly by considering degrees. So we obtain G::::o S2.
2) (X2 - 2)(X2 - 3) has splitting field 1Q[v'2, V3] over IQ. As this extension
is of degree 4 over IQ, we find that G = Gal(lQ[ v'2, V3]IIQ) has four elements;
thus it is isomorphic to either Zj4Z or Zj2Z x Zj2Z. To decide which of
these is the right group, we need to study G; its elements 0'1, 0'2, 0'3, 0'4
are extensions of IQ-automorphisms of 1Q[v'2]. They are are defined by the
images of v'2 and V3, and these images are conjugates, so they must be
equal to ±v'2 and ±V3 (Table 8.4).
v'2 V3
0'1 v'2 V3
0'2 -v'2 V3
0'3 v'2 -V3
0'4 -v'2 -V3
TABLE 8.4.
We see that the element 0'1 is of course the identity element, and 0'2, 0'3,
0'4are of order two. Thus G is isomorphic to Zj2Z x Zj2Z; its subgroups
are {ad, (a2), (a3), (0'4), and G. The corresponding fields of invariants are
Q[v'2, v'3], Q[v'3], Q[J2J, Q[V6] and Q. We can guess these intermediate
extensions easily and check that they work: for example, Q[v'3] is invariant
under (0"2) and its degree is equal to the index of (0"2) in G. Without this
intuition, we would need to work with a basis of Q[v'2, v'3] over Q, for
example 1, v'2, v'3, V6, and solve the equations x = O"i(X), i = 2,3,4, with
x = a + bv'2 + cv'3 + dV6.
The factorizations of (X2 - 2)(X2 - 3) are given by
(X - v'2)(X + v'2)(X 2 - 3) over Q[v'2];
(X - v'3)(X + v'3)(X2 - 2) over Q[v'3];
(X2 - 2)(X2 - 3) over Q[J6].
Choosing the correspondence defined in Table 8.5, we obtain the elements
of the subgroup of 84 isomorphic to G (Table 8.6).
TABLE 8.5.
0"1 0"2 0"3 0"4

id (1 2) (3 4) (1 2)(34)
TABLE 8.6.
3) X 3 - 1 has splitting field Q[j] over Q, and we can reason exactly as in

part 1).
4) X 3 + 2 has the same splitting field over Q as X 3 - 2, since its roots are
the negatives of the roots of X 3 - 2. So we can reason as in §8.3 and §8.6.
5) We saw (Exercise 7.1) that the splitting field of X4 + lover Q is N =

Q[(] = Q[i, v'2] with ( = ei 7r:/4 = (1 + i)/v'2. We find the intermediate
extensions Q[iJ, Q[v'2] and Q[iv'2] by the same procedure as in 2), and the
elements of the Galois group are defined by Table 8.7.
i J2 ( (3 (5 (7
0"1 i J2 ( (3 (5 C
0"2 -i v'2 C (5 (3 (
0"3 i -v'2 (5 C ( (3
0"4 -i -v'2 (3 ( C (5
TABLE 8.7.
The factorizations of X4 + lover an intermediate extension L are ob-

tained by regrouping the roots of X4 + 1,which are conjugates over L. To
compute the action of G on these roots, it is useful to add columns giv-
ing their images to Table 8.7; for example, the conjugate of ( over lQ[i] is
a3 (() = (5, etc. The factorizations of X4 + 1 follow from this:
• over lQ[i] :
• over IQ[ v'2] :
• over lQ[iv'2] :
6) X4 - 1 has splitting field lQ[i] over IQ; the solution works exactly as in
1). By numbering the numbers 1, -1, i, -i as 1, 2, 3, 4, we identify the
group G with the subgroup of S4 given by id and (34).

1) The quotient Gal(NIK) / H is not necessarily a group, but it is a coset
space of order IGal(NIK)I/IHI = [N : K]/[N : I(H)] = [I(H) : K].
2) The number of K-homomorphisms from L to C is equal to [L : K], so

the number of K-automorphisms of L is ~ [L : K]. If we have equality, this
means that a(L) C L for every K-homomorphism a of L in C, which in
turn means that L is a normal extension of K.
3) The different factors do not bring any more information than does the
first. Examples: II(X2 + k 2) whose splitting field is lQ[i]' II
(X2_
l:Sk:Sn
2k2) whose splitting field is IQ[ v'2], etc.
4) Assume that P and Q are relatively prime. If P(X + a)/Q(X + a) =

P(X)/Q(X), we have
P(X + a)Q(X) = P(X)Q(X + a).

P(X), which is relatively prime to Q(X), divides P(X + a), so that we
have P(X +a) = AP(X). Comparing the highest degree terms implies that
A = 1; if deg(P) = n 2': 1, then comparing the terms of degree n - 1 gives

a contradiction for a =1= o.

1) If L' u L" generates N, sup(L', L") = N in the trellis of extensions, so
inf(G',G") = {id} , i.e. G' n Gil = {id}.
If L' n L" = K, then inf(L',L") = K in the trellis of extensions, so
sup(G',G") = G, i.e. G' and Gil generate G.
If Land L' are normal extensions of K, then G' and Gil are normal
subgroups of G.
These three conditions imply that G '::= G' X G".
2) The converse is shown simply by running the argument backward.
3) The first question asserts that Gal(NIK) '::= Gal(NIL') x Gal(NIL").

Proposition 8.4 shows that we have both Gal(NIK)/Gal(NIL') '::= Gal(L'IK)
and Gal(NIK)/Gal(NIL") '::= Gal(L"IK). It follows that Gal(NIL')
Gal(L"IK) and Gal(NIL") '::= Gal(L'IK), which gives the result.

1) If a, b, c lie in K, then G = {id} and d E K. If a and b are of degree 2
and conjugate over K, then c lies in K, d rf- K and G '::= Z/2Z.
2) a) If d E K, [K[a, b, c] : K] = [K[a, d] : K] = 3 so G '::= Z/3Z, and there

is no non-trivial intermediate extension.
If d is of degree 2 over K, then [K[d] : K] = 2 and [K[a] : K] = 3, so that
[K[a, b, c] : K] = 6. Thus G is of order 6. As G is isomorphic to a SllbglOup
of 8 3 , it must be isomorphic to 8 3 . There are four intermediate extensions
corresponding to the four non-trivial subgroups of 83. The one of degree 2
is K[d]. The ones of degree 3 are K[a], K[b] and K[c]. The elements of G
are given in Table 8.8.
a b c d
(i1 a b c d
(i2 b c a d
(i3 C a b d
(i4 a c b -d
(i5 c b a -d
(i6 b a c -d
TABLE 8.8.
3) a) As X 3 -2X -1 = (X + 1)(X2 -X -1), we see that the splitting field

of X 3 - 2X - 1 is Q[J5]; its Galois group over Q is isomorphic to 7L/27L,
and its Galois group over Q[J5] is trivial.
b) As X 3 - 3X + 1 is irreducible over Q (the possible rational roots,
namely ±1, do not work), and moreover D = 9 2 , we find that G ::: 7L/37L.
c) Similarly, X3 - 4X + 1 is irreducible over Q and, as D = 229 is not a
square in Q, we have G::: S3.
d) X3 - 2X + 2 is irreducible over Q (use Eisenstein's criterion); con-
sequently, it is also irreducible over = QQ[iJ19]. As D = -76, its Galois
group over Q is isomorphic to S3, and its Galois group over Q[iJ19] is
isomorphic to 7L/37L.
e) X 3 - 5 is irreducible over Q and Q[iV3]; as its discriminant is D =
-3 X 15 2 , its Galois group over Q is isomorphic to S3 and its Galois group
over Q[iV3] is isomorphic to 7L/37L. Over Q[ ij5], X3 - 5 is reducible and
its Galois group is isomorphic to 7L/27L.
f) The polynomial factors as (X2 + 3) (X 3 - 3X + 1). The splitting fields of
the two factors are of degrees 2 and 3 over Q, and their Galois groups over Q
are isomorphic to 7L/27L and 7L/37L respectively. By Exercise 8.6, the Galois
group of X 5 + X 2 - 9X + 3 over Q is isomorphic to 7L/27L x 7L/37L ::: 7L/67L.
A generator of this group is the Q-automorphism defined by a(a) = band
a(iV3) = -iV3, where a and b are two distinct roots of X 3 - 3X + 1.

1) a) If N = K[y'a, Vb], then N is the splitting field of the polynomial
(X2 -a)(X 2 -b) over K, so it is a normal extension of K. The Galois group
has four elements, namely the elements 0'1 = id, 0'2, 0'3 and 0'4 defined in
Table 8.9.
0'1 0'2 0'3 0'4
y'a va va -va -va

vb vb -vb vb -vb
TABLE 8.9.
Clearly, G has three elements of order two, so it is isomorphic to 7L/27L x

7L/27L. The subgroups of order two of G are generated by these three ele-
ments. They correspond to three intermediate extensions:
b) Let HI and H2 be two distinct subgroups of order two of Gal(NIK).

The fields I(Hd and I(H2 ) are distinct quadratic extensions of K, so they
generate N; there exist a and bE K such that I(Ht} = K[v'a] and I(H2) =
K[Vb] (see Exercise 4.7). Hence N = K[y'a, Vb].
2) a) Clearly 0: is a root of the polynomial P(X) = X4 - 12X2 + 25.
We have P(X + 1) = X4 +4X3 - 6X 2 - 20X + 14. Applying Eisenstein's
criterion with p = 2 shows that P(X + 1) is irreducible over Q, so P(X)
is also irreducible. We can also try to factor P in Z[X] using Descartes'
method, but this turns out to be impossible.
b) As Jll = 0: 2 - 6, we have Q[v'IT] C Q[o:]. The inclusion is strict

because Q[o:] is an extension of degree 4 of Q by a). Furthermore, 0:f3 = 5
shows that f3 = 5/0:; thus f3 lies in Q[o:]. As 0:(0:3 - 120:) = -25, we have
f3 = (12/5)0: - (1/5)0: 3 .
c) The roots of Pare ±o: and ±f3. As f3 E Q[o:] by b), Q[o:] is a normal
extension of Q.
d) When 0" is in G, 0"(0:) is a conjugate of 0:, so 0"(0:) equals ±o: or ±f3.

The group G contains the four elements 0"1 = id, 0"2,0"3 and 0"4 defined by
Table 8.10.
0"1 0"2 0"3 0"4

0: 0: -0: f3 -f3
f3 f3 -f3 0: -0:
TABLE 8.10.
Since we know that
and
0"4(0"4(0:)) = 0"4 ( -f3) = 0"4 ( -~) = %= 0:,
we see that G contains three elements of order 2, so it is isomorphic to
Z/2Z x Z/r2Z.
e) The conjugates of 0: + f3 are 0: + f3 and 0"2(0: + f3) = -0: - f3. The
conjugates of 0: - f3 are 0: - f3 and 0"2(0: - f3) = f3 - 0:. Thus 0: + f3 and
0: - f3 are of degree 2 over Q. We find that (0: + f3)2 = 12 + 10 = 22,
(0: - f3)2 = 12 - 10 = 2. But 0: + f3 and 0: - f3 are> 0, so 0: + f3 = y'22 and
0: - f3 = .12. Finally, we have 0: = Jll/2 + Jlfi and f3 = Jll/2 - Jlfi.
f) The intermediate extensions of degree 2 over Q are given by


1) We have D = _4p 3 - 27q2 = -216 and d = 6iv'6. Now, P is a cubic
polynomial with real coefficients and negative discriminant, so it has one
real root and two complex conjugate roots.
2) Cardan's formulas give a = u + v, b = ju + j 2 v, c = j 2 u + jv (the

inequalities u > 0 and v < 0 make it possible to see which of j 2 u + jv and
ju + lv has positive imaginary part).
We check that P is irreducible over Q by making sure that none of the
possible candidates for rational roots works: ±1 and ±2 are not roots of P.
As -216 is not a square in Q, r is isomorphic to 8 3 .
3) As u3 - 1 = h, u is algebraic over Q[h], so it is algebraic over Q. As

a = u + v and v = -lju, we see that a E Q[u]. Thus Q[u], which contains
the extensions Q[h] and Q[a], of degrees 2 and 3 over Q, is of degree a
multiple of 6 over Q. The equality (u 3 - 1)2 = 2 shows that u is of degree
::::; 6. We conclude that the degree of u over Q is equal to 6 and that its
minimal polynomial over Q is given by X 6 - 2X 3 - 1.
4) The conjugates of u over Q are the solutions of (x 3 - 1)2 = 2; we find

u, ju, j 2 u, v, jv, j 2 v. We check that N c Q[u,j] because v = -lju, and
that Q[u,j] eN because j = juju. This gives N = Q[u,j].
5) We first mention Q[a], Q[b], Q[c], of degree 3 over Q, Q[dj = Q[iv'6],

of degree 2 over Q, and Q[a, b, c], of degree 6 over Q. Other intermediate
extensions are easy to see: Q[u], of degree 6 over Q. We can also mention
Q[j, h], of degree 4 over Q, Q[ju] = Q[j 2 v], Q[j 2 u] = Q[jv], of degree 6
over Q.
6) As u is real and j is not, j is of degree 2 over Q[u] and [N : Q] = 12. As

N is a normal extension of Q, we have IGI = [N : Q] = 12.
7) We see that 17 2 = id so a is of order 2. As uv = -1, p(uv) = -1,

p( v) = -1 j p( u) = -1 fjv = j 2 u. We can then fill out the first three columns
of Table 8.11. We see that p is of order 6 and that the 12 elements of G
can be written pk and pka with 0 ::::; k ::::; 5. As h = u3 - 1, p( v'2)
(jv)3 - 1 = -v'2. Finally, we have
a( d) = 6a( iV3)a( h) = 6( -iV3)( h) = -d,
p(d) = p(6iv'6) = 6p(iV3)p(h) = 6(-iV3)(-h) = d.
8) Table 8.11 shows that the Q-automorphisms pk and pka for 0 ::::; k ::::;
5 are all different, so that we have indeed found all the elements of G.
Consequently, p and a generate G.
Moreover, ap = p5a, as it is easy to check on u and j. With the relations
p6 = 17 2 = id, this shows that G is isomorphic to the dihedral group D 6 •
9) As [Q[u) : Q) = 6, Gal(NIQ[u]) has two elements. Table 8.11 shows that

these are exactly id and a. As Q[u] is not a normal extension of Q, this
subgroup is not normal in G.
10) (a,p2) has six elements: id, p2, p4, a, p2a and p4a, so we are looking
for a quadratic extension of Q. Table 8.11 shows that it is given by Q[v'2).
(pa, p2) has six elements: id, p2, p4, pa, p3a and p5a , so we are looking
for a quadratic extension of Q. Table 8.11 shows that it is given by Q[j).
(p) has six elements, so we are looking for a quadratic extension of Q.
Table 8.11 shows that it is Q[d).
u j v v'2 d a,b,c
id u J v v'2 d a,b,c
p jv j2 j 2u -v'2 d c,a,b
p2 ju j j 2v v'2 d b,c,a
p3 v j2 U -v'2 d a,b,c
p4 j 2u j jv v'2 d c,a,b
l j 2v j2 ju -v'2 d b,c,a
a u i v v'2 -d a,c,b
pa jv j j 2u -v'2 -d c,b,a
p2a ju j2 j 2v v'2 -d b,a,c
la v j u -v'2 -d a,c,b
la j 2u j2 jv v'2 -d c,b,a
p5 a j 2v j ju -v'2 -d b,a,c
TABLE 8.11.
11) An isomorphism of G with D6 is given by associating to p the rotation

l
r of angle 7r /3 and center 0, and to a the symmetry rk s with respect to
the line making an angle of k7r / 6 with (0 A).
This isomorphism gives the list of subgroups of G by making the list of
subgroups of D 6 , which is easier thanks to geometric considerations.
Subgroups with two elements
These subgroups are generated by elements of order 2 and correspond
to the symmetries with respect to 0 or with respect to lines. We find:
HI = (p3), Hk+2 = (pka) with 0:::; k :::; 5.
Subgroups with four elements
As there is no element of order 4, these subgroups all correspond to the
subgroups of D6 generated by symmetries with respect to two perpendic-
ular axes (which thus commute).
They are given by Hs = ((j, p3(j), Hg = (P(j, p4(j), HlO = (p2(j, p5(j); they
all contain p3.
Subgroups with three elements
The only such subgroup is given by Hll = (l), because the only elements
of order 3 in D6 are p2 and p4.
Subgroups with six elements
As p and p5 are the only elements of order 6, H12 = (p) is a cyclic
subgroup of order 6. If there are other subgroups with six elements, they
are isomorphic to S3, which is isomorphic to D 3, the dihedral group of
isometries of an equilateral triangle. There are two ways of placing an
equilateral triangle with center 0 with respect to the hexagon: a vertex at
A or the rotation of this by 1r /6. Thus we find two other subgroups of G of
order 6:
Hl3 {id,(j,p2(j,p4(j,p2,p4} = ((j,p2)
H14 {id,p(j,p3(j,p5(j,p2,p4} = (p(j,p2).
Let Li denote the extension corresponding to the subgroup Hi, 1 ::; i ::; 14.
To find L i , we will look for an extension L invariant under Hi (so L C L i )
of suitable degree: [Li : Q] = [N : Ql/[N : L i] = IGI/IHil.
Among the subgroups of G, only HI, H ll , H 12 , H 13 , and H14 are normal;
the corresponding extensions are thus normal extensions of Q. It is useful
to add a column to Table 8.11 to describe the action of the elements of G
on the roots a, b, c of P. We find that
LI = Q[a, d] = Q[a, b, c], L2 = Q[u] = Q[v] = Q[a, v'2],
L3 = Q[b,j], L4 = Q[j 2 u] = Q[c, v'2J,
L5 = Q[a,j], L6 = Q[ju] = Q[b, v'2],
L7 = Q[c,j], Ls = Q[a],
Lg = Q[bJ, LlO = Q[c],
Lll = Q[j, d] = Q[j, v'2], L12 = Q[d] = Q[iv'6],
L13 = Q[v'2J, L14 = Q[j] = Q[iJ3].

1) If J5 E K, then Q[v's] is one of the three intermediate extensions Q[V2],
Q[J3]' Q[v'6], but it is easily seen that this is impossible.
2) L is the splitting field of (X2 - 2)(X2 - 3)(X2 - 5) over Q.
3) As [L : Q] = 8, the order of G' is 8, so G' ~ (Z/2Z)3.
4) The complete list is easy to establish: Q[V2], Q[J3J, Q[v's]' Q[v'6J,
Q[ v'iO]' Q[ 05], Q[ v'35].
5) We use induction on the cardinal n of E, and show that Q[{ Vp; P E E}l
is an extension of Q of degree 2n and Galois group (Z/2Zt, in which
{Vp; pEE} is a free system over Q. The result is obvious if n = 1. As-
sume it holds for some integer n :::: 1, and let E have cardinal n + 1. If
PI E E, then we apply the induction hypothesis to E - {PI}. The field
Q [{ y1P,p E E,p =I- pI}] has 2n - 1 quadratic subextensions, given by the
Q [II vp] for non-empty subsets F of E - {pd, and none of these exten-
pEF
sions is equal to Q[v'Pil. This concludes the proof.
9
Roots of Unity
In this chapter, we give the first example of a family of extensions whose

Galois group is actually computable: this is the family of extensions of a
field by roots of unity. The earliest work on this subject is due to Vander-
monde (1770). It was followed by work of Gauss, in particular his beautiful
discovery, on March 30, 1796, at the age of 19, of the construction of the
regular polygon with 17 sides with ruler and compass (see Exercise 9.7)
and its consequences.
9.1 The Group U(n) of Units of the Ring ZjnZ

9.1.1 Definition and Background
The invertible elements of a ring A are often called units; they form a group
under the multiplication law of A.
Let n be an integer> 1. The group of units of the ring ZjnZ is written
U(n). The map associating the order of the group U(n) to the integer n
is written cp and called the Euler function. Euler introduced it in 1760,
to generalize results of Fermat, and proved the statements 2) and 4) of
Proposition 9.1.2 below. In our notation, we make no distinction between
an integer and its class modulo n.
Recall that k is invertible in U(n) if and only if k is relatively prime to
n. Indeed, if k is prime to n, Bezout's identity gives integers u and v such
that uk + vn = 1, so that uk = 1 in U(n), and if k is not prime to n,

150 9. Roots of Unity
there exist d, u, v -=I- 0 mod n such that k = ud, n = vd. Because kv = 0 in

U(n), k is not invertible in U(n).
EXAMPLES. -
1) The underlying set of the group U(8) is {I, 3, 5, 7}, and we have
!p(8) = 4. Its elements are all of order 2, so we have U(8) ~ (2/22)2.
2) The underlying set of the group U(15) is {I, 2, 4, 7, 8, 11, 13, 14}, so
!p(15) = 8. Its elements are of order 2 or 4, which is the case for
exactly one of the three abelian groups of order 8, so we know that
U(8) c:::' 2/22 x 2/42.
Fortunately, there exists a general theorem giving the structure of the

group U(n).
9.1.2 The Structure of U(n)

PROPOSITION. - The group U(n) is abelian. Moreover:
1) If n = r s where rand s are relatively prime, we have U (n) c:::' U (r) x

U(s) and !pen) = !p(r)!p(s).
2) Ifn = pk, where p is a prime greater than 2, then U(n) is cyclic of

order !p(pk) = pk _ pk-l.
3) For k 2: 2, we have U(2k) c:::' 2/22 x 2/2 k - 2 2; for k = 1, we have

U(2) c:::' {O}.
4) If {Pi; i E I} is the set of distinct prime factors ofn, we have !pen) =

n ll(1- -).
iEI
1
Pi
PROOF. - The proof is entirely given in the exercises (see Exercise 9.8). <>
EXAMPLE. - We have 2,800 = 16 x 25 x 7, so
U(2,800) U(16) x U(25) x U(7) c:::' 2/22 x 2/42 x 2/202 x 2/62,
!p(2, 800) 2,800 (1 - ~) (1 - ~) (1 - ~) = 960.

9.2 The Mobius Function 151
9.2 The Mobius Function

Let D (n) denote the set of divisors of n.
9.2.1 Multiplicative Functions

DEFINITION. - A function f : N* -> Z (or JR) is called multiplicative if for
every pair (m, n) ofrelatively prime numbers, we have f(mn) = f(m)f(n).
EXAMPLE. - The Euler function is a multiplicative function by §9.1.2.
9.2.2 The Mobius Function

Let p, : N -> Z denote the function, known as the Mobius function, defined
by
p,(I) 1,
p,(n) 0 if n has a square factor,
p,( n) ( -1 r if n is a product of T distinct primes.
9.2.3 Proposition
The Mobius function is a multiplicative function, and fOT all n > 1, we
have
L p,(d) = o.
dED(n)
PROOF. - If m and n are relatively prime integers, then either one of

them has a square factor or neither one does. In both cases, we check that
p,(mn) = p,(m)p,(n).
If
n= II (Pi)k;
iE{l, ... ,r}
is the decomposition of n into a product of distinct primes, and if m =

II Pi, we have
iE{l, ... ,r}
L p,(d) = L p,Cd) = L (~)( _1)k,

dED(n) dED(m) O~k~r
which is the expansion of (1 - lr = o.

9.2.4 The Mobius Inversion Formula
PROPOSITION. - Let G be an abelian group written additively. Let g : N* .......

G be a map, and let f : N* ....... G be the map defined by fen) = g(d). L
dED(n)
Then
g(n) = L }L(d)f (~) = L }L (~) fed).

dED(n) dED(n)
REMARK. - If the group law of G is written multiplicatively, then f is

defined by f (n) = II
g( d) and the inversion formula is written as
dED(n)
g(n)= II n)J.L(d)
f (d = II f(d)J.L(n/d).
dED(n) dED(n)
PROOF. - Let us give a direct proof. We have
L }L(d)f (~) L (}L(d) L gee))

dED(n) dED(n) eED(n/d)
L (g(e) L }L(d)).
eED(n) dED(n),eED(n/d)
For every e E D(n), we have
{ d; d E D( n) and e E D (~) } = {d; d E D( n) and d E D (~) } = D (~) ,
so we have
L J.£(d)f (J) = L (g(e) L }L(d)) = g(n)

dED(n) eED(n) dED(n/e)
since L }L(d) = 0 for e < n by the formula of Proposition 9.2.3. <;

dED(n/e)
COMMENTARY. - August Mobius is better known for his band, or for the
invention of barycenters. He introduced his function in 1832, using it for
an inversion formula that was generalized independently by Dedekind and
by Liouville in 1857.
9.3 Roots of Unity 153
9.3 Roots of Unity

9.3.1 n-th Roots of Unity
DEFINITION. - Let n > 1 be an integer. An n-th root of unity in C is a
complex number ( such that C = 1. Any such ( is of the form ( = e 2i k-rr/n
for some integer k with 0 :S k < n.
9.3.2 Proposition
The set of n-th roots of unity in C forms a cyclic group Iln isomorphic to
('Lln'L, +).
PROOF. - Consider the group homomorphism f : ('L, +) --+ C* defined by
f(k) = e2ik1r/n. The kernel of this homomorphism is exactly n'L, and its
image is Iln, which gives the result. 0
9.3.3 Primitive Roots

DEFINITION. - Let n > 1 be an integer. A primitive n-th root of unity in
C is a generator of the cyclic group Iln.
9.3.4 Properties of Primitive Roots

Using the isomorphism defined by f above, we see that the primitive n-th
roots of unity are of the form e 2i k-rr/n with 1 :S k < nand k prime to n,
since the generators of ('Lln'L, +) are the classes of integers relatively prime
to n. Thus, there exist tp(n) primitive n-th roots of unity in C.
If ( and (' are two primitive n-th roots of unity, there exists u prime to
n such that (' = (u.
9.4 Cyclotomic Polynomials

9.4.1 Definition
The n-th cyclotomic polynomial is the polynomial n defined by the fol-
lowing expression:
II
(;EI-'n, (; primitive
(X - ().
9·4.2 Properties of the Cyclotomic Polynomial

We will show how to compute n, and prove that n is the minimal poly-
nomial of every primitive n-th root of unity in C.
The fact that n is irreducible over Q whenever n is prime was proved
by Gauss in No. 341 of his Recherches arithmetiques, after the definition
of the cyclotomic polynomial in No. 339 (Figure 9.1). His proof is not as
rapid as the proof using Eisenstein's criterion.
SSg. VequatiOll Z' - I = 0 (en supposant, comme n taut tou-

jours Ie faire par.la suite, que 11 est un Dombre premier impair),
ne renferme qu'une sewe racine reeUe a;;::; 1; .les. "' - 1 autrcs,
«lui sou donnes par l'equation
...::;~+~= . + etc. +z+ 1 = 0 , . • . . . . (X)
IOnt toutes imaginaires ;
54!. TsEoREaB. Sl la function X (71:' ~) e~, divisible par
une fonction d'UII degre infericur
p =xA +AxA--1 + BxA-Ia +etc.+I.U+L-s'
les col[/ficiem..A, B, ....L ne peUII'" plItI~tre 10118 enli-ers ni
raiionnels.
FIGURE 9.1. Articles 339 and 341 of Recherches arithmetiques
The irreducibility of n over Q for non-prime n was proven by Gauss

somewhat later, in 1808. This problem was studied more deeply by several
19th-century mathematicians: Eisenstein, Dedekind, Kronecker, etc.
PROPOSITION. - For every n ~ 1, we have
1) xn - 1 = II d(X)
dED(n)
2) n(X) = II (X d - l)/L(n/d), and when p is a prime, we have

dED(n)
p(X) = L Xk
OSkSp-l
3) n(X) E Z[X]
4) n (X) is irreducible over Q.
PROOF. -
1) We compute
xn -1 II (X - () = II II (X - ()
dED(n) (E/Ln, ( of order d
9.4 Cyclotomic Polynomials 155
II II (X - () = II d(X).
dED(n) (EJ.Ld, ( primitive dED(n)
2) Consider the multiplicative group G = qX)* of non-zero rational

functions in one indeterminate with complex coefficients. Let f, g :
N* --* G be the maps defined by fen) = xn - 1 and g(n) = n(X).
By 1), fen) = II g(d) and the multiplicative form of the Mobius
dED(n)
inversion formula gives the result.
3) In the preceding formula, the exponents of the polynomials belong to

{O, 1, -I}; thus n(X) is a polynomial which is the quotient of two
monic polynomials in Z[X], which means that it lies in Z[X].
4) Let w be a primitive n-th root of unity, and let P be its minimal

polynomial over Q. Let E be the set of roots of P in C, and let F be
the set of primitive n-th roots of unity. Let us show that E = F.
Because n(w) = 0, we see that P divides n, so E c F.

To prove the converse inclusion, we first show that E is stable
under raising to p-th powers whenever p is a prime not dividing n.
To show this, suppose it is not the case, so there exists an element
0:E E and a prime p not dividing n such that o:P ~ E. Let 8 be the
polynomial such that Xn - 1 = P(X)8(X). As o:P ~ E, P(o:P) =/:- 0,
so 8(o:P) = 0, which shows that 0: is a root of 8(XP). As P is the
minimal polynomial of 0:, it divides 8(XP). Set 8(XP) = P(X)T(X).
The polynomials 8 and T lie in Z[X], since P is monic. We now
consider the situation in (ZjpZ)[X], adding a subscript 1 to denote
the images of polynomials of Z[X] in (ZjpZ)[X]. We have 8 1 (XP) =
(8 1 (X))P = P1 (X)T1 (X) (to see why, consult §14.4.2). As (ZjpZ)[X]
is a factorial ring, every irreducible factor U1 of PI divides 8 1 ; the
equality xn - 1 = P1 (X)8 1 (X) shows that (U1 )2 divides xn -1; it
follows that U1 divides the derivative nX n- 1 of Xn -1 in (ZjpZ)[X].
As p is prime to n, these two polynomials are relatively prime in
(ZjpZ)[X], which is a contradiction.
Now, we can finish the proof that FeE. Let ( E F; there exists
a number u prime to n such that ( = wU • Let u = (Pi)k i be II
l<i<r
the decomposition of u as a product of prime factors. Th~ fact that
E is stable under raising to p-th powers whenever p is a prime not
dividing n shows that 0: = W U = (... (w P1 ) .•. )pr lies in E. Thus, we
have E = F, so n = P, and n is irreducible over Q. <>
EXAMPLE. -
tP28(X) = IT (Xd _1)JL(28 j d)

dED(28)
(X 28 _1)JL(1)(X 14 _1)JL(2)(X 7 _ 1)JL(4)(X4 _1)JL(7)
.(X2 _1)JL(14)(X _ 1)JL(28)
(X 28 _1)(X 14 _ 1)-1(X7 _ 1)0(X 4 _1)-1(X2 _ l)(X _1)0
(X 28 - 1)(X2 - 1)
(X14 - 1)(X4 - 1)
(X14 + 1)
(X2 + 1)
X12 _ X lO + X 8 _ X 6 + X4 - X2 + 1.
A quicker way of computing the cyclotomic polynomials is given in Exercise
9.4.
9.5 The Galois Group over Q of an Extension of Q

by a Root of Unity
PROPOSITION. - Let n ~ 2 be an integer and ( a primitive n-th root of
unity in C. Then
1) IQ[(] is a normal extension of (;
2) [IQ[(] : IQ] = cp(n);
3) Gal(IQ[(]IIQ) ~ U(n); in particular, this group is abelian.

PROOF. -
1) Indeed, IQ[(] is the splitting field of Xn - 1, or of tP n , over IQ.
2) The minimal polynomial of ( over IQ is n, which is of degree cp(n).
3) Set G = Gal(IQ[(]IIQ) and let a E G; then a is determined by a(O

which is a conjugate of ( and therefore a primitive n-th root of unity.
Thus, it is of the form (k for some k prime to n. We use this k to
construct a map 'I/; : G ---> U(n), defining it by 'I/;(a) = k.
If a'(() = (k', we have (aoa')(() = a((k') = (kk', so that 'I/;(aoa') =
'I/;(a) . 'I/; (a') , which proves that 'I/; is a group homomorphism.
Now, 'I/; is injective, because if 'I/; (a) = 1 then a = id. Furthermore,
we have
IGI = (IQ((] : IQ] = cp(n) = IU(n)l,
so 'I/; is an isomorphism.
COMMENTARY. - One of the most beautiful theorems of Galois theory,

the Kronecker-Weber theorem, states that every normal extension N of
finite degree of Q whose Galois group is abelian (such extensions are called
abelian extensions, cf. §10.1), is contained in a cyclotomic extension, i.e.
an extension generated by a root of unity (. In Exercise 9.9, we study a
particular case of this theorem, namely, the case of quadratic extensions.
The reader can find a proof of the general theorem in the last chapters of
the book by Paulo Ribenboim listed in the bibliography. The study of the
abelian extensions of Q is the object of what is known as class field theory.

Exercise 9.1. Roots of unity
1) Complex numbers: are the following numbers roots of unity in C?
3V6+7iV5
a)
17
7 +4iJ2
b)
9
c)
..)2 + v'3 + i..)2 - v'3
2
2) Let n be an integer, and let ( be a primitive n-th root of unity. Let

K be an intermediate field between Q and Q[(]. Show that K is a
normal extension of Q.
3) Let m and n be two relatively prime integers. Let ( be a primitive

m-th root of unity and 'fJ a primitive n-th root of unity. Show that
Q[(] n Q['fJ] = Q.
4) Let K be a field contained in C, n ~ 2 an integer and ( a primitive

n-th root of unity in C. Show that Gal(K[(]IK) is isomorphic to a
subgroup of U(n).
Exercise 9.2. Algebraic numbers of modulus 1 that are not roots

of unity
Set P(X) = (X2 + X + 1)2 - 2X2, and let t and u denote the real
roots of P, and v and w the non-real roots of P.
1) Show that P is irreducible over Q.

2) Show that v and ware algebraic over Q and of modulus 1, whereas t

and u are algebraic over Q but not of modulus 1.
3) Deduce that v and w are not roots of unity.
Exercise 9.3. The Mobius function

1) Check, by computing the terms, that L p(d) = o.
dED(360)
2) a) Show that n = L cp(d).

dED(n)
b) Deduce that cp(n) = L dp (J).

dED(n)
Exercise 9.4. Computation of cyclotomic polynomials

1) For each n :s; 12, decompose the polynomial Xn - 1 as a product of
irreducible factors in Z[X] and in Q[X].
2) Using the formula of Proposition 9.4.2 2), determine the cyclotomic

polynomials <.I>n(X) for n = 30, n = 81.
3) In this question, we propose a rapid way of computing <.I>n(X) for

n>1.
a) Show that <.I>n(X) = X'P(n) <.I>n (1/ X). Deduce that the coeffi-
cients of <.I>n satisfy a'P(n)-k = ak for 0 :s; k:S; cp(n).
b) Show that <.I>n (X) is determined by its value modulo X['P(n)/2]+1,

where the square brackets indicate the integral part, and that
this computation can be done in the ring A = Z[X]/(X['P(n)/2]+l).
4) Prove the formulas:

<.I>n(X) = <.I>m(xn/m) whenever m is the product of the distinct
prime factors dividing n;
<.I>pn(X) = <.I>n(XP)/<.I>n(X) whenever p is a prime not dividing n;
<.I>2n(X) = <.I>n( -X) whenever n is an odd integer> 1.
5) Now, use these results to compute the polynomials <.I>n(X) for n = 30,
81,105.
Exercise 9.5. Fifth roots of unity

Set (= e2i7r / 5.
1) Solve the equation x4 + x 3 + x 2 + X + 1 = 0 by using the change of

variables y = x + (1/ x). Deduce algebraic expressions for (k, 1 ::::; k ::::;
4.
2) Without using the results of Chapter g, define an isomorphism of the
group G = Gal(IQ[(]IIQ) onto a well-known group. Show that G has a
non-trivial subgroup H of order 2 and index 2.
3) Find an element generating the field I(H), which (we recall) is the
set of elements in IQ[(] invariant under H.
Exercise 9.6. Fifteenth roots of unity

Set (= e2i7r/15 and 'TJ = e2i7r / 5, j = e2i7r / 3 . Recall that
27r
cos -
-1
= ---,--
+ v'5
5 4
1) What is the degree of IQ[(] over IQ?

Compute the minimal polynomial of ( over IQ.
Give the decomposition of X I5 -1 as a product of irreducible factors
in IQ[X].
Set G = Gal(IQ[(]IIQ), and let CTk denote the element of the group G
such that CTk(() = (k with 1 ::::; k ::::; 14.
2) What are the possible values for k?
3) a) Show that the Galois group G is isomorphic to a product P of

two cyclic groups, and construct this isomorphism.
b) What are the orders of the elements of P?
4) a) Show that 1Q[J5] C IQ['TJ].
b) Show that IQ[(] is an extension of degree 2 of IQ [cos(27r/15)].
5) a) Show that lQ[j], 1Q['TJl, IQ [v'5] , IQ [j, J5] are extensions of IQ

contained in IQ[(].
b) For each of the four fields K listed above, determine the sub-
group Gal(IQ[(]IK) of G (give the elements CTk of each of these
groups).
6) a) Determine the field of invariants of the subgroup of G generated

by a 14.
b) Solve the same problem for the subgroup of G generated by a2

(show first that the desired field is a subfield of Q [j, Y5]).
7) Give the trellis of subgroups of P and the trellis of extensions of Q
contained in Q[(]' by making the Galois correspondence completely
explicit.
8) Compute an algebraic expression for cos(27r /15) by using the group
9) Using the problems above, decide if the regular polygon with 15 sides
(and radius 1) is constructible with ruler and compass.
Exercise 9.7. Seventeenth roots of unity

Set ( = e2i7r/17 and G = Gal(Q[(lIQ). Let ai be the Q-automorphism
of Q[(] defined by ak(() = (k for k = 1, ... ,16. Note that in this
exercise, as in the preceding one, the procedure we use is not entirely
algebraic: it is necessary to consider orders of elements to distinguish
the roots.
1) Find the smallest integer n such that the class of n modulo 17 gener-
ates (1::/171::)*. Use this to determine an isomorphism from G to the
group (1::/161::, +), and then the trellis of subgroups of G.
2) For every intermediate extension K between Q[(] and Q, give an

element ak of G such that K = I( (ak), the set of elements invariant
under (ak).
3) In terms of (, determine an x such that Q[x] is a quadratic extension

of degree 2 of Q contained in Q[(]. Find the minimal polynomial of
x over Q. Deduce the value of x.
4) In terms of (, determine y such that Q[x,y] is an extension of degree

2 of Q[x] contained in Q[(]. Find the minimal polynomial of y over
Q[x], and use it to deduce the value of y.
5) In terms of (, determine z such that Q[x, y, z] is a quadratic extension

of Q[x, y] contained in Q[(]. Find the minimal polynomial of z over
Q[x, y], and use it to deduce the value of z and of cos(27r /17).
6) Deduce the value ( from the previous computations.
7) Is the regular polygon of radius 1 with 17 sides constructible with

ruler and compass?
Exercise 9.8. The structure of U(n)

Let n be an integer > 1.
1) a) Suppose that p is a prime. What is the structure of U(p)?

b) Suppose n = rs with rand s relatively prime. Show that we
have U(n) ~ U(r) x U(s) and rp(n) = rp(r)rp(s).
c) Determine the structure of U(n) for n = 2, 4, 6, 8, 9, 12, 15.
2) Assume that n is of the form pk, where p is an odd prime and k is an

integer> 1.
a) Let a be an integer whose class modulo p generates (71/p71r.

Set b = a pk - 1 • Show that the class of b modulo pk is of order
p -1 in U(n).
b) Show that for every r ~ 0, we have (l+p)pr = l+p r +l mod pr+2.

c) Deduce that 1 + P is of order pk-l in U(n), and then that U(n)
is cyclic.
d) Determine a generator of U(25), and a generator of U(125).
3) Assume that n is of the form 2k.
a) Show that for every r ~ 0, we have 52r = 1 + 2r+ 2 mod 2r +3 .

b) Deduce that the class of 5 is of order 2k - 2 in U (n).
c) Show that, for k ~ 2, a power of 5 is never equal to -1 modulo
n.
d) Deduce that for every k ~ 2, U(n) is isomorphic to 7l/2 k - 2 71 x

7l/271.
e) Determine the order of 3 in U (n).
4) a) What is the structure of U(200)?

b) Show that the group U ( n) is cyclic if and only if n is of the form
pk or 2pk, where p is an odd prime (decompose n as a product
of prime factors).
Exercise 9.9. Quadratic and cyclotomic extensions

For every integer n ::::: 2, set (n = e 2i7r In. We propose to prove the
following proposition: for every prime p in N, there exists n in N such
that v'P E Q[(nl·
1) Show the result for p =2 (take n = 8).

From now on, we assume that p > 2. Set P(X) XP -1,
S(X) = P(X)/(X -1), and ( = (p.
2) Show that the discriminant D(S) is the square of an element d of
Q[(J, and that D(S) is rational.
3) a) Show that S'(x) = (pxP-1)/(x - 1) for every root x of S.

b) Deduce that D(S) = (_1)(p-l l/ 2pp-2.
4) a) Deduce from the two preceding questions that v'Pld or v'Plid
is rational, by arguing according to the values of p modulo 4.
b) Show that Q[i, (nl = Q[(4nl for every odd integer n.
c) Prove the proposition.
5) Show that for every NEZ, there exists an integer n such that v'ii E
Q[(n].
Exercise 9.10. Factorization of <Pp over a quadratic extension of

Q
This exercise uses the results of Exercise 9.9.
Let p be an odd prime. Set ( = e2i7r/p, N = Q[(J, and G = Gal(N[Q) ~
U(p). Let <pp(X) denote the minimal polynomial of ( over Q, and let
ak be the element of G defined by ak(() = (k. Let a be a generator
of U(p).
Let L denote either the field Q [.JP] if p = 1 mod 4 or the field Q [i.JP]
if P = 3 mod 4.
Recall that for every divisor d of n, a cyclic group of order n has a
unique subgroup of order d.
1) Show that LeN.
2) What is the set of quadratic extensions of Q contained in N?
3) What is the order of G' = Gal(N[L)? What are its elements, in terms
of a?
4) Using 3), show that q}p factors as a product of two irreducible poly-
nomials over L[X]. Denote them by Sand T; give their roots in N
and their constant terms.
5) Compute the expression of Sand Tin L[X] by using the preceding

questions:
a) for p = 7,
b) for p = 13.
1) a) The modulus is not 1.
b) The modulus is 1, but the minimal polynomial over IQl is X2 -

C14j9)X + 1. As it does not have integral coefficients, the number we are
considering cannot be a root of unity in C, since the minimal polynomials
of roots of unity are the cyclotomic polynomials, and we know that they
have integral coefficients.
c) The square of the number is e i7r /6, and its real part is strictly positive,
so it must be e i7r / 12 .
2) 1Ql[(] is a normal extension of IQl with Galois group U(n). K is the field
of invariants of a subgroup H of U(n). As U(n) is abelian, H is a normal
subgroup of it, so K is a normal extension of 1Ql.
3) Let w be a primitive mn-th root of unity. The degree of lQl[w] over IQl
is <p(mn) = <p(m)<p(n) (this equality holds because m and n are relatively
prime).
As Q[(] is of degree <p(m) over 1Ql, and 1Ql[7]] is of degree <pen) over Q,
the tower rule shows that lQl[w] is of degree <pCm) over 1Ql[7]] and <pCn) over
Q[(]. Set L = Q[(] nQ[7]]' r = [Q[(] : L]. As L is an intermediate extension
between Q and 1Ql[(], we have r :s; <p(m) (Figure 9.2). Furthermore, the
degree of ( over L is greater than or equal to the degree of ( over 1Ql[7]]; as
lQl[w] = 1Ql[7]][(]' we have r 2: <p(m) and thus the result.
~Q[ro)~(m)
Q[~] 02[11 ]
~11
/K[~\
Q[~] K
Q ""Q/
FIGURE 9.2. FIGURE 9.3.
We can also use the isomorphism U(mn) ':::' U(m) x U(n); the intermedi-
ate extensions Q[(] and Q[1]] corresponding to the subgroups U(m) x {id}
and {id} x U(n) of U(m) x U(n) have intersection equal to Q (see Exercise
8.6).
4) Because K[(] is the splitting field of xn -lover K (Figure 9.3), it is a

normal extension of K. As K[(] ~ Q[(] and Q[(] is a normal extension of Q,
every element of Gal(K[(]IK) induces an element of Gal(Q[(]IQ) ':::' U(n)
by restriction. This restriction is a group homorphism, which is injective
because if a E Gal(K[(]IK) induces the identiy on Q[(], then a(() = (, so
a = id.
1) In the field Q [J2] [X], we have
P(X) = (X2 + (1 - J2)X + 1)(X2 + (1 + J2)X + 1).
The roots of P are given by
-1-J2±J2y2-1
t,u
2
y2 - 1 ± i y''-2y2-=-2-+-1
v,W = 2
Note first that P has no linear factor in Q[X]. Next, note that a decom-
position into two quadratic factors is impossible in Q[Xl, since one of the
factors must be X2 + (1 + J2)X + 1.
2) The computations are not difficult.
3) The conjugates of roots of unity over Q are also roots of unity; as t and
U do not have modulus 1, their conjugates v and w over Q are not roots of
unity.
1) J-L(d) is not zero when d is squarefree, so we have
L J-L(d) = J-L(1) + J-L(2) + J-L(3) + J-L(5) + J-L(6) + J-L(10) + J-L(15) + J-L(30)

dED(360)
= 1 - 1 - 1 - 1 + 1 + 1 + 1 - 1 = O.
2) a) In 'Lln'L, we know that the order of every element divides n. For

a divisor d of n, let Ed denote the set of elements of order d. The family
(Ed)dED(n) is a partition of'Lln'L, and we know that Ed has r.p(d) elements.
This gives the result.
b) To obtain the results, it suffices to apply the Mobius inversion formula
with f = id and g = r.p.
1) We can use the formula Xn -1 = II

<Pd(X), or simply make a direct
dED(n)
computation. Up to invertible factors, the decompositions are the same in
'L[X] and in Q[X]:
X2 -1 (X - l)(X + 1),
X 3 -1 (X_1)(X2+X+1),
X4-1 (X - l)(X + 1)(X2 + 1),
X 5 -1 (X - 1)<P5(X),
X 6 -1 (X - 1)(X2 + X + l)(X + 1)(X2 - X + 1),
X7 -1 (X - 1)<P 7(X),
X8 -1 (X - l)(X + 1)(X2 + 1)(X4 + 1),
X9 -1 (X 3 -1)(X 6 + X 3 + 1),
X lO -1 (X - 1)<P5(X)(X + 1)<P5( -X),
Xll -1 (X -l)<Pll(X),
XI2 -1 (X6 - 1)(X2 + 1)(X4 - X2 + 1),
(X -1)(X2 + X + l)(X + 1) .
(X2 - X + 1)(X2 + 1)(X4 - X2 + 1)
2) We have
(X30 - 1)(X5 - 1)(X 3 - 1)(X2 - 1)
[t]<P30(X)
(XI5 - l)(XIO - 1)(X6 - l)(X - 1)
(XI5 + l)(X + 1)
(X5 + 1)(X3 + 1)
X 10 _ X5 +1
X2 - X +1
X 8 + X7 - X 5 _ X4 - X 3 + X + 1,
X81 -1
X27 -1
X 54 + X27 + 1.
3) a) Set S = L J-l(njd). We know that S = 0 for n > 1. Furthermore,

dED(n)
by the preceding exercise, cp(n) = L dJ-l(njd). Consequently, we have
dED(n)
x",(n) II (X- d - l)J.t(n/d)

dED(n)
(_1)8 II (Xd _l)J.t(n/d)
dED(n)
<pn(X).
If <Pn (X) = L akXk, the preceding equality shows that ak = a",(n)-k'

O:-:;k:-:;",(n)
cp(n)
b) It suffices to compute the terms of <pn(X) of degree::::; -2-' The
polynomials of the form 1 - Xk are invertible in A, since the Xk are nilpo-
tent.
4) We have
<pn(X) II (X n / d - l)J.t(d)
dED(n)
II
dED(m)
((xn/m)m/d _ l)J.t(d)
<pm(x n / m ),
<ppn(X) II (Xpn/d - l)J.t(d)

II (Xpn/d _ 1 )J.t(d)
dED(n) dED(pn)-D(n)
II ((xp)n/d - l)J.t(d) II ((X n / d _l)-!L(d)
dED(n) dED(n)
<pn(XP)
<pn(X) ,
<pn(X2)
<P2n(X)
<pn(X)
II (X 2n / d _l)Jl(d)(X n/ d - l)-Jl(d)
dED(n)
II (X n / d + l)Jl(d)
dED(n)
(-1) L /L(d) II (X n / d _l)Jl(d)

dED(n) dED(n)
since n/d is odd and L /L(d) = 0 for n > 1.

dED(n)
5) To compute 30, we can reason modulo X 5, since '1'(30) = 8. Modulo

X 5 , we have
(X30 - 1)(X 5 - 1)(X 3 - 1)(X2 - 1)
(X15 - 1)(XI0 - 1)(X6 - l)(X - 1)
(-1)( -1)(X3 - 1)(X2 - 1)
(-1)( -1)( -l)(X -1)
_(X3 - l)(X + 1)
- X4 - X 3 + X + 1,
which gives the result.
To compute 81, we can write 81(X) = 3(X27). Because '1'(81)/2 =

27, we can compute modulo X 28 : 81(X) = (X 81 _1)/(X 27 -1) = X27 +1,
etc.
To compute 105, we compute modulo X 25 since '1'(105) = 48. Modulo

X 25 , we have
(X 105 _ 1)(X5 - 1)(X3 - 1)(X7 - 1)

(X15 - 1)(X21 - 1)(X35 - l)(X - 1)
(X 5 _ 1)(X2 + X + 1)(X7 - 1)(1 + X I5 )(1 + X21)
_X24 _ X22 _ X 20 + X17 + X 16 + X 15 + X14 + X 13 +
X12 _ X 9 _ X 8 _ 2X7 - X 6 _ X 5 + X2 + X + 1,
so
X 48 + X47 + X 46 _ X 45 _ X44 _ 2X 43 _ X42 _ X41 +
X 38 + X 37 + X 36 + X 35 + X 34 + X 33 _ X 30 _
X 28 _ X 26 _ X24 _ X22 _ X 20 + X17 + X 16 +
X 15 + X14 + X 13 + X12 _ X 9 _ X 8 _ 2X7 _
- X 6 - X 5 + X2 + X + 1.
It so happens that the coefficients of <Pn with n < 105 all lie in the set
{ -1, 0, I}; however, this result shows that this is not always the case.
1) The equation becomes y2 + y - 1, so we have y = (-1 ± F5)/2.

For the two values y, the equations x2 - xy + 1 = 0 give
-1+V5 .V1O+2V5 -1-V5 .V1O-2V5

X= ±z---- or 4 ±z 4 .
4 4
We need an additional argument, using the ordering on JR, to be able to
deduce the values of the (k for 1 ::; k ::; 4 from these four numbers.
For this, we consider the values of the real parts 2 cos(27r /5) = (-1 +
F5)/2 and 2cos(47r/5) = (-1 - F5)/2 and the values of the imaginary
parts. We find that
-1+V5 .V1O+2V5
( 4 + z 4 '
(2 -1-V5 .V1O-2V5
4 +z 4 '
(3 -1- V5 . V10 - 2V5

-z
4 4
(4 -1+V5 . V1O+ 2V5
-z
4 4
2) As [Q[(] : Q] = 4, G has four elements. They are determined by the

image of (. Let (j denote the Q-homorphism defined by (j(() = (2. As
(j2(() = (4, (j3(() = (3, and (j4(() = (, we have (j4 = id, (j is of order 4 in
G, and G is isomorphic to Z/4Z. This last group has only one non-trivial
subgroup, generated by the class of 2, which corresponds to the subgroup
H of G generated by (j2.
3) As F5 = 2(( + (4) + 1, we see that Q[F5] must be the extension

corresponding to H.
1) We know that [Q[(] : Q] = '1'(15), where 'I' denotes the Euler function.
As '1'(15) = '1'(3)'1'(5) = 8, Q[(] is an extension of degree 8 of Q.
The minimal polynomial <P of ( over Q is the cyclotomic polynomial
<P15(X). The formulas of Proposition 9.4.2 give
(X15 - l)(X - 1)
(X5 - 1)(X3 - 1)
XS - X7 +X 5 - X4 +X3- X + 1,
<P15(X)<P5 (X)<P3(X)<P1 (X)
(X s _ X7 + X 5 - X4 + X 3 - X + 1) .
(X4 + X 3 + X2 + X + 1)(X2 + X + I)(X -1).
2) The image (k of ( under ak is a primitive 15th root of unity, so k is

prime to 15 ; thus, the possible values of k are the eight numbers 1, 2, 4,
7, 8, 11, 13, and 14.
3) a) We know that G is isomorphic to the group U(15) of invertible

elements of the ring ::£/15::£. This last group is isomorphic to U(3) x U(5),
so to P = ::£/2::£ x ::£/4::£. The isomorphism is thus the composition
G ~ U(15) = (::£/15::£)* ~ (::£/3::£)* x (::£/5::£)* J!.. ::£/2::£ x ::£/a,

where <p(a) = k such that a«() = (k, X(k) = (k mod 3,k mod 5) and 'ljJ is
the isomorphism defined by choosing the classes of 2 modulo 3 and modulo
5 as generators of (::£/3::£)* and (::£/5::£)* . This is made explicit in Table 9.1.
a a1 a2 a4 a7 as all a13 a14

<p(a) 1 2 4 7 8 11 13 14
X(<p(a)) (1,1) (2,2) (1,4) (1,2) (2,3) (2,1) (1,3) (2,4)
'ljJ(X(<p(a))) (0,0) (1,1) (0,2) (0,1) (1,3) (1,0) (0,3) (1,2)
TABLE 9.1.
b) Here, we let 2 and 4 denote the classes modulo 2 and 4; P has three
elements of order two, namely (0,2), (1,0), (1,2), and four elements of
order 4, namely (0,1), (1,1), (0,3) (1,3).
4) a) As cos(27r/5) = (-1 + v's)/4 = (1] + 1]-1)/2, we have Q[v's] c Q[1]].

b) As 2 cos(27r/15) = (+ e
1 , we see that Q[cos(27r/15)] C Q[(]. This
inclusion is strict since ( is not real; as (2 - 2( cos(27r /15) + 1 = 0, Q[(] is

a quadratic extension of Q[cos(27r/15)].
5) a) As j = (5 and 1] = (3, Q[j] and Q[1]] are extensions of Q contained

in Q[(]. We know that Q[cos(27r/5)] C Q[1]]' which shows that Q[v's] is
contained in Q[(]. The preceding inclusions show that Q[j, v's] C Q[(].
b) K = Q[j]. We have ak E Gal(Q[(]IQ[j]) if and only if ak(j) = j; as
j = (5, this means that 5k = 5 mod 15 or k = 1 mod 3, so k = 1, 4,7, 13.
The Galois group is cyclic, generated by a7.
K = <01[1]]. We have ak E Gal(Q[(]IQ[1]]) if and only if ak(1]) = 1]. As
1] = (3, this means that 3k = 3 mod 15 or k = 1 mod 5, so k = 1,11.
K = Q[v5]. As Q[v5] = Q[1] + 1]-1], CTk E Gal (Q[(]IQ[v's]) if and only

if CTk( 1] + 1]-1) = 1] + 1]-1. This means that
27r 27r
cosk- =cos- i.e. k = ±1 mod 5,
5 5 '
so k = 1, 4, 11, 14. The Galois group is not cyclic; it has three elements of
order 2.
K = Q[j, v5]. As Gal (Q[(]IQ[j, v5]) is the intersection of the two
groups Gal(Q[(]IQ[j]) and Gal(Q[(]IQ[v5], it contains CTk for k = 1,4.
At this stage of the problem, we can begin to construct the trellises.
6) a) Set K = I( (CT14))' As CT14 is of order 2, Q[(] is a quadratic extension

of K. We see that cos(27r/15) = (( + C 1)/2 is invariant under CT14, so we
have Q[cos(27r/15)] C K. Then, by 4 b) of this exercise, we can conclude
that K = Q[cos(27r /15)].
b) Set L = I((CT2))' As CT2 is of order 4, Q[(] is an extension of degree 4

of L, so L is an extension of degree 2 of Q. As (CT2)2 = CT4, L is a subfield
of I((CT4)) = Q[j, v5]. Note that v5 does not lie in L, since CT2 is not
in the group Gal(Q[(]IQ[v5]). Thus, by the preceding question, we have
CT2( v5) = -v5; furthermore, CT2(j) = j2 = -1 - j. It follows that those
elements a + bj + cv5 + djv5 of Q[j, v5] which are invariant under CT2 are
of the form a + cv5(1 + 2j), i.e. a + ciM. This gives L = Q[iv'15].
7) The trellis of subgroups of P contains three subgroups with two ele-

ments, generated by the elements of order 2 of P, and three subgroups with
four elements: two generated by elements of order 4 of P (each containing
two elements of order 4 and the element (0,2)) and the last one containing
the three elements of order 2 of P, isomorphic to 7L/27L x 7L/27L.
Questions 5 and 6 contained several hints for the construction of the
trellis corresponding to the extensions of Q contained in Q[(] (Figure 9.4).
«0,0»
/l~
«0,2» <(1,0» <(1,2»
/~ ~ i /
<(1,1» «0,1» «1,0),(0,2»
~~/
p Q[l;;] "'-
/ 1 'x
Q[j,VS] Q [11] Q[cos 21t 114]
/~~~
Qi,Vf5] Q[j] Q [1/5]
~+/
Q
FIGURE 9.4. The Galois correspondence
8) The Galois group Gal (Ql[cos(27r/15)lI Ql[.J5]) is of order 2, generated

by the restriction of (/4. Thus, cos(27r 115) is a root of the polynomial
whose coefficients lie in Ql [J5] .

Because 15«() = 0 implies that (4 - (3 + (1 -1 + C 1 - C 3 + C 4 , we
have
(+ (-1 + (4 + (-4
2
C + 1 +(-3
2
27r 1 1 +.J5
cos-+-=---
5 2 4'
« + (-1 )«4 + (-4)

4
(3 + (-3 + (5 + (-5
4
~(cos 27r + cos 27r) = -3 +.J5.
253 8
It follows that cos(27r/15) = (1+Y'S+V30 - 6Y'S)/8, since cos(27r/15) > o.

Thus,
0"4 (cos 27r) = (4 + (-4 = cos 87r = 1 + V5 - V30 - 6V5.

15 2 15 8
We could find the same result more rapidly by using
cos(27r /15) = cos (( 7r /3) - (7r /5)) ...

but that method does not employ the beautiful results of Chapter 9!
9) The regular polygon with 15 sides and radius 1 is constructible with

ruler and compass. As in the preceding problem, we can note that the
regular polygons with 6 and 10 sides are constructible, and that (27r /15) =
(7r/3) - (7r/5). We can also use Proposition 5.7, and then the preceding
problems show that lQ[cos(27r /15)] has an increasing sequence of subfields
IQ IQ[Y'S]' lQ[cos(27r/15)], each of degree 2 over the preceding one (this is
exactly what is expressed by the formula for cos(27r/15), which only uses
extractions of square roots).
Recall that IQ[(] is an extension of degree <p(17) = 16 of IQ, which has basis
((khSk96 as a IQ-vector space.
1) We find 28 = 1 mod 17; 3 is of order 16 and works. The isomorphism

of (2/162, +) onto G is the composition
2/162 ~ U(17) J:.. G,

with <p(r) = 3T and 'IjJ(k) = O"k. It is given in Table 9.2.
0 1 2 3 4 5 6 7
id 0"3 0"9 0"10 0"13 0"5 0"15 O"ll
8 9 10 11 12 13 14 15
0"16 0"14 0"8 0"7 0"4 0"12 0"2 0"6
TABLE 9.2.
Since the trellis of subgroups of (2/162, +) is given by
{O} C (8}=(82/162,+) C (4) = (42/162,+)

C (2) = (22/162, +) C (2/162, +) ,
and we have 0"9 = (0"3)2,0"13 = (0"3)4, 0"16 = (0"3)8, we find that the trellis
of subgroups of G is given by
{id} C (0"16) C (0"13) C (0"9) C (0"3) =G

(we can also note that (0"9) = (0"2), (0"13) = (0"4)).
2) By the Galois correspondence, we can give the trellis of extensions of IQ

contained in IQ[(] :
1({id}) = IQ[(] :> 1((0"16)):> 1((O"d):> 1((0"9)):> 1((0"3)) = I(G) = IQ.

3) As (0"9) is a subgroup of index 2 of (0"3) = G, 1((0"9)) is a quadratic
extension oflQ. The elements of 1((0"9)) are the U E IQ[(] such that 0"9(U) =
u. As u is of the form L
ak(k, 0"9(U) = u implies that the coefficients
l:S;k96
ak of index 1, 9, 13, 15, 16, 8, 4, 2 are equal, as well as the coefficients ak
of index 3, 10, 5, 11, 14, 7, 12, 6.
Set x = ( + (9 + (13 + (15 + (16 + (8 + (4 + (2 (we could have set
x = (3 + (10 + (5 + (11 + (14 + C + (12 + (6). We have x E 1((0"9)) and
x tf- IQ since IQ = I ( (0"3)) and 0"3 (x) =I- x. Thus, x is of degree 2 over IQ
and 1( (0"9)) = lQ[x]. As Gal(lQ[x]llQ) = (0"3) / (0"9) is a group of order 2, the
conjugate of x over IQ is 0"3(X) since 0"3 tf- (0"9). The minimal polynomial
of x over IQ is X2 - (x + 0"3(X))X + X0"3(X). We see that x + 0"3(X) =
L (k = -1, since L (k = 0. The computation of X0"3(X) is longer:
l:S;k:S;16 0:S;k:S;16
the expansion of the product has 64 terms. After regrouping them, using the
equality L(k = 0, we find X0"3(X) = -4; x is a root of the polynomial
0:S;k96
X2 + X - 4, so x = (-1 ± v'l7)/2. Note that lQ[x] = lQ[vTI].
We now need to use an order argument to tell which of these two val-
ues is actually x. As x = 2cos(2n/I7) + 2cos(4n/I7) + 2cos(8n/I7) +
2cos(I6n/I7), we have x > (0,5 + 0,5 - 1) > 0, so x = (-1 + v'l7)/2,
0"3(X) = (-1 - v'l7)/2.
4) We use the same procedure as in the preceding question.
The extension I ( (0"13)) is a quadratic extension of degree 2 of I ( (0"9)) =
lQ[x] since (0"13) is a subgroup of index 2 of (0"9). It is generated by any
element y E 1((O"d) not in lQ[x].
The search for u = L ak(k satisfying 0"13(U) = u leads to setting
1<k<16
y = (+ (4 + C 4 + C 1 ,fo; example.
We have y E 1( (O"d) and y tf-1Q[x] since 0"9(Y) =I- y. Thus, y is of degree
2 over lQ[x] and 1((0"13)) = lQ[x,y]. As Gal(lQ[x,y]llQ[x]) = (0"9)/(0"13), the
conjugate of y over lQ[x] is 0"9(X) since 0"9 tf- (0"13). The minimal polynomial
of y over lQ[x] is X2 - (y + 0"9 (y))X + Y0"9(Y).
174 9. RootsofUnity
We see that y + ag(y) = x and yag(y) = -1, so y is a root of the

polynomial X2 - xX - 1. Thus, y = (x ± "';x 2 + 4)/2. We have Q[x, y] =
Q[m, v'34 - 2m] since x 2 + 4 = 8 - x = (34 - 2V17)/4.
As y = ( + (4 + (13 + (16 = 2 cos(27r /17) + 2 cos(87r /17) > 0, we have
y = (x + "';x 2 + 4)/2 and ag(y)y = (x - v'x2 + 4)/2.
5) Still using the same procedure, we determine the extension I( (a16)).
The search for the u = L
ak(k satisfying a16 (u) = u leads to setting
1S:k96
z = (+ C 1, for example, checking that z f/- Q[x, y] since a13(z) =I- z. The
minimal polynomial of z over Q[x, y] is X 2 -(z+a13(z))X +za13(z). We see
that z+a13(z) = y. Furthermore, za13(Z) = (3+(5+(12+(14 is an element
of Q[x, y], which we need to express in the basis (1, x, y, xy). Computing
xy in terms of (, we find xy = 3+(2 + (8 +(g +(15 +2((3 +(5 +(12 +(14),
so za13(Z) = (xy - x + y - 3)/2. Denote this quantity by a; then z is a
root of the polynomial X2 - yX + a, so z = (y ± v' y2 - 4a) /2 = (y ±
"';2x - 2y - xy + 7)/2.
As z = 2cos(27r/17) > a13(z) = 2 cos(87r/17) , we have
y - "';2x - 2y - xy + 7
z = y + "';2x - 2y - xy + 7
-=---'-----::-=--~- adz) = 2 .
2 '
After completing the computations, we obtain
27r Z
cos-=-=
17 2
-1 + m + v'34 - 2m + J68 + 12m - 4v'34 - 2m - 8v'34 + 2y'1 i

16.
This expression can be found, in varying forms, in the literature. The cosine
of the other multiples of 27r /17 are obtained with analogous computations.
The formulas that develop are analogues of the preceding ones, up to some
sign changes.
6) Finally, ( = cos(27r /17) + i sin(27r /17) = (z + iJ 4 - z2)/2.

All of these results are due to Gauss (Figure 9.5).
CG.I~;::~""+ir\b7+nV'~:-:aV'11)""'lV{{1'1+SVIVJ""'V@+-.V11)
-eVti4+U/l7)} ;
Jes cosinua des al1altip'les de eet 8JJgle ()nt 'one forme semJ.Iable,
lea uus oat DR radical ., plus. n,. a certain.meat bien mil
-de s'etoBller Ilue la. div.isibilitcl du oe~le til 1) . , S puties ayaat
et~ CODDue des Ie temps d'Eucl/dtJ, on Il'ait rim ajotlM a eel
d~couvettes daJls DR intervalle de deux mille ans, ef· '!\10" tous les
g~metres aient lnoonce comme certain, qU'exeepti cos divisions
et ceRa qui s'ea d«!duisent (les divisions en 2(A., 15, 5.2'f, 5.21'- ,
lS.2(A. parties), on De ~Ul'ait en eJfeetUCf auanne par des
comtr)1.ctlDns .,~o~triques.
FIGURE 9.5. Gauss: Recherches arithmetiques (value of cos~;)
1) As (s = e i7r / 4 = (1 + i)/V2, we have V2 = (1 + (~)/(s so V2 E Q[(sj.

2) The roots of S are of the form (i with 1 :::: i :::: p - 1, so D(S) =
II ((i - (j)2 is the square of d = II
((i - (j), which is ob-
l:Si<j:Sp-1 l:Si<j:Sp-1
viously an element of Q[(j. We know that the discriminant of S lies in the
ring generated by its coefficients; thus it is an integer.
3) a) Every root x of S is a root of P different from 1. We have P'(X) =

pXp-l and P'(X) = S'(X)(X - 1) + SeX), so S'(x) = P'(x)/(x - 1) =
pxP-I/(x - 1).
b) D(S) = (_1)(p-l)(p-2)/2 II
S'((i), by §3.6. On the one hand,
l:Si:Sp-1
(_1)(p-l)(p-2)/2 = (( _1)p-2)(p-I)/2 = (_1)(p-I)/2
since p - 2 is odd and (p - 1) /2 is an integer.

On the other hand, II
S'((i) = II
p(i(p-I) /((i-I). In the nu-
l:Si:Sp-1 l:Si:Sp-1
merator of this expression, we have ( II (iY-\ the exponent of ( in
l:Si:Sp-1
the part between parentheses is L i = (p(p - 1))/2. As (P = 1, the
numerator is thus pp-I .
The denominator is given by
II ((i - 1) = (_l)p-1 II (1 - (i) = S(1) = p.

l:Si:Sp-1 l:Si:Sp-1
Thus d2 = D(S) = (_1)(P-l)/2pp-2.
4) a) If p = 4k + 1, then d 2 = pp-2, so p/d2 = l/PP-l = U/p2k)2 and

,jP/d = ±(1/p2k) is rational.
If p = 4k + 3, then d 2 = _pp-2, so p/d2 = _(l/PP-l) = _(1/p2k+l)2
and ,jP/id = ±(1/p2k+l) is rational.
b) First, i = ((4n)n and (n = ((4n)4 show that Q[i, (n] C Q[(4n]. Then,
as 4 and n are relatively prime, we see that i is of order 4 and (n is of order
n in C*, and i(n is of order lcm (4, n) = 4n in C*, which proves that i(n is
a primitive 4n-th root of unity, so we have shown the converse inclusion.
Finally, Q[i, (n] = Q[(4n].
c) If p = 1 mod 4, then part a) shows that ,jP/d E Q. As d E Q[(], we
have ,jP E Q[(] = Q[(p].
If p = 3 mod 4, then part a) shows that fo/id E Q. As d E Q[(p], we
have fo E Q[i, (p[= Q[(4p],
5) If N > 0, the decomposition of N as a product of primes shows that

IN is a product of roots of primes. Because each of these roots lies in a
cyclotomic field, their product also lies in a cyclotomic field (indeed, given
cyclotomic fields generated by m-th roots of unity mi with 1 ::; i ::; r, the cy-
clotomic field generated by the m-th root of unity m = lcm {mi; 1 ::; i ::; r}
will contain them alL This extends easily to negative integers N.
1) The inclusion is a consequence of Exercice 9.9.
2) L is a quadratic extension of Q contained in N; it is the only Oile since

there is only one subgroup of index 2 in the cyclic group C.
3) C' is the subgroup of index 2 of C. It has (p - 1)/2 elements, namely

the (jk for k = a 2r , 1::; r ::; (p - 1)/2.
4) The roots of S are the (jk (() for (jk <f. C'; those of T are the (jk (() for
(jk <f. C'. We note that T = (ja(S). The irreducibility of Sand T over L is
a consequence of the fact that ( is of degree (p - 1) /2 over L. The constant
term of Sis (_1)(p-l)/2(N with N = L a 2r = 0 mod p; thus it is
1$r$(p-l)/2
1 if p = 1 mod 4 and -1 if p = 3 mod 4.
5) a) If p = 7, then 3 is a generator of (7L/7'1l)* and its successive powers

are 3, 2, 6, 4, 5, 1. We have SeX) = (X - (2) (X - (4)(X - () = x 3 - f3X 2 -
(f3 + l)X - 1. We compute f3 = ( + (2 + (4 by noting that the conjugate of
f3 over Q[V7] is (ja(f3) = (3 + (5 + (6. As f3 + (ja(f3) = -1 and f3(ja(f3) = 2,
we see that f32 + f3 + 2 = 0, which gives f3 = (-1 + iV7)/2, taking into
account that 1m f3 > O.
We can determine T in the same way, and with (3' = (-1 - iv7)/2, we
find
<P7(X) = (X 3 - (3X 2 - «(3 + I)X - I)(X 3 - (3' X2 - «(3' + I)X - 1).
b) If p = 13, then 2 is a generator of (2/132)*, and we have
S(X) (X - (4)(X - (3)(X _ (12)(X _ (9)(X _ (lO)(X - ()

= X6 - I'X 5 + 2X4 - (1 + I')X 3 + 2X2 -I'X + 1.
The computation of I' = (2 + (5 + (6 + C + (8 + (11 is done as above: we
see that ,.,? + { - 3 = 0, which gives { = (-1 + v'i"3) /2, taking into account
that { > o.
We can determine T similarly, and now, with {' = (-1- Vi3)/2, we find
S(X) X 6 -I'X 5 + 2X4 - (1 + {)X 3 + 2X2 - {X + 1,

T(X) = X6 - {' X 5 + 2X4 - (1 + {')X 3 + 2X2 - {'X + 1.
10
Cyclic Extensions
After having studied extensions by roots of unity in Chapter 9, we now

proceed to study extensions by roots of arbitrary elements of the base
field, and consider in particular when such extensions have cyclic Galois
group.
10.1 Cyclic and Abelian Extensions

DEFINITION. - A normal extension N of a field K contained in C is said to
be cyclic if the Galois group Gal(NIK) is cyclic, and abelian if Gal(NIK)
is abelian.
EXAMPLE. - The extension of a subfield K of C by a root of unity is abelian

(see §9.5).
COMMENTARY. - Generally, we will use the term cyclic (resp. abelian)

extension of a field K for a Galois extension N of K whose Galois group
Gal(NIK) is cyclic (resp. abelian); for more details, see §15.6.
10.2 Extensions by a Root and Cyclic Extensions

PROPOSITION. - Let n ;::: 1 be an integer, ( a primitive n-th root of unity
in C, K a subfield of C containing (, and a an element of K. Let b be a
root of xn - a in C.

180 10. Cyclic Extensions
1) K[b] is a cyclic extension of K of degree a divisor d of n, and bd E K.
2) In particular, if xn - a is irreducible over K, then Gal(K[b]IK) is

cyclic of order n.
PROOF. -
°
1) The roots of Xn - a in C are the n numbers bC;k, ~ k ~ n - 1. As
they are all elements of K[b], K[b] is a normal extension of K. Set
°
G = Gal(K[b]IK). For every a of G, a(b) is a conjugate of b over K,
which can be written uniquely as bC;k avec ~ k ~ n - 1. The map
1j;: G ....... 7l,/n7l, defined by 1j;(a) = k is a group homomorphism, since
if a(b) = bC;k and a' (b) = bC;k' , then (a 0 a')(b) = a(bC;k') = bC;k+k'. It
is injective, since if 1j;(a) = 0, then a(b) = b, so a = id. It follows that
because G is isomorphic to a subgroup of the cyclic group 7l,/n7l" it
is itself cyclic of order a divisor d of n.
Let a be a generator of G and set a(b) = bC;k. As a is of order d,
= (bC;k)d = bdC;kd = bd. Because the
k is of order d in 7l,/n7l" so a(bd)
element bd is invariant under a, it is invariant under every element
G; thus it lies in K.
2) If xn - a is irreducible over K, [K[b] : K] = n so IGI = n. <>
EXAMPLE. - Consider the example of the field K = Q[j, v2], which contains
the sixth roots of unity. The polynomial X 3 - v2 is the minimal polynomial
of b = V"2 over K since [K[b) : K] . [K : Q[v2] = [K[b] : Q[bll·[Q[b] :
Q[v2] = 6 shows that b is of degree 3 over K. The group Gal(K[b]IK) is
thus isomorphic to 7l,/37l,; its elements send b to b, jb, j 2 b.
10.3 Irreducibility of XP - a
PROPOSITION. - Let p be a prime such that p ~ 2, K a subfield of C, C; a
primitive p-th root of unity, and a an element of K.
1) If K contains C;, the polynomial XP - a
• either is irreducible in K[X]

• or factors as a product of linear factors in K[X].
2) If K does not contain C;, the polynomial XP - a
• is either irreducible in K[X]

• or admits at least one root in K.
10.4 Hilbert's Theorem 90 181
PROOF. -
1) Proposition 2 shows that if b is a root of XP - a in C, then K[b] is a

cyclic extension of K of degree a divisor d of p. If d = p, then XP - a
is irreducible in K[X]; if d = 1, then XP - a factors as a product
of linear factors in K[X] (which are pairwise distinct since XP - a is
prime to its derivative in K[X], given that K has characteristic zero).
2) If XP -a is not irreducible in K[X], then XP -a = P(X)S(X), where

P and S are non-constant polynomials of K[X].
If b is a root of XP - a in C, we have XP - a = II (X - b(k) in

O::;k::;p-l
qX]. The constant term of P(X) is equal (up to sign) to a product
c of terms of the form b(k; thus c E K and c = bT(8, with 1 S; r < p.
We have cP = bTPC P = aT. By Bezout's identity, there exist integers
u and v such that ur+vp = 1; hence a = aUT +VP = cUPa vp = (ctiav)p.
As cUa v E K, we see that XP - a has a root in K. <>
10.4 Hilbert's Theorem 90

10.4.1 The Norm
DEFINITION. - Let K be a field, and let L be a normal extension contained
in C, of finite degree over K. Let G = Gal(LIK). For every a E L, we
define the norm of a over K to be the product NL/K(a) = II
a(a). When
aEG
there is no risk of confusion, we simply write N for the norm. As N(a) is
invariant under every K-automorphism of L, we have N(a) E K. If L is
the normal closure of K[a], N(a) is the product of the conjugates of a over
K. Finally, it is clear that N(ab) = N(a)N(b).
The norm depends on the extension L: if L' is a normal extension of
finite degree of K containing L, then NU/K(a) = (NL/K(a))[L':Lj for every
a E L.
The norm can be considered as a determinant (see Exercise 10.4).
EXAMPLES. - If K = IQ, then a, b E IQ, and we have:
1) if L = lQ[i], N(a + ib) = (a + ib)(a - ib) = a2 + b2 ;
2) if L = IQU], N(a + jb) = (a + jb)(a + j 2 b) = a2 - ab + b2 .
3) if L = IQ [v2] , N ( a + bv2) = (a + bv2) (a - bv2) = a 2 - 2b 2 •

10.4.2 Hilbert's Theorem 90

THEOREM. - Let K be a field, and let Lee be a cyclic extension of degree
n of K. Let G = Gal(LIK), and let a be a K -automorphism of L that
generates G. For every x of L, we have the equivalence
N(x) =1 {=} 3y -=I- 0 such that x = ary).
PROOF. - We have G = {a k ; 0 ::::: k ::::: n - I} , so if x = ary) , then

y an-1(y)
N(x) = II
O$k$n-l
k
a (x) = a( ) ... a n () =
y y
1
because an(y) = y.
Conversely, suppose that N(x) = 1. The linear combination
id + xa + ... + [xa(x) ... a n- 2 (x)] a n- 1

of K-automorphisms is not identically zero (Dedekind's theorem; see §6.6.3).
Let z be an element of L where it does not vanish; let y be its value at z.
As y = z + xa(z) + ... + xa(x) ... a n - 2 (x)a n- 1(z) -=I- 0, we have
xa(y) = xa(z) + '" + xa(x) ... an-1(x)an(z) = y- z + N(x)z = y,
which gives the result. <>
COMMENTARY. - This theorem appears in a text by Hilbert entitled Die
Theorie der algebraischen Zahlkorper, which first appeared in 1897, and
whose French translation appeared in 1909 under the title Theorie des
corps de nombres algebriques. The whole of the theory is remodeled and
restructured there, in 200 pages and 169 numbered theorems.
10.5 Extensions by a Root and Cyclic Extensions:

Converse
Let us now prove a converse of Proposition 10.2.
PROPOSITION. - Let n ~ 1 be an integer, ( a primitive n-th root of unity,

K a field containing (, and Lee a cyclic extension of K of degree nand
Galois group G with generator a. Under these conditions, there exist a E K
and bEL such that bn = a and L = K[bJ. In particular, G is the Galois
group of xn - a over K.
PROOF. - Let us consider the norm map N : L -> K. We have
10.6 Lagrange Resolvents 183
Hilbert's Theorem 90 implies that there exists bEL such that C 1 =

bjo-(b), so that o-(b) = be. We have o-k(b) = b(k for 0 :::; k :::; n - 1,
and the elements b(k are pairwise distinct conjugates of b. Consequently,
[K[b] : K] = n, hence L = K[b]. Moreover, o-(bn ) = (b()n = bn shows that
bn is invariant under 0-, so it is invariant under every element of G, which
proves that bn E K, by §8.5. <>
REMARK. - We can give a proof of this result that does not use Hilbert
90, but the eigenvalues of 0- as a K-linear map from L to L (see Exercise
10.3). Here, also, the result remains true for an arbitrary field of arbitrary
characteristic.
10.6 Lagrange Resolvents

10.6.1 Definition
Let n be an integer, n :::: 1, K a field containing the n-th roots of unity,
and Lee a cyclic extension of K of Galois group G with generator 0-. For
every root of unity c: and every element x of L, the sum
is called a Lagrange resolvent.
COMMENTARY. - The history of the resolvents used by Lagrange in 1770,

in particular in the study of third-degree equations, lies at the heart of the
algebraic resolution of equations. Resolvents can be found, in varying forms,
in the work of Euler (1750), Bezout (1765), Vandermonde (1770), and Gauss
(around 1800). In the proof above, the application of Hilbert's Theorem 90
to C 1 leads to the linear combination id + (-10- + ... + [(-1 t-
1o-n-1 and
thus to a Lagrange resolvent.
10.6.2 Properties
With the notation of §6.1, we have:
1) o-((c:,x) = c:- 1 (c:,x);
2) (l,x) E K;
3) (c:, x)n E K;
4) (c:,x)(c:- 1 ,x) E K;
5) L c:-r(c:,x) = no-r(x) for 0 :::; r :::; n - 1.

c:EJ.Ln
PROOF. -
1) We have
U«e,X)) u(x) + ... + en-1Un(X)

e-l[eU(X) + ... + en-1Un-l(X) + x]
e-l(e, x).
2) As u(l,x) = (l,x), by 1), (l,x) is invariant under u, and so under

every element of G. Thus, §8.5 gives the result.
3) U«e,x)n) = [U«e,X))]n = [e-l(e,X)]n = (e,x)n, by 1), which gives

the result exactly as in 2).
4) We have
U«e, X)(e- l , x)) U«e,X))U«e-l,X)) = e-l(c, x)c(cl, x)

(c,x)(Cl,x)
by 1), which again gives the result just as in 2).
5) Set Sr = L c-r(c,x).
eEl-'n
We have Sr = L c- Lr ckuk(x) = L uk(x) L c k- r .

eEl-'n O:Sk:Sn-l O:Sk:Sn-l eEl-'n
Now, L ck - r = n if k = rand L ck - r = 0 if k =J- r, since if w is

eEl-'n eEl-'n
a primitive n-th root of unity, then
'"' c k-r
L-. =
O:SI:Sn-l
This gives the result.
10.7 Resolution of the Cubic Equation

. In this section, we will show how the results of the preceding sections and
of Chapter 8 guide the resolution of cubic equations.
A cubic equation can always be brought (by translation) to the form
x 3 + px + q = 0, with p,q E <C. Set K = Q(p,q,j); we need to adjoin the
cube roots of unity because we will encounter a cyclic cubic extension.
Let a, b, c be the roots of the equation in Co Recall (see VII.l and III.6)
that K[a, b, c] = K[a, dJ with d = (a-b)(b-c)(c-a) and d 2 = _4 p3 -27q2.
10.7 Resolution of the Cubic Equation 185
We know that the Galois group of X 3 + pX + q is, in general, 8 3 , identified

with the group of permutations of {a, b, c}; in what follows, we fix this
identification once and for all. Let L = I(A3) be the field of invariants of
the alternating group A 3 , i.e. the group of even permutations of {a, b, c}.
This group contains three elements: the identity, the 3-cycle (Y = (a b c)
and its square (Y2 = (a c b). As A3 is a subgroup of index 2 of 8 3,L is
a quadratic extension of K and there exists an element z E L such that
Z2 E K and L = K[z].
Furthermore, K[a, b, c] is a cyclic extension of L, since K[a, b, c] is a nor-
mal extension of K, so it is a normal extension of L, and the alternating
group A3 = Gal(K[a, b, c]IL) is a cyclic group of order three. By Propo-
sition 10.5, there exists an element z' E K[a, b, c] such that z,3 ELand
K[a, b, c] = L[z']. The trellis of subgroups of the equation thus shows the
existence of an intermediate extension which we will use to compute a, b, c
in steps (Figure 10.1). To determine elements like z and z', we will use
Lagrange resolvents and results from §1O.6.
K[a,b,c]=L[z']
I
L=K[z]
K
FIGURE 10.1.
To determine z, we use the discriminant d; the even permutations of

{a, b, c}, namely id, (Y, (Y2, leave d invariant, and an odd permutation like
(a b) transforms d into its conjugate -d =I d. This shows that dEL and
d rf. K,
so L = K[d].
To determine z', we follow the ideas of §10.4 and §1O.5. We choose a cube
root of unity, for example j. The linear combination of K-automorphisms
f = id + j(Y + j2(Y2 is non-zero by Dedekind's theorem. It does not vanish
at a, otherwise we would have f(b) = j2 f(a) = 0, f(c) = jf(a) = 0, but as
f(l) = 0, f would be zero. Consider (j, a) = a + jb + j 2c = f(a). We have
(j,a) rf. L, since (Y(j,a) = j2(j, a) =I (j,a) and (j,a)3 E L, by Property
10.6.2 c); thus, (j, a) is a primitive element of the extension K[a, b, c] of
K[d].
It remains to apply the techniques of Chapter 3 to compute the equation
satisfied by (j,a)3 = (a+jb+j 2c)3. As
we find
(j, a)3=a 3 + b3 + e3 + 3j(a2b + b2e + e2a) + 3j2(ab2 + be2 + ea 2) + 6abe

·VS
=a3 + b3 + e3 -
3
2[a 2b + b2e + e 2a + ab 2 + be2 + ea 2]- 3 T d + 6abe
=_ 27q _ 3iVS d.
2 2
We obtain three values of (j, a), corresponding to the three possible values
of a. Similarly, we find (l, a)3 = -(27q/2) + (3iv'3/2)d.
We need to determine the roots of the equation. By Property 10.6.2. 4),
(j,a)(j2,a) E L:
(a + jb + le)(a + j 2b + je) = a 2 + b2 + e2 - ab - be - ea
-3p.
Thus, we obtain three values of (j2, a) corresponding to the three values of

(j, a). Now, because (1, a) = a + b + e = 0, the formulas of 10.6.25) yield
3a (j,a) + (j2,a),
3b l(j,a) + j(j2,a),
3e j(j, a) + j2(j2, a).
We recover Cardan's formulas this way, since
( (j,a))3
3
= _~ ±
2
J+ q2
4
p3
27'
etc.
10.8 Solution of the Quartic Equation

A quartic equation with complex coefficients can always be brought (by
translation) to an equation of the form X4 +px 2+qx+r = 0 with p, q, r E C.
Set K = lQ(p, q, r, j); again, we need to adjoin the cube roots of unity
because we will encounter a cyclic extension of degree 3. Let a, b, e, d denote
the roots of the equation in C. We know that the Galois group of the
equation is, in general, the group 8 4 identified with the permutations of
{a, b, e, d}; again, we fix such an identification once and for all. Let us show
how knowing the Galois .group guides the resolution of the fourth-degree
equation.
10.8 Solution of the Quartic Equation 187
~T~
H=«ab)(cd» H'=«ac)(bd» H"=«ad)(bc»
~I~
V
I / K[a'l'C,d~
A4
T=M[a+b] T'=M[a+b] T"=M[a+b]
I I
S4
M=L[u]
/
I
L=K[O]
I
K
FIGURE 10.2.
Let A4 be the alternating subgroup, and V the normal subgroup of order

4 generated by products of two disjoint transpositions (these are called
double transpositions). Let H, H', H" denote the three subgroups of V
having two elements, and let L, M, T, T', T" be the corresponding subfields
of invariants. Figure 10.2 gives the Galois correspondence for the subgroups
of 8 4 just described, and their corresponding subfields.
As in §1O.7, L is a quadratic extension of K and L = K[8], where 8 is a
square root of the discriminant of the equation
8 = (a - b)(a - e)(a - d)(b - e)(b - d)(e - d).
M is a cyclic extension of degree 3 of L, since V is a normal subgroup of
index 3 of A 4 . Lagrange's method for finding an element generating Mover
L is to consider
u = (a + b)(e + d), v = (a + c) (b + d), w = (a + d)(b + c)

(we can also work with the resolvents (i,a), (-l,a), (-i,a) or with ab+
cd, ae + bd, ad + be). The actions of V and A4 over u, v, ware the desired
ones: they are invariant under V but not under A4 (recall that the elements
of A4 are 3-cycles or double transpositions), so they generate M. Moreover,
considering the action of the 3-cycle (a be), we see that they are conjugates
over L, so they are the roots of a cubic equation over L, which we can
determine by computing u + v + w = 2(ab + ae + ad + be + bd + cd) = 2p.
Similarly,
uv+vw+wu
uvw
The equation y3 - 2py2 + (p2 - 4r)y + q2 = 0 is called the resolvent equation

of the original equation. As it is cubic, we know how to solve it (see §10.7).
Note that [M : L] = 3 and not 6; we can check by an easy computation
that (u - v) 2 (v - w) 2 (w - u)2 = tP, so the discriminant of the resolvent lies
in L.
Let us continue: a + b rj. M since the element (a c)(b d) of V modifies
it; we do, of course, have a + b E I(H). Thus, a + b is of degree 2 over
M and T = M[a + b]; its conjugate is c + d, the image of a + b under the
permutation (a c)(b d). As a + b + c + d = 0 and (a + b)(c + d) = u, we
set a + b = Fu and c + d = -Fu.
Similarly, a + c = 0 and b + d = -0, a + d = J -wand b + c =
-J-w.
The numbers Fu, Fv, J -w should not be selected independently.
The (easily checked) equality (a+b)(a+c)(a+d) = -q shows that we have
the choice between two of the roots.
We finally obtain
2a a+b+a+c+a+d = Fu+ 0+ J-w,

2b Fu-0-J-w,
2c -Fu + 0 - J-w,
2d -Fu - 0 + J-w.
10.9 Historical Commentary

Lagrange was not the first to make the attempt to unify the methods for
solving equations of degrees 2, 3, and 4. Tschirnhaus also tried; however,
his ideas did not extend to equations of degree 2': 5 (see Exercise 3.2).
Lagrange also wrote: "We can be assured that even if we succeeded in
giving a general solution of the fifth degree equation and the following ones,
we would have only algebraic formulas, precious in themselves, but not very
useful for effective resolution .... "
For further information on the methods of Lagrange, see for example the
book by J.-P. Tignol (pp. 163-201) listed in the bibliography.

Exercise 10.1. Cyclic and abelian extensions
1) a) Show that a quadratic extension is cyclic.
b) Determine the set of elements x of Q[j] such that Q[x] = Q[j]

and x 2 E tQ.
2) Let N be an abelian extension of a field K. Show that every interme-

diate extension L is an abelian extension of K.
3) Give an example of a non-cyclic abelian extension of Q.
4) Let a = 15, b = a + ia, and let N be the normal closure of Q[b] over
Q.
a) Is N a cyclic extension of Q?
b) Give a strict subfield K of N such that Gal(NIK) is cyclic of
. order 4.
5) Determine the splitting field N of the polynomial X 6 - 2 over Q[j].

Describe the elements of the Galois group G = Gal(NIQ[j]), and
determine the structure of G.
Exercise 10.2. The splitting field of (XP - 2)(Xq - 2)

1) Let K = Q[j], set P(X) = (X2 - 2)(X3 - 2), and let N be the
splitting field of P over K.
a) Show that N = K[a], where a is a suitable (non-integral) power

of 2; determine [N : K], and show that N is a cyclic extension
of K.
b) Give the trellis of intermediate extensions between K and N;

are they all normal extensions of K?
2) Now let p and q be two distinct primes n = pq, ( a primitive n-th root
of unity, K = Q[(]' P(X) = (XP - 2)(XQ - 2), and N the splitting
field of P over K. Show that N is a cyclic extension of K.
Exercise 10.3. Cyclic extensions without Hilbert's Theorem 90

Let K be a field contained in <C. A polynomial P of K[X] is said to
split if it is a product of linear factors in K[X]. Recall that if f is
an endomorphism of a K-vector space such that there exists a split
polynomial Pin K[X] with simple roots such that P(f) = 0, then f
is diagonalisable.
1) Let n ~ 1 be an integer, ( a primitive n-th root of unity, and K a

field containing (. Let L be a cyclic extension of K of degree n, with
Galois group G generated by a. Consider a as an endomorphism of
the K-vector space L.
a) Show that the eignvalues of a are n-th roots of unity, and that
1 is an eigenvalue of a.
b) Show that a is diagonalisable.
c) Deduce that 1 cannot be the only eigenvalue of a.
d) Show that the eigenvalues of a form a cyclic group isomorphic
to the group J.Ln of n-th roots of unity.
e) Show that there exists a E K, bEL such that bn = a and
L = K[b].
2) Let P(X) = X 3 + pX + q be a polynomial with coefficients in C,

irreducible over K = Q(p, q). Let a, b, c denote its roots in C, and let
d be a square root of its discriminant. Set L = K[j, d], N = L[a, b, c].
a) Show that (l,a, b) is a basis of N over L.
b) Show that G = Gal(NIL) is cyclic.

c) Let a be a generator of G. Determine the eigenvalues and eigen-
vectors of a.
d) Check that we obtain Lagrange resolvents.
Exercise 10.4. The norm as determinant

Let Lee be a normal extension of a field K and a an element of L.
Let P(X) = LO<k<n akXk be the minimal polynomial of a over K,
and L' the norm~l ~losure of K[a] in L. We write m : L ---t L for the
map defined by m(x) = ax, and set m' = miL'.
1) Check that m is K -linear.
2) a) Express N U / K ( a) in terms of the coefficients of P.

b) Show that NU/K(a) = det(m').
3) Deduce that NL/K(a) = det(m).

1) a) We know that such an extension is normal and has Galois group

isomorphic to Z/27L
b) Q[j] is a cyclic extension of Q, since it is a normal extension and its

Galois group over Q is 7l/271. Thus, there exists x such that Q[x] = Q[j]
and x 2 E Q. Writing x = a + bj, we see that x 2 = a 2 + 2abj + b2( -1 - j)
is real if 2ab - b2 = 0. As b must be non-zero, it is necessary and sufficient
that b = 2a. We find x = a + 2aj = ai J3 with a -I- 0. Thus, we have
x 2 = -3a 2 E Q.
2) Gal(NIK) is an abelian group, and Gal(NIL) is a normal subgroup of
it; thus L is a normal extension of K. Moreover, Gal(LIK) is a quotient of
Gal(NIK), so it must be abelian, which gives the result.
3) Take the field Q [J3, h]; its Galois group over Q is 7l/271 x 7l/271.
4) a) The minimal polynomial of b over Q is X4 + 20, whose roots are
±b and ±ib. Thus we have N = Q [i, -Y5]. We see that [N : Q] = 8 and
Gal(NIQ) contains the elements (T and T defined by a(i) = i, (T(a) = ia,
T(i) = -i, T(a) = a.
We note that (TT -I- Ta, since aT(a) = ia and T(T(a) = -ia. The group
Gal(LIQ) is thus neither commutative nor cyclic.
b) Set K = Q[i]; then K contains the fourth roots of unity. Thus, N is
a cyclic extension of K, by §10.2, with Galois group isomorphic to 7l/471
since [N : Q[i]] = 4.
5) As -l = e'7f/3 is a sixth root of unity, the splitting field is N =

Q[±-v2,±j-v2,±j2 -v2] = Q[-v2,j], so we have [N: Q] = 12,[N:
Q[j]] = 6. A direct application of §1O.2 shows that N is a cyclic extension
Q[J], with Galois group isomorphic to 7l/671. The elements of Gal(NIQ[j])
are defined by (T( -v2) = (_j2)k-v2, O::S k::S 5.
1) a) We already saw, in part 2) of Exercise 4.5, that a = 21 / 6 is a primitive

element of Q [h,
{f2]. As [N: Q] = [Q[a,j] : Q [a]][Q [a] : Q] = 12, it is
clear that [N : K] = 6 and a is of degree 6 over K. As j is a sixth root of
unity, Proposition 10.2 shows that N is a cyclic extension of K.
b) The only non-trivial subgroups of 7l/671 are {O, 2, 4} and {O, 3}; by the
Galois correspondence, they correspond to extensions K[h] and K[ {f2],
which are thus the only non-trivial intermediate extensions between K and
N; note that they are all normal since Z/6Z is abelian.
2) We have N = K[21/p, 2 1/ q ]. Set a = 21/n. As a q = 2 1/ p and aP = 21/ q ,

we have N C K[a]. As there exist integers u and v such that up+vq = 1 by
Bezout's identity, we have 2v / P 2u / q = a, which gives the converse inclusion
and N = K[a].
1) a) If ). is an eigenvalue of (T, there exists a non-zero x E L such that

(T(x) = ).x. Thus (Tk(x) = ).kx for every integer k ~ 0; for k = n, we obtain
).n = 1. Moreover, we have (T(1) = 1.
b) As (Tn - I = 0, we have P«(T) = 0 in K[X], with P(X) = xn - l.
As P is split over K because K contains the n-th roots of unity, the fact
recalled above implies that (T is diagonalisable.
c) Because it is diagonalisable, (T cannot have 1 as its only eigenvalue,
unless it is the identity, which is false.
d) If ). and /-L are eigenvalues of (T and if x and yare the associated non-
zero eigenvectors, we have xy =J 0 and (T(xy) = )./-Lxy, (T(x- I ) = ).-IX- I .
The eigenvalues of (T thus form a subgroup H of /-Ln.
If H is strictly contained in /-Ln, it is a cyclic group of order d dividing n,
and thus contains only d-th roots of unity. As (T is diagonalisable, we have
(Td =id, which is false.
e) By d), H contains (. Let b be a non-zero eigenvector of (T associated
to (. We have (T(b) = (b, so (T(b n ) = Cb n = bn ; as a = bn is invariant under
(T, it is invariant under all the elements of G, so it lies in K.
The conjugates of b are the (Tk (b), 0 $ k < n, i.e. the (k b for 0 $ k < n.
They are all distinct, so b is of degree n over K and L = K[b].
2) a) If b lies in the L-vector space generated by 1 and a, we see that c E L

since c = -a - b. Thus [N : L] $ 2, which is false.
b) By a), G is cyclic of order 3.
c) As (T induces a permutation of the roots of P, we have (T( a) = b or c.
Consider the first case. The matrix of (T in the preceding basis is given by
(o~ ~ -~).
1 -1
The characteristic polynomial is 1 - X 3, the eigenvalues are 1, j and j2,

and the corresponding eigenvectors are (1,0,0), (0,1, -j) and (0,1, _j2),
i.e. a - jb and a - j 2 b.
d) As (1 - j)(a - jb) = a + lb - ja - jb = a + j 2 b + jc and similarly,
(1 -l)(a - j 2 b) = a + jb + j 2 c, we find Lagrange resolvents up to the
coefficients 1 - j and 1 _ j2.
1) This is obvious.
2) a) NL'/K(a) = n
aEGal{L'IK)
b) It is enough to compute the determinant of the matrix of m' in the

basis (ak)o::;k::;n.
3) The determinant of m can be computed in a basis of the form
where (bl)o<l<r is a basis of £ over £'. We check that it is equal to

(det(m')Y, ;'hich gives the result since NL/K(a) = (Nu/K(a»[L:L'l.
11
Solvable Groups
This chapter and the next one are devoted to the problem of resolving
algebraic equations by radicals. Given a polynomial with coefficients in a
field K, together with its splitting field N over K, the solvability of the
equation P(x) = 0 by radicals can be expressed in terms of the existence
of a particular sequence of intermediate extensions between K and N (see
Chapter 12). By the Galois correspondence, this translates to a property
of Galois group Gal(NIK). In this chapter, we introduce the groups having
this special property, called solvability.
11.1 First Definition

A finite group G is solvable if it has a finite decreasing sequence (GihSiSr
of subgroups such that
1) Go = G :J ... :J G r = {e} ;
2) G H1 is a normal subgroup of G i for 0 ::; i ::; r - 1;
3) GdG H1 is commutative for 0 ::; i ::; r - 1.
EXAMPLES. -
1) An abelian group G is always solvable: it is enough to take Go = G

and G 1 = {e}.

196 11. Solvable Groups
2) 8 3 is solvable: take Go = 8 3 , G 1 = A3 and G 2 = {e} .
3) 8 4 is solvable: take Go = 8 4 , G 1 = A 4 , G 2 = V, and G 3 = {e}

where V is the subgroup consisting of the three double transpositions
(products of two disjoint transpositions) and the identity.
4) An and 8 n are not solvable for n 2: 5 (see §11.6).
5) Any p-group, and any group of order pq for distinct primes p and q,
is solvable (see Exercise 11.3).
6) Any group whose order has only two prime divisors is solvable (Burn-
side's theorem, 1904).
11.2 Derived or Commutator Subgroup

Recall that a commutator in a group G is an element of the form [a, b] =
a-1b-1ab, for any a, bEG. The set of commutators of G generates a
subgroup of G called the commutator subgroup of G, or the derived subgroup
of G. We denote it by D(G).
If f : G ----t G' is a group homomorphism, we have
f(D(G)) = D(f(G)) C D(G') because f[x, y] = [f(x), fey)]·

Recall that D(G) is a normal subgroup of G, and even a characteristic
subgroup of G, i.e. a subgroup stable under every automorphism of G (not
only inner automorphisms).
Finally, let us recall a property that will be quite useful later on: for
every normal subgroup H of G, H ::J D(G) is abelian if and only if G/H
is abelian.
11.3 Second Definition of Solvability

PROPOSITION. - A finite group G is solvable if the sequence Di(G) defined
inductively by DO(G) = G and for i 2: 0, DHl(G) = D(Di(G)), eventually
stabilizes at { e} , i. e. if there exists an integer s such that D S (G) = {e} (so
that Di(G) = {e} for all i 2: s).
PROOF. - Assume that G is solvable (as defined in §11.1). Then, by in-
duction, we have Di(G) C G i . Indeed, this is clear for i = 0, and if we
assume that for some integer i, Gi/G H1 is abelian, then we obtain Gi+l ::J
D(G i ) ::J D(Di(G)) = DHl(G). As G r = {e}, we have Dr(G) = {e}.
Conversely, the sequence Di (G) is a sequence satisfying the conditions of
the first definition, since conditions 1) and 2) are obvious, and also because
11.4 Examples of Solvable Groups 197
the fact that Di+l(G) = D(Di(G)) is normal in G i shows that the quotient
Di(G)IDi+l(G) is commutative for every i. <>
11.4 Examples of Solvable Groups

PROPOSITION. - The following groups are solvable:
1) a subgroup of a solvable group;
2) a quotient of a solvable group by a normal subgroup;
3) an extension of a solvable group by a solvable group;
4) a finite product of solvable groups.
PROOF. -
1) Let H be a subgroup of a solvable group G. We have Di(H) c Di(G)

for every i 2:: o. As G is solvable, there exists an integer r such that
Dr (G) = {e}, so Dr(H) = {e}, which gives the result.
2) Let K be a normal subgroup of a solvable group G, and let p :
G ----> GIK be the canonical homomorphism. We have p(D(G)) =
D(p(G)) = D (GIK) , so by induction, wehavep(Di(G)) = Di (GIK)
since p(G) = GIK. If Dr (G) = {e}, then Dr (GIK) = {e}, which
gives the result.
3) Let {e} ----> K ~ G ~ H ----> {e} be an exact sequence of groups, with

K and H solvable (we say that G is an extension of K by H, or an
extension of H by K; it depends on who is writing). There exists r
such that Dr(H) = {e}, so p(Dr(G)) = {e}; hence Dr(G) C i(K).
There exists s such that DS(K) = {e} so Dr+s(G) C i(DS(K)) = {e},
which gives the result.
4) It suffices to note that the product of two solvable groups Hand K is

solvable since H x K is an extension of H by K. Indeed, the sequence
{e} ----> K ~ H x K ~ H ----> {e} is exact with i(k) (e,k) and
p(h,k) = h. <>
11.5 Third Definition

PROPOSITION. - A finite group G is solvable if and only if there exists a
finite decreasing sequence (Hj)o"5.j"5.r of subgroups of G such that
1) Ho=G-:J··<)Hr={e};
2) Hj+! is a normal subgroup of H j for 0:::; j :::; r - 1;
3) H j f Hj+! is cyclic of prime order for 0 :::; j :::; r - 1.
PROOF. - A group satisfying the above properties is solvable; this is clear

using the first definition of solvability. Let us show the converse. Take a
sequence (Gi)O~i~r of subgroups of G satisfying the conditions of the first
definition. For every i, 0 :::; i :::; r - 1, construct by induction a sequence
(Kj ) of intermediate groups between G i and G H1 : Ko = G i , and KHI
is a maximal normal subgroup of K j containing G H1 . The sequence (Kj )
is decreasing, and finite because G is finite. Because the normal subgroups
of Kjf KHI are in bijection with the normal subgroups of K j containing
Kj+!, K j f Kj+! is a group that has no proper normal subgroup, i.e. a
simple group. Moreover, K j fKH1 is a subgroup of G;jG H1 . Thus, it is
abelian, and consequently, it is cyclic of prime order. Putting together the
sequences (Kj ) obtained for each i, we obtain the result. <>
11.6 The Group An Is Simple for n > 5

11.6.1 Theorem
For n 2': 5, the alternating group An is simple (i.e. it has no normal sub-
groups except for itself and the identity) and non-solvable; the group Sn is
also non-solvable.
COMMENTARY. - This fundamental theorem is due to Galois. One can show

directly that An is not a solvable group (see §l1.6.2).
PROOF. - Let n 2': 5. Recall (see Exercise 11.1) that An is generated by
the set of 3-cycles or by the set of double transpositions. Let us show that
two 3-cycles a = (a b c) and a' = (a' b' c') are conjugate in An. There
exists a in Sn such that a(a) = a', a(b) = b', a(c) = c'. If a belongs to
An, the equality aaa -1 = a' shows that the two 3-cycles are conjugate in
An; we obtain the same result if a does not belong to An, by changing a
to j3 = a 0 (d e) where d and e lie in the complement of {a, b, e}. Similarly,
we see that two double transpositions (a b) (e d) and (a' b') (c' d') are
conjugate in An, by taking any permutation a E Sn such that a(a) = a',
a(b) = b', aCe) = e', a(d) = d'. If a does not belong to An, set j3 = ao(a b).
Let H be a normal non-trivial subgroup of An. Let us show that H
always contains a 3-cycle or a double transposition. By the above, it will
contain all of them, and therefore be equal to An.
Let a be an element of H different from the identity element e, and let
al ... ak be its decomposition into a product of disjoint cycles, indexed so
that the sequence of lengths l(ai) of these cycles is decreasing.
11.7 Recent Results 199
1) Assume that l(al) = r > 3. Set
We have
17- 117 (al· .. a r )-l(al a2 a3)(al ... ar )(al a2 a3)-1

(a r al a2)(al a3 a2)
(al a3 ar ).
Then a' lies in H, because H is normal in An, SO a-la' = (al a3 ar )

lies in H; H contains a 3-cycle, so H = An.
2) Assume that l(al) = 3 and l(ai) = 2 for i > 1. Then H contains the
3-cycle 172, so H = An.
3) Assume that l(al) = l(a2) = 3. Set 171 = (a be), 172 = (x y z),

and a' = (b c x)a(b c X)-l. Because a-la' lies in H and a-la' =
(a b x c z), we can conclude using part a).
4) Assume that l(ai) = 2 for i = 1, ... , k. Set 171 = (a b), 172 = (c d),
and a' = (b c d)a(b c d)-I. Because a-la' lies in H and a-la' =
(a d)(b c), H contains a double transposition, so H = An.
In every case, H = An; thus An is a simple group. As it is not
commutative, it is not solvable, and by Proposition 11.4, Sn is also
non-solvable. <>
11.6.2 An Is Not Solvable for n ~ 5, Direct Proof
PROPOSITION. - An is not solvable for n ?:: 5.

PROOF. - The proof consists in showing that the normal subgroup D(An)
is equal to An. Because An is generated by the 3-cycles (see Exercise 11.1),
it suffices to show that every 3-cycle (a b c) belongs to D(An). Let d and
e be two elements of {I, ... ,n}, different from a, b, c; then
[(a cd), (b c e)] = (a d c)(b e c)(a c d)(b c e) = (a be). <>
11.7 Recent Results

The classification of all finite simple groups was completed in 1981. Galois
discovered the first examples of these, namely, the alternating groups An for
n?:: 5. Emile Mathieu discovered others in 1861. Beginning with a lecture by
Richard Brauer at the International Congress of Mathematicians in 1954,
together with seminal work of Claude Chevalley from the same period, the
proof that the list of known finite simple groups was actually a complete list
represents an enormous effort on the part of the mathematical community.
Counted all together, the different parts of the proof fill nearly 10,000 pages;
it is the longest proof on record. Given that even the best mathematicians
make errors every 50 pages or so .....
Certain simple groups fall naturally into some infinite series, whereas
others, called sporadic, seem to be the only ones of their kind. Among
the sporadic simple groups, the most famous and largest one is known as
the monster (or sometimes, the friendly giant). It possesses spectacularly
beautiful properties; its order is equal to 246 X 320 X 59 X ~ X 112 X 133 x
17 x 19 x 23 x 29 x 31 x 41 x 47 x 59 x 71, which is roughly 1054.
The proofs of theorems concerning solvable groups are also sometimes
extremely long. In 1963, Walter Feit and John Thompson published a fa-
mous 255 page proof of the theorem "All finite groups of odd order are
solvable."

The first exercises recall basic properties of permutation groups.
Exercise 11.1. The permutation groups Sn and An

1) The conjugate of a cycle.
Let n be an integer, £7 a permutation of {I, ... ,n}, and (Xb .•. ,Xk)
a k-cycle of Sn, i.e. the permutation associating Xi+l to Xi for i =
1, ... ,k - 1, Xl to Xk and leaving the other elements of {I, ... ,n}
fixed. Show that £7(Xb .•. ,Xk)£7- l = (£7(Xl), ... ,£7(Xk)).
2) Let n ::::: 3 be an integer. Knowing that Sn is generated by the set

of transpositions of {I, ... , n}, show that Sn is generated by the
following sets of permutations:
a) the set of transpositions (1 2), ... , (1 n);
b) the set of transpositions (1 2), (2 3), ... , (n - 1 n);

c) the set of two permutations (1 2) and (2 3 ... n);
d) the set of two permutations (1 2) and (1 ... n);
e) the set consisting of an arbitrary transposition and an arbitrary

p-cycle £7.
3) Let n ::::: 3 be an integer.

a) Show that the product of two distinct transpositions of Sn is

either a 3-cycle or a product of two 2-cycles. Deduce that An is
generated by the set of 3-cycles of Sn.
b) Show that for n ~ 3, An is generated by the set {(I 2 3), ... ,

(1 2 n)}.
c) Show that for n ~ 5, An is generated by the set of permutations

known as double transpositions, i.e. permutations of the form
(a b)(c d) with a, b, c, d pairwise distinct. Show that this result
is false for n = 3, 4.
d) Show that, for every n ~ 1 and every odd integer k with k :::; n,
An is generated by the k-cycles, by showing that every 3-cycle
is a product of two k-cycles.
4) Take an action of Sn on a set E. Let p be the largest prime number

less than or equal to n. Show that the orbits for this action have
cardinal equal to 1, 2, or a number greater than or equal to p.
Exercise 11.2. The groups A4 and S4

1) Let G be a subgroup of A4 containing two 3-cycles defined over dis-
tinct subsets of E = {I, 2, 3, 4}. Show that G = A 4 .
2) a) Show that the elements of A4 different from the identity element

are eight 3-cycles and three double transpositions.
b) Show that A4 has a unique normal subgroup, describe it, and

show that it is normal in S4.
3) Show that A4 is the only subgroup of order 12 in S4.
4) Find the transitive subgroups of A4 and S4.
Exercise 11.3. Examples of solvable groups

1) Show that a p-group is solvable.
2) Show that a group G of order pq for distinct primes p and q is solvable.
3) Show that a group G of order pqr for distinct primes p, q, and r is

solvable.
4) Show that a group G of order p2q for distinct primes p, q is solvable.

Exercise 11.4. Transitive and solvable subgroups of Sp

Let p be a prime. We know that (ZjpZ, +,.) is a finite field with p
elements. Let E = {O, ... ,p -I} denote the set of its elements, and
(for this exercise) consider Sp as the group of permutations of E.
Let GA(p) denote the group of bijective affine maps from ZjpZ to
itself, i.e. the maps fa,b : ZjpZ --> ZjpZ defined by fa,b(X) = ax + b,
with a of. 0, a, b E ZjpZ. Set t = hI and ma = fa,o. We will show
that a subgroup G of Sp having the following properties:
1) the action of G on E is transitive,

2) G is solvable,
is conjugate to a subgroup H of GA(p), i.e. of the form (JH (J-I for a

permutation (J of Sp; we will also prove the converse.
1) The group GA(p)
a) Show that mat = tama.
°
b) Show that every element of GA(p) can be written uniquely in
the form tbm a , with 1 :S a :S p - 1, :S b :S p - 1.
Deduce that IGA(p) I = p(p - 1).
c) Show that (t) is a normal subgroup of GA(p).
d) Show that GA(p) is transitive and solvable.
2) Let G be a transitive subgroup of Sp. Show that a normal subgroup

H of. {id} of G acts transitively on E.
3) a) Let G be a solvable subgroup of Sp acting transitively on E.

Thus, we assume that there exists a finite decreasing sequence
(H j )O::;j::;r of subgroups of G, satisfying the conditions of defi-
nition 11.5.
Show that H r - I is conjugate to the group (t).
b) Deduce that G is conjugate to a subgroup of Sp containing t.
4) Let (J E Sp be such that (Jt(J-I E GA(p). Show that (J E GA(p).
5) Show that a group having properties 1) and 2) is of the stated form.
6) Let G be a transitive subgroup of Sp.
a) Show that if G is solvable, the only permutation of G having at

least two fixed points is the identity.
b) Now let us prove the converse. By counting the elements of G

that have a fixed point, show that there exists an element T E G
that has no fixed point. Show that T is a p-cycle. Conclude.
COMMENTARY. - These results are due to Galois. Later (see Exercise 12.3),
we will prove the main result on equations that he deduced from them.

Partial solution to Exercise 11.1.
1) Set 0: = O"(Xl, ... , Xk)O"-l and (3 = (O"(Xl), ... , o"(Xk». We see that
o:(y) = (3(y) in each of the two cases y tf- {O"(Xl), ... ,O"(Xk)} and y E
{O"(Xl), ... , o"(Xk)}.
2) d) Using the equalities
(l. .. n)k(l 2)(1. .. n)-k=(k+1 k+2) for k=1, ... ,n-2,
we find ourselves in the situation of b).

e) Let (i j) denote the transposition; there exists an integer k such
that O"k (i) = j. If we set T = O"k, then via the bijection cp defined by
cp(l) = i, cp(2) = j, cp(r + 1) = TT(i) for 2 ::; r ::; n - 1, we return to the
previous situation.
3) d) The k-cycles are even permutations when k is odd, and they generate
the set of 3-cycles, via the formula
4) The result is clear if p = 2, so let p be an odd prime, and take x E E.

For every p-cycle 0", the orbit of x under the action of (0") has either one
or p elements. If for every p-cycle 0", the orbit of x under the action of (0")
has only one element, then the orbit x under the action of An has only one
element by 3) d), so the orbit of x under the action of 8 n has at most two
elements.
Partial solution to Exercise 11.2.

4) The order of a transitive subgroup G of A4 is a multiple of 4, which is
the order of the unique orbit, and it divides 12, so it is equal to 4 or 12. If
it is 12, then G = A 4 • If it is 4, then G is the subgroup generated by the
double transpositions.
Similarly, a transitive subgroup G of 8 4 has order a multiple of 4 and a
divisor of 24, so it is equal to 4, 8, 12 or 24. If it is 24, then G = 8 4 , and if
it is 12, then G = A 4 . If it is 4, then G can be the subgroup generated by

the double transpositions, but also one of the three subgroups generated
by a 4-cycle. If it is 8, then G is a Sylow 2-subgroup of 8 4 , i.e. one of the
three subgroups isomorphic to the dihedral group described in §8.7.2.

1) If G is a p-group, we know that its order is of the form pn. Let us use
induction on n. If n = 1, then G ::: ('lllp'll, +), and the result is clear.
Suppose the result holds for every p-group of order pk with k < n, and let
G be a p-group of order pn. The center Z(G) of G is a non-trivial abelian
subgroup of G, so it is solvable. As Z(G) is normal in G, the quotient
GIZ(G) is a p-group, solvable by the induction hypothesis. We conclude
that G is solvable by Proposition 11.4 3).
2) Suppose that p > q. The number mp of Sylow p-subgroups of G divides
pq and is equal to 1 mod p. Thus, mp = 1 and G has a unique Sylow p-
subgroup H. We know that H is solvable and normal in G, and that the
quotient G I H is solvable, so we conclude that G is solvable by Proposition
11.4 3).
3) Suppose that p > q > rand G does not have any normal subgroups.
Then by the Sylow theorems, we find:
a) mp = 1 mod p, mp > 1 and mplpqr, so mp = qr;
b) mq = 1 mod q, mq > 1 and mqlpqr, so mq 2: p;
c) mr = 1 mod r, mr > 1 and mrlpqr, so mr 2: q.
Now, Sylow l-subgroups for distinct primes l = p, q, r have trivial inter-
section, so we find that
IGI 2: qr(p - 1) + p(q - 1) + r(q - 1) + 1 > pqr.
This contradiction shows that G must have a normal subgroup of prime
order, so it must be solvable. Denote it by H. The group G I H is solvable
by 2), and we can thus again use Proposition 11.4 3) to conclude.
4) If p > q, we see that mp = 1, G has a normal subgroup H of order p2,

so solvable, by 1), and the quotient G I H is of order q and is also solvable.
Again, we can conclude using §9.4 3).
Let p < q, and assume that G has no normal subgroups. The Sylow
theorems give mp = 1 mod p, mp > 1, and m plp2q, so m~ = q. But they
also give mq = 1 mod q, mq > 1, and m qlp2q, so mq = p . As two Sylow
l-subgroups for distinct primes l = p, q have trivial intersection, we find
IGI 2: q(p2 - 1) + p2(q - 1) + 1 > p2q.
This contradiction shows that G must have a normal subgroup H of order
p2 or q. Because H is solvable, the group GIH is also solvable, and once
again we conclude using §9.4 3).

1) a) This follows from the fact that a(x + 1) = ax + a. Note that t is the
p-cycle (01 ... p - 1).
b) As fa,b = tbm a , it suffices to check that if ax + b = a'x + b' for every
x E IFP' then a = a' and b = b', which is easy.
c) Part a) shows that matm;;1 = t a , which suffices since the ma and t
generate GA(p).
d) The quotient G A(p) / (t) is commutative since the representatives of
the classes are ma's, which commute with each other. Thus, it is solvable.
The same holds for GA(p) and its subgroups, by Proposition 11.4.
2) Let x and y E E. As G is transitive, there exists f E G such that

f(x) = y; for every h E H, we have f-Ihf E H, so f-I(OH(Y)) C OH(X).
Consequently, IOH(X)1 = IOH(y)l. Under the action of H, E is thus the
union of r disjoint orbits having the same number of elements s. As s > 1
(because H =f:. {id}), P = rs, so that we must have s = p and r = 1. This
proves the result.
3) a) By the preceding question, the groups Hi for i = 0, ... , r - 1 act

transitively on E. The only possibility for H r - I is a cyclic group generated
by ap-cycle c. If c = (xo ... Xp-I), and if we define <p: E ---t E by <P(Xi) = i,
then <pC<p-1 = t.
b) <pG<p-1 answers the question.
4) If crtcr- I = tbm a , we see, by an easy computation, that crtkcr- I = tnmak
for k 2:: 1, with n = L aib. If k = p - 1 and a =f:. 1, we find that
O$i$k-I
n = 0 mod p; hence, crtp-Icr- I = id. It follows that t p- I =id, which is
absurd. Thus, we have a = 1 and crtcr- I = t b, i.e. crt = tbcr. For every
i E E, we thus have cr(i + 1) = cr(i) + b, hence cr = to"(O)mb which indeed
lies in GA(p).
5) Let G have the two properties. By 3), there exists <p such that <pG<p-1
contains t. As <pHr_I<P-1 = (t) is a normal subgroup of <pHr _ 2<P-I, part 4)
shows that <pHr _ 2<P-1 C GA(p). For the same reasons, <pHi<P-1 C GA(p)
for 0 ::; i ::; r - 1, which gives the result.
6) a) If G is solvable, it is conjugate to a subgroup r of GA(p). As the

only element of GA(p) having at least two fixed points is the identity, the
same holds for G.
b) For every i E {O, ... ,p - I}, let O(i) denote the orbit of i under
G, and SCi) the stabilizer of i; set q = IS(O)I. Because G is transitive,
O(i) = {O, ... ,p -I} and IGI = IO(i)IIS(i)1 show that IS(i)1 = q for every
i. By the hypotheses, G is the disjoint union of the SCi) - {id} , together
with the set A of elements having no fixed points, and {id}. Thus IAI =
pq - p(q - 1) - 1 =p- 1, which proves the existence of 7.
Let n be the order of 7. For every k < n, 7 k E A since if 7 k fixes i, it
also fixes 7(i), which is impossible. Thus, the orbits under the action of (7)
have the same cardinal n. As {O, ... ,p - I} is the disjoint union of these
orbits, we see that n divides p. Consequently, n = p, 7 is a p-cycle and
Au {id} = (7).
Up to conjugation, we can assume that 7 = t. If a E G, ata- 1 has no
fixed points (otherwise t would), so ata- 1 E (t) and by 4), we show that
G C GA(p). Thus, G is solvable.
12
Solvability of Equations by Radicals
Using the correspondence constructed in the preceding chapters, together

with group-theoretic results, Galois obtained his famous criterion of solv-
ability by radicals. "This material is so entirely new that new names and
new characters are necessary to express it," he wrote, adding later that the
true value of his criterion is essentially theoretical, as it is often impossible
to compute the Galois group of a given polynomial: "In a word, the compu-
tations are not practicable." However, he adds, the applications generally
lead to "equations all of whose properties are known beforehand," so that
the computations are possible, as in Chapters 9 and 10.
12.1 Radical Extensions and Polynomials Solvable

by Radicals
12.1.1 Radical Extensions
DEFINITION. - A field Lee is said to be a radical extension of a field K
if there exists a tower (Ki)O~i~r, called a radical tower, of extensions of K
such that for each i = 0, ... , r -1, KHI is an extension of Ki by a root of
an element of K i . In other words, we have
hence, for each i = 1, ... , r, there exists an ni such that (ai)n i E K i - 1 .

208 12. Solvability of Equations by Radicals
EXAMPLE. - {j
'\/ 1 + P + V-5 belongs to a radical extension of Q
(this formula actually defines 360 distinct complex numbers!)
12.1.2 Polynomials Solvable by Radicals

DEFINITION. - A polynomial P(X) E K[X] is said to be solvable by radicals
over K if there exists a radical extension L of K containing the splitting
field N of P.
COMMENTARY. - All of the elements of L can be written using radicals,

but L can contain N strictly. We require that all of the roots of P lie in
L; in the case of a polynomial irreducible over K, we can show that it is
equivalent to require that one root of P lie in L (see, for example, the book
by J.-P. Tignol, p. 345 and following).
OUTLINE OF THIS CHAPTER. - We will show in §12.2 that if P(X) is a

polynomial of K[X] which is solvable by radicals over K, with splitting field
N over K, then Gal(N[K) is a solvable group. For this, we first need to
construct a suitable radical extension of K (following §12.1). We will then
show the converse of this fundamental result, in §12.4.
12.1.3 First Construction

Let us start from a radical extension L of K, containing N, defined by a
radical tower T = (Ki)o<i<r with notation as in §12.1.1. Let us define the
tower T' = (KDo:Si:Sr+~ by Kb = K, K~ = K[(] where ( is a primitive
n-th root of unity, and n = lcm {ni; 1 :S i :S r}, K:+ I = K[(, aI, ... ,(!,] for
1 :S i :S r.
In this situation, KI+I is, by §1O.2, a normal extension of KI for i =
0, ... , r, but we do not know if L' = K;+1 is a normal extension of K. For
example, the radical tower Q c Q[ v'2] C Q[ {12] is not in this case. Let us
show that we can define a tower Til that also satisfies this last property.
12.1.4 Second Construction

Let Pi be the minimal polynomial of ai over K for i = 1, ... , r, and let Ai be
the set of conjugates in C ofai over K. The extension L" = K[(, All ... ' Ar]
is a normal extension of K, since it is the splitting field of (xn -1) II Pi.
l:Si:Sr
Let us show that it is a radical extension of K of the same form as T'. It
suffices to show that for every i with 1 :S i :S r, and every a E Ai, we have
ani E K [( , Uj<iAj].
Because a is conjugate to ai over K, there exists a K-homomorphism a:
K[aiJ ----> C such that a(ai) = a and we can extend a to a K-homomorphism
12.2 If a Polynomial Is Solvable by Radicals, Its Galois Group Is Solvable 209
u': K[(, Uj:'OiAj] ~ <C such that u'(ai) = a. Since K[(, Uj<iAj] is a
normal extension of K (because it is the splitting field of (xn -1) II Pj )
l:'Oj<i
and (ait' belongs to it, the same holds for an, which is conjugate to it,
since an, = (u'(ai))n, = u'((ai)n,).
Note that we can refine the tower (K [( , Uj<iAj]) in such a way
- l<i<r
that the successive extensions are abelian (this is the caSe-for K[(] over K,
by §9.5) or cyclic (this is the case for extensions by each conjugate of ai,
by §10.2).
12.2 If a Polynomial Is Solvable by Radicals, Its

Galois Group Is Solvable
THEOREM. - Let K be a field contained in <C, and let P be a polynomial in
K[X] with splitting field N. If P is solvable by radicals, then Gal(NIK) is
a solvable group.
PROOF. - We just saw that there exists a radical extension L of K con-
taining N, defined by a tower (Ki)o<i<s such that Ko = K, Kl = K[(]
is an abelian extension of K, and Ki+1 = Kdxi] is a cyclic extension of
Ki for 1 ::; i ::; s - 1, with (Xi)n, E K i , ( a primitive n-th root of unity,
n = lcm {ni; 1 ::; i ::; s - I} and L = Ks a normal extension of K.
Set G i = Gal(LIKi ). The sequence (Gi)OS;i:'Os satisfies the properties of
§11.1. Indeed, condition 1) is clear; by §8.5 3), G i +1 is a normal subgroup
of G i , and as GdGi+l c:::'. Gal(Ki +1IKi ) is abelian, conditions 2) and 3) of
§11.1 are satisfied. Gal(LIK) is thus a solvable group. By §11.4, the same
holds for the group Gal(NIK), which is a quotient of Gal(LIK) by §8.4. <>
12.3 Example of a Polynomial Not Solvable by

Radicals
PROPOSITION. - The polynomial P(X) = X 5 - lOX +5 E Z[X] is not
solvable by radicals.
PROOF. - By the preceding theorem, it suffices to determine its Galois
group and show that it is not a solvable group. Let N denote the splitting
field of P in <C, and set G = Gal(NIQ). P is an irreducible polynomial by
Eisenstein's criterion. It has three real roots and two complex conjugate
roots, which we call a and b; this can easily be determined by studying the
sign of the derivative of P.
Complex conjugation on C induces an element of G that exchanges a

and b and fixes the three other roots of P. Because IGI = [N : Q] = [N :
Q[a]][Q[a] : Q], the group G has order a multiple of 5, which by Cauchy's
theorem (see Exercise 8.2) implies the existence of an element (J' of order 5
in G. The group G is then identified with a subgroup of S5 that contains
a 5-cycle and a transposition. Thus, we have G ~ S5 (see Exercise 11.1.2)
e)). As we know that S5 is not solvable by Theorem 11.6.1, we can apply
Theorem 12.2. (:;
12.4 The Converse of the Fundamental Criterion

PROPOSITION. - Let K c C be a field, and let P be a polynomial in K[X]
with splitting field N over K and Galois group G = Gal(NIK). Assume
that K contains a primitive n-th root of unity ( with n = [N : K] = IGI· If
G is a solvable group, then P is solvable by radicals.
PROOF. - By §11.5, there exists a finite decreasing sequence (Gi)O:<:::i:<:::r of
subgroups of G such that
1) Go = G ::J ... ::J G r = {e} ;
2) Gi+l is a normal subgroup of G i for 0 :S i :S r - 1;
3) GdGi+1 is cyclic of prime order Pi for 0 :S i :S r - 1.
Set Ki = I(G i ) for 0 :S i :S r; we have Ko = K, and by §8.4, we also have

Gal(Ki+IIKi) ~ Gal(NIKi)/Gal(NIKi+I) ~ GdGi+l for 0 :S i :S r - 1.
Thus, Gal(Ki+IIKi)) is cyclic of prime order, and by §1O.5, there exists an
element ai E Ki+l such that (ai)pi E Ki and Ki+l = Kdai]. This concludes
the proof. (:;
12.5 The General Equation of Degree n

12.5.1 Algebraically Independent Elements
DEFINITION. - Let K be a field, and let L be an extension of K. Let
Xl, ... , Xn E L; these elements are said to be algebraically independent over
K ifthe following homomorphism is injective: we define f : K[Xl, ... , Xn] ----+
L by taking flK to be the inclusion of K into L, and setting f(Xi ) = Xi
for i = 1, ... , n. In other words, Xl,"" Xn are algebraically indepen-
dent if there exists no non-zero polynomial P E K[X I , ... ,Xn] such that
P(XI, ... ,xn) = O.
12.5 The General Equation of Degree n 211
12.5.2 Existence of Algebraically Independent Elements

PROPOSITION. - For every integer n, there exist n algebraically independent
complex numbers over Q.
PROOF. - Let us start with Q, the field of rational numbers. It forms a
countable set; its algebraic closure (see §14.1) C(Q) inside <C is also count-
able (check this); so its complement is non-empty. Choose an element Xl
in the complement. The extension of C(Q) by Xl is again countable; its
algebraic closure C(C(Q)(XI)) inside <C is again countable, so again its
complement is non-empty; choose an element X2 in this complement. Con-
tinuing in this way, we obtain any number of algebraically independent
elements. We can even choose the Xi to be real. <>
12.5.3 The General Equation of Degree n

Let XI, ... ,X n be algebraically independent elements over Q, and let N =
Q(XI,'" ,xn ) be the field generated by these elements over Q. Let P be
the monic polynomial defined by
P(X) = II (X - Xi).
l-:;i-:;n
The equation P(X) = 0 is called the general equation of degree n over

Q. Set
P(X) = xn + L akXk and K = Q(ao, ... , an-I).

O-:;k-:;n-l
12.5.4 The Galois Group of the General Equation of Degree n
PROPOSITION. - 1) The Galois group Gal(NIK) of the general equation of

degree n is isomorphic to Sn.
2) The general equation of degree n is not solvable by radicals for n 2: 5.
PROOF. - 1) Every permutation s E {I, ... , n} induces a Q-automorphism
a of N defined by a(xi) = xs(i) for 1 :S i :S n; thus, we can consider Sn as
a group of Q-automorphisms of N. Set L = J(Sn). Emil Artin's theorem
shows that [N: L] = n!. Furthermore, as ak E L for 0 :S k :S n, by §3.2.2,
we see that K c L. Finally, as N is the splitting field of P over K, we have
[N : K] :S n!.
We obtain L = K, so
Gal(NIK) = Gal(NIL) ~ Sn.
2) It suffices to apply Theorem 12.2.

COMMENTARY. - Figure 12.1 shows the beginning, written in an astonish-

ingly modern style, of Abel's first memoir on the impossibility of resolving
the general equation of degree five. He quickly extended this result to the
general equation of arbitrary degree greater than or equal to five (in 1826).
These results correspond to part 2) of Proposition 12.5.4. This impossibility
had actually already been proved, in 1799, by Ruffini, but his enormously
long proof had not convinced his colleagues, even after several modifications
and simplifications.
MEMOIRE SUR LES EQUATIONS ALGEBRIQUES, OU L'ON DEMONTRE

L'IMPOSSIBILlTE DE LA REsOLUTION DE L'EQUATION GENERALE
DU CINQUIEME DEGRE.
Broehate lmprimH chez Grimdabl, Christiania 18U.
Le., geom~tres se sont beaucoup occupes de 1& resolution generale des

equations alge1>riques, et plusieurs d'entre eux ont eberche a en prouver
l'impossihilitl!; mais si je ne me trompe pas, on n'y a pas reussi jusqu'a
present. J'ose done esperer que les geom~trea recevront avec bienveillance
ee memoire qui a pour but de remplir eette laew,e dans 1& tbl!orie des
equations aIgebriques.
Soit
y' - ay'+by' - cy'+ dy - e=O
l'equation generale du cinquime degre, et supposons qn'elle soit resoluble
algebriquernent, e'est-a-dire qu'on puisse exprimer y par une fouction des
quauti~ a, b, C, d et e, fonnee par de.. radicaux.
FIGURE 12.1. The beginning of Abel's first memoir

Exercise 12.1. Examples of polynomials not solvable by radicals
over Ql
1) Show that the following polynomials are not solvable by radicals over
Ql.
a) X5 - 14X + 7;
b) X5-7X2+7;
c) X7 - lOX5 + 15X + 5.
2) Let P be an irreducible polynomial of Ql[X] of prime degree p ~ 5.
Assume that P has exactly two non-real roots. Show that P is not
solvable by radicals.
Exercise 12.2. Cubic radical extensions

1) Let K be a field contained in C, and let P(X) = X 3 + pX + q be
an irreducible polynomial in K[X]. Let x be a root of P, and let
D = d 2 = _4 p 3 - 27 q2 be the discriminant of P. Take an element
u = a + bx + cx 2 of K[x] with not lying in K, of minimal polynomial
X 3 + a' X2 + b' X + c' over K.
a) Determine a' and b' in terms of a, b, C,p, q.
b) Show that K[x] is a radical extension of K if and only if -3D

is a square in K.
c) Extend the condition of part b) to the case where x is a root of

an arbitrary irreducible polynomial P(X) = X 3 + aX2 + bX +
cEK[X].
2) Is the extension Q[cos(21r/7)] of Q radical?
3) Let Xl, X 2, X3 be indeterminates, and let Sl, S2, S3 be the elementary

symmetric polynomials in these indeterminates.
a) Show that Q(X I , X 2, X 3) is not a radical extensionofQ(sl, S2, S3)'
b) Show that Q[j](X1 , X 2, X 3) is a radical extension OfQ(Sl, S2, S3).
Exercise 12.3. A result of Galois

In this problem, we will make use of the results on transitive and
solvable subgroups of Sp given in Exercise 11.4.
1) Let K be a field contained in C, P an irreducible polynomial of prime

degree p ~ 5 in K[X], E the set of roots of Pin C, and N the splitting
field of P over K.
a) Assume that P is solvable by radicals. Show that N = K[x, y]

for every pair of distinct elements x and y E E.
b) Show the converse.
2) Let K c R Deduce from the above that a polynomial in K[X], of

prime degree greater than or equal to five and having exactly two real
roots, is not solvable by radicals.
Exercise 12.4. Irreducible cubic equations: the necessity for non-

reals
Consider the polynomial P(X) = X 3 + pX + q with p and q real.
Set K = Q(p, q), let a, b, and c denote the roots of P, and let D =
_4p 3 - 27q2 be its discriminant. Assume that P is irreducible over
K and that D > 0, and let d be a real number such that d2 = D.
In this case, the equation P(x) = 0 has three real roots; however,
Cardan's formulas involve a square root of a negative number, i.e. a
non-real number (see Exercise 2.4). Let us show that this cannot be
otherwise. In order to show it, we first assume the contrary.
Suppose there exists a radical tower defined by a sequence
of strictly positive real numbers and a sequence PI, ... ,Pn
UI, ... , Un
of prime numbers such that for 1:::: i :::: n, (Ui)P; E K[ PVUl,···, P;--vtUi-l]
and a E L = K[ PVUl, ... , p~.
1) Show that K[a, b, c] C L[d].

Define the sequence of fields Ko = K, KI = K[d], ... , Ki+1 =
K[d, P-vtul, ... , ~ for 1 :::: i :::: n, and let r be the smallest in-
dex i such that i such that a E Ki+I.
2) Show that r > O.
3) Show that K[a, b, c] C Kr+1 and that P is irreducible over K r .
4) Show that Pr = 3, and then that K r+1 is a normal extension of K r .

5) Deduce a contradiction.
COMMENTARY. - Otto Holder was the first to clarify the problem posed
by Cardan's formulas in 1891.

1) Use the arguments of §12.3 in the three cases.
2) The Galois group G of P contains the Q-homomorphism induced by

complex conjugation, and because P is irreducible, IGI is a multiple of p.
Identifying G with a subgroup r of Sp, we see that r contains a transpo-
sition and an element of order p, i.e. a p-cycle. Consequently, r = Sp (see
part 2) of Exercise 11.1) and G ':::; Sp. Now we can conclude, using §12.2
and §1l.6.
1) a) As u does not lie in K, it is of degree 3 over K. If x, y, z denote

the roots of X 3 + pX + q in C, then the conjugates of u over K are u,
v = a + by + cy2, W = a + bz + cz 2, and the coefficients of the polynomial
(X - u)(X - v)(X - w) lie in K.
Using computations on symmetric polynomials, we find
a' -(u+v+w)=-3a+2pc,
b' uv + vw + wu = 3a 2 + pb 2 - 4pac + p 2c2 + 3qbc.
b) In order for K[x] to be a radical extension of K, it is necessary and

sufficient that there exist an element u = a + bx + cx 2 E K[x] such that
K[x] = K[u], u 3 E K, i.e. such that bore -# 0 and a' = b' = O. As a' = 0 if
and only if a = 2pc/3, these conditions are equivalent to _(p 2c 2/3) + b2p +
3bcq = O. This equation has a non-zero solution (b, c) in K2 if and only if
the discriminant of pX 2 + 3qX - (p2/3) is a square in K, which gives the
result. Note that u is not unique.
c) By a variable change of the form Y = X + a, we transform P into a
polynomial of the form X 3 + pX + q with the same discriminant D as P;
we then recover the preceding conditions.
2) The minimal polynomial of 2 cos(27r /7) is given by X 3 + X2 - 2X - 1

(see Exercise 2.6), which becomes y3 - (7/3)Y - (7/27) after setting Y =
X + 1/3. Its discriminant is D = 49. We deduce that Q[cos(27r/7)] is a
non-radical extension of Q.
3) a) Set d = (Xl - X 2 )(X2 - X 3 )(X3 - Xl) and D = d 2 . We know

that the Galois group Gal(Q(X l , X 2 , X 3 ) IQ(8l, 82, 83)) is isomorphic to S3,
so the intermediate extensions are Q(8l' 82, 83)[d] and Q(8l, 82, 83) [Xi] for
i = 1,2,3. As the quadratic extensions are radical, we still need to ex-
amine the case of extensions of degree 3. The extension Q(Xl, X2, X 3 ) of
Q(81, 82, 83)[d] is not radical since -3D is not a square in Q(8l, 82, 83)[d]
(indeed, D = d 2 is a square and -3 is not a square). Let us now show
that the extension Q(81' 82, 83) [Xi] of Q(8l' 82, 83) is not radical either. If
it were, -3D would be a square in Q(8l, 82, 83), so D would be a square in
C(8l,82,83), and we would have
which is false.
b) In this case, -3 is a square, which gives the conclusion.
The proof of course uses the Galois correspondence. We need to trans-

late the property in the text into a condition on the Galois group G =
Gal(NIK), or rather on the isomorphic subgroup of Sp obtained start-
ing from an indexation of the roots as in §8.1.4. If we take a bijection
r.p : {I, ... ,p} -+ E with Xi = r.p(i) , we obtain an injective group homo-
morphism : G -+ Sp such that s = (a) is the permutation given by
a(xi) = Xs(i)'
1) a) If P is solvable by radicals, we know that (G) is conjugate to a

subgroup H of GA(p). Now, an element of GA(p) that has two fixed points
is the identity. Consequently, an element of G which fixes two roots is the
identity; in other words, Gal(NIK[x, y]) = {id} for every pair of distinct
elements X and y E E. Thus N = K[x, y].
b) We know that ( G) is transitive and the fact that N = K[x, y] for
every pair of distinct elements x and y E E is equivalent to the condition
in (G) that a permutation that fixes two elements of {I, ... ,p} is the
identity, hence the result (see Exercise 11.4).
2) If a and b denote the two real roots of P, then we cannot have N = K[a, b]
(with notation as above), since K[a, b] C lR. and N ct. lR.. The result then
follows from the first question.
COMMENTARY. - The result of this exercise is due to Galois, who stated it

as follows: "For an equation of prime degree, which has no commensurable
divisors, to be solvable by radicals, it is NECESSARY and SUFFICIENT
that all the roots be rational functions of any two of them."
1) We know that K[a,d] = K[a,b,c]. As K eLand a E L, we have

K[a,d] C L[dj, hence K[a, b, c] C L[d].
2) r = 0 means that a E Kl = K[d]; this is impossible since a is of degree
3 over K and d is of degree 2 over K.
3) As a and d lie in K r+1 , we have K[a, b, c] C K r+1 by 1).

Because Kr contains d, if it contains a root of P, then by 1), it must
contain them all. But this contradicts the definition of r. Thus, P has no
roots in Kr; as it is of degree 3, this suffices to show that it is irreducible
over K r .
4) As Kr+l ::J Kr[a, b, c] and Kr[a, b, c] is of degree 3 over K r , we have

Pr = 3 and Kr+l = Kr[a, b, c]. K r+1 is a normal extension of K r .
5) Set U = U r . Because Kr+l is a normal extension of K r , and it contains

one root of X 3 - U, it must contain the other roots, namely j qu and j2 qu.
But this contradicts Kr+l c R
13
/
The Life of Evariste Galois
The life of Evariste Galois is the most famous, fascinating, and commented
life of any mathematician. It has even become something of a myth, like
the lives of the immortal poets Rimbaud, Byron, or Keats.
Our knowledge of Galois' life contains enough gaps to allow imagination
(and historians of science) to flourish. The books by Bourgne and Azra
listed in the bibliography denounce some dangerous hypotheses and out-
right errors; see also their edition of Galois' complete works, containing
portraits, reproductions of his writing, and all of his extant articles.
Evariste Galois was born on October 25, 1811, in Bourg-la-Reine, a town
(renamed Bourg-l'Egalite during the Revolution!) located about 10 km
south of Paris. His father, Nicolas Gabriel Galois, was a political liberal
and the mayor of the commune during the Hundred Days; his strong per-
sonality assured him re-election under the Restoration. Apart from acting
as mayor, he was also the director of a school. Galois' mother, Adelalde
Marie Demante, was the daughter of a magistrate. She appears to have
played an important role in the education of her young son, particularly in
the domain of Latin culture.
At the age of 12, Evariste entered the royal school of Louis-Ie-Grand.
He was a brilliant student, but his teachers also commented that he had
"somewhat bizarre manners" and was "rebellious" ... (is this really extraor-
dinary?) In October 1826, he entered the advanced rhetoric class, but at
the beginning of the second trimester, he was demoted to the previous year
because of his mediocre performance. At that time, all study was heavily
based on classical culture, and sciences were studied only as extra work;
this actually represented a regression with respect to the Napoleonic and

220 13. The Life of Evariste Galois
Revolutionary periods, during which mathematics teaching played a fun-

damental role in education.
Galois discovered mathematics thanks to his demotion, because he was
allowed to take the extra courses. Alone, he read the whole of Legendre's
Elements of geometry and Lagrange's texts on the resolution of equations,
as well as works by Euler, Gauss, and Jacobi. In 1827, he obtained first
prize in a national mathematics competition (the Concours General, which
still exists today); the following year, he obtained an accessit (awarded
to the best papers after the first, second, and third prizes) in the same
competition, as well as two successive accessits in Greek.
In 1828, he attempted the entrance examination for the Ecole Polytech-
nique, but failed, and remained at Louis-Ie-Grand, in the advanced math-
ematics class. His teacher, Monsieur Richard, was 33 years old; he soon
came to deeply admire the genius he perceived in his student. He kept
all of Galois' homework, and later gave them to another student of his,
Charles Hermite. He encouraged Galois to publish his first research results;
an article appeared on April 1, 1829, in the Annales de mathimatiques, the
journal founded by Joseph Gergonne.
It was at this point that difficulties and dramatic events began to accu-
mulate around Galois. An article he sent to the Academy of Sciences was
delivered to Cauchy and lost (Cauchy had already lost an article by Abel).
On July 2, Galois' father Nicolas committed suicide, unable to endure
the attacks of the cure of Bourg-la-Reine, who among other things wrote
a series of anonymous letters that he attributed to no other than Nicolas
Galois! His funeral was the scene of a small riot.
A few days later, Galois once again attempted the entrance examination
for the Ecole Polytechnique. It was catastrophic; to the stupefaction of his
professor, Galois failed. One of the examiners, either Dinet or Lefebure de
Fourcy (probably Dinet), asked a question about logarithms which Galois
found too simple or even stupid; it is said that he flung the blackboard
eraser in the examiner's face. The mathematician Joseph Bertrand (born
in 1822) denies this incident, but mentions the "crazy laughter of the gentle-
men of the jury examining the candidates to the Ecole Polytechnique (who
do not, to my surprise, each occupy a chair at the Academy of Sciences,
because their place is certainly not in posterity)."
Following the advice of his professor, Galois entered the Ecole Normale.
At that time, this school was known as the Ecole Preparatoire, and was
considered to be on a much lower level than the Ecole Polytechnique. While
there, he wrote the results of his research and presented them for the Grand
Prize in Mathematics of the Academy of Sciences. Fourier (the Fourier of
Fourier series, the theory of heat, the expedition to Egypt, etc.) took Galois'
manuscript home with him, but died shortly afterward. The manuscript is
now lost, but part of the results contained in it appeared in the Bulletin
des sciences mathematiques published by the Baron de Ferussac, in April
13. The Life of Evariste Galois 221
and June 1830. In the end, the Grand Prize was awarded to Abel (who had
died the previous year) and Jacobi.
Galois' political opinions appear to have evolved very rapidly, and from
that point on, he began to live a political life as intense as his mathematical
one. During the famous days of July 27,28, and 29, 1830, he and the other
students were locked into their school to prevent them from participating
in the action outside, while the students at the Ecole Polytechnique battled
on the barricades and made history. By October 1830, at the beginning of
the academic year, Galois had turned into an active, ardent, and intrepid
Republican, ready to defend the "rights of the masses", according to one
member of his family. He joined the Society of Friends of the People on
November 10 and openly criticized the opportunism of the director of the
Ecole Normale and the philosopher Victor Cousin, who from faithful fol-
lowers of Charles X had become faithful followers of Louis-Philippe. He
mingled criticism of their teaching with his political criticism and found
himself indefinitely suspended.
The last mathematical article published in Galois' lifetime appeared on
December 1. On December 5, he apparently authored a long letter published
in the Gazette des Ecoles, signed "a student at the Ecole Normale", in
which the director is derided in the following terms: "Everything in him
announces the narrowest ideas and the most absolute routine." In early
January, Galois was expelled by an exceptional decision of the board.
On January 2, 1831, a letter appeared (once again in the Gazette des
Ecoles) , entitled On the teaching of the sciences, subtitled Professors. Books.
Examiners. In the letter, Galois denounced the mediocrity of the teaching
available to students: "When will students be given time to meditate on
the mass of acquired knowledge .... Why do the examiners ask questions in
a twisted way? It would seem that they are afraid to be understood by
those who are being examined ..... Do they fear that science is too easy?"
With no income, Galois opened a public course on higher algebra, on Jan-
uary 13, at the Caillot bookstore, 5 rue de la Sorbonne. The course probably
did not last long. The advertisement, which appeared in the Gazette des
Ecoles, ran as follows: "This course will take place every Thursday at 1:15;
it is aimed at young people who, feeling the incompleteness of algebra as
taught in the colleges, wish to study the subject more deeply. The course
will consist of theories, some of which are new, and none of which has ever
been lectured on in public. We mention here a new theory of imaginaries,
the theory of equations solvable by radicals, the theory of numbers and
elliptic functions treated by pure algebra." Thirty people attended the first
lecture.
The academician Denis Poisson advised Galois to write a new version
of the memoir that had been presented a year earlier to Fourier and lost.
On January 17, the Academy gave Poisson the task of reading the new
manuscript, together with Sylvestre Lacroix. On March 16, Galois wrote
to the academicians, pressing them to read his manuscript. Meanwhile,
political tensions ran high, as Louis-Philippe had managed to maneuver

the Republicans out of power (in December 1830, he dissolved the National
Guard, of which Galois was a member).
On May 9, 1831, after the acquittal of some young Republicans on trial,
Galosi attended a banquet in the salon of the restaurant Aux Vendanges
de Bourgogne; Alexandre Dumas and Fran<;ois-Vincent Raspail were also
present. During the banquet, Galois proposed an unplanned toast. "To
Louis-Philippe!" he said, raising his glass in one hand - ''the fumes of the
wine had removed my reason," he later explained - and a knife in the
other. Those who did not see the knife protested, and during the following
moments of effervescence, Alexandre Dumas became frightened and fled.
The next day, Galois was arrested at his mother's home, and sent to the
Sainte-Pelagie prison (near the Jardin des Plantes), to be judged on June
15.
The full text of the trial still exists. Galois' testimony reads: "Here are
the facts. I had a knife which I had been using to cut food during the meal.
I raised the knife, while saying To Louis-Philippe, if he betrays. The last
words were heard only by my nearest neighbors, because of the whistles
excited by the first part of my toast." The jury was indulgent, and Galois
was acquitted.
An anonymous article published in the journal Le Globe mentions Galois'
mathematical genius. On July 4, Poisson and Lacroix finally published their
report on Galois' memoir, writing "We have made every effort to under-
stand Galois' proof. His reasoning is neither sufficiently clear nor sufficiently
developed to allow us to judge of its correctness ... We can wait until the
author has published his entire work in order to form a definitive opinion."
They conclude that, for the moment, "we cannot propose approval of this
memoir."
Galois was disappointed. But Poisson and Lacroix were not entirely mis-
taken; the text is extremely difficult to understand, and the author could
certainly have provided additional explanations.
On July 14, Galois was arrested on the Pont-Neuf, at the head of a large
group of demonstrators. He was again sent to Sainte-Pelagie, and this time
condemned, on October 23, to six months of prison because he was not a
first offender. His friend Ernest Duchatelet, who was arrested together with
him, was condemned to just three months.
Galois continued to work in prison, where he also socialized with people
like Nerval and Raspail.
In December, he submitted a new effort for publication. But the pref-
ace he wrote for his article was so polemical that the complete text ended
up being published only in 1948, by Rene Taton. Bitter over the loss of
his manuscripts and Poisson's lack of understanding, he violently attacked
politicians and scientists, placing them on the same level: "If I had to ad-
dress some words to the great men of this world or the great men of science ...
I swear that it would not be thanks." In the second part, he analyzes the
procedures he used to construct his theory, emphasizing that computations,

even very elegant ones, have their limits, as do algebraic transformations
to such a point that in order to do them, they must have been foreseen. He
adds: "Jumping into the computations, grouping the operations, classifying
them according to their difficulties and not according to their form: this,
according to me, is the mission of future geometers [mathematicians]; this
is the road I have followed in this article." To conclude, he emphasizes what
he was unable to understand, and dreams of a time in which egoism will
no longer reign in the sciences, and "people will study together, instead of
sending sealed letters to the Academy; people will hasten to publish even
their smallest observations, if they are new, and will not hesitate to add 'I
do not know the rest.' "
Because of the cholera epidemic of early 1832, Galois was transferred on
March 16 to a pension or sanatorium run by a certain Faultrier, near the
Place d'Italie, along the Biiwre River, not far from the Croulebarbe Mill.
At the time, this was located in the commune of Gentilly; later, it became
part of the 13th arrondissement of Paris. In theory, Galois was scheduled to
be set free on June 1, but it seems certain that he left prison earlier. In May
1832, he had a brief love affair with a young woman, Stephanie D., whose
true identity is still under discussion. He broke it off on May 14, and this
appears to have been the cause of the duel he fought a few days later. The
night before the duel, on May 29, Evariste assembled his latest discoveries
in a splendid letter, addressed to a faithful friend, Auguste Chevalier. It is a
dramatic scene to imagine. Foreseeing his death, pressed by time, he wrote
this letter containing a summary of his mathematical work, in desperate
urgency.
Paris, May 29, 1832
My dear friend,
I have done several new things in analysis.
Some of these things concern the theory of equations, others concern
integral functions.
In the theory of equations, I looked for conditions for the equations
to be solvable by radicals ....
He recalls the set of results that he obtained, concluding seven pages later
with an obscure sketch of some notions later created by Riemann (many-
sheeted Riemann surfaces are referred to as ''the theory of ambiguity").
My main meditations for some time now have been directed towards
the application of the theory of ambiguity to transcendental analysis ...
But I do not have time now and my ideas on this immense terrain are
not yet well developed ...
You will publicly request Jacobi or Gauss to give their opinions, not
on the truth but on the importance of these theorems.
After that, I hope there will be people who find profit in attempting
to decipher this mess.
Rereading or wishing to modify one statement, he wrote in the margin:
"There is something to be completed in this proof. I have no time."
He wrote other short letters, for example: "I am dying, the victim of an
infamous coquette, and two fools of this coquette. My life is extinguished
in a miserable cancan. Oh ! why die for so little .... Adieu! I had a lot of
life for the public good."
The exact circumstances of the whole adventure are not known, nor is
the name of his adversary. In the morning of the May 30, Galois, grievously
wounded and abandoned after the duel, was picked up by a peasant and
carried to the Cochin hospital where he died of peritonitis on May 31, in the
arms of his young brother Alfred, saying "Don't cry, I need all my courage
to die at twenty." He was buried in the mass grave of the Montparnasse
cemetery.
Accounts of his death appeared in a few newspapers, but they gave con-
tradictory details. His friends organized a demonstration, which they post-
poned on hearing of the death of General Lamarque; it took place on June
5 and led to the massacre of the Saint-Merri cloister. Victor Hugo recounts
the event in the chapter The epic of the rue Saint-Denis of his book Les
Miserobles.
Thanks to the devotion of Auguste Chevalier and his brother, the papers
of Evariste Galois were collected, and his letter-testament was published in
September 1832. It did not, however, attract any attention, even though
it was published by Auguste Chevalier together with a presentation of the
life of his friend ("A second condemnation threw him behind bars for six
more months. Death awaited him at the exit"), to which Nerval added, in
1841: "He was killed in a duel the day after he was given his freedom." The
romantic myth surrounding Galois was born from these lovely writings.
In 1835, Lacroix mentioned, in a note near the end of the sixth edition
of his Complements of elements of algebro Galois' memoir, which he had
read together with Poisson, saying:
In 1828, Abel wrote to Legendre: "I have been happy enough to find
a sure rule for recognizing if a given equation is by radicals or not. A
corollary of my theory shows that generally, it is impossible to resolve
equations superior to the fourth degree." (Journal de Grelle, year 1830,
1st cahier, p. 73.) This discovery was announced by Legendre to the
Academy of Sciences on February 23, 1829; but Abel did not publish
anything on the subject, and nothing to do with it was found in his
papers ....
In 1831, a young Frenchman, Evariste Gallois (sic), who died the
following year, announced, in a memoir presented to the Academy of
Sciences, that 'for an irreducible equation of prime degree to be solv-
able by radicals, it is necessary and sufficient that given any two of its
roots, the others can be deduced from them rationally', but his memoir
appeared practically unintelligible to the commissaries who examined

it.
On September 4, 1843, Liouville announced to the Academy of Sciences
that he just discovered, in the papers of Galois, a solution "as precise
as it is deep" of the problem of the solvability of equations by radicals
(Comptes rendus hebdomadaires des seances de l 'Academie des sciences,
voL 17, pp. 448-449). Wishing, undoubtedly, to understand it better, he
waited until October 1846 before publishing the texts by Galois, in his
Journal de mathematiques pures and appliquees, with no commentary at
all.
The truncated version of Galois' life presented by Liouville (who defended
the institutions attacked by Galois, condemned his political activities, etc.)
was the accepted version for 50 years, until a more precise version, based
on forgotten and rediscovered documents, was published by Paul Dupuy in
the Annales de l'Ecole Normale Superieure.
In the 1850s, the texts of Galois' memoirs finally became accessible to
mathematicians. They initiated a great deal of work by Serret, Betti, Kro-
necker, Dedekind (who taught from them in G6ttingen, in the winter of
1857-1858), Cayley, Hermite, Jordan (see his Traite des substitutions from
1870), etc. Their importance in the development of 20th century mathe-
matics is immense.
Figure 13.1 is a reproduction of the portrait of Evariste Galois, drawn
long after his death, from memory, by his brother Alfred, and published in
Le magasin pittoresque in 1848.
FIGURE 13.1. Portrait of Evariste Galois by his brother Alfred, 1848

14
Finite Fields
In this chapter, we drop the assumption that the fields we consider are
subfields of C. We will make use of analogues of some of the definitions
and results of previous chapters, which adapt to the case of finite fields;
we do not always give the new proofs for these results. Note, however, that
Theorem 14.1.3 proves the existence of an algebraic closure for each of the
fields we will study; it plays a role analogous to that of C in the previous
chapters. Fields of characteristic 2 are a particularly exciting subject of
current research.
14.1 Algebraically Closed Fields

In this section, we consider arbitrary finite or infinite commutative fields
of arbitrary characteristic (see §14.3).
14.1.1 Definition
A field C is said to be algebraically closed if every polynomial of degree
greater than or equal to 1 in C[X] factors into a product of linear factors
or, equivalently, if every polynomial of degree greater than or equal to 1 in
C[X] has at least one root in C.
EXAMPLE. - The field C of complex numbers is an algebraically closed
field, by d'Alembert's theorem (see Exercise 7.4).

228 14. Finite Fields
14.1.2 Algebraic Closures

DEFINITION. - Let K be a field. An algebraic extension G of K which is
algebraically closed is called an algebraic closure of K.
EXAMPLES. - 1) te is an algebraic closure of JR.

2) Q has an algebraic closure that is a subfield of te, namely the union
of all algebraic extensions of finite degree of Q contained in C.
14.1.3 Theorem (Steinitz, 1910)

Let K be a field.
1) There exists an algebraic closure G of K. It is unique up to a non-

unique K -isomorphism.
2) Let L be an algebraic extension of K, not necessarily contained in

G, and let (J" : K ---> G be a homomorphism. Then there exists an
extension of (J" to L.
REMARK. - Note that a finite field K cannot be algebraically closed since

the polynomial 1 + II
(X - a) has no roots in K.
aEK
PROOF. - Let us sketch a proof (see the book by A. and R. Douady listed
in the bibliography for more details).
1) Let E be a set containing K, with sufficiently large cardinal. The
algebraic extensions of K whose elements are in E form an ordered set,
by extension, which is inductive. By Zorn's lemma, this set contains a
maximal element G. Assuming that G is not algebraically closed leads to
a contradiction. If K is finite, we can even avoid having recourse to Zorn's
lemma, replacing it by a direct construction.
The isomorphism of two algebraic closures G and G' of K is a conse-
quence of part 2), with L equal to G or G', noting that a K-homomorphism
of an algebraic extension L (whether of finite or infinite degree) of a field
K is an automorphism. This isomorphism is not unique.
2) Consider the set of pairs (L', (J"'), where K c L' eLand (J"' : L' ---> G
is an extension of (J"; put an order on it by the extension relation. It is
inductive, so we can use Zorn's lemma, which asserts that it has a maximal
element (L1' (J"1). We prove that L1 = L by assuming the contrary and
obtaining a contradiction. 0
COMMENTARY. - This theorem can be found in an article by Steinitz dating
from 1910, which Bourbaki refers to as "having given birth to the current
conception of algebra" .
14.2 Examples of Finite Fields 229
14.2 Examples of Finite Fields

1) The set 7l./p7l., equipped with the natural addition and multiplication,
is a finite field for every prime number p. From now on, we will denote
this field by IF p.
2) Let K be a finite field with q elements, and let P be an irreducible

polynomial of degree n in K[X] (such polynomials exist). Recall from
§4.7 the construction of the extension K[X]/(P) of K. A basis of this
new field as a K-vector space is given by the set of classes of the
powers Xk for 0 ~ k ~ n - 1; thus, it is finite-dimensional and has
qn elements.
For example, if K = 7l./27l. and P(X) = X2+X +1, then K[X]j(P)

has four elements, namely the classes of 0, 1, X, X + 1; let us denote
them by 0,1, x, x + 1. The multiplication law in K gives
x 2 = -x - 1 = x + 1,
x 3 = x(x + 1) = x 2 + X = 1,
(x + 1)(x + 1) = x 2 + 1 = x, X4 = x, etc.
In §14.5.1, we will show the existence and uniqueness, up to isomor-

phism, of a finite field with pr elements for every prime number p 2:: 2
and every integer T 2:: 1.
14.3 The Characteristic of a Field

14.3.1 Definition
Let K be a field, and let f : 7l. -> K be the ring homomorphism defined by
f(l) = 1. Then the image of f is a subring of K that is an integral domain,
so its kernel is a prime ideal of 7l.. The non-negative integer generating this
ideal is called the characteristic of K, and written char(K).
EXAMPLES. ~ 1) Q, lR, C are all fields of characteristic o.

2) An example of a field having characteristic equal to a given prime
number p is the field IFp, a quotient of IFp [X] by an irreducible polynomial,
or the (infinite) field of rational functions IFp(X).
14.3.2 Properties
A field of characteristic 0 contains a ring isomorphic to 7l., so it has a
subfield isomorphic to Q, whereas if ker(f) = (p) for some p > 0, then
p must be prime and K contains a subfield isomorphic to IFp. Thus, a
field of characteristic 0 is infinite, and every finite field has non-zero prime
characteristic. Every extension of a given field K has the same characteristic
asK.
14.4 Properties of Finite Fields

14.4.1 Proposition
Let K be a finite field of characteristic p > 0 having q elements.
1) K is a finite-dimensionallF p-vector space. If r ?': 1 is its dimension,
then q = pT.
2) The additive group of K is isomorphic to the group (Zip'll, + r.
3) K* is cyclic of order q - 1 = pT - 1.
4) Every element x E K* satisfies
x q- 1 = 1,
and every element x E K satisfies
COMMENTARY. - Part 3) gives the Primitive Element Theorem for finite

fields; if L is a finite field extension of K, there exists an element x which
generates the multiplicative group L*, so L = K[x].
PROOF. - 1) Because K is of characteristic p, K contains a subfield K'
isomorphic to lFp . We can then equip K with the structure of a finite-
dimensional K'-vector space. If this dimension is r, the cardinal of K is
pT.
2) This follows from the vector space structure mentioned above.
3) Let us follow Gauss' original proof (pp. 53-54 of the Recherches
arithmetiques). Let a be an element of maximal order s E K*. If s = q - 1,
then a generates K*, which is thus cyclic. Suppose s < q - 1. The set
E = {a k ; 0::; k ::; s - I} has s elements, all roots of X S - 1. Now, since
a polynomial of degree s over a field has at most s roots, every element
b E K* not in E is not a root of XS - 1. Thus, the order t of b does not
divide s, and lcm(s, t) > s. Let us write the decompositions of sand t as
products of prime factors Pi, for 1 ::; i ::; r, raised to positive or zero powers
s = II p~i and t = II
p~i, and suppose the Pi are ordered in such a
l$i$T l$i$T
way that k i < Ii for 1 ::; i ::; j and k i ?': Ii otherwise. Set
u = II p~i, U
I S
=-,
u
v= II p~i, and v'
t
v
l$i$j l$i$j
As u' and v have no common prime factor, they are relatively prime. We
see that aU is of order u' and bV ' is of order v. The element aUbv ' is of order
u'v = lcm(s, t) > s, which leads to a contradiction.
Unfortunately, there exists no algorithm for rapidly computing a gener-
ator of K*, even in the case K = IFp'
14.5 Existence and Uniqueness of a Finite Field with pT Elements 231
4) We know that in a finite group, the order of an element divides the

order of the group. As K* is of order q - 1, we have x q - 1 = 1 for every
x E K*. Thus, for every x E K*, we have x q = Xi this property also holds
for x = 0. Fermat's well-known "little theorem" is a special case of this
statement. <)
14.4.2 The Frobenius Homomorphism

PROPOSITION, DEFINITION. - 1) Let K be a (finite or infinite) field of
characteristic p > 0. The map Fp : K ~ K defined by Fp(x) = x P is a
(necessarily injective) field homomorphism, called the Frobenius homomor-
phism of the field. Thus, for every x and y E K, we have
(x + y)P = x P + yp.
2) If L is an extension of a finite field K with q elements, the map Fq :
L ~ L defined by Fq(x) = x q is a K-homomorphism called the Frobenius
homomorphism of the extension.
3) If L is a finite extension of a finite field K with q elements, the Frobenius
homomorphism Fq is a K -automorphism.
PROOF. - 1) Let k be such that 1 ::::: k ::::: p - 1. As p! = k!(p - k)! (D,
and furthermore the prime number p does not divide k!(p - k)!, it must
divide (~). The binomial formula then shows that (x + y)p = x p + yP, so
Fp(x + y) = Fp(x) + Fp(Y). The equality Fp(xy) = Fp(x)Fp(Y) is obvious.
2) If char(K) = p, then Fq is a power of Fp and x q = x for x in K.
3) It suffices to note that Fq is a K-linear injective map between vector
spaces of equal (finite) dimension. <)
REMARK. - If K is a field of characteristic p > 0, the Frobenius homo-

morphism of K is the Frobenius homomorphism of the extension of lFp by
K.
14.5 Existence and Uniqueness of a Finite Field

with pr Elements
14.5.1 Proposition
Let p 2: 2 be a prime and r 2: 1 an integer.
1) Let C be an algebraic closure of lFp (such a C exists by Steinitz'
theorem 14.1.3). Then there exists a unique subfield K of C such that
IKI =q =pT.
2) K is the splitting field of the polynomials X q - X and X q- 1 - 1.
3) Every field with pT elements is isomorphic to K.
NOTATION. - We write IFq for the field with q elements.

PROOF. - 1) If such a field exists, its elements are roots of the polyno-
mial xq - X in C. Let K be the set of these roots. K is stable un-
der multiplication and taking inverses; it is also stable under addition,
since if F = Fp denotes the Frobenius homomorphism of C, we have
(x +y)q = (x+y)pr = Fr(x + y) = Fr(x) + Fr(y) = x q +yq = x +y. Thus,
K is a subfield of C.
K has q elements because P(X) decomposes into linear factors in C, and
because it is prime to its derivative P' (X) = -1, it has only simple roots.
2) This follows from the construction of K.
3) Proposition 14.4.1 3) ensures the existence of a generator x of the
multiplicative group K*. Let rp : IFp[X] ~ K be the surjective homomor-
phism defined by rp(X) = x, P the minimal polynomial of x over IFp, and
'ljJ : IFp[X]/(P) ~ K the factorization of rp by the canonical projection
7r : IFp[X] ~ IFp[X]/(P). Then 'ljJ is an isomorphism (Figure 14.1). As x is
of order q -1, x is a root of the polynomial Xq-l -1, so P divides Xq-l -1.
Set Xq-l - 1 = P(X)S(X).
IF [X]--<p----;oo K
~~
<p IF [X]/(P)
p
K '~
'II'
FIGURE 14.1.
Let K' be a field with pr elements. The q - 1 elements of K'* are roots
of Xq-l - 1. As P is not a constant, there exists at least one element
y E K'* which is a root of P. Let rp' : IFp[X] ~ K' be the homomorphism
defined by rp'(X) = y. As rp'(P) = 0, there exists 'ljJ' : IFp[X]/(P) ~ K'
such that rp' = 'ljJ'7r. The map 'ljJ' is an injective field homomorphism, so it
is an isomorphism since IFp[X]/(P) and K' both have q elements. 'ljJ''ljJ-l is
then an isomorphism of K onto K'. (>
14.5.2 Corollary
For every irreducible polynomial Pin IFp[X] of degree r ::::: 1, K = IFp[X]/(P)
is a jield with q = pr elements, isomorphic to the subjield of the roots of
xq - X in the algebraic closure C ofIFp.
PROOF. - The first assertion is proved using the arguments of §4.5.2, and
the second follows from §14.5.1. (>
14.6 Extensions of Finite Fields 233
14.6 Extensions of Finite Fields

PROPOSITION. - Let K and L be two finite fields of characteristic p, and
set IKI = q = pr, r ~ 1.
1) If L is an extension of K, there exists s ~ 1 such that ILl = psr = qS.
Every element of L is algebraic over K, of degree less than or equal to s.
2) If there exists s ~ 1 such that ILl = psr = qS, then there exists a
unique subfield of L isomorphic to K.
PROOF. - 1) ILl = qS is an immediate consequence of the finite-dimensional
K-space structure s of L. If x E L, then the family {x\ 0:::; k :::; n} is not
independent, which shows (as in §4.5.4) that x is algebraic over L.
2) Let C be an algebraic closure of L, and let K' be the subfield of C
generated by the roots of xq - X, as in §14.5.1. For every x E K', we
have x q = x, so we have x qS = x, which shows that x E L. Thus, K' is
isomorphic to K, by §14.5.1. <>
EXAMPLE. - Thus, a field of order p4 cannot lie inside a field of order p6.
However, each of these two fields contains a unique subfield of order p2,
and these two subfields are isomorphic.
14.7 Normality of a Finite Extension of Finite

Fields
PROPOSITION. - Let K be a finite field and L an extension of finite degree
s E K with qS elements. Then L is a splitting field of one of the polynomials
X qS _ X and X qS - 1 - lover K, and thus it is a normal extension of K.
PROOF. - If an irreducible polynomial in K[X] has a root x in L, then
it divides xqs - X, so it factors into linear factors in L[X]; moreover, its
roots are simple. Thus, the conjugates of x over K lie in L. <>
14.8 The Galois Group of a Finite Extension of a

Finite Field
14.8.1 Proposition
Let K and L be two finite fields of characteristic p such that K c L.
Suppose that IKI = q = pr and ILl = qS for integers rand s ~ 1. Then the
Galois group G = Gal(LIK) is cyclic of order s, generated by F = F q •
PROOF. - F is a K-automorphism of L, by Proposition 14.4.2 3). The

powers of F are also K-automorphisms of L, so they are elements of G.
Let us consider whether or not they are distinct.
We have Fk (x) = xqk for every x E L. In particular, F S = id( L). For
1 :S k :S s - 1, we have Fk :j; id( L), otherwise every x E L * is of order
less than or equal to qk - 1. Now, because L * is a cyclic group, it has an
element of order qS - 1.
° Thus, G has at least s elements, namely the K-automorphisms Fk for

:S k :S s - 1. So if x generates L, then its s images under the powers
of F are distinct; these are the conjugates of x over K. As [L : K] = s, x
cannot have any other conjugates, so G cannot have any other elements.
Thus IGI = s, which gives the result. <)

Let us keep the notation of the preceding proposition. The subgroups of
the group G are all cyclic, generated by elements of the form Fk, where
k divides s. The subfield of L invariant under Fk consists of the elements
x E L such that x q = x. For every k dividing s, L has a unique such subfield
k
(cf. §14.5.1), so the Galois correspondence that we learned in Chapter 8 for

sub fields of C continues to hold for finite fields.
14.8.3 Example
Proposition 14.8.1 proves that Gal(lF212IlF2) ~ Z/12Z.
For example, we have I((F2)) ~ lF4, I((F 6 )) ~ lF64, etc. The roots of
unity different from 1 in lF212 are the n-th roots for n dividing 4,095
5 x 7 x 9 x 13, i.e. n = 5,7,9,13,35,45,63,65, ... ,4,095.
IF
"
= iF
212 4096 {O}
/ ~
/
~6=1F64 iF 4= iF
2 16
6ZI17Z 4ZI12Z
I ~
if 3= IF
2 8
1
iF22=~
I
3ZI12Z
~ I
2ZI12Z
~E/ ~ /
2 ZI12Z
FIGURE 14.2. Galois correspondence for the extension IF 12 of lF2

Exercise 14.1. The algebraic closure
1) Show that the algebraic closure of IQ in C is the union of the algebraic
extensions of finite degree of IQ contained in C. Show that this closure
is a countable set.
2) Let K be a field, L an algebraic extension of K, and C an algebraic

closure of L. Show that C is an algebraic closure of K.
3) Let C be an algebraically closed field. What is the algebraic closure

of C? What are the algebraic extensions of C? Give an example of
an algebraically closed field C and a field L strictly containing C.
Exercise 14.2. Finite fields

1) Let K be a field and L a sub field of K. Show that K and L have the
same characteristic.
2) Let K be a field. What is the intersection of the subfields of K?
3) Does there exist a field containing exactly 51 (resp. 129, 243, 1,024,
65,536, 65,537) elements?
4) Consider the polynomial P(X) = X(X - l)P in IFp[X]. What is the

multiplicity of 1 as a root of P? What about as a root of pI?
5) Set K = IF2(X). Show that the Frobenius homomorphism F of K is

not an isomorphism of K.
Exercise 14.3. x generator, electronic transmission

Let p be a prime and r ~ 1 an integer.
1) Show that there exists an irreducible polynomial P of degree r in

IFp[X] such that the multiplicative group K* of the field K defined
by K = IF p [Xli (P) is generated by the class x E X in K.
2) Assume that pr - 1 is prime.
a) What are the possible values of p?
b) Show that for every irreducible polynomial P of degree r in

IFp[X], the class x E X generates the multiplicative group K*
of the field K = IFp[X]/(P).
3) Show that the class x E X in K = 1F 2 [X]/(X 4 + X 3 + X2 + X + 1)

does not generate K* .
4) The goal of this question is to give a glimpse into the use of the field
K = 1F2 [X]/(X 7 + X 3 + 1), isomorphic to 1F 128 , in the transmission
of electronic messages. We follow an article by Pierre Arnoux pub-
lished in the March 1988 issue of the magazine Pour La Science. The
transmission is based on cyclic codes conceived in 1959-1960. A mes-
sage to be transmitted is a sequence of O's and l's, and statistics show
that transmission errors, i.e. reception of a 0 instead of a 1 or vice
versa, are not very frequent; messages are cut into sections of 120
elements each.
a) Check that K is a field and that the class x E X in K generates

K*.
If we wish to transmit a sequence a = (akh:;k:S;1l9 of elements

of {O, I}, we transmit the sequence <p(a) = (ak)o:s;k926, where
the last terms are defined by x 7 L ak xk = L a12o+k xk .
O:S;k:S;119 O:S; k:S; 6
b) Show that if <p( a) is transmitted with an error, the message can

nevertheless be reconstituted.
c) Show that if two messages are distinct, the sequences trans-

mitted differ for at least three indices. What is the minimum
number of errors that must be made to transmit a message
incorrectly?
Exercise 14.4. ax 2 + (3y2 = -1

Let K be a finite field of characteristic p with q elements, and let
f: K ---> K be the map defined by f(x) = x 2 •
1) a) Show that f is surjective if p = 2.

b) Show that the cardinal of the image of f is (q + 1) /2 if P > 2.
2) We want to show that given two non-zero elements a and {3 of a finite

field K, there exist elements x and y E K such that ax 2 + (3y2 = -1.
a) Show that such elements x and y exist if p = 2.
b) Assume p > 2. Determine the number of elements in the sets

{I + ax 2 ; x E K} and { -{3x 2 ; x E K} . Deduce the existence of
x and y.
Exercise 14.5. Zech's logarithm

1) Show that K = lF 2 [Xl!(X 3 + X + 1) is a field.
2) Let x denote the class of X in K. Show that x generates K*, and
construct the table of powers of x in the basis {I, x, x 2 }.
3) Define the function Z : {I, ... , 6} -+ {I, ... , 6} by 1 +Xi = XZ(i). The
function Z is called Zech '8 logarithm.
a) Give the table of values of Z.
b) Show that using Z, we can compute sums in K.
c) Show that Z(Z(i)) = i.
COMMENTARY. - Zech's logarithm is an economical way to program the

computation of addition in the field IF q' but unfortunately, the place needed
to stock the tables limits this procedure to small values of q.
Exercise 14.6. The field with 343 elements

1) Let a be an element of lF 7 . For which values of a is the polynomial
X 3 - a irreducible in lF 7 [Xj? Deduce that K = lF 7 [Xl!(X 3 - 2) is a
field.
2) What are the possible orders of an element of the multiplicative group

K*? What is the order of the class x E X in K*?
3) Find an element y of order 19 in K*, of the form a + bx, with a and

bin lF 7 .
4) Use the above to find a generator u of the multiplicative group K*.

Determine the minimal polynomial of u over K.
Exercise 14.7. Sums of two squares

Let K be a finite field of characteristic p. In this problem, we propose
to show that every element of K is a sum of the squares of two
elements of K.
1) Prove a more precise result in the case p = 2.

Now we assume p > 2.
2) Give an explicit formula in the case where -1 is a square in K.

3) Assume that -1 is not a square in K, and set L = K[i] with i 2 = -l.

Let N : L -+ K denote the norm map (whose value at x is equal to the
product of the conjugates of x), and F the Frobenius homomorphism
of L.
a) Determine the relation among N(x), F(x), and x; deduce an

expression for N(x) in terms of x.
b) Show that N induces a homomorphism from L * to K*. Deter-

mine its kernel. Conclude.
4) Express each of the elements of IF q as a sum of two squares for q =

7,13,19.
COMMENTARY. - The method we propose here is different from the one

used in Exercise 14.4.
Exercise 14.8. Cyclotomic polynomials and finite fields

Let J..lk denote the group of k-th roots of unity in K.
Let K be a finite field of cardinal q, of characteristic p, and let
C be an algebraic closure of K. Let n ::::: 1 be an integer which is not
a multiple of p (see part 1 of this problem for the case where n is
a multiple of p). Let cI> = cI>n be the n-th cyclotomic polynomial in
Z[X], and let f : Z[X] -+ K[X] be the homomorphism defined by
f(l) = 1 and f(X) = X. Let J..l = J..ln, L = K[J..l]' s = [L : K], and
F the K-automorphism of L defined by F(x) = x q • Finally, let G be
the Galois group Gal(LIK).
1) In this question, we assume that n is a multiple of p, and we set

n = mpr with (m,p) = 1 and T > O. Show that J..l = J..lm.
2) a) What is the cardinal of J..l? What is the number of primitive

roots in J..l?
Set <Pd,K(X) = II
(EJ,ld,C; primitive
(X - () for d dividing n.
b) Show that J(xn - 1) = II cI>d,K(X) in K[X].

dED(n)
c) Deduce, using induction on n, that f(<pn(X)) = cI>n,K(X).
From now on, let cI>n denote the polynomial cI>n,K if no confusion
is possible.
3) a) Show that s is the order of q in U(n), by considering the Fk(),

where ( is a generator of {L.
b) Deduce that the cyclotomic polynomial n factors into a prod-

uct of irreducible polynomials of degree s in K[X].
4) Assume n = 8, and set = s, so (X) = X4 + 1.

a) What happens to if p = 2? Assume now that p =1= 2.
b) Show that is reducible over lFq whatever the value of q. Note,
however, that X4 + 1 is irreducible in Z[X].
c) Determine L when q = 5,7,17, and factor in K[X] and in

L[X] as a product of irreducible factors.
5) Show that 12 is reducible in every finite field.
6) a) Give a necessary condition on n for n to be irreducible in lFq.
b) Under what conditions on q is the polynomial n irreducible in

lFq for n = 3,5, 14?
Exercise 14.9. The field with 16 elements

1) What are the irreducible polynomials of degree 4 in IF 2[X]?
2) For each of the preceding polynomials P, let x denote the class of

X in lF 2 [Xl!(P). Construct the table of values of powers of x with
respect to the basis {I, x, x 2 , x 3 }. Detail the case where x generates
lFi6·
From now on, we set P(X) = X4 + X + 1 and we define IF 16 as

lF 2 [Xl!(P). Recall (Exercise 9.6) that 15(X) = XS_X7 +X5_X4+
X 3 - X + 1 in Z[X]. We continue to write 15 for the polynomials
in lFdX] or lF I6 [X] obtained by reducing the coefficients modulo 2 or
modulo 16.
3) a) Give a generator of the Galois group Gal(lFI6IlF2).
b) Determine the factorization of 15 into a product of irreducible

factors in IF'2[X] (use the preceding exercise).
c) Give the roots of these factors in IF'16[X] in terms of x.

Exercise 14.10. Roots of unity and "cosine"

1) Let p be a prime, IF q a field of characteristic p, C an algebraic closure
of IF q' k an integer prime to p, J.l the group of k-th roots of unity in
C, and ( a generator of J.l. Show that c = ( + C 1 E IF q if and only if
q = ± 1 mod k (we distinguish the cases ( E IF q and ( rf. IF q ) •
2) Take q = 16 and k = 17.

a) Determine the polynomial in lF 2 [X] whose roots are the distinct
non-zero values w + w- 1 as w runs through J.l.
b) Determine the possible values of c in terms of x for the field

lF 2 [Xl!(X 4 + X + 1) of exercise 14.9.
NOTE. - See the article by Minae and Reis listed in the bibliography for
further developments on this subject.
Exercise 14.11. Irreducibility of XP - X +a

Let p be a prime, K a finite field of characteristic p, a an element of
K, and P(X) = XP - X + a. Let C be an algebraic closure of K.
1) Let x be a root of P in C. Determine the other roots of P in C.
2) Deduce that either the polynomial XP - X + a has all its roots in K,

or it is irreducible in K[X].
3) a) Let a -lOin IFP' Show that the splitting field of the polynomial
XP - X + a over IF p is an extension of degree p of IFp'
b) Let n be an integer. Show that the polynomial XP - X + n is

irreducible in Q[X] for an infinite number of values of n.
Exercise 14.12. Irreducible polynomials over a finite field

Let nand r be integers, and let p be a prime; set q = pr. Let Iq (n)
denote the number of polynomials irreducible of degree n over lFq •
1) Show that Iq(n) 2: 1, by considering a generator of an extension of

degree n of IF q'
2) Show that qn = L dlq(d).

dED(n)
3) Deduce a formula giving Iq(n).

4) Show that xqn - X = II II P, where A(d) denotes the set of

dED(n) PEA(d)
irreducible polynomials over IFq of degree d.
5) a) Show that Iq(n) ':::0 qn In in the neighborhood of +00.
b) Compute Iq(n) in the cases q = 2,1:::; n :::; 9, q = 3,1:::; n :::; 5,

q = 5, 1:::; n :::; 3, q = 7, n = 1 or 2.
Exercise 14.13. The quadratic reciprocity law

In this exercise, we assume that p is a prime number different from 2.
We will use the same notation for an integer and for its class modulo
p.
1) Let f : U (p) ---+ U (p) be the homomorphism defined by f (x) = x 2 ,

and set A = '25(1).
a) Let 9 be a generator of U(p). Show that x E A if and only if x

is an even power of g.
b) Deduce that x E A if and only if X(p-l)/2 = 1, and that x (j A

if and only if X(p-l)/2 = -1.
c) Considering {-I, I} as a multiplicative group, check that set-

ting <p(x) = x(p-l)!2 gives a group homomorphism <p : U(p) ---+
{-I, I}.
From now on, we set <p(x) = (~); (~) is called the Legendre symbol
of x. The Legendre symbol extends to every integer x -::J 0 mod p.
2) Show that, for x and y in U(p), we have
(x:) (~) (~) (~) =1
(~) = 1 (X~l)=(~)
(-pI) = (_1)(p-l)/2 L (::.) = O.
xEU(p) P
3) Computation of (~)=
Let C be an algebraic closure of IFp, and let ( be a primitive

eighth root of unity in C. Set a = (+ C I .
a) Compute a 2 .
b) Deduce the values of (~) according to the values of p mod 8.
4) Main formula:
Let p and q be distinct primes: then (~) (~) = (-1) ( zs.! )( q;1 ).
Let p and q be distinct odd primes, and let C be an algebraic
closure of IFp and ( a primitive q-th root of unity in C. Set
a = L (~) (k; in this expression, k simultaneously denotes
q
kEU(q)
an element of U(q) and the corresponding integer between 1 and
q -l.
a) Show that
a2 =(_1)(q-I)/2 L m L
O::;m::;q-I (kEU(q),k#-m
C- mk - I )
q .
b) Show that
L C-mk- I ) --1 for 0 < m ::; q - 1,

kEU(q),k#-m q
""
L (l-mk- l ) -q-1 for m = O.
kEU(q),k#-m q
c) Deduce that a 2 = (_1)(q-I)/2 q.
d) Show that a P - 1 = (~) (compute aP ).
e) Conclude.
5) Show that 30 is a square in 1F 65 ,537 (note that 65,537 = 224 + 1 is a

prime number).
COMMENTARY. - Special cases of the quadratic reciprocity law were indi-

cated by Fermat, namely the values of (~2) and of (~3). They were
proved by Lagrange (1775) and Euler (1760) respectively. The general law
was conjectured by Legendre (1785), who gave an incomplete proof of it;
Gauss gave several proofs, the first one in 180l.
The quadratic reciprocity law relates the properties: "p is a quadratic
residue modulo q " and" q is a quadratic residue modulo p ", which pro-
vides a rapid method for computing the Legendre symbol. The proof given
above uses what are known as "Gauss sums". The ninth of Hilbert's famous
23 problems proposed in 1900 concerns generalizations of the quadratic reci-
procity law for cubic residues, etc., which are currently objects of research.

1) The conditions n . 1 = 0 in K and n . 1 = 0 in L are equivalent.

2) The intersection of the subfields of K is the smallest sub field of K; it is
Q if char(K) = 0, 7l/p71 if char(K) = p.
3) As the numbers 51 and 129 are not prime powers, there is no field with
51 or 129 elements.
To see that there exist fields of cardinal 243 = 35 , 343 = 73 , 1,024 = 2 10 ,
as well as 65,536 which is prime (65,536 = 216 and 65,537 is the Fermat
number F 4 ), it suffices to note that they are all prime powers.
4) The multiplicity is the same in the two cases, namely p.
5) We know that F is injective. It is not surjective: for example, X is

not in the image of F. Indeed, if X = (P/Q)2, we would have XQ2 = p2,
which is impossible since the first term is of odd degree and the second of
even degree.
1) Let L be a field with pr elements, y a generator of L *, and P the

minimal polynomial of y over lFp. As L = lFp[y] is of degree rover lFp, Pis
of degree r. Let 7r : lFp[X] -> K = lFp[X]/(P) be the canonical projection.
The ring homomorphism f : lFp[X] -> L defined by f(X) = y induces an
isomorphism 'P : K -> L and x = 7r(X) = 'P- 1 (y) is a generator of K of
minimal polynomial P over lFp.
2) a) The primes of the form 2 r - 1 are the Mersenne numbers, which

are prime, for example, for r = 2,3,5, 7 but not for 11; the largest known
Mersenne primes are for r = 132,049 (1983), r = 216,091 (1985), r =
1,398,269 (1996), etc. If p -I- 2, the condition that pr - 1 is prime is equiv-
alent to p = 3 and r = 1, otherwise pT - 1 is an even number greater than

2, so it cannot be prime.
b) Because the order of K* is prime, it is generated by its non-trivial

elements.
3) We have x 5 -1 = 0 and IK*I = 15.

4) a) The polynomial X7 + X 3 + 1 is irreducible: to see this, it suffices
to check that it is not divisible by an irreducible polynomial of degree less
than or equal to 3. Furthermore, x generates K* since 127 is prime.
b) Let b = (b k )0::;k::;126 be the sequence received. On reception, of the

message, one computes the numbers Ck, 0 ::; k ::; 6, defined by
x7 L bkX k = L Ckxk.
09::;119 0::;k::;6
If the error concerns the l-th rank for 0 ::; I ::; 119, we have
L Ck xk + L b 120+k xk x7 L bk Xk + x7 L ak xk
0::;k::;6 0::;k::;6 0::;k::;119 0::;k::;119
xl+ 7 .
If the error concerns the rank I, 120 ::; I ::; 126, we have
'L" ' Ck X k + '"'

L b 120+k x k = x 1-120 .
0::;k::;6 0::; k::; 6
As x generates K*, we can always find I, and if 0::; I ::; 119, we can correct
the received message.
c) We have x7 L ak xk + L a120+k xk = 0 for every transmitted

0::;k::;119 0::;k::;6
sequence (ak)0<k<126. If two transmitted sequences differed for a single
index I with 0 -;; 1 ::; 126, we would have xl = 0; if they differed for two
indices I and If, we would have xl = xl'. But this is impossible since x is of
order 127 in K*.
Three errors suffice: we can receive the sequence ao = a120 = a123 = 1,
ak = 0 otherwise, instead of the zero sequence, and we will not perceive
the error, since
x7 L ak xk + L a120+k xk = o.
0::;k::;119 0::;k::;6
1) a) f is surjective since it is the Frobenius automorphism of K.

b) Let 9 be the restriction of f to K*. It is a homomorphism of multi-
plicative groups. Its kernel has two elements: 1 and -1, which are distinct
since p > 2. Thus, its image has (q - 1)/2 elements (we can also see that
if x =I- 0, then x =I- -x since p > 2 and x 2 = (_x)2, which enables us to
count). The image of f also contains 0, which gives the result.
2) a) As 0: is non-zero, it is invertible. By 1) a), _0:- 1 is the square of an

element x; we obtain the result by setting y = 0.
b) Because 0: and f3 are non-zero, the two sets have (q + 1) /2 elements,

as does the image of f. Thus their intersection is non-empty. Let z be
an element of this intersection; there exists x and y in K such that z =
1 + o:x 2 = _f3y2, which gives the result.
1) The polynomial X 3 + X2 + 1 has roots in 1F2, so it is irreducible over

1F 2. The quotient 1F2 [Xl!(X 3 + X2 + 1) is an extension of degree 3 of 1F2'
so it is a field with eight elements.
2) x generates K*, which is of prime order. The table of powers of x,

expressed in the basis (1, x, x 2 ), is given in Table 14.l.
3) a) The values of Z are given in Table 14.2.

b) Because two elements are both powers of the generator x, their sum
is computed by factoring out the smallest power of x. For a < b, we have
x a + x b = x a (1 + x b- a ) = xa+Z(b-a). For example, x 3 + X4 = x 3(1 + x) =
X 3 X 5 = x 8 = X.
c) XZ(Z(i» = 1 + XZ(i) = 1 + 1 + Xi = Xi, hence Z(Z(i)) = i.
1) We compute the cubes modulo 7 (Table 14.3).

X 3 - a is irreducible in 1F7 [Xl if a is not a cube modulo 7, i.e. if a =
2,3,4,5.
2) K is a field with 73 = 343 elements and K* has 342 = 2 x 9 x 19

elements. The order of an element of K* is a divisor of 342. As x 3 = 2 and
23 = 8 = 1 mod 7, x is of order 9.
1 X X2
1 1
X 1
x2 1
x3 1 1
x4 1 1 1
x5 1 1
x6 1 1
x7 1
TABLE 14.1. Powers of x
i 1 2 3 4 5 6
Z 5 3 2 6 1 4
TABLE 14.2. Zech's logarithm
3) If y = a+bx is of order 19, we have y19 = 1, so y21 = y2. As (a+bxf =

a + bx 7 = a + 4bx, this gives
a3 + 12a 2bx + 48ab2x 2 + 64b 3 x 3 = a 2 + 2abx + b2x 2 ,

a 3 + 5a 2bx - ab 2x 2 + 2b 3 = a 2 + 2abx + b2x 2 ;
hence, a 3 + 2b 3 = a 2, 5a 2b = 2ab, -ab 2 = b2.

We do not have b = 0 since y would lie in IF 7 and would not be of order
19. So we can simplify the last condition by b2 , which gives a = -1. The
second condition is then satisfied, and the first implies that b3 = I, i.e.
b=l,2,4.
4) A generator of K* is the product of three elements of order 2, 9, and 19

respectively; for example, u = -xy = X - bx 2 . Let us compute the powers
of u (Table 14.4). We immediately see that u is of degree 3 over F7 since it
is a generator of K*. As u 3 = bu - 2, X3 - bX + 2 is the minimal polynomial
of u over F 7 .
X 0 1 2 3 4 5 6
x3 0 1 1 -1 1 -1 -1
TABLE 14.3.
1 X x2
1 1 0 0
u 0 1 -b
u2 -4b 2b2 1
u3 -2 b _b2
TABLE 14.4.
1) If p = 2, the Frobenius morphism F defined by F(x) = x 2 is an au-

tomorphism of K. Its surjectivity means that every number is a square in
K.
2) If -1 = b2 , then because 2 is invertible, we have
3) a) The Galois group of Lover K has two elements, id and F. Hence,

N(x) = xF(x) = xx q = x q+ 1 •
b) N(xy) = N(x)N(y) is clear; ker(N) = {x,x q +1 = I}.

Let 9 be a generator of L*; the elements ofker(N) are clearly the gk(q-l),
o~ k < q + 1. There are q + 1 of them. Thus the image of N contains
(q2 _ 1)/(q + 1) = q - 1 elements, so it is K*. As an element of L can be
written u + iv with u, v E K, for every x E K*, there exists u and v in K
such that N(u + iv) = x, i.e. u 2 + v 2 = x.
4) If q = 13, then q = 1 mod 4, so -1 is a square; we see easily that

-1 = 52, so a = (6a + 6)2 + (4a - 4)2.
If -1 is not a square, we can still work using successive tries. For example,
the squares mod 19 are 1, 4, 9, -3,6, -2, -8,7,5, so 2 = 16+5 = 42 +9 2,
3 = 16 + 6 = 42 + 52, etc.
1) It suffices to note that Xn - 1 = (xm - l)pr.
2) a) As xn - 1 is prime to its derivative nX n - 1 since n is not a multiple

of p, it has n distinct roots and IILI = n. As IL is a cyclic subgroup of the
cyclic group K*, there are <1>( n) primitive n-th roots of unity.
b) The proof is the same as the one given in §9.4.2.

c) The property is obvious for n = 1. Let n > 1, and suppose that the
property holds for every k < n. We have Xn -1 = n(X) II d(X)
dED(n),d#n
in Z[X]. Hence f(xn -1) = f(n(X)) II
dED(n),d#n
f(d(X)), i.e. by the the
induction hypothesis,
II
dED(n),d#n
We conclude by comparing with the formula of b).
3) a) We have F(() = (q so, by induction, Fk(() = (qk. As F is of order

s in G, we see that qk i= 1 for 1 ::::: k < sand qS = 1. This gives the result.
b) As K[IL] = K[(] for every generator ( of IL, every root of n is of

degree s over K. Thus, the irreducible factors of n, which are minimal
polynomials of these generators, all have the same degree s.
4) a) factors into linear factors: (X) = (X + 1)4.

b) If P i= 2, we have q = 1, 3, 5, 7 mod 8. As these elements are of order
1 or 2, we have s = 1 or 2; thus ( is never of degree 4 over IF q and X4 + 1
is reducible over lFq whatever the value of q.
c) The cases proposed here lead to different situations.

• q = 5. As 5 is of order 2 in U(8), X 4 + 1 factors into a product of two
irreducible factors of degree 2 in lF5[X], Noting that -1 = 4 mod 5,
we have X4 + 1 = (X2 - 2)(X2 + 2) in lF5[X]. If we set 2 = a 2 and
L = lFda], we have X4 + 1 = (X - a)(X + a)(X - 2a)(X + 2a).
• q = 7. Because 7 is of order 2 in U(8), the degree of the irreducible

factors is again 2. As -1 is not a square in lF7, we will translate the
equality of JR[X] : X4 + 1 = (X2 -.J2x + 1)(X2 +.J2x + 1). We see
that 2 = 32 mod 7, so X4 + 1 = (X2 - 3X + 1)(X2 + 3X + 1). If we
set i 2 = -1 and L = lF7[ij, we obtain X4 + 1 = (X + 2 - 2i)(X + 2 +
2i)(X - 2 - 2i)(X - 2 + 2i).
• q = 17. As 17 = 1 mod 8, X 4 + 1 factors into linear factors and

L = lF17 : X4+ 1 = (X2 -4)(X 2 +4) = (X -2)(X +2)(X -8)(X +8).
5) Distinguish the cases p = 2, p = 3, and p 2: 5, and use the same

reasoning as in 3) b); as U(12) ~ (Z/2Z)2, we see that s = 1 or 2.
6) a) The condition " is irreducible over K" is equivalent to s = deg( <1» =

cp(n), in other words to the fact that q generates U(n). This means that
U(n) is cyclic, i.e. (see Exercise 9.8,4 b» that n is of the form 2pk or pk,
for an odd prime p.
b) As q must generate U (n), we find that
q = 2 mod 3 if n = 3; q = 2, 3 mod 5 if n = 5;
q = 3,5 mod 14 if n = 14.
1) A polynomial of degree 4 is reducible if it has a root or 2 irreducible

factors of degree 2. We know that the only irreducible polynomial of degree
2 over IF 2 is X2 + X + 1. Since the only reducible polynomial of degree 4
has no root over lF2' it is X4 + X2 + 1 = (X2 +X + 1)2. Now, a polynomial
of degree 4 over IF 2 which has no roots must be one of the four polynomials
X4 + X + I, X4 + X 3 + I, X4 + X 3 + X2 + X + 1 and X4 + X2 + 1. The
preceding remark shows that only the first three of these polynomials are
irreducible.
2) Tables 14.5 and 14.5bis give the powers of x expressed in the basis I, x,
x 2 , x 3 in each case. We see that x is a generator in the two first cases (to
obtain this result alone, it would suffice to check that x is not of order 3 or
5, which is shown by the computation of x 3 and x 5 ) but not in the third.
X4+X +1 X 4 +X 3 +1
k 1 x x2 x3 1 X X2 x3
0 1 1
1 1 1
2 1 1
3 1 1
4 1 1 1 1
5 1 1 1 1 1
6 1 1 1 1 1 1
7 1 1 1 1 1 1
8 1 1 1 1 1
9 1 1 1 1
10 1 1 1 1 1
11 1 1 1 1 1 1
12 1 1 1 1 1 1
13 1 1 1 1 1
14 1 1 1 1
15 1 1
TABLE 14.5.
1 X x2 x3
1
1
1
1
1 1 1 1
1
TABLE 14.5.bis
3) a) The Galois group Gal(lF16llF2) is cyclic of order 4, generated by the

Frobenius automorphism F, defined by F(z) = z2.
b) Because x is a root of clI 15 having minimal polynomial X4 + X + 1, we

know that clI 15 is divisible by X4 + X + 1. This shows that the irreducible
factors of clI 15 in lF 2[X] have degree 4 (see Exercise 14.8). Thus clI 15 is
a product of two irreducible factors over <P2, and the second one is the
quotient <P15(X)/(X 4 + X + 1) = X 4 + X 3 + 1. Hence,
<P15(X) = (X4 + X + 1)(X4 + X 3 + 1).
c) The distinct roots of <P15 are all the generators of the cyclic group
IFi6' i.e. the xk where k is prime to 15. Thus, we find that <P15(X) is equal
to
The conjugates of x over IF 2 are the images of x under the elements of the
Galois group Gal(IF16IIF2); thus they are x = id(x),x 2 = F(x),x 4 = F2(X),
and x 8 = F3(x). These are the four roots of X4 + X + 1. '
The roots of X4 + X 3 + 1 are the other roots of <P15 : x 7, xu, x 13 , X14.
We can check that these are the conjugates of x 7, since F(x 7) = x14,
F2(X7) = x 13 , and F 3(x) = xu.
1) Let us first assume that c E IF q. If ( E IFq, we have (q-l = (k = 1, so

q - 1 = 0 mod k. If ( rt IFq, we have cq = c = ( + C 1 = (q + C q , so
((q-l _ 1)(( - C q ) = o. As the first factor is non-zero, we have ( = C q ,
so (q+ 1 = 1 and q + 1 = 0 mod k.
Now assume that q = ±1 mod k.
If q = 1 mod k, then (q-l = 1 shows that ( lies in IFq, so c does as well.
If q = -1 mod k, then (q+l = 1 implies that ( = C q , so cq = (q + C q =
C 1 + ( = c. Thus c E IF q.
2) a) The primitive 17th roots of unity are all distinct in an algebraic
closure of IF 16 , since X17 - 1 is prime to its derivative. If c = W + w-l, we
have w2 = cw + 1; if c = 0, then W = 1, otherwise c can take eight distinct
values since if W +w- 1 = "1 + "1-1, we have (w - "1)(W"1 -1) = 0, so "1 = W or
"1 = w- 1. We have w 2 = cw + 1, w 4 = c 3w + c 2 + 1, w 8 = c7 W + c6 + c4 + 1,
w 16 = C15 W + c 14 + c 12 + c8 + 1, W 17 = c 15 + w(c l6 + c 14 + C12 + c8 + 1). As
w 17 = 1 = C15 , C is a root of the polynomial Q(X) = X 8+X7 +X6+X4+1,
whose roots are exactly the eight possible values of c.
b) We see that x is a root of Q. Let ( be a 17th root of unity such that

x = (+(-1. As ( generates the group of 17th roots of unity in IF 16 [(] ~ IF 256,
the possible non-zero values of c are the (k + (-k with k = 1, ... ,8. Table
14.6 gives the result of the computations; the first line gives the value of k,
and the second gives the expression for the quantity (k + (-k in terms of x.
We can also use the factorization Q(X) = (X4 + X + 1)(X 5 - 1)/(X -1).
1 2 3 4 5 6 7 8
X X2 X+X 3 l+x X2 +X 3 x3 1 +x+x2 +x3 1 +x2
TABLE 14.6.

1) For every u E IFp, we have (x+u)P - (x+u) +a = x P+u-x-u+a = O.
The p roots of P are thus the x + u with u E IFpo
2) If P is reducible in K, there exist polynomials S, T in K[X] such that
P(X) = S(X)T(X) with deg(S) = s, 1 ::::; s ::::; p - 1. The roots of S are
among the roots of P; they are of the form x + u, with u in IFp' Their sum
is of the form sx + v with v E IFp; as it belongs to K (up to sign, it is a
coefficient of S), we have sx E K, so x E K since s -I- o. Because one root
of the polynomial P lies in K, they all do, by 1).
3) a) The polynomial XP - X + a has no root in IFp, since x P - x = 0 for
every x E IFp. Thus, it is irreducible in IFp[X] , which gives the result.
b) The preceding question shows that XP - X + n is irreducible in IFp [X],
so it also is irreducible in Z[X] and in Q[Xl, whenever n is an integer not
divisible by p.

1) Let C be an algebraic closure of IF q, and let L be the subfield of C with
qn elements; it is an extension of degree n of IFq' Let x denote a generator
of L *; x is of degree n over IFq, so its minimal polynomial over IF q, which is
irreducible over IFq, is of degree n, and consequently, Iq(n) 2: 1.
2) Let L be as above. Consider the equivalence relation: "x and yare
equivalent if they have the same minimal polynomial over IF q". If x is of
degree dover IFq, then x has d distinct conjugates over IFq, and d divides
n. The desired equality follows from the partition of L into equivalence
classes.
3) The Mobius inversion formula gives nlq(n) = Lp,(d)qn/d.
dED(n)
4) If P is an irreducible polynomial of degree d for some d dividing n, then
every root of P is in an intermediate extension between IF q and IF qn, so it
satisfies xqn - x = O. This proves that P divides xqn - X. By 2), the sum
of the degrees of such polynomials is equal to the degree of xqn - X, which
concludes the proof.
5) a) We have 0 < 1 - (nlq(n)/qn) =
dED(n),d#l
b) The results are given in Table 14.7. Note the form of the equivalence
of a).
n 1 2 3 4 5 6 7 8 9
qn 2 4 8 16 32 64 128 256 512
Iq(n) 2 1 2 3 6 9 18 30 56
n 1 2 3 4 5
qn 3 9 27 81 243
Iq(n) 3 2 8 18 48
n 1 2 3 n 1 2
qn 5 25 125 qn 7 49
Iq(n) 5 10 40 Iq(n) 7 21
TABLE 14.7.
COMMENTARY. - These tables show that we can choose an irreducible

polynomial of degree n over IFq in many ways. However, certain polynomials
are more useful than others.
1) a) Every x E U(p) can be written as a power of g; the result follows

since p is odd.
b) On the one hand, gP-l = 1; on the other, if a = g(p-l)/2, we have
a =I- 1 and a 2 = 1. Thus, because IFp is a field, we have a = -1.
2) The properties follow from 1). The last one follows from the equality
between the number of even powers and the number of odd powers of 9 in
U(p).
3) a) As (4 = -1, we have (2 = -C 2 , so a 2 = (2 +2+C 2 = 2.
b) We know that (~) = 2(p-l)/2, so (~) = a P- 1 = aPia. As a P =

(( + C1)P = (P + C P, a P depends on the values of p mod 8.
If p = ±1 mod 8, then a P = a, so (~) = a P- 1 = 1; 2 is a square modulo

p.
If p = ±5 mod 8, as (4 = -1, aP = (5 +C5= -a, so (~) = -1; 2 is
not a square modulo p.
4) a) Let k, I, m denote both an element of lFq and the corresponding

integer between 0 and q - 1. The multiplicativity of the Legendre symbol
implies that
a2 = L ('5.) (i) (HI - L (m L ('5q.) (m q- k)

k,IEU(q) q q O::;m::;q-l kEU(q),k#m
since(~) (m;k) = (_:2) C-;k- 1

) =(_1)(q-l)/2C-;k-
1
).
b) If m = 0, then L (
1
-;k-l) =q- 1. If 1 :::; m :::; q - 1,
kEU(q),k#m
we have
L
kEU(q),k#m
C-mk- q
1
)_ L
- IEU(q),I#l
(i)--l'
q - ,
this follows because {I -

and from the last equality of 2).
mk-\
k E U (q), k m} = ~ {I; lEU (q), I ~ I}
c) We obtain a 2 = (_1)(q-l)/2 (q -1 - L (m) = (_1)(q-l)/2 q.

l::;m::;q-l
d) To begin with,
aP = L ('5.) (k p - L (IP-l) (I
kEU(q) q IEU(q) q
( p-l)
q
L
IEU(q)
(i) q
(I - (E)q a
Furthermore, a ~ 0 by c).
e) The preceding computations allow us to write
5) The number 65,537 is prime (see Exercise 14.2), and it suffices to show
that (65~~37 ) = l.
By 3), (65,2537 ) = 1 since 65,537 = 1 mod 8.

By 4), we have
(65,~37) = (65,:37) = (~) =-1
and
(65,5537) = (65,:37) = (~) =-l.
This gives the result. But it does not produce the numbers with square
equal to 30 modulo 65,537. A formal computation program gives ±27, 135.
15
Separable Extensions
In this chapter, we consider arbitrary commutative fields, i.e. finite and

infinite fields of arbitrary characteristic.
15.1 Separability
DEFINITION. - Let K be a field and L an extension of K. An element a E L
that is algebraic over K is said to be separable over K if it is a simple root
of its minimal polynomial.
An algebraic extension L of a field K is separable if every element of L
is separable over K.
An irreducible polynomial in K[X] that has no multiple roots in an
algebraic closure of K is called separable; if it has multiple roots, it is
called inseparable.
EXAMPLES. -
1) If L c <C is an algebraic extension of a field K, then L is a separable

extension of K (see §4.4.5).
2) Algebraic extensions of finite fields and of characteristic 0 fields are

always separable, by Proposition 15.5 below.

258 15. Separable Extensions
COMMENTS. - The notion of separability dates back to Steinitz (1910). A

non-separable algebraic element of degree n over a field K has less than n
conjugates over K.
15.2 Example of an Inseparable Element

Let IF p (U) be the field of rational functions in U with coefficients in IFpo
Let K be the image of the Frobenius homomorphism F: lFp(U) -+ lFp(U)
defined by F(x) = x P (Figure 15.1). Set V = F(U) = Up. Then V belongs
to K, and U is algebraic over K since it is a root of the polynomial XP - V E
K[X]. However, we can easily check that it does not lie in K.
Let us show that the polynomial XP - V E K[X] is irreducible over K.
The proof is immediate for p = 2. In the general case, we have XP - V =
F(XP - U) and FllFp(U) is an isomorphism, so it suffices to show that
XP -U is irreducible in lFp(U) [X]. But lFp[U] is a factorial ring, with fraction
field lFp(U), in which U is an irreducible element, so it is prime and we can
apply Eisenstein's criterion.
Thus U is of degree p over K, with minimal polynomial XP - V. In
lFp(U)[X], we have XP - V = XP - UP = (X - U)p. Thus, U is a unique
root, of order p, of its minimal polynomial, so U is not separable over K.
IF (U)
p
F
IF (U) " Im(F)=K

p
FIGURE 15.1.
15.3 A Criterion for Separability

PROPOSITION. - Let p be a prime, K a field of characteristic p, L an
extension of K, and a an element of L which is algebraic over K, of
minimal polynomial P over K.
1) a is separable over K if and only if pl(a) =I- 0 or, equivalently, pI =I- o.
The conjugates of a over K are then also separable over K.
2) If a is not separable over K, then P is of the form
P(X) = L ak XkP .
O:S;k:S;n
15.4 Perfect Fields 259
PROOF. -
1) We have P(X) = (X - a)S(X), so PI(X) = S(X) + (X - a)S'(X). If

a is separable over K, we know that S(a) =1= 0, so p' (a) is non-zero.
The converse is immediate. Finally, pI =1= 0 is equivalent to pI (a) =1= 0
since P is the minimal polynomial of a over K and p' cannot vanish
at a if pI =1= O.
2) We have p' = 0 by a), which implies that P(X) = L akXkP. <)

Osksn
15.4 Perfect Fields

Let K be a field of characteristic p > O. Let KP denote the image of the
Frobenius homomorphism F: K -+ K defined by F(x) = x p • As F is
injective, it is a subfield of K.
A field is said to be perfect if it is of characteristic 0 or of characteristic
p =1= 0 with the property that KP = K.
EXAMPLE. - Every finite field is perfect.
15.5 Perfect Fields and Separable Extensions

PROPOSITION. - A field K is perfect if and only if every algebraic extension
of K is separable.
PROOF. - Assume that K is perfect. The case char(K) = 0 was considered
in §4.4.5. Assume p = char(K) =1= 0, and suppose that L is an algebraic
extension of K, and a is an element of L, non-separable over K, of minimal
polynomial P over K. By Proposition 15.3, P(X) = L
akXkp (with
Osksn
an =1). As KP = K, there exist elements bo, ... , bn E K such that (bk)P =
ak for 0::; k ::; n; hence, P(X) =( L bkXky. Thus, P is not irreducible
Osksn
over K, which contradicts the hypothesis.
Conversely, suppose that every algebraic extension of K is separable. If
char(K) = 0, then K is perfect by definition. Suppose p = char(K) i- 0;
let us show that the Frobenius homomorphism F is surjective. Let a E K,
and let b denote a root of XP - a in an algebraically closed extension of K.
Let S be the minimal polynomial of b over K. As XP - a = (X - by, S
divides (X - by, so it is of the form (X - b)8. But as K[bJ is a separable
extension of K, we have s = 1, so b E K. <)
260 15. Separable Extensions
15.6 Galois Extensions

15.6.1 Definition
An algebraic extension N of an arbitrary field K is said to be Galois if
K = I(Gal(NIK)), the field of invariants of Gal(NIK).
15.6.2 Proposition
An algebraic extension is Galois if and only if it is normal and separable.
COMMENTARY. - This proposition explains why we used the term Galois

extensions for normal extensions in Chapter 8 (cf. §8.2). Indeed, as we saw
in earlier chapters, the Galois extensions of a perfect field are exactly the
normal extensions since all extensions of such a field are separable.
PROOF. - Let N be a Galois extension of K. Let x E N have minimal
polynomial P over K, and let Xl = x, ... , Xn be the distinct conjugates
of x in N. The polynomial S(X) = II
(X - Xi) is invariant under the
l~i~n
elements of Gal(NIK), so its coefficients lie in I(Gal(NIK)) = K. It follows
that P = S and the roots of P are all distinct and lie in N. The extension
N of K is thus normal and separable.
Now, let N be a normal separable extension of K. We have the inclusion
K C I(Gal(NIK)). If x E N does not lie in K, then its minimal polynomial
over K is of degree strictly greater than 1. As the extension is separable,
there exists a conjugate y of x, different from x, in an algebraic closure C
of N. There exists a K-homomorphism a: K[x] ---> C such that a(x) = y.
We can extend a to N; let a' denote this extension. As N is a normal
extension of K, a' defines an element of Gal(NIK) such that a'(x) of. x;
consequently, x ~ I(Gal(NIK)). <>

The main theorem of the Galois correspondence given in §8.5 extends to
all Galois extensions of finite degree, using the same proof as the one given
in §8.5.
Toward Chapter 16
It would be possible to develop the different aspects of Galois theory almost
indefinitely; we choose to stop at this point and devote one final chapter
to giving some idea of two domains of current research.
16
Recent Developments
16.1 The Inverse Problem of Galois Theory

16.1.1 The Problem
Is every finite group the Galois group of an extension of the field Q? The
answer to this question is not yet completely known.
The study of Galois groups of polynomials of degree 2, 3, and 4 enables
us to assert that any subgroup of S2, S3, S4 is the Galois group of an
extension of Q. We also gave examples of polynomials in Q[X] with Galois
group Sp, p 2': 5 prime (see §12.3 and Exercise 12.1).
In 1981, Jean-Pierre Serre indicated that aside from abelian groups, the
answer is actually known for very few groups; until now, however, it has
been positive in every case. The groups An and Sn (Hilbert, 1892), solvable
groups (a result whose proof is extremely difficult, due to Shafarevitch,
1954), the groups PSL (2, IF p) for certain values of p ... all have been shown
to occur as Galois groups of extensions of Q.
The situation has developed considerably over the last few years. In
1985, work by Belyi, Fried, Llorente, Matzat, Thompson, and others showed
that 18 of the 26 simple groups known as the sporadic groups occur as
Galois groups of extensions of the field Q. Many other results have been
proved since then (see the books by B. Matzat and J.-P. Serre listed in the
bibliography) .

262 16. Recent Developments
16.1.2 The Abelian Case

PROPOSITION. - Let G be a finite abelian group. Then there exists a finite
Galois extension K ofQ havivg Galois group isomorphic to G.
PROOF. - We know that G is a finite product of cyclic groups: G
II Z/n/Il, where the ni mayor may not be distinct. For every i E I,
iEI
choose a prime Pi such that Pi = 1 mod ni. By Lejeune-Dirichlet's theo-
rem of arithmetic progressions (which states that if a and b are relatively
prime, then the arithmetic progression (an + b)nEN contains an infinite
number of primes) we can choose the Pi all distinct.
Set N = II
Pi and ( = e 2i7rfN . By §9.5, group Gal(Q[(]IQ) is iso-
iEI
morphic to the group U(N) of invertible integers modulo N. By §9.1.2,
U(N) ~ IIU(Pi) ~ II Z/(Pi - l)Z. Because Pi - 1 = kini, U(N) has
iEI iEI
H = II niZ/(Pi - l)Z as a subgroup. Writing H' for the corresponding
iEI
subgroup of Gal(Q[(]IQ), K = I(H') is a normal extension of Q, since H'
is a normal subgroup of G; its Galois group is U(N)/H ~ G. <>
16.1.3 Example
Let us consider the case where G = Z/3Z x Z/3Z x Z/5Z. We have nl =
n2 = 3, n3 = 5; we can take PI = 7, P2 = 13, P3 = 11, which leads to
N = 1,001.
We know that U(N) ~ U(7) x U(13) x U(ll) ~ Z/6Z x Z/12Z x Z/10Z,
and we set H ~ 3Z/6Z x 3Z/12Z x 5Z/10Z.
16.2 Computation of Galois Groups over Q for

Small-Degree Polynomials
The whole of this section concerns Galois groups over Q. The first prob-
lem is to effectively compute the Galois group of a given polynomial in
Q[X]. The computations are actually more approachable than what Galois
believed, thanks to computers; the search for algorithms to compute the
Galois group of a polynomial has developed a great deal since the 1970s,
using ideas that can be traced back to work of Lagrange from 1770. By
1995, algorithms to determine the Galois groups of all polynomials of de-
gree less than or equal to 11 had been implemented. For higher degrees,
Galois' remark remains valid: for polynomials of degree 15, for example,
one is dealing with groups that are subgroups of S15, a group having more
16.2 Computation of Galois Groups over Q for Small-Degree Polynomials 263
than 1.3 x 10 12 elements. The following sections are based on an article by

Richard Stauduhar, listed in the bibliography.
16.2.1 Simplification of the Problem

Let P be a polynomial of degree n of Q[X], and let N be the splitting field
of P. To determine Gal(NIQ), we can:
1) Reduce to a polynomial having only simple roots. Indeed, we know
that the polynomial S = P/gcd(P, PI) has the same roots as P, but
they are all simple roots: thus, its splitting field is still N, and its
Galois group is the same as that of P.
2) Next, reduce to a polynomial having integral coefficients. It suffices

to multiply P by the least common multiple of the denominators of
the coefficients of P.
3) Finally, reduce to a monic polynomial with integral coefficients. To

accomplish this, if P(X) = L
akXk E Z[X], it suffices to con-
sider the polynomial P 1 (Y) = yn + L ak(a n )n-k-lyk. As

O:'Ok:'On-l
P1 (a n X) = (a n )n- 1 p(x), the polynomial PI has the same Galois
group as P.
Because of this, we will assume from now on that P is a monic polynomial
in Z[X].
16.2.2 The Irreducibility Problem

Recall that there exist algorithms to decompose any polynomial in Z[X]
into a product of polynomials in Z[X], all of which are irreducible in Q[X].
These algorithms apply to all polynomials of small degree which we consider
here. The starting point of these methods is the article by Berlekamp on
the factorization of polynomials with coefficients in a finite field (1967).
If a polynomial is not irreducible, it is not always easy to find its Galois
group in terms of the Galois groups of its irreducible factors. One can try
to adapt the methods described below.
From now on, we assume that P is a monic irreducible polynomial in
Z[X].
16.2.3 Embedding ofG into Sn

Recall what we did in §8.1.4. If E denotes the set of roots of Pin C, then
a bijection 'P : {I, ... , n} - t E defines an indexation of the roots of P,
Xi = 'P(i), giving an embedding : G -+ 5 n . Two bijections {I, ... ,n}-+

E define embeddings G ----> 5 n whose images are conjugate subgroups of
5 n . We will only identify the Galois group as a subgroup of 5 n up to
conjugation.
16.2.4 Looking for G Among the Transitive Subgroups of Sn

The group (G) is a transitive subgroup of 5 n (cf §8.1.4); to know which
one, we can undertake to make a list of all the transitive subgroups of 5 n
up to conjugation, and to find criteria for eliminating subgroups from the
list one by one, until only one remains. Such criteria can be obtained by
considering polynomial functions of the roots, the first examples of which
were given by Lagrange. This procedure gives the structure of G without
giving an explicit isomorphism from G to the corresponding subgroup of
5n ·
In the remainder of this chapter, we will restrict ourselves to the case
n = 4, except in §16.2.6, where we state a general result.
16.2.5 Transitive Subgroups of S4

The list of transitive subgroups of 54 is as follows:
1) Order 24: 54;
2) Order 12: the alternating group A4 whose elements are the identity,
the eight 3-cycles, the three double transpositions;
3) Order 8: three subgroups isomorphic to the dihedral group D 4 ; these

groups are mutually conjugate, since they are the 2-Sylow subgroups
of 54; the elements of these subgroups can be obtained by numbering
the vertices of the square described in §8.7.2 in different ways via the
bijections j, g, h (Table 16.1).
j 9 h
A ----> 1 A---->l A---->l
B---->2 B---->2 B -+ 3
C---->3 C-+4 C---->2
D---->4 D---->3 D---->4
TABLE 16.1.
Hl = {id, (12), (34), (12)(34), (13)(24), (14)(23), (1324), (1423)},

H2 = {id, (13), (24), (12)(34), (13)(24), (14)(23), (1234), (1432)},
H3 = {id, (14), (23), (12)(34), (13)(24), (14)(23), (1243), (1342)}.
4) Order 4: V = {id,(l 2)(3 4), (1 3)(2 4), (1 4)(2 3)}, the Klein
Vierergruppe;
5) Three mutually conjugate cyclic subgroups of the subgroups of order

8 of S4, generated by 4-cycles:
a) Xl : id,(l 3 2 4), (1 2)(3 4), (1 4 2 3),
b) K2 : id,(l 2 3 4), (1 3)(2 4), (1 4 3 2),
c) K3 : id,(l 2 4 3), (1 4)(2 3), (1 3 4 2).
Our results are shown (up to conjugation) in Figure 16.1.
S4
/~
D4 A4
/~/
Z14Z ~
FIGURE 16.1.
16.2.6 Study oJ if> (G) cAn

Set
D = IT (Xi - Xj)2
l<O;i<j<O;n
and
d = IT (Xi - Xj)
l<O;i<j<O;n
(D is the discriminant of P). Let T(X l , ... ,Xn) = IT

(Xi - Xj), and
l<O;i<j<O;n
let t be a transposition of Sn. We have t(T) = -T; we can assume that
t = (n - 1 n). In this case, the sign change comes from t(Xn - l - Xn) =
-(Xn-l - Xn). It follows that s(T) = T if sEAn and s(T) = -T if
s rf:. An·
PROPOSITION. - We have IP(G) c An if and only if dE Z.

PROOF. - Suppose IP(G) C An. For every a E G, we have s = (a) E An
so a(d) = S(T)(XI, ... ,Xn ) = T(xl, ... ,xn ) =d. Consequently, d E Q. If
we write d = p/q with (p, q) = 1, we know that p2/q2 = D E Z since it is
the resultant of a monic polynomial in Z[X], so q = ±1 and d E Z.
Suppose IP(G) ct An- There exists a in G such that s = lP(a) tf. An, so
a(d) = S(T)(XI' ... ,Xn ) = (-'[')(XI' ... ,Xn ) = -d. Hence d tf. Z. <)
The case n = 3. If d E Z, we have G ~ A3 ~ Z/3Z. If d tf. Z, we have

G~S3·
The case n = 4. If d E Z, we can have G ~ A4 or G ~ V If d tf. Z, we can

have G ~ S4, G ~ D4 or G ~ Z/4Z.
REMARK ON THE TEST d E Z. - This test comes down to seeking the

integral roots of the polynomial
R(X) = X2 - D = (X - d)(X + d),
known as the A 4 -resolvent, a polynomial whose coefficients are known to
be integral. We can compute D exactly, by successive divisions, and look
to see if it is a square in Z. We can also compute a good approximation
(say on the order of 10- 9 ) of each of the roots, both real and complex, of
P. This gives a good approximation .6. of D; then, knowing that D is an
integer, we take D to be the nearest integer to the approximate value .6..
16.2.7 Study of Cf>( G) C D4

More precisely, the point is to determine if IP( G) is contained in one of
the three subgroups of S4 isomorphic to D 4. Consider the polynomial
U(X I , ... ,X4) = X I X 2 + X 3 X 4 introduced by Lagrange (cf. §10.8). If
s E S4, then S(U)(XI"'" X 4) can take one of the three values
U(Xl, . .. ,X4),
(23)(U)(X I , ... , X 4) = X I X 3 + X 2X 4,
(24)(U)(XI" .. ,X4) = X IX 4 + X 2X 3 .
The elements of S4 that leave U fixed are those of the group HI men-
tioned in §16.2.5, and those that leave (23)U fixed are those of the group
(23)HI(23) = H 2. Finally, those that leave (24)U fixed are those of the
group (24)HI(24) = H 3 .
Let us form the D 4-resolvent, the polynomial defined by
We now need an elementary result on algebraic integers; this is an extremely

important notion in number theory which we have not used up to now. A
algebraic number over Q is an algebraic integer if its minimal polynomial

has integral coefficients. We will assume the (true) result that the sum and
the product of two algebraic integers are also algebraic integers; this result
can be shown using resultants.
PROPOSITION. - R(X) E Z[X].

PROOF. -
1) By the result just cited, the roots of R are algebraic integers, so the
coefficients of R are also algebraic integers.
2) For every a E G, we have a(R)(X) = R(X); the coefficients of R

are invariant under every element of G, so they lie in Q and are
computable using results on symmetric polynomials (see Chapter 3).
3) The only algebraic integers in Q are the elements of Z, which gives

the result. ¢
PROPOSITION. - Assume that R has only simple roots. Then,
cJ>(G) CHI if and only if + X3X4

XIX2 is an integral root of R,
cJ>(G) C H2 if and only if XIX3 + X2X4 is an integral root of R,
cJ>(G) C H3 if and only if XIX4 + X2X3 is an integral root of R.
PROOF. - Suppose that cJ>(G) C HI. For every a E G, we have s = cJ>(a) E

HI, so
a(XIX2 + X3X4) = xIX2 + X3X4·
Consequently, XIX2+X3X4 E Q. Because XIX2+X3X4 is an algebraic integer,
it follows that XIX2 + X3X4 E Z.
Suppose that cJ>(G) ct. HI. Then there exists a in G such that s = cJ>(a) (j.
HI, so a(XIx2 + X3X4), which is a conjugate of XIX2 + X3X4, is one of
the two other roots of R. Because the roots of R are distinct, we have
a(xIx2 + X3X4) =f. XIX2 + X3X4·
Consequently, XIX2 + X3X4 (j. Q, so XIX2 + X3X4 (j. Z. ¢
REMARK. - In the case where R has a multiple root, the preceding criterion
is not applicable. This difficulty can be avoided, for instance, by modifying
the polynomial without changing its Galois group.
16.2.8 Study ofCP(G) C Z/4Z

Here, again, we want to determine if cJ>( G) is contained in one of the three
subgroups of 8 4 isomorphic to Zj 4Z. This study can be done only in the
case where the preceding paragraph gives rise to a positive answer. Assume,
therefore, that we have shown that (G) C H2, and let us seek to determine
if (G) C K 2 . Consider the polynomial
V(X1, ... , X 4) = x1xi + x 2xi + x 3xl + x 4xl-

If s E H 2, then s(V)(X 1, ... , X 4) can take one of the two values
V(X 1, ... ,X4 ) or

(13)(V)(Xl' ... ,X4) = x 3xi + X 2Xr + x1xl + x4xi.
The elements of H2 that leave V fixed are those of the group K 2 .
Let us form the (71/471)-resolvent, the polynomial given by
R(X) = (X -(XIX~+X2X~+X3X~+X4Xi))(X -(X3X~+X2Xi+XIX~+X4X~)).
PROPOSITION. - Suppose that R has only simple roots. Then (G) C K2

if and only if y = XIX~ + X2X~ + X3X~ + x4xi is an integral root of R.
PROOF. - The proof is the same as the proof of the proposition of the
preceding section.
If (G) C K 2 , then for every a E G, we have s = (O') E H 2 , so O'(y) = y.
Consequently, y E IQ. As y is an algebraic integer, it follows that y E 7l.
If (G) ct. K 2 , there exists a in G such that s = (O') r¢ K 2 , so O'(y),
which is a conjugate of y, is the other root of R. Because the roots of R
are distinct, we have o'(Y) i- y. Consequently, y r¢ 7l. <>
REMARK. - We can also study (G) C Kl or (G) C K3 by this method,
by renumbering the roots.
16.2.9 An Algorithm for n = 4

Let P be a polynomial of degree 4 in IQ[X].
1) We first reduce to the case of a monic polynomial with integral coef-
ficients (see §16.2.1).
2) We check that P is irreducible over IQ.
3) We compute good approximations of the roots of P in Co
4) We test to see if (G) C A4 (see §16.2.6).
5) We test to see if (G) C D4 (see §16.2.7).

If the answer is
4 : yes, 5 : yes, then G c:::' V;

4 : yes, 5 : no, then G c:::' A4;
4 : no, 5: no, then G c:::' 54;
4 : no, 5: yes, we do test 6) below.
6) We test to see if cI>(G) c '.£/4'.£ (see §16.2.8).

If the answer is
6: yes, then G c:::' '.£/4'.£;

6 : no, then G c:::' D4.
BIBLIOGRAPHICAL NOTE. - To study recent developments of these meth-

ods, see the articles by A. Valibouze and their bibliographies.
Bibliography
The recent books cited here are available in most university libraries. For
the older books, it takes a bit of luck to find them, and you may find
yourself searching for years, like the scholars of the Middle Ages seeking
for rare manuscripts.
Recall that an introduction to Galois theory can be found in every basic
algebra book; we have made use of such books in writing this one, for
example, N. Bourbaki (chapter V of Algebra), the more advanced book
Algebra, volume 2 by MacLane and Birkhoff, Jacobson's Basic algebra, and
Lang's Algebra.
ABEL Niels Henrik (1802-1829). - Oeuvres completes (French, two vol-

umes). Imprimerie de Grondhal & Son, Christiania, 1881.
The two articles on the impossibility of resolving equations of degree 5
and higher are contained in volume 1, pp. 28-33 and 66-94. The second
of these articles first appeared in Journal fur die reine und angewandte
Mathematik, founded in 1825 by a friend of Abel named August Leopold
Crelle. Many other articles by Abel were published in this journal. For a
biography of Abel, see the book by Oystein Ore listed below.
ABHYANKAR Shreeram S. - Fundamental group of the affine line in pos-

itive characteristic. In Geometry and analysis (Bombay, 1992). Tata Inst.
Fund. Res., Bombay, 1995, pp. 1-26.
In this article, the author discusses what he calls the surprise of the
century. Let P be a polynomial of degree n in K[X], irreducible over K.
Choose a root Xl of P, and set PI = PIX - -Xl. If PI is irreducible
over K[X1J, choose a root X2 of P 1 and set P2 = Pd X - -X2' Iterate

this procedure until you obtain a reducible polynomial. Among the results
obtained by this method, note the following surprise: if Pl, P 2 , P3 , P4 , and
P s are all irreducible, then P6 , ... ,Pn - - 3 are irreducible as well!
AD DIN AT TUSI SHARAF. - (Euvres (French, 2 volumes), R. Rashed,

ed. Les Belles Lettres, Paris, 1986.
APMEP. - Presence d'Bvariste Galois. no. 48, 1982, 57 pages.

This booklet by the French Association of Mathematics Teachers in
Public Education reproduces about 15 texts by Galois and also contains
texts by R. Taton: "Evariste Galois et ses contemporains", J. Dieudonne:
"L'influence de Galois", and A. Dahan-Dalmedico: "Resolubilite des equa-
tions par radicaux et premier memoire d'Evariste Galois".
ARTIN Emil. - Galois theory. Dover Publications, Mineola, NY, 1998 (orig.
publ. 1942).
Artin is one of the founders of the modern exposition of Galois theory.
ARTIN Michael. - Algebra. Prentice-Hall, Englewood Cliffs, NJ, 1991.

A recent and very agreeable book, at the level of juniors and seniors in
college, covering various different parts of algebra, Galois theory in partic-
ular (pp. 492-584).
BAKER Alan. - Transcendental number theory. Cambridge Mathematical

Library, Cambridge University Press, Cambridge, U.K., 1990.
The transcendence of e and 7r is proved in the very first pages of this
small book.
BOURGNE Robert, AZRA J.-P. - Bcrits mathmatiques d'Bvariste Galois.

Gauthier-Villars & cie, Paris, 1962,541 pages.
This book (reedited in 1976) contains every extant work by Galois,
down to his smallest computations, and 15 pages of reproductions of his
manuscripts. Additional details can be found in the articles written by the
authors/editors for the 1983 Abel-Galois conference in Lille.
CARDAN Girolamo. - Hieronymi Cardani, prtEstantissimi mathematici,

philosophi, ac medici, ARTIS MAGNJE, sive de regulis algebraicis ... Ioh.
Petreium, Nuremberg, 1545.
This book treats the resolution of all of the different cases of the resolu-
tion of third-degree equations in detail. An English translation is available
(M.LT. Press, 1968, 267 pages).
CARREGA Jean-Claude. - Theorie des corps. La regIe et Ie compas. Her-

mann, Paris, 1981 (new ed. in 1989), 277 pages.
This is an elementary book that contains a great deal of geometry.
CHILDS Lindsay. - A concrete introduction to higher algebra. Springer-

Verlag, New York, 1979.
This book contains a presentation of Berlekamp's algorithm.
DALMAS Andr. - Evariste Galois revolutionnaire et geometre. Le nouveau

commerce, Paris, 1982 (new ed. of the 1956 orig.), 184 pages, 16 pages of
documents.
This book contains an account of Galois' life, some texts by Galois, and
some documents, written by a writer who is in general rigorous, although
his tentative identification of the "infamous coquette" may be questionable.
DOUADY Adrien, DOUADY Regine. - Algebre et theories galoisiennes.

Cedic, Paris, 1979.
The Galois theory can be found on pages 75-117 of volume 2 of this
advanced book.
EDWARDS Harold M. - Galois theory. Springer-Verlag, New York, 1984.

In this book, the theory is presented according to the methods of those
who created it: Lagrange, Gauss, Abel, Galois, and the others.
EULER, Leonhard (1707-1783). - Complete works. 29 volumes.
GAAL, Lisl. - Galois theory with examples. Chelsea Publ. Co, New York,
1973, 248 pages.
GARLING D.J.H. - Galois theory. Cambridge University Press, New York,

1986, 167 pages.
Elementary.
GAUSS Carl Friedrich (1777-1855). - Recherches arithmetiques (trad. par

A.-C.-M. Poullet-Delisle). Courcier, Paris, 1807.
The original work by Gauss was in Latin, published in 1801. It contains
his research on the quadratic reciprocity law and roots of unity, in partic-
ular, on the regular polygon with 17 sides.
GIRARD Albert (1595-1632). - Invention nouvelle en l'algebre par Albert

Girard mathematicien. Tant pour la solution des equations, que pour recog-
noistre le nombre des solutions qu 'elles rer;oivent, avec plusieurs choses
qui sont necessaires it La perfection de ceste divine science, suivi de: de la
mesure de La superfice (sic) des triangles et polygones sphericques, nouvelle-
ment inventee. Guillaume Iansson Blaeuw, Amsterdam, 1629,64 pages.
HOUZEL Christian. - Les principaux themes dans l'histoire des equations

algebriques. In Actes de l'Universite d'ete sur l'histoire des mathematiques,
1984, Commission Inter-IREM Epistemologie.
This article was the basis of Chapter 1 of this book.
HOUZEL Christian. - La resolution des equations algebriques. In Sciences

a l'epoque de la Revolution franr;aise. Albert Blanchard, Paris, 1988, pp.
17-37.
This is a very precise study of work on algebraic equations from Euler
to Vandermonde, Lagrange, and Ruffini, with a complete bibliography.
IREM de Paris VII. - M.A. T.H., vol. 2, pp. 70-77.

This is a presentation of part of al Khwarizmi's work on second-degree
equations.
LACROIX, Sylvestre Fran~ois (1765-1843). - Complement des elements

a
d'algebre l'usage de l'Ecole centrale des Quatre Nations. Bachelier, Paris,
1835.
This sixth edition contains the final note on Galois quoted in Chapter
13.
LAGRANGE Joseph Louis (1736-1813). - Refiexions sur la resolution alge-

brique des equations. In Nouveaux memoires de l'Academie de Berlin, vol-
umes of 1770 and 1771.
LAGRANGE Joseph Louis. - (Euvres. Gauthier-Villars (et fils), Paris,

1867-1892, vol. 3, pp. 205-421.
A reproduction of the 1770-1771 article.
LAGRANGE Joseph Louis. - De la Resolution des equations numefiques

de to us les degres. Duprat, Paris, an 6, 268 pages.
LAGRANGE Joseph Louis. - Traite de la resolution des equations nume-

-riques de to us les degres avec des notes sur plusieurs points de la theorie
des equations algebriques, 3rd edition. Bachelier, Paris, 1826.
LIDLR., NIEDERREITER H. - Finite fields. Cambridge University Press,

New York, 1994.
This new edition of a book first published in 1983 contains beautiful
chapters on finite fields and code theory.
MATZAT B. H. - Konstruktive Galoistheorie. Lecture Notes in mathemat-

ics, vol. 1284, Springer-Verlag, New York, 1987.
This book gives a complete, advanced exposition of recent results on the
inverse Galois problem introduced in Chapter 16.
MINAC Jan, REIS Clive. - Trigonometry in finite fields. Exposition Math.,

11, 1993.
MUTAFIAN Claude. - Equations algebriques et theorie de Galois. Vuibert,

Paris, 1980, 264 pages.
A detailed, concrete, tranquil progression containing many examples.
ORE Oystein. - Abel. Un mathematicien romantique. Berlin-Paris, 1989.

A biography of the famous, deeply imaginative Norwegian mathematician
who was celebrated by Crelle, Legendre, and Jacobi but never obtained a
job in his lifetime, either in Norway or in Berlin. He remained too poor to
marry the girl who had been his fiancee since 1823, and finally died in her
arms on April 5, 1829, at the age of 26.
RlBENBOIM Paulo. - Algebraic numbers. Wiley-Interscience, New York,

1972, 300 pages.
ROTMAN Joseph. - Galois theory. Springer-Verlag, New York, 1990, 108

pages.
Elementary.
SAMUEL Pierre. - Theone algebrique des nombres. Hermann, Paris, 1967.

This classic textbook is mainly devoted to the study of algebraic inte-
gers. It contains the proof of d' Alembert's theorem given as an exercise in
Chapter 7 of this text.
SERRE Jean-Pierre. - Gours d'arithmetique. Paris, P.U.F., 1977, 2nd edi-

tion.
Exists in English.
SERRE Jean-Pierre. - Topics in Galois Theory. Jones and Bartlett, Boston,

Londres, 1992, 117 pages.
An advanced exposition of recent result on the inverse Galois problem
introduced in Chapter 16 of this book.
STAUDUHAR Richard. - The determination of the Galois groups. In Math-

ematics of computation, vol. 27, no. 124, 1973, pp. 981-996.
STEINITZ E. - Algebraische Theorie der Korpern. Journal de Grelle, 1910,

pp. 167-309.
STEWART Ian. - Galois theory. Chapman and Hall, 1973, 226 pages.
The author of this book is well-known for his tales and drawings. His
book is extremely attractive and agreeable.
TATON Rene. - Evariste Galois et ses biographes. In Golloque Abel-Galois.

Lille, France, 1983, 30 pages.
Nothing appears to have been omitted from this extremely precise article,
which is followed by an extensive bibliography.
TIGNOL Jean-Pierre. - Ler;ons sur la theorie des equations. UCL, Louvain

la neuve, 1980.
TIGNOL Jean-Pierre. - Galois' theory of algebraic equations. Longman

Scientific & Technical, 1987, 429 pages.
This book, which is based on the 1980 "Lessons", is very pleasant. It
provides a profound explanation of the history of the theory of algebraic
equations through the work of Galois.
VANDERMONDE Alexandre TMophile (1735-1796). - Histoire de I 'Acade-

-mie Royale des Sciences de Paris. Volumes de 1771.
VAN DER WAERDEN Bartel Leendert. - Modern algebra. Frederick Ungar

Publ. Co, New York, 1949, 2 volumes.
This famous textbook on modern algebra was first edited in German,
in the 1930s; the author was strongly influenced by Emmy Noether. The
presentation of Galois theory follows Emil Artin's approach.
VAN DER WAERDEN Bartel Leendert. - A history of algebra. Springer-

Verlag, New York, 1985.
VIETE Franois (1540-1603). - Opera mathematica in unem volumen con-

gesta . .. opera atque studio Francisci Schooten Leydensis . .. Bonaventurreet
Abrahami Elzeviriorum, Lugduni Batavorum, 1646.
This book was edited in Latin by F. Van Schooten, 43 years after Viete's
death; Viete's original notation was not preserved. A literal translation by
Jean Peyroux exists, published by the bookstore A. Blanchard, 1991. Older
texts by Viete are extremely rare.
WANTZEL Pierre Laurent (1814-1848). - Recherches sur les moyens de

reconnaitre si un probleme de geometrie peut se resoudre avec la regIe et
Ie compas. In Journal de mathematiques pures et appliquees, vol. 2, 1837,
pp. 366-372.
WEBER Henri. - Traite d 'algebre superieur-e. Gauthier-Villars, Paris, 1898.

Galois theory of 100 years ago, presented without the use of linear alge-
bra.
YOUSCHKEVITCH Adolf P. - Les mathematiques arabes (VlIIieme -

XVieme siecles). Vrin, Paris, 1976.
A very precise and complete book on the subject.
Index
Abe~ 3, 6, 51, 134, 212, 221 Bombelli, 4, 17, 19

Adjoining a root, 62 Bourbaki, 228
al Khwarizmi, 5 Brauer, 199
al Uqlidisi, 4 Budan,5
Algebraic closure, 228 Burnside, 122, 134
Algebraic element, 55, 56 Byron, 219
Algebraic extension, 59
Algebraically independent elements, Cardan, 14, 18-20
211 Cauchy, 6, 134, 220
Arab,2 Cayley, 122, 225
Archimedes, 2, 13, 89 Character, 101
Argand,6 Characteristic, 229
Artin, 101, 122 Chevalley, 200
at Thsi, 2, 13 Chinese, 2
Chuquet,4
Babylonians, 1, 6 Class field theory, 157
Baker, 56 Commutator, 196
Belyi, 261 Conjugate, 93
Berlekamp, 57, 263 Constructible point, 80
Bernoulli, 5 Cousin, 221
Bertrand, 220 Cramer, 4
Besicovitch, 139 Cubic equation, 13, 19-21, 136,
Betti, 225 184
Bezout, 6, 183
Biquadratic extensions, 136 d'Alembert, 6, 114
278 Index
Dedekind, 51, 62, 93, 101, 152, Feit, 200

154,225 Fermat, 3, 149, 242
Degree Ferrari, 6, 14, 18
of an algebraic element, 56 Field
of an extension, 52 algebraically closed, 227
Descartes, 2-5, 13, 19, 21, 43, 87 of invariants, 122
Dihedral group, 128 perfect, 259
Dinet, 220 Field extension, 51
Direct product, 135 Finite-degree extension, 52
Discriminant, 37, 42 Fiore, 14
Double transpositions, 201 Fourier, 5, 220
Dumas, 222 Fried,261
Duplication of the cube, 85 Frobenius homomorphism, 231
e,65 Galois, 3, 4, 7, 51, 99, 121, 198,

Eisenstein, 154 199, 203, 207, 219, 262
Eisenstein's criterion, 58 Galois correspondence, 126, 234
Electronic transmission, 235 Galois extension, 260
Element Galois group
algebraic, 55, 56 of an extension, 119
inseparable, 258 of a polynomial, 120
primitive, 99 Galois resolvent, 100
separable, 257 Gauss, 2, 6, 43, 84, 149, 154, 183,
Elementary symmetric polynomial, 220, 243
27 Gelfond,56
Embeddings, 97 Generated extension, 54
Equation Gergonne, 220
cubic, 13, 19-21, 136, 184 Girard, 4, 5, 32
general, of degree n, 211 Greeks, 2, 13
quartic, 18, 21 Group
Eratosthene, 85 commutator, 196
Euclid,88 dihedral, 128
Euler, 4-6, 15, 22, 41, 149, 183, simple, 198
220,243 solvable, 195-197
Euler function, 149 of units, 149
Extension Group actions, 133
abelian, 179
Hermite, 21, 55, 220
algebraic, 59
Hilbert, 181, 261
cyclic, 179
Hudde, 120
Galois, 260
generated, 54 Inseparable element, 258
normal, 108 Intermediate, 51
quadratic, 52 Irreducible case, 17, 20, 21
radical, 207
separable, 257 Jacobi,220
Index 279
Jordan, 225 Polynomial

cyclotomic, 153
Kashi, 4 minimal,56
Keats, 219 separable, 57, 257
Khayyam, 2, 13 solvable by radicals, 208
K-homomorphism,94 symmetric, 26
Kronecker, 62, 112, 122, 154, 157, Primitive element, 99
225 p-subgroup, 134
Lacroix, 221 Quadratic extension, 52

Lagrange, 6, 121, 183, 188, 220, Quadratic reciprocity law, 241
243,262 Quartic equation, 18, 21, 186
Laplace, 6
Lefebure de Fourcy, 220 Raspail, 222
Legendre, 220, 243 Regular pentagon, 87
Leibniz, 4, 5 Resolvent, Lagrange, 183
Leonard of Pisa, 2, 14 Resultant, 35, 40
Lindemann, 55 Richard,220
Liouville, 6,55, 64, 152, 225 Richard J., 139
Llorente, 261 Rimbaud, 219
Rolle, 6
Mathieu, 199
Root
Matzat, 261
n-th of unity, 153
Menechme, 13
primitive n-th of unity, 153
Minimal polynomial, 56
of unity, 153
Moivre,17
of unity, 5th, 159
Moore, 51
of unity, 15th, 159
Neper,4 of unity, 17th, 160
Nerval, 222 Ruffini, 6, 212
Newton, 2, 32 Rupture field, 62
Newton's formulas, 32
Normal closure, III Schneider, 56
Normal extension, 108 Scipio del Ferro, 6, 14
Notation, 3 Separability, 257
Separable element, 257
Orbit, 133 Separable extension, 257
Origami,90 Separable polynomial, 57, 257
Oughtred,4 Serre, 261
Serret, 225
Pacioli, 14 Seventh roots of unity, 21
Pappus, 89 Shafarevitch, 261
Perfect field, 259 Splitting field, 107
Point constructible in one step, Stabilizer, 133
79 Steinitz, 228, 257
Poisson, 221 Stevin,4
280 Index
Stifel, 4 Transit ive, 133

Sturm, 5,43 Trisect ion, 2, 85
Sylow, 134 Tschirn haus, 6, 39, 188
Sylow p-subg roup, 134
Sylvest er, 41 Unit, 149
Symme tric polyno mial, 26
Symme tric rationa l functio n, 26 Vander monde, 2, 6, 149, 183
Viete, 2, 18, 19, 21
Tartagl ia, 14, 19
Thomp son, 200, 261 Wantze l, 2, 84
Tower rule, 53 Waring , 6
Towers, 52 Weber, 51, 62, 122, 157
Transc endenta l, 55
Zech's logarith m, 237
Graduate Texts in Mathematics
(continued from page ii)
65 WELLS. Differential Analysis on Complex 97 KOBLITZ. Introduction to Elliptic Curves

Manifolds. 2nd ed. and Modular Forms. 2nd ed.
66 WATERHOUSE. Introduction to Affine 98 BROCKERiTOM DIECK. Representations of
Group Schemes. Compact Lie Groups.
67 SERRE. Local Fields. 99 GROVE/BENSON. Finite Reflection Groups.
68 WEIDMANN. Linear Operators in Hilbert 2nded.
Spaces. 100 BERG/CHRlSTENSENiRESSEL. Harmonic
69 LANG. Cyclotomic Fields II. Analysis on Semigroups: Theory of
70 MASSEY. Singular Homology Theory. Positive Definite and Related Functions.
71 FARKASiKRA. Riemann Surfaces. 2nd ed. 101 EDWARDS. Galois Theory.
72 STILLWELL. Classical Topology and 102 VARADARAJAN. Lie Groups, Lie Algebras
Combinatorial Group Theory. 2nd ed. and Their Representations.
73 HUNGERFORD. Algebra. 103 LANG. Complex Analysis. 3rd ed.
74 DAVENPORT. Multiplicative Number 104 DUBROVINlFoMENKoINOVIKOV. Modem
Theory. 3rd ed. Geometry-Methods and Applications.
75 HOCHSCHILD. Basic Theory of Algebraic Part II.
Groups and Lie Algebras. 105 LANG. SL2(R).
76 IrTAKA. Algebraic Geometry. 106 SILVERMAN. The Arithmetic of Elliptic
77 HECKE. Lectures on the Theory of Curves.
Algebraic Numbers. 107 OLVER. Applications of Lie Groups to
78 BURRIS/SANKAPPANAVAR. A Course in Differential Equations. 2nd ed.
Universal Algebra. 108 RANGE. Holomorphic Functions and
79 WALTERS. An Introduction to Ergodic Integral Representations in Several
Theory. Complex Variables.
80 ROBINSON. A Course in the Theory of 109 LEHTO. Univalent Functions and
Groups. 2nd ed. Teichmiiller Spaces.
81 FORSTER. Lectures on Riemann Surfaces. 110 LANG. Algebraic Number Theory.
82 BOTTiTU. Differential Forms in Algebraic 111 HUSEMOLLER. Elliptic Curves.
Topology. 112 LANG. Elliptic Functions.
83 WASHINGTON. Introduction to Cyclotomic 113 KARATZAS/SHREVE. Brownian Motion and
Fields. 2nd ed. Stochastic Calculus. 2nd ed.
84 lRELANDIROSEN. A Classical Introduction 114 KOBLITZ. A Course in Number Theory and
to Modem Number Theory. 2nd ed. Cryptography. 2nd ed.
85 EDWARDS. Fourier Series. Vol. II. 2nd ed. ,115 BERGERIGOSTIAUX. Differential Geometry:
86 VAN LINT. Introduction to Coding Theory. Manifolds, Curves, and Surfaces.
2nd ed. 116 KELLEy/SRINIVASAN. Measure and Integral.
87 BROWN. Cohomology of Groups. Vol. I.
88 PIERCE. Associative Algebras. 117 SERRE. Algebraic Groups and Class Fields.
89 LANG. Introduction to Algebraic and 118 PEDERSEN. Analysis Now.
Abelian Functions. 2nd ed. 119 ROTMAN. An Introduction to Algebraic
90 BR0NDSTED. An Introduction to Convex Topology.
Polytopes. 120 ZIEMER. Weakly Differentiable Functions:
91 BEARDON. On the Geometry of Discrete Sobolev Spaces and Functions of Bounded
Groups. Variation.
92 DIESTEL. Sequences and Series in Banach 121 LANG. Cyclotomic Fields I and II.
Spaces. Combined 2nd ed.
93 DUBROVINlFoMENKOINOVIKOV. Modem 122 REMMERT. Theory of Complex Functions.
Geometry-Methods and Applications. Readings in Mathematics
Part I. 2nd ed. 123 EBBINGHAuslHERMEs et al. Numbers.
94 WARNER. Foundations of Differentiable Readings in MathematiCS
Manifolds and Lie Groups. 124 DUBROVINlFoMENKOINOVIKOV. Modem
95 SHIRYAEV. Probability. 2nd ed. Geometry-Methods and Applications.
96 CONWAY. A Course in Functional Part III.
Analysis. 2nd ed.
125 BERENSTEIN/GAY. Complex Variables: An 156 KECHRIS. Classical Descriptive Set
Introduction. Theory.
126 BOREL. Linear Algebraic Groups. 2nd ed. 157 MALLIAYIN. Integration and
127 MAsSEY. A Basic Course in Algebraic Probability.
Topology. 158 ROMAN. Field Theory.
128 RAUCH. Partial Differential Equations. 159 CONWAY. Functions of One
129 FULTONiHARrus. Representation Theory: A Complex Variable II.
First Course. 160 LANG. Differential and Riemannian
Readings in Mathematics Manifolds.
130 DODSONIPOSTON. Tensor Geometry. 161 BORWEINIERDELYI. Polynomials and
131 LAM. A First Course in Noncommutative Polynomial Inequalities.
Rings. 162 ALPERINIBELL.Groupsand
132 BEARDON. Iteration of Rational Functions. Representations.
133 HARRIs. Algebraic Geometry: A First 163 DrxON/MORTIMER. Permutation Groups.
Course. 164 NATHANSON. Additive Number Theory:
134 ROMAN. Coding and Information Theory. The Classical Bases.
135 ROMAN. Advanced Linear Algebra. 165 NATHANSON. Additive Number Theory:
136 ADKINS/WEINTRAUB. Algebra: An Inverse Problems and the Geometry of
Approach via Module Theory. Sumsets.
137 AxLERIBOURDONiRAMEY. Harmonic 166 SHARPE. Differential Geometry: Cartan's
Function Theory. Generalization of Klein's Erlangen
138 COHEN. A Course in Computational Program.
Algebraic Number Theory. 167 MORANDI. Field and Galois Theory.
139 BREDON. Topology and Geometry. 168 EWALD. Combinatorial Convexity and
140 AUBIN. Optima and Equilibria. An Algebraic Geometry.
Introduction to Nonlinear Analysis. 169 BHATIA. Matrix Analysis.
141 BECKERiWEISPFENNINGIKREDEL. Grabner 170 BREDON. Sheaf Theory. 2nd ed.
Bases. A Computational Approach to 171 PETERSEN. Riemannian Geometry.
Commutative Algebra. 172 REMMERT. Classical Topics in Complex
142 LANG. Real and Functional Analysis. Function Theory.
3rd ed. 173 DIESTEL. Graph Theory. 2nd ed.
143 Doo8. Measure Theory. 174 BRIDGES. Foundations of Real and
144 DENNIS/FARB. Noncommutative Abstract Analysis.
Algebra. 175 LICKORISH. An Introduction to Knot
145 VICK. Homology Theory. An Theory.
Introduction to Algebraic Topology. 176 LEE. Riemannian Manifolds.
2nd ed. 177 NEWMAN. Analytic Number Theory.
146 BRIDGES. Computability: A 178 CLARKEILEDYAEV/STERN/WOLENSKI.
Mathematical Sketchbook. Nonsmooth Analysis and Control
147 ROSENBERG. Algebraic K-Theory Theory.
and Its Applications. 179 DOUGLAS. Banach Algebra Techniques in
148 ROTMAN. An Introduction to the Operator Theory. 2nd ed.
Theory of Groups. 4th ed. 180 SRIVASTAVA. A Course on Borel Sets.
149 RATCLIFFE. Foundations of 181 KRESS. Numerical Analysis.
Hyperbolic Manifolds. 182 WALTER. Ordinary Differential
150 EISENBUD. Commutative Algebra Equations.
with a View Toward Algebraic 183 MEGGINSON. An Introduction to Banach
Geometry. Space Theory.
151 SILVERMAN. Advanced Topics in 184 BOLLOBAS. Modern Graph Theory.
the Arithmetic of Elliptic Curves. 185 COX/LITTLE/O'SHEA. Using Algebraic
152 ZIEGLER. Lectures on Polytopes. Geometry.
153 FULTON. Algebraic Topology: A 186 RAMAKRIsHNANNALENZA. Fourier
First Course. Analysis on Number Fields.
154 BROWNIPEARCY. An Introduction to 187 HARRIS/MORRISON. Moduli of Curves.
Analysis. 188 GOLDBLATT. Lectures on the Hyperreals:
155 KASSEL. Quantum Groups. An Introduction to Nonstandard Analysis.
189 LAM. Lectures on Modules and Rings. 198 ROBERT. A Course inp-adic Analysis.
190 ESMONOEIMURTY. Problems in Algebraic 199 HEoENMALMifKORENBLUNUZHU.Theory
Number Theory. of Bergman Spaces.
191 LANG. Fundamentals of Differential 200 BAO/CHERN/SHEN. An Introduction to
Geometry. Riemann-Finsler Geometry.
192 HrRscHiLACOMBE. Elements of Functional 201 HlNORY/SILVERMAN. Diophantine
Analysis. Geometry: An Introduction.
193 COHEN. Advanced Topics in 202 LEE. Introduction to Topological
Computational Number Theory. Manifolds.
194 ENGELINAGEL. One-Parameter Semigroups 203 SAGAN. The Symmetric Group:
for Linear Evolution Equations. Representations, Combinatorial
195 NATHANSON. Elementary Methods in Algorithms, and Symmetric Function.
Number Theory. 2nd ed.
196 OSBORNE. Basic Homological Algebra. 204 ESCOFIER. Galois Theory.
197 EISENBUOIHARRrS. The Geometry of 205 FELIX/HALPERINITHOMAS. Rational
Schemes. Homotopy Theory.

Jean-Pierre Escofier Auth. Galois Theory PDF

Uploaded by

Copyright:

Available Formats

Jean-Pierre Escofier Auth. Galois Theory PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Jean-Pierre Escofier Auth. Galois Theory PDF

Uploaded by

Copyright:

Available Formats

Graduate Texts in Mathematics 204

(continued after index)

Translated by Leila Schneps

Mathematics Subject Classification (2000): l1R32, I1S20, 12FIO, 13B05

Library of Congress Cataloging-in-Publication Data

Printed on acid-free paper.

© 2001 Springer-Verlag New York, Inc.

Production managed by Francine McNeill; manufacturing supervised by Joe Quatela.

ISBN 0-387-98765-7 SPIN 10711904

Springer-Verlag New York Berlin Heidelberg

• a digression on constructions with ruler and compass (Chapter 5),

• beautiful applications (Chapters 9 and 10), and

1 Historical Aspects of the Resolution of Algebraic

2 Resolution of Quadratic, Cubic, and Quartic Equations 9

2.2.7 Fran<;ois Viete . 18

4.4.2 Transcendental Numbers . . . . . . . . . . . . 55

5 Constructions with Straightedge and Compass 79

7 Normal Extensions 107

8 Galois Groups 119

8.1.5 A Short History of Groups 121

9 Roots of Unity 149

10 Cyclic Extensions 179

10.4.1 The Norm . . . . . . . . . . . . . . . . . . . . . 181

11 Solvable Groups 195

12 Solvability of Equations by Radicals 207

13 The Life of Evariste Galois 219

14 Finite Fields 227

15 Separable Extensions 257

16 Recent Developments 261

16.2.3 Embedding of G into Sn . . . . . . . . . . . . . . . . 263

1.1 Approximating the Roots of an Equation

It is not possible to give here the full history of approximations as devel-

1.2 Construction of Solutions by Intersections of

1.3 Relations with Trigonometry

1.4 Problems of Notation and Terminology

{+;P +ll:nn BHp} requabitur E.

B in A quadratum plus D plano in A requari Z solido,

1.5 The Problem of Localization of the Roots

1.6 The Problem of the Existence of Roots

Done i1 [c faut refouven·ir· d'ub[crvcr tousjours ecla : on pourroit dire 1I

has n real or complex roots (1749: Researches on the imaginary roots of

1.7 The Problem of Algebraic Solutions of

2.1 Second-Degree Equations

J.-P.Escofier, Galois Theory

This problem discusses the quadratic equation llx 2 + 7x = 6.15; the

Base 60 Base 10 Computation of

You will multiply 3.30 by 12.15

You will add it to 1.8.45 1.21 81

It is the square of 9 9 Vb: - ac

The inverse of 11 cannot be

1 _~2 + Vb24 -ac

The side of the square is 30.

TABLE 2.1. Method for solving a quadratic equation

OTHER EXAMPLES. - Here are the equations corresponding to other prob-

COMMENTARY. - In these problems, the solutions are always positive num-

2.1.2 The Greeks

2.1.3 The Arabs

2.1.4 Use of Negative Numbers

2.2 Cubic Equations