Number Theory UNSOLVED

CONTENTS
PAGE
PREFACE
.......................................
Chapter I
FROM PERFECT KGXIBERS TO T H E QUADRATIC
RECIPROCITY LAW
SECTION
1. Perfect Xumbcrs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2. Euclid . . . . . . . . . . . . ............................. . 4
3. Eulers Converse P r ................ ............... 8
SECOND EDITION
4. Euclids Algorithm . . . . . . . ............................... 8
5. Cataldi and Others. . . . . . ............................... 12
6. The Prime Kumber Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Copyright 0.1962. by Daniel Shanks
7. Two Useful Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Copyright 0. 1978. by Daniel Shanks 8. Fermat. and 0t.hcrs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
9. Eulers Generalization Promd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
10. Perfect Kunibers, I1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
11. Euler and dial.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Library of Congress Cataloging in Publication Data
12. Many Conjectures and their Interrelations. . . . . . . . . . . . . . . . . . . . 29
Shanks. Daniel . 13. Splitting tshe Primes into Equinumerous Classes . . . . . . . . . . . . . . . 31
Solved and unsolved problems in number theory. 14. Eulers Criterion Formulated . . . . . . ....................... 33
Bibliography: p. 15. Eulers Criterion Proved . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Includes index . 16. Wilsons Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
1. Numbers. Theory of . I . Title. 17. Gausss Criterion ................................... 38
[QA241.S44 19781 5E.7 77-13010 18. The Original Lcgendre Symbol . . ..................... 40
ISBN 0-8284-0297-3 19. The Reciprocity Law . . . . . . . . . . ..................... 42
20 . The Prime Divisors of n2 a . . . + ..................... 47
Chapter I1
Printed on long-life acid-free paper T H E C S D E R L T I S G STRUCTURE
21 . The Itesidue Classes as an Invention . . . . . . . . . . . . . . . . . . . . . . . . . 51
Printed in the United States of America 22 . The Residue Classes 3 s a Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
23 . The Residue Classcs as n Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
24. Quadratic Residues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
V
vi Solved and Unsolved Problems in N u m b e r Theory Contents vii
SECTION PAGE SECTION PAGE
25 . Is the Quadratic Recipro&y Law a Ilerp Thcoreni? . . . . . . . . . . . 6.2 61. Continued Fractions for fi. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
26 . Congruent.i d Equations with a Prime Modulus . . . . . . . . . . . . . . . . 66 62. From Archimedes to Lucas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
27 . Eulers 4 Eunct.ion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 63. The Lucas Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
28. Primitive Roots with a Prime i\Iodulus . ........... 71 64. A Probability Argument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
29 . mpas a Cyclic Group . . . . . . . . . . . . . . . . ........... 73 65. Fibonacci Xumbers and the Original Lucas Test . . . . . . . . . . . . . . 198
30. The Circular Parity Switch . . . . . . . . . . . ........... 76
31. Primitive Roots and Fermat Xumhcrs., ........... 78
32 . Artins Conjectures . . . . . . . . . . . . . . . . . . ........... 80 Appendix to Chapters 1-111
33. Questions Concerning Cycle Graphs . . . . ........... 83
34. Answers Concerning Cycle Graphs. . . . . ........... 92 SUPPLEMENTARY
COMMENTS.
THEOREMS. ...
AND EXERCISES . . . . . . 201
35. Factor Generators of 3% . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 Chapter IV
36. Primes in Some Arithmetic Progressions and a General Divisi-
bility Theorem . . . . . . . . . . ............................ 104 PROGRESS
37 . Scalar and Vect.or Indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 SECTION
38 . The Ot.her Residue Classes . . . . . . . ...................... 113 66. Chapter I Fifteen Years Later . . . . . . . . . . . . . . . . . . . . 217
39 . The Converse of Fermats Theore ................ 67. Artins Conjectures, I1 . . . . . . . . . . . . . . . . . . . . . . . . . 222
40 . Sufficient Coiiditiorls for Primality . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 68. Cycle Graphs and Related Topics . . . . . . . . . . . . . . . . . . .225
69. Pseudoprimes and Primality . . . . . . . . . . . . . . . . . . . . . . 226
Chapter 111 70. Fermats Last Theorem, I1 . . . . . . . . . . . . . . . . . . . . . 231
71. Binary Quadratic Forms with Negative Discriminants . . . . . . 233
PYTHAGOREAKISM AKD ITS MAXY COKSEQUESCES 72. Binary Quadratic Forms with Positive Discriminants . . . . . . .235
41. The Pythagoreans . . . . . . . . . . . . . . . . . . . ............ 121 73. Lucas and Pythagoras . . . . . . . . . . . . . . . . . . . . . . . . . 237
42 . The Pythagorean Theorein . . . . . . . . . . . ................ 123 74. The Progress Report Concluded . . . . . . . . . . . . . . . . . . . 238
43. The 4 2 and the Crisis . . . . . . . . . . . . . . . . . . . . . . . . . .
44 . The Effect upon Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
45 . The Case for Pythagoreanism . . . . . . . . . . . . . . . . . . . . . . . . .
46 . Three Greek Problems . . . . . .................. Appendix
47 . Three Theorems of Fermat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 .......................
ON FUNDAMENTALS
STATEMENT 239
48 . Fermats Last Theorem . . . . . . . ....................... 144 OF DEFINITIONS
TABLE ............................ 241
49 . The Easy Case and Infinite Desce ....................... 147
50. Gaussian Integers and Two Applications . ........... 149
...................................
REFERENCES 243
51. Algebraic Integers and Kummers Theore ........... 1.51 ......................................
INDEX 255
52. The Restricted Case, S Gcrmain, and Wieferich . . .
53. Eulers Conjecture . . ................................ 157
54 . Sum of Two Squares . . . . . . . . . . . . . . . . . . . . . . . ...... 159
55. A Generalization and Geometric Xumber Theory . . . . . . . . . . . . . . 161
56. A Generalization and Binary Quadratic Forms . . . . . ........ 165
57 . Some Applicat.ions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
58. The Significance of Fermats Equation . . . . . . . . . . . .
59. The Main Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
60. An Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 178
PREFACE TO THE SECOND EDITION
The Preface to the First Edition (1962) states that this is a rather
tightly organized presentation of elementary number theory and that
I number theory is very much a live subject. These two facts are in
I
conflict fifteen years later. Considerable updating is desirable a t many
places in the 1962 Gxt, but the needed insertions would call for drastic
surgery. This could easily damage the flow of ideas and the author was
reluctant to do that. Instead, the original text has been left as is, except
for typographical corrections, and a brief new chapter entitled Pro-
gress has been added. A new reader will read the book a t two
levels-as it was in 1962, and as things are today.
Of course, not all advances in number theory are discussed, only those
pertinent to the earlier text. Even then, the reader will be impressed
I with the changes that have occurred and will come to believe-if he did
not already know it-that number theory is very much a live subject.
The new chapter is rather different in style, since few topics are
developed a t much length. Frequently, it is extremely hrief and merely
gives references. The intent is not only to discuss the most important
changes in sufficient detail but also to be a useful guide to many other
topics. A propos this intended utility, one special feature: Developments
in the algorithmic and computational aspects of the subject have been
especially active. I t happens that the author was an editor of Muthe-
matics of C m p t a t i o n throughout this period, and so he was particu-
larly close to most of these developments. Many good students and
~
professionals hardly know this material at all. The author feels an

obligation to make it better known, and therefore there is frequent
emphasis on these aspects of the subject.
To compensate for the extreme brevity in some topics, numerous
references have been included to the authors own reviews on these
topics. They are intended especially for any reader who feels that he
must have a second helping. Many new references are listed, but the
I following economy has been adopted: if a paper has a good bibliogra-

phy, the author has usually refrained from citing the references con-
tained in that bibliography.
The author is grateful to friends who read some or all of the new
chapter. Especially useful comments have come from Paul Bateman,
Samuel Wagstaff, John Brillhart, and Lawrence Washington.
DANIELSHANKS
December 1977
ix
PREFACE T O THE FIRST EDITION
It may be thought- that the title of this book is not well chosen since
the book is, in fact, a rather tightly organized presentation of elementary
number theory, while the title may suggest a loosely organized collection
of problems. h-onetheless the nature of the exposition and the choice of
topics to be included or. omitted are such as to make the title appropriate.
Since a preface is the proper place for such discussion we wish to clarify
i
~
this matter here.
Much of elementary number theory arose out of the investigation of
three problems ; that of perfect numbers, that of periodic decimals, and
that of Pythagorean numbers. We have accordingly organized the book
into three long chapters. The result of such an organization is that motiva-
tion is stressed to a rather unusual degree. Theorems arise in response to
previously posed problems, and their proof is sometimes delayed until
an appropriate analysis can be developed. These theorems, then, or most
of them, are solved problems. Some other topics, which are often taken
up in elementary texts-and often dropped soon after-do not fit directly
I
into these main lines of development, and are postponed until Volume 11.
I
Since number theory is so extensive, some choice of topics is essential, and
while a common criterion used is the personal preferences or accomplish-
ments of an author, there is available this other procedure of following,
rather closely, a few main themes and postponing other topics until they
become necessary.
Historical discussion is, of course, natural in such a presentation. How-
ever, our primary interest is in the theorems, and their logical interrela-
tions, and not in the history per se. The aspect of the historical approach
which mainly concerns us is the determination of the problems which sug-
gested the theorems, and the study of which provided the concepts and the
techniques which were later used in their proof. I n most number theory
books residue classes are introduced prior to Fermats Theorem and the
Reciprocity Law. But this is not a t all the correct historical order. We have
here restored these topics to their historical order, and it seems to us that
this restoration presents matters in a more natural light.
The unsolved problems are the conjectures and the open questions-
we distinguish these two categories-and these problems are treated more
fully than is usually the caw. The conjectures, like the theorems, are in-
troduced a t the point at which they arise naturally, are numbered and
stated formally. Their significance, their interrelations, and the heuristic
xi
x ii Preface
If, as they say, we can succeed even partly in this direction we mill consider
evidence supporting them are often discussed. I t is well known that some
ourselves well rewarded.
unsolx ed prohlrms, c w h as Ecrmat~Last Thcorern and Riclmanns Hy-
The original presentation of this material was in a series of t w n t y public
pothesis, ha\ e t)ccn eiiormou4y fruitful in siiggcst ing ncw mathcnistical
lectures at the ?avid Taylor RIodel Basin in the Spring of 1961. Following
fields, and for this reason alone it is riot desirable to dismiqs conjectures
the precedent set there by Professor F. Rlurnaghan, the lectures were
without an adeyuatc dimission. I;urther, number theory is very much a
written, given, and distributed on a weekly schedule.
live subject, arid it seems desirable to emphasize this.
Finally, the author wishes to acknowledge, with thanks, the friendly
So much for the title. The hook is largely an exposition of known and
advice of many colleagues and correspondents who read some, or all of the
fundamental results, but we have included several original topics such as
notes. I n particular, helpful remarks were made by A. Sinkov and P.
cycle graphs and the circular parity switch. Another point which we might
Bateman, and the author learned of the Original Lcgcndrc Symbol in a
mention is a tcndeney here to analyze and mull over the proofs-to study
letter from D. H. Lehmer. But the author, as usual, must take responsi-
their strategy, their logical interrelations, thcir possible simplifications, etc.
lt happens that sueh considerations are of particular interest to the author, bility for any errors in fact, argument, emphasis, or presentation.
and there may be some readers for whom the theory of proof is as interest-
D.%;~IEL SHZXKS
May 1962
ing as the theory of numbers. Hovever, for all readers, such analyses of
the proofs should help to create a deeper understanding of the subject.
That is their main purpose. The historical introductions, especially to
Chapter 111, may be thought by some to be too long, or even inappro-
priate. We need not contest this, and if the reader finds them not to his
taste he may skip them without much loss.
The notes upon which this book was based were used as a test a t the
American University during the last year. A three hour first course in
number theory used the notes through Sect. 48, omitting the historical
Sects. 41-45. But this is quite a bit of material, and another lecturer may
prefer to proceed more slowly. A Fecond semester, which was partly lecture
and partly seminar, used the rest of the book and part of the forthcoming
Volume 11. This included a proof of the Prime Sumber Theorem and would
not be appropriate in a first course.
The exercises, with some exceptions, are an integral part of the book.
They sometimes lead to the next topic, or hint a t later developments, and
are often referred to in the text. X o t every reader, however, will wish to
work every exercise, and it should be stated that nhile some are very easy,
others arc not. The reader should not be discouraged if he cannot do them
all. We would ask, though, that he read them, even if he does not do them.
The hook was not written solely as a textbook, but was also meant for
the technical reader who wishcs to pursue the subject independently. It is
a somewhat surprising fact that although one never meets a mathematician
who will say that he doesnt know calculus, algebra, etc., it is quite common
to have one say that he doesnt know any number theory. Tct this is an
old, distinguished, and highly praised branch of mathematics, with con-
tributions on the highest levcl, Gauss, Euler, Lagrangc, Hilbcrt, etc. One
might hope to overcome this common situation by a presentation of the
subject with sufficient motivation, history, and logic to make it appealing.
CHAPTER I
Chapter I :: FROM PERFECT NUMBERS TO THE QUADRATIC

RECIPROCITY LAW
FROM PERFECT NUMBERS TO
Perfect Numbers
Euclid THE QUADRATIC RECIPROCITY
Euler's Converse Proved
Euclid's Algorithm
Cataldi and Others
Cataldi
LAW
The Prime Number Theorem
Useful Theorems
Two Useful
Fermat and Others
Generalization Proved
Euler's Generalization 1. PERFECT
NUMBERS
Perfect Numbers II I1 Many of the basic theorems of number theory-stem from two problems
and M31
Euler and M31 investigated by the Greeks-the problem of perfect numbers and that
of Pythagorean numbers. In this chapter we will examine the former,
Conjectures and
Many Conjectures and their Interrelations
and the many important concepts and theorems to which their investiga-
Splitting
Splitting the Primes into Equinumerous
Primes into Equinumerous Classes
Classes tion led. For example, the first extensive table of primes (by Cataldi)
Euler's Criterion
Euler's Criterion Formulated
Formulated and the very important Fermat Theorem were, as we shall see, both direct
Euler's Criterion Proved
Euler's Criterion Proved consequences of these investigations. Euclid's theorems on primes and
Theorem
Wilson's Theorem on the greatest common divisor, and Euler's theorems on quadratic resi-
Gauss's Criterion
Gauss's Criterion dues, may also have been such consequences but here the historical evidence
The Original
The Original Legendre
Legendre Symbol
Symbol is not conclusive. In Chapter 111we will take up the Pythagorean numbers
and their many historic consequences but for now we will confine ourselves
The Reciprocity
The Reciprocity Law
Law to perfect numbers.
The Prime
The Prime Divisors
Divisors of n"2 ++ aa
of n^2
Definition 1. A perfect number is equal to the sum of all its positive
Appendix to
Appendix to Chapters 1-111
Chapters I-III
divisors other than itself. (Euclid.)
EXAMPLE:
Since the positive divisors of 6 other than itself are 1, 2, and
3 and since
1 + 2 + 3 = 6,
6 is perfect.
The first four perfect numbers, which were known to the Greeks, are
PI = 6,
Pz 5 28,
P, = 496,
r.,= 8128.
1
2 Solved and Unsolved Problems in N u m b e r Theory F r o m Perfect N u m b e r s to the Quadratic Reciprocity Law 3
I n the Middle Ages it was asserted repeatedly that P , , the mth perfect the proof can be simplified. And, if we state that Theorem T is particularly
number, was always exactly i n digits long, and that the perfect numbers important, then we should explain why it is important, and how its funda-
alternately end in the digit 6 and the digit 8. Both assertions are false. I n mental role enters into the structure of the subsequent theorems.
fact there is no perfect number of 5 digits. The next perfect number is Before we prove Theorem 1, let us rewrite the first four perfects in
binary notation. Thus:
Pg = 33,550,336.
Decimal Binary
Again, while this number does end in 6, the next does not end in 8. It also
ends in 6 and is P1 6 110
Pg = 8,.589,869,056. PZ 28 11100
We must, therefore, a t least weaken these assertions, and we do so PI 496 111110000

as follows: The first me change to read Pq 8128 1111111000000
Conjecture 1. There are injbiitely m a n y perfect numbers. + + +
ow a tinary numher coiisisting of n 1's equals I 2 4 . . . 2n-1 = +
The second assertion we split into two distinct parts: 2" - 1. For example, 1 11 11 (binary) = 25 - 1 = 31 (decimal). Thus all
Open Question 1. B r e there a n y odd perfect numbers? of the above perfects are of the form
Theorem 1. Ecery even perfect number ends i n a 6 or an 8. 2n-1(2n - I ) ,
By a conjecture we mean a proposition that has not been proven, but e.g.7 496 = 16.31 = 24(25- 1).
which is favored by some serious evidence. For Conjecture 1, the evidence
is, in fact, not very compelling; we shall examine it later. But primarily we Three of the thirteen books of Euclid were devoted to number theory.
will be interested in the body of theory and technique that arose in the I n Book IX, Prop. 36, the final proposition in these three books, he proves,
attempt to settle the conjecture. in effect,
An open question is a problem where the evidence is not very convincing Theorem 2. T h e number 2n-1 (2" - 1) i s perfect i f 2" - 1 i s a primc.
one way or the other. Open Question 1has, in fact, been "conjectured" in both It appears that Euclid was the first to define a prime-and possibly
directions. Descartes could see no reason why there should not be an odd in this connection. A modern version is
perfect number. But none has ever been found, and there is no odd perfect
number less than a trillion, if any. Hardy and Wright said there probably Definition 2. If p is an integer, > 1, which is divisible only by f1 and
are no odd perfect numbers a t all-but gave no serious evidence to support by f p , it is called prime. An integer > 1, not a prime, is called composite.
their statement. *4bout 2,000 years after Euclid, Leonhard Euler proved a converse to
A theorem, of course, is something that has been proved. There are Theorem 2 :
important theorems and unimportant theorems. Theorem 1 is curious but - 1) with
Theorem 3. Ecery e m n perfect number i s of the f o r m 2rL-1(2rb
not important. As we proceed we will indicate which are the important 2" - 1 a prime.
theorems.
The distiiictioii between open question and conjecture is, it is true, We will make our proof of Theorem 1 depend upon this Theorem 3
somewhat subjective, and different mathematicians may form different (which will he proved later), and upon a simple theorem which we shall
judgments concerning a particular proposition. We trust that there will prove a t once :
he no similar ambiguity coiiceriiing the theorenis, and we shall prove many Theorem 4 (Cataldi-Fermat). If 2 n - 1 i s a prime, then n i s itse(f a
such propositions in the following pages. Further, in some instances, we primp.
shall not nierely prove the theorem but also discus the nature of the proof, PROOF.
We note that
its strategy, and its logical depeiitleiicc upon, or independence from, some
concept or some previous tlieorem. We shall sonietinies inquire whether an - 1 = ( a - 1)(an-' + an-2+ . . . + a + I ) .
4 Solved and Unsolved Problems in Number Theory From Perfect Numbers to the Quadratic Reciprocity Law 5
If n is not a prime, write it n = rs with r > 1 and s > 1. Then Euclid, recognizing that this needed proof, provided two fundamental
underlying theorems, Theorem 5 and Theorem 6 (below), and one
2" - 1 = (27)s - 1, fundamental algorithm.
and 2" - 1 is divisible by 2r - 1, which is > 1 since r > 1. Definition 3. If g is the greatest integer that divides both of two integers,
Assuming Theorem 3, we can now prove Theorem 1 . a and b, we call g their greatest common divisor, and write it
PROOF OF THEOREM 1 . If N is an even perfect number,
B = (a, b).
N = 2?'(2" - I)
I n particular, if
with p a prime. Every prime > 2 is of t,he form 41n +
1 or 4m +
3, since (a,b) = 1,
otherwise it would be divisible by 2. Assume the first case. Then
N = - 1)
24m(24m+l we say that a is prime to b.
EXAMPLES
:
= 16"(2.16" - 1) with m 2 1.
2 = (4, 14) 1 = (1, n )
But, by induction, it is clear that 16'" always ends in 6. Therefore 2.16" - 1
ends in 1 and N ends in 6 . Similarly, if p = 4m +
3, 3 = (3,9) 1 = ( n - 1, n)
N = 4.16"(8.16" - 1 ) 1 = (9,201 1 = (P, Q) (any two distinct primes)
and 4.16'" ends in 4, while 8.16" - 1 ends in 7. Thus N ends in 8. Finally Definition 4. If a divides b, we write
if p = 2, we have N = P1 = 6, and thus all even perfects must end in alb;
6 or 8.
if not we write
2. EUCLID
a@.
So far we have not given any insight into the reasons for 2"-'(2" - 1)
being perfect-if 2" - 1 is prime. Theorem 2 would be extremely simple EXAMPLE
:
were it not for a rather subtle point. Why should N = 2p-'(2p - 1) be 23 12047.
perfect? The following positive integers divide N :
Theorem 5 (Euclid). If g = ( a , b) there i s a linear combination of a and
1 and (2" - 1) b with integer coeficients m and n (positive, negative, or zero) such that
2 and 2(2" - 1) g = ma + nb.
2' and 22(2p- 1 )
Assuming this theorem, which will be proved later, we easily prove a
2?' and 2p-1(2p- 1 ) Corollary. If (a, c) = (b, c) = I , the2 (ab, c) = 1.
Thus 8, the bum of these divisors, including the last, 2"-'(2" - 1) = N , is PROOF.We have
equal to mla + nl c = 1 and m b + n2c = I ,
Z = (1 + + +
2 22 . . . +
2 9 1 +
(2" - I)].
and therefore, by multiplying,
Summing the geometric series we have
Z = (2" - I ) .2" = 2 N .
Mab Nc = 1 +
Therefore the sum of these divisors, but not counting N itself, is equal to
with M = mlmz and N = mln2a m2n1b + +
n1n2c. Then any common
divisor of ab and c must divide 1, and therefore (ab, c) = 1.
8 - N = N . Does this make N perfect? Kot quite. How do we know We also easily prove
there are no other positive divisors?
6 Solved and Unsolved Problems in Number Theorg From Perfect Numbers to the Quadratic Reciprocity Law 7
Theorem 6 (Euclid). If a, b, and c are integcrs such that
This generates Eq. (1). NOTV if thcre were a second represcntation, hy the
clab a i d (c, a ) = I, corollary of Theorem 6, each p , must equal some '1% , since p , [ N . Likcwise
lhen each q2 must equa1 some p , . Thercfore p , = qt and V L = n. If b, > a, ,
divfde p:& into Eqs. ( I ) and ( 2 ) . Then p , would divide the quotient in
clb.
Eq. ( 2 ) but not in Eq. ( I ) . This contradiction shons that a, = b , .
PROOF.
By Theorem 5,
Corollary. T h e only positive divisors of
'111c + na = 1. N = p;"' . . . p?
Therefore
are those of the f o r m
nzcb -l- nab = 6,
p;'p;' . , . p;; (3)
but since clab, ab = cd for some integer d. Thus
where
c(mb + nd) = 0,
or clb.
Corollary. I f a prime p dzvides a product of n nziiiibers,
pIa1a2 . . . a, ,
it niust divide at least one of them.
PROOF. If p i j a l , then (a1 , p ) = 1. If now, p+a2 , then we must have
pljalal, for, by the theorem, if plalar, then pja2.It follons that if p j a l ,
p+az , and p + a n , then p+alazaB. By induction, if p divided none of a's i t
could not divide their product.
Euclid did not give Theorem 7, the Fundaiuental Theorem of Arzthi?ietic,
and it is not necessary-in this generality-for Euclid's Theorem 2 . But
we do need it for Theorem 3.
Theorem 7. Every integer, > 1, has a unique jactorization into
primes, p , i n a standard form,
N = p;'p;z . . . P 2 , (1)
with a , > 0 and pl < p2 < . . . < p,, . That i s , if
N = &iqi2 . . . qnLbm (2)
r
f o r primes 41 < 42 < . . < qm and exponents b, > 0, then p ,

' = q, , v i = n, Theorem 2 Theorem 3
and a L = b, .
PROOF. First, N must have a t least one represcntatioii, Eq. ( 1 ) . Let a Theorem 7
be thc sinallest divisor of N which is > I . It must be a prime, >iiic.e if not, Theorem 4
a would hare a divisor > 1 and <a. This divisor, <a, ~ ~ o i i divide ld N
aiid this contradicts the dcfiiiition of a . Write a now as p , , and the quotient, Theorem 6
N / p l , as N1 . Repeat the process uith N 1 . The process must terminate,
since
Theorem 5
N > Ni > N , > ... > 1,
8 From Perfect Numbers to the Quadratic Reciprocity Law 9
Solved and (insolved Problems in Number Theory
They support the theorems which rest upon them. In general, the impor- a = qlal + a2
tant theorems will have many consequences, while Theorem 1, for in- a1 = q2a2+ a3
stance, has almost no consequence of significance.
The proofs of Theorems 3 and 5 will now be given. an-2 = qn-1G-1 + an
3. EULER'S PROVED
CONVERSE an-i = qnan.
PROOF 3 (by L. E. Dickson). Let N be an even perfect
OF THEOREM
This must occur, since a > al > a2 . . . > 0. Then the greatest common
number given by divisor, g = ( a , 6) , is given by
N = 2,-'F g = a,. (5)
where F is an odd number. Let 2 be the sum of the positive divisors of F. For, from the first equation, since gla and gib, we have g l a l . Then, from
The positive divisors of N include all these odd divisors and their doubles, the second, since g l a and g [ a l , we have g1a2 . By induction, gla, , and
their multiples of 4, . . . , their multiples of 2"-'. There are no other positive therefore
divisors by the corollary of Theorem 7. Since N is perfect we have (6)
9 6 a,.
N = 2"-'F = (1 + 2 + . . - + 2,-')2 - N But, conversely, since a, lanpl by the last equation, by working backwards
or through the equations we find that a,la,-2 , anlan-3, . . . , anla and
a,\b. Thus a, is a common divisor of a and b and
2N = 2F = (2" - 1 ) Z . a, 5 g (the greatest).
Therefore With E q . (6) we therefore obtain Eq. (5). Kow, from the next-to-last
Z = F + F/(2" - I), (4)
equation, a, is a linear combination, with integer coefficients, of a,-l and
an-2. Again working backwards we see that a, is a linear combination of
and since Z and F are integers, so must F / ( 2 " - 1) be an integer. Thus a,-i and an-i.-l for every i. Finally
(2" - 1 ) [ F g = a, = ma nb + (7)
and F / ( 2 " - 1) must be one of the divisors of F . Since Z is the sum of for some integers m and n. If, in Theorem 5, a and b are not both positive,
all the positive divisors of F , we see, from Eq. ( 4 ) , that there can only be one may work with their absolute values. This completes the proof of
two, namely F itself and F / ( 2 " - 1). But 1 is certainly a divisor of F. Theorem 5, and therefore also the proofs of Theorems 6, 7, 2 , 3, and 1.
Therefore F / ( 2 " - 1) must equal 1, F must equal 2" - 1, and 2" - 1 EXAMPLE:
Let g = (143, 221).
has no other positive divisors. That is, 2" - 1 is a prime.
Then 221 = + 78,
1.143
143 = 1.78 + 65,
4. EUCLID'S
ALGORITHM
PROOF
78 = 1.65 + 13,
OF THEOREM 5 (Euclid's Algorithm). To compute the greatest
common divisor of two positive integers a and b, Euclid proceeds as follows.
Without loss of generality, let a 5 b and divide b by a : 65 = 5.13,
b = qoa + a1
and g = 13. Now
13 = 78 - 1 . 6 5
with a positive quotient q o , and a remainder al where 0 5 a1 < a. If
= 2 . 7 8 - 1.143
al f 0, divide a by al and continue the process until some remainder,
a,+l , equals 0. = 2.221 - 3.143.
The reader will note that in the foregoing proof we have tacitly assumed it is clear from his books, prefers the nonconstructive. This type of proof
several elementary properties of the integers which we have not stated is often shorter, more elegant. The const.ructiue proof, on the other
explicitly-for example, that alb and alc implies alb +c ; that a > 0, hand, is practical-that is, it givcs solutions. It is also richer, that is,
+
and bia implies b 5 a, and that the al in b = qoa al exists and is unique. it develops more than is (immediately) needed. The mathematiciarl who
This latter is called the Division Algorithm. For a statement concerning prefers the nonconstructive will give another name to this richness-he
these fundamentals see the Statement on page 217. will say (rightly) that it is irrelevant.
It should be made clear that the m and n in Eq. (7) are by no means Which type of proof has the greatest clarity? That depends on the
unique. I n fact, for every k we also have algorithm devised for the constructive proof. A compact algorithm will
often cast light on the subject. But a cumbersome one may obscure it.
I n the present instance it must be stated that Euclids Algorithm is
remarkably simple and efficient. Is it not amazing that we find the greatest
Theorem 5 is so fundamental (really more so than that which bears the common divisor of a and b without factoring either number?
name, Theorem 7), that it Twill be useful to list here a number of comments. As to the richness of Euclids Algorithm, we will give many instances
Most of these are not immediately pertinent to our present problem-that below, ( e ) , ( f ) , (g) , and Theorem 10.
of perfect numbers-and the reader may wish to skip to Sect. 5 . Finally it should be not8ed that some mathematicialls regard noncon-
(a) The number g = ( a , 6 ) is not only a maximum in the additive sense, structive proofs as objectionable on logical grounds.
that is, d 5 g for every common divisor d, but it is also a maximum in the (d) Another point of logical interest is this. Theorem 7 is primarily
multiplicative sense in that for every d multiplicative in statement. In fact, if we delete the standard form,
pl < p z < . . . , which we can do with no real loss, it appears to be purely
dlg. (8)
multiplicative (in statemcnt) . Yet the proof, using Theorem 5, involves
This is clear, since dlu and dlb implies dlg by Eq. ( 7 ) . addition, also, since Theorem 5 involves addition. There are alternative
(b) The number g is also a minimum in both additive and multiplicative proofs of Theorem 7, not utilizing Theorem 5, but, without exception,
senses. For if addition int,rudes in each proof somewhere. Why is this? Is it because the
mla + nlb = h (9)
demonstration of even one representation in the form of Eq. (1) requires
the notion of the smallest divisor?
for a n y m1and nl, we have, by the same argument, When we come later to the topic of primitive roots, we will find another
glh. instance of an (almost) purely multiplicative theorem where addition
intrudes in the proof.
Then it is also clear that (e) Without any modification, Euclids Algorithm may also be used
g 5 every positive h. to find g(x) , the polynomial of greatest degree, hich divides two poly-
(11)
nomials, a ( x ) and b ( z ) .In particular, if a ( x ) is the derivative of b(x), g(z)
(c) This minimum property, ( 11), may be made the basis of an alterna- will contain all multiple roots of b(x) . Thus if
tive proof of Theorem ti, one which does not use Euclids Algorithm. The
most significant difference between that proof and the given one is that b(x) = 2 - 5xz + 7x - 3,
this alternative proof, a t least as usually given, is nonconstructive, while
Euclids proof is constructive. By this we mean that Euclid actually con-
and b(z) = a(z) = 3 2 - lox + 7,
structs values of nz and n which satisfy Eq. ( 7 ) , while the alternative then g(s) = Ul(Z) = -+(x - 1).
proves their existence, by showing that their nonexistence would lead to a
contradiction. We will find other instances, as we proceed, of analogous Therefore (x - 1)21b(x).
situations-both constructive and nonconstructive proofs of leading (f) Without elaboration at this time we note that the quotients, q L,
theorems. in the Algorithm may be used to expand the fraction a/D into a continued
Which type is preferable? That is somewhat a matter of taste. Landau, fraction.
12 Solved and Unsolved Problems in Number Theory From Perfect Numbers to the Quadratic Reciprocity L a w 13
Thus Any even perfect number is of the form
a
- _- -1 2P-l(2p - 1 )
1
b qo+-
q1 + q-2 + . . . -1,
1
with p a prime. If there were only a finite number of primes, then, of course,
Qn there would only be a finite number of even perfects. Euclid's last con-
and, specifically, tribution is
Theorem 8 (Euclid). There are injinitely m a n y primes.
PROOF.I f pl , p2 , . . . , p , are n primes (not necessarily consecutive),
then since
Similarly from (e) above, we have N = pip2 * . . p , +1

is divisible by none of these primes, any prime P,+~ which does divide N ,
3x2 - lox +
7 - 9 (and there must be such by Theorem 7 ) , is a prime not equal to any of
x3 - 5s' + 7~ - 3
-
(32 - 5) - 8 the others. Thus the set of primes is not finite.
(32 - 7) *
EXERCISE 2. ( A variation on Theorem 8 due to T. J. Stieltjes.) Let A
(g) Finally we wish to note that, conversely to Theorem 5, if
be the product of a n y r of the n primes in Theorem 8, with 1 5 r 5 n,
ma + nb = I and let B = plpz . . . p n / A .
then a is prime to b. But likewise m is prime to n and a and b play the Then A + B is prime to each of the n primes.
role of the coefficients in their linear combination. This reciprocal relation- EXAMPLE:
pl = 2, p , = 3, p , = 5. Then
ship between m and a , and between n and b, is the foundation of the so
called modulo multiplication groups which we will discuss later. 2.3.5 + 1, 2.3 + 5, 2.5 + 3, 3.5 +2
Now it is high time that we return to perfect numbers. are all prime to 2, 3 , and 5.
5. CATALDIAND OTHERS EXERCISE
3. Let A , = 2 and A , be defined recursively by
The first four perfect numbers are A,+1 = A: - A, + 1.
2(2* - l ) , Show that each A , is prime to every other A , . HINT:Show that
22(23- 11, An+l = A1A2 . . * A , +1
24(25 - I ) ,
and that what is really involved in Theorem 8 is not so much that the p's
26(27- 1). are primes, as that they are prime t o each other.
We raise again Conjecture 1. Are there infinitely many perfect numbers? EXERCISE
4. Similarly, show that all of the Fermat Numbers,
We know of no odd perfect number. Although we have not given him a
great deal of background so far, the reader may care to try his hand at:
F, = +
22m 1
EXERCISE
1. If any odd perfect number exists it must be of the form for m = 0, 1, 2, . . . , are prime to each other, since
D = (p)4"+1N2 F,+1 == FoFl . . . Fm + 2.
where p is a prime of the form 4m + 1, a 2 0, and N is some odd number Here, and throughout this book, 22mmeans 2(2m),not (2') = 2'" = 4".
not divisible by p . In particular, then, D cannot be of the form 4m +
3. EXERCISE 5. Show that either the A , of Exercise 3, or the F , of Exercise
(Descartes, Euler) . 4, may be used to give an alternative proof of Theorem 8.
I
14 Solved and Cnsolved Problems i n Number Theory From Perfect Numbers to the Quadratic Reciprocity L a w 15
Thus there are infinitely many values of 2 - 1 with p a prime. If, as rather laborious, aiid since Mnirirreascs so very rapidly, it virtually forces
Leibnitz erroneously believed, the converse of Theorem 4 were true, that the creation of other methods. To cstinintc the l a l m involved in proving
is, if ps primality implied 2 - 1s primality, then Conjecture 1 would some ill, a primc by Cataldis method, we must know the number of
follow immediately from Euclids Theorem 2 and Theorem 8. But the primes < ynl, .
converse of Theorem 4 is false, since already
Definition 7. Let
23[211- 1,
dn)
a fact given above in disguised form (example of Definition 4 ) .
be the number of primes which satisfy 2 5 p 5 n.
Definition 5. Henceforth we will use the abbreviation
EXAMPLE
:
Afa = 2 - 1.
~ ( 7 2 3 )= 128.
ATn is called a Afersenne number if n is a prime.
Skipping over an unknown computer who found that M13 was prime, There is no shortage of primes. A brief table shows the trend.
and that Ps = 212f1113was therefore perfect, we now come to Cataldi n r(n)
(1588). He shom-ed that A I l i and MI, were also primes. Xow Mlg = 524,287,
10 4
and we are faced witcha leading question in number theory. Given a large
102 25
number, say AT,, = 2147483647, is it a prime or not? 103 168
To show that N is a prime, one could attempt division by 104 1,229
2, 3, . . . , N - 1, and if N is divisible by none of these then, of course, i t 105 9,592
is prime. But this is clearly wasteful, since if N has no divisor, other than 1, 106 78,498
107 664,579
which satisfies 5,761,455
108
50,817,534 ( U . H. Lehmer)
d 5N
<
455,052,511 (11. H. Lehmer)
then N must be a prime since, if
This brings us to the prime number theorem.
N = fg,
f and g cannot both be > fl. Further, if we have a table of primes which 6. THE PRINENUMBERTHEOREM
includes all primes S f l , it clearly suffices to use these primes as trial I n Fermats time (1630), Cataldis table of primes was still the
divisors since the snmllest divisor ( >1) of N is always a prime. largest in print. I n Eulers time (1738), there was a table, by Brancker,
Definition 6. If z is a real number, by u p to 100,000. I n Legendres time (1798), there was a table, by Felkel,
up to 408,000.
[XI The distribution of primes is most irregular. For example (Lehmer),
we mean the greatest integer 5 s . there are no primes betweeii 20,831,323 and 20,831,533, while on the other
EXAMPLES
: hand (Kraitchik) , 1,000,000,000,061 and 1,000,000,000,063 are both
primes. No simple formula for n ( n ) is either known, nor can one be ex-
1 = [1.5], 2 = [2], 3 = [3.1417], pected. But, in the large, a defitiite trend is readily apparent, (see the
foregoing table), and on the basis of the tables then existing, Legendre
-1 = [-+I, 724 = [GI. (1798, 1808) conjectured, in effect, the Prime ?;umber Theorem.
To prove that 6119 = 524,287 is n prime, Cataldi constrncbtcd the first Definition 8. If f(.c) and y(x) are two functions of the real variable
extensive tahle of primes--up to 750-and he simply tried division of IL, ) asymptotic to g ( x ) , and write it
we say that ~ ( I L is
by all the primcs <[-\/GI = 723. There arc 128 snch primcs. This was
f(z) - Q(Z),
!
i
1 16 Solved and Unsolved Problems i n Number Theory From Perfect Numbers to the Quadratic Reciprocity Law 17
if Theorem gl.
. f(.)
Lim -= 1.
i
! =-* g ( x )
1 Theorem 9. (The Prime Number Theorem, conjectured by Legendre, Theorem g2.

I Gauss, Dirichlet, Chebgshev, and Riemann; proven by Hadamard and de la
Vall6e Poussin in 1896). r(n> - l"*. log 2
1
!
r(n) -- n
log n '
These three versions are all equally true. Which function on the right
is the best approximation?
No easy proof of Theorem 9 is known. The fact tha7 it took a century to P. Chebyshev (1848) gave both Theorems g1 and 9. , but proved neither.
i prove is a measure of its difficulty. The theorem is primarily one of analysis. C. F. Gauss, in a letter to J. F. Encke (1849), said that he discovered
I Kumber theory plays only a small role. That some analysis must enter is Theorem g2 at the age of 16-that is, in 1793-and that when Chernac's
I clear from Definition 8-a limit is involved. The extent to which analysis is factor table to 1,020,000 was published in 1811 he was still an enthusiastic
I involved is what is surprising. We shall give a proof in Volume 11. prime counter. Glaisher describes this letter thus :
j For now we wish to make some clarifications. Definition 8 does not "The appearance of Chernac's Cribum in 1811 was, Gauss proceeds, a
1 mean that f(x) is approximately equal to g ( x ) . This has no strict mathe- cause of great joy to him; and, although he had not sufficient patience
I
matical meaning. The definition in no way indicates anything about the for a continuous enumerat,ion of the whole million, he often employed
difference unoccupied quarters of an hour in counting here and there a chiliad."
f(z) - d z ) , EXERCISE 6. Compute N/log N - 1 (natural logaritkm, of course!)
II merely about the rafio
Thus
f(x)lg(x).
for N = 1071,n = 1, 2, . . . , 10, and compare the right and left sides of
Theorem 91.
7. Two USEFULTHEOREMS
i n2 +1 - n2
Before we consider the work of Fermat, it will be useful to give two
theorems. The first is an easy generalization of an argument used in the
n2 + 100n - nz
proof of Theorem 4, page 3. We formalize this argument as
and n2 + n'.' log n - n2

Theorem 4 0 . If z # y , and n
2
> 0, then
- ylx" - y".
are equally true. Which function, on the left, is the best approximation to
I n particular, if y = 1,
n2is quite a different problem.
If 2 - 11s" - 1,
I f(.) -d 2 ) and, if y = - y , and n is odd,

t
I and g(x> - h(z)

2 + ylz" + y", ( n is odd). (13b)
f(s) -
The proof is left to the reader.
then Wz).
! Theorem 9 may therefore take many forms by replacing nllog n by any
Theorem 10. If a, 6 , and s are positil'e integers, we write
1
! function asymptotic to it. Thus sn - 1 = B,, sb - 1 = Bb.
I:
1 P
I
T h e n if ( a , b) = g, aiid obtain a remainder cf r 1's. On the other hand, the ancient interpreta-
tion of Eq. (17) is that a stick b units long is measured by a stick a units
( B a , Bb) = Bg L ( 14) long, q times, with a remainder r units long.
and in particular if a i s prime to 6, then s - 1 i s the greatest common divisor
of sa - 1 and sb - 1.
PROOF. In computing g = ( a , b ) by Euclid's Algorithm, the ( m 1)st + (17a)
equation (page 9) is
am-i = qmam + am+]. The quotient Q, of Eq. (18), consists of the q marks (bits) made in meas-
It follows that uring I f b by Ma!
Bam-i = &,Barn + Bam+l Q: (000010o0o1000~
for some integer Qm , for the reader may verify that
8. FERMAT
Ah'D OTHERS
Barn-,= sam-' - = '%+I BPmQm + BPmam + BQm+l
Kow we come to Pierre dc Fermat. I n the year 1610, Fraiice was the
leading country of Europe, 110th politically and culturally. The political
leader was Cardinal Richelieu. The leading mathematicians mere Ren6
But BamlBy,,, by Eq. (13a) with 5 = s"", and n = q m , and thus Descartes, G6rard Desargues, Ftriiiat, and the young Blaise Pascal. I n
1637, Descartes had published La Geometrie, and in 1639 the works of
+ 1) Bgmam
__
Barn
Desargues and Pascal on projective geometry had appeared. From 1630
011, Father 8Iarin PIIerseniic, a diligciit correspondent (with an inscrutable
is an integer. Call it Qm and this proves Eq. (16). handwriting) had been sending challenge problems to Descartes, Fermat,
But were we to compute ( B , , &) by Euclid's Algorithm, Eq. (16) Frenicle, and others coiiceriiiiig perfect numbers arid related concepts.
mould be the m +
1st equation and the remainder, B,,,, , of Eq. (16) By his pcrsevcrance, he eventually persuaded all of them to work on perfect
corresponds to the remainder, am+l, of Eq. (15). Therefore if ( a , b ) = g , numbers.
(B, Bb) = Bg . At this time AI, , 1113 , A 1 5 , ill7, A l l 3 , I I I 7, and Af19 were known to be
prime. But
Corollary. Every illersenne number, M , = 2' - I, i s prime to every other
illersenne number.
The correspondence between Eqs. (1.5) and (16) has an interesting and Fermat found tJhat
arithmetic interpretation. For simplicity, let s = 2 and thus B, = Ma =
2" - 1. Let
The ot)vious numerical relationship between p = 11 and t,he factors 23
b=qa+r (17) and 89, in the first instance, and hctween 23 aiid 47 in the second, may
and well have suggested to Ferniat the following
k f b = QAI, + I/, . (18) Theorem 1 1 (Fermat, 1640). If p > 2 , a r q prime which divides A l p
Now I f , , in binary, is a string of z ones, and if the division, Eq. (18), is +-
must be of f h r f o r m 2Xp 1 with I; = 1, 2 , 3, . ' . .
carried out in binary we divide a string of a 1's into a string of b 1's At the same time Fermat found:
100001000 Theorem 12 (Fermat, 1640). Eiwy primp p diilides 2' - 2 :
11111 11111111111111 ( 1%) p12p - 2. (19)
1111111111
111 These two important theorems are closely related. That Theorem 11
implies Theorem 12 is easily seen. Since the product of two numbers of Theorem 14 (Euler). For a n y positive integer m, and any integer a prime
the form 2 k p +1 is again of that form, it is clear by induction that Theorem to m,
11 implies that all divisors of M , are of that same form. Therefore M ,
- 1. (24)
itself equals 2Kp + 1 for some K , and thus ill, - 1 is a multiple of p .
?7Zlu+(m)
And this is Theorem 12. The case p = 2 is obvious. Later we will prove Theorem 14, and since, for a prime p , +( p ) = p - 1,
But conversely, Theorem 12 implies Theorem 11. For let a prime q this will also prove the special case Theorem lR1 . That nil1 complete the
divide AZ, . Then proofs of Theorems 1 3 , 12, and 11. For the moment let us consider the
significance of Fermats Theorem 11 for the perfect number problem.
Q12P- 1, (20) The first Mersenne number we have not yct discussed is J f J g . To de-
and by Theorems 12 and 6, termine whether it is a prime, it is not necessary to attempt division b y
q129-I - 1. (21) 3, 5 , 7, etc. The only possible divisors are those of the form 58k +
1. For
h-ow by Theorem 10, (2 - 1, 2q-1 - 1) = 2 - 1 where g = ( p , q - 1 ) .
k = 1, 2, 3 , and 4 we have 5% +1 = 59, 117, 175, and 233. But 59-fAf?g.
Again, 117 and 175 are not primes and therefore need not be tried, since
Since q > 1, we have from Eqs. (20) and (21) that g > 1. But since p is
the smallest divisor ( >1) must be a prime. 14inally 23?iilJ29 . Thus n.e
a prime, we therefore have plq - 1, or q = s p +
1. Finally if s were odd,
q would be even and thus not prime. Therefore q is of the form 2 k p 1. + i
find that A l 2 9 = 536,870,911 is composite with only 2 trial divisions.
EXERCISE 7. Assume that p = 1603.5002279 is a prime, (which it is),
To prove Theorems 11 and 12, it therefore will suffice to prove one of the
two. and that q = 32070004559 divides ill, , (which it does). Prove that q is
Several months after Fermat announced these two theorems (in a a prime.
letter to Frenicle), he generalized Theorem 12 to the most important EXERCISE
8. Verify that
Theorem 13 (Fermats Theorem). For ezwy prime p and a n y integer a, 3.74 + l/Af37.
plap - a. (22) (When we get to Gausss conception of a residue class, such computations
as that, of this exercise will be much abbreviated.)
This clearly implies Theorem 12, and is itself equivalent to
It has been similarly shown that At,, , i l l 4 3 , M,, , 11153 , and A t 5 9 are also
Theorem 131. If p j a , then composite. Up to p = 61, there are nine Mersenne primes, that is, M ,
plaP- - 1. (23) for p = 2, 3, 5, 7, 13, 17, 19, 31, and 61. These nine primes are listed in
the table on page 22, together with four other columns.
For if p(u(aP-l - 1) and p j a then by Theorem 6, p[aP- - 1. The con- The first two columns are
verse implication is also clear. Kearly a century later, Euler generalized
Theorem 13 and in doing so he introduced an important function, + ( n ) . s, = Imp] (25)
and
Definition 9. If n is a positive integer, the number of positive integers
prime to n and 5 n is called +( n ), Eulers p h i function. There are therefore cp = P(SP). (26)
+ ( n )solutions nz of t8hesystem: The number c, is the number of trial divisions-i la Cataldi (see page 14)
needed to prove M , a prime.
/ (m,n) = 1
11 5 m 5 n. ak
Definition 10. By a,,b(n)is meant the number of primes of the form
+ b which are sn.
EXAMPLES
:
EXAMPLES
:
+(1) = 1, + ( 2 ) = 1, 443) = 2, +(4) = 2, + ( 5 ) = 4,
~ , , ~ ( 5 0=) 6; the six primes bring 5 , 13, 17, 29, 37, ill
+(G) = 2, +(7) = 6, +(8) = 4, +(9) = 6, +(lo) = 4. 7r43(50) = 8; thr eight primes being 3, 7, 1 I , 19, 23, 31, 13, 47
~ ~ ~ ( =1 19552 0 ~ )
For any prime, p , + ( p ) = p - 1.
a 8 , 3 ( 1 O 6 ) = 19653
aB,s(1O6) = 19623 or, equivalently, for a n y two numbers prime to a, b and b, we have
a8,7(1O6)= 19669.
By Theorem 11, the only primes whirh may divide d l , are those counted
ra,b(n) - a,,b(n). (29)
We postpone the proof of Theorem 15 to Volume 11, but a special case
by the function ~ ~ , , ~ (The
n ) .next column of the table is which we need later is proven in Section 36. The more difficult Theorem 16
f, = %,l(S,). (27) will be used as a guide in the following investigations but will not be used
logically and will not be proven. We note that although Eq. (28) is an
The last column, e, , we d l explain later. (hlnemonic aid: c p means
asymptotic law, we may nonetheless employ it for even modest values of
Cataldi, f p means [Fermat, e, means Euler.)
n with a usable accuracy. Thus r#~(38) = 18; more generally, for any prime
TABLE O F THE FIRST NINE MERSENNE PRIMES p , 4 ( 2 p ) = p - 1. Then ~ ( s , , ) = 128 and Aa(s19) = 7.1. The number
sought is ?r38,1(s19) = fig = 6, a reasonable agreement considering the
P P JP =P
__ smallness of the numbers involved. Generally we should expect
n
2 3 1 0 0 0 1
3 7 2 1 0 0 f P =-
P-1
CP
5 31 5 3 0 0
7 127 11 5 0 0 but it is clear that this is not an exact statement, since we give no bound
13 8,191 90 24 2 1 on the error.
17 131,071 3G2 72 4 3
19 521,287 (Cataldi, 158s) 724 128 6 3 EXERCISE 10. The ratio s,/cp may be regarded as a measure of the
31 2,147,483,647 (Euler, 1772) 46,340 4,702 157 84 improvement introduced by Cataldi by his procedure of using only prinzcs
61 2,305,843,000,213,6~3,O51 1.5.109 75.1()6* 1.25.106** 0. fi2. 10G**
as trial divisors (page 14). Similarly, c p / f p measures the improvement
* Estimated, using Theorern 9. made by Fermat. Now note that the second ratio runs about 3 times the
** Estimated, using Theorem 10 first, so that we may say that Fermats improvement was the larger of
the two. Interpret this constant ( ~ 3 as) 2/log 2 by using the estimates
We see in the table that had Cataldi known Theorem 11, the 128 di-
for c p and f p suggested by Theorems 9 and 16. Evaluate this constant to
visions which he performed in proving dl,, a prime could have been re-
several decimal places.
duced to 6 ; j l p = 6 .
EXERCISE 9. Identify the two primes in f 1 3 , namely those of the form 9. EULERS
GENERALIZA4TIONP R O V E D
26k +
1 which are <90. Also identify the 4 primes in j 1 7 . We now return t o Eulers Theorem 14,
We inquire now whether the ratio j,/c, will always be as favorable as
the instances cited above. RIore generally, how does a,,h(n) compare with mIa+(m- 1, (a,m) = 1
~ ( n )Since
? ah; + b is divisible by g = ( a , b ) it, is clear that the form which we will prove by the use of the important
OX- +b cannot contain infinitely many primes unless b is prime to a . But
Theorem 17. Let m > 1. Let a , , 1 5 i 5 +(m),be thc +(nz) potdilte i n -
suppose ( a , b ) = I ? If we hold a fixed there are + ( a ) values of b which are
<a and prime to a. Does each such form possess infinitely many primes? tegers <in and prime to m. Let a be any intcgw prime to 711. Lct the + ( ? I L )
Two famous theorems answer this question : products, aal , aa2 , . . . , aa+(,) be divided by nz, giving
Theorem 15 (Dirichlet, 1837). If ( a , b ) = 1, there are infinitely many I aaz = q,m 4- r, (31)
primes of the form alib. + i with 0 5 r z < m.
A stronger theorem which implies Theorcm 15 (and also Thcorcm 9) is
T h e n the +( m) values of r , are distinct, and are equal to the +( m) values of
Theorem 16 (de la Vall6e Poussin, 1896). I f (a, b ) = I , then a Lin some rearrangement.
PROOF
OF Tmmnmq 17. Since a and a, are both primc to m , so is their
(28) product-by Theorem 5 , Corollary. Therefore, from Eq. (31), r I is also
I
prime to m and thus is equal to one of the ai . If r , = r j we have from Our logical structure so far (not including Theorem 8 and the unproven
Eq. (311, Theorems 9, 15 and 16) is given by the diagram on the previous page.
a(ai - a j ) = (q; - qj)m 10. PERFECT
NUMBERS,
I1
Thus from Theorem 6, since ( a , m) = 1, I n the previous sections we have attempted to look a t the perfect num-
mlai - a j bers thru the eyes of Euclid, Cataldi and Fermat, and to examine the
consequences of these several inspections. In the next section we take
or a; = aj . Thus the ri are all distinct. u p other important implications which were discovered by Euler. The
PROOF OF THEOREM
14 (by Ivory). The product of any two equations in reader may be inclined to think that we have no sincere interest in the
Eq. (31) is perfect numbers, as such, but are merely using them as a vehicle to take
us into the fundamentals of number theory. We grant a grain of truth to
a'aia, = Qm + rirj this allegation-but only a grain. For consider the following:
for some integer Q, and by induction, the product of all 4 ( m ) equations in If N is perfect it equals the sum of its divisors other than itself.
Eq. (31) can be written
a4(m)
alaz * . a+,,) - rlrz . . * rdfrn)= Lm
Dividing by N, we find that the sum of the reciprocals of the divisors, other
for some integer L. But (Theorem 17) the product of all the ri equals than 1, is equal to 1.
the product of all the ai . Since
(aQ'"' - l)alaz . . . a+,,)
is divisible by m, and each a, is prime to m, by Theorem 6
For Ps = 28, we have, for instance,
mlaQ(m)- 1.
This completes the proofs of Theorems 14, 131 , 13, 12, and 11. 1 = -1+ - +1- + - +
1- 1 1
7 14 28 4 2'
1 Now write these fractions in binary notation. Since 7 (decimal) = 111

(binary), we have
7-
1 -- .001001001001 ...
7
A = .000100100100 . . . (shift right one place)

$B = .000010010010 . . . (shift right one place)
1 -
- .01oO00oO0000 ...
+ = .100000000000 . . .
sum = 1 = .111111111111
The fractions not only add to 1, but do so without a single carry! And
as it is with 28, so is it with 496. Is that not perfection-f a sort?
11. EULERAND M,,
We continue to examine the Mersenne numbers, M , , and our attempt
to determine which of these numbers are prime. In Theorem 11 we found
that any prime divisor of M , is necessarily of the form 2kp + 1. We now
seek a sufficient condition-that is, given a. prime p and a second prime
q = 2kp + 1, what criterion will suffice to guarantee that q \ M , ? Consider
the first case, k = 1. Given a prime p, q = 2p +1 may be a prime, as for
p = 3, or it may not, as for p = 7 . If it is, y may divide M , , as In view of the discussion above we can at once write the
23iMll and 471MZ3 Corollary. If p = 4m + 3 i s a prime, with m > 0 , and if (I = 2p +1
i s also a priine, then qlill, -and thus 2p1ilf, i s not perfect.
! or it may not, as
Like Fermats Theorem 12, we will not prove Theorem 18 directly,
I 11+M6and 59+M29*. but deduce it from a more general theorem. This time, however, the
I
What distinguishes these two classes of y? To help us discover the criterion, generalization is by no means as simple, and we shall not prove Theorem 18
consider a few more cases: until Section 17. For now we deduce a second important consequence.
~
i 71M3** and 1671Ms3 Theorem 19. Eiiery divisor of M P ,

for p > 2 , i s of the jorm 8X: =!= 1.
but
PROOF.
Let y = 2Q + 1 be a prime divisor of I f p . Then
i 831M41 and 107tM63.
q12ilfp = 2pf - 2 = N 2 - 2 (34)
i N = 2(P+1)12
I
The reader may verify (in all these cases) that if p is of the form 4nz 3 + where
I and thus q = 8m +
7 , then q \ A f p ,whereas if p is of the form 4ni 1 and + Thus 2 = N 2 - Kq
! thus y = 8m +
3, then q+M, . Does this rule always hold?
i
Consider the question in a more general form. Let q = 2& 1 be a+ for some integer K . Then
prime with Q not necessarily a prime. When does 22 = N 4 - Kzq
y12Q - l ? for some integer K 2 , and, by induction
N Z Q- Ly.
II
By Fermats Theorem we had 2 =
y122Q- 1, now ( I ~ Nsince

, 412, and thus, by Fermats Theorem, qINZQ- 1. There-
fore ~ 1 - ~ and, by Theorem 18, q must be of the form 8k =!= 1. Fi-
2 1,
and factoring the right side: nally, since the product of numbers of the form 81; f 1 is again of that
i q1(2Q - + 11, form, all divisors of A l p are of the form 8k f 1.
We were seeking a sufficient condition for q l M p and found one in the
we find from Theorem 6, Corollary that either corollary of the previous theorem. Here instead we have another necessary
condition. Let us return to the table on page 22. We may now define
or ep , the last column. Prom the primes counted by f, = q,, I ( s,), we delete
those of the form 8k f 3. By Theorem 19 only the remaining primes can
qualify to be the smallest prime divisor of If, . We call the number of these
It cannot divide them both since their difference is only 2. Which does it primes e p .
divide? To give the answer in a compact form we write the class of integers As an example, consider M31. For nearly 200 years, Cataldis 11119 had
8k +7 as 8k - 1 and the class 8k +
5 as 8k - 3. Then we have been the largest known Mersenne prime. To test N 3 1 , we examine the
primes which are <46,340, of the form 62k +
1, and of the form 8k f 1.
I
t Theorem 18. If q = 2Q + 1 i s prime, then Let k = 4 j + nz with m = 0, 1, 2 , and 3. Then the primes of the form
q12Q - 1 i f q = 8k f 1, (32)
I
62k + 1 are of four types:
and 248j+ 1 = 8(31j)+ 1
I 248j + 63 = 8(31j + 8) - 1
4124 +1 if q = 8k f 3. (33)
248j + 125 = 8(31j + 16) - 3
* Nonctheless M z is~ composite, since 2 3 3 / M ~.,
** Norletheless Mt is prime, since 7 = Ms . I 2481 + 187 = 8(31j + 23) + 3.
28 Solved and Unsolved Problems in Number Theory From Perfect hTumbers to the Quadratic Reciprocity Low 29
The last two types we eliminate, leaving for which primes p = 4m +1 and primes q = Gp +
1, does qlM, ? But
e31 = T248. ~(sd + T 2 4 8 , ds31).
this time the answer is considerably more complicated than was the crit,erion
for y = 2 p + 1 above. A short table is offered the reader:
Euler found that no prime q satisfied
y < 46,340*
+ 1 or
248k p = 5,37, 73, 233
I p = 13, 17, 61, 101, 137, 173, 181
( + 63
248k
!l=
EXERCISE 16. Can you find the criterion which distinguishes these two
and qIJf31 . classes of q? This was probably first found (at least in effect) by F. G.
Thus il13] = 2147483647 was the new largest known prime. It remained Eisenstein. It is usually stated that the three greatest mathematicians
so for over 100 years. were Archimedes, Kewton and Gauss. But Gauss said the three greatest
were Archimedes, Newton and Eisenstein! The criterion is given on page
EXERCISE 11. Show that if p = 4m + 3, q 2 k p + I, and q l M p ,
=
169.
then k = 4r or k = 4r +
1. If p = 4na + 1, and qlM,, then k = 4r or
k = 4r +3. 12. MANYCONJECTURES .4ND THEIRI S T E R R E L A T I O X S
EXERCISE
12. Show that i p + I never divides ill,. So far we have given only one conjecture. But recall the definitions of
EXERCISE
13. Show that if p 4m + 3,
= conjecture and open question given on page 2. Since by Open Question 1
we indicate a lack of serious evidence for the existence of odd perfects,
ep = T8p. 1 b p ) + T ~ Pf p,t l ( S p ) , it is clear that if we nonetheless conjecture that there are infinitely many
while if p = 4m + 1, perfects, what we really have in mind is the stronger
Conjecture 2. There are infinitely m a n y Mersenne primes.
ep = T8p. 1(sp) + TSp, 6p4l(sp).
Contrast this with
EXERCISE 14. Show that ep is approximately one half of f p . Com-
Conjecture 3. There are infinitely m a n y Aiersenne composites, that i s ,
pare the actual values of c31, f S 1 , and ex1on page 22 with estimates ob- composites of the f o r m 2 - 1, with p a prime.
tained by Theorems g1 and 16.
Is this a conjecture? Yes, it is. It has never been proven. It is clear that
EXERCISE15. Identify the 3 primes in e19 . at least one of these two conjectures must be true.
A glance a t M 6 1 , the last line of the table on page 22, shows that a By Theorem 18, Corollary, Conjecture 3 would follow from the stronger
radically different technique is needed to go much further. Eulers new Conjecture 4. There are infinitely m a n y primes p = 4m +
3 such that
necessary condition, e p , only helps a little. But the theory underlying e p
is fundamental, as we shall see.
q = 2p + I i s also prime.
The other advance of Euler, Theorem 18, Corollary, seems of more But this is also unproven-although here we may add that the evidence
(immediate) significance for the perfect number problem. It enables us for this conjecture is quite good. We listed on page 28 some small p of this
to identify many M , as composite quite quickly. For the following primes type. Much larger ps of this type are also known. Some of these are
p = 4m + 3, q = 2p +
1 is also a prime: p = I I , 23, 83, 131, 179, 191, p = 16035002279, 16045032383, 16048973639, 16052557019, 16086619079,
16118921699, 16148021759, 16152694.583, 16188302111, etc.
239, 251, 359, 419, 431, 443, 491, 659, 683, 719, 743, 911, . . . . All these
hi, are therefore composite. For any of these p , y = 2 p + IlJf,, arid M, is a number, which if
I n Exercise 12, we saw that 4 p +
l-fJi,. But if p = 4m 1, then + written out in decimal, would be nearly five billion digits long. Each such
q = 6p -t 1 = 8(3m) +
7 is not excluded by Theorem 19. Again we ask, number would more than fill the telephone books of all five boroughs of
New York City. Imagine then, if Cataldi were alive today, and if he set
* Note that Branckers table of primes sufficed. It existed then and included himself the task of proving these M , composite-by his methods! Cant
primes <100,000-see page 15.
you see the picture-the ONR contract-the thousands of graduate as-
30 Solved and Unsolved Problems in Number Theory From Perfect N m i b e r s to the Quadratic Reciprocity Lalo 31
sistaiits gainfully employed-t he Beneficial Suggestion Committee, etc.? to Artins Conjecture and to Fermats Last Theorem, but it would be too
But we are digressing. digressive to give explanations a t this point.
We had occasion, in the proof of Theorem 19, to use the fact that
Conjecture -4 also implies the weaker
Conjecture 5. There are infinitely m a n y primes p such that q = 2 p 1 + 2M, = N2 - 2
i s also przme. Or, equiaalently, there are infinitely m a n y integers n such that for some N.Thus Conjecture 2 implies the much weaker
? L f 1 i s prime, and n i s twice a prime.
Conjecture 9. There are injinitely-many n for which n2 - 2 i s twice a
Conjecture 5 is very closely related* to the famous prime.
Conjecture 6 (Twin Primes). There are infinitely m a n y integers n such This is clearly related to
that n - 1 and n +
1 are both primes.
Conjecture 10. There are infinitely m a n y primes of the f o r m n2 - 2.
While more than one hundred thousand of such twins are known, e.g.,
7~ = 4. 6, 13, 18, 30, . . . , 1000000000062, 1000000000332, . . . , While more than 15,000 of such primes are known, e.g. n = 2, 3, 5,
140737488333508, 140737438333700, a proof of the conjecture is still 7, 9, . . . , 179965, . . . , a proof of the conjecture is still awaited. Yet it is
awaited. Yet it is probable that a much stronger conjecture is true, namely probable that a much stronger conjecture is true, namely
Conjecture 7 (Strong Conjecture for Twin Primes). Let z ( N ) be the Conjecture 11. Let P-Z(N) be the number of primes of the f o r m n2 - 2
number of pairs of twin primes, n - 1 and n 1, f o r 5 5 n 1 5 N.+ + for 2 5 n j N . T h e n
Then
z(N) - 1.3203236
N
dn
(35)
PP2(N)- 0.9250272 SNdn
z log n
(36)
The constant in Eq. (3.5) is not empirical but is given by t8heinfinite On page 48 we will return to this conjecture. I t is known to be related to
product Conjecture 12. Let P,(IV) be the number of primes of the f o r m n2 +1
m I -
1
\ for 1 5 n 5 N . T h e n
1.3203236 . . .
taken over all odd primes.

= 2
p=3 {1 -
(35%)
Pl(N) - 0.6864067 S -.
2
dn
log n
(37)
In Exercise 37S, page 214, we will return to this conjecture. It is knon-n
to be intimately related to the famous As in Eq. ( 3 5 ) , the constants in Eqs. (36) and (37) are given by certain
infinite products. But we must postpone their definition until we define
Conjecture 8 (Goldbach Conjecture). Every euen number > 2 i s the s u m the Legendre Symbol.
of two primps.
EXERCISE 17. On page 29 there are several large primes p for which
EXAMPLES
: q = 2p + 1 is also prime. These were listed to illustrate Conjecture 4.Now
4 = 2 + 2 show that the qs also illustrate Conjecture 10.
6 = 3 + 3 But we do not want to leave the reader with the impression that number
theory consists primarily of unsolved problems. If Theorems 18 and 19 have
8 = 3 + 5 unleashed a flood of such problems for us, they mill also lead to some beauti-
10 = 5 + 5 = 3 + 7, etc. ful theory. To that we now turn.
Returning to Conjecture 5 , we will indicate now that it is also related 13. SPLITTING THE P R I M E S INTO EQUINUMEROUS cL4SSES
* By related M emean here t h a t t h e heuristic arguments for t h e two conjectures Definition 11. Let A and R be two classes of positive intrgers. Let, A(n)
are so similar t h a t if we succeed in proving one conjecture, t h e other will almost be the number of intrgers in iZ \\-hich are 5 n ; and let B ( n ) be similarly
surely yield to t h e same technique.
defined. If There are, of course, 8 different bs, since +(24) = 8. It will be usefill for
A(n) - B(n)
the reader at this point, t.0 know a formula of Euler for his phi function.
In Sect. 27, when we give the phi function more systematic treatment, we
we say A and B are equinumerous. will prove t,his formula. If N is written in the standard form, Eq. ( 1 ) , then
By this definition and Theorem 16 the four classes of primes: 8k -t- 1,
8k - 1,8k +
3, and 8k - 3 are all equinumerous. Now Theorem 18 stated
that primes p = 2Q +
1 divide 2 - 1 if they are of the form 8k 1 + As an example
or 8k - 1. Otherwise they divide 2Q +
1. Therefore the two classes of
+(24) = 24(1 - $)(I - 5) = 8.
primes which satisfy
But this does not end the problem of the generalization. Still another
q12 - 1 and ~ 1 2 1~ + base, e.g., 5 , 6, 7, etc., will introduce still another cleavage plane. The
are also equinumerous. problem is this: What criterion determines which of the odd primes q,
We expressed the intent (page 27) to prove Theorem 18 not directly, (which do not divide a ) , divide aQ - 1, and which of them divide aQ I ? +
but, following the precedent: By Theorem 131exactly one of these conditions must exist.
Theorem 13 -+ Theorem 12, 14. EULERS
CRITERIOX FORMULrlTED
to deduce i t from the general case. The difficulty is that the generalization The change of the base from 2 to 3 changes the divisibility laws from
is not a t all obvious. For the base 3, there is Eqs. (32) and (33) in Theorem 18 to Eqs. (38) and (39) in Theorem 20.
Theorem 20. If y = 2Q+ 1 # 3 i s a prime, then Euler discovered what remains invariant. In the proof of Theorem 19 the
following implication was used: If there is an N such t,hat ylN2 - 2, then
~ 1- 2 1.~ The reader may verify that the number 2 plays no critical role
and in t,his argument,, so that we can also say that if there is an N such that
qIN2 - a, and if y j a , then pjaQ - 1. The implication comes from Fermats
Here, again, we find the primes, (not counting 2 and 3 ) , split into equi- Theorem 131 , and the invariance stems from t,he invariance in that theorem.
numerous classes. But this time the split is along quite a different cleavage Now Euler found that the converse implication is also true. Thus we
will have
plane-if we may use such crystallographic language. Thus 7/2a - 1,
+
while 7133 1. Theorem 21 (Eulers Criterion). Let a be a n y integer, (positive or nega-
Since primes of the form 8k +
1 are either of the form 24k +
1 or of tive), and let q = 2Q +
1 be a prime which does not divide a. If there i s a n
the form 24k + 17; and since primes of the form 12k - 5 are either of integer N such that
the form 24k + +
7 or of the form 24k 19; etc., the reader may verify q1N2 - a, then plaQ - 1.
that Theorems 18 and 20 may be combined into the following diagram:
+
If there i s n o such N , then qlaQ 1. I t follows that the converses of the last
For p = 24k + b = 2& + 1 = prime: two sentences are also true.
Before we prove this theorem, it will be convenient to rewrite it with a
q)ZQ - 1-1 b = 1, 23. I b = 7, 17. I notational change introduced by Legendre.

Definition 12 (Legendre Symbol-the current, but not the original
definition). If q is an odd prime, and a is any integer, then the Legendre
Symbol (i) has one of three values. If y / a , then
(3 = 0. If not, then
(i) = +1 if there is an N such that qIN2 - a, and = - 1 if there is

not.
EXAMPLES
: planes, 2Q + I , 3 f 1, and 5 f 1. I n each of the eight cubes there will be
+
(s) = +1 since 713 - 2.
four values of b, corresponding to four classes of primes, q = 120k
All toget,her there will be 32 classes, corrcsporiding to +( 120) = 32.
15. EULERS PROVED
CKITERIO?;
b.
(;) = -1. Our proof of Throrem 211 will he based upon a theorem related to
Theorem 17.
(i) = +1
since, for every p, q 1 1 2 - 1. Theorem 22. Let q be prime, and Id a , , i = 1, 2 , . . . , q - 1, be the posi-
title integers < q . Let a be a n y integer prime to q. Gicen a n y one of the aL,
there i s a unique j such that
)(: = +1 if qja, since, for every q, qla2 - a2
qla,aj - a. (43)
Now we may rewrite Eulers Criterion as
PROOF.By Euclids Eq. (7), page 9, there is an m and an n such that
Theorem 211 . If q = 2Q + 1 i s a prime, and a i s a n y integer,
ma, + n q = 1,
pla - (;) or
We may remark that usually Eulers Criterion is presented as a method nmaL + naq = a. (44)
of evaluating (;) by determining whether qla - I or not. The reader Since ( m , q ) = 1, we have q t m a and if me divide ma by q we obtain
may note that we are approaching Eulers Criterion from the opposite ma = sq + a, (45)
direction. The fact is, of course, that Eulers Criterion is a two-way im- for some j and some s. From Eqs. (44) and (45),
plication, and may be used in either direction.
qla,a, - a.
18. From Theorems 18 and 211 show that for all odd primes p ,
EXERCISE
Now, for any li such that
qla& - a,
Likewise we have
qladak - a>),
and, since q t a , , we have ql(ak - a,), that is, k = j.
where the square bracket, [ 1, is a s defined in Definition 6. Now we can prove Theorem 211 .
EXERCISE19. Determine empirically the cleavage plane for q [ S Q f 1,
which is mentioned on page 33, by determining empirically the classes of
OF THEOREM
PROOF 211 (by Dirichlet). Assume first that t)
With rcferencc to Definition 12, this implies that the j and i in Eq. (43)
= -1.
primes q which divide N 2 - 5, and those which do not. That is, factor N 2 -
5 for a moderate range of N , and conjecture the classes into which the can never be equal. Therefore, by Theorem 22, the 2Q integers a Lmust
prime divisors fall. You will be able to prove your conjecture after you fall into Q pairs, and each pair satisfies an equation:
learn the Quadratic Reciprocity L a w . a, a j = a + Kq (46)
EXERCISE 20. On the basis of your answer to the previous exercise, for some integer K . The product of these Q equations is therefore
extend bhe diagram on page 32 to three dimensions, with the three cleavage
( Z Q ) ! = a* + Lq
for some integer L. Therefore Therefore ( 2 Q ) ! = -1 + Kq for some K , and Eq. (51) becomes
(;) = -1 implies qlaQ - (2Q) ! . (47) qla - (;) (53)
Now assume (i) = + l . Then q1N2 - a for some N , and, since q+N
(3
Finally if - = 0, qla, and Eq. ( 5 3 ) is still true. This completes the proof
we may write N = sq + al. for some s and r . Therefore of Theorem 211.
It may be noted, that if b2 = a, then by Eq. (40), and the last example
qlar2 - a. (48) of Definition 12, we again derive
If, for any t,
glb2Q - 1,
q l d - a, which is Fermats Theorem 131. This theorem is therefore a special case
then from Eq. (48), both of Eulers Theorem 14, and his Theorem 211.
!?la? - a:, or ql(at - a,)(at + a,). EXERCISE 21. There have been many references to Fermats Theorem
in the foregoing pages. With reference to the preceding paragraph, review
Thus either t = r , or at + a, = mq. In the second case, since a t and a, the proof of Theorem 211 to make sure that a deduction of Fermats Theo-
are both < q , m = 1, and therefore at = q
are exactly two values of a, which satisfy the equation

- a, . Thus if
(3
- = +1, there rem from Eulers Criterion is free of circular reasoning.
We have set ourselves the task of determining the odd primes q = 2Q 1
which divide uQ - 1. Eulers Criterion reduces that problem to the task of
+
These two values, a, and a t = q

q12 - a.
- a, , satisfy
evaluating (i). This, in turn, may be solved by Gausss Lemma and the
Quadratic Reciprocity Law. It would seem, then, that Eulers Criterion
-apt = u + Kq (49) plays a key role in this difficult problem. Upon logical analysis, however,
it is found to play no role whatsoever. Theorem 21 and Definition 12 will
for some K .
be shown to be completely unnecessary. Both are very important-for
The remaining 2Q - 2 values of a , fall into Q - 1 pairs (as before) and
other problems. But not here. If we have nonetheless introduced Eulers
each such pair satisfies Eq. (46). The product of these Q - 1 equations,
Criterion at this point it is partly to show the historical development, and
together Qith Eq. (49), gives partly to emphasize its logical indepcndcnce.
-(2Q)! = aQ + Mp 16. WILSONSTHEOREM
I n the proof of Theorem 211 we have largely proven
for some M . Therefore -
Theorem 23 (Wilsons Theorem). Let
(;) = +1 implies N = ( q - I)! + 1.
Equations (47) and (50) together read Then N i s divisible by q i f and only i f q i s a prime.
PROOF (by Lagrange). The if follows from Eq. (52) if q is an odd
prime, since p - 1 = 2Q. If Q = 2 , the assertion is obvious. If q is not
a prime, let q = rs with r > 1 and s > 1. Then, since s i ( q - 1) !, s+N.
If we let a = 1, by the third example of Definition 12, we have, for every q, Therefore q+N and qlN only if q is prime.
The reader will recall (page 14) that when we were still with Cataldi,
q1(2Q)! + 1. (52) we stated that a leading problem in number theory was that of finding an
38 Solved and Unsolved Problems in N u m b e r Theorg F r o m Perfect N u m b e r s to the Quadratic Reciprocity L a w 39
efficient criterion for primality. In the ahmice of such a criterion, we have A large step in this direction stems from the simple
used Fermats Theorem 1 I , and Eulcrs lheorcm 19, to alle\iatc the
problem. Kow MTilsonsTheorem is a necessary aiid hufficient condition Theorem 24. Let a,(i = I , 2, . . . , Q ) bc the positave odd integers less t h a n
for primality. But the reader may easily verify that it, is not a practical a prime q = 2Q + 1, and let a be a n g integer not divisible b y q. Let
criterion. Thus, to prove MI, a prime, we n-ould h a w to compute: aa, = qtq + (0 < r L< 4 ) (55)
5242871524286! + 1. (54) as in Eq. (31) of Theorem 17, ( p a g e 2 3 ) . In addition to the result given
But the arithmetic involved in Eq. (54) is much greater than tvcn that there that all Q of these rt are distinct, it is also true that n o two of t h e m add
used in Cataldis method. K e will return to this problem. to q:
EXERCISE
22. If 4 = 2Q + 1 is prime, arid Q is even, T, + f q. (56)
d(Q9z + 1. PROOF.
If r I + r j = q, then, from Eq. (55),
EXERCISE23. a(a, + a,) = Ky (57)
($) = (-l)Q,
for some integer K . But a , +
a j is even and <2y. Therefore q t a , a, +
and Eq. (57) implies yla. Since this cannot be, we obtain Eq. ( 5 6 ) .
+ 1 are of the form 4nz + 1.
and therefore all odd divisors of n2
From this simple observation we obtain an important result which we
will call Gausss Criterion.
24. For a prime y = 4m + 1, find all integers R: which satisfy
EXERCISE
Theorem 25 (Gausss Criterion). Let y , a, a , , and r , be as in the previous
+
qlxZ 1. theorem, a n d let y be the number of the r2 which are even-and therefore not
EXERCISE23. We seek t o generalize Wilsons Theorem in a manner equal to some a , . T h e n
analogous to Eulers generalization of Fermats Theorem. Let m be an qlaQ - ( - 1 ) 7 ; (W
iiiteger > 1 and let a, be the + ( m ) integers I , . . . , m - 1 which are prime
to m. Let A be the product of these +(m)integers a , . Then for m = 9 i.e., qlaQ - 1 or qlaQ + 1 according as y i s even or odd.
or 10, say, we do find ntlA + 1 analogous to p l ( p - 1) ! + 1 for p prime. PROOF. The set of Q remainders, r L, given by Eq. ( 5 5 ) , consists of y
But for nt = 8 or 12 n e have, instead, mlA - 1. Find one or more addi- even integers and Q - y of the odd integers a , . Let each of the y even
tional composites m in each of these categories. JiTewill develop the com- integers, r , , be written as q - ak for some k. But this a k cannot be r m ,
plete theory only after a much deeper insight has been gained-see Ex- one of the Q - y odd remainders, since, if it were, we would have r , +
ercise 88 on page 103. T~ = y in violation of Theorem 24. Therefore, for each a , , either a Lis one
of the odd r , or q - a, is one of the even r , , but not both. I n the first case
17. GAUSSS
CRITERION
we have I
After our digression into Eulers Criterion, we return to the problem
posed 011 page 33. Which of the primes q = 2Q +
1, which do not divide aa, = q,q + a, (59)
a, divide aQ - l ? The similarity of Theorems 18 and 20, for the cases for some j , and in the second case we have
a = 2 and a = 3, may creak the imprtssion that the problem is simpler
than it really is. But consider a larger value of a-say a = 17. Then it will auk = qky + ( q - ail
be found that primes of the form 31k f 1 , 3 4 k f 9,34k =!= 13, and 34k & 15 = (qk + 1 ) q - ai
divide 17Q - I , while 341; f 3, 3.21~f 5, 34li f 7, and 34-X:f 11 divide
+
17Q 1. Such complicated rules for choosing up sides seem obscure in- for some k .
If we now take the product of the y equations of type Eq. (60) and the
deed. Thus the complete, and relatively simple solution for every integer
a , a t the hands of Euler, Ltgendre, and Gauss, may well be corisidered a Q - y equations of type Eq. (59) we obtain
Solved Problem par excellence. aQ(ala, . . . a*) = ~y + (-1) (alaz . . . qQ)
for some integer L. Proceeding as we did in Theorem 14 (page 24) we Definition 13 (Original Legendre Symbol). If q = 2Q + 1 is prime, and
obtain Eq. (58). a is any integer, then (alq) has one of three values. If qla, then (alq) = 0.
EXERCISE 26. Derive Fermats Theorem from Gausss Criterion, and,
+
If not, then (alq) = +1 if q(aQ- 1 and (alq) = -1 if qlaQ 1. Inevery
case
as in Exercise 21, check against circular reasoning.
With Gausss Criterion we may now easily settle Theorem 18. !?laQ- (aid. (62)
PROOF OF THEOREM 18. Let a = 2 in Theorem 25. If Q is odd, there are This looks very much like Eulers Criterion. But of course it isnt. It is
(Q +
I ) /2 odd numbers, 1, 3, 5 , . . , Q, whose doubles
1
merely a definition, not a theorem. Further, there is nothing in this defini-
2.1, 2.3, . . . , 2 - Q tion about an N such that q1N2 -
a, etc. In view of Theorem 211,
Eq. (40) , it is clear that
are less than q. Therefore qi = 0 in Eq. (55), and these even products,
2ai, are themselves the r , . The remaining products (63)
2(Q + 2 ) , 2 ( Q + 41, *. . 2(2Q - 1)
1
We stated above, however, (page 37) that the solution of the problem
will have qi = 1 and therefore their ri will be odd. Thus, if Q is qlaQ f 1 is logically independent of Eulers Criterion and Definition 12.
odd, y = (Q + 1)/2. Likewise, if Q is even, the Q/2 products For the present then, we will ignore Eqs. (63) and (40), and confine
2.1, 2.3, . . . ,2 - ( Q - 1) ourselves to Definition 13 and Eq. (62).
I n terms of the original Legendre symbol we may rewrite Gausss
have even r , , and y = Q/2. Both cases may be combined, using Definition Criterion as
6, in the formula
Theorem 2S1. W i t h all symbols having their previous meawing, we have
(aid = (64)
i.f n4a.
From Eq. (58) we therefore have The symbol ( a lq) has two important properties-it is multiplicative and
periodic.
41Z4 - (-1) [q+141 (compare Exercise 18). (61)
Theorem 26.
Finally if q = 8k f 1, = 2 k . And if q =
2k f 1. This completes the proof of Theorem 18, and therefore also of

Cq J I
8 k f 3, - -
-
(a
(ablq) = (alq) (big).
+ kqlq) = (aln) ?
( 65)
(for a n y integer k). (66)
Theorem 19.
18. THEORIGINAL SYMBOL
LEGENDRE
With the proofs of Theorems 18 and 19, we might consider now whether
we should pursue the general problem, qlaQ f 1, or whether we should
return quickly to the perfect numbers. But there is little occasion to do the
latter. We have already remarked (page 28) that a radically different
technique is needed to go much further. Such a radically different tech-
nique is the Lucas Criterion. But to obtain this we need some essentially
new ideas. And to prove the Lucas Criterion we will need not only Theorem
18, but also Theorem 20-the case a = 3. We therefore leave the perfect
numbers, for now, and pursue the general problem.
Legendres original definition of his symbol was not Definition 12, but
Again, by Eq. (13), The theorem may also be stated as follows: ( p l q ) = ( a l p ) unless p
ql(a + lid - a and q are both of the form 4172 +
3. In that case, PQ is odd, and ( p l q ) =
-(qlp).
for every n. Therefore q [ ( a + Icq) - aQor Before we prove Theorem 27, let us state right off that it completely
aI(a + kqld - (aid.
solves our problem, qjaQ=t1. We stated above that what remained was to
(68)
Since the right sides of Eqs. (67) and (68) are less than q in magnitude,
evaluate all (pjq). But if p > q we may write p = sq +
r and therefore,
by Eq. (66), ( p l q ) = ( r l q ) . Without loss of generality we may therefore
they must both vanish, and therefore Eqs. (65) and (66) are true. assume p < q. But in that case we may use Eq. (76) and obtain
To solve the problem qlaQ f 1, we must evaluate (alp). If qja, there is
no problem. Let qtu and let a be a positive or negative integer written (PI!?) = ( - l ) p Q ( q l P )
in a standard form so that if q = sp + r,
a = *pyp;z . . . p, (69) (Pld = ( -UPQ(rlP).
By factoring out p? for every even a , , and p:I- for every odd a j > 1, Thus we reduce a symbol whose right argument is a prime p to one whose
we are left with right argument is a smaller prime p . By continuing this reduction we must
a = j=pJpk ... p a 2 , eventually get down to a symbol
(70)
a product of primes times a perfect square N 2 . Now, from Eq. (65), (-1lq), (lip), or (2lq)
(N21d ( N l q ) ( N I q ) = +1 which we can evaluate by Eqs. (73), (74) or (75).

= (71)
To illustrate these reductions we will evaluate several (aiq) and prove
since q % N , so that, from Eq. (65), we have one theorem. I n carrying out any step of a reduction it will be convenient
(lq) (+/q)(pJlq) (Ptnlq). to write
= * (72)
If the first factor is ( - 1la), from Eq. (62) we have (Ila) = Iv
(-llq) = (-1)Q. (73) (-1iq) = (-1);
Otherwise ( 2 / q ) = (-1)p+1)41
(1Id = 1. (74) (a%d = ( a ~ q ) s(if d N ) (77)

If p j = 2, from Eq. (61) we have (abla) = (a@ (bIdM
(21d = (-1)
l(q+1) 141
(75)
Therefore to evaluate any (aiq) there remains the problem of evaluating
(a + klq) = ( 4 d P
( p l q ) for any two odd primes p and q. ( P l d = (-l)Q(nlP)R

depending on whether that step uses the unit, negative, double,
19. THERECIPROCITY
LAW square, multiplicative, periodic, or reciprocity rule. There is no
By examining many empirical results (such as those of Exercise 19), unique method of reduction. Thus
Euler, Legendre and Gauss independently discovered a most important
theorem. But only Gauss proved it-and it took him a year. I n terms of the (8117) = (2117)s = + I D ,
original Legendre Symbol we write or
Theorem 27 (The Reciprocity Law). I j p = 2P + 1 and q = 2Q +1 (8117) = (-9117) = (-1117)s = +IN,
are unequal primes, then
or
(PlP)(qlP) = (76) (8117) = (25117)p = (1117)s = + l u ,
or The simplest and most direct proof of the Reciprocity Law is perhaps
the following modification of a proof by Frobenius. It is based on Gausss
(8117) = (-26117)p = (26117)MN = (13117)MD
Criterion.
= (17113)R = (4113)p = + I a v . PROOF OF THEOREM 27. Let q = 2Q +
1 and p = 2P 1 be distinct+
Any path leads to the same answer, (8117) = 1, and this implies 17/86 - 1. primes. Let a be an odd integer satisfying 0 < a < p such that
Again, qa = pa +r (78)
(17147) = (47117)R = (13117)p = +I (as above). with r an even integer satisfying 0 < r < p . If y is the number of such a,
(17147) = (6411i)p = +ISo. by Eq. (64) we have
Therefore 4711723- 1 . (41P) = (--1l7.
A third example raises a new point. We have It follows from Eq. (78) that for each such a, the corresponding a is also
( 15/47) = (3147) (5147)M = - 1 (4713) (4715) odd, is unique, and satisfies 0 < a < q.
By symmetry
= -1(-113)(215)p = (215)N = -10.
(Pld = ( - V
But (15147) L -1(47115)RL -1(2115)pL -1,.
I n the second reduction we did not factor 15 and we applied the rules where y is the number of odd a such that
(77) to (47115) and (2115). But 15 is not a prime! Nonetheless we obtained
the correct answer. We will return to this pleasant possibility in Volume
pa = qa +r (79)
I1 when we study the Jacobi Symbol. Wow let us prove Theorem 20. with 0 < a < q, 0 < r < q, 0 < a < p , and with a odd. Again, for each
PROOF
OF THEOREM
20.
such a, a is unique.
If we now consider the function
(3112k + 1) = (12k + 113)R = (1/3)p = I U .
R ( a , a) = qa - pa
(3112k - 1) = -1(12k - 113)R = -1.(-113)p = IN
where a = 1, 3, 5, . - . , p - 2, and a = 1, 3, 5 , ... , q - 2, we see that
(3112k + 5) = (1% + 513)R = (-113)p = -IN. there are y of these R which satisfy 0 < R < p , and y of these R which
(3112k - 5) -1(12k - 513)R -1.(113)p -lo. satisfy -q < R < 0. Since there is no R = 0 (because a < p and p is a
= = =
prime), we see that there are y +y values of R such that
Therefore q = 12k f 113 - 1 and
-4 <R G p. (80)
q = 12k & 513 + 1.
But if R1 = qal - pal is one of these, then so is Rz = qaz - pa, where
We note, in passing, that Theorem 20 makes an assertion concerning
(319) for infinitely many q, while in the proof we need evaluate only finitely a1 + az = p - 1,
many Legendre Symbols. It is, of course, the Reciprocity Law, together
with Eq. (M), that brings about this economy.
EXERCISE 27. Verify the statements on page 38 concerning 17 f 1. and therefore
EXERCISE 28. Investigate the possibility of always avoiding the double Ri + R2 = p - 4.
rule, inasmuch as
For, the mean value of R1 and R2 equals the mean value of the limits of
(21q) = (-114)(4 - 2lq) Eq. (SO), - q and p . Therefore if R1 is even, and between these limits,
If so, it means that our original motivation, 412 f 1, is the one thing we so will Rz be even and between the limits. And likewise if al is odd and
do not need in determining qla 1. between 0 and p - 1, so is az . And similarly with al and a,.
46 Solved and Unsolved Problems in Number Theory From Perfect Numbers to the Quadratic Rcciprocity L a w 47
Therefore each R in Eq. (80) has a companion R in Eq. (80), given by Jf the number in each set is A , then PQ = y y + +
2A. Therefore we
Eq. ( S l ) , unless have another variation on the proof of Theorem 27.
al = a2 = ( p - 1)/2 = P , and al = a= Q. (82) EXERCISE 33. Examine the con~pnnions,JCq. (81),in scveral numerical
cases and verify that sometimes the y solutions of Eg. (78) choose their
Bat, since every a and a is odd, Eq. (82) cannot occur unless P and Q are
companions solely from the y solutions of Eq. (79) , while sometimes some
both odd. Conversely, if P and Q are both odd, there is a self-companioned
of the y companions of Eq. (78) are themselves from the set Eq. (78).
R q P - pQ P -Q
= =
20. THEPRIME
DIVISORS
OF n2 +a
given by Eq. (82) , which does satisfy Eq. (80). Now that we have completed the solution of the problem q[aQf 1,
Thus y + y is even unless P and Q are both odd. But so is P Q even,
we will lift our ban against Eulers Criterion and Definition 12. Henceforth,
unless P and Q are both odd. Therefore
(alq) and (f) are identical, mTi11 be designated the Legendre Symbol, and
may be written in either notation.
and Theorem 27 is proven. If q = 2Q +
1 is a prime which does not divide a, we now have at once
Gauss gave seven or eight different proofs of the Reciprocity Law. All of that
them were substantially more complicated than the one we have given-
and the first proof, as we have said above, took him a year to obtain.
qtn2 +a
Yet the given proof, based on Gausss Criterion, seems quite straight-
forward and simple. We will return later to this question-since we are
for some n, if (T) = 3-1, and
interested, among other things, in the reasons why some proofs are com-
plicated, and in the feasibility of simplifying them.
We may note that the proofs of Theorem 25 (Gausss Criterion) and of
Theorem 27 just given, are similar in strategy to parts of Dirichlets proof of for a n y n, if (,) = -1. The symbol (i)
~ we may evaluate by the rules
Eulers Criterion (page 35). I n both cases we multiply Q equations
together, and in both cases n-e set up c o mp a n io n s ~ x c e p tthat in
Eulers Criterion the companions are multiplicative, as in Eq. (46), while
in Theorem 27 they are additive, as in Eq. (81). Again, in both cases, the
self-companioned singularity (which may or may not occur) is the
critical point of the proof.
EXERCISE 29. Show that if the Q numbers a, in Theorem 24 are the num-
bers 1, 2, . . . , Q instead of the odd numbers, the theorem is still true. U b
EXERCISE 30. Modify Theorem 25 in accordance with the different set
5, 13, 17
(a)
of a , in the previous exercise. (For this different set and with the use of +1 1,
+2 I, 11, 17, 19
instead of (alq), this result is called Gausss Lemma.) Carry out the -2 1, 7, 17, 23
details of the new proof. +3 1, 7, 13, 19
-3 1, 11, 13, 23
EXERCISE 31. With the variation on Theorem 23 of the previous example,
+6 1, 5, 7, 11
carry out another proof of Theorem 27-with companions, etc.
-6 I, 5, 19, 23
EXERCISE 32. Consider Eq. (80) and show that each R such that p < R
can be put into one-to-one correspondence with an R such that R < - q . COMMEKT:
In each of these seven cases one-half of the primes divide
P
+
the numbers of the form n2 a, since +(24) = 8. (When we get to modulo Gauss Sums, and Dirichlet Series. We therefore postpone further considera-
multiplication groups, these seven sets of b will constitute the seven sub- tion of this conjecture until Volume 11.
groups of order four in the group modulo 24. Why the special role of b = l ? We offer, however, without further comment, a little table for the
Because 1 is the identity element of the group.) reader's consideration.
EXERCISE 35. Prove the conjecture you made concerning the prime
divisors of n2 - 5 in Exercise 19. Or, if your conjecture was erroneous, dis- a P.(lO,OOO) P.(lOO,OOO)
prove it. But if you haven't done Exercise 19, don't do it now. You already
know too much. +7 167 1238 9521
-__ -----_--
The reader no doubt asked himself, while reading Conjectures 11 and -2 157 1153 8888
12, why there should be more primes of the form n2 - 2 than of the form -- - _____
+
n2 1, and what the general situation would be for any form n2 a. + -5 148 1088
__
8579
With what he knows now the reader may begin, if he wishes, to partially
formulate his own answer. I n particular, from the table in Exercise 34, -3 120 850 6664
+
should there be (relatively) few primes of the form nz 6, or (relatively) +1 112 841 6656
many? - __
Dehition 14. By P,(N) is meant the number of primes of the form d a + +4 125
-
870 6517
+
for 1 5 n 5 N . If a is negative, and if for some n, n2 a is the negative +3 109 711 5426
of a prime, we will, nonetheless, count it as a prime.
Now that we have the Legendre symbol we can define the constants -6 91 643 5010
____-__
in Conjectures 11 and 12, and state a general conjecture of which these -7 68 440 3627
two are special cases. ---___-
Conjecture l Z 1 (Hardy-Littlewood). +2 68 446 3422
__
If a z -m2, $6 53 444 3420
-
+5 48 339 2567
where the constant ha i s given by the infinite product -4 n

L 2 2
-
~
h, =
*{
W
1 - (-alw)
L}
w - 1
-1 I 1 I 1
taken over all odd primes w. Here (-alw) is the Legendre symbol.
0 0 0
I 0
EXAMPLE :
From ( - l [ w ) = ( -l)(w-1)'2 we have
hi = (1 + 3 > ( 1 - 4>(1 + 6 > ( 1 + &)(1 - h ) ( 1 - A) * *
= 1.37281346 . * . ,
and we thus obtain Eq. (37) for primes of the form n2 1. + IL
But to evaluate such slowly convergent infinite products we will need

many things which we have not yet developed-Mobius Inversion Formula,
CHAPTER II
Chapter II
Chapter I1 ::THE
THE UNDERLYING
UNDERLYING STRUCTURE
STRUCTURE THE UNDERLYING STRUCTURE
The Residue
The Residue Classes
Classes as
as an
an Invention
Invention
The Residue Classes as a
The Residue Classes as a ToolTool
The Residue
The Residue Classes
Classesasas aa Group
Group
Quadratic Residues
Quadratic Residues 21. THERESIDUECLASSESAS AN INVENTION
Is the
Is the Quadratic
Quadratic Reciprocity
ReciprocityLawLaw aa Deep
Deep Theorem?
Theorem? In July 1801, Carl Friedrich Gauss of Braunschweig completed a book
CongruentialEquations
Congruential Equationswithwith aa prime
primeModulus
Modulus on number theory, written in Latin, and entitled Disquisitiones Arithme-
ticae. He was then 24, and largely unknown. He had been writing this book
Eulerso function
Euler's function
for five years. Upon publication, it was a t once recognized as a work of the
PrimitiveRoots
Primitive Roots with
with aa Prime
Prime Modulus
Modulus highest order, and, from that time until his death many years later, Gauss
Mp as
Mp asaa Cyclic
CyclicGroup
Group was generally regarded as the worlds leading mathematician. Since Gauss
TheCircular
The CircularParity
ParitySwitch
Switch was the director of the astronomical observatory a t Gottingen for 48 years,
PrimitiveRoots
Primitive Rootsand
andFermat
FermatNumbers
Numbers his death was recorded with appropriate accuracy: February 23, 1855 a t
ArtinsConjecture
Artin's Conjecture 1:05 a.m.
QuestionsConcerning
Questions ConcerningCyclic
CyclicGraphs
Graphs We should make it clear that his early reputation stemmed equally (or
perhaps principally) from quite a different source. On January I, 1801,
AnswersConcerning
Answers ConcerningCyclic
CyclicGraphs
Graphs Giuseppe Piazzi had discovered a minor planet in the general vicinity
FactorsGenerators
Factors GeneratorsofofMmMm predicted by Bodes Law. This planetoid was named Ceres, but, being
Primein
Prime inSome
SomeArithmetic
ArithmeticProgressions
Progressionsand
andaaGeneral
GeneralDivisibility
Divisibility only of 8th magnitude, it was lost 40 days later. From the data gathered
Theorem
Theorem during these 40 days, and with new methods of reducing these which he
Scalarand
Scalar andVector
VectorIndices
Indices devised, Gauss managed to relocate the planet. And since celestial me-
TheOther
The OtherResidue
ResidueClasses
Classes chanics was the big thing in mathematics a t that time-say as topology
TheConverse
Converseof ofFermat's
FermatsTheorem
Theorem is today-this relocation too was regarded as a work of first magnitude.
The
But if fads in mathematics change qu?ckly, certain things do not. Of these
SufficientConditions
Sufficient ConditionsforforPrimality
Primality two works of Gauss in 1801, his book is still of first magnitude, and Ceres
is still of eighth.
At that period, France was once again the leading center of mathematics
with such luminaries as Lagrange, Laplace, Legendre, Fourier, Poncelet,
Monge, etc., and consequently Gausss book was first translated into
French (1807). It is perhaps through this translation that the work of
Ch. Fr. Gauss (de Brunswick) became known to the mathematical
world. It is said that Dirichlet carried his copy with him wherever he
went, that he even slept with the book under his pillow, and that many
years later, when it was out of print, he regarded it as his most precious
possession-even though it was coinpletely in tatters by then. For ap-
proximately $9.50 one may purchase a 1953 (Paris) reprint of this transla-
51
52 Solved and Unsolved Problems in Number Theory The Underlying Structure 53
tion, with a n unsubstantial cover, and with pages so well oxidized that it Could so much formal similarity be fortuitous? And if not, what could be
may well attain this "Dirichlet Condition" even if it encounters a more its significance?
casual reader. There also exists a German translation(( l889), but, a t this Where we first came upon such expressions we know well enough-if
writing, the book is still not available in English. N = 2n-1F is to be perfect, the sum of divisors 1 2 + +
. . . 2"-' = +
We ask now, what was in it; and why did it make such a splash? Well, 2" - 1 must be a divisor of N , and must also be a prime. But 23[211- 1,
many new things were in it-Gauss's proof of the Reciprocity Law, his and therefore MI1 was not a prime, etc. It is another question, however, if
extensive theory of binary quadratic forms, a complete treatment of we ask why the expressions AIB - C should be so persistent.
primitive roots, indices, etc. Finally it included his most astonishing dis- We should make it clear, a t this point, that though we have followed one
covery, that a regular polygon of F, = 2'" +1 sides can be inscribed in path in the previous chapter, that starting from the perfect numbers, much
a circle with a ruler and compass-provided F, is a prime. other ground had been gone over by this time. I n particular, consider
But the most immediate thing found in Gauss's book was not one of Gauss. Gauss could compute as soon as he could talk-in fact, he jokingly
these new things; it was a new way of looking a t the old things. By this claimed he could compute even earlier. He rediscovered many of the
new way we mean the residue classes. Gauss begins on page 1 as follows:
"If a number A divides the difference of two numbers B and C, B and C
I theorems given in the previous chapter before he had even heard of Fermat,
Euler or Lagrange. It is clear that no computing child could reinvent
are called congruent with respect to A , and if not, incongruent. A is called anything as esoteric as the perfect numbers, and therefore Gauss could
the modulus; each of the numbers B and C are residues of each other in the not have followed the path which we have sketched. To the Greeks a
first case, and non-residues in the second." divisor of a number, other than itself, was a "part" of the number; and
for a perfect number, the whole was equal to the sum of its parts. Such a
Does i t seem strange that Gauss should write a whole book about the
Greek near-pun could well engage the classicists of the Renaissance, but
implications of
would not be likely to occur to a self-taught Wunderkind.
A(B - c? (83) What was available to Gauss was such material as
1 -- .142857142857142857 . * *
7
It surely is not clear a priori why Eq. (83) should be worthy of such pro-
tracted attention. I n fact, these opening sentences are completely un-
and = .076923076923076923 . . . .
motivated, and hardly understandable, except in the historical light of the
previous chapter. But in that light, the time was ripe-and even overripe- Now if 3 is a periodic decimal with a period 6, then since 1 = 0.999999 . . . ,
for such an investigation. We will review four aspects of the situation then it means that 71999999, or
existing.
71106 - 1.
(a) First, it will not have escaped the reader that we were practically
surrounded by special instances of (83) in the previous chapter. Thus Likewise for any prime p , not equal to 2 or 5, we find
Fermat's Theorem 13] reads: p110*' - 1.
p1uP-l - 1, Therefore, say,
and his Theorem 11 : 13110'' - 1.
q(2p - 1 4 2plq - 1,
But we have just seen that also has a period of 6, so that
can go it one better by having both hypothesis and conclusion in that 131106 - 1.
form. So likewise Euler's Criterion:
From the foregoing theory we know that
QIN' - u f~ qluQ - 1,
p110'p-'"2 - I means (IOlp) = +I
and his Theorem 19: and that
q12' - 1 -+81q - 1 or 81p - (-1). (10113) = (-3113)p = (3113)MN = (13[3), = +1pu.
54 Solved and Unsolved Problems in Number Theorg T h e Underlying Structure 55
It is clear, however, that whether Fermat and Euler were interested in (d) A final, and most important point. Fermats Theorem quickly let
perfect numbers-and 231211 - I ; or Gauss was interested in periodic its power be seen. Thus
decimals-and 131106 - 1, the basic underlying theorems are identical, fig = 6 << ~ 1 9= 128
and AIB - C arises in either case. was most impressive. Similarly Euclids Theorem 5 and its immediate
(b) There is another case of persistence in the previous chapter. On consequence Theorem 6 have, by their constant use, become quite in-
pages 24, 27, 35, etc., we are saying, repeatedly, for some integer, Q, I,, dispensable. Yet can we say, a t this point, that we can see clearly the
K , K2 etc., and that seems almost paradoxical a t first. Isnt number theory source of this power and this indispensability? There is suggested here the
an exact science-dont we care what Q, L, etc., are equal to? The answer existence of a deeper, underlying structure, the investigation of which
is, generally,* no. If we are interested in AIB this implies some integer X deserves our attention.
such that B = A X , but which integer is quite irrelevant. We want then, in (b) , an algebra of ambiguity; in (c) , an arithmetic of
It is instructive to examine the additive analogue of divisibility, A < B. remainders; and in (d) , an interpretation in terms of an underlying struc-
This implies a positive X such that B = A +
X , but which X is again ture. It is the merit of the residue classes that they answer all three of these
irrelevant. If this were not the case, Analysis would be quite impossible. demands.
It is difficult enough to show that a certain quantity is less than epsilon- We could, it is true, have introduced them earlier-and saved a line here
it would be totally unfeasible if we always had to tell how m u c h less. The and there in the proofs. But History did not introduce them earlier. Nor
analyst embodies this ambiguity in X by working with classes of numbers, would it be in keeping with our title, Solved and Unsolved Problems,
- E < X < E , and any X in the class will do. Likewise in divisibility theory for us to do so. To have a solved problem, there must, first be a problem,
we should consider the advantages of working with classes of numbers, and then a solution. We could not expect the reader to appreciate the
which would embody the ambiguity presently in question. solution if he did not already appreciate the problem. MoreoTrer, if we have
A variation on this theme concerns the algebra of such ambiguity. On gone on a t some length before raising the curtain (and perhaps given undue
page 27 we square one ambiguous equation, 2 = N 2 - Kq, to obtain a attention to lighting and orchestration) it is because we thought it a matter
second, 2 = N 4 - K2q. On page 36 we substitute the ambiguous N = of some importance to analyze those considerations which may have led
sq + a, into qIN2 - a to obtain q1a; - a. Such persistent, redundant, and Gauss to invent the residue classes. Knowing what we do of Gausss great
rather clumsy algebra virtually demands a new notation and a new algebra. skill with numbers, and while we can not say for certain, the consideration
(c) Again, consider the arithmetic of page 26 : most likely to have been the immediate cause of the invention would seem
to be item (c) above.
1671283- 1,
EXERCISE 36. Using the results of Exercise 35 and of Exercise 18, deter-
I
or the seemingly impossible operation, mine the odd primes p = 2P +
1 # ?such that l / p has a decimal expan-
sion which repeats every P digits. The period of some of these primes may
- 1,
32070004j59~21603500227~
be less. Thus & = .027027 . . . does repeat every 18 digits, but its period
of Exercise 7. The first seems a little long and the second virtually im- is 3.
possible-but only because the dividend, and therefore the quotient is so
22. THE RESIDUECLASSES AS 9 TOOL
large. But IW ?aid that in questions of divisibility the quotient is irrelevanf,
that only the remainder is of importance. Thus, if Definition 15. If a, b, and c are integers, with a > 0, and such that
b = qa + r, alb - c, (84)
we may write, equivalently,
divisibility depends only on r. And r is less than a. And a, even in the
second case, is not too largc to handle. What we want, then, is an arithmetic b =c (m oda), (85)
of remainders.
and the latter is read b is congruent to c modulo a. We may also say b is
* An important exception will be discussed in Sect. 25. a residue of c modulo a. Conversely, given Eq. ( 8 3 ) , we may write Eq.
(84).If b is not congruent to c modulo a, we write coejkients. That is, f is a sum of a finite number of terms, nu% . . , each
b +c (moda). (86) being a multiple of a product of powers of the variables. Here n is an integer
and a, /3, . are nonnegative integers. I f al , b1 , c1 , . . are integers, and i f
If
b = pa + r, (87)
then b =r (moda)
al = a2, bl = b z , c1 = c 2 , - . . (mod M ) (W
independently of the value of q. As q takes on all integral values, . . . , -2, then
-1, 0, 1, 2, . . , each such b is congruent to r, and all such b form a class
of numbers which we call a residue class. a is called the modulus. Nz = f(a2 , bz , cz , * . ) = N 1 (mod M ) . (91)
EXAMPLES
: PROOF.
The reader may easily verify that if Eq. (90) is true, then so are
2 = 1 (mod 23). al + bl = a2 + b2 (m odM ),
229= - 1 (mod 59).
a1 - bl = az - bz (m odM ), (92)
(Fermats Theorem) albl = a2bz (mod M ) .
a +0 (mod p ) + aP1 = 1 (mod P I . By induction, it is clear that any finite number of these three operations
(Eulers Criterion) may be compounded without changing the residue class, and since any
polynomial, Eq. (89), may be thus constructed, the theorem is true.
N2= a (mod p ) ++ =1 (mod p ) .
Corollary. If f ( a ) i s a polynomial in one variable, then
For any a > 0, and any b we can always write Eq. (87) with 0 5 r < a.
Corresponding to a modulus a, there are therefore a distinct residue classes, a = a (mod M ) implies f ( a ) = f ( a ) (mod M ) .
and the integers 0, 1, 2, . . . , a - 1 belong to these distinct classes, and This simple theorem allows us to use the residue classes as a tool for
may be used as names for these classes. Thus we may say 35 belongs to those arithmetic and algebraic problems which we discussed on page 54.
residue class 3 modulo 16. Consider some simple examples.
Congruent to is an equivalence relation, in that all three characteristics (a) To verify that 71106 - 1, we may write
of such a relation are satisfied. Specifically:
Rejlexive. For all b, (88a) lo6 = 36 (mod 7)
b=b (moda). since 10 = 3. But \
Symmetric. (88b) 3 = 2 and thus lo6 = 36 = 23 = 1 (mod 7).

Therefore
b =c (mo d a) implies c 3 b (moda).
71106 - 1.
Transitive. ( 88c)
(b) To determine if 167 divides M83 , we may proceed as follows:
b=c (moda) and c = d (moda)
2* = 256 = 89 (mod 167)
implies b = d (mod a ) .
:. 16
2 = 8g2 = 7921 = 72 (mod 167)
All the numbers in a residue class are therefore congruent to each other
(mod a ) . :. 232 = 72 = 5184 = 7 (mod 167)
The utility of residue classes comes from the fact that this equivalence :. 2 64
= 49, and 26 7 = 49.8 = 58 (mod 167)
is preserved under addition, subtraction and multiplication. Thus we have ... 283 =
67 16
2 . 2 = 58.72 = 4176 = 1 (mod 167)
Theorem 28. Let f(a,b, c, . . . ) be a polynomial in r variables with intbger .: 16712 - 1.
T h e Underlying Structure 59
58 Solved and Unsolved Problems in Number Theory
Therefore if
The advantage of the congruence notation is clear. What we really want to bc = bd, nbc = nbd, or c = d (mod a ) .
know here is whether 283and 1 are in the same residue class, and in our
computation of 283we continually reduce the partial results to smaller Equation (93) is the key to our next topic, the Residue Classes as a
members of the residue class, thus keeping the numbers from becoming Group.
unduly large. EXERCISE 37. Prove Theorem 22, page 35, and Theorem 211 , page 35, in
(c) Aside from advantages in the computation of results, there is also the congruence notation.
an advantage in their presentation. Thus to show that G411232 1, the + EXERCISE
38. Verify that
presentation
67004 17 18231M911.
641 14294967297 23. THE RESIDUECLASSESAS A GROUP
lacks the property of being easily checked mentally. But consider In the previous sections the integers were the sole objects of our atten-
(mod G41). tion, and, as long as we considered the residue classes merely as a tool,
640 = 5.128 = 5.2 E -1
this remained the case. We now consider a system of residue classes as a
.: 54.228E 1 (mod 641). mathematical object in its own right, and, in particular, we study the
multiplicative relationships among these classes.
But 54 = 625 E -16 = -24 (mod 641).
For a modulus nt there are m residue classes, which we designate 0, 1, . . . ,
.: -2 32
E 1 (mod 641) m - 1, the ath class being that which contains the integer a. The system
of these m classes is therefore not infinite, like the integers, but is a finite
or 232 + 1= 0 (mod 641).
system with m elements. By the product of two classes a and b we mean the
Here the arithmetic is easily verified mentally. class of all products albl where
(d) The proofs of some of the theorems in the previous chapter could al = a and bl = b (modm).
have been written more compactly in the new notation. For example, on
page 27, if q1N2 - 2, then By Eq. (92) all these products lie in a residue class, say c, and we write
N2= 2 (mod q ) ab = c (modm).

For example, for m = 7, we have the following multiplication table:
and directly we may write
24 N2Q 1 (mod 4 ) . 3 4
__
Thus by setting up an algebra of ambiguity (page 55) we have simul- 0 0
__
taneously rid ourselves of the some integer K (page 27) which is
3 4
clearly redundant and merely extends the computation. __
But to complete our algebraic tools we need division also, and for this we 6 1
have __ ab = c (mod 7).
2 5
Theorem 29 (Cancellation Law). If bc = bd(mod a ) and (b, a ) = 1 then _-
c = d(mod a ) . 5 2
This is only a restatement of Theorem 6 in the new notation. We will _-
reprove it using this notation. 1 6
PROOF.If ( b , a ) = 1, from Eq. (7) , page 9, we have
nb = 1 (mod a ) . (93)
If ( a , m) = 1 and a = a, (mod m),we have (al , m ) = 1. Thus we that

may say that the residue class a is prime to na. Now if ( a , m) = 1 we have d'a = 1
an a' and 7n' such that
for every a.
a'a +
m'm = 1 (94)
Thus the + ( m )residue classes prime to m form a group under the binary
and conversely. Thereforc
operation multiplication modulo m. The postulates (B) and (C) are
a'a = I (mod m) (95) trivially true, while closure (A ), from Eq. (97), and inverses (D) , from
Definition 16. We may call the a' and a in Eq. (95) the reciprocals of Eq. (96), both stem from Eq. (94), that is, from Euclid's Theorem 5.
each other modulo m, and write Definition 18. If the operation in a group is commutative, that is, if
a-1 = a' (mod m ) . (96) ab = ba
We may therefore characterize the +(nz) residue classes prime to m as
for each a and b, the group is called Abelian. If the number of elements in
those which possess reciprocals. If ( a , m) = ( b , m ) = 1, then so
a group is finite, the group is finite, and the order of the group is the number
is (ab, m) = 1, by Theorem 5, Corollary. I n fact, since
of elements.
a-'ab-'b =1 (mod m ) ,
Definition 19. The group of +(m) residue classes prime to m, under
we have explicitly multiplication modulo m, we call a modulo multiplication group, and we
E a-'b-' (mod m ). (97) write it 3n, . It is a finite, Abelian group of order + ( m ).
The theory of finite groups is a large subject, into which we shall scarcely
We will have occasion, say in Eqs. (103a) and (104a) on page 66, and
enter. We shall confine ourselves primarily to 3n,. Nonetheless, there is a
in Eq. (136) on page 100, to calculate the reciprocal of a modulo m. This
value here in introducing the more abstract Definition 17, and that lies
we do by obtaining Eq. (94) from Euclid's Algorithm as on page 9.
in the economy of this definition. I n any theorem, say for m,,, , which we
Equivalently, one may utilize the continued fraction (12) on page 12
deduce from these four postulates, we have a certain assurance that re-
with the term 1/qn omitted. This fraction we evaluate by the method on
dundancies and irrelevancies have not entered into the proof. Pontrjagin
page 183 below. The denominator so obtained, or its negative, is the
reciprocal of a modulo b. This follows from the analogue of Eq. (271). puts it this way:
"The theory of abstract groups investigates an algebraic operation in
Definition 17. A group is a set of elements upon which there is defined its purest aspect."
a binary operation called multiplication which
(A) is closed, that is, if Several of our foregoing theorems have a simple group-theoretic in-
terpretation. We will illustrate thew using the multiplication table for m7 .
c = ab,
then c is in the group if a and b are ; and 1 2 3 4 5 6
___ -__
(B) is associative, that is, 2 4 6 1 3 5
( a b ) c = a(bc)
___ ___
3 6 2 5 1 4
for every a, b and c. ___
Further, 4 1 5 2 6 3
_-- -_-
(C) the group possesses an identity element (write it 1) 5 3 1 6 4 2
such that _ _ _ -_-
la = a 6 5 4 3 2 1
for every a ; and also (Note that the row and column headings are omitted, since the first row
(D) it possesses invcrsc. elements (write these a-') such and column also serve this purpose.)
Theorem 17 says that if EXERCISE

40. If (a, m) = 1, show that
-
aa, = r , (mod 7) a 1 ad- (mod m) (99)
the r , are a permutation of the a , -that is, each row in the table contains Further, if ( a , m ) = g, a = cug, and m = pg,
every element. But this is true for every finite group.
Again, Theorem 22 says that then
xu, = a (mod 7) and

are integers that satisfy
has a unique solution-that is, each column in the table contains every
element. Again, this is true for every finite group. aa + mm = g.
Since in an Abelian group the rows and columns are identical, we now 24. QUADRtlTIC RESIDUES
realize that TheGrem 22 is essentially a restatement of Theorem 17. We
have seen previously that Fermats Theorem 13 may be deduced either Definition 20. Any residue class lying on the principal diagonal of the
from Eulers Theorem 13, or from Eulers Theorem 211 , and we now note 312, multiplication table is called a quadratic residue of m. That is, a is a
that the corresponding underlying Theorems 17 and 22 are also equivalent. quadratic residue of m if
Eulers Theorem 14 says that ( a , 7) = 1 implies x2 = a (m odm )
a6 E 1 (mo d7 ). has a solution x which is prime to m. If ( a , m) = 1, and a is not a quadratic
Again, for every group of order n, an = 1 is valid for every element a. residue of m it is called a quadratic nonresidue. When the meaning is clear,
I n fact, the whole subject of finite group theory may be thought of as a we will sometimes merely say residue and nonresidue.
generalization of the theory of the roots of unity. It is not surprising, then, EXAMPLES
: From
that Fermats Theorem plays such a leading role, seeing, as we now do,
that it merely expresses the basic nature of any finite group. 1 3 7 9
The three theorems just discussed hold for m, whether m is a prime or ____
not. But Eulers Criterion does not generalize so simply. This criterion 3 9 1 7
(mod 8) ____ (mod 10)
states that 7 1 9 3
- 1
ad(P)/2 =
(mod p ) - n 2 = a ____
(mod p ) . (98)
9 7 3 1
/ . . , ,
But consider m = 8 and m = 10. In both cases ++(m)= 2. Now for the
modulus m = 10, the implication (98) still holds. But for m = 8, we have <
we see that 8 has 1 as its only quadratic residue, while 10 has both 1 and 9.
32 = 1 (mod 8) From the table on page 61, 7 has 1 , 2 , and 4 as quadratic residues.
while From Definition 12, page 33, it is clear that if p t a , a is a quadratic
n2 = 3 (mod 8)
has no solution. This is a difference which we shall investigate. It is as-
residue of p , or is not, according as
(f)+ =
(9 =
1 or - 1 according as a is or is not a square modulo p .

+1 or -1. Or, we may say,
sociated with a particular characterization of the MI,, groups for every m

which is prime, and for some m which are composite; namely, that these Theorem 30. Every prime p = 2 P +
1 has exactly P quadratic residues,
groups have a property which we shall call cyclic. and therefore also, P quadratic nonresidues.
EXERCISE 39. Write out the multiplication tables for 3% and m ~(If. PROOF.
I n the proof of Eulers Criterion on page 36 we showed that if
you use the commutative law, and the generalized Theorems 17 and 22
mentioned above, you will save some arithmetic.) (:) = +1 there are exactly two incongruent solutions of x2 = a (mod p ) .
64 Solved and Unsolved Problems in Number Theory T h e Underlying Structure 65
Since each of the 2P classes 1, 2, * . , 2P has a square, there are exactly are clearly somewhat more involved than those of the form
P distinct squares.
qlaQ- 1,
Delinition 21. If
(9
- = 6(mod p ) for either solution of
+1 we write since aQ - 1 is a specific number, while in N 2 - a, N is unspecified and
may range over 2Q possibilities. Therefore it is not surprising that the
x2 = a (mod p ) . For a
modulo p .
= 0, fi = 0. For
* (9
= -1, 6does not exist Quadratic Reciprocity Law lies a little deeper than does Legendres Re-
ciprocity Law.
But even in the best of Gausss many proofs, the theorem still seemed
EXERCISE 41. For every modulus m, the product of two residues is a far from simple. It is of some interest to analyze the reasons for this.
residue, and the product of a residue and a nonresidue is a nonresidue.
For every prime m and for some composite m, the product of two non- (a) I n his simplest proof, the third, Gauss starts with the Gauss
residues is a residue, while for other composite m, the product of two Lemma, (Exercise 30). From this, and a page or so of computation, he
nonresidues may be a nonresidue. derives another formula. If a is odd:
EXERCISE 42. Theorem 30 may be generalized to read that the number

of residues is ++(m)for some composite m, but not for others.
EXERCISE 43. For which primes p = 24k +
b does mpcontain g,
where
&, or ? Examine all eight possible combinations of the existence
and the nonexistence of these square roots.
M = q3.
z= 1
25. Is THE QUADRATIC LAW A DEEP THEOREM?

RECIPROCITY Here [ ] is the greatest integer function, defined on page 14. Now it ap-
We interrupt the main argument to discuss a question raised on page pears that with Eq. (100) Gauss has already dug deeper than need be. What
46. The Quadratic Reciprocity Law states that for any two distinct primes, we need is the parity of the sum, y + y, (page 46). The individual ex-
p = 2P +
1 and q = 2Q +
1, p and q are both quadratic residues of each ponent, M , is not needed, and, if it is obtained nonetheless, it is clear that
this is not without some extra effort.
other, or neither is, unless P Q is odd. In that case, exactly one of the primes
is a quadratic residue of the other. The theorem follows a t once from (b) Gauss then proceeds to prove that
Theorem 27 with the use of Definition 20 and Eulers Criterion.
The Quadratic Reciprocity Law is often refered to as a deep theorem.
We confess that although this term deep theorem is much used in books i
on number theory, we have never seen an exact definition. In a qualitative by the use of various properties of,the [ ] function. Here we see irrele-
way we think of a deep theorem as one whose proof requires a great deal vancies. What has the [ ] function to do with the Quadratic Reciprocity
of work-it may be long, or complicated, or difficult, or it may appear to Law? Later Eisenstein simplified the proof of Eq. ( l O l ) , but only by bring-
involve branches of mathematics the relevance of which is not a t all ap- ing in still another foreign concept-that of a geometric lattice of points.
parent. When the Reciprocity Law was first discovered, it would have This is all very nice theory-but it all takes time.
been accurate to call it a deep theorem. But is it still? (c) Finally there is a point which we may call abuse of the congruence
Legendres Reciprocity Law (so named by him), involves neither the symbol. We have shown many uses of the notation, = (mod p ) . But
concept of quadratic residues, nor the use of Eulers Criterion, as we have this symbol may also be misused. Suppose we write Eq. (78) as follows:
seen. With the simple proof given on page 45, we would not consider it a i
deep theorem. qa = r (mod p ) , ( 102)
u.
Now divisibility questions of the form
and inquire as to the number of odd as for which r is even. There are three
q\N2 - u things wrong with such an approach.
(1) We are interested not in one group m, , but in the interrelation The algebra here is so much like ordinary algebra because the residue
between t x o groups m, and m, , and, for this, the congruence notation is classes modulo a prime form a field, just as the real or rational or complex
not helpful. numbers form a field. Thus, just as group theory applies to m,,, so does
(2) There are no even and odd residue classes. If a is even, then field theory apply here. An important theorem in field theory states that
a + p = a is odd. an nth degree polynomial can have at most n roots.
(3) Most important is the following. The concept congruent to is Theorem 31. A t most n residue classes satisfy the equation:
of value when, (as on page 54), we dont care what the quotient is. But
in Eq. (78), f(z) = a,zn + an-lzn-l + . . . + a. = 0 (mod p ) (105)
pa = pa + r, with a, f 0 ( m o d p ) .*
the quotient a, for the divisor p , is also a coefiient of p in evaluating PROOF.Let Eq. (105) have n roots, z1, x 2 , . . . , zn . Dividing f ( z ) by
( p l y ) . And the quotient a is a coefficient of y for (ylp). This is precisely x - z1 we obtain f ( z ) = fl(z)( z - zl) +
c1 . But since p l f ( z l ) we find
where the reciprocity lies, and, if we throw it away, as in Eq. (102), we p i c l . Therefore
must work the harder to recover it. f(z) = f1(z)(. - 21) + lip.
EXERCISE
44. Evaluate (13117) by Eq. (100). Compare page 44.
Repeating this operation with f l ( z ) , then fZ(x), etc., we obtain
EQUATIONS
26. CONGRUENTIAL WITH PRIMEMODULUS
A
f(z) = a,(z - z l ) ( x - z2) ... (z - 2,) + pg(z)
I n Sects. 23 and 24 we developed reciprocals and square roots modulo m.
With these we may easily solve the general linear and quadratic con-
for some polynomial g(z) . Xow if there were an n + 1st root zn+l, not
congruent to one of the others, we would have
gruential equations for a prime modulus. These are
0 = f(z,+d = a,(zc,+l - x l ) ( x,,+~- z2) . . . ( xn+l - zn) (mod p )
ax +b =0 (mod p ) (pta) (103)
Therefore, by Theorem 6 , Corollary, a, = 0 (mod p ) , contradicting the
and
hypothesis.
+ +
ax2 bx c =0 (mod p) ( p t a ) . (104) We will use this theorem later when we investigate primitive roots.
The reader may easily verify that the solutions are the same as those given We could have used it earlier, together with Fermats Theorem, to prove
in ordinary algebra, that is, Eulers Criterion.
x = -a& (mod p ) (103a) If N 2 = a (mod y) , then N2Q= aQ (mod y) and, by Fermats Theorem,
a = 1 (mod y) . The converse is the more difficult. But from Theorem 30
and
there are Q quadratic residues. Therefore, from what we have just shown,
x = (2a)-( -b f 4-c) (mod p ) . (104a) there are Q solutions of aQ - 1 = 0 (mod q ) . But by Theorem 31, there
Therefore, as in ordinary algebra, Eq. (103) has precisely one solution, can be no other solutions. Therefore aQ = 1 (mod q) implies N 2 = a (mod
while Eq. (104) has 2, 1, or 0 solutions depending on whether a) *
( b 2 - 4aclp) = +1, 0, or -1. If p is not a prime, in Theorem 31, there may be a greater number of
EXAMPLES :
solutions. (Where does the proof break down?) Thus
(a) 3z + 2 = 0 (mod 7). z2 = 1 (mod 24)

Since 3- = 5, x = - 10 = 4 (mod 7 ). has 8 solutions, and so does
(b) 3 x 2 + 42 +1 = 0 (mod7). x2 = z (mod 30).
Since b2 - 4ac = 4 is a quadratic residue of 7, with square roots 2 and 5,
we have z = 6-( -4 f 2) = 6 or 2 (mod 7 ). The equation z2 = z (mod m) is particularly interesting, because in any
(c) 22 + +
32 2 = 0 (mod 7). * Since X P = X , X P + ~= 2 2 , ctc., for every z (mod p ) , any polynomitll of order higher
Since 9 - 16 = 0 (mod 7) there is only one solution, namely 1 (mod 7). than p - 1 may be reduced t o one of order not higher than p - 1.
68 Solved and Unsolved Problems in Number Theory The Underlying Striicti~re 69
field the two roots, 0 and 1, are the identities for addition and multiplica- differences. For this analysis we will want a better knowledge of Euler's
tion respectively. If nz is divisible by more than one prime, we shall see that function.
Our first result is
z2 = z (mod m) (106)
Theorem 32 (Euler). If
has more than 2 solutions, and that each one may serve as an identity
element in a multiplicative group. Thus N = p;"'pi2 . . . P2, (107)
z2 = z (mod 10) then
has 4 solutions 0, I , j,and 6. I n addition to the set mm,o,

of elements 1, 3 , 7 ,
and 9, which form a group with 1 as the identity, so likewise 2, 4, 6, and
+(N) = N (1 - ;-
)(I - $) . . . (1 - k). (108)
8 form a group modulo 10 with G as the identity, and 0 and 5 form groups I n the proof of Eq. (108) the main work is done (and constructively) by
of one element each, by themselves.
Theorem 33. If A > 0, B > 0, and ( A ,B) = 1, the A B numbers
EXERCISE
45. Show that
m = Ab + Ba
52
2 . ~ ~ + +5 =0 (mod 7)
with a = 0, 1, . . . , A - 1, and b = 0, 1, . . . , U - 1, belong to distinct
has no solution. residue classes naodulo AB. Further, if in nz, the a's are co?LJned to the +(il)
EXERCISE
46. Find the 8 solutions of numbers prime to A , and the b's to thP + ( B ) numbers prime to B , then the
corresponding + ( A ) + ( B )numbers m are all prime to AB.
x2 = z (mod 30)
PROOF.If
and shorn that corresponding to each solution there is a multiplicative
group of residue classes, modulo 30, with that solution as the identity.
Ab, + Bal = Abe + Ba:, (mod i l B )
EXERCISE47. Just as in Exercise 40, Eq. (%), we have a n explicit then Ab, = A bl (mod B ) and Ba, = Bas (mod A ) . But since ( A ,B ) = 1'
formula for a reciprocal, 6' modulo m, so, for some prime moduli, we by Theorem 29, bl E bz (mod B ) and al = a? (mod A ) . Furthermore'
have an explicit formula for a square root. Show that if p = 4m 3, + p = Ap +Ba = AP (mod B ) .
and ( a l p ) = +1, then 6E a"" (mod p ) . I n particular Since ( A , B ) = I, and if ( p , B ) = I , we have p prime to B. Likem-ise
4"' = 4 (mod 8k + 7). if ( a , A ) = 1 , p is prime to A . Therefore if ( a , A ) = (0,B ) = 1, p i s
prime to AB.
Also show Corollary. If A > 0, B > 0, and (A, B) = 1, then
ah + a-' = 4 + 1)
(mod8k +(AW = +(A)+(B). ( 109)
where a is any quadratic nonresidue of the prime 8k + 1. Thus we may PROOF. The + ( A ) + ( B )numbers p just indicated are prime to A B , and
compute 4 explicitly for all the prime moduli for which it exists. not congruent modulo A B . Furthermore, each such p is congruent to
exactly one integer satisfying 0 < z < AB. KO other of the A B numbers
27. EULER'S+ FUNCTIOU m = Ab + B a arc prime to AB, for if ( a , A ) # I , then m is not prime to
On page A2 we noted that while certain theorems for m, with m A , nor therefore to AB. Similarly for h and B. This proves Eq. (109).
a prime, could he extended to all mm,or even to all finite groups; others, PROOF OF THEOREM 32. If N 1 = p;I, the numbers 5 N1 and not prime
such as Euler's Criterion, could be extended to m m for some composite to N , are the multiples of p , 5 N 1 , Since thcre are pP'-' of these, we have
m, (say m = 9, l o ) , hut not for others, (say m = 8, 12). I n Exercises
41 and 42 there were closely rclatcd cxtcnsions, again valid only for some
composite m. Tikcwise, back in Exercaise 25 there was such an extension.
We are concerned now with the underlying structural reasons for these and by applying Eq. (109) n - 1 times we ohtain Eq. (108).
Another important result concerning +( N ) is wherein the infinite product on the right is taken over all primes. This
identity, in the hands of Riemann and others, led eventually to a proof of
> 0,
Theorem 34 (Gauss). If N
Theorem 9. If s = 1, the harmonic series on the left, 1 + + +
$ ... ,
diverges. If there were only a finite number of primes, the product on the
right would remain finite and yet equal to the series on the left. This con-
where the sum on the left is taken ouer all positive divisors d of N . tradiction gives another proof of Theorem 8. Again, if s = 2, we have
I~XAMPLE:N = 341 has four positive divisors, 1, 11, 31, and 341.
+(I) + +(11) + +(31) + +(341) = 1 + 10 + 30 + 300 = N .
PROOF.
Consider the equation so that if Theorem 8 were false we would have T equal to a rational num-
ber. This is known to be false, and if this latter does not already assume
N Theorem 8, we have still another proof.
(z,N)= -
d (c) Equation (108) also leads to mean value theorems for + ( N ) and
+ ( N ) / N as N 4 00, and to an interesting relationship between + ( N )
where d is a positive divisor of N and x can be 1, 2, . . . , N . Any solution
and a ( N ) , the sum of the positive divisors of N .
x of Eq. (1 11) must be a multiple of N N
- ,z = k - , where 1 5 k 5 d . Further (d) Theorems 32 and 34 have a relationship, via the so-called Mobius
d d Inversion Fornzula, which has an important generalization.
any such x will be a solution if and only if (k,d ) = 1, since ink nd = 1 + But we shall follow none of these diverging leads a t this time. What is
(3+
N
implies nz k- nN = - and conversely. There are therefore +( d ) solu- now in order is a deduction of primitive roots using Theorem 34.
d
tions. Since every 1 5 z 5 N satisfies an Eq. ( 111) for one and only one d , EXERCISE 48. Since (a, N ) = 1 implies ( N - a, N ) = 1, + ( N ) is even
we obtain Eq. ( 110). if N > 2 .
Theorems 32 and 34 could lead us off in several dircctions. Thus EXERCISE 49. Verify Theorem 34 for N = 561.
(a) From Theorem 32, Euler proved Euclids Theorem 8 as follows. If
EXERCISE 50. Verify Theorem 34 for N = 30. What is the relationship
211 = 2 . 3 . 5 , . . Pn between this partition of 30 and that of Exercise 46? HINT:Compare the
proof of Theorem 34 with the membership in the eight groups.
and if there mere no primes <M other than 2 , 3, . . . , p , we would have
EXERCISE 51. Find several multiplicative groups modulo 561 other
+(M) = (
1. But + ( & I ) = A 1 1 - -
;)(1 - ;) . . (1 - ); > 1. than m661.
On the other hand, we now have an upper bound: ROOTSWITH
28. PRIMITIVE A PRIME~ \ ~ O D U L U S
For every a prime to m
-
= 1 (mod i n ) ,
As n 3 m, we see that
but for some a, a smaller exponent, s, may suffice for
u8 E 1 (modm)
to be satisfied. Thus for any quadratic residue of m, (if m > 2 ) we have
decreases monotonically and, if wc are investigating T ( N ) we are led to -1
a9()12 = (m odm ).
the question of estimating thc right side of Ey. (112).
( b) Perhaps it was such a consideration which led Euler to his famous Definition 22. If (a, m) = 1 and e is the smallest positive exponent such
identity : that
ae = I (m odm ) (114)
(113) vie say a is of order e modulo m.
Ex.IruPLE:: If u = 10 and m is a prime # 2 or 5, theii the ordcr e is also Definition 23. If (a, m) = 1, and a is of order +(m)modulo m, we call
the period of the periodic decimal l/m. Thus 10 is of order 3 modulo 37, a a primitive root of m. In particular, for a prime modulus p , a primitive
as on page 55. (It is probable that this definition, and Definition 23, root of p is a residue class of order p - 1.
Theorem 35, and Theorem 36 which follow, all stem from Gauss's early EXAMPLE: +
Since, on page 53, the decimal expansion of is of period 6,
studies in periodic decimals mentioned on page 53. See Exercise 8s on 3 = 10 is a primitive root of 7.
page 203 for a plausible reconstruction of Gauss's line of thought.) The importance of Theorem 36 is that it guarantees (nonconstructively !)
Theorem 35. If (a, m) = 1, and a i s of order e , then is a primitive root for every prime modulus. This result-that is, every prime
modulus has a primitive root-is one of the fundamental theorems of
a'= 1 (modm) number theory. It is the basis of the theorems which we shall obtain in
we have elf. I n particular el+(m). Further, a', a', a3, . . . , ae belong to e this chapter concerning the structure of the 311, groups. I n particular, it
distinct residue classes modulo m. is the basis of the structural differences which we sought at the end of
Sect. 23 and the beginning of Sect. 27. It implies that 311, is a cyclic group.
PROOF. We have w / u e - 1 and mlaf - 1, and by Theorem 10, nilaU - 1
where g = ( e , f ) . Therefore g 5 e. But g Q e by the definition of e . There- EXERCISE 52. For every divisor d of 12, determine the +(d) residue classes
fore e = g and elf. Further, if ael = ae2 (mod nz), and e 2 el > e2 2 I , of order d modulo 13, in particular, determine the 4 primitive roots of 13.
we have ue"-ez = 1 (mod m),which again contradicts the definition of e . EXERCISE 53. For every prime p > 2, 1 is of order 1 and p - 1 is of
Theorem 36 (Gauss). I f dip - 1, where p i s a prime, there are +(d) order 2 and these are the sole residue classes of these orders.
residue classes of order d modulo p .
29. 311, AS A CYCLIC GROUP
PROOF. From Theorem 35, if a is of order e modulo p , then a', a2,u3, . . . ,
Definition 24. A group is cyclic if it contains an element g, called a
a' are e distinct residue classes. They are thus e distinct solutions of
generator, such that every element a in the group may be expressed as
2. =1
(mod P I , a=g
n
and, by Theorem 31, there can be no others. Each class of order e modulo
for some integral exponent, positive, negative, or zero.
p is therefore contained among these e classes. But if r 5 e and ( r , e ) # 1,
By Theorem 36, p has + ( p - 1) distinct primitive roots. Let g be any
let T = sg and e = tg with g > 1. Then
one of these. Since, by the last sentence of Theorem 35, g , g2, . . . , g P 1
(a')t = (a")" = 1
(mod P ) , are all distinct, g serves as a generator for m,, and thus 311, is cyclic.
By rearranging the rows and columns of the table for 3117 on page 61,
and we find that ur is of order S t < e . Let + ( e ) be the actual number of
and since 3 is a primitive root of 7, we obtain
classes of order e. Then, by Theorem 35, if e { p - 1, + ( e ) = 0, and
if elp - 1, we have just shown that
\
_
2 !- 6
+(el 5 d e ) . (115)
But since every class, 1, 2, - - - , p - 1 is of some order modulo p we have 614
__
c+ ( d )
d
= P -1
4
__
5
5
1
where the sum is taken over all positive divisors of p - 1. Since from __ _-
Theorem 34 we now have 2 6
_-
c[+(dl
d
- +(dl1 = 0, 6 4
and since, from Eq. (115), each [ d ( d ) - + ( d ) ] 2 0, we obtain

where the kth elernent in the first row is congruent to 3k-1. Here 3 is the
+(d) = 444 generator, and the ( n + 1)st row is obtained from the first by a left, n
for every d. shift, cyclic permutation.
T h e Underlying Structure 73
Some composite m may also have a primitive root; thus 2 is one for 9.
Or, if me prefer, under
112
--__
4 8
(mod 7) 5" - 5" (mod 91,
2 4 8 7 since 5 is a primitive root of both moduli.
____ The group of the mth roots of unity, e
2aaz/m
, for a = 0, 1, . . . , m - 1,
4 8 7 5
--__ under ordinary multiplication; the group of rotations of the plane through
8 7 5 1 360a/m degrees, for a = 0, 1, . . . , m - 1, undcr addition of angles; and
---_ the group of the m residue classes under addition modulo m, are all iso-
7 5 1 2 morphic. They all are the same abstract group-namely, the cyclic group
---_
of order m. We designate this group as e m .
5 1 2 4
The isomorphism between MI, for a prime p and e p-l suggests a circular
For a n y modulus m > 2 which possesses a primitive root g, regardless representation of m, , which eliminates the obvious redundancy in the
of whether m is prime or composite, it is almost immediate that if a = g" multiplication table for m p ,and which we illustrate for p = 17:
(mod m j , then a is a quadratic residue of m or not according as n is even
or odd. Further, there are exactly ++(m)residues. Further (Euler's Cri-
terion generalized), a is a residue if and only if = 1 (mod m) . Further,
the product of two nonresidues is a residue. We will determine later which
composite m have a primitive root, and therefore also these other properties.
EXERCISE 54. Prove the "if" part of Wilson's theorem (page 37) using
a primitive root of the prime p. HIST: evaluate the sum 1 2 + +... +
( q - 1) modulo q - 1. With reference to Exercise 25, generalize the proof
here to those composites m which have primitive roots.
Definition 25. Two groups a and a are said to be isomorphic if every
element a of a may be put into one-to-one correspondence with an ele- Here 3 is the generator and successive powers of 3 correspond to successive
ment b of a, rotations thru 224'. Or 3-' = 6 may be considered the generator and its
a-b powers are strung out in the opposite direction. Two residue classes a t
in such a way that if a1 - -
bl and az 6 2 , then a1a2 blbz . That is, the
angles a and fl have a product a t an angle a +
0.I n particular, reciprocals
4
lie a t a n equal distance from 1 in opposite directions. The residue - 1 = 16
correspondence is preserved under the group operation. Starting with the is thus its own reciprocal, and the only class of order 2. It follows that
a's and performing first the mapping, a -+ b, and then the product, we will residues on opposite ends of a diameter add to 17; each is congruent to
obtain the same result as if we first perform the product, and then the the other's negative. The quadratic residues are 1,9, 13,15, etc.
mapping. I n an isomorphism, therefore, these two operations may be com-
muted. If a and a are isomorphic, we write
It is well known that historically f i = a did not attain full re-
spectability until it was interpreted as a rotation of 90". If p is an odd
ae63 prime, m, will have a a if and only if p = 4nz +
1. We now see the
and we consider them to be the same "abstract group." significance of this, in that only e4, allows a rotation of exactly 90". Thus
for p = 17, in the diagram, we have 4 and 13 as the two values of 1/-1.
It is easily seen that two cyclic groups of the same order are always We see also that Euler's Criterion,
isomorphic. Thus
m7 e 3x9 up 5 (:) (modp),
under the mapping
and his even more celebrated formula,
(mod 7) 3" 2" (mod 9 ) .
en*i = (-I)",
are very intimately related. Euler was no doubt the worlds most prolific locations are arbitrary except that no two contacts lie on a diameter. There
mathematician. A modern mathematician, looking a t the last two equa- is a rotor ( R ) which may assume the 2N angular positions, and attached
tions, may be tempted to say, No wonder, he works both sides of an rigidly to R, at any of the 2N divisions on the hub, are N hands ( H ) .
isomorphism. But better judgment a t once prevails-had Euler not Again, their location is arbitrary except that no two lie in the same diam-
worked both sides, the isomorphism may not have been discovered. eter. Let m hands be touching contacts in a particular position of the rotor.
EXERCISE 55. Show that m14 a and give two distinct mappings. Theorem 37. As the rotor turns, (in either direction), m will be alternately
EXERCISE 56. Show that other circular representations of X17may be even and odd.
obtained from the given one by starting a t 1 and taking steps of k.224O EXAMPLE: I n the special case for N = 8 in the diagram, a clockwise
where (k, 16) = 1. More generally, if g is a primitive root of p , gk is also, rotation will give the follouing periodic m sequence: 5, 2, 5 , 4, 5, 4, 3, 4,
if and only if ( k , p - 1) = 1. 3, 6, 3, 4, 3, 4, 5, 4, repeat.
EXERCISE 57. Show that PROOF. Opposite each hand in a rotor is a space. Let a complete group
ms e %lZ of contiguous hands with no spaces in between be called a bunch, and
reading clockwise let the first hand in a bunch be called a trailing hand,
but and the last hand, a leading hand. Let a complete group of contiguous
ms * 3x10.
spaces be called a gap. Put each trailing hand T , into correspondence with
the leading hand L , immediately preceding the space opposite T , . There
Show that %& is not cyclic. is such an L, since preceding T , there is a space S, . Opposite X, is a hand.
30. THECIRCULAR SWITCH
PARITY Since this is followed by the space S , which is opposite T , , the hand is a
leading hand.
I n 1956 the author invented the following unusual switch. Now as the rotor turns one division (clockwise) , the only changes in m
which need be counted are those in which a leading hand picks up a contact
or a trailing hand drops one. For if a nonleading hand picks up a contact,
it was dropped by the hand ahead of it; and if a nontrailing hand drops a
contact, it is picked up by the hand behind it. But there was either a
contact under T , or in S , , but not both. Therefore either T , will drop this
contact, or L , will pick it up, but not both. The contribution of the pair
of hands towards changing m is therefore f l .
But starting a t T , , and going clockwise to L , , we will pass k bunches
and k - 1 gaps. And the remaining bunches in the other half of the rotor
may be reflected into these k - 1 gaps. Thus the total number of bunches,
2k - 1, is odd, and the number of pairs, T , and I,, , is therefore also odd.
But a change in m by an odd number of f l means a change of parity.
We now ask, how many distinct rotors of order N are there-that is,
rotors that cannot be transformed into tach other merely by rotation?
Call this number R ( N ) . If N is an odd prime, we obtain an old friend.
Set aside the special rotor R1consisting alternately of one hand and one
space. Consider any other rotor of order N , and in particular consider the
J
Si pattern of hands and spaces in a Hock of N consecutive divisions. This
pattern may be represented by an N-bit binary number, with ones for
Definition 26. A circular parity switch of order N has a stator (S) with hands, and zeros for spaces. Excluding the two possible patterns in RI :
2N equally spaced divisions. At N of these there are contacts (12). Their 1010 * . . 01
and 0101 . . . 10,
any of the 2N - 2 remaining patterns is a legitimate one, and will occur

in precisely one rotor R , . It cannot occur in two, since the remaining N
divisions of R , must have the complementary pattern, and therefore R ,
is completely defined. If a different block of N consecutive divisions in R ,
is examined, a different pattern must be found. For if two patterns in Ri
were identical, R , would have to be periodic, with a period less than 2N. But until we compute a primitive root we cannot (completely) assign the
This period must divide 2N. The period cannot be the prime N , since we residue classes to suitable billets. (Where p - 1 = -1 goes is simple
know that complementary blocks of N divisions have complementary enough.)
patterns, not the same. The period cannot be 2, since we excluded those Given a prime p , it is always possible to compute a primitive root by
two patterns. Thus Ri must have 2N different patterns. Therefore trial and error, since m, is finite. For p > 2, a quadratic residue of p is
clearly not a primitive root of p . For if a is a quadratic residue of p = 2P 1 +
2N - 2 we have up = 1 (mod p ) by Eulers Criterion. Thus the order of a modulo
R(N) =
2N
~
+ 11 p is P or smaller. Further, for p > 3 , p - 1 = -1 is not a primitive root,

since ( - l ) z = 1. But with these obvious exceptions, and with no deeper
and since R ( N ) is an integer, we have reproven Fermats Theorem 12. theory, one might now examine the remaining residue classes in search of a
A second application of the parity switch is this. Consider the circular primitive root. Gauss, and others, have devised more efficient techniques,
diagram for mp (page 7 5 ) as a stator, with contacts at the even numbers. but no general, explicit, nontentative method has been devised, and this,
This is a legitimate stator since opposite each even e is the odd p - e like a good criterion for primality, remains an important unsolved problem.
as we showed on page 73. Let the rotor have hands which, in one position, The converse problem is even harder. Given an integer 9 , for which
point to every odd number. If the hand pointing to 1 is now brought primes p is g a primitive root? Xot even in a single instance is it known
around to the number a, the ( p - 1 ) / 2 hands will point to the ( p - 1 ) / 2 that there are infinitely many such primes p . For example consider
products Theorem 38. If p = 4m + 3 and q = 2p + 1 are both prime, -2 is a
l . a , 3.a, 5 . a , - . ., ( p - 2 ) . a primitive root of q.
(mod p ) ,
EXAMPLE
:
and let m of these products be even. Since in the rotors original position -2 = 5 is a primitive root of 7 .
m is 0, by Theorem 37 m will be even or odd according as a is a quadratic
PROOF. There are + ( 2 p ) = p - 1 primitive roots of q. None of the p
residue or not. That is,
quadratic residues, a , of q can be a primitive root, as above. Nor can -1,
(;) = (-l)m.
which is not a quadratic residue, be a primitive root. Thus any other
quadratic nonresidue is a primitive root, and -2 is always one, since
Thus we have reproven a combination of Eulers and Gausss Criteria (-2Iq) = -(2\q) W N = - - I D .
with the aid of a switch. Therefore if Conjecture 4 were true we could prove the existence of
infinitely many q with -2 as a primitive root.
31. PRIMITIVE ROOTSAND FERMAT
NUMBERS Similarly, if the weaker Conjecture 3 were true, we could utilize
By characterizing m, as a cyclic group, for every prime p , we have gone Theorem 39. If p and q = 2 p +
1 are both odd primes, -4 i s a primitive
the limit in its structural analysis. A cyclic group is the simplest type; root of q.
and we may say that there remain no questions concerning its structure. EXAMPLE :
But the content of that structure is quite another matter. Thus we know, -4 = 3 is a primitive root of 7.
a t once, that m7 e6:
The proof of Theorem 39 is left for the reader. Another theorem of
slightly different character is 3; +1, and all odd squares, are primitive roots only for the prime 2; and
Theorem 40. If F , = 22m + 1 i s a prime, with m h 1, 3 i s a primitive any even square is never a primitive root. I n spite of the negative results
of the previous section, the evidence is sufficient to warrant our stating
root of F, .
Coqjecture 13 (Artin). Every integer a, not equal to - 1 or to a square, i s
EXAMPLE : a primitive root of infinitely many primes.
3 is a primitive root of 5 = F, and of 17 = F2. It is likely that a stronger result is true:
PHOOF. Since + ( F m - 1 ) = $ ( F m - 1 ) ) we see that in this (unusual) Conjecture 14 (Artin). If a # b with n > 1, and i f v a ( N ) i s the number
case any quadratic nonresidue of F, is also a primitive root. But of primes 5 N for which a i s a primitive root, then
II
F , = 5 (mod 12) -
v O ( N ) 0.3739558 r ( N ) .
This conjecture was made by E . Artin in a conversation with H. Hasse
(117)
by induction, since F1 = 5, and
F,+I(F, -
= + 1. in 1927. It states that for a = 2, 3, 5, 6, 7, 10, etc., approximately of all
primes will have a as a primitive root, and that this asymptotic ratio,
Therefore, by Theorem 20, (3lF,) = -1. 0.37 . . . , is independent of a. (If a is a cube or some other odd power,
Here, again, we do not know whether there are infinitely many Fermat there is a minor complication, which need not concern us here.)
numbers, F,, which are prime. Fermat thought all F, might be prime, We shall explain presently the coefficient in Eq. ( 117), and the heuristic
but said he couldnt prove it. Euler showed, however, that 6411F6, as on reasoning behind Eq. (117). But first we examine two tables based on
page 58. Aside from the five primes, F, for 0 5 m S 4, no other prime counts, va( N ) , given by Cunningham ( 1913).
F, has been found. On the contrary, F, for 5 S m 5 16, a t least, are
all composite. Any prime F , corresponds to a constructable regular polygon, a Vo (10,oOo)
(Gauss, page 52). Like the Mersenne numbers, (page 18)) the Fermat
2 470 ,3824
numbers, (page 13) , are all prime to each other.
3 476 .3873
There are three possibilities: 5 492 .4003
(a) Only finitely many F, are composite. 6 470 ,3824
7 465 .3784
( b) Only finitely many F, are prime.
10 467 ,3800
(c) Infinitely many F , are prime, and infinitely many are composite. 11 443 .3605
If (a) or (c) were true, we could find infinitely many primes with 3 as a 12 459 ,3735
primitive root, but actually possibility (b) is the most likely. We will
return to this question in Exercise 36S, page 214. 3806 av.
EXERCISE 58. Criticize the word explicitly in the last sentence in ~
Exercise 47. Investigat,e possibilities of remedying this flaw. N/10,000 VAN) vz (A3 /r(N)
EXERCISE 59. Find a primitive root of p = 41. 1 470 ,3824 467 ,3800
2 840 .3714 8G5 .3824
EXERCISE 60. Find 16,188,302,110 primitive roots of q = 32,376,604,223.
.3803
3 1205 .3713 1234
EXERCISE 61. If p = 4m + 3 > 3 and q = 2 p +1 are both primes, 4 1570 .3735 1587 .3776
there are a t least three successive integers, g , g + 1, and g + 2, which 5
6
1923
2263
.3746
.3736
1947
2296
,3793
.3791
are all primitive roots of q.
7 2589 .3733 2639 ,3805
EXERCISE
62. Using residue arithmetic, show that 8 2928 .3736 2975 .3796
9 3274 ,3758 3291 ,3777
2741771Fc. 10 3603 ,3756 3618 ,3772
32. ARTINSCONJECTURES I
,3745 av. ,3794 av.
It is easily seen that -1 is a primitive root only for the primes 2 and
A/
I n the smaller table we see that v, is substantially independent of a tions. This may be noted in both tables above, and also, more clearly, in
for the eight smallest positive integers not equal to a power. I n the larger the following data: v 2 ( N ) / a ( N )= 0.3988, 0.3861, 0.3857, and 0.3849 for
table, for the two most studied cases, a = 2 (related to perfect numbers),
N = 1000, 2000, 3000, and 4000. This tendency has an interesting ex-
and a = 10 (related to periodic decimals), we see that v , ( N ) / n ( N ) changes
planation* If a prime does not have 2 as a primitive root, the reason, four
only slightly with N. times out of five, is that (2jp) = + l . These latter primes are those of the
A probability argument which makes Conjecture 14 plausible runs as
forms 8k f 1. While it is true that these primes are equinumerous to those
follows. Consider a = 2, and the primes p 5 N . For every p choose a of the forms 8k 3, nonetheless there is a definite tendency for the class
primitive root g and write gm = 2 (mod p ) and ( m ,p - 1) = G. What is
the probability that 2/G?Except for p = 2, p - 1 is always even, and m
of primes 8k + 1 to lag behind the other three classes. See page 21 for
some data. This interesting lag (which we will discuss in Volume 11) has
is even in one half the cases-that is, when 2 is a quadratic residue of p .
the consequence that (1 - 3) , the first factor in A , is too small for these \
Since G must be 1 if 2 is to be a primitive root of p , we delete these cases,
modest N, and therefore, in general, v 2 ( N ) runs too high.
leaving, in the mean, (1 - + ) T ( N ) primes. What is the probability that -
3jG? Except for p = 3, all primes are 3k +1 or 3k + 2 , and therefore 33. QUESTIONS
CONCERNING
CYCLEGRAPHS
3 / p - 1 in one-half the cases, while 31m in one third the cases. Eliminating
We now concern ourselves with the structure of 311, with m not neces-
the remaining primes where 31G me are left with sarily a prime. A good insight into these structures will be gained by the
primes. Continuing with 5/G, 71G, etc., we are left with study of the cycle graphs of these groups.
Definition 27. If ( a , m ) = 1 and u is of order e modulo m, the e residue
A.n(N) classes d,u2, u3, . . , ue are called the cycle of a modulo m. The definition
primes with G = 1, where the coefficient A (called Artins constant), is may be clearly generalized to any finite group.
given by the infinite product: s
Definition 28. If a set s of elements in a group is closed under the
group operation, and contains the identity and the inverse of each of its
elements, it is called a subgroup of s. s
In particular, itself is also a sub-
group of s.
The argument may he improved somewhat by using Theorem 16 and It is clear that each cycle of m, is a cyclic subgroup of 311,. A diagram
analogous results, but this improvement does not suffice to constitute a of a group, which shows every cycle in the group, and the connectivity among
real proof of Conjecture 14. For any other nonpower a , the argument is these cycles, is called a cycle graph of the group. It generalizes the circular
unchanged, but for u = 8, say, we have 31m in all the cases where diagram of m17 on page 75. On pages 87-92 we show cycle graphs for 14
nonisomorphic m, groups. We will first make some comments, and we will
p = 3k + 1. This changes the factor then raise some questions.
instead Let our point of departure be the cycle graph of 31155on page 88. It is
vs(N) - +An(N), etc.
of only moderate complexity, and thus is best adapted to illustrate the
concept. The powers of 2 (mod 55), namely 1, 2, 4, 8, 16, 32, 9, 18, etc.,
J. W. Wrench, Jr., has recently completed a highly accurate computa- constitute the cycle of 2 modulo 55. This cyclic subgroup of 3n55is of order
tion of Artins constant. He gets 20, and is easily seen in the graph. Now 53 = -2 (mod 55) is not in this
subgroup. Therefore the cycle of 53, which is also of order 20, is connected
A = 0.37395 58136 19202 2880,5 47280 54346 41641 51116 . (119) to the cycle of 2 only a t their even powers, that is, at the quadratic residues.
If Artins Conjecture 14 proves as obdurate as the conjectures of Sect. 12- Similarly 51 = -4 has a cycle of order 10 which is connected to that of 4
and there is little doubt that it will-Wrenchs Eq. (119) should suffice as a t their even powers. Finally, the cycle of 29 completes the 40 = 9(55)
a check on any empirical studies of v , ( N ) for quite a long time. residue classes in m56 . Xo residue class is of order 40 modulo 55 and there-
There is distinct tendency for v a ( N ) / r ( N )to run high for small values fore m65 is not cyclic.
of N-that is, for this ratio to approach A from above, aside from fluctua- Now let us back up to some smaller composite moduli. The smallest m
for which m, is not cyclic is 8. This is a well-known group of order 4-the (e) Can we characterize m, by a formula? Given m, we wish to deter-
Tour group. Here 3 , 5, and 7 are all of order 2, and their 3 cycles are mine the structure of m, by a n (easily computable) formula. We recall,
connected only at their common square, 1. Since Em12 $ m8, their cycle in this connection, that the structure of 3x17 is clear even before we compute
graphs look alike-in fact, if 3 is replaced by 11, they are identical. a primitive root.
The next noncyclic group is mI6.Here four residue classes are of the (f) If 311, is cyclic there is an a of order + ( m )modulo m. But if m, is
highest order, 4, and the cycles for 2 and 7, say, are connected a t their not cyclic what is the largest order possible within the group?
common square, 4, and common fourth power, 1. Two other cycles are (g) If m, is cyclic there are & ( m ) quadratic residues, but if mmis not,
those of 11 and 14. It is clear that in the cycle graphs we are concerned how many are there?
only with the ordering in, and topology of, the cycles. The actual size, (h) Finally we note, from group theory, that every group of order 4 is
shape, or location of the various cycles is not meant to be of significance. either isomorphic to 3 2 8 or to the cyclic 3 n 5 . There are only two abstract
As with the circular diagram for m17, we can easily read off the powers, groups of order 4. Of order 8, there are five abstract groups, with cycle
order, and inverse of every residue class. graphs as follows:
It may be seen that
m
15 31116 31120 m30.
m2
4 is also of order 8, but is not isomorphic to 31115 , or to any other Em,.
It has only one quadratic residue.
~ Z , which
I is isomorphic to mZ8, 31136, and m42 , may be generated by
the three cycles of 10, 11, and 17. These three cycles are connected a t the
three quadratic residues.
31154 is cyclic and isomorphic to m19.
m2
4 Q 9 4
m 63 really needs three dimensions. The four bunches, of three cycles

We see that two of them are isomorphic to 311, groups. The cyclic group
each, regroup, after passing through the quadratic residues 4, 25, 37, and
(38 f 311,, but it is a subgroup of an YE, group. Namely, (3, is isomorphic
22, into three bunches of four cycles each. After passing through the square
to the group of quadratic residues of xl7 -that is, to the cycle of 2 modulo
roots of unity 62, 8, and 55, they again regroup, etc. By needs three I
17 (see page 75).
dimensions we mean, of course, that it cannot be drawn in two dimen- The remaining two groups are well-known non-Abelian groups; Q is the
sions without some cycles crossing each other. I n three dimensions 31163 may quaternim group, and a)4 is the octic group (the symmetries of a square).
be neatly represented as four mzl-like structures, in four planes separated
Since their multiplications are not commutative, they cannot be isomorphic
by angles of 45) and joined together at the four square roots of unity, 1,
to any am , or subgroup thereof. Therefore every Abelian group of order
62, 8, 55.
8 is isomorphic to a subgroup of an m, .
Now we wish to ask several questions.
We now ask, is every finite Abelian group isomorphic to a suhgroup of
(a) For which m are the m,, cyclic?
m +
( b) Which m, are isomorphic? Generally when we pass from m to
I, we obtain a totally different pattern, e.g., m = 54, 55, 56, 57. But
I
1
a n F111, ?
We close this section with a useful theorem.
3n3 $ 3R4 , 3Rls 31116 , and, more spectacularly, mIo4 I
Theorem 41. I n every Jinite Abeliun group, i f x2 = a possessfs n solufions
(c) For which m are the cycle graphs three-dimensional; as in m = 63, 1
x, then every square, y2 = b, possesses n solutions. I n particular, in 311,,
and, even more intricate, in nz = 91? every quadratic residue has an equal number of square roots modulo m.
(d) We note definite lobal patterns. Thus m57has nine lobes of the PROOF. Let a have n square roots, zl , x2 , . . . , zn . Let b have at least
same type of which 3 1 1 ~has ~ three, and m8 , one. Again, m56 has three lobes one, yl . Then each element
of the type of which 31124 has one; and 3nb5possesses five ml5-type lobes.
y. = y,xl-lx, (120)
We ask, what is the structure of the various types of lobes, and how many
2 -2 2
such lobes may a group have? for i = 1, 2, . . . , n satisfies 7/12 = 11 since 2/12 = 11, T I xt = hCn = b.
Further, if yt = y, , we have x1y1-'y2 = xlyl-'y, , and thus x2 = x, .There-

fore no square in the group can have fewer square roots than any other
square.
It follows that if the cycle graphs for Q and D4 represent groups, (and
they do), these groups cannot be Abelian, since in the octic group the
identity has 6 square roots, while a second element has only 2. In the
quaternion group the situation is reversed.
EXERCISE
EXERCISE
group of
63. Show that + ( m ) = 8 has exactly five solutions m, and
that therefore m24 is isomorphic to no other m, .
64. Each of the 7 rows in the table on page 47 form a sub-
3 ~ 2 isomorphic
EXERCISE
4 to me.
65. 3Kis has both abstract groups of order 4 as subgroups.
A
EXERCISE 66. The quadratic residues of m constitute a subgroup of
m,, . Call it Q m . Then Q 5 5 3% and Q65 m21. But Q 6 3 is isomorphic to
no m, . Also Q54 Q 6 7 , etc.
EXERCISE 67. Draw a cycle graph for ms3.

EXERCISE 68. Determine the periods of the decimal expansions of & and
by examining the cycle graphs of m57 and mE3 .
EXERCISE 69. Determine 11-', 47-', and the four square roots of -1
modulo 65.
EXERCISE 70. Determine the order of 2 modulo 85. Interpret the result
+
in terms of the equation F o F I F ~ 2 = F , . Compare Exercise 4.
EXERCISE 71. Let a finite group of order m contain a subgroup of order
s. Then slm. This is called Lagrange's Theorem-it generalizes Theorem 35.
EXERCISE 72. There is only one abstract group of a prime order-the
cyclic group.
The Underlying Structure 89
31167
ma
3
mlo6 m = plelpznz . . . P,"". (121)

(B) For each odd prime p , write +(p,"') = ( p , - 1)pL"'-' in a modified
standard form
$I(P,"') = <qlb1><q2b2> . . . <q."'><Ptaa-'>, (12iL)
by factoring p , - 1 into the prime powers qzb', and, if a , > 1, by including
the last factor. The symbol <q;'> means that the prime power is written
as a single number, e.g., <5'> = 25.
(C) I f p l = 2 a n d i f a l > lwewrite+(4) = 2 , + ( 8 ) = 2 . 2 , + ( 1 6 ) = 2 . 4 ,
and, in general,
$1(2"')= 2.<2a1-2> (al 2 3). (123)
If al = 1, we omit this step.
(D) Now combine (C) and (B) into a modified standard factorization
of $I(m):
$Im -- 2 . 2 . . . 4 . 4 . . . 8 . . . 3 . 3 . . . 9 , . . 5 . . . (12.2)
Here +(m) is factored into primes, and powers of primes, and we take
care not to multiply factors of 2 with those of 4, etc.
If m = 2, we write& = 1.
EXAMPLES
:
m = 105 = 3.5.7 $I105 = 2.2.4.3
m = 104 = 8.13 $104 = 2.2.4.3
m = 65 = 5.13 6%= 4 . 4 . 3
m = 15 = 3-5 $115 = 2.4
34. ANSWERS
CONCERNING
CYCLEGRAPHS m = 16 $I16 = 2.4
We shall prove
m = 24 = 8.3 424 = 2.2.2
Theorem 42 (Gauss). Sn, i s cyclic-that i s , m has a primitive root-if
and only if m i s one of the following: m = 63 = 9 . 7 $63 = 2.2.3.3
m = 2, 4, p", or 2p" m = 17 417 = 16
where p i s a n odd prime and n 2 1. Now we can state

EXAMPLE. Theorem 43. m, and Sn,~~are isomorphic if and only if +ml and $,, I are
identical.
5nb4is cyclic, since 54 = 2.2.
EXAMPLES :
Which X , are isomorphic? To answer this we nee1 311104 m1
06 ?4 m65
Definition 29. By +m we mean a particular factorization of +(m)ob-
tained as follows if m > 2 :
m1
5 * m1
6 f m2
4
(A) Factor m into it,s standard form: m4

0 @ m4
8 $X 1 7 (verify).
The Underlying Stru.cture 95
EXERCISE 73. If k is odd, 311k $ 3 n 2 k . If k is prime to 3 and 4, 3 n a k

EXAMPLE
:
3114k
For m = 15, = 2.4, and we may take g1 = -1 = 14 (mod 15), of
If k is prime to 7 and 9, m7k x&. order 2, and g2 = 2, of order 4.Then each of the 8 residue classes in 31116 is
EXEI~CISE
74. Show that the X, are isomorphic for rn = 35, 39, 45, 52, E ( - 1)"2*(mod 15) for one a and b such that 0 5 a < 2 , and 0 5 b < 4.
TO, 78, and 90. The representation, Eq. (l26), of 3% as a direct product of r cycles is
EXERCISE 75. Show that the 5n, are isomorphic for m = 51, 64, 68, the characterization we sought in question (e) on page 85. We shall see
and 102, but for m = 51, 80, 96, and 120 we obtain 4 distinct abstract presently that Theorems 42 and 43 are consequences of Theorem 44. But
groups of the same order. so are two others:
The last two theorems both follow from a more powerful result. To Theorem 45. If f z are characteristic factors of 311, , then
state this, it is convenient to modify the last definition to
fllfl (127)
Definition 30. By @, we mean a particular factorization of #(m) ob-
i tained as follows: i f i 5 j . It follows that i f jr i s the largest characteristic factor of 311, ,
For each distinct prime qt which divides #( rn) we take the largest power afr -
= 1 (mod 112) (128)
<qtb'> which appears explicitly in 4, and multiply these powers together.
The product we call a characteristic factor of mrn. Setting this factor aside for every residue class a in 311, .
we repeat this operation with the remaining <q%"'> in # m . Then a,,, is COMMENT : Equation (128) gives us a sharpening of Euler's generalization
the product of these characteristic factors of Fermat's Theorem. Further, it is clear f r o n Eqs. (128) and (126) that
f,. is the answer to question (f) on page 83.
@m = fl.j2 . . . jr (125)
Theorem 46. Ij m > 2, and mrnhas r characteristic factors, m has
where jl 5 j2 5 j3 . . . 5 j,. # ( m ).2-' quadratic residues, and each of these has 2' square roots.
EXAMPLES
: EXAMPLE :
@104,]05 = 2.2.12 mloshas = 6 quadratic residues.
a.65 = 4.12 PROOF OF THEOREM 4.5. Equation (127) is clear from the construction
of the jtin Definition 30. Then Eq. (128) follows at once from Eq. (126).
@15,16 = 2.4 PROOF OF THEOREM 4G. If m > 2, each contribution to & , Eq. (122)
%.24 = 2 .2 .2 in step B, and Eq. (123) in step C of Definition 29, is even. Therefore it
follows that each jtis even. It is then apparent, from Eq. (12G), that u j is
a
63 = 6.6 a quadratic residue of m if and only if each of its exponents, s,,, is even.
Since, by Theorem 41, each quadratic residue has an equal number of
= 16
square roots, Theorem 46 follows.
Then we mill have PROOF OF THEOREM 42. If m = 2 , 4 , p " , or 2p" we find that @, = # ( n z )
Theorem 44. Ij @% i s the product of r characteristic factors jz, for each f l with only one characteristic factor, and therefore g1 is of order #(m)-that
there i s a residue class g , , of order fz modulo m, such that every residue class is, g1 is a primitive root. Whereas if ~ i is
z divisible by two distinct odd primes
a , in mrncan be expressed as or equals 4k with k > 1, we find at least two characteristic factors. Since
the largest, jr, is less than #( m ) , by Eq. (128) there is no primitive root.
glsl.ig2s2.i .. . grsr3t
a, (mod m) (126) PROOF OF THEOREM 43. First me note, by the construction, that +!, and
with 0 5< ji SL,j
4," are identical if and only if @,. and am,, are identical. Then if 4%. and
#m" are identical, by the ohvious mapping
in one and
of the g , .
only one way. We say that mmis the direct product of the r cycles gl'slg2fsz . . . grlsr - g1f~s1g2f~s2 . .. g,f181
96 Solved and Unsolved Problems in Number Thcory The Underlying Structure 97
we find that mrn, and mrnRare isomorphic. Conversely if they are isomorphic graph of 3n, is three dimensional if mmhas a t least two characteristic
it is clear that +(m) = 4(m) and also, from Theorem 46, mmtand 311,. factors which are not powers of 2. Thus m mis three dimensional for m :
must have the same number of characteristic factors. We say further that G3, 91, 275, and 341, since a63 = 6 . 6 , ag1= 6.12, @215 = 10.20, and =-
am.and a,. must in fact be identical, for, if not, we compare 10.30. See Exercise 19S, on page 206, for a sketch of the proof.
On the other hand, if
with a, = <aa>. <2b> . . . <2y> .<2w>
where N is an odd number 2 1, the cycle graph will have N lobes, and each
from right to left, and let fj # fj be the largest factors which differ. lobe is characterized by the formula ( <2>. <2b> . * . <2>. <a*>).
Assume There are two different lobes of order 4:
F = fj < f j = G
a.nd let P be the product
f;.+1.f;+2
Then the R = +(m)/Presidue classes,

. . . fr = f:(,1.f;+2
g1s1g282. . gjJ .
. . . fi..
fq.
obtained by allowing the s, to take on all values, all satisfy zF = 1(mod m) .
But all R of the residue classes
Thus the cyclic m13has 3 of the { 4 1, while Snzl (page 87) has 3 lobes { 2 ~ 2 ) .
There are three different lobes of order 8: the cyclic (8); and
do not satisfy xF = 1 (mod m) since q l n is of order G > F . Let there be
S < R residues, Eq. (130), which do satisfy x F = 1 (mod m ) . All in all
there are exactly RFT- solutions of x F = 1 (mod m) since any of the R
solutions of Eq. (129) may be multiplied by
la
Is,+i
g3+1 g3+P .. . grtg9
t o yield another solution, if, and only if f,+kIs3+k.F for each k such that
j + 1 =( j+ k 5 T . That is, each s,+k can take on the F values as in VZ4], 3p166 and 3 1 1 5 6 respectively.
fi+lc 2f?+k
- . . . ( F - l)f,+k There are five different lobes of order 16: the cyclic { l 6 ) in m17; 12.2.41
7 F f F F * in 3?llOs
; ( 4 . 4 ) in 31165 ; (2.8) in 3K3* (not shown) ; and ( 2 . 2 . 2 . 2 ) in 311168
(not shown).
Likewise there are exactly SF- solutions of zF = 1 (mod mf). Since How many different lobes are there of order 2? The answer is p ( n ) , the
S < R it follows that 311,t and 311, are not isomorphic unless and +,I
number of partitions of n. Thus p(4) = 5, since 4 may be partitioned (into
9,. are identical, since, in any isomorphism, 1 must map into 1, and the positive integers) in five ways:
z such that z = 1 must map into similar x.
Theorem 44 is also one of the keys to the answer to the last question in 4 = 4
Sect. 33, page 85. This answer is given by 4 = 1 + 1 + 2
Theorem 47. Eilery finite Abelian group i s isomorphic to a subgroup of 4 = 2 + 2
m,, for infinitely m a n y different values of m.
4 = 1 + 3
The two remaining questions in Sect. 33, (c, arid ( d ) , we shall here
answer with less formality. We will state, without proof, that the cycle 4=1+1+1+1
We will return to the theory of p ( n ) in Volume 11. To each partition of and the square bracket has p - 1 terms, each of which is =hW2(mod p ) .
n = nl + +
n2 ' . . +
nk there is a lobe of order 2": But p + ( p - 1)hP2 and we thus have
{ <2"l>. <2"*> . . . <2"k>). +
( h P)*' f hlT' (mod p') .
It will follow from Theorem 47 that for any such lobe, and for any odd Therefore a t least one of the numbers h and h + p-call it g-satisfies
N , there are infinitely many Em, which have subgroups with a correspond-
ing cycle graph.
gP-' 1 +
(mod p'). (132)
But it is not possible to have two lobes of { 2 . 2 ) : By Theorem 35, if g is of order e modulo p 2 , el+(p2) = p ( p - 1). But,
since g is of order p - 1 modulo p , \ye have p - l l e . From Eq. (132),
e # p - 1 and we therefore find e = ( p - l ) p , that is, g is a primitive
root of p2.
We thus have
(f-1 = 1 kp +
(with p t k ) .
By the binomial theorem

since we have seen that this group is non-Abelian (page 86). And four
lobes of (2.2) does not represent a n y group-ven a non-Abelian. It may
be shown that it violates the associative law.
There remain the tasks of proving Theorems 44 and 47.
EXERCISE 76. Find the relationship between r, the number of charac-
teristic factors of X,,and the number of odd primes which divide m, and
the power of 2 which divides m. since
EXERCISE 77. For any ai , in Eq. (126)) which has every s ~ even,
, ~ find
explicitly its 2' square roots.
EXERCISE
78. If (a, 561) = I, then By induction, for every odd p ,
=1 (mod 561). 9
(P-l)pd-' = 1 + kap8 (p+ks).
I n particular, 56112561- 2. It follows, by the same argument as for p2, that g is also a primitive root
35. FACTOR
GENERATORS
OF Em,
of p" for every s. For
We will prove Theorem 44 in three (rather long) steps.
g(P-l)P'-2 P1 (modp')
Lemma 1. Theorem 44 i s true i f m equals a prime power p". That i s , i f and thus g is of order ( p - 1)pS-' modulo p'.
p i s odd, or if m = 2 or 4, m has a primitive root. If m = 2" with n 2 3, For p = 2 , we note
@ m = 2.<2"-'> and we have the representation 32=1+8
a 3. -- ( - 1 ) 8 J 3 t J ( m o d 2 " ) (131) 34 = 1 + 16 + 32t

where -1 i s of order 2 , and 3 i s of order 2n-2.
3' = 1 + 32 + 6 4 ~
PROOF. I T c know that each p has a primitive root h . 11-e first show that
either h or h + p is a primitive root of p2. For, if p > 2 , from Eq. (13) for integers t and u. By induction, 3 is of order 2"-' modulo 2", if n 2 3.
with x = h + p , y = h, and n = p - 1, we have But none of the 2n-2 classes
+
( h + P ) ~ - ' - hp-' - p [ ( h P)*' + +
( h pIp-% + +
. . . hpz1 a 3. -= 3tl (mod 2")
can be congruent to an For any Jixed 7 %in Eq. ( 1 3 5 ) , since h y t = 1 (mod A ) , the + ( A ) values
of c, are = to the9(A) valiiesof a , (mod A ) , and therefore are incongruent
an 3 -3'k (mod 2 7 ,
modulo A . On the other hand, for two different values of y , , the cz are
for, if so, we would have incongruent modulo B . Therefore each of the + ( i i ) + ( B )values of c, are
813" +1 incongruent modulo AR, since they are either incongruent modulo A , or
modulo B , or both. Since each cz is prime to both A and B, each is prime
where a = I tj - t k [ . This is not possible since 3" 1 = 2 or 4 (mod 8) + to A B , and since + ( A B ) = d ( A ) + ( B ) the
, Lemma is proven.
for every a. Therefore the representation given by Eq. (131) gives every Therefore, given any
residue class in 3112..
On page 90 we see the cycles of 3 and 63 = - 1 (mod 64). Each residue m = plalpzaz . . . P,"",
class -3" has been placed close to +3". we can construct a representation of mmin the form Eq. (135) by treating
Lemma 2. If the + ( A ) classes ai in mAcan be written each p,"' by Lemma 1, and compounding them by Lemma 2. There re-
a . = g l d l , ig2az.i . . . gnu".
t -
mains the problem of putting the representation into the characteristic
(mod A ) (133) factor form, Eq. (126), of Theorem 44.
where th.e factor generator g j i s of order mj modulo A , and Lemma 3. If g is of order A B modulo m with ( A , B ) = 1, the A B residue
m1m2. . . m, = 4(A); classes gc ( 0 5 c < A B ) can be written as a direct product
c a b
and if the + ( B ) classes bi in a cyclic m a ,with B prime to A , are written g = st (modm) (138)
b 1. =
-9
Bi
(mod B ) 7 where s is of order A and t is of order B modulo m. Conversely, given two
(134)
residue classes s and t , of orders A and B , with ( A , B ) = 1, :he A B classes
then the + ( A B ) classes ci in X A B can be written on the right of Eq. (138) are all distinct, and constitute the cycle of some g of
c . = hlrl,*h272.i. . . h,Yn.ihYi order AB modulo m.
I -
(mod A B ) (135)
where the factor generator hj i s of order mj , and h is of order + ( B ) ,modulo PROOF. Given g , set s = g B and t = g A . Then s is of order A , t of order
AB. B , and the A B classes
PROOF.
Let sat b -
= g aBfbA (OZa<A) (OSbbB)
k = B-'(l - 9) (mo d A) are distinct by Theorem 33. Conversely, given s and t , consider the right
side of Eq. (138). Now if
and kj = A-'( 1- gj) (mod B )
Saitbi -
= Sa2th2 7
for j = 1, 2, * * * , n. Then set h = Bk + g , hj = Akj + gj , and we have

we have
h = Bk +g = 1 (mod A ) , Sai-az -
= tb2-b1 1
hi = A k j + gj = 1 (mod B ) , ( j =
(137) 1, 2, . . . , n ) . and
We now say that Eq. (135) has the stated properties. For, since hi 3 1 S(a~-az)A -
= t(bz--bi)A
g j (mod A ) , hj is of order m j modulo A . Therefore
Then Bl(b2 - bl) A , and, since ( B , A ) = 1, we have bl = b 2 . Likewise
hjmi = 1 + sA and also = 1 + tB. al = az . Thus the A B classes on the right of Eq. (138) are distinct. Now
Therefore sA = tB and, since ( A , R ) = 1, we have Rls. Thus hj is also set g = st (mod m ) , and if g is of order e , then
of order mj modulo A B , since, if it were of a smaller order modulo A B , sete = 1.
this would imply a smaller order modulo A . Likewise h is of order + ( B )
modulo A B . Thus, as before, Ale, and Hle. Therefore e = -4B. Further if A-'A 1
P
102 Solved and Gnsolved Problems i t a Number Theory
The Underlying Structure 103
(mod B ) , 11e have
"Proof." Any primitive root h of p may be lvritten h = gk where k is prime
A-'A = kB + 1, and K ' ( 0 - a ) A +a = k(b - a)B + 0. to p - 1, and k < p - 1. But
Let gP ' = 1 + sp (P4-s).
f = A-'(b - a ) A + a (mod A B ) Therefore he' = g'p"'k = 1 + ksp + tp2, or, since p t k s , we may write
or hp-' =
1 + up (Pb).
f = k(b - a)B +b (mod A R ) . Therefore h is of order p ( p - 1) modulo p'.
Therefore g' = s't' = satb (mod 712) so that the cycle of g contains the A B EXERCISE
81. Given g, a primitive root of p", with p odd, find a primitive
residue classes, and no others. root of 2p".
Thus, given any representation of 3% , obtained by Lemmas 1 and 2, EXERCISE 82. Determine a representation, Eq. ( 135), of X L 3 5 by Lemmas
we may decompose the cycles into cycles of prime-power order, as in + m , 1 and 2. It will be a product of two cycles of orders 4 and 6. S o w decom-
and then recompose them into the characteristic factors, as in @ m . This pose and recompose into a product 2.12 and thus map %7A5 isomorphically
completes Theorem 44, and therefore also Theorems 42, 43, 45 and 46. into m39.
Previously u-e made the point that a primitive root for a prime modulus EXERCISE 83. Investigate the degeneration of Eq. (131) into one char-
mas proven to exist nonconstructively. %7e should now add that the sub- acteristic factor for 2" = 2 or 4. Note: - I = 3 (mod 4 ) .
sequent steps in proving Theorem 44-that is, the foregoing three Lemmas-
are all constructive, and involve explicitly given computations. EXERCISE 84. Show that 44-z' +
1, and therefore 1/-1does not exist
modulo 2" for n > 1. But if Emzn were cyclic, would exist if n 2 3.
WTe note that a representation of m, in the form of Eq. (135) is not
necessarily unique, even as to the number of factors, and can involve as Thus 2" has no primitive root if n 2 3.
many generators as the number of factors in +", or as few generators as EXERCISE 85. Let n 2 3. Show that r is a quadratic residue of 2" if, and
the number of factors in +". only if, r = 812 +
1. Thus 17 is the smallest positive integer, not equal to
We may also note that the last Lemma can assist us in the finding of a square, which is a quadratic residue of 2". Xote that in the cycle graph
primitive roots. Thus 2 is of order 3 modulo 7 and - 1 is of order 2. There- on page 90 the quadratic residues of 64 are strung out in numerical order!
fore -2 = 5 is of order 6, that is, 5 is a primitive root, etc. What do you make of that? Also, the smallest positive a for which x2 a +
EXERCISE 79. A primitive root of p which is not a primitive root of p' is is divisible by every power of 2, for some x, is a = 7.
hard to come by. Show that 10 is a primitive root of 437 but not of 487' by EXEI~CISE 86. For m = 2" with n > 3, show that the two classes of
computing (101487), and with reference to the congruences: numbers 8k +
1 and 8k - 1 play special roles in the structure of 3%.
loo3 = 189 (mod 487) loo3 = 51324 (mod487')
But 8k + 3 and 8k - 3 play similar roles. How many subgroups of order
2"-' are contained in 311, ? Are they all isomorphic? Show that 3 may be
18g3= 475 513243 = 100797 replaced by 5 in Eq. (131).
47s3 = 220 1007973= 145833 EXERCISE 87. If p is an odd prime, r is a quadratic residue of p" if and
only if r is a quadratic residue of p .
2203 = 232 14.58333 = 78152
EXERCISE 88. If mrnis not cyclic, a"'"' = 1 (mod m ) for every a prime
2323 = 1 (mod 487) 781523 = 1 (mod 487') to m. Thus Euler's Criterion can be generalized only to composites of the
Find a primitive root of 4872. Determine the periods of the decimal ex- form 2p" and p" with p odd, and to 4. Further, if mrnis represented by Eq.
pansions of 487T' and 487T2. (126) the product A of all the residue classes a, is given by
t+( m)
EXEI~CISE
80. Find the fallacy in the following: If g is a primitive root -4 = (9192 * . . gr> (mod m ) ,
both of p and of p2, then every primitive root of p has the same property.
and thus A = - 1 or A I +1 according as 3% is cyclic or not. Therefore
(compare Exercise 25 on page 38) Wilmi's Theorem, p l ( p - 1) ! I,+ But Eq. (139) may have an arbitrarily large number of factors of the
like Euler's Criterion, only generalizes to these same composites (Gauss). same order, and therefore Theorem 47 follows if, and only if, there are
infinitely many primesof the form kp,"' +
1 for every prime power p,"'.
36. I'RIRIE:,I N SOME ARITHMETICPROGRESSIOM This is a special case of Dirichlet's Theorem 15 (page 22). But we have
A N D A GEKERAL THEOREM
DIVISIBILITY not proven Theorem 15.
To prove Theorem 47 (page 96), me mill assume, from group theory, Special cases of Dirichlet's Theorem may be proven by variations on
that every finite hbelian group a can be written as a direct product of the proof of Euclid's Theorem 8. Thus if 11 is a product of primes of the
cyclic subgroups. That is form 4k - 1, 4211 - 1 must be divisible by a different prime of that form.
a, = glul.*g2a2r~. . . 9,"",' For if every prime divisor of 4M - I were of the form 4k +
I, so would
(139) their product be of that form. With a similar definition of 212, we may use
for every a , in a. The generator g, is of order m, and the order of Q. is 6 M - 1 to prove that there are infinitely many primes of the form 6k - 1.
the product mlm2. . . m,, . This implies that the cycles of any two generators Using our knowledge of quadratic residues, we may similarly show that
g, and g k have no element in common except the identity g: = gk .
Kow the reader may verify that Lemma 3 above holds for every finite
0
M2 + 2, ill2 - 2 , and +4 M2
Abeliaii group, so that any representation, Eq. (139), may be decomposed must contain a t least one new prime of the form 8k + 3 , 8k - 1, and
into cycles of prime-power order. Assume this done, and that m, is now 8k - 3 respectively. Again
equal to plaJfor p , prime and a , 2 1.
Kow let
$(M2 + I) and $ ( M 2 + 3)
N qiqz ... qn
are divisible only by primes of the form 4k +
1 and 6k +
1 respectively.
1
But it is clear that by such individual attacks we can never prove
n here q , is a prinie of the form kp,"' +
1. Then 3Rn will contain a cycle of Dirichlet's Theorem, since this encompasses infinitely many cases. For
order q , - 1 = kp,"' generated by a residue class s , . Further t , = s,k our present purpose we do not need Dirichlet's Theorem in its full glory:
(mod N ) has a cycle of order p,"' and the subgroup of m Ngenerated by "Thereare infinitely many primes of the form ak +
b for every ( a , b ) = 1."
flQ1*yZa%l . . . t,"",' It suffices if b = 1 and a is any prime power, and this we may obtain by a
very useful generalization of Fermat's Theorem 11.
is isomorphic to a. Theorem 48. Let m = p"-' with p prime and n positive. Let ( a , b ) = 1,
= a , y = bm, and
m
EXAMPLE : x
Let a be an Abelian group of order 9 represented by
a = x"ys
where z and y are elements of both being of order 3. Then is isomorphic Then
toasubgroupof3n91since91 = 7 . 1 3 a n d 7 = 2 . 3 + 1while13 = 4 . 3 + 1.
( x - y, z ) = 1 or p (141)
Specifically, starting with 3 and 2 as primitive roots of 7 and 13 respectively,
and using Lemma 2 with A = 7, g1 = 3, R = 13, g = 2 , etc., n e obtain according as
a representation of 3ng1as
( a - b, p ) = 1 or p.
a = G6"15' (mod91) Secondly,
with 66 of order 6, and 15 of order 12. Then 66' = 79, and = 29 are PI2 or P b
both of order 3, and a is isomorphic to the subgroup of 3 Z g 1 given by:
according as
a = 79"29' (mod 91).
pl(a - b) or p + ( a - b ) .
(Kote that a is also isomorphic t o the subgroup of quadratic residues Q63,
but this mapping is not obtained by the construction given above.) Thirdly, all other prirne divisors of z are of the form kp" + 1.
Before we prove Theorem 48 we shall give several applications.
( a ) If a = b +
I we find that all divisors of
has a t least one prime divisor of the form p"lc +
1. Given &I, the product
of a number of such primes, if a = M and b = &I - 1, we find from Ey.
(b + 1)"" - bm" (140) that the z there contains a t least one more. For every prime power
z =
(b + 1)" - bm (143) p", there are therefore infinitely many primes of the form p"k 1. By the +
construction on page 104 there are therefore infinitely many m, with
are of the form kp" + 1. subgroups isomorphic to any finite Abelian group.
( b ) If, in ( a ) , b = 1, n = 1, and ?u = 1, we obtain Fcrmat's Theorem 11.
(c) If a = 2s, b = I , p = 2 , and n - I = t , we find that PROOF OF THEOREM 48. Let g = (x - y, z ) . Then y = x (mod g ) , and,
from Eq. (140), z = pxp-' (mod 9 ) . Thus glpzp-' and also gl(z - y, p x p l ) .
1 But since ( a , b ) = 1, we have (2, y ) = 1, (x - y, x) = 1, and
z =
(2S)2f+1 -
(2sy - 1
= (2Sy +1
(x - y, 2"') = 1.
has divisors only of the form k2"' + 1. Therefore g = 1 or p . Now, for every c,
(d) I n particular, t = I,
(2SY +I C E C p e C p Z E
by Fermat's Theorem. Therefore

. . . = c (m odp)
is divisible only by divisors of the form 4lc + 1.

(e) And x-y = am - b" 3 a - b (m odp),
w 4+ 1 and if p+(u - b), p+(x - y) and g = 1. But if p l ( a - b ) , y = x (mod p )
and, by Eq. (140), z = pz"-' (mod p ) or plz and g = p . This proves Eq.
is divisible only by divisors of the form 8k 1. + (141) and the first part of Eq. (142).
( f ) And the Fermat Number, obtained from Eq. (144) by s = 1 and
If q is a prime divisor of z, ylxp - yp or a"" = bPn (mod q). Thus q j a and
t = m,
q t b , for if it divided either, it would divide the other also, and this con-
F, = +
22m 1, tradicts ( a , b ) = 1. Let b-' satisfy b-'b = 1 (mod q ) . Then
has divisors only of the form k2"" 1. + (b-'a)'" = 1 (mod y),
(g) If a = 3s, b = 1, p = 3, and n = 1, we find that
and since
(b-'a)'-' =1 (mod y)
by Fermat's Theorem, we obtain, by Theorem 10,
= 4[(6~ + + 31
yl(b-'a)h - 1
has divisors only of the form 6k + 1.
( h) If p = 2 and n = 2, then, if ( a , b ) = 1, where h = ( p " , y - 1 ) . If h # p", we must have hlnz = p"-'. Then
4 4 yl(b-'a)" - I
u - b
Z I P -
a' - b2
- a2 + b2 (14.5) I
or
has only 2 and primes of the form 4k + 1 as possible prime divisors. I Yl(X - Y).
(i) Finally we complete the But, by Eq. (141), q can then be only p , and that only if p l ( a - b ) . All
PROOF
OF THEOREM
47. I n Eq. (143) let b = 1. Then other prime divisors of x , [that is, all prime divisors if p + ( a - b ) ] , have
2mp - 1 h = p" and therefore are of the form p = lip" +
1. This proves the third
t part of the theorem and the second part of Eq. (142).
2m - 1
With the foregoing theory we are now in position (in principle) to map
any finite Abelian group isomorphically into a subgroup of an X, and we may map 3 and 1 into 501 and 751 respectively, then multiply
therefore to carry out algebraic computations within the group by ordinary 501 and 751 decimally. The last three digits of the product are 251
arithmetic. An example is given in the following exercise. We quote from and by mapping backward we find the answer, octal 5. Thus
a recent article in a digital computer newsletter. 3*1 = 5
EXERCISE 89. BINARY AKD DECIMAL MACHIKES as before.
AND ISOMORPHIC OPERATIONS. Now the reader is asked to examine octal biconditional and, by com-
Certain operations, which are easy on binary machines, are paring this with @looo, to show that an isomorphic mapping such as that
awkward on decimal machines, and conversely. In particular, the given follows from the theory above. Is there another mapping into X l o ~
logical AND, OR, and CO?(lPLENENT are naturals for binary which does not use the same eight decimal numbers? Could we use m =
machines while long numerical tables are often more quickly done 100 instead of m = 1000? What is the smallest modulus possible? Find a
on decimal machines since otherwise much machine time is used in mapping for this modulus. From the remarks concerning lobal patterns on
binary-decimal conversion. page 97 describe the cycle graph for Xlooo. Where, in this pattern, are
Sometimes a very binaryish operation can nonetheless be done the eight. decimal numbers utilized above?
decimally by using isomorphic operations. To illustrate this, consider EXERCISE 90. Find a prime of the form 9k +
1 by the recipe given in
the following example.
Let octal biconditional be an operation which is designated by *
the proof of Theorem 47. Find the two smallest primes of the form 3k 1 +
given by that recipe, and compare these with the two smallest primes of
and which is performed on two (three-bit) binary numbers, from 000 the same form which were used in the example on page 104.
to 111. Let
EXERCISE 91. From a book on group theory or modern algebra obtain
A*B = c definitions of quotient group and group of automorphisms. Ir3t (3, be the
where A and B are two such numbers and the result C is a third. group of all integers under addition. Let m > 0, and let em(m) be the mul-
Then the first bit of C is a 1 if the first bits of A and B are equal. tiples of m. Let e, be the group of m residue classes under addition modulo
Otherwise, it is 0. The same rule holds for the second and third bits. m. Then a, is the quotient group e,/e,. And 5n, is isomorphic to the
Examples : group of automorphisms of a,. And therefore every finite Abelian group
is isomorphic to a subgroup of the group of automorphisms of a quotient
3*1 = 5 (octal) group of an infinite cyclic group.
since 011*001 = 101 (binary). Can this characterization of Abelian groups-which seems to involve
only group-theoretic concepts-be proven independently of the number-
Again 5*4 = 6 (octal) theoretic results in Theorem 48?
since 101*100 = 110 (binary). From the relationship between a, and 311, explain the coincidence)
that the number of primitive roots of p and the order of mP-1both equal
This operation, octal biconditional, arose in a practical problem, d P - 1).
namely, clipped autocorrelation. It would seem to be very awkward
to carry it out on a decimal machine. 37. SCALA4RAND V E C T O R I N D I C E S
However, it is isomorphic to multiplication modulo 1000 according If 3 is chosen as the primitive root of 17 we may have two tables:
to the following mapping:
i O 1 2 3 4 5 6 7 9 10 11 12 13 14 15
octal 0 1 2 3 4 5 6 7 ~ _ _ _
a 1 3 9 1 0 1 3 8 7 4 1 2 2 6
1 : : : 5 1 : 5
decimal 999 751 749 501 499 251 249 1
For example, to compute
3*1
110 Solved and Unsolved Problems in Number Theory I The Underlying Structure 111
I n both tables where

a = 3i (mod 17).
ind 0 - ind a )
9
(nlod p-__
1)
. (150)
The exponent i is called the index of a modulo 17 and written
EXERCISE
93. Solve
i = ind a. ( 146)
3x6 = 5 (mod 17).
Similar tables have been worked out for all moduli <2000 which are primes
or powers of (odd) primes. They enable one to multiply, divide, and solve If the modulus does not have a primitive root we must replace the
binomial congruences quite easily for these moduli. For example, scalar indices i with vector indices (i, j , . . . ) . For example, each of the 24
residue classes prime to 33 can, by the foregoing theory, be expressed as
al% = x (mod 17) a = 8'26' (mod 35)
is solved by
with i = 0, 1, . . . , 3 a n d j = 0 , 1, . . . , 5. When the vector index (i,j )
ind z = ind al + ind az (mod 16) has only 2 components a two-dimensional representation is handy. Thus
Thus, for
5.6 =x (mod17)
6 116 131 I
in d x = 4=5 + 15 (mod 16),
and therefore x = 13 (mod 17). Similarly
ax = b (mod17)
271 2 ) l /
is solved by
ind x E ind b - ind a (mod 16).
With indices, as with logarithms, multiplication, division, evolution, and
involution are replaced by addition, subtraction, multiplication, and
division respectively. The general binomial congruence :
axn =b (mod p ) , (147) we have

is treated in
ind ab = (is ,j 3 )
EXERCISE 92. If, in Eq. (147), n is prime to p - 1, there is a unique
solution given by with i3 = il + it (mod nl) and j 3 = j , + j , (mod n,) where the generators
ind x = n-'(ind b - ind a ) (mod p - 1) ( 148)
where n-l is the reciprocal of n modulo p - 1.
If (n,p - 1) = g and g+(ind Z, - ind a ) , there is no solution. But if
gl(ind b - ind a ) , there are g solutions given by
since
(149)
I
Alternatively, we may considrr the table to be continued periodically in EXERCISE 96. The transformation from a 4 X 6 representation to a
both directions. Then ordinary vector addition suffices. 2 X 12 representation of 3n35(Exercise 82) may be interpreted as a linear
The problem of binomial congruences we leave to the reader. transformation whereby a fundamental 4 X 6 rectangle becomes a funda-
We note that by the use of Lcmma 3, page 101, the 4 X 6 table above mental 2 X 12 parallelogram.
can be transformed into a 2 x 12 table, etc. Even for a prime modulus,
38. THEOTHERRESIDUE
CLASSES
say 7, we may modify its one dimensional index into a two dimensional
2 X 3 diagram. Thus After this detailed treatment of m,,, it is natural to ask What of the
residue classes not prime to m? This can be answered quickly. Consider
m = 21. Then besides the 12 solutions of ( 5 , m) = 1, in 3 n 2 1 , there are 6
solutions of (5, m) = 3 , 2 solutions of (z, m ) = 7, and 1 solution of (5,m )
= 21. These three sets of residue classes constitute three other groups under
multiplication modulo 21. These groups have the cycle graphs
1 1 1
D E F /G-!
and the identities 15, 7, and 0 respectively. More generally we have
Theorem 49. If m = A B with ( A , B ) = 1, the 4 ( A ) multiples of B , aB,
we a t once obtain where ( a , A ) = 1, f o r m a group under multiplication modulo m isomorphic
to m A .We caU this group %,,,(). If 0 i s the reciprocal of B modulo A , and
a0 = & (mod A ) , (151)
the isomorphic mapping i s
(modA) a-cB (modm), (152)
and, in particular, PB i s the identity of 311rn(B).
For the cycle of A is obtained by continued repetition of the vector dis- EXAMPLE: Let m = 21, A = 7, and B = 3. Then ,B = 5 (mod 7) and 15
placement from 1 to A , giving us 1, A , B , C, and then, reducing the i co- is the identity of Em,: as shown above. nt:? 3% under the mapping
ordinate modulo 4, back to 1. The continued repetition of the vector (mod 7) a- 3e (mod21)
displacement from 1 to E , again reducing i by 4, or j by 2 when necessary,
gives us 1, E , B, G, 1, etc. The elaborate pattern Sne3is most easily ob- where c E 5a (mod 7).
tained not by multiplication modulo 63 but by addition of two dimensional PROOF.
By Theorem 17 the &sare a rearrangement of the as.
vectors modulo (6, 6 ) . If
EXERCISE 94. Find the pattern of the cycle graphs for SnG3and for Y K ~ ~ ala2 = a3 (mod A ) ,
(Exercise 67) by the use of modulo vector addition. elBc2B= L Y ~ ( Y ~ (I B ) ~ = c3B
, B a3PB (mod A ) .
EXERCISE 95. Show that if the octal numbers of Exercise 89 are sub-
tracted from 7 and written in binary, they may be interpreted as vector And clearly
indices of the corresponding decimal numbers. d ? & B = s3B (mod B ) .
Theref ore PROOF.Any such x is the identity of a multiplicative group modulo m.
-
nzioclBoc2B- &B
( Y I B ~ 2 R c3B
or
(mod m ) .
Thus the + ( A ) multiples of B prime to A form a group isomorphic to
EXAMPLE:
mrn@)
If m = 36 there are four mm(B)
for m = 21. (mm(')
isomorphic to the four
is now our former Mt, .) The remaining 15
residue classes modulo 36 have powers in one of the m36@)although
, they
themselves remain outside. We may diagram these appendages as follows:
m, under multiplication modulo m.
Definition 31. An integer not divisible by a square greater than 1 is
called quadratfrei.
Theorem 50. If m i s divisible by n distinct primes, there will be exactly
2'&multiplicative groups modulo m of the type 3nm(B)described in Theorem
49. I f m i s quadratfrei, each of the m residue classes i s contained in exactly
one of these groups. I f m i s not quadratfrei, those residue classes a satisfying
(a, m) = g with (9,m/g) # 1 (153)
are contained in n o multiplicative group.
PROOF.
If
m = plalpZaZ . . . P,"", mrn(l)never has appendages. These extra residue classes "join" the group
we may clearly choose the B , (and therefore the A ) , of Theorem 49 in irreversibly. Their powers get in, but can't get out.
2" different ways. If na is quadratfrei, each a , equals 1 , and therefore Let us also note
B may be any divisor of m. Since the residue classes in mm(B) satisfy Theorem 51. The 2" multiplicative groups Mt,,,(B) are isomorphic to sub-
(x,m ) = B , no residue class belongs to two of these groups, and, if m groups of =,(') = m, .
is quadratfrei, every possible greatest common divisor, g = (2, m) , occurs
The proof is left to the reader.
as a B. I n this case, then, each residue class is in exactly one group.
But if one or more a, # 1, and if g = spza with 1 5 a < a , , and pLJrs, EXERCISE
97. Interpret the proof of Theorem 1, on page 4, in terms of
let a be a residue class satisfying Eq. (153). Then a = tp," with pl#. 3n10(2).
It follows that 39. THECONVERSE

OF FERMAT'S
THEOREM
T
a z a (mod pl"') If N is a prime # 2,
for no T > 1, and therefore 2N-1 =
- 1 (modN). ( 154)
r
a -a (mod m ) The converse is not true. Thus
for no such r. But if a were in a group of order h, and that group had an 2560= 1 (mod 561)
identity e, we would have
ah = e and a"+' - a (mod m).
Thus if m is not quadratfrei there are still only 2" groups, and all remaining
as in Exercise 78, but 561 = 3.11.17 is not a prime. The smallest com-
posite N which satisfies Eq. (154) is 3-21 = 11.31. In fact
residue classes, Eg. (153), are in no group. It z = 2a3b (mod 341) (155)
Corollary. Thcre are exactly 2" solutions of

x2 = z (mod m)
So 2" = 1 -
is a representation of m341 where 2 is of order 10, and 3 is of order 30.
2340(mod 341).
Definition 32. A fermatian is an integer N which satisfies Eq. (154).
if m i s diviviblc by czactly n distinct primes. +
'Definition 33. A Fermat number F,,, is one of the form 22m 1.
Definition 34. A Carmichael number m is a composite whoselargest charac- which is divisible by 2" - 1 = fz . Further, by Theorem 4, f2 is composite
teristic factor, f , , divides m - 1. See Definition 30. if f1 is. Since 2'' - 1, say, is a composite fermatian, there are infinitely
Definition 35. A Wieferich Square is the square of a prime p such that many of them.
p72p-' - 1. The first ten composite fermatians are
Wieferich Squares enter into the theory of Fermat's Last Theorem. 341 = 11.31 561 (a Carmichael)
Theorem 52. All odd primes, Fermat numbers, Mersenne numbers, Car- 645 = 3 - 5 . 4 3 1105 ( a Carmichael)
michael numbers and Wiejerich Squares are fermatians. There are other 1387 = 19.73 1729 ( a Carmichael)
fermatians, also, since 341, for instance, i s none of these. 1905 = 3.5.127 2047 (a Mersenne)
PROOF. Odd primes are obviously fermatians. Since Fm12'm+1 - 1, and 2465 (a Carmichael) 2701 = 37.73
2m+112'm,we find The 43rd composite fermatian, 31417 = 89-353, belongs to none of the
Fm12P"-1- 1. foregoing distinguished classes, but is perhaps distinguished in its own
right:
Again 314171231416- 1.
M , = 2' - 1 = k p + 1. P. Poulet and D. H. Lehmer have tabulated all composite fermatians
<lo8. We give a table showing C ( N ) , the number of such composites
Then
I N . This is compared with s ( N ).
-
M, = 2, - 112k" - 1 = p p - 1 - 1.
N
A Carmichael number m must be odd, since f, is even, (proof of Theorem
46), and thus could not divide m - 1 if m were even. Therefore, by Eq. 10 -
0 4
( 128) , 102 0 25 -
168
(mod m). 56
103
2m-' E 1 104 22 1229 56
105 79 9592
MSO,if ~ ~ 1 -2 1,~ p 2'1 2 p 2 - 1 - 1. Finally 341 is none of these. It is not a 106 247 78498
121
318
Carmichael number since f r = 30j340. 107 750 664579 886
108 2043 5761455 2820
It has never been proved that
(a) There are infinitely many Mersenne composites, or
(b) There are infinitely many Fermat composites, or Apparently composite fermatians are relatively rare. Of these 2043 com-
(c) There are infinitely many Carmichael numbers, or posites we may note that 252 are Carmichaels, 2 are Mersennes, 2 are
(d) There are infinitely many Wieferich Squares. Wieferich Squares, and none are Fermat numbers. For the entries in the
table we have
Of the last there are only two examples up to p
These are 1093' (Meissner) and 35112 (Beeger) .
= 100,000; (S.Kravitz). C(N) < m. (156)
Nonetheless it is easy to prove Definition 36. If a class of positive integers A contains a subclass B ,
and if A and B are equinumerous, we say almost all A numbers are B
Theorem 53. There are injinitely many composite fermatians. numbers.
PROOF.
Suppose fl is a composite fermatian. Then EXAMPLE : From the prime number theorem, almost all positive integers
f2 = 2'1 - 1 are composite.
While it has not been proven that Eq. (156) remains valid as N + ,
is also one. For if f112"-' - 1, f l f i - 1, and j 2 = 1 + kj1 . Then one is tempted to risk
Q,
p - 1 -1 = -1 Conjecture 15. Almost all jermatians are prime.

Composite fermatians have some interesting properties (Poulet). Their COMMENT: By implication the test is truly infantile since the number
distribution is very irregular. Thus 65,350,801 and 65,350,821 are SUC- doesnt know its mlam - a from its pjaP - a.
cessive composite fermatians, and so are 95,452,781 and 96,135,601-a PROOF. A Carmichael number m is quadratfrei, for if p2/m,we have
gap of 20, and another of more than a half of a million. Very unexpected pl~$(m),and therefore plf, , its largest characteristic factor. But if m = ps,
is the fact that more than one half of these numbers end in the digit 1. and fi = p t , we see that f,+m - 1. Now, since ni is quadratfrei, by Theo-
EXERCISE 98. Prove that 1105 is a Carmichael number. rems 50 and 51, every residue class a modulo m is in an Fnm(B)isomorphic
to a subgroup of Fn, . Thus
EXERCISE 99. The divisibility relation defining Wieferich Squares re-
minds one of the rare primitive roots of p which are not primitive roots .fr =I (mod m)
of p 2 . But show that 2 is not a primitive root of 3511. Nor is it of 1093,
but that is not as easy. where I is the identity of 311m(B).Then a- = I and am= a (mod m ) .
We now seek a better criterion and we decide that Eulers Criterion is
40. SUFFICIENT FOR PRIMALITY
CONDITIONS twice as good a test as Fermats Theorem. If 341 m-ere a prime, since it is
When we left the perfect numbers we were in need of a good criterion of the form 8k + 5, we would have (21341) = -1, and
for the primality of M , . Wilsons Theorem: 2l i 0 = -1 (mod341).
( N - l ) ! = -1 (mod N ) , (157) But since 2 = 1 and 21i0 = 1, we see that 341 does not pass this test.
is a necessary and sufficient condition, but it is not practical. Fermats If a composite N passes Eq. (158), it may be expected to pass
Theorem: 2(N-l)/2 -
= (21N) (mod N ) (160)
- 1
2N-1 = (modN), (158) only one-half the time. Here the Legendre symbol, ( 2 l N ) , is computed
is a necessary and practical condition, but it is not sufficient-as we have as if N n-ere a prime. Sonetheless, Eq. (160) is not sufficient either, and,
just seen. We may even say that it is particularly useless for Mersenne and in particular, all Alersenne numbers satisfy this congruence.
Fermat numbers, in view of Theorem 52. This is unfortunate, for while I n contrast, Eulers Criterion, with a base 3, is a necessary and sufficient
2N-, like ( N - 1) ! , also grows rapidly, it is relatively easy to compute- condition for the primality of Fermat numbers.
% by successive squarings and residue arithmetic. Theorem 55 (Pepins Test). F , = 22m + 1 i s a prime i f and only i f
We note that while N = 341 passes the test of Eq. (158), it does not 3(Fm-1)/2 -
=
pass the test: (mod F,).
-1 (161)
3340= 1 (mod 341), PROOF. I n Theorem 40 m-e showed that if F , is a prime, (31F,) = -1,
and, by Eulers Criterion, Eq. (161) follows. The converse interests us
since, by Eq. (155), 3 is of order 30 and thus 3340= 31 # 1 (mod 341). more. If Eq. (161) is true, so is
But a Carmichael number m passes the test 3Fm-l =
- 1 (mod F,).
urn- = 1 (mod m )
Then if @IF, , 3Fm- = 1 (mod p ) , and the order of 3 modulo p divides
for every a prime to ni. Because of this these numbers are also called F , - 1 = 22m.This order is thus a power of 2. But it cannot divide 2zn-1 =
pseudoprimes. By the results of Sect. 38 we may state an even stronger (F, - l ) / 2 since that would contradict Eq. (161). Therefore the order
result. is F , - 1, and since it must be 5 p - 1, we have F , 5 p . Thus p = F ,
Theorem 54. For every Carmichael number m, and any a, and Eq. (161) is also a sufficient condition for the primality of F , .
The reader mill hear a familiar ring in the argument. We use the fact
mjam - a; (159) that a divisor d of pn, with p a prime, divides pn-, if it does not equal
just as in Fermats Theorem 13: p. If this leads to a contradiction, d must equal p. In Theorem 55 p = 2,
but in Theorern 48 p is any prime.
plaP - a. With this success for Fermat numbers we again inquire about Mer-
senne numbers, 2, - 1 = M , . Here M , again involves a power of 2 ,

but this time M , - 1 is not that power of 2 . Instead M , + 1 is. Here we
CHAPTER III
see the difficulty. What we need are not divisibility theorems like Fermat's
Theorem and Euler's Criterion, since these involve N - 1. We need a
divisibility theorem involving N + 1. Lucas found such a theorem, and
by the use of it he obtained the Lucas Criterion for Mersenne numbers.
The theorem is associated with rational approximations to the &.
When the 4, and earlier, the 1/2,were found to be irrational, there was a
PYTHAGOREANISM AND ITS
great crisis in Greek mathematics and philosophy. We close the present
chapter, and start a new one, which discusses this crisis, and, associated
MANY CONSEQUENCES
with it, another important source of number theory.
EXERCISE
100. If 2" +
1 is prime, m is a power of 2 .
101. From case ( f ) of Theorem 48, page 106, if a prime
EXERCISE
+
plF,, p = 1 k2"+'. Show that 2 is of order 2"" modulo p , and also, 41. THEPYTHAGOREANS
that if m > 1, (21p) = 1. Then 2(p-1)'2-
= 1 (mod p ) , and k is even. We now examine a third source of number theory, one much older than
+
Thus p = 1 ~ 2 if ~m > ' 1.
~ periodic decimals, and even older than perfect numbers.
EXERCISE 102. From Exercises 100, 101, and 4, if we search for the Definition 37. Pythagorean numbers are three positive integers that
smallest prime which divides F s , our first trial divisor is 641. satisfy the equation
EXERCISE 103. Prove that every RIersenne number passes the Euler
Criterion test, Eq. (IGO), as stated on page 119.
a2 + b2 = c2. (162)
The name has a twofold significance. First, it refers to the Pythagorean
Chapter
Chapter III
I11 ::PYTHAGOREANISM
PYTHAGOREANISM AND
AND ITS
ITS MANY CONSEQUENCES
CONSEQUENCES
Theorem concerning a right triangle, and the three integers give us such a
triangle :
The
The Pythagoreans
Pythagoreans
The
The Pythagorean
Pythagorean Theorem
Theorem
The
The Square
Square Root of 2 and
and the Crisis
Crisis
The
The Effect
Effect upon
upon Geometry
Geometry
The
The case
case for Pythagoreanism
Pythagoreanism
a
Three
Three Greek
Greek Problems
Problems
Three
Three Theorems
Theorems of Fermat
Fermat whose sides have an integral relationship to each other. Second, it refers
to the fact that the Pythagoreans gave a formula for infinitely many such
Fermat's
Fermat's Last "Theorem"
"Theorem"
triangles. Namely, if m is odd and > 1, set
The
The Easy
Easy Case
Case and
and Infinite
Infinite Descent
Gaussian
Gaussian Integers
Integers and
and Two
Two Applications
Applications a = m, b = + ( m 2- 1 ) ) and c = +(nz' + 1) (163)
Algebraic
Algebraic Integers
Integers and
and Kummer's
Kummer's Theorem
Theorem EXAMPLES
:
The
The Restricted
Restricted Case,
Case, Sophie
Sophie Germain,
Germain, and
and Wieferich
Wieferich 32 + 42 = 52
Euler's
Euler's "Conjecture"
"Conjecture"
Sum
Sum of
of Two
Two Squares
Squares 52 + 122 =
A Generalization
Generalizationand
and Geometric
Geometric Number
Number Theory
Theory - But there are also two senses in which this name, "Pythagorean" num-
A Generalization
Generalizationand
and Binary
Binary Quadratic
Quadratic Forms
Forms bers, is seriously misleading. First, Neugcbauer has shown that the Baby-
lonians knew of the numbers of Ey. (162)-not merely those given by
Some
Some Applications
Applications
The
The Significance
Significanceofof Fermat's
Fermat's Equation
Eauation 121
Pythagoreanism and its M a n y Consequences 123
But the direct relation between number and spherics, without music
Eq. (163)-at least 1,000 years before Pythagoras. Second, such a desig- as a middleman, was also known to Pythagoras from his travels in Egypt,
nation does not suggest, and indeed tends t o conceal, the fact that origi- and is worth more of our time. We shall not discuss Pythagorean astron-
nally the Pythagoreans thought that every right triangle would have its omy in full. What we need to do is to understand a simple instrument
three sides in an integral relationship by a proper choice of the unit length. called a gnomon, because it exemplifies the Pythagorean synthesis of
Furthermore, this belief was not a casual one but instead fundamental to spherics, geometry and arithmetica.
the whole Pythagorean philosophy. When it was shattered by a number-
theoretic discovery which the Pythagoreans made themselves, a profound
crisis arose in this philosophy and in Greek mathematics.
Pythagoras (570?-500? B.c.) was born on the Greek island of Samos,
traveled in Egypt, and perhaps in Babylonia, and founded a school and
secret brotherhood in southern Italy. We need not go into the ethical doc- /
trines that he expounded. On the scientific side, four subjects were studied;
arithmetica (the theory of numbers), geometry, music, and spherics
(mathematical astronomy). Of these four, arithmetica was considered the
fundamental subject. In fact, the point of the Pythagorean philosophy was
that Number is everything. We should make it clear a t once that Number
here means positive integer. There were no others. Since we are writing
here on the theory of numbers, it behooves 11s to examine this far-reaching
assertion in some detail.
I The gnomon is an L-shaped movable sundial used for scientific studies.
The relationship between number and musical intervals was one of It rests on one leg; the other is vertical. The length and direction of the
Pythagorass first discoveries. If a stretched string of length, say, 12, shadow is measured a t different times of the day and year. If the shadow
I falls directly on the horizontal leg a t noon (when the shadow is shortest),
sounds a certain note, the tonic, then it sounds the octave if the length is
reduced to 6. It sounds the jifth (do to sol) if the length is reduced to 8, ~
that leg points north. The noon shadow changes length with the seasons-
I
and the fourth (do to f a ) if reduced to 9. So Harmony is Number. There minimum a t summer solstice and maximum a t winter solstice. The sunrise
follows a study of means. The fourth is the arithmetic mean of the tonic
I shadow is perpendicular to the horizontal leg during the vernal or autumnal
and octave, 9 = *(12
+
$(A i),
+ 6), while the fifth is their harmonic mean, $ =
since its pitch is half-way between theirs. There also follows a
study of proportion. The fifth is to the tonic as the octave is to the fourth,
i equinox. Thus the gnomon is a calendar, a compass and a clock. Pytha-
goras knew the world was a sphere-the gnomon measures latitude, it
measures the obliquity of the ecliptic, etc. Here we have Solar Astronomy
and the criterion of such proportionality is found in with Number (measurements) as the basis.
8.9 = 12.6. 42. THE PYTHAGOREAN
THEOREM
Since we may write this as I n all such shadow measurements the geometry of similar triangles and
of right triangles is essential. A generation before Pythagoras, Thales of
9.8 = 12.6, Miletus (a commercial center near Samos) also went to Egypt, studied
we also have that the fourth is to the tonic as the octave is to the fifth, mathematics, and started a school of philosophy. It is sometimes said that
etc. The study of means and proportion was an important ingredient of Pythagoras was one of his students. Plutarch tells the story that Thales
Pythagoreanism. determined the height of the Great Pyramid hy comparing the length of
The Pythagorean relationship between music and spherics is less con- the shadows cast by the Pyramid and by a vertical stick of known length.
vincing. The intervals between the seven planets-the Moon, the Sun, Some writers of mathematical history contest this, claiming that Thales
Venus, Mercury, Mars, Jupiter and Saturn-correspond to the seven inter- did not know of the l a m of similar triangles. We believe that he did, but
vals in the musical scale. This cxplains the Celestial Harmony, and shows we need not argue the point. It suffices for the argument which follows
that the Heavens too are essentially Number. We will see later how this that the Pythagoreans did know about similar triangles, a d this fact is
mystic nonsense played a most important role in the history of science. not in question.
am aE
124 Solved and Unsolved Problems in Number Theorg Pythagoreanism and its M a n y Consequences 125
Nor do we raise the questions as to how and where Pythagoras dis- b a
covered the Pythagorean Theorem. He may actually have learned of it
from Egypt, for the rope stretchers there had long known how to con- a
struct right angles with a rope triangle of sides 3, 4 and Fj; perhaps the
Great Pyramid (2700 B.c.) had already been laid out in this way. But we
do raise the question as to how Pythagoras proved (or thought he proved) b
the theorem, since this proof appears to be a critical step in the subsequent b b a b
events.
We conjecture, on the basis of what we have already related, and upon a b
subsequent events which we will relate presently, that the original proof
ran as follows. A number of historians have favored a different opinion-that Pythag-
orass proof was a dissection proof such as that shown above. A square of
side a +b can be dissected into four triangles and the square c2, or into
four triangles and the two squares a2and b2.
We think that this opinion is incorrect on three grounds.
(a) The suggested proof has none of the elements of Pythagoreanism-
no proportion, no means, no Number-as-Everything, no relation to
spherics.
D (b) The suggested proof is very clever, and appears to be of a sort
that could be concocted after one knew the theorem to be true. But this
implies a prior proof-or a t least some serious evidence in the theorems
favor.
(c) The subsequent events, and their culmination in Euclids Elements,
are best explained in terms of the (fallacious) proof which we have sug-
gested.
The Pythagorean derivation of Eq. (163) may date from the same (early)
period as the Pythagorean Theorem. The names square number, cube
number, triangular number, etc., all derive from the Pythagorean
study of the relation between Number and form. The triangular numbers,
1, 3, 6, 10, etc., are the sums of consecutive numbers:
Draw the perpendicular COF. Find the greatest common measure of

e m
the four lines BC, CA, BO and O A . In terms of this length as a unit, let
m e *
the four lines be of length a , b, d , and c-d respectively. Sinre COB and
m e e e
ACB are both right angles and CBO equals itself, the triangles CBO and
ABC are similar. Thus c is to a as a is to d . Here we have a third type of 10 = 1 2 + + + 3 4,etc. The square numbers, 1, 4,9, 16, etc., are the
mean, a is the geometric mean of c and d , and sums of successive odd numbers:
a2 = cd.
Therefore the square C D equals the rectangle OG. Siniilarly CE equals A F ,
and the square on the hypothenuse equals the sum of the squares on the
sides.
126 Solved and Unsolved Problems in Number Theory Pythagoreanism and its M a n y Consequences 127
16 = I + + +
3 5 7, etc. The odd numbers the Pythagoreans called (a) The proof is fallacious.
gnomons. It follows a t once that if m is odd, and if m2 is thought of as a (b) The theorem is put in doubt.
+
gnomon of side + (m 2 I ) , then (c) The theory of proportion, and of similar triangles, is put in doubt.
- I)] = [+(m2+ I)].
m2 + [+(m2 (d) The Pythagorean philosophy is largely undermined. For if Number
(that is, positive integers), cannot even explain a 45 triangle, what
This proves Eq. (163) geometrically. And the first case of Eq. (163) becomes of the much more far-reaching claims?
is the Egyptian triangle, 3 4 - 5 .
If we now look back a t the illustration on page 123, we see the right The Pythagoreans were a secret society, and it is said that their dis-
triangular shadow and, framing the square on one side, the gnomon- coveries were kept secret. But it is also said that Pythagorass lectures
which is really a n odd number, etc. This was the Pythagorean synthesis were well-attended by the townspeople of Crotona. However contradictory
a t its best, and in its happy days-before the trouble began. this may appear, it is clear that Theorem 56 was highly embarrassing.
The (unnamed) Pythagorean who first divulged this startling result is
43. THE 2/z AND THE CRISIS said to have suffered shipwreck in consequence, for the unspeakable and
The source of the trouble is attributed to Pythagoras himself. It is his invisible should always be kept secret.
At a later date a new embarrassment arose. While it was not of quite the
Theorem 56. The equation same crucial character it may also have been considered important. The
2a2 = c2 Pythagoreans knew of four regular polyhedra. and they associated these
with the four elements. The tetrahedron was fire, the cube was earth,
has n o solution in positive integers. the octahedron was air, and the icosahedron, water. But Hippasus, a
PROOF.Assume a solution with ( a , c) = g. Let a = Ag and c = Cg and member of the society, discovered the fifth regular polyhedron, the dodec-
ahedron. By an ominous coincidence Hippasus, for divulging this dis-
(A,C) = 1. (165) covery, was also shipwrecked and perished.
Then 2A2 = C2. Far be it from us to suggest foul play on the basis of such flimsy evi-
dence. Still, we recall that this was in southern Italy-the home of the
But since C2 is even, so must C be even. Let C = 2 0 and Mafia-and that a cardinal principle of the Mafia is silence or quick
retribution. The latter-day Mafia, in Chicago during the Prohibition era,
2A2 = 4D2, or A 2 = 20. was, as we know, involved in the iiunibers racket, and was also inter-
Then A is also even, and since this contradicts Eq. (165), there is no ested in fifths and fourths, and if squealers were seldom shipwrecked,
solution. they were often found, well-weighted, a t the bottom of the Chicago river.
This means that Yet the parallel does not quite run true; it takes a rather vivid imagination
4 z c/a. to picture Little Caesar striding into the back room of the garage on Clark
Street, and snarling, OK, Louie, so you told about Godels Theorem!
It is not a ratio, therefore, from the modern point of view, it is a n irrational Now take dat !
number. But an irrational number is no number a t all-it is (via the But returning to more solid ground, therc is no questioning the fact
Dedekind Cut) a class of classes of ordered pairs of numbers. It is totally that the problems raised by the 4 2 were most scrious. We will examine the
man-made, as L. Kronecker said, and thus is of dubious significance effects of this crisis upon geometry, spherics,) and arithmetica in the
philosophically. next three sections.
To the Pythagoreans, Theorem 56 was a terrible shock. It implies that
in a 45 right triangle (with b = a ) , the hypothenuse and the side are 44. THEEFFECTUPON GEOMETRY
incommensurable. There is no common measure such as we presumed in 1f.our supposition is correct, the order of the day a t this point must have
proof of the Pythagorean Theorem ! The following serious consequences been to
ensue. (a) Devise a sound proof of the Pythagorean Theorem, and
128 Solved and Unsolved Problems in Number Theory Pythagoreanism and its Many Consequences 129
(b) Devise a sound theory of proportion, which could handle incommen-

surate quantities, and therefore restore the important results concerning 47
similar triangles.
Geometry as a deductive science probably began with the Pythagoreans.
We see now that they had a strong motivation. When naive mathematics
leads to paradoxes and contradictions, the day of rigorous mathematics
begins. I n the nineteenth century the paradoxes of the Fourier Series I 35
played a similar role in the motivation of rigorous mathematics; were it lP
not too digressive, we should expound here on the parallelism of the prob-
lems created and of the answers found.
Instead, we skip over 200 years of Greek mathematics, and examine
briefly the Greek answers to problems (a) and (b) above, as they appear
in Euclids Elements.
23P
Euclid gives two proofs of the Pythagorean Theorem-in Book I,
Prop. 47, and in Book VI, Prop. 31. Both proofs use (essentially) the
same figure as we show on page 124. Keither proof has any relation what-
soever to the dissection figure on page 125. The first proof has nothing to 9Pj llP
do with similar triangles-these require a sound theory of proportion, 8
and this is postponed to Book V. Book I is, so to speak, more elementary.
It is clear, by reading it, that the main point of Book I is to prove the
Pythagorean Theorem. This theorem is I, 47, and I , 48, the last proposi-
tion in Book I, is its converse. With few exceptions almost all of the pre-
vious theorems enter into the chain of proof leading to I, 47.
We show this in the following logical structure. The propositions labelled 3P I
{-
p are the problems. We will discuss their role presently. The blank 4
block under 46, and 37 is inserted because both of these propositions
depend upon both 31, and 34.
The proof in I, 47 is based not on similar triangles, but on congruent
triangles. Draw A D and CG in the figure on page 124. Then the triangles I n conclusion we would point out that three important peculiar aspects
ABD and GBC are congruent. But the first equals half of the square CD, of the Elements all bear testimony to the original Pythagorean proof
and the second, half of the rectangle OG. And so C D equals OG, etc. and to the subsequent crisis over the &.
The three theorems concerning congruent triangles-I, 4; I , 8; and I, (a) I n elementary teaching the problems are often thought of as exer-
26; well-beloved of all high school geometry students-all play leading cises, or as applications. Euclid has no use whatsoever for exercises or appli-
roles, as we see in the logical structure. The problems (bisect a line, an cations. The problems are proof that any construction called for in the
angle, construct a perpendicular, etc.) also play leading roles. Number proof of a theorem is indeed possible. The original mistake of Pythagoras,
Find the greatest common measure, etc., was not to be repeated.
plays no role. Proportion plays no role.
(b) numbcr is expelled from Geometry. Aluch nonsense has been written
Book V gives the Eudoxus theory of proportion, the answer to problem
on this point. It has hccn called a peculiarity of the Greek mind-a
(b), and in Book VI we find a second proof of the Pythagorean Theorem,
preference for form rathcr than iuimber-a greater ability in geometry
similar to the one which we have attributed to Pythagoras-but now
thanarithmctic, etc. There is no basis for this. Euclid has three books on
based upon the logically sound Eudoxus theory. There can be no doubt the theory of numbers. The origins of Greek mathematics in Egypt and
that Euclid knew of the earlier proof, and also what was wrong with it. Babylonia were definitely numerical. Pythagorass opinion of Number we
know. The expulsion of number from geometry was solely due to the prob- (a) Galileo (1590) found that during successive seconds from the time
lems raised by the 4 9 . a t which it starts falling, a body falls through distances proportional to
(c) Euclids proof of I, 47 is seldom appreciated in its historical context. 1, 3, 5, 7, etc., so that the total distance fallen is proportional to the square
No doubt Euclid liked the logical simplicity of the fallacious Pythag- of the time. Here we have square numbers arising as sums of the odd
orean proof. But to postpone a proof of the Pythagorean Theorem until numbers (gnomons!).
after the advanced Eudoxus theory can be studied is undesirable. (b) Johannes Kepler was an out-and-out Pythagorean*-one who really
Therefore Euclid gives the most elementary proof he can find, while believed in the Harmony of the Spheres (page 122), etc. He sought for
keeping as close as possible to the original Pythagorean structural frame- many years to find accurate numerical laws for astronomy expressing such
work. When Schopenhauer criticized this Euclid proof of I, 47 as a mouse- harmonies and in 1619 he discovered his important Third Law-the
trap proof, a proof walking on stilts, etc., he showed that he had little squares of the periods of the planets are proportional to the cubes of their
appreciation of the historical, mathematical, and even philosophical points mean distances from the sun.
which were involved. (c) Even before Newtons Principia (1687) it was known to Robert
45. THE C.4SE FOR PYTHAGOREANISM Hooke, Christopher Wren, and others, that the integer exponents in (a)
and (b) imply that each planet has an acceleration toward the sun which
The most important problem concerning the integers is the determination
is inversely proportional to the square of its distance from the sun.
of their role in Nature. The Pythagoreans said Number is everything,
(d) Inspired by Newtons Law of Gravity (c), Charles Coulomb (1785)
but, aside from the analysis of music, we cannot say that they made a good
determined, with a torsion balance, that electrostatic forces were also
case for this assertion. Nor could they be expected to do so, with science
inverse square. Henry Cavendish (1773, unpublished) had already ob-
a t such a primitive level. The mystic and numerological aspects of Pythag-
tained the same law by another method-one which is most instructive
oreanism we now regard most unfavorably. However, these aspects can
for our present discussion. The experiment was repeated by Maxwell a
be ignored. The real difficulty with Pythagoreanism stems from the 4
hundred years litter. The experimenter enters a large hollow electrical
and its corollary that in the analysis of continuous magnitude the integers
conductor. The conductor is charged to a high potential and the experi-
(as such) do not quite suffice.
menter attempts to measure a change of potential on the inside surface.
If we ask whether modern physical scientists believe that the world can He finds nothing-within experimental error. In this way Maxwell es-
be best understood numerically, the answer is yes-practically all of them
tablished that the exponent is -2 with a probable error of 4=1/21600.
do. But here numbers are no longer confined to integers; they also in-
Later experiments reduced the possible deviation from -2 even further.
clude real numbers, vectors, complex numbers, and other generalizations.
The point involved is this: a New Pythagorean might say that Cou-
The founders of modern physical science (Galileo, Kepler, and others)
lombs results merely indicate that the exponent is approximately equal
did not have a rigorous theory of real numbers, but they had the practical
to -2. But the Cavendish-Maxwell experiment not only suggests that it
equivalent, namely, decimal fractions. These, of course, the Greeks did
is exactly -2 but also suggests the reason for this. Mathematically the
not have. The formulation of the laws of nature in terms of ordinary dif-
only law of force which would behave in this way is one whose divergence
ferential equations (Newton), and in terms of partial differential equations
is zero-that is, one that falls off radially in such a way as to just compen-
(Euler, DAlembert, Fourier, Cauchy, Maxwell), appeared to further
sate for the increase in the area of a spherical shell with its radius. Now
weaken the role of integers in Nature and to strengthen that of real num-
this area increases with the square of the radius, and this is so because we
bers. But even here we may note that while the variables in a n equation
live in a space of three dimensions. In effect, then, the fact that the ex-
are continuous, the order of the equation, and the number of variables in it,
ponent -2 is an integer is directly associated with the fact that the dimen-
are integers-a point that should not be neglected.
sionality of space is an integer.
A philosophy which interprets the world numerically, in the general (e) From this interpretation of Coulombs Law (the divergence is zero),
sense of real numbers, we may call New Pythagoreanism, whereas one that
from a similar, inverse square, electromagnetic law due to Andre Marie
insists that the integers are fundamental-not only mathematically, but
-Amp&re(1822), and from other experimental results, James Clerk Maxwell
also physically-we call Old Pythagoreanism. We now inquire whether a
was led to the electromagnetic wave equations in 1865. While the dependent
case can be made for Old Pythagoreanism. To determine this we must
examine a list of some of the key discoveries in physical science. * He even suggested the possibility t h a t the soul of Pythagoras may have migrated
into his own.
and the independent variables here are both continuous, we shall see that of its absolute temperature. We note that although electromagnetism and
in some respects the number of independent variables, 4, and of dependent thermodynamics are both theories of continua (New Pythagoreanism) the
variables, 6, is more fundamental. We will pick up this thread presently. real point of the law is the exponent. Here again the exponent 4 is said to
(f) Prousts Law of Definite Proportions (1799) and Daltons Law of be exact and, in fact, even a casual examination of Boltzmanns derivation
Multiple Proportions (1808) in chemistry directly imply an Atomic Theory shows that this exponent equals the number of independent variables in
of matter. The integral ratios in the second law exclude any other inter- the wave equation-the three of space and one of time. Just as 2 = 3 - 1
pretation. Further, it appears that chemical affinity involves integers in (d), so does 4 = 3 + 1 here.
directly, namely the valence of the elements. (n) But this Boltzmann theory involving continua (m) and his other
(g) I n exact analogy, Lisles Law of Constant Angles (1772) and Haiiys theory (h) involving atoms contradict experiment when combined theo-
Law of Rational Indices (1784) for crystals directly imply that a crystal retically. Thus if electromagnetic radiation is produced by oscillating
consists of a n integral number of layers of atoms. Again, the integral ratios electrons (l), the statistical theory of equilibrium which Boltzmann de-
in the second law exclude any other interpretation. Further, there is a veloped for (h) does not imply Stefans Radiation Law (m). It implies
direct relationship between number and form, e.g., the six-sided symmetry instead the so-called Rayleigh-Jeans Law, which does not agree with
of frozen H20. experiment, and in fact asserts that a n infinite amount of energy will be
(h) The ratio of the two specific heats of air is 7/5 and of helium is 5/3. radiated! I n plain language this erroneous law implies that equilibrium is
While the (New Pythagorean) phenomenological theory (thermodynamics) not possible a t temperatures above absolute zero.
cannot explain these integral ratios a t all, the atomic theory (f) explains To save the situation, that is to preserve both Stefans (m) and Lorentzs
them easily (Boltzmann). By a similar argument Boltzmann explains the (l), Max Planck (1900) found it necessary to assume that the energy was
Dulong-Petit Law for the specific heats of solids. not radiated continuously but discretely in quanta. He gives
(i) Faradays Law of Electrolysis (1834) states that the weight of the
E = hv
chemical deposited during electrolysis is proportional to the current and
time. If chemical weight is atomic, from (f), then this law implies that where E is the energy of the quantum, v is its frequency, and h is a con-
electricity is also atomic. Such electric particles were called electrons by stant. It is interesting to note that this Planck constant h enters into a
Stoney (1891). We will pick up this thread presently. related radiation lam (Wiens Displacement Law) in the form of a ratio,
(j) I n 1814 Joseph von Fraunhofer invented the diffraction grating. A k / h , where k is the Boltzmann constant. Just as h is a measure of the energy
glass plate is scratched with a large number of parallel, uniformly spaced, per quantum, so k is a measure of the energy per atom, The ratio k / h is
fine lines. This integral spacing produces an optical spectrum, since parallel determined experimentally. If atoms are small, then so are quanta
light of a given wavelength, shining through the successive intervals on small, but if matter is not continuous-that is, if k > O-then neither
the glass, will be diffracted only into those directions where the successive is energy continuous, since h > 0.
beams have path lengths that differ by an integral number of wavelengths. But Planck was a New Pythagorean and did not like his (discrete)
(k) The simplest spectrum is th at of hydrogen. The wavelengths of its quanta. He sought for years to circumvent his own (fundamental) dis-
lines have been accurately determined, (j). In 1885 Balmer found that these covery. But the logic is clear. Just as discrete matter implies discrete
wavelengths are expressible by a simple formula involving integers. electricity in (i) so does discrete matter imply discrete energy here-for
(1) Pieter Zeeman (1896) discovered that the lines of a spectrum are the ratio may be determined experimentally in either case.
altered by a magnetic field, and H. A. Lorentz at once devised an ap- (0)Einstein accepted quanta heuristically and in 1905 he used them
propriate theory. The radiating atoms (f) contain electrons (i) whose to explain photoelectricity.
oscillations produce the spectrum by electromagnetic radiation (e). The (p) In the same year, but in quite a different vein, he also developed
frequency of the oscillations (and therefore also their wavelength) is relativity. The Michelson-Morley experiment (1887) had suggested that
changed by the action of the magnetic field upon the electrons. Maxwells Equations (e) must remain invariant to observers traveling
(m) From Maxwells Equations (e) and thermodynamics, Ludwig with different velocjtics. The consequences of such an assumption are that
Boltzmann (1884) derived Stefans Law of Radiation (1879). This states time and space are no longer absolute and distinct, but are related by the
that a blackbody radiates energy at a rate proportional to the fourth power Lorentz Transformation. In the hands of H. Minkowski (1908) this led to
134 Solved and Unsolved Problems in Number Theory Pythagoreanism and its Manu Consequences 135
the four-dimensional space-time continuum. In this theory particular should all be of the form 2n2 was not clear. Indeed, how could it be-for
importance is attached to vectors with four components. One such vector what can mere weight have to do with these other properties?
is a space-time displacement. Another, which we will need soon, is the (s) I n 1911 C. G. Barkla found, by x-ray scattering, that an atom con-
momentum-energy vector, three components of momentum and one of tains a number of electrons approximately equal to one-half of its atomic
energy. A skew-symmetric tensor in this four-dimensional world has six weight.
components-four things taken two a t a time. The most important ex- In the same year E. Rutherford found, by alpha particle scattering, that
ample is the electromagnetic field-three components of electric field, and the (compensating) positive charge, and with it most of the mass, was con-
three of magnetic. centrated a t the center of the atom. This positive charge was about one-
(9) This recalls the fact that the Pythagoreans also considered four to half of the atomic weight. There followed Rutherfords theory of the atom-
be especially important. Thus (they say), the soul is related to fire, and a miniature solar system with the light, negatively charged electrons
fire, as we indicated before, is a tetrahedron, and a tetrahedron has both bound to the heavy, positively charged nucleus by inverse square Coulomb
four vertices and four faces, and is the smallest regular polyhedron. The forces (d).
reader may well consider that we should hastily stow this back in the (t) I n 1913 Niels Bohr assumed that the hydrogen atom had this
closet-and lock the door. But we have our purpose, and since we have (simplest) Rutherford structure (s)-one proton as a nucleus and one
raised the point let us examine it for a moment. electron as a satellite. With the use of Plancks E = hv, ( n ) , he deduced
The Pythagoreans say that a point is of no dimension, two points form the Balmer formula (k) with great precision. However, he had to assume
a line, three points a surface, and four a solid. A tetrahedron has two that the electron could have a stable orbit only if its angular momentum
special properties: it is the smallest polyhedron, and it has the same num- were an integral multiple of h / 2 ~ That
. is,
ber of vertices and faces (i.e., it is self-dual). Both properties follow from
mvr = nh/2a
the fact that its number of vertices is one more than the dimensionality
of space. Let us admit, then, that four is important to Pythagoras for the with m the electrons mass, r the orbits radius, v the electrons velocity,
same simple reason that it is important to Einstein, Minkowski, Stefan and h Plancks constant. The integer n, the principal quantum number,
and Boltzmann: 4 = 3 + 1. made no sense in the Kew Pythagorean theories then in vogue, but its
But why should fire be a tetrahedron? The reader knows that the spec- acceptance was forced by the remarkable accuracy of the theorys pre-
tacular part of fire is the radiant heat and light, and that this is electro- dictions.*
magnetic, and that the six components of this field are obtained by taking (u) 1913 was a good year for Old Pythagoreanism. Soddy and Fajans
the four dimensions of space-time two at a time (p). So likewise the six found that after radioactive emission of an alpha particle (charge +2) the
edges of the tetrahedron join the four vertices two a t a time, and also are resulting element is two places to the left in the periodic table, whereas
the intersections of the four faces two a t a time. But we do not insist upon emission of a beta particle (charge - 1) results in a daughter element one
it. If the reader can find a more fitting regular polyhedron for fire let him place to the right. Together with the earlier results in (s) this Displacement
do so. We now close the closet and return to experimental facts. Law makes it clear that atomic number, not atomic weight, is the important
(r) A most important discovery, and one which is very instructive for factor. This integer is the positive charge on a nucleus, the equal number
our present investigation since it combines New and Old Pythagoreanism, of electrons in that atom, and the true place in the table of elements. This
is Mendelkeffs Periodic Table of the chemical elements (1869). If the was explicitly stated by van den Broek and rapid confirmation was ob-
elements are listed in order of their atomic weights (f), then chemical, tained by Moseley (w).
spectroscopic, and some other physical similarities recur periodically. But (v) In 1912 von Laue made the very fruitful suggestion that a crystal
there were many imperfections and many questions arose. Tellurium (g) would act like a diffraction grating ( j ) for radiation of a very short
weighs more than iodine. But if placed in the table in that order these wavelength.
elcments clearly fall into the wrong groups. Again, the position of the rare -
earths and the numerous radioactive dccny products was not clear. The * While Niels Bohr was applying numbers t o the analysis of spectra, his brother,
raw gases wcrc entirely unanticipated. Further, the table is not strictly Hnrald Bohr, was applying a generalized spectral analysis (almost periodic func-
periodic but has periods of length 2, 8, 18, and 32. Why these periods tions) t o t h e analysis of number (prime number theory).
(w) Henry Moseley (1913) used von Laues suggestion ( v ) to measure For n = 4,for example, we have 16 possible states:
the (very short) wavelengths of x-rays. Optical spectra, like chemical values of ni
behavior, are due to the outer electrons in an atom, and thus have a periodic
character. But x-ray spectra are due to the inner electrons, and these
electrons are influenced almost solely by the charge on the nucleus.
Moseleys photographs show a most striking monotonic variation of the
x-ray wavelengths with atomic number.
Atomic number a t once cleared up most of the difficulties in (r). But
what about 2nZ ?
We note in passing a remarkable neck-and-neck race of x-rays and radio-
T T T T
1 = 0 1 2 3
active radiation :
Gnomons !
X-Rays Radioactivity
(z) But a fourth quantum number was already waiting. In 1925 Uhlen-
Discovery. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ,1895 Roentgen 1896 Becquerel I
Atomic Structure.. . . . . . . . . . . . . . . . . . . . . . ,1911 Barkla 1911 Rutherford beck and Goudsmit discovered the spin of the electron. This gives rise to a
Atomic Number. . . . . . . . . . . . . . . . . . . . . . . . ,1913 Moseley 1913 Soddy fourth number s which can take on two possible values. When this fourth
coordinate is added, with its astonishing rounding out of the little solar
(x) I n 1923 L. de Rroglie applied relativistic invariance of four-vectors system by rotating the planets and thus simulating time, we obtain the
( p ) to Plancks E = hv, ( n ) . The energy E and the time associated with 2n2 states which correlate with the periods in the periodic table. But me
the frequency v are merely single components of two four-vectors. The must distinguish-and also associate-two different harmonies here.
remaining three components of momentum and of space, respectively, In one atom an electron can go from state to state; thus giving rise to the
(p ) , must be similarly related. Thus a particle of momentum mu should spectrum. This is the first harmony. On the other hand, as we go through
have a (de Broglie) wavelength A given by the periodic table, adding one new electron each time, the new electrons
will also take on these distinct quantum states according to the Pauli
A=- h Exclusion Principle (1925). This gives rise to the periodic table-the
mu * second harmony.
Before the rare gases were discovered it seemed as though the (lighter)
When this is applied to Bohrs
elements in the periodic table had a period of 7, not 8, and Newland (1864)
mvr = nh/2a called this the Law of Octaves. He was an Old Pythagorean, but he lacked
the facts.
one obtains
If we thought it necessary to strengthen the case we could continue and
nX = 2ar. discuss isotopes (Soddy) ; hc/2ae2 = 137 (Eddington) ; magic numbers
(Mayer) ; lstrangeness numbers ( Gell-Mann) ; etc. It is not a coinci-
Thus the matter wave has exactly n periods around the circumference of dence, for example, that the three nuclei which are fissionable with slow
the orbit and the interpretation of the electrons stability is that it con- neutrons, U233,U235, and P u ~ ~all
, contain an even number of protons and
stitutes a standing wave. an odd number of neutrons.
( y) This conception was refined in the Schroedinger Wave Equation However it is not our purpose to write a history of science. We asked
(1926). Here there nrr three qiinntum numbers n, I , and m corresponding to whether there is a case for Old Pythagoreanism. We conclude that there is
the dimensionality of space. I n polar coordinates the wave functions cor- -and a strong one. Henceforth we shall call it Pythagoreanism.
responding to I and m are sphcricnl harmonics-no, not Harmony of the
Spheres-but very close to it. I t further develops that the integer I can EXERCISE 104. Draw a diagram showing the historical-logical structure
equal 0, 1, 2, . . . . n - 1 while m ran equal -1, - I 1, . . . . l - 1, 1. + of the discoveries (a) to (z) discussed above.
138 Solved and Unsolved Problems in Number Theorp Pythagoreanism and its M a n y Consequences 139
46. THREEGREEKPROBLEMS
and diagonal c, with the right side of Eq. (169) having the smallest
We now return to number theory and consider three problems which are magnitude possible. The corresponding isosceles triangle approximates a
immediately suggested by (the troublesome) Theorem 56. We recall that right triangle and the ratio c / a is a rational approximation of the d2.
this theorem stated that the equation
c2 = 2a2 (166)
It is interesting that the opposite strategy leads to (essentially) the
same problem. Let the triangle be a right triangle whose (integral) sides
differ by as little as possible, that is, let 6 = d +
1 in Eq. (162). Then
has no solution in positive integers. The first problem is that of generalizing +
from d 2 h2 - ?I = 0 we have
this theorem. While the & is encountered in a 45 right triangle (one- 2a2 + 2cz + 1 - c2 = o or
half of a square), the & is similarly encountered in a 30-G0 right tri-
angle (one-half of an equilateral triangle), and the corresponding equation (26 + 1)2 - 2 2 = -1.
is
We therefore require a solution of Eq. (169) with c odd ( = 2d I), +
c2 = 3a2. (167) a = t, and -1 on the right.
The Pythagoreans knew a t least some of the solutions of Eq. (169).
Equation (167) again has no solution in positive integers, or, we may say, But Theon of Smyrna (ca. A . D . 130) gave
the & is irrational.
Plato states that Theodorus the Pythagorean (ca. 400 B.c.) showed Theorem 58. Let the side and diagonal numbers a, and c, be de-
that &,&, &, fi,4, fl, m, a,
a7 a,
a, and fl Jined by
were all irrational, beyond which for some reason he did not go. The al = 1, c1 = 1;
implication is that Theodorus had no general approach to the problem.
With the use of the unique factorization in Theorem 7, however, it is very a2 = 2 , c2 = 3 ;
easy to prove the more general
a3 = 5 , c3 = 7;
Theorem 57. T h e equation
and, in general,
cn = Nan (168) a,+i = a, + Cn , Cn+l = 2an + cn. (171)
has n o solution in positive integers unless N i s the nth power of a n integer.
Then
PROOF.If c and a are written in standard form:
c,2 - 2an2 = (-1). (172)
c = pl=lp2Q2. . . a = q!lqt2 ...
PROOF.
From Eq. (171)
we see that c, an, and thus also cn/an, have all the exponents in their
standard factorizations divisible by n. Therefore N = cn/an is an nth c2,+1 - 2 ~ 2 , +=~ (2a, +c +
~ -~a ( a) , ~ CnI2
power. = 2a,2 - cn2

There are many deeper solved and unsolved problems concerning irra-
tional and transcendental numbers, but it would be digressive to discuss = - (cn2 - 2a,2).
them now.
The second problem arises by modifying Eq. (166) to read Since c: - 2a: = - 1, Eq. (172) follows by induction.
Several comments are in order. Equation (171), in fact, gives all the
c2 - 2a2 = %I. (169) solutions of Eq. (169), but that has not yet been demonstrated. The
source of the solution Eq. (171) is not indicated here but will be revealed
The motivation is clear. The right side of Eq. (169) cannot be replaced by
rater. Finally we note that the right triangles obtained by Eq. (170),
zero. To best approximate an isosceles right triangle we seek sides a,
from the side and diagonal numbers for n odd (and > 1) , are given by
the triples: This is solved by

(3, 4, 5 ) ; (20, 21, 29); (119, 120, 169); etc. Theorem 59. If a, 6 , and c are positive integers which satisfy
These agree with the Pythagorean sequence, Eq. (163) : a2 + b2 = c2, (177)
(3, 4, 5) ; (5, 12, 13) ; (7, 24, 25) ; etc., it is su&ient i f they are given by
only in the first triangle. a = s(2uv), b =
2
s(u - v ) ,
2
and c = s(u2 + v2), (178)
Theorem 58 has an important generalization but some modification is with u > v, and Eq. (178) is also necessary providing we are willing to inter-
necessary. For example, if we replace 2 by 3 in Eq. (169) and choose the change the formulae for a and b i f this is necessary.
negative sign:
COMME~VT:The sufficiency was given by Euclid, Book X, Prop. 28, 29,
c2 - 3a2 = - 1 (173) but was known to the Babylonians more than 1,000 years earlier (see
we obtain an equation with no solution. That is clear since it implies page 121).
c2 I - 1 (mod 3 ) , and we know that is impossible. But if we choose the PROOF.Since
plus sign and if N is not a square, the equation
c2 - Nu2 = I
(2uv)2 + ( u 2 - v2))"= (u2+ v2))"
(174)
always has infinitely many solutions. This important theorem of Fennat is an identity, the sufficiency of Eq. (178) is obvious. Suppose ( a , b ) = s
we postpone until later. in Eq. (177). Then SIC and let a = sA, b = sB, and c = sC. Then
If N = 3, we have A' + B2 = C2 (179)
c2 - 3a2 = I (175) with A , B , and C all prime to each other. A and B are not both odd, for
and, while Eq. (173) is impossible,
if so A2 + B2 = C2 is of the form 4m +
2, and this is impossible. h'or are
they both even, since ( A , B ) = 1 . Without loss of generality let A be even
c2 - 3a2 = -2 (176) and B be odd, and therefore C is odd. Then
is not. I n his famous Measurement of a Circle Archimedes obtains
33 > ?r > 338
(); =
C
-. - B
2
C +B
___
2 -
(180)
and in deducing these inequalities he uses But ( C +

B ) / 2 and ( C - B ) / 2 are prime to each other, for, if not, their
sum C and difference B would not be either. By Theorem 7 and Eq. (180),
-->a>-.
1351
780
265
153
+
( C B ) / 2 and (C - B ) / 2 are therefore squares, say u2and v2.
Therefore
The reader niay verify that these good approximations to 4 3 , (call them 2 2
c / a ) , satisfy Eqs. (175) and ( 176) respectively, so that Archimedes knew A = 2uv, B = u -v, and C = u 2 + v2. (181)
a t least some solutions of these equations. Then Eq. (178) follows.
EXERCISE 103. From one of these Archimedean approximations to the
Corollary. All Pythagorean numbers
4 3 , and by an approach similar to Eq. (170), deduce the fact that (451,
780, 901) gives a right triangle which is approximately 30400. A2 + B2 = C2
The last exercise, and the two series of Pythagorean numbers given
with A , B , and C prime to each other, and with A even, are given by Ep. (181)
above, suggest the third problem-that of finding all solutions of
where u and v are prime to each other, one being odd and one even. These
a2 + b2 = c2. triples are called primitive triangles.
The 12 smallest primitive triangles listed according to hypothenuse are : concept. What can primality have to do with a sum of squares? We will
return to this point and theorem.
A B C U With Eqs. (170) and (171) we obtained infinitely many primitives with
1 A - B 1 = 1. The column I A - B 1 above has familiar looking numbers,
4 3 5 2 from our studies of the factors of M , ,and suggested to Frenicle and Fermat
12 5 13 3
8 15 17 4
that every prime of the form 8m f 1 is the difference of the legs of in-
24 7 25 4 3 17 finitely many primitive triangles. Since
20 21 29
12 35 37
5
6
2
1
1 IA - B I = I(u - v ) 2 - 2v2 I
23
40 9 41 5 4 31 the implication is that every prime p of the form 8m f 1 can be written as
28 45 53 7 2 17
60 11 61 6 5 49 *p = x2 - 2y2
56 33 65 7 4 23
16 63 65 8 1 47 in infinitely many ways. Together with Theorem 60 we are led to consider
48 55 73 8 3 7 the numbers
x2 N y 2 +
EXERCISE 106. I n how many primitive triangles is 85 the hypothenuse? I
for every integer N . This brings us to an extensive subject-that of binary
What about 145?
quadratic forms. We may note that while perfect numbers and periodic
EXERCISE 107. If un and vn are prime to each other and both odd, show decimals lead to quadratic residues only a t a deeper level, with Pythagorean
that the A , B , C obtained from Eq. (181) equal 2B', 2A', 2C' for some numbers they arise a t once. For if a prime p is given by
primitive triangle: A', B', C'. Determine the u and v for this triangle in
terms of un and vo .
p = x2 + Ny',
then, since (yl p ) = 1, we have
47. THREETHEOREMS
OF FERMAT
Just as Theorem .56 led the Greeks to the t'hree problems discussed above, (y-lx) E (mod p ) .
so did Theorem 59 lead Fermat to three important theorems. Each of It is interesting to note that the two square roots which were most fruitful
these, in turn, led to an important branch of number theory. We will prove historically in forcing an extension of the number system, namely Ir-1
none of these theorems in this section but will state all three-in a survey and dlwere also those which arose earliest in these binary quadratic
fashion.
Perhaps the most important is
+
forms, x2 N y 2 .
Further examination of the column C raises other questions. The hy-
Theorem 60. Every prime of the f o r m 4 m + 1 i s the sum of two squares pothenuse 65 arises twice:
in a unique way. 65 = 72 + 4' = 82 + 12.
EXAMPLES :
I n how many ways is an integer a sum of two squares?
5 = l2 + 22, 17 = l2 + 4') 37 = + G2,
1' And, of course, some numbers cannot be written as sum of two squares.
But, of these, some are a sum of three squares, and some of four. Thus
13 = 22 + 32, 29 = 2' + s2, 41 = 42 + 5'.
14 = 9 + 4 + 1 ; 7 = 4 + 1 + 1 +l.
This theorem, which had already been stated by Girnrd several years
Following an earlier statement by Bachet, Fermat proved
earlier, is, of course, suggc,i.tcd hy the third column of the foregoing table
+
and the formula C = u2 v2. In example ( h ) of Theorem 48, page 106, Theorem 61. Every positive integer n i s expressible as
we have seen that if C is prime it is of t,he form 4m +
1. But to prove
II
Theorem GO we need thc (harder) converse and also the uniqueness. The
theorem is rather surprising, since primality is a purely multiplicative
I where w , x , y , and z are integers, positive or zero.
Like Theorem 60, Theorem 61 is rather surprising, and rather hard to The Corollary of Theorem 62 is, of course, the case 72 = 4. The reader
prove. Euler was unable to prove it although he worked on it for years. probably knows that no general proof has been found, although it has
Through its generalization, Warings Problem, it became a major source been attempted by Euler, Legendre, Gauss, Abel, Dirichlet, Cauchy,
of additive number theory. A sketch of a proof of Theorem 61 is given in Bummer, etc.; that Paul Wolfskehl, a wealthy German interested in
Exercises 315-335, on page 209. The first published proof is due to La- number theory, offered a reward of 100,000 marks in 1908; that Hugo
grange. Stinnes, a wealthy German not interested in number theory, helped bring
From a sense of symmetry the reader probably can guess what comes on the German Inflation in the 1920s and thus (incidentally) reduce the
next. If a sum of two squares leads us to consider a sum of four squares value of this prize considerably; and that (nonetheless) much further
on the one hand, it should also lead us to consider a sum of two fourth effort has been expended by thousands of professionals and amateurs with
powers on the other. I n the foregoing table either A (as in 4, 3, 5) or B no conclusive result. According to Professor Mordell, there are easier
(as in 40, 9, 41) may itself be a square, but not both simultaneously. This ways to make money than by proving Fermats Last Theorem.
result is closely related to an impossible problem of Bachet-to find a We will first give an interesting approach which makes the conjecture
Pythagorean triangle whose area is also a square. (We may call this the plausible. The reader knows that if g ( x ) is a rational function of x ,
problem of squaring the triangle-in integers.) This problem may be
shown to imply the following condition:
a4 - b 4
= c .
2
is integrable in terms of elementary functions-that is, a finite combination
Fermat proved Eq. (183) impossible, and similarly he proved of algebraic, trigonometric and exponential functions together with their
Theorem 62. The equation inverses. Or, again, say,
a4 + b4 = c2 (184)
has no solution in positive integers.
is so integrable. But
Corollary. The equation
a4 + b4 = c4 (185)
has no solution in positive integers.
is not elementary-it is an elliptic integral.
PROOF OF THE COROLLARY. A fourth power is a square. Chebyshev has proven that if U , V , and W are rational numbers, then
We will prove Theorem 62 later. The corollary is in striking contrast
with Theorem 59 where there are infinitely many solutions. The corollary
is the only easy case of Fermats Last Theorem. We will consider this
celebrated conjecture in the next section. While it is sometimes stated to is integrable in terms of elementary functions if and only if
be a n isolated prohlem-of no special significance-it was, in fact, one of
the main sources of algebraic number theory.
48. FERMATSLAST THEOREM
It is well known that Fermat wrote that he had a remarkable proof of is an integer. I n Eq. (188) we have A = --B = I , U = 0, V = 4, and
W = 3. But neither 2 , nor 3, nor $ is an integer.
Conjecture 16. The equation If in Theorem 59 we set x = a / c and y = b / c and t = v/u we find that
an + b = cn all rational solutions x and y of
has no solution in positive integers i f n > 2. x2 + y2 = 1
146 Solved and Unsolved Problems in Number Theorg Pythagoreanism and its Many Consequences 147
are given by
Now it is clear that if k could be an integer >2, Eq. (192) would have
infinitely many rational solutions (by choosing any rational t ) and thus
(191)
ak + bk = ck
where t is an arbitrary rational number.
Now, in Eq. (186), let us similarly write x = a/c and y = b/c and would have infinitely many integral solutions. But although Eq. (192) is
generalize the exponent n to be any rational number k . Thus not solvable in rational functions, this does not preclude, a t least according
to any known argument, a solution in terms of rational numbers. Although
xk yk = 1. + (192) the existence of such rational functions would disprove Conjecture 16,
We now ask, following the example of Eq. (191) : Are there rational func- their nonexistence does not prove it. The approach here is therefore only
tions suggestive-it proves nothing about Conjecture 16, but it does show the
special role of n = 1 and n = 2 .
z = f(t) and y = g(t) Three comments of general mathematical interest are these.
such that Eq. (192) is satisfied identically? (a) The use of the transformations, Eq. (191), for rationalizing in-
If k = l / q , for q a nonzero integer, the answer is yes, since we may set tegrands involving y = 4- is well known to calculus students. We
see here the intimate connection with Pythagorean numbers.
z = t* and y = ( 1 - t)'. (b) The reader notes that we have not previously used methods in-
And, if k = 2/q the answer is yes, since we may set volving functions and integration, and may well ask, "What have these
to do with number theory?" The question is well taken and in fact it may
be stated that here, at least, the influence really goes in the opposite direc-
tion. The proof of Chebyshev's result, Eq. (190)--see Ritt, Integration in
But for any other rational number k no such rational functions exist. Finite Terms, Columbia University Press, 1948, p. 37-43 based on certain
For consider y = ( 1 - z k ) l i k
and the integral characterizations of the algebraic functions z u ( A +
B Z " ) in
~ terms of
integers-namely, the number and order of the so-called branch points.
1 (1 - xk)lik dx = / ydx. (193)
It is not so much that algebraic functions have number-theoretic implica-
tions as that numbers have functional implications.
If z = f ( t ) , and y = g ( t ) , by the change of variable z = f ( t ) the integral (c) We are impressed here with the fact that although Conjecture 16
becomes has so far resisted all attempts a t proof, the analogous theorem in terms
of functions is relatively easy. There are other examples of this phenomenon
in number theory. For example, there is a theorem analogous to Artin's
Conjecture 13 which concerns functions, not numbers, and this has been
and since this integrand is a rational function, the integral is elementary. proven by Bilharz. It would take us too far afield to elaborate.
But, by Eqs. (189), (193) and (190), we must have
49. THEEASYCASE AND INFINITE
DESCENT
1 1 2
- or - or - To prove Conjecture 16 it would clearly suffice to restrict the variables
k k k
in
an integer, say q. Therefore we must have
a" + b" = C" (TZ > 2)
k = l/q or k = 2/q
as follows:
and this condition is not only sufficient, but also necessary. I n particular
(A) a, b , and c are prime to each other, and
k # 3,4,5, * * * . ( B ) n = 4 or n = p , an odd prime.
For if ( a , b ) = s > 1 we may proceed as on page 141 in the proof of The- of positive integers
orem 59, and if n # 4 or p , it equals 4k or p k for some k > 1. But c > y > y1 > yz > *.. > 0.
a + b = C Since this infinite descent is impossible, there is no solution.
is then impossible if We now analyze this proof for any light it may throw on Conjecture 16,
and note three features:
+
(ak)> (bk14 = (2)) (A) The proof leans heavily on Theorem 59, but this is possible only
and because 4 = 2 . 2 and thus is not extendable to odd exponents.
(B) As in Theorems 56, 57 and 59, the unique factorization of Theorem
(ak)> + = (2)) 7 plays an important role-in distinction to, say, Chapter 11, where the
are impossible. Fundamental Theorem was hardly used a t all. We should expect unique
The only easy case is n = 4, and therefore also n = 4k.The impossibility factorization to be important for Conjecture 16.
of this case me now prove. The proof is similar to that which Fermat gave (C) The infinite-descent strategy, like point B, is not peculiar to n = 4,
for Eq. (183),and this latter proof is noteworthy in two ways: and we may expect it to be useful for Conjecture 16.
(A) Of all Fermats theorems this is the only one for which his proof is Despite its rather exotic name it should be noted that infinite descent
known. is essentially the Well-Ordering Principle, i.e., every nonempty set of
(B) The proof uses infinite descent, a method Fermat recommended positive integers contains a smallest member. As is well known, this prin-
highly, which he used both for negative propositions such as Theorem 62, ciple is equivalent to the principle of induction-and thus is the most
and with some modification, for positive propositions such as Theorem 60. characteristic of all the laws concerning the integers. The reader may note
that in the proof of Theorem 7 itself (page 6), and of the mderlying
OF THEOREM
PROOF 62. Assume
Theorem 5 (page 9), the Well-Ordering Principle is used several times.
A4 + B4 = C2 (194) 50. GAUSSIANINTEGERS
AND Two APPLICATIONS
where A , B , and C are prime to each other, and, without loss of generality,
To attempt Conjecture 16, the analysis above suggests that we utilize
let A be even. Then, by Theorem 59, Eq. ( l S l ) , we have
points B and C there while dropping point A. We introduce this possibility
A 2 = 2uu, B2 = u2 - v2, and C = u2 + v2 by returning first to the paradox raised on page 143. Given a prime p of
with u prime to v. Then B2 + v2 = u2,and since B is odd, v is even. Thus
the form 4m + 1, and given, by Theorrm 60,
v = 2rs and u = r2 + s2 p = a2 + b2, (196)

we repeat, What has the multiplicative concept of primality to do with a
with r prime to s. Since
s u m of two squares? We can write Eq. (196) in a purely multiplicative
A 2 = 2uv = 4rs(r2 + s2), manner:
by Theorem 7, r , s, and u = r2 + s2 must all be perfect squares. Let p = (a + & ) ( a - bi) (197)
r = a2, s = @, and u = y2. where i = J? and p is a product of the two complex factors. This is a
Then rather ironic solution of the paradox, since in terms of these factors p is no
longer a prime !
a4 + p4= y2 (195) Definition 38. Gaussian integers are numbers of the form a +
bi, where
with ySu<C. a a,nd b are integers.
EXAMPLES
:
Given a solution, Eq. (1941, n e could thus find a second solution, Eq.
(195), whose right side is smallrr. But this implies an infinite sequence 2 + 3i, 4 - 7i, -2i, 7
We will give only a brief sketch of these integers. The sum, difference, for some a and h. Since p is real the quotient must be c ( a - b i ) . Thus
and product of two Gaussian integers are Gaussian integers, but +
p = c ( a 2 b2), and since c must he 1, wc have p = u2 b2. +
a + bilc + d i Pitooi~THAT EQ. (183)
PARTIAL IS IMPOSSIBLE.
Assume
only if there is an e + f i such that A 4 - B4 = C', B 0 (m od2) (198)
(a + b i ) ( e +fi) = (ae - bf) + (af + be)i = + di.
c with A , B, and C prime to each other. Then C is odd and
That is, the e and f obtained by solving ae - bf = c and af + be = d A4 = (C + i B 2 ) ( C - iB').
must both be integers. A u n i t y is a divisor of 1, namely, 1, - I, i, or -i.
Two numbers are associated if their ratio is a unity. A prime is not a unity, By unique factorization C +
iB2 is associated with a perfect fourth power.
and is divisible only by associates of I and of itself. Prime to each other Assume first that it equals a fourth power or its negative:
means having no common divisor other than a unity.
Consider all ordinary integers, positive, negative, and zero. Let +1 and
C + iB2 = (D + iE)4 or = -(E + iD)4.
-1 be the unities. Let a and -a be associated, and let Then
=t2, f3, f5, etc. B2 = 4DE(D2 - E 2 ) , ztC = D4 - 60%' + E4.
be the primes. The fundamental theorem (Theorem 7) may be extended Since D, E , and 0' - E' are prime to each other they are perfect squares.
to all integers as follows: Let D = a', E = @,and D2 - E' = y'. Then
Theorem 63. Each integer not zero or a unity can be factored into a product a4 - p4 = y2, (198a)
of primes which i s unique except for a possible rearrangement, and except and, since 4P21B2,fl < B. Since C is odd, D and E are not both odd. And,
f o r a possible substitution of associated primes. since D' - E' is a square = 4m +
1, we must have E and therefore /3 even.
EXAMPLE : Finally, C + +
iB2 # & i ( D iE)4, since equality here implies that C is
even. Thcn, by infinite descent, from Eq. (198a) and /3 = 0 (mod 2), Eq.
-15 = 3(-5) = 5(-3), etc. (198) is impossible.
Now we state, without proof, that Theorem 63 is also true for Gaussian By a somewhat similar approach, using a generalized unique factoriza-
integers. Assuming this, we will give two applications. tion, and infinite descent, we now examine Conjecture 16.
PARTIAL PROOF OF THEOREM 60. on the basis of this unique factori- EXERCISE 108. Show that Bachet's problem (page 144) is equivalent
zation we will show that every p = 4m +
1 is a sum of two squares. Since to Eq. (198) and therefore impossible.
(-1 I p ) = 1 there is an s such that
AND KUMMER'STHEOREM
IKTEGERS
51. ALGEBRAIC
PIS2 + 1. We generalize Gaussian integers and sketch the following. A root z of a
Let the quotient be q and polynomial with integer coefficients is called an algebraic number. The set
of all numbers
pq = s2 +1 = (s + i ) ( s - i).
Now p cannot be a Gaussian prime, for if it were, by the unique factoriza-
+
tion of s2 1, we would have pis +
i or pls - i. Since neither quotient which are rational functions (with integer coefficients) of x is called an
is a Gaussian integer, p is not a Gaussian prime. But it is not divisible by algebraic number Jield, k ( z ) . The numbers of such a field which are roots
a real prime. Therefore of a polynomial
a + biip wn + awn-' + . . . + s = 0
152 Solved and Unsolved Problems in Number Theory Pythagoreanism and i t s M a n y Consequences 153
with integer coefficicnts, and leading coefficient, 1, are the algebraic integers EXAMPLES
:
of that field. 1 1 1 7709321041217
B,=gl
Bz = - ,
30
BS ...
=7, , =
510
EXAMPLE : 42
The Gaussian integers arc the algebraic integers of k(-) for if a Definition 40. A prime p is regular if it divides none of the numerators of
and b are ordinary iiitcgrrs, a + bi is a root of
w 2 - 2aw + +
a2 b2 = 0, BI , Bz , B3 , * . . , B(p-3)/2
and i t may be shown that all other numbers in k ( d q ) are not roots of when these numbers are written in their lowest terms. Otherwise p is
a polynomial with leading coefficient 1. irregular.
Unities, associated numbers, and primes are defined as beforc. If Theorem
63 held for the algebraic integers of any field then Conjecture 16 could be EXAMPLE
:
shown to follow. Since 3717709321041217, and the larger number is the numerator of
As2wme BIG, and 16 5 a(37 - 3 ) , 37 is irregular.
A + BP = C p (199)
Theorem 64 (Kummer, 1850). Fermats Last Theorem i s true for every
exponent which i s a regular prime. The only irregular primes u p to 100 are
with .4, R , and C prime to each other, and p an odd prime. Let, 37, 59, and 67.
2raip
p = e (200) COMMENTS :
and w e may then factor thc left side of Eq. (199) as follons (A) The definition of regular is explicit but complicated; it has no ap-
parent relation to the problem. There is a more basic definition in terms of
(A + B ) ( A + p B ) ( A + p2R) . . . ( A + ppP1B) = Cp. (201) the so-called class number but this is less explicit numerically and would
It may be shown that the algebraic integers of k(p) are take longer to explain. This latter concept is fundamental, but is beyond
our scope.
a + bp + cp2 + . . . + spP2 ( B ) The name irregular is really misleading. Although only 3 of the
where a , b, . . * are ordinary integers. first 24 odd primes, 2 < p < 100, are irregular, larger primes are ir-
We have, therefore, as in Eq. (197), turned an additive problem into regular more often. Of the 367 primes, 2 < p < 2520, 144 are irregular;
a purely multiplicative problem. and of the next 183 primes, 2520 < p < 4002, 72 are irregular. These
Now if these algebraic integers had unique factorization we could deduce ratios:
from Eq. (201) that each factor on the left is associated with a perfect 144 -
_ - .392 and
72
= .393,
~
pth power of an algebraic integer. If this were always true, Fermats Last 367 183
Theorem would follow. E. E. Kumnier, A. L. Cauchy and G. Lam6 all are substantial.
assumed that such uniqueness did exist. However, Dirichlet pointed out (C) Other criteria have been found, besides Theorcm 64, and applied
that this must be proven. I n fact, it is not true in general-the first counter- to the irregular prime cases. With the aid of the SWAC, Selfridge, Nicol
example being p = 23. To overcomc this lack of unique factorization into and Vandiver proved that Conjecture 16 is true for all exponents 54002.
primes, Kummer was led to introduce the important, underlying ideal Rut with Kummers regular primes, and other primes allowed by other
numbers, a development we cannot enter into here. known criteria, it has not yet been proven that there are infinitely many
With this theory Kunimer obtained a proof of Conjecture 16 for many valid prime exponents.
prime exponents n. We will state his remarkable result but not attempt to Before leaving nonunique factorization let u s examine a few examples.
prove it, as the proof is long arid difficult. Consider the quadratic fields k ( 4 N ) where, without loss of generality,
Definition 39. The Bernoulli number B , is a rational number defined by N is qiiadratfrei. Of the 12 cases, N = -7, -6, -5, -3, -2, - 1 , I , 2,
the power series: 3, 5, 6, 7, only in k ( 4 q ) and k ( d 3 ) do the integers not have unique
factorization. Wc $;how two well-1cnon.n examples. In k ( v q ) ,
21 = 3.7 = (1 + 2&-5)(1 - a n )
154 Solved and Unsolved Problems in N u m b e r Theory Pythagoreanism and i t s M a n y Consequences 15.5
although the four factors here may all be shown to be primes. I n k ( 4 16), Theorem 65 (Germain, modified). T h e eyuation
6 = -dq(,/q) = 2 . 3 A + B = C
and again the factors are primes. Finally we note, in passing, that in the
corresponding two quadratic forms,
has n o solution in integers prime to p i f p is a n odd prime, and q = 2p +1
is also a prime.
a2 + 56 = (a + -\/--5b)(a - m), PROOF. Assume a solution. We may take A , B, and C prime to each
a2 + 6b2 = ( a + -)(a - &%), other, and since p is odd we may write Eq. (203) symmetrically:
if we set a = n and b = 1, and consider R -/- S + T =0 (204)

n2 +5 and n2 +6 where A = R, B = S, and - C = T . Consider
we obtain forms with an exceptionally low density of primes. See the table + T = (-R).
Sp
on page 49. This is not a coincidence-the low density is really related
to the nonunique factorization-but the argument is a long one. Both sides are divisible by S + T by Theorem 40 on page 17, and since
p+R, we obtain p + S + T . Now let 1 in Theorem 48, page 105, and
=
52. THE RESTRICTED
CASE,SOPHIEGERMMIL,
AND WIEFERICH let
Sophie Germain, a Parisian lady, was a contemporary of Gauss. Since S = x = a and T = -y = -b.
the &ole Polytechnique did not accept women in the school she took corre-
spondence courses. She wrote Gauss after his Disquisitiones appeared Since p+S + T = a - b, by Eq. (141), S + T = x - y is prime to
telling him how much she liked the book. She included some of her own +
S P T P- ( - R I P
discoveries and signed-as talented ladies did in those days-with a
male pseudonym, hl. Le Blanc. Gauss was impressed. Only later, under
= S+T S+T
interesting circumstances, did he learn that M. Le Blane was a lady. Therefore, since
Gauss was astonished and pleased. Henceforth their correspondence was
not strictly technical; he told her his birthday, etc. ( -R) = FT( S +
(-R)p * T),
There is a special case of Conjecture 16 which is substantially easier.
I n this, the restricted case, it is assumed, in both factors on the right are perfect pth powers. Write
A + B = C
that p does not divide A , B , or C. Since it is possible that this case is true
Similarly, by symmetry,
while Conjecture 16 is false we state it separately.
Conjecture 17. T h e equation T + R = sP, +
T P RP -
- up, -8 = su.
T+R
a + b = c +
(206)
. .
has n o solution in integers not divisible by p . R +S = tP, RP Sp = p,

R+S
-T = t7.
The Encyclopaedia Britannica ( 1960) states erroneously that Sophie Therefore

Germain proved this conjecture; the article should add: for p < 100.
We give a cut-down version of her result. It shows 2R = + t - r.
s (207)
(A) How far one (*ango with w r y elementary arguments, Now, by Eulers Criterion, if q = 2 p + 1 does not divide R , S, or T ,
(B) That the restricted case is indeed easier, and we have
(C) There is a relation to our Conjecture 5. See pages 30 and 31. R, S, T P all = f l (mod 4).
Pythagoreanism and its Many Consequences 157
136 Solved and Unsolved Problems in N u m b e r Theory
difficulty comes when the restriction is waived. Further, as we shall soon
By Eq. (204) this is impossible. On the other hand, q cannot divide two
see, there can be little doubt that Conjecture 17 is true. Still, it has not bern
or thrce of R , S , and T since they are prime to each other. Therefore q
proven-not even for infinitely many p a s already stated in point ( U ) .
divides exactly one of them. J,et it be R. From Eq. (207) it therefore
follows similarly that q divides exactly one of r , s, and t , and by Eq. (206)
(E) Unique factorization is again fundamental. (Where does it enter?)
it milst be r . In 1909 A. Wieferich showed that Conjecture 17 is true if
Then, since q 1 R, from p2+2P- - 1, (208)
aP(T + R ) = T P + RP that is, if p2 is not a Wieferich square (see pages 116, 118). This criterion
we have aPT = TP (mod q ) , is therefore sufficient for all p < 100,000 except 1093 and 3.511. However,
= P _= y p - 1 despite the fact that these squares are so rare, no one has proven that
or (mod a ) . there are infinitely many p which satisfy (208). Further, D. Mirimanoff
Kow, since q t a, subsequently (1910) showed that
Tp- = & l (mod q ) . p2+3P- - 1 (209)
From Eq. (203), S = - T (mod a ) , and therefore, from was a n equally valid criterion. Therefore the rare Wieferich squares must
also violate the equally prevalent (209) if we are to discover a counter-
example for Conjecture 17.
With these and other similar criteria, and following many previous
pp = pSP = pT* = f p (mod q ) . authors, D. H. and Emma Lehmer showed that Conjecture 17 is true for
This is impossible, since (p I a) = & l , and by Eulers Criterion the all primes <253,747,889.
theorem is proven. EXERCISE 109. Show that Conjecture 17 is true for p = 3 since if 3+a,
COMMEXTS :
a3 3 f l ( m o d 9 ) .
(A) Therefore Conjecture 17 is true for p = 3,5,11,23, * . . ,16188302111, EXERCISE 110. Show that the 24 odd primes < 100 satisfy one or another
... of the six criteria in point (C) , page 156.
(B) If Conjecture 5 were true, Conjecture 17 would be true for in- EXERCISE 111. If Conjecture 17 is true for all prime exponents does it
finitely many primes. But the latter has never been proven. follow that it is true for all exponents, as on page 148?
(C) By a modification of the argument, the criterion:
53. EULERSCONJECTURE
2p + 1 is a prime, Although
may be supplemented by other criteria. It suffices if any of the following
are true: A3 + B3 # C3
4p +1, 8p +
1, lo p + 1, 14p + 1, or 16p +1 in positive integers we do have
is a prime. For example, since 63 = 33 + 43 + s3 and
29 = 4.7 + 1 is a prime, 2g3 = 113 + 153 + 273.
Conjecture 17 is true for p = 7. The above criteria, taken together, suffice There are in fact infinitely many solutions of
for all p < 100. Therefore, as S. Germain proved, Conjecture 17 is true for
all p < 100.
D3 = A 3 + B3 + C.
(Ul Theorem 65 has about the easiest proof of any significant result In our proof of
v b t a i r d on Fcrmats Last Theorem. That the restricted case is much easier
is aL-3 shown by the fact that in Kummers Theorem 64 the greatest A4 + B4 # C4
we utilized the fact that we had all solutions of 54. SUMOF TWOSQUARES
a' + b2 = c2 On pages 150-151 we gave a nonconstructive, partial proof of Theorem
60 based upon Gaussian integers. We now give two complete proofs, the
and that, of these, a and b could not be squares simultaneously. The strategy first explicitly constructive, and the second implicitly constructive. Both
suggests itself to find all solutions of Eq. (212) and, by specialization, are based upon (-1[p) = + l . The first proof-it may be Fermat's-
to show that Fermat's Last Theorem is true for n = 3. Further, one could uses the method of descent and also a famous identity which goes back
hope for a similar approach to n = 5, 7, etc. (at least) to Diophantus:
We know of no serious progress in this direction. I n this connection there
is a "conjecture" of Euler. While it has an attractive ring to it we know of Theorem 66.
no serious evidence and so shall call it
Open Question 2. Can an nth power ever equal the s u m of fewer than n
nth powers? That is, can
PROOF. Clear.
A" = Bln + Bzn + + Bkn PROOF of THEOREM 60. If p is a prime = 1 (mod 4 ) , there is an s <p
for 1 < k < n? such that pjs' + 1. Write s = ao , 1 = bo , and
Euler "conjectured" no. If his "conjecture" were true, Fermat's Last
Theorem would follow as a special case.
pqo = at + bt. (216)
112 (From Dickson) . Write Eq. (212) in symmetrized form

EXERCISE
It follows that qo < p . If pa'= 1, p
= a: +
b;. If not, divide ao and bo by
9 0 choosing remainders, positive or negatire, which have a minimum
w 3+ x3+ y 3 + z3= 0. (213) magnitude :
Substitute an = roqo + ao , bo = soqo + Po (217)
w = Lz (w + x + y + z ) , X=3(w+x-y-z) Both remainders, a. and P o , therefore satisfy I z 1 5 ~ Q O and
, not both
are zero. For if a0 = Po = 0, we have qolp, and since 1 < Yo < p this is
Y = +(w - 2 + y - z), 2 = +(w - x - y + 2)
impossible.
and show that Eq. (213) becomes the determinantal equation: Now define ql by
qoq1 = (Yo2 + Po2 (218)
and we have 0 < q1 5 + q ~ But,
. from Eqs. (216) and (215), we have
This is the condition that

pqtq1 = (ao2 + bO2) + Po")
((Yo2
( a o a + bopo)' + (aopo - boao)'.

wa + 3zb - 3yc = 0, -za + wb + 3xc = 0, ya - xb + wc = 0 =
have solutions a, b, and c not all zero. Substituting Eq. (217), and dividing by 902, now yields
Solve for x, y , and z in terms of w and obtain solutions of Eq. (214) :
w = -6pabc x = pa(a'+ 36' + 3c') Thus, if
y = pb(a' + 3b2 + 9c') z = 3pc(a2 + b2 + 3 2 ) .
Now with a = b = c = 1 and a proper choice of p obtain Eq. (211).
Conversely, from Eq. (210), obtain an a, b, c, and p which gives that
we have
solution. Finally, can all solutions of Eq. (212) in integers be obtained by
these formulae? pql = a12 + b12.
If q1 # I , we continue, and obtain A shorter, more modern proof of Theorem 60 is related to the idea in
Exercise 116. It uses Thues Theorem, and this, in turn, uses the
pa > q1 > . . . > q n = 1.
Dirichlet Box Principle. If more than N objects are placed in N boxes,
Finally at least one box contains two or more objects.
p = + bn2.
an2 Theorem 67 (Thue). If n > 1, (a, n) = 1, and m i s th,e least integer
To show the uniqueness asserted in Theorem GO, we assume a, b, c , and
> &, there exist an x and y such that
d are positive and ay = +x or - x (modn)
where
0 <x < m, 0 < y < m.
p 2 = (ac + b d y + ( a d - bCy (223)
PROOF. Consider ay - x for the m2possibilities: y = 0, 1 , 2 , . . . , m - 1
and x = 0, 1, 2, . . . , m - 1. Since m2 > n, by the Dirichlet Box Principle
and at least two of these possibilities must be congruent modulo n. Let
p2 = (ac - bd)2 + (ad + bc). (224) ayl - x1 3 ay2 - x2 (mod n )
By Eq. (222), ( p - a) d2= ( p - c2)b2or with yl > y2 . Further x1 # x 2 , for otherwise, since ( a , n ) = 1, we have
y1 = y 2 . Let y = y1 - y2 and x = f ( x l - 5 2 ) > 0 and we have
p ( d 2 - b2) = (ad - bc)(ad + bc).
a y = +x or -x (modn)
Sow if plad - bc, by Eq. (223) we have ad - bc = 0 , and thus d2 - b2 = 0
or d = b. Whereas, if plact +
bc, by Eq. (224) we have ac = bd, and since as required.
(a,b ) = 1, we have ald and blc. By Eq. (222) we now have d = a. Since SECOND PROOF OF THEOREM +
60. Let pls2 1. By Thues Theorem there
p is prime, we must have one of these two cases. exist positive integers x and y < 4 such that
Finally, to make the determination of p = a: +
b: completely con-
structive-but not necessarily efficient-we note that we may take ys =f x (mod p ) .
Since ( y , p ) = 1, we have
s2 +1= +1=0 (mod p )
by ~ ~ i ~ s oTheorem,
ns and Exercise 22, page 38. or
EmRcI6h 113. Determine 29 = aI2 + b: and 89 = + 122, given
aZ2 x2 + y2 = 0 (mod P I -
+
29 1 122 1 and 89 I 342 1. + But 0 < x2 + y2 < 2p. Therefore p = x2 + y2. The uniqueness we prove
EXERCISE
114. From the previous exercise find the two representations as before.
Of
EXERCISE 117. Apply the Dirichlet Box Principle to Gertrude Steins
29.89 = 2581 = A 2 + B2 surrealist opera, Four Saints in Three Acts, and draw a valid inferewe.
using Eq. (215). 55. A GENERALIZATION NUMBERTHEORY
AND GEOMETRIC
EXERCISE 115. Given p = a2 +
b2, determine 2p = A2 B2, and + Fermat, in a letter to Frenicle (1641), called Theorem 60 the fundtt-
+
5 p = C2 D2 = E2 F2. + mental theorem on right triangles. Compoundillg factors by Eq. (215), he
EXERCISE 116. Using the results of Exercise 113, find, conversely, an obtained numerous results such as:
x and y such that 2912 +
1, and 891y2 +
1 by x = a,bl- (mod 29), and A prime = 4m + 1 is the hypothenuse of a Pythagorean triangle in a
y I a&- (mod 89). single way, its square in two ways, its cube in three ways, etc.
162 Solved and Unsolved Problems in Number Theorg Pythagoreanism and its M a n y Consequences 163
EXAMPLE
: EXAMPLES
:
52 = 42 + 32 r ( 2 ) = 4 sinceA = 1; (1). B = 0.
= 24' + 7'
r ( 5 ) = 8 since A B 0.
25' = 20'+ 152 = 2; (1, 5). =
r(7) = O sinceA 1; (1). B 1; (7).

1252 = 1202 + 35' = 117' + 44' = loo2 + 75', etc.
= =
r(65) = 16 since A = 4; (1, 5, 13, 65). B = 0.

It is clear, from Eq. (215), that the product of two distinct primes of
the form 4 m + 1 is a hypothenuse in two ways, and, it may be shown, Theorem 68 contains Theorem 60 as a special case when allowance is
that a product of 12 such primes is a hypothenuse in 2k-' ways. made for the different conventions. We now apply this generalization to
derive the famous Leibnitz series:
EXERCISE 118. Obtain 4 distinct representations of n = A' B' for + 1 - 13 + 1 5- 1 +7 4 - ... (227)
(the Carmichael number) n = 5.13.17 = 1105.
We asked, on page 143: I n how many ways is n a sum of two squares? Equation (227) was one of the first results obtained by Leibnitz from his
The answer takes a particularly neat form if me alter the convention of newly discovered integral calculus. I n the subsequent priority controversy
what we mean by "how many ways." concerning the calculus, Newton's supporters pointed out that Gregory
Definition 41. By r ( n ) we mean the number of representations n = had already given
x2 + y' in integers x and y, which are positive, negative, or zero. The
+
arctan x = x - +x3 +,x5 - . . .
representations are considered distinct even if the x's and y's differ only and Eq. (227) follows by taking x = 1. Our present interest concerns
in sign or order. Further we define R ( N ) by quite a different point-a remark by Leibnitz concerning Eq. (227). He
N
suggested that with Eq. (227) he had reduced the mysterious number a
R(N) = Cr(n). to the integers. We may contest this claim. The derivation of Eq. (227)
n=O
(225) using integration and Taylor's series does not reveal the number-theoretic
EXAMPLES
: relation between a and the odd numbers. One may ask, "What has a circle
to do with odd numbers?" and receive no convincing answer from this
r(0) = I since o = 0' + 0'. derivation. The real insight is given by Theorem 68.
Consider the number of Cartesian lattice points ( a , b ) in or on the circle
r ( 4 ) = 4 since 4 = ( ~ 2 ) + ' 0' = 0' + ( ~ 2 ) ' . x2 + y2 = N . We show these points for N = 12. There are 37 of them.
r ( 8 ) = 4 since 8 = ( ~ 2 ) + ' (f2)'.
r(10) = 8 since 10 = ( ~ 1 ) + ' ( 3 3 ) ' = ( ~ 3 ) ' + (&1l2.

r ( p ) = 8 if p is a prime = 4m + 1.
R(12) = 1 + 4 + 4 + 0 + + 0 = 37.
. * .
It can be shown, by elementary methods, that the following result holds.

Theorem 68. If n, 2 1, has A positive divisors = 1(mod 4)and B positive
divisors = - 1(mod 4), then
r(n) = 4(A - B). (226)

W e mean here all divisors, not merely prime divisors.
x2 + y2 5 12
Pythagoreanism and its M a n y Consequences 165
It is clear, by Definition 41, that the number of such points equals R ( N ) ,
where
since each point corresponds to exactly one representation of one n = a' +
b2 5 N. Further, if we associate each point ( a , b ) with the unit square of K = [fl]
- 1.
which it is the center, ( a f 3, b f i), we see that R ( N ) approximates the
area of the circle, aN. The reader may show that the difference, R(N) - There are [v'x]terms in the first sum and we have
TN, vanishes with respect to aN as N -+ 00 , since this difference is associ-
ated with the (relatively small) region along the circumference. I n this way
he will obtain
where I 0 I < 1, since the error made by removing the square brackets in
Theorem 69. each term is < 1 . On the other hand the magnitude of the second sum is
less than, or equal to, the magnitude of its leading term-since the terms
R(N) - T N . (228) are alternating in sign and monotonic in magnitude. Therefore it equals
Corollary. The mean number of representations of n = a2 + b2, for n u p 0'fl where I e' I < 1. Therefore, dividing by N, Eq. (231) now becomes
t o n = N, tends to T as N m.--+
But, from Theorem 68, we may obtain a neat and exact formula for
R ( N ) . Each n 5 N receives a contribution of 4 representations from its
*T -c K
k=O
(-l)*-
1
+ feNl
-
divisor 1. Each n I N , which is divisible by 3, loses 4 representations from where I 0" I < 2, and, letting N -+ 0 0 , Eq. (227) follows.
this divisor 3, and there are [N/3] such values of n. Similarly, there are
EXERCISE
4[N/(2k + l ) ] contributions, or 4[N/(2k + l ) ] losses, corresponding to
119. Gauss gave R(100) = 317 and R(10,OOO) = 31417.
the odd divisor 2k + 1, according as k is even or odd. Counting the single
Verify the former, using Eq. (230).
representation of 0 = O2 + 02,we thus obtain EXERCISE 120. Jacobi's proof of Theorem 68 was not elementary but was
based upon an identity which he obtained from elliptic functions :
R(N) = + 4{[:]
1 - [:] + ];[ - ]:[ + a - . } . (229) (I + 2x + 2x4 + 2x9 + 2x16+ . . .)'
Further, since [N/(2k + l ) ] = 0 if 2k + 1 > N, we may write Eq. (229)
as a n infinite series, and thus obtain
Show that if the left side is written as a power series,
Theorem 70.
1 + alx+ a x 2 + a3x3+ . . . ,
then a,, = r ( n ), while if the right side is
EXAMPLE
:
1 + biz + b2x2 + b3x3 + . . * ,
then b,= 4(A - B ) ,where A and B are as in Theorem 68.
R(12) = 1 + 4{12 - 4 + 2 - 1 + 1 - 1) = 37.
56. A GENERALIZATION FORMS
A N D B ~ N A RQUADRATIC
Y
With Theorem 69, dropping the 1 as N -+ w , we obtain We now (start to) generalize Theorem 60 in a different direction. We
consider numbers of the form x2 +
Ny2 as suggested on page 143. At first
things go easily. Theorem 66 becomes
Now split the right side into two sums: Theorem 71.
(a2 + Nb2)(c2+ N d 2 ) = (ac + N b d ) 2 + N(ad - b c ) 2 (232)
= (UC - Nbd)' + N(ad + bc)2.
166 Solved and Unsolved Problems in Number Theory Pgthagoreanism and its Many Consequences 167
PROOF.
Consider Therefore
(a +m b ) ( a - -b)(c +m d ) ( c-m d ) . (5 + 2& - 2(x +yy =p
By pairing the 1st and 2nd terms, and the 3rd and 4th we obtain the left or
side of Eq. (232). By pairing the 1st and 4th terms, and the 2nd and 3rd we a - 2b2 = p. (233)
obtain the first right side; while pairing the 1st and 3rd terms, etc., gives
the second right side. Now let aZnand c2,, be the side and diagonal numbers of Theorem 58,
For N = 2 and 3, Theorem 60 generalizes easily to page 139. Then by Eqs. ( 2 3 2 ) and (172),
Theorem 72. Every prime p of the forms 8 m +
1 and 8m +
3 can be (2,- 2 a 9 (a - 2b2) = p
+
written as p = x2 2y2 in a unique way. Every prime p of the form 6 m 1 + and
can be written as p = x2 +
3y2 in a unique way.
p = (cnna- 2 ~ ~ , , -
b )2(c2,b
~ - a2,~)~.
PROOF. If -N is a quadratic residue of p there is an s, prime to p , such
+
that s2 N = 0 (mod p ) . By Thues Theorem, as on page 161, there are Likewise
positive integers x and y < fisuch that p = (a,a + 2 ~ 2 , b )-~ 2(c2,b + az,a)2.
s2+ N = x2y/-2+ N = x2 + N y 2 = 0 (mod p ) . Therefore from each of the infinitely many pairs (c2, , a2,), and from Eq.
Now if p = 8m + 1 or 8nz + 3, ( - 2 ( p ) = +l, and x2 + 2y2 is a multi-
( 2 3 3 ) we obtain two other solutions of Eq. ( 2 3 3 ) .
ple of p which is <3p. If x2 + 2y2 = p , we have our solution; but if x2 + EXAMPLE :
2y2 = 2 p , since x must be even, = 2w, we have y2 + 2w2 = p a s o u r From 32 - 2.1 = 7, and (c2, , a2,) = ( 3 , 2 ) and (17, 1 2 ) , we find:
solution. 7 = s2 - 2 . 3 2 = 1 3 -
~ 2 . g 2 = 272 - 2 . 1 9 ~= 752 - 2 . ~ 3 ~ .
Again, if p = 61n + 1, ( -31p) = +1, and x2 + 3y2 is a multiple of
p < 4p. Now x2 + 3y2 # 2 p , for if equality holds, x and y are either both EXERCISE 121. From jZ- 2 . Y = 17, find four other representations of
odd or both even and therefore x 2 + 3y2 is divisible by 4, that is, 21p. a2 - 2b2 = 17.
Therefore either p =x 2 + 3y2, or x = 3w and y 2 + 3w2 = p , as before. To generalize Theorem 74 to a2 - 3b2, a2 - 5b2, etc., we would need the
The uniqueness follows from the more general generalization of Theons Theorem 58 known as Fermats Equation, i.e.,
Theorem 73. If N > 0 there i s at most one representation of a prime p as Eq. (174). This we will investigate in Sect. 58 below. We may also note
p = a2 + Nb2 in positive integers a and b. that the infinite number of solutions in Theorem 74, in distinction to the
single solution in Theorem 72, is associated with the fact that the alge-
PROOF. This is left for the reader, who will utilize Theorem 71.
Kow, the natural generalization of Theorem 72 would be this-if
braic number field k( 4 has inJinitely many unities-see pages 152, 150. -
That is,
+
( -NIP) = + l , then p = x 2 Ny2 in a unique way-but this supposition
Cn + fianI1
is not true. The generalization breaks down a t two points.
First, as hinted by the qualification, N > 0, in Theorem 73, uniqueness for any side and diagonal numbers a, and cn .
need not hold if N < 0. Thus we have (see page 143) the Fermat-Frenicle A second, and more difficult, point which precludes the simple generaliza-
Theorem 74. Every prime p of the form 8 m f 1 can be written as a2 - 2b2 tion of Theorem 72 mentioned on page 166 is this. I n the proof of Theo-
in infinitely many ways. rem 72-say with N > O-one finds an x and y such that x2 N y 2 = r p , +
PROOF. Since (2173) = + I , we have, by Thues Theorem, x2 - 2y2 is a where the coefficient r satisfies 1 5 r < N +
1. It is not clear that, with
multiple of p , < p and > -2p. Since x 2 - 2y2 # 0 by Theorem 56, we have these many possibilities for r , one can always obtain an r = 1.
Indeed, for N = 5 and 6 this is impossible. Thus (-61p) = +1 for
x2 - 2!j2 = -p. p = 24m+ l , 5 , 7 , o r l l (see tableon page 47).Inparticular(-615) = 1.
168 Soloed and Unsolved Problems in Number Theory 169
Pythagoreanism and its Many Consequences
But i t is clear that 5 f a' +
6b2.Similarly, (-513) = +1, but 3 # a2 + (A) (For those who know vrctor algebra.) Diophantus's formula, Eq.
56'. The partial proof of Theorem 60 using the unique factorization of
(215), has an interesting interpretation in vector algebra. Let
Gaussian integers (page 150) suggests that the "difficulty" stems from the
lack of unique factorization in k ( G ) and k ( a ) (see page 153). Vl = ai + bj, Vz = ci + dj.
This is indeed the case. The following may be shown.
Then the scalar and vector products are Vl.Vz = ac bd, Vl x V2 = +
+
Theorem 75. If (-61p) = 1, p = u2 6b2in a unique way if p = 24m + (ad - bc)k. But the magnitude of Vl X Vz is the length of Vl times the
1, 7. But 2p = a2 +
6b2 if p = 24m +
5, 11. Similarly, if (-51p) = 1, length of V, times the sine of the angle between them. And V1. Vz is the
p = a' + 5b2 or2p = a' +
5b2 accordingasp = 20m +
1 , 9 o r p = 20m+ length of Vl times the length of V2 times the cosine. Therefore
3, 7.
The two classes of primes, in either case of this theorem, are related to
I v1 1 I vz I = (Vl. VdZ +I v
1 x vz I 2,
the so-called class number (see page 153), which is > 1 when unique fac- and we obtain the first part of Eq. (215). On the other hand, if V, = ci - di,
torization is absent. We cannot do justice to this most interesting concept while I V3 1 = 1 Vz I , the sine and cosine of the angle between Vl and V3
in a few pages. Instead we pass on to other subjects. will now be different, generally, and we obtain the second representation
in Eq. (215).
EXERCISE 122. Prove that for N = 7 everything is "OK" again-that is,
if ( -71p) = +1, there is a unique representation p = a' +
7b'. The fact
(B) (For those who know partial differential equations.) If the lowest
frequency with which an elastic square membrane can vibrate is
that the relatively large value N = 7 is still "OK" is related to the specially
large density of primes of the form n' +
7. See table on page 49 and com- wo = fik
+
pare remarks about n2 5, 6 on page 154.
where k is a constant, then it is well known that every possible frequency
EXERCISE 123. For N = 10, find a p such that (-lOlp) = +1, but is given by
p # a' + lob'. w = 4 S T F k (234)
EXERCISE 124. I n general, if p < N, p # a' 4-Nb'. What does this sug-
where a and b are positive integers. Corresponding to this frequency, Eq.
gest concerning unique factorization in k ( d7V)in general? Investigate
(234) , the shape of the membrane is given by
the literature to confirm or reject any hypothesis you develop. Caution:
If N = -1 (mod 4) the integers of k ( d - ) are of the form C sin (n-ay/L) sin (n-by/L)
+
3( a d - b ) . By unique factorization one could therefore only conclude
where L is the length of the side. For the frequency Eq. (234) , there will
that 4p = a' +
Nb'. An example is p = 3, N = 11. The integers
therefore be s different niodes of motion if n = a' +
b' can be written as a
in k ( m ) do have unique factorization.
sum of squares in s different ways-where a and b are positive, but m-here
EXERCISE 125. Analogous to Theorem 68, for N = 2 there is the the order is counted. Thus for w o , s = 1; for w = d k , s = 2; for w =
following: a k , s = 4, etc.
+
The number of representations of n = x2 2y2is equal to 2 ( A - B ) if (C) (For those who attempted Exercise 16, page 29.) By Theorem 72
n has A divisors = 1, 3 (mod 8) , and B divisors _= - 1, - 3 (mod 8). the prime q = Gp +
1 may be written
By a n argument similar to that above (page 164) but now using ellipses
+
z2 2y' = n, show that
q = a2 + 3b'
in a unique way. The criterion sought is this: qiM, if, and only if, 3jb.
-1
n- 1 + -1 - - 1- - +1 - +1- - -1 - - .1. . 1
2 d 3 5 7 9 11 13 15 EXriMPLES :
EXERCISE 12G. Conjecture the results analogous to the previous exercise p = 5, q = 31 = 2 ' + 3.3'; Since 313, 311M5
for N = 3. Investigate the literature to check your conjecture.
p = 13, q = 79 = 22 + 3.5'; Since 31;s) 791;MI3
57. SOMEAPPLICATIOSY
p = 17, q = 103 = 10' + 3.1'; Since 3+l, 1031;M17
We now give several applirations of the foregoing results.
We shall not prove this rule, but we will indicate its source.
Let g be a primitive root of q, and let 2 = g e (mod q ) , and therefore For N = 1, likewise-since otherwise a' +
b' would be even. Therefore a t
2' = g e p (mod a ) . Since q = 6 p +
1, we have that qjM, if, and only if, least one of the representations in Eq. (235) has x > 0 and y > 0. By
6 / e . Since p = 4 m +
1 (see page 29), we have (2lq) = + I , and e is induction every divisor of n > 1 equals x' +
Ny' in positive integers.
even. Therefore qlM, if, and only if 31e. Therefore the necessary and Therefore if n is composite, write it as a product, Eq. (235), with a, b,
sufficient condition sought is that 2 is a cubic residue of q : c, d > 0. Then there are a t least two distinct representations of n in non-
negative integers, since ac +
Nbd > ac - Nbd. For N = 2, 3 this suffices.
x3 3 2 (mod 9). For N = 1, we must also show that ac Nbd = ac + bd # ad + bc. +
Prior to the time that the theory of cubic residues was developed, Gauss This is so because
found that it was necessary in developing the theory of biquadratic res- a(c - d) = b(c - d )
idues, x4 = a (mod p ) , to introduce the Gaussian integers-namely, those
~ ) k ( i ) . Similarly, under this stimu-
of the algebraic number field 1 ~ ( e ~ * " = implies c = d , or a = b, and thus that n is even. This completes the proof.
lus, Eisenstein developed the theory of cubic residues withthe field k(e2*"3). With Theorem 76 we have a method for determining the primality of
Since n = 4m +
1 by N = 1 , and of n = 8m +
3 by N = 2. The method is
useful if n is not too large. One uses subtraction and a table of squares,
e2r2'3= ;(-I + -1, instead of division and a table of primes. To test the remaining numbers,
we are not surprised to find criteria involving namely n = 8 m +7, one would want to use N = -2. But as we have
seen in Theorem 7 4 we now lack uniqueness. To clarify the number of
a' + 3b2 = (a + b)(a - b). representations of n = a' - 2b' we now investigate Fermat's Equation.
The criterion that 2 is a cubic residue of q = 6m +
1 is: 31b, where q = EXERCISE 127. Show that Theorem 76 may be easily extended to the
a' + 3b'. case N = - 1 and n = 2m 1. +
(D) (Necessary and Sufficient Conditions for Primality.) EXERCISE 128. 45 = a ' + b'
in a unique way, but 45 is not a prime.
Theorem 76. For n > 1, and 25 = a' +
b' in a unique way in positive integers, but 25 is not prime.
+ 1;for +
21 # a2 b', and therefore 21 is composite. Again, 21 is composite since
N = 1: assume n = 4m
it equals a' +
5b2 in two ways. But neither 3 nor 7 equals a2 5b'. From +
N = 2: assume n = 8m -l- 1 or 8m + 3; for Theorem 75,
N = 3: assume n = 6m + 1. 3 = $(I2 + 5.1'), 7 = 3(3' + 5.1').
I f n i s prime, n = a' +
Nb2 in a unique way in positive integers a and b, T h u s 3 .7 = ( 4 +
G)( 4 - G)= (1 2 G)(1 - 2 G). +
and ( a , b ) = 1. Conversely, i f n = a' +
Nb' in a unique way in nonnegative Compare page 153. Construct a similar example: p q = a' 6b' in two +
integers, a and b, and i f ( a , b ) = 1 , then n i s prinze. ways, while neither p nor q equals a' 6b'. +
PROOF. For n prime we have shown a unique representation. Further EXERCISE 129. One half of the numbers 8 m +
7 may be tested by n =
( a , b) = I since ( a , 6) In. a' +3b'.
Now, conversely, let n = a' +
Nb2 and ( a , 6) = 1. Then ( b , n ) = 1 EXERCISE 130. All M, for p an odd prime fall in the class indicated in
and (ab-')' = -N (mod n ) . Thus every prime divisor of n is of the form
listed above corresponding to N . By Theorem 71,
the previous exercise. I n particular Ml1is not a prime, since MI1 # a2 3b2. +
But for p large, say p = 61, the test is impractical.
(a' + N b 2 ) ( c 2+ N b 2 ) = (ac + Nbd)' + N ( a d - bc)' OF FERMAT'S
58. THESIGNIFICANCE EQUATION
(235)
= ( a c - N b d ) ' + N ( a d + bc)'. The equation :
Therefore a product of two primes satisfying ( -NIP) = +1 is also of 2' - Ny' = 1
the form 'x +Ny' with x and y positive. For if (ac - N b d ) and (ad - bc)
for N > 1, and not a square, is called Fermat's Equation. I n older writings
were both zero, we find a' = Nb2. For N = 2, 3 this is clearly impossible.
it is often called "Pell's Equation." If N = n', it is clear that Eq. (236) implies infinitely many. All this bec*ause 1 . M = A 1 on the ltft side of
has no solution in positive integers since no two positive squares differ Eq. (241).
by one. Fermat stated that Eq. (236) has infinitely many solutions for (C) This special role of Af = 1 is also indirated-it is really the same
every other positive N . He suggested the cases N = 61 and 109 as challenge point in different languagc-by the fact that for any solution x and y
problems. Later Frenicle challenged the English mathematicians with of Eq. (236),
N = 151 and 313.
For some N a solution is easily obtained. For N = 2 we have 3' - N2' = X f O Y
1 from Theorem 58, and, more generally, if N = n2 1, + is a unity of the algebraic field k ( fl). See pages 152, 167.
(2n' + 1)' - N ( 2 n ) 2= 1. (237) (D) Again, the solutions of Eq. (236) are intimately related to the
rational approximations of fl, as we already noted on pages 139, 140.
But for N = 61, x = 1766319049 and y = 226153980 is the smallest solu- Thus, from a larger solution for N = 3:
tion, and for N = 313 the smallest 2 has 17 digits. Such an x is not some-
thing one would like to obtain by trial and error. 70226' - 3.40335' = 1,
EXERCISE 131. Verify the following generalization of Eq. (237). If we get
N = ( n m ) 2f m, then
70226/40545 = 1.7320508077 . . . , (242)
(2n'm 1)' - ~ ( 2 n ) =
' 1. (238)
which agrees with 4 to ten figures.
And if N = (nm)' f 2na, then (E) Further, these approximations, and the solutions of Eq. (236))
(n'm d= 1)' - N ( n ) ' = 1. are obtained by infinite continued fractions, and Fermat's Equation was
(239)
the occasion for the introduction of this technique into number theory.
Show that by a proper choice of in and n, Eqs. (238) and (239) suffice to ( F ) The same continued fractions may be used expeditiously to obtain
yield solutions for all nonsquare N where 2 5 N 5 20 except for two
cases. Likewise for 30 5 N 5 42. p = a' + b2
I n the next section we state and prove the main theorem by a lengthy +
for primes of the form 47% 1.
implicit construction. Later we give an efficient algorithm. We now list (G) If we factor the left side of Eq. (242) :
some reasons why Eq. (236) is important.
(A) If Eq. (236) is generalized to -70226 . - 37
_ - .-_ 26 73
40545 15 51 53
a2 - Nb2 = M (240)
we obtain convenient gear ratios to approximate 4:
for any integer M , there can be no solution unless M is a quadratic residue
of every prime which divides N ; the example N = 3, M = -1 was men-
tioned on page 140. (We note that while this condition is necessary, it
is not sufficient. Thus
a2 - 34b' = -1
has no solution even though - 1 is a quadratic residue of 2 and 17.) M = 1
is, of course, a quadratic residue of all primes.
(B) But if Eq. (240) has a solution, it has infinitely many. Using the
method in the proof of Theorem 74, with the identity from Theorem 71,
(5' - Ny') ( a z - Nbz) = (xu f Nyb)' - N ( x b f ya)', (241)
and with any solution of Eq. (236), one obtains another solution of Eq. ( H ) But to carry out such factorizations it is desirable to know the
(240). Further, since we may take LIT = 1, one solution of Eq. (236) divisibility properties of the solutions (x,y) of Eq. (236). These properties
are given by interesting and useful divisibility theorems for the infinite > 1, and thus, by Eq. ( 2 4 4 ) , xn + fiyn < + dRYn+l ,Zn+l
sequence of solutions of Eq. (236). For N = 3 these theorems were used assume
by Lucas to obtain his criterion for the primality of Mersenne numbers.
It was this consideration (page 120) which led us into this chapter.
xn + flyn <x + fly < xn+l + f l y n + 1 * (247)
Since (zn - fiy,) (2, fly,) +
= 1, we note that the first factor here
59. THEMAINTHEOREM is > O . Multiply Eq. (247) by this positive number, x, - fly,, , and
Theorem 77. I f N > 1, and not squure, we have
X' - Ny2 = 1 (243)

1 < (z + fly)(xn - fly,) <( +
~ 1 f l ~ 1 ) " + ~-
( ~f 1l y l ) "
has injinitely many solutions in positive integers. I f x1 fl y1 i s the + = x1 + dNy1.

smallest value that x N<+ y takes on, with x, y a solution, then every solu- Let
tion i s given by (X + fly) (5, - Z/Nyn) = a +fib
xn + flyn = (XI + fi ~ 1 n.) (244) where a = xxn - Nyyn and b = yx, - xyn . But
The x, and yn may be computed explicitly by a' - Nb2 = (x2 - Ny2)(2: - Nyn') = 1,
xn = +[(XI + 0 + ( ~ -1 flyl)"]
1 ~ 1 ) ~
and since 1 < a +
f i b we find 0 < a - f l b < 1. Thus 1 < 2a and
0 < 2 flb. Therefore we have a solution of Ey. (243) in positive integers
a and b with a +
f l b < x1 +
f l y l . Since by the definition of x1 +
<Nyl this cannot be, there is no other solution x fly. +
Finally we come to the real problem, to show that Eq. (243) has a t
or recursively by
least one solution. The first published proof is by Lagrange. Our proof
xn+l = x1xn + N ylyn rests on a lemma which we will prove later in two ways.
ylxn + xlyn.
(246) Lemma. There i s a n integer M such that x2 - Ny2 = M has infinitely
yn+1 =
many solutions in positive integers.
COMMENT: If x, y are positive integers which satisfy Eq. (243) we will
We assume this now and consider M2 boxes Bo.b with 0 5 a < I M 1,
sometimes use the expression: x +
d??y "is" a solution of Eq. (243).
0 5 b < I M I. Choose 11.1' +
1 solutions of z2 - N y 2 = M and place each
PROOF. First we prove that x12 - Ny? = 1 implies 22- Ny; = 1.
pair (x, y) in the box if x = a, y = b (mod 1 M I). By the Box Principle
From Eq. (241), with x = a = x1 and y = b = y1 , and choosing the
we therefore have two different solutions:
plus sign on the right, we see that the x2 and yz of Eq. (246), with n = 1,
satisfy Eq. (243) if x1 and yl do. By induction, the x, and yn of Eq. (246) x12 - Ny? = x: - Ny: = M (248)
also satisfy Eq. (243). Also, by induction, these integers satisfy Eq. (244),
and likewise
with 21 = xz , yl = y2 (mod 1 M I). Thus
21x2 - Nyiy2 X: - Nyl2 0 xlyz - xzyi (mod I M I),
xn - f l y n = - fly11 n-
and we have x1x2 - Nyly2 = U M and x1y2 - x2y1 = v M with u and v in-
Then Eq. (245) follows a t once. tegers.
Next we prove that there are no other solutions. Assume another s o h - But, by Eqs. (241) and (248),
tion, Eq. (243). Then
M Z = (21x2 - Nylyz)' - N(zly2 - ~2?/1)'= (u2- Nv2)M2.
x + dGy # xn +fiyn 9
Then
(249)
for, if equality held, x - xn = dp(y, - y) , and, since flis irrational

by Theorem 57, we must have y = yn , x = x, . Therefore, since x1 + u
' - Nu2 = 1.
Now, if v = 0, xly2 xzyl and, by Eq. (249),
=
Thus, since vm is irrational, there are infinitely many solutions of
M = zt (21x2 - Nyiyz).
Thus
0 <x - fly < - .1 (2.50)
IYI
Mxiyz = f(xl%zyz - N Y ; X ~ Y=
~ )f M ~ z y 2 . It follows that
Since x1 > 0 and x2 > 0 we have x1 = x2 , and likewise yl = y 2 . Thus 2.\/Ny <x +f i y <2fiY + 1,
v2 > 0 and Eq. (243) has a t least one solution u2 - Nu2 = 1 in positive
and thus, whether y and x are positive or negative,
integers. By the Well-Ordering Principle there is therefore a smallest
solution: x1 m + 1 . Ix +f l y I <2 f l I Y I + 1.
The reader may note that the device used in Eq. (249) of multiplying
Therefore we find infinitely many solutions of
two equations, and then dividing them by M 2 , is analogous to the strategy
utilized in Theorem 60, both after Eq. (218) and after Eq. (222). 0 < 1 x2 - N y 2 I = ( X - fly) Ix + &y I < 2\/T + 1.
A proof of the Lemma using a continued fraction algorithm will be
By the Box Principle (extended) we therefore have infinitely many solu-
given later. A shorter, and now standard, proof runs as follows:
tions of x2 - N y 2 = M for some 0 < 1 M I < 2 f l +
1. This completes
PROOF OF THE LEMMA. For y = 0, 1, 2, . * . , let x = [ a y ] + 1. Since the proof of Theorem 77.
flis irrational we have The reader notes the curious character of the proof given here for Theo-
rem 77. A solution of Eq. (243) implies
0 <2 = x -f l y < 1.
- - f l =1
X
For any positive integer n1 consider the n1 boxes:
Y Y(X +f l y )
that is, x / y is a ((good rational approximation of fl.I n the proof of
the Lemma, we first find that there exist approdmations:
and nl + 1 values of z given by y = 0, I , . . . , n1 . At least two xs are
- f l < p1
in one box, and they are unequal since flis irrational. Call them z1 > z2 .
Then their difference satisfies: Y
1 then, by the Box Principle, better approximations :

0 < 21 - 2 2 = (a - X d - *(y1 - yz) <-
n1
This may be written
Finally, using the Lemma, we attain the required approximations. It could
be called a proof by convergence, and this suggests that an explicit
where x3 = x1 - x 2 , y3 = y1 - y 2 . Now choose n2 by and more efficient algorithm for finding good rational approximations of
fl could lead to an explicit and more efficient construction of solutions
1 of Eq. (243). This we now examine.
- < 23
n
2 EXERCISE 132. By the same technique as that used on page 17-1to show
and by the same process we obtain a < z3 with that there is no other solution x fly, +
show that if u2- N y 2 = -1
24
+
has a solution, and if u1 + T y l is the smallest value possible, then all
solutions of u2 - N y 2 = -1 are given by
un +
f l y * = (u1 myl) + (251)
for n odd, while for n even one obtains the solutions of Fermats Equation: It is clear, by the rules (Eqs. 254-256), that since Cn-l , B, and C, repeat
here for n = 1 and 7, that A , , B, , and C, will henceforth be periodic
~ r +n f l ~ z r n = xm +flym (252) with a period of 6.
given by Eq. (244). Our immediate interest in the algorithm-there will be other points
later-is in the important relation which we will prove in the following
133. Show that Theons rule, Eq. (171) , gives all solutions of
EXERCISE section:
x2 - 2y2 = & l . P; - NQ; = (-l),C,. (259)
60. AN ALGORITHM If C , = 1, with n even, we obtain a solution of Fermats Equation: P:-
For any positive nonsquare N we define five sequences of nonnegative NQ; = 1. It will be shown that for every N there are infinitely many n
integers A , , B, , C, , Pn , and Q, as follows. Let with ( - 1)C, = I, and for the smallest such n > 0 we obtain the smallest
solution of Eq. (243) :
C-1 = N ; Co = 1; Bo = 0; P-1 = Qo = 0; Po = Q-1 = 1. (253)
For n = 0, 1, 2, . . . , define the sequences recursively by
Pn + f i Q n = XI + f l ~ 1 .
Thus 170 +
d 1 9 . 3 9 is the smallest solution for N = 19.
An+1 = [flcT , or, since A, = [4V], If C , = 1 with n odd, the smallest such n yields the smallest solution of
u2 - Nu2 = - 1 by
Pn + f l Q n = UI + flu1
using the notation of Exercise 132. We show such a case for N = 13:
Bn+l = An+lCn - Bn.
Cn+l = Cn-1 + An+l(Bn - Bn+1)- n C, An B, 1 Qn
Pn+1 = P,-1 + A,+iP,.
i :
-1 13 - - 0 1
Qn+l = Qn-1 + An+lQn- 0
1
1
4
-
3
0
3
1
3
2 3 1 1 4 1
I n Eq. (254) we use the [ ] function of page 14. 7 2
3 3 1 2
EXAMPLE: 4 4 1 1 11 3
For N = 19 we show the sequences: 5 1 1 3 18 5
6 4 6 3 119 33
~ _ _ _ _ _ _ _
n Cn B,
-
Qn
Then 18 +
6 . 5 is the smallest solution of u2 - 13v2 = -1. Corre-
-1
0
19 -
-
- 0 1 spondingly, a s in Exercise 132, 649 +
6 . 1 8 0 = (18 6 ~ 5 is ) + ~
1 0 1 0 the smallest solution of x2 - 13y2 = 1. Alternatively, one could continue
1 3 4 4 4
2
1 the table until Clo = C6 = 1, since now there is a period of 5. Then PI0 =
5 2 2 9 2
3 2 1
649, Q1o = 180. I n fact, by periodicity we have
3 13 3
4 5 3 3 48
5 3
11 PEk - 13&& = ( - l ) k .
1 2 61 14
6 1 2 4 170 39
7 3
134. Obtain solutions of x2 - 31y2 = 1, and of x2 - 41y2 =
EXERCISE
8 4 1421 326
fl.
EXERCISE135. For N = (nm)' +

m (compare Exercise 131) carry which we abbreviate as
out the algorithm algebraicly and obtain
Pz = 2n2m + 1, QZ = 2n.
Similarly carry out the algorithm for N = (nm)' - m. I n this case what The fractions
is the period of A , , B, , C, if m = 1 ; if m > I ?
We shall see that A , , B, , and C, are periodic, from some point on, for
-1 = I ,
1
3
2
1
5
- 1 + y1 - - 1
-7 -
2+2
+ 1 z,
+1 etc.
all N . This will follow from the inequalities:
are called the convergents of the continued fraction. The reader may note
0 < A, < 2 f l , 0 < B, < fl, O < C, <2fl (260) that these convergents are c,/a, , the ratios of Theon's diagonal and side
which hold for all positive n. Granting these for now, it is clear, by the numbers.
Box Principle, that C,,-1, B, , and C, must eventually repeat. Then A , ,
If 4 = 1 + +
( l / x ) , we have x = l / ( & - 1) = ( 4 3 1)/2. Let
B, , and C, will be periodic henceforth. We designate the period p ( N ) ,
+ +
( 4 1)/2 = 1 ( l / y ) and y = a/(& - 1 ) = 4 1 = 2 + +
( l / z ) . Thus
as in p ( 19) = 6, p( 13) = 5. Assuming Eq. (259), we then have another
proof of the Lemma (page 175).
(263)
We will also obtain the useful invariant:
B: + C,C,-I = N. (261) The reader may verify that the convergents now:
Since N # m2,we see that C, f 0. This justifies the division in Eq. (254).
Again, since C,-l = ( N - Bn2)/C, we need not have stipulated the repeti-
tion of CnV1in the previous paragraph. Also, if C, = CnP1we have N =
B: + C
.: I n particular, for n = 3, we have 13 = 2' +
3'. It can be shown
are alternately solutions of Eqs. (176) and (175), and, conversely, Archi-
that for every prime N = 4m +
1 there is such a C, = C,-l .
medes' approximations (page 140) are later convergents.
It may be easily shown that the convergents of any continued fraction
FRACTIONS
61. CONTINUED FOR fl 1 1 1
Consider A'+x+A,+A,+
...
(4- 1 > ( 4 + 1) = 1. form a convergent sequence if the A's are positive integers. Also that if
Then x is irrational and > 1 it has a unique representation of this type:
1 x = A 1 + ; I1 ; + x1+ ...
\/z=1+
1 + d
and, by substitution we have Further, if such an x is given by
d5 = 1 + -1 1
2+1+1/2'
By continuation we obtain the infinite continued fraction: where y > 1, then these n values of A are those of its unique representa-
tion. It follows that Eqs. (262) and (263) are the representations for
2/2 and 4.
One could proceed with flas with 4.But there is much redundancy,
notationwise and otherwise, in such algebra. If one seeks an algorithm
with the redundancy removed one obtains that given in Sect. 60. Thus Since a1 = flwe obtain
we shall see that
1 1 1
fi=Al+-
AZ+A3+Z+ * * .
where the As are given by Eq. (254). Further the convergents are by induction.
Next we show 0 < A,, and 1 < a, . From a, > 1, and [a,] = A , , and
Pn
- = A + -1 1 1 Eq. (268), we find A , > 0 and anfl > 1. Since a1 = fl > 1, the re-
Qn A z + A 3 + ... +A, quired results follow by induction. Then from Eqs. (269) and (264) we
where P, and Qn are given by Eqs. (257-258). Thus, from page 178, derive Eq. (265).
To complete the proof of Eq. (260) assume
1 1 1 1 1 1 1
43=4 +3 +7 +++7 +5 +g + 3 + .... 0 < B, < dz, 0 < C, < 2 f l (270)
We may indicate the periodicity neatly by the symmetric formula: for some positive n. From
Similarly we find B,+, < 9.

Then from B i f l + C,+ICn = N , we obtain 0 < Cn+i.
1 1 1 1 1 Thus from
a = 3 + i + i + i + i + 3 + a
and
18--
- 1 1 1 1 we have Cn+l < 2 f l . But if B,+1 S 0, from Eq. (255) Cn S An+1Cn 5
5 3+ i+i+i+i Bn < d R . This implies
We now prove Eqs. (259) , (260) , (261) , (265) and (266). The subject
of continued fractions is a large one. It is not our purpose now to expound
upon it a t length. Our primary interest concerns its relation to Theorem
77. At that, our treatment is brief and we leave numerous computations and this contradiction implies 0 < BR+I . Then Eq. (270) follows by in-
for the reader. duction for all n > 0. Finally, since 1 5 C,, we get A,, < an < 2 f l
First, from Eqs. (255) and (256) me have from Eq. (267). This completes the proof of Eq. (260).
Next, from Eqs. (257), (258) and (253) we obtain a second important
Cn+1 = Cn-1 + 2An+1Bn - A:+lCn 3 invariant by induction :
and from this, and Eq. (255), we obtain Eq. (261) by induction. Then (-l)n(PnQn-~ - Pn-lQn) = 1. (271)
we define
This implies that ( P , , Q,) = 1 and the fraction P,/Q, is in its lowest
a, = fl+
Bn-1 terms.
Cn-1
1
Now we prove Eq. (266) , slightly generalized. Let a, , n = 1, 2, . * .
and using Eq. (261) we find that be any positive numbers, not necessarily integers, and let
1
p-1 = qo = 0, po = p-1 = 1,
an=A,+-.
%+I pn+l = pn-1 + an+ipn, qni-1 = qn-1 + an+ipn j
184 Solved and Unsolved Problems in Number Theory Pythagoreanism a n d its M a n y Consequences 85
analogous to Eqs. (253), (257) and (258). Then an alternating manner:
-p1_--a 1 and = a2 a1 + = al + -
1
q1 1 Qz a2 a2
are identities. Assume, for some n > 1, We have shown that the algorithm yields a convergent, periodic, con-
tinued fraction for fl, and, if ( - 1)"Cn = 1, we have a solution of
Eq. (243). These fractions were used by Fermat, Frenicle, Wallis and
Brouncker to obtain solutions. N o one prior to Lagrange, however, (except
for any positive a's. Thus we may replace a, by a, + and obtain possibly Fermat), proved that such an n always existed. We have seen
that the algorithm implies the Lemma, and this implies a solution: x2 -
N y 2 = 1 . Therefore
and
2 1 1
Therefore Eq. (272) is true for all positive n and a's. In particular, Eq. -Y - f i = y z
(266) is true. Further, from Eq. (269), we obt'ain fl+
Now any rational number b / a > 1 may be expanded into a finite con-
tinued fraction as on page 12. We have
-b = q 0 + - 1 -
1 1
-
a q1 + 42 + + qn
where the q's are given by Euclid's Algorithm on page 9. Further qn > 1,
and, a t our option, we may also write
b 1 1 1
-=qo+-
a q1 + +
. . . (qn - I) i +
(274) Using one or the other, s l y can be written
2
-
Y
= a1 + az1 + * * *
1
-
an
with n even. If z is defined by

and from Eq. (271) we prove Eq. (259).
It is easy to show that the right side of Eq. (265) converges to the left
side, for from Eqs. (273) and (271) me also obtain
we have, analognlls to Eq. (275),
Pn 1
fl =
- -
Qn
(-l)n
&n(Qn-1 + an+1 Q n ) '
(275)
Since Q,,increases without bound the convergents converge to dN in

186 Solved and Unsolved Problems in Number Theory Pythagoreanism and its Many Conseyuences 187
1 1 1 + P-1 (284)
where y is the denominator of the next to the last convergent of Eq.
(277). Therefore 0 < y/y < 1, and comparing Eqs. (278) and (276) ...
A1+;i;+ +%+A,+ + x1 - -Pm Qm
Qmz + Qk-i
Q-1.
we find z > 1. Then, by Eq. (264), ai = Ai and x/y is a convergent Pn/Qn . EXERCISE 141. Use one of the results of the previous exercise as a short-
It follows that every solution of Eq. (243) is given by the algorithm. cut in solving x2 - 61y2 = - 1. What do you note about the Ps and Qs
EXERCISE 136. Solve 61 = a +
b2 and xz - 61y2 = - 1 by the algorithm. +
used, in relation to the 61 = a2 b2 of Exercise 136?
Solve x2 - 61y2 = +1; compare page 172. Obtain the representation of EXERCISE 142. From Eq. (273) and the periodicity of the As rederive
a. the recurrence relations, Eq. (246).
EXERCISE 137. Let n be the smallest positive index for which C, = 1. EXERCISE143. There are infinitely many solutions of x2 - 34y2 = M for
From Eqs. (261), (260), etc. show B, = = A , and An41 = 2A1. M = +2 and -9, but none for M = -1.
The representation may be written
EXERCISE144. If N 3 1 (mod 4) and prime, and the smallest solution
of Eq. (243) is x1 + flyl , then
and the period p ( N ) = n. The sequence of As is u1 = ,

(q and vl = /% (285)
A , , Az, A, ... A,, 2A1, Az , A , , . . . A , 2A1, etc. +

are integers and u1 d N v l is the smallest solution of u2 - Nu2 = - 1.
EXERCISE
138. The representation, Eq. (279), is always symmetric: EXERCISE 145. If N = 2k + 1 is prime, the period p ( N ) of flis even
or odd according as k is odd or even.
EXERCISE
146. If N 3 1 (mod 4) and prime, and if p ( N ) = 2k - 1, then
To prove Eq. (280) show that one may replace N < by - 0in Eq. N = B: + C:. (286)
(279). Then solve for the lower radical in terms of the upper. Alternatively,
use Eqs. (279) and (273) to derive EXERCISE
147. If N is an odd prime, and Nls2 + 1 with 0 < s < + N ,
write
Pn = A,&, + Qn-1 .
- _- a l + - a21
N + + azn
1 c
Then, with Eq. (282) below, show that S
P
- ,_ - A 1 + - 1 - 1 1 by Euclids Algorithm. Then the as are symmetric and
Qn A, A,-1 + -
... A2 * + + N = p2,-1 + p,2.
EXERCISE
139. Show that if one runs the indices backwards one obtains
For example: 14291620 + 1.
EXERCISE 148. Conversely, if an odd prime N is a sum of two squares,

consider its representation by Eq. (287). Then expand by Euclids
Algorithm :
1 1
-
pn = a,+1+
plb-1
-
an+z + . . + aZn
*
-
+
If the next to the last convergent is u/u and s = upn upn-, , then 0 <
s < N, and Nls2 1.+ Theorem 78. For all positive n, r, and s,
QnIQm i P n IP(Zs+l)n *
EXERCISE 149. I n Exercise 147 it is not necessary to complete Euclid's
Algorithm in order to determine n. The largest numerator < flis p , . It will be convenient in such investigations to introduce two new se-
EXERCISE 150. quences. From Theorem 77 we obtain
P z n f &Qzn = ( P z f ~ Q z ) "= (2 f @)". (292)
p(N) < 2N. ( 288)
EXERCISE
151. This implies
P, then On
If Pn = - N +Bn Pn-1 P2nt.2 = 2P2n + 3Q~n 1 Qz~+z= Pzn + 2Q2n (293)
Qn
=
+
Bn Pn-1 . But since AZn+2= 1 for all n, we also have, from Eqs. (257) and (258) ,
EXERCISE 152. If xn +f l y n is the nth solution of Eq. (243), then the Pzn+z = P z n + Pzn+l 2 Qz~+z = Qzn + QZn+l *
2nth solution is given by Newton's Algorithm for taking square roots:
Then, from Eq. (293), we obtain the odd-order convergents:
xZn/Yzn = [x*/yn +Z
N J
P ~ n + l= P z n + 3Q2n 1 QZn+l = Pzn + Qzn
if the right side is in its lowest terms. or

62. FROM
ARCHIMEDES
TO LUCAS P2n+l f &Qz~+I (294)
From = d)= (2 & &),(I
(Pzn f d Q z n ) ( 1 A f~5).
Now = 2(2 f 4)
( 1 f fl)' , and Eqs. (292) and (294) may be
written
we obtain the approximations P,/Qn :
2"(Pzn * dQ2n) = (1 f d)2n,
r$] rq]
4 6 2n(PZn+1f @QZn+l) = ( 1 f @)z"+l-
Using the square bracket = = n, we therefore have, for

all m,
8 9 10 11
2'''(Pm f d Q m ) = (1 f d)". (295)
If we now define
2 ["I2 Pn = t n , 2
El Qn = un 1 (296)
Archimedes' approximations on page 140 are those for n = 12 and 9. Our we have
gear ratio on page 173 is
t, f d u n = (1 f d)". (297)
- 70226 - 26 37 73
Pig
--- __._.- By this definition we override the pulsing character of P, dQn-due +
Qi8 40545 15 51 5 3 '
to the period, p( 3) = 2-and may transfer our investigation to the smooth
and we note that the first factor on the right is PC/Qc. This is not ail iso-
lated result, for we shall prove
sequence tn +a u n instead. For, if we can factor tn and un , we can also
factor Pnand Qn by Eq. (296).
190 Solved and Unsolved Problems in Number Theory Pytiiagoreanism and its M a n y Consequences 191
From Eq. (297) we obtain at once some useful identities:
Corollary. If Qni s a prime, then n i s a prime.
tn+m = tntm + 3unum, We find that Q 3 , Q 5 , Q 7 , Q l l , and Q13 = 2131 are indeed primes. But
Un+m = Untm + - Umtn
Q17 = 67.443. Q19 = 110771 is again prime. This corollary, the numerical
behavior ( 6 primes and one composite) , and the exponential growth of the
Then Qn are all reminiscent of Mn = 2" - 1. Since, from Eqs. (296) and (297) ,
tzn = t,2 + 3un2, we have
Uzn = 2untn .
Since ( 1 + 6 ) . ( 1- - G)"= (-2)", we have
tm - 4 3 u m = (1 - 6)"
= (-2)"(1 + d>-". and see that their formulas are somewhat similar. Let us pursue this
analogy.
Therefore From Theorem 35, on page 72, if m > 1 and odd, and if mi21 - 1, and
- G u m > ( t n+ d u n )
if e is the smallest positive 5 such that m12" - 1, then elf. The analogous
(tm = (-2>"(1 + &>n-m,
result is
and, if m < n, Theorem 79. If m > 1 and odd, and if mlQf, and i f Qe i s the smallest posi-
( -2) mUn-m = Untm - Umtn , tive Qn which m divides, then e If.
( -2) "tn-m = t n t m - 3unum ,
PROOF.Assume the contrary, and let f = qe + r with 0 < r < e. Con-
sider Eq. (300) with n = f, m = r . Then
while, if m = n,
(-2)7Uqc = ujt7 - U J f
( -2) = tn2 - 3~,2. Since m divides Q e , it divides Qne and therefore u p , . Likewise mluf , and
Now we can give the thus nz/u,tf . Then, since ( t f , m) = 1 by Eq. (301), ??zju,, and mi&, . Since
PROOF OF THEOREM 78. From Eq. (298), if m = r n , we see that un(um this contradicts the definition of e, we have r = 0 and elf.
implies U , I U ( ~ + . ~ ) ~ . By induction, unlum for all positive r. Now t,luzn by Now we investigate the analogue of Fermat's Theorem. Let p be an odd
Eq. (2991, and therefore, by what has just been proven, tnluZsn.Thus, if prime, and, using the binomial theorem, we expand
m = 2372, we see from Eq. (298) that tnlt(28+l)n . Then Eq. (291) follows
from Eq. (296) directly if Qn , or Pn , respectively, is odd. To determine
their divisibility by powers of 2, we obtain from Eq. (298), with m = 4,
and from Eq. (296),
Pn+4 = 7Pn + 12Qn Then
Qn+4 = 7Qn + 4Pn.
It follows, by induction, that P 4 k + 2 is divisible by 2, but not by 4, and all
other P n are odd. Likewise Qn is even only for n = 4k. From Eqs. (299)
= 1 + p(p - 1) 3 + _.. P(P - 11.e.2 3'"'/2
and (296) we have 1.2 + 1 . 2 . . . ( p - 1)
2PzkQzk .
Q4k =
But every term except the first is divisible by p , since these binomial co-
It follows, by induction, that for k > 1 and n = 2k(2s + l) , Qn is di- efficients are integers, and the factors in their denominators are < p .
visible by 2k but not by 2k+1. Therefore
Thus Eq. (291) is true for all n. (303)
t, = 1 (modp).
192 Solued and Unsolved Problems in Number Theory Pythagoreanism and its M a n y Consequences 193
Similarly Thus
By Eulers Criterion, 3(p-1)/2

-
= (:) (mod p ) . Therefore We evaluate the Legendre Symbols-say from the table of (,)
~ on Page
up= (i) (mod p ) .

47-and find:
If p = + 1 or 11,
24m u;,-l)/l =0 (mod P I .
Kow we use Eq. (300) with n p , nz 1, and, since tl
24m + 13 or 23,
= = = u1 = 1,
we have
If p = 3UqP-1)/z = 1 (mod PI.
If p = 24m + 5 or 7, 3UqP+1),z = -2 (mod p ) .
(305)
If p = 24m + 17 or 19, u ~ , + ~=) /0~ (mod p ) .
By Theorem 20, (!) = 1 if p = 12m =t1. Therefore for these primes I n the first and last case plu(,Fl)~2. In the two middle cases, since ~ + U W I ) / Z,
while from Theorem 80, pIupTI, we see, from uzn = 2tc,tn , that pjt(,w/z .
we do get a Fermat Theorem, since p 1 2 ~ , - ~ and
, therefore p~&,-l. For
We have therefore proven
the remaining primes # 2 or 3 we have
we find
(3 = - 1. But from Eq. (298) Theorem 81. Assume p prime. T h e n
PI&(*-l)/2 if p = 247n+ 1, 111
upt1= u, + t, = 1+ (;)
- (modp). (306) pIP(,_i,/a if p = 2 4 +
~ 13,
~ 23,
(308)
Together with Eys. (305), (304), and Theorem 79 we have thus proven PIP(,+l,/Z if P = 24m + 5, 7,
Theorem 80. I f p i s a n odd prime, p1&(,+~),2 if P = 24m + 17, 19.
PI&, PI&,-1 1
7 pIQ,+l These and similar results have been obtained by Lucas and by D. H.
according as p = 3, 12111=tI , or 1 2 1 f
~ 5. Further i f plQe and e i s the smallest Lehmer.
(z)
such positive index, EXERCISE
153.
e = 3, e / p - 1, or ejp +1 P, = , &, = (i) (mod p ) .
respectively.
Next we investigate the analogue of Eulers Criterion. From Eqs. (299) EXERCISE 154. For n = p or p f 1, P, = either f l or f 2 , Qn = either
and (301) 0 or f l (mod p ) .
Gun2= tzn - (-2) n. EXERCISE
155. Every prime Qn except Q3 = 3 ends in the digit 1.
If 12 = ( p & 1)/2 we use Eys. (298) and (300) to obtain 63. THELUCASCRITERION
-12uqp_1),2 = t, - 3up + a( - 2 ) (p-1) / 2 With the third case in Theorem 81 we have obtained that which we sought
at the end of the last chapter. We analyzed Pepins Theorem 55 thfre, and
6uqp+l)/2= t, + 3up - ( - 2 ) ( p + 1 ) / 2 . found that this test succeeded as a necessary and sufficient criterion for
the primality of the Fermat number F , because, in If e < 2 we have e/2-, and, by Theorem 78,
F m l (3(F-1)2
+ l)(3(Fm-U/2 -
11, qIQw-1 = Q(M~+I)/?.
Fmdivides only the first factor on the right, and also F m - 1 is a power of 2. , and (Ps, QS)
This cannot be, since q/P(Mp+1)/2 = 1 for every s, since
For A l p we have instead M, +
1 as a power of 2. While Eulers Criterion
P: - 3Q: = 1 or -2.
is therefore useless our new Eiiler Criterion yields
Theorem 82 (Lucas Criterion). A necessary and s u f i i e n t condition that Therefore e = 2. But, by Theorem 80, the index e for any odd q satisfies
211,> 3 i s prime i s e 5 q + 1. Then M,= 2 - 1 5 q. Since qIM,, we have q = M,, that
is, M , is a prime.
A4plp(Mp+l)/2. (309) Finally, since P2 +
&Qz is the smallest solution of x2 - 3y2 = 1, we
T h i s test m a y be carried out eficiently as jollows. Let S1 = 4, SZ = 14, . . , have
S,+1 = S?- 2. T h e n the condition becomes
P2m = P, + 3Qm2 = 2P, -I
M,IS,-l, (310) for any even m, by Eq. (292). If we define S, = 2Pz. we therefore have
or, using residue arithmetic, S1 = 2Pz = 4, and S,+l = Sn2- 2. Since ( M , 1)/2 = 2@, Eq. (310) +
is equivalent to Eq. (309).
Spp1=0 (mod ill,). (3104 We now give a brief account of the Mersenne numbers after Euler. There
E X ~ ~ M P L :E S were then eight known Mersenne primes, the Greek primes:
7 = M,IP, = 7 31 = M5IP16 = 18817. M2 = 3, M3 = 7, M5 = 31, M7 = 127; (311)
To test A17 = 127 we use Eq. (310a) and arithmetic modulo 127. Then the medieval MI3 = 8191; and the modern M17 , M l g , and MB1. Mersenne
stated in 1644 that for 31 5 p 5 257 there were only four such primes,
81 = 4, & = 14, 8 3 67, Sq 42, 8, I 111, 86
0 (mod 127). M31 , M67 , MI27 , and MZs7 . While Euler had verified h f 3 1 the remaining
G
For such a small M , this test requires more arithmetic then Fermats j , and three were beyond his technique. There now ensued a pause of over a
Eulers e p on page 22. But consider 3161. Then eG1 implies about a million century.*
divisions-and also a table of primes of the forms 48% 1 and 488k 367+ + I n 1876 E. A. Lucas used a test which is related to Theorem 82 and is
out to 1.5 billion. However, Eq. (310a) requires only about 60 multiplica- described below. He found that Me7 is composite and MI27 is prime. With
tions, 60 subtractions, and 60 divisions. Arithmetically speaking, a Lucas one or another of these Lucas-Lehmer criteria, and with extensive compu-
test for M , is comparable with an Euler test for M31 , and a Cataldi test tations by hand or desk computers, all doubtful M , were settled by the
for Mlg . year 1947 for 31 < p 5 257. It was found that
PROOF OF THEOREM 82. If n = 2m + 1, M , = 2 - I = 7 (mod 24) M61 , M89 , M107 , and M127
by induction, since M 3 = 7 and
are prime while the other A1, including M2,7 are composite.
4(7 + 1) -I =7 (mod 24). The arithmetic necessary for a Lucas test of M , is roughly proportional
If M, is prime for p = 2nz +
I we have Eqs. (309) by (308). Conversely, to p 3 , since that in the multiplication of two n digit numbers is proportional
assume Eq. (309) and suppose a prime q divides Af, . Then qIP(Mp+l)/zand
qlt(Mp+l)/?.Since u ? ~
= 2u,t, we obtain
* Peter Barlow, in the article Perfect Puumber in A New Mathematical and
Philosophical Dictionary (London, 1814), says Euler ascertained t h a t z3* - 1 =
2147483647 is a prime number; and this is the greatest a t present known t o be such,
PI& ~ , + 1 *
and consequently the last of the above perfect numbers, which depends upon this,
Let e be the smallest positive integer where qlQe. By Theorem 79 is t h e greatest perfect number known a t present, and probably the greatest t h a t
ever will be discovered; for as they are merely curious, without being useful, i t is
elM, + 1 = 2,. not likely t h a t any person will attempt t o find one beyond it.
196 Solved a n d Unsolved Problems in Number Theory Pythagoreanism and its Many Consequences 197
to n2. It is clear, then, that it becomes prohibitive to go much beyond 64. A PROBABILITY
ARGUMENT
p = 257 without a high-speed computer. The Lucas prime M127therefore The Lucas test of M4,,, on an IB M 7090 took about 50 minutes. It is
remained the largest known prime for three-quarters of a century. Further, clear that once again we are up against current limits of theory and tech-
a test of Catalans conjecture was not possible. On the basis of Eq. (311), nology. Suppose one had a computer 1,000times as fast. Then one could
Eulers , and Lucass MI27 , Catalan had conjectured that if P = M , test a n M , for p about 50000 in about one hour. However, there are about
is a prime then M , is prime. If this were true, Conjecture 2 (and therefore 10 times as many primes to be tested in each new decade, so that one would
Conjecture 1 also) would follow a t once. But, for instance, is A!f8191 = MM,, really want a computer 10,000times as fast to do a systematic study out
a prime? to p M 50000.
A. M. Turing in 1951 utilized the electronic computer a t Manchester, How many new Mersenne primes can be reasonably expected for 5000 <
England to test Alersenne numbers, but obtained no new primes. I n 1952 p < 50000? A related question is this: Why do we call it Conjecture 2?
Robinson used the SWAC in California and found five new primes: Surely 20 Mersenne primes do not constitute some serious evidence.
Jf521 , Mm7 , Mi279 , M2203 , Mzzei . The answer is suggested by the prime number theorem:
There are no others for 127 < p < 2309.I n 1953 Wheeler used the ILLIAC
and proved that M s l g l is composite. The computation took 100 hours!
a(N) N p. log n
Although it cannot be said that Catalans conjecture was nipped in the One can give a probability interpretation of this relation. However, it is
bud, i t was definitely nipped. It reminds one of the English philosopher not rigorous mathematics. The probability that an n chosen a t random is
Herbert Spencer, of whom it was said that his idea of a tragedy was a prime is l/log n. The heuristic argument goes as follows. Consider an
theory killed by fact. In 1957 Riesel used the Swedish machine BESK to
show that if 2300 < p < 3300 there is only one more Mersenne prime,
interval of positive integers, M - + A M 5 m 5 M +$An[, with AAf small
compared with $1, but large compared with log M . Then the number of
M 3 2 1 7 . Finally, in 1961, Hurwitz used an IB M 7090 to show that for
primes in this interval we estimate by
3300 < p < 5000 there are two more Mersenne primes, M 4 2 6 3 and M 4 4 2 3 .
The first of these is the first known prime to possess more than 1000 digits
in its decimal expansion, while the twentieth known perfect number,
1M+~AM
M--:AM dmllog m.
- 11,
pz0= 24422(2M23 By the mean-value theorem this integral equals Ahf/log ( M +E ) for a
is a substantial number of 2663 digits. small e. Thus the ratio of the number of primes to the number of integers
here, which we call the probability, we may estimate as l/log If.
EXERCISE 156.The reduction of S: modulo Air, is facilitated by binary Suppose now the Mersenne numbers A t p are tentatively considered num-
arithmetic. For let S, modulo M , be squared and equal Q2 R. If, there- + bers chosen a t random. Since log Jt, M p log 2 the probable number of
fore, R is the lower p bits of the square and Q is the upper p bits, then Mersenne primes M , for p , 5 p 5 p , would then be estimated by
8: = Q +R (mod Af,) . Or, if the right side here is > M , , then Sn2=
+
Q R - M , . Thus the Lucas test requires n o division if done in binary.
EXERCISE 157. (For those who know computer programming.) Estimate
the computation time-say on an IBM 7090-to do a Lucas test on A f 8 1 9 1 . The series on the right can be shown to be divergent, so that by choosing
(For those who have used desk computers.) Estimate the computation p , large enough the probable number P could be made arbitrarily large.
time-using residue arithmetic on a desk computer-to verify the following Now, in fact, the error in our assumption can only rcinforce this conclusion.
the following counter-example of Catalans conjecture : The unrandomness of the At, is all in the direction of greater tendency
1 + 120 towards primality. Thus q + M , if q < 2 p +
I . Again, any divisor of M , is
that is,
M191MM19
of the forms 2pk + 1 and 8k & I , arid all JL, are prinie to carh other.
Everything we know suggests that our assumption errs on the conserva-
2 = 2 (mod 62914441). tive side.
Were such a random assumption valid it would follow, from the It can be shown, by induction, that
1
known rate of divergence of - , that if M p i are the successive Mersenne
P (314)
primes, then log p i would grow exponentially. Empirically, the sequence
p i = 2, 3, 5, 7, 13, 17, 19, 31, 61, 89, 107, 127, 521, 607, 1279, 2203, 2281, The analogue of Theorem 80 is
3217, 4253, 4423 suggests a slower, linear growth of log p i . A reasonable Theorem 83. If p i s an odd prime,
guess is that there are about 5 new prime M , for 5000 < p < 50000.
We know much larger composite M , than prime M p . For example, as PIUP, PIuP-1, or PlUP+l
on page 29, M 1 6 1 8 8 3 ~ 2 ~ 1 is
1 composite. Primes of such a size are completely according as p = 5 , 10m f 1, or 10m f 3. Further, i f p J U , , and e i s the
inaccessible to us with our current theory and technology. The Lucas test, smallest such positive index,
when done in binary, appears so simple (see Exercise 156) that it may be
hoped that one could penetrate more deeply into its meaning, and thereby e = 5 , elp - 1, or elp +1
effect the next breakthrough. Alternatively, however, it is also conceivable respectively.
that one could obtain a (metamathematical ?) proof that the number of
elementary arithmetic operations here is the minimum needed to decide the The original Lucas test was based on this Fermat-type theorem for 6.
primality of M , . But, to date, neither of these things has been done, and If M = 1Om - 3, and MIUM+I, but M%Ud for every divisor d of M + 1,
it is an Open Question which is the more likely. it may be shown that A1 is a prime. Since
EXERCISE 158. Give a heuristic argument in favor of infinitely many 2 - 1 E - 3 (mod l o ) ,
Wieferich Squares, p212- - 1. On the other hand, exp1ain their rarity. providing p = 3 (mod 4), the test is suitable for one-half of the Mersenne
numbers, including A167 and M 1 2 7 , but not MZ57. By computing UZP-I and
65. FIBONACCI
NUMBERSAND THE ORIGINALLUCASTEST
UZpmodulo M , one can determine the primality of the latter if p = 3
Why do we single out 4
as a basis for a test; can we not use d$, say, (mod 4).
instead? The answer is that the original Lucas test did use 6, via the Lucas then modified this procedure into an Euler Criterion-type test
so-called Fibonacci numbers. as in Theorem 82. Let
Consider the continued fraction
v,= 1, vz = 3, vn+l= v, + v,4. (315)
x = + -i + i + i + . .
1
1 1 1
Then
Sincex = 1 + ( l / x ) , o r x 2 = x + 1, we have
x = $(1 f &),
If R, Vzn, then it may be shown that
but since x > 0 we must take the + sign. The corresponding convergents =
to (1 + &)/2 are R1 = 3, Rn+l = Rn2- 2.

l 2i 32 5 38 13
1
- 5 21
- g 34
~ . .~- . It follows, if p = 3 (mod 4 ) , that M, is a prime if and only if
Jf,IRp-i.
The denominators (call them ?In)are the Fibonacci numbers. They are
clearly definable by Therefore Eq. (310) in Theorem 82 is also valid if we set X 1 = 3 instead
U1 = Uz = 1, Un+l = Un + Un-1. (312)
of 4, but only if p = 3 (mod 4 ) .
The difference between 4and 6as the basis of a test therefore comes
The numerators are U n + l ,and we have +
to this-all M , are of the form 12712 7, while some areof the form lOm 1, +
Un+,/Un - +~5). (313)
and others are of the form loin +
7. Another reflection of this difference is
that all even perfect numbers, except the first, end in 4 when written in the SUPPLEMENTARY COMMENTS, THEOREMS,
base 12, but they end in 6 or 8 in decimal.
EXERCISE 159. Prove the results stated in this section. More generally,
AND EXERCISES
+
let x1 f l y 1 be the smallest solution of x2 - Ny2 = 4. Let S 1 = x1 and
examine the sequence S,+ = S: - 2. Note that x = z + ( l / z ) where
z = $(x + d r y ) . Specifically examine N = 3, 5, and 6, and develop a
Lucas test based on Sl = 10. Why cant 4 be used as the basis of a We utilize this section to tie down some loose ends developed in the
Lucas test? Relate this to the fact that the @ exists in SnM,-specifically, foregoing three chapters, and also to give some further comments and
(2(pf2)2= 2 (mod M,). exercises of interest in their own right. These results could have been in-
EXERCISE 160. Use Eq. (232) with N = 1 and a + +
bi = cos 0 i sin e cluded earlier, in the appropriate sections, but it seemed better not to
to derive the trigonometric addition laws for cos ( e f +), etc. Interpret attenuate the main argument. The 40 exercises which are given follow the
Eq. (244) as a generalized De Moivres Theorem. Interpret the vectors order of the corresponding topics in the text.
( J , , yt) of Theorem 77 as an infinite cyclic group under the operation de-
termined by Eq. (244). Reduce these vectors modulo a prime p and discuss
EXERCISE 1s. On page 15 we noted a gap of 209 between successive
primes. Show that there exist arbitrarily large gaps by considering the
the corresponding finite cyclic groups.
+
sequence m ! k with k = 2 , 3, ... , m for a large value of m (Lucas).
EXERCISE161. (Lucass Converse of Fermats Theorem.) If mJam-- 1,
and m+ad- 1 for every divisor d of m - 1 which is <m - 1, then m is EXERCISE 2s. A less tricky, but also less simple proof of the foregoing
prime. result may be obtained by assuming the existence of a largest possible gap
m, and showing that a consequence of this is contradicted by. the Prime
Number Theorem.
EXERCISE3s. With reference to Conjecture 5, page 30, consider the
sequence:
Ui+l = 2a; + I,
with a1 = 89. Then a l , az = 179, a3 = 359, a4 = 719, a5 = 1439, and
a6 = 2879 are all primes. But show that in any such sequence, regardless
of the starting value a1 , the a, cannot all be prime. In fact, infinitely many
ai must be composite.
EXERCISE 4s. ( Aus der ballistichen Zahlentheorie )
Two missiles, pl and 112 , are moving parallel to the 2 axis, and, a t time
t = 0, they pass each other in the following kinematic attitudes:
&(O) = X,(O) = 1
X1 (0) = ** ( 0 ) = -1.
p1 hasa sharp nose and many control surfaces, and therefore is dccclcrated
by a skin-friction drag force lvhich is proportional to its velocity. p 2 , a
much older model (circa 1850), has a blunt nose and no control surfaces,
201
202 Solved and Unsolved Problems in Number Theory Supplementary Comments, Theorems, and Exercises 203
and therefore is decelerated by an air-inertia drag force which is propor- 7s. Let N be written in decimal:
EXERCISE
tional t o the square of its velocity.
N = a, 10 + a,-l lonp1+ ... + al 10 + .a0
Let the sum of the digits be
is the mean relative velocity, and is an analytic function of t . Show that the
S N = a, + U,-I + ... + + a1 a0
initial value of its nth derivative, that is and the alternating sum and difference be
D N = a, - + ... + ( - l ) a o .
dV
dt
(O),
Using residue algebra prove the divisibility criteria :
+
is an integer if and only if n 1is a prime. 3INw 3 1 s N ,
EXERCISE 5s. Consider the seven sets of four residue classes b modulo 24 9/Ntt g(SN,
in the table on page 47. Omitting the residue 1 the remaining residues may
be diagrammed as follows: 11(N 11(DN.
EXERCISE 8s. (Gauss, Reciprocals, and Fermats Theorem) On pages
53-54 we indicated that Gauss independently discovered Fermats Theorem
n2+6 from his studies, as a boy, of a table of reciprocals. Let us put ourselves in
his place and reconstruct his discovery. Gauss computed a table of recipro-
cals l / m out to m = 1000. If p ( m ) designates the period of l / m in decimal,
the period for all m < 100, and prime to 10, is given in the following table:
PERIOD OF I/m IN DECIMAL
m m m m
1 - 3 1 7 6 9 1
11 2 13 6 17 16 19 18
21 6 23 22 27 3 29 28
Six of the seven sets are shown as straight lines through three points. But 31 15 33 2 37 3 39 6
+
the seventh set, n2 1, is represented by the dotted line. 41 5 43 21 47 46 49 42
This would appear to give special roles to the form n2 + 1 and the res- 51 16 53
63
13
6
57
67
18
33
59
69
58
22
idue class in the center, 23 = - 1 (mod 24). But a priori no residue class in 61 60
8 77 6 79 13
71 35 73
3 n 2 4 except 1, and no subgroup of order 4, has a special role. Show, in fact, 41 87 28 89 44
81 9 83
that any of the seven n2 + a may be given this special role, and any E 91 6 93 15 97 96 99 2
not on it may be placed in the center. There are thus 28 such diagrams.
But is there a configuration of seven straight lines and scven points, with Now, being Gauss, the reader a t once notes: p ( m ) < m always; p ( m ) =
each line on three points, and three lines through each point, so that we m - 1 only if m is prime; but p ( m ) # m - 1 for every prime m ; however,
+
could draw a diagram with no n2 a in a special role? if p ( m ) # m - 1 for some prime m , then p(m)jm - 1 for that prime; and
EXERCISE
6s. Show that Conjecture 121, on page 48, implies that this implies rn11Om- - 1 for every prime m other than 2 and 5.
i
Pl ( N ) - P4 ( N ) .
Now prove this conjecture by noting, first, that the significance of the
p ( m ) = m - 1, for some prime m, is, that during the division involving
p ( m ) digits, each remainder r (that is, residue of lo), from r = 1 to
Compare with the empirical data.
r = m - I, occurs exactly once. If p ( m ) = e < m - 1 for some prime m,
e different remainders occur. If a is not one of these, a / m also has a period by the formula
of e and these e remainders are all distinct from the foregoing. By continua- M
tion, and exhausting all possible remainders other than zero, elm - l . But x= a; - ci (mod M ) . (319)
i=l VLt
the base 10 plays no essential role in the argument so that, for any prime
m, ( a , m ) = 1 implies mlarn- - 1 or a- = 1 (mod m ) . As an example find the four square roots of unity modulo 2047 = 23.89
Now, reader, relinquish your role as Gauss, resume that of a student and by solving all four cases of
verify that Gausss proof of Fermats Theorem, in his book Disquisitones
x = f l (mod23)
Arithmeticae, is essentially that which we have just reconstructed, and
further, with a slight abstraction, this is the classic proof of Lagranges
{x = f l (mod89).
Theorem (Exercise 71 on page 86) given in any book on group theory. Further, two solutions of x2 - 2 = 14 (mod 2047) are obviously x = =t4.
Note that whether one is led to Fermats Theorem via the perfect num- Find two others.
bers, or via periodic decimals, the problem does not initially concern
EXERCISE 12s. Investigate the parallelism between the proofs of Theo-
itself with the concept of primality. The concept asserts itself, and enters
rems 34 and 36, both of which are due to Gauss. But also consider the
the problem whether the investigator wishes it or not.
significant difference whereby the + ( d ) solutions z in the former theorem are
EXERCISE 9s. Let p be prime and p t a . If p = 1 (mod 4), a and -a are given explicitly, while the + ( d ) residue classes of order d in the latter are
both quadratic residues of p , or neither is. If p = -1 (mod 4), exactly shown to exist nonconstructively.
one of the two, a or -a, is a quadratic residue of p.
EXERCISE 10s. In Exercise 34, page 47, we saw that any value
EXERCISE 13s. If g is a primitive root of p , a prime of the form 4m + 1,
then so is p - g a primitive root of p .
of a = - 1, f 2 , f 3 , or f 6 is a quadratic residue for one-half of the primes,
and a quadratic nonresidue for one-half of the primes. Investigate this EXERCISE 1 4 s . Show that the two proofs of the if part of Wilsons
problem for all a. Start with any a = ( - l ) ( z r l ) z pwhere
, p is an odd prime, Theorem, that by Dirichlet, equation ( 5 2 ) , and that of Exercise 54, page
such as a = -3, +5, -7, -11, +13, etc. Use Theorem 30 on page 63, 74, are not as unrelated as they seem a t first.
and the Quadratic Reciprocity Law in the form preferred by Gauss: For the classical trick of summing s = EzZ: n is to write the same sum
backwards and associate integers with a common sum, thus:
+
s =
s = ( p - 1) + ( p -1 2 ) ++ ( p -2 3) ++ ..... . ++ ( p -12 ) + ( p - l )
( p - 1 ) + ( p - 1 ) + ( p - 1 ) + . . . + ( p - 1 ) + ( p - 1) .
and let q be a prime of the form k p b. Now let a = ( - l ) ( M - l ) z Mwhere
,
M is a product of distinct odd primes, and use Theorem 33 and its Corollary. 2s =
Then let a be the negative of the foregoing, with q of the form k ( 4 M ) b. + On the other hand Dirichlets proof associates integers with a common
Introduce a factor of 2 with q = Ic ( 8 M ) +
b, and finally introduce any product, and one proof is a logarithmic version of the other.
square factor. As an aside the reader may note that the same classical trick, ab-
EXERCISE 1 IS.Generalize the ideas in Theorem 33 to obtain the famous stractly speaking, is also at the foundation of Euclidean metric geometry.
Euclids I, 34 states that the diagonal of a parallelogram divides it into
Chinese Remainder Theorem. Consider n moduli m, prime to each other:
two equal parts:
( m , , m,) = 1 (i # j ) .
T h e n the set of congruential equations:
x = ci (mod m,) (i = 1 , 2 , ... , n ) (317)
has a unique solution x modulo the product M = ml . m 2... m, . T h e solution
mag be obtained from the incerses: The parallel postulate comes in at I, 29, and the reader may verify, in the
diagram on page 129, that all further consequences of I , 29 leading up t o
the Pythagorean Theorem utilize this I, 31.
EXERCISE 133. A student, S. Ullom, notes in the diagram on page 75, This portion (sub-graph) of the cycle graph is already three-dimensional
that if we take differences modulo 17 we get the cyclic group again, rotated (nonplanar). To see this, let us attempt to place these six residue classes
through a certain angle: in a plane and connect them without any crossing lines. First draw
lo 3
8 9
The path shown is a so-called Jordan Curve, and by the Jordan Curve
Theorem, which see, the third quadratic residue can topologically go into
Prove that this property holds for every prime p and primitive root g, only two places, the inside or the outside. Any two points in the
EXERCISE 16s. In the definition of subgroup on page 83 it is redundant inside may be connected by a continuous arc lying wholly within the in-
to stipulate that the set contains the identity. Further, if the group is side. Similarly for the outside. But if one point in the inside is joined to a
finite, it is also redundant to stipulate the presence of every inverse. A sub- point in the outside the connecting arc must cross the Jordan Curve.
set of a finite group therefore is a subgroup of that group if it merely Choose the inside for the residue class 37 and connect to 8 and 62:
satisfies the closure postulate, (A) on page 60.
EXERCISE 17s. Ullom asks if the converse of Theorem 41 on page 85 is
true. If all the squares in a group have an equal number of square roots
El
I .
is the group necessarily Abelian? Answer by A. Sinkov, no. There exists a

non-Abelian group of order p 3 for every odd prime p wherein each element
has one square root.
EXERCISE 18s. If 311, m,t, their cycle graphs may be drawn so that
that they look alike, i.e., they may be superimposed. Show that the con-
verse is true; if they look alike, they are isomorphic. Now, by the Jordan Curve Theorem, we have three options for locating
55. Complete the proof that this sub-graph is nonplanar.
EXERCISE 19s.To prove the criterion for the three-dimensionality of the Since it is nonplanar it is clear that completion of the cycle graph, by
cycle graphs of certain 311, given on page 97 proceed as follows. First, adding other residue classes and lines, cannot undo this property, and there-
note in 31163 the configuration involving the 3 square roots of unity other fore %63 is also nonplanar .
than 1,namely, 62,8, and 55, and any three of the quadratic residues other Finally show that if 32, has at least two characteristic factors, f, and
than 1, say, 4, 25, and 37: fr-l , which are not powers of 2 the cycle graph of 32,,, must contain a sub-
graph similar to the foregoing and therefore m, is three-dimensional.
EXERCISE 20s. If, as on page 97,
@
,,, = <2.> * <2b> . . . <ay> . <2N>
for some odd N 2 1, and likewise if
= <2> . <2b> . . . <2> . <2N>
with the same characteristic factors except for the last, prove that the EXERCISE
30s. Attempt to prove Theorem 75 on page 168.
cycle graphs of 311, and Sn, may be drawn so that the latter will contain EXERCISE 31s. (Euler's Identity) Unlike the result of Diophantus on
N' lobes of the same structure as the N lobes in the former. page 159 for sums of two squares, m = a' 6' + +
c2 and n = e' f' + +
g2
EXERCISE 21s. If N = 1 in the previous exercise we may say Fm, is one- do not imply that mn = j' + +
li2 12. Find a counter-example. Derive from
lobed. Examples illustrated on pages 87-91 are ms, m15, 3 T h , % 4 , %&,and the vector algebra on page 169 the true relation:
me.Gauss proved that an m-sided regular polygon may be constructed
with a ruler and compass if, and only if, m is a power of two times a product
of distinct Fermat primes. An m-sided regular polygon is therefore SO
constructable if, and only if, m, is one-lobed.
But for four squares we again have a result analogous to that for two
EXERCISE 22s. Prove the statement on page 98 that a cycle graph which squares, for show that the last equation is a special case of Euler's Identity:
contains four lobes of ( 2 . 2 ) does not represent a group since it implies a
violation of the associative law. (az + b' + c' + d 2 ) ( e 2+ f' + g2 + h')
EXERCISE 23s. On page 102 we indicated that the computations for ob- = ( a e + bf + cg + dh)' + (af - be + ch - dg)' (321)
taining the representation of m, from the primitive roots of the corre-
sponding primes were indicated explicitly in the proof of Theorem 44,
+ ( a g - bh - ce + d.)' + ( a h + bg - cf - de)'.
However, on page 99, only one of the two, h or h + p , was proven to be a With reference to a textbook of modern algebra examine the parallelism
primitive root of p k . k > 1. Remove this tentative feature by showing, first, between Diophantus's Identity and complex numbers on the one hand,
that if and Euler's Identity and quaternions on the other.
h, = h + np, ( n = 0, 1, 2, ... p - 1) EXERCISE 32s. (A false start on Theorem 61) In view of Euler's Identity
in the previous exercise, if
the p values of h;-' modulo p' are all incongruent and exactly one of them,
say h, , satisfies hz-' = 1 (mod p'). Now solve for m and give an explicit p = w2 + 2 + y2 +
formula for h,+l. The latter is a primitive root of p k for all k > 0.
for every prime p , the Fermat-Lagrange Four-Square Theorem on page
EXERCISE 24s. If the fallacious result in Exercise 80 on page 102 were 143 would follow by induction. Now, by Theorems 60 and 72,
true, it would follow from Exercise 79 that if 3 is primitive root of 487, it
would not be a primitive root of 4872.But show by computations like those p = + b2 + 0' + 02,
az
on page 102 that 3 is a primitive root of both 487 and 487'.
or p =a2 + b2 + b2 + 0'
EXERCISE 25s. There are infinitely many primes of the form 12k - 1.
EXERCISE 26s. The integer 2047 is a fermatian but not a Carmichael for all primes except those of the form 8k + 7. Of the latter one-half are of
number. If(a, 2047) = 1, the probability that = 1 (mod 2047) is a. the form
EXERCISE 275. Prove Theorem 57 on page 138 without explicit reference p = a' + b2 + b' + b'
to unique factorization. For if ( c , a ) = g , lct c = Cg, a = Ag, and utilize by Exercise 129. Now attempt to express a t least some of the remaining
( A , C ) = 1. primes in the form
EXERCISE 28s. Attempt to prove Theorem 63 on page 150 for the
Gaussian integers. These are the algebraic integers in the field k ( e2ra'4)-see
a* + bz + (2b)' + 0' or a2 + b2 + b2 + (26)'
page 152. If you succeed, try also k ( e2r"3),and if you succeed here attempt by use of Theorem 75. The attempt fails.
to prove Conjecture 16 for the osponent 3. See page 152. EXERCISE: 33s. (Lagrange's Four-Squarc Theorem) A proof of Theorem
EXERCISE 29s. Attempt to prove Theorem 68 on page 162 by elemen- 61 is known which is remarkably like that of Theorem 60 on page 1.59.
tary means. Alternatively, investigate the elliptic theta functions and There are small differences, due first, to the fact that Theorem 60 applies
attempt to rederive Jacobi's proof mentioned on page 165. only t o primes = 1 (mod 4 ) while Theorem 61 applies to all primes, and
210 Solved and Unsolved Problems in Number Theory Supplementary Comments, Theorems, and Exercises 21 1
second, because 4(3) = I , while 2 ( 9 2 < 1. We first prove the

quadrates* respectively. Verify the generalization stated by one of Eulers
Lemma. Por every prime p there i s a qo such that 1 5 qo < p and sons that if
pgo = a: + b? + c? + d?. (322)

k k
3 = 2 qk +
rk with 0 < r k < zk, ( 323 )
For p = 2 this is obvious, and for p = I (mod 4 ) we proceed as on page then the integer 2k qk - 1 cannot be written as a sum of fewer than I ( k )
I59 with co = do = 0. For p = -1 (mod 4 ) let a be the smallest positive positive kth powers where
quadratic nonresidue of p . Then a - 1 and p - a are both quadratic
residues; and, by adding these, find an a,, and 6 0 such that
pyo = a? + b: + 1 + 0 Waring had earlier implied that every positive integer is the sum of I ( k )
with q o < p . non-negative kth powers.
If qo is even in ( 3 2 2 ) , 0, 2, or 4 of the integers there, a. , bo , co , and do , A great deal of modern work in this direction has succeeded in nearly
are even. By associating integers of the same parity, and renaming them if proving Warings Conjecture. Hilbert proved that for every positive k
necessary, show that there is a smallest g ( k ) such that
g(k)
n = C xmk
m= 1
But if qo is odd proceed as in equation ( 2 1 7 ) , page 159, etc., using the for every positive n, with non-negative 2,. But he did not show that
identity of Euler instead of that of Diophantus, and thereby obtain a q1less g( k ) = I ( k ) , nor even give it an upper bound.
than qo . Now complete the proof, again using (321 ). Wieferich proved that g ( 3 ) = 1 ( 3 ) = 9, and Pillai proved that g(6)
The foregoing proof of the Lemma uses ( - I [ p ) = ( - 1 )p for both = 1 ( 6 ) = 73. Dickson and Niven proved that if
classes of odd primes, p = 2 P +
1 3 & I (mod 4). A different proof uses
the Box Principle on the p +
I residue classes rk 5 zk - qk (325)
x2 an d - 1 - x2 and k 2 7 , then g ( k ) = I(k). Verify ( 3 2 5 ) for 1 5 k 5 10. It is now con-

jectured that ( 3 2 5 ) is true for all positive k , and, if this were true, Warings
with 0 5 x 5 P . Then show that Conjecture would be proven for every k except 4 and 5 . If ( 3 2 5 ) is false
xi2 E - 1 - x.2 for some k , and if f k = [ ( t ) kDickson
], showed that, for that k ,
3 (modp)
for a t least one i and j, and thus th at g(k) = I(k) + .fk 7
pqo = xi2 + X; + 1 + 0 or g(k) = I(k) + f k - 1,

for a qo 5 P. according aS 2 = f k qk + +
f k qk O r zk < j k qk jk + +
qk . Verify that
EXERCISE 34s. (Warings Conjecture) The integer 7 = 4 1 1 I + + + fk + +
qk f k qk - 2k > - 1 and therefore one of these two conditions must
cannot be expressed as a sum of fewer than 4 squares. Similarly prove hold. I n these cases, if any such exist, Warings Conjecture would be
that false.
With reference to the ideas suggested by the cycle graph for m64 , show
23=8+8+1+1+1+1+1+1+1 that, if k >= 3 ,
and rk 5 2k - 5 .
79 = 4.16 + 15.1 Estimate by heuristic probability considerations the probability that
cannot be written as a sum of fewer than 9 positive cubes, and 19 bi-
* Biquadrate means fourth power.
(325) is violated for a particular k . Therefore show that the odds would Morc generally, a representation in the form
favor the truth of Warings Conjecture for all k 2 10.
There remain the hardest cases, k = 5 and k = 4. Dickson showed that n = C =t( a , ) k
1
I(5) = 37 5 g ( 5 ) 5 54, is allowed in this easier Problem (E. M. Wright et al).

and Chandler showed that Show that, with this new degree of freedom, the 4 squares necessary in
Lagranges Theorem may always be reduced to 3 , e.g. :
I(4) 19 5 g ( 4 ) 5 35.
=
28 = 14 - 13 + 1.
It is curious that the easiest exponent for Fermats Conjecture, namely,
4 , is the hardest exponent for Warings. Nonetheless the earliest and More generally, a representation as a sum and/or difference of three
simplest result, due to Liouville, is for this exponent, and like the same squares is also possible for those algebraic integers which may be written
exponent in Theorem 62 it utilized the theory for the exponent 2. as a sum and/or difference of any number of squares (R. M. Stemmler).
Liouville showed that g ( 4 ) 5 53, that is Examine the Gaussian integers and show that not all of them are repre-
sentable as a sum and/or difference of squares.
63
EXERCISE 333. (Theorem 76 for N = -2, see page 171)
n= Ex,,,. 4
m=l
Theorem 76, Let n = f1 (mod 8 ) and be > 1. If n i s prime, n = a2 - 26
Let n = 6p + r , and, using the Four-Square Theorem, in a unique way in positii,e integers a and b such that 6 5 m 2 . Further
( a , b ) = 1. Concersely, i f n = a2 - 26 i n a unique way in non-negative
n = 6a2 + 66 + 6c2 + 6d + r. integers with b S m 2 , and if ( a , b ) = 1 , then n es prime.
Again use this theorem on a, b, c, and d so that First show, for any positive n, that if n = a - 2b2,and if a > d%, we
n = 6 ($ x?)2 + 6 ($ x ? y + 6 (8x?)2 + 6 ($ x?>. + r,

also have n = a: - 26; with
al = 3a - 4b,
where T = 0, 1, 2, 3, 4, or 5. Now Liouville uses the identity: b1 = 3b - 2a,
a n d 0 < al < a. For the smallest a > 0 we therefore must have a 5 6,
and b 5 f l 2 follows.
Show uniqueness, for n prime, somewhat as on page 160 with Eq. (232)
+ (xi + + ~ 4 ) (51
~ - ~ 4 + (Q +
) ~ ~ 3 + -
) (ZZ
~ ~ 3 ) (326)
~ instead of (215). From the analogue of Eqs. ( 2 2 2 ) and (223), and the in-
equalities for a and 0, obtain a contradiction with a solution of
+ .( + x d 4 + (x2 - x4I4 + (53 + 24)* + (53 - x4)4.
a - 2b = 1
Verify this identity and the proof follows a t once. having 0 < b < 2.
When Liouvilles recipe is applied to n = 79 we get exactly 19 positive Finally, one must prove the converse.
biquadrates, not 53. We get, in fact, the representation on page 210. Further, for n prime, if 5 and y are obtained from Thues Theorem, as
The reader who wishes to pursue these problems will find a vast litera- in the proof of Theorem 7 4 on page 166, show that 2y - x = a and
ture. There is not only the extensive analytic theory (due to Dickson y - x = b give the unique solution indicated in the above theorem. From
et al) mentioned above, but Warings Problem has also been extended to
- A l p = 1 - 2(2(p--1))
algebraic numbers by C. L. Siege1 et al. There is also the so-called easier
Warings Problem. Note that if we allow negative integers, 23 is the sum obtain a solution of
of only 5 cubes: ill, = a - 26
23 = 33 + (-113 + (-1)3 + (-113 + (-1)3 with b < d m .Find such a representation for M7 and two such for Mi1 .
Develop a result analogous to Th eo r~ ni761 for u2 - 3b2. Study the transformation

EXERCISE 36s. Using heuristic probability considerations similar to xl+l = Xt2 - 2 (327)
those used for Merseniie numbers on page 197 argue that there are only a
finite number of Fermat primes as is suggested on page 80. Why is the acting upon every residue class modulo a prime Af, . For A16 verify the
argument less convincing in this case? following diagrams:
EXERCISE 37s. Obtain the constant (35a) for Conjecture 7 by a prob-
ability argument. If the probability of n +
2 being a prime were inde-
pendent of the probability of n being a prime, we could assign l/(log n ) z
as the probability that both are prime. But if n > 2, prime, and therefore
if 2#n, we automatically have 2l;n +
2 . We therefore (tentatively)
correct the probability to 2/(log n)' since on this ground, if n is known to
+
be prime, n 2 now has twice the probability. But, again, if n > 3, prime,
and therefore 3+n, n +2 has 1 chance in 2 of being divisible by 3, not 1
chance in 3. We again correct t o
Q
1
l-5
2--
1
1 (logn)2.
1-3
By continuation, obtain (35a), and by integration obtain the conjectured

asymptote in (35).
For large N it is known that the agreement in ( 3 5 ) is good. Thus D. H.
Lehmer finds 2(37-106) = 183728, while the right side of (35) for
N = 37.106is183582. - 1 . .
EXERCISE 38s. If Z i k ' ( N )is the number of pairs of primes of the form Here the + means application of the transformation (327).
n - k and n + k for n + k 5 N , advance a n argument to show th a t n'ow note the following: 52 = - 10 (mod i116 ). The repeated applica-
Z'*'(N) - Zil)(N),
tion of the transformation
x,+1 = 2%(X,2 - 3) (328)
but
Z(3)( N ) - 22"' ( N ) ,
to any of the 8 possible starters in the top row of the main pattern gives
a cyclic sequence of period 8 which runs through these 8 starters. Applica-
tion of (328) to the second row gives the second row in a cycle of period
EXERCISE 39s. Develop a strong conjecture which bears the same rela-
tion to Conjecture 4 as Conjecture 7 does to Conjecture 6. Using the 4, etc. Omitting the residues 0 and f 2 all +(M6 - 3) of the remaining
datum Z(lOO0) = 35 estimate the number of M , , with p < 1000, for residues in the main pattern satisfy = - 1 while the $( M 5 - 3)
which 2p + 1IMP . Compare with the list on page 28. . " I
EXERCISE 40s. (Lucas Sequences) From page 199 the S 1 = 4 in Theo- residues in the spiral patterns satisfy
rem 82 may be replaced by S1 = 3 for one-half of the hf, . But in Exercise
159 it develops that S1 = 10, like 51 = 4, is valid for every M , . Show Develop a general theory for all prime M, , proving the main theorems,
that besides S1 = 4, and S 1 = 10, there are infinitely many such universal if you can.
starters. For instance 52 is one such, and if x is one, so is x ( x 2 - 3 ) . Hint:
Note, on page 188, that 4 = 2Pz while 52 = 2 P 6 .
C H A P T E R IV
PROGRESS
PROGRESS
Chapter II Fifteen
Chapter Fifteen Years
Years Later
Later
Artin's Conjecture
Conjecture III1
Artin's
Cycle Graphs and Related Topics
Cycle Graphs and Related Topics
PROGRESS
Pseudoprimes and
Pseudoprimes and Primality
Primality
Fermat's Last Theorem
Fermat's Last Theorem II I1
Binary Quadratic
Binary Quadratic Forms
Forms with
with Negative
Negative Discriminants
Discriminants
Binary Quadratic
Binary Quadratic Forms
Forms with
with Positive
Positive Discriminants
Discriminants
Lucas and
Lucas and Pythagoras
Pythagoras 66. CHAPTER I FIFTEEN YEARSLATER
The Progress
The Progress Report
Report Concluded
Concluded First, read the Preface to the Second Edition. Square brackets below
indicate references: [1]-[34] are the annotated references of the first
edition, while [35]-[154] have been added for this chapter.
There has been work on Open Question 1,page 2. In [35] Hagis shows
that no odd perfect number is less than 1050. His long, detailed 83-page
notebook [36] has been carefully checked by his principal competitor
Tuckerman, and so we must accept it as valid. In [37l Buxton and
Elmore claim 1o200, but I do not know that their proof has been
similarly authenticated.
Does this 105' bound change the status of Open Question 1 to that of
a Conjecture? Not in my opinion; 1050is a long way from infinity and
all we can conclude is that there is no small odd perfect number. In
fact, of the 24 known even perfect numbers, only the first nine are
smaller than 1P0,so we cannot even state that P,, = P M , = 1.9 . 1053
is the tenth perfect number. When one examines the elaborate [36] it
certainly seems doubtful that anyone will overtake P, = 219936M19937
9.3 . by such methods. But Hagis himself graciously implies [38]
that Tuckerman's algorithm [39] may be more powerful than his.
There has been work on the table of n(n), page 15. Lehmer's ~ ( 1 0 ~ ' )
listed there is correct as shown, although [3] erroneously gave it as 1
larger. Bohman [40] worried about this discrepancy at length, but he
then continued, using the same method, to compute
n( lo1') = 004118054813,
n( 10") = 037607912018,
n( = 346065535898.
The gap of 209 consecutive composites on page 15 is the largest gap
[4] that occurs up to 37 million. Skipping over intermediate work, which
is referenced in Brent's [41], we find in [41] that the prime p =
217
218 Solved and Unsolved Problems in Number Theory Progress 219
2614941710599 is followed by 651 composites and that all gaps that as N .+ 0 0 , where x(N) i s the right side of (35)and the product above is
occur before p are smaller. Every possible gap 1 , 3 , 5 , . . . up to 533 taken over all odd primes q, i f any, that divide k.
occurs below p , and its first occurrence has been recorded. The evidence In [44] the conjecture is stated in a stronger form: the order of the
in [41] and elsewhere supports the conjecture that I gave in [42] and I error term is given. The heuristic arguments and data given in [44]
now wish to add make Conjecture 19 very plausible. We return to it presently.
Conjecture 18. Let p(g) be th.e first prime that follows a gap of g OT Conjectures 6 and 7 about twin primes are truly key questions. The
more consecutive composites. If all gaps that occur earlier are smaller twin primes 140737488353700 ? 1were the largest known to me in 1962
than g we call g a maximal gap, and we have the asymptotic law but one of the new primality criteria alluded to above has yielded [45]
the much larger pair 76 . 313' 2 1. These primes have 69 decimal digits
1% P( s) -G (329) but no doubt even larger pairs could be found by the same method.
asg+co. Brent [46] (see also [47], [48]) has counted the twins up to lo1' and finds
More general and stronger conjectures are discussed in [41] and in x(10") = 224376048,
papers cited there.
Section 10 made the point, like it or not, that the perfect numbers had so that we could now give one pair to every American.
a great influence in the development of number theory. Aliquot The evidence for Conjectures 6 and 7 is overwhelming, and although
sequences are closely related to perfect numbers. One iterates the they remain unproved, interest has already shifted to the second-order
operation term
r 3 ( N )= f ( N )- z ( N ) . (332)
s ( n ) = u ( n ) - n, (330)
where u( n ) equals the sum of the positive divisors of n. See [43] for an This difference oscillates [46, Fig. 31 around zero in an unpredictable
introduction. If s(n) = n, then n is perfect. Study of these sequences way; it is not understood at all [48].
has surely been one of the causes of the many remarkable new develop- In his famous paper [49] that initiated sieve theory, Brun proved that
ments in primality theory and in factorization methods that have the series
occurred in recent years. So we see the same forces acting before our B = -1+ - 1
+-+ 1- +1- + -1+ - +1- + .1. . 1
very eyes (at a lower level, to be sure). The reason is clear: the perfect 3 5 5 7 11 13 17 19 (333)
numbers (always) and the aliquot sequences (frequently) grow very converges. The denominators here are the twin primes. The accurate
rapidly, and if one is to handle them one is constantly forced to invent computation of Bmn's constant B is a real challenge [50]. Assuming
stronger and stronger methods. The sequences a n 2 1 are also related, (35), Brent [46] estimates
and a project for factoring them has been another cause of these new
developments. Their exponential growth creates the same situation and, B = 1.9021604 ? 5 . lop7. (334)
as before, Necessity becomes the Mother of Invention. This is probably correct, or nearly correct, but the unpredictable r 3 ( N )
Now consider Conjecture 4 and Exercise 16 on page 29, and the makes it very difficult to obtain greater accuracy. While B is a well-de-
answer to the latter on page 169. Exercise 39s calls for a stronger fined real number, its evaluation to, say, 20 decimals would not only
quantitative version of Conjecture 4, and we could also ask for a require a proof of Conjecture 7 but would require the understanding of
stronger modification of Exercise 16. The generalization was given in r3(N ) besides.
r441 and we call it For all primes, the analogous
Conjecture 19. Let f k ( N )be the number of Mp with p 2 N that have a
p r i m divisw d = 2kp + 1. Then
can be expressed in terms of the complex zeros of the Riemann zeta
function [51]. That is bad enough, but for r 3 ( N )we lack even that.
I
I
! 220 Solved and Unsolved Problems in Number Theory Progress 221

1
I
The generalization Z ( k ) ( N of
) Conjecture 7 referred to in Exercise Except for the single linear polynomial f, = an + b, with ( a , b) = 1,
! 38S, which counts the prime-pairs where (338) reduces to the (28) in de la Vallee Poussins Theorem 16, no
case of (338) has been proved. Nonetheless, one can be quite confident,
i n-k , n+k (335)
for example, that althoughf, = n4 + 2 has never been studied, one can
I for n + k S N , had been examined in [4] for k = 1 , 2 , . . . ,70. The now compute its C accurately (say, to 12 decimals) by Epstein zeta
more difficult problem functions and would find that that C and k = 1 in (338) would ac-
Pn+l - ~n = 2k (336) curately estimate its Q(N)for large N .
concerning consecutive primes has been impressively studied by Brent. An unusual result of Hensley and Richards may offer a different type
In [41] he estimated the value of P , + ~ , where (336) is first satisfied, and of evidence. If thefi in (337) are all linear, and if we assume infinitely
in [52] he estimated the number of solutions of (336) for pnt1 I N. His many k-tuples of such primes for each suitable array (337), without
extensive empirical data convincingly agrees with the conjectures de- requiring the stronger result (338), Hensley and Richards [57l show that
duced there from reasonable heuristic arguments. Of course, none of for some integers x and y 2 2 we have
these conjectures was proved. 4%+ Y) > d x ) + dY). (339)
Going beyond the linear polynomials (335) to Conjecture 12, and the Since this contradicts a frequently suggested property of ~ ( x )it, would
table on page 49, let us add [53] as another source of data on P J N ) be desirable to find such a counter-example. There is none with x = y,
besides the earlier [16]. For P,(N) alone, that is, for primes of the form since it was recently proved [57a] that ~ ( 2 %<)2m(x) if x 2 11. While
n2 + 1, Wunderlich [54] has gone much further and we record his an example of (339) would certainly not prove the k-tuple prime
PI(lo6) = 54110 and P,( lo7) = 456362. conjecture, it would at least verify a predicted but unexpected con-
As expected, they agree well with Conjecture 12. sequence.
Goldbachs Conjecture 8 has been verified [58] by Stein and Stein for
The Bateman-Horn Conjecture [34] is a most important generaliza-
all even numbers up to 10. The historically important variant
tion. Briefly (but see [34]), if
4n + 2 = p , + p 2
f,(n),f2(49 . . .f k ( 4 (337)
are k independent, irreducible polynomials in n, and if Q(N) is the with p , = p , = 1 (mod 4), which was mentioned on page 244, was also
number of n 2 N for which all of the k fi( n) are simultaneously prime, verified to the same limit if we allow p , = 1 for a few small n. Hardy
then and Littlewood [9] also gave a strong version of Conjecture 8. If P(2n)
is the number of solutions of
Q(N) - c J N(logdnn)k
2
(338)
then
2% = P l + P2
as N + 00, where C depends upon the array (337) and is given by a very
slowly convergent product.
The linear and quadratic cases above are all special cases of (337) and in the notation of (331). This has been satisfactorily verified to n = 105
all other polynomials that have been studied, such as in [59]. See also [60] for a different version.
fl = n4 + 1, f, = n3 + 3, f, = n6 + 1091, The extensive development of sieve methods since Bruns time (cf.
[Sl]) has been largely directed towards the proofs of weakened conjec-
f, = ( n - + 1, f2 = ( n + I ) +~ 1, tures. The result that is closest to (340) was obtained by Jing-run Chen
etc. have given results consistent with (338). An accurate computation [62] (cf. [Sl]). He showed that, for all n greater than some no, the
of the appropriate C is frequently difficult, but in [55] Davenport and number of solutions of
Schinzel give a useful first approximation. Recently [56], Epstein zeta 2n = p , + Pz,
functions have been found to be very effective in computing many such where Pzis the product of at most two primes, is greater than one-third
constants C accurately. of the right side of (340).
As this is being written, there has just appeared [63] a result of plying by the factor
Pogorzelski which states The Goldbach Conjecture is provable from
the following: The Consistency Hypothesis, The Extended Wittgenstein
Thesis, and Churchs Thesis. I t seems unlikely that (most) number-the-
orists will accept this as a proof of Conjecture 8 but perhaps we should But these p are all = 1(mod 5) and since (515k + 1)= + 1 by the
wait for the dust to settle before we attempt a final assessment. Reciprocity Law, they also have 2 I G and we have already deleted them,
Schinzel [64] has generalized the strong Goldbach conjecture (340). with the factor
This complements the Bateman-Horn conjecture and, although it has
not been studied as extensively as the latter, there is no reason to think
that it is not equally reliable.
(1 - f ).
Finally, returning briefly to (331), we note that only for k = 1is this For a = 5, being a quintic residue is not independent of being a
contained in the Bateman-Horn Conjecture. For k = 3 , 4 , 5 , 7, . . . (331) quadratic residue. That is the only erroneous factor for a = 5, and so
takes us into a new realm and thus suggests [44] that the Bateman- we should expect
Horn conjecture can and should be generalized further. Also of interest
in [44] is the speculation there that it may now be possible to prove
Conjecture 3, which states that there are infinitely many Mersenne
composites. Of all the conjectures in Section 12, Conjecture 3 is cer-
tainly the hardest to doubt and perhaps the easiest to prove. It is instead of (117). Therefore, a = 5 should have a density of primitive
embarrassing that none of the conjectures in Section 12 are yet proved roots that is about 5% higher.
and good strategy therefore suggests a serious attack on Conjecture 3. What is really embarrassing here is that it is just what one finds in
Cunninghams table on page 81! We accepted the high v,(iO,OOO) = 492
EXERCISE162 (Hard Times). In the 4000 numbers fi(n) = n6 + there because it exceeded An(10,000)= 459.6 by less than 2- and
1091, ( n = 1to 4000), there is only one prime. Identify it, and estimate by an imprecise probability estimate such an excess seems to be an
the small constant C in (338) for this fi(n); [65], [66]. acceptable fluctuation. If Cunningham had continued his table for
67. ARTINSCONJECTURES, I1 a = 5 until N = lo5 or lo6 the error would certainly have become
obvious.
Artins Conjecture 13 remains as before: there is little doubt that it is For the seven other a in Cunninghams table, Conjecture 14 needs no
true, but it has not been proved, not even for a single base a, including change. But for a = 13, 17,29, . . . or a = - 3, - 7, - 11, . . . , that is,
the values a = - 2, - 4, + 3 cited in Theoreas 38-40. For a = 3 in for any prime = 1(mod 4), we have the same coupling between 21 G and
Theorem 40, while many new factors have been found for various F,, la\ J G ,and (341) generalizes to
[67l, [a],[69] including the spectacular factorization
F7 = 59649589127497217 . 5704689200685129054721, (342)
not a single new prime F, has been found and perhaps there are none.
Had Cunningham computed the data for a = - 3, the fact that its
Nonetheless, there is little doubt that Conjecture 13, and Conjecture 14
density runs 20% higher than that for a = 2 and 3 would surely have
also, are true for a = 3. exposed the error much earlier.
But Conjecture 14 is not true, as it was stated, for all a # b with D. H. and Emma Lehmer discovered and analyzed these errors in
n > 1; in particular, it is false for a = 5. The heuristic argument for
their aptly entitled paper Heuristics, Anyone? 701, where they did
a = 2 on page 82 is sound, and it also applies to a = 3. But for a = 5 it include data for a = - 3. For most small a the correction needed for
is not sound; Artin has an oversight here and we have followed him too
(117), if any, is rather obvious; but the general case is somewhat
uncritically. Those p that have 5 as a quintic residue, i.e., those for
complicated, and for brevity we refer the reader to Heilbronns for-
which one has 51G in the notation above, were deleted there by multi- mulation in [71, secs. 23, 241.
EXERCISE 163. Show that for a = - 15 there is coupling between the Conjecture 20. If vF(N)i s the number of primes 5 N that have a
cubic and quintic residues and therefore the conjecture should be Fibonacci primitive root, then
-
v - 1&N) 94A~(N)/95.
vF( N)- 27
- AT(N)
38 (347)
Let us now record
asN-+oo.
Conjecture 14 (Amended). If a i s not - 1 or a square, then I t was suggested in [74] that Hooley's conditional proof of (343) could
va(N) -f a A 4 N ) , (343) probably be modified to be a conditional 'proof of (3457, and this was
recently done by Lenstra [75].
where fa i s a rational number given by Heilbronn's rules [71]. Fre-
quently, e.g., for a = - 6, -5, -4, -2, 2, 3, 6, fa i s simply equal to 1. 68. CYCLEGRAPHS AND RELATED TOPICS
The next big development was that Hooley [72] proved Conjecture 14
On page 84 we indicated that tm, and tm,+' are isomorphic for n = 3,
(Amended) conditionally. He showed that (343) follows if one assumes
15, and 104. This sequence continues with n = 495,975,. . . . For n <
that the Riemann Hypothesis holds for certain Dedekind zeta functions. lo8 there are twenty-three examples, the last of which is n = 48615735
(Clearly, that implies that Conjecture 13 also follows under these
(verify). It is not known whether the sequence is infinite, and that is
conditions.) His proof goes well beyond our subject matter and we
also true of the much larger set of n for which H n ) = dn + 1).The
confine ourselves to one remark: Hooley's bound for the error term
latter condition is necessary but not sufficient; for n < 108 there are
306 examples [76].
IJ@) - faA4Ol The cycle graphs have proved to be useful when working with finite
is rather large compared with the known empirical data. Abelian groups; and I have used them frequently in finding my way
Baillie computed both sides of (343) for all a between - 13 and + 13 around an intricate structure [77, p. 8521, in obtaining a wanted multi-
inclusive and all N up to 33 . lo6. In my review [73] of this extensive plicative relation [78, p. 4261, or in isolating some wanted subgroup [79].
table, I point out that Any two Abelian groups that have superimposable cycle graphs are
isomorphic, as in Exercise 18s.That is true for any groups, Abelian or
not, that are of order < 16; but for order 16 one can display an Abelian
and a non-Abelian group that have the same (abstract) cycle graph [80].
is valid for all a and N in this range. While we certainly do not know The non-Abelian one gives a nicer example for Exercise 17S, since its
that (344) remains valid for larger N , this does seem to suggest that it two square elements each have eight square roots. There is a second
may be possible to reduce Hooley's error term, assuming, as before, all pair of such nonisomorphic look-alike groups among the fourteen
needed Riemann Hypotheses. groups of order 16.
An elementary variation [74] on (343) of interest is given by Cyclic groups have such a simple structure that one is surprised when
Definition 42. If g is a primitive root of p that satisfies they yield an important new application. In many problems, one wants
and needs a very efficient solution of
g2 = 1 + g (modp), (345)
xz=a (modp). (348)
we call g a Fibonacci primitive root.
Since (345) implies If p = 4m + 3, the answer is fi = a"+', as in Exercise 47. But
suppose p = 8 m + 5 or (harder) p = 8 m + 1. The importance of (348)
I the
g3 = g+ gz, g4 = g 2 + g3, etc. (mod p ) ,
sequence g o = 1,g' = g , g2, . . . , which would normally
be com-
(346) was obvious to Gauss [81, p. 3731 and to his best English expositor
Mathews [82, p. 531 but neither came up with a particularly efficient
method. Sometimes an efficient method is absolutely essential. In "7'7, p.
puted by repeated multiplication by g (modp), can also be computed 8471 I am analyzing a certain subgroup and must solve (348) for
additively by (346). An example is g = 8 for p = 11, and we have p = (P1 + 3)' - 8, a prime of 37 digits. Unless the algorithm is highly
g2 = 1 + 8 = 9, g3 = 8 + 9 = 6, etc. Now we state efficient, that is impossible. But when one analyzes the location of a in
226 Solved and Unsolved Problems i n Number Themy Progress 227
the cyclic group mP,a very efficient algorithm is not difficult to Note that (156) remains valid in this much extended range;
construct. For brevity, the reader is referred to [83, sec. 51. C 2 ( N ) / l / m has maxima near N = 3 lo6 and 11 - lo6 that are
+
Gauss's book finally got translated into English [84] but unfor- < 1, and it then falls steadily. The earlier (156) suggested Conjecture
tunately the translation was not the best possible [&la]. The German 15, but that conclusion had already been proved by Erdos [SS].We have
edition, which contains considerable additional material, has been re- Theorem 84 (formerly Conjecture 15). Almost all 2-psetulqmmes are
printed [81]. p.ime.
This year (1977) Gauss is 200 years old and I am much tempted to Erdos proved that
have a longish section discussing him, his work, and even his errors. But
we have more pressing topics and for brevity we'll move on. C,( N)e('OgN)1'4/3/ N is bounded. (351)
69. PSEUDOPRIMES AND PRIMALITY
Theref ore
What we called a fermatian in Definition 32 is usually called a (C2") log N U N
2-pseudopmme in the literature. Let us write must approach 0, and the theorem is proved.
But (351) clearly does not prove the much stronger (156) and, in fact,
Definition 43. If Erdos has repeatedly conjectured (cf. 1891) that C2(N)/N'-' and even
-1
== (modn), (349) c(N)/N'-' will increase without bound for every positive E . If he is
n is called an a-pseudqn-ime whether it is composite or not. We correct, C 2 ( N ) / d m will stop decreasing at some N and then will
abbreviate this as a-psp. Let C,(N) be the number of composite a-psp increase without bound. What is that N ?
not exceeding N. If The matter is of interest. If a 40-digit n is a 2-psp, and if (156) holds,
the probability that n is composite is less than lo-''. But if
a(n-1)/2 - (4%)
= (mod 4, (350) C , ( N ) / l / m increases without bound starting a t some unknown N ,
where (aln) is computed as if n were prime, n is called an Eukr a-psix,; we lose that estimate. Erdos's "conjecture" remains controversial; it is
we let E J N ) be the number of these that are composite. Let c ( N ) be not a conjecture as we defined it on page 2.
the number of Carmichael numbers. John Selfridge [87] has improved the subject with his
Poulet's [23] dates from the pre-computer age and has many errors.
Definition 44. If n = t . 2" + 1 with t odd, n is a strong a-psp if
Our table on page 117 reflects all the corrections known at the time of
our first edition, but further errors have been found subsequently [85], at = ?1 (modn) or
[86]. Sam Wagstaff has now gone much further, and Poulet's table atr = -1 (mod n )
should be retired. We show an excerpt from Wagstaff's data [87l. I
for some positive r < s. Let S,(N) be the number of composite strong
have included the ratio C 2 ( N ) / l / m from our inequality (156),
a-psp that do not exceed N .
E 2 ( N )as far as I computed it on an HP-65, and S 2 ( N )(which is defined
Note that when one computes an-' (mod n) one first computes a t
later). (mod n) and then squares this residue s times. Any x that we thus
N C2(N) C2(N)/ m S2(N) c ( N )
encounter which satisfies '2 = 1 must equal +1 if n is a strong a-psp
just as it does if n is a prime.
103 3 0.231 0 1
7 EXERCISE 164 (SELFRIDGE).If n is a strong a-psp it is also an Euler
104 22 0.628 5
a-psp. The two concepts are equivalent if n = 3 (mod 4) but not if
105 78 0.796 16 16 n = 1(mod 4).
lo6 245 0.874 46 43 Selfridge and Wagstaff have found that
107 750 0.920 162 105
Nl = 2047 = 23 . 89
lo8 2057 0.857 488 255
1282 646 is the first composite strong 2-psp, that
109 5597 0.785
1o'O 14885 0.698 3291 1547 N , = 1373653 = 829 . 1657
I
228 Solved and Unsolved Problems i n Number Theory Progress 229
is the first composite strong a-psp for both a = 2 and 3, that Additional 3-psp are generated by
N3 = 25326001 = 2251 . 11251 n'(m) = (12m + 7)(24m + 13),
is the first for a = 2, 3 and 5, and that and clearly these are not 2-psp.
N4 = 3215031751 = 151 * 751 * 28351 For every a,
is the only composite strong a-psp for a = 2, 3, 5 and 7 that does not n,(m) = (6am + 1)(12am + 1)
exceed 25 . lo9. is both an a-psp and a 3-psp, so that there is little doubt that Conjecture
21 is true.
EXERCISE 165. Show that N4 is a Carrnichael number. Show that N4is
a strong a-psp for a = 2, 3, 5, and 7 but not for a = 11 simply by If (156) remains true (or nearly true) as N -+ co, (353) shows that
showing that C,(N) is neatly trapped between fl/lo$ N and fl /dFN.
However, there is insufficient evidence to designate (156) a conjecture,
and we are aware of Erdos's opinion. Numbers a t infinity are quite
is true for a = 2, 3, 5 and 7, but not for 11. different from those that we see down here: the average number of
their prime divisors increases as log log N and, while that increases
EXERCISE 166. Examine the cycle graph of the subgroup C, X C, in very slowly, it increases without bound. People say that Erdos under-
mN,.
If (a, N,) = 1, the probability that N , is an a-psp is 16/32; the stands these numbers. We do note that the Erdos construction [89] that
probability that it is an Euler a-psp is 8/32 and the probability that it is is said to yield so many Carmichael numbers is decidely peculiar in that
a strong a-psp is 6/32. N2 is an Euler 67-psp but not a strong 67-psp. they all are products of primes ri for which each ri - 1 is square-free.
Our table of C,(N), etc. suggests several questions, all of which are That is most untypical of the known Carmichael numbers; among the
open. We note that E,( N)/ C,( N ) is running a little less than l/2, but first 300 only three have that character, namely:
we do not know what happens as N + co. (We should emphasize that
67 * 331 * 463, 23 43 . 131 . 859, 131 * 571 . 1871.
-
this ratio is an average: for n G 1 (mod 8) alone the fraction is much
larger.) It is probable, but unproved, that c( N)/ C,( N) -+ 0. It is plausi-
ble, but unproved, that Sa(N)/Ca(N) 0 very slowly, say as (log
log N)-'.
In contrast with Erdos's C2(N)/N1-', even C,(N)/log N has not
All told, we regard the Erdos conjecture as an (unlisted) Open Question.
The n in (352) are not Carmichael numbers, since n is not an a-psp for
any a that satisfies (a(24m + 1) = - 1. The numbers
been shown to increase without bound. Nonetheless, we list

n(m) = (6m + 1)(12m + 1)(18m + 1)
Conjecture 21. The ratio Ca(N)/N'/2p' increases without bound fo. are all Carmichael numbers if the three factors are prime, since
all a and any positive E . n( m) - 1 = 36m(36m2 + l l m + 1).
For consider the numbers Therefore, [go, p. 1991 although it remains unproved that there are
n(m) = (Ern + 1)(24m + l), (352) infinitely many Carmichael numbers, there is little doubt that c(N)
where both factors on the right are prime. Then n(3) is the 10th increases at least as fast as CN'/3/(log N)3for some constant C.
composite 2-psp on page 117 and n(69) gives Selfridge's N, above. Since The Wieferich Squares (page 116) are much rarer; for p < 3 . lo9
(2124m + 1) = (3124m + 1) = 1, Theorems 44 and 46 show that each there are still only the old examples of Meissner and Beeger [91].
n(m)is a 2-psp and a 3-psp. How many such n(m)are there < N? As we indicated above, primality and factorization theory have
EXERCISE 167. Adapt the heuristic argument in Exercise 37s to these advanced greatly in recent years. An exposition would require a whole
n( m). Then the desired number should be asymptotic to book, and we merely give some key references here. If n is a strong
a-psp for a = 2, 3, 5 and 7, then n is a prime if it is < 25 . lo9 and
1 . 3 2 0 3 m / (log N),, (353) # N4. But this is based on Wagstaff's table, which required much
where the coefficient is that in (35a). Show that the 25th number in computer time and is therefore not extendable to very large n.
(352) is n(213) and N = n(213) in (353) gives 25.14. Show that the 50th As an example, consider c937in Theorem 58. I t arises in the analysis of
number is n(519), and now (353) gives 49.84. Not bad. a certain simple group [92] and it is essential there that it be prime. But
230 Solved and Unsolved Problems in Nurnber Theory Progress 231
cB7 has 359 decimal digits and it surely would have defied all tech- Theorem 85. The column number i s a prime i f and only i f all
niques known prior to the recent developments. A sketch of its primal- coefficients in it are printed in bold-face.
ity proof is in [92, sec. 41. The key reference is [93], an important paper For a proof, see [98].
of Brillhart, Lehmer and Selfridge. To be very brief, this combines
generalizations of our Exercise 161 on page 200 and of our Theorem 82.
It uses known factors of both n - 1 and n + 1, together with a bound 70. FERMATS LASTTHEOREM, I1
B such that n 5 1 have no other prime divisors < B, and combines all The ratios
this into a powerful primality criterion for n. This has been imple- -144
- - 0.392 and - 72
= 0.393
mented in computer programs and it is now routine to prove primality 367 183
for large primes of, say, 50 digits. Our cg37 is much larger, but its on page 153 are very suggestive; they are nearly equal and one asks:
algebraic source (172) greatly assists us in factoring cgS7k 1, and that What is this number? Since a prime p must pass a gauntlet of
suffices. ( p - 3)/2 numbers B, in Definition 40 (page 153) in order to be
Besides the references in [93], which includes Pocklington, Robinson, regular, we may heuristically estimate the probability P of regularity
Morrison, Riesel, etc., other pertinent references are Williams [94], [95], by
[95a] and Gary Miller [96]. The last contains an idea related to strong
pseudoprimes. Certain factorization methods that give a cornplete fac-
torization may also be used for primality tests if n is not too large. We
P = (1 - a, ( ~ - 3 ~ 2
if we assume that the numerators of the B, are equidistributed (mod

(354)
return to them later; see [65], [78], [97J

In contrast to these highly technical, but very effective, methods we -
p). Then P e m 1 I 2 = 0.60653 as p +. 00, and the density of irregular
primes is therefore given by
close this section with a new necessary and sufficient condition for
primality that has more charm than utility [98]. Conjecture 22 (Lehmer [99], Siege1 [loo]). If I(N) i s the number of
Consider Pascals Arithmetical Triangle with each row displaced two irregular primes S N then
places to right from the previous row. The n + 1 binomial coefficients
of ( A + B) are
p , k = 0, 1, . . . , n, and are found in the n-th row -
I( N ) (1 - e - 1 / 2 ) ~ (N ) = 0.39347~(
N) (355)
between columns n and 3n inclusive. Each coefficient in the n-th row as N - + co.
is printed in bold-face if it is divisible by n. If Conjecture 22 were true, then by Theorem 64, Conjecture 16 would
Then we have be true for a t least three-fifths of all prime exponents.
Conjecture 16 itself is now true for all exponents 6 125000 by
Column No. Wagstaffs calculations [loll. Further, he gives I(125OOO) = 4605 and
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 ~(125000)= 11734. Their ratio equals 0.39245, in good agreement with
0 1 (355). The index of irregularity j( p) is the number of B, in Definition
1 11 40 divisible by p; regular primes have j ( p) = 0 and irregular primes
2 1 2 1 havej(p) 2 1. A related conjecture is
3 1331
4 14 6 4 1 Conjecture 23.
5 151010 5 1 J(N)= 2 j(p)-2~(N).
N
1
(356)
6 1 6 15 20 15 6 1 p=3
7 1 7 21 35 35 21 7 1
8 182856705623 8 The heuristic argument is now even simpler if the same equidistribu-
9 1 9 36 84 126 126 tion is assumed. Wagstaffs data gives J(125000) = 5842 and
10 1 10 45 120 J(125000)/~(125000)= 0.49787, in good agreement with (356). More to
11 1 11 the point is the fact that N = 125000 is not exceptional: J ( N ) / a ( N )
Row No. and I ( N ) / l r ( N )both have only small fluctuations up to this limit.
I 232 Solved and Unsolved Problems in Number Theory Progress 233
~
Of the three conjectures, Conjectures 16, 22, and 23, the last is the Since (361) is already better than the present bound 3 . lo9, the order of
1 the day seems clear: investigate (357) for q = 37. If it is true, then we
t weakest, but conceivably it may be the least difficult to prove. If it is
have a new bound; if not, there must be an interesting mathematical
I proved, then Conjecture 16 is true for at least one-half of all prime
exponents. reason for this failure.
1
Turning to Conjecture 17, it is now true for all p < 3 . lo9 since, as Concerning Eulers generalization designated as Open Question 2 on
we indicated above [91], the only violations of Wieferichs (208) for page 158, I am pleased with my intuition there. I refused to call it a
p < 3 . 109 remain the old cases, p = 1093 and 3511. Prior to [91], conjecture, since I said that there was no serious evidence for it.
everyone quoted the Lehmers smaller bound 253,747,889, but this may Several years later a counter-example
have become invalid shortly after they computed it [27] in 1941. The 1445 = 275 + 8 4 5 + 1105 + 1335 (362)
reason is that they not only assumed the validity of the criteria in (208)
and (209) but also that of all such criteria: was found by Lander and Parkin [103]. Curiously, no other counter-ex-
ample is known. The most probable reason is that further computations,
1
p21 q p - 1 - (357) as in [104], have simply not gone far enough.
for every prime q 2 43. In 1948 Gunderson [lo21 questioned the valid- Since Open Question 2 is settled, let us replace it with
ity of the proofs that had been given for (357) for the last three cases: Open Question 3. Is there a nontrivial solution of
q = 37,41, and 43.
Nonetheless, using (357) only for q 2 31, he deduced a bound for A4 = B4+ C4 + D4? (363)
Conjecture 17 that was larger than 253,747,889, namely Although (363) has been investigated frequently, there is insufficient
< 1.1. 109. evidence to warrant a conjecture. One often reads that the methods of
algebraic geometry are very powerful. Perhaps it is not too unfair to
He showed (Theorem N ) that if challenge the algebraic geometers with (363): find a solution or prove
p2lqip-1 - 1 (358) that none exists. No doubt algebraic geometry itself would be the main
beneficiary, since new developments would probably be required.
for the first n primes: q1 = 2, q2 = 3, q3 = 5, . . . , qn, then p satisfies the
inequality EXERCISE 168 (W. JOHNSON [105]). Determine the probability of j ( p )
= n, using the previous assumption. For n = 0, we gave P = e-I2
(log-&) above.
(2n - 2)! 2
-
.. . -
( n - l)!(n - l)! n! log q l . log q2 . . . log q,
-
I e.
2
(359) EXERCISE 169. The absence of Wieferich Squares p 2 for 3511 < p <
3 . lo9 does not contradict Exercise 158, since the probable number in
this interval is only 0.983. Using (208) and (209) and the sum
Designating the left side by f,( p ) , one finds that the iterative sequence
P = 2fJP) + 1 (360) 5
p = 3 109
p P 2 , what is the probability of a counter-example for Conjec-
converges fairly rapidly to the desired bound for p . Since 31 = qll, the ture 17?
use of (360) for n = 11gives Gundersons bound for Conjecture 17 more 71. BINARY QUADRATIC FORMS WITH NEGATIVE DISCRIMINANTS
precisely, namely,
The most classical of classical number theory is the theory of binary
p < 1,110,061,000. quadratic forms. Yet even here there has been significant development.
If the validity of (357) is proved for q12 = 37 one gets a new bound: We cannot adequately treat all of these topics here, since we largely
confined ourselves above to the classical problems
p < 4,343,289,000. (361)
I p = a2 + Nb2, x2 - Ny2 = + I
If q13 = 41 and 414 = 43 are also good, this becomes
p < 57,441,749,000, that initiated the subject and to their immediate generalizations.
Starting with Fermats Theorem 60, we might add a survey of
and if q15 = 47, . . . , qm = 71 are also good, we have computational methods [lo61 and one new short-cut [log.For Theorem
p < 32,905,961,000,000. 69, let us extend the data for R ( N ) given in Ex. 119 with the results
given in [lo81 and the references cited there: our previous text; the class groups are Abelian and their cycle graphs
R(106) = 3141549, R(108) = 314159053, are particularly informative. We continue with other references for
new factorization methods in the next section.
R(10) = 31415925457, R(lo1) = 3141592649625, In view of the historical importance of Pythagorean numbers (see
Fermats statement to Frenicle, on page 161) it is curious that the
R( = 314159265350589. obvious three-dimensional analogue was not examined earlier. As far as
There has also been interest in Landaus function B(x), which counts I know, it is new. In how many ways can we solve
each integer n = a2 + b2 2 x only once no matter how many repre-
p2=a2+b2+c2 for O<alblc (367)
sentations it may have [109].
In the generalization if p is an odd prime? The elegant answer is in
Theorem 86. Write p uniquely as p = 8n 2 1 or p = 8n i 5. Then
rp = a 2 + Nb2 (364) (367) has e m t l y n solutions.
on page 167, we wish to make r = 1 if possible and to minimize it
otherwise. This relates, as we indicated on pages 153, 154, and 168, to EXAMPLES: p = 3, 7, and 13 each have one solution, p = 5 has none,
questions involving unique factorization and to those concerning the and p = 19 has three.
density of primes generated by quadratic polynomials. In an important 9 = 1 + 4 + 4, 49 = 4 + 9 + 36, 169 = 9 + 16 + 144.
development, H. M. Stark proved [110] that for negative N the 361 = 1 + 36 + 324 = 36 + 36 + 289 = 36 + 100 +225.
quadratic field k( ) has unique factorization only for For a proof, see [117]. I t is based on known classical results involving
N = -1, - 2 , -3, - 7 , -11, -19, -43, -67, -163. (365) ternary forms (see page 246).
A. Baker [lll] and K. Heegner [112] have given other approaches to EXERCISE 170. Determine the nine solutions of (367) for p = 73. Note
this long-sought theorem. Correspondingly, the famous polynomial that 732 - 1Z2 = 61 85 gives rise to four of them.
n2 + n + 41, which has -163 for its discriminant, must have a very
72. BINARY QUADRATIC FORMS WITH POSITIVEDISCRIMINANTS
high density of primes. In [56] we find that we should take C = 3.31977
in (338) with k = 1. Paul T. Rygg [113] has counted these primes up to In contrast with (365) we list
n = lo6, and his count does agree very well with (338). Conjecture M. There are infinitely many quadratic fields k ( V N ) for
For computational developments on (364) we refer to published tables N > 0 that have unique factorixatk.
such as [114], to reviews thereof, such as [115], and to improved This is an important conjecture, since its proof will require a deep
algorithms, such as [83, sec. 61. An example in the latter solves (364) for insight not now available. For the large p in (366), k( G ) has class
every N from 1 to 150 inclusive for a remarkable prime number h = 1 and therefore unique factorization. Empirically, that is
p = 26437680473689 (366) !;ot surprising; for about 80% of known k ( G ), where p is a prime
that we will refer to repeatedly below. Such solutions are possible only = 1(mod 4), we have h = 1 [118]-[120], and this empirical density
because ( - NI p ) = + 1for all N between 1and 150 for this prime. The decreases only very slowly as p increases [121]. Therefore, the a-priori
generalization of Landaus B(x)to n = a2 + Nb2 6 x has been studied odds actually favor h = 1 for the prime in (366). While there are only
in [116]. nine cases in (365), many thousands of such fields have been recorded
Much (but not all) of the recent development in factorization methods for N > 0.
involves binary quadratic forms either explicitly or implicitly. Our The difference arises from the fact (page 173) that one has infinitely
Theorem 76 above is closely related to the Lehmers algorithm [97], many units when N > 0. We must generalize Fermats equation (236)
which may be used both for factoring and for primality tests. The to include the possibilities indicated in Exercises 124 and 144. That
previously cited [78] has these same features; however, it derives its done, we have
greater efficiency not from Theorem 76 but from more advanced ideas Definition 45. If T and U are the smallest positive integers that
involving class groups and composition that we did not study above. We satisfy
must therefore drop the topic, even though it would fit in nicely with
T 2- U 2 N = + 4 (368)
then power but all of the complexity. It is now so simple and requires so
little storage that one can factor
= ( T +U r n ) / 2 (369)
is called the fundamental unit of k ( m ) and R = log c is called its 260 + Po- 1 = 139001459 8294312261,
*
regulator. F1- 7 = 17174671 . 131111671

To be brief, it is known that the product 2 R h / m plays the same on a little HP-67 even though this only computes with 10-digit numbers.
role if N > 0 that the product r h / c N does if N < 0. In the latter Other recent developments in factorization are by Lehman [137),
case, the class number grows (on the average) proportionally to Pollard [138], and J. C. P. Miller [139]. For a survey article, see Guy
C N ; while in the former, Rh grows (on the average) proportionally [140].
to m. Thus, if R is big enough, there is no reason why h cannot equal
EXERCISE171. Since 17174671 has a unique representation A +
1 no matter how big N becomes. So the real question is this: Why are
the fundamental units (369) frequently so large? This takes us back to 190B2 for A = 3991 and B = 81, it is prime. Why do we select 190?
the very beginning. When Fermat and Frenicle challenged the English [141].
(page 172) with N = 61,109,151,313, . . . , they may not have realized 73. LUCASAND PYTHAGORAS
it, but k(m) has h = 1 in all of those cases. Our estimate on page 198 that there will be about 5 new prime Mp
For the p of (366) the smallest u that satisfies u2 - pv2 = - 1 has for 5000 < p < 50000 needs little revision, if any. Four have been found
9440605 decimal digits [122]. That makes even the answer to the famous for p < 21000 and about 5 still seems a reasonable guess. Gillies [la]
Archimedes Cattle Problem [123], [la]look small. The regulator of has found prime Mp for p = 9689,9941 and 11213, and Tuckerman [143]
k ( G ) is 21737796.4. It is that large because (a) the class number is 1, has found M,,,.
(b) p is large, and (c) ( p l q ) = + 1 for q = 3,5, 7 , . . . , 149. This last Gillies included a statistical theory, based upon unproved hypotheses,
point is significant, since an average p this size, not having this which implies that about six or seven prime Mp should be expected in
unusual property, would have a u with only 1116519 digits. each decade: A < p < 10A. Ehrman studied these Gillies hypotheses
Digressing briefly, it is the last point (c) that gives p its mathematical [144] and interpreted previous data [145] on the distribution of the
interest (not its gigantic u). It is the smallest prime = 1 (mod 8) that number of divisors Mp has below a given bound B. These distributions,
has ( p l q ) = + 1 for q = 3 to 149. The Riemann Hypothesis puts a limit and those in (331), constitute first steps in understanding Mp.
on how long a run of residues a prime of a given size can have, and p There has been no computation of Mp to my knowledge since that of
was computed by the Lehmers and myself precisely to test this limit Tuckerman [143]. That is surprising, since it was a t that very time that
[1251. Knuth had begun to publicize [146] the new Strassen-Schonhage fast
Had Frenicle persuaded Lord Brouncker to compute the continued Fourier multiplication algorithm for which one has
fraction for G , they would have found [126] that its period is
18331889. But a new development makes it possible to compute R Theorem S. It i s possible to multiply two n-bit numbers in
accurately in a few seconds of machine time. Exercise 141 shows how to O(n log n log log n) steps.
use symmetry to cut the computation in half. It turns out (surprisingly) This leaves open the pertinent question: For what n does this become
that symmetry is not essential here; the use of composition and competitive with the older O(n) multiplication? It does seem .to offer
quadratic forms allow a doubling operation anywhere in the period, and an escape from our statement on page 195 that the Lucas arithmetic for
therefore repeated doubling is also possible [122]. Mp is roughly proportional to p 3 , and I do not know why this has not
For h = 1 in cubic and quartic fields, see [120] and [l]-[129], while been exploited.
for three interesting continued fractions, see [130]-[132]. We should add that the theory of Lucas sequences plays a large role
Returning to factorization, the continued-fraction method [133] is in many of the new primality tests referenced above, not merely in
complicated but extremely powerful. An interpretation of it in terms of tests for Mp.
quadratic forms [134] is of interest; and subsequently this led to a Returning to the beginning of Chapter 111, the Case for Pythag-
greatly simplified method [135], [136], which loses much of the previous oreanism remains an important philosophical proposition. I know of no
serious discussion or refutation that has appeared anywhere; Eves [147] STATEMENT ON FUNDAMENTALS
merely copied our list without advancing the question. It is therefore
unnecessary to strengthen the case here, but two additions and one
subtraction should be made. The genetic code in DNA and recent
theories of elementary particles are almost pure Pythagoreanism, and it
is hard to conceive of two more fundamental things in the universe. On The logical starting point for a theory of the integers is Peanos five
the other hand, let us delete Eddingtons speculation that hc/2re2 = axioms. From these one can define addition and multiplication and prove
137 from our page 137. I t mars a good case, since subsequent measure- all the fundamental laws of arithmetic, such as
ments [la]have given
hc/2re2 = 137.0388 ? 0.0019.
a + b = b + a,
a ( b + c ) = ab + ac,
74. THEPROGRESS REPORTCONCLUDED
a(bc) = (ab)c,
We are nearly done; but even the Supplement and the commentary in
the first edition References need updating. For more on Exercises 4 s etc. The reader knows that we have not done this. We have assumed all
and 8S, see [149] and [150], respectively. Finite geometries, as in these fundamentals without proof, and even without explicit statement.
Exercise 5S, arise in interesting number-theoretic situations; cf. [120, Sometime, however, if he has not already done so, the reader should go
page 301. through this development, and he can hardly do better than to read
Warings Conjecture (page 211) that every positive integer is the sum Landaus Foundations of Analysis, Chelsea, 1951.
of I( k) non-negative k-th powers is now even more nearly proven- Similarly we have skipped over the simpler properties of divikibility. We
but still not completely. Rosemarie Stemmler [El]completed a verifi- have not defined divisor, divisible, even, etc. If there is an integer
cation for all k up to 200,000, excluding the two hard cases k = 4 and 5. c such that
Mahler [152] had already shown that g(k) # I(k) for at most a finite ac = b
number of k. Continuing developments of Bakers method [111]suggest
that a proof will be found for all k > 200,000, but this has not yet been we say a is a divisor of 6. If 2 is a divisor of b we say b is wen, etc. For these
done. As we indicated on page 212, k = 4 is the hardest case. Chen [153] elementary definitions, and for such theorems as
has now proved that g(5) = I(5) = 37 and, while there has been pro- alb and bic implies alc,
gress on g(4), its value remains unsettled. It seems likely that Warings
Conjecture will be completely proved in due course. alb and alc implies alb + c,
Dicksons valuable History [l] has been reprinted by Chelsea; and the etc., the reader is referred to Chapter I of a second book of Landau, Ele-
dedicated scholar we called for on page 243 has turned out to be Wm. J. mentarg Number Theory, Chelsea, 1958.
LeVeque. His six-volume [154] collection of reviews, while not quite One of these elementary theorems should, however, be singled out for
equivalent to Dicksons History, is certainly a valuable aid to research. special mention. This is the Dizision AZgorithm:
This progress report confirms the statement in the 1962 preface that
number theory is very much a live subject. Even within the limited Theorem. If a > 0, then for every b there are unique integers q and r , with
confines of our previous subject matter, the progress made since then is 0 5 r < a, such that
impressive. b = pa + r;
that is, there i s a unique quotient q and a unique remainder r .
This theorem is indeed a fundamental one in the theory of divisibility.
It enters the theory via the Euclid Algorithm (page 8 ) and elsewhere.
The proof runs as follows. Let b - z1 a be the smallest non-negative
2 39
integer of all integers of the form b - xa, with x an integer. Set q = x1 and
T = b - pa. Then one shows that 0 5 T < a, whereas for any other x2 ,
T~ = b - x 2 a would not satisfy one, or the other, of the inequalities in
0 5 T2 < a.
The key argument is the fact that there exists a smallest non-negative
b - xu. This is guaranteed by the Well-Ordering Principle (page 149). As
is stated on page 149 the latter is equivalent to the principle of induction
(Peanos fifth axiom), and thus is the principle which gives the integers
their special (discrete) characteristics.
TABLE OF DEFINITIONS
De$nition Page
m t )c e fu r1 e P n 1
r ) 2 p c o m p o s i t e 3
m ) o3 g c r d , ) (I se! ai a i op , at vI r t i 5i e s sm e t er ,
) 4 U o P + , h 5
m
) 5 & u, MI n n e r s 4e 1 n n e
) 6 b l 4 1
) n 7( r 5 1
) 8 a ) x ( ) j x s (- g y m 5 p 1 t o t
) s 9 r+(n),
e l uE p f h i u n c0 2 t i o n
b , a 1 0
)T n ( ) 1 2
1 e 1 )q u i n u m 1 e3 r o u s
1 , Ll2 o b m y) S e t g e n 3 3 d r e
0
1 O3 ,)Ll o brm y )S @ ca ( i g g e i 1 n4 n d a r l e
1W P 4 ( C ) I 8 4
= ,1 d bo 5m c ( s) a), c o ,ns ag lr uutdeeoum
nd taoi s e re, u d i l so e r f c,
s u l u d om 5 5
p i d 1 o c am 6 ( c- ) m r ) , 0 6
r 1, d eg s7o l c ) a se si rd ee vnns ti i t y o, c 0i 6 a t i v e ,
ou n a i l1ro e b A 8 g
)r f g fo m o t g hr ie r 1d t 6 o c e u r p
u o n
1ooi tam
c i9l p i t lum
) r gm m 1 6
i s 2 e q 0u a rd )r ra tni c q uo a ed r na t i c s i 3 d6 u e ,
d2 o vm1 )( ) P 4 6
d s2 f a 2io o o ) em m
r d e 1 7 r
2 p
t 3r oi m
o )i rt i v e 3 7
o y r 2 c 4 g) g e n e r a t o r 3 7
o s 2 i, 5s p u o) r g a -A b, TIL ti 3L8 4 7
r i 2a c 6w p
) is r c u 6 7 l a r
d 2 o fc 7 o )a m y m c l 3 e 8
2 s 8 ) u b g r 3 8 o u p
2 &9 I) l 2 9
cm
i t s3i re t@ca0ra,hc , ) r o t c a f j , . 4 9
3 q1 ) u a d r 1 1 4 a t f
3 f e2 r m )a t i a n 1 1 5
e 3 , rFe3 b mu n ) Fm 1 1 5
241
242 Solved and Unsolved Problems in Number Theorg
De$nition Puge
34) C n uam b e r r m i1 1 6 c h a
e i 3 W5 ) S q u a r e 1 1 6
3 a 6 l ma) o s1 t 1 1 1 7
3 P7 )y n t h ua g om r e 1b 2a 1 n e r
3 G8 a u s) is ni at ne g e r s 1 4 9l 1
e 3 B9 ) n & u m b 1 5 2 e o f r ,
e 4m ri 0 r p) e g u 1 l5 3 1,a1 r
1 6 2 1
, ) n 4( r 1 R) ( n )
42) Fibonacci primitive root. . . . . . . . . . . . . . . . . . . . . . . . 224 c
p - 4 a 3 ) r e Ca(N), l u E a . - ) p&N s (( pZc , V ) . ,. .6 . .2 . 2. . . d c I
44) strong u-psp, $(N). . . . . . . . . . . . . . . . . . . . . . . . . . . 227 a b l t
45) fundamental unit, regulator. . . . . . . . . . . . . . . . . . . . . .236 a 5 s i
REFERENCES
The best single source for the historical aspects of number theory is,
of course, the monumental
1. LEONARDEUGENEDICKSON,History oj the Theory oj Numbers, Chelsea,
New York, 1971, Volumes 1, II, III.
Our account here of the origin of Fermats Theorem, of Cataldis table
of primes, etc. is largely drawn from this source. It would be highly desir-
able if a dedicated scholar, with access to appropriate funds and graduate
assistants, should undertake to bring this history up-to-date.
VVe will not attempt, in this section, to indicate the source of everything
in the present book, but will concentrate on the more modern theorems,
conjectures, and tables, and on a few historically interesting points. The
references will be numbered and listed under appropriate section headings.
SECTION5
The primes, out to greater than 108, have been recently tabulated on
microcards in
2. C. L. BAKER & F. J. GRUENBERGER,The First Six Million Prime
Numbers, Madison, 1959.
For the last two entries in the table on page 15, see
3. D. H. LEHMER, On the exact number of primes less than a given
limit, IZZ.J. oj Math., 3, (1959) p. 381-388.
SECTION6
4. D. H. LEHMER, Tables concerning the distribution of primes up to
37 millions 1957, reviewed in MTAC, 13, (1959) p. 56-57.
5. M. KRAITCI~K, Les Grand Nombres Premiers, Sphinx, Mar. 1938,
p. 86.
For a complete account of the early work on Theorem 9 see
6. E. LANDAU, Hanclbuch der Lehre von der Verteilung der Prim~~ahlen,
Chelsea, New York, 1974, Chapters 1 and 2.
SECTION8
For a table of +(n) see
7. J. W. L. GLAISHER,Number-Diviser Tables, Cambridge, 1940, Table 1.
There have been numerous tables of ra,b(n). lor discussion and further
references, see
8. DANIEL SHANKS, Quadratic residues and the distribution of primes,
MTAC, 13, (1959) p. 272-284.
243
244 Rejerences
SECTION11
Historically, Theorem 18 required a long time to get proved-analogous
to the delayed proof in our treatment. It was first proven in its entirety by
Lagrange in 1775-3 years ujter Euler determined the primality of A!fzl.
But, of course, this proof did not use Gausss Criterion, as we do on page
40. It does use Eulers Criterion, and theorems on the prime divisors of
binary quadratic forms similar to Theorems 72 and 74 on page 166. See
reference 12 below, page 209 for an account. By these latter-t)ype theorems,
Theorem 19 may follow directly, and net, as we show here, as a consequence
of the barder Theorem 18. Thus if q 12 M* = Nz - 2, 4 must be of the
form p = x2 - 2~2, and it follows at once that q = 8k f 1. See, in this
connection, the remark on page 143 concerning the fact that quadratic
residues arise most obviously in connection with binary quadratic forms.
SECTIOX 12
The large composite Mp on page 29 were obtained as a byproduct of
the studies in reference 16. See Exercise 17 for the connection.
The two smallest pairs of twin primes > 1012on page 30 are from refer-
ence 5, while the two largest pairs <247 are from unpublished work (June,
1961) by P. M. Fitzpatrick et al.
Conjectures 7 and 12 are from
9. G. H. HARDY B J. E. LI~TLEWOOD,Partitio numerorum III: On the
expression of a number as a sum of primes, Actu. Muth., 44, (1923)
p. 48. ,
The Goldbach Conjecture must surely strike a thoughtful reader as
exceedingly curious. One muZtipZies primes; why add them? When Euler
was attempting to prove Theorems 60 and 61 (pages 142,143) in the 1740s
he was in frequent correspondence with Goldbach. The latter presumably
conceived the idea that ij every integer of the form 4k + 2 is the sum of
two primes of the form 4m + 1, (a variant of the Conjecture), then Theo-
rem 61 would follow for a11integers of the form 4k + 2 from Theorem 60.
From this conclusion it easily follows that Theorem 61 is true for a11posi-
tive integers. Empirically, Goldbachs notion seemed valid, at least if one
allows 1 to be considered a prime. This was the accepted convention then.
Thus 14 = 1 + 13 = O2+ l2 + 2z + 32.
For interesting modern work in additive numbcr theory arising out of
Goldbachs Conjecture, see
10. KHIX~I~N, Three Pearl8 of Number Theory, Graylock, Rochester, 1952,
Chaptcr II.
For the m.ich more difficult Vinogradoff theorem that every sufficiently
large odd number is the sum of three primes, sec
11. T. ESTERMANN,Introduction to Modem Prime Number Theory, Cam-
bridge, 1952, Chapter 3.
Rejerences 245
NO I T C E S 18
F t o Oh r ,eLl o brm y S e si e g e g e i n n d a r l e
, 1 dA 2 r . a DL E Gh E N DTht?orie
cR E~, des
n Nombres,
a E Tome
l I, B N - M A
a P ..1 7 9p 1 9 5 5 ,
NO I T C E S 9 1
y
ao Ul ro t r cn r t n e m i petf r eo eto hc T 7es i rx 2e i ew d tl n b t o l o y k
e r o m c T of h p e f m i o t u o h p erF r obrb e n lp i u si ial e r tg i x o i .c o n a
k on eo t b
t i rhwns i l gis nrEa e p p a o e nt b i
y e 1 l . W 3s J MLe. W. - Topics
En in oNumber s VTheory,
i d Vol.d E 1, A Q U E ,
R .1 ep 7 09 a . d
5 i 6 n , g ,
cd ne soP i e tht t t r ui hr eba a n ftid ix it ep er n h r t e p o tb a eo r et a i r n e gd
n r e F r t u o h ebt i w u f a
t de r e t a i l e d
A . 1 H 4H . Vodesuvgen ber Zahlentheorie, ,Snpi r l i rne gB e r , 1 9 5 0
. p 9 3 .
t t n e su Te r p h ea h n nea ye o d s e tr e 4tf 2 n
e r 51en n
9i ec 1e h ew h i n d e
o p oe n d e n f et m lr yta s h op e u n d
A 1 , D SK5 N A H .eS h T q u a d r ra t i c l e A , a3 c b
t3 6 w si ,t p r r
, ) 2 5 58,
Bull. A.M.S., 91 ( . p 452.
NO I T C E S 0 2
n o C s i i a 2r e w f1 o r do i anp g n i r re . f e re9e n
rF c oea om
o r p o s
e s ir d a p ne r t d i d
n e ns t e ae t a ,
d A c o 1 e, D oS K
6N A
jHw.S n e o l tO c htn f e t o H i & L - na o c r d y
m m g n i n r ie uc t rnh f eo pf m ro to f h nz e+ a, Muth. Camp., 1 4 ,
, ) 0691 ( . p 3 2 1 - 3 3 2 .
NO I T C E S 0 3
c A i 1 t, r~DK7NrA H S. A
ie a c w b p is m a an ru o d p t nc p lu i l c a a t r i
3 t 4 A h 5 b Notices,
e A s Mo ) ,)S 8 m
5t951 (r .a po r 9e y 6t .a c r , h c . . . t ,
NO I T C E S 1 3
r oe p eF m
r o For c r e h t ar u f n dr s ee e f e r e
t n t. 1, .NeO A AaeeP . h T t m rc o im
G~8X rf p ho o t s e t ih t ee - Fnm eu ns s
b ) ) e1 6 9
Muth. Comp., 1 1 ( .. 5 0 2p ,r 4 ,
NO I T C E S 2 3
- S r ee 8f e . re 1e npc 6 4e ,a n d
e. b 1. .A 9J mCmC . i u U t Or n h
f n eoN pf e m
o - tu
ad Nsi hs e r e I N G H
, y t i ca Proc. London Math. . rSoc.,
e )s ) 42,1 9
13,1 ( . p 2 5 8 - 2
w T RvE B2R E Hi 0 ne.rB
tos i v i dmi rPi I m m
i L v i r t o HP r Ag e R g Ze b , e
) ) 7 3114,
Muth. Ann., 91 (. p 4 7 6 - 4 9 2
246 References
21. JOHN W. WRENCH, *JR.,Evaluation of Artins constant and the twin-

prime constant, MU& CW~., 15, (1961), p. 396-398.
SE~TI~X 33
The cycle graphs were investigated by the author in the early 195Os,
but have not been previously published.
SECTION39
22. SI~NEY KRAVITZ,T~~ congruence 2~-l = 1 (mod p2) for p < 100,000,
Muth. Camp., 14, (1960), p. 378.
23. P. POULET, Table des nombres compos& v&ifant le thhorme de
Fermat pour le module 2 jusqu lOO,OOO,OOO,Sphixc, Mar. 1938,
p. 42-51.
24. D. H. LE~MER, On the converse of Fermats Theorem, Il, Amer.
Muth. MonthZy, 56, (1949), p. 300-309.
SECTION47
On pages 142-143 we indicate the historical background of xz + Ny2
for N = 1 and N = -2. The case N = +2 is also intercsting. Euler (and
Goldbach) also attempted Theorem 61 as follows. If one cari show that
every integer of the form 8k + 3 is the sum of three squares,
Sk + 3 = (2u + 1)2 + (2b + 1)2 I (2c + 1)2,

then that proves its equivalence, a theorem of Fermat which states that
every integer is the sum of three triangular numbers:
k = o(o + 1) + b(5 + 1) + C(C + 1)

2 2 2 .
Further, one deduces
4lc + 2 = (u - !_))Z+ (u + b + l)Z + cz + (c + l)Z,
SO that Theorem 61 is valid for each such integer. As we indicatecl above,

this implies its validity for a11integers.
, New, by Theorem 72, a11primes of the form 8k + 3 cari be represented
by the binary form !
SIC+ 3 = (2u + l)* + 2(2b + 1)2,

and this may be easily extended to a11 integers of the form 8k i- 3 that
have prime divisors only of the form 8k + 1 and 8k + 3.
But this binary form does not suffice to obtain such representations as
8.4 + 3 = 35 = 5.7 = 12 + ? + 52.
For these one needs ternury forms. The attempt therefore failed. Although
R 2 e 4 7 j e r
E d mne i r po ouedh T t 1 6r lu L y d aoeehrn le aa a

n o sddvi r thS
g . O e, r i a l n g
t p r o l o nan
gs e md
i eohl bt o r p l a r eh v e sa d i m p o r t ba n t y p r o
NO I T C E S 8 4
, . 2. e L 5Je Mge . r dh T Oi L e c r t ou R rF ebe sr nm m
La t D aT sah se o trC e m
E , L L ,
1 9 2 1 .
ce rn T roh e e cz r a h l f i B ne g oae p 1 ro 4 t 7 r e f e r e2 n c e0 .
NO I T C E S 1 5
f L . 2. E Jo 6L S . . , o.L OC H ~AI N r . . & H SPV A N D I V fE R , o F e r m a
t ms e ar, oL e h T 2 f a op 0 1 r es x1 spn oe0an rel hn .t s c 4o r i P N m a e t i o
A , S
) 5 5 49 1 ( .c 1 pc, 9 a 7
i d 0 . . - , 9 7 3 .
NO I T C E S 2 5
S r ee5f e re 2e na c e n d
. 2 . AD M
7H ,M&
RE. MH E L e t Os f ahf nc eo tF s i a L eT h e ro r e mr , s m t a t
B Ad M
. , S) 1m4 49 1 ( a. 7 op , 1e t c3 r h 9. . . - , 1 4 2
NO I T C E S 3 6
s .o 2, . RD EM 8 Ha E eL. c h eA nu t f x o L ft . n n jA e u o n n d c e t d
M, ) 0 a 3 t39 1h.( 8. .14 ,4 9p1, 4
m a Fi t o l h rr Me pr e eh t ar u f n r dr ss ee e e f n e n r e e
L 2 A 9 . HwT eJ RN~ I T Z , M . hp te a M r Cr a m1 p i . 6s, , m e e n s n
(. p 2 1 4 9 9 -6 2 2 5 ) 1 , .
E. s 4 3 X E R C I S E
u F a osn ro W C a s o er e ni nj g e c s t
e . 3 . HZ 0 H O . h Aa dS d i Zt i v e T z tM , Sne pi rl wir&ne BgA e r , eN N i , t
1 .1 p 8 9 5 6 ,
t x F e o o r t a l g e b nr a i c a ut n h d em p c s b
r e a e oe s rb i sl e
r . e,3.L EnCGE 1LIe S .G f om eW l b o r p o a t a- l mg eu rb nr a i c i n g
. b r f ee m Ar iJ M, ) 4a 4. t69 1he( . .6 , p , 1 l 2 d 2 s - , 1 3
n . 3,s. T. HEG2M IoRW.a r d eAu W nn o p r ooab l eJ m , L r M i a n t g h s
, ,)S4 3 991 ( . po 2 6c 7 . - , 2 7 2 .
n 3i R O 3i S E. M .s A M R I rSE Te E haM T
M LaE R , e m e W l b o r p n i a l g e b r a i c
u n . a f, t .c A h & , Ai ) 1 6 961 ( . e , p 4 l 4 d 7 s - , 4 6 8
l u F ,cLso u o mia r f if e tf u nb i ( l d n pf e o i H i l rb m e r tk s o e o sn f t
+ ) 1( 1T 2 s r peeh f e re 1e. nC Ic 0eIea I , ho g ra e e p m .
E. s 7 3 X E R C ~ ~ E
e , cS l e . e e 4 F no i ataozri l aren reg f o C , 7 1o 1 1 2 , n, j e
e s e te c . ,
L Ui 3AG NP .A Mt E TTA B oO t & R .p Am H A y h se uOr ias - t ir c o fR N ,
Q
an l uo m c t dh f e i o p s . n hu mtt b aer rMs r, C a i m i p . b, mu t e i o n
, ) 2 6 91 1 ( . 6 p, 3 6 3 - 3 6 7 .
4 2 R e f e r e n c
5R E T 3 E P H A r G eJ wdI RS
on , lu. o A , b f t o s h ro e oe P
t fd d e r f
57m u9 . n & -9- , & & 11 Comp., 2 5(
7 p 9 p .
6 3 B R Y TA N T s U i g oa H RC fU M
5 3&T K , ipe ld p , . WE v 5 R i- M e
0 1
7s 3n M B eU .~ aT O t SnNE LdxM . O ~ erA, e w ? oo l fb o u f n d os r
o Pd dn u m b eeA r s , . rNottieq e. brmh A,t .a M c os fS 2 3 t e, r c a t
. (5 5 -p A .1 9 7 6 ) ,
r 8R E e T 3 E Pk H Ac G J I RSu , . o, T Rf U 5M M 2e T , Comp., uv t i h e
& 1 0 2 ( 07 p, 1 p .
d9 3
e B R c Y TA N o T rU s e Aa p r r c eCh w d a on ul n o bKd f o E r R M
b m o uPd dn e& p d 9 p, r . 4 f - e 9 c 4 t
0 4 J BA N t O On h un m eob p e r r H nif l me etev sihsa g as M nl Ai Nm , i
9 1 BIT, 12, ( p 5 7p 6 - .5 7 7 . 59,
1 4 ,R T N ~P B t s I .r i Tf h o e eCg r ao l c H
gf ab p c s A e u R t r D w r
3 7 s 9p r iu m M e s 1, Comp., c 27,- ( p c 9 5pd 9 - .9 6 3e . z s 60. s i
2 4 D A NS I E L mO H n g a b ap A s s e xp r Niu m e s , t i cK w mc S e ea ,
& 1 d ( 8, p , 6 p4 6 1- .6 5 1 . 9 6 , 1 4 6 ) ,
3 4 R K G Ia .u J nyL t S d a. Ch. W E d rH i a v a L e sn A Fl R Ri D I q
s e & q 2 (& 9u p , e1 p0 1 1-n.1 0 7 c. 9e ?, , 27 6 5 )
4 4 D A N S I HE AL aN KSnS I Dd K N RE AYV I T ZtO, dhins t r ieb u t ioo n f
M e r s e dn n e & 2i d ( 1 , p , 9 vp 1 . i7 9 s - 6 o 1 7 r 0
5 4 H C W I. L .L I aA MC~ne R mZdoe. Sm. i r pA n o Rt uhf e N 63 m K b E
s m r o f 2 A + 13 a m2 n A - d13 z m, 2& ( 6& p , 9 9p 5 1- .9 9 8 . 9 7 2
r 6 4r R I P B R I .E N T , Ci t d hni s t r iebHu t oi o pn r if mA e s 64 R D
a t n wp dr ii mn & e s ,i2 (d 9 , p , 4 p3 1- .5 6 . 9 7 5 )
7 4 D A NS I E Lt n e Hro B , RdfU , & M
4 A i ,T p e 3 p 3 N .1 .v K 65 i S e ,
8 4 D A NS I E L t n e Hor B RfU M2A 1&T ,e3X ( 0 ,N p, v .1 K i 9 S e 7 ,
3 7 9 . 66
9 , n 4 u Vr e B i r . s L 1 a + /1 +5 / l /7+ 1l / l+ 11 / 3+ 11 / 7+ 1 9
1 / + 12 / 9+ 13 / 1+ 14 / 1+ 14 / 3+ 15 / 9+ .6. . ,1 o les d - 67,
n o s m o n no im t bp r ren si jmu am e es a eut xc s e ot ou n ru v s e r g e n
f ,Bd. SA.3 Ma%., i 4 ( np 1 p0 1 1- .11 0 i 4 , 9 2 e 1 & , 9 1
0 5 D A N S I HE AL aN KJ nS O W d HW N . J R R. c, B E o r Nn u Cs
9 M 1 Comp., 28,z ( p B pt . 3h .- 2 9
1 5 H AR N I E Sa S, EL Gn LHU
e m N
~d N G
o AS R c a l c u l a t i or n s t oe l a
l e u i m
e Rm i rr p on u m f b e r & 2 d ( 4, p , 9 p6 9 1- .9 8 3 . 9 7
2 5 R P B RI .E N T d ,i hsl t rCileb ua t iomo ns f gH a b p s A e R t D w
s p r iu m & e s ,2 X ( 8c , p , 3 1p 5 1-c .3 2 4 . e9 s 7 s 4 i
p 3 p 5 uD A NSS I E L H d aa A tr n a d c Ne a o K
m n Sa c ,r
a H c o n j e c t u r&e , 1d ( 7, p , 1 p 1 . 8 9 % 6 1 3
1
4 5 M C W . . U N t O G hDan u se spEi ra in) oR mz t e( lsmhnL iI en I = e C H ,
1 & , 2 d ( 7 , p , 3 9p 9 1- .4 0 0 . 9 7 3 )
R e 249 f e r e n
a
AI 5 H DH.
,c
. Ca
. . o h f t n uJ Ms
An S d . n oA
o ct en a r i t h m ee t i c a l
l 1 t ( 0 ap , 1 Zp8n 1 1- .1 8t 5 . & s 9 ,
r
& 6
t
. 6
a
s )
5 D6 A N. S I E L C a l c uHl a t iao n an A
dp o E p f lNz e i pt ac K a s t S i t o , n e
, . f . h pt u, um M o n9 C c 2 ( t p i 2 7p 1 1o- .2 8 7n . s9 , 7 5 )
A 5 m HD7 . CH o OIa c I n ERA d nUN tO
N i h nG e So L f L A E S Y
o t c w p oo c mn . ~pj royoi rme, P e s ., Sc h n t t P a M u cu r re e se r n
4 2 ( p 1 p 1 . S a e 2l es 9 o % 7 1 3 2 ) 7 , .
5 7 J Ba ~ . K. LR EO YS a S ELnRO Wd SE L L C HS h a rbOp oe ru En d s N F E L D ,
e c . hf n t o Chuhr e f t O, (au. X#np (,) m xMd )a , 9C 2
( p %p 1 . 3 9 - 7 2 5 6 ) 9 , .
a t N 5nI eM E8 m TL S.i . r .a e Pnp R xSd E .T .E I N , r o a d dnei t i v e s u l
- 2 & 1 d ( 9, p , 4 2p 7 1- .4 3 4 . 9 6 5 ) ,
5 D9 A N. S I E Ln i e Hotn Si eRaf t Sn A Ud eM
3 6&T N, ip d pv , . K i S e , w
3 3 2 - 3 3 4 .
L 6 RJ 0 BA .OA HN M a ACnN d F R ~ B E R G , r N o t hne u e sm eu r l
7 G o 9l d bc, ao nc jhe c1t5u rBe , 1 ( I p ~T p . , 9 - 2 4
o 6 H1 H h . . at A Hn e - R vd IEeeC LHi . ES R T ,M B A c a d e Em i c R S T
P . r 4eL 7s s9 ,1 o n d o n ,
, rN 6 E H J 2p C . .e tO r h n ee g r o a a l f e vi e na t .hns e t e g
e sm i ou r a pm f a t n ph d e o a m rf ot s pw rt i mo oeS s , & d u c
, S &1 t (i u6 , p , 1 p5 7 1- .1 7 6 . 9 7 3 ) ,
l6 H3 A o .P O . GOG
. RZELSK I, C o n,j e c t u r e.C, y e J& 2 s 9 2 ,
, ( 2 1 p- 1 p .
C6 A4 S . . rr ee A mp o aa a rp k n o . B h f t a zHn oca rM dn , t e m
m o C 1 ( 7 p , 4 p4 5 1- .4 4 7 . 9 6 3 ) ,
6 . D5 A N. S rI E L l u H dAo woo p r i A fem eJ s , R nN e sK c iS r te
,
, M 5 ( p 2 7p 2 1- .2 7 5 . 9 7 1 ) ,
6 D6 A N. S I HE AL aN KMnS A R d W~ I N U Nl A
doD w oE Ref L I n C H s ,
m r ai p ur ikram te nescP h d e apT P p e( uT O s t i n
6 J 7 O C. HHAL~YBURTON,
N . J a R Jn . O B dH N RT I w o I L L H
b h5 fm t o7 uF e , fra unm. 9 a pt , mM oc19 C 2 ( t p 1p o. r s
1 1 2 .
6L G8 M L A. ~ . Ha IE H ~n Ce W md .o .S n pe r i wom t e fs hf o er m
s 6
rD9 r E uS. H
oe. IF.P P Eb E n, fme wo uF e rfa nm a t & c 3& 2& , t o r
( 1 9 7 8 ) .
s7 D0 H kL. E. H. aM dEE nR M Ld M u A HEt a n ney o nS eu H
, i r i M s t Ei c R
s
ua Mi nl d pAun e u o tR T h . Sv t ai n n f U eo r d m a t i
a t P r, eS2 s 6 s 9 , 1 p 2 0p 2 - .2 1 0 .
c 7 Ai1 E W. . d . a J nCEn
ePd v
dM. i . t. nI,i mSs i u Irt P o o T R L E L R E N R
. C v i n U a P r .eC s8 s6 m,9 1 a t m , b r r i i d d
\
250 R e f e r e n c e s
R H 72. C H AO O nc o n, j e c t ru rO
.C
e , r e l JZ e2 t s L 2 i 5E , n Y
. 0 2 2 - ( 9 0p 2 p 1 . 9 6 7 ) ,
73. D A NS I E L Ho B Xf U AM
5 a1MT e, U & N i v K l i S l e ,
t . i t A N S Ii E rL r
74. D
m . H anp Fo b r uiA F i Qo N 1b 0 o , Ko t nS s a ,
( p 1 p 1 .1 8 61 .9 % 7 1 2 6 ) 8 ,
75, H W L .E N. S TJ R A R
, .AO , nc r a- oEngu cl dl ai td ns i j ne c
e m h t vi r i ,gn l . onf bih e,a l tIld s u ,2 M 4 ( 1 p 29 0p 1 7- .2 27 4 . ) ,
76. D A, NS I E L . Ho B p Xf U. %mM 6Am , M.a T, e C0 v 3i
N v K l i S l e ,
( 1p 19 p8 9 7- .1 96 0 . ) ,
i t 77. Dc A NS I u E L Gd a u Hts s e s f o r rA e m a t n 2 hrNd e - nK S aS y r,
s & u 2. ( 5& p , b8 p3 7 1- .8 5 3 g. 9 r 7 o 1 u ) p
a z i D A r N S I Eo L C t l nHa sc s a t ah ueAoo f r y f
78. mNa n d bK eS r,
0 g7 e n e ,hr a .,o9S yhc m, q.t xP xu1i0uuM r 2e ( p 4 1p 5 - ,4 4 0 .
e i d 79.t eD A N cDS I E Ln uH z ef t aA t h h a a s t vNu oe m s f K S ,
r c o u na t . E t uee c z f pa p a f p s s( uT O t e i n
80. D A NS I r E L eH o s H p aFfdu Sl n r l Af X dk r Gkh e eo Q N v n K i i S e o ,
n P. (h St 6 a M )C o m1 ,p ( . 9 , p , 3 p3 5 1- .3 3 7 . 9 6 5
s e r , ek81.
h Cit F Gt n .A
i e.UU Sm S ,h t t i H rb A e C r h e l
r e, d ki t.rNie5o n6Ye , 9 1w p r i n t
y 82.r G ,B o sM eA
r. fTeh
. H bETWn Sr , u o N C r e hd i t Nie o ne , ew p l r s i
, k .r 1o 6Y 9 1
m 83. D Ae NvuS I i E F L nH a.Al g~o roi t hr m Ps , NS e c - K S ,
n e mn u p om o Mnin du, Cn . o hNn t a iM W ( p p 1 . 9 7
5 1 - 7 0 .
i c u i q t84. msC F hGi A . t .UD S i S , r A . v iY n aU l P e r eN s se , w
. 6 H6 a 9 v 1 e n ,
8 4 D aA N. S.I E L h H o t t Rtfh Aer u a ae MnN bs v l K ao i t S i v e o ,
. 9 p , m a10 C 2 ( p 6 p1 7 - .6 1 8 .
85. D H L . . T a eEb r l r 4e a ft 8 aP oH P5 o ur . l & e t , 2d M ( 5, , 1E 9R 7,
p 9 4p 4 - .9 4 5 .
86. J D S . . T a e bIr l r 4e a ft 9 aP oP7Vo ur . l & e t , 2 d ( I 6, p, .1 F 9T 7,
8 1 4 .
r AL p TE oJ L SSd . . Gu a
87. SA
ne -d Ws L P J R . ,
r ps o d r u es t es r po n b g ma Cn u d n a- &p ( i r uTd O m i c
) p . e u r
. Z . r h P hE R eD
88. . t ~ taSO l, mmnnp or iu s mt Ae so , , M M7 5 ( 1 9 5 0 ) ,
p 4 0p 4 - .4 0 7 .
, , s &89. D RP rE . e pOs e ub dno p r i mmea s Cn u d n a A &r . m i c
u M , L 4l ( 1 p e 29 0p 1 5- .2b 06 6 . ) r , e c e n ,
n i 90. aA S C Ht. I Na Z EWrnL S I dE .~..e & K I S,c u rh c y o p no tc
9 J1 B . . J T or n . a sa c i Pan W i d . E l I tO N Fh ne r leBm a t E hR G aE R r ,
q Computers
u in Number o Thory, t A c a d e m P iir c e s s , e n t
, 1L 7 9 1 p o2 1p 3 - .2 2 2 . n d o n ,
M 9 AM2 O R. NIR I S L D A NESLI HE AL aN I KHnS~C W d . . M S A i N m ,
g r ro oeu sdp q sr uf o a r a e a n i d n n s e tq u eo ne pc re i fm r e s .e s t i n g
Acta A&h. (TO uppeur.)
93. J O B H N D RH L . E H . M a I E JnR L S d . L. EN L e w L H F A R R
o r t p c c r ai t ae r f in a d o p ? 1f Math. , Cbmp.,

( p 6 p2 0 1- .6 4 7 . 9 7 5 ) ,
tr 9 eiH4 C W.t Il. L .L, I eaD A aMDJ SnUSD J md . . i r o t p hf e
g n o iN sb uf yf o N ? 1fa 2& , i3 (d 0 , c p , 1 p5 7 1- .1 7t 2 . 9 o 7 r 6 s
9 H5 C W . .I L . L I, aADMD JeSn UmS Jod . S . a e m i f r lp o r g o r i
g n ti s u g e e L en h sfm ee r u r& n it3 a (d 0c , l p , t ipi 1 i . z o e n n9 d s g , 7
8 6 7 - 8 8 6 .
t 9 5i H aC W. l I. L .L I aAaeMRm~n Ho m OSd .L T iE o, b s e r rv a t i oo n sp n
e t & i3 (d 2 , , 1 9 7 8 ) .
9 G6 AL. M R Y. I h sR t s ae L yt n i f d p o pL r e i omEm a t l aR i h tn , y
Jour. Computer and System SC&,1 ( 3 p, 3 p 1 . 0 9 & 7 3 6 1
u 9n qo D7 iH Lt. iE. aH. azM nEE i nR M r LdohM t A c c na fEAe e w t H M E R
n i su u q f Muth. Comp., o 28, ( pr 6 p 1 . m % 9 s ~ 7 , 6 4 3
R Ns 9 E H8 s .B M e A a. N Dcn NA NdS I e E L n HA a sn dA u N f K f Si
o c f po r a ri n s tod u r s cJour.ie , Combinutoral m a l i
Theory, 13, ( p 1 p3 1 1- .1 3 4 . 9 7 2 ) ,
9 , D9 sH cL . . i . t a mA E e ha u t pn a ut m Hd r oe mAppZica-
M a t E i o Rn ,
tiens of Digitul , Computers,
n n i G, 3 B 6 9 1 p 2o 1p 9 - .2 3 1 . s t o n
r1 0 C eL 0 S. . m . z Zm w I B u e u ei Km
E e Nachr. r G Ahxd. k u E n g Le n ,
Miss. GcWingen, ( N 6, op 15 .p1 - .5 7 . 9 6 4 ) ,
u 1G g 0 S A 1A M e . SU WE Lr . J r R T
. i ,h e p r i tm 1e 2s 5o 0 0 0 ,
Muth. Comp., 32, (1978).
102. N O R M G G A N . Derivution
U of Criteria
N for the FirstD C u s eE R S O
of Ferwzats Last Thecrrim und the Combinatmof these Criteria
to Produce a New Lcwer Bound fw the Exponent, T C h o e r s n i
. U t . p 8ne 4S 9 i 1 v e r s i t y ,
1 e 0 L J3 t L . A. N . n aD ETnR u R Pd . o . c AA t ER o u K l I e N
s ou p om w f .c oe hn rj etsc t u rM e , C o m2 p ( . 1 , p , 1 p0 1 1- .1 0 3 . 9 6 7
1 0 L J4 L . . . T R P AA . R . aK IJ nNL SNd . . E s D u Ar ov e yLf E F R R , I
a u q e s uo m l ips fo kw e eibid,r s , p 4 4p 6 - .4 5 9 .
L L1 E0 W 5 J. O H N SO N , p r I i am ec n s rd i n vya r i a nr t s , c e l g o u t
ibid, 29, (1975), p 1 p1 3 - .1 2 0 .
1 0 D 6A N.S I E Ll h e Hoi D Raf J n oA r dUd ae M 1n & 9 TN ,i d v , K i S e , w
( p 2: 6po0 1s- .2Sl 6 2a e. e 9 6 7 ) ,
6s i 0 1 a t D A er NsaS I J Ef T L r n o H t d i hf ae A ou a r feq a N l Kl S ,
252 References
p d ml e i of i c n p = A2 + B2. (TO appear.)

1 0 J 7Oe B .tH o NN R o r eep rmne s ie In tr ianpg a aL s sou t mwL f o H A
u q . sh t a M Comp., 2 ( 6 p, 1 p 1 . 0 9 1 7 1 2 -
s T n 1 I 0W e 8C M.m. . i Tn udh m eob l-e r fp ?o ii naa/ t sn t t i
s r e p y h & 2z ( 0& p , 3 O p C1 - . 3 1 0 . 9 6 6 )
1 0 D 9A NS . I E L Ts e cHho n de - o r td e r e i Atr amhn e e N xs p a n - Ky mS p ,
s io B(x),
o nfibid, 1 ( 8 p , 7 p5 1- .8 6 . 9 6 4 )
ln o i 1p t 1 Ha m 0Mn . Si. lom.X rR eKc , t A e d o t cfh e o m p
a u q f i eo l c lda s,ssf - ne u nm boe r . hl &t c ha @ M a n J., 1 4 ,
. ( 7 2 p- 1 p 1 . 9 6 7 ) ,
i a1 1 A r1B As. .K mbEr Ro , f e i L tg l nh l e io oa f n n ug m - ea ra i
b e rM s , 1a ( 3 p , 2 pt 1 . 0h 9 e& 6 m 2 6 t 1 )
p o 1 i1 K 2H D E. E. G N E R , A n a l yu s i M sn d o d u l
Math. Zeit., 56, ( p 2 2p 7 1- .2 5 3 . 9 5 2 )
1 1 D 3A N. S I E L H o , RT R Myf U g Ag Math. e Gwnp.,N(TO v K i S e ,
appear.)
1 1 H G4 U . PMT S A AC ,T. H. A E M H. a 0 n P Gd U . E .Ph ?Te pAr e, s e n tM u- M A
tien of Primes by Quadratic l a Forms, y o. Rh t S a M o c T . a bN l e o s .
m a 5. C v i , n U P r . eN 0s 6Y se 9, o1w r k
1 1 D 5A N.S I E L Ho p r R e fv i t o Au s Math. e Ccrnp.,
a N 15, v b K i l S e e ,
. ( 4 8 p% 8 p 1 . 9 6 1 ) ,
1 1D A 6 NY .S I HRE ARLaN A KLn S P
d S . VC o aa t n r H i h a M t e i Io on
o L f P a1 & ar , ti2 (d 0 , p n, 5 5p 1 1- .5 6 9 d. 9 a 6 u 6 .
1 1 D 7A N.S I E L H o F oR fr ab eALns T ad e 1& N 2& a 5&v , K b i S l e ,
( p 6 3 .1 0 . 9 7 1 ) ,
1 1 D 8A N.S I E L Ho K l RfoN se sw A , ma a O nen r d d m UN a nM v T K i S e ,
1 &
0 ,2 d ( 3, p , 2 1p 3 1- .2 1 4 . 9 6 9 )
1 1 M 9D H. .E .N D Y T d ,i lsh t raiebseu tsdoi oani l cf n o r eu f a l m b
a3 u1 q1 %f i e l &d2 s i2, 4( 19 & p , 1 p 1 . 9 7 5 )
1 2 D 0A N.S I E L s u H Ar ocv q ei yb f u cA u-a e q gn l dN a a u K
d a rS r a ,
l ae ni oa ri b t nam uot mruf b fipe e(rm l oda cs t n i o p o v f & O iC . e w
Seventh Southeastern Conf. on Comtinatis, Graph Theory and
i u Computing,
o e Lt a t nS o Ut na iB v . R, o u ( g e p, 1 p 15 . - 9 7
2R 1 1 . P L I. Ao KC u Rfr Uo d K 9a HMath.
M , e Comp.,
T E A vj I R i N D e
29, (1975), p 3 3p 5 - .3 3 6 .
i a t 1 r 2aD A 2 f NS
. rI En L d Ti Hh a e ou a rAd feql e a i l f N a n d1 K S ,
p p i a t s Proc. 1972 Number Thwy Conf., B o u l C d e ro , l -
o r a( d o p, 2 1p 7 1- .2 2 4 . 9 7 2 ) ,
o L i 1 I 2 H tC
3 W. . u. R A l G E. R. aoM ACnNR SZ d . . Ao t hf eR N K
c a t p t l e o A r c rhfi m e d e sMath.
, oComp., 1 ( 9 b p, p 1 .l 9 e 6 m
6 7 1 - 6 7 4 .
g 1 2D e 4A N.S Ir E L ob v Hio ef w f , 4U A 1 M 1 &T pZ N@ p, 6 . - K S ,
6 8 7 .
1 2 D 5A N.S I E L SH e y x s A o aL i t t l femw oe o bd N ois mu n n d a s aK t t i i cS o n ,
o L x( n Pro~.
l) , Symposiu
Pwe Muth., 24, (1973).
E iM1r A 2 D l6e .S l d in H o We R f HA ea W n N r id g vh t K i S e , w
U M1 1 T , UComp.,
M &2 ( 8 p , 3 3p 3 1- .3 3 4 . 9 7 4 )
E M1 A 2D 7 .S Ho B Rf A
W ia l l ieaa m Bsn Nar nd i vu k rK i u S e c , w
U 2M 0& T , 3 Z ( 0 , p , 3 7p 7 1- .3 7 9 . 9 7 6 )
L 1 I2 H 8C W . . . C c ei r btp a u i unc r f e i e w l di c l sat s s -hn u m b e r
, e n o & 3 d ( 1, p , 5 7p 8 1- .5 8 0 . 9 7 7 ) ,
1 2 R I9 C . HB AL ~ . C o m p uAlt a tai oosne tsd pai hful Koc er g o E f I N ,
c c qe fo I ur & I ,2zim ( 9& pat , 1 p3e p7 1- .1 4 a4r . ll 9 ti de 7 in sx 5 c
1 3 J 0O W. HW N . J a R D Rn A. NdS I E L E H c Q N Ao u C nN e H c Ks , e tS
c K h i n t c h i n fec os n s tfa a nt nt e h d e c o o mr f p ue t a g t i uo n
o c f r & ai2 (d 0c , p , 4t 4p 4 1- i.4 4 8 o. 9n s ,6 6 )
u1 3 R n1 . P iB R I E . CNt om T p,u t an t iCoo n t o r hf ec H ef r Aa c g R u D l
t if oE o n cr u& oi3 (d 1 , pl , 7n p7 1 1- .7 7 7e .s 9 rt 7 a 7 n s )
1n 3 J 2Oi W. HW Nt . J Rn R. o, o G o Rfs C pE e r F r ea c Ntf i o on rv C i H e ,
7 i U nM
1 &
5 T ,i3. (4d 14 , 0p , 1 .1 9 7 7 )
1 3M 3 . A M O RI . R I aS OJ Nn O B C dH N m R eHAt oh f o d I f A a L E cL L H
t
i r o it ac n at n gf h d e o F T&f , 2 d ( 9, p , 1 p 1 . 8 9 % 7 2 5
ea 1 u3nD 4A N.nA S I E L i H ta i n n dm A o o p t c hfr N e o v Ke m Se n t,
c f r a ca m t i eo tn oh fo d f . rNotices, .e m h A,t .a M c o S 2 2 ,
( .A 8b s6 tr ro-1ea Acp t anA p w
9e lit t st hi ha t em 7 w e ai l l 5 p ) p , e
t1 3 Ds 5A NS . aI E L r H
a pf p l i c R a nt i oo nt s Ai hf ee oN r e fc a l K e S n , t
u q f i eQ l d (.) s rN n,.eo m &
h Ae,t s.a ,M c o S 2 ( 3 , 1 9 7
.A b9 s 5 t ro- a Ac t n
r 1 o3 D 6A Nt.S I E Lc S q u a rHe - f o rfm A a s i m Op Nl e ( K N S 1 ,
a l ( TOgctppear.)o r i t h m .
1 3 R 7S . . L He g r Ea l EF i H M aR n UComp.,&M c M t At e A oN g N r ,e
2 ( 8 p , 6 p3 7 1- .6 4 6 . 9 7 4 ) ,
1o 3 J M 8 Pe .tO. L t . Lon A loR M D
cr , aA C am e t fh of od r BIT,
1 ( 5 p , 3 3p 1 1- .3 3 4 . 9 7 5 ) ,
ot 1 t s3 J C 9 cP .e . M . a . g fO g i n uw i a lst h n lae pw - e r ,
p r o a Math. c h , Camp., 29, ( p 1 p5 5 1- .1 7 2 . 9 7 5 )
1 4R 0 . K G I U. YH t, o f w Ca oc a t n ou rm b He Proc. r , Fi$th A R D
p Manitoba
i nCon$ on Numericaln i Math., W ( p 4 p9 -1 .8 9 . 9 7 6
i 1 4D l A1 N.S I aE L s smUoa H ossl ucie nf n pe e gr e A i r p n tN eK s S t ,
( ctppear.)
T O
1 4D O 2 N . BA GL I DL .L ~ TE Sh n, r eMe ee r w s e pn rn ei a m ae ns s d t a t i s t i c
t h e o Muth. r y , Comp., 18, ( p 9 p3 1- .9 5 . 9 6 4 )
2 References
5 4
1 B R 4Y A T N 3 T U. T 2 h 4 Me et rC .sh e~pn on e r PK r V &E . i R m M

7 9 Acad. Sci.
1 . USA, 68, p( p 2 3 1 9 - 2 3 2
1 J O4 R HE N 4. T.enHmhu ime orb ep r fRd i v i so o cr sM f e A r N t
b m Mu e r s e nn n e M &Comp., / z2 ( 1 p , 7 p0 0 1- .7 0 4 . 9 6 7
1 D A4 N S I E 5 L .Ho K R f a AMn r d, e U 3 1 M N
1 a aT v Kd v i Sa i e ,c
& 1Z ( 9, p , 1 1C . 1 3 o; & r 2 X9 ( 7r , p , 4i 5 6. 1g3 . e 5n9 d a)7 , ,3
- 1 D A4 N S I E 6
N A h t u.H on K
L T R f R eA p So r e t N v K i S e ,
& 2 (& 5 p , 9 3p 7 1- .9 3 8 . 9 7 1 ) ,
~ A W1 O H , s4 ~ W v E 7 In . Mutkm,uticul
. CircZes, Quadrants 1 und II,
r eP rb i en W d l e ,a Sn c h, m d9B i 6d t 9, 1 S o5 e 5 e 65 ,O o 7s , O n t o
p 3 p6 - .3 7 .
1 D A4 N S I E 8 L .H o M cR f N Ri A se hp o o Fre t n uN nv d K a i m S e e n ,
6 C 9 o, Muth. 1n 8 Camp., s 1 ( t p a3 .n 3 t 5 s ,.
1 D A4 NS I E 9 L d H.A r e l a it u ie oxAnt ne br ai l ol i ras t i SIAM
cNs , l K i S t ,
9 1 Rev., 6, ( p 5 p4 - .5 6 .
1, S DK A5 ~ NS I E 0 L .o Y a RftPrime e s Period
, Lagths,
e Muth. v i e
Comp., 29, ( p 1 p 1 . 1 9 6 7 2 5 - ) 1
S M O 1 E R 5T M 1 S l . a .e d T i h We a r t-i no gp fx e o h r e o
0 0s 2t n- e 1 n 0 4 & 1 d ( 8, p , 1 p 1 . 4 9 - 6 1 4 4
1 K M5 . 2 t Os A f t h n.r ae p r oH t p hf o waeo e aLrr as ft i o cn a E l t R i , o
n u m ( b I e Muthemutiku,
Ir ) , 4, (1957), p 1 p2 2 - .1 2 4 .
i r 1 , aJ NI NE 5 GHW C U 3N
- R . p f g o,(r 5Sr ) 3 ?Sintiu,
, i 1
.o b l e
1 ( p 1 p . 5 4 7 - 1 5
t i1 Wd5 J L E 4 I. ( . E Reviews L in Number
V. r Theory,
Le m A E I Q A U M
. h ,t .a . Mc 4 o 7S 9 1
INDEX
Abeban, 61 Characteristic factor, 94, 102

Abelian group, 85, 107, 109, 225 Chebyshev, 16, 17, 145
Abstract group, 74,85 Chen, 221,238
Additive number theory, 144 Chinese Bemainder Theorem, 2.04
Algebraic integers, 151, 152,208 Circular Parity Switch, 76
Algebraic numbers, 212 Class number, 153,168
Algebraic number field, 151, 170 Clipped autocorrelation, 108
Aliquot sequences, 218 Closed, 60
Almost ah, 117 Composite, 3
Archimedes, 29, 140, 188 Congruent, 55
Arithmetic mean, 122 Congruential equations, 66
Arithmetic progressions, 104 Conjecture, 2
Artin, 81 Continued fraction, 11, 173, 180
Artins Conjecture, 31, 80, 147, 222 Converse of Fermats Theorem, 115
Artins constant, 82 Cubic residues, 170
Associative, 60 Cunningham, 81,223
Associated, 150
Asymptotic, 15
Bachet, 143, 144 Cyclic group, 73, 86, 225

Bachets problem, 151
Baillie, 224 Davenport, 220
Bateman-Horn Conjecture, 220, 222 Deep theorem, 64
Beeger, 116 De la Vallee Poussin, 16, 22
Bernoulli number, 152 Descartes, 2
Bilharz, 147 Descartes Perfect Number Theorem, 12
Binary quadratic forms, 143, 165, 233 Dickson, 8, 158, 211
Binomial congruence, 110 Diophantus, 159
Biquadratic residues, 170 Dionhantuss formula. 169
Bohman, 217
Brent, 217,219 Direct product, 94
Brancker, 15 Dirichlet, 16, 22, 35, 51, 152
Brihhart, 230 Dirichlet Box Principle, 161
Brouncker, 185,236 Dirichlet Series, 49
Bruns constant, 219 Dirichlets Theorem. 22. 105
Carmichael number, 116, 118, 226, 229 Division Algorithm, 10, 239
Catalan, 196
Cataldi, 1, 14, 21,25 Eisenstein, 29, 65, 170
Cataldi-Fermat, 3 Elastic sauare membrane. 169
Cauchy, 152
Chandler, 212 Equivalence relation, 56
255
2 5 6 x e d n I
Erdos, 227, 229 Group, 59, 60

Euclid, 3, 4, 5, 6, 13, 25 Gunderson, 232
Euciids Algorithm, 8 Guy, 237
Euclids Elements, 125, 128
Eudoxus, 123 Hadamard, 16
Euler, 3, 12, 21, 25, 28, 42, 69, 76, 195 Hagis, 217
Etrier u-pseudoprime, 226 Hardy-Littlewood, 48, 221
Eulers Conjecture, 157, 233 Harmonie mean, 122
Eulers Converse, 8 Hasse, 81
Eulers Criterion, 33, 35, 41, 47, 62, 67, Hensley, 221
68, 74, 103, 192 Hilbert, 211
Eulers Generalization, 23 Hooley, 224
Euiers Identity, 209 Hurwitz, 196
Eulers phi function, 20, 68
Identity, 60
Factor generators, 98, 100 Index, 110
Felkei, 15 Index of irregularity, 231
Fermat, 19,25, 142, 143, 185 Infinite descent, 147
Fermatian, 115 Inverse, 60
Fermat numbers, 13,78,80,115,119,222 Inversion Formula, 71
Fermats Equation, 167, 171, 179,235 Irrational, 138
Fermats Last Theorem, 31,144,153,231 Irrational number, 126
Fermats Theorem, 1,20, 62, 78, 191,203 Irregular primes, 153
Fibonacci numbers, 198 Isomorphic, 74, 84, 93
Fibonacci primitive root, 224 Ivory, 24
Field, 67 Jacobi, 165
Finite groups, 61 Jacobi Symbol, 44
Four group, 84 Jordan Curve, 207
Frenicle, 20, 143, 185, 236 Jordan Curve Theorem, 207
Frobenius, 4.5
Fundamental Theorem, 150
Kraitchik, 15
Fundamental Theorem of Arithmetic, 6 Kronecker, 126
Fundamental unit, 236 k-tupie Prime Conjecture, 221
Kummer, 152
Gauss, 16, 17, 29, 42, 51, 70, 72, 92, 165, Kummers Theorem, 151
203,204,205,226
Gausss Criterion, 38, 39, 41 Lagrange, 37, 144, 175, 185
Gaussian integers, 149, 213 Lagranges Four-Square Theorem, 209
Gausss Lemma, 46, 65 Lagranges Theorem, 86,204
Gaussian prime, 150 Lame, 152
Gauss Sums, 49 Lander, 233
Gear ratios, 173 Legendre, 15, 16, 42
General Divisibility Theorem, 104 Legendre Symbol, 33
Generator, 73 original, 40,41
Geometric mean, 124 Lehman, 237
Geometric number theory, 161 Lehmer, D. H., 15, 117, 157, 193, 214,
Germain, 154, 155 223,230,231,232
Girard, 142 Leibnitz, 14, 163
Gnomon, 123 Lenstra, 225
Goidbach Conjecture, 30, 221,222 Liouviile, 212
Greatest common divisor, 5 Loba1 patterns, 84
hiex 257
Lucas, 120,188 Period, 179

Lucas Criterion, 40, l20, 193, 194, 195 Periodic, 41
Lucass Converse of Fermats Theorem, Periodic decimal, 53, 72
200 Pillai, 211
Lucas-Lehmer criteria, 195 Pogorzelski, 222
Lucas Sequences, 214,237 Pollard, 237
Lucas test, 198 Pontryagin, 61
Poulet, 117,226
Prime modulus, 66,71
Mahler, 2.38
Prime Number Theorem, 16, 197
Maximal gap, 218
Primes, 3, 13, 15, 104
Meissner, 116
Prime to, 5,60
Mersenne, 19, 195
Primitive roots, 71, 73, 74, 76, 78, 81, 92,
Mersenne composites, 29, 222
98, 102
Mersenne number, 14,18,25,80,120,195
Primitive triangles, 141
Mersenne primes, 22,29, 197,237
Proportion, l22
Miller, 237
Pseudoprimes, 226,227
Miramanoff, 157
Pythagoras, 122
Mobius Inversion Formula, 48, 71
Pythagoreanism, 130,237
Modula, 55
Pythagorean numbers, 1,121,141
Moduio multiplication group, 61
Pythagoreans, 127
Modulo vector addition, 111
Pythagorean Theorem, 123, 124, 126,
Modulus, 56
128,130
Multiplication table, 59, 61
Multiplicative, 41
Quadratfrei, 114
Quadratic fields, 153
Necessary and sufficient conditions for
Quadratic nonresidue, 63
primality, 170, 230 Quadratic Reciprocity Law, 34, 37, 47,
Neugebauer, 121
New Pythagoreanism, 130 64,204
Quadratic residues, 63,83,85,86,95, 143
Newton, 29 Quaternion group, 85
Newtons Algorithm, 188
Nicol, 153
Niven, 211 Rational approximations, 173
Nonunique factorization, 153 Reciprocals, 60, 66, 203
Reciprocity Law, 42,45,46,64
Regular polyhedron, 127
Octal biconditional, 108 Regulator, 236
Octic group, 85 Residue, 55
Odd Perfect numbers, 2,12,217 Residue classes, 51, 52, 55, 56, 59
Old Pythagoreanism, 130 Restricted Case, 154
One-lobed, 208 Richards, 221
Open question, 2 Riemann, 16, 71
Order (of element), 71 Riesel, 196
Order (of group), 61 Right triangles, 161
Robinson, 196
Parkin, 233 Rygg, 234
Partitions, 97
Pells Equation, 172 Scalar and vector indices, 109
Pepins Test, 119 Selfridge, 153, 227, 230
Pepins Theorem, 193 Shinze, 220,222
Perfect numbers, 1,25,53,196 Side and diagonal numbers, 139
258 Index
Siegel, 212, 231 Ullom, 206

Similar triangles, 123, 127 Unique factorization, 6, 138, 150, 152,
Sinkov, 206 168,234
Square numbers, 125 Unity, 150, 173
Square roots modulo m, 66 Universal starters, 214
Squaring the triangle, 144
Standard form, 6
Stark, 234 Vandiver, 153
St&, 221 Ve&r algebra, 169, 209
Stemmler, 231, 238
Stieltjes, 13
Strong u-pseudoprime, 227 Wagstaff, 226,227,231
Subgroup, 83,85 Wallis, 185
Sum of squares, 143, 159,233,235 Warings Conjecture, 210,238
SWAC, 153, 196 Warings Problem, 155, 212
Well-Ordering Principle, 149, 240
Thales, 123 Wheeler, 196
Theodorus, 138 Williams, 230
Theon, 139 Wilsons Theorem, 37, 38, 74, 104, 118,
Thues Theorem, 161,166 205
Triangular numbers, 125 Wieferich, 154, 157,211
Tuckerman, 217,237 Wieferich Square, 116,229
Turing, 196 Wright, 213
Twin primes, 30, 219 Wunderlich, 220

Number Theory UNSOLVED

Uploaded by

Copyright:

Available Formats

Number Theory UNSOLVED

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Number Theory UNSOLVED

Uploaded by

Copyright:

Available Formats

CONTENTS

professionals hardly know this material at all. The author feels an

I following economy has been adopted: if a paper has a good bibliogra-

Chapter I :: FROM PERFECT NUMBERS TO THE QUADRATIC

We must, therefore, a t least weaken these assertions, and we do so PI 496 111110000

f o r primes 41 < 42 < . . < qm and exponents b, > 0, then p ,

Similarly from (e) above, we have N = pip2 * . . p , +1

1 Theorem 9. (The Prime Number Theorem, conjectured by Legendre, Theorem g2.

II merely about the rafio

and n2 + n'.' log n - n2

I f(.) -d 2 ) and, if y = - y , and n is odd,

I and g(x> - h(z)

1 Now write these fractions in binary notation. Since 7 (decimal) = 111

A = .000100100100 . . . (shift right one place)

i 71M3** and 1671Ms3 Theorem 19. Eiiery divisor of M P ,

y122Q- 1, now ( I ~ Nsince

taken over all odd primes.

q)ZQ - 1-1 b = 1, 23. I b = 7, 17. I notational change introduced by Legendre.

(i) = +1 if there is an N such that qIN2 - a, and = - 1 if there is

are exactly two values of a, which satisfy the equation

These two values, a, and a t = q

2k f 1. This completes the proof of Theorem 18, and therefore also of

(N21d ( N l q ) ( N I q ) = +1 which we can evaluate by Eqs. (73), (74) or (75).

(1Id = 1. (74) (a%d = ( a ~ q ) s(if d N ) (77)

( p l q ) for any two odd primes p and q. ( P l d = (-l)Q(nlP)R

where the constant ha i s given by the infinite product -4 n

But to evaluate such slowly convergent infinite products we will need

b=b (moda). since 10 = 3. But \

Symmetric. (88b) 3 = 2 and thus lo6 = 36 = 23 = 1 (mod 7).

N2= 2 (mod q ) ab = c (modm).

If ( a , m) = 1 and a = a, (mod m),we have (al , m ) = 1. Thus we that

Theorem 17 says that if EXERCISE

xu, = a (mod 7) and

1 or - 1 according as a is or is not a square modulo p .

sociated with a particular characterization of the MI,, groups for every m

EXERCISE 42. Theorem 30 may be generalized to read that the number

25. Is THE QUADRATIC LAW A DEEP THEOREM?

(a) 3z + 2 = 0 (mod 7). z2 = 1 (mod 24)

z2 = z (mod 10) then

has 4 solutions 0, I , j,and 6. I n addition to the set mm,o,

and since, from Eq. (115), each [ d ( d ) - + ( d ) ] 2 0, we obtain

and 0101 . . . 10,

any of the 2N - 2 remaining patterns is a legitimate one, and will occur

+ 11 p is P or smaller. Further, for p > 3 , p - 1 = -1 is not a primitive root,

EXERCISE 58. Criticize the word explicitly in the last sentence in ~

m 63 really needs three dimensions. The four bunches, of three cycles

Further, if yt = y, , we have x1y1-'y2 = xlyl-'y, , and thus x2 = x, .There-

EXERCISE 67. Draw a cycle graph for ms3.

mlo6 m = plelpznz . . . P,"". (121)

m = 2, 4, p", or 2p" m = 17 417 = 16

where p i s a n odd prime and n 2 1. Now we can state

(A) Factor m into it,s standard form: m4

EXERCISE 73. If k is odd, 311k $ 3 n 2 k . If k is prime to 3 and 4, 3 n a k

@104,]05 = 2.2.12 mloshas = 6 quadratic residues.

Then the R = +(m)/Presidue classes,

By the binomial theorem

a 3. -- ( - 1 ) 8 J 3 t J ( m o d 2 " ) (131) 34 = 1 + 16 + 32t

for j = 1, 2, * * * , n. Then set h = Bk + g , hj = Akj + gj , and we have

by Fermat's Theorem. Therefore

is divisible only by divisors of the form 4lc + 1.