L.J. Mordell Diophantine Equations

DIOPHANTINE
EQUATIONS
This is Volume 30 in
PURE AND APPLIED MATHEMATICS
A series of monographs and textbooks
Edited by PAUL A. SMITH and SAMUEL EILENBERG,
Columbia University, New York
A complete list of the books in this series appears at the end of this volume.
DIOPHANTINE
EQUATION S
L. J. MORDELL
ST. JOHNS COLLEGE
CAMBRIDGE, ENGLAND
1969
ACADEMIC PRESS London and New York

ACADEMIC PRESS INC. (LONDON) LTD.
Berkeley Square House, Berkeley Square, London, W 1X 6BA
US.Edition published by
ACADEMIC PRESS INC.
111 Fifth Avenue, New York, New York 10003
0 1969, BY ACADEMICPRESSINC.(LONDON)LTD.
COPYRIGHT
ALL RIGHTS RESERVED.
NO PART OF THIS BOOK MAY BE REPRODUCED IN ANY FORM,
BY PHOTOSTAT, MICROFILM, OR ANY OTHER MEANS, WITHOUT
WRITTEN PERMISSION FROM THE PUBLISHERS.
LIBRARY
OF CONGRESS
CATALOG
CARDNUMBER:
68-9112
PRINTED IN GREAT BRITAIN BY WILLIAM CLOWES AND SONS LTD

LONDON AND BECCLES
Most books on number theory treat with more or less detail various aspects
of Diophantine analysis, a subject which can be described briefly by saying
that a great part of it is concerned with the discussion of the rational or integer
solutions of a polynomial equation f(xl, xz. . . , x,) = 0, with integer co-
efficients. It is well known that for many centuries, no other topic has engaged
the attention of so many mathematicians, both professional and amateur,
or has resulted in so many published papers. Nevertheless, apart from reports
or surveys, there are very few books dealing only with Diophantine equations,
and sometimes these are either very elementary or deal only with special
aspects of the subject. It seems therefore desirable to produce a more repre-
sentative account, and an attempt is now made to do this. I have tried to
preserve the old spirit and traditions of Diophantine analysis, and to these I
attach great importance.
It is hoped that all readers will find herein some results which are of interest
to them and which do not require too much knowledge. Some well known
elementary results have been included for the sake of completeness. Accounts
have been given without undue generality of many results which seemed to
me of greatest significance, and most representative of the subject. Note has
been taken of various Diophantine problems, both classical and recent ones,
which seemed worthy of special attention. I have included a number of results,
really part of the subject but not usually found in accounts of Diophantine
equations, and indicating something about their role in number theory. The
material is arranged in a systematic way such that a basic idea runs through
each chapter, and it is hoped that the presentation is reasonably self-contained.
The proofs of the results vary considerably. Some are very elementary and
others are rather simple, making little demand upon specialized knowledge.
This does not apply to the demonstration of the really important theorems.
Many require a knowledge of the fundamental results for an algebraic number
field, for example, the basis for the integers, the finite generation of the
units and the finiteness of the number of classes of ideals. When more know-
ledge of algebraic numbers is required as for Fermats last theorem, I state
the results assumed in the proof. Occasionally a brief introductory account
vi PREFACE
is given of relevant results or principles required, for example, of some

algebraic geometry and the invariants and covariants of binary cubic and bi-
quadratic forms, and also for local or p-adic applications.
This book had its origin in the lectures on Diophantine equations given at
Cambridge (England), Toronto (Canada), Varenna (Italy), and Urbana
(Illinois), and so the reader must not expect an exhaustive treatise. However,
it seemed desirable to enlarge the scope of the book and in particular some
mention is made of the more important and worthwhile results, especially
recent ones.
A great deal of number theory also arises from the study of the solution in
integers of a polynomial equation f ( x l , x2,. . ., x,) = 0, and so our subject
is coextensive with most of number theory. Hence it becomes necessary to
make a choice of material, both classical and modern.
There is no need in this book to go into detail about general Diophan-
tine problems whose study forms part of a general theory to be found in the
usual treatises such as the representation of number by quadratic forms in
several variables, or by norm forms in algebraic number fields.
No account has been given of theorems whose proofs are of an advanced
analytical character such as the applications of the Hardy-Littlewood-
Vinogradoff circle method, though many interesting and important results
have been found in this way, for example, Warings theorem on the represen-
tation of an integer as a sum of powers, and Davenports theorem that every
homogeneous cubic equation in at least 16 variables has an infinity of integer
solutions.
Unless the results are of fundamental importance, arithmetical proofs have
also been usually omitted when they are lengthy and complicated, or involve
considerable details, or excessive numerical work, or are out of place in this
book. Among such instances are Linniks proof of Warings theorem, and
+
Siegels theorem on the solution of the binomial equation a x 1 by = c.
Many striking results on the solution of equations of the form ax + by = c
can be stated very simply but, unfortunately, their proofs require a surprisingly
large amount of calculation and so are not discussed here. This applies to
proofs of many simple and long sought results given by Ljunggren. An out-
line, however, is given of some of the specially interesting ones.
I have also omitted a discussion of results which require too much pre-
vious preparation and knowledge such as the modern developments linking
Diophantine analysis with the new algebraic geometry and homological
algebra. These have led to many new points of view as can be seen from
Langs book and Cassels recent report.
Most of the book is concerned with the classical theory of the subject, in
which the unknowns were in general elements of the rational field Q, and the
solution of the equation involved only operations in this field. Occasionally
solutions were considered in some quadratic and cubic fields, and more
PREFACE vii
importantly in cyclotomic fields in connection with Fermats last theorem.
Further developments led to emphasis on solutions in fields other than Q,
and in particular, in finite fields. This required a discussion of the solvability
and the number of solutions of a polynomial congruence,
f(4=f(xl, x2,. . ., x,) = 0 (modp3,
where p is a prime number. Such congruences had previously arisen in the
solution of some equationsf(%) = 0, and in particular in the Hardy-Little-
wood analytic solution of Warings problem that every positive integer n
can be expressed as a sum of k r-th powers of non-negative integers where k
is independent of n.
Great advances have recently been made by the association of con-
gruences with types of zeta-functions, and many important and suggestive
results have been found. In particular, this has led to highly probable con-
jectures whenf(x) = 0 is a plane cubic curve of genus one, and a new field
of research has been opened up.
Sufficient has been said to show that Diophantine analysis draws upon
resources from many branches of mathematics, for example, the higher
arithmetic, algebra, geometry, analysis ;not only the classical aspects of these
subjects but also the most recent developments.
I have much pleasure in acknowledging my great indebtedness to the many
writers on Diophantine equations, both ancient and modern, many of whom
are not mentioned here. From these, I have learnt much and also derived
inspiration. In fact, my interest in the subject was aroused in my school days
by reading the chapters on Diophantine equations usually to be found in
many of the algebra books of the first half of the nineteenth century. These
except for Eulers algebra have long since been forgotten.
Among the writers on Diophantine Equations whom I have studied are
Borevich and Shafarevich, Carmichael, Delone and Faddeev, Dickson, Lang,
Nagell, Skolem. Many books on number theory also contain useful sections
on Diophantine analysis. There may be mentioned those by Bachmann,
Hardy and Wright, Landau, Le Veque, Nagell, Sierpenski, Uspensky-
Heaslet. I have also consulted many papers and memoirs, in particular those
by Billings, Birch and Swinnerton-Dyer, Cassels, Erdos, Fueter, Ljunggren,
Nagell, Pocklington, Roth, Selmer, Skolem, Siegel. Many references have
been given but the list is by no means complete, especially for the older results.
For these Dicksons invaluable History of the Theory of Numbers may
be consulted.
Nov. 1968 L. J. MORDELL

This Page Intentionally Left Blank
Contents
Preface V
1. Introduction 1
2. Equations Proved Impossible by Congruence Considerations 3
3. Equations Involving Sums of Squares 12
4. Quartic Equations with only Trivial Solutions 16
5. Some Linear Equations 30
6. Properties of Congruences 34
7. Homogeneous Equations of the Second Degree 42
8. Pells Equation 53
9. Rational Solutions Derived from Given Ones 66
10. Rational Points on Some Cubic Curves 76
11. Rational Points on Cubic Surfaces 82
12. Rational and Integer Points on Quartic Surfaces 90
13. Integer Solutions of Some Cubic Equations in Three Variables 100
14. Simple Algebraic Considerations 111
15. Applications of Algebraic Number Theory 121
16. Finite Basis Theorem for the Rational Points on a Cubic Curve
f(x, y , z ) = 0 of Genus one 138
17. Rational Points on Curves of Genus g = 0 or 1 and g > 1 150
18. Representation of Numbers by Homogeneous Forms in Two
Variables 157
19. Representation of Numbers by Special Binary Quadratic and
Quaternary Quadratic Forms 164
20. Representation of Numbers by Homogeneous Forms in Several
Variables 174
21. Representation of Numbers by Polynomials 181
22. Thues Theorem on the Integer Solutions off(s, y ) = nz 186
I*
X CONTENTS
23. Local Methods or p-Adic Applications 200

24. Binary Cubic Forms 213
25. Binary Quartic Forms 232
+
26. The Equation y2 = x3 k 238
+ + +
27. The equation y2 = ax3 bx2 cx d 255
28. Some Equations of Degree > 3 262
29. Fermats Last Theorem 280
30. Miscellaneous Results 287
Bib1iogr aphy 306
List of Equations and Congruences 307
Acknowledgments
I have much pleasure in acknowledging my very great indebtedness to many

colleagues for their numerous comments and suggestions. Among these are
Dr A. Baker, Prof. Cassels, Prof. Chalk, Prof. Davenport, Mr Makowski,
Mr H. Montgomery, Prof. Nagell, Dr M. Newman, Dr A. Rotkiewicz,
Prof. Schinzel, Prof. de Witte. I am very grateful to all of them for their
valuable assistance with the manuscript and proof sheets and the resulting
clarification in the exposition. Prof. de Witt and Mr Montgomery should be
given special mention.
Mr Montgomery also prepared the list of equations.
The book has profited from the great knowledge and the generous help of
Prof. Nagell in reading the proof sheets.
I am especially indebted to Prof. Schinzel who made a most painstaking
reading and careful scrutiny of the proof sheets. He corrected many errors,
noted many obscurities, and suggested numerous improvements.
Finally, I should like to thank the Academic Press for the great help they
have given me in preparing my manuscript for publication and for meeting my
every wish. It has been a great pleasure to deal with them.
Feb. 1969 L. J. MORDELL

This Page Intentionally Left Blank
CHAPTER 1
introduction
Preliminary
1. Let xl, x2,. . . ,xn,say x, be n variables, and let f(xl,x,, . . . , x,), say f(s),
be a polynomial in these variables with rational coefficients. There will be no
loss of generality in supposing that the coefficients are integers since we shall
be concerned with equations f(:z) = 0. An obvious question is the
Problem. To find some or all of the solutions of f(x) = 0 in
I. rational numbers, i.e. in the field Q ;
11. rational integers, i.e. in the ring Z.
Suppose first that f(z) is a homogeneous polynomial. We ignore the trivial
solution z = 0. Then the questions I and I1 are clearly? equivalent, and we
can confine ourselves to integer solutions with (xl,x2, . . ., x,) = 1.
Suppose next that f(z)is a non-homogeneous polynomial. On putting
=yl/~ntl,...>xnYn/yn+l,
x1
we have a homogeneous equation
dY1,Y,, . . Yn,Y n t 1) = 0.
. 3
There is now a 1-1 correspondence between the rational values of z, and

those integer values of y with # 0, and (yl,y,, . . .,yn+ = 1.
We may also have simultaneous equations of the typef(rc) = 0, e.g.
f1(z) = 0,. . . ,fr(x) = 0.
These can be written as the single equation

f?(X) + ... + f?(X) 0.=
More generally we can impose restrictions upon the variables. Thus we can
require them to be positive integers. The simplest significant problem then
arising is x l x 2 = n, and the solutions of the natural questions arising from
this lead in due course to the classical results on the divisor problem and the
theory of prime numbers. We are also led to Warings problem on the repre-
sentation of integers as sums of rth powers. We can require variables to be
prime numbers and then problems associated with Goldbachs theorem arise.
We may also allow the variables to be algebraic integers, for example, Gaussian
integers of the form x + yi where x and y are rational integers.
t This statement due to Gauss is disputed by Dickson.1,2
2 DIOPHANTINE EQUATIONS
The general problem suggests many questions. Can we find reasonably

simple necessary conditions for the solvability of f ( x ) = O ? Are these con-
ditions sufficient? Can we, having found some solutions, deduce from these
others or an infinity of others, or all the solutions? What can be said about the
number of solutions or their magnitudes in terms of the coefficients of f ( x ) ?
Questions may also be asked about the arithmetical properties of solutions.
2. Necessary conditionsfor solvability.
Theorem 1
Rational or integer solutions of f ( x ) = 0 can exist only if f ( x ) = 0 can be
satisfied by real values of x.
Proof. Obvious.
Theorem 2
Integer solutions of the inhomogeneous equationf ( x ) = 0 can exist only ifthe
congruence
f ( x ) = 0 (mod M )
has solutions for all integers M.
Proof. Obvious.
The elementary properties of congruences show that we need only consider
M = pa where p runs through the primes and a = 1,2, . . ..
Theorem 3
Integer solutions x # 0 of the homogeneous equationf ( x ) = 0 can exist only
if the congruence
f ( x ) = 0 (mod M ) ,
has solutions for which (xl,x2,. . ., x,, M ) = 1 for all integers M ; and if in
particular when M = pa, it has solutionsfor which not all x are divisible by p.
This is obvious since we may suppose ( x l , x2,. . ., x,) = 1 in a homogeneous
equation, and this x must satisfy f ( x ) = 0 (mod M ) .
It is also obvious that if an inhomogeneous equationf(x) = 0 implies that
x - =x = - . . = x, = 0 (mod pa), for given p and arbitrary large a, then
x = 0 is the only solution.
Finally, if the equationf(x) = 0 implies that one variable, say x l , is divisible
b y p for an infinity of primesp, then there can only be solutions with xl = 0.
REFERENCES
1. L. E. Dickson. Fallacies and misconceptions in diophantine analysis. A new
method in diophantine analysis. Bull. Amer. Math. SOC.,27 (1921), 3 12-3 19.
2. L. E. Dickson. Modern Elementary Theory of Numbers. Univ. Chicago
Press, Chicago (1939), Chapter IX.
CHAPTER 2
Equations Proved Impossible by

Congruence Considerations
1. We now consider the

Problem. To find by congruence considerations, equationsf ( x ) = 0 , with either
no integer solutions or only the solution z = 0.
This requires the application of some elementary results in number theory.
Many of the results now given are classical. We begin with a preliminary dis-
cussion of the equation
x; = a + bx2, (1)
where r is an integer > 1. First, let r = 2 and so
X: = u + bx,.
This is solvable if and only if the congruences
x: = a (mod 29, x: = a (mod p s ) , x: = a (mod @'),. . .
are solvable, where 2" 11 b, i.e. 2" is the highest power of 2 occurring in b,
p s 1 b . . . . For simplicity, we suppose (a, b) = 1. When a = 1, the congruence
is always solvable; when CI = 2 only if a = 1 (mod 4); when CI = 3 only if
a = 1 (mod 8), and then it is also solvable for CI > 3. The congruence is solv-
able for all fl if it is solvable for fi = 1. Then a is called a quadratic residue of
p , i.e. (alp) = 1. If the congruence is insoluble, a is called a quadratic
non-residue of p , i.e. (a/p) = - 1. From these results, it folIows that every
prime divisor p of x2 - a for integer x is either a divisor of a, or can be repre-
+
sented by a finite number of arithmetic progressions. Thus ifp I ( x 2 c2)and
p # 2, then if ( p , c ) = 1, p = 1 (mod 4), but if p = 3 (mod 4), p I c.
is impossible in integers fi f ( x l , x2, . . ., x,) has a prime factor p which cannot

be a diuisor ofg(x), e.g. g(x) = x2 - a where (alp) = - 1.
Proof. Obvious.
Suppose next that r = 3 in (1). Then
X: = (I + bx2, (4)
and for simplicity we suppose that (a, b) = 1. Let 3" (1 6, p B 1 b, . . . . Then (4)
is solvable if and only if
x3 -
= a (mod 3"), x; = a (mod p8), x: = a (mod qy),. . . (5)
are solvable. When a = 1, the first congruence is always solvable. When
a = 2, it is solvable only when a = & 1 (mod 9), and then it is solvable for all
a. Take next the second congruence. When p = 2 (mod 3), it is solvable for
all ,8. When p E 1 (mod 3), the condition for solvability is not so simple and
it is not easy to specify the values of a for which the congruence is solvable.
Then a is called a cubic residue o f p and we write (alp), = 1. It might be noted
that (2/p), = 1 if and only if p can be represented in the form p = x2 27y2 +
with integers x and y. Similar remarks apply for other values of r.
2. Congruences mod M .
We now apply to various equations, congruences mod M , where M is a
prime power. We commence with M = 2".
M = 4.
The equation
x: + x i = 4x3 + 3
has no integer solutions.
For x: = 0, 1, x i E 0, 1, x: + xi 3 0, 1, 2.
The equation
x: + x i = (4a + 3)xi (7)
has only the integer solution x1 = x2 = x3 = 0.
This is obvious if a < 0, and so we need only consider a > 0. From (6),
x3 f 1 (mod 2), and so x3 = 0 (mod 2), and then x1 = x2 = 0 (mod 2).
Since we may suppose (xl, x2,x3) = 1, we have a contradiction unless
X I = x2 = x3 = 0.
M = 8.
The equations
x: + 2xi = 8x3 + 5 or 8x3 + 7,
x? - 2xi = 8x3 + 3 8x3 + 5
(8)
and or
have no solutions.
For to mod 8, x: = 0, 1,4 and so x: + 2xi f 5,7, x? - 2xi f 3, 5.
The equation
X: + X: + x i = 4" (8xP + 7) (9)
has no solutions.
2. CONGRUENCE CONSIDERATIONS 5
If ct 2 1, x1 3 xz = x3 = 0 (mod 2), and so we need only consider CL = 0.
+
But then x: + x i x i f 7 (mod 8).
The equation
ax: + bxi +
C X : = 0, abc # 0, (XI, ) 1
~ 2~, 3 = (10)
has only the trivial solution x 1 = 0, xz = 0, x3 = 0 if either
a = b = c = 1 (mod 2) and a = b = c (mod 4),
or a/2 3 b =c = 1 (mod 2) and b +c E a or 4 (mod 8).
The result is obvious in the first case since
xf + xi + xi f 0 (mod 4).
In the second case, we write
ax? + bxi + cx,Z = 0 (mod 8).
Clearly x l , x2, xg are not all odd since a + b + c $ 0 (mod 8). Also two of
the x cannot be even and so only one can be even. This cannot be x 1 since
b + c f 0 (mod 8), nor x2 since cx3 is odd, nor x3 since bx2 is odd.
The equation
z2 = (ax2 -t by2)2 - 2k(cx2 + dy2)' (1 1)
has no integer solutions #(O, 0, 0 ) if a + b = 0 (mod 2), cd = 1 (mod 4),
k = 1 (mod2).
Suppose first that x = 1 (mod 2), y = 0 (mod 2). Then
= a2 - 2c2 = - 1 or 2 (mod 4),
z2
and this is impossible. Similarly if x = 0 (mod 2), y = 1 (mod 2).

Suppose next that x = y = 1 (mod 2). Then
(I)2(T) = a + b 2 - 2k (1) + '(mod4)
= -1 or 2 (mod 4),
since c + d = 2 (mod 4).
M = 16.
The equation
ax: bx:+ CX; + dx: + = , x3, x,) = 1
0, ( x ~x2, (12)
has only the trivial solution s = 0 if
I. a f O(mod 16), etc.
11. a +
b f 0, a + c $ O(mod 16), etc.
111. a + +
b c f 0, a + b + d f 0 (mod 16), etc.
1V. a + +
b + c d $ 0 (mod 16).
Since x: = 0 or 1 (mod 16), these conditions exclude in turn one odd value,
two odd values etc. for the variables. A simple instance occurs if we take
a, b, c, d to be congruent mod 16 to any distinct four of 1, 2, 3,4,5, 6.
M = 2".
The equation
UX: + bxz + C X ; = 2dx1~2~3, (13)
where a =b =c= & 1 (mod 4) has only the solution x1 = xz = x3 = 0.
Obviously x,, xz, x3 cannot all be odd, for if one is even, so are the others
since then
x: xi x: = O(mod4).+ +
Put x1 = 2X,, xz = 2Xz, x3 = 2X3. Then
aXf + b X i + cX," = 4dX1X2X3.
Similarly X,, A', X 3 are all even. Hence all the x must be zero since they are
divisible by 2" where a is arbitrary.
M = 32.
The equation
(alxf + azxi + a,xg + a,x:)(b,x: + bzxi + b3x; + b4x3
= 2k(c,x: + C ~ X ;+ + ~ 4 x : ) ( d l +~ f d 2 ~ ;+ d 3 ~ ;+ d 4 ~ 3 (14)
C~X;
has only the trivial solution x = 0 if the a, b, c, dare all odd, and
a, = az = a3 = a4 (mod 8), = b2 = b3 = b, (mod 8),
bl
(c, + c2 + cg + ~4)(d1+ dz + d3 + d4) = O(mod 16).
We may suppose without loss of generality that
alx: + azxi + a3xi + a4x: = 0 (mod 2).
Then either
I. XI xa 3 x3 = x4 = 1 (mod 2),
or, say,
11. x1 = x2 = 1 (mod 2), x3 = x4 = 0 (mod 2).
+ + +
For I , alxy a2xi a3x; a4x: 3 4 (mod 8), and so the left-hand side of
(14) is = 16 (mod 32). The right-hand side is =O (mod 32), i.e. a contradiction.
+ + +
For 11, alxT + azx$ a3xi alx: = a, a, = 2 (mod 4), and so the left-
hand side of (14) is =4 (mod 8). The right-hand side is =O (mod 8), i.e. a
contradiction.
M = 3".
M = 3.
The equation
(3a + l)x: + (3b + l)xg = 3c, c $ 0 (mod 3) (15)
Here xf + xz = 0 (mod 3), and since x: = 0, 1 (mod 3), x1 = x2 =
0 (mod 3), and this is impossible.
M = 9.
The equation
x; + x i + xi: = 9x4 f 4 (16)
This is obvious since x: = 0, ? 1 (mod 9).
The equation
x; + 2x; + 4x: = 9x2, (XI, x2, x3, x4) = 1 (17)
has only the trivial solution 5 = 0.
+
Here x; + 2x; 4x: = 0 (mod 9). The only solution of this is x1 = x2 =
x3 = 0 (mod 3), and then x4 = 0 (mod 3).
The equation
ax3 + 3bx2y + 3cxy2 + dy3 = z3 (18)
has no integer solutions if
a = d = 4 (mod 9), b = 0 (mod 3), c = 5 1 (mod 3).
We may suppose ( x , y , z ) = 1. Clearly x y f 0 (mod 3), since 4x3 =
z3 (mod 9) requires x= z = y = 0 (mod 3). From (18), z = a x + dy
(mod 3),
z3 = a3x3 + 3a2dx2y + 3ad2xy2 + d3y3(mod 9),
and so
Since to mod 3, x3 = x, x2 = 1, we have

x + (b - 1)y + (c - l)x +y = 0 (mod 3), or x = 0 (mod 3),
and this has been excluded.
The result implies that the equation
ax3 + 3bx2y + 3cxy2 + dy3 = 1
has no rational solutions.
8 DlOPHANTlNE EQUATIONS
M = 7.
The equation
(7a + 1 ) ~ :+ (76 + 2 ) ~ ;+ ( 7 +~ 4 ) ~ :+ (7d + I ) x ~ x= ~0,x ~
(xi, xa, = 1,
~ 3 ) (19)
Here x? + 2x; + 4x: + x1xzx3= 0 (mod 7).
Also x: = 0, f 1 (mod 7). It suffices to show that x1 = x, = x3 = 0 (mod 7);
if x3 = 0, x i + 2xi = 0 and then x1 = xz E 0: if x3 f 0, we can put
x1 E X1~3,x, = XZx3, and then
x: + 2x; + X,X, + 4 = 0.
Clearly XlX2 f 0, and so X : = 5 1, X i= f 1. This leads to four impossible
cases since
1. l
'
: E Xi:= 1 gives X , X , = 0,
2. x: = x: E -1 gives XIX, = -1, i.e. X ; X ; = -1,
3. x: = -xi 1 gives XIX, = 4, i.e. X : X ; = 1,
E
4. X : = - X i = - 1 gives X1Xa E 2, i.e. X : X ; = 1.
M = 72.
The equation
X: + 2 ~ ;= 7 ( ~ : + 2x3, (XI, xa, ~ 3 ~, 4 ) 1
= (20)
Here x: +
2xi E 0 (mod 7) and so x1 = 7x1, x, = 7 X z . Then
7yx; + 2 x 3 = x: + 2x:.
This gives x3 = 7 X 3 , x4 = 7X4 and then (x,, xZ,x3, x4) 2 7.
An obvious deduction is that the equation
x3 + 2y3 = 7(z3 + 2)
has no rational solutions.
M = pa, p a prime.
The equation
x: + 1 = px,, p = 3 (mod 4)
is impossible.
Obvious since (- l / p ) = - 1.
The equation
x: + 1 = axz,
is impossible i f a has a prime factor E3 (mod 4).
So for the equations
x: - 2 = px, if p = k 3 (mod 8)
and x: + 2 = px, if p = 5, 7 (mod 8).
The equation
X; -2 = 7x2
is impossible.
Proof. Obvious.
It is more difficult to deal with the equation
x; - 2 = px, (27)
if p = 1 (mod 3). As already stated on page 4, (27) is solvable, i.e. 2 is a
cubic residue of p if and only if for integers a and b
p = a2 + 276,.
The equation
x: + xi - p(xg + x:) = 0, p = 3 (mod 4), (xl,x,, x3,x,) = 1 (28)

has only the solution 3: = 0.
Here x: + xi = 0 (modp), and so x1 = p X 1 , x2 = pX2. Then x3 = p X 3 ,
x4 = p X , etc.
The equation
ax? + bpxi + cp2x: = 0, (xl, x2,x3) = 1, abc $ 0 (modp) (29)
has only the solution m = 0.
Here x1 = 0 (mod p ) and so x1 = p X l . Then
ap2X; + bxg + cpxz = 0.
Similarly x, = 0 (modp) and x, = pX,. This leads to x3 = 0 (mod p).
A more general result' is given by the
Theorem 2
Let f(x, y , z ) be a cubic homogeneous polynomial such that the only solution
of f(x, y , z ) = 0 (modp), p a prime, is x = y = z = 0 (mod p). Let
fs(x,, ys, ZJ, s = 1, 2, 3, be three such polynomials each associated with the
same prime p. Then the equation in nine variables
flfi(X1, y1, Z l ) + Pfi(X2, Y2,z2)
+ pY3(x39 ~ 3 + p3
~3 3 ) 5
I.m.n=l
almnXlYmZn = 0, (301
where the sufixes I, m, n in any one term are all different, has only the trivial
solution x, = y , = z, = 0, (s = 1,2, 3).
A congruence mod p shows that x1 = pX,, y1 = p Y l , z1 = pZl. Similarly
we obtain x, = pX,, y , = p Y,, z2 = p Z , and x3 = pX,, y , = p Y,, z3 = pZ3,
etc.
A similar result holds for equations of the nth degree in n variables.
An application with p = 7 is given from (19) by
+ + + +
f ( x , y , z ) = (7a l)x3 (7b 2)y3 (7c + 4)z3 (7d 1)xyz. (31)+ +
For other instances, take an equation irreducible over the rational field Q
with three roots O,, 02, O,, and a prime p such that p is a prime in the field
Q(Ol, 02, 0,). Let my), wg), w g ) be a basis of the integers in the field Q(O,),
(r = 1,2,3), the w y ) denoting the conjugates of wil) = wl. Then we can take
f ( x , Y , 4 = N x w , + yw, + zw,),
N denoting the norm in the cubic field, i.e. the product of three conjugate
expressions. For if f ( x , y , z ) E 0 (modp), since p is a prime in Q(Ol, O,, O,),
we must have for some r
+
xwy) ywg) + zwy) = 0 (modp).
Since the w are a basis, x E y = z E 0 (modp).
For the application above, we note that
x3 + 2y3 + 4z3 + xyz E 0 (mod 7 )
can be written as
N(x + y q 2 + z q 4 ) = 0 (mod 7).
Since 2 is not a cubic residue of 7 , we have x E y E z E 0 (mod 7).
The method of proof for equation (29) in nine variables is of some interest
since Artin has stated the
Conjecture
A homogeneous equation f ( x ) = 0 of degree r with n > r2 variables always
has a non-trivial integer solution in any p-adic Jield, i.e. as a congruence
mod pa,for all primes p and integers a > 0.
This has been long known for r = 2, n = 5 and is proved in Chapter 7; and
it has been proved recently for r = 3.
A counterexample with r = 4, n = 18, has just been found by Terjanian,:
Theorem 3
Let f ( x ) = x: + xi + xt - xXxi - xix: + x,
- xzx: - x1x~x3(x1 + x3).
Then the equation
f ( 4+ fb)+ f ( 4 - 4 U W + f ( 4+ f(l.l)) = 0, (32)
where (x,y , x ,(c, v, u)= 1, has only the trivial zero solution since there is no
2-adic solution.
We prove first that f(z)5 1 (mod 4), unless x1 = x2 = x3 = 0 (mod 2).
We have three cases.
1. If x 1 3 x2 3 x3 = 1 (mod 2), say, x1 = 2 X , +1, etc.,
f(x) = - ~ 2 x3x1 - xlx2 (mod 4)
~- 3
= -2(1 + 2X2+ 2X3) (mod 4)

= -3 (mod 4).
2. If x1 3 x2 = 1 (mod 2), xa = 0 (mod 2),
f ( x ) = 1 - x2x3 - x1x3 = 1 (mod 4).
3. If x1 = 1 (mod 2), x2 = x3 = 0 (mod 2),
f ( x ) = 1 (mod 4).
Hence f(z) = 0 (mod 2) only when x = 0 (mod 2). The equation requires
m = 0 (mod 2), y = 0 (mod 2), z = 0 (mod 2) since otherwise the equation
is impossible when considered as a congruence (mod 4). Hence we must have
x = 2%, y = 2CV, z = 2 3 , and then
f&> + f(.) + f(u>- 4(S(9") + fW) + f(m>= 0.

This gives u = 2%, v = 2"y;w = 2% and this contradicts ( x , z, ~, u,V , U ) = 1.
It has been shown by Shanuel that for largep, there exists r such that for a
form f ( z ) of degree r, with more than r2 + 1 variables, f ( z ) f 0 (modp")
for a certain a unless x = 0. This has been proved for all p by J. Browkin3
who has also found other results.
By ideas completely different from those above, a non-constructive proof4
has been given of the
Theorem 4
For every natural number r, there exists afixed set S ( r ) of primes such that
for every prime p not in S(r), every homogenous equation f ( x ) = 0 of degree
rand with n > r2 variables, always has a non-trivial solution in thep-adicfields.
REFERENCES
1. L. J. Mordell. A remark on indeterminate equations in several variables. J.
Lond. Math. SOC.,12 (1937), 127-129.
2. G . Terjanian. Un contre-exemple A une conjecture d'Artin. Comptes Rendus
Acad. Sci. Paris, 262 (1966), 6 12.
3 . J. Browkin. On forms over p-adic fields. Bull. Acad. Polon. Sci. Sir. des sci.
math. astr. et phys., 14 (1966), 489-492.
4. J. Axil and S. Kochen. Diophantine problems over local fields I. Amer.
J. Math., 87 (t965), 605-630.
CHAPTER 3
Equations Involving Sums of Squares
Theorem 1
The general integer solution of the equation
xl + x; + . . . + x: = x2, (XI, X Z , . . * ) x,) = 1, x > 0, (1)
is given by
where the X are arbitrary integers with (X 1 , . . . , X,) = 1 and d > 0 is taken so
that ( x l , x2,. . ., x,) = 1. For given xlrx2, . . . ,x, there are two sets of values
for the ratios Xl:X 2 : . . . : X,.
Put x1 = tX,, . . ., x,-1 = txn-1, x, = tx, - x.
Then t(x: + xi + . + Xi) = 2 x x , ,
* *
t(x: - x: - * . .- XE-1) = 2xnXn,

and so
and (2) follows.

Conversely, suppose that the x l , . . ., x , are given in (3). We now solve for
the X. From homogeneity considerations, it suffices to take X, = 1, x = 1.
Then
= ux1, . . . , xn-l = vx,-1,
x,
say, whence from (3),
2v = 1 + v2(x: + . + X i - l ) ,
*
vy1 ); - 20 + 1 = 0,
(u(1 + x,) - l)(v(l - x,) - 1) = 0,
and the result follows.
We now examine some particular cases. We take n = 2 and write
x2 + y2 = z2, (x,y ) = 1, z > 0. (4)
3. EQUATIONS INVOLVING SUMS OF SQUARES 13
Here
x = -2 X Y Y2 - X2
d Y = d 9 z= y2 X 2 , ( X , Y ) = 1. (5)
d
+
Clearly d = ( 2 X Y , Y 2 - X2,Y 2 X)= 1 or 2. +

If d = 1, then X , Y cannot both be odd or even, and so we have the solution
when x is even. If d = 2, A, Y are both odd and replacing them by Y f X
we have from ( 5 ) ,
x = Y2 - X2, y = 2XY, z = Y2 + X2. (6)
This is the solution when y is even.
The equation
X2 + y2 = 2z2, (x,y ) = 1, z > 0. (7)
We cannot have z even since then x = y = 0 (mod 2).
Hence x, y are both odd, and so we can put
x = X + Y, y = x - Y, ( X , Y ) = 1 ,
and X2 + Y2 = 22.
Then X = u2 - b2, Y = 2ab or X = 2ab, Y = a2 - b2.

The solution of (4) may also be found by arithmetical considerations. We
need only consider non-negative solutions. Clearly x,y are not both odd since
then x2 = y 2 = 1 (mod 4), and z 2 = 2 (mod 4). Hence z is odd. We consider
only the solutions with even y. Write
y2 = 2 2 - x2 = ( z - x)(z + x),
then ( z - x , z + x) = 2 since x, z are odd. Hence
z + x = 2Y2, z - s = 2X2, y2 = 4X2Y2,
and so x = Y2 - y = 2XY,
X2, z = Y2 + x2.
Clearly every non-negative solution (s,y ) leads to a non-negative (A, Y ) .
The general solution of (4) can also be found very simply by applying the
arithmetic theory of the Gaussian integers x + iy where x, y are rational in-
tegers. The main results required are as follows. The units, that is divisors of
unity, are f 1, f i. The primes p = 1 (mod 4) can be expressed in the form
x2 + y 2 with integers x, y. The primes q = 3 (mod 4) cannot be expressed in
this way. Since 2 = i(l - i), 2 is not a prime. Hence the primes in the com-
plex theory are typified by the factors x + iy of p , by q, and by 1 - i. A
Euclidean algorithm holds for the Gaussian integers, and so factorization is
unique if units and the order of the prime factors are disregarded.
We write (4) as
(x + iy)(x - iy) = z 2 .
As z is odd, the two factors are relatively prime for a common factor would
divide 2x, 2iy. Hence
x + iy = ( X + i Y ) 2 or i(X + iY)2,
since - 1 = i 2 can be absorbed in the square. This gives the result.
The equation
X2 + y2 = pz2, z > 0,
where p is a prime = 1 (mod 4) and so p = r2 + s2.
All the integer solutions are given by
x + y i = ( f r f i s ) ( Z + iW)2,
where Z , Ware integers and the ~f:signs are independent of each other.
We now take n = 3 and write
x2 + y2 + 22 = w2, (x, y, 2) = 1, w > 0, (9)
Clearly only one of x, y , z can be odd, say z, and so it suffices to consider the
equation
4x2 + 4y2 + 22 = w2, (x, y , z ) = 1.
(10)
We give a general solution different from (2) by using complex numbers.
This has been presented by Carmichael' without using them as
x = xz - YW, y = xw + YZ,
(1 1)
z = x2 + Y2 - 2 2 - w2, w = x2 + Y2 + 2 2 + w2.
We have + 4y2 = w2 -
4x2 22.
Hence w - z = 2f, w + z = 2(x2 + y"/f, (12)

where, by adjusting the sign of z, we may suppose thatfis odd. Suppose that
x = y = 0 (mod q"), but not (mod qa l), where q is a prime = 3 (mod 4). It
+
follows then that either ( f , q) = 1 orf 3 0 (mod q2"). For iff = 0 (mod q B ) ,
but not qB+ and 3, < 2cr, then also w = z = 0 (mod 4). Write
x + iy = n (a, + ib),
for the decomposition into complex primes. Then f as a divisor of x2 + y2
is either prime to q or divisible by qZaand so can be expressed as a sum of two
squares, and then has a representation
f= n (a? + b?) = z2+ w ~ ,
where the as, 6, are included among the arrb,, and in which we can take
Z + iW = n (a8 + ib,).
3. EQUATIONS INVOLVING SUMS OF SQUARES 15
Then x + iy = ( X + iY)(Z + iW)
where X , Yare rational integers. This with (12) gives (11).
The equation
x2 + y2 = z2 + w2.
The complete solution is given by

2x + d2Y,
= d1X 22 = dlX - dzY,
2w = d1Y + dzX, 2y = d1Y - dZX,
where d,, d2 are arbitrary integers.
We deduce this from the solution of
AB = CD.
Write ( A , C) = dl and so A = d l X , C = d,Z,
(B, 0)= d2 and so B = dzY, D = dzW,
then XY = zw.
Since(Y, W ) = 1, and ( X , Z ) = 1, X = k W, Y = 5 2 , and so A = d,X,
B = dzY, C = f dl Y, D = 5 d2X. Write now
(x + z)(x - z ) = (w + y)(w - y).
Hence x + z = dIX, x - z = dzY,
w + y = 5d,Y, w -y = +d,X.
Then the result follows on replacing k w, fy by w,y.
REFERENCE
1. R. D. Carmichael. Diophantine Analysis. John Wiley & Sons, New York
( 19 1 5).
CHAPTER 4
Quartic Equations with only Trivial Solutions
1. We shall consider some quartic equations of the form f ( x 2 ,y 2 , z") = 0

which have only trivial solutions, e.g. those given by x y = 0. Equations of the
form
+
ax4 + bx2y2 cy4 = dz2 (1)
have a long history going back to Fermat and Euler as can be seen from the
second volume of Dickson's "History of the Theory of Numbers". Many of
them can be dealt with by elementary means requiring only the obvious
Lemma
Let k = klk2 be the factorizations of k with ( k l , k,) = 1. Then the general
integer solution of the equation
XY = k Z 2 , (X,Y ) = 1, Z # 0,
is given by
X = +k,P", Y = +k2Q2, Z = kPQ,
where P, Q are arbitrary integers with (Pk,, QkJ = 1.
The really classical equation is that due to Fermat and given as
Theorem 1
The equation
x4 + y4 = z2
has no integer solutions with x y # 0 (and so no rational solutions with xy # 0).
We need only consider the solutions with x 2 0, y b 0,z 2 0 and (x,y ) = I .
Clearly x and y cannot both be odd, and so we need only consider .Y = 1,
y = 0 (mod 2). Hence by ( 6 ) of chapter 3,
x2 = a2 - b2, y2 = 2ab, z = a" + b2,
a 2 b a O , ( a , b ) = l ; ( x , b ) = 1 ; a r 1,b=O(mod2).
Hence x = p2 - q2, b = 2pq, a = p2 + q2,
where p 2 q 2 0, ( p , q ) = I , and p and q are not both odd.
Now y 2 = 2ab = 4pg(p2 + 9").
4. QUARTIC EQUATIONS WITH ONLY TRIVIAL SOLUTIONS 17
Since p , q, p2 + q 2 are relatively prime to each other in pairs,
p=r2, q=s2, p2+q2=t2, r2O,s2O,taO,(r,s)= I,
and so r' + s4 = t 2 .
Also
x = r' - s4, y = 2rst, z = r' + 6r4s4 + .'s
If xy = 0, then rst(r - s) = 0. Clearly r # s and so solutions with x y # 0
correspond to solutions with rs # 0 and vice versa. Hence any solution x, y , z
with xy # 0 leads to a solution r, s, t with rs # 0 and t < i%,since z >
(r4 + s4)2 = t4. The process can be continued and leads to a new solution
r l , sl,tl with rlsl # 0 and tl < qt.The t, tl, etc., form a sequence of de-
creasing positive integers and so we arrive at a stage when t , = 1 and a con-
tradiction arises, and then r,s, = 0.
This method is the so-called method of infinite descent and is due to
Fermat. It has many important applications in number theory. The proof
could have been slightly shortened if we had assumed in the beginning that
.Y, y , z was a solution of equation (2) with least z, but it is often convenient in
some questions to make the descent.
Corollary
The equation
Y2 = x4 +1
has only the rational solutions X = 0, Y = & 1.
For if X = x / y where ( x ,y ) = 1, then Y = z/y2, and
x4 + y4 = 22.
Theorem 2
The equation
x4 - y4 = 22,
(.v, Y ) = 1 (4)
has no integer solutions except x2 = y 2 = 1 and x2 = 1, y = 0.
For these solutions xyz = 0, and we suppose now that xyz # 0. Let
x > 0, y > 0, z > 0 be that solution for which x is least. Clearly x is odd and
then y may be even or odd.
If y G 1 (mod 2), then
x2 = a2 + b2, y2 = a2 - b2, z = 2ab, (a, 6 ) = 1, a > b > 0,
and so x2y2 = a4 - 64.
This is a case of equation (4) and since a < x , we have a contradiction.

If y = 0 (mod 2), then

x 2 = az + b2, y 2 = 2ab, (a, b) = 1, a > 0, b > 0,
and we may suppose that a is even and b is odd. Then
a = 2p2, b = q2, q = 1 (mod 2), ( p , q) = 1, p > 0, q > 0
and so x2 = 4p4 + 94, y = 2pq.
Hence p 2 = rs, q2 = rz - s2, (r, s) = 1, r > s > 0;
and so r = u2, s = u2, (u, v) = 1 , u > u > 0.
Then u4 - 04 = 42.
Also u= l k p < dX,

and this contradicts the definition of x.
Corollary
The only integer solutions of
x4 + y4 = 222, ( x ,y ) = 1
are given by x2 = y 2 = 1.
For x , y are both odd and
Corollary
If we replace x , y by x ~f:y , we see that the only integer solutions of
x4 + 6x2y2 + y 4 = 2, (x,y) = 1 (6)
are given by x2 = 1, y = 0 and x = 0, y 2 = 1.
Using this we now prove
Theorem 3
The only integer solutions of
x 4 - 6xay2 + y 4 = z2, (x,y ) = 1
are given by x = 0 or y = 0.
We may suppose that x > y > 0.
Write (x2 - y2)2 - 4x2y2 = 22,
and so (x2 - y2 + z)(x2 - y2 - 2) = 4x2y2.
We show now that 2 is the greatest common divisor of the two left-hand
factors, We cannot have x = 1, y = 1 (mod 2), for then z2 = - 4 (mod 16).
If x = 1, y = 0 (mod 2), then z is odd and the greatest common divisor
divides 2z, and so is 2. Similarly, if x = 0, y = 1 (mod 2). Both factors are
obviously positive and so
x2 - y 2 + z = 2a2, xa - y 2 -z = 2b2, xy = a b , a > 0, b > O,(a,b) = 1.
Then + bZ,
x2 - y a = a2
and so (x2 + y2)2 = (a2 + b2)2 + 4a2b2
= a4 + 6a2b2 + b4.
Hence ab = 0 from equation (6) and then xy = 0.

Equations of the form
x4 + kX2y2 + y4 = z2, ( x , ~=) 1, (7)
where we may suppose that x 2 0, y 2 0, have played an important part in
the early history of the subject and Euler wrote many papers on the equation.
The problem has usually been to show that the only solutions are given by
xy = 0, and occasionally further solutions x = y = 1. A list of the values of
k for which this is so has been given by Pocklington' who has also found some
general results. Thus there are only solutions with xy = 0 of
(xz + y2)2 - nx2y2 = z2
if n = 1, 5, 7 (mod 8), n - 4 = p2n+1,
p a prime, and if n is not divisible by
a prime = 1 (mod 4). We give a few instances.
k = I.
The equation
x4 + x"2 + y4 = 22, ( x , y ) = 1, (7')
has only the solutions x 2 = 1, y = 0 ; x = 0, y 2 = 1.
Here x and y are of different parities and we may suppose that y is odd
and has the least possible value.
Write 4z2 - (2x2 + y2)2 = 3y4,
or + 2x2 + y2)(2z - 2x2 - = 3y4.

(2z y2)
The two factors are relatively prime for (2x2 + y 2 , 3y) = 1 since y is odd, and
we cannot have x2 = y 2 = 1 (mod 3). Hence, either
2z + 2x2 + y 2 = a4, 22 - 2x2 - y 2 = 3b4, y = ab,
or 22 + 2x2 + y 2 = 3a4, 2z - 2x2 - y 2 = b4, y = ab.
The first pair of equations gives
4x2 = a4 - 2a2b2 - 3b4 = - 4 (mod 16)
and so is impossible. The second gives

4x1 = 3a4 - k2b2 - b4
= (a2 - b2)(3a2 + b2).
Since we may assume that x # 0, and a = b = 1 (mod 2), then
a2 - b2 = c2, 3a2 + b2 = 4d2.
Then u = p2 + q2, b = pa - q2
and p 4 + p2q2 + q4 = d 2 .
This is the same as equation (7') and
p < da,q < di and so p < dy,q < dy,
and this contradicts the definition of y .
We deduce another result. From
(x2 4- xy + y2)(x2 - xy + y2) = z2,
we have X2 + xy + y2 = x2, .P - xy + y2 = Y2
and so 3x2 - Y2 = 2(x + y)2 = 2u2,
3Y2 - x2 = 2(x - y)2 = 2v2,
whence u2 + 3 v 2 = 4Y2, v 2 + 3u2 = 4x2.

These equations can only have the solutions X 2 = Y 2 = 1, i.e. U 2 = 1,
v2 = 1.
k = -1.
The equation
x4 - x"y2 + y4 = 22, (x, y ) = 1
hasonlyrhesolurionsx2= 1 , y = O ; x = 0 , y 2 = 1 a n d x 2 = y 2 = 1 .
Write the equation as
(x2 - y2)2 + xZy2 = 22.
Suppose first that x and y are not both odd.

Then x2 - y2 = a' - b2, xy = 2ab, (a, b) = 1.
Let dl = ( x , b), d2 = ( y , a). Then
x = d,X, b = dlB, y = d2Y, a = d2A, XY = 2AB.
Since (X,B) = 1, and ( Y , A) = 1, then ( X , Y) = (2A, B) or ( A , 2B).
Hence x = 2Ad,, b = dlB, y = d2B, a = d2A,
or .Y = Ad,, b = d,B, y = 2d2B, a = d2A.

Then for the first alternative,
4A2d: - diB2 = d i A 2 - d:B2,
and d:(4A2 + B2) = &(A2 + B2),
and since A2 + B2 $ 0 (mod 3 ) and ( A , B ) = I ,
A' + B2 = C 2 , 4A2 + B2 = D2.
We may suppose that B is odd since, if B were even, we could put B = 2 8 ,
and have a similar pair of equations. Hence B = p 2 - q2, A = P4 and
+
p4 - p2q2 q4 = C 2 . Also p q C a < xy/2, and so the method of descent
applies since p and q are not both odd.
The other alternative gives
A2d: - 4diB2 = d;A2 - d : B 2
and so d:(A2 + B 2 ) = dz(A2 + 4B2).
Now A = p 2 - q2, B = pq and pq < b < x y / 2 and so the method of descent
applies to the product xy.
Suppose next that x and y are both odd.
Then xy = a2 - b2, x2 - y 2 = 2ab, (a, 6 ) = 1 ,
and so a and h are not both odd. Then
Hence ab = 0, x = y giving the solution

x2 = y 2 = 1.
Corollary
On replacing 2x, 2y by x f y , we see that the only solutions of
x4 + 14x2y2 + y4 = z2
aregiuenbyx2 = 1,y2 = 0, I , o r y 2 = 1,x2 = 0, 1 .
An interesting corollary is Fermat's
Theorem 3
There cannot be ,four squares in arithmetical progression.
Let the squares be x2, y 2 , z 2 , w2.
Then x2 + z2 = 2y2, 222 = y2 + w2
and so x2(222 - y2) = wZ(2y2 - z2),

2(x222 - y2w2) = x2y2 - w222.
2+
Put xz = a, yw = 6, 2c = xy + wz, 2d = xy - wz.Since obviously x,y , z , 11

are all odd, c and d are integers.
Then a2 - b2 = 2cd, ab = c2 - d2,
a4 - a2b2 + b4 = (c2 + d2)2.
Hence a = 0, b = 0, 1, etc.
It is easily shown that other values of k, e.g. k = f 3, -4, are also easily
dealt with in a similar way.
Theorem 4
Let a, b, c be rational numbers where ac # 0. Then if there are only a finite
number of rational solutions of
(b2 - 4ac)X4 - 2 b X 2 Y 2 + Y4 = Z2, (9)
there are only a j n i t e number for
ax4 + bx2y2 + cy4 = z2.
These are given by y = X , z = x Y and possibly x y = 0.
If x # 0, replace the second equation by
ax4 + bx2y2 + cy4 = x2z;,
Then ax4 + (by2 - zDx2 + cy4 = 0.
Hence (by2 - z : ) ~- 4acy4 = v2,
say, and this is the first equation with y = X , z1 = Y. Since X = 0, Y = 1 is
a solution of (9), y = 0 is a possible solution of (10).
Take b2 - 4ac = 1 and suppose that the only solutions of
X4 - 2 b X 2 Y 2 + Y 4 = Z2,
are given by X Y Z = 0. Then the only possible solutions of
ax4 + bx2y2 + cy4 = z 2
are given by xy = 0.
Putting b = f 3, ac = 2 in equation (9), the only possible solutions of
2
ax4 f 3x2y2 + - y4 = z2
a
are xyz = 0. In particular if a = 1 and b = -3,
(x2 - y2)(x2 - 2y2) = 22,
and we have ?I = f 1, y = 0 and x2 = y2 = 1.
2. An interesting special case of equation (1) is the equation
ax4 + by4 = cz2, (x, y ) = 1.
Most of the results on the impossibility of non-trivial integer solutions do not
go too far back. They are due to Pocklington l, Nagel12, and Lind3. Many of
them can be proved by a simple application of infinite descent. Among these
are the equations
x4 + dy4 = z2, (x, y ) = 1, y # 0, (1 1)
for the following values of d. Let p be a prime number. Then
I. d = p, p = 7, 11 (mod 16),
11. d =2p, p = k 3 (mod 8),
111. d = 4p, p = k 3, - 5 (mod 16),
IV. d = -p, p = 2 3 , -5(mod 16).
From Theorem 4, it follows that equation (12) is also impossible when d is
replaced by -4d where now d is defined above. Hence 111 follows from IV.
Let us take 11,
x4 + 2py4 = z2, > 0,y > 0, > 0.
A congruence mod 8 shows that y f 1 (mod 2), and so x =z E I (mod 2).
From
(2 x2)(z T x2) = 2py4,
since the two factors have greatest common divisor 2, we have either
x2 = 2a4, z T x2 = 16pb4, y = 2ab,
or z k .x2 = 2pa4, z T x2 = 16b4, y = 2ab,
The first set gives
a4 - 8pb4 = kx2.
The - sign can be rejected, and then
(az + x)(a2 - X) = 8pb4
gives either
a2 +x = 2pc4, a2 -x = 4d4, b = cd,
or a2 +x = 2c4, a2 -x = 4pd4, b = cd.
The first of these gives
a2 - 2d4 = pc4,
and this is impossible since (2/p) = - 1.
The second gives

c4 + 2pd4 = a2,
and since d < di < dy, the method of descent applies.
The second set above gives
pa4 - 8b4 = &x2.
A congruence mod 8 shows that the + sign must be rejected. The - sign
must also be rejected since (2/p)= - I .
We now take IV,
x4 - py4 = 22, x > 0, y > 0, z > 0.
I f y = 1 (mod 2), then x = z = 1 (mod 2) and x4 = - 2 , 4 (mod 8). Hence

y = 0 (mod 2), x = z = 1 (mod 2). Then
(x2 f z)(x2 T z) = py4,
and so x2 f z = 2a4, x2 T z = 8pb4, y = 2ab,

or x2 & z = 2pa4, x2 T z = 8b4, y = 2ab.
The first gives
a4 + 4pb4 = x2.
Here a = 1 (mod 2) and so
x T a2 = 2pc4, x & a2 = 2d4, b = cd,
d4 - p c 4 = +a2.
The - sign must be rejected since d4 k 3c4 = - 1 (mod 8) is impossible.

The + sign gives a descent since c 6 b < y.
The second gives
pa4 + 4b4 = x2.
Since a = x = 1 (mod 2), a congruence mod 8 shows that this is impossible
if b = 0 (mod 2). Hence b = 1 (mod 2), and
p + 4 = x2 (mod 8).
This is impossible unless p = - 3 (mod 16). But then,
x & 26' = c4, x T 2b2 = pd4, a = cd.
and so &4b2 = c4 - bd4.
A congruence mod 16 shows that the - sign is impossible. For the + sign,
the method of descent applies since cd = a < y.
Sometimes the application of biquadratic residues may be useful. Let p be a
prime = 1 (mod 8) and so p = a2 + b2 where we may suppose a is even and
so a = 0 (mod 4). Then 2 is a biquadratic non-residue or residue mod p
according as a E 4 (mod 8) or a E 0 (mod 8). We have now
V. The equation
x4 - py4 = 222
has no integer solutions if 2 is a biquadratic non-residue of p .
Let z = 2'_7,, z , = 1 (mod 2).
Then
by the law of quadratic reciprocity and since (2/p) = 1. Hence z = s2 (mod p )

and then x4 E 2s4 (mod p ) , and this means 2 is a biquadratic residue.
It may be remarked that it is well known that solutions of equation (11)
can be found when d has special values, e.g. when d = u(02 - u), a solution
is given by the identity
(02 - 244 + u(oz - 4(244 = (v4 + 4uu2 - 4u2)2.
3. A different type of result is given by5
A = U,
Pa
bl
4 ' c1
where k3 = k, = I (mod 8), are taken from a finite set and M, ivl are integers.
The equation F = k3w2is impossible since it gives
x2 + y 2 + z2 + w2 = 0 (mod 8),
and so x = y = z = w = 0 (mod 2).
Suppose next that A is even. Then besides equation (13) we have also
F = 2k5w2, F, = 2k6w:. (14)
We consider the first of these; clearly w f 0 (mod 2) as this would give
x2 + y2 + z2 = 0 (mod 8), x = y = z = 0 (mod 2).
Hence w = 1 (mod 2) and
x2 + y 2 + z2 = 2 (mod 8).
This gives three possibilities typified by x = 0 (mod 4), y = z = 1 (mod 2).
The equation F, = 2k6w7 gives the three corresponding congruences
b, + c, = 2w: (mod 8), c, + a, = 0 (mod 2),
a , + 6 , = 0 (mod 2)
all of which are impossible from IV.
We show now there are no solutions with F, = 0. Then G = 0 and
X2
- Y2 -
-
22
b,r - c1q CIP - a,r a1q - b,p
Since (x, y , z ) = 1, we have with integer d,
b,r - c,q = dx2, c,p - a,r = dy2, a,q - b l p = dz2.
Then A = a(b,r - c,q) + .. .
= d(ax2 by2+ + CZ) = dF,
and so the odd factors of d and F are = 1 (mod 8).
Also F = -x2 - y2 - z2 (mod 8).
We can exclude x = y = z = 1 (mod 2) since then F = - 3 (mod 8) and so
all the factors of F cannot be = 1 (mod 8). Also x = 1, y = 0, z = 0 (mod 2)
can be excluded since then F = - 1 (mod 4). Similarly x = 0 (mod 4),
y = z = 1 (mod 2),since now F = b +
c (mod 8) = -2 (mod 8). Finally let
x = 2 (mod 4), y = z = 1 (mod 2). Then there are the three possibilities
Fl = 4u, + b , + c,, a, + 46, + c,, a, + b, + 4c, (mod 8)

= 4 , 6 (mod 8), = 1 (mod 2), = 1 (mod 2),
and so F, # 0.
We can now deduce
Theorem 6
There are no integer solutions of
br - cq cp - ar aq - bp
(15)
where
and the positive odd factors of these three terms are all = 1 (mod 8).
To reduce equation (12) to this form, we take k , = k2, and
b,c + bc, = 2qr, c,a + ca, = 2rp, a,b + bla = 2pq.
Then bca, = -aqr + brp + cpq, etc.
and so a,, b,, c, will be integers if 111 is satisfied.
The equation (12) now takes the form
(aa, - p2)x4+ . . . = 0,
or a(pb - aq)(pc - ar)x4 + . . . = 0.
On replacing x by (qc - rb)x/a, we have equation (15).
(-aqr + brp + cpq)/bc . -
Now -A = a . . I
P . .
also aqr brp CP9
A = 2 1 1 1 ,
pbc qca rub
Hence all the odd factors of A are = 1 (mod 8).

We have now to show equations (13) and (14) have no integer solutions, and
we need only consider (14). We write F, = 2klwI in the form
(uqr + brp + cpq) (;:+ :+ 3rz

- -
-
-
2 x2 -y2 + -
+ bpr CP9 z2) = 2k4wT,
ca ab
and from F = 2k,w2, this is

(qr + rp + pq)w2 + qrx2 + rpy2 + pqza = w: (mod 8). (16)
Since from equation (14)
x2 + y 2 + z2 = -2w2 (mod 8),
one of x , y , z must be even, say x , and then x = 2 (mod 4) y = z = w = 1
(mod 2), etc.
Then (16) becomes
5qr + 2rp + 2pq = w f (mod 8).
Similarly if y = 2 (mod 4), etc.
5rp + 2pq + 2qr = wl (mod 8),
5pq + 2qr + 2rp = MI: (mod 8).
These are all impossible if p = 0 (mod 8), qr = 1, 3, 7 (mod 8).
We now examine the conditions that the odd factors of
L = -qc - rb 1 M=- ra - pc i N = -Pb - qa

a C
should be = 1 (mod 8). We take q = r = -1 (mod 8) arbitrarily such that

the odd prime factors of (qc - rb)/a are = 1 (mod 8). We take p = 0 (mod 8)
so great that
b Y = p c - ra > 0, c Z = p b - qa > 0.
On putting p = 8bcP + PI,
say, where PI is a value of p = 0 (mod 8) giving integer values for Y and 2,
then
Y = 8c2P + P2, Z = 8b2P + P,,
say, where P2 = P, = 1 (mod 8).
Our problem now is to find P such that Y > 0 , Z > 0, and YZ is divisible
only by primes = 1 (mod 8).
Many numerical instances have been given me by Mr. K. Kloss. For
example, when a = 7 , b = 15, c = 23, we can take
P 4 r L M N
8280, 4991, 13335, 12176, 6473, -3881,
8280, 16583, 15855, -20512, 5297, -353,
11040, 3703, 14175, 18208, 10313, -6073.
+
These equations L3x4 + M 3 y 4 N3z4 = 0, which have n o solutions, can-
not be proved impossible by taking congruences mod 16 since A4 +N =
0 (mod 16). Cassels points out that the impossibility of the three equations
follows at once by taking congruences t o mod 761, 353, 6073 respectively.
One would have thought that it should not be difficult to find numerical in-
stances for which congruences considerations d o not suffice to show the im-
possibility. Cassels however finds this is not so for a large number of values
for P, 4, r.
REFERENCES
I . H. C. Pocklington. Some diophantine impossibilities. Proc. Camb. Phil. SOC.,
17 (1914), 110-118.
2. T . Nagell. Sur Iirnpossibilite de quelques tquations a deux indeterminkes.
Norsk Mat. For. Skriftern I NO. 13, Kristiania 1923.
3. C. E. Lind. Untersuchungen uber die rationalen Punkte der ebenen kubischen
Kurven vorn Geschlecht Eins. Diss. Uppsala, (1940).
+
4. L. J. Mordell. The diophantine equation x4 my4 = z2. Q.J. Math., (2) 18
1-6 (1967).
5 . L. J. Mordell. Some quartic diophantine equations of genus 3. Proc. Am. Math.
SOC.,17 (1966), 1152-1158; also 18 (1967), 190.
2+
CHAPTER 5
Some Linear Equations
1. We now discuss some linear equations.
Theorem 1
Integer solutions of the equation
a,x, + a2x2 + . + a,x,
' '
= a, (1)
where the a's are integers, a1a2,. ,a, # 0, exist i f and only f
(a19 a29 * * ., an) I a.
This condition is obviously necessary and so we may assume that
(a,, a2, . . ., a,) = 1.
Suppose first that n = 2. If a = 0, the general solution is x, = --a2t,
x2 = alt where t is an arbitrary integer. If a # 0, the equation is equiva-
lent to
alx, + a2x2 = a, (a,, a2) = 1, (2)

where we may suppose that a, > 0, a2 > 0 on writing -x2 for x2 if need be.
The solution is given by the Euclidean algorithm for finding the greatest
common divisor of a, and a2. We may write
a2 = qla, + rl where 0 < rl < a,, (rl, a,) = 1.
- r l x Z + aa
Then x1 = -91x2
-q1x2 + -r,xza1a1
= -91x2 + x3,
= -91x2 x3,
Continuing this process, we have a decreasing set of positive integers

r,, r 2 , . . . until we come to a stage where
rn-2x,-1 + r,,-3x, = a, rn-2= 1.
5. SOME LINEAR EQUATIONS 31
Then x,- ,,x , - ~ ,. . . , x2, x1 are given successively in terms of a parameter x,.
If one solution ( p , q ) is known, all the solutions are given by
.=
i p + ta2, y = q - ta,,
where t is an integer parameter.
This is obvious since
+
al(x - p ) a2(y - q ) = 0.
2. The argument when n = 2 can be adapted for the general equation (l), say,
L ( z ) = alxl + . . . + anx, = a. (3)
We may suppose that a, is one of the a, for which Jar[is least. We apply a
substitution
x1 = X , kzxz + +
. . . k,x,, +
where the k are integers satisfying
lalkr + all < lull, ( r = 2 , . . ., n).
Suppose now that in the new form, a2 plays the role of a,. A substitution
X2 = X2 + 1 1x1 + 13x3 + . . . + lnXn
makes the new coefficients all have moduli < la,[. Continuing this process, we
arrive at a unimodular integral substitution (since X,, X2, . . . are easily given
in terms of x,, xz,. . .), which changes L ( z ) into, say,
L(z) = qx,+ . . . + e,xn,
where the E = 0 or -t 1, and then finally to a stage where, say, el = 1,
e2 = 0, . . ., E, = 0. Then X , = a and the general solution is given linearly in
terms of the n - 1 variables X,,. . ., X,.
3. Theorem 2
Let L,(z) = allxl + . . . + arnx,,( r = 1, 2,. . . , n), be n linear forms with
integer coefficients and determinant D > 0. Then the equations
Lr(z)= a, (r = 1 , 2 , . . ., n) (4)
have an integer solution if and only if the congruences
L,(z) = a, (mod M ) (r = 1 , 2 , . . ., n) (5)
are solvable for all M .
The condition is necessary. For the sufficiency, we reduce the L,(x) to
canonical forms given by the
Lemma
A unimodular integral substitution changes the Ls into the forms
~11x1~
,21x1 + ~ 2 2 x 2 , .. .) CnlXI + * * * + Cnnxn,
where cii > 07 0 Crl, crz, . . 3 cii-1 < cii.
Let c,, be the greatest common divisor of the coefficients ofL,. Then L, may
be transformed into c,,X, by a unimodular integral substitution. This substi-
tution changes L, into c,,X, + L,(X,, X,, . . ., X,) where L2 is a linear form
in X,, . . ., X,. This can be changed into czaX, where c, is the greatest
common divisor of the coefficients of La. Then replacing X , by X , kzczlXl +
if need be, we may suppose that 0 Q c,, < c,,. This process can be continued
and so the lemma follows. It is also clear that cllcaa.. . c,, = D .
We now continue with the proof of the theorem. We can replace the
equations by
cllX, = a,, cZlXl + czaX2 = a,, . . . .
Since cl1X, = a , (mod cl,) is solvable, a, = 0 (mod cL1),and so the equation
for X , has an integer solution. We eliminate X , by writing
51CllXl + 5 , ( C , l ~ , + CZZX,) = 5101 + 52%,
and taking (,ell + facZl= 0. Since the equation is solvable as a congruence
to mod cllcZafor all 5,, Say X , is an integer.
Similarly we can eliminate X , , X , from the first three equations, etc.
It may be noted that the Theorem 1 can be extended to the case when the
number m of equations is < n . When all the right-hand a are zero, we can
find an estimate for the magnitude of a solution.
4. Theorem 3
The m simultaneous equations
L, = a,lxl + arzx, + * . . + arnx, = 0, (6)
(r = 1, 2, . . . , m), ( m < n) where the a's are integers, have an integer solution
5 # 0, for which the x satisfy the inequality
Ix,[ Q ( A 1 A 2 . ..A,)l'(,-rn), (7)
where Ar = lar1l + lar21 + . + larnl.
*
Let the x run through the integer values in the interval [0, N ] where N is an
integer defined later. Then if B, is the sum of the positive a,,, ar2,. . ., a,, and
- Cr is the sum of the negative ones,
-C,N Q L, Q B,N.
Write A , - B, + C,. Then L, assumes at most ArN + 1 values. Hence if
( N + 1)" > n (A,N + l),
m
r=l
the Lr take the same set of values for two different S , say 5' and z". Then
c = z' - x'' is a solution of L, = 0, (r = 1,2,. . ., n), and Ix,I Q N for
s = 1, 2 , . . ., n.
5. SOME LlNEAR EQUATlONS 33
An estimate for N is given by
+ 1) n A , ,
m
(N + 1) > (N
r=1
or
We can take N t o be the integer part of n (A,)l(n-m).

rn
r=1
It is sometimes useful t o know that linear forms can be made t o satisfy

various inequalities for integer values of the variables. This is a problem in the
geometry of numbers and a fundamental result due t o Minkowski is given by
Theorem 4
n
Let L, = 2a,,x, (r = 1 , 2, . . ., n) be
s=1
n linear forms with real coeficients
and determinant A = Ila,,))# 0. Let 11, I, , . ., I,, be n real numbers such that
Ill2.. .I,, 2 In(.
Then integers x # 0 exist such that
lLrl < I,, (r = 1, 2 , . . ., n).
Further any n - 1 of the equality signs may be omitted,
CHAPTER 6
Properties of Congruences
1. Congruences play an indispensable role in the discussion of Diophantine

equations. They can usually be considered as Diophantine equations in
which the variables are elements of a finite field. We have seen in Chapter 2
that many equations
f(4= f(X1, x2, . . ., x,) = 0
have been shown to have no solutions by proving the impossibility of a
congruence
f(z) = 0 (modp).
There are, however, many important applications of congruences which
give affirmative results. Thus solvability of congruences and the number of
their solutions are essential in applying the methods of analytical number
theory, not only in proving the existence of solutions of an equation but also
formulae for their number. Further, in recent years, it has been shown that
there seems to exist for cubic equations f(x, y ) = 0 a close connection be-
tween the number of solutions off(x, y ) = 0 (modp) and the existence of
rational solutions off(x, y) = 0.
It seems worth while to give some results on congruences required later but
also of interest in themselves.
Theorem 1
Let f(x) be apolynomial in one variable with integer coefficients and let p be a
prime. Suppose there exist x1 and 6 2 0 such that
f(xl) = 0 (modp26+1), dfdx,

(x,) = 0 (modpd), f 0 (modpd+l).
Then for s 2 1, x exists such that

f ( x ) E 0 mod^^^+^), x = x1 (modpd+). (1)
We prove the result by induction. It is true for s = 1 with x = xl. We prove
that if the result is true for s with x = xs, then it is true for s + 1. We write
x = x,+ = x, (modpdtS)and have to solve
f ( 4 = f(xs) + (x - xs)f(xs) + (x *! x2J.(xs)
- + . . . = 0 (modp2d+S+1).
6. PROPERTIES OF CONGRUENCES 35
Put x - x, = X P ~ +thenf(x,)
~; = Ap where A f 0 (modp), since
f(x,) = f(x,) + (x, - xl)f(xl) (modpdtl).
Hence AX + f ( x , ) / ~ =~ 0~ (modp),
+~
and this has a solution for X . We notef(x) = 0 (mod pa)>,but not (mod pd ).
The solvability off(x) = 0 (modp) for all integers r 3 1 can be expressed
by saying that f ( x ) = 0 has a root, say x = X , in the p-adic field Q,. This
means that the solutions of f(Xr) = 0 (mod p) can be written in the form
X, = xo f x,p + . . . + xr-lp,-l,
where the x are integers with 0 6 x < p and where, say,
X+, - X , = xrp.
It proves convenient to mention the solvability off(x) = 0 in real numbers,
and then we can say that f(x) = 0 is solvable in the p-adic field Q,, .
Example
.x2 = a,
where a is ap-adic unit, i.e. (a, p) = 1. Ifp is an odd prime, the equation has
a solution in Q, if and only if a is a quadratic residue of p , i.e. (alp) = 1.
If p = 2, the condition is a = 1 (mod 8).
Theorem 2
Let f(z) = f(xl, x2 . . . , x,) be a polynomial with integer coeficients in n
ijariables and of total degree 1 < n. Then if the congruence
f(S) = 0 (mod p),

p a prime, has one solution, it has another.
In particular if there is a solution .z = .c0, we can always suppose xo = 0
on putting x = X + .yo.
The theorem is due to Chevalley l. For the proof, we recall Fermats result
that if x is any integer then x p = x (mod p ) . It follows that if f ( x ) is a poly-
nomial in x such thatf(x) = 0 (mod p) for all x, thenf(x) is divisible mod p
by x p - x, and so identically,
f(x) = .fl(x>(xP- x) (mod PI,
wherefl(x) is a polynomial in x with integer coefficients, and the coefficients
of the powers of x on both sides are congruent mod p.
Suppose now that g(s) is a polynomial such that
g(z) = g(xl, x2, . . . , x,,) = 0 (mod p)
for all integers xl, x2, . . ., x,. On replacing xf by xl, xf+ by x:, etc., we may
suppose that no x occurs in g(5) to a power > p . Then considered as a poly-
nomial in xl,
g ( s ) = (x? - xl)gl(Xz, * * * 5 xn) (mod P)
= (x! - X I ) . . .(x: - x,)g,
on continuing the process, Clearly g, is a constant.
Suppose now that f(~) is a polynomial such that f ( z ) $ 0 (modp) except
when x1 = x2 = . . . 3 x, E 0 (modp). Then the congruence
1- (f(~))p-~ = (1 - x f - l ) . . .(l-x:-l) (modp)
holds for all Z. It is obviously true for cz = 0 since f ( z ) = 0. For the other 2,
(f(s))P-' = 1 , and at least one of the right-hand factors is ~ 0The . right-hand
side is of degree (p - 1)n in the variables and so we have a contradiction if
( p - l)n > ( p - 1)1, i.e. n > 1.
An illustration is given by
alx: + . . . + a,xf, = 0 (modp),
where (ala2.. .a,,p) = 1, n > 1. The congruence always has a solution
different from z = 0.
2. The case 1 = 2 is of special interest and we give a different discussion for
f ( x ,y, z ) = ax2 + by2 + cz2 = 0 (modpl), (x, y , z, p ) = 1. (2)
W e prove
Theorem 2
Suppose that a, 6, c are relatively prime in pairs, and so abc # 0,and are
square free. Then the congruence
ax2 + by2 + cz2 = 0 (mod M )
is solvablefor arbitrary M with (x, y , z , M ) = 1, i f i t is solvablefor M = 4abc.
This implies that integers A , B, C exist such that
bA2 + c = 0 (mod a), cB2 + a 3 0 (mod b), aC2 + b E 0 (mod c).
Suppose first that p is an odd prime and p'llM. There are two cases ac-
cording as ( p , abc) = 1 or p. Suppose first that r = 1. Then the congruence
is solvable when (p, abc) = 1. For, write
x2 = -ba-'y2 - ca-'za (modp).
Take z = 1 and let y run through the values 0, 1, . . .,p - 1. Then the right-
hand side runs through 1 + ( p - 1)/2 = (p + 1)/2 values, and so one of
these must be either zero or a quadratic residue ofp. Since c f 0 (modp), we
cannot have both ax = 0, by = 0. Hence Theorem 1 applies for r > 1 on
considering f ( x , y , 1) as a function of x or y.
Suppose next that p I abc. For simplicity, we suppose that ( p , bc) = 1,
and p 1 a. Then
by2 + cz2 = 0 (mod p ) .
+
This is solvable with y z 0 (mod p ) if and only if ( - b c / p ) = 1. Then
Theorem 1 applies on considering f(1, y , 1). If, however, y = z E 0, the
congruence (2) is impossible for r = 2.
Suppose next that p = 2. We may take r 3 2 since the congruence is
solvable for r = 1. For simplicity, we suppose that at most one of a, 6, c is
even and that this is not divisible by 4.
Case I. a, b, c are all odd. Then the congruence (2) is solvable for r = 2 un-
less a = b = c (mod 4). If this is not so, then we may assume that a =
- b (mod 4 ) and so a solution is given by x = y = 1 (mod 2), z = 0 (mod 2).
A solution is then given when r = 3 by taking z = 0 (mod 4) when a + b =
+
0 (mod S), and z = 2 (mod 4) when a b = 4 (mod 8). By Theorem 1, the
congruence is solvable for r > 3.
Case II. One of a, 6, c is even, say a = 2 (mod 4). Then the congruence is
solvable unless either b + c = 4 (mod 8) or b + c 3 a (mod 8). Suppose
first that a =_ 2 (mod 8). Then a solution when r = 3 requires y E z I
+
I (mod 2), and then x = 0 (mod 2) if b c E 0 (mod 8), but x = 1 (mod 2)
+
if b c = -2 (mod 8), i.e. b + c = - a (mod 8). The result for a =
- 2 (mod 8) follows in the same way. Solvability for r > 3 follows by applying
the theorem to the congruence applied as a function of y.
3. The congruence
ax3 -I- 3bx2y + 3 c x y 2 + dy3 = z3 (mod 9).
This has only the trivial solution x =y =z E 0 (mod 3) if
I. a f 0, f k I (mod 9), d $ 0, f 2 1 (mod 9),
11.
fl - a3
3
+ c - ad2 ? ( b - a 2 d + -
- d 3 ) $ 0 (mod 9).
3
We have
z = ax + dy (mod 3), z3 = (ax +d ~(mod
) ~9).
Then
(a - a3)x3 + 3(b - a2d)x2y + 3(c - ad2)xy2 + (d - d3)y3= 0 (mod 9),
and
- a3 d - d3
a
3
x3 + (b - a2d)x2y + (C - ad2)xy2 +- 3
y 3 = 0 (mod 3).
From I, x y 3 0 (mod 3), and so x2 = y 2 = 1 (mod 3), and

- a3
a
3
x + (b - a2d)y + (c - ad2)x + 3
y = 0 (mod 3).
Since x E -t y (mod 3), the condition for insolubility becomes
- a3
a
3
+ c - ad2 f ( b - a2d + 7
- f 0 (mod 3). d3)
Corollary
The congruence
ax3 + 3bx2y + 3cxy2 + dy3 = 1 (mod 9)
has no solutions if 1 and 11 hold with the omission of a f 0 and d f 0.
The congruence
ax3 + 3bx2y + 3cxy2 + dy3 = 1 (mod 7 ) ,
is impossible if to mod 7 , I and either I1 or 111 holds.
1. a f &I, d $ fl,
11.
b2 + c(d f a -
7
111. brO, CEO, d k a f 0 , fl.
From I, x y f 0. Hence x3 = k 1 , y3 = k 1. Then we can write the con-
gruence with X = x / y , 2 = l / y as
aX3 + 3bX2 + 3cX + d Z 3 = 3 E,
and so 3bX2 + 3cX + d ? a - = 0. E
If b f 0, this will be impossible if the quadratic character

( 9 ~ ' - 12b(d _+ Q -
= -1,
7
or
(c2 + b(d7k u - E) )=-I
If h = c = 0, the congruence is impossible if none of the four conditions

d _+ a _+ 1 E 0 are satisfied.
4. It is often of importance to find estimates for the magnitude of the solutions
of congruences. It will suffice to take a particular case2 given by
6. PROPERTIESOF CONGRUENCES 39
Theorem 3
Let L,, L,, . . ., L, be n homogeneous linear forms in x, y , z with integer
coeficients. Let ql, q2, . . . , qn be any npositiiie integers and rlr r,, r3 any positive
numbers such that r1r2r3 2 q1q2.. .qn. Then the congruences L, = 0 (mod qs),
(s = 1, 2, . . . , n) halie a non-trivial solution, i.e. not (0,0, 0), such that
1x1 < r1, IYl < r2r I Z I < r3.
Consider the sets of integers (x, y , z ) lying in the intervals

0 < x < rl, 0 < y < r2, 0 < z < r3.
Denote by [r] the integer part of r. Write r' for [r] when r is not an integer,
and for r - 1 when r is an integer. There are
+ Lrl1)(l f rh)(l + rh)
> r1r2r3
sets of integers x, y , z lying in the intervals, and so there are more than
qlq2.. .qn sets of residues for L, (mod qs),(s = 1,2, . . . , n). Hence at least two
different sets ( x ~y,,
, zl) and (x,, y,, z2) have the same residue set, and so the
differences x = x1 - x, y = y , - y,, z = z1 - z, are not all zero. For
these we have
Ldx, Y , z ) = 0 (modq,), (s = 1 , Z . . ., n ) , 1x1 < rl, lyl < r2, IzI < r3.
In Chapter 2, some equations were proved impossible by congruence con-
siderations. These involved congruences of the form
a,x? + a,xi + . . . + anxk = 0 (modp),
p a prime, for which it was shown that at least one of the x, say x,, is divisible
byp. I f this holds for an infinity of primesp, then only solutions with x1 = 0
are possible. This possibility can be ruled out from3
Theorem 4
The congruence
f(z)= a,x? + a,x; + . . . + anxf:= 0 (mod p ) ,
(3)
where p is a prime, a1a2.. .an $ 0 (mod p ) always has a solution with
x l x z . . .x, f 0 (mod p ) i f
( p - l),-lp-,'Z > f1f,. . . I , ,
There is no loss of generality if we suppose that
f , > O ,...) / , > O , l l I p - 1 )..., l n l p - l .
Denote by N ( a ) = N(a,, a,,. . ., a,) the number of solutions of (3) with
xlx2. . .x, f 0 (mod p). Write e ( X ) = e ( 2 ~ i X / p ) Then
. the number of
solutions can be written in the form
since the sum is zero if.f'(z) f 0 (mod p ) , and is p iff(;.) = 0 (mod p ) . Here
2' means that no x is zero.
Write this as
t=1 2
Sum now for each a with 0 6 a < p . The sum is zero except when the x, y
satisfy
t,x? = t2y:,
I . . .) t,xf: = fay>, (modp). (6)
Denote by M the number of solutions for tl, f a , cc, y . Then 11, t2 can take at
most ( p - 1)2 sets of values, and x can take at most I values for given y.
Hence
M < ( p - 1)2+nlll2.. .I,.
For each solution, the sum for the a gives p".
Hence 2 IpN(a) - ( p - l y y < p y p - 1)2+nll12.. .I,.
(a)
On replacing x1 by fix,, . . ., x, by fnxnr we see there are at least

( p - 1)"/1112.. .I, congruences with the same N(a) if a1a2.. .a, $ 0 (modp),
since f ' assumes only ( p - 1)/1 values for f = 1,2, . . . ,p - 1. Hence
and so IpN(a) - (p - I)"[ < ~ " / ~-( 1)1112.

p . .l,
Hence N(a) # 0 if
( p - 1)" > p y p - l)1112. . .I,,
and so if
( p - l)n-lp-n/2 > &I2.. .I,.
Theorem 5
The congruence
f(z) = alx? + . . . + a,xf: + a = O(modp),
where p is a prime, aala2. . .a, f 0 (modp) always has a solution with
x1x2.. .x, f 0 if
( p - 1)n-1'2p-n/2 > 1112.. .I,.
The proof proceeds as for Theorem 4 except that now t, = t2 (mod p ) . Then
M +
< ( p - l)l+VIl2.. .I,,, and the exponent 2 n is replaced by 1 + n.
A result4 required in Chapter 14 for the integer solutions of the equation
+
z3 = a.y2 + by2 c is given by
5. Theorem 6
The congruence
y2 = x3 + k (mod a),
when a is odd, has a solution with ( y , a) = 1 except when a = 0 (mod 7),
k = 6 (mod 7).
It suffices to consider the case a = p' where p is an odd prime and r is a
positive integer. If there is a solution (xo,yo) with ( y o , p ) = 1 when r = 1,
then by Theorem 1 there is a solution for r > 1 .
Three cases arise, p = 3, p = 2 (mod 3), p = 1 (mod 3) when r = 1. When
+
p = 3, since x3 = x (mod 3), y 2 3 x k (mod 3) and this has a solution
y = 1.
When p = 2 (mod 3), the congruence x3 = y 2 - k (modp) is solvable for
arbitrary y.
When p = 1 (mod 3), we can use an estimate for N the number of solutions
of the congruence given by Theorem 4, but there will be less numerical detail
if we use the fairly deep estimate5 given by Hasse and Davenport,
IN - pI < 2dp, or N > p - 2dp.
Then i f p 2 13, N > 4 and so there will be at least two different residues of
x3 (modp), and so there exists a value of y f 0 (modp). The only prime
p < 13 is 7, and it is easily verified that a solution y f: 0 (mod 7) exists except
when k = 6 (mod 7).
When a is even and a/2' is odd, there are solutions with ( y , a) = 1 if there
are solutions of y 2 = x3 + k (mod 2') with y odd. When r = 1, we can take
y = I . When r = 2, we have solutions if k $ 3 (mod 4) since we can take y
odd. When r = 3, we have solutions with y odd if k $ 3, 5, 7 (mod 8). This
result holds for r 2 3.
REFERENCES
I . C. Chevalley. DCmonstration d'une hypothese de M. Artin. Abh. Math. Sem.
Hamburg, 11 (1935), 73-75.
2. L. J. Mordell. On the equation ax2 + by2 - cz2 = 0. Monatshefte fiir Math.,
55 (1951), 323-327.
3. L. J. Mordell. The number of solutions of some congruences in two variables.
Math. Z . , 37 (1933), 193-209.
4. L. J. Mordell. Note on cubic equations in three variables with an infinity of
integer solutions. Annuli mat. puru uppl. (4) 29, (1949), 301-305.
5. H. Davenport and H. Hasse. Die Nullstellen der Kongruenzetafunktionen
in gewissen zyklischen Fallen. J . Reine Angew. Math., 172 (1934), 151-182.
CHAPTER 7
Homogeneous Equations of the Second Degree
1. Let
be an indefinite quadratic form with integer coefficients and of determinant

D = IIursll # 0. We discuss the solvability in integers of the equationf(f(m) = 0.
Results for n = 3 were given by Legendre for the diagonal forms, i.e.
a23 = a31 = a,, = 0, and various modifications of his proof have been given
by other writers. Results for diagonal forms in four and five variables were
found by Meyer nearly a hundred years after Legendre's work. The general
case was considered by Minkowski and Hasse' culminating in the latter's
result given below as Theorem 1 , and embodying a new and important
principle, namely, that solvability of the equation f(x) = 0 for all local (i.e.
p-adic) fields implied global solvability (i.e. solution in integers).
Other results give estimates for the magnitude of a solution in terms
of the coefficients. For n = 3 , the first sharp results are due to Holzer,. His
method is rather advanced, and so further on we shall give the less sharp re-
sults due to Mordel13 which can be proved by elementary means. These were
also found later by S k ~ l e mMordel15
.~ has since found a very elementary proof
of Holzer's results also given further on. An estimate sometimes better than
Holzer's and depending upon results in quadratic fields was given by M.
Kneser ".
Estimates for the magnitude of a solution of the general quadratic equa-
tion have been given by Cassels7, Davenport8, Birch and Davenport', but
they have no place here. These results are of great importance in many
investigations.
Theorem 1
The equation f(x) = 0 is solvable in integers if and only if the congruence
f(x) = 0 (modp') is solvable for all primes p and integers r 2 1 with
(x1, x2,. . ., xn, P) = 1.
Stated otherwise the conditions imposed uponf(s) mean that the equation
f(5) = 0 is solvable in every p-adic field including of course the real field pa.
The condition is obviously necessary.

It is well known that by means of a linear substitution
+ +
X, = c,,xr . . . c,,x,, (r = 1 , . . ., n),
7. HOMOGENEOUS EQUATIONS OF THE SECOND DEGREE 43
where the c are integers and c,, # 0, that the solution of the equationf(.c) = 0
is equivalent to that of an equation
F(%) = A,Xf + . .. + AnXZ = 0. (2)
This is easily proved by induction. We may suppose that a,, # 0 by a linear
substitution if need be.
Then aIIf(4 = X ? f fib),
where
Similarly
h22f1(.7.) = X; + S2(T),
2 b2rxr, etc.
n
where X, =
r=2
It can be shown that the substitution does not affect p-adic solubility.
2. The case n = 2 is trivial and so we commence with n = 3 and write
f ( x ,y , Z) = ax2 + by2 + C Z ~= 0. (3)
We now may suppose that a, 6, c have no common factor, and also that
x, y , z have no common factor. We also suppose that a, b, c do not all have
the same sign and thatf(x, y , z ) 3 0 (mod p') is solvable for all primes p and
integers r >, 1 with (x, y , z , p) = 1.
We now reducef(x, y , z ) to a canonical form. Firstly, we may suppose that
a, 6, c are square-free. For let a = U , C X ~b, = b,p2, c = c,y2 where a,, b,, c1
are square-free. We have now a 1-1 correspondence between the solutions of
equation (3) and of
m,,z1) y1, = alx? + b,Yf + clz? = 0
given by
n,x, = ax, cllyl = py, d,z, = yz, (x1, y,, Zl) = 1,
d.Y = pyx,, fly = yay,, dz = apz1, ( x , y , z ) = 1.
y , z ) = 0 is p-adically solvable so is f,(x, y , z ) = 0 and con-

Clearly if ,/(s,
versely.
Next we may suppose in equation (3) that (6, c) = (c, a ) = (a, b) = 1.
For let p be a prime factor of (b, c) and so b = pbl, c = p c l . Then ax2 =
0 (modp), and so ,Y = p x l . Hence with a , = pa, (3) becomes
a,s: + b,y2 + c,z2 = 0.
Here alblcl = abc/p and so lalblcll< label. Clearly the process, applied to
all the prime factors of (6, c), (c, a), (a, b), must terminate at a stage when
(6, c) = (c, a) = (a, b) = 1. Further the final equation is solvable as a con-
gruence for all p'.
We now write equation (3) in the form
ax2 + by2 - CZ' = 0, u > 0, b > 0, c > 0. (4)
Let p be an odd prime divisor of a. Then by2 - cz2 = 0 (mod a) is solvable
and so as in Mordel13 and Skolem4,
by2 - cz2 = b(y + kz)(y - kz) (modp),
say, and so f(x, Y , 4 = LlM1 (mod P),
identically in x, y , z where L1,M 1are linear forms in x, y , z.
Similarly if q, r are odd prime divisors of 6, c, respectively.
f(x, y , 4 = L2M2 (mod 41,
f(x, y , z ) = L3M3(mod r).
Also ax2 + by2 - cz2 = (ax + by + C Z ) ~(mod 2).
Hence since abc is square-free, on taking into account the various p, q, r, we
have by the Chinese remainder theorem,
f ( x , y, z ) z L M (mod abc),
where L,M are linear forms.
We show now that we can find integers x, y, z # 0, 0, 0, such that
ax2 + by2 - cz2 = 0 (mod abc), (5)
and 0 < 1x1 < dbc, 0 6 (yI < d/ca, 0 < IzI < d/ab,
provided we exclude the trivial case a = b = c = 1. Here the numbers
dbc, dG, dab are irrational. The number of sets of integers x,y, z with
o<x<d/bC,o<y< z/Ca,0~~<2/'abis~(I+[d~])>~(db
abc. But L has only abc residues mod abc, and so for two sets of values for
x,y , z, say, xl, yl,z1 and x2, y2, z2,
Uxi, yi, 21) L(x2,~ 2 z2)
, (mod abc).
Hence L(xl - x2, yl - y2, z1 - z2) = 0 (mod abc),
and Ixl - x21 < dbc, Iyl - y21 < dG, Izl - z21 < dab.
Then for x = x1 - x2,etc.
-abc < axa + by2 - CZ' < bbc,
and so ax2 + by2 - cz2 = 0 or abc.
In the second case,

(ax2+ by2)(z2 + ab) = c(z2 + ab)2,
or a(xz + by)2 + b(yz - ax)2 = c(z2 + ab)2.
i.e. a x 2 + bY2 = cz2.
This proves the solvability of equation (4).

The result gives an estimate for the magnitude of the solution, but a better
one is given by Mordells
Theorem 2
There exists a non-triaial solution with
1x1 ,< d S C , lyl < d%, IZI < 2dab. (6)

This is not so precise as Holzers2 result 1x1 < 6, (yl < di,IzJ < dab,
where the equality signs can be omitted unless two of a, b, c, are equal to one.
It suffices for a proof of this to show that we can find a solution not (0, 0,O)
of the congruence
f ( x , y , z) = ax2 + by2 - cz2 = 0 (mod 4abc),
with 1x1 < dZC, Jyl < dZa, IZI < 2dab.
Then If(x, y , z)l < 4abc, i.e. f(x, y , z) = 0.
We do this by applying Theorem 3 of Chapter 6 to five linear congruences.
Suppose first that a, b, c are all odd. Then there exist integers A , B, C such
that
bA2 - c = 0 (mod a), - cB2 + a = 0 (mod b), aC2 + b = 0 (mod c).
Three of the congruences are given by
y - Az = 0 (mod a), z - Bx = 0 (mod b), x - Cy = 0 (mod c).
Then from the first,
ax2 + by2 - cz2 = (bA2 - c)z2 = 0 (mod a).
Similarly for mod 6, mod c.
We can satisfyf(x, y , z) = 0 (mod 4) by taking two linear congruences each
to mod 2. We cannot have both a - c G 2 (mod 4), b - c = 2 (mod 4)
for then a = b = - c (mod 4), and sof(x, y , z ) = 0 (mod 4) is not solvable.
We may suppose that h - c = 0 (mod 41, and then the two congruences are
x = 0 (mod 2), y = z (mod 2).
It now suffices to take

rl = d/2bc, r2 = .\/2ca, r3 = 2 4 2 .
in Theorem 3 of Chapter 6.
Suppose next that one of a, b, c is even, say a. We can now make
f(x, y , z ) = 0 (mod 3abc) if x, y , z satisfy the three congruences:
y - Az = O(mod+a), z - Bx = O(modb), x - Cy = O(modc).
We now satisfy f(x, y , z ) = 0 (mod 8) by taking two linear congruences one
to mod 2 and the other to mod 4.
These are
x = 1y (mod 2), z = m y (mod 4),
where I, m satisfy
a12 + b - em2 = 0 (mod 8).
+
If a b - c = 0 (mod 8), we take I = m = 1. If b - c = 0 (mod 8), we
take I = 0, m = 1. Then as before, Theorem 3 of Chapter 6 gives a solution.
The equation is impossible when b - c = 4 or a (mod 8 ) as follows on taking
congruences mod 8.
This finishes the proof.
To summarize, we have proved
Theorem 3
rfa, b, care square free, (a, b) = (b, c) = (c, a ) = 1 , anda, b, c do not have
the same sign, then the equation
ax2 + by2 + cz2 = 0
has non-trivial integer solutions if and only if - bc is a quadratic residue of a,
and so for every prime factor of a, i.e. x2 + bc = 0 (mod a) is solvable, and
similarly for - ca and - ab, and if
ax2 + by2 + cz2 = 0 (mod 8)
is solvable.
These give N , say, necessary and sufficient conditions. It can be shown that
if any N - 1 of these conditions are satisfied, so is the remaining one. This is
a consequence of the laws of quadratic reciprocity and the supplementary
laws, namely, that if p , q are two different odd primes
and
There is, however, no need here to go into details. The result is well known
from Hilberts theory of the norm-residue symbol.
Theorem 4
If one integer solution not (0, 0,O)of the irreducible equation
f ( x ,y , z ) = ax2 + by2 + cz2 + 2fyz + 2gzx + 2hxy = 0 (7)
exists, then the general solution with ( x ,y , z ) = 1 is given by a j k i t e number of
expressions of the form
x = Alp2 + Bipq + Clq2, y = A2p2 + B 2 ~ 4+ C2q2,
z = A3p2 + B3pq + c3q2, (8)
where p , q take all integer values with ( p , q) = 1, and the A , B, C are integer
constants.
For let a solution (x,, yo, z,) be known. Write
x = rx, p, +
y = ry, q, + z = rz,,
where p , q, r are rational numbers. Then we have a linear equation in r from
which r = f(p, q, O)/L(p,q ) where L is a linear function of p and q. Hence
ignoring denominators, we have with an appropriate factor 8,
8 . ~= f l ( P , q),
8Y = f X P , q), 8,z = f 3 ( P , q),
where fi,f2,f3 are binary quadratics with integer coefficients, and we may
now suppose that p , q are integers and that ( p , q ) = 1. A simple elimination
procedure shows that the common factors of f l ,f 2 , f 3are independent of p , q.
These and S can be removed on replacing p , q by a linear substitution on
P. 4.
The method of Theorem 4 has been used by Mordel15to give an elementary
proof of Holzers
Theorem 5
The solvable equation ax2 + by2 + cz2 = 0 taken in the canonicalform with
a > 0, b > 0, c < 0 has a non-trivial solution with
1x1 I dm, IyI IdW, IZI I dZ
It is easily seen that all the equality signs may be removed unless two of
the a, b, c are unity.
Let (x,, yo, z,) be a solution with (x,, yo, z,) = 1. We show that if for
this solution, ( z o ( > dab, then another solution (x, y , z) can be found with
IzI < lzol. Hence there exists a solution ( x , y , z ) with IzI I dab. The
inequalities for x , y are now obvious. Put
x = x, + tx,
y = y, tY, + z = zo + tz,
where X , Y, 2, are integers to be determined later. Then
(ax2 + b Y 2 + cZ2)t + 2(axoX + byoY + czoZ) = 0.
1
Hence, on neglecting a denominator, a solution, say, (x, y , z), is given by
6~ = zo(aX2 + 6 Y 2 + cZ') - 2Z(axoX + by0 Y + CZOZ),
ax = xo(aX2 + b Y2 + - 2X(axoX + by, Y + C Z O Z ) ,
CZ2) (9)
Sy = yo(aX2 + 6 Y 2 + cZ') - 2 Y(uxOX + by0 Y + CZOZ),
where S is a common divisor of the three expressions on the right.
Further x , y , z are integers if
Ale, 6l(Xyo - Yxo),
For from
ax; + by; + cz; = 0,
it easily follows that (S, abxoyo) = 1. From (9) it suffices to show that
P = a x o X + 6yoY = O(mod6), Q = a X 2 + b Y 2 = O(mod6).
Since X = xo Y/yo, P = Y(axi + by;)/yo = 0 (mod 8);
also Q E (ax; + by:) Y z / y ; E 0 (mod 6).
From (9)
Suppose now that z i > ab. Take X , Y as a solution of y o X - xoY = 8.

First let c be even. Take 6 = +c, and Z so that
Then
Hence on continuing the process we have a solution with z2 I ab. Secondly

let c be odd. Now impose the condition
aX + b Y + CZ E 0 (mod 2).
This defines the parity of 2. Since 6 is odd, the three expressions on the right-
hand side of (9) are divisible by 26, and so we now have (10) with 6 replaced
by 26. Take 6 = c and Z with assigned parity so that
Then (1 1) is replaced by
This completes the proof.

If, I
2 - <1+1 and
3. Quaternary quadratic equations'O.

By means of a linear substitution, we may take the equation in the form
.f(.z) = alx: + a2xg + a3xz + a4xi = 0,

where the a's are integers.
Theorem 6
The equation f(z) = 0 is sohable in rational integers i f and only i f f ( x ) =
0 (mod p') is soh>ablefor all primes p and,for ail r with (xl,x2, x3, x4,p) = 1,
anrlj(.r) = 0 is soloable in real numbers.
We may suppose that the a are square free since if a, = A1a? etc. where A ,
etc. are square free, a 1-1 correspondence is established between the solutions
of
5
r=l
arx; =o, $ A , X ; = 0,
r=l
by the relations
X, = . .,
alxl,. x1 = a2a3a4X1.. .
We may also suppose as in the case of the ternary quadratic that no three of
the a's have a common factor.
Since the coefficients a,, a2,a3,a4 do not all have the same sign, we may
suppose that a, > 0, u4 > 0. Write
g = alx? + a2x& h = -a3xz - a4xi.

We shall prove the theorem by showing the existence of an integer a repre-
sentable by both g and h. This will follow from the p-adic solvability of
f ( z ) = 0.
Let p,, p2,. . . ,p s be all the different odd primes dividing a1a2a3a4.Then
for everyp from among pl, . . . ,ps,and forp = 2 there is a p-adic solution of
a14? + a& + a,(; + a,(: = 0.

We may suppose in this solution (ti,(h, &, 6;) that none of the 6' are zero.
For if = 0, the p-adic equation is solvable for arbitrary f4. This is obvious
on replacing (,, t2,f 3 by t f i q-,, t & + +
T ~t G
, +
73, where the 71 are
arbitrary, and this gives a linear equation in t.
Write
6, = a,[? + a24X = -a3[; - a45i.
We can choose the solution 5 so that b, # 0 in the p-adic field Q,. For if
6 , = 0, then both g and h represent any integer in Q, since g and h
factorize in the p-adic field. We may also suppose that b, = 0 (mod p%) but
not p A p + l, and also when p = 2. Consider the congruences with a > 0,
a = 6 , (mod a = b, (modpApP1), p =pl, . . . ,ps.
These determine a uniquely mod m = 2A2+3p:1+l. . .p;$+ Since b, f 0
(mod p A p P + l ) b,a-
, = 1 (modp). Hence the quadratic character (b,a-l/p) =
1, and so b,a- is a quadratic residue in Qpand so is ap-adic square. Similarly
b,a- = 1 (mod 8) and so b,a-l is a quadratic residue in Q2 and so is a
2-adic square. Since b, and a differ by a square factor in all the fields Q,,
Q,, both the forms
-ax: + a,X: + a,XZ, - a x : - a3XE - a4Xi
represent zero in all the Q, and Q,. These forms have a zero in every p-adic
field, except possibly when p I a. Also the forms are indefinite since a, > 0,
a, < 0. Hence if a has only one prime factor q different from 2, p , , . . .,ps, the
forms will also have a zero in QQ.Then from the ternary case of Theorem 1,
we have
a = ale? + a2c& a = -a3c32 - a&,
with rational c, and so finally
a,cl + a2c: + a3cf + a4ci = 0.
The existence of such an a is proved by Dirichlets theorem that there are
an infinity of primes in an arithmetical progression in which the first term is
prime to the difference. Let a be an arbitrary one of the a defined by the
congruences above; write d = (a, m). Then there exists an integer k 2 0 such
that a/d + km/d is a prime q. We take a = a +
km = qd.
We now state without proof the conditions for the p-adic solvability of the
equation
f ( x ) = alx: + a,x$ + a3xi + a 4 4 = 0,
wheref(x) is in the normal form, i.e. the a are square free integers and no
three of the a have a common factor.
The a of course must not all have the same sign. The p-adic equation is
solvable for all p when (p, a,, a,, a3, a,) = 1, as is obvious from the result
for the ternary quadratic. Next let p = p , , be an odd common prime factor
of a, and a, for which ( -a3a4/p)= 1. Then we must have (( -a,a2p-,)/p) = 1.
Similarly for the other prs.
Finally 2-adic solvability is required. This is easily seen to be the solvability
of f ( x ) = 0 (mod 8). The conditions are as follows.
+
If a1a2a3a4= 1 (mod 8), then we require a, -k a, + a3 a, 3 0 (mod 8).
There are no conditions if
ala2a3a4= 2, 3, 5, 6 , 7 (mod 8).
Suppose next that
ala2a3a4= 4(mod 8) and a, = a, = O(mod2),
say. Then if
~a,a,a3a4= I (mod 8),
we require
+a, + j a , + a3 + a4 = j(agaz - 1) (mod 8).
No conditions are required if
~a,a,a,a, = 3, 5, 7 (mod 8).
We prove now the
4. Theorem 7
Every indefinite quadratic form j ( x ) in fiiie iiariables represents zero.
It suffices to take
f(z)= a1x: + ' .. + a&,
where the a are square free, a, > 0, us < 0.
Write
g = alx: + a&, h = -a3xg - a4xi - a&.
We argue as in the case of four variables. We find by Dirichlet's theorem a
positive rational integer a which is represented by both g and h in the real
field, and in all p-adic fields Q, with the exclusion perhaps of a field Q, where
q is an odd prime number not dividing any of the coefficients a,, a2, . . . . Then
g represents a by the theory of the ternary quadratic. Also since h represents
zero in Q,, it represents a rationally by the theory of the quaternary quadratic.
The theorem follows as for n = 4.
It is now trivial that an indefinite form f ( x l , x2,. . ., x,) with n > 5
variables represents zero. It is also easily proved that solutions exist with
XlX,. . . .Y, # 0.
5. A much more difficult question is to investigate the solvability of two
simultaneous homogeneous quadratic equations f ( 2 ) = 0, g(z) = 0 in n
variables. Mordell" has shown by an application of Meyer's theorem that
+
subject to some simple natural conditions, for example, that Af(s) p g ( z ) is
not a definite form for all real values of A, p not both zero, that non-trivial
solutions, i.e. not all zeros, exist if n 2 13. Swinnerton-Dyer12 has improved
this to n 3 11. It is conjectured that n = 9 is the best possible result, but some
equations in 9 variables are such that four of the variables must be zero.
An interesting question is whether the two simultaneous equations in five
variables are solvable when they are solvable in every p-adic field includingp,.
REFERENCES
1. H. Hasse. Uber die Darstellbarkeit von Zahlen durch quadratische Formen
im Korper der rationalen Zahlen. J. reinp angew. Math., 152 (1923), 129-148;
also 205-224.
2. L. Holzer. Minimal solutions of diophantine equations. Can. J. Math., 11
(1950), 238-244.
3. L. J. Mordell. On the equation ax2 + by2 - cz2 = 0. Monatshefte fur Math.,
55 (1951), 323-327.
4. T. Skolem. On the diophantine equation ax2 + bya + cz2 = 0 . Rendiconti di
Matemuticu e delle sue applicazioni (9,11 (1952), 88-102.
5. L. J. Mordell. On the magnitude of the integer solutions of the equation
ax2 + by2 + cz2 = 0 . J. number theory, I (1968), 1-3.
6. M. Kneser. Kleine Losungen der diophantischen Gleichung ax2 + bya = cz.
Abh. Math. Sem. Hamburg, 23 (1959), 163-173.
7. J. W. S. Cassels. Bounds for the least solution of homogeneous quadratic
equations. Proc. Camb. Phil. Soc., 51 (1955), 262-264. Addendum, ibid. 52
(1956), 664.
8. H. Davenport. Note on a theorem of Cassels. Proc. Camb. Phil. SOC., 53
(1957), 539-540. Addendum, ibid. 52 (1956), 664.
9. B. J. Birch and H. Davenport. Quadratic equations in several variables. Proc.
Camb. Phil. SOC., 54 (1958), 135-138.
10. Z. I. Borevich and I. R. Shafarevich. Number-theory, Chapter I. Academic
Press, New York and London (1966).
1 1 . L. J. Mordell. Integer solutions of simultaneous quadratic equations. Abh.
Math. Sem. Hamburg, 23 (1959), 124-143.
12. H. P. F. Swinnerton-Dyer. Rational zeros of two quadratic forms. Acta
Arith., 9 (1964), 260-270.
CHAPTER 8
Pell's Equation
1. Theorem 1
Let D be a positive integer which is not a perfect square. Then the equation
y2 - D,v2 11 (1)
has an infinity of integer solutions. If (.Y, y ) = ( U , T ) where T > 0, U > 0 is
the solution with least positive x, all the solufions are gioen by
XY dD = f(T + UdD)", (2)
where n is an arbitrary integer.
We ignore the trivial solution y = k 1, x = 0 given by n = 0.
The proof of ( I ) follows easily from a result on Diophantine approximation
given by
Lemma 1
Let 8 be an irrational number and q > 1 an arbitrary positive integer. Then
there exisi integers x and y such that i f L = y - x0,
ILI < l / q , 0 < .Y 6 q. (3)
Let x take the values 0, I , 2, . . ., q and let y be such an integer that 0 6 L < 1.
Then q + 1 values for L arise lying in the q semi-open intervals
Hence two of the values of L corresponding to say ( x l , y,) and (x2. y2),where
s1 # x2, say x1 > x2, lie in the same interval and so
lY, - J'2 - (x1 - .Y2)0l < I/q.
Then (3) follows on putting y = y 1 - y,, x = x1 - s2.On replacing (3) by
ly - ~ 8 1< I/x,
it follows that there are an infinity of integer solutions of this inequality.
Lemma 2
A number m = m(D), e.g. m = 1 + 240, exists such that
Iy2 - Dx21 < m,
for an infinity of integers, x,y ,
3+
Take 0 = dD in (3) and so integers (x,y ) exist such that

ly - XdB1 < l/lxl.
Then I y + XdZI = I y - XdB + 2xdBI
< 2lxldB + l/Jxl,
and so Iy2 - Dx21 < 2d0 + 1/x2
< 2dB + 1.
We now deduce the existence of an integer solution of equation (1).
There exists an integer k such that Ikl < m and
y2 - Dx2 = k
has an infinity of integer solutions. We may suppose there are two, say
(xl,Y,) and (x2.y2), such that
xz = x1, Yz = Yl (mod k), (x2,Y2) # ( - & , -YJ.
From a
y1 - Dx: = k, y i - D x ~= k,
we have by multiplication,
( ~ 1 -~ D ) ~D(yix2 - ~2x1) = k2.
2 x1~2-
Write ~ 1 -~D x2 ~ x= ~k Y , ~ 1 x 2- ~ 2 x 1= kX.

Clearly X , Yare integers, X # 0 and
Y 2 - D X 2 = 1.
We now deduce an infinity of solutions. Let (x,y ) = (U,T ) where U > 0,
T > 0, and U is the least value of X . Then an infinity of solutions (x,, y,,) with
x, > 0, yn > 0 are given by taking
y, + x,d/D = (T + UdO), y, - x,dB = (T - UdB),
where n is any positive integer.
All such solutions are given by these formulae.
For suppose (x,y ) is a solution not so given. Then for some positive integer n,
(T + U f i ) < y + x a < (T + U d D ) n + l .
Then 1 < ( y + x d B ) ( y n - xndB)< T + u ~ B .
Write ( y + x d B ) ( y , - x n d B ) = Y + XdB.
and so Y+ X O <T+ UdZ, Y 2 - D X 2 = 1.
+
Since Y XdB > 1, and 0 < Y - X d / o < 1, then X > 0, Y > 0. This
contradicts the definition of T, U.
8. PELL'S EQUATION 55
The solutions with
s < 0, y < 0 are given by y + x d / D = -(T + UdB).,
x < 0, y > 0 are given by y + x d D = (T + Ud/O)-.,
x > 0, y < 0 are given by y + x d / D = -(T + UdD)-",
where n is a positive integer.
Corollary
I f d is a giren integer, there exists an injinity of solutions with x = 0 (mod d ) .
This is obvious from Y 2 - D d 2 X 2 = 1.
In the study of the units of quadratic fields, say Q ( d d ) ,a Pellian equation
takes the form
y 2 - dx2 = 4.
If now (x, y ) = (u, t ) , u > 0, t > 0 is the solution with least x, then it can be
shown similarly that the general solution is given by
y + 2xd/d = & ( t + u d d) '
where n takes all integer values.
The solution of the equation
y 2 - Dx2 = - 1 (4)
is a much more difficult question and simple explicit conditions for solvability
are not known. A necessary condition is that D is not divisible by 4 or by any
prime = 3 (mod 4). It is easily proved that the equation is solvable when
D = p is a prime = 1 (mod 4).
For let (x, y ) = (U,T ) be the fundamental solution of y 2 - px2 = 1.
Then U = 0, T = 1 (mod 2). Write
T+l T - 1
-.-=
2 2
Then either
T-+ 1 T - 1
- 2 - Pa2, -- - b2,
2
T
--+1 T- 1
or -a2,
= ph2,
2 2
where a, h are integers. The second set gives a2 - pb2 = 1 , and contradicts
the definition of the fundamental solution.
The first gives

b2 - pa2 = - 1.
Then an infinity of solutions is given by
Y + X d i = (b + adi)2n+1,
where n is an arbitrary integer.
The fundamental solution U,T above is obviously given by
T + Ud; = (b + ad;)z.
Pells equation arises very naturally in the study of the units in a quadratic
field say Q(l/d), where d > 0 and is square free. We have the known results
for the units E ,
d = 2 , 3 (mod 4), E =y + xdd, y2 - dx2 = 1,
1 + l/d
d = 1 (mod4), E=y+-
2 xy
y2 + xy + 4
1-d
x2 = f 1,
where x and y are rational integers. Both equations are included under the
form Y 2 - dX2 = f 4. The fundamental unit e0 corresponds to the solution
with smallest positive X and Y. Of course when d = 2, 3 (mod 4), eo =
T + U q d . The result for d = 1 (mod 4) can be proved in a similar way.
The study of the class number of the field has led t o the
Conjecture (Ankeny, Artin, Chowla)
Let p be a prime = 1 (mod 4) and so now
Y2 -px2 = -4.
Let u be the least positive value of X . Then u $ 0 (mod p).
This has been verified for p < 2000 when p = 5 (mod 8), and for p <
100,OOO by Dr Goldberg whenp 5 1 (mod 8). It has been shown by Mordelll
whenp = 5 (mod 8), and by Chowla whenp = 1 (mod 8), that u f 0 (mod p)
if and only if
%-1m f 0 (modp),
where the Bernouilli numbers are defined by the expansion
There is also the further
Conjecture
Let p be a prime = 3 (mod 4) and so now
Y2 - px2 = 1.
Let U be the least positive calue of X . Then U $ 0 (mod p).

8. PELLS EQUATION 57
This has been verified by K. Goldberg for p < 18,000. It has been shown
by Mordel12 that U $ 0 (modp) if and only if the Euler number
4,-3,,4 f 0 (modp),
where Enis defined by
There is a similar conjecture when p = 1 (mod 4).

2. Theorem 2 (Gauss)
Given one integer solution (x,,yo) of the equation
f ( x , y ) = ax2 + bxy + cy2 + Ci.r + ey + f = 0, (5)
where the coeficienrs are integers, there exists an injinity of integer solutions if
D = b2 - ~ U >
C 0,
is not a perfect square, and if the discriminant
A = 4acf + bde - ae2 - cd2 - fb2 # 0.
The equation can be written as
Y2 - DX2 = M,
where M # 0 since A # 0, and
Y=px+qy+r, X=sy+t,
say, where the coefficients are integers. From
Yz - 0x2, = M , T 2 - DU2 = 1,
where now T, U is any integer solution of Pells equation,

(TYo - - D(TX0 - UYo)2 = M .
Then new values of X , Yare given by
X = TXO - UYO, Y = TYO - DUX,.
These give values of x,y satisfying
px + qy + r = T(pxo + qy, + r) - DU(syo + t ) ,
sy + r = T(syo + t ) - U(pxo + qyo + r). (6)
There are an infinity of solutions of Pells equation with U = 0 (mod p s ) , and
so T 2 = 1 (modps).
We may suppose T = 1 (mod ps), for if T $ 1 (mod p s ) , we need only take

a new solution given by
Tl + U,d/o = (T + U d D ) ' , or Tl = T 2 + DU2, Ul = 2TU.
Then p x + qy + r = p x , + qy, + r (modps),
+ t = sy, + t (modps).
sy
This gives y = yo (mod p), p(x - x,) + q ( y - yo) = 0 (mod ps), and so an
integer value for x. Hence equation (6) is satisfied by integer values o f x , y.
If A = 0, the equation ( 5 ) is reducible. Then either both linear factors have
rational coefficients, and so at least one linear factor is zero for an infinity of
integer values of x, y , or both linear factors have coefficients in the quadratic
field Q(d3). Then there exists only the solution given by X = 0, Y = 0.
If D is a perfect square or if D < 0, there are obviously only a finite number
of solutions.
There are many interesting applications of Pell's equation to Diophantine
equations.
Theorem 3
The equation
x3 + y 3 + z3 + w 3 = n (7)
has an infinity of integer solutions i f there exists one solution (a, b, c, d ) such
that
-(a + b)(c + d ) > 0,
is not a perfect square, and a # b, or c # d.
Put x=a+X, y=b-X, Z = C + Y, w=d-Y.
Then (a + b)X2 + (a2 - b2)X + (c + d ) Y 2 + (c2 - d2)Y = 0.
Write
(a +b)(X+ y)2
+ + (c d)( Y + - d)a
- (a + b)(a - b)2 - (c + d)(c - d)' = 0.
4 4
The result now follows from equation (1) since
+
D = -(a b)(c + d), and (a b)(a - b)2 + + (c + d)(c - d)2 # 0
unless a = b, c = d.
It is very surprising that this is the only known method of proving the
existence of an infinity of integer solutions of the general equation (7).
Another application of Pell's equation is given in Chapter 13 where it is
shown that the equation
a(x3 + y 3 ) + b(z3 + c3) = 0,
where a, b, c are integers, has an infinity of integer solutions other than
x+y=o.
3. Of special interest is the application of the Pel1 equation to the discussion
of the integer solutions of the simultaneous equations
+ bxlxz + C X ~= d,
UX: (8)
a1x: + b,XlX, + clx; = d1Z', (9)
where the constants are integers. We may suppose that none of the left-hand
sides are perfect squares, and that
b2 - 4ac = D >0
is not a perfect square.
These equations include as particular cases many equations of current
interest.
We first of all note that the equations have only a finite number of integer
solutions. For from Chapter 7, the general solution of equation (9) with
( x l , x2, z ) = 1, is given by a finite number of expressions of the form
x, = 1,x2 + m,xy + n,y2, (r = 1,2),
where the I, m, n are integer constants and x , y are integers. On substituting
in equation (8), we have a finite number of binary quartic equations of the
formf(x, y ) = A say. By a theorem of Thue in Chapter 20, there are only a
finite number of integer solutions for x,y. To find these is in general a diffi-
cult problem, and most results require detailed and complicated proofs.
However, A. Baker has recently found estimates (very large ones, indeed) for
the magnitude of the solutions.
The Pellian method of approach is to note that the general solution of
equation (8) can be written in the form
x = pu + qu, y = ru + sv,
where thep, q, r, s refer to a finite set of rational numbers, and u, v are integer
solutions of a Pellian equation. Then from equation (9), we have a finite set of
equations
z 2 = Au' + BUV+ CV',
where A , B, C are rational numbers.
Suppose for simplicity that the Pellian equation takes the form
U' - Dv2 = 1,
and denote by uo, vo the fundamental solution. Then the general solution is
given by
u + udD = (u, + u,d/D)".
Write a = uo + vodO, p = u, - voz/O, ffp = + 1,
Then 2u = an + B", 2vdB = ffn - B", z = z,,
and z," = Ala2" + B1F2"+ Cl( f 1)".
Hence we are led to the following
Problem
Let a sequence u, be given by a recurrence formula
+
u , , ~ au,+l + bun = c, (1 1)
where the a, b, c are given integer constants. When is u, a perfect square ?
We give some applications to the Fibonacci sequence u,, where u, = 0,
u1 = 1, and
un+2 = un+1 + un,
and to the Lucas sequence v,, where uo = 2, u1 = 1, and
Theorem 4
v, = x2 only when n = 1, 3, fx = 1,2,
v, = 2x2 only when n = 0, + x = 1, 2.
Theorem 5
u, = x2 only when n = 0, f 1,2, 12, fx,= 0, 1, 12.
u, = 2x2 only when n = 0, f 3 , 6 , and fx = 0, 1,2.
Proofs for u, = xa were given about the same time by Cohn4 and Wylie5.
We give here the later proof by Cohn6 who also proved the other results.
We need some properties of u, and v,. From equation (12), on taking the
residues of u, mod 4, we have a periodic sequence of order 6 given by
2,1,3,0,3,3,2,1, . . .
and so v, = 0 (mod 2) only when n 3 0 (mod 3), and uk = 3 (mod 4) when
k = i-2 (mod 6). We see that v, = 0 (mod 3) only if n 5 2 (mod 4), since
the residues of 21, mod 3 form a periodic sequence of order 8 given by
2 , 1 , 0 , 1 , 1 , 2 , 0 , 2 , 2 , 1 , 0 ,....
We note that
Hence
L n= (-I),vn, u-, = (-l)"-lUn.
Also (P,, u,) = 1 if n $ 0 (mod 3),
(rn, u,) = 2 if n = 0 (mod 3),

follows from
We easily deduce the relations

2Um+n = Umun + Unum, 211m+n = SUmUn + VmUn,
uz, = u; + 2(- 1y-1, Ugm = umum.

We denote hereafter by k an integer positive or negative for which k =
+ 2 (mod 6). We have now two results fundamental for the proofs of the
theorem,
P,,+~,,.~ = (- l)fzin(mod v,),
u ~ = (-
+ 1)bn
~ (mod
~ uk).
~
For 211,+2k = 5UnU2k + VnC'Zk
= 5U,UkU, + un(v2 - 2)
= - 2 e , (mod uk).
The first result above follows since rk is odd.
so 2 L l n + 2 k = unz'Zk + uZkz'n
= U,(Lf - 2) + UkL'kU,
= - 214, (mod D~).
The second result now follows.

We proceed to the proof of the theorems.
First for = 2. There are no solutions if n = 2m is even since
* 2.
13,
l'zm = 17;
We show there are solutions n = 1, 3, s = 1, 2 if /I is odd,

3*
Write n E c (mod 4), c = 1, 3. We exclude n = 1, 3 and write with k > 0,

+
n = c 2.3'k, where r 3 0, k = f 2 (mod 6 ) and when c = I , u, = 1 and
when c = 3, u, = 4.
Then V, (- 1)3'V, G - V , (mod vk)
= - 1, - 4 (mod vk).
Since vk = 3 (mod 4), ( v n / v k ) = - 1 and so v, cannot be a square.
Take next v, = 2x2. There are now only the solutions n = 0, _+6 and
+ x = 1 , 3. Suppose first that n is odd. Since
v; = 5u2, - 4,
then x4 = 5(3un)2 - 1.
A congruence mod 8 shows that this is impossible.

Suppose next that n is even. Take first n = 0 (mod 4). Then excluding n = 0,
we can write
n = 2.3'k, k = _+2(mod6).
Hence 2V, -2Uo (mod Uk)
- 4 (mod Vk).
and so v, # 2x2.
Take next n = 6 (mod 8), n # 6. Then we can write
n = 6 + 2.3'k, k = 2 (mod 6).
Hence 2V, = -2V0 = -36 (mod vk),
and so v, # 2x2.
Take finally n = 2 (mod 8). Since v - , = v, and -n = 6 (mod 8), v, = 2x2
only if -n = 6 (and so also n = 6).
We come next to Theorem 5. Now u, = x2 only for n = 0, k 1,2, 12 and
+ x = 0, 1 , 1 , 12.
Suppose first that n = 1 (mod 4). Since u1 = 1, we exclude n = 1, and
write
n = 1 + 2.3'k, k = + 2 (mod 6).
Then U, -U1 - 1 (mod Vk),
and so u, # 2.
If n = 3 (mod 4); u - , = u, and so u, = x2 only if -n = 1 (and so also if
n =1).
Suppose next that n = 2m is even. Then uzm = umvm = x2.
If m = 0 (mod 3), u, = 2y2, v, = 2z2.
The second gives only m = 0, - 6 , and m = - 6 gives no value for y.
If 171 $ 0 (mod 3), u, = y2, u, = z 2 . The second gives only m = 1, 3 and
in = 3 gives no value for y.
In a similar way it can be shown that if u, = 2x2, then n = 0, & 3 , 6 and
fx = 0, 1, 2.
Of special interest is the application of such ideas to the equation
dy2 = ax4 + bey2 + C,
which reduces to a particular case of equations (8), (9).

Write z = 1 and
dy2 = ax4 + bx2z + cz2.
Then x2 and z are given by a finite number of expressions of the form
IX2 + rnXY + nY2.
Alternatively,
4ady2 = (2ax2 + b)2 + 4ac - b2,
and so 2ax2 + b = Xyn + pa".
where y, 6 are of the form uo f uod/D,
and A, p are a finite set of conjugate constants in Q(d3).
We have then an
equation of the form (10).
Some simple illustrations due to Cohn6 are given now, but some of the
results are due to Ljunggren9 and Mordel18.
The equation
y2 = 5x4 +1
has only the solutions x = 0, rt2.
For
and then 2x2 = U6n.

The equation
y2 = 5x4 - 1
has only the solution fx = 1.
Here y + x2d3 = (2 + d 3 y n - 1 = (+1 + d3 6n-3

,
and then 2x2 = U6n-3.
The equation
y 2 = 5x4 +4
has only the solutions +x = 0, 1, 12.
y + x W 3+ d3)n 1 + 45 2n
Here
2 (
= -
= ( 7 , and )
x2 = uZn.
The equation
y2 = 5x4 -4
has only the solutions & x = 1.
Here , and x2 = u ~ , , - ~ .
2
The preceding results arose from the discussion of the case a = 1 of the
more general sequence
unt2 = aunt1 + un.
Cohn7 has discussed this in detail and has found results similar to those
given here. In the course of his investigation, the equation
2y2 = 3x4 - 1
arises. His method could not prove that the only integer solutions are given by
x = 1, 3. This was however, proved by Bumbylo by an extension of Cohn's
method. For now with a = 5 2 4 6 , +
x2 =
an + al-"
l + a
The proof depends upon an application of the law of quadratic reciprocity in
the quadratic field Q(d/-2).
Other methods of dealing with quartic equations are given in Chapter 28.
REFERENCES
1. L. J. Mordell. On a Pellian equation conjecture. Acta Arithmetica, 6 (1960),
137-144.
2. L. J. Mordell. On a Pellian equation conjecture. J. Lond. Math. SOC.,36
(1961), 282-288.
3. L. J. Mordell. On the four integer cube problem. J. Lond. Math. Soc., 11
(1936), 208-218. Addendum 12 (1937), 80. Corrigendum 32 (1957), 383.
4. J. H. E. Cohn. On square Fibonacci numbers. J. Lond. Math. SOC.,39 (1964),
537-540.
5. 0. Wylie. Solution of the problem. In the Fibonacci series Fl = 1, Fa = 1,
+
F,, + = F,, F,,- I, the first, second and twelfth terms are squares. Are there
any others? Am. Math. Monthly, 71 (1964), 220-222.
6. J. H. E. Cohn. Lucas and Fibonacci numbers and some diophantine equations.
Proc. Glasgow Math. Assoc., 7 (1965), 24-28.
7. J. H. E. Cohn. Eight diophantine equations. Proc. Lond. Math. SOC., 16
(1966), 153-166. Addendum, ibid, 17 (1967), 381.
8. PELLS EQUATION 65
8. L. J. Mordell. The diophantine equation y2 = Dx4 + 1. J. Lond. Math. Soc.,
39 (1964), 161-164.
9. W. Ljunggren. Some remarks on the diophantine equations x2 - Dy4 = 1,
X* - Dy2 = 1. J. Land. Math. SOC., 41 (1966), 542-544.
10. R. T. Bumby. The diophantine equation 3x4 - 2y2 = 1, Math. Scan., 21
(1967) 144-148.
CHAPTER 9
Rational Solutions Derived from Given Ones
1. Problem
Given a rational solution of f ( x , , x2,. . ., x,) = 0, where f is a polynomial
with rational coeflcients, to deduce other solutions.
Suppose first that the equation represents a curve, say f ( x , y ) = 0. We refer
to a rational solution ( x , y ) of f ( x , y ) = 0 as a rational point P on the curve
f(x, y ) = 0. If we write the equation in the homogeneous form F ( X , Y, 2) = 0,
where x = X/Z, y = Y/Z, we still refer to rational points ( X , Y, Z ) on the
curve F(X, Y, Z ) = 0 when the ratios X / Z , Y / Z are rational. We have a 1-1
correspondence between (x, y ) and ( X , Y, 2) except when 2 = 0. We can,
however, say this corresponds to an infinite rational point (x,y).
Theorem 1
Let f ( x , y ) = 0, g(x, y ) = 0 be two curves of degrees r, s. Zfrs - 1 of their
intersections are rational points, then the remaining intersection is also a
rational point.
On eliminating y , x is determined by an equation of the rsth degree with
rational coefficients. Since rs - 1 of its roots are known, the remaining one is
determined by an equation of the first degree and so is also rational, as was to
be proved.
If, however, only rs - 2 of the intersections are rational, the remaining two
are determined by a quadratic equation and so belong to a quadratic field.
Theorem 2
L e t f ( x , y , z ) = 0, g(x, y , z ) = 0, h(x, y , z ) = 0 be three surfaces of degrees
r, s, t. I f rst - 1 of their intersections are rational points, then the remaining
intersection is also a rational point. Ifrst - 2 of their intersections are rational,
then the remaining two intersections hace coordinates in a quadratic field. I f
one of the surfaces is a cubic surface, then the line joining these points meets it
in a rational point.
The last part is obvious since the line can be defined by two planes with
rational coefficients.
Applications will be given in Chapters 12 and 13.
We come back to curves.
Letf(x, y ) = 0 be a conic with the known rational point P(xo,yo). We take
for g(x, y) = 0, the line y - yo = t(x - xo)through P, where t is a rational
9. RATlONAL SOLUTIONS DERIVED FROM GIVEN ONES 67
parameter. This line meets the conic in one other point Q, and so Q is a
rational point whose coordinates are given in terms of the parameter t. All the
rational points Q onf(x, y ) = 0 are given in this way since the slope of the
line PQ defines t. There is also a 1-1 correspondence between the points Q and
the rational values of t including t = co. Also P corresponds to itself for the
value of t for which the line is a tangent at P . The intersection of the line
x = xo other than P corresponds to t = co, but when x = xo is a tangent, P
then corresponds to t = co.
On eliminating y , we find, say,
Alt2 + Blt + C, Azt2 + B2t + C2,
s=
At2 + Bt + C '
=
At2 + Bt + C
where the A , A,, A2 etc. are integer constants.
Hence the solution of F ( X , Y , Z ) = 0 is given by putting x = X/Z,
y = Y / Z , t = p/q, where p , q are coprime integers, and then
dX = Alp2 + Bipq + Clq2, dY = A2p2 + Bzpq + Czq2,
dZ = Ap2 + Bpg + Cq2.
The factor d can be removed by an appropriate substitution p -+ hp dq. +
Then the solutions X, Y, Z are given by a finite set of binary quadratic forms.
Theorem 3
All the rationalpoints on an irreducible cubic curve with a double point can be
expressed in terms of a rational parameter.
There can be only one double point P, for if P, P1 were both double points,
the line P P , would meet the curve in four points. Also P is a rational point,
say (so,yo).This is easily proven since P is determined by the equations,
Then every other rational point Q on the curve is determined parametrically

by the equation
y - yo = t ( x - s o ) ,
since the double point counts for two of the intersections of the line and the
curve.
Example 1
ax3 + bs'y + cxy2 + dy3 = ex2 + fxy + gy2,
say g(x, r) = h(x,y>.
Here y = t s gives
x = 4 1 , t)lg(l, t>, Y = t w , t)lg(l, t ) .
We note that the origin (0,O) corresponds to both t = 0 and t = a.
Theorem 4
From a known set S of rationalpoints P,,Pz, . . ., P, on a cubic curve, others
can in general be found.
The tangent at P , meets the cubic in three points. Two of these are at P ,
and so the third point, say P , , , , being determined by an equation of the first
degree, is rational and is different from P if P , is not a point of inflexion. So
the tangent at P,,,may be expected to lead to a new rational point P 1 l , l l etc.
Next the line P,Pzmeets the curve in a third rational point P1,2,in general,
different from P , and P2. This tangential and chord process can be applied to
all the points of S and also to all of the points found in this way. In general,
one would expect an infinity of rational points to arise, but this must be
proved in each case.
The method above suggests the conjecture that all the rational points on a
cubic curve of genus one can be deduced from a finite number by the chord
and tangent process. This will be proved later, in Chapter 16.
2. We now give a number of classic instances due to Diophantus, Fermat,
Euler, Lagrange, Cauchy.
Example 2
y2 = x3 + k.
Let P ( p , q), q # 0 be a rational point. The tangent a t P is
and so x3 + k = (q + ?f (x - p))'.
Since two roots are x = p , q, the sum of the roots of this equation gives for
the third root, x ,
x = -9p4
- 2
4q2
p , and then y =q +@
2q
(v-
4q2
3p).
Alternatively, put x =p + X , and
y2 = + 3p2X + 3pX2 + x3,
q2
and take y = q + - 3P2

X
2q
Then
442-- 3p + X ,
9 ~ 4 etc.
When k = -2, p = 3, q = 5 , we find x = 129/100.

Fermat asserted that an infinity of rational solutions will arise by this method
of Bachet, but this statement requires proof, which will be given later in
Chapter 23.
9. RATIONAL SOLUTIONS DERIVED FROM GIVEN ONES 69
When q = 0, the equation takes the form y 2 = x3 - p 3 . Now the tangent
at P ( p , 0) meets the curve again at infinity, i.e. x = co,y = 00, and this must
be considered as a rational point. The homogeneous form of the equation
is Y 2 Z = X 3 - p 3 Z 3 , and P is the point ( p , 0, 1). The tangent at P is
X - pZ = 0, and meets the cubic again at the point (0, 1, 0). The tangent at
this point is Z = 0 and so the point (0, 1,O) is a point of inflexion, and so
the process cannot be continued.
Example 3
+ bx + cx2 + d.y3,
p3 = a3 a # 0.
A solution is given by putting y = a + bx/3a2.
Example 4
y3 = a + bx + ex2 + d3x3, d # 0.
A solution is given by y = dx + c/3d2.
Example 5
Let y2 = ax4 t bx3 + c s 2 + dx + e,
with the rational solution ( p , q ) , q f 0.
On putting x = p + ,'A we may suppose that the solution is (0, q ) and that
e = e:, el # 0. Put
y = A 2 + BX + el,
where 2e,A + B2 = c, 2e1B = d.
Then A2x t 2AB = ax + 6,
giving in general a new value for x .
Example 6
y2 = ax4 + bx3 + cx2 + dx + e.
Let a = a: # 0 be a perfect square. Put
y + C x + D,
= a1x2
where 2a,C = b, 2a1D + C2 = c.
Then 2CDx + D 2 = dx + e , etc.
Example 7
y2 = ax4 + bx3 + cx2 + dx + e.
Let two rational points ( p , q), ( p , , 4,) be known. Put
Then the equation takes the form

Y2 =q2x4 + bX3 + cx2 + dX + q:,
and now both of the methods above can be applied.
Example 8
y2(ax2 + bx + c ) + y(alx2 + b,x + c,) + a,x2 + b2x + c2 = 0.
Solve for y. Then
2y(ax2 e bx + c) + alx2 + b,x + c1 = z,
say, where
z2 = (alxz + b,x + c,)~- 4(ax2 + bx + c)(a2x2+ b,x + c,),
a quartic curve in z, x .
We may proceed more simply as follows. Let (x,, y,) be a known solution.
If we put x = x,in the equation, the quadratic in y has two roots one of which
is y,. The other is a rational number y , given by
yz + y1 = -(alx2 + b,x + cl)/(ax2+ bx + c).
Hence in general we have a new solution (x,, y 2 ) different from the first. We
can continue the process. Put y = y , in the equation. Then the quadratic in x
gives us a rational x2 where
xz + xl = -(by: + b,y, + b,)/(ayZ + sly, + a,).
Then in general we have a new solution (x,,y,). So we find rational solutions
(xz, Y 3 ) , (x39 Y 3 ) , (x39 Y4), (x4, Y4), * . * *
Example 9
The simultaneous equations
y 2 = alx2 + b,x + c,, z2 = a2x2 + b,x + c2.
Suppose that a rational solution (x,, y,, zl) is known. On putting x = x1 X , +
we may take x , = 0 and then replace cI, c, by c,2, ci. In the first equation, put
y = t x + c,. Then
t2x + 2rc, = a l x + bl, x = (b, - 2c,t)/(t2 - al).
Then
z2(t2 - al)2 = +
s2 = a2(bl - 2 ~ , t ) ~b2(bl - 2tc,)(t2 - a,) + cZ(t2 -
The quartic in r becomes a square when 6, - 2tcl = 0, i.e. for t = b,/2c,
unless c1 = 0, and so in general other rational values of t can be found.
If, however, c, = c2 = 0, put y = Y / X , z = Y / X , x = 1/X.
Then Y 2 = a, + blX, Z 2 = a, b2X. +
and b z Y 2 - b l Z 2 = a1b2 - a2bl.
The solution of this equation is given by the classical theory of quadratic
forms.
An alternative method is to note that
cgy2 - c;zz = s((alcz"- a2c:)x + blcg - b&) = x(px + q), say.
Take czy + ClZ = 2C,C,(P.\: + 4)/4,
czy - c,z = qx/2c,c,,
Then y = el + rx, say, and (rx + c , ) ~ = u1x2 + b,x + cf, etc.
An ancient problem which had its origin1 among the Arab scholars before
972 A.D. is, when k = 5, to find rational solutions of the simultaneous
equations:
Example 10
y 2 = x2 + k, Z' = x2 - k.
It is easily verified that if k can be written in the form
k = ab(a + b)(a - b),
then a solution is given by
2x = u2 + b2, 2y = a2 + 2ab - b2, 22 = az - 2ab - b2.
From this solution, others can be derived by the method of Example 9.
For a = 5, b = 4, k = 5-fj2,x = 41/2. Hence when k = 5, a solution is
given by
x = j 41
-y y=iz,
49
z=-
31
12'
This solution was found by Leonard0 of Pisa, about 1220 A.D. and also by
Euler who considered the general equation.
The equations can be dealt with by arithmetical considerations on writing
them in the following forms:
x2 - 5y2 = u2, x2 + 5y2 = 02%
Then Genocchi has shown that from a solution (x, y , u, v ) , another

( X , Y , U , V ) is given by
x= 212x2 + 5v2y2, Y = 2xyu0,
u = u2x2 - 502y2, v= u4 - 2x4.
From the first solution of Leonardo's equation, we find the second solution
3,444,161 4,728,001 113,279
= 1,494,696' = 1,494,696' = 1,494,696'
A complete solution can be found by more complicated arithmetical methods
which show that all the solutions can be derived from the first by rational
operations. The third solution has 15 digits in the numerators and denomi-
nators.
3. One of the first results in which the method of infinite descent of Chapter 4
shows that all integer solutions can be derived from rational functions of one
solution, was given by Euler and Lagrange.
Theorem 5
AN the integer solutions of the equation
x4 - 2y4 = -z2, > o,y > 0, z > 0, (x,y) = I,
can be so derived from the solution (x,y , z ) = (1, 1, 1).
Suppose x, y is the solution with least y and x > 1. If z 3 0 (mod 2), then
x = y E 0 (mod 2), and so we have
x = y E z = 1 (mod 2).
x2+z2 x2-22
Since +(
(7)T= y 4)
,
we have
y2 = u2 + b2, u 2 0, b 3 0, (a, b) = 1.
Clearly ab # 0, since then z = x2 = y 2 = 1. First, for + k b , we have
x2 = a2 - ba + 2ab, y2 = a2 + b2.
The first equation shows that a f 0 (mod 2), and so a = 1 (mod 2), b =
0 (mod 2). Hence from the second equation
u p 2 - q2,
1 b = 2pq, y = pa + q2, ( p , 4) = 1.
Alsopq # 0. Then
x 2 + 2b2 = (a + b)',
and so
x = +(c' - 2d2), b = 2cd, u + b = c2 + 2d2,
c 5 0, d 2 0, (c, d ) = 1 and cd # 0. Hence
p 2 - q2 + 2pq = c2 + 2d2, pq = Cd.
From the second
p = rc/s, q = sd/r, (r, s) = 1 , rs # 0,
and so
c=sX, d = r Y , p = r X , q = s Y , (A, Y ) = 1 , X Y # O .
Substituting in above, we have
(r2 - s2)X2+ 2rsXY - (s2 + 2r2)Y2 = 0.
6 = (rs & w , r 2 - s2),

where r2s2 + (r2 - s2)(s2 + 2r2) = w2,
or s4 - = --2.
Here y = p 2 + q2 = r 2 X 2 + s 2 y 2 > r2, or r < dj,

and so we have, from the solution (x, y ) , deduced a smaller solution (r, s).
Take next -2ab above. Then
x2 = a2 - b2 - 2ab, a > b.
Hence a - b = c2 + 2d2 and so
p2 - q 2 - 2pq = c2 + 2d2, pq = cd.
As before we deduce a smaller solution (r, s).
We finally come to a stage with r = 1 , s = 1 and so all the solutions can be
expressed rationally in terms of this.
We now prove an associated
Theorem 6
All the integer solutions of the equation
.y4 - 2y4 = > 0, y > 0, ( x , y ) = 1,
z2,
except x = 1, y = 0, can be deduced by rational processes from the solution

( x ,y ) = ( 3 , 2 ) and the solution (x, y ) = ( 1 , 1) of x4 - 2y4 = - z 2 .
Clearly x, z are odd and y is even. Hence
x2 + z = 2a4, x2 - z = 16b4, y = 2ab,
ab # 0, (a, b) = 1, and so a = 1 (mod 2) and
a4 + 8b4 = x2.
Hence +a2 = p 2 - 2q2, b2 = pq, p q # 0, ( p , q ) = 1.
Then p = c2, q = d2, and
c - 2d4 = f a 2 .
The equation with the - sign has been dealt with. For the equation with the +
sign, d2 < b2 < t y 2 or d < i y , i.e. there is a solution of the original equation
with a smaller y. Continuing the process, we have two equations such as
e4 - 2f4 = +g2.
It may be remarked that the solution x = 1, y = 0 leads to b = 0.
In all these examples, new rational solutions were derived from known
rational solutions. It is sometimes possible to deduce rational solutions from
known solutions given as numbers in an algebraic number field.
We have Heegner's3
4. Theorem 7
If the equation
y2 = ax4 + bx3 + ex2 + dx + e, (1)
where a, b, c, d, e are rational numbers, has a solution ( p , q ) in an algebraic
number field K of odd degree b 3 , then there exists a rational solution.
We may suppose that p satisfies an equation with rational coefficients of
degree r, sayf,(p) = 0, and that K is generated by p . For if p generates a sub-
field Kl of X of degree r l , then from equation (l), q must be an element of an
algebraic number field of degree 2r, and so of rl since r is odd. Then q =
fr-d(p), a polynomial of degree r - 6 with rational coefficients, and where in
general 6 = 1. Since the relation
= dp4 + bp3 + cp2 + dp + e
is a consequence off,(p) = 0, we have if now x is an arbitrary variable, the
identical relation
f ,"-&) - ax4 - bx3 - ex2 - dx - e =f r ( ~ ) f r - ~ ~ ( x ) ,
wheref,-,,(x) is a polynomial of degree r - 26.

Hence a solution of equation (1) is given by taking x to be a root of the
equation5 - 26(x)= 0. Since r - 26 is odd, the process can be continued until
a stage is reached when r - 26 = I .
In Heegner's applications, the field K is derived by very abstruse and com-
plicated considerations depending upon equations connecting various modu-
lar functions. He gives a number of equations which are said to have rational
solutions. Among these, where p is a prime, are
+ 27,
p y 2 = x3 p E 3 (mod 4), y # 0 ;
py2 = 2x4 + 1, p = 3(mod8);
p y 2 = 4x4 - 1, p = 3 (mod 8);
py2 = 8 - x4, p = - 1 (mod 8).
REFERENCES
1. L. E. Dickson. History of the Theory of Numbers. Carnegie Institute,
Washington, 2 (1920), 459-472, for the history and accounts of the subject and
references.
2. J. V. Uspensky and M. A. Heaslet. Elementary number theory. McGraw-
Hill, London. (1939), 419-427.
3. Kurt Heegner. Diophantische Analysis und Modulfunktionen. Math. Z., 56
(1952), 227-253.
CHAPTER I0
Rational Points on Some Cubic Curves
1. Equivalent curves. Two curves are called equivalent if they can be trans-
formed into each other by a birational transformation with rational coeffi-
cients. This means that if the curves are given by f ( x , y ) = 0, F ( X , Y ) = 0,
we have, except for a finite set of values for the variables,
x = a(X, Y ) , y = b(X, Y ) ; X = A(x,y), Y = B(x,y),
where a, b, A, B are rational functions of their arguments with rational
coefficients.
Theorem 1'
A cubic curuef ( x , y ) = 0 with a rationalpoint is equivalent to a cubic whose
equation is in the Weierstrass normal form
ya = 4x3 - gax - g3.
If the curve is written in the homogeneous form, F ( X , Y, Z) = 0, then
ga,g, are well known invariants of the ternary form F(X, Y , Z ) . We give
Nagell'sa proof.
Let P be the rational point and take as origin Q, the point where the tangent
at P meets the curve again. The line y = tx through Q, where t is a parameter,
meets the cubic in two other points determined by the equation,
Lxa + 2Mx + N = 0,
where L, M , N are polynomials in t of respective degrees 3, 2, 1, and with
rational coefficients. Hence
(Lx + M ) a = M a - LN.
Since P and Q are rational points, and QP is a tangent, the quartic in t ,
M a - LN = 0 has a rational root, say t = to. Write to t = to + 1/T,Then
(Lx + = F(T)/T4,
where F(T) is a cubic in T.Apply a linear substitution T = CX + D with
rational coefficients C, D, and put L x + M = N Y / ( C X + D)a, and then
we have an equation of the form
ya = 4 x 3 - g a x - g3.
10. RATIONAL POINTS ON SOME CUBIC CURVES 77
Then r takes the form
t = (FA+ G)/(CX + 0) and y = rx.
Clearly rational X , Y lead to rational x , y and conversely.
It proves convenient to regard ( X , Y ) = (co,co) as a rational point. This
arises from the homogeneous form of the equation
Y2Z = 4 x 3 - gJZ2 - 8 3 2 3 ,
which has the rational point ( X , Y, 2) = (0,1,0), so that Y / Z = co,X / Z = co.
Theorem 2
If the quartic curve
y2 = ax4 + bx3 + cx2 + dx + e
has a rational point, it is equivalent to a cubic curve
y2 = 4x3 - g2x - g3.
We may assume that e is a perfect square. Then on writing x = 1/X,
y = Y / X 2 ,we may assume that a is a perfect square. If a = 0, the result is
obvious. If a # 0, we may assume that a = 1 on writing x = X/&, y =
Y/&. Then on replacing x by X - +b, we may assume that b = 0. We now
write the quartic as
y2 = x4 - 6cx2 + 4dx + e.
A 1-1 correspondence between this quartic and the cubic is given by
y = -2 + 2 x + c, 2x = (Y - d ) / ( X - c).
For on substituting for y,
.Y* - 6cx2 + 4dx + e = x4 - 2x2(2X + c ) + ( 2 X + c)~,
x2(X - c) + dx = X 2 + c X + +(c2 - e).
2x(X - c) = - d & (d2 + 4 ( X 3 - c z X ) + (c2 - e)(X - c))lI2
= - d f ( 4 X 3 - g2X - g3)l,
+
where g2 = e 3c2, g , = -ce - d 2 + c3, d 2 = 4c3 - g2c - g,.
Hence we can write
y2 = 4 x 3 - g2X - g3, 2x(X - C) = Y - d.
We note that this normal cubic has the rational point ( X , Y ) = (c, d ) .
A sometimes more convenient form for the quartic is
z2 = ax4 + 4bx3y + 6cx2y2 + 4dxy3 + ey4 = F(x, y).
The binary quartic F(x, y ) has the two well known invariants g,, g, given by
a b c
g, = ae - 4bd + 3ca, g3 = b c d .
c d e
Suppose now that the normal cubic has no double point.
We notice some obvious consequences of these results. If the cubic curve
F ( X , Y , Z ) = 0 has one rational point, then all the rational points can be
found if all the rational points of the normal cubic are known. Further if the
curve F ( X , Y, Z) = 0 has a finite number 2 1 of solutions, the normal cubic
has only a finite number of solutions and so any other cubic curve
F,(X, Y, 2 ) = 0 with the same invariants as F ( X , Y, 2 ) can have only a
finite number of rational points.
Similar results apply to the quartic za = F(x, y).
2. One would expect that the general cubic curve has an infinity of rational
points if it has one, but this requires proof for any given curve.
Theorem 3
Let a, b, c be square-free integers, relatively prime in pairs, abc # 0 and d an
integer. Then the curve
ax3 + by3 + cz3 + dxyz = 0,
when at most one of a, b, c is f 1, has either none or an infinity of rational
points.
When a = b = 1, c # k 1, the curve has either one or two or an infinity of
integer solutions. Only two solutions may be possible when d = -(c f: 2) or
d = - (4c f l), and so ifsuch an equation has a third solution, there will be an
infinity of solutions.
When a = b = c = 1, and d # 1, -3, - 5 , the curve has either three
rational points or an infinite number. Three of the points are (1, - 1,O) etc.
The case d = - 3 is trivial since then x + y + z = 0. When d = 1, there are
only six solutions, (1, - 1,O) and (1, 1, - 1) etc. When d = - 5 , there are
only the six (1, - 1,0), (1, 1,2), etc.
The results for d = 1, - 5 are due to Mordel13, the others are due to
Hurwitz 4.
We prove first that the tangential Q ( X , Y, 2 ) of any point P(x, y, z) on
the curve is given by
6 X = x(by3 - cz3), 6 Y = y(cz3 - ax3), 62 = z(ux3 - by3),
where 6 is an integer such that (X, Y, 2 ) = 1. On writing x = X q i etc., it

suffices to prove the result when a = 6 = c = 1, and then to verify that Q
lies on the tangent at P and also on the curve. The first is obvious since the
tangent at P is in current coordinates 2,J. 2
T(3X2 + dyz) + . . * = 0,
and C x 3 ( y 3 - z3) = 0, C x y z ( y 3 - z3) = 0.
The second follows from
2x3(y3- z +
~ ) ~ X~ Y Z n ( ~ -
3 ~ 3 )
= (x3 + + + dxyz) n (y3 -

y3 23 z3) = 0.
We may suppose that (x, y , z ) = 1. Then x, y , z are relatively prime in
pairs. For if a prime p divides x and y , then p 2 I cz2, i.e. p I z contrary to
hypothesis. Suppose now that there are only a finite number b 1 of rational
points on the curve, and denote by ( x , y , z ) one for which lxyzl has its
greatest value.
We return to the point Q and show that (6, xyz) = 1. For if a prime p
divides 6 and x , then ( p , y z ) = 1. Then since p divides both cyz3 and bzy3,
p I b, p I c contrary to hypothesis. Hence
6 I (by3 - cz3), 6 I (cz3 - ax3), S I (ax3 - by3).
Now suppose lab1 > 1. Then xyz # 0, for if x = 0, then by3 + cz3 = 0
whose only solution is y = z = 0 since (6, c ) = 1 and b and c are square free.
Similarly for y = 0 etc. Hence IXYZl/lxyzl < 1 and so
1
by3 - cz3 cz3 - ax3 ax3 - by3
6 * S * 6 1< 1.
Then by3 - cz3 = 0, f 6; cz3 - ax3 = 0, k 6 ,

ax3 - by3 = 0, f 6 ,
where the signs are independent of each other. The 0 can be rejected since
( y , z ) = 1, and (6, c ) = 1 etc. Finally the t- 6 can be rejected since the sum
of the left-hand sides of the equations is zero.
Suppose next that a = f 1, b = f 1, IcI > 1, say a = b = 1. The curve
has a rational point (1, - 1,O) and this is a point of inflexion and is also the
only rational point for which xyz = 0. Excluding this point, we have a con-
tradiction unless XYZlxyz. Then
X = Y = 1, 2+cz3+dz=O.
Hence either
z = kl, 2 f (c + d ) = 0,
or z = f2, 1 5 (4c + d ) = 0.
80 DIOPHANTINE EQUATIONS ~
I
This means that a finite number of rational points are possibly only for the
four equations,
x3 + y3 + cz3 - (c f 2)xyz = 0, I
satisfied by ( x , y, z ) = (1, - 1,013 (1, 1, f 11,

and x3 + y 3 + cz3 - ( 4 ~k l ) X p = 0,
satisfied by (x, y, 2 ) = (1, - 1, O), (1, 1, k2).
Finally let a = b = c = 1. As before, we must have XYZJxyzand so two
of x , y, z must be equal, say x = y = 1.
Then 2 + z3 + dz = 0,
giving Z = +I, 2+(1+d)=O, d=l,-3;
or z = i-2, 2 f (8 + 2d) = 0, d = -5, -3.
This completes the proof of the Hurwitz results.
Example
The equation
x3 + + 23 - xyz = 0
y3
has only the trivial solutions with xyz = 0.

This was proved by Cassels6by extending a method similar to that used by
Mordell for d = 1, -5. He and Sansone also gave another proofs. Their
proof involves algebraic number theory and will be discussed in Chapter 15.
We note that all the equations
ax3 + by3 + C Z ~= 0
with abc = k have the same invariants. We deduce from Theorem 1 that if
the equation
x3 + y 3 + k z 3 = 0 (2)
has only the trivial solution x = 1, y = - I, z = 0, then the equations (1)
have only a finite number of solutions. It also follows at once that if a, b, c
satisfy Hurtwitzs conditions, namely, that a, 6, c are square free, relatively
prime in pairs, and at most one of them is 1, then the equations (1) have no
rational solutions. It is easily verified that the equation (2) has rational
solutions with z # 0 if and only if
XY(X + Y ) + k23 = o
has rational solutions with 2 # 0. Thus we can take
x = x3, Y = y3, 2 = xyz,
or
x = X3 + 6XY + 3XY - Y3, y = -X3 + 3XaY + 6 X Y a + Y 3
z = 3Z(Xz + XY + Y).
Fueters result on the infinity of solutions of the equation y2 = x3 + k,will
be given in Chapter 23.
REFERENCES
1. L. J. Mordell. Indeterminate equations of the third and fourth degrees. Q. J.
Pure Appl. Math., 45 (1913), 170-181.
2. T. Nagell. Sur les propriktks arithmktiques des cubiques planes du premier
genre. Acta Math., 52 (1929), 93-126.
+ + +
3. L. J. Mordell. The diophantine equation x 3 y 3 z3 kxyz = 0. Colloque
sup la thkorie des nombres. Bruxelles (1955), pp. 67-76.
4. A. Hurwitz. Ueber ternare diophantische Gleichungen dritten Grades. Viertel
jahrschrijit Naturf. Ges. Zurich. 62 (191 7) 207-229. Math. Werke (Birkhauser Cie,
Basel) 2 (1933), 446-468.
5. J. W. S. Cassels. On a diophantine equation. Acta Arithmetica, 6 (1960), 47-51.
6. G. Sansone and J. W. S. Cassels. Sur la probleme de M. Werner Mnich.
Acta Arithmetica, 7 (1962), 187-190.
CHAPTER 11
Rational Points on Cubic Surfaces
1. Let&, y , z ) be a cubic polynomial and g(x, y, z, w ) a homogeneous cubic

polynomial, both with rational coefficients. In Chapter 1, we have shown that
some equations f = 0 and g = 0 have only the trivial solution, e.g. if p is a
prime,
ax3 + bpy3 4- cpaz3 = 0, abc f 0 (mod p),
x3 + 2y3 = 7(z3 + 2w3).
No method is known for determining whether rational points exist on a
general cubic surface f ( x , y , z ) = 0, or for finding all of them if any exist.
Geometric considerations may prove very helpful, and sometimes by their
help an infinity of solutions and even all the solutions may be found. Thus it
is known that a cubic surface may have four double points, i.e. points for
which
Suppose that one of these points, say Pl(xl, y,, z,) is rational. Then ob-
viously all the rational points on the surface are given by the third intersection
with the surface of the line
-x _- -x- l - Y - Y l -21,
1 m n
where I, m,n are rational parameters.
Results may also be deduced' when the double points are defined by
irreducible equations of the second, third and fourth degrees.
When one rational point PI is known, an infinity of rational points can in
general be found. Thus any tangent line to the surface at Pl meets it at a
point Pz different in general from P1.The process can be continued. Further,
the tangent plane at PI meets the surface in a cubic curve with a singular
point at P,. The curve, if irreducible, has a double point and so its coordinates
can be expressed rationally in terms of a rational parameter. A similar result
holds if the cubic splits into a conic and a straight line. A more interesting
case, considered further on, arises when the cubic curve splits into three
straight lines and then these are defined by a reducible or irreducible cubic
equation. The tangential process may be continued by taking another rational
11. RATIONAL POINTS ON CUBIC SURFACES 83
point on the cubic curve defined by P1, but there seems no analogue corre-
sponding to the finite basis theorem for cubic curves. In fact, Segre2 has
proved the surprising result that cubic surfaces exist containing a rational
point PI which cannot be obtained by the tangent plane process starting from
any given set of rational points of the surface not including P1.
The existence of straight lines on a cubic surface may prove useful. In
general, there are 27 such lines, determined by an equation of the 216th
degree, and various sets of these lines may be determined by equations of
lower degrees.
Theorem 1
AN the rational points on a cubic surface can be found if it contains two lines
whose equations are defined by conjugate numbers of a quadratic field and in
particular by rational numbers.
Let P1, P, be a pair of conjugate points on conjugate lines L1,L, on the
surface. Then the line PIPzmeets it in a third point P whose coordinates are
rational since if a cubic equation with rational coefficients has a rational
quadratic factor, it will also have a rational linear factor. Conversely let a
rational point P be given. A plane drawn through P and L1 will meet La in a
point P,. Then PP, meets Ll in a point P1 and clearly PI, P, are conjugate
points in the quadratic field.
Before the later general results were found, solutions of some equations3
were found by special devices.
The equation
(x + y + z ) ~- dxyz = m.
We can find easily the special solutions with m = dyz,. Then the equation
can be written as
+ +
(x y z ) = ~ dyz(x +
z).
This equation defines a cubic curve in homogeneous coordinates. It has a
+
double point at x z = 0, y = 0, and so its coordinates can be expressed
rationally in terms of a parameter.
Put x + z = py, and ( p + l)3y = dpz.
Then +
m = dyz2 = d2pz3/(p l)3.
This requires p = dmt3 and then x, y , z are rationally expressable in terms of
a parameter 2.
This equation was the first really significant equation for which a parametric
solution was found in the 1 15years since the time of Ryley, who in 1825 found
a parametric solution of
X3 + Y3 + Z3 = n. (2)
This equation is included in the equation (1) on putting d = 24, m = 8n,

x = Y + Z, etc.
Richmond4 has given other solutions of Ryleys equation, namely
3(1 - t + t a ) X = ~ ( +l t3), 3(1 - t + ta)Y = s(3t - 1 - t3), (3)
3(1 - t + t2)Z= s(3t - 3t2), t = 3n/s3.
A particular case of Ryleys equation is
Euler s equation
x3 + y3 + z3 = 1. (4)
The complete solution can be found from Theorem 1 since two lines on the
surface are given by
x+py=o, z=p; x + p2y = 0, z = pa,
where pa + p + 1 = 0.Alternatively, write the equation as
(x + Y ) ( X + P Y ) b + P2Y) = (1 - z)(l - ZP)(l - zPa).
The solution in terms of rational parameters p, q is given by
(x + Y)(P + Pd)(P + P a d = 1 - z. (6)

This also holds when p = q = 0 giving z = 1, x = y = 0.
On excludingp =q = 0, we can write equations (9,(6) as
x + PY + P(P + P4)Z = P + P4,
x + P2Y + P2)(P + P 2 d Z = P + p2q, (7)
(x + y)(p2 - pq + qa) = 1 - z.
Eliminating x, y from the first two equations, we find
x =P + qz, Y =4 + (4 - P>Z,
(8)
z((p - q)3 - 4 3 - 1) = p3 + 13 - 1.
Then x((p - q)3 - q3 - 1) = (pa - pq + qa)2 - p - q,
y((p - q)3 - q3 - 1) = 1 - (p2 - pq + q 2 y + p - 2q. (9)
It is obvious on solving equations (8) that (p, q) are uniquely determined

when (x, y, z) are given since the denominator 1 + z + z2 # 0. Conversely
let (p, q) be given. Since the only solutions of
(p - q ) 3 - 4 3 - 1 = 0,
I 1. RATIONAL POINTS ON CUBIC SURFACES 85
are given byp = q = I ;p = 1 ,q = 0 as will be shown in Chapter 15, then the
equations (7) are inconsistent.
2. We return to the general equation f ( x , y , z ) = 0. We suppose that
f ( x , y , z ) is an algebraically irreducible cubic polynomial and is not a function
of two independent variables, nor a homogeneous polynomial in linear
functions X, Y, Z of x, y , z. In the latter case, the equation takes the form
F(X, Y , Z ) = 0, which is a cone, but which in homogeneous coordinates
represents a cubic curve. This may contain only a finite number of rational
points or an infinite number. We give now a proof of Segres5
Theorem 2
Zff(x, y , z ) is as above, the general cubic equation, f ( x , y , z) = 0 has either
none or an infinity of rational solutions.
Take the equation in the homogeneous form F ( X , Y , Z , W ) = 0, and
suppose there is a rational solution (Xl, Y,, Z1, W,) # (0,0, 0, 0), and we
may assume that Z , # 0.
Put
x = XI5 + t, Y = Y15 + 7, z =z15, w = w15 + p.
The equation becomes
t2L1 + 5L, + L3 = 0, (10)
where L1,L2,L3 are homogeneous forms of degrees 1, 2, 3, respectively, in
t , ~p, ,and L1, L2 are not both identically zero. If L, is identically zero, the
equation in 5 gives at once all the rational solutions of F ( X , Y , Z , W ) = 0.
Suppose next that L1 is not identically zero. An infinity of solutions arise on
imposing the condition L1 = 0 on t , ~p, ,provided that L2 is not identically
divisible by L1. When this is so, a linear substitution on t , ~ p, ,replaces L1
by p, say. Then the 5 equation takes the form
(f + A 4 1 ($) + M3 = 0,
where M1, M 3 are linear and cubic polynomials in 5 / p , v / p . On solving for
( / p , we see that the problem of finding the rational solutions of this equation
is equivalent to that of finding rational solutions of the equation
z2 = S(X, y), (1 1)
where f ( x , y ) is the general cubic polynomial in x, y . This is the type of
equation which would have arisen when the tangent plane at a rational point
of the surface F(X, Y, Z , W ) = 0 meets the surface in three straight lines.
We give Whiteheads6 solution of equation (1 1).
Write
z2 = ax3 + bxzy + cxy2 + dy3 + ex2 + f x y + gy2 + hx + iy + j , (12)
4 f
where we may suppose that a # 0 by a linear transformation if need be, and

then that a = 1 on replacing x, z by xla, z/a.
On replacing x by x - py - q, say, we may suppose that the equation
takes the form
z2 = x3 + x(ay2 + by + c ) + dy3 + ey2 + fy + g.
The right-hand side is a function of only one variable, which can only be
x + ky, k a constant, if and only if,
a = b = d = e = f = 0,
a case we have excluded.

Put x = y2 + 2ty where t is a parameter. Then
z2 = ys + 6ty6 + (12t2 + a)y4 + (8t3 + 2at + b + d)y3
+ (2bt + c + e)yz + (2ct + f)y + g.
Put now z = y3 + Ay2 + By + C,
where
2A = 6t, 2B + A 2 = 12t2 + a, 2C + 2AB = 8t3 + 2at + b + d.
and so A = 31, 2B = 3 t 2 + a, 2C = b + d - at - t3.
Substituting for z2,we have

Py2 + Qy + R = 0,
where + f(3t2 + a)2 + 3t(b + d - at - t3) = 2bt + c + e.
P
or P = 3t4 + +at2 - (6 + 3d)t + c + e - fa2.
so Q + 3(3t2 + a)(b + d - at - t3) = 2ct +f,
or Q = 3t6 + 2at3 - $(b + d)t2 + $(a2 + 4c)t + f - +a(b + d).
Finally R = g - f ( b + d - ata - t 3 ) a
= - a t e + . . ..
We note that
2Pt - Q = at3 - +(b + 9d)t2 + (2e - aa)t + +a(b + d ) - J
and this is not identically zero unless a = e = f = 0, b + 9d = 0.
On solving the quadratic for y, we have two values of y which are finite and
different, except for a finite number of values of t since neither P nor A =
Q2 - 4PR = 3t1 + . . . is identically zero.
11. RATIONAL POINTS ON CUBIC SURFACES 87
Hence 2Py = - Q f dK,
2P2x = 2P2y2 + 4P2ty
= -2PQy - 2PR + 4P2ty
= y(4P2t - 2PQ) - 2PR
= (- Q f d L ) ( 2 P t - Q ) - 2PR
= Q2 - 2PR - 2PQt f (2Pt - Q)d&
Since z = y3 + . . - ,we have, say,
x =a f qdA, y = p f pld& z =y k ndh.
These two points define a straight line and any point on it can be written as
x = + clle, Y =p + pie, z = + yle,
where 8 is a parameter. The points where this line meets the cubic surface
Z 2 = f ( X , Y ) are given by a cubic equation in 8. This has a factor B2 - A,
and the remaining factor is linear, and determines a rational point. The third
root will not be infinite except for a finite number of values of t. For if the
coefficient of O3 is zero,
+ alpla + dp; = 0,
a;
or (2Pt - Q)3 + aP2(2Pt - Q ) + dP3 = 0.

On noting the degrees of 2Pt - Q and P and the coefficient of t12 in the
equation, we find d = 0. If 2Pt - Q is identically zero, a = e = f = 0,
+
b 9d = 0, i.e. b = 0, and this is the excluded case. If 2Pt - Q is not
identically zero,
(2Pt - Q)2 + aP2 = 0.
From the coefficient of t e , a = 0, and then 2Pt - Q = 0, a contradiction.
3. Mordell has proposed the plausible

Conjecture
The cubic equation F ( X , Y,2, W ) = 0 is solvable ifand only ifthe congruence
F(X, Y, 2, W) = 0 (mod pr), (1 3)
is solvable for all primes p and integers r > 0 with ( X , Y, Z, W,p ) = 1.
The conjecture holds in at least two cases:
I. when the equation takes the form
N(X + 8Y + P Z ) = mW3,
where N is the norm form in the cubic field Q(0).
This can be proved by the theory of algebraic numbers.

11. when the equation takes the form
X3 + a Y 3 = b(Z3 + aW3).
This follows from case I since the relation
establishes a 1-1 correspondence between (X,Y, Z, W) and (4,7,5, p).

The conjecture, however, is not true for all F ( X , Y,2, W). Thus as shown by
Swinnerton-Dyer', the equation
wx(x + w)= N ( X + ye + zea),
where 83 + 7(e + i y = o
is locally solvable but has no integer solutions.
Other instances have been given by Mordel18. The arguments are similar
to those of Swinnerton-Dyer. A more interesting counter-example has been
given by Cassels and Guye,
Theorem 3
The equation
5x3 + 9 ~ +3 1023 + 1 2 ~ 3= o
is solvable everywhere locally but has no rational solutions except (0, 0, 0,O).
The proof depends upon the application of the cubic fields k, = Q($'%),
kz = Q(V%) and their composition K = klkz,and requires the theory of
relative fields. This seems to be the only known instance of an equation of this
form which has been proved impossible otherwise than by simple congruence
considerations.
Previously Selmer lo stated that he has verified the solubility of
ax3 + by3 + cz3 + dw3 = 0
in all the relevant cases when labcdl < 500.
4. Prior to the discovery of Theorem 2, various classes of cubic equations
had been shown to have parametric rational solutions. This has been done by
Mordell, by combining both quadratic and cubic irrationalities as for equation
(lo), and by Segre by the use of geometric considerations. Further results by
the use of more abstruse geometrical ideas were found by Segrea. He de-
veloped these in a memoir in which he studied in great detail the configurations
arising from the 27 lines on a cubic surface and the sets of 216 doublets, of 720
triplets, of 72 sextuplets, the sets being such that no two lines in a set intersect.
1 1. RATIONAL POINTS ON CUBIC SURFACES 89
The determination of a set depends upon the solution of an algebraic equation
in one variable of degrees 27,216, 720, 72. The existence of a rational root in
one of these equations, say conditions pl, pz, p3, p4, gives results on the exis-
tence of rational solutions. Thus, for p l , pz, there is an infinity of solutions,
and for p z all the rational solutions are given as quotients of polynomials of
degree 4. He finds numerous results for which his memoir, which contains
many references t o the literature, should be consulted.
REFERENCES
1. T. Skolem. Einige Bemerkungen uber die Auffindung der rationalen Punkte
auf gewissen algebraischen Gebilden. Math. Z., 63 (1955), 294-312.
2. B. Segre. On the rational solutions of homogeneous cubic equations in four
variables. Math. Notae (Rosario Argentina), 11 (1951), 1-68.
+ +
3. L. J. Mordell. On Ryleys solution of x 3 y 3 z3 = n. J. Lond. Math. SOC.,
17 (1942), 139-144.
+ +
4. H. W. Richmond. On rational solutions of x 3 y 3 z3 = R. Proc. Edin-
burgh Math. SOC.,(2) 2 (1930), 92-100.
5 . B. Segre. A parametric solution of the indeterminate cubic equation za =
f ( x , y ) . J. Lond. Math. SOC.,18 (1943), 226-233.
6. R. F. Whitehead. A rational parametric solution of the cubic indeterminate
equation za = f ( x , y ). J. Lond. Math. SOC.,19 (1944), 68-71.
7. H. S. P. Swinnerton-Dyer.Two special cubic surfaces. Mathematika, 9 (1962),
54-56.
8. L. J. Mordell. On the conjecture for the rational points on a cubic surface.
J . LO^. Math. SOC.,40 (1965), 149-158.
9. J. W. S. Cassels and M. J. T. Guy. On the Hasse principle for cubic surfaces.
Mathematika, 13 (1966), 111-120.
10. E. S. Selmer. Sufficient congruence conditions for the existence of rational
points on certain cubic surfaces. Math. Scand., 1 (1953), 113-119.
CHAPTER 12
Rational and Integer Points on Quartic Surfaces
1. Only a few scattered results are known for the rational points on quartic
surfaces, and the first now given is due to Euler.
The equation
+ y4 = 24 + w4.
x4 (1)
Putx = at + c, y = bt - d, 2 = at + d, w = bt + c, wherea,b, c, dare
parameters. A quartic in t arises in which the coefficients of t4,t O are zero. The
coefficient of t will also be zero if
c(a3 - b3) = d(a3 + b3).
Take c = a3 + b3, d = a3 - b3.
Then 3t(a2 - ba)(c2- da) = 2(ad3 - ac3 + bc3 + bd3).
This gives, on ignoring the denominators,
x = a1 + a6b2 - 2a3b4 + 3aab5 + abe.
y = asb - 3a5b2 - k 4 b 3 + a2b6 + bl.
+ a6b2 - 2a3b4 - 3a2b6 + abe.
z = a7
w = aab + 3a6ba - 2a4b3 + a2b6 + bl.
Very little is known about the more general equation
ax4 + by4 + cz4 + dw4 = 0, abcd # 0, (2)
Richmond has shown that if arational point P(xo,yo, zo, wo)is known, others
can be found if abcd is a perfect square. His proof depends upon geometrical
considerations, but we present the proof rather differently. This condition
implies that a rational line can be drawn through P to meet the surface Q in
three points at P, and so if the line does not lie on the surface, then Q, the
fourth point of intersection, will also be rational.
When xoyozowo= 0, the method does not give a new solution, and
Richmond tacitly assumes that xoyozowo# 0.
Let the line be
x---
XO y-yo --
2 - 2 0 w-wwg
- t. (3)
1 --= m n P
12. RATIONAL AND INTEGER POINTS ON QUARTIC SURFACES 91
Then
a(xo + It)' + b(yo + mt)4 + c(zo + nt)4 + d(wo + ~ 2 =)0. ~ (4)
This equation will have a triple root t = 0 if
+ by: + cz; + dwt = 0,
ax6
axtl + by3,m + czin + dw;p = 0, (5)
axil2 + by$m2 + czgn2 + dwip2 = 0.
Suppose first that wo = 0, xoyozo# 0.
Eliminating a, b, c from equation ( 5 ) we have
Take 1 = xo, m = y o , and so from equation (3, n = zo.

Then from equation (4), the line meets the surface where
(ax: + by: -t- cz:)(l + t)' + dp4t4 = 0
that is, p t = 0, and no new solution arises whether p = 0 or p # 0.
Suppose next that zo = 0, wo = 0 and so xoy, # 0. Now wecan take 1 = xo,
m = yo. Then
+ by:)(l + t)' + (cn' + dp4)t4 = 0.
(ax:
Hence t = 0 and no new solution arises unless cn4 + dp4 = 0.
Then the surface will contain lines of the form
x-ry=O, z-sw=O.
By a slight change of notation, we can write the equation of the surface (2)
in the form
a(x4 - y') + c(z' - w') = 0. (6)
and so (1) is a special case of this. The equation (6) is a particular case of the
equation
xy(mxa + ny2) = zw(mza nwa) +
dealt with by Euler. A solution rather different from his is given here since the
method can be applied to more general equations.
We show that there is a parametric solution, and in general an infinite set of
parametric solutions in a variable 0.
Write the equation (6) as
a(x + y)(x3 - x2y + xy2 - y3) + c(z + w)(z3 - z 2 w + zw2 - w3) = 0.
Put v(x + y) = z + w,
where u is a parameter. The quartic curve in homogeneous coordinates

x, y, z so obtained contains the cubic curve
42 - xay + xya - y 3 ) + C U ( Z ~- za(ux + uy - Z) + Z ( U X + UJJ - z ) ~
- (ux + uy - = 0. z)3)
This contains the point x = 1, y = 1, z = u.

Write x + y = X, x - y = Y, z = u. Then
f ( X , Y ) = &aY(X2 + Y2)+
cu(u3 - u y u x - u ) + u(uX - u)2 - (UX - u ) 3 ) = 0.
This curve contains the point (X, Y) = (2, 0). The inflexions on the cubic
are given by
Now at (X, Y) = (2, 0),
-
a?- - a Y + 2cu4 - 6cu4(uX - u) = -4cu4 2
axa - a = 3aY = 0
~2
and so no inflexion occurs at (X, Y) = (2,O). Hence the tangential of this

point gives a new solution, and we may expect an infinity of solutions each
involving the parameter u in continuing the process.
The method is easily generalized and can be applied to the equation
L i f i k Y , 2, W) = L2fa(x, Y , Z, w), (7)
+ + +
where L1 = alx bly clz dlw, L2 = a2x bay Caz d2w, and + + +
f1,& are cubic polynomials in x , y, z, w. There are of course trivial solutions
given by L1 = La = 0.
Put La = uL1,and then
If we know one solution of this cubic equation, we can proceed as before.

Suppose for example that x = y = z = 1 is a solution. This will be so if
a1 + bi + + d1 = 0,
~1 + 62 + + CZ d2 = 0,
Si(1,1, 1, 1) = z f a ( l , 1, 1, 1).
Many examples can be constructed.
12. RATIONAL AND INTEGER POINTS O N QUARTIC SURFACES 93
We now return to equation ( 5 ) when xoyozowo# 0.
Put 1=LX, +-
''09 m = MY, + PY0,
- n = NZ, -.
+ PZO
WO WO WO
Then axtL + b y f M + cz:N = 0,

ax,4L2 + by:M2 + czfN2 = 0,
or, say, AL + BM + CN = 0,
AL2 + B M 2 + C N 2 = 0.
Eliminating L,
+ A B ) M 2 + 2BCMN + (C2 + A C ) N 2 = 0.
(B2
The ratio M / N is rational if -ABC(A + B + C ) is a perfect square, i.e. if
abcd = k2,k rational.
Suppose first that
a14 + bm4 + en4 + dp4 # 0.
Then the process gives from equation (4)a value for t and so another solution.
This will be the same as xo,yo, zo, wo only if either l/xo = m / y o = n/zo =
p/wo which is not so, or if t = 0, that is,
ax0I3 + byom3 + cz0n3 + dwop3 = 0.
Eliminating a, b, c, d from this equation and (9,we have
But m/yo - n/zo = 0 gives M - N = 0 and then

(B+ C)' + A(B + C ) = ( A + B + C ) ( B + C ) = 0.
Clearly A + B + C # 0 since w o # 0. If B + C = 0, then
byf + czf = 0, ax: + dwf = 0,
and the surface would contain the lines
-=-z , _x -- w
-I
Yo zo xo wo
This case has been dealt with.
Suppose finally that
a14 + bm4 + en4 + dp4 = 0.
On again eliminating a, b, c, d from this equation and (9,we have
4*
We have already dealt with I/xo - m/yo = 0, etc. and so we have

l +-
- m +-
n + 2- = 0.
xo Yo zo wo
Hence the solution I, m, n, p of equation (2) is different from xo,yo, zo, wo.
A similar result holds for Segresa equation derived from equation (2) on
replacing a, x, etc., by A . + AIO + A202 + A303, X + YO + 2 0 2 We3, +
etc., where O is the root of a quartic equation,
O4 + p03 + q02 + rO + s = 0,
where all the coefficients are rational numbers. The only difference is that in
abcd = k2, k must be a number in the field, reducible or irreducible, defined
by 0.
The coefficients of the quartic equation are rational numbers since the
coefficients a,, bl,. . . , are now conjugate numbers in the field.
2. A few equations of the form (2) may be noted.
The equation
x4 + y4 + 24 = w4. (8)
It has been conjectured by Euler that the equation has no solutions except
those with xyz = 0. Morgan Ward3*4has shown that there are no integer
solutions with w < 10,OOO. This is true5 for w < 220,000.
The equation
x4 + y4 + 424 = 1 (9)
has a rational solution given by
x=- t4-2 2t3 2t
t4 + 2 Y = Z2z = -
t 4 + 2,
where t is a rational parameter.

This is a particular solution for which
X2 + 2yz = 1.
Then y4 + 424 = 1 - = 1 - (1 - 2yz),
x4
and (y2 + 2292 = 4yz,

y2 + 222 = 2djG.
Put y = r2z. Then
(t4 + 2)z = 2t.
Since (y - 2~) = 4yZ(l - ~ Y Z=
) ~YZX,
- - 222 --t4 - 2 z
x = -ya
2t
= -.
t4 - 2
+2
2djG t4
12. RATIONAL AND INTEGER POINTS ON QUARTIC SURFACES
The equation
x4 + y4 + 24 = 2.
Put z = x + y, then
x4 + y4 + ( x + y)4 = 2,
2(x2 + xy + = 2, y2)2
x2+xy+y2= 1.
A parametric solution is given by putting y = 1 - tx giving
x(l - t + t2) = 2t - 1.
The solutions of (9), (10) and (11) below are due to Carmichaele.
3. The equation
x4 + ay4 + bz4 = w2.
New solutions can be found in general from a given solution (xl, y , , z,,wl).
The equation is satisfied by the assumption
x2 = X 2 - a Y 2 - bZa, y2 = 2XY, 'Z = 2XZ,
w = X 2 + a Y 2 + bZ2.
A solution of the first of these is given by
X = X: + uY: + bZ:, Y = 2X1Y1, Z = 2X1Z1,
x = X: - aY: - bZ:.
Then y2 = 4X1Yl(X: + a Y: + bZf),
Z2 = 4X,z,(X:+ U Y ? + b z f ) .
These will be satisfied if we put

X , = x:, Y, = y:, z,= z:,
and X: + ay: + bz: = w:.
The solution of the equation (11) is included in that of finding rational
solutions of
ax4 + by4 + c = z2, (12)
when one (xo,yo, zo) is known.
Put x = xo + pt, y = yo + qt,
wherep and q are parameters. Then the equation (12) becomes a quartic curve
in (z,t) with a known rational point (zo,0). The classic procedure gives in
general another solution in terms of two parameters p, q.
4. Some quartic equations f ( x , y , z ) = 0 which have an infinity of integer

solutions, have been given by Mordell'.
Theorem 1
The equation
z2 = u: -k U2u3, (13)
where Ur = arx2+ h,xy + b,y2 + f r v + grx (r = 1, 2, 3), and the coejicients
are integers, has an injinity of integer solutions fi h: - 4arb, > 0 and is not a
perfect square for either r = 2 or r = 3, and either U2 or U , is absolutely
irreducible.
We have from (13)
z + u -%,, 4 ,,
z-u,=-u
l - 4 P
where p , q are integers and (p, q) = 1. Then
For integer solutions of equation (14), U2 = 0 (mod q), U, = 0 (mod p ) , and

then z is also an integer. Write (14) as
P ( x , y ) = ax2 + hxy + bya +fy + gx = 0, (15)
where a = p2a2 - 2pqal - q2a3, h = pah2 - 2pqhl - q2h3,
b = P2b2 - 2P9bl - gab,, f = pafa - 2pqf1 - qy3,
g = P"2 - 2PRl - q2g,.
The equation (15) has a solution x = y = 0. Hence from Theorem 2,

Chapter 9, it follows that (15) will have an infinity of integer solutions if
P ( x , y ) is absolutely irreducible, and h2 - 4ab > 0 and is not a perfect square.
The condition for reducibility is
2a h g
2A== h 2b f = 0.
g f 2c
This is a binary sextic in p , q and is not identically zero since the coefficient
of p6 is obtained by replacing a, b etc. in A by a2,b, etc. Hence if either U2 or
U, is irreducible, there will be only a finite number of values of p , q for which
P ( x , y ) is reducible.
Next ha - 4ab = (p2h2 - 2pqh1 - qah3)'
- 4(p2a2 - 2pqal - qaa3))(pab2- 2pqb, - q2b3).
If hi - 4a2b2 > 0 and is not a perfect square, this holds for h2 - 4ab if p is
large compared with q and for an infinity of p .
There are many special cases not included in the theorem. We need only
mention
Theorem 2
The equation
z2 = k2 + x2(ax2 + by2), abk # 0 (16)
has an injinity of integer solutions if either b > 0, or b < 0, b2 < 4ak2.
We have
z + k = P9- (ax2 + by2), z -k =f

4
~ 2 ,
where p , q are integers and ( p , q) = 1. Then

(aq2 - p2)x2 + bq2y2 = 2kpq.
This will have the solution x = 0, y = t, where t i s an arbitrary integer if
bqt2 = 2kp, and so if 6 = (b, 2k), we can take
Ap b
= -
6
t2

hq = S
2k A = (P,F).
Hence there will be an infinity of solutions for x , y if b(p2 - aq2) > 0 and is
not a perfect square, i.e. if b(b2t4- 4ak2) > 0 and is not a perfect square. This
is possible if b > 0 for an infinity of values of t, and also if b < 0 and
b2 < 4ak2 for t = 1.
The case b < 0, b2 > 4uk2 seems difficult.
Theorem lo
3 8 9 Qp
The equation
(x2 + a)(y2 + a) = (az2 + b2)2,
where a and b are integers, has integer solutions given by
x = bu + Cdv, y = bu - CdU, z =u = (X + y)/2b.
where c, dare any integers for which
a k b2 = c2d,
and the integers u and v satisfv the equation
f u 2 - do2 = 1.
The equation with the plus sign has an infinity of solutions if d > 0 and is
not a perfect square. The equation with the minus sign may have an infinity of
98 DIOPHANT'INE EQUATIONS
solutions if d c 0 and - d is not a perfect square, for example if - d is a

prime = 1 (mod 4).
Put x = X + Y, y = X - Y. Then
(Xa- Y 2 ) , + h ( X a + Y 2 ) + aa = (azz + b2)2
(Xz- Y a ) , - 2a(X2 - Y 2 ) + a2 = (azz + ba), - 4 u P .
Take X = bZ. Then
b 2 Z 2 - Y 2 - u = f ( a Z 2 - b2).
Hence either
Y2 - (a + b2)ZZ = - (a + b2),
or 1'2 + (a - b2)Z2= - ( a - b2).
The theorem follows on putting Y = cdu, 2 = u.
More generally the method applies to a quartic surface f ( x , y, z ) = 0 if its
section by a plane consists of two conic sections, for example, if identically,
f ( x , Y , ax + + c> = g2(x,u) - h2(X,y),
where g(x, y), h(x, y) are polynomials of the second degree, or its equivalent
f ( x , y, '4 = gYx, Y ) - h2(X, Y ) + I @ , y, z>m(x,y, 4,
where I(x, y , z), m(x, y, z ) are linear and cubic polynomials respectively in
x , y , z. This remark is due to Mordell.
A quartic equation not without interest is given by
arsx'ys = dz2, (18)

r.s=o
where the a's and d are integers (and we may suppose that d > 0). It seems
difficult to find the integer solutions since no general criterion for solvability
exists. However, if aZ2< 0, the term a2,x2y2implies that y, say, is bounded
and then x and z are given by a quadratic equation. When one integer
solution (xo,yo, z,,) of (18) is known, an infinity can be found. This leads to
solutions (xo,y1,zl), ( x l , y1,z2), ( x l , y2, 4, (x2,y2, 4 etc. where from
a Pellian equation, y,, x l , y,, x 2 , . . . may each have an infinity of values.
For rational values of x , y , z, Mordell has proposed the
Conjecture
There exists in general an injinity of rational values for x, y , z.
Put x = X / Y, y = P / Q , z = Z / Q Y where X , Y, P, Q, Z are integers and
( X , Y ) = 1, (P,Q ) = 1.
Then
$ arsXryz-rpaQa-s= dZ2,
r,a = o
or
X2(~z2P2 + azlPQ + a2oQ2) + XY(a12P2 + a l l P Q + U10Q2)
+ Y 2 ( ~ o z P+2 aolPQ + ~ o o Q =~ )dZ2. (19)
Necessary and sufficient conditions can be imposed upon P, Q so that this
equation is solvable for X , Y, Z. Suppose for simplicity that a,, = all = a,,
= 0, and write the equation as
AX2 + B Y a = dZ2. (20)

The common divisors of A and B are finite in number and can be divided
out by applying a linear substitution to P and Q, and replacing2 by k2. Hence
we may suppose that equation (20) is in the normal form, i.e. (X, Z ) = (Y, Z )
= (X, Y ) = 1, ( A , B) = 1, (A, d) = (B, d) = 1, and that A, B, d a r e square
free since square factors can be absolved in X 2 , Y 2 ,Z2.Suppose for simplicity
that (a2z,azl,az0)= 1 = (uo2,uol,uoo) and that d = 1. Then solvability of
(20) is possible only if P, Q are such that A and B are not both negative;
that (20) is solvable mod 8 i.e. we must exclude A = B = - 1 (mod 4); A = 2
(mod 4), B = 5 (mod 8) or B = A + 1 (mod 8), and similarly for B = 2
(mod 4), and that B should be a quadratic residue of every prime factor of A,
and similarly for A and the prime factors of B.
These conditions will be simpler if A and B are both primes. It should not
be too difficult to prove that they can be satisfied.
REFERENCES
+ +
1 . H. W. Richmond. On the diophantine equation F = ax4 by4 cz4 dw4 = +
0, the product abcd being a square number. J. Lond. Math. SOC.,19 (1944),
193-194.
2. B. Segre. On arithmetical properties of quadratic and quartic surfaces. J. Lond.
Math. SOC.,19 (1944), 195-200.
3. Morgan Ward. Eulers three biquadrate problem. Proc. Nut. Acad. Sci., 31
(1 943, 125-1 27.
4. Morgan Ward. Eulers problem on sums of three fourth powers. Duke Math. J.,
15 (1948), 827-837.
5 . L. J. Lander, T. R. Parkins, and J. L. Selfridge. A survey of equal sums of like
powers. Math. Cornp., 21 (1967), 446453.
6. R. D. Carmichael. Diophantine Analysis. John Wiley & Sons, New York
(1915), pp. 4 4 4 8 .
7. L. J. Mordell. On some ternary quartic diophantine equations. Elemente der
Mathernatik., 31 (1966), 89-90.
8. H. C. Williams. A note on a diophantine problem. Unpublished.
9. A. Schinzel et W. Sierpinski. Sur lequation diophantienne ( x 2 - 1) ( y 2 - 1)
= ( ( b - ~ ) / 2 ) - 1). El. Math. 22 (1963), 37-38.
10. K. Szymiczek. On a diophantine equation. El. Math. 22 (1967), 37-38.
CHAPTER 13
Integer Solutions of Some Cubic Equations

in Three Variables
1. Comparatively little is known about the integer solutions of the equation

f(x, Y , 2) = 0, (1)
wheref(x, y , z ) is the general cubic polynomial with integer coefficients. There
are, however, a number of special results, but in general they give some but
not all the solutions. Thus a solution may be given in the form
x = fd0, Y = f&), z = f3(t), (2)
wheref,,f,,f, are polynomials in an integer parameter t .
A solution may also exist in the form
x = f d u , 01, Y = f d u , u), z = f3(u, 01, (3)
wheref,,fz,f3 are polynomials in integer parameters and u and u satisfy a
Pellian equation.
A particularly interesting equation is
x3 + + y3 23 = n. (4)
When n = a3, there are the solutions
x = t, y = -t, z = a;
x = gat4, y = 3at - gat4, z =a - gat3.
When n = 2a3,
x = a(l + 6t3), y = a(l - 6t3), z = --at2.

There are no solutions of equation (4) when n = If: 4 (mod 9), as is obvious
from a congruence (mod 9).
A question arises. If n $ & 4 (mod 9), does the equation (4) have an infinity
of integer solutions? When n = 3, there are solutions given by
(x, Y , z ) = (1, 1, 11, (434, - 5 ) .

Excluding these four solutions, there are no others with max (1x1, I y l , l z l ) 6
3164. This result is due to Miller and Woolettl who have given a table of
solutions for 0 < n < 100.
13. INTEGER SOLUTIONSOF CUBIC EQUATIONS IN THREE VARIABLES 101
These results led to a more extensive table due to V. L. Gardiner, R. B.
Lazarus, and P. R. Stein and produced at the Los Alamos Scientific
Laboratory of the University of California, Los Alamos, New Mexico.
Write equation (4) in the form
x3 + y3 = z3 - d.
The range chosen was
0 Q x Q y Q 65,536, 0 < z - x Q 65,536, 0 < [dl < 999.
No new solution was found for n = 3. It would be of great interest to prove
there are no other solutions and also if one of the equations for which no
solution was found, for example, n = 30, could be proved to have no solutions.
It has been shown2 that the cases n = a3, 2u3 are the only ones for which
the equation (4) has a parametric solution given by polynomials in t of degree
6 4. Segre by geometric considerations has proved the more general
Theorem 1
The equation
ax3 + by3 + cz3 + d = 0, ubcd # 0, (5)
where a, b, c, dare integers, has in general no solutions with x, y , z as relatively
prime polynomials with rational coeficients of degree < 4, in a parameter t.
There are three essentially distinct exceptions,
X3 + y3 + cz3 + c = 0, (6)
x3 + y3 + cz3 + 2 = 0,
x3 + + 2z3 + 2 = 0,
y3
where c # 2r3, r rational.

The solutions of (6) are x = t, y = - t , z = - 1, and
The solutions of equation (7) are

6
1, z = - t2.
C
The equation (8) has in addition to the three solutions given by c = 2, the
further solutions
x = 4t2 - 6t - 1, V. = 4t2 - 2t - I , z = -4t2 + 4t - 1;
and 2 7 , ~= 2(4t4 - 4f3 - 6t2 + 17t - 2),
27y = 4(2t4 - 8 t 3 + 61' + 4t - 13),
27: = -8t4 + 20t3 - 24t2 - 16t + 37.
By putting t = 1 + 9r,, we can reduce the denominators of x, y , z to 3.
An instance of equation (3) is now given.
Theorem 2
The equation
ax3 + ay3 + bz3 = bc3, abc # 0, (9)
where a, 6, c are integers, has an infinity of integer solutions other than the
trivial ones x + y = 0, z = c.
The result is a simple consequence of the existence of the rational generator
+
x y = 0, z = c on the cubic surface defined by the equation (9).
Put z = c + t(x + y ) ,
where t is an arbitrary integer parameter. Then dividing out by x + y,
a(.? - xy + y 2 ) + 3bc2t + 3bct2(x + y ) + bt3(x + Y ) =~ 0.
Put x+y=u, x-y=v.
- (u2 + 3 ~ ' )+ 3bc2t + 3bct2u + bt3u2 = 0,

a
Then
4
or -3av2 = (a + 4bt3)u2+ 12bct2u + 12bc2t.
If an equation
Do2 = Au' + BU + C,
where A , B, C, D are integers, has one integer solution, it will have an
infinity if A D > 0 and is not a perfect square, and if B 2 - 4AC # 0. Here a
solution will be given by u = 0 if 3av2 = - 12bc2t,that is, if t = -abk2 where
k is an integer. Then A D = 3a(4a3b4ks- a) > 0 for k # 0, and is not a
perfect square for an infinity of values of k.
Also B 2 - 4AC = -48bc2t(bt3 + a) = 0
only when t = 0 or bt3 + a = 0. Hence there are an infinity of integer solu-
tions for u, u. We have now to show that we can find integer values for x, y.
The condition for this is u = v (mod 2). Suppose that a = 2aa',CI 3 0, a' odd.
Take t = 0 (mod 2"). Then the u, v equation gives
-3a'v' = a'u2 (mod 2), or u = v (mod 2).
+ +
In the particular case x3 y3 z3 = 1, the well known solution
(x, y, z ) = (9, -8, - 6 ) corresponds to t = -7. We have then also solutions
(x, y , Z ) = (- 103, 94, 64), (904, -823, - 566), (3097, -2820, - 1938).
Lehmer5 has dealt in some detail with this equation. He has given re-
currence formulae which lead to an infinity of solutions expressed as poly-
nomials in a parameter t .
13. INTEGERSOLUTIONS OF CUBIC EQUATIONS I N THREE VARIABLES 103
2. Theorems 2 and 3 are practically the only known results about equations
of the form ax3 + by3 + cz3 = d. Except for the trivial case n = & 4 (mod 9),
there are not known any impossible equations,
x 3 + y3 + z 3 = n. (4)
However, Hurwitz6 has shown by considering the number of lattice points
and the volume enclosed by the obvious surface, that there are an infinity of
values of integers n f 2 4 (mod 9) for which there are no solutions in
positive integers.
Among the simplest cubic equations are those given by z2 = f ( x , y ) . There
arises the
Conjecture
The equation z2 = f ( x , y), where f(x, y ) is a cubic polynomial in x, y with
integer coefficients has an infinity of integer solutions $it has one.
We may suppose the solution is given by x = y = 0. There are some special
cases when the conjecture is true.
Theorem 3
The equation
z2 = p 2 + Ix + my + ax3 + bx2y + cxy2 + dy3, (10)
where (1, m) = 1 , has an infinity of integer solutions.
By making a linear transformation, we may suppose that I = 1, m = 0.
If p = 0, solutions are given by x = 0, z2 = dy3.
I f p # 0, put x = 4p2X, y = 2pY, z = pZ. Then
Z2= 1 + 4X + 64ap4X3 + 32bp3X2Y + 16cp2XY2 + 8dpY3.
Put now 2 = + 2 X - 2 X 2 , Z 2 = 1 + 4X - 8 X 3 + 4 X 4 .
1
Then X4 = (2 + 16ap4)X3+ 8bp3X2Y + 4cp2XY2 + 2dpY3.
Solutions of this are given by putting Y = t X , t an integer parameter, since
then
X = 2dpt3 + . . . + 2 + 16ap4.
Similarly we can prove
Theorem 4
The equation
z3 = p 3 + Ix + nzy + a.v2 + bxy + cy2, (I, m) = 1 (1 1)
has an infinity of solutions.
A parametric solution is given by taking
1 = 1, m = 0, x = 3p2X, z =p + X, y = tX.
Theorem* 5
The equation
z2 = p 2 + Ix + my + ax2 + bxy + cy2 + Ax3 + Bx2y
+ Cxy2 + Dy3, p # 0, (12)
has an injinity of integer solutions ifp I (I, m), and if the equation
ax2 + bxy + cy2 - (Ix + m ~ ) ~ / =4 p+ ~2 p ,
has an injinity of integer solutions.
This will be so if there is one solution, e.g. x = 0, y = 1, when m = 0,
c = + 2 p , and if
p2(b2 - 4ac) + am2 + cI2 - blm > 0,

and is not a perfect square.
Write the equation as
z2 = p 2 + L , + L2 + L3,
where L1,L2,L3 are the terms of the first, second and third degrees respec-
tively. Put
z = p + L1
- + s,
2P
2ps = L2 - - L?
*
4P2
Then z2 = p 2 -+
+ L1 + 4P
L2
+ 2 p s + L1
P
S
+ s2.
Hence x, y satisfy the equation
Put x = t X , y = t Y where t is an integer parameter.

Then
By hypothesis there are an infinity of integer values for X , Y for which the
coefficient of t is one. Hence the theorem.
3. Even in the simplest case of the equation z2 = f ( x , y), namely,
za = ax3 + by3 + c, (13)
13. INTEGER SOLUTIONS OF CUBIC EQUATIONS IN THREE VARIABLES 105
very little is known. In special cases, solutions may be found. Thus if c =
( 4 ~ b d ~a )solution
~, is given by
x = up2, y = bq2, z = a2p3 + b2q3, pq = 2d2,
wherep, q are integers. This gives a finite number of solutions depending upon
the number of factors of d, and so equations can be constructed having at
least as many integer solutions as can be desired.
There are some cases where an infinity of integer solutions can be found, as
is shown in Chapter 14, for the equation
) ~ab2x3 + y 3 .
z 2 - ( 2 7 ~ b c= (14)
Results may sometimes be found when a solution is known, occasionally
rather trivially as, for example, with
22 = x3 + y3 - 1.
Put x = I + w , y = 1 - wand so z2 = 6w2 1. +
Hence there are an infinity of integer solutions with x y = 2. +
More generally let (x, y, z ) = ( p , q, r), r # 0 be a solution of equation (13).
+
Put x = p + t X , y = q t Y, where X , Y, t are integers.
Then
z2 = r 2 + 3(up2X + bq2Y)t + 3(upX2 + bqY2)t2+ ( a x 3 + bY3)t3.
z = r + - (up2X + bq2Y)t + kt2,
3
Put
2r
where k is an integer, and
2kr + 4r9
2- (ap2X + bq2Y)2= 3(upX2 + bqY2),
3k
and so k2t +7 (ap2X + b$Y) = aX3 + by3. (16)
It may happen that for suitable k , the quadratic equation (15) in X , Y may
have an infinity of solutions. Then t is determined by equation (16), and its
only possible denominator is a divisor of k2r. If an integer value of t arises,
equation (15) shows that z is an integer.
We apply these results to prove
TheoremD6
The equation
z 2 = (612 + 61 - l ) x 3 + (612 - 61 - l)y3 + 11 - 12f2, (17)
where I is an integer # O , has an injnity of integer solutions.
Take k = f 1, r = 3, in equation (16), then

t = a x 3 + b Y 3 T (ap2X + bq2 Y )
is an integer. Also
3(apX2 + bqY2) - ~ ( u P ' X + bq'Y)' = -16.
If there is one integer solution in X , Y there will be an infinity if the quad-
ratic in X , Y is indefinite and rationally irreducible. Suppose, for example,
that p = q = 1, and that X = 1, Y = - 1 is a solution.
Then + b) - +(a - b)' = f 6 .
3(a
Put u - b = 121, u + b = 121' & 2,
and so u = 61' + 61 & 1, b = 61' - 61 f 1.
The discriminant A of the quadratic is given by
4A = a2ba - ( 1 2 ~- a2)(12b - b'),
= 12ab(a + b - 12).
Then A > 0 if a and b are positive and a + b > 12, or
if ab < 0 and
a + b < 12.
The quadratic is irreducible if A is not a perfect square. Clearly
(a, b) = 1 , (ab, 6) = 1 , (a + b - 12, 3 ) = 1.
Also
+ b - 12) = (a, u - b + 12) = (a, 121 + 12) = (a, I + 1) = 1.
(a, u
Hence A can be a square only when a, 6, 3(a + b - 12) are all perfect squares.
This is impossible since (a + b, 3 ) = 1. Hence we may take with I # 0,
a = 61' + 61 + 1 , b =612 - 61 + I , if c = 7 - 1212,
and a = 61' + 61 - 1 , b = 61' - 61 - 1, if c = 1 1 - 121'.
The discussion of the equation
z3 = UX' + by2 + c
requires a simple application of algebraic numbers. This will be given in
Chapter 14, with simple conditions for the existence of an infinity of integer
solutions.
4. We consider finally the integer solutions of the cubic equation
x2 + y2 + 'Z
b. - U X ~ Z= (19)
The case a = 3, b = 0 is classic. Markoff proved that there are an infinity of
solutions, and that these were given by a simple algorithm, the so-called
Markoff Chain. When b < 0, the equation (19) is a particular case of the
equation
X? + Xg + . . . + x;. . . - XXIXZ.. .x, = 0 (20)
in n + 1 positive unknowns x , sl,. . ., x,, some of which are equal to
one. The equation (20), it will be noted, is related to the equation z = f ( x , y ) / g
( x , y ) considered in Chapter 30. Hurwitzs made an exhaustive study of the
equation (20). An account is given of his ideas as well as of those of Mordell'O
and Schwartz and Muhlyll as applied to equation (19).
Let ( x , y , z ) be any solution of equation (19). A trivial solution is one in
which two of the unknowns are zero. It can occur only when b is a perfect
square. A positive solution is one in which x > 0, y > 0, z > 0. From any
solution (x, y , z ) of (19), other solutions can be derived by the following
elementary operations: permuting x, y , z ; changing the signs of any two of
s,y , z ; and what is most important, replacing x by ayz - x, or y by azx - y,
or z by axy - z. The last statement is obvious on considering equation (19) as
a quadratic in x for given y , z.
There is no loss of generality in supposing that a > 0. We show now that
we may suppose that the non-trivial solution ( x , y , z ) is a positive solution
and we shall call x + y + z its height. It suffices to assume that xyz # 0, for
if x = 0, another solution is (ayz, y , z). If x > 0, y > 0, z < 0, the solution
(x,y , asy - z ) is a positive solution. The solution ( x ,y , z ) is called funda-
mental if 0 < x I y Iz , and if no elementary operation produces a positive
solution of smaller height. The definition of the height gives at once the
Theorem 7
I f a non-tricial solution exists, it can be derived from a fundamental solution
by a$nite number of elementary operations.
The cases b > 0 and b I 0 must be considered separately. Suppose first that
b > 0.
Theorem 8
I f (x,y , z ) is a fundamental solution of equation (19), then x2 + y2 < b
except when a = 1, b = 4, or a = 2, b = 1.
For either axy - z I 0 and then from (19), x2 y2 I + b, or
axy - z 2 z, axy 2 2z. (21)

From (19),
3z2 - axyz 2 6, z(3z - axy) 2 b, 3z - axy > 0.
Then from (21),
Write (19) as
Then (22) gives

x2 + y 2 + &-a2x2y2- &a2x2y22 b,
and since x 5 y ,
(
2y2 1 - -
T)>b.
Hence ax < 3 and so a = 1, x = 1,2, or a = 2, x = 1 .

When a = 1, x = 1, and then equation (21) gives y 2 22 whereas y I z.
Hence the only possible fundamental solutions are those with x2 + y2 I b.
When z = 1, x = 2, equation (21) gives 2y 2 22 and soy = z. Now equation
(19) becomes ( y - z ) ~= b - 4 and so b = 4. Then all the solutions (2, y , y ) ,
y L 2 are fundamental since their heights cannot be decreased. These are the
only possible fundamental solutions with x2 + y2 > 4. When y = 1 , the
solution (2, 1, 1) is a permutation of the fundamental solution (I, 1 , 2). This
and (2, y , y ) , y L 2 are the only fundamental solutions of x2 + y 2 + z2 -
xyz = 4. When a = 2, x = 1, equation (19) becomes ( y - z ) = ~ b - 1. Also
2y L 22 and so y = z, and b = 1. The solutions (1, y , y), y L 1 are funda-
mental.
Theorem 9
I f a non-trioial solution exists, there will be an infinity of solutions except when
u = l , b = 2.
For the special cases a = 1, b = 2, i.e. x2 + y2 + z2 - xyz = 2, there
exists only one fundamental solution, namely (1, 1 , l), since only for this is
x2 + y2 5 2. The elementary operations produce only the further solutions
s u c h a s ( k 1 , k l , O ) , ( l , - l , O ) , ( l , -1, -1).
We shall now show that a solution can be found with a height greater than
that of a given solution. We consider in order the cases a > 2; a = 2, b = 1 ;
a = 2 , b > 1;a= 1,b>2.
a > 2
If a non-trivial solution exists, there will be a solution (x, y, z ) with
x 5 y I z. Then ( y , z, ayz - x ) is a solution of greater height since ayz -
x > x from yz > x.
The infinity of solutions is also obvious from a Pel1 equation in two of the
variables.
a=2,b=l
There are solutions ( I , y , y ) with infinitely many fundamental solutions.
a=2,b>l
Since (1, 1, I) is not a solution, at least one of x, y , z exceeds 1, say z 2 2.
Then ayz - x > x as before since yz 2 2x.
~ = l , b > 2
Since ( 1 , 1, 1 ) is not a solution, we may suppose that z 2 2 in a fundamental
solution (x, y , z). The cases ( I , 1, 2), (1,2, 2), (2, 2,2) are the only possible
ones with z I2. If either (1, 1, 2) or (2, 2 , 2 ) is a fundamental solution, b = 4,
and it has been shown that then there are an infinity of solutions. For the
solution (1,2,2), and those with z > 2, we have as before uyz - x > x etc.
Trial of the possible fundamental solutions of equation (19) seems to be the
only method for determining whether (19) is solvable. A necessary condition
is that if b = 3 (mod 4), then a = 0 (mod 4). This follows from the con-
gruence
x2 + y2 + z2 - axyz E 3 (mod 4).
This is impossible if any of the x, y, z are even. Hence
x =y =z = 1 (mod 2), axyz E 0 (mod 4), a = 0 (mod 4).
Suppose next that b I0, say b = -c. Then from equation (19)
z(axy - 32) I c.
Two cases arise according as axy - 3z > 0 and then x Iy I z I c, and
such values can be found by trial, or axy - 32 I 0.
Then from equation (21),
3uxy Iz I faxy,
2yy1 - +Zx) 2 -c.
Suppose first x < 3 / a and so a = I , x = 1 , 2 or a = 2, x = 1.
These values are impossible since they make the left-hand side of equation
(19) positive. Hence
x 2 3/a, 2y2(4u2x2 - 1) I c,
and so if c # 0, x, y , z are bounded and the fundamental solutions are easily
found by trial.
If c = 0, then x = 3/a and so a = 3 or 1.
If a = 3, x = 1 and then y I z I +y. But
0 = 1 + +
y2 22 - 3yz I 1 + + $y2 - 3y2.
y2
The only solutions are y = 0, 1 and so we have a fundamental solution

( I , 1, 1) for Markoffs e q ~ a t i o n l ~ ~ ~ .
If a = 1 , x = 3 and then 3y I z I 3y/2, and now
0 = 9 + y2 + 22 - 3yz.
We have y = 3, z = 3 and the fundamental solution (3, 3, 3).
I10 DIOPHANTINE EQUATIONS
We note that when c = 0 and a > 0, solutions occur only when a = 1, 3.

The more general equation
ax2 + b y 2 + cz2 + dxyz = e (24)
is not easily amenable to treatment.
A simple result is given by
Theorem 10
If a = b = c = - d $ 0 (mod 3), e = f 3 (mod 9), the equation (24) has
no integer solutions.
Obviously x = y = z = 0 (mod 3) is impossible.
From (24),
x2 + y 2 + z 2 - xyz = O(mod 3).
Then x f 0 (mod 3) since otherwise y = z = 0 (mod 3).
Also xyz = f 1 (mod 3) gives 1 3 0 (mod 3), and can so be ruled out.
REFERENCES
1 . J. C. P. Miller and M. F. C. Woolett. Solutions of the diophantine equation
+
x3 y 3 + z3 = k . J . LO^. Math. SOC., 28 (1955), 500-510.
2. L. J. Mordell. On sums of three cubes. J. Lond. Math. Soc., 17 (1942), 139-144.
3. B. Segre. On the rational solutions of homogeneous cubic equations in four
variables. Mathematicue Notae (Rosario, Argentina), 11 (1951), 1-68.
+ +
4. L. J. Mordell. On the infinity of integer solutions of ax3 ay3 bz3 = bc3.
J. Land. Math. SOC., 30 (1955), 1 1 1-1 13.
+
5. D. H. Lehmer. On the diophantine equation x3 + y 3 z3 = 1 . J. Lond.
Math. SOC.,31 (1956), 275-280.
6. A. Hurwitz. Uber die Darstellung der ganzen Zahlen als Summen von den
Potenzen ganzer Zahlen. Math. Annufen., 65 (1908), 424-427; also in Math.
Werke (Birkhauser-Cie, Basel), 2 (1933), 421-426.
7. L. J. Mordell. On cubic equations z2 = f ( x , y) with an infinity of integer
solutions. Proc. Am. Math. SOC.,3 (1952), 210-217.
8. L. J. Mordell. Note on cubic diophantine equations z2 = f ( x , y) with an
infinity of integer solutions. J. L o r d Math. SOC.,17 (1942), 199-203.
9. A. Hurwitz. Uber eine Aufgabe der unbestimmten Analysis. Archiu der
Math. und Phys., 3 (14) (1907), 185-196; also Math. Werke (Birkhauser-Cie,
Basel), 2 (1933), 410-421.
+ + +
10. L. J. Mordell. On integer solutions of the equation x2 y 2 z2 2xyz = n.
J. Lond. Math. Soc., 28 (1953), 500-510.
1 1 . H. Schwartz and H. T. Muhly. On a class of cubic diophantine equations.
J . Lond. Math. SOC.,32 (1957), 379-383.
12. A. A. Markoff. Sur les formes quadratiques binaires indkfinies. Math. Ann., 15
(1879), 381-409.
13. J. W. S. Cassels. An Introduction to Diophantine Approximation.
Cambridge University Press (1957).
CHAPTER 14
Simple Algebraic Considerations
1. Some results require only the definition of an algebraic number, namely,

that 6 is an algebraic number if it is the root of an equation, reducible or
irreducible, over the rational field Q,
6" + + a26n-2+ . . . + a, 0, =
where the a are rational numbers. Many of these results are classical and are
due to Euler and Lagrange.
The equation
ax2 + by2 = cz", (1)
where a, b, and c are integers.
Suppose first that n is odd. Consider the classical case when c = 1.
Some integer solutions may be found by putting
z = up2 + bq2,
where p and q are arbitrary integers, and taking
xda + yd-6 = ( p d a + qd-b)",
Then x, y are expressed as polynomials in p and q.

Occasionally as will be seen in Chapter 15 all the integer solutions may be
obtained in this way. The proof requires arithmetic theory.
Suppose next that n = 2m is even. Then unless a = 1 when a solution is
given above for all n, the question becomes more difficult. From Chapter 7,
it is seen that now the problem is to find integer solutions for X,Y of the
equation
zm = A X 2 + B X Y + CY2
for a finite set of integer values of A and B.
These can be reduced to the form (1). Then if c = c; + abc; or c =
ac; + bc; solutions may be found as before by factorizing c.
The equation
where 0 = el, 02, e3 are roots of the equation

t3 + at2 + bt + c = 0,
where a, b, c are integers.
A partial integer solution is given by
x + ye1 + 28: + qe, + re:).,
= (p
x + ye, + 20;= ( p + qe2 + re:).,
x + ye3 + 2% = ( p + 4e3 + re;).,
w = n ( p + qe + rez),
e
where p, q, r are arbitrary integers and n runs through the integers.

The general solution depends upon the theory of algebraic numbers and is
connected with the units in an algebraic number field.
The equation
Z2 = ax3 + bx2y + cxy2 + dy3, (3)
where a, b, c, dare rational numbers.
All the rational solutions are given at once by putting x = pz, y = qz
where p and q are arbitrary rational numbers. Finding the integer solutions
is a different matter. However, as shown by Lagrange, some integer solutions
are found without much difficulty when a, b, c, d are integers and a = 1. Let
0 = B,, Oz, O3 be roots of the equation, reducible or irreducible,
t3 + bt2 + ct + d = 0.
Write z2 = ( x - e,y)(x - 8zy)(x - 03Y),
and x - ty = ( p + qt + rt2)2, t = d,, 02, 8,,
where p, q, r are integers. This implies that when we replace t 3 by - bt2 -

ct - d and t4 by
-bt3 - eta - dt = - b ( - b t 2 - ct - d ) - Ct2 - dt,

the coefficients of t 2 in the above must be zero.
Hence q2 + 2pr - 2qrb + r2(b2- c) = 0.
This gives an integer value for p if r(b2 - c ) is even and 2r I qa. These con-
ditions are easily satisfied, for example, by r = 2rl, q = 2rlq1. Clearly then
x, y and z = N ( p + qe + re2) are integers.
The complete integer solution requires arithmetic theory and is given in
Chapter 25.
14. SIMPLE ALGEBRAIC CONSIDERATIONS 113
2. The application of both quadratic and cubic irrationalities is sometimes
useful. This introduces many parameters satisfying a few equations, and so it
is not difficult sometimes to find particular solutions. A simple instance' is
given by
Theorem 1
The equation
y2 - ax2 = z3 + bz + c, (4)
where a # 0, and b, c are rational, has a two parameter rational solution.
Let O = 01, O,, O3 be the roots of the equation
t3 + bt + c = 0.
Put y xd/a = n e
( p + 982 (r + s g d i ) ,
where p , q, r, s are rational parameters. Clearly these two equations define

x, y as rational numbers and
y 2 - ax2 = n((p + q q 2
E
- a(r +
Hence equation (4) is satisfied if
z -t = (p + qt2)2 - a(r + st)',
for t = el, 02, 03, since
z3 + bz +c=n t=e
(Z - t).
Multiply out, put t 4 = - bt2 - ct and equate coefficients of to, t, t 2 on both

sides. Then
z = p2 - ar2, -1 = -cq2 - 2ars, 0 = 2pq - bq2 - as2.
These give
P =
bq2 + as2 1 - cq2
r=--------,
29 24lS
i.e. a two parameter solution.

Another application2 is to the
Theorem 2
The equation
z 3 = ax2 + by2 + c (5)
has an injinity of integer solutions f a , b, c are odd integers, (a, b) = 1, and if
when a = 0 (mod 7), c f b3 (mod 7 ) ,or when b = 0 (mod 7), c f u3 (mod 7)
and so ifab = 0 (mod 7), it sufices if c f i-1 (mod 7).
Denote the three roots of t 3 = c by 0 = el, 02, 0,. Then, say,

ax2 + by2 = ( z - - e2)(z - e), =J-(~),
Put
z - 8 = a(l + me + n02)2 + b(p + 48 + r02)2 = a x 2 + b y 2 , (6)
say, where I, m, n, p , q, r are rational integers. Then f ( z ) can be expressed in
the form
f(z) = (ax2 + bY:)(aXi + b Y i ) ( a X $ + b y ! ) ,
and so we can take
x d i +y d Z = n ( X , d i + YJmb).
3
r=l
Then x = aX1X2X3 - b X l Y z Y 3 - bX2Y3Yl - b&YlY2,

y = -bYiY2Y3 + aXIX2Y3 + aXzX3YI + u X ~ X ~ Y Z ,
and so x, y are rational integers.
Expanding equation (6) and equating coefficients of 8, O2 on both sides, we
have the two equations,
2mal + 2qbp = -1 - an2c - br2c,
(7)
2rbp
2nd i- = -am2 - bq2.
These are linear in I, p and we have four variables in m, n, q, r at our disposal
to ensure that I and p are integers. The right-hand sides of equations (7) will
be even numbers if
m = q = 1 (mod 2), n + r = 1 (mod 2).
We impose the condition mr - qn = 5 1. Then I, p will be integers if
bq3 - r - bcr3 = O(modu), -am3 + n + ucn3 = O(mod b).
Solutions are shown to exist for these congruences from the result in Chapter 6
on the solvability of
y2 = x3 + k (mod a).
No difficulties arise in the proof which now is straightforward.
When a = b = 1, a solution is given in terms of an integer parameter t, by
x + iy = n(1+ te - +(t2 + i ) P + i(t + e)),
e
z = 1 - ct(t2 + 1) + 1 + 2t + 4-C ( P + 1)2

where i 2 = -1, e3 = c.
A particular case3 of the equation
z 2 = ax3 + by3 + c
is given by
Theorem 3
The equation
z2 - ( 2 7 ~ b d=
) ~ab2x3 + y3, ab # 0, (8)
where a, b, d a r e integers, has an injnity of integer solutions.
Consider the equation
z2 - k2 = ab(x3 + cy3), abc # 0, (9)
Denote by f3 = B,, 02, O3 the roots of t 3 = c. Take
z+k = an ( p + q6' + r02),
e
(10)
z -k = bn + qlB + r102),
(p1
R
where the p , pl, etc. are integers.

Then
z2 - k 2 = ab n(P +
R
Q6' + R P ) = ab(P3 + cQ3 + c2R3 - 3cPQR),
and we take
P = ppl + c(qrl + q l r ) = x,
Q = pq1 + Pi4 + crri = Y ,
R = prl + p l r + qql = 0.
From equations (10) and ( 1 I),
2k = a(p3 + cq3 + c2r3 - 3cpqr) - b(py + cq,3 + c2r: - 3cplqlrl). (13)
The six variables p , q, r, pl, q l , rl satisfy the two equations (12) and (13) and
particular solutions may be found without difficulty.
Take p 1 = q, q1 = -r, rl = 0. Then equation (13) is satisfied.
Also x = pq - crz, y = -pr + q2, z - k = b(q3 - cr3), (14)
2k = a(p3 + cq3 + c2r3 - 3cpqr) - b(q3 - cr3). ( 1 5)
Take now c = b/a. Then equation (15) becomes
2k = up3 + 2b2r3/a- 3bpqr,
and equation (9) becomes
z2 - k2 = abx3 + b2y3.
Also in equation (14), s and z will be integers if br2/a and b2r3/aare integers.
Take nowp = 3bX, q = Y, r = 3aZ, k = 27ab2d. Then x and z are integers
and x 3 0 (mod b), z E 0 (mod 6).
Replacing x and z by bx, bz, equation (17) becomes
z2 - ( 2 7 ~ b d )=
~ abax3 y3. + (18)
Also equation (16) becomes
2d = b X 3 + 2aZ3 - X Y Z .
This equation is of the first degree in Y and has an infinity of integer solutions
as will be noted in Chapter 30,
3. Simple algebraic considerations sometimes suffice to find the solutions of
an equation in a given algebraic number field K. Consider the equation
f(x,y) = 0 in the field K = Q ( t ) defined by an irreducible equation F(t) =
0 of degree n.
Then x = xo + x,t + +
' . * x , - p l = x(t>,
y = yo + y,l + - + y,-ltn-l
* * = yo),
where the xo,y o etc. are rational numbers.

Since f(x(t),y(t)) must be divisible by F ( f ) , we have an identity in a
variable z given by
S(X(Z)>, Y ( d ) = F(z)G(z),
where G(z) is a polynomial in 2. In some instances, G(z) is linear and so the
left-hand side must vanish for a rational value of z.If the equation f ( x , y ) = 0
has only a finite number of rational solutions, for example, when f ( x , y ) =
+
x4 y 4 - 1, some progress may be expected. In this way, we can prove
results due to Aigner6 and Fadeevs respectively.
Theorem 4
When K is a quadratic field, the only non-rational solutions of the equation
x4 + y4 = 1 are given by
K = Q(i), x = fi, y = 0, x = 0, y = ki,
& a -- .?2I -
- e2a -- 1.
When K is a cubic field, then the only solutions, except for rational ones,
are given by
y = +i+ tX, x4 + ( i - 1 + t 4 4 = 1,
(20)
or x = f 1 + ty, y4 + (It 1 + ty)4 = 1.
where t is a rational parameter.
For the tirst part. on putting I - .P = ty2, we obtain the parametric
solution,
I - t2
- = 2t
-
x2 =
1 + t2
42
1 + 12
and so t is in K. Write
x = (1 + t2)xy, Y =+ ty.
(1
Then x 2 = 2r(l - t2), Y 2 = 2t(l + P),

and it is known that the only rational solutions of the first equation are
t = 0, k 1, and of the second, t = 0, 1. If t is rational, then since either
P / Y 2 or Y 2 / X 2is rational, K can only be the quadratic field Q ( i ) arising
from t = - 1 or co. Suppose then that t is not rational.
We can put
X =a + bt, Y = a, + b,t,
where a, b, a,, b, are rational.
Then we have the two identities in z,
+ b ~ -)2z(l~ z)
(a - = F(z)(P + Qz),
(a1 + b,~) - 2 ~ ( 1+ z) 1 F(z)(Pl + Q~z).

Hence the first left-hand side must vanish for z = 0, i-1 and the second for
z = 0, 1 , and so there are six possibilities for the pairs P + Qz, P, + Qlz.
Thus if these are 1 + z, 1 - z , then a = b, and we may take a, + bl = 2,
and on dividing out by 1 + z, 1 - z, respectively,
a2(1 + z ) - 2z( 1 - z ) = PF(z),
(2 - b,)2 + (2 - b?)Z + 222 = P,F(z).
Hence == (2 - b,), a2 + bf = 4,
and so a = 0, 6, = 2 which must be rejected, and

a = k2 = b, al = 2, b, = 0.
These give
F(z) = Z + 2 + 2, t =
-1 + d-7 K = Q(2/-7),
2
and then the values for x, y.
It is easily shown that the other five possibilities do not lead to a solution.
Similar Considerations apply to the proof of the results for a cubic field.
5+
Another application is given by

Theorem 5
Let, a, b, c be rational integers where (b, c) = (c, a ) = (a, 6 ) = 1, and a and
b are square free.
Then the integer solutions of the equation
ax2 + by2 + c = 0 (21)
can exist in a quadratic field Q ( t ) if and only if there exist rational integers
p , q, d, dl such that
up2 + bq2 = d, (UP, bg) = d l , (22)
and either (A), d is some divisor of abc, and t satisfies the equation
+ abk2/dl + c/d = 0,
t2 (23)
and k is a rational integer such that abk2/dl + c/d is an integer.
Then
x = Pt + bqk/d,, Y = qt - apk/d,; (24)
or (B), d = 0 (mod 2), d is some divisor of 2abc, and t satisfies the equation
t2 + t + t ( l + abk/d,2) + c/d = 0, (25)
where k is an integer such that
(1 + abk/d:)/4 + c/d
is an integer.
Then
x = Pt + 3(P + b q k l 4 , Y = qr + 3(4 - aPk/dl). (26)
We can put
x = Pt + P1, Y = qt + 41, (27)
where p , q, p l , q1 are rational integers. There is no loss of generality in sup-
posing that ( p , q ) = 1, and so t is an integer. Then since (a, b) = 1, from
(221, di I ah.
On substituting in (21) for x , y from (27), t becomes a root of the equation
(up2 + bq2)t2+ 2(appl + bqql)t + up: + bq? + c = 0. (28)
From (22), since t is an algebraic integer,
2@PP, + bqq,) = 0 (mod 4,
up: + bq: + c = 0 (mod d ) .
The solution of (29) for p l , q1 can be written as
14. S I M P L E ALGEBRAIC CONSIDERATIONS 119
since it is easily verified that these equations give integer values for I, k in
terms of p l , ql.
Substituting from (31) in (28), we have
(upz + bq) t2 + ,(upz + bqz)t + $((up2 + bqz)12 + ubk2(ap2+ bqz)/d:)
+c=o,
or
t2 + It + $(I2 + abk2/d,2)+ c/d = 0, (32)
where k is an integer such that
L = &(12 + abkz/dI) + c/d (33)
is an integer.
On writing t - t o for t in (32), we need only consider the values I = 0, 1.
Suppose first that I = 0 giving case (A). Then from (31),
2p1 = bqk/d1, 2q1 = -apk/dl.

Since (bqld,, upld,) = 1 , k is even. Replacing k by 2k,
p1 = bqk/d,, 41 = -UPk/d,.
and (33) becomes L = ubk2/dl + c/d.
Since ubL and ubld, are integers, d I ubc. Then we have (23) and (24).
Suppose next that I = 1 giving case (B). Then from (33),
L = &(l + abk2/d:) + c/d.

From (31), on multiplying the equations by up, bp and adding, d = 0 (mod 2).
Suppose first that dl = 0 (mod 2). Then since up and bq are both even and
(u, b) = ( p , q ) = 1, we have, say,
a = 0 (mod 2), b = 1 (mod 2), p = 1 (mod 2), q = 0 (mod 2),

and so u/2, and d/2 are both odd. Hence +(I +
ubk2/d:) must not have a
denominator divisible by 2 but this is impossible since ubkz/df will have a
2 in the numerator. Hence d, = 1 (mod 2), and then three possibilities arise
of which two are typified by up = 0 (mod 2), bq 2 1 (mod 2), and from (22),
produce the contradiction d = 1 (mod 2). Hence up = bq = 1 (mod 2), and
then from (31), k is odd.
Now
2 L ~ b= +(ab
z+I:(- k) + 7.
2ubc
The first term on the right is an integer a n d so d 1 2abc. We now have (25)
and (26).
This finishes the proof.
Finally, we notice some results by Fjellstedt' on the solvability in a
quadratic field Q(d-m), m > 0 of equations of the form x 2 - ciy2 = /I
where a,/3 are integers in the field Q ( d G )H.e discusses the cases /3 = 1
by methods due to Nagell but his paper contains many misprints. He gives
many references.
REFERENCES
1. L. J. Mordell. A rational parametric solution of z2 - k = ax3 + bx2y +
cxy2 + dy3. J . Lond. Math. Soc., 18 (1943), 222-226.
2. L. J. Mordell. Note on cubic equations in three variables with an infinity of
integer solutions. Ann. Mat. Pura Appl., IV., 29 (1949), 301-305.
3. L. J. Mordell. The congruence ax3 + by3 + c = 0 (mod xy), and integer
solutions of cubic equations in three variables. Acta Mathernatica, 88 (1952),
77-83.
4. L. J. Mordell. The diophantine equation x4 + y4 = 1 in algebraic number
fields. Acta Arith., 14 (1968), 347-355.
5. A. Aigner. Uber die Moglichkeit von x4 + y4 = z4 in quadratische Korper.
Jahresbericht der deutschen Math. Verein., 43 (1934), 226-228
6. D. K. Fadeev. Group of divisor classes on the curve defined by the equation
x4 + y4 = 1. American Math. SOC.Trans. Soviet Maths. Dokl., 1 (1960), 1149-
1151.
7. L. J. Mordell. The integer solutions of the equation ax2 + b y 2 + c = 0 in
quadratic fields. Bull. Lond. Marh. Soc., 1 (1969), 43-44.
8. L. Fjellstedt. On a class of diophantine equations of the second degree in
imaginary quadratic fields. Ark. Mat., 2 (1954), 435-461.
CHAPTER 15
Applications of Algebraic Number Theory
1. A great many results have been found by applying this theory, and many
of the most important results in Diophantine analysis have been proved in
this way. Let K = Q(8) be an algebraic number field defined by a number 8
given as a root of an irreducible equation of degree n with rational integer
coefficients and leading coefficient unity. Then every integer w in K can be
expressed in the form
W = XlW, + .. + xp,,
where the x are rational integers and the w are a basis of the integers. The
integers of the form
= xo + x , +~ . . . + x,-len-l
form a ring Z[O],say.
A unit 7 in K or Z[O] is an integer divisor of unity and so N ( 7 ) = & 1. All
the units are given by Dirichlet's
Theorem 1
Suppose that K has rl real conjugate fields and r2pairs of conjugate imaginary
+
fields. Write r = rl r2 - 1. Then every unit 7 in the field K can be expressed
in the form
t t
rl = 7:7:...7:,
where qo is a root of unity in the field, ql,7 2 ,. . . , yr are the so called funda-
mental units, and the t take all integer values, positive, negative or zero.
An obvious consequence is that if I is a given integer, all the units can be
expressed in the form
7 = tot',
where to,5 are units and tobelongs to a finite set.
Similar results hold for the units in the ring 2 [ O ] . This Theorem 1 and the
next are of the greatest importance in the applications.
Theorem 2
The number h of ideal classes in K is finite. I f a is any ideal in K , then
ah = (PI,
where P is an integer in K. I f a' is a principal ideal and ( I , h) = 1, then a is a
principal ideal (a).
These theorems are easily applied. There are others which require more
detailed arithmetical knowledge, such as the properties of K considered as an
extension of a field k, and in particular a study of the relative discriminant.
Sometimes the application of the reciprocity laws in the fields proves useful.
The proof of theorems often depends upon intricate numerical calculations.
We recall the results when K is a quadratic field, say K = Q(d/d), where d
is a square free integer.
Two cases arise:
I. d = 2 , 3 (mod 4). Then the integers are given by x = x 1 x2d/d,where+
x l , x2 are rational integers. Also x is a unit if x: - dxi = f 1, and this is the
Pel1 equation.
11. d = 1 (mod4). The integers are given by x = x 1 + x2(1 +
dd)/2, and
x is a unit if
X : + ~ 1 x 2+ a(l - d)x: = f 1.
Then all units 7 are given by 7 = k 5" where 5 is a fundamental unit and n
takes all integer values.
If d < 0, the only units are f 1, except that when d = - 1, there are four
units given by i n ,n = 0, 1, 2, 3, i.e. f 1, f i, and that when d = - 3, there are
six units f pn, n = 0, 1, 2, where p = (- 1 + d 2 ) / 2 ,i.e.
f l , f(-1 + d 3 ) / 2 ,f ( - 1 - dr3)/2.
2. Many important Diophantine equations are particular cases of the equa-
tion
X Y = CZ',
where X , Y, Z are integers in K and c is a given integer in K. The solution of
this depends upon ideal theory. Denote by h the number of classes of ideals.
We commence with the case h = 1, and then and only then does unique
factorization exist for the integers in K . We suppose that the common divisors
of X , Y belong to a finite set, and then there is no loss of generality in
supposing that ( X , Y ) = 1. We have now
X = 5,c1P', Y = 52~2Q', Z = t3c3PQ,
where 5152 = 5;, C l C 2 = cc$.
Here C1, t2,5, are units, cl, c2, c3 are algebraic integers, each trio being taken
from a finite set, and P, Q are arbitrary algebraic integers with (P, Q ) = 1.
We consider some applications in the simpler Euclidean fields, namely
Q(0, Q ( C 2 ) , Q(d3).
Thefield Q(i)
The equation
x2 + y2 = ZI, I > 1,
where x , y are rational integers, ( x , y ) = 1 and z > 0.
15. APPLICATIONS OF ALGEBRAIC NUMBER THEORY 123
Here (x + iy)(x - iy) = zI.
+
Write d = (x iy, x - iy). Then d I (2x, 2y) and so d I 2, and d = 1, 1 + i, 2.
If d = 2, x = y = 0 (mod 2). If d = 1 + i,
x
-=
+ iy x + y + i(-x + y )
2
9
I + i
is an integer and so x = y (mod 2 ) ; but then ZI = 2 (mod 4), and this is
impossible. Hence d = 1 and so
x + iy = 'i (a + ib)I, x - iy = ( -ijr(u - ib)I, r = 0, 1, 2, 3,
z = 2 + b2,
where a, b are rational integers.
In particular, for the equation
1 + y2 = 21,
a and b must satisfy the equation
& 2 = ir(a + ib)' + (- i)'(a - ib)l.
We prove that when I = 3, the only solution is y = 0.
Since i = ( -i)3, the i' can be absorbed in (a ib)3, hence +
+1
- = a3 - 3ab2 = a(a2 - 3b2),
andsou= ? I , b = O , z = I,y=O.
We prove that when I = 4, the only solution is y = 0.
Here 1 + iy = &(a + ib)4 or ? i(a + ib)4,
The first gives
+- 1 = a4 - 626' + b4 = (a2 - b2 + 2ab)(a2 - b2 - 2ab),
and so
a2 - b2 + 2ab = tl, a2 - b2 - 2ab = +1, T l ,
giving a = & 1, b = 0 or a = 0, b = f 1.
The second gives
&1 = 4 2 6 - 4ab3,
and is impossible.
It will be shown in Chapter 30 that, for all I, the only solution is y = 0.
The equation
y2 +4 = 13,
has only the solutions z = 2, 5.

If y is odd, we have
2 + iy = (a + ib)3,
2 = a(ua - 3b2),
givinga = 2,b = 1,z = 5 ; o r a = - 1 , b = l , z = 2,andanevenvaluefory.
If y is even, put y = 2 Y, z = 22, and so
~2 + I = 223.
Then Y is odd and since ( Y + i, Y - i ) = 1 + i,
then 1 + iY = (1 + i)(u + ib)3,
and 1 = u3 - 3a2b - 3ab2 + b3,
= (a + b)(aa - 4ab + b2).
Hence u+b= fl, a 2 - 4 a b + b 2 = 21,

and so a = f l , b = 0 or a = 0, b = +1,
and z=1, z=2.

The result is due to Fermat.
The field Q ( G 2 )
The equation y 2 + 2 = x3 has only the integer solutions x = 3, y = & 2.
Clearly x f 0 (mod 2), and so
y + d x = (a + b d - ) 3 , x = a2 + 2b2,
where a, b are integers. Hence
1 = b (3a2 - 2b2),
and so b = 1, a = f 1, x = 3.
The result is due to Fermat.
Thefield Q( dr3)
The application of this field, say Q(p), where p2 p + +
1 = 0, is of great
importance in the discussion of many cubic equations, and in particular of
equations of the form
+
x3 y3 = az3.
This equation has attracted the attention of many mathematicians over a
period of years as can be seen from the second volume of Dicksons History
of the Theory of Numbers.
We give a brief rCsumC, sufficient for our applications, of the arithmetical
properties of the field, and for which one may consult Bachmann and
Marshall Hall, Jr.2
The integers in the field are given by a -t bp where a and b are rational
integers and the units by k p", n = 0, I , 2 . The primes are:
I. the rational primes q = - I (mod 3),
11. the factors x = a + bp, x1 = a + bp2 of p = xnl = a2 - ab + b2,
wherep is a prime = 1 (mod 3).
111. the prime h = 1 - p where h2 = -3p.
+
We define a primary integer x py as one with x = 1, y = 0 (mod 3), and
+
then each integer a bp has a unique primary associate x y p given by +
+
a bp = k p"(x + yp), n = 0, 1, 2. The primary integers form a subgroup.
We require some results on cubic residues. There are N(n) = p residues
mod x , and so these can be taken as 0, I , . . ., p - 1. If a is any integer
$ 0 (mod x ) in Q(p), then
aN(n)-l = 1 (mod x),
and so a(P-1)13 - pt
= (mod x), t = 0, I , 2.
Then the cubic character of c( (mod n) is defined by
The cubic law of reciprocity states that if x , x' are two primary primes, then
This also holds when n' = -4. There is the supplementary law,
(;)3= pb'3.
The cubic character has the usual properties, i.e.
We give criteria for the cubic characters of 2 and 3. When n' = - 2 ,
(52)3 = (3(3, =
Now (5) = 1 if and only if b = 0 (mod 2). For

+y +
3
(x ~ =)a ~ bp (mod 2),
gives x2y - xy2 = b (mod 2), and so b = 0 (mod 2), and then we can take
.Y
b
= a, y = 0 (mod 2 ) . Also
= 661, and
b 3 0 (mod 3), and so
(3 = 1 if and only if
p = a2 - 6abl + 366: = +
(a - 3 b 1 ) ~ 276:.
52
If x3 = 2 (mod T), then, since we can take x to be a rational integer, x3 =

2 (modp) is solvable if and only if
p = X2 + 3 Y2, with Y = 0 (mod 3).
Further 3 is a cubic residue of p , i.e. x3 = 3 (mod p ) is solvable, if and only if
4p = X 2 + 3 Y 2 , with Y = 0 (mod 9).
The condition for this is b = 0 (mod 9). For
4p = (2a - b)2 + 3b2.
For immediate application, we require the
Lemma
vq is a prime = 2, 5 (mod 9), then p is not a cubic residue of q, i.e. x3 =
p (mod q) is not solvable.
Write q = 3r + 2, and suppose x3 = p (mod q ) is solvable. Then
- I
XN(q)-l = = X(3rt1)(3rt3) = 1(
x42-l
- 1, mod q),
and so p(3r+ 1 X r t 1 ) = 1, pTtl = l(modq), p r t l = 1.
This is impossible since r f - 1 (mod 3).
Theorem 3
Let a = p or p2, where p = 2, 5 (mod 9 ) is a prime, and let E be a unit in
Q(p). Then the equation
x3 + y 3 + caz3 = 0,
has no solutions (x,y , z , c) in thefield Q(p) except
z=o, x = -y, -PY, -p2.YY,
unless a = 2, when there are also the solutions
X3 = y 3 = - &Z3, E = & 1.
We may suppose that x, y, z are integers in Q(p), that xyz # 0, ( x , y ) = 1 ,

( y , z ) = 1 , (2,x ) = 1. Let ( x , y , z , E ) be a solution for which N(xyz) is a
minimum, where N denotes the norm in the field. Write
a =x + y, p = px + p2y, y = p2x + py.
Then 6 = (a, j3,y ) = 1 or 1 - p or 1 - p2.
Also a + +y = 0, -.-.
aB2:
+ Ea (g)3 = 0.
Here .IS, P / S , y/S are relatively prime in pairs since their sum is zero. Hence if
a is a factor, say, of y/S,
where el, e2, c3 are units and xl, y,, z1 are integers in Q(p). Also x l y l z l # 0 as
otherwise z = 0. Hence
.,x: + E2y; + e3az; = 0,
or say, x; 72y; +
73az; = 0, +
where vz, v3 are units. Now y l is prime to a as otherwise x1 = 0 (modp).
Hence T~ is a cubic residue o f p and so v2 = k 1. Then
x: f y; + v3az; = 0,
*
i.e. an equation of the same type as the original one.
Also N ( ~ y z <) ~N ( ~ ~ y , z =
, )N
~ (s3a) = N (f)3*
hence, 1 3 N(Sxy),
and so x, y , S are units and x3 = k 1, y 3 = k 1.
If a > 2, then z = 0 and the only solutions are those with xyz = 0. If a = 2
there is now also a solution x3 = y 3 = --z3, E = f 1. These and x3 + y 3
= 0, z = 0 are the only solutions.
It has also been long known that there are a number of other values of a for
which the equation
x3 y 3 = az3 +
+
has no rational solutions except x y = 0, z = 0. Let p = 5 (mod IS),
q = 11 (mod 18) be primes, and s o p and q are primes i n Q(p). Then there are
no rational solutions except z = 0 when
a = P,2P, 9P, P 2 , 9P2, 4P2,P4, PIP: >
and also when

a = 494% 99, 2q2, q2, 9q2, 91q22, p2q2.
We prove the result when a = 9p. We may suppose x, y , z are rational
integers, z # 0, and (zI has its least value and that (x,y ) = 1. Since
(x + Y>(X + yp>(x + YPZ) = 9 P 3 ,
and (x + y , x + y p ) = 1 or 1 - p, etc.,
we have x + y = 3pw3,
x + y p = p1 - p)(u + x + yp2 = p 2 y 1 - p y u + vp2)3, A = 0, 1, 2,
where ti, u, w # 0 are rational integers and (u, v ) = 1.
Suppose first that h = 0.

The last two equations give
x = u3 + 3u2u - 6 u u 2 t v3, y = -u3 + 6u2u - 3uu2 - LJ,
and so x + y = 9uu(u - v) = 3pw3,
whence w = 0 (mod 3), w = 3w1 # 0. Then
UD(U - c) = 9pw;.
Hence we have five possibilities with (xl, y l ) = 1.

I. u = x;, v = y;, u - u = 9pz:, x l y l z , = w1.
This gives x: - y: = 9pz;, where
IXlYlZlI Iw1l = 311.1 W I Z I ,
since 9plz13 = 3plw13(x2 - x y + y2) 2 3p1wy.

Hence lzll 6 M J ~ 6 ++/31zI, a contradiction.
11. u = x,; v = 9y7, u - u = pz;,
and so x; - 9y; = pz;.
Clearly z1 $ 0 (mod 3), and so z? = k 1 (mod 9), xy = & 5 (mod 9), and
this is impossible.
111. u = x;, u = py:, u - 0 = 9z;.
Here x: - py: = 9z;.
This is the same as 11.
IV. li = x;, u = 9py7, u - v = z;.
Since .Y: - 9pyy = z:, this is really the same as I.

v. 11 = 9x;, 0 = py;, u - 1) = z;.
Here 9x7 - py: = z;.
This is the same as 11.

Suppose next that h = 1, 2. Then by addition
3pw3 + p"+l(l + + p2"2(1 - p"(u +
- p)(u up2)3 = 0,
or (1 - p2)pw3 + p"+'(u + vp)3 - + p2"4(u vp2)3 = 0.
This is impossible since it implies that p is a cubic residue of p .
15. APPLICATIONS OF ALGEBRAIC NIJMBER THEORY 129
Ex-aniple
A solution of
x3 + y 3 = az3
can be deduced from any solution of
AX3 + B Y 3 + C Z 3 = 0, where ABC = a.
For the substitution
x = u3 + 3u2v - 6uu2 + v3, y = -u3 + 6u2u - 3uu2 v3,
gives 9uv(u - v)3(u2 - uo + v2)3 = az3.
This is satisfied if we take now
u = AX3, u = -BY3, u-u = -CZ3.
Then z = 3XYZ(U2 - uu + u.
We now take the case a = 1 and prove the
Theorem 4
+ +
The equation x3 y 3 z 3 = 0 has no solutions in Q(p) except those given
by X ~ =
Z 0.
We may suppose that x, y , z are integers and (x, y ) = 1, etc. Write
A = 1 - p so that A is a prime and
3 = (1 - p)(l - pz) = A2(1 + p) = -A2p2, and A2 = -3p.
We prove first that xyz = 0 (mod A). For suppose that xyz f 0 (mod A).
Put with rational integers a and b,
+ bp = a + b(l - A) = a + b (mod A).
x = a
Since x $ 0 (mod A), a + b E k 1 (mod 3), x3 = f 1 (mod 3A). Then
x3 + y 3 + z 3 = f 1, f 3 (mod 3 4 $ 0 (mod 3A).
Hence we may assume that z = 0 (mod A).

We prove the more general result that the equation
x3 + y3 = ? ~ 3 9 3 ,
where 7 is any unit, n > 0, z # 0, ( z , A) = 1, ( x , y ) = 1 and so (x, A) = 1,

has no solutions.
We show first that if x, yare any integers in Q(p) and (x, y ) = 1 , (x, y , A) = 1
+
and x3 y 3 = 0 (mod A3), then x 3 y 3 = 0 (mod A*). +
For + y)((x + y)z - 3xy) = 0 (mod A3).
(x
Then s + y = 0 (mod A), x = - y + At, say.
Then x3 + y3 = h"3 - 3A2tZy + 3Aty2

= A3t(t2- pzy2)(mod A4)
= 0 (mod A4),
since y 2 = 1, p (mod A) and t = 0, 5 1 (mod A).
The next stage is to show that if the equation is solvable for an exponent n
it is also solvable for n - 1, and so for n = 1. Since z $ 0 (mod A), this
+
contradicts x3 y 3 = 0 (mod A4).
Write (x + y)(x + py)(x + p2y) = 17A3=Z3.
Then one and so all of the three factors must be divisible by A, since x + y =
x + +py hy etc. Also only one of them can be divisible by a power of A
greater than one. On writing y for py or p2y, if need be, we may suppose that
x + y = 71A3n-2z;, x + py = T ~ A Y ; , x + p2y = q3Ax;,
where rll, 7 2 ,173 are units, ( x l y l , 4 = 1, (xl, y , , zl) = 1, x l y l z l = z # 0.
Multiply by p, p2, 1 and add. Then
or x; + 5,y; = 5 h 3 n - 3 z : ,
where 5, t1 are units. Then as n > 1,
x: + C1y: = 0 (mod X3).
Since ( x l y l , A) = 1, x: = & 1 (mod A2), y; F & 1 (mod h2),
and so 5, = 5 1 (mod h2), and t1 = & 1.
Hence we have a similar equation with n - 1 in place of n. This concludes the
proof.
There are applications of the field Q ( d 3to
) some equations of the form
ax3 + by3 + cz3 - dxyz = 0 (x, y, z ) = 1. (1)
Such equations also occur in the discussion of the integer solutions of the
simultaneous equations
X + Y+Z=dW, (2)
XYZ = eW3, (3)
Thus equation (3) leads to
X = ax3, Y = by3, Z = CZ~, W = XYZ,
where abc =e is some factorization of e and so (1) follows from (2).
It is very difficult to prove the non-existence of integer solutions of the
general equation (1). Sometimes progress can be made with the special case
a = 6, and in particular with
x3 + y 3 + C Z ~- dxyz = 0, (x,y , Z ) = 1. (4)

The question now is whether there exist integer solutions other than the
trivial ones given by xyz = 0 which we exclude hereafter. Suppose for
simplicity that c is not divisible by the cube of any prime divisor of d. Then
obviously (x,y , d) = 1. Also (x,z ) = ( y , z ) = 1. Write 2 p = - 1 k d r 3 .
In equation (4), put
X = ppx + pp2y + qz, Y = pp2x + ppy + qz, (5)

z= px + p y + qz.
Then if pd - 39 = 0, (6)
XYZ = pyx3 + y3) + - 3p2qxyz
$23
= p 3 ( - c z 3 + dxyz) + q3z3 - 3p2qxyz
= (43 - cp3)z3.
Suppose first that d = 0 (mod 3). Then we take p = 3, q = d, and
X = +
3 ( p ~ p2y) + dz, Y = 3(p2x + py) + dz,
(7)
Z = + + dz.
3 ( ~ y)
Then X + Y + 2 = 3dz,
XYZ = Az3,
where A = d3 - 2 7 ~ .
Here X , Yare integers in Q ( d r 3 ) , and Z is a rational integer # O from equa-
tion (9). Write 6 = ( X , Y, Z).Then from equation (7), since 6 I p2X p Y Z + +
etc.
S I 9x, 9y, 3dz,
and so we may take 6 = 1, d r 3 , 3 . We cannot have 6 = 4-3.
For then z = 0 (mod d3), z = 0 (mod 3) and 3 I 6.
On replacing X by 3"X etc., z by 3 a ~ cc, = 0, 1, we may suppose that
equations (S), (9) hold and that ( X , Y, 2 ) = 1. This is obvious when cc = 0.
When cc = 1, equation (7) becomes
X = p~ + p2y + dz, Y = p2s + py + dz, Z = x + y + dz.
Now ( X , Y) = 1, for if ( X , Y) # 1, d-3 I ( X , Y) and X + Y + Z = 3dz

shows that (X,Y, Z ) # 1. Similarly (Y, Z ) = ( Z , X ) = 1. Also z f 0 (mod 3 )
since X = Y = Z (mod d-3) and equation (9) would give X = Y = Z =
0 (mod dr3).
The equations (8), (9) are of the forms (2) and ( 3 ) but now X , Yare conju-
gate integers in Q ( d 3 ) . Further progress depends upon the residues
mod 3 of the prime factors of A.
Suppose first that A is divisible only by rational primes T = 2 (mod 3). Then
either X and Y are both divisible by T or neither is. If A is square free, the
first alternative is impossible since it would give z = Z = 0 (mod n) from
equations (8) and (9). Hence with /? = 0, 1, 2,
X = pox:, Y = pyy, Z = AZ~, z = xly,z,,
where x,, y , are conjugate integers in Q ( d T )and
, z , # 0. Then
pBx; + pZByy+ Az: = 3d~,y,z,. (10)
We show now that /? = 0 and so equation (10) takes the form (4). Since
( z , 3 ) = 1, x,, y , , z , are all prime to d-3. Hence
xy = f 1 (mod 3d3), yy = C 1 (mod 3 d / ) , z: = f 1 (mod 9).

Hence since A = & 1 (mod 9),
rtpB * p2B f 1
f 0 (mod d2).
3
This occurs only when /? = 0.
We now apply to equation (10) the transformation used for (4).
We takep = 1 , q = d, and then
Xi = p1-1 + pzyyl + dzl, Y1 = p2.y1 + py1 + dz,, (1 1)
21 = s, + y , + dz,,
and so X,, Y,, 2, are all rational integers. Also
XlY,Z1 = ( d 3 - A)z: = 27~~:, (12)
X, + Y, + Z 1 = 3dz1. (13)
One of X 1 - Y,, Z 1 is = 0 (mod 3 ) and since from equation (1 1) X , = Y, =
Z (mod d-3),X1 = Y , = Z , = 0 (mod 3). Hence we can dispense with the
factors 27 and 3. Also Z , # 0.
Now account must be taken of the factorization of c. If L is a prime, we
deduce the same equation (4). Hence the method of infinite descent applies.
Then there may be either no solution at all, or a finite number from which all
the solutions can be deduced by rational processes.
An illustration is given by Ward's3 result that the equation
x3 + +y3 5z3 - 5xyz = 0 (14)
+
has only the solution x y = 0, z = 0.
Here c = 5, d = 5 , A = - 10 and no other solution arises.
The equation (14) arises from the equation
2x3 + 2x2y + xyz = 0
on writing X = x 2y + +
2z etc., and so the only solutions are x y +
= 0, z = 0, etc.
Suppose next that A is divisible by rational primes r = 1 (mod 3). Now
r = 7r1rzwhere rl,7r2 are conjugate primes in Q(d-3). This implies that
from equations (8) and (9), we may deduce a relation of the form
AX: + By: + CZ: = 3dXiyiz1,
where A , B, C are conjugate. It is problematic if this equation reduces to the
form (4).
Cassels and Sansone4 show that this occurs for the equation
x 3 + y3 + t3 - X ~ =
Z 0,
which has only the solutions x +y = 0, z = 0 etc.
Here XYZ = -26z3, (15)
X + Y+Z=3z.
Clearly we may suppose z $ 0 (mod 3). Hence we have two possibilities.

The first is
X = pyx:, Y = p"~:, Z = -26zf, z = ~ 1 ~ 1 ~ 1 ,
and so p'.Yy + p"yy - 26~: = 3X1ylZl.

Since ,vlylzl f 0 (mod 3), we have as before y = 0 and so
.Y; + J*: - 2 6 ~ := 3.\-,y,zl.
This leads to X , Y1Z, = (3zIy,
X , f Y1 -k 21 = 321,
and as before, XI, Y , , Z , are all divisible by 3. Hence the method of infinite
descent applies and no non-trivial solution emerges.
The second possibility, since 13 = ( 3 p - 1)(3p2 - 1). is given by
x = py3p - I)YY, Y = pZfi(3pZ - l)y?, 2 = -2:;, := .\-1y1=1,
and so pd(3p - I).Y: + p2'(3p2 - I),$ - 2Z: = 3Sly12,. (17)

As before, since x1ylzl f 0 (mod 3), 6 = 0 and so
(3p - + (3p2 - 1))'; - 2.2: = 3.V,ylzl.

Now we apply the substitution
Xl = px1 + pay, + z1, Yl = p2x1 + py1 + z1,

z1 = x1 + y , + z1.
and so 3x1 = p2Xl + pYl + Z 1, 3Yi = P2Xi 4- py1 z 1,
321 = Xi + Y1 + z1.
Then x1y:' + YlZ? + ZlX:' = XlYlZl.
We show that this gives

x1 = Y h y1 =Z h , z1 = XZY,,
and xi + y; + zz" = xzyzz,. (19)

Hence the method of infinite descent applies. For let x, = ( y l , z l ) etc. and
then
x1 = pyzz,, Y l = qzzxz, z1 = rx2y2,
where
( p , q ) = 1, (x2,y2)= 1 etc. and (XZ,P) = ( y 2 , q ) = (z2, r ) = 1.
Then +
pq2x2z,2 qr2y2xz + rp2z2yi = pqrx2y2z2.
On dividing by y,, we see that since (qx2z2,y,) = 1, we can put
p = Pyz, q = Qz,, = Rx,,
and so PQ'Z; + QR'x; + RP2y; = PQRxzyzzz.
Dividing by R shows that R = 1 = P = Q and so we have equation (19).
No solutions emerge from the descent.
3. We can now consider some results when unique factorization is not
assumed to hold in K . We consider the equation
XY = CZ',
where 1 is prime to h, the number of classes of ideals in K . For simplicity, we
suppose that ( X , Y ) = 1 and that all the factors of c in K are principal ideals.
We have now the ideal equations
(X) = (C1)"l, (Y) = (C2)P,
where cl, c2 refer to all possible factorizations.

Now c = c1c2, (el, c2) = 1. Hence a',G' are principal ideals, and since (I, h) =
1 , n , d are principal ideals. Then we have as before,
X = c151P', Y = c2t2Q', 2 = t3PQ,
where cl, c2 are a finite set of integers in K with clcz = c, (cl, c2) = I , and
tl, t2,t3 belong to a finite set of units with t1t2= (k. Also the algebraic
integers P, Q satisfy the condition (clP, c,Q) = 1.
The simplest application is to the equation
y2 = x3 + k.
and full details are given in Chapter 26. A simple case may be given here. We
suppose that h(Q(dt/k))f 0 (mod 3), d = 2, 3 (mod 4) and is square free.
Then s $ 0 (mod 2), and from
(y + .\/k)(y - dk)= 2.
we deduce with rational integers a, b,
y + dk = &(a+ bdk)3,
where E = 1, 7 , 7 - l where 17 is the fundamental unit in the field, i.e.
7 = T + Udk, T2- kU2 = ? 1.
We now have binary cubics in a, b which require further discussion.
The real difficulty with the equation
XY = CF, ( X , Y ) = 1,
arises when h is not prime to 1. Suppose for simplicity that h 1 I. Then
(X) = qcc:, (Y) = r&,
where (c) = ~1CZ, ((1, 4 = 1,
refers to all the ideal factorizations of c. Since ( X ) and a: are principal ideals,
r l , c2 must be principal ideals. Hence solutions can occur only for factoriza-
tions of c not involving ideal factors. It may happen that only (c) = rl need be
considered. This also arises sometimes when ( X , Y ) # 1. We may then de-
L ~ a l is an ideal and if I = h the equation is
duce an equation (A') = L L ~ Cwhere
impossible if dl is not a principal ideal.
An application is given in Chapter 26 to the equation
y 2 = x3 +k
when k = 1 (mod S), k < 0,k # - 7 , h(Q(&)) = 3, e.g., whenk = -31.
We now give an application to a cubic field.
Theorem 5
The equation x3 + dy3 = 3z3, where d is square free, d = L- 2, f4 (mod 9),
and h ( Q ( f i ) ) = 3, has no rational solutions if 3 = c3 and the ideal G is not
principal.
We require some results on the cubic field Q(0) where = d. The integers
have the form a + be + cB2 where a, b, c are rational integers. Also (3) = z3
is given by (3) = (3, Vd 5 1)3 or (3, vz
f 1)3 according as d = f 2 or
5 4 (mod 9).
We may suppose that x, y , z are integers and that (x, y ) = 1 and so
( y , z ) = 1. We prove that (z, 3) = 1 .
I f 3 I z, x3 + dy3 E 0 (mod 9).
Since 3 y , y 3 = k 1 (mod 9), and x3 rt d E 0 (mod 9), which is impossible.
Write (x + ey)(x2 - exy + e2p) = 3z3.
Since x2 - exy + ezyz = (x + ey)2 - 3exy,

3 cannot be a common divisor of the two left-hand factors. Clearly x + Oy =
0 (mod c ) but not (mod c2). Hence
(x + ey) = ca3, (x2 - exy + e y ) = 263.
Since a3 is a principal ideal, so is C, a contradiction.
An illustration is given by d = 22. This is of special interest since the con-
+
gruence x3 22y3 - 3z3 = 0 (mod M ) is solvable for all M .
4. When no restrictions are placed on the class number h, we have the
fundamental
Theorem 6
The solution of X Y = cZ1,where ( X , Y ) I 8 where 6 is a given ideal, has the
form
X = AClP', Y = pCZQ', Z =V ~ ~ P Q ,
where P , Q are arbitrary integers in K, and tl, la, t3 are units, A, p, v are
numbers in K not necessarily integers, all six being taken from a finite set and
such that
XPi-152 = 6V'G.
For now we have the ideal equation

(X) = 61c1a1, (Y)= S2C2C',
where S1, 6, are elements of a finite set of ideals all of whose prime ideal
divisors divide 6 and where (c) = r 1 r 2 is an ideal factorization of c.
There are only a finite number of classes of ideals, say those typified by
al, a2,. . ., a,,.
Let the inverse class of a be typified by a-'.Then aa-'= (P)
15. APPLICATIONSOF ALGEBRAIC NUMBER THEORY 137
where P is an integer. Hence a - ' ( X ) = 6,c,(P)' and so S 1 c , / a - I = ( A ) is a
principal ideal. Hence
( X ) = (A)(P)', x = AtP',
where 5 is a unit. Since 5 = 5,5: where 5, takes only a finite number of
values, we have the theorem. Also A is one of a finite set of numbers in K .
+
An application to the equation y 2 = x 3 k will be given in Chapter 26.
There are obvious extensions to the solution of the more general equation
x,x,.. . x, = CZ',
where we suppose now that
(X,, X,) I 67,s ( r , s = 1, 2 , . . ., n, r f 8.)
where SreSis an ideal of a finite set.

The most important application is to Fermat's last theorem, namely, that if
1 > 2, the only solutions in rational integers of the equation
XI + J+ = Z f
are given by X ~ = Z 0. This will be investigated in Chapter 28.
For some equations reducing to the form X Y = c Z 1 ,it may prove con-
venient to consider solutions in a field K' containing K as a subfield, and so
the relative units must be studied. Often considerable numerical detail may
be involved.
REFERENCES
1 . P. Bachmann. "Die Lehre von der Kreistheilung". B. G. Teubner, Leipzig
(1872), pp. 185-199 and 220-224.
2. Marshall Hall, Jr. Some equations y 2 = x 3 - k without integer solutions.
J. Lond. Math. Soc., 38 (1953), 381.
3. M. Ward. The vanishing of the homogeneous product sum of the roots of a
cubic. Duke Math. J., 26 (1952), 553-562.
4. G . Sansone and J. W. S. Cassels. Sur le probleme de M. Werner Mnich. Acta
Arith., 7 (1962), 187-190.
CHAPTER 16
Finite Basis Theorem for the Rational Points on a

Cubic Curvef(x,y,z) = 0 of Genus One
1. Theorem 1
AN the rational points of a cubic curve f ( x , y , z ) = 0 of genus one, can be
found from a finite number by the chord and tangent process.
A sketch is given of the original proof by Mordelll. It may be supposed that
there are an infinity of solutions, as otherwise there is nothing to prove. We
show first that the problem is equivalent to that of finding the integer solu-
tions in x, y , z of an equation with integer coefficients.
ax4 + bx3y + cx2y2 + dxy3 + ey4 = z2. (1)
Let (x, y, z) be a rational solution of f ( x , y , z ) = 0; a solution with z = 0
may be considered trivial, and so we may suppose z # 0. Put
x = XZ + X, y = yZ + Y, z = ZZ.
Then S1Z2 + s,z + s, = 0, (2)
where S1,S,, S3 are homogeneous functions in X , Y of degrees one, two,
three. Since the curve is of genus one, S, # 0. Hence
2s1z = -S2 (q - 4s1s3)12.
Here S, - 4S1S3 is a homogeneous binary quartic in X , Y which becomes a
perfect square for values of X, Y given by S, = 0. This square cannot be zero
for then S, = 0, S, = 0, and so a linear factor would divide out in equation
(2). Hence by a linear transformation on X, Y, we are led to an equation (1)
in which a is a perfect square # O . By a further transformation, we have now
the problem of finding the solution in integers x, y , z where (x, y ) = 1 of an
equation, say,
x4 + bx3y + cx2y2 + dxy3 + ey4 = f z 2 , (3)
where b, c, d, e, f are integers.
Let the algebraic integer 0 be defined by
O4 + be3 + cd2 + dd + e = 0. (4)
Write equation (3) as
( x - ey)(x3 + ( e + b)xay + . . .) = fz2.
16. RATIONAL POINTS ON A CUBIC CURVE.~(SJ,:) = 0 OF GENUS ONE 139
The common ideal factors of the two factors on the left-hand side are finite in
number. Hencc from the general result of Theorem 4, Chapter 15,
s - fly = pT2/m, (5)
where p is one of a finite set of integers in Q(8), and m is one of a finite set of
rational integers, and
T =p + qe + + s~3,
where p , q, r, s are rational integers. Since there are an infinity of solutions

for x, y , the same p, in must also occur for a particular solution, say ( x ~y).
Then mz(x - 8y)(x - By) = (P + Q8 + R P + SB3)2y (6)
where P, Q, R,S are rational integers.
If the quartic for 0 is irreducible, equation (6) will also hold for all the
conjugates of 8. This will still be true if the quartic is reducible. On making a
linear transformation on x, y , we may suppose that x = 1, y = 0. On multi-
plying out equation (6) and replacing 06,05, O4 in terms of d3, B2, 0 from (4),
and equating to zero the coefficients of 03,02, we have two equations in
P, Q, R,S. On solving for two of the unknowns in terms of the others, we
find that another solution (xl, yl) of equation (3) can be expressed linearly
in terms of P, Q, R,S. Hence from equation (6),
max(Ix1l, IYll) 6 kmax(Ixl, lYo12,

where k is a constant. But from a sequence s, > 0, where s , + ~ < ks;,, s, is
bounded, and if any of the s have s > k, then after a certain stage we arrive
at an n such that s, 6 k. This proves that the infinity of solutions can be
expressed in terms of a finite number by rational processes.
The geometrical meaning of the process is that if a rational point P is known
on the quartic (3), another rational point is found from the intersection with
+ +
the quartic of a parabola z 2 = gx2 hx j having three point contact
with the quartic at P. The connection with the chord tangent process follows
from the 1-1 correspondence between the points of the quartic
y 2 = x4 + 6cx2 + 4dx + e,
and the cubic
t2 = 4s3 - g,s - g,
given by
~X(S + C ) = t - d, y = 2s - x2 - C,
where g,, g, are the invariants of the quartic.

2. We now prove by Weils, method the finite basis theorem.
Theorem 2
There exists a jinite set S of rational points P1,Pa, . . . , P, on the cubic curue
y2 = 4x3 - g,x - g3, (7)
such that all the rationalpoints on the curve can be derivedfrom the P by rational
processes.
This is given by taking the point where the tangent at P1 meets the
curve again, and the point P1,2where the line joining P1,P2 meets the curve
again, adding these points to the set S, and then proceeding similarly with all
the points of the enlarged set.
On replacing x, y by x/4, y/4, the equation (7) takes the form
y2 = x3 - h2x - h3. (8)
We may suppose that h2 and h3 are integers, for if h2 = H 2 / H , h3 = H 3 / H ,
it suffices to replace x, y by x / H 2 , y / H 3 .
Since the curve is of genus one, the equation
x3 - h2x - h3 = (x - el)(x - e2)(x - e3) = 0
does not have equal roots. The equation can be considered as defining cubic
fields K = Q(e), (e = el, e,, e3), reducible or irreducible.
We require the coordinates of the points P1,2,Pl,l. Let e denote any of
el, e2, e3. We prove that PlS2is given by
The equation of the line joining P1 and P2 is
+- yx2 -
2
.
Hence x3 - h,x - h3
(
= y1
x1
(x - X I ) )
Two roots are xl, x2 and the third is x ~ , ~ .

+ + +
Put x = e X,x1 = e Xl, x2 = e A,. Then, say,
X3 + H1X2 + H2X = ( y , + Y2
x2
- y1
- Xl
(X - XI,).. (1 1)
The product of the three values of X gives
Another form for Xla follows from equation (10). This gives
16. RATIONAL POINTS ON A C U B I C C U R V E ~ ( X , Y , Z =
) 0 OF GENUS O N E 141
This can be written as
on putting y: = xy - h2xl - hB,etc.

We now deal with PlPl.The equation (1 1) takes the form
x3 + H1X2 + H2X = (y1 + 3x2Yl- ( X - X1)),
on replacing ( y 2 - y1)/(x2- x l ) by dyl/d.xl.Now the product of the roots gives
+ ex, + e2 - h,) - 3xf + h,

xll - e = (2(x2
2Yl )
= (x + h2 - 2exl -
2Yl 2e2)2*
We write equation (7) as

yz = (x - el)(. - e,)(x - e3) = N ( s - e).
We use the term norm for N ( x - e) even when the cubic is reducible. Here e
stands for each of the numbers e l , e,, e3. Our problem is to find in K , = Q(e)
those numbers of the form x - e, where .Y is a rational number, whose norm
is the square of a rational number, and this is to hold for e = el, e2, e3.
These numbers are included i n the set S(e) of numbers xo x l e x2e2, + +
so, x1, x2, rational, whose norm is a perfect square. Clearly S(e) includes
the squares of the numbers of K . We divide S ( e ) into classes putting into a
class C, say, all those numbers of S(e) which differ from an element of C by
a squared factor of a number of K. There are infinitely many classes.
Two classes C, and C , can be multiplied together since any element of C ,
multiplied by any element of C, belongs to a uniquely defined class, say
C1C2.The principal class is formed by those numbers of K which are perfect
squares and plays the role of a unit class E in the multiplication.
If C is any class, clearly C2 = E. The classes form an infinite Abelian group
in which every element is of order 2.
We prove now the fundamental.
Theorem 3
The nuinbers of the jomr .\- - ci ( e = el,e2, e3), iikm norins are ralional
squares, belong to only ajinite number of classes.
In y2 = N ( x - e), put x = U / Wl where U and Wl are rational integers

and (U, W,) = 1. Then
( y W f ) 2= W I N ( U - e W J ,
and since (Wl,
U ) = 1, Wlis prime to N ( U - eWl), and so Wl = W 2 .
Then y = V /W 3 and ( V , W ) = 1. Hence
V 2 = N ( U - eW2)
= ( U - elW2)(U- e2W2)(U- e,W2).
In the field Q(el), U - el W 2and ( U - e2W2)(U- e3W 2 ) ,which is also a
number in Q(el), have only a finite number of common ideal factors since +
if U - el W 2 = 0 (mod+), then (el - e2)(e1- e,) = 0 (modh). Hence by
the general theory
U - el W 2 = pP2,
where p is one of a finite set of numbers in Q(el) and /3 is an integer in Q(el).
Hence on dividing by W 2 ,we have, say,
x - el = pla:,
where p1 is one of a finite set and a1 is a number in Q(el). A similar result
holds if el is replaced by e2 or e3, This proves the theorem.
The numbers x - e do not form a multiplication group since the product of
two such numbers need not be of the same form.
An addition process can be defined for the rational points giving them a
group structure.
Thus we define P1 + P,, the sum of two rational points, as the image of P12
in the x axis and write P3 = P1 P,. +
We can say that the rational point P(x, y ) is associated with the number
x - e and that P belongs to the class C if x - e does.
Theorem 4
U P , , P2 belong to the classes C,, C,, then P1 + P2 belongs to the class C,C,.
This is obvious from equation (9).
Theorem 5
The point P2 = 2P1 belongs to the principal class, i.e. x2 - e = v2, where v is
in Q(e).
This is obvious from equation (14).
The converse of this theorem is given by
Theorem 6
I f P , belongs to the principal class, there exists a rational point P, such that
Pz = 2P,.
Let x2 - e = ( A + Be + Ce)), (15)
16. RATIONAL POINTS ON A CUBIC CURVE f(.u,y,z) = 0 OF GENUS ONE 143
where A , B, C are rational numbers. This holds for e = el, e,, e3. On multi-
plying out, replacing e3 by h2e + h3, and e4 by h2e2 h3e, and equating +
coefficients of I , e, c2 on both sides,
+A' 2BCh3 = x2,
2AB + 2BCh2 + C2h3 = - 1, (16)
B 2 + 2AC + C2h2 = 0.
On eliminating A from the second and third equations,
B 3 - h,BC2 - h3C3 = C,
(:I3
we have
or as C # 0,
1 = - h, (g) - h3
Write 1jC = y,, BjC = .xl. Then xl, y , is a rational solution of
y21 -- X; - h2x - h3. (17)
The third equation in (16) gives 2Ay1 + x: 3- h2 = 0. Then substituting
for A , B, C ,
A+Be+Ce -
,
- -xf - h,
+ -x1
e + - - ,e2
2Yl Y1 Y1
and so from equations (15), (14), x2 = x l , and so P2 = 2P,.
Theorem 7
UP,,P2 belong to the same class, then
Pi + P2 = 2P3,
where P3 is also a rational point.
This is obvious from Theorems 4 and 6, since Pl + P, belongs to the
principal class.
Theorem 8
All the rationalpoints on the curre y 2 = s3 - h,x - h, can be obtained from
a finite number by the addition process.
We distribute all the rational points into classes putting in the same class all
those points for which x - e belongs to a given class as previously defined.
From Theorem 3, the rational points then fall into a finite number of classes,
and these may be represented by the points
Al(x1, Y AAdx2, ~ 2 1 ,. . .? An(xn7 Yn).
We now apply the following reduction process to a rational point P(x,y).

We take the point A = A j , say, belonging to the same class as P. Then from
Theorem 7,
+
P A , = 2P1, (18)
with a rational point P,. Similarly

Pi + Aj.1 = 2P2,
where A,,, is one of the As and P , is rational.
. . . . ....
P, + A,,, = 2 p s + , ,
where the Ps are all rational points. Hence P is expressed linearly in terms of
the A and P, Thus eliminating P I , Pz,. . . , P,, we have
P + A , + 2Aj.I + . . * + 2Aj,, = 2P,+,. (19)
We prove now that the process leads after a finite number of steps to a
point P s + l also belonging to a finite set.
We change our notation slightly, i.e. write ( x / z 2 ,y / z 3 )for ( x , y ) and now
( x , z ) = ( y , z ) = 1, and equation (7) in the form
YZ = x3 - h2xz4 - h3z6,
so that the homogeneous coordinates of the point P are now (xz, y , z3). We
suppose that x , y, z are integers and that (x, z ) = 1.
We turn now to equation (18) and consider the relation
P + A = 2PI
connecting any three rational points P, A , P,. Suppose that the coordinates
of P, A , P,,2P1 are ( x , y , z), (a, b, c), (x,, y,, z l ) , and ( X , Y, Z ) respectively.
Write p = max( 1x1, z 2 ) , cc = max( la/, c2), pl = max( Ixll, z:), M =
max( 1x1,Z 2 ) .
Clearly IyI = O(p3/,)where the constant in 0 (and hereafter) depends upon
h,, h3. Here it can be taken as (1 + lh,l +
lh31)12.
We require the abscissa of P + A . From equation (13), on using the homo-
geneous coordinates, we have
_
X - (ax - h2c2z2)(c2x+ az2) - 2h3c4z2- 2bcyz
2 2 - (c2x - .z2)2
Since ( X , Z ) = 1, 1x1,IZI are respectively less than the moduli of the
numerator and denominator of the right-hand side. Hence
M = O(JJ,~)>.
We now prove that p1 = O(M1l4).The formula (14) gives
16. RATIONAL POINTS ON A CUBIC CURVE,f(X,J?,Z) = 0 OF GENUS ONE 145
say. Since ( X - eZ2)'12 is an integer in K , so is p qe re'. Hence if A is+ +
the discriminant of the e equation, Ap, Aq, Ar are rational integers. So also
are the numbers
Since (sl,
zl) = 1, h Z / y l z , is also an integer. Hence
xf I h(2p + hr), 2: I Ar.
But A(2p + h,r) and Ar can be expressed linearly in terms of ( X - eZ2)ll2
with e = el, e,, e3, and so they are O ( d M ) . Hence xf = O(da),z: =
O ( d M ) ,and so p, = O ( M " 4 ) ,and then p, = O(p112).
Applying the same process to the equations similar to (18), and denoting
,
the corresponding p by pl,p,, . . ., we have pr+ = O(p:12). Such a sequence
is bounded and so we come to a stage P , + , , a point with bounded coordinates
and so only with a finite number of possibilities for P,,,. The theorem now
follows from equation (19).
There is a simple analytical interpretation of these arithmetical results. The
coordinates (.Y, y ) of any point P on the curve
y2 = .v3 - h2.y - h3
can be expressed in terms of an elliptic parameter u by means of

.v = p(u), 24' = p'(u),
where p(u) is the Weierstrass elliptic function with invariants 4h2, 4h3 and
periods w , , u p ,say. Then there is a 1-1 correspondence between the elliptic
parameter u mod(w,, w 2 j and the points P = P(u), say. Clearly the point
P( - u) is the image of P(u) in the .Y axis.
The addition theorem for the elliptic function p(u) shows that the three
points with parameters u,, u,, u3 are collinear if
u1 + u1 + u3 = 0 (mod wl, w,).
+
We now define the sum P, P p of two points P,(ul) and P,(u,j as the point
with parameter u, +u, = -u3. Hence the sum is the image in the x axis of
Pl.,,the third point of intersection of the line P,,Pz with the cubic.
The finite basis theorem can now be stated in the form:
Theorem 9
AN the rational points on the cubic (7) are given parametrically by
Zl = nllUl + .. . + 177,3/,,
where u,, . . ., 11, are afixedjnite set and the m take all integer iwlires.
An infinity of rational points can be derived from a given one u unless
there exists an integer n such that nu equals a period of the elliptic function
and then u is of finite order and is called an exceptional point. The parameters
u of the exceptional points form a finite Abelian group of the cubic.
The minimum number g of generators of infinite order of the basis is called
the rank of the curve.
We have Nagell's
Theorem 10
I f ( x l ,y l ) is an exceptional point not at infinity of the curve
y 2 = x3 - A X - B, (21)
where A , B are integers, then x l , y 1 are integers and either y , = 0 or y : is a
divisor of thediscriminant D = 4A3 - 21B2.
It is an open question whether curves exist with an assigned number N of
exceptional points. There exist a number of nonequivalent cubics with
N = I , 2, 3 , 4 (cyclic or non-cyclic group), 5, 6,7, 8 (cyclic or not), 9 (cyclic)
10, 12, 16 (non-cyclic), but none with N divisible by 1 1 , 14, 15, 16 (cyclic) 20,
24, 32. For proofs and references, see Linda and Nagelllo.
Estimates have been given by Billing4 for g , the number of generators of
infinite order for the rational points on a cubic curve in his comprehensive and
valuable memoir. He has also found all the generators for 0 < IAI, IBI < 3.
Birch and S~innerton-Dyer~ have given g for most of the sets 0 < I A1 < 20,
0 < IBI < 30, IAl < 200, B = 0.
3. Two particular cases of the equation (21) are of special interest and have
been discussed by many writers, namely, the equation y 2 = x3 k which +
will be discussed in Chapter 26, and the equation
y 2 = x3 - A X (22)
which we deal with here. Many results have been found by Billing4, Birch and
Swinnerton-Dyer 5 , and Lind 6 . The equation (22) is associated with the
equation
Y 2 = X 3 + 4AX (23)
by the relation
and so a solution of equation (23) can be derived from one of (22). Conversely
on applying the transformation (24) to (23), a solution of (22) can be deduced
from one of (23).
Further equation (22) is birationally related to the equation
X4 + 4A = 2' (25)
by the relation
2x = x 2 - z, y = xx.
16. RATIONAL POINTS ON A CUBIC CURVE,f(.Y,),:) = 0 OF GENUS ONE 147
Billing4 gives the basis points for equation (22) for J A l 5 50 with, however,
at1 error for A = 33. The range has been extended by Birch and Swinnerton-
Dyer5.
Lind6 has dealt exhaustively with the question of finding values of A for
which equations (22) and (23) have only exceptional solutions. His results
depend upon the study of the equation (25) already considered in Chapter 4.
He gives, inter aha, the instances
A =p, p = f 3 , -5 (mod 16),
A = -p, p = -5, 7(mod 16),
p = 1 (mod 8), (5) 4

= -1,
the symbol being a biquadratic character,

A = -2p, p = t 3 (mod 8).
He gives also results for A = +pq, 2pq, e.g. A = pq, p = 3 (mod 16),
q = - 5 (mod 16). Here p and q are either primes or cubes of primes.
4. Recently Birch and Swinnerton-Dyer have found interesting results and
conjectures for the number of generators of infinite order. These depend on the
association of a curve with a congruence zeta-function of a complex variable s.
This takes a comparatively simple form for the cubic curve.
y2 = x3 - A X - B.
Let p be a prime and denote by N p - 1 the number of solutions of the
congruence
y 2 = x3 - A x - B (mod p).
Then
IN,, - p - 11 < 2dP,
and excluding a finite number of exceptional values of p , the zeta function
takes the form
f(s) = n
P
(1 + (N, - p - 1 ) p + p1-2s),
where the exceptional primes are excluded from the product. The product
converges absolutely for R(s) > 3/2, and it has been conjectured by Hasse
thatf(s) can be continued meromorphically over the whole plane. Then there
is the
Conjecture
The function f ( s ) has at s = 1 a pole whose order is the number of generators
of the rational points of infinite order.
They have given a great deal of numerical evidence in support of this

remarkable conjecture for special cubics, for example, those with AB = 0.
Confirmation* is also available for the equation
x3 + y? = uz3.
Now N,, where p is a prime, is the number of solutions of the congruence
.x3 + y 3 = uz3 (modp).
The zeta function associated with the curve turns out to be
S(s)i(s - l)/La(s),
where {(s) is the Riemann zeta-function, and La(s)is a Hecke L series.
As above, the behaviour of L,(s) at s = 1 leads to the conjectures for the
number g of generators of the rational solutions, i.e. Lc,(s)has a zero at s = 1
of order precisely g. This was verified for special values of cube-free a =
2'3", r, s = 0, 1, when the product of the prime divisors #2, 3 of M is
< 100. There were 262 instances with g = 0, 297 with g = 1, 56 with g = 2,
and 2 with g = 3.
It may be noted that Wiman11*12.'3ha s constructed an infinity of non-
equivalent cubics with g = 4,5, and a number with g = 6.
As can be seen from Cassels'9 exhaustive survey article, the question of the
rational points on a cubic curve has of late engaged the attention of many
mathematicians. No criterion, however, has actually been proved for deter-
mining whether such points exist. The recent work of Birch and Swinnerton-
Dyer leads to very useful empirical methods which are frequently successful
in finding solutions.
The two last papers contain many references.
REFERENCES
1. L. J. Mordell. On the rational solutions of the indeterminate equations of the
3rd and 4th degrees. Proc. Camb. Phil. SOC.,21 (1922), 179-192.
2. A. Weil. Sur un thCorkme de Mordell. Bull. Sci. Math., (2) 54 (1930), 182-191.
3. T. Nagell. Solution de quelques probltmes dans la thCorie arithmktique des
cubiques planes du premier genre. Vid. Akad. Skrifer Oslo, I (1935), Nr 1.
4. G . Billing. Beitrage zur arithmetischen Theorie der ebenen kubischen Kurven
vom Geschlecht Eins. Nova Acta. Reg. SOC.Scient. Upsaliensis., ser IV. 11
(1938), Chapter IV.
5. B. Birch and H. P. F. Swinnerton-Dyer. Notes on elliptic curves I. J . reine
angew Math., 212 (1963), 7-25.
6. C. E. Lind. Untersuchungen uber die rationalen Punkte der ebenen kubischen
Kurven vom Geschlecht Eins. Dissertation, Uppsala (1940).
7 . B. Birch and H. P. F. Swinnerton-Dyer. Notes on elliptic curves 11. J. reine
angew Math., 218 (1965), 79-108.
8. N. M. Stephens. Conjectures concerning elliptic curves. Bull. Am. Math. SOL-.,
73 (1967), 160-163.
16. RATIONAL POINTS ON A CUBIC CURVE,f(S,),Z) = 0 OF GENUS ONE 149
9. J . W. S. Cassels. Diophantine equations with special reference t o elliptic
curves. J . Land. Math. Sac., 41 ( 1966), 193-291.
10. T. Nagell. Recherches sur Iarithmetique des cubiques planes d u premier
genre dans un doinaine d e rationalite quelconque. Nova Acto Reg. Sac. Scient.
Upsaliensis, Ser. IV 15, N r 6, 1952.
1 I . A . Wiman. Uber den Rang von den Kurven y 2 = x(x + a ) (x + h). Acta
Math., 76 ( 1944), 225-25 1.
12. A. Wiman. Uber rationale Punkte auf den Kurven y 2 = x(xz - 2).Acta
Math., 77 (1945), 281-320.
13. A. Wiman. Uber rationale Punkte auf Kurven dritter Ordnung vom
Geschlechte Eins. Actri Math., 80 (l948), 223-257.
6-t
CHAPTER 17
Rational Points on Curves of Genus g = 0 or I and g >I

1. Let f(x, y, z) = 0 be the equation in homogeneous coordinates x, y, z of
an irreducible curve of the nth degree where f is a homogeneous polynomial
of degree n with rational coefficients. Such polynomials or curves will be
called rational. The general polynomial of degree n contains t ( n + l)(n + 2)
terms.
A curve of degree n is in general uniquely defined by +n(n + 3) linearly-
independent linear conditions on its coefficients, for example, if it is to pass
through +n(n + 3) points. This, however, is not always so. Thus a cubic curve
is in general determined by nine points, but an infinity of cubics S, + kS2 = 0,
where k is an arbitrary parameter, pass through the nine points of intersection
of the cubics S1= 0, S2 = 0.
Definition
A point on the curve f = f(x, y, z ) = 0 which also satisfies the equations
is called a singular point.

Such a point is called a double point if it does not also satisfy all of the six
equations typified by a2f/ax2 = 0, a2f/axay = 0. We shall consider only
curves all of whose singular points, say N in number, are double points. Then
each such point is counted at least twice in the number of intersections of
f = 0 with another curve. The singular points of a rational curve are deter-
mined by rational equations, and so all the symmetric functions of the co-
ordinates of these N points are rational numbers. A set of n points with such a
property will be called a rational n set.
Theorem 1
N < +(n - l)(n - 2). (2)
For if N > +(n - l)(n - 2), a curve of degree n - 2 can be drawn through
+
+(n - l)(n - 2 ) 1 of the double points, and any other n - 3 points of the
curvef = 0 of degree n, since
+(n - 2)(n + I) = +(n - I)(n - 2) + 1 + n - 3.
17. RATIONAL POINTS ON CURVES OF GENUS g = 0 OR 1 AND g > 1 151
These two curves would intersect in at least
n-3+(n- I)(n-2)+2=n2-2n+ 1 >n2-2n
points. This is impossible by Bezouts theorem.
Hence an integer p > 0, called the genus of the curve, is defined by
N = +(n - I)(n - 2) - p . (3)
Definition
A curw passing through the N double points of the curiv f = 0 is called an
adjoint curve o f J
When the double points form a rational set, the conditions on the adjoint
curve can be expressed linearly with rational coefficients in terms of the
coefficients of the adjoint curve, which are supposed hereafter to be rational
numbers.
We recall that two curvesf(s, y , z ) = 0, F ( X , Y, 2 ) = 0 can be birationally
transformed into each other if their points, except for a finite number, are
connected by a relation of the form,
.V = A(X, Y , Z ) , y = B(X, Y,Z), z = C(X, Y,Z),
X = a ( s ,y , z ) , y = b(x, Y , z), z= 4 x 9 Y , z),
where As etc. are polynomials in the arguments A, Y, 2,etc. This is a 1-1
correspondence, except for the singular points. Such curves have the same
genus. When rational, they are called equivalent if the polynomials A , a, etc.
have rational coefficients.
It is well known that a curve with singularities of order greater than two can
be transformed by a birational transformation into a curve with only double
points.
2. A fundamental result due to Hilbert and Hurwitzl is given by
Theorem 2
E w y rational curw C of degree n and genus zero is equiilalent to a rational
curre of degree n - 2 and genus zero.
An adjoint curve of degree n - 2 can be made to pass through n - 2
further points, for example, arbitrary rational points, or a rational n - 2
point set, since
f(n - 2)(n + 1) = +(n - l)(n - 2) + n - 2.
Hence a family of adjoint curves of degree n - 2 is given by
441 + A 2 4 2 + . * + An-l+n-l = 0,
where the X are arbitrary constants, and the 4 are rational polynomials in
s,y, z of degree n - 2. By the choice of the n - 2 points defining the 4, we
may suppose that the 4 are irreducible and linearly independent.
Define a curve C , by
(1 - r, = . . . = -, 6n-1
;67 - 4 2 Cll-1
where the 5 are homogeneous coordinates in n - 2 dimensional space. Then

C1is of degree n - 2. For it meets an arbitrary hyperplane
A151 + ... + hn-lSn-l = 0
in points determined by the intersection of
A141 + . . . + Xn-,&-l = 0,
a curve of degree n - 2, and f ( x , y , z ) = 0, a curve of degree n. Of these
n(n - 2) intersections, (n - l)(n - 2) are given by the double points leaving
n - 2 variable intersections depending on the X and forming a rational set.
Thus if we take any two of the equations, say
&-&=&
dl - 4 2 $3,
we have, on eliminating x , y , z , a curve of degree n - 2

g(&, (2, 53) = 0.
The transformation is birational in the t and x , y , z .
It follows that a rational curve of odd degree and genus zero has an infinity
of rational points since it is equivalent to a line.
A rational curve C of even degree and genus zero is equivalent to a conic. If
the conic has one rational point, it has an infinity of rational points and so does
C. If the conic has no rational points, the only possible rational points on C
are the singular points, since these have been excluded in the birational trans-
formation.
It may be remarked that the conic, and so also the curve C , have an infinity
of rational 2 sets, i.e. points (xl,yl) and ( x z , y 2 ) for which x1 x,, x l x z , +
y , + y z , y l y , are all rational. This is obvious from the intersection of a conic
f ( x , y ) = 0 and an arbitrary line Ix + my + n = 0. It becomes a difficult
question if the solutions are required to be in a given quadratic field.
A quartic curve of genus zero whose only possible rational points are its
singular points is easily constructed.
Let F, F,, F,, F3 be four homogeneous rational quadratic forms, and L a
linear rational form, in x, y , z . Suppose that F = 0 has no rational points,
that Fl = 0 and Fz = 0 each pass through the intersection of F = 0, L = 0,
that F, F,, F2 are linearly independent, and that F3 = 0 does not pass through
either point of intersection.
The quartic is given by
F ( X , Y, 2 ) = 0, X = F1(s,y , z), Y = FZ(x,y , z), Z = F ~ ( xy, , z).
The curve is of genus zero since X , Y , Z , and so x , y , z , can be expressed in
terms of a variable t but with algebraic coefficients. The two intersections of
F ( s , y , z ) = 0 and L(s, y , z) = 0 both give X = Y = Z = 0 and so this must
be a double point on F(X, Y, 2 ) = 0. Rational points on F ( X , Y, 2 ) = 0
other than the singular points would lead to rational points on F ( x , y , z ) = 0.
3. Results for curves of genus > O have been given by PoincarC2.
Theorem 3
Every rational curve C of genus one and degree n > 3 , which has a rational
point P, is equivalent to a cubic curve.
The curve has fn(n - 3) double points since
I = +(n - I)(n - 2 ) - fn(n - 3).
Draw an adjoint curve of degree n - 2 which meets the curve C in three
points at P. Its equation contains n - 4 arbitrary constants since
f(n - 2)(n + 1) - fn(n - 3 ) - 3 = n - 4.
The adjoint meets C in a rational n - 3 set since
n(n - 2 ) - n(n - 3 ) - 3 = n - 3.
We now draw another adjoint curve of degree n - 2 through the rational
n - 3 set. This family of adjoint curves involves two arbitrary parameters
since
+(n - 2)(n + I ) - fn(n - 3 ) - (n - 3) = 2,
and so takes the form
A141 + A 2 4 2 + A343 = 0,
where the A are rational. These adjoints intersect C in a rational 3 set since
n(n - 2 ) - n(n - 3) - (n - 3 ) = 3.
Then f1:f2:f3 = 41:42:43
is a birational transformation with rational coefficients which changes C into a

cubic curve.
4. Very little is known about the rational points on curves of genus > 1. It
may be that some curves of genus two or three are more amenable to attack,
and so two well known relevant results in algebraic function theory may be
mentioned.
Theorem 4
A rational curce ofgenus two is equivalent to a curve
y2 = aox6 + a1x5 + . . . + a6.
Theorem 5
The quartic curve
ax4 + by4 + cz4 + 2fy2z2 + 2gz2x2 + 2hx2y2 = 0
is in general of genus three.
Mordell put forward in 1922, the
Conjecture
There are only a finite number of rational points on a curije of genus > 1.
Much attention has been given to this in recent years but it is still unproved.
There are not known many non-trivial instances of this conjecture apart from
those given by Fermats last theorem. The known results for the quartic
above are found by replacing z 2 by z and so reducing the problem to that of
rational points on a curve of genus 1.
In Chapter 4 a quartic curve of genus three was given which has no rational
points except trivial ones.
Some curves of genus two which have no rational points have been
found by Mordel13. The equations may be written in the form
z2 = (alx2 + b1y2)(a2x2+ b,y2)(a3x2 + b3y2),
where the a and b are integers and are all positive, except that in one pair, a
and b may have opposite signs.
Theorem 6
The equation (4) has no integer solutions except (0, 0, 0) if all the following
conditions are satisfied:
+ +
(i) a, + a2 + a3 = - 1 (mod 4), b, b2 b3 E 0 (mod 4),
(ii) only one b is even,
+
(iii) none of the sets a,, a2, a3, or a , + bl, a2 b2, a3 + b3, are congruent
mod 4 to 1 , I , 1 , or to a permutation of (0,1 , 2).
(iv) a2b3 - a3b2, a3b1 - a,b3, a,b2 - a2bl have no common divisor, and all
their odd factors are = 1 (mod 4).
The equations so found are of special interest since they may be everywhere
locally solvable.
Suppose that ( x , y ) = 1 . Then we easily deduce that
a1x2+ bly2 = d,d,z:,
a2x2+ b2y2 = d3d,zi,
a3x2 + b3z2 = dld2z&
where d, I a2b3 - a3b2,etc.
Two cases arise according as all or only two of the d a r e odd.
In the first case,
-
z21 = a1x2 + b,y2,
zl +
= a2x2 bzy2, (mod4)
zd = a3x2 + b3y2.
By addition
z: + zi + zi + x2 = 0 (mod 4).
Either z1 = z2 5 z3 = .Y = O(mod 2), and since at least one b is odd,
y = 0 (mod 2) and so (x, y ) # 1 ; or z1 = z2 = x3 = x = 1 (mod 2). Then
according as y is odd or even,
a, + 6, = a2 + b, = a3 + b3 = 1 (mod4),
or
a, = a, = a3 = I (mod 4),
and these have been excluded by (iii).
The case when, say d3,is even can be dealt with similarly. A simple numerical
illustration is given by
(a,, a29 a3) = (4a + 1,3,4b + 3), (bl, bz, b3) = (2, 1, 1).
Then the only condition to be satisfied is that the odd factors of 5 - 4 4
8b - 4a + 5 should be = 1 (mod 4). Some instances are given by (a, b) =
(0, 11, (1, 21, (-2, 2).
These equations are everywhere locally solvable, that is, as a congruence
mod p', p a prime. On giving x or y a fixed value, the curve has genus two, and
it is easily shown by Weil's theorem, that the congruences are solvable for
p > 17. The smaller values o f p must be considered separately.
New r e s ~ l t s * ,of
~ , considerable
~ interest have been found very recently
concerning rational points on some curves of genus greater than one. More-
over the methods suffice to determine the points and these are also shown to
be finite in number.
Let f ( s , y ) = 0 be a curve I' with coefficients and variables defined in an
algebraic number field K or in a finite extension of K . Let a rational mapping
4, defined over K of r into a curve I', of genus one be given by
d:{u, v > = {Ml(X, MAX, Y)>.
The mappings when M,(x, y ) , M,(x, y ) are constant, form a subgroup G of

the group cf, of the mappings, and this is the group of the rational points of r,.
The mappings are called independent if their images in I?, are linearly inde-
pendent from the view point of the group law of addition of the points of rl.
Then if the rank of the quotient group @ / G is greater than the rank of the
generators of G1,there are only a finite number of points in K on r and these

can be effectively determined.
Illustrations are given by
+ y4 = A , r1:u4 + 02 = A,
with the two independent mappings.
41 :{u, 4 = {x, Y, 4 a : k 4 = { Y , x2>
if G has at most one generator,
and by
r:x6 + y6 = A , G:u3 + u2 = A , or u3 + 1 = Au2 or v2 + 1 = Au3,
with the respective mappings,
if any of the G have at most two generators.

A simplified version has just been given by Cassels.
REFERENCES
1. D. Hilbert and A. Hurwitz. Uber die diophantischen Gleichungen von
Geschlecht Null. Acta. Math., 14 (1890), 217-224.
2. H. Poincark. Sur les proprietks arithmktiques des courbes algkbriques. J .
Marh. (3), 7 (1901), 161-233.
3. L. J. Mordell. On some sextic diophantine equations of genus two. Proc. Amer.
Math. SOC.,1969-1970.
4. V. A. Demjanenko. Rational points on some classes of algebraic curves.
(Russian) Izuesriju Akad. Nauk. (ser. Mat.), 30 (1966), 1373-1 396.
5. J. I. Manin. The Tate height of points on an Abelian variety, its variants
and applications. (Russian) Iztlesrija Akad. Nauk. (ser. Mar.), 28 (1964),
1363-1 390.
6. V. A. Demjanenko. Rational points on some curves of higher genus. (Russian)
Acra. Arith., 12 (1967), 333-354.
7. J. W. S. Cassels. On a theorem of Demjanenko. J . Lond. Math. SOC.,43 (1968).
61-66.
CHAPTER 18
Representation of Numbers by Homogeneous Forms

in Two Variables
1. Problem
Let f(x, y) = aoxn + a1.Y -'y + . . . + a n y , (1)

where the a are integers, and (a,,, a , , . . ., a,) = 1. To solre in integers x, y ,
.k
y ) = m, (2)
where m is any gicen integer.
If m = 0, the only solution is x = y = 0 unlessf(x, y ) has a rational linear
factor and then the solution is obvious. We suppose hereafter that m # 0.
The solution of f ( x , y ) = m is equivalent to the solution of any other
equation F ( X , Y) = m if we can establish a 1-1 correspondence between
integer sets (x,y ) and ( X , Y). Such a one is given by the substitution
if a , ,8, y , 6 are integers and

A=aS-/3y= tl.
These conditions are necessary and sufficient. For on putting (A', Y) =
( I , 0) and (0, l), then a , /3, y , 6 are integers. On putting (x, y ) = (1, 0) and
(0, I), then ./A, jI/h,y / A , S/A and so (a6 - jIy)/A2 = _+ ]/A are integers.
HenceA = f l .
It suffices to consider only the case A = 1. Then the forms f ( x , y ) and
F ( X 7 Y) = f ( a X + 8, Y , y X + S Y) are called equivalent. We write
f(X5.Y) F(x7y),
The concept -
and say that all the forms equivalent to a given form define a class of forms.
of equivalence satisfies the usual equivalence axioms.
We need only consider solutions with (x,y ) = 1. For if (x,y ) = d,
x = dX, y = dYthen
f(X7 Y ) = m/dn,
and so only a finite number of values of d arise.
6*
We show that the solution of equation (2) can be reduced to the case when
tu = 1. We may suppose that (ao, m) = 1 by means of a unimodular substitu-
tion if need be. We may also suppose that ( y , m) = 1 on writing dy for y when
( y , m) = d. For any such solution (xo,yo), we can find I so that xo =
fy,, (mod m) and f ( l , 1) = 0 (mod m). Then the substitution
x = IX -t m Y , y = X,
gives an equation (2) in which m divides out. There are as many such equations
as there are roots of the congruence f ( l , 1) = 0 (mod m).
Inrarian ts
-
When f ( x , y ) F ( x , y ) , algebraic theory shows that certain functions of
the coefficients of f ( x , y ) , say, Z(f), are unaltered by the substitutions of
determinant unity.
Thus if Il(f>= Il(F), I2Cf 1 = 12(F), . . . 9
then 11,12,.. ., are called invariants of f ( x , y).

There are also certain functions of the coefficients and the variables, say
J ( x , y , f ) , which are unaltered by the substitution. Thus if
J(x, y , f ) = 4x9 y , F ) ,
J is called a covariant of f ( x , y).
-
If, however, f ( x , y), F(x, y ) have the same invariants and covariants, it does
not necessarily follow that f ( x , y ) F(x, y). There arises the following
Problem
To construct all the forms with a giren set of inoariants and cotlariants, and
then to separate these into classes of non-equivalent forms.
This is facilitated by picking out a suitable representative of each class of
equivalent forms, say a reduced form, and preferably in such a way that,
firstly, its coefficients are bounded in terms of the invariants, and secondly
that there is only one reduced form in each class.
This leads to other questions:
1. the discussion of the minimum value of I f ( x , y)l for integer values of x, y
not both zero.
This question is a particular case of a general problem in the Geometry of
Numbers or in Diophantine Approximation.
11. to investigate whether the number of classes is finite.
111. to find all the substitutions S changing f ( x , y ) into an equivalent form
F(X, Y ) .
Those substitutions, say A , for which f ( x , y ) = f ( X , Y ) , are called the
automorphs off(x, y). Then all the S are given by
S = AS1,
where S , is any particular substitution changing f ( x , y ) into F ( X , Y ) .
18. NUMBERS BY HOMOGENEOUS FORMS I N TWO VARIABLES 159
We now come back to
f ( s ,y ) = m, (x,y ) = 1.
Suppose that a particular solution is given by (x, y ) = ( a , y). Take any two
integers /3, 6 such that a6 - By = 1. As before, denote by S the substitution
s = ax + BY, y = yX + 6Y.
Then f(x9.Y) = F ( X , Y ) . (4)
say, and the coefficient of the highest power of X in F( X , Y ) is m. Hence we
must now construct the forms F ( X , Y ) whose leading coefficient is mX" and
which has the same invariants and covariants asf(x, y ) . We must then find
those of the F ( X , Y ) which are equivalent tof(x, y). If S above is the corre-
sponding substitution, then (x, y ) = (a,y ) gives a representation of m by
f(X9 Y ) .
We have the same representation independently of the choice of 8, 6. All
the solutions /3', 6' of a6 - ,By = 1 are given by 6' = fl + ta, 6' = 6 + ty,
where& 6 is a fixed solution, and t is any integer. We have now the substitution
s + t Y ) + p Y , y = y ( X + t Y ) + SY,
a(X
-
=
and so f(x,y) = F(X + tY, Y ) F(X, Y).
Hence the same solution (a,y ) arises from F ( X , Y ) and F ( X + t Y, Y ) .

2. We give now a brief rCsumC of the results for the binary quadratic form
f ( x , y ) = ax2 + bxy + cy2. (5)
These are given in the usual textbooks.
There is a single invariant
d = b2 - ~ U C . (6)
If d < 0, the form is definite and it suffices to consider only the case a > 0,
i.e. a positive definite form. The reduced form satisfies
c 2 a 3 lbl, whence a 6 d ( - d / 3 ) , (7)

and so the number of classes of forms of discriminant d is finite. These con-
ditions define a unique reduced form except when c = a or b = & a. We then
have a unique reduced form if we impose the supplementary condition b > 0.
It suffices to consider primitive forms, i.e. those with (a, b, c) = 1.
The automorphs are easily found.
When d < -4, these are
When d = -4, there are also two further ones

0 +1
( & 1; - 0 ) .
When d = -3, there are the four further ones
0, T 1 +1, + 1
L l , ill' ($1, 0).
When d > 0, the form is indefinite. Write
f=
-b + dd
2# 9 s=
-b - dd
(8)
2a
The formf(x, y ) is called reduced if If I < 1 , Is1 > 1 , fs < 0. These conditions
are equivalent to
0 < dd- b < 21al < .\/a+ b. (9)
This implies b < d2,and so the number of classes of forms of discriminant d
is finite. Now there is not a unique reduced form but a chain of reduced forms,
and two forms are equivalent if and only if their reduced forms occur in the
same chain.
The automorphs of a primitive form are given by
where Q = f-(t - bu), /3 = - CU, y = 6u, 6 = f-(t + bu),

where t , u run through the integer solutions of X 2 - d Y 2 = 4. Denote by
T, U that solution with least positive X and Y. Then
t + udd T + Udd
, n = O , & l , f 2, . . .
2
gives all the solutions for t , u.
We remark that in the quadratic field R(d/d), where d has no squared
factors, the units, i.e. the algebraic integers E satisfying N ( E )= f 1, are given
by E = f y", n = 0, t I , t 2, . . . , where y is the fundamental unit.
When
d = 1 (mod 4), E =x + y (T),1 + dd x2 + xy + y2 (T)

1-d = + 1,
d = 2,3 (mod 4), E = x + ydd, x2 - dy2 = t 1,
where x, y are integers and the minus sign is taken if the corresponding equa-
tion is solvable.
18. NUMBERS BY HOMOGENEOUS FORMS IN TWO VARIABLES 161
The solution of f ( x , y ) = in, where m is square free, requires the construc-
tion of forms
F ( x , y ) = mx2 + Bxy + Cy2
equivalent to f(x, y). Since
B2 - d
B 2 - 4mC = d, ~
4m
= c,
a necessary condition for the representation is that there exist integers B for
which Cis an integer. In particular, dmust be a quadratic residue of every odd
prime factor p of m, i.e. ( d / p ) = 1. N o other conditions arise when m is odd
since d = 0, 1 (mod 4).
Two simple illustrations may suffice:
If m is odd, every prime factor p of m must satisfy (- I/p) = I , i.e. p --=

1 (mod 4). Since there is only one class of forms when d = -4, typified by
+
x2 y 2 , every such odd number m can be represented as x 2 y 2 = m. +
B2 - 8
d=8, --
4m
- c.
If m is odd, every prime factor p of m must satisfy (2/p) = 1, i.e. p =
? 1 (mod 8). Since there is only one class of forms when d = 8, every such m
can be represented as x2 - 2y2 = m.
If the number of classes of forms with discriminant d is greater than 1, it is
in general a very difficult problem to determine which of the classes represents
m.
3. We shall now give Hermite's's2 method of reducing the general binary
form. Considerable numerical work may be required in applying it except
perhaps for the binary cubic which will be dealt with separately further on.
Theorem 1
The binary forms with integer coefficients,
f ( x , y ) = aoxn + a,x"-' + . . . + anyn, (1 1)
with giivn invariants, can be arranged in afinite number of classes.
We may suppose that a,, # 0. Suppose thatf(x, y ) has X real linear factors
x + city, and p pairs of conjugate imaginary factors x &y, x + +
y j y . Con-
sider the quadratic form
where the t and u are real positive variables. Let the unimodular integral
substitutions ;
x =pX + qY, y = rX + sY,
changef(x, y ) into the equivalent form
F ( X , Y ) = AoX" + A I X " - ' Y + . . . + Any".
Then A0 = a0 n (P + n
A
i=l
air)
U
j=1
(P + Bjr)(p + y j r ) .
Let S change g(x, y ) into, say,
2 Tf(X + +2 2
A
+ /-$Y)(X + y i Y ) ,
U
G(X, Y ) = U:(X (13)

1=1 j=l
= PX2 + QXY + RY2, (14)

of discriminant D.
We note that the set of forms g(x, y ) corresponding to real t , u, is the same
as the set of forms G ( X , Y ) corresponding to real T, U . For
x + uy = ( p + ur)X + ( q + us)Y,
and so we need only write
+ air) = Ti, u,"(P + Bjr)(p + =
~ j r ) u,Z. (15)
Then, say,
A0 -
- a0 -
- -. a0 (16)
T I . . .TAUT,. . U : t l . . . t A u z . . .U; h(t, U)
We shall now show that the coefficients 1, u can be chosen so that g(x, y )
is a real quadratic covariant off(x, y). We then call F ( X , Y ) reduced if the
quadratic form G ( X , Y ) is reduced. We show that in this case the coefficients
A,,, A l , . . ., A , are bounded in terms of the invariants off(x, y).
From equation (1 3),
P = T: +
+ TZ + 2U: + . . . + 2 U ; ,
9 ' .
R = ui2TT + . . + ai2T: + 215;yiUz + . . . + 2 f ( , ~ ~ U ~ .

Put Tl = t1d\/P, u1= vldP,etc.
U'T~ = +dR, B'U, = +,dXefW, yf y1 = +ldXe - l w ,
etc., where I/J, etc. are real. Then
tz + " ' + t?+ 217:: + . . . + 2 1 ; = I , (17)
@ + . . . + @ + 2 * ? + . . . + 2*;= 1. (18)
18. NUMBERS BY HOMOGENEOUS FORMS IN TWO VARIABLES 163
Here the E, 7, 4, + are positive variables.
Now
T i . . .TAU;. . . U E F ( X , Y )
= A. nA
i=l
(TiX + u;T,Y) n ( U j X + p i U j Y ) ( U j X + y l U j Y ) ,
11
j = 1
or
Hence we can write for 0 < m < n,
where I1 is a real polynomial in 5, 7, 4, $. Then as seen from equations (17)

and (18), 12 has an attained maximum value Q0 depending only on n.
Suppose now that G ( X , Y ) is reduced. Then PR < 3 0 , and so
The discriminant D involves only the T, U and so we take for T, U such values
as make
E = DnizAE/h(T,U ) 2
a minimum. This process defines a quadratic covariant off(x, y).
It does not follow at once that the inequality above establishes bounds for
A , since A n - , may be zero. This difficulty was noted by Julia3 who produced
a satisfactory treatment based upon the principles given in Hermites first
paper.
REFERENCES
I . C . Hermite. Note sur la rkduction des fonctions homoghes a coefficients
entiers et a deux indkterminees. Oeuvres, 1 (1905), 84-93.
2. C. Hermite. Sur lintroduction des variables continues dans la theorie des
nombres. Oeuvres, 1 (1905), 173-178.
3. C. Julia. Etude sur les formes binaires non quadratiques. Mem. Acad. Sci.
Ilnst. Franc(>,55 (1917), 1-293.
CHAPTER 19
Representation of Numbers by Special Binary Quadratic

and Quaternary Quadratic Forms
1. We deal here with the representation of numbers by special quadratic

forms in 2 and 4 variables. Such questions are included in the general theory
of the representation of numbers by quadratic forms which is discussed in
Chapters 18 and 20. Many results however can be found very simply without
using the general theory, and this had been long known when a = b = 1.
The method now expounded is due to Thue, who applied it in a few instances,
as did also Nagell, but it is capable of wider application.
The equation
ax2 + by2 = m.
We suppose that m > 0 and that (ab, m) = 1.
Write +
f ( x , y ) = ax2 by2 = m,
We shall suppose that ( x , y ) = 1 and so ( x y , m) = 1.
Solvability of equation (l), if ( x , y ) = 1, implies the existence of an integer
k such that
+
ak2 b = 0 (mod m). (2)
This imposes conditions on a and b. Thus if m is odd, the Legendre symbol
( - a b / p ) = 1 for every prime p dividing m. We first prove the l s 2
Theorem 1
If the congruence
ak2 +b 3 O(modm)
is soloable, integers x , y , not both zero, exist such that
ax2 + by2 = M m (3)
for some integer M with M < 2dab if a > 0, b > 0, and IMI < d/labl if
ab < 0.
Put x = k X + m Y,y = X . Then by Minkowski's theorem on linear forms
in Chapter 5, there exist integers (A', Y ) # (0,O)such that
1x1 < fdm, lyl < I-wm,
where 1 is any positive constant.
19. BINARY A N D QUATERNARY QUADRATIC FORMS 165
Hence if a > 0, b > 0.
f(x,y ) < (UP+ b/I2)m < 2 d a b m ,
if we take l 2 = dbz.
If ah < 0,
1f ( x , y)l < max( la1I2rn,lb1Pm) < d l q m,
if we take l 2 = dml.
Further
f ( ~y ), = + m Y)2+ bX2
a(kX
= (ak2 + b ) X 2 + 2akmXY + am2 Y 2 = 0 (mod m )
= Mm.
The problem now is to determine whether M = 1 is possible. Many of the
possibilities for M can be eliminated by congruence considerations. Suppose
that m is odd.
I. congruences mod 4 or mod 8. Thus all to mod 4, a 3 1, b = -1
excludes M = 2 (mod 4) since then x and y are odd and x2 - y 2 I
0 (mod 4). If a = I , b = 1, 2, then solvability for M = 0 (mod 4) implies
solvability for M / 4 since now x = y = 0 (mod 2).
11. congruences mod q, q = 3, 5,7,. . ., a prime. If (-ab/q) = - 1, then
M $ 0 (mod q).
111. sometimes it may be shown that if M = M,, say, is possible, then
M = M 2 is also possible and conversely. For this we use the results,
if ax2+ by2 = m, ax: + by? = ml,
then (axx, + byy# + ab(xyy,- x , ~ =) mm,;
~ (4)
and if x 2 + aby2 = m, as: + by: = m,,
then a(xx, + byy,)' + b(sy, - ays,)2 = mm,. (5)
1V. If (m, ab) = 1, and a is square free, then solvability of ax2 by2 = am+
+
implies that of x2 aby2 = m and conversely.
We suppose for simplicity that m is an odd prime p since solvability when
171 is composite follows easily from the identities in 111. Let us now consider
the detailed results.
We begin with the case a > 0, h > 0. We take a = 1 and then
.'i2 + by2 = My, M < 2db,
and (- b/p) = I is the necessary condition for solvability. We write this in a
slightly different form, for example, if b e 1 (mod 4), then (- l / p ) ( p / b )= 1,
and ifb E 3 (mod 4), then (p/b) = I . l f t b = r = 3 (mod 4), then ( 2 / p ) ( p / r )=

1, but if 46 = r = 1 (mod 4), then ( - 2 / p ) ( p / r ) = 1.
2. It will suffice if we give the results for square free b < 20 and give details
for a few values of b.
M = I i f b = I , 2,3, 7 and (-b/p) = I
Take for example b = 7 and so M < 2/28 = 1 , 2 , 3,4, 5 . We can exclude
M = 2 , 4 since then x E y = 1 (mod 2 ) and x2 + 7y2 = 0 (mod 8). We can
exclude M = 3, 5 since (-7/3) = - 1 = ( - 7 / 5 ) .
M = I i f b = 5providedthat(-l/p) = ( p / 5 ) = l . I f ( - l / p ) = ( p / 5 ) = -1,
then M = 2 or 3 and either value implies the other. The condition (- 5 / p ) = 1
gives (-lip) = ( p / 5 ) . Here M < d% = I , 2, 3,4. If M = 4, then x =
y = 0 (mod 2) and implies M = 1. For this, ( p / 5 ) = 1 and so (- l/p) = 1.
Then M = 2 or 3 can arise only when (2p/5) = 1 or (3p/5) = I , i.e. ( p / 5 ) =
- 1.
+ +
Suppose now x2 5y2 = 2p. Multiply by l 2 5. l 2 = 6. Then from
equation (4),
x2 + 5Y2 = 3p,
where .x + 5y = 2 x , x - y = 2Y,
and so 3 s = x + 5Y, 31' = x- Y

If x and y are given, x = y (mod 2 ) and so A', Yare integers. If X and Yare
given, we can choose Y so that X = Y (mod 3), and then x, y are integers.
M = I ifb = 6 provided that ( 2 / p ) = (p/3) = 1 but if ( 2 / p ) = (p/3) = - 1,
then 2 . 2 + 3y2 = p .
M = 1 ifb = I0 provided that ( - 2 / p ) = ( p / 5 ) = 1 but if ( - 2 / p ) = ( p / 5 ) =
- 1, then 2x2 +5y2 = p .
M = I or 4 jf h = I 1 or 19, but we cannot remove the ambiguity. The
+
solvability of s2 by2 = Mp means that
s 2 + sy + 31'2 =p, A-2 + xy + 5y2 = p
are both solvable.
M = l i f b = 13providedthat(-l/p)=(p/13)= l b u t i f ( - l / p ) = ( p / 1 3 ) =
- 1, then A4 = 2 or 7 and M = 2- M = 7.
M = I i f b = 15 provided that (p/3) = ( p / 5 ) = 1 but if ( p / 3 ) = ( p / 5 ) = - I ,
then 5 2 + 3y2 = p .
Different kinds of results arise when b = 14, 17. Suppose first that b = 14.
Then if (2/p) = (p/7) = 1, either M = 1 or 2 2 + 5y2 = p and we cannot re-
move the ambiguity. If (2/p) = (p/7) = - I , then either M = 3 o r 2x2 +
5y2 = 3p, and we cannot remove the ambiguity.
When h = 17, M = 1 o r 2 if ( - lip) = (p/17) = 1 and we cannot remove
the ambiguity. I f (- I/p) = (p/17) = - 1 , M = 3, 6, 7, each implying the
other two.
19. B I N A R Y A N D Q U A T E R N A R Y Q ~ J A D K A T I CFORMS I67
We now consider b = 1 1 in detail. Here A.i < 2d71 = I , 2 , 3, 4, 5, 6. We
can exclude M = 2, 6 since s2 + 1 ly2 5 0 (mod 4). If M = 4, the case when
x, y are even reduces to M = 1.
We show that the case when s and y are both odd, and the cases M = 3, 5,
each imply the other two.
Suppose s2 + 1 l y 2 = 4p. From l 2 + 1 I . l 2 = 12, we deduce
A2 + l l B 2 = 3p,
where
4A=.~+Ily,4B=.~-y; 3y=A-B, 3.\-=A+lIB.
By choice of sign of y , A and B are integers. If A , B are given, we can take
A = B (mod 3), and then x, y are integers.
+
Suppose next that x2 1 1y2 = 5p. From 22 11. l2 = 15, we deduce +
C2 + 110' = 3p,
where
5C = +
2 . ~ 1 IJ, 5D = s - 2 ~ ; 3.u = 2C + 11D, 3 , ~= C - 2 0 .
We can take x = 2y (mod 5) and later C = - D (mod 3), and then have
integer correspondences.
We show that if these three cases occur, then M = 1 cannot occur and con-
+ +
versely. This is easy t o see from X 2 XY 3Y2 = p, but the result is a
particular case of the
Theorem 2
The equations
ax2 + by2 = Ip, ax: + by: = mp, (.u, y) = 1, (x,, y , ) = 1.
cannot both be solrable if a > 0, b > 0, (ah, p ) = I, I # in, ltiz < ah.
We may take to mod p,
x E ky, s1 = k y , where ak2 + b = 0.
Since (asx, + byy,)2 + ah(.uy, - = Imp2,
on putting axx, + byy, = p X , xy1 - s l y = p Y ,

where X , Yare integers, then
X2 + ahY2 = In?.
Hence Y = 0, i.e. s / y = and then s = ? sl,
xl/jll, y = f y , , and this is not
so.
3. We turn now to the case ab < 0. We write equation ( I ) in the form
5 2 - cy2 = Ep, E = f 1.
Now
Some results may be mentioned.

M = I if c = 2, 5, 13, 17. We also have here M = - 1 since x2 - cy2 = -1
is solvable for these c on noting y = 5 for c = 13.
M = k I w h e r e c = 3 , 6 , 7 , 11,14,19,andif
c E 1 (mod 2),
($) k 1 , = (5) = +I,
and if c = 0 (mod 2), ($) *1, = (6) = +I.
The sign corresponds to M = k 1.

M = k I i f c = 15and
(t) = (5) = 1, p = f 1 (mod 4).
M = k 1 or * 2 when c = 10 according as
(f)= (5) = I or - I .
The details are comparatively simple except for c = 19 and so we give them.
Here
x2 - 1 9 ~ '= Mp.
We can exclude IM1 = 4 since this reduces to IM I = 1. We show that any of
M = 1, - 2, - 3 implies each of the other two.
For from X' - 19y3 = p , 4' - 19.1' = -3,
A' - 19B2 = -3p,
where
A = 4~ - 19y, B = x - 4 ~ ; 3~ = -4A + 19B, 3y = -A + 4B.
Since we can take A = B(mod 3), M = 1- M = -3.
Next M = - 2 tf M = -3. For from
S' - 1 9 ~ '= -2p, 5' - 19.1' = 6,
C' - 1 9 0 ' ~-3p,
where
2C = 5~ - 19y, 2 0 = x - 5y; 3~ = 5C - 19D, 3y = C - 5D.
19. BINARY AND QUATERNARY QUADRATIC FORMS 169
The result is now obvious since we can take x = y (mod 2), C =
- D (mod 3 ) . Similarly, we can show that each of M = - 1 , 2 , 3 implies each
of the others. Hence when (- I/p) = I , we can take M = 1 since M = - 1 is

excluded by the congruence x2 - 19y2 = - p = - 1 (mod 4).
So when ( - I/p) = - I , we have M = - 1.
4. We now consider the representation of numbers m, which we may suppose
square free, by the special quaternary forms
j = f(x, y , z , w) = x2 + bcy2 + caz2 + abw2. (6)
These forms admit a composition process. Thus if
fl = f(s1, y,, z,, w,) = xf + bcy: + caz: + abw:,
then fz = ffl = f(-hY 2 , z2, )+>2), (7)
where XZ = . Y X ~- (bcyy, + CaZ.7.1 + abrtlw,),
yz = y s , + xy, + U(ZM, - WZ1),
~2 = + + b(rty, yw!,),
ZS, XZ, -
112 = wx1 + + c(yzl zy,).

XM, -
Hence iff represents both m and m,, thenSrepresents mm,.
Theorem 3
Integers (s,y , z , it,) # (0, 0, 0, 0 ) exist such that
j ( x , y , z , w) = Mm, I M 1 < d2FFC1, (8)
prolitled that the congruence
cA2 + bB2 + a E O(modrt7) (9)
is solaable.for A and B.
It is shown in Chapter 6 that the congruence is solvable if (m, abc) = 1,
m f. O(mod4).Puts = c A X + b B Y + m Z , y = B X - A Y + mW,z = X,
11 = Y. From equations (6) and (9),
,f(s,y, 2 , I+!) = F ( X , Y, 2,W ) = Mm, (10)

where M is an integer, and the determinant D of F is a2b2c2m4.
There are several methods for estimating a minimum value p o r lower
bound for a quaternary form F ( X , Y , Z, W ) of determinant D. The most
elementary is to use Hermites* lpl < (4)32qrq, and so M < 1 . 5 4 d m l .
Next perhaps is the application of Minkowskis theorem on linear forms in
Chapter 5. Thus integers (X,Y , Z , W ) # (0, 0, 0, 0) exist such that, if
6 = q/labcl, then
61x1 < 2- labanl,
f ~~ ~~
Sly1 < ,I%[%

SIzI < djhnll, 61wl < dlGl.
This gives IMI < 4dlad;l. The estimate (8) follows from the well known
more precise results of Korkine and ZolotarefP lpl < 2/4\01. These
4
estimates are usually given for positive definite forms, but they also hold
for indefinite forms although the results are not very sharp.
We commence with the case c = 1 and so
f = x2 + by2 + az2 + abw2 = Mm, IMI < dm, (1 1)
and +
A 2 + bB2 a = 0 (mod m), (m, ab) = 1 (12)
The classical case is a = b = 1, Hermites result gives M < 1-54, i.e. M = 1.
Minkowskis estimate gives A4 < 4 and so M = I , 2, 3.
(F)
The case M = 2 can be written as
+ (y)
+ + + (T) ( Tz ) z U Z-MM = m,
since either x, y , z , M: are all even or odd or only two odd, say x, y. The case
M = 3 can be written as
( + ; +; + 3
(y +; + M
)Z + - 2)Z (x - w) + (x
since we may suppose s = 0 (mod 3), y = z E w = 1 (mod 3).

-y = 171, +
1
The method shows that m is representable by f in equation ( 1 I ) when
b = I , a = &3, 5 5 , -7, -11 and also when a = 5 2 , b = 3, -3. It will
suffice to take a few of these cases.
b = l,a = 3 M < d6 = 1, 2, (m,3) = 1.
If M = 2, s2 + y2 + 3(z2 + w2) = 2m. (13)
If x = y (mod 2), then z = w (mod 2), and
and this is the case M = 1.

If x = y +
1 (mod 2), then z = 14 + 1 (mod 2), and so from equation (13)
1 + 3 = 2m (mod 4), m = 0 (mod 2).
Hence M = 1 if m = 1 (mod 2). I f m = 0 (mod 2), the argument applies to
t m , and since 2 is representable b y f so is m .
Finally if m z 0 (mod 3), +m is representable by f and so is 3 and hence
also m.
b = I , a = -3 ] M I = 0, 1,2, (m,3) = 1.
Since - 1 is representable, we need only consider M 3 0.
Clearly M = 0 is impossible since this requires x 2 y = 0 (mod 3) and then
z = w = O(mod 3). If M = 2, the argument above applies when x =
y (mod 2), z = iti (mod 2). For .Y E y +
1 (mod 2), z = it 1 (mod 2), we +
need only consider
x = 1 (mod 2), y = 0 (mod 2), z = I (mod 2), w = 0 (mod 2),
Since 12 - 3.12 = -2,
we have
x + 3z 7
( - 3 ) y + 3w 7
+'
( -3 y)
+ w 2 ='
-m,
(x+)2 (T)
and this is the case M = - 1. The case m = 0 (mod 3 ) is dealt with as above.
b = I , a = 5, then (m, 5) = 1 and this case is of some interest.
Here x2 + y 2 + 5(z2 + w2) = Mm, M = 1,2, 3.
Clearly M # 1 for all m since 3 is not representable. It can be shown from the
theory of the ternary quadratic that all integers > 3 are representable with
M = 1. However, all integers are representable with M = 2. Thus if M = 1,
+ y)2 + (x - y ) 2 + 5(z + w ) +~ 5(z - w ) =~ 2m.
(x
If M = 3, compound with l 2 + 5. l 2 = 6 . Then
(x + 5z 7 + 5 )+' y + 5w + 5 (7)'
(x+)2 = 2m
(T)2
is an integral representation. For

x2 + y 2 = 2 + w2 (mod 3),
and we can take
= y = 1,
x z = w = 1 (mod 3),
and .Y = 0, y = 1, z = 0, w = 1 (mod 3).
If m = 0 (mod 5), say m = 5m,, from a representation
+ J J ~+ z2 + w2 = 2m,,
xy
(2x1 + J J , ) +
~ (x, - ~ J J , +
) ~5(z2 + w 2 ) = lorn,.
b = l,a = -7 [MI = 0, 1,2, 3, (m, 7 ) = 1.
Clearly M = 0 is impossible and we may suppose M > 0, since 32 + 22 -
7(12 + 1') = -1.
For M = 2, x2 + y2 - 7(z2 + w2) = 2m.

The case x = y (mod 2), z = w (mod 2) is dealt with as for a = & 3. We need
only consider x = 1, y = 0, z = 1, w = 0, all to mod 2.
Since 3' - 7 . l 2 = 2, we need only note
( T3x
)2 + 7z - 7 (x*)2 + 3y
(-z-)2 + 7w - (T*)
7 y+3cr = m.
For M = 3, x2 + y 2 - 7(z2 + w2) = 3m.

From 22 - 7 . I' = -3, we deduce

(+2s' + 7z - 7 (V)'
+ 2y
(7)2 + 7w - 7 y(+ + 2w = -m.
The brackets are integers since it suffices from x2 + y2 = z 2 + w2 (mod 3), to

consider only
x = z = l , = w E 0 (mod 3),
y
and x = y = z = w = 1 (mod 3).
When m = 0 (mod 7), it suffices to put x = 7 X , y = 7 Y .
We take finally a = k 2, b = 5 3, c = 1. It is well known that the result
for a = 2, b = 3, c = 1 follows from a = b = c = 1,
since
x2 + 2y2 + 3z2 + 6w2
= x' + ( y + z + w)' + ( - y + z + ')M + (z - 242.
We have now the solvable equations with (m, 6 ) = I ,
I . x2 - 3y2 - 22' + 6w2 = M m ,
11. x' + 3y2 - 2z' - 6w2 = M m ,
111. x2 - 3y2 + 2 2 - 6w2 = M m .
Here 1 M I = 0, 1, 2, 3. Since each of the forms represents - 1, we need only
consider M = 0 , 2 , 3.
Consider I. Clearly M # 0 since x2 - 2z2 = 0 (mod 3), gives x = 3X,
z = 32, and then y = 3Y, w = 3 W . I f M = 3, x = 3 X , z = 3 2
and so 3 X 2 - y 2 - 6 Z 2 + 2w2 = m,
and this is I with M = - 1.
Suppose next M = 2. Then x =y (mod 2). If x = 2X, y = 2 Y,
2 X 2 - 6 Y 2 - z2 + 3w2 = m,
and this is I with M = - 1. If, however, x =y = 1 (mod 2), apply com-
position with
1' - 3.1' - 2.12 + 6.1' = 2.
Then x2 = x + 3y + 22 - 6w, y2 =x +y - + 2w,
22
z2 = x + z + 3y - 3w, W
' =x +y - z + w,
Hence if w = z (mod 2), x2 = y 2 = z2 = w 2 = 0 (mod 2), and this gives I

with M = 1.
We can reject IV =z + 1 (mod 2) since from I
-2 - 2(w - 1)2 + 6w2 = 2m (mod 4), i.e. m = 0 (mod 2)
which has been excluded.
When m = 0 (mod 6) the representation holds for mi6 and for 6 a n d so for
1H.
When m = 0 (mod 2), the representation holds for j m and for 2 a n d so for
m.
When m = 0 (mod 3), the representation holds for +m and for 3 a n d so for
m.
We show now that each of 1 and I1 implies the other.
Multiply I with M = 1 by l 2 - 2 . l 2 = - 1. Then
+ 2 ~ - )2(x~ + z ) +~ 3y2 - 6w2 = -m.
(s
Since x + 2z = X , x + z = 2 gives a 1-1 correspondence between integer

sets (x, z ) and (A,Z ) , and we have I1 with m replaced by -m.
Take finally 111 and multiply by l 2 - 3 . l 2 = -2. Then
(x - 3 ~ -)3(x~ - Y ) -
~ 4z2 + 12w2 = -2m.
Write this a s P Q - 3RS = -2m,
where P = x - 3y -t 2z, Q = x - 3y - 2z,
R = x -y + 2w, S = x -y -2 ~ ,
Then x, y , z , IV will be integers if
P = Q (mod 4), R = S (mod 4), P = R (mod 2).
If m E 1 (mod 2), we can solve 111 with P = Q = R = S = 1 (mod 4).
If m = 0 (mod 2), 111 holds with M = 1 for +m, for 2 and so for m.
REFERENCES
1. B. A. Venkoff. Elementary Theory of Numbers. (Russian.) (1937), 51.
+
2. L. J. Mordell. Solvability of the equation ax2 by2 = p . J. Lond. Math. Soc.,
41 (1966), 517-522.
3. L. J. Mordell. The representation of numbers by some quaternary quadratic
forms. Actu Arith., 12 (1966), 47-54.
4. J. W. S. Cassels. An Introduction to the Geometry of Numbers. Springer-
Verlag, Berlin (1959), 31.
5 . L. J. Mordell. Observation on the minimum of a positive quadratic form in 8
variables. J. Lond. Math. Soc., 19 (1944). 3-6.
CHAPTER 20
Representation of Numbers by Homogeneous Forms

in Several Variables
We give a few brief remarks about the general theory, for which various
account^^-^ can be consulted.
1. Let f(x) = f ( x , , x2, . . . , x,) be a homogeneous form with integer co-
efficients of degree d in n variables. We discuss the integer solutions for
(z) = (xl, x2, . . . , x,) of the equation
f(r)= m (1)
for given m. We may suppose that the representation of m is primitive, i.e.
(xl, x2, . . ., x,) = 1. If ( p 1 , p 2 ,.. ., p , ) is such a solution, then by means of a
P1
unimodular substitution with integer coefficients and first column (!'), we

Pn
can transformf(2) into a form g ( r ) in which m is the coefficient of x;'. Hence
the problem of representation requires the following three steps.
1. To find all the forms g(s) with the same invariants and covariants asf(s)
and with m as the coefficient of xf.
2. To find those of the g ( r ) which are equivalent tof(x).
3. To find the substitutions S which transform g ( r ) intof(z).
P1
If the first column of S is (p'), then z = j gives all the representations.

Pn
We remark that all these substitutions are given by S = S'A where S' is
any one of them and A runs through all the automorphs off(z).
It can be shown2 that reduced forms exist forf(x) and that the class number
is finite. This requires that the coefficients of the reduced form be bounded in
terms of the invariants of f(s).Then questions arise about the minimum
value of f ( r ) for integer r . There are very few results for general forms of
degree d 2 3. The theory of quadratic forms which arises when d = 2 has
been thoroughly investigated. We outline the method l.
Let f(x) = 2
r.s= 1
arsxrxsi ars = asr, (2)
with integer coefficients ur,s( r , s = I , 2, . . . , n) and determinant

D = IlarsII z 0.
20. REPRESENTATION OF NUMBERS BY HOMOGENEOUS FORMS 175
We can then write g(.r) as
g(s)
1
= -((my1
m
+ h,(z))2 - h2(.r)), (3)
where h,(s) is a linear form in x 2 , . . ., x, and h,(z) is a quadratic form in

x2,. . ., x, of determinant mn-,D. Then g ( z ) has integer coefficients if and
only if the congruence
hT(.r) = h,(.r) (mod m), (4)
holds identically in the variables s,, . . ., x,. Next we have to find all the
quadratic forms h2(z), say h2,,(.r), in n - 1 variables and of determinant
n F Z Dfor which the congruence (4) is possible, and then all the solutions for
h,(x) incongruent mod m. Let there be, say, w , solutions of h?,,(x) = h2,,(x)
-
(mod m) for which g ( x ) f ( x ) . This of course requires a knowledge of all
the classes of quadratic forms i n n variables of determinant D. Then if
6, is the number of automorphs of h2,s and 6 those off(rT), the number of
primitive representations, is SU,/S,.xs
2. We illustrate the method by taking the simplest case
f ( x ,y, z ) = x2 + y 2 + z2 = m,
for which no really elementary treatment is known.
Theorem 1
Every integer m > 0 not of the form 4"8p + 7), h 3 0,p 3 0, is the sum o j
three integer squares, i.e.
x2 + y 2 + z2 = m (5)
is soloable.
Here D = 1, and we show that the class number is one. We require an
estimate for the minimum value taken by a positive definite form for integers
(x,y , z ) f (O,O,0).
Lemma 1
Let g(x, y , z ) = ax2 + by2 + cz2 + 2fyz + 2gzx + 2hxy
be a positive definite form with determinant
Then integers .Y,y, 2 not all zero exist such that

f(x, y , z ) ,< 4D''3.
We may suppose that the minimum is attained for integers x, y , z for which
(x, y , z ) = 1 and, by means of a unimodular substitution, we may suppose
that a is the minimum. Then for all integers (x, y , z ) # (0,0, 0),
g(x, Y , z ) 2 (1.
Hence
(ax + hy + g z y + (ab - h2)yZ + 2(af - gh)yz + (ac - g2)z2 2 2.
We give y , z integer values Z(0, 0 ) ,which make the binary quadratic in y , z a
minimum. Since we can satisfy
Py2 + Qyz + Rz2 < p3- Q2,
this minimum is
< (+)12((ab- h2)(ac - 8) - (af - gh)) = (+aD)l.
We then give x an integer value such that

lax + hy + gzl < +a.
Hence
4a D
-3
2 ($a2)2,
64D
and so a3 < -.
27
It may be noted that the best possible estimate is a3 < 2 0 .
We prove now that the class number is one.
Lemma 2
If a, b, c, f, g , h are integers and D = 1, then
g(x, y, z ) - x2 + y2 + z 2 . (6)
We may suppose that a < 4/3 and so a = 1. On writing x - hy - gz for
x, we may suppose that g = h = 0,
and so g(x, y , z) - x2 + by2 + 2syz + cz2.

+
Here by2 2fyz + cz2 is a positive definite binary quadratic form of determi-
nant - 1 and so it is equivalent to y2 + z2.
This proves the lemma.
We now proceed with the representation of m. We must construct positive
definite ternary forms g ( x , y , z) of determinant 1 and first coefficient m.
Then mg(x, y , z) = (mx + qy + rz)2 + Qy2 + 2Syz + Rz2,
say, where QR - S 2 = m,and to mod m,
q2+ Q=O, r2+ R=O, qr+S=O.
I t is, however, simpler to take
g(x, y, z) = mx2 + 2xz + by2 + 2fyz + cz2,
where b > 0, c > 0, f a r e to be found, and A = bc - fz > 0.
m 0 1
Also 1 = 0 b f = m(bc - f 2 ) - 6.
1 f c
A+f2
Then b=mA- 1, c=--.
b
Now we need only find A > 0 and f such that c is an integer. We shall take A
to be a linear form defined mod 4 or mod 8, and b = p or 2p where p is an
odd prime which will be specified later. We have to satisfy (- A/p) = 1 and we
note that @/A) = (- 1/,4), the symbols being the quadratic characters.
Various cases must be considered depending upon the form of m. Write
m = 4*mm,, m, > 0, m, = 1, 2, 3 (mod 4). Then m is representable if and only
if m , is representable. For from
x2 + y 2 + z2 = m E 0 (mod 4),
clearly x = y = z = 0 (mod 2),
and so x = 2x,, y = 2yl, z = 22,. Hence we need only consider A = 0.
The representation is impossible if m = 7 (mod 8), since x2 = 0, 1, 4
(mod 8), and x 2 + y2 + z2 g 7 (mod 8).
Suppose next that m = 1, 2, 3, 5, 6 (mod 8). Take first m = 2, 6 (mod 8).
We put A = 4t + 1, b = 4im + m - 1. Since (4m, m - 1) = 1, we can take
t by Dirichlets theorem on the primes in an arithmetic progression so that b
is a prime p , and p = 1 (mod 4).
Then
Suppose next that m = 1 , 3 , 5 (mod 8). Take

A=8t+r, 6=(8t+r)m-I, r = l i f m ~ 3 ;r = 3 i f m = 1 , 5 .
We can take t so that 6 = 2p, wherep is an odd prime,
(4m,7)rm - 1 = 1.
since
178
--
DIOPHANTINE EQUATIONS
Hence we have
m
m
m
-
- -
I
= 3,
1,
5,
A
A
A
= 3, (mod 8),
1, (mod 8),
3, (mod 8),
p
p
p -
1 (mod 4),
1 (mod 4),
3 (mod 4).
Hence in all cases (- 2 / A ) = 1, and either - A = 1 (mod 4) o r p = 1 (mod 4).
Then
($) (5) = = (2)(i)
(2) = = = 1.
This concludes the proof.

3. There are other simple results about the numbers which can be represented
by the form
f ( a , 6) = x2 + ay2 + bz2,
for particular values of a and b.
These can be proved by the general theory and sometimes by special
methods. Thus an integer m > 0 cannot be represented byf(a, b) for the a
and b below if and only if,
a = 1, b = 2, m = + 14),
4(16p
a = 1, b = 3, m = 99p + 6),
a = 2, b = 3, m = 4(16p + 10).
These results can be used to prove5 that an integer m can be represented by
some quaternary forms
f ( a , b) = a(x2 + y2 + z2) + bw2.
We may suppose that m f 0 (mod 4), on putting x = 2Xetc.
-
classical case. For we need only take w = 0 if m
m 3 (mod 4).
-
Let us take a = 1. Then every positive integer is represented when b = 1, the
I , 2 (mod 4) and w = 1 if
When b = - 1, every integer positive, negative or zero is represented. Thus if
-
m = 0, we take w = 1. For other m, we take w such that w 2 + m > 0, and
according as m = 1, 2 (mod 4), or = 3 (mod 8) or -7 (mod 8), we take
w 0 (mod 2), w = 0 (mod 4), w = 2 (mod 4).
If m > 0, a representation is also possible for 1 Ib I 7. We may suppose
that m = 4(8p + 7) as otherwise we can take w = 0. Suppose first that
b = 1,2,4, $ 6 . Take w = 2, then m - bw2 = 48p + 7 - b) and is not of
+
the form 4(8p 7). Suppose next that b = 3. If p = 0 we take w = 2Aand
then m - bw2 = 4+, and if p 2 1 take w = 2l and then m - bw2 =
4(8p - 5).
Finally, let b = 7. If p = 0, I or 2, take w = 2 and then m - bw2 = 0, 2
4+ or 4A+2.If p 2 3, take WJ = 2l. Then m - bw2 = 4(8p - 21).
Some results on the representation of complex integers by special binary
and ternary quadratic forms are very simple. Niven has proved the
Theorem 2
A representation of m = a + 2ib, where a and b are rational integers, as
s2 + y 2 = m,
where x , y are complex integers, is always possible except when a = 2 (mod 4),
b = 1 (mod 2).
We give a simpler proof6. Write
(x + iy)(x - iy) = a + 2ib.
If a is odd, a solution is given by
s + iy = a + 2ib, s - iy = 1.
Suppose a = 2a, is even. If a, and b are both even, take
x + iy = a, + ib, s - iy = 2.
If a, and b are of different parity, take
x + iy = (1 + i)(a, + ib), x - iy = 1 - i.
If a, and b are both odd, the representation is impossible. For if
x=X+iY, y=Z+iW,
X2 - Y2 + 2' - W 2 3 2(mod4), XY + Z W = 1 (mod2).
Then say Z W E 1 (mod 2), Z 2 - W 2= 0 (mod 4), and X 2 - Y 2 = 2 (mod 4)
is impossible.
We use this theorem to prove
Theorem 3
A representation of m = a + 2ib, where a and b are rational integers, as
s2 + y2 + z 2 = m,
where x , y , z are complex integers, is always possible.
For take z = X + iY. Then we have a representation from Theorem 2, if
the integers A', Y satisfy
a - X2 + Y 2 $ 2 (mod 4).
We take X = Y 3 1 (mod 2) if a is odd, and X = 1, Y = 0 (mod 2) if a is
even.
+
Mr. Makowski notes that a solution when m = 2a 2ib is (1, a + ib, b +
+ +
i- ai), and when M = 2a 1 2ib, is (2,a - 1 + ib,b + 2i - ia).
REFERENCES
I . P. Bachmann. Die Arithmetik der quadratischen Forrnen, Part 1. Teubner
(1898).
2. L. E. Dickson. Studies in the Theory of Numbers, University of Chicago
Press (1930). Also his History, Vol. 3, Chapter XIV.
3. B. W. Jones. The Arithmetic Theory of Quadratic Forms. Math. Assoc.
of America and John Wiley and Sons (1950).
4. G. L. Watson. Integral Quadratic Forms. Cambridge University Press (1960).
5. S, Ramanujan. On the expression of a number in the form uxa + by2 + cza
+ dw2. Proc. Cumb. Phil. SOC.,19 (1917), 11-21. Also in Collected Papers
(1927), 169-1 78.
6. L. J. Mordell. The representation of a Gaussian integer as a sum of two squares.
Maths. Magazine, 40 (1967), 209.
CHAPTER 21
Representation of Numbers by Polynomials
1. Let f ( x ) be a polynomial in x which assumes only integer values when x

is an integer. Such are
x ( x - 1) x ( x - l)(x - 2)
....
2! 3!
9
Interesting problems arise about the existence of an integer s independent

of n such that every integer n can be expressed with integers x in the forms
I. n =f(xd + f(x2) + . . . + f(xs),
11. n = f ( x l )+ f(x,) + . . . + f(x,), x1 3 0, . . ., x, 3 0,
111. n = f ( x d k f ( x 2 >k . . . f f ( x J
with any combination of signs.
In 11, when f ( x ) = x', we have Waring's problem. We will not discuss this
and some other aspects of these questions since they involve considerations
rather remote from those dealt with here. Some discussion of 111 when
f ( x ) = x' is more relevant to our methods.
Theorem 1
Every integer can be expressed as an algebraic sum of three squares, i.e.
n = k x f ? xi: f xg,
is solvable for some combination of signs.
The proof is obvious from
2n + 1 = (n + - n2, 2n = n2 - (n - 1)' + 1,
and so s < 3. Since 6 = x2 ? y2 is impossible, s 2 3, and so the best possible
value of s is 3.
Theorem 2
Et'ery integer can be expressed as a sum o f j v e integer cubes.
The result is obvious with four cubes for integers G O (mod 6), since
( x + 1)3 + (x - 1)3 - 2x3 = 6x.
The theorem holds for all integers n as follows on putting for x , the integer
(n - n3)/6.
I+
2. The question arises whether fewer cubes are sufficient.

We have the still unproved
Conjecture
Every integer n can be expressed as a sum of the cubes of four integers.
This has been proved for all integers n $ 2 4 (mod 9).
Some results due to Richmond l, and Mordel12 are given by
Theorem 3
AN integers = 3 (mod 6), a 1, & 7, 5 8 (mod 18), can be expressed as a sum
of four integer cubes.
We can easily verify that
k3 + ( - k+ 4)3 + (2k - 5)3 + ( - 2 k + 4)3 = 6k + 3,
(3k + 30)3 + ( - 3 k - 26)3 + (-2k - 23)3 + (2k + 14)3 = 18k + 1 ,
( k + 2)3 + (6k - 1)3 + (8k - 2)3 + ( - 9 k + 2)3 = 18k + 7 ,
(k - 5)3 + ( - k + 14)3 + (3k - 30)3 + (-3k + 29)3 = 18k + 8.
These results are found by putting
x=ak+p, y=bk+q, z=ck+r, w=dk+s
in x3 + y 3 + z3 + w3 = n.
Take 2 a3 = 0, 2 a"p 0.
=
Then n = 3 2 ap2k + 2 p 3 .
A special instance is given by (a, b, c, d ) = ( 1 , 6, 8, -9), and then

p + 36q + 64r + 81s = 0.
An obvious solution is ( p , q, r , s) = (2, - I , -2, 2), giving 2up2 = 6,
x p 3 = 7 and so the formula for 18n + 7.
It can be shown from Theorem 3, Chapter 8, that an infinity of integer
solutions exist.
It has been shown very recently by Demjanenko3 that integers of the form
18k & 2 can be expressed in an infinity of ways as the sum of four cubes of
either linear or quadratic polynomials in k. The proof is rather complicated
since the coefficients are large numbers and depend upon the solution of the
Pellian equation x2 - 3420y2 = 1.
There remains now only the question whether integers n of the form
9k & 4 can be expressed as the sum of four cubes, He has verified that this is
so for n < 1000.
21. REPRESENTATION OF NUMBERS BY POLYNOMIALS 183
Representations by linear polynomials do not exist when n = & 4 (mod 9),
i.e. n = 18k k 4, 18k k 5. For if they did, 2 up2 = 6 and so
2p3= +4(mod9), i . e . p - q = r = s = +l(mod3),
say, p = 3p, f 1, etc. Then
2 a2(3p1 f 1) 0, 2 a($, k 1)' = 0 (mod 3),

=
and so 2 u2 = 0 (mod 3), 2 a = 0 (mod 3).

Hence only one or all four of a, 6, c, d are = O (mod 3).
For the first, we can take
a = b = c = k I, d = O(mod 3).
But then 0 = u3 + b3 + 2 + d 3 E f 3 (mod 9).
For the second, we take CI = 301, etc. and then
20: = 0, z a i ( 3 p 1 k 1)' = 2.
These give the contradiction
2 a, = 0, 2 a, = 2 (mod 3).
It has just been shown by Schinze14that there does not exist a representation
of 9k ? 4 as a sum of cubes of four quartic polynomials.*
3. Theorem 4
I f r is a giiwn integer > 1, there exists a tialue of s independent of the integer n
such that
+xi & S; . . . & X: = n
for integers x.
This result is an immediate consequence of the identity
(.Y + r - I)' - (r I ) (x + r - 2)' + ( r 2 ') (x +r - 3)' + -..

+ (-1),-lxr = r!x + cr,
where c, is an integer independent of x. In fact 5 xr above can be replaced
by an integer valued polynomial
f(x) = a,x' + u,xr-l + . . . + a,,
with a, = a / r ! ,say.
Write Af(x) =f ( s + 1) - f ( x ) ,
A2f(x) = A(Af(x)) = f(x + 2) - 2f(x + 1) + f ( s ) , etc.
* See Addendum.
Then Af(x) = aorx'-' + a * . ,
A2f(x) = aor(r - l)x'-' + * . .,

Ar-'f(x) = a,r!x + cr, say.
This proves the identity above since its left-hand side is Ar-lxr.The theorem
follows at once since any integer n can be expressed in the form
n = r!x + c, + 1' + 1' + - . . + l',
with integer x and at most r ! - 1 of the 1'.
We consider the particular case r = 4. Then we have the
Theorem 5
I f s is the best possible value for all n, then
9 Q s Q 10.
We show that s 2 8 when n = 8 (mod 16). Since x4 = 0, 1 (mod 16),
clearly n cannot be expressed as an algebraic sum of less than eight fourth
powers, and for eight powers only when these are all odd and have the same
sign.
Then s 2 9 since z,"=l
xt = 24 is impossible.
The estimate s Q 12 follows from
(X + 314 - 3(x + 214 + 3(x + 114 - x4 = 24x + 36.
This shows that eight fourth powers suffice when n = 12 (mod 24). Four more
will suffice for all n since the congruence
f a 4 f b4 f c4 k d4 = n (mod 24),
is solvable for a, b, c, d. Since a4 = 0, 1 (mod 3), the congruence is solvable
mod 3. Since a4 = 0, 1 (mod 8), it is also solvable for n = 0, k 1, 2, k 3,
4 (mod 8).
To prove s 6 10, Hunter5 uses identities such as
4 8 ~4 + = +
2 ( 2 ~ 3)4 + (2x + 6)4 + 2(2xa + 8~ + 11)4
- (2xa + 8x + - (2x2 + 8x + 12)4.
The problem of finding the best value of s in Theorem 4 is a very difficult

one which has been solved only for r = 2. Some results are given in Hardy and
Wright but these are out of place here.
REFERENCES
1. H. W. Richmond. An elementary note upon Waring's problem for cubes,
positive and negative. Messenger Math., 51 (1922), 177-186.
21. REPRESENTATION OF NUMBERS BY POLYNOMIALS I85
2. L. J. Mordell. On the four integer cubes problem. J . London. Math. Soc., 11
(1936), 208-218. Addendum. Also in 12 (19371, 80; Corrigendum. Also in
32 (1957), 383.
3. V. A. Demjanenko. On sums of four cubes. (Russian). Zzo. Vysi. Ufebrr.
Zmed. Muternatiku, 5 (54) (1 966), 63-69.
4. A. Schinzel. On the sums of cubes of polynomials. J. Lond. Math. Soc.,
43 (1968), 143-145.
5 . W. Hunter. The representation of numbers by sums of fourth powers. J . Lond.
Moth. SOC.,16 (1941), 177-179.
6. G . H. Hardy and E. M. Wright. An Introduction to the Theory of Numbers.
Clarendon Press, Oxford (1 949, $27-28.
Addendum
Mordell has found a simpler proof of Theorem 4. It may appear as On the
sum of cubes of polynomials in the Acta Arith. for 1970.
CHAPTER 22
Thue's Theorem on the Integer Solutions of f ( x , y) = rn
1. A fundamental result is given by Thue'sl
Theorem 1
The equafion
f(x, y ) = a& + alxn-'y + ... + anyn = m # 0, (1)
where n b 3 , and f(x, y ) is irreducible in the rational field, has only a finite
number of integer solutions.
The cases when f(x, y ) is reducible are easily disposed of. If
f(x, y ) = a(bx + cy)", or a(bx2 + cxy + dy2)ni2, (2)

there may be an infinity of solutions. If
f (x,Y ) = g(x1 y)h(x, Y)l (3)
where the polynomials g(x, y ) , h(x, y ) are relatively prime to each other as
functions of x and y , there can be only a finite number of integer solutions.
For we have
g ( x , y ) = 6, h(x,y) = m P ,
where 6 is any divisor of m. Hence the result since these equations are
independent of each other.
The case f(x, y ) = g(x, y)' is included in the cases (l), (2), (3).
A trivial case of (1) is given by the
Theorem 2
If all the roots of f(z, 1) = 0 are complex numbers, there are only a finite
number of solutions of f(x, y ) = m.
If y = 0, there are at most two values for x .
Suppose then y # 0. For real z , If(z, 1)1 > c > 0, where c is independent
of z. Hence
and so y is bounded.
22. THUE'S THEOREM ON THE INTEGER SOLUTIONS OF f ( x , y ) = m 187
Thue's proof depends on Diophantine approximation to an algebraic
number 0 of degree n 3 3 by means of rational fractions. His result is in-
cluded in Siegel's2
Theorem 3
There are only afinite number of integer solutions for x, y of the inequality
+ +
where v > n/(s 1) s, and s is any one of 1, 2, . . . , n - 1.
O n t a k i n g v = P + E , / 3 = m i n S = , , . . . ,.-l(n/(s+ l ) + s ) , a n d e , O < ~ d1,
such that v is an integer,
n n
[di] + 1 + [ d i ] < l/n + di = 2di.
Thue proves the result for v > n/2 1. +
It has been shown by Roth3 that this theorem holds if v > 2 + e where
E > 0 is arbitrary. The 2 here is the best possible constant.
A particular instance of (4) is that the inequality
where A > 0 is any given constant, has only a finite number of integer solu-
tions.
Take s = 1, Y = 4 2 + +
1 ). Then for all large I yl,
Further if A is sufficiently small, the inequality (5) has no integer solutions

in x, y. The result is trivial if 0 = 5, +
iC2 is complex. For then 1C21 < A / [yln,
and so there are no solutions if A <
Suppose next that 0 is real. We may suppose y > 0, and that ( 5 ) holds. Let
the roots of f(z, 1) = 0 in equation (1) be O1 = 8, B,, . . ., On. Then for
i = 2 , . . ., n,
Iei 51
--
A
I I; +
e -- lei - el,
d -
Y"
+ 2A1, A l = max pi\,i = I, 2 , . . ., n
6 A + 2 A 1 since lyl >, 1.
Hence
or
But for integers (x, y ) # (0,0), If(x, y)l 2 1, and so

A(A + 2A1Y > l / ~ a o ~ .
This inequality, however, requires A b A, = A2(A1,a,,), and so if A < A z ,
there are no solutions of (5).
Assuming ( 5 ) for the present, we can prove Theorem 1. At most two solu-
tions arise when y = 0, and so we can suppose y > 0. Then
lao nn
1=1
(x - k~)l
= Iml,
and so for at least one i, say i = 1, and 8, = 0, say,

1x - Oyl G Im/a,I"" = B say.
Then for i # 1, and large Iyl,
IX - e,yl = IX - ey + (8 - e,)yl 2 I le - e,ly - IX - eyl I
2 lBzy - BII, B2 = min 10 - Oil, i = 2 , . . ., n.
> 92lY1, f o r y > 2B1IBz.
Hence lao(x - OY)(BzY/2)"-'I < Iml,
that is,
Then from ( 5 ) only a finite number of values of x, y are possible.

We can deduce in a similar manner, the more general
Theorem 4
Let P ( x , y ) be an irreducible binary form of degree n > 2, and let Q(x, y ) be
a polynomial of degree m < n - 2 d n . Then the equation
P ( x , Y ) = Q(x, Y ) , (6)
has only aJinite number of integer solutions.
Suppose that this is not so. By a linear transformation if need be, we may
suppose that a,, the coefficient of the highest power xn in P ( x , y ) is not zero,
and so then
P(X,Y) = a0 n ( x - w,
n
i=l
say. Hence for one of the O,, say 0, = 0, x - Oy is very small for an infinity
of integers x, y , corresponding to the solutions, and so
n
22. THUE'S THEOREM ON THE INTEGER SOLUTIONS OF f(S,J.') = ??7 189
where k,,k , are constants independent of .Y and y .
Then
and
This contradicts Siegel's theorem.

If the more precise result of Roth is used, the theorem will hold form < n - 2.
Schinzel has proved in the addendum to Chapter 28 that it suffices if m < n.
2. We now prove Siegel's Theorem 2. We require a number of preliminary
lemmas.
Let fi(x),f2(x),. . . ,fm(x) be rn polynomials with integer coefficients. They
are called linearly independent or dependent according as a relation
2
m
A = l
c,+~A(x> = o (7)
with rational constants c implies c1 = c, = ... = cm = 0 or not. Denote by
D the determinant (Wronskian)
D = Dcf1,f2, . . .,fm) = IIfik'(x)II, A = 1 , 2 , . . ., m ; k = 0, 1 , . . ., m - 1,
where
=f(x), k = 0.
Lemma 1
f l(x), f 2 ( x ) ,. . . ,f m ( x )are linearly dependent i f and only if D = 0 identically.
For if a relation (7) holds, on differentiating m - 1 times, we have
) = O , k = 0 , 1 , ..., m - 1.
h=l
The elimination of the c's gives D = 0.

Suppose next that D = 0. We may assume fl f 0 (i.e. not identically zero),
as otherwise the lemma is true. We note, as is easily proved, that if g(x) is a
function of s differentiable m - 1 times, then
O(gf1, gf2, . . g f m )
' 9 = grnD(f1,f2, . * .,fin).
Thus the second row on the left is
This becomes on subtracting (dg/dx)/gtimes the first row,

We prove the lemma by induction, assuming it holds for m - 1, and then

showing that it holds for m. Suppose that x lies in an interval I in whichfl(x)
does not vanish.
Take g = l/fl, then
Since the first column has 1 in the first row and zeros in the other rows, we have
Hence
are linearly dependent, and so for rational constants c2, . . . , c,,
Then c2f2 + . . . + C m f m = --c f

1 1
where cl is a constant obviously rational. Since this relation holds for x in an

interval I, and the f ' s are polynomials, it must hold for all values of x.
Lemma 2
Let y 1 = ailxl + at2x2+ . .. + ai,x,, i = 1 , 2 , . . ., m
be m linear forms in n > m variables with integer coefficients, with absolute
values all 6 A , say. Then the equations
y l = O , ( i = 1,2 ,..., m),
have integer solutions for the x's, not all zero, and with absolute ualues satisfying
1x1 < 1 + (nA)m'cn-m). (8)
Let the x take the values 0, f 1, k 2,. . ., f H , where H will be defined
presently, and so the y's take (2H + I)" sets of integer values. Since IyI <
+
nAH, there are at most (2nAH 1)" different sets of values for they. Hence if
(2nAH + 1)"' < (2H + l)",
the same set of y values will arise from two different x sets. By subtraction, we
obtain an x set not all zero and with 1x1 < 2H for which all they are zero. The
H inequality will be satisfied if
(2H + 1)"
2 (2nAH + I Z A ) ~ ,
or (2H + l)"-"' 2 (nA)",
2H + 1 >, (nA)m'(n-m).
22. THUE'S THEOREM ON THE INTEGER SOLUTIONS ) m
o ~ f ( x , y= 191
Then we can take H to be an integer such that
(nA)m'(n-m)- 1 < 2H < (nA)m'(n-m)+ 1,
since the interval for 2H is of width 2.
This lemma is a corollary to Theorem 3 in Chapter 5.
Lemma 3
Let 0 be an algebraic integer of degree n 2 3, and a, b, r non-negative integers
such that
(a + l)(b + 1) > rn.
Then there exist polynomials not identically zero,
where the c,,~ are rational integers, such that

~ ( xe), = (x - eys(X),
and S(x) is a polynomial in x. Further, if
RA(B, 8) = 0 for A = 0, 1,. . ., r - 1.

The vanishing for x = 6 of a polynomial R ( x ) with rational coefficients
requires n conditions given by expressing the powers of 6' in R(0) in terms of
1, 0 , . . ., e n - 1 , and then equating to zero the coefficients of these powers.
These conditions are equivalent to the set
R(0,) = 0, R(6,) = 0 , . . ., R(0,) = 0,
where 8, = 8, O,, . . ., On are the conjugates of 8, and so since the 0 are
different, to the set,
8:R(O,) + 8t,R(O,) + .+. + 8hR(On) = 0, t = 0, 1,..., n - 1.
These relations give linear conditions with rational coefficients for the co-
efficients of R(0). Hence each condition RA(8,0) = 0 imposes n such con-
ditions on the c,,~, and so there are rn conditions for the (a 1)(b +1) +
variables c,,~. These can be satisfied by integer values of the c , , ~not all zero
if (a + I)(b + 1) > rn.
It is necessary to find an estimate for the c,,~.
This is given by
Lemma 4
Let b he ajiseil integer with 0 6 b < n, 6 a constant with 0 < 6 < 1, and
where the square brackets denote the integer part. Then there exists in Lemma
3, an integer c = c(8, a), and a polynomial R(x, y ) in equation (9) such that
Ica,El < c, 0< a < a, 0 < fl < b. (13)
We note that
(U + l)(b + 1) > (i- 1

;) r(b + 1 ) > nr.
The conditions on the coefficients c,,~ are given by
8\&(8,, 8,) + e;R,(Bz, 0,) + +
. . . e;R,(e,, en) = 0,
t = 0 , 1 , ..., n - l , X = O , l , ..., r - 1.
Since
the coefficient of c,,E in R,(B, 13)has an absolute value

< (1 4- 1 ) amax(l, 8a+b)
< c;, c1 = cl(e, 6).
Hence the coefficient of c,,~ in the conditions RA(8,8) = 0, has an upper
bound given by
c;(pllt + ... + pnlt) < cj, c3 = c3(4 61,
where c2 = max c1 for 8 = B,, Oa, . . .. Then Lemma 2 gives the estimate
IC.,~ <1 + ((a + l)(b + l ) ~ j ) n ( ( ~ + ) ( ~ + l ) - ~ ~ ) ,
Also (a + l)(b + 1 ) - rn > 6r,
and (a + l)(b + 1)cj <

(:-r + 11(6 + 1)cj
< (n + l)(r + 1)cj < ,:c c4 = ~ ~ (6).8 ,
Finally
\ c , , ~ J< I + (ci~ld < c:, c5 = c5(e, 6).
Corollary
If
22. THUE'S THEOREM ON THE INTEGER SOLUTIONS OFf'(S, y) = nz 193
We now have an estimate for RA(x, y), when x , y are near 8, given by
Lemma 5
Writex - 0 = u , y - 0 = 11, where IuI i1, IvI < 1. Then
IRA(x,Y)I 6 c;(IuI~-~ + Ivl), h = 0,1,. . ., r - 1, c7 = ~ ~ ( 6).
0 ,
a=O B = O
say, where,
Since ea,o = 0 for 0 6 a < r - 1 - A,

RA(x,y) = u r - A zA
a=r-A
e,,oua-r+A v + 2 2 e,,Buuvn-l.
a-A b
a=O 8 = 1
From this, since IuI 6 1, )uI < 1,
d l ~ l ' - ~ ( ( a- r + 1)c: + lul(a + 1)(b + I)c&

G C;(IUI~-A + c7 = 4 8 , e).
Let now pl/ql, p2/q2be two rational fractions with
(PI, 41) = 1, 1,
(pz, 42) = 41 > 0,9 2 > 0.
We shall show that p l / q l , p2/q2cannot both be good approximations to 8.
Put x = pl/ql, y = p2/q2in the inequality for RA(x, y).
The left-hand side is an integer. If it is not zero, and the proof of this which
now follows is the most difficult part of the investigation, then the right-hand
side 2 1. Then one of the two terms 23 and so
This proves the statement.

Write now
2Ym(x>ym= 2 g,(y)xl,
b a
R(X,Y) = (16)
m=o I=O
wherefm(x), g l ( y ) are polynomials in x , y respectively, and neither all the

fm(x)nor all the gl(y) are identically zero.
The polynomialsfJx), ( m = 0, 1, 2, . . . , b) are not necessarily independent.
Suppose that b + 1 (0 < b 6 b) o f them, say
Fo(x), F,W, . . ., M X ) ,
are linearly independent. Then
A = [IFjp(x)II # 0, CI = 0, 1,. . ., b; fl = 0, 1,. . ., b.
Write
where the coefficients in F,(x) are rational integers, and those in G,(y) are
rational numbers. The G,(y) are also linearly independent since if they were
not, then R(x, y ) could be expressed with less than b of the F(x).
Lemma 6
$0 if q2 > cTlo. (17)
For otherwise
5 g, (t)= 0.
l=O
XI
Then for some I for which g,(x) is not identically zero,
say.
On multiplying out by q;, we see that q2 I c ~ , where
~ fl is the greatest
number for which e l , , # 0. This is impossible since I C , , ~ ~6 cT by Lemma 4.
Lemma 7
Let k be any given rational number. If there exists an integer y such that
Ay)(h)# 0, and y + b < r, then there exists an R,(h, p2/q2)wirh 0 < X < r
such that R,(h,p2/q2) # 0.
Since
22. THUES THEOREM ON THE INTEGER SOLUTIONS OFf(X,Y) = m 195
there exists a ,f3 such that G,(pz/q,) # 0, or by a slight change of notation, say
Go(Pzlq2) # 0.
By repeated differentiation for x,we have, say,
2 Flp)(x)G, (t),a
b
a ! R , Y-():, = = 0, 1,. . ., b.
B=O
Here the b + 1 linear equations in G,(p,/q,) can be solved on multiplying the

equations by the usual cofactors, since the determinant A(x) formed from
F p ) ( x )is not identically zero.
Hence
where the H,(x) are polynomials with integer coefficients.

Differentiate y times and put x = h. Then
AY(h)Go (t) =
h=O
h,R, (h,:),
where the h, are rational numbers. Since 6 y < b y < r, the Rh are + +
defined, and clearly not all the R,(h,p,/q,) are zero for 0 < h 6 b y. +
We now find y from
Lemma 8
Suppose r > n. Then there exists an integer y = y(0, b, 6, r, h) with
06 y 6 6r + n2 - n, A(Y)(h)# 0. (18)
b
From a!R,(x,y) = FF)(x)G,(y), ( a = 0, 1,. . ., b),
B=O
2 FFyx)cB(e).
b
a ! ~ , ( ~ ~=, e)
8=0
The polynomial R,(x, 0) in x is divisible by (x - e)r-@,and so by (x - O)r-b

since a < b < r. On multiplying the last equations by the appropriate minors
ofA(x) and adding, A(x)G,(e) is also divisible by (x - O)r-b. Now Go(y) $ 0,
and Go(@ # 0 since G0(Q is of degree b < n. Hence A(x) is divisible by
(.Y - B)r-b, and so if
f(x) = xn + ulxn-l + . . . + a, = 0,
where the a are rational numbers, is the irreducible equation satisfied by 0,
A(x) = f r - b ( ~ ) D ( ~ ) ,
where D(x) is a polynomial with rational coefficients and of degree d, say.

Hence since the elements of A(x) are polynomials of degree a at most,
+ n(r - b) < a(b + I),
d
d < - r(b + 1) - nr + nb < Sr + nb

b + l
+
< 6r + n2 - n.
Since h is rational,f(h) # 0, and so A(x) has a zero at x = h of order y at
most equal to d.
Lemma 9
Zf (pl, q l ) = 1, ( p 2 ,q2) = 1, q1 > 0, q2 > c, r 2 2n2, 6 < 4, there exists
cg = c,@, e), and an integer h = h(0, 6 , 6, r, pl, ql, p a , q2) with 0 < <
6r + n2, i.e. h < r, such that
The lemma is trivial if either 10 - pl/qll > 1 or 18 - pa/q21 > 1 with

cg = 1, and so we may assume that
From Lemma 7 with h = p l / q l , we can now find an R,(x, y ) such that

&(Pr/qa, P a / d # 0.We take h = pJq1 in Lemma 8, and so Y = Y(& by& r,pl,ql).
Then O < y < Sr+n2-n,
h < b + y < 6r + na - n + 6 ,
r
h<-
2
+ -2r = r.
We have now disposed of the preliminary preparations and are in a
position to prove Siegels theorem.
Suppose first that 0 is an integer. We need only consider solutions with
( x , y ) = 1. For if ( x , y ) = d,, x = dlX, y = d l Y , (A, Y ) = 1, and the
inequality (4) becomes
This has only a finite number of solutions from the case dl = 1. Also in
equation (l), dl can have only a finite number of values.
Let us suppose there are an infinity of solutions of the inequality (4)
22. THUE'S THEOREM ON THE INTEGER SOLUTIONS O F ~ ( X J ) = m 197
where we may suppose that y > 0. We select two solutions (pl, ql), ( p 2 , q2),
q2 > q l , and define the integer r by
We can take q1 > c, and q2 > c' and also so great that r >, 2n2. We shall
impose conditions on the choice of 6 and q1 which will contradict the assertion
in Lemma 9, i.e. we shall make
Write v = n/(b + 1) + b 4- E, where we can take 0 < E < 1, and so

n
v <-
2
+n - 1 + 1 < 2n.
Now ~ ~ <, q i 7rt b ( r t l ) - u ( r - - h ) >
< q l bn+ tl dr t br t b - vr t v h
,
d b vhr
< (qlbTi"+T+r)
We can take 6 so small that 6/(b + 1) + 6v < $ E , and r so great that b/r +
n2v/r < +E. Then A < ( ~ ~ q ; " /and
~)~we, can suppose q1 so great that A < 1.
Next
qlbe+d r t r(b - v )
, since b < v,
+
Hence T ( ~ ~ q ~< -I ~ ~ ~ ) r
B c ( C ~ ~ ; & / ~ )<
from the choice of q1 for A.
This proves Siegel's theorem when 6 is an algebraic integer.
Suppose next that 0 is not an integer. Write 0 = $ / j , where $ is an algebraic

integer and j is a rational integer. The inequality (4) becomes
Then _
j -- j <- 1
yv Yktb+* Yi+ltbti
for large y. Hence there can be only a finite number of values f o r j x and y .
3. The Thue-Siege1 approximation result can be applied to equations other
than (1). A corollary to Theorem 4 is given by
Theorem 5
Let F(x, y ) be a homogeneous irreducible polynomial of degree n and G(x, y )
a polynomial of degree m. Then the equation
F(X,Y) = C(X,Y)
has only a j n i t e number of integer solutions for x, y i f m < n/2 - 1.
The result is easily proved if all the linear factors of F ( x , y ) are complex.
Suppose next that there are an infinity of solutions when there are real
linear factors of F(x, y). Then for large 1x1, there exists a real linear factor
y - a x such that F(x, y ) / ( y - ax)xn-l --f c1 where c1 is a constant. Hence
say,
lY - ax1 Ix"-'l < CzlG(X,Y)l,
and so
for arbitrary E > 0. Hence the result.

If Roth's result is used, we can take m < n - 2.
Other applications require an extension of the Thue-Siege1 approximation
theorem to approximation by algebraic integers. SiegeP has proved the
Theorem 6
Let K be an algebraic numberjeld of degree n and a be the root of an equation
of degree d > 2 irreducible in K . Let 5 # 0 be an algebraic number in K and let
H(5) be the maximum tlalue of the absolute values of the coeficients of the
irreducible equation satisfied by 5. Then the inequality
I[ - trl ,< l/(H([))d""fl)+S+E, O < s < d
has only ajinite number of solutions in integers E of degree n.
22. THUES THEOREM ON THE INTEGER SOLUTIONS OFf(X,y) = I77 199
The proof of the existence of only a finite number of integer solutions of
f(X,Y) = m > 0, (21)

wheref(x, y ) is a homogeneous polynomial of degree n irreducible over the
rationals, can give estimates for the number of solutions. This, however, does
not help in finding the solutions since the method is non-constructive.
Recently, Baker4 has shown that bounds can be given for the magnitude of
the solutions. H e has proved the
Theorem 7
Let 01 > n + I . Then the solutions of (21) satisfy the inequality
max (IxI, I y l ) < ~ e ( ~ ~ ~ ~ ) ~ ,
where c is an effectiuely computable constant depending only on n, CY and the
coeficients of f ( x , y ) .
Previously he had proved some special results, for example, when
f(x, y ) = .x3 - 2y3, then 1x1 < M , I yI < M where M = (3. 105m)23.
Bakers results also imply that bounds can be found for the solution of
some other equations whose solution can be made to depend upon equations
such as (21), for example,
y2 = ax3 + bx2 + c s + d,
as this can be reduced t o the question of representing unity by a finite number
of binary quartics as is shown in Chapter 27.
REFERENCES
1 . A. Thue. uber Annaherungswerte algebraischer Zahlen. J. reine nngew. Math.,
135 (1909), 284-305.
2. C . L. Siegel. Approximation algebraischer Zahlen. Math. Z., 10 (1921), 173-
213. Also Gesammelte Abhandlungen, 1 (1966), 6-46.
3. K . F. Roth. Rational approximations to algebraic numbers. Mathematika,
2 (1959, 1-20. Also Corrigendum. Malhernatika, 2 (1955), 168.
4. A. Baker. On the representation of integers by binary forms. Phil. Trans. R.
SOC.,263 (1968), 173-191.
5. A. Baker. Rational approximations to v2 and other algebraic numbers.
Q.J. Math., 15 (1964), 375-383.
CHAPTER 23
Local Methods or p-Adic Applications
1. One of the most important applications ofp-adic numbers to Diophantine

equations deals with the
Problem
Let wl, w 2 , . . ., w , be a basis of the integers in an algebraic number field
K = Q(0). Denote by N(w) the norm of w in the field. Then it is required to
discuss the solution in rational integers x of the equation
N(xlwl + x2w2 + . . . + x,w,) = a, (1)
where the x satisfy the n - m equation
g,(a) = 0, (1 = 1, 2 , . . ., n - m) (2)
for given m < n, and the g ( x ) are polynomials in x with rational coeflcients.
The simplest problem is when x,+ - -
= x , + ~ = . = x, = 0.
From algebraic number theory, it is known that the general solution of
equation (1) is given by
x1 1 + xz w(*)
a + . + X,W~) = d S P , (s = 1, 2, . . ., n), (3)
where denotes the conjugates of w , the c belong to a finite set of numbers
in K and E is a unit in K . We recall that every unit E can be written in the form
E = [rl:17):a. . .+, (4)

where [ is a root of unity in K, the 7 are a set of fundamental units, and the u
+
are arbitrary rational integers. Also r = rl r2 - 1 where rl, r2 denote
respectively the number of real fields and pairs of complex fields among the
conjugates of K . On substituting for the a from equations (3) and (4) in (2),
we have n - m equations in the r unknowns u which occur as exponents in
the powers of the units 7. If r < n - m, we should expect there to be only a
finite number of solutions if the equations resulting from (2) are independent.
As these equations are best dealt with by p-adic methods, we give a brief
rCsumt of p-adic number theory where p is a rational prime. For simplicity
we define the p-adic numbers over the rational field Q.Let x be any number
in Q. We define a valuation 1x1, as follows. If x = 0, 1x1, = 0. If x # 0, x
23. LOCAL METHODS OR ~ - A D I CAPPLICATIONS 201
can be written in the form x = pax1/x2, where xl,x2 are prime to p . Then
1x1, = l/p". The valuation has the properties
1. 1x1, 2 0, and 1x1, = 0 if and only if x = 0,
2. IxlxZIP = IXllPlxzlP,
3. 1x1 k XZlP G IXllP + IXZlP.
This is the triangular inequality and may be sharpened to

1x1 k XZlP G max ( b l I P 3 1x21,).
The p-adic field Q , is defined as a field which contains the field Q and is
such that if X E Q,, then
I. There is a valuation (A'(, on Q , which satisfies 1, 2 and 3 and coincides
with 1x1, when X = x is an element of Q .
11. Limits are defined as for complex number theory, except that an absolute
value 1x1 there is replaced by [XI,
and corresponding results exist for the
p-adic field. Thus Cauchy's convergence principle holds, i.e. a sequence
XI, A', . . ., tends to a limit if and only if IX, - Xnlp< E for m,n >
N(E).This limit will be in Qp and so the field Q , is complete.
111. Every element of Qp is a limit of a sequence of numbers of Q.
The discussion of convergence is now very simple since a series 2a, con-
verges if and only if Ja,J ,-+ 0, and so there is no need for the concept of
absolute convergence. The usual results on functions of a complex variable
carry over to p-adic variables. There is, however, greater simplicity since if
lan\,--f 0, the series
m
converges for 1x1, G 1.

We can define series which have the characteristic properties of the ex-
ponential and logarithmic functions.
Thus we have
1
if 1x1, < -
P-1
log(1 + x) = s - -
X2
2
+ " . + (-1)n-lx"
n
+ ... if 1x1, < 1
Similar results hold when x is a number in an algebraic number field K.

If 7 is a unit in K, there exists a rational integer a such that if p is an odd
prime, then 7' = 1 (mod p ) , but if p = 2, va = 1 (mod 4). Then = eulog Ila
can be expanded as a power series in v with coefficients in K which converges
for allp-adic integers u. Since u = au + b, 0 < b < a, we have a expansions
for qu. It follows at once that the product q;q;a, . . can be expressed as a
finite number of power series in v l , v 2 , . . .with coefficients in K which
converge for all p-adic integers v l , u2, . . ..
2. We now show by a method due to Skolem howp-adic theory can be applied
to the equations arising from (2). It is obvious that equations will have only a
finite number of rational solutions if they have only a finite number of p-adic
solutions. The equations (2) can be replaced by a finite number of equations
g,(ul, u 2 , . . ., u,) = 0, (1 = 1 , 2 , . . ., n - m), (5)
where the g(u) are power series converging for all p-adic integers u. The
coefficients may be taken as rational numbers whose denominators are prime
to p since we are considering a set of conjugate equations.
The finiteness of the number of p-adic solutions can be proved from the
following theorems.
Theorem 1
Let f ( x ) = 2; anxn where lanlp--f 0 so that the series converges when
1x1, < 1. Then the equation f(x) = 0 has only afinite number of solutions for
1x1, < 1, i.e. for p-adic integers x.
More generally we have a result given by Strassmann
Theorem 2
Let fo(x),fi(x), . . . be polynomials whose coeficients are p-adic integers, and
suppose that fo(x)has at least one coeficient & 0 (mod p). Then the equation
(6)
has only afinite number of solutions in p-adic integers.

There are extensions to several variables.
Theorem 3
Let fo(x,y ) , f i ( x ,y), . . ., g,(x, y), g,(x, y ) , . . . be polynomials whose co-
efficients are p-adic integers and suppose that
Then the simultaneous equations
have only a finite number of p-adic solutions.

This also holds if fo, go are relatively prime mod p .
23. LOCAL METHODS OR P-ADIC APPLICATIONS 203
Skolem has proved many results by his methods. When the question is only
to prove the existence of a finite number of integer solutions, the work may
not be too complicated as can be seen from proof of
Theorem 4
Let a,, a2, . . ., a,,, b,, b2, . . ., b, be numbers in an algebraic number field K.
Then the equation
2 alb:
1=1
= 0
has only a finite number of rational integer solutions for x, if none of the
quotients bJb, is a root of unity.
When the numerical values of the fundamental units of the relevant
algebraic number field play no part in an investigation, and only an estimate
for the number of solutions is required, the details may not be troublesome
as in Skolem's proof of Theorem 5
Theorem 5
I f d is an integer > 1, there is at most one rational integer solution of x3 +
dy3 = 1 other than x = 1, y = 0.
Let us suppose there are two solutions (xl,y l ) and (x2,y2).Write 0 = $2.
+ +
Then v1 = x1 y,B, q 2 = x2 y20 are positive units with norm 1 in the
field Q(0). Since there are no roots of unity in the field and there is only one
fundamental unit, we have
7;1 = 7 2 (9)
with rational integers, u l , u2. We may suppose that u,, u2 are not both
EO (mod 3), for if so, we would have + 1 3 = 7 ; 2 1 3 . Hence we may suppose
that u2 f 0 (mod 3). Then u,/u2 is a 3-adic integer and so we have, say, the
3-adic equation
7; = 7 2 -
Suppose first that y , = 0 (mod 3) and so x1 f 0 (mod 3).

Then 7; is defined in the 3-adic field over Q(0) and so
and the left-hand side can be expanded as a binomial series. Equating the
coefficient of O2 to zero, we have
(")2 x1
+
(.?i)2
5 x1
(")
(Y')5 d + (")8 (&)*
x1
d2 + . . . = 0.
We show that the only solutions of this equation are 14 = 0, 1. For on

dividing out by
we have
This gives an impossible congruence mod 3, since 4, 5, 7, 8, . . . are 3-adic

units.
Suppose next that y, $ 0 (mod 3). Since
T: = X: + 3x:y1e + 3x1yqe2 + y:e3
= 1 + 3x:y1e + 3x,y:ea = 1 + 36,
say, 7: is defined for u = 0 (mod 3). Write u = 3v + uo, uo = 0, 1, 2.
Then $VfUO r12 (mod 3), $0 = v2 (mod 3).
Comparing coefficients of B2, we see that uo = 0, 1.
Take first uo = 0. Then
(1 + 3s) = x2 + y2e.
Denote by bt the coefficient of O2 in 6. Then
t-0
3bt (y) = 0,
or
Divide out by 3x1y:, and we have, say,
u + 3B2 (3+ (9) + 32B3 s.. = 0,
where the B are polynomials in xl, y, with integer coefficients. This is im-
possible, for if 3* is the highest power of 3 dividing v , all the other terms are
divisible by 3+. For the general term is
and 3t-2/t is a 3-adic integer. This is obvious on putting t = to3 where 3 is

the highest power of 3 dividing t.
Suppose next uo = 1. Then
23. LOCAL METHODS OR p-ADICAPPLICATIONS 205
Denote by ct the coefficient of 6 in S'. Then
s1
t=O
2 (y) +
3'bt y, 2 3'ct (y)
m
t=o
= 0.
Dividing out by 3s: y:, we have, say,
and this is impossible as before.

3. When it is required to find all the integer solutions of an equation, the
process may not be very complicated if fundamental units are not involved as
in the following theorem. This deals with a conjecture enunciated by Rama-
nujan and first proved by Nagell. Many other proofs have been given and
these have been analysed and discussed by Hasse. He has given a simpler
version of Nagell's4 proof as well as a generalization. We give Hasse's5 proof
of
Theorem 6
+
The equation x2 7 = 2" has only the positive integer solutions given by
.Y = 1, 3, 5, 11, 181 corresponding to n = 3 , 4 , 5 , 7 , 15.
When n is even, n = 4 is the only solution since
(2"/2)2 - x2 = 7, 2"'2 f Y = 7, 2"/2 T x = 1.
We may now suppose that n is odd, and we write the equation as
x2
- + 7 - 2y,
-
4
where y is odd and y 2 3.
We factorize the equation in the field Q ( d r 7 ) , in which the integers have
+
the form (m n d 7 ) / 2 where m = n (mod 2), and in which unique
factorization holds. Since
we have
x + 24-7 = * (1 * 24-7 ) ,
We show that the positive sign is impossible in equation (1 1).

Write this as
ay - by = a - b.
Then a2 = (1 - b)2 = 1 (mod b2),
since ab = 2, and so
aY a(a2)'Y- 1)/2 -
= a (mod b2),
a = a - b (mod b2),
which is false.
Hence we have
and so = y (mod 7).

-2Y-1 -
This has the odd solutions y = 3, 5, 13 (mod 42). We prove that 3, 5, 13 are
the only remaining solutions for y in equation (10).
It suffices to show that there cannot be two solutions y , y , with y , - y E
0 (mod 42). Suppose that 7' is the greatest power of 7 dividing y , - y . Then
aY1 = ay.ay1-y = aY(+)Yl-Y(I +d q 1 - Y . (13)
Now (+)yi-y = ((+)6)(yi-y)/6 1 (mod 71+l).
Also (1 + 2/-7)y1-y = 1 + ( y , - y)d/-? (mod 7'+l)

as follows on raising 1 + dr7 successively to powers 7 , 72,. . ., 7' and then
to the power (y, - y)/7'.
Since
on substituting in equation (13), we have
Similarly bY1 = by + (Y
1 - Y ) d-7
2y
(mod 7'+').
Since from equation ( I l),

a Y - by = ayi - bYi,
then ( y , - y)d\/-7 = 0 (mod 7'+l),
and so since yl, y are rational
y, - y = O(rn0d7'+~).
This contradiction establishes the theorem.
4. When the numerical values of the fundamental units must be considered,
in general, a great deal of detailed numerical work, and several cases may arise.
23. LOCAL METHODS OR P-ADIC APPLICATIONS 207
It is only exceptionally that little detail is required and then the result may
perhaps be found more simply by classical methods. Thus we have the proof
by Skolem3 of
Theorem 7
The only integer solutions of the equation
x4 - 2y4 = 1
are giwn by y = 0, x = & 1.

Clearly y = 0 (mod 2) and so we may replace the equation by the equation
x4 - 32y4 = 1. (16)
It can be shown that the fundamental units in the field Q(e), where 0 = @,
are given by 1 + 6 and 1 + P. Since x + 2yO is a unit in the field, we have
+ e y ( 1 + 02) = + 2 ~ 4
(1 (17)
u.u - 1
(1 + ue + ___
2!
+ . . . ) (I + + . ..) = s (mod 2),
2 5 ~ 2
and so u = 0 (mod 2).

Since the coefficients of P, O3 vanish in the left-hand side of equation (17),
we deduce the p-adic equations:
(); +u+2 ((z) + (3 + (); t (4) + (9) + 22(. . . ) = 0,
(3+ uu + 2 ((I;)+ (;) 2 + (;) (;) + u (3)) + 22(. . .) = 0.

Multiply the first equation by u and subtract from the second. Then
-+(u + l)u(u - 1) + 2 {(;I (d

- ((;) - (3)
(3)(57)
+
+ ((i)- u + 22(. . .) = 0. (18)
If u # 0, suppose that 2A11 u and so h 3 1. Then
Hence all the terms of equation (14) are = O (mod 2Atl) except the first and
so we must have u = 0.
Then 2 + 2 (;) + 22 (;) + = 0,
and a similar argument shows that u = 0.

In the same way, it can be shown that the only integer solutions of
x4 - 8y4 = I , (19)
are given by y = 0, x = 1. For now we have
(I + e)yi + 82)" = + ye3,
and the result is easily proved.
The most troublesome numerical details arise perhaps when Skolem's
method cannot be applied directly because there are not as many equations
as there are unknowns. It then becomes necessary to introduce relative units
in an extension of the original algebraic number field. An illustration is given
by the equation x3 - 3xy2 - y 3 = 1 noted by Nagell, and mentioned by
Skolem as a possible application of his method. The equation was disposed of
by Ljunggren who proved
Theorem 8
x3 - 3xy2 - y3 = 1
are given by
(&Y) = (1, O), (0,-119 (- 1, 11, (1, -31, (-3,219 (2, 1).
Here we have
N(x + 6y) = 1,
taken in the cubic field Q(0) where
e3 - 3e +I = 0.
This equation is well known as an Abelian equation associated with the 9th
roots of unity. Call the roots 5, v, 5. Then 5, r ) may be taken as the funda-
mental units and so
x + 5y = 5"r)".
This gives
(.I- 5)E"v" + (5 - 0 v " P + ( E - d5"P = 0.
This is the only equation connecting the p-adic variables u, v. On writing
x = 5u- ~ i z u ~+ w
( a, y = 5'" - 21J)/ZvU - u/2,
we find the equation

5x2 - &Y2 = 5.
23. LOCAL METHODS OR p-ADICAPPLICATIONS 209
We now consider the extension field arising from the adjunction of 42 to
the field Q(E). The new field is of degree 6 and there are now four fundamental
units since only one of t , ~5 ,is negative. Then ( X d z + YdE)zis a unit in
Q(d/r> with relative norm 1, and can be written in the form + E : E ~ , where
e l , e2 are two of the fundamental units with the same property, and can be
taken as
Ljunggren finds X , Y in terms of E:, ek and their relative norms. Since

N ( X ) = N ( Y ) = 1, two equations are obtained for the h, k. The results
follow on working in the 2-adic field.
The equation arises from the discussion of the equivalent equations
y 3 = x 2 + x + 1,
ya = x3 - 48.
The theorem shows that the only solutions of the first one are x = 0, - 1,
18, - 19. A simpler method of proof is obviously desirable.
Skolem has given many other applications. He has proved Thue's theorem
in Chapter 22, on the finiteness of the number of integer solutions of a poly-
nomial equation f ( x , y ) = m, except in the case when all the roots off(x, 1 ) = 0
are real. An account of this is given by Borevich and Shafarevich' in their
book on number theory which also contains an excellent account of p-adic
numbers and also of local methods.
5. Skolem has also shown that results can be obtained for diophantine equa-
tions in more than two variables. Thus he has proved
Theorem 9
Let a, 6 , c, be three linearly independent numbers in an algebraic numberfield
K of degreefive whose conjugates include only one realfield. Then the equation
N(ax + by + cz) = n, (21)
where N denotes the norm in K, has only ajinite number of rational integer
solutions f o r x , y , z .
In particular, he proved that the equation
N(x + YVT + z V ~=) 1 ,
i.e. xs + 2 p + 4z6 - +1 0 ~ ~ 13 0 ~ ~ 2 ~I,
=~ 2
has at most six solutions in integers s,y , z , and of these three are known,
namely ( s , y , 2) = (1,0,0), ( - I , I , 0), (1, I , -2).
210 DlOPHANTINE EQUATIONS
Skolem's theorem 9 is included in the generalization due to Chabauty' and

proved by applying results due to Skolem and Siegel, namely,
Theorem 10
Let K be aprimitive algebraic numberfield of degree n, and let r = rl i- r2 - 1
where rl is the number of realjields conjugate to K and r2 is the number ofpairs
of conjugate complex fields. Then if r < n - 3 and a , /3, y are three linearly
independent integers in K, the equation
+ + yz) = A ,
N ( ~ x /3y (22)
where A is a given rational integer has only afinite number of rational integer
solutions for x, y , z.
Recently other results have been found for equations typified by (22).
Thus Nagell' has proved the
Theorem 11
Suppose that K is a biquadratic field having as its only quadratic subfield an
imaginary one. Then if@, /3, y are as in Theorem 10, its result for (22) holds.
Though some of the new results are not proved by p-adic methods, an
account of them may not be out of place here. By combining arithmetic
methods with an application of Thue's theorem, Nagells has proved
Theorem 12
Let p be a primes > 5, and m a rational integer such that m(m - 1) Z 0
(modp). Then if5 is a primative pth root of unity, the equation
N(x + 5y + 5"z) = 1
has only afinite number of rational integer solutions for x, y , z.

Previously Siegel'O had used his results on diophantine approximation,
for example, the approximation of algebraic numbers by other algebraic
numbers. Thus he proved
Theorem 13
Let 0 be an algebraic integer of degree n and let h be a positive integer such
that
n > hZ($ + s), 2s =1- - 1.
Then the equation

+
N ( X ~ x,e + .. . + Xheh) = A,
where A is a given rational integer, has only afinite number of rational integer
solutions for xo, xl,. . ., x h .
23. LOCAL METHODS OR p-ADICAPPLICATIONS 21 1
Then if n < 57, 17 = 1 ; if 57 < n < 307, h = 1 o r 2; if 17 2 307, we can
takc h = 3. I n general, we can take / I = 4h4 - 2h2 1. +
Very recently, new results on diophantine approximation have been given
by Schmidtl who has found results for simultaneous approximation similar
to those by Roth. H e proves the
Theorem 14
Let a, p, be algebraic numbers such that a , p, 1 are linearly independent over
the rationaljeld Q. Then for every E > 0, there are only a j n i t e number of
rational fractions x/z, ylz satisfying the inequalities
Let now K be a n algebraic number field of degree n > 3, and let a, F, y

be three elements of K linearly independent over Q and such that K =
Q ( a / y , ply). Schmidt12 applies theorem 14 to show that if f ( x , y , z) is a poly-
nomial, then subject to certain conditions, the equation
wax +1
. + YZ) = f ( x , Y , z ) ,
Y
has a t most a finite number of rational integer solutions for x, y , z.In partic-
ular, he proves
Theorem 15
+
Suppose that us + By + yz is not of the form AR pS where R is a rational
form and S is a real form whose coeficients are in a quadraticjeld. Then if
f(.u, y , z ) is a polynomial of degree < nl4, the equation
ax + FY + y z ) = f ( x , Y , z),
has only ajinite number of rational integer solutions for x, y , z.
REFERENCES
1. R. Strassmann. uber den Wertevorrat von Potenzreihen im Gebiet der p -
adischen Zahlen. J. reine angew. Math., 159 (1928), 13-28
2. T. Skolem. Diophantische Gleichungen. Springer-Verlag, (1938), pp. 114-
120. See also the references cited in this book.
3. T. Skolem. The use of p-adic method in the theory of diophantine equations.
Bull. SOC.Math. Belg., 7 (1959, 83-95.
+
4. T. Nagell. The diophantine equation x2 7 = 2. Arkiv matematik, 4
(1960), 185-187.
5 . H. Hasse. uber eine diophantische Gleichungen von Ramanujan-Nagell und
ihre Verallgerneinerung. Nag. Math. J., 27 (1966), 77-102.
6. W. Ljunggren. Einige Bernerkungen iiber die Darstellung ganzer Zahlen
durch binare kubische Formen mit positiver Diskriminante. Acfa Math., 75
(1942), 1-21.
7. Z. I. Borevich and I. R. Shafarevich. Number Theory. Academic Press,

New York and London (1966).
8. C. Chabauty. Sur certaines Cquations diophantiques ternaires. Cotnples
rendus Acud. Sci. Paris 202 (1936), 2117-2119.
9. T. Nagell. Remarques sur les formes a plusieurs variables dkcomposables
en facteurs lineaires Ark. mar., 7 (1967), 313-329.
10. C. L. Siegel. Approximation algebraischer Zahlen. Math. Zeit. 10 (1921)
173-213 or Gesammelte Abhandlungen 1 (1966), 6-46.
11. W. M. Schmidt. On simultaneous approximations of two algebraic numbers by
rationals. Acta Math., 119 (1967), 27-50.
12. W. M. Schmidt. Some diophantine equations in three variables with only
finitely many solutions. Mathematika, 14 (1967), 1 1 3-120.
CHAPTER 24
Binary Cubic Forms
1. We now apply the general theory of Chapter 20 to the binary cubic

f ( x , y ) = ax3 + bx2y + cxy2 + dy3, (1)
with integer coefficients and discriminant
D = -27a2d2 + 18abcd + b2c2 - 4ac3 - 4bd3, (2)
where we suppose D # 0. Here D is the invariant
D = a4IIn(/3- Y)~,
and a, 8, y are the roots of the equation
at3 + b t 2 + ~6 + d = 0.
There are quadratic and cubic covariants H ( x , y ) and G(x, y ) given by
H = -- = (bx + CY)~- ( ~ U X + by)(cx + 3dy),
= (b2 - 3ac)x2 + (be - 9 a d ) ~ y+ (c2 - 3bd)y2,

= Ax2 + Bxy + cy2,
say, of discriminant B 2 - 4AC = - 3 0 .
Also G =
aH aH
= - (27a2d - 9abc + 2b3)x3 . . .. + (4)
-
aYlIx
The covariants G, H , and f are not algebraically independent.
Theorem 1
G2 + 27Df2 = 4 H 3 . (5)
By a linear substitution of determinant unity with real or complex co-
efficients, we can write, say,
f = UX' + dy3, D = -27a2d2,
H = -9adxy, G = 27ad(ax3 - dy3).
The identity (5) follows at once.
8+
The following theorem is included in the general theorems, and has been
known for many years, but an independent proof is worth while.
Theorem 2
The number of classes of binary cubics with given discriminant D # 0 isfinite.
It suffices to prove thatf(x, y ) is equivalent to a binary cubic all of whose
coefficients are bounded in terms of D.
We consider two cases according as D > 0 or D < 0. Suppose first that
D > 0 and so the roots a, /I, y are all real. Then H ( x , y ) is a positive definite
+
form since H > 0 unless 3ax + by = 0, and cx 3dy = 0, bx + cy = 0 have
a solution other than x = y = 0. This requires D = 0. We now reduce
H ( x , y ) by means of a unimodular integral substitution, and so we may
assume that
C 2 A 2 IBI, A < dz,AC < D . (6)

We apply the same substitution tof(x,y), and we then call the new form
reduced. We show that its coefficients are bounded in terms of D .
Write H(x,y) = A p e ,
P=x+
B + 2A
4-30 y, Q = x +
B - 1/70
2A y'
Then
where p , q are rational numbers whose denominators are seen from (6) to be
bounded in terms of D .
Hence a, b, c, d are expressed as polynomials in p , q, A , B, D. Equating
discriminants for the two forms of f ( x , y),
D = -27(p2 + 3q2D)2(4Ad/D)8.
Since p , q have bounded denominators, their numerators are also bounded.
Hence a, b, c, d are bounded in terms of D .
Suppose next that D < 0. Let a be the real root and /3, y the conjugate
complex roots of f ( x , 1) = 0.
Suppose first that a = 0. Write
f(x, y ) = y(bx2 + CXJJ + dy2), A = c2 - 4bd.
Then D = b2C2 - 4db3 = b2A,
and so b, A are bounded in terms of D . As we may suppose IcI <b# 0. then
c, d a r e also bounded.
Suppose next that a # 0. Write
f(x,Y ) = k (x - ay)(Px2 + Q x +~ Ry2),
24. BINARY CUBIC FORMS 215
where the sign is taken so that Px2 Qxy + + Ry2 is positive definite. We
reduce this form and so we may suppose that
-1 < p + y < 1, P y 2 1.
We apply the same substitution tof(x, y ) and call the new form reduced. We
show that its coefficients are bounded in terms of D.We show that
IP - YI2 2 3, (a - B)(. - Y) 2 $.
The first follows from
(B - YI2 = (B + YI2 - 4PY < -3,
and the second from
( a - P)(a - y) =
( a - B
- + Y
2 )2-(T)*
B - Y 2
Hence a and so P and Q are bounded in terms of D. Since

(DI 2 3a4 (a T)
-B + Y 4 ,
a is bounded in terms of D.Since
27a2
then ID1 > - P R ,
16
and so R is bounded. This proves that the coefficients off(x, y ) are bounded.
To prove thatf(s, y ) in equation (1) is equivalent to a reduced binary cubic
involves a special case of a general problem in the Geometry of Numbers.
This is to find the minimum value m of a function Ig(x, y)I for integer values
of x, y. For (l), it was shown by Mordell that if D < 0, then ni < - D/23,
and that equality occurs when
f(x,y) N m(x3 - xy2 - y3).
If D > 0, then m < q/o/49and for equality
f(x,y) - m(x3 + xy2 - 2xy2 - y 3 ) .
2. We can use these results in I to find the solution in integers X , Y, Z of
x2 +~ Y =Z2 3 , (7)
where k is a given integer.
This is suggested by the relation

G2(x,Y ) + 27Df 2(x,y ) = 4H3(x,y ) ,
which gives an integer solution in two parameters x , y of the equation
X2 + 27DY2 = 4Z3.
It proves convenient in dealing with equation (7) to write f ( x , y ) with
binomial coefficients and to modify slightly the definition of the invariants and
covariants.
Write now f ( x , y ) = ax3 + 3bx2y + 3cxy2 + dy3,
where a, b, c, dare integers. Use D,, H,, G1 to denote invariants etc. for non-
binomial coefficients, and D, H , G for the invariants for binomial coefficients.
Then D1 = 2 7 0 , Hi = 9H, G1 = 27G.
Here D = -a2d2 + 6abcd + 3b2c2- 4ac3 - 4db3,
H = (b2 - ac)x2 + (bc - ad)xy + (c2 - bd)y2
= AX2 + Bxy + cy2, = ( A , B, C ) , (8)
say, of discriminant B 2 - 4AC = - D,
G = -(a 2d - 3abc + 2b3)x3+ ....

Now G2(X,y ) + Df2(x,Y ) = 4H3(x,Y). (9)
I f D = 4k, C(x, y ) = 0 (mod 2), and so integer values of X , Y, Z in equation
(7) are given by taking
X = 3G(x, Y), Y = f ( x ,Y ) , Z = H ( x , y), (10)
where f ( x , y ) is any binary cubic with binomial integer coefficients of dis-
criminant D = 4k.
We prove now the
Theorem 3
All the integer solutions of
X2 + k Y 2 = Z3, (X,Z)= 1 (7)
are given by the formulae (10) above.
For let ( X , Y , Z ) = (g,J h) be a solution such that
g2 + k f 2 = h3, ( g , h ) = 1. (1 1)
We shall construct a binary cubic f ( x , y ) with integer binomial coefficients
and of discriminant 4k such that g,S, h are values assumed by C(x,y ) ,f ( x , y ) ,
H ( x , y ) for integers x, y. Then all the solutions of equation (7), are given by
24. BJNARY CUBIC FORMS 217
taking for f ( x , y ) a set of non-equivalent binary cubics of discriminant
D = 4k, and letting x, y run through all integer values for which ( X , 2 ) = 1.
Since from equation (1 l), - k is a quadratic residue of h, there exist binary
quadratics of discriminant -4k with first coefficient h, for example, say,
(h, 2B, C ) = hx2 + 2Bxy + Cy2, where B 2 - hC = -k.
We take for B any solution of the congruence f B = - g (mod h3).
We shall now construct a binary cubic (,L b, c, d ) with discriminant D = 4k
and H in ( 8 ) given by (h, 2B, C).
b2 - h
Since It = b2 - f c , c = -3
f
we can take b = g / h (modf) and, in particular,
bh = g + Bf, and so b = 0 (mod h2).
Then
= -kf + 2gB + B'f.

We now find d. Since bc - f d = 2B,
On simplifying,
h3d = - k g - 3kfB + 3gB2 + fB3.
We show now that c and d a r e integers. We have
h2c = - k f + 2g ( 7 )+ 7
-- - (mod h2)
- h f 2 - 8'
- = 0 (mod h2).
- f
Hence c is an integer. Next
h 3 d - -kg 3 2 -
+ 3kg + -
f'
5 (mod h3)
= ( k f 2 + g 2 ) (mod h3)
f 3
= 0 (mod h3).
Then d is an integer. This completes the proof.
There is an immediate application of Theorem 3 to the equation

y2 = x3 + k.
This shows at once that there are only a finite number of integer solutions
if (x, k) = 1, as happens for example when k is square free. We must solve
f ( x , y ) = 1 wheref(x, y ) is a binary cubic with discriminant - 4k. Since there
are only a finite number of classes of forms with discriminant -4k, Thues
theorem in Chapter 22 shows at once that there are only a finite number of
integer solutions for x, y.
The result holds for all k as is shown in Chapter 26.
3. We discuss in more detail the integer solutions of the general cubic
equation
f ( x , y ) = ax3 + bx2y + cxy2 + dy3 = m. (12)
It is known as a particular case of Thues theorem of Chapter 22 that there are
only a finite number of solutions if we suppose that f ( x , y ) is irreducible. No
necessary and sufficient condition for the existence of integer solutions of
equation (12) nor a finite algorithm for finding them when they exist are
known. However, Baker3 has recently found an upper bound (a very large
one indeed) for the magnitude of the solutions.
Some results are known about the number of solutions and some special
equations of the form ax3 + dy3 = m can be completely solved. We have
seen that it suffices in (12), to consider the case when m = 1. Then if one
solution is known, we may by means of a linear substitution suppose that
a = 1.
Let 8, 8, 0 be the roots of the cubic equation
t 3 - bt2 + ct - d = 0, (13)
and let K = Q(8) be the cubic field over the rational field generated by 8. Two
cases arise according as the discriminant D < 0 or >O.
Suppose first D < 0. Then combining results of Delaunay and Nagell 5,
we have the
Theorem 4
--
The equation (12) for m = 1 has at most three integer solutions except when
f ( x , y ) = (a, b, c, d ) (1,0, 1, 1) or (1, - 1, 1, l), when there are exactly four
solutions, or whenf(x, y ) (1, 0, - 1, l), when there are exactlyfive solutions.
The proof requires a detailed investigation and so we shall give only a
sketch of the method. We operate in the ring 2 [ 8 ] . In this there is one funda-
mental unit v, and this may be taken to satisfy 0 < 9 < 1. From equation (1 3).
+ +
( x ey)(x ey)(x + e y = 1,
and so x + 8y is a unit in the ring. Hence
x + 8y = T*, m = 0, f:l, + 2,... (14)
and we have to find the integer values of m for which this is possible. The first
stage in the proof is to show that equation (14) can be replaced by
x + 7)Y = Trn, (15)
and that 7 is a root of an equation with d = 1. The next stage is to show that
i n 3 0 except for the excluded case y3 +
q Z = 1 when v W 2= 1 + 7.
We show there is at most one unit for which xy # 0. Suppose that m, is the
least positive integer for which a relation
xo + 7)yo = rlm0 (16)
holds, and that x, y , m in equation (15) is a different set.
Put m = qm, + r, 0 6 r < m,.
It can be shown that 3 < r < m, - 2. Put
E=7)'= X+7)Y+7)2Z.
Then Z # 0 since r < m,. From equations (15), (16)
x + 7)y = (x, + 7)yo)"X + 7) Y + q 2 Z ) .
A congruence mod yo shows that Z = 0 (mod yo) and so IZI > yo. From
equation (1 7) and the conjugate equations,
+
- dDz = (7)' - 7)")e + (7)" - 7))E' + (7) - 7)"".
Also 773 2 7)r = & 3 7y0-2,
Then I$EI <1 since 1% I 3 n3

7) 7)
3
I = 1.
Hence, since 17'1 > 1,
WGoI < 2 + ~ ( I Y o I+ IYoI - 1) = -2 + 81yol.
Then - D < 64, and then there can be at most three units.
The cases D > -64 arise only for D = -23, -31, -44 and these must be
investigated in detail.
4. Theorem 4 does not enable us to find the existing integer solutions. All the
solutions, however, can be found for some equations of the form
ax3 + by3 = c.
In particular, as was first shown by Delaunay, the equation
s 3 + dy3 = 1,
does not present too much difficulty. The integer solutions are trivial when d
is a perfect cube. Then if [dl > 1, the only solution is x = 1, y = 0, and when
Id1 = 1, there is another solution x = 0, dy = 1. We may suppose now that
d > 1 and is free from cubed factors since these can be absorbed in y3.
We consider the cubic field K = Q ( a ) . The integers in K of the form
x + y q d + zf@, where x, y, z are rational integers, form a ring Z [ q d ] ,
the units in which are those integers 77 whose norm N ( 7 ) = f 1. Let E be
the fundamental unit in the ring chosen so that 0 < E < 1. Then all the units
in Z [ q d ] are given by
7 = &En,
where n takes all integer values. The + sign must be taken for the positive
units. Then if rational integers x, y satisfy
x3 + dy3 = 1, i.e. N ( x + ym)= 1,
7=x + y q d is a positive unit in the ring. Such a unit will be called a
binomial unit.
Theorem 5
The equation x3 + dy3 = 1 (d > 1) has at most one integer solution with
xy # 0. This is given by the fundamental unit in the ring when it is a binomial
unit, i.e. E takes the form E = x + y q d .
We require four lemmas.
Lemma 1
There cannot exist units
7 =P + Qqd + RVZ, 171 > 1, Q R = 0.
Suppose first that R = 0. Then P Q < 0, and
1/171 = P 2 - P Q V d + Q2YZ3 1 + Vd + Vdz > 3,
a contradiction.
So if Q = 0, PR < 0, and
1/171 = P 2 - P R Y d + R2VP > 3.
Hence we have to find the positive integers n such that
En =x + yVd.
Lemma 2
No unit of the types
(x + yVd>, n = +2, + 3,...; (.Y + yV/d), n = 51, + 2 ,...
can be a binomial unit.
We may suppose that the units are positive and so n > 0. Take the first
type. Expand and equate to zero the terms in q/d2,and put t = dy3. This
gives three different cases depending on the residue of n (mod 3). Then for
n = 2 (mod 3),
n = 1 (mod 3),
n = 0 (mod 3),
Since t = dy3, (x, t ) = 1. Suppose first that d > 2 and so 1x1 > 1. Let p be
a prime divisor of x and let p a be the greatest power o f p dividing the binomial
coefficient in the highest power of t in the equation above under discussion.
We shall show that the remaining terms in the equation are divisible byp"+'
thus giving a contradiction.
It will suffice to take the case n = 1 (mod 3). The general term with r > 0 is
given by
3' n.n - 1 . . . . .n - 3r - 1
-.
+
t(n-3'-"/3
X
(3r 2)!
Write this as
n - 2.n - 3.....n - 3r - 1
)
X3'
+ 1)(3r + 2)
t(n-3r- 1)/3
1.2 . . . . .3r ((3r
The first two terms in brackets are integers. Now ifp 2 2, p3' > 3r 2 since +
+
113' > 3r 2, and so the term in the third bracket is divisible byp.
We now take the second type of units, and write t = d2y3.
Three cases arise:
n = 2 (mod 3),
('I) t(n-2)13 + s
t(n-5)/3 3 (3+ .. ' + (X3)(n-2)'3 ('I)
= 0,
n = 1 (mod 3),
n = 0 (mod 3),
An argument similar to that for the first type applies.

8*
Suppose finally that d = 2. Then it is known that the only rational solutions
of,y3 + 2y3 = 1 a r e x = 1,y = 0; x = - I , y = 1 (see Chapter IS), and that
- 1 + v2 is the fundamental unit.
Lemma 3
The square of a unit 7 # k 1 cannot be a binomial unit.
Suppose that
+ QVJ + RVZ.
=P
Then Q2 + 2PR = 0. (18)

Also P3 + Q3d + R3d2 - 3dPQR = 1, (19)
since 7 is a unit.
We shall show that the only integer solutions of these simultaneous equa-
tions are given by P = l, Q = R = 0, and P = Q = 0, R = l , d = l , and
these are obviously excluded.
From equation (19), (P, R ) = 1, and so equation (18) gives four cases.
(A) P = q2, R = -2r2, Q = +2qr,
( B ) P = -q2, R = 2r2, Q = +2qr,
(C) P = 2q2, R = -r2 , Q = +2qr,
(D) P = -2q2, R = r2, Q = k2qr.
Here q, r are integers. Substituting these in equation (2), we have, on putting
p = dr3,
(A) -8p2 & 20pq3 + 4 = 1,
(B) -8p2 t 20pq3 + q6 = - 1,
(C) p 2 & 20pq3 - 8g6 = 1.
(D) p 2 f 20pq3 - 8qs = - 1.
Clearly (B) and (D) are impossible on taking residues mod 4.
From (A),
(4p + 5q3)2 = 27q6 - 2.
This is Fermats equation of which the only solutions are q = k 1, p = 0.
T h e n r = O , a n d P = 1, Q = R = 0.
Next equation (C) can be written as
(p + 10q3)2 - 1 0 8 ~ 6= 1,
or say p: - 4q; = 1.
Hence P1 + 1 = 2q;,
Then
q; - 43" = 1, q2 = , q3 = 0 or q2 = 0, q3 = -1.
Hence q = 0, p = f1, P=O, Q=O, R=d=l.
Lemma 4
The cube of a unit 7 # k 1, cannot be a binomial unit.
Let = P + Q V +~RVZ.
Then equating to zero the coefficient of vdz in q3, we have
P2R+ PQ2 + R 2 Q d = 0. (20)
Also P 3 + Q3d + R 3 d 2 - 3PQRd = 1. (21)
We shall show that the only solutions of these equations are P = 1, Q =
R = O o r P = R = 0, Q = 1 = d , o r P = Q = 0, R = 1, d = 1.
Write ( Q , R ) = 8 and so ( P , dS) = 1.
Then from equation (20)
R = S2r, Q = Sq, (q,r ) = 1,
and so + Pq2 + S3r2qd = 0,
P2r
or -r2qS3d = P(Pr + q2).
But (r, Pr + q 2 ) = 1, and so P = pr2,
and -qS3d = p(q2 + pr3).
Also ( P , dS) = 1 and so ( p , d8) = 1, and q = ps.
Then -S3sd = p(ps2 + r3).

Hence s = tp, -S3td = t2p3 + r3.
Now t I s I q, and since t I r and (q,r ) = 1, then t = k 1.

Hence k S3d = p 3 + r3. (22)
Substituting in equation (21) the values
R = rS2, Q = kp26, P = pr2, PQR = f r 3p 3 8 3,
we get p3r6 k p6S3d + r3S6d2 ? 3p3r3S3d= 1.
Replacing S3d from equation (22), we have
p3r6 - p 6 ( p 3 + r3) + r3(p3 + r3)' + 3p3r3(p3 + r3) = 1
or -p9 + r9 + 6 p V + 3p6r3 = 1.
There are several ways of dealing with this equation.
Writingpr2 = I, r3 - p3 = m,it becomes

913 + m3 = 1.
This is our equation with d = 9. The fundamental unit is E = - 2 + q9.
Also E" is a binomial unit only when n = 1. This proves the lemma.
Alternatively putting p 3 = u, r3 = u, we can verify the identity
(u3 + ~ U ' U + ~ U U ' - v3)U3 = V 3 + W 3 , (24)
where
u = u2 + uu + 02, v = u3 + 3u2v - u3, w = 3u2v + 3uv2.
Hence the only solutions of equations (23) and (24) are given by UVW = 0
a n d s o u = O , v = - 1 ; u = l , u = O ; u = - 1 , v = 1.
We can now prove that the nth power of a unit 7 # f 1, is not a binomial
unit. We may assume that 0 < 7 < 1 and n > 0. We need prove the statement
for n = 1 (mod 2) and n $ 0 (mod 3).
Write x~=(P+QV~+RVZ)~=X+YV~.
Then X" + pY" + p2zn = 0,
where p = t?I3, and Y, 2 are the conjugates of X in K. This equation can be
written in various forms depending upon the residue of n (mod 3). Thus when
n = 2 (mod 3),
Xn + (p")" + (pZ)" = 0.
Hence p2 Y + pZ divides the unit X and so must be a unit, i.e. on substituting
for Y, 2, then - P + 2 Q m - R q d z is a unit.
Hence -P3 + 8Q3d - R 3 d 2 - 6PQRd = f 1,
also P3 + Q3d + R 3 d 2 - 3PQRd = 1.
By addition 9Qd(Q2 - P R ) = 0,2,
and so either Q = 0 or Qz - PR = 0.
If Q = 0 , 7 = P + R V Z , and is excluded by Lemma 2.
If Qz - PR = 0,
1
- =
1
7 P+QV/~+RVZ
= P2 + Q W d z + R 2 V 8 - dQR - P R G - PQVd
= P2 dQR + (R'd - PQ)q/C,
-
is a binomial unit > 1 and is excluded by Lemma 1.

Suppose next that n E 1 (mod 3). Then
X" + (pY)" + (p")" = 0,
and so p Y + p2Z, i.e. - P - Q v d + 2R4'22 is a unit. Hence

P 3 + Q3d 8R3d2 + 6PQRd = f 1.
-
Also P 3 + Q3d + R 3 d 2 - 3PQRd = 1.
Hence 9Rd(r2d - P Q ) = 0.
Then either R = 0 and P + Q V ~ i i sa binomial unit and so n = 1 is the only

possibility from Lemma 2, or R2d - PQ = 0.
Then I/? = P 2 - d Q R + (Q2 - PR)Vdz
is a binomial unit > 1 and this is impossible.
Nagell* and Ljunggreng have generalized the method above and have found
more comprehensive results given as
Theorem 6
Let a, b, c bepositive integers, a > b > 1, c = I , 3, (ab, c) = 1, b = 1 ifc =
3. Then the equation
ax3 + by3 = c (25)
has at most one integer solution (x, y ) , and for this
c - yxva +yeq3
is either thefundamental unit or its square in the cubicfieldQ (vd)
defined by
,)"ba/Q
v( +
excluding however, the equation 2x3 y 3 = 3 which has the two
solutions ( I , 1) and (4, - 5).
Further there is at most one equation (25) with given d which has integer
solutions with xy # 0 except when d = 2, 20.
5. Suppose next that D > 0 in equation (12). There are now two fundamental
units ql, qz and so
x + ey = ?:2m. (26)
It is not often that equations such as (26) are so easily dealt with as in Chapter
27. In general, they prove rather difficult and troublesome as was seen for
the equation
x3 - 3xy2 + y3 = 1
considered in Chapter 23.
Baulin? following Ljunggren's treatment of this equation showed that the

only integer solutions of the equation
x3 + xzy - 2xy2 - y3 = 1
aregiven by (1, O,), (0,11, (- 1, 11, (- 1, - 11, (2, - 11, (- 4 % (5,4), (4, -91,
(- 9,5).
These two equations arise from the 7th and 9th roots of unity, and the
last one gives the cubic field of least positive discriminant, namely 49.
The important only other known result when D > 0 is given by Siegel's'O
Theorem 7
r f D is suficiently great, there are at most eighteen integer solutions of the
equationf(x,y ) = 1.
The proof depends upon ideas similar to those used in the proof of the
Thue-Siege1 theorem, but now the approximations to algebraic numbers are
given by hypergeometric series.
6. We give a few results on the rational solutions of the equation
ax3 + bxZy + cxy2 + dy3 = e,
or in the homogeneous form
ax3 + bxZy + cxy2 + dy3 = ez3, (27)
where now x , y , z are integers and (x. Y. z ) = 1.
As seen in Chapter 6, the non-existence of integer solutions of equations of
this form may be shown by simple congruence considerations, for example,
by taking equation (27) as a congruence mod 9.
Most of the known results about equation (27) deal with special cases of
the equation and equations derived from these, and in particular the equation
Ax3 + By3 + Cz3 = 0, (28)
and the derived equation
x3 - 3xy2 + y3 = pz3.
The equation (28) is a particular case of the equation
Ax3 + By3 + Cz3 + Dxyz = 0,
already considered in Chapter 10.
Some results have already been given for special cases of equation (28).
Selmer" has written an extensive memoir on this equation. The results are
usually found by algebraic number theory. By a slight change in the variables,
the equation (28) may be written in the form
x3 - my3 = nz3. (29)
The left-hand side is then split into linear factors. If m = If: 1, we can apply
the quadratic field Q(p). If m # ~f:1, the cubic field Q(yh) is required and its
arithmetical properties must be studied. Thus if 6' = $'%, equation (29) leads
to an ideal equation
(x - ey) = ?&3, (30)
where 7 is an ideal taken from a finite set. Then we have an equation
x - y6' = pa3 = (e + f 8 + go2)@ + 06' + w P ) ~ ,
where u, w are rational integers and e , f , g are taken from a finite set of
21,
rational numbers. Multiplying out and equating the coefficient of O2 on the

right-hand side to zero, we arrive at a homogeneous cubic equation, say,
f ( u , u, w) = 0. (31)
Sometimes congruence considerations applied to equations (30) or ( 3 1) may
show there are no solutions. This may require the reciprocity laws in quad-
ratic and cubic fields.
The equation (28) of most interest is the special case
x3 + Y3 = AZ3 (32)
where A is cube free. The application of the field Q(d/-3) gives a method of
descent which leads to equations of the form
ax3 + by3 + cz3 = 0, abc = A , (a, 6) = (b, c )
(33) = (c, a) = 1.
If any of these equations with 1 < a < b < c are solvable, so is the equation
(32). In fact, from a result due essentially to Euler, we can take
x+ Y = -9abcx3y3z3, x- Y = (ax3 - by3)(by3 - cz3)(cz3 - ax3),
2 = 3(abx3y3 + bcy3z3 + ~ a z ~ x ~ ) x y z .
Some such relation might be expected since both the equations (32) and (33),
when a solution is known for a set a , b, c, are birationally equivalent to the
cubic
y2 = 4x3 - 27A2.
Some of the equations (33) may be of special interest since they may be every-
where locally solvable but have no solutions. The simplest such equation is
3x3 + 4y3 + 5z3 = 0.
We now give some equationslz with no rational solutions.
Theorem 8
The equation
(px + 4 ~ + )kz3~ = ( r s + sy)(x2 + xy + 7y2), (34)
where k = 2 or 4, and the coeficients p , q, r, s are integers, has no integer

solutions except (0,0,O)i f
p -r = 1 (mod 2), p +q f 0 (mod 3),
and if either
q2 - qp + 7p2 $ 0 (mod P2)or s2 - sr + 7 r 2 f 0 (mod P2)
where P2 is a prime = 1 (mod 3) for which 2 is a cubic non-residue.
From this is deduced
Theorem 9
The equation
Ax3 + Dy3 + kz3 = 0, k = 2,4,
has no integer solutions except (0, 0,O)if
A =1Ph + P)(4 - 2P), D = 3d7P + d ( 7 P - 2q),
P + q $ 0 (mod 31,
and either
p = 1 (mod 2), q(q - p ) = 0 (mod 4),
or p = 2 (mod 4), q = 1 (mod 2),
and if either t
+
q2 - qp 7p2 f 0 (mod Pz) or pq(63p2 - 34pq 9q2) $ 0 (mod Pz). +
It is shown in due course that an instance is given by the equation
23 + iiy3 + 20x3 = 0.
This is of special interest since it has no integer solutions except (0, 0, 0), but
it is solvable locally everywhere. This is obvious since the congruence is
solvable to moduli 9, 5, 11 since 5 = 11 = 2 (mod 3). This result is given by
Selmer and is proved in an entirely different way.
In equation (34), we may suppose that ( x , y, z ) = 1 and so (x, y ) = 1.
Write f ( x , y ) = x2 xy + +7y2.
We consider the possible prime factors d of f ( x , y ) . Clearly d # 2 since this
requires x = y = 0 (mod 2). Next d # 3, since this requires x = y = 5 1
(mod 3), and so
j ( x , y) = (x - 4 ~ =) 0 ~(mod 9).
Then z = 0 (mod 3), p x + qy = 0 (mod 3) contrary to the end of (35). Hence
the only prime divisors of f ( x , y) are the primes P = 1 (mod 3), and the odd
primes P' = 2 (mod 3) and then ,Y = y = 0 (mod P') since (- 3/P') = 1.
t See addendum,
The only prime divisors of the left-hand side of (34) that need be considered
+
the common divisors of p x qy and z, the primes PI = 1 (mod 3) for
which 2 is a cubic residue; and this is so if and only if
P, = X: + 27Yf
with integers X I , Y,; the primes Pz = I (mod 3) for which 2 is not a cubic
residue and then z E 0 (mod P z ) ,p x + qy F 0 (mod P2);and the primes P'.
Let us consider next the divisor d = Pz.
Thenpx + qy E 0 (mod P z ) , f ( x ,y ) = 0 (mod P2)and sof(q, -p) = 0 (mod P2).
+
By (36),f ( s , - r ) $ 0 (mod P2)and so rx sy $ 0 (mod P2).Hence f(x, y )
= 0 (mod P i ) . Since P2 can be expressed as
P2 = x: + x2y2 + y;,
we have a factorization
f(x, y ) = nP'' n
P' pa
(x: + ~ 2 + y 3 3 n ( X : + 27 Y:).
~ 2
p1
Also f(x,y) =
Let 6 be a common divisor of these two factors in Q(d\/),and so

2x + y = 0 (mod S), yd\/-27 = 0 (mod S), 2 x d - 2 7 = 0 (mod 6).
Now 6 is prime to 3 sincef(x, y ) is prime to 3. If 6 = 2, y = 0 (mod 2). Then
x E 0 (mod 2), since equation (34) taken as a congruence mod 2 becomes
(PX)~E rx3 (mod 2).
Hence 6 = 1.
Since unique factorization exists in Q(d-),
x + $ ( I + dl-27)
= pn n P' n (x2 + $
P' P2
(1 + d 3 ) 3 n p1
(XI + Ylz/=z), (35)
where p = &(- 1 + dr3),a = 0, & 1.

Write x2 + y-"(i + d q )= A + B ~ T ,
2
where A , B are either both integers or both halves of odd integers.
Then (A + Bd-3I3 = A3 - 9AB2 + 3 d T 3 ( A 2 B - B 3 ) .
Here A(A2 - 9 B 2 ) and B(A2 - B 2 )are both integers and so

(A + Bd/-3)3 = C + 04-3,
where C, D are integers. Hence equation (35) can be written as
x + 22 (1 + d 3 7 ) = pa n pP'(X + YdT7).
Suppose first that CI = 0. Then y = 0 (mod 2) and this is impossible. Suppose
next that CI = & 1. Then
3y= * x - 3Y.
Then X = 0 (mod 3), andf(x, y ) = 0 (mod 9 ) and this is also impossible.
To prove Theorem 9, write equation (28) in the form
Ax3 + Bx2y + C,yy2 + Dy3 = kz3,
where
A = r - p3, B = r + s - 3p2q, C = 7r + s - 3pq2, D = 7s - q3.
Suppose now that B = C = 0. Then
7r + s = 3pq2, r + s = 3p2q,
and so r = t P d 4 - PI, s = 3Pd7P - 4).
Hence A = t P d 9 - P> - P 3 = + PN4 - 2P),
tP(4
D = 2Pd7P - 4) - q3 = t d 7 P + d ( 7 P - 2q).
We still have the condition p +q f 0 (mod 3). The condition p - r =
1 (mod 2) becomes
P - *P4(4 - PI = 1 (mod 2).
This gives either
p = 1 (mod 2), q(q - p) = 0 (mod 4),
or p = 2 (mod 4 ) , q = 1 (mod 2).
The condition f(s, - r ) $ 0 (mod P,) becomes
pq(63p2 - 34pq + 9q2) f 0 (mod P2).
Takep = 1,q = 4 , r = 6 , s = 6 , A = 5 , D = -22.Thenf(q, - p ) = 19 = P ,
and f(s, - r ) = 9 , 62 $ 0 (mod P2).On writing 2x for x etc., we get the
equation
z3 + iiY3 + 20.~3= o
previously noted.
24. BINARY CUBIC FORMS 23 I
Addendum to Theorem 9
Prof. Schinzel points out that the condition, either q2 - qp p 2 f 0 +
+
(mod P2)or pq(63p2 - 34qp 9q2) f 0 (mod P2)can be replaced by the con-
dition q f 0 (mod 7). For if neither statement is true, the identity
63p2 - 34pq + 9q2 = 9(q2-qp+7p2) - 25pq
shows that p q E 0 (mod P2),and then i f p = 0 (mod P2), also q = 0 (mod P z ) ;
and if q = 0 (mod PJ, then 7p = 0 (mod P2).These are impossible if
( q , 7 p ) f 0 (mod P J . Since a common divisor of p , q can be absorbed in x
and y , we may suppose that ( p , q ) = 1 and so we require that q f 0 (mod 7).
H e also notes that the equation is locally solvable everywhere it either
q = 4p (mod 9) or the cube free-kernel of pq = k 3 (mod 9). Thus if p = 2,
+
q = 3, we have the equation lox3 51y3 z3 = 0. +
REFERENCES
I . L. J. Mordell. The minimum of a binary cubic form I, 11. J. Lond. Math Soc.,
18 (1943), 201-219 and 210-217.
2. L. J. Mordell. The diophantine equation y 2 - k = x3. Pror. Lond. Math.
Soc., (2) 13 (1913), 60-80.
3. A. Baker. On the representation of integers by binary forms. Phil. Trans. R .
Soc. Lond., 263 (1968), 173-208.
4. B. Delaunay. Uber die Darstellung der Zahlen durch die binare kubische
Formen mit negativer Discriminante. Math. Z . , 31 (1930), 1-26.
5. T. Nagell. Darstellungen ganzer Zahlen durch binare kubische Formen mit
negativer Diskriminante. Muth. Z . , 28 (1928), 10-29.
6. W. Ljunggren. Einige Bemerkungen uber die Darstellung ganzer Zahlen durch
binare kubische Formen mit positiver Diskriminante. Actu Math., 75 (1942),
1-21.
7. V. I. Baulin. On an indeterminate equation of the third degree with least
positive discriminant (Russian). Tiil'sk Gos. Ped. Znst. U'ten. Zap Fiz Mat.
Nalrk Vyp., 7 (1960), 138-170.
8. T . Nagell. Solution complete de quelques Cquations cubiques a deux indeter-
minees. J . lie Math., (9) 4 (1925), 209-270.
9. W. Ljunggren. On an improvement of a theorem of T. Nagell concerning the
diophantine equation Ax3 + B y 3 = C. Math. Scan., 1 (1953), 297-309.
10. C. L. Siegel. Uber einige Anwendungen diophantischei Approximationen.
Abh. preuss. Akad. Wiss. Phys. Math., KI (1929), Nr 1.
11. E. S. Selmer. The diophantine equation ax3 + by3 + c z 3 = 0. Actn Math.,
85 (1951), 203-362.
+
12. L. J. Mordell. Equations ax3 + bx2y cxy2 + d y 3 = e with no rational
solutions. J . Lond. Math. SOC.,43 (1968), 433-438.
CHAPTER 25
Binary Quartic Forms
1. Let
F = F(x, y ) = ax4 + 4bx3y + 6cx2y2 + 4dxy3 + ey4 = (a, b, c, d, e)(x, y)*,

where a, b, c, d, e are integers. We shall require some results on the invariants
and covariants of F(x, y).
There are two invariants,
a b c (1)
g2 = ae - 4bd f 3c2, g, = b c d .
c d e
There are quartic and sextic covariants, H ( x , y), G(x, y), defined by
I a2F a2F I
=I + 2bxy + cy2 bx2 + 2cxy + dy2

ax2
+ 2cxy + dy2 cx2 + 2dxy + ey2
bxa
= (b2 - ac)x4 + (2bc - 2ad)x3y + (-ae - 2bd + 3c2)x2y2
+ (2cd - 2be)xy3 + (d2 - ec)y4. (2)

Also
aF aF I
1
G = G(x,y) = --
8
By means of a linear substitution with real or complex coefficients, F(x, y ) ,

if it has no squared factor, as we suppose hereafter, can be reduced to the
canonical form
F(x,y) = x4 + 6rnx2y2 + y4.
25. BINARY QUARTIC FORMS 233
Here g2 = 1 + 3m2, g3 = m - m3,
H ( x ,y ) = -mm(x4 + y4) + (3m' - 1)x2y2,
C(S,y ) = (1 - 9m2)xy(.y4 - y4).
The relation between the covariants of the quartic is given by
Theorem 1
G2 = - g2HF2 - g3F3
4H3
= 4(H - e,F)(H - e2F)(H - e3F), (4)
where e l , e2, e3 are roots of the equation
4t3 - g,t - g, = 0.
It suffices to prove the result for the canonical form of the quartic. The e are
roots of the equation
4 t 3 - (1 + 3m')t - nz + m3 = 0,
or (t + m)(4t2 - 4mt + m2 - 1) = 0,
m + l in - 1
and so el = -m, e2 = - e3=-.
2
9
2
We verify at once that
H - e,F = (9m2 - l)x2y2,
H - e2F = -#(3m + I)(x' + y2)',
H - e3F = -3(3m - 1)(2 - y')',
and the theorem follows.
We have already proved in Chapter 18 the
Theorem 2
The class nirnzber for binary quartics with given inoariants is finite.
The conditions for a reduced form cannot be expressed in very simple
terms. Birch and Swinnerton-Dyer in their exhaustive study of rational
points on elliptic curves, found it convenient to develop a theory of reduction
which is independent of Julia's geometric presentation but is more suitable
for computational purposes.
2. We can now find' the integer solutions of the equation
2' = X 3 - G 2 X Y 2 - G3Y3, (5)
where G2, G, are given integers and the right-hand side has no squared linear
factor.
Let F(x, y ) be a representative of the binary quartics with invariants

g, = 4G2, g, = 4G3. Then from equation (4), a solution is given by
X = H ( P , 41, y = F(p, 41, 22 = G(p, 41, (6)

and X , Y, Z are all integers if p, q are integers.
If (p, q ) = 6, then X/a4, Y/iS4, Z/Ss are integers and so we may take S = 1.
We shall prove that all the integer solutions for which ( X , Y ) = 1 are included
in these formulae. For let (h, a, g ) be such a set of values for ( X , Y , Z ) , and so
g 2 = h3 - G2h a - G3a3, (a, h) = 1.
We shall construct a quartic (a, b, c, d, e)(x,y)4 with invariants 4G2, 4G3 with
leading coefficients h, g for the quartic and sextic covariants respectively. Then
if the quartic is equivalent to F(x, y ) , we have
a = F(p, d , h = H(P,d , 2g = G(p, 41,
and since (a, h) = 1, then (p, q ) = 1. We have

h = b2 - ac, and so (a, 6 ) = 1.
We take b = g/h (mod a2),
and then ac = ( g 2 - h3)/h2(mod a2),
and so c is an integer.
We define d by
2g = a2d - 3abc + 2b3.
Since h = b2 - ac, hb = b3 - abc, then
a2d = b3 - 3bh + 2g = g3/h3 - g (mod a2)
= 0 (mod a2),
and so d is an integer.
Finally we define e by
4G2 = ac - 4hd + 32.
On substituting the values of r, d in here, we find
a3e = 4a2G2 - 3h2 + b4 - 66% + 8bg.
Substituting in here the values of g, h, G2, we find
4G, = are + 2bcd - ad2 - b2e - c3,
and so the quartic has invariants G2, G,.
Since ae and (ac - b2)e are integers and (a, b) = 1, then e is an integer.
Considerable numerical detail may be involved in applications as is seen
for the equation
2
, = s3 +j 3 ,
whose solution when y is odd and prime to x, is given by

x = -4p3q + 494, Y = P4 + 8pq3,
x = -p4 + 6p2q2 + 3q4, y + 6p2q2 - 3q4,
= p4
s = p4 + 6p2q2 - 3q4, y = -p4 + 6p2q2 + 3q4,

x = 2p4 - 4p3q - 4pq3 + 2q4, 4 = p4 + 4p3q - 6p2q2 + 4pq3 + q4,
s = 4p3q + 24p2q2 + 48pq3 + 36q4, y = p4 + 8p3q + 24p2q2 + 24pq3.
Here p, q take such integer values that y is odd and prime t o x.
Results for equation ( 5 ) may sometimes be found by factorizing the right-
hand side in the cubic field given by - G20 - G3 = 0, and in particular
when unique factorization holds in the field.
Thus all4 the rational integral solutions of
2 2 = x3 + 4y3,
where x, y , I are relative prime in pairs is included in
x = p(p3 + 431, y = q(q3 - 2p3), + z = p6 - 1 0 ~ 3 ~-3 2q6.
The equation
2z2 = s3 + 13
has also been solved completely.

Three o r four sets of formulae are required according as (s,y ) = 1 o r 2.
3. We can apply the theory of the binary quartic to prove the
Theorem 3
The equation
ey2 = a s 3 + h.r2 + C.Y + cl, a # 0,
(7)
where the constants are integers and the right-hand side has no squared linear
factor in s,has only afinite number of integer solutions.
The proof will be given in Chapter 27 and depends on Thues theorem that
the equation
.f(s,y) = a s 4 + hs3j>+ cx2y2 + d,vy3 + ey4 = m,
(8)
where f ( s ,y ) is not a square i n x, y , has only a finite number of integer
solutions. In general, it is not too easy to find them, but thep-adic methods of
Chapter 23 are sometimes useful.
A particular case of a theorem by Siegel, given in Chapter 28, states that

the equation
kz2 = f ( x , 1)
has only a finite number of integer solutions. This is easily proved for the
special case
kz2 = ax4 + cx2 + e. (9)
For, writing it as
kz2 = ax4 + cxzy + ey2,
where y = 1, the general solution as shown in Chapter 8 is given by a finite
number of expressions of the form
x2 = lp2 + mpq + nq2, 1 = 11p2 + mlpg + nlq2,
where p , q are integer parameters. The solution of the first is given by a finite
number of expressions of the form
p = rX2 + sXY + t Y 2 , q = rlX2 + s,XY + f l y 2 ,
where X , Yare integer parameters. Substituting in the second equation gives
finally a finite number of equations of the form
F ( X , Y ) = AX4 + B X 3 Y + C X 2 Y 2+ DXY3 + E Y 4 = 1. (10)
Alternatively, as shown in Chapter 8, the solution of equation (9) can be
expressed by a finite number of expressions of the form
.? = klTn + k2Un,
where, say, D = 4ak and
T, + U,d/o = (T + Ud/o)",n = 0, f 1, k2,. . .
and T, U is the fundamental solution of
Y2- DX2 = f l .
The problem now is to find n. In some instances, the solution may be found
by discussing various congruences satisfied by T,, U,,. In others, the p-adic
methods of Chapter 23 must be applied.
Some results on the number of solutions of irreducible equations such as
(lo), say ( A , B, C, D, E ) where A > 0, E > 0, have been given by Nagell.5
Suppose that the equation f(x, 1) = 0 generates only imaginary biquadratic
fields with at least one quadratic subfield. Thus he has proved
Theorem 4
The number N of solutioris of equation (10) is always even and N < 8.
Also N = 8 ifand only if
F(X, Y ) - ( 1 , 0 , - 1 , 0 , 1 ) o r ( l , - I , - 1 , 1 , 1 ) ; a n d N = 6 i f a n d o n l y i f
F ( X , Y) (l,O, r2, -2r, 1) or (1,2r, r 2 + s, rs, 1) or (1, 1, 1, 1, 1) where r is
N
an integer and s = I , the case r = 2, s = 1 being excluded.
REFERENCES
1. L. J. Mordell. Indeterminate equations of the third and fourth degrees. Q.J.
Pure appl. Maths., 45 (1914), 170-186.
2. B. J. Birch and H. P. F. Swinnerton-Dyer. Notes on elliptic curves, I. J. reine
angew. Math., 212 (1963), 7-25.
+
3. F. E. G. Rodeja. On the diophantine equation x3 y 3 = 2z2. Riuista Mat.
Hisp. Am. (1953), 4-13, 229-240.
4. C. Georgikopoulous. Rational solutions of the equations x3 4y3 = za, +
+
x3 2y3 = z2. Bull. SOC.Math. Grkce., 24 (1948), 13-19.
5 . T. Nagell. Sur les representations de l'unitk par les formes binaires biquadra-
tiques du premier rang. Arkivfor Mat., 5 (1965), 477-521.
CHAPTER 26
The Equationf = x3 +k
1. The Diophantine equation
y 2 = x3 + k.
This has played a fundamental role in the development of number theory.
The earliest recorded result was given in 1621 by Bachet who noted that when
k = -2, and so x = 3 is a solution, other solutions can be found by the usual
tangent method. Then Fermat posed a problem for the English mathema-
ticians to show that the only integer solutions of y 2 = x3 - 2 are given by
s = 3. He asserted that, at first sight, this seemed to be a difficult question,
although an infinity of fractional solutions can be deduced by the method of
Bachet. He stated that he had discovered an exceedingly beautiful and subtle
method which enabled him to solve such questions in integers and which was
not known to Bachet. His proof was never published. A faulty proof was
given many years later by Eu1er.t
It may be remarked that the first proof of the existence of an infinity of
rational solutions was given by Fueter in 1930, and will be considered further
on.
We consider first various methods for discussing integer solutions of the
equation (I).
Congruence considerations
It has been known2 for nearly a century that many equations can be proved
to have no integer solutions when they can be written in the form
y2 + lm2 = x3 + n3, (2)
This is considered as a congruence mod 2l. Simple congruence considera-
tions may eliminate some of the residues of x mod 4 or mod 8, or mod I , and
the other residues may often be eliminated by considering the quadratic
characters of x + I?, x2 - x n + n2 mod 8 or mod 1.
Some illustrations are now given.
The equation
y2 + 4a2 = x3 + (4b - 1)3, (3)
where a has no prime factors = 3 (mod 4).
Clearly
x f 0 (mod 2), x $ - I (mod 4), and so x = 1 (mod 4).
t For the early history, see Dicksons History of the Theory of Numbers 11.
26. THE EQUATIONy 2 = x3 +k 239
Then
.P - s(4b - I ) + (4b - I)2 = 3 (mod 4),
and is > O and so cannot divide y 2 + 4u2. Hence there are no solutions.
The equation
y2 + (2a + = x3 + (46 + 2)3,
where 2a +
1 has no prime factors = 3 (mod 4).
Here x $ 0 (mod 2), x f - 1 (mod 4) and so x = 1 (mod 4).
Then
x2 - x(4b + 2 ) + (46 + 2)2 = 3 (mod 4),
and so cannot divide y 2 (2a ++
We give instances of equation (2) when I = ? 2.
The equation
y2 - 2b2 = x3 - a3,
where a = 2, 4 (mod 8), b = 1 (mod 2), and b has no prime factor
= k 3 (mod 8).
Here x f O(mod 2), x $ 1 (mod4) and so x = - 1 (mod4), i.e. x =
- 1 , 3 (mod 8). When .Y = - 1 (mod S), x - a = t 3 (mod 8) and since
x - a > 0, it cannot divide y 2 - 2b2. When x = 3 (mod 8),
x2 + a x + a2 = 1 + 3a + u2 = k 3 (mod 8),
and so cannot divide y 2 - 2h2.
The equation
y2 + 2b2 = x3 - a3,
where a = 4 (mod 8), b = 1 (mod 2) and b contains no prime divisors
= 5, 7 (mod 8).
Here also x = - 1, 3 (mod 8). When x = - 1 (mod S), y 2 = 5 (mod S),
When s = 3 (mod 8), x - a = 7 (mod 8) and cannot divide y 2 2b2. +
Results may also be found if k can be written in several forms.
The equation
y2 = x3 + 45. (7)
Clearly x f 0 (mod 2), x f 1 (mod 4) and so x = 3, 7 (mod 8).
Write y' - 2.6' = x3 - 27.
Here x $ 0 (mod 3) since if x = 3X, y = 3 Y , Y 2 = 3 X 3 + 5.
240 DIOPH ANTINE EQUATIONS
Then if x 5 3 (mod 8), x2 + 3x + 9 = 3 (mod 8) and so cannot divide

y2 - 2.62.
Write next J" - 2 . 32 = ,P + 27.
Then if x = - 1 (mod 8), x2 - 3x + 9 = 5 (mod 8) and so cannot divide
y 2 - 2.32.
Hence there are no integer solutions.
The equation
y 2 - 3b2 = .'c3 - a3,
where a = 1 (mod 4), h = L-2 (mod 6) and b has no prime .factors
= & 5 (mod 12).
Here x $ 0 (mod 2), ,Y $ - 1 (mod 4) and so x = 1 (mod 4). Also x =
+
a, a 1 (mod 3). We reject x = a since then x3 = a3 (mod 9) and so
y 2 = 3 (mod 9). Hence
x2 + a s -t a2 = 1 (mod 3), =3 (mod 4), = 7 (mod 12).
Then x2 + ax + u2 contains a prime factor p = f 5 (mod 12) and since
(3/p) = - 1, it cannot divide y2 - 3b2.
We have the more general2
Theorem 1
The equation
y2 - kb2 = x3 - k3a3
has no integer solutions when
a = - 1 (mod 4), b = 0 (mod 2),
k is squarefree, k = 3 (mod 4), ( k , 6 ) = 1 , and b $ 0 (mod 3) i f k = 2 (mod 3);
and lastly a and b hatie no common prime factor p i f the quadratic character
(k/p) = - 1.
+
Since y2 3 x3 - 1 (mod 4), x 0, 2, 3 (mod 4), and so x 3 1 (mod 4).
Write y 2 - kb2 = (x - ka)(x2 + kax + k2u2),
and F = x2 + kax + k2u2 > 0.
Now (x, k ) = 1 for if q I ( x , k), where q is a prime, then q I y 2 but q2 t kb2
since ( k , b) = 1. Hence (F, k ) = 1.
Then since F = 3 (mod 4),
and so Fcontains as a factor a primep to an odd power such that ( k / p ) = - 1.

26. THE EQUATION y2 = x3 +k 24 1
Hence y = b = 0 (modp), and so p occurs to an even power as a divisor of
y 2 - kb2 and to an odd power in x - ka. Then
x - ka = 0 (modp), x2 + kax + k2a2 = 0 (modp),
and so 3kax z 0 (modp).
If p = 3 , then k = 2 (mod 3 ) since ( k / p ) = - 1 and we would have b =
0 (mod 3), and this has been ruled out.
If x 3 0 (modp), ka E 0 (modp) whereas k is prime to p, and a is also
prime to p since p 1 (a, b).
2. Applications of the quadraticjield Q(dk>
We recall some well known results. We suppose that k is a square free
integer. Two cases arise.
k = 2 , 3 (mod 4). Then the integers x in the field are given by x = x 1 x , d / k+
where x 1 and x2 are rational integers. Also x is a unit if x: - kxi = f 1 .
k E 1 (mod 4). The integers are given by x = x 1 x2
1+*
+
2 ( andx )
is a unit if x: + +
xlx2 fl(1 - k)xi = t 1.
It is known that all units are given by E = t 7f"' where 7 is a fundamental
unit and n takes all integer values.
If k < 0, the only units are t 1 except that when k = - 1, there are four
units given by in, n = 0, 1 , 2 , 3, i.e. k 1, t i , and that when k = - 3 , there
- 1 + d-3
are six units fp", n = 0, 1, 2, where p =
2 '
namely
f l , f ( -1 + "3)
, f( - 1 -2d")
We begin with the case when h , the number of classes of ideals in the field
Q ( d k )is not divisible by 3, and this of course is so when unique factorization
holds in the field, for example, when k = - 1, - 2 , - 3 , -7, . . ., k =
2 , 3 , 6, . . . .
Suppose now that k = 2 , 3 (mod 4 ) or 5 (mod 8).
This means that x f 0 (mod 2) in
y2 - k = x3.
Write (y + d % ) ( y - dk)= x3.

The two factors are relatively prime since any common divisor would
divide 2 d k but (x, 2k) = 1. Hence by the general theory of Chapter 15
y + dk = &(a + bw)3, y - dk = &'(a + bw')3,
(9)
.'i = N(u + bw),
where a, h are rational integers, e , el are conjugate units, w , w' are conjugate
numbers and w = d k if k = 2, 3 (mod 4), w = (1 +
d k ) / 2 if k = 5 (mod 8).
Also if k < 0, E = f I , and if k > 0, E = I , ? , 7 - l where 7 is a fundamental

unit of the field Q(Uk) since a factor - 1 can be absorbed in the cube.
Suppose, for example, that k < 0, k = 2, 3 (mod 4).
Then y + dk = (a + b d k )3 , x = a2 - kb2,
and so 1 = b ( 3 2 + kb2). (10)
Solutions arise only when b = f 1, k = & 1 - 3 2 .
In particular, when k = - 2 , we have only a = t- 1 and x = 3, as stated by
Fermat. When k = - 13, h = 2, and a = 5 2 , x = 17, y = f 70 are the
only solutions.
Suppose next that k > 0, k = 2 , 3 (mod 4). We may always take 7 to be
the unit defined by 7 = T + U d k where (T, U ) is the fundamental solution
of T 2 - k U 2 = 1. For if yo = t + u d k where t 2 - ku2 = - 1 , =~ 720 and
then yo = ~ : / 7and so T$ can be absorbed in the cube. No solutions arise from
E = 1 since equation (10) holds. The units 7, 7 - l give
1 = T(3a2b + kb3) f U(a3 + 3kab2). (1 1)

As shown in Chapter 7, congruences mod 9 or mod 7 sometimes suffice to
prove that no integer solutions exist. We prove this for the two cases,
k = 4 (mod 9), = 0 (mod 9),
if U
and k = 7 (mod 9),
if U = f 3 (mod 9).
From T 2 - k U 2 = I , we have U = 0 (mod 3), T = &, 1 (mod 9), and so
from equation (1 1) as a congruence mod 3, Tb = 1 and so b = f 1. Then
from equation (1 1) as a congruence mod 9,
1 = 3a2 + k k Ua,
and so a f O(mod 3 ) since k f I (mod9). Also a $ & I (mod 3 ) since
1 $ 3 + k f U(mod9).
Instances are given by k = 7, 34, 58, 70.
On taking (1 1) as a congruence mod 7, we can show that the equation is
impossible if U E 0 (mod 7) when k = 4 (mod 7). Then x = 0 (mod 7) since
y2 - 4 = x3 (mod 7), and from x = a2 - kb2, a = f 26 (mod 7). Then
equation (1 1) becomes
1 = f (3a2b + 4b3) = f2b3 (mod 7),
and this is impossible. An instance is given by k = 159, T = 1324, U = 105.
We now consider some cases when the class number is divisible by 3. We
suppose now that k is square free, k = 2, 3 (mod 4) and h(Q(d/k))=
0 (mod 3).
In (y + d L ) ( y - dk)= x3,
26. THE EQUATION y2 = X3 k 243
the two factors are relatively prime since (x, 2k) = 1. Hence we have the ideal
equations
(u + d k ) = a:, (u - dk) = a:.

Let 8, be an ideal in the inverse class of al.Then
df(y + dk)= = (X + ~dk)3,

where X , Yare rational integers. Also
tf = ( p + qdk), p 2 - kq2 = (N(d1))3= n3, say,

where p and q are rational integers.
Hence we have an ideal equation
+
N ( ~ , ) ~ ( J Jdk)= ( p - q l / k ) ( X + yl/k13,
i.e. an equation,
n3(y + dE)= ?)(p - q d k ) ( X + Yl/k)3, (12)
where 71 is a unit in Q(dk).If k < 0, we can take 7 = 1, and if k > 0, then
+ U d k is the fundamental solution of T 2 -
71 = 1 , e , e - l where e = T
-
k U 2 = 1 . There will be as many sets of equations of this kind as there are
classes of ideals & whose cubes give the principal class. When & 1 , we take
n = 1. If it is convenient to take b to be a prime ideal, then n will be a prime.
The resulting cubic equation in X , Y takes the form
AX3 + 3BX2Y + 3 C X Y 2 + D Y 3 = m3,
and can sometimes be shown to be impossible by simple congruence con-
siderations.
Let us now consider the equation
y2 - k = x 3
when k is square free, k < 0, k = 1 (mod 8), k # -7, h = 3.

An integral basis in Q(d/i;) is (1, w ) where w = (I + &)/2. We have an
ideal factorization of 2 given by
(2) = (2, 1 + w)(2, 1 + w ) = VV,
where w , w are conjugate as are the ideals %?, V . Then V is not a principal
ideal since for integers a, b,
2 # (a + bw)(a + bw),
follows from
+
8 # ( 2 ~ b) - kb2.
Now ( y + dk,y - d i ) = 2 or 1 since % # v'. We have two cases. In

the first, x is even and hence the ideal equation
(y + dk) = (2)ga3, (13)
which is impossible since a3 is a principal ideal.
In the second case, x is odd and we have
(y * dk) = a3, (14)

and this may be difficult to deal with. Other methods may be used to dispose
of odd values for x. Suppose for instance that
y2 + 31 = x3.
Then x f 1 (mod 4) since y 2 f 2 (mod 4). Also x f - 1 (mod 4) from
y2 + 4 = (x - 3)(x2 + 3x + 9).
+ +
Here x 2 3x 9 = 3 (mod 4) and is > O and so cannot divide y 2 4. +
3. Application of cubic residues
Another method of approach is to use cubic residues to a primep > 3. We
recall that the congruence
x3 = I (modp), (I, p ) = 1,
is always solvable if p 3 2 (mod 3). I f p z 1 (mod 3), solvability requires a
condition on I , namely that I should be a cubic residue of p , written as
(lip), = 1 and so l(P-1)'3 = 1 (modp). If I = 2, the condition becomes the
existence of integers r, s such that
r2 + 21s2 = p.
If I = 3, the condition i s
r2 + 243s2 = 4p.
We now prove3 the
Theorem 2
The equation
y 2 = x3 + k , k = 2a3 - 3b2, (15)
has no integer solutions if a, b are integers, ab # 0, a $ 1 (mod 3), b $
0 (mod 3), and if when h = 0 (mod 2), then a = 1 (mod 2 ) ; and i f 2 is a cubic
residue of every odd prime p = 1 (mod 3) dividing a.
Write x3 + 2a3 = y 2 3b2. + (16)
The possible primes dividing the left-hand side are the common prime factors
of x and a, 2,3, primes q = 2 (mod 3), and primes p = 1 (mod 3) for which
@/PI3 = 1.
26. THE EQUATIONy 2 = .x3 +k 245
We show first that the left-hand side, say L, is prime to 6 . For if .Y = 2s,,
then y = h (mod 2 ) , and so eithery = h E 0 (mod 2 ) or y = b = 1 (mod 2).
The first case is impossible since a = I (mod 2). The second case gives
x3 + 2a3 = 4 (mod 8) and this is also impossible.
+
Next if .x3 2a3 = 0 (mod 3), y = 0 (mod 3), and so x3 2a3 E +
3 (mod 9). This is impossible since a3 = 0, - 1 (mod 9). Suppose now that a
prime q =- 2 (mod 3) divides L. Then y = b = 0 (modq). Hence we can
write y = Qy,, b = Qb,, where Q is the product of factorsq dividingL. Then
Q2 must divide out in both sides of equation (16) and the only other prime
factors of L are primesp = A 2 + 27B2.
Since factorization in the quadratic field Q ( Y ' I 3 ) is unique, we have
(y1 + b,d-3)pa = n
P
(A + Bd-27) = C +DdZ7,
+
where p = f(- 1 d z 3 ) , a: = 0, f. 1. I f a: = 0, then b, = 0 (mod 3)
contrary to hypothesis. If a = -t 1, y, + b, E 0 (mod 2), and then L =
0 (mod 2), and this has been excluded.
Instances are given by k = 51, with a = 3, h = 1 , and k = 13, with a = 2,
b = 1.
A similar result is
Theorem 3
The equation
s3 + 4a3 = y2 + 3b2, ab # 0, (17)
has no integer solution if b = rf: 1 (mod 6), a e 0 , 2 (mod 6), and a has no
prime factors p = 1 (mod 3).
A variant using the cubic character of 3 is given by
Theorem 4
The equation
x3 + 3a3 = y 2 + 3b2, (18)
has no integer solutions if k / 3 = a3 - b2 $ 0 (mod 3) unless that one of
h, h & ( 1 - k ) \t,hich is dirisihle by 3 is also diiisible by 9.
The left-hand side is not divisible by 3, for this would require y = 0 (mod 3),
x = 0 (mod 31, a3 - b2 = 0 (mod 3). Hence
y = k 1 (mod 3), x3 = 1 (mod 3), .x3 = 1 (mod 9),
and then
y2 = 1 + k (mod 9), y = f (1 - k ) (mod 9).
Hence the only possible divisors of the left-hand side are the common
divisors of s and a, the primes p = 1 (mod 3) of which 3 is a cubic residue
9+
and the primes q = 2 (mod 3). If q is a divisor, then since ( - 3 / q ) = I,

y = 0 (mod q), b = 0 (mod q).
Hence x3 + 3a3 = Q2(~: + 363,
where Q is a product of primes q. The factor Q2 divides out and so
where C, D are integers and a = 0, 4 1.

If a = 0, b, = 0 (mod 9) and so b = 0 (mod 9).
I f a = f l , y l f b , r O ( m o d 9 ) a n d s o y f b-O(mod9).
The theorem now follows since y = -t (1 - k) (mod 9).
An illustration is given by y 2 = x3 - 24 where 24 = 3 .32 - 3 a . 13.
4. Application of binury cubic forms

We consider now in more detail than in Chapter 24, the application of the
binary cubic form to the equation
y 2 = x3 + k.
Let (x, y ) = (p, q ) be a solution. With it, we associate the binary cubic form
f ( x , y ) = x3 - 3 p x p 2qy3 + (19)
with binomial coefficients and so of discriminant
D = -4q2 t 4p3 = -4k.
Hence f ( x , y ) is equivalent to one of the finite number of binary cubics
of discriminant -4k representing unity. Conversely, since every such cubic
is equivalent to a cubic
g ( x , y ) = x3 + 3bx2y + 3cxy2 + dy3, (20)
this gives at once a solution of y 2 = x3 + k on writing x - by for x in
equation (20), and reducing it to the form (19). Since by Thues theorem there
are only a finite number of representations of unity by g(x, y), y 2 = x3 k +
can have only a finite number of solutions.
In general, it is not an easy matter to find all the solutions, and even for
some small values of k, a surprising amount of numerical detail may be
required. When k > 0, the cubic g(x, y ) has negative discriminant, and so the
representation results by Delaunay and Nagell in Chapter 21 are applicable.
Thus all the equations with 0 < k < 100 have been solved, and solutions are
given in Hemers thesis4. In particular, Nagel15 has shown that there are
exactly eight solutions when k = 17, namely,
x = -1, -2,2,4, 8, 43, 52, 5234,
y = 4, 3, 5, 9, 23, 282, 375, 378661.
26. THE EQUATION y2 = x3 + k 247
When k < 0, the problem becomes much more difficult since there are no
useful general theorems about the representation of unity by binary cubic
forms of positive determinant. There are twenty unsolved equations for
0 > k 3 - 100. These correspond to irreducible cubic fields having two
fundamental units. Since there is only one relation connecting them, the
p-adic methods of Chapter 23 cannot be applied directly. It becomes necessary
to introduce new units in fields containing the cubic field as a subfield. An
illustration is given by the equation y 2 = x3 - 48 which has only the solu-
tions (x, y ) = (4, k4), (28, It 148) and is of particular interest. On replacing
.Y by 4x and y by 8y + 4, it becomes
y2 + y + 1 = x3 = ( y - p)(y - p"),
which has only the solutions (x, y ) = ( I , 0), (1, - I), (7, 18). This follows
since from the equation
Y-P = 2 P"(P + 4PI3,
we easily deduce an equation
x3 - 3XY2 + Y3 = 1.
Ljunggren6 solves this completely in the manner indicated above, and some
details are given in Chapter 23.
It is sometimes desirable to use the binary quartic equations which have
been seen to arise. Now there are two fundamental units and two relations
connecting them and so the p-adic method can be applied directly. In this
way, Ljunggren7 has shown that when k = -7, the only solutions are
(s,y ) = (2, I), (32, 181); and that when k = - 15, the only solution is
( - Y , Y ) = (1, 2).
+
When solutions exist for y 2 = x3 k , and we have no criterion for this, a
computer may be very useful unless the solutions are exceedingly large numbers.
Thus tables6av6bhave been compiled for Ikl < 9999 and 0 < y < lolo.
This makes the range for s, -21 < .Y < 4, 641, 589.
5. Rational solutions
+
When one rational solution of the equation y 2 = x3 k is known, it has
been shown that in general a new solution can be found, and so an infinity of
rational solutions can be expected. The first result, however, is due to Fueter*
who proved the
Theorem 5
Let k befreefrom sixth power primefactors. Then if the equation y 2 = x3 + k
has a rational solution (x,y ) = ( p , q ) with p q # 0, there are an infnity of
rational solutions unless k = -432 when there is only the solution x = 12,
y = k 36, and k = 1 n,hen there are only the solutions x = 0, - 1,2.
His proof, however, is rather complicated since it depends upon the

association of the equation with the related equation y 2 = x 3 - 27k. Another
proof0 can be given based on Bachets idea. It may be recalled that Fermat
stated without proof that when k = - 2 , this idea leads to an infinity of
rational solutions.
It has been shown by putting x = p + +
X , y = q 3paX/2q, that when
p q # 0, a solution is given by
u = -9P4
-2
p, I =
8q4 -36q2p3 + 27p6
- 492 8q3
We have to investigate when this solution is different from ( p , 4). The
numerator of y shows that y # 0, and x = 0 can arise only from 9p3 = 8q2,
i.e. q = 3t3, p = 2t2, k = t s , and so effectively from k = 1. On putting
p = P/D2,q = Q / D 3 ,
Q2 = P3 + kD6,
where(P, Q , D ) = I, (P, D ) = I, ( Q , D ) = 1.
9P4
Also Dx = - - 2P.
4Q2
Hence we certainly have a new solution if 3P2/2Qis not an integer, and so we
may suppose that 3P2/2Q is an integer. Then the special cases must be con-
sidered when P = 0 (mod 2) or Q = 0 (mod 3 ) or (P, Q ) > 1. It can be
shown that a similar procedure with the special cases leads to a solution which
is not a special case. In this way, we find a solution which either has a greater
denominator than D , or, if the denominator is D , the numerator is greater
than P.
Equations with no rational solutions

Two general methods are known for proving the non-existence of rational
solutions. The first depends essentially upon the study of quadratic fields, and
the second upon cubic fields.
Some proofs may require a deep result from the theory of relative fields
given by the
Lemma
ZfQ(Z/k) and Q(d-k) are two quadraticfields, and ifthe class number for
the imaginary field is prime to 3, then the class number for the real field is also
prime to 3.
We need not use this lemma if we postulate the non-divisibility by 3 for
both class numbers.
26. THE EQUATION y 2 = x3 + k 249
The first general result is due to Fueter8 who proved the
Theorem 6
The equation y 2 = 53 + k has no rational solutions i f k is a negative integer
free from sixth power factors, and contains all odd prime factors to an odd
power, and if also
k = 2 (mod 9), k f 1 (mod 4), k f 4 (mod 16),
and the class number h of Q(dk)is not diiisible by 3.
It suffices to consider the equation
y 2 = .y3 + kt6,
and to show that there are no integer solutions in ( x ,y , r ) with t # 0 and
(s,y , t ) = 1. With such a solution, we could associate a cubic equation
u3 - 3xu - 2y = 0
of discriminant d = - 108kt6.
The solution of the cubic is given by
u =(y + t31/k)1'3 + ( y - t 3 d k p 3 .
Two cases arise since either the equation is irreducible in the field Q(z/-),
or is reducible and so splits into a linear factor and a quadratic factor which is
then reducible in Q(d\/).
In the first case, it is shown that
(y + t3dk) = a3,
where a is a non-principal ideal. Then h = 0 (mod 3), a contradiction.

In the second case, it is obvious that
y + t3dk= (r + s ~ ' / k ) ~ / n ~ ,
where r, s, n are integers and (r, s, n) = 1.
This gives
n3t3 = s(3r2 + ks2), n3y = r(r2 + 3ks2).
The discussion of the first of these leads to an equation
y: = X: - 21ktf
with an associated cubic equation
V3 - 3.Ylt7 - 2yl = 0
with discriminant dl = (54t,3)2k.
Here again two cases arise according as the cubic is reducible or irreducible
in the field Q(d\/k).
In the first case the method of infinite descent applies and
so no solution is possible. In the second case, it is shown by results on the
relative discriminants of algebraic number fields that h = 0 (mod 3), a contra-
diction.
The second general method lo*11*12is illustrated by the following Theorems
7, 8, and 9.
Theorem 7
+
The equation y 2 = x3 k has no rational solutions ifall the following con-
ditions are satisfied.
I. k is positive and square free and k 3 2 or 3 (mod 4) and k = - 2 or
- 4 (mod 9).
11. The number h of classes of ideals in the quadraticfield Q(d-) is not
divisible by 3.
111. Thefundamentalsolution (X,Y ) = ( U , T ) of the equation Y 2 - k X 2 = 1
satisfies the conditions
111'. i f k = -2 (mod 9), then U = f 3 (mod 9), andso T = & 1 (mod 9),
111". i f k = - 4 (mod 9), then either U = f 3 (mod 9) and so T =
f 1 (mod 9), or U = + 4 (mod 9) and T 3 5 3 (mod 9).
Instances are k = 7 since h = 4, and U = 3 , and k = 14 since h = 4,
U = 4, T = 15.
Theorem 8
+
The equation y2 = x3 k has no rational solutions i f the following con-
ditions are satisfied.
I. k is negative and square free, k = 2 or 3 (mod 4), and k = 2 , 4 (mod 9).
11. The number of classes of ideals in the quadraticfield Q ( d k ) is not divisible
by 3.
+
111. Thefundamentalsolution ( X , Y ) = ( U , T ) of theequation Y 2 3kX2 = 1
satisfies the condition U f 0 (mod 3).
Instances are given by k = - 5 , -14, -34, -41 with U = 1,2, 10, 11,
respectively.
Theorem 9
The equation
y a = x3 + k
has no rational solutions if the following conditions are satisfied.
I. k is negative, k is square free, k = 2, 3 (mod 4), k 5 3 (mod 9).
11. The number h of classes of ideals in thequadraticfield Q ( d z ) i snot divisible
by 3.
+
111. Thefundamentalsolution ( X , Y ) = ( U , T ) of the equation Y 2 +kX2 = 1
satisfies the condition, either U = & 3 (mod 9) or T = f 3 (mod 9).
26. THE EQUATIONya = x3 +k 25 1
Instances are given by k = -33, -42, -69, -78, - 105, - 144 corre-
sponding to U = 3, T = 15, T = 24, T = 51, T = 6, U = 6, respectively.
The method will be sufficiently illustrated if we prove Theorem 8.
It suffices to consider the integer solutions of
y2 = x3 + kZ6, z # 0, (x, y , Z ) = 1,
and suppose that z has its least positive value.
Then with integers A , B and ( A , B ) = 1,
y + z3dk= ( A + B d k ) 3 , z3 = B(3Az + kB2).

Suppose first that B is prime to 3. Then
B = BY, 3A2 + kBf = z?.
As a congruence mod 9 this gives

3A2 +k = z;,
and is impossible if k = 2 or 4 (mod 9), since z: $ 2, 4, 5, 7 (mod 9).

Suppose next that B is divisible by 3. Then clearly
B = 9BY, A2 + 27kBf = z?.
Write this as
(A + 3B:d-)(A - 3B7d-3k)
~~
= 2:.
Then ( z , , 6 ) = 1 since ( A , B , ) = 1 and so the two factors are relatively prime.

From the lemma, the class number for Q ( d X )is prime to 3, and so
A +3 B : d x = (T + Ud-k)"(C + Dz/?E)3, (C,D ) = 1, CI = 0, f 1.
Suppose first that CI = f 1. Then from a congruence mod 3-k
applied to the irrational terms,
-. .
0 = C3Ud-3k.
Now C $ 0 (mod 3) since A $ 0 (mod 3), and so U = 0 (mod 3), and this
has been excluded.
Suppose finally that CI = 0. Then
B; = C 2 D - kD3.
Hence D = D:, C 2 - kDy = B;, and this is the same as the original
equation for z .
Also D , = <ID < 3'B: < g B / 9 < zIV9, and so this contradicts the defini-
tion.
6. Further application of algebraic number fields

The finite basis theorem for cubic curves given in Chapter 16 applies to the
special equation
y2 = x3 + k, (21)
and so depends on the consideration of the cubic field Q(V/k). A simpler
treatment has been given by PodsypaninL3 by using the quadratic field
Q(dE),and we give an outline of his method. It is of special interest since it
associates with equation (21) the related equation
y2 = - 27k, (22)
as had been previously done by Fueter. Thus if P ( x , y ) is a solution of equation
(21), then Pl(xl,y,), where
x3 + 4k y(x3 - 8k)
X1 = - y1 = > (23)
X2 x3
is a solution of (22). Applying this result again to equation (22), it is easily
shown that a solution of (21) is given by Pz(x2,y,) where
x, =
X? - 108k
Y, =
yl(x? + 216k)
x? x::
This gives a solution of equation (21) if (xl,y,) is any solution of (22) even
though this is not derived from (21) by (23).
Denote by G the additive group of the rational solutions of equation (21)
and by H that of (22). Then if P , is derived from P by (23) and (24), it follows
that P , = 3P,, and so we have an inclusion relation
G 3 H 3 3G.
An equivalence relation for solutions of equation (21) now arises. It can be
shown that ( x , y ) is a point 3P if and only if
y + dk = a3, (25)
where a is a number in the field Q(d/k). Further, let P,(x,, yl), P2(xz.y,) be
any rational solutions of (21), and denote by P12the point P , + P,. Then
(Y, + dF)(~2 + dk) = ~ ~ ( +~ dk),

1 2
where PXl2 = XlYZ - X2Y1 + (x1 + dJ.

For let x = Ay + B be the equation of the line joining P1,P,. Then identically,
Y 2 - k - (AY + = - A 3 ( y - YlMY - Yz)(Y + Y12)a
The result follows on putting y = - dk.

26. THE EQUATION 4 = 5 f k 253
I t can then be shown that the solution P12 is a solution of equation (22)
derived from (21) if and only if
(y1 + d Q ( y 2 + dk)= y3, (26)

where y is a number in the field Q(d/k>.
The relations (25), (26) define an equivalence process. In this way, Podsy-
panin finds for Ikl < 90, the number of fundamental solutions and their
values.
Several other authors have treated the equation by using cubic fields and
Weils ideas. Billing l4 has dealt exhaustively with the problem of finding the
generators and has given them for Ikl < 25.
The equation (21) was considered afresh by Cassels l 5 who applied the ideas
of Weils proof. Cassels replaces (21) by
y2 = x3 + kZ6,
where .Y, y , z are integers. From this is deduced in the usual way
x - klVk, = paz, (27)
where p > 0 is an integer in Q(Vk)taken from a finite set and 6a is an integer
in Q(Vk).Several cases are considered depending upon the parities of x, y , z.
Then equation (27) is taken as a congruence whose modulus is an ideal factor
of 2 and many special results are found. In this way, he finds for Ikl < 50, the
number of generators and solutions of the equations.
His work was amplified and completed by Selmer 17.
The most extensive tables have been given by Birch and Swinnerton-Dyer
who have found the number of generators for all Ikl < 400 except -309.
Their method depends upon recent ideas as expounded in Cassels l9 report
and requires a detailed discussion of the reduction of binary quartic forms,
both locally and globally. Cassels found the number of generators for
k = -309.
A surprising result has been found very recently by Bakerz0 who shows
that all the integers solutions of the equation (1) satisfy
max (1x1, l y l ) < exp (10101k1104).
This depends upon his new proof of Thues theorem which relies essentially
on the study of linear forms in the logarithms of algebraic numbers.
REFERENCES
1. L. J. Mordell. A Chapter in the Theory of Numbers. (1947). Cambridge
University Press.
2. L. J. Mordell. The diophantine equation y2 - k = x3. Proc. Lond. Math. SOC.,
(2) 13 (1 913), 60-80.
9*
2 54 DIOPHANTINE EQUATIONS
3. Marshall Hall. Some equations y 2 = x3 - k without integer solutions. J .

Lond. Math. SOC., 28 (1953), 379-383.
4. 0. Hemer. On the diophantine equation y 2 - k = x3. Diss. Uppsala (1952).
See also Notes on the diophantine equation y 2 - k = 2.Arkiv for Mat., 3
(1954), 67-77.
+
5 . T. Nagell. Einige Gleichungen von der Form ay2 by + c = dx3. Vid. Akad.
Skrifter Oslo, Nr. 7 (1930).
6. W. Ljunggren. Einige Bemerkungen uber die Darstellung ganzer Zahlen
durch binare kubische Formen mit positiver Diskriminante. Acta Math., 75
(1942), 1-21.
6a. M. Lal, M. F. Jones, W. J. Blundon. Tables of solutions of the diophantine
equation y3 - x2 = k. Dept. of Maths., Memorial University of Newfoundland,
St. Johns, Newfoundland, (1968)-and 6b, Numerical solutions of the diophan-
tine equation y3 - x = k . Math. Comp., 20 (1966), 322-325.
7. W. Ljunggren. On the diophantine equation y2 - k = x3. Acta Arith., 8
(1963), 451-463.
8. R. Fueter. uber kubische diophantische Gleichungen. Commentarii Math.
Helv., 2 (1930), 69-89.
+
9. L. J. Mordell. The infinity of rational solutions of y2 = x3 k . J. Lond.
Math. SOC.,41 (1966), 523-525.
+
10. L. J . Mordell. On some diophantine equations y2 = x3 k with no rational
solutions. Archiv math. natur., 49 (1947), 143-150.
+
1 1 . K. L. Chang. On some diophantine equations y 2 = x3 k with no rational
solutions. Q. J . Marhs., 1 (1948), 181-188.
+
12. L. J. Mordell. On some diophantine equations y 2 = x3 k with no rational
solutions. Abhaiidlungen aus Zahlentheorie und Analysis zur Erinnerutig an
Edmund Landau (1968), 225-232. V E B Deutscher Verlag der wiss. Berlin.
+
13. V. D. Podsypanin. On the indeterminate equation x 3 = y2 Azs. (Russian).
Mat. Sbornik N.S., 24 (1949), 391463.
14. G. Billing. Beitrage zur arithmetischen Theorie der ebenen kubischen Kurven
vom Geschlecht Eins. Diss. Uppsala (1938).
15. J. W. S. Cassels. The rational solutions of the diophantine equation y 2 =
x 3 - D. Acta Math., 82 (1950), 243-273.
16. E. S. Selmer. The diophantine equation y 2 = x3 - D. A note on Cassels
method. Math. Scan., 3 (1955), 68-74.
17. E. S. Selmer. On Cassels conditions for rational solubility of the diophantine
equation y2 = x3 - D. Archiv marh. natur., 53 (1956), Nr. 7.
18. B. J. Birch and H. P. F. Swinnerton-Dyer. Notes on elliptic curves. I . J. rehe
angew. Math., 212 (1 963), 7-25.
19. J . W . S. Cassels. Diophantine equations with special reference to elliptic
curves. J. Lond. Math. SOC.,41 (1966), 193-291.
+
20. A. Baker. The diophantine equation ya = x3 k . Phil. Trans. Roy. SOC.,
London, 263 (1968), 193-208.
CHAPTER 27
The Equation y2 = ux3 + bx2 + cx +d

1. The study of this equation has led to important developments and great
advances in diophantine analysis.
The most important result 1 * 2 about its integer solutions is given by
Theorem 1
The equation
ey2 = ax3 + bx2 + cx + d, a # 0, (1)
where the constants are integers and the right-hand side has no squared linear
fuctor in x,has only afinite number of integer solutions.
Multiply by 81e, and put y , = 9ey and x1 = 3x. The equation takes the
form
y ; = als: + 3b,xf + 3ClXl + d , ,
Put t = 2u,y1, s = a,xl + b,. The equation becomes, say,
t2 = 4s3 - g,s - g , (2)
or (4t)2 = ( 4 ~ -) 4gZ(4s)
~ - 16g3.
There are only a finite number of integer solutions for s, t . For in Chapter 25,
we have seen that all the integer solutions of
2 2 = x 3 - G J Y 2 - G3Y3 ( X , Y ) = 1,
where the right-hand side has no squared linear factors, are given by
X = H ( p , q), y = F(P, 91,
where F ( p , q) is a representative of the finite number of binary quartics with
invariants
g2 = 4Gz, g3 = 4G3.
Then by Thues theorem, there are only a finite number of integer solutions of
F ( p , q ) = 1.
We can show more directly that the equation
t 2 = 4 3 - g2.Y - g3, gi + 27& # 0,
has only a finite number of integer solutions.
With the solution (s, t ) , we associate the binary quartic written with binomial
coefficients,
f(X>Y ) = ( 4 0 , -s, t, e>(x,Y)', (3)
where e = g, - 3s2. Its invariants are actually g,, g, since
g, = e + 3s2,
1 0 -s
-s t e
The quartic (3) is equivalent to one of a finite number of binary quartics

(u, b, c, d, e)(x, y)' with invariants g,, g,. Hence
(1,0, -s, t , 4(x, u)' = (a, b, c, 4 e)(px + ry, qx + v)',

where p , q, r , s are integers and p s - qr = 1, and so
1 = (a, b, c, 4 e)(p, q)',
and by Thue's theorem, there are only a finite number of integer values for
p , q. Further r, s are uniquely determined in terms of p , q since
Hence there are only a finite number of values for r, s.

There is an alternative method of dealing with equation (1).
Write it as
ey2 =f(x) = a(x - e,)(x - e,)(x - 0,).
Then the 8, which are roots off(x) = 0, define three fields Q(Ol), Q(8,), Q(8,)
whose nature depends upon the reducibility off(x). From the general theory
of Chapter 15, we have for each 0 and the corresponding field Q(8), a relation
x - 8 = m(p + 48 + re2),, (4)
where p , q, r are rational integers and m takes only a finite number of values
in Q(8), and can be written asp, + q,B + r,02. Hence from equation (4) on
replacing 8', O3 in terms of P, 0, 1, and equating coefficients of 0, O2 on both
sides, we have two equations of the form
fdp, 4 , r ) = 0, ~ A Pq,, r ) = 1,
where fi,f2 are homogeneous quadratic forms, The solutions of the first of
these is given by a finite number of expressions
p = j&t, 4, q =j h ,4 r = j&, 4,
27. THE EQUATION y 2 = ax3 + bx2 + c x + d 257
where the j are binary quadratics in the integers u, u. Substituting in the
second equation, we have a finite number of equations h(u, 1:) = 1 where h is a
binary quartic. Since equation (1) is of genus one, h(u, v ) cannot be a perfect
square. Then by Thue's theorem, there are only a finite number of integer
solutions for u, u and so for x, y .
From the particular case
y2 = X(UX~ + bx + c>, b2 - #
~ U C 0,
it is obvious that the equation
y2 = ax4 + bx2 + c
has only a finite number of integer solutions.
2. The problem of finding all the integer solutions of an equation ( 1 ) is in
general a very difficult one, and only a few results have been found. An
interesting one3 is given by
Theorem 2
The equation
y(y + 1) = x(s + l)(x + 2) (5)
has only the integer solutions x = - 1, - 2 , 0 , I , 5 .
Hence 0, 6 and 210 are the only integers which can be expressed as both a
product of two and three consecutive integers.
+
On replacing 2x 2 by x and 2y +
1 by y , the equation becomes
2y2 = x3 - 4x f 2, (6)
and we have to prove that its integer solutions are given by
x = 0, - 2 , 2 , 4 , 12.
The equation requires the study of the cubic field Q(8) where
83 - 40 + 2 = 0.
We require the following arithmetical properties of the field Q(0).
1. The integers are of the form x = a + +
b8 cd2, where a, b, care integers.
11. There are two fundamental units E = 6' - I , 7 = 28 - 1.
111. Unique factorization exists in Q ( 0 ) since the number of ideal classes is
one.
We note that 2 = where 5 is a unit and it easily follows that 8 is a prime
in Q(0).
Write equation (6) in the form
(x - e)(x2 + ex + 82 - 4) = 103y2.
Then 0 is a common divisor of the two factors. Also the prime 48 - 3 with
norm 37 is a possible common divisor since
382 - 4 = (383 - 4eye = (se - 6 ) p = 2(4e - 3ye.
Hence x -e = 5 q4e - 3 y 4 y a + be + ce2)2,
where n is an integer 2 0 and I, m are any integers, positive, negative or zero.

If n is odd, on taking the norm of both sides, we find the impossible con-
gruence
x3 - 4x +2= k2.37y2 (mod 8).
If n is even, (48 - 3)" can be absorbed in the square term. Hence it suffices to
consider the four sets of simultaneous quadratic equations arising when
( I , m) = (0, O), (LO), (0,I), (1, 1).
These are given explicitly on noting that
(a + be + c02)2 = a2 - 4bc + O(2ab + 8bc - 2c2) + P(b2 + 2ac + 4c2).
Fortunately these equations can be dealt with by elementary considerations
and congruence conditions, and then the desired result is obtained.
In problems of the present type, the simultaneous quadratic equations may
lead to a large number of biquadratic equations. These may require a detailed
study of arithmetical properties of many quartic fields. In general the equa-
tions can only be dealt with by thep-adic considerations of Chapter 23 applied
to the units of the corresponding quartic fields, and this may involve con-
siderable numerical work. Thus Avanesov4 has proved a conjecture put for-
ward by Sierpinski, here given as
Theorem 3
The only positive integer solutions of the equation
$y(y + 1) = &x(x + l)(x + 2) = n
are given by n = 1, 10, 120, 1540,7140.
The first interesting equation to be completely solved is given in
Theorem 4
6y2 = X(X + 1 ) ( 2 ~+ 1)
are x = 0, -1, 1, 24.
This is Lucas' problem of the square pyramid. Watson5 gave a very compli-
cated solution depending upon elliptic functions.
A proof based upon arithmetical considerations has been given by
Ljunggren8. It depends upon a Pel1 equation in a quadratic field. A quartic
27. THE EQUATION y 2 = ax3 + bx2 + cx + d 259
field arises all of whose conjugate fields are imaginary. There is one funda-
mental unit and its properties must be investigated in detail.
Problern
Are the only integer solutions of the equation
6y2 = x3 + 5~ + 6 = ( X + 1 ) ( 2 -x + 6)
given by
x = - 1,0,2, 7 , 15, 74?
The equation arises from
y2=1+x+-
x..x - 1
2!
+ x . x -3!1.x - 2
3. Congruence considerations7 enable us to construct many classes of
equations
y 2 = ax3 + bx2 + cx + d
for which bounds can be given for the magnitude of the solutions. Equations
can also be given with only one or two integer values for x.
Theorem 5
Let a > 0, 6, c be arbitrary integers and let d be odd and hare no prime
factors = 3 (mod 4). Then the equation
y2 = 2ax3 + (6 - 2a - 2c + 8b)x2 + 2cx - d2 (7)
can only hare integer solutions for uihich x < 0, x = 3 (mod 4), and so these
are bounded in ternis of a, b, c, d.
In particular, the equation
y 2 = 2ax3 + (4 + 4b)x2 + 2(1 - a + 26)x - I (8)
has only the solutions ( x , y ) = (1, & 1 ) $ a > 0, 3a > 1 + b; and the equation
y 2 = 2ax3 + (12a + 8)s2 + (10a + 6)x - 1, (9)
has only the solutions (.Y, y ) = (- 1, rf: l ) , (- 5, -t 13) $ a > 0.
Clearly in equation (7), x is not even since y 2 $ - 1 (mod 4). Also x $
1 (mod 4 ) since then
y 2 = 2a + 6 - 2a - 2c + 86 + 2c - 1 = 5 (mod 8).
Hence s = 3 (mod 4) and since x I y 2 + d2, x must be negative.
If in equation (7) we take .Y = -1, y = 1, d = 1, then c = 1 - a + 2b,
then c = 1 - a + 26, and we have with a > 0,
y2= 2ax3 + ( 4 + 4b)x2 + 2(1 a + 2blx - 1.
-
Any further solutions must satisfy x < -5. We show then that y2 < 0.
The condition for this can be written as
(X + 1)(2ax2+ (4 - 2~ + 2b)x - 2) < -1.
This holds if
IOU - (4 - 2a + 26) > 0, i.e. 3a > 1 + 6.
Another solution x = - 5 of equation (8) arises if b = 3a + 1 , and then
with a > 0,
y 2 = 2ax3 + (12a + 8)x2 + (IOU + 6)x - 1.
Any further solution of this must satisfy x 6 -9.
Write y2 -1 = (x + 1)(2ax2 + (10a + 8)x - 2).
+
Then y 2 < 0 i f ax2 (5a + 4)x - 1 > 0. This is true if 9a > 5a + 4 or if
a > 1. If a = 1 , the equation becomes
yz = 2x3 + 20x2 + 1 6-
~ 1.
Then x # -9, since y 2 $ - I (mod 3). Another solution would satisfy
x 6 - 13. This is impossible since then
y2 = 2x(x2 + lox + 8 ) - 1
+
and x2 lox + 8 > 0.
The method is capable of considerable generalization8. Suppose that the
cubic equation ( I ) can be written in the form
y2 = ax3 + bX2 + cx + d = g(x)h(x)- k f 2 ( x ) (10)
wheref(x), g(x), h(x) are polynomials with integer coefficients and k is a
square-free integer. Simple results arise if the factors of k are taken from the
product 2.3.5.7. Then conditions are imposed upon the coefficients of f ( x ) ,
g(x), h(x) so that h(x) cannot be a divisor of y 2 -t kf2(x),unless h(x) < 0. If
h(x) = x2" + hlxZn-l+ . . an upper bound is easily found for the magni-
+ ,
tude of x in a solution (x, y).

A simple illustration of equation (10) is given by
y2 + 2(x2 + f i x + f,), = (2x2 + g,x + g2)(x2 + h,x + h d ,
where g , = 0 (mod 8), g , = 5 (mod 8). Then g(x) = 2x2 + g,x + g , = 7 or
5 (mod 8) according as x is odd or even, and so if g(x) > 0, then - 2 is a quad-
ratic non-residue of g(x). Now suppose g(x)and h(x) have no common factors
=5,7 (mod 8). The condition for this is easily found since any common
divisor must divide
(.f? - 4fz)(g? - 8g2) - ( 4 h 4- 2g2 - higi)2.
27. THE EQUATION y 2 = ax3 + bx2 + cx + d 26 1
Hence solutions of the equation can occur only when g(x) < 0. This means
that an upper bound for x can be found from the roots of g(x) = 0.
The question arises whether such a result exists for all equations (1).
It has been noted in Chapter 22 that Baker has found bounds (very large
ones indeed) for the magnitude of the solutions of equation (1).
REFERENCES
1. L. J. Mordell. Note on the integer solutions of the equations Ey2 = Ax3 +
Ex2 + Cx + D.Messenger Maths., 51 (1922), 169-171.
2 . L. J. Mordell. On the integer solutions of the equation ey2 = ax3 bx2 + +
cx+ d. Proc. Lond. Ma/h. SOC.( 2 ) , 21 (1923), 415-419.
3. L. J. Mordell. On the integer solutions of y ( y + +
1) = x(x l)(x + 2).
Pacific J . Maths., 13 (1963), 1347-1351.
4. E. T. Avanesov. Solution of a problem on polygonal numbers (Russian).
Ac/a Arith., 12 (1967), 409419.
5. G . N. Watson. The problem of the square pyramid. Messenger Maths., 48
(1919), 1-22.
6. W. Ljunggren. New solution of a problem proposed by E. Lucas. Norsk Mat.
Tidsskrift, 34 (1952), 65-72.
+
7 . L. J. Mordell. The diophantine equation y 2 = a x 3 6 . 2 cx+ + d or fifty
years after. J . Lnnd. Mu/h. Soc., 38 (1963), 454458.
+
8. L. J. Mordell. The diophantine equation y 2 = ax3 6x2 c x + + d. R . Circ.
Mat. Palermo (II), 13 (1964), 1-8.
CHAPTER 28
Some Equations of Degree > 3
1. We show now that some general classes of equations of degree > 3 have
only a finite number of integer solutions. We commence with a result * due to
Runge l.
Theorem 1
Let f ( x , y ) be a polynomial of degree n with integer coeficients which is
irreducible in the rational field Q, and suppose that the homogeneous part
fn(x, y ) of degree n of f ( x , y ) is reducible in Q but is not a constant multiple of a
power of an irreducible polynomial. Then the equation
f(x, Y ) = 0 (1)
has only afinite number of integer solutions.
For simplicity, we shall deal only with the case when the linear factors of
fn(x,y ) are all different. We may suppose that at least one linear factor has real
coefficients as otherwise the theorem is trivially proved. Let a! be a root of the
equation f(1, t ) = 0, and so a! is of degree n , c n if f ( x , y ) is of degree n in y.
On writing f(x,Y ) = (v - axlfn- dx, r) + g(x, Y ) ,

where g(x, y ) is of degree < n - 1, we see that we have for large values of x a
convergent expansion for y in descending powers of x given by
where a,,,al,.. . are elements of the field Q(a).

Suppose now that the equation (1) has an infinity of integer solutions. For
these, (2) will hold with an appropriate a. We shall show that from (2), we
can deduce a polynomial equation for x, y , say h(x, y ) = 0, which is of degree
< n in y and so h(x, y ) = 0 is independent of f ( x , y ) = 0. Hence these two
equations give only a finite number of values of x and y.
Consider the polynomial
2 2
P nl-1
h(s, y ) = cp,qxpyq, (3)

p=o a=o
* See Addendum 1 .
28. SOME EQUATIONS OF DEGREE >3 263
where we have at our disposal the integer P and the n,(P 1) integer con- +
stants cP,*.On substituting from equation (2) in (3), we have an expansion
2
-03
h(x,y) = drY.
rSP+nl
The coefficients d, can be expressed linearly in terms of the c ~ and , ~

a,, a,,. . ., a ( , ~ - ~We
) . now impose the d, = (n, - 1)(P n, + +
1) conditions
on the cP,*that the coefficients of a , a2,. . . , an,- in d, for r = P n,, . . ., 1,0 +
should be zero. This will be possible if
nl(P + 1) > (n, - 1)(P + n, + l ) ,
or P 2 n: - n,.
Then we have an identity
+ n1
2 2 drxr.
P
h(x, y ) - erxr =
T=O T<O
As the left-hand side is an integer, it must be zero since x is large. This

completes the proof.
The most general result on the existence of only a finite number of integer
solutions for an irreducible polynomial equationf(x, y ) = 0 is due to Siege],
in 1929. He proved the
Theorem 2
There are only a j n i t e number of solutions i f the equation f ( x , y ) = 0 is not
identically satisjed by a parametric representation,
x = AIL", y = B/L", or x = C/Qn, y = Die",
where A , B, C , D are polynomials in t with integer coeflcients and L is a linear
polynomial and Q is an indefinite quadratic polynomial.
The proof is of a very advanced character.
2. We now consider some equations of the form
y 2 = aox" + alx"-l + . . + a,,
(4)
+
where the a's are integers.

By applying the methods of Chapter 24, it is a simple matter to construct
equations of the form
y2 + k = aOx2"+l+ alxZn+ . . . + u2,x =f(x),
say, where a, > 0, which have only a finite number of integer solutions and
for which an estimate in terms of the coefficients can be found for the magni-
tude of the solutions. Suppose for simplicity that k is a prime = 7 (mod 8) and
that a,, $ 0 (mod k), a,, # i-1 . Clearly x f 0 (mod k). Denote by r the
quadratic residues rlr rz, . . . , r(k-1),2 of k and by n the quadratic non-residues
n l , n29 . . ., n(k - I)/,*
Suppose first that s is congruent to a quadratic residue, say .Y = r . This will

be impossible iff(r) is a quadratic non-residue of k , i.e. if
+ alr2, + . . . + aZnr= n (mod k),
aorZn +
where n is an arbitrary quadratic non-residue. If 2n + 1 > (k - 1)/2, these

congruences have a solution for the a.
Suppose next that x is congruent to a quadratic non-residue n. Let us
assume first that x > 0. Let x = 2"x, where x1 is odd and is also a quad-
ratic non-residue of k since 2 is a quadratic residue. Then from equation (5),
This contradiction shows that x < 0. Then obviously from (9,x must be
bounded and so can be easily found.
Results can also be found for equation (4)when n is even. Suppose for
example that
ya + ka = (axz + b)(aoxan+ . . . + a 2 n - 1+~ aZn)= flf,,
say, where k has no prime factors = 3 (mod 4), and
a = 1 (mod 4), = -2 (mod 8), 2azn-1+ a,, = 3 (mod 4).
b
Then if x = 1 (mod 2), fl = - 1 (mod 4); if x = 0 (mod 4), 3fi = - 1 (mod
4); if x = 2 (mod 4), fa 3 3 (mod 4). Hence the only possible solutions must
satisfy fl < 0, f, < 0 and so are finite in number,
We now prove an earlier Siegel's4 result given by
Theorem 3
The equation
y 2 = aoxn + a1xn-' + . . + an a
(5)
has only ajinite number of integer solutions * if the right-hand side has at least
three direrent linear factors.
Write y2 = ao(x-a,) ( x - a 2 ) . ( x - a,),
where we suppose the a define a field K and that al,a2, a3 are all different.
The general theory of Chapter 15 gives
x - a1 = 1.161, x - a2 = 1.,t:, x - a3 = 1.36&
where the 5 are integers in K and the p are a finite set of numbers in K .
Hence pz5% - p362 = a3 - a2 # 0,
p35: - 1.16: = a1 - a3 # 0,
PI61 - pzt; = az - a1 # 0.
* See Addendum 2.
d&) = K ' , say,

Then in the field K(d/.1, da,,
tZdE - t 3 G = B1.L
534i-G - E I G = 132&
t,dZ - t 2 4 G = s34,
where I is an arbitrary large integer, el, cZ, e3 are units in K' and PI,p2, P3 are
numbers in K' belonging to a finite set. Then
PI.: +sz4+ s3.5 = 0,
or (R) ();! + (k)):( 1

+I = 0.
By Siegel's Theorem5 2, this has only a finite number of solutions for

and EJe,, and then for t4/t3and t2/t3 and then for El, E2, E3.
A particular case for which a simpler proof is known is given by Landau
and Ostrowski6, and Thue7 as
Theorem 4
Ifa, b, c, d a r e integers, and ad # 0, b2 - 4ac # 0, n B 3, the equation
ay2 + by + c = dx",
has only afinite number of integer solutions.
+
On writing y for 2uy b, the equation takes the form
y2 - k = Ix", kl # 0.
Suppose first that k is a perfect square, say k = m2.
Then (y + m)(y - rn) = Ix".
Since y + m, y - m have only a finite number of common factors, we have
with integers Y , 2,
y +m =fly", y - m = fzZ", x = fYZ,
wheref,,,f,,f belong to a finite set of integers, and f l f2 = If". Then
fly" - fiZn = 2m.
By Thue's theorem, there are only a finite number of integer values for Y, Z.
Suppose next that k is not a perfect square.
Then (y + d E ) ( y - a)= Ix".
Then the results in Chapter 15 give at once
f(y + d k ) = ( g i-
hl/k)(u + d/k)",
wheref, g, h are integers taken from a finite set, and u, t' are integer variables.
From this
2fdk = (g + hdE)(u + vdEy - ( g - hdk)(u - UdE)".
This reduces to a binary equation in u, v, say f(u, u) = m, where f(u, u) is
clearly not a power of another form. Again by Thue's theorem in Chapter 22,
there are only a finite number of integer values for u, u.
We give another proof, essentially Thue's, which is really independent of
algebraic number theory. We may suppose that k is not a perfect square
and so x # 0.
We require the
Lemma
Integers (q, r ) # (0,O)exist such that
Proof obvious from Minkowski's theorem on linear forms.

Write s = qy - rx. Then Is1 < d/Tsll,s = qy (mod x).
Put t = (s2 - kq2)/x".Then t # 0 since k is not a perfect square and q # 0.
Also t is an integer since s2 - kq2 = q2y2 - kq2 = 0 (mod x). Next t is
bounded independently of the values of x, y , and so is restricted to a finite
set. For
Write now A + B d k = (S + qdL)"(y - dk)/x".

We prove that A , B are bounded integers independent of x and y. For
s + qdk = + dE)- rx,
q(y
(s + qdk)" = ( - ~ ) ~ + r ~( yx+~ dE)(c+ N E ) ,
where C , D are rational integers. Since ( y + d / k ) ( y - d/k)= Ix", A and B
are integers. Next for both signs,
] A f Bdkl 6 (dm+ dmdm)"(l+ d m ) l y l / l x l " .
But IyI < lk11/2+ 1Z11/21x1"126 (lk11/2+ 1Z11'2)Ix1"'2
and so [Ak Bdkl 6 (1 + dNy+l(lkl1la+ ll11/2).
Hence A and B are bounded.
Finally
(A + BdE)(s - qdky = (8' - kq2)"(y- &)/x" = t(y - dE).
Then
(A + B d k ) ( s - qdk)" - ( A - B d k ) ( ~+ qdk)"= -2tdk.
There are only a finite number of such equations for s, q since A , B, t are
bounded. Since the left-hand side is not the power of a polynomial, Thue's
theorem applies.
In general it is not easy to find the integer solutions of equation (4), or to
prove that there are none. However, Baker has recently given an estimate for
the magnitude of the solutions. An interesting example due to Riggee and
Erdoss is given by
Theorem 5
The equation
y 2 = x(x + l)(x + 2 ) . . .(x + n) = f ( x ) , (7)
say, has only the integer solutions giaen by y = 0.
On excluding these values, we easily deduce that
x + r = a , x , 2 ( r = 0 , 1 , ..., n - 1 ) ,
where the a are square free and have as factors only primes < n . Arguments
rather different from those usually employed for diophantine equations lead
to a proof. The first stage is to show that the a are all different. We do not
proceed beyond this stage since a great deal of detail is involved. Further the
proof shows first that the theorem holds for n > 100, and then the remaining
values of n require special treatment.
We first prove Oblcith's l o result that x > n2. Suppose first that x < n. By a
theorem of Tchebycheff, there is a prime p such that
x + n > p 3 $(x + n) 3 x,
and so p but not p 2 dividesf(x), and this is impossible.
Since x > n, then by a theorem of Sylvester and Schur,f(x) has a prime
+
factor q > n. Then q2 I x r for some r < n - 1, and so
x+r (n + 1)2, x > r2.
Suppose next that the a's are not all different, say, a, = a, where r > s. Then
n > a,x,2 - a,x," = a , ( x ~- x,") > 2a,s,
3 22/a,X,2 = 2d(X + s) > di,
and this is not so.
A more general result is
Theorem 6
The equation
y" = x(x + 1 ) . . .(x + n - 1)
has only the integer solutions gioen by y = 0.
A proof for m = 2 was found by Erdos," and for all m,by Erdos and Sel-
fridge, but this has not been published. Previously many special results had
been proved by several authors, for example, Obldth'O, Rigge12, Erdos 13,
Johnson 14. These in general require too much numerical detail for any account
to be given here.
3. A few words may be said about the integer solutions of the quartic equa-
tion
ky2 = ax4 + bx3 + cx2 + dx + e. (9)
By Siegel's Theorem 3, there are only a finite number of integer solutions
when the right-hand side has no squared linear factors. Apart from this result,
very little is known about the equation, and it is surprising what a variety of
methods, some very complicated, have been employed in dealing with it. It
had seemed impossible to find all the solutions except in particular cases.
Recently Baker has formed an estimate (very large) for the magnitude of the
solutions.
A simple instance 14a of equation (9) is the equation
y2 + k2 = ( p x 2 - p - 1 - 4q)(rx2 - s),
where k has no factors = 3 (mod 4), p > 0, q > 0, s > r > 0. When the
two factors in x are positive, the first excludes x = 1 (mod 2), and the second,
x = 0 (mod 2) if s = 1 (mod 4), and so solutions can arise only when these
factors are negative. It is easy to construct equations with two values for x2,
but not with three values. In particular, the equation
y2 + 1 = ( 4 2 - 17)(60x2 - 1025)
has only the integer solutions x = 0, & 1, f 2.
An interesting example is given by
Theorem 7
The only positive integer solutions of the equation
are (x, y ) = (1, I), (2,2), (4, 9). The others are x = 0, 1, y = 0, 1 and
( x , ~ )= (-1, - 1 ) (-3, -8).
This means that the only triangular numbers which are squares of triangular
numbers are 0, 1, 36.
A complicated proof was given by Ljunggren l5 depending upon the p-adic
methods of Chapter 23 applied to a quartic field. On putting 2x - 1 = X ,
2y - 1 = Y , we have
Y2=
xz-1
2(-) + 1.
He uses the unit 1 + d? to construct integers in Q(V'z)with relative norm 9
over Q(d/.,. He then deduces an exponential equation in only one variable
and shows that the only positive solutions are (A', Y ) = (I, 1) and (7, 17).
A simple proof was given by Cassels"j depending upon the properties of
the quartic field Q ( v - 2 ) .
Special cases of equation (9) when h = d = 0 have been considered in
Chapter 8, and also when b = c = d = 0. Then the equation becomes
ky2 = ax' +e (10)
which is of special interest. Results of various degrees of generality have been
found by many authors, for example, Ljunggren Mordel121,Cohn22,
17918*19*20,
Podsypanin 22a (cf. Ljunggren la) especially for the equations.

y2 = Dx4 f 1.
Sometimes elementary methods suffice for the rational solutions as was
shown in Chapter 4. We now prove similarly
Theorem 8
The equation
y2 = Dx' +1
has no integer solutions except x = 0 if
I. D f 0, - I , 3, 8 (mod 16).
11. D has no factorization D = pq, ( p , q ) = 1 wherep is an odd number > 1
such that either
p = 21, or p E 4 f 1, or p = 49 f l(mod 16).
We may suppose s > 0. Clearly equation (1 1) cannot be satisfied by odd
values of .Y since D +
1 $ 0, 1, 4, 9 (mod 16), and so x is even. We now
consider that solution of (1 1) for which x has its least positive value. Then
y f I = 2pa4, y T 1 = 8qb4, x = 2ab, (12)
and ( p , q ) = I , a = p = I (mod 2 ) for some decomposition D = pq.
Suppose first that p = 1, and so from equation (12)
a4 - 4Db' = f 1. (13)
The negative sign can be rejected and so
0' - 1= 4Db'.
Then with D = rs, b = cd,
a2 + 1 = 2rc4, a2 - 1 = 2sd4,
and so c = r = I (mod 2).
Also rc4 - sd4 = 1.
Suppose first that r = 1 and so

c4 - Dd4 = 1. (15 )
Then a solution of equation (1 1) is given by x = 2acd. Since x # 0, d # 0,
then d < +x, and so equation (15) gives a contradiction.
Suppose next that r > 1 in equation (14). Then d cannot be even since
from 11, r f f 1 (mod 16). Also dcannot be odd, for then from (14), r - s E
1 (mod 16), and this is excluded by 11.
Suppose finally that p > 1. Then from equations (12),
pa4 - 4qb4 = f 1.
Then b cannot be even since p f i-1 (mod 16), and cannot be odd since
p - 4q f f 1 (mod 16). This completes the proof.
If D is a prime, q = 1, and then the proof does not apply when D =
k 1, 3, 5 (mod 16).
By the use of less elementary results, it can be proved that when D is a
prime E 1 (mod 4), there are no integer solutions of y2 = Dx4 + 1 except
x = 0 unless D = 5 when also x = 2. Two results are required. The first is
that when D = 2, the only integer solutions are given by x = 0. This is easily
proved for we deduce at once,
y f 1 = 2p4, y 7 1 = 16q4, x = 2pq,
p4 - 8q4 = 1, p2 + 1 = 2r4, p 2 - 1 = 4s4.
Clearly s = 0, p 2 = 1, q = 0, x = 0.
The other is a much deeper result noted a little further on that the only
integer solutions of y 2 = 2x4 - 1 are 5 x = 1, 13.
LjunggrenZ0 proves the result for D = 1 (mod 4) by expressing x , y in
terms of the solution of the Pel1 equation Y 2 - D X 2 = - 1.
A proof, however, follows by a slight modification of the general proof.
When r > 1 in equation (14), then r = D,s = 1, and we have a2 - 1 = 2d4
giving a = k 1 etc. Also equation (16) becomes
Da4 - 4b4 = 1
+ +
Da4 = (2b2 26 1)(2b2 - 2b + 1).
Then +
2b2 f 2b 1 = e4
(26 t- 1)2 = 2e4 - 1, etc.
Ljunggren lo has proved some results about the number of solutions of some
equations of the form (10).
Theorem 9
The equation
y2 = Dx4 + 1, (16)
where D > 0 and is not a perfect square, has at most tu'o solutions in positice
integers.
28. SOME EQUATIONS OF DEGREE >3 27 1
Denote by E the fundamental unit in the quadratic field Q(d/O).
If there are
two solutions, these are given by
either y + x 2 d o= E, or by y + x2dD= E, E ~ ,
the latter occurring for only a finite number of values of D.

It is much more difficult to deal with the equation
y2 = Dx4 - 1.
In particular for the special case

y = 2x4 - I
it has been known for two centuries that solutions are given by (x, y ) = (1, 1)
and (13, 239). It was proved by LjunggrenZ3that these are the only positive
integer solutions. The proof is exceedingly complicated. It is a particular case
of his more general
Theorem 10
If the fundamental unit of the quadratic field Q(dd)is not also the funda-
mental unit of the ring Z[d/d],then the equation
y 2 = dx4 - 1
has at most two possible solutions, and these can be found by a finite algorithm.
We assume of course that the Pel1 equation
y2 - dX2 = -1
is solvable, and so we write
E = u + Vd2, E' = u-v f i
for the fundamental unit in the ring Z[dd].

Then EE' = - 1 and so u is odd. We now take x > 0, y > 0 and so
y + xzdd = er, y - xWd = Ell, r = I, 3,5,. . ..
Hence xZ=v (&I I11')

~
We show that from this we can deduce an equation of the form

&S - E's
x20 -- -*
e - E
This is obvious if u is a square. Suppose next that u = l k 2 where 1 > 1 and is

odd and square-free.
Then
say. Since E - e' = 0 (mod I ) , it follows that r 3 0 (mod I ) , say r = Is. Write
If (P,I ) = 1, then P = x i as asserted. If (P,I) > 1 suppose that a prime p

divides ( P , I ) . On putting (2 - e'")i(e - e') = p"q where ( p , q ) = 1, then
since 1 is square-free, it is easily seen that Q is exactly divisible by p . Then Q is
exactly divisible by I and so again
&S - &IS
I = x:.
& - &
To find the integer values of xois a particular case of the problem in Chapter 8
of finding the square numbers in a sequence defined by a recurrence formula.
Ljunggren shows that the p-adic considerations of Chapter 23 can be
applied if we use an extension field of Q(Y'2). Thus from
(& - &I)X: = ES - E'S,
+ 1 = F 1 ( & 2+ l)x;,
2 s
(&"2 - (&2+ 1)(Xo&(S-1)'2)2= - 1.

Then es + x ~ E ( ~ - ~ ) ~ ~ ( ~ ~
+ l)ll2 is a unit with relative norm
- 1 in the ring
Z[1, vdz, d-1, vY'2 d e 2 + 11. There are three fundamental units in the
ring but he shows that it suffices to use only two of them.
4. Binomial equations
The equation
axm - by" = c.
We may suppose that a, b, c, m, n are positive integers, and that either m or n

is not less than 3.
Very little is known about the integer solutions of equation (19) for general
m and n. The known results are of considerable interest, and in general are
concerned with the existence of only a few solutions, often none, one or two.
They can often be stated very simply in terms of fundamental units of algebraic
number fields. Proofs are usually very complicated and may involve con-
siderable detail, both theoretical and numerical, and so cannot be given here.
28. SOME EQUATIONS OF DEGREE > 3 273
A few results, however, may be mentioned, Most of them deal with the case
when 111 = n.
Some are included in the results for the integer solutions of the inequality
lax" - by"l < c.
The first ones are due to ThueZ4and are particular cases of his general
theorem for the inequality
laox" + a,x"-ly + ... + anyn]d c.
This states * that if an integer solution is known for which 1x1, lyl are greater
than a determinable constant M , then a constant N can be found such that all
the solutions are numerically less than N . The method was extended and
generalized by SiegelZ5who proved the fundamental
Theorem 11
The inequality
lax" - by"/ < c,
where a, 6 , c are positive integers and n 3, has at most one solution in
positive co-prime integers (x, y ) if
where p runs through all the diferent prime factors of n.

In particular when n is a prime, we can replace condition (19) by the weaker
and more practical condition
lablliZ b 188c4. (20)
The idea in the proofs by Thue and Siegel can be made to depend essentially
upon the approximation of algebraic functions by means of rational functions.
This can be presented in different ways. Let A,(z), Br(z) be polynomials with
rational coefficients and of degree r. Thue proved a relation of the form
A,(z") - zB,(z") = (2 - l)"+'C,(Z),
where C,(z) is a polynomial of degree n(r - 2). His work is purely algebraic.
Siegel, however, used a relation of the form
A,(z) - (1 - z)l/"B,(z) = 22r+lC,(z).
Now C,(z) and the polynomials A,(z), B,(z) appear as hypergeometric series.
Their properties are easily investigated, and the relations between them are
given at once by using the hypergeometric differential equation.
* Hyyro2'" points out that N depends upon the first hypothetical solution. Since there is
no limit to the first solution, the method fails to give an N .
DomarZ6 by sharpening some of the inequalities in Siegel's proof, has

obtained the two results given in
Theorem 12
The equation
ax" - by" = f 1, (21)
where a, b are positive integers and n 2 5, has at most two solutions in positive
integers.
Theorem 13
The equation
X" - Dy" = f 1, (22)
where D > 0 and n 2 5, has at most one solution in positive integers x , y
except possibly when D = 2, and when n = 5 or 6 and D = 2" & 1.
Theorems 11 and 13 have been used by af Ekenstama7 to prove a result
given without proof by Tartakowski, namely, are
Theorem 14
If the equation
x2" - Dy2" = 1, (23)
where D is a positive non-square integer and n > 3, has a solution (xl, y l ) in
positive integers, then i f E is the fundamental unit of the ring Z[do]and
N ( E ) = 1,
x; + ~ $ 0= e or 82,
the E~ occurring in only a finite number of cases. If N(e) = - 1 there are only
solutions with y = 0.
Some results for equation (19) may be noted when m,n have small values,
for example 2,4, 6 and c also has special values, for example, 1,2,4, 8,. . ,.
Thus Ljunggren 1 7 * 1 5 has proved
Theorem 15
The equation
where a, b, c are positive integers, has at most one solution in positive integers
x, y when
n = 2, c = 1,2,4, 8;
n = 3, c = 1, 2, 3, 4, 6.
Further, among all the equations (24) corresponding to a given field
there is at most one solvable in positive integers except when n = 2, d = 5 .

~ ~ utilizing the results in Siegel's 25 paper, supersedes
af E k e n ~ t a m ,by
Theorem 12 by
Theorem 16
Among all the equations
lax" - by"] = 1, n 2 5
~~
corresponding tofields Q(Vd)= Q ( V a b - l ) with given d, there are at most

two solvable in positiveintegers x, y .
He obtained a similar but more complicated result for the equations
lax" - by"] = c where 1 < c < c', a number depending upon lab/. Another
of his results has been improved by H ~ y r o , and
~ ~ "is given as
Theorem 17
Let d 3 2, n 3 5 be given integers. Then the equation
Ix" - dsy"l = 1,
where x 3 2, y 2 I , 0 < s < n, and x 2 3 for n = 5, 6 has at most one

solution s, x, y .
He has also improved Theorem 16.
Theorem 18
The equation
x4 - Dy4 = 1, (25)
where D is apositive non-square integer has at most one integer solution. Ifthis
is ( x l ,y,) and E is the fundamental unit in Q(z/o> and N ( e ) = 1, then
x: + y:z/Z = E or e2,
the latter occurring in only afinite number of solutions. I f N ( e ) = - 1, there is

no solution except when D = 5.
Ljunggren has shown that the e2 arises only in the case D = 7140 when
239' + 26'4714O = (169 + 247140)' = e2.

By different methods Delone and Faddeev 29 have proved
Theorem 19
The equations
X* - Dy4 = f 1 (26)
have at most one solution in positice integers. This will be given by .IC= A,
y = B fi the fundamental unit of the ring Z [ q x ] takes the form
= AZ + ABKXD+ B W / - .
The impossibility of non-trivial solutions of the equations
y2=xmf1, m > 3
y3=xmf1, m > 2
will be noted in Chapter 30.
Finally we give a result by S k ~ l e m ~ ~ .
Theorem 20
The equation
x5 + Dy5 = 1
has no solution in non-zero integers when D = 4, 8, 16, and only the solution
(-1, 1) when D = 2.
The proof depends upon the p-adic methods of Chapter 23.
From Siegel's theorem, it follows that there is at most one non-trivial
solution if D > 1250VB. It has been shown by Haggmark3' that there are at
most two solutions, a result contained in Theorem 12.
Addendum 1
Theorem 1 has been improved very recently by Schinzel 32 who following a
suggestion by Davenport and Lewis, proved as follows,
Theorem 21
Let f ( x , y ) be a polynomial of degree n with integer coeficients which is
irreducible in the rationa1,field Q and let fn(x, y ) be its homogeneous part of
degree n. If f ( x , y ) = 0 has infinitely many integer solutions, then except for a
constantfactor,f n ( x ,y ) is apower of a linearform or of an irreducible quadratic
form.
Suppose that f ( x , y ) = 0 has infinitely many integer solutions. Then by
theorem 1, fn(x,y ) is, except for a constant factor, a power of an irreducible
form, and so either
fn(x,y ) = ax" or by",
38. SOME kQUA'IIONS O F DFGREE >3 277
in accordance with the theorem, or
/"(.Y, y) = 00s'' + . + any,
' ' (aoa, # 0). (27)
By Siegel's theorem 2, we have identically in t ,
f ( W )W))
, = 0, (28)
where either
I, R(r), S(t) are polynomials, not both constant, or
where A , B, L are polynomials, L is linear, ( A , B, L ) = 1, HZ > 0,
where C, D, Q are polynomials, Q is a quadratic, (C, D, Q ) # Q,m > 0.

For I, from equations (27), (28), R and S are of the same degree and so if
r and s are the leading coefficients of R and S, then fn(r, s) = 0. Hence
f;,(x, y ) is divisible by sx - ry and so from theorem 1,
f i ( x , y ) = c(sx - ry)".
For 11, let t o be the zero of L(t). Multiplying (27) by L(t)mnand then
putting t = to, we see thatfn(A(to), B(to))= 0. Hencef,(x, y ) is divisible by
B(to)x - A(to)y and so from theorem 1,
/n(& Y) = c(B(to).y - A(rolY)"
For 111, let t,, t2 be the zeros of Q ( t ) . We cannot have C(t,) = 0, D(tJ = 0,
(i 1, 2) say, D(tJ # 0. Multiplying (28) by Q(t)"'" and then putting
=
r = t l , we obtainfn(C(tl), D(tl)) = 0, and sofn(x,y ) is divisible by D(rl)x -
C(tl)y. Then either C(t,)/D(t,) is rational, and
f,d& 1.) = 4D(t,),Y - C(t,)Y)",

or
;ire conjugates in a quadratic field, and so
C(t,)/D(t,)and C(t2)/D(t2)
f,,(.v. I.) = c
t
n
= 1.2
(D(t ,)x- C(/,)r)'''Z,
and the product is a power of an irreducible quadratic.

A consequence of this theorem due to Davenport and Lewis replaces
theorem 4 of Chapter 22 by
10
Theorem 22
The equation f ( x , y ) = g(x, y ) where f and g are polynomials of degrees n, m
with integer coefficients, has only afinite number of integer solutions f n > 2 ,
t n < n, and f is an irreducible form.
Addendum 2
Siegel's theorem 3 does not enable one to find all the integer solutions
of the equation
y2 = aoxn + alx"-l + . . . + a , = J'(x),
say.
These can now be found from the following result just proved by Baker.33
Theorem 23
r f f ( x ) has at least three simple zeros, then all the integer solutions satisfy
the inequality
max (1x1, Iyl) < exp exp exp (n10f13An2),
where A = max / a , [ ,(r = 0, 1,. . .n).
REFERENCES
I . C. Runge. Uber ganzzahlige Losungen von Gleichungen zwischen zwei
Veranderlichen. J . reine angew. Math., 100 (1887), 425-435.
2. C. L. Siegel. Uber einige Anwendungen diophantischer Approximationen.
Abh. preuss. Akad. Wiss.Phys-Math. KI. (1929), Nr. 1.
3. L. J. Mordell. The diophantine equation y 2 = ax" bx2 cx+ + + d or fifty
years after. J. Lond. Math. SOC.,38 (1963), 454-458.
4. C. L. Siegel. The integer solutions of the equation y 2 = ax" + bx"-l +
+
. . . k . J. Lond. Math. Soc., 1 (1920), 66-68.
5 . C. L. Siegel. Approximation algebraischer Zahlen. Math. Z., 10 (1961),
173-213 (205).
6. E. Landau and A. Ostrowski. On the diophantine equation ay2 + by + c =
dx". Proc. Lond. Math. SOC.,(2), 19 (1920), 276-280.
+
7. A. Thue. Uber die Unlosbarkeit der Gleichungax2 bx + c = dy" in grossen
ganzen Zahlen x und y . Arch. Math. Naturv. Kristiania Nr. 16, 34(1917).
8. 0. Rigge. uber ein diophantisches Problem. ZX.Skan. Math. Kongr. Helsing-
fors (1938).
9. P. Erdos. Note on products of consecutive integers. J . Lond. Math. SOC.,
14 (1939), 194-198.
10. R. Oblith. Uber Produkte aufeinander folgender Zahlen. T6hoku Math. J . ,
38 (1933), 73-92. See also Berichtigung.
1 I . P. Erdos. Note on the product of consecutive integers. 11. J. Lond. Math. SOC.,
14 (1939)) 245-249.
12. 0. Rigge. On a diophantine problem. Ark. Math. Astr. Fys., 27A, 3 (1940).
13. P. Erdos. On the product of consecutive integers, 111. KoninkI. Nederl. Akad.
van Wetenschappen (1955), 85-90.
28, SOME EQUATIONS OF DEGREE >3 279
+ +
14. L. Louis Johnson. On the diophantine equation y k = x(x 1). . . (x n - I).
Am. Math. Mon., 47 (1940), 280-289.
+ +
14a. L. J. Mordell. The diophantine equation dy2 = ax4 by2 c. Acfa Arith.
15 ( I 969).
15. W. Ljunggren. Solution complete de quelques Cquations du sixiime degre a
deux indeterminkes. Arch. Math. Naturv., 48 (1946), Nr. 7, 26-29.
16. J. W. S. Cassels. Integral points on certain elliptic curves. Proc. Lond. Math.
SOC.(3), 14 A (1965), 55-57.
17. W. Ljunggren. Einige Eigenschaften der Einheiten reeller quadratischer und
rein biquadratischer Zahlkorper. Oslo Vid-Akad Skrifter, 1 (1936), NO. 12.
18. W. Ljunggren. Uber die unbestimmte Gleichung Ax2 - By4 = C. Arch.
Math. Naturv., 41 (1938), Nr. 10.
19. W. Ljunggren. Uber die Gleichung x4 - Dy2 = I . Arch. Math. Naturv., 45
(1942), Nr. 5.
20. W. Ljunggren. Some remarks on the diophantine equations x2 - Dy4 = 1
and x4 - dy2 = I . J . Lond. Math. Soc., 41 (1966), 542-544.
2 I . L. J. Mordell. The diophantine equation y2 = Dx4 + I . J . Lond. Math. SOC.,
39 (1964), 161-164.
22. J . H. E. Cohn. Eight diophantine equations. Proc. Lotid. Math. SOC.(3), 16
(1966), 153-166. Addendum Ibid. (3) 17 (1967), 381.
22a. V. D. Podsypanin, On an indeterminate equation (Russian). Izv. Akad. Nauk
SSSR Mat., 5 (1941) 305-324.
23. W. Ljunggren. Zur Theorie der Gleichung x2 + 1 = Dy4. Auh. Norske
Vid. Akad. Oslo, No. 5 1 (1942).
24. A. Thue, Berechnung aller Losungen gewisser Gleichungen von der Form
axT - by' = J: Vid. selskap Skrifter Kristiania mat. natur. Kl. (1918), No. 4.
25. C. L. Siegel. Die Gleichung ax" - by" = c . Math. Ann., 144 (1937), 57-68.
Also Gesanimelte Abhandlungen, I1 (1966).
26. Y . Doniar. On the diophantine equation 1A.P - By"I = I , II > 5. Math.
SCUII2 ~ .(1, 954), 29-32.
17. A. af Ekenstam. Contributions to the theory of the diophantine equation
A s " - By" = C. Dissertation (1959) Uppsala, Alnrqvist and Wiksells.
27a. S. Hyyro. Uber die Gleichung ax" - by" = z und das Catalansche Problem.
Anti. Acaci. Sci. Fennicae, Series A.1.355 (1964) 32.
28. V. A . Tartakowski. Auflosung der Gleichung x4 - py4 = 1. Izv. Akad. Nauk
SSSR, 20 (1926), 301-324.
29. B. N. Delone and D. K. Faddeev. "The Theory of Irrationalities of the Third
Degree. Tran.slationofMath. monographs, Am. Math. Soc., 10 (1964), 370-380.
30. T. Skolem. En metode ti1 behandling av ubestenite ligninger. Chr. Michelsens
Itist. Beretti, 1V (1934), Nr. 6, Bergen.
3 1. P. Hgggmark. On a class of quintic diophantine equations. Dis.srrtatioti
Uppsala (1952).
32. A. Schinzel. An improvement of Runge's theorem on diophantine equations.
Conimentarii Acaci. PontiJjciae Scientiarirtn, 2 (1969). (A generalization will
appear in the same journal.)
33. A. Baker. Bounds for the solutions of the hyperelliptic equation. Proc. Camb.
Phil. Soc.. 65 (I 949). 439444.
CHAPTER 29
Fermat's Last Theorem
1. The most famous of all Diophantine problems is due to Fermat and is

called Fermat's theorem, namely,
Theorem 1
I f n is an integer > 2, the equation
y + y" = Zn (1)
has no integer solutions except those given by xyz = 0.
Fermat stated that he had a proof of this, but it seems unlikely that this is
so. The proof given when n = 4 in Chapter 4 is in principle due to him. The
proof given when n = 3 in Chapter 16 goes back to Euler, but his proof was
incomplete. Kummer made the greatest contributions to the subject. He
introduced the algebraic numbers arising from the pth roots of unity, and this
was the beginning of algebraic number theory which was later generalized by
Dedekind and Kronecker.
It is obviously sufficient to prove the theorem when n is an odd prime p .
The proof is comparatively easy when p is a prime < 19 since then unique
factorization holds in the field Q(5) where 5" = 1. Kummer was at first under
the impression that this held for all p .
No proof is known for general p but, as shown later, the theorem has been
proved for many special values for p . We may suppose that p > 3 and write
Let 5 = ezniiPbe a complex pth root of unity. Then the equation (2) can be
written as
(s + .r)(.v + (.I,).. .(.v + i"-'y) = -.

-- p ( 3)
To apply the methods of Chapter 15, we require some knowledge of the

cyclotomic field Q(4). For this the usual books on algebraic number theory
may be consulted. For the applications to Fermat's last theorem, references
may be made to the accounts by Hilbert', Bachmann2, Dickson3, Mordel14,
Landau5, LeVeque6, Borevich and Shafarevich', and Vandiver*. A complete
study of the solvability of the equation (1) requires a detailed, intricate and
advanced arithmetical investigation into which we cannot enter here.
29. I-ERMAT'S LAST THEOREM 28 I
We prove the impossibility in some comparatively simple cases and for this
we require the following results. Most of them will be stated without proof,
but a few will be proved.
I. The degree of the field is 17 - 1 since the equation
--1' ~ 1 + :p-2 +'.'+z-t-l=O
is irreducible in the rational field Q.
11. The integers in the field are given by
5 = (10 + (Il< -k ' ' ' i- ( I , , 3,
where the n are rational integers.

Clearly .$p = a (modp) where a is a rational integer.
111. The only roots of unity in the field are kc', r = 0,1,. . . , y - 1.
IV. The numbers E~ = (tr - 1)/(5 - 1) are units. For E, is an integer and
has norm N(E,) = 1.
V. X = 1 - 5 is a complex prime and p = d P - l , where E is a unit. For put
u=lin
r=l
then /J = n
P - 1
r=l
(I - c') = n (H)
P- 1
1 5 -
(1 - [)p-l
VI. Every unit E can be expressed in the form E = P.1 where 0 < s < p and
is a real unit.
Write E = E~ = f(t),wheref(5) is a polynomial with integer coefficients,
and el = f(r;l). Then
is also a unit, and so
has coefficients i n Z . Since I E ~ / E - , . I = I , is a root of unity by a well

known theorem of Kronecker, and so
1 = *<J&-*,
say. We show that the negative sign does not hold, i.e. t', + 4'. ~, =. 0
is impossible.
For if this holds, we have identically i n I\'.
(I,, + (/,\I' k ' .. + n,,- 2 1 1 J - ?!
+ lI,'fP(l7,, -+- f I , l I ' ' ! ' ' ' f- (/,, . , n ~ - " - " ' )
= I,,(( 1 + 11' + ... +- \\J' - l<?(ll,)
for some integer t and a polynomial g(w) with integer coefficients. Put
w = 1, then
a,+ a, + . . . + = O(modp).
Hence E = a, + a,{ + + ap-25p-2
9 . .
= a,(5 - 1) + . . . + a p - 2 ( 5 p - 2- 1) (mod A)
= 0 (mod A),
and this is impossible.
VII. If p is a regular prime and e is a unit in Q(5) for which E = a (mod Ap),
where a is a rational integer, then e = E; where E, is a unit in Q({).
The proof requires too much detail to be given here.
The primes p are called regular or irregular according as the ideal class
number h is not divisible byp. Kummer has shown that h f 0 (modp), if p
does not divide the numerator of the first $(p - 3) Bernoulli numbers.
These are defined by
-- t
et - 1
-I-z+c t
n=1
(-l)n-1Bnf2n
(24!
All the primes < 100 exceptp = 37, 59,67 are regular. It has been proved that
there are an infinity of irregular primes but it is not known if there are an
infinity of regular primes. We shall prove Fermat's theorem for the regular
primes. Kummer has proved the theorem for some classes of irregular primes.
2. We come back to equation (3). The common factors of x + Cy, x Py, +
s > r, are divisors of (lr- P)y, and since (x, y ) = 1, of Cr(l - P-'), and
these can only be 1 or A. In the first case, all the factors x {'y are relatively +
prime. In the second, all the factors are divisible by A, and so z is divisible by h
and hence by p. The cases z $ 0 (mod p ) , z = 0 (mod p) are called the first
and second cases respectively.
Theorem 2
I f p is regular, theJirst case is impossible.
Each of the factors in equation (3) is a pth power except for a unit. Hence
s + 5y = &UP, (4)
where E is a unit in Q ( { ) .We may take E = P.1 where 0 < s < p, and TI is a real
unit. Since c8' = a (mod p ) , where a is a rational integer,
s + 5y = 5"( (mod p),
and 5 is a real integer in Q(5). Hence also
.Y + {-'J? = [-'[(mod p).
Then 5-s(s + 2;y) - C'(s + <-',v) = O(mod p).

29. FERMATS LAST THEOREM 283
Since 1 , i , . . . , 5 P - 2 form an integer basis for Q([),then if ao,a,, . . ., ap- are
rational integers,
uu + u,< + I + u p - 2 5 p - 2= 0 (modp)
if and only if
U() =a, = . .-
- u p - 2 E 0 (modp).
Write <V-(.Y + ir) - (ex + is-

y) 3 0 (mod p).
If 2 < s ,< p - 2 and s # (p -t 1)/2,the powers of 5 are all different and less
t h a n p - 1, and s o x = y = O(modp), i.e. z = O(modp).
If s = (p + 1)/2, then .Y + ( y = 5.x +
y (modp), i.e. x = y (modp).
+
Similarly x = z (modp) and SO x P y P + z P = 3.uP = 0 (modp), and this
does not hold sincep > 3.
If s = 0, .v + (J - (.v +[ - y) = 0 (mod p).
Then y = 0 (mod p), and similarly x = 0 (modp).
If s = I. 5 -.v + y 5 5s + y (modp), i.e. .Y = 0 (modp),
and similarly y = 0 (mod p).
If s = 2, s + i y = i4.\- + c3y and so .Y = y 0. E
Ifs = p - I, ((.Y i [ y ) = + {-2y(modp),

(-I.<
1.e. i3(\-+ ij,) = i x + y (mod p).

Since p 3 5, this gives s = y = 0 (mod p).
Theorem 3
I f p is regular, the second case is impossible.
It proves convenient now to consider solutions not in Z , but integer solu-
tions in Q ( [ ) .We also consider the more general equation
S + 4 = EhnpZP,
where E is a unit and n > 1 . We shall show that n must be greater than 1 and
that if the equation is solvable for n, it is also solvable for n - 1 . This leads
to a contradiction. We have
n
P-1
r=O
(.Y + cry)= EhnPZP.
(5)
One of the factors on the left-hand side, and so all of them, must be divisible
by X since x + try - (x + Py) = (5 - P)y. This also shows that X is the
greatest common divisor of any two of the factors. Hence one of these must be
divisible by X n P - p By a slight change in the variables, we may suppose this
+
occurs when r = 0. We now replace ex, Cby in the equation by x, y and show
that we can choose a, b so that x = x,, y = yo (mod A2) where xo, yo are
rational integers.
In I1 we replace 5 by 1 - A, and so with rational integers xo, . . . , x p - z ,
+ s,h + . . . +
.Y = s o SP-2hP-2.
Then since 5 = 1 - A, 5"s = so + h ( s , - axo) (mod A2). Since so $

0 (mod A), we can choose a so that x1 - QX, = 0 (modp).
Similarly for y, - by, = 0 (modp). Hence x + y = xo + yo (mod A'), and
so xo + yo = 0 (modp). Then x + y = 0 (mod ha) and since s + y =
0 (mod Anp-p+l), n > 1.
Write S = ( x , y ) for the ideal common factor of x, y . Then we have the ideal
equations
+ y ) = S h n p - p + l a g , (x + try) = sxa;,
(x
or, say, (X + 5 " ~ )= 6Atra,P,
where a0,a, are ideals and t , = tip p + 1 when r = 0, t, = 1 when r # 0.

Hence a: a;and since (p, h) = 1, a,
N
integers S,, y,. Also we can write

-
- a,, and so Srar = yraofor algebraic
where is a unit, and so

(S,yJP(x + Cry) = ErsAfr-ts(Ssrr)P(-x + 5'~).
In particular with ( r , s) = (0, 1) and (r, s) = (2, l), we can write, say,
up(x + y) = E,(X + 5y)j3;h("-1)P,
4(s + C2r)= E Z G + 5r)B%,
where e l , c2 are units, and the a, jl are algebraic integers. Multiply the equa-
tions by Saf(,a; respectively and add. Since
5(x + y) + (s + 5") = (5 + l)(s + 5y),
we can cancel a factor x + i y and so
aPap2(5 + 1) = 5E1(a2/3,)*A'n-1)P 4- EZ(C(1132)P.
Since cl, cZ and 5+ 1 = - I)/([
(i2 - I) are units, we have an equation of
the form
,yp + E3 yp = E4h(" - 1 , p z p
where X, Y, 2, are integers in Q(5). Now X p = so, Y p = y o (mod p ) where

xo,yo are rational integers and xoyo f 0. Hence since n > 1,
.yo + e3yo I0 (mod hP).
29. FERMAT'S LAST THEOREM 285
Then e3 = -ao/yo (mod hP) and so c3 = eE where e5 is a unit, and
x p + (E5Y)P = c4XP(n-1)Zp,
an equation with n replaced by n - 1. This proves the theorem.

3. Other methods are known for dealing with the first case when x and y are
rational integers. These depend upon arithmetical considerations different
from those previously considered and we can give only brief mention of the
results. Kummer showed by multiplying together an appropriate set of the
equations (4), that solvability implies some polynomial congruences. He
proved
Theorem 4
Denote by t the quotient of any two of s,y , z , where xyz $ 0 (mod p ) . Write
+,(t) = t - p - l t 2 + 3n-lt3 - .. . + (-1)~-2(-
~ 1)n-lyp-l.
Then soli1nbility of equation (1) itiiplies
= 0 (modp), n
+n(t)B(p-n)12 = 3, 5, . . . ,p - 2. (6)
From this, polynomial congruences can be found for the t if some or all of
the B ( p - n ) ,are
2 not divisible byp.
The congruence (6) can be replaced by
+p-n(t)+n(t)= 0 (modp), n = 2, 3,. . ., '

2
3,
c $ ~- l ( t ) = 0 (mod p).
From these Wieferich deduced the criterion
2p-1 = 1 (modp2). (7)
Mirimanoff showed that also
3p-l = 1 (modp2). (8)
For p < 31,059,000 congruence (7) is satisfiedlO only by p = 1093, p =
3571, and (8) only by p = 11, p = 1,006,003 for p < 10,752,000.
Furtwangler has also proved (7), (8) by applying the general law of
reciprocity in the cyclotomic field Q(5). He showed that if a prime q is a
divisor of xyz, then
qp-' = 1 (modp2). (9)
It has been proved by various writers that this result holds for all primes
q < 43. The Lehmers have used such results to prove the impossibility of the
first case for p 6 253, 747, 887.
Results for the irregular primes were also found by Kummer. It has been
proved by K. L. JensenO that there are an infinity of irregular primes. Of the
Ilf
302 primes less than 2,000, only 118 are regular. The work of Vandiver and
the Lehmers has settled the insolvability for p < 2000. Selfridge, Nicol
~ this for p < 4002. It now holds12 for p < 25,000.
and V a n d i ~ e r do
REFERENCES
1. D. Hilbert. Die Theorie der algebraischen Zahlkorper. Jahresbericht der
Deutschen Mathematiker Vereinigung, 4 (1 897), 175-546. Also Gesammelte
Abhandlungen, I (1932), Springer, Berlin. Also ThCorie des Corps de
nombres algkbriques (1913), A. Hermann et fils. (Translation by A. Levy
and Th. Got.)
2. P. Bachmann. Das Fermatproblem in seiner bisherigen Entwicklung
(1919). Walter de Gruyter & Co., Berlin und Leipzig.
3. L. E. Dickson. History of the Theory of Numbers, TI. Carnegie Inst.
Washington Pub. 256 Roman Cap, 1920. (Reprint: Stechert, New York, 1934).
4. L. J. Mordell. Three Lectures on Fermats Last Theorem (1921) Cambridge
University Press, Cambridge.
5 . E. Landau. Vorlesungen iiber Zahlentheorie, 111 (1927), S. Hirzel, Leipzig.
6. W. J. Le Veque. Topics in Number Theory, I1 (1956), Addison-Wesley,
Reading, Massachusetts.
7. Z. I. Borevich and I. R. Shafarevich. Number Theory (1966), Academic
Press, New York & London. (Translation from the Russian.)
8. H. Vandiver. Fermats last theorem; its history and the nature of the results
concerning it. Am. Math. Monrh., 53 (1946), 555-578. See also A supple-
mentary note to a 1946 article on Fermats Last Theorem. Amer. Math.
Monthly, 60 (1953), 164-167.
9. K. L. Jensen. Om talteoretiske Egenskaber ved de Bernoulliske Tal. Nyt.
Tidskr. Math., 26B (1915) 73-83.
10. K. E. Kloss. Some number theoretic calculations. J. Res. Nut. Bur. Standards
Sect. B, 693 (1965), 335-336.
11. D. H. Lehmer, E. Lehmer, H. S. Vandiver. An application of high speed
computing to Fermats Last Theorem. Proc. Nut. Acad. Sci. U.S.A., 40
(1954), 25-33.
12. J. L. Selfridge and B. W. Pollock. Notices Amer. Marh. SOC.,11 (1967), 97.
(Abstract 608-138).
13. J. L. Selfridge, C. A. Nicol, H. S. Vandiver. Proof of Fermats last theorem
for all prime exponents less than 4002. Proc. Nut. Acad. U.S.A., 41 (1959,
970-973.
CHAPTER 30
Miscellaneous Results
1. The equation z = f ( x , y).

Let f(x, y ) be the quotient of two polynomials in x, y with integer co-
efficients. Two kinds of problems arise according as the integers x , y , z are all
non-negative or not. Two of the first type will be discussed. The first deals
with the exceedingly interesting
Conjecture (Erdos, Straus),

The equation
4-
n
- -1x+ - 1y+ - ,1z
-
where n is an integer > 3 is solvable in positive integers x, y , z.

It has been verified for n < 5000 by Straus, n < 8000 by Bernsteinl,
n < 20,000 by Shapiro, n < 106,128by Oblith2, and n < lo7by Yamamoto3.
The proofs show that the equation is solvable if n satisfies one of an infinite
set of congruences, or equivalently, if n is representable by one of an infinite
set of arithmetic progressions.
Obviously if equation (1) is solvable for n, and in particular when n is a
prime, it is also solvable for any multiple of n.
For some simple general results, let a, b, c, d be arbitrary positive integers.
Then if
a + bn + cn = 4abcd, (2)
there is a representation
(x, y , z ) = (bcdn, acd, abd).
If nu + b + c = 4abcd, (3)
there is a representation
(x, y, z )
(bcd, nabd, nacd).
=
Representations when n = 2, 3 (mod 4 ) follow from (3). For if

a = 2, b = 1, c = 1, then n = 4d - 1,
a = 1, b = 1, c = 1, then n = 4d - 2.
When n = 0 (mod 4), we have x = y = z = 3n/4.
Hence we need only consider the case n = 1 (mod 4).
If in equation (3), a = 1, b = 1, c = 2, n E -3 (mod S), and so we need only

consider n = 1 (mod 8).
Write equation (3) as
+ b = ~ ( 4 a b d- 1) = C q , (4)
where q + 1 = 0 (mod4ab).
Take q = 3, a = b = 1, then ri = - 1 (mod 3).
Since for n = 3, we have ( x , y , z ) = (3,2, 2), equation (1) is solvable for
n = 0 (mod 3), and so we may suppose that n = 1 (mod 3).
Take q = 7, and then ab12, and equation (4) gives
n = -1, -2, -4(mod7).
Since for n = 7, we have (x, y, z) = (2,28,28), we may suppose that n E
1,2,4 (mod 7).
Take q = 15, then ab14, and with (a, b) = (1,2), (2, l), we have
n + 2 = 0 (mod 15), n + 8 = 0 (mod 15).
Since n = 1 (mod 3), we have
n = -2, - 3 (mod 5).
Hence we may suppose that n = 1 , 4 (mod 5).
Gathering these results together, the conjecture is proved except possibly
when
n = 1, 121, 169,289, 361, 529 (mod 840).
Since the first prime in this set is 1009, the conjecture holds for n < 1009.
Write 1 = a(n + 3). Then with arbitrary m,
4m + 3
4
-=- 1
n I+m
+-.
n(Z + m)
Hence a representation exists if for positive integers m, d, d,, d,, d3,
d(4m + 3) = dl d2, +
+
1 m = dld2d3,
Many results can be found in this way.
The conditions (2) and (3) are sufficient for solvability. We now investigate
necessary conditions. Let n be a prime p > 3. Then all three of x, y , z cannot
be divisible by p. Hence either only one, say x, is divisible by p , or only two,
say, y , z, are divisible by p. We replace equation (1) by the two equations
- - + - + -, xyz $ O(modp),
-- 1 1
P P X Y Z
-=-+-+-, xf:O(modp).
P x PY P Z
30. MISCELLANEOUS RESULTS 289
We now prove a result due to Rosati3" who also verified the conjecture for
n < 171,649.
Theorem 1
A representation ( 5 ) exists if and only if for positive integers a, b, c, d,
( x ,y , z ) = (bed, acd, abd),
where (a, 6 ) = (b, c ) = (c, a ) = I , p j abed, and
a + bp + cp = 4abcd. (7)
A representation (6) exists ifand only if'.for positizie integers a, b, c, d,
(x, y , z ) = (bed, abd, acd),
where (a, b) = (b, c ) = (c, a ) = 1, p j bed, and
pa + b + c = 4abcd.
To prove representation (9, write ( y , z ) = 6, y = 6b, z = Sc and so
(bcS,p) = 1. Then
6bc + p(b + C)X = 4Sbcx.
Since ( b + c, be) = 1, x = bed, say, and so
6 + pd(b + = 46110~. C)
Then 6 = da, say, and (ad, p ) = 1. Hence

a + p(b + c) = 4abcd.
To prove representation (6), write ( y , z ) = 6, y = Sb, z = Sc. Then
pSbc + cx + bx = 4.xSb~.
Here x = bed, p t bcd
p6 + (b + c)d = 4xSbcd.
Then 6 = ad and
pa + b + c = 4abcd.
We note that a = 0 (modp), i.e. y = I = 0 (modp) is a possibility.
L. Bernstein has shown that an infinity of congruences can be derived for
p from equations (7) and (8). To illustrate his method, it will suffice to deal
with (8).
We may suppose p = 8s +
I . We shall find values for b and c. From (8),
b+ c = a(4bcd - p ) = a(4t - l), say. (9)
Then 2s + t = bcd, 2s + t = 0 (mod bc). (10)
We require another equation for 6, c. We can take for this

c - b = 42g/ + I ) - 25
since a + b + c is even. Then
c = at(2 + g) - 5 b = 42t - rg - 1) + f.
Then equation (10) becomes
2s + t = 0 mod (at(2 + g) - f)(a(2/ - tg - 1) +f)). (1 1)
Here we can take g = 0, f 1, rt 2, . . ., a an arbitrary integer, and ( f , a ) = 1 ,
f= + 1 , k 2 , . . ..
Yamamoto3 deals rather differently with equations (7), (8). Write (8) as
pa + b = c(4ubd - 1) = cq.
Then if ( a , p ) = 1, and this is a restriction,
b
p = --
U
(mod q), q = - 1 (mod 4ab). (12)
Then the Legendre-Jacobi quadratic character

(P/d = - 1.
Suppose first that a and b are odd. Then
Suppose next that ab is even, say ab = 2"k, k odd, q = -1 (mod 8). Then
In (7), put q = 4abd - p . Then p = - a / b (mod q), and p = -q (mod 4ab).

This can be written as
p = -s (mod 4ubq), s = q (mod 4ab), s = a/b (mod q). (14)
If we use the Kronecker quadratic character, we can show similarly that
(s/4ubq) = 1.
+
In (8), put b c = as. Then
p +
s = 4bcd, p = --s (mod 4bc). (1 5 )
In (7) put b + c = as. Then
1 + p s = 4bcd, p = - l/s (mod 4bc). (16)
In both of these cases, the Kronecker character
(2) = -1.
30. MISCELLANEOUS RESULTS 29 1
2. The equation y z + +
zx xy = d
The problem here is to find explicit expressions for non-negative integer
solutions of the equation
yz + z x + XY = d, (17)
and what is more important, their number. The problem is connected with the
class number for binary quadratic forms. Let
[x, y , z ] = X X 2 + 2yXY + z Y2 (18)
be a binary quadratic form with negative determinant - d and so
x z - y2 = d > 0. (19)
The results will be simpler if weights are attached to special forms, thus
[x, 0, ,x] and [s,x, x ] are reckoned as +
and 3 respectively. Call a form
[s,y , z ] odd or even according as x, z are not both even, or are both even.
Denote by F ( d ) the class number of odd forms, and by G(d) the class number
of all forms. Then G(d) - F ( d ) is the number of even forms.
If the binary quadratic form had been written as
[x, y , z] = .xX2 + y X Y + zY2 (20)
of discriminant -d, and so
4xz - yz = d,
we write H ( d ) for the class number. Then [2s,y , 2z] is an even form of
determinant - d, and so
H(d) G(d) - F(d).
While there is no difficulty in finding a few obvious solutions of equation (17)
for special forms of d, the association with the class number suggests that it
may be difficult to find others.
A famous difficult problem is: when does the quadratic field Q ( 4 - d ) have
class number one, i.e. when is H ( d ) = 1 ? This requires that d should satisfy
various congruences, e.g. d = 3 (mod 8). It is easy to see that d must be 1 or
a prime, and the nine values
C/ = 1,2, 3,7, 11, 19,43,67, 163
are known. It was known that there was at most one other value of d. It has
recently been shown by H. M. Stark by analytic methods that this does not
exist. I n fact, a proof had been given by Heegner which seemed to depend
upon an unproved assertion. However, it has been shown by Birch, Siege1
and Dering that Heegners method of proof is essentially sound.
When d = 3 (mod S),

G(d) = +F(d), H ( d ) = +F(d). (21)
Then if H ( d ) = 1 , G(d) = 4, and so there can only be twelve solutions of
equation (17). These are typified by six solutions ( 1 , 0, d ) each of weight 4,
three solutions ( 1 , 1 , (d - 1)/2)if d > 1 , and six solutions ( 1 , 3 , (d - 3)/4)if
d > 7 . When x = 1 , and so ( y + I)(z + 1) = d + 1 , there would be other
solutions unless i)(d + 1) is a prime. By Starks result, there are other solutions
when d > 163.
Theorem4 2
The number of solutions of
yz+zx+xy=d, x>O,y>,O,z>O, (22)
is 3G(d) i f a weight 3 is attached to a solution with xyz = 0.
To prove Theorem 2, we establish a 1-1 correspondence5 between the
reduced quadratic forms [ A , B, C ] of determinant -d, and the solutions
x, y, z of equation (20) given by
A = x + ~ , IBI = x , C=X+Z.
The reduced forms are given by
B=O, A < C, (23)
B = 0, A = C , weight 3, (24)
B > 0, [ A , +_ B, C ] , B < A < C , (25)
B > 0, [A,A, C ] , C > A, (26)
B > 0, [2A, A , 2 4 , weight 3. (27)
For (23) x = 0, z > y . By permuting, we have six solutions, each of weight 4.
For (24) x = 0, z = y. By permuting, we have three solutions, each of
weight 3.
For (25) we have six solutions for x, y , z.
For (26) we have y = 0, and so six solutions each of weight 4.
For (27) we have x = y = z = A , one solution.
In each of these cases, the number of solutions is three times the number
of the corresponding reduced forms. Hence the theorem follows.
It may be remarked that results such as Theorem 2 had their origin in the
researches of Hermite and Liouville who also found the number of solutions
+
for some assigned residues r of x y (mod 4). We note4 that N the number
of solutions of equation (17) is 3G(d) - 3F(d) when r = 0, and that N =
F(d) when r = 1,2, 3 , provided that d is not a perfect square when r is odd.
3. We now come back to the equation
z =f(x, r) = f d x , Y)/fa(X,Y ) , (28)
where,f,(x, y),.f2(s, y) are polynomials with integer coefficients which have no
polynomial as a common factor.
When h ( x , y ) and f2(x,y ) are homogeneous polynomials of the same
degree, there is no loss of generality i n supposing that (x, y ) = 1. Since the
possible common divisors d of fl(x, y ) , fi(x, y ) divide their resultant, the
finding of the integer solutions of equation (28) reduces to solvingf,(x, y ) = d
in integers. By Thue's Theorem, Chapter 22, there will be only a finite
number of integer solutions except possibly when f2(x, y ) is a power of a
linear or quadratic form.
We may suppose now that f , ( x , y), f2(x, y), are not both homogeneous
forms. In general, it is difficult to prove the existence of solutions of equation
(28). In some cases, an infinity of solutions can be found, e.g. whenf,(x, y ) =
xy. Then by division if need be, we may suppose without loss of generality
that
fl(X,Y) = g(x) + h(y) + c = zxy, (29)
where g(x), h ( y ) are polynomials in x, y respectively, without a constant term.
We may suppose that (x, y ) = 1, for if (x, y ) = d, then d I c, and so we have a
finite number of equations similar to (29) with (x, y) = 1.
The equation (29) is equivalent to the congruence
g(x) + h(y) + c = 0 (mod xy), (30)
or to the pair
g(x) + c = 0 (mod y), h(y) + c = 0 (mod x). (31)
The conditions (31) are necessary, but they are also sufficient. For suppose
g(x) + c = uy, h(y) + c = ox. (32)
Then a solution of (29) is given by
zxy = uy + ux - c, (x,y) = 1,
for z is an integer since (x, y ) = 1 and uy - cis divisible by x etc. The method
of solution of equation (29) will be sufficiently illustrated if we consider a
congruence arising in the discussion6 of integer solutions of a ternary cubic
equation, namely,
ax: + bxz + c = O(modx,x,), (33)
where (a, b, c) = 1. We have supposed (xl, x2) = 1 and we now also suppose
that (xl, c) = 1, (.'i2,c) = 1. These conditions involve no loss of generality.
Thus if (xl, c) = d, x1 = dX,, c = dc, and
ad3X: + bx; + dc, = 0 (mod dXlx,),
or ad2X: + bxjl/d + c1 = 0 (mod X,x,).
Since (XI, s2)= 1, we have d i x2 and so d I b.
11*
We now replace congruence (33) by the pair of congruences
bxi + c = 0 (mod x l ) , (34)

ax: + c = 0 (mod x,). (35)
The existence of an infinity of solutions of (34), (35) is proved by means of a
recursive algorithm for a sequence x l , x2, x3, . . , , and congruences for n 2 2
of the form
3
xn+z + An+, = 0 (mod xn+J (36)
+
x : + ~ An = 0 (mod x,+~). (37)
We satisfy (34) by putting
bxi +c= ~1x3, (38)
where x, will be presently assigned with (x2, c) = 1, and taking for x1 any
+
divisor of bxi c prime to c, e.g. x1 = 1. Clearly (x2, x3) = I.
To satisfy ( 3 9 , we see from (38),
a r-)3 + c = 0 (mod x2), (39)
or xz + ac2 = 0 (mod x,), (40)
and bxg + c = 0 (mod x3). (41)

From (40), (41), a particular solution for x l , x2, x3 can be deduced by taking
x2 = x: + ac2, b(xg + ac2)3 + c = 0 (mod x3),
and so + 1 = 0 (mod x3).
ba3c5
If we take for x3 any divisor of ba3c5 + 1, then (x3, ac) = 1 and so (x2,x3) = 1.
Take
x3 = ba3c5 + 1, x2 = (ba3c5 + 1)3 + ac2.
Then from (38)
=
~1x3 + c = b(x; + ac2))"+ c,
bx;
and so x 1 = bx! + 3abcax,5 + 3a2bc4xi + c.
We can deal more generally with (40), (41).
Write x$ + aca = ~ 2 x 4 .
Then b ('' :4ac2)3 + c = 0 (mod x3).
Suppose now that (x3, x4) = 1 which is so if (x4,ac) = 1.
Then xi+ a3bc5 = 0 (mod xg)
and xz + ac2 = 0 (mod xd).
These congruences are similar to (40), (41).
A particular solution is given by
x3 = x: + a36c5, a863c13 + 1 = O(modx,),
and then (x3,x,) = 1.
The process can be continued and thus an infinity of solutions arise for
s l , x2 as polynomials in a, b, c.
It can be shown that
A, = ahnbWv*, n = I, 2 , . . .,
where A,, p,,, v, are given by the sequences
h=0,1,3,8,21, ...
/L= -1,0,1,3,8 ,...
I)= 1,2,5,13,34,....
These are practically sequences of alternate Fibonacci numbers
0 , 1 , 1 , 2 , 3 , 5 , 8 , 1 3 , 2 1 , ....
4. Some interesting results can be found for the equation z = fi(x,y)/f2(x, y ) ,
whenf,(.u, y ) andf,(u, y ) are special quadratic polynomials. These are due to
Barnes7, Goldberg et Millss*lo~ll.There may be an infinity of com-
paratively trivial solutions. Apart from these, there are only a finite number of
possible values for z. For such 2 , we have a quadratic equation zf2(x,y ) =
f i ( x , y). The values of x, y are either finite in number or, if not, can be found
from a Pel1 equation or an equivalent recursive algorithm.
We consider the case when
f1(x,y) = ax2
f 2 k Y)
z =
= PXY
flk Y)/fZ(X,Y ) ,
+ bxy + cy2 + Clx + e . +f,
+ qx + r.Y + s,
~ } (45)
where the coefficients are integers, subject to ac # 0, and to the divisibility

conditions
a I 0))d, P,41, c I (b, e, P, r ) . (46)
When p = 0, the equations (45) can be replaced by the equation,
fXX, Y ) = cy2 + ey + f, f2(x, Y ) = qx + 3,
by a change of variables and division.
+
This defines the quadratic character of qx s (mod e2 - 4fc), and the
solution presents no difficulty.
Let now (x, y, z) = (uo, ul,u) be a solution, and so
flk u1) = uf2(x, u1) (47)
is a quadratic in x with roots x = uo, u2 where u2 is given by
U(U~ + 242) =PU~O + qu - bu, - d. (48)
From (46), u2 is an integer. Hence (x, y, z) = (u2, u l , u) is also an integral
solution of (47). Continuing in this way, we obtain a sequence. . . uo, ul, u2,
. . ., extending infinitely in both directions, and defined by
a(u2n + u2n+ 2) =~ u 2 +n1~ + qu - bu2n + 1 - d, (49)
C(U~,,-~ + u ~ , , , ~=)puanu + ru - buzn - e. (50)
Also (x, y, z) = (u2,,, u2,,* 1, u) are solutions of (45) for all integers n.
Lemma
If lu21 2 IuoI, then either
pul + q = 0, or luol < Ic/allalulI+ k,

where k is a constant depending only on the coefficients of fl(x, y), f2(x, y).
Since uo, u2 are the roots of the equation (47) in x, we have not only (48) but
also
auOu2= cuf + eul + f - u(rul + s).
Eliminating u from these equations, we find
auaf2(uo, ui) + auo(rui + S) = CPU? + Q(ui),
where Q(ul) is a quadratic polynomial with coefficients depending upon those
infdx, Y),fi(X, r). Since lu2l 2 IuoI,
ICP: + Q<ui>l 2 ui)l
lau2f2(~0, - lauo(rui + s)l
2 lauE(pu1 + q)1 - 21auo(rul + s)l.
If pu,+ q # 0, we deduce from this quadratic inequality in uo that
Ipui + 41 IuoI Irui + $1 + {Irui + s12 + I~-(PJ + q)(c& + Q(ui))l>12.
Hence b o l < k + Ic/allalull,
where k is a constant depending only on the coefficients of fl(x, y) and
fZ(X9 Y).
We now arrange the sequences uo, u l , uz, . . . in four classes:
I. Those where either f2(u2,, = 0 or fi(u,,,, uZnW1) = 0 for some
integer n. Such sequences exist if and only if the equationsf,(x, y ) = 0,
.fz(x,y ) = 0 have an integral solution. Then solutions of (45) exist for
every z, and there are at most four sequences for any z.
11. For these
P U Z ~ ~ I4+ = 0, rUZnt1 +s # 0
for some integer n. This implies
P 14, PS # qr, fi(uzn, - q / p ) = ~ P -S qr)/p,
and so the congruence
PfI(X1, - 4/P) = 0 mod (PS - qr)
has integral solutions. Then sequences exist for every integer z which can
be represented by pfl(x, - q / p ) / ( p s - qr) with integer x.
111. Similar results hold when
PUzn + r = 0, quzn + s # 0.
IV. These are sequences not of the comparatively trivial classes I, 11, 111.
We now prove
Theorem 3
There exist sequences of type IV for onIy afinite number of values of v.
On interchanging x and y if need be, we may suppose without loss of
generality that u1 is an element of the sequence with least absolute value, and
also that luol ,< Iuzl. Since the sequence (u,) is not of types I or II,pul q # +
0, and so from the lemma,
+
lull 6 luol 6 I ~ / a l l / ~ l u ~k.
l
Two cases arise. First,
IPOUlI + r u 1 + SI,
6 21PO
and so IPuoU1I 6 2(lql Ic/all/z + + 21qlk + 21sI.
1~1)1~11
If u1 # 0, then from this, luol 6 kl for some constant kl, and if u1 = 0,

1 ~ 0 1 6 k.
Hence uo, u1 are both bounded, and so there are only a finite
number of possibilities for u.
Secondly, IPuoU1l > 21quo + + $1.
Now Ifz(u0, d l > 3lPuoU,l, uou1 # 0,
and so
Then v is bounded since luo/ull< Ic/all/z + k.

We now discuss the solutions of equations (45) for given z = v.

Put
6 = b - pi, d = d - qv, e = e - rs, f = f - sv.
Then equation (47) becomes
ax2 + bxy + cy2 + dx + ey +f= 0,
and the divisibility conditions (46) give,
a I (b, d), c I (b, e).

Also equations (49) and (50) become
4u2n + Uzntz) = -b U2n-1 - d,

~ ( ~ 2 1n - + U2n t 1 ) = -b uZn- e.
On eliminating u1 from
a(uo + u2) = -bul - d,

auOuz= cul + eu, + f,
then G(uo,u2) = 0 where,
G(X, Y ) = X 2 + BXY + Y 2 + D X + D Y + F,
B = 2 - b2/ac, D = (2cd - be)/ac,
F = (cd2 + bl2Y - bde)/a2c.
From conditions (51), B and D are integers and so F is also an integer.
Eliminating uZnklfrom (52), (53), we obtain
ac(u2n + 2~2nt2+ ~ 2 n t 4 )= -bc(uzn+l + ~ 2 n + 3 )- 2cd

= b + be
~ - ~2cd. ~ ~ ~
Hence U2n + U2nt4 = -BUzn+z - D. (54)
It follows that x = uo, u4 are the roots of G(x, u2) = 0, and so
uOu4 = U: + D u +~ F. (55)
We may suppose now without loss of generality that luzl < luZnlfor all n,
and in particular that luol 2 Iu21, Iu41 2 Iu21. Then from equation ( 5 3 ,
uo = El42 + 6, u4 = &U2 + 6,
where 8 = -t 1, E = f 1, &Ua8 2 0,EU28 > 0. (56)
Substituting in (54) with n = 0 and in (55), we obtain
-B u ~- D = (E + E ) u ~+ 6 + a, (57)
+ D u +~ F = EEU; + ~ + E S U ~+ 66.
6 ~ ~ 2 (58)
From (56), all the non-zero terms on the right-hand side of (58) have the same
sign. Then for EE = k 1,
161 + 181 < ID1 + IFI,

and so there are only a finite number of possibilities for E , E , 6, 6.
For fixed E , E , 6, S, there are at most two values for uz from (57), (58),
unless
E + E = -B, 6 + 6 = -D,
EE = 1, ~ 8+
= D, 66 = F. (60)
Suppose that (60) is not satisfied. Then each value of u2 leads to at most one u
sequence. Hence there are only a finite number of u sequences. This also holds
for u2 = 0. The sequences can be found from the above.
When (60) is satisfied, then E = E = k 1. It can now be shown that there
are only a finite number of c sequences for fixed u except in some special cases
when the sequences are of order 1, 2, or 4, with one exception.
It can be shown from (54), that the L sequence is cyclic if B = 0, f 1, 2, and
then it has period 1, 2, 3, 4, 6, 8, 12.
We now give few particular results.
The equatiotz
x2 + y2 + 1 = xyz.
We suppose the u, in the sequence uo, ulr u2, . . . to be positive, u1 minimal
and u2 3 uo. Now (u0u2,u,) = 1. We cannot have u1 > 1, for then
uou2 3 (u, + 1)2 > u: + 1 = uou2.
Hence u1 = 1, u0u2 = 2, uo = 1, u2 = 2. The u sequence becomes 5, 2, 1, 1,
2, 5 , 13, . . . with z = 3. Hence x,y are consecutive terms in the sequence of
alternate terms of the Fibonacci sequence 1, 1, 2, 3, 5, 8, 13, 21.. . .
The equation
x2 + y 2 + x + y + 1 = xyz, x > 0, y > 0,
Here XIY2+Y+ 1, y J x 2 + x + 1,
and x, y are consecutive terms of the sequence 1, 1,3, 13,. . . where u, =

5 U , - l - U , - 2 - 1.
The equation
x2 + y2 - x - y + 1 = x y z , x > 0, y > 0,
Now xIya-y+ 1, ylx2-x+ 1, and x = y = 1.
We conclude by noting that when the divisibility conditions (46) are not
satisfied, the problem becomes a very difficult one. Thus for z(l + xy) =
+
x2 2ya, the only procedure seems to be to try if there is a solution for
various values of z, e.g.,7 when z = 43, x = 30905, y = 663738.
5. The equation xm - y" = 1
This equation seems to have been first noticed by Catalan who proposed in
1842 the
Conjecture
The only solution in integers m > 1, n > 1, x > 1, y > 1 of the equation
xm - y" = 1 (61)
is m=3, n=2, x=2, y=3. (62)
This still remains unproved. An alternative form of the conjecture is
I f p and q are prime numbers and x > 1, y > 1, a similar result holds for
xp - y* = 1. (63)
Very few really general results are known, and even when they are simple,
their proofs are usually rather complicated and often out of place here. Some
of the results give estimates for possible values of m, n,p , q.
Perhaps the most general result is Cassels' la
Theorem 4
I f in equation (63), p > q > 2 or q > p > 2, then q I x, p I y.
Another proof has been given by Hyyro 13. The result when p = 2 is due to
Nagell 14.
Most of the other results deal with the cases when some of the variables
p , q, x , y are given. Thus Le Veque l5 established a result proved later by
Cassels l6 as the
Theorem 5
Suppose that (x,y) are given in (61), and (62) is excluded, Let p, v be the
least positiue solution of
x' = 1 (mod Y), yv = 1 (mod X ) ,
30. MISCELLANEOUS RESULTS 30 1
where X , Y are the products of the odd primes dividing x , y respectively. Then
m=p,n=v,exceptthatm=2,n= Imayoccurifp=v= landx+ lis
apower of 2.
When m, n are given in equation (61), the equation has only a finite number
of solutions but these are not easily found. This is a particular case of Siegels
Theorem in Chapter 25. The result also follows from Mahlers16a result.
Theorem 6
The greatest prime factor ofa?cm+ byn, ab # 0, ( x , y ) = 1, m 3 2, n 3 3
tends to injinity as max (/XI), ( I y l ) tends to infinity.
Proofs for equation (61) when m, n are primes satisfying various conditions
have been given by LeVeque by using the methods for dealing with Fermats
last theorem.
Suppose next in equation (63) that p and q are given odd primes. Then
Hyyro l 7 has shown that all the solutions can be given explicitly in terms of the
convergents to the simple continued fraction for ql-lppl -liS.
There are many results for special values of m, n and some are now given.
+
The equation y 2 1 = x p , p an odd prime.
This was proved impossible for x > 1 in 1850 by Lebesgue18. A slightly
dissimilar proof was given by Cassels 16. Clearly y = 0 (mod 2), x = 1 (mod 2).
Then from
(1+ iy)(l - iy) = s,
1 + iy = ir(u + iv) r = 0, 1, 2, 3.
x = u2 + v2.
The factor i r can be absorbed in the pth power, and so we need only
consider r = 0. Then
Hence u = t I , and from a congruence mod 4, since v is even, u = 1 (mod 4)

and so u = 1. Then
P.P - 1 p . p - 1.p - 2 . p - 3 112 + ..
2! 4!
Then, since u is even, the general term can be written for k 2 1 as
(- l ) k - l p . p - 1 . p - 2 . p - 3 . . . . . p - 2k + 1. 1.2v2-
(65)
2! (2k - 2 ) ! 2k - 1.2k
To prove that equation (64) is impossible (which is obvious if ( p . p - 1)/2 is
odd since u is even), we need only show that for k > 1 the term (65) is
divisible by a higher power of 2 than f o r k = 1. The first two parts of (65) are
integers. Also 2 cannot occur in the reduced denominator of the third part
since, for k > 1,
2'"' > 1 + 2k - 2 > k.
The equation y 2 - 1 = xp
The special case p = 3 of the equation
ya - l = x P
(66)
was considered by Euler. The only solutions are x = 0, - 1,2.
We may suppose now thatp is a prime > 3. The first general results are due to
Nagel114*1g.He proved very simply that p must be = 1 (mod 8) and y =
0 (mod p ) . He showed that equation (66) leads to
y f 1 = 2xp, y T 1 = 2p-1x;,
and so XP1 - 2P-ZxP2 -- +

-1. (67)
Since y = 0 (modp), it is easy to see that (x + I , xp + 1) = p, and
+
(xp + l)/(x 1) = 0 (modp) but f 0 (modp2).
When p = 1 (mod 8), Nagell14 showed by cyclotomic considerations
applied to equation (68) that a necessary condition for solvability is that
u + u = 1 (mod 8), where u + vdi is the fundamental unit > I in the field
generated by dp. This condition is equivalent to his other one that 2 is a
biquadratic residue of p . Since Nagell proved the impossibility of equation
(66) when p = 3 , 5 , 7 (mod 8), he thus showed that equation (66) is impos-
sible for a set of primes of density t + t + 4 = 4. +
It may be noted that from (67),estimates (rather large ones) had been found
for the magnitude of possible p , xl, x2 by Oblath20, HyyrOl7 and InkeriZ3.
Some thirty years after Nagell's results, Chao KO24 proved the impossibility
of equation (66). We give his proof which for p = 3, 5, 7 (mod 8) is similar to
Nagell's but not so simple.
It easily follows from (66) that
+ 1 =py:, -
xp +1
x
x + l = PYi,
where y = p y , ~ ,is odd.
Suppose first that p = 5 , 7 (mod 8), and s o p = 8n + a , a = $ 7 . Then from
(67), x = a - 1 (mod 8). Write (66) as
y2 = (x2 - 1 +1 ) 4 +~ 1
= x a + 1 (mod x2 - 1).
Hence the quadratic character
(-)xax -+1 1 (-)x -21

= = 1.
This is impossible since x - 1 = 3, 5 (mod 8).

Suppose next that
p = 8n + 3 = 24m + a, a = 11, 19.
Now y2 = (x3 - 1 + l)*mx=+ 1,
= xa + 1 (modx3 - 1).
Hence (-)xax3 +- 11 = 1.
Suppose first that a = 11 and so x = 2 (mod 8). Sincexl' - x2 = x2(x9 - l),
1=
x2
(m)
=
+1
r+) = = (-)
-x - 1
(m)
x2 +1
=
2
(3 )
= -1.
Take a = 19, s = 2 (mod 8), and then

+1
= - (&)
x + l - 2
1 = (m)
XI9
= (m)
x3
= (sfi)= - l *
There remains p = 1 (mod 8), and now from (68), x = 0 (mod 8).
From (68),
py; = xp-1 - xp-2 + '*.-x + 1. (69)
Let q < p be a positive odd integer. Write p = kq + a, 0 < a < q, and so
(a941 = 1.
(-x)t -1
Put E(t) =
(-x) - 1'
and so E ( r ) = 1 (mod 8). By (68),

py; =
xp + 1 -
- -
Xkq+= +1
x + 1 x + l
Since x4 + 1 = (x + 1) E(q), this becomes
((x + 1) E(q) - 1)kxa + 1
PY; = x + l
since k =a + 1 (mod 2). Now
We show that the quadratic character (E(a)/E(q))= 1. We apply the Euclidean

algorithm
q = + r l , 0 < rl < Q
kla
a = k2r1 + r 2 , 0 < r2 < rl
rl = k3r2 + r 3 , 0 < r3 < r2
. . . . . . . . .
rs-l = kstlrs-l + rstl, 0 < rs < r s - l
rs = ks+2rs+1.
Then since E(kla + r l ) - E(rl) = 0 (mod E(a)),
From (70),
Since x = - 1 (modp) from (67),

(&))(y (;).
= =
We have a contradiction if q is taken as an odd quadratic non-residue of p .

This proves the result.
We conclude this section by mentioning the impossibility of the equations
y3 = xp + 1, 1x1 > 1, Nagell'*
y 3 = x p - 1, 1x1 > 2, NagellZ4
y4 = xp + 1, Selberg 25.
The last result is now a special case of Chao KO's theorem.
REFERENCES
1. Leon Bernstein. Zur Losung der diophantischen Gleichung m / n = l / x +
l/y +l/z insbesondere irn Fall m = 4.5. reineangew. Math., 211 (1962), 1-10.
2. R. Oblhth. Sur l'kquation diophantienne 4/n = l/xl 1/ x z+ +
1/x3.Mathesis.
59 (1949), 308-316.
3. K. Yamamoto. On the diophantine equation 4/n = l/x + l/y +
l/z. Mem.
Fac. Sci. Kyushu University. Ser. A, 19 (1965), 37-47.
3a. L. A. Rosati. Sull'equazione diofantea 4/n = l/xl + +
l/xz l/xg. Boll.
Union Mat. Ztal., (3) 9 (1954), 59-63.
4. L. J. Mordell. On the number of solutions in positive integers of the equation
y z + zx + xy = I?. Am. J . Marh., 45 (1923), 1 4 .
5 . R. F. Whitehead. On the number of solutions in positive integers of the
equation yz + zx + xy = n. Proc. Lond. Math. SOC.,(2), 21 (1923), xx.
+ +
6. L. J. Mordell. The congruence ax3 by3 c = O(modxy), and integer
solutions of cubic equations in three variables. Acra Math., 88 (1952), 77-83.
+ +
7. E. S. Barnes. On the diophantine equation x2 y 2 c = xyz. J . Lond.
Math. SOC.,28 (1953), 242-244.
8. K. Goldberg, M. Newman, E. G. Straus, and J. D. Swift. The representations
of integers by binary quadratic rational forms. Archiv Math., 5 (1954), 12-18.
9. W. H. Mills. A system of quadratic diophantine equations. Pacijc J . Math., 3
(1 953), 209-220.
10. W. H. Mills. A method for solving certain diophantine equations. Proc. Am.
Math. SOC.,5 (1954), 473-475.
11. W. H. Mills. Certain diophantine equations linear in one unknown. Can. J.
Math., 8 (1956), 5-12.
12. J. W. S. Cassels. On the equation ax - b y = 1,II. Proc. Camb. Phil. Soc., 56
(1960), 97-103. This contains a history of the conjecture and many references.
13. S. Hyyro. uber die Gleichung axm - by" = z und das Catalansche Problem.
Ann. Acad. Sci. Fennicae, Series A. 1, 355 (1964), 3 9 4 8 .
14. T. Nagell. Sur l'irnpossibilite de l'equation indeterminee zp +
1 = y2. Norsk
Mat. Forenings Skrifter., 1 (1921), Nr. 4.
15. W. J. Le Veque. On the equation ax - by = 1. Am. J. Math., 74 (1952),
325-33 1.
16. J. W. S. Cassels. On the equation ax - b y = 1. Am. J. Math., 75 (1953),
159-1 62.
+
16a. K. Mahler. On the greatest prime factor of axm by". Nieuw Arch. Wisk.,
(3) 1 (1953), 113-122.
17. S. Hyyro. Uber das Catalansche Problem. Ann. Univer. Turku, Series A, 79
(1964), 3-9.
18. V. A. Lebesgue. Sur I'impossibilite, en nombres entiers, de l'equation xm =
y2+ 1. N o w . Ann., Math., (1) 9 (1850), 178-181.
19. T. Nagell. Sur une equation a deux indCterminCes. Norsk Vid. Selsk Forh.,
7 (1934), 136-139.
20. R. Oblath. Sobre ecuaciones diofinticas imposibles de la forma xm 1 = y". +
Riu. Mat. Hispano-Am., 14 (l), (1941), 122-140. This contains many references.
21. C. L. Siegel. Die Gleichung ax" - by" = c. Math. Ann., 114 (1937), 57-68.
22. R. Oblith. Uber die Zahl x2 - 1. Muthematica Zutphen. B8 (1939), 161-172.
23. K. Inkeri and S. Hyyro. On the congruence 3P-' = 1 (mod$) and the dio-
phantine equation x2 - 1 = y p . Ann. Univer. Turku, Series A, 50 (1961),
34.
24. Chao KO. On the diophantine equation x2 = y" + 1, xy # 0. Scientia
Sinica (Notes), 14 (1964), 457-460.
25. T. Nagell. Des Cquations indeterminkes x2 + x + 1 = y" et x2 x + + 1=
3y". Norsk. Mat. Forenings Skr. Series 1, 2 (1921), 12-14.
26. S. Selberg. Sur I'impossibilitk de Yequation indeterminke zp 1 = y2. +
Norsk. Mat. Tidskrift, 14 (1932), 79-80.
Bibliography
References to books on Diophantine equations and to books on number theory
containing important sections on Diophantine equations.
1. P. Bachmann. Niedere Zahlentheorie, 11. B. G. Teubner (1910).
2. P. Bachmann. Das Fermatproblem in seiner bisherigen Entwicklung.
Walter de Gruyter & Co., Berlin und Leipzig (1919).
3. S. I. Borevich and I. R. Shafarevich. Number Theory. Academic Press,
New York and London (1966).
4. R. D. Carmichael. Diophantine Analysis. Mathematical Monographs
No. 13, Wiley, New York (1913). Reprint by Dover Corporation.
R. D. Carmichael. Analyse Indeterminee. Les Presses Universitaires, Paris
(1929).
5. B. N. Delone and D. K. Faddeev. The Theory of Irrationalities of the Third
Degree, Vol. 10. Translations of Mathematical Monographs, American
Math. SOC.,Providence, Rhode Island (1964).
6. L. E. Dickson. History of the Theory of Numbers, Vol. 2, Diophantine
Analysis. Carnegie Institution of Washington (1920). Reprint Chelsea.
7. L. E. Dickson. Introduction to the Theory of Numbers. Univ. Chicago
Press, Chicago (1929).
8. L. E. Dickson. Modern Elementary Theory of Numbers. Univ. Chicago
Press, Chicago (1 939).
9. A. 0. Gelfond. The Solution of Equations in Integers. Freeman & Co.,
San Francisco and London (1961).
10. G. H. Hardy and E. M. Wright. The Theory of Numbers. Clarendon Press,
Oxford (1 945).
10a. M. Kraitchik, Thdorie des nombres 111, Analyse diophantienne et applications
aux cuboides rationnels, Gauthier Villars et Cie, Paris (1947).
11. E. Landau. Vorlesungen iiber Zahlentheorie (I) und (111). S. Hirzel,
Leipzig (1927).
12. E. Landau. Diophantische Gleichungen mit endlich vielen Losungen-neu
herausgegeben von Arnold Walfisz. VEB. Deutscher Verlag der Wissen-
schaften Berlin (1 959).
13. S. Lang. Diophantine Geometry. Interscience, New York (1962).
14. W. J. LeVeque. Topics in Number Theory, (11). Addison-Wesley Pub-
lishing Co. Inc., Reading, Massachusetts (1956).
15. L. J. Mordell. Three Lectures on Fermats Last Theorem. Cambridge
University Press, Cambridge (1921).
L. J. Mordell. Le Dernier T h e o r h e de Fermat. Les Presses Universitaires
de France, Paris (1929).
16. L. J. Mordell. A Chapter in the Theory of Numbers. Cambridge University
Press, Cambridge (1947).
17. T. Nagell. Lanalyse indeterminee de degrd supkrieur . Gauthier Villars
et Cie, Paris (1929).
18. T. Nagell. Introduction to Number Theory. Almqvist and Wiksell,
Stockholm; John Wiley and Sons Inc., New York (1959).
19. W. Sierpinski. Elementary Theory of Numbers. Panstwowe Wydawnictwo
Naukowe. Warsaw (1964).
20. T. Skolem. Diophantische Gleichungen. Springer, Berlin (1938).
21. J. V. Uspensky and M. A. Heaslet. Elementary Number Theory. McGraw
Hill Book Co. Inc., New York (1939).
List of Equations and Congruences
Degree 1
+
alxl ~ 2 x 2+. . . U,X, = o + 30
L,(x) = a, (r = 1 , 2 , . . . , n ) 31
L,(x) = 0 (v = 1,. . ., m), (m < n) 32
Degree 2
Homogeneous
+
ax2 by2 + C Z ~+ 2 f y ~+ 2gzx + 2 h ~ y= 0 47
+ + = 0 5,43-48
by2 C Z ~
= O(modM) 36
x: + x2" = (4a + 3)x: 4
x2 + y 2 = 2z2 13
x2 + y 2 = p z 2 14
alx: + azxf + ~ 3 x 3+ a4x; = 0 49-51
x2 + y2 = + 22 w2 15
x: + xf - p(x: + xi) = 0 9
.y2+ = 2y2, 2Z7 = +
Z2 y2 21
1212
2arsxrys
r.s
= 0 42
alx: + +
~ 2 x 2+ . * * U ~ X ; = 0 51
x:+x;+...+x;=x2 12
Inhomogeneous
+ + +
ax2 bxy cy2 dx ey f = 0 + + 57-58
ax2 bxy+ + cy2 = m 159
+
ax2 by2 = m 7, 118-120, 164-169
y2 - Dx2 = 51, t-4 53-57
+
x2 y 2 = m 179
X: = u + bxz 3, 8
+
x2 ay2 + 6z2 = m 175-178
+ +
x2 y 2 z2 = m 4, 175, 179
X: ? X: f X: = n 181
YZ +zx + ~y = d 291-292
X; + ~f = 4x3+ 3 3, 8
+ +
xf 2x2" = 8x3 5 or 8x3 7 + 4
+
x: - 2xz = 8x3 3 or 8x3 5 + 4
+ + + +
U X ~ ~ x ~ x Z C X ~= d, U ~ X ? blx1~2 C ~ X ; = d1z2 59
y2 = Ul.K2 + bl.y + cl, = U ~ X ' + h2x + c2 70

y2 = x2 + k, z2 = .y2 - k 71-72
u n + 2 + sun+ 1 + bid, = C, U, a square 60-63
u, twice a square 63
x2 + bcy2 + caz2 + a b =~nz ~ 169-173
Degree 3
Homogeneous
+
ax3 bxZy + cxy2 + dy3 = ez3 226-228
+
ax3 bx2y + cxy2 + dy3 = z3 7
= z 3 ( m o d 9 ) 37
+
Ax3 By3 + Cz3 = 0 226,230
UX: + bpx; + cp2x; = 0 9
+
x3 y 3 = az3 124-130
+
x3 dy3 = 3z3 136
+
ax3 by3 + cz3 + ~ X Y Z= 0 8,78-80, 130, 226
u =b = 1 131-134
x3 + y3 + 2 3 - xyz = 0 80
UX' + by3 + cz3 + dw3 = 0 88
5x3 + 9y3 + 10z3 + 12w3 = 0 88
x: + 2x; + 4x; = 9x: 7
x: + 2x; = 7(x; + 2 x 3 8
fi(x1, ~ 1 ~ , 1 +
) ~ f 2 ( ~ ~2 , 2 ~ , 2 +) ~ 2 f 3 ( ~ 3~ 9 3 ~ , 3 +
) p3 2 almnxiyrnzn = 0 9
Inhomogeneous
ax3 + bx2y + cxy2 + dy3 = m 218
m =1 218-219, 226
+
ax3 + bx2y cxy2 + dy3 = 1 (mod 9 ) 38
x3 - 3xy2 - y 3 = 1 208-209
ax3 + by3 = c 219, 225
x3 + dy3 = 1 203-205, 219-225
+ + +
ax3 bx2y cxy2 dy3 = ex2 + f x y + gy2 67
y 3 = a3 + bx + ex2 + dx3 69
+
y 3 = a bx + cx2 + d3x3 69
ey2 = ax3 + bx2 + cx + d 235, 255
y2 = ax3 bx2 cx + + + d 199, 255-261
y2 = x3 - h2x - h3 143-146
py2 = x3 27 + 74
y ( +~ 1) = X ( X + +
l ) ( x 2) 257
3y(y +1 ) = +x(x l)(x 2) + + 258
6y2 = X ( X +1)(2x 1 ) + 258
6y2 = x3 5~ + 6 + 259
y 2 = x3 + k 135, 218, 238-254
y 2 = x3 + k (mod a) 41,48
LIST OF EQUAIIONS AND CONGRUENCES 309
+
y 2 = x3 k -continued
k = n3 - lm2 238-241
k = 2 8 - 3h3 244-245
k = 4u3 - 3h3 245
k = 3n3 - 3h3 245-246
k = -432 247
-144 251
-105 251
-78 251
-69 251
-48 247
-42 251
-33 251
-31 244
-24 246
-15 247
-7 247
-4 123
-2 124
I 247
7 242
17 246
34 242
45 239
58 242
70 242
+
X: = u bx2 3-4,9
+ +
ax3 by3 + cz3 dxyz = e 110
+
ux3 + uy3 bz3 = bc3 102
+ +
.x3 y3 z3 - U X ~ Z= b 106-110
+
ax3 + by3 + cz3 d=0 101-103
+ +
ax2 by2 cz2 = 2dxyz 6
+ +
u(x3 + y3) b(z3 c3) = 0 59
+
.x3 + y 3 z3 = KI 83, 100-101
n = l 84
(x + y + z ) - ~ dxyz = m 83
z2 = f ( x , y ) , fcubic 85-87, 103-106
+ + + + + + +
z2 = p 2 Ix my ax2 bxy cy2 Ax3 Bx2y Cxy2 + Dy3+ 104
+
z2 = p2 + 1.u + my ax3 + bxZy + cxy2 + dy3 103
+
= ax3 + bs2y r.t-y2 + dy3 112
z2 = u s 3 + h ~ 3 + ~c 104-106, 1 15-1 16
? = X 3 - G X Y 2 - G3Y3 233
z2 = .y3 +
1j3 235
2z2 = x3 + y3 235
y 2 - ax2 = z3 bz + c 113+
z3 = p 3 + Ix my + ax2 bxy + cy2+ + 103
+
z3 = ax2 by2 + c 106, 113-114
z3 = x2 + ky2 215-217
+ + +
x3 y3 z3 w3 = n 58, 182
x i + xi + .xi = 9x4 & 4 7
x: x3, + x," = n 181
+ +
+ . a * +
x2 y 2 1 = XYZ 299
+ +
x2 y 2 & x f y 1 = XYZ 299-300
Degree 4
Homogeneous
+ +
kl(ax2 by2 CZ2)(aiX2 blY2 + C1z2) = +
k2(px2 + qy2 + rz2)2 25
( v ) 3 7 ) " y . + (?)aq - bp z4 = 0 27
~ +4 (
( U ~ X : + ~ 2 x 2+ agxf + a&)(bixf + b 2 4 + bgxf + b 4 ~ 3
= 2k(cix; + C ~ X ;+ C ~ X ;+ c,x;)(dlxf + d 2 ~ :+ d 3 ~ :+ d 4 ~ 3 6
a,x: + uzx; + agx: + a4x: = 0 5, 90-94
U(X' - y4) + C(Z' - w4) = 0 91
x4 + y 4 = z4 + w4 90
x4 + y4 + 2 4 = w4 94
xy(mx2 + ny2) = zw(mz2 + nw2) 91
+
f ( z )+ f(y) fk)- 4(f(cl) f(u) + +
f ( ~ )=) 0 where
f ( s )= x: + +
x; x: - xgx; - x:x; - x f x f - x1xzxg(x1 x2 x g ) + + 10
Inhomogeneous
+
Ax4 Bx3y + Cx2y2 + Dxy3 + Ey4 = 1 236-237
ax4 - by4 = c 274
x4 - Dy' = & 1 275-276
x4 + y 4 = 1 116-117
x4 - 2y4 = 1 207
x4 - 8y4 = 1 208
+
ky2 = ax4 bx3 + cx2 + dx + e 268-272
y2 = ax4 + bx3 + cx2 + dx + e 69-70, 74, 77, 268
dy2 = ax4 bX2 + c + 63-64,236,268-272
py2 = 2x4 + I 74
py2 = 4x4 - 1 74
py2 = 8 - x4 74
y2 = Dx4 & 1 269-271
Y 2 = X4+ 1 17
+ + + +
y2(ax2 bx C) y(alx2 blx ~ 1 ) a2x2 b2x c2 = 0 + + + + 70
LIST OF EQUATIONS AND CONGRUENCES 31 1
268
ax4 + b 2 y + cx2y2 + dxy3 + cy4 = z2 138
ax4 + bx2y2 + dy4 = cz2 16-25
b =0 23-25
c = l 22
+
(ax2 by2j2 - 2k(cx2 +
dy'))" = z2 5
x4 - py4 = 2z2 25
x4 - 2y4 = + z 2 72-74
.r4 - y 4 = z 2 17
-x4 + .v4 = 2z2 18
x4 + kX2y2 + y4 = z 2 19
k=O 16
= +6 18
= 1 19
= -1 20
= 14 21
arsxrys= dz2 97
r,s=O
z2 = k2 + .x2(ax2 by2) + 97
ax4 + by4 + c = z 2 95
z 2 = U: +
U2U3,where Ur = arx2 h,xy + + b,y2 + f r x + g,y 96
s 4 + y4 + 424 = 1 94
+
.r4 + y 4 2 4 = 2 95
.r4 + ay4 + bz4 = z 2 95
+ +
( x ~ a)(y2 a) = (UZ' +
b2)2 95
Degree > 4
Homogeneous
XI + y' = Z' 137,280-286
Inhomogeneous
f ( x ) = m,f homogeneous of degree ci in n variables 174-180
+ +
aox" a l x n - l y + . . . anyn = m 157-163, 186-199, 218
ax" - by" = c 273-275
c= +I 274-275
X" - Dy" = & 1 274-276
n=5 275
f (x,y ) = 0 262-263, 276-277
y 2 = f ( ~ ) 263-265
+ +
ay2 by c = dx" 265
y m = x(x + 1). . .(x n)+ 267-268
in = 2 267
ax"' - by" = c 272-276, 300
y2 +
I = x"' 276, 301
y 2 - 1 = x"' 276, 302
y3=x - + 1 276, 304
y4 = X P +
1 304
X; = a +
bx2 3-4
P(x, Y ) = Q<x,Y ) 188, 198,278
Zrn = AX2 + +
B X Y CZ2 111
UX' +
by2 = CZ" 111
x2 +
y2 = 21 122
+ +
N(CLX p~ Y Z ) = n 209-21 1
+ +
N(ax py y z ) = f ( x , y , 2 ) 21 I
+ + +
z2 = (alx2 b,y2)(a2x2 b2y2)(u3x2 b3y2) 154
Miscellaneous equations
2 atb: =0 203
+ 7 = 2"
t=1
x2 205-206
Pure and Applied Mathematics
,4 Series of Monographs and Textbooks
Edited by PAUL A. SMITH and SAMUEL EILENBERG,
Columbia University, New York
1. ARNOLD SOMMERFELD. Partial Differential Equations in Physics. 1949 (Lectures

o n Theoretical Physics, Volume VI)
2. REINIIOLD I ~ A E RLinear
. .4lgebra and Projective Geometry. 1952
3. HERBERT BUSEMANN and PAULKELLY. Projective Geometry and Projective
hletrics. 1953
4. STEFANI ~ E R G M A N 31. SCHIFFER. Kernel Functions and Elliptic Differential
and
Equations in Mathematical Physics. 1953
5. RALPEIPHILIPBOAS,JR. Entire Functions. 1954
6. HERBERTBUSEMANN. The Geometry of Geodesics. 1955
7. CLAUDECHEVALLEY.Fundamental Concepts of Algebra. 1956
8. SZE-TSENHu. IIomotopy Theory. 1959
9. A. 11. O~TROWSKI. Solution of Equations and Systems of Equations. Second
Edition. 1966
10. J. D I E U D O N NFoundations
~. of Modern Analysis. 1960.
11. S. I. GOLDBERG. Curvature and Homology. 1962
12. SICURDUR HELGASON. Differential Geometry and Symmetric Spaces. 1962
13. T. H. HILDEBRANDT. Introduction to the Theory of Integration. 1963
14. SHREERAW ABIIYANKAR. Local Analytic Geometry. 1964
15. RICHARD I,. BISHOPand RICHARD J. CRITTENDEN. Geometry of Manifolds. 1964
16. STEVENA. GAAL.Point Set Topology. 1964
17. BARRYMITCHELL. Theory of Categories. 1965
18. ANTHONY P. MORSE. A Theory of Sets. 1965.
19. GUSTAVE CHOQUET.Topology. 1966
20. %. I. BOREVICII and I. R. SIIAFAREVICII. Number Theory. 1966
21. Jose LVIS MASSERA and JUAN .TORGE SCHAFFER.Linear Differential Equations
and Function Spaces. 1966
22. RICHARDD. SCIIAFER. An Introduction to Nonassociative Algebras. 1966
23. MARTINEICIILER.Introduction to the Theory of Algebraic Numbers and Functions.
1966.
24. SIIREERIM ABEIYANKAR. Resolution of Singularities of Embedded Algebraic
Surfaces. 1966
25. F R A N ~ TREVES.
OIS Topological Vector Spaces, Distributions, and Kernels. 1967
26. PETER D. LAX and RALPHS. PHILLIPS. Scattering Theory. 1967
27. OYSTEIN ORE. The Four Color Problem. 1967
28. MAURICE HEINS. Complex Function Theory. 1968
29. R. M. BLUMENTHAL and R. K. GETOOR. Markov Processes and Potential Theory.
1968
30. L. J. MORDELI,. Diophantine Equatioris. 1969
I n preparation
HANSFREUDENTHAL
and H. DE VRIES.Linear Lie Groups.
J. DIEUDONNB.
Foundations of Modern Analysis (enlarged and corrected printing)
WILLIAM F. DONOGHUE, JR. Distributions and Fourier Transforms
MARSTONMORSEand STEWART S. CAIRNS.Critical Point Theory in Global and Differen-
tial Topology.
J. BARKLEY ROSSER.Simplified Independence Proofs: Boolean Valued Models of Set
Theory.

L.J. Mordell Diophantine Equations

Uploaded by

Copyright:

Available Formats

L.J. Mordell Diophantine Equations

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

L.J. Mordell Diophantine Equations

Uploaded by

Copyright:

Available Formats

DIOPHANTINE

ACADEMIC PRESS London and New York

PRINTED IN GREAT BRITAIN BY WILLIAM CLOWES AND SONS LTD

is given of relevant results or principles required, for example, of some

Nov. 1968 L. J. MORDELL

23. Local Methods or p-Adic Applications 200

I have much pleasure in acknowledging my very great indebtedness to many

Feb. 1969 L. J. MORDELL

There is now a 1-1 correspondence between the rational values of z, and

These can be written as the single equation

The general problem suggests many questions. Can we find reasonably

2. Necessary conditionsfor solvability.

Equations Proved Impossible by

1. We now consider the

is impossible in integers fi f ( x l , x2, . . ., x,) has a prime factor p which cannot

and this is impossible. Similarly if x = 0 (mod 2), y = 1 (mod 2).

Since to mod 3, x3 = x, x2 = 1, we have

x: + xi - p(xg + x:) = 0, p = 3 (mod 4), (xl,x,, x3,x,) = 1 (28)

= -2(1 + 2X2+ 2X3) (mod 4)

f&> + f(.) + f(u>- 4(S(9") + fW) + f(m>= 0.

Equations Involving Sums of Squares

t(x: - x: - * . .- XE-1) = 2xnXn,

and (2) follows.

Clearly d = ( 2 X Y , Y 2 - X2,Y 2 X)= 1 or 2. +

Then X = u2 - b2, Y = 2ab or X = 2ab, Y = a2 - b2.

Hence w - z = 2f, w + z = 2(x2 + y"/f, (12)

The complete solution is given by

Quartic Equations with only Trivial Solutions

1. We shall consider some quartic equations of the form f ( x 2 ,y 2 , z") = 0

This is a case of equation (4) and since a < x , we have a contradiction.

If y = 0 (mod 2), then

Also u= l k p < dX,

Hence ab = 0 from equation (6) and then xy = 0.

or + 2x2 + y2)(2z - 2x2 - = 3y4.

and so is impossible. The second gives

and so 3x2 - Y2 = 2(x + y)2 = 2u2,

3Y2 - x2 = 2(x - y)2 = 2v2,

whence u2 + 3 v 2 = 4Y2, v 2 + 3u2 = 4x2.

Suppose first that x and y are not both odd.

or .Y = Ad,, b = d,B, y = 2d2B, a = d2A.

Hence ab = 0, x = y giving the solution

and so x2(222 - y2) = wZ(2y2 - z2),

Put xz = a, yw = 6, 2c = xy + wz, 2d = xy - wz.Since obviously x,y , z , 11

The second gives

I f y = 1 (mod 2), then x = z = 1 (mod 2) and x4 = - 2 , 4 (mod 8). Hence

and so x2 f z = 2a4, x2 T z = 8pb4, y = 2ab,

The - sign must be rejected since d4 k 3c4 = - 1 (mod 8) is impossible.

by the law of quadratic reciprocity and since (2/p) = 1. Hence z = s2 (mod p )

Fl = 4u, + b , + c,, a, + 46, + c,, a, + b, + 4c, (mod 8)

Hence all the odd factors of A are = 1 (mod 8).

(uqr + brp + cpq) (;:+ :+ 3rz

and from F = 2k,w2, this is

L = -qc - rb 1 M=- ra - pc i N = -Pb - qa

should be = 1 (mod 8). We take q = r = -1 (mod 8) arbitrarily such that

Some Linear Equations

1. We now discuss some linear equations.

alx, + a2x2 = a, (a,, a2) = 1, (2)

Continuing this process, we have a decreasing set of positive integers

We can take N t o be the integer part of n (A,)l(n-m).

It is sometimes useful t o know that linear forms can be made t o satisfy