Professional Documents
Culture Documents
CAMBRIDGE
UNIVERSITY PRESS
Published by the Press Syndicate of the University of Cambridge
The Pitt Building, Trumpington Street, Cambridge CB2 1RP
40 West 20th Street, New York, NY 10011-4211, USA
10 Stamford Road, Oakleigh, Melbourne 3166, Australia
v
vi Contents
0.6 Algebraic integers 39
0.6.1 Definition 39
0.6.2 Basic properties 40
0.6.3 Class numbers 41
0.6.4 Ideal numbers and ideals 42
0.7 The reception of ideal theory 44
0.7.1 How the memoir came to be written 44
0.7.2 Later development of ideal theory 45
Acknowledgements 47
Bibliography 48
Part two: Theory of algebraic integers 51
Introduction 53
1 Auxiliary theorems from the theory of modules 62
§1. Modules and their divisibility 62
§2. Congruences and classes of numbers 64
§3. Finitely generated modules 67
§4. Irreducible systems 71
2 Germ of the theory of ideals 83
§5. The rational integers 83
§6. The complex integers of Gauss 84
§7. The domain o of numbers x + y/ 86
§8. Role of the number 2 in the domain o 89
§9. Role of the numbers 3 and 7 in the domain o 91
§10. Laws of divisibility in the domain o 93
§11. Ideals in the domain o 95
§12. Divisibility and multiplication of ideals in o 98
3 General properties of algebraic integers 103
§13. The domain of all algebraic integers 103
§14. Divisibility of integers 105
§15. Fields of finite degree 106
§16. Conjugate fields 108
§17. Norms and discriminants 111
§18. The integers in a field Il of finite degree 113
4 Elements of the theory of ideals 119
§19. Ideals and their divisibility 119
§20. Norms 121
§21. Prime ideals 123
§22. Multiplication of ideals 125
§23. The difficulty in the theory 126
§24. Auxiliary propositions 128
Contents vii
§25. Laws of divisibility 129
§26. Congruences 134
§27. Examples borrowed from circle division 138
§28. Classes of ideals 146
§29. The number of classes of ideals 147
§30. Conclusion 149
Index 153
Part one
Translator's introduction
Translator's introduction
3
4 Translator's introduction
times, and their basic theory was laid down in Euclid's Elements (see
Heath (1925)) around 300 BC. Yet even ancient number theory contains
problems not solvable by Euclid's methods. Sometimes it is necessary
to use irrational numbers, such as v,"2-, to answer questions about the
ordinary integers. A famous example is the so-called Pell equation
x2-cy2=1
where c is a nonsquare integer and the solutions x, y are required to be
integers. Solutions for certain values of c were known to the ancients, but
the complete solution was not obtained until Lagrange (1768) related the
equation to the continued fraction expansion of He also showed that
each solution is obtained from a certain "minimum" solution (xo, yo) by
the formula
xk+ykf -±(xo+yo.)k.
The irrational numbers xk + in this formula are examples of alge-
braic integers, which are defined in general to be roots of equations of
the form
0.2 Squares
0.2.1 Pythagorean triples
Integers a, b, c such that
a2 + b2 = C2
into primes p1, P2, , pz and q1, q 2 ,... , qj respectively, where one of the
primes p is different from all the primes q.
Nonunique prime factorisation is ruled out by the following proposi-
tion of Euclid (Elements, Book VII, Proposition 30).
Prime divisor property. If p is prime and p divides the product ab of
integers a, b, then p divides a or p divides b.
An interesting aspect of the proof is its reliance on the concept of
greatest common divisor (gcd), particularly the fact that
gcd(a, b) = ua + vb for some integers u, v.
The set {ua + vb : u, v E 7L} is in fact an ideal, and unique prime factori-
sation is equivalent to the fact that this ideal consists of the multiples
of one of its members, namely gcd(a, b).
It should be mentioned that Euclid proves only the prime divisor prop-
erty, not unique prime factorisation. In fact its first explicit statement
and proof are in Gauss (1801), the Disquisitiones Arithmeticae, article
16. As we shall see, this is possibly because Gauss was first to recognise
generalisations of the integers for which unique prime factorisation is
not valid.
8 0.2 Squares
0.2.3 Irrational numbers
As everybody knows, Pythagorean triples also have significance as the
sides of right-angled triangles. In any right-angled triangle, the side
lengths a, b, c satisfy
a2 + b2 = c2
0.2..4 Diophantus
The equations a2+b2 = c2 and x2-2y2 = 1 are examples of what we now
call Diophantine equations, after Diophantus of Alexandria. Diophantus
lived sometime between 150 AD and 350 AD and wrote a collection of
books on number theory known as the Arithmetica (Heath (1910)). They
0. 2.4 Diophantus 9
This gives
and hence
4
x1 +y 4= 2
1 - zl.
Thus we have found another sum of two fourth powers equal to a square,
and by retracing the argument we find that the new square z2 is smaller
than the old, z2, but still nonzero. By repeating the process we can
therefore obtain an infinite descending sequence of positive integers,
which is a contradiction.
Fermat called the method used in this proof infinite descent, and used
it for many of his other theorems. He claimed, for example, to have
proved that any prime of the form 4n + 1 is a sum of two squares by
supposing p = 4n + 1 to be a prime not the sum of two squares, and
finding a smaller prime with the same property. However, it is very hard
to see how to make the descent in this case. Euler (1749) found a proof
only after several years of effort. In 0.3.4 we shall see an easier proof
of the two squares theorem due to Lagrange. Lagrange's proof does
use another famous theorem of Fermat, but it is the easy one known
as Fermat's "little" theorem: for any prime number p, and any integer
a 0- 0 (mod p), we have aTi-1 - 1 (mod p) (Fermat (1640b)).
The proof of Fermat's little theorem most likely used by Fermat uses
induction on a and the fact that a prime p divides each of the binomial
coefficients
p p(p - 1)(p - 2) (p - i + 1)
i - i!
for
--
1 < i < p - 1,
as is clear from the fact that p is a factor of the numerator but not of
the denominator. This proof implicitly contains the "mod p binomial
theorem",
(a + b) P - a1' + b'' (mod p),
which has its uses elsewhere (see for example Gauss's proof of quadratic
reciprocity in 0.5.4).
The proof more often seen today is based on that of Euler (1761),
12 0.3 Quadratic forms
which implicitly uses the group properties of multiplication mod p, par-
ticularly the idea of multiplicative inverses. An integer a is nonzero
mod p if gcd(a, p) = 1, in which case 1 = ar + ps for some integers
r, s by the Euclidean algorithm. The number r is called a multi-
plicative inverse of a (mod p) since ar - 1 (mod p). It follows that
mod p multiplication by a nonzero a is invertible, and in particular the
set {a x 1, a x 2, ... , a x (p - 1)} is the same set (mod p) as the set
{ 1, 2, ... , p - 11. Hence each set has the same product mod p,
ax2ax...x(p-1)a-lx2x...x(p-1) (modp),
and cancellation of 1, 2, ... , p - 1 from both sides (which is permissible,
since 1, 2, ... , p - 1 have inverses) gives Fermat's little theorem:
aP-1 - 1 (mod p).
Since numbers that end in 3 or 7 are of the form 10n + 3 or lOn + 7, and
0.3.3 Reduction of forms 13
[y s-]-[0 1
since the product of a substitution and its inverse is the identity. There-
fore, taking determinants of both sides,
a a' /3'
=1.
y S y' S'
Finally, since the determinants on the left are integers and the only
integer divisors of 1 are ±1, it follows that the invertible substitutions
for which the pairs (x', y') run through Z x Z are precisely those with
a, 0, y, S E Z and determinant aS - Qy = f1. Such substitutions are
now called unimodular.
The result a'xj2+b'x'y'+c'y'2 of a unimodular substitution a'x'+/3'y'
for x and y'x' + S'y' for y in ax 2 + bxy + cy2 is therefore a form that
takes the same values as ax 2 + bxy + cy2. Forms transformable into
each other by unimodular substitutions are equivalent, as we would say,
because the unimodular substitutions form a group. Lagrange observed
that equivalent forms have the same discriminant
D=b2-4ac=b'2-4a'c',
as can be checked by computing bj2 - 4a'c' and using aS - 0y = ±1.
(Incidentally, the old term for the discriminant of a quadratic form was
determinant. I have retained this term in the translation of Dedekind's
memoir because he refers to a slightly different definition, due to Gauss.)
Observing the invariance of the discriminant is a first step towards decid-
ing equivalence of forms. To go further we need to answer the question:
how many inequivalent forms have the same discriminant?
Lagrange found a way to answer this question for forms with negative
discriminant. He showed that any ax 2 + bxy + cy2 can be transformed
into an equivalent form a'x'2 + b'x'y' + c'y'2 that is reduced in the sense
that Ib'I < a' < c'.
It follows that
-D=4a'c'-bj2>4aj2-a'2=3a'2
and therefore, in the case of negative discriminant, only finitely many
values of the integers a', b' and hence c', can occur in reduced forms.
For any particular D < 0 it is then possible to work out the inequivalent
reduced forms of discriminant D. The number of them is called the class
number h(D). The first few calculations yield the following results.
0.3.4 Lagrange's proof of the two squares theorem 15
and pairing the first factor with the third, and the second with the
fourth. Lagrange's own identity can be derived quite mechanically from
(x2 + 5y2)(xj2 + 5yj2) = (xx' - 5yy')2 + 5(xy' + yx')2
(Brahmagupta's identity for c = 5), and
(the result of completing the square on 2x2 + 2xy + 3y2). Use the latter
to rewrite each of 2x2 + 2xy + 3y2 and 2x'2 + 2x'y' + 3yj2, in the form
2 [X2 + 5Y2}, multiply them out using Brahmagupta's identity, then
absorb the factors of 2.
There is a related identity for the product of the two different forms
of discriminant -20:
(x2 +5 Y2) (2x'2 + 2x'y' + 3 yj2) = 2X2 + 2XY + 3Y2
0.3.7 The class group 19
where
X = xx' - yx' - 3yy',
Y = xy' + 2yx' + yy'
and it can be derived in a similar way. These identities show that the
forms x2 +5 y2 and 2x2 + 2xy + 3y2 are "closed under products" in a
certain sense. The product operation is known as composition of forms.
Legendre (1798) managed to show, in fact, that any two quadratic
forms with the same discriminant could be "composed" in this fashion.
Something very interesting was going on, but what?
x=a3-6ab2,
1 = 3a2b - 2b3
only if b = ±1 and a = ±1, since 1 and -1 are the only integer divisors
of 1. This gives x = 5 as the only positive integer solution for x. Q.E.D!
Euler gave several examples of this kind, generally splitting quadratics
into irrational complex factors and treating the factors as integers. Why
is this permissible, if indeed it is? To answer this question we need to
recall how ordinary integers behave, particularly as divisors.
The behaviour of the ordinary integers is ruled by unique prime fac-
torisation, which in turn depends on the prime divisor property: a prime
divides a product ab only if it divides one of a, b (0.2.2). For example,
this is crucial in proving that relatively prime integers are squares when
their product is a square (0.2.1). To justify similar propositions about
"quadratic integers" such as x + , we have to decide which of them
are "primes" and then see whether they have a prime divisor property
like the ordinary integers. This is easiest when the quadratic integers in
question have a "Euclidean algorithm", because one can then follow the
trail blazed by Euclid in his proof of the prime divisor property.
The first to carry out such a program was Gauss (1832), who studied
the divisibility properties of the numbers x + y , where x, y E Z.
These numbers are now known as the Gaussian integers. They are the
simplest kind of quadratic integers, and they tie up nicely with quadratic
forms and some other threads in our story, so it is worth looking at them
first.
The division property is clear as soon as one realises that the multiples
p,3 of 3 lie at the corners of a lattice of squares in the complex plane. A
typical square is the one with corners 0, ,3, i,3, (1 +i)/3. The remainder
p is the difference between a and the nearest corner p/3 in the lattice,
so 0 < lpl < 1,31 because the distance IpI between any point in a square
and the nearest corner is less than the length 1,31 of a side.
To find gcd(a, 0) one can therefore let a = a1, ,3 = N1 and repeatedly
compute
aj+1 = ,aj,
Oj+i = remainder when aj is divided by (3
until a zero remainder /3k is obtained. (We use j, k.... for indices from
24 0.4 Quadratic integers
now on, since i will be reserved for .) Then gcd(a, /3) = 3k, and
there are Gaussian integers p and v such that gcd(a, /3) = pa + v/3 by
an argument like that used for Z (0.2.2). In fact, the rest of the route
to unique prime factorisation is essentially the same as in Z. We get the
prime divisor property by arguing that
1 = gcd(7r, a) = µ7r + va
when it is a Gaussian prime not dividing a, and multiplying both sides
by /3. Unique prime factorisation (up to unit factors) is obtained by
supposing
7172 ... 7rr = 0102 ... Os
are two prime factorisations of the same number, and cancelling 7r1, 7r2, .. .
in turn until only units remain.
Thanks to unique prime factorisation, and the fact that the units of
Z[v-2] are ±1, the proof that relatively prime numbers are cubes when
their product is a cube is essentially the same as in Z. This in turn is
like the proof for squares mentioned in 0.2.1.
If I wanted to proceed with the use of imaginaries in the way that earlier
mathematicians have done, then one of my earlier researches which is very
difficult [composition of forms] could have been done in a very simple way.
(Excerpt from Kummer (1846a) in Edwards (1977), p. 335.)
y3 = z3 - x3
and then factorises the right hand side into
(z-x)(z- (3x)(z-(3x)
where
-1+v/ 53--1
(3 =
2
2_
are the imaginary cube roots of 1. At this point Euler makes an inter-
esting mistake. He wants to argue that the factors z - x, z - S3x, z - sax
are relatively prime factors of a cube, hence cubes themselves, but uses
this argument in the wrong setting - in Z[v/-3] rather than Z[(3]. It so
happens that unique prime factorisation fails in Z[-3], as can be seen
from the example
4=2x2=(1+')(1-"),
whose factors all have minimal norm, 4, for nonunit elements of 7G[.
However, unique prime factorisation is valid in Z[(3], for geometric rea-
sons like those that apply to Z[i] and Z[/], and Euler's idea can be
made to work.
Sn = cos - -
2a
n
+ i sin
21r
n
is a complex nth root of 1. It was first studied by Gauss (1801), for
geometric reasons. The points (n, (n2, ... , (n = 1 are equally spaced
0.5.2 The cyclotomic integers 31
around the unit circle, hence the name "cyclotomic", from the Greek for
"circle-dividing". Since
( n-1+...+Cn+1=0,
(S3+(2)2+(S3+(2) - 1 =0,
since ((3 + (2)2 = (4 + 2(5 + ( and (5 = 1. Thus we have a quadratic
equation for (3 + (2, which can be solved by square roots, and we can
then find 3 and (2 individually by solving quadratic equations, since
we know their sum and product.
Of course, if p - 1 is not a power of 2 one cannot repeatedly halve
the number of terms in the Gauss sums until only one term remains.
However, in the nontrivial case where p is odd, the first halving is always
possible. From now on we shall use (, without subscript, to denote an
arbitrary root of the cyclotomic equation. The set of all roots (k is
partitioned into those for which k is a square mod p, and those for
which k is a nonsquare. As we know from the existence of primitive
roots (0.3.5), these sets are of equal size. In fact, we know that the
squares mod p are the even powers of a primitive root mod p, and the
nonsquares are its odd powers. In the case p = 5, 2 is a primitive root,
so its odd powers are 2 and 23 = 8 - 3 (mod 5), and the corresponding
Gauss sum is the one used above: (3 + (2. We could also have used its
complement, S4 + (, whose exponents are the squares mod 5.
In the Disquisitiones, article 356, Gauss showed that the sums whose
exponents are the squares and nonsquares mod p are the two roots of
the equation
2p41 =0,
X +xf
with the - sign when p - 1 (mod 4) and the + sign when p = 3 (mod
4). Since the roots are -2 ± 4 in the former case and -2 ±i$ in the
latter, it follows that, if a is a primitive root mod p,
(a2i (a2j+1 ±Vf-
when p - 1 (mod 4),
i
(azi - Ca29+1
=
P
when p - 3 (mod 4).
i
The left hand sides of these equations can be written more concisely with
the help of the Legendre symbol or quadratic character symbol defined
by
k +1 if k is a square mod p
p -1 if k is a nonsquare mod p.
34 0.5 Roots of unity
Quadratic characters have the following multiplicative property,
(k
\p/ \p/ - pl/ '
which follows immediately from the fact that the squares mod p are
precisely the even powers of a primitive root.
In terms of quadratic characters, the Gauss sum on the left of the
equations is simply S = EP-1 (p) (k, and Gauss's theorem is equivalent
to:
S+2 _ f +p if p = 1 (mod 4)
-p if p 3 (mod 4).
P-1-1
S (l
k-1 \ k) (k,
? \p/ ()Ck+t
1
P-
S2 = k,l-1
P-1
E kl ck+l
k,1=1 p
by the multiplicative property of quadratic characters.
Now each k 0- 0 (mod p) has a multiplicative inverse mod p (0.3.1),
hence as k runs through the nonzero congruence classes mod p, so does
kl, for fixed 1 # 0 (mod p). We may therefore replace k by kl, and
P-1
S2 = kl2 ((k1+1
k,1=1 p
P-1 (k) S1(k+1)
7,
- k,t=1 p
since k is a square mod p t* kl2 is a square mod p
0.5.3 Cyclotomic integers and quadratic integers 35
P-1
p - 1) (iP + E (k)
P-2 P-1
(t(k+l)
1 pJ k=1 p1=1
1 2
1 +
1=1 `pl k=11 -
P
since l(k + 1) for fixed k < p - 1 runs through the
nonzero congruence classes
_1 -2-2 (kl
(p (-1)
= \ p / (p - 1) + k=1 p/
by the cyclotomic equation
(tJ)p
p
since half the (P) are +1 and half are -1, by 0.3.5
-p p1
since p - 1 -1 (mod p).
This surprising relationship can be extracted from the qth power of the
Gauss sum S = EP-=1 (k) (k by using the "mod q binomial theorem"
As mentioned in 0.3.1, this holds for q prime because in that case all the
binomial coefficients (i), ... , (q? 1) are divisible by q. This leads to the
"mod q multinomial theorem"
(x1+x2+...+xk)qxi+xz+...+xq (mod q)
Sq (k
P
PE-1 (k q
(kq
(mod q)
k=1 p)
(,)n
= E (k)c-kq since _ (P for any odd power n
(p)2
(kq since (p)2 = 1
_
_ ()q)
k=1
P-'
p
(p)
S
0.5.4 Quadratic reciprocity 37
since, for fixed q, kq runs through the nonzero congruence classes. We
now multiply both sides by S and use SZ = (Pl) S, obtaining
C
l
p/
a
if q - 1 (mod 4),
if q - 3 (mod 4),
k
1 if k=a3n
if k = a3n+1
q 3
C3 if k=a3n+2
It is greatly to be lamented that this virtue of the real numbers [i.e. the rational
integers], to be decomposable into prime factors, always the same ones for a
given number, does not also belong to the complex numbers [i.e. the integers of
cyclotomic fields]; were this the case, the whole theory, which is still laboring
under such difficulties, could easily be brought to its conclusion. For this
reason, the complex numbers we have been considering seem imperfect, and
one may well ask whether one ought not to look for another kind which would
preserve the analogy with the real numbers with respect to such a fundamental
property. (Translation by Weil (1975) from Kummer (1844).)
The first sentence is the one quoted by Dedekind in Latin in his in-
troduction: "Maxime dolendum videtur ..." (Kummer's paper was one
of the last important mathematical works written in Latin.) Dedekind
goes on to say:
But the more hopeless one feels about the prospects of later research on
such numerical domains, the more one has to admire the steadfast efforts of
Kummer, which were finally rewarded by a truly great and fruitful discovery.
This was the discovery of ideal numbers. The rest of the story can be
left to Dedekind himself. He explains Kummer's ideal numbers with
admirable clarity, and his own distillation of the concept of ideal from
them. He also gives a candid account of the difficulties en route to unique
prime factorisation of prime ideals. The concept of ideal is very simple,
and so are the concepts of divisor and product, but unfortunately it is
not clear that if ideal b divides ideal a then a = be for some ideal c.
Dedekind was stymied by this difficulty for a long time, and here he
explains how he overcame it after several years of struggle. The memoir
is not only a superb exposition, but also a rare opportunity to see a great
mathematician wrestling with a problem, with blow by blow comments
up to the final victory.
44 0.7 The reception of ideal theory
0.7 The reception of ideal theory
0.7.1 How the memoir came to be written
Dedekind's first exposition of ideal theory, Dedekind (1871), included
proofs of unique prime ideal factorisation and finiteness of the class
number, together with a very impressive application of the theory: a
correspondence between the quadratic forms of discriminant D and the
ideals of Q(om), under which the product of ideals corresponds to com-
position of forms. Replacement of the complicated and mysterious forms
by objects that behaved like integers should have been a revelation, but
Dedekind's contemporaries were slow to appreciate his achievement.
There was great resistance to the idea of treating infinite sets as math-
ematical objects. Dedekind had tried this in 1857 when he introduced
congruence classes in place of specific residues, and again in 1872 with
his definition of real numbers as Dedekind sections. But he was fighting
2000 years of tradition (plus a formidable modern opponent, Leopold
Kronecker). The "horror of infinity" that had haunted mathematics
since Zeno was not to be dispelled overnight. Most mathematicians
were not even willing to consider a theory based on infinite sets, let
alone appreciate its power or elegance.
Nevertheless, Dedekind tried again. The memoir on algebraic integers
first appeared in instalments in the Bulletin des sciences mathematiques
et astronomiques in 1877. The reason Dedekind chose this unusual outlet
was a letter he received from Rudolf Lipschitz in 1876:
Acknowledgements
I am indebted to George Francis for sending me a copy of the Lipschitz-
Dedekind correspondence when it was not available in Australia. Thanks
also to Peter Stevenhagen and John McCleary for corrections and com-
ments, and to two anonymous reviewers at Cambridge University Press
for valuable advice.
48
Bibliography 49
=0,
with finite degree n and rational coefficients a1, a2, ... , an_1i an. It is
t Elements, VII, 32.
t Theoria residuorum biquadraticorum, II; 1832.
53
54 Introduction
called an algebraic integer, or simply an integer, when it satisfies an equa-
tion of the f o r m above in which all the coefficients a1, a 2 ,-- . , an_1i an
are rational integers. It follows immediately from this definition that the
sum, difference and product of integers are also integers. Consequently,
an integer a will be said to be divisible by an integer /3 if a = /3ry, where
ry is likewise an integer. An integer a will be called a unit when every
integer is divisible by e. By analogy, a prime must be an integer a which
is not a unit and which is divisible only by units a and products of the
form ea. However, it is easy to see that, in the domain of all integers
we are considering at present, primes do not exist, since every integer
which is not a unit is always the product of two, or rather any number,
of integral factors which are not units.
Nevertheless, the existence of primes and the analogy with the do-
mains of rational or complex integers re-emerges when we restrict our-
selves to an infinitely small part of the domain of all integers, in the
following manner. If 9 is an algebraic number there is, among the in-
finitely many equations with rational coefficients satisfied by 0, exactly
one
on +a19n-1 +...+ an-10 +an = 0,
of minimal degree, and which we call for this reason irreducible. If xo,
Si, x2, ..., xn_1 denote arbitrary rational numbers, the numbers of the
form
0(9) = xo + x19 + x292 + + xn-19n-1,
the set of which we call 0, will also be algebraic numbers, and they enjoy
the fundamental property that their sums, differences, products and
quotients also belong to the set 0 . I call such a set St a field of degreet n.
The numbers 0(9) belonging to a field 1 are now partitioned, following
the definition above, into two large sets: the set o of integers and the
nonintegral, or fractional, numbers. The problem we set ourselves is to
establish the general laws of division that govern such a system o.
The system o is evidently identical with the system of all rational
integers when n = 1, or with the complex integers when n = 2 and
0 =. Certain phenomena which occur in these two special domains
o occur again in every domain o of this nature. Above all, the unlimited
decomposition which prevails in the domain of all algebraic integers is
never encountered in a domain o of the kind indicated, as one easily
sees by consideration of norms. If we define the norm of any number
t Dedekind calls it "finite, of degree n". (Translator's note.)
Introduction 55
f The legitimacy, or rather the necessity, of such demands, which must always be
imposed with the introduction or creation of new arithmetic elements, becomes
more evident when compared with the introduction of real irrational numbers,
which was the subject of a pamphlet of mine (Stetigkeit and irrationale Zahlen,
Brunswick, 1872). Assuming that the arithmetic of rational numbers is soundly
based, the question is how one should introduce the irrational numbers and define
the operations of addition, subtraction, multiplication and division on them. My
58 Introduction
This problem is essentially simplified by the following considerations.
Since a characteristic property A serves to define, not an ideal number
itself, but only the divisibility of the numbers in o by the ideal number,
one is naturally led to consider the set a of all numbers a of the domain
o which are divisible by a particular ideal number. I now call such a
system an ideal for short, so that for each particular ideal number there
corresponds a particular ideal a. Now if, conversely, the property A of
divisibility of a number a by an ideal number is equivalent to the mem-
bership of a in the corresponding ideal a, one can consider, in place of
the properties A, B, C... defining the ideal numbers, the corresponding
ideals a, b, c,..., in order to establish their common and exclusive char-
acter. Bearing in mind that these ideal numbers are introduced with
no other goal than restoring the laws of divisibility in the numerical
domain o to complete conformity with the theory of rational numbers,
it is evidently necessary that the numbers actually existing in o, and
which are always present as factors of composite numbers, be regarded
first demand is that arithmetic remain free from intermixture with extraneous
elements, and for this reason I reject the definition of real number as the ratio
of two quantities of the same kind. On the contrary, the definition or creation of
irrational number ought to be based on phenomena one can already define clearly
in the domain R of rational numbers. Secondly, one should demand that all real
irrational numbers be engendered simultaneously by a common definition, and
not successively as roots of equations, as logarithms, etc. Thirdly, the definition
should be of a kind which also permits a perfectly clear definition of the calculations
(addition, etc.) one needs to make on the new numbers. One achieves all of this
in the following way, which I only sketch here:
1. By a section of the domain R I mean any partition of the rational numbers
into two categories such that each number of the first category is algebraically less
than every number of the second category.
2. Each particular rational number a engenders a particular section (or two
sections, not essentially different) in which each rational number is in the first or
second category according as it is smaller or larger than a (while a itself can be
assigned at will to either category).
3. There are infinitely many sections which cannot be engendered by rational
numbers in the manner just described. For each section of this kind one creates
or introduces into arithmetic a special irrational number, corresponding to the
section (or engendered by it).
4. If a, ,3 are any two real numbers (rational or irrational), then one easily
defines a > 03 or a < 33 in terms of the sections they engender. Moreover, one can
easily define, in terms of these two sections, the four sections corresponding to the
sum, difference, product and quotient of the two numbers a, 0. In this way the
four fundamental arithmetic operations are defined without any obscurity for an
arbitrary pair of real numbers, and one can really prove propositions such as, for
example, f = f , which had not previously been done, as far as I know, in
the strict sense of the word.
5. When defined in this way, the irrational numbers unite with the rational
numbers to form a domain 91 without gaps and continuous. Each section of this
domain 91 is produced by a particular number of the domain itself; it is impossible
to engender new numbers in this domain 91.
Introduction 59
62
§1. Modules and their divisibility 63
6±6' = (afa')+()3±,3);
and, since the numbers a ± a' are in a and the numbers ,3 ±,3' are in b,
the numbers 6 ± 6' also belong to the system a. That is, a is a module.
Since the number zero is in every module, all the numbers a = a + 0
64 Chapter 1. Auxiliary theorems from the theory of modules
of the module a and all the numbers ,Q = 0 +,3 of the module b belong
to the module D. Consequently, the latter is a common divisor of a and
b. Also, if the module D' is any common divisor of a, b, so that all the
numbers in a and all the numbers in b are in D' then (by virtue of 1) all
the numbers a+,3, that is, all the numbers in the module D, also belong
to the module D'. Thus D is divisible by D'.
Having carried out these rigorous proofs, we need not explain further
how the notions of least common multiple and greatest common divisor
can be extended to any number (even an infinity) of modules. Neverthe-
less, it may be useful to justify the terminology chosen, by the following
remark. If a, b are two particular rational integers, m their least com-
mon multiple and d their greatest common divisor, it follows from the
elements of number theory that [m] will be the least common multiple,
and [d] the greatest common divisor, of the modules [a], [b]. In any case
we shall soon see that the number-theoretic propositions relevant to this
case can also be deduced from the theory of modules.
p = a (mod a),
and all such numbers w form a class of numbers modulo the module m.
If there is a number w satisfying the two congruences then the numbers
w - p, w - a will be in a, b respectively, and since the latter are both
in a, the difference p - or of the numbers will likewise be in Z. That
is, the above condition p - or (mod i) is necessary. Conversely, if this
condition is satisfied then (by virtue of the definition of a in §1,4) there
is a number a in a and a number 3 in b whose sum a +,3 = p - ar, hence
the number w = p - a = a +,3 satisfies the two congruences. Thus the
condition is also sufficient. Moreover, if w' is a number satisfying the
same conditions as w, then w' - w will also be in both a and b, and hence
also in m, which means that w' = w (mod m). Conversely, each number
w' in the class represented by w modulo m will satisfy the congruences.
Q.E.D.
13= y1 N1 +Y2132+y3133+"'+yn13n,
where yi, y2, y3, .... yn are arbitrary rational integers, evidently form a
module, which we call a finitely generatedt module [131, /32,133, ... , Qn]
The complex of constants /31 i /32,133, ... , /3n will be called a basist of the
module.
This module [/31i /32, ... , /3n] is evidently the greatest common divi-
sor of the n finitely generated modules [/31], [ 32], ... , P,,]. It is easy to
see that each multiple of a finitely generated module is itself a finitely
generated module, but here I confine myself to proving the following
fundamental theorem, which will later have important applications.
2. If all the numbers in a finitely generated module b = [/31,132,. .. , )3n]
can be transformed into the members of a module a by multiplication
by nonzero rational numbers, then the least common multiple m of a
t Dedekind calls them "finite". (Translator's note.)
t Note that Dedekind's basis elements need not be independent. However, he re-
quires them to be independent for fields (§15). (Translator's note.)
68 Chapter 1. Auxiliary theorems from the theory of modules
and b will be a finitely generated module, and one can choose a system
of 2(n + 1)n rational integers a such that the n numbers
µl = a1/3l
µ2 = al /31 +a2/32
a2")Q2 a3(n)03
An = ain)/31 + + + ... + a(n)/3n,
form a basis of m, and at the same time
(b, a) m) = a'1a2a3' ... ann),
By hypothesis there are n nonzero fractions
Si 82 83 Sn
t1, t2,
t3 tn)
whose numerators and denominators are rational integers, such that the
n products
81 82 83 8n
/31, -/7 /33, ...7
on
ti t2 t3 tn
belong to the module a. Since members of a module a are changed into
other members of a when multiplied b y rational integers tl, t2, t3, ... , to
(§1,1), the products s1/31, 82/32, 83/33, ... , sn/3n likewise belong to a, and
if s denotes the absolute value of the product 818283 sn, the numbers
s/31, 8/32i 8,03, .... s,3n, and consequently all products s/3, belong to the
module a, where /3 denotes any number in the module b.
Now let v be a particular index from the sequence 1, 2, ... , n. Among
the numbers in the module [,31,/32, ... , /3,] divisible by b let
Then if we put
i (v) , (v) i (v)
yl = yl - xval , y2 = y2 - x"a2 , ..., yv-1 = yv-1 - xvav-1
the number
theorem. Each number µ in the module in, that is, each number Mn in
both a and b = [,31,(32, ... , /n], is of the form
A = µn-1 + xn/Ln
where xn is a rational integer and µn_1 is a number belonging to the
two modules a and [31,02 , ... , /3n_l], and hence also to the module M.
Each number µ'n-1 of this nature is of the form
Al = xuILl
where xl is a rational integer. Thus it is proved that each number µ of
the module m can be represented in the form
of the module b for which the rational integers zi, z2, . . . , z, satisfy the
n conditions
0<z',<a(").
We shall show that these numbers ,0', the number of which is evidently
equal to a' a2 a,(" ), form a complete system of representatives of the
module b modulo m (or a).
In the first place, all the numbers Q' in the module b are incongruent
modulo m. If
4f1 + ... + zn/33 = z',01 + ... + zn'fn (mod a),
then the numbers z' l', z2 , ... , z;' satisfy the same n conditions as the
numbers z'1, z2',. .., zn. Then if the n differences
zn - zn, n I II
zn-1 - Zn-11 Z2/ - z2, z1 - z1
are not all zero, let z - be the first of them with a nonzero value,
a value which we can assume to be positive by symmetry, and which is
also < aU") since both numbers z' and z' are < a( v). Then the difference
(zi - zi)al + ... + (z' _ zv)QV
is evidently a number p' in a and [01,02 , ... , .3n] for which the coefficient
of 3, is positive and < a,(,", contrary to the definition of the number
µ,,. Thus any two different systems of n numbers z'1, z2..... zn, which
satisfy the conditions above, also produce two numbers 0' in the module
b which are incongruent modulo a.
In the second place, it is easy to see that an arbitrary number
0 =x101+x202+ ..+znOn
in the module b is congruent modulo a (or m) to one of the numbers
,0' since, if z1, z2, ... , zn are given, it is clear that we can successively
choose n rational integers
xn,xn-1,...) x2ix1
so that the n numbers
zn = zn + ann)xn,
(n) (n-1)
zn-1 = zn-1 + an-lxn + an-1 xn-1 ,
§4. Irreducible systems 71
....................................
z2 = z2 + a2n)xn + a2n-1)xn-1 ... +
a2x2,
(n) xn+a1(n-1) i
z1 =z1+a1 x1,
we have
)3' = 0 +x1/- 1 +x2/ 2 +... +xnAn,
W. Irreducible systems
1. A system of n numbers al, a2i ... , an will be called an irreducible
system, and its members will be called independent, when the sum
a=x1a1+x2a2+"'+xnan
is nonzero for any system of rational numbers x1i x2, ... , xn which are
not all zero. It then follows that any two different systems of ratio-
nal numbers x1i x2, ... , Xn produce unequal sums a. In the contrary
case, that is, when there is a system of rational numbers x1i x2, ... , xn,
not all zero, for which the sum a is zero, then the system of numbers
a1, a2i ... , an will be called reducible, and the numbers themselves will
be called dependent on each other. If one wants to retain this terminol-
ogy in the case n = 1, a single number evidently forms a reducible or
irreducible system according as it is zero or not. The definition above
easily yields the following theorems, whose number can be increased
enormously, on the determinants of rational numbers.
2. If the n numbers a1, a2,. . ., an are independent, then the n num-
bers
is nonzero or not.
Since al, a2i ... , an are independent, the sum
Cnxl+C'n'x2+...+Cnn)xn=0,
which is impossible when C is nonzero. Hence the numbers al, a2, ... , an
are independent in that case. But if we have C = 0 there is always a
system of rational numbers xl, X 2 , ... , xn satisfying the preceding equa-
tions, and not all zero. This is seen immediately when all the n2 coeffi-
cients c vanish. If this is not the case then, among the minor determi-
nants of C that do not vanish there will be one, say
n
f C1iC2 ... Cr(r)
r
xr = pr+lxr+l + +pnr)xn,
where the r(n-r) coefficients p are rational numbers. Now if we give the
n - r quantities xr+l, ... , xn arbitrary rational values then not only can
we ensure that they are not all zero, the quantities x1,. .. , xr will likewise
take rational values. Thus we have a system of n rational numbers
X17 X21 ... , xn, not all zero, for which the sum a' is zero. Hence in this
case the n numbers a', a'2, ... , a'n are dependent. Q.E.D.
§4. Irreducible systems 73
aµ =rlµ)w1
where the system of coefficients
rl,I I
r2, ..., ,
rn,
(r) rlIf, r2If i ..., rn,
... ... ., .,
(m) (m)
r1 , r2 , ..., rn(m)
consists of rational numbers, at least one of whose
m(m-1)...(m-n+1)
1 2 n
n x n partial determinants R is nonzero. Otherwise, any n of the m num-
bers aA would be dependent. Conversely, it follows from the hypothesis
of the theorem, that the m numbers aµ can always be expressed in terms
of n independent numbers wV, by choosing the latter to be, for example,
n numbers among the m numbers aµ which form an irreducible system.
Then, since the n + 1 numbers a.., wl, ... , wn are dependent there is an
equation, for each index p, of the form
with rational coefficients a for which the product aia2 ann is nonzero.
If we now re-express the numbers 3v in terms of the n numbers w we
can conclude that the assertion above is true, which at the same time
proves the theorem.
6. To the preceding proof I add the following remarks. Since the m
numbers a,, form a basis for the module a just as much as the n numbers
a',,, there are m equations of the form
aµ = p1µ>a1 +p2µ) a2 + ... + p, an,
and n equations of the form
a'
where the 2mn coefficients p and q are all rational integers. By substi-
tuting the first expressions in the second, and bearing in mind that the
n numbers a' are independent, we deduce that the sum
qp' , + q Pv, + ... + q(m)p(,7)= 1 or 0,
according as the m indices v, v' from the series 1,2,. .. , n are equal
or not. Then if P denotes any n x n partial determinant formed from
the system of coefficients (p), and if Q denotes any determinant formed
similarly from the system of coefficients (q), then we know that the sum
E PQ,
§4. Irreducible systems 77
taken over all combinations of n different upper indices, is equal to 1, and
consequently the determinants P have no common divisor. Conversely,
this property of the determinants P is necessary if the n numbers a,,,
and also the m numbers
aµ = plµ) al + ... + pn ) an
are to form a basis of the module a.
A system of coefficients such as (p) is evidently just a special case of
the preceding coefficients (r). Now since the n numbers a can likewise
be represented in the form
aV = e1V)wl + e2V )w2 + ... + e(V)wn,
They are of the form R° = R1 + xR2, from which we easily deduce that
the greatest common divisor E of the determinants R is the same as
that of the determinants R°. Thus the determinants R° cannot simul-
taneously vanish. We shall now use these basis transformations of the
module a as follows:
The m coefficients rnµ) of the number wn cannot all be zero, since
then all the determinants R would be zero. Now if two of these co-
efficients, say rn and rn, are nonzero, and if Ir,'n I > jraI, then we can
choose a rational integer x such that I rn + xrn I < I rn . t The elementary
transformation above therefore gives us a new basis in which all the m
coefficients rnµ), except the first, rn, remain the same, and this single
coefficient is replaced by one of smaller absolute value. By repetition of
this procedure we necessarily arrive at a basis in which all but one of
the m coefficients of wn are zero. We denote the member of the basis
a(nn)
for which the latter coefficient is nonzero by
an =a (n)
1 wl +a 2n)w2 + ... + a(n)wn,
and m - n = s numbers ai , a2 , ... , a9' which are all zero, and which can
therefore be omitted. The n nonzero coefficients a(V) can be taken to
be positive, since a' can be replaced by -a' without alteration of the
t Here again is the same principle which is fundamental in the theory of rational
integers.
W. Irreducible systems 79
giving
r' = 21, r" =7,
i
r"'=9, r""=8
i ,
fir)
2= 0,,
r' r"
2 = 7, 2-
r2 - -3 , r""
2 = 2.
whence, conversely,
Since 61i 62, 63, 64 are the quantities which, in the general theory, we have
denoted by ai, a2, ai, a2, we have
pi = 1, pi = 6, pi/ _ -2, P, - 2,
(p) i n-7
p2 - 0, p2 - , p2
m=-3 , AT an- 2 .
Thus we obtain, for the determinants proportional to the R,
(P) P1,2= 7, P1,3 = -3, P1,4 = 2,
1 P3,4 2,
= P2,4 = -2, P2,3 = -4,
and likewise
I qi = 1, q1 = 0, q1 f - 0, g111 - 0,
q2 = 0, q2 - 1, q2 - U, q2 3,
and
Q1,2 = 1, Q1,3 = 0, Q1,4 = -3,
(Q)
Q3,4 = 0, Q2,4 = 0, Q2,3 = 0
Finally, from the systems of coefficients
f hi = 0, hi = 0, hip = 1, hip' = 0,
h2/1 = 1,
(h) l h2 = 0, h2 = 3, h2I _ -2,
and
ki = _2 kl = _1, k1I - 1 k111 5,
(k) f k2- 2, k2 =-2, k2' -0: k2"=7
we derive the determinants Hµ,µ, = Q µ and K,.,µ, = Pµ,µ comple-
mentary to Pµ,µ, and Qµ,µ' respectively,
82 Chapter 1. Auxiliary theorems from the theory of modules
H1ni
,2 = h'1na
h2 - h'1mi nnh2" - h1nh2nn H1,4 = h1
h2m> H1,3 = h1 n ni
> h2 - h1inh2,
n
(H){ i n n i
H3,4 = h1h2 - h1 h2, m i ' w ' mi nn i
H2,4 = h1 h2 - h1 h2 , H2,3 = h'h1111 - h1 h2,
k'1 k2"-kil2o
(K) k' kill
K34=k'k2
1 2
-k'i12
k K2,4=k'1"ki
1 2 - 1 2 ' K2,3 1 '
83
84 Chapter 2. Germ of the theory of ideals
factor. This property depends essentially on the theorem that a prime
divides a product of two factors only when it divides one of the factors.
The simplest way to prove these fundamental propositions of number
theory is based on the algorithm taught by Euclid, which serves to find
the greatest common divisor of two numbers.t This procedure, as we
know, is based on repeated application of the theorem that, for a positive
number m, any number z can be expressed in the form qm + r, where
q and r are also integers and r is less than m. It is for this reason that
the procedure always halts after a finite number of divisions.
The notion of congruence of numbers was introduced by Gauss.t Two
numbers z, z' are called congruent modulo the modulus m, written
z - z' (mod m),
when the difference z - z' is divisible by m. In the contrary case z and z'
are called incongruent modulo m. If we arrange the numbers in classes,
with two numbers in the same class§ only if they are congruent modulo
m, then we easily conclude from the theorem recalled above that the
number of classes is finite and equal to the absolute value of the modulus
m. This also follows from the studies of the preceding chapter, since the
definition of congruence in Chapter 1 contains that of Gauss as a special
case. The system o of all rational integers is identical with the finitely
generated module [1], and likewise the system m of all numbers divisible
by m is identical with [m]. The congruence of two numbers modulo m
coincides with congruence modulo the system in. Thus (by §3,2 or §4,4)
the number of classes is (o, m) = ±m.
where all the Greek letters denote rational integers. And it follows im-
mediately, by virtue of the equation µa2 = µ'a'2, that the four numbers
fi, f2, gi, g2 appearing in the relations (3) will likewise be integers.
These decompositions are simplified if we make the additional assump-
tion that a is prime to b and c, since this implies p = p' = 1, a = a'
and hence the fifteen numbers can be expressed as follows, in terms of
five numbers a, /31, /32, yl, y2:
t Since these decompositions do not seem obvious to me, I include the following
proof of the consequences of (1) as an example. Note first that abi = d2 and
b1 b2 = b2 are both squares. Suppose that
a = pat, bh = µ1t1, b2 = p202,
where µ, Al, 92 are squarefree. Then abl = µµ1a2(31 is not a square unless
IA = µl. Similarly, b1b2 is not a square unless Al = µ2. Thus in fact µ = µl = p2
and hence
a=µa2, bh =u,Qi, b2 = µ(32
Forming products of these, we get
2
dl =
2
- abl = p a tl
22 dl =path,
d2 = ab2 =µ2a212 d2 = µa(32,
b2 = bhb2 = p2t1t2 b = ptlt2,
which completes the proof of the decompositions claimed by Dedekind. (Transla-
tor's note.)
§8. Role of the number 2 in the domain o 89
and at least one of the two rational integers x1, yl will be odd. If both
are odd, w1 will be divisible by a and we shall have
W1 = xl - 5y1 + 2x1y19 = 2w2,
f Luckily, since, for example, trying analogously to determine the role of the number
2 in the domain of numbers x + y/ leads to complete failure. Later we shall
clearly see the reason for this phenomenon.
§9. Role of the numbers 3 and 7 in the domain o 91
the product of a and ,(31 i and 1- 0 like the product of a and /32. In fact
this presumption is plainly confirmed: each number w = x + y0 divisible
by 1 + 0 is in fact divisible by a and 31, because
x+y0=(1+0)(x'+y'0)
implies
x=x'-5y', y=x'+y',
and consequently
x =- y (mod 2), x =- y (mod 3).
Conversely, each number w = x + y0 divisible by a and '31, that is, sat-
isfying the two preceding congruences, is also divisible by 1 + 0, because
we have y = x + 6y' and consequently
x+y0= (1+0)(x+5y'+y'0).
We can now also introduce the powers of the ideal prime numbers ,131,
,132, as we have done above for powers of the ideal number a. By analogy
with the theory of rational numbers, we define divisibility of an arbitrary
number w by flln or 02n by the respective congruences
//pi) w(1-0)n -0 (mod 3n),
Q2 ) w(1 + 0)n - 0 (mod 3'' ),
and this yields a series of theorems which agree perfectly with those of
the theory of rational numbers. We treat the ideal prime numbers y1,
rye in the same way.
where x, y, z1, z2 are rational integers, are closed under addition, sub-
traction and multiplication. The domain o' of these numbers contains
the domain o, and all the ideal numbers needed for the latter can be
replaced by actual numbers of the new domain o'. For example, by
putting
a=,31+/32, 7'1 =2,31+/32, 7'2 =)31+2,32
all the equations (4) of §7 are satisfied. Likewise, the two ideal prime
factors of the number 23 in the domain o are replaced by the two actual
numbers 2,31 - /32 and -,31 + 2,32 of the domain o', and it is the same
for all the ideal numbers of the domain o.
Although this way is capable of leading to our goal, it does not seem
to me as simple as desirable, because one is forced to pass from the given
domain o to a more complicated domain o'. It is also easy to see that
the choice of the new domain o' is highly arbitrary. In the Introduction
I have explained in detail the train of thought that led me to build this
theory on quite a different basis, namely on the notion of ideal, and it
would be superfluous to come back to it here; hence I shall confine myself
to illustrating the notion by an example.
The five ideals above, a, b1i b2, c1, c2 are evidently of this form, since
(b + 9) can also be replaced by -(b + 0).
The set of all numbers conjugate to the numbers in an ideal m is
evidently also an ideal
ml = [ma, m(-b + 0)].
Two such ideals in, m1 may be called conjugate ideals.
Let p be any number in the domain o. The system [p, p9] of all
numbers divisible by p forms an ideal which we call a principal ideal,t
and which we denote by o(p) or op. It is easy to give it the above
form [ma, m(b + 9)]; m is the greatest rational integer that divides p =
m(u + v9) and we have, moreover,
with the help of the methods indicated in the first chapter (§4, 5 and 6).
It is clear from the definition that the product mm' is a finitely generated
module with basis consisting of the four products
mm'aa', mm'a(b' + 9), mm'a' (b + 9),
mm'(b + 9)(b' + 9) = mm'[bb' - 5 + (b + b')9],
of which only two are independent. Thus for the ideals considered above,
for example
b1=[3,1+9}, C2=[7,3-0],
we find the product
b1c2 = [21,9-39,7+79,8+29].
This module is derived from the one considered at the end of the first
chapter (§4,6), and by setting w1 = 1, w2 = 9 we conclude
b1c2 = [21,-17+9] _ [21,4+9] = o(4+9).
In the same way we obtain all the following results, completely analogous
to the hypothetical equations (4) of §7:
0(2) = a2,o(3) = bib2, o(7) = c1c2;
o(-2+0) = bi, o(-2 - 9) = b2;
o(2+30) = ci, o(2-30) = c2;
o(1 + 9) = ab1, o(1 - 9) = ab2;
We have
as
m" = pmm' a' = 2 = 4q
and b" is determined by the congruences
qb" = q'b', qlb"
- q'b, q"ibii = bb' - 5
P
(mod a").
It is also necessary to note the special case where m' is the ideal m1
conjugate to m. The preceding formulas then yield the immediate result
mm, = oN(m).
The two notions of divisibility and multiplication of ideals are now con-
nected in the following manner. The product mm' is divisible by both
m and m' since, by property II of ideals, all the products pp' whose
factors belong respectively to m, m' are in both these ideals; the same
then holds for the product ideal itself. Conversely, if the ideal m" _
[m"a", m"(b" + 0)) is divisible by the ideal m = [m, m(b+ 0)] then there
is exactly one ideal m' such that mm' = m". In fact, if we let ml denote
the ideal conjugate to m and form the product
of finite degree n with rational coefficients al, a2i ... , an_1 i an. If the
coefficients of this equation are rational integers, that is, numbers from
the sequence 0, ±1, ±2,..., then 0 is called an algebraic integer, or simply
an integer. It is clear that the rational integers are also algebraic integers
and, conversely, if a rational number 0 is at the same time an algebraic
integer then, by virtue of a known theorem, it will also be one of the
rational integers 0, ±1, ±2,.... From the definition of integers we easily
derive the following propositions:
1. The integers are closed under addition, subtraction and multipli-
cation, that is, the sum, difference and product of any two integers a, 3
are also integers.
Proof. By hypothesis, there are two equations of the form
O(a) = as + piaa-1 + ... + pa-la + pa. = 0,
gl,ab-1 +
V)(3) _ ,3b + ... + qb-1Q + qb = 0,
in which all the coefficients p, q are rational integers. We now put ab = n
103
104 Chapter 3. General properties of algebraic integers
and let
w1,w2,...,wn
denote the n products as /36' formed from one of the a numbers
1,a,a2 ,...,a a-1
and one of the b numbers
P,Q2......
1, (36-1
where the coefficient of the highest degree term is unity and the others
a, /3, ... , e are integers, is likewise an integer.
Proof. By hypothesis, the coefficients a, ,(3, . . . , e are roots of equations
O(a)
=as+pla' 1+...+Pa =0,
W(Q) _Ob+ql)3b-1+...+qb=0,
.................................
X(C) = Ee + s1Ee-1 + ... + se = 0,
n = mab e and let w1, w2, ... , wn denote all the n products of the
form
e'
wm'aatOb' ...
where the exponents are rational integers satisfying
0<m'<m, 0<a'<a, 0<b'<b, ..., 0<e'<e,
then it is easily seen that, with the help of the equations F(w) = 0,
O(a) = 0, 0(0) = 0, ..., x(e) = 0, all the products WW1, WW2, ... , wwn
reduce immediately to the form
k1w1 + k2w2 + ... + knwn,
where k1, k2, ... , kn are rational integers. It then follows, as in the pre-
vious proof, that w is an integer. Q.E.D.
It follows from the latter theorem that, for example, if a is any integer
and r, s are positive rational integers then B a' is also an integer.
is a n y polynomial in t with rational coefficients xo, x1, x2, ... , xn_1 and
degree < n. We first remark that each such number w = 0(9) is uniquely
expressible in this form, by virtue of the irreducibility of f (t). We then
see easily that the numbers w are closed under rational operations, that
is, addition, subtraction, multiplication and division. For the first two
operations this evidently follows from the common form 0(9) of all the
numbers w, and for multiplication it suffices to remark that each number
of the form '(9), where V)(t) is a polynomial of any degree with rational
coefficients, is likewise a number w, because if we divide fi(t) by f (t)
the remainder will be a function 0(t) of the kind above, and at the same
time we have 0(9) = 0(9). Finally, to treat the case of division it is
enough to show that if w = 0(9) is nonzero then its reciprocal w-1 also
belongs to the system Q. Well, since ¢(t) has no common divisor with the
irreducible function f (t), the method for finding the greatest common
divisor of the polynomials f (t), 4(t) gives, as we know, two polynomials
fi (t), 01(t), with rational coefficients, satisfying the identity
9' to be any root of the equation 0 = f (9'), and replace each number
w = 0(9) in the field Sl by the corresponding number w' = 0(9'), then
this substitution will really be an isomorphism of S2, that is, it will satisfy
conditions (1) and (2). To show this, we let 01(t), 02(t), ... denote any
particular functions of the form 0(t). Now if we have
a=01(0), 0=02(0), a+,3=03(0), a,3=04(0),
and consequently
a'=01(0% Y=02(0% (a +,3)' = ()3(9'), (a,3)'=04(0'),
it follows from the equations
03(0)=01(0)+02(0), -04(9) = 01(9)Y'2(9)
we conclude that LJ
(7) N(p) _ fm1,1m2,2 ... mn,n,
because the determinant
fWr1 W2 (n) _-
n ... Wn A(Wl, W2, ... , Wn)
is nonzero.
It follows that every norm is a rational number and, by virtue of (4)
and (5), so is every discriminant. These two propositions could also have
been deduced from the theory of transformations of symmetric functions,
but I wish to avoid relying on this.
If we replace the p in the equations (6) by p - z, where z is any rational
number, then the coordinates mi,i, are unchanged except for the mi,i on
the diagonal, each of which is replaced by mi,i - z. Theorem 7 is then
changed into the equation
m1,1 - z m2,1 ... mn,l
M1,2 m2,2 - z ... mn,2
...(µ(n) - z),
= (µ - z)(µ " - z)
module b = [,31, 132,.. . , ,an] and which are therefore integers in the field.
We then have n + 1 equations of the form
= cl)3l + C2,32 + ... - COn,
al = C1,101 + c2,1,32 + ... + Cn,l
a2 = c1,201 + C2,2132 + ... + Cn,2Nn,
CO0
....................................
an = C1,01 + C2,02 + ... + en,nQn,
whose n(n + 1) coefficients are rational integers and whose n + 1 partial
n x n determinants, obtained by suppressing one horizontal line, have
no common divisor (§4,6). If we put
But if we have d - 1 (mod 4), z can also equal 2,t and we have
O= = 1, 12 , and D =
1 1
1
2
d 2
= d.
These two cases can be combined into one by noting that o = [1, D 2 D ]
in each. It is also clear that a quadratic field is completely determined
by its discriminant D. This is not so for the next case, namely n = 3,
where invariants additional to the discriminant are necessary for the
complete determination of a cubic field. However, we shall be able to
give a complete explanation of this fact only with the help of the theory
of ideals.
We now return to considering an arbitrary field Il of degree n, and
we add the following remarks on divisibility and congruence of numbers
in the domain o. Let A, p be two such numbers, and suppose that A
is divisible by p. By the general definition of divisibility (§14) we then
have A = pw, where w is an integer, and since the quotient w of the two
numbers A, p belongs to S2, by definition of a field, w will likewise be a
number in the domain o. The system m of all numbers in the field 0
t It follows, for example, in the case d = -3 that the integers of the field are not all
of the form x + y/ where x, y are rational integers.
118 Chapter 3. General properties of algebraic integers
divisible by it therefore consists of all numbers of the form pw, as w runs
through all numbers of the domain o = [W1, W2, ... , wn], that is, through
all numbers of the form
w=h1w1+h2w2+"'+hnw,,,,
whose coordinates hl, h2i ... , ham. are rational integers. Consequently we
have m = [pwl, pw2i... , pwn]. We now say that two integers a, /3 in the
domain o are congruent modulo p, and write
a - /3 (mod p),
when the difference a - /3 is divisible by it and hence in m. Thus this
congruence is completely equivalent to the following:
a - /3 (mod m),
the meaning of which was explained in §2. In the contrary case, a, /3
are called incongruent modulo p. If we understand a class modulo p to
be the set of all numbers in o congruent to a particular number, and
hence congruent to each other, then, in the notation of §2, the number
of these classes will be (o, m). And since the integers pw1i pw2, ... , pwn
forming the basis of m are connected to the numbers w1i w2, ... , wn by
n equations of the form (6), (§17), in which the coefficients mz,i, are
necessarily rational integers, it follows from the equation following (7),
together with Theorem 4 of §4, that the number of these classes is
(o, m) = ±N(p).
The system m is identical with o if and only if p is a unit, in which case
±N(p) = (o, o) = 1.
In this conception of congruence, where an actual number p appears
as divisor or modulus, there is a complete analogy with the theory of
rational numbers. However, it is plain, as we have already indicated
in detail in the Introduction and in Chapter 2, that completely new
phenomena concerned with the decomposition of numbers into factors
arise in this same domain o. These phenomena are brought back under
the rule of simple laws by the theory of ideals, the elements of which will
be covered in the next chapter.
4
Elements of the theory of ideals
In this chapter we shall develop the theory of ideals to the point indicated
in the Introduction, that is, we shall prove the fundamental laws which
apply to all fields of finite degree, and which regulate and explain the
phenomena of divisibility in the domain o of all integers in such a field
Q. The latter are what we refer to when we speak of "numbers" in
what follows, unless the contrary is expressly indicated. The theory is
founded on the notion of ideal, whose origin has been mentioned in the
Introduction, and whose importance has been sufficiently illuminated by
the example in Chapter 2 (§§11 and 12). The exposition of the theory
that follows coincides with the one I have given in the second edition of
Dirichlet's Vorlesungen fiber Zahlentheorie (§163). It differs mainly in
external form; however, if the theory has not been shortened it has at
least been simplified a little. In particular, the principal difficulty to be
surmounted is now thrown more clearly into relief.
119
120 Chapter 4. Elements of the theory of ideals
ideal and denote it by o(p), or more simply by op or po. It is evident
that this ideal will be unchanged when p is replaced by an associate,
that is, a number of the form ep, where a is a unit. If p is itself a unit we
have op = o, since all numbers in o are divisible by p. It is easy to see
that no other ideal can contain a unit. Because if the unit a is in the ideal
a then (by II) all products ew, and hence all numbers w in the principal
ideal o, are in a. But since, by definition, all numbers in the ideal a are
likewise in o, we have a = o. The ideal o plays the same role among the
ideals as the number 1 plays among the rational integers. The notion
of principal ideal op also includes the singular case where p = 0, where
the resulting ideal consists of the single number zero. However, we shall
exclude this case from now on.
In the case n = 1, where our theory becomes the old theory of num-
bers, every ideal is evidently a principal ideal, that is, a module of the
form [m] where m is a rational integer (§§1 and 5). The same is true
for the special quadratic fields considered in Chapter 2 (§6 and the be-
ginning of §7). In all these cases, where every ideal of the field Il is
a principal ideal, numbers are governed by the same laws that govern
the theory of rational integers, because every indecomposable number
also has the character of a prime number (see the Introduction and §7).
This will follow easily from the results below, but I mention it now to
encourage the reader to make continual comparisons with the special
cases, and especially with the old theory of rational numbers, because
without doubt it will help greatly in understanding our general theory.
Since each ideal is a module (by virtue of I), we immediately carry
over to ideals the notion of divisibility of modules (§1). We say that an
ideal m is divisible by an ideal a, or that it is a multiple of a, when all
the numbers in m are also in a. At the same time we say that a is a
divisor of in. According to this definition, each ideal is divisible by the
ideal o. If a is a number in the ideal a then the principal ideal oa will
be divisible by a (by II). For this reason we say that the number a, and
hence every number in a, is divisible by the ideal a.
Likewise we say that an ideal a is divisible by the number 77 when a
is divisible by the principal ideal o77. When all the numbers a in an
ideal a are of the form 77P it is easy to see that the system r of all the
numbers p = a/77 will form an ideal. Conversely, if p runs through all
the numbers in an ideal r while 77 is a fixed nonzero number, then all the
products 77P will again form an ideal, divisible by or7. We shall denote
an ideal formed in this way from an ideal r and number 77 by rq or 77r.
We evidently have (rr7)r7' = r(7777') = (r77')r7, and 77r' will be divisible by
§20. Norms 121
r7r if and only if r' is divisible by r. Thus the equation 77r' = it entails
the equation r' = r. The notion of a principal ideal op is the special case
of rp where r = o.
We finally remark that divisibility of the principal ideal op by the
principal ideal or7 is completely equivalent to divisibility of the number
p by the number i . The laws of divisibility of numbers in o are therefore
included in the laws of divisibility of ideals.
The least common multiple m and the greatest common divisor a of
two ideals a, b are also ideals. Certainly m and a are modules (§1, 3
and 4), and they are divisible by o, since a and b are divisible by o.
Also, if p = a = 3 is a number in m and hence also in a and b, and if
6 = a' +/3' is a number in the module a then the product pw = aw = /3w
will likewise be in m and the product 6w = a'w +,a'w will be in a since
(by virtue of II) the products aw, a'w are in a and the products /3w, /3'w
are in b. Thus m and a enjoy all the properties of ideals. At the same
time it is clear that mr7 will be the least common multiple of the ideals
all, bq, and ar7 their greatest common divisor.
If b is a principal ideal o77, then the least common multiple m of a, b
will always be of the form r7r, where r is another ideal and in fact a divisor
of a, since 77a is a common multiple of a and o77, and hence divisible by
77r. This case occurs frequently in what follows, and for that reason we
say, for brevity, that the ideal r dividing the ideal a corresponds to the
number 77. If r' is the divisor of r corresponding to the number 77', then
r' will also be the divisor of a corresponding to the product q?7'. This
is because 7rr7'r' is the least common multiple of 77r and or777', and hence
also of a and or777', since 77r is the least common multiple of a and o77,
and or777' is divisible by or7.
§20. Norms
Since each ideal a is also a module, we say that two numbers w, w' in
the domain o are congruent or incongruent modulo a according as their
difference w - w' belongs to a or not. We express the congruence of w
and w' modulo a (§2) by the notation
w - W (mod a).
As well as the theorems on congruences established previously for ar-
bitrary modules, we must also note that two congruences modulo the
122 Chapter 4. Elements of the theory of ideals
same ideal a,
W=-W, 'w" -W "' (mod a),
can be multiplied to give the congruence
ww" - w'w'" (mod a),
since the products (w-w')w" and (w"-and hence also their sum
ww" - w'ware numbers in the ideal a. Moreover, if m is a principal
ideal op, then (by virtue of §18) the congruence w - w' (mod m) will be
identical with the congruence w - w' (mod p).
A very important consideration is the number of different classes, mod-
ulo the ideal a, which make up the domain o. If p is a particular nonzero
number in the ideal a then the principal ideal op will be divisible by a,
and since a is divisible by o it follows (§2,4) that
(o, op) _ (o%a)(a, op).
But the number (o, op) = ±N(p) by §18, and hence the domain o con-
tains only a finite number of mutually incongruent numbers modulo the
ideal a (§2,2). This number (o, a) will be called the norm of the ideal a
and we denote it by N(a). The norm of the principal ideal op is equal
to ±N(p), and o is evidently the only ideal with norm 1.
If p runs through a complete system of N(a) incongruent numbers
(mod a) then so does (1 + p), and adding the corresponding congruences
1 + p = p', where p' runs through the same values as p, yields N(a) = 0
(mod a). That is, N(a) is always divisible by a. As a special case, this
result includes the evident theorem that N(p) is divisible by p (see §17).
Now suppose that r is any ideal and 77 is a nonzero number. Always,
(o77, ri7) = (o, r) = N(r),
since two numbers i'w' and 77w" in the principal ideal 77o are congruent
(mod 77r) if and only if the numbers w', w" in o are congruent (mod t).
Let a, b be any two ideals, let m be their least common multiple, and
let a be their greatest common divisor. By §2, 3 and 4, we have
(b, a) m) a)
and, since a is divisible by o,
(o, a) = (o, D) (D, a), (o, m) _ (o, b) (b, m),
hence
and
N(m)N(0) = N(a)N(b).
If we apply these theorems to the case where b is a principal ideal oi',
so that m is of the form rrl, then, since the ideal r is the divisor of a
corresponding to the number i (§19), we get
(b, a) _ (oi', rrl) = N(r),
and consequently
N(a) = N(r)N(D).
The ideal r can also be defined as the system of all roots p of the con-
gruence q7P - 0 (mod a), as is easy to see.
same way as with a, and choose a number 77' in such a way that the
divisor r' of r corresponding to r7' again has norm less than that of r,
while likewise being divisible by p. But since (§19) r' is at the same time
the divisor of a corresponding to the number 77r7' this contradicts the
assumption we have made about r7 and r. Thus r = p, that is, q p is the
least common multiple of a and or7. Q.E.D.
(a) The system o is a finitely generated module [w1, w2, ... , wn] whose
basis is also a basis for the field Q.
(b) The number 1 is in o, hence so are all the rational integers.
(c) Each product of two numbers in o is also in o.
When a domain o enjoys these three properties we shall call it an
order. It follows immediately from (a) and (c) that an order consists
entirely of integers from the field Sl, but it does not necessarily contain
all the integers (except in the case n = 1). Now if a number a in the
order o is called divisible by a second such number p only when a = µw,
with w also in o, and if we modify the notion of congruence of numbers in
o in the same manner, then one sees immediately that the number (o, op)
of mutually incongruent numbers of o modulo p is again ±N(µ) (§18).
It is also easy to see that all the definitions and all the theorems of the
present chapter retain their meaning and truth if we always understand
number to mean a number in the order o. Thus, in particular, each order
o in field 1 has its own theory of ideals, and this theory is the same for
all orders (which are infinite in number) up to the point we have carried
it so far. However, while the theory of ideals in the order o of all integers
in the field 1 leads finally to general laws which coincide completely with
the laws of divisibility for the rational numbers, the theory of ideals in
other orders is subject to certain exceptions, or rather, it requires a
certain limitation of the notion of ideal. The general theory of ideals in
an arbitrary order, whose development is equally indispensable for the
theory of numbers and which, in the case n = 2, coincides with the theory
of orders of binary quadratic forms,t will be left aside in what follows,t
and I shall content myself with giving an example to call attention to
the character of the exceptions just mentioned. In the quadratic field
resulting from a root
+vf ---3
2
)(mod w),
LO C-)r w ( )3
implies that the number
v
µ
in the field S2 satisfies an rth degree equation of the form
7]r = 77s + W
P-W (AV)r,
with r > 1, is an integer, and if s is any one of the r exponents
0, 1, 2, ... , r - 1 then the term
s
_ (v
= WC
µ)
§25. Laws of divisibility 129
and hence
N(a) = N(pi)N(p2) ... N(pm)
Moreover, if we have
gig2...qr,
b =
and consequently
N(b) = N(gl)N(g2) ... N(q,),
N(ab) = N(pi) ... N(pm)N(ql) ... N(qr),
which obviously implies
N(ab) = N(a)N(b).
Q.E.D.
8. An ideal a (or a number a) is divisible by an ideal a (or a number
6) if and only if each power of a prime ideal which divides a (or 6) also
divides a (or a).
Proof. If p is a prime ideal and p7z is a divisor of an ideal a then we
have (by 6) a = alp"°, where al is an ideal. If we suppose the latter
decomposed into all its prime factors, then a will also be expressed as
a product of prime ideals, among which the factor p appears at least m
times. Conversely, if the decomposition of Z into prime factors includes
the prime ideal p at least m times, then a will evidently be divisible
by pm. Thus if we suppose that every power of the prime ideal which
divides a also divides an ideal a, this amounts to saying that all the
prime factors in the decomposition of a appear, at least as often, as
factors in the decomposition of a. The factors of a include first of all
the factors of a and, if we denote the product of the other factors by a',
then we have a = aa', and consequently a is divisible by D. The converse
proposition, that if a is a divisor of a then each power of a prime ideal
that divides a also divides a, is verified easily. Q.E.D.
If we combine all factors of the same prime in the decomposition of
an ideal a then we find
a = pagbrc . .
§26. Congruences
Having established the laws of divisibility for ideals, and hence also for
numbers in o, we shall add some considerations on congruences which are
important for the theory of ideals. For the moment we shall be content
simply to give indications of the proofs.
1. Since o is the greatest common divisor of two relatively prime ideals
a, b, and ab is their least common multiple, then (§2,5) the system of
two congruences
w - p (mod a), w - or (mod b),
where p, or are two given numbers in o, always has roots w, and all these
roots come under the form
w - T (mod ab),
where r represents a class of numbers modulo ab which is determined by
p and o, or by their corresponding classes modulo a and b respectively.
Conversely, each class r (mod ab) is determined in this way by precisely
one combination p (mod a), Q (mod b).
We now say that the number p is prime to the ideal a when op and a are
relatively prime ideals, and we let b(a) denote the number of mutually
incongruent numbers modulo a which are prime to a. One easily derives
the theorem that
1'(ab) = V)(a)V)(b)
the function vi(a) in the case where a is a power pm of the prime ideal
p. The total number of mutually incongruent numbers modulo p7z is, in
the case m > 0, equal to
N(pm) = [N(p))m = (o,
pm) = (o, p) (p, pm) = (p, pm)N(p)
It is necessary to subtract from this the number of numbers not prime
to p"°, and hence divisible by p. Since this number is equal to
[N(p)]m-1,
(p, pm) =
we get
X(d) = N(a)
where the summation is taken over the divisors a' of a. But since we
also have (1)
E O(d) = N(a),
it follows immediately that X(d) is never zero and always 0(a'). Thus
we have proved the following very important theorem:
"If a and a' are any two ideals we can always, by multiplying a by an
ideal b' prime to a', change it into a principal ideal OW = 077. ))
Putting aa' = a, the fact that z'(a') is nonzero means that there is
always a number 77, corresponding to the divisor a' of a, such that a will
be the greatest common divisor of a and o17. If we then put o77 = ab', b'
will be an ideal prime to a'. Q.E.D.
3. Since each product pp' of numbers p, p' prime to an ideal a is
likewise a number prime to a, and since, as p remains constant and p'
varies, pp' runs through a system of /1'(a) mutually incongruent numbers
(mod a), we deduce by the well-known method,t and for each value of
the number p, the congruence
pO(a) = 1 (mod a),
which represents the highest generalisation of a celebrated theorem of
t See Dirichlet's Vorlesungen fiber Zahlentheorie, §19.
§26. Congruences 137
op = p1 p22 ... pe
where p1, p2, ... , Pe are distinct prime ideals with respective degrees
fl, f2, .... fe. We deduce from this the following extremely important
theorem:
"The rational prime p divides the fundamental number 0(1l) of the
field S2 if and only if p is divisible by the square of a prime ideal."
This theorem is still true, although the proof is more difficult, when
the numbers k corresponding to all possible numbers 0 are all divisible
by p. Such a case is actually encountered,t and this is one of the reasons
I was determined to build the theory of ideals, not on congruences of
higher degree, but on entirely new principles which are at the same time
as simple as possible and better suited to the true nature of the subject.
to introduce his ideal numbers in the first place, how the elements of the
theory explained above lead to the goal with great facility.
Let m be a positive rational prime number, and let Il be the field of
degree n resulting, in the manner described above (§15), from a primitive
root of the equation 9' = 1, that is, a root of the equation
f(o)=om-1+em-2+...+92+e+1=0.
Since the coefficients are rational, we have n < m - 1. Moreover, since
0,02'. .., 0--1 are all roots of this equation we have
B,n-1),
.f(t) = tt 1 = (t - 9)(t - 92) ... (t -
where t is a variable, and consequently
m=(1-9)(1-02)...(1
The m - 1 factors of the right-hand side are integers and associates of
each other, because if r is one of the numbers 1, 2, ... , m - 1 then
1-or =1+0+0 +...+er- 1
1-e
will be an integer, and if s is positive and chosen so that rs - 1
(mod m), then
1-9 1-9''e _ e(s-1)r
= - or 1 + er + 92r + +
1 9T 1
we get
m = Eµ.,-1
where ko, k1i k2, ... , km-2 are integers, is not divisible by m, and hence
also not by µ7a-1, unless all the numbers ko, kl,... , km-2 are divisible
by m. Because an w divisible by m must also be divisible by µ, and
this makes ko divisible by µ, and hence also by µ2, and this makes k1
divisible by p, and hence also by m. Continuing in this way, we conclude
that the other numbers k1, k2, ... , km-2 are divisible by m.
With the help of this result it is easy to show that the m - 1 numbers
1, 9, 92, ... , 9m-2 form a basis for the domain o of all integers in the field
Q. Since we have
mom_1 = (9
tm - 1 = (t - 1)f (t), - 1)f'(9),
it follows, by excluding the uninteresting case m = 2, that
mm-2,
N[f'(9)] =
and since N(9) = 1 and N(9 - 1) = m it follows from §17 that
0(l, 9, 02'. .., 9m-2) _ (-1) 21 mm-2_
Moreover, since p = 1 - 9, 0 = 1 - p it is clear that the two modules
[1, 0.... , 9m-2] and [1, µ, ... , µm-2] are identical, whence it follows (§4,3
and §17,(5)) that we also have
.2'..., 7a-2) _ (-1) m- 1
mm_2.
Since the numbers 1, p, p2, ... , µm-2 are independent, each number in
the field Q can be put in the form
ko + kip + k2p2 + ....+ km-2µ"Z-2 w
k k
where k, ko, k1, k2, ... , km-2 denote rational integers without common
divisor. For this number to be an integer, that is, for w to be divisible
by k, it is necessary (§18) that k2 divide the discriminant of the basis
1, µ, µ2, ... , µ7a-2, and therefore k cannot contain prime factors other
§27. Examples borrowed from circle division 141
than the number m. Moreover, since it has been shown that w is not
divisible by m unless the numbers ko, k1,.. . , k ,,n-2 are divisible by m,
k must also not be divisible by m. Thus we must have k = ±1, and all
integers of the field are of the form
w = ko + k1µ + k2µ2 + ... + km-2µm-2,
whence we have
o = [1, L, ... , µm_2] = [1, 0, ... , em-2],
or again, since 1 + 0 + 02 + ... + 9m,-2
+ Om_ 1 = 0,
0 0(Q) = (_1)--r-'Mm-2.
Now let p be any prime ideal different from oµ. The positive rational
prime p divisible by p is different from m, and we have
N(p) =pf,
where f is the degree of the prime ideal p. Two powers 0', 09 are
congruent modulo such a prime ideal only if they are equal, that is,
if r s (mod m). This is because r 0 s (mod m) implies 0' - 03 =
or (1- 9''-S) = ep where e is a unit, in which case or cannot be congruent
to 93 (mod p). Now since we have (§26,3)
9N(p) = 0 (mod p),
it follows that
pf 1 (mod m).
Let a be the divisor of O(m) = m - 1 to which the number p belongs
modulo m, that is, let a be the least positive exponent for which
pa = 1 (mod m).
As we know, f must be divisible by a, and hence f > a. But since all
integers in the field Il have the form
w = F(9) = x19 + X2 02 + + x7,,,_19"°-1
where x1i x2, ... , x,,,, are rational integers, it follows from well-known
theorems, true for every prime number p, that
wP - F(OP), wP' - F(9P'") (mod p),
and hence
wP a
- w (mod p).
142 Chapter 4. Elements of the theory of ideals
We conclude first of all that the ideal op is a product of distinct prime
ideals, because if op = p2q there would be a number w divisible by pq but
wpa
not by p, so w2 and hence also would be divisible by p2g2 = pq, and
then also by p, contrary to the preceding congruence. Moreover, since p
is divisible by p, every integer w in the field Sl satisfies the congruence
wra - w (mod p),
which therefore has N(p) = pf mutually incongruent roots w. And since
its degree is pa we must have pf < pa, thus implying f < a. But it has
already been shown that f > a, so f = a. We have therefore arrived at
the following result, which is the main theorem of Kummer's theory:t
"If p is a prime number different from m and if f is the exponent to
which p belongs modulo m, so that O(m) = e f for some e, then
P1p2...pe,
op =
where pl, p2, ... , pe are distinct prime ideals of degree f ."
The rest follows easily. The general case where m is an arbitrary
composite number can be treated similarly. The degree of the normal
field Sl is always equal to the number O(m) of those numbers among
1, 2,3,..., m that are prime to m. The preceding law is proved without
any change, and the determination of the prime ideals that divide m
does not present any extra difficulty.
From the very general researches that I am going to publish shortly,
the ideals of a normal field Sl immediately allow us to find the ideals
of an arbitrary subfield $ of fl, that is, any field H whose members all
belong to Q. For example, this enables us to know the ideals of any field
H resulting from division of the circle and, to give a more precise idea
of the scope of these researches, I mention the following case.
Again let m be a prime number, so ¢(m) = m - 1, and let e be any
divisor of m -1 = e f . In the theory of rational numbers, the congruence
kf - 1 (mod m)
has precisely f mutually incongruent roots h, which are closed under
multiplication and which, in that sense, form a group. If 0 is again a
primitive root of the equation 0 = 1 and if Sl is the corresponding field
of degree m - 1, then all the numbers F(9) in this field satisfying the
t Kummer's researches may be found in Crelle's Journal, 35, in Liouville's Journal,
XVI and in the memoirs of the Berlin Academy for the year 1856.
t Dedekind calls it a "divisor", as in his footnote in §15. (Translator's note.)
§27. Examples borrowed from circle division 143
9h
m) = (:,:p )'
where ±m = 1 (mod 4), and to the theorem
Cm)
_ (-1) s.
ma-1
and consequently
WP = (x + yi)P - x + (-1) 2 yi (mod p).
Now if p - 1 (mod 4), each integer w will satisfy the congruence
wP - w (mod p),
whence it follows immediately that op is the product of two different
prime ideals of degree 1. But if p - 3 (mod 4) we have
2
wP=w WP -w (modp),
where w' is the number conjugate to w, and we conclude easily that op
is a prime ideal of degree 2. But every ideal a of this field must be a
principal ideal. In fact, if ao is a member of the ideal a with minimum
norm, then each number a in the ideal will be divisible by ao. This is
because (§6) we can choose the integer w so that
N(a - wao) < N(ao),
and since the numbers a, ao, and hence also a-wao, belong to the ideal
a, we must have N(a - wao) = 0, whence a = wao and consequently
a = oao. Q.E.D.
Now, since op is the product of two prime ideals of degree 1 in the
case where p is a rational prime =- 1 (mod 4), it follows that
p = N(ao) = N(a + bi) = a2 + b2,
which is the celebrated theorem of Fermat.
146 Chapter 4. Elements of the theory of ideals
(ab)c = a(bc) that (AB)C = A(BC) for any three classes A, B, C. Then
we can apply the same reasoning as for the multiplication of numbers
or ideals, and show that, in the composition of any number of classes
Al, A2, ... , Am, the order in which pairs of classes are combined has no
influence on the final result, which we can denote simply by A, A2 A,,,.
If the ideals al, a2i ... , am represent the classes A1, A2, ... , Am then the
ideal a, a2 an, represents the class Al A2 Am. If all the m factors
equal A, then their product is called the mth power of A, and we denote
it by A'. In addition, we put Al = A and A° = O. The following two
cases are particularly important:
The equation oa = a yields the theorem that OA = A for any class A.
Moreover, since each ideal a can be transformed into a principal ideal
am by multiplication by an ideal m, for each class A there is a class M
satisfying the condition AM = 0, and only one, because if the class N
is such that AN = 0 it follows that
N = NO = N(AM) = M(AN) = MO = M.
The class M is called the class opposite or inverse to A, and we denote
it by A. Conversely, it is clear that A will be the class inverse to A.
If in addition we define A-' to be the class inverse to A' then we have
the following theorems for any rational integer exponents r, s:
ArAs = Ar+s, (Ar)s = Ars, (AB)r = ATBr.
Finally, it is evident that from AB = AC we can always deduce B = C,
multiplying by A-'.
If we now allow each of the n coordinates hl, h2i ... , hn to take all k + 1
values 0, 1, 2, ... , k, then we obtain distinct numbers w and, since their
number is (k+1)n and hence > N(a), there are necessarily two different
numbers w,
a = blwl + ... + brawn 'Y = C1w1 + ... + Cnwn,
§30. Conclusion
We shall derive some further interesting consequences of the fundamental
theorem proved above. (See Disquisitiones Arithmeticae, art. 305-307.)
Let h be the number of classes of ideals of the field S2, and let A be a
particular class. The h + 1 powers
0, A,A2,...,Ah-i,A'
are all different, and we say that the class A belongs to the exponent m.
Obviously Al-1 = A-1, and more generally we have Ar = A8 if and
only if r - s (mod m). Then, if B denotes any class, the m classes
(B) B, BA, BA 2'. .., BA--1
will be all different, and any two complexes of m classes, such as the
preceding (B) and the following
(C) C, CA, CA2,... , CAm-1,
where a', j3' are integers satisfying the condition aa' +,130' = 6. Q.E.D.
152 Chapter 4. Elements of the theory of ideals
If at least one of the two numbers a, /3 is nonzero, then the number
6, and any of its associates, deserves the name greatest common divisor
of a, 3. If 6 is a unit then a, /3 may be called relatively prime, and
two such numbers enjoy the characteristic property that any number p
divisible by a and /3 is also divisible by a/3. This is because the equations
µ = aa" = /3/3" and 1 = aa' + /3/3' imply
p = a/3(a'/3" +,3'a"),
and the converse is equally valid, since a, /3 are both nonzero.
Index
153
154 Index
of forms, 17, 19, 28, 44, 100, 102 Diophantine equation, 8
of ideal classes, 146 Diophantus, 8
congruence Arithmetica, 8
higher order, 57 identity, 9, 18, 23
modulo a module, 64 Dirichlet, 56, 149
modulo an algebraic integer, 118 class number formula, 42
modulo an ideal, 97, 121 theorem on primes, 42, 149
of algebraic integers, 55 Vorlesungen, 5, 21, 40, 45, 53, 61, 84,
of Gaussian integers, 85 87, 98, 102, 119, 125, 135-137,
of quadratic integers, 87 143, 149
of rational integers, 84 discriminant, 112
roots of, 137 determines quadratic field, 117
conjugate is rational, 113
algebraic numbers, 55 of cyclotomic field, 141
and norm, 112 of field, 116
fields, 110 of Gaussian field, 145
ideals, 97 of quadratic form, 14
in quadratic field, 89, 116 Disquisitiones
number, 41, 110 of Gauss, 7, 33, 40
numbers in quadratic field, 87 divisibility
periods, 143 by ideal number, 58
construction by nth power, 90
straightedge and compass, 31, 32 of algebraic integers, 54, 105
continued fraction, 4 of ideals, 60, 98, 120
continuity, 58 of modules, 63
coordinates of quadratic integers, 87
with respect to a basis, 108 of rational integers, 83
correspondence divisor, 7
of numbers and ideals, 121 behaviour as a, 89
cubic ideal, 60
field, 117 of a module, 63
reciprocity, 144 of an ideal, 133
cyclotomic domain, 5
equation, 31, 32, 139 Dedekind, 5
field, 139
integers, 30, 32, 140 Eisenstein, 40, 144, 149
elementary transformations, 77
decomposable numbers, 55 Elements
Dedekind of Euclid, 6, 7, 53
and Weber theory, 46 equation
avoidance of symmetric functions, 41 cyclotomic, 31, 32, 139
class number theorem, 42 Pell, 4, 8
definition of algebraic integer, 39 hagorean, 29
domain, 5 T+
X y3 = z3, 30
invention of ideals, 3 x4 + y4 = z2, 10
proof of quadratic reciprocity, 37 x4 + y4 = z4, 10
proof of two square theorem, 25 y3=x2+2,9,21
section, 44, 58 equivalence
Supplement to Dirichlet, 20 of ideal factors, 28
theory of ideals, 29, 37 of quadratic forms, 13
degree equivalent
of a field, 54, 108 ideal numbers, 56
of a form, 56 ideals, 146
of a prime ideal, 124 Euclid, 53
Descartes, 10 algorithm for gcd, 12, 84
determinant, 71 Elements, 6, 7, 53
of a quadratic form, 14, 98 formula for Pythagorean triples, 6, 10
Index 155
root
of congruence, 137
of unity, 28, 38
primitive, 33, 137, 139, 142
Schonemann, 137
section, 44, 58
Serret, 137
Stark, 42
subfield, 142
substitution, 108
inverse, 13
unimodular, 14
symmetric functions, 41, 113
Newton theorem, 40
two square theorem, 9
and Gaussian primes, 24
Dedekind proof, 25, 145
Euler proof, 11
Fermat proof, 11
Lagrange proof , 15
unimodular substitution, 14
unique prime factorisation
and equivalence of forms, 28
failure in [v/-], 30
failure in Z[./- ], 5, 27
failure in Z[C23], 32
failure in cyclotomic integers, 56
failure in quadratic field, 87
for ideals, 3, 130
in complex integers, 56
in Disquisitiones, 7
in Gaussian integers, 22, 24, 85
in rational integers, 22, 56, 84
in Z[v/- ], 26
in Z[C3], 30
of ideals, 5, 102
units
in algebraic integers, 54, 106
in Gaussian integers, 23, 85
in quadratic field, 86
in rational integers, 83
vector space, 41
Weber, 45
and Dedekind, 46
Weil, 13
zero module, 63