i

i
IrvingBook 2013/5/22 15:39 page i #1

Beyond the
Quadratic Formula

i
i

i
IrvingBook 2013/5/22 15:39 page ii #2

c 2013 by the Mathematical Association of America, Inc.

Library of Congress Catalog Card Number 2013940989
Print edition ISBN 978-0-88385-783-0
Electronic edition ISBN 978-1-61444-112-0
Printed in the United States of America
Current Printing (last digit):
10 9 8 7 6 5 4 3 2 1

i
i

i
IrvingBook 2013/5/22 15:39 page iii #3

Beyond the
Quadratic Formula

Ron Irving
University of Washington

Published and Distributed by

The Mathematical Association of America

i
i

i
IrvingBook 2013/5/22 15:39 page iv #4

Council on Publications and Communications

Frank Farris, Chair
Committee on Books
Gerald M. Bryce, Chair
Classroom Resource Materials Editorial Board
Gerald M. Bryce, Editor
Michael Bardzell
Jennifer Bergner
Diane L. Herrmann
Paul R. Klingsberg
Mary Morley
Philip P. Mummert
Mark Parker
Barbara E. Reynolds
Susan G. Staples
Philip D. Straffin
Cynthia J Woodburn

i
i

i
IrvingBook 2013/5/22 15:39 page v #5

CLASSROOM RESOURCE MATERIALS

Classroom Resource Materials is intended to provide supplementary classroom material for studentslaboratory exercises, projects, historical information, textbooks with unusual approaches for presenting mathematical
ideas, career information, etc.
101 Careers in Mathematics, 2nd edition edited by Andrew Sterrett
Archimedes: What Did He Do Besides Cry Eureka?, Sherman Stein
Beyond the Quadratic Formula, Ronald S. Irving
Calculus: An Active Approach with Projects, Stephen Hilbert, Diane Driscoll
Schwartz, Stan Seltzer, John Maceli, and Eric Robinson
Calculus Mysteries and Thrillers, R. Grant Woods
Conjecture and Proof, Miklos Laczkovich
Counterexamples in Calculus, Sergiy Klymchuk
Creative Mathematics, H. S. Wall
Environmental Mathematics in the Classroom, edited by B. A. Fusaro and
P. C. Kenschaft
Excursions in Classical Analysis: Pathways to Advanced Problem Solving
and Undergraduate Research, by Hongwei Chen
Explorations in Complex Analysis, Michael A. Brilleslyper, Michael J. Dorff,
Jane M. McDougall, James S. Rolf, Lisbeth E. Schaubroeck, Richard L.
Stankewitz, and Kenneth Stephenson
Exploratory Examples for Real Analysis, Joanne E. Snow and Kirk E. Weller
Exploring Advanced Euclidean Geometry with GeoGebra, Gerard A. Venema
Geometry From Africa: Mathematical and Educational Explorations, Paulus
Gerdes
Historical Modules for the Teaching and Learning of Mathematics (CD),
edited by Victor Katz and Karen Dee Michalowicz
Identification Numbers and Check Digit Schemes, Joseph Kirtland
Interdisciplinary Lively Application Projects, edited by Chris Arney
Inverse Problems: Activities for Undergraduates, Charles W. Groetsch
Keeping it R.E.A.L.: Research Experiences for All Learners, Carla D.
Martin and Anthony Tongen
Laboratory Experiences in Group Theory, Ellen Maycock Parker

i
i

i
IrvingBook 2013/5/22 15:39 page vi #6

Learn from the Masters, Frank Swetz, John Fauvel, Otto Bekken, Bengt
Johansson, and Victor Katz
Math Made Visual: Creating Images for Understanding Mathematics, Claudi
Alsina and Roger B. Nelsen
Mathematics Galore!: The First Five Years of the St. Marks Institute of
Mathematics, James Tanton
Methods for Euclidean Geometry, Owen Byer, Felix Lazebnik, and Deirdre
L. Smeltzer
Ordinary Differential Equations: A Brief Eclectic Tour, David A. Sanchez
Oval Track and Other Permutation Puzzles, John O. Kiltinen
Paradoxes and Sophisms in Calculus, Sergiy Klymchuk and Susan Staples
A Primer of Abstract Mathematics, Robert B. Ash
Proofs Without Words, Roger B. Nelsen
Proofs Without Words II, Roger B. Nelsen
Rediscovering Mathematics: You Do the Math, Shai Simonson
She Does Math!, edited by Marla Parker
Solve This: Math Activities for Students and Clubs, James S. Tanton
Student Manual for Mathematics for Business Decisions Part 1: Probability and Simulation, David Williamson, Marilou Mendel, Julie Tarr, and
Deborah Yoklic
Student Manual for Mathematics for Business Decisions Part 2: Calculus
and Optimization, David Williamson, Marilou Mendel, Julie Tarr, and
Deborah Yoklic
Teaching Statistics Using Baseball, Jim Albert
Visual Group Theory, Nathan C. Carter
Which Numbers are Real?, Michael Henle
Writing Projects for Mathematics Courses: Crushed Clowns, Cars, and
Coffee to Go, Annalisa Crannell, Gavin LaRose, Thomas Ratliff, and
Elyn Rykken

MAA Service Center

P.O. Box 91112
Washington, DC 20090-1112
1-800-331-1MAA
FAX: 1-301-206-9789

i
i

i
IrvingBook 2013/5/22 15:39 page vii #7

To my parents,
Florence and Herbert Irving

i
i

i
IrvingBook 2013/5/22 15:39 page viii #8

i
i

i
IrvingBook 2013/5/22 15:39 page ix #9

Preface
I have also omitted here the demonstration of most of my
statements, because . . . if you take the trouble to examine
them systematically the demonstrations will present themselves to you and it will be of much more value to you to
learn them in that way than by reading them.
Rene Descartes [19, p. 192]

Every student learns the formula for the solution of a quadratic, or degree two, polynomial equation in a high school algebra course. It is one
of the few mathematical topics that many adults remember years later, at
least by name. However, the study of cubic, or degree three, and quartic, or
degree four, polynomial equations has largely disappeared from the mathematical curriculum. In the rush to calculus, high school students do not see
it. At the university level, undergraduate mathematics majors often crown
their algebraic studies with Galois theory, which provides the tools needed
to show that there is no formula for the solution of degree five equations
analogous to the quadratic formula for degree two equations. Galois Theory can also be used to show that formulas exist for solutions in degrees
three and four, but these may be skipped over.
What are the formulas? The answer is at the heart of this book. The results are both elementary and beautiful. Moreover, they are an essential part
of the history of mathematics, representing the high point in mathematical
developments of the sixteenth century.
This book has evolved from notes used in a class for in-service and
prospective secondary mathematics teachers. It is intended to be suitable
ix

i
i

i
IrvingBook 2013/5/22 15:39 page x #10

Preface

for such an audience as well as for undergraduates who desire to look at

a topic they may have unwittingly skipped over, or anyone mathematically
curious and willing to do some work. Included in this category are high
school students eager to go beyond the standard curriculum. An important
feature of the book is the sections devoted to history. They are intended to
enrich the readers learning by revealing the key discoveries in the study of
polynomial equations as milestones in intellectual history across cultures.
The mathematical content of the book parallels that found in a few chapters of a mathematical classic from 1914, Elementary Theory of Equations,
by Leonard Eugene Dickson [20]. Dickson was one of the leading American mathematicians of the first part of the twentieth century. This book is
more leisurely than Dicksons, but is similar in approach and borrows some
of the examples.
The book is designed for self study. Much of the material is presented as
exercises, with outlines for working through them. Mathematics is learned
only by doing, and that is what is asked of the reader. Doing is not making
calculations. It is necessary to learn how to write results in organized, intelligible, logical prose and, ideally, to practice describing the results to others
orally. In teaching, I try to keep lecturing to a minimum, asking students to
read the book on their own, to work through the ideas orally in class with
each other, and to present the results in writing.
In developing material on polynomial equations, how many of the foundational results on real numbers and elementary calculus should we assume? On occasion, it is useful to be able to refer to the intermediate value
theorem or basic results on turning points of a polynomials graph. For instance, we need to know, given a positive integer n, that any positive real
number has an nth root, or that if the graph of a polynomial lies below the
horizontal axis at one point and above at another, then the graph crosses
the axis somewhere in between. From calculus, knowing the relationship
between a polynomials derivative and the behavior of its graph allows conclusions to be drawn about the connection between the discriminant of a
quadratic or cubic polynomial and its roots.
I answer this question by including two sections, Section 1.5 on the
intermediate value theorem and Section 1.6 on graphs of polynomial functions, in which key ideas are sketched and theorems stated in a form that
makes them available for reference later. This allows us, for instance, to
treat discriminants early, with a small dose of calculus, returning later with
additional algebraic tools to study them purely algebraically and see them
in their proper light. Readers knowing calculus will already be familiar with
much of the material. Others I hope will be able to grasp the issues and be
comfortable using the theorems when the time comes.

i
i

i
IrvingBook 2013/5/22 15:39 page xi #11

Preface

xi

Lets take a quick tour of the book.

Chapter 1 consists of definitions and basic results on polynomials. Some
of the material, such as the discussion of graphing, is more advanced or
abstract than the material in the rest of the book and need not be understood
in detail on a first reading. A first reading gives orientation, and the chapter
can be returned to as needed.
Chapter 2 treats the quadratic formula for the solution of degree two
equations and examines it from several perspectives. The chapter closes
with a selective history of the quadratic formula, focusing on the work of
the ancient Babylonians nearly 4000 years ago and Al-Khwarizmi around
825 C.E.
The heart of the book is Chapters 3 through 6, in which we study polynomial equations of degrees three and four. The discovery of solutions to
them by several mathematicians in sixteenth-century Italy represents a high
point in mathematical history and was the most significant mathematical
accomplishment in Europe for centuries. It also makes for a good story.
Chapter 3 introduces the solution to (reduced) cubic equations that has
come to be known as Cardanos formula, in honor of Girolamo Cardano, a
co-discoverer. Following the derivation of the formula and its use in some
examples, we discuss graphs of cubic polynomials and the discriminant of
a cubic, concluding with the dramatic story of the formulas discovery and
publication.
One lesson of Chapter 3 is that Cardanos formula, when applied to
certain cubic equations, expresses the solutions in terms of square roots of
negative real numbers, obliging us to make sense of them. We do this in
Chapter 4, learning how to work with an expansion of the system of real
numbers known as the complex numbers and how to compute nth roots of
complex numbers using trigonometry. The history of complex numbers is
interesting. We close with a short account.
With complex numbers, we return in Chapter 5 to cubic equations, refining Cardanos formula and giving an algebraic description of the discriminant. We see how trigonometry can be used for the most difficult family
of cubic equations to dispense with complex numbers altogether. Our historical review of cubics picks up where it left off two chapters earlier, as
we learn how Cardanos successors came to terms with the appearance of
complex numbers in his formula.
After our three-chapter-long struggle with the complexities of cubic
equations, we enter smoother waters in Chapter 6, which is devoted to quartic equations. Out of respect for history, we begin with the original method
of solving them, due to Cardanos assistant Lodovico Ferrari, but then turn

i
i

i
IrvingBook 2013/5/22 15:39 page xii #12

xii

Preface

to Rene Descartes approach of a few decades later and Leonhard Eulers

formula from a little over a century after that. All three methods require
an intermediate step of solving a cubic equation that depends on the coefficients of the quartic. This is why the quartic waters are so much calmer;
nearby, the cubic channel continues churning, but once we master the navigation of its shoals, the quartic provides little in the way of new obstacles.
Eulers formula for the solution of a quartic equation leads to the calculation
of the quartics discriminant.
A recurring theme of the book is the abundance of information the coefficients of a polynomial encode about its roots. The quadratic formula
is the first illustration, with more appearing throughout the discussion of
cubic polynomials. The theme reaches its climax in Chapter 6, where we
learn how the discriminant and other polynomial expressions in a quartics
coefficients can be used to determine the nature of its roots. The chapter
concludes with a section inviting the reader to consolidate understanding
of the material on cubic and quartic polynomials through a series of essay
questions, followed by a historical section on the contributions of Descartes
and Euler.
Chapter 7 is an introduction to some topics in the theory of polynomials
of higher degree: quintic polynomials, the fundamental theorem of algebra,
polynomial factorization, and symmetric polynomials. The level of difficulty rises, but full understanding is not the aim. The treatment is intended
as a peek at what lies ahead, with results sketched, history surveyed, and
little proved. We conclude on a highpoint that draws many of the chapters
elements together: a proof of the fundamental theorem of algebra that is
essentially the one given by Pierre-Simon Laplace in 1795.
Acknowledgements: I thank my Math 497 students for working with early
versions of this material. Their responses guided my selection of topics and
my choice of emphasis. I am particularly grateful to Andrew Grzadzielewski
for his continued encouragement.
Another student, Spencer Hubbard, read the manuscript and did all the
exercises. His careful review and eye for detail led to many improvements,
as well as enjoyable conversations. I thank Spencer in particular for his
question about the roots of a quartic polynomial with zero discriminant,
which motivated an expansion of Section 6.6.
The panel of reviewers that the Mathematical Association of America
assembled to read an earlier version of the book did the most extraordinary
of jobs, entirely transforming my vision of what it could be. No doubt I have
fallen short, but the book is far better thanks to their advice and criticism, as

i
i

i
IrvingBook 2013/5/22 15:39 page xiii #13

Preface

xiii

well as later extensive comments by the copy editor, criticism of the history
sections by an anonymous MAA reader, and transformation of my primitive
diagrams by Beverly Ruedi. I am indebted as well to Don Albers and Jerry
Bryce for their willingness to consider this project and for their ongoing
support.
Exploring mathematics, even elementary mathematics, is a privilege,
connecting us to fellow humans across millennia and cultures in our search
for fundamental truth. (I hope this book illuminates these connections.) My
greatest debt is to the members of my family, who have allowed and encouraged me to enjoy this privilege. My parents, to whom I have dedicated this
book, arranged for me to arrive on a leap day, thereby inspiring my childhood interests in mathematics and astronomy. Their gift of Irving Adlers
The Giant Golden Book of Mathematics [2] on my second birthday sealed
my fate.

i
i

i
IrvingBook 2013/5/22 15:39 page xiv #14

i
i

i
IrvingBook 2013/5/22 15:39 page xv #15

Contents
Preface
1

ix

Polynomials
1.1 Definitions . . . . . . . . . . . . . . . . . . . . . .
1.2 Multiplication and Degree . . . . . . . . . . . . . .
1.3 Factorization and Roots . . . . . . . . . . . . . . .
1.4 Bounding the Number of Roots . . . . . . . . . . .
1.5 Real Numbers and the Intermediate Value Theorem
1.6 Graphs . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

1
1
4
8
10
12
16

Quadratic Polynomials
2.1 Sums and Products . .
2.2 Completing the Square
2.3 Changing Variables . .
2.4 A Discriminant . . . .
2.5 History . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

21
22
24
28
29
33

Cubic Polynomials
3.1 Reduced Cubics . .
3.2 Cardanos Formula
3.3 Graphs . . . . . . .
3.4 A Discriminant . .
3.5 History . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

47
47
50
58
61
66

Complex Numbers
4.1 Complex Numbers . . . . . . . . . . . . . . . . . . . . . .
4.2 Quadratic Polynomials and the Discriminant . . . . . . . .
4.3 Square and Cube Roots . . . . . . . . . . . . . . . . . . .

73
73
77
81

.
.
.
.
.

.
.
.
.
.

xv

i
i

i
IrvingBook 2013/5/22 15:39 page xvi #16

xvi

Contents

4.4
4.5
4.6
4.7
4.8
4.9
5

The Complex Plane . . . . . . . . . . . . .

A Geometric Interpretation of Multiplication
Eulers and de Moivres Formulas . . . . . .
Roots of Unity . . . . . . . . . . . . . . . .
Converting Root Extraction to Division . . .
History . . . . . . . . . . . . . . . . . . . .

Cubic Polynomials, II
5.1 Cardanos formula . . . . .
5.2 The Resolvent . . . . . . .
5.3 The Discriminant . . . . .
5.4 Cardanos Formula Refined
5.5 The Irreducible Case . . . .
5.6 Vi`etes Formula . . . . . .
5.7 The Signs of the Real Roots
5.8 History . . . . . . . . . . .
Quartic Polynomials
6.1 Reduced Quartics . . . . .
6.2 Ferraris Method . . . . . .
6.3 Descartes Method . . . . .
6.4 Eulers Formula . . . . . .
6.5 The Discriminant . . . . .
6.6 The Nature of the Roots . .
6.7 Cubic and Quartic Reprise .
6.8 History . . . . . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

84
88
92
98
101
103

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

109
109
113
115
120
124
125
130
133

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

143
143
146
149
154
157
162
167
169

Higher-Degree Polynomials
7.1 Quintic Polynomials . . . . . . . . . .
7.2 The Fundamental Theorem of Algebra
7.3 Polynomial Factorization . . . . . . .
7.4 Symmetric Polynomials . . . . . . . .
7.5 A Proof of the Fundamental Theorem .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

179
179
185
191
200
211

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

References

217

Index

223

About the Author

227

i
i

i
IrvingBook 2013/5/22 15:39 page 1 #17

1
Polynomials
We will be studying polynomial equations throughout this book, especially
those of degrees 2, 3, and 4. In this chapter, we introduce terminology and
obtain some basic results that hold for all polynomials. The reader familiar
with this material may wish to skip ahead. Others may wish on first encounter to read through this chapters definitions and results, returning for a
closer reading as they are used.

1.1 Definitions
What is a polynomial? We know that
3x 2

4x C 7

is one, as is
5x 17 C 12x 11
So are
and

4x 7 C 13x 4 C x 3
4 100 2 88
x
C x
3
5

p 333
2x

x C 113:

1
2x 200 C x 111 C x 4
3

 2:

But
x 4 C sin x
is not a polynomial, and neither is
10x :
1

i
i

i
IrvingBook 2013/5/22 15:39 page 2 #18

1. Polynomials

(Why not? For now, we can agree that they certainly dont look like polynomials. We will return to this question in Exercises 1.12 and 1.14.)
The formal definition of a polynomial is: a polynomial is an expression
of the form
an x n C an

1x

n 1

C an

2x

n 2

C


C a2 x 2 C a1 x C a0 ;

where n is a non-negative integer and an ; an 1 ; : : : ; a1 ; a0 are numbers. As

indicated in the examples, the numbers might be integers, fractions, or more
generally real numbers. (The real numbers are those numbers that mark positions on a line.) The numbers are called the coefficients of the polynomial.
The coefficient a0 is the constant term of the polynomial. Real numbers are
themselves polynomials, the constant polynomials, with 0 called the zero
polynomial.
For a non-zero polynomial, the largest integer d for which there is a
non-zero coefficient ad is called the degree of the polynomial. The expression ai x i is the degree i term of the polynomial. By the definition of degree,
the non-zero real numbers, viewed as polynomials, are degree 0 polynomials. By convention, 0 is said to have degree 1.
Ordinarily, when we write a polynomial as an x n C


C a1 x C a0 ,
we anticipate that n is its degree and an 0. Because this is not required
by the definitions, we must exercise some care. Two polynomials an x n C



C a1 x C a0 and bm x m C


C b1 x C b0 are equal if the corresponding
coefficients are equal. More precisely, they have the same degree, d say,
and ai D bi for each i D 0; 1; : : : ; d . (If m or n is greater than d , then the
indices ai or bj with i > d or j > d will be 0, by definition of degree.)
Exercise 1.1. What are the degrees of the four polynomials displayed at
the start of the section?
Polynomials of low degree have special names: degree 1 polynomials
are linear polynomials, degree 2 polynomials are quadratic polynomials
or quadratics, degree 3 polynomials are cubics, degree 4 polynomials are
quartics, and degree 5 polynomials are quintics.
A polynomial can be regarded as a function, taking on numerical values
when numbers are substituted for x. This leads to the use of functional
notation, with polynomials represented by expressions such as f .x/. Given
a polynomial f .x/ and a number a, we write f .a/ for the number we get
when we replace x by a. For example, suppose f .x/ is the polynomial
x3

3x C 2:

i
i

i
IrvingBook 2013/5/22 15:39 page 3 #19

1.1. Definitions

Then
f .2/ D 23

3
2C2D8

6 C 2 D 4:

There is nothing special about x. We may write other letters for the
variable (also called the indeterminate) of a polynomial, or any symbol in
place of x. For example,
y 3 3y C 2
is a polynomial, as are
t3

3t C 2

and
|3

3| C 2:

Although written differently, they describe the same function.

Exercise 1.2.

Is

p
3
2411

3
7
a polynomial? If so, what is its degree? If not, why not?
437 C

4 C

Given a polynomial f .x/ and a real number c, there is a natural way to

form their product c
f .x/: we multiply each coefficient of f .x/ by c. For
example,
3
.4x 3

3x 2 C 2x C 8/ D 12x 3

9x 2 C 6x C 24

and

1
3 2 1

.4x 3 3x 2 C 2x C 8/ D x 3
x C x C 2:
4
4
2
A non-zero polynomial is monic if the coefficient of its highest-degree
term is 1. As we did in the example, we can multiply any non-zero polynomial by a suitable real number to obtain a monic polynomial: Given
an x n C an

1x

n 1

C


C a1 x C a0

with an 0, multiply by 1=an .

Algebra has developed from the need to solve polynomial equations. A
polynomial equation is an equation of the form
f .x/ D 0;
with f .x/ a polynomial. A solution to this equation is a real number a
satisfying
f .a/ D 0:

i
i

i
IrvingBook 2013/5/22 15:39 page 4 #20

1. Polynomials

The number a is also called a root of f .x/. For example,

x5

3x 3

6x C 4 D 0

is a polynomial equation, and one can check that x D 2 is a solution.

Polynomial equations need not have solutions. For example,
x2 C 1 D 0
doesnt. Substituting any real number for x, we obtain as its square a positive real number or 0. Adding 1 yields a positive real number, which therefore cant be 0.
In saying a polynomial equation has no solution, we are implicitly assuming that the domain of allowable solutions is the set of real numbers. In
Chapter 4, we will introduce complex numbers, and we will find that equations such as x 2 C 1 D 0 have solutions in this expanded domain. It would
be more accurate to say that x 2 C 1 D 0 has no real number solution rather
than that it has no solution.
For any non-zero polynomial f .x/ and non-zero real number c, the
polynomial equations
f .x/ D 0
and
c
f .x/ D 0
have the same solutions. If a is a real number satisfying f .a/ D 0, then
cf .a/ D 0 also, and similarly, if cf .a/ D 0, then since c is non-zero, f .a/
must be 0. Therefore, whenever we want to solve a polynomial equation
f .x/ D 0 (with f .x/ non-zero), we can replace f .x/ by its associated
monic polynomial and solve the resulting equation instead. We will make
use of this elementary observation throughout the book.

1.2

Multiplication and Degree

Adding and multiplying polynomials is done in a first class in algebra.

Nonetheless, it is important to state carefully what the operations are in
general. The notation may seem complex, but the ideas and the results will
be simple and should be familiar.
Two polynomials can be added and multiplied to obtain new polynomials. Addition is easy to define: we simply combine terms of like degree.
Suppose r .x/ and s.x/ are polynomials given by
r .x/ D an x n C an

1x

n 1

C


C a1 x C a0

i
i

i
IrvingBook 2013/5/22 15:39 page 5 #21

1.2. Multiplication and Degree

and
s.x/ D bn x n C bn

1x

n 1

C


C b1 x C b0 :

(We do not assume that n is the degree of r .x/ or of s.x/, so an or bn may

be 0.) The sum r .x/ C s.x/ is defined to be the polynomial
.an C bn /x n C .an

C bn

1 /x

n 1

C


C .a1 C b1 /x C .a0 C b0 /:

The product is also as expected, but its description is more involved.

Given real numbers a and b and non-negative integers m and n, we define
the product of ax m and bx n to be
.ax m /
.bx n / D abx mCn :
If m D n D 0, we are multiplying the constant polynomials a and b,
and their product is the usual product ab. This rule, combined with the
distributive law, completely determines how two polynomials r .x/ and s.x/
are multiplied. If one polynomial is 0, the product is 0. Otherwise, their
product r .x/
s.x/ is the sum of the terms obtained by multiplying each
term in r .x/ by a term in s.x/.
Exercise 1.3. Let us be more explicit about the coefficients of the product
of two polynomials. Suppose r .x/ is a non-zero polynomial of degree m,
with
r .x/ D am x m C am 1 x m 1 C


C a1 x C a0 :
Suppose s.x/ is a non-zero polynomial of degree n, with
s.x/ D bn x n C bn

1x

n 1

C


C b1 x C b0 :

(i) The coefficients am and bn are non-zero. Why?

(ii) Show that the constant coefficient of r .x/s.x/ is a0 b0 .
(iii) Show that the degree 1 coefficient of r .x/s.x/ is
a0 b1 C a1 b0 :
(iv) Show that the degree 2 coefficient of r .x/s.x/ is
a0 b2 C a1 b1 C a2 b0 :

i
i

i
IrvingBook 2013/5/22 15:39 page 6 #22

1. Polynomials

(v) Show, for any non-negative integer k, that the coefficient of x k in the
product r .x/s.x/ is
ak b0 C ak

1 b1

C ak

2 b2

C


C a2 bk

C a1 bk

C a0 bk :

(It is understood that ai D 0 for i > m and bj D 0 for j > n.)

(vi) Check that the coefficient of x mCn in r .x/s.x/ is am bn .
(vii) Deduce Theorem 1.1.
Theorem 1.1. Suppose m and n are non-negative integers, r .x/ is a polynomial of degree m, and s.x/ is a polynomial of degree n. Then the product
r .x/s.x/ has degree m C n.
Exercise 1.4. Practice multiplying polynomials. Determine the product of
x 12 C 3x 7 C 4x 2

and
4x 5

3x 4 C 12x:

Write other polynomials and multiply them.

Given polynomials f .x/ and r .x/, the polynomial r .x/ divides the
polynomial f .x/ if there is another polynomial s.x/ such that
f .x/ D r .x/s.x/:
We say that r .x/ is a factor of f .x/. As a consequence of Theorem 1.1, a
factor of a non-zero polynomial f .x/ has degree less than or equal to the
degree of f .x/.
Given two positive integers a and b, long division is an algorithm for
dividing a into b, yielding a quotient q and a remainder r , with q and r nonnegative and r < a. For example, if a D 17 and b D 892, long division
produces the quotient q D 52 and the remainder r D 8. We learn this
algorithm early in our mathematical studies. From the equation b D aq C r
and the fact that r < a, it follows that a divides b precisely when r D 0.
The long division process can be adapted to polynomials, as in high
school algebra. Given two non-zero polynomials a.x/ and b.x/, we divide
a.x/ into b.x/ to obtain a quotient q.x/ and a remainder r .x/, with r .x/
having lower degree than a.x/. The next exercise serves as review.

i
i

i
IrvingBook 2013/5/22 15:39 page 7 #23

1.2. Multiplication and Degree

Exercise 1.5. Let a.x/ and b.x/ be non-zero polynomials. Let q.x/ be the
quotient obtained by long division when one divides a.x/ into b.x/ and let
r .x/ be the remainder.
(i) By the definition of quotient and remainder,
b.x/ D a.x/q.x/ C r .x/:
(ii) Using the definition of divisibility, show that if r .x/ D 0, then a.x/
divides b.x/.
(iii) Suppose instead that a.x/ divides b.x/. Show that r .x/ must be 0.
(This requires a little care. Use the definition of divisibility to produce
a polynomial p.x/ for which b.x/ D a.x/p.x/. Compare with the
equation in the first part of the exercise and use Theorem 1.1 to deduce
that r .x/ D 0.)
(iv) Conclude that to test if a.x/ divides b.x/, carry out long division and
see if the remainder is 0.
(v) Decide if x

2 divides x 3 C 6x

20.

(vi) Decide if x 2 C 2x C 1 divides x 5 C 4x 2 C 2.

Let us collect some facts about multiplication and division that are analogues of familiar facts about number multiplication and division.
Theorem 1.2. Suppose f .x/, g.x/, and r .x/ are polynomials.
(i) If f .x/ 0 and g.x/ 0, then f .x/g.x/ 0.
(ii) If f .x/g.x/ D 0, then at least one of f .x/ and g.x/ is 0.
(iii) If r .x/ 0 and r .x/f .x/ D r .x/g.x/, then f .x/ D g.x/.
Exercise 1.6.

Prove Theorem 1.2.

(i) Prove the first part either directly from the definition of multiplication
or from Theorem 1.1.
(ii) Explain why the second part follows from the first part.
(iii) Use the second part to prove the third part.
The third part of Theorem 1.2 says that when we have an equality of
polynomial products, we can cancel a non-zero factor.

i
i

i
IrvingBook 2013/5/22 15:39 page 8 #24

1.3

1. Polynomials

Factorization and Roots

The material in this section connects finding the roots of a polynomial f .x/
with finding degree 1 factors of f .x/. We will work in greater generality
than we will need in our treatment of quadratic, cubic, and quartic equations. The generality allows us to see more clearly what the issues are,
though we do more than is needed in later chapters.
Suppose we are given a polynomial f .x/ and a real number a such that
x a divides f .x/. According to the definition of division, this means there
is a polynomial s.x/ such that
f .x/ D .x

a/s.x/:

If we substitute a for x on both sides, we find that

f .a/ D .a

a/s.a/ D 0
s.a/ D 0:

Lets record what we have just found as a theorem:

Theorem 1.3. Given a polynomial f .x/ and a real number a, if x
divides f .x/, then f .a/ D 0.

Theorem 1.3 provides motivation for developing techniques to factor

polynomials, as is done in a first algebra course. To factor the quadratic
polynomial x 2 13x C 36, recognize that 36 D 4  9 and 13 D 4 C 9. This
leads to the factorization
x2

13x C 36 D .x

4/.x

9/

and shows that 4 and 9 are roots of x 2 13x C 36.

Factorization problems in a first algebra course usually deal with polynomials having integer coefficients. However, in this book we allow not just
integers as coefficients but all real numbers. Thus, when presented with the
polynomial x 2 2, we are interested in factoring it as a product
.x

r /.x

s/

for real numbers r and s. If we allow integers only, there is no such factorization, but with real numbers, we find the factorization
p
p
.x
2/.x C 2/:
p
p
Theorem 1.3 then tells us (as is already evident) that 2 and
2 are roots
of x 2 2.

i
i

i
IrvingBook 2013/5/22 15:39 page 9 #25

1.3. Factorization and Roots

Given a polynomial f .x/ and a real number a, it is useful to know that

if x a divides f .x/, then a is a root of f .x/. But will every root of f .x/
arise in this way? That is, if a is a root of f .x/, must x a divide f .x/?
Before answering this, let us review a point of logic. Many mathematical
statements, such as Theorem 1.3, take the form, If P is true, then Q is
true, where P and Q are mathematical statements. For a statement of this
type, the new statement If Q is true, then P is true is called its converse.
A statement and its converse are logically independent of each other:
both may be true, one may be true while the other is false, or both may be
false. Thus the fact that a statement is true sheds no light on the truth of its
converse. For example, we know it is true that if an integer n is divisible by
10, then n is even. The converse states that if an integer n is even, then n is
divisible by 10, and this is not true. Care is needed never to assume that the
converse of a statement is true when the statement is true.
Theorem 1.3 is an example of a statement whose converse is true:
Theorem 1.4. Given a polynomial f .x/ and a real number a, if f .a/ D 0,
then x a divides f .x/.
Theorem 1.4 allows us to limit the possibilities for roots. If we can find
all the factors of f .x/, we will be able to find its roots by examining the
degree-one factors. The proof of Theorem 1.4 is given in the next exercise.
Exercise 1.7. Let f .x/ be a polynomial of degree n and let a be a real
number. Let x D y C a.

(i) Suppose f .x/ D an x n C


C a1 x C a0 : Substitute y C a for x to get

a polynomial in y:
g.y/ D an .y C a/n C


C a1 .y C a/ C a0 :

(ii) Use Theorem 1.1 to show for each positive integer i that .y C a/i has
degree i .
(iii) Deduce that g.y/ has degree n and that therefore real numbers bn ;
bn 1 ; : : : ; b1; b0 exist for which
g.y/ D bn y n C bn
(iv) From y D x

1y

n 1

C


C b1 y C b0 :

a, get a new expression for f .x/:

f .x/ D bn .x

a/n C bn

1 .x

a/n

C


C b1 .x

a/ C b0 :

i
i

i
IrvingBook 2013/5/22 15:39 page 10 #26

10

1. Polynomials

(v) Lets determine b0 . Substitute a for x on both sides and show that
f .a/ D b0:
(vi) Conclude that
f .x/ D f .a/ C .x

a/ bn .x

a/n

C


C b2 .x

a/ C b1 :

That is, given f .x/ and a, there is a polynomial h.x/ such that
f .x/ D f .a/ C .x

a/h.x/:

(vii) Using this, prove Theorem 1.4.

Given a root a of a polynomial f .x/, we can ask how high a power of
x a divides f .x/. The root a is called a simple root if x a divides f .x/
but no higher power does. It is a multiple or repeated root if a higher power
of x a divides f .x/. The largest integer m such that .x a/m divides f .x/
is called the multiplicity of a as a root of f .x/; that is, m is the multiplicity
of a if f .x/ can be factored as .x a/m g.x/ for some polynomial g.x/ but
not as .x a/mC1 h.x/ for any polynomial h.x/.

1.4

Bounding the Number of Roots

It is natural to pursue the ideas of Section 1.3 further, and we shall do so.
However, with two exceptions, we will not use the results of this section
until Section 7.3. The exceptions are the proofs of Theorems 1.15 and 4.18,
neither of which is essential for the development of the material in this book.
Thus, this section can be omitted on a first reading.
We begin with an extension of Theorem 1.4:
Theorem 1.5. Suppose f .x/ is a polynomial and a1 ; : : : ; ak are k distinct
real numbers that are roots of f .x/. Then the polynomial
.x

a1 /.x

a2 / : : : .x

ak /

divides f .x/; that is, there is a polynomial g.x/ such that

f .x/ D .x

a1 /.x

a2 /


.x

ak /g.x/:

We will prove Theorem 1.5 in a sequence of exercises. First we comment on the use of the word or in logic or mathematics. Given two assertions
P and Q, the statement P or Q means that at least one of the two holds.

i
i

i
IrvingBook 2013/5/22 15:39 page 11 #27

11

1.4. Bounding the Number of Roots

The possibility that both hold is not excluded. For example, the statement
6 is even or 7 is even is correct, since 6 is even, and the statement 6 is
even or 8 is even is correct also. If the statement P or Q is true but Q is
not true, then P must be true.
We prepare for the proof of Theorem 1.5 with some polynomial divisibility facts.
Exercise 1.8.

Let b and c be real numbers.

(i) Show that if c 0, then the only polynomials that divide c are nonzero constant polynomials. (Use Theorem 1.1.)
(ii) Show that if x b divides x c, then b D c. (If x
for some polynomial g.x/, what must g.x/ be?)

c D .x

b/g.x/

Exercise 1.9. Suppose r .x/ and s.x/ are polynomials and a is a real number. If r .a/s.a/ D 0, then r .a/ D 0 or s.a/ D 0. (Why?) Use this and
Theorem 1.4 to deduce that if x a divides r .x/s.x/, then x a divides at
least one of r .x/ and s.x/.
We can now prove Theorem 1.5.
Exercise 1.10. Suppose f .x/ is a polynomial and a1 ; : : : ; ak are k distinct
real numbers that are roots of f .x/.
(i) Use Theorem 1.4 to show that there is a polynomial g1 .x/ satisfying
f .x/ D .x

a1 /g1 .x/:

(ii) Suppose k > 1. Since a2 is a root of f .x/, use Theorem 1.4 and
Exercise 1.9 to deduce that x a2 divides x a1 or g1 .x/.
(iii) By assumption, a1 and a2 are distinct. Use this and Exercise 1.8 to
deduce that x a2 cannot divide x a1 , and so therefore x a2
divides g1 .x/.
(iv) Conclude that there is a polynomial g2 .x/ satisfying g1 .x/ D .x
a2 /g2 .x/, and therefore that
f .x/ D .x

a1 /.x

a2 /g2 .x/:

(v) Suppose k > 2. Make a similar argument to show that x a3 divides

.x a1 /.x a2 / or g2 .x/. Show that the first possibility cant hold,

i
i

i
IrvingBook 2013/5/22 15:39 page 12 #28

12

1. Polynomials

so that x a3 must divide g2 .x/. Conclude that there is a polynomial

g3 .x/ satisfying g2 .x/ D .x a3 /g3 .x/, and therefore that
f .x/ D .x
(vi) Repeat k

a1 /.x

a2 /.x

a3 /g3 .x/:

3 additional times to obtain a polynomial gk .x/ satisfying

f .x/ D .x

a1 /.x

a2 /


.x

ak /gk .x/:

An immediate consequence of Theorem 1.5 is the following important

result about a polynomial and its roots:
Theorem 1.6. Suppose f .x/ is a polynomial of positive degree n. Then
f .x/ has at most n distinct roots.
Exercise 1.11. Prove Theorem 1.6 using Theorem 1.5 and a comparison
of degrees.
We noted in Section 1.1 that sin x is not a polynomial. This now follows
as a consequence of Theorem 1.6.
Exercise 1.12. Deduce from Theorem 1.6 that sin x cannot be equal to
a polynomial. (Hint: How many solutions are there to sin x D 0?) Show
similarly that cos x and tan x are not polynomials.

1.5

Real Numbers and the Intermediate

Value Theorem

In this section, we temporarily leave algebra to introduce the intermediate

value theorem. We will make no effort to prove it, since the proof depends
on foundational results about the real numbers whose development would
take us too far afield. But the theorem is easily stated, matches our intuition,
and will be useful to have for reference, as is explained in more detail at the
start of Section 1.6.
The primary question we wish to answer about a polynomial f .x/ is
whether it has a root. Viewing f .x/ as a function, we are asking whether
f .x/ takes on the value 0 when a real number is substituted for x, or equivalently, whether the graph of y D f .x/ crosses the x-axis.
The answer to this question depends on an understanding of the properties of the real numbers. To see why, lets consider the simplest nontrivial example for which we may find ourselves searching for a root. Every

i
i

i
IrvingBook 2013/5/22 15:39 page 13 #29

1.5. Real Numbers and the Intermediate Value Theorem

13

degree-one polynomial ax C b has a root (what is it?), so we may turn to

degree 2. Given a positive real number r , does the polynomial x 2 r have a
root? Equivalently, does r have a square root? Lets be specific: does 2 have
a square root?
The first numbers we learn about are the integers: 0; 1; 2; : : : ; and their
negatives. Next we learn about rational numbers, the ratios a=b of two integers a and b, with b 0. No integer has a square equal to 2. Nor is
there a rational number whose square is 2. This is a famous discovery often
credited to Pythagoras some 2500 years ago. You can rediscover it in the
following exercise.
Exercise 1.13. Prove that 2 does not have a rational square root.
(i) Suppose a and b are positive integers such that a=b is a square root
of 2. Assume further that a=b is a reduced fraction; that is, no positive
integer divides a and b other than 1. Square and clear denominators to
obtain a2 D 2b 2.
(ii) Show that 2 must divide a.
(iii) Show that 2 must divide b.
(iv) This is a contradiction, proving that a and b cant exist.
We learn early in our mathematical education that 2 does have a square
root, and that it is given by a decimal expansion that we can describe to as
many terms as we wish. Squaring 1 and 2, we find that 12 is less than 2 and
22 is greater than 2, so we anticipate that the square root, if it exists, lies
between them. With further testing, we find that 1:42 is too small and 1:52
is too large, so the square root should
p lie between them. Iterating, we get
better and better approximations to 2. For instance, after a few minutes
work (or almost no work at all, when armed with a calculator), we get the
approximation 1:4142, whose square is 1:99996164, or the poorer approximation 1:4143, whose square is 2:0002449, and we conclude that the square
root lies between the two numbers. Continuing, we obtain as many terms as
we wish of a decimal expansion that we take to be the desired square root.
The process of determining a square root of 2 in this way may seem natural, but its rigorous justification requires laying a proper foundation for the
definition and properties of the real numbers. This book is not the place to
lay such a foundation. We will instead review some essential facts about real
numbers and functions on real numbers, along with their consequences for
the study of polynomials. Many introductory real analysis texts cover this

i
i

i
IrvingBook 2013/5/22 15:39 page 14 #30

14

1. Polynomials

material. Our standard reference will be Steven G. Krantzs Real Analysis

and Foundations [38].
We must start with a formal definition of the real numbers. We first
learn that a real number consists of an integer or whole number part and a
fractional part given by a possibly infinite decimal expansion. By moving
from rational numbers to this larger class, we fill the holes. Any real number
can be approximated by a rational number as closely as desired: we truncate
the real numbers decimal expansion farther and farther out to get better
and better approximations. But only by allowing the full infinite expansion
do we succeed in filling the holes, assigning a real number to every point
on the number line. We will omit a detailed development of this process,
relying instead on the readers intuition about real numbers as given by
decimal expansions. (Krantz provides a treatment in two parts, introducing
the properties of the set of real numbers and stating a theorem that such a
set of numbers exists [38, pp. 5862], then providing the technical proof of
the theorem in an appendix [38, pp. 7173].)
Given two real numbers a < b, we introduce two sets of real numbers
associated to a and b, the closed interval a; b, which consists of all numbers x satisfying a  x  b, and the open interval .a; b/, consisting of all
numbers x satisfying a < x < b.
Once the real numbers are introduced, we can study properties that sets
of real numbers may satisfy, such as connectedness. We will omit the formal definition, but from it we can prove two basic results [38, pp. 145147]:
open and closed intervals are connected, and any connected set of real numbers that includes the numbers r < s will contain the entire closed interval
r; s.
We also introduce the notion of continuity, for functions defined on
the real numbers or a suitable subset. We will not go into the details (see
[38, pp. 159164]). Informally, a continuous function is one whose graph
doesnt skip around or have gaps. From the formal definition, we can verify
that any constant function f .x/ D c is continuous, as is f .x/ D x. We can
also prove that sums and products of continuous functions are continuous,
from which it follows that any polynomial function is continuous. This is
sufficiently important to record as a theorem:
Theorem 1.7. A polynomial with real coefficients, regarded as a function
on the real numbers, is continuous.
Another key result in the study of continuous functions is the theorem
[38, pp. 169170] that a continuous function takes a connected set of real

i
i

i
IrvingBook 2013/5/22 15:39 page 15 #31

1.5. Real Numbers and the Intermediate Value Theorem

15

numbers to a connected set. Combining this with the result that a connected
set of real numbers containing r and s contains all the numbers in between,
we can easily deduce the intermediate value theorem [38, p. 170].
Theorem 1.8 (intermediate value theorem). Let a < b be real numbers and
let f be a continuous function defined on the closed interval a; b. Given
a real number d between f .a/ and f .b/, there is a real number c in a; b
such that f .c/ D d .
Geometrically, the intermediate value theorem tells us that if the graph
of a continuous function goes through two different heights, then the graph
goes through every height in between as it passes from one height to the
other. Although the theorem is fundamental to the development of calculus,
its proof is usually deferred to an advanced calculus or real
analysis course.
Since polynomials are continuous functions, the intermediate value theorem applies to them, yielding the following result.
Theorem 1.9 (intermediate value theorem for polynomials). Let a < b be
real numbers and let f .x/ be a polynomial. Given a real number d between
f .a/ and f .b/, there is a real number c in a; b such that f .c/ D d .
As an example of the intermediate value theorem in action, given a positive real number d that is not the square of an integer, we can find nonnegative integers a < b with a2 < d < b 2. By the intermediate value
theorem, there is a real number c satisfying a < c < b and c 2 D d . So it
is a consequence of the intermediate value theorem that every positive real
number has a positive square root. This is sufficiently important to record
as a theorem.
Theorem 1.10. Given a positive real number d , there exists a real number
c such that c 2 D d , so every positive real number has a square root.
The same argument works when we replace the exponent 2 by any positive integer n, showing that every positive real number has a positive nth
root.
Here is another application of the intermediate value theorem, stated for
polynomials although it holds for any continuous function:
Theorem 1.11. Let a < b be real numbers and let f .x/ be a polynomial
for which f .a/ < 0 and f .b/ > 0. Then there is a real number c between
a and b such that f .c/ D 0.

i
i

i
IrvingBook 2013/5/22 15:39 page 16 #32

16

1. Polynomials

In terms of graphs, the theorem tells us that if there is a point on the

graph of y D f .x/ that is below the x-axis and another point that is above
the x-axis, then there must be a point of the graph on the x-axis.

1.6

Graphs

In Section 1.5, we introduced some results on real numbers. Here, we will

continue our excursion into realms beyond algebra, this time turning to analytic geometry to study the shape of the graph of a polynomial function.
We will give merely an overview. Calculus gives the results of this section
as easy exercises. Some of the results can be obtained by more elementary
considerations, albeit at the cost of a little more work, but either way, foundational theorems on the construction and properties of the real numbers
are essential for the proofs.
We will use the results of this section in Sections 2.4, 3.3, and 3.4, where
the discriminants of quadratic and cubic polynomials are studied. In Sections 4.2 and 5.3, the results will be proved again algebraically, so that their
derivation does not depend on the results of this section. Nonetheless, the
approach to discriminants using graphs in Sections 2.4 and 3.4 provides
additional insight and intuition.
The polynomial functions whose graphs are the simplest are the powers
of x. We know that the graph of y D x is the line forming a 45 angle with
the x-axis. Suppose n is an integer greater than 1. For positive real numbers
a < b and r < s, the inequality ar < bs holds. Setting r D a and s D b,
we find that a2 < b 2. For a positive integer n, we can repeat the argument
n times to obtain an < b n . This tells us that the graph of y D x n is always
increasing in height as x increases, for x  0.
y

Figure 1.1. Graph of y D x n

i
i

i
IrvingBook 2013/5/22 15:39 page 17 #33

17

1.6. Graphs

We also know that the graph of y D x n increases in height without

bound; that is, for any positive number N , there is a real number r for which
r n > N . (For N > 1, the inequality will hold if we take r to be an integer
greater than N .) We can use the intermediate value theorem, Theorem 1.9,
to deduce that as x increases, with x > 0, the function x n assumes all
possible positive real number values. This leads to the next theorem, and to
the sketch in Figure 1.1 illustrating the behavior of the graph of y D x n to
the right of the y-axis.
Theorem 1.12. Let n be an integer greater than 1.
(i) For x  0, the graph of y D x n is strictly increasing; that is, an < b n
for any real numbers a and b with 0  a < b.
(ii) The graph of y D x n goes through all non-negative heights; that is, for
any non-negative real number s, there is a non-negative real number r
satisfying r n D s.
One feature of the graph in Figure 1.1 that requires calculus to justify
is the way the shape has been drawn. Its what is called concave up. Those
familiar with calculus will recognize that this follows from the fact that for
x > 0, the second derivative n.n 1/x n 2 of x n is positive.
y

y = x4

y = 3x 3

Figure 1.2. Graphs of even and odd powers of x

We have discussed x n only for x  0. If x < 0, then . x/n D x n
for n even. This tells us that the graph to the left of the y-axis is a mirror
image of the graph to the right. Thus, as x increases, the graph falls steadily
from arbitrarily large heights to .0; 0/, then rises arbitrarily high. If instead
n is odd, then . x/n D x n , so that the graph to the left of the y-axis is

i
i

i
IrvingBook 2013/5/22 15:39 page 18 #34

18

1. Polynomials

obtained from the graph to the right by taking the mirror image across the
y-axis and then reflecting again across the x-axis. We see from this that the
graph rises steadily as x increases, from arbitrarily low heights to .0; 0/ and
then onwards to arbitrarily high heights. This is illustrated in Figure 1.2.
The graph of a general polynomial will be more complicated, but some
features remain unchanged from that of a simple power of x. For example,
consider the graph of the polynomial x 7 6x 5 C 11x 3 6x depicted in
Figure 1.3. Like the graph of an odd power of x, it rises to infinity to the
right and drops to infinity to the left. The complicating feature is that it rises
and falls in between. Counting, we find that it makes six turns on its way
from minus infinity to infinity. (Those familiar with calculus will recognize
that a seventh-degree polynomial can have no more turns, though it may
have fewer. For instance, x 7 has no turns.)
y

Figure 1.3. Graph of y D x 7

6x 5 C 11x 3

6x

Lets discuss the general picture. Suppose that f .x/ is a monic polynomial of positive degree n,
x n C an

1x

n 1

C


C a1 x C a0 :

We can factor x n out to obtain


an 1
a1
a0 
f .x/ D x n 1 C
C


C n 1 C n :
x
x
x

For x very large, the terms other than 1 inside the parentheses will be small
in absolute value and f .x/ will be close to x n . Using estimate arguments
typical of calculus, we can convert this idea into a proof that there is a positive real number N such that for x  N , the graph of f .x/ rises steadily,
and without bound. Since the graph rises without bound, we can further
conclude by the intermediate value theorem that the graph goes through every height above f .N /. For x negative but large in absolute value, the same

i
i

i
IrvingBook 2013/5/22 15:39 page 19 #35

19

1.6. Graphs

argument allows us to show, if n is even, that the graph of f .x/ decreases

from unbounded heights as x increases through negative values and, if n
is odd, that the graph increases from arbitrarily low levels. We summarize
what we have found in the next theorem.
Theorem 1.13. Let f .x/ be a monic polynomial of positive degree n. There
exists a positive real number N for which f .x/ satisfies:
(i) Given real numbers b > a  N , we have f .a/ < f .b/.
(ii) For any real number s > f .N /, there is a real number r > N such
that f .r / D s. That is, the graph of f .x/ increases for x  N and
goes through all heights above f .N /.
(iii) Given real numbers a < b 
and f .a/ < f .b/ if n is odd.

N , we have f .a/ > f .b/ if n is even

(iv) If n is even, then for any real number s > f . N /, there is a real
number r < N such that f .r / D s; if n is odd, then for any real
number s < f . N /, there is a real number r < N such that f .r / D
s. That is, if n is even (odd), the graph of f .x/ decreases (increases)
for x  N and goes through all heights above (below) f . N /.
When n is even, since the graph of f .x/ falls, then rises, it may never
cross the x-axis. This is the graphical way of stating that an even-degree
polynomial may have no roots. For instance, as we have seen, x 2 C 1 has
no roots.
In contrast, by Theorem 1.13, an odd-degree polynomial will assume
negative values for x negative of large absolute value and positive values
for x positive and large. By Theorem 1.11, the polynomial must assume the
value 0 somewhere. This is important enough to record as a theorem:
Theorem 1.14. Let f .x/ be a polynomial of odd degree. Then there is a
real number r such that f .r / D 0. That is, every odd-degree polynomial
has a root.
Theorem 1.13 tells us that the graph of a degree-n polynomial f .x/ behaves like the graph of x n in its extremes. Away from them, the graph may
behave differently from the graph of x n . It may shift from rising to falling to
rising multiple times, as illustrated by the seventh-degree polynomial graph
of Figure 1.3.
A graph changes from falling to rising or rising to falling at what are
called turning points. Let us define this precisely. A local maximum of a
function f .x/ is a point .a; f .a// on its graph with the property that there

i
i

i
IrvingBook 2013/5/22 15:39 page 20 #36

20

1. Polynomials

exists a positive number r such that every c in the open interval .a r; aCr /
satisfies f .c/  f .a/. Similarly, .b; f .b// is a local minimum if there is a
positive number s such that every c in the open interval .b s; b C s/
satisfies f .b/  f .c/. A turning point for f .x/ is a point that is either a
local maximum or a local minimum.
An example of a turning point is the point .0; 1/ on the graph of f .x/ D
x 2 C 1, for which it is a local minimum. In fact, since f .x/ > 1 for any
x 0, the point .0; 1/ is more than a local minimum. Its what is called a
global minimum.
Readers familiar with calculus will know, given a differentiable function
y D f .x/, that for .a; f .a// to be a turning point, f 0 .a/ must equal 0.
Thus, the turning points, if there are any, will be among the points where the
derivative vanishes. The converse does not hold: f 0 .a/ may equal 0 without
.a; f .a// being a turning point. An example is x 3 , whose derivative at 0 is
0, but .0; 0/ is not a turning point, since the function f .x/ D x 3 is always
increasing. (The graph is rising when x 0 but flat at x D 0.)
If f .x/ is a polynomial of positive degree n, we learn in calculus how to
compute its derivative f 0 .x/, and find that f 0 .x/ is itself a polynomial, of
degree n 1. By Theorem 1.6, f 0 .x/ can have at most n 1 distinct roots;
that is, the derivative of f .x/ vanishes at at most n 1 points. Since these
are the only candidates for turning points, we obtain the next theorem.
Theorem 1.15. A polynomial of positive degree n has at most n 1 turning
points.
We can use Theorems 1.13 and 1.15 to see again that the sine function
cannot be a polynomial, and to see that the exponential function 10x isnt a
polynomial.
Exercise 1.14. This exercise assumes familiarity with sin x and 10x as
functions defined for all real values of x.
(i) The graph of y D sin x has the shape of an infinite wave, oscillating
between peaks of height 1 and valleys of height 1. Describe the turning points of sin x and deduce that it has infinitely many local minima
and infinitely many local maxima.
(ii) Deduce from Theorem 1.15 that sin x cannot be a polynomial.
(iii) Use Theorem 1.13 to give an alternative proof that sin x cannot be a
polynomial.
(iv) For x < 0, we have 0 < 10x < 1. Deduce from Theorem 1.13 that the
function y D 10x cannot be a polynomial.

i
i

i
IrvingBook 2013/5/22 15:39 page 21 #37

2
Quadratic Polynomials
The heart of this book is the study of solutions to cubic and quartic equations, which we will begin in Chapter 3. This chapter is devoted to quadratic
equations. Even though they are familiar from a first algebra course, a close
look is warranted, as a warmup before we tackle the greater difficulties of
cubic and quartic equations and to introduce themes that will recur as we
study cubics and quartics.
The general quadratic equation has the form
ax 2 C bx C c D 0;
for real numbers a, b, and c, with a 0. The quadratic formula for the
solutions of this equation takes the form
p
b
b 2 4ac
xD

:
2a
2a
Since a 0, we are free to divide both sides of the equation ax 2 C
bx C c D 0 by a before searching for solutions. This amounts to setting
a D 1 and studying quadratic equations of the form
x 2 C bx C c D 0:

We will take this approach throughout this chapter. The quadratic formula
then takes the simpler form
p
b
b 2 4c
xD

:
2
2
We will obtain the quadratic formula (in this simpler form) by three
different approaches, and turn to some history at the end.
21

i
i

i
IrvingBook 2013/5/22 15:39 page 22 #38

22

2. Quadratic Polynomials

2.1

Sums and Products

Our initial approach to the quadratic formula depends on solving a seemingly different problem, the determination of two numbers given their sum
and product. Lets begin with a related but simpler problem.
Given the sum and difference of two numbers, can we determine the
numbers? For instance, suppose I tell you that I am 36 years older than my
son and that our ages sum to 84. Can you determine how old we are?
Exercise 2.1.

Solve the age problem.

Exercise 2.2. Solve the general form of the age problem. Suppose r and s
are unknown numbers, with u D r C s and v D r s. Determine r and s
in terms of u and v.
We see from the solution to Exercise 2.2 that we can determine two
numbers from their sum and difference. It is also possible to determine two
numbers from their sum and product. Doing so requires the calculation of
a square root, which leads to an ambiguity in sign. But it turns out to be
harmless, since the two solutions it yields are the two we seek.
Exercise 2.3. Given two numbers r and s, suppose that u D r C s and
v D r s, as in Exercise 2.2. Set p D r s.
(i) Verify that
.r C s/2 .r s/2 D 4r s;
so

u2

v 2 D 4p:

(ii) Write this as

v 2 D u2

4p

and take square roots to obtain a formula for v in terms of u and p,

correct up to an ambiguity in the sign.
(iii) Using Exercise 2.2, find r and s in terms of u and p and conclude that
Theorem 2.1 holds.
Theorem 2.1. Given two real numbers r and s, let u be their sum and p
their product. Then r and s are the two quantities
p
u
u2 4p

:
2
2

i
i

i
IrvingBook 2013/5/22 15:39 page 23 #39

23

2.1. Sums and Products

Because of the uncertainty in sign arising from taking a square root, we

cant specify which expression in Theorem 2.1 is r and which is s. However,
this doesnt matter, since we are interested only in identifying the pair of
numbers r and s, not in knowing which is which.
Theorem 2.1 is essentially the quadratic formula. Before we see this,
we will obtain a general result about quadratic polynomials that is a consequence of Theorem 1.4, which stated that if a polynomial f .x/ (of any
degree) has a root a, then x a divides f .x/.
Exercise 2.4. Let b and c be real numbers and let f .x/ be the polynomial
x 2 C bx C c.
(i) Suppose f .x/ factors as .x r /.x s/ for distinct real numbers r and
s. Show that r and s are the only roots of f .x/. (Hint: Suppose t is a
root. Substitute it for x.)

(ii) Conversely, show that if distinct real numbers r and s are roots of f .x/,
then f .x/ D .x r /.x s/. (Hint: Use Theorem 1.5.)

r /2. Show that r is the only root of f .x/.

(iii) Suppose f .x/ factors as .x

(iv) Conclude by proving Theorem 2.2.

Theorem 2.2. Let f .x/ D x 2 C bx C c for real numbers b and c. Exactly
one of the three possibilities occurs:
(i) x 2 C bx C c has two distinct real roots r and s, and factors as
.x

r /.x

s/:

(ii) x 2 C bx C c has only one real root r , and factors as .x

r /2 .

(iii) x 2 C bx C c has no real roots, and cannot be factored as a product of

degree 1 polynomials.
Given a quadratic polynomial x 2 C bx C c with real coefficients b and c
and with real roots r and s, to find a formula for r and s in terms of b and c,
it would suffice by Theorem 2.1 to find formulas for r C s and r s in terms
of b and c. We could then recover r and s from the values for r C s and r s.
This turns out to be easy to do.
Exercise 2.5. Suppose that the polynomial x 2 C bx C c has real roots r
and s, distinct or equal. Theorem 2.2 yields a factorization of x 2 C bx C c
as
.x r /.x s/:

i
i

i
IrvingBook 2013/5/22 15:39 page 24 #40

24

2. Quadratic Polynomials

(i) Multiply the expression .x

r /.x

s/ to show that

r Cs D b
and
r s D c:
(ii) Use Theorem 2.1 to obtain expressions for r and s in terms of b and c.
This proves Theorem 2.3.
Theorem 2.3 (Quadratic Formula). Let b and c be real numbers and suppose x 2 C bx C c has real roots r and s (distinct or coincident). Then r and
s are the two quantities
p
b
b 2 4c

:
2
2
We see from Theorem 2.1 and Exercise 2.5 that the quadratic formula
is the expression for the roots of x 2 C bx C c in terms of their sum b and
their product c.
Let us consider how the quadratic formula of Theorem 2.3 provides
a solution for quadratic equations. In the simplest case, with b D 0, the
equation has the form
x2 C c D 0
p
c. If c > 0,
and the quadratic formula tells us that the solution is x D
then c has no square roots and there is no solution. If c < 0, then c is
positive and has two square roots. The quadratic formula does not tell us
how to compute them. It tells what we already know, that the solutions to
the equation are the square roots. Thus, rather than regarding the formula
as a way to solve an arbitrary quadratic equation, we should view it as a
reduction technique. It gets us to the point of having to do a square root
calculation, then leaves us on our own. As we saw in Section 1.5, calculating
square roots is not a problem of algebra.

2.2

Completing the Square

We now turn to a second approach to deriving the quadratic formula, completing the square.
Exercise 2.6.

Solve the quadratic equation

x 2 C 10x

39 D 0:

i
i

i
IrvingBook 2013/5/22 15:39 page 25 #41

25

2.2. Completing the Square

(i) Write the equation as

x 2 C 10x D 39:
(ii) Divide 10 by 2 to get 5 and add 52 to both sides to obtain
x 2 C 10x C 52 D 39 C 52 :
The left side is the square of x C 5.
(iii) Write the equation as
.x C 5/2 D 64;
take square roots on both sides, and conclude that the equation has
solutions x D 3 and x D 13.
We can now take up the general case.
Exercise 2.7.

Let a, b, and c be real numbers.

(i) Verify that

.x C a/2 D x 2 C 2ax C a2 :
(ii) Deduce that the polynomial
x 2 C bx C

b2
4

is the square of a degree one polynomial.

(iii) Rewrite
x 2 C bx C c
by adding and subtracting b 2 =4 and find that solving the equation x 2 C
bx C c D 0 is equivalent to solving
.x C b=2/2 D d=4
for a suitable real number d . Write d in terms of b and c.
(iv) Deduce for d < 0 that there is no solution to x 2 C bx C c D 0.
(v) Deduce for d D 0 that x 2 C bx C c factors as .x C b=2/2 and the one
solution to x 2 C bx C c D 0 is x D b=2.
(vi) Deduce for d > 0 that there are two solutions to x 2 C bx C c D 0,
involving the square root of d . Write the solutions in terms of b and c,
giving Theorem 2.4.

i
i

i
IrvingBook 2013/5/22 15:39 page 26 #42

26

2. Quadratic Polynomials

Theorem 2.4. Let b and c be real numbers.

(i) If b 2

4c < 0, then the equation

x 2 C bx C c D 0

has no solutions.
(ii) If b 2 4c D 0, then the only solution of x 2 Cbx Cc D 0 is x D b=2.
(iii) If b 2 4c > 0, then the equation x 2 C bx C c D 0 has two solutions,
given by
p
b
b 2 4c
xD

:
2
2
The quadratic formula takes on a simpler form if we alter how we write
the coefficients in our initial quadratic equation. Given the real numbers b
and c, let B D b=2 and C D c. Then b D 2B, c D C , and
x 2 C bx C c D 0
can be written as
x 2 C 2Bx

C D 0;

or
x 2 C 2Bx D C:
For the remainder of the section, we take this as our standard form for
a quadratic equation. With this notation, we can appreciate the process of
completing the square geometrically.
Algebraically, to make the left side of x 2 C 2Bx D C a perfect square,
we add B 2 to both sides, yielding
x 2 C 2Bx C B 2 D B 2 C C
or
.x C B/2 D B 2 C C:
Taking square roots, we recover the quadratic formula, now in the form
p
x C B D B2 C C ;

or

xD B

p
B2 C C :

The change in our coefficients results in a quadratic formula without the

cluttering appearances of 2 and 4.

i
i

i
IrvingBook 2013/5/22 15:39 page 27 #43

27

2.2. Completing the Square

Lets reconsider completing the square geometrically. Lengths cant be

negative numbers. Let us assume, then, that B and C are positive, and in
taking the square root of a positive number, we allow only the positive
square root.
We begin with the equation
x 2 C 10x D 39
of Exercise 2.6, so that B D 5 and C D 39. Draw a square whose sides
have an unknown length x and place rectangles atop it and to its right with
sides of lengths 5 and x, as in Figure 2.1.

5
5
5

Figure 2.1. Completing the Square

Since the square has area x 2 and each rectangle has area 5x, the figure
drawn has area x 2 C 10x. We are given that
x 2 C 10x D 39;
so the area is also 39.
Looking at the figure, we can complete it by filling in the missing square
in the upper right corner, which has side length 5. If we do fill it inthereby
completing the squarewe get a square with side length x C 5. Its area is
.x C5/2. But also, since we are adding a 55 square to a figure that has area
39, we know that the area is 25 C 39, or 64. Equating the two expressions
for the area, we obtain
.x C 5/2 D 64:

i
i

i
IrvingBook 2013/5/22 15:39 page 28 #44

28

2. Quadratic Polynomials

p
The larger square has side length
p given by 64, so the original square has
side length x given by 5 C 64, or 3. We have solved the quadratic equation geometrically, and we have seen that the algebraic process of completing the square has a geometric counterpart.
The general case is handled similarly. Given x 2 C 2Bx D C , with B
and C arbitrary positive numbers, we draw the square whose sides have
an unknown length x, then place rectangles atop it and to the right with
sides of lengths B and x. The resulting figure has area x 2 C 2Bx, which
we recognize as equal to C , thanks to the given equation. We complete
the square by placing a square of side length B in the upper right corner,
yielding a larger square with side length x C B. Its area is both .x C B/2
and B 2 C C , yielding
.x C B/2 D B 2 C C:
From this we obtain
xCB D
or

xD BC

B2 C C

p
B2 C C :

We have derived the quadratic formula geometrically.

2.3

Changing Variables

Let us turn to our third approach for deriving the quadratic formula, the
technique known as change of variables. We begin with a now-familiar example.
Exercise 2.8.

Solve the equation

x 2 C 10x

39 D 0

by changing variables.
(i) Let y related to x by x D y 5. Substitute y
and obtain an equation in y.

5 for x in the equation

(ii) The new quadratic equation has the form y 2 d D 0 for a constant d .
We have a quadratic equation without a degree-one term.
(iii) Take square roots to obtain two values for y, then use x D y
obtain two solutions to the original quadratic equation.

5 to

i
i

i
IrvingBook 2013/5/22 15:39 page 29 #45

2.4. A Discriminant

29

We now take up the general case, with the goal of finding a change of
variable that, as in Exercise 2.8, eliminates the degree-one term.
Exercise 2.9. Let f .x/ D x 2 C bx C c. Let a be a real number. Introduce
a new variable y by x D y C a or y D x a.
(i) Substitute y C a for x in f .x/ and get a polynomial g.y/ in y. Determine it explicitly in terms of a, b, and c.
(ii) Examine g.y/ and determine a, in terms of b and c, so that g.y/ has
the form y 2 d . The number d will be expressible in terms of b and
c.
(iii) With this a, the solutions to g.y/ D 0 are the square roots of d .
p
(iv) Use the relationship between x and y to obtain that x D a d .
(v) Show that this is the quadratic formula.
We will use this technique in our treatment of cubic and quartic equations. It will allow us to drop the term of second-highest degree from the
polynomial, as here we were able to drop the degree-one term.

2.4 A Discriminant
A quadratic polynomial x 2 C bx C c may have two distinct real roots, one
repeated real root, or no real roots, as we saw in Theorem 2.2. It is possible
to determine which occurs from b and c, without finding the roots.
This is evident from the quadratic formula, since the roots, if they exist,
are
p
b
b 2 4c

:
2
2
We see that if b 2 4c > 0, there are two distinct real roots, if b 2 4c D 0,
there is one multiple root, and if b 2 4c < 0, there are no real roots. We can
obtain this independently of the roots by studying the shape of the graph of
y D x 2 C bx C c.
To do so, we first need to review the shape of the graph of y D x 2. This
was discussed in Section 1.6, where we saw that there is a turning point
at .0; 0/, with the graph falling to .0; 0/ as x increases through negative
values to 0 and rising as x increases through positive values, the shape being
concave up throughout. (See Figure 2.2.)
The graph has an alternative description from the theory of conic sections. We wont be using it, but the issues are worth reviewing. The distance

i
i

i
IrvingBook 2013/5/22 15:39 page 30 #46

30

2. Quadratic Polynomials
y

Figure 2.2. Graph of y D x 2

between two points .a; b/ and .c; d / is the square root of .c a/2 C.d b/2.
(This is essentially the Pythagorean Theorem.) The distance from a point P
to a line ` is defined to be the distance between P and the closest point Q
on ` to P , with Q being the intersection of ` and the line perpendicular to
` that passes through P .
Exercise 2.10. Verify that the points in the plane equidistant from .0; 1=4/
and the line y D 1=4 are precisely those satisfying the equation y D x 2 .
(Hint: Set the squared distances equal to each other rather than the distances.)
In the theory of conic sections, a parabola is defined as the locus of
points equidistant from a point, the parabolas focus, and a line not containing the point, the parabolas directrix. Using this language, Exercise 2.10
states that the graph of y D x 2 is the parabola with focus .0; 1=4/ and
directrix y D 1=4.
The unique line that passes through a parabolas focus and meets the
directrix at a right angle is the axis of symmetry for the parabola. The point
where the axis of symmetry intersects the parabola is the parabolas vertex.
For the graph of y D x 2 , Exercise 2.10 shows that the focus is .0; 1=4/ and
the directrix is y D 1=4. It follows that the axis of symmetry is the y-axis
and the vertex is .0; 0/. (See Figure 2.3.)
The essential results about the behavior of the graph dont require calculus or the theory of conic sections, but do require results on the real numbers
such as the intermediate value theorem. It lets us prove that x 2 assumes all
positive real number values as x increases through positive real numbers,

i
i

i
IrvingBook 2013/5/22 15:39 page 31 #47

31

2.4. A Discriminant

y
y = x2
Focus
(0, 1/4)
(x, y)
x

Vertex (0, 0)
Directrix y = 1/4
(x, 1/4)

Axis of symmetry x = 0

Figure 2.3. Focus and Directrix of the Parabola y D x 2

and similarly as x increases through all negative real numbers. Geometrically, this means the graph of y D x 2 goes through all positive heights as
it falls to .0; 0/ on the left side of the y-axis and goes again through all
positive heights as it rises on the right side of the y-axis.
Once we know the shape of the graph of y D x 2 , we can use it for
the graphs of other monic quadratic polynomials. Doing so depends on the
following result, which tells us how to obtain equations for graphs that are
obtained from known ones by horizontal or vertical translations.
Exercise 2.11. Suppose y D f .x/ is a function defined for all real number
values of x.
(i) Given a real number C , verify that the graph of y D f .x C / is
obtained from the graph of y D f .x/ by shifting every point to the
right a distance of C .
(ii) Given a real number D, verify that the graph of y D f .x/ C D is
obtained from the graph of y D f .x/ by shifting every point on it
upward a distance of D.
(iii) Combine (i) and (ii) to obtain the result that the graph of y D f .x
C / C D is obtained from the graph of y D f .x/ by shifting every
point to the right a distance of C and upward a distance of D.
By combining Exercise 2.11 with the technique of completing the square,
we can show that the graph of any monic quadratic polynomial is obtained
from the graph of y D x 2 by vertical and horizontal translation.

i
i

i
IrvingBook 2013/5/22 15:39 page 32 #48

32

2. Quadratic Polynomials

y = x2 + 4

y = x2

y = x2 4
x

Figure 2.4. Graphs of three parabolas

Exercise 2.12. Let b and c be real numbers. By completing the square, we
can write y D x 2 C bx C c as


b2
b
:
y D .x C /2 C c
2
4
Deduce by Exercise 2.11 that the graph of y D x 2 C bx C c is obtained
from the graph of y D x 2 by a shift b=2 to the right and c b 2=4 upward.
Exercise 2.12 tells us that the graph of y D x 2 C bx C c has the same
shape as the graph of y D x 2 , but is shifted up or down, left or right. Hence,
the graph, like that of y D x 2 , has a single turning point. This turning point
may lie above, on, or below the x-axis, as illustrated in Figure 2.4 for x 2 C4,
x 2 , and x 2 4. To decide which, we compute the y-coordinate of the turning
point of y D x 2 C bx C c, which Exercise 2.12 allows us to do.
Exercise 2.13. Let b and c be real numbers. Using Exercise 2.12, conclude
that the graph of y D x 2 C bx C c has a local (and global) minimum at
. b=2; c

b 2 =4/:

The graph of x 2 C bx C c is the parabola with focus at . b=2; c b 2=4 C

1=4/, directrix y D c b 2 =4 1=4, axis of symmetry x D b=2, and
vertex at . b=2; c b 2=4/.
Now that we have found the turning point on the graph of a monic
quadratic polynomial, we can use it to get information on the polynomials
roots.

i
i

i
IrvingBook 2013/5/22 15:39 page 33 #49

2.5. History

33

Exercise 2.14. Using Exercise 2.13, show that the nature of the roots of
x 2 C bx C c is determined by the sign of c b 2 =4:
(i) If c b 2 =4 < 0, then the graphs turning point is below the x-axis.
Conclude that the graph crosses the x-axis twice, so that x 2 C bx C c
has two distinct real roots.

(ii) If c b 2 =4 D 0, then the turning point is on the x-axis. Conclude that

the graph of x 2 C bx C c touches the x-axis once, so that x 2 C bx C c
has one real root. Since x 2 C bx C c must factor as the product of two
degree-one polynomials, deduce that the real root has multiplicity 2.
(iii) If c b 2=4 > 0, then the turning point is above the x-axis. Conclude
that the graph of x 2 C bx C c does not touch or cross the x-axis, so
that x 2 C bx C c has no real roots.
(The three cases are illustrated in Figure 2.4.)

Exercise 2.14 shows that the nature of the roots of x 2 C bx C c is

determined by the quantity c b 2 =4. Any non-zero constant multiple of
c b 2=4 will determine their nature as well. Lets multiply by 4, obtaining
the quantity b 2 4c. We will write for this and call it the discriminant of
x 2 C bx C c. We see that it is the product of 4 and the y-coordinate of the
turning point on the graph of y D x 2 C bx C c. Exercise 2.14 yields the
result we are after:
Theorem 2.5. Let f .x/ D x 2 C bx C c for real numbers b and c and let
be the associated discriminant b 2 4c.
(i) If > 0, then f .x/ has two distinct real roots.
(ii) If D 0, then f .x/ has a repeated real root.
(iii) If < 0, then f .x/ has no real root.

2.5 History
In this section, we will look at the history of quadratic equations from ancient Babylonian civilization four millennia ago to Italy in 1500, touching
occasionally on developments in algebra beyond the quadratic formula.
Every ancient civilization developed a body of mathematical knowledge, in part to serve the practical needs of measurement, construction,
and commerce. The work of the ancient Greeks may be the most familiar,
thanks to Euclid, but important contributions were made as well by Babylonian, Egyptian, Indian, and Chinese civilizations. Mathematical problems

i
i

i
IrvingBook 2013/5/22 15:39 page 34 #50

34

2. Quadratic Polynomials

and solutions were often described through words, and geometry was used
to express concepts that we might now address algebraically through notational systems not then available. General methods would be laid out implicitly through the working of a series of examples. All these civilizations
addressed problems, in different guises, that we can now interpret as fitting
under the heading of solving quadratic equations.
Lets begin with a look at some Babylonian work on problems reducible
to quadratic equations. The first Babylonian dynasty, in the years ranging
from around 1900 B.C.E. to 1600 B.C.E., made many contributions to human culture. Perhaps most notable is the code of Hammurabi, consisting of
282 laws that were written on clay tablets. Some of the earliest preserved
mathematical texts date to this time. Babylonians used cuneiform, a wedgelike script that evolved from pictographs, for their writing. They also used
base 60 in writing numbers, a choice that persists today in our methods for
measuring time (minutes, seconds) and angle (degrees, minutes, seconds).
Surviving on tablets are examples of the mathematical problems that Babylonians posed and solved. Many reduce to solving quadratic equations.
Our understanding of the mathematics on the cuneiform tablets owes
much to the pioneering work of Otto Neugebauer, an Austrian scholar who
studied mathematics at Munich and Gottingen in the 1920s. While at
Gottingen, Neugebauer shifted his interest from mathematics to its history and did his doctoral research on Egyptian mathematics. He stayed
on at Gottingen and began to study the Babylonian tablets that can be
found at many of the major museums in Europe and the United States.
His three-volume work Mathematische Keilschrift-Texte [44], published in
1935, translated and interpreted tablets found in museums from London and
Paris to Berlin and Istanbul.
Neugebauer came to the United States in 1939 and continued his work,
publishing Mathematical Cuneiform Texts [46] jointly with Abraham Sachs
in 1945. He provided an overview of his findings in The Exact Sciences
in Antiquity [45], published in 1952 based on a 1949 lecture series he presented at Cornell. These books are well worth a look, both for their content
and for the tablet photographs.
In surveying the Babylonian treatment of algebra, Neugebauer highlighted its abstract nature, divorced from both geometric considerations and
practical meaning:
[G]eometrical concepts play a very secondary part in Babylonian algebra, however extensively a geometrical terminology may be used. It
suffices to quote the existence of examples in which areas and lengths
are added, or areas multiplied, thus excluding any geometrical inter-

i
i

i
IrvingBook 2013/5/22 15:39 page 35 #51

2.5. History

35

pretation . . . . Indeed, still more drastic examples can be quoted for the
disregard of reality. We have many examples concerning wages to be
paid for labor according to a given quota per man and day. Again, problems are set up involving sums, differences, products of these numbers
and one does not hesitate to combine in this way the number of men
and the number of days. It is a lucky accident that if the unknown number of workmen, found by solving a quadratic equation, is an integer.
Obviously the algebraic relation is the only point of interest, exactly as
it is irrelevant for our algebra what the letters may signify.
Neugebauer further explained [45, p. 42] that some of the Babylonian
texts contain collections of problems only, sometimes more than 200 on
a single tablet of the size of a small printed page. These collections of
problems are usually carefully arranged, beginning with very simple cases
. . . and expanding step by step to more complicated relations. Regarding one such series [45, p. 42], one finds that they all have the same pair
x D 30; y D 20 as solutions. This indicates that it was of no concern to the
teacher that the result must have been known to the pupil. What he obviously had to learn was the method . . . . From actually computed examples
it becomes obvious that it was the general procedure, not the numerical result, which was considered important.
A tablet of particular interest is one catalogued in the British Museum
as BM 13901. It originally contained twenty-four problems. Some are damaged, but as A. E. Berriman notes in his article The Babylonian quadratic
equation [10, p. 185], those that remain, however, are sufficient to reveal
a carefully graduated course of instruction. Neugebauer studied the tablet
in the third volume of Mathematische Keilschrift-Texte [44]. A discussion
can also be found in the 2009 English translation of Jacques Sesianos 1999
French work, An Introduction to the History of Algebra: Solving Equations
from Mesopotamian Times to the Renaissance [59, pp. 1016].
Lets look at the first problem of BM 13901. We will use the Babylonians sexagesimal number system, as Neugebauer did in his translations.
Commas separate the numbers in the 1 and 60 columns and a semi-colon
separates the integer part from the decimal part, with the first position after
the semi-colon being the number of sixtieths. For example, 1; 2 is sexagesimal notation for the decimal number 62 and 0I 20 is the fraction 20=60, or
1=3.
The problem is stated simply [59, p. 10]: I added the area and the side
of my square: 0I 45. In standard fractional notation, this number is 45=60,
or 3=4. The unknown is the length of the side of the square. Thus, we are

i
i

i
IrvingBook 2013/5/22 15:39 page 36 #52

36

2. Quadratic Polynomials

being asked to solve the quadratic equation

x2 C x D

3
:
4

What follows on the tablet is a recipe: You put 1, the unit. You divide in
two 1: 0I 30. You multiply by 0I 30: 0I 15. You add 0I 15 to 0I 45: 1. It is
the square of 1. You subtract 0I 30, which you multiplied, from 1: 0I 30, the
side of the square.
To make sense of this, it is perhaps better to think of the equation in the
form x 2 C bx C c D 0, with b D 1 and c D 3=4. We are asked first to
divide b by 2, yielding b=2. We are then told to square it, obtaining b 2 =4.
2
We add
p this to c, obtaining b =4 c. Finally we take the square root of
this, b 2 =4 c, and subtract the number we earlier multiplied, which was
b=2. The final answer, then, is
r
b
b2
C
c;
2
4
or 1=2. The recipe is the quadratic formula!
Lets take the same approach to the second problem of BM 13901 [59,
p. 10]: I subtracted from the area the side of my square: 14; 30. The sexagesimal number 14; 30, in decimal form, is 14  60 C 30, or 870. Thus, we
are asked to solve the quadratic equation
x2

x D 870:

The solution is laid out as follows: You put 1, the unit. You divide in
two 1: 0I 30. You multiply by 0I 30: 0I 15. You add to 14; 30: 14; 30I 15. It
is the square of 29I 30. You add 0I 30, which you multiplied, to 29I 30: 30,
the side of the square.
Once again thinking of the equation as x 2 C bx C c D 0, this time with
b D 1 and c D 870, we can describe the solution as follows: We first
form the quantity b=2. We then square it and add the result to c to get
b 2 =4 c. The next step is to determine the square root, to which we add
the number we earlier used for multiplying, which was b=2. The result
30is the desired answer, and once again it is what we obtain by using the
quadratic formula.
It turns out that the two problems are not typical. Few examples of single
quadratic equations have been found. More common are problems involving
two variables and two equations, one equation having the form xy D 1, the
other being a linear equation in x and y. They can be reduced to solving a
single quadratic equation.

i
i

i
IrvingBook 2013/5/22 15:39 page 37 #53

37

2.5. History

Consider for instance Problem-Text T in Neugebauer and Sachss Mathematical Cuneiform Texts. They open their commentary [46, p. 116] with
the observation that the text can best be compared to an extensive collection of problems from a chapter of a textbook. It is obvious that a collection
of this sort was used in teaching mathematical methods. The student is
asked to reduce each problem to what Neugebauer and Sachs refer to [46,
p. 117] as the normal form of quadratic equations which gives the product of the unknown quantities and their sum or difference. . . . The whole
system of a main example with all its variants serves the purpose of giving
the general rule of solution; this corresponds in a certain sense to an algebraic formula in which the letters can be replaced by special numbers in
each individual case.
The normal form is the problem we studied in Section 2.1. A representative example is the first problem on tablet AO 8862 at the Louvre in Paris,
treated by Neugebauer in the first volume of Mathematische KeilschriftTexte [44, pp. 108123] and by Sesiano as well [59, p. 12]. Van der Waerden
opens the chapter on Babylonian mathematics in Science Awakening with
this example [64, pp. 6365]: Length, width. I have multiplied length and
width, thus obtaining the area. Then I added to the area, the excess of the
length over the width: 3; 3 (i.e., 183 was the result). Moreover, I have added
length and width: 27. Required length, width and area.
We are asked to solve the two equations
xy C .x

y/ D 183

and
x C y D 27:
The tablet then lays out a sequence of numerical calculations that lead to
the answer. They can be interpreted as instructing the student to add the
equations, yielding x.y C 2/ D 210, and to add 2 to both sides of the
second equation, yielding x C .y C 2/ D 29. This transforms the original
equations to normal form, x and y C 2 being two numbers whose product
is 210 and whose sum is 29.
It is important to note that scholarship on Babylonian mathematics in
recent decades has yielded a perspective sharply different from that engendered by Neugebauers work. Jens Hyrup, in his 2002 study Lengths,
Widths, Surfaces: A Portrait of Old Babylonian Algebra and Its Kin [35],
observes that
The Babylonian algebra which most historians of mathematics found
in Neugebauers works looked astonishingly modern and similar to

i
i

i
IrvingBook 2013/5/22 15:39 page 38 #54

38

2. Quadratic Polynomials

ours. It is the purpose of the present book to replace this standard interpretation by a less modernizing reading. . . . The mathematical texts are
school texts. They contain no theorems and no theoretical investigations. . . . [Their authors] were teachers of computation, at times teachers of pure, unapplicable computation . . . but they remained teachers,
teachers of scribe school students who were later to end up applying
mathematics to engineering, managerial, accounting, or notarial tasks.
Hyrups own analysis of the first problem on AO 8862 #1 [35, pp. 169
170] emphasizes its concrete nature: The text starts by stating that a rectangular surface or field is built, that is, marked out; after pacing off its dimensions, the speaker appends the excess of the length over the width to
it; the outcome is 3; 3. Even this is done quite concretely in the terrain. Then
he turns back and reports the accumulation of the length and the width to
be 27.
In Mathematics in Ancient Iraq: A Social History [56], Eleanor Robson
offers an enlightening side-by-side comparison of van der Waerdens and
Hyrups analyses [56, pp. 276278]. In van der Waerdens translation of
1954 the problem is entirely numerical . . . In his reading the Babylonians
are just like modern mathematicians: they use symbols and equations,
which means that the problem can safely be expressed as modern algebra. Hyrup, by contrast, opens his comments on the same problem with an
interpretative diagram that does not appear on the cuneiform tablet and continues [as quoted above]. All of van der Waerdens apparently arithmetical
numbers turn out to have dimension: they are particular lengths and areas
that are manipulated physically.
Another common form taken by problems on the tablets is to find two
numbers from known values of the sum of their squares and of their sum
(or difference). An example is problem 9 of BM 13901 [59, p. 15]: I added
the area of my two squares: 1300. The side of one exceeds the side of the
other by 10. Here we are asked to solve the equations
x 2 C y 2 D 1300
and
x

y D 10:

This is easily converted to a quadratic equation in one variable. In reinterpreting the problem this way, we are placing a modern overlay on the data
of the tablet. Hyrup offers his own analysis [35, pp. 6670], attempting to
interpret the tablets prescription as a series of instructions for tearing out,
inscribing, and appending squares of given dimensions.

i
i

i
IrvingBook 2013/5/22 15:39 page 39 #55

2.5. History

39

We have given more attention to Babylonian mathematics and the implicit treatment within it of quadratic equations than might seem necessary.
However, it is difficult to avoid the temptation. As Eleanor Robson concludes [56, p. 290]:
Compared to the difficulties of grappling with fragmentary and meagre
nth-generation sources from other ancient cultures the cuneiform evidence is concrete, immediate, and richly contextualised. . . . This opens
a unique window onto the material, social, and intellectual world of
the mathematics of ancient Iraq that historians of other ancient cultures can only dream of.
Around 600 B.C.E., a new mathematical current arose within the Greek
civilization of the Mediterranean, starting with the work of Thales and
Pythagoras. The Greek tradition would have enormous influence on the development of mathematics, and much has been written about it. Perhaps
of greatest importance was the introduction of the axiomatic method and
deductive proof. For a concise introduction, William Berlinghoff and Fernando Gouvea have an informative survey [9, pp. 1424]. (See S. Cuomos
Ancient Mathematics [15] for a more detailed study.) The most famous of
the Greek mathematicians is Euclid, who lived around 300 B.C.E. in Alexandria. Little is known about him, or about the origins of his greatest work, the
Elements. For instance, no definitive answer can be given to the question of
how much of the Elements is due to him and how much is a compendium
of earlier work.
The Elements provided the model for the axiomatic method and laid the
basis for geometry for centuries. Most of its thirteen books are devoted to
geometry, with plane geometry treated first and regular polyhedra in threespace covered in the concluding Book 13. Books 7 to 9 contain some of the
most famous results of elementary number theory, such as Euclids proof
that there are infinitely many prime numbers. Book 2 provides solutions to
geometric questions about area that amount to solving simultaneous equations x C y D a and xy D b for constants a and b [65, pp. 7780]. As we
know, this reduces to solving a quadratic equation.
Jumping ahead five centuries to the late stages of the Greek tradition, we
come to Diophantus, another Alexandrian. Diophantus wrote Arithmetica,
a collection of about 200 problems and their solutions. Arithmetica consisted of thirteen chapters, of which six were preserved in Greek and four in
an Arabic translation from the ninth century, but the Arabic chapters were
essentially lost until their rediscovery in 1968.
In contrast to Euclids geometric algebra, Diophantus works in a manner
we would recognize as more strictly algebraic, with symbols for the lower

i
i

i
IrvingBook 2013/5/22 15:39 page 40 #56

40

2. Quadratic Polynomials

positive and negative powers of an unknown, rules of multiplication, and

rules of substitution. Many of his problems reduced to equations in two
unknowns. Among them were problems studied by the Babylonians, such
as finding two numbers whose sum and product are given. Solutions were
to be found in rational numbers.
The Greek chapters of Arithmetica had little influence until a revival of
interest in them among Italian mathematicians late in the sixteenth century
and even more so by the French mathematician Pierre de Fermat in the
seventeenth century. In Book II, Diophantus poses the problem to divide a
square into two squares, that is, to find solutions to x 2 C y 2 D z 2 . Although
we have not discussed this problem, it has a history as long as that of the
quadratic equation. All ancient civilizations studied it, including two we
have not discussed, those of Egypt and China.
Fermat owned a copy of a 1621 Latin translation of Arithmetica made
by the French mathematician and scholar Claude-Gaspard Bachet (1581
1638). (Bachets best known work is Probl`emes plaisants et delectables qui
se font par les nombres [6], his 1612 collection of arithmetic puzzles and
tricks that was the basis for many subsequent books on recreational mathematics.) Some time around 1637, Fermat wrote the famous marginal note
that for integers n greater than 2, the equation x n C y n D z n has no rational solution, adding the tantalizing remark that he had a truly marvelous
proof that his margin was too narrow to contain.
Whether Fermats statement was true would remain one of the great
open questions in mathematics until it was settled in the affirmative in 1995
by Andrew Wiles and Richard Taylor. In the meantime, Diophantus became the eponym for an entire branch of mathematics, Diophantine analysis, which brings number-theoretic and geometric methods to bear on the
question of whether algebraic equations in several unknowns have integer
or rational solutions. (See [7], [8, pp. 3557], and [59, pp. 3151] for more
information on Diophantuss work.)
In tracing the development of algebra, let us move east to India, where
the mathematician and astronomer Aryabhata lived from 476 C.E. to about
550 C.E. In his work Aryabhatiya, he solves some problems that reduce to
quadratic equations. Satya Prakash observes in a book on the later Indian
astronomer-mathematician Brahmagupta [53, p. 215] that Aryabhata has
given the solutions of a few quadratic equations, but he nowhere gives the
procedure of solving these equations.
Brahmagupta, who lived from around 598 to 665, wrote several major works, including Brahmasphutasiddhanta. Some of his work on arithmetic and algebra was translated into English in 1817 by H.T. Colebrooke

i
i

i
IrvingBook 2013/5/22 15:39 page 41 #57

2.5. History

41

[14]. Here is one problem Brahmagupta poses [14, p. 346]: When does
the residue of revolutions of the sun, less one . . . equal to the square root of
two less than the residue of revolutions, less one, multiplied by ten and augmented by two? Translating into modern algebraic notation, Brahmagupta
is looking for a solution of the equation x 2 86x D 249. Following Colebrooke in his translation, let us call 249 the absolute number and x the
middle term. With this terminology, Brahmaguptas proposed solution
amounts to a statement of the quadratic formula:
Now, from the absolute number, multiplied by four times the coefficient of the square, and added to the square of the coefficient of the
middle term, the square root being extracted, and lessened by the coefficient of the middle term, the remainder is divided by twice the coefficient of the square, yields the value of the middle term.
Satya Prakash comments [53, p. 215] that Brahmagupta undoubtedly is
not the discoverer of these rules, but perhaps for the first time in the history
of algebra we find the process of solving a quadratic equation so clearly
indicated.
Brahmaguptas greatest contribution to algebra is his study of integer
solutions to the equation x 2 ny 2 D 1, where n is a positive integer that
is not a square. Such an equation is now known as Pells equation and arises
naturally in a variety of contexts. The twelfth century Indian mathematician
Bhaskara obtained more definitive results. More information on this work
can be found in V. S. Varadarajans Algebra in Ancient and Modern Times
[67], which has much to say about other cultures as well. See too B. L. van
der Waerdens Geometry and Algebra in Ancient Civilizations [65].
The subsequent centuries were ones of exciting intellectual developments in the Islamic world. Most notable for us is the work of Muhammad
ben Musa al-Khwarizmi, a Persian who lived roughly between 780 and 850
near Baghdad. Al-Khwarizmi contributed to many fields, including mathematics, astronomy, and geography. Perhaps his greatest work is his treatise
on algebra, The Compendious Book on Calculation by Completion and Balancing, written in Arabic around 825. In it, he systematically studied linear
and quadratic equations using the techniques of al-jabr and al-muqabala.
The process of al-jabr is that of adding equal terms to both sides of an equation in order to eliminate negative terms. In contrast, al-muqabala is the reduction of positive terms by subtracting equal amounts from both sides of
an equation.
Al-Khwarizmis work would have a significant influence on European
mathematics several centuries later, thanks to its translation into Latin in the

i
i

i
IrvingBook 2013/5/22 15:39 page 42 #58

42

2. Quadratic Polynomials

twelfth century by Robert of Chester and Gerard of Cremona and in the thirteenth century by William of Luna. Robert of Chesters version, Liber algebrae et almucabola, was translated into English in 1915 by Louis Charles
Karpinski, who also provided notes and an introduction [36]. Karpinskis
edition is well worth a look, as is the more recent Al-Khwarizmi: The Beginnings of Algebra [54] by Roshdi Rashed, which includes a translation
(with the original Arabic text on facing pages) and commentary. Rashed
also writes about the traditions of calculation in the eighth century and alKhwarizmis knowledge of Greek and Indian mathematical literature.
At the outset of his book, al-Khwarizmi introduces three types of quantities: roots, squares, and numbers, the root being what we would label x,
the square being x 2 , and number being number. He then gets right to business, classifying linear and quadratic equations into six forms and illustrating how to solve each with examples. We would regard all six types as one,
but since al-Khwarizmi does not use negative numbers or 0, he is obliged to
consider separate cases, with a, b, and c always positive:
(1) Squares equal roots, or what we would describe as ax 2 D bx.
(2) Squares equal numbers, or ax 2 D c.
(3) Roots equal numbers, or bx D c.

(4) Squares and roots equal numbers, or ax 2 C bx D c.

(5) Squares and numbers equal roots, or ax 2 C c D bx.

(6) Roots and numbers equal squares, or bx C c D ax 2 .

(In addition to [36] and [54], see also the discussion of al-Khwarizmis work
in The Beginnings and Evolution of Algebra by I.G. Bashmakova and G.S.
Smirnova, from which the above summary is drawn [8, p. 50].)
What is novel about the opening of al-Khwarizmis book is that in it he
lays out the mathematical issues in purely algebraic terms before turning to
the examplesdrawn from commerce and inheritancesthat fill the later
pages of the book. The focus is on equations in the abstract, classified by
degree. Rashed explains in the introductory essay to his translation [54,
p. 24] that what al-Khwarizmi
does cannot be reduced to anything to be found in other traditions,
such as those of the Babylonians, of Diophantus, of Heron of Alexandria, of Aryabhata or of Brahmagupta. It is not in the course of solving
problems that al-Khwarizm finds these equations. The classification
in fact precedes the problems. It is introduced deliberately as the necessary first step in the construction of a theory of equations of the first

i
i

i
IrvingBook 2013/5/22 15:39 page 43 #59

2.5. History

43

and second degree; and this theory will become the nucleus of a mathematical discipline.
The first of al-Khwarizmis six cases that interests us is .4/, and the
example he uses to illustrate it is none other than the quadratic equation
x 2 C 10x D 39 that we studied in Sections 2.2. This equation is not distinguished in any mathematical sense, but since it is the first one for which
al-Khwarizmi employs the technique of completing the square, it has resonated through history. Or, as Karpinski suggests [36, p. 19], the equation
runs like a thread of gold through the algebras for several centuries, appearing in the algebras of the three writers mentioned, Abu Kamil, al-Karkhi and
Omar al-Khayyami, and frequently in the works of Christian writers, centuries later. Here is al-Khwarizmis own description of the example [54,
p. 100]:
Squares plus roots are equal to a number, as when you say: a square
plus ten roots are equal to thirty-nine dirhams [a unit of measure];
namely, if you add to any square [a quantity] equal to ten of its roots,
the total will be thirty-nine.
Procedure: you halve the number of the roots which, in this problem,
yields five; you multiply it by itself; the result is twenty-five; you add
it to thirty-nine; the result is sixty-four; you take the root, that is eight,
from which you subtract half the number of the roots, which is five.
The remainder is three, that is the root of the square you want, and the
square is nine.
Upon completing his treatment of the remaining types of quadratic equation, al-Khwarizmi returns to the equation x 2 C 10x D 39 and illustrates
two ways of solving it geometrically. The second is the one we used in
Section 2.2.
Rashed concludes his introductory essay by turning to the question of
Indian influence on al-Khwarizmi. Two centuries ago, in translating the
work of the Indians, Colebrooke suggested [14, pp. xxxxi] that al-Khwarizmi was a mere borrower. This assessment is misguided, but the extent to
which Islamic mathematicians and astronomers were familiar with and influenced by Indian literature is an interesting question. Rashed emphasizes
what is original to al-Khwarizmi [54, p. 77]:
Comparing the two texts, that of Brahmagupta and that of al-Khwarizmi, reveals irreducible differences. . . . Brahmagupta arrived at the
quadratic equation in one unknown in the course of solving a problem
in astronomy. In other words, he did not give himself the equation as

i
i

i
IrvingBook 2013/5/22 15:39 page 44 #60

44

2. Quadratic Polynomials

such with a view to solving it. This link between problem and equation,
which is found in other mathematics, this as it were empirical grounding for the equation, has vanished in al-Khwarizmis programme. From
the start, al-Khwarizmi proceeded by defining basic terms, which were
then combined to give him the ideal canonical equations with which his
theory is concerned. This new approach breaks the close link between
problems and equations. As for problems, al-Khwarizmi turns to them
later, as exercises in algebra, that is as providing an area in which he
can apply the theory of equations that he has already constructed. . . .
. . . al-Khwarizmi wanted to construct a form of calculation for unknowns that was independent of what they represented, that is a new
mathematical discipline, that by its very nature was subject to the rules
of proof. There is no trace of any such programme in the work of his
predecessors.
Al-Khwarizmi was one of several Islamic scholars who made important
contributions to algebra. Others include Thabit ibn Qurra, who spent much
of his career in Baghdad, studying medicine, philosophy, mathematics, and
astronomy as well as translating Euclid and other Greeks into Arabic. He
died in 901. Also worth mentioning is the famed Persian poet, philosopher,
and mathematician Omar Khayyam, who studied in Samarkand and worked
in Bukhara (both in modern-day Uzbekistan), dying in 1131. He wrote a
book on algebra, studying quadratic equations from a geometric viewpoint
as in Euclid. The work of Thabit and Omar Khayyam is discussed in detail
in van der Waerdens A History of Algebra From Al-Khwarizmi to Emmy
Noether [66, pp. 1531].
While mathematics was flourishing in Arabic, Persian, and Indian lands,
the medieval era in Europe was mathematically more quiescent. The revival
of significant European mathematical activity occurred first in Italy, perhaps because its cities were major trade centers with connections to Arabic
ports along the Mediterranean. The great Italian mathematician Leonardo
da Pisabetter known to us as Fibonacciprovides testimony to this effect
in the prologue to his influential 1202 work Liber Abaci [28, pp. 1516], or
the Book of Calculation:
As my father was a public official away from our homeland in the
Bugia [the Algerian port city of Bejaia] customshouse established for
the Pisan merchants who frequently gathered there, he had me in my
youth brought to him, looking to find for me a useful and comfortable
future; there he wanted me to be in the study of mathematics and to
be taught for some days. There from a marvelous instruction in the

i
i

i
IrvingBook 2013/5/22 15:39 page 45 #61

2.5. History

45

art of the nine Indian figures, the introduction and knowledge of the
art pleased me so much above all else, and I learnt from them, whoever was learned in it, from nearby Egypt, Syria, Greece, Sicily and
Provence, and their various methods, to which locations of business I
travelled considerably afterwards for much study, and I learnt from the
assembled disputations.
Fibonacci is best known for the sequence of numbers that bears his
name, the Fibonacci numbers, which arise in his solution to the following
problem [28, pp. 404405]:
A certain man had one pair of rabbits together in a certain enclosed
place, and one wishes to know how many are created from the pair in
one year when it is the nature of them in a single month to bear another
pair, and in the second month those born to bear also.
This is sufficiently famous that we will take a moment to review
Fibonaccis conclusion. He works out the numbers month by month, then
explains that we added the first number to the second, namely the 1 to the
2, and the second to the third, and the third to the fourth, and the fourth to
the fifth, and thus one after another until we added the tenth to the eleventh,
namely the 144 to the 233, and we had the above written sum of rabbits,
namely 377, and thus you can in order find it for an unending number of
months.
In Chapter 15 of Liber Abaci, Fibonacci examines the algebra of quadratic equations, following al-Khwarizmis treatment closely. For example
[28, p. 554], Fibonacci introduces roots, squares, and simple numbers and
explains that in the solutions of problems there are six modes of which
three are simple, and three are composite, just as in al-Khwarizmis classification of quadratic equations. Of particular importance is Fibonaccis use
of Hindu-Arabic numerals, which led to their adoption in Europe.
For a brief survey of the algebraic work of Italian mathematicians in
the subsequent three centuries, concluding with Luca Pacioli (14451514),
one can turn to [66, pp. 4247]. Pacioli provided an overview of the mathematical knowledge of the time in his 1494 book Summa de arithmetica,
geometrica, proportioni e proportionalit`a [50], published in Venice. The
most influential material in this book would turn out to be his account of
double-entry bookkeeping, thanks to which he has been called the Father of
Accounting. We will return to Pacioli for a moment, in Section 3.5, to begin
our survey of sixteenth-century Italian contributions to the solving of cubic
equations.

i
i

i
IrvingBook 2013/5/22 15:39 page 46 #62

46

2. Quadratic Polynomials

Let us close this historical discussion with an excerpt from Karpinskis

introduction [36, pp. 1112] to his translation of al-Khwarizmi, in which
he describes the study of quadratic equations as an inevitable stage in the
history of human thought.
Algebra is not, as often assumed, an artificial effort of human ingenuity, but rather the natural expression of mans interest in the numerical side of the universe of thought. Tables of square and cubic
numbers in Babylon; geometric progressions, involving the idea of
powers, together with linear and quadratic equations in Egypt; the socalled Pythagorean theorem in India, and possibly in China, before the
time of Pythagoras; and the geometrical solution of quadratic equations even before Euclid in Greece, are not isolated facts of the history
of mathematics. While they do indeed mark stages in the development
of pure mathematics, this is only a small part of their significance.
More vital is the implication that the algebraical side of mathematics
has an intrinsic interest for the human mind not conditioned upon time
or place, but dependent simply upon the development of the reasoning
faculty. We may say that the study of powers of numbers, and the related study of quadratic equations, were an evolution out of a natural
interest in numbers; the facts which we have presented are traces of the
process of this evolution.

i
i

i
IrvingBook 2013/5/22 15:39 page 47 #63

3
Cubic Polynomials
In this chapter, we will take our first look at cubic equations and the famous
formula for their solution known as Cardanos formula. Girolamo Cardano,
for whom the formula is named, was a sixteenth-century Italian scholar. The
story of the formulas discovery is complex, as we will see in Section 3.5,
and credit must be shared with Scipione del Ferro and Niccol`o Fontana.
Our results in this chapter will be imprecise, because we are lacking
what turns out to be an essential tool: complex numbers. However, it is this
first look that will reveal the need for complex numbers. After developing
the basic facts about them in Chapter 4, we will return to cubic equations in
Chapter 5 and treat them with appropriate care.

3.1 Reduced Cubics

Quadratic polynomials need not have roots. In contrast, a cubic polynomial
must have a root. This is a special case of Theorem 1.14, which stated that
any odd degree polynomial has a root. (Alternatively, once we have Cardanos formula and complex numbers, the existence of a root will follow
automatically from algebra alone.) Let us record this as a theorem.
Theorem 3.1. Let f .x/ be a cubic polynomial. Then there is a real number
a such that f .a/ D 0.
Theorem 1.4 states that, given a polynomial f .x/ and a real number a,
if f .a/ D 0, then x a divides f .x/. Combining this with Theorem 3.1,
we find that any cubic polynomial can be factored as a product of linear and
47

i
i

i
IrvingBook 2013/5/22 15:39 page 48 #64

48

3. Cubic Polynomials

quadratic polynomials.
Exercise 3.1.

Suppose f .x/ is a monic cubic polynomial.

(i) Use Theorems 1.4 and 3.1 to deduce that f .x/ factors as
.x

a/.x 2 C mx C n/

for some real numbers a, m, and n.

(ii) Deduce that any real root of f .x/ besides a is a root of x 2 C mx C n.
(Hint: Suppose r is a root and r a. Substitute r in f .x/, using the
factorization in (i). What do you get?)
(iii) Apply Theorem 2.2 to x 2 C mx C n to prove Theorem 3.2.
Theorem 3.2. Let f .x/ be a monic, cubic polynomial. Exactly one of the
following occurs.
(i) f .x/ has one real root a, of multiplicity 3, and factors as
.x

a/3 :

(ii) f .x/ has two distinct real roots a1 and a2 , of multiplicities 1 and 2,
and factors as
.x a1 /.x a2 /2 :
(iii) f .x/ has three distinct simple real roots a1 , a2 , and a3 , and factors as
.x

a1 /.x

a2 /.x

a3 /:

(iv) f .x/ has one simple real root a and factors as

.x

a/s.x/;

where s.x/ is a quadratic polynomial that does not factor as a product

of linear polynomials.
A cubic polynomial of the form
y 3 C py C q
is called a reduced cubic. In the next exercise, we will see that a change of
variable allows us to pass from an arbitrary cubic polynomial to a reduced
one. This is analogous to the change of variable used in Exercise 2.9 to pass
from a quadratic polynomial to a new one without a degree-one term, and
will simplify the task of solving cubic equations.

i
i

i
IrvingBook 2013/5/22 15:39 page 49 #65

49

3.1. Reduced Cubics

Exercise 3.2.

Let
f .x/ D x 3 C bx 2 C cx C d;

for real numbers b, c, and d . Given another real number a, lets see what
happens under the change of variable x D y C a, or y D x a.
(i) Substitute yCa for x and obtain a polynomial g.y/ in the new variable
y. Write it as
y 3 C By 2 C Cy C D
and obtain formulas for the coefficients B, C , and D in terms of a and
the old coefficients b, c, and d .
(ii) Observe that there is a choice of a for which B D 0. Thus for this a,
changing variables provides a new polynomial g.y/ that is reduced.
(iii) Verify that g.y/ is the reduced polynomial described in Theorem 3.3.
(iv) What is the relation between a root of f .x/ and a root of g.y/? In
particular, given a root s of g.y/, describe a root r of f .x/.
(v) Conclude that solving g.y/ D 0 allows us to solve f .x/ D 0.
Theorem 3.3. Let b, c, and d be real numbers. Given a cubic polynomial
x 3 C bx 2 C cx C d , the change of variable x D y b=3 yields the reduced
cubic polynomial
 2

 3

b
2b
bc
3
y C
Cc yC
Cd :
3
27
3
We have shown that we can pass from the problem of solving an arbitrary cubic equation to the equivalent problem of solving a cubic equation
of the form
y 3 C py C q D 0;
and that being able to solve equations of this simpler type allows us to solve
arbitrary cubic equations.
Exercise 3.3.

Suppose we wish to solve the cubic equation

x3

3x 2

4x C 12 D 0:

(i) What reduced cubic equation would we solve in order to find the solutions?

i
i

i
IrvingBook 2013/5/22 15:39 page 50 #66

50

3. Cubic Polynomials

(ii) How are the solutions of the reduced cubic equation related to the solutions of the original cubic equation?
In Section 3.2, we will solve the reduced cubic equation in Exercise 3.3,
allowing us to find the solutions of the original cubic equation.
There are two families of reduced cubic equations that take on an even
simpler form, those for which p D 0 and those for which q D 0. Lets
discuss these.
Exercise 3.4.

Consider the cubic equation y 3 C py D 0.

(i) Observe that y D 0 is a solution and that the other solutions are solutions to y 2 C p D 0.
(ii) Deduce that if p > 0 then y D 0 is the only solution and 0 is a simple
root of y 3 C py; if p D 0 then y D 0 is the only solution and is a
repeated root of y 3 C py; and if p < 0 then there are three distinct
p
p.
solutions: y D 0 and y D
Exercise 3.5. Consider the cubic equation y 3 C q D 0. If q D 0, we are
studying the equation y 3 D 0. Obviously the only solution is 0, which has
multiplicity 3 as a root of y 3 . Assume for the remainder of this exercise that
q 0.
(i) By Theorem 3.1, the equation must have a solution, and any solution
is a cube root of q.
(ii) Suppose r is such a cube root: r 3 D q. Thus, we can rewrite the
polynomial y 3 C q as y 3 r 3 . Show that
y3

r 3 D .y

r /.y 2 C ry C r 2 /:

(iii) Show that y 2 C ry C r 2 has no real roots. (What is its discriminant?)

(iv) Conclude that r is the lone root of y 3 C q and that r has multiplicity
1.

3.2

Cardanos Formula

We have seen in Section 3.1 that every cubic polynomial has at least one
root and that by change of variable we can reduce the problem of finding
a root of a given cubic polynomial x 3 C bx 2 C cx C d to the problem of

i
i

i
IrvingBook 2013/5/22 15:39 page 51 #67

51

3.2. Cardanos Formula

finding a root of a cubic polynomial of the form y 3 C py C q. What would

be a suitable solution to the reduced problem?
At the end of Section 2.1, we observed that the quadratic formula does
not so much solve quadratic equations as provide a reduction procedure, one
that uses algebra to reduce the problem of solving a quadratic equation to
the problem of calculating square roots of real numbers. Algebra offers no
guidance on finding the square roots. Similarly, in solving cubic equations,
we should be content if we reduce the problem to one of calculating square
and cube roots. The calculation of cube roots of real numbers, like the calculation of square roots, is not a problem we should expect algebra to solve.
In factand this is an essential part of our storywe will discover that we
must be content with something less: a reduction to square root calculations
of real numbers and cube root calculations of entities more general than real
numbers.
No point delaying. Heres the formulaCardanos formulafor a solution of the reduced cubic equation y 3 C py C q D 0:
s
3

yD

q
C
2

p3
q2
C
C
27
4

s
3

q
2

p3
q2
C
27
4

We will explore its history in Section 3.5. For now, let us just note that
Cardano did not write down such a formula explicitly. Rather, he illustrated
how to solve reduced cubic equations through examples.
Exercise 3.6. Substitute the expression for y given by Cardanos formula
into the polynomial y 3 CpyCq and verify, after expanding and simplifying,
that it equals 0.
Cardanos formula expresses the solution to y 3 C py C q D 0 in terms
of the coefficients p and q, addition, multiplication, division by some constants, and the taking of square and cube roots. Thus, just as the quadratic
formula reduces the solution of quadratic equations to square root calculations, Cardanos formula reduces the solution of cubic equations to square
root and cube root calculations. Once we use Cardanos formula to find a
real root r of y 3 C py C q, we can divide y 3 C py C q by y r to obtain
a quadratic polynomial and use the quadratic formula to find any additional
real roots.
The formula has one worrisome feature, the appearance of the square
root of p 3 =27 C q 2 =4. In the quadratic formula, whenever the quadratic
polynomial being studied has a real root, the quantity whose square root

i
i

i
IrvingBook 2013/5/22 15:39 page 52 #68

52

3. Cubic Polynomials

must be calculated is non-negative. Cubic polynomials, in contrast to quadratics, always have real roots. If the value of p 3 =27 C q 2=4 for a reduced cubic
polynomial is negative, we may be in for trouble in attempting to use Cardanos formula. Well soon see what issues arise.
Before using Cardanos formula in specific examples, lets see how we
might derive it. There are many ways. In the next exercise, we will work
through an approach described by Francois Vi`ete in 1591. More will be
said about Vi`ete in the historical discussion of Section 5.8.
Exercise 3.7. We begin with the cubic polynomial y 3 C py C q. The key
idea (by no means obvious!) is to introduce a new variable z satisfying
yDz

p
:
3z

If p D 0, this change of variable accomplishes nothing, so lets assume

until the last part of the exercise that p 0.
(i) Substitute z p=3z for y in y 3 C py C q D 0, expand the cubed term,
simplify, and obtain an equation in z:
z3

p3
C q D 0:
27z 3

(ii) Multiply by z 3 to clear denominators and obtain

z 6 C qz 3

p3
D 0:
27

(iii) This is a quadratic equation in z 3 . Use the quadratic formula to obtain

p
q
q 2 C .4p 3 =27/
3
z D

:
2
2
(iv) Introduce R as an abbreviation for p 3 =27 C q 2 =4 and rewrite the last
equality as
q p
z3 D
R:
2
p
(v) p
There are two possible values for z 3 , either .q=2/C R or .q=2/
R. Multiply them and simplify to get
 q p  q
C R
2
2

p   p 3
R D
:
3

i
i

i
IrvingBook 2013/5/22 15:39 page 53 #69

53

3.2. Cardanos Formula

(vi) Take cube roots of both sides and deduce that the two values of z have
a product satisfying
r
 r

q p
q p
p
3
3
C R
:
R D
2
2
3
(vii) This means that if z is the cube
p root of .q=2/ C
the cube root of .q=2/
R.

p
R, then p=3z is

(viii) Recall that z was introduced to satisfy y D z p=3z. The two terms on
the
p right of this equation,
p z and p=3z, are the cube roots of .q=2/C
R and of .q=2/
R.
(ix) Conclude that y is the sum of the two cube roots:
r
r
q p
q p
3
yD
R:
C R C 3
2
2
(x) We have assumed that p 0. If p D 0, then the equation we are
solving is y 3 C q D 0 and the formula still holds. (Check that one
summand becomes 0 and the other is the cube root of q.)
We have obtained Cardanos formula! That wasnt hard at all. However,
in the derivation of the formula, we have not been precise on certain points.
The imprecisions will be addressed in Exercise 5.1, where we review the
derivation with the benefit of complex numbers. For now, let us ignore these
imprecisions and give the formula a try.
Exercise 3.8.

Solve y 3

3y C 2 D 0.

(i) Use Cardanos formula to find one solution r .

(ii) Use it to factor y 3 3y C 2 as the product of linear and quadratic
polynomials and find the other two solutions.
The next equation is a famous example of Cardano.
Exercise 3.9.

Solve y 3 C 6y

20 D 0.

(i) Use Cardanos formula and obtain

q
q
p
3
3
y D 10 C 6 3 C 10

p
6 3:

i
i

i
IrvingBook 2013/5/22 15:39 page 54 #70

54

3. Cubic Polynomials

(ii) Observe that the function y 3 C 6y 20 is always rising as y increases.

(Consider the behavior of the separate terms y 3 and 6y.)
(iii) Deduce that the graph of y 3 C6y 20 crosses the y-axis (the horizontal
axis) only once, so that y 3 C 6y 20 D 0 has only one real solution.
(iv) Conclude that the expression we have obtained for y using Cardanos
formula must be the lone real solution.
We can stop here, content that we have found the single real root of
y 3 C6y 20. Or we can use a calculator to determine at least approximately
what the sum of the two cube roots is. If we do this, we discover that it is
close to 2. Substituting 2 for y in y 3 C 6y 20 to see how close to 0 we
are, we further discover that
23 C 6
2

20 D 8 C 12

20 D 0:

In other words, 2 is not just approximately a root of y 3 C 6y 20. It is a

root! Since y 3 C 6y 20 has only one real root, the complicated expression
we found in Exercise 3.9 for the root as a sum of cube roots must equal 2:
q
q
p
p
3
3
10 C 6 3 C 10 6 3 D 2:
This is at least a little disappointing. What if we didnt recognize that
the sum of cube roots offered by Cardanos formula is simply 2? It turns
out, for this example at least, that we can calculate the cube roots arising
p
as summands in Cardanos formula explicitly as expressions involving 3
and then verify that their sum is 2. Lets do so.
Exercise 3.10. We will calculate the cube roots that appear in Exercise 3.9
by guessing and verifying.
p
p
(i) Guess that the cube root of 10 C 6 3phas the form a C b 3. Cube
this. Combine terms that dont involve 3 and terms that do to get two
equations in a and b with integer coefficients.
(ii) Theres no systematic way to solve two simultaneous equations such
as these in general. Instead, turn again to guessing. Try small integers
as values for a and b to find a solution.
p
(iii) Find a real cube root of 10 6 3 in a similar way.
(iv) Verify that the sum of the two cube roots is 2.

i
i

i
IrvingBook 2013/5/22 15:39 page 55 #71

55

3.2. Cardanos Formula

Lets try another example where the same difficulty arises.

Exercise 3.11. Solve y 3

2y

4 D 0.

(i) Use Cardanos formula to obtain

s
s
r
3
100
3
y D 2C
C 2
27

100
:
27

(ii) Rewrite this as

r
3

10 p
2C
3C
9

r
3

10 p
3:
9

(iii) Taking a hint from Exercise 3.10, lets see

p if we can obtain a simpler
expression. Find
the
cube
root
of
2
C
10
3=9 by guessing
that it has
p
p
the form a C b 3. Cube this, set it equal to 2 C 10 3=9, and obtain
two equations for a and b. Try small integers for a and possibly small
fractions for b to find cube roots.
p
(iv) Obtain a cube root of 2 10 3=9 as well, and deduce from Cardanos
formula that 2 is a solution of y 3 2y 4 D 0.

(v) Factor y 3 2y 4 as the product of y 2 and a quadratic polynomial.

Because the quadratic polynomial has no real roots, 2 is the only real
root of y 3 2y 4.

Once again, Cardanos formula has given us a solution to a cubic equation as a frightful sum of cube roots that turns out to be a simple number.
Lets consider yet another example. A surprise awaits, one of the great
surprises in the history of mathematics.
Exercise 3.12. Solve y 3

7y C 6 D 0.

(i) Use Cardanos formula to obtain

s
s
r
3
100
3
3C
C
3
yD
27

100
27

as one root.
(ii) Rewrite this as
r
3

3C

10 p
3C
9

r
3

10 p
3:
9

i
i

i
IrvingBook 2013/5/22 15:39 page 56 #72

56

3. Cubic Polynomials

The form of the solution in Exercise 3.12 is similar to the form of the
solution in Exercise 3.11, the only essential difference being the minus sign
under the square root. But what a difference! Because of the minus sign, the
solution makes no sense. After all, there is no square root of 3.
Lets not worry about the meaninglessness of our solution
pjust yet. Instead, taking
3 the way
p a hint from Exercises 3.10 and 3.11, lets treat
to
add
to obwe did 3 and try to find cube roots that we might be able
p
tain a simpler answer. Even though we dont know what
3 means, lets
assume in the exercises that when we square it, the result is the number 3.
p
Exercise 3.13.
3=9 has a cube root in
p Lets guess that 3 C 10
p the
form a C b
3. We defer any concerns about the meaning of a C b
3,
choosing for now just to work with it formally.
p
p
(i) Cube a C b
3. Combine terms that dont involve
3 and terms
that do to get two equations in a and b with integer coefficients.
(ii) Make guesses for a and b. Small positive integers for a and fractions
involving thirds for b should give a solution quickly.
p
(iii) Find the cube root of 3 10
3=9 similarly.
p
p
3=9 and 3 10
3=9
(iv) The expressions for the cube roots of 3C10
involve the square root of 3. Lets not worry yet. Lets proceed.
p
(v) Add the two cube roots: the troublesome
3 terms cancel, leaving
us with a meaningful real number, 2. Verify that 2 is a solution to y 3
7y C 6 D 0.
(vi) Conclude that Cardanos formula has led us to a correct solution of the
equation y 3 7y C 6 D 0, namely, y D 2.
(vii) Since 2 is a solution, the polynomial y 3 7y C 6 factors as the product
of y 2 and a quadratic polynomial. Divide y 2 into y 3 7y C 6 to
find the quadratic polynomial, then determine its roots, thereby finding
all three roots of y 3 7y C 6.
Cardanos formula has successfully produced the roots of y 3 7y C 6,
provided
we are willing to work formally with the meaningless expression
p
3.
We can use Exercise 3.13 to solve the cubic equation introduced in Exercise 3.3.

i
i

i
IrvingBook 2013/5/22 15:39 page 57 #73

57

3.2. Cardanos Formula

Exercise 3.14. Solve

x3

3x 2

4x C 12 D 0

by using the change of variable of Theorem 3.3 and applying the result of
Exercise 3.13.
Lets conclude this section withpone more illustration of what becomes
possible once we open the door to
3.
Exercise 3.15. Solve x 3

1 D 0.

(i) Observe that 1 is a solution.

(ii) Factor x 3

1 as .x

1/.x 2 C mx C n/ and determine m and n.

(iii) Use the quadratic formula to solve x 2 C mx C n D 0. Obtain in this

way two additional solutions of x 3 D 1, the two numbers
1

(iv) For the displayed numbers to make sense, there must be a square root
of 3. Lets continue not to worry about this. It will be convenient to
have a special name for the hypothetical number
1
C
2

p
2

We will use the

p lower case Greek letter omega for this purpose and
write 1=2 C
3=2 as !.
(v) Verify that the other hypothetical number,
p
1
3
;
2
2
is ! 2 .
(vi) Multiply ! by ! 2 to get 1. In other words, verify that ! 3 D 1. Similarly, verify that .! 2 /3 D 1. Thus, ! and ! 2 are both solutions to
x 3 D 1.
(vii) Conclude that the three cube roots of 1 are 1, !, and ! 2 .

i
i

i
IrvingBook 2013/5/22 15:39 page 58 #74

58

3. Cubic Polynomials

Exercise 3.16. Let c be any non-zero real number. From Exercise 3.5, c
has one real cube root. Call it a. Provided that the new number ! introduced
in the preceding exercise makes sense, show that !a and ! 2 a are also cube
roots of c. Conclude that if we decide to treat ! as an allowable number, we
will find that the polynomial x 3 c has three distinct roots, one real and the
others of a new form.

3.3

Graphs

In this section, we will look at the shapes of cubic polynomial graphs. The
results of this section are needed only in Section 3.4, where we take our
first look at the discriminant of a cubic. We will return to the study of the
discriminant in Section 5.3 and obtain results purely algebraically, without
reference to this section. Thus, we might choose to omit both this section
and Section 3.4. However, the principal results enhance our visual or geometric understanding of cubic polynomials.
To give full proofs of the results of this section, some background from
the foundations of real numbers and calculus is needed. Readers with that
background, including familiarity with the connection between derivatives
and turning points summarized in Section 1.6, will be able to prove the results easily. For those unfamiliar with calculus, we will indicate an approach
that reduces the calculus to a minimum, although some foundational theorems on the real numbers are needed.
We will restrict ourselves to reduced cubic polynomials, those of the
form x 3 C px C q. As we saw in Section 3.1, this is in fact no restriction
at all. Recall from Section 1.6 the notions of local maximum, local minimum, and turning point for the graph of a function y D f .x/. We know
from Theorem 1.15 that the graph of x 3 C px C q has at most two turning
points. We know from Theorem 1.13 that the graph rises from arbitrarily
low heights on the left of the y-axis to arbitrarily high heights on the right.
The theorems tell us that the graph of a cubic has two possible behaviors: it
will rise steadily as x increases or it will rise to a local maximum, fall to a
local minimum, then rise. Which one occurs depends on the sign of p.
The possibilities are illustrated in Figure 3.1, which shows the graphs of
the cubic polynomials y 3 C px for p D 4; 2; 0; 2, and 4. For p D 2 or
4, (or, more generally, p positive), the graph rises steadily, with no turning
points. For p D 2 or 4 (or, more generally, for p negative), the graph
rises, then falls, then rises. An exceptional case occurs when p D 0. Here,
the graph never turns, but at x D 0, it stops rising for an instant, being
essentially flat. More precisely, the x-axis is tangent to the graph at .0; 0/,

i
i

i
IrvingBook 2013/5/22 15:39 page 59 #75

59

3.3. Graphs

y = x 3 + 4x
y = x 3 + 2x
y = x3
y = x 3 2x
y = x 3 4x
x

Figure 3.1. Shapes of cubic polynomial graphs

so that the slope of the graph there is 0.
Let us return to arbitrary reduced cubic polynomials.
Exercise 3.17. Let p and q be real numbers with p  0. Let f .x/ D
x 3 C px C q.
(i) Show for any two real numbers a < b that f .a/ < f .b/.

(ii) Conclude, using Theorem 1.14, that Theorem 3.4 holds.

Theorem 3.4. Let p and q be real numbers with p  0. Let f .x/ D
x 3 C px C q.

(i) The graph of y D f .x/ never falls as x increases. In particular, the

graph of y D f .x/ has no turning point.

(ii) The graph crosses the x-axis exactly once, so that f .x/ has exactly
one real root.
The graph of y D x 3 C px C q crosses the y-axis at .0; q/. Those
familiar with calculus will recognize that the slope of the graph at .0; q/ is
p. Thus, the larger p is, the steeper the graph is at .0; q/ and the smaller
p is, the flatter the graph is at .0; q/. In the extreme case that p D 0, the
polynomial f .x/ is just x 3 C q and the slope at .0; q/ is 0. Thus, the graph
is flat there, with tangent line y D q.
When p < 0, the graph of x 3 C px C q has two turning points. We can
use elementary calculus to prove this and determine where they are:

i
i

i
IrvingBook 2013/5/22 15:39 page 60 #76

60

3. Cubic Polynomials

Theorem 3.5. Let p and q be real numbers with p < 0 and let f .x/ D
x 3 C px C q. Let a be the positive square root of p=3.
(i) The graph of y D f .x/ has two turning points, a local maximum at
x D a and a local minimum at x D a.
(ii) As x increases, the graph rises to a turning point at


2ap
a; q
;
3
falls to a turning point at


2ap
a; q C
;
3
then rises.
Exercise 3.18. We can use elementary calculus to prove Theorem 3.5:
(i) Calculate where f 0 .x/ takes on the value 0 and deduce that x D
and x D a are the only candidates for turning points.

(ii) Determine the sign of f 0 .x/ for values of x satisfying x < a, a <
x < a, and a < x. Deduce that there are indeed turning points at
x D a and x D a.
(iii) Calculate the heights f . a/ and f .a/ of the turning points.
(iv) Deduce that the theorem holds.
We can prove much of Theorem 3.5 without calculus. Lets do so, to
get an enhanced understanding of why Theorem 3.5 holds. The shape of the
graph of x 3 C px C q is independent of q. As q varies, the graph slides up
and down (as described in Exercise 2.11), but otherwise remains unchanged.
Exercise 3.19. Let p and q be real numbers with p < 0 and let a be
the positive square root of p=3. In this exercise, we will find two turning
points for the graph of y D x 3 C px C q. We can rewrite the cubic as
x 3 3a2 x C q.
(i) Let g.x/ D x 3 3a2 x C 2a3 . Because a is a root, x a divides g.x/.
Factor g.x/ as .x a/2 .x C 2a/. Use this factorization to verify for
x > a that as x increases, so does g.x/. Also check that g.a/ D 0, but
g.x/ > 0 for any x  0 besides x D a.

i
i

i
IrvingBook 2013/5/22 15:39 page 61 #77

3.4. A Discriminant

61

(ii) Deduce that .a; 0/ is a local minimum for g.x/ and that the graph of
g.x/ rises to the right of x D a as x increases.
(iii) Let h.x/ D x 3 3a2 x 2a3 . Because a is a root, x Ca divides h.x/.
Factor h.x/ as .x C a/2 .x 2a/. Use this factorization to verify for
x < a that as x increases, so does h.x/. Also check that h. a/ D 0,
but h.x/ < 0 for any x  0 besides x D a.
(iv) Let q be a real number and let f .x/ D x 3 3a2 x C q. The graph of
f .x/ is just a vertical shift of the graphs of g.x/ and h.x/. Conclude
that f .x/ has a local maximum at . a; q 2ap=3/ and a local minimum at .a; q C 2ap=3/, with the graph rising as x increases to a and
as x increases from a.
Lets continue with the notation of Exercise 3.19. We might guess from
the exercise that the graph of y D x 3 C px C q falls as x goes from a to a.
Theorem 3.5 tells us that this is the case, as we easily check using calculus,
since we can verify that the derivative 3x 2 C p of x 3 C px C q is negative
for all values of x satisfying a < x < a.
Without calculus, we can still show with a little more work (and an
appeal to basic facts from the foundations of real numbers) that x 3 Cpx Cq
decreases as x goes from a to a. For instance, we can show for each real
number r between a and a that there is some open interval around r on
which x 3 C px C q is decreasing. This local information, together with
basic results on the real numbers, shows x 3 C px C q is decreasing across
the entire interval . a; a/. Rather than pursuing this point further, we will
accept that it is true, as we already know from calculus.

3.4 A Discriminant
For a quadratic polynomial x 2 C bx C c, we know that there are three
possibilities for the roots: two distinct real roots, one real root repeated, or
no real roots. We also know that we can decide which case holds from the
coefficients b and c, as we saw in Theorem 2.5. The three cases correspond
to the quantity b 2 4c, the discriminant of x 2 C bx C c, being positive,
zero, or negative. This can be proved by using elementary calculus or the
quadratic formula.
For a cubic polynomial x 3 C bx 2 C cx C d , as for a quadratic polynomial, it is possible to determine from the coefficients the nature of the
cubics roots: whether the cubic has one simple real root, three distinct real
roots, or repeated real roots. This is governed by an expression in b, c, and

i
i

i
IrvingBook 2013/5/22 15:39 page 62 #78

62

3. Cubic Polynomials

d , known as the discriminant. We will discuss an algebraic approach to the

discriminant in Section 5.3. An alternative approach based on an understanding of the graph of a cubic polynomial (and therefore depending on
calculus) is also possible. In this section, we will describe this approach for
a reduced cubic polynomial, using the results of Section 3.3.
We continue to work with a cubic polynomial x 3 Cpx Cq, where p and
q are real numbers, and we will write it as f .x/. Our analysis, as the results
of Section 3.3 would suggest, depends on whether p is positive, negative,
or 0. In case p D 0, we analyzed the polynomial x 3 C q in Exercise 3.5.
It has one real root, the real cube root of q, which is of multiplicity 3 if
q D 0 and of multiplicity 1 otherwise.
Exercise 3.20. Let f .x/ D x 3 C px C q with p > 0.

(i) Recall from Theorem 3.4, or observe anew, that the graph of y D f .x/
is always increasing and that therefore f .x/ has only one real root.

(ii) Let r be the real root. From Theorem 3.2, either r has multiplicity 1 as
a root of f .x/, in which case f .x/ factors as the product of x r and
a quadratic polynomial that is not a product of linear polynomials, or
r has multiplicity 3, in which case f .x/ factors as .x r /3 . Show that
f .x/ cannot factor as .x r /3 and conclude that r has multiplicity 1.
We will handle the case p < 0 in Exercise 3.21, using Theorem 3.5.
It will be helpful to have a picture in mind of the possible behaviors of the
graphs. We will use Figure 3.2 as a guide. The figure shows the graphs of the
cubic polynomials x 3 3x C q for q D 4; 2; 0; 2, and 4. All five graphs
y

y = x 3 3x + 4
y = x 3 3x + 2
y = x 3 3x
y = x 3 3x 2
y = x 3 3x 4
x

Figure 3.2. Intersection of y D x 3

3x C q with x-axis

i
i

i
IrvingBook 2013/5/22 15:39 page 63 #79

63

3.4. A Discriminant

have the same shape, as we would expect. They are translations of each
other, up or down. How far each one is shifted determines how the x-axis
meets it and therefore how many roots the corresponding polynomial has.
Imagine a video showing the graph of y D x 3 3x C q as q increases
from 10 to 10. At the start of the video, the graph of y D x 3 3x 10
is shown. As time goes by, the curve steadily rises, until at the end of the
video we arrive at the graph of y D x 3 3x C 10.
As suggested in Figure 3.2 with q D 4, at the videos start, the curve
will cross the x-axis only once. When q D 2, the curve will cross the
x-axis off to the right, but suddenly the turning point on the left makes
contact with the x-axis, producing a second root. Once q increases above
2, the turning point rises above the x-axis and is replaced by two points
of intersection on the left, along with the point of intersection on the right.
Thus, as the figure illustrates in the case of q D 0, whenever 2 < q < 2
the graph of x 3 3x C q crosses the x-axis three times and there are three
roots. The behavior changes when q D 2. If we were to watch the video as
q increases from 2 to 2, we would see the middle of the three intersection
points approach the rightmost intersection point, until they merge when q D
2. The graph of x 3 3x C 2 therefore meets the x-axis only twice and the
polynomial has two roots. Finally, as q increases beyond 2, the graph no
longer makes contact with the x-axis to the right, crossing only on the left,
with x 3 3x C q having only one root.
Exercise 3.21. Let f .x/ D x 3 C px C q with p < 0. Let a be the positive
square root of p=3.
(i) Recall from Theorem 3.5 that the graph of y D f .x/ has two turning
points, a local maximum at . a; q 2ap=3/ and a local minimum at
.a; q C 2ap=3/.
(ii) The real roots of f .x/ occur at the x-coordinates of the points where
the graph of y D f .x/ meets the x-axis. Geometrically there are five
cases, as depicted in Figure 3.2 and described in the discussion that
preceded this exercise:
(a) The x-axis crosses the graph above the local maximum:
f .a/ < f . a/ < 0:
(b) The x-axis crosses the graph at the local maximum:
f .a/ < f . a/ D 0:

i
i

i
IrvingBook 2013/5/22 15:39 page 64 #80

64

3. Cubic Polynomials

(c) The x-axis crosses the graph between the local minimum and the
local maximum:
f .a/ < 0 < f . a/:
(d) The x-axis crosses the graph at the local minimum:
0 D f .a/ < f . a/
(e) The x-axis crosses the graph below the local minimum:
0 < f .a/ < f . a/:
(iii) For each of these five cases, describe the number of real roots of f .x/
and their multiplicity, referring to Theorem 3.2 for guidance on the
options.
(iv) Express the last result by saying that there are three possibilities for
the nature of the roots, depending on whether f . a/ and f .a/ have
opposite sign, f . a/ and f .a/ have the same sign, or one of them is
0.
(v) The three cases can be described more compactly in terms of the sign
of the product of f . a/ and f .a/:
(a) If f . a/f .a/ < 0, then f .x/ has three distinct real roots.
(b) If f . a/f .a/ > 0, then f .x/ has one real root, of multiplicity 1.
(c) If f . a/f .a/ D 0, then f .x/ has two distinct real roots, of multiplicities 1 and 2.
(vi) Calculate the product f . a/f .a/ to get
f . a/f .a/ D q 2 C

4p 3
;
27

so
27f . a/f .a/ D 4p 3

27q 2 :

Given the cubic polynomial x 3 C px C q, whatever the sign of p, lets

introduce a name and a notation for the quantity that appeared at the end of
Exercise 3.21. Set
D 4p 3 27q 2
and call the discriminant of x 3 C px C q. We will discover an algebraic
interpretation of this quantity in Section 5.3.

i
i

i
IrvingBook 2013/5/22 15:39 page 65 #81

65

3.4. A Discriminant

Exercise 3.22. Lets continue the analysis of the roots of the cubic polynomial x 3 C px C q.
(i) Assume p D 0. Then x 3 C px C q has one real root, which has
multiplicity 1 if q 0 and multiplicity 3 if q D 0. Verify that < 0 if
q 0 and D 0 if q D 0.

(ii) Assume p > 0. From Exercise 3.20, x 3 C px C q has one simple real
root. Verify that < 0.

(iii) Assume p < 0. Deduce from Exercise 3.21 that when the graph of
y D x 3 C px C q crosses the x-axis three times, so that x 3 C px C q
has three distinct real roots, > 0. Deduce that when the graph crosses
the x-axis only once, so that x 3 C px C q has one real root, < 0.
Deduce that when the graph crosses the x-axis once and is tangent to
the x-axis once, so that there is a repeated real root, D 0.
We have considered all possible cases. Reviewing them, we see that is
positive precisely when there are three distinct real roots, negative precisely
when there is only one simple real root, and 0 when there is a repeated real
root. Thus, we have proved the following theorem:
Theorem 3.6. Let f .x/ D x 3 C px C q for real numbers p and q. Let
D 4p 3 27q 2 , the discriminant of f .x/.
(i) If > 0, then f .x/ has three distinct real roots.

(ii) If D 0, then f .x/ has a repeated real root.

(iii) If < 0, then f .x/ has one simple real root.
Exercise 3.23. Using Theorem 3.6, describe the nature of the roots of each
of these polynomials:
(i) x 3

3x C 2.

(ii) x 3 C 6x

20.

(iii) x

2x

4.

(iv) x 3

7x

6.

Cardanos formula for a root of the reduced cubic polynomial x 3 Cpx C

q includes the square root of the quantity p 3 =27 C q 2=4, which we denoted
by R. The discriminant and R are constant multiples of each other:
D 108R:

i
i

i
IrvingBook 2013/5/22 15:39 page 66 #82

66

3. Cubic Polynomials

If R is negative, Cardanos formula contains square roots of negative

numbers. Since R is negative when is positive, and is positive precisely
when f .x/ has three distinct real roots, we see that the troublesome situation of having to deal with square roots of negative numbers in using
Cardanos formula arises precisely when f .x/ has three distinct real roots.
We will discuss this further in Section 5.4.

3.5

History

We closed our account of quadratic equations in Section 2.5 with Luca Pacioli, who summarized the mathematical knowledge of the time in 1494 [50].
Regarding cubic and quartic equations, Pacioli wrote [66, p. 47] that it has
not been possible until now to form general rules. This was the setting at
the dawn of the sixteenth century.
By the centurys end, general rules would be in place, thanks to the work
of Cardano and other Italian mathematicians. The story of these discoveries
is a wonderful one. Given the excellent accounts in [67] and [66], as well
as the 1953 biography Cardano: The Gambling Scholar [49] by the mathematician ystein Ore, only a brief account will be given here. (Ores biography is unfortunately no longer in print. For more on Cardano, one can
also turn to Anthony Graftons Cardanos cosmos: the worlds and works
of a Renaissance astrologer [33] as well as Cardanos autobiography, The
Book of My Life [13], available in a 2002 edition that contains both Jean
Stoners 1929 English translation and an introduction by Grafton.)
An important point to keep in mind in preparation for the story is that
the academic culture at the time was not anything like that to which we are
accustomed. If someone were now to solve a centuries-long problem, he or
she would immediately announce it, lecture on it, and publish it, with the
written account being made available on the internet long before it actually
appears in print. Other mathematicians would study the solution, make sure
there are no errors, and pay close attention to the methods used for the solution, in anticipation that the methods and the new ideas they contain may be
applicable to other problems. In contrast, in sixteenth century Italy, scholars
would keep techniques of solution to themselves, perhaps employing them
to succeed at public competitions.
Scipione del Ferro was the first person to obtain a formula for solutions
to cubic equations. Del Ferro lived from 1465 until 1526, serving as a professor at the University of Bologna for the final thirty years of his life. He
found a way to solve cubic equations of the form
x 3 C px D q:

i
i

i
IrvingBook 2013/5/22 15:39 page 67 #83

67

3.5. History

Del Ferro and his contemporaries worked with positive numbers only, both
as coefficients and as solutions, so p and q are understood to be positive
and we seek positive values of x. An example would be x 3 C 6x D 20, the
cubic equation of Exercise 3.9.
We saw in Section 2.5 that seven centuries earlier, al-Khwarizmi had
classified quadratic equations into three forms (excluding cases in which
one of the coefficients is 0). Similarly, because of the restriction to positive
coefficients, there are three forms for cubic equations with no degree 2 term,
the form x 3 C px D q that del Ferro studied as well as
x 3 D px C q
and
x 3 C q D px:
Were 0 available, we might momentarily be tempted to add a fourth case,
x 3 C px C q D 0:
But since p and q are positive, it cannot have a positive solution. Hence, we
are missing nothing by omitting it.
Del Ferro did not publish his solution, but he did communicate it to his
son-in-law, Annibale della Nave, and his colleague Antonio Maria Fior. In
1535, Fior challenged the Venetian mathematician Niccol`o Fontana (1499
1557) to a contest in which each would pose thirty problems to the other, the
loser paying for a banquet for thirty. Fontana, better known by his nickname
Tartaglia, or the stutterer, prepared problems of varying types. In contrast,
all of Fiors problems were cubic equations of the type x 3 C px D q, with
del Ferros solution as his secret weapon.
Just before time expired, on the night that ran from February 12 to 13,
Tartaglia found a way to solve the equation x 3 C px D q on his own! So
much for Fiors secret weapon. Tartaglia solved all thirty of Fiors problems,
whereas Fior could solve only some of Tartaglias. Victory was sufficient
satisfaction for Tartaglia, who chose to forgo the banquet. Of course, he had
also won eternal fame as independent co-discoverer, with del Ferro, of the
cubics solution.
The story now shifts to another participant, the one for whom the solution to the cubic equation is named, Girolamo Cardano. Cardano was a
prominent scholar in many fields, famous as a doctor, astrologer, philosopher, and mathematician. He lived from 1501 to 1576 and spent most of
his life in the city of his birth, Milan. In 1539, having heard of Tartaglias
solution to the cubic, he asked Tartaglia through an intermediary what the

i
i

i
IrvingBook 2013/5/22 15:39 page 68 #84

68

3. Cubic Polynomials

solution was, but Tartaglia chose not to tell. Cardano then invited Tartaglia
to come to Milan as his guest, enticing him with the opportunity to meet Alfonso dAvalos, the military commander of Milan, to whom Tartaglia would
be able to show some of his military inventions.
Like Cardano, Tartaglia was a man of many talents, including expertise
in ballistics and military engineering. In 1546, he would publish some of his
military work in Quesiti et Inventioni Diverse, or New Problems and Inventions [63]. (Its drawings of cannons, cannonball paths, and fortifications are
reason enough to seek out a copy of the book.) Naturally, Tartaglia accepted
Cardanos invitation. During his stay, Tartaglia told Cardano how he solved
the equation x 3 C px D q, with Cardano swearing an oath on March 29,
1539 never to publish it.
We should note at this point that the principal source for many details
of this story, including Cardanos oath, is Tartaglia himself, as he would
devote the final pages of Quesiti et Inventioni Diverse to an account of his
dealings with Cardano.
After Tartaglia left, Cardano saw how to use Tartaglias ideas in order
to obtain solutions to the cubic equations of the forms x 3 D px C q and
x 3 C q D px. Difficulties arise in these two cases because for certain
values of p and q, the solutions may contain expressions with square roots
of negative numbers, as we saw in Exercise 3.12. (In contrast, this cannot
occur in the expression for a solution of the cubic equation x 3 C px D
q.) Nonetheless, in principle Cardano had found a solution to any cubic
equation without a degree 2 term, and as we know, it is then an elementary
matter to solve all cubic equations.
The fourth participant in our story is Lodovico Ferrari, who lived from
1522 to 1565. Ferrari came from Bologna to Milan at the age of 14 to work
as a servant in Cardanos household. Cardano quickly realized Ferraris
talent, and Ferrari moved from servant to student to collaborator. Ferrari
learned from Cardano the method of solving cubic equations, then made
his own great contribution: the solution of quartic equations. What Ferrari
discovered is that one can reduce the problem of solving a quartic equation
to that of solving an auxiliary cubic equation, one whose coefficients are
expressed in terms of the coefficients of the original quartic. We will study
Ferraris discovery in Section 6.2.
Ferraris result was an advance of the greatest importance, but it posed
a difficulty for Cardano. He had given his oath to Tartaglia that he would
not publish Tartaglias solution to cubic equations of a special form. Yet,
Cardano had extended this to other cubic equations and his disciple Ferrari
had shown how to use it to solve quartic equations. This was too important

i
i

i
IrvingBook 2013/5/22 15:39 page 69 #85

3.5. History

69

to keep secret. Moreover, he may have had a way out, since the formula was
initially due to del Ferro, not Tartaglia.
Cardano decided to publish the solution, which he did in the book Ars
Magna, written in Latin and published in 1545. (It is available in an English
translation by T. Richard Witmer, with the title The Great Art; or, The Rules
of Algebra [12].) He states clearly at the beginning of Ars Magna [12, pp. 8
9] that the solution to the cubic equation in the special form x 3 C px D q
was discovered by del Ferro and re-discovered by Tartaglia:
In our own days Scoipione del Ferro of Bologna has solved the case
of the cube and first power equal to a constant, a very elegant and admirable accomplishment. Since this art surpasses all human subtlety
and the perspicuity of mortal talent and is a truly celestial gift and a
very clear test of the capacity of mens minds, whoever applies himself to it will believe that there is nothing he cannot understand. In
emulation of him, my friend Niccol`o Tartaglia of Brescia, wanting not
to be outdone, solved the same case when he got into a contest with
his [Scipioines] pupil, Antonio Maria Fior, and, moved by my many
entreaties, gave it to me. For I had been deceived by the words of Luca
Paccioli, who denied that any more general rule could be discovered
than his own. Notwithstanding the many things which I had already
discovered, as is well known, I had despaired and had not attempted
to look any further. Then, however, receiving Tartaglias solution and
seeking for the proof of it, I came to understand that there were a great
many other things that could also be had. Pursuing this thought and
with increased confidence, I discovered these others, partly by myself
and partly through Lodovico Ferrari, formerly my pupil. Hereinafter
those things which have been discovered by others have their names
attached to them; those to which no name is attached are mine. The
demonstrations, except for the three by [al-Khwarizmi] and the two by
Lodovico, are all mine.
Cardano turns to cubic equations in Chapter 11, On the Cube and First
Power Equal to the Number, which is devoted to equations of the form
x 3 C px D q. At the start of the chapter, he once again credits del Ferro
and Tartaglia [12, p. 96]:
Scipio Ferro of Bologna well-nigh thirty years ago discovered this rule
and handed it on to Antonio Maria Fior of Venice, whose contest with
Niccol`o Tartaglia of Brescia gave Niccol`o occasion to discover it. He
[Tartaglia] gave it to me in response to my entreaties though withholding the demonstration. Armed with this assistance, I sought out its

i
i

i
IrvingBook 2013/5/22 15:39 page 70 #86

70

3. Cubic Polynomials

demonstration in [various] forms. This was very difficult. My version

of it follows.
By modern standards, Cardano had given proper credit for the results
on which he and Ferrari built, and had satisfied all scholarly expectations.
But this was a different era, and Tartaglia was furious. As already noted, he
told his side of the story a year later in Quesiti et Inventioni Diverse [63],
making the story of the oathand its textpublic.
Cardano did not respond to Tartaglias accusations, but Ferrari, on Cardanos behalf, issued a public challenge. The dispute was argued in Milan
on August 10, 1548, with Tartaglia leaving before it was settled and evidently being named the loser. (See Ores discussion [49, pp. 99105] for an
attempt to sort out the details from the available records.)
These discoveries make for a richly entertaining story, worthy of the
greatness of the discoveries themselves. As Varadarajan notes in concluding his account of the dispute [67, p. 62], Tartaglia, through his penchant
for secrecy, represents the middle ages, while Cardano, with his generous
views about authorship and sharing of knowledge, represents the modern
view. Mathematics was very fortunate that a person with the great vision
and world-view like Cardano was able to get his hands on the original discovery and was then able to build a wonderful structure on top of it that led
to the beginning of Algebra as we know today. Ore offers the following
conclusion [49, p. 106]:
Cardano and Ferrari represent by far the greater mathematical penetration and wealth of novel ideas. Tartaglia was also, doubtless, an excellent mathematician, but his great tragedy was the head-on collision
with the only two opponents in the world who could be ranked above
him. Without them, he would most likely have been reckoned as the
foremost mathematician in the middle of the sixteenth century.
As for Cardanos actual account of his formula, let us see what he says
in Chapter 11 of Ars Magna in his treatment of equations of the form x 3 C
px D q [12, pp. 9899]. Some of his terminology has been modernized
in the English translation. The terms binomium and
the
p apotome, which
p
translation retains, refer to pairs of numbers such as 5 C 3 and 5 3.
Cube one-third the coefficient of x; add to it the square of one-half the
constant of the equation; and take the square root of the whole. You
will duplicate this, and to one of the two you add one-half the number
you have already squared and from the other you subtract one-half the
same. You then have a binomium and its apotome. Then, subtracting

i
i

i
IrvingBook 2013/5/22 15:39 page 71 #87

71

3.5. History

the cube root of the apotome from the cube root of the binomium, the
remainder [or] that which is left is the value of x.
For example,
x 3 C 6x D 20:
Cube 2, one-third of 6, making 8; square 10, one-half of the constant;
100
p results. Add 100 and 8, making 108, the square root of which is
108. This you will duplicate: to one add 10, one-half the constant,
and frompthe other subtract the same.
p Thus you will obtain the binomium 108 C 10 and its apotome 108 10. Take the cube roots
of these. Subtract [the cube root of the] apotome from that of the binomium and you will have the value of x:
q
q
3 p
3 p
108 C 10
108 10:
One should compare this description with Exercise 3.9.
Cardano continues in Chapter 11 with two more examples [12, pp. 99
100], the equations x 3 C 3x D 10 and x 3 C 6x D 2. He then returns to the
first example and notes that the complicated difference of cube roots can be
calculated to be 2.
We will return to the cubic equations history in Section 5.8. We conclude for now with the opening words of Graftons introduction to Cardanos autobiography [13, p. ix]:
Cardano dazzled readers across sixteenth-century Europe. His original
and influential books dealt with medicine, astrology, natural philosophy, mathematics, and moralsto say nothing of devices for raising
sunken ships and stopping chimneys from smoking. They won the attention of popes and inquisitors, Catholics and Protestants, theologians
and playwrights. And nothing did more to enhance their appeal than
the polished stories about Cardanos past that glittered like enticing
gems in his most technical treatises, or the longer, dramatic retellings
of his whole life story that he offered the public at suitable intervals.

i
i

i
IrvingBook 2013/5/22 15:39 page 72 #88

i
i

i
IrvingBook 2013/5/22 15:39 page 73 #89

4
Complex Numbers
Our experience with Cardanos formula has taught
p us that to solve cubic
equations, we need to work with numbers such as
3, that is, square roots
of negative numbers. Cubic equations were where they were first encountered. They are now called complex numbers, and have been found to have
many important uses in mathematics and science. In this chapter, we will
introduce them and see how to calculate their nth powers and roots, which
is needed in our study of polynomial equations.

4.1 Complex Numbers

In this section we will develop the arithmetic
of complex numbers. We bep
gin bypintroducing a new number,
1, to serve as a square root of 1.
Once
1 is admitted topthe roll
numbers,p
we are led to more
p of allowable
p
new numbers, such as 3
1, 2 C
1, andp4 
1, when we try
to perform arithmetic operations that combine
1 and our old numbers.
In general, given real numbers
p a and b, we must admit to the roll of numbers the new number a C b
1. Numbers of this type are called complex
numbers. Real numbers are themselves complex numbers:
the real number
p
a can be thought of as the complex number
a
C
0
1.
p
Given a complex number a C b
1, the real number a is called its
real part and the real p
number b is called its imaginary part. For example,
the real part of 2 C 3
1 is 2 and the imaginary part is 3. Real numbers
are complex numbers with imaginary part 0. Also
p of note are the complex
numbers with real part 0, numbers of the form b
1, called pure imaginary
numbers.
73

i
i

i
IrvingBook 2013/5/22 15:39 page 74 #90

74

4. Complex Numbers

p
p
Suppose we are given two complex numbers, aCb
1 and c Cd
1.
How do we add and multiply them? Lets do what comes naturally and see
what happens. This will lead us to definitions of sum and product.
For addition, if we rearrange and combine terms, we obtain
p
p
p
p
.a C b
1/ C .c C d
1/ D a C c C b
1Cd
1
p
D .a C c/ C .b C d /
1:
This would appear to be a sensible definition of addition, and it is what we
adopt. For example,
p
p
p
.2 C 3
1/ C .8 C 2
1/ D 10 C 5
1:
How about multiplication? Lets use the distributive law and multiply
the same two numbers:
p
p
p
p
p
p
1/
.8 C 2
1/ D 2
8 C 2
2
1C3
1
8C3
1
2
1:
.2 C 3
p
It is natural to p
expect that
1 should commute with real numbers so that
p
1
8 D8

1. Assuming this and rearranging terms, we obtain
p
p
p
p
p
.2 C 3
1/
.8 C 2
1/ D 2
8 C 2
2
1C3
8
1 C 3
2.
1/2 :
p
If
1 is to make sense as a number, then it must have the property that its
square is 1:
p
.
1/2 D 1:
If we assume this, then we find that
p
p
p
p
p
.2C3
1/
.8C2
1/ D 16C4
1C24
1C6. 1/ D 10C28
1:
p
p
More generally, suppose we wish top
multiply aCbp 1 and cp
Cd
1.
Proceeding in the same way, assuming
1
c Dc

1 and .
1/2 D
1, we will obtain
p
p
p
.a C b
1/
.c C d
1/ D .ac bd / C .ad C bc/
1:

Let us adopt this as our definition of multiplication. The product of two

complex numbers is again a complex number, as would need to be the case
if the definition is to be of any use.
p
p
Lets summarize. Given complex numbers a C b
1 and c C d
1,
we have adopted the rules that their sum and product are given by
p
p
p
.a C b
1/ C .c C d
1/ D .a C c/ C .b C d /
1
and
.a C b

1/
.c C d

1/ D .ac

p
bd / C .ad C bc/
1:

i
i

i
IrvingBook 2013/5/22 15:39 page 75 #91

75

4.1. Complex Numbers

Exercise 4.1. Our definition of multiplication

for two complex numbers
p
was motivated
the assumptions that
1 commutes with real numbers
p by
and that .
1/2 D 1. Now that weve defined multiplication, we should
verify that these properties actually hold.
(i) Given a real number a, use the definition of multiplication to prove
that
p
p
a

1D
1
a:
p
(Hint: The real number a is the complex
1, while
p
p number a C 0

1 is the complex number 0 C 1

1.)
(ii) Use the definition of multiplication to prove that
p
.
1/2 D 1:
p
Often the symbol i is used in place of
1 to denote the square root
of 1. Theres no need to do this, but it does make our expressions easier
to write. In this notation, a complex number has the form a C bi , for real
numbers a and b, and i satisfies i 2 D 1. The sum of two complex numbers
a C bi and c C d i is
.a C bi / C .c C d i / D .a C c/ C .b C d /i
and their product is
.a C bi /.c C d i / D .ac
Exercise 4.2.
(i) 3 and 17
(ii) 3i and 17

bd / C .ad C bc/i:

Add and multiply the pairs of complex numbers.

15i .
15i .

(iii) i and i .
(iv) 4 C i and 3 C 7i .
(v)

4 C 5i and 3 C 6i .

(vi) 2 C 3i and 2

i.

(vii) 5 C 2i and 5

2i .

p
A positive real number r has a real square root r . When we square
p
a pure imaginary number r
i according to our multiplication rule, we
obtain
p
. r
i /2 D r:

i
i

i
IrvingBook 2013/5/22 15:39 page 76 #92

76

4. Complex Numbers

p
Thus r
i is a square root of r . By creating the new number i to serve
as a square root of 1, we have obtained square roots for all negative real
numbers.
A notion that we will need is that of complex conjugate. The complex
conjugate, or simply the conjugate, of a complex number a C bi is the
complex number a bi . The conjugate of a complex number r is denoted
by putting a bar over r to get r . In this notation, we would write for example
that
2 C 3i D 2 3i
and
12i D 12i:
The complex conjugate of a real number a is a itself and the complex conjugate of a pure imaginary number bi is its opposite bi .
A multiplicative inverse of a number r is a number s satisfying r s D 1.
For instance, the multiplicative inverse of 1 is 1 and the multiplicative
inverse of 2 is 1=2. What is the multiplicative inverse of  ? It is 1= , of
course. But this isnt an answer. Its a notation. The fact that there is a real
number s for which 
s D 1 is by no means obvious. It requires a proof, one
that depends on foundational results about the construction of real numbers.
(See, for instance, [38, pp. 7173] for a discussion.)
Let us accept the truth of the statement that every non-zero real number r has a multiplicative inverse, which we will write as 1=r . Thanks to
conjugation, we can deduce from this that non-zero complex numbers have
multiplicative inverses.
Exercise 4.3.

Let a and b be real numbers.

(i) Check that .a C bi /.a

bi / D a2 C b 2.

(ii) Assume that a C bi 0. Describe (in terms of a and b) the real and
imaginary parts of a complex number that is a multiplicative inverse of
a C bi .
Lets derive the basic facts on conjugation and the arithmetic operations
for complex numbers.
Exercise 4.4.

Let r and s be complex numbers.

(i) Show that

r Cs Dr Cs

i
i

i
IrvingBook 2013/5/22 15:39 page 77 #93

4.2. Quadratic Polynomials and the Discriminant

77

and
r
s D r s:
(ii) Use this to show that
r2 D r 2
and
r 3 D r 3:
(iii) We can use mathematical induction to show more generally that
rn D rn
for any positive integer n.
Lets restate one part of Exercise 4.4 as a theorem, for later reference.
Theorem 4.1. Let r and s be complex numbers.
(i) If r 2 D s, then r 2 D s.

(ii) If r 3 D s, then r 3 D s.
One consequence of Exercise 4.4 is Theorem 4.2. We wont use it until
Exercise 7.1, yet its proof is worth working out now as an exercise on conjugation facts. For quadratic, cubic, and quartic polynomials, we will obtain
Theorem 4.2 independently by direct determination of the roots.
Theorem 4.2. Let f .x/ be a positive-degree polynomial with real coefficients and let r be a root of f .x/ that is a non-real complex number. Then
r is also a root of f .x/.
Exercise 4.5.
f .r /.)

Prove Theorem 4.2. (Hint: Apply Exercise 4.4 to calculate

4.2 Quadratic Polynomials and the

Discriminant
We can use complex numbers to obtain a better understanding of the roots
of quadratic polynomials.
Exercise 4.6. Consider the quadratic polynomial x 2 C bx C c, with real
numbers b and c as coefficients. Assume that b 2 4c < 0. In this case
x 2 C bx C c has no real roots.

i
i

i
IrvingBook 2013/5/22 15:39 page 78 #94

78

4. Complex Numbers

(i) The quadratic formula gives as roots of x 2 C bx C c the two numbers

p
b
b 2 4c

:
2
2
Until now, we have viewed these as non-existent, since we had no
numbers available to serve as square roots of the negative real number b 2 4c. With the introduction of complex numbers, we now have
square roots.
(ii) Rewrite the two roots of x 2 C bx C c given in the previous part as
complex numbers, using i .
(iii) Substitute the resulting complex numbers into x 2 C bx C c and verify
that they are roots of x 2 C bx C c.
(iv) Draw the conclusion that if complex numbers are allowed, the quadratic
formula makes sense even when b 2 4c < 0.
(v) The two roots we have obtained in this way are complex conjugates of
each other. Write the roots as r and r . Verify that
x 2 C bx C c D .x

r /.x

r/

by multiplying out the right side of the equation.

(vi) Deduce that Theorem 4.3, a refinement of Theorem 2.2, holds.
Theorem 4.3. Let f .x/ D x 2 C bx C c for real numbers b and c. Exactly
one of three possibilities occurs:
(i) x 2 C bx C c has two distinct real roots a1 and a2 , and factors as
.x

a1 /.x

a2 /:

(ii) x 2 C bx C c has only one real root a, and factors as .x

a/2 .

(iii) x 2 C bx C c has two distinct complex roots r and r, and factors as

.x r /.x r/.
Moreover, the first possibility occurs if b 2 4c > 0, the second possibility
occurs if b 2 4c D 0, and the third possibility occurs if b 2 4c < 0.
We saw in Theorem 2.5 that the nature of the roots of a quadratic polynomial x 2 C bx C c is determined by the sign of b 2 4c, which we denoted
. In the next exercise, we will see that the nature is determined by the sign
of the square of the difference of the polynomials roots.

i
i

i
IrvingBook 2013/5/22 15:39 page 79 #95

4.2. Quadratic Polynomials and the Discriminant

79

Exercise 4.7. Suppose b and c are real numbers. Write r1 and r2 for the
roots of x 2 C bx C c and for .r1 r2 /2 .
(i) Conclude from Theorem 4.3 that there are three mutually exclusive
possibilities for r1 and r2 : they are real and distinct, or real and coincident, or distinct complex conjugates of each other.
(ii) If r1 D r2 , then D 0.
(iii) If r1 and r2 are real and distinct, then > 0.
(iv) The only remaining possibility is that r1 and r2 are not real but are
complex numbers, each the conjugate of the other. Suppose r1 D m C
ni and r2 D m ni , with m and n real numbers and with n 0. (If
n D 0, the roots are real.) Calculate and show that it is a negative
real number.
(v) We wish to show that the converses of the results in (ii)-(iv) hold as
well; that is:
(a) If > 0, then r1 and r2 are real and distinct;
(b) if D 0, then r1 is a real number and r1 D r2 ;

(c) if < 0, then r1 and r2 are non-real, complex numbers, each the
conjugate of the other.

(vi) Assume that > 0. We are to prove that r1 and r2 are real and distinct. Suppose this is not the case and deduce from the first part of the
exercise that either r1 and r2 coincide and are real or they are non-real
complex conjugates. Use other parts of the proof to obtain the contradiction that  0. Conclude, as desired, that r1 and r2 are real and
distinct.
(vii) Make similar arguments to prove the other two converses.
(viii) Conclude that the sign of determines the nature of the roots of x 2 C
bx C c.
Both b 2 4c, which we write as , and .r1 r2 /2 , which we write as
, determine the nature of the roots of x 2 C bx C c. Lets compare the two
quantities.
Exercise 4.8. With the notation of Exercise 4.7, recall that x 2 C bx C c
factors as .x r1 /.x r2/. Use this to express b and c in terms of r1 and
r2 and deduce that .r1 r2 /2 D b 2 4c, so that and coincide.

i
i

i
IrvingBook 2013/5/22 15:39 page 80 #96

80

4. Complex Numbers

We see that the discriminant of x 2 C bx C c has two descriptions. It is

the square of the difference of the roots and it is the quantity b 2 4c. We are
free to take either of these as the definition of the discriminant. However,
the more fundamental quantity is .r1 r2 /2 . One reason is that it can be
generalized for higher-degree polynomials. Hence, it is the one we adopt as
the defining expression for a quadratic polynomials discriminant.
The fact that .r1 r2 /2 equals b 2 4c can then be interpreted as the
statement that we can calculate the discriminant of x 2 C bx C c, defined as
the square of the difference of the roots, in terms of the coefficients b and
c, without knowledge of the roots. We record this in the next theorem.
Theorem 4.4. Let b and c be real numbers and let r1 and r2 be the roots
(real or complex) of the quadratic polynomial x 2 C bx C c, so that x 2 C
bx C c factors as .x r1 /.x r2 /. Let be the discriminant of x 2 C bx C c,
which by definition is .r1 r2 /2 .
(i) determines the nature of the roots of x 2 C bx C c: If > 0, then
the roots are real and distinct; if D 0, then there is one root, real of
multiplicity 2; if < 0, then the roots are a pair of non-real complex
conjugates.
(ii) can be calculated in terms of the coefficients:
D b2

4c:

Complex numbers have allowed us to obtain the refinement Theorem

4.3 of Theorem 2.2. Similarly, we can refine Theorem 3.2 on the roots of
cubic polynomials.
Theorem 4.5. Let f .x/ be a monic, cubic polynomial. Exactly one of (i)
(iv) occurs.
(i) f .x/ has one real root a, of multiplicity 3, and factors as
.x

a/3 :

(ii) f .x/ has two distinct real roots a1 and a2 , of multiplicities 1 and 2,
and factors as
.x a1 /.x a2 /2 :
(iii) f .x/ has three distinct simple real roots a1 , a2 , and a3 , and factors as
.x

a1 /.x

a2 /.x

a3 /:

i
i

i
IrvingBook 2013/5/22 15:39 page 81 #97

81

4.3. Square and Cube Roots

(iv) f .x/ has one real root a and two distinct non-real, complex roots r
and r, and factors as
.x
Exercise 4.9.

a/.x

r /.x

r /:

Prove Theorem 4.5.

Exercise 4.10. Suppose x 3 C px C q is a cubic polynomial. We saw in

Theorem 3.1 that it must have a real root. Suppose you have found such a
root, r , by using Cardanos formula or other methods. Explain how you can
find the other two roots, whether real or complex.

4.3 Square and Cube Roots

At the end of Section 2.1, we saw that the quadratic formula reduces the
problem of solving a quadratic equation to that of calculating square roots
of real numbers. At the beginning of Section 3.2, we hoped that solving
cubic equations could be reduced, similarly, to the problem of calculating
square and cube roots of real numbers. Our initial experience with Cardanos formula suggested that we might need to find cube roots of non-real
complex numbers as well. Lets take a look at the problem of calculating
square and cube roots of complex numbers.
We begin with square roots. How do we find square roots of a complex
number a C bi ? There are two issues here. Does a C bi have another complex number as its square root? If so, can we calculate it? Regarding the
second question, we already have noted that algebraic techniques alone do
not allow us to compute square roots of positive real numbers. We must use
techniques of approximation. We cannot expect the calculation of square
roots of complex numbers to be any easier. We should be content if we can
reduce the problem of calculating square roots of complex numbers to the
problem of calculating square roots of positive real numbers.
Fix real numbers a and b; they should be regarded as known constants,
not as variables. Suppose we wish to find square roots of a C bi . If b D 0,
then finding the square roots of a C bi reduces to finding the square roots
of the real number a. This is trivial if a D 0. If a 0, it reduces to the
problem of finding the square root of a if a is positive or the square root
of a if a is negative (in which case the square roots of a are i times the
square roots of a). Thus, we can assume that b 0. Lets deal with this
case in an exercise.

i
i

i
IrvingBook 2013/5/22 15:39 page 82 #98

82

4. Complex Numbers

Exercise 4.11. Let a and b be real numbers with b 0. We wish to find

a complex number r C si such that .r C si /2 D a C bi . We can introduce
variables x and y for the unknown real numbers r and s, so that finding a
square root amounts to solving the equation
.x C yi /2 D a C bi
for real values of the unknowns x and y. We view a and b as known, fixed
constants.
(i) Expand the equation .x C yi /2 D a C bi and obtain two simultaneous
equations for x and y with real coefficients.
(ii) Using one of the equations and the fact that b 0, show that if these
equations have solutions, x and y will be non-zero.
(iii) Using this, divide both sides of one equation by x to get an equation
expressing y in terms of x and substitute this into the other equation
to get a single equation in x.
(iv) To find a square root of a C bi , we need to solve a single equation in x
involving only real numbers. Clear denominators in this equation and
obtain a degree 4 or quartic equation in x.
(v) It can be regarded as a degree 2 or quadratic equation in x 2. Use the
quadratic formula to obtain two values for x 2 in terms of a and b.
(vi) We are looking for a real value of x that solves the equation. If one of
our expressions for x 2 is a positive real number, then its two square
roots are our desired values for x. Verify that one of the expressions
for x 2 is positive.
(vii) Take square roots to obtain two values of x. Using an equation you
obtained earlier in the problem, obtain the two corresponding values
for y.
(viii) Write the formulas for the two square roots of a C bi . Notice that each
is 1 times the other.
We have just shown that any non-zero complex number has two complex numbers as its square roots, each the opposite of the other. We will
record part of this.
Theorem 4.6. Given a complex number s, there exists a complex number
r satisfying r 2 D s; in other words, every complex number has a complex
square root.

i
i

i
IrvingBook 2013/5/22 15:39 page 83 #99

83

4.3. Square and Cube Roots

The existence of square roots allows us to extend the quadratic formula

to quadratic polynomials with complex coefficients. We wont need the result for our analysis of cubic and quartic polynomials, but we will use it
in the final section, Section 7.5. In anticipation of that, lets work out the
details here.
Exercise 4.12. Let b and c be complex numbers.
(i) Rework Exercise 2.7 to show that the roots of x 2 C bx C c are
p
b
b 2 4c
xD

:
2
2
(ii) Use Theorem 4.6 to deduce that x 2 C bx C c has a root in the complex
numbers, and so factors as a product of two linear polynomials.
(iii) Conclude that Theorem 4.7 holds.

Theorem 4.7. Let b and c be complex numbers. There exist complex numbers r1 and r2 (possibly coincident) such that
x 2 C bx C c D .x

r1 /.x

r2 /:

In Exercise 4.11, we have proved that complex square roots exist and
shown how to determine them explicitly in terms of square roots of positive real numbers. We wish to be able to compute cube roots of complex
numbers as well, since this is essential in using Cardanos formula. Our experience in Exercise 4.11 suggests that we should be able to calculate cube
roots of complex numbers in terms of cube roots of real numbers. Lets
mimic what we did for square roots and see how far we can get.
Exercise 4.13. Fix real numbers a and b; they should be regarded as constants, not as variables. Assume b 0. We wish to determine the cube roots
of the non-real complex number a C bi ; that is, we wish to find a complex
number m C ni such that .m C ni /3 D a C bi . We can introduce variables
x and y for the unknown real numbers m and n, so that finding a cube root
amounts to solving the equation
.x C yi /3 D a C bi
for real values of x and y.

i
i

i
IrvingBook 2013/5/22 15:39 page 84 #100

84

4. Complex Numbers

Expand the equation .x C yi /3 D a C bi and obtain two simultaneous

equations for x and y with real coefficients. Conclude that the problem of
finding a complex cube root of aCbi is equivalent to the problem of finding
real solutions to the two equations.
In Exercise 3.13, we solved the equations that arise in Exercise 4.13
by trial and error for certain choices of a and b. Now we want a general
solution, but there is no obvious way to proceed. Lets experiment.
Exercise 4.14.
ercise 4.13.

Begin with the two equations for x and y obtained in Ex-

(i) Using one of the equations and the assumption that b 0, show that
if these equations have solutions, then y will be non-zero.
(ii) Introduce a new variable s and set x D sy, so that s D x=y. This
makes sense, since we know that y cant be 0. Substitute sy for x in
the two equations for x and y to get two equations in s and y. Write
them so that each is an expression in s times y 3 . Using the assumption
that b 0, show that 3s 2 1 0.

(iii) Solve one equation for y 3 , substitute in the other, and obtain a cubic
equation in s with coefficients involving a, b, and 3.
We have reduced the problem of finding a cube root of a C bi to the
problem of solving a cubic equation in s with real coefficients. If we can
solve the cubic equation for s, then we can determine y 3 , take cube roots,
find a value for y, and then a value for x. This is analogous to what happened in Exercise 4.11, where we reduced the problem of finding square
roots of a C bi to the problem of solving a quadratic equation with real coefficients. We could solve that equation using square roots of real numbers.
In our current situation, however, it is not clear how to proceed. Indeed,
we seem to have made a circle. We wish to compute cube roots of complex
numbers in order to solve cubic equations involving real numbers. Now we
find that computing cube roots of complex numbers leads us to the problem
of solving cubic equations with real number coefficients, and this may lead
us back to the calculation of the cube root of a complex number. We will
return to this issue in Section 5.5.

4.4

The Complex Plane

We discovered at the end of Section 4.3 that an algebraic approach to calculating cube roots of complex numbers does not seem to work, since it leads

i
i

i
IrvingBook 2013/5/22 15:39 page 85 #101

85

4.4. The Complex Plane

to the problem of finding roots of a cubic polynomial with real coefficients,

which leads back to the problem of computing cube roots of complex numbers. We need a new idea.
There is such an idea, one that will allow us to calculate not just cube
roots but nth roots of complex numbers for any positive integer n, provided
we supplement algebra with trigonometry. The idea depends on a geometric
interpretation of complex numbers, which we develop in this section.
A complex number a C bi is encoded by the two real numbers a and b,
its real and imaginary parts. This allows us to identify a C bi with the pair
.a; b/. But we can also regard .a; b/ as the cartesian coordinates of a point
on the plane: a, as usual, is the x-coordinate, specifying how far the point is
to the left or right of the y-axis, and b is the y-coordinate, specifying how
far the point is above or
pexample, we can identify the
pbelow the x-axis. For
complex number 1 C 3i with the pair .1; 3/, which we think of as the
coordinates of the point on the plane depicted in Figure 4.1.

2
p/3
(0, 0)

(1, 0)
p
Figure 4.1. The Point 1 C 3i in the Plane

Points in the plane can also be specified by their polar coordinates. The
polar coordinates of a point p in the plane are numbers r and , with r representing the distance from p to the origin of the plane and  representing
the angle formed by the positive ray of the x-axis and the line connecting p
to the origin. The angle  is measured in radians, proceeding counterclockwise from the x-axis to the line with p on it. Negative angles indicate that
we proceed in a clockwise direction from the x-axis.
Lets write a pair of polar coordinates as r; , using the brackets to
indicate that the pair ofpnumbers is not the usual cartesian coordinates. For
example, the point .1; 3/ (in cartesian coordinates) can be written in polar
coordinates as 2; =3. See Figure 4.1.
Polar coordinates have some disadvantages. The principal one is that a
point p is described by infinitely many different pairs of polar coordinates.
If we move around the plane along a circle, tracing an angle of 2 , we return
to the point at which we started. Thus, in addition to r; , the point p also

i
i

i
IrvingBook 2013/5/22 15:39 page 86 #102

86

4. Complex Numbers

has polar coordinates r;  C 2 . One more trip around the circle and we
return to the same point, but this time with polar coordinates r;  C 4 .
More generally, for any integer n, the point has polar coordinates r;  C
2n . The origin can be described in still more ways, since it has polar
coordinates 0;  for arbitrary .
Polar coordinates have advantages as well (or else we would not introduce them). One advantage is that certain figures in the plane can be described as the graphs of equations that take on an especially simple form in
polar coordinates. For example, r D 1 is the equation in polar coordinates
of a circle of radius 1 with the origin as its center.
In cartesian coordinates, the circle is described by the equation x 2 C
2
y D 1, and we can use it to define the cosine and sine functions for arbitrary real numbers . Given a point p on the circle such that the radius
from the origin to p forms an angle  with the x-axis (measured counterclockwise from the x-axis to the radius), the x and y coordinates of p are
the cosine and sine of :
p D .cos ; sin /:
This is the key to relating cartesian and polar coordinates.
Exercise 4.15. Let r be a non-negative real number.
(i) The equation of the circle with radius r and center the origin is x 2 C
y 2 D r 2 . Check that a point q on the circle for which the radius from
the origin to q forms an angle  with the x-axis (measured counterclockwise from the x-axis) has cartesian coordinates
.r cos ; r sin /:
(ii) Conclude that a point with cartesian coordinates .a; b/ and polar coordinates r;  satisfies
a D r cos I b D r sin :
These equations show how to recover the cartesian coordinates of a
point from its polar coordinates.
To describe the polar coordinates of a point in terms of its cartesian
coordinates, we use inverse trigonometric functions. Any one will do. Lets
choose the inverse cosine function, arccos t.
Given a real number t between 1 and 1, there are infinitely many angles  for which cos  D t. Thus, there are infinitely many possibilities for

i
i

i
IrvingBook 2013/5/22 15:39 page 87 #103

87

4.4. The Complex Plane

arccos t. In defining arccos t, by convention we regard  as restricted to the

interval 0;  . Within this interval, as  increases from 0 to  , its cosine
will decrease from 1 to 1, assuming all possible values in this range exactly once. The inverse function arccos t is then defined, for t in 1; 1, as
the unique angle  between 0 and  for which cos  D t. It will decrease
from  to 0 as t increases from 1 to 1. It is positive for t < 1 and zero for
t D 1.
Exercise 4.16. Let a and b be real numbers. We wish to find polar coordinates r;  for the point described in cartesian coordinates by .a; b/.
(i) If .a; b/ has polar coordinates r; , explain why .a; b/ has polar
coordinates r; .
(ii) Describe r in terms of a, b, and square roots.
(iii) Assume b  0. Describe  in terms of a, b, and the inverse cosine
function.
(iv) What is  if b < 0?
Exercise 4.17. Let a and b be real numbers.
(i) Using the identification of the complex number a C bi with the point
.a; b/ in the plane, show that a non-zero complex number a C bi can
be written as
r cos  C .r sin /i
for some non-negative real number r and some real number .
(ii) Write the following complex numbers in this form:
(a)
(b)
(c)
(d)

1 and 1.
i and i .
The four complex numbers 1 i .
p
The four complex numbers 1 3i . (See Figure 4.1.)

We have found that any complex number a C bi can be written as

r cos  C .r sin /i
for real numbers r and , with r and  chosen to be polar coordinates of the
point with cartesian coordinates .a; b/. Exercise 4.16 shows how to express
r and  explicitly in terms of a and b. We call r the length or magnitude of
the complex number a C bi and  the angle of inclination of a C bi , since it

i
i

i
IrvingBook 2013/5/22 15:39 page 88 #104

88

4. Complex Numbers

represents the angle at which the line through .a; b/ and .0; 0/ inclines with
respect to the x-axis. Often,  is called the argument of a C bi .
Recall that in Exercise 4.1 we proved that real numbers commute with
i under multiplication. Thus, the complex number a C bi is the same as
a C i b. Sometimes, when working with complicated expressions for the
imaginary part, the a C i b form is preferable, in that it eliminates the need
for extra parentheses. Thus, for example, we might prefer r cos  C i r sin 
to r cos  C .r sin /i . Sometimes its convenient to write r cos  C r i sin .
For example, we might write 2 cos.=3/ C 2i sin.=3/ rather than putting
the i before the 2 or after the sine term.

4.5

A Geometric Interpretation of
Multiplication

Having identified complex numbers with points on the plane, we will see
in this section how to interpret multiplication of complex numbers geometrically. Lets first review how to interpret multiplication of real numbers
geometrically. We identify real numbers with points on the line in the usual
way and wish to understand what multiplication by a non-zero real number
r does to the line. Begin with three special cases.
If r D 2, multiplication of a real number a on the line by 2 yields
the number 2a, which is twice as far from 0 as a is. Thus, multiplication
by 2 expands the line by a factor of 2. If r D 1=2, multiplication of a
by 1=2 yields the number a=2, which is half as far from 0 as a is. Thus,
multiplication by 1=2 contracts the line by a factor of 2. If r D 1,
multiplication of a by 1 yields the number a, which is the mirror image
of a on the opposite side of the line from the origin, 0. Thus, multiplication
by 1 reflects the line across 0.
In general, given a positive real number r , we can interpret multiplication by r as an expansion of the real number line if r > 1, as a contraction
if r < 1, and as leaving the line unchanged if r D 1. Multiplication by r
performs the same scale change on the line as well as reflecting it across 0.
Multiplication of complex numbers by a non-zero real number has a
similar geometric interpretation, but with the real line replaced by the plane.
We identify a complex number a C bi with the point .a; b/ on the plane.
Given a positive real number r , multiplication of the complex number aCbi
by r yields ra C r bi , which lies on the line through 0 and a C bi in the
plane but is r times as far from the origin as a C bi . Thus, multiplication by
r is an expansion of the plane if r > 1, a contraction if r < 1, and a fixing
of the points on the plane if r D 1.

i
i

i
IrvingBook 2013/5/22 15:39 page 89 #105

89

4.5. A Geometric Interpretation of Multiplication

Lets see what multiplication by 1 does to the plane. Suppose s;  are

polar coordinates for the point .a; b/, so that
a C bi D s cos  C si sin :
Using the trigonometric identities cos. C  / D
sin , we obtain

cos  and sin. C  / D

.a C bi / D s cos. C  / C si sin. C  /:
Thus, multiplication by 1 leaves the length of each point in the plane unchanged and adds  to its angle of inclination. This results in a rotation of
the plane around the origin by  , or 180. Given a positive real number r ,
it follows that multiplication by r is an expansion or contraction of the
plane followed by a 180 rotation.
We wish to extend this analysis to obtain a geometric description of
multiplication of complex numbers by an arbitrary complex number. The
essential case to handle is that of multiplication by i , the most basic of
non-real complex numbers. First we check that multiplication of complex
numbers satisfies the associative law.
Exercise 4.18. Verify that multiplication of complex numbers is associative; that is, given complex numbers p, q, and r , verify that
.pq/r D p.qr /:
To do this, introduce real numbers a, b, c, d , e, and f with p D a C bi ,
q D cCd i , and r D eCf i , then carry out the multiplications and compare,
using the associative law for multiplication of real numbers. (The fact that
multiplication of real numbers is associative requires proof, but we will take
it as a known property of the reals.)
Thanks to associativity, whatever action multiplication by i performs on
the plane, performing it twice must produce the same action as multiplication by i 2 . The action of i on a complex number p followed by a second
action of i yields i
.i
p/, whereas the action of i 2 on p is given by .i 2 /
p,
and they are equal. But i 2 D 1 and we just saw that multiplication by
1 is a 180 rotation. Thus, performing the action of i twice in succession
produces a 180 rotation. This suggests a candidate for the action of multiplication by i on the complex plane: rotation by 90 . There are two choices,
rotation clockwise and rotation counterclockwise. Thats good, since there
are two square roots of 1, the imaginary numbers i and i . We might

i
i

i
IrvingBook 2013/5/22 15:39 page 90 #106

90

4. Complex Numbers

imagine that one rotation corresponds to multiplication by i , the other to

multiplication by i . Lets sort this out in the next exercise.
Exercise 4.19. Lets identify a complex number a C bi with the point
.a; b/ on the plane and let s;  be its polar coordinates.
(i) Using polar coordinates and the trigonometric identities
cos. C =2/ D

sin ; sin. C =2/ D cos ;

verify that
i.a C bi / D i.s cos  C i s sin /
D s sin  C i s cos 

D s cos. C =2/ C i s sin. C =2/:

(ii) Conclude that multiplication by i rotates the complex plane counterclockwise 90.
(iii) Similarly, verify that
i.a C bi / D i.s cos  C i s sin /
D s sin 

D s cos.

i s cos 

=2/ C i s sin.

=2/

and conclude that multiplication by i rotates the complex plane clockwise by 90 (or counterclockwise by 270).
We introduced complex numbers as formal entities in order to deal with
the lack of a square root of 1, declaring i to be such a square root. Additional numbers were introduced so the ordinary rules of addition and multiplication hold, thereby bringing the complex numbers into existence. No
concrete meaning was attached to i . It was indeed imaginary. Now, however, having identified complex numbers with points on the plane and multiplication by non-zero real numbers with geometric actions on the plane,
we have found a meaning for i . No longer imaginary, it is revealed to be
the number that, when used to multiply other complex numbers, produces a
counterclockwise rotation by 90 .
To complete our picture, we next need to extend our interpretation of
multiplication by i (and i / to multiplication by any complex numbers of
length 1. Geometrically, these are the numbers that lie on the unit circle
centered at the origin. The angle of inclination of i is 90 and the angle

i
i

i
IrvingBook 2013/5/22 15:39 page 91 #107

4.5. A Geometric Interpretation of Multiplication

91

of inclination of i is 270 . We have seen that multiplication by i and i

produces rotations of exactly those angles. It is not difficult to guess the
general result.
Theorem 4.8. Let u be a complex number with length 1 and angle of inclination , so that
u D cos  C i sin :
Let v be a complex number with length s and angle of inclination , so that
v D s cos  C si sin :
Then the product uv has length s and angle of inclination  C :
uv D s cos. C / C si sin. C /:
Thus, multiplication by u effects a rotation of the complex plane counterclockwise through the angle .
Theorem 4.8 will follow from the standard trigonometric identities for
the sine and cosine of the sum of two angles. These are sufficiently important that we should record them.
Theorem 4.9. Let  and  be two real numbers. Then
and

sin. C / D sin  cos  C cos  sin 

cos. C / D cos  cos 

sin  sin :

We should also take a moment to review how Theorem 4.9 is proved, to

ensure that we avoid any circularity in our reasoning. Some books provide
a proof of Theorem 4.9 as a consequence of de Moivres formula, which
we will prove in Section 4.6 as a consequence of Theorem 4.9. This is the
circularity we wish to avoid.
Fortunately, there are many other ways to prove Theorem 4.9. See Chapter 6 of Eli Maors Trigonometric Delights [40, pp. 8794] for one based
on the classical result of plane geometry known as Ptolemys theorem.
This states, for a quadrilateral inscribed in a circle, that the product of the
lengths of the quadrilaterals two diagonals equals the sum of the products
of lengths of the two pairs of opposite sides. (Ptolemy is the great Alexandrian astronomer and mathematician of the second century C.E.)
Exercise 4.20. Use Theorem 4.9 to show that
.cos  C i sin /
.cos  C i sin / D cos. C / C i sin. C /:
Deduce that Theorem 4.8 holds.

i
i

i
IrvingBook 2013/5/22 15:39 page 92 #108

92

4. Complex Numbers

Lets do some calculations to illustrate Theorem 4.8.

Exercise 4.21. Compute the product of each pair of numbers:
(i) cos.=2/ C i sin.=2/ and cos.3=2/ C i sin.3=2/.
(ii) cos.=4/ C i sin.=4/ and cos.=4/ C i sin.=4/.
(iii) cos.=6/ C i sin.=6/ and cos.=3/ C i sin.=3/.
(iv) cos.=18/ C i sin.=18/ and cos.=5/ C i sin.=5/.
For the first three examples, rewrite the numbers in more familiar terms,
using the explicit values of the trigonometric functions.
We conclude the sequence of ideas in this section with a description of
the product of two complex numbers.
Exercise 4.22. Suppose we are given a complex number u with length r
and angle of inclination  and a complex number v with length s and angle
of inclination . Thus
u D r cos  C r i sin 

and
Show that

v D s cos  C si sin :

.r cos  C r i sin /
.s cos  C si sin / D r s cos. C / C r s i sin. C /:

Reinterpret this to say that the product uv is the complex number whose
length is the product r s of the lengths r and s of u and v and whose angle
of inclination  C  is the sum of the angles of inclination  and  of u
and v.
We have found that multiplying two complex numbers amounts, geometrically, to multiplying their lengths and adding their angles of inclination.

4.6

Eulers and de Moivres Formulas

Our motivation in studying complex numbers is our desire to compute their

cube roots. The result that will allow us to do thisde Moivres formula
provides at the same time a procedure for computing nth roots for any positive integer n. (The formulas name honors Abraham de Moivre (1667
1754), a French mathematician who spent much of his life in England.)

i
i

i
IrvingBook 2013/5/22 15:39 page 93 #109

93

4.6. Eulers and de Moivres Formulas

Thus, we may as well consider the general case. However, we will lay out
the issues in such a way that the reader can ignore the general case and
focus on the cases n D 2 and n D 3.
One way to derive de Moivres formula is to use trigonometry and Theorem 4.9. We will take this approach in a moment. But first lets see how easily it follows from Eulers formula, a result due to Leonhard Euler (1707
1783), the greatest mathematician of the eighteenth century.
Eulers formula involves the number e, whose central role in mathematics Euler first discovered and whose notation he introduced. Readers who
have studied calculus will be familiar with it as the number most naturally
used for exponentiation. What makes it so natural is that the exponential
function e x with base e is its own derivative, in contrast to other exponential functions such as 2x and 10x . Another attractive feature is that its
inverse function loge x is the antiderivative of 1=x. For readers unfamiliar
with e, suffice to say that it is an irrational number with decimal expansion
that begins 2:71828 and that for theoretical reasons it is the best base for
studying exponentiation.
Eulers formula relates exponential and trigonometric functions:
Theorem 4.10 (Eulers formula). Let  be a real number. Then
e i D cos  C i sin :
At first glance, the formula may appear both mysterious and wondrous.
For those readers familiar with power series expansions of functions, the
mystery is easily addressed, but the wonder remains.
In a calculus course, it is shown that the sine and cosine functions have
power series expansions
cos x D 1
and

x4
x2
C
2
4

x3
x5
C
3
5
and the exponential function satisfies
sin x D x

ex D 1 C

x6
C



6
x7
C


;
7

x
x2
x3
x4
x5
C
C
C
C
C


:
1
2
3
4
5

The expansions are valid for any real number x.

Once complex numbers are introduced, it is not a large leap to replace
real numbers in the power series expansion of e x with pure imaginary numbers. Suppose for instance that  is a real number and let x D i. If we

i
i

i
IrvingBook 2013/5/22 15:39 page 94 #110

94

4. Complex Numbers

substitute i for x in the power series expansion of e x and use the equalities i 2 D 1, i 3 D i , and i 4 D 1, we find that
e i D 1 C i

2
2

3
4
5
C
Ci
C


:
3
4
5

Collecting real and imaginary terms on the right side of this equation, we
obtain cos  C i sin , yielding Eulers formula. Thats all there is to it.
Given a real number  and a positive integer n, the rules of exponentiation yield
.e i /n D e i n :
Reinterpreting the left and right sides of this equation via Eulers formula,
we see that
.cos  C i sin /n D cos n C i sin n:
This is de Moivres formula:
Theorem 4.11 (de Moivres formula). Let n be a positive integer and  a
real number. Then
.cos  C i sin /n D cos n C i sin n:
In words, de Moivres formula says that the nth power of a complex
number of length 1 is the number obtained by multiplying the original numbers angle of inclination by n. In particular, squaring amounts to doubling
the angle and cubing amounts to tripling the angle.
We were led to de Moivres formula from Eulers formula, but the proof
of Eulers formula requires ideas from calculus. An alternative, more elementary proof of de Moivres formula is desirable. As de Moivre understood, the underlying issue is one of trigonometry. The proof sketched in
Exercise 4.23 uses the classical angle sum formulas of Theorem 4.9 (and an
implicit induction argument). The cases n D 2 and n D 3, which are all we
need, are treated separately before we handle the general case.
Exercise 4.23. Prove de Moivres formula.
(i) The formula holds trivially for n D 1.
(ii) Square cos  Ci sin , combine real and imaginary terms, and use Theorem 4.9 to show that
.cos  C r i sin /2 D cos 2 C i sin 2:

i
i

i
IrvingBook 2013/5/22 15:39 page 95 #111

4.6. Eulers and de Moivres Formulas

95

(iii) Multiply both sides of the last equality by cos  C i sin  and use Theorem 4.9 to show that
.cos  C r i sin /3 D cos 3 C i sin 3:
(iv) We can continue this process step by step to obtain the general result.
Suppose k is a positive integer less than n and we know that
.cos  C i sin /k D cos k C i sin k:
Multiply both sides of this equality by cos  C i sin  and use Theorem
4.9 to deduce that
.cos  C i sin /kC1 D cos.k C 1/ C i sin.k C 1/:
(v) Deduce that we can make this argument n 1 times successively to
obtain the formula for k D 2, then k D 3, then k D 4, and so on, until
we obtain the formula for k D n.

Exercise 4.24. Lets do some computations using de Moivres formula.

(i) Compute
.cos.=4/ C i sin.=4//n
for n D 2, 3, 4, and 8, thus finding
p
p !n
2
2
C
i
2
2
for these values of n.
(ii) Compute

p !n
3
1
C
i
2
2

for n D 2, 3, and 6.
Exercise 4.25. As a minor extension of de Moivres formula, prove Theorem 4.12, which describes the nth power of a non-zero complex number.
Theorem 4.12. Let n be a positive integer, r a positive real number, and 
a real number. Then
.r cos  C r i sin /n D r n cos n C r n i sin n:

i
i

i
IrvingBook 2013/5/22 15:39 page 96 #112

96

4. Complex Numbers

De Moivres formula and Theorem 4.12 allow us to compute nth powers

of complex numbers instantly, once the numbers are expressed in the form
r cos  C r i sin . They also reveal how to calculate nth roots: we do the
reverse, calculating the real nth root of the length of a complex number and
dividing the angle of inclination by n.
The procedure has the drawback that it is not purely algebraic. It requires trigonometry to determine the angle of inclination of a given complex number and trigonometry again, once nth roots have been taken, to
calculate the real and imaginary parts.
Lets work through the details, dealing first with square roots and cube
roots. We already found how to compute the two square roots of a nonzero complex number in Exercise 4.11, using algebra, but now well use the
trigonometric approach.
Exercise 4.26. Suppose we wish to compute the square root of a non-zero
complex number c. Write c as
r cos  C r i sin 
for some positive real number r and some real . Write
real square root of r .

p
r for the positive

(i) Using Theorem 4.12, verify that the two complex numbers
p
 p

r cos C r i sin
2
2
and

p
p


r cos. C  / C r i sin. C  /
2
2
are square roots of c.
(ii) Check that each is the opposite of the other, as we would expect.
(iii) Use the formula to compute square roots of
(a) 4 and 4.
(b) i and i .
(c) 1 C i and 1 C i .
Next lets find cube roots.
Exercise 4.27. Suppose c is a non-zero complex number. Write it as
r cos  C r i sin 

i
i

i
IrvingBook 2013/5/22 15:39 page 97 #113

97

4.6. Eulers and de Moivres Formulas

for some positive real number r and some real .

(i) Recall that r has a unique real cube root. Write it as

p
3
r.

(ii) Using Theorem 4.12, show that

p
p


3
r cos C 3 r i sin
3
3
is a cube root of c, as are
p
p


2
2
3
r cos. C
/ C 3 r i sin. C
/
3
3
3
3
and

p
p

4

4
3
r cos. C
/ C 3 r i sin. C
/:
3
3
3
3

(iii) As in Exercise 3.15, let

!D

1
C
2

p
2

Show, using the formulas, that the three cube roots of 1 are 1, !, and
! 2 . (Hint: First express 1 as cos 0 C i sin 0.)
(iv) Suppose d is one of the cube roots of c. Verify that the other two are
!d and ! 2 d .
(v) Suppose c is a non-zero real number, with real cube root d . Conclude
that c has one real cube root, d , and two non-real cube roots, !d and
!2 d .
We can use Exercise 4.27 to redo some cube root calculations that we
made earlier when we used Cardanos formula, as illustrated next.
Exercise 4.28. In Exercise 3.13, we found a cube root of
3C

10 p
3
9

10 p
3:
9

and a cube root of

Lets now find all three cube roots.

(i) Use the method
p of Exercise 4.27 and a calculator to find a cube root
of 3 C 10
3=9. Then use Exercise 4.27 to find its other two cube
roots.

i
i

i
IrvingBook 2013/5/22 15:39 page 98 #114

98

4. Complex Numbers

(ii) In the same way, find the three cube roots of 3

10

3=9.

(iii) Alternatively, use Theorem 4.1 to find its three cube roots.
(iv) Compare the results to those arising from the calculation of cube roots
in Exercise 3.13.
We have focused so far on finding square and cube roots of a non-zero
complex number, but the same approach allows us to find nth roots for any
positive integer n. As a consequence of Theorem 1.12, any positive real
p
number has a unique positive real nth root, which we write n r. Here is the
general result:
Theorem 4.13. Let r be a positive real number. The complex number
r cos  C r i sin 
has n distinct nth roots, given by




p
p
2j
2j


n
r cos
C
C n r i sin
C
n
n
n
n
as j D 0; 1; : : : ; n

1.

Exercise 4.29. Prove Theorem 4.13:

(i) Verify that the n quantities are distinct.
(ii) Verify that they are nth roots.

4.7

Roots of Unity

The only result from this section that we will use later is Theorem 4.16,
which can be proved by direct calculation. Thus, the reader may wish to
go straight to Theorem 4.16, and its proof, and move ahead. However, a
natural generalization of Theorem 4.16 is easy to obtain, and it enhances
our understanding of nth roots.
A complex number that has an nth power equal to 1, where n is a
positive integer, is called a root of unity. Thanks to Theorem 4.13 (or de
Moivres formula), we have identified them all.
Theorem 4.14. The n distinct nth roots of unity are




2j
2j
C i sin
;
cos
n
n
as j D 0; 1; : : : ; n

1.

i
i

i
IrvingBook 2013/5/22 15:39 page 99 #115

99

4.7. Roots of Unity

Exercise 4.30. Prove Theorem 4.14.

Exercise 4.31. For small n, we can write down the nth roots of unity as
ordinary complex numbers, without trigonometric functions in their expressions. We have already done so in several cases. Lets review and extend
what we know. Describe without cosines and sines the two square roots of
1, the three cube roots of 1, the four fourth roots of 1, the six sixth roots of
1, and the eight eighth roots of 1.
The two square roots of 1 are the powers of 1, the three cube roots of
1 are the powers of the number we have called !, and the four fourth roots
of 1 are the powers of i . For an arbitrary positive integer n, let n be the
number
2
2
cos
C i sin
:
n
n
Here,  is the Greek letter zeta. Notice that 2 is 1 and 3 is !, while 4 is
i.
We learn different parts of Theorem 4.15 early in our mathematical education. One might not think to state it as a theorem, but it provides the
template for the results that follow.
Theorem 4.15. Let 2 be the number cos.2=2/ C i sin.2=2/. For simplicity, write 2 in its more familiar form as 1.
(i) The two distinct square roots of 1 are 1 and 1, with . 1/2 D 1.
(ii) 1 C . 1/ D 0.
(iii) x 2

1 D .x

(iv) A2

B 2 D .A

1/.x

. 1//.

B/.A

. 1/B/.

Theorem 4.15 has an analogue for any integer n > 2, with 1 and 1
replaced by the nth roots of unity. We will need the appropriate analogue
for n D 3 in Section 5.3.
Theorem 4.16. Let 3 be the number cos.2=3/ C i sin.2=3/. For simplicity, write 3 in its more familiar form as !.
(i) The three distinct cube roots of 1 are 1, !, and ! 2 , with ! 3 D 1.

(ii) 1 C ! C ! 2 D 0.
(iii) x 3
(iv) A3

1 D .x

B 3 D .A

1/.x

!/.x

B/.A

! 2 /.

!B/.A

! 2 B/.

i
i

i
IrvingBook 2013/5/22 15:39 page 100 #116

100

4. Complex Numbers

Exercise 4.32. Prove Theorem 4.16. The first part has already been shown
in Exercise 3.15 (or Exercise 4.31). Verify the second part by direct calculation, and use it to prove the third and fourth parts.
The analogous result for n D 4 is even easier to prove.
Theorem 4.17. Let 4 be the number cos.2=4/ C i sin.2=4/. For simplicity, write 4 in its more familiar form as i .
(i) The four distinct fourth roots of 1 are 1, i , i 2 , and i 3 , with i 4 D 1.
(ii) 1 C i C i 2 C i 3 D 0.
(iii) x 4 1 D .x 1/.x i /.x i 2 /.x i 3 / D .x 1/.x C1/.x i /.x Ci /.
(iv) A4

B 4 D .A
D .A

B/.A iB/.A i 2 B/.A i 3 B/

B/.A C B/.A iB/.A C iB/.

Exercise 4.33. Prove Theorem 4.17.

Here is the theorem that generalizes Theorems 4.15, 4.16, and 4.17:
Theorem 4.18. Let n be a positive integer and let n be the number cos.2=n/C
i sin.2=n/.
(i) The n distinct nth roots of unity are 1; n ; n2 ; : : : ; nn
1.
(ii) 1 C n C n2 C


C nn
(iii) x n

1 D .x

(iv) An

B n D .A

1/.x

C nn

, with .n /n D

D 0.

n2 /


.x

n /.x

B/.A

nn

n2 B/


.A

n B/.A

/.
nn

B/.

Exercise 4.34. Prove Theorem 4.18.

(i) Use Theorem 4.14 and de Moivres formula to obtain the first part.
(ii) For the second part, verify that x n
xn

1 D .x

1/.x n

1 factors as

C xn

C


C x 2 C x C 1/:

Then substitute n for x and use the fact that n 1 to deduce the
desired result.

i
i

i
IrvingBook 2013/5/22 15:39 page 101 #117

101

4.8. Converting Root Extraction to Division

(iii) The proof of Theorem 1.5 works just as well for complex roots of a
polynomial as for real roots. Deduce that Theorem 1.5 can be applied
to show that
.x

1/.x

n /.x

n2 /


.x

nn

divides x n 1. Compare degrees and deduce that x n 1 must be a

constant multiple of the product. Then compare the coefficients of x n
and deduce that equality holds.
(iv) Deduce the fourth part from the third using the substitution x D A=B.

4.8 Converting Root Extraction to Division

Throughout our discussion of the quadratic formula and Cardanos formula,
we have emphasized that the role of algebra is to reduce the problem of
solving quadratic and cubic equations to that of calculating square and cube
roots, but that algebra does not provide us with the means to make the root
calculations. Lets discuss how we can in principle make the calculations
(and more generally nth root calculations) through the use of exponential
and trigonometric functions and their inverses. We will once again assume
familiarity with the exponential function e x , where the base e is Eulers
number, which we met in Section 4.6. We will also use the inverse function
loge .x/, the natural logarithm function, which well write as log.x/.
Exercise 4.35. Let r be a positive real number and n a positive integer.
Suppose we wish to calculate the real nth root of r .
(i) Set c D r 1=n , the real nth root. Apply log.x/ to both sides to obtain
log.c/ D log.r /=n:
(ii) Exponentiate both sides to obtain
c D e log.c/ D e log.r /=n:
(iii) Rewrite this as
r 1=n D e log.r /=n:
(iv) Conclude that this reduces the problem of calculating nth roots of positive real numbers to that of computing certain values of the exponential
and logarithm functions and dividing by n.

i
i

i
IrvingBook 2013/5/22 15:39 page 102 #118

102

4. Complex Numbers

The lesson of Exercise 4.35 is that applying the logarithm converts extraction of an nth root of a real number to division by n.
Exercise 4.36. Let a and b be real numbers satisfying a2 C b 2 D 1 and let
n be a positive integer. Suppose we wish to calculate an nth root of a C bi .
For simplicity in the trigonometric considerations to follow, assume that
b  0. Then a C bi can be written as cos  C i sin  for an angle  given by
 D arccos a:
(i) Use Theorem 4.13 to deduce that one nth root of cos  C i sin  is
cos.=n/ C i sin.=n/:
(ii) By substitution, conclude that one nth root of a C bi is given by
cos..arccos a/=n/ C i sin..arccos a/=n/:
(iii) We may rewrite this in another way. The sine and cosine functions
satisfy
sin x D cos.x =2/:
Rewrite the expression for an nth root of a C bi as
cos..arccos a/=n/ C i cos..arccos a/=n

=2/:

(iv) Conclude that this reduces the problem of calculating nth roots of complex numbers of length 1 to that of computing certain values of the
inverse cosine and cosine functions and dividing by n.
A non-zero complex number aCbi can be written in the form r .cos  C
i sin / for a positive real number r and a real number . Exercises 4.35 and
4.36 give us procedures for computing the nth roots of r and cos  C i sin ,
allowing us to compute an nth root of a C bi .
The calculational methods of the last two exercises have a common feature. In each nth root calculation, we make use of a familiar function f .x/
and its inverse f 1 .x/, applying f 1 to a real number c, dividing what
we obtain by n, then calculating f on the result. Both times, this process
converts nth root extraction to division by n. The cost of this simplification
is the requirement that we have the means to compute f 1 and f on given
real numbers.

i
i

i
IrvingBook 2013/5/22 15:39 page 103 #119

103

4.9. History

4.9 History
Complex numbers are so widely used that it can be difficult to understand
why their acceptance took centuries, stretching from their appearance in
Cardanos Ars Magna [12] in 1545 to the development of the subject now
called complex analysis in the late nineteenth century. Two entertaining
accounts of the intellectual challenge complex
p numbers posed are Paul J.
Nahins An Imaginary Tale: The Story of
1 [43] and Barry Mazurs
Imagining Numbers (Particularly the Square Root of Minus Fifteen) [41].
They serve as natural complements to each other, with rich historical discussions that are especially worth reading. In this section, we will touch on
just a few historical highlights.
Lets begin with de Moivres formula. It does not appear that de Moivre
wrote the formula down explicitly, but he surely knew it. Or at least he
understood that taking nth roots of complex numbers produces formulas
similar to trigonometric formulas for sines and cosines, allowing him to
relate taking nth roots with dividing angles by n. For instance, in a 1707
note with the title The analytic solution of certain equations of the third,
fifth, seventh, ninth and other higher uneven powers, by rules similar to
those called Cardans, he writes [16], [60, pp. 443444]:
If the equation were
5y
yD
1
2

s
5

61
C
64

20y 3 C 16y 5 D

375
1
C
4096
2

s
5

61
64

61
;
64
r

375
:
4096

And if by any means the fifth root of the binomial can be extracted the
root will come out true and possible, although the expression seems to
include an impossibility. Now the fifth root of the binomial
61
C
64
is

375
4096

1
1p
C
15;
4
4

and of the binomial

61
64

375
4096

i
i

i
IrvingBook 2013/5/22 15:39 page 104 #120

104

4. Complex Numbers

is

1 1p
15
4 4
whose semi-sum 1=4 D y. But if the extraction cannot be performed,
or should seem too difficult, the thing may always be effected by a
table of natural sines in the following manner.

To the radius 1 let a D 61=64 D 0:95112 be the sine of a certain

arc which therefore will be 72 230 , the fifth part of which (because
n D 5) is 14 280 ; the sine of this is 0:24981; near ly D 1=4. So also
for equations of higher degree.
p
De Moivre recognizes that .61=64/C i 375=4096 has the form sin  C
i cos  for a suitable , and adding the fifth roots of this number and its
conjugate amounts to computing the sine of =5. As Mazur comments ([41,
p. 199]), Although there were earlier hints of the link between the solution of polynomial equations and trigonometry, we know that Abraham De
Moivre, by 1707, had perceived the analogy between the geometric problem of cutting an arc of a circle into n equal parts and taking the nth root of
a complex number.
In 1748, Leonhard Euler stated de Moivres formula in its standard
form, deriving it using the angle-sum formulas for sine and cosine. Here
is the start of his discussion:
Since .sin z/2 C .cos z/2 D 1, on decomposing into factors we get
.cos z C i sin z/.cos z

i sin z/ D 1:

These factors, although imaginary, are of great use in the combination

and multiplication of arcs. For example, let us seek the product of these
factors.
Euler then continues in a familiar way to find the square, the cube, and the
nth power of cos z i sin z.
Euler made many contributions to the understanding of complex numbers, including his famous formula relating the exponential, cosine, and
sine functions. Beyond the results themselves, he played an essential role
in bringing about the acceptance of their use. For example, in his 1770 text
Elements of Algebra [26], Euler introduces complex numbers early, in Chapter XIII of Section 1, where he writes [26, pp. 4243]:
When it is required, therefore, to extract the root of a negative number,
a great difficulty arises; since there is no assignable number, the square
of which would be a negative quantity. : : : it is evident that we cannot

i
i

i
IrvingBook 2013/5/22 15:39 page 105 #121

4.9. History

105

rank the square root of a negative number amongst the possible numbers, and we must therefore say that it is an impossible quantity. In this
manner we are led to the idea of numbers, which from their nature are
impossible; and therefore they are usually called imaginary quantities,
because they exist merely in the imagination. : : : But notwithstanding
this, these numbers present themselves to the mind; they exist in our
imagination, and we still have a sufficient idea of them; : : : for this
reason also, nothing prevents us from making use of these imaginary
numbers, and employing them in calculation.
Elements of Algebra would become an influential text. A translation
from German to French was prepared by Johann III Bernoulli, with about
one hundred pages of additions by Joseph-Louis Lagrange. An English
translation of the French edition was begun by Francis Horner, who also
wrote a short biography of Euler, and completed by Reverend John Hewlett.
The 1840 edition of the English translation containing Bernoullis notes,
Lagranges additions, and Horners biography remains available thanks to
Springer-Verlag.
A good account of Eulers work on complex numbers can be found in
Chapter 5 of William Dunhams Euler: The Master of Us All [23]. Dunham
concludes it with the observation [23, pp. 101102] that complex numbers
were here to stay. A concept only dimly understood for its role in solving
cubic equations had been legitimized by the discoveries and influence of
Leonhard Euler. Without apology or embarrassment, he treated these numbers as equal players on the mathematical stage and showed how to take
their roots, logs, sines, and cosines.
As the eighteenth century gave way to the nineteenth, the geometric
description of complex numbers presented in Sections 4.4 and 4.5 first appeared. How it did is a fascinating story, well told in the books of Nahin
[43] and Mazur [41]. We will provide a brief sketch. But first we should
note that an earlier effort to describe pure imaginary numbers as points on
the plane that lie off the real number line appears in a 1685 work on algebra
[69] by the English mathematician John Wallis (16161703). His approach
was inconclusive. We will return to some of the material in Walliss book in
Section 5.8.
The first person to describe complex numbers geometrically in the way
we now understand was the Norwegian Caspar Wessel (17451818). He
presented a paper to the Royal Danish Academy of Sciences in 1797 that
was published in Danish two years later in the Academys Memoires [70].
Wessel recognized that in multiplying two complex numbers, one adds their
angles of inclination. It appears, however, that Wessels work went unno-

i
i

i
IrvingBook 2013/5/22 15:39 page 106 #122

106

4. Complex Numbers

ticed for nearly a century, until it was rediscovered in 1895 [43, pp. 4849,
243].
In 1806, the Swiss-born Jean-Robert Argand (17681822) published a
pamphlet, Essay on the Geometrical Interpretation of Imaginary Quantities, in which he also identified complex numbers with points on the plane.
(We will quote from A.S. Hardys 1881 translation [4].) Before tackling
complex numbers, Argand reviews the one-time challenge of making sense
of negative numbers [4, pp. 1722], concluding that
the difficulty of the subject will not be questioned if we remember
that the exact sciences had been cultivated for many centuries, and had
made great progress before either a true conception of negative quantities was reached or a general method for their use had been devised.
The notion of a negative number, Argand explains, might seem imaginary,
but when we compare two quantities, we consider not just the ratio of their
absolute values, but also a relation of direction, or of the sense in which
they are estimated, a relation either of identity or opposition.
Next Argand poses the problem [4, p. 23] of finding the geometric
mean between two quantities of different signs, that is, to find the value of
x in the proportion C1 W x WW x W 1. In effect, he asks for a a value of x
satisfying 1=x D x=. 1/, or x 2 D 1.
Here we encounter a difficulty . . . ; but, as before, the quantity which
was imaginary, when applied to certain magnitudes, became real when
to the idea of absolute number we added that of direction, may it not
be possible to treat this quantity, which is regarded imaginary, because
we cannot assign it a place in the scale of positive and negative quantities, with the same success? On reflection this has seemed possible,
provided we can devise a kind of quantity to which we may apply the
idea of direction, so that having chosen two opposite directions, one
for positive and one for negative values, there shall exist a third
such that the positive direction shall stand in the same relation to it
that the latter does to the negative.
Having opened the door to additional directions, Argand explains that
the quantity x which is to be the geometric mean of 1 and 1 should be
perpendicular to the line containing them. This yields two choices, which
are related to each other
p as C1 and
p 1. They are, therefore, what is ordinarily expressed by C
1 and
1. So begins Argands identification
of complex numbers with points on the plane.
Argands work was ignored initially. Guillaume-Jules Houel tells the
story in his preface to the 1876 reprint of Argands book [4, pp. iiixvi]. (See

i
i

i
IrvingBook 2013/5/22 15:39 page 107 #123

4.9. History

107

also [43, pp. 7374].) Argand was living in Paris at the time, and he sent a
copy to the great French mathematician Adrien-Marie Legendre. Without
mentioning Argands name, Legendre described Argands ideas in a letter
to Francois Francais, a professor of mathematics. On Francoiss death in
1810, his younger brother Jacques inherited his papers. Jacques published
an article in 1813 in the Annales de Mathematiques describing Argands
ideas as well as Legendres letter, and asked who the unknown author of the
pamphlet might be. Argand saw Francaiss article and identified himself,
receiving full credit from Francais.
Carl Friedrich Gauss (17771855), the greatest mathematician of the
era, had come up with the same ideas, perhaps even earlier than Wessel. He
chose not to publish his work until 1831, yet he came to share naming rights
with Argand. The plane, when its points are identified with complex numbers, has come to be called the Gaussian plane, Argand plane, or Argand
diagram.

i
i

i
IrvingBook 2013/5/22 15:39 page 108 #124

i
i

i
IrvingBook 2013/5/22 15:39 page 109 #125

5
Cubic Polynomials, II
We derived Cardanos formula for roots of reduced cubic polynomials in
Section 3.2, only to discover that using it may require us to compute cube
roots of complex numbers. This phenomenon arose in Exercise 3.12, when
we tried to solve the cubic equation y 3 7y C 6 D 0. Our study of the discriminant in Section 3.4 revealed that this difficulty will occur whenever we
work with a cubic whose roots are real and distinct. With that, we brought
Chapter 3 to a close and turned to a study of complex numbers in Chapter 4.
Now that we have learned how to use trigonometry to compute cube roots
of complex numbers, we are ready for a more systematic treatment of cubic
polynomials.

5.1 Cardanos formula

Given real numbers p and q, Cardanos formula for a root of the reduced
cubic polynomial y 3 C py C q takes the form
s
s
r
r
3
3
q
p3
q2
q
p3
q2
yD
C
C
C
C
:
2
27
4
2
27
4
Our derivation of the formula in Exercise 3.7 lacked precision, first because
we didnt yet know how to interpret the expressions within the cube root
signs if p 3 =27 C q 2 =4 < 0 and second because we didnt yet understand
that there are three possibilities for each cube root. We can proceed now
with confidence.
p
Let us once again write R for p 3 =27 C q 2=4 and ! for 1=2 C
3=2,
which is a cube root of 1. We learned in Exercise 4.27 how to compute cube
109

i
i

i
IrvingBook 2013/5/22 15:39 page 110 #126

110

5. Cubic Polynomials, II

p
roots of q=2 C R. Let A be one of them. Then
p the other two are !A and
2
R, then the other two are
! A. Similarly, if B is a cube root of q=2
!B and ! 2 B. In the next exercise, we determine which cube roots to add
together in order to obtain roots of y 3 C py C q.
Exercise 5.1. Let p and q be real numbers with p 0. (We can handle
the p D 0 case directly.) Since any non-zero number has three distinct
cube roots, each summand on the right side of Cardanos formula has three
possible values.
p
(i) Check,
as in Exercise 3.7, that the product of q=2 C R and q=2
p
R is p 3=27.
(ii) The derivation inpExercise 3.7 showed
that for the sum of the cube
p
roots of q=2 C R and q=2
R to be a root of y 3 C py C q, the
two cube roots must be chosen so thatptheir product is p=3. Select
one of the three cube roots of q=2 C Rany one, it doesnt matter
whichand call it A.
(iii) Although A is chosen arbitrarily,
we will exercise care in choosing a
p
R so that A C B is a root of y 3 C py C q.
cube root B of q=2
We have three choices for it. We will want to choose a specific one,
given the choice of A. For now, let us choose one arbitrarily and call it
B. The other two are !B and ! 2 B.
(iv) Check that A3 B3 D p 3 =27 and deduce that AB equals one of the
numbers p=3, !p=3, and ! 2 p=3.
(v) If AB D p=3, set B D B. If AB D !p=3, set B D ! 2 B; and if
AB D ! 2 p=3, set B D !B. Verify in all cases that AB D p=3.
(vi) Show, with A and B chosen
the numbers !A and ! 2 B
pin this way, that p
are cube roots of q=2 C R and q=2
R with the property that
2
their product equals p=3.pShow also that the
p numbers ! A and !B
are cube roots of q=2 C R and q=2
R with the property that
their product equals p=3.
p
(vii) Conclude
p that if A and B are chosen as cube roots of q=2 C3 R and
q=2
R satisfying AB D p=3, then the three roots of y CpyCq
have the form
r1 D A C BI r2 D !A C ! 2 BI r3 D ! 2 A C !B:

i
i

i
IrvingBook 2013/5/22 15:39 page 111 #127

111

5.1. Cardanos formula

Exercise 5.1 removes the imprecision that was present in our earlier
derivation of Cardanos formula. We can summarize what we have found in
a theorem.
Theorem 5.1 (Cardanos formula). Suppose p and q are real numbers, with
p 0. Let A be a cube root of
r
q
p3
q2
C
C
2
27
4
and let B be the unique cube root of
r
q
p3
q2
C
2
27
4
satisfying AB D p=3. Let ! be the cube root 1=2 C
the three roots of the polynomial y 3 C py C q are

3=2 of 1. Then

A C B; !A C ! 2 B; ! 2 A C !B:
Exercise 5.2. Use Cardanos formula, as clarified in Theorem 5.1, to solve
the cubic equation
y 3 C 6y 20 D 0
that we first studied in Exercises 3.9 and 3.10. Theres no need to redo work
already done. Use the result of Exercise 3.10 to write a cube root for each
term in the formula. Then use what we have learned since then to write two
more cube roots for each term, pair them correctly, and take their sums to
find the three solutions to the equation. One solution is real and two are not.
Given a cubicp
polynomial y 3 C
ppy C q for which R is negative, the
R will not be real, and the calculation
numbers q=2 C R and q=2
of the roots given by Cardanos formula requires us to compute cube roots
of non-real numbers.
Exercise 5.3.

Use Cardanos formula to solve the cubic equation

y3

7y C 6 D 0

that we first studied in Exercise 3.13.

(i) Write the solution given by the formula as a sum of cube roots. It
involves the cube roots of
10 p
3C
3
9

i
i

i
IrvingBook 2013/5/22 15:39 page 112 #128

112

5. Cubic Polynomials, II

and
3

10 p
3:
9

(ii) Using thepnumbers ! and ! 2 and the determination of one cube root of
3 C 10
3 in Exercise 3.13 or Exercise 4.28, write expressions
for
9
p
3.
Write
the three complex numbers that are cube roots of 3 C 10
9
p
also the three complex numbers that are cube roots of 3 10
3.
9
p
p
10
10
(iii) Pair the cube roots of 3 C 9
3 and 3
3 as specified in
9
Theorem 5.1 to get three pairs such that the product of the complex
numbers in each pair equals 7/3.
(iv) Add the complex numbers in each pair to obtain all three real solutions
of y 3 7y C 6 D 0.
When making calculations
of cube roots of complex numbers, ! can be
p
written as 1=2 C
3=2 or as cos 120 C i sin 120 , depending on the
form used for complex numbers, and similarly for ! 2 . It may be convenient,
given angles  and that add up to 360, to view not as 360  but as
, keeping in mind that cos.360 / D cos. / D cos  and sin.360
/ D sin. / D sin .
Exercise 5.4.

Solve the cubic equation

y3

3y C 1 D 0:

(i) Using Cardanos formula, show that the solutions have the form
s
s
p
p
3
1
3
3
1
3
C
i C
i:
2
2
2
2
(ii) The numbers inside the cube roots signs are our familiar friends ! and
! 2 , the non-real cube roots of 1. Thus, the solution can be rewritten as
p
p
3
y D 3 ! C !2 :
(iii) Find three cube roots of ! and three cube roots of ! 2 , expressed in
terms of the cosine and sine of suitable angles.
(iv) Form pairs of cube roots in accordance with Exercise 5.1 and add them,
taking into account the advice that preceded this exercise.
(v) The three solutions to y 3 3y C 1 D 0 have the form 2 cos , for three
particular angles . What are they?

i
i

i
IrvingBook 2013/5/22 15:39 page 113 #129

5.2. The Resolvent

113

(vi) We now have precise expressions for the three roots of y 3 3y C

1, expressed in terms of cosines. To proceed further, use a calculator
to determine the values of the cosines, at least approximately. Then
substitute the approximate answers into the polynomial y 3 3y C 1.
As the last part of Exercise 5.4 may suggest, once we turn to calculators,
we are settling for approximate answers, and these can be obtained in a
variety of ways independent of Cardanos formula. For example, we can
solve cubic equations using Newtons methoda standard application of
calculusor turn to a selection of mathematical programs without even
worrying about the nature of the underlying algorithms.
The need to approximate already reared its head in our use of the quadratic formula. The finding of square roots of positive real numbers is an
algorithmic process, one we carry out until we arrive at as close an approximation to the answer as we wish.
What the quadratic formula does for us is reduce the problem of solving
arbitrary quadratic equations to that of calculating square roots. Likewise,
Cardanos formula reduces solving cubic equations to the calculation of
square and cube roots. It is only at this stage, and because we may be calculating cube roots of complex numbers, that we turn to trigonometry and
calculators.
In addition, both the quadratic formula and Cardanos have theoretical
value, representing roots of polynomials in terms of their coefficients. We
will make use of this in Section 5.3 in obtaining a formula for the discriminant of a cubic polynomial.

5.2 The Resolvent

Given the reduced cubic polynomial y 3 C py C q, one can rewrite the terms
r
q
p3
q2

C
2
27
4
whose cube roots we compute in using Cardanos formula as
p
q
q 2 C 4p 3 =27

:
2
2
These are the roots of the quadratic polynomial t 2 C qt p 3 =27, which has
q 2 C 4p 3=27 as its discriminant. This suggests an alternative approach to
the derivation of Cardanos formula, one presented by Euler in Elements of
Algebra [26, pp. 263264].

i
i

i
IrvingBook 2013/5/22 15:39 page 114 #130

114

5. Cubic Polynomials, II

Exercise 5.5. Let b and c be real numbers and let U , V be the roots of the
quadratic polynomial x 2 C bx C c. From Exercise 2.5, b D .U C V / and
c D UV.
(i) Let d be the real cube root of c. Choose cube roots A of U and B of
V so that AB D d .

(ii) Using the formulas A3 D U , B 3 D V , and AB D d , expand .ACB/3

and verify that
.A C B/3 D 3d.A C B/ b:
(iii) Conclude that Theorem 5.2 holds.

p
Theorem 5.2. Let b and c be real numbers and let 3 c be the real cube
root of c. Let U and V be the roots of the quadratic polynomial
x 2 C bx C c:
p
p
p
p
p
Choose cube roots 3 U and 3 V of U and V so that 3 U
3 V D 3 c.
Then the sum
p
p
3
3
UC V
is a root of the cubic polynomial
p
3 3 c y C b:

y3

It is now an easy matter to rederive Cardanos formula:

Theorem 5.3 (Cardano again). Let p and q be real numbers. Let U and V
be the roots of the quadratic polynomial
t 2 C qt
Choose cube roots
Then the sum

p3
:
27

p
p
p
p
3
3
3
3
U and V of U and V so that U
V D

p=3.

p
p
3
3
UC V

is a root of the cubic polynomial

y 3 C py C q:
Exercise 5.6. Verify that Theorem 5.3 is an immediate consequence of
Theorem 5.2. Verify also that Theorem 5.3 becomes Cardanos formula in
its usual form once we use the quadratic formula to express U and V in
terms of p and q.

i
i

i
IrvingBook 2013/5/22 15:39 page 115 #131

115

5.3. The Discriminant

The polynomial t 2 C qt
its resolvent quadratic.

p 3 =27 associated with y 3 C py C q is called

5.3 The Discriminant

For a quadratic polynomial x 2 C bx C c with real or complex roots r1 and
r2 , we saw in Exercise 4.8 and Theorem 4.4 that the squared difference
.r1 r2 /2 of the roots is expressible in terms of the coefficients:
.r1

r2 /2 D b 2

4c:

This allows us to obtain information on the roots from the coefficients alone:
if b 2 4c is positive, the roots are real and distinct; if b 2 4c is 0, the roots
are real and coincide; and if b 2 4c is negative, then the roots are a pair of
non-real, complex conjugate numbers.
We would like similarly to be able to obtain information about the roots
of a cubic polynomial in terms of its coefficients. The discussion that followed Exercise 4.8 showed that the two quantities b 2 4c and .r1 r2 /2
associated with x 2 C bx C c can both be regarded as its discriminant, but
.r1 r2 /2 is the more fundamental quantity, the one we chose as the discriminants definition.
Given a cubic polynomial x 3 C bx 2 C cx C d , lets write r1 ; r2; r3 for
its roots, real or complex. In analogy with the quadratic case, we define the
discriminant of x 3 C bx 2 C cx C d to be
.r1

r2 /2 .r1

r3 /2 .r2

r3 /2

and denote it by .
For the special case of a reduced cubic polynomial x 3 C px C q, we
introduced the quantity 4p 3 27q 2 in Section 3.4, denoted it by , and
called it the discriminant. We also saw, in Exercise 3.22 and Theorem 3.6,
that the nature of the roots of x 3 CpxCq is determined by : the polynomial
has three distinct real roots if is positive, a repeated real root if is 0,
and one simple real root plus two complex conjugate roots if is negative.
Lets show for a general cubic polynomial that our newly introduced
determines the nature of the roots in the same way.
Exercise 5.7. Let f .x/ D x 3 C bx 2 C cx C d . We will relate the sign of
the discriminant of f .x/ to the nature of the roots of f .x/.
(i) Suppose f .x/ has a multiple root. Show that D 0.
(ii) Suppose that the three roots are real and distinct. Show that > 0.

i
i

i
IrvingBook 2013/5/22 15:39 page 116 #132

116

5. Cubic Polynomials, II

(iii) Suppose that one root is real and the other two are complex conjugates.
Show that < 0. (Hint: Suppose the roots are r , a C bi , and a bi ,
with r , a, and b all real numbers. Using this notation, calculate by
calculating the product of the root differences before squaring.)
(iv) Prove the converses: if D 0, then f .x/ has a multiple root and all
its roots are real; if > 0, then f .x/ has three distinct real roots;
and if < 0, then f .x/ has one real root and two non-real complex
conjugate roots. (Hint: These follow from the three statements by the
logical argument that we used in Exercise 4.7 to prove the quadratic
analogue.)
(v) Conclude that you can determine the nature of the roots of f .x/ from
the sign of the discriminant of f .x/, as described in Theorem 5.4.
Theorem 5.4. Let f .x/ be a degree 3 polynomial with discriminant .
(i) If is positive, then f .x/ has three distinct real roots.
(ii) If D 0, then f .x/ has only real roots, one of which occurs with
multiplicity at least 2.
(iii) If is negative, then f .x/ has three distinct roots. One is real and two
form a complex conjugate pair.
We would like to be able to compute the discriminant of a cubic polynomial in terms of its coefficients, so that we can determine the nature of the
roots without computing them. Lets first consider a reduced cubic polynomial, one of the form y 3 C py C q. We just reviewed that the nature of its
roots is determined by the sign of the quantity , with D 4p 3 27q 2. If
we can prove that D , then we can put aside any concerns that the two
uses of the word discriminant conflict. At the same time, we will have an
algebraic proof that the quantity 4p 3 27q 2 determines the nature of the
roots. This would contrast with our proof of Theorem 3.6, which depended
on calculus.
Exercise 5.8. Suppose p and q are real numbers and consider the polynomial y 3 C py C q. We will use the notation and results of Exercise p
5.1.
Accordingly, we choose A to be one of p
the three cube roots of q=2 C R
and B to be the cube root of q=2
R satisfying AB D p=3. With
these choices the three roots of y 3 C py C q are
r1 D A C B; r2 D !A C ! 2 B; r3 D ! 2 A C !B:

i
i

i
IrvingBook 2013/5/22 15:39 page 117 #133

117

5.3. The Discriminant

(i) Verify that

r1

r2 D .1

!/.A

! 2 B/;

that
r1

r3 D ! 2 .1

!/.A

!B/;

r3 D !.1

!/.A

B/:

and that
r2
(ii) Verify that
.1

p
!/ D 3 3i

!/3 D 3.! 2

and use Theorem 4.16 to deduce that

.r1

r2 /.r1

r3/.r2

r3 / D .1 !/3 .A3

p
B 3/ D 3 3i.A3

B 3 /:

p
(iii) p
The definitions of A and B yield A3 D q=2C R and B 3 D q=2
R. Deduce that
p
A3 B 3 D 2 R
and that
.r1

r2 /.r1

r3 /.r2

p p
r3 / D 6 3i R:

(iv) Deduce that the discriminant of y 3 C py C q is given by the formula

D

108R:

(v) Conclude that Theorem 5.5 holds.

Theorem 5.5. Given real numbers p and q, the discriminant of the polynomial y 3 C py C q is
4p 3 27q 2:
Exercise 5.9. Calculate the discriminant of each cubic polynomial and
state whether it has a multiple root, a real root and two non-real roots, or
three distinct real roots.
(i) y 3

3y C 2.

(ii) y 3 C 5y C 1.
(iii) y 3

5y C 1.

i
i

i
IrvingBook 2013/5/22 15:39 page 118 #134

118

5. Cubic Polynomials, II

It remains to show for a general cubic polynomial x 3 C bx 2 C cx C d

that its discriminant can be computed in terms of b, c, and d . This will
allow us, as in the quadratic case, to determine the nature of the roots of a
cubic from the coefficients alone.
Exercise 5.10. Consider the cubic polynomial x 3 C bx 2 C cx C d , with
real numbers b, c, and d as coefficients.
(i) From Exercise 3.2 and Theorem 3.3, the change of variable x D y
.b=3/ transforms x 3 C bx 2 C cx C d into a polynomial of the form
y 3 C py C q. Write the roots of y 3 C py C q as r1 , r2 , and r3 . In
terms of r1 , r2 , and r3 , what are the roots of x 3 C bx 2 C cx C d ? (See
Exercise 3.2.)
(ii) Observe that although the roots of x 3 C bx 2 C cx C d and y 3 C px C q
differ, their differences are the same.
(iii) Deduce that the discriminant of x 3 Cbx 2 CcxCd and the discriminant
of y 3 C px C q are equal.
(iv) Using the formulas of Theorem 3.3 for p and q in terms of b, c, and
d and the formula for the discriminant of y 3 C px C q, obtain the
formula in Theorem 5.6 for the discriminant of x 3 C bx 2 C cx C d in
terms of b, c, and d .
(v) Conclude that the nature of the roots of x 3 C bx 2 C cx C d can be
determined in terms of b, c, and d , without explicit knowledge of the
roots.
Theorem 5.6. Given real numbers b, c, and d , the discriminant of x 3 C
bx 2 C cx C d is
18bcd

4b 3d C b 2 c 2

4c 3

27d 2:

Exercise 5.11. Calculate the discriminant of each cubic polynomial and

state whether it has a multiple root, a real root and two non-real roots, or
three distinct real roots.
(i) x 3 C 2x 2
3

(ii) x C 3x

4.
5x C 1.

The discriminant separates cubic polynomials into three families according to the nature of their roots, as described in Theorem 5.4: there are

i
i

i
IrvingBook 2013/5/22 15:39 page 119 #135

5.3. The Discriminant

119

three distinct real roots when is positive, there is one real root along with
a distinct pair of complex conjugate roots when is negative, and there is
a repeated root when is zero. Theorem 5.6 allows us to determine which
occurs for a cubic polynomial from its coefficients alone. If D 0, we can
use the coefficients to do a little more.
Theorem 5.7. The polynomial x 3 Cbx 2 Ccx Cd has one real root, of multiplicity 3, if D 0 and b 2 3c D 0. It has two real roots, of multiplicities
1 and 2, if D 0 and b 2 3c 0.
Exercise 5.12. Prove Theorem 5.7.
(i) Check that the only reduced cubic polynomial with a root of multiplicity 3 is y 3 . (Hint: Given a real number r , what is the coefficient of y 2
in .y r /3 ?)
(ii) Deduce that y 3 C py C q has a root of multiplicity 3 if D 0 and
p D 0, but roots of multiplicities 1 and 2 if D 0 and p 0.
(iii) Observe that x 3 C bx 2 C cx C d has a root of multiplicity 3 precisely
when the reduced cubic polynomial associated with it by the change
of variable of Theorem 3.3 does.
(iv) Show that the theorem follows.
The sign of a cubics discriminant determines whether or not we will
have to calculate cube roots of non-real complex numbers when using Cardanos formula. We saw this at the conclusion of Section 3.4 in terms of
, the discriminant we obtained using calculus. Having now developed the
theory of the discriminant algebraically, we can restate the situation in terms
of .
Exercise 5.13. Consider the reduced cubic polynomial y 3 C py C q. As
usual, let R D p 3 =27 C q 2 =4.
(i) Using Theorem 5.5, observe that and R have opposite signs.
(ii) Review the relation of the sign of to the nature of the roots and
rephrase it as a relation between the sign of R and the nature of the
roots.
(iii) Deduce that if y 3 C py C q has one real root and two non-real roots,
then Cardanos formula expresses the real root as the sum of cube roots
of real numbers.

i
i

i
IrvingBook 2013/5/22 15:39 page 120 #136

120

5. Cubic Polynomials, II

(iv) In contrast, deduce that if y 3 Cpy Cq has three distinct real roots, then
Cardanos formula for the roots expresses all three of them as sums of
cube roots of non-real complex numbers.
The appearance of cube roots of non-real complex numbers in Cardanos formula when is positive can be made explicit by rewriting Cardanos formula with in place of R:

yD

s
3

s
p
q
q

3
C p C
2
2
6 3

p :
6 3

Our examples of polynomials for which such cube roots arose werent anomalies. They were inevitable.

5.4

Cardanos Formula Refined

The nature of a reduced cubic polynomials roots depends on the sign of

its discriminant. For each of the three possible caseszero, negative, or
positivewe can obtain a refined version of Cardanos formula. Lets start
with the case of zero discriminant.
Exercise 5.14. Let y 3 C py C q be a reduced cubic polynomial with discriminant equal to 0; that is, 4p 3 27q 2 D 0. The case p D q D 0 is not
interesting, so lets assume p and q are both non-zero.
(i) Show that p must be negative.
(ii) Solve the equation 4p 3

27q 2 D 0 for q in terms of p. Show that

2p
qD
3

p
:
3

(iii) To
p simplify notation, lets introduce a new constant a, with a D
p=3; that is, a is the positive square root of p=3. Check that
p D 3a2 and q D 2a3 . The choice of sign in the expression for
q in terms of a is determined by the sign of q: if q is positive, then
q D 2a3 ; if q is negative, then q D 2a3 .
(iv) Assume that q is positive. Rewrite the cubic polynomial y 3 C py C q
as
y3

3a2 y C 2a3:

i
i

i
IrvingBook 2013/5/22 15:39 page 121 #137

121

5.4. Cardanos Formula Refined

Verify that y D a and y D

factorization
y3

2a are roots. Show that we have the

3a2 y C 2a3 D .y

a/2 .y C 2a/:

(v) Assume that q is negative. Rewrite y 3 C py C q as

y3

3a2 y

2a3 :

Verify that its roots are a and 2a, and that we have the factorization
y3

3a2 y

2a3 D .y C a/2 .y

2a/:

(vi) Deduce Theorem 5.8.

Theorem 5.8. Let y 3 C py C q be a reduced cubic polynomial with discriminant equal to 0.
(i) If q D 0, then p D 0 and 0 is a root of multiplicity 3.
p
p
(ii) If q > 0, then
p=3 is a root of multiplicity 2 and 2
p=3 is a
root of multiplicity 1.
p
p
(iii) If q < 0, then
p=3 is a root of multiplicity 2 and 2
p=3 is a
root of multiplicity 1.
Lets next consider the case of negative discriminant. We saw in Exercise 4.27 that a real number c has one real cube root, say d , and two nonreal cube roots, !d and ! 2 d . This allows us to be more explicit about our
choices of A and B in the version of Cardanos formula stated as Theorem
5.1.
Exercise 5.15. Let y 3 C py C q be a reduced cubic polynomial with
negative discriminant; that is, 4p 3 27q 2 < 0. Let R as usual equal
p 3 =27 Cp
q 2 =4, so D 108R
p and R > 0. In particular, the quantities
q=2 C p R and q=2
R are real. Let A be thepreal cube root of
q=2 C R and let B be the real cube root of q=2
R.
(i) Verify that AB D p=3. (This can be done directly. Or, from Exercise
5.1, the product of A with one of B, !B, and ! 2 B is p=3. Consider
which product is a real number.)
(ii) Deduce in the case of negative discriminant that Theorem 5.1 takes on
the special form of Theorem 5.9.

i
i

i
IrvingBook 2013/5/22 15:39 page 122 #138

122

5. Cubic Polynomials, II

Theorem 5.9. Let y 3 Cpy Cq be a reduced cubic polynomial with negative

discriminant. Let A be the real cube root of
r
q
p3
q2
C
C ;
2
27
4
let B be the real cube root of
r
p3
q
q2
C ;
2
27
4
p
and let ! D 1=2 C
3=2. Then y 3 C py C q has one real root, A C B,
and two non-real roots, !A C ! 2 B and ! 2 A C !B.
In using Cardanos formula to find the roots of a cubic polynomial
y 3 C py C q with positive discriminant, we can cut the work of cube root
calculations in half once we recognize that the two expressions whose cube
roots we need to calculate are complex conjugates of each other.
Exercise 5.16. Let y 3 C py C q be a reduced cubic polynomial with positive discriminant; that is, 4p 3 27q 2 > 0. Let R D p 3 =27 C q 2 =4, so
that R < 0.
p
p
(i) Since R is negative, p
R has two real square
R and
R.
proots,
Observe that q=2 C
R i and q=2
R i are complex conjugates.
(ii) If s is a complex number and r is one of its cube roots, then Exercise
4.1
cube root of q=2C
p yields that r is a cube root of s. Suppose A is a p
R i . Deduce that A is a cube root of q=2
R i.
(iii) Verify that AA D p=3. (Use the fact that the product of A with one
of A, !A, and ! 2 A is p=3. Consider which product is a real number.)
(iv) Use Exercise 4.4 to verify that !A and ! 2 A are complex conjugates
and that ! 2 A and !A are complex conjugates.
(v) Deduce that Theorem 5.1 takes on the special form of Theorem 5.10.
Theorem 5.10. Let y 3 CpyCq be a reduced cubic polynomial with positive
discriminant. Let A be a cube root of
r
p3
q2
q
C
C
2
27
4
p
and let ! be the cube root 1=2 C
3=2 of 1. Then the three roots of
y 3 C py C q are A C A, !A C !A, and ! 2 A C ! 2 A.

i
i

i
IrvingBook 2013/5/22 15:39 page 123 #139

5.4. Cardanos Formula Refined

123

Let us return to the cubic y 3

5.10.

7y C 6 and find its roots using Theorem

Exercise 5.17. Use Cardanos formula, as refined in Theorem 5.10, to

solve the cubic equation
y3
p

7y C 6 D 0:

(i) One cube root of 3 C 10

3=9 was determined in Exercise 3.13 or
Exercise 4.28. Add this cube root to its conjugate to obtain one solution
to y 3 7y C 6 D 0.

(ii) Using ! p
and ! 2 , write expressions for the other two cube roots of
3=9. Add each to its conjugate to obtain the other two
3 C 10
solutions.
We close this section with two exercises that provide an opportunity to
tie together the ideas we have learned so far.
Exercise 5.18. Solve the equation
x 3 C 6x 2 C 3x C 18 D 0:

The solution should be presented in essay form, with explanations of what is

being done at each step and why it is being done. The solution should not be
merely lines of calculation. There should be English prose. The following
steps should be included.
(i) Change variables to obtain a presumably simpler cubic equation to
solve.
(ii) State Cardanos formula clearly in general and then explain how to
apply it.
(iii) Use Cardanos formula to obtain a real solution.
(iv) Obtain the other two solutions in two ways. First use factorization and
the quadratic formula. Then obtain the other two solutions using cube
roots of 1 and Cardanos formula again.
(v) Obtain three solutions to the original equation.
Exercise 5.19. Solve the equation
x3

3x 2

10x C 24 D 0:

i
i

i
IrvingBook 2013/5/22 15:39 page 124 #140

124

5. Cubic Polynomials, II

Again the solution should be presented as an essay. The points listed in

Exercise 5.18 apply to this equation as well, but there are some differences
in the details:
(i) This time it may be necessary to calculate cube roots of non-real complex numbers, which can be done using trigonometry, along with a
calculator.
(ii) Calculator solutions are only approximate. Point this out. Even though
the solution is approximate, it may in fact be correct. Check to see if it
is.

5.5

The Irreducible Case

Reduced cubic polynomials with positive discriminant are precisely those

with three distinct real roots, as we saw in Theorem 5.4. Thus, they factor as
products of three distinct linear polynomials and are said to be completely
reducible in modern terminology. Yet, they are also the polynomials whose
roots Cardanos formula expresses as sums of cube roots of non-real complex numbers. Cardano called this the irreducible case. To understand the
sense in which irreducible applies, lets continue an analysis that we began in Exercises 4.13 and 4.14.
Exercise 5.20. Fix real numbers a and b with b 0. We wish to calculate
cube roots of a C bi . This amounts to solving the equation .x C yi /3 D
a C bi for real values of x and y,
(i) Review Exercises 4.13 and 4.14, where we found that the equation
.x C yi /3 D a C bi yields two equations for x and y:
x3

3xy 2 D aI 3x 2y

y 3 D b:

(ii) The substitution x D sy led to a single equation in s,

s3

a
3 s2
b

3s C

a
D 0:
b

The equation is a cubic in s with coefficients expressed in terms of

a=b.
(iii) Using Cardanos formula to solve a cubic equation of irreducible
type requires us to calculate a cube root of a non-real complex number.
We now find that calculating such a cube root leads to a new cubic
equation. Have we made progress? Lets examine the new cubic.

i
i

i
IrvingBook 2013/5/22 15:39 page 125 #141

125

5.6. Vietes Formula

(iv) Introduce another variable t, with s D t C .a=b). Substitute t C .a=b/

for s in the equation for s to obtain a reduced cubic equation for t,
t3

3kt

a
2 k D 0;
b

with k D 1 C .a2 =b 2/.

(v) We can try to solve it using Cardanos formula. Calculate R, or for
this new reduced cubic. Verify that R has the form k 2 . This means the
new cubic, in t, is also of irreducible type, so that solving it will require
another calculation of the cube root of a non-real complex number.
(vi) Show that in fact using Cardanos formula to solve the cubic in t requires us to compute the cube root of
a
k C ki:
b
(vii) Write this as
k

.a C bi /;
b
a real number times a C bi .
(viii) Conclude that in order to use Cardanos formula to find roots of the
reduced cubic equation in t, we must be able to compute the cube root
of none other than a C bi , the complex number whose cube root we
were trying to calculate in the first place!
The calculation in the last exercise shows that we have come full circle.
To solve an irreducible cubic equation, we need to calculate the cube root
of a non-real complex number a C bi . To compute it, we are led to solve a
different cubic equation. Passing from this different cubic to its associated
reduced cubic equation and applying Cardanos formula, we find that we
must compute the same cube root with which we began. It is in this sense
that the problem is irreducible.
We need not despair. We must accept that non-algebraic techniques are
needed. For example, we can use trigonometry to compute cube roots, as
worked out in Exercise 4.27.

5.6 Vietes Formula

We closed Section 5.5 with the recognition that we must leave the realm of
algebra for trigonometry to find the roots of reduced cubic polynomials in

i
i

i
IrvingBook 2013/5/22 15:39 page 126 #142

126

5. Cubic Polynomials, II

the irreducible case. We previously observed (Exercise 5.16 and Theorem

5.10) in this case that the two summands in Cardanos formula are complex
conjugates of each other. The sum of a complex number and its conjugate
is twice the real part of the two numbers: .a C bi / C .a bi / D 2a.
Thus, given a cubic polynomial y 3 C py C q with positive discriminant,
p
if we can determine the real parts of the cube roots of q=2 C R in
terms of p and q, we will have a new formula for the cubics roots. We
will obtain such a formula, using the standard trigonometric functions and
inverse trigonometric functions.
From the discussion before Exercise 4.16, the inverse cosine function
arccos x is defined only for real numbers in the interval 1; 1. As x increases from 1 to 1, the value arccos x decreases from  to 0, taking each
value once. The following inequality ensures that the inverse cosine is defined on expressions that will arise in Exercise 5.22.
Exercise 5.21. Suppose p and q are real numbers. Assume p < 0 and let
p
p=3 be the positive square root of p=3. Show that
1<

2p

3q
p

p=3

<1

precisely when 4p 3 27q 2 > 0. (Hint: Use the fact, for a real number r ,
that 1 < r < 1 precisely when r 2 < 1.)
Exercise 5.22. Suppose p and q are real numbers satisfying
4p 3 27q 2 >
p
3
2
R to be the pos0. Let R D p =27 C q =4 (which is negative) and take
itive square root of R.
p
(i) Observe that p must be negative. Write
p=3 for the the positive
square root of p=3.
(ii) By Exercise 4.17, we can write
q p
C
Ri
2
in the form
r cos  C .r sin /i
for real numbers r and , with r > 0. Using Exercise 4.16, show that
r
r
p3
p
p
rD
D
;
27
3
3

i
i

i
IrvingBook 2013/5/22 15:39 page 127 #143

127

5.6. Vietes Formula

and that one choice of  is

 D arccos

q
D arccos
2r

(By Exercise 5.21, arccos is defined here.)

2p

3q
p

p=3

(iii) Letsp
simplify our notation by introducing new constants a and b, with
aD
p=3 and b D 3q=p. Then r D a3 and  D arccos.b=2a/.
(iv) From Exercise 4.27, one cube root of

r cos  C r i sin 
is

p
p


3
r cos C 3 r i sin :
3
3
Deduce, using our shorthand notation, that

 

 
1
b
1
b
a cos
arccos
C ai sin
arccos
3
2a
3
2a
p
Ri .
is a cube root of q=2 C

(v) By Theorem 5.10, one root of y 3 C py C q is given by adding the

complex number displayed in (iv) to its conjugate. Adding a complex
number and its complex conjugate yields twice the real part of the
complex number. Conclude that

 
1
b
2a cos
arccos
3
2a
is a root of y 3 C py C q.
The formula for a root of the reduced cubic y 3 C py C q in Exercise
5.22 is due to Francois Vi`ete, whose work we will discuss further in Section
5.8. Lets record it as a theorem, replacing a and b with the expressions in
p and q that they represent.
Theorem 5.11 (Vi`ete). Suppose p and q are real numbers. Let y 3 C py C q
be a cubic polynomial with positive discriminant. Then one of the roots of
y 3 C py C q is given by the formula
!!
r
p
1
3q
2
cos
arccos
p
:
3
3
2p
p=3

i
i

i
IrvingBook 2013/5/22 15:39 page 128 #144

128

5. Cubic Polynomials, II

From Exercise 4.27, we can express the other two roots of y 3 C py C q

by similar expressions, by adding 2=3 and 4=3 to the argument of the
cosine.
Vi`etes formula allows us to avoid calculating cube roots of non-real
complex numbers in finding roots of cubic polynomials of irreducible type.
This is reminiscent of the situation we discussed in Section 4.8, where we
examined how to calculate nth roots of real and complex numbers using the
exponential and logarithm functions or using the cosine and inverse cosine
functions. We concluded that discussion with the observation that in both
cases the calculation involved using a familiar function f the exponential
or cosine functionin the same way. We calculate the value of f 1 for
some constant c, divide the resulting number by n, and then apply f to the
result or a minor variant. For example, the nth root of a positive real number
r is
e log.r /=n :
Vi`etes formula fits into this theme, with the cosine function as f . We calculate an inverse cosine, divide by 3, and calculate a cosine, thereby replacing
a cube root calculation (of a complex number!) with division by 3. Lets try
it.
Exercise 5.23. Verify that y 3 7y C 6 has positive discriminant. Then use
Vi`etes formula and a calculator to solve
y3

7y C 6 D 0:

Exercise 5.24. Verify that y 3 3y C 1 has positive discriminant. Then use

Vi`etes formula to express one of the solutions to the equation
y3

3y C 1 D 0

in terms of cosines. We get the same answer as in Exercise 5.4 using Cardanos and de Moivres formulas.
Vi`etes formula applies also to a cubic y 3 C py C q with zero discriminant. We can rework Exercise 5.21 p
(still assuming p < 0) to find that
4p 3 27q 2 D 0 precisely when 3q=2p
p=3 equals 1 or 1 . We obtain
1 when q is positive and 1 when q is negative. The inverse cosine
p of 1
is  and of 1 is 0. Thus, when q > 0, Vi`etes formula
yields
y
D
p=3,
p
while when q < 0, the formula yields y D 2
p=3. This proves again
part of Theorem 5.8.

i
i

i
IrvingBook 2013/5/22 15:39 page 129 #145

129

5.6. Vietes Formula

Although Vi`etes formula allows us to bypass the calculation of cube

roots of non-real complex numbers, such cube roots did appear in its derivation. Once we knowor guessthat a formula for y exists expressing it as
twice a positive multiple of a cosine, we may wish to derive it directly, without reference to Cardanos formula or complex numbers. This can be done,
using the cosine triple angle formula:
Theorem 5.12. Let  be a real number. Then
cos 3 D 4 cos3 

3 cos :

Exercise 5.25. Prove Theorem 5.12.

(i) Using the angle sum formulas of Theorem 4.9, write formulas for
sin 2 and cos 2 in terms of sin  and cos .
(ii) Using Theorem 4.9 again, write cos 3 in terms of sines and cosines
of  and 2.
(iii) Obtain the identity
cos 3 D cos3 

3 sin2  cos :

(iv) Use
cos2  C sin2  D 1
and the equality in (iii) to obtain the desired identity.
We can now use Theorem 5.12 to derive Vi`etes formula independently
of Cardanos formula.
Exercise 5.26. Let y 3 C py C q be a cubic polynomial with positive discriminant. Recall that p must be negative. We wish to find a solution in the
form y D 2a cos , where a and  are real numbers expressedpin terms of
p and q and a is positive. (We have been using a to represent
p=3, but
for now we assume that a is unknown).
(i) Substitute 2a cos  for y in the equation y 3 C py C q D 0 to obtain
8a3 cos3  C 2ap cos  C q D 0:
If we can find values of a and , in terms of p and q, for which this
holds, then we will have the desired formula.

i
i

i
IrvingBook 2013/5/22 15:39 page 130 #146

130

5. Cubic Polynomials, II

(ii) Use the triple angle formula and collect terms to rewrite the equation
as
.6a3 C 2ap/ cos  C 2a3 cos 3 C q D 0:
It will suffice to find values of a and  for which 6a3 C 2ap D 0 and
2a3 cos 3 C q D 0.

(iii) We have assumed that a is positive, so that a 0. Using this, solve

6a3 C 2ap D 0 for a and show that
p
aD
p=3:
(iv) Substitute

p=3 for a in the equation

2a3 cos 3 C q D 0;

solve for , and show that

!
3q
1
 D arccos
p
:
3
2p
p=3

(v) Rewrite 2a cos  as an expression in p and q using the expressions we

have just obtained for a and .
We have succeeded in deriving Vi`etes formula directly, without introducing complex numbers.

5.7

The Signs of the Real Roots

We have seen how to compute the discriminant of x 3 C bx 2 C cx C d in

terms of the coefficients b, c, and d . We have also seen how to use to decide whether x 3 Cbx 2 Ccx Cd has a repeated root, three distinct real roots,
or one real root and two complex conjugate roots. Thus this information can
be determined using only the coefficients.
It is possible to obtain more information from the coefficients, such as
how many of the real roots are positive and how many are negative. We will
close our treatment of cubic polynomials by working this out, because it is
interesting in its own right and because we will need the result in our study
of quartic polynomials. We begin by obtaining analogous results for linear
and quadratic polynomials.
Exercise 5.27. Let c be a real number. Show that the only root of the linear
polynomial x C c is c. Deduce that the sign of the root can be determined

i
i

i
IrvingBook 2013/5/22 15:39 page 131 #147

131

5.7. The Signs of the Real Roots

in terms of the sign of the constant coefficient. (Yes, this is trivial, but it is
the first in a sequence of results.)
Exercise 5.28. Let b and c be real numbers and let r1 and r2 be the roots
of the quadratic polynomial x 2 C bx C c. Since we are interested in the case
that the roots are real, we assume that b 2 4c  0.
(i) If c D 0, the roots are 0 and b, so their sign is determined by the sign
of the coefficient b.

(ii) Assume c 0. From Exercise 2.5, c D r1 r2 and b D .r1 C r2/.

Show that if r1 and r2 have the same sign, then c > 0, and if they have
opposite sign, then c < 0.
(iii) Assume that c is positive. Show that r1 and r2 are both positive precisely when b < 0 and both negative precisely when b > 0.
(iv) Deduce Theorem 5.13.
Theorem 5.13. Let b and c be real numbers satisfying b 2
c 0.

4c  0 and

(i) The polynomial x 2 C bx C c has two real roots, both non-zero.

(ii) If c > 0 and b < 0, then both roots are positive.

(iii) If c < 0, then one root is positive and one is negative.
(iv) If c > 0 and b > 0, then both roots are negative.
To handle cubic polynomials, we need a cubic analogue of Exercise 2.5.
Theorem 5.14. Let x 3 C bx 2 C cx C d be a cubic polynomial with real
coefficients b, c, and d , and suppose its three roots are r1 , r2 , and r3 . Then
b
c
d

D
r1 r2 r3 ;
D Cr1 r2 C r1 r3 C r2 r3 ;
D
r1 r2 r3 :

Exercise 5.29. Prove Theorem 5.14. (Hint: Use the factorization of x 3 C

bx 2 C cx C d as .x r1 /.x r2 /.x r3 /.)
A cubic polynomial of the form x 3 C bx 2 C cx factors as the product
of x and x 2 C bx C c, so that one of its roots is 0 and the others are roots of
x 2 C bx C c. Thus, we can determine the signs of its roots using Theorem
5.13. We will exclude this case.
Lets consider cubic polynomials that have non-real roots.

i
i

i
IrvingBook 2013/5/22 15:39 page 132 #148

132

5. Cubic Polynomials, II

Theorem 5.15. Let x 3 C bx 2 C cx C d be a cubic polynomial with real

coefficients b, c, and d whose discriminant is negative. Assume further that
d 0. Then the real root of x 3 C bx 2 C cx C d is positive if d is negative
and negative if d is positive.
Exercise 5.30. Prove Theorem 5.15. (Hint: Use Theorem 5.14 and the fact
that the product of a non-zero complex number and its conjugate is a positive real number.)
We now turn to cubic polynomials whose roots are all real.
Exercise 5.31. Assume that x 3 C bx 2 C cx C d is a cubic polynomial with
real coefficients b, c, and d . Assume that d 0 and that the discriminant
of x 3 Cbx 2 Ccx Cd is non-negative, so that the roots of x 3 Cbx 2 Ccx Cd
are all real (whatever their multiplicities).
(i) Show that if b, c, and d are positive, then all the roots are negative.
(Hint: If x > 0, what is the sign of x 3 C bx 2 C cx C d ?)

(ii) Conversely, show that if all the roots are negative, then b, c, and d are
positive. (Hint: Examine the expressions for b, c, and d in terms of the
roots given in Theorem 5.14.)
(iii) Show that if b is negative, c is positive, and d is negative, then all the
roots are positive. (Hint: If x < 0, what is the sign of x 3 C bx 2 C cx C
d ?)
(iv) Conversely, show that if all the roots are positive, then b is negative, c
is positive, and d is negative.
(v) Show that d > 0 if there are 0 or 2 positive roots and d < 0 if there
are 1 or 3 positive roots.
(vi) Deduce Theorem 5.16.
Theorem 5.16. Let x 3 C bx 2 C cx C d be a cubic polynomial with real
coefficients b, c, and d whose discriminant is non-negative. Assume further
that d 0.
(i) If b, c, and d are positive, then all the roots are negative.

(ii) If b is negative, c is positive, and d is negative, then all the roots are
positive.
(iii) If neither condition holds, then there are both positive and negative
roots. In this case, if d > 0, then there are two positive roots; if d < 0,
then there is one positive root.

i
i

i
IrvingBook 2013/5/22 15:39 page 133 #149

133

5.8. History

In the next exercise, we will apply what weve learned.

Exercise 5.32. For the cubic polynomials in (i)-(iv), use Theorems 5.15
and 5.16 to determine from the coefficients how many real roots there are
and how many of them are positive or negative.
(i) x 3
(ii) x

3
3

6x 2 C 11x
2

5x C 9x

6.
5.

(iii) x C 6x C 3x C 18.
(iv) x 3

3x 2

10x C 24.

Lets extract one small piece of Theorems 5.15 and 5.16 that will play
an essential role in our study of quartic polynomials. (See Exercise 6.8.)
Theorem 5.17. Let f .x/ be a cubic polynomial with non-zero constant
coefficient d .
(i) If d > 0, then f .x/ has at least one negative real root.
(ii) If d < 0, then f .x/ has at least one positive real root.
Exercise 5.33. Verify that Theorem 5.17 follows from Theorems 5.15 and
5.16. It is valid regardless of whether f .x/ has one or three real roots.

5.8 History
We concluded Section 5.5 with the observation that we need not despair
when faced with a cubic equation to solve in the irreducible case. How did
Cardano react? Lets find out, as we continue the historical account of cubic
equations that we began in Section 3.5.
The cubic equations whose solution Cardano learned from Tartaglia
have the special form
x 3 C px D q
with p and q positive. Since p is positive, the discriminant is negative and
the formula of del Ferro and Tartaglia involves cube roots of real numbers.
Cardano, however, extended Tartaglias solution to cover cubic equations of
the additional forms
x 3 D px C q
and
x 3 C q D px:

i
i

i
IrvingBook 2013/5/22 15:39 page 134 #150

134

5. Cubic Polynomials, II

Again, the coefficients are taken to be positive. But then, rewriting them in
the form x 3 C P x C Q D 0 by moving the appropriate terms to the left
side, we see that P is negative, so the discriminant 4P 3 27Q2 may
be positive. Thus Cardano opened the door to the irreducible case, with its
square roots of negative numbers and cube roots of non-real numbers.
Cardano was aware of the new difficulty. However, uncertain how to
handle it, he omitted examples of cubic equations of this type in Ars Magna
[12]. Thus, the answer to our opening question is that he suppressed the
irreducible case.
In an interesting passage later in Ars Magna, Cardano does make some
calculations with square roots of negative numbers. We know from Section
2.1 that solving a quadratic equation x 2 bx C c D 0 (the minus sign in
front of the b is intentional) is equivalent to finding two numbers whose
sum is b and whose product is c, and that special cases of this problem can
be found on Babylonian clay tablets from over three thousand years ago. In
Chapter 37, titled On the Rule for Postulating a Negative, Cardano poses
and solves just such a problem [12, p. 219]:
If it should be said, Divide 10 into two parts the product of which is
30 or 40, it is clear that this case is impossible. Nevertheless, we will
work thus: We divide 10 into two equal parts, making each 5. These we
square, making 25. Subtract 40 if you will, from the 25 thus produced
. . . leaving a remainder of 15, the square root of which added to or
subtracted
from 5 gives
p
p parts the product of which is 40. These will be
5C
15 and 5
15.
By competing the square, Cardano has found two numbers that have
sum 10 and product 40. Equivalently, he has solved the quadratic equation
x2

10x C 40 D 0;

despite its discriminant being negative. Next Cardano writes [12, pp. 219
220],
p
Putting
aside the mental tortures involved, multiply 5 C
15 by 5
p
15, making 25 . 15/, which is C15. Hence this product is 40.
. . . This truly is sophisticated . . . . So progresses arithmetic subtlety
the end of which, as is said, is as refined as it is useless.
Recognizing that the problem makes no physical sense, Cardano is content
to label the result useless.
Complex numbers enter into the solving of quadratic equations precisely when the equations have no real solutions. Though sophisticated, the

i
i

i
IrvingBook 2013/5/22 15:39 page 135 #151

135

5.8. History

new numbers arent needed. Whats different about cubic equations is that
the new numbers appear precisely when the equations have only real (and
distinct) solutions. This was a puzzle Cardano was not prepared to address.
In contrast, Cardanos near-contemporary Rafael Bombelli (15261572)
did not shy away from the irreducible case. In the influential three-volume
work LAlgebra [11], printed in 1572 and again in 1579, Bombelli set out to
present the results of Ars Magna in a manner more accessible to beginners.
Bombelli gives as one example the irreducible cubic equation
x 3 D 15x C 4:
(See the discussion in [66, pp. 6061].) Using Cardanos formula, Bombelli
writes the solution
q
q
p
p
3
3
2C
121 C 2
121:
He describes the imaginary square roots, following Cardano, as sophistic,
but he also notes that the cubic equation can be solved, since after all x D
4 is a solution. This motivates finding a way to make the formula work,
and Bombelli proceeds much as we did in Exercise 3.13 when we used
Cardanos formula topsolve the cubic equation y 3 7y C p
6 D 0.
b
as
a
possible
cube
root
of
2
C
121 and a
Lets
write
a
C
p
p
bpas the corresponding
cube
root
of
2
121.
By
setting
the cube of
p
aC
b equal to 2 C
121 and collecting terms, we get

a3 3ab D 2:
p
p
By
setting the product
of a C
b and a
b equal to the product of
p
p
p
p
3
3
2C
121 and 2
121, we get
a2 C b D 5:

Bombelli found
pa D 2 and b D 1 to this pair of equations.
p the solution
1 and 2
1, he obtained x D 4 as the solution to the
Adding 2 C
cubic equation, prompting him to comment, At first, the thing seemed to
me to be based more on sophism than on truth, but I searched until I found
the proof. [66, p. 61]
Bombelli also gave rules for the manipulation of these new numbers,
including the rule that
p
p
1

1 D 1;
and provided examples of calculations involving them. He didnt use our
notation, but the essential ideas were all there.

i
i

i
IrvingBook 2013/5/22 15:39 page 136 #152

136

5. Cubic Polynomials, II

We have focused on Italian mathematicians, but as the sixteenth-century

neared its end, the work of the French mathematician Francois Vi`ete came
to the fore. Some of his writings are available in a 1983 translation by T.
Richard Witmer under the title The Analytic Art [68]. In In Artem Analyticem Isagoge, which appeared in 1591 and is translated by Witmer as
Introduction to the Analytic Art, Vi`ete introduced the idea of using letters
for known and unknown quantities, differentiating between the two by using vowels for unknown quantities and consonants for the known quantities.
This allowed him to treat algebraic problems in generality rather than describing general methods implicitly through specific examples.
Another of Vi`etes contributions, which we studied in Section 5.6, was
the development of a trigonometric approach to cubic equations in the irreducible case that allows us to avoid complex numbers altogether. With A as
his variable and B and D as constants, he studies ([68, p. 174]) the equation
A3

3B 2 A D B 2 D

under the assumption that B is greater than half of D. Taking B and D

to be positive, we can check that the condition B > D=2 is the requirement
that the given cubics discriminant is positive. He then describes how to
obtain a solution geometrically in terms of two right triangles, one having
an acute angle three times that of the second. Next, Vi`ete illustrates the
method by solving the cubic equations
x3

300x D 432

and
x3

300x D 432:

In [48, pp. 203205], R.W.D. Nickalls relates this passage of Vi`etes

to our more familiar trigonometric view, explaining that Vi`etes approach
stems from his familiarity with the then equivalent of the trigonometric
triple-angle identities, since he himself had established formulae for chords
of multiple arcs in terms of chords of simple arcs, and hence he was aware
that solving a cubic with three real roots was analogous to trisecting an
angle.
Before moving on, let us recognize the greater significance of Vi`etes
work, as he was the first to write down cubic equations (in the irreducible
case) in general form and solve them.
Another French mathematician, Albert Girard (15951632), who would
spend most of his life in the Netherlands as a religious refugee, published Invention Nouvelle en lAlg`ebre in 1629 [32]. Its a short work, just 64 pages,

i
i

i
IrvingBook 2013/5/22 15:39 page 137 #153

5.8. History

137

with a central section on algebra that is full of new ideas. We will discuss
Girards contributions further in Sections 7.2 and 7.4. For now, we will be
content with a few words about his account of cubic equations.
Girard doesnt use letters for constants or variables, yet his text is easy to
understand. For example, he divides cubic equations into a variety of cases,
one of which he describes as the equations with 1.3/ equal to .1/ C .0/.
The parenthetical numbers should be interpreted as powers of x, so that
this refers to equations in which a cube equals a degree-one term plus a
constant, or x 3 D px C q. For this case, Girard provides a rule for solving
the equation 1.3/ equal to .1/ C .0/ when the cube of a third of the number
of .1/ is larger than the square of half of .0/ with the aid of a table of
sines. This is, of course, the irreducible case. Without a general notation,
Girard must describe the rule by way of example. The example he uses is
1.3/ equal to 13.1/ C 12, which we would write as x 3 D 13x C 12. He
displays a series of calculations that leads to the solution x D 4, using what
is essentially Vi`etes approach.
We mentioned John Wallis in passing in Section 4.9 with regard to his
early effort to provide a geometric description of complex numbers. This is
contained in his wonderfully titled 1685 book, A Treatise of Algebra, both
Historical and Practical: Shewing the original, progress, and advancement
thereof, from time to time, and by what steps it hath attained to the heighth
at which now it is; with some additional treatises, which would have great
influence. J.F. Scotts The Mathematical Work of John Wallis, written in
1938, is a valuable guide to Walliss mathematical work, with a chapter
devoted to the Treatise of Algebra [58, pp. 133165]. Scotts concluding
appraisal [58, p. 165] is that Walliss book constituted a reservoir from
which contemporary and later algebraists drew much inspiration. It may
be mentioned that this treatise did a great deal towards popularizing the
notation which was now rapidly becoming current in Europe.
Of particular significance is Walliss unreserved adoption of complex
numbers as solutions to algebraic equations. Scott writes [58, pp. 156157],
In his quadratic equations he discusses every type, and the rules he evolves
for determining the nature of the roots by a mere inspection of the equation
would not be out of place in a modern text-book. He was quite at home
with imaginaries, and he knew that such roots always occurred in pairs.
Moreover he would not allow the use of the word Impossible as applied to
an equation with imaginary roots. An equation, for example, such as
aa C 8a D 25;
p
p
of which the roots are 4 C 16 25 and 4
16

25, imaginaries

i
i

i
IrvingBook 2013/5/22 15:39 page 138 #154

138

5. Cubic Polynomials, II

had hitherto been styled Impossible. Yet, avers Wallis (and now we turn
to Walliss own writings [69, p. 174]):
Imaginary quantities, when they occur, have been taught to imply an
impossible case; and algebraists have been wont so to teach. Yet is
not this so to be understood . . . . For it was before thought (and so
delivered by diverse algebraists) that whenever (in pursuance of the
resolution) we are reduced to an impossible construction, (such as the
square root of a negative quantity), the case proposed is to be judged
impossible. Which is yet here discovered to be otherwise.
As for instance, the equation aaa
s
3

3C

7a D 6 should have for its root

100
C
27

s
3

100
;
27

which should therefore be judged an impossible case. Yet hath it a real

root, a D 3, beside which it hath also two negatives, a D 1 and
a D 2. And if we change but the sign of the absolute number (which
is the only even place not vacant), the equation
aaa

7a D 6

will have two affirmative roots, a D 1, a D 2 (and one negative root,

a D 3).
Walliss example, the equation aaa 7a D 6, should look familiar. It is
the one we used in Exercise 3.12 to illustrate the appearance of complex
numbers in Cardanos formula.
Half a century later, in the 1739 paper On the reduction of p
radicals
to simpler
terms,
or
the
extraction
of
roots
of
any
binomial
a
C
Cb or
p
aC
b, de Moivre sets himself the problem
[17],
[60,
p.
447]
to
extract
p
the cube root of the impossible binomial aC
b. He reduces this problem
to solving a reduced cubic equation, but in a different way from how we
proceeded in Exercises 4.13 and 4.14. Rather than quoting from his entirely
readable text [60, p. 447], lets reproduce his approach as an exercise.
Exercise 5.34. Let a and b be real numbers with b 0 and suppose x Cyi
is a cube root of a C bi .
(i) Set .x C yi /3 D a C bi , as in Exercise 4.13, and obtain two equations
in x and y by setting the real and imaginary parts equal to each other.

i
i

i
IrvingBook 2013/5/22 15:39 page 139 #155

139

5.8. History

(ii) Square both equations, take their sum, and obtain the equation
x 6 C 3x 4y 2 C 3x 2 y 4 C y 6 D a2 C b 2:
The left side is .x 2 C y 2 /3 .
(iii) Let m be the real cube root of a2 C b 2 and take cube roots of both sides
to obtain x 2 C y 2 D m.
(iv) Substitute m x 2 for y 2 in one of the equations of the first part to get
a reduced cubic equation in x:
4x 3

3mx

(v) Calculate the discriminant of .4x 3

b 0 to show that it is positive.

a D 0:
3mx

a/=4 and use the fact that

(vi) Conclude that cube roots of complex numbers can be calculated by

solving reduced cubic equations in the irreducible case.
Following de Moivre, we have converted the calculation of the cube
root of a complex number to the solving of a reduced cubic equation of
irreducible type. At this point, de Moivre observes that this will best be
done by means of a table
p of sines, illustrating the process by determining
the cube roots of 81 C
2700. The cubic equation to be solved is
4x 3

63x

81 D 0:

De Moivre suggests that this be compared with the equation for the cosines,
namely 4x 3 3r 2x D r 2c. To understand the sense in which this is an
equation for cosines, we can rewrite it as
 x 3
x
c
4
3 D :
r
r
r

Taking x=r to be cos  and c=r to be cos 3 for a suitable angle , the
equation becomes the trigonometric identity
4 cos3 

3 cos  D cos 3

of Theorem 5.12.
Following de Moivre again, we use the comparison
of 4x 3 63x D 81
p
3
2
2
to 4x
3r x D r c to obtain the values r D 21 and c D 27=7. To find
x, we use the triple cosine identity and the identifications x=r D cos  and
c=r D cos 3. Applying the inverse cosine to c=r , dividing by 3, computing

i
i

i
IrvingBook 2013/5/22 15:39 page 140 #156

140

5. Cubic Polynomials, II

the cosine, and multiplying by r yields x. De Moivre does the equivalent,

thereby obtaining the equations
three solutions. From these, he determines
p
three cube roots of 81 C
2700. (See [60, pp. 447449] for the complete
discussion.)
In our approach to solving reduced cubic equations, we have relied on
Cardanos formula to produce a number whose cube root we need to calculate, and in the irreducible case applied de Moivres formula to make
the cube root calculation. In his 1739 paper, de Moivre has reversed the
flow. He starts from a complex number whose cube root we want to calculate, produces a reduced cubic equation we wish to solve that falls under
the irreducible case, then takes Vi`etes approach to solving the equation.
De Moivre concludes his discussion with the observation that the use of
trigonometry is essential [60, p. 449]:
There have been several authors, and among them the eminent Wallis,
who have thought that those cubic equations which are referred to the
circle, may be solved p
by the extraction of the cube root of an imaginary
2700, without regard to the table of sines, but
quantity, and of 81 C
that is a mere fiction and a begging of the question. For on attempting
it, the result always recurs back again to the same question as that first
proposed. And the thing cannot be done directly, without the help of
the table of sines.
He recognizes that computing cube roots of complex numbers and solving
cubic equations in the irreducible case are essentially the same problem. We
can loop back and forth between the two, but to break the cycle, we must
introduce trigonometry.
Let us close our historical account of the cubic by examining Eulers
treatment in Elements of Algebra [26]. Euler derives Cardanos formula in
Chapter XII of Section 4, Of the Rule of Cardan, or of Scipio Ferreo. His
approach is simple and elegant, relying on the introduction of an associated
quadratic polynomial. Compare the following excerpt [26, pp. 263264]
with Theorems 5.2 and 5.3 (ignoring the regrettable clash of notation).
p
Whenever we meet with an equation of the form x 3 D 3x 3 pq C
p
p
p C q, we know that one of the roots is 3 p C 3 q. Now, we can
p
determine p and q, in such a manner, that both 3 3 pq and p C q
may be quantities equal to determinate numbers; so that we can always
resolve an equation of the third degree, of the kind which we speak of
[i.e., reduced].
Let, in general, the equation x 3 D f x C g be proposed. Here, it will
p
be necessary to compare f with 3 3 pq and g with p C q; that is, we

i
i

i
IrvingBook 2013/5/22 15:39 page 141 #157

141

5.8. History

p
must determine p and q in such a manner, that 3 3 pq may become
equal to f , and p C q D g; for we then know that one of the roots of
p
p
our equation will be x D 3 3 p C 3 3 q.
Following his derivation, Euler presents a sequence of examples of cubic equations with negative discriminant, starting with two for which the
cube root calculations are trivial. He introduces a third example [26, p 265],
x 3 D 6x C 40, with the immediate comment that x D 4 is one of the
roots. Applying Cardanos formula, he expresses the same surprise we did
after Exercise 3.9 on comparing the complicated answer it produces with
the known simpler answer:
Consequently one of the roots will be . . .
q
q
p
3
3
x D 20 C 14 2 C 20

p
14 2I

which quantity is really = 4, although,

pupon inspection, we
p should not
suppose it. In fact, the cube of 2 C p2 being 20 C 14
p 2, we have,
reciprocally,
the
cube
root
of
20
C
14
2
equal
to
2
C
2; in the same
p
p
p
p
3
20 14 2 D 2
2; wherefore our root x D 2 C 2 C
manner,
p
2
2 D 4.

Next Euler solves a non-reduced cubic, discussing the need (in Hewletts
translation) to destroy the second term and illustrating how to do so. The
resulting reduced cubic is treated like the example above, after which Euler
observes [26, p 268]:
It was, however, by chance, as we have remarked, that we were able,
in the preceding example, to extract the cube root of the binomials that
we obtained, which is the case only when the equation has a rational
root; : : : But when there is no rational root, it is, on the other hand,
impossible to express the root which we obtain in any other way, than
according to the rule of Cardan; so that it is then impossible to apply
reductions.
For example,
in the equation x 3 D 6x C4, we have : : : x D
p
p
p
p
3
3
2C2
1C 2 2
1, which cannot be otherwise expressed.

Thus, on arriving at the irreducible case, Euler declares that we can go no

further than to leave the solution in such a complicated form.
In contrast to Cardano, Euler was comfortable with the use of complex
numbers. We already quoted a passage from earlier in his book [26, p. 43]
in which he wrote that nothing prevents us from making use of these imaginary numbers, and employing them in calculation. Perhaps he chose not
to pursue the example further because he preferred to avoid trigonometry as

i
i

i
IrvingBook 2013/5/22 15:39 page 142 #158

142

5. Cubic Polynomials, II

a tool. In a footnote [26, p 268] to Eulers passage, Bernoulli explains that

the cubic is an example of
the well-known irreducible case; a case which is so much the more
remarkable, as the three roots are then always real. We cannot here
make use of Cardans formula, except by applying the methods of approximation. . . . In the present work of Euler, we are not to look for
all that might have been said on the direct and approximate resolutions
of equations. He had too many curious and important objects, to dwell
long upon this.
We have dwelled long upon this indeed. It is time to move beyond cubics.

i
i

i
IrvingBook 2013/5/22 15:39 page 143 #159

6
Quartic Polynomials
We learned in Section 3.5 that Lodovicio Ferrari discovered a way to solve
quartic equations not long after Cardanos work on cubic equations, and that
Cardano presented Ferraris method in Ars Magna [12]. We will take a brief
look at Ferraris approach, then turn to the approach of Rene Descartes.
Both reduce the solving of a reduced quartic equation to the determination
of the roots of an auxiliary cubic polynomial. A closer look at Descartes
solution will allow us to obtain a formula due to Euler that expresses the
roots of the reduced quartic in terms of the roots of the auxiliary cubic. In
turn, Eulers solution leads to a formula for the discriminant of a quartic
polynomial. We will then have our final look at the effectiveness of coefficients as a tool to glean information about a polynomials roots.

6.1 Reduced Quartics

From our study of graphs of polynomials in Section 1.6, we have an idea of
what the graph of a quartic polynomial looks like and the number of roots
it may have. Assuming the quartic is monic, its graph will fall from infinity,
turn one or three times, then rise to infinity. The number of real roots will
depend on the number of turning points and their location above, on, or
below the x-axis. For instance, if there is one turning point, then we are in a
situation similar to that of quadratic polynomials, with two, one, or no real
roots depending on whether the turning point is below, on, or above the xaxis. If there are three turning points, there are many possibilities. We will
not discuss them all, satisfying ourselves with those that can be illustrated
through one family of examples. The interested reader can use them as a
starting point and complete a survey of the options.
143

i
i

i
IrvingBook 2013/5/22 15:39 page 144 #160

144

6. Quartic Polynomials

Figure 6.1. Graph of y D x 4

x3

4x 2 C 4x

Consider the infinite family of quartic polynomials x 4 x 3 4x 2 C4x C

e, where e takes on all real number values. Figure 6.1 depicts the graph of
y D x 4 x 3 4x 2 C 4x, corresponding to e D 0. We see from the graph
that this polynomial has four distinct real roots. Or, we see it directly from
x4

x3

4x 2 C 4x D x.x C 2/.x

1/.x

2/:

Imagine allowing e to increase from negative values of e with large absolute value to 0 and on through positive values of e. As e grows, the graph
moves upward in the plane. For a negative value of e that is large in absolute
value, the graph will meet the x-axis in two points and the quartic has two
real roots. As e increases, it assumes the value for which the graphs one
local maximum is on the x-axis, the case of three real roots. As e increases
further, the local maximum rises above the x-axis, which then intersects the
graph in four points, as illustrated in Figure 6.1. This corresponds to the existence of four real roots. Then e assumes the value at which the higher of
the two local minima makes contact with the x-axis, giving us once again
a quartic with three real roots. As e continues to increase, the higher local
minimum rises above the x-axis and we have just two real roots. This remains the case until e assumes the value at which the lower local minimum
makes contact with the x-axis. In this case, the quartic has one real root. For
larger e the graph lies entirely above the x-axis and the quartic has no real
roots.
Before analyzing quartic polynomials in general, we need some elementary facts that are similar to those we obtained to begin our treatment of
cubics.
Exercise 6.1.

Given A and B, we know that

.A C B/2 D A2 C 2AB C B 2

i
i

i
IrvingBook 2013/5/22 15:39 page 145 #161

145

6.1. Reduced Quartics

and
.A C B/3 D A3 C 3A2 B C 3AB 2 C B 3 :

Show that

.A C B/4 D A4 C 4A3B C 6A2B 2 C 4AB 3 C B 4 :

A quartic polynomial of the form
z 4 C qz 2 C r z C s
is called a reduced quartic.
Exercise 6.2.

Let
f .x/ D x 4 C bx 3 C cx 2 C dx C e;

for real numbers b, c, d , and e. Given another real number a, lets see what
happens under the change of variable x D z C a, or z D x a.
(i) Substitute z Ca for x and obtain a polynomial g.z/ in the new variable
z. Write it as
z 4 C Bz 3 C C z 2 C Dz C E
and obtain formulas for the coefficients B, C , D, and E in terms of a
and the coefficients b, c, d , and e.

(ii) There is a choice of a for which B D 0, giving a new polynomial g.z/

that is reduced. Verify that g.z/ is the reduced polynomial described
in Theorem 6.1.
(iii) What is the relation between a root of f .x/ and a root of g.z/?
(iv) Conclude that solving g.z/ D 0 gives solutions of f .x/ D 0.
Theorem 6.1. Let b, c, d , and e be real numbers. Given a quartic polynomial x 4 C bx 3 C cx 2 C dx C e, the change of variable x D z b=4 yields
the reduced quartic polynomial


 3



3b 2
b
bc
3b 4
b 2 c bd
4
2
z C
Cc z C
C d zC
C
Ce :
8
8
2
256
16
4
Exercise 6.3. What is the reduced quartic equation associated to the quartic equation
x 4 C 4x 3 2x C 4 D 0

How are the solutions of the reduced quartic equation related to the solutions of the original quartic equation?

i
i

i
IrvingBook 2013/5/22 15:39 page 146 #162

146

6.2

6. Quartic Polynomials

Ferraris Method

A method for solving quartic equations was first discovered Lodovico Ferrari. In recognition of the historical importance of Ferraris approach, we
devote this section to it. However, we will not use his approach in the remainder of the chapter. Thus, this section can be omitted.
Suppose that we wish to solve the reduced quartic equation
z 4 C qz 2 C r z C s D 0
with real coefficients q, r , and s. Moving the lower-degree terms to the right
to obtain
z 4 C qz 2 D r z s;
we can regard the left side as a quadratic polynomial in z 2 and complete the
square by adding qz 2 C q 2 to both sides. This yields
.z 2 C q/2 D qz 2 C q 2

rz

s:

Ferraris idea is to introduce an auxiliary term t, which is best thought

of not as a variable but as a to-be-determined real number, and use it to
change the left side from .z 2 C q/2 to .z 2 C q C t/2 . The goal then becomes
to find a value of t that makes the right side a square as well, allowing us
to take square roots of both sides and reduce to a simpler problem. Lets do
this.
To begin, we add 2.z 2 C q/t C t 2 to both sides of .z 2 C q/2 D qz 2 C
2
q
r z s. The left side becomes a square and we obtain the equation
.z 2 C q C t/2 D qz 2 C q 2

rz

s C 2.z 2 C q/t C t 2 :

Keep in mind that q, r , and s are specific real numbers, as is t, although

we dont yet know what t is. The only variable is z. It is natural, then, to
collect terms on the right side of the equation so that it is exhibited as a
quadratic polynomial in z with coefficients involving the constants q, r , s,
and t:
.z 2 C q C t/2 D .q C 2t/z 2

r z C .q 2

s C 2qt C t 2 /:

Our goal is to find a value of t for which the right side becomes the square
of a linear polynomial in z.
We know that a monic quadratic polynomial z 2 C bz C c is the square
of a linear polynomial precisely when it has a real root of multiplicity 2, or
equivalently, when its discriminant b 2 4c is 0. A polynomial of the form

i
i

i
IrvingBook 2013/5/22 15:39 page 147 #163

147

6.2. Ferraris Method

az 2 C bz C c, with a 0, will factor in the same way as z 2 C .b=a/z C c=a.

Thus, it will be the square of a linear polynomial precisely when .b=a/2
4.c=a/ D 0, or b 2 4ac D 0.
We can apply this to
.q C 2t/z 2

r z C .q 2

s C 2qt C t 2 /

viewed as a quadratic polynomial in z. It will be the square of a linear

polynomial in z if t is chosen so that
. r /2

4.q C 2t/.q 2

s C 2qt C t 2 / D 0:

Rewriting the expression on the left side as a cubic polynomial in the unknown constant t and multiplying by 1, we find that t must satisfy
8t 3 C .20q/t 2 C .16q 2

8s/t C .4q 3

4qs

r 2 / D 0:

Choosing such a t ensures that the quadratic polynomial in z is the square

of a degree-one polynomial, so that both sides of the earlier equation
.z 2 C q C t/2 D .q C 2t/z 2

r z C .q 2

s C 2qt C t 2 /

are squares.
Using the techniques we have learned for solving cubic equations, we
can find t, then factor the quadratic in z as the square of a degree-one polynomial dz C e. In this way, we obtain the quartic equation
.z 2 C q C t/2 D .dz C e/2 :
Taking square roots on both sides, we obtain quadratic equations
z 2 C q C t D dz C e
and
z 2 C q C t D dz

e:

Solving for z yields four solutions to the original quartic equation. We (and
Ferrari) have reduced the problem of solving the original quartic equation
z 4 C qz 2 C r z C s D 0 to that of solving the cubic equation in t.
Lets try Ferraris approach in an example, one we will pursue only to
the point of obtaining the reduced cubic equation that needs solving.
Exercise 6.4.

Solve the equation

z 4 C 6z 2

60z C 36 D 0:

i
i

i
IrvingBook 2013/5/22 15:39 page 148 #164

148

6. Quartic Polynomials

(i) Use Ferraris method to show this can be done by finding t satisfying
t 3 C 15t 2 C 36t

450 D 0:

(ii) Change variables to transform the cubic equation to a reduced cubic

equation,
u3 39u 380 D 0:
(iii) Find a solution using Cardanos formula.
(iv) Explain how to obtain a solution to the original quartic equation.
Lets work out the next example to conclusion. It is missing both degree
three and degree two terms, so that the cubic equation to solve is already
reduced.
Exercise 6.5.

Solve the quartic equation

z4

12z C 3 D 0:

(i) Following Ferraris procedure with q D 0, r D 12, and s D 3,

introduce a new variable t and show that the equation takes the form
.z 2 C t/2 D 2tz 2 C 12z C .t 2

3/:

(ii) Show that the right side of this equation will be the square of a degreeone polynomial in z if t is a solution of the cubic equation
8t 3

24t

144 D 0;

t3

3t

18 D 0:

or
(iii) Use Cardanos formula to show that one solution is
q
q
p
p
3
3
t D 9 C 4 5 C 9 4 5:

p
(iv) Use the method of Exercise
3.10 to guess that a cube root of 9 C 4 5
p
has the form a C b 5, obtain two equations in a and b, and guess
values of a and p
b, leading to the p
discovery that the two cube roots we
want are 3=2 C 5=2 and 3=2
5=2. Alternatively, guess that there
is some small positive integer solution to the cubic equation, without
reference to Cardanos formula, and find it.

i
i

i
IrvingBook 2013/5/22 15:39 page 149 #165

149

6.3. Descartes Method

(v) Substitute t into the equation in z and t at the start of the exercise to
obtain
.z 2 C 3/2 D 6z 2 C 12z C 6:
(vi) As Ferraris method ensures, the right side of this equation is the square
of a degree-one polynomial in z:
6z 2 C 12z C 6 D 6.z C 1/2 :
(vii) Using this, write the equation as
.z 2 C 3/2 D 6.z C 1/2 :
(viii) Take square roots on both sides to get
p
p
z2
6z C .3
6/ D 0
and
z2 C

p
p
6z C .3 C 6/ D 0:

(ix) Solve them for z, thereby obtaining four solutions to

z4

12z C 3 D 0:

6.3 Descartes Method

We turn in this section to an approach to solving reduced quartic equations
introduced by Rene Descartes. The idea is to factor a reduced quartic polynomial z 4 C qz 2 C r z C s as a product of quadratic polynomials with real
coefficients. If we can do this, then we can find the quartics roots by applying the quadratic formula to its quadratic factors.
Lets first treat a special case, one for which we will find the desired
factorization, although the case is sufficiently simple that we could find the
roots directly.
Exercise 6.6. Let q and s be real numbers and consider the polynomial
z 4 C qz 2 C s.
(i) Assume that q 2
l /.z 2 C m/.

4s  0. Show that z 4 C qz 2 C s factors as .z 2 C

i
i

i
IrvingBook 2013/5/22 15:39 page 150 #166

150

6. Quartic Polynomials

(ii) Assume that q 2 4s < 0 and observe that this implies s > 0. Verify
that a factorization of the form .z 2 C l /.z 2 C m/ cant exist. Therefore,
to show that z 4 C qz 2 C s factors as a product of quadratic polynomials
with real coefficients, we will instead seek a factorization of the more
general form
z 4 C qz 2 C s D .z 2 C kz C l /.z 2 C k 0 z C m/;
where the coefficients k, k 0 , l , and m are real and k and k 0 are nonzero. Our search will occupy the remainder of the exercise.
(iii) Multiply the two quadratic polynomials, obtain expressions for the coefficients of each power of z, and show that for equality to hold, we
must have k 0 D k and l D m. Thus, we can simplify our task and
look for a factorization of the form
z 4 C qz 2 C s D .z 2 C kz C l /.z 2

kz C l /:

(iv) Show next that for equality to hold, k and l must satisfy l 2 D s and
2l D k 2 C q. We view these as equations in the unknowns k and l ,
with q and s regarded as known constants.
(v) Square both sides of 2l D k 2 C q, substitute s for l 2, and show that k
must satisfy
k 4 C 2qk 2 C .q 2 4s/ D 0:
(vi) This is a quadratic equation in k 2 . Set j D k 2 and write it as
j 2 C 2qj C .q 2

4s/ D 0:

(vii) Check that the discriminant of j 2 C 2qj C .q 2 4s/ as a quadratic

polynomial in j is 16s, which is positive. Deduce that the quadratic
polynomials roots are real and distinct. Using the assumption that q 2
4s < 0, deduce that there is a positive real root. (Recall Theorem 5.13.)
(viii) Let k and k represent the square roots of the positive root. Conclude
that k and k give a value for l and allow us to factor z 4 C qz 2 C s as
a product of quadratic polynomials with real coefficients.
Exercise 6.6 shows that a quartic polynomial of the form z 4 C qz 2 C s
factors as a product of quadratic polynomials. We can use the factorization
to find the roots of z 4 C qz 2 C s. Alternatively, we can find them directly.

i
i

i
IrvingBook 2013/5/22 15:39 page 151 #167

151

6.3. Descartes Method

Exercise 6.7.
qz 2 C s are

Use the quadratic formula to show that the roots of z 4 C

p
q 2 4s
:
2

The argument in Exercise 6.6 showing when q 2 4s < 0 that z 4 C

qz C s can be factored as a product of quadratic polynomials with real
coefficients can be extended to quartic polynomials with non-zero linear
term.
2

Exercise 6.8. Let q, r , and s be real numbers, with r 0, and consider

the polynomial z 4 C qz 2 C r z C s.
(i) We wish to find real numbers k, k 0 , l , and m for which z 4 C qz 2 C
r z C s factors as
.z 2 C kz C l /.z 2 C k 0 z C m/:
Show that for this to hold, we must have k 0 D
for a factorization of the form

k. Thus we can look

z 4 C qz 2 C r z C s D .z 2 C kz C l /.z 2

kz C m/:

Check also, since r 0, that k must be non-zero.

(ii) Multiply the two quadratic polynomials and show that for equality to
hold, k, l , and m must satisfy
l Cm

k 2 D qI k.m

l / D r I l m D s:

(iii) Since k cant be zero, we are free to divide by it. Do so in the second
equation and obtain from the first two equations the new equations
m C l D q C k2I m

lD

r
:
k

Using these, obtain

2m D q C k 2 C

r
I 2l D q C k 2
k

r
:
k

Conclude that l and m are determined once k is known.

(iv) We can multiply both sides of the equation l m D s by 4 to obtain
.2l /.2m/ D 4s. Substitute the expressions for 2l and 2m and obtain
k 6 C 2qk 4 C .q 2

4s/k 2

r 2 D 0:

i
i

i
IrvingBook 2013/5/22 15:39 page 152 #168

152

6. Quartic Polynomials

(v) This is a cubic equation in k 2 . Set j D k 2 and write it as

j 3 C 2qj 2 C .q 2

4s/j

r 2 D 0:

(vi) Deduce from Theorem 5.17 that the cubic equation in j has a positive
real solution.
(vii) Let k and k represent the square roots of this solution. Conclude that
k and k give values for l and m and allow us to factor z 4 C qz 2 C
r zCs as a product of quadratic polynomials with real coefficients. The
roots of the quadratic polynomials yield the solutions of the original
quartic equation.
(viii) If s D 0, then solving the original quartic equation reduces to solving
z 3 C qz C r D 0. The degree-six polynomial in k becomes k 6 C
2qk 4 C q 2 k 2 r 2 , which factors as .k 3 C qk C r /.k 3 C qk r /.
Hence, setting it equal to 0 and solving for k is essentially the same as
solving the original equation.
We have obtained a procedure for finding roots of quartic polynomials.
First we pass from a quartic polynomial
x 4 C bx 3 C cx 2 C dx C e
to its associated reduced quartic polynomial
z 4 C qz 2 C r z C s:
If r 0, we write the cubic polynomial
j 3 C 2qj 2 C .q 2

4s/j

r 2;

which is called the resolvent cubic of the quartic z 4 C qz 2 C r z C s. If r D 0

and q 2 4s < 0, we can work with the same cubic, or its quadratic factor
j 2 C 2qj C .q 2 4s/. We find a positive real root of the resolvent cubic,
take its square roots, use them to factor the reduced quartic as a product
of quadratic polynomials with real coefficients, find their roots, and thereby
obtain the roots of the original quartic polynomial. If r D 0 and q 2 4s  0,
we obtain a factorization of z 4 C qz 2 C r z C s directly as a product of
quadratics, using the quadratic formula.
We restate some of this discussion in the next theorem.

i
i

i
IrvingBook 2013/5/22 15:39 page 153 #169

153

6.3. Descartes Method

Theorem 6.2. Let z 4 C qz 2 C r z C s be a reduced quartic polynomial with

real coefficients. Assume that r 0 or that r D 0 and q 2 4s < 0. Let
j 3 C 2qj 2 C .q 2

4s/j

r2

be the given quartics associated resolvent cubic polynomial and let k and
k be square roots of a positive real root of the resolvent cubic. Then z 4 C
qz 2 C r z C s factors as



1
r
1
r
z 2 C kz C
q C k2
z 2 kz C
q C k2 C
:
2
k
2
k

The roots of z 4 Cqz 2 Cr zCs are the roots of the two quadratic polynomials
in the factorization.
The procedure for finding roots of quartic polynomials can be long and
complicated, but in principle it works. The hardest part is likely to be the
calculation of roots of the resolvent cubic, a problem we already understand.
Lets try the procedure on some examples. They are designed so that
the polynomials arising as resolvent cubics have positive real roots that are
easy to find. Thus, in solving the quartic equations, we can focus on the
new features of the procedure and not get distracted by the now-familiar
difficulties of the cubic.
Exercise 6.9.

Find the four solutions of

z4

3z 2 C 6z

2 D 0:

To do so, write the resolvent cubic, find a real root by guessing (squares of
integers are good guesses), take its two square roots, factor z 4 3z 2 C6z 2
as a product of two quadratic polynomials, and find their roots.
Exercise 6.10. Find the four solutions of
z4

2z 2

8z

3D0

by following the procedure outlined in Exercise 6.9.

Our third example is the quartic equation of Exercise 6.5. We will solve
it using Descartes method.
Exercise 6.11. Find the four solutions of
z4

12z C 3 D 0:

i
i

i
IrvingBook 2013/5/22 15:39 page 154 #170

154

6. Quartic Polynomials

In this case, since there is no square term, the resolvent cubic will be a
reduced cubic, so that Cardanos formula can be applied directly. Find a
real root by using Cardanos formula or by guessing. (Guessing may not
lead to a root that is the square of an integer, but it may lead to a positive
integer.) Then proceed as in the preceding exercises. Compare the solutions
to the result in Exercise 6.5 to make sure they are the same.
We have found that a reduced quartic polynomial
z 4 C qz 2 C r z C s
with real coefficients factors as a product of quadratic polynomials with real
coefficients. From this, we can obtain information on the nature of the roots
of any quartic polynomial.
Theorem 6.3. Any quartic polynomial f .x/ with real coefficients factors
as a product of quadratic polynomials with real coefficients. Exactly one of
the following possibilities occurs for the roots of f .x/:
(i) f .x/ has four distinct real roots.
(ii) f .x/ has two distinct real roots and a pair of non-real complex conjugate roots.
(iii) f .x/ has no real roots, but two pairs of distinct non-real complex conjugate roots.
(iv) f .x/ has repeated roots.
Exercise 6.12. Deduce Theorem 6.3 for reduced quartic polynomials from
Exercise 6.6 and Theorem 6.2. Then explain why it holds for quartic polynomials in general.

6.4

Eulers Formula

We can refine Descartes method of solving a reduced quartic equation (Exercise 6.8 and Theorem 6.2) to obtain a formula for its solutions in terms of
the solutions of the associated resolvent cubic equation. The formula is due
to Euler, who obtained it directly rather than as a consequence of Descartes
result. We will be content to derive it from Descartes factorization. Eulers
approach will be discussed at the end of Section 6.8.
We consider once again the reduced quartic equation
z 4 C qz 2 C r z C s D 0:

i
i

i
IrvingBook 2013/5/22 15:39 page 155 #171

155

6.4. Eulers Formula

In light of the explicit description of its solutions in Exercise 6.7 when r D

0, lets assume that r 0. Suppose the three solutions of the resolvent cubic
equation
j 3 C 2qj 2 C .q 2

4s/j

r 2 D 0:

are j1, j2 , and j3 . Our goal is to describe the roots of z 4 C qz 2 C r z C s in

terms of j1, j2 , and j3 .
We have already shown that at least one of the resolvent cubics roots
is positive and real. Lets assume it is j1 , and let k1 and k1 be its square
roots. Because j1 is positive, k1 and k1 are real. The other two roots of
the resolvent cubic, j2 and j3 , may be real or may be complex conjugates.
Whether or not j2 and j3 are real, their square roots are opposites of
each other. Thus we can write k2 and k2 for the square roots of j2 and k3
and k3 for the square roots of j3 .
Do not be fooled by the notation. Seeing the pair ki and ki , you should
not assume that ki is positive or that ki is negative. In fact, they may not
even be real. The notation should convey only that whatever they are, each
is the opposite of the other.
Exercise 6.13.
sion.

Use the notation and assumptions of the preceding discus-

(i) Explain why there are eight possible choices for the triple of numbers
.k1 ; k2; k3 /, depending on choice of signs.
(ii) Show that
j1 C j2 C j3 D 2q; j1 j2j3 D r 2 :
(Hint: Use Theorem 5.14 on how the roots and coefficients of a cubic
polynomial are related.)
(iii) Deduce that the product k1 k2 k3 equals either r or r , depending on
which square roots of the ji s are chosen as the ki s. More precisely,
deduce that four of the eight choices of .k1 ; k2 ; k3/ result in k1 k2 k3 D
r and that the other four choices result in k1 k2 k3 D r .
Exercise 6.14. Continue within the setting of Exercise 6.13. For definiteness, lets fix the choice of triple of square roots .k1 ; k2 ; k3 / to be one of
the four satisfying k1 k2 k3 D r .
(i) Given our choice of k1 as one of the square roots of the positive real
number j1 , we obtain from Theorem 6.2 the factorization of z 4 Cqz 2 C

i
i

i
IrvingBook 2013/5/22 15:39 page 156 #172

156

6. Quartic Polynomials

r z C s as


1
2
z C k1 z C
q C k12
2

r
k1

 
z2



1
r
2
k1 z C
q C k1 C
:
2
k1

Use the quadratic formula to obtain expressions for the roots of



1
r
z 2 C k1 z C
q C k12
:
2
k1
Show that
zD

k1
1

2
2

k12


2 q C k12


r
:
k1

(ii) Use the formulas from Exercise 6.13 for 2q and r in terms of the ki s
as well as the choice of the triple .k1 ; k2; k3 / to satisfy k1 k2 k3 D r
to write this equality as
zD

k1
1p

.k2
2
2

k3 /2 :

(iii) Conclude that the roots of the quadratic factor are

zD

1
. k1 C k2
2

k3 /

and z D

1
. k1
2

k2 C k3 /:

(iv) Follow the same procedure to find the roots of



1
r
2
2
z
k1 z C
q C k1 C
2
k1
and show that you get
zD

1
.k1 C k2 C k3 /
2

and

zD

1
.k1
2

k2

k3 /:

(v) Conclude that once the three roots j1, j2 , j3 of the resolvent cubic
polynomial are found, the roots of the quartic polynomial can be expressed in terms of square roots of j1 , j2 , and j3 , as described in Theorem 6.4.
Theorem 6.4. Let q, r , and s be real numbers with r 0. Let j1 , j2 , and
j3 be the roots of the resolvent cubic polynomial
j 3 C 2qj 2 C .q 2

4s/j

r2

i
i

i
IrvingBook 2013/5/22 15:39 page 157 #173

157

6.5. The Discriminant

associated with the quartic polynomial

z 4 C qz 2 C r z C s:
Then the roots of z 4 C qz 2 C r z C s are the four sums
p
p 
1 p
j1 C j2 C j3
2
p p p
arising from the choices of square roots j1 , j2 , j3 that satisfy
p
p
p
j1
j2
j3 D r:

Lets solve two quartic equations using Theorem 6.4. The equations
have been chosen so that the roots of the resolvent cubic polynomials are
easily found by trial and error.
Exercise 6.15. Find all four solutions of
p
z 4 3z 2 C 6z

1
D0
2

and all four solutions of

z4

3z 2

p
6z

1
D 0:
2

In the last exercise, the two quartic equations have the same resolvent
cubic equation. This is a special case of a general phenomenon.
Exercise 6.16. Explain why the quartic equations
z 4 C qz 2 C r z C s D 0
and
z 4 C qz 2

rz C s D 0

have the same resolvent cubic equation. Describe how the solutions of the
first are related to the solutions of the second.

6.5 The Discriminant

The discriminant of a quadratic or cubic polynomial is defined as the square
of the product of differences of the roots. The discriminant of a quartic
polynomial
x 4 C bx 3 C cx 2 C dx C e

i
i

i
IrvingBook 2013/5/22 15:39 page 158 #174

158

6. Quartic Polynomials

is defined in the same way. Suppose its roots are r1 , r2 , r3 , and r4 . Then its
discriminant is the product
.r1

r2 /2 .r1

r3 /2 .r1

r4 /2 .r2

r3 /2 .r2

r4/2 .r3

r4 /2 :

Because the root differences are squared, we get the same result regardless
of how the roots are ordered.
For quadratic and cubic polynomials, the discriminant is important for
two reasons. First, its sign gives us information on the nature of the roots:
how many are real numbers and how many are non-real complex numbers.
Second, there is a formula expressing the discriminant in terms of the coefficients of the polynomial, allowing us to calculate the discriminant from
the coefficients and thereby obtain information about the roots from the coefficients alone.
We will show for a quartic polynomial that the discriminant gives information on the nature of the roots, although not as complete as in the
quadratic and cubic cases, and that it can be calculated from the coefficients.
Exercise 6.17. In this exercise, we will find out what the discriminant of
a quartic polynomial tells us about the nature of its roots. Let f .x/ be the
polynomial, with discriminant , and let r1, r2 , r3 , and r4 be its roots.
(i) Check that D 0 precisely when at least two of the roots coincide.
(ii) Assume in the remainder of the exercise that the four roots are distinct, so that 0. From Theorem 6.3, there are three possibilities
for the roots: all are real, two are real and two are non-real complex
conjugates, or there are two pairs of non-real complex conjugates.
(iii) Show that if all four roots are real, then > 0.
(iv) Suppose two roots are real and two are complex conjugates. Show that
< 0. (Hint: The product of a non-zero complex number and its
conjugate is a positive real number. Name the four roots r , s, a C
bi , and a bi , where r , s, a, and b are real numbers. There are six
root differences. Four of the six differences occur as pairs of conjugate
complex numbers, so that their products are real. One of the remaining
root differences is real and the other is pure imaginary. From this it
follows that is negative.)
(v) Suppose that the four roots occur in two pairs of non-real complex
conjugates. Show that > 0. (Hint: Pair four of the differences so
that they are conjugates and their products are real. The other two root
differences should be pure imaginary.)

i
i

i
IrvingBook 2013/5/22 15:39 page 159 #175

159

6.5. The Discriminant

(vi) Show that the converses of the statements in (iii)(v) hold; that is,
prove Theorem 6.5. (Hint: As in Exercises 4.7 and 5.7, these follow
from the statements themselves by an elementary logical argument.)
Theorem 6.5. Let f .x/ be a polynomial of degree 4 with discriminant .
(i) If is positive, then f .x/ has four distinct roots. Either they are all
real or they are all non-real, in which case they form two complex
conjugate pairs.
(ii) If D 0, then f .x/ has a root occurring with multiplicity at least 2.
(iii) If is negative, then f .x/ has four distinct roots. Two are real and
two form a complex conjugate pair.
When the discriminant is positive, Theorem 6.5 does not completely
settle the nature of the roots. Theorem 6.9 will describe how to make further
use of the coefficients to determine whether the roots are all real or all nonreal.
We wish to calculate in terms of the coefficients of the polynomial.
We begin with the special case of a reduced quartic with no linear term.
Exercise 6.18. Let q and s be real numbers. Use the description of the
roots of
z 4 C qz 2 C s
in Exercise 6.7 to calculate the discriminant, obtaining Theorem 6.6.
Theorem 6.6. Let q and s be real numbers. The discriminant of z 4 Cqz 2 Cs
is given by
D 16s.q 2

4s/2 D

128q 2s 2 C 16q 4 s C 256s 3:

The key to computing the discriminant of a general reduced quartic

polynomial is to relate it to the discriminant of the quartics resolvent cubic.
Exercise 6.19. Let q, r , and s be real numbers, with r 0.

(i) Using the notation we employed in Exercise 6.14, write the roots r1 ,
r2 , r3 , and r4 of z 4 C qz 2 C r z C s as halves of sums of ki . Show
that the six root differences, up to sign, are
k1 C k2 ; k1 C k3 ; k2 C k3 ;
k1

k2 ; k1

k3 ; k2

k3 :

i
i

i
IrvingBook 2013/5/22 15:39 page 160 #176

160

6. Quartic Polynomials

(ii) Deduce that

D .k12

k22 /2 .k12

k32 /2 .k22

k32 /2 :

(iii) The ki s are the square roots of the solutions j1, j2 , j3 of the resolvent
cubic equation. Using the ji s, write the equation for as
D .j1

j2 /2 .j1

j3 /2 .j2

j3 /2 :

Since the ji s are by definition the roots of the resolvent cubic polynomial, the product on the right side is in fact the discriminant of the
resolvent cubic.
Since the discriminant of z 4 C qz 2 C r z C s (r 0) coincides with the
discriminant of its resolvent cubic, we can compute it using our results on
discriminants of cubics. Lets do this in a simple special case first.
Exercise 6.20. Let r and s be real numbers with r 0.

(i) Check that the resolvent cubic of z 4 C r z C s is the reduced cubic

j3

4sj

r 2:

(ii) Use the formula for the discriminant of a reduced cubic polynomial
y 3 C py C q in terms of p and q to show that its discriminant is
256s 3
(iii) Deduce that 256s 3

27r 4 :

27r 4 is the discriminant of z 4 C r z C s also.

Exercise 6.21. Compute the discriminants of the quartic polynomials below and say what you can about their number of real roots.
(i) z 4 C z C 1.
(ii) z 4 C z

1.

(iii) z C z C s. (The answer depends on s.)

We now obtain a formula for the discriminant of a general reduced quartic polynomial.
Exercise 6.22. Let q, r , and s be real numbers, with r 0.
(i) The resolvent cubic of z 4 C qz 2 C r z C s has coefficients 2q, q 2 4s,
and r 2. Use the formula in Theorem 5.6 for the discriminant of x 3 C
bx 2 C cx C d , substitute for b, c, and d , and obtain a complicated
expression in q, r , and s for the discriminant of the resolvent cubic.

i
i

i
IrvingBook 2013/5/22 15:39 page 161 #177

161

6.5. The Discriminant

(ii) Conclude that it is also the discriminant of the reduced quartic polynomial z 4 C qz 2 C r z C s with which we began.
(iii) Expand and simplify the expression in q, r , and s. Combine with Theorem 6.6 to conclude that Theorem 6.7 holds.
Theorem 6.7. The discriminant of the reduced quartic polynomial z 4 C
qz 2 C r z C s is given by
D 144qr 2s

128q 2s 2

4q 3 r 2 C 16q 4 s

27r 4 C 256s 3:

Exercise 6.23. Compute the discriminants of the cubic polynomials below.

What can be said about their number of real roots?
(i) z 4

3z 2 C 6z

(ii) z 4

2z 2

(iii) z

2.

8z 3.
p
3z C 6z 12 .
2

The discriminant of an arbitrary quartic polynomial can be computed,

although the process is laborious.
Exercise 6.24. Consider the quartic polynomial
x 4 C bx 3 C cx 2 C dx C e:
From Exercise 6.2 and Theorem 6.1, the change of variable x D y .b=4/
transforms x 4 C bx 3 C cx 2 C dx C e into a polynomial of the form z 4 C
qz 2 C r z C s.
(i) Verify, as in Exercise 5.10, that although the roots of x 4 C bx 3 C
cx 2 C dx C e and the roots of z 4 C qz 2 C r z C s are not the same,
their differences are.

(ii) Deduce that the discriminant of x 4 C bx 3 C cx 2 C dx C e is the same

as the discriminant of z 4 C qz 2 C r z C s.
(iii) Conclude that a formula for the discriminant of x 4 Cbx 3 Ccx 2 CdxCe
in terms of b, c, d , and e can be found by substituting into the formula
for the discriminant of z 4 C qz 2 C r z C s the expressions given by
Theorem 6.1 for q, r , and s in terms of b, c, d , and e.
(iv) Carry out this calculation, simplify, and conclude that Theorem 6.8
holds.

i
i

i
IrvingBook 2013/5/22 15:39 page 162 #178

162

6. Quartic Polynomials

Theorem 6.8. The discriminant of the quartic polynomial x 4 C bx 3 C

cx 2 C dx C e is given by
D 18bcd 3 C 18b 3cde
C 144b 2ce 2
4c 3d 2

6.6

80bc 2de

128c 2 e 2

6b 2 d 2 e C 144cd 2e

192bde 2 C b 2 c 2 d 2

4b 2c 3 e C 16c 4e

27d 4

4b 3 d 3

27b 4e 2 C 256e 3:

The Nature of the Roots

We saw in Theorem 6.5 that the discriminant of a quartic polynomial gives

us partial information on the nature of its roots. In this section, we will
do better by supplementing the discriminant with other expressions in the
coefficients. We know for example that a quartic with positive discriminant
has either four distinct real roots or four distinct non-real roots occurring in
complex conjugate pairs. We will refine this, then turn to a closer analysis
of quartic polynomials with zero discriminant.
Exercise 6.25. Consider the quartic polynomial
z 4 C qz 2 C s
and assume its discriminant is positive. We wish to determine, in terms
of q and s, whether the roots are all real or all non-real.
(i) Use the fact that > 0 and the formula for (Theorem 6.6) to deduce
that s > 0.
(ii) Deduce that the roots of z 4 C qz 2 C s are non-zero, and that they
will all be real precisely when the roots of the quadratic polynomial
t 2 C qt C s are real and positive.
(iii) Using Theorem 5.13 and the fact that s > 0, deduce that this occurs
precisely when q < 0 and q 2 4s > 0.
Next we examine reduced quartic polynomials with non-zero linear term.
Exercise 6.26. Consider the quartic polynomial
z 4 C qz 2 C r z C s:
Assume that its discriminant is positive and that r 0.

i
i

i
IrvingBook 2013/5/22 15:39 page 163 #179

163

6.6. The Nature of the Roots

(i) Deduce, since is also the discriminant of the resolvent cubic polynomial
j 3 C 2qj 2 C .q 2 4s/j r 2 ;
that its roots j1, j2 , and j3 are real and distinct.
(ii) From Theorem 5.14, deduce that
j1 j2 j3 D r 2 :
(iii) Conclude that either all the roots ji are positive or one is positive and
the other two are negative.
(iv) From Theorem 5.16, deduce that the ji are all positive precisely when
q < 0 and q 2

4s > 0:

(v) If the ji are all positive, then their square roots ki are all real numbers. Conclude from Theorem 6.4 that the four roots of the quartic
z 4 C qz 2 C r z C s are all real.
(vi) Suppose j1 is positive and j2 and j3 are negative. Conclude that the
two square roots k1 are real while k2 and k3 are pure imaginary.
Show in this case that the four roots of z 4 C qz 2 C r z C s are all
non-real complex numbers.
(vii) Conclude that the roots of z 4 C qz 2 C r z C s are all real precisely when
q < 0 and q 2 4s > 0.
Combining the results of Exercises 6.25 and 6.26, we obtain:
Theorem 6.9. The roots of a reduced quartic polynomial z 4 C qz 2 C r z C s
with positive discriminant are all real if q < 0 and q 2 4s > 0, and all
non-real otherwise.
This result is due to Lagrange, who stated it in his Traite de la Resolution

des Equations
Numeriques de Tous les Degres [39, p. 68], which first appeared in 1798.
Exercise 6.27. For each of the quartic polynomials, use the discriminant
and the coefficients to decide how many roots are real.
(i) z 4 C z C 1.

(ii) z 4 C z

1.

i
i

i
IrvingBook 2013/5/22 15:39 page 164 #180

164

6. Quartic Polynomials

(iii) z 4 C z C s. (The answer depends on s.)

(iv) z 4
(v) z

(vi) z 4

3z 2 C 6z
2z

2.

8z 3.
p
3z 2 C 6z 12 .

A version of Theorem 6.9 for arbitrary quartic polynomials is easily

derived. A polynomial x 4 C bx 3 C cx 2 C dx C e has the same discriminant
as its associated reduced quartic polynomial z 4 C qz 2 C r z C s. Moreover,
the two are related through a change of variable by the real number b=4,
as described in Theorem 6.1. Hence, if their shared discriminant is positive,
then both have all real roots or all non-real roots. Which of the two occurs,
according to Theorem 6.9, is determined by the signs of q and q 2 4s, and
these can be expressed in terms of b, c, d , and e, as in Theorem 6.1. This
yields a theorem describing whether the roots of x 4 C bx 3 C cx 2 C dx C e
are all real or all non-real in terms of b, c, d , and e.
Combining this with Theorem 6.5, we can decide from a quartic polynomials coefficients when it has roots that are real and distinct, non-real and
distinct, or a mix of real and non-real but distinct. The remaining possibility,
which occurs when the quartic has zero discriminant, is that of a repeated
root. We conclude our development of the quartic with an examination of
this case, leaving many of the details for the reader to work out.
Once again, we look first at quartics whose linear term is zero, recalling
from Theorem 6.6 that z 4 C qz 2 C s has discriminant 16s.q 2 4s/2 . Hence,
for the discriminant to be 0, either s D 0 or q 2 4s D 0.
Theorem 6.10. Let z 4 C qz 2 C s be a quartic polynomial with zero discriminant.
(i) Assume s D 0. Then z 4 C qz 2 factors as z 2 .z 2 C q/.
(a) If q D 0, then 0 is a root of multiplicity 4.
(b) If q 0, then 0 is a root of multiplicity 2 and the square roots of
q are roots of multiplicity 1, real if q < 0 and pure imaginary if
q > 0.
(ii) Assume s 0, so that q 2
factors as .z 2 C q=2/2 .

4s D 0 and q 0. Then z 4 C qz 2 C s

(a) If q < 0, then z 4 C qz 2 C s has two real roots of multiplicity 2,

each the others opposite.
(b) If q > 0, then z 4 C qz 2 C s has two pure imaginary roots of
multiplicity 2, each the others opposite.

i
i

i
IrvingBook 2013/5/22 15:39 page 165 #181

165

6.6. The Nature of the Roots

Exercise 6.28. Prove Theorem 6.10

To continue, we need the quartic analogue of Theorem 5.14, which describes the coefficients of a cubic polynomial in terms of its roots.
Theorem 6.11. Let x 4 C bx 3 C cx 2 C dx C e be a quartic polynomial with
real coefficients b, c, d , and e, and suppose its roots are r1 , r2 , r3 , and r4 .
Then
b D r1

r2

r3

r4 ;

c D Cr1 r2 C r1 r3 C r1 r4 C r2 r3 C r2r4 C r3r4 ;

d D r1 r2 r3

r1 r2 r4

r1 r3 r4

r2 r3 r4 ;

e D Cr1 r2r3 r4 :

Exercise 6.29. Prove Theorem 6.11. (Hint: Use the factorization of x 4 C

bx 3 C cx 2 C dx C e as .x r1 /.x r2 /.x r3 /.x r4/.)
Theorem 6.11 tells us in particular that the sum of the roots of a reduced
quartic polynomial is 0. We can now complete our analysis of the roots of a
such a polynomial when it has zero discriminant.
Theorem 6.12. Let z 4 C qz 2 C r z C s be a reduced quartic polynomial
with zero discriminant.
(i) If z 4 C qz 2 C r z C s has a root of multiplicity 4, or two roots of
multiplicity 2, or 0 as a root of multiplicity 2, then r D 0 and the roots
are as described in Theorem 6.10.
(ii) Assume r 0. Then z 4 C qz 2 C r z C s has either a non-zero real root
of multiplicity 3 and a real root of multiplicity 1 or a non-zero real root
of multiplicity 2 and two roots of multiplicity 1.
(a) If q < 0 and q 2 4s > 0, the roots are all real. If q 2 C 12s D 0,
there is a non-zero real root a of multiplicity 3 and 3a is a root
of multiplicity 1. If q 2 C 12s 0, there is a non-zero real root a
of multiplicity 2 and there are two real roots of the form a C c
and a c for some non-zero real number c with c 2a.

(b) If it is not the case that q < 0 and q 2 4s > 0, then there is a
non-zero real root a of multiplicity 2 and there are two complex
conjugate roots of the form a C bi and a bi , for some nonzero real number b.

i
i

i
IrvingBook 2013/5/22 15:39 page 166 #182

166

6. Quartic Polynomials

Exercise 6.30. Prove Theorem 6.12.

(i) From Theorem 6.3, if a non-real complex number occurs as a quartics
root, so does its conjugate. Using this and Theorem 6.11, prove (i) by
calculating r in terms of the roots. (Remember, the roots add to 0.)
(ii) Assume r 0. Show that there is a non-zero real root a of multiplicity
2 or 3.
(iii) Show that if all the roots are real, they have the form a; a; aCc; a
c for some non-zero real number c, but if they are not all real, they have
the form a; a; a C bi; a bi for some non-zero real number b.
(iv) Use Theorem 6.11 to calculate q, r and s in terms of the roots. Verify
that q < 0 and q 2 4s > 0 if the roots are all real, but not otherwise.
(v) Compute q 2 C 12s in terms of the roots and show that it is the square
of a simple expression. Check that it cant be 0 unless the roots are all
real, in which case it is 0 precisely when c D 2a. This is exactly the
case of a multiplicity 3 root.
In the case of a quartic polynomial z 4 C qz 2 C r z C s with zero discriminant and r 0, Theorem 6.12 focuses attention on the quantity q 2 C 12s.
Whether it is zero or not determines whether the polynomial has a triple
root or not. A similar result holds for z 4 C qz 2 C s: for its discriminant to
be zero, either s D 0 or q 2 4s D 0. Either way, if q 2 C 12s D 0, then
q D s D 0, and conversely. Thus, q 2 C12s D 0 precisely when z 4 Cqz 2 Cs
has a quadruple root. This yields the following theorem.
Theorem 6.13. The polynomial z 4 C qz 2 C r z C s has a root of multiplicity
3 or 4 precisely when D 0 and q 2 C 12s D 0.
We can use Theorem 6.1 on change of variable to extend our results
on reduced quartic polynomials with zero discriminant to arbitrary quartic
polynomials with zero discriminant. As one example, we obtain:
Theorem 6.14. The polynomial x 4 C bx 3 C cx 2 C dx C e has a root of
multiplicity 3 or 4 precisely when D 0 and c 2 3bd C 12e D 0.
Exercise 6.31. Prove Theorem 6.14.
These theorems provide our final illustration of the theme that a polynomials coefficients encode rich information on its roots.

i
i

i
IrvingBook 2013/5/22 15:39 page 167 #183

167

6.7. Cubic and Quartic Reprise

6.7 Cubic and Quartic Reprise

We close our treatment of cubic and quartic polynomials with three exercises that provide an opportunity to review what we have learned. In each
exercise, the reader is given a list of items to address and asked to write
an essay. The essay should be seamless, treating the items in an integrated
manner rather than separately and in isolation.
Exercise 6.32. Write an essay on solving cubic equations. Illustrate by
computing solutions to
x 3 C 6x 2 C 3x C 18 D 0
and

x3

3x 2

10x C 24 D 0:

Interweave a concrete treatment of the equations with the general discussion, including the following points.
(i) The reduction of arbitrary cubic equations to reduced cubic equations,
so that Cardanos formula can be applied.
(ii) How Cardanos formula can be used to obtain not just one but three
solutions to the reduced cubic equation. In particular, explain which
cube roots to pair in the formula in order to obtain the three solutions.
(iii) How to compute cube roots of non-real complex numbers.
(iv) How to sidestep such computations by relying instead on trigonometric
and inverse trigonometric functions.
Exercise 6.33. Write an essay on solving quartic equations. Illustrate by
computing solutions to the quartic equations
x4
and

x4

4x 3 C 3x 2 C 8x

10 D 0

42x 2 C 64x C 105 D 0:

Interweave a concrete treatment of the equations with the general discussion, including the following points.
(i) The reduction of arbitrary quartic equations to reduced quartic equations, and its application to the first quartic.
(ii) The solving of a reduced quartic equation, such as the one obtained,
by writing the reduced quartic polynomial in it as a product of two

i
i

i
IrvingBook 2013/5/22 15:39 page 168 #184

168

6. Quartic Polynomials

quadratic polynomials with unknown coefficients, obtaining equations

that the coefficients must satisfy, and reducing the solution of the equations to the problem of solving a cubic equation.
(iii) Solving the cubic equation in the example by guessing, using the solution to obtain a factorization of the quartic polynomial as a product
of quadratic polynomials, obtaining solutions to the reduced quartic
equation, and then finding solutions to the original quartic equation.
(iv) A description of the general form of the cubic equation to solve in
studying a reduced quartic polynomial, illustrating this with the second
example and obtaining
j3

84j 2 C 1344j

4096 D 0:

(v) The observation, after guessing a solution to this cubic equation, that
we can use the procedure described for the first quartic to factor the
second one as a product of two quadratic polynomials, followed by
an explanation that we can instead describe the roots of the reduced
quartic in terms of the three solutions of the cubic equation.
(vi) A list of eight candidates for solutions to the original quartic equation
in terms of the three solutions to the cubic equation, and an explanation
of which four to choose. (Provide an explanation that does not depend
on substituting all eight candidates in the original equation.)
(vii) A description of a quartic equation whose solutions are the other four
candidates, and an explanation of why they are solutions.
(viii) A discussion of how these ideas can be used to solve any quartic equation, once a method is available for solving cubic equations.
Exercise 6.34. Write an essay about the role the discriminant plays in understanding the solutions of quadratic, cubic, and quartic equations. Address the following issues.
(i) The definition of the discriminant of a polynomial of degree 2, 3, or 4
in terms of its roots, and the information it gives about the roots.
(ii) The formula for the discriminant of a polynomial of degree 2, 3, or 4
in terms of its coefficients, and how the formulas make it possible to
obtain information about the roots from the coefficients.
(iii) The quadratic formula for the roots of a quadratic polynomial and the
appearance of the discriminant within the formula.

i
i

i
IrvingBook 2013/5/22 15:39 page 169 #185

169

6.8. History

(iv) Cardanos formula for the roots of a reduced cubic polynomial and the
appearance of the discriminant within the formula.
(v) The way in which the sign of the discriminant of a cubic polynomial
affects the calculations in Cardanos formula, with the need to compute
cube roots of non-real complex numbers in certain cases.
(vi) How to combine information from the discriminant of a quartic polynomial with information from the quartics coefficients to refine our
understanding of the nature of the quartics roots.

6.8 History
Lodovico Ferrari was the first person to develop a method for solving quartic equations. We already introduced him in Section 3.5, learning that he
was Cardanos assistant, that Cardano described his quartic solution in Ars
Magna, and that he participated in the famous dispute in 1548 with Tartaglia.
Let us look at Cardanos presentation of Ferraris work before moving on to
contributions of two intellectual giants from the following centuries, Rene
Descartes and Leonhard Euler.
Cardano presents the quartic solution in the penultimate chapter of Ars
Magna, making the following transition [12, p. 237]:
There is another rule, more noble than the preceding. It is Lodovico
Ferraris, who gave it to me on my request. Through it we have all the
solutions for equations of the fourth power, square, first power, and
number, or of the fourth power, cube, square, and number, and I set
them out here in order.
What Cardano sets out are twenty families of quartic equations, depending
on the signs of the coefficients.
After describing a procedure for handling the equations, Cardano works
nine examples. The first example is introduced by [12, p. 239]:
For example, divide 10 into three proportional parts, the product of the
first and second of which is 6. This was proposed by Zuanne de Tonini
de Coi, who said it would not be solved. I said it could, though I did not
yet know the method [for doing so]. This was discovered by Ferrari.
Let the mean be x. The first, then, will be 6=x and the third will be
x 3=6. These equal 10. Multiplying all terms by 6x, we will have
60x D x 4 C 6x 2 C 36:

i
i

i
IrvingBook 2013/5/22 15:39 page 170 #186

170

6. Quartic Polynomials

Cardano spends several pages on the details of this examples solution.

It is the equation of Exercise 6.4. More examples follow, until he comes to
z 4 C 3 D 12z;
which we encountered in Exercise 6.5, and to an example of a non-reduced
quartic. He concludes, By this you know the methods for these rules if you
have paid careful attention to the examples and the operations.
Let us turn to Descartes. He is one of the founders of both modern philosophy and modern mathematics. It is impossible to do justice to his life,
his work, and his impact in a short space. We will have to be content with a
brief survey of his contributions to algebra. But first, a few words about the
broader context of his work and how algebra fit into it.
Descartes was born in 1596 in La Haye, France. In 1619 he had the nowfamous dreams that inspired him to find a new basis for scientific inquiry.
He moved to the Netherlands in 1628 and published Discours de la methode
pour bien conduire sa raison et chercher la verite dans les sciences, or A
Discourse on the Method of Correctly Conducting Ones Reason and Seeking Truth in the Sciences [18], in 1637. Ian Maclean opens the introduction to his translation [18, p. vii] with the observation that it marks one
of the pivotal moments of Western European thought; it was the work of
a formidably clever, radical, rigorous thinker, who in this short, informally
presented introduction to his work threatened the very foundations of many
prevailing philosophical beliefs, and set an agenda for enquiry into man and
nature whose effects have lasted up to the present day.
The work consists of a preface and three essays. The preface, A Discourse on the Method, is the most famous part, for it is here that Descartes
seeks to reform my own thoughts and build on a foundation which is mine
alone. [18, p. 15]. Algebra is one of the areas to which he intends to apply
his new foundation [18, p. 17]:
As for ancient geometrical analysis and modern algebra, even apart
from the fact that they deal only in highly abstract matters that seem to
have no practical application, the former is so closely tied to the consideration of figures that it is unable to exercise the intellect without
greatly tiring the imagination, while in the latter case one is so much
a slave to certain rules and symbols that it has been turned into a confused and obscure art that bewilders the mind instead of being a form
of knowledge that cultivates it. This was why I thought that another
method had to be found.
In the next paragraph, he proposes his first precept, never to accept
anything as true that I did not incontrovertibly know to be so. This is a stiff

i
i

i
IrvingBook 2013/5/22 15:39 page 171 #187

6.8. History

171

challenge. Some pages later, in one of the most historic passages of western
literature, Descartes arrives at his first item of knowledge [18, p. 28]:
Because our senses sometimes deceive us, I decided to suppose that
nothing was such as they lead us to imagine it to be. And because
there are men who make mistakes in reasoning, even about the simplest elements of geometry, and commit logical fallacies, I judged that
I was as prone to error as anyone else, and I rejected as false all the
reasoning I had hitherto accepted as valid proof. Finally, considering
that all the same thoughts which we have while awake can come to us
while asleep without any one of them being true, I resolved to pretend
that everything that had ever entered my head was no more true than
the illusions of my dreams. But immediately afterwards I noted that,
while I was trying to think of all things being false in this way, it was
necessarily the case that I, who was thinking them, had to be something; and observing this truth: I am thinking therefore I exist, was so
secure and certain that it could not be shaken by any of the most extravagant suppositions of the skeptics, I judged that I could accept it
without scruple, as the first principle of the philosophy I was seeking.
This is Descartes famed statement of certainty, Cogito, ergo sum, traditionally translated as, I think, therefore I am.
Having developed a method for obtaining knowledge from secure foundations, Descartes illustrates its effectiveness in three essays: La Dioptrique
(treating optics), Les Meteores (treating meteorology), and La Geometrie
(treating, yes, geometry, but also algebra). Our interest is in La Geometrie.
There is much more to be said about Descartes philosophical work and his
life, but let us leave it at that, adding only that in 1649, he moved from the
Netherlands to Sweden to tutor Queen Christina, becoming ill a few months
later and dying in 1650.
We turn to La Geometrie. (An English translation by David Eugene
Smith and Marcia L. Latham with the title The Geometry of Rene Descartes
was published in 1925, the translated pages interleaved with a facsimile of
the original edition. It remains available in a reprinted version [19].) La
Geometrie has three parts. One might not guess from the third parts title,
On the Construction of Solid or Supersolid Problems, that it is devoted
to algebra. This makes sense once one learns that in Descartes terminology, solid refers to equations of degree 3 and supersolid to equations of
higher degree. This third part contains the first treatment of algebra written
with language and algebraic notation that we would recognize. Lets survey
some of its highlights.

i
i

i
IrvingBook 2013/5/22 15:39 page 172 #188

172

6. Quartic Polynomials

In the fifth paragraph [19, p. 156], Descartes begins a sequence of general statements . . . concerning the nature of equations. The next sentence
introduces a radical change in perspective: An equation consists of several
terms, some known and some unknown, some of which are together equal
to the rest; or rather, all of which taken together are equal to nothing; for
this is often the best form to consider.
What Descartes is proposing, and what we now take for granted, is that
polynomial equations are best written in the form of a polynomial on one
side of the equal sign and 0 on the other. Until this point equations would
be written with the restriction that coefficients are positive and terms should
be added, not subtracted, forcing the consideration of multiple forms of
quadratic and cubic equations, as we have seen.
Another of Descartes innovations is the use of letters near the end of the
alphabet for variables and letters near the beginning or middle as constants,
a notational convention that we take for granted. (Just before Descartes
birth, Vi`ete had taken the first step in this direction, using vowels for variables and consonants for constants.)
In the ensuing pages, Descartes makes the first-ever statements of a series of now-standard results about polynomials, and in language we would
find largely familiar. He speaks of the dimension of a polynomial or equation where we say degree. He distinguishes between positive and negative roots of a polynomial, calling the positive ones true, the negative ones
false. And, as we have seen, Descartes uses the word nothing in place
of zero. Given this terminology, the passage that comes after the sentence
about setting polynomials equal to nothing is completely understandable
[19, pp. 159160]:
Every equation can have as many distinct roots (values of the unknown
quantity) as the number of dimensions of the unknown quantity in the
equation. . . . It often happens, however, that some of the roots are false
or less than nothing. . . . It is evident from the above that the sum of
an equation having several roots is always divisible by a binomial consisting of the unknown quantity diminished by the value of one of the
true roots, or plus the value of one of the false roots. In this way, the dimension of an equation can be lowered. On the other hand, if the sum
of the terms of an equation is not divisible by a binomial consisting
of the unknown quantity plus or minus some other quantity, then this
latter quantity is not a root of the equation.
We find here the first statements of two of our theorems: Theorem 1.6, that
the number of roots of a polynomial is bounded by its degree, and Theorem
1.4, that if a is a root of the polynomial f .x/, then x a divides f .x/.

i
i

i
IrvingBook 2013/5/22 15:39 page 173 #189

173

6.8. History

A few paragraphs later, Descartes explains how to pass from a polynomial equation with a given set of roots to a new one with roots shifted by a
fixed quantity through a change of variable [19, pp. 163164]:
If the roots of an equation are unknown and it be desired to increase or
diminish each of these roots by some known number, we must substitute for the unknown quantity throughout the equation, another quantity greater or less by the given number. Thus, if it be desired to increase by 3 the value of each root of the equation
x 4 C 4x 3

19x 2

106x

120 D 0

put y in place of x, and let y exceed x by 3, so that y 3 D x.

Then for x 2 put the square of y 3, or y 2 6y C 9; for x 3 put its
cube, y 3 9y 2 C 27y 27; and for x 4 put its fourth power, or y 4
12y 3 C 54y 2 108y C 81: Substituting these values in the above
equation, and combining, we have . . . y 4 8y 3 y 2 C 8y D 0, or
y 3 8y 2 y C 8 D 0, whose true root is now 8 instead of 5 since it
has been increased by 3.
New in this discussion is the abstract idea of transforming an equation
with one set of roots to an equation with a different set of roots without
knowing what the roots are. Descartes applies this procedure to pass from
an arbitrary polynomial equation to a reduced one [19, p. 167]:
Now this method of transforming the roots of an equation without determining their values yields two results which will prove useful: First,
we can always remove the second term of an equation by diminishing
its true roots by the known quantity of the second term divided by the
number of dimensions of the first term, if these two terms have opposite signs; or, if they have the same signs, by increasing the roots by
the same quantity.
Descartes works out some examples of this method, then explains how
to transform a polynomial equation to a new one whose roots are all multiplied or divided by a given quantity. It is exciting to read these pages and
see now-familiar algebraic notions and formulations coming to life for the
first time.
Just a few pages later [19, pp. 180183], before discussing cubic equations, Descartes turns to factoring quartic polynomials as products of quadratics. He passes to a reduced quartic (with p, q, and r as coefficients rather
than our q, r , and s), then introduces the resolvent cubic without ado:

i
i

i
IrvingBook 2013/5/22 15:39 page 174 #190

174

6. Quartic Polynomials

Again, given an equation in which the unknown quantity has four dimensions . . . we must increase or diminish the roots so as to remove
the second term, in the way already explained, and then reduce it to
another of the third degree, in the following manner: Instead of
x 4 px 2 qx r D 0
write
y 6 2py 4 C .p 2 4r /y 2

q 2 D 0:

For the ambiguous sign put C2p in the second expression if Cp occurs
in the first; but if p occurs in the first, write 2p in the second . . . .
For example, given
x4

4x 2

8x C 35 D 0

replace it by
y6

8y 4

124y 2
4

64 D 0:

For since p D 4, we replace 2py by 8y 4 ; . . . .

Similarly, instead of

x4

17x 2

20x

6D0

we must write
y6

34y 4 C 313y 2

400 D 0;

for 34 is twice 17, and 313 is the square of 17 increased by four times
6, and 400 is the square of 20.
We do not bother with the distinction Descartes maintains between positive
and negative coefficients, and therefore would dispense with his clarifications on how to choose signs in writing the resolvent cubic equation.
Descartes presents a third example of a quartic and its resolvent cubic, one with coefficients that are algebraic expressions rather than specific
numbers, then explains how to use the resolvent cubic to factor a quartic
[19, p. 184]:
If, however, the value of y 2 can be found, we can by means of it separate the preceding equation into two others, each of the second dimension, whose roots will be the same as those of the original equation.
Instead of x 4 px 2 qx r D 0, write the two equations
Cx 2

1
1
q
yx C y 2 p
D0
2
2
2y

i
i

i
IrvingBook 2013/5/22 15:39 page 175 #191

175

6.8. History

and
1
1
q
Cx 2 C yx C y 2 p
D 0:
2
2
2y
. . . It is then easy to determine all the roots of the proposed equation.
This is our Theorem 6.2.
Up to this point, Descartes has yet to discuss Cardanos formula. It has
not been needed, as the values of y in his examples can be found directly
and the quartics then factored as products of quadratics, as he illustrates in
the sentences that immediately follow. Rather than continuing to read his
discussion, lets do the work ourselves:
Exercise 6.35. Solve two quartic equations from La Geometrie. In both
cases, a root of the the resolvent cubic can be found by guessing. (Try
squares of small positive integers.)
(i) x 4

4x 2

(ii) x 4

17x 2

8x C 35 D 0.
20x

6 D 0.

Descartes closes La Geometrie with a wish [19, p. 240]: I hope that

posterity will judge me kindly, not only as to the things which I have explained, but also as to those which I have intentionally omitted so as to leave
to others the pleasure of discovery.
A little over a century later, Euler gave his own account of quartic equations in Elements of Algebra [26]. In Section 5.8 we discussed his treatment of Cardanos formula, which appears in Chapter XII of Section 4. The
next three chapters treat the quartic. In Chapter XIII, Euler deals with some
special families of quartics, then in Chapter XIV he presents the rule of
Bombelli, which is actually Ferraris approach, misattributed by Euler. It
is in Chapter XV, Of a new method of resolving equations of the fourth
degree, that Euler obtains the result that constitutes our Theorem 6.4.
Rather than using Descartes approach, as given in Theorem 6.2, Euler
hypothesizes from the start [26, p. 283] that a reduced quartic equation has
a root that can be expressed as the sum of three square roots, then searches
for an expression for the cubic polynomial whose roots these would be:
We will suppose that the root of an equation of the fourth degree has
p
p
p
the form, x D p C q C r, in which the letters p, q, r express the
roots of an equation of the third degree, such as z 3 f z 2 Cgz h D 0;
so that p C q C r D f ; pq C pr C qr D g; and pqr D h.

i
i

i
IrvingBook 2013/5/22 15:39 page 176 #192

176

6. Quartic Polynomials

We based our treatment of the quartic on Descartes resolvent cubic,

which led to the expression in Theorem 6.4 that is one-half the sum of
three square roots. We could instead have absorbed those pesky 2s into
the square root signs, obtaining three numbers that are one-fourth the roots
of Descartes resolvent. This would lead to a new resolvent cubic that is
slightly different, and that is the cubic that Euler obtains.
Euler proceeds much as we did in Section 5.2 when we derived Cardanos formula for the roots of a reduced cubic polynomial from its resolvent quadratic. (See Theorems 5.2 and 5.3.) After a half-page of elementary
calculations [26, p. 283], which can be done as an exercise, Euler arrives at
the conclusion [26, p. 284]:
Since, therefore, the equation
x4

ax 2

bx

c D 0;

gives the values of the letters f , g, and h, so that

f D

p
1 2 1
1 2
1
1
a; g D
a C c; and h D
b ; or h D b;
2
16
4
64
8

we form from these values the equation of the third degree z 3 f z 2 C

gz h D 0, in order to obtain its roots by the known [Cardanos] rule.
And if we suppose those roots, 1. z D p, 2. z D q, 3. z D r , one of the
roots of our equation of the fourth degree must be, by the supposition,
xD

p
p
p
p C q C r:

This method appears at first to furnish only one root of the given equap
tion; but if we consider that every sign may be taken negatively, as
well as positively, we immediately perceive that this formula contains
all the four roots.
A short discussion follows, like the last part of Exercise 6.14 and of Exercise 6.16. Euler explains that there are eight possibilities for the sum of
the three square roots, but the correct four are chosen by using the rule that
the square roots product must equal b=8. Eulers notation doesnt match
ours, nor does his resolvent cubic, but it is easy to check that with the appropriate change in the coefficient labels and signs and a change of variable
by a factor of 4, his result coincides with Theorem 6.4.
Procedure in hand, Euler solves the equation x 4 25x 2 C 60x 36 D 0
[26, pp. 284285]. Lets do so too.

i
i

i
IrvingBook 2013/5/22 15:39 page 177 #193

177

6.8. History

Exercise 6.36. Solve

x4

25x 2 C 60x

36 D 0

using the version of Eulers approach given in Theorem 6.4. One root of
its resolvent cubic can be found by testing various integer squares. Using
it, factor the cubic and find the other two roots. Euler constructed a very
simple example indeed.
After that example, Euler demonstrates how a general quartic equation
can be converted to a reduced quartic equation, thereby showing that his
new rule is applicable to all quartics. Next comes the following observation
[26, p. 286]:
This is the greatest length to which we have yet arrived in the resolution of algebraic equations. All the pains that have been taken in
order to resolve equations of the fifth degree, and those of higher dimensions, in the same manner, or, at least, to reduce them to inferior
degrees, have been unsuccessful: so that we cannot give any general
rule for finding the roots of equations, which exceed the fourth degree.
We will take up the story of the quintic next.

i
i

i
IrvingBook 2013/5/22 15:39 page 178 #194

i
i

i
IrvingBook 2013/5/22 15:39 page 179 #195

7
Higher-Degree
Polynomials
We have devoted chapters to quadratic, cubic, and quartic polynomials. This
pattern cannot continue through all degrees, and not just because the resulting book would be infinitely long. It turns out that results of the sort
we have obtained do not exist for polynomials in degree greater than four.
Therefore, we will content ourselves with a survey of some central results
about higher-degree polynomials, combining proof sketches (or no proofs
at all) with historical discussions. The chapter ends with a proof of the fundamental theorem of algebra.

7.1 Quintic Polynomials

We have obtained the quadratic formula, Cardanos formula, and Eulers
formula for solutions of polynomial equations up through degree four. What
about quintic polynomials? By Theorem 1.14, quintics have real roots. It is
natural to seek a formula for a root like our other formulas, one involving
the polynomials coefficients and the operations of sum, product, quotient,
and nth root extraction. If we had such an expression, we could solve quintic
equations by radicals, the word radical referring to an nth root. Through
the seventeenth and eighteenth centuries, progress was made in simplifying
the problem of solving quintic equations, but no solution by radicals was
found. Mathematicians began to suspect that a solution may not exist, and
this was shown to be true in the nineteenth century. In this section, we will
describe some of the highlights of the quintics history, sketching results but
providing no proofs.
Our first step in dealing with a cubic or quartic equation was to change
179

i
i

i
IrvingBook 2013/5/22 15:39 page 180 #196

180

7. Higher-Degree Polynomials

variables to pass to a reduced equation, one in which the term of second

highest degree drops out. This can be done for quintics as well. In fact, it
can be done in general: Given
x n C an

1x

we can replace x by y D x
y n C bn

n 1

.an
2y

n 2

C


C a1 x C a0 D 0;
1 =n/

to obtain

C


C b1 y C b0 D 0;

which has no y n 1 term. It is called a reduced equation of degree n. The

coefficients bi are expressible in terms of the ai . If we can solve the reduced
equation, we can add an 1 =n to each solution to obtain the solutions of the
original equation.
The reduction of the problem of solving a quintic equation to that of
solving a reduced quintic equation is straightforward, relying on an obvious linear change of variable. Ehrenfried Walter von Tschirnhausen (1651
1708) introduced the idea of using non-linear changes of variable to eliminate additional terms in an equation. He showed how to pass from a reduced
equation of degree n to an equation of the form
z n C cn

3z

n 3

C


C c1z C c0 D 0;

where the new coefficients are expressible in terms of the old ones. The
equation has no terms of degrees n 1 or n 2 and is called a principal equation. If we can solve a principal equation, then Tschirnhausens change of
variable process allows us to pass back to solutions of the reduced equation,
perhaps having to calculate some square roots to do so. From the solutions
to the reduced equation, we get solutions of the original equation.
Applying Tschirnhausens idea to a reduced equation of degree 3, we
obtain a cubic equation of the form z 3 C c D 0, which can be solved by
computing cube roots. Square root calculations lead to the roots of the reduced equation and adding a suitable constant leads to the roots of a general
cubic. Cardanos formula can be derived in this way.
The process of passing from a reduced equation to a principal equation
is an example of a general process called a Tschirnhausen transformation.
Applying it to quintic polynomials, we are led to quintic equations of the
form
z 5 C cz 2 C dz C e D 0:
Erland Bring (17361798) used more complicated Tschirnhausen transformations to go one step further. He showed in 1786 that the problem of solving a quintic equation can be reduced to that of solving a quintic equation

i
i

i
IrvingBook 2013/5/22 15:39 page 181 #197

181

7.1. Quintic Polynomials

of the form
w 5 C dw C e D 0:
Such equations are now called quintic equations in Bring-Jerrard form, in
honor of Bring and of George Jerrard (18041863), who used the same idea
to study equations of higher degree.
Several eighteenth-century mathematicians tried to solve the general
quintic equation, including Bring, Etienne Bezout, Edward Waring, Lagrange, andno surpriseEuler. We saw in Section 5.2 that Euler solved
reduced cubic equations in Elements of Algebra [26] by determining, for a
quadratic polynomial with roots U and
p V , the
p coefficients of a cubic polynomial whose roots have the form 3 U C 3 V . Going backwards, he associated to a reduced cubic polynomial its resolvent quadratic polynomial
and found the roots of the cubic as sums of cube roots of the roots of the
quadratic. He had a similar approach to solving reduced quartic equations
using resolvent cubics, as we discussed in Section 6.8. We concluded that
section with his remark that efforts (up to 1770) to resolve equations of the
fifth degree had been unsuccessful.
Earlier, in a 1732 paper, Euler had conjectured that the roots of a quintic
polynomial take the form
p
p
p
p
5
A1 C 5 A2 C 5 A3 C 5 A4 ;

where A1 , A2 , A3 , and A4 are roots of a quartic polynomial associated to

the quintic. More generally, he conjectured such a result for polynomials
of degree n, each root being the sum of nth roots of n 1 quantities that
would themselves be roots of an associated resolvent polynomial of degree
n 1. This would be a powerful extension of his approach to cubic and
quartic polynomials. However, he was unable to make any progress on the
conjecture, except for special families of polynomials.
The Italian mathematician Paolo Ruffini (17651822) was the first to
put aside efforts to solve quintic equations and instead attempt to prove that
there is no general solution by radicals. The title of his 1799 work says it all:
General theory of equations in which it is shown that the algebraic solution
of the general equation of degree greater than 4 is impossible [57]. In the
introduction, Ruffini wrote [5, p.263]:
The algebraic solution of general equations of degree greater than 4 is
always impossible. Behold a very important theorem which I believe I
am able to assert (if I do not err); to present the proof of it is the main
reason for publishing this volume. The immortal Lagrange, with his
sublime reflections, has provided the basis of my proof.

i
i

i
IrvingBook 2013/5/22 15:39 page 182 #198

182

7. Higher-Degree Polynomials

Ruffini sent Lagrange a copy of the book, but received no reply. In 1803,
Ruffini published a paper with another proof. In 1806, he published yet
another proof, and in 1813 he published a paper in which he expressed his
disappointment at the poor reception his work received. As it turns out,
Lagrange had read his work, but did not think the proof was complete and
chose not to express his approval. Other contemporaries also did not find
the proof complete or correct.
Raymond Ayoub has given an excellent appraisal of Ruffinis work in
Paolo Ruffinis Contributions to the Quintic [5], along with an account
of the work of Lagrange, Gauss, and others. It is clear in retrospect, as
emerges from Ayoubs account, that Ruffini did not receive proper credit
for his contributions. One exception was the response of the great French
mathematician Augustin-Louis Cauchy, who wrote to Ruffini in 1821 [5,
p. 271] that your memoir on the general resolution of equations is a work
which has always seemed to me worthy of the attention of mathematicians
and which, in my judgement, proves completely the insolvability of the
general equation of degree > 4.
It fell to the Norwegian mathematician Niels Abel (18021829) to provide the first widely recognized proof that there can be no general solution of quintic equations by radicals. His result was published in 1824 [1].

A few years later, the French mathematician Evariste

Galois (18111832)
provided another approach, one that would revolutionize algebra [31]. His
proof can be found in every graduate-level text on algebra (such as Abstract
Algebra, by David S. Dummit and Richard M. Foote [22, p. 629]) and some
undergraduate texts as well, and his ideas continue to influence mathematics
today.
The lives of Abel and Galois are of great interest. Both were mathematical geniuses of the first rank. Both died young. Galois is notorious for his
dramatic death at 20 from wounds incurred in a duel, and for the image of
him feverishly writing out his mathematical ideas in a letter on the eve of
the duel. In both cases, we can only dream of the glorious discoveries they
would have made had they lived longer, and reflect on the unfairness of life.
Since their stories are well told elsewhere, we will move on. Especially recommended for further reading is Peter Pesics Abels Proof: An Essay on
the Sources and Meaning of Mathematical Unsolvability [52].
There is one more twist to the tale. Ruffini deserves credit as the first
mathematician to focus attention on proving that the quintic cannot be solved
by radicals rather than searching for such a solution. But Gauss had similar
inklings. In his 1799 doctoral dissertation, Gauss wrote [5, pp. 262263]:
After the labors of many geometers left little hope of ever arriving at
the resolution of the general equation algebraically, it appears more

i
i

i
IrvingBook 2013/5/22 15:39 page 183 #199

183

7.1. Quintic Polynomials

and more likely that this resolution is impossible and contradictory. . . .

Perhaps it will not be so difficult to prove, with all rigor, the impossibility for the fifth degree. I shall set forth my investigations of this at
greater length in another place. Here it is enough to say that the general
solution of equations understood in this sense [i.e., by radicals] is far
from certain and this assumption [i.e., that any equation is solvable by
radicals] has no validity at the present time.
Gauss never pursued the matter. Ayoub speculates that Gauss did not
attach very much importance to solvability by radicals, referring again to
Gausss dissertation, in which he wrote [5, p. 275]:
what is called a solution to an equation is, in reality, nothing but the
reduction of the equation to prime equations the solution is not
exhibited but symbolized
and if you express a root of the equation
p
x n D H by n H , you have not solved it nor done anything more
than if you devise some symbol to denote a root of the equation x n C
Ax n 1 C


D 0 and place the root equal to this symbol . . . .
We made much the same point for quadratic equations at the end of Section 2.1, describing the quadratic formula as providing not a solution to a
quadratic equation but a reduction of the solution to the problem of calculating square roots. Gauss dismisses the entire enterprise. His youthful views
notwithstanding, the work of Abel and Galois has been justly celebrated.
The story of quintic equations does not end with Abel and Galois. As
Gauss reminded us, a solution by radicals requires the extraction of nth
roots, a process that is a problem of analysis, not algebra. Using standard
functions of analysis, such as the exponential and logarithm, or cosine and
inverse cosine, we can calculate nth roots of real and complex numbers.
Perhaps by allowing an enlarged stock of functions from analysis, we can
use algebra to reduce the solution of quintic equations to a form that can be
handled. In the decades after the deaths of Abel and Galois, several mathematicians did this.
The first was Gotthold Eisenstein (18231852), in 1844. We have seen
that the problem of solving the general quintic equation can be reduced by
Tschirnhausen transformations to that of solving the Bring-Jerrard equation
w 5 C dw C e D 0:
One can reduce further to the problem of solving the quintic equation
v5 C v

f D 0;

where f is a constant.

i
i

i
IrvingBook 2013/5/22 15:39 page 184 #200

184

7. Higher-Degree Polynomials

In a short note on solutions of equations of degree up to 4 [24], Eisenstein introduces functions  and that satisfy ./2 D  and ./3 D ,
observing that  is defined up to multiples of 1 and is defined up to
multiples of the powers of !. These are the square root and cube root functions. He expresses solutions of general cubic and quartic equations in terms
of  and . He then concludes his paper with the remark that the roots of a
general equation of degree 5 have a similar form, using in place of  and
a function  satisfying
./5 C ./ D :
In a footnote, he offers a power series formula for :
./ D 

5 C 10

9
2

15
14

13
17
C 20
19
18
3
4

:

We can think of  as a generalized fifth-root function. Given a real number f , we can apply  to obtain a new number .f / that is not quite a fifthroot of f : rather than satisfying .f /5 D f , it satisfies .f /5 C.f / D f .
But this means that .f / is a solution to our equation v 5 C v D f , and with
such a solution available, we can solve any quintic equation.
Although quintic equations cant be solved by radicals, Eisenstein has
demonstrated that we can come close. We simply need his power series,
which lets us calculate a slight extension of fifth roots: the numbers v satisfying v 5 C v D f .
Papers by S.J. Patterson [51] and John Stillwell [62] contain illuminating discussions of Tschirnhausen transformations, the Bring-Jerrard form
of the quintic, and Eisensteins solution of the quintic. Stillwell comments
[62, p 61] that Eisenstein added his solution to v 5 C v D f only as an
afterthought, speculating that Eisenstein may well have have considered
this solution to be childs play, because it is based on a method he discovered for himself at age 14. (In another of Eisensteins 1844 papers [25], he
had described how a certain power series arose from his first mathematical
research, in his fifteenth year.)
In 1858, Charles Hermite (18221901) used the Bring-Jerrard reduction to solve quintic equations, but in terms of a type of function known
as a modular function [34]. Felix Klein (18491925) described another approach in his famous 1884 book Lectures on the Icosahedron [37].
A century later, Peter Doyle and Curt McMullen published a paper,
Solving the quintic by iteration [21], in which they showed that quintic equations can be solved by a special iteration process involving generally convergent algorithms based on rational functions. (Rational functions are quotients of polynomial functions.) This is a result in the relatively

i
i

i
IrvingBook 2013/5/22 15:39 page 185 #201

7.2. The Fundamental Theorem of Algebra

185

new theory of complex dynamics, a subject of great contemporary interest.

They also proved a modern analogue of the Abel-Galois unsolvability result, stating that polynomial equations of higher degree cannot be solved by
their iteration process. In 1998, McMullen was one of four recipients of the
Fields Medal at the International Congress of Mathematicians. The Fields
Medal is the highest honor a mathematician can receive and is regarded as
the equivalent in mathematics of the Nobel Prize.

7.2 The Fundamental Theorem of Algebra

The discovery of solutions to cubic equations in the sixteenth century led
to the introduction of complex numbers, which in turn led to the realization that we can find solutions to all equations of degree four or less. What
happens for higher degrees? Thats the subject of this section.
The relevant mathematical result is known as the fundamental theorem
of algebra. It has several formulations, which we will discuss without providing proofs. What we do discuss, in place of proofs, is the need to bring
non-algebraic information to bear in order to find a proof. We then turn to
the fundamental theorems history, with special attention to a gap in some of
the early proofs that can be filled through the introduction of the algebraic
notion of a field. We will return to the fundamental theorem in Section 7.5,
in order to provide a proof.
A quadratic equation has either two real roots or a pair of conjugate
complex roots; a cubic equation always has a real root and either another
two real roots or a pair of conjugate complex roots; a quartic equation has
zero, two, or four real roots, any others occuring in conjugate complex pairs.
For higher degrees, we know that every odd degree polynomial has a real
root (Theorem 1.14), but we have no results for even degrees. Given our
experience with polynomials of degrees 2 and 4, we might expect that even
degree polynomials have roots provided we allow complex numbers as well
as real numbers. The truth of this is the fundamental theorem of algebra.
Theorem 7.1 (fundamental theorem of algebra). Any polynomial of positive
degree with real coefficients has a root in the complex numbers.
Theorem 4.2 states that if a complex number r is a root of a polynomial
with real coefficients, then so is r . We can use this to prove Theorem 7.2.
Theorem 7.2. Let f .x/ be a positive-degree polynomial with real coefficients and let r a root of f .x/ that is a non-real complex number. Let

i
i

i
IrvingBook 2013/5/22 15:39 page 186 #202

186
bD

7. Higher-Degree Polynomials

.r C r/ and c D r r. Then b and c are real and f .x/ factors as

.x 2 C bx C c/g.x/;

for a polynomial g.x/ with real coefficients.

Exercise 7.1.

Prove Theorem 7.2.

Theorem 7.2 has the following consequence:

Theorem 7.3. Let f .x/ be a polynomial of degree n > 0 with real coefficients.
(i) f .x/ factors as a constant multiple of a polynomial of the form
.x

r1 / : : : .x

rj /.x 2 C b1 x C c1 / : : : .x 2 C bk x C ck /;

where j and k are non-negative integers satisfying j C 2k D n and

the coefficients occurring are real numbers.
(ii) The numbers r1 ; : : : ; rj are the real roots of f .x/, each value occurring as often as its multiplicity as a root.
(iii) The non-real roots of f .x/ are the roots of the quadratic factors x 2 C
bi x C ci . There are k conjugate pairs of such roots, allowing repetitions.
The principal point of Theorem 7.3 is that any positive-degree polynomial with real coefficients factors as a product of linear and quadratic polynomials with real coefficients. This both follows from and implies Theorem
7.1, and is thus also called the fundamental theorem of algebra.
The fundamental theorem extends to polynomials with complex coefficients:
Theorem 7.4. Let f .x/ be a polynomial of degree n > 0 with real or
complex numbers as coefficients.
(i) f .x/ has a root in the complex numbers. In fact, f .x/ has exactly n
complex numbers as roots, counting multiplicities.
(ii) f .x/ factors as a constant multiple of a polynomial of the form
.x

r1 / : : : .x

rn /;

where r1; : : : ; rn are the complex numbers that are roots of f .x/.

i
i

i
IrvingBook 2013/5/22 15:39 page 187 #203

7.2. The Fundamental Theorem of Algebra

187

Theorem 7.4 appears at first to be a significant extension of the fundamental theorem, but it follows almost immediately via a simple trick. Write
f .x/ for the polynomial obtained by conjugating each of the coefficients of
f .x/. Its easily checked that the product f .x/f .x/ is a polynomial with
real coefficients. But then Theorem 7.3 can be applied to it, yielding Theorem 7.4.
Despite its name, the fundamental theorem cannot be proved by algebra
alone. This should not be surprising. Even proving that the polynomial x 2
c has a real root for a positive real number c (Theorem 1.10) requires the
intermediate value theorem, which is also needed to prove that every cubic
polynomial with real coefficients has a real root (Theorem 1.14). We can
hardly expect to prove the full fundamental theorem with anything less than
the intermediate value theorem as a tool.
Typically, the more powerful the tools used from calculus or the broader
field of analysis, the shorter the proof of the fundamental theorem. For example, the fundamental theorem can be proved in just two or three sentences [3, p. 122] from Liouvilles theorem, which states that a bounded entire function must be a constant. (Joseph Liouville was a nineteenth-century
French mathematician.) The hard work comes in proving Liouvilles theorem. A proof also follows from the result from topology that a continuous
real-valued function on a disk in the plane takes on a minimum and a maximum value, but a longer argument is needed. See for example the proof that
Charles Fefferman published in 1967 while still an undergraduate [27].
From an algebraic perspective, the most attractive approach to proving
the fundamental theorem would be to rely on the intermediate value theorem to prove that every positive real number has a real square root and
every cubic polynomial with real coefficients has a real root, and then rely
on algebra alone for the full theorem. A short, elegant proof follows from
more advanced results of group theory and Galois theory, both of which
grew from Galoiss ideas in proving the unsolvability of quintic polynomials. See, for example, the exposition in [22, pp. 615617]. Alternatively, we
can give a proof using more elementary algebraic ideas. We will introduce
these ideas in Section 7.4, then discuss the proof in Section 7.5. A useful
reference is The Fundamental Theorem of Algebra, by Benjamin Fine and
Gerhard Rosenberger [29], which presents proofs of the fundamental theorem using analysis, algebra, and topology.
Lets turn to some high points in the history of the fundamental theorem.
Reinhold Remmerts article on the theorem provides an excellent overview
[55, pp. 97122]. See also Chapter 14 of John Stillwells Mathematics and
its History [61, p. 266] and Chapter 6 of Bashmakova and Smirnovas The
Beginnings and Evolution of Algebra [8, pp. 9899].

i
i

i
IrvingBook 2013/5/22 15:39 page 188 #204

188

7. Higher-Degree Polynomials

The first mathematician to state the theorem was Albert Girard, in 1629,
in Invention Nouvelle en lAlg`ebre [32]. We briefly examined Girards account of cubic equations in Section 5.8. After discussing cubics, he states
a theorem that begins, All algebraic equations have as many solutions as
the size of the highest quantity. Girard makes no effort to provide justification, but he illustrates the statement with examples. For the first one,
Girard writes, Given the equation x 4 D 4x 3 C 7x 2 34x 24, the size of
the highest quantity is 4, which signifies that there are 4 certain solutions,
4
neither more nor less. Girards p
final example
p is x D 4x 3, for which
he lists the solutions 1; 1; 1 C
2; 1
2. From this, we can infer
that Girard intends his theorem to be understood in the context of complex
numbers, with roots counted according to their multiplicities. Thus, he has
provided a correct statement of the fundamental theorem.
Several of the leading mathematicians of the eighteenth centuryJeanle-Rond dAlembert, Euler, Lagrange, and Pierre-Simon Laplace (1749
1827)attempted to prove the fundamental theorem, but their arguments
were not complete. (See Chapter 6 of Dunhams Euler: The Master of Us
All [23] for an account of Eulers efforts.) One problem with these early
proofs was their reliance on the implicit assumption that a polynomial has
roots somewhere, where not being clear. Within this mysterious domain,
they would then show that the roots actually lie in the complex numbers. As
Remmert puts it [55, pp. 9899], until Gauss all mathematicians believed
in the existence of solutions in some sort of no-mans land . . . and tried
imaginatively to show that these solutions were in fact complex numbers.
It was Gauss who first pointed out this problem, in his doctoral dissertation
of 1799 [55, p. 104]:
Gauss begins his dissertation by a detailed critical examination of all
previous attempts to prove the theorem known to him. This is not the
place to discuss in detail the objections raised by the twenty-two year
old student against the proofs of dAlembert, Euler, and Lagrange
and thus against the leading mathematicians of the time . . . . Gausss
main objection was that the existence of a point at which the polynomials take the value zero is always assumed and that this existence
needs to be proved. Thus for example he reproaches Euler for using
hypothetical roots.
Gauss was not yet aware of Laplaces proof. In 1815, he would subject
Laplace to the same criticism [55, p. 105]: The ingenious way in which
Laplace dealt with this matter cannot be absolved from the main objections
affecting all these attempted proofs.

i
i

i
IrvingBook 2013/5/22 15:39 page 189 #205

7.2. The Fundamental Theorem of Algebra

189

Gauss offered not just criticism in his 1799 dissertation, but also a proof
of his own, one that did not rely on the existence of roots in some unspecified domain. Rather, he set out to prove their existence from scratch, within
the complex numbers. Although Gauss avoided the error of his predecessors, his proof, which is topological in nature, had gaps.
The first proof of the fundamental theorem that appears to be free of
error is one given by Argand in 1814, relying on the existence of a minimum
for a continuous function. He doesnt justify the existence of the minimum,
something Cauchy would later do. Gauss would continue to give additional
proofs using a variety of methods. His second proof, from 1816, is algebraic
in nature, and correct. It completes Eulers argument.
Regarding the gap in the proofs of Gausss predecessors, Remmert observes [55, p. 109]:
Nowadays one can only speculate about how mathematicians before
the beginning of the nineteenth century had visualized the solutions
of equations in their minds eye. It is difficult for us to understand
why, until the time of Gauss, they had an unshakable belief in a kind
of extraterrestrial existence of such solutions somewhere or other,
and then sought to show that these solutions were complex numbers.
With the development of algebra in the nineteenth century, it became a simple matter to construct the extraterrestrial solutions.
In studying polynomial equations, we typically wish to find solutions
in some familiar domain of numbers, such as the rational numbers, the real
numbers, or perhaps the complex numbers. As our historical discussion suggests, it may be convenient, at least provisionally, to search for solutions in
a broader domain of numbers, one that may contain the complex numbers
and be large enough to contain roots of the polynomial. If we can introduce
such a domain, we can then work within it to show that the roots are in fact
complex numbers. This leads to the algebraic notion of a field.
The families of numbers with which we are most familiarintegers,
rationals, reals, and complexesall have certain basic arithmetic properties
in common. To describe them, let us write K to denote any of these four
sets of numbers. Then K satisfies:
(1) Commutativity: Any two numbers a and b in K satisfy a C b D b C a
and a
b D b
a.
(2) Associativity: Any three numbers a, b, and c in K satisfy .aCb/Cc D
a C .b C c/ and .a
b/
c D a
.b
c/.
(3) Distrtibutivity: Any three numbers a, b, and c in K satisfy a
.b Cc/ D
a
b C a
c.

i
i

i
IrvingBook 2013/5/22 15:39 page 190 #206

190

7. Higher-Degree Polynomials

(4) Additive Identity: There is an additive identity in K, that is, a number

o satisfying o C a D a D a C o for any number a.
(5) Additive Inverses: For any number a in K, there is an additive inverse;
that is, a number a0 in K such that a C a0 D o.
(6) Multiplicative Identity: There is a multiplicative identity in K, that is,
a number ` satisfying `
a D a D a
` for any number a.
There is a seventh property that K may satisfy:
(7) Multiplicative Inverses: For any number a in K other than the additive
identity o, there is a multiplicative inverse; that is, a number a00 in K
such that a
a00 D `.
Any set K of numbers that satisfies properties .1/ to .6/ is called a ring.
The integers, rationals, reals, and complexes all form rings, with 0 as the
additive identity, a as the additive inverse of a number a, and 1 as the
multiplicative identity. If property .7/ is satisfied also, and if the additive
and multiplicative identities of K are distinct, then K is a field. The integers arent a field, since only 1 and 1 have multiplicative inverses. The
rational, real, and complex numbers are fields. (The existence of multiplicative inverses for complex numbers was treated in Exercise 4.3.) It will be
convenient to use standard notations for these fields: Q is the field of rational numbers, R is the field of real numbers, and C is the field of complex
numbers. The letter Q is chosen for the rational numbers since they are
quotients of integers.
Rings and fields are treated in any standard undergraduate algebra course.
We will not pursue the derivation of the basic results on them here. What
matters for us is that a field is the natural setting in which to study polynomials and their roots.
We now ask, given a polynomial f .x/ whose coefficients are real numbers, if we can construct a field K containing, but possibly larger than, C
that contains a root of f .x/. Once we have developed the basic results on
rings and fields, it is not difficult to prove this and more:
Theorem 7.5. Let f .x/ be a polynomial of positive degree n with real
coefficients. There exist a field K containing C and elements r1 ; r2 ; : : : ; rn
in K, repetitions allowed, such that
f .x/ D .x

r1 /.x

r2 /


.x

rn /:

We call K a field extension of C and say that f .x/ splits into linear
factors over K. We may also call K a splitting field for f .x/, although this

i
i

i
IrvingBook 2013/5/22 15:39 page 191 #207

191

7.3. Polynomial Factorization

term is usually reserved for a field of this type that is minimal in a suitable
sense.
Theorem 7.5 is the result that the mathematicians of the eighteenth century were missing, or unjustifiably took for granted. It is purely algebraic,
the proof requiring no property of the real numbers other than the fact that
they form a field. Once it is available, the proofs of the fundamental theorem that Gauss criticized can now be completed [55, pp. 107108]: The
Gaussian objection against the attempts of Euler-Lagrange and Laplace was
invalidated as soon as Algebra was able to guarantee the existence of a splitting field for every polynomial. From that moment on, as Adolf Kneser already observed in 1888, the attempted proofs became in effect fully valid.
We will discuss Theorem 7.5 in Section 7.3 and use it in Section 7.5,
where we present Laplaces 1795 proof of the fundamental theorem.

7.3 Polynomial Factorization

In studying roots of a polynomial f .x/, we have had occasion to look at
factorizations of f .x/ as a product of lower-degree polynomials. For instance, our proof that a polynomial of degree n has at most n roots relied
on Theorem 1.5, which states that a polynomial f .x/ with distinct roots
a1 ; : : : ; ak can be factored as
.x

a1 /.x

a2 /


.x

ak /g.x/

for a polynomial g.x/. In this section, we will study factorizations of f .x/

more generally. The principal theme that emerges is that polynomial factorization is analogous to integer factorization. Moreover, a result on polynomials can often be proved by mimicking the proof of the corresponding
integer result. A second theme is that the results do not require the polynomial coefficients to be real numbers. As we will see, the coefficients can
come from any set satisfying the axioms of a field. Working with more
general fields will allow us to sketch a proof of Theorem 7.5, the missing
ingredient in some early proofs of the fundamental theorem of algebra.
Lets begin our study of polynomial factorization by determining the
polynomial analogue to prime number. Given a non-zero polynomial f .x/
of degree n and a non-zero number c, we can factor f .x/ as the product
of the degree 0 polynomial c and the degree n polynomial c 1 f .x/. This
is as interesting as the factorization of an integer n > 1 as n
1. In both
cases, we call such factorizations trivial. A factorization of n as a product
r s of smaller positive integers is called non-trivial, as is a factorization of
f .x/ as a product of lower-degree polynomials g.x/h.x/. An integer n > 1

i
i

i
IrvingBook 2013/5/22 15:39 page 192 #208

192

7. Higher-Degree Polynomials

is prime if it has no non-trivial factorizations. Equivalently, n cannot be

factored as the product of two smaller positive integers. Similarly, we say
that a positive-degree polynomial f .x/ is irreducible if it has no non-trivial
factorizations; that is, it cannot be factored as the product of two lowerdegree polynomials.
Any degree 1 polynomial is irreducible, since constants are the only
lower-degree polynomials and no product of constants can have degree 1.
An example of an irreducible polynomial of degree 2 is x 2 C 1. If it could
be factored as a product of lower degree polynomials, then the resulting
degree 1 factors would correspond to roots, but x 2 C 1 has no real roots.
More generally, a quadratic polynomial x 2 C bx C c either factors as a
product .x r /.x s/, corresponding to real roots r and s, or has no such
factorization. Which of the two cases occurs can be determined by the sign
of the discriminant b 2 4c, allowing us to conclude that x 2 C bx C c is
irreducible precisely when b 2 4c < 0.
We are assuming that the numbers we allow as coefficients are real numbers. However, we may wish to select the allowable coefficients from other
classes of numbers, and this will lead to different factorization results. For
example, if we allow complex numbers as coefficients, then x 2 C bx C c
is never irreducible. If we allow only rational numbers
then
p
p as coefficients,
2/.x C 2/ when we
x 2 2 is irreducible, even though it factors as .x
allow real numbers as coefficients.
If we choose to work with real coefficients, then the fundamental theorem of algebra (in the form of Theorem 7.3) states that any polynomial of
positive degree factors as a product of degree 1 and degree 2 polynomials,
from which it follows that the irreducible polynomials are the degree 1 polynomials and the degree 2 polynomials with negative discriminant. Theorem
7.4 implies that when we allow complex numbers as coefficients, the irreducible polynomials are precisely the degree 1 polynomials. In contrast, if
we allow only rational numbers as coefficients, there are irreducible polynomials of every positive degree. For instance, for any positive integer n,
x n 2 is irreducible. These examples make it clear that the study of polynomial factorization will depend on our choice of allowable coefficients.
It turns out that there are many natural choices besides the rational numbers, the real numbers, and the complex numbers. What they have in common is that they are examples of fields.
We have described fields as collections of numbers with addition and
multiplication rules that satisfy seven properties. A review of the properties reveals that nothing in their wording requires that what we are adding
and multiplying be numbers. They can be any sort of objectpolynomials,

i
i

i
IrvingBook 2013/5/22 15:39 page 193 #209

7.3. Polynomial Factorization

193

M&Ms, Lego bricksas long as addition and multiplication rules are introduced that satisfy the properties. We will have a better perspective on the
notion of a field if we introduce two examples besides rational, real, and
complex numbers.
Binary arithmetic gives one example. Write F2 for the set of symbols 0
and 1 with the familiar addition rules 0 C 0 D 0 and 0 C 1 D 1 C 0 D 1
along with the not-so-familiar rule 1 C 1 D 0. For multiplication, we will
adopt the usual rules 0  0 D 0  1 D 1  0 D 0 and 1  1 D 1.
Exercise 7.2. Under the definitions for addition and multiplication, verify
that F2 is a field, with 0 as the additive identity and 1 as the multiplicative
identity.
We have constructed a field with just two elements. It is important to
recognize that the elements, 0 and 1, are not the usual numbers zero and
one shared by Q, R, and C. Rather, they are symbols satisfying the addition
and multiplication rules just introduced. We may wish to give them concrete
meaning, but F2 as described is a field whether we do so or not.
One way to give meaning to 0 and 1 is to regard 0 as a symbol for the
collection of all even integers and 1 as a symbol for the collection of all odd
integers. The addition and multiplication rules for 0 and 1 then represent
familiar arithmetic facts about the addition and multiplication of even and
odd integers. For example, 0 C 1 D 1 represents the fact that the sum of an
even integer and an odd integer is odd, while 1  1 D 1 represents the fact
that the product of two odd integers is again odd. But we neednt have this
in mind when working with F2 .
A second example of a field is the set R.t/ of rational functions. A rational function is a fraction f .t/=g.t/, where f .t/ and g.t/ are polynomials
with real coefficients and g.t/ is non-zero. Just as in describing rational
numbers as the fractions formed from integers, we have to be precise, identifying two fractions f .t/=g.t/ and h.t/= k.t/ as the same if f .t/k.t/ D
g.t/h.t/. Addition and multiplication are performed in the usual way, using
common denominators for addition. The fraction f .t/=g.t/ is 0 precisely
when f .t/ D 0. Assuming f .t/=g.t/ is non-zero, its multiplicative inverse
is g.t/=f .t/.
Once a field K is chosen, we can study polynomials with coefficients
in K. The standard notation for the set of polynomials is Kx. Thus, Rx
consists of polynomials whose coefficients are real numbersthe polynomials with which we have worked for most of this bookand F2 x consists
of polynomials whose coefficients lie in the field F2.

i
i

i
IrvingBook 2013/5/22 15:39 page 194 #210

194

7. Higher-Degree Polynomials

Given two polynomials f .x/ and g.x/ in Kx, the addition and multiplication laws for K allow us to add and multiply f .x/ and g.x/ in the
way defined in Section 1.2, obtaining new polynomials f .x/ C g.x/ and
f .x/g.x/ that also lie in Kx. (Under these addition and multiplication
rules, Kx satisfies the defining properties of a ring, the polynomial ring
with coefficients in K.) For example, we can add and multiply polynomials
whose coefficients lie in our new fields F2 and R.t/.
Exercise 7.3. Let x C 1 and x 2 C x C 1 be polynomials in F2x. (That
is, we regard the coefficients as elements of F2 . Thus, 1 is not our usual
number 1. It is the element 1 of F2 .)
(i) Verify that .x C 1/ C .x 2 C x C 1/ D x 2 .
(Hint: 1 C 1 D 0.)
(ii) Verify that .x C 1/.x 2 C x C 1/ D x 3 C 1.
Exercise 7.4.

Let

2tx 2

1
x C 1 and
t

3t 5 C 4t 4
x C .3t 2
t2 C 1

5/x

be polynomials in R.t/x. Compute their sum and product.

Given a field K, lets write 0 for its additive identity. In all our examples
of fields besides F2, the additive identity is the actual number 0, but sometimes, as is the case with F2 , we cant identify 0 with a number. It is simply
a distinguished member of the field. We can define the notion of root for any
polynomial f .x/ in Kx: a root of f .x/ is an element a of K satisfying
f .a/ D 0. For example, the polynomial x 2 C 1 in F2 x has 1 as a root,
since
12 C 1 D 1 C 1 D 0:
The definition of irreducible polynomial makes sense for polynomials
with coefficients in any field: a polynomial f .x/ of positive degree in Kx
is irreducible if it cannot be factored as the product of two polynomials of
lower degree in Kx. In the next exercise, we determine the irreducible
polynomials of low degree in F2 x.
Exercise 7.5.

Consider polynomials in F2 x.

(i) Show that x and x C 1 are the only polynomials of degree 1 in F2 x,

and that both are irreducible.

i
i

i
IrvingBook 2013/5/22 15:39 page 195 #211

195

7.3. Polynomial Factorization

(ii) Show that x 2 , x 2 C x, and x 2 C 1 are not irreducible. Show that x 2 C

x C 1 is the only other polynomial of degree 2 and that it is irreducible.

(iii) Write all eight polynomials of degree 3 in F2 x. Determine which

are irreducible. For those that are not, factor them as products of irreducible polynomials of lower degree.

We can show, just as with Qx, that for every positive integer n, there
is an irreducible polynomial of degree n in F2 x.
Once we broaden the study of polynomials to allow any field as the set
of coefficients, we become interested in two types of theorems: those that
are specific to the choice of K, and those that hold no matter what field K
is chosen. The fundamental theorem of algebra is an example of the first
type of theorem, as it is a statement about polynomials with real or complex
coefficients. The results of Sections 1.2, 1.3, and 1.4 are examples of the
second type. They make sense for polynomials with coefficients in any field
K and the proofs work when real numbers are replaced by elements of
K. For example, combining Theorems 1.3 and 1.4, and extending them to
arbitrary fields, we obtain:
Theorem 7.6. Let K be a field and let f .x/ be a polynomial in Kx. If
x a divides f .x/, then a is a root of f .x/. Conversely, if a is a root of
f .x/, then x a divides f .x/.
Also, Theorem 1.5, which we restated at the beginning of this section,
holds for polynomials with coefficients in any field K:
Theorem 7.7. Let K be a field and let f .x/ be a polynomial in Kx. If
a1 ; : : : ; ak are distinct roots of f .x/ in K, then there is a polynomial g.x/
in Kx such that
f .x/ D .x

a1 /.x

a2 /


.x

ak /g.x/:

In the remainder of this section, we will look at results on polynomial

factorization that, like these theorems, hold for any choice of coefficient
field K. Let us begin with a refinement of Theorem 7.7. The multiplicity
of an element a as a root of a polynomial f .x/ is the largest integer m for
which .x a/m divides f .x/.
Theorem 7.8. Let K be a field and let f .x/ be a polynomial in Kx with
exactly k distinct roots a1 ; : : : ; ak in K. For 1  i  k, let mi denote the
multiplicity of ai as a root of f .x/. Then f .x/ has a factorization
f .x/ D .x

a1 /m1 .x

a2 /m2


.x

ak /mk g.x/

for a polynomial g.x/ in Kx that itself has no roots in K.

i
i

i
IrvingBook 2013/5/22 15:39 page 196 #212

196

7. Higher-Degree Polynomials

We might try to prove Theorem 7.8 by extending the argument in the

proof of Theorem 1.5. Certainly we can show by that argument that f .x/
factors as
.x a1 /r1 .x a2 /r2


.x ak /rk g.x/
for some positive integers r1 ; : : : ; rk and for a polynomial g.x/ not divisible
by any of the x ai . However, we must do a little more work to verify that
each exponent ri equals mi . Lets sketch the argument, using Exercise 1.9.
We will take i D 1, but the argument is the same for any index.
By the definition of m1 , there is a polynomial h.x/ not divisible by
x a1 such that f .x/ D .x a1 /m1 h.x/. Also by the definition of m1 , we
must have r1  m1 . Setting our two factorizations of f .x/ equal to each
other and canceling .x a1 /r1 , we obtain
.x

a2 /r2


.x

ak /rk g.x/ D .x

a1 /m1

r1

h.x/:

If m1 D r1 , we are done. Otherwise, we find that x a1 divides the product on the left side. Making an argument like the one used in the proof of
Theorem 1.6, we find that either x a1 divides one of the factors x ai
for i > 1, which is impossible, or it divides g.x/, which is also impossible.
Therefore we have m1 D r1 .
Recall the following fact about factorization of integers:
Theorem 7.9. Any integer n > 1 can be factored as a product of prime
numbers.
It is easy to take this for granted. Yet, it requires proof. The proof is
straightforward. If n is prime, we are done. Otherwise, by the definition of
prime, n factors as a product r s for smaller positive integers r and s. If r
or s is prime, we leave it as is. If not, we factor it as the product of smaller
positive integers. We continue in this way as we test each new factor for
primeness. Any non-prime factors become smaller with each round, so that
after at most n 1 rounds, we will have factored n as a product of prime
numbers.
Theorem 7.9 has a polynomial analogue, with essentially the same proof:
Theorem 7.10. Let K be a field. Any polynomial f .x/ of positive degree
in Kx can be factored in Kx as a product of irreducible polynomials.
Lets sketch the argument. If f .x/ is itself irreducible, we are done.
Otherwise, by the definition of irreducibility, f .x/ factors as a product
g.x/h.x/ for lower-degree polynomials g.x/ and h.x/ in Kx. If g.x/
or h.x/ is irreducible, we leave it as is. If not, we factor it as the product

i
i

i
IrvingBook 2013/5/22 15:39 page 197 #213

197

7.3. Polynomial Factorization

of lower-degree polynomials in Kx. After at most n rounds, where n is

the degree of f .x/, we will have factored f .x/ as a product of irreducible
polynomials.
We can refine the statements of the last two theorems to obtain the next
two.
Theorem 7.11. Let n be an integer greater than 1. There exist prime numbers p1 ; : : : ; pk and positive integers m1 ; : : : ; mk satisfying
m

n D p1 1


pk k :
Theorem 7.12. Let K be a field and let f .x/ be a polynomial of positive
degree in Kx. Then there exist irreducible polynomials p1 .x/; : : :, pk .x/
in Kx and positive integers m1 ; : : : ; mk satisfying
f .x/ D p1 .x/m1


pk .x/mk :
The irreducible factors of f .x/ of degree 1 in the factorization of Theorem 7.12 correspond to roots, whereas irreducible factors of higher degree
have no roots.
A counterpart to the existence of prime factorizations for integers is the
famous theorem that such factorizations are unique. The formal statement
of this requires some care. Heres one version:
Theorem 7.13. Let n be an integer greater than 1. Suppose
m

p1m1


pk k
and
n

q1n1


q` `
are two prime factorizations of n, for prime numbers p1 < p2 <


< pk
and q1 < q2 <


< q` and for positive integer exponents mi and nj . Then
k D ` and for each i between 1 and k, we have the equalities pi D qi and
mi D ni .
The key to proving Theorem 7.13 is a property of prime numbers that
is analogous to the result we proved in Exercise 1.9 for polynomials of the
form x a: if a prime number p divides a product r s of positive integers,
then p divides r or p divides s. This result goes back to Euclid. It is a
consequence of the result known as the euclidean algorithm for finding the
greatest common divisor of two positive integers. Given it, we can prove
Theorem 7.13 in much the same way as Theorem 7.8. The difficulties are
more ones of organization than of conception.

i
i

i
IrvingBook 2013/5/22 15:39 page 198 #214

198

7. Higher-Degree Polynomials

Euclids key property of prime numbers has its counterpart for irreducible polynomials: given a field K, if an irreducible polynomial p.x/
in Kx divides the product r .x/s.x/ of two polynomials r .x/ and s.x/
in Kx, then p.x/ divides r .x/ or s.x/. This is an extension of Exercise
1.9. We will not prove it. The proof is not difficult, paralleling as it does
the proof of the integer result. We introduce the notion of greatest common
divisor for polynomials, obtain a polynomial analogue of the euclidean algorithm, and prove the result, mimicking the classical integer arguments.
Once this property of irreducible polynomials is available, we can mimic
the proof of Theorem 7.13 to obtain what is essentially an extension of Theorem 7.8, a unique factorization theorem for polynomials. The statement is
more complicated than its integer counterpart because we can alter a factorization of a polynomial by inserting constant factors without changing
it in an essential way, but this is the only issue: up to constant factors, the
same irreducible polynomials occur in any factorization of a polynomial
into irreducible polynomials with the same exponents.
With these general factorization theorems in place, we may next wish to
study factorization problems with particular fields of coefficients, such as Q
or F2 . They turn out to be important, and difficult. Let us instead conclude
by returning to the issue that prompted us to introduce the notion of a field
in Section 7.2.
We saw that several eighteenth-century attempts to prove the fundamental theorem of algebra failed because of a missing ingredient, Theorem 7.5.
This states, given a polynomial f .x/ in Rx of positive degree n, that there
exists a field K containing C and elements r1; r2 ; : : : ; rn in K satisfying
f .x/ D .x

r1 /.x

r2 /


.x

rn /:

The proof does not depend on working with the real numbers. It yields more
generally, with no additional work, the following theorem.
Theorem 7.14. Let K be a field and let f .x/ be a polynomial of positive degree in Kx. There exists a field L containing K and elements r1 ; r2 ; : : : ; rn
in L, repetitions allowed, such that
f .x/ D .x

r1 /.x

r2 /


.x

rn /

in Lx.
The proof is not difficult, and can be found in many algebra texts (for
example, [22, p. 536]). A full discussion would take us too far afield. Let us
take a brief look at the essential issue, which is the following partial result:

i
i

i
IrvingBook 2013/5/22 15:39 page 199 #215

199

7.3. Polynomial Factorization

Theorem 7.15. Let K be a field and let f .x/ be a polynomial of positive

degree in Kx. There exists a field L containing K and a root r of f .x/ in
L.
If we can prove Theorem 7.15, we can iterate its construction to prove
Theorem 7.14.
The construction of L is surprisingly easy. All we need to do is mimic
the construction of C from R, which was done to build a field larger than
R that contains a root of x 2 C 1. In that construction, we created a new
element, i , took C to be the set of polynomial expressions of degree at most
one in i with real coefficients, and defined addition and multiplication in the
obvious way, taking into account that i must satisfy i 2 C1 D 0, or i 2 D 1.
For Theorem 7.15 we can assume f .x/ is monic. Suppose it has the
form
x n an 1 x n 1


a1 x a0 ;
where we choose minus signs to simplify what follows. We introduce a new
element r that will be the analogue for f .x/ of i for x 2 C 1. We want r to
be a root of f .x/, which means r satisfies
rn

an

1r

n 1

a1 r

a0 D 0;

or
r n D an

1r

n 1

C


C a1 r C a0 :

We then take L to be the set of all polynomial-like expressions in r of degree

at most n 1 with coefficients from K.
Addition of expressions is defined in the obvious way. For multiplication, we first treat r as a polynomial indeterminate and multiply as usual.
The result will be a polynomial in r , but it may have degree n or greater. We
then use the last equation, which we can think of as a rewrite rule telling us
how to replace r n by a sum of lower-degree terms in r . Iterating, we can
write any polynomial in r as a polynomial in r of degree less than n. Having
defined addition and multiplication in L, we easily verify that L satisfies the
defining conditions of a ring.
What requires closer examination is the additional requirement that fields
must satisfy: that every non-zero element has a multiplicative inverse. Is this
the case for L? It is not difficult to see that the answer is no whenever f .x/
fails to be irreducible. However, if f .x/ is irreducible, then L is a field. The
proof of this, like the proof of unique factorization for polynomials, depends
on polynomial analogues of integer results on greatest common divisors and
the euclidean algorithm.

i
i

i
IrvingBook 2013/5/22 15:39 page 200 #216

200

7. Higher-Degree Polynomials

Theorem 7.15 follows. If f .x/ is irreducible, L is a field and we are

done. Otherwise, we can replace f .x/ with one of its irreducible factors
g.x/ in Kx and apply the same construction to g.x/. The result will be a
field containing a root of g.x/ and hence a root of f .x/ as well.
The idea behind enlarging R in order to create a new field containing
a square root of 1 is a powerful one. We see now that it can be used to
expand any field to a larger one containing roots of a given polynomial.

7.4

Symmetric Polynomials

In our study of quadratic, cubic, and quartic polynomials, we found formulas for the discriminant in terms of the polynomials coefficients. The
discriminant can be defined for a polynomial of any degree, and the lowerdegree examples suggest the possibility that it can again be written in terms
of the coefficients. There are two issues. Does such an expression exist? If
so, what is it? The answer to the first is yes, as we will see in this section.
Finding the formula turns out to be a more difficult problem, which we will
not pursue.
The theorem that the discriminant of a polynomial can be expressed in
terms of the polynomials coefficients is a special case of a far-reaching
result on symmetric expressions in the roots of a polynomial. An example
of a symmetric expression is r 2 s C r s 2, which remains unchanged if we
swap r and s. In contrast, r 2s C 2r s 2 changes under the swap and therefore
is not symmetric. The general resultwhich will apply to discriminants
is that any symmetric expression in a polynomials roots can be written in
terms of the polynomials coefficients. We will develop the relevant ideas in
this section, and conclude with some history.
Lets begin by defining the discriminant. Suppose f .x/ is a polynomial
of positive degree n, of the form
xn

a1 x n

C a2 x n

C . 1/n

an

1x

C . 1/n an :

We index the coefficients so that the subscript of a coefficient and the exponent of its associated power of x sum to n, and also we write the coefficients
with alternating signs. The convenience of these choices will soon be clear.
Assume that the coefficients are real numbers.
By the fundamental theorem of algebra, f .x/ factors as
.x

r1 /.x

r2/


.x

rn /;

where r1 ; : : : ; rn are the real or complex roots of f .x/, listed with possible
repetitions. The discriminant of f .x/ is defined to be the square of the

i
i

i
IrvingBook 2013/5/22 15:39 page 201 #217

201

7.4. Symmetric Polynomials

product of differences of all the pairs of roots. In product notation,

Y
D
.ri rj /2 :
1i <j n

We take the product over all roots, with repetitions, not over the set of distinct roots. If there is a repeated root, which means there is a pair of indices
i and j with i j and ri D rj , then the difference ri rj is 0 and D 0. If
no root is repeated, then the factors in the product are non-zero and 0.
We are assuming that coefficients are real numbers. We could choose to
work in the generality of Section 7.3, with an arbitrary field K as the field of
coefficients for the polynomial f .x/. Theorem 7.14 ensures the existence
of a larger field L that contains a set r1 ; : : : ; rn of roots for f .x/ for which
f .x/ factors in Lx as
f .x/ D .x

r1 /.x

r2 /


.x

rn /:

We can then define the discriminant of f .x/ as the product

Y
D
.ri rj /2 :
1i <j n

However, we would have to prove that doesnt depend on the field L and
roots ri . This is a journey we choose not to take.
For polynomials f .x/ of degree 2, 3, or 4, we found formulas for the
discriminant that express it as a sum of integer multiples of products of the
coefficients. For example, we found in Theorem 5.6 that the discriminant of
x 3 C bx 2 C cx C d is
18bcd

4b 3 d C b 2c 2

4c 3

27d 2:

We wish to show for f .x/ of any degree n that a formula like this exists:
the discriminant of f .x/ can be written as a polynomial expression in the
coefficients of f .x/ with integer coefficients.
The discriminant is not the only expression in the roots of f .x/ that we
have written as a polynomial expression in its coefficients. In Theorem 5.14
we saw for a cubic polynomial x 3 a1 x 2 C a2 x a3 with roots r1 , r2 , and
r3 that
a1 D r1 C r2 C r3 ;
a2 D r1 r2 C r1 r3 C r2 r3 ;
a3 D r1 r2 r3 :
The proof amounts to factoring the cubic as .x r1 /.x r2 /.x r3 / and carrying out the multiplication of the degree-one terms. The simpler quadratic

i
i

i
IrvingBook 2013/5/22 15:39 page 202 #218

202

7. Higher-Degree Polynomials

analogue was obtained for real roots in Exercise 2.5 and in general in Exercise 4.8, and the quartic analogue was Theorem 6.11.
A similar result holds for polynomials of higher degree. We have factored a general degree n polynomial f .x/ as
.x

r1 /.x

r2 /


.x

rn

1 /.x

rn /:

When we carry out the multiplication of the n terms, the result is the sum
of the expressions we obtain by choosing from each term x ri either the
x or the ri and then multiplying them together. For an integer k between
1 and n, the coefficient of x n k will be the sum of the terms we obtain
when we choose x from n k of the factors and choose ri s from k of
the factors. We might for instance choose constants from the factors with
indices i1 ; i2; : : : ; ik . This would yield
. 1/ri1
. 1/ri2


. 1/rik
x n

or
. 1/k ri1 ri2


rik x n

To get the entire coefficient of x n k , which we have written as . 1/k ak ,

we need to add all these expressions together. The result is
X
ak D
ri1 ri2


rik :
1i1 <i2 <


<ik n

The notation is formidable, but what it says is that ak is the sum of

all products of k of the roots ri . This gives us a sequence of polynomial
expressions in the roots that equal the simplest of polynomial expressions
in the coefficients: the coefficients themselves, up to sign. Two special cases
are
a1 D r1 C r2 C


C rn
and
an D r1 r2


rn :
We have now introduced several polynomial expressions in the roots
of a polynomial f .x/ that can be written in terms of its coefficients: the
discriminant and the sums of products of roots. What connects them? Each
remains unchanged if we reorder the roots. The roots are specific complex
numbers, but we have assigned names to themr1, r2 , and so onthat do
not take into account which is which. Our naming is arbitrary. When we
sum them, or sum their two-by-two products, or sum their fifth powers, we
are forming quantities that are independent of our choice of names. This

i
i

i
IrvingBook 2013/5/22 15:39 page 203 #219

203

7.4. Symmetric Polynomials

is the case as well for the discriminant, a point that is worth a moment of
thought.
By definition,
Y
.r1 ; : : : ; rn / D
.ri rj /2 :
1i <j n

If we reorder the roots ri in some way, the order of the factors in the product
will change, but the list of factors is unchanged: we choose every possible
pair of roots, take its difference, square it, and multiply. Since multiplication doesnt depend on the ordering, the product that defines .r1 ; : : : ; rn /
doesnt either. If we were to omit the squares in the product, the resulting
product
Y
.ri rj /
1i <j n

would fail to be independent of the ordering of the ri s. For example, consider what happens if we switch the order of r1 and r2 but leave the order
of r3 ; : : : ; rn intact. The factor r1 r2 changes to its opposite, r2 r1 . For
any i > 2, the factors r1 ri and r2 ri switch with each other, but their
product is unchanged. The remaining factors ri rj with i; j > 2 remain
unchanged. The product as a whole, then, is changed to its opposite.
Expressions in the roots of f .x/ that remain unchanged under re-naming
of the roots, such as the examples we have just examined, are called symmetric.
To proceed, we pass to a more abstract framework. The formulas we
have obtained relating roots and coefficients of a polynomial make sense
whatever the values of the coefficients are. In effect, we can regard them
as variables, to be replaced by particular real numbers when we identify
a specific polynomial of interest. If we are studying the polynomial x 3 C
3x 2 2x C 7, for example, then we substitute 3 for a1 , 2 for a2 and 7
for a3 (keeping our sign convention in mind).
This suggests the path we should take, which is to start over, working
with a generic polynomial of degree n whose coefficients are variables or
indeterminates. We can in this way discuss all degree n polynomials at once,
specializing the variable coefficients to actual ones when we have a specific
polynomial in mind.
Let us begin anew, then. Not only should the coefficients of a degree
n polynomial be generic, but the roots should be too. In other words, we
replace the roots ri by variables. Fix a positive integer n and introduce the
new variables or indeterminates t1 ; t2 ; : : : ; tn , which we think of as standins for the roots of a degree n polynomial. We anticipate that when we study

i
i

i
IrvingBook 2013/5/22 15:39 page 204 #220

204

7. Higher-Degree Polynomials

a specific polynomial, we will substitute the complex numbers that arise as

roots of the polynomial for the ti s.
We are interested in polynomial expressions in the ti s with integer coefficients. The most general such expression can be written as
X
ai1 ;i2 ;:::;in t1i1 t2i2


tnin :

It is understood that the sum is finite and the coefficients ai1 ;i2 ;:::;in are integers, all but finitely many of them being 0. Just as the ti s are stand-ins for
the roots of a given degree n polynomial, the expressions are stand-ins for
polynomial expressions in the roots.
We already know some expressions that will interest us. For each k
between 1 and n, lets define ek .t1 ; : : : ; tn / by
X
ti1 ti2


tik
ek .t1 ; : : : ; tn / D
1i1 <i2 <


<ik n

and call it the kth elementary symmetric polynomial. Repeating the analysis
we made for sums of products of roots, we obtain:
Theorem 7.16. Let n be a positive integer and let t1 ; t2 ; : : : ; tn be n indeterminates. Then
n
X
.x t1 /.x t2 /


.x tn 1 /.x tn / D x n C
. 1/k ek .t1 ; : : : ; tn /x n k
kD1

We also introduce, for each positive integer k, the power sum polynomials pk .t1 ; : : : ; tn /, defined by
pk .t1 ; : : : ; tn / D t1k C


C tnk ;
with p0 .t1 ; : : : ; tn / D 1. For n  2, we introduce the polynomial .t1 ; : : : ; tn /
defined by
Y
.t1 ; : : : ; tn / D
.ti tj /2 :
1i <j n

We call this the discriminant.

Lets return for a moment to the polynomial f .x/ with which we worked
earlier in this section. It had the form
f .x/ D x n

a1 x n

C a2 x n

C . 1/n

an

C . 1/n an :

with roots r1 ; : : : ; rn and factorization

.x

r1/.x

r2 /


.x

rn /:

i
i

i
IrvingBook 2013/5/22 15:39 page 205 #221

205

7.4. Symmetric Polynomials

If, for each index i from 1 to n, we substitute ri for ti , then we pass from
the generic polynomial
xn C

n
X

. 1/k ek .t1 ; : : : ; tn /x n

kD1

to the specific polynomial

f .x/ D x n

a1 x n

C a2 x n

C . 1/n

an

C . 1/n an :

The coefficients ai , which are real numbers, can be expressed in terms of

the roots rj :
ai D ei .r1 ; : : : ; rn /:
Similarly, when we specialize the variables ti to be the roots ri of f .x/, the
resulting number
.r1 ; : : : ; rn /
is the discriminant of f .x/.
We have observed that certain expressions in the roots of a polynomial,
such as its discriminant, remain unchanged when we reorder the roots. We
can now study such expressions.
We say a polynomial p.t1 ; : : : ; tn / in the variables ti with integer coefficients is symmetric if any reordering of the variables leaves the polynomial
unchanged. We can introduce notation to make this more precise. We define
a permutation of the set of numbers 1; : : : ; n to be a function  that takes
the set to itself so that any two numbers i j are moved to two numbers .i / and .j / that remain unequal: .i / .j /. With that notation,
p.t1 ; : : : ; tn / is symmetric if for every permutation  ,
p.t.1/ ; : : : ; t.n// D p.t1 ; : : : ; tn /:
Our earlier observations made for a specific polynomials roots carry
over: the elementary symmetric polynomials, power sum polynomials, and
discriminant are all examples of symmetric polynomials. In contrast, the
square root of the discriminant (obtained by omitting the squares in the
definition of the discriminant) fails to be symmetric, as will just about any
polynomial expression in the ti s with integer coefficients that we write down
at random.
It is easy to see that integer multiples, sums, and products of symmetric
polynomials are symmetric. We can form sums of integer multiples of products of the elementary symmetric polynomials and obtain new symmetric

i
i

i
IrvingBook 2013/5/22 15:39 page 206 #222

206

7. Higher-Degree Polynomials

polynomials. For example, with n D 2, since e1 .t1 ; t2/ and e2 .t1 ; t2/ are
symmetric, so is
e1 .t1 ; t2 /2 4e2 .t1 ; t2 /:
Expanding, we find that
e1 .t1 ; t2 /2

4e2 .t1 ; t2 / D .t1 C t2 /2

4t1 t2 D .t1

t2 /2 D .t1 ; t2/:

Thus, we have written the discriminant .t1 ; t2 / as a polynomial expression

in elementary symmetric polynomials with integer coefficients.
This expression is familiar. Whats new is the context we have now established in which we can ask whether such expressions exist for any n.
That is, for a positive integer n, is .t1 ; : : : ; tn / equal to a polynomial expression in the elementary symmetric polynomials ei .t1 ; : : : ; tn / with integer coefficients? If such an expression exists, then when we specialize the
ti s to the roots ri of a polynomial f .x/, the elementary symmetric polynomials specialize to the values ei .r1 ; : : : ; rn/, which are up to sign the
coefficients of f .x/.
Our main result is that every symmetric polynomial, not just the discriminant, can be obtained as a polynomial expression in the elementary
symmetric polynomials. This is the fundamental theorem on symmetric
polynomials.
Theorem 7.17. Let n be a positive integer. Every symmetric polynomial
in the variables t1 ; : : : ; tn with integer coefficients is an integer-coefficient
polynomial expression in the elementary symmetric polynomials
e1.t1 ; : : : ; tn /; : : : ; en .t1 ; : : : ; tn /:
That is, given a symmetric polynomial p.t1 ; : : : ; tn /, there is an n-variable
integer-coefficient polynomial q for which
p.t1 ; : : : ; tn / D q.e1 .t1 ; : : : ; tn /; : : : ; en .t1 ; : : : ; tn //:
As a special case of Theorem 7.17, we obtain:
Theorem 7.18. Let n be an integer, n  2. The discriminant
Y
.t1 ; : : : ; tn / D
.ti tj /2
1i <j n

is a polynomial expression in the elementary symmetric polynomials with

integer coefficients.

i
i

i
IrvingBook 2013/5/22 15:39 page 207 #223

207

7.4. Symmetric Polynomials

We obtain the same result for power sum polynomials, too, but we can
say more, as we will later in the section.
To see the power of Theorem 7.18, lets return to our standard degree n
polynomial
xn

a1 x n

C a2 x n

C . 1/n

an

C . 1/n an

with roots r1 ; : : : ; rn . Suppose we are able to find the explicit n-variable

polynomial q for which
.t1 ; : : : ; tn / D q.e1 .t1 ; : : : ; tn /; : : : ; en .t1 ; : : : ; tn //:
We know that when we substitute the roots ri for the variables ti , the elementary symmetric polynomials ek specialize to the coefficients ak :
ek .r1 ; : : : rn / D ak . We can therefore use .t1 ; : : : ; tn / to calculate the
discriminant .r1 ; : : : ; rn / of f .x/ in terms of the coefficients ak :
.r1 ; : : : ; rn / D q.e1 .r1 ; : : : ; rn /; : : : ; en .r1 ; : : : ; rn //
D q.a1 ; a2 ; : : : ; an /:

We have obtained expressions for .t1 ; : : : ; tn / in terms of the elementary symmetric polynomials for n D 2; 3; 4. Theorem 7.18 ensures that
there is such an expression for any n. They can be quite complicated. For
example, the discriminant of the principal quintic polynomial
x 5 C cx 2 C dx C e;
is given by
D 108c 5e

27c 4 d 2 C 2250c 2de 2

1600cd 3e C 256d 5 C 3125e 4:

The formula for the discriminant of a general quintic polynomial is a sum

of 59 terms! As n increases, it becomes increasingly difficult to determine
the expression for the discriminant.
The proof of Theorem 7.17 is not difficult conceptually, but requires a
good choice of notation. Lets see how it is proved for n D 2.
Since we have only two variables, we will dispense with indices. Lets
write the variables as t and u. The elementary symmetric polynomials are
t C u and tu. What we are to prove is that any symmetric polynomial in t
and u is a polynomial expression in t C u and tu.
Suppose p.t; u/ is a symmetric polynomial. Since there are only two
variables t and u, being symmetric means that if we switch t and u in the
expression for p.t; u/, we leave the polynomial unchanged. In particular, if

i
i

i
IrvingBook 2013/5/22 15:39 page 208 #224

208

7. Higher-Degree Polynomials

t m un occurs in p.t; u/ with a non-zero coefficient a, then t n um must also

occur in p.t; u/ with coefficient a. It follows that p.t; u/ must be a sum of
integer multiples of terms of the form
t m un C t n um ;
for non-negative integers m and n with m < n, and integer multiples of
terms of the form
t m um ;
for non-negative integers m. Therefore, if we show that these expressions
can be written as polynomials in t C u and tu, it will follow that every
symmetric polynomial in t and u can be written in terms of t C u and tu.
We can simplify further. Given non-negative integers m < n, let n D
m C k. Then
t m un C t n um D t m um .t k C uk / D .tu/m .t k C uk /:
Thus we can combine the two cases into one, and it suffices to show for all
non-negative integers m and k that
.tu/m .t k C uk /
is a polynomial expression in t C u and tu. Since .tu/m is already such
an expression, we need only show for every non-negative integer k that the
sum t k C uk is a polynomial expression in t C u and tu.
This is easily done. If k D 0, the sum t 0 C u0 is 1, and if k D 1, it is
t C u. These are both polynomial expressions in t C u. Suppose we already
know for an integer k > 1 that t k 2 Cuk 2 and t k 1 Cuk 1 are polynomial
expressions in t C u and tu. Then
t k C uk D .t k

C uk

/.t C u/

tu.t k

C uk

shows that t k C uk is also a polynomial expression in t C u and tu. In this

way, starting from the knowledge that our result holds for k D 0 and k D 1,
we can work our way through all possible positive integers k, concluding
for any positive integer k that the sum t k C uk is a polynomial expression
in t C u and tu. (We are implicitly using the principle of mathematical
induction, which will be discussed in Section 7.5.) By the earlier reductions,
we have proved that every symmetric polynomial in t and u is a polynomial
expression in t C u and tu. This proves Theorem 7.17 for polynomials in
two variables.

i
i

i
IrvingBook 2013/5/22 15:39 page 209 #225

209

7.4. Symmetric Polynomials

Theorem 7.17 is proved for polynomials in more variables by a similar

bootstrap process. The organization of the proof is more complicated, but
the idea is the same. See [22, pp. 621622] for an outline.
In proving Theorem 7.17 for two variables t and u, we were able to
reduce to the case of the power sum polynomials t k C uk , which we then
proved to be polynomial expressions in t C u and tu. For general n, Theorem 7.17 guarantees that the power sum polynomials pk .t1 ; : : : ; tn / can
be expressed as polynomials in e1 .t1 ; : : : ; tn /; : : : ; en.t1 ; : : : ; tn / with integer coefficients. But we can do better, obtaining an explicit formula that
relates power sum polynomials and elementary symmetric polynomials. In
this formula, Theorem 7.19, it is understood that e0 .t1 ; : : : ; tn / D 1 and
ek .t1 ; : : : ; tn / D 0 for k > n. We omit the proof. (See [22, pp. 618619]
for a discussion.)
Theorem 7.19 (Girard-Newton identities). Let k and n be positive integers.
Then
kek .t1 ; : : : ; tn / D

k
X

. 1/i

ek i .t1 ; : : : ; tn /pi .t1 ; : : : ; tn /:

i D1

By using e0.t1 ; : : : ; tn / D 1 and moving terms around, we can re-write

the kth Girard-Newton identity as
pk .t1 ; : : : ; tn / D . 1/k

kek .t1 ; : : : ; tn /

k 1
X

. 1/k i ek i .t1 ; : : : ; tn /pi .t1 ; : : : ; tn /:

i D1

When k D 1, this is
e1 .t1 ; : : : ; tn / D p1 .t1 ; : : : ; tn /;
which holds since both sides are equal to t1 C


C tn by definition. From
this starting point, as k increases, the other Girard-Newton identities allow
us to express each power sum pk .t1 ; : : : ; tn / recursively as a polynomial in
the elementary symmetric polynomials. When we specialize to a degree n
polynomial
xn

a1 x n

C a2 x n

C . 1/n

an

C . 1/n an ;

we are able to write the power sums pk .r1 ; : : : ; rk / in terms of the coefficients ai . Girard wrote such formulas explicitly for k  4. Lets conclude
the section with a look at his work.

i
i

i
IrvingBook 2013/5/22 15:39 page 210 #226

210

7. Higher-Degree Polynomials

Prior to Girard, Vi`ete obtained formulas in special cases for the coefficients of a polynomial as sums of products of the roots. Girard was the first
to state such formulas in general. In his 1629 book Invention Nouvelle en
lAlg`ebre [32], Girard stated the theorem that any polynomial of degree n
has n roots. Only with the presence of n roots can we even contemplate expressions for the coefficients of the polynomial in terms of the roots. Girard
provided them in the second part of the same theorem.
Before stating the theorem, Girard introduced a notion he called a faction:
When several numbers are proposed, their sum is called the first faction; the sum of all the products two-by-two is called the second faction; the sum of all the products three-by-three is called the third faction; and so on up to the end, with the product of all the numbers called
the last faction. There are as many factions as the numbers proposed.
The second part of the theorem then states:
The first faction of the solutions [of a polynomial equation] equals the
first coefficient, the second faction of the same is equal to the second
coefficient, the third the third, and so on, so that the last faction is equal
to the last, and this according to the signs, which one observes are in
alternating order.
Girard offers no proof. But the proof is clear, provided we understand a
polynomial of degree n to have n roots, as Girard states, and provided we
understand that the polynomial factors as the product of terms x r , as r
runs through the n roots, which he surely did.
A few pages later, Girard introduces the expressions in the roots ri that
we call the power sum polynomials. He then writes formulas for them in
terms of the coefficients ai , for powers up to 4:
r1 C


C rn
r12 C


C rn2
r13 C


C rn3
r14 C


C rn4

D
D
D
D

a1 ;
a12 2a2;
a13 3a1 a2 C 3a3 ;
a14 4a12a2 C 4a1 a3 C 2a22

4a4 :

These are the first four Girard-Newton identities of Theorem 7.19. Girard
doesnt continue with higher exponents, but it is clear that a sequence of
such formulas exists.
Isaac Newton (16431727) obtained the same result in 1707 [47, pp.
392393], again without proof. (In supplementary notes immediately after
Newtons statement of the result, Reverend Wilder provides a proof [47, pp.

i
i

i
IrvingBook 2013/5/22 15:39 page 211 #227

7.5. A Proof of the Fundamental Theorem

211

393396].) The identities are often named after Newton alone, but credit is
due to both.
Girards observation that certain symmetric expressions in the roots of a
polynomial can be written in terms of the polynomials coefficients comes
in for praise in H. Gray Funkhousers 1930 paper, A short account of the
history of symmetric functions of roots of equations [30, pp. 358, 360]:
As is the case with the development of algebra as a whole, the subject of symmetric functions to a perhaps greater extent waited upon
the introduction and improvement of symbolism. It was hard to perceive with any clarity relations between roots and coefficients when
the equation was wrapped up in a paragraph of words representing it.
Prior to the time of Franciscus Vieta, who was among the first to employ letters to represent numbers, we find very little trace of the thing
we are seeking. . . .
We can appreciate how great an advance Vieta made when there is
considered the few fragmentary statements that have come from his
predecessors. He paves the way and provides an introduction to the
first man who really has a place in the history of symmetric functions
of roots of equations, a man, who for clearness and grasp of material
at hand in not only this topic but also in other phases of algebra could
well hold his place a century later.
The man was Albert Girard and his work on algebra is a little 34-leaf
pamphlet called Invention Nouvelle en lAlg`ebre, published in 1629.

7.5 A Proof of the Fundamental Theorem

The fundamental theorem of algebra states that every polynomial with real
coefficients has a root in the complex numbers. As we discussed in Section
7.2, the theorems name is misleading. Any proof depends on some information about the real numbers, hence on results from calculus or analysis.
At the end of Section 7.2, we mentioned the existence of proofs that are entirely algebraic if we take two results as given: every positive real number
has a real square root (Theorem 1.10) and every cubic polynomial with real
coefficients has a real root (Theorem 1.14). With the ideas of Section 7.4,
we can now describe one such proof, essentially due to Laplace in 1795.
That is the purpose of this section.
In addition to Theorems 1.10 and 1.14, our proof of the fundamental
theorem will use two results developed in this chapter: Theorem 7.5, which

i
i

i
IrvingBook 2013/5/22 15:39 page 212 #228

212

7. Higher-Degree Polynomials

states for a polynomial f .x/ with real coefficients that a field K exists containing both C and a set of roots of f .x/, so that f .x/ factors in Kx as
a product of linear factors, and Theorem 7.17, the fundamental theorem on
symmetric polynomials. We will also need the principle of mathematical
induction.
Induction has been used implicitly on occasion, for instance in the proof
of de Moivres formula in Exercise 4.23 and the proof of the two-variable
case of the fundamental theorem on symmetric polynomials in Section 7.4.
Its application in proving the fundamental theorem is more subtle. Hence, in
preparation for that proof, an explicit discussion of induction is warranted.
Lets write N for the set of positive integers. (This is a traditional mathematical notation, the letter N suggesting the natural numbers, another
name for the positive integers.) Here is the statement of the principle:
Theorem 7.20 (principle of mathematical induction). Let S be a subset of
N. Assume that S contains 1 and that if S contains a positive integer k,
then S contains k C 1. Then S D N.
It is perhaps misleading to call the induction principle a theorem. In
developing the foundations of arithmetic, we typically adopt the principle
as an axiom to be satisfied by the positive integers rather than a theorem to
be proved about them. But this is a subject for another book (for instance,
[42, pp. 149150]). Whatever the principles logical status, we will accept
it as a valid statement.
The principle becomes a proof technique when we want to verify a family of statements indexed by the positive integers. De Moivres formula is
an example. We wish to prove for every positive n that
.cos  C i sin /n D cos n C i sin n:
Let S be the set of positive integers for which the equality holds. We see immediately that S contains 1. Suppose we can show for any positive integer
k that if k is in S , so is k C 1. Then the principle of mathematical induction
implies that S D N and de Moivres formula holds for all positive integers.
What we must show, then, is that if
.cos  C i sin /k D cos k C i sin k;
then
.cos  C i sin /kC1 D cos.k C 1/ C i sin.k C 1/:
This is essentially what we checked in Exercise 4.23.

i
i

i
IrvingBook 2013/5/22 15:39 page 213 #229

7.5. A Proof of the Fundamental Theorem

213

Sometimes we wish to prove a family of statements indexed not by N

but by the set N [ f0g of non-negative integers. The principle of induction
works. We prove that the statement associated with 0 is true, then we prove
for any non-negative integer k that the validity of the .k C 1/st statement
follows from the validity of the kth statement.
We are ready to prove the fundamental theorem of algebra: any polynomial f .x/ of positive degree in Rx has a root in C. We might think of
proceeding inductively using the degree of f .x/ as the induction index. But
we already know the fundamental theorem is true for odd-degree polynomials, so induction on degree isnt likely to be a successful strategy. Instead,
following Laplace, we do something more clever, applying induction not to
the degree of f .x/ but to its degree of evenness.
A positive integer n factors as 2k s for some non-negative integer k and
positive odd integer s. What interests us is the exponent k. The smaller k is,
the closer n is to being odd. Or, the larger k is, the more even (so to speak)
n is. This exponent k is the basis for our induction argument.
The starting point for our induction is the case k D 0. The positive
integers for which k is 0 are the odd ones. Thus, we begin by considering
the odd-degree polynomials in Rx. By Theorem 1.14, they have real roots,
so the fundamental theorem holds for them. Our induction argument is off
to a successful start.
We next choose an integer k  0 and assume for all odd positive integers s that the fundamental theorem holds for polynomials of degree 2k s.
This is called the inductive hypothesis. Using it, we wish to prove for all
odd positive integers t that the fundamental theorem holds for polynomials
of degree 2kC1 t. If we do, then by the principle of mathematical induction,
we will have shown that the fundamental theorem holds for polynomials of
degree 2` u, for all non-negative integers ` and all odd positive integers u.
In other words, it holds for all polynomials of positive degree, as desired.
Suppose f .x/ is a polynomial in Rx of degree n, with n of the form
2kC1 t and t odd. The induction argument requires us to show that f .x/ has
a root in C. By Theorem 7.5, there is a field K containing C that contains
roots r1 ; r2 ; : : : ; rn of f .x/ (repetitions allowed), with f .x/ factoring in
Kx as
.x r1 /


.x rn /:
Fix an arbitrary integer m. For a pair of integers i and j satisfying 1  i <
j  n, we can form in K the element
ri C rj C mri rj :
Since there are n.n

1/=2 pairs i and j , there will be n.n

1/=2 such

i
i

i
IrvingBook 2013/5/22 15:39 page 214 #230

214

7. Higher-Degree Polynomials

elements. Therefore, the polynomial

Y
F .x/ D
.x .ri C rj C mri rj //
i <j

in Kx has degree n.n 1/=2.

If we permute the roots r1 ; : : : ; rn , we permute the n.n 1/=2 elements
ri C rj C mri rj , leaving them unchanged as a set. Therefore we also leave
unchanged the coefficients of the polynomial F .x/. But these are polynomial expressions in the roots r1 ; : : : ; rn of f .x/ with integer coefficients.
Therefore, by a suitable application of the fundamental theorem on symmetric polynomials, they are polynomial expressions in the coefficients of
f .x/ with integer coefficients, which is to say, they are real numbers.
We have proved that F .x/ lies in Rx. Recall that f .x/ has degree n,
with n D 2kC1 t and t odd. Since k C 1 is positive, n is even. Hence, n 1
and t.n 1/ are odd. But F .x/ has degree n.n 1/=2 and
n.n

1/
2

2kC1 t.n
2

1/

D 2k t.n

1/:

Thus, the degree of F .x/ is exactly of the form for which we assumed by
the inductive hypothesis that the fundamental theorem holds: the product of
2k and an odd number. We conclude that F .x/ has a root in C, which means
that there is a fixed pair fi; j g (depending on m) for which the element ri C
rj C mri rj of K lies in the smaller field C. At the risk of over-complicating
the notation, we might write this element as ri.m/ C rj.m/ C mri.m/ rj.m/ to
emphasize its dependence on the choice of the integer m.
We chose m arbitrarily. The argument we just made applies for each
choice of m, and there are infinitely many such choices. But there are only
n.n 1/=2 pairs of integers fi; j g between 1 and n. Therefore, there must be
distinct integers m and m0 yielding the same pair; that is, there exist integers
m m0 for which i.m/ D i.m0 / and j.m/ D j.m0 /.
Lets simplify our notation, writing i and j for the integers in this pair.
Then ri C rj C mri rj and ri C rj C m0 ri rj are both complex numbers, so
their difference .m m0 /ri rj is as well. Since m m0 is a non-zero integer,
we can divide by it to deduce that ri rj is complex. Hence mri rj is complex
and we can subtract it from the complex number ri C rj C mri rj to deduce
that ri C rj is complex. We have proved that f .x/ has two roots ri and rj
in the field K whose sum ri C rj and product ri rj lie in the smaller field C.
There is one last step. The polynomial .x ri /.x rj / equals
x2

.ri C rj / C ri rj :

i
i

i
IrvingBook 2013/5/22 15:39 page 215 #231

7.5. A Proof of the Fundamental Theorem

215

Since ri C rj and ri rj are complex, it lies in Cx. By design, its roots are
ri and rj . We proved in Theorem 4.7 that any quadratic polynomial in Cx
has complex roots. Hence, ri and rj lie in C. But ri and rj are roots of
f .x/, so f .x/ has roots in C.
We have proved, under the assumption that polynomials whose degree
is a product of 2k and an odd number have roots in C, that polynomials
whose degree is a product of 2kC1 and an odd number also have roots in C.
We also proved the base case, that polynomials of odd degree have roots in
C. By the principle of mathematical induction, polynomials of any positive
degree have roots in C, proving the fundamental theorem.
Lets review where we used our two assumptions. The existence of real
roots for odd-degree polynomials entered the stage at the beginning, as the
base case of the induction argument. The existence of real square roots for
positive real numbers was used, implicitly, at the end, when we quoted Theorem 4.7 on quadratic polynomials in Cx having roots in C. This followed
by a completing-the-square argument from the existence in C of square
roots of complex numbers (Theorem 4.6), which followed from the existence of real square roots for positive real numbers.
What a beautiful proof! It depends on several of the ideas we have introduced in this chapter, plus a touch of genius. It is a perfect peak on which
to bring our study of polynomials to a close.

i
i

i
IrvingBook 2013/5/22 15:39 page 216 #232

i
i

i
IrvingBook 2013/5/22 15:39 page 217 #233

References
[1] Niels Abel, Memoire sur les e quations algebriques ou lon demontre
limpossibilite de la resolution de lequation generale du cinqui`eme degre,
(1824) in uvres Compl`etes, Volume I, Grndahl & Sn, Christiania, 1881,
28-33; translated by P. Pesic in [52, pp. 155169].
[2] Irving Adler, The Giant Golden Book of Mathematics, illustrated by Lowell
Hess, Golden Press, New York, 1960.
[3] Lars V. Ahlfors, Complex Analysis, 3rd ed., McGraw-Hill, New York, 1979.
[4] Jean-Robert Argand, Imaginary Quantities: Their Geometrical Interpretation,
translated by A.S. Hardy with a preface by J. Houel, D. Van Nostrand, New
York, 1881.
[5] Raymond G. Ayoub, Paolo Ruffinis contributions to the quintic, Archive for
History of Exact Sciences, 23 (1980) 253277.
[6] Claude-Gaspard Bachet, Probl`emes Plaisants et Delectables, 4th ed.,
Gauthier-Villars, Paris, 1905.
[7] I.G. Bashmakova, Diophantus and Diophantine Equations, translated by Abe
Shenitzer with the editorial assistance of Hardy Grant, Mathematical Association of America, Washington, D.C., 1997.
[8] I.G. Bashmakova and G.S. Smirnova, The Beginnings and Evolution of Algebra, translated by Abe Shenitzer with the editorial assistance of David A. Cox,
Mathematical Association of America, Washington, D.C., 2000.
[9] William P. Berlinghoff and Fernando Q. Gouvea, Math Through the Ages, Oxton House Publishers, Farmington, Maine, 2002.
[10] A.E. Berriman, The Babylonian quadratic equation, The Mathematical
Gazette, 40 (1956) 185192.
[11] Rafael Bombelli, lAlgebra, Giovanni Rossi, Bologna, 1579.
[12] Girolamo Cardano, The Great Art; or, The Rules of Algebra, translated and
edited by T. Richard Witmer, The M.I.T. Press, Cambridge, Massachusetts,
1968.

217

i
i

i
IrvingBook 2013/5/22 15:39 page 218 #234

218

References

[13] Girolamo Cardano, The Book of My Life, translated by Jean Stoner with an
introduction by Anthony Grafton, New York Review Books, New York, 2002.
[14] Henry Thomas Colebrooke, Algebra with Arithmetic and Mensuration from
the Sanskrit of Brahmegupta and Bhascara, John Murray, London, 1817.
[15] S. Cuomo, Ancient Mathematics, Routledge, London, 2001.
[16] Abraham de Moivre, The analytic solution of certain equations of the third,
fifth, seventh, ninth and other higher uneven powers, by rules similar to those
called Cardans, Philosophical Transactions, 25 (1707) 23682371; portions
translated in [60, pp. 441444].
[17] Abraham de Moivre, On the reduction
to simpler terms, or the expof radicals p
traction of roots of any binomial a C Cb or a C
b, Philosophical Transactions, 40 (1739) 463-478; portions translated in [60, pp. 447450].
[18] Rene Descartes, A Discourse on the Method of Correctly Conducting Ones
Reason and Seeking Truth in the Sciences, translated by Ian Maclean, Oxford
University Press, Oxford, 2006.
[19] Rene Descartes, The Geometry of Rene Descartes, translated by David Eugene
Smith and Marcia L. Latham, Open Court Publishing Company, La Salle, Illinois, 1952.
[20] Leonard Eugene Dickson, Elementary Theory of Equations, John Wiley and
Sons, New York, 1914.
[21] Peter Doyle and Curt McMullen, Solving the quintic by iteration, Acta Mathematica, 163 (1989) 151180.
[22] David S. Dummit and Richard M. Foote, Abstract Algebra, 3rd ed., John Wiley
& Sons, New York, 2004.
[23] William Dunham, Euler: The Master of Us All, Mathematical Association of
America, Washington, D.C., 1999.
[24] Gotthold Eisenstein, Allgemeine Auflosung der Gleichung von den ersten vier
Graden, Journal fur die reine and angewandte Mathematik 27 (1844) 81-83;
reprinted in Mathematische Werke, Vol. I, Chelsea Publishing Company, New
York, 1975, 79.
[25] Gotthold Eisenstein, Entwicklung von , Journal fur die reine and angewandte Mathematik 28 (1844) 49-52; reprinted in Mathematische Werke, Vol.
I, Chelsea Publishing Company, New York, 1975, 122125.
[26] Leonhard Euler, Elements of Algebra, translated by Rev. John Hewlett,
Springer-Verlag, New York, Berlin, Heidelberg, Tokyo, 1984.
[27] Charles Fefferman, An easy proof of the fundamental theorem of algebra,
American Mathematical Monthly, 74 (1967) 854855.
[28] Fibonacci, Fibonaccis Liber Abaci: A Translation into Modern English of
Leonardo Pisanos Book of Calculation, translated by L.E. Sigler, SpringerVerlag, New York, 2002.
[29] Benjamin Fine and Gerhard Rosenberger, The Fundamental Theorem of Algebra, Springer-Verlag, New York, 1997.

i
i

i
IrvingBook 2013/5/22 15:39 page 219 #235

References

219

[30] H. Gray Funkhouser, A short account of the history of symmetric functions of

roots of equations, The American Mathematical Monthly, 37 (1930) 357365.

[31] Evariste
Galois, Memoire sur les conditions de resolubilite des e quations par

radicaux, in Ecrits
et Memoires Mathemathiques dEvariste
Galois, Robert
Bourgne and J.-P.Azra, eds., Gauthier-Villars & Cie, Paris, 1962, 4371.
[32] Albert Girard, Invention Nouvelle en lAlg`ebre, Guillaume Iansson Blaeuw,
Amsterdam, 1629.
[33] Anthony Grafton, Cardanos Cosmos, Harvard University Press, Cambridge,
Massachusetts, 1999.
[34] C. Hermite, Sur la resolution de lequation du cinqui`eme degre, Comptes Rend.
Acad. Sci. Paris 46 (1858) 508515.
[35] Jens Hyrup, Lengths, Widths, Surfaces, Springer-Verlag, New York, 2002.
[36] Louis Charles Karpinski, Robert of Chesters Latin Translation of the Algebra
of Al-Khowarizmi, with an Introduction, Critical Notes and an English Version,
MacMillan Company, New York, 1915.
[37] Felix Klein, Lectures on the Icosahedron and Solutions of Equations of the
Fifth Degree, translated by George Gavin Morrice, Dover Publications, New
York, 1956.
[38] Steven G. Krantz, Real Analysis and Foundations, 2nd ed., Chapman and Hall,
Boca Raton, Florida, 2005.
[39] J.L. Lagrange, Traite de la resolution des e quations numeriques de tous les
degres, fourth edition, reprinted from the second edition of 1808, in uvres de
Lagrange, Tome Huiti`eme, Gauthier-Villars, Paris, 1879.
[40] Eli Maor, Trigonometric Delights, Princeton University Press, Princeton, New
Jersey, 1998.
[41] Barry Mazur, Imagining Numbers (Particularly the Square Root of Minus Fifteen), Farrar, Straus and Giroux, New York, 2003.
[42] Elliott Mendelson, Introduction to Mathematical Logic, 5th ed., Chapman and
Hall, Boca Raton, Florida, 2010.
p
1, Princeton University
[43] Paul J. Nahin, An Imaginary Tale: The Story of
Press, Princeton, New Jersey, 1998.
[44] Otto Neugebauer, Mathematische Keilschrift-Texte, reprinted by SpringerVerlag, New York, Heidelberg, Berlin, 1973. (Originally printed by Julius
Springer Verlag, Berlin, 1935.)
[45] O. Neugebauer, The Exact Sciences in Antiquity, Princeton University Press,
Princeton, New Jersey, 1952.
[46] O. Neugebauer and A. Sachs, eds, Mathematical Cuneiform Texts, American
Oriental Society and the American Schools of Oriental Research, New Haven,
Connecticut, 1945.
[47] Isaac Newton, Universal Arithmetic, or, a Treatise of Arithmetical Composition and Resolution, Translated from the Latin by Mr. Ralphson, Revised and
Corrected by Mr. Cunn, Illustrated and explained in a series of notes by Rev.
Theaker Wilder, W. Johnston, London, 1769.

i
i

i
IrvingBook 2013/5/22 15:39 page 220 #236

220

References

[48] R.W.D. Nickalls, Vi`ete, Descartes and the cubic equation, Mathematical
Gazette 90 (2006) 203208.
[49] ystein Ore, Cardano: The Gambling Scholar, Princeton University Press,
Princeton, New Jersey, 1953.
[50] Luca Pacioli, Summa de arithmetica, geometria, proportioni e proportionalit`a,
Paganino de Paganini, Venice, 1494.
[51] S.J. Patterson, Eisenstein and the quintic equation, Historia Mathematica 17
(1990) 132140.
[52] Peter Pesic, Abels Proof, The MIT Press, Cambridge, Massachusetts, 2004.
[53] Satya Prakash, A Critical Study of Brahmagupta and His Works, The Indian
Institute of Astronomical & Sanskrit Research, New Delhi, 1968.
[54] Roshdi Rashed, editor, translator, commentator, Al-Khwarizm: The Beginnings of Algebra, SAQI Books, London, 2009.
[55] Reinhold Remmert, The fundamental theorem of algebra, in Numbers,
Springer-Verlag, New York, 1990.
[56] Eleanor Robson, Mathematics in Ancient Iraq, Princeton University Press,
Princeton, New Jersey, 2008.
[57] Paolo Ruffini, Opere Matematiche di Paolo Ruffini, 3 volumes, edited by Ettore
Bortolotti, Volume 1, Palermo, 1915, Volumes 2 and 3, Edizioni Cremonese,
Rome, 1953 and 1954.
[58] J.F. Scott, The Mathematical Work of John Wallis, Chelsea Publishing Company, New York, 1981.
[59] Jacques Sesiano, An Introduction to the History of Algebra: Solving Equations
from Mesopotamian Times to the Renaissance, translated by Anna Pierrehumbert, American Mathematical Society, Providence, Rhode Island, 2009.
[60] David Eugene Smith, A Source Book in Mathematics, Dover Publications, Mineola, New York, 1959.
[61] John Stillwell, Mathematics and Its History, 3rd ed., Springer Science+Business Media, New York, 2010.
[62] John Stillwell, Eisensteins footnote, The Mathematical Intelligencer 17
(1995) 5862.
[63] Niccol`o Tartaglia, Quesiti et Inventioni Diverse, fascimile reproduction from
the 1554 edition, Ateneo di Brescia, Brescia, 1959.
[64] B.L. van der Waerden, Science Awakening, translated by Arnold Dresden, P.
Noordhoff, Groningen, Holland, 1954.
[65] B.L. van der Waerden, Geometry and Algebra in Ancient Civilizations,
Springer-Verlag, Berlin, New York, 1983.
[66] B.L. van der Waerden, A History of Algebra from Al-Khwarizm to Emmy
Noether, Springer-Verlag, Berlin, New York, 1985.
[67] V.S. Varadarajan, Algebra in Ancient and Modern Times, American Mathematical Society, Providence, Rhode Island, 1998.

i
i

i
IrvingBook 2013/5/22 15:39 page 221 #237

References

221

[68] Francois Vi`ete, The Analytic Art, translated by T. Richard Witmer, The Kent
State University Press, Kent, Ohio, 1983.
[69] John Wallis, A Treatise of Algebra, both Historical and Practical, John Playford, London, 1685.
[70] Caspar Wessel, On the analytical representation of direction, Memoirs of the
Royal Danish Academy of Sciences 1799; portions translated in [60, pp. 55
66].

i
i

i
IrvingBook 2013/5/22 15:39 page 222 #238

i
i

i
IrvingBook 2013/5/22 15:39 page 223 #239

Index
Abel, Niels, 182183
Adler, Irving, xiii
al-Khwarizmi, 4144
Annibale della Nave, 67
Argand, Jean-Robert, 106107, 189
Aryabhata, 40
associative law, 89
Bezout, Etienne, 181
Babylonian mathematics, 3339
Bachet, Claude-Gaspard, 40
Bhaskara, 41
Bombelli, Rafael, 135
on the irreducible case, 135
Brahmagupta, 4041
Bring, Erland, 180
Cardano, Girolamo, 47
dispute with Tartaglia, 6770
formula, 5157, 109115, 120123
on the irreducible case, 133135
change of variables
cubic polynomials, 48
general polynomials, 180
quadratic polynomials, 28
quartic polynomials, 145
completing the square, 2428
complex numbers
addition, 74
conjugate, 76
cube roots, 8384, 9698

de Moivres formula, 9395

definition, 73
Eulers formula, 9394
geometry of multiplication, 91
imaginary part, 73
multiplication, 74
real part, 73
square roots, 8183, 96
complex plane, 8588
connected set, 14
converse, 9
cubic polynomials
Cardanos formula, 5157, 109115,
120123
irreducible case, 124130
reduced, 49
resolvent quadratic, 113115
signs of roots, 130133
Vi`etes formula, 125130
dAlembert, Jean-le-Rond, 188
de Moivre, Abraham, 93, 103104, 138
140
formula, 9395
Descartes, Rene
La Geometrie, 170175
cogito, ergo sum, 171
quartic solution, 149154
Dickson, Leonard Eugene, x
Diophantus, 3940
discriminant

223

i
i

i
IrvingBook 2013/5/22 15:39 page 224 #240

224
of a cubic polynomial, 115120
of a generic polynomial, 204
of a quadratic polynomial, 33, 80
of a quartic polynomial, 157162
of a reduced cubic polynomial, 64
Doyle, Peter, 184
Eisenstein, Gotthold, 183184
quintic solution, 184
Euclid, 39
Euler, Leonhard, 104105, 113
formula, 9394
fundamental theorem of algebra,
188
on the cubic, 140142
on the quartic, 175177
quartic solution, 154157
Fefferman, Charles
fundamental theorem of algebra,
187
Fermat, Pierre de, 40
Ferrari, Lodovico, 68, 169170
quartic solution, 146149
Fibonacci, 4445
numbers, 45
field
definition, 189190
examples, 192193
field extension, 191
splitting field, 191, 198200
Fior, Antonio Maria, 67
fundamental theorem of algebra, 185
191
proof, 213215
Galois Theory, ix, 187

Galois, Evariste,
182183
Gauss, Carl Friedrich, 107
fundamental theorem of algebra,
189
on solutions by radicals, 182
Gerard of Cremona, 42
Girard, Albert, 136137
fundamental theorem of algebra,
188
symmetric polynomials, 209211

Index
Hermite, Charles, 184
indeterminate, 3
induction, 212213
integers, 13
intermediate value theorem, 15
existence of roots, 15
polynomials, 15
interval
closed, 14
open, 14
Jerrard, George, 181
Klein, Felix, 184
Lagrange, Joseph, 163, 181, 188
Laplace, Pierre-Simon, 188
proof of fundamental theorem, 211
215
Legendre, Adrien-Marie, 107
Leonardo da Pisa, see Fibonacci
Liouville, Joseph, 187
theorem, 187
logic
converse, 9
or, 10
mathematical induction, principle of,
212213
McMullen, Curt, 184
multiplicity, 10
Neugebauer, Otto, 34
Newton, Isaac, 211
Omar Khayyam, 44
or, 10
Pacioli, Luca, 45
parabola, 30
axis of symmetry, 30
directrix, 30
focus, 30
vertex, 30
polar coordinates, 85
polynomial
coefficients, 2

i
i

i
IrvingBook 2013/5/22 15:39 page 225 #241

225

Index
constant, 2
constant term, 2
cubic, 2
definition, 2
degree, 2
division, 6
factor, 6
irreducible, 192, 194
linear, 2
monic, 3
product, 56
quadratic, 2
quartic, 2
quintic, 2
sum, 5
polynomial equation, 3
solution, 3
solution by radicals, 179
polynomial factorization
existence, 196197
non-trivial, 191
trivial, 191
uniqueness, 198
Ptolemy, 91
theorem, 91
pure imaginary numbers, 73
quadratic formula, 21, 24, 26
complex coefficients, 83
quartic polynomials
Descartes solution, 149154
Eulers solution, 154157
Ferraris solution, 146149
nature of roots, 162169
reduced, 145
resolvent cubic, 152, 173, 175176
quintic polynomials
Bring-Jerrard form, 181
Eisensteins solution, 184
solvability by iteration, 184
unsolvability by radicals, 181183
rational numbers, 13
resolvent polynomial
general notion, 181
of a cubic, 113115
of a quartic, 152, 173, 175176

ring
definition, 189190
polynomial ring, 194
Robert of Chester, 42
root of a polynomial, 4
in a field, 194
multiple, 10
multiplicity, 10
repeated, 10
simple, 10
root of unity, 98
Ruffini, Paolo, 181
on the quintic, 181182
Scipione del Ferro, 47, 66
symmetric polynomials
definition, 205
discriminant, 204
elementary symmetric polynomials, 204
fundamental theorem, 206
Newtons identities, 209
power sum polynomials, 204
Tartaglia, 47, 67
dispute with Cardano, 6770
Thabit ibn Qurra, 44
Tschirnhausen, Ehrenfried Walter von,
180
Tschirnhausen transformation, 180
turning point, 20
global minimum, 20
local maximum, 19
local minimum, 20
Vi`ete, Francois, 127, 135136
formula, 125130
symmetric polynomials, 209
Wallis, John, 105, 137138
on the cubic, 138
Waring, Edward, 181
Wessel, Caspar, 105

i
i

i
IrvingBook 2013/5/22 15:39 page 226 #242

i
i

i
IrvingBook 2013/5/22 15:39 page 227 #243

About the Author

Ron Irving is a mathematics professor at the University of Washington in
Seattle. He was born in suburban New York City, studied mathematics and
philosophy at Harvard, and received his Ph.D. in mathematics at MIT. Following a postdoctoral position at Brandeis and a National Science Foundation postdoctoral fellowship year at the University of Chicago and UC San
Diego, Irving came to Seattle. He has been a visiting faculty member at
UCSD and Aarhus and a member of the Institute for Advanced Study in
Princeton. His research interests have ranged over several areas of algebra,
including ring theory and the representation theory of Lie groups and Lie
algebras.
When Irving began teaching the departments senior algebra course for
majors planning on secondary teaching careers, he developed an interest in
the preparation of pre-service and in-service teachers. His work with this
audience led to receipt of the universitys Distinguished Teaching Award
and to his book Integers, Polynomials, and Rings.
Irving spent seven years in academic administration, serving as department chair for a year, divisional dean of natural sciences for four, and interim dean of the College of Arts and Sciences for two. During this time,
he established the Summer Institute for Mathematics at UW, a six-week
residential program that brings talented high school students in the Pacific
Northwest to the university to share in the excitement of doing mathematics.
He continues to serve as the programs executive director. He also began the
planning for a new undergraduate degree in integrated sciences designed to
meet the needs of future secondary science teachers.
Since returning to the department, Irving has taught a variety of algebra
courses and written this book. He has been the director of the Integrated Sci227

i
i

i
IrvingBook 2013/5/22 15:39 page 228 #244

228

About the author

ences program, completing its planning, approval, and implementation. He

serves as the secretary-treasurer of the Astrophysical Research Consortium,
which oversees Apache Point Observatory in the Sacramento Mountains of
New Mexico, and is president of the board of the Burke Museum of Natural
History and Culture.
Irving is a member of the Mathematical Association of America and the
American Mathematical Society. As this book goes to press, he is beginning
a second stint as department chair.

i
i