r.

o
512.81 N7313 73-50666 Niven An introduction to the theory of numbers

kansas city

public library
kansas
city,

missouri

Books

will

be issued only

on presentation of library card.

Please report lost cards and

change of residence promptly.

Card holders are responsible for
all

books, records, films, pictures

or other library materials

checked out on their cards.

DATE DUE

:i6
DEC 06

IL

2 o 1994

t993

MAI SEP

16

An

Introduction to

the Theory of

Numbers

Books by

Ivan

Niven

Calculus:

An

Introductory Approach (Van Nostrand)

Diophantine Approximations (Wiley)

Irrational

Numbers (Wiley)

Mathematics of Choice

(Random House)
Irrational

Numbers Rational and

:

(Random House)

An

Introduction to

the Theory of
Third Edition

Numbers

IVAN N.IVEN
University of

Oregon
S.

HERBERT

ZUCKERMAN

University of Washington

JOHN WILEY & SONS INC

New York
London
Sydney

Toronto

Copyright

I960, 1966, 1972

by John Wiley

& Sons, Inc.

All rights reserved. Published

simultaneously in Canada.

No part of this book

without the

nor transmitted, nor translated into a

written

may be reproduced
permission

by any means, machine language

of the publisher.

Library of Congress Catalog Card

Number: 73-178149

ISBN 0-471-64154-5
Printed in the United States of America.

10

987654321

Preface

In this third edition Sections 4.4, 6.3, 6.4 and 9.10 have been added, and
Sections 5.7 and 10.3 have been modified considerably. Following the last chapter a set of miscellaneous problems has been added, and also a collection

of eleven special topics in outline form but with sufficient explanation for easy filling in of details. Sincere thanks are expressed to Paul T. Bateman, Bruce C. Berndt,

W. Curtis, Edwin Hewitt, Ralph D. James, Emma Lehmer, Roy W. Ryden, Sigmund Selberg and John Steinig for helpful suggestions since the
Charles
first

version. Herbert S.

Zuckerman shared

following his untimely death in 1970 I have laboration in seeing it through the press.

in the planning of this edition; missed very much his able col

February 1972

IVAN NIVEN

350666

Preface to the First Edition

complete introduction to the theory of numbers within the compass of a single volume. The basic concepts are more specialized material presented in the first part of the book, followed by

Our purpose

is

to present a reasonably

in the final three chapters. Paralleling this progress

more

particular discussions,

we have attempted

leisurely

set forth in

pace than we have followed later. Thus a more compact and sophisticated presentation than are the

from general topics to to begin the book at a more the later parts of the book are

earlier parts.

The book is intended for seniors and beginning graduate students in American and Canadian universities. It contains at least enough material
for a full year course; a short course can be built by the use of Sections 1.1 to 1.3, 2.1 to 2.4, 3.1, 3.2, 4.1, 5.1 to 5.3, 5.5, 6.1, and 6.2. Various other

arrangements are possible because the chapters beyond the fourth are, apart from a very few exceptions, independent of one another. The final
three chapters are entirely independent of each other. To enable the student to deepen his understanding of the subject, we have provided a considerable number of problems. The variety of these exercises
extensive, ranging from simple numerical problems to additional develop ments of the theory. The beginner at number theory should take warning that the subject is noted for the difficulty of its problems. Many an innocentlooking problem gives, by the very simplicity of its statement, very little
is

tion.

notion of the considerable ingenuity or depth of insight required for its solu As might be expected, the more difficult problems are placed toward the
sets.

ends of the
stitute

In

many

a related series in which the last ones can be solved

first

instances three or four consecutive problems con more readily by use ones.

of information from the

text itself entirely

As a matter of principle we have made

the

independent of the problems. In no place does the proof of a theorem depend on the results of any problem.
In choosing methods of proof,

we have

tried to include as

many methods

as possible. have tried to state the proofs accurately, avoiding statements that could be misleading and also avoiding unduly long discussions of un important details. As the reader progresses he will become familiar with

We

more and more methods, and he should be able by patterning them after our proofs.
vi

to construct accurate proofs

Preface to the First Edition

vii

The reader
one

bibliography at the

interested in further exploration of the subject will find the end of the book of considerable use. In particular, any interested in the history of the subject should consult O. Ore, Number
Its

more specific information, L. E. Dickson, Our approach is analytical, not historical, and we make no attempt to attribute various theorems and proofs to their original discoverers. However, we do wish to point out that we followed the suggestion of Peter Scherk that we use F. J. Dyson's formulation of the proof of Mann's a/? Theorem. Our proof is based on notes graciously placed at our disposal by Peter Scherk. For permission to use several problems from the American Mathematical Monthly, we are indebted to the editors. We also
Theory and
History, and, for

History of the Theory of Numbers.

appreciate the careful reading of the manuscript by Margaret Maxfield, efforts resulted in numerous improvements. Finally we would like to record our deep appreciation of and our great debt to the mathematicians

whose
whose

lectures

were

vital to

our introduction to the theory of numbers: L. E.

Dickson, R. D. James, D. N. Lehmer, and Hans Rademacher.

June 1960

IVAN NIVEN HERBERT S. ZUCKERMAN

Contents

1.

Divisibility
1.1

Introduction
Divisibility

1.2
1.3

Primes

2 10

2.

Congruences
2.1

20
20 27 29 34 37 39
43

Congruences
Solutions of Congruences Congruences of Degree
I

2.2
2.3

2.4
2.5

The Function

</>(n)

2.6

2.7
2.8 2.9

Congruences of Higher Degree Prime Power Moduli Prime Modulus Congruences of Degree Two, Prime Modulus

Power Residues

46 46
51

2.10
2.11

Number Theory from

an Algebraic Viewpoint

Multiplicative Groups, Rings, and Fields

56

3.

Quadratic Reciprocity
3.1

63
63

Quadratic Residues
Quadratic Reciprocity

3.2
3.3

67
71

The

Jacobi

Symbol

4.

Some
4.1

Functions of

Number Theory

78
78

Greatest Integer Function

4.2
4.3

Arithmetic Functions

84
88
IX

The Moebius

Inversion Formula

Contents
4.4
4.5

The

Multiplication of Arithmetic Functions

91

Recurrence Functions

96

5.

Some Diophantine Equations

5.1

100
100

Diophantine Equations

5.2
5.3

The Equation ax

by

101

Positive Solutions

5.4
5.5

5.6

5.7
5.8

TheEquationx + y = z The Equation x4 + y4 = * 2 Sums of Four and Five Squares

2

Other Linear Equations

2

102 104
2

105 107

109

5.9

Waring's Problem Sum of Fourth Powers

112 112
113

5.10
5.11

Sum

of

Two

Squares

5.12

The Equation 4x 2 The Equation ax 2

+f=n + bf + cz =
2

118

120
123

5.13 Binary Quadratic Forms 5.14 Equivalence of Quadratic

Forms

127

6.

Farey Fractions and Irrational

6.1

Numbers

134
134 137 142 147

Farey Sequences
Rational Approximations
Irrational

6.2
6.3

Numbers

6.4

Coverings of the Real Line

7.

Simple Continued Fractions

7.1

150
150

The Euclidean Algorithm

Infinite

7.2
7.3

Uniqueness Continued Fractions

Irrational

152
153

7.4
7.5

Numbers

157
159
163

7.6

Approximations to Irrational Numbers Best Possible Approximations

Periodic Continued Fractions
Pell's

7.7
7.8 7.9

166

Equation Numerical Computation

172
175

Contents
8.

Elementary Remarks on the Distribution of Primes

8.1

178
178
181

8.2
8.3

The Function TT(X) The Sequence of Primes

Bertrand's Postulate

185

9.

Algebraic
9.1

Numbers

188
188

Polynomials
Algebraic
Algebraic

9.2
9.3

Numbers

192
Fields

Number

195

9.4
9.5

Algebraic Integers

199
201

Quadratic Fields
Units
in

9.6

Quadratic Fields
Quadratic Fields

203

9.7
9.8
9.9

Primes

in

204
206

Unique Factorization
Primes
in

Quadratic Fields Having the Unique

Factorization Property

208
3

9.10

The Equation x 3

+ y3 = z

213

10.

The
10.1

Partition Function

219
219

Partitions

10.2
10.3

Graphs
Formal Power Series and Euler's Identity
Euler's
Jacobi's

220
223

10.4
10.5

Formula Formula
Property

227
233

10.6

Divisibility

236

11.

The Density of Sequences of Integers

11.1

240
241

Asymptotic Density
Square-Free Integers
Sets of Density

11.2
11.3

243

Zero
a/?

246

11.4

Schnirelmann Density and the

Theorem

250

xii

Contents

Miscellaneous Problems

255

Special Topics
Periodic
decimals,
n

259
Unit
fractions,

The equation

x~ n

4.

y~

= z~ n

Gauss's generalization of Fermat's

theorem, Primitive root

mod

group of
of the

rational points

p by group theory, The on the unit circle, The day

week from the

date,

Some number

theoretic

determinants, Gaussian integers as sums of squares, Unique factorization in Gaussian integers, The
Eisenstein irreducibility criterion.

General References

269

Answers to Problems

271

Index

285

An

Introduction to

the Theory of

Numbers

1
Divisibility

1.1

Introduction

The theory of numbers is primarily concerned with the properties of the natural numbers, 1, 2, 3, 4, also called the positive integers. However, the theory is not strictly confined to just the natural numbers or even to the
,

In fact, some theorems of number theory are most easily proved by making use of the properties of real or complex numbers even though the statement of the theorems may involve only natural numbers. Also, there are theorems concerning real numbers that depend so heavily on the properties of integers that they are properly included in the theory of numbers. An integer n greater than 1 is called a prime if it has no divisor d such that
set
all

of

integers: 0,

1,

2,

3,

<

<

n.

The

greater than

is

fact that for there is a prime every given positive integer stated in terms of integers, and it can be from the

proved

properties of the natural numbers alone. The fact that every natural number can be expressed as a sum of, at most, fifty-four fifth powers of integers is also stated in terms of natural numbers, but known any proof depends on
properties of complex numbers. Finally, the question as to how many primes there are that do not exceed x clearly belongs to the theory of numbers but its

natural numbers.

answer involves the function log x and is well outside of the realm of the The last two examples are beyond the scope of this book. we do not restrict ourselves to the integers but will use real and However,

complex numbers when it is convenient. The questions discussed in this book are not numerical computations or numerical curiosities, except insofar as these are relevant to general propositions. Nor do we discuss the foundations of the number system; it is assumed that the reader is familiar not only with

Divisibility

the integers, but also with the rational and real numbers. However, a rigorous logical analysis of the real-number system is not prerequisite to the study of

number theory. The theory of numbers relies for proofs on a great many ideas and methods. Of these, there are two basic principles to which we draw especial attention. The first is that any set of positive integers has a smallest element if it contains any members at all. In other words, if a set S of positive integers is not empty,
then
s
it

contains an integer s such that for any

holds.

member a of

S, the relation

principle, mathematical induction, is a logical consequence of the first.* It can be stated as follows: if a set S of positive

^a

The second

integers contains the integer 1, and contains n then 5 consists of all- the positive integers.
It

whenever

it

contains

be well to point out that a negative assertion such as, for example, positive integer can be expressed as a sum of the squares of three the number 7 integers," requires only that we produce a single example

may

"Not every

cannot be so expressed. On the other hand, a positive assertion such as "Every positive integer can be expressed as a sum of the squares of four integers," cannot he proved by producing examples, however numerous. This result is Theorem 5.6 in Chapter 5, where a proof is supplied.
Finally, it is presumed that the reader is familiar with the usual formulation of mathematical propositions. In particular, if A denotes some assertion or collection of assertions, and B likewise, the following statements are logically

equivalent

they are just different ways of saying the same thing. implies B. If A is true, then B is true.
is is

In order that

B A
If

A be true it is necessary that a necessary condition for A. a sufficient condition for B. B

and

be

true.

implies

sufficient condition for

implies A then one can say that A to hold.

9

is

a necessary and

In general,

we
9

m,

/2,

as,

z,

shall use letters of the roman alphabet, a, b, c, to designate integers unless otherwise specified.

1.2

Divisibility

Definition 1.1

An integer b is divisible by an integer a, not zero, if there integer x such that b ax, and we write a\b. In case b is not divisible we write a b.

is

an

by a

Jf

Compare G. Birkhoff and

S.

MacLane,

Survey of Modern Algebra, Macmillan,

third edition, 1965, pp. 10-13.

1.2

Divisibility

a
is

is

Other language for the divisibility property a b is that a divides b, that a divisor of b, and that b is a b and < a < b then a multiple of a. If called a proper divisor of*. It is understood that we never use as the left
\

pair of integers in a b. On the other hand, not only may occur as the right member of the pair, but also in such instances we always have divisibility. Thus a for every integer a not zero. The notation a K b is sometimes used to indicate that b but a K+l b.
\

member of the

||

Theorem
(1)

1.1

a
a
| |

b implies a

be for any integer c;

\

(2)

b and b
\

c imply a

c;

(3) a
(4) (5) (6)

b and a
\

c imply a

(bx

+
b;

cy)for any integers x and y;

a a
|

b and b
\

a imply a
b

b, a

>

0,

>

0,

imply a <i
is

b.

ifm

5* 0,

a
\

b implies and

implied by

ma
\

mb.

Proof.

The proofs of

these results follow at once

divisibility.

Property 3 admits
a
|

from the definition of an obvious extension to any finite set, thus:

bjXj for any integers
:?=!

b n imply a

Property 2 can be extended similarly.

Theorem

1.2 The division algorithm. Given any integers a there exist unique integers q and r such that b r, qa then r satisfies the r a. stronger inequalities

and

b,

with a
#.

>

0,

^r<

< <
+

If a\b,

Proof.

Consider the arithmetic progression

,

30, b

20, b

a, b, b

a, b

2a, b

3a,

extending indefinitely in both directions. In this sequence, select the smallest non-negative member and denote it by r. Thus by definition r satisfies the inequalities of the theorem. But also r, being in the sequence, is of the form b qa, and thus q is defined in terms of r.

To prove

satisfying the

the uniqueness of q and same conditions. First

r,

suppose there
that r x

is

we prove

another pair q 1 and

r.

rx

For

if not,

a (<l

r < a, and then we see that presume that r < r so that < r x ^ ?i) an d so a fa r), a contradiction to Theorem 1.1, part Hence r = r 1? and also q = q lm
|

we may
r

(5).

We

have stated the theorem with the assumption a

is

>

0.

However,

this

not necessary, and we may formulate the theorem without it: given any integers a and Z>, with a 0, there exist integers q and r such that a.
hypothesis

Divisibility

Theorem

1.2 is called the division algorithm.

An

algorithm

is

a mathe

1.2 matical procedure or method to obtain a result. that we this in the form "there exist integers q and r," and wording suggests an than have a so-called existence theorem rather algorithm. However, it

We have stated Theorem

may be observed

+ q and r, need be examined only in part to yield the smallest positive member r. In actual practice the quotient q and the remainder r are obtained by the arithmetic division of a into b.
-

that the proof does give a method for obtaining the integers b because the infinite arithmetic progression a, b, b a,
,

Definition 1.2

The integer a
is

is

a common divisor ofb and

c in case a
\

b and

a
|

c.

Since there
is

only a finite number of divisors of any non-zero integer,

only a finite number of common divisors ofb and c 9 except in the case 0. If at least one ofb and c is not 0, the greatest among their common divisors is called the greatest common divisor of b and c and is denoted by
there

=c=

(b, c).
15

62

'

Similarly we denote the greatest b n not all zero, by (b l9 4 a ,

'

common
,

divisor

g of

the integers

'

*).

Thus the
b, c except

greatest common divisor (b, c) is defined for every pair of integers 0, and we note that (b, c)ll. 0, c

Theorem

1.3 Ifg is the greatest common divisor ofb and % andy Q such that g = (b, c) = to + cy Q integers
.

c,

then there exist

Proof.

This set of integers {bx values, and also by the choice x

all integers.

where x and y range over and negative y = 0. Choose X Q and y Q so that bx + cy Q is the least positive integer / in the set; thus / = bx + cy$. Next we prove that / b and / c. We establish the first of these, and the second follows by analogy. We give an indirect proof that / b, that is, we assume lj(b and obtain a contradiction. From / Jfb it follows that there exist integers q and r, by Theorem 1.2, such that b = < r < /. Iq + r with Hence we have r = b - Iq = b - q(bx + cy ) = 6(1 - qx + c(-qy Q ), and thus r is in the set {bx + cy}. This contradicts the fact that / is the least
Consider the linear combinations bx
cy,

cy} includes positive

~)

= g(Bx + Q/ Thus g and so by 5 of Theorem 1.1 we conclude part thatg- ^ Nowg < /is impossible, since g is the greatest common divisor, and so g = = bx + cyQ
b
c
9

positive integer in the set {bx since g is the greatest

+
+

cy}.

Now

common
cy Q

divisor of b
).

and

c,

= gB,

= gC

and

we may
/,
\

write

bx

/.

Theorem
in the

1.4

The greatest common

it is

divisor

gofb and

c can be characterized

following two ways:

is divisible

andy
which

the least positive value ofbx (1) cy where x over all it the is common divisor range integers; (2) positive ofb and c

by every common

divisor.

1.2

Divisibility

Proof.

Part

observe that

if

follows from the proof of Theorem 1.3. To prove part 2, we d is any common divisor of b and c, then d g by part 3 of
\

Theorem
2,

1.1.

Moreover, there cannot be two

1.1,

distinct integers with property

because of Theorem

part 4.

Theorem

common

1.5 Given any integers b ly divisor g, there exist integers

b2

b n not
,

all zero, with greatest

x l9 # 2

x n such that

Furthermore, g is the least positive value of the linear form 25U b$j where the bn yj range over all integers; also g is the positive common divisor ofb ly b 2 , which is divisible by every common divisor.
-

Proof.

is a straightforward generalization of the preceding two the and theorems, proof is analogous without any complications arising in the passage from two integers to n integers.

This result

Theorem

1.6

For any positive integer m,

(ma, mb)

m(a,

b).

Proof.

By Theorem

1.4

we have

(ma, mb)

= least positive value of max + mby = m {least positive value of ax + by} = m(a, b).
\

Theorem

1.7

If d

a and d b and d
\

>
d

then

d d
If(a,b)

=g,

then

The second assertion is the special case of the first obtained by using Proof. the greatest common divisor g of a and b in the role of d. The first assertion
in turn
in that
is a direct consequence of Theorem 1.6 obtained by replacing m, theorem by d, (aid), (bid) respectively.

a,

Theorem
Proof.

1.8

If (a, m)

(b,

m)

1,

then (ab,

m)
,

1.

ax

+ my = bx + rny-^. = ^2/2 where y

l
1

By Theorem

1.3 there exist integers

X Q y 0? x l9 y such that 1 Thus we may write (ax^(bx-^ = (1 myQ )(l = 3/0 + 2/1 2 is defined by the equation 2/2

6
Divisibility

From the equation abx^ + my^ = 1 we note, by part 3 of Theorem 1.1, that any common divisor of ab and m is a divisor of 1, and hence (ab, m) = 1.
Definition 1.3

We
-

a l9 a^ that a say l9 a 2
that
,

and
i

9 -

say that a and b are relatively prime a n are relatively prime in case (a l9 a 2

in
-

case
,

(a, b)

1,

a n)

=
1

1.

We
all

a n are relatively prime in pairs

2,

n and]

The

fact that (a, b)

= =

in case (a iy

a^

for

2,

n with

^ j.

1 is

coprime, or by saying that a

sometimes expressed by saying that a and b are is prime to b.

Theorem
Proof.
(a,

1.9

For any x

(a, b)

(b, a)

(a,

-b)

(a,b

ax).

hence

Denote (a, b) by d and (a, b + ax) by g. It is clear that (b, a) = -b) = d. By application of Theorem 1.1, parts 3 and 4, we obtain d\g,g\d, and

d=

g.

Theorem 1.10

Ifc\ab and (b,

c)

1,

then c
\

a.

Proof. By Theorem 1.6, (ab, ac) so c a by Theorem 1.4.

|

= a(b, c) =

a.

But c ab and
\

c
\

ac,

and

example, that 6, g = 2, and one pair of values for x y is 2, -3. But if b and c are large, inspection is not adequate except in rather obvious cases like (963 963) 963
,

Given two integers b and c, how can the greatest common divisor g be found? Definition 1.2 gives no answer to this question, and neither does Theorem 1.3 which merely asserts the existence of a pair of integers x and y Q such that g = bx Q + CyQ Ifb and c are small, values of and y can be g, z found by inspection. For if b = 10 and c = it is obvious
.

and (1000, 600)

'

200.

Consider the case b

q

=
r
1

=
c

963, c

657. If we divide c into

1,

and remainder

306.

ticular 306

963

Thus b
(b, c)

657.

Now

by

and

# in

Theorem

= c# + r, or r = b - cy, in par = (b - cq, c) by replacing a and *

b we get a quotient

1.9, so

we

see that
-

(963, 657)

(963

657, 657)

(306, 657).

The integer 963 has been replaced by the smaller integer 306, and this suggests that the procedure be repeated. So we divide 306 into 657 to get a quotient 2
and a remainder 45, and
(306, 657)

(306, 657

306)

(306, 45).
is

Next 45

divided into 306 with quotient 6 and remainder 36, then 36 divided into 45 with quotient 1 and remainder 9. conclude that
is

We

(963, 657)

(306, 657)

(306, 45)

(36, 45)

(36, 9).

1.2

Divisibility

Thus

(963, 657) = 9, and we can express 9 as a linear combination of 963 and 657 by eliminating the remainders 36, 45, and 306 as follows:
9

= 45 - 36 = 45 - (306 -45-6) = -306 + 7-45 = -306 + 7(657 - 306 2) = 7 657 - 15 306 = 7 657 - 15(963 - 657) = 22-657- 15-963.
,

1.3 where g = (b, c) = bx + cy beginning with 657 we have used a procedure, called the "Euclidean algorithm," to find g = 9, x = -15, y Q = 22. Of course these values for x and 2/ are not unique: - 15 + 657A: and 22 - 963k will do where k is any

In terms of

Theorem

963 and c

integer.

In general, to find the greatest

integers x and y satisfying g done in the special case above.

common
(6, c)

divisor (b, c) of b

bx

o/

we

By Theorem 1.9, we may presume c positive, because the case c =

Theorem
1.11

(b, c}
is

and also generalize what is (b, c), and hence

c,
:

and

very special

(4, 0)
0,

\b\.

The Euclidean algorithm. Given integers b and c > a repeated application of the division algorithm, Theorem 1.2, series of equations

we make
a

to obtain

= cq + r = c /i? + r = ?i Ms +
b
l 2

l9 a,

>*3>

<r < < < <r <

><

c,

r l9
ra ,

/V_a

0-i-

<

r.

<

_,

rAe greatest common

in the division pro cess.

divisor (b, c}

Values ofx
-

ofb and c is rj9 the last non-zero remainder andy in (b, c) = bx Q + cy^ can be obtained
set

by eliminating

rf __ l9

r2 ,

r^from the

of equations.

into c, Proof. The chain of equations is obtained by dividing c into 6, r 2 into r l9 The process stops when the division is exact, that , r^ into ^_ x is when the remainder is zero. Thus in our application of Theorem 1.2 we have written the inequalities for the remainder without an Thus, for equality

sign.

r x < c, because if r x were < r < c in place of example, equal to zero, the chain would stop at the first = b in which the greatest case equation cq l9 common divisor of b and c would be c.

Divisibility

We now
Theorem
(4, c)

prove that

1.9

we

the greatest observe that

r,
is

common

divisor

g of

b and

c.

By

(b

cq l9

c)

(r 1? c)

(r x , c

rtfj
Ol> r 2)

Ol

r 2?3,

>*

2)

(>*3> 'a).

(r^, r,-) induction we get (b,c) Continuing by mathematical because r is a divisor of r,_ l9 and r is positive. and c, we need To see that r, is expressible as a linear combination of b the chain, last from the second and third equations of merely eliminate r all of until on so and then eliminate r,_ a from the third and fourth last, te form the + in <% are eliminated. Thus we r,
/-,-

'

'

>Vi, 0-2,

'2, 'i

get

Definition 1.4

3Tte

w/^w
\

^, a 2

an

all different
9

from

zero, have a

common multiple b if a, b for i do exist; for example the product

common
[>!,
2,

2,

n.
is

a&

an

(Note that common multiples the positive one.) The least of

multiples
,

is

called the least

common

multiple,

and

it

is

denoted by

a n ].

Theorem 1.12
a n]
0,
6.
|

Ifb
the

is

any common multiple ofa^

that if

a,

7%w w
2/7,

same as saying
comprise

h denotes

a n then [a l9 a& a n ], then [a l9 a 2

,
,

/7,

3A,

all the

common

multiples of

a^

<z a

an

by h. By Theorem 1.2 be any common multiple and divide Let Proof. h. r that r, r such flA there is a quotient q and a remainder / each For as follows. we argue 0. If r must prove that r we know that a t h and a t m, so that a, r. Thus r is a positive 1

m m =
|

^ <

We
,

2,

common
Theorem

h , contrary to the fact that multiple of a l9 j a all the common multiples. positive of
,

is

the least

1.13

Ifm

>

0,

[ma, mb]

m[a,

b].

Also

[a, b]

(a, b)

\ab\.

of m, Since [ma, mb} is a multiple of ma, it is & fortiori a multiple Proof. that note A we , and so can be written in the form mh^ Denoting [a, b] by 2 Thus 1.12. Theorem m/? so 2 by a /z 25 ft /* 2 am m/7 2 im w/z 2 and bm mh^ a h l9 b h and so A 2 h v 0ra m// the other
, ,

m^
\

A!

We conclude that A x =
|

A2

On

hand,

x,

of a, say ma. Then b ma and (a, b) = 1 so by Theorem [a] b] is a multiple a 1.10 we conclude that fe m. Hence b^m,ba^ma. But Z?a, being positive common least the than less be multiple, common multiple of b and a, cannot
,
\
|

[a,

theorem is established. part of the a and b, since for It will suffice to prove the second part positive integers 1. Now where case the with (a, b) special b].
A 8 and thus the
first

-b]

[a,

We

begin

and so ba

= ma =

[a, b].

1.2

Divisibility

Turning to the general case where (a, Z>) = g > 1, we have ((a/g), (%)) = by Theorem 1.7. Applying the result of the preceding paragraph, we obtain

g
2

g\g

g/

gg
first

Multiplying by g and using Theorem 1.6 as well as the ab. present theorem, we get [a, b\(a, b)

part of the

PROBLEMS
1.

By using

the Euclidean algorithm find the greatest

(b)

common

of
(a)
(c)

divisor (g.c.d.)

2.

7469 and 2464; 2947 and 3997; Find the greatest

2689 and 4001

divisor

(d) 1109

and 4999.
of the numbers 1819 and 3587, and

common

then find integers x and y to satisfy

1819a5
3.

3587s/

= g.

Find values of x and y to (a) 243x + l9Sy = 9;

43x

satisfy

llx-50y = 1; + 64y = 1 (</)93s-81y =3;

(b)
(c)
;

0) 6z + 10y + I5z = 1. Find the least common multiple (l.c.m.) of (a) 482 and 1687, (b) 60 and 61. 5. How many integers between 100 and 1000 are divisible by 7? 6. Prove that the product of three consecutive integers is divisible by 6; of four consecutive integers by 24.
4. 7.

Exhibit three integers that are relatively prime but not relatively prime in

pairs.

same parity if they are both even or both even and the other odd, they are said to be of opposite parity, or of different parity. Given any two integers, prove that their sum and their difference are of the same parity.
8.

Two
if

integers are said to be of the

is

odd:

one

9.

Show

10.

11. 12.

13.

that if ac be then a b. Given a b and c d, prove that ac bd. Prove that 4 -f (rc2 + 2) for any integer n. Given that (a, 4) = 2 and (b, 4) = 2 prove that (a Prove that n 2 - n is divisible by 2 for
\
\

b, 4)
;

4.
3

every integer

that

is

divisible

14.
15.

16.

by Prove that Prove that Prove that

b.

5 6; that n

is divisible
2

by

30.

if if

is odd, /z 1 is divisible by 8. x and y are odd, then x 2 + y* is even but not divisible by 4. if a and b are (a, b) = [a, b] then positive integers

satisfying

10

Divisibility

17. Evaluate (n, n 1) and [, n + I] where n is a positive integer. 18. Find the values of (a, b) and [a, b] if a and 6 are positive integers such that a b.
|

19.

Prove that any

set

of integers that are relatively prime in pairs are relatively

is

prime.
20. Given integers a and /3, a number n there is an integer k such that ak b
said to

n.
.

be of the form ak + b if Thus the numbers of the form

3k + 1 are Prove that every integer is of the -8, -5, -2, 1, 4, 7, 10, form 3k or of the form 3k + I or of the form 3k + 2. 21. Prove that if an integer is of the form 6k + 5 then it is necessarily of the form 3k 1 but not conversely. 22. Prove that the square of any integer of the form 5k + 1 is of the same form. 23. Prove that the square of any integer is of the form 3k or 3k + 1 but not of the form 3k + 2. 24. Prove that no integers x, y exist x + y - 100 and (x, y) = 3. satisfying 25. Prove that there are infinitely many pairs of integers x, y satisfying x + y = 100 and (x, y) = 5.
-

be given integers. Prove that integers x and y exist satisfying = g if and only ifg\s. 27. Find positive integers a and b satisfying the equations (a, b) = 10 and [a, b] = 100 simultaneously. Find all solutions. 28. Find all triples of positive integers a, b, c (a b,c) = 10 and

26.

+y = s and

Letsand^-

>

(x 9 y)

[a, b, c]

satisfying

100 simultaneously.
/

29. Let

g and

satisfying (x, y) 30. Let b and^-

b can be solved simultaneously if and only ifg*\b. 31. Let n > 2 and k be any positive integers. Prove that (n 32. Let 72 > 2 and k be Prove that any positive

xy

= g and [x, y] = / if and only if g L > be given integers. Prove that the equations
\

be given positive

integers.

Prove that integers x and y

(x 9 y)
k

exist

= g and
1).

1)

and only

if

(/

integers.

1)

k.
|

k Suggestion: n

{(n

(n

(n
|

I)

(72*

1) if

1)

33. Prove that (a, b, c) ((a, b), c). 34. Extend Theorems 1.6, 1.7, and 1.8 to sets of 35. Prove that if (b, c) 1 and r then

1}*.

more than two

b,

(r, c)

integers.

1.

36. Prove that if

a,
777,

> n then a* n are positive integers with

,

is

a divisor of a*

1.

Show

that

if

m ^ n,
\

then
if

2m

-,2"

Jl 12 if

#
a

is is

even,

odd.

1.3

Primes

An integer p>\is called a prime number, or a prime, in case no divisor d of p satisfying \<d<p.Ifan integer a > I is not a prime, it is called a composite number.
Definition 1.5
there
is

1.3

Primes
3, 5,

11

Thus, for example, 2,

composite.

and 7 are primes, whereas

4, 6, 8,

and

9 are

Theorem 1.14

Every integer n greater than ofprimes (with perhaps only one factor).

can be expressed as a product

Proof.

If the integer n is a prime, then the integer itself stands as a "product*' a with single factor. Otherwise n can be factored into, say, n^n^ where n and 1 < 2 < n. If n is a prime, let it stand; otherwise it will 72 1 < X <

factor into, say, n z n

where

<

<^

and

<

/7

<

n^\ similarly for

/7 2 .

This process of writing each composite number that arises as a product of factors must terminate because the factors are smaller than the composite

number

itself,

and yet each factor

is

an integer greater than

1.

Thus we can

write n as a product of primes, and since the prime factors are not necessarily distinct, the result can be written in the form

where /? 1? /> 2

oc r are positive. ,?r are distinct primes and oc l5 a 2 This expression is called a representation of n as a product of primes, and it turns out that the representation is unique in the sense that, for fixed
'
'
'

any other representation is merely a permutation of the factors. It may appear obvious to the reader that the representation of an integer as a product of primes is unique, but it is a fact requiring proof. Indeed there are mathematical situations where it might appear equally "obvious" that factorization will be unique, but where in fact it is not. We digress from our main theme to discuss two of these situations where factorization is not
n,

unique.
First consider the class

E are 2, 4,
to
8

6, 8, 10,

of any two elements in

E of positive even integers, so that the elements of that E is a multiplicative system, the product E being again in E. Now let us confine our attention
.

Note

E in the sense that the only "numbers" we know are members of E. Then = 2 4 is "composite," whereas 10 is a "prime" since 10 is not the product
or
-

the more "numbers." The "primes" are 2, 6, 10, 14, has two 60 the "number" Now are numbers" 4, 8, 12, "composite = 2 30 = 6 10, and so factorization factorings into "primes," namely 60

of two

is

not unique.

but also rather more complicated, example 6 where a and b is obtained by considering the class C of numbers a + b-J C is closed under addition this We that all over system say integers. range and multiplication, meaning that the sum and product of two elements in C

somewhat

less artificial,

are elements of C.

By taking b

we note

that the integers

form a subset

of the class C.

12
Divisibility

First

we

establish that there are primes in C,

and that every number

6 in

in

C
be

can be factored into primes. For any number a + b\] convenient to have a norm, N(a + b\f^6) defined as
9

it

will

N(a
Thus
a

Z>V^6)

(a

bj^6)(a

- &V^6) =

a2

6b*.

thejiorm b^j6 and

of a number in Cis the product of the complex

its

number
this,

conjugate a

b^J6.

Another way of saying

perhaps in more familiar language, is that the norm is the square of the abso lute value. Now the norm of every number in C is a positive integer greater than 1, except for the numbers 0, 1, 1 for which we have N(G) = 0, N(l) = 1, N( 1) = 1. We say that we have a factoring of a + b 6 if we can >/
write
(1.1)

b^/^6
1

(a?!

W^fe + W^),
+ yj~^6) >
1.

where tffo
the

+ y 1% /^6) >
the factors

and N(x 2

This restriction on
trivial

(l)(a The norm of a product can be readily calculated to be the product of the norms of the so that in = N(x l + we have N(a + factory the^ factoring (1.1)

+ iV-6 =

norms^of

is

needed to rule out such

+ 6V :i Q = (-l)(-fl-W :: 6).

factorings as

W^6)

yd-6)N(x 2 +

2/ 2

V~6).
1

It

follows that

< N(*i + yJ^S) < N(a + K//Z 6), < N(x + y v^6J < N(a + bj^6),
2 2

and so any number a

factors since the

W-6 will break up into

factor
is

only a

finite

number of

norm of each

an

We remarked
1, is

integer.

a2

6b 2

above that the norm of any number in C, apart from and greater than 1. More can be said. Since N(a + W^6) has the value
,

we

observe that

0-2)
that
is,

N(a
the

+ bJ^6)^6

if

0,

in the sense of (1.1), is called a prime in C. For example 5 is a prime in C. For in the first place 5 cannot be factored into real numbers in C. In the

norm of any complex number in C is not less than 6. number of C having norm > 1, but which cannot be factored

second

place,

we had a factoring 5 = (^ + y^^6)(x 2 numbers, we could take norms to get

if

+ y*J~^6)
y 2 J^6)
9

into complex

25

= Nfa + y^^6)N(x, +
Thus
5
is

which contradicts
establishes that 2

(1.2).
is

a prime in C, and a similar argument

a prime.

1.3

Primes

13

We are now in a position to show that not all numbers of C factor uniquely
into primes. Consider the

number
2
5

10
(2

and

its

two

factorings:

10

+ J^6)@ -

V-6)-

The first product 2 5 has factors that are prime in C, as we have seen above. Thus we can conclude that there is not unique factorization of the number 10 in C. Note that this conclusion does not depend on our knowing that
2
in

V~-6 and 2

V
-

6 are primes; they actually are, but

it is

unimportant

our discussion.

We now
integers 0,

return to the discussion of unique factorization in the ordinary It will be convenient to have the following result. 2, 1,
.

Theorem 1.15
if p
|

a:a 2

'

If p ab, p being a prime, then p a or p b. More generally, a n then p divides at least one factor j z of the product.
|

Proof.

then (a,p) 1 and so by Theorem 1.10, p b. may regard this as the first step of a proof of the general statement by mathe matical induction. So we assume that the proposition holds whenever p
If

p )( a,

We

divides a product with fewer than n factors.

Now
c.
\

if

p ac where
|

a n then^? we apply the induc lip a some that for conclude tion hypothesis to subscript L p i
c

p
c
\

a^

a n that
,

is,

#2

<2

a l or p

Theorem 1.16
tion theorem.

The fundamental theorem of arithmetic, or the unique factoriza The factoring of any integer n > 1 into primes is unique apart

from the order of the prime factors.

First proof.

Suppose that there

is

Dividing out any primes an equality of the form

(1-3)

common

an integer n with two different factorings. to the two representations, we would have

where the factors p and

t

q,

no prime on the
because

left

side occurs
,

are primes, not necessarily all distinct, but where on the right side. But this is impossible

pl

q^
.

one of the q^ That

is,

q s and so by Theorem 1.15, /? x is a divisor of at least p must be identical with at least one of the q^.

Suppose that the theorem is false and let n be the smallest as the product of positive integer having more than one representation
Second proof
primes, say
(1-4)
It is clear

rt=Pip2-'Pr
that r and s are greater than

= M2'"qs1.

Now the primes p\,p^

,

'

'

'

>Pr

no members
a

common

with q^ q^ then we could divide prime,

in
-

common

it

q s because if, for example, p were out of both sides of (1.4) to get two

14
Divisibility

ofnlp v But this would contradict our assumption that all smaller than n are uniquely factorable. integers Next, there is no loss of generality in presuming that 1 < lt and we p q
define the positive integer
(1.5)

distinct factorings

N as

N=

(q l

- pjqfa

,= p^(p^

pr

q s).

It is clear that

Pi

is n, so that uniquely factorable into primes. But so (1.5) gives us two factorings of N, one involving^ and the other not, and thus we have a contradiction.

\ (?i - Pi),

N<

In the application of the fundamental theorem we frequently write any 1 in the form, sometimes called the integer a "canonical

>

form,"

P?pt '-p"r

where the primes p i are distinct and the exponents <x< are positive. However, sometimes convenient to use a slight variation of the canonical form and to permit some exponents to be zero. For example, if we want to describe the greatest common divisor g of a and b in terms of the prime factors of a and b, we would write
it is

(16)

a
;>

o ai

a2

ar

where a z

and
g

&

;> 0.

Then

the greatest

common
. .

divisor
ar ./? r )

is

seen to be

(a, b)

p^iMp
6

n**<**.f*).

^min(
.

where min (oc, 0) denotes the minimum of b = 225, we would have

a

oc

and

In case a

108 and

=2

3 3 5,

= 23

52

= 23 2 5 =
and b
is

9.

Similarly the least

common

multiple of a

seen to be

where max

(a,

/?)

denotes the

maximum

of a and

we

/?.

If

no a f

is

greater than

say that a

is

square-free.
is

Theorem 1.17 Euclid. The number of primes end to the sequence of primes
2,3,5,7, 11,13,
Proof.

infinite.

That

is,

there

is

no

Suppose there were only a

the

finite

Then form

number

number of primes F p. F p,

'

n Pr

'

1.3

Primes
that n
is

15

Note
of n

not divisible by p l or p 2 or

orp r Hence any prime

. .

divisor p

is

a prime distinct

fromp^p^

,p r Since n

is

either a

prime

factor/?, this implies that there

is

a prime distinct

prime or has a ,p r fiomp l ,p 2y

.

Theorem 1.18

There are arbitrarily large gaps in the series of primes. Stated otherwise, given any positive integer k, there exist k consecutive composite
integers.

Proof.
(k

Consider the integers

1)!

2, (k
is

1)1

3,

(k

1)1

k, (k

1)!

Every one of these

composite because j divides (k

1)1

+ k+l. + j if 2 ^ y

fc+

1.

The primes

we denote

the

are spaced rather irregularly, as the last theorem suggests. If number of primes that do not exceed x by ir(x) we may ask
9

about the nature of this function. Because of the irregular occurrence of the primes, we cannot expect a simple formula for TT(X). However, one of the most striking results in advanced number theory, the prime number theorem, gives

an asymptotic approximation for

lim
that

TT(X). It states

that

7rOr)^^

1,

is,

that the ratio of IT(X) to #/log x approaches

as x

becomes

indefinitely

large.

PROBLEMS
1.

With a and b

as in (1.6)

what conditions on the exponents must be

b)

satisfied

ifa\b? What
2.

conditions

if (a,

divisible

Observe that the definition that a is square-free amounts to stating that a is by the square of no integer greater than 1. What is the largest number

of consecutive square-free positive integers? What is the largest number of consecutive cube-free positive integers, where a is cube-free if it is divisible by the cube of no integer greater than 1.
3. In

any positive integer, such as 8347, the last digit is called the units digit, the next the tens digit, the next the hundreds digit, etc. In the example 8347, the units digit is 7, the tens digit is 4, the hundreds digit is 3, and the thousands
2 if and only if its units digit is digit is 8. Prove that a number is divisible by divisible by 2; that a number is divisible by 4 if and only if the integer formed
tens digit and its units digit is divisible by 4; that a number is divisible by and only if the integer formed by its last three digits is divisible by 8. 4. Prove that an integer is divisible by 3 if and only if the sum of its digits is divisible by 3. Prove that an integer is divisible by 9 if and only if the sum of

by

its

8 if

its

digits is divisible

by

9.

16
5.

Divisibility

the

Prove that an integer is divisible by 11 if and only if the difference between sum of the digits in the odd places and the sum of the digits in the even
is

places
6.

divisible

by

1 1

Show

a positive odd integer. 0, 7. Let a/b and c/d be fractions in lowest terms, so that (a, b) = Prove that if their sum is an integer, then \b\ = \d\. 8. Prove that any prime of the form 3k + I is of the form 6k + 1 n
r

that every positive integer n has a unique expression of the

form

2 r m,

(c,

d)

1.

Prove that any positive integer of the form 3k 4- 2 has a prime factor of the same form; similarly for each of the forms 4k H- 3 and 6k H- 5. 2 2 10. If x and y are odd, prove that x + y cannot be a perfect square. x 11. If and y are prime to 3, prove that x 2 + y 2 cannot be a perfect square. 12. Prove that (a, b) = (a,b,a + b) and more generally that (a, b) (a, b, ax + by) for all integers x, y.
9.
9

(a, a + k) Prove that (a l9 a a 15. Prove that (a, a 4- 2)

13.

Prove that

k
,

for all integers a,

k not both
,

zero.

14.

an )

((a l9

fl

2,

16. If (a, b)

= /?,

1 or 2 for every integer a. a prime, what are the possible values of (a2 b)l
,

a^

an ).

Of

(a
3

3
,

b)l

)? 4 17. Evaluate (#,/? ) and (a where/? is a prime.

(0
,Z>

Of

A,/?

2 given that O,/? )

/?

and

(Z>,/? )

= p2

(1.6), what conditions must be satisfied by the 2 2 exponents if a is to be a perfect square ? A perfect cube ? For a b ? For a \b ! 19. Prove the second of Theorem 1.13 use of the and l.c.m. part by g.c.d. formulas following (1.6).
\

18. If a

and b are represented by

20. Prove that (a 2 , b 2 ) = c 2 if (a, b) c. 21. Let a and b be 1 and ab is a positive integers such that (a, b) perfect square. Prove that a and Z> are perfect squares. Prove that the result generalizes to kih powers.

22. Given (a, b c)[a b, c] abc, prove that (a, b) (b, c) (a, c) = 23. Prove that [a b, c](ab, be, ca) = \abc\. 24. Determine whether the following assertions are true or false. If true,
9 9
9

1.

prove

the result, and

(1) If
(fl,

if false,

b)

(2) If (a, b)
(3) If (a, b)

= = =

give a counterexample.
[a, b]
,

(a, c)
(a, c)

then

[a, c].
,

(*, c)

then (a 2 A 2 ) = (a 2 c 2 ). then (a, b) = (a, 6, c).

|

(4) If/? is
(5) If/? is

(6) If (7)
(8)

a3 3 If a c 2 then a c. Ifa 2 |c 3 thena|c.

1 | 1
|

a prime and/? a prime and/? c 3 then a c.

a and/? (a 2 a 7 then/? a.
|

b 2 ) then/?

b.
|

(9) If/? is

(10) If/? is (11) If (a, A) (12)

2
[fl

a prime and/? (a 2 a prime and/? (a 2

| |

then (a2

+ b 2) and/? + b2) and/? 2 ab, b ) = 1.

2
|

(A
2
|

(6

+ c 2 ) then/? + c 2) then/?

(a
|

2
2

(a

- c2 + c2

). ).

,^,^]"^,^].

1.3

Primes
(13) If b
|

17
(a
2
2

(14)

Ifb\(a

(15) (a, 6, c)

+ 1) then b - 1) then = ((a, (a,

Z>
,

(a (a

4 4

+ TZ

1).

1).

c)).
is it

25.

For which

positive integers

true that

Given positive integers a and b such that <2 6 2 6 2 a 3 a 3 M, Z? 4 a 5 , , b. prove that a = am + bn 9 27. Given integers a,b,c,d, m, n, #, v satisfying^ be ~ 1, u
26.
|

= cm

-\-

dn, prove that (m,ri)

if

(u 9

v).

composite has a prime divisor^ satisfying/? ^ >/. 29. Obtain a complete list of the primes between 1 and n, with n = 200 for convenience, by the following method, known as the "sieve of Eratosthenes." By the "proper" multiples of A: we mean all positive multiples of k except k itself. Write all numbers from 2 to 200. Cross out all proper multiples of 2, then of 3, then of 5. At each stage the next larger remaining number is a prime.
28. Prove that

is

it

Thus 7 is now the next remaining larger than 5. Cross out the proper multiples of 7. The next remaining number larger than 7 is 11. Continuing, we cross out the proper multiples of 11 and then of 13. Now we observe that the next
problem
which
remaining number greater than 13 exceeds V200, and hence by the previous all the numbers remaining in our list are prime.
is

30. Consider the set

the highest

of integers 1, 2, 3, -,. Let 2 k be the integer in S k 2 of 2. Prove that is not a divisor of power any other

integer in S. 31. Prove that %f**i \lj is not an integer if n 1. 32. Consider the set T of integers l,3,5,---,2

>

in

T which is

the highest

power of 3. Prove

that 3 r is

Let 3 r be the integer not a divisor of any other

1.

integer of T. 33. Prove that

*-34.

1) is not an integer if n > 1. ^F=i l/(2/ a that Say positive integer n is a sum of consecutive integers and k so that n + (m (m + 1) + positive integers

if

there exist

that n

is

so expressible

35. If 2n 36. If 2 n

My =

Prove not a power of 2. is an odd prime, prove that n is a power of 2. is a prime, prove that n is a prime. (Numbers of the form
4- k).
if

=m +

and only

if it is

2P

1,

where

is

stated that the only primes p for which y 3, 5, 7, 13, 17, 19, 31, 67, 127, 257. However, 67 and

Mersenne

a prime, are called Mersenne numbers, because = 2, is a prime are p

example, and

M M

257

61 is

a prime. The question of which numbers

are not primes, for are primes is

M%

not

settled.)

37. If a

and b

>

2 are any positive integers, prove that 2

/.
1

1 is

not divisible

& by2 -1.

and / be given with^= of pairs positive integers x, y satisfying (#, y)

38. Let positive integers^

Prove that the number of

[x, y]

g and

/ is

2 k where
,

18

Divisibility

is

the

number of

distinct

x2 different pairs if x 1 3 be a fixed 39. Let k

prime factors of or y ^ y z .)

l\g.

(Count x l9 y and x 2} y 2 as
-

integer.
triplet

Find
is

all sets

a lt

a%,

a k of positive integers
of the
triplet.

such that the

40.

sum
2

of any
6

divisible

by each

member

Proyejhat

+V

and 2

- V -6

are primes in the class

C of numbers

+ W-6.
Prove Theorem 1.13 by use of the fundamental theorem.
is

41.

42. Prove that every positive integer

2'o

uniquely expressible in the form

+ 2h +

2>

+
<jm
.

+ 2V

where

m>

and

<y <j\ </2 <

43. Prove that

any positive integer a can be uniquely expressed in the form

where each b 3
(a

0, 1,

or

1.

44. Prove that there are n n

no

positive integers a, b, n

>

such that (a n

b n)

Prove that no polynomial /< of degree >1 with integral coefficients can = then if represent a prime for every positive integer x.
45.

Suggestion:

/(/
46.

+ kP)

-/(/)

is

a multiple of/? for every k and so/(j

9

/(/) p kp) has the

same

property.

Prove that there are

infinitely
24

2 21

1,2

22

many primes by
Suggestion:

+1,223 +

1,2

+ !,-.

considering the sequence use Problem 36 of

Section
47.

1.2.

The numbers

in the

Fermat numbers,

after Pierre

Verify that the fifth

9

sequence in the preceding problem are called the Fermat who thought they might all be primes. number in the sequence is not a prime by out
multiplying
1)(2
23

(2

27

2 21

2 19

2 17

14

1)

to get 2 32 1. 48. Prove that there are infinitely

many primes

of the form

4/z

of the

form 6/2+5. Remark. The

last

theory (the proof of which deeper than the methods of this book), due to Dirichlet, is that the arithmetic progression a, a + b, a + 2b, a + 3Z>, contains infinitely many are relatively prime, that is if (a, b) = 1. primes if the integers a and b >
lies
-

and 5, 11, 17, 23, 29, gressions 3, 7, 11, 15, 19, of primes. One of the famous theorems of number
-

problem can be stated thus: each of the arithmetic pro contains an infinitude
-

NOTES
and

ON CHAPTER

The symbol Z is widely used

because a rational

to denote the set of all

zero. Likewise the set of all rational

integers, positive, negative,

numbers

is

number

denoted by g, presumably

is

a quotient a/b of integers.

Notes on Chapter
It

19

Theorem
this

can be noted that the second proof of Theorem 1.16 does not depend on 1.15 or indeed on any previous theorem. Thus the logical arrangement of Theorems 1.14 and 1.16 in an chapter could be altered by
considerably
putting

and then using the formulas for (, c) and [b, c] following (1.6) to prove such results as Theorems 1.6, 1.7, 1.8, 1.10, 1.13, and 1.15. The prime number theorem, stated at the end of Section 1.3, is not proved in this book. A weaker version due to Tchebychef is given in Theorem 8.1. For a proof
early position,

of the prime number theorem

itself,

the reader

is

referred to the excellent accounts

by Hardy and Wright (listed in the General References on page 269) and by Norman Levinson, Amer. Math. Monthly, 76, 225-244 (1969).

2
Congruences

2.1

Congruences
1

It is

apparent from Chapter

theory, one that sets

that divisibility

is

number

a fundamental concept of

matics. In this chapter

slightly different point

we

A congruence is nothing more than a state ment about divisibility. However, it is more than just a convergent notation It often makes it easier to discover proofs, and we will see that congruences can suggest new problems that will lead us to new and
of view.
interesting topics.

apart from many other branches of mathe continue the study of divisibility, but from a
it

Definition 2.1
that a is
divisible

If an integer m, not zero, divides the difference a congruent to b modulo m and write a = b (mod
is

b,

we say
is

by m, we say that a

m).Ifa-b

not

we

write

a^b
-b

not congruent to b modulo m, and

in this

case

(mod m).
is

divisible by if and b is divisible by only if a confine our attention to a positive modulus. Indeed generally shall assume throughout the present chapter that the modulus is a

Since a

we can

-m
we

positive r

integer.

Congruences have many properties in

properties that follow easily

from the

theorem.

with equalities Some definition are listed in the followine

common

Theorem
(a)

2.1

Let

a, b, c, d, x,

y denote

integers.

Then:

b (mod m), b

a (mod m), and a

-b=

(mod m)

are

equivalent statements.

20

2.1

Congruences
(b)
(c)

21

If a

If a ==
m).

(mod m) and b b (mod w) #/

c ==

(mod m), then a d (mod m), f/ze

(mod

TW).

##

cy

bx

dy

(mod
(</)

Ifa~b
If a

(e)

(mod w) at/ c = d (mod m), ^e/z <zc = bd (mod (mod m) and d m, d > 0, ^/ze/z a = b (mod </).
\

m).

Theorem 2.2

Let f denote a polynomial with integral

coefficients.

If a

(mod m)
Proof.
integers.
,

then

(a}

=/()

(mod

m).

We

a 2 5= A 2 and finally

can suppose f(x) = c Q x n + cx n -^ + + cn where the ^ are Since a = b (mod m) we can apply Theorem 2.ld repeatedly to find a 5 == Z? 3 an = n (mod m), and then tyz *-' == (mod m), "1 -1 n = n
1

c a

+ ^a

71

cn

cG b

c^

71

c^
+

'

'

'

cn

(mod m),

by Theorem ax

2.1c.

The reader is, of course, well aware of the property of real numbers that if = ay and a ^ then x = y. More care must be used in dividing a con
a.

gruence through by

Theorem
(a) (b)
(c)

2.3

ax

ay (mod w)

if and only if x
(a,
i

=
1
,

2/

mod
(a,

1 .

If ax

y (mod [m l5
(<3)

(mod m) and (mod m for

ay
z -)

m)

then x =.
,

1, 2,

//

m)/ y (mod m). <m/ cw/y if x

=y
2.

2,

ra r ]).

Proo/.

If ax

ay (mod m) then ay

ax

= mz

for

some

integer

Hence we have
a
(a,
f x

(y

#)

m
;
:

m)

(a,

m)

and thus

m
(a,

m)
by Theorem
1.7

But
(y

(<2/(<z,

m), ml(a,m))
1.10.

=
x

and therefore {m/(a, m)}

x)

by Theorem

This implies

(mod m\(a,

m)).
x)
x),

If
(a,

a;

=y

m)(y

(mod m/(# m)) then {ra/(<2, m)} (y x). Using Theorem 1.1 we get m a(y
,
|

and

hence

and then

w ax =

ay (mod m) follows.
(b)

This
it

use

a special case of part (a). It often. very

is

is

listed separately

because

we

will

22
(c) If
i

Congruences

=y
,

(mod
r.

JH<)
is,

for

=
a

1, 2,

r,

then

/H<

(y

a?)
,

for
/w r ,

1, 2,

That
(see

is

common
[m^
a;

multiple of

/w x ,
|

and

therefore

Theorem
,

1.12)
,

mr

w (y

8,

a).

This

x y (mod [m 1? m 2 implies If x == y (mod [m^ w 2

,

wz r ]).

/w r ])

then

y (mod

/H<)

by Theorem

2.1e,

since

m$

[m l?

r ].

In dealing with integers

modulo

we

are essentially performing the

of m. In a ordinary operations of arithmetic but are disregarding multiples sense we are not distinguishing between a and a mx, where x is any

integer.

Given any integer a, let q and r be the quotient and remainder upon division by m; thus a = qm + r by Theorem 1.2. Now a = r (mod m) and,

since r satisfies the inequalities ^ r < m, we see that every integer is m 1. Also it is clear congruent modulo m to one of the values 0, 1, 2, that no two of these m integers are congruent modulo m. These m values
,

constitute a complete residue system definition of this term.

modulo m, and we now

give a general

Definition 2.2

Ifx^y
is
is

(mod m)

then y

is

called a residue

of x modulo m.

Asetxly x 2 ,-'-,x m
integer
It
is

there

called a complete residue system modulo one and only one xs such that y x^ (mod m).
-

iffor every

obvious that there are

the set
1, 2,
,

infinitely
1,

many complete

residue systems

modulo m,
if

being another example.

A set of m integers forms a complete residue system modulo m if and only

no two integers
in the set are congruent

modulo m.

Theorem 2.4
Proof.
(x,
|

Ifx

y (mod m) then

(x,

m)

(y,

m).

We
m) m)
\

have y

m) m, we have
(x,

find (y,

for some integer z. Since (x, m) x and and hence (x, m) (y, m). In the same way we (x, m) y = (y, m) by Theorem 1.1 since (a?, m) and then have m) (x, m)

mz

and

(y,

are positive.

Definition 2.3
that (ri9

A
1
,

m)

ri

reduced residue system modulo is a set of integers r i such ^ r, (mod m) ifi 5^ y, and such that every x prime to m, is
i

congruent modulo

m to some member r

of the

set.

In view of Theorem 2.4 it is clear that a reduced residue system modulo m can be obtained by deleting from a complete residue system modulo m those members that are not relatively prime to m. Furthermore, all reduced residue
will contain the same number of members, a number that systems modulo is denoted by <f>(m). This function is called Euler's ^-function, sometimes the
totient.
-

By applying
,

this definition

of

<f>(m)

1, 2,

m mentioned in the paragraph following Definition 2.2, we can get

to the complete residue system

2.1

Congruences

23

what amounts
theorem.

to an alternative definition of </>(m), as given in the following

Theorem 2.5
equal to

The number

<j>(m) is

the
to

that are relatively

prime
Let

number of positive m.
*

integers less than or

Theorem 2.6
residue system

Let

(a,

m)

1.

rl9 r 2 ,
,

rn

modulo m. Then ar l9 ar2 residue system, respectively, modulo m.

Proof.
If (r i9

ar n

be a complete, or a reduced, is a complete, or a reduced,

m)

then (ar i9 m)

by Theorem
,

1.8.

rn ar n as of r l9 r 2 There are the same number of ar l9 ar 2 we need only show that a^ ^ ar^ (mod m) if i ^ j. But Theorem 2.3b shows that ar i = ar3 (mod m) implies r i = rj (mod m) and hence z =7.
, ,

Therefore

Theorem 2.7 Fermat's theorem. Let p denote a prime. Ifp p = a (mod/>). (mod/?). For every integer a, a

)f

a then

<2

P ~~1

We will postpone the proof of this theorem and will obtain it as a corollary
to

Theorem
2.8

2.8.

Theorem

Euler's generalization of Fermafs theorem. If

(a,

m)

then

a* (m) as
Proof.

(mod

m).

Let

r 1? r 2

r^ (m}
,

by Theorem 2.6, flr l9 0r 2 m. Hence corresponding

ri

be a reduced residue system modulo m. Then is also a reduced residue system modulo flr^ (TO)
is
i

to each r i there

one and only one

,

ar^

such that

(mod m). Furthermore, different r will have different corresponding ar jt This means that the numbers ar l9 ar 2 ar^ (w) are just the residues modulo m of r l9 r%, but not necessarily in the same order. r^ (m) Multiplying and using Theorem 2.ld we obtain
flry
,
-

and hence
4>(m)

J
I,

=l

)-!!^ (modm)
i=l

a <f>M

JJ r
j=l

= JJ T
j=l

<}>(m)

(mod

m).

Now

(r^

m)
1

so

we can

use

Theorem

2.3b to cancel the rs and

we obtain

a^ (m) ES

(mod

m).
2.7.

Corollary.

Pr00/ o/ Theorem
find ^(/?)

If

/>

^a

then (a,/?)

and a* (p)

1
,

(mod

p).

To

we

refer to

Theorem

2.5. All the integers 1, 2,

p with the exception of p are relatively prime = 1 and the first part of Fermat's theorem <[>(p) p is now obvious. part
1,
,

to p. Thus we have follows. The second

24
Corollary 2.9 If (a, m) ^4// solutions are given by x
Pr00/.

Congruences

//z<?

ax

==

x^
1

Since (1,
set
is

~ 00^ = b b = (mod m) and hence any solution then ax #(# o^) (mod m). Using Theorem 2.3Z? we get x ^ = (mod m) = which implies x o^ + ym. The fact that all these actually are solutions follows from Theorem 2.2.

merely If #

xl

w)

a+

(m) - l

and

+ jm ^ m, we

& (mod where j =

/w) /zas

a solution x
2,
.

2^.

0,

see that

</>(/w)

1.

Then we need

b.

Ruler's function <j>(m)

is

of considerable

interest.

We

will consider

it

further in section 2.4

and

section 4.2.

Theorem 2.10

Wilson's

theorem.

If

is

a prime then (p

1)!

(mod p).
Proof.

Ifp

2 or p
little

3 the congruence
g: 5.

is

easily verified.

Now we can suppose p

we must
and
use a
care.

We

idea behind the proof is quite simple, but consider the integers whose product is (p 1)!

The

a way that the product of the two members of each pair is congruent to 1 modulo p. Given an integer j satisfying 1 <jy ^ p 1, then (/,/?) = 1 and we see by
try to pair

them

off in such

Corollary 2.9 that there

is

With each j we will associate the corresponding integer z. Since if = ji = 1 (mod/?) we see that/ is the integer associated with z. The integer 1 is associated with itself, and so is/? 1. Omitting these values for a moment we consider = (7 + 1 p) = 1 and hence 2 <j ^p 1, For these j we have (j 1 /?) 2 - l)(y + 1)^0 (mod/?) by Theorem 1.8. Hence each of these y 1 = / (j is associated with an 2, and the associate of is y itself. 7^7, 2 ^ f <;/? Thus the integers 2, 3, 2 can be p paired off,y' and its associate z, and
, ,

< ^p
z

exactly one integer

i

1.

Obviously

is

such thatyY impossible so we have

z

^ <p
z

(mod/?) and
1.

/f
1

(mod/?). Wilson's theorem and Fermat's theorem can be used to determine those 2 1 primes/? for which # (mod/?) has a solution. This is a special case of

(mod/?). Multiplying all these pairs together we get 2 3 (mod/?), and Wilson's theorem follows because 1 (/? 1)
1

==

(^
1

2)

==

some

results that

we

will take

up

later (see

interesting to see that this special case

Chapter 3). However, it is can be handled by fairly simple means.

==
1

Theorem 2.11 Let p denote a prime. Then x 2 = 2orp = l (mod 4). if and only ifp
Proof.

(mod

/?)

has solutions

we have the solution x = 1. For any odd prime p we can write Wilson's theorem
Ifp
2

in the

form

2(l-

..;...^)(i. ..(,_;-). ..(p-zxp-l))

(modp).

2.1

Congruences

25

The product on the left has been divided into two parts, each with the same number of factors. Pairing off j in the first half with the second half, we can rewrite the congruence in the form

p-jw

(p-D/2

II
Buty(/>

7*0 -J")

s -1
we

(mod

jp).

-/)

EE

-/
,

(mod/?), and so

have, ifp ==

(mod

4),

n
If/?

-/,

n,
(mod

. (1)/2

and so we have a solution,

^
1

2 and/?
/?)

=1

II&
4), ==

2 7 of x
?

then/?

= -1

(mod

4).

== integers, x*

-I

(mod/?), then

-.1
x

^ =

(mod

since (p

l)/2

(mod/?) by Theorem

we have a;*-1 == ( x *yv-v& == (~ i)to-i)/2 = 1 (mod 2). But clearly pjfx, so we have
This contradiction shows that x 2

(mod/?). In this case,

if

for

some

2.7.

= -1

(mod /?) has no

solution in this case.

PROBLEMS
1. List all

integers

2. Exhibit a

x in the range 1 ^ x g 100 that a; = 7 (mod 17). satisfy complete residue system modulo 17 composed entirely of multiples

reduced residue system for the modulus 12; for 30. an integer x is even, observe that it must satisfy the congruence x = (mod 2). If an integer y is odd, what congruence does it satisfy? What con gruence does an integer z of the form 6k + 1 satisfy? 5. Write a single congruence that is equivalent to the pair of congruences x = 1 (mod 4), x = 2 (mod 3). 2 2 6. Prove that ifp is a prime and a = b +
4. If

3. Exhibit a

Show that if f(x) is a polynomial with integral k (mod m), then f(a + tm) = k (mod 777) for every
7.
8.

(mod/?), then/?

(a

6) or/?

(a

).

coefficients

and

if

/(a)

Prove that any number which

its

integer

t.

is

for
9.

a square must have one of the following

units digit: 0, 1, 4, 5, 6, 9.

Prove that any fourth power must have one of 0, 1, 5, 6 for its units digit. 10. Evaluate ^(w) for m = 1, 12. 2, 3, 11. Find the least positive integer x such that 13 (x* + 1). 12. Prove that 19 is not a divisor of 4n 2 + 4 for any integer n. 13. Exhibit a reduced residue system modulo 7 composed entirely of powers
-

14. Solve 3x

15. Illustrate the

5 (mod 11) by the method of Corollary 2.9. proof of Theorem 2.10 for /? = 11 and

p =

13

by actually

determining the pairs of associated integers.

26
16.

Congruences

The
~= 1

5x
17.

(mod

integers 12, 23, 34, 45, 56 are congruent to 1 modulo 11. 5 9 and hence that x 11) we merely note that 45

To

solve

is

solution. Solve

ax

== 1

Prove Prove 19. Prove 20. Prove

18.

that n*

that n 7 that that

/z

- 1 is divisible by 7 if - n is divisible by 42, for 12 1 is divisible by 7 if w is divisible by 7 if

6fc

(mod

11) for a

2, 3, 7)

10.

(72,

=
=

1.

any integer
7)
1.

n.

(/z,

(n, 7)

1, A:

being any positive

n.

integer. 21. Prove that

7z

13

22. Prove that n12

23. Prove that 24. Prove that 25.
7Z

12
5

-i/?

by 2, 3, 5, 7, and 13 for any integer by 13 if n and a are prime to 13. # 12 is divisible by 91 if n and a are prime to 91. 3 4- J/z + ^72 is an integer for every integer n.
n
is

divisible

<a

12

is

divisible

What

is

400 the last digit in the ordinary decimal ? representation of 3

4 Suggestion: 3 4 that 3 implies

= 1 (mod 5) by Fermat's theorem, and this with 3 4 = = 1 (mod 10). Hence 3 4n - 1 (mod 10) for n >
any

(mod 2)

1.

400 the last digit in the ordinary decimal ? representation of 2 400 27. are the last two digits in the ? ordinary decimal representation of 3 20 E= use Theorem to 2.8 establish that 3 1 (mod 25). In addition, Suggestion:

26.

What What
1

is

20 = 1 (mod 4), whence 3 1 (mod 100). (mod 4) so that 3 20 28. Show that -(m - l)/2, -(TTZ - 3)/2, (m - 3)/2, (m - l)/2 is a com plete residue system modulo m if m is odd, and that (m 2)/2, (m 4)1 2, (m 2)/2, 777/2 is a complete residue system modulo m if m is
-

3 2 E=

even.
29.
is

Show Show
2.

that 2, 4, 6,

2/w

is

a complete residue system

modulo

if

m
if

odd.
that
I

30.

2
,

22 ,

ir? is

m >
32.
33.

not a complete residue system modulo

777

31. If n

is 1)! = composite, n > 4, prove that (n (mod TZ). Show that an integer 777 > lisa prime if and only if 777 divides (m -!)!

!.

For

positive integers a,

m, n with

m ^

n,

prove that

1, if

a a

is

even,

2, if

is

odd.
this to the

2m Suggestion: if p is a common divisor, a n~ m 2 777 77. < , power presuming

34.

= -1

(mod/?). Raise

For 77z odd, prove that the sum of the elements of any complete residue system modulo 777 is congruent to zero modulo m; prove the analogous result for any reduced residue system for m > 2.
35.

Find
b

all sets
c),

of positive integers a,

/3,

==

(mod
is

(mod

a), c

a (mod

b).

satisfying all three congruences Suggestion: if a, b, c is such a

set,
all

ka, kb, kc for any positive integer k. Hence it suffices to determine "primitive" sets with the property (a, b, c) = 1. Also there is no loss in
so also

b ^ c. generality in assuming that a 36. Find all triples a, b, c of non-zero integers such that a b c (mod |a|), c = a (mod |/3|).

(mod

|c|)

2.2

Solutions of Congruences
37. If/?
is

27

an odd prime, prove that:

I

32

52

(/?

2)

2 EE

(_i)<3H-D/2 (mod/?),
(_l)<*+-i>/a (mod/?).

and
22

42 6 2

(p

2
I)

38. Prove that (a 4- b)* =z a v

39. If r l9 r 2

r^^

is

b (mod/?), where/? is a prime. reduced residue system modulo a prime any

2-1

/?,

prove that

II>V
5=1

= ~1
,

(mod/?).

40. If r l9 r 2 ,

r^

and

ri, ^2,

r'p

are any two complete residue systems

modulo a prime p

set r^i, r 2 /*2, cannot be a , modulo residue system complete /?. 41. If/? is any prime other than 2 or 5, prove that/? divides infinitely many of the integers 9, 99, 999, 9999, . Ifp is any prime other than 2 or 5, prove that

>

2,

prove that the

r^

/?

divides infinitely
is

many

42. If /?

a prime, and
h
if

if

of the integers 1,11, 111, 1111, h + k 1 with /z ^ and

=p

0,

prove that

h\k\
43. 44.

(~-l)

=0(mod/7).
ap

which to base ten n, prove that there is an integer and 1 such that n m. Prove that the same holds for contains only the digits or and 9, but for no other pair of digits. and 2, or and 3, , digits
\

For any prime/?, Given an integer

b p (mod/?), prove that a p

6 P (mod/?2).

45. If n
46. If 47. If
1

is

^k <
3,

p =
(/?

/?

composite, prove that (n n - 1 prove that (n is a prime, prove that (/?

or
5.

1)!
2

+
-T

1 is

I)

(rf
1

not a power of n.
1).

1)!

+
(/?

2,

Suggestion:
is

if

/?

>

5,

(/?

l)/2,

and so (/?-!)!

divisible

by
72)

a power of/? if and only if 1, and !)! has factors 2, /?

is

I)

48. Prove

JJ

(mod

if

>

2.

the left denotes the product of all the positive integers x less than or equal to n such that both x and x + 1 are relatively prime to TZ. 49. Prove that there are infinitely many primes of the form 4/z + 1. 50. Prove that (/? - 1)! =/? - 1 (modi + 2 + -+(/?- 1)) if/? is a prime.

The symbol on

51.

For

positive integers
itself.

let r(/z)

denote the number of positive divisors of

\

n, including n

For d such that d

n,

^d^V

pair flfwith

/<f

to prove

that

T(/I)

<

2\Jn.

2.2

Solutions of Congruences

In analogy with the solution of algebraic equations it is natural to consider the problem of solving a congruence. In the rest of this chapter we will let

28

Congruences

f(x) denote a polynomial with integral coefficients, and we will write * -* n a n If u is an integer such that f(u) f(x} = aQx + a-p
1

(mod m). (mod m) then we say that u is a solution of the congruence/^) Whether or not an integer is a solution of a congruence depends on the modulus m as well as on the polynomial/^). If the integer u is a solution of (mod m), and if v = u (mod m), Theorem 2.2 shows that v is also a f(x) =
f(x)
u (mod m) is a solution of (mod m), meaning that every integer congruent to u modulo m 2 x + 4 = satisfies f(x) = (mod m}. For example, the congruence x = 3. solution 8. We can It also has the (mod 10) has the solution say x = = a; 8 3 (mod 10) and 8 (mod 10) are solutions. In this case, since 3 (mod 5), we can even say that x = 3 (mod 5) is a solution. In the general case, if ~ (mod m) has a solution u, it has infinitely many solutions all f(x) = u (mod m). It is more reasonable to count the integers v such that v
will say that

solution. Because of this

we

solutions in a different way. u (mod rri). In the example 3 3

We

will

not count v as distinct from u

separately.

if v

and 13 are not counted

However,

and

8 are

both counted since

Let
r l9 r 2 ,
,

3^8
r

(mod

10).

Definition 2.4

The number of solutions off(x)

f(r,)

denote a complete residue system modulo m. (mod m) is the number of the r i such that

(mod

m).

It is clear from Theorem 2.2 that the number of solutions is independent of the choice of the complete residue system. Furthermore, the number of solutions cannot exceed the modulus m. Ifm is small it is a simple matter to for each r of the and thus to determine the number of i just compute /(rj

solutions. In the

above example the congruence has just two solutions. Some

x2

other examples are

+ +

x2
Definition 2.5

= = =

(mod

7)
5) 8)

has no solutions, has two solutions, has four solutions.

-

(mod (mod

Letf(x)

a xn

(mod degree of the congruence f (x) the smallest positive integer such that a 3
-

+ a&^ +
m)

a n If a*
.

is

n.Ifa (mod m);

Q

& (mod m) the (mod m), let j be

congruence

is

j.

If there

is

no such integer j, that

is

is,

then the degree of the if all the coefficients

off(x) are multiples

ofm.no

degree

assigned to the congruence.

It should be noted that the degree of the congruence /(a?) = (mod m) is not the same thing as the degree of the polynomial /(#). The degree of the congruence depends on the modulus; the degree of the polynomial does not. Thus if g(x) = 6x* + 3x* + 1, then g(x) = (mod 5) is of degree 3, and == is of whereas is of degree 3. (mod 2) g(x) degree 2, g(x)

2.3

Congruences of Degree
\

29
0,

Theorem 2.12
then u
is

If d m, d > a solution off(x) =

and

if

is

solution off(x)

(mod

m),

(mod

d).

Proof.

This follows directly from Theorem 2.1 e.

PROBLEMS
1.

If/(#)

(mod p) has no
solutions.
2.

(mod/?) has exactly j solutions with/? a prime, and^Or) = == solutions, prove that f(x)g(x) (mod p) has exactly j
==

Denoting the number of solutions of/ (x)

k (mod

777)

by N(k), prove

that

a congruence /(#) = (mod m) has m solutions, prove that any integer whatsoever is a solution. (In such a case the congruence is sometimes called an
3. If
4.

identical congruence.) The fact that the product of

3 leads to the identical
this,

any three consecutive integers congruence x(x + \)(x +2) =0 (mod and write an identical congruence modulo m.

is

divisible

by

3).

Generalize

2.3

Congruences of Degree
1

Any

b (mod m), a ^ then ax ~ b (mod m) (mod m). From Corollary 2.9 we see that if (a, m) has exactly one solution, x = x (mod m). Let g denote (a, m). If ax = b (mod m) has a solution u, then b = au (mod m) and hence Z? = au = (mod g). Therefore, ax = b (mod m) has no solution if ^ i. However, if # b then, for w an integer, au == & (mod m) holds if and only if (a]g)u = (&/#) (mod m/#) by Theorem 2.30. Now ~ (&/g) (mod mjg) has just one 1 and the congruence (a/g)x (a[g, m/g) = x words the solutions of ax = b (mod m) In other solution #! (mod m/g-). = u u such that the are ^ (mod m/g), that is w = a;x + t(mlg) integers 1 then u takes on r the values 0, 1 If is 1 t = 0, 2, given g m. If t is modulo are which two of values no congruent given any other g one of these g m modulo to u be will the value, congruent corresponding values. Thus the solutions of ax = b (mod m) are x = x l + t(mfg) (mod m),

congruence of degree

can be put in the form ax

0<t^g-l.
Theorem 2.13

g g

)(

b.
1

Ifg where X Q
|

it

Let g denote (a, m). Then ax has g solutions x = (bjg)x

is

any solution of(a/g)x

+
1

(mod m) has no solutions = 0, 1, t(m]g) (mod m), t (mod m[g).

b
-

if
,

Proof.
(ajg)x

This theorem follows from what has already been proved since = (b/g)xQ is a solution of 1 (mod m/g) has a solution a: and then x^
(big)

5=

(mod

m/g).

30

Congruences

For reasonably small numbers the solution of a congruence can often be obtained by inspection or by trying all integers of a complete residue system

modulo m. However,

if

the

numbers

of a congruence of the form ax

are large the actual numerical solution (mod m) can be rather lengthy. The

hardest part is solving congruences ax 1 (mod m) with (a, m) 1. The solution as given in the proof of Corollary 2.9 is not usually very practical. number of special methods of solution have been developed, but perhaps

the best general method is to use Euclid's algorithm. Using Theorem 1.11 we determine g (a, m) and at the same time obtain integers u and v such that

au

mv

= = g. Then we can take u for the X

in

Theorem

2.13,

and the

rest is

easy.

Another way to go about solving a congruence of degree 1 is to factor the modulus m XILiX*- Writing m i p%* we note that the m i are relatively m k] = m. From Theorem 2.3c we see prime in pairs and that [w 3 ra 2
,

that the problem of solving ax

, k, simultaneously. The individual congruences ax b (mod raj may be easier to solve since their moduli i are smaller than m. Suppose the congruences ax b (mod 777,-) have the solutions x u i (mod t ). There still remains the problem of finding the simultaneous solution x of the set of congruences. The proof of the next

congruences

ax

(mod m)
i

is

(mod

ra z-),

equivalent to solving the set of

1, 2,

theorem

will

show how

this

can be done.

Theorem 2.14

Chinese remainder theorem. Let

m m
l9

2,

m
-

r
,

denote r

positive integers that are relatively prime in pairs, and let a l9 a z , r the x Then a i any integers. i (mod raj, 1, 2, congruences common solutions. Any two solutions are congruent modulo m-jn^

a r denote
,

r,
.

have

mr

Remark.
there

If the

moduli

l9

ra 2 ,

are not relatively prime in pairs,

not be any solution of the congruences. Necessary and sufficient conditions are given in Problem 14(c) in the next problem set.

may

Proof.

Writing
nij)

m=

(m/m i9

=
1

m^m^

mr

we

see that

m\m^

is

an integer and that

if /

1.

Therefore, by Corollary 2.9, there are integers b f such that

m,). Clearly

(m/m^bf s= define X Q as

(mod

(m/m^ =

(mod

m,)

^ j. Now

if

we

3=1

we have

2m
3=1

0^-

=m
__

m,-

m.

bia i

=^

(mod

ra f )

so that x Q

is

common

solution of the original congruences.

2.3

Congruences of Degree
If #

31

and x1 are both common solutions of x = a (mod m^, i = 1, 2, r, r and hence # = % (mod m) by then x Q = x l (mod m^, for z = 1, 2,
,
,

Theorem

2.3c. This

completes the proof.

we are to find the common solutions of x = 1 or 2 (mod 3), x = 1 or 2 = 1 or 2 (mod 5). We have m^ = 3, m 2 = 4, m z = 5, m = 60, (mod 4), x and each a is 1 or 2. To find b^ we solve (60/3)^ = 1 (mod 3); that is 20^ = 1 and i x = 1 (mod 3). We can take b = 1 (mod 3), which is the same as 20. Similarly we obtain b z = 1 and (m/m 2 )b 2 = then have (mlm l )b l = = 2, and can 15. For 3 we have 12 3 = 1 (mod 5), 2Z? 3 = 1 4Z? 3 = 2, 3 24. Using (2.1) we have merely to insert the take 3 = 2, (ra/w 3 ) 3 = 15a 2 20<z x 24# 3 (mod 60). Doing this we obtain values of the a in x =
i
Z?
,

The proof of this theorem provides us with an efficient method for solving a certain kind of problem. As an example let us rind all integers that have remainders 1 or 2 when they are divided by each of 3, 4, and 5. In other words

the values given in the following table.

(mod

60)

The
3

integers having remainders

1,

2, 17, 22, 26,

1 or 2 when divided by -14, -19, -23 (mod 60).

3, 4, 5 are

given by

PROBLEMS
Find all solutions of the congruences 2Qx =4 (mod 30); (a) 2Qx == 30 (mod 4); (b) = 254 (mod 400). (c) 353a; 2. How many solutions are there to each of the following congruences: (a) \5x = 25 (mod 35);
1.

(b)
(c)

I5x

=24 (mod

35);

I5x

=0

(mod 35)?

32

Congruences
3. Find the smallest positive integer (except x = 1) that satisfies the following = 1 (mod 7). congruences simultaneously: x = 1 (mod 3), x == 1 (mod 5), x 4. Find all integers that satisfy simultaneously: x == 2 (mod 3), x =3 (mod 5),
a?

==

(mod
all

2).

5. Solve the set

6.

of congruences: x

(mod 4),

a?

z=

Q (mod

Find

integers that give the remainders 1, 2, 3

3), x =5 (mod 7). when divided by 3, 4, 5

random from 1, 2, 3, 14, and 6 is selected at random = b (mod 15) has at least from 1 2, 3, 15, what is the probability that ax one solution? Exactly one solution? 8. Given any positive integer k, prove that there are k consecutive integers each
,
, ,

respectively, 7. If a is selected at

divisible
9. If

by a square

>

a solution of the congruence ax = b (mod m) (obtained perhaps by = x2 + application of Euclid's algorithm to a and 777), prove that x t(mlg) (mod m) gives all solutions as t runs through 0, 1, 1, where g is ,g

x2

is

= 1, and let % denote a solution of ax = 1 (mod m). For = \\a (1/0)(1 - axj) s Prove that x s is an integer and s that is a solution of ax = (mod m ~ (mod m ) is 11. Suppose that (a, m) = 1. If a = 1, the solution of ax = ^(1 - )J<z = = x a m If a then is odd and (mod 2, obviously
10.

defined as^-

=
,

(a,

m).

Suppose
-

(a,

m)

1, 2,
it

let

xs

).

).

772

a;

77z

s (mod 772 ) is the solution of ax = to show that the solution of ax =

1 1
s

s (mod m ). For all other a use Problem 10 s (mod m ) is x = k (mod 777 s) where k is the
.

nearest integer to 12. Solve 3>x = 1

13. Let
772

(\la)(\

ax^)

(mod
,

l5

2,

congruences b t x = congruences have a simultaneous solution. e 14. (a) Consider the set of congruences x = a t (mod , r, p *), i = 1, 2, == with e 1 ^ e 2 = x a t is a simultaneous solution of these = e r Prove that e a for i = 2, 3, r. (a congruences if /> (b) Let the canonical factoring of TT? be/?^2 Prove that any simul .^ taneous solution of the set of congruences x = a (mod 1, 2, -,, /^')> * is a solution of # = ^ (mod 772). (c) Prove that the set of congruences x == (mod 777,-), / = 1, 2, TZ, has a
,
'
'

by Problem 11, taking x = 2. 77? be r relatively prime in pairs. Assuming that each of the a t (mod 77? f ), i = 1, 2, r, is solvable, prove that the
125)

"

ft-

simultaneous solution

if,

and only

if,

(m^

^ m)
3
,

(a^

a}) holds for every pair

1

of moduli, that

is

for every pair of subscripts i,j such that

-

two solutions are congruent modulo [m ly m z 15. Let (a, b) = 1 and c > 0. Prove that there
(a

<j ^

n.

Any

TW W ].
is

bx, c)

an integer x such that

1.

2 16. Consider a square divided up into n equal squares. Number the columns of small squares 1, 2, n from left to right. Similarly number the rows 1 TZ and let 2, {c, r} denote the small square in the cth column and rth row. Let <z 6 a, b, a, be positive integers less than or equal to 72 and such that (a, n) = (b, n) = (a, TZ) = (/?, 72) = 1. Write 1 in square {# /3 }. Then count
-

columns to the right and b rows up from

this square. If this takes

you outside

2.3

Congruences of Degree

33

you

of the large square, count as if the large square were bent into a cylinder. Thus will arrive at {aQ + a, b + b} or {aQ + a - TZ, b + b] or {a + a, b + Q Q b + a TZ, /3 + Z> #}, whichever is one of the small TZ} or {a

actually squares. Write 2 in this square. Count a to the right and b up from 2 and insert 3 in the square. Continue until you have written in 1, 2, 3, , n. Prove that 772 will have been written in de {x m ,y m } where x m and y m are

uniquely

termined by

2/m

!) ( 1)

mod ")>

1 1

(mod TZ),

^ ^m ^ ^ m ^
2/

n
TZ, 1

^m ^

TZ.

Also show that

the process one

all

these squares {x m ,

y m] are
1

different but that continuing

more step would put n

,

in

{ar

which is already occupied

by
n

1.

Now having reached {a b ] again, count a to the right and p up and write + 1 in this square which may or may not already be occupied. Then revert to the original process with step a, b to insert n + 2, n + 3, 2n. Continue in this manner, using the extra step a, p just for n + l,2n + 1, l)n + ,(
-

and stopping when r? has been inserted. For 1 ^ m ^ n2 prove that ;w is in {# m 2/ m } where
,

#0

^-

1)

Fir]
1~|

(mod/2),

^ Xm g

72,

[m
and
(a/3
1

(mod 72),
J
TZ

l^y m
m
1.

^n,

\m /3a,
T72

11
is

the quotient
1

when

is

divided into

Also prove that

if

TZ)
TZ

then each square contains one and only one integer

(ap
/3a,
TZ)

T72,

2
.

From now on assume

1.

gs ^ qn + s +

column are just the m = = s ^ TZ 1 for which aQ 1 and as ^ c aq ^q :g TZ 1 (mod TZ). Prove that there is one and only one s for each q and that each s is distinct from all the others. Then show:
TZ

Writing

TTZ

= ^TZ +

s,

show

that the entries in the cth

,

Sum

of

m in

cth

column

v >
<7=0

^TZ

TZ

+ x^ /s
s=0

4-

TZ

Prove the same for the sum in a row. Since 7i(72 2 + l)/2 is independent of c, the sums in each row and in each column are the same, and the square array of integers is a so-called magic square.

Note

that the initial square

{<z

/3

} is

essential conditions are that a, b, a,

p9

aft

subject to no conditions. The only /3oc are relatively prime to TZ. Show
TZ

that these conditions cannot be fulfilled if

is

even. However, for

odd

TZ

the

values a

/3

= l,/?=2 always

give a magic square.

34

Congruences

2,4

The Function

4>(n)

We will

section. In this section

return to the discussion of the solution of congruences in the next we will use the Chinese remainder theorem to obtain
(f>(ri)

an important property of the function

of Definition 2.3.
positive, relatively

Theorem 2.15 Let m andn denote any two Then c/>(nm) = <j)(ri)<f>(m).
Proof.

prime

integers.

Denote <(ra) by j and let modulo m. Similarly write k for

residue system

r 1? r 2 ,
</>(ri)

r,-

and

let

be a reduced residue system s l9 s%, sk be a reduced

-

modulo n. If a; is in a reduced residue system modulo mn, then = (x, n) = 1, and hence x = rh (mod m) and x = si (mod /z) for some (#, m) h and z. Conversely, if x = 7^ (mod m) and a; = ^ (mod 72), then (a;, mn) = 1. Thus a reduced residue system modulo mn can be obtained by determining all x such that x = rh (mod m) and x = s i (mod 72) for some h and z. According
to the Chinese remainder theorem, each pair h,
i

modulo mn.

Clearly, different pairs h,

yield different

determines a single x x modulo mn. There are

jk of these pairs. Therefore a reduced residue system modulo mn contains jk = <f>(m)</>(ri) numbers, and we have <j>(mn) = cf>(m)cf)(n). = 1 and It is essential that m and n be relatively prime. In fact <f>(2)

Theorem 2.16
Remark.

Ifn

>

then

</>(n)

= w H^k (1 ~

I/?)all

^^ <(!)

1-

The symbol IX,

n

Thus

if

an analogous -)II^=i (1 notation regarding sums. We will also write to mean the sum over all d n the positive divisors of or not. Furthermore we will sometimes use prime the convention that an empty sum is 0, an empty product is 1. Had we done
5

= p^pl? We l//?

means the product over

primes that divide

1/p)

n.

*^r in canonical form, then IJj,| w (1 will often use this notation as well as

means

so here

we would not have had to

treat n

as a special case in the statement

of the theorem.
Proof.
If
It is

>

obvious that <(!) we can write n

'
'

(P?>P#LP'#i
repeatedly

'Pr

= = pl pl = for 7 = 1, 2,
1.
l

pr r
r

in
1.

canonical

form.

Now
2.15

Applying Theorem

we obtain

e In order to compute <f>(p e ), p a prime, we recall that </>(p ) e e 6 integers x such that 1 <; x <,p (x,p ) = 1. There are/?
,

is

the

number of

integers x between

2.4
1

The Function

35
<f>(n)

and/>% and we must count all of them except />, 2p, 3p,

"1

j^

^. Therefore

and hence

3=1

Another way of writing the

result of

Theorem

2.16

is

Theorem 2.17
Proof.
If
rc

Forn^l

we have

^dln

<f>(d)

n.

=/?%/? a prime, then

(p

1)

p(p

1)

p '\p

1)

n is a power of a prime. Now we proceed by induction. We suppose the theorem holds for integers with k or fewer distinct prime factors and consider any integer TV with k + 1 distinct prime 6 factors. Let/? denote one of the prime factors of TV, and let/? be the highest
Therefore the theorem
is

true

if

power ofp that

(/?, TZ)

divides N.

Then

N = p n, n has k distinct prime factors, and

e
,

1.

Now as Granges

over the divisors of

2 the set d,pd,p d,

e ,p d

Hence ranges over the divisors of N.

we have

2
d\N

#<*)

2 #<o + 2
d\n
d\n d\n

2 ^^ +
d\n

"

"+

2
d\n

^oou + KP> +

fa*)

d\n

d\p

In Chapter 4 we

will obtain a different

proof of

this

theorem.

It will

be

this section, independent of the results of be reversed, that we can begin by proving Theorem 2.17 and obtain Theorem Theorem 2.1 6 from it. It is then an easy matter to obtain Theorem 2.15 from

and we

will find that the

order can

2.16.

36

Congruences

PROBLEMS
1.

2.

3.

For what values of n is <() odd? Find the number of positive integers ^ 3600 that are prime to 3600. Find the number of positive integers g 3600 that have a factor greater than

in
4.

common

with 3600.

5.

Find the number of positive integers ^ 7200 that are prime to 3600. Find the number of positive integers ^ 25200 that are prime to 3600. (Observe that 25200 = 7 x 3600.) 6. If m and k are positive integers, prove that the number of positive inte

gers rg
7.
8. If

mk

that are prime to

m is k(/>(m).

Show

= P<f>(m)<j>(n)l<f>(P). Hence if (m, n) > <j>(mri) 9. If </>(m) = <f)(mri) and n > 1, prove that n

that c/>(nm) n<j>(m) if every prime that divides n also divides m. and n 9 prove that denotes the product of the primes common to

m
is

1,

prove 2 and

</>(mri)

>
>

</>(m)<l>(ri).

odd.

10. Characterize the set of positive integers n satisfying <f>(2ri) 11. Characterize the set of positive integers satisfying </>(2ri) 12. Prove that there are infinitely many integers n so that 3 < 13.

<f>(ri).

4>(ri).

<t>(n).

Find all solutions x of <(#) = 24. 14. Prove that for a fixed integer n the equation
solutions.
15.

<f>(x)

n has only a

finite

number of
exactly

Find the smallest positive integer n so that <j>(x) = n has no solution; two solutions; exactly three solutions; exactly four solutions. (It has been conjectured that there is no integer n such that (#) = n has exactly one solution, but this is an unsolved problem.)
16. Prove that there
is

no solution of the equation

(/>(%)

14,

and

that 14

is

the least positive even integer with this property. Apart from 14, what is the next smallest positive even integer n such that ^(x) = n has no solution?
17.

Prove that for n

is /z</>(/7)/2.

^2 the sum of all positive integers less than n and prime

odd prime
factors,
k prove that 2
\

to n

18. If n has

distinct

<f>(ri).

positive integers less than n that Prove that/(w) n. f(ri) implies f x such that 20. Let (ri) denote the number of integers
19. Define /(/?) as the

sum of the

and prime
1

to n.

<f>

n and

(x, n)

(x 4-

n)

1.

Prove

2>|n

21. (a) Let the canonical factorization of n be n

positive integer y, define

= H*=

!/??*'.

For eyery

otherwise.

2.5

Congruences of Higher Degree

Prove that

37

2?= i e j(pi)

n lpi> and

more

generally

for

^ k.

'Pi
(b)

Prove that

and hence that

(c)

#(/i)

J-Li

TTLi tt ~

Deduce

that

3=1

22. If

and so obtain an independent proof of Theorem 2.16. d n and < d < n, prove that n - <f>(ri) > d - <j>(d). 23. Prove the following generalization of Euler's theorem:

for any integer

a.

2.5

Congruences of Higher Degree

There is no general method for solving congruences. However, certain reductions can be made so that the problem finally becomes that of solving can use the method of the Chinese congruences with prime moduli.

We

remainder theorem in the

If

first

m = pl p
l

step of this reduction.

cf

p*r then the congruence /(x)

to the set of congruences /(#) 1, 2, (mod /?/), f that solutions of one are solutions of the other. If for some 7,
,

= =

(mod m)
r,
1

is

equivalent
the

in the sense

(mod m) has congruence /(a;) (modp**) has no solution, then /(a) no solution. On the other hand, if all the congruences/^) = (mod/?^) have solutions, we can suppose that the zth congruence has exactly k i solutions,

^j^r,

af \ No two of these are congruent modulo p*, by say a^\af\ Definition 2.4, and every solution of/(x) (modpl ) is congruent to some
(

ai 3) modulo^-. Now an integer u

(

(mod m) if and only if for each z the moduli p\ l are relatively Since a^^ (modp^). in the Chinese remainder theorem is prime pairs, applicable. We determine ei & b 1 and can then find u by means such that (mod p?*) integers z mp^ i
is

a root of f(x)

there

is ay'f

such that u

38

Congruences

of (2.1):
r

(2.2)

~ ba
-^
*

i}

i*

mod

)'

When
&i

best to compute the coefficients actually solving a problem it is usually first since they are independent of the choice of the/,-. It is then easy mp^ Ji) in (2.2), and the problem is solved. to insert the various values of the a-

bi

be a different u modulo m for each choice of the integers take on any of k i values. Therefore the con >Jr> anc* each 7;. can J\9J29 - k r solutions. Since k t is the number of = has kjk^ (mod m) gruence/^) solutions of /(a?) = (mod />**)> we have the following theorem.

There
'
*

will

"

Theorem 2.18
(x)

Let N(m) denote the number of solutions of the congruence r = P* is the (mod m). Then N(m) = II<=i ^O ?*) tf m
7

P^

'

'

'

canonical factorization

ofm.

the case in which N(p*f) Perhaps it should be remarked that 2.18. Theorem in out not is ruled j

for

some

Example.

Solve x 2

x Trying the values 3 solution. Since 15

+x + = 0,
5,

7
1,

=
2,

(mod 15). we find that a2

(mod

5)

has no

the original congruence has no solution.

Solves 2 + x + 7 = (mod 189), given thata = 4, 13, -5 (mod 27) 2 are the solutions of x + x + 7 = (mod 27) and that x = 0, -1 (mod 7) are the solutions of x* + x + 7 = (mod 7). 3 = 27, p^ = 7, a[1] = 4, In this example we have 777 = 189 = 27 7 = 3 7, /?Ji a) _ __ _i ^ o .g ncj ^ we (3) _. _ 0(2) Q^ ^2) i _5^ a multiply the congruence 13^ fl = s 4 b obtain = and 1 (mod 27) (mod 27). For Z> 2 we have 27Z> 2 = 28Z?! : 7&i by 4 = now write (2.2) as We can -1 us Z? 1 (mod 7) which 7). (mod 2 gives
Example.
-

3 -

4<4V

27(-l)4^>

zSfl^^

27'4'V

(mod

189)
77,

for the required roots w. Using the known values of the a(^

we quickly find u =

-14, -140,
If the

50, 13,

-113 (mod

189).
b%.

numbers had been larger we might have had more trouble finding b^ and

In any case the methods mentioned in Section 2.3 can be used.

PROBLEMS
1.

Solve the congruences:

x* x*
2,3

+2x -3 =0 (mod 9); +2x -3 =0 (mod 5); + 2x - 3 = (mod 45).

2.6

Prime Power Moduli

2 Solve the congruence x + 4x + 8 = 3 x 9x 2 + 23x 3. Solve the congruence

39

2.

(mod

15).

(mod 503) by observing a prime and that the polynomial factors into (x 5). 3)(x \)(x 3 9x2 + 23x - 15 = 4. Solve the congruence x (mod 143).
that 503
is

15^0

2.6

Prime Power Moduli

The problem of solving a congruence has now been reduced to that 'of solving a congruence whose modulus is a power of a single prime. If r is a solution of /(a?) = (mod /) for t = (mod /), then/(r) =~ 1 ,s. Let z' ', x *\ o^ be the solutions of /(a) = (mod /). 1, 2, There may be no such solutions, or there may be many. Consider s ^ 2. If = (mod /7 s"1 ) such then there is a solution there is a solution x of/(z) = ^1 (modp 8-1). Therefore o?^ = rr^i + v^p*-1 (mod/) for some thatrcj
(

integer

v^.
f(x)
-

Remembering that/ (x) is a polynomial of degree n with integral coefficients,

we

see that

f'(x)

are polynomials with integral coefficients

t

and that/ u) (o;) is identically zero for f(x) is finite and we have

>

n.

Thus the Taylor's expansion of

f(x

h)

=f(x) +f(x)h

f"(x)h*

^f \*W,
1

(n

and then
EE

f(x) = f(xll +
==

v s _ lP
)

-1
)

= /(a;^) + /'(x^Dv^p-

(mod p

).

But/(o?iffl)

(mod^

-1

so

we have

(2.3)

/'(^)^ = - -/(i)
if

Conversely,

(2.4)

/'(e> s - -^/(^-i) =
s-i

(mod

p),

all the solutions i/' ) = (mod/). This shows us how to find 1 = of those we know if s (mod/* ). For f(x) (mod/), ^ 2, off(x) each root x^ we find all the solutions v of (2.4), and then the integers x i + i/"" 1 will be solutions off(x) = (mod /). It can, of course, happen that there are no v corresponding to some x 8 _i. In this case we have no 5 solutions off(x) = (mod/? ) arising from this particular ^i\.

40

Congruences

$*z2, we start with the solutions x[ off(x) = (mod Fixing upon a ^ we must first solve = for v with s For each v we 2, x[ (2.3), particular have a root 4 = a4' + v^p (mod/? ) off(x) = (mod/? Using each one of these 4 we must then solve (2.3) with s = 3,j = k, in order to find solutions of/ (x) = But the congruence for has modulus/? and (mod/? 4 = x[ (mod/?), and hence we can write it as
3}

We can say a little more about the solutions.

3 fc)
)

In solving/(#)
/?).
:.

(modp
1

),

).

fc)

).

i?

fc>

3*}

f'( x

i) v

j
z

f( x zj

(mod

p).

P
This happens at each stage, and hence we

may

determine t?^ from

(2.5)

/'(^K-i = -

;r~ /(^-i)

(md P)
}

= (mod/?). all the a;^ that ultimately arose from the solution x[** off(x) The congruence (2.5) is a linear congruence. Iff'fai *) 7^ ( m od/?), then there will be exactly one v s _ : for each of the 0^1 arising ultimately from x[ Ji \ 1 Iff'tyi**) = (mod/?), then there will be/? or no t?^ according as/^^)//?^ is or is not to modulo congruent p.
for
3

Example.

Solve x 2
find that

+
x

+
=

(mod
3)
1

27).

By

trial

we

== 1

(mod
2x

is

the only solution off(x)

==

=
is

(mod

3) for

the present/^). Then/'(oO and (2.5) reduces to

and/'(l)

(mod

3).

There

only one x l9

which means that there are no

-1 (mod

3)

if/(a^i)
1}

v^ if /(^) ^
3 s).
1]

(mod

3 s)

and that

v s _^

0, 1,

(mod

We now find

4 = (mod 3),f(4 4 = (mod 32),/(4 4 = 4 (mod 3 ),/(4

1 1
1}

)
1)

=9,
)

^ = 0,
no
z; z;

1,

-1 (mod
-1 (mod

3)

2)

2)

= =

9,

27,
9,

=
i?

0, 1,

3)

^ 23)

4 4
Example.

X)

2)

43)

s -2 (mod 3 2),/(43) ) = = 4 (mod 3 3) = 13 (mod 3 3) s _ 5 (mod 33^ +# +

)

no

Solve x*

(mod

3 4).

Continuing from the previous example, we find

/(4

X)

27,

/(4

2)

189,

/(4
189

3
>)

27.
3 4).

The congruence has no

solution since 27

0,

&

(mod

2.6

Prime Power Moduli

Solve
oc

41

Example.

+
2)

==

3 (mod 7 ).

The
/'(O)

solutions of f(x)

=0

(mod
will
1.

7) are

==

0,
1}

!,/'(-!)

-1. There

one

2)

corresponding to
u s_!

4 =

be just one

(mod 7). Furthermore, = Oand corresponding to a?f

1

Now
7)

(2.5)

becomes

= -

^/(4-i) (mod
7)

corresponding to

1}

=
-1-

y s _i ss

zi/(4-i) (mod

corresponding to

4 =
2)

Then we
4*>

find

=
=

0,

/(4) /(4
2)

7,

!>!

-!,

1}

-7,
i>

1} /<4 ) = = -1, 2)

49,

4 =

-56,

2)

-1,

=7,

^=
7

1,

4
7
3
)

2)

=
x

6,

/(4

v2

= 49, = l,
7
3
)

42) =55.

The solutions of 3 (mod 7 ).

2 cc

4-

a?

(mod

are

= -56

(mod

and x

55

When

solving numerical problems one often

must determine whether or

1 and k is not too not one integer k divides another integer n. If (k, 10) there is a rather simple way to do this. As a first example consider large,

=
|

31

it

23754.
Since

Then n
(31,

=
31

10-2363.
2363.

- 4k = 23754 - 4 = we see that 10)

1

30
31
|

4 = 10(2375 -4-3) 23754 if and only if

still

We can repeat the argument to reduce 2363

23754
12

further.

The entire

process can be put in a more convenient form.

2363
9

227"

T
+
bj),

21

31

1 23754.

This process can be used for any k whose = Wa b. Then n bk 10/ + 1 and n

last digit is

We
IQbj

can write k

10a

+
+

10(0

= -

and k n
\

if

and only

if

k
\

(a

bj).

If the last digit of

is 9,

we can

write

we have n
if

+ + bj). If the last digit of k is 3, we can write 3k = 10; n 4. ibk = 10(0 + /), and hence k\nit and only if (0 + */) Similarly, n if and only if is 7, we write 3k = Wj + 1 and obtain k if the last digit of
bk
10a b b
10(0
bj).

IQbj

1Q/

and n

=
\

100
if
1

Then k n

(a

b, and and only and find

fc

A:

k\(a-

bj).

42
Example.
19

Congruences

=
6

10

10

20513

8638
_16

2057

847
14

J4
219
18

70

39

19^20513

7
|

8638

PROBLEMS
but in general the 1. The foregoing method determines whether k divides number we finally reach is not congruent to n modulo k. Consider the following
,

scheme, exemplified for n

1234.

1234 369
108 30
9

Here we have written down 1234, then 3 123, 3 36, 3 10, 3 3, in turn. We drop oif the right digit at each step and multiply what is left by 3. Now we have n = 1234 30 = 2 (mod 7) and also n = 1234 =4 9+8 0+9 = 12 (mod 13). Show why this works for all positive integers n. What multiplier should be used instead of 3 if the modulus k is 9 or 1 1 if k = 17; if A: = 19? For k = 17 and 19 the procedure is likely to be too long to be of any practical value. Find more satisfactory variations of the method. For example:

=4+9+8+0+9=

1734562 86725 4335 215

10
1734562
2.

62

25

Show

that for

+ 35 + 15 + 10 = 147 E= = 9 the method in the text

and that

47

52

14

(mod

19).

are essentially the same "casting out nines."

3.
all

and the method of problem 1 they amount to the familiar process of

Using the fact that 1001

=7-11-13, and assuming that you can recognize

work out a

multiples of 7 or 11 or 13 having no more than three digits, scheme for testing for divisibility by 7 or 11 or 13 simultaneously. 1 1 1 s 4. Prove (y y* vp*- )' /y'- ip- (mod/? ) if s ^ 2.

<

2.7

Prime Modulus

43

Prove
n-I

f(y

vp*-

-/(#)

- i)aP~

i ~'l

vp*~'

(mod/>

) if

and

This can be used to replace the use of Taylor's expansion at the beginning

of this section.
5.

Apply the method of this

section to solve

ax

(modp
11

),

(a,p)

1.

How are these solutions related to those given by Problem with m replaced by p s ? 4 5 4 6. Solve x + # + 1 ==0 (mod 3 ). 3 3 7. Solve a + x + 57 = (mod 5 ). 2 s # 8. Solve + 5z + 24 (mod 36).
9.

of section 2.3

Solve x z

10. 11.

+ 10z2 + x + 3 = (mod 3 Solve x* + x2 - 4 EE (mod 7 3 2 5 = Solve x* + a (mod 7

).
).

3 3).

2.7

Prime Modulus

We

reduced the problem of solving f(x) = (mod ra) to its last we will not be able to here that It is moduli. stage, congruences with prime some are there find a general method. However, general facts concerning

have

now

the solutions, and

matters.

we

will find that

we

are led to

some new and

interesting

As

before,

we

write/(oj)
=

a Qx n

+ a^^ + =

a n and we assume that

is

a prime and a Q

(mod/>).

Theorem 2.19

If the degree n off(x)

(mod p)

is

to p, then either every integer is

a solution off(x)

(mod p)

greater than or equal or there is a

and such polynomial g(x) having integral coefficients, with leading coefficient 1, that g(x) = (mod p) is of degree less that p and the solutions of g(x)

(modp)
Proof.

are precisely those off(x)

(modp).

- x we obtain f(x) q(x)(x*> x) + r(x) Dividing f(x) by x* where q(x) is a polynomial with integral coefficients and r(x) is either zero or a less than/?. Fermat's theorem polynomial with integral coefficients and degree hence and u ss shows that u* (mod p), f(u) EE r(u] (mod p) for every
integer u. Therefore if r(x)
is zero, or if every coefficient in r(x) is divisible == is a solution of/(z) then (mod/?). The only other every integer by/?, m m b is r(x) = b mx m_^x \-b^m<p, with at least one possibility

^+

44
coefficient

Congruences

k such

not divisible by p. Let b k be the coefficient with largest subscript b k) = 1. Then there is an integer b such that bb k == 1 (mod/?), (/?, = (mod /?) have the same solutions. and clearly r(x) (mod/?) and br(x)
that

The requirements of the theorem

are satisfied if

we

define g(x) as

In the statement of Theorem 2.19, g(x)

thatg(o;)
==

(mod/?) and f(x)

described as having the property (mod/?) have the same solutions. From
is

the proof of the theorem

However,

modulo p
is

see that g(u) bf(u) (mod/?) for every integer u. we do not say that the polynomials g(x) and bf(x) are congruent we will use this last statement to mean that each coefficient in g(x)

we

congruent modulo

to the corresponding coefficient in bf(x).

Theorem 2.20
solutions.

The congruence f(x)

(mod p) of degree n has

at

most n

(mod p). I by induction on the degree of f(x) = and hence the is just a with a ^ (mod p), = has the n If has no solution. 1, congruence exactly one congruence solution by Theorem 2.13. Assuming the truth of the theorem for all con gruences of degree <w, suppose that there are more than n solutions of the = (mod p) of degree n. Let the leading term of f(x) be congruence f(x) u n u n+I be solutions of the congruence, with a x n and let w 1? u z
Proof.

The proof

is

0, the polynomial /(#)

Uj

Uj

(mod/?) for

^j.

We

define g(x)

by the equation

n noting the cancellation of a x on the right. Then either g(x) is identically k n. zero, or it is a polynomial of degree k, We wish to prove that g(x) is either identically zero or is a polynomial

<

having
g(x)

all its coefficients divisible

by p. If

this

were not so the congruence

But g(x) by our induction hypothesis. Now what we have proved about g(x) shows that g(u) = = a (w u^(u w 2) (u integers u and hence that/(w)
, ,

(mod p) would have a degree, say = (mod p) has n solutions u l9 u 2

h,

and we see that h ^ k < n. u n9 and this is impossible

(mod/?) for all u n ) (mod/?) for

all

integers u. In particular,

*o(w+i

Wi)(+i

wj

(t/ n+1

- wj =/(w n+1) s

(mod/?).

But

this contradicts Theorem 1.15, and hence (mod p) has more than n solutions is false.

the assumption that f(x)

n 1 bn Corollary 2.21 If b Q x (mod n solutions then all the coefficients b 3 are divisible by p.

+ b^' +
-

/?)

Aos more than

2.7

Prime Modulus

45

The congruence f(x) = (mod p) of degree n, with leading p = x n and has solutions a 1, only iff(x) is a factor of x if coefficient p x = f(x)q(x) + ps(x) where q(x) ands(x) modulo p, that is, if and only ifx

Theorem 2.22

have integral coefficients, and where either s(x) than n or s(x) is zero.
Proof.

is

a polynomial of degree

less

If/(#)
n.

f(x),
less

we
/?),

find x p

than

x by (mod/?) has n solutions, then n^.p. Dividing x* x has is zero or where r(x) r(x) r(x) degree f(x)q(x) p u For every solution u off(x) = (mod p) we have u

(mod

and hence

r(u)

(mod

p). Therefore if r(x) is

not zero,

it is

than n having n solutions. According to Corollary 2.21 all the coefficients of r(x) are divisible byp, and we can write r(x) ~ ps(x). u x = f(x)q(x) + ps(x) then f(u)q(u) == u* Conversely, if x*

polynomial of degree

less

ps(u)

(mod/?) hasj? of degree p-n with leading coefficient 1 and hence by Theorem 2.20 the congruence q(x) (mod p) has at most p-n solutions, If u is any of the other p-k residues with v k Vt,k t?i, ^p-n. say, modulo p, then (q(u),pj) = 1 and f(u)q(u) == (mod p), and we have
solutions.

==

(mod/?) for every integer u. Therefore f(x)q(x)

But q(x)
,

is

(mod p) has at least with Theorem n solutions. 2.20, shows that This, (p p n solutions. has (mod exactly p) f(x) The restriction a = 1 in this theorem is needed so that we may divide x by /(a) and obtain a polynomial q(x) with integral coefficients. How xp much of a restriction. We can always find an integer a such ever, it is not
f(u)
1.15.

(mod p) by Theorem
n)

k^p
=

Hence f(x)

very

that aa

(mod/?).
==

Then

af(x)
/?),

(aa Q

l)x

n ==

solutions as f(x)
coefficient
1.

(mod

and af(x)

(aa

(mod/?) has the same n has its leading l)x

PROBLEMS
1.

Reduce the following congruences to equivalent congruences of degree

z 11
a:

6:

+ 13 + x 7 + = 2 (mod 7); a 10 + 4a - 3 = (mod 7). (c) 2 3 2. Prove that 2a; + 5ic + 6x + 1 =0 (mod
20
15

(a)
(b)

+x* +5 =0(mod7);
rr

a;

a;

7)

has three solutions by use of

Theorem
3.

2.22.

Prove that #14

12#2

(mod

13) has thirteen solutions

and so

it is

an

identical congruence. 4. Prove that if f(x)

.

x = = (mod x == a lt x = 2 p) has j solutions - ^(a? - a 2) a s (mod p), there is a polynomial q(x) such that /(a?) = (x _ ^ 5-)?(^) (mod /?). Suggestion: begin by showing that there is a (^ ^(z) - ajq^x) (mod/?) and that ^(a) = (mod/?) has solutions suchthat/(tf) s (a? x = dj (mod/?). Then use induction. x = a^ x = 3
, ,
. .

46
5.

Congruences

the degree off(x)

6.

With the assumptions and notation of the previous problem, prove that if then q(x) is a constant and can be taken as the leading is j,
Prove that Fermat's theorem implies that
x v-i

coefficient of /(#).

(p

\)( X

2)

(a?

-p +

1)

(mod;?)
(mod/?).

and
xp
7.

= x(x

\)(x

2)

(x

p +

1)

By comparing
Let

coefficients of

in the previous

problem, give another proof

is

of Wilson's theorem.
8.

m be composite. Prove that Theorem 2.20 is false if "mod/'

replaced

by "mod m."

2.8

Congruences of Degree Two, Prime Modulus

If f(x)

(mod
prime

/?)

is

of degree two, then /(a)

will

ax 2
2

bx

relatively

to p.
is

We

Hence u = if 2au b v and + (mod p) (mod/), where only 4#c (mod/). Furthermore, since (20,/>) = 1, for each solution v there is one, and only one, u modulo p such that 2au + b = v (mod/). Clearly different v modulo/? yield different u modulo p. Thus the problem of solving the congruence of degree two is reduced to that of solving a congruence of the form x 2 = a (mod p). In Chapter 3 we will consider the congruence x 2 = a (mod/) in detail. For the present we will merely obtain some general results concerning the more n general congruence x = a (mod p) and certain related concepts.
difficulties.
is
z;

Then p

a solution of f(x) 2 2 is a solution oft; == b

suppose odd, and 4af(x)

if

p>2
=

since the case/?

+ c, and a is = 2 offers no
b2
.

(lax

b}

4ac

PROBLEM
1.

Reduce the following congruences to the form x2 = a (mod/?): - x + 5 = (mod 7); (a) 4x* + 2x + 1 == (mod 5); (b) 3x* 2 s a; 2x* 7x 10 + + x - 1 EE (mod 13). (c) (mod 11); (</)

2.9

Power Residues
n If x

Definition 2.6

a (mod /)

/z&s

solution, then

is

called

an nth

power

residue

modulo p.

Definition 2.7

Let denote a positive integer and a any integer such that Let h the smallest positive integer such that a n = 1 (mod m). denote (a, m) We say that a belongs to the exponent h modulo m.

2.9

Power Residues
1

47

Since a* (m} ==

prime to

(mod m) by Euler's theorem, we see that every a relatively m. Dividing <f>(m) belongs to some exponent A ^ <f>(rri) modulo
<f>(m)
is

by A we obtain

= qh +

/*,

^r<

A.

But then a r

=
n

a r+**

0* (m)

s= 1

(mod
since

m). Since A

^r<
first

have the

1 the least positive integer such that a Therefore r be cannot see that we A, positive. assertion in the following theorem.

(mod m), and r = 0, and we

then h
\

Theorem 2.23
Furthermore a j

If a belongs a k (mod m)

to

the exponent h
\

modulo
k).

<f>(rri).

if and only ifh

(j

1 , k and since (a, m) in assuming j Proof. There is no loss in generality k k j the Thus 1 a^~ to is a (mod m) the congruence a (mod m). equivalent second assertion in the theorem follows as in the proof of the first assertion.
9

>

Theorem 2.24

If a belongs

to the

exponent h modulo
k

then a k belongs to the

exponent A/ (A, k] modulo m.

Proof.

But A
|

According to Theorem 2.23, kj if and only if {A/(A, k)}

\j.

(a
\

(mod m)

{fc/(A, k)}j

if and only if A kj. and hence if and only if

|

{A/(A, k)}

Therefore the least positive integer./ such that

(dy

(mod m)
is

Definition 2.8

If a belongs to the exponent

<f>(m)

modulo m, then a

called

a primitive root modulo m.

1) primitive roots Theorem 2.25 If p is a prime, then there exist <f>(p e e modulo p. The only integers having primitive roots arep 2p 1 2, and 4, with p an odd prime.
,
, ,

Proof.

Each

integer

a,

l<a<p
,

1,

belongs to

some exponent A
k h

1 to the exponent A, then (a ) modulo p with h\(p 1). If a belongs h~l 2 modulo distinct are a a Therefore, , p. (mod p) for all k 9 and 1, a, h from Theorem 2.20, these A numbers are all the solutions of x == 1 (mod/?).

A modulo By Theorem 2.24 just ^(A) of these numbers belong to the exponent that a to smaller exponents. Also, any integer belongs p. The others belong h to the exponent A modulo p is a solution of x == 1 (mod /?). Therefore for

a 1 there will be either </>(A) or no integers a, 1 each A that divides p the denote Let modulo A the to that a such 1 y(A) p. exponent belongs p Then modulo A the to a that the of number p. exponent belong integers
,

^ ^

=
divide
first

1- But 1 and <(A) for each A that divides p 2i-i V( A ) have we so Theorem to 1 (y(A) A 2.17, 2*10-1 according ( ) ? for all A that <; 0. This implies y(A) and ^(A) <f>(h) </>(A) 0. This proves the and in 1) 1)

=^

1,

particular
is

y(p

<(p

= >

It

part of the theorem. is easily seen that <f>(n)

i

even for n
0, e i

the

are distinct

odd primes,/^

> 2. Let m = > 0, and ^

fc

2/

n?=iA
m)

fi

where
1

1.

If (a,

we

48

have a*

^=

Congruences
1

^2 or/ ^2.

(mod /*) and Then both <j>(p{

ai a 4*<ii )*</3i'i)

is congruent to 1 m. Therefore we do not have a primitive root if k : 2, or if fc = 1 and/ ^ 2. Next we prove that 2 n has no primitive root if n ^ 3. First note that for 2 any odd integer a, say a = 2b + 1, then a = 46(6 + 1) + 1 = 1 + 8c,
,

*^= I (mod mjp$). Suppose that 1 and ) ^(m// ) are even, and therefore 1 1 modulo/ and modulo m// hence modulo
a * (ml
1

<2

(1

+
we

8c)

1
7z

duction

get for

+ I6d, ^ 3,
71

a*

(I

16rf)

32r.

Continuing by in

"2

a*

=
,

+ 2 ng, a*

w-2

n (mod 2 ).

and hence 2 n has no primitive root if n ^ 3. This together with the preceding paragraph establishes that the only positive integers that can possibly have primitive roots are/, 2/, I, 2, and 4, with p an odd prime. Finally we prove that there is a primitive root in each of these But
(/>(2 )

71

-1

>

2W

-2

cases.

Consider

m =p
6

and

let

a be a primitive root modulo p. Let b

+ pt.

Then by

the binomial theorem,

33

-1

=
to
1

a*~ l

+
j

(p

l)a
1

p- 2

pt

(mod/)

= + n^jp with n^ ^ (mod p). Noting + np )v + np (mod p ~ we use induction to see that JP'-^P-D = + w^- w h ^ = n ._^ (mod/- Then = ^ ^ (mod/?). For e ^ 2 let 6 belong to the exponent h modulo p Then h p ~ (p 1) and
and we can choose
j

make b*~ l

that

~l

(1

==

2j

it

).

?z^

hence h =p'd, s I, d\(p 1), e +1 1 + ^s +i/ = 1 (mod p ), and therefore s^.e have b d = b pSd = 1 (mod/, which implies that (/?
since b

^e

which implies b* *-v

8(

s
|

=
e

(mod p

),

1.

We

also

1)

dby Theorem
1.

2.23

=p
Now

a (mod p) and a belongs to the exponent p 1 h = <^(/> e ) and 6 is a primitive root modulo
,

Then we have
is

Notice that b

independent of e.
consider
e
,

e e 2p and let a be a primitive root modulo p Let 6 = h n a or <z + p whichever is odd. Then b = 1 (mod 2) for all A, and b = a n = I 1 1) 7z. This implies that b is a primitive root (mod/) if and only if/"
.

m=

modulo 2/.
Finally,

we

note that 3
1.

is

a primitive root modulo 4, and

is

a primitive

root modulo 2 and

Theorem 2.26 Suppose that m has a primitive root g. Then g j = g k (mod m) j == 1 (mod if and only ifj = k (mod <f>(m)); in particular, g m) if and only if 2 (m) The set a reduced residue <j>(m) \j. g, g forms system modulo m, g* j so that if a is any integer satisfying (a, m) = 1, there is one and only one g
-

in the set

such that

a (mod m).

2.9

Power Residues

49

The first part of the theorem is a special case of Theorem 2.23. It Proof. (m) are follows that g, g 2 , g^ incongruent in pairs modulo ra, and so this set forms a reduced residue system modulo m.
,

The exponent j such that g j = a (mod m) is called the index of a. The index depends on m and g as well as on a. Indices behave very much like logarithms, and they are sometimes useful as an aid to computation as well as being of
theoretical interest.

Theorem 2.27

If p (mod p) has (n,p

is

1)

a prime and (a, p) = 1, then the congruence x n solutions or no solutions according as

or

==

Let b denote (,/?

a "- 1)/&
(

1).

n If x ==

a (mod
)

/?)

has a solution u, then

= u ^-u/ =

tt

<*-i)(/ B

s
(

(mod p ^

a (mod/?) has no solution if a *-l)/b ^ 1 (mod p). (p ~1)/b = 1 Conversely, suppose that a (mod/?). By Theorems 2.25 and 2.26, there is a primitive root g modulo and an exponent j such that g* == a p Thus we have (mod/?).
Therefore x

and
b
\j.

this implies that jr(/>

as a

any v power of g, say g

Now

solution of x n
,

= (mod/? 1) by Theorem 2.26, so that = a (mod/?), if any exist, can also be written n = a modulo/?. Hence the solutions in x if any, ofx
1)/Z?
9

(mod p) correspond to the solutions in y of gyn == g j (mod /?). This con gruence, by Theorem 2.23, has solutions if and only if yn ==/ (mod/? 1) has solutions, which it does by Theorem 2.13 since b Moreover there are \j. 1) solutions by Theorem 2.13, and thus we have (n,p (n,p 1) solutions of x n = a (mod /?).
Corollary 2.28 If p is an odd prime and ~ has two or no solutions according as (p 1)/2
Proof.
(a,

p)
1

1 1

then x*

a (mod

/?)

or

(mod/?).

From

Fermat's theorem

we have
!)

a -i

_i=

(mod

p),

and hence

a< p 1)/2
is

(mod/?).

not useful for numerical calculations unless p is very small. For example, if we ask whether the congruence x 2 = 19 (mod 61) has solutions, Corollary 2.28 leads into a very lengthy calculation. In Chapter 3
better

This result

methods for answering such questions are developed.

50

Congruences

PROBLEMS
Find a primitive root of the prime 3 ; the prime 5 ; the prime 7 ; the prime 1 1 ** the prime 13. ^
1.
;

2.
3.

& does the prime 13 have? 6 belong modulo 7? each of do 4. To what 3, 4, 5, 1, 2, exponents To what exponents do they belong modulo 11?

Find a primitive root of 23.

How many primitive roots

^>
'

5.
if

Let p be an odd prime. Prove that a belongs to the exponent 2 modulo p 1 and only if a s (mod/?). 2 6. If a belongs to the exponent h modulo m, prove that no two of a, a
3
,
,

an odd prime, how many solutions are there to rc 1 = 1 (mod/?); toa^ = 2 (mod/?)? 8. Prove that 3 is a primitive root of 17 by observing that the powers of 3 are
7. If

a h are congruent modulo m.

is

3?

congruent to 3, 9, 10, 13, 5, 15, 11, 16, 14, 8, 7, 4, 12, 2, 6, 1 modulo 17. Then use Theorem 2.27 to decide how many solutions each of the following con
gruences has:
(a)
(c)

z12
z20

= =

16
13

(mod (mod

17) 17)
8

(b)

a 48

(d) aP-

= =

9 9

(mod (mod

17)
17).

3 (mod 17). 3 , 13 Suggestion: 16 9. Using the data of the previous problem, decide which of the congruences x2 == 16 (mod 17), have solutions. x2 = 1, x 2 == 2, x 2 == 3, , = then #n is a 10. Prove that if and 1 and

prime

(a,/?)

(,/?

1)

1,

=
1

a (mod/?) has exactly one solution. 11. Prove that if g is a primitive root
then g*
is

modulo a prime/?, and if

(k,

p
1

1)

=
a

also a primitive root. 12. Prove that if belongs to the exponent 3 modulo a prime/?, then a* s (mod/?), and 1 -1- a belongs to the exponent 6.
13.

+
is

Prove that if a belongs to the exponent h modulo a prime/?, and if h

(mod/?).

even,

then^ 72 = -1

14. Prove that if a belongs to an exponent /z, and b to an exponent k, modulo m, then ab belongs to an exponent which is a divisor of hk. Furthermore, if = 1, then ab belongs to the exponent hk modulo m. Suggestion: if ab (/z, fc)

belongs to the exponent

1

r,
r

then

(ab)

(ab)

nr

(a

h r hr
)

b hr

(mod

m),

so that
15.

hr.

Given that ab = 1 (mod m), and that a belongs to the exponent h modulo m, prove that b belongs to the exponent h. Then prove that if a prime/? > 3, the product of all the primitive roots of/? is congruent to 1 modulo/?. 16. Let a and n > 1 be any integers such that 71 - 1 s I (mod ri) but ax & 1 (mod n) a
1.

for every proper divisor x of n

Prove that n

is

a prime.

2.10

Number Theory from

an Algebraic Viewpoint

51

17. For any prime p and any integer a such that (a,p) 1, say that a is a cubic residue ofp if # 3 = a (mod/?) has at least one solution. Prove that ifp is of the form 3k + 2, then all integers in a reduced residue system modulo p are

cubic residues, whereas if p is of the form 3 + 1, only one-third of the members of a reduced residue system are cubic residues. 18. Prove Wilson's theorem using primitive roots.
r _ 1 be the Letp be an odd prime. Let rl9 r 2 p integers in any order. Prove that at least two of the numbers 1 r l5 2 r 2 are congruent modulo p.

19.

1, 2,
*

,/?

(p

l)/^

2.10

Number Theory from an

Algebraic Viewpoint

In this and the next section

we discuss some of the forms in which the of number elementary concepts theory turn up in algebra. The theory of numbers provides a rich source of examples of the structures of abstract algebra. We shall treat briefly three of these structures: groups, rings, and
fields.

Before giving the technical definition of a group, let us explain some of the language used. Operations like addition and multiplication are called "binary operations" because two elements are added, or multiplied, to
b, is produce a third element. The subtraction of pairs of elements, a likewise a binary operation. So also is exponentiation, a 5 in which the
,

a group consists of a set of elements on those with a elements, such that certain binary operation together which we deal will have with theoretic number hold. The groups properties

element a is raised to the bih power.

Now,

integers or sets of integers as elements,

and the operation will be either addition or multiplication. However, a general group can have elements of any sort and any kind of binary operation, just so long as it satisfies the
conditions that

we

shall

impose

shortly.
,

and we shall begin with a general binary operation denoted by is single-valued. This means that for each that this binary operation presume
set of b has a unique value or is not defined. pair a, b of elements, a or closed elements is said to be "closed" with respect to an operation , b is defined and is an element of the set for "under" the operation, if a

We

every pair of elements a, b of the set. For example, the natural numbers are closed under addition but are not closed under subtraction. 1, 2, 3,

An

element e

is

said to be

an "identity element" of a

set

with respect to the

operation

if

the property

holds for every element a in the set. In case the elements of the set are is ordinary addition, 0, if numbers, then e is the zero element, e

52

Congruences

whereas e is the unity element, e if 1 ing the existence of an identity element 1 "inverse," written cr if the property
, ,

is

ordinary multiplication.

Assum
an

e,

an element a

is

said to have

a
holds. If the elements are

a~ l

a~l

numbers and

write a
is

b for a

b and

is ordinary addition, we usually a for the inverse ar 1 because the additive inverse

number a. On the other hand, if the operation is a b. In this for case we write a b the notation ordinary multiplication,
the negative of the
is

dr 1

the customary one in elementary algebra for the multiplicative inverse. Here, and throughout this section, the word "number" means any sort of

number,

integral, rational, real, or

complex.

Definition 2.9

group

is

set

of elements

a, b, c,

together with a

single-valued binary operation

(1) the set is closed

such that

under the operation;

(2) the associative law holds, namely,

a (b c) (a b) cfor all elements a, b, c in the set a has (3) unique identity element, e; each in G has a unique inverse in G. element (4)

G;

A group G

is

called "abelian" or
in
is

"commutative"
group"
is

pair of elements a, b elements; otherwise it

G.

"finite

an

"infinite

b a for every one with a finite number of group" If a group is finite, the number b

if a

of its elements

is

called the "order" of the group.

Properties 1, 2, 3, and 4 are not the minimum possible postulates for a group. For example, in postulate 4 we could have required merely that each element a have a left inverse, that is an inverse a' such that a' a e, and

then

we could prove

to avoid too lengthy a discussion of to the books on algebra.

the other half of postulate 4 as a consequence. However, group theory, we leave such refinements

set of all integers 0, is a 1, 2, group under addition; in fact an abelian group. But the integers are not a group under multiplication because of the absence of inverses for all elements 1. except Another example of a group is obtained by considering congruences modulo m. In case 777 6, to give a concrete example, we are familiar with such simple congruences as
it is

The

(mod

6),

4 (mod

6).

We

get "the additive group

1, 2, 3, 4, 5,

say 0,

modulo 6" by taking a complete residue system, and replacing congruence modulo 6 by equality:

3+4=1,

4.

2.10

Number Theory from

an Algebraic Viewpoint
is
:

53

The complete addition

table for this system

Of

course, any complete residue system

1, 2, 3, 4, 5, 6,

thus

or 7,

modulo 6 would do just as well; 2, 17, 30, 8, 3, could serve as the elements,

provided
7,

we perform

additions

2, 17, 30, 8, 3, the addition table

the two groups are essentially

modulo 6. If we were to use the system would look quite different. However, the same; we have just renamed the elements:

is now called 30, 1 is 7, and so on. We say that the two groups are "isomorphic," and we do not consider isomorphic groups as being different. Thus we speak of "the" additive group modulo 6, not "an" additive group

modulo

6.

Definition 2.10

with operation and G with operation O, are said to be isomorphic if there is a one-to-one correspondence between the elements ofG and those ofG', such that if a in G corresponds to a' in G'', and b
in

Two groups, G

corresponds to b in G', then a

in

corresponds to a

b' in G'.

Another way of thinking of the additive group modulo 6 is in terms of the two integers a and b into the same residue class modulo 6 if a = b (mod 6), and the result is to separate all integers into six
so-called residue classes. Put

residue classes:

Q
C3

,-18, -12, -6,0, 6,12,18,' ,-17, -11, -5,1, 7,13,19, ,-16, -10, -4,2, 8,14,20, ,-15, -9, -3,3, 9,15,21, ,-14, -8, -2,4, 10,16,22, ,-13, -7, -1,5,11,17,23,-

54
If

Congruences

any element in
that

element in class

we observe

class C2 is added to any element in class C 3 the sum is an C so it is reasonable to write C2 + C3 = C5 Similarly C3 + C4 = Cl9 C5 + C3 = C2 etc., and so we could make
,

up an addition table for these classes. But the addition table so constructed would be simply a repetition of the addition table of the elements 0, 1,2,3, 4, 5 modulo 6. Thus the six classes C C1? C2 C3 C4 C5 form a group under this addition that is isomorphic to the additive group modulo 6. This residue class formulation of the additive group modulo 6 has the advantage that such
,
,

a peculiar equation as 5
.

meaning than in elementary arithmetic) form C5 + C5 = C4

4 (in which the symbols have a different is replaced by the more reasonable

Theorem 2.29

addition modulo m.

Any complete residue system modulo m forms a group under Two complete residue systems modulo m constitute iso

morphic groups under addition, and so we speak of "the" additive group modulo m.
Proof.

Let us begin with the complete residue system

0, 1, 2,

modulo m. This system is closed under addition modulo m, and the associative property of addition is inherited from the corresponding property for all
integers, that is c (mod m). b)

+
is

(b

c)

(a

b)

c implies

(b

c)

(a

The

identity element

additive inverse of

unique. Finally, the 0, and the additive inverse of any other element a is
is 0,
it is

and

a.

These inverses are unique.

,

ro> ?!,

Passing from the system 0, 1, " 9 r m-i> we observe that

,

to

any complete residue system

all
1,

the above observations hold with a

so that

0, 1, replaced by ra a group with new notation.

we have

essentially the

same

PROBLEMS
1.

Which of the following

are groups?

fa)i

the even integers under addition;

(b)'
(c)

he odd integers under addition; the integers under subtraction;

all

(d) the

integers (jQ all rational of the form a\b

(g) the

OX

even integers under multiplication; which are multiples of 7, under addition; numbers under addition (recall that a rational number where a and b are integers, with b ^ 0);

is

one

same set as in (/), but under multiplication; the set as in (/) with the zero element deleted, under multiplication; t*i all rational numbers a\b having b = 1 or b = 2, under addition;

^
(/)

all

rational

numbers alb having

=1,6=2,

or 6

3,

under addition.

2.10
2.

Number Theory from

Let

an Algebraic Viewpoint
1), (1,

55

and

-1), (-1, 1), (-1, -1), a group. 3. Using the complete residue system 7, -2, 17, 30, 8, 3, write out the addition table for the additive group modulo 6. Rewrite this table replacing 7 by 1 b as the one that this table gives the same values for a 30 0, etc.
let (a, b)

have as elements the four pairs (1, = (ac bd). Prove that (c, d)
9

is

by

Verify

in the text.
4.

Prove that the

set

of elements

e, a, b, c,

with the following table for the

binary operation,

a group. Prove that this group is isomorphic to the additive group modulo 4. 5. Prove that the set of elements e, u, v, w, with the following table for the
is

binary operation,

a group. Prove that this group is not isomorphic to the additive group modulo 4, but that it is isomorphic to the group described in Problem 2. 6. Prove that the set of elements 1 2, 3, 4, under the operation of multiplication
is
,

modulo

a group which is isomorphic to the group in Problem 4. 2 z, where 1, +/, 7. Prove that the set of complex numbers +1,
5, is
z

1,

is a group under multiplication and Problem 4.

that

it

is

isomorphic to the
is, if

group in

8.
is

Prove that the isomorphism property is "transitive," that to G3 then G isomorphic to G 2 and if G 2 is isomorphic
,
,

a group

is

isomorphic to

08-

56
9.

Congruences
Prove that the elements 1, 3, 5, 7 under multiplication modulo 8 form a group which is isomorphic to the group in Problem 5. 10. Prove that there are essentially only two groups of order 4, that is that any group of order 4 is isomorphic to one of the groups in Problems 4 and 5.
11.

For any
,

Cw_ 1}

> 1, separate all integers into classes C Q, positive integer = s (mod 777), thus putting integers r and s into the same class if r
,

'

'

2777,

777,

0,

777,

2777,

Q:
etc.

-2777

1, -777

1, 1,

772

1, 2777

1,

Prove that

if any two integers, one from class Ca and one from class C& are added, the sum is always an integer in a unique class, namely, either Ca+6 or Define the sum Ca + C& = Ca+&_ m according Ca+b or Ca + Cb = Ca+6_ m accordingly, and prove that these classes form a
,

2Lsa+b<mora+b^m.
Prove that
this

group under

this addition.
777.

group

is

isomorphic to the additive

group modulo

2.11

Multiplicative Groups, Rings, and Fields

Theorem 2.30
modulo

Let

m>

be a positive integer.

Any reduced
is

is

a group under multiplication modulo m. The group

residue system of order </>(ra).

Any two such groups are isomorphic, and so we speak of "the multiplicative group modulo m."
Proof.
72

Let us consider any reduced residue system r l9 r 2 <f>(m). This set is closed under multiplication modulo m by
,

rn

where
1.8.

Theorem

= (ab)c implies that a(bc) = (ab)c property for integers, because a(bc) (mod m). The reduced residue system contains one element, say rj9 such that
r,

The associative property of multiplication is inherited from the corresponding

(mod m), and

Finally, for each

Theorem
ri, r 2

unique identity element of the group. the xr r5 (mod rn) has a solution i /,-, congruence by 2.13, and this solution is unique within the reduced residue system
this is clearly the

r n Two different reduced residue are con systems modulo element modulo and so we have an element, m, gruent, by isomorphism

,--,

between the two groups.

Notation.

We have been using the symbol for the binary operation of the and we have found that in particular groups group, may represent addition
is

or multiplication or some other operation. In dealing with general groups it convenient to drop the symbol just as the dot representing ordinary multiplication is usually omitted in algebra. We will write ab for a b, abc
,

for

(b

c)

(a

b)

c,

a 2 for a
c)

a,

a z for a
b)
(c

abed can be written for (a

d=

(a

(a

a), etc. Also, d) etc., as can be

2.11

Multiplicative Groups, Rings, and Fields

57

will even use the word seen by applying induction to the associative law. it must be remembered that we do but the for , operation multiplication

We

not mean the ordinary multiplication of arithmetic. In fact we are dealing with general groups so that a is not a number, it is just an abstract element 1 3 2 1 2 of a group. It is convenient to write a for e, a~ for (a" ) a~* for (a" ) , etc. It is not difficult to show that the usual laws of exponents are valid under this
,

definition.
c. ac implies b In any group G, ab If a is any element of a (unique) smallest there is with element then a finite group G e, identity r such that a r e. positive integer

Theorem 2.31

Proof. on the

To

ac part of the theorem is established by multiplying ab l c. a~l (ac), (crtyb left by ar thus ar\ab) ec, b (ir a)c eb of obtained the series elements the second consider by repeated part, prove

The

first
1

= =

multiplication by a,
e, a,

a 2 a z a*,, ,

Since the group is finite, and since the members of this series are elements s of the group, there must occur a repetition of the form a = a* with, say, s s s whence s < t. But this equation can be written in the form a e = a <2*~ 1 -* = e. Thus there is some positive integer, t s, such that a*-8 e and a
,

the smallest positive exponent with this property

is

the value of r in the

theorem.
Definition 2.11

If a

finite

be any group, finite or infinite, and a an element ofG. a is of efor some positive integer s, then a is said to be offinite order. If r e. a r that such order, the order of a is the smallest positive integer

Let

If there
order.

is

no positive integer s such that a

group powers of a

is

said

to

be cyclic if it

e, then a is said to be of infinite contains an element a such that the

a~ z a~*, a-1 a*
,

e, a,

a 2 a*,--,

element a comprise the whole group; such an is called a generator.

is

said to generate the group

and

all the elements of a finite group are of finite or order. Every group, infinite, contains at least the single element e that is of finite order. There are infinite groups consisting entirely of elements

Theorem

2.31

shows that
finite

of

finite order.

If a cyclic
e,
2 az a, a
, ,

group
,

is finite,
,

and has generator

r
is

a, then the
a.

a^1

where

the order of the element

group consists of All other powers

of a are superfluous because they merely repeat these.

58

Congruences

Theorem 2.32

order of the group. If the order of the group

Proof.

The order of an element of a finite group G is a divisor of the n = is denoted by n, then a e.

r. It is

Let the element a have order

readily seen that

(A)

e,a,cP #,"*,cf-*
9

are r distinct elements of G. If these r elements

there
is

some other element, say

ia
,

Z? 2

do not exhaust the group, Then we can prove that

,

OB)
are r distinct elements,

M, b*a\ b&\
different

b z a r~i
elements of A. For in the
first

all

from the

place b 2a s
If

if

b^a

Z> 2 0*,

=
G

a*,
is

then

Z? 2

a* by Theorem 2.31. And on the other hand, then a s * a 1 3 so that b z would be among the powers of a.
,

if

6 3 which gives rise to r

not exhausted by the sets A and B, then there new elements

is

another element

all different

from the elements in

A
,

and
,

B by
9

a similar argument. This

since

process of obtaining new elements Z> 2 3 So if the last batch of new elements is, say
b k b ka, b ka*,
,

must terminate

G is finite.

3
>

'

'

>

*" 1
,

then the order of the group G is fcr, and the first part of the theorem is kr and a r = e by proved. To prove the second part, we observe that n

Theorem

2.31,

whence a

e.

It can be noted that Theorem 2.32 implies the theorems of Euler, where the set of integers relatively prime to the modulus the group. In making this implication, the reader will see the

Fermat and

m is taken as

necessity of the and notation of into that of number "translating" language group theory In the same we note that the of Definition theory. way 2.7, that language

"a belongs to the exponent h modulo w,"

is

translated into group theoretic

language as "the element a of the multiplicative group modulo A." Also the "primitive root modulo m" of Definition 2.8
"generator" of the multiplicative group modulo
Definition 2.12
operations,

m
is

has order
called a

m in group theory.
,

ring

is

set

and o, such

that

it is

of at least two elements with two binary a commutative group under is closed

under o, and such that

is

associative
is

and

distributive with respect to

The

identity element with respect to

called the "zero" of the ring. If all the

elements of a ring, other than the zero, then it is called a field.

form a commutative group under O,

2.11

Multiplicative Groups, Rings, and Fields

59

It is

b for a

customary to call b, ab for a

(ai)c, a(b
c,

a(bc)

elements a, h,

multiplication and to write for a ring are then ab ba ca. In general, the ac, (b c)a c) are not numbers, and the operations of addition and

addition and

b.

The conditions on O

multiplication are not the ordinary ones of arithmetic. However, the only rings and fields that will be considered here will have numbers for elements,

and the operations will either be ordinary addition and multiplication, or addition and multiplication modulo m.
Theorem 2.33

Zm of elements 0, 1, 2, m is modulo a m, ring for any integer multiplication defined is afield if and only ifm is a prime.
The
set
,

1,

with addition
1.

and

m>

Such a ring

Proof.

have already seen in Theorem 2.29 that any complete residue system modulo m is a group under addition modulo m. This group is commutative, and the associative and distributive properties of multiplication

We

modulo

are inherited

from the corresponding properties for ordinary

a ring.

multiplication.

Therefore Zm is

Next, by Theorem 2.30 any reduced residue system modulo m is a group under multiplication modulo m. If m is a prime/?, the reduced residue system
of Zp
is
is 1,

2,

,/?

1,

that

is, all

the elements of Z^ other than 0. Since

m
a

is not a the zero of the ring, Z^ is a field. On the other hand if prime, then m. Then the elements of m other than b a is of the form ab with 1

<

<

do not form a group under multiplication modulo m because there is no inverse for the element a, no solution of ax = 1 (mod m). Thus Z m is not
field.

y For questions can be settled very readily by using the fields 3 for that any prime p example consider the following problem: prove

Some

thesum

ill
2

>

1
,

1 if

^2

(p-1)

'

number a\b has the property that p a. In the field 2 the above sum isy" 2 or x 2 where x is the least positive term in I// Zp the == 1 such that xj (mod/?). Hence in Z P the problem can be put in the integer 1~ 2 + 2~ 2 4the sum that + (? 1)~ 2 is the zero element form, prove 1 are of the field. But the inverses of 1, 2, 3, ,/? just the same elements we can write so in some order, again
written as a rational
|

I- 2

2- 2

+
there

+
is

(/?-

I)-

22

+
But

(p

I)

For

this final

sum

a well-known formula for the

l)/6.

sum of the

squares

of the natural numbers giving p(p l)(2p because of the factor/?, except in the cases/?

this is zero in

Z 99

2 and/?

where division

by 6

is

meaningless.

60

Congruences

PROBLEMS
1. Prove that the multiplicative group modulo 9 is isomorphic to the additive group modulo 6. 2. Prove that the additive group modulo in is cyclic with 1 as generator. Prove that any one of <f>(m) elements could serve as generator. 3. Prove that any two cyclic groups of order m are isomorphic. 4. Prove that the group of all integers under addition is an infinite cyclic group. k = 5. If a is an element of order r of a e if and only if group G, prove that a

r\k.

modulo

m such that the multiplicative group not cyclic ? 7. subgroup S of a group G is a subset of elements of G which form a group under the same binary operation. If G is finite, prove that the order of a subgroup S is a divisor of the order of G. 8. Prove Theorem 2.32, for the case in which the group is commutative, in a
6.

What

is

the smallest positive integer

is

manner analogous 9. Prove Theorem

10. Let

to the proof of Theorem 2.8. by the method used in the proof of consist of all possible sequences (a l9 a 2 , # 3 ,
2.8
,

Theorem
)
-

2.32.

or -1. Let (a l9 a z aB
that

(b l9 b z ,

Z> 3 ,

= (a^,

with each # 2 6 2 as b z ).
,
,

'=

Show

is

an

infinite

group

all

of whose elements are of

<?,/and
let

finite order.

11. Let

G consist of a,

b, c, d,

@ be defined by the following table.

Show
12.

that

is

a noncommutative group.
is

Prove that the multiplicative group modulo/?

cyclic ifp is a prime.

13. Exhibit the addition

and multiplication

tables for the elements of the field

of residues modulo 7. 14. Prove that the set of all integers under ordinary addition and multiplication is a ring but not a field.

2.11

Multiplicative Groups, Rings, and Fields

15.

61

Prove that the

is

set of all

even integers under ordinary addition and mul

6, 9 is

a ring. tiplication 16. Prove that the set 0, 3,

a ring under addition and multiplication

modulo
17.

12.

Prove that

in

any

field

aO

Qa
l)a,

elements

1 < a < m. Prove that the set of with addition and multiplication modulo 777, forms a ring. Under what circumstances is it a field? 19. Prove that the set of all rational numbers forms a field. 20. Prove that the set of all rational functions /(#)/(>), where f(x) are polynomials with integral coefficients, and^O) ^ 0, forms a field.

18. Let a be a divisor of

0, a, 2a, 3a,

m, say
,

m = aq with

for every element a.

(q

21. If x, y,

xy

z, are any numbers, real or complex, the cancellation law states that z if x 7* 0. There is a "weak" cancellation law that states xz implies y that x*y = x2 z implies xy = xz whether x is zero or not. Consider the set of all > \. Show that the integers modolo 777 with multiplication modulo 777,

a prime, and that the "weak" law holds if and only if 777 is a square-free, that is 777 is a product of distinct primes. 22. Consider the system of all integers modulo 777 under multiplication modulo
cancellation law holds if
if
777

and only

is

777, 777

>

1.

Show

that:

(a) (b)
(c)

It is

not a group.

It is associative.

It is
is

commutative.
closed.

(d) It
(e) It

has a unique unity an element u such that ux == x for all x. an element z such that zx = z for all x. (f) It has a unique zero 23. For 777 = 30 find all idempotent elements (x such that x2 = x). Also find the elements w such that wx = 2, the zero element, for some x ^ z. 24. An integral domain is a ring with the following additional properties: (i) there is a unique identity element with respect to multiplication; (ii) mul = ac and a 7* 0, then b = c. Prove that tiplication is commutative; (iii) if ab

any
(b)
(c)

field is

an integral domain. Which of the following are integral domains?

of
all

(a) the set

integers ;
;

the set

Zm of Theorem 2.33
all

the set F[x] of

polynomials in a variable or indeterminate x, with

coefficients in a field F.

25. Let

numbers

set of all the divisors of m. For as a O b = (#, b), two operations O and are associative and a b = [a, b] g.c.d. and l.c.m. Prove that O and = (a O b) (a O c) commutative. Prove the distributive law a O (b c) Show that a O a = a and its dual a c). (b O c) = (a b) O (a a a = a. Also prove 1 O a = 1 and 1 a, so that 1 behaves like an ordinary as a b if a O a = m. Define a relation zero, and 777 O a = a, and 777 is transitive, and that a b if and only if a b = a, Prove a a, that
777

be a positive integer and consider the

set

in this
9

define

to each divisor a there

not divisible by any square other than 1 then corresponding = 1, a a = 777. (These is a divisor a such that a O a of Boolean m are algebras.) algebras with square-free examples

Prove that if 777

is

62
26. Prove that for any prime p
1
t

Congruences

>
1

2 the

sum
1
i

"*"

1^

__ 3

___^______

(p

'

I)

if

written as a rational

Interpret the

sum

in Z^,
js

number a/6, has the property and use the result

23

that

a.
\

Suggestion:

_j_

= n*(n +

2 l) /4.

NOTES ON CHAPTER
In 2.1

== a (mod m), (ii) a = 6 (mod m) if and (i) a only if b (mod m) and 6 s c (mod m) imply a = c (mod /w). Thus the congruence relation has the (i) reflexive property, (ii) the symmetric and (iii) the transitive property, and so the congruence relation is a

was noted (mod 7?z), and (iii)

it

that

property,
If (a,

so-called equivalence relation.

m)

the congruence ax
1

m) or a? = ba~ where a" is do not use the symbol a"1 in this way, reserving it to mean 1 1 a. A generalization by Gauss of Fermat's Theorem 2.7 can be found in the Special first part of Theorem Topics, page 262 ff. Also on page 263 is given a proof of the
l

l b (mod m) could be solved as x == ba~ (mod an integer satisfying aa~* = 1 (mod 777). However, we

2.25 by a group theoretic approach. For interesting discussions of magic squares see Chapter 4 of the book by Harold M. Stark, listed in the General References on page 270, and the paper by D. N. Lehmer, "On the congruences connected with certain magic squares," Trans.

Amer. Math. Soc., 31, 529-551 (1929).

3
Quadratic Reciprocity

3.1

Quadratic Residues

Definition 3.1

For alia such that (a, m) = 1, a is called a quadratic residue modulo m if the congruence x 2 = a (mod m) has a solution. If it has no solution, then a is called a quadratic nonresidue modulo m.
a quadratic residue or nonresidue modulo according as consider as distinct residues or nonresidues not, only those that are distinct modulo m. The quadratic residues modulo 5 are 1 and 4, whereas
Since a or
is is

+ m is
we

2 and 3 are the nonresidues.

Definition 3.2
f

If p denotes an odd prime

1

and (a> /?)

=
1

!,

the

Legendre symbol
a quadratic non-

a\
is
|

defined to be

if a is a quadratic residue,

if

is

residue

modulo p.
3.1

Theorem
prime
(a)

to p.

Let p be an odd prime and Then

let

a and b denote integers relatively

=a<^"

(modp),

(c)

b (mod/?) implies that

''
=
J
j

\(0-D/2

63

64

Quadratic Reciprocity
If

Remark.
x2

=
f-J

then a (3)

~ 1)

(mod/?) and Corollary 2.28 shows that

a (mod /?) has exactly two solutions; alternatively, this conclusion can be drawn from Theorem 2.20, because if x 2 = a (mod/?) has one solution XQ then x
a distinct solution because/?
is

is

odd.

Proof.

parts are

Part (a) of the theorem follows from Corollary 2.28. The remaining all simple consequences of part (a). Part (a) can also be proved
:

without the use of Corollary 2.28 as follows If

has a solution, say x
(mod/?).

- =
J

then x 2

a (mod/?)
1

Then, by Theorem

2.7,

a (p

~ i}/2

x^~~

= =

(~\

On

the other hand,

if

==

(-)

1,

then x 2

a (mod

/?)

has no solution,

and we proceed
1

as in the

proof of Theorem 2.10. To each j satisfying

^7 =P ^ ^p
/

we

1.

We

is j.

Since x 2

associate the unique integer i such that.//" a (mod /?), see that i is and that the associate of i impossible

a (mod

/?)

itself.
/,

Thus the

integers 1,2,
1)! ==

and.//

has no solution, no integer j is associated with 1, can be paired off,/ and its associate ,/?
l)/2 pairs. Multiplying all these pairs (/? a *~l)/2 (mod/?). Using Theorem 2.10 we obtain
(

# (mod/?). There are

get
(/?

together

we

s _i =
Theorem
Consider
3.2
the

(mod/?).

Lemma

of Gauss. Let p be an odd prime and

-

let (a, /?)

!.

, {(/? integers a, 2a, 3a, l)/2}a afld rte/r least nonnegative residues modulo p. Ifn denotes the number of these residues that exceed

Proof. ^i 3 s 29

Let r x
m

r2 ,

rn

denote the residues that exceed

/?/2,

and
ri

let

sk denote the
is

and none
/?/2,
is
/

=
Sj

remaining residues. The r i and s i are zero. Furthermore n + k = l)/2. Now (p

all distinct,

< /?
p
cr,

<
ri

1, 2,

,, and
r

the numbers

an

(P

for if/?

=^
0,

then r z
l)/2,
/>

/?

pa,

!)/2 > 1

this implies a(/> r l9 r 2 ,p

^ (p + a) =
-

and/? a =
-

^-

aa (mod/?). Since (a,/?) = 1 (mod /?), which is impossible. Thus

pa
sk are all distinct, are all at least
(/?
1

= =

are distinct. Also no

aa, for

some

p,

,p
1, 2,

r n , s l9 s 2 ,

and

less

than/7/2, and they are n

,

are just the integers

+k= - l)/2 (p

l)/2 in

number. That

is,

they

in

some

order. Multiplying

them

3.1

Quadratic Residues

65

together

we have
(P

ri)(P

ra)

(p

r Js^a

sk

and then

=
r n sis 2
Sfc

(mod

p),

=
a

^
2

(mod p),
(mod p).
w (2)-1)/2 (-l) a

3a

^
2, 3,

=
,

2^
(2\

We

can cancel the factors

(p

l)/2 to obtain

(mod/?) which gives us (-1)"

Definition 3.3

fl

<-i)/a

(modp) by Theorem

3.1a.

For real x, the symbol

[x]

denotes the greatest integer less than

or equal to x.

For example,

[15/2]
is

7,

[-15/2]

-8, [-15]
2p)

-15.

Theorem 3.3

Ifp

an odd prime and

t

(a,

1,

then

=(-!)<

where

^
j=I

also

~
\p/

(-1)

1)/8
.

\pJ

Proof.

and

Si

use the same notation as in the proof of Theorem 3.2. The rz are just the least positive remainders obtained on dividing the integers
y"

We

ja by/?, to be q

1, 2,

(p

l)/2.

[jalp].

Then

for (a,p)

The quotient
1,

in this division
is

is

easily seen

whether a

odd or even, we have

5=1

5=1

5=1
n

and

^w
5=1

^w
5=1

(p

-r +
.)

7c

^Lf
3=1

Sf

np

_ y Tf + Z-<
5=1

Z-<
5=1

and hence by subtraction,

66

Quadratic Reciprocity

I'so

we have
2

F
(
1

-(mod2).
*

If

is

odd, this implies n

l)/8

(p*

follows by

3.2, and the first part of Theorem 3.3 are of considerable importance in theoretical considerations, they are too cumbersome to use for calculations unless p is very small. However, the

(mod 2) since Theorem 3.2. Although Theorem 3.1 (a), Theorem

= 2 ~ moc 3=1 = for ^ 7 ^ [2//p]

2). If

it

implies

(p

l)/2.

Our theorem now

other parts of Theorems 3.1 and 3.3 are useful in numerical cases. The ~ second part of Theorem 3.3 involves ( 1) (3)2 1)/8 and this can be easily
,

is reduced modulo computed ifp 2 a and (-l)<*> -i)/8 = (_i)<3 -i)/8

B

8.

For example ifp

59 then/?
that the
1

(mod

8)

Finally,

we point out

problem of

numerical evaluation of
the next section.

fa\
I

apart from the cases a

2, is treated in

PROBLEMS
1.

Find

[3/2],

[-3/2],
is

M,

[-7],

M for

rg

<

1.

2.
3.

With reference to the notation of Theorem

Prove that 3
a quadratic residue of 13,
I

= [b/a]. prove that q but a quadratic nonresidue of

1.2

7.

4.

Find the values of I-

in each of the 12 cases,

1, 2,

~2,

and

/>

!!, 13, 17.

,

Prove that the quadratic residues of 1 1 are 1 3, 4, 5, 9, and list all solutions of each of the ten congruences x 2 = a (mod 11) and # 2 == a (mod II 2 ) where a = 1,3, 4, 5, 9.
5.
6. List the
7.

quadratic residues of each of the primes 7, 13, 17, 29, 37.

Which of the following congruences have solutions ? = 2 (mod 61) (a) x* (b) x* =2 (mod 59) _ _ 2 (moci 6 _ _ 2 moci ( C) #2 i) y) X2 ( 59) = = 2 (mod 118) x* 2 (mod 122) (e) (/) & a 2 = -2 (mod 118). () z2 = -2 (mod 122) (/z)
8.

How many ?

How many solutions are there to each of the congruences: = -1 (mod 61) 0) ^ = -1 (mod 59) 2 = = -1 (mod 3599) ^ C -1 (mod 365) ( ) (</) ^ 2 = = x* a -1 -1 (mod 244)? (e) (mod 122) (/)
(a) x*

3.2

Quadratic Reciprocity
9. Let/?

67

be a prime, and let (a,p) = (b,p) = 1. Prove that if a? 2 = a (mod /?) and x 2 = 6 (mod/?) are not solvable, then #2 = ab (mod/?) is solvable. 10. Prove that if p is an odd prime then x* == 2 (mod/?) has solutions if and
only
11.
if/?

or 7

(mod

8).

Denote quadratic residues by r, nonresidues by /?. Prove that r^ and /z^ are residues and that rn is a nonresidue for a prime p. Show that there are l)/2 quadratic residues and (/? l)/2 nonresidues for an odd prime /?. (p 12. Let g be a primitive root of an odd prime /?. Prove that the quadratic p ~ l and the nonresidues residues modulo/? are congruent to^ 2 ,^ 4 ,^ 6 ,g ~2 3 5
-

2> are congruent to g, ^ ^ g g\ 13. Prove that if r is a quadratic residue modulo > 2, then r^ (m)fz == 1 (mod m). Suggestion: use the fact that there is some integer a such that
-

<

(mod
is

w).
if

14.

Show
/3

that

is

a quadratic residue modulo m, and

then

also a quadratic residue.

is

#/3 = 1 (mod m), Then prove that the product of the quadratic

residues modulo/?

congruent to
1.

+1

or

according as the prime/?

3,

is

of the

form4& +
15.

3 or

4k

Prove that

if/? is

of quadratic residues less than

a prime having the form 4k + 3 5 1 /?/2, then

and

if

772

is

the

number

(/?-

2)

= (-1)+*+*
I

(mod/?). 16. Prove that the quadratic residues modulo p are congruent to
2
,

(mod/?), and 2

(/?is

1) E=

(-!)+*

2
,

22

2
,

{(/?

l)/2}

where/?

an odd prime. Hence prove that

if/?

>

3, the

sum

of the quadratic residues

17.

is

divisible

by p.
16 (mod/?)
if
is

For

all

B primes/? prove that X

solvable. Suggestion: use

Theorem
18. Let/?
/?
|

2.27.

be an odd prime. Prove that

1)

there

is

an integer x such that

(x*

p p

(a (z

+ + +

then then then then

2)
2)
1)

2
4

p Show
|

(#

p = = /? = /? = /?

(mod
or 7
or 3

4);
8); 8);

(mod (mod
8).

(mod

that there are infinitely many primes of each of the forms 872 4- 1, + 5, 8/z 7. Suggestion: use Theorem 2.27 for the case/? (x* 1). 3, 8 19. Let/? be an odd prime. Prove that every primitive root of/? is a quadratic nonresidue. Prove that every quadratic nonresidue is a primitive root if and
8

only
if/?

if/?

is

of the form 22
is

"

where n

is

a non-negative integer,

i.e., if

and only

=3

or/?

a Fermat number.

3.2

Quadratic Reciprocity
The Gaussian reciprocity
ll\ (2)
/

Theorem 3.4
primes, then

law.

If

p and

q are distinct odd

l\{(

68
Proof.

Quadratic Reciprocity

Let
1

S be

(p

the set of
(q

all

l)/2,

^y^ <

pairs of integers (x y) satisfying

9

<;

1)12.

The

set

has (p

l)(q

l)/4

member!

qx
(P

Separate this set into two mutually exclusive subsets S^ and S2 according as py orqx py. Note that there are no pairs (x y) in S such that qx py. The set S l can be described as the set of all x pairs (x y) such that 1

>

<

l)/2

^ y < qx/p.

2a*
(q

1)/2

[<F/p].
l)/2,

\^x

we have

then seen to be Similarly ^2 consists of the pairs (x y) such that I < py/q, and the number of pairs in S2 is 2iii1)/2 Thus
pairs in
S' 1

The number of

is

^y ^

[pylq]-

and hence

by Theorem

3.3.

This theorem, together with Theorem

3.3,

3.1

and the second part of Theorem For example, we have

makes the computation of |- J

fairly simple.

Ur)

ur/taJww'

A shorter way is

1- This computation demonstrates a number of different |-gj-j sorts of steps; it was chosen for this purpose and is not the shortest possible.

Hence

/-42\

,
61

/19\

UJ

_ ~

/61\

UJ

=L

\19J

3.2

Quadratic Reciprocity

42\

by use of Theorem 3.2 or the

part of Theorem 3.3, but the computation would be considerably longer. is another kind of problem that is of some importance. As an example, let us find all odd primes p such that 3 is a quadratic residue modulo p. We have
first

There

_
(1)

()

M
\3/
if
\
'

==

(mod

3),

and
if
if

p
p

(mod

4)
4).

== 3 ==
1

(mod

Thus
|-J

=
(mod

if and

only ifp

==

(mod 3),/?
(mod

(mod 4),

or/?

2 (mod

3),

=
/?

4); that is/?

or 11

12).

Results of this type are sometimes useful if one is trying to determine whether or not a certain number is a prime. Consider the number 9997. We

might notice that 9997

100 2

3, so that 3

100 2

(mod

/?)

if

9997.
\

That is,

either/?

= 3 or
1

(- )

and hence/?
If

=
all

1.

Since 3

> 9997 and p

and

2 we have (-] \pj

95,

=1,

or 11

(mod

12).

We need test only/?

,

such that/? <; \/9997.

,

we

list
1

the

numbers

1, 13,

25,

97,

11, 23, 35,

we can

delete

and
It

be tested.

composite numbers, and we find just eleven primes that must will be found that 13 divides 9997 and that 9997 = 13 769. Is
\ |

769 a prime? Ifp 769 then/? 9997 and it will be in our list. The only/? in our list such that/? <| V769 are 13, 11, and 23. None of these divides 769, and hence 769 is a prime.
Just as we determined which primes have 3 as a quadratic residue, so for any odd prime p we can analyze which primes have /? as a residue. This is done in effect in the following result.

Theorem 3.5 Let p be an odd prime. For any odd prime q > p let r be deter mined as follows. First ifp is of the form 4n + 1, define r as the least positive remainder when q is divided by p; thus q = kp + r, < r < /?. Next ifp is of
the

form 4n

3, there is
1

a unique r defined by the relations q

4kp

r,

<r<

4p, r EE

(mod

4).

Then

in

both cases

- =
1

W/

i^\ \P)

70

Quadratic Reciprocity

Proof.
{

Ifp

4n

+
4n

1,

by Theorems
3,

3. 4

and 3.1c we

see that

W/

= i\ =
\P)

r\
.

In case/?

+
1

we

first

prove that

r exists to satisfy
is

the conditions
4/? ?

stated. Let r

be the
If r
==

least positive

remainder when q

<
r
.

rQ

<

4/?.

(mod 4), take r

is

if r

(mod 4)

divided by take r

so

4p

1
.

The uniqueness of r
If q

readily established.

4kp

+
r

r,

then #

=r=
r

(mod
(mod

4)

and again

If

4kp

then q

- = - = W/ \P1 \P]
1

4)

and by Theorems

3.4, 3.1c,

and

3.ld

we have

For example, suppose we want to determine

as a quadratic residue.

all

<r<

44 and

A complete set of quadratic residues r of

(mod
4) is 1, 5, 9, 25, 37.

odd primes q

that have 11
1 1

Hence by Theorem

satisfying 3.5 the

odd primes q having 1 1 as a quadratic residue are precisely those primes of the form 44k r where r 1, 5, 9, 25, or 37.

PROBLEMS
1.

In the example preceding

Theorem

3.5,

why was

it

unnecessary to

test

any

primes/?
2.

>

\/9997?
,

Prove that ifp and q are distinct primes of the form 4k + 3 and if x* =p 2 solutions, then # = q (mod/?) has two solutions. 3. Prove that if a prime/? is a quadratic residue of an odd prime q, and/? is of the form 4fc + 1, then q is a quadratic residue of p. 4. Which of the following congruences are solvable ?

(mod q) has no

(a) x? 55
(c) #2

(mod

227)

(6)

z2

(mod 2 29)
229)
1009)

(e)

z2

s _ 5 (mod 227) = 7 (mod 1009)

(d) x*
2 (/) z EE

-5 (mod -7 (mod

(Note that 227, 229, and 1009 are primes.)

5.

Find the values of

in the nine cases obtained

from

all

combinations of

p = 7,
6.

7.

11, 13 and q 227, 229, 1009. Decide whether a 2 = 150 (mod 1009) is solvable or not. Find all primes/? such that x* = 13 (mod/?) has a solution.

\q/

8.

Find

all

primes/? such that

I J

=1.

3.3

The
9.

Jacob!

Symbol
primes p such that
I I

71

Find

all

2 a quadratic residue? is a quadratic nonresidue of each of the a (mod pq) solvable ?

10.
11. If

Of which primes
a
is

odd primes/? and

q, is

x2

==

12. In the

Let 0, A, B,

proof of Theorem 3.4 consider the pairs (#, y) as points in a plane. C denote the points (0, 0), (/?/2, 0)(/7/2, ^/2), (0, q/2), respectively,
lines

and draw the

OA, OB, OC, AB, and EC. Repeat

the proof of

Theorem

3.4 using geometrical language pairs of points, etc. 13. Prove that there are infinitely many primes of each of the

forms 3n
I
I

+
1.

and 3n

1.

Suggestion:

first

determine the primes/? such that

3.3

The Jacob! Symbol

Let (P,

Definition 3.4
the qi are

1,

Q >

0,

odd, so that

Q =

q:q 2

qs where
/p\
I I

odd primes, not

necessarily distinct.

Then

the Jacobi

symbol

is

defined by

Q)

(L\ (Q)
where
If

P ~ -TT(

/p\
I

is
J

the

Legendre symbol.

an odd prime, the Jacobi symbol and Legendre symbol are in distinguishable. However, this can cause no confusion since their values are

is

the

same

in this case. Clearly

=
[ J

but

it is

not true that

j~~

=
J

\y/
implies that

\*/

is

a quadratic residue modulo Q. For example, j~ J

but

x2

2 (mod
if (a,
/

9)

has no solution.

A number a
is

is

a quadratic residue modulo

all

only

0=1
=
J

and a
1,

is

a quadratic residue modulo

not a quadratic residue.

primes

that

divide Q. If

then a

Theorem
QQ')

3.6
1.

Suppose that Then

and

are odd and positive and that (PP' y

72

-"
/P'

Quadratic Reciprocity

0)P' = P (modg)
Proof.
First (a)
is

imp/to

^
/p\
I

obvious from the definition of

and
j

(&) follows

from

the definition and

also does (d).

Theorem
(e)

3.1i.

Then

(c)

follows
'

To prove

we

write

Q = #i#

'

from (b) and (a) and so ^ P ( m od ^) ? Tnen P'

(e)

so that (
3.4.

w=w
\

/~\ by Theorem

3.1c,

and then we have

from Definition

Theorem

3.7

J/ g

is

odd and

Q >

0,

\Q7
".

\ft

We have

(1\ Q
If a

-| ) /

= =1 n
I

i\

;=i V q* I

|-^| (

*(-1)""-"" = (-D n (-D

(

i)/2

II

(.-!)!

^=1

and b are odd, then

1 ab- 1 _

la /g I 2

-1 1

+
,

6 *LI=-1\

-1\ = _
2
)

(a ^

~ ~ - ^^ l)(fc
2

1) *>

(mod2)

and hence

22
Applying
this repeatedly

(mo d2).

we

obtain

(3-1)

n^= (-I)".
a and b are odd, then
\
7

(mod

2),

and thus fcl\

Similarly, if

= (lzLl)(^=J) a
8

mod

8),

3.3

The

Jacobi Symbol

73

so

we have
a
2

-!

+
,

b*-l - =

aV-1

(mod 2),

..

and hence,
s

(-)
Theorem 3.8

=n

(-)

= (-i>^

(ff '

~1)/8

(-i)
if(P,

(Q2 -i)/8 .

IfP and Q
_\

are odd

and positive and

/

1,

then

l^\ (P\ Q) (P)

f.

fn\

~ ( ~ 1\{(P }
we^
as

1)/2}{(Q

1)/2>

Writing

P=

TIi=iPi as

Q =

IT5=i

?5?

we have
W-D/2)

where we have used Theorem

PJ

3.4.

But
1

gj

p<

and

as in (3.1) in the

proof of Theorem

3.7.

Therefore

we have

/P\
\e/

= /\/
\P/

j\{(P-l)/2}{(Q-l)/2>

which proves the theorem. The theorem we have just proved shows that the Jacobi symbol obeys the law of reciprocity. It is worthwhile to consider what has been done. In this The definition of the chapter we have been interested in quadratic residues. is a natural one to make. We then proved the useful and Legendre symbol for this symbol. The Jacobi symbol is an ex celebrated law of
reciprocity

tension of the Legendre symbol, defining

for composite Q. However,

74

Quadratic Reciprocity
to be

for quadratic

P and 1 for nonresidues modulo Q. Had this been done, there would have been no reciprocity law (P = 5, Q = 9 is an example). What we have done is this we have dropped the connection with quadratic residues
residues
:

in favor of the

law of reciprocity. This does not mean that the Jacobi symbol cannot be used in computations like those in Section 3.2. In fact, the Jacobi symbol plays an important role in such calculations. In Section 3.2 we used
to
(
j

the reciprocity law to invert the symbol

/
)

but

we could do

it

only

if q

was a prime. In order

to

compute

we had
J
|

to factor a

and consider a

product of Legendre symbols.

not need to factor a
if it is

Now

however, using Jacobi symbols we do

positive.

odd and

We

compute

as a Jacobi
is

\PI

symbol and then know the quadratic character of a modulo p ifp For example:
/105\ \3177

a prime.

/317\

/_2\
\1057

U057

and hence 105

is

a quadratic residue modulo the prime number 317.

PROBLEMS

Which of the following congruences are solvable: x 2 = 10 (mod 127); (>) x* = 73 (mod 173); 2 (c) a s 137 (mod 401)? 3. Which of the following congruences are solvable: = 11 (mod 61); (a) x* (b) & = 42 (mod 97); z = - 31 =0 (mod 103)? x -43 (mod 79); (c) (d) x* 4. Show that if and are odd p q primes one of which is of
2.
(a)

the form

4k

1,

*5.

g - GJ
3>

/
1

/\
I

Prove that

0, /?

an odd prime.

6. Ifp is an odd 1, prove that ax + bx + c == prime and (a,p) (mod/?) has two, one, or no solutions according as b 2 4ac is a quadratic residue, is congruent to zero, or is a quadratic nonresidue modulo p.
z

3.3

The Jacobi Symbol

7.

75
to prove that
if/?
is

Use Wilson's theorem

1- 2- 3

a prime of the form 4n

3, then

^-^

s(-l)* (mod /?)

the factors

where
8.

TH is the

number of quadratic nonresidues among

on the

left side.

(a)

Let

p be an odd prime
)

a (mod
(b)

with a quadratic residue has exactly two solutions, by writing x = x

a.

solution of x 2

x2
9.

a (mod/?). Generalize by the use of mathematical induction, and establish that a (modp*) has exactly two solutions.

+ py>

Prove that x 2 = where x is a

Let the prime divisors of the odd integer

be

/?i,/? 2 ,

p n and
,

let

(a,

m)

Prove that x 2
,72.

a\

a (mod m) has a solution

if

and only

if

=1,

for/
10.

\Pi/

1,2,

For which primes p do there

(y 9 p)

1,

such that x 2

exist integers

x and y with
and y with

(a?,/?)

= =
=

1,

(mod;?)?
integers x
(#,/?)

11.

(2/ 3 /7)

12.

into

an odd prime, be divided ,/? /?, /? two nonempty sets 6^ and S2 so that the product of two elements in the same set is in Sl9 whereas the product of an element of S1 and an element of S2
.

2 2 (y 9 n ) = 1, such that a? + y 13. Let the integers 1, 2,

For which prime powers/? do there exist = 1, such that x 2 + y2 = (mod/? a)? For which positive integers n do there exist

1,

integers x

andy with

(#,

77)

1,

(mod )? 1 modulo

is in S2 Prove that S1 consists of the quadratic residues, S2 of the nonresidues, modulo/?. Suggestion: use ^ primitive root modulo/?. a 14. Let k be odd. Prove: if a ^ n, then x 2 = 2 k (mod 2n) has at least one solution. If a < n, then the congruence has a solution if and only if a is even ~ and x 2 = k (mod 2n a) has a solution. 15. Let k be odd. Prove that x 2 = k (mod 2) has exactly one solution. Further

2 more, x

there are
16.

= k (mod 22) is solvable if and only if k = 1 (mod 4), in which case two solutions. Let k be odd, and let n ^3. Prove that x 2 = k (mod 2n ) is solvable if and
1

if k s (mod 8). Suggestion: use mathematical induction. Assuming that = k (mod 2n) has a solution w, show that an integer / can be found so that + 2n~1 2 s (mod 2n+1). ( u of 17. Assume that ^ 3 and & s 1 (mod 8). Prove that any solution n~ 1 and x 2 = k (mod 2 n ) gives rise to three other solutions, M, w + 2 ~ u + 2 n 1 Prove that these four solutions are incongruent modulo 2". v and u + v is 18. Prove that if w and u are any odd numbers, then one of u of the form 4m + 2. 19. Let ^ 3 and & = 1 (mod 8). Prove that if u and v are two incongruent n~ 1 solutions of x 2 = k (mod 2 W ), then v has one of the three forms u, u + 2 ~l n n K + 2 modulo 2 Hence the congruence has exactly four solutions. n 2 2 - v)(u + = (mod 2n), Suggestion: analyze u = v (mod 2 ), and hence (u

only

x2

TZ

z;)

in the light of the preceding

problem.

Quadratic Reciprocity
20. Consider the a (modp s ) with/? a prime, s congruence x* 1. Prove: if t s the congruence is solvable. If t s the (b,p)

>

>

= p*b

<

is

solvable

if

and only

if / is

21. Consider the congruence

let/?'*

even and x 2 = b (mod/? 5 "*) is solvable. xz = a (mod m). For each prime factor/? of
.

congruence

denote the highest power of/? that divides m, and/>**> the highest power that divides a, so that s ^ 1, t ^ 0. Write c for 9 v y a/pt p Prove that the con
gruence is solvable if and only if (1) for each prime factor/? of m such that
=
t

<

s^, the integer

is

even and

i;

(2) in case

is

even and
/ 2 ).

min

(3, s 2

t2

< s2

then

t 2 is

even and

c2

r (mod 2 ) where

x2

22. Let/? be any odd prime. Let f(a) denote the number of solutions x, y of a (mod /?), where two solutions x l9 y and o? 2 2/ 2 are counted y* o; 1 unless y 2 (mod/?). Prove that 2 and separately unless a? x

/?

which case the result isf(a) = 2/? 1. For the next few problems we extend the range of meaning of the symbol
#, in
it

/(a)

=p -

-I by defining

to be

whenever the prime/?

a.
|

23. Prove that

(mod p). Also prove that assuming a & /b\ = 1-1 if a b (mod p). \Pf
24.

)
|

1 ,

and

that

I
I

=
\
1

\P)
I

\P1\PJ
/j\

\Pl
/
|

For any odd prime p consider the sequence

/2\
,

\PJ \PJ \ P )' Any consecutive pair of terms of the sequence is of one of the four types, 1 or 1, 1 or -1,1. Denote the number of occurrences of 1, 1 or -1, each
of these types by N(l,
1),

N(-1

-1), N(l, -1), and

y>

#(-1,1)

Prove that
2

respectively.

25. Prove that

V |-J
a;=l

I\

=
' J

\r/

-lif/>isanyoddprime.5'jgycj/w:define

and

P
Then evaluate

P rove

that

s<fi,p)-a(l,p)

if

^<a.

^li ^(,/?).

3.3

The Jacob! Symbol

26.

77
,

Using the notation of the two preceding problems show that -s(l p)

is

the
'

excess of the
/
(

number of changes of

sign in the sequence

| ,

-1\
over the
]

number of times

that the sign does not change.

Hence prove

that

Then

+ N(-l, 1) - AX1, 1) establish in case/? s 1 (mod 4) that

N(1
9

-1)

tf(-l, -1)

+1.

N(l,

1)

N(l, -1)
4) that

7V-(-l, 1)

= #(-1, = N(-l,
/

-1)

(p

i)/4j

and

in casej?
tf(l,

(mod

-!)-!= N(l, 1) = N(-l,

if/? is

1)

-1)
L\
|

(p

3)/4.

27.

Prove that

an odd prime then

1.

V
\

"""

= -1

unless/?

A, in
|

m=i
which case the sum has value/?

NOTES ON CHAPTER
Theorem
3.5
is

a variation of a result by Peter Hagis, Amer. Math. Monthly,

77, 397 (1970).

in Special Topics there is a discussion of the calendar problem of the of the week finding day given the date, using a simple congruence involving the function of Definition 3.3. greatest integer

On

page 264

4
Some Functions of Number Theory

4.1

Greatest Integer Function

The function [x] was introduced in Definition 3.3. It is defined for all real x and it assumes integral values only. Many of its properties are included in the
following theorem.

Theorem
(a)

4.1
[x]

Let x and y be real numbers. Then we have

<x<

[x]

1,

- 1<

[x]

x,

^x-

[x]

<

1.

+ m] = [x] -\-mifm is an integer. 00 [x] + [y] ^ [x + y] ^ [x] + [y] + 1.

(c)

[x

"0

ifx

is

an

integer,

(e)

[*]+[-*]

[1

otherwise.

[\x\~\
(g)
(K)

m\
[

= ri
\m\
is

r x "i

tf mis

a positive

integer.

x
[x

M^
+

()
to x,

it is

the fractional part ofx. x. the least integer x] the is nearest i] integer to x. If two integers are equally near the larger of the two.

78

4.1

Greatest Integer Function

79

.[_ x (j) near to x, it

.

+
is

i]

&

?/ze

the smaller

nearest integer to x. If two integers are equally of the two.

is

(k)
1
,

Ifn and a are positive integers, [n/a] n that are divisible by a. 2, 3,

,

the

number of integers among

Proof.

The

first

part of (a)

is

just the definition of [x] in algebraic form.

The

two other parts are rearrangements of the first part. In (&) the sum is vacuous if x < 1. We adopt the convention that a vacuous

sum
z

is zero. Then, for x 0, the sum counts the number of positive integers that are less than or equal to x. This number is evidently just [x]. Part (c) is obvious from the definition of [x].

To prove
and
[x]

<

(d)
1,

we

write x

<
n

<p<
[n

=n+

v,y

=m+

/u,

where n and

m are integers

1.

Then

[y]

+m^
x

+ v + m + ft] = [x + y] = n + m + [v + < n + m +
IJL]

=
n

[re]

+
1

[y]

1.

Again

<
[X]

-v<

writing
1.

+
1

v,

we

also

have

v,

Then

[-X]

72

[-72

V] fO if v

and we have

(e).

To prove
1,

(/)

we

write x

+ v,n=qm + r,Q^v<
,1 g

and have

PL] LmJ
since

_|> L

+r+ m

+ rr+JHL m J

r 4- v

<

m. Then (/) follows because

Part (#)

is

"fractional part nothing more than a definition of the words

of s."
#

<

Replacing a by
[

a in
1,

(a)

we

get
(/z).

<
or,

[a?]

a?

and hence
if

07]

<
(i)

which proves
/z

be the nearest integer +o two are equally distant. Then n = a 6, = -fl L since < fl + [-6 + i TI, ^ J] The proof of (j) is similar to that of (z).
let

To prove

we

taking the larger one

6

<

<i

and

[z

J]

80

Some

Functions of
-

Number Theory

,ja are all the positive part (k) we note that if a, 2a, 3a, n that are divisible then we must integers prove that [n/a] =/. But by a, we see that (j exceeds and so l)a n,

To prove

ja<*n<
Theorem 4.2
Let

(j

1)0,

j<n/a<j+l,

[n/a]

= j.

denote a prime. Then the largest exponent e such that

e
|

is

-mProof.
If p*

>

n,

then [nip
series.
1 !.

0.

Therefore the
is

sum

terminates;

it is

not

really an

infinite
It is

The theorem

induction.

true for

Assume
n.
\

it is

easily proved by mathematical true for (n 1)! and let/ denote the

largest integer such that p*

Since n\

(n

1)!,

we must prove

that

I [//]- 2

[(-!)//]=;.
[ZLl
l

But
1

_ p*"
I

lp \

and hence

] J

/I

if if

P*

io

p^n

the preceding proof is short, but it is rather artificial. different proof can be based on a simple, but interesting, observation. If a l9 a 2 a n are non-negative integers let /(I) denote the number of them that are greater than or equal to l,/(2) the number greater than or to 2, etc. Then
*

'

equal

we let aj be the largest integer such that p"3 Then \j. counts the number of integers ^ /z that are divisible byp,f(2) the number /(I)

Now,

for

^7 ^
,

'

72,

divisible by;? 2

etc.

so

that/()

Hence/(^) counts the [//]. Thus we see that

fc
,

integers/?*,

2p

k
,

k
,

3/?

[nlp ]p

=l

=1

Formula (/) of Theorem 4.1 shortens the work of computing e in Theorem 4.2. For example, if we wish to find the highest power of 7 that divides 1000! we compute
[1000/7]

142,
164
1

[142/7]

20,

[20/7]

2,

[2/7]

0.

Adding we

find that 7

1000!, 7

165

^1000!.

4.1

Greatest Integer Function

81

The applications of Theorem 4.2 As an example, let us prove that

are not restricted to numerical problems.

nl

is

to
it

+ ar = n. To do this we merely have numerator to at least as high a power as divides the denominator. Using Theorem 4.2 we need only prove
an integer
if

ai

0, a^

az

show that every prime

divides the

But repeated use of Theorem 4.1rfgives us

,

+
.

y-L

ffrl

<

r<ii

a.

+
,

a,]

j-y-

fn -]

Summing

this over

we have our

desired result.

special case of what has just been proved is that the product of any k consecutive integers is divisible by k\. To see this we first observe that if one

of the k consecutive integers is 0, the result is immediate because is divisible by any nonzero integer. Thus we need consider only k consecutive positive

k consecutive negative integers, and the latter case can be sub sumed under the former. Taking the largest of the k consecutive positive we look at the product integers as
integers, or
,

n(n- 1)0 -2)

Taking
r

(/!

-k +
k

1)

or

{!}/(/!

A;)!.

2, a x

k, a%

{"}/{^i- <2 2-}

r n\l(k\(n

k)\}

is

in what was proved above we see that an integer, and so k\ is a divisor of

n\l(n-k)\. Another way

>

to

draw

this

conclusion

is

is a binomial coefficient, specifically the coefficient n of (x and all such coefficients are integers. y)

to use the fact that nl/{kl(n k)\} ~ ofxhy n k in the expansion

to prove that

+ A slightly more complicated example

is

(ab)l
a

a\(b\)
is

an integer.

We

must show that

82
for every prime p. Let r

Some
and

Functions of

Number Theory

s denote the integers

such that

f=

< /+1

and/

^6 </

+1 .

Then

-*

+
,

x^ z

["*!

vi~ -Z

a i

since [ai//]

a [A//] by repeated use of Theorem

PROBLEMS
power of 2 dividing 533 ? The highest power of 3 ? The of 6? The highest power highest power of 12? The highest power of 70? 2. If 100! were written out in the ordinary decimal notation without the factorial sign, how many zeros would there be in a row at the right end?
1.
is

What

the highest

3.

For what
I*

real

numbers x
!>];

is it

true that

- 3 + [a]; (c) 3] = 3 + x; = !>]; (d) D*+i] + [*0) [9a] =9? = M + 4. Given that [x +
[a?

GO (W

M+M
+ +
3]

2/1

fr]

and [-as

- y] =

[-#]

[-y], prove that

x or y
5.

is

an

integer.

Find formulas for the highest exponent e of the prime p such that/ divides the product 2 4 6 (2w) of the first TZ even numbers; (Z>) the product of the first n odd numbers.

O)

6.
7.
8.

For any For any For any

number x prove that [x] + [x + J] = [2z]. positive real numbers x and y prove that [#] \y] positive real numbers x and y prove that
real

[jcy],

9.

10. Let

Prove that (2)!/(!) 2 is even if TZ is a positive integer. m be any real number not zero or a positive integer. Prove that an
equation of Theorem
4. 1/ is false.

exists so that the

4.1

Greatest Integer Function

11. If p

83

the terms of

prove that the divisors ofp*<f coincide with 2 q + <f) when the latter is multiplied out. a and m ^ 2, prove that a 12. For any integers m[a/m] is the least nonnegative residue of a modulo m. Write a similar expression for the least positive residue of a modulo m. 13 If a and b are positive integers such that (a, b) = 1 and /> is a real number such that ap and bp are integers, prove that p is an integer. Hence prove that
distinct primes,
(1

and q are

+ p + #2)(1 +

n\l(a\b\) p to prove that

is

an

integer

if (a,

b)

and #
nl
-

1.

Generalize this

a^.a^l
is

ar

an integer

if (a l7

a z ,-

,a r)

14.

Consider an integer n
find the

= 1 and a^ 4- a z 4and the integers /, 1

that are divisible

+
^
i

ar

1.

<n. For each k

.

0,

1, 2,

number of f s

k k+1 Thus by 2 but not by 2

prove

we

and hence that we get the correct value for the sum nj2 + w/4 + w/8 + replace each term by its nearest integer, using the larger one if two 15. If n is any positive integer and any real number, prove that

if

exist.

16. Prove that [2<x] + [2/5] ^ [a] numbers, but that [3 a] + [30] ^

+
[a]

[0]

+
[0]

[a

[2a

0] holds for every pair of real 2/3] does not.

17.

For every positive integers, prove that n \(n

1)! is

a divisor of (2n

2)!.

18. If (m, n)

1,

prove that
71-1

a =i L

2bd
[(1

^r fw^l
-I

(m =i

l)(ii

1)

19. If

m>

1,

prove that

20. Let

be

real,

and

<

=
Prove that
21. Let n be an

+ V3) 2m+1 is divisible by < 1. Define if [nd] = [( n - 1)0],

]

2 W+1 but not by 2 W+2

JO

\l otherwise.

lim

^ +^
>5.
v

+
If

'

odd

integer

n
l]o;

uv, with u

>

and u

+
2

^64n, prove

n factors into the product of two integers, that the roots of # 2 2[v +
H- u)/2}
2

u)/2}

y to get

are integers. Suggestion: use the identity {(u bounds on the integer (u v)/2.

{(M

84

Some

Functions of

Number Theory

22. Let a be a positive irrational number. Prove that the

[1

two sequences,

a], [2 4- 2cc],

-,[/!+

a],
[1

and

4-

or*], [2

2a- 1 ],

[n 4- nor*],
is

together contain every positive integer exactly once. Prove that this
is

false if

rational.

23. Let
that

S
if

S be the set of integers given by [<xx] and [px] for a? 1 , 2, Prove consists of every positive integer, each appearing exactly once, if and
.

only
24.

a and

ft

are positive irrational

numbers such that a

4-

For

positive real

numbers

a,

/?,

y define/(a, p, y) as the

sum of all positive

terms of the series

4-

(If there are

/(&
CLX

no positive terms, define/(a, p, y) = 0.) Prove that /(a, p, y) = Suggestion: /(a, p, y) is related to the number of solutions of py ^ y in positive integer pairs x y.
,

y)'

25. Prove that ifp

is

a prime and
\

^ />*
if if
/i

then
fc

= H "
(o

(mod/?)

5=

n
where

(mod

17)

72

or/? , ^ D* -

1,

is

the binomial coefficient.

is
2

26. Prove that if/?

a prime and
n

^
fc
,

^ /?
1

fc

then

(mod/?

) if

^'(-l)
(1 (mod /?
27. Prove

-1

=0 or/?
2

(mod/?
2
)

) if 72

a/?*-

<p -

1,
r

<3(2

(mod/?),

otherwise.
is

that
1,

if

<

Oj

^p -

Aj

^/?

prime
1,

and

777

J5=o

J
">

25=o

V'>

then

Suggestion: consider

(1

4- #)

(mod/?).

4.2

Arithmetic Functions
<j>(n)

Functions such as
numericalfunctions.
is

of Theorem 2.5 that are defined for

integers n are called arithmetic functions, or

Specifically,

number

all positive theoretic functions, or

the positive integers and

an arithmetic function f'is one whose domain whose range is a subset of the complex numbers.

4.2

Arithmetic Functions

85
definitions.

Definition 4.1

r(n)

is

the

For positive integers n we make the following number ofpositive divisors ofn.
the

o(ri) is the

ak (ri)

is

sum of the positive divisors ofn. sum of the kth powers of the positive

divisors ofn.
all

For example, r(6)

functions.

4, a(6)

12,

cr 2

(6)

50.

These are

arithmetic

The value of A: can be any real number, positive, negative, or zero. The functions r(n) and o(ri) are merely special cases of ak (n), because r(n) =
<r

(X),

a(ri)

0i(ri).
tf 16

It

is

IXz|n/(^0

f r

sum and product off(d)

convenient to use the symbols d n f(d) and over all positive divisors d of n.
\

Thus we write

d\n

d\n
T

d\n

Theorem

4.3

Ifn

e = p^p% - pT

then
-

r(n)

=
/

(ei

!)(*,

+
,

1)
TZ

(e r
if,

1).

^to
if,

r(l)

Proof.

<fi
(e r

A positive integer d divides Thus ^ ^ for = 1, 2,

r.

and only

there are just

4.3 that

= p{*p{* 'pf fo + 1)(^ 2 +!)

' '

1)

such
)

d.
1

If

(m,

it

follows from
is

Theorem

r(m)

r(m)r(n).

Definition 4.2

= f(m)f(n) for every pair of'positive integers m,n satisfying (m,ri) = 1, f(mri) thenf(n) is said to be multiplicative. Iff(mn) =f(m)f(ri) whether m andn are
relatively

Iff(n)

an arithmetic function not

identically zero such that

prime or not,f(ri)

is

said to be totally multiplicative or completely

multiplicative.

If/is a multiplicative function,/(/i) =/()/(!) for every positive integer n, 1. and since f(ri) is not zero for all values of n, we see that/(l)

From the definition

in pairs, then

of a multiplicative function/it follows by mathematical

,

induction that ifm l9 ra 2

mr are positive integers which are relative prime m ) =/(

r

/(/HX/MS

In particular, this result would hold if the integers ls powers of distinct primes. Since every positive integer
into a product of

m m2 >

prime powers of distinct primes, it a and we know the value of/(p ) for every prime/? and function multiplicative for every positive integer every positive integer a, then the value of f(ri)

mr are prime can be factored follows that if /is a

,

/(2

n can be readily determined by multiplication. For example /(3600) 2 4 2 if g is a totally multiplicative function and we ), Similarly,

know

)/(3 )/(5 the value of g(p) for every

prime p, then the value of g(ri) for every

86
positive integer n can

Some

Functions of

Number Theory

be readily determined. For example #(3600)

These basic properties can be stated in another way. First, if /and g are j = functions such that for all multiplicative f(p*) g(p ) primes p and all
positive integers 7,

then/()

Second,
for
all

if

/and g

primes/?,

then/ =

g(n) for all positive integers n, so that/ g. are totally multiplicative functions such that/(/?) g(p)

g.
let F(ri)

Theorem 4.4
Then
F(ri) is

Let f(ri) be a multiplicative function and

multiplicative.

= ^d n f(d).
\

1. If 1 or n 1 then Proof. Suppose (m, n) F(m) so that F(mri) F(m)F(n) is clear in these cases. Otherwise canonical representations

m=

or ^(72) 1, and n have

with positive exponents a< and ft and where l9 2 p p ,pr q l9 q 2 qs are distinct primes. The positive divisors dI of n are the numbers dl just l r pl pl**"pl for all possible choices of the y's satisfying 5^ y,- <^ a z are given by Similarly the positive divisors rfa of grji^ q* 9
-

-.

4= ^
ftl

d*

^
d

A- Therefore as
all
2
-

runs through

all

positive divisors of n,

and d2

runs through
values

= d^ = /#/#

positive divisors of

m,
-

their

product
2

d^
a i5

'/>J>?^

#,
nm

these are just all the positive divisors of other words

^ y, ^ = p**p*

runs through the <5, ft; but

^
-

p q

ar

ftz

ps .

In

(iiln^lw
It is clear that

d\nm

(dl9 4)

1,

and hence we have

^
d\nm

2 2/^^) = 2
dilndzlm
di\n

could have used this theorem and Definition 4.1 to prove that r(n) is multiplicative. Since r(n) 2<*|n 1 is of the form 2*, n /(/), and since the 1 is function/(/i) multiplicative, Theorem 4.4 can be used, and we see that

We

r(n)

is

multiplicative.
'
' '

Then Theorem

t is

a prime, then r(pp)

/>/'

1>

^pfj

and no

to prove. If 1 since/;? has the e t positive divisors more. Then, since r(n) is multiplicative, we have

4.3

et

would have been easy

1,

r(^0

fe

l)(* a

1)

(er

l).

4.2

Arithmetic Functions

87

This exemplifies a useful method for handling certain arithmetic functions. We will use it to find a formula for a(n) in the following theorem. However,
it

same manner

should be pointed out that a(ri) can also be found quite simply in the as we first obtained the formula for r(ri).

Theorem 4.5

Ifn

= p[ pl*
l

/?;*,

then

Proof.
y(fl)

By
,

definition a(ri)

F(ri)

2<z|n

^
is

so

we can apply Theorem

and a(n)
-

4.4 with

<r(n).

Thus a()

multiplicative

= JJJUi cr(pfOwhose sum

is

But the positive divisors of

p^

are just I,p i9 pi,

p**

I)-

PROBLEMS
Find the smallest integer x for which <f>(x) = 6. Find the smallest integer x for which r(x) = 6. 3. Find the smallest positive integer n so that a(x) = n has no solutions; exactly one solution exactly two solutions exactly three solutions. 4. Find the smallest positive integer m for which there is another positive
1.

2.

such that a(tn) integer n j^ n T(n)l2 5. Prove that Y[d\n d

=
.

a(ri).

6.

Prove that

^ d n d = ^^\n njd,
\

and more generally than

7. 8.

Prove that o_ (X) = n~ k a k (n). Find a formula for ff (/z).

fc
fc

for every n, show and^(/z) are multiplicative functions, and^() 5^ that the functions F(n) = f(n)g(n} and G() = f(n)lg(n) are also multiplicative. 10. Give an example to show that if f(ri) is totally multiplicative, F(ri) need
9. If/(/z)

not also be totally multiplicative, where F(ri) is defined as *]?a\ n f(d). 11. Prove that the number of positive irreducible fractions ^ 1 with denomi
nator
12.

< n is 0(1) + 0(2) + 0(3) + + 0(). Prove that the number of divisors of n is odd
k

if

square. If the integer or double a square.

13.

prove that

ff

k (ri) is

odd

if

and only if n is a perfect and only if n is a square

infinitely

Given any positive integer n

>

1,

x satisfying r(x) n. integers 14. Given any positive integer x satisfying a(x) integers 15. Prove that if (a, b) >
16.
n.

prove that there are

is

many

prove that there

only a

finite

number of

We

is, if

1 then ak (ab) < c k (a)e k (b) and r(ab) < r(a)r(b). = 2m, that is a perfect number if a(m) say (following Euclid) that n 1 is a than other itself. If 2 divisors is the sum of all its positive

88

Some
prime p prove perfect numbers. 17. Prove that an integer q
9

Functions of

Number Theory

that 2 n ~^p is a perfect

is

number. Use

this result to find three

a prime if and only if a(q) = q + 1. & where & y and & < jr, then A: = 1. <j(q) 19. Prove that even every perfect number has the form given in Problem 16. n Suggestion: assume that 2 ~^q is a perfect number, where n > 1 and q is odd. Write a(q) = q + & and so deduce from = 2 n that y = (2n - 1). crp"-^) Thus k ^ and k <q. 20. For any integer n g 2 define i>() as (-1)', where is the total number of
18.

Show

that if

=q +

prime factors of n. For example, define v(\) = 1. Prove that v(ri)

if
is

=
is

16,

then/

4;

if

72, y

5.

Also

a totally multiplicative function and that

n

2(1
i'(d)

if

a perfect square, l
.

(Q otherwise.

21.

Ifd\n and S (nfd), then d (n/d). Prove that the set of ordered pairs (d, d) where Granges over all positive divisors of a fixed integer n and, for each value
\ \

of d, 6 ranges over all positive divisors of n/d, is a symmetric set in the sense that if (a, b) is in the set, so is (b, a). 22. Prove that the set of pairs in the preceding problem is the same as the set of pairs (d, 6) over all positive d and 8 such that dd n.
\

23. Consider the set of ordered pairs (d, y) where d ranges over all positive divisors of a fixed integer n, and for each such d, y ranges over all positive divisors of d. Prove that this is the same as the set of ordered pairs (fly, y) where y ranges over all divisors of n and, for each such y, ft runs the

through

positive divisors of

72/7.

4.3

The Moebius Inversion Formula

The Moebius function
/j,(ri)

Definition 4.3

is

defined by

i/=l
if a
r
2
1

nfor some a

>

{(

l)

ifn
/u,(ri)

= pipz
is

pr

pi distinct primes.

Theorem 4.6

The function

multiplicative

and

O
Proof.
It is clear

if

>

1. is

from the
is

definition that ju(n)

multiplicative. If F(ri)

=
1

2d| w

X<0,

th en F(ri)

and F(p e)

= 2;=0

multiplicative by Theorem 4.4. Since F(l) = /*(!) = 1 + (_i) = o, we have the desired result. /^(/)

Theorem 4.7
integer n,

Moebius

thenf(n)

inversion formula. IfF(n)

= ^d nf(d)for every positive

\

4.3

The Moebius Inversion Formula

89

Proof.

We

have

2
d\n
d\n
5\(n/d)

=2
5\n d\(n/S)

8\n

d\(n/5)

by Theorem

4.6.

Theorem
F(n)

4.8

If f(ri)

= ^^(^(n/d)

for every positive integer

n,

then

ZdinfWAgain by Theorem 4.6 we find

Proof.

d\n

d\n 5\d

d|n y\d

y\npy\n

^'
its

y\n

0|(n/y)

=
It

F(n).

should be noted that Theorem 4.7 and

converse,

Theorem

4.8,

do not

generalization of these two require that f(ri) or ^(72) be multiplicative. of the next section. in 4.14 at the end Theorem results is given As an example we will obtain useful. The last two theorems are often very 2.4 Euler's the results of Section ^-function in a different way. In concerning

Theorem

we saw that <f>(ri) is the number of positive integers less than or that are relatively prime to n. Let S denote the set of integers equal to n i n. that is, the set of integers i satisfying 1 72, 1, 2, separate S
2.5
,

^ ^

We
d.

into subsets

element

Sd where d\n, by putting into Sd if (z, 77) of S is in exactly one Sd Moreover, is in Sd if and
,

Then each
if

form jd with

^y <
.

^(n/d) elements in Sd

z is of the only are Therefore there exactly (_/', n\d) Since there are TZ elements in 5, we have n dln (f>(n/d) y
.

n\d and

1.

which we can write as n

= ^d n <f>(d).
\

=^

This

is

Theorem

2.17.

Then, by

Theorem

4.7,

'"

V<
Also the function ^(rf)/^
is
is

multiplicative,

Theorem

4.4.

Hence

j>(n)

and therefore so is <f>(n)jn by 2.15. multiplicative, and we have Theorem

n replaced Finally, using the foregoing equation with

by^

e
,

we have

90
if

Some
e

Functions of

Number Theory

and hence
1
'
'

<KA P?
since
<f>(n) is

P/)

= &PF
is

A*(I \

-} pj

f
\

- -M
pj

f
\

- -M
pT /
f

multiplicative. This

Theorem

2.16.

PROBLEMS
1.

2.
3.

Find a positive integer n such that ju(n) + n(n Prove that p(ri)ti(n + 1)^0* + 2)^0* +3) =
Evaluate

+
if

1)

is

p(n + 2) = 3. a positive integer.

2&1

^(/!).

4.

Prove Theorem 4.8 by defining

G()

as

2d| n /W) then a PPty* n g Theorem

4.7 to write/(ji)

= ^l

p(d)G(nJd).Thus

%dln v(d)G(n\d)
=
=

2a\n f*(d)F(nld).

Use

this to

5. If

that F(l) G(l), F(2) G(2), F(3) G(3), and so on. denotes the number of distinct prime factors of a positive integer n,

show

prove that 2d|nlM^)l

6. If
\

2*.

F (n) = ^?d n f(d)

for

every positive

integer

n,

prove that

/(/*)

1.

Let

have the distinct prime factors Pi,p 2

,pjc-

Prove that

Similarly, evaluate 2d[

8. If
72

is

any even

integer,

prove that ]d|n

1)2

9.
(

By

use of the algebraic identity (x

1)2

12
a?

result

2-i
+

= 2JU {( + = w ( + l)/2.

z2}

p(d)<l>(d) = 0. + I) 2 - #2 = 2x + 1, establish = 25.! (2 + 1) and so derive

that

the

10.

By

use of the algebraic identity (x

3

that (n

I)

derive the result

+ I) 3 x* = 3#2 + 3# + 1, establish = 22=i {(a? + I) 3 - *3 } = 2S-i ( 3 ^ 2 + ^ + 1), and so 2 22-i ^ = ( + 1)(2 + l)/6. (The results of this and the
3

previous problem can be established by other methods, mathematical induction,

for example.)
11. Let S(ri) denote the

sum

of the squares of the positive integers

n and

prime to n. Prove that

n into classes, so that all Suggestion: separate the integers integers that (k, n) are in the same class. 12. Combine the results of the two preceding problems to get

=d

k such

fn

i\
-

n)

4.4

The

Multiplication of Arithmetic Functions

to get

91

Then apply the Moebius inversion formula

1

-'-"-4
= that ^&\ n dp(d) where the are factors of n. distinct l) <f>(ri)pipz ( Pk/n Pi,p 2 ,pk prime z 14. Combine the results of the two preceding problems to get S(ri) = n <f>(ri)/ 3 4- ( ,pic are the 1)^00/^/72 Pkl& for 7i > 1, where as before pip 2 = <f>(n)/n. d n f^(d}/d prime factors of n. Suggestion: use the formula 15. Given any positive integer k, prove that there exist infinitely many integers
13.

Prove that/00
k
-

= n^ (n) is a multiplicative function and

'

'

'

'

'

'

'

n such that

4.4

The

Multiplication of Arithmetic Functions

There are two standard kinds of multiplication of arithmetic functions. The first is the straightforward one, where the product^ of two functions /and

defined by (fg)(n) =f(n)g(n). If /and g are multiplicative so is/-. now turn to a more fruitful definition of multiplication.
is

We

If/ and g are arithmetic functions, define the Dirichlet product (or convolu n tion) f*g as the arithmetic function whose value at any positive integer
is

given by

d\n

did z =n

where the

first

sum

is

over

all

positive integer divisors

d of n, and

the second

/* g

ordered pairs dl9 d2 of positive integers with product n. It is clear that the two sums are equal and that multiplication is commutative, g */ because the set of ordered pairs d2 dl is the same as the set of

sum

is

over

all

ordered pairs dl9 d2 Furthermore, if h

.

another arithmetic function, the associative property (f*g)*h=f*(g*fy can b e established as follows. For any positive
is

the definition gives integer n a straightforward application of

where the sum is over and (/* (g product

,

all

* K))(ri) is

ordered triples dl9 dz dz of positive integers with seen to be the same thing.
,

92

Some

Functions of

Number Theory

will

In addition to the functions p, r, <f>, a already examined in this chapter we need the functions 7, U, 1 and for defined by 7(1) 17(1) E(\) all w 1,

>

7(/2)

0,

U(n)=I,

E(n)

n.

easy to verify that these are multiplicative functions. The function 7 is seen to have the property 7*/ /*7=/for every arithmetic function/, and so 7 is called a multiplicative identity for Dirichlet multiplication of
It is

arithmetic functions. Furthermore, 7

is

there were another function 7X such that 7X

function/, If two arithmetic

we could

= 7* 7X = 7. functions / and g have

write 7t

the multiplicative identity because if * 7X for every arithmetic */

=/

=/

the property

f*g = g */=

7,

we

say that

/ and g are multiplicative inverses of each other and we write

multiplicative inverse if and only

Theorem
iff (I)
Proof.
7(1)

4.9 An arithmetic function f has a ^Q.lfan inverse exists it is unique.

If/has an inverse/- then/*/1,

so that /(I)

0.

Conversely

from these equations. Next we can 1 1 C/*/- )^ =/(2)/-Hl) +/(1)/~ (2). Continuing
duction

= 7and/(l)/- (l) = (f*f~*)(l) = if /(I) ^ we can calculate /^(l) = 7(2) = calculate /~ (2) from
1

we

^(n)

is

if/^O) has been evaluated for 7 determined by

see that

by
1
,

mathematical
2,
,

in

then

=
because this

J(n)

(/*/-')<

sum contains f~\n) only in the term f(l)f~l (n). Thus this can be solved to give a unique value forf~ l (n). equation Next we note that Theorem 4.6, if restated in the notation of this section,
can be written as follows.

Theorem 4.10

7*^ = 7,^= U- andU={T\

l

The functions

(JL

and

U are multiplicative inverses,

thus

U=

Furthermore, Theorem 4.7 and 4.8 can be stated more succinctly in the following way.

Theorem 4.11
thenf =
It
p,

If

f and F are

any arithmetic functions such that

F=

F and conversely.
we do not need
it

17*/,

may

be noted that
because

this result,

follows

of Dirichlet multiplication.
/j,

F =/

to use Theorems 4.7 and 4.8 to prove from Theorem 4.10 and the associative property That is, if we multiply F = C/*/by /* we get

and the process

is

reversed by multiplying by U.

4.4

The Multiplication of Arithmetic Functions

93

The set of all arithmetic functions f with /(I) forms a Dirichlet multiplication. Similarly, the set of all multiplicative under group arithmetic functions is a group.

Theorem 4.12

First consider the arithmetic functions /with /(I) 5* 0. This set Proof. then (/*#)(!) closed undermultiplicationbecauseif/(l) 7^ Oandg(l) 5^

is

have already observed that the associative property holds f( l)g(l) 7* 0. functions. The identity element is the function /, and arithmetic for all
multiplicative inverses have been taken care of in Theorem 4.9. Turning to the multiplicative functions, we note that the associative property holds as before, and that the function / is multiplicative. So we need only prove first that the product of two multiplicative functions is
multiplicative,

We

and second that the inverse of a

multiplicative function is

multiplicative. First consider

two

multiplicative
is

we

multiplicative functions /and g. To prove that/*g is parallel the notation and the proof of Theorem 4.4, which

not surprising since Theorem 4.4 is just the special case of what we are now proving, with g replaced by the very special function U. Thus for relatively

prime positive integers

m and n we get

d\nm

g
j

>

1 a function Finally, to prove that if/is multiplicative so also is/" , we define 1 5 and as follows. First set g(pf) ( p ) for every prime p every integer ?* 0. we want g to be multiplicative, so for any integer n H_p*

=/~

Now

and g are the primes p i are distinct. we define g(n) *), where the is both multiplicative, so/*g preceding paragraph. multiplicative by k Also for any prime power p we see that

= IJ g(pf

Now/

2
* *

Thus/* g coincides with /on prime powers, both are multiplicative functions, and so/* g = /by the basic observations made immediately after Definition

94
4.2.

Some

Functions of
it

Number Theory

But the inverse of/is unique by Theorem 4.9, and

is

follows that/-1

= g,

and so/"1

multiplicative since

is.

Theorem 4.13
[*>

The following

relations hold

among

the functions

/,

U, E,

r,

<f>,

<r:

(1)

(ji

(4)

or

= =

J7-i
17 *

(2) r

(5)

or

= U* U = *r
<

(3)

<f>

(6) o*<j>

= * = E*E.
/*

Proof. Item (1) was proved in Theorem 4.10, and is repeated for complete ness. Since all the functions are the results for multiplicative, we need

prove

prime powers only. Thus items

(2), (3), (4)

are established

by noting that

(C7*

C*
(17 *

)(/)

= >

E(d)U(p

/d)

Finally, items (5)

and

<*

= <>*)*(J7*
identities in

(6)

follow at once from the others; for example

in terms of

The

J7)= C/*^. Theorem 4.13 can be elaborated

(3)

any positive

integer n.

Thus item

becomes

Item 3 implies, on multiplication by C/", that ,&= Z7*^, and this gives Theorem 2.17 when applied to 72. Thus we have given an independent proof of Theorem 2. 17. Items (5) and (6) can be written in the forms

a(n)

=
d\n

and
("/^)
d
l

2w=
d\n

WT( " )

'

d\n

Finally, this approach by algebraic structure is now used to derive a more general inversion formula than Theorems 4.7 and 4.8. Let p(x) be a complexvalued function defined for all real x 1. If arithmetic function

/is any

4.4

The

Multiplication of Arithmetic Functions

95

define the product //?

by

(4.1)

for

Thus//?, like /? itself, is a function whose domain is all real is the complex numbers, or a subset thereof. If g is another arithmetic function it turns out that g(ffi) (g */)/? To prove this
all real
1

1.

and whose range

we

first

evaluate g(f{i) for any real x

Co?]

thus

3=1

}=1 k=l

We

collect terms with equal values of/fc in this

ofjk, say jk n is any positive integer

we

note that n

x,

double sum. For any fixed value jk ^j[xfj] ^j(x/j) = x. Conversely, if so n -^ [x] and if /z = /fc is any factoring of n
9

we
4.1.

see that/

^ yl ^

[re]

and k

//

[a;]//

[a;//]

Hence we can

rewrite the double

sum above

in the

by part (/) of Theorem form

*AO

2
9

/)() #*/)

(te

function

Having established that g(//3) = (g*f)p, we take /to be the Moebius ^ and we write y for pf} so that y(rc) = (l*>fi)(x) for all real a; 2: 1.
5

Then we

see that

where the

Uy

last step

here follows at once from Definition

j8

implies /t(Z7y)

(4.1).

or y

Conversely

^^3.

Thus we have proved the following

result.

Theorem 4.14
x

Le/

/?(#)

and y(x) be complex-valued functions defined for

then

all

>

1.

jjT |8(x)

= B=i r(^//)/^ allx^l,

w
(4.2)

rW =
x^.
1
,

/or all

arf

conversely.
is

This result implies, but

that

Theorem

4.14 implies

Theorem

not implied by, Theorems 4.7 and 4.8. To see 4.7, suppose we are given any arithmetic

function/. Define y(x) in this way: For any positive integer k take y(k) 0. Next (i(x) is defined in terms 1 define y(x) /(*); for any noninteger x

of y(x) by the

first

equation in Theorem 4.14, so that

]8(a?)

if

is

not an

96
integer,

Some
and ifx
is

Functions of

Number Theory

an integer, say x

w,

/./)

J=l

2
5=1
5|w

j\n

Then

(4.2)

gives us the conclusion of Theorem 4.7

if

we

define the arithmetic

function

F by

taking ^(72)

fi(ri)

for

To

establish that

Theorem 4.14

positive integers w. implies Theorem 4.8, the argument

all

is

very

is now the similar, except that the arithmetic function starting point, and is identified with .F on the with for x 1 but x j8 positive integers, /?(#)

not an integer. Then

f(ri)

(4.2) is

used to define y(x), and

finally /is defined

by

y(n) for

all

positive integers n.

PROBLEMS
1.

Prove that

7,

E and
is

2.
it

If/ and g are

follow that/" 1

totally multiplicative but that p,T,<f> totally multiplicative, does it follow that/*^ totally multiplicative?

Z7 are

and a are not.

is

also?

Does

3.

Prove that p * r

= U and
*
/*

hence

^d n f^(d)r(n/d)
\

1.

4. 5.
6.

Prove that
Prove that
Define the

E=
<r

= 2d|7i j*(d)<y(n/d). * U = E* r and hence = n ^d\n r(d)/d. 2<z|n ^W) arithmetic function A by A() = 2f=i (y, ), where
or

and hence

/z

(y,

A() by using the fact that the number of such that (/, n) = d is ^(/i/^/), and thus establish integers/ among 1, 2, that A = E* Then use Theorem 4.12 to prove that A is a multiplicative function. Finally, prove that the analogous function to A with the l.c.m. in
n.
,

denotes the g.c.d. ofy and

Evaluate

<t>.

place of g.c.d., namely ]T?=i [/, n], is not multiplicative. 7. Let /, g, h be three arithmetic functions satisfying f(ri) = n g(d) 9 = for all positive integers n. Prove that h is the g(ti 2,d\nf(d)h(n/d),f(l) ^ Moebius function.

^\

8.

9.

Prove that the Prove that the

set
set

of

all all

arithmetic functions /with /(I)

1 is

a group.

multiplication,

and

arithmetic functions forms a ring, with Dirichlet addition defined by (f g) (n) =f(ri) g(n) for all

of

positive integers n. 10. Prove that (/)

=f(gP), where the product

is

defined as in (4.1).

4.5

Recurrence Functions

particular sort of arithmetic function can be defined as follows. If a, b, x l are arbitrary numbers, x09 perhaps even complex, we let /(O) x and l9 1. This determines /(I) 1) /(it af(n) 1) for n bf(n

xQ

4.5

Recurrence Functions
b,

97

f(ri) uniquely,

depending only on a, x n in place of/() and we have x n+1

a recurrence or recursion formula.

ax n

x v For convenience we will write + bxn_: Such a relation is called

.

In order to get a simpler relationship

we write this last equation in the form

x n+I
If ki

kx n

(a

)(:&

and & 2 are the roots ofk 2

- fc^) + (b + ak - fc^a?^. ak b = Q, then + k = a and we have

fc

and hence

Substracting
fore, if
/: 2

5^

we find (Ar2 kl9 we have

M

fcO^

(^

k&^kl

(^

k&J)k*. There

/C 2

/Cj

Thus we have a formula for finding the value of x n directly in terms of a; _ a, b, x Q x without having to compute the values of # 2 x z n 1 How = this formula has no if k k In this case we ever, z meaning might try holding ki and n fixed and letting k% approach k l9 hoping that this will suggest a solution that we can then verify. We regard the above equation as having the form
, ,

with g(kj)

= h(ki) =

0,

and we apply L'Hospital's

rule in the

form

if

the limit on the right exists. This gives us

lim x n
kz^-ki

=
is

lim [n(x l
k2-+7ci

k^k^ + x k?} = nx^k^

Q

Now we
y n actually we have a

can

set y n nxjc^ the solution. Since A: x

nx k^

+ x k^
Q

fc 2 is

2kl} b
(n

k\.

We

have y
1

the only root of fc 2

and check whether or not ak b = = for n and, 1, x^ ^ y^

IKfc?

(n

+ l)^^ +
2

xjk*+*

ay n

fcj{(n

- i^fcr -

- 1)^^" + ^fcr
1

fey^.

98

Some
for n i> 1,

Functions of

Number Theory
for

Now,
2/o

we have y n+l
#1

0,
,

2/1

x n+i

a(y n

0.

This

implies

yn +

O+=
# w+1
fci

b(yn _^

- ^_ ) and = n +
x 1

2, 3 9 4,

and we have

^=
We
then so are

2/n

n^ifcr

(n

iKfci,

/c 2 .

have found formulas for xn in both cases. If all the x n9 since x n = ##_! + bx n _ 2
.

a, b,

XQ x are integers
,

The Fibonacci numbers

^n+i
fc a

,F

F^

are

defined

by
0,

=
(l

Fn

Fn-i- In thi s case

J

we have

k*

F = 0, F^ = 1, - \/5)/2, k^ = (1
Q

+V5)/2,and
W

by(4.1),

MA + ysr
""5ll
2 /

/i
I
-

As another example we consider the sequence 0, 1 3, 8, 21 x = 0, ^ = 1, a = 3, A = -1. Then & x = (3 - V?)/2, ^2

,

for

which

(3

+ >/5)/2,

and

But

<

w {(3

\/5)/2}

and x n
1

is

an integer so that

in this case

we can

write the solution as

/5\

PROBLEMS
1.

Without using the

results of this section find

a formula for xn ifxn+1

ax n

Find a formula for x n ifxn+1 = 2xn xn _, x = 0, x = 1. Also if x = 1, = = x Then do 1. the same for 2a; #! n + 3#n _ 1 n+1 3. Write the first ten terms of the Fibonacci series. Prove in general that any two consecutive terms are relatively prime. 4. Prove that the Fibonacci numbers satisfy the inequalities
2.
.

if
1

n > 1. > F2 +

F!

Suggestion: write a for 3 = a2 + a > .Fg, a

(1

F3 + F2 = F4
n

4- ->/5)/2,

and observe that a 2 = a Then use induction.

.

5.

Prove that for n

/i

2,
//z

1\

2\

3\

-4
3

Notes on Chapter 4
where the sum of the binomial
largest/ such that 2j
coefficients

99

on the

the right terminates with

1.

Suggestion: use the fact that

m -1
6. 7.

F + F2 + F3 + + Fn = F^ - 1. Fn_Jn+l - F* = (-!). F _iFn + Fm Fn+ for any positive integers m and n. 8. Prove that Fm+n Then prove that Fm Fn ifm\ n. Suggestion: let n = mq and use induction on q. = F F3 F4 F5 F Prove 9. Consider the sequence 1, 2, 3, 5, 8,
Prove that Prove that
1
OT

2,

that every positive integer can be written as a sum of distinct terms from this sequence. Suggestion: for k not in the sequence, let n be such that n _ x

<

A:

k - /v^ < Fn _ 2 Prove the statement by induction. a n that can be con 10. Let/(/0 denote the number of sequences a l9 a% structed where each a j is +1, -1, or 0, subject to the restrictions that no two 1. Prove consecutive terms can be +1, and no two consecutive terms can be

< Fn Show
.

that

<

that/(/z)

is

the integer nearest to J(l

+
1

n+1 >/2)
.

Suggestion: prove that/Oz)

11. (a) Let

divides another

Sn be a set of integers #, member of Sn Show

,

tc

that

^ such that no member of 5n Sn can have [(n + l)/2] members

,

but no more.
(b)
(c)

Find the

maximum
k

size

of
is

Sn

if

the x are also restricted to be odd.

Show

then n
(d)

<

that if 2 a, a odd, 3 k+l a.

in the

Sn

of (a) and

if

Sn

is

of

maximum

size,

Show that an Sn

of (a) can be less than

maximum
to

size,

and yet be maximal

in the sense that

12.

no new member can be adjoined

Sn

Let/(w) be the sum of the first n terms of the sequence 0, 1, 1, 2, 2, 3, 3, Construct a table for/(). Prove that/() = [ 2 /4], For x y integers 4, 4, x > y, prove that xy =f(x + y) y). Thus the process of multiplica f(x tion can be replaced by an addition, a subtraction, looking up two numbers
.

in the table,
13.

Use

and subtracting them. the ideas of this section to find a formula for x n

if

xl

2,

#2

1,

and x n+l

xn

4ccn _ 1

4#n _ 2 Suggestion: consider the expression

.

NOTES ON CHAPTER

inversion formula

The observation following Theorem 4.14 that this result implies the Moebius was pointed out by our colleague Robert B. Burckel The evaluation of certain determinants whose elements are number theoretic
is

functions

264 given in Special Topics, page

if.

5
Some Diophantine
Equations

5.1

Diophantine Equations

There are many problems and puzzles whose solutions require more than
just finding all the solutions of some equation. They are worded in such a way that the desired solutions must satisfy other conditions. For

example,

we

notice that

26 2 1$ --=--=-

is

correct although this cancelling of the 6

violates the rules of algebra. Our problem is to find all the positive fractions that behave in this way. That is, we are to determine x, y, z in such a way that

+y= l$y + z
lOx

This reduces

(y

- s> =
and
is

100

>, but we

are only

interested in solutions such that x 9 y, will not carry out the solution. It

We
19

z are positive integers less than 10. not hard to see that the solutions are

16

26
J

49
' >

o7 95

TA 64

Z? 65

and the

IQx
fractions of the

98

form
IQx

+ +

x
.

In the foregoing problem the x)z W(y equation (y z)x is an indeterminate equation. It has many algebraic solutions, and we are required to sort out the solutions in which x, y, and z are less than 10.
positive integers

problem and we say we are solving a Diophantine equation. This particular problem is merely a curiosity, but there are many important Diophantine equations. In general, the added
is

Such a problem

called a Diophantine

restrictions are that the solutions are to be integers or, Frequently the solutions are also to be positive.

sometimes, rational.

100

5.2

The Equation ax+ by

101

There are endless

general method of

varieties

solution.

We will

We

of Diophantine equations, and there is no discuss a few of the simplest equations.

will also consider

some

related problems.

For example Theorem

5.6

x\ x\ x\ integers, for each positive integer n. will not attempt to find all the integral solutions.
solution, with x l9

states, in effect, that the equation x\

n has at least one

x2 # 3 x
, ,

However, we
7.

Pell's

2 equation, x

dy

N,

will

be discussed in Chapter

5.2

The Equation ox
by = suppose a ^

by

two variables having integral coefficients can be put The problem is trivial unless neither a nor b is zero, so we can 0, b ^ 0. We let g denote (a, b). Then Theorem 1.3 shows that there exist integers x and y such that axQ + by = g. Numerical values for X Q and y can be obtained conveniently by applying the Euclidean

Any

linear equation in

in the

form ax

c.

algorithm,

Theorem

Now if g)( c,
we
2/i

then ax

1.11, to the integers c clearly by

\a\

and

\b\.

use our solution X Q , y

==
( clg)y<)

of ax
c.

f ax

by

by

has no solution in integers. If g c, g to get a solution x (clg)x Q

In order to find

denote any integral solution. Then hence

(5.1)

we have

integral solutions ar c bs ax
all

we

let r, s

= =

+ by^

and

-(r-sO^-^-jfr).
|

8 8 y 1) and (alg,bjg)=l by Corollary 1.7, and hence (a/g)\(s = Theorem 1.9. This xl = s r and i) by y (big) (r implies (alg)u some for and then u t. Therefore (5.1) implies (blg)t integers u, t, every = c can be written in the form r = integral solution r, s of ax + by
But
x\

(blg)t, s

yi

(alg)t.

Since these values clearly satisfy ax

by

c,

we have
'

solved the Diophantine equation. Note that the equation has solutions and only if (a, b) divides c. if If a, b, c have a common divisor, it can be divided out so we can presume that (a, b, c) 1. With this hypothesis our earlier conclusion can be stated

this

way: that ax

that (a, b) 1, thus g solution in integers of ax

by

= c is solvable if and only if (a, b) = Assuming = in the previous discussion, we see that every + by = c is given by x = x + bt, y = y at,
1. 1 : : t is

where x ly

y^ is

any particular solution and

an arbitrary

integer.

PROBLEMS
1.

Prove that

all

solutions of 3x

5t,

= -1 -

5y

3f; also in the

form x

can be written in the form x 2 - 5t, y = -1 + 3t; also in

Some Diophantine Equations

the form x = -3 + 5t, y = 2 - 3t. Prove that x = a + bt, y = c + J/ is a form of the general solution if and only if a, c is a solution and either b = 5, </ = -3, orZ> = -5, d = 3. 2. Find all solutions of IQx 7y = 17.
3. If <2#

is

solvable, prove that

it

has a solution x

y Q with

*o
4.

<

^
c is

1*1-

Prove that ax

by

c is solvable if

and only

if

ax + 6y

solvable.
5.

Prove that the conclusion that ax

+ by =

c is solvable if

and only

if (a, b)

divides c can be

drawn from the conclusions of Theorem

2.13, with a,

m, b

= (a, b, c). if and only if (a, b) Given that ax + by = c has two solutions, (XQ y ) and (x l9 yj with 1. #! = 1 + # and given that O, b) = 1, prove that b = 8. Interpreted geometrically, the solutions of ax + by = c in integers are
7.
,
,

of that theorem replaced by a, b, c. 6. Prove that ax + by = c is solvable

certain points on the straight line represented by the equation in an x, y coordinate system. If (a, b) = 1 prove that any segment of the line of length 2 2 (a + )^ contains at least one of these points with coordinates.
,

integral

9.

Find necessary and

sufficient conditions that

c-,2

&
have
at least

+ biV +

=d
^

l9

b 2y

c2 z

= dz
assuming that the

one simultaneous solution in integers

x, y, z,

coefficients are integers with b^ bz . 10. Give an independent proof that ax

integers x,

<
is

<

+ by = c has at least one solution in y if (a, b) c, by using induction on max (a, b). Suggestion: if ^thenao; + by = c is solvable if and a)y = c only if a(x + y) + (b
\

solvable.

5.3

Positive Solutions

Let us suppose that a, b, and c are positive, that (a, b) c, and that we are asked to find all solutions r, s of ax by = c in positive integers. We solve as in Section 5.2, and have only to restrict t in such a way as to make r and s
\

positive.

We

merely

restrict t to
t is
[

the range

(g/iX

< <
t

(g/a)^.

The

smallest allowable value for

[(

(g/iX

1]

and the

gl a}y-L

largest value is

!]

The number of

solutions

is

then

from which we

find,

by use of Theorem

4.ld, that

5.3

Positive Solutions

103

Since

-(?/%i

- (/#K =

-(gjab)(byi

+ axj =

-(gc)l(ab)

we

finally

have

_r_ii
L

_i<^< - ~

-f--^]
L

ab\

abl

These inequalities do not constitute a precise formula for N, but they do consecutive integers one of which must be N. Note that there will specify two at least one positive solution of ax + by = c if g c and gc > ab. be always
\

PROBLEMS
1.

(a)
(b)
(c)

Find all solutions 5x + 3y = 52;

I5x

in positive integers:

4Qx

+ +

ly = 111; 63y = 521

(d)
(e)

123o?

+57y =531;

Ux +5012/ = 1; Ux + 501y = 274; - 1000. (g) 97x + 98y 2. Prove that Wlx + 31y =
(/)
3.

Given that

(a, b)

and

ax
4.

has infinitely be positive integers. Prove that there is no solution of ax in positive integers if a + b > c. 5. The theory in the text states that one of the formulas
c
b, c

+ by Let a,

3819 has a positive solution in integers. that a and b are of opposite sign, prove that many positive solutions for any value of c.

by

positive integers a specific solution [c/ab]. Suggestion: if c/a is an integer, then prove that 0. c can be found easily, for example, x l of ax + by c/a,

Prove that neither formula is correct in all cases. be positive integers such that (a, b) = 1. Assuming that c/ab is of solutions of ax + by = c in not an integer, prove that the number 1. or + is [c/ab] [c/ab] Assuming further that cfa is an integer,
is

correct.

6.

Let

a, b, c

N=

=
1.

7.

Let

a, 6, c

= 1 4- c/<zZ?. that integer, prove 8. Modify the theory of this section so as to treat the number, say JV , of solutions in non-negative integers of ax by = c, where a, b, c are positive that t is restricted to the range observe c. Suggestion: integers such that (a, b) value for t is -[Qf/^K] and allowable The smallest * a (gl )yi-Qr/^K ^ the largest value is [(#/fl)2/i]. The final inequalities turn out to be
an

be positive integers such that

(a, b)

^=

Assuming that c/ab

is

*r

+1-

[S]

104
9.

Some Diophantine Equations

Let a and b be positive integers satisfying (a, b) = 1. Consider the set S of + by], where x and y range over all non-negative integers. Prove S contains all integers greater than c = ab a b, but not c preceding problem, restrict x and y to be positive integers, yielding {ax + by}. What is the largest integer missing from the set S'?

integers {ax that the set

itself.

10. In the

the set S'

5.4

Other Linear Equations

Consider the equation

(5.2)

a^ + a x +
2

---- a x k k
\~

c,

k
,

>
,

2,

and

let

g denote the
'
'

greatest

has a solution, then clearly , y k such that 2/i, 2/2,

ak ). If the equation common divisor (a l9 a 2 c. 1.5 there exist Theorem g Conversely, by = a a + kyk g.Ifg\c then c = gr a^ + 2y2 +
\

some integer r and x1 Thus (5.2) has solutions

for

ry l9 x 2
if

n/ 2 ,
if

xk

ryk

is

a solution of

(5.2).

and only

c.
\

To
(a l9

find solutions of (5.2)

we

reduce

it

to the case of Section 5.2 with

two unknowns.
a 29
'

ak )

(5.3)

We can suppose that the c. We write xk = #_! = aw + fiv,

a t are not zero, and that

yu

dv

where we

shall
will

Then we

integers if

choose integers a, have u = <3^-i and only if xk_ly xk are.

^ we
If
,

/?,

y,

<5

in such

an d

=
take

a way that

a<5

/?y

=
z;

1.

7%-i

<*#&

Thus,

w,

are

=
then
(/S, <5)

6=
fly

and we can solve ad

for a, y

by the method of

Section 5.2. However,

we need only one

reduces to
-

pair of values for a, y, not the

general solution.

Equation
(5.4)

(5.2)

now
z

a^ + a x +
2

+ ak_^k_ +
2

(ak

_^ +

aky)u

c,

with one

less

unknown, and we note

that

(a l9

a2

ak_ 2 (ak^ l9 akj)

,

(a i9 a 2 ,-

ak).

Thus the new equation,

(5.4), has the same properties as equation (5.2), that the g.c.d. of its coefficients divides c and that no coefficient is zero. If k 3 this reduction process can be applied to equation (5.4) to produce an

>

5.5

The Equation x2

+ f = z2

105

2 variables, and hence repetitions of the process finally equation with k an to lead equation with two unknowns. Furthermore we may note from Section 5.2 that if a linear equation in two unknowns has a solution, its general solution is given in terms of a single
parameter in terms
t.

ofkl parameters. This can be proved by induction on k. For if

(5.4), in

Similarly the solutions of (5.2), with

k unknowns,

are expressed

any equation such as

u in terms of k

2 parameters, v l9 v 2 ,--' xk_ 2 <xu of (5.2) are given by x l9 * 2

,

k
the

form x i

parameters v l9 v 2 = b t + dii3 Vi
,

xk_ z unknowns, has solutions xl9 x 2 v k_ 2 then, by (5.3), the solutions + f)v yu + dv. These involve the v. It is -2> easy to see that the solutions have diiZ v z + + ^.fc-i^-i where we have written
1
, ,
, y
,

t?

fc

v k_i for v. If

we

are asked to solve a system of s equations in r

unknowns,
-

(5.5)

a^&i

ait &*

+
first

aJiTxr

ci9

j
1.

=
it

1, 2,

, 5-,

we

by solving the be of the form

start

equation, j

If

has a solution,

it

will

We substitute these in the remaining equations and solve the second equation,
j
this process allows

2 new parameters. Repetition of v r_! in terms of r us to solve the system (5.5). Of course, if we encounter an equation that has no solutions, then the system (5.5) has no solutions.
for v l9 v%,
-

= 2,

PROBLEM
1.

Solve the equations

(a)
(c) (e)

x + 2y + 3z = 5x - 2y - 4z = 3x -6y +5z =

(b)
1

11

(d) 5x (/) 6x

+ 2y + 3z = 10 - 2y - 4z = 10 + 4Sy - l$z = 5

5.5

The Equation x 2

z2

We

2 z wish to solve the equation x 2 y = ^ in positive integers. Consider 2 such a solution a, y, 2 and write g for (z, y). Then g z* and hence g z, and in general, we see that (x, y, z) since (x, y z) g. By ((ar, y), 2;) holds

symmetry we have (x

y, z)

(x, y)

(y, z)

(x 9 z)

=g

and

(s.i). W g^

(I..')
*/

_(;,)_,. ^ s/

106

Some Diophantine Equations

A solution #1,2/1, z\ having the property that these three are relatively prime
in pairs is called a primitive solution. Thus every solution x, y, z can be written in the form gxl9 gyly gz^ where x i9 y ly z is some primitive solution.
if g is

Conversely, if x l9 y ly z 1 is a primitive solution, then gxl9 gy ly gz 1 is a solution a positive integer. Thus we need look only for primitive solutions, and

this

we now

do.

x and y cannot both be even. They cannot both be odd either, for if were we would have x 2 = 1 (mod 4), y 2 = 1 (mod 4), and therefore they
z2

Now

2 (mod 4), which is impossible. Since x and y enter the equations symmetrically we can now restrict our attention to primitive solutions for which y is even, x and z odd. Then we have

(5.6)

^r*^

Now
lz

22
x
x

+x
'

+
+

and
z

Iz

and therefore

This with equation (5.6) shows that

(z

x)/2

r2

and

(z

some
y
is

also see that (r, s) r 1, r positive integers r, s. s, x r2 s 2 Also, since z is odd, r and s are of 2rs, z opposite parity, even, the other odd.

We

x)/2

=
2

s 2 for

>

s2 ,

one

and

Conversely, let r and s be any two integers such that (r, 2 2 ,y of r sy 2r,s, z opposite parity. Then if x

s)

=
r2

r
2
,y

>s>
,

0,

we have

#> y? 2 positive

and

1 and that y is even. easy to see that (x 9 y) Therefore, x, y, z primitive solution with y even. Thus we have the following result.

It is

is

Theorem x = r2

5.1 s
2
,

The positive primitive solutions ofx 2 + y 2 = z 2 with y even are 2 2 y = 2rs z = r + 5- w/zere r a/zJ s are arbitrary integers of
9
,

opposite parity with r

It

>s>
=

0/2^ (r, j)

1.

defined

may be noted 2 byx = r

that if r
s 2 ,y

and

s are

2rs, z

r2

any integers whatsoever, then x, y, z + s 2 give a solution of # 2 + y 2 = z 2

,

but not necessarily a primitive solution.

5.6

The Equation x4

= z2

107

PROBLEMS
1.
2 2 2 < z < 30. Find all primitive solutions of a? + y = z having 2 2 = z2 then one of x, y is a multiple of 3 and one of Prove that if x + y
,

2.

x, y, z
3.

is

Any
all

a multiple of 2 solution of x
is

5.
2
2

triple

because there

Find
(b)

+ y = z in positive integers is called a Pythagorean a right triangle whose sides have corresponding lengths. arithmetic progression, Pythagorean triples whose terms form (a) an
is

4.

a geometric progression. If n is any integer ^3, show that there

its

a Pythagorean

triple

with n as one

members. 2 2 y = n! 5. For which integers n are there solutions to the equation x 2 = x form n + y2 z2 the 6. Prove that every integer n can be expressed in 4 2 = 2 = 1. with z solutions x has y,z) that (x, +y 7. Prove infinitely many 2 2 2 = 8. Show that all solutions of x + 2y = z in positive integers with (x, y, z) 2 2 2 2 = = v are where u and u + 2v 2u \, y 1 are given by x = \u 2uv, z that u is odd and (u, v) = 1. Suggestion: any arbitrary positive integers such 2 - x2 = 2 (mod 8), which is solution has y even, because y odd implies z Hence x and z are odd, and the proof of Theorem 5.1 can be used
of
.
9

impossible. as a model.
9.

to an isosceles right triple of integers belongs are infinitely many primitive Pythagorean triples for that there but triangle, which the acute angles of the corresponding triangles are, for any given e of ?r/4. positive e, within

Prove that no Pythagorean

5.6

The Equation x 4
now

z2

integers

z 2 in the impossibility of solving the equation x* y* The argument is indirect, in that a solution in positive positive integers. is assumed, and this will lead to a contradiction. Assuming at least

We

prove

denotes a positive solution such that no other positive solution has a smaller value of z. The contradiction is obtained with a smaller z value. First by deriving another positive solution in integers 1. For if a prime p divides each of a;, y, z we establish that g.c.d. (x y, z) 2 4 4 then + t/ ) so that p* z and p 2 z. It would follow that (x/p)* 2 2 the minimal character of z.

one solution, we presume that

x, y, z

/ O
=
1

and y were odd we is impossible. which would have z == 2/ = 1 (mod 8) and so z = 2 (mod 8) that one of a; conclude and y were even then z would be even, and so we If and x is even y is odd, and and y is even and the other odd. We presume that
In particular
4

(yip)*

(zjp

we

contradicting see that x, y and z are not

4
2

all

even. If a;

note that z

is

also odd.

108

Some Diophantine Equations

Next we write y*
are relatively prime.
,

(z
if

)(z

For

divides

x 2 and z and observe that z 2 2 x and z + x it divides both z

)
.

x2

their

4 2 product 2/ their sum 2z and their difference 2x This is impossible because 2 x and z + # 2 are fourth powers y is odd and g.c.d. (x, y, z) = 1. Hence z 4 4 = 1. Thus we have of positive integers, w and say, with (w, z?)
i;

x2

u\

x2

v\

(v

u 2)(v 2

u 2)

= 2z

2
.

x 2 are odd, and hence v 2 + u 2 = 2 (mod 4). Also no odd prime p divides both v 2 u 2 and v 2 + u 2 since 2 2 2 = 1 and so v u is a perfect square, where y + w 2 is twice a square, (u, v) 2 2 2 2 = u a u + u 2 = 2b 2 with positive a and A. By Theorem 5.1 we say t? 2 2 2 = r + s u = r2 2rs where r and are positive integers. s and a get v

Now u and y are odd because 2

,

z 2 and z

5-

Now v 2 +
a;

u2
z2 ,

+
+

4
t/

=
w

Now
w4
v*

=
2

2b 2 implies r 4 + s 4 = b 2 and since this has the same form as we have a contradiction if we prove that z > b. 2 v = 1 is impossible because this implies x = 0, and so
,

>

y2

and we have

which completes the proof of the following proposition.

Theorem 5.2 solutions x =

77ze
0, y, z

ow/y integral solutions of x*

2
2/

#rf a, y

+
#
2
.

y*

z2

are /Ae

0, z

The method used

in the

proof of

this

theorem

is

sometimes called "proof

by descent" or "Fermat's method of infinite descent." This type of proof, which also occurs at other places in number theory, is based on the principle that every nonempty set of positive integers contains a least element.

The
z
4

fact that x*

y^

z2

has no positive solutions implies that x*

y*

has no positive solutions. This is a particular case of a famous statement of Fermat, in which he asserted in a marginal note that he could prove that for

n z n has no solutions in 2 the equation x n y every integer n integers other than the trivial ones in which at least one variable is zero. This proposi

>

still a conjecture for many values of n, is known as Fermat's theorem or Fermat's big theorem, as contrasted with Fermat's little theorem (Theorem 2.7). In addition to the case n 3 is 4, a proof for n

tion, although

last

given in

9.10 using simple ideas in algebraic

number

theory.

PROBLEMS
1.

For any positive integer n solutions with xy 0.

Prove that x*

(mod

4),

n prove that x

+ y n = zn
0.

has no

2.

^ + 4y 4 = z 2

has no solutions with xy

5.2.

Suggestion: use

the

method of proof of Theorem

5.7

Sums
3.

of Four and Five Squares

4
2/

109

Prove that #

4 the equation implies 2

= +

z2

has no solutions with yz

0.

4(xy)*

= O 4 + y 4) 2

Suggestion: note that

4.

whose
5.

Consider an integral right triangle, that is, a right triangle the lengths of sides form a Pythagorean triple. Prove that the area is not a perfect
<z

square.

Prove that there are no positive integers a and b such that both a b 2 are perfect squares.
2

b 2 and

5.7

Sums

of Four and Five Squares

is

Our aim

in this section

not to solve a Diophantine equation but only to

n has at least one integral x\ x\ x\ prove that the equation x\ n is a The solution may include zeros whenever solution positive integer. # 15 #2 # 3 , # 4 would be n where with 5 for as the #'s, example among

So as a simple extension of the four square 2, 0, 1, result we establish that every positive integer (apart from 12 exceptions between 1 and 33) are sums of five positive squares.
in

some

order.

Lemma
(5.7)

5.3

We
xl

have
x\

(xl

x*)(y*

y\

yj

y%

This identity, discovered by Euler, can be verified by just multiplying

out both sides.

and Y can be expressed as sums of four This identity shows that if O 2 and O2 O2 XY. Since 1 = I 2 squares, then so can their product 2 O 2 , we need prove only that every odd prime can be O2 I 2 = I2

+
p

expressed as a

sum of four squares. The proof can be broken up into two

Let
denote any odd prime. There
x\
is

steps.

Theorem
1

5.4

an integer
a?

m
a,

such that

< m < p andmp =

Consider the

x\

+ xl +
-

x\for some integers x^

2
,

# 3 #4
,

Proof.

sets

2 1. 22 52 consisting of -O 2 1, l)/2} 1, 1, -I -{(p 2 2 two that no we see or x + Since y), y} p (x y (mod;?) implies;? (x numbers of S^ are congruent modulo p. Also no two numbers of S2 are 1 of p integers. Since congruent modulo p. Now Si and S 2 together consist
-

Sl

consisting of O

2
,

2
, -

{(p

l)/2}
2

and

there are only

distinct residue classes

of iSu say x 2 2 x2 1 y

is

congruent modulo

(mod

p) 9

<

<

to

modulo p we see that spme number some number, y z 1, of *S2 Then

.

(p

l)/2,

<;

(p

l)/2,

and we

110

Some Diophantine Equations

have
x*

y*

mp,

l^m = ~(x

y*

1)

Theorem 5.5
Proof.

Ifm

is

the least integer satisfying

Theorem

5.4, then

m=

1.

There certainly is at least one such m. Ifm is even, then so is mp = %l + %l + %l + #4, and hence either none, two, or four of the x i are even. If exactly two of the x i are even, we can number the x i in such a way that # 4 are even, and x and # 2 are the even ones. Then in all cases x : #2 and x z

fa +
Therefore

a
a? 2

\~2~)
Now
For
(5.8)
1

[X-L

* 2 \2

\~T~7
even.

m is not least if it is
4,

consider the possibility

i

m>

1.

Since

is

odd we have

^ m < p.

^ ^

we

define

numbers y i by

yt
j/J

x{

(mod m),
a?J

-^
x\

^%^^
+
n
x\

Then
%l

XB

+ y\ + y? + y\ == + ^ + + %l = mp, and we can write

(mod m)

since

a?J

m J / < m. m 4(\ 2 If were zero, we would have y = y = y = y = by (5.9), and then = = = = xl + = This would 3 x^ #! (mod m) by (5.8). imply mp = = which m hence xl + x\ + x\ (mod (mod m), ), p impossible since 3 ^ m <p. Therefore we have n > 0.
(5.9)

y\

y\

y\

y\

mn,

^
a

^
z

TI

o;

o;

is

Using
(5.10)

Lemma

5.3

we

see that

m*np

= =

+ x\ + A\ + A\ +
(x\

+ x%(yl + AI + Al
x\

y\

y\

&
z

where the
of
(5.7).

denote the expressions whose squares appear on the right side

(5.8)

Using

Dividing (5.10)

we easily by m we get
2

find that

A =
t

(mod m)

for

1,

2, 3, 4.

with

<

<
is

m. This shows that

proved.

is

not least

if

m>

1.

Hence

and

the theorem

5.7

Sums

of Four and Five Squares

111

2 2 l 22 requires Collecting our results and noting that 7 the theorem. first sentence of the obtain we four squares, following
2

+l +l +

Theorem 5.6

Every positive integer

is

four is a sum of Jive positive squares of integers;

replaced by "four."
It is

in general. squares will not suffice

a sum offour squares, and fewer than Every sufficiently large positive integer
this result is false if "five" is

about possible to be quite specific

what is meant by

"sufficiently large"

exceptional positive squares. with ease, and on the other hand it is readily checked that all the rest of the to 169 can be expressed in five square form. We now positive integers up this holds for every integer ^170. that prove 169 as a sum of four squares by the For any integer n 170, we write n

here; in fact, every positive integer except 1, 2, 3, 4, 6, 7, 9, 10, 12, 15, 18, cases can be verified These 33 is a sum of five

first

part of

Theorem

5.6,

169 =: x*

+
2

z2

+w

2
.

We may
we
>v

+w + + x + y* + z Third, if x 0, we write n = = n = 12 + 4 + 3 + z + y that note we w but z 0, = 10 + 8 + 2 + + x n we the only positive one, get
write n

presume that = 13 2 + x2

#^?/>z^w^O.
+
y
2

z2

If these integers are all positive Second, if x 9 y, z are positive but

2
.

12 2

52

2
.

y are positive Finally if x is

In every case n
it is

has been expressed as a sum of five positive squares. To complete the proof of Theorem 5.6 we must show that
every sufficiently large positive integer that is, we must show that there are infinitely
is

false that

sum

of four positive squares;

many exceptions. To do this, we the that k, first prove integer 2k is a sum of four for any positive integer x2 sum. a is such &k and Assuming that 2k only if positive squares if
z

= + assume = + + + Conversely, (2w) (2z) + (2y) (2s) y + Then u, v, s, are all even because otherwise that 8r = u + v + s + == would be false that u + v + s + (mod 8), in view of the illustration = n u = 2m + l, (2w + I) = (mod 8). = It follows that 2r (w/2) + (v/2) + (s/2) + (*/2)
Z2

W 2 weget 8fc
2 2

it

of integers, Next we observe that 8 is positive squares four of sum = a is not 32 = that we conclude 32 and taking 2r positive 8, 8r infinite chain of integers 8, 32, 128, 512, squares. This leads to an which are not sums of four positive squares. 22c+i

not a

sum of four

PROBLEMS
1.

Prove that no integer of the form Sk

7 can be expressed as a

sum of three

squares.

112
2.

Some Diophantine Equations

Prove that no integer of the form 4 m (8k + 7) is expressible as a sum of three squares. (Remark: these are the only integers not expressible as sums of three squares, but the proof of this result is not given in this book.)
3. Verify that 169, every positive integer listed following Theorem 5.6, is expressible as

apart from the 12 exceptions a sum of five positive squares of

is

integers. 4. Prove that every positive integer, apart

from a few exceptions,

expressible

as a

sum of six

positive integral squares.

What

are the exceptions ?

5.8

Waring's Problem
as to whether the results of Section 5.7

Waring considered the question

can

conjectured that 9 cubes, 19 fourth etc. suffice. Hilbert would powers, proved that for each positive integer k there is a number g such that g fcth powers will suffice. If g(k) denotes the
least

be generalized to higher powers.

He

seen. It

number of kth powers that will suffice, then g(2) = 4, as we have = 9, and the value ofg(k) is also known for k is known that ^(3)

just 6

apart from a possible exceptional circumstance relating to the fractional 4 and 5 it is known that 19 g(4) rg 35 and part of (3/2)*. For the cases k 37 the 54. for and known However, except g(3), g(5) proofs of g(2) these results involve much more complicated methods. The proofs lean

heavily on the theory of functions of a complex variable, and they belong to the part of number theory called the analytic theory of numbers. There is a good account of the history of the problem in the notes at the end of

Chapter 21 of Hardy and Wright's An Introduction Numbers.

to

the

Theory of

PROBLEM
Prove that g(k) ^ 2 4- [3 /2 ] - 1. Suggestion: n = 2 k [3 k /2 k ] 1 as a sum of kth powers.
1.
fc fc

fc

express

the

number

5.9

Sum

of Fourth Powers
result

There

is

one small

obtain simply.
g(4)

We

will

<

connected with Waring's problem that we can 50. This is far from the known result, prove g(4)

35, but
it

because

of some interest because the proof the existence of g(4). proves

it is

is

elementary, and

5.10

Sum

of

Two

Squares

113

Summing
find

the identity (x i

+ x^ +

(x i

~ x^ = 1x\ +

\2x\x}

2x]

we

((*<

^ + & - x^

t=l

4-1 3=1

-622^z=l 3=1

\2

"5 2*5

H=l

The

left side is

every

sum of 12 fourth powers. This with Theorem 5.6 shows that number of the form 6m 2 is a sum of 12 fourth powers. But Theorem
a

5.6 also

sum of

shows that every positive integer of the form 61 can be expressed as four numbers of the form 6/w 2 and hence as a sum of 48 fourth
,

powers.

Furthermore the integers./ = 0, 1, 2, 81, 16, 17 form a complete residue Then system modulo 6, and each one is the sum of at most 2 fourth powers. 67 +j with / positive, and in the form can be written n 81 > every integer hence can be expressed as a sum of at most 50 fourth powers. For < n ^ 50 we have n = 22* 14 and for 50 < < 81 we have n = 24 + 24 + 24 +
>

In

all

cases n

is

expressed as a

sum of at most 50

fourth powers.

5.10

Sum

of

Two

Squares

Not every positive integer is a sum of two squares. We will determine just which integers are sums of two squares and will find the number of solutions of x 2 + y 2 = n. A solution x, y of x2 + 7/ 2 = n will be called primitive if (#, y) = 1. We
restrict

n to be positive, and we

let
2
2

7V(72)

P(n) Q(n)

= number of solutions of x + y = n, = number of non-negative, primitive solutions of x* + y = n, = number of primitive solutions of x + y = n.

2

114
,

Some Diophantme

Equations

In counting solutions we consider x l9 y l and x 2 y 2 as distinct if x 1 x 2 or It be noted for n 1 is the 7^ number > of 2/i 2/2that, P(ri) may actually 2 2 a; x of solutions since neither nor can be zero. + n, y y positive primitive
,

Theorem 5.7

We

have N(l)

g(l)

= = 4, P(l)

=* 2, and,

for

n>!

Q(ri)

4P()

Since
tions,
If
72

==

(1) 2 +

O2

we have

JV(1)
x,

>
0;,

and
y
is is

and

Q(l) a non-negative primitive solution, then x a primitive solution for all choices of the signs.

+ (1) = 4 and P(l) =

O2

and there are no other solu

2.

is

1,

^
2
\

1,
it

From
then

this

follows that Q(n)

If x,
(

4P(ri).
if

xlg> ylg)

2 2 any solution of x + y = 72, and 2 2 = 2 1, and (x/g) + (y[g) nig From

.

=
it is

(x, y),

n,

this

easy to see that

Theorem 5.8 Suppose n > 1. .E^c/z non-negative primitive solution of x 2 + y 2 = n determines a unique s modulo n such that sy = x (mod n). Further
2 more, s

(mod

n),

and

different non-negative primitive solutions de

termine different s modulo n.

Proof.
If x, y is a non-negative primitive solution, then (y, n) = 1 and hence sy == x (mod /x) determines a unique j modulo n. Furthermore if y' is a solution of yy' 1 (mod ri), then s = xy (mod ), and we have s 2 =

y'

= -2/V = -1
2

(mod/i).
that different solutions determine different
s.

We
sy

must

still

show
u,

Suppose

that x, y

and

sx

(mod

are both non-negative primitive solutions and that == u n) and sv (mod ). Then we have xv =~ ^y yw (mod ri).
v

=
,

any non-negative primitive solution is a positive solution and so we have 1 ^ x < V, 1 <j < V- Hence 1 <i zy < and similarly I ^yu <n. Therefore xv = = = = and x hence since yu, u, y v, (a?, y) (M, t?) = 1 and all the numbers are positive.
1
,

But since n

>

t?

Theorem 5.9

Suppose n

primitive solution

> 1, s = (mod n). TTzere w x,yofx + y = n such that sy = x (mod

2

a non-negative
ri).

Proof.

Consider the

set

of integers u

sv

where u and

run through
(1
,

all

integral values such that

O^w^V^O^y^ V^
^=

There are

+
.

[V])

>

n different pairs sv i = W 2 i

u, v.

^2

Therefore, there are two pairs u l9 v and u 2 v 2 such that z; w2 then (mod ). Let us set a Wl MO
,

We

5.10

Sum
5-z;

of

Two

Squares
ri)

115
|w

have

(mod

and

^
|t?
|

\v

and w 2

>

^2 are different pairs,

w and

z;

^ n. Furthermore, since u l9 v^ cannot both be zero. Also, we can

\

show that

at least

one of \u

and
|

is less

than *Jn. This

is

obvious

if n is
(

not

a square. If 72 is a square and w == 1 (mod and hence j s -Jri), s*

1

\V Q
I

=
\

\Jn,

we have j>/
But s 2

Vw m
1

d/2) 3

(mod
1

^ri).
1

(mod

72),

so

we

have s

and

7z

= = 4,

(mod

*Jn),

and hence

but this cannot occur since there

for w

(mod V)- Therefore V = 2 is no integer s such that s ~

2
I 2
.y

-1 (mod

4).

The bounds
congruence ^y

and V Q imply the inequality

^u% +

VQ

<
2

2n.

The
1)

(mod
).

72)

implies ul

(mod

72).

Now let ^ =
g

Together these imply ul

(t/o,
i?

v\

+
72.

VQ

?;^

+# =
2

f?^

Then ^ 2

72
1

and

5-(y

/g)

(w /g)

(mod

72/g),

and hence

g/

\g/ g/

s. _fs.V +
\g/
(w
,

(a| \g/
)

=0 (mod
V

). g/

This

is

Finally, if
i?

w and V Q have the same sign we let x = |w |, y = |y |. If w and = l^ol- I n fc ot h cases we see that x, y have opposite signs we let x |u |, y

possible only if g

1,

and we have

t;

1.

is

a non-negative primitive solution. In the first case we have sy = w = x (mod 72). In the second case we have sy = s(w ) = =f i; = x (mod 72).

s(v } =
^(^^0)

The last two theorems show that there is a one to one correspondence between the non-negative primitive solutions of # 2 + y 2 = n and the solutions of the congruence s 2 = 1 (mod 72). Combining this with Theorem 5.7 we have the following.
Theorem 5.10
Then
P(ri)

Let

R(ri)

denote the number of roots of

1,
tf .,

s*

forn^l,

R(n)for n > and N(n) = 4

P(l)
w

2,

R(\)

1,

g(l)

= 4,

(mod

72).

Q(n)

4R(n)

Theorem 5.11
Proof.
2.18.

The functions R(n) and

is
is

N(ri)/4 are multiplicative functions.

The fact that R(n) To show that N(n)j4

prime integers
72!

multiplicative follows directly

multiplicative,
72 2 .

from Theorem

we

consider any two positive

relatively

and

Then

AXi) 7 7 4 4

116

Some Diophantine Equations

Leth(V)

~ 1)/2 } e (33 (_l){ p an odd prime, h(p ) Let a n be determined in such e^.1. h(n) for composite way that h(n) is a

Theorem 5.12

1,

h(2*)

0,

multiplicative function.

Then

N(ri)

Proof.

It is interesting to

note that h(n) as defined above

is

actually totally

multiplicative.

By Theorem 4.4, ^\ n h(d) is multiplicative. Since N(ri)/4 is also multiplica we have only to verify that N(n) = 4 ^>, d n h(d) for n a prime power. The 2 = = 1 (mod 2). For e > 1 1 (mod 2) has the one solution s congruence s e 2 2 = = 1 (mod 4) the congruence s 1 (mod 2 ) has no solutions since s has none. Therefore by Theorem 5.10,
tive,
\

N(2

2^ R(2

~ 2f
)

since nonzero terms will occur only for e

2f=

or

1.

Correspondingly

we have

=
We now
consider an

4
/=0

odd prime p. From Theorem 2.11 we see that s2 = 1 has two solutions if p ^ 1 (mod 4), and no solutions if (mod p) = x 2 + 1, with = 3 We (mod 4). p apply Section 2.6 to the polynomial /(#) e 2 = = = 2#. Since 1 that S we find 1 (2s, p) (mod p ) has the same f'(x) number of solutions for all e ^ L That is
r\ f 10 if
\

]?

_ =

A\

(mod

4),

Then, by Theorem 5.10,

e

if e is
e/2

even

we have

(5.11)

N(p

=
_

J ^(P
/=o

"2

=
= s
1

^ 2

J?(p)

f4e
14

if

(mod (mod

4)
4),

if

and

if

is

odd we have

(5.12)

N(/)

=4
4e

-2

(p

=
1

-4)

R(p)

+ 4ifp= = if
j>

(mod

(mod 4).

5.10

Sum

of

Two

Squares
this

117
also

Corresponding to
(5.13)

we
4

have

h(d)

KP f) =

4/z(l)

(-l
4)
4), e
4), e

+
4

4e

if
if

== 1

(mod (mod (mod

(4

if j7

= =

3 3

even
odd.

comparison of (5.11) and (5.12) with (5.13) completes the proof of the theorem.
Corollary 5.13
N(ri) is
1

four times the excess of the number of divisors ofn

of the form 4j

1 , then h(d) 1. If d is even, then h(d) 0. If d is odd and Proof. J is the product of the primes p l9 p 2 not k ,p necessarily distinct, then 1 or 1 h(d) according as an even or an odd number of the p i are of the
, ,

+ = If d +

over those of the form 4j

3.

form 4j
N(n)

3.

But

also, in this case,

according as an even or an odd

d = p^p 2 -p k = I or 3 (mod 4), number of the p are of the form 4j + 3. Since

t

4 2d|n h(d), the corollary follows.

2 2 n is solvable if and only if the Corollary 5.14 The equation x y canonical factoring ofn into prime powers contains no factor p* with p of the 3 and e odd. form 4j

Proof.

This follows at once from (5.11) and (5.12) and Theorem 5.11.

PROBLEMS
1.

2.
3.

Evaluate N(ri), Prove that if n

P(ri) 9
is

and

Q(ri) for

100, 101,

and

102.

square-free, N(ri)

Q(ri).

Prove that the number of representations of an integer m > 1 as a sum of two squares of positive relatively prime integers equals the number of solutions of the congruence x 2 = 1 (mod m).
divisors of the

4.

Corollary 5.13 implies that every positive integer has at least as many form 4j + 1 as of the form 4/4-3. Prove this fact directly.

5.

For a given

(a)
(b)
(c)

N(n) = K if and only if K = (mod 4); = K if and only if K has the form 2 m with m ^ 0; P(ri) m Q(n) = K if and only if K has the form 2 with m ^ 2.
if

positive integer K, prove that there

is

an integer n such that

6.

Prove that

an

fi()
7.

= o.
a2

integer n
is

is

divisible

by a prime of the form

4fc

3,

then

n
c2

+ b z with (a, b) = L +d z = ? with(c,d) = 1.

Suppose that q

any positive divisor of n, and that n is expressible as Prove that there exist integers c and d such that

118

Some Diophantine Equations

5.11

The Equation 4x 2

The ideas and methods of Section 5.10 can be greatly extended. The quadratic form x 2 + y 2 is only a special case of an extended theory. However, it is probably the most interesting case, and we will do no more than consider a
few direct consequences and applications. In this section we will restrict our attention to positive n = I (mod 4). If x 2 + y 2 = n, then one of x and y is odd, the other is even. If a; is even, we let u = x/2, v = y; if y is even, we let u = y/2, v = x. In both cases we have 4u 2 + v 2 = n. Since x 2 -f y 2 = y 2 + x 2 and x ^ y, we see that exactly two solutions of x 2 + y 2 = 77 correspond to one solution of 4 2 + v 2 = n. Also
(#, y)

(2u, u)

the equation 4x 2
n,

= (w, y) +y =
2

since

t?

is

n, just as

odd. If we define N'(ri), P'(n), Q'(n) for we defined N(ri), P(ri), Q(n) for x 2 + 7/ 2 =

we have

n be an integer, n > 1, n 1 (mod 4). If n is a prime, n has exactly one non-negative solution, and it is a primitive y 2 solution. Ifn is not a prime, then 4x 2 n has either no primitive solutions, y

Theorem 5.15
then 4x 2

jCer

more than one non-negative primitive solution, or it has one non-negative primitive solution and at least one non-negative nonprimitive solution.
Proof.
If n
is

a prime,

we

use Theorems 5.12 and 5.10 to obtain

=
Therefore 4# 2

4,

Then, changing signs of u and v, we find three other solutions. Since N'(n) = 4 we see that 4x 2 + y 2 = n has just one non-negative solution, and it is
primitive. If n is composite,

n has just one non-negative primitive solution, say u,

v.

and

if

Theorem

5.10,

Q'(ri)

some prime p

== 3
0.

Q(ri)/2

=
2

2jR(/i)

(mod

4) divides n, then,

Thus 4z 2

= -

by

72

has no

= R(n)j2 =
If n

primitive solutions in this case. If n EE 1 (mod 4), e t pfyl* />,

-

2 1
7

^,
p

>
>

and 4x2
==
1

=n

> 0, r 1 then P'(n) P(n)/2 has more than one non-negative

,

>

primitive solution.

=p

e
,

1,

(mod

4),

then

/=o

5.11

The Equation 4x 2
P'(ri)

+f =
e

119

and

= P(n)/2 =

R(p

)/2

1.

Therefore 4x 2
it
,

n has just one

It

non-negative primitive solution,

and

has more than four solutions.

must

have some nonprimitive solution, solution. This completes the proof. primitive The problem of deciding whether or not some number
v.
|w|, |y| is

Then

a non-negative non-

is a prime has always can been large, present quite a problem, many. of methods and machines. No various the led to has it and development there is someone these far how matter trying to push a little go, always n = 1 (mod 4), we a Given criterion for 5.15 is Theorem further. primality.

of interest to

If the

number is

it

2 n. If we find a nonprimitive look for non-negative solutions of 4x + y* we can or two solution non-negative solutions, stop looking further; n is If we find no solutions, n is again composite. If we find exactly composite. one non-negative solution, and it is primitive, then n is a prime.

n we not only know find a nonprimitive solution x, y of 4x y that n is composite, but also we know a factor (x, y) of 77. If we find two nonv we can also find a factor of n. Since negative primitive solutions u, v and \JL, 2 2 different are v and n, v y 2u, non-negative primitive solutions of z 2/z,
If

we

they determine different s and

such that sv
1

2u (mod n\
s

tv
ri)
9

2/z

(mod

n)

by Theorem
2( s

5.8.

Then

==

(mod

),

(mod

and

f)(uv

=
VJLC)

(s

i)($vv

vtv)

(s

00 + O y " s (s - )^ =
2
f

(mod

n),

which implies that uv

(uv

vp,

ri),

then since

vp and n have a common factor >1. If g denotes V/2 is not an integer,

and g
^ to

is

When
square.

a proper factor of TZ. 2 = 2 looking for non-negative solutions of 4x + y a; < V/2 and we need only check whether /i
2

/z,

we can
2

restrict

4a;

But we don't even need to

try all these values of x.

a perfect Since y is odd,

is

we have y
which is
re

(mod
x2

8).

odd if n = must be tried.

is

equivalent to
5

=
8).

4x2 Therefore x must satisfy n (n l)/4 (mod 2). Thus x is even if

In either case

==

1 1

==

(mod 8), (mod 8),

(mod

we have

halved the number of x that

4x 2 = y 2 (mod c), and we odd and positive then, n 2 4x is a quadratic nonresidue modulo c. For can exclude all x for which n modulo 5, and we can exclude x for which are nonresidues and 3 2 example 2 x 2 = 2 - n or 3 - TZ (mod 5). If n s 3 is == that or 3 2 4x n (mod 5), 2 = == 1 or 0, 2, (mod 5), that is x (mod 5) this excludes a such that x

Going

further, if c is

(mod

5).

120

Some Diophantine Equations

= 0, 2, 3 (mod 5). Writing down the even == 34 and 0, 2, 4, 0, 2, 3 (mod 5), we find scratching out the a that we need test only x = 4, 6, 14, 16, 24, 26, 34. This list can be cut down further if we use other values of c. For example, c = 7 eliminates 6 and 34,
numbers
,

Let us consider a simple example. < x < 71/2 36. Since n to n s= 3 (mod 5) we can exclude all re

<

We take n = 4993. = I (mod 8), we

Then we can

restrict

take x even. Since

c 1 1 eliminates 14. This is hardly worthwhile for an n of this size. In event we have a few to check. It will be found that x values 16 any only 2 n 4x* 3969 63 and that there are no other solutions. Therefore gives

and

4993

is

a prime.

larger values of n one would probably use more values of c. If one is to use this method very much, it is advisable to make short tables going x are excluded by various values of c. If the method is carried which showing

For

out systematically, and if a table of squares primality of n that are not too large. This
4x*

is

available,

it is

a useful test for

test

n, and it is useful only for n = I based on other equations, valid for other n.
2

+y =

was based on the equation (mod 4). There are other tests,

5.12

The Equation ox 2

by

cz 2

Although the theorem we give concerning this equation goes back to Legendre, our proof is a recent one, adapted from a paper of Mordell (see the Notes at the end of this chapter).
be nonzero integers such that the product abc is that ax 2 + by 2 cz 2 have square-free. a solution in integers x, y, z, not all zero, are that a, b, c do not have the same
a, b, c

Theorem 5.16

Let

Necessary and sufficient conditions

be, ac,

sign,

and

that

ab are quadratic residues modulo a, b,

c,

respectively.

Before giving the proof of this result

we

establish

two lemmas.

Lemma
an

5.17

integer.

Let k, //, v be positive real numbers with product hp,v = m Then any congruence ax + 0y + yz = (mod m) has a solution
\x\

x, y, z, not all zero, such that

<* A,

\y\

^ p,
-

\z\

v.

Proof.
z

Let x range over the values 0, over 0, 1, , |>]. This gives us

-

1,

[A],

y over

0, 1,

[u],

and

(1

triples x, y, z.

Now
-f

(1

[A])(l

[>])(!

+ [u])(l + + M) > fyv = m

[A])(l

[v]) different

4.la and hence there must be

that
**}
oca?!

+
<;

P(Vi
11,

^ -

some two

yz l

triples
2

xl9 y l9
777).

z1

by Theorem and x 2 y 2 z z such

, ,

y2

fa

2/ 2 )

z2

+ Xzi < v.
|

+ ftfe + yz - ^2) s (mod

az 2

(mod
m),

fa

Then we have afo

a?

[A] <; A,

\y

5.12

The Equation ax2

by

cz 2

=
by*

121

Lemma 5.18

2 Suppose that ax is that and also modulo n;

cz 2 factors into linear factors

modulo

ax 2

ax

+ +

by by

+ +

cz 2 ==
cz
2

fax
fax
by
2

=
2

+ ^y + + fi& +

y^z)fax

yzz)fax

+ ^y + + p# +
y, a',
7'

y&) (mod m),

y 4z) (mod
n).

lf(jYi^n)^\ then ax
Proof.

cz factors into linear factors

/3,

modulo mn.
y' to satisfy
)>
/z).

Using Theorem 2.14, we can choose a,

a
a

ft',

= =

a l9 a3
,

j8

s =

j?!,

y
7

==

y lf

a' ==

a2
a4

j8'

3,

=
2

73, a' ==

|8'

= =

/? 2 ,

/3 4 ,

= 72 m d / = 74 (mod
(

Then

the congruence

holds

+ by + cz == modulo m and modulo

ax 2
2

(as

+
2

j8y

+
cz
2

yaf)(a'a?
it

+ fy +

y'z)

/z,

and hence

holds modulo mn.

has a solution
, ,

Proof of Theorem
all

5.16.

If

ax 2

+% +
same

= =
c

a?

>

2o

not
,

zero, then a, b, c are not of the

sign.

so there would be a prime and aZ?c is square-free. There p dividing both c and xv Thenp/pZ? since j? 2 then and fore p by\ and p ^ A, hence p\yl,p\ 2/i, p (ax\ + 6yJ) so that 2 that p is a factor of concluded have z. We czf. But c is square-free so p
|
1

we have a solution # 15 yl9 ^ with (^ 19 t/ ls Next we prove that (c, x-^ = 1. If this were not
1.
|

Dividing XQ y Q

ZQ

by (XQ y Q
,

Z Q)

rc

+ (mod c). 2 we u b this and get by (mod c), multiplying cz\ implies ax\ + by\ ab is a quadratic ab (mod c). Thus we have established that U 2 b 2 yl = ac are quadratic be and residue modulo c. A similar proof shows that b a and residues modulo respectively.
Let u be chosen to satisfy ux^
1

ls 2/ l9

and

^ contrary to

(a^,

yx

^=

1.

Consequently,

we have

(c,

x^

== 1.

Then

the equation ax\

by\

Conversely,

let

us assume that

be,

ac,

ab are quadratic residues

modulo a, b, c respectively. Note that this property does not change if a, b, c are replaced by their negatives. Since a, b, c are not of the same sign, we can
in order to have one positive change the signs of all of them, if necessary, a change of notation, we can with them of and two negative. Then, perhaps it so that a is positive and b and c are negative. arrange 2 ab (mod c), and a : as a solution of aa = Define r as a solution of r = on a, b, c. Then 1 (mod c). These solutions exist because of our assumptions

we can
ax 2

write

by

= aa^ax + by = fli^ + aby = a (a x - r y = a^ax - ry)(ax + ry) = (x - ary)(ax + ry) (mod c),
2 2
2 2
2
2

ax 2

by

cz 2

(x

ajy)(ax

ry)

(mod

c).

122

Some Diophantine Equations

2

Thus ax

+ by* + +
2

cz 2

is

the product of two linear factors

b.

similarly that ax 2

modulo a and modulo

by
2

+
2

cz 2
is,

modulo
(5.14)

abc.

That
by

Applying can be written as the product of two linear factors there exist numbers oc, /?, y, a', /?', / such that

Lemma

5.18 twice,

modulo c, and we conclude

ax

cz 2

(KX

+ +

@y

yz)(<z'x

{3'y

/z) (mod

abc).

Now we
(5.15)

apply

Lemma
ax

5.17 to the congruence

+
9

f}y

yz ==

(mod

a&c),
l9

using A

= V&c

/*

= V \ac\
\x
is

v
\

congruence (5.15) with

square-free, so -Jbc
\l\ab\. Therefore

= Vl^l- Thus we get a solution o^, y % of the ^ ^Jbc, \y\ ^ VW|, ^ Vl^l- But abc
\z^\

is

an integer only

if it is 1,

and

similarly for \J\ac\

and

we have
with equality possible only = be = "~ ac w ^ equality possible only ^ w ^ equality possible only =
"~"
fl

l^il

l2/il

= V^ c ^ Vl ac
'

xi
l'

2/i

fl

2i
is

Hence, since ^

positive

and b and

c are negative,

= c = 1, if a = 1, c = 1, if a = 1, b = 1. we have, unless b = c =
if

-1,
fla? i

by*

+
cz\

C2 i

ax i

and
fl^

byl

cz\

Leaving aside the special

> 6(-flc) + c(-afo) = case when b = c = we have

byl

-2abc

<

ax\

+
2

byl

cz\

<

abc.

Now xl9 y l9 z

l is

a solution of (5.15) and so also, because of (5.14), a solution of

ax 2

+
2

by

cz

(mod

abc).

Thus the above

axl

inequalities imply that

+ cz = -abc. In the first case we have our solution of ax + by + cz = 0. In the second case we readily verify that x y z defined by x = by + XjZ y = = = z form a solution. x In case + then + <z#i zl ab, 2/A, y = z = = = = ab and ab because like + is 0, z\ zj ab, ^ abc, square-free. = is a solution. Then = 1, = 1, and x = 1, y = 1, 2 The conditions Finally we must dispose of the special case b = c =
+
byl

cz

or

ax\

+
2

byl

2,

z,

2,

l9

Z>

1.

on a,

b, c

now imply that

Theorem
5.10

1 is

a quadratic residue modulo a, in other words,

positive.

that R(a) of

is

By Theorem

5.10 this implies that

5.13

Binary Quadratic Forms

is

123

Q (a)
A

positive

Then x

and hence that the equation y 2

1,

z2

2/1,

=^

is

a solution of ax 2

= a has a solution yl9 z. + by* + cz = since

2

= c=

-1.

PROBLEM
1.

Show

theorem

that in the proof of Theorem 5.16 we have established more than the stated, that the following stronger result is implied. Let a, b, c be
is

non-zero integers not of the same sign such that the product abc

square-free.

Then the following

(a)
(b)
(c)

ax 2 ax 2

+ +

by

2 2

three conditions are equivalent. 2 has a solution x, y, z not all zero; 4- cz

by

cz

factors into linear factors

modulo abc;
a, b, c respectively.

be,

ac,

ab are quadratic residues modulo

5.13

Binary Quadratic Forms

is

A form

variables,

all

a homogeneous polynomial, that is, a polynomial in several of whose terms are of the same degree. quadratic form

has terms of degree two, and

/(*!,

is

therefore an expression of the type

(5.16)

a?

2,

s n)

We
If

our attention to quadratic forms with integral coefficients o#. xn x n) assumes only positive values whenever x l9 x Z9 are replaced by any set of integers other than 0, 0, 0, then/is said to be & positive form. Similarly / is called a negative form if its value is negative
restrict

f( x i> x z>

'

'

when x l9 x 2

x n are replaced by integers not all zero. A definite form is 2 2 one which is either positive or negative. For example x\ + x\, or x + y in 2 2 x both other notation, is a positive form, and 3y is a negative form; 2 2 are definite forms. The form x y is an indefinite form. Sometimes called positive definite, and negative forms negative positive forms are
'
'

'

>

>

definite. Clearly if /is positive

then

/is

is no need to study both positive of the other. properties of one kind follow from the properties if there are an said to integer represent quadratic form (5.16) is m. For Z> that such , 6J , bn example, /(A l9 2 integers b l9 6 a ,

there

negative, and conversely. Hence and negative forms because the

because Every quadratic form represents = 4 form a zero called 0. The form/is *J = if/(A l9 2 0) /(O, 0, form a definite definition zero. not all A b for some integers 6 1} 2 n By conditions and sufficient 5.16 is not a zero form. Theorem gave necessary
x*
7/2

represents 5 but not

, *
'

6.

'

'

>

>

that ax?

by

cz 2 be a zero

form for a wide

class of integers a, b, c.

124

Some Diophantine
said to be universal if
it

Equations

A positive form is
Thus, according to

Theorem

5.6, x\

+ Although we will not prove it, it is 2 2 forms includes of arithmetic can universal. The cz be by + theory quadratic such problems as to determine which forms are universal, to determine or characterize the class of integers represented by a quadratic form that is not universal, and to determine how many ways an integer can be represented
by a quadratic form. For example, Corollary 5.14 determined the class of 2 2 and Corollary 5.13 determined integers represented by the form x + y the number of representations. A form involving two variables is called a binary form. The rest of this
,

positive integers. a universal form. x\ x\ a fact that no form of the type ax 2

represents

all

x\ +

is

chapter deals with binary quadratic forms, that

(5.17)
/(a?, y)

is

forms of the type

2
.

ax 2

bxy

cy

not intend to do more than just give an introduction to a part of the theory dealing with binary quadratic forms. The use of matrices would little that we shall do, the simplify a few of our proofs. However, for the
simplification is not enough to offset the introducing matrices.
9

We do

work

that

would be required
2

in

Theorem 5.19 The quadratic form f(x y) = ax 2 4ac is negative, a and only if its discriminant b 2
Proof.

+ bxy + cy > 0, and c >

>
0,

is

positive if

0.

Since /(I, 0)

a and/(0,

1)

c,

we

see that

/is not

positive if

or

if c

0.

We

can

now

suppose a

>

0, c

and can write

(5.18)

/(*, y)

4a

f- ((2ax

by?

(4ac

- &V)-

This shows that/( not positive if 4ac

that /(#, y}
is

b, 2d)

b 2)a and hence, since a (4ac 2 4ac 0. On the other hand, if b

> 0, that /is < 0, it shows

never negative no matter what integers may be substituted for and x and y. Moreover /(#, y) then holds if and only if 2ax by 2 2 x if a that and 0. These b )y 0, y (4ac equations imply 0. 4ac b2

>

<

The quadratic form x 2

is

2 dy with d
2

>

has discriminant 4d

>

0,

and

it

For fixed d and n it turns out that the equation has either no solution or an infinite number of solutions. On 2 = the other hand it is easy to see that x 2 y = n has no solutions if n 2 (mod 4), and only a finite number of solutions, x = (t + n/t)/2, y =
(t

clearly indefinite. Pell's equation x square, will be discussed in Section 7.8.

dy

>

0,

d not a

perfect

fl/0/2,

t\n,t

n/t

(mod

2), if

&2

(mod

4).

The

first

situation does

not arise in the case of definite forms, as shown in the following theorem.

5.13

Binary Quadratic Forms

125

Theorem 5.20

Let

be a positive quadratic form. Then the number of

an integer representations of
will Proof. for which f(x, y)

m by f is finite, possibly zero.

is only a finite number of pairs of integers can suppose m > 0. Using (5.18) we see that 2 2 Z> )y ^ 4am, and hence

We

show

that there

f(x

y)

^ = m implies (4ac
m.
2 ^1

We

/_m_f < =
27

2r

am

This restricts y to a
(2ax

finite

+ by) ^ 4am
2

(4ac

number of values. For each such y we then have 2 which then restricts 2ax + by and Z>%
,

hence # to a

finite

number of values.

Corollary 5.21

Let f be a positive quadratic form. Then the smallest positive can be found in a finite number of steps. integer represented by f
Since a
is
1 in the a proof of represented by/ we just take the least x and then determine that and are allowed, y

Proof.

m=

Theorem

5.20, find all

y).

value of /(a,

For example, \ff(x y) = 5z 2 + 14o^ + lip 2 then b* - 4ac = -24 < and 2/ is limited to - V30/3 ^ y ^ V30/3. Hence y = -1, 0, 1. For y = -1 = 2 or we have (lOa? - 14) 2 ^ 56 from which we find x = 1 or 2 f(x, 1) we find 3 = 0, /(O, 0) = 0. For y = 1 we find x = 2 or 3. For ?/ =
9

Therefore the least positive integer represented by/is 2. 5z 2 4- I4xy + ll?/ 2 we replace x by form f(x, y) 2Z2 3T2 Now 27 and y by -JT 7, we get the form F(X, 7) the transformation
1

,/0,

1)

3 or 2.

If in the quadratic

Z*

(5.19)

= _z + 27,
Y to

2/

= jr-r

can be solved for

X and

give the inverse transformation

X = x + 2y,
The transformation

Y=x +

y.

(5.19) has the special property that

its

inverse also has

and its inverse are integral transformations. integral coefficients; both (5.19) and Fare replaced by integers, then (5.19) gives a pair of integers for x If and y to make the value of f(x y) the same as that of F(X, Y). The inverse

transformation works the other way, giving integers to integers x and y.

X and Y corresponding

It follows that any integer represented by /can also be represented by F, and conversely. Furthermore, the number of representations is the same in

126

Some Diophantine
cases.

Equations

both

The form Fis considerably simpler than/. By

all

verify that
(

the solutions of
(5.19)

F(X Y)
9

trial, it is easy to 14 are (1, 2), (1, -2), (-1,2),

namely -3), (-3,1). The quadratic form F is said to be equivalent to/. Anything we can say about representations by Fcan be carried over to /by virtue of (5.19) and its inverse. If we study F there is no need to study /or any other equivalent form. This example suggests the desirability of studying transformations and
1,

2).

Then

gives

us the solutions of f(x, y)

14,

(3,

-1), (-5,

3), (5,

equivalence of forms

more

closely.

PROBLEMS
1.

(a)

Determine the 2x 2 + 2y 2 2 2 2x 2y (K) 2 (c) x xy;

;

class of integers represented

by each of the following forms:

(d)
2.

2x 2

2y

2z2

+
2xy

2t

2
.

Prove that x 2

+ y2

is

a zero form. Determine the class of integers

finite

represented by this form.

3. If

tained by multiplying each integer of

class

infinite, let mC denote the class ob by the integer m. Prove that if C is the of integers represented by any form /, then mC is the class represented

is

any

class

of integers,

or

by mf. Prove that ax 2 + bxy + cy 2 is a positive form if and only if a > and 2 4ac < 0. (Observe that this problem shows that the condition c > b
4.

in
5.

Theorem

5.19

is

superfluous.)

Prove that ax 2 + bxy + cy 2 is a positive form if and only if c > and 2 4ac < 0. b 2 2 6. Prove that the form ax + bxy + cy is and negative if and only if a < b 2 - 4ac < 0. 2 2 2 7. Prove that the form ax 4ac < 0. +- bxy + cy is definite if and only if b 2 2 8. Prove that x + Ixy + y is neither a definite form nor a zero form.
9.

(a)
(b)

Find all solutions of the Diophantine equations 5x2 + I4xy + lly 2 = 35;
5x 2

I4xy
2
.

lly

=46.
2 by the positive form 7x

10.

Find the

least positive integer represented

25xy

23y

11. If/(#, y) is a positive binary quadratic form, prove that/(a, ft) is positive for every pair of real numbers a, ft except 0, 0. ax 2 + bxy + cy 2 be a quadratic form with a 12. Let /(a?, y) and

>

b2

4ac

0.

Given

also that (a,b,c)

, precisely the numbers the restriction that (a, b, c) 1, prove that the integers represented

by / are
all

m2

prove that the integers represented with m = 0, 1, 2, Hence, removing

1,
.

numbers of the form

dm2

where d

by /are

(a, b, c).

5.14

Equivalence of Quadratic Forms

127

5.14

Equivalence of Quadratic Forms

The inverse of an integral transformation x

Theorem 5.22
(5.20)
is

<x.X

fiY,

also an integral transformation terminant of the transformation,

= yX + dY A= if and only if
y
"

where

A &

the de

A=
Proof.
to get

a<5

/?y.

The

inverse exists if
if

and only
0.

if (5.20)

and 7, hence

and only

if

A^

Then if the

inverse does exist

can be solved uniquely for we can solve

If

Let/? be any prime, and let/ be the highest power of p that divides A. If p does not divide A, then k = 0. Since the coefficients in (5.21) are integers, we conclude that

A = 1 then (5.21) is an integral transformation. Now suppose that (5.21) is an integral transformation.

It

follows that

But/
implies

is

the highest

0.

2k power of;? that divides A, so we have = 1. A hence and divides Thus no prime A,

fc,

which

This theorem shows that if = by means of (5.20) with A

F is
1,

a quadratic form obtained from a form/ then, just as in Section 5.13, the problem

of representing integers by/can be reduced to that of representing them by F.

Definition 5.1

formg(x,

y)

A quadratic form f(x, y) = ax 2 + bxy -f cy* is equivalent to a = Ax 2 + Bxy + Cy* if there is an integral transformation (5.20),

with determinant
equivalent to

A=

1,

that carries

f(x

y) into g(X, Y). In case

is

g we

write

f~ g.
+
cy
2
,

out we find Applying (5.20) to /and multiplying

(5.22)

A =

aoc

&ay

B =

2a<x/? 4-

b(ad

+
c<5

]8y)
2
,

2cy<5,

C=
if (5.20) carries /(a?, y) into

ap

bfid

g(X, Y).

it

The next theorem shows that this equivalence is a true equivalence relation: is reflexive, symmetric, and transitive.

128

Some Diophantine Equations

Let /
g,

Theorem 5.23

and h be binary quadratic forms. Then

(*)
(c)

iff iff r

g and g ~h thenf~ h.
x

Proof,

and

it

(a) The identity transformation carries f(x, y) intof(X 9 Y).

X, y

= 7

has determinant

(b)

We

can suppose that (5.20)

is

the transformation that carries

f(x
9

y)

into g(X, 7). Then,

by

(5.21) the inverse transformation that carries

g(X

Y)

back into/(#,

y) has determinant

AA
(c)

AA
X=
9

We

suppose that (5.20) with

A A =

carries

and that the transformation

a

A
find

OL^U

ftz;,
9

Y=
If

f(x

y) into

g(X

ftyi

y^u

6]V 9 with

A =
x

Y)

carries

g(X

7) into h(u

v).

we

eliminate

and 7,

we

x y

= =

( aai

(yod

dy^u

(yft

ddjv,

transformation carries /(#, y} into h(u, v). An easy compu tation shows that the determinant of this transformation is equal to

and
(cud

this

/?y)(aA

ftyj

1.

Therefore/Or, y)

is

equivalent to h(x y).

9

Theorem 5.24
Proof.

Iff

~ g,

then the discriminants

off and g
9

are equal.
as in Definition

We

suppose that (5.20) carries /(rr, y) into

g(X 7)

5.1. It is possible to

to obtain

B - 4AC =
2

compute the discriminant

(a<5

B2
but

2
(Z>

/?y)

- 4ac)

4AC
it

is

of g, using (5.22), simpler to use the

ordinary rule for multiplying determinants,

Evaluating the determinants we find 2 Since (a<5 A 2 1 we have b 2 /Sy)

(a<5

4ac

=B

py) (4ac
2

b 2)

= 4AC

B2

44C.

In light of

Theorem

5.23

we can

separate

all

binary quadratic forms into

equivalence classes, putting two forms /and g into the same class if/~g. Part (6) of Theorem 5.23 assures that / and g represent the same set of integers if/and g are in the same class, and that there is a one to one corre spondence between the representations by/and those by g. In other words it

5.14

Equivalence of Quadratic Forms

129

one representative form from each equivalence class. in form a class is positive then so are all the others in that one Clearly, class. Also two forms in the same class have the same discriminants.
suffices to consider just
if
it is desirable to have is some way of picking one form from each class. In the case of unique representative equivalence this can be a done in positive binary quadratic forms, fairly simple way. The whole theory connected with positive forms is quite elegant. For this reason we will restrict our attention to positive forms from now on.

The next thing that

Definition 5.2

A positive binary quadratic form ax 2 +

2
2

bxy

cy

is

a reduced

form

ifQ^b^a^c. For example, x + y

is

a reduced form.

Theorem 5.25

To each

positive binary quadratic

form

there corresponds

an

equivalent reduced form.

Proof.

Cy

2
.

Consider any positive binary quadratic formg(o;, y) = Ax 2 + Bxy + We shall show that there is an equivalent form ax 2 + bxy + cy 2 with

^
tion

\b\

^ a ^ c. This will suffice because in case b negative, the transforma = X,y Y will carry ax + bxy + cy into a reduced form.
is
2 2

Let a denote the least positive integer represented by

g.

In Corollary 5.21

we saw how a can be found. There

Note Bay Cy 2 = which a/d implies d are integers /? and d such
a.
,

are integers a and y such that Aa 2 1 that (a, y) for if (a, y) d, then g(a/J, y\d) 1 since a was minimal. Then by Theorem 1.3 there

+ =

that

a<5
fiv,

the

transformation x

/3y

and we use these

dv

to construct
1.

aw

yu

of determinant

This

transformation carries g(x, y) into

h(u, v)

= A(*.u + fiv) + B(<w + = au + kuv + mv

2 2

fiv)(yu

dv)

C(yu

dv)

say,

where a

is

the integer defined above.

for any integer j the transformation u terminant 1, and it carries h(u, v) into

Now,

jy, v

y has de

f(x, y)
If

= ax +
2

(fc

- 2aj)xy +

(aj

kj

rri)y

we take/

to be the nearest integer to kj2a,

we have
a,

--^
2

2a

-j ^

-a ^

fc

- 2aj ^

|fc

- 2aj\ ^

and we can write

== ax 2 f(x, y)

bxy

cy

2
,

\b\

a.

130

Some Diophantine
positive,
:

Equations

Now g ~/, g is
positive, hence
c

and/(0,

1)

c.

Therefore

represents c,

and

c is

a.

Theorem 5.26
Proof.

are identical. If two reduced forms are equivalent they

and Ax 2 + Bxy + Cy 2 be equivalent reduced forms. We can suppose that a ^ A and that (5.20) is the transformation carrying one into the other. Then the coefficients satisfy (5.22). First we prove that a = A. We have ^ b ^ a ^ c, and using the simple
Let ax 2

bxy

cy

+ y ^ 2 |ay| we find ^ = fla + 6ay + cy ^ 0a + cy - b (5.23) ^ 0a + tzy 6 |ay| ^ 2fl |ay| * |ay| ^ a |<xy|. follows that Since ^ 1. If |ay| = we have ^A> ,4 = + cy ^ 0a + ay ^ (5.24) because a and y are not both zero. If |ocy| = 1, then (5.23) reduces im mediately to the same result, A ^ a. Thus we have A ^ a in both cases, and this with a ^ A implies that a = A. Now that we have a = y4 > we need only prove b = B or c = C, since 4 AC by Theorem 4ac = j5 one follows from the other because 6 C requires no further proof. Since a = A we can 5.24. Thus the case c interchange the forms if necessary and can assume c > C. Then we have c > a since C ^. A = a. This rules out the possibility |ay| = 1, for if = a. Now = then cy > fly and (5.23) would then imply A > a |ay| that we have |ay| = 0, we can prove that y = 0, for otherwise we would again have cy > ay and (5.24) would imply A > a. We have narrowed the possibilities down to a = A, a < c, y = 0. Since = Then by the determinant A = a<5 1, we also have a(5 /?y is = are There two cases. b. we have B (5.22) 2a&/3 <i First, suppose that B = 2fl<x/? + b. Then ^ a and <; B ^ ,4 = = is a multiple of 2a so we But B b that a b ^a. ^B 2aa/? imply have 5 b = 0, 6 = B. 6. This time we find Second, suppose that B = 2fl<x/3 ^ B + 6 ^ 2a or B + b = 2fl. and 5 + 6 is a multiple of 20. We have either B + b = Since ^ i < a and ^ J5 ^ a we have = 5 = 0if^ + i = 0, and b = B = aif B + b = 2a. Again we have b = B. This completes the proof since we have a = A,b = B, c = Cin every case.
2 inequality a 2 2 2

|ocy|

it

|ocy|
2

floe

|ocy|

1.

Z>

Z>

two theorems show that Definition 5.2 does just what we wanted. us with one and just one representative for each equivalence class of positive binary quadratic forms. It should be noted that Definition 5.1 is not the only way in which the positive binary quadratic forms can be put

The

last

It supplies

5.14

Equivalence of Quadratic Forms

131

tion have determinant

into equivalence classes. In fact, some authors demand that the transforma 1 in their definition of equivalence. They then use a definition a different of reduced form in order that Theorems 5.25 slightly

and 5.26

still

hold under their definitions of equivalence and of reduced forms.

Theorem 5.27
discriminant.

There are only finitely

many reduced forms having a

given

Proof.

Let

d be any

form with discriminant

2 negative integer. If ax 2 4ac d d, then b

we have d
there are at

4ac

b2

^ 4ac

most ^d/3 possible for b corresponding to each a. Finally there d. b2 such that 4ac

+ cy is a reduced = and ^ b ^ a ^ c. Thus ac ^ 3d as well as ^ b ^ a, so that values values for a > 0, and at most a +
+
bxy
2 2 1

is

at

most one

c,

for each a, b,

all reduced forms with d ^ 16. We have 1 ^ a ^ we have d = 4c, and we can Corresponding to a = 1, b = = 1 we have d = 4c 1, = = take c 1, 2, 3, or 4. Corresponding to # 1, b = = 2, A = 0, c can take 4. and again we 1, 2, 3, or Similarly we find a = = = = = = = c and 6 rf c rf 8e 2, 6 = 2, 2, 1, 1, 8c, 2, 2, and a = 2. the 6? = 8c c reduced to the forms according values of d 4, Listing

Let us find
1

or

2.

we

get the following table.

J= 3, x + xy +
2

2
?/

+ 7, a* + xy + 2y* 8, x + 2y d= Il,x + xy + ly d = 12, + 32/2 2z + 2*2/ + 2y d = 15, x + xy + 4y 2x + xy +

rf

= d= d=

4,

z2

a/

2y

d. For d = 3, 4, 7, 8, 11, there is just one reduced form of discriminant d will discriminant form with of values For these d, any positive represent the same set of integers as the corresponding reduced form represents. For

example, Corollary 5.14 determined the set of integers represented by x 2 + y 2 and this is the set represented by any positive form of discriminant
,

-4. The case d

discriminant

12

is

little

different since there are

two reduced forms of

form x 2 + 3y 2 the coefficients are relatively 2 2 prime whereas in the second form 2x + 2xy + 2y the coefficients are all divisible by 2. From (5.22) it is obvious that any form equivalent to
12.

But

in the first

132

Some Diophantine

Equations

2x

have all its coefficients divisible by 2. Also by 12, having all its coefficients any positive form with discriminant divisible by 2, will be equivalent to a reduced form having all its coefficients 2 2 divisible by 2, hence it is equivalent to 2x + 2xy + 2y Therefore we can make the statement: Suppose/is a positive form of discriminant 12. If the coefficients of /are relatively prime, then /represents the same class as does X 2 ^_ 3^ if t k e coefficients are not relatively prime, then / represents the same class as does 2x 2 + 2xy + 2y 2 The case d = 16 can be treated in a
2y
will also

+ 2xy +

(5.22)

similar way.

The case d

15

is still

different,

and harder. Here we have two reduced

forms, and in both of them the coefficients are relatively prime. There are ways of distinguishing between forms equivalent to one or the other reduced form without having to actually produce the reduced form by going

through

the steps of the proof of Theorem 5.25. However, we shall not go into this. This is as far as we will pursue the topic of quadratic forms. There is more that can be done. For one we have not discussed the question of how thing,
to determine the class of integers represented by a reduced form. The methods used to obtain Corollary 5.14 can be extended to give some information.

further study of the transformation introduced in the present section

is

also

very useful.

PROBLEMS
1.

Prove that the following forms are equivalent:

ax 2 ax
2.
2

bxy

bxy

+ cy 2 + cy2

ex 2

bxy bxy

ex 2

+ ay 2 + ay 2

Find the reduced form equivalent to

3x 2

(a)
(b)
(c)

2x

+ -

2x 2
3x 2

(d)
3.

+ 5y 2 2 5xy + 4y + xy + 6y 2 + xy + y 2
Ixy
;
.

Prove that there are no binary quadratic forms with discriminant congruent

to 2 or 3
4. 5.

modulo

4.

Find the reduced form equivalent to 7x 2 Prove that to any given positive binary

25xy

-f

23y

2
.

quadratic form there corresponds

infinitely
6.

many

equivalent forms.

is 43. Hence only one reduced form of discriminant 43 are prove that any two positive binary quadratic forms of discriminant

Prove that there

equivalent.

Notes on Chapter 5
7.

133

Prove that any two positive binary quadratic forms of discriminant

2
2

-67

are

equivalent.
8. Denote the positive form ax + buy + cy by [a, A, c]. Prove that, as a variation of the method of Theorem 5.25, the reduced form equivalent to [fl, b, c] can be obtained by a finite sequence of operations of three types:

(1) in case

>

c,

the replacement of
a, the
\b

[a, b, c]

by

[c,

b, a];

(2) in case
is

\b\

>
2

replacement of

[a, b, c]
is

chosen so that

2aj\

^
(b

a,

and
2

Cj_

discriminants, b
(3) in case b

4ac

[a, b 2aj, cj, where; determined by the equality of the

by

2a/)

40^;
[a, b, c]

<

0, the replacement of

by

[a,

-b,

c].

NOTES ON CHAPTER

The methods of this chapter are not suitable for dealing with Pell's equation, 3 z 3 discussed in Section 7.8, and the y = z in positive integers, impossibility of x given in Section 9.10. The Diophantine equations or* + y~ n z~ n 9 where n is a positive integer, are n n shown to be very closely related to x + y z n in the Special Topics, page 261, For further reading on the subject of this chapter, see the book "Diophantine

Equations" by L.
following:
J.

J.

Mordell

listed in the

General References on page 269, and the

L.

Hunter, Number Theory, Edinburgh, Oliver and Boyd, 1964. 2 2 J. Mordell, "On the cz 2 = 0," Monatsh. Math., Bd. 55, equation ax + by 323-327 (1951).

Th. Skolem,

ax 2

by

"A simple proof of + cz2 = 0," Norsk

the solvability of the Diophantine equation

Vid.

Selsk.

Fork.,

Trondheim

24,

102-107

(1952).

H.

S.

Vandiver, "Fermat's

results concerning it,"

last theorem, its history, and the nature of the known Amer. Math. Monthly, 53, 555-578 (1946).

6
Farey Fractions and Irrational

Numbers

A rational number is one which is expressible as the quotient of two integers.

Real numbers which are not rational are said to be irrational. In
this

chapter

the Farey fractions are presented; they give a useful classification of the rational numbers. Some results on irrational numbers are given in Section 6.3,

and

this material

can be read independently of the other two sections. The

discussion of irrational

numbers is limited to number theoretic considerations,

with no attention given to questions that belong or the foundations of mathematics.

more properly

to analysis

6.1

Farey Sequences

Let us construct a table in the following way. In the first row we write 0/1 and 1/1. For n we use the rule: Form the nth row by copying 2, 3,

the (n l)st in order, but insert the fraction (a a')l(b b') between the consecutive fractions alb and a'/b' of the (n l)st row if b 4- b' Thus,

since

^2
+

we

insert (0

+
(1

!)/(!

1)

between 0/1 and

is

1/1

and obtain

0/1, 1/2, 1/1, for the

second row. The third row

insert

0/1, 1/3, 1/2, 2/3, 1/1.

3)

To
1)

obtain

the
(1

fourth
l)/(3

but not
134

row we + 2) and

(0 +
2)/(2

!)/(!

and

(2

l)/(3

3).

The

first five

rows of the

6.1

Farey Sequences

135

table are:

Oil 323
1
1

I!?,
1

1
1

11

43

23 34

I
1

Piii2j.32341.
^

15435253451
^
=

Up to this row, at least, the table has a number of interesting properties. All the fractions that appear are in reduced form; all reduced fractions a/b 1 and b n appear in the nth row; if a/b and a'\b' are such that a/b r n. 1 and b b ab' consecutive fractions in the nth row, then a'b

>

We will prove
Theorem
6.1

all

these properties for the entire table.

a'/b'

If a/b and

are consecutive fractions in the nth row, say

ab'

with a/b to the left ofa'/b', then a'b

1.

1. l)st row. Any Suppose it is true for the (n Proof. It is true for n consecutive fractions in the nth row will be either ajb, a'\V or a/b, a '-)i(b b'), or (a a')/(b b'), a'\V where a/b and a '/b' are con (a

secutive fractions in the (n a'b a(b (a b') a')b

l)st

row. But then we have a'b

ab'
r

ab'

1,

a'(b

b')

(a

a')b

a'b

1,

ab'

1,

and the theorem

Every a/b

is

proved by mathematical induction.

reduced form, that
in order
is, (a,

Corollary 6.2
Corollary 6.3

in the table is in in
f

b)

The fractions

each row are listed

f

of their

size.

Theorem 6.4

If a/b and a /b

among
is

all rational fractions

are consecutive fractions in any row, then with values between these two, (a b') a')/(b

the unique fraction with smallest denominator.

b') will be the first a')/(b Proof. In the first place, the fraction (a fraction to be inserted between a/b and a'\V as we continue to further rows f b )th row. Therefore we have of the table. It will first appear in the (b

a
b

a
b

+ a' +V

d_

by Corollary

6.3.

136

Farey Fractions and Irrational

Numbers

Now
a' IV.

consider any fraction x/y between alb and a'jb' so that alb

<

x/y

<

Then

(6.D

_2 = ('_*) + (*_ b \b' \y y)

_
a'y

b.

- Vx
[_

by
and therefore
b

bx-ay ^ ^
fry

+
y

ft

fr'
,

fry

00 y

Wy

+V< =

a'fe

- aV J_ =
bV'
r
f

which implies y zz b + b'. If y > 6 + V then x/y does not have least denominator among fractions between alb and a /b If y == + V then the inequality in (6.1) must become equality and we have a'y Vx = 1 and Z?x ay = 1. Solving, we find x = a + a', y = b + b', and hence (a + a')l(b + V) is the unique rational fraction lying between alb and a' IV
.

with denominator

Theorem 6.5 I/O yth #t/ all later rows.

Proof.

+V ^x<
.

y, (x, y)

1,

then the fraction x/y appears in the

This

is

obvious

>

1.

definition

Then if y and so it
of the (y

1, with Suppose it is true for y = y y the fraction x/y cannot be in the (y l)st row by must lie in value between two consecutive fractions alb

if

1.

and

a' fV

l)st

row. Thus afb a b

a b

<
f

x/y
tf

<

a'fV. Since

+ +

a'

and
(y

ajb,
l)st

y
(b

part

+ V by Theorem 6.4, and so we have y = b + V. Then the uniqueness of Theorem 6.4 shows that x = a + a'. Therefore x/y == (a + #')/
b'} enters in the

a jb are consecutive, the fraction (a b') is not in the a')/(b row and hence b + V 1 y by our induction hypothesis. But

>

yth row, and

consists
5j!

it is

then in

all later

rows.

Corollary 6.6
that

The nth row

alb

andO

<

n.

of all reduced rational fractions alb such The fractions are listed in order of their size.

Definition 6.1

The sequence of all reduced fractions with denominators not exceeding n, listed in order of their size, is called the Farey sequence of order n.

that

The wth row of our table gives that part of the Farey sequence of order n lies between and 1 and so the entire Farey sequence of order n can be obtained from the /zth row by adding and subtracting integers. For example,
,

6.2

Rational Approximations
is

137

the Farey sequence of order 2

"*'

-3-5-2-3-1-10113253
'

'

'

'

'

"'

l'2'l

l'2'l'

This definition of the Farey sequences seems to be the most convenient. However, some authors prefer to restrict the fractions to the interval from
they define the Farey sequences to be just the rows of our table. reduced fraction with positive denominator :</i is a member of the Any of order n and can be called a Farey fraction of order n. Note Farey sequence
to
1
;

that consecutive fractions a\b and a'\V in the Farey sequence of order n n. satisfy the equality of Theorem 6.1 and also the inequality b

+V>

PROBLEMS
\b' be the fractions immediately to the left and the right of the r fraction 1/2 in the Farey sequence of order n. Prove that b 1 b 2[(n l)/2], that is, b is the greatest odd integer ^ n. Also prove that

1.

Let a/b and a

'

a
2.

a = b. Prove that the number of Farey fractions a/b of order n satisfying the

inequalities of this value.

3.

a\b

is 1

2?=i

<(/)> an(i

^ at ^^ sum

*s

exactly half

be any three consecutive fractions in the Farey sequence Prove that a\b' = (a + a")/(b 4- b"). 4. Let ajb and a \b' run through all pairs of adjacent fractions in the Farey sequence of order n > 1 Prove that
Let ajb,
n. a'/b', a'jb"
'

of order

mm
5.

(a
1

~~

\b

17

a\ I b)

= ~f

la'

n(n

^ 1)

and

max

\b

17

1 ~ a\ = " 1
1

bj

n
,

d >

Consider two rational numbers a/b and c/d such that ad be = 1 b > 0, 0. Define n as max (b,d), and prove that a/b and c/d are adjacent fractions in the Farey sequence of order n. 6. Prove that the two fractions described in the preceding problem are not
1. necessarily adjacent in the Farey sequence of order n Consider the fractions from 0/1 to 1/1 inclusive in the Farey sequence of order n. Reading from left to right, let the denominators of these fractions
, -

7.

be a l9 a 2
6.2

a k so that di

and a k

1.

Prove that ]T"=i (a^+i)"1

1.

Rational Approximations
If a/b and c/d are Farey fractions of order n such that no other lies between them, then a

Theorem 6.7

Farey fraction of order n

a
b

b(b

d)

b(n

'

1)

138

Farey Fractions and Irrational

Numbers

and
c
__

d^ b +
Proof.

d(b

+ d)~

d(n

+
!_
1

We

have
a
b

+c +d

by Theorem 6.1 and the proved in a similar way.

- bc\ b(b + d) fact that b +

\ad

b(b + d)~ b(n + 1) d ^ n + 1. The second

formula

is

Theorem

6.8

Ifn

is

alb such that

<

a positive integer and x ^ n and

is real,

there

is

a rational number

x
Consider the
set

a
b b(n

Proof. fractions (a fractions a\b

of

all

Farey fractions of order n and

all

of the

c)/(b

d) as described in

Theorem

and c/d, the number x will lie and c/d if necessary, we can say that x lies in the closed changing a/b between a/b and (a + c)/(b + d). Then, by Theorem 6.7,
x
a
b b
1

For some Farey between or on, and so by inter

6.7.

interval

b(n

1)

Theorem
rational

6.9

If

is

real

and

irrational, there are infinitely

many

distinct

numbers a/b such that

j

For each n Proof. such that bn n

1, 2,

we can

find

an a n and a b n by Theorem

6.8

<

and
1

.1

Many
many
would

of the

distinct ones.

72=

1,

to each other, but there will be infinitely there were not infinitely many distinct ones, there be only a finite number of distinct values taken by |f a n /b n 9 Then there would be a least one among these values, and it 2, 3,

ajb n may be equal

For
if

would be the value of

|_- ajb n
f
is

|f

|f

ajb n
for
all
TZ

for

some

%/^|

n, say
.

1, 2, 3,

n = k. We would have But |f - aj^| > since

irrational,

and we can

find

an n

sufficiently large that

6.2

Rational Approximations

139

This leads to a contradiction since

On b,

we would now have

1

b n (n

1)

The condition that f be irrational is necessary in the theorem. For if x is any rational number, we can write x = r/s, s > 0. Then ifa/b is any fraction such that a/b ^ r/s, b > s, we have

2 Hence all fractions a/6, 6 > 0, satisfying \x a/b\ < l/b have denominators b ^ s, and there can only be a finite number of such fractions. The result of Theorem 6.9 can be improved, as Theorem 6.11 will show. Different proofs of Theorems 6.11 and 6.12 are given in Section 7.6.

Lemma 6.10

Ifx and y are positive integers then not both of the two inequalities

^-7=(-;
can hold.
Proof.

+ ~l

and

i-)

The two

inequalities can be written as

V
Adding these
2y
2

xy

#2

v5 z(# +
2

y)

(a;

+
3z 2

y)

x2

inequalities,

we

get
2

^/~5

(x

2xy)

2xy

2y

2
,

hence
it
2

2(V5
2

the
0.

form 4y
is

- V5> ^ 0. Multiplying this by 1>2/ + (3 - 4(^/5 - \)xy + (5 - 2^5 + l)a; ^ 0, (2y 2

(>/5

2 we put
1

in

This

impossible for positive integers x

and

because
,

\J~5 is

irrational.

Theorem 6.11

(Hurwitz) Given any irrational number

/#ere e^rw? infinitely

many
(6.2)

differemt rational numbers h/k such that

Proof.

Let n be a positive integer. There exist two consecutive fractions a/b and cjd in the Farey sequence of order n, such that a[b f c/J. Either f (a c)/(4 d) or f d). c)l(b Case I. | (a c)/(b d). Suppose that

<

<

<

+ <

>(a +

a
b

+c +d

140

Farey Fractions and Irrational

inequalities

Numbers

Adding

we

obtain

_5>J_ + J_ d b~ dy y
5

a
'

+ c _a > ~ +d b

(b

+ d)V5
dv

'

y5

hence
=

bd
and
1

^2>j^a + i\ =
6
a

V5\6

= ""
d)

(g

c)b
6(fc

(fc

d)g

6(6

> = J = /l +
v 5\fc
/

L_
(fc

d)

<3

These two inequalities contradict

a/h, c\d, (a

Lemma

6.10. Therefore at least

one of

c)j(b

d) will serve as h\k in this case.

c)l(b

Case

II.

>(a +

+
a
b

d).

Suppose that
c
1

+ +

d\/5

Adding

as before,

we obtain

V
hence

'

A/ 5
+

d)

d(h

V5
at
least

$
one of
a/b,
c\d,
satisfies (6.2).

which also contradicts

(a

Lemma

6.10.

Again

+
We

c)l(b

d) will serve as h\k.

have shown the existence of some h\k that depends on our choice of n. In fact h/k is either a/b,

This h\k

c/d,

or (a

c)/(b

d),

where a\b and c/d are consecutive fractions in the Farey sequence of order n, and a/b < f < eld. Using Theorem 6.7 we see that
a
e

~i
+
d(n

+1)
many

4(/i

""
1)

We

want to

establish that there are infinitely

h\k that satisfy

-^
is

(6.2).

Suppose that we have any h 1 /k l that

satisfies (6.2).

Then

positive,

and we can choose n

>

h,
.

The Farey sequence of order n then

6.2

Rational Approximations

141

yields

an h/k that

satisfies (6.2)

and such

that

"n
This shows that there exist infinitely
(6.2) since, given

many

rational

numbers h/k that

satisfy
.

any

rational

number, we can find another that is

in

closer to f

Theorem 6.12
words Theorem
Proof.

The constant >/5

6.1
1

Theorem
is

6.1

1 is

the best possible. In other

does not hold if\f5

replaced by any larger value.

We

need only exhibit one f for which

V5

larger value. Let us take

cannot be replaced by a

J(l

+ Vs).

Then

(x

fe

For

integers h,

k with

>

0,

we then have

(6.3)

The expression on the left in (6.3) are irrational. The expression |/z 2
Therefore
2
|/z

is

not zero because both f and

Afc
2
|

is

AJk

VI f a non-negative integer.

k*\

and we have

(6.4)

Now
j

suppose we have an

infinite

>

0,

and a

positive real

number

sequence of rational numbers such that

(6.5)

mfe!

Then
finite

fc.-f

<
mkj
h,>

h,

<

A:,|

mk

,
.

and

this implies that there are

.

only a

number of
as/

kj

* oo

> oo.

corresponding to each value of k3 Therefore Also, by (6.4), (6.5) and the triangle inequality

we have we have

142

Farey Fractions and Irrational

Numbers

hence

m<
and therefore

d;

^
= J5.
'

m<

lim I ""^ooVrnfe?

J5\

PROBLEMS
1.

a,
2.

Prove that for every real number x there are infinitely many pairs of integers l b, with b positive such that \bx a\ < (*Js b)~ Take as on p. 141. Let A > and a > 2 be real numbers. Prove that there
.

are only finitely

many

rationals h/k satisfying

3.

rational

k
4.

= a, k = b is a solution of the inequality (6.2) for some ir Suppose h f. Prove that only a finite number of pairs h, k in the set {h = ma,
mb;

m=
1

1, 2, 3,

} satisfy (6.2).

Let a

>

infinitely
/?

be a real number. Suppose that for some real number /? there are a Prove that h\k\ < k~ many rational numbers h/k such that \p
.

is

irrational.

5.

Prove that the following are irrational: JJLi 2

^ 2i

2 ~'

6.3

Irrational

Numbers
is

A
is

rational

number, as

well-known,

is

one which can be expressed as the

quotient of two integers, ajb with b

said to be irrational.

0.

A real number which is not rational

That \/2
that a 2

is

irrational

tion theorem.

For
.

if

V^

can be concluded at once from the unique factoriza could be represented in the form a/b, it would follow

2b 2 But

this is

power of 2 that
that divides 2b 2

divides a 2
is

impossible with integers a and b because the highest is an even power, whereas the highest power of 2 an odd power, by the unique factorization theorem.

more general argument

Theorem 6.12
(6.6)

for deducing irrationality

is

formulated next.

If a polynomial equation with integral coefficients

n

c nx

cn

_^ +

c 2z 2

+ c& +

0,

cn

0,

6.3

Irrational

Numbers

143

has a nonzero rational solution a/b where the integers a and b are relatively prime, then a c and b c n
.
\ \

Proof.
n

71 - 1 we note that Replacing x by a\b in (6.6) and multiplying by b c n a /b is an integer, and hence b c n since (a, ft) 1. On the other hand, n n replacing x by a/b in (6.6) and multiplying by b /a, we observe that c b /a is an integer, so a C Q
,

Corollary 6.13

rational solution, that solution

If a polynomial equation (6.6) with c n is an integer which divides CQ

1
.

has a nonzero

Corollary 6.14
any,

ofx

For any integers c and n 0, the only rational solutions, if c are integers. Thus x n c has rational solutions if and only if

>

c is the nth
It

power of an

integer.

there are

V$, VI, V^ are irrational because no integral solutions of x2 = 2, x 2 = 3, and z3 = 5. Another application of Theorem 6.12 can be made to certain values of the
Let 6 be a rational multiple of rr, thus 6 = rrr where r is Then cos 6, sin 6, tan 6 are irrational numbers apart from the cases
is

follows at once that such numbers as

trigonometric functions, as follows.

Theorem 6.15
rational.

where tan 8

undefined,
1/2,

and
1;

the exceptions
sinfi

cos0
Proof.

0,

0,

1/2,

1;

tan

0,

1.

induction that there

Let n be any positive integer. First we prove by mathematical is a polynomial/^) of degree n with integral coefficients and leading coefficient 1 such that 2 cos nd = /n (2 cos 6) holds for all real

numbers

known

z2 note that/^z) 2 because of the wellx, and/ 2 (#) 2 2 cos 26 2. cos The (2 0) identity identity
6.

We

2 cos (n
is

1)0

(2 cos 0)(2 cos nO)

2 cos

(n

1)0

by elementary trigonometry, and this reveals that = x x n ( ) /n+i(X) f /n-iO*0 which completes the proof by induction. Next, let the positive integer n be chosen so that nr is also an integer. With
easily established

r-n-

it

follows that

/n (2 cos
where the plus sign holds

0)

= 2 cos

= 2 cos nrrr =

2,

if nr is even, the minus sign if odd. Thus 2 cos is a solution of n (x) 2. Setting aside the cases where cos 0, we apply Corollary 6.13 to conclude that 2 cos 0, if rational, is a nonzero integer. But

cos
2.

and

As

to

^ 1 so the only possible values of 2 cos 0, apart from 0, are 1 So Theorem 6.15 has been established in the case of cos 0. sin 0, if is a rational 0, and from the multiple of TT so is 7r/2
=
cos
(77/2

identity sin

0)

we

arrive at the conclusion stated in the

theorem.

144

Farey Fractions and Irrational

Numbers

tan 2 tan 2 6)/(l (1 Finally, the identity cos 26 tan 6 is rational so is cos 26. In view of what was just
cosine function,

(9)

reveals that if

proved about the

1. cos 26 0, 1/2, possibilities 26 1 cos 6 when tan that calculated it is When cos 26 readily 1 tan 6 is undefined; when cos 26 1 tan 1/2, 0; when cos 26

we need look only at the

tan 6

is

one of the irrational values

6. 15.

>/3,

l/>/3.

This completes the proof

of Theorem

to a positive rational base positive rational number for example, Iog 6 9. If is easily classified as rational or irrational. Consider, b are a and this were a rational number a/b, where positive integers, this

The logarithm of any

would imply

that 9

6 a/b or 9

&

a
.

The unique

factorization theorem can

&

2 These 3 and 1 be applied to separate the primes 2 and 3 to give 9 9 is irrational. that conclude we b 0, and so Iog 6 equations imply that a The basic mathematical constants TT and e are irrational. proof of this 7 and 8 Problems in reader it the to we leave for e is simple that
.

= =

sufficiently
TT

below. For

the matter

is

not quite so easy, so we precede the proof with

a lemma.

Lemma 6.16
divisible

Ifn

is

any positive

integer,

n coefficients, then x g(x)

and all

its derivatives,

andg(x) any polynomial with integral evaluated at x = 0, are integers

by

nl.

Proof. c

in g(x) is of the form ex* where c and j are integers with The corresponding term in x ng(x) is cx* +n if we prove the lemma for this single term, the entire lemma will follow because the derivative of a finite sum is the sum of the derivatives. At x = 0, it is readily seen that cx i+n and all its derivatives are zero, with one exception, namely the (j + /i)th derivative. The (j+ )th derivative is c{(j + )!}, and since j ^ 0, this is divisible by n\.

Any term

and;

0.

Theorem 6.17

TT is

irrational.
TT

Proof. Suppose that the polynomial

(6.7)

a/b,

where a and b are positive

integers. Define

f(x)

x n (a

bxf\n\

= bVfr -

n
a?)

/7i!,

where the second form of f(x) stems from the first by simple algebra. The integer n will be specified later. We apply Lemma 6.16 with g(x) in the form n n n bx) and all its derivatives, evaluated at bx) to conclude that x (a (a

by n!. Dividing by nl, we see that/(#) and all its 0, are integers. Denoting the;-th derivative of w) W) and writing /(z) =/(a;), we can state that/ (0) is an f(x) by/ (),
x
0, are integers divisible

derivatives, evaluated at x

integer for every/

0, 1, 2, 3,

6.3

Irrational

Numbers

145

By

the second part of (6.7)

we

derivatives

we

(-ly/^fr
is

get
*)

-/'(>
)

x)

an integer for every j = 0, 1, 2, 3, Next the polynomial /"(a;) is defined by

F(x)

=/'

(^).

Letting

=/,/ x =

find that/(7r
(2)

(7r

x)

z)

=/

(2)

=/(X), and taking Cr), and in general

we
.

obtain the result

= /(*) -/>(
equation
is

Now

if this

differentiated twice the result

is

+
because /
<an+a)

(-l)-i/<W(aO

a polynomial of degree 2/z. Adding these get F(x) (x) =/(*). Also, by the preceding paragraphs equations we observe that F(G) and F(ir) are integers, because they are sums and
(a)

since /(a;)

is

we

+F

(2)

differences of integers.

Now

by elementary calculus
x

it is

seen that

dx

(F'O)

sin

F(x) cos x}

F"(x) sin

a?

+ F(x) sin = /(a;) sin x.

a;

Thus we are able to

integrate f(x) sin x, to get

(6.8)

f(x) sin
Jo

a:

Ja;

[F'(x) sin

a;

- F(a;) cos

a;];

= F(TT) + jF(0).
+

contradiction arises from this equation, because whereas ^(TT) F(G) is integer, we demonstrate that the integer n can be chosen sufficiently large in the definition off(x) in (6.7) that the integral in (6.8) lies strictly between

an

Oand 1. From

(6.7)

we

see that

from x

to x

TT,

/(*)
Also/(a;) sin x

< --

nn
and /O) j ^ j
interval
~ sin

>

in the

open

<x<
<
n

TT,

and hence

<

f /(*) si sin x dx
Jo

it is

because the interval of integration is of length TT. From elementary calculus well-known that for any constant such as tra, the limit of (ird) n lnl is zero

as n tends to infinity.

in (6.8) lies strictly between stated above. It follows that

Hence we can choose n sufficiently large that the integral and 1, and we have obtained the contradiction
TT is

irrational.

146

Farey Fractions and Irrational

Numbers

PROBLEMS
1.

Prove that the irrational numbers are not closed under addition, subtraction,

multiplication, or division. 2. Prove that the sum, difference, product, and quotient of irrational and the other a nonzero rational, are irrational.
3.

two numbers, one

establish

Prove that >/2

it is

+ >/3

is

a root of z 4

10x z

0,

and hence

that

irrational.

4. (a)

2 positive integer h, note that /z ends in an even number of zeros whereas 10/z ends in an odd number of zeros in the ordinary base ten notation.

For any
2

Use
h2

this to

10& 2

h\k so that prove that A/ 10 is irrational, by assuming y 10 Extend the Extend this to ^lO. (b) (c) argument to prove argument

is irrational, where n is a positive integer not a perfect square, by n as of the the number base taking system instead of ten. 5. (i) Verify the details of the following sketch of an argument that A/77 is

that A//2

irrational.
let
is

Suppose that A/77 is rational, and among its rational representations a\b be that one having the smallest positive integer denominator b, where a
an
integer.
8Z>).

also

(77Z>

Sa)l(a
is

Prove that another rational representation of A/77 Prove that a 86 is a smaller positive integer than
(ii)

is

b,

which

a contradiction,
is

a positive integer not a perfect square, by assuming n = a/b and then getting another rational representation of n with denominator
irrational if n

Generalize this argument to prove that

A/TZ is

kb where k

[A/TC],

the greatest integer less than ^/n.

(An

interesting

aspect of this problem is that it establishes irrationality by use of the idea that every nonempty set of positive integers has a least member, not by use of the
6.

unique factorization theorem.) Let a/b be a positive rational number with a

>

0, b

>
l/n
,

0, g.c.d. (a, b)

1.

Generalize Corollary 6.14 by proving that for any integer n > 1 the equation x n === a/b has a rational solution if and only if both a and b are nth powers of
l/n integers. Suggestion: If (a/b)
is

rational so

is

the equation x n 7. Prove that a

ab 71

"1

b(a/b)

which

is

a root of

number a

is

rational if

a. Prove that a number a is rational if integer k such that [fca] there exists a positive integer k such that [(A:!)a] (&!)a.

and only

if

there exists a positive

and only

if

8.

Recalling that the mathematical constant e has value

2JL W*> prove

that

Hence prove
9.

that e

is

irrational.

where "1" is in radian measure. Suggestion: use the infinite series for cos x, and adapt the ideas of the two preceding problems.
1 is irrational,

Prove that cos

6.4

Coverings of the Real Line

147

6.4

Coverings of the Real Line

The rational numbers are dense on the real line, that is, between any two of them there is another. This is obvious, because if r and s are rational numbers, (r + s)/2 lies between them on the real line. It might appear therefore that if every rational number is covered by an interval on the real line, the entire line would be covered. Whether this is so or not depends on the lengths of
the intervals, as

we

shall see.

numbers a and /J with a < {}, the open interval from a to x satisfying a < x < /?; this interval is denoted is the set of real numbers /? by The closed interval is the set of real numbers x satisfying a ^ x ^ (I, (a, /?).
Given two
real

denoted by

[a,

/?].

Now

Theorem
is

6.11 can

be interpreted as stating that

if

every rational

number h\k

covered by an open interval

(--F-.-+-M \k /S k -v/JK. *J J
\/V
If

/C

/-S K, L' / /

then every irrational number f is covered, in fact covered by infinitely of the intervals. Thus the entire real line is covered.

many

However,
open
(6 9)
-

if

every rational

number h/k

is

covered by the slightly shorter

interval

(irs-l
=

+
4l?)

then not every irrational number is covered. Specifically we prove that not covered. Without loss of generality we may presume that h/k is in lowest terms, i.e., that g.c.d. (A,&) 1, because the interval (6.9) covers
is

number hjk written in the form (ch)l(ck) where c is a positive integer. Also k > may be presumed. First if k = 1 the intervals (6.9) extend for a distance 1/4 on each side of every integer A, so
any
interval with the rational
,

they do not cover \jl\2 because it 2. Second if A/A: 1 or h/k

is

>

<

approximately .707. Henceforth we take then the interval (6.9) does not overlap
1

the interval (0,

1)

because

if A/A:

>
A

then

h__J_
k 4k
2

= /LZLl> = k
=:

'

and

if A/A:

<

then

4k 2<

+
lc

148

Farey Fractions and Irrational

if

Numbers

Third
for

h\k

lies in

the interval (0,

1),

then

<

h/k
is,

<

I.

Now suppose that

some h\k

the interval (6.6) covers x/2/2, that

_
k
Squaring
all

'

4k*

2
see that

4fe

three expressions

we

k*

2k 3

16k*

2k

2k 3

2 Multiplying by 2k and then subtracting

2/z

from each expression we get

(6.10)

_* + _L <jfe _2A*< + -L
k
2

8/c

8fc

However we note

that

But A 2

2A 2 in (6.10)
2/z
2

is

must have k 2

an integer, and a view of these last inequalities, we 0, which is impossible because %/2 is irrational.

PROBLEMS
1.

Prove that

if

every rational

number h\k

is

covered by an open interval of

the type (6.9) but with the constant 4 replaced by 3, then >/2/2 is covered. 2. Prove that the irrational numbers are dense on the real line, i.e., between

every two irrational

numbers there

is

another irrational.

NOTES ON CHAPTER
6.3.

irrationality

6.15 follows that of E. A. Maier, "On the of certain trigonometric numbers," Amer. Math. Monthly 72, 1012 (1965). Further results on the topic of this section can be found in Ivan Niven,
6.3.

The proof of Theorem

Irrational

Numbers, Carus Monograph 11, New York, John Wiley and Sons, 1956. The relationships between rational and irrational numbers and their decimal expansions are given on page 259 ff in Special Topics. The fact that
every
is

positive rational number is expressible as a sum of distinct "unit fractions," established also in Special Topics, page 260 ff.

Notes on Chapter 6
6.4.

149
possible statement that can be

The sharpest
is

made

type considered

this: if every rational

number h\k on

for coverings of the the real line is covered by

2 h\k + l/#c ] with ft = (V? + 3)/2, then the covered by these intervals. For any larger value of ft, the entire real line is not covered. This is part of what is established by A. V. Prasad in "Note on a theorem of Hurwitz," Journal London Math. Soc., 23, 169-171 (1948).

the closed interval

entire real line
is

[h/k

l/#c

7
Simple Continued Fractions

7.1

The Euclidean Algorithm

fraction
Z/Q/WI,
i

Given any rational

u1

n lowest terms so that

(w

i)

==

and
1.11

>

0,

we apply

the Euclidean algorithm as formulated in

Theorem

to get
(7.1)

w
Wl

w2

= = =

W^Q
Mafll

w 3 <2 2

+ + +

w2
r/

3,
,

w4

< < <

1?

w2 w3

w4

< Wi <w <w

3

+
The notation has been
of b, c by by <ZQ> a i5
i/
,

< ui <

1.11 by the replacement w w and of q l9 q2 " 2 s r, by z/,+1 q The form (7.1) is a little more suitable for our present #:> purposes. If we write ^ in place of z/$/ i+ i for all values of / in the range z ^7, then equations (7.1) become

altered

from that of Theorem

, , ,

w 1? of
3

r l9 r 2 ,

'

If

we take the first two of these equations, those for which and eliminate g l9 we get
?o

and

1,

ao

150

7.1

The Euclidean Algorithm

its
,

151

In this result we replace f 2 by the replacement of f 3 4

,
,

value from (7.2), and then

we continue with

to get

(7.3)

This is a continued fraction expansion of f or of uju^ The integers a t are called the partial quotients since they are the quotients in the repeated that application of the division algorithm in equations (7.1).
,

We

presumed

the rational fraction u^u^ had positive denominator u l9 but we cannot make a similar assumption about z/ Hence a Q may be positive, negative, or zero.
.

However,

since

< <

ti

<

w 1?

we note

similarly the subsequent quotients a 29 a 3 case j that is if the set contains 1, (7.1)
,

that the quotient a^ is positive, and af are positive integers. In ,

-

u3 imply that a,-> 1 (a a x a,) to designate the continued fraction in (7.3). In general, if # x l9 a, are any real numbers, all a we shall write positive except perhaps
Ujluj+i

more than one equation, then

.

and

w; +1
.

<

We

shall use the notation

Such a

finite

continued fraction

is

The following obvious formulas are often

said to be simple if useful

:

all

the

x i are

integers.

'

'

'

*!

^-

25

/-!

/-

The symbol

[x 09

x l9

o;^]
-

is
-

We use the notation (#

# ls

often used to represent a continued fraction. ^-} to avoid confusion with the least common
,

multiple and the greatest integer.

152

Simple Continued Fractions

PROBLEMS
1.

Expand the

rational fractions 17/3, 3/17,

set (7.1) consists
is

and

8/1 into finite simple

con

tinued fractions.
2.

Prove that the

Under what circumstances

3.

of exactly one equation

?

if

and only

if

u:

1.

4.

<2, 1,4>; <-3,2, 12>; <0, 1, 1, 100). Given positive integers b, c, d with c > d, prove that <<z, c> < (a, d} but (a, b, c} > <a, b, d} for any integer a. 5. Let l9 # 2 a n and c be positive real numbers. Prove that
flf ,

Convert into rational numbers:

<<7

a l9

fl

n>
is

>

<0

*!,

an

c>

holds

if

is

odd, but

is false if

even.

7.2

Uniqueness

In the last section we saw that such a fraction as 51/22 can be expanded into a simple continued fraction, 51/22 = (2, 3 7). It can be verified that 51/22 can also be expressed as (2, 3, 6, 1), but it turns out that these are the only two representations of 51/22. In general, we note that the simple continued
,

fraction expansion (7.3) has

an alternate form,

(7.4)

=
M!

(0

a l9

fl

y _!,

dj)

(a

a l9

a,-_ 1?

a,

1, 1).

The following
Theorem
z

result establishes that these are the only

two simple continued

fraction expansions of a fixed rational

7.1

number.
,

If (0
TZ.

al9

tinued fractions are simple,

/(?r

as ) = and if a^
,

<A

>

and b n

>

1, ?/ze72/

=n
,b n )

0,

Proof.

We

write y i

for the

continued fraction

(b i9

b i+l9

and

observe that

(7.5)

%=

<>

& i+1

b n)

=
1

fe,

+i

/?

i+2 ,
, ,

=
1,

fe.

+ -.
yi+l

(o tf+1 ,

b n)

Thus we have y i

>

i n. The \y t ] for all values of i in the range Consequently b t hy that the continued are fractions can be written in the form pothesis equal f o> where we are using the notation of equation (7.3). the definition 2/o

b and y i

>

for

1, 2,

and y n

bn

>

1.

^ ^

Now

of

^ as

Wf/Wf+i implies that f z +1

-

>

for all values of/

0,

and so

^=

[f f ]

7.3

Infinite

Continued Fractions
(7.2). It

153
follows from y
(7.2)

for

<
=

parts, b

<:j by equations
[y
]

f that, taking integral

(7.5)

By equations
,

and

we

get

^o

2/o

*o

=
2/i

2/i,

fli

[fi]

[2/1]

= =

&!-

fi

This gives us the start of a proof by mathematical induction. establish that f, y i and a, 4^ imply that f <+1 y +1 and a i+1 To see this, we again use equations (7.2) and (7.5) to write

We now
i f+1
.

St

ai

yi

bf

must also follow that the continued fractions have the same length, that = n. For suppose that, say, j < n. From the preceding argument we have f, = yf a = b^ But ^ = as by (7.2) and ^ > 6^ by (7.5), and so we have a contradiction. If we had assumed y* > a symmetrical contradic tion would have arisen, and thusy must equal 77, and the theorem is proved.
It
is,

that j

Theorem 7.2 Any finite simple continued fraction represents a rational number. Conversely any rational number can be expressed as a finite simple continued fraction, and in exactly two ways.
Proof.
the

The first assertion can be established by mathematical induction on number of terms in the continued fraction, by use of the formula
'
' '

<a

ai,

a y>

+-

The second

assertion follows

from the development of

u^u-^ into

finite

simple continued fraction in Section 7.1, together with equation (7.4) and

Theorem

7.1.

PROBLEM
1.

Let

<7

a ly

a n and b

b^

*
,

b n+l be positive integers.

What

are the

conditions for

7.3

Infinite

Continued Fractions
be an
infinite

Let

<2

a l9
.

<2

2,

sequence of integers,

all

perhaps a

We

define two sequences of integers {h n }

and {k n } inductively

positive except as

154
follows
(7.6)
:

Simple Continued Fractions

= l, ^ = 0A-i+^= -2=1, &-i 0, kf = ajc^ + k,_ = 1, ^ = We note that ^& & >k = kQ <; &! < & < 3 < < kn <
/L_ 2

0,

A_!

for/>0,
for
i

fc

|> 0.
,

l9

kz

>

k2

etc.,

so

that

fc

Theorem

7.3

For any positive real number

x,

Proof.

If n

0, the result is to

be interpreted as

xk_!

+
1,

k_ 2
the result
is

which

is

true by equations (7.6). If n

which can be

verified

We

establish the
,

holds for (a

a ly

from (7.6) and the fact that (0 x) stands for a + \\x. theorem in general by induction. Assuming that the result an_ ly x), we see that
,
-

<0o>

<*!,

w > #>

a ls \ ^o

n _ l5

an

xh n

Theorem

7.4

/jTw<? define rn

(a

an )for all integers n

0,

We apply Theorem 7.3 with x replaced

(7.6) thus:

by an and then use equations 4

kn

Theorem

7.5

The equations

7.3

Infinite

Continued Fractions
1
.

155

hold for

The

identities

-r_ =
is
,

^^
=
1.
.

hold for

i^l.

The fraction AJA^

is

reduced, that

(h t fcj

= 1. Continuing the P/w/. The equations (7.6) imply that /z.^ 2 we assume = that induction, h^-Jc^ proof by A^fe^ ( 1)*~ Again we - A fc, = (0A-i + h+_Jk^ - h^ajc^ use equations (7.6) to get A< fc +

A.^

(A<-i*<-2 &z- 2 ) stated in the theorem.

A*-2^-i)

(I)*"1

This

We divide by &,_!&,. to get the second result, the formula

.

proves

the

first

result

for

Furthermore, the fraction hjk^ is in lowest terms since any 1 is also a factor of ( i I)*" The other formulas can be derived in much the same way from (7.6), al though we do not need induction in this case. First we observe that A fc_ 2
Ti

/<_!.

factor of

^ and k

h_ 2 k

a 09
4-

and that

in

^-2(^-1

^_ 2 ) = a^h^Jc^ - h^k^)
The values
r n defined in

general

hjk^ - h^Jc
==

(fl^-i

^--2)^-2

(-1)^-. The

final

identity

can be obtained by dividing by k^^k^

Theorem 7.6

Theorem

7.4 satisfy the infinite chain

r2 r4 r6 of inequalities r Q the r n with even subscriptsform an

<

<

<

<

<
0, r 2j

r7

<

r5

<

r3

<

rx

Stated

in

words,

form a decreasing sequence, and lim^n r n exists, andfor every]

Proof.
r *s

increasing sequence, those with odd subscripts every r 2n is less than every r2j_. Furthermore,

< lim^n r n < r2M

The
^z
-

identities

of Theorem 7.5 for

< r*w
z

r zj-i

>

r 2j+i,

and

r 2j
i

and the
^i

are positive for

< ^

rt r f-1 and r i r f_ 2 imply that r^_^ because the k t are positive for i

>

rs

>

rs

>

To prove

that

Thus we have r < r 2 < r4 < r 2n < r w _ l9 we put the previous

r
-

and
results

together in the form

The sequence r r 2 r4 is monotonically increasing and is bounded above by r l9 and so has a limit. Analogously, the sequence r l9 r 3 rs is r and is bounded below and so has a limit, monotonically decreasing by These two limits are equal because, by Theorem 7.5, the difference r i rf-1
,

tends to zero as

Another way of looking

is

the integers k i are increasing with to observe that (r r L ), (r 2 , r 3 ), (r4 r 5 ), a chain of nested intervals defining a real number, namely lim^^ r n
i

tends to

infinity,' since
is

z.

at this

These theorems suggest the following

Definition 7.1

definition.
-

o/ integers, all positive sequence aQ aly a%, a an infinite simple continued fraction determines except perhaps for is The value of (a a l9 a 2 (a a i9 a 2 ). > defined to be lim^^
infinite
, , ,
,

An

<a

a l9 a 2 ,---, a n ).

156
This
limit,

Simple Continued Fractions

being

the

same

as

lim^^

rn ,

exists
.

Another way of writing

(#
,

this limit is

#!,--, a n )

h n /k n

limn^^ hjk n

by Theorem 7.6. The rational number

r w is called the 72th convergent to the infinite

continued

fraction.

We

say
.

that

the infinite

continued fraction con

' ' '

verges to the value lim n _>00 r n In the case of a finite simple continued fraction the rath a n ) we similarly call the number (<2 #1, l9 (a
, , >

convergent to

<<2

l9

a n >.
0/" #/zy infinite

Theorem 7.7 27ze va/z/^ a2 -) is irrational.

-

simple continued fraction (#

o^,

Proof.

Writing 6 for
rn

(<2

a l5 # 2 >

between

|0 rj and making use of the result from Theorem

and

r w+1 ,

so that

<

we observe by Theorem

7.6 that

lies
fc n ,

<

|r n+1 7.5 that

rj. Multiplying by
rn

|r n+1

(^n/^+i)"

we have

= a[b with integers a and b, b suppose that 6 were rational, say d Then the above inequality would become, upon multiplication by b,
Now

>

<
<

\kn a

h n b\

< -would

The integers k n increase with , so that b k n+l Then the integer \k n a

.

we could choose n
h n b\
lie

sufficiently large so

between

and

which

is

impossible.

Suppose we have two different infinite simple continued fractions, # 15 fl 2 and (& Z? l5 b 2 } }. Can these converge to the same value? The answer is no, and we establish this in the next two results.
(#
-

Lemma
ao

7.8

Let 6

(<z

aly a 2

>

be a simple continued fraction. Then

,

[#]

Furthermore if6 1 denotes (a l9 a z

Proof.

Now

flj

By Theorem 7.6 we > 1, so we have

6

see that r

<

6
1
,

<

<
a n>

+ 1/6^. < r that a < 6 < a + and hence a = [0]. Also

8, )

then 6
is

l9

= lim
->>

(<z

fli,

= lim
(

+-

n-^oo \

(a l3

Q
Q
,
.

l/a x

a n )/

hm <a
n-*oo

l8

a n)

ex

Theorem

7.9

7V0

distinct infinite simple

continued fractions converge to

different values.

7.4

Irrational

Numbers

157

Proof.

by

Lemma

Let us suppose that (a^ a i and 7.8, [6]

a^ a 2

>

==

<&

J l9

Z>

2,

6.

Then

(a l9

<2

2,

(b l9

2,

Hence (a ly (b l9 ) }. Repetition of the argument gives a and so by mathematical induction a n = b n for all n.
a2
,

b2

b ly

PROBLEMS
1.

Evaluate the
7.8,

infinite

continued fraction

<1, 1, 1, 1,

>.

Lemma
2.

we

see that 6

Suggestion: by

I/O in this case. This gives a quadratic

equation, only one of whose roots is positive. Evaluate the infinite continued fractions
<2, 3, 1, 1, 1, 1,
>.

(2, 1, 1, 1, 1,

>

and

Suggestion: use the result of the preceding problem

along with

Lemma

7.8.

3. Evaluate the infinite

(a)
(c)

continued fractions
(b)

(2,2,2,2,

>;

<1, 2, 1, 2, 1, 2,

>;

(2,1,2, 1,2, !,--->; >. (d) <1, 3, 1,2,1,2,1,2, 4. For n ^ 1, prove that kjkn^ = (a n , a^^ ,a2 a-,}. Find and prove a similar continued fraction expansion for /z n //tn_i, assuming a g: 0.
,

7.4

Irrational

Numbers

irrational

have shown that any infinite simple continued fraction represents an number. Conversely, if we begin with an irrational number f or we can f o, expand it into an infinite simple continued fraction. To do this we - a ), and next al = [fj, f2 define a = [ ], f x = 1/(
,

We

and so by an inductive
(7.7)

definition

a,

[fJ f

The at are integers by definition, irrationality of f x is implied by that

Furthermore,
a^

and the
,

are
2

all

irrational since the

for

of f because

that of

a^ =

by that of f 1? and so on. [f z-_i] and the fact that f _! is

irrational imply that

158

Simple Continued Fractions

(7.7) in the

Next we use repeated application of

get the chain

form f z

ai

l/i i+1 to

0o

+T=
si

(a*,

f i>

-, a m_ 2

a m_

(<z

a l9

a m _ l9

m >.

is the value of the infinite This suggests, but does not establish, that determined by the integers a t continued fraction (a a l9 a 2) } To prove this we use Theorem 7.3 to write
-

f-i o\ (7.8)

f-

'

'

<0

*i,

a n -i> fn>

f-

n""nl

-i"

""n2
fe

f n n -l
fe

-2

with the

/z^

and kt defined

as in (7.6).

By Theorem

7.5

we

get

(-I)

""1

This fraction tends to zero as

/z

are increasing with n, and g n is positive. Hence f tends to infinity and then, by Definition 7.1,

tends to infinity because the integers k n r n_! tends to zero as n

= Km r n = lim
n-*-co
n-+<x>

a ls

a n)

(a

a l9 a 2)

}.

We summarize the results of the last two sections in the following theorem.
Any irrational number $ is uniquely expressible, by the procedure that gave equations (7.7), as an infinite simple continued fraction # 1? <2 2? (<2 ). Conversely any such continuedfraction determined by integers are which ai positive for all i represents an irrational number, f. The
,

Theorem 7.10
*

>

finite simple continuedfraction (a , a l9 r n , and is called the nth convergent to

to the

a n ) has the rational value

hjk n
-

a^ For n

= 0, 2, 4,

Equations (7.6) relate the

/z

and k4

sequence with

these convergentsform a monotonically increasing as a limit. Similarly , for n the convergents form 1, 3, 5,

a monotonically decreasing sequence tending to The denominators k n of the convergents are an increasing sequence of positive integers for n > 0.
.

7.5

Approximations to

Irrational

Numbers
<0
,

159

Finally, with f, defined by (7.7),

we have

fn

a l9

<0

^,

a n _ l9 f n >

(a n , a n+l9 a n+2

-).

Only the last equation is new, and it becomes obvious if Proof. to g n the process described at the opening of this section.

we apply

PROBLEMS
1.

Expand _each of
Given that two
-

the following as infinite simple continued fractions: ^~2

722.

1,72/2,73,1/73.
irrational numbers have identical convergents hQ /k up to h n /kn prove that their continued fraction expansions are
,
,

hjkl9
3.

identical

up

to a n .

Let

a, p,

identical convergents h ]k , h^k^ same convergents up to h n jk n .

4.
<fl

y be irrational numbers satisfying a < p < y. If a and y have up to hnjk n prove that p also has these
-

Let
,

| be a l9 a 2 a z
, ,

an
>.

irrational
,

number with continued

-

fraction

expansion

Let b l9 b 2 bz ,

be any

finite

or infinite sequence of

positive integers. Prove that

lim <a
n
5.
*c

a l9 a 2

an 6 X
,

Z?

2,

63 ,

>

In the notation used in the text, prove that

6.

Prove that for n

>
n

~
7.

(~l)

n {f?H-l

<0

an> an-l> '''>&%, ^i)}"1

Prove that
^n l^n-if

V-ll

+ ^7i-i l^n^ ~

hn

I.

7.5

Approximations to Irrational Numbers

Continuing to use the notation of the preceding sections, we now show that the convergents h n /k n form a sequence of "best" rational approximations to the irrational number
.

Theorem 7.11

We have for
C

any n

0,

and

!.

ftj

<

160
Proof.

Simple Continued Fractions

k n By
.

The second inequality follows from (7.9) and (7.7) we see that

the

first

by multiplication by

Using

(7.6),

we

replace a n+l k n

k n ^ by k n+1

to obtain the first inequality.

closer to f , that
is

Theorem 7.12

The convergents hjk n are successively

In fact the stronger inequality

Proof.
first,

\k n

hn

<

|f& n
is

A n _il holds.
it

To see that the second we use k n^ ^ k n to write

inequality

stronger in that

implies the

|ffc n

h n \<

tfk^

-h n_,

Now

to

prove the stronger inequality we observe that a n

(7.6),

+ >
1

fn

by

(7.7),

and so by

This inequality and (7.9) imply that

1
.

k n_l k n+l

We

multiply by

A: w _!

and use Theorem

7.11 to get

The convergent hjk n

fractions with
different

is

the best approximation to

less.

of

all

the rational

denominator kn or

The following theorem

states this in a

way.
Ifa/b
is

Theorem 7.13
If

|fA: n

- hjkn \for some n ^ 1, then a/b\ < |f h n \for some n ^ 0, f/zen 6 ^ n+1
fc
.

a rational number with positive denominator such that

b

>

kn

In fact if |f&

a|

<

7.5

Approximations to
First

Irrational

Numbers

161
first.

Proof.

we show
first

Suppose that the

that the second part of the theorem implies the part is false so that there is an a/b with

and

<
< \k n ^ k n+1
h n \. But the
,

The product of

these inequalities gives

\b

a\

second part of the theorem says that this implies b contradiction, since k n < k n+l for n : 1.

so

we have a

To prove the second part of the theorem we proceed again by indirect h n and b < k n+l Consider a\ < \k n argument, assuming that \b
.
\

the linear equations in x

and

y,

+
The determinant of

= 4,
1

xh n

yh n+l

a.

coefficients is

by Theorem

7.5,

and consequently

these equations have an integral solution x, y. Moreover, neither x nor y is then b == yk n+l which implies that y zero. For if x 0, in fact that k and b in contradiction to b /: If then a xhn g: y y n+1 n+1

> =

<

^ =

xk n and
,

\b
since ||

fl |

xh n

=
\

AJ

-h

^ 1, and again we have a contradiction. Next we prove that x and y have opposite signs. First, if y < 0, then xk n = b yk n+l shows that x > 0. Second, if y > 0, then b < fcft+1 implies that b < y^n+i, and so xk n is negative, whence x < 0. Now it follows from h n+I have opposite signs, and hence h n and f^^+x Theorem 7.10 that k n A n+ i) have the same sign. From the equations h n ) and y(k n+l x(gk n h n ) + y(k n+l h n+l ). Since the a = x(gk n defining x and y we get gb two terms on the right have the same sign, the absolute value of the whole equals the sum of the separate absolute values. Thus
a

= = >

\x(k n \x(k n
\x(k n

h n)
h n )\

h n )\

n
is

A tt

^ \kn -

hn

This

is

a contradiction, and so the theorem

established.
is

denote any Theorem 7.14 Let number ajb with b^.1 such that

irrational number. If there

a rational

<-;
then a[b equals one of the convergents of the simple continuedfraction expansion

oft.

162
Proof.
It suffices to

Simple Continued Fractions

prove the result in the case (a, b) = 1. Let the conof the vergents simple continued fraction expansion of f be h s lkj9 and that b < & w+1 de a/b is not a convergent. The inequalities k n suppose termine an integer n. For this the h is \gb a\ < \k n

inequality

impossible because of Theorem 7.13. Therefore we have

- AJ ^

|#

fl|<

^ 20

Using the
that

facts that <z/A 7^

AJ& n and

that bh n

afc n is

an

integer,

we

find

_!_

bk n

^ -

\bh n

ak n
bk n

h*

k
is

This implies b

<kn which

a contradiction.
(n
l)st

Theorem 7.15
convergent
Proof.

ofx ifx
have

TAe wth convergent of Ifx is the reciprocal of the is any real number >1.
a?

We

=
1,

<0

flfi,

n are the convergents for

i

= (0 * fll > and I/a x and l/x respectively, then

-

>.

If h n /k n

and

0,

fci

A;_ 2

The theorem now follows by mathematical induction.

PROBLEMS
1.

2.
is

Prove that the first assertion in Theorem 7.13 holds in case n = Prove that the first assertion in Theorem 7.13 becomes false if

if

>

1.

"Z>

3.

replaced by "6 ^ /c^," Suggestion: use I Say that a rational number a/b with 6 > the irrational number f if
|f
a|

> kn "

TT~

and

/z

1.

is

a "good approximation" to

= min

|fv

7.6

Best Possible Approximations

163

where, as indicated, the minimum on the right is to be taken over all integers x and all y satisfying < y ^ b. Prove that every convergent to $ is a "good
*'

approximation 4. Prove that every "good approximation" to f is a convergent. 5. (a) Prove that if rjs lies between a/b and cfd, where the denominators of 1 then s > b and these rational fractions are positive, and if ad be =
,

s>d.
(b)

Let

be an irrational with convergents {h n /kn }. Prove that the sequence

hn-i

hn-i

kn-l
is

kn-1

+ hn + kn

h n_ 1

^n-I

+ 2hn + 2&w

h n_ 1 &n-l

+ +

an+

A=

h n +i
^n+l

an+l^n

increasing if TZ is odd, decreasing if n is even. If a/b and c/J denote any be = 1. The terms of consecutive pair of this sequence, prove that ad this sequence, except the first and last, are called the secondary convergents; here n runs through all values 1, 2,
.

a/b Say that a rational number a/b is a "fair approximation" to f if min| x[y\, the minimum being taken over all integers x and y with < y ^ &. Prove that every good approximation is a fair approximation. Prove that every fair approximation is either a convergent or a secondary
(c)
|

convergent to f
(J)

Prove that not every secondary convergent

is

a "fair approximation".
*

Suggestion: consider
(e)

= *J2.
, ,
|
|

with limit Say that an infinite sequence of rational numbers, r l9 r2 r3 is an /v+1 < "approximating sequence" to an irrational number f if = the of if the denominators and 1, 2, 3, r^ are |f r^|, y positive form an to the "fair Prove with that I approximations" increasing j.
,

"approximating sequence." with the first term deleted, so (/) Let Sn_i denote the finite sequence of (b) that Sn_i has a n+:L terms, the last term being /z n+1 //:n+1 Prove that the infinite the terms of SQ in order, sequence of rational numbers obtained by first taking is also an then the terms of 52 then 54 , then 56 "approximating sequence"
.

to

Prove also that

this
is

rational

number

<

sequence is maximal in the sense that if any other introduced into the sequence as a new member, we no

longer have an approximating sequence. for the sequence obtained by taking the (g) Establish analogous properties terms of *S1 1} Si, Ss S5 be irrational, I = <a a lt a 2 6. Let >. Verify that
,
,
-

and

-? = (-a = < -a a2 + 1
I
,

1, I
,

a1
*

I,a 2 ,at,
> if

"
-)

if

at

>

#3 ,

<?

4,

#1

7.6

Best Possible Approximations

7.11 provides another
7.1
1

Theorem

statement of Theorem

method of proving Theorem 6.9. For in the we can replace k n+i by the smaller integer k n to

164
get the weaker, but
still

Simple Continued Fractions

correct, inequality

kn K

k ^n

Moreover the process described

any given
irrational f as

in Section 7.4 enables us to

determine for

convergents hjk n as we please. We can also use continued fractions in other proofs of Theorems 6.11 and 6.12. First we give a simple lemma.

many

Lemma

7.16

If x

is real,

>

1,

and x

+
1-

or*

<

^/~5,

then x

<

^(^5

1)

Proof.

For

real

we note

that

x- increases with

x,

and x

x~l

Theorem 7.17

many
(7.13)

rational

(Hurwitz) Given any irrational number f ?Aere exist numbers h\k such that
,

infinitely

Proof.

We will establish that, of every three consecutive convergents of the continued fraction expansion of at least one satisfies the simple inequality. Let q n denote kjk n _^ We first prove that
,

(7.14)
if (7.13) is false

4,

+ tf <
A,_i/^-i and h\k

for both h\k

A y /^- Suppose (7.13)

is

false for these

two values of h\k.

We

have
1 1

But f
that

lies

between

A^/A^ and Aj/Ar

and hence we

find, using

Theorem

7.5,

Vi
Combining
these results

we

get

ki K

k f-l K

Since the
follows.

left

side

is

rational

we

actually have a strict inequality,

and

(7.14)

Now suppose
have
(7.14) for

(7.13) is false for h\k

both j

=n

and 7

hjk i9 i = n - 1, n, n + 1. 72+1. By Lemma 7.16 we

We then
see that

7.6

Best Possible Approximations

1) and q n This gives us

165
1),

q? > i(V5
a n+I

^ < J(s +

and, by (7.6)

we

find

+ g\

n+l

and

this is

a contradiction.

The constant -J5 in the preceding theorem is best possible. In other words Theorem 7.17 does not hold zf \/5 is replaced by any larger value.

Theorem 7.18

Proof.

It suffices to exhibit

largest possible constant. Consider the irrational see that ). expansion is (1, 1, 1,

an irrational number for which ^fs is the whose continued fraction

We

>

Using

(7.7)

we can prove by induction

(V5

that

^=

(V5

l)/2 for all

^ 0,

for if ^ =

+ l)/2 then - a,)' = (K>/5 + 1) - I)"1 = K>/5 + 1). f*n = (Si A simple calculation yields A = = = 1, A = = 2- Equations (7.6) & = ^_! + ^_ and so by mathematical induc become /^ = -_! + A tion k n = n _! for ^ 1. Hence we have
1
/: fc x

A: 2

/z

._ 2 t

25

/z

7z

r lim

k^ = ^= kn
lim
n _i

=
2

r lim

t U + B1 ^

^
I

=
2

If c

is

any constant exceeding

x/5, then

holds for only a

finite

number of values of n. Thus, by

1
1

(7.9),

finite number of values of n. Thus there are only a finite 2 number of rational numbers hjk satisfying |f A/fc| < l/(cA: ), because any such A/A: is one of the convergents to S by Theorem 7.14.

holds for only a

166

Simple Continued Fractions

PROBLEMS
1.

Find two rational numbers a/b

satisfying

V2-I
2.

1
'

ITb 2

Find two rational numbers alb

satisfying

3.

Prove that the following

is

false for

any constant
rational

rational

number

?,

there exist infinitely

many

c > 2: Given any ir numbers h\k such that

4.

Given any constant

c,

prove that there

exists

an

irrational

number

and

infinitely

many

rational

numbers h/k such that

i
~k
c
'

5.

at least

Prove that of every two consecutive convergents one satisfies

1
'

hjkn

to g with n

0,

2k 2
Suggestion:

Use

the idea of

Lemma

7.16.

7.7

Periodic Continued Fractions

simple continued fraction (0

An

infinite
is

a l9 a 2

) is

there

an integer n such that ar

said to be periodic if

a n+r for

all sufficiently

large integers

r.

Thus a periodic continued

(7.15)
(i
,

fraction can be written in the

form
a n_ 1?
)

b l9 b 29

b j9 a

a l9

a n _ 1?

fl

a l9

where the bar over the a

is

a l9

a w_! indicates that

(2,

this

block of integers
}

repeated indefinitely.

For example

3} denotes

(2, 3, 2, 3, 2, 3,

and

7.7

Periodic Continued Fractions

167
6 for (2, 3}

its

value

is

easily

computed. Writing
6

we have

This

is

value 6
this

quadraticjjquation

in 6,

and we discard the negative root

(4, 1,

to get the

we

Vl5)/3. have f (4,

(3

As
1
,

a second example consider 6), with 6 as above, and so

271). Calling

+
These two examples

illustrate the following result.

is

Theorem 7.19 Any periodic simple continuedfraction number, and conversely.

Proof.
its

a quadratic irrational

Let us write f for the periodic continued fraction of (7.15) and for a n _J = (a a l9 (a Q9 a l9 purely periodic part, a^ l9 0). Then

equation (7.8) gives

fl

0ft n -i

h n_2

and

this is

number or a

a quadratic equation in 6. Hence 6 rational number, but the latter

is
is

either a quadratic irrational

ruled out by

Theorem

7.7.

Now

f can be written in terms of 0,

00

+
to

4
(
,

where
is

ra'/?'

an<i

^/? are the

last

two convergents

#!,, ^).
,

But

6
0,

of the form (a + ^/b)[c, and hence f is of similar form because, as with we can rule out the possibility that f is rational.

To prove

the converse, let us begin with any quadratic irrational f or

of the form f = f 0. The (a + V*)/c, with integers a, 6, c, fc > 0, c integer b is not a perfect square since f is irrational. We multiply numerator

and denominator by

|c|

to get

ac

+ ^^ g2
c

^o

ac

+
c

Vfrc

according as c

is

positive or negative.

Thus we can
\/d

write g in the

form

m +

168

Simple Continued Fractions

(d ml), d,

where q

d not a perfect and q Q are integers, ? 5^ of its formulation a can we form simple get square. By writing | We sha11 P rove that tne continued fraction expansion (a a l9 a 29 >

>

in this

equations

(7.16)

fl

,=

[fj,

f,

m + V^
i

^^^

a i9 and irrationals Si in such a define infinite sequences of integers i9 q i9 will have the continued fraction we hence and that hold, way equations (7.7)
expansion of |
In the
first
.

a* = [So]- If S^ ma = - /wf+l )/?0 ? +1

.

place,

we
f

start

with

o>

?o as

determined above, and we

let

?,

a. are

known, then we take

m i+1 = a#

m,-,

a [f,+1 ]. That is, (7.16) (m,+1 + Jd)lq are at least real. a that m determine i i9 q i9 sequences Si, actually does the m i and q i are that to induction mathematical we use Now prove 0. If it is true i for holds This and q i (d such that q i n%).
(rf
,

^=

M Mm
t

integers at the zth stage,

we observe
d

that

equation

- mLi

m i+l = a^
d

is

an

integer.

Then

the

2
, *

+
,

2a i m i

- a^
2

to

establishes that

q i+l

is

an

were,

we would have d
q
i

integer.
,

Moreover q i+I cannot be zero, since if it m^+i whereas d is not a perfect square. Finally,
q i+l
\

we have

(d

/wj+1 )/y i+1 , so that

(d

/wj+1 ).

Next we can verify that d

__

which
the

verifies (7.7)

and so we have proved that

'

<a

a l9 a 2

'

)>

with

^ defined by
^-

(7.16).

the conjugate of Si, that is, v ^)/?i- Since the (m i the we get the of the of a conjugates, quotient conjugate quotient equals

By

we denote

equation

7.7

Periodic Continued Fractions

(7.8).

169
g' n

by taking conjugates in

Solving for

we have

As

TZ

different

tends to infinity, both h nr. l /k nr . l and /z w _ 2 /& n _ 2 tend to from f, and hence the fraction in parentheses tends to

which

is

L Thus

for

sufficiently large n,
is

say n
is

> N where

JV

is

fixed, the fraction in parentheses

is

positive,

and
for

g' n
/I

negative.

But g n

positive for n

and hence

fn

- &>
It

>

A?".

and hence q n

>

for n

>
d

Applying (7.16) N.

we

see that this gives

2^lqn >

also follows

from

(7.16) that

- mz n+l ^d,

qn

q n q n+1

> N. Since 6? is a fixed positive integer we conclude that qn and N. Hence the can assume only a fixed number of possible values for n ordered pairs (m n q n ) can assume only a fixed number of possible pair values and for n TV, and so there are distinct integers/ and k such that m s = ^ We can suppose we have chosen y and k so thaty < k. By (7.16) this q.
for n

>

>
fc

=m

fc

implies that

,-

| fc

and hence that

The proof of Theorem 7.19 is now complete. Next we determine the subclass of real quadratic
purely periodic continued fraction expansions, that

irrationals that
is,

have

expressions of the

form

'

'

(a

0i,

<O-

Theorem 7.20 The continuedfraction expansion of the real quadratic irrational 1 < f < 0, w/zere f number f is purely periodic if and only if > 1 and
denotes the conjugate of
P/-00/.
.

First

we assume

that I

>

and

<

'

<

0.

As usual we

write

for

and take conjugates

in (7.7) to obtain

(7.17)

77t+i
f

*<

Now
1/fi+i
==

^ +

for

all

z,

even for

=
<

0, since

<

-1, and we have -1

l/

<
1

+1

mathematical induction, that

a,

> 1. Hence if ^ < 0, then < 0. Since -I < |Q < we see, by < holds for all ^ 0. Then, since
i

+1

by

(7.17),

we have

170

Simple Continued Fractions

Now f is <y <

fc.

a quadratic irrational, so f, and Then we have f} =

for

some

integers

j and & with

Thus
us

^-

=
t-

fc

implies

^_ x

f^x. Ay-fold iteration of

this implication gives

k _ i9

and we have

To prove
(a
,

the converse, let us

,

assume that f
-

is

purely periodic, say

positive
integers.

=
Then

<%,
flr

^_i),

where

<2

a l9

<z

n _i

are

>

1.

Also, by (7.8)

we have
n-

f fcn-l

fc

n-2

Thus f

satisfies

the equation

and its conjugate % Since f This quadratic equation has two roots, 1, we need only prove that/(#) has a root between 1 and in order to establish
.

>

that

-1 <

'

<

0.

We

will

do

this

opposite signs. First we observe that/(0) 0. Next we see that for n for i I

by showing that /(-I) and/(0) have = h n_ 2 < by (7.6) since at >

>

/(-I)

_!

k n_ z

+
>

h n _i

h n_ 2

^ k n_s +
Finally, if n
this

h n_ 3
1)

0.

1,

we have /(

=k

k_^

h_

>

0,

and

completes the proof.

We now
integer

d not a

turn to the continued fraction expansion of \[d for a positive perfect square. get at this by considering the closely

We

related irrational

of Theorem 7.20,
(7.18)

number Jd + [\ld]. This number satisfies the conditions and so its continued fraction is purely periodic,

We
which

can suppose that we have chosen r to be the smallest integer fot V^ + [Jd\ has an expansion of the form (7.18). Now we note that

7.7

Periodic Continued Fractions

(a t ,

171

= fr = ) is purely periodic for all values of z, and that Furthermore are all different from since other f 2r S r -i Si, 2 f wise there would be a shorter period. Thus Si = if and if of the is only form mr. =
a i+ i,
.

^=

Now we
because
1
|

can

start
[Vrf]

with So
2

(rf
jr

= V^+

[V5], #

1,

m =

[VS] in (7.16)

).

Then, for ally

0,

m + ya =

Sjr

=
i

for

and hence ^ r = 1 since the left side is rational and \J~dh irrational. Moreover = 1 for no other values of the subscript /. For qi = 1 implies ff = q

m +
i

\fd,

but

we have -I
Thus
Si

=
t

^ has a purely < m - ^Jd <Q

i i

periodic expansion so that, by

y

Theorem

7.20

^/d r.
1

<m <
i

^/d,

and hence m,
z.

[V5].

S o and

is

a multiple of

We
/Wf_

also establish that q i

does not hold for any

implies

S = -mi V^ by < V^ > and

1

(7.16),
i

and by Theorem
0.

7.20

For j = 1 we would have

m + ^d <

But

this implies

^<m

<

^d
S

1,

which

is

impossible.

Noting that

[^/d+ [Jd]]

2[Vrf] 9

we can now

turn to the case

\/d.

Using

(7.18)

we have

with a

=
Q
Q
ar

gr

= apply (7.16) to ^d + [^d], q = 1, m = [V5] we have = = d But we can also apply (7.16) to V5 with 2[V5], /wj [V5], q [V^] = 1, TW = 0, and we find a = [V5], m = [V5], ?! = [V^] The value of # is different, but the values of m and of q are the same in both cases. Since S = (/w< + V^)/^ we see that further application of (7.16) the same values for the a for the m and for the q^ in both cases. In yields other words, the expansions of \ld + \\jd\ and \jd differ only in the values of a and m Stating our results explicitly for the case V5 we have the follow
When we
2 l
.

</

l5

l9

i9

i9

Q.

ing theorem.

Theorem 7.21

If the positive integer

is

not a perfect square, the simple

continuedfraction expansion of^fdhas theform \j~d

(0

a l9 a 2

a^, 2a

172
with a Q
(7.16),

Simple Continued Fractions

=
we

[V5]. Furthermore with

A0z;e qt

=
/,
\

jd, q

1 1

if and only ifr

andq

0, in equations holds for no subscript L

m =

Here

r denotes the length

of the shortest period

in the

expansion of^fd.

PROBLEM
1.

For what positive

the quadratic irrational ([>/<*] integers c does

+ \fd)fc

have a purely periodic expansion ?

7.8

Pell's

Equation

The equation z 2 dy* = N, with given integers d and N and unknowns x and y, is usually called Pell's equation. If rf is negative, it can have only a a 2 the equation finite number of solutions. If d is a perfect square, say d = N and again there is only a finite number of reduces to (x ay)(x + ay) solutions. The most interesting case of the equation arises when d is a positive continued fractions are integer not a perfect square. For this case, simple
,

very useful.

expand \Jd into a continued fraction as in Theorem 7.21, with conwith f = vergents hjk n and with q n defined by equations (7.16)
,

We

V^
=

qQ

1,

m =
Q

0.

Theorem 7.22
n ~l
(

Ifd is a positive
all integers

integer not
1.

a perfect square, then h n

dk n

l)

q n+ ifor

Proof.

From
2

equations (7.8) and (7.16)

we have
n+l

+ln
f n+l^n

n-1

fc-l

(W n+1

We
two

irrational part

simplify this equation and separate it into a rational and a purely much as we did in (7.19). Each part must be zero so we get n+l from them. The final result is equations, and we can eliminate

where we used Theorem 7.5

in the last step.

Corollary 7.23 Taking r as the length of the period of the expansion of y d, as in Theorem 7.21, we have for n : 0,

7.8
It

Pell's

Equation

173

certain values of
2 solutions of x

can be seen that Theorem 7.22 gives us solutions of Pell's equation for N. In particular, Corollary 7.23 gives infinitely many
dy
2

by the use of even values

is

nr.

Of course

if r is

even,

odd, Corollary 7.23 gives infinitely many 2 2 1 the use of odd integers n 1 solutions of x The next dy by 2 1 can be obtained from theorem shows that every solution of x 2 dy =
all

values of nr are even. If r

the continued fraction expansion of Jd. But first we observation: apart from such trivial solutions as

make

x2

dy combinations of signs solutions x > 0, y > 0.

1
,

all

solutions of x 2
x,
y.

dy

=N

#=1,

this simple

of
all

fall

into sets of four

by

Hence

it is

sufficient to discuss the positive

Theorem 7.24

conver'gents to the continued fraction expansion of^/d be

Let d be a positive integer not a perfect square, and let the hjk n Let the integer 2 s y = t ofx 2 dy = satisfy \N\ < \fd. Then any positive solution x
.

with (s 9

1)

satisfies s

hn

k n for some positive

integer n.

Proof.

real

be positive integers such that (E, M) = 1 and < a < Vp- Here p and a are cr, where Vp is irrational and numbers, not necessarily integers. Then
Let
2

and

pM =

! M
and hence

M(E
Also

< E/M

V/> implies EI(M^/p)

>

1,

and therefore

M
By Theorem
If
7.14,

2M 2

jE/M

is

a convergent in the continued fraction expansion

p
2

N>
t,

0,

we

take a
2

=N

d,

E=

s,

M=

t,

and the theorem holds

in this case. If

N < 0,

then
s.

(l/rf>

= - JV/rf,
is

and we take a

-JV/J, p

1/rf,

E=

M=
=
is

We
7.15

find that

f/s-

Then Theorem
Theorem 7.25

shows that

s/t

a convergent in the expansion of l/^. is a convergent in the expansion of

^.

among x
\J~d.

h n9 y

1 are to be All positive solutions of x dy found &, w/zere /z n /A;n are /Ae congruents of the expansion of

Ifr

the period

of the expansion of^fd, as

in

Theorem

7.21,

and ifr

is

<I74
2 even, then x

Simple Continued Fractions

dy*

has no solutions, and all positive solutions of On the k nr _^for n = 1, 2, 3, x == h nr_ l9 y x I are dy given by = // solutions iz* & x h /w#*? other hand if r is odd, then nr _i nr _ l9 y z = 1, 3, 5, a/?J a// positive solutions of by use of n ofx
1

dy*=l
=
I

x2

dy

by use ofn
is

=
,

2, 4, 6,

Proof.

This result
2

= [V^l > so the sequence A /z 1? A 2 the first solution that appears, & is strictly increasing. If 19 y x denote x l9 and hence y x have will x we y^ then for every other solution y frac continued of means solution least this found also. by
Of X2

The sequence of pairs

a corollary of Theorems 7.21, 7.22, and 7.24. wil1 include all positive solutions <7z )> (*i> ^i)>
"
'
*

dy

I.

Furthermore,

we

let

>

>

Having

tions,

we can

positive find all the remaining positive solutions

by a simpler method.
2

dy = 1, d Let xly y be the least positive solution of x* are solutions not a perfect square. Then all positive being a positive integer are x the = 1,_2, 3, where n andy n integers defined by given by x n y n for n

Theorem 7.26

values of x n and y n are determined by expanding the power and and the purely irrational parts. For example, equating the rational parts, 3 and y3 y\d. Src^ z3 x\ 2/iV^) so that x 3 2/ 3 \/5 G*i

The

=
is

+ Ix^d

Proof.

First
71

we
,

establish that

xn yn
,

a solution.

We

have x n

y n \/d

(%

2/iV^)

since the conjugate of aproduct is the product of the conjugates.

Hence we can write

is
i

Next we show that every positive solution can be obtained. Suppose there a positive solution s, t that is not in the collection {x n y n }. Since both
,

2/iV^ and

+ V^
?

are g reater than

such that (x I

y^fd)

^s +
.

tjd

<
1

(x^

be some integer m there_must m+l We cannot have + y^d)

.

+ 2/iV^) = s + tJd> for this would imPty x + 2/mV^ = ^ + ^\/5, w and hence S = x m9 = y m Now (x - y ^/d^ = (^ + 2/iV5)- and we can
w
(*i
t x
,

multiply the above inequality

by

(x^

y-fjd)

to obtain

\<(s
Defining integers a and i by a

+ bjd =

(s

t\/d)(x^

y-L ^fd)

m we have

7.9

Numerical Computation
is

175

so a, b

a solution of x 2

But then

<

such that < a + b\J~d < o^ + y-cJd. and hence a < + < (a b^~d)~ V5 < Now we have a = i(fl + iVrf) + i( - &V5) > 4 + > 0, W5 = i(fl + W5) - 4(a - ^V^) > 4 - 4 = 0,

dy
1

Z>

fl

so a, b

is

a positive solution. Therefore a

>

x l9 b

>

2/ l9

but this contradicts

false.

<s + &V^ < + 2/iV^ and hence our supposition was are solutions given by x n ,y n ,n = 1, 2, 3,

xi

All positive

It

may be
n.

negative

For

N different
dsl

noted that the definition of x n y n can be extended to zero and They then give nonpositive solutions.
,

from

there are certain results that can be proved, but

they are not as complete as what we have shown to be true in the case 2 For example, if x l9 y^ is the smallest positive solution of # 2 dy
if
(''o

rl

N, then

x2

dy = N. However, there is no be obtained in this way starting from a fixed

2

StfJd)(x

integers n
,

r n9

sn

can be defined by

rn

N= = I, and + sn\fd =
1
.

yi\fd)

and

it is

easy to

that r n , s n are solutions of assurance that all positive solutions can

r
,

show

PROBLEMS
The symbol d denotes a positive integer, not a perfect square. 2 2 1 is solvable, let x l9 y l be the smallest 1. Assuming that x dy =
solution.

positive
is

Prove that x 2

yz

defined by #2

2 smallest positive solution of x

x2
n

dy

1, 3, 5,

with n
2.

Also prove that all solutions of -1 are given by xn yn where xn + yn jd = (^ + y^d) 71 with 2 and that all solutions of x 2 dy = 1 are given by x n9 yn 7,

dy
,

+ y^j~d =

(x

+ y^Jdf

the

1.

2, 4, 6, 8,
if

Prove that

dy

=N

has one solution,

(x\

it

has infinitely many.

Suggestion:

use

the

identity

- dy$(x\

dy%)

(x

xz

- dy:y^ 2 -

2 1 has no solution if d s 3 (mod 4). Prove that x 2 dy = 4. Let J be a positive integer, not a perfect square. If k is any positive integer, 2 x2 dy = 1 with prove that there are infinitely many solutions in integers of

3.

k
|

y.

7.9

Numerical Computation
in finding a simple continued fraction (7.7) must be used. However

The numerical computations involved

can be rather lengthy. In general the algorithm

176
if
is

Simple Continued Fractions

a quadratic irrational the work can be simplified. It is probably best to use (7.16) in a slightly altered form. From (7.16) we have

= d-n^ _

Starting with f

rm
L

+
^

(/M O

V5)/^

(^

wg),

we

obtain, in turn,

Jdl
,

m =
2

-ml

a^!

-m

[wii

fm,.!
fl<-l=
L

y<q
'

^-i

=
The formula ^^ i+1

ft_ 2

fli-i

rf

numbers, this procedure is For quite small numbers

/n|+i serves as a good check. out. fairly simple to carry

it is

Even

for large

often easiest to obtain the expansion directly.

as follows
:

For

example for

V^ we can compute

&- 3-1
In this case f3
fl

'

f x so

we

stop.

We

have

=l,

fli=l,

^2

= 2,

a3

= 0! =

1,

V3

<1, 1,2).

Notes on Chapter 7

177
is

When
from

a continued fraction

known

the convergents can be obtained

(7.6).

The work can be

systematized.

The following example,

for A/3>

demonstrates a convenient method.

hQ
hi
/z

= 1, = 2, = 5,

kQ
k,

k2

= = =

NOTES ON CHAPTER
A

completely different approach to continued fractions, specifically with the continued fractions arising naturally out of the approximations rather than the other way about, can be found (for example) in Chapter 1 of the book by J. W. S. Cassels listed in the General References, page 269.

s
Elementary Remarks on the Distribution of Primes

8.1

The Function

Tt(x)

The

makes it clear that the primes are distributed the natural a very irregular way. Theorem 1.18 shows that numbers in among there are arbitrarily large gaps in the sequence of primes. The proof of
discussion in Section 1.3

Theorem
r

that the rth prime p r

shows that there are infinitely many primes but also no greater than Hi/?j + 1, the product of the first I 1 primes plus LA minor change in the proof shows that p r ^ TI^Pi
is

1.17 not only

ifr>2.
In this chapter

we abandon our convention

that letters of the

roman

alphabet represent integers.

Definition 8.1

For real x we exceed x. Thus for example

TT(-l)

let TT(X)

denote the number of primes that do not

= 77(1) =

0,

7T(2)

77(5/2)

1.

Theorem

8.1 (Tschebyschef)

There exist positive constants a and b such that

log x

log x

for
178

x^2.

8.1

The Function

TT(X)

179
integer and/? a
I

Proof.

For n a positive

prime
\
.

let j^A

be the largest power ofp

/o n \

that divides the binomial coefficient

According to Theorem 4.2

*
We
define the integer

v by the inequalities p

<;

< p l+v
for;

Clearly v9

exists

and

is

unique.
9

Then
\-~j)

P
2^P
j

- =

>

v We also

have for every j

and hence

[2/t//]

2[//]

for ally i>

1.

Using

this in (8.1)

we

obtain

(8.2)

ft, =

and therefore

On

the other

hand
,

if

72

<

j?

<;

then

/?

(2n)\

and ^

</z! so

we have

n
I

which, along with

(8.3), gives us

*<IL

and hence, changing each;?

in the first product to n,

(M)
2/Z

.-~a)a<w~>.
[
]

But

^
n /

(1

2n
l)

2 2n and

/2n\
\

(2w)(2n

1)

(n
i

1)

TT 2 = = -pr n + 7 > "~

'

n
.

J=i

=1

Using these
(8.5)

inequalities in (8.4)

and taking logarithms, we obtain

where we must assume

>

in the first inequality.

180

Elementary Remarks on the Distribution of Primes

let

exceed

For any real x ^ 2 we x. Then we have x

2n be the greatest even integer that does not 2n, n ^ 1, 2n + 2 > x, and hence
(2n

2)log2

log_2

^_
log x

log (2n)

log x

4 log z

To
(8.5)

we

get the other half of the inequality in the theorem, in the first part of take 2n to be an exact power of 2, say 2 r with r 3, to get

^(2
Replace r by 27; 2/ obvious result ?r(2 2 )
1,

r-x

^ 2'/(r 3,

1).

2/

2,

and add these

inequalities to the

<

2 2 to obtain, for j

^ 2,

In the final two sums

we

replace r

by/ and

respectively to get

Now
(8.6)

2^'

+1

<

2^+1 // since/
2
7r(2

<

2',

and hence we have for/

2

0<2-2

/;

or

^<7^ 2 there
4
is

Note that

this

holds also for/

"- 2

=
1?'

y^

such that 2 23

<x^

1. Now for any real x and we can write

an integer

o-^og*
2
Substituting these in (8.6)

81og2
-

we

get

^0*0
a;

<

16
7

""

32 log 2
log
Z>

cc

Therefore
proved.

we can

take #

(log 2)/4,

32 log

2,

and the theorem

is

We

have found values of a and b that

suffice,

but they are by no

means the best possible values. The theorem we have just proved tells us something about how numerously and how scarcely the primes are distributed. Since there are infinitely many primes, we cannot say that there are more natural numbers than there are primes. However the ratio 7r(n)/n represents the proportion of primes in
the
first

creases, we

n natural numbers. Since 7r(n)/n i/log n tends to zero as n in are led to say that the primes are scarcer than the natural numbers.

<

8.2

The Sequence
is

of Primes
all

181
positive integers are composite."

This

often stated as: "almost

tells

Of

course the theorem

us a good deal more. Since

x/log x
the function TT(X) is of order re/log x. nor too scarce. Since

The primes

are neither too

numerous

^/x

<
is

logx = ->

as x

->

oo,

the sequences of squares

number theorem mentioned

theorem.

scarcer than the sequence of primes. The prime in Section 1.3 is a refinement of the present

PROBLEM
1.

Prove that v, as defined following

(8.1), is

equal to

riog2n~|
.

Llog/?J

8.2

The Sequence of Primes

,

Results concerning the size of the rth prime, pr can also be used to describe how numerous the primes are. Our first result is essentially a corollary to

Theorem
Theorem

8.1.

8.2

There exist positive constants c and d such that cr log r

<

p <
r

dr log r for r

2.

Proof.

Using Theorem

8.1

and the

fact

that^ r

^ r we have

The other way around we have r = 7r(p r ) >^ap r l\ogp T If r is large so isp r and there is a constant k such that log/v/V/V < a if r = k Then for r = k
.

>

>

Pr

182

Elementary Remarks on the Distribution of Primes

r

hence
is

>

>//V>

l5Pr

<

larger than the largest

2 log r and therefore ap r number among

^3
'

<

logp r

<

2r log

r.

If

2
9

#2

...
'

JPfc-l

2 log 2'

3 log 3

(fc

1) log

(fc

'

1)

then

/? r

<

Jr log r for r

^ 2.

Theorem

8.3

The

series

diverges.

Proof.

For

>

we have
pr
dr log r

and the

series

2JL 2
.

l/( r

r ) diverges.
is

Alternate proof
8.1,
it is

Whereas the proof just given

based on Theorem 8.2 and

possible to establish follows. First we prove that

all positive integers

Theorem
'

l/

terms of simpler concepts, as where diverges, 2' denotes the sum over
8.3 in

r2

that are square-free. Given any positive integer m, let 2 be the largest integer square that divides m, so that r 1, and mfr is

square-free.

of a perfect

Thus every positive integer is uniquely expressible as a product square and a square-free positive integer. It follows that for any

positive integer n

3<n

because when the two sums on the

left

are multiplied together the result

is

collection of reciprocals of integers 1/ra, n. including all integers as n tends to infinity the sum on the right is unbounded because the infinite

m<

Now

sum 1/Jw is a divergent series. But the first sum on the left is bounded as n 2 tends to infinity because the infinite sum a convergent series. Hence I// is the second sum on the left is unbounded, and this proves that 2' 1/& is a

divergent series.

Next suppose, contrary

to a value ft, where the

to what we want ta prove, that 2 1/P converges sum is over all primes p. In the next step of the argument

we

use the fact that by dropping all terms beyond x in the power series x x holds for all 1 expansion of e or exp (x), the inequality exp (x)

>

positive real

numbers

x.

Hence for any

positive integer n

exp

(/J)

>

exp

l^l/p] W<n /

* <W

n exp

(l/j>)

> ft (1 + *<"

1/p)

>

V
t?n

(1/fe)

8.2

The Sequence

of Primes

183

from multiplying out the factors (1 + l/p). Now unbounded as n tends to infinity, whereas exp (ft) is a fixed real number. Thus we have a contradiction, and the proof is complete, If a series 2=i a r f positive terms converges for all values of k >/and diverges for all values of k </,/being fixed, then/is called the exponent of convergence of the sequence a r The sequence \\p r has exponent of convergence 1. If k > 1, then 1/pJ < 1/r* and 2rli 1/r* < k ^ 1, it converges. If follows from Theorem 8.3 that 2JL X I//?* diverges. The sequence a r = 1/r 2 has exponent of convergence In fact, if ar = 1+e 1+ 1+E if r i> 2 and the a 6 then < > 0, r ], 1)< 2/r l/(r l/[r sequence
where the
the last
last step follows
is

sum

ar has exponent of convergence (1 + s)" 1 < denser than the sequence consisting of [r 1+c ].

1.

In a sense the primes are

Theorem

8.4

There

is

a constant k such that

By Theorem

<

/clog log x

if

3.

Proof.

8.2

2c log 2

Jr-i

log

2c log 2

c Js

log

+ - log log x - - log log 2,

2clog2
and this is less than k log log x for x ^ 3, if fc is large enough. The method used in this proof occurs frequently in number theory. It is the method ordinarily found in proofs of the integral test in the theory of series. For a monotonic function /(#), we compare ^n=M/(n ) with dx. Geometrically, J^i/(^) dxis the area under the curve y =/(), jjf-.i/(aO
whereas 2^-Jif/( /I ) re P resents the area covered by rectangles having unit N. + 1, bases and altitudes /(n), n = Af In contrast with Theorem 8.3, Theorem 8.4 shows that the primes are not too numerous.
,

Theorem
Proof.

8.5

Ifx^l

then

Tlv^x p

<

4*.

This theorem

an odd integer n

<

4n

4a: -

x 3. If it is true when x is obviously true for 2 x 2 since Tlp^xP == n 3, then it is true for n Therefore we need to consider only odd integers n with
is

^ < ^ < +

184

Elementary Remarks on the Distribution of Primes

^ 3. The proof is now by induction on the odd integer n. Noting that the theorem holds for n = 3, we assume the result for all odd integers greater than 1 that are less than some odd integer n ^ 5. We define k = (n l)/2 where the sign is chosen so that k is odd. Then k ^ 3. Now
n
(8.7)

(?)
is

W = kl(n-k)l
"1

that

and n k k </?
(8.7)

even and n
77,

then/?

k = 2k ~=f 1 k odd and/? \n\ p -f

9

^k +
&!,/? f

1.

If p

is

a prime such

Jfc)!.

Hence from
so

we

see that the product of all such primes divides j^l

and

b th these binomial coefficients appear in the n This implies that < 2 n ~l Using this and the expansion of (1 + l) \ ' induction hypothesis we have
(k)
(n
an(i
.

But

k)

m
=

JJ p
because
72

JJ p
1
.

JJ p

< 4*

2"-

2 n+2fc

-1

2 2n

4n

2A:

PROBLEMS
1.

Find the exponent of convergence of the sequence of the reciprocals of the

increasing positive integers that lack the digit 9

when

written in ordinary

decimal notation.

Give an independent proof of Theorem 1.18 ("There are arbitrarily large gaps in the series of primes") by using Theorem 8.2. 3. Write s r for the sum of the first r Prove that there are
2.

primes.

positive

constants a l and b : such that for r

2

>

<2jr

log r

<

sr

<

log

r.

4. (a)

Let

{a^}, {bj} 9 {c^}

limit infinity.

lim Cjjbj
is false.

be increasing sequences of real numbers, each with Say that a t is asymptotic to b jy written a j bj, if and only if Prove that a s bj implies log a,log bj, but that the converse

(b) Ifcij
(c)

bj prove that bj a^ ifaf Prove that lim (log a s)ja s = 0.

~ by and GJ ~ d
if

prove that

(d) If lim

cja s

0,

prove that a s

~ bj

and only

if

*,

~ bj +

8.3

Bertrand's Postulate

185

sketch is now (e) The prime number theorem states that rr(ri) /log . n log n. Verify the steps in this given that this is implied by p n proof and (n + 1) log (n + 1) ~7zlogrc ~/? n prove the converse result. Note that p n+1 For integers k > 1 define n by p n ^ k < l9 so that n is a function of k,

and
log
/:

k~pn

Also n

p^

ir(k)

and hence &

log 7r(k)

log log rr(k)

~ log

TT(^). It

and TT(/;) log TT(&) log follows that k n(k) log k or

8.3

Bertrand's Postulate

The following theorem was conjectured by Bertrand, but proved by

Tschebyschef.

Theorem
n

8.6
2n.

For every positive integer n

>

there

is

a prime p such that

<p<

Proof.

For n

To verify the theorem for the smaller values of n we take;?

respectively,

128 the result will be established by a general argument. 3 5 for n 2, 3

and

7
is

for

4^<^6
?<;<; ^ ^ 23 ^ n ^ 43 ^ ^
13
TI

12

23
43
83
131

22
42
82
127.

83

^<;

Next, suppose the result is false for some integer n and (8.2), with this supposition we have of

128.

By

the definition

For any prime p

in the range 2n/3

<p ^

we have

p^3,
and so by
(8.1)

a
J

>-jip^2n,
3

l^-<-, 2
p

2^p

For any prime p ^ = 1 and PQ ^

in the range
1.

^2n

<p ^
/?

2/2/3

we have

/?

> 2,

hence

For any prime

satisfying

V^n we have p" z

186

Elementary Remarks on the Distribution of Primes

v"

< In.

Using these

facts in (8.8)

we

obtain

P"

_n

P"

n ^

n
If

P.

>

3, the

number of primes
rr(y) last
:

integers rg y,

and so

(1

^ y is at most the number of positive odd + y)/2. Applying this simple observation to
we
see that the

the

first

product in the
\J2ri)j2.

expression

number of factors

is

at

most (1 + we have

We

apply Theorem

8.5 to the

second product and then

2n
(8.9)

<

(2 W )<

+V

^3
.

is
I (2n\

the largest of 2n
2/z

terms in the binomial expansion of

if

(1

2n
l)
,

or the largest of

terms

we combine

the

first

and

last

terms

into a single term 2, so

we have

This with (8.9) implies

n l (2n)~ 2*

Also

(3

>/2n)/2

<

(2V2/0/3 for n

128,

and hence

Taking logarithms and then dividing by V^? we get

(8.10)

Vfl log 2

But V/z log 2

-V2 log (2)

is

zero

< V? log (2ri). for n = 128 = 2

and

which
n

is

128,

128 because log positive for n and so the Theorem holds for n
is

>

1/4.

Hence

(8.10) is false for

128.

This proof

fairly typical

of a great

the proof one uses inequalities

and

many proofs in number theory. In estimates the sizes of various expressions.

Notes on Chapter 8

187

Often these estimates are good enough to prove the theorem for large values of n, say, but are too crude to yield the desired result for smaller n. We are
then forced to take care of these smaller n by more special methods.

PROBLEMS
1.

Prove that for every positive

real

number x

>

there

is

x
2.

<p <

a prime/? such that

2x.

impossible in integers m, n > 1, k > 1. be positive integers, k >l,r >l. Prove that there is a prime whose digital representation to base r has exactly k digits. 4. For this problem include 1 as a prime. Prove that every positive integer can be represented as a sum of one or more distinct primes.
3. Let

Prove that n\

= mk is

k and

5.

Prove that the following three properties of a positive integer n are equivalent
primes

(i) all

^ ^Jn
1
;

are divisors of n\
(iii)

(ii) all

to n are primes or

every composite integer

positive integers <n has a factor in

<n and prime

common

Furthermore, prove that only a finite number of positive integers have these properties, and find them. Suggestion: if n is large enough there are
n.

with

distinct

primes /? 1 ,/? 2 ,/7 3 ,/?4 such that

<p, <

-^ for; =

1, 2, 3, 4.

NOTES ON CHAPTER

The final part of the proof of Theorem 8.1 has been improved by use of an arrangement by E. L. Spitznagel, Jr., "An elementary proof that primes are scarce," Amer. Math. Monthly, 77, 396-397 (1970).

9
Algebraic

Numbers

9.1

Polynomials

Algebraic numbers are the roots of certain types of polynomials, so it is natural to begin our discussion with this topic. Our plan in this chapter is to proceed from the most general results about algebraic numbers to stronger numbers. In this process of specific results about special classes of algebraic

proving more and more about less and less, we have selected material of a number theoretic aspect as contrasted with the more "algebraic" parts of the theory. In other words, we are concerned with such questions as
divisibility,

uniqueness of factorization, and prime numbers, rather than

questions concerning the algebraic structure of the groups, rings, and fields arising in the theory.
that we will consider will have rational numbers for Such polynomials are called polynomials over Q, where Q denotes the field of rational numbers. This collection of polynomials in one variable x is often denoted by Q[x], j ust as a ^ polynomials in x with integral coefficients is denoted by Z[x], and the set of all polynomials in x with coefficients in any set of numbers F is denoted by F[x]. That the set of rational numbers forms a field can be verified from the postulates in Section

The polynomials

coefficients.

2.10. In a

polynomial such as

f(x)

a xn

+ a^-i + =

.--

an ,

0,

the non-negative integer n is called the degree of the polynomial, and a is 1 the polynomial is called "mo/izc." called the leading coefficient. If a
,

Since

we

assign no degree to the zero polynomial, we can assert without

188

9.1

Polynomials
is

189
the

of two polynomials exception that the degree of the product the polynomials. degrees of

sum of the

A polynomial/ (z) is said to be divisible by a polynomial g(x), not identically = g(x)q(x) and we write zero, if there exists a polynomial q(x) such that/(V)
*(*) !/(*)

Also, g(x) is said to be a divisor or factor off(x). The degree of g(x) here 0. does not exceed that of/(x), unless/Or) is identically zero, written /(a) This concept of divisibility is not the same as the divisibility that we have

considered

earlier.

In fact 3
|

of degree zero, whereas

it is

7 holds if 3 and 7 are thought of as polynomials not true that the integer 3 divides the integer 7.

Theorem

9.1

To any polynomials f(x) and g(x) over

with g(x)

such thatf(x) correspond unique polynomials q(x) andr(x) == lower or r(x) is of where either r(x) degree than g(x).
Proof.

= g(x)q(x) +

^ 0,

there
r(x),

== In case/(V) = orf(x) has lower degree than g(x), define q(x) and r(x) =/(). Otherwise divide g(x) into f(x) to get a quotient q(x) and a remainder r(x). Clearly q(x) and r(x) are polynomials over Q, and either = or the degree of r(x) is less than the degree ofg(x) if the division has r(x) been carried to completion. If there were another pair, q^(x) and r (x) then we would have
9

f(x)

= g(x)qi(x) + r^x)

r(x)

r (x)

= g(x){q

l (x)

jr(a?)}.

Thus g(x) would be a

follows that q(x)

divisor of the polynomial r(x)

r : (x)

which, unless

than g(x). identically zero, has lower degree

Hence

r(x)

r (x)

0,

and

it

qi(x).

Theorem

9.2

Any polynomials f(x) and

have g(x), not both identically zero,

a common

divisor h(x) which is

a linear combination off(x) and g(x). Thus

h(x)\f(x),h(x)\g(x),and
(9.1)

h(x)

= f(x)F(x) + g(x)G(x)

and G(x). for some polynomials F(x)

Proof.

the polynomials of the form (9.1) that are not identically it by h(x). If h(x) were zero, choose any one of least degree and designate us would r(x) 9.1 h(x)q(x) /(a?) not a divisor of /(a?), Theorem give then But = than lower r(x) =f(x) and r(x) of degree h(x). with r(x) is of the form (9.1) in g(x){G(x)q(x)} which F(x)q(x)} h(x)q(x) =/(){! contradiction with the choice ofh(x). Thus h(x) \f(x) and similarly h(x) g(x).

From

all

190

Algebraic
9.3

Numbers
zero,

Theorem

there corresponds
(1)

To any polynomials f(x) and g(x), not both identically a unique monic polynomial d(x) having the properties
9

d(x)\f(x) d(x)\g(x); as in (9.1); (2) d(x) is a linear combination off(x) andg(x), is a divisor of d(x), and thus (3) any common divisor off(x) and g(x) there is no common divisor having higher degree than that of d(x).
Define d(x) c'^x), where c is the leading coefficient of h(x), so Proof. that d(x) is monic. Properties (1) and (2) are inherited from h(x) by d(x).

properties
d^(x)

equation c-*f(x)F(x) + c~^g(x)G(x), Equation (9.1) implies d(x) shows that if m(x) is a common divisor of f(x) and g(x), then m(x) d(x). that d(x) and d(x) both satisfy Finally, to prove that d(x) is unique, suppose d: (x) and d^(x) d(x), hence have then We d(x) (1), (2), (3).

and

this

= q(x]d(x)

and d(x)

This implies q(x)q^(x)

d(x).

= qi(x)d^(x)

1,

degree zero. Since both d(x)

some polynomials q(x) and q^(x). from which we see that q(x) and q^x) are of and d^(x) are monic, we have q(x) = 1, d(x) =
for
called the greatest
d(x).
is irreducible,

Definition 9.1

f(x) andg(x).
Definition 9.2

We
is

The polynomial d(x) write (f(x\g(x))

is

common

divisor

of

A polynomialf(x\

not identically zero,

over

if there

no factoring, f(x)

= g(x)h(x)

off(x)

into

or prime, two polynomials

g(x) andh(x) of positive degrees over Q.

For example
(x

x*

is

irreducible
field

A/2) (a;

-s/2)

over the

over Q. It has the factoring of real numbers, but it has no factoring

over R.
If an irreducible polynomial p(x) divides a product f(x)g(x), then p(x) divides at least one of the polynomials (x) and g(x).

Theorem 9.4

or g(x) == the result is obvious. If neither is identically Proof. lff(x) zero, let us assume that/?(x) </(#) and prove that p(x) g(x). The assumption that p(x) ^f(x) implies that (p(x) f(x)) 1, and hence by Theorem 9.3 there exist polynomials F(x) and G(x) such that 1 p(x)F(x) +f(x)G(x).
9

Multiplying by g(x)

we
g(x)

get

= p(x)g(x)F(x)
I

+f(x)g(x)G(x).

Now

p(x)

is
(x)>

a divisor of the right and hence P(x) <X>-

member of

this

equation because

Theorem
into

9.5 Any polynomial f(x) over Q of positive degree can be factored a product f(x) = cp^(x)p 2 (x) pk (x) where the PJ(X) are irreducible monic polynomials over Q. This factoring is unique apart from order.

9.1

Polynomials

191

Clearly/(X) can be factored repeatedly until it becomes a product of Proof. irreducible polynomials, and the constant c can be adjusted to make all the

Let us consider another factoring, into irreducible monic polynomials. According q^x), cq-L (x)q 2 (x) f(x) to Theorem 9.4, pi(x) divides some q z (x), and we can reorder the q m (x) to

factors monic.

We must prove uniqueness.

make /?!(#)
p(x)

= q\(x). A repetition

q(x). Since/?! (x) and q^x) are irreducible and monic, of this argument yields

we have

Definition 9.3
is

A polynomialf(x) =

aQ x n

a n with integral coefficients

.

said to be primitive if the greatest common divisor of its coefficients is 1 a^ Obviously, here we mean the greatest common divisor of integers as defined in
Definition 1.2.

Theorem
Proof.

9.6

The product of two primitive polynomials

n

is

primitive.

Let a Q x
is
.

a n and b x m
c Q x n+m

+ +
is

b m be primitive polynomials
.

Suppose that this product a prime p which divides every polynomial n coefficient ck Since a Q x + + a n is primitive, at least one of its coefficients is not divisible by/?. Let a i denote the first such coefficient and let b,- denote the first coefficient of b^x + + b m not divisible by p. Then the coefficient
their

and denote

product by not primitive, so that there

c n+m

of a-+w-M in the product polynomial

(9.2)
c, 4

is

m. In this sum, summed over all k such that ^ k ^ n, ^ i +j k any term with k < z is a multiple of p. Any term with k > / that appears in k < j and will also be a the sum will have the factor b i+ ^k with z + j = /, appears in the sum, and we have multiple of p. The term ajb^ for k c i+j = apj (mod/?). But this is in contradiction with p c i+ji p f ai9 p -f bj.
j
\

Theorem
f(x)

9.7

Gauss'

Lemma. If a monic polynomial f(x)

with

integral

coefficients factors into

two monic polynomials with rational then g(x) and h(x) have integral coefficients. g(x)h(x),

coefficients,

say

Proof.

Let c be the least positive integer such that cg(x) has integral 1. Then cg(x) is a coefficients; if g(x) has integral coefficients take c its is of a divisor if coefficients, then /? c p primitive polynomial, because and because c is the leading coefficient, (clp)g(x) would have integral
[

minimal property of c. Similarly let c x be least that cji(x) has integral coefficients, and hence ch(x) positive integer such 9.6 the product {cg(x)}{cji(x)} Theorem is also primitive. Then by ccjf(x) 1 cc : has since is integral coefficients, it follows that /(x) primitive. But 1. cl and c
coefficients contrary to the

Algebraic

Numbers

PROBLEMS
1.

If/00 g(x ) and -00 f(x), prove

| |

that there

is

a rational number c such that

g(x)
2.

cf(x).
|
| |

that /(a?) h(x). If/00 g(x) and ^00 AOO, prove either g(x) 3. If /?00 is irreducible and ^00 p(x) prove that c. number = rational c/?00 for some g(x)
9
|

is

a constant or

0. for any rational c 4. If /?00 is irreducible, prove that cp(x) is irreducible a into factors product g(x)h(x) 5. If a polynomial/00 with integral coefficients

of two polynomials with coefficients in Q, prove that there with integral coefficients. 1 OO^i 00
6.

is

a factoring

If/ 00

and^OO

are primitive polynomials,

and if/00 #00 and #00 \f(x),

|

prove that /(a;)

#00-

9.2

Algebraic

Numbers

an algebraic number if it satisfies complex number is called = where some polynomial equation f(x) f(x) is a polynomial over Q.
Definition 9.4

Every rational number r in this case. as x

is

an algebraic number because/00 can be taken

irreducible monic satisfies a unique over Q. Furthermore every polynomial equation

Theorem

9.8

An

algebraic
==

number f

polynomial equation g(x) over Q satisfied by f is divisible by g(x).

From all polynomial equations Q by Proof. 0. If the leading coefficient of G(x) is c, define lowest degree, say G(x) and g(x) is monic. The polynomial g(x) is c~ 1 G(x), so that g() s (x) and h 1 (x)h 2 (x) 9 then one at least of A x (f) irreducible, for if g(x) are and that to the fact would hold, contrary G(x) 7z 2 (f) g(x)
over
satisfied
f,

choose one of

= =

least degree satisfied by f polynomial equations over Q of be any polynomial equation over Q having f as a root. Next let/(a?) = gOO?00 r <X>- The remainder r(x) Applying Theorem 9.1 we get/00 the degree of r(x) would be less than otherwise for must be identically zero, = 0. Hence of a root be would that of g(x), and f g(f) r(x) since /(f) of a divisor is /(re). g(o:)
.

that g(x) is an irreducible Finally to prove that g(x) is unique, suppose the argument Then 0. that such monic polynomial g(a?) ^(a?) by ^(f) of the But irreducibility g^x) then implies above, say g^x) g(x)q(x). monic. Thus and 1 since that q(x) is a constant, in fact q(x) ^(cc) are g^x)

we have ^00

= g(x).

9.2

Algebraic

Numbers

193

Definition 9.5

g(x)

The minimal equation of an algebraic number f is the equation described in Theorem 9.8. The minimal polynomial of f is g(x). The
is

degree of an algebraic number

the degree
is

of its minimal polynomial.

an algebraic integer
...
if
it

Definition 9.6 An algebraic number f some monic polynomial equation

(9.3)

satisfies

f(x)

x-

+ b^-i +

bn

=Q

with integral coefficients.

Theorem 9.9

Among

the rational numbers, the only ones that are algebraic

1,

integers are the integers 0,

2,

Proof.

Any

m.

On

is an integer algebraic integer because /(#) can be taken as the other hand, if any rational number mfq is an algebraic

integer, then

we may suppose (m,q)

n

=
1

1,

and we have
bn

-}

+b feT +--- +
1

= =

o,

Thus q m so The word "integer"

n
,
|

mn + that q =

brfm"-

b nq

0.

1 , and /w/y is an integer. in Definition 9.6 is thus simply a generalization of

our

are often 1,2, previous usage. In algebraic number theory, 0, referred to as "rational integers" to distinguish them from the other algebraic
integers, that are not rational.

For example, \/2

is

an algebraic integer but

monic with

not a rational integer.

Theorem 9.10

The minimal equation of an algebraic integer

is

integral coefficients.

Proof.

The equation

is

monic by

definition, so

we need prove

and let its minimal equation be g(x) = 0, monic and irreducible over Q. By Theorem 9.8, g(x) is a divisor of /(a?), say /(a) = g(x)h(x), and the quotient h(x) like/(#) andg(a;), is monic and has coefficients in Q. Applying Theorem
9

coefficients are integers. Let the algebraic integer f satisfy/ (a;)

only that the

as in (9.3),

9.7,

we

see that g(x) has integral coefficients.

Theorem 9.11
equations
(9.4)
0,

Suppose that the

Let n be a positive rational integer and | a complex number. W not all zero, satisfy the complex numbers B l9 2
' '

"

>

= a^A +

ai>2 6 2

a,. n

n,

1, 2,

n,

is an algebraic number. where the n* coefficients ajti are rational. Then an is are rational a the f Moreover, if algebraic integer. integers, iti

194
Proof.

Algebraic

Numbers

Equations (9.4) can be thought of as a system of homogeneous 6 n Since the d i are not all zero, the de 2 terminant of coefficients must vanish
linear equations in O l9
, ,
.
:

0,

n Expansion of this determinant gives an equation f + b^ n ^ + + bn = where the 6 4 are polynomials in the ^ Thus the b i are rational, and they
.

>fc

are rational integers if the

a^ are.

Theorem 9.12 If a and $ are algebraic numbers, so are a a a</ /? are algebraic integers, so are a + /? aw d a/?.
.

ft

ad

a/?.

Suppose that

oc

and

/?

satisfy

with rational coefficients

numbers

l9

ra

as the

and Z> numbers

-. ;

Let

mr, and define the complex

oc

2
,

in

any order. Thus and r = 0, 1,

-

O l9
,

1.

6 n are the numbers Hence for any 0,,

a'jS*

with $

0,1,---,AH-

In either case

we

a^

Aj.

H---that

see that there are rational constants h ftl9

+
'

A J>n such that

kjtn
C^-.i

such

+ is an algebraic integer. We also have oc/W, = oc^A + + ^ n J = ^.^ + -+*, n a0 n from which we find a00, = + ^ w e n where c = ^- A,. + c,,^! + H---- + kg.nhn.i- Again we Theorem 9.11 to conclude that a/9 apply ^.A,<
a
/?
. .
.

^,i) fl i (9.4), and so ft are algebraic integers, then the ai9 b i9 A, ti

.

= fc^A + + ^ n n and hence (a + 0)0, = + + (Aj. w + fcj.)0 B These equations are of the form we conclude that a + is algebraic. Furthermore, if and
-

A ; >n
-

w . Similarly there are rational

fl
,

constants

kj

15

)5fl,

'

oc

ft

fc

j< are

all

rational integers,

and

fl

iti

is

algebraic,

and that

it is

an algebraic integer

if

a and

are.

9.3

Algebraic

Number

Fields

^95

This theorem states that the set of algebraic numbers is closed under addition and multiplication, and likewise for the set of algebraic integers. The following result states a little more.

Theorem 9.13 The set of all algebraic numbers forms afield. The class of all algebraic integers forms a ring.
Proof.

are defined in Definition 2. 12. The rational numbers and unit for the system. Most of the postulates are seen to be satisfied if we remember that algebraic numbers are easily complex numbers whose properties we are familiar with. The only place where
fields

Rings and

and

serve as the zero

any

difficulty arises is in

inverses. If

^
1

is

proving the existence of additive and multiplicative a solution of

aQ x n

+ ax n -i +
1

an

then

a and or are solutions of

aQ x n

aQ

^a;"-

+
+

a 2 x n ~*

+ (-l)X =

and

a^x

atfc*

a nx n

0,

respectively. Therefore if a is an algebraic number, then so are a and a" 1 . 1 If a is an algebraic then so is but not a, integer, necessarily or Therefore
.

the algebraic

numbers form a

field,

the algebraic integers a ring.

PROBLEMS
Find the minimal polynomial of each of the following algebraic numbers: + (1 + tyT)/2, 1 + Which of these are algebraic integers? VJ VJ. 1 2. Prove that if a is of athen and a - 1 are also of n, algebraic -a, degree in the case of a"1 degree n, assuming a ^ 3. Prove that if a is of algebraic of degree n, and ft is m, then
1.

7,

^7,

a
4.

algebraic
a/?.

degree

ft

is

of degree

mn. Prove a similar result for

Prove that the set of all real algebraic numbers (i.e., algebraic numbers that are real) forms a field, and the set of all real algebraic integers forms a ring.

93
The field

Algebraic

Number

Fields

discussed in Theorem 9. 1 3 contains the totality of algebraic numbers. In general, an algebraic number field is any subset of this total collection that is a field itself. For number, then it can be example, if is an
algebraic
readily
7* 0,

numbers of the form/(f)/A(f), A(f) h over constitutes a field. This field is denoted by g, /and polynomials and it is called the extension of <2 by
verified that the collection of all
.

196

Algebraic

Numbers
in

Theorem 9.14

If

is

an algebraic number of degree

in the form

n,

then every

number

Q()
(9.5)

can be written uniquely

a,

*if

'

'

a n _^ n

~l

where the a i are rational numbers.

Proof.

Consider any number /(f)//z(f) of >(?) If the minimal polynomial But #(&) is irreducible, so the (s), then g(x) f h(x) since A(f) 7^ 0. and of divisor common h(x) is 1 and so by Theorem g(x) polynomial greatest 9.3 there exist polynomials G(x) and H(x) such that 1 g(x)G(x) + h(x)H(x). = #(f) and = we that x the fact and 0, get l//z() g(g) using Replacing by f
of f
is
,

/(f)/A(f)

fc(f). k(x)=f(x)H(x) so that /()/A(f) = = hence and + r(z), /(f)/A(f) (%(#) get A:(z) Dividing fcfc) by g(z), = the form r of is where (9.5). r(f) &() (f) To prove that the form (9.5) is unique, suppose r(f) and r^f) are ex

=/()#)

Let

we

form (9.5). lfr(x) r^x) is not identically zero, then it is a the minimal polynomial of f has n. Since than less of degree polynomial have we 7^ /^(f), unless r(x) and ^(a) are 5^ 0, /z, r(f) ^(f) r(f) degree
pressions of the

the same polynomial.

The

field <2(f)

congruences modulo the polynomial

2.1, for

can be looked at in a different way, by consideration of g(x). That is, in analogy with Definition
least

any polynomial G(x) of degree at

one we

will write

/!(*)= /2 (oO (mod G(z))

in order to get back to Q(f) we will take 2 (aO). Ultimately, G(x) (/i(z) the minimal polynomial g(o;) of f for G(x). However, the theory of con gruences is more general, and we start with the polynomial G(x) over Q,
if
|

irreducible or not.

The properties of congruences in Theorem 2.1 can be extended at once to the polynomial case. For example, part (c) of the theorem = h 2 (x) (mod G(:r)), has the analogue: If/! (a?) 2 (z) (mod (/(#)) and h(x)

=/

then/iCr)/*^)

(mod G(x)). By the division algorithm Theorem 9.1, any polynomial/^) over Q mapped by division by G(x) onto a unique polynomial r(x) modulo G(x);
z

=f (x)h
=

z (x)

is

f(x)

G(x)q(x)

r(x\

f(x)

r(x)

(mod

G(*)).

Thus the

of polynomials r(x) consisting of and all polynomials over Q than n constitute a "complete residue system modulo (/(#)" in the sense of Definition 2.2. Of course the present residue system has
set

of degree

less

infinitely

many members, whereas

the residue system

modulo

contained

precisely

elements.

Theorem 9.15
of polynomials
(9.6)

Let G(x) be a polynomial over

of degree n

1.

The

totality

r(x)

<*&

'

'

9.3

Algebraic

Number

Fields

197

with coefficients in Q,

and with addition and multiplication modulo G(x), forms

ring.

Proof.
its

This theorem is the analogue of the first part of Theorem 2.33, and proof is virtually the same. First we note that the polynomials (9.6) form a group under addition, with identity element 0, the additive inverse of r(x) r(x). Next, the polynomials (9.6) are closed under multiplication being modulo G(x), and the associative property of multiplication comes from the

corresponding property for polynomials over

that
is

Q with ordinary multiplication,

{r 1 (x')r 2 (x)}r 3 (x')

r^(x){r2 (x)r z (x)}

implies
}

(mod
is

G(xJ).

modulo Similarly, distributive property of polynomials over Q.

the distributive property

G(x)

inherited

from the

we extend Definition 2.10 to the concept between Two fields F and F are isomorphic if there fields. of isomorphism the elements of F and the elements between is a one-to-one correspondence r if a and in F that b such of F' correspond respectively to a' and b in F' then a + b and ab in F correspond respectively to a + b' and db' in F'. A virtually identical definition is used for the concept of isomorphism between a direct analogue of the second part of Theorem rings. The following result is
Before stating the next theorem,
r 9

2.33.

The ring of polynomials modulo G(x) described in Theorem and only if G(x) is an irreducible polynomial. If G(x) is the afield minimal polynomial of the algebraic number f then this field is isomorphic

Theorem 9.16
is

9.15

if

to

Proof.

If the polynomial G(x) is reducible over g, say G(x) = G^ where G^(x) and G 2 (x) have degrees between 1 and 721, then G : (x) and G z (x) are of the form (9.6). But then d(a?) has no multiplicative inverse

modulo G(x)

since

G l (x)f(x) =

(mod G(z))
GI(X)
|

implies

G(x)

{G(x)f(x)

1},

{G^(x)f(x)
is

1},

G^(x)

1.
\

Hence the ring of polynomials modulo G(x)

not a

field.

On

the other hand, if G(x)

is

r(x) of the

form
is

(9.6)

has a unique
this

irreducible over g, then every polynomial modulo G(x), r multiplicative inverse : (x)

of the form

(9.6).

To show
and so

G(x) and r(x)

h(x) such that
(9.7)

1,

we note that the greatest common divisor of and by Theorem 9.3 there exist polynomials /(a?)

r(x)f(x)

G(x)h(x).

198

Algebraic

Numbers

= G(x)q(x) Applying Theorem 9.1 to f(x) and G(x) we get f(x) where r^x) is of the form (9.6). Thus (9.7) can be written
1

r^x)

r(x)ri(x)
is

G(x){h(x)

r(x)q(x)},

r(x)r^x)

(mod G(x)\
This inverse
is

so r : (x)

unique because
r(x) ri (x)

a multiplicative inverse of r(x) of the form 1 if r(x)r 2 (x) (mod G(x)) then

(9.6).

r(a?)r a (aO

(mod G(x)\ G(x)

\

r(x){r^(x)

r 2 (z)}.

Since G(x)

<r

r(x)

we have G(x)

{r t (x)

r^(x)}

by Theorem
is

9.4.
less

But the

than n, polynomial r^x) r z (x) the degree of G(x). Hence r^x) 0, r : (x) = r 2 (#). Finally, if G(oO is the minimal polynomial g(x) of the algebraic number f we must show that the field is isomorphic to Q(f). To each r(rc) of the form (9.6) we let correspond the number r(f) of Q(). Theorem 9.14 shows that
r 2 (x) is either identically

zero or

of degree

this

correspondence

is

one-to-one. If
rs(x),

r^)r^(x)
then

r^(x)

r z (x)

r 4 (a)

(mod G())

and hence

since G(f)

0.

Therefore the correspondence preserves multiplication and

addition.

The theorem we have just proved is significant in that it makes possible the development of the theory of algebraic numbers from the consideration of polynomials without any reference to the roots of the polynomials. The fundamental theorem of algebra states that every polynomial of positive
degree over

has a root that

is

a complex number. Therefore the algebraic

obtained by means of Theorem 9.16 are essentially the same to the fields R(g) of Theorem 9.14, but one does not need a isomorphic the of fundamental theorem of algebra to use the method of knowledge

number

fields

Theorem 9. 16. The fundamental theorem of algebra implies, and is sometimes stated in the form, that every polynomial/^) of degree n over Q has n complex roots.
If f(x)

irreducible over Q, then the n roots, say f l5 f n are called and the of one of them are numbers, conjugate algebraic conjugates any the all others. 9.16 Theorem does not make simply any distinction
is
, ,

Now

between conjugates, whereas Theorem 9.14 allows for such a distinction. For example, lttg(x) be the irreducible polynomial x 3 2. In Theorem 9.14 we can take f to be any one of the three algebraic numbers which are solutions
of a 8

- 2 = 0,

namely ^2, 0)^2, coH/2 where

o>

(-1

z\/3)/2.

Thus

9.4

Algebraic Integers

199

there are three fields

(9.8)

Rffi),
first

(0)^2),

R(c

The

of these consists of real numbers, whereas the other two contain

nonreal elements. Therefore, the first is certainly a different field from the others. It is not so apparent, but can be proved, that the last two differ from each other. On the other hand, if we apply Theorem 9.16 to the polynomial x3 ax a zx 2 2, we obtain a single field consisting of all polynomials a Q

over

Q modulo

x*

2.

According to Theorem 9.16,

this field is

isomorphic

to each of the fields (9.8). Since isomorphism is a transitive property, the fields (9.8) are isomorphic to each other. They differ in that they contain
different elements, but they are essentially the

same except

for the

names of

their elements.

PROBLEMS
1.

Prove that the

fields

of
is

(9.8),

although isomorphic, are

distinct. Suggestion:
a)
2

to

2 prove that Q(co ^2)

different

element of the
that
2.
co

latter field, that is,

from assume
2
.

Q(wv 2),

assume that

^2

is

an

that there are rationals a, b, c such

^2 =

Prove that the

polynomials a + 3. Prove that any algebraic number field contains Q as a subfield. 4. Assuming the fundamental theorem of algebra, prove Theorem 9.10 by the following procedure. Let the algebraic integer satisfy some monic polynomial

Prove that no such rationals exist. where / 2 = 1, is isomorphic to the field of bx with a and b in Q, taken modulo x 2 + 1.
c(o>^2)
field Q(z),

bco^2

all

= with integral coefficients. Then equation f(x) field of complex numbers, say
f(x)
Ifg(x)
is

we can

factor f(x) in the

(x

fi)(a>

- &(x ,

'

'

ft)

(*

- W9.8,

the minimal polynomial of

theng(x) f(x) by Theorem

|

and so

g(x)

= (x- &(x
-

62)

(x

r)

where

2,

6r

are a subset of

2,

fn

Thus

2,

6r are algebraic

integers,

and by Theorem 9.12 the

9.9.

coefficients of g(x) are algebraic integers.

Then apply Theorem

9.4

Algebraic Integers

and 1, and so, by the algebraic number field contains the elements numbers. Thus any rational all the contain must a for field, postulates field contains at least some algebraic integers, the rational number algebraic

Any

200
integers 0,
1,

Algebraic
2,
.

Numbers
an

The following

result

shows

that, in general,

algebraic

number

field also

contains other algebraic integers.

is

Theorem 9.17
that bo.
is

If a. is any algebraic number, there an algebraic integer.

a rational integer b such

0. Let/(#) be a polynomial over Q such that/(a) Proof. may presume that the coefficients of/(V) are rational integers, since we can multiply by the least common multiple of the denominators of the coefficients. Thus we can

We

take/(#)

in the

form

f(x)

bx n

n -1

a^x

+
is

an

bx n

with rational integers b and

ar

Then b&

a zero of

and hence b& Theorem 9.18

is

an algebraic integer.

The

integers

of any algebraic number field form a

ring.

Proof. If a and /? are integers in such a field F, then since F is a field. But by Theorems 9.12 and 9.13, a
algebraic integers. Thus the integers of identity elements of addition and multiplication.
Definition 9,7

and a/9 are in F and a are form a ring with and 1 as the

+ +

/?

/?,

a/?,

divisor

of an integer
j

In any algebraic number field an integer oc there an exists /? if integer y such that
j8. Any of the integer are called associates if a//? is

F
1

is

said to be

case we write a
integers a

divisor

is

$ ay. In this called a unit of F. Nonzero

and ft

unit.

but

This definition of associates does not appear to be symmetrical in a and we shall establish that the property really is symmetric.
is

/?,

Theorem 9.19 The reciprocal of a unit number field form a multiplicative group.
Proof.
If
a

unit.

The

units

of an algebraic

Hence

e2

is

a unit, then there exists an integer s 2 such that also a unit, and it is the reciprocal of e. If, similarly,
is 4,

s^ =
3 is

1.

an algebraic number field form a multiplicative group where the identity element is 1, and the inverse of e is the reciprocal of e. If a and /? are associates, then a//? is a unit by definition, and by the above theorem /?/a is also a unit. Hence the definition of associates is symmetric if a and /? are associates, then so are and a. /?
:

unit with reciprocal Hence the units of

then the product

s^ is a unit because (fiifisK^^) =

any
1-

9.5

Quadratic Fields

201

PROBLEMS
1.

Prove that the units of the rational number field

ft

2 are
p.

and that

integers

a and
2.

are associates in this field

if

and only

if

a ==

For any algebraic number a, define m as the smallest positive rational integer such that met. is an algebraic integer. Prove that if &a is an algebraic integer, where b is a rational integer, then m b. 3. Let a *= 4- a 2 / be an algebraic number, where a t and a 2 are real. Does it follow that o^ and a 2 are algebraic numbers ? If a is an algebraic integer, would a x and a 2 necessarily be algebraic integers?
\

OC-L

9.5

Quadratic Fields
one of the form Q(f) where
is

A quadratic field is
numbers. Since f

a root of an irreducible

quadratic polynomial over Q. By Theorem 9.14 the elements of such a field are the totality of numbers of the form a a^, where a and a I are rational

is

of the form (a

b\Jm)lc where

a, b, c,

are integers,

we

see that

= Q (a +
other hand,

bjm)

Q(bjri)

Q(V^)-

Here we have presumed that c if m and n are two

is 1,

1. On the Is square-free, and that different square-free rational integers, neither

it is

m^

of which

then

Q(^m)

impossible to find rational

<2(V) since \/m is not in Q(\fn}. That is, numbers a and b such that ym == a + b\fn.
-^

is a Every quadratic field is of the form 2(\/m) where not to 1. Numbers but or rational equal negative positive square-free integer,

Theorem 9.20
the

of

form a
(a

b\Jm with

rational integers

a and b are

integers

of <2(vw).
1

These are the only integers ofQ(^/m) ifm

the

2 or 3 (mod

4).

Ifm

(mod 4),

numbers

b^/m)j2, with

odd rational

integers a andb, are also integers

o/2(Vm), and
Proof.

there are no further integers.

first

We
is

have already proved the

part of the theorem.

AH
is

that

remains

to identify the algebraic integers.

(a

Any number in Q(Jm)

of the

form is no

oc

Z?Vra)/c

where

a, b, c are rational integers with c

>

0.

There

loss in generality in

terms. If b
integer if

and

1 so that a is in its lowest assuming that (a, b, c) Theorem rational a is then 9.9, is an algebraic and, 0, by 1. If b 7* 0, then a if it is a rational integer, that is c only

202
is

Algebraic
its

Numbers

not rational, and

a

minimal equation
I

is

quadratic,
2

bm

a-

bjm\

2a

b*m

According to Theorem 9.10, a will then be an algebraic integer if and only if this equation is monic with integral coefficients. Thus a is an algebraic integer if and only if
(9.9)

c
|

la
b

and

c2
\

(a

b 2 m),

and
(a,
2
1

this includes the case

0, since (a, b, c)

=
\

1.

If (a, c)

>

and c 2a,
\

then a and c have some

\

common prime
,
\

m is square-free. Therefore (9.9) p If c 2a and c > 2 then (a, c) > 1, so that (9.9) can hold only if (a, c) = or c = 2. It is obvious that (9.9) holds for c = 1. For can hold only if c = c = 2 condition (9.9) becomes a s b m (mod 4) and we also have a odd since (a, c) = 1. Then (9.9) becomes b m = a = (mod 4), which requires that b be odd, and then reduces to m = b m = (mod 4). To sum up: (9.9) is satisfied if and only if either c = or c = 2, a odd, & odd, m = (mod 4),
1

2 2 2 2 b,c)=l. Then p a and p c and 2 2 b m, p m, which is impossible since

factor, say p, and 2 2 if c b 2 m), we (a

-f

b since

would have

1.

and

this

completes the proof.

Definition 9.8
the product

The norm
ex.

of

and its

of a number a = (a + b-^/m)/c = (a b\fm)/c, conjugate, a

N(VL)

in

Q(\m)

is

JV(a)

aa

bJrn a
^
c

bJm ^ =
c

fc

m
.

Note
its

Theorem 9.20 the number a is an integer in Q(\fm) if and only if = a. conjugate a is an integer, and that if a is a rational number then a
that by

Theorem 9.21
#(<*/?)

=
is

N(<z)N(ff).

The norm of a product equals the product of the norms, N(a) = if and only z/a = 0. The norm of an integer in
is

2(Vm)

a rational integer. If y

an integer

in Q(*Jrri), then

N(y)

if

and only
Proof.

if

is

unit.

= a/9. Then we in Q(\[m) is easy to verify that (a#) = aa/?/? = N(*)N(P). If a = 0, then a = and A (a) = 0. Conversely if N(a) = 0, then aa = so that a == or a = 0; but a = = 0. implies a
For a and
have
JV(aj8)

/?

it

a^a/5

it

Next, if y is an algebraic integer in <2(Vw), it has degree either 1 or 2. If has degree 1 , then y is a rational integer by Theorem 9.9, and N(y) yy 2 y so that N(y) is a rational integer. If y is of degree 2, then the minimal

2 equation of y, x

and again N(y)

= yy

(y
is

+ y)# + yy 0, has rational integer a rational integer.

coefficients,

9.6

Units

in

Quadratic Fields
1

203
, ,

To prove

an integer, then y y = 1 y 1 so that y is a unit. be a unit. Then there is an integer s such that y 1 since JV(y) 1. This implies N(y)N(s) = N(l) = 1, so that W(y) = ye and N(e) are rational integers.
If 7V(y)

and y

is

the converse,

let

Remark.

The

integers of 2(z) are often called Gaussian integers.

PROBLEMS
1. If

an integer a in Q(\]ni)
7?2

is

neither zero nor a unit, prove that \N(cc)\

>

1.

2. If

(mod

4),

prove that the integers of

Q(-Jm) are

all

numbers of

the form

+bwhere a and b are rational

3. If a is
4. If

integers.
\

any integer, and s any unit, in Q(^/m), prove that e a. a and ft 7* are integers in 2(V/w), and if a 0, prove that a
|

J5

and

JV(a)

N(#).

_
number
if

5. If a is

Show
6.

an algebraic that this is false

in

Q(\jm) with
is false

m < 0,

prove that N(<x)

is

0.

m >

0.

Prove that the following assertion a is an algebraic integer. integer, then

7.
field.

in fi(/): If 7V(a)

a rational

is false in every quadratic = that 7V(a) is certainly an a so Define (x 2^m)/y, Suggestion: 2 m 1 x and 2/ satisfy x z y = 4/rz. Choose # = m + l, y integer if 4 be can cases 4. The a if > not an is that ^ 1| so \m 1] \m integer

Prove that the assertion of the preceding problem

treated specially.

9.6

Units
field

in

Quadratic Fields
is

m>

quadratic
I.

Q(*Jm)

called imaginary if

m<

0,

and

it is

called real if

fields.

We

There are striking differences between these two sorts of quadratic shall see that an imaginary quadratic field has only a finite number

1 are the of units; in fact for most of these fields only units. units. has field infinitely many hand, every real quadratic

On

the other

Theorem 9.22

Let
1

m
,

be a negative square-free rational

integer.

The field

Q(\fm) has units and m = -3. The

1, (1

and these are

the only units except in the cases

1

m=

units for Q(i) are

and

L The

=3) are units for Q(\T

V^/2,

and (-1

V^

Algebraic

Numbers

Theorem 9.21 we look for all integers a in Proof. Taking note of we can write a in one of 1. such that tf(a) According to Theorem 9.20 the two forms x yjm and (x W"0/2 where x and y wt rational 1 (mod 4). in the second form, x and y are odd and integers and where, 2 2 is Since ^2/ )/4 respectively. (* Then JV(oc) my* or JV(a) -1. For 2 with 2 a no are .tf(a) so there re my negative we have 2 2 and the only solutions of 2?/ a*
,

m=

=&-

m < _i
X2 X
2

we have
! i

o,

my*
x

-/m/

__
__

m2/

m y*

= =

are y

in this case.

has t he solutions x

others For
1

- m</ 1 with odd x and y since x* we see that the solutions of the equation (x*
2

(mod4),m

< -3

For m = -1, the equation = 1, y = and no 1, and & 0, y - mf)\*t = there are no solutions of (x*

- m > 4. Finally, for m = -3, + 32/ )/4 = with odd x and y

2
1

are just ys the units described in the theorem.

s=l,

=l,an<U=-l,y=l.
infinitely

These solutions give exactly

Theorem 9.23
Proof.

There are

many

units in

any real quadratic field.

= x + y^m with integers x, y are integers in - my*. If z - wy = 1, then a is a unit. But 2(V^) with norms 7V(oc) = x* in Theorems 7.25 and 7.26 >m/ = 1, m > 1, was treated the equation x*
The numbers a
2
2 8

where

it

was proved that

it

has infinitely

many

solutions.

PROBLEM
1.

Prove that the units of GG/2) are

(1

+ J2) n

where n ranges over

all

integers.

9.7

Primes

in

Quadratic Fields
a
unit, in

Definition 9.9
is

An

not algebraic integer a,

is divisible

called a prime if it
is

only by

its

associates

and

a quadratic field Q^Jrri) the units of the field.

almost the same as the definition of primes among the This definition In Q all primes are rational integers. There is this difference, however. IT is a prime in Q(-Jm) no such property is required. Thus if positive, whereas For 7r is an associated prime in 2(Vm). then in unit a e is and 2(V"0>
example,
77-

is

an associated prime of 77.

If the norm of an integer
is
oc

Theorem 9.24

in

Q(Jm)

is

p, where

is

rational prime, then a

a prime.

Proof.

= $y where ft and y are integers in Q(Jm). By Suppose that a Theorem 9.21 we have tf(a) = N($)N(y) = p. Then since N(ft and N(y)

9.7

Primes

in

Quadratic Fields
,

205
ft

are rational integers, one of them must be 1 so that either and the other an associate of oc. Thus a is a prime.

or y

is

a unit

Theorem 9.25 Every integer into a product of primes.

in Q(\/tri),

not zero or a unit, can be factored

Proof. If a is not a prime, it can be factored into a product /ty where neither ft nor y is a unit. Repeating the procedure, we factor ft and y if they

are not primes. The process of factoring must stop since otherwise we could f) n with n arbitrarily large, and no factor fa a unit. get a in the form ft^ 2 that would this But imply
g:

n
,

n arbitrary,

;=1

3=1

since |JV(^)|

is

an integer

>L
into primes, this Section 1.3 that

Although we have established that there is factorization factorization may not be unique. In fact, we showed in
factorization in the field

g(V 6) is not unique. In the next section we prove unique in the field Q({). The general question of the values of m for which Q(*Jm) has the unique factorization property is an unsolved problem. There is, however, a close connection between unique factorization and the Euclidean algorithm, as we now show.
that factorization
is

Just as in the case of the rational field, a unique factorization theorem will have to disregard the order in which the various prime factors appear. But now a new ambiguity arises due to the existence of associated primes. The two factorings
OC

==

"

TT'i'77'2

77"*.

==
1

'

(.^l^"ly

\^2^2J

y^r^f)

where the
same.

,-

are units with product

will

have to be considered as being the

Definition 9.10

A quadraticfield Q(\fni) is said to have the uniquefactorization

be factored into property if every integer a in Q(\/m), not zero or a unit, can and the the order ambiguities between primes of primes uniquely^ apart from
asso dated primes.
Definition 9.11

quadratic field

g(Vm)

is

said to be Euclidean if the integers

is,

of

QkJrri) satisfy a Euclidean algorithm, that

with
ft
|

Q(Jm)
j8y

<J,

|#(<5)

^ <

0, there exist

a and $ are integers of that a = integers y and d of Q(Jm) such

if

\N(P)\.

Theorem 9,26
property.
Proof.

Every Euclidean quadratic field has the unique factorization

similar to the procedure used in Theorem 1.16. First we

The proof of

this

theorem

is

of arithmetic, establishing the fundamental theorem

206
establish that if a

Algebraic

Numbers

and

are any two integers of Q(<Jm) having

no common

such that factors except units, then there exist integers A and ^ in Q(Jm) od form the of of jip where 1- Let S denote the set integers aAo #1*0 The norm A^od #a) of any >l and p range over all integers of Q(V/n)e /fy^ is a rational integer, so we can choose an integer, cd^ in S integer value taken on by \N(^ is the least such that

+ + +

positive \N(e)\ to a and e Applying the Euclidean algorithm say,

we

get

<5,

\N(d)\

<

\N(e)\.

Then we have
an integer in S. Now this requires \N($)\ = by the definition of = &y and hence s a. and we have 6 = by Theorem 9.21. Thus a s~ l is also a unit by Then unit. a e is therefore and we find e
<5

so that
e,

is

Similarly

ft,

Theorem

9.19,

and we have

say.

Next we prove that if TT is a prime in Q(\]rn) and if TT a/3, then TT a or TT and a have no common factors except units, and 0. For if TT f a, then = hence there exist integers A and /* such that 1 = 7rA + a/v Then /3
| |

TT

7r/3A

ocjfyio

and

TT
|

^ because
if TT
|

TT
|

oc/?.

induction to prove that of the product. ocj

(oc^
is

This can be extended by mathematical 77 divides at least one factor ), then

From
1.16,

this

point on the proof

is

identical with the first

proof of Theorem

and there

no need

to repeat the details.

PROBLEMS
1. If

2.

3.
4. 5.

that TT is a prime and e a unit in G(V/n), prove Prove that 1 + i is a prime in Q(i). Prove that 11 + 2^/~6 is a prime in g(V 6 )-

STT is

a prime.

3 is a prime in 2(0, but not a prime in Prove that there are infinitely many primes in any quadratic

Prove that

field

9.8

Unique Factorization
various_ quadratic fields, Theorem_9.26 Q(J~2), fiCV^S), Q(J~V, fi(>/2), fi(\/3). We shall show
shall apply

In this section

we

to

namely

g(i),

9.8

Unique Factorization

207

that these fields have the unique factorization property are Euclidean fields. There are other Euclidean

by proving that they

quadratic fields, but we focus our attention on these few for which the Euclidean algorithm is easily

established.

Theorem 9.27

The fields Q(^n) for

m=

-1, -2, -3, -7,

2,

3,

are

Euclidean and so have the unique factorization property.

Consider any integers a and ft of fi(Vw) with ft ^ 0. Then a//? = where u and v are rational numbers, and we choose rational integers x and y that are closest to u and v, that so that
Proof.

+ vjm

is,

(9.10)

0^|w-s|^J,
x

0^|0-y|J.
<5,

If

we_denote

y^ by y and a - $y by
=
tf a

and

tf<J

=N

then y and d are integers in

i/

a?)

(9.H)

\N(d)\--

By Equations

(9.10)

we have

rn\y

y)

__

11

^> 0,

0^(M
and hence, by
fi(V/w)
is

-o;) -m(i;~2/)
\N(d)\

^ +
4
4
if

(~ m )

if

m<

(9.11),

<

\N(ft)\

w=

2,3, -1, -2.

Therefore

Euclidean for these values of w.

3

For the cases m = With M and u defined

as possible as close to

and

772

as above,

7 we must choose y in a different way. we choose a rational integer j as close to 2v

r,

and then choose a rational integer 2u as possible. Then we have \2v

such that r

- s\ ^

(mod
r|
:

2),
1,'

J and |2w

and the number y

since
is

m=

(r

+ sjm)/!

(mod
in

4) in

an integer of Q(Jm) by Theorem the cases under discussion. As before, d == a

is

9.20,
fly

an integer

Q(\Jm) and

JV(<3)

= +
(-m)

<

for TW

and

m=

7.

208

Algebraic

Numbers

PROBLEMS
Q(V~ H) has the unique factorization property. Prove that Q(^/5) has the unique factorization property. 3. Prove that in g(0 the quotient y and remainder 6 obtained in the proof of Theorem 9.27 are not necessarily unique. That is, prove that in Q(i) there
1.

Prove that

2.

exist integers a, p, y, 6, y 1? d l

such that

4. If a

and

/?

are integers of Q(i), not both zero, say that y

ft

is

a greatest

com

mon divisor
of a and
]3.

of a and

if

N(y)

is

greatest

among norms of all common

divisors

fixed pair a,

Prove that there are exactly four greatest common divisors of any ft, and that each of the four is divisible by any common divisor.
in

9.9

Primes

Quadratic Fields Having the Unique

Factorization Property
If a field

more about

Q(\Jm) has the unique factorization property, we can say much the primes of the field than we did in Section 9.7.
Let Q(\Jni) have the unique factorization property. Then to Q(^l m) there corresponds one and only one rational prime p

Theorem 9.28
any prime
rr in

such that TripProof. The prime rr is a divisor of the rational integer Nfa), and hence there exist positive rational integers divisible by TT. Let n be the least of these. Then n is a rational prime. For otherwise n n^n Z9 and we have, by the

unique factorization property,

since
TT

TT
\

n,

TT
|

2) 5 TT
|

or

TT
|

2,

a contradiction

by Theorem 1.3 such that 1 = px + qy. Since rr (px which is false, and hence the prime p is unique.
|

a rational prime, call it p. And, if were a divisor of another rational prime q, we could find rational integers
n
77,

<

<

<

n2

<

n.

Hence n

is

+ qy) this implies rr

1
|

Theorem 9.29
(1)

Let Q(\/m) have the unique factorization property. Then:

is

Any

rational prime p
,

either a

prime

rr

of the field or a product

1

rr^rr^

of two primes not necessarily (2) The totality of primes rr,

distinct.,
TT I? rr 2

of Q(\l m).
to all

obtained by applying part

rational primes, together with their associates, constitute the set

of

all

primes ofQ(\]m).

9.9

Primes
(3)

in

Quadratic Fields Having the Unique Factorization Property

rational
in

209

An odd

prime p satisfying
if and only if
I

(p,
I

m)
1.

is

a product

TT^TT^

of

two primes
the product
are,

Q(\m)

Furthermore ifp

= ^773,

and of two primes, then TT^ and 7r 2 are not associates, but and 7r2 and T^ are. = 1> then 2 is the associate of a square of a prime if m == ( 4 ) tf ( 2 #0 3 (mod 4); 2 w a prime ifm=z5 (mod 8); <W 2 fj r/ze product of two
7? 2
>

distinct
(5)

primes

ifm~l

(mod

8).

Any
in

rational prime p that divides

m is the associate of the square

of a

prime
Proof.
for

Q(\m).
prime p is not a prime in Q(<Jm\ then p = <rrp and some integer ft of g(V/n). Then we have N(Tr)N(fi) = Since JV(TT) ^ 1 we must have either N(f}) 1 or
TT
,

(1) If the rational

some prime
z
.

N(p)=p
JV(]5)

db/?.

If JVX]5)

then

ft

is

a unit by Theorem 9.21, and

TT is

an

associate of/?, which then must be a prime in Q(\fm). If #05) then /? /? is a Theorem and so is a of two 9.24, by TT/? p product prime primes in

(2)

The statement
(1).
is

(2)

now

follows directly from

Theorem

9.28

and

statement
(3) If

an odd rational prime such that

(p,

m)

and (~) \pj

1,

there exists

a rational integer x satisfying

==

x2

(mod/?), p\(x*
it

m),

(a?

yfm)(x

Jm). x

Ifp were a prime of Q(\/m), x + Vra, so that one of

x

would divide one of the

factors

V#* and

^m
'

P
this is impossible

would be an
hence p
is

by Theorem 9.20, and not a prime in Q(\[m). Therefore, by statement (1), p

integer in

Q(^m). But

p p

= 1, and that suppose that/? is an odd rational prime, that (p,ni) is not a prime in QG/m). Then from the proof of statement (1) we see that = TT/J, N(j3) = /?, and #(77) = /?. We can write TT = a + b\]~m where
Now
Then
2

a and b are rational integers or,

integers.
4/?,

a2
2

mb*

if m = = JV(TT) =
/?).

(2a)

m(2b)

(mod

(mod 4), halves of odd rational 2 and we have (2d) 2 m(2b) == /?, Here 2a and 2b are rational integers and
1

2^Q
neither
is

Algebraic

Numbers

and we
(2b,p}

divide the other a multiple of/?, for ifp divided either one it would 2 2 2 4mb*), p 4p. Therefore would have p* 4a\ p* 4b\ p (4a 1 (mod 2bw that such w p) 1, and there is a rational integer

2 2 (2<w) == m(2bw)

=m

fm\
(mod;?), and

we have

=1.

\-J

we prove that Furthermore, with the notation of the preceding paragraph and T? and are. From;? = np 77 and ]8 are not associates, but 77 and ]8 are, we have and #(77) = a 2 - mZ? 2 = /?
*

=
/?

--=
a

( fl

hand we note that

2

so

77

and

are associates.

On

the other

TT_ "

""

m(2&)
[

4p
$ab. and this is not an integer, and so not a unit, because^ does not divide Thus 77 and /? are not associates. then (4) If m = 3 (mod 4),

m -m=
*

2^=^ = (m + V^)(m - Jm),

Q(Jm). Hence 2
2.
is

and 2 f (m

of VOT), so 2 cannot be a prime

divisible

must have norm by a prime a; + y-Jm and this prime == that this But 2. implies my*

Therefore x*

+ yjm
and, similarly,

--

_
x*
9

_
^
'

my
x
2

-^ - x

+ yJm =

+ -my

and therefore

(a?

y\/m)(x

Hence
If

(x

y\[m)(x

+ y^Jm)~ and its inverse are integers of Q(\lrri). + W^)"1 is a unit and ^ ~~ y^ and x + y^~ are
l
?

associates.

__
( mo d 4) and

m=
J(a?

if

is

not a prime in
2.

prime

yVw) having norm

y,

2(Vz) then 2 is divisible by a This would mean that there are rational
2

integers x
(9.12)
If

and

both even or both odd such that

& - my =
=
2.

8.

xl

x and y are even, say x

myl

But, since

= 2z m ==
1

2y
4),

(mod

x2

then (9.12) would require my\ is either odd or a

9.9

Primes

in

Quadratic Fields Having the Unique Factorization Property

211

multiple of

4.

Thus
x
2

z2

2
2/

(9.12) can have solutions only with odd x and y Then

(mod

8),

and
2

(9.12) implies
1

- my =
(mod
8)

-m=
Q(Jm)

0,
if

m=

(mod
8).

8).

It

follows that 2

is

Now if m =

a prime in

m=

(mod

we observe

that
1
*

^UH = 2
4
8

= LlU/ ~
2

+V
'

and 2

-r

(1

solutions in

odd

\(x ?/Vm) are not associates in In fact their quotient is

Vm)/2, so 2 cannot be a prime in fi(\/w). Hence (9.12) has integers x and y. Now the primes %(x + y^w) and

2(Vm)

because their quotient

is

not a unit.

=
8

which
(5)

Ifp = \m\ then/ = Vra and hence/ is the associate of the square of a prime in Q(Jm) by Theorem 9.24. Ifp < |m|, we note that

not even an integer in Q(\fni). Let/ be a rational prime divisor ofm.

is

Vm

(9.13)

in

.p-

= VSi-VS.
is

But/

is

not

ajdivisor

of >/m in g(\//w) by Theorem 9.20 and hence/

not a
9

of/. proved provides a method for determining the primes of a quadratic field having the unique factorization property. For such we look at all the rational primes /. Those / for which =1 (/, 2m)

prime in Q(\/m). Therefore/ is divisible by a prime IT, with TV(TT) = p and hence ?r is not a divisor ofm//. But, by (9.13), -n is also a divisor of Vra, TT* is a divisor ofm, and hence ?7 2 is a divisor

The theorem we have

*

just

together with
for which

all their

associates in Q(*Jm)

are primes in

m). Those

(/,2m)

p = 77l7r2j a product Any other factoring

primes

and (-\ \Pf

= +1

will factor into

of two primes of Q(\//w), with Nfa) N(ir 2 ) /. of / will merely replace and ?7 2 by associates. The

=
/

for which (/,

2m)

>

will either

be primes of g(v m) or products

of two primes of 2(Vm).

Suppose that a is an integer in Q(Jm) and that JV(oc) = prime. Then a is also an integer in Q(^/m) and oca = 7V(oc)

/,

/ a

rational
this

p, and

212
necessitates that a be a prime in Q(\Jm). If

Algebraic

Numbers
write
4),

m^

(mod

4),

we can

y\[m, 7V(a)

we can

write a

(a;

= +

x2

my , with integers x and y. If m 2 2 with x and /m/ yJm)/2, 4N(oc) = #

2
,

(mod

?/

integers,

both odd or both even.

Combining

these facts

we have
let/?
I

the following. Let

factorization property,

and

be a rational prime such that

4),

Q(\lm) have the unique = 1, (/>, 2m)

of the two equations

=
(
)

+1. Then

if

(mod

one at
a,

least

\P/

my 2 =

has a solution. Let x

b be such a solution.

Then

the numbers a

b^fm, a

b\/m, and the associates of a and a

are primes in Q(\lm), and these are the only primes in Q(\Jm) that divide/?. 1 On the other hand, if (mod 4), one at least of the two equations

m~

= a, y = b, we can say that the numbers denoting such a solution by x = (a b^m)/2, and their associates are primes in oc = (a + b\/m)/2, a
Q(\/rri), and these are the only primes in Q(*Jni) that divide p. It is worth noting that our consideration of algebraic number fields has thus given us information concerning diophantine equations.
It must be remembered that these results apply only to those Q(-Jni) that have the unique factorization property.

my =
2

4/7

has a solution with x and

?/

both odd or both even. Again

Example,

m
/m\

1.

Gaussian primes. The

I

field is

Q(i) and

we have

2m = -2,

+
4fc

= 2,
I

/,

_ r+1

ifp
if/i

= 4k + ==

(p)
solution since # 2

"1-1
/>

+3.

For each rational prime p of the form 4k

solution x

+ y2 = = a^y = b p
.

the equation x 2

2
/

= />
+

has a

is

clearly

impossible. For each such p choose a

The primes
^
of
ib pt

in Q(i) are

/,

all

rational primes

p = 4k +
1
/

3, all
1
1

ib^

all

together with
1
/

all their associates.

Note

that

has not been

included since

/(I

-f /), / is

a unit of Q0") and hence

/is

an associate

!+/.
Example,

m =

__
3.

The

field is

Q(\l
3y
2

3)

and we have

2m = -6,

x2

4-2

has no solution,

if/?

(-;)= -lif/7

=3^ + = 3k +

1, (j7,6)

2,

(/?,

6)

= =

1.

9.10

The Equation x3

=z 3
,

213
,

For each odd p

3k

choose g g 6 g such that a|

3Z>|

= 4p.
=
3/r -h 2,

The primes
all (flp

in

Q(^f^3)
3)/2, all
(3

are 2, (3

+V

3)/2, all

odd

rational primes/?

+ ^V
we^omit
3)/2.

(^

b^

3)/2, together
it

with

all their associates.

Here,

again,
(3

^j

3)12 because

can be shown to be an associate of

the/?

+ ^/

We

could have included 2

among

3k

4-

2 by just omitting

the

word "odd."

PROBLEMS
In the second example, where m = 3, we know from the theory that if/? any prime of the form 3k + 1 then there are integers x and y such that # 2 + 3?/ 2 = 4/7. Let x = 2u y and establish that any such prime can be 2 2 in the form + uy y expressed
1.
is
,
.

2.

The

rational

prime

13 can be factored in

two ways

in

13

Prove that

this is

not in conflict with the fact that Q(-\f

1

3)

has the unique

factorization property.
3.
4.

Prove that

->/3

and \/3

are associates in Q(\J3).

= Prove that the primes of fi(\/3) are \/3 1, ^/3, all rational primes p 1 (mod 12), and 5 (mod 12), all factors a + b^/3 of rational primes/? =
associates of these primes.

all

5.

Prove that the primes of Q(^/2) are ->/2, all rational primes of the form 1 and 3 and all factors a + b\J2 of rational primes of the form Sk 8k
,

all

associates of these primes.

6. Prove that if m is square-free, m < 1, \m\ not a prime, then Q(\/ni) does not have the unique factorization property. Suggestion: use part (5) of Theorem

9.29.

9.10

The Equation x 3

z3

3 = z* has no solutions in positive rational prove that x + y* x it will z. be established that a 3 + /? 3 + y 3 = has Even more, y, integers no solutions in nonzero integers in the quadratic field 2(v--3). Note that 3 3 3 this amounts to proving that oc + y? = y has no solutions in nonzero

We

shall

integers of

-) =
3

2(V

3),

because

this

3 equation can be written as a

3
/9

o.

Algebraic

Numbers

For convenience throughout this discussion we denote 2 and co + 1 = by to, which satisfies the equations o> +
notation the units of

1 3

+
1.

co

3)/2

In this
9.22.

QCV 1^) are_l

co,

co

2
,

as given in

Theorem
9.24.
6.

Also, in this
this

field the integer

V-3

is

a prime, by

Theorem

Because

the discussion we denote it by prime plays a central role in associates of 6 are by the six units, we observe that the
(1

Multiplying

(9.14)

d.

co),

(1

co ),

(o>

O=
+

V-3.

Lemma

9.30
1

Every integer

in

g(V-3)

is

congruent to exactly one of 0,

+ 1,

modulo

a and 6 are b6)/2 in g(V-3), where Proof. Consider any integer (a an integer, also is a0)/2 rational integers, both even or both odd. Then (b

and so
\(a

b6)

= K* +
is

^)#

2a

2a
or

mod
1

)'

Now

the rational integer 2a so the lemma is proved.

congruent to

0, 1,

modulo

3,

and

6
\

3,

Lemma 9.31
(mod
3

Let
f
3

and 77 be
1

integers
4
).

of g(V-3), not

divisible

/y

6) then ==

(mod
0) rte
3

^s

o (mod
f
3

(mod
Proo/.

0) fAe

// f s -1 3 I + ^ =
4
).

(mod (mod

8 0) rtcw f 4
).

s -1

byO.Ifg=l
3

Finally if f

(mod
^
3

4 ).

=
1

(mod
it

From Lemma
(9)

9.30

follows that |

=
/?.

(mod

then f

00 for some integer

2

1 (mod Then

(9).

First if |

9^S

3
/3

300

3
/?

93

(mod

4
(9

because

9.

Also we note that

But

is

a divisor of

/?(/$

1)0?

1)
(9)

by

Lemma

9.30
1

and hence f 3
(9),

(mod (mod

6 6

4
).

4
)

= I (mod 0) because is a divisor of f (f - l)(f + 1), and so = If f s (mod 0) then P + ^ o (mod 0) implies f + = (mod = ~1 (mod 0) and hence & + ^ = 1 - 1 == (mod 4 Finally if 3 P - rf = (mod 0) then f3 + (-^?) = (mod 0) and so f + (-^) =
Now
|
3
<9

Second if f (mod and | 3 = -1 (mod 4 ).

= -1

then (-f)

(mod

(-^)

= =

17

(9).

6>

r]

).

(mod

6 4 ).

Lemma
a
3

9.32
y*

+^+

= o.

Suppose there are integers

Ifg.c.d. (a,
j8,

a,

ft,

y)

?/ze

duf&j

y of <2(\/^3) such that 0/7^ <3J o/y one of

a, 0, y.

9.10

The Equation x 3

z3

215

Proof.

Suppose that

6 divides

none of
y
3

a,

ft,

y.

Then by Lemma
l

9.31

=
Considering
divisor of 3,
at least
all
1,

a3

3
/3

= =

(mod

4
6>

).

possible combinations of signs, -1, or 3. But O 4 9, and hence

ft,

we conclude that we conclude that 6

is

divides

one of

a,
if

y.

Furthermore
to hypothesis.

6 divides

any two of them,

it

must divide

the third, contrary

Lemma

9.33

6 f a/?y,

and

units

Suppose there are nonzero integers a, f} y of g(v and a positive rational integer r such that 15 2
9
,

3),

with

0.

Then

sl

an d

^
3

2.
3
3

Since r

we

see that
co, co )
2

a3

1,

> we see that a 3 + e^ = (mod Using Lemma 9.31 + e^ = The unit s is one of + ^(1) = (mod co and so + fii(l) is one of 2, 0, -2, (1 co),
(9

).

(9

).

with all possible combinations of signs. But 6 s divides none of 2 2 these except 0, because 1 co and 1 co are associates of 0, 1 co co 2 co and 1 o> are units, and 4 whereas 7V(0 3 ) == 27. It
(1

follows that

By Lemma

+ ei(l) = 0, 3 + s^ = 9.31,
1
oc

so

Sj.

N(2) = =
1.

From

this

it

follows that

is

(mod ) implies a r a divisor of s^(6 y) 3 and r

a,
/?,

+ s^ = ^ 2.
7
in

(mod

4
<9

).

Lemma
e, a/7

9.34

d#

rational integer r

There do not exist nonzero integers ^ 2 .ywc/z

oc

Q(v

~ 3),

(9.15)

3
/?

e(6 y)*
r
/?,

0.

We may presume that g.c.d.

more,
select

(a,

6 y)

and that 6

<r

y.
oc

Further

6 does not divide

both a and
that 6 f
ft.

/3,

and

so, interchanging

and

ft

if

necessary,

we may presume
such that

If there are integers satisfying (9.15)

set

(9.16)

N(a

3
/S

e3

VO
.

minimum. This can be done because every norm in <2(v 3) is a non== + 1 negative integer. Note that e in (9. 1 5) is omitted in (9. 1 6) because N(s) We now construct a solution of (9.15) with a smaller norm in (9.16), and
is

this will establish the

lemma.

Since r
(9.17)

2,

we have a3
a3

3
/3

(mod
j8)(a

6 6 ). Also

+P=

(a

a>0)(a

co /?).

216

Algebraic
first
2

Numbers
co/?,

We
a

co /?, it
TT
|

prove that if any prime TT divides any two of a + /?, must be an associate of d. First if TT (a + ft) and
|

<x

+
\

and
co/?)

TT
1

then

/?(!

co)

and
TT
\

TT
|

oc(l

co).

But
|

g.c.d. (a,

/?)

(a
2

and
|

co is

an

associate of 6 by
/?(!
a> )
2

(9.14).

Second
co ).
\

if TT

(a

+
|

/?)

and

77

(a

+
|

co /?)
2

then
2

and
2

a(l
if TT

by
|

(9.14).

Third

(a

Again we see that TT (1 2 (a -f co /3) then to/?) and TT

|
|

co )
TT

and so

/?(co

co )

TT co ), and oc(o) again by (9.14) we get TT 9. Furthermore, because of (9.14) and the fact that 6 f /?, we notice that the 2 co /? are divisible differences between oc {3, a + co/?, and a + by 6, but not

and

by
6
a
,

2
.

The product of
d
c

b
,

by are the highest powers of 6 dividing a 4-

these three

is

divisible

6
,

as in (9.17).

/?,

-f co/?,

Hence if and a + co 2 /?
that a, b, c

respectively, then from this argument are 1 , 1 3r 2 in some order, and

-

and

(9.15)

we conclude

/?

co/?

co /?

are integers with no written as

common prime

factor in <2(V

3).

And

(9.15) can be

so each of the factors

on the

left is

an associate of the cube of an integer, say

(9.19)

j8

s^l
+
co^S)

tx

+
2

co

&

A|,

co

2
/?

3 (9

^,

where
(a

e l9 e 2 , e 3 are units.

Also we note that

/?)

co(a

co

(a

co

^)

(a

+ =

/?)(!

co

co )

0,

and so
(9.20)

s^Xl
4

+ s^l +
.

s 5 6 Xl

0,

where

=
4

co 2

and

co e 3

Thus
left side

and

6 5 are units,

and

(9.20)

is

symmetric in the three terms on the

,

of the equation. Thus we can assign the values 1 1 , 3r 2 to a, b, c in any order, say a c 3r 2. these values in 1, 6=1, Substituting

=
7

(9.20)
(9.21)

and dividing by

s-^6

we

get
6 A^

#+

+ 8 (r-%) =
3

0,

where 6 and 7 are the units 4/ x and fig/^. Since y ^ we see that A^/lg 5^ from (9.18) and (9.19). Also 6 f 2 ;Lj) so by Lemma 9.33 we conclude that = r 1 and ;> 2. 1 But 6 (9.21) is of the form (9.15) because 6 A| is

CM

either A| or

A 2 ) 3 Taking the
.

norm analogous

to (9.16)

we have by

(9.19),

Notes on Chapter 9
(9. 18),

217
3r,

and a

+c=

(a

flXa

a>/?)(a

2 co /?))

3 and 7V(oc) because JV(6) This completes the proof of

1,

N(P)

1.

Lemma

9.34.

Theorem 9.35
as
_j_

^93

_|_

_- Q y3

There are no nonzero integers a, /?, y in Q(\j 3) such that JT^T^ re no positive rational integers x, y, z such that

Proof.

assertion follows from the first. To prove the first, 3 = 3 3 0. We nonzero integers a, /?, y such that a + /? + y suppose there are = 6 divides Lemma 1. Then that 9.32, by g.c.d. (a, f} y) may presume r 6 y. Let d be the highest power of 6 dividing y, exactly one of a, /?, y, say = ry! where 6 -r y x Then by Lemma 9.33 we conclude that r ^ 2, and say y

The second

oc

+ p* +
9.34.

(6

yJ*

0.

But

this contradicts

Lemma

PROBLEMS
1.
3 3 3 3 ^a + 2 + e3 y = 0. Since e^ can be written -^(-a) we may 2 e that = Likewise for e 2 and 3 Prove 2 es are l9 1, o>, or co presume that ^

Suppose there are nonzero integers

3
.

a,

/?,

y in <2(y
.

3)

and

units

19

2>

such that

2 1, co, co in

some

order.
1

2.

Prove that there are nonzero integers and units as in Problem

such that

0.

NOTES ON CTHAPTER

we It can be noted that after Sections 9.1-9.4 on algebraic numbers in general, turned our attention to quadratic fields. Many of our theorems can be extended to to fields of algebraic numbers of higher degree, but of course it is not possible of brief Our fields. for those algebraic survey obtain results as detailed as quadratic numbers has omitted not only these generalizations but also many other aspects of
algebraic
9.2.

number theory
The

that have been investigated.

it is

A complex number is said to be nonalgebraic or transcendental if

basic mathematical constants
rr

not

algebraic.

proofs are given in and Ivan Niven, listed in the General References.

and e are transcendental numbers; the books by G. H. Hardy and E. M. Wright, W. J. LeVeque,

218
9.8.

Algebraic

Numbers

The only fields (Q^rri) with m < having unique factorization are the -1, -2, -3, -7, -11, -19, -43, -67, -163. This result was only settled completely quite recently. Harold M. Stark, who contributed to establishing and status of this and related this result, gives an excellent account of the history
cases

m =

problems in Chapter 8 of his book, "An Introduction to Number Theory" in the General References on page 270; see especially his Theorems 8.21,

listed

8.22,

and

8.23.

An

alternative proof to that given in

Theorem

9.27 of the unique factorization

property of Gaussian integers is given in Special Topics, page 268. The Diophantine problem of expressing Gaussian integers as sums of squares can be found

on page 267. The Eisenstein irreducibility criterion for polynomials is set forth on page 268. For further reading on the topic of this chapter, see Chapter 8 of the book by Harold M. Stark listed in the General References, page 270, and:
Ethan D. Bolker, Elementary Number Theory, Chapter
Benjamin, 1970;
6,

New

York, W. A.

Harry Pollard, The Theory of Algebraic Numbers, Carus Monograph 9, New York, John Wiley and Sons, 1950; Abraham Robinson, Numbers and Ideals, San Francisco, Holden-Day, 1965.

10
The
Partition

Function

10.1

Partitions

Definition 10.1

defined as the number of a sum of positive integers. as ways that the positive integer n can be written be are not considered to Two they differ only in the order

The partition function

p(ri) is

partitions

different if

of their summands.

It is convenient to define

p(Q)

1.

For example
1

+
,

5=5=4+1=3+2=3+1+1=2+2+1=2+ =
+
1

1,

and

p(5)

7.

Similarly,

/?(!)

1,

satisfy

Other partition functions can be defined for which the summands must certain restrictions. We will make use of some of these.

Definition 10.2

p m (n)

= the number of partitions ofn into summands less than or equal to m. = the number of partitions ofn into odd summands. p(ri) = the number of partitions ofn into distinct summands. p*(n) = the number ofpartitions ofn into an even number of distinct summands. q*(n) = fa num b er Ofpartitions ofn into an odd number of distinct summands. qo^} We make the convention m (0) = />(0) =/(0) = f(0) = 1, <f(0) = 0.
t

Since

p 2 (5)
4

= 3. Also + = 3 + 2,
1

5=2 + 2+1=2+1 + + = + + + +
1
1
1
1

we have

and

= 3 + + = + + + + 1, = 3 + 2, and 5 = 5, so that 5 = 4 +
1

and 5

/(5)

3,

'=1.

219

220

The
10.1

Partition Function

Theorem

We

have

(c)

(d)

== p m _(n) + p m (n p m (n) d = f(n) + q(n). p (n)

-m)ifn^m>l

Proof.

With the possible exception of

(c),

these are all obvious

from the

that each partition of n counted by p m (ri) definitions. To prove (c) a summand have not either has or does equal to m. The partitions of the

we note

second sort are counted by p m -i(n). The partitions of the first sort are m into summands obtained by adding a summand m to each partition ofn in If n = m, the number. are hence and to or less than AM) m, p m (n equal = = m. n 1 counts the single partition m) term/7 m (

Theorem
Proof.

10.2

Forn^l

m have p
,

(n)

= p(n).
,

Consider any partition counted by p(ri). It will consist of, say, r x r s summands a s where the a i are distinct summands a l9 r 2 summands 2
,

odd
r,

integers

and
b/
( ]

bf

2>,

]T?=i r ^ a ^

or
are
d

1.

Now we can write each r in Then n = 2*-i J, b^Ua, gives us a

n
-

the

form

of n whose

summands

the a t are distinct

partition
is

and

the integers 23'a, for which b f y odd, the summands 2 'a i are distinct,
all

1.

partition Also, since

this

and

new
write

counted by p (n\
If

We can reverse the process.

each
ck as 2
ek
,
,

=
,

2?Ui

ck an(1 the c fc are distinct,

-

we

dk dk odd. Then we to be found among the dl9 d2

-

let
-

a l9 a^
t,

d and we

b\

otherwise.

the partition

Then 2*-i2; *f >2 Xof n into r l summands a l9 r 2 summands a 2

}

let b\^ = = SLi^* =

,

,a s be

all

the different integers

1

7Z
-

some ck and we S et back

if
is
,

3 2 a{

,r s

summands as

have found a one-to-one correspondence between d the partitions counted byp(n) and those counted by p (ri) and hence we have

where

ri

^ bf

2>.

We

10.2

Graphs
If n
t

A partition of n can be represented geometrically.

is
fl

al

a2

ar

a partition,

= "=
'

fl

we can arrange the summands a Then the g raP h of tri is partition is r-

in such a

way

that a l

a : points in the top row, a 2

the array of points having in the next row, and so on down to a r in the

10.2

Graphs

221

bottom row.

19
If we read the

+ 2+1.
we
5

graph

vertically instead

of horizontally,

different partition. For example, from 19 4 3 1. From a 3 5 3 19 partition n

obtain a possibly

+ + 2+1 we get = a + a + ----h ar

2

a we obtain a partition consisting of r summands with largest summand l9 of n into a summands with largest summand r. Since this correspondence is
reversible

we have

the following theorem.

Theorem 10.3 The number of partitions of n into m summands is the same as the number of partitions of n having largest summand m. The number of into at most m summands is p m (n). partitions ofn

Theorem 10.4

Ifn^.0

then

((_!)'

ifn

(3/
.

j)!2for some j

0, 1, 2,

?()-*"(*)=.
[0

otherwise.

Proof.

= 1. We now suppose and q e (G) For n = we have y = f(0) = a n a + ar into distinct + 2+ n ^ 1 and consider a partition the point farthest denote A summands. In the graph of this partition we let l there will be no are distinct in the first row. Since the summands to the
right

point directly below

A^
.

< below the point that is = = will be no such point A 2 the point that is immediately to the left of A 2 If point A* directly below a 2 = a l - 1 and a z < a 2 - 1, there will be no point A*. We continue this
.

there will be a point A 2 directly a 1, there immediately to the left of A^ If a 2 be a will there al 1 and a: If 3 2, then a 2
If

a2

=a

1,

, As , thus obtaining a set of points A l9 A 2 ,process as far as possible, the of with slope 1. We also label the points 1 5lying on a line through AI Bt t ar Notice that Bt and A, may be the same bottom row Bl9 B 2

>

point.

A, A*

222

The

Partition Function

Now we wish to
into distinct

change the graph into the graph of another partition of n

summands.
etc. It

placing them to the

right of AZ,

we try taking the points Bly B 2 ,'-',B t and A t B to the right of A l9 B% to the A of l9 A 2 right s or if t s and is obvious that we cannot do this if /
First,
-

B = As
t

However we can do

it

if

<
,

s or if

obtain a graph of a partition into distinct

and B ^ A a9 and we summands. Second, we try the

t

>

A s underneath reverse process, putting A l9 2 I OT s t will give a proper graph if and only if s s or if t In other words, we can move the B^ if t
,

Bl9 B2

.Z?

This
a
.

<

<

= =

and
t

s
.

and

B ^A B ^ A We
t

can move the

is
t

in which both the

t 1 and B t ^ A s There is no graph can be moved. The graphs in which neither = s and Bt = A s and those for which s = possible are those for which t = I and B t A 8 Starting with a partition P for which the A can be moved,

if

<

or

if s

A and
t

the

we obtain a partition P' having the s points that were just moved in the bottom row. These points are in the B t position and, of course, can be moved back
to return to the partition P. similar situation arises if the partition P is such that the B t can be moved. In both cases the number of summands in
1. This pairs off all the partitions of n for which can be moved, and in any pair P, P', one has an even number of summands, the other an odd number. Let us consider the exceptional partitions for which neither the A i nor the Bi can be moved, those for which B t = A s s = t or t 1. Since B t = A s the graph consists of s rows. The bottom row has t points and the partition is n = #! + <2 2 + + as as = t, a5_ x = t+I,a 1 = t + s 1. Therefore n = st + (s - l>/2. If s = t, this is n = (3s 2 - s)[2; and if s = t 1, it is

P'

differs

from that

in
t

P by

either the

or the

'

'

n
2,

(3s
,

+ s)J2.
if

are

by q

(ri)

2 1, j)/2, s easy to verify that the integers (3s Therefore we have paired off the partitions counted with those counted by q(ri) except for a single partition into s

It is

all distinct.

summands

(3s

s)f2.

This means
if
/I

q*(n)

q(n)

((-1)
10

(3J

5)/2 for

some

1,

2,

otherwise.

PROBLEMS
1.

Let p'(n) denote the number of partitions, n = a x 4- a 2 + + a r , of n into summands a x ^ a 2 ^ a z ^ ^ a r = 1 such that consecutive at differ by at most 1. Read the graphs of such partitions vertically in order to prove
*

that,/(/i)
2.

=/Oz).

Consider n dots in a row, with a separator between adjacent dots, so n 1 1 separators in all. By choosing j separators to be left in place while the
others are removed,

and then counting the number of dots between adjacent

10.3

Formal Power Series and Euler's Identity

separators, prove that the equation

223

a?!

x2

xj

In

1\
i
I

has

solutions in positive integers, where two solutions x l9 x Z9

f
-

xs

#J are counted as distinct if x k ^ xk for at least one subscript k. (Note that the order of summands is taken into account here, so these are

and

x[, x%,

not partitions of n.)

3.

= 1, 2, 3, in the preceding problem, taking y prove that the number n ~~ 1 of of ways where again the order writing n as a sum of positive integers is 2
By
,

of

summands

is

taken into account. For example,

if

4 the sums being

counted are
1

+1 +1 +1,1 +

2,

1,

1, 1

3, 3

1,

2, 4.

10.3

Formal Power Series and Euler's Identity

depend on the theory of this book. In the sections and are the of functions beyond scope analytic without the use of we will obtain some results one this interesting following

Many

results concerning the partition function

very

much analysis.

In this section

we get several results on partition

functions

by using power
needed.

series in

a formal way, so that no theorems on analysis are

,

The power series that we use are of the form a + a^x + a^ + a zx* + where a Q = 1 in most cases. Such a power series is treated "formally" if no is a dummy variable, numerical values are ever substituted for x. Thus
a l9 a^ a z and the power series is really just a sequence of constants of the x identification for the to retain convenient it is very However, easy e ^ to ^ e sa are a and series Two q ua ^ if M' ]T f& power general term. defined is series two = of these The all a power product subscripts y. bj for
*

to be

With

these definitions of equality

and

multiplication of formal
5^
is

power
(and b

series,

the set

of

all

power

series

with real coefficients with a

0)

forms an abelian group. The associative property

follows from

easy to prove; in fact it the associative property for polynomials in x because the

224
coefficient of x
n in

The

Partition Function

terms in

n to x so that all any product is determined by the terms up in all power series in proving higher powers of x can be discarded

n that the coefficients of x are identical.

The
of any

identity element of the group ai x* l $ a given power series

is 1

or

-f

Oz

po wer

series

Oz 2

The

inverse

(^ ^

exists)

such that

(2 ^=0
holds.

that the coefficients 6

The definition of multiplication of formal power series suggests at once can be calculated in turn from the sequence b^ b 29
,

of equations
aQ bQ

1,

#0^1

a ibv

3=0

because # 5^
of the

0.

Thus the inverse

exists. Finally, the

group

is

abelian because

symmetry of the definition of multiplication. x is readily calculated to be 1 The inverse of 1

+x+

x2

+x +
3

As in analysis, this is called the power series expansion of (1 Under suitable circumstances an infinite number of power
multiplied.

a)-

1
.

series

can be

An

illustration of this is

71=1

a product that will be used in what follows. The reason that this infinite x m for any positive integer product is well-defined is that the coefficient of

depends on only a

finite

number of factors,

in fact

it

depends on

71=1

In general let leading term

PIt P2 P 3
,

be an

infinite

sequence of power series each with

is well-defined if for the infinite product a finite number of the in only every positive integer k the power $* occurs m it that the x term in the is clear power series. For if this condition is satisfied
'

1.

Then

JWV
'
'

Pr where r is chosen so determined by a finite product PiP^z has P P P series that none of the power r+3> r+1 any term of degree m r+2 1 in each series. term the constant of course or lower, except
*

product

is

'

>

10.3

Formal Power Series and Euler's Identity

225

x n )~ l has the expansion The function (1 m and 2, multiplying we find

,

2L

^'

n
-

Taking n

1,

(I
oo

X'
oo

1=0 32=0

where c

is

the

number of
9 ,

solutions of j\
-

non-negative integers j\ j2

,j m

That

is

+j 2 + +jm m = jin = p m (j) and we have c,

1
-

In a similar

way we

find

1 icn )~ is called the generating function for p(ri) The function JJ=i (1 and application of analytic methods to this function leads to results about p(ri). ^ n )~ 1 Similarly the The generating function for p m (ri) is IKU (1
9
.

generating function for p(ri)

is

found to be

n=0
d and the generating function foip (n)
is

^ 2n ~1 )-1 This formula can be proved directly and then used to give another proof of Theorem 10.2. Formally, at least, we have

Theorem

10.2

is

equivalent to

n?-i

C1

^ n)

= IKU

(1

1.

226

The

Partition Function

Taking n

1, 2, 3,

and multiplying we
00

find

00

and

In a similar

way we can

multiply out

U=i

~ *") formally to get

n a-**) =
K=1

Then Theorem

10.4 implies

ft
This
is

(1

*")

;=i

known

as Euler's formula.

formal sense that the coefficients

Whereas here we have proved it only in the of the power series are identical, an analytic

proof is given in Theorem 10.9 with convergence indicated for suitable values of x. Since a variable is never assigned a numerical value in formal power
series, questions

of convergence never

arise.

Theorem 10.5
p(n)

Euler's identity. For any positive integer n,

1)

= p(n -

p(n

+ p(n - 2) - p(n - 5) - p(n - 7) - 12) + p(n - 15) -

-where each

sum extends over all positive integers j for which the the partition function are non-negative.
(1
a;
71

arguments of

Proof. From Euler's formula and the fact that II function for p(n) we can write

)-

is

the generating

{i

+ >X-i)V

3'

+ ' )/i

or

10.4

Euler's Formula

227

Equating

coefficients

of

xn

on the two

sides

we

get
7)

p(n}-p(n

1)

-p(n

~ 2) + P(n -

5)

+ p(n 12)

-p(nand thus the theorem

is

-/?(-

15)

+ -" =

0,

established.

PROBLEMS
1.

Show

that the infinite product

(1

%)(!

+ a^Xl +

'

'

o^a^Bj)

V a^i
-

ajg*

where a,

a is or 1, and a fc 1. Count the number of terms in the that are of #2 xz x to show that expansion degree n. Set x 1 # 2)(1 #3) is the (1 oO(l function generating for//(/i) of Problem 1,

Section 10.2.
2.

Compute a short table of the values of p(n), from n = 1.) of Theorem 10.5. (Recall that p(0)

to n

20, by use

10.4

Euler's

Formula
will

In this section

we

functions, not just in the formal sense. serious limitations, so it is convenient

prove Euler's formula as an equality between two Formal power series arguments have now to use a few rudimentary facts

concerning

infinite series

and

limits.

reader familiar with the theory of

\x\

analytic functions will recognize that our functions are analytic in and will be able to shorten our proofs.

<

1,

Theorem 10.6
n
(ri)x

Suppose

<;

<

and

let

<f>

m (x)

Un-itt

^ n). Then

converges and

2n
00

=
<f>

1
*

'

n=0

m(x)

Proof. into at

By Theorem 10.3, p m (n) is equal to the number of partitions of n most m summands. This is the same as the number of partitions into exactly m summands if we allow zero summands. Then each summand is or I m n or 2 or or n, and we have/7 m (X) ^ (n + l) m The series (n + l) # n the ratio and hence so does ^S=Qp m (ri)x by the com test, converges, by
.

2U
9

parison

test.

228

The

Partition Function

Now
-i
1

xn

M
ch

xn

3=0

where the
2,
,

last

sum

is

a
if

finite
/i

and

ch

sum and

/?

m (/z)

< m \k.

^p m (h)

for

all

0,

1,

Therefore we have

wlfe-1

I
As k ->
oo

we have

"fe 1
^
j*

^
**n
^

and hence

Theorem 10.7

We let
Prao/
the
1

<j>(x)

ForOx<l, lim m = Iim m ^ ^ m(a; and dejine

co

m(x)

exists
(1

and is

jjjo^

^
=

different from zero.

fo

6g ^ (a;)

Since

mean
xn

<

1 the result is obvious for x <^ m (0) 0. For x we apply value theorem to the function logz to obtain a y such that

>

<

and

log

'

Therefore

- log (1 ~ (1 ~ *">

xn )

_ =

and hence

-log ^(r)

y -log

(i

- ^) ^

< y ^L ~
^m (x)-i,

*m+1

This shows that -log #(*), and hence

>

is

bounded

for

c~ oo.

fixed as

But

10.4

Euler's

Formula

229

-> oo. Since ^(X)" 1 increases monotonically for x fixed as z) 1/(1 l this shows that linv.^ <f> m (x)~ exists and is different from zero. Therefore lim^^oo <f> m (x) exists and is also different from zero.

>

Theorem 10.8

For

<

the series

'^

sss

^p(n)x

converges,

and

Proof.

We
m

have, using

Theorem

10.6,
oo

m
n=0
77

2 x)*" =
7i=0

2 *<)*" = 2 j^*)** = ^ *rl = #*>"*

(

n=0

For #

fixed,

JJLoX )^
n

increases as

m-^oo.
1
.

Therefore 2*=oJ 3 (' 2 ) a;n

lim^oo *2%=<>p(ri)x But now

exists

and

is

^
oo

^(a)"

oo

2 p()^
W=0

2p
71=0

( " )a:

"

= ^*>~
1
5

1
-

Letting

m ->
10.9

oo

we have 2S=o?(")^ n

=
#

^(:c)~"

anc* hence

^SLo^W^

11

Theorem

Euler' s formula. For

<

we

5=1
(3:i 2J o; converges; therefore so does Proo/. The ratio test shows that 2.i the above series. Let q^,(n) be the number of partitions of n into an even number of distinct summands no greater than m, and let q m (n) be the number
' '

)/2

of partitions of n into an odd number of distinct summands no greater than = 0. Then m. As in Definition 10.2 we will take ^(0) 1, ^(0)

(10.1)

<f>

m (x)

(1

x)(l

*2)(1

x*)

(1

xm)

finite
e

sum. But for n

have q m (ri)

+ qm

(ri)

^ m we have q m (n) = q(n ), qm (n) = ^(w) < p(ri) for all n. Therefore
e

and we

also

2<
" w (*)la' :Sy ^_/

m(n)|*-^

230
Since

The

Partition Function

we get 2n=o (?*() - ?()K = <(*) oo. Using Theorem 10.4, we have the present theorem. by letting m We will have occasion to multiply power series. For this we need the

2.H-i />()*" ~>

->

as

m -^

oo,

following lemma.

Lemma

10.10

Let

2?-o^

and

SwV*

0<a?<L
^x<

Then

2%*> Q#-o

^A-;K
1

converges

be absolutely convergent for and has the value

The condition
1

The sums 27=o a&*

polynomials

could be replaced by \x\ < 1 we take lemma in agreement with our other theorems. just to keep the are polynomials and can be multiplied by and

re

<

2.

the usual rules of algebra.

The terms of degree

m or less in the product of the

are precisely the terms in

m. k All other terms in the product are of the form <*$#?+* with j see that we k exceeds and one least [m/2], k of/ implies that at Since;

>

>m

w
3=0

TO

+
*=[m/

fl

when
y

fc

a sum of terms including all the terms multiplied out, will be m, and possibly others. This implies that

>

a ^''
j=Q

z '^ |+ 2
'^^ +

Jfc=[wi/2]

5=[m/2]

fc=0

=
m

2 '"^i z

2
j=[m/2]
fc

1^'

since all four infinite series in this last expression are convergent. Letting -> oo we see that J?=o CS?-o converges and is equal to

^A-^

This
for

lemma
x

implies that if 2?=o <V*

and

2,

a:
fc

converge absolutely

and

2.
ro

S=o (2?o ^^K- Applying it to 2?=o 1^1 ^^ IW ^, we find that 2^=0 (2?-o kA-;lK converges forO <j < 1.

<

1,

then so does

a:

Since |2 ; =0 a,V,-l

^ 2?=o l^^-l, we see that^^o (2J-o ^V,0^ converges

10.4

Euler's

Formula

231

absolutely for

<

Then

the

lemma can be extended

to the product

XJ that are absolutely convergent finite number of power series 2=o forO <:x< 1. Another fact that we will use is the following:

of any

Lemma

10.11
j

If
<;

2=
x

2L

bjX for
If

<

1,

aj xj an d 2?=o bftf converge absolutely and 2j=o Q>f& then a, 0, 1 2, b/for all]

Pr00/.

[a^l + |Ztyr'|, and hence ^LoCjX j x 1 and we need only show that 2 =0 CjX = for < ^ converges absolutely = 0. Setting # = we have c = 0. Suppose the lemma is false. implies Cj Then there is some Cj ^ 0, and we can let k be the smallest positive integer j = and, because this series for which ck ^ 0. Then k c x cfxf = x is there an for |, integer m > k such" that converges absolutely -1 = x 1 we have Now for 2~ < < < |c |. 2S=jt <?X 2JLm+i l^2"^"| z we get dividing by
c,,-

aj

6j,

then

\c3 x

<|

2L

^L

fc

fc

fc

Then

for

<

we can

write

V 2

|c*U

and

finally

But ck+i x

ck+ &*

c mx

m -k

is

a polynomial, and

we can make

its

to value less than the positive number \ck \j2 by choosing x close enough therefore the lemma is proved. and contradiction a is x This zero, J. The next theorem gives for the sum of divisors function, a(n), an identity

<

similar to that for p(ri) in

Theorem

10.5.

232

The
For n
1)

Partition Function

Theorem 10.12
o(n)

^
a(n

we have
2)

o(n

+ -

a(n

5)

cr(n

7)

a(n

12)

a(n

15)

=
VO

2
otherwise

where the sum extends as far as the arguments are positive.

Proof.

Taking the derivative of log

/ / v\

<f>

m (x)

log

ni
Z^

C1

*n )

we g et

A *.-! x^x^ = Vz2i^_ = y y ^ n ^-i ^

where
f
/2

^^ yy
sL-

^
*

for

a?

<

1,

if

72

A:

otherwise.

There are m series ^=1 ^n,^" 1 eac ^ can be added term by term to give

which converges absolutely. They

c>
we have
infinite series,

m^;

g- 2 (2 <">)"-'
fc=1 \ n=1

since TO (V) is a (^(/z) Using (10.1) ^(x) #. But can write in the we also form of an finite sum, a polynomial in (10.1)
<

= 2n

- q m (n))xn~i

in which

all

the terms

from a

certain

72

on are

zero.

Then equation
oo
/

(10.2)

can

be put in the

form

2n-l
WroW
QmV ))
1

^ /^\Am\
oo
e

m
ij+l

o )

4mV

n^C / X"
))

\
]

n=0

2-4\-i
3=0 \z=l

=
by

y
7i=0

\n=0

Lemma

10.10.

Then Lemma

10.11 gives us
fc-l

2-^
m
Wm(w)

m())

Cz,fc-n-

10.5

Jacobi's

Formula
e

233

For any given k we can choose m > k. Then q m (k) = q e (k), q m (k) q(k), - *) = = = = d and -a(k m (n) ?'(/*), ?'(), qm(n} 2<-i *<.*q I^-n for ^ fc 1. This with Theorem 10.4 gives us
e

-cr(fc)

c;(/c

1)

cr(fc

2)

cr(/c

5)

a(k

- 7) +

^0

otherwise

and the theorem

is

proved.

PROBLEM
1.

Compute a

short table of the values of or(),

10.12. Verify the entries

from n

to n
a(n)

20,
\

by

means of Theorem
directly.

by computing

=^Ld n d

10.5

Jacobi's

Formula

Theorem 10.13

Jacobi's formula. For

^x<

1,

3=0

The formula
For

is

obvious for x

0, so

we can suppose

<

<

1.

0<#<

1,0

<z<

l,we

define

(10.3)

fn (z) = f[
fc==1

{(1

2
<?

*-V)(l

**V*)}

J
^
2
-

^'
<JL,

where the a, are polynomials and it is easy to see that

(10.4)

in

gr.

Since

/n (l/) ==/()
1)

we have

=a

y,

an

= (-l)Y+3+5+> " +(2n - = (-l)T

we

In order to obtain the other a s

replace z by qz in (10.3) and find

/n(*)

= IT id n+1
fc=2

na-^ym
3=0

234

The

Partition Function

and hence

r *1

f[

(1

2 '- 1

If

we

write the functions

fn

in terms of the

a,-,

using (10.3), and equate the

coefficients

of z

2fc
,

we

find

and then

This, along with (10.4) allows us to find a n_ l9

n_ 2

in turn. In fact, for

<j ^

n we find

(_iy'n

4n
<7

4n ~ 2
)(l
<?

4n- 2 ' +2
)

(!

(n_^)2

and hence

(10.5)

ak

- *=^
is

a -

**)

(_ 1)

*
,

^
=

^
<

(-l)

This formula

valid for

<;

if

we agree
is

to take

(?

1-

Returning to (10.3),

we

see

that/n (z)

2 qz~ ), which has the value A derivative and then setting z q we have

which

is (1

a product of 2n factors, one of l at z q <*. Therefore taking the

= nao^-H

On

the other hand,

we
==

also have,

from

(10.3),

X
f't&A
)

/ ycLjq Zw

'-X

-A =^ / y&j<i ^-^
1

'

(Q.*

^ 0-

10.5

Jacobi's

Formula

235

Thus we

find

3=1

and hence, by

(10.5),

-aNow

")

3=1

n (-DVV - <ro -&&?*

n
a-">=si,
>m

and ^jLij<f

\<f

#~

J
|

converges, so

we have

for n

- (i -

*o

2 c^

wv
j=m+l

9-0

^
We
keep

2 ^'
let

'

lg3

oo.

r'i

^
5=m+l

'

'

1?5

"''

m fixed but arbitrary

and

n ->

By Theorem

10.7

we have

and

so that

we

get

Now letting m -*

(-D

^^

'

~j
')

gA

co

we

find

y-l

tf-0

2 (-D

where we can make the

toy
it

1,

we

last step because both series converge. Changing j write the last series as 2?=o (-1X0'+ 1 W*** and can then add

to the first series to obtain

?=o

This

is

z our theorem with x replaced by q

236

The

Partition Function

PROBLEM
Replace z by q in (10.3), multiply by proof of Euler's formula.
1.
l/6
<f>

n (f),

and use

(10.5) to obtain

10.6

Divisibility
is

Property

Theorem 10.14

Ifp

a prime and

<

then

rv

=i

where the
Proof.

a$

are integers.

For

<

we have

the expansion

3=1

'

5=0

say,

and therefore

=
say.

2
+
(P

b >u

+2

( fc ^

*'->'

2^
c
5=0

But
-

__
'

2)

g+^so
that

1)

|l

(modp)if;

1)!

ta

(mod
c

p) if j

= ^
c3
-

(modp)
(mod
p) 9

and &

< (modjp) fory >

<
bl

Z? 2

<
<
1,

we have

1,

>

0,

c3

0.

Now,

for

<:x

10.6

Divisibility

Property

237
10.10,
2^

where

a^ =
oo

c j and,

by
oo

Lemma
x

2 a^ = 2 c^"* 2 'i- =22

1

L^]
<

'**

ft=0

3=0

fc=0

7i=0

3=0

By Lemma

10.11

we then have

3=0

and hence
ah
(

m)

m-1}
al

ss ai

1}

4 ^ ai-" ^ ai
Therefore

W)

1}

= c^ (mod = ^ > 0,

jp),

Since the
verges and

sum on

the

left

increases as

m ~>

oo

we

see that

]T=o a

xJl

con

But we also have

oo

in

a^z* =

co^

a"** +

and

finally

Since

^ =
0)

and al

ft)

CA

(mod p)

for A

>

1,

the theorem

is

proved.

< x < 1 we toe z<(o;) 4 Theorem 10.15 For (mod are integers and b m = Q (mod 5) ifm =

=
5).

2U^

238
Proof.

The

Partition Function

We can

write

Theorem

10.9 in the

form
if
/c

ck

((-I)*'
\

(3/

10

otherwise

and Theorem

10.13 as

otherwise,

and then apply

Lemma

10.9 to obtain

7i=0

Then 6 m
TZ

= ^~Q
,

cfc rfw _!_ fc

0,

such that
in

=m

can be written as 1. But dn is

it is

2 ^4 summed
unless
==

over
2

all

0,

0, 1, 2,

which case
it is

ck

by

saying that
it is (
1)*.

unless k

(-l) j (2j
2

1).

+;)/2, 7 Furthermore we can describe

(j

(3z

z)/2, /

0,

2,

in

which

case

Then we can

write

(10.6)

summed

over

all

w-

and y such that y

and

(3z

+ 0/2 +

(j

+;')/2

1.

But

so that if
2(j

1)2

m= + (2/ +
2
is

(mod
2

I)

5), the

terms in (10.6) will have to be such that

5).

(mod

That

is

(2y

I)

-2(i

I)

(mod

5).

However,
2j

===

a quadratic nonresidue modulo (mod 5), and hence b m = (mod 5)

5, so this == if

condition implies

(mod

5).

Theorem 10.16
Proof.

We

have p(5m

4)

(mod
10.8

5).

By Theorems

10.15, 10.14,

and

we have

*5fc

Notes on Chapter 10

239

where the aj and b m are integers and b m = Using Lemmas 10.10 and 10.11 we find that
[n/5]

(mod

5) for

m=

(mod

5).

5k

(mod

5)

and hence /?(5m

This theorem
function.

+
is

4)

==

(mod

5) since b 5m + 5 _ 5k

(mod

5).

(mod
k p(5 n

7).

The With

only one of several divisibility properties of the partition == methods of this section can be used to prove that/?(7w 5)

the aid of

more
24r

r)

(mod

fc

) if

extensive analysis,
1

it

can be shown that

,

(mod

fc

),

&

2, 3, 4,

and there

are

other congruences related to powers of 5. There are somewhat similar congruences related to powers of 7, but it is an interesting fact that
still

k p(l n

for

+ r) = (mod 7 ) if 24r == (mod 7 is valid for k = 1,2 but is false = 3. There are also divisibility properties related to the number 11. An
fc

fc

identity typical of several connected with the divisibility properties

is

PROBLEMS
1.

Write Euler's formula as

~~"
00

r-iyys^)/2
Use and
2.

Jacobi's formula as in
verify (10.6).

Theorem

10.13, multiply x$(x)<j>(x)

out formally

Obtain a congruence similar to that in Theorem 10.16 but for the modulus

35, using

Theorem 10.16

and/?(7/z

5)

(mod

7).

NOTES ON CHAPTER

10

For a more extensive discussion of the methods of Section 10.3, including a of convergence, see I. Niven, Formal proof of Theorem 10.12 avoids all questions 871-889 Amer. Math. Power Series, (1969). This paper also treats Monthly, 76, recurrence functions in a quite different way from that in Section 4.5. For some of the original basic work on congruence properties of partitions, see
S.

Ramanujan, Collected Papers Cambridge

p

Press, 1927.

11

The Density
of Integers

of Sequences

it is

is meant by the density of a sequence of integers necessary to use certain concepts from analysis. In this chapter it is assumed that the reader is familiar with the ideas of the limit inferior of a

In order even to define what

sequence of real numbers and the greatest lower bound, or infimum, of a set of real numbers. Also, in Section 11.2 we make use of the fact that
/^- These are discussed in many texts, for example in Mathematical Analysis, by ApostoL* Two common types of density are considered in this chapter, asymptotic density and Schnirelman density. The first is discussed in Sections 11.1 to 11.3, and the second in Section 11.4. Density will be defined for a set A of

2w=i

I/"

"

77

distinct positive integers. will think in a sequence according to size,

We

of the elements of A as being arranged

(11-1)

ai<a <*s<
a
z

set
it

and we will also denote A by {a }. Furthermore we will use both the terms and sequence to describe A. The set A may be infinite or finite. That is,

It
is

may contain infinitely many elements or only a finite number of elements. may even be empty, in which case it will be denoted by 0. If an integer m an element of A we write m e A if not we write m $ A. The set A is con
,

tained in B,

c:

or

=>

A,

if

every element of

is

an element of B.

We

Tom M.

Apostol, Mathematical Analysis, Addison- Wesley, 1957.

240

11.1

Asymptotic Density

241

write

elements.

B and B c: A, that is if ^4 and B have precisely the same U B of two sets y4 and 5 is the set of all elements m such that me A or me B. The intersection A n B of A and .5 is the set of all
A
<=

= B if A

The union A

m
v4

such that

>4

\jQ and
all
is

me A and me B. Thus, for example, ^U^ = ^n^==^4, = A,A n = 0. If A and B have no element in common, A n = 0,
A
of
^4

of

are said to be disjoint. By the complement positive integers that are not elements of A.
all

we mean

the set

Thus

C\

A=

and

the set of

positive integers.

11.1

Asymptotic Density

The number of positive integers in a set A that are less than or equal to x is denoted by A(x). For example, if A consists of the even integers 2, 4, 6, then ,4(1) = 0, A(2) = 1, A(6) = 3, A(l) = 3, .4(15/2) = 3; in fact A(x) = {#J we have A(as ) =j. [x/2] if x ^ 0. On the other hand, for any set A
,

Definition 11.1

The asymptotic density of a set

3 1 (^l)

is

(11.2)

= Uminf^.
n^-oo

n
that

In case the sequence A(n)jn has a limit,

we say

has a natural density,

d(A). Thus
(11.3)
w-*.oo

n
it is

if

has a natural density. If A

11.1

is

a finite sequence,
sequence, then

clear that 6 (A)

0.

Theorem

If A

is

an

infinite

d^A)
exists, then

lim inf
rc-- oo

an

Ifd(A)
Proof.

d(A)

lim n _ co n/a n
is

The sequence k/ak

-.

a subsequence of A(ri)/n and hence

.

lim inf
w-*oo

~A(n)
n

...

lim

r mf
.

k
.

k-xx>

a^

If n is

any integer ^ a and ak %_! -^.n < ak and

is

the smallest integer in

that exceeds

then

a*

242
It follows that

The Density

of Sequences of Integers

k
ak

^+ <.A(n)
,

1 ,

r f hm mf
.

k
a

n
is

fc~>oo

r mf f A(n) ^ lim n-+ao n

.

and so the theorem

proved.

PROBLEMS
1.

(a)
(6)
(c)

Prove that each of the following sets has a natural density, and find the set of even positive integers;
the set of

its

value:

odd positive integers ; the positive multiples of 3 ; (d) the positive integers of the form 4
(e)
all

+ s
b

2;

positive integers a

satisfying

(mod

rri),

where b and

m >

are

fixed;

(/) the set of primes; n (g) the set {ar } with n the set of all perfect (/z)

1, 2, 3,

and

fixed a

1,

fixed r

>

(0
(/)

the set of the set of

all

squares; positive cubes;

positive powers, that
is,

all

all

numbers of the form a n with

exists

al,n^2.
d(A)

2. If the natural density d(A) exists,

= 1. if and only if d : (A) + d^(A) A, prove that d^A) + d(A) < 1. 5. Define A n as the set of all a such that (2n)l ^ a < (2n + 1)1 and let A be the union of all sets A n , n = 1, 2, 3, Prove that d : (A) + d(A) = 0. 6. Let A* be the set remaining after a finite number of integers are deleted from a set A. Prove that d^A) = d^A*), and that d(A) exists if and only if 6(A*) exists. 7. If two sets A and B are identical beyond a fixed integer n, prove that
4.

d(A) = 1. 3. Prove that 8 (A) exists

prove that 6 (A) also

and that

For any

set

8.

= d^B). Given any set A that ^04) = d^B).

d,(A)

{<z ;-}

and any

integer b

0, define

B =

{b

a s }. Prove

be the set of all even positive integers, B^ the set of all even positive with an even number of digits to base ten, and B2 the set of all odd integers positive integers with an odd number of digits. Define B = B u B2 , and
9.

Let

prove that d(A) and 6(B)

10. If

exist, but that d(A u B) and 6(A n B) do not exist. prove that 6 t (A v B) ^ d^A) + ^(5). 11. Let 6* denote a m Prove that any finite set of positive integers a l9 a 2> the set A of all positive integers not divisible by any member of S has natural

AnB =

0,

density

11.2

Square-Free Integers
12. Let

243

be a

4-

=m

set of positive integers such that for every integer m, the equation has at most one solution not counting order with x and y in A.

has density zero. Even more, prove that A(ri) ^ 2\Jn. = 1 define fc+1 as the least positive {a^} as follows. With a t - 0,, with 1 ^ A ^ AT, that is different from all the numbers a h + at integer 1 ^ z ^ &, 1 &. Prove that ^4 satisfies the ^ ^y inequality of the preceding

Prove that 13. Define

,4

A =

1. problem, and that A(ri) ^ ^/z k 14. Let P be the set of integers {m } with

m=
.

1,2, 3,

-and&

2, 3, 4,

Let P! be the subset with k

3, 4,

Prove that
n 0.

PI (W) -r- = ^aoPto

r lim
15.

Find the asymptotic density of the

digits in

set positive

of integers having an odd

number of

the base ten representation.

11.2

Square-Free Integers
2

An

integer is square-free if it is divisible by no perfect square a 2 will prove that the set of square-free integers has natural density 6/vr

>

1.
.

We

Lemma

11.2

The function r(n\ representing the number of positive divisors

ofn, satisfies the inequality r(n)

Proof.

2\Jnfor n^.1.
d,

Consider the positive divisors,

is

of
1

n.

vn

the distinct divisor d'

n/d,

and

Corresponding to each d' ^ \//2. Therefore r(n)

that
1

cannot exceed twice the number of divisors

the

d such

^ v n.
2v n.

Clearly

number of these d cannot exceed

Vn,

and we have

r(n)

Theorem 11.3

We

have

Proof.

Writing

we

consider
z

first
is
.

a fixed integer j

^m.lfdis

any divisor of y, say/

z

then

a term of R m and l/^ 2 is a term of S m and [J,(d)ld occurs in P m Then I// 2 occurs in P m with coefficient
l if;

fj,(d)l(dq)

= dq,

244

The Density

of Sequences of Integers

2 by Theorem 4.6. In case j> m the product ft(d)/(dq) may appear for some divisors d of y. Therefore we can write

in

Pm

2
Lemma

J=l

where 2' denotes a sum over the appropriate

divisors

d of y. Thus we have

y-m+1

d\j

j=m
that

and, using

11.2,

we observe

^ 2' M<OI ^ 2
d\3
d\j

}|

Now we

have

ii

IK ^

m w= < y 2,
, ,

-2

Applying Cauchy's condition to the convergent series ]T 2/j% we see that 1 tends to zero as m tends to infinity. In view of (1 1.4) this establishes \P m
1

the theorem.

Corollary 11.4

We

have

Proof.

It is

well

known from elementary

2

series that

2n=i

l/^

results in the theory

it

of Fourier

^fi* For

instance,

follows by setting x

in

the result

-=

7rX

-^ + 2 ^ COSnX

rr in the ^ x < 2rr. This result is to be found in range Mathematical Apostol's Analysis.* This with Theorem 11.3 proves the

which

is

valid for

corollary.

Theorem 11.5
Proof.
integers.

The

set

of square-free integers has natural density

the sequence
1, 2,

6/7r

Let

S denote

3, 5, 6, 7, 10,
,

of square-free
all

For any

positive integer n let/7 1 ,/? 2

,p r denote

the primes

Tom M.

Apostol, Mathematical Analysis, Addison- Wesley, 1957, p. 501.

11.2

Square-Free Integers

245
to prove

such that/?*

n.

We first wish

(11.5)

S(n)

y (-i
T terms obtained by setting each
2
oc

where the sum ranges over the

Now

or

1.

2 [fl/r

interpret n that are divisible

m^
If

number of integers ^ H that are divisible by t*, and we can each term on the right side of (11.5) as a count of those integers
]

is

the

no

square-free, other terms. If 1

m is
m

by (p^pf

/?^)

^m^n, then m
and
[n]

is

^.m^n

m
0.

is

divisible

counted by the term [n] and by 2 by p\ but by no other p i9

positively

then

counted by the terms - 1 negatively, a net count of 1

is

and ~[n/pl], once

and once
but not

To

integer

m, 1 ^ m ^ /z, that is divisible by /??1?jp 2 ,/, by any of the other p J. Then m is counted by the terms
-

take the general case, consider an

,

>

1,

The

net count for this

m is

thus

snce

This establishes (11.5). Next we note that (11.5) can be written as

(11.6)

S(n)=
d

sum any term restrict d in (11.6) to

In this
(11-7)

for which

d2

>

n has the factor

2 [rc/rf ]

0,

and we can

be such that d 2

n.

In fact,

we have

where d ranges over

2 all positive integers such that d 1 77, since any term in to a d that is not square-free. that value of is not in (1 1.7) (1 1.6) will belong 0. In this case the term has the factor p(d)

Using Corollary 11.4

and

(11.7)

we

find

The Density

of Sequences of Integers

as n

>-

oo

and
as n

-> oo

since

we

of (11.8) tends to zero, and converges. Therefore the right side 2 -> oo. have S(ri)/n -+ 6/7r as n

2 1M

PROBLEMS
of integers divisible by no square >4. of integers divisible by no square >100. 3. (a) In Corollary 11.4 separate the infinite series into two parts Si and S2 where S^ is the sum over odd values of d, and S2 over even values. Prove that Sl = 8/7T2 (b) Find the density of the set of positive integers divisible by no
1.

2.

Find the density of the Find the density of the

set set

odd square >1.

11.3

Sets of Density Zero

infinite

need the following well-known result from the theory of products. For convenience we prove it here.
shall

We

Lemma
JJ?=1
Proof.
(1

11.6

Let

2 ^ be a divergent series with <

real

cs

<

Iforj

1, 2,

Then, given any

Cj)

number

>

0,

there

is

an integer

such

that

<

sfor every integer

that

n^.N.

We note

3!

4!

5!

and hence that

3=1

1 1 .3

Sets of Density

Zero

247

Since

Cy

diverges

we can choose
e

N so
?N ~1 C
Z-'i J
1
3

that
^ <

and

the

lemma

follows.

one item of special notation. For any set of the set of those elements a of A integers A and any prime p, Ap will denote 2 such that p a but p i a.
In this section

we

shall use

Theorem 11.7

If there

is

a set of primes {pj such that

^pT

diverges

and

AP

has natural density zero for

i=
of

1, 2, 3,

then

has natural density

zero.

Proof.
2

Let / denote the

r '
, -

set

all

positive

integers,

let
,

u4 4 u U A u X
x

and

let
,

B
(r)

(r}

= C^.

Then A

n I9( = A 9i

C = /PI U A nC =
(r)

(r)

4p

and hence

(11.9)

A c
for at least

Now B M
p*

^u^u
,

U A 9r

in

consists of all positive integers except those such that r, and we will prove that 1, 2,

pj

n,

oney

(H.io,

^,.

(-OP
3
r

1^

'

where the sum extends over the

or
2.

The proof is

< similar to that of (1 1.5). Any positive integer m

jtf'fc,
(fe,

terms obtained by taking each

a,-

0, 1,

n can be

written as

m=

pf

pr)

1,

ft

0.

= or 1, and y i 2 if ft Let y i ft if ft terms on the right side of (11.10) for which

and
it is

^
r.

2.

Then
af

is
,

counted with the sign (-l)i+

a *+-+ a

y f z = 1, 2, The net count for m

<

counted by the
-

r,

is

then

=
a 1==0

02=0

a r =0

z=

n
< i

But

= if y = 1 - 1 + 1 = 1 if y, = 2 and we see that m has a count of if any y = and otherwise has a count if and only if ft = 1, the right side of (11.10) counts the Since y< = of < in B n that are m \ and (11.10) is established.

a-

=lify =

1-1

JL

1.

(r

248

The Density
the greatest integer symbol in (11.10)

of Sequences of Integers

Removing

we

get the inequality

hence

To prove

the theorem

such that A(n)fn

we must show that for any real e > <sforn^N. First we choose r so that

there

is

an

which can be done by

diverge. The sets such that

Lemma

.6 since

2 p7

l
9

and hence

also

2 Of

pT*)>

Av

have natural density zero so we can find an integer NI

(11.13)

2r
ls

ifn

JVj.

Taking

A^ JV

AT

^3r -4/,

and using

(11.9), (11.11), (11.12),

and

(11.13)

we

see that

=
n

^
n
n

Theorem 11.8 Let k be a fixed positive integer. If each integer divisible by k or fewer distinct prime factors, then d(A) = 0.
Proof.

in

set

is

all positive integers having k or fewer c Then A D A(n) ^ D (ri), and we need only prime (k) is by induction on k. For k = 1 the The the for D theorem prove proof (1) (1) = {/>*}. We apply Theorem 11.7, consists of all prime powers, D set D 1 taking the^ to be all primes. The series 2/f diverges by Theorem 8.2, and = since D consists of the single element^,-. Thus d(D (l) ) = 0. <5(Z> ) 1. Turning to general k, we assume that the theorem holds in the case fc The elements of D f are the positive integers that are divisible by p but not z by p and that have k or fewer distinct prime factors. If ae D!?\ then k ~1 (k alpeD^v. Therefore D \n) <> D ~v(nlp), and hence d(D^ ^ = * = 0. Now we can implies d(D ) apply Theorem 11.7 as before, and it let

We

(k)

denote the set of

distinct

factors.

(lc)

(7c)

follows that

d(D^)

0.

1 1 .3

Sets of Density

Zero

249

As another
Theorem 11.9
density zero.

application of

Theorem

11.7

we prove

the following:
-

The

set

of integers

{<f>(m)} 9

m=

1,

2,

3,

has natural

Denote the set under consideration by A. Given s Proof. k so that 2~ k sets g/2 and separate A into two

>

0,

we choose

<

disjoint

consists of those

members of

that are divisible

by

2*.

and C where B Hence B(ri) ^ 2~kn

for

all n.

Now C consists
let

of the numbers

<j>(m)

C* denote the set ofm for which

of ^4 that are not divisible by 2*. We </>(m) e C. If # 1? # 2 qr are the distinct

, 9

prime factors ofm, then

which shows that 2 fore, if m e C* then

7

"- 1
1

<j>(m) since all

<k and

but one of the q t must be odd. There

say.

Hence
and

if <f>(m)

and

</>(m)

then

m^

72/cfc

and hence

C(/z)

C*(nfck ). But
factors,

now

the elements of the set

C* have k
is

so by
fc

Theorem

11.8

we

see that there

or fewer distinct prime an integer such that

C*(m)/m < sck/2 for m ;> N. Therefore C(n) <i C*(/2/c ) < and B(ri) ^ 2- < en/2. Finally we have ^(n) = B(ri) n ^ c^TV", and this implies 6 (A) = 0.
fc

ew/2 if w

c^,
if

+ C() <

ns

PROBLEMS
1.

Let k be a fixed integer. Prove that d(A)

if

every integer in

has the

form

where the p i are any primes,

1,2,
2. If
-,,5.

/:,

5-

is

arbitrary,

and a t

2 for

a sequence of integers ^4 = {# J has the property that converges, 0. Prove that the converse is false. prove that d(A) 3. Assuming the proposition that the set of primes (q& of the form 4n + 3 has 1 the property that 2?? diverges, prove that the set of integers each of which is a sum of two squares has density zero. as representable

2^

250

The Density of Sequences

of Integers

11.4

Schnirelmann Density and the

ocp

Theorem
set

Definition 11.2
integers
is

The Schnirelmann density d(A) of a

of non-negative

d(A)

inf

where A(n)

is

the

number ofpositive

integers

^n

in the set

A.

Comparing this with Definition 11.1 we immediately d^A) ^ 1. Schnirelmann density differs from asymptotic density
see that
sensitive to the first terms in the sequence.

d(A)

in that

it is

Indeed

if

then d(A)

0, if

that d^A) is unchanged if the then d(A) i, whereas it is easy to see 1 if and numbers 1 or 2 are removed from or adjoined to A. Also, d(A)

<

considering sets A consisting only of positive can contain 1 1.2 is worded in such a way that Definition integers. However, is not counted by A(ri). the number that noted be it should but 0,

only Until

if

contains
-

all

the positive integers.

now we have been

Definition 11.3

Assume

that

A and
of

e B. The sum
the

A +B
b where

of

the sets

and

is

the collection

of

all integers

form a

aeA

and

beB.
Note that

<=

A +

B,

<=

A +

=L Then by Theorem 5.6 we see that S + The sum A + B has not been defined unless e A and e B. We will assume that is in both A and B in the rest of this chapter. However, the sum could be defined for all A and B as the sum of the sets obtained from A and B by adjoining the number to each. This is equivalent to defining the

the set of squares 0,

1, 4, 9,

B. As an example let us take S to be and 7 the set of all non-negative integers.

S+S+S

sum

The

as the collection {a, b, a b} with a e A, b G B. result that is proved in the remainder of this section

is

the

oc/?

theorem

of H. B. Mann, which was conjectured about 1931, with proofs attempted and B subsequently by many mathematicians. The theorem states that if A the are if and are sets of non-negative integers, each containing 0, a, ft, y oc In ther min then A A Schnirelmann densities of B, B, (1, ft). y 1 1 oc in which case y words y : a ft ft unless
9

+
,

>

which we

prove a somewhat stronger result, Theorem 11.15, from deduce the oc/? theorem. We start by considering any positive and two sets A and B^ of non-negative integers not exceeding g. integer g We assume throughout that A and B l are such sets and that belongs to both A! and B Denoting A l + B by Cl5 we observe that C I may have
Actually

we

will

will

11.4

Schnirelmann Density and the

a/?

Theorem

251
6,

elements

>g even though A

and B^ do not.

We also assume that for some

2,
-

0<0<1,
(11.14)

A^m) + B^m) ^

6m,

m=

1,

,g.

OUT

W
way
the

idea is to first replace A l that (11.14) holds for A 2

and B^ by two new and 2 that C2 =

,

sets,
^4 2

A 2 and
2

in such a

+B c C

l5

and that

<

J?i&).

Lemma 11.10 Let A and B^ satisfy (11.14). IfB1 <t Al9 then there exist sets A 2 andB 2 with C2 = A 2 + B 2 such that C2 c Cl9 B2 (g) < B^g) and A 2 (m) +

m=

1,2,

g.

Proof.

We
r

Define

B =B n

merely

shift to

all

elements of
B'',

/4i,

A2 = Al u

B2 =
.

that are not already in

A.
.

5', where by

we mean

complement of v4 1} now the set of all non-negative integers not in A : Thus belongs to both A 2 and j9 2 Then A 2 (m) = ^(/H) + ^'(m) and
2

(m)

for

= Bi(m) m = I, 2,
and 6 E
AI

B'(m)so we have ^ 2 (m)

,

ae
y4 2

y4 2

B2
b,

U
a

.5',

we
a

write h

.B 2 (m) 6m ^(/M) ^(w) consider h 6 h a b C with Then 2 g. any Noting that B 2 is contained in both A and B and that c ^^ In the first case we can have either a 6 ^ x or a 6

Now

+ =

'

EA l9 b e B

in the

second case h

=
c:

+
.

a,b e
Since

A ly

B^;

hence in both cases we have h e

that

Q. Thus we have
is

C2

Ct

it is

obvious

^2 (^) <

B^g), the lemma

proved.

We

will get a similar result for the case

<=^

A ly

but

it is

a b

little

more

complicated. b in B!. Then

We

assume B^g)
if

>
A

0,

which implies that there

is

some
is

integer

>

let

not in AI. There may a denote the smallest a e

the largest integer in A l9 the sum a be other pairs a e A l9 b e B l such that a

is

certainly

+
B

A^ We

such that there

is

for which

b $A!. Since

^ ^ we
c:

see that a Q
results.

0.

Before defining

A 2 and ^2

we

will obtain

two preliminary

11.11 Let A l and B: satisfy B <= A l and B^(g) 0. Let a be defined as above. Suppose that there are integers b and z such that b G B l and z aQ b b, we have <^g. Then for each a e A such that 1

Lemma

>

<

^z

^a ^z

e A I9 and

(11.15)

A^^A^ + A^z-b).
We
have a

Proof.

<

and a

^ z ^ g,

hence a

b E

Al

is minimal. Now there are A^z b) positive integers a belonging b also z a the corresponding a a each to A l with such b, and to ^ <s b a b b satisfies to Furthermore each such < z, and belongs

because a

+ + ^

hence ^(z)

^(i)

A^(z

b),

and we have

(11.15).

252

The Density
11.12

of Sequences of Integers

Lemma

Let

and
is

satisfy (11.14),

B1 c A

19

and B^g)

a Q as before. If there
Proof. Since

an integer y

^g

such that

A(y}

<

0. Define aQ then y By,

>

>

such that A^z) < 6z. Then y^z^l. have we B^z) > 0, and hence there is a b e B A^z) + B(z) ^ such that < b ^ z <; g. If z <; a Q9 we would have z-a <&^z<g-, - b). Now and we could apply Lemma 11.11 to get A : (z) ^ A^b) + A^z since 1) + 1 1) + 1 ^ 0(* b eB l c X 19 so we have A^b) = ^(6 - 1 < z. Also, ^(z - 6) ^ 6(2? 6), and we are led to the contradiction - i) = 0(z - 1) + 1 0*. Therefore we have - + 1 +

Let

be the

least integer

6z

Z>

AI(Z)
2

^
,

B(b

1)

6(2
.

>

and hence y
11.13

>

Lemma

Let

A : and Bl

satisfy
f

B c

the set of all b e B I such that a + b $ A l9 such that b e B and a + b < g. Finally integers a + b ad J?2 (g) < ^i(g). = B 1 C^ ~B' Then C2 c: and B2
.

owd ^(g) > 0. Let and let A' denote the

let

denote

set

A2

=A u

of aU A!

Proof.

Note that
,

A 2 and
ly

/zeC2

= a 4- b, aeA a + beC^ since aeA be and we have A = a + * +

then h

so that the s_um C2 is well defined. If then h If 1 C\B'. l \j A', some for b a *i e 5/ i B^. If a e 4', then a we would otherwise since 6 e *i- Here a o

2,

beB

ae^,

^
.

have b e

'.

Since b 1 e

J? l9

we

again have h e

Cx

Finally

B 2 (g) < B
:

B(g),

since

the definition of a ensures that B'(g)

>

0.

Lemma

11.14

For

A l9 B19 A* B2

as in

Lemma

11.13, if A^

satisfy (11.14)

(11.16)

A 2 (m)
From
the

+B

(m)

^
,

6m

for

m=

2,--,

Proof.

way

A', B',

A 2 B2

were chosen we have

,4'(
^[ 2

m = B'(m- a ), = ^f^rn) + (/w) + B 2 (m)

)
,

B!(/M)

(B'(m)

B'(m

- a^,
)-

for

_ jj'( w
S^(m)

m=
-

1, 2,
Q ).

g.

Therefore (11.16) holds for

g-

Consider any

Bi(m

fl

smallest element of BI
(11.17)

m ^ for which w ) ~ 5/ ( w - a o) > 0, and we let b a <b ^m. Therefore such that m
5/ (

m for which B'(m) = a Then J5'(m) > B'(m

all
Q

denote the

A 2 (m) +

Bt(m)

^ = Ai(m) + = Xx(i) 4-

BI(W

fl )

^(60

1).

11.4

Schnirelmann Density and the

oc

Theorem
11.11 with b

253

Now m
and
z

a < b ^ m ^ g, = m to get

so

we can apply Lemma

We

also have

<

# so

Lemma
6
)

11.12 shows that

A(m Thus we can reduce

(11.17) to

>

fl(w

60).

But b
(11.14)

A 2 (m) + B2 (m) ^ ^(i ) + 6(m - 6 ) + J?^ - 1). eJ?! c ^ so we have ^(60) = ^(^ - 1) + 1. Using
we have,

this

and

^
Theorem 11.15

6(60

1)

0(m

.For tf/ry positive : and integer g let <; Let to both sets non-negative integers g. belong

B1 denote fixed sets of A and BI9 and write Cx

for

AI +

Si-

#/br ^om^

6 such that

0< ^

1,

+
then

J^C/H)

em,

m=

1, 2,

Q
If

^
^(g)

then B^ consists of the single integer 0, Cx A l9 and 5x (g) %. We prove the theorem for general Ci(g) ^i(g) ^i(g) sets by mathematical induction. 1 and that the theorem is true Suppose k for all A 19 B! with B^g) < k. If A^m) Om for 1 B^m) 2,
0,

k, then Lemma 11.10 or Lemmas 11.13 and 11.14 supply us ^i(g) with sets A 2 B 2 such that B 2 (g) < k, C2 c Cl9 and ^ 2 (m) 52 (/w) 6/w for m 1, 2, ,g. Therefore, by our induction hypothesis, we have

and

if

m=
+

g,

>

Qsfe)

= ^ Og,

which implies QCg)

The
oc/S

^
a,

Og.

Theorem 11.16
A, B,
Proof.

theorem. Let

A
y

and
(1,

integers, each containing 0,

A+B

respectively.

and let Then y

/?,

dlewo/e

r/ze

be any sets of non-negative Schnirelmann densities of

/S).

min

Let and 5 X consist of the elements of A and J5, respectively, that do not exceed g, an arbitrary positive integer. Then A^m) ^ aw and ^i(m) ^ = min (1, a + /?), the conditions of for m = 1, 2, /?/ g. If we take 6 Theorem 1 1.15 are satisfied and we conclude that C^g) ^ 6g. Since Q(g) ^ Og for every positive integer g, we have y ^ 6 = min (1, a + /?).
,

254

The Density

of Sequences of Integers

PROBLEMS
1.

What

is

the Schnirelmann density of the set of positive

odd

integers?

set set

of positive even integers ? The set of positive integers of positive integers s 1 (mod rri) ?

(mod

3) ?

The The

2.

Prove that the analogue of Theorem 11.1 for Schnirelmann density, namely, d(A) = inf w/tfn , is false. 3. Prove that the analogue of Theorem 11.16 for asymptotic density is false. even integers, and consider A + A. Suggestion: take A as the set of all positive 4. Prove that if d(A) = a, then A(ri) ^ aw for every positive integer n. Prove
that the analogue of this for asymptotic density is false. 5. Establish that Theorem 11.16 does not imply Theorem 11.15 by considering

the sets

A =

{0, 1, 2, 4, 6, 8, 10,

},

11.16 asserts that the density of

A +B

B = {0, is >

2, 4, 6, 8, 10,

}.

Theorem

whereas Theorem 11.15 says

much more.
6. Exhibit 7.

> a + ft - a/3. B). Prove that y d(B\ y d(A 8. Consider a set A with positive Schnirelmann density. Prove that for some

two sets A and B such that d(A) = d(B) = Q,d(A+B) = l. For any two sets A and B of non-negative integers, write a = d(A),

/?

positive integer n

nA
where / is the set of with n summands.

=
all

(n

l)A

(n

2)A

/,

non-negative integers, and

nA=A+A+''*+A

NOTES ON CHAPTER

11

For further reading on the subject of this chapter, see the book by Halberstam and Roth listed in the General References on page 269 and the following:
E. Artin and P. Scherk,

"On

the

sums of two

sets

of integers," Ann. Math. (2) 44,

/.

138-142 (1943). F. J. Dyson, "A theorem on the densities of sets of integers," 20, 8-14 (1945).

London Math. Soc.

sets

H. B. Mann,

"A proof of the

fundamental theorem on the density of sums of

of positive integers," Ann. Math. (2) 43, 523-527 (1942).

Miscellaneous Problems

Several of these problems are not easy,

and so can be regarded by the

of the

reader as "research" problems rather than routine exercises.

1.

Let

Ln

denote the least

common

multiple

[1, 2,

ri\

first

positive integers. Prove that

^ L~

converges, where the

sum

is

taken

over
2.

all

positive integers n.

Given any integers a, b, c, and any prime p not a divisor of ab, prove + by 2 = c (mod p) is solvable. 3. If g is a divisor of each of ab, cd, and ac + bd, prove that it also divides ac and bd, where a, b, c, d are integers. 4. Let p and q be twin primes, i.e., primes satisfying q = p + 2. Prove 2 that there is an integer a such that/? (a q) if and only if there is an 2 a unsolved problem to is famous that b such (There p). q (b integer
that ax 2
| \

prove that the number of pairs of twin primes is infinite. What is known is that the sum of the reciprocals of all twin primes is, if not a finite sum, with Theorem certainly a convergent series; this result can be contrasted 8.3. A proof of this result can be found in Chapter 15 of the book by
General References on page 269.) has exactly 2 solutions if p is an odd = 2 and prime and k is a positive integer, but exactly 4 solutions if p x 2 == 1 of 1 the solutions m > k g: 3. Hence prove that for any integer and determine its order. (mod m) form a multiplicative group, 6. For any positive integers a, b, and n, prove that if 72 is a divisor of n n b )/(a a n _ frn then w i s a divisor of (a b). Suggestion: Write g

Hans Rademacher
Prove that x 2

listed in the
1

5.

(mod p

for g.c.d. (a

and prove that g 2 (a - b ). 1 1. For any positive integers n and k prove that Sf-i/" is not an (This is a generalization of Problem 31 of Section 1.3.) 8. Find all positive integers n such that <f>(ri) n.

b, n)

inte g er

255

256
9.
if

Miscellaneous Problems

For any positive integer n prove that and only if n = 1 or n is a prime.

<f>(ri)

o(n)

^ 2, with equality

denote the number of positive integers a ^ n such that (mod ) for some integer x > 1 Prove that p(ri) is a multiplicative function. (This function is akin to the Euler ^-function which can be
10. Let p(ri)

ax

defined in an analogous

ax
11.

way

but with the congruence replaced by

(mod

).)

if

Prove that an odd integer n ^ 1 is a prime if and only if it is not expressible as a sum of three or more consecutive positive integers. n 12. Let q = 4 + 1 where n is a positive integer. Prove that q is a prime ~ 1)/2 =

and only

if

3 (<?

-1 (mod
if

q).

13.

Prove that n

is

a prime

except the first and last 14. Let d be the greatest

except the d p if n
first
is

and only if are divisible by

all
.

the coefficients of (x

+ +

y)
n

common
where n

divisor of the coefficients of (x

y)

and

last,

a power of a prime
is

any positive integer >1. Prove that and that d = 1 otherwise. p

is
9

if (n,k) 1. \K] 16. If a positive integer n is written to base b prove that the digit in the yth position, counting from the right hand end of the digital formulation,

15.

Prove that n

a divisor of the binomial coefficient

(,

is

[nib*-

b[n/b>].

Given any prime p and positive integers m and r with m > r, say that the triple/?, m, r has property Q if the following condition holds: r are written to base p there is at least one digit a in the if r and m
17.

expression for

such that

if

is

the corresponding digit in

then
I 1

a
if

+ b^p.
and only
r,

Prove that
if

is

a divisor of the binomial coefficient

holds. (The property

Q amounts

to saying that

if r

and

m
is

written to base/?, are added as in elementary arithmetic, there at least one "carry" in the addition.)

18. Evaluate the integral JJJJJJ [x

z]

dx dy dz where the square

brackets denote the greatest integer function. Generalize to ^-dimensions, with an -fold integral.
19.
K/rc

Prove that of the two equations

/+] =

[/i

TZ

2],

[Vn
the
first

+ -Vn+l]=

_
[jfi]

[Vn

3/n

_
+
2],

[x] is

holds for every positive integer n, but the second does not, where the greatest integer notation.
integers.

20.

Let/ and k be positive

Prove that

[(]+

)a]

[(/+

A:)fl

[joe]

[kai

kft]

Miscellaneous Problems

257
ft

for

all real

numbers a and

if

to take

0^oc<l,0^/3<l,

and only if j = k. Suggestion: It suffices and g.c.d. (;, &) = 1. (This generalizes

Problem 16 of 4.1.) 21. Prove that among any ten consecutive

is

positive integers at least one to the of the nine. (Remark: if the "ten" other relatively prime product in this statement is replaced by "," the result is true for every positive

integer n

16, but false for n > 16. This is not easy to prove; cf. Evans, Amer. Math. Monthly, 76, 48-49 (1969).) 22. Prove that the sum of the first n natural numbers is a perfect square

Ronald

J.

for infinitely many values of n. 2n 1 is a 23. Prove that 2 2 perfect square for infinitely integer values n.

many

24. If

that

n, prove any [n/2] -f 1 integers are selected from 1, 2, 3, two of them are relatively prime. Establish that if only [n[2] integers are selected, it does not follow that two of them must be relatively prime. 25. Given any set of n positive integers, prove that there is a non-empty subset whose sum is a multiple of n. 26. Let n and k be positive integers with k < n and (k, n) = 1. Prove that if k distinct integers are selected at random from 1, 2, n, the n that their is divisible is sum l/n. by probability 27. Let/(#) and g(x) be polynomials in Z[x], i.e., polynomials with
, *

integral coefficients. Suppose that g(m) \f(m) for infinitely many positive integers m. Prove that g(x) \f(x) in Q[x], that is, there exists a quotient

= g(x) q(x). polynomial q(x) with rational coefficients such that f(x) to 9.1 Theorem After get polynomials q(x) and applying Suggestion: a suitable in <2M> multiply by r(x) positive integer k so that kq(x) and and use the fact that g(m) > kr(m) for have kr(x) integer coefficients, m. (Remark: The example g(x) = 2x + 2, sufficiently large integers
=
xz
1

f(x)

with

odd shows

that q(x) need not have integral

in

coefficients.)

28. Let/(#)

and g(x) be primitive nonconstant polynomials

Z[x] such

1 for divisor (f(m),g(m)') infinitely many that to show such polynomials an Construct m. example positive integers 1 in the exist with polynomial sense. (/(#), g(aO)

that the greatest

common

>

g.c.d.

29.
(

For any odd prime p except

4. i).
is

Prove that p
that if
1

1/'(0).

3 define f(x) (x 2) T)(x alternative formulation of this

An
'

problem

+
2
1

p-

r
I

is

added to give the

a. rational fraction a[b, then/? nonconstant Given 30. polynomial f(x) with integer coefficients, any are there that (mod/>) infinitely many primes p such that/(#) prove

is

solvable. Suggestion: If there

were only

finitely

many

such primes/?,

258
let

Miscellaneous Problems

be their product, define X Q

= P n/(0),

and examine /0

with n

large.

an increasing sequence of positive integers a n_-^\a n = as n with positive natural density, prove that lim (a n
31. If

A =

{%,

2,

3,

} is

be an infinite sequence of disjoint sets of positive with respectively. Prove that integers, asymptotic densities ft, ft, /? 3
,

tends to infinity. 32. Let 5 l5 B 2 5 3

2 ft
#3
33.

converges, and
.

that

<5

X (5)

^ 2 ft

where

is

the set

^U

.Bg

(This

is

a generalization of Problem 10 in Section

ju,

11.1.)

where ^ is the Moebius function, r is the number of divisors function, and the equation contains both kinds
r,

Prove that r 2 *

=^

of multiplication of arithmetic functions given in Section 4,4, so that r 2 (n) means r(n) r(). Suggestion: The product of multiplicative
functions
is

multiplicative for both ordinary products

and

Dirichlet

products. Cf. Problem for the source.

34. Let

E 2235, Amer. Math. Monthly, 78, 406, 407

(1971)

a minimal set
class

n be a given positive integer. What is the number of elements in S of distinct integers having the property that every residue

modulo n occurs at least once among the sums of the non-empty subsets of S? (For example, if n = 6 the answer is 3 because {1, 3, 5} is a minimal set, with every residue class modulo 6 occurring among

+ 5,1 + 3 + 5.) Prove that no n points with rational coordinates (#, y) can be chosen in the Euclidean plane to form the vertices of a regular polygon
1,3,5,1+3,1 + 5,3
35.
4. Suggestions: If n with n sides, except in the case n 3, the area of such a triangle can be shown to be rational by the use of one standard elementary formula, but irrational by another. For values of n other

than 3, 4, or 6, a similar contradiction can be obtained by applying the law of cosines to a triangle formed by two adjacent vertices and the center of the polygon.

Special Topics

With the miscellaneous

but in
filling in

sufficient detail so that the

topics below, only outlines of proofs are given reader should find no great difficulty

arising solution.

the gaps. Thus these sketches may be regarded as elaborate problems from extensions of the theory, with detailed suggestions for their

Periodic Decimals

We now

sketch the proofs of two basic results about periodic or repeating decimals to base ten, namely decimals of the sort 0.06272727 which is
,

written 0.0627 where the bar indicates that the digital pair 27 is repeated indefinitely. The least number of digits in the repeating block is called the

period of the repeating decimal, where the word "least" must be included so
as to avoid writing 0.0627 in the form 0.062727, and so interpreting the period here as 4 when of course it is 2. terminating decimal is one with a finite number of nonzero digits following the decimal point, such as 0.25. Any terminating decimal can be written as an infinite periodic decimal with

an endless succession of nines, as for example 0.25 = 0.249. The first basic result is that any rational number is expressible as a terminating decimal or an infinite periodic decimal, and conversely. The proof
the decimal ex straightforward. First, given any rational number a/6, to create a division with b into obtained a, continuing by dividing pansion is the decimal expansion. If the remainder is zero at some stage, the decimal
is

expansion terminates; remainders are 1, 2,

if

not, the remainder must repeat, since the possible , b 1, and so the periodic decimal is formed.
is

Conversely, any terminating decimal

obviously expressible as a rational

259

260

Special Topics

number.
part, as
(1)

Any

infinite periodic

decimal can be written,

if

we omit

the integer

Denoting
It will

this

number by

/3

we note

that 10 r+s/3

10 r/?

is

an integer, and the

result follows.

minimal length of the nonrepeating part and

repeating part.

be convenient to write every decimal expansion (1) with r as the s the minimal length of the Thus for 0.263 the values of r and s are 2 and 1 not 3 and 1
,

as suggested by 0.2633, and not 3 and 2 as suggested by 0.26333. The second result relates the length of the period to the rational number. and 1, because the confine attention to rational numbers between

We

is apparent; for example the decimal expansion to the right of the decimal point. Furthermore same of 1/6 and 13/6 are the numbers we set aside all rational c\d with denominators d divisible by no numbers have terminating decimal rational such and 2 than 5; primes other numbers rational the are and (in lowest terms) whose only expansions,

extension to

all

other cases

is as follows. Let c/d be a rational such that some prime p d with and < < d, (c, d) 1, decimal has the 5. expansion (1), where r and s are the If c/d p ^ 2, p 7^ minimal lengths of the nonrepeating and the repeating parts, then r and s are

decimal expansions terminate. The result

c

number with

the least integers such that

S

(2)

d\

10*-(10

1),

r^O,
d

s^l.
9

2*5 p k where (k 10) Another way of stating this is 1, with to 10 which the 5is and max the conclusion that r (a, /S), belongs exponent I0 r )c/d is an integer modulo k. To prove this result we note that (10 r+s
to write

because of

(1),

and so d is a divisor of 10 r (10s

\

1).

To prove
1)

that r

and

s are
r

the least values satisfying (2), suppose that d 10*00* r t and 1 s. Then it must be established that j as 10*00*- l)(d so that

with
t

^ ^

and

<j ^

s.

Define q

10*00*- l)c/d=cq
is

an

integer.

Thus

10'"+* (c/d)

and

10*(c/rf)

have the same decimal part

r Substituting the decimal expansion (1) for c\d here, we conclude that/ and t s because r was the minimal length of the nonrepeating part and s is

the minimal length of the repeating part of the decimal expansion (1).

Unit Fractions

unit fraction, or Egyptian fraction,

is

positive integer. Every positive rational

one of the form l/, where n is a number is expressible as a sum of a

Special Topics finite

261
unit fractions.

number of distinct

We now

sketch a proof of this result,

beginning with rationals <1. Consider the rational number a\b with < a b and (a, b) 1. To prove that alb is expressible as a sum of distinct unit fractions we use induction on a. If a 1 the result is obvious.

<

Next suppose the proposition holds for all fractions with numerator a 1 or less. Now dividing b + a by a to get a quotient q and a remainder r, we note that r 7* because (a, 6) = 1. Hence b + a = aq + r with < r < a, and a/b = l/# + (a The numerator last fraction has <#, r)/&#. positive and so can be expressed as a sum of distinct unit fractions, each with
denominator larger than q because (a r)fbq < Ijq. Next if a/b is any rational > 1 define the integer n by
,

--

23
_
i

.-

-rr

-x

23
i
.,

nn+1
i

so that

c ^ ^ < < ~~

where
1

c -

= db
is

23

...__.
n

By the first part of the proof c\d, if not 0, unit fractions each with denominator >w

expressible as a
1.

sum of distinct

The Equation x~ n
For any

y~

= z~ n
n

z~ has a solution in x~ y~ positive integer n the equation n n n z x This does. is easy to establish if if and y only positive integers n n n to is because the latter equation (xy)~ , and (yz)~~ (xz)~ equivalent n n n the former is equivalent to (yz) (xy) (xz) to characterize all solutions in positive 2 difficult it not n and is 1 For

integers. If a solution has (x, y, z)

>

divisor to get a primitive solution,

is

i.e.,

we can remove the greatest common 1. This condition one with (x y,^)
9

now assumed, and we

|

equivalent to yz

yz.

Write a
if

+ = (x, z),
xz

begin with x~i

xy.

+
it

From
(y, z),

this

z~l by noting that follows that z\xy y\ xz,

y-

it is

and

and c

(x, y)

and

it

follows that

(a,6)=(M)=(c,fl)=l.
the is a prime that divides x and z, then /? a but p j y. Then of exact the same that z x and conditions p power xy imply yz divisibility that divides x also divides z. This holds for the other pairs a?, 2/ and z, y, and it 1 1 z- 1 ab. Substituting these in oryfollows that x be, z ac, y

Also,

we

arrive at a

= +

and so #

a (a

b),

i(a

ft),

+ =

0fc is

the

solution of the equation.

262

Special Topics
in positive integers

Every solution

ofx-

+ y-1 =

z' 1 with
(a, b)

given by taking arbitrary positive integers a, b with x a2 b2 06. ab, y ab, z

(x, y, z)
1

is

and

=
j

writing

As an immediate consequence of
x -i

__ 3-1 y-i

this we note that for n p OS itive integers, the result

all

solutions of

(a?,

z)

(y, z)

==

(a, y)

>

holds.
2 2 2 Turning to the case or + y~ = z~ it follows from the results above that 2 any primitive solution of this equation in positive integers has z = ab, x 2 = a(a + b), y 2 = b(a + b) for some positive a and b with
,

(a, 6)

1.

Hence

a, b,

with

(c,

d)

1.

Also c 2

and a + b are perfect squares, say a = c 2 b = d 2 + d 2 is a perfect square, so by Theorem 5.1 there
,

integers

are integers r and s such that

(*)

r2

-s

2
,

d=2rs,

>s>

0,

(r, j)

1,

+ =
5-

(mod

2).

By

and x

assigning rf the value Irs we are taking, in effect, y to be the even value the odd, it being clear that x and y must have opposite
in positive integers
2 of x~

Every solution

y~

parity.

z~ 2 with

(x, y, z)

and y even

is

given by

= r4 -

s*,

20

),

2rs(r

-s)
2

where r and s are arbitrary integers subject only

(*) above.

to the conditions

imposed

in

Gauss' Generalization of Fermat's

First establish that ifp is a is divisible

Theorem
is

(i)

then a m *
result of

any prime and a

if

any integer not

,

by/ m is divisible by/-1

divisible by/?,

(ii)

Hence prove the

is

Gauss that

if

a and n are any positive integers, then n

a divisor of

d\n

a prime, this is the theorem of Fermat. Suggestions: Prove the property k separately for each prime factor p ofn. Ifp does not divide a use the property (i), noting that the sum in (ii) can be separated into a collection of pairs of terms as in (i).
is

If n

Special Topics

263

A
The

Primitive Root

mod p

by Group Theory

existence of a primitive root modulo p, a prime, can be approached by group theory, to give an alternative proof of the first assertion in Theorem
2.25.

By showing
1

without difficulty that

the existence of one primitive root g, we can conclude r g is a primitive root for integral values of r from 1 to
if (/%/?

p
<j)(p

if

and only

1)

1,

and hence

it

follows that there are

1) primitive roots

modulo p.
is

The group
there
is

theoretic language for the existence of a primitive root

that

I in the an element g of order p multiplicative group modulo p. This can also be stated in the form that the multiplicative group modulo p is cyclic, and this can be proved as follows. ord a the order of any element a of a finite abelian group. (i) Denote by r If a = e for some integer r, where e is the identity element of the group then d r is divisible by ord a. (ii) If ord a = h and d h then ord (a ) = h/d. (iii) If ord a = A, ord b = k and (h, k) = 1 then ord (db) = hk. (iv) Let h be the maximum order of any element of a finite abelian group, say ord b = A. Then h if a is any element, a = e because h is divisible by ord a by the following that argument. Write k for ord a. If k -f h then there is a prime p such a and r for k IP*- Then ord ^ = ^ Write a * and fc > = P P" that = we see so are orders ord 6^ by (iii) relatively prime, h/p*. These ord (a r b^) = hp a ~^ > h, thus contradicting the maximal property of A.
\

||

II

(v)

Consider

If h denotes the

now the multiplicative group (1, 2, -,/> maximum order of any element then x ==
71

1}
1

modulo

p.

(mod p)

for

every element in the group, by

having

solutions,
1

Hence h

=p

(iv). But by Theorem 2.20 this congruence, 1. must have degree at least p 1, so h^.p -1 = x in the 1 because a? group (mod/?) for every element
31

by Fermat's theorem.

The Group

of Rational Points on the Unit Circle

2 Let x and y be rational numbers satisfying x

+ y* =
is

1,

so that (x 9 y)

is

rational point on the unit circle. (Thus (x, not the greatest common divisor symbol.) (i)

y) here

a pair of coordinates, Prove that these points form a

group

under the multiplication

where this multiplication has been suggested by the product of the complex numbers + iy l and x 2 + iy2 (ii) Prove that if < y l < y^ ^ 1, then a

264
rational

Special Topics

number

u/v can be chosen so that

Vi

^ <

u*

u 2__ +

-v* =
v*
2

< TTT +
r*
1

y*

Prove that the elements (x, y) in the group G are dense on the unit circle (meaning that between any two points of the group there lies another) by 2 2 2 2 2 2 taking x = 2uv/(u + v ), y = (w )/(w + y ) in part (ii) above.
(iii)
t;

The Day of the Week from the Date

Any date, such N, M, C, Y as

as

December

follows. Let

N = 25.

Let

M=

10 for December,

M be the number of the month beginning with March, so that M = for March, M = for January, M = 12 for
1

N be

25, 1984, can be separated into integer parts the number of the day in the month, so
1 1

February. (The peculiarity of starting with March is needed because the extra leap year day is added at the end of February.) Let C be the hundreds in the year, and Y the rest, so that C = 19 and Y = 84 for 1984. If ^denotes the day of the week, where d = for Sunday, d = 1 for Monday,

= J=
d

6 for Saturday, then

TV

[2.6M

0.2]

+ Y+

[Y/4]

[C/4]

- 2C =

(1

L)[M/11] (mod

7),

where L

for a leap year and L =

25, 1984

the case of

December

we

for a nonleap year. have L 1 and so

For example,

in

s=

25

[25.8]

84

[21]

[19/4]

38

2[10/11] == 2

(mod

7),

so Christmas day in 1984 falls on a Tuesday. This formula holds for any date after 1582, following the adoption of the Gregorian calendar at that time. The leap years are those divisible by 4,

except the years divisible by 100 which are leap years only if divisible by 400. For example, 1984, 2000, 2004, 2400 are leap years, but 1900, 1901, 2100, 2401 are not.
Verify the correctness of the formula by establishing (i) that if it is correct for any date then it is also correct for the date of the next succeeding day,

and

(ii)

that

it is

correct for one particular day selected

from the current

calendar.

Some Number
The purpose here
is

Theoretic Determinants
about determinants of order n j position (intersection of zth row and y'th
results

to establish
f,

some

such that the element in the

Special Topics

265
related to the g.c.d.
(/,/).

column)

is

some expression

For example,

in

Theorem 2 below

we

evaluate such a determinant with

of notation we position. For brevity theof in technically correct /((,/)) to denote the value of the place /(r,y) x state a succession of results with for function f(x) g.c.d. (z,y)-

the i,j shall write the functional notation

itself in (z",y)

We

sketches of proofs.

Theorem

1.

positive integers

For any arithmetic function m by

d\m

2^
'

all /(ra), let g(ni) be defined for

'

-^-w

d\m

square

these two equations being equivalent by Theorems 4.7 and 4.8. Let matrix of order n with the element f(i,j) in the z'th row, yth

M be

the

column

position.

Then the determinant of

is

given by
n

Proof.

otherwise. Let A be the square 1 if Define <% j i, and a tj matrix of order n with a tj in the z, j position, and B the matrix of order n with in the i-j position. Denoting the transpose of A by A*, we see that lit
\

has the element in the

z ,

j position

fc=l

Jc\(i,j)

the last

two formulations being readily

calculated. Also

by evaluating the

determinants

we

get

3=1

and the theorem follows because

Corollary.

M = BA* implies det M = det B

andf(p)

det A.

Iff is

totally multiplicative

0/br every prime p, then

Proof.

We note that

266

Special Topics
2.

Theorem
the

Let

be the matrix of order n with the element g.c.d.

(z,y) in

Then i,j position.

This follows from the Corollary to Theorem

by taking f(m)

m.
[i,j]

Theorem

3.

Let

be the matrix of order n with the element l.c.m.

in the i,j position.

Then

Proof.

Define /(m)

1/ra,

and note that

Then use
that a

the Corollary to

Theorem

1,

and

l/(z,y)

[i,j]/ij,

and the

fact

common multiplier of the elements of a row or column of a determinant

4.
is

can be factored out.

Theorem

Let
the

where r(m)

number of positive

be the matrix of order n with r(i,j) in the divisors ofm. Then det 3

i,j

position,

1.

This follows from

Theorem

with g(m)

1.

Theorem
where a

5.

Let

M M

be the matrix of order n with

is

the

sum of divisors function. Then det

cr(z,y) in the i,j position,

= n\.

This follows from

Theorem

with g(m)

= m.
ju,(i,j)

Theorem 6. Let Then for n = 1, 2,

3,

be the matrix of order n with 7 we have

,

in the

ij position.

det
respectively,

M
5

1,

-2,

4, 4,

-8, -32, 64

and det
this

ifn

8.

To prove

multiplicative

we use Theorem 1 with f(m) Theorem 4.12, and by

p(m), so that g(m)

is

(p)=-2,
Next we generalize Theorem

(/)=!,
2.

(/*)

if

r>3.

Theorem

7.

Le

i,j position, where a

is

6e /Ae matrix of order n with the element any real number. Then

(/,/)*

n the

This can be proved by taking /(m)

= ma

in the Corollary to

Theorem

1.

Special Topics

267
7 an analogous result can be found at once for the case

From Theorem
with
ft

by

j] in the i,j position, where ft is any real number. This can be done p foi (ij)^ ft in Theorem 7, next substituting [i 9 j]*li*j replacing a by
[i
9

in the determinant

and then factoring out the i p and/* multipliers from the denominators of the elements in the rows and columns. We omit the
B9

details.

Finally,

we

use

Theorem

in the case

where /(m)

is

defined as the largest

square-free divisor of
(3)

m, so that
g(P)

/(m)
8.

Il>>

/>-!,

S(/)

if

r>l.

Theorem

Let

be the matrix of order n withf(i,j) in the ij position,

with fas defined in (3) above. Then det and det 7 ifn ;> 4.

l,2ifn= 1,2,3

respectively,

Gaussian Integers as

Sums

of Squares

Let a and b be rational integers. Prove that the Gaussian integer a + 2bi is expressible as a sum of two squares of Gaussian integers if and only if not both a/2 and b are odd integers, by the following steps. First if a/2 and b are

odd
a

integers

we look

at the possibility that a

(u

+
u*

2bi

is
2

sum of two
v*,

squares,
uv.

2bi

(r

+ si)* +

w)

r*

-s ,

= rs +

The impossibility in integers of the last two equations here can be shown by a simple examination of the evenness and oddness of r s, u, v. Conversely, assume that a/2 and b are not both odd integers. Then we seek Gaussian integers f r\ such that 2 + rf = a -f 26f by factoring f 2 + rf
,

into (f

of a

??0(f 2bi:
(i)
(ii)
(iii)

"~~

^0

an<i writing assorted possible corresponding factorings

f f

+ + +

?]i=a
ijz

ijz

= =

Jfl

+ 2bi, 1-971= 1; + W, f - v = 2; -v= f (j* + W)(l + 0,

1

i.

If a

is

odd,

(i)

can serve because the equations in

(i)

can be solved for Gaussian

a
is

(a integers f and 97, for example f the cases where a/2 and b are both even

l)/2

hi. If

even,

we

separate

on the one hand, and where aj2 and b are of opposite parity on the other. If a [2 and b are even integers, (ii) can be used because these equations can be solved for Gaussian integers f and if a/2 and b are integers of opposite parity, then (iii) will do, r\. Finally, again yielding Gaussian integers f and r\. The above proof was given by L. J. Mordell, Math. Magazine, 40, 209
(1967).

268

Special Topics

Unique Factorization

in

Gaussian Integers
1

An alternative proof of the case m =

of Theorem 9.27 can be constructed

by analogy with the second proof given of Theorem 1.16, the unique
factorization theorem, as follows. First, given

any two nonzero Gaussian

/?

integers a and ft, establish that some associate /?' of/?, perhaps be chosen so that the triangle formed by the three points a,

itself,

can

/?',

in the

complex plane has an angle N(P) ^ N(oc) then N(& - /?')

^ <

ir/4 at the origin 0.

Hence prove
this result,

that if

N(<x).

With the use of

unique

factorization of integers in Q(f) can be proved by presuming that if unique factorization fails, it fails for some integer with a least norm. This integer then plays the role of n in the second proof of Theorem 1.16.

The

Eisenstein Irreducibility Criterion

let

Let/? be a prime, and

f(x)

+
.

atf

+ a^ +

a nx

be a polynomial with integer coefficients such that p is a divisor of all co 2 efficients a so a If also a * then -r n n except p /? /(a) is irreducible in that
,

it

cannot be factored into two polynomials with integer coefficients, excluding of course the trivial case where one factor is a constant. By Theorem 9.7 it

follows that /(a;) cannot be factored into two polynomials with rational
coefficients, excluding again trivial cases.

To prove

this result,

suppose on the contrary that/()

= g(x)h(x)
l,

where

g (x)

V
2,
3=0
v

'*>

+
a

n,

1,

j=0

where the

coefficients are integers.

Now
|

=
,

b c

and aQ
.

is

divisible

but not by /, so we may presume that/? and/? #! so/? b Similarly/? a 2

.
|

6 but/? f c Next implies/? A 2 and by induction

|

% = Vi + Vo
we establish

by p

that/?

is

divisor of every coefficient of g(x), the last step being

P ar>
|

P\ (Vr
is

+ Vr-l +

'

'

'

b r CQ )

SO

/?

br

It

follows that/?

also a divisor of every coefficient off(x).

General References

J.

W.

S. Cassels,

An

Introduction to Diophantine Approximation,

to

Cambridge

Tract 45, 1957.

K. Chandrasekharan, Introduction

Analytic

Number Theory, New York,

Springer- Verlag, 1968. L. E. Dickson, History of the Theory of Numbers, Washington, Carnegie Institution of Washington, 1919; reprinted, New York, Chelsea, 1950.
L. E. Dickson, Introduction to the Theory of of Chicago Press, 1929.
L. E. Dickson,

Numbers, Chicago, University

Modern Elementary Theory of Numbers, Chicago, University of Chicago Press, 1939. Emil Grosswald, Topics from the Theory of Numbers, New York, Macmillan,
1966.

H. Halberstam and K. F. Roth, Sequences,

1966.

vol. 1,

Oxford, Clarendon Press,

G. H. Hardy and E. M. Wright, An Introduction to the Theory of Numbers, 4th ed., Oxford, Clarendon Press, 1960. B. W. Jones, The Arithmetic Theory of Quadratic Forms, Carus Monograph 10, New York, John Wiley and Sons, 1950.

D. H. Lehmer, Guide

to Tables in the

Theory of Numbers, Washington,

Bulletin, National Research Council,

No.

105, 1941.

W.
L.

J.

LeVeque, Topics Addison- Wesley, 1956.

in

Number Theory,

vols. I

and

II,

Reading, Mass.,

Mordell, Diophantine Equations, New York, Academic Press, 1969. Ivan Niven, Irrational Numbers, Carus Monograph 11, New York, John Wiley and Sons, 1956. O. Ore, Number Theory and its History, New York, McGraw-Hill, 1949. Hans Rademacher, Lectures on Elementary Number Theory, New York,
J.

Blaisdell Publishing

Company,

1964.

269

270

General References

Daniel Shanks, Solved and Unsolved Problems in Number Theory, Washington,

D.C,

Spartan, 1962.

W.

Sierpinski, Elementary Theory of Numbers, New York, Hafner, 1964. Harold M. Stark, An Introduction to Number Theory, Chicago, Markham,

1970.
J.

V. Uspensky and

I.

M. H. Heaslet, Elementary Number Theory, New York, McGraw-Hill, 1939. M. Vinogradov, Elements of Number Theory, translation of 5th Russian
edition,

New

York, Dover, 1954.

Answers

Section
1.

1.2, p.

2.
3.

g =
(a)

(a) 77, 17,

(b)
a;

1,

(c) 1,

(d)

1.

71,

= 9,2/= -11, <W)z = 7,y=8,

z
(a)

y=

-36.
x x

(b)
(e)

= 3l,y = 44, = \,y= 1, z =

(c)x

= 3,y=-2.

-1.

4.
5.

3374

(6) 3660.

128.
6, 10, 15.

7.

17.

l;n(n
a; b.
a;

l).

18.

25. 27.

will do. + 5, y = 95 - 100, = 2, a 10, i = 100 is a solution in positive integers. All solutions are given = 10; 100, 50; a = 20, b = 100; a = 10, b = by = a = 20; with all arrangements of signs. There are 16 50, 6 =

= =

lOOn

<z

Z>

solutions in
28.

all.

10, b

100, c
all

10, 20, 50, or 100;

= 20,

50, c
all.

10, 20,

50, or 100;

and

permutations of these, 36 answers in

Section
1.

1.3, p.

15

a,-

<; 0j for
z

every;
3

=
3
.

1, 2,

r; a,

or

&=

for every;.

2.

3; 7.

16. 17. 18.

,p*;

P ,p ,p ;p*,P
3
|

P
2

,p.
<x
|

3 ;

a,.;

a,

ft; a,

ft; for all;,

<; <

r in

each part.
271

272
24.

Answers

Counterexamples for
(1) a (8) a

false statements are

(10)^
25. 39.

(13) a All ft not of the form

a, a,
,

= 2, 6 = 6, c = 10. = 8, c = 4. = 5,^ = 2,4=1,^ = = 2, 6 = 5.

p
,

3.

a or a, a,

a,

where /> is an odd prime. 2a or fl, 2a, 3<2.

Section
1.

2.1, p.

25
92.

7,24,41, 58,75,
1, 5, 7, 11

2.
3.

0, 18, 36, 3, 21, 39, 6, 24, 42, 9, 27, 45, 12, 30, 48, 15, 33.

4.
5.

y x

==

10.

=5 m=

(mod 12); 1, 7, 11, 13, (mod 2); z = 1 (mod 6). (mod 12).
1,

17, 19, 23,

29 (mod 30).

2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12.

11.
13. 14.

s 3
a

#/)= = 5.
= = =
9
2,

1,

1,2,2,4,2,6,4,6,4,10,4.

1,9,3,81,243,27.

11;.

16.

3,4,

5,

6,7, 8,9,

10.

3
25.
26. 27.
35. 36.
1.

6,4,3,9,2,8,7,5,10.

6.

0, 1.

Primitive solutions with a

^b^ca.rea

=
\a\

Solutions such that (a,b,c)

any positive integer.

1, c

a a
tf <2

= 1, c = or 2; = = 1, 6 = 2, c = 3; =b= with any c > 0; = &= c with any c > 2; =z2,b= 2n + 1, c = 2n + with any >
Z?

\b\

are

1.

Section
4.

2.2, p.

29
(3

3(3

1)(3

2)

+m-

1)

(mod m).

Section
1.

2.3, p. 31

(a) (b)
(c)

no solution no solution

3=

-82 (mod

400).

Answers
2.
3.

273
(6) 0,
(c) 5.

4.
5. 6.

= 106. 23 + 30;. x = 33 (mod - 2. 60;

x

(a) 5,

84).

7 '
12.

21*
105'

42 (mod 125).

Section

2.4, p.

36

1.

1,2.
960.

2.
3.

2640.
1920.

4.
5.

6720.

10.

n odd. n even.

11.
12. 13. 15.

k
,

1,2,

-will do.

35, 39, 45, 52, 56, 70, 72, 78, 84, 90.

3,1,2,4.

Section
x x

2.5, p.

38

1.

= = = =

1,2, 6
1
,

(mod (mod 5)

9)

ss
2.
3.

1,6, 11,28,33,38
solution.
1, 3,

(mod

45).

No
x
x

(mod

4.

1, 3, 5,

503). 14, 16, 27, 122, 133, 135

(mod

143).

Section

2.6, p.

42

6.

No
*
x

solution.

7.
8.

= =

(mod

5 3 ).

9.

z x

7, 15, 16, 24 3 == 15 (mod3 ).

(mod

36).

10.

No

solution.

11.

23

3 (mod 7 ).

274

Answers

Section
1.

2.7, p.

45

(a)
(6)
(V)

x5 z2
a;

+ x* + 5 = + 3x - 2 EE - a* - 4x +
2.8, p.
2

(mod 7), (mod 7), 3 == (mod

7).

Section
1.

46

(a)
(6)
(c)

(d)

= 2 (mod 5) = 4 (mod 7) (x + I) = g mo d 11) (4^ + 7)2 (2x + I) = 5 (mod 13).

(4x+
1)
2

Section
1 1

2.9, p.

50

Z., Z,,

22322
J,
5.

Z,, Z,.

2.
3.

4.

4.

1,3, 6,3,6,2.

1,10,5,5,5,10.
7.
8.

/>

1.

0.

(a) 4,

9.

z2
a;

= =

(4) 0,
a;

1,

(c) 4,

2,

z2

4,

9,

a;

13, x*

15,

16

(mod

17).

Section 2.10,
I-

p.

54

(),

(*), (/), (*), (0-

3.

Answers

275

Section
6. 8.

2.11, p.

60

23.

24.

(a)

and

(c)

are integral domains; (b)

is

an integral domain

if

and only

if

m is

a prime.

Section

3.1, p.

66

1.

1,

-2,

3,

-7,

0.

4.

'

2\

n)
(

= ~ if =
+1,
(

y
/

2\

= ~ ij

y=
/

2\

+i>

111/
77
5.

\13/
T;

-i,

) \17/

+1,

=+L
1

=+L

71

= -!

6.

2 (mod 11), 5 (mod 11), x = (mod 11), x = = x 3 (mod 11). a; 4 (mod 11), 2 2 (mod IP), 27 (mod II 2 ), x = x i (mod II ), a; = 2 2 x = II 3 x 48 (mod II ), ). (mod 8 (mod 4 (mod 13), 4, 1, 2, 3, 1, 1,2, 4 (mod 7), 13 7, 4, 3, 1, 7, 9, (mod 29), 6, 5, 1, 4, 16 12, (mod 37). 10, 11,

= = = =

17),

9,

7.
8.

(d)2,
(a) 2,

(h)2.
(b) 0,
(c) 4,

(d) 0,

(e) 2,

(/) 0.

276

Answers

Section

3.2, p.

70

4.

(i), (c), (d), (e,) (/).

^)
11) 2277
6.

= +1
= _j

'

=
(^)
/

+ls

" ~ lf
(low)

m=_

/j3\ . _ L
\10097

\2297

Yes.

7.
8-

9.

10.

= 2,/> = 13, and/7 = 1, 3, 4, 9, p = 1, 3, 9, 13 (mod 40). Odd primes p = 2, 3 (mod 5). p=l,3 (mod 8).
/?

10, 12

(mod

13).

Section

3.3, p.

74

L
2.
3.

= _ 1?
(6).
(c).
/?

(51) \717

. _ lf

fZl) Y737

. +1

10. 11. 12.

= 2 and^ =
a

2and/?

for/?
-

(mod 4). (mod 4) and a

ax

2 ai/?2 2

-^
82

or \,ps

= =

1,2,3,
1

(mod

4),

af

1, 2, 3,

Section
1.

4.1, p.

529,263,263,263,87.
24.
(a) all
re

2.
3.

such that x

[#]

<

J,

(6) all a,
(c)
(rf)

all all
all

integers,

(e)

x such that x a; such that 1

[x] ^ ^ < 10/9.

,

a;

Answers
5.

277

(a)

_
-<
if

pis odd,

I (I?] -[?])'"*
Oifj>

2.

Section
1.

4.2, p.

87

7.

2.
3.

12.

2,1,12,24.
6.

4.

10.
13.

/() = -n
x

will do.

=p

71

will do,

where j?

is

any prime.

16.

6,28,496.

Section
1.

4.3, p.

90

n
1.

33 will do.

3.

7.

'

'

'

A-

Section
2.

4.4, p.

96

No;

no.

278

Answers

Section

4.5, p.

98

xm

b n/ *xQ
.

ifn

is

even,
is

&! if

odd.

3.

0*1,1,2/3,
(A)

11.

- [-71/3]
?

5, 8,
if

is

13,21,34. odd.
5

12.

72=

2
1

f(n) = /(5 + 3)
ample.
13.

4 4
3)

7
12

10

16

20

25
1

f(5

=/(8)
n~2

/(2)
(1 if
/2

=
is
71

16

15

3,

for

ex

xn

= + 2n~ 1

(-2)

even,
if
is

[1

-1

odd.

Section
2.

5.2, p.

101

*=1 +7f,y= -1 +

10*.

Section
1.

5.3, p.

103

( fl )

x
a;

(A)

= =

8,
6,

y y

= =

4;
3.

a?

5,

= 9; = 2, y =
a;

14.

(d)

= 2,y =

5.

(e) no solution. (/) no solution. (g) no solution.

10.

ab.

Section
x

5.4, p.

105

(a)
(i)
(c)

= + w, y = 2w + 2 = w 20, = 10 + w y = 2u + 3v,z = u 2v, x = + 2t, y = 6 + 15* - 20, 2 = -2 1

3z;,

a?

5f

v,

Answers
(d)
(e)

279

x x

= =

t,y

2t,y

I5t

-2v,z

-11

3t

50, z

= -5f + v = -11 + 3t +
9

6v.

(/) no

solution.

Section
1.

5.5, p.

107

x
a;

= = = =

3,

y #
?/

=15,
5,

= 4, 2 = 5, = 8, 3=17,
=12,z=13,

x
a;
a; rr re

= = = =

4,
8,

s * 3
3.

((3)

a;

= 20, z = 29, = 7, 24, 2 = 25. # = 3/:, = 4k, z = = 4k, y = 3k, z =

21, y
?/
?/

=12,
20,

*/

24,

= 3, 2 = 5, = 15, z = 17, = 5, 2=13, y = 21, 2 = 29, = 7, z = 25.

2/
*/
/

5&,
5k.

(Z?)

none.
0,

5.

1,3 (mod

4).

Section
4.

5.7, p.

111

1,2,3,4,5,7,8,

10, 11, 13, 16, 19.

Section
1.

5.10, p.

117

#(100) #(101)
#(102)

= = =

12,
8,

P(100)
P(101) P(102)

0,

= = =

0,

2(100)
2(101)

2,
0,

2(102)

= 0, = 8, = 0.

Section
1.

5.13, p.

126
k

(a) all
(b) all

= 2 p^ = 0, 2,
e

p^
6

1,

e i even ifp t

(mod

4).

(mod

8).

(c) all integers. (J) all non-negative even integers.

2.

All perfect squares.

x
a? a?

= = =

4,
6,

y y y

8,

= = =

1,

5,

-5,

= = x =
3?
a;

4,
6,

2/

y -8, y

= = =

5,
5.

(Z>)

no

solution.

10.

1.

280

Answers

Section
2.
(a) x*
(b)

5.14, p.

132
2
,

+ +

xy xy

+ 3</ + 2y
2
.

(c)
(rf)

2x*
a;

2
,

+ xy + 6y + xy + 3y\

2
,

4.

a;

xy

5y

Section
6.

6.1, p.

137
will do.

= d=l,c =
7.1, p.

Section
1.

152
3/17

17/3
<2,

<5, 1, 2},

(0, 5, 1, 2), 8/1

(8).

3.

1,4}=

14/5,

(-3,

2, 12)

-63/25,

(0, 1, 1,

100}

101/201.

Section
1.

7.2, p.

153

for

The following conditions are necessary and sufficient. In case as = bj ^y ^ then n must be even. Otherwise define r as the least value
,
-

of j such that a3 n bj. In case r ar 6 r but for r odd, #r & r In case a n < Z? w but for n odd we require a n

1,

then for r even

72,

we

<

>

= then for even we require > + i n or a n = + 6 n with

1
,

require

Section

7.3, p.

157

2.
3.

+ V5)/2, (25 + V2, (b) (1 + (a)

(3
1

V3)/2,
'

(c)

+
a

Vs,
'

(rf)

- ^3.

4.

AJA^ = r

n'

an~ 1

'

a*'

ai>

a
\

*_

Section
1.

7.4, p.

159

V^ =

<1, 2, 2, 2,

>,

V2 - 1 =
>

(0, 2, 2, 2,

>,

= V2/2
V3

(0, 1, 2, 2,

<1, 1, 2, 1, 2, 1, 2,

>,

-^ V3

<0 ?

1, 1, 2, 1, 2, 1, 2,

->.

Answers

281

Section
1.

7.6, p.

166

1/1, 3/2 will do.

2.

3/1

,22/7 will do.

Section

7.7, p.

172

Section
1.

8.2, p.

184

log 9 (base 10).

Section
5.

8.3, p.

187

1,2,3,4,6,8,

12,

18,24,30.

Section
1.

9.2, p.

195
a;

x
7,

7, x*
1

^7,

- 7, x* - 3x*[2 + 3z/4 - 1, 4 - 4z + -v/2 + \/3~ are algebraic integers.

3

4x*

\6x

8.

Section

9.4, p.

201

3.

Yes; no, for example a

%(\

zX/3).

Section
6. 7.

9.5, p.

203

(1

7i)/5 will do.

The
.

suggestion also works in case

t , ^ t. u be handled by such numbers as:

m = 2. The other special cases can 27 + V3 9 + 4V2 ! + 4vCI^ rr

-- -,

10V5

'

282

Answers

Section

10.3, p.

227

2.

w
p(ri)

p(n)

= = = =

1,2, 3,4, 5,

6,

7,

8,

9,10,11,12.
18,
19,

1, 2, 3, 5, 7, 11,

15, 22, 30, 42, 56, 77.

17,

13,

14,

15,

16,

20.

101, 135, 176, 231, 297, 385, 490, 627.

Section
1.

10.4, p.

233

a(n)

CT(TZ)

= = = =

1,2,3,4,5,

6,7

8,

9,10,11,12.
28.

1, 3, 4, 7, 6, 12, 8, 15, 13, 18, 12,

13, 14, 15, 16, 17, 18, 19,20.

14, 24, 24, 31, 18, 39, 20, 42.

Section

10.6, p.

239

2.

p(35m

19)

(mod

35).

Section

11.1, p.

242

1.

(a) i, (6) i, (c)

(rf)

J,

0)

(/) 0, (g) 0, (A) 0,

(/)

0, (j) 0.

15.

Jr.

Section
2

11.2, p.

246

15/(27T

).

3.

(6) 8/T7

Answers

283

Section

11*4, p.

254

1.

1/2, 0, 1/3, 1/m.

Miscellaneous Problems,
5.

p.

255

has exactly b distinct odd prime factors, the order is 2 & , 2 &+1 or 2 &+2 according as m[4 is not an integer, is an odd integer, or is an even
If
,

integer.
8.

ii=l,2'or
l;(n-l)/2.
l

18.

34.

Index

Abelian group, 52 Additive group, 53

Algebraic integer, 193, 199

Congruence (continued) identical, 29

number of solutions,
prime modulus, 43

28, 44,

45

Algebraic number, 193 closure properties, 194

degree of, 193

field of,

ax=b (modm),
x 2 =-1 (mod

prime power modulus, 39 24, 29

p),

195

24

minimal equation, 193 Algorithm, 3, 4

Euclidean, 7
o/3

.+a n
(modm),
28, 37

=0

theorem, 250, 253

irrationals, 137,

x 2 =a (mod p), 49
Conjugate numbers, 198 Continued fraction, 1 50 convergents of, 156
finite,

Apostol, T. M., 240

Approximation by

159, 164 Arithmetic function, 84

152
158

254 Associate, 200

Aitin, E.,

infinite, 153,

periodic, 166

Asymptotic density, 241

Belonging to exponent, 46, 58 Bertrand's postulate, 1 85
Binary operation, 51 Binary quadratic form,
Bolker, E. D., 218
1

simple, 151

Convergent, 156
secondary, 163

Coprime, 6
Coverings of the real Cubic residue, 51
Cyclic group, 57
line,

147

23

Calendar problem, 264 Canonical form, factorization, 14

Cassels,
J.

Day of the week, 264

Decimals, periodic, 259
Definite form, 123 Degree, of a congruence, 28

W.

S.,

269

Chandrasekharan, S., 269 Chinese remainder theorem, 30

Closed under, 11, 51

Common

divisor,

4
8

of an algebraic number, 193 of a polynomial, 188 Density, 241

a/3

of polynomials, 189

Common multiple,

theorem, 250, 253 asymptotic, 241

natural,

Commutative group, 52 Complete residue system,

241

22,

54

Completely multiplicative function, 85 Composite number, 10 Congruence, 20

degree
of,

Schnirelmann, 250 sets of density zero, 246

of square-free integers, 244 Descent, proof by, 108

28

Determinants, 264
Dickson, L. E., 269

degree two, prime modulus, 46

285

286
Diophantine equations, 100 ax + by = c, 101 a l x +...+akx k = c, 104
l

Index
Functions,
<t>(n),

ir(x),

15,

178

22, 34, 89,

249

[x],

65, 78

/ft;, 85, 86,

243

aft/ .85, 87,

232

MftA 88, 244 pftl, 219, 226 Fundamental theorem of arithmetic, 13

+ cz 2 =0, 120 x a -d> 2 = 1,173 = z3 x3 ,213 x .i +y .i = z -i 262 2 2 2 x- +jr =z262
Gaussian integers, 203, 267, 268
primes, 212 reciprocity law, 6 7

+/

Gauss's generalization of Fermat's

theorem, 262

Dirichlet convolution, 91

lemma, 64
polynomial lemma, 191 Generating functions, 225
Generator, 57, 263

product, 91
Dirichlet's theorem, 18

Discriminant, 124
Divisibility, 2

Graph, 220
Greatest

of algebraic integers, 200 of polynomials, 189

Division algorithm, 3
Divisors, 3

common

divisor, 4,

265

Euclidean algorithm, 7 of polynomials, 190

Greatest integer function, 65, 78

sum

of,

85
J.,

Grosswald,

E.,

269

Dyson,

F.

254

Group, 52, 263

abelian,

52

Egyptian fraction, 260

Eisenstein irreducibility criterion, 268

commutative, 52 cyclic, 57
finite,

Empty sums and

products, 34
1

52 52

Equivalent quadratic forms,

Eratosthenes, sieve of, 17

27

generator, 57
infinite,

Euclidean algorithm, 7
quadratic
fields,

205 229

isomorphic, 53 order of, 52

order of an element, 57
Hagis, P., 77

Euclid's theorem, 14
Euler's formula, 226,

generalization of Format's theorem, 23,

58

^function, 22, 34, 89, 249

Evans, R.
J.,

Halberstam, H., 269

257

Hardy, G. H., 269 Heaslet, M. A., 270

Hunter,
in,
J.,

Factorial,

80

133

power of prime

80

Hurwitz' theorem, 139, 164

Factorization, unique, 11, 13, 205

Farey sequence, 134 Fermat number, 18 Format's last theorem, 108, 213 method of infinite descent, 108
theorem, 23,58 Fibonacci numbers, 98 Field, 58, 195 Form, 123

Idempotent, 6 1
Identical congruence, 29

Identity element, 51
Indefinite form, 1 23

Index,

49
continued fraction, 153, 158

Infinite

Infinite descent,

108

Formal power

223 Function, recurrence, 96

series,

Infinitude of primes, 14
Integer, 1

Index
Integer (continued}
algebraic, 193,

287
Notation, see Symbols

199

Number,
perfect,

algebraic,

193

Gaussian, 203
quadratic, 201
Integral domain, 61

87 prime, 1, 10

Number

theoretic function, 84

Inverse element,

52
146

Numerical function, 84
Order of an element, 5 7 Ore, (X, 269

Inversion formula, 88, 92, 95

Irrationality of e,
ofrr,

144
number, 142 of, 157

of trigonometric functions, 143

Irrational

expansion

Parity, 9

Irreducible polynomial, 190,

268

Partial quotient,

151
236, 238

Isomorphism, 53, 197

Jacobi's formula, 233

Partitions,

219

divisibility properties,

Jacobi symbol, 71
Jones, B. W., 269

equation, 172 Perfect number, 87

Pell's

Periodic continued fraction, 166

Periodic decimals,
8
Pollard, H.,

259
1

Least

common multiple,

218
96

Legendre's theorem, 1 20 Legendre symbol, 63

Polynomial congruence,
monic, 188 primitive, 191

Polynomials, irreducible, 190

Lehmer, D. H., 269 Lehmer, D. N., 62 Lemma of Gauss, 64

Polynomials over the rationals, 188

Positive form, 1 23

on polynomials, 191
Le Veque, W.
J.,

269

Power

residue,

46

Levinson, N., 19

Prasad, A. V., 149

Prime, relatively, 6

Magic square, 33 Maier, E. A., 148

Prime number theorem, 15, 178

Primes, 1,10,119, 178

Mann, H. B., 250, 254 Mann's theorem, 250, 253

Mathematical induction, 2 Minimal equation, 193
Modulus, 20 Moebius function, 88 inversion formula, 88

contained in factorial, 80
distribution of, 14, 15,

182
1

Gaussian, 212
in arithmetic progression,
infinite

number

of,.

14

in quadratic fields, 204, 208 Primitive polynomial, 191

Monic polynomial, 188

Mordell, L. Multiple, 3
J.,

root,

4 7, 58, 263

133, 267, 269

solution, 106, 113,

118

Pythagorean

triple,

107

Multiplication of functions, 91
Multiplicative function, 85

Quadratic

field,

201
1

group, 56

Quadratic forms,
equivalent, 127

23

nih power residues, 46

Natural density, 247

reduced, 129 Quadratic irrationals, 167

number,

1
1

Negative form,

23

Nonresidue, 63

expansion of, 167 Quadratic nonresidue, 63 reciprocity, 67, 73

residue,

Norm, 12, 202

63

288
Rademacher, R, 269 Ramanujan, S., 239
Rational approximation, 137, 159, 164 Rational integer, 193
Reciprocity, quadratic, 67, 73

Index
Symbols (continued)
a

= b (mod m),a^b (mod

\ 20

0(m), 22 II, 2, 34
p\n d\n

Recurrence function, 96

Reduced quadratic form,

residue system, 22, 56
Relatively prime, 6
in pairs, 6

29

ftp P
f

ll

\QI
[x],65,78
r(n\ 85 a(/t), 85
M(W), 88

Remainder theorem, 30 Representation by quadratic forms, 123 Residue, 22

quadratic, 63

-,127
.,
jc/>,

system, complete, 22, 54 reduced, 22, 56

Ring, 58

151

^0^1
t

>,

155

Robinson, A., 218 Roth, K. F,, 269

Scherk,

Q[x],Zlx] F[x],l*B
,

195

P.,

254

q(n),219
e,u, n, 240, 241

Schnireimann density, 250

Secondary convergent, 163 Shanks, D., 270

Sierpinski, W.,

Totally multiplicative function, 85

Totient, 22, 34

270
1

Sieve of Eratosthenes,

Simple continued fraction, 151 Skolem, Th., 133

Spitznagel, E.

Unique

factorization, 11, 13

in quadratic fields, 205,

268

L,

Jr.,

187

Unit, 200, 203

Square-free, 14
integers, density,

244

Unit fraction, 260 Universal form, 1 24

Stark, H. W.,

218,270

Uspensky,

J.

V.,

270
133

Subgroup, 60 Sum of five squares, 111

of four squares, 109 of fourth powers, 112 of two squares, 113, 267

Vandiver, H.

S.,
I.

Vinogradov,

M, 270

Waring's problem, 112

Symbols,

a\b,

a^b.a^-

\\

b, 3,

Wilson's theorem, 24
Wright, E. M., 269

(b,c),(b lt b 2t ... ,b n \4 a a ...,<z n ],8 [<z lf

,

a z,
ir(jc),

is
15,

178

Zero element, 58 form, 123

02 746