You are on page 1of 167

THEORY OF ALGEBRAIC INTEGERS

Other books available. in the Cambridge Mathematical Library:

A. Baker Transcendental number theory
H.F. Baker Abelian functions
N. Biggs Algebraic graph theory, 2nd edition
S. Chapman & T.G. Cowling The mathematical theory
of non-uniform gases
G.H. Hardy A course of pure mathematics, 10th edition
G.H. Hardy, J.E. Littlewood Inequalities, 2nd edition
& G. Po1ya
D. Hilbert Theory of algebraic invariants
W.V.D. Hodge & D. Pedoe Methods of algebraic geometry,
volumes I, II & III
R.W.H.T. Hudson Kummer's quartic surface.
A.E. Ingham The distribution of prime numbers
H. Lamb Hydrodynamics
F.S. Macaulay The algebraic theory of modular systems
G.N. Watson A treatise on the theory
of Bessel functions, 2nd edition
E.T. Whittaker A treatise on the analytical dynamics
of particles and rigid bodies
A. Zygmund Trigonometric series

Theory of Algebraic Integers Richard Dedekind Translated and introduced by John Stillwell CAMBRIDGE UNIVERSITY PRESS .

New York. Cambridge CB2 1RP 40 West 20th Street. 1831-1916.D43 1996 512'. Australia First published in French 1877 English translation and introduction © Cambridge University Press 1996 First published in English 1996 Library of Congress cataloging in publication data Dedekind. cm. translated and with an introduction by John Stillwell p.74-dc2O 96-1601 CIP British Library cataloguing in publication data available ISBN 0 521 56518 9 paperback Transferred to digital printing 2004 . 2. Theory of algebraic integers / Richard Dedekind. Integral representations. Trumpington Street. NY 10011-4211. Richard. ISBN 0-521-56518-9 (pbk. Title QA247.) 1. 1. USA 10 Stamford Road. Includes bibliographical references and index. Oakleigh. Algebraic number theory.Published by the Press Syndicate of the University of Cambridge The Pitt Building. Melbourne 3166.

2.2 Divisors and prime factorisation 7 0.2 The grit in the oyster 12 0.3 Gaussian primes 24 0.3.1 Pythagorean triples 6 0.4.3.4 Quadratic reciprocity 36 0.5.4 Lagrange's proof of the two squares theorem 15 0.1 General remarks 3 0.2 Gaussian integers 22 0.6 Composition of forms 17 0.3 Irrational numbers 8 0.3.3.7 The class group 19 0.3 Reduction of forms 13 0.2.2.4.5 Primitive roots and quadratic residues 16 0.3 Cyclotomic integers and quadratic integers 32 0.2 The cyclotomic integers 30 0.5.4.1 Fermat 10 0.5 Roots of unity 29 0.2. Contents Part one: Translator's introduction page 1 T'ranslator's introduction 3 0.2 Squares 6 0.4 Diophantus 8 0.5.3.4.5 The failure of unique prime factorisation 27 0.4 Imaginary quadratic integers 25 0.3.1 Fermat's last theorem 29 0.5.4.1 The need for generalised "integers" 21 0.5.3.5 Other reciprocity laws 38 v .3 Quadratic forms 10 0.4 Quadratic integers 21 0.

Ideals and their divisibility 119 §20. The rational integers 83 §6. Congruences and classes of numbers 64 §3.6. Divisibility and multiplication of ideals in o 98 3 General properties of algebraic integers 103 §13. The integers in a field Il of finite degree 113 4 Elements of the theory of ideals 119 §19. Divisibility of integers 105 §15. Role of the numbers 3 and 7 in the domain o 91 §10. Norms and discriminants 111 §18.2 Basic properties 40 0.6. The difficulty in the theory 126 §24. Finitely generated modules 67 §4. Conjugate fields 108 §17.7. Ideals in the domain o 95 §12. Irreducible systems 71 2 Germ of the theory of ideals 83 §5. Prime ideals 123 §22.3 Class numbers 41 0. The domain o of numbers x + y/ 86 §8.6. The complex integers of Gauss 84 §7. Role of the number 2 in the domain o 89 §9. The domain of all algebraic integers 103 §14.1 Definition 39 0.4 Ideal numbers and ideals 42 0. Laws of divisibility in the domain o 93 §11.1 How the memoir came to be written 44 0. Modules and their divisibility 62 §2.7 The reception of ideal theory 44 0. Auxiliary propositions 128 . Fields of finite degree 106 §16.6 Algebraic integers 39 0.7. vi Contents 0.2 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. Norms 121 §21. Multiplication of ideals 125 §23.6.

Examples borrowed from circle division 138 §28. Congruences 134 §27. Laws of divisibility 129 §26. Classes of ideals 146 §29. Contents vii §25. Conclusion 149 Index 153 . The number of classes of ideals 147 §30.

.

Part one Translator's introduction .

.

Emmy Noether used to say "Es steht schon bei Dedekind" ("It's already in Dedekind"). Dedekind wrote for an audience that knew number theory . before going on to develop a general theory and to prove the theorem on unique factorisation into prime ideals. to- gether with groups. Today this is still worthwhile. and of course most are not fluent in German and French. I have translated the most accessible of Dedekind's works on ideal theory. These concepts. Translator's introduction 0. rings and modules were generalised beyond the realm of numbers by Emmy Noether and Emil Artin.1 General remarks Dedekind's invention of ideals in the 1870s was a major turning point in the development of algebra. His aim was to apply ideals to number the- ory. which became the temporary home of algebra while its core concepts were growing up. the numbers m + n/ where m. In an attempt to overcome these problems. Dedekind (1877). Algebra finally became independent in the 1920s. This memoir shows the need for ideals in a very concrete case. n E Z. which he wrote to explain his ideas to a general mathematical audience. The algebraic integers in Dedekind's title are a generalisation of the ordinary integers . At the same time. rings.created in response to certain limitations of classical number theory. when fields. but not so easy. Today's readers probably have the opposite qualifications. he created algebraic number theory. modules and vector spaces. field or module. Sur la Theorie des Nombres Entiers Algebriques. But even then.especially quadratic forms - but not the concepts of ring. but to do this he had to build the whole framework of commutative algebra: fields. and urged her students to read all of Dedekind's works in ideal theory. were to form the core of the future abstract algebra. The ordinary integers have been studied since ancient 3 .

Sometimes it is necessary to use irrational numbers. A famous example is the so-called Pell equation x2-cy2=1 where c is a nonsquare integer and the solutions x. yo) by the formula xk+ykf -±(xo+yo. 4 Translator's introduction times. ao are ordinary integers.)k. Yet even ancient number theory contains problems not solvable by Euclid's methods. 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. see 0. Solutions for certain values of c were known to the ancients. so theorems about Z may be obtainable as special cases of theorems about R (we shall see . The first fact implies that the algebraic integers form a ring.yEZ} and Z[i]={x+yi:x. . and the rational algebraic integers are just the ordinary integers (for more details."2-. that is.and x. the ordinary integers are also known as rational integers. to answer questions about the ordinary integers. In particular. ao E Z.. difference and product. and their basic theory was laid down in Euclid's Elements (see Heath (1925)) around 300 BC..2 and §13 of Dedekind's mem- oir). we are not interested in the ring of all algebraic integers so much as rings like Z[v]={x+yV':x. which are defined in general to be roots of equations of the form an_ 1. an_ 1 i . Because of the second fact. The reason for working in rings Z[a] is that they more closely resemble Z. In general we use the notation Z[a] to denote the closure of the set Z U {a} under +. The irrational numbers xk + in this formula are examples of alge- braic integers.. y are required to be integers. and hence are more likely to yield information about Z. they are closed under sum.. Any ring R of algebraic integers includes Z.6. . . such as v. Algebraic integers are so called because they share some properties with the ordinary integers. However.yeZ}. .

0.1 General remarks 5

several examples later). However, useful theorems about R are provable
only when R has all the basic properties of Z, in particular, unique
prime factorisation. This is not always the case. Z[-5] is the simplest
example where unique prime factorisation fails, and this is why Dedekind
studies it in detail. His aim is to recapture unique prime factorisation by
extending the concept of integer still further, to certain sets of algebraic
integers he calls ideals. This works only if the size of R is limited in some
way. The ring A of all algebraic integers is "too big" because it includes
f along with each algebraic integer a. This gives the factorisation
a = \I-a-vra- and hence "primes" do not exist in A, let alone unique
prime factorisation.
Dedekind found the appropriate "small" rings R in algebraic number
fields of finite degree, each of which has the form Q(a), where a is an
algebraic integer. Q(a) denotes the closure of Q U {a} under +,-,x
and = (by a nonzero number), and each Q(a) has its own integers,
which factorise into primes. In particular, Z[\] is the ring of integers
of Q(/ ), and 6 = 2 x 3 is a prime factorisation of 6. Not the prime
factorisation, alas, because 6 = (1+/)(1-) is also a factorisation
into primes (see 0.4.5). However, unique prime factorisation is regained
when one passes to the ideals of Z[v/-5], and Dedekind generalises this
to any Q(a). The result is at last a theory of algebraic integers capable
of yielding information about ordinary integers.
A lot of machinery is needed to build this theory, but Dedekind ex-
plains it well. Suffice to say that fields, rings and modules arise very
naturally as sets of numbers closed under the basic operations of arith-
metic. Fields are closed under +, -, x and =, rings are closed under +,
- and x, while modules are closed under + and -. The term "ring" was
actually introduced by Hilbert (1897); Dedekind calls them "domains"
here, and I have thought it appropriate to retain this terminology, since
these particular rings are prototypes of what are now called Dedekind
domains. Dedekind presumably chose the name "module" because a
module M is something for which "congruence modulo M" is meaning-
ful. His name for field, Korper (which also means "body" in German),
was chosen to describe "a system with a certain completeness, fullness
and self-containedness; a naturally unified, organic whole", as he ex-
plained in his final exposition of ideal theory, Dedekind (1894), §160.
What Dedekind does not explain is where Z[/] comes from, and
why it is important in number theory. This is understandable, because
his first version of ideal theory was a supplement to Dirichlet's number
theory lectures, Vorlesungen caber Zahlentheorie (Dirichlet (1871)). In

6 0.2 Squares

the present memoir he also refers to the Vorlesungen frequently, so his
original audience was assumed to have a good background in number
theory, and particularly the theory of quadratic forms. Such a back-
ground is less common today, but is easy and fun to acquire. Even
experts may be surprised to learn how far back the story goes. The
specific role of can be traced back to the anomalous behaviour of
the quadratic form x2+5y2, first noticed by Fermat, and later explained
in different ways by Lagrange, Gauss and Kummer. But the reason for
Fermat's interest in x2 +5 y2 goes back much further, perhaps to the
prehistory of mathematics in ancient Babylon. Let us begin there.

0.2 Squares
0.2.1 Pythagorean triples
Integers a, b, c such that
a2 + b2 = C2

are one of the oldest treasures of mathematics. Such numbers occur as
the sides of right-angled triangles, and they may even have been used to
construct right angles in ancient times. They are called Pythagorean
triples after Pythagoras, but they were actually discovered indepen-
dently in several different cultures. The Babylonians were fascinated
by them as early as 1800 BC, when they recorded fifteen of them on a
tablet now known as Plimpton 322 (see Neugebauer and Sachs (1945)).
Pythagorean triples other than the simplest ones (3,4,5), (5,12,13) or
(8,15,17) are not easily found by trial and error, so the Babylonians
probably knew a general formula such as
a=2uv, b=u2-v2, c=u2+v2
which yields an unlimited supply of Pythagorean triples by substituting
different integers for u, v.
The general solution of a2 + b2 = c2 is in fact
a = 2uvw, b = (u2 - v2)w, c = (u2 + v2)w,
as may be found in Euclid's Elements Book X (lemma after Proposition
29). A key statement in Euclid's proof is: if the product of relatively
prime integers is a square, then the integers themselves are squares.
Euclid first used the general formula for Pythagorean triples in his the-
ory of irrational numbers, and it is in a different book from his theory
of integers. The assumption that relatively prime integers are squares

0.2.2 Divisors and prime factorisation 7

when their product is a square is justified by a long chain of proposi-
tions, stretching over several books of the Elements. However, a direct
justification is possible from his theory of integer divisibility, which is in
Book VII. This theory is fundamental to the theory of ordinary integers,
and also the inspiration for Dedekind's theory of ideals, so we should re-
call its main features before going any further. Among other things, it
identifies the important but elusive role of primes.

0.2.2 Divisors and prime factorisation
An integer m divides an integer n if n = ml for some integer 1. We also
say that m is a divisor of n, or that n is a multiple of m. An integer
p whose only divisors are ±1 and ±p is called a prime, and any integer
can be factorised into a finite number of primes by successively finding
divisors unequal to ±1 but of minimal absolute value. However, it is not
obvious that each factorisation of an integer n involves the same set of
primes. There is conceivably a factorisation of some integer
g1g2...qj
n = p1p2...pi =

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.

2. 1. uses unique prime factorisation to see that each prime appears to an even power in a square.cy2 = 1. Hence it is tempting to try to interpret a right-angled triangle as a Pythagorean triple by choosing the unit of length so that a.2 Squares 0. The simplest counterexample is the triangle with sides 1. 8 0. via increasingly large integer solutions. yn. b and c are integers (Pythagoras' theorem). They found it could used to approach f rationally. where c is a nonsquare integer. Dedekind even used it to prove the irrationality of VFc (Dedekind (1872). 0. Section IV).3 Irrational numbers As everybody knows. They .4 Diophantus The equations a2+b2 = c2 and x2-2y2 = 1 are examples of what we now call Diophantine equations. c satisfy a2 + b2 = c2 whether or not a. f and so on. the quotient xn/yn necessarily tends to v f2-. It is impossible to interpret this triangle as a Pythagorean triple because is not a rational number. b and c all become integer lengths. after Diophantus of Alexandria. In any right-angled triangle. A proof of this fact. can similarly be used to approach the irrational number V /c-. The irrationality of led the Greeks to study the so-called Pell equation x2-2y2=1. Since xn .2yn = 1.. Pythagoras or one of his followers made the historic discovery that this is not always possible. xn. v'2-.2. the side lengths a. Pythagorean triples also have significance as the sides of right-angled triangles. f . which also proves the irrationality of i. and an even number of times in the prime factorisation of m2. b. The general Pell equation x2 . This equa- tion later proved fruitful in many other ways. Then the equation 2n2 = m2 is impossible because the prime 2 occurs an odd number of times in the prime factorisation of 2n2. Diophantus lived sometime between 150 AD and 350 AD and wrote a collection of books on number theory known as the Arithmetica (Heath (1910)).

p. 268. Euler would. p. For Diophantus it is usually a rational solution.1 for the astonishing sequel to this solution. off and on. a sum of squares] in one way only . It appears from this that Diophantus is aware of the identity (a2 + b2)(c2 + d2) = (ac ± bd)2 + (ad bc) 2 though he makes no such general statement. but his com- ments were as cryptic as the Arithmetica itself. which is due to the fact that 65 is the product of 13 and 5. each of which is the sum of two squares. Euler spent about 40 years. there is hardly any method yet invented in this kind of analysis not already traceable to Diophantus.2. the integer solutions are of more interest. the actual methods that he uses for solving any of his problems are as general as those in use today . but he mentioned an integer solution to another remarkable equation: the solution x = 5. 0. It shows how much theory can be latent in a single numerical fact. but for some equations.. until he could reconstruct most their methods and prove their theorems. Fermat saw much deeper than this. namely into 72 +4 2 and 82 + 12.) . Fermat and Euler more thoroughly later. Book III. Diophantus remarks 65 is naturally divided into two squares in two ways.. 429-430. with Diophantus. In the Arithmetica.. but one example is worth mentioning here. The term "Diophantine" refers to the type of solution sought: either rational or integer. that the identity reduces the representations of a number as a sum of two squares to the repre- sentations of its prime factors. such as the Pell equation.4 Diophantus 9 consist entirely of equations and ingenious particular solutions. The first mathematician to understand Diophantus properly was Fermat (1601-1665). 2. his comment on the problem is: A prime number of the form 4n + 1 is the hypotenuse of a right-angled triangle [that is. However. The Pell equation was actually not studied by Diophantus. Problem 19. y = 3 of y3 = x2 + 2. (See 0. If a prime number which is the sum of two squares be multiplied by another prime number which is also the sum of two squares.) Although all Diophantus' solutions are special cases. Noticing. they usually seem chosen to illustrate general methods. (Euler Opera Omnia 1.4. reading between the lines of Fermat and Diophantus. the product will be the sum of two squares in two ways. We shall study the connection between Diophantus. (Heath (1910).) And if anyone would know.. Euler (1756) went so far as to say Nevertheless.

1). u is also a Pythagorean triple. He took up number theory only in his late 30s. Suppose that there are positive integers x. involving just Euclid's theory of divisibility. y. z = u2 + v2.v2. it is a beautiful piece of work.3 Quadratic forms 0. z such that x4+y4 = z2. which we can take to be primitive. z such that x4 + y4 = z4. in the posthumously published Fermat (1670).1 Fermat Unlike Euclid or Diophantus. (x2)2 + (y2)2 = z2 . This says that x2. and it is also primitive. The argument is by contradiction. Fermat's proof shows that there are no positive integers x. It also has a place in our story. y. z is a Pythagorean triple. Gauss and Dedekind all used this theorem of Fermat to test the strength of new methods in number theory. It turns out to be the only instance of Fermat's last the- orem with a really elementary proof.3. and his marginal notes on Diophantus. z such that x4 + y4 = z2. hence there are integers u. However.3 Quadratic forms The restriction to primes of the form 4n + 1 is understandable because a prime p 54 2 cannot be the sum of two squares unless it is of the form 4n+ 1 (by a congruence mod 4 argument). y. . But Fermat's claim that any prime p = 4n + 1 is a sum of two squares comes right out of the blue. and fully estab- lishes his credentials as both an innovator and a student of the ancients. and left only one reasonably complete proof. 10 0. as an application of Pythagorean triples. y = s2 . his letters. u = s2 + t2.2. No one knows how he proved it and the first known proof is due to Euler (1756). and the gist of it is as follows (omit- ting mainly routine checks that certain integers are relatively prime). The middle equation says that v. Fermat never wrote a book. 0. v such that x2 = 2uv. As we shall see later. t such that v = 2st. hence there are integers s. His reputa- tion rests on a short manuscript containing his discovery of coordinate geometry (independent of Descartes).t2. y2. by showing that there are not even positive integers x. or in other words. y. by Euclid's formulas (0. Lagrange. and as the first proven instance of Fermat's last theorem. y2 = u2 .

In 0.4 we shall see an easier proof of the two squares theorem due to Lagrange. 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 . but still nonzero. However. and any integer a 0. and hence 4 x1 +y 4= 2 1 . It follows that each is itself a square. which is a contradiction. which has its uses elsewhere (see for example Gauss's proof of quadratic reciprocity in 0. .3. Lagrange's proof does use another famous theorem of Fermat. This proof implicitly contains the "mod p binomial theorem". we have aTi-1 .1. and finding a smaller prime with the same property.1 Fermat 11 This gives x2 = 2uv = 4st(s2 + t2). for example. and used it for many of his other theorems. Fermat called the method used in this proof infinite descent.a1' + b'' (mod p).2 s2+t2=zi. He claimed. and by retracing the argument we find that the new square z2 is smaller than the old. (a + b) P .i + 1) i . 0. as is clear from the fact that p is a factor of the numerator but not of the denominator.4). The proof more often seen today is based on that of Euler (1761). Thus we have found another sum of two fourth powers equal to a square. but it is the easy one known as Fermat's "little" theorem: for any prime number p. .1 (mod p) (Fermat (1640b)). it is very hard to see how to make the descent in this case. 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. t and s2 + t2 have product equal to the square (x/2)2.1)(p . By repeating the process we can therefore obtain an infinite descending sequence of positive integers.0 (mod p).5.3. so the relatively prime integers s. say s=xi2 t=yi. Euler (1749) found a proof only after several years of effort.zl. z2. i! for -- 1 < i < p .2) (p .

1 (mod p).11. 2.1 (mod p)..3.. . . .. .. 0. p .4): Theorem 1.3 Quadratic forms which implicitly uses the group properties of multiplication mod p.1)} is the same set (mod p) as the set { 1. 2. a x (p . p . p = x2 + y2 p = 1 (mod 4) (Fermat (1640c)) Theorem 2. If two primes which end in 3 or 7 and surpass by 3 a multiple of 4 are multiplied. . in which case 1 = ar + ps for some integers r. . (Fermat (1654).. and could only conjecture the following less satisfying fact. and he was probably right. It follows that mod p multiplication by a nonzero a is invertible. p = x2 + 2y2 p 1 or 3 (mod 8) (Fermat (1654)) Theorem 3. p) = 1. x2 +2 y2 and x2 +3 y2 for integer values of x and y (the first prompted by Diophantus' remark on sums of squares. Hence each set has the same product mod p.. and in particular the set {a x 1. as proofs eventually published by Euler were based on Fermat's method of infinite descent.. . since 1. the next theorem should be about x2 +5 y2 .1 from both sides (which is permissible.. a x 2. p = x2 +3 y2 p 1 (mod 3) (Fermat (1654)) Fermat thought he could prove these theorems. Since x2 +4 y2 = x2 + (2y)2. p . 2. s by the Euclidean algorithm. 12 0. as mentioned in 0. which is "of the form" x2 + y2. ax2ax.x(p-1) (modp). par- ticularly the idea of multiplicative inverses. then their product will be composed of a square and the quintuple of another square.) Since numbers that end in 3 or 7 are of the form 10n + 3 or lOn + 7. .2. and cancellation of 1. This is the grit in the oyster. Fermat was unable to prove a theorem about primes of the form x2 +5 y2 . An integer a is nonzero mod p if gcd(a. the anomalous behaviour of the quadratic form x2 +5 y2 . and .2 The grit in the oyster The mathematical pearl that is Dedekind's theory of ideals grew in re- sponse to a tiny irritant.. Between 1640 and 1654 Fermat discovered three beautiful the- orems about the representation of odd primes p by the forms x2 + y2..1 have inverses) gives Fermat's little theorem: aP-1 . The number r is called a multi- plicative inverse of a (mod p) since ar ..x(p-1)a-lx2x..

3. Euler (1744) found another clue to the puzzle. c E Z).3 Reduction of forms The first to account rigorously for the two-faced behaviour of x2+5y2 was Lagrange (1773). What about primes of the form x2 + 5y2 ? There are some. This is very puzzling. 0. The substitution x'=ax+/3y. As Weil (1974) says: When there is something that is really puzzling and cannot be understood. 61=42+5x32.and he conjectured that the following holds for all prime values p. He noticed two "faces" to the behaviour of x2 + 5y2 . 41=62+5x12. Fermat's conjecture can be restated as follows: Fermat's conjecture. This situation begs for an explanation. p = x2 +5y2 p m 1 or 9 (mod 20) 2p = x2 +5 y2 p m 3 or 7 (mod 20). it usually deserves the closest attention because some time or other some big theory will emerge from it. they lie in the classes 20k + 1 and 20k + 9 that seem conspicuously absent from the conjecture above..3. sometimes 2p . y' =ryx+Sy maps Z x Z into Z x Z provided a. he had the very fruitful idea of finding those forms a'x'2+b'x'y'+c'y'2 equivalent to ax2 + bxy + cy2 via a change of variables. Euler's conjecture. such as 29=32+5x22. If two primes are of the form 20k + 3 or 20k + 7 then their product is of the form x2 + 5y2. 0.3 Reduction of forms 13 such a number is also of the form 4m + 3 if and only if it is of the form 20k + 3 or 20k + 7. . y=7x +6'y'. 6 E Z and it is one-to-one provided there is an inverse substitution x = a'x' + /3'y'.sometimes it represents a prime p. b. 0. Studying the general question of which integers n could be represented by a given quadratic form axe+bxy+cy2 (where a.y. and furthermore.

3 Quadratic forms In this case S)[ry' [y s-]-[0 1 since the product of a substitution and its inverse is the identity. S E Z and determinant aS . in the case of negative discriminant. For any particular D < 0 it is then possible to work out the inequivalent reduced forms of discriminant D. as can be checked by computing bj2 . I have retained this term in the translation of Dedekind's memoir because he refers to a slightly different definition. taking determinants of both sides. The number of them is called the class number h(D). It follows that -D=4a'c'-bj2>4aj2-a'2=3a'2 and therefore. it follows that the invertible substitutions for which the pairs (x'. 14 0.0y = ±1. the old term for the discriminant of a quadratic form was determinant. Such substitutions are now called unimodular. y. a a' /3' =1. since the determinants on the left are integers and the only integer divisors of 1 are ±1. There- fore. y S y' S' Finally. The first few calculations yield the following results. only finitely many values of the integers a'. due to Gauss. 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'. y') run through Z x Z are precisely those with a. b' and hence c'.) Observing the invariance of the discriminant is a first step towards decid- ing equivalence of forms. can occur in reduced forms. (Incidentally. Lagrange observed that equivalent forms have the same discriminant D=b2-4ac=b'2-4a'c'. 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. 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. . because the unimodular substitutions form a group. 0. as we would say.4a'c' and using aS .Qy = f1.

provided z is chosen so that p does not divide z2n .1.4 Lagrange's proof of the two squares theorem Given a prime p = 4n + 1.2n = 2n values of z which make z2n . hence h(-4) = 1. All forms with discriminant -8 are equivalent to x2 +2 y2 . p then divides z4n .3. Before following up this suggestion. the forms x2 +5 y2 and 2x2 + 2xy + 3y2. we want to choose one of the 4n nonzero values of z (mod p) so that z2n . A modern proof of this theorem (not much different from Lagrange's own proof) uses the fact that the nonzero integers mod p have multi- plicative inverses. Aha! There is something different about x2 +5 y2f Perhaps the previously invisible companion 2x2+2xy+3y2 accounts for its "two-faced" behaviour. which is m2 + 1 with m = zn. and hence p divides z2n + 1 = m2 + 1 as required. article 182). Hence there cannot be more roots than the degree of p(z). 0. hence h(-8) = 1. since Lagrange has shown we can transform this form into x2 + y2. In terms of congruences. By Fermat's little theorem. let us see how Lagrange used the uniqueness of the form x2 + y2 to prove the two squares theorem. Thus p will divide z2n + 1. however. and one can show as in classical algebra that each root zi of a polynomial p(z) corresponds to a factor (z .1 = (z2n . This is where it is crucial that p be of the form 4n + 1 and prime. its discriminant is -4. and of course it takes the value p for x = 1. 0. It follows that the congruence classes mod p form a field. (The use of this particular form is a simplification of Lagrange's argument due to Gauss (1801).zi) of p(z). . y = 0. But there are two inequivalent reduced forms with discriminant -20.1 nonzero mod p. hence h(-12) = 1.1 is nonzero (mod p). It suffices in turn to find an integer m such that p divides m2 + 1.1)(z2n + 1) for any integer z relatively prime to p. All forms with discriminant -12 are equivalent to x2 +3 y2 . This is possible by an earlier theorem of Lagrange (1770): a congruence of degree q modulo a prime p has at most q solutions.3. so h(-20) = 2. it suffices to find any form with discriminant -4 that represents p.4 Lagrange's proof of the two squares theorem 15 All forms with discriminant -4 are equivalent to x2 + y2. because in this case the form px2 + 2mxy + M2+1 y2 has integer coefficients. It follows in the present case that we have at least 4n .

2. Like the existence of primitive roots (also conjectured by Euler). p . F o r example.. The first proof was given by Gauss (1801).1 are squares. and it also throws more light on the two squares theorem. since it will be important later. but was unable to prove it. which means a is a primitive root mod p. called "quadratic reciprocity".3. The quickest proof looks at elements of relatively prime order. was conjec- tured by Euler. a4.. the primitive root is not constructed explicitly . article 55. The existence of a primitive root a mod p means that {l.. An integer a is called a primitive root mod p if aP-1 1 (mod p).1. We digress a little further here to discuss this concept... 2.1..1 this is a congruence of degree 1 with more than 1 solutions. More generally.. Euler conjectured that a primitive root exists for each p.. The traditional term for squares mod p is quadratic residues mod p. .the existence of primitive roots.. of a primitive root are squares mod p. a2. and first shows that the least common multiple 1 of the orders of nonzero integers mod p is itself the order of some element. if we define the order of element a mod p to be the least positive n such that an = 1 (mod p). 16 0.a. a. exactly h a l f the integers 1. it was first proved by Gauss. Existence of a primitive root means that the group of nonzero congruence classes is cyclic.5 Primitive roots and quadratic residues Lagrange's result that an nth degree congruence mod p has at most n dif- ferent solutions has another important consequence in mod p arithmetic . . the even powers 1. contrary to Lagrange's result.. its existence is shown by counting the number of possible solutions of a congruence..2.a 2.aP-2} _ {1. .p-1. Indeed. Since 1 is the least common multiple we then have x1=-1 (modp) for x=1.1} This makes many facts about multiplication mod p quite transparent.. and conversely. and their fundamental theorem.. We shall say more about this .. If 1 < p .3 Quadratic forms 0.. Like all proofs since. Quadratic residues arise naturally in the study of quadratic forms.. a' 1 (mod p) for l < q < p . Hence in fact 1 = p .p. mod p.1.rather. then a primitive root is an element of order p .

for example 3=2x02 +2x0x1+3x12. accounts for much of the anomalous behaviour of x2 +5 y2 . Quadratic character of -1. 23 = 2 x (-2)2 +2 x (-2) x 3+3 x 3 2. m is square mod p = m=a 2i for some j m221 = ai(P-1) = 1 (mod p). 0. m is a square mod p e* m 2 1 (mod p) If a is a primitive root mod p. that p = 4n + 1 divides some m2 + 1.3. 0.6 Composition of forms The second reduced form of discriminant -20. namely 2x2 + 2xy + 3y2.6 Composition of forms 17 in Section 0. Two important preliminaries to the general discussion of squares mod p are the following: Euler's criterion. The primes not represented by x2 +5 y2 . m is nonsquare mod p #> m = a2j+1 for some j mp2 = ai(P-1)+p21 . are represented by 2x2 + 2xy + 3y2.ap21 -1 (mod p).2) about products and doubles of primes in the classes 20n+3 and 20n + 7.2) that all primes p in the classes 20n + 1 and 20n + 9 are of the form x2 +5 y2. namely those in the classes 20n + 3 and 20n + 7. with the help of another observation: the product of two . Lagrange (1773) proved this. He used these two theorems to establish the conjectures of Fermat and Euler (0. P = 4n + 1 for some n.3.3.-1 (mod p) for some x x has order 4 (mod p) e* x = a4 where a is a primitive root e#0.5. It also follows easily from the existence of primitive roots: -1 is a square mod p x2 . Notice that this statement strengthens the result used in the proof of the two squares theorem. 7=2x12+2x1x1+3x12. and also Euler's conjecture (0.3. The number -1 is a square mod p p = 4n + 1 for some n.

Such an identity can be checked mechanically by multiplying out both sides. We have already seen one known to Diophantus (0. Lagrange's own identity can be derived quite mechanically from (x2 + 5y2)(xj2 + 5yj2) = (xx' .yy')2 + It has a generalisation (x2 + cy2)(x 2 + cy'2) = (xx' .cyy')2 + c(xy' + yx )2 discovered by the Indian mathematician Brahmagupta around 600 AD. 363. 18 0. in the form 2 [X2 + 5Y2}.2yy'. 14. Use the latter to rewrite each of 2x2 + 2xy + 3y2 and 2x'2 + 2x'y' + 3yj2.5yy')2 + 5(xy' + yx')2 (Brahmagupta's identity for c = 5). (See Colebrooke (1817). and Weil (1984). p. This crucial observation is based on the following algebraic identity: (2x2 + 2xy + 3y2)(2x'2 + 2x'y' + 3yj2) = X2 + 5Y2 where X = 2xx' + xy' + yx' . 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 . multiply them out using Brahmagupta's identity.4): (xY1 + yx)2 (x2 + y2)(x 2 + y'2) = (xx . but how did Lagrange find it in the first place? Probably by past experience with products of forms.2. and 2x2+2xy+3y2 = 2 [(x+ 2)2+5(2)21 1 (the result of completing the square on 2x2 + 2xy + 3y2).) This would have been easy for Lagrange to derive using the complex factors (x2+cy2)(x2+cyj2) = and pairing the first factor with the third. p. some of which were known much earlier. and the second with the fourth. then absorb the factors of 2. Y=xy'+yx'+yy'.3 Quadratic forms numbers of the form 2x2 + 2xy + 3y2 is a number of the form x2 +5 y2.

that any two quadratic forms with the same discriminant could be "composed" in this fashion. rewritten by Cox (1989)) p. C = 5x2 + 6xy + 10y2. E = 6x2 + 2xy + 7 y2.yx' . BB = A by Lagrange's identity. in fact. B = 2x2 + 2xy + 21y2.3.3. 0. 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 table would be A B A A B B B A because AA = A by Brahmagupta's identity.7 The class group 19 where X = xx' . if we take the forms with discriminant -20.3.6. B = 2x2 + 2xy + 3y2. and AB = BA = B by the last identity of 0. D = 3x2 + 2xy + 14y2. Legendre (1798) managed to show. All he could was draw up "multiplication tables" for the forms with particular discriminants. The product operation is known as composition of forms. For example. Legendre had found a set (the forms with fixed negative discriminant) and a "product" operation on it (composition) but there was no reason to expect the operation to have simple or interesting structural properties. A more complicated example actually given by Legendre was for the forms of discriminant -164. . A = x2 +5y2 .3yy'. Something very interesting was going on. such as groups.7 The class group Composition of forms came on the scene decades before the axiomatic properties of abstract structures. 42 as: A = x2 + 41y2. but what? 0. were considered in mathematics.

With Gauss's definition of composition. Yet all this was accomplished with definitions and proofs so cumber- some it took 70 years for the rest of the world to understand them. The proof is monstrous. Ambiguity would be avoided in any modern attempt to define an operation on a set. 20 0. Gauss even came close to finding a decomposition of the class group into cyclic factors. the form 13' from F' and f'. namely f"(ff') = (f"f)f'. article 240. and it was avoided in the next study of quadratic forms. Gauss's analysis of forms under composition is amazingly modern in some ways. The two- valued entries are due to an ambiguity of sign in Legendre's definition of composition. It requires the derivation of 37 equations. Even the statement of associativ- ity in the Disquisitiones is clumsy. that the equivalence classes of forms constitute an abelian group under com- position. as if Gauss had not really grasped what associativity is about: If the form F is composed of the forms the form a from F and f". showed that it is well-defined on equivalence classes of forms and showed. He defined composition unambiguously.. by Gauss (1801) (the Disquisitiones Arithmeticae). A polished exposition was eventually given by Dedekind in his Supplement X to . it is rather disturbing to see AC = BC when A # B. for example.) It is only by using the commutativity of composition (which is mercifully obvious) that this statement can be rewritten in a form recognisable as associativity. and worse still that some of the products have two values. It is now called the class group.3 Quadratic forms Legendre's multiplication table for these forms can then be written: AB C D E AAB C D E BBA E CD C C C AorB DorE DorE D D E DorE AorC BorC E E D DorE BorC AorC To us. it is a major problem to prove that composition is associative. of course. (Gauss (1801). in effect. a' will be . the form F' from f. then the forms %. f". equivalent.. This put the subject of composition of forms out of the reach of most mathematicians until Dirichlet and Dedekind simplified it enough to make associativity obvious. most of which Gauss leaves to the reader.

they are cubes themselves.4. the approach via algebraic identities. Kronecker (1870) showed that the decomposition of the class group into cyclic factors follows purely from axioms for finite abelian groups (which he was the first to state). For example. since facts could be deduced from it without reference to the definition of composition. and more usable. since their product is a cube. Dedekind's Supplement X to Dirichlet (1871) was in effect the swansong of the old theory of compo- sition of forms. He assumes they have gcd 1. 0.cyy')2 + c(xy' + yx)2./)(x + s). the theory of algebraic integers. However. The abstract structure of the class group was seen as more important. b E Z = a3 + 3a2b/ + 3ab2(-2) + b3(-2) . This is equivalent to saying that y = 3.4 Quadratic integers 0. The most interesting and provocative example is the Euler (1770) proof of the following claim of Fermat (1657): 27 is the only cube that exceeds a square by 2. Euler's solution is breathtaking. and proceeds to treat x . Lagrange (1768) and Euler (1770) began using complex or irrational factorisations to find integer solutions of equations.1 The need for generalised "integers" 21 the 2nd edition of Dirichlet's Vorlesungen caber Zahlentheorie (Dirichlet (1871)). such as (x2 + cy2)(x 2 + cy'2) = (xx . was becoming irrelevant by this time. and x . He fac- torises the right hand side into (x . 0. even if not exactly rigorous. This presumably means that x+ = (a + bV"--2)3 for some a. and therefore. x = 5 is the only positive integer solution of the equation y3=x2+2.4.1 The need for generalised "integers" We have already seen how convenient it is to use factorisations involving square roots to prove identities about integers. no matter how slick. because in the same work he showed how to rebuild the class group on a simpler and more general basis. The first important examples of such integers were studied by Lagrange and Euler. as if they were integers.

2b3 equating real and imaginary parts.2. which in turn depends on the prime divisor property: a prime divides a product ab only if it divides one of a. The norm of x + yi . y E Z. We can do this immediately with the help of the norm. Why is this permissible. is defined by N(x+yi) _ Ix+yi12 =x2+y2.4. 22 0. The behaviour of the ordinary integers is ruled by unique prime fac- torisation. To justify similar propositions about "quadratic integers" such as x + .2 Gaussian integers Making the usual abbreviation i for. Q. a measure of size in Z[i] introduced by Gauss. This is easiest when the quadratic integers in question have a "Euclidean algorithm".b2) only if b = ±1 and a = ±1. These numbers are now known as the Gaussian integers. where x. and they tie up nicely with quadratic forms and some other threads in our story.4 Quadratic integers and therefore x=a3-6ab2. This gives x = 5 as the only positive integer solution for x. The first step towards unique prime factorisation in Z[i] is to show that primes exist. this is crucial in proving that relatively prime integers are squares when their product is a square (0. 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).2). b (0. But 1 = 3a2b . we have to decide which of them are "primes" and then see whether they have a prime divisor property like the ordinary integers. .D! Euler gave several examples of this kind. since 1 and -1 are the only integer divisors of 1. particularly as divisors. They are the simplest kind of quadratic integers. if indeed it is? To answer this question we need to recall how ordinary integers behave. 1 = 3a2b . we denote the set of Gaussian integers x + yi by Z[i].20 = b(3a2 . so it is worth looking at them first. For example..1).2. generally splitting quadratics into irrational complex factors and treating the factors as integers. 0. who studied the divisibility properties of the numbers x + y .E. N(x + yi).

Oj+i = remainder when aj is divided by (3 until a zero remainder /3k is obtained. (1 +i)/3. We call any such divisor a Gaussian prime because it is not the prod- uct of Gaussian integers of smaller norm. This is similar to the situation in 7L. To find gcd(a.3. so 0 < lpl < 1. The remainder p is the difference between a and the nearest corner p/3 in the lattice. Euclid's proof of the prime divisor property for Z (0. The division property is clear as soon as one realises that the multiples p. -i. where prime factors are determined only up to the unit factors ±1.4.yy')2 + (xy +yx )2 = (x2 +y2)(xj2 +Y..31. The proof for Z[i] is similar.2) involves his algorithm for finding the gcd by repeated subtraction.2.2. .2). by Diophantus' identity (0. By repeatedly removing prime divisors. the so-called units 1.31 of a side. A typical square is the one with corners 0.3 of 3 lie at the corners of a lattice of squares in the complex plane.Q # 0 there are Gaussian integers p and p ("quotient" and "remainder") such that a=µ. which terminates because it yields a decreasing sequence of positive integers. for indices from . as in 7L. The decrease is guaranteed by the following division property of Z[i]: for any Gaussian integers a and ..4). is to establish the prime divisor property: if a prime divides a product then it divides one of the factors. which terminates because it yields a sequence of remainders that decrease in norm. 0.2 Gaussian integers 23 and has the multiplicative property N((x + yi)(x' + y'i)) = N(x + yi)N(x' + y'i). The real problem in 7L[i]. (We use j. we obtain a factorisation of any Gaussian integer into Gaussian primes.3 = N1 and repeatedly compute aj+1 = . . because (xx' . -1. each Gaussian integer of norm > 1 has a divisor x + yi of minimal norm > 1. i.d+p where 0 < 1pI < 1. since the norm is a positive integer.. i. k.It follows that any divisor x + yi of X + Yi has N(x + yi) < N(X + Yi) and therefore. 0) one can therefore let a = a1.aj. Obviously these factors are determined only up to factors of norm 1.31 because the distance IpI between any point in a square and the nearest corner is less than the length 1.3. but the algorithm for gcd requires repeated division with remainder.

yi. so each such x2 + y2 gives us two Gaussian primes. by unique prime factorisation in Z[i].yi yields a factorisation of x + yi by conjugation. and hence unique up to unit factors. if x2 + y2 is an ordinary prime then x + yi is a Gaussian prime.2. Moreover. in turn until only units remain. that the partition x2 + y2 of p into two squares is unique. Os are two prime factorisations of the same number. 7r2.2). because it cannot be the product of numbers of smaller norm. We get the prime divisor property by arguing that 1 = gcd(7r. a result also stated by Fermat. since i will be reserved for . and Gauss observed that it is closely connected with Fermat's two squares theorem.yi) is unique up to unit factors.yi. Hence there are exactly two Gaussian primes. 0. and cancelling 7r1... . being the square of the Gaussian prime 1 + i. 24 0. Conversely. Fermat's two squares theorem tells us that the primes of the form x2 + y2 are exactly the primes p = 4n + 1 and p = 2. up to unit factors. /3) = pa + v/3 by an argument like that used for Z (0.) Then gcd(a. x + yi and x . Thus x2 + y2 = (x + yi)(x .. and there are Gaussian integers p and v such that gcd(a. .yi) is a Gaussian prime factorisation. 7rr = 0102 .. for each or- dinary prime p = 4n + 1.3 Gaussian primes Identifying the actual primes of Z[i] is a separate question. so divisors of x + yi have norm dividing x2 + y2. a) = µ7r + va when it is a Gaussian prime not dividing a. Unique prime factorisation (up to unit factors) is obtained by supposing 7172 .. the rest of the route to unique prime factorisation is essentially the same as in Z.) The ordinary prime 2 is exceptional. .4. In fact. This is because N(x + yi) = x2 + y2. the factorisation x2 + y2 = (x + yi)(x . and multiplying both sides by /3. up to a unit factor. In particular. /3) = 3k. if x + yi is a Gaussian prime then so is its conjugate x . (This shows. incidentally.4 Quadratic integers now on. since any factorisation of x .

since these have no divisors of the form x2 + y2 (by Fermat's theorem again) and hence no complex Gaussian prime divisors x + yi (since a divisor x + yi implies a divisor x . with norm defined by N(x + yam) = Ix + yI2 = x2 + cy2. for any positive integer c.4 Imaginary quadratic integers 25 It follows that x2 + y2 is an ordinary prime. y E 7L}. By Fermat's two squares theorem these are the primes p = 4n+1 and p = 2. which we have already described in 0. The latter proof is short enough to describe here. 0.4. Thus the "properly complex" Gaussian primes are (up to unit factors) the conjugate factors x + yi and x . which gives p=x2+y2. p is not a Gaussian prime.4. since it builds on the result of Lagrange that p = 4n + 1 divides a number of the form m2 + 1. Since p does not divide either of the Gaussian factors m + i. by the prime divisor property of Z[i]. goes on to deduce the equivalence of these forms directly from unique prime factorisation in Z[i]. one in the memoir below (§27) and another in his final version of ideal theory (Dedekind (1894). m .4 Imaginary quadratic integers As an obvious generalisation of the Gaussian integers we can consider. §159. 7L[] = {x + yv/-c : x. The close relation between Gaussian primes and the ordinary primes of the form x2 + y2 suggests that the theory of 7G[i] may be used to prove Fermat's two squares theorem. . The real Gaussian primes are the ordinary primes p = 4n+3. 0.yi.i of m2 + 1 (the quotients p ± I are not Gaussian integers). In fact. Dedekind (1894).3.yi). Dedekind gave at least two such proofs. We therefore have a Gaussian factorisation p = (x + yi)(x . by conjugation). since any factorisation of it into ordinary primes would yield a different factorisation into Gaussian primes.4.yi of ordinary primes of the form x2+y2. This argument from unique prime factorisation replaces Lagrange's ar- gument from the equivalence of quadratic forms with discriminant -4. §159).

and prime factorisation is unique up to sign (as in Z. any solution x of y3 = x2 + 2 must be odd.yV---2 of Z[V/-2].'--2./----2 and x . because an even x yields .2) that the (ordinary) primes of the form x2 + 2y2 are those of the form p = 8n + 1 or p = 8n + 3. like that in Z[i].31 and '1.Q. x-/) = 1. and hence unique prime factorisation. /3. A typical rectangle is the one with corners 0. Each ordinary prime p = 8n + 1 = x2 + 2y2 or p = 8n + 3 = x2 +2y2 splits into primes x + y. 26 0.4 Quadratic integers The multiplicative property of the norm. is based on the division property: for any a and /3 # 0 there are p and p in Z[vr-2] ("quotient" and "remainder") such that a=p/3+p where 0<JpJ <1/31. N((x + y\c)(x' + y'V)) = N(x + y )N(x' + y' ).(3. Since N(x+yv) = x2 +2 y2.31. so the maximum distance of any point from a corner (attained by the centre point) is a IQI < 1131. The lengths of its sides are 1. The actual primes in Z[/] can be described using Fermat's theorem (0. Also. However. note that any common divisor of x + and x .x-v)=1 and that relatively prime numbers in Z[\] are cubes when their prod- uct is a cube.. to justify Euler's solution of y3 = x2 + 2 it remains to prove that gcd(x+v. Finally. (1 + / ). as in Z[i].6) and one proves. Unique prime factorisation also holds in Z[y]. is also a divisor of their difference 2 . the uniqueness of the prime factorisation depends on the value of c. It follows that Z[\/-2] has a Euclidean algorithm. the only units in Z['] are ±1. To see why gcd(x + V/'--2. After the Gaussian integers (c = 1).. by the argument used for Z[i]. The division property holds in Z[\] for similar geometric reasons.3. as required for the division property. since Z[v/-2] contains the numbers x± used by Euler in his solution of the equation y3 = x2 + 2. The proof. and hence "more unique" than in Z[i]). The multiples p/3 of 0 lie at the corners of a lattice of rectangles in the complex plane. that each element of Z[V/-c] has a factorisation into primes. the case c = 2 is also of interest. as we shall see in the next section.3. and each ordinary prime p = 8n + 5 or p = 8n + 7 remains prime in Z[i]. is equivalent to the Brahmagupta identity (0.

This in turn is like the proof for squares mentioned in 0. 3.\ and 2 f . (1 + V1----5)1 (1 . The nonuniqueness of prime factorisation in Z[V/-5] can be traced to the existence of ordinary integers x2 + 5y2 with divisors not of the same form. then we should have the following factorisation of its norm in Z: x2 + 5y2 = (x2 +5 Y2)(X2 + 5y2).1.4. 0. . the proof that relatively prime numbers are cubes when their product is a cube is essentially the same as in Z.5 The failure of unique prime factorisation It should not be a great surprise that something goes wrong in 7G[].(xr + 5Y.y/ belong to a set Z[/] of "integers" without unique prime factorisation. and they certainly differ by more than unit factors. N(3) = 9. But this implies that 1 is the gcd of x . The norms of the factors are N(2) = 4. Consider the following two factorisations of 6: 2x3=(1+V)(1-/). hence the factors 2. which are not norms. If each number x + y/ had a unique prime factorisation in Z[v/-5]. These factors are therefore primes in Z[v/-5].V/'--5) = 6. and hence its gcd with N(2 v/ ) = 8 is 1..2). x+y/ = (x1+y1V-5)(x2+y2V).2. since the norms of the factors on the left are different from the norms of the factors on the right. 0. And each factor (x + 5y?) would be an ordinary prime or the square of an ordinary prime. since any further decomposition of x2 + 5y2 into . Thus N(x + / ) must be odd. in which case 8 divides y3 and does not divide x2 + 2.4. Its ir- rational factors x + y and x .(xr+Yr\). The only nontrivial divisors of these norms are 2 and 3. because the norm of x+y is the anomalous quadratic form x2+5y2. ) themselves have no nontrivial divisors.. prime factorisation is not unique. So here we have another anomalous behaviour of x2 + 5y2. N(1 + V/--5) = 6 and N(1 .. Can this be related to the fact that x2 + 5y2 is one of two inequivalent quadratic forms with discriminant -20? Indeed yes. and the fact that the units of Z[v-2] are ±1. Thanks to unique prime factorisation.. and hence of x+ and x .5 The failure of unique prime factorisation 27 an even y. In short.

4 Quadratic integers ordinary primes would yield a different prime factorisation of x + y in Z[/]. (Excerpt from Kummer (1846a) in Edwards (1977). which will be discussed in the next section. which are not of the form x2 + 5y2. Thus inequivalence of forms is actually the same thing as failure of unique prime factorisation. Then. under which ideal factor classes correspond to classes of forms with the same discriminant. if not for the conservatism of Gauss.) Having invested so much energy in composition of forms. On the other hand. this work also belongs to the theory of roots of unity. As we shall now see. As we know. He later told Dirichlet If I wanted to proceed with the use of imaginaries in the way that earlier mathematicians have done. 28 0. he was unable to find a rigorous description of ideal factors. A second major influence was the theory of roots of unity. Instead. from the theory of forms to the theory of ideal factors. However. It is possible that number theory could have changed direction much earlier. However. Gauss was aware of the failure of unique prime factorisation for quadratic integers when he wrote the Disquisitiones.the theory of regular polygons (or circle division) and the search for reciprocity laws . According to Kummer (1846a).was crucial in the development of the more general concept of algebraic integer. but the failure of unique prime factorisation in Z[v/-5] was not explicitly pointed out until Kummer (1844) proposed the introduction of "ideal factors" to save it. Gauss perhaps became unwilling to pursue the alternative theory of quadratic integers. . since the product of numbers of the form (x + 5y?) is another number of the same form (by Brahmagupta's identity). Kummer (1846b) went on to define a notion of equivalence for ideal factors. then one of my earlier researches which is very difficult [composition of forms] could have been done in a very simple way. But of course 6 has divisors 2 and 3. each divisor of x2+5y2 would also have this form. and could see that something like ideal factorisation was needed. it must be admitted that Gauss's work on other topics . it would be an oversimplification to say that this happened just because of quadratic forms like x2 + 5y2 . namely 2x2 + 2xy + 3y2. Kummer's discovery changed the whole direction of number theory at this point. Most of this could have been noticed by Lagrange. they are of the other form with discriminant -20. 335. and invented the theory of composition of forms as a substitute. p. it was only in 1832 (the paper on Gaussian integers) that he got as far as proving unique prime factorisation for Z[i].

or a biquadrate into two biquadrates.5. as by then it was the only claim of Fermat's still to be proved or disproved. Fermat wrote his famous marginal note: It is impossible to separate a cube into two cubes. Diophantus threw new light on the Pythagorean equation by considering rational solutions. The algebraic approach. or in general any power higher than second into powers of like degree. while falling short of Fermat's last theorem. Although these two claims are logically equivalent. which stems from Fermat and Euler. 0. b. is "geometric" in a broad sense.5 Roots of unity 0.3. having been the inspiration for important results of Euclid (0. I have discovered a truly marvellous proof of this which however this margin is too small to contain.4) and Fermat (0. Fermat's own contribution to Fermat's last theorem was slight. and it is the only case of Fermat's last theorem that has been proved by elementary methods. analysis and topology as well. . is "al- gebraic". they suggest different viewpoints and different approaches to proving the the theorem.1). and by showing that any nonzero square in Q splits into two nonzero squares. The geometric approach seems to have won the day.5. This claim of Fermat's became known as Fermat's last theorem in the early 19th century. but only by calling on large amounts of algebra. as he did not repeat the claim later. it was the main stimulus for the development of Dedekind's theory of ideals. Next to this result in Diophantus' Arithmetica. Since any rational solution becomes an integer solution when multiplied through by a common denominator. an equivalent claim is that an + bn = cn has no nonzero integer solution.3.2. Diophantus (0. In particular. This proof does not generalise to other values of n. and left only the proof for n = 4 (0.1).2. in- stead of just integer solutions.1 Fermat's last theorem The Pythagorean equation is probably the most fruitful equation in num- ber theory. which stems from Diophantus. The approach through rational numbers. The "marvellous proof" he claimed in his marginal note was probably based on a mistake. The approach through integers. c E Q when n is an integer > 2.1). as far as we know. Fermat's last theorem is the claim that the equation an + bn = Cn has no nonzero solution a. was an outstanding success in the development of algebra as a whole.1 Fermat's last theorem 29 0.

x.S3x. (n = 1 are equally spaced . 0. However.. It also contains a treatment of x3 + y3 = z3. and pregnant with possibilities for generalisation.2 The cyclotomic integers The numbers Z[(3] in the proof of Fermat's last theorem for n = 3 are examples of cyclotomic integers. hence cubes themselves. Euler rewrites the equation as y3 = z3 .5. but uses this argument in the wrong setting .in Z[v/-3] rather than Z[(3]. The reasoning is not quite so sound. this book contains a brilliant. treatment of the equation y3 = x2 + 2. the Fermat equation for n = 3. The points (n.1. . 30 0. z .x3 and then factorises the right hand side into (z-x)(z. It so happens that unique prime factorisation fails in Z[-3]. .] where Sn = cos . whose factors all have minimal norm. for geometric reasons. As we have seen in 0. z . At this point Euler makes an inter- esting mistake. It was first studied by Gauss (1801).5 Roots of unity The search for an algebraic proof of Fermat's last theorem really begins with the Algebra of Euler (1770). and Euler's idea can be made to work. unique prime factorisation is valid in Z[(3].. and basically sound. as can be seen from the example 4=2x2=(1+')(1-"). He wants to argue that the factors z .. (n2.(3x)(z-(3x) where -1+v/ 53--1 (3 = 2 2_ are the imaginary cube roots of 1. 4.4.- 2a n + i sin 21r n is a complex nth root of 1.. for geometric rea- sons like those that apply to Z[i] and Z[/]. but it is extremely thought-provoking. The general ring of cyclotomic integers is Z[(. for nonunit elements of 7G[.sax are relatively prime factors of a cube.

When n is a prime. and the primes among them are called Fermat primes. Gauss found . which is possible when Cn is expressible in terms of rational operations and square roots. Gauss was initially interested in Z[Cn] because of its bearing on the construction of the regular n-gon by straightedge and compass.. When n is not prime the cyclotomic equation is of degree lower than n . but the latter three yield regular n-gons not previously constructed.5. but it opened a new one in the history of number theory. zn-1 + + z + 1 = 0 is called the cyclotomic equation. 0. article 341.. the theory of cyclotomic integers created new links between different parts of number theory.what is true is that no two numbers 22h + 1 have a common prime factor. Since Cn . This was observed by Polya and Szego (1924). and is not so easy to describe. 5. Yet in a way Fermat was not far wrong . .+Cn+1=0.. and since then no more such primes have been found..2 The cyclotomic integers 31 around the unit circle.. an-2 E Z}. Perhaps the most surprising feature of this discovery is that primes 22h + 1 had already been considered .for an entirely different reason . 65537. 17. In particular. In fact. from the Greek for "circle-dividing". Euler found that 641 divides 225 + 1. with integer coefficients.by Fermat.. This is false.1. and used by them to give a new proof that there are infinitely many primes. proved that it is the equation of minimal degree. satisfied by Cn.1)(Cn -1 + .. The only known primes of this form are 3.. hence the name "cyclotomic". and was an important factor in his decision to become a mathematician. and hence Z[Cn] can be described explicitly as Z[Cn] = {ao + a1Cn + . Gauss's construction of the regular 17-gon was the first such construction since ancient times. Gauss's discovery may have closed a chapter in the history of geom- etry. Fermat mistakenly believed the numbers 22h + 1 to be prime for all values of h (Fermat (1640a)). a1 i . 257. + an-2Cn-2 ao.1 = ((n . Gauss (1801). He made the amazing discovery that this occurs for precisely the n that are products of a power of 2 with distinct primes of the form 22h + 1. The numbers 22h + 1 are now called Fermat numbers. + (n + 1) and (n :A1 it follows that Cn satisfies the equation ( n-1+. Quite apart from drawing attention to the problem of finding all Fermat primes (intractable so far).

32 0. In fact. Gauss solves the equation by stepwise evaluation of sums of progressively smaller subsets of the roots. Thus in 2h steps one reaches a single root of the cyclotomic equation. As we shall explain in the next sec- tion.3 Cyclotomic integers and quadratic integers Since a straightedge creates lines (which have equations of degree 1) and the compass creates circles (which have equations of degree 2).x) ..+z+1=0 by rational operations and square roots.. Unbeknownst to Lame. Lame (1847) used this formula in an attempt to prove Fermat's last theorem for arbitrary n > 2. because of the factorisation z" . Unfortunately. <P-1.(. In the case p . and this omission turned out to be fatal. there is necessarily a solution of the cyclotomic equation zp-1+. he failed to check uniqueness of prime factorisation in Z[C ]. ..x)(z .. Unique prime factorisation is crucial. . hence they solve linear and quadratic equations. hence the equation says that the sum of all the roots is -1.. the connection between cyclotomic integers and quadratic integers came to light because a solution of the cyclotomic equation by square roots is needed to construct the regular p-gon. and it fails in Z[(281.5 Roots of unity connections between arithmetic modulo p. and had taken steps to deal with the problem. x). this is getting ahead of our story. Let us return to Gauss.. The roots are gyp. For example a root C of the cyclotomic equation for p = 5 satisfies (4+(3+(2+C+1=0. Kummer (1844) had al- ready discovered this.. Cyclotomic integers also arise naturally in connection with Fermat's last theorem. and its sum is obtainable from the preceding Gauss sum by square roots. (z .x" _ (z . which generalises the factorisation used by Euler in his attempt to prove Fermat's last theorem for n = 3. expressible by square roots alone. However. . now known as Gauss sums. (P. He as- sumed. construc- tions find intersections of lines and circles. Thus if the regular p-gon is constructible by straightedge and compass.1 = 22h each subset is half the size of the previous one. as would follow from unique prime factorisation. the quadratic integers 7G[ p1 and the cyclotomic integers Z[(]. 0. that relatively prime factors of an nth power are themselves nth powers. and this can be done by rational operations and square roots.5..

Thus we have a quadratic equation for (3 + (2. Gauss showed that the sums whose exponents are the squares and nonsquares mod p are the two roots of the equation 2p41 =0. Ca29+1 = P when p . . Since the roots are -2 ± 4 in the former case and -2 ±i$ in the latter. which can be solved by square roots.1 (mod 4) and the + sign when p = 3 (mod 4). if p .1 =0. since ((3 + (2)2 = (4 + 2(5 + ( and (5 = 1. in the nontrivial case where p is odd. these sets are of equal size. S4 + (.5. (a2i (a2j+1 ±Vf- when p . The set of all roots (k is partitioned into those for which k is a square mod p. 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. it follows that. and those for which k is a nonsquare.3. whose exponents are the squares mod 5. We could also have used its complement. In the case p = 5. we know that the squares mod p are the even powers of a primitive root mod p.sign when p . From now on we shall use (. In the Disquisitiones.3 (mod 5). the first halving is always possible. Of course. In fact. i (azi . if a is a primitive root mod p. since we know their sum and product.3 (mod 4). article 356.5). and the corresponding Gauss sum is the one used above: (3 + (2. 2 is a primitive root. and we can then find 3 and (2 individually by solving quadratic equations. so its odd powers are 2 and 23 = 8 .1 is not a power of 2 one cannot repeatedly halve the number of terms in the Gauss sums until only one term remains. X +xf with the . As we know from the existence of primitive roots (0.3 Cyclotomic integers and quadratic integers 33 which can rewritten (S3+(2)2+(S3+(2) . 0. However.1 (mod 4). to denote an arbitrary root of the cyclotomic equation. without subscript. and the nonsquares are its odd powers.

Since P-1-1 S (l k-1 \ k) (k.l-1 P-1 E kl ck+l k.34 0.1=1 p by the multiplicative property of quadratic characters. Even more concisely. We may therefore replace k by kl. .k. In terms of quadratic characters.1=1 p P-1 (k) S1(k+1) 7.1). the Gauss sum on the left of the equations is simply S = EP-1 (p) (k. The proof that S2 = ( i) p goes as follows.5 that -1 is a square mod p if and only if p = 1 (mod 4).3.5 Roots of unity Quadratic characters have the following multiplicative property. hence as k runs through the nonzero congruence classes mod p. and Gauss's theorem is equivalent to: S+2 _ f +p if p = 1 (mod 4) -p if p 3 (mod 4). S2 \p /p since we know from 0. pl/ ' which follows immediately from the fact that the squares mod p are precisely the even powers of a primitive root. Now each k 0. so does kl. and P-1 S2 = kl2 ((k1+1 k.3. (k \p/ \p/ . ? \p/ ()Ck+t 1 P- S2 = k.0 (mod p) has a multiplicative inverse mod p (0. for fixed 1 # 0 (mod p).t=1 p since k is a square mod p t* kl2 is a square mod p .

by 0. (3 3 2.does not belong to any Z[(P]. and hence Z[('P] contains either or v. However. Automorphisms of Q((p) extend functions of the form r2((P) = (P.3 Cyclotomic integers and quadratic integers 35 P-1 p .3. V2. Z[(P] and Z[V/I]. and hence any two of them commute. they do show that Galois theory is just over the horizon.1) + k=1 p/ by the cyclotomic equation = (tJ)p p since half the (P) are +1 and half are -1. ( 3 2}. For example. One way to see this is to extend Z[(] to the field Q((p). P. any field containing r2 and (3 has noncommuting automorphisms. cubic integers. We shall not elaborate on these hints. extending all permutations of the set { 3 2. which Dedekind wishes to avoid. Remark. and to consider automorphisms of fields. The relationship of cyclotomic integers with quadratic inte- gers is special. say. . and there is no comparable relationship with.1 -1 (mod p). On the other hand.5 -p p1 since p .1 runs through the nonzero congruence classes _1 -2-2 (kl (p (-1) = \ p / (p . It follows that S = f (Pl )p. Thus Gauss's computation forges links between squares mod 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 . In the next section we shall how Gauss exploited these links to give a proof of quadratic reciprocity.5. 0. by allowing arbitrary quotients. since the ideas belong to Galois theory.

..4 Quadratic reciprocity The quadratic reciprocity law is the following symmetric relationship between the quadratic characters of two odd primes p.. (p-1 as indeterminates because the mini- mality of the cyclotomic polynomial (0.36 0.3.5. As mentioned in 0..5 Roots of unity 0..)n = E (k)c-kq since _ (P for any odd power n (p)2 (kq since (p)2 = 1 _ _ ()q) k=1 P-' p (kq by the multiplicative property (p) S . q. .+xk)qxi+xz+... . (2..+xq (mod q) by induction on k. (p) (q) _ ( -1 if p. . This leads to the "mod q multinomial theorem" (x1+x2+.q=3 (mod 4) q p 5l +1 otherwise. this holds for q prime because in that case all the binomial coefficients (i).. Xk are indeterminates. when x1i x2. .xq + yq (mod q).2) means that each element of Z[C] is uniquely expressible as a sum E'-1 ak(k. Here is the qth power calculation: P(pk) q Sq (k P PE-1 (k q (kq (mod q) k=1 p) (. (q? 1) are divisible by q.1. .5.. We are able to treat the powers (. 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" (x + y)' ..

1 (mod 4). because of the way it relates squares mod p to quadratic integers and cyclotomic integers. . as in the modern theory of functions. for fixed q.5. \p/ \ P The left hand side is (S2)2 _ =11 s2 p 2 P J and hence we can cancel (P1) p from both sides to obtain Q= C l p/ a p2 \p) (mod q). and his own proof (§27 of the memoir) shows how Gauss's ideas can be transformed with the help of ideals.3 (mod 4). We now multiply both sides by S and use SZ = (Pl) S. known from 0. but the one using Gauss sums is especially enlightening. and indeed to construct the theory in such a way that it is able to predict the results of calculation.5. It was certainly an inspiration to Dedekind. The law of quadratic reciprocity has probably been proved in more ways than any other theorem in mathematics except the theorem of Pythagoras. Dedekind says he is giving the sixth proof of Gauss. of which the above is the sixth (Gauss (1818)). Remarks. to seek proofs based immediately on fundamental characteristics. 0. Gauss himself gave eight proofs. rather than on calculation. As he says in §12: It is preferable. Euler's criterion (0. if q .3. from which the statement of quadratic reciprocity follows. the quadratic character of -1. Finally.4 Quadratic reciprocity 37 since. and since (Pl) = ±1 we have if q . but he succeeds in eliminating all the computations.5) allows us to replace p-21 by the quadratic character (q) . This is exactly what he intends to accomplish with the theory of ideals. kq runs through the nonzero congruence classes. obtaining Sq+i S (mod q).3. using the value of GO . There are shorter or more elementary proofs.

In fact. the obvious values of the "biquadratic character" (k)4 are 1. which in the present state of number theory stands as the main problem and pinnacle of the discipline. the obvious values are k 1 if k=a3n if k = a3n+1 q 3 C3 if k=a3n+2 Similarly.5. he said this when he was still flushed with success over reci- procity . (Kummer (1850b). -1 and -i. In the long run not everyone agreed with him. a3n+1 or a3n+a (where a is a primitive root mod p). one wants to define a multiplicative "cubic character" (k)3.5 Roots of unity 0. he gave pride of place to Kummer's results on Fermat's last . (Kummer (1847b). which I have communicated to the Academy of Sciences over the past three years. When Hilbert surveyed algebraic number theory in his influential Zahlbericht (Hilbert (1897)). Thus there is a more pressing reason to use roots of unity when one seeks reciprocity laws for 3rd and 4th powers. for the reasons just men- tioned.5 Other reciprocity laws The appearance of roots of unity in the proof of quadratic reciprocity is very interesting and historically important. then real values of the character will not work. when he had made progress on reciprocity laws: Through my investigations in the theory of complex numbers [cyclotomic in- tegers] and their applications to the proof of the Fermat theorem.) However.) And a few years later. Likewise. One reason it is easy to avoid complex numbers is that the real numbers + 1 and -1 can serve as values of the quadratic character (k). with three different values according as k is of the form a3n. q If. but it is not strictly necessary. Eisenstein (1844) found a cubic reciprocity law through investigation of Z[(3].and when his work on Fermat's last theorem was running out of steam. 38 0. There are many other proofs involving just arithmetic mod p and q. The latter was in fact what Gauss was doing when he developed the theory of unique prime factorisation in Z[i]. I have succeeded in discovering the general reciprocity law for arbitrary power residues. on the other hand. Kummer even declared that cyclotomic integers were more important for their application to higher reciprocity laws than their application to Fermat's last theorem. i. In 1847 he said: The Fermat theorem is indeed more of a curiosity than a main point in the science [of number theory].

'a-. its ideas are at the core of algebraic number theory. a2.5. but it also has one undesirable property: if a is an algebraic integer then so is V/a-. Thus we would be happier if all members of Z[(3] were classed as aintegers.. which can be most simply defined as the closure .1 Definition 39 theorem. . t E Q}.3]. This is very different from the behaviour of ordinary integers. As already mentioned in 0.1. A definition which draws the line at just the right place is the one given by Dedekind in the introduction to his memoir. but it is not clear how it should be defined. unique prime factorisation fails in Z[vr-3]. Dedekind escaped this situation by restricting attention to the integers in a field of finite degree.6. An example which underlines the subtlety of the concept is 7L[ ] = {m + of : m. when it satisfies an equation of the form above in which all the coefficients al. and the theory of all algebraic integers is therefore not very useful in the study of Z.6. or simply an integer.. but it is valid in Z[t.1 Definition The examples of quadratic and cyclotomic integers hint at a general concept of algebraic integer. where b3 = -1 . Even today. we do not want to admit all the members of Q[/] = Is + t/ : s.. a2. He states it in a way that makes it a natural specialisation of the concept of algebraic number: A number 0 is called an algebraic number if it satisfies an equation on +a10n-1 +a20n-2 +. It is called an algebraic integer.1.. when Kummer's approach to Fermat's last theo- rem has been abandoned.6 Algebraic integers 0.. for the reasons given in 0. an. of course. 0.. and not just the members of Z[. +an-10+an = 0 with finite degree n and rational coefficients a1. it is not even clear how quadratic integers should be defined in general. an_ 1. This definition has some desirable properties. . an are rational integers. since all rational numbers would be included. as we shall see in the next section. n E Z}. This would certainly be going too far. an-1. 0. In fact.. At the same time. hence every algebraic integer has a nontrivial factorisation a = V/'a-v. .

to the second edition of Dirichlet's Vorlesungen 2iber Zahlen- theorie (Dedekind (1871)).. Kummer. first pointed out by Eisenstein (1850). unless a divides b as an ordinary integer. an) is any polynomial in the roots with ordinary integer coefficients. then g(al.n Obviously h(x) is monic.6 Algebraic integers Q(a) of Q U {a} under rational operations.. .. and if g(al. Taking ... ac(n))) = 0.].. .. with symmetric coefficients. where a is an algebraic number... . This is obviously crucial when results about ordinary integers are being derived as special cases of results about algebraic integers. an) is also a root of a monic polynomial equation h(x) = 0 with ordinary integer coefficients. . all permutations o of 1. working without a general definition of algebraic integer. hence it actu- ally has integer coefficients.g(aa(1). In article 11 of the Disquisitiones he assumes the reader knows that a rational solution of a monic polynomial equation with integer coefficients is itself an integer. The integers of Q(a) are simply the algebraic integers in Q(a).2 Basic properties Dedekind first stated his definition of algebraic integer in his Supplement X. If f (x) = 0 is a monic polynomial equation with ordinary integer co- efficients and roots a1.. Eisenstein's proof uses the theorem of Newton (1665) that symmetric polynomials in the roots are polynomial functions of the coefficients. §160. When a is an algebraic integer they include all members of Z[a] but sometimes more. a2.. There he added that: It follows immediately that a rational number is an algebraic integer if and only if it is a rational integer. actually defined his integers to be the members of Z[C.. . so in a sense he was lucky to hit on the "correct" integers. These follow from another property of monic polynomial equations. . applying it to the equation h(x) = H (x . It follows that an ordinary integer a does not divide an ordinary integer b. For example. and its roots include g(a1..2. . This result goes back at least as far as Gauss.) is indeed Z[(p]. the set of integers of Q((..and x. an. On the other hand. . the set of integers in Q(/) is not Z[\] but Z[(31. an). as an algebraic integer.. 40 0.6. 0. Similarly crucial properties of algebraic integers are closure under +.

along with the concepts of basis and independence he also introduced. and hence they are algebraic integers. a2.3. a..3 Class numbers 41 g(al. His reduction process for quadratic forms gives an easy way to find the class number of quadratic forms with negative discriminant.6.. 0. It turns out that N(w) is always a rational number. The reason is simple: if the minimal rational polynomial equation satisfied by a is of degree n. Dedekind remarks in §17 of the memoir that he wishes to avoid the theory of symmetric functions. In fact in §13 (and previously in §160 of his Supplement X to Dirichlet (1871)) he has already derived clo- sure from the fact that a system of homogeneous linear equations with nonzero solution must have nonzero determinant. and N(w) is a rational integer when w is an integer of Q(a). ala2 all satisfy monic equations with ordinary integer coefficients. were important in the development of linear algebra as a self-contained disci- pline. Again generalising the situation for quadratic and cyclotomic integers. a2. al.3. al . If a. ¢(a2). because these fields are vector spaces of finite dimension over the rationals.a2 or ala2 shows that al +a2. because the reduced forms are inequivalent.3 Class numbers As already mentioned in 0. O(an_l)... .cy2 can be redefined and reinterpreted as the class number of the field Q(/) (and generalised to any field of finite degree. This means that every integer has a divisor of minimal norm. The class number of x2 . These proofs. an-l is a basis for Q(a) over Q. apart from some simple exceptions. and hence a factorisation into "primes" (though the factorisation is not necessarily unique). and O(a) is the rational function of a equal to w. . then N(w) is the product of O(a). . . . a2) = al +a2. 0. As Dedekind explains in §16 and §17 of the memoir. Setting the theory of algebraic integers within the theory of fields of finite degree gave this development an extra push. The latter fact makes it easy to see the multiplicative property of the norm: N(wlw2) = N(wi)N(w2). Lagrange (1773) gave the first rigorous results bearing on unique prime factorisation of algebraic integers. . the conjugates of w are also its images under all the isomorphisms of Q(a) onto other number fields. Fields of finite degree are important to the concept of algebraic inte- ger because they admit a suitable concept of norm. the so-called conjugates of w..a2.6.. Dedekind defines the norm N(w) of any number w E Q(a). . then 1. al . an_l are the roots of the irreducible monic equation satisfied by a. d(al).

13. (See Baker (1966) and Stark (1967). and Kummer (1850a) gave his own version. Gauss made some progress. There is a similar reduction process. to all intents and purposes. article 303). also due to Lagrange. Apparently.-19. Thus unique prime factorisation holds in these fields. Fermat saw trouble coming .-163 (Disquisitiones.) An equivalent statement is that these are the only quadratic fields with negative discriminant and unique prime factorisation. and the first failure is in Q((23). Dirichlet's result does indicate the depth of the problem. The proof is similar to the one given in the memoir below. While analytic methods are beyond our scope. in his first account of ideal theory. presumably deferring to Kummer.-11. He gives the following specific values for which the class number is 1: p = 5.-2. can now be summarised as follows. a conjecture which remained open until 1966. The finiteness of the class number for all fields of finite degree was first proved by Dedekind (1871).17. Lagrange's computations were extended by Gauss (1801). 42 0.E Ideal numbers and ideals The long story of unique prime factorisation.)..-43. The situation is more difficult for quadratic fields with positive dis- criminant.19. Thus we do not immediately know the class number. when he showed that a prime divides a product only if it divides one of the factors. He also conjectured that these are the only negative c for which Q(/) has class number 1. Dirichlet (1839) made even greater progress.-67. lost and regained. Kummer (1847a) mentions this. 0.6. while giving a direct proof that the class number of Q(Cp) is finite. 7. giving an algorithm in article 195 of the Disquisitiones to decide equiv- alence of reduced forms. and his related theorem on primes in arithmetic progressions (Dirichlet (1837)) has likewise not been proved in a more elementary manner. who showed that the class number of Q(om) is 1 for c = -1. using analysis to discover a formula for the class number of quadratic fields.-7.-3. Euclid discovered unique prime factorisation in Z. when it was proved by Baker and Stark. which shows there are only finitely many reduced forms. No more elementary approach has yet been found. which Kummer shows to have class number 3.11. only that it is finite.6 Algebraic integers see §29 of the monograph). Dirichlet also found an analytic formula for the class number of the cyclotomic field Q((. Dirichlet never published his class number formula for Q((p). but it is no longer clear which of them are equivalent.

. which were finally rewarded by a truly great and fruitful discovery. the integers of cyclotomic fields]. the complex numbers we have been considering seem imperfect. Lagrange found what the trouble was (class number greater than 1). 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. the more one has to admire the steadfast efforts of Kummer. does not also belong to the complex numbers [i. to be decomposable into prime factors. the rational integers]. the whole theory. which is still laboring under such difficulties. 0. Gauss's way out was a retreat from the direction already opened by Lagrange . was Kummer. Dedekind was stymied by this difficulty for a long time. However.. The memoir is not only a superb exposition.4 Ideal numbers and ideals 43 in the behaviour of x2 + 5y2. and here he explains how he overcame it after several years of struggle.) Dedekind goes on to say: But the more hopeless one feels about the prospects of later research on such numerical domains. The first to embrace the algebraic integers. for all their faults.) The first sentence is the one quoted by Dedekind in Latin in his in- troduction: "Maxime dolendum videtur .e. with blow by blow comments up to the final victory.the use of algebraic integers in the study of Z. For this reason. (Translation by Weil (1975) from Kummer (1844). but unfortunately it is not clear that if ideal b divides ideal a then a = be for some ideal c. and his own distillation of the concept of ideal from them. could easily be brought to its conclusion. but also a rare opportunity to see a great mathematician wrestling with a problem. The rest of the story can be left to Dedekind himself." (Kummer's paper was one of the last important mathematical works written in Latin. He explains Kummer's ideal numbers with admirable clarity. The concept of ideal is very simple. but he hoped to regain it: It is greatly to be lamented that this virtue of the real numbers [i. always the same ones for a given number.6. He also gives a candid account of the difficulties en route to unique prime factorisation of prime ideals. He could see that unique prime factorisation was lost. and so are the concepts of divisor and product. and Gauss found an elaborate way round it (composition of forms).e.. This was the discovery of ideal numbers. were this the case.

and since Herr Hoiiel has always been very attentive and obliging. a translation of which is] Our bulletin would receive with great pleasure a detailed analysis of M. leaving the language up to me. made a careful French translation. and that is the purpose of my letter. You have perhaps seen that Darboux' Bulletin has brought out a series of my works on the theory of homogeneous functions of an arbitrary number of differentials. In my opinion they are of rare value. Nevertheless.7. The second editor of the Bulletin. The memoir on algebraic integers first appeared in instalments in the Bulletin des sciences mathematiques et astronomiques in 1877. from §159 onward. let alone appreciate its power or elegance. To make the purpose of these lines clear. 44 0. it is necessary to mention a few things first. Herr Darboux invited me to write such a series in 1872. Herr Hoiiel. who also translated Dirichlet's German works for Liouville's Journal. and again in 1872 with his definition of real numbers as Dedekind sections. together with a very impressive application of the theory: a correspondence between the quadratic forms of discriminant D and the ideals of Q(om). namely concerning your own investigations in the second edition of Dirichlet's lectures on number theory. under which the product of ideals corresponds to com- position of forms. As I recently had cause to write to him. Dedekind (1871). it would be highly desirable if you could arrange to give a detailed analysis of these investigations in the Bulletin. . Therefore. and in Germany itself they have not received their full due. Dedekind had tried this in 1857 when he introduced congruence classes in place of specific residues. which I can best convey to you in his own words.1 How the memoir came to be written Dedekind's first exposition of ideal theory. 11 March 1876 Dear Colleague. He writes: [French text follows. but Dedekind's contemporaries were slow to appreciate his achievement. Leopold Kronecker). Dedekind's work. I have remained in touch with him since. But he was fighting 2000 years of tradition (plus a formidable modern opponent.7 The reception of ideal theory 0. Replacement of the complicated and mysterious forms by objects that behaved like integers should have been a revelation. included proofs of unique prime ideal factorisation and finiteness of the class number. There was great resistance to the idea of treating infinite sets as math- ematical objects. Most mathematicians were not even willing to consider a theory based on infinite sets. Herr Hoiiel has now sent a reply.7 The reception of ideal theory 0. I brought up an idea I have long cherished. The "horror of infinity" that had haunted mathematics since Zeno was not to be dispelled overnight. The reason Dedekind chose this unusual outlet was a letter he received from Rudolf Lipschitz in 1876: Bonn. Dedekind tried again.

with whom I have worked closely as editor of the forthcoming collected works of Riemann. With cordiality and deep respect. the first part of which is: Dear Colleague. as the details were worked out and Hoiiel was engaged to trans- late Dedekind's manuscript into French. working out a more detailed presentation of the theory of ideals. Your letter brought me great and unexpected joy. who will himself write to M. Weber in Konigsberg. and who has expressed his intention to acquaint himself with this theory. pp. Dedekind. and who will in any case be happy to publish his work. He . and I have come so far as to obtain a somewhat improved form of the essential foundation (the content of §163). since for years I had more or less given up hope of interesting anybody in my general theory of ideals. Darboux. namely. and expanding its interaction with Galois theory. After their publication in the Bulletin. Lipschitz (Translation of Lipschitz (1986).) Correspondence about the series of articles continued until 12 August 1876. but also in such a practical way. you are the first. but little by little I have become convinced that the presentation itself is to blame for the failure of this plan. Yours. I can only suppose that the presentation deterred readers through excessive brevity and terseness. If you have occasion to write to the learned professor. and thank him in advance for anything he is disposed to send us. pp. I cannot know whether you are inclined to undertake such a work. But I hope you will understand the true motive for the steps I have taken.) On April 29 Dedekind wrote a long reply. you may tell him how flattered we will be by his collaboration. I thought that the inclusion of this investigation in Dirichlet's Zahlentheorie would be the best way to attract a wider circle of mathematicians to the field.] Of course.2 Later development of ideal theory 45 I have communicated the substance of your letter to M.7. 0. not merely to show interest in the subject. that it revives hope that my work may not have been in vain. the articles were also published as a book (Dedekind (1877)). and since autumn I have been spending my free time. 48-49. Lipschitz (1986).7. R. 47-48. revising it substantially for the third (1879) and fourth (1894) editions of Dirich- let's Zahlentheorie. 0. (Translation of Dedekind's letter to Lipschitz. obtained by resigning my three year directorship of the local polytechnic. [End of French text.2 Later development of ideal theory Dedekind continued to present new versions of his ideal theory. With the exception of Professor H. the keen desire for the mathematical public to learn the true value of first rate research.

) Dedekind and Weber developed an ideal theory for functions." is undoubtedly another reference to Riemann's work. Once again. and less readable. Kronecker (1882) published a long account of his own theory of fields and integers. rather than on calculation. though expressed in a very different. he used ideal theory to simplify Kummer's results on the Fermat problem. which he had been developing for some decades. Kronecker was almost diametrically opposite to Dedekind in mathematical philosophy.not only a turning point in the development of complex function theory. p. with their arithmetic theory of Riemann surfaces (Dedekind and Weber (1882)).7 The reception of ideal theory and Weber also made a major application of ideal theory outside number theory. as its name suggests.) The Zahlbericht. he opposed nonconstructive methods. but with the clear aim of elucidating properties of numbers. Thus the appearance of Hilbert's Zahlbericht in the Jahresbericht der Deutschen Mathematikervereinigung in 1897 was greeted with great joy. It was only in the 1920s that Artin and Emmy Noether created mod- ern ideal theory by abstracting the properties of rings that make unique . As Artin (1962) said: Dedekind's presentation is easy to read and elegant for us today. This is the great fruit of their collaboration in the publi- cation of Riemann's works . fields and modules heavily. language. but at the time it was too modern. He even claimed not to believe in irrational numbers. but in fact neither the Kronecker nor the Dedekind approach attracted an immediate following. In partic- ular. and in particular the use of infinite sets as mathematical objects. infinity. (The "modern theory of functions" cited by Dedekind in §12 of the memoir as the inspiration for proofs "based on fundamental char- acteristics. 46 0. Hilbert presented all the results known up to that time and. somewhat easier to prove). His definitions of irrational numbers (as pairs of sets of rationals) and "ideal numbers" (as the sets he called ideals) should have made the convenience of his position clear. made the results of Kummer accessible to a wider circle of readers. His theory included an equivalent approach to unique prime factorisation. (Translation of Artin (1962). rings. prime ideals are important. and unique prime factorisation is valid (in fact. Dedekind favoured the free use of infinite sets wherever it was necessary or convenient. 549. and entire algebraic functions playing the role of "algebraic" integers. through great simplifications. In the same year. Hilbert used groups. but also the beginning of modern algebraic geometry. with polynomials playing the role of "rational" integers. was intended to be a report on number theory.

one still finds the best of both worlds in the theory of algebraic numbers. In practice. Acknowledgements 47 factorisation into prime ideals possible. Victoria. The subject then became part of ring theory. and using these properties as axioms. Clayton. and to two anonymous reviewers at Cambridge University Press for valuable advice. Acknowledgements I am indebted to George Francis for sending me a copy of the Lipschitz- Dedekind correspondence when it was not available in Australia. Australia John Stillwell . Thanks also to Peter Stevenhagen and John McCleary for corrections and com- ments. at least in theory. and Dedekind's theories of algebraic numbers and algebraic functions became mere spe- cial cases.

313-342. Preuss. (1877). (1872). Vieweg. Springer-Verlag. Wiesbaden. 45-81. 59-80. volume 1. Edwards. Also in his. Math. Math. P. Algebra. Dedekind. volume 1. second edition. (1882). E. (1871). P. Dedekind. Reprinted by Chelsea. Baker. Dirichlet. 19. 13. fourth edition. Gottingen. deren erstes Glied and Differenz ganze Zahlen ohne gemeinschaftliche Factor sind. R. 92. H. Gauthier-Villars. Vieweg. 204-216. Die Bedeutung Hilberts fur die moderne Mathematik. G.. Dirichlet. 27. 1973. dass jede unbegrentze arithmetische Progression. (1977). Chicago. Collected Papers. A.. Beweis des Reciprocitatssatzes fur die cubischen Reste in der Theorie der aus dritten Wurzeln der Einheit zusammengesetzten complexen Zahlen. J. G. Beweis der Satz. in Essays on the Theory of Numbers. G. with Arithmetic and Mensuration. (1817). 411-496. G. Stetigkeit and Irrationalzahlen. (1871). (1894). Eisenstein. from the Sanscrit of Brahmegupta and Bhdscara. (1850). 547-551. (1962). 380-497. L. Wiley. M. English translation Continuity and Irrational Numbers. Abh. L. Uber einige allgemeine Eigenschaften der Gleichung. R. P. 289-310. (1839). Colebrooke. (1844). Jahrbuch Akad. 181-290. D. R. J. Also in his Mathematische Werke. L. Wiss. Dirichlet. and Weber. Bibliography Artin. Mathematika... Recherches sur diverses applications de l'analyse infinitesimal a la teorie des nombres. Math. Also in his Mathematische Werke. F. In Dirichlet (1871). H. (1894). Also in his Mathematische Werke. London. volume 1. Supplement XI. 48 . G. P. Reprinted by Martin Sandig. (1837). 1901. T. Dirichlet. Theorie der algebraischen Functionen einer Veranderlichen. unendliche viele Primzahlen enthalt. John Murray. Fermat's Last Theorem. Supplement X. Primes of the Form x2 + ny2. 1968. (1989). Dedekind. F. Open Court. von welcher die Theorie der ganzen Lemniscate abhangt. 324-369. Vorlesungen fiber Zahlentheorie. Sur la Theorie des Nombres Entiers Algebriques. Vorlesungen caber Zahlentheorie. Eisenstein. R. Linear forms in the logarithms of algebraic numbers. L. Wiss. H. A. In Dirichlet (1894). J. der Konigl. R. Dedekind. Cox. reine angew. Akad. Dedekind. refine and angew. reine angew. 434-657. G. (1966).

531-555.. Gauss.. I. 49-64. St. Petrop. In Fermat (1894). 1956. Comm. (1870). L. (1843). August(?) 1640. Comm. In Euler (1843). Berlin. P. Also in his Werke I. Fermat. Math. L. Soc. C. volume 1. 8. Petrop. Monatsber. Gauthier-Villars. 175-546. Theoria residuorum biquadraticorum. 1966. Springer-Verlag. Novi comm. (1756). 1984. Konigl. F. Fermat. (1744). C. Novi comm. Gratulationschrift der Univ. P. Euvres. 15 August 1657. Fermat. Also in his Werke. Breslau zur .. Cambridge University Press. 63-363. Euler. Correspondance Mathematique et Physique. Euler. In Fermat (1894). volume 2. (1657). (1910). D. 241-276. Gauthier-Villars. (1832). Fermat. I. Solutio generalis quorundam problematum diophanteorum quae vulgo nonnisi solutiones speciales admittere videntur. L. C. Letter to Frenicle. Petersburg. Wiss. volume 2. Sci. volume 2. 209. 7. tBuvres. qui radicibus unitatis et numeris realibus constant. 1-122. volume 2. P. The Thirteen Books of Euclid's Elements.. (1749). Euler. 224-287. F. L. 881-889. Kronecker. 1968. Disquisitiones Arithmeticae. Letter to Mersenne. Cambridge University Press. (1818). Also in his Mathematische Werke.. Sci. Also in his Werke. Jber. 212. 194-222. Euler. (1640a). 556-619. 310-314. 14. T. L. 493-495. 345. 92. (1640b). Observations sur Diophant. 12 April 1749. Comm. refine and angew. Also in his Gesammelte Abhandlungen. P. T. Grundziige einer Theorie der algebraischen Grossen. Reg. Math. 39. Rec.. L. De numeris complexis. Fermat. Rec. Gott. 4. volume 2. Reg. Gauss. (1654). Diophantus of Alexandria. acad. 271-282. L. Also in his Opera Omnia ser. Heath. L. Reprinted by Dover. 205-206. Hilbert. Fermat. (1897). Theoremata arithmetica nova methodo demonstrata. Petrop. Auseinandersetzung einige Eigenschaften der Klassenzahl idealer complexer Zahlen. L.. acad. J. Theoremata circa divisores numerorum in hac forma paa ± qbb contentorum. volume 3. P. (1801). Letter to Digby. Gott. Bibliography 49 nebst Anwendungen derselben auf die Zahlentheorie. volume 2. Letter to Frenicle. ser. Kronecker. 4. J. (1640c). In Fermat (1894). (1882). Fermat. I. In Fermat (1894). P. acad. 25 September 1654. Letter to Goldbach. English translation. English translation Elements of Algebra. Also in his Opera Omnia ser. (1894). Euler. Die Theorie der algebraischen Zahlkorper. volume 2. Letter to Pascal. 67-148. 25 December 1640. P. sci. second edition. Fermat. 6. Euler. Vollstandige Einleitung zur Algebra.. L. (1770). sci. (1761). (1896). Soc. (1925). 151-181. Reprinted by Johnson Reprint Corporation. 155-184. 428-445. In Fermat (1896). Math. deutsch. P. sci. Kummer. 18 October 1640. F. Yale University Press. Heath. (1670). 74-104. Theorematis fundamentalis in doctrina de residuis quadraticis demonstrationes et ampliationes novae. Akad. (1844). In Fermat (1894). Verein. reine angew. E. In his Opera Omnia. Gauss. E.

Konigsberg. 154-165. (1773). Weierstraf3 and anderen. Newton. Michigan Math. Also in his Collected Papers. Lipschitz. Birkhauser. Nouvelle methode pour resoudre les problemes indetermines en nombres entiers. Allgemeine Reciprocitatsgesetze fur beliebig hohe Potenzreste. The Mathematical Papers of Isaac Newton. E. Enseign. Briefwechsel mit Cantor. volume 1. Also in his Euvres. Berlin. 1976. (1945). R. 671-731. reine angew. de l'acad. E. roy. sci. E. Kummer. Paris. Yale University Press. G. de l'acad. and Sachs. volume 3. A. 305-319. E. Bestimmung der Anzahl nicht aquivalenter Classen fur die aus Aten Wurzeln der Einheit gebildeten complexen Zahlen and die idealen Factoren derselben. 279-302. E. 655-726. past and present. E. Kummer. Nouv. Wiss. Newton. Also in his Collected Papers. (1984). Uber die Zerlegung der aus Wurzeln der Einheit gebildeten complexen Zahlen in ihre Primfactoren. Also in his Collected Papers. 87-110. 24. 14. E. E. 24. H. 519-520. Deutsche Mathematiker Vereinigung. sci. Kummer. 98. volume 3. Zur Theorie der complexen Zahlen. mem. (1665). A. 265ff. (1850a). Miscellanea Taurinensia. E. volume 1. J. Monatsber. 203-210. J. Lagrange. E. Letter to Kronecker. volume 1. Springer-Verlag.. Berlin. J. (1967). reine angew. Also in his Euvres. Legendre. A. Stark. Akad. Monatsber. and Szego. Introduction to Kummer (1975). Kummer. L. Also in his (Euvres. (1768). 50 Bibliography Jubelfeier der Univ. Two lectures on number theory. 14 June 1846. Berlin. Helmholtz. Also in his Collected Papers. Lame. 93-116. 19ff. (1770). entitled Theorie des Nombres (1830). Springer-Verlag. Kummer. (1975). Akad. J.. (1847a). Mathematical Cuneiform Texts. Akad.. E. volume 1. Recherches d'arithmetique. G. 299-322. Weil. A: M. volume 2. Berlin. volume 1. volume 1. Neugebauer. Beweis der Fermat'schen Satzes der Unmogligkeit von xA + yA = z" fur eine unendliche Anzahl Primzahlen A. (1986). (1850b). Third edition. 1-27. In Newton (1967). 87-96. Cambridge Universtity Press. M. . E. Solution d'un probleme d'arithmetique. 20.. E. (1967). (1846a). 4. Polya. Mem. I. Essai sur la Theorie des Nombres. In Kummer (1975). A. (1847b). Lagrange. O. Monatsber. Kummer. Dedekind. Math. 274-297. (1974). (1924). J. English translation Problems and Theorems in Analysis. 345-357. Also in his Collected Papers. Collected Papers. Of the nature of equations. (1846b). Comptes Rendus. (1975). reprinted by Blanchard. Weil. L. Number Theory: an Approach Through History. I. 327-367. 132-141. Wiss. 211-251. Duprat. G. Also in his Collected Papers. E. Also in his Collected Papers. Aufgaben and Lehrsatze aus der Analysis. Kronecker. L. volume 1. Berlin. Kummer. 695-795. Demonstration general du theoreme de Fermat. volume 1. 310-315. Math. 1955. Springer-Verlag. Weil. A complete determination of the complex quadratic fields of class-number one. Wiss. J. (1847). Lagrange. 165-192. Math. (1798).

Part two Theory of algebraic integers .

.

where a and b are any rational integers. for the first time. a2. . Because of the extraordinary scope of this field of mathematical research.. 1832. were extended when Gauss introducedt complex integers of the form a+b . in the present memoir. Introduction In response to the invitation it has been my honour to receive. previously going under that name.. proved by Euclid.. A number 0 is called an algebraic number if it satisfies an equation =0. was established in its essentials by Euclid. ±1. which I shall try to clarify in the remarks that follow. the fundamental theorem that each integer is uniquely decomposable into a product of primes is an immediate consequence of the theorem. I propose. It is t Elements. Two thousand years later. II. however. and which I shall call rational integers from now on. an_1i an. to develop the fundamental principles of the general theory of algebraic integers I published in the second edition of Dirichlet's Vorlesungen fiber Zahlentheorie. ±2. Gauss gave. He showed that the general laws of divisibility of these numbers are identical with those that regulate the domain of rational integers. The theory of divisibility of numbers. t Theoria residuorum biquadraticorum.t that a product of two numbers is not divisible by a prime unless the prime divides one of the factors. The broadest generalisation of the notion of integer is the following. VII... . 32. with finite degree n and rational coefficients a1. The numbers 0. an extension of the notion of integer. I restrict myself here to the pursuit of a single goal. 53 . which is the basis of arithmetic. At any rate.

numbers. It follows immediately from this definition that the sum. I call such a set St a field of degreet n. when it satisfies an equa- tion of the f o r m above in which all the coefficients a1.) . . If 9 is an algebraic number there is. products and quotients also belong to the set 0 . of degree n". the numbers of the form 0(9) = xo + x19 + x292 + + xn-19n-1. 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. differences. Consequently.. of integral factors which are not units. exactly one on +a19n-1 +. and which we call for this reason irreducible. a 2 . an_1i an are rational integers. If xo. since every integer which is not a unit is always the product of two. The numbers 0(9) belonging to a field 1 are now partitioned. If we define the norm of any number t Dedekind calls it "finite. The system o is evidently identical with the system of all rational integers when n = 1. following the definition above. the set of which we call 0. or with the complex integers when n = 2 and 0 =. an integer a will be said to be divisible by an integer /3 if a = /3ry. Certain phenomena which occur in these two special domains o occur again in every domain o of this nature. 54 Introduction called an algebraic integer. difference and product of integers are also integers.. However.. of minimal degree.. Si. The problem we set ourselves is to establish the general laws of division that govern such a system o. in the following manner.. Above all. among the in- finitely many equations with rational coefficients satisfied by 0. into two large sets: the set o of integers and the nonintegral. will also be algebraic numbers. the unlimited decomposition which prevails in the domain of all algebraic integers is never encountered in a domain o of the kind indicated. and they enjoy the fundamental property that their sums.. primes do not exist. xn_1 denote arbitrary rational numbers. in the domain of all integers we are considering at present. An integer a will be called a unit when every integer is divisible by e. where ry is likewise an integer. 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. it is easy to see that. (Translator's note. .+ an-10 +an = 0. Nevertheless.-. By analogy. or simply an integer. or fractional. or rather any number. x2. as one easily sees by consideration of norms.

whereas . then the conjugate numbers µ1.0) where a and 0 are any two numbers of the field Q.. such that each class is the set of all numbers congruent to a given number (representing the class). let any two numbers a. In fact. An immediate consequence of this result is that N(p) = ±1 just in case p is a unit. In his researches on the domain of numbers belonging to the theory of circle division. 9n-1 denote all the roots of the irreducible nth degree equation.. ttl = Y'(01). equal at most to the absolute value of N(p). as we know. neither of which is a unit..3) = N(a)N(. it evidently follows from the above that each decomposable number can always be represented as the product of a finite number of indecomposable factors. They differ in a manner so complete and so essential as to leave little hope of preserving the simple laws that govern the old theory of numbers. and hence a number in o.µ2. An-1 = G(9n-1) where 9.. just as in the theory of rational or complex integers. .. hence corresponding to equations of the form 9' = 1. Now if it is an integer. called the modulus. Introduction 55 p = 0(9) belonging to the field fI to be the product N(p) = pµ1µ2 . This result again corresponds completely with the law holding in the theory of rational or complex integers./3) is or is not divisible by p. 91i 92. Then we can. At the same time. a rational number.. and so N(p) will be a rational integer. one always has N(a. . observed un- til now with the old theory. partition all numbers of the system o into number classes. . according as their difference ±(a . µ2 = 0(92). And a deeper study shows us that the number of classes (with the exception of the case p = 0) is always finite. pn-1. But at the same time this is the point where the analogy.. In fact. µn_1 will likewise be integers. namely that each composite num- ber is representable as the product of a finite number of prime factors. Now if a number in the system o is called decomposable when it is the product of two numbers of the system. then N(p) will always be. This norm plays an extremely important role in the theory of numbers in the domain o. /3 in this domain be called congruent or incongruent relative to the third p. . . whose factors are the conjugate numbers p = 0(0). is in danger of being irrevocably broken.. and never 0 unless p = 0.. Kummer noted the existence of a phenomenon distinguishing the num- bers of that domain from those considered previously.

Section 0. one discovers that. Kummer said "Maxime dolendum vide- tur. The set of all actual numbers. 1844. quae magnis adhuc difficultatibus laborat. as a prime or as a product of prime factors. considered as a special case of ideal numbers. Translator's note. which are decomposable into n linear factors with algebraic coefficients. actual or ideal.t the more one has to admire the steadfast efforts of Kummer.4. for an English translation of this passage. forms the principal class. To each class there corresponds an infinite system of equivalent homogeneous forms. The great success of Kummer's researches in the domain of circle di- vision allows us to suppose that the same laws hold in all numerical f In the memoir: De numeris complexis qui radicibus unitatis et numeris integri realibus constant (Vrastislaviae.56 Introduction each number in the domain of rational or complex integers decomposes uniquely into a product of primes. Or what amounts to the same thing. which yield two actual numbers when combined with the same ideal number are called equivalent. . in n variables and of degree n. he obtained the surprising result that the laws of divisibility in the numerical domains studied by him were now in complete agreement with those that govern the domain of rational integers. quod haec numerorum realium virtus. in the numerical domains considered by Kummer.) t Zur Theorie der complexen Zahlen (Crelle's Journal.6. inasmuch as a prime cannot divide a product of two or more factors without dividing at least one of the factors. The number of these classes is finite. and Kummer succeeded in determining their number by extending the principles used by Dirichlet to determine the number of classes of binary quadratic forms. But the more hopeless one feels about the prospects of later research on such numerical domains. quae si esset tota haec doctrina.t By considering the indecomposable numbers which lack the characteristics of true primes to be products of ideal prime factors whose effect is only apparent when they are combined together. Two ideal numbers. (See the Introduction. whether as divisor or dividend. which were finally rewarded by a truly great and fruitful discovery. 35). That geometer succeeded in resolving all the apparent irregularities in the laws. one discovers that the indecompos- able numbers lack the characteristic of genuine primes. facile absolvi et ad finem perduci posset". non eadem est numerorum complexorum. whether prime or composite. and all the ideal numbers equivalent to the same ideal number form a class of ideal numbers. §8). a number may be representable in sev- eral entirely different ways as the product of indecomposable numbers. Each number which is not a unit behaves consistently in all divisibility situations. ut in factores primes dissolvi possint qui pro eodem numero semper iidem sint.

until I abandoned the old formal approach and replaced it by another. The power of this concept resides in its extreme simplicity. I began by building on the theory of higher order congruences. by this route. . it is nevertheless to be feared at first that the language which speaks of ideal numbers being determined by their products. and it is sufficient to consider a system of actual numbers that I call an ideal. he says that a is divis- ible by an ideal number corresponding to the property A.. B corresponding to particular ideal numbers. seems all the more necessary since the ideal numbers do not actually exist in the numerical domain o. Brunswick. to the effect that a satisfies one or more congruences. In the latter approach I need no concept more novel than that of Kummer's ideal numbers. In my researches. However. which was the subject of a pamphlet of mine (Stetigkeit and irrationale Zahlen. while this method leads to a point very close to my goal. Assuming that the arithmetic of rational numbers is soundly based. a fundamentally simpler conception focussed directly on the goal. without exceptions. C. My . may lead to hasty conclusions and incomplete proofs. presumably in analogy with the theory of rational numbers. and then to indicate how one can derive. or rather the necessity. multiplication and division on them. but only the divis- ibility of these numbers. If a number a has a certain property A. 1872). . Kummer did not define ideal numbers themselves. I have not been able to surmount. a precise definition covering all the ideal numbers that may be introduced in a particular numerical domain o. that I first published in the place mentioned above. B. the property C corresponding to their product. which must always be imposed with the introduction or creation of new arithmetic elements. On the other hand. the goal of which has been to arrive at a definitive answer to this ques- tion. the question is how one should introduce the irrational numbers and define the operations of addition. that serve to introduce the ideal numbers. and my plan being above all to inspire confidence in this notion. Introduction 57 domains o of the most general kind considered above. To satisfy these demands it will be necessary and sufficient to establish once and for all the common characteristic of the properties A. While this introduction of new numbers is entirely legitimate. from properties A. of such demands. I did not achieve the general theory. I shall try to explain the train of thought that led me to it. subtraction. and at the same time a general definition of their multiplication. certain exceptions to the laws holding in other cases. since I had previously noticed that the latter theory allows Kummer's researches to be shortened considerably. And in fact this danger is not always completely avoided.t f The legitimacy. becomes more evident when compared with the introduction of real irrational numbers.

c. it is impossible to engender new numbers in this domain 91.. Since a characteristic property A serves to define. For each section of this kind one creates or introduces into arithmetic a special irrational number. Thirdly. in terms of these two sections. and one can really prove propositions such as. 2.. in place of the properties A. be regarded first demand is that arithmetic remain free from intermixture with extraneous elements. 4. in the strict sense of the word. f = f . corresponding to the section (or engendered by it). 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. the corresponding ideals a. and for this reason I reject the definition of real number as the ratio of two quantities of the same kind. B. and not successively as roots of equations. When defined in this way. Now if.58 Introduction This problem is essentially simplified by the following considerations. and which are always present as factors of composite numbers. etc. then one easily defines a > 03 or a < 33 in terms of the sections they engender.3 are any two real numbers (rational or irrational). 5. b. the definition should be of a kind which also permits a perfectly clear definition of the calculations (addition. .. Each section of this domain 91 is produced by a particular number of the domain itself. etc. defining the ideal numbers. 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. On the contrary. one can easily define. but only the divisibility of the numbers in o by the ideal number. 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). I now call such a system an ideal for short.) one needs to make on the new numbers. which I only sketch here: 1. 3.. Secondly. product and quotient of the two numbers a. the irrational numbers unite with the rational numbers to form a domain 91 without gaps and continuous. the four sections corresponding to the sum. one can consider. it is evidently necessary that the numbers actually existing in o. which had not previously been done. Each particular rational number a engenders a particular section (or two sections. difference. for example. 0. so that for each particular ideal number there corresponds a particular ideal a.. 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. not an ideal number itself.. in order to establish their common and exclusive char- acter. If a. There are infinitely many sections which cannot be engendered by rational numbers in the manner just described. In this way the four fundamental arithmetic operations are defined without any obscurity for an arbitrary pair of real numbers. Moreover. one should demand that all real irrational numbers be engendered simultaneously by a common definition. as far as I know. as logarithms. One achieves all of this in the following way. . C. 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. 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. conversely.

Thus if p is a particular number of o. Now. II. We have therefore found a common characteristic of all ideals: to each actual or ideal number there corresponds a unique ideal a. but also when it is ideal. If a = pw is divisible by µ. by introducing ideal numbers and a corresponding lan- guage. conversely. the notion of inte- ger established above immediately yields the following two elementary theorems on divisibility: 1. A fact of the highest importance. That is. The sum and difference of any two numbers in the system a are always numbers in the same system a. but also sufficient conditions for a numerical system a to be an ideal. 2. Now. since each product ww' of integers w. to the numbers w of our numerical domain o. Any product of a number in the system a by a number of the system o is a number in the system a. The latter system is evidently not altered when one replaces p by eµ. it is apparent that the definitions of these ideal numbers and their divisibility should be stated in such a way that the elementary theorems 1 and 2 above remain valid not only when the number p is actual. the system a of all numbers a = pw in the domain o divisible by p likewise has the essential character of an ideal. but for all ideals. Properties I and II are therefore not just necessary. then so are their sum a+a' = p(w+w') and their difference a-a' = µ(w-w'). If two integers a = µw.w' of two integers w. is that. since the sum w + w' and difference w . as we pursue the goal of restoring the laws of divisibility in the domain o to complete conformity with those ruling the domain of rational integers. Introduction 59 as a special case of ideal numbers. it is the set a of all numbers a of the domain o divisible by a particular number. enjoying the properties I and II. and it will be called a principal ideal. with p denoting a particular one of these numbers and a the corresponding principal ideal. Consequently. where a is any unit in o. If we apply these theorems. we obtain the following two fundamental properties of such a numerical system a: I. Any other condition imposed on the . each system enjoying properties I and II is also an ideal. and after surmounting the greatest difficulties. each number aw' = p(ww') divisible by a will also be divisible by p. the properties I and II should hold not only for principal ideals. true for all integers. w' is itself an integer. a' = pw' are divisible by the integer µ. either an actual number or an ideal number indispensable for the completion of the theory. w' are themselves integers. which I was able to prove rigorously only after numerous vain attempts.

60 Introduction
numerical system a, if it is not simply a consequence of properties I and
II, makes a complete explanation of all the phenomena of divisibility in
the domain o impossible.
This finding naturally led me to base the theory of numbers in the
domain o on this simple definition, entirely free from any obscurity and
from the admission of ideal numbers.t
Each system a of integers in a field S2, possessing properties I and II,
is called an IDEAL OF THAT FIELD.
Divisibility of a number a by a number p means that a is a number
pw in the principal ideal corresponding to the number p and which can
be conveniently denoted by o(p) or op. At the same time it follows from
property II or theorem 2 that all the numbers in the principal ideal oa
are also numbers in the principal ideal op. Conversely, it is evident that
a is certainly divisible by p when all numbers in the ideal oa, and hence
a itself, are in the ideal op. This leads us to establish the following
notion of divisibility, not just for principal ideals, but for all ideals:
An ideal a is said to be divisible by an ideal b, or a multiple of b, and
b is said to be a divisor of a, when all numbers in the ideal a are also in
b. An ideal p, different from o, which has no divisors other than o and
p, is called a prime ideal.t
Divisibility of ideals, which evidently includes that of numbers, must
at first be distinguished from the following notion of multiplication and
the product of two ideals:
If a runs through all the numbers in an ideal a, and Q runs through
all the numbers in an ideal b, then all the products of the form a,3, and
all the sums of these products, form an ideal called the product of the
ideals a, b, which we denote by ab.§
One sees immediately, it is true, that the product ab is divisible by a
and b, but establishing the complete connection between the notions of
divisibility and multiplication of ideals succeeds only after we have van-
quished the deep difficulties characteristic of the nature of the subject.
This connection is essentially expressed by the following two theorems:
If the ideal c is divisible by the ideal a, then there is a unique ideal b
such that ab = c.
t It is of course permissible, though not at all necessary, to let each ideal a correspond
to an ideal number which engenders it, if a is not a principal ideal.
t Likewise the ideal number corresponding to the ideal ab is said to be divisible by
the ideal number corresponding to the ideal b, and corresponding to a prime ideal
one has a prime ideal number.
§ The ideal number corresponding to the ideal ab is called the product of the two
ideal numbers corresponding to a and b.

Introduction 61

Each ideal different from o is either a prime ideal, or uniquely ex-
pressible as a product of prime ideals.
In the present memoir I confine myself to proving these results in a
completely rigorous and synthetic way. This provides a proper founda-
tion for the whole theory of ideals and decomposable forms, which offers
to mathematicians an inexhaustible field of research. Of all the later de-
velopments, for which I refer to the exposition I have given in Dirichlet's
Vorlesungen fiber Zahlentheorie and other memoirs still to appear, I have
included here only the partition of ideals into classes, and the proof that
the number of classes of ideals (or of classes of the corresponding forms)
is finite. The first section contains only the propositions necessary for
the present goal, extracted from an auxiliary theory, also important for
other researches, which I shall expound fully elsewhere. The second sec-
tion, which aims to clarify the general notions by numerical examples,
can be omitted entirely. However, I have kept it because it may help in
the understanding of the later sections, where the theory of integers in
an arbitrary field of finite degree is developed from the above viewpoint.
To do this one needs to borrow just the elements of the general theory of
fields, a theory whose further development leads easily to the algebraic
principles invented by Galois, which in their turn serve as a basis for
deeper researches into the theory of ideals.

1
Auxiliary theorems from the theory of
modules

As I have emphasised in the Introduction, we shall frequently have to
consider systems of numbers closed under addition and subtraction. The
general properties of such systems form a theory so extensive that it can
also be used in other researches; nevertheless, for our purposes just the
elements of this theory are sufficient. In order to avoid later interruption
to the course of our exposition, and at the same time to make it easier to
understand the scope of the concepts on which our theory of algebraic
numbers is based, it seems appropriate to begin with a small number
of very simple theorems, even though their interest lies mainly in their
applications.

§1. Modules and their divisibility
1. A system a of real or complex numbers will be called a module when
all the sums and differences of these numbers also belong to a.
Thus if a is a particular number in the module a, all the numbers
a + a = 2a, 2a + a = 3a, . .

a-a=O, 0-a=-a, -a-a=-2a, .. 7

and consequently all numbers of the form xa also belong to a, where x
runs through all the rational integers, that is, all the numbers
0,fl,±2,±3,....
Such a system of numbers xa itself forms a module, which we denote
by [a]. Consequently, if a module includes a nonzero number then it

62

and. and thus composed entirely of numbers belonging to both a and b. b because each common multiple of a.3. and hence also to the system in. b is also a divisor of a. whence 6±6' = (afa')+()3±. it is evident that any module is divisible by itself. Each of the two numbers p ± p' will belong (by 1) not only to the module a but also to the module b.. b divisible by each other are identical. . if the module m' is any common multiple of a. 4.3). The zero module is therefore a common multiple of all modules. and b a divisor of a. a is a module.3 will form a module.3 to all the numbers in a module b. which we shall denote by a = b. and that two modules a. whence it follows that m is a module. m is a common multiple of a. p' be any two numbers in the system in. forms a module by itself. is divisible by its immediate successor then it is clear that each will be divisible by all its successors. b. Finally. 0' to the module b. Moreover. a' belong to the module a and . b. 3. if a is a particular number in a module a. b because every common divisor of a. It will be called the least common multiple of a. m' is divisible by in. Since all members of this module m are in a and also in b. if each of the modules a.3 ±. which is in each module. . c. the numbers 6 ± 6' also belong to the system a. let p. 2. Moreover. S' in the system a can be put in the form 6 = a + /3. Indeed. Modules and their divisibility 63 includes an infinity of different numbers.3' are in b. all the numbers a = a + 0 . It is also evident that the number zero. b is divisible by m. §1. then the module [a] will be di- visible by a. and hence in both a and b. That is. Indeed. that is. then the system D of all numbers a +. when all the numbers in the module a are also in the module b. a. any two numbers 6. Also. A module a will be called divisible by the module b or a multiple of b. b be any two modules. since the numbers a ± a' are in a and the numbers . Let a. then (by virtue of the definition of the system m) these numbers will also be in in. Since the number zero is in every module. If a becomes equal in succession to all the numbers in a module a. b. and. The system m of all the numbers that belong to both modules will itself be a module. This module is called the greatest common divisor of a. 6' = a'+ 0' where a.

all the numbers in the module D.Q = 0 +. b. of the modules [a]. We immediately deduce the following simple propositions. 64 Chapter 1. Thus D is divisible by D'. 2. it may be useful to justify the terminology chosen. 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+. by which we mean the set of those numbers congruent to a particular number. modulo a. If w w' (mod a) and x is any rational integer then xw .w' (mod D).w') is in a or not. 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. also belong to the module D'.w" (mod a) then w . Congruence of the numbers w. Also. w' will be called congruent or incongruent modulo a according as their difference ±(w . b are two particular rational integers.w" (mod a). and hence to each other.w' ± w" (mod a). The first of the preceding theorems leads to the notion of a class of numbers relative to a module a. If a.w' (mod m). w' with respect to the module a will be indicated by the notation w .w' (mod b) then w . the latter is a common divisor of a and b. [b]. whose proofs we can omit: If w w' (mod a) and w' . and [d] the greatest common divisor. b. Having carried out these rigorous proofs.w' (mod a). then w . Neverthe- less. m their least com- mon multiple and d their greatest common divisor. Let a be a module. where m is the least common multiple of a. that is. If w w' (mod a) and D is a divisor of a. if the module D' is any common divisor of a.3 of the module b belong to the module D. 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. Such a class modulo a is completely determined by giving a . Consequently.3.xw' (mod a). If w . by the following remark. Congruences and classes of numbers 1. If w w' (mod a) and w .w' (mod a) and w" = w"' (mod a) then w ± w" . Auxiliary theorems from the theory of modules of the module a and all the numbers . it follows from the elements of number theory that [m] will be the least common multiple. Numbers w. §2.

A deeper examination of such a system (/3r) of representatives now leads to the following theorem: 3. §2. it is evident that any numbers /3. 0' in the module b which are congruent modulo a are congruent modulo m. and hence also modulo m. 33. with least common multiple m and greatest common divisor Z. a). and. a). will be denoted by (b. consequently we have (b. they are incongruent modulo a.. on the contrary. a) _ (b. represented by the number zero. The second part is proved in absolutely the same way: since b is divisible by a. Let a. but which represent all classes having members in b. because f3 . where a is in a and /3 is in b. when it is finite. And. and since any two different representatives are incongruent modulo a and hence also modulo m. and the number of numbers or. since each number 3 in the module b is congruent to one of the representatives . b be any two modules. Now. or of the classes they represent. since each number 6 in a is of the form a + /3. a). by hypothesis. it will be convenient to assign the value zero to the symbol (b. and to only one. . in such a way that each number in b is congruent modulo a to one of these numbers. First of all. Congruences and classes of numbers 65 single member. for example. we have 6 =a+0=f3 (mod a). Any complete system of representatives of the module b modulo a will at the same time be a complete system of representatives of the module b modulo m. the numbers Nr 3are likewise in a and. Q. since /3 and consequently 6 is congruent to one of the numbers /3r modulo a.3. the numbers /3r form a complete system of representatives for the module a modulo a.D. Now if b is a second module we can always choose in b a finite or infinite number of numbers (or) 01) 32. If. and each member can be regarded as a representative of the whole class. and hence also in m. . . modulo a. I call a complete system of representatives of the module b modulo the module a. which are mutually incongruent modulo a./3' is in a as well as in b. m) = (a. form such a class.E.. The numbers in the module a. and for the module a modulo a. these numbers Nr in b form a complete system of representatives of the module b modulo m. the number of representatives Or is infinite. Such a system of numbers 3r in the module b.

it remains to see that each number y in c is con- gruent to one of the numbers /3r + ys modulo a. since y'.D. In the first place.Qr runs through the representatives of b modulo a and if ys runs through the representatives of c modulo b. w . any two of which are incongruent modulo b. Q. 5. they are all incongruent modulo a. If a is a divisor of a and at the same time a multiple of c. since all numbers in b are congruent to each other and hence . all the numbers Nr+ys belong to the module c. of two modules a. because all numbers in b are .Qr + ys form a complete system of representatives of the module c modulo a./3" (mod a). a = a. 4. a) = (c. Auxiliary theorems from the theory of modules If b is divisible by a then (b. since . In the second place. if (b. In fact. since each number y is congruent to one of the numbers 'ys modulo b. Conversely. We then have y=a+'Ys =a+Qr+ys =fir+ys (mod a). /3" likewise are particular members of the series /3r. b) (b. a).o (mod b) has a common root if and only if .0 (mod a). Also. and lys is likewise in c. a) = 1. that y' . and let p. we can choose /3 = a + or. /3" be particular values of fir. In fact if we let /3'. and let y'. y" are particular members of the series ys. The system of two congruences w . We evidently have at the same time m = b. which proves the assertion above. 66 Chapter 1. Now since /3'. and consequently (c. then the numbers . we must have /3' = /3". y" be particular values of 'ys.Qr is in b and hence also in c. since a is divisible by b and /3' .0 (mod b). where a is a number in the module a.0 (mod a). b. Let m be the least common multiple. since each of these numbers /3 is congruent to one of the numbers or modulo a. where /3 is a number in the module b. o be given numbers. However.-y" (mod b). we can choose y = /3 + ys. any two of which are incongruent modulo a. In the third place. a) = 1 then b will be divisible by a.E. moreover if . and a the greatest common divisor.p (mod a). then the hypothesis /3' + y' = /3" +'y" (mod a) implies. we must have y' = y" and hence the hypothesis above becomes /3' .

. . Conversely. /3n will be called a basist of the module. . Let /31i . . . but here I confine myself to proving the following fundamental theorem. Finitely generated modules 1. if this condition is satisfied then (by virtue of the definition of a in §1.) . and hence also in m. be particular numbers. )3n] can be transformed into the members of a module a by multiplication by nonzero rational numbers. if w' is a number satisfying the same conditions as w.132. .Q. yn are arbitrary rational integers.a = a +. . Thus the condition is also sufficient. where yi. (Translator's note.or (mod i) is necessary.w will also be in both a and b. y3.a will be in a. This module [/31i /32... . Moreover.133. If all the numbers in a finitely generated module b = [/31. Qn] The complex of constants /31 i /32.. . he re- quires them to be independent for fields (§15). § 3.. y2. b respectively. If there is a number w satisfying the two congruences then the numbers w . It is easy to see that each multiple of a finitely generated module is itself a finitely generated module. each number w' in the class represented by w modulo m will satisfy the congruences.]. and all such numbers w form a class of numbers modulo the module m. . That is...4) there is a number a in a and a number 3 in b whose sum a +.ar. . (Translator's note. hence the number w = p . All the numbers 13= y1 N1 +Y2132+y3133+"'+yn13n. . P. which we call a finitely generatedt module [131.3 = p . §3.. which means that w' = w (mod m). which will later have important applications.3 satisfies the two congruences. and since the latter are both in a.... the difference p .or of the numbers will likewise be in Z. w . [ 32]. /32.E.. the above condition p ..133. 2. .) t Note that Dedekind's basis elements need not be independent.D.. Conversely. N3.. . Q..p. /3n] is evidently the greatest common divi- sor of the n finitely generated modules [/31]. then w' . However. evidently form a module. then the least common multiple m of a t Dedekind calls them "finite"... . C 3 2 . Finitely generated modules 67 p = a (mod a).

t2. and if s denotes the absolute value of the product 818283 sn...] divisible by b let denote those belonging to module a and hence also to module b.31.. the numbers s/31. t3 tn) whose numerators and denominators are rational integers. the coefficient y is divisible by av(v).3n. + a(n)/3n. and consequently all products s/3. . the products s1/31.. . a) m) = a'1a2a3' .. belong to the module a. t3. .. sn/3n likewise belong to a. t2....68 Chapter 1.. This is because one can always put yv = xvavv) +' 1/v where xv and y. One can then see that.. form a basis of m. s. 2. . . .1). where /3 denotes any number in the module b. By hypothesis there are n nonzero fractions Si 82 83 Sn t1. . in all the numbers µv.. 8/32i 8.. /3. for which y takes the smallest positive value av(v).. such that the n products 81 82 83 8n /31. to (§1. and at the same time (b. . Now let v be a particular index from the sequence 1.03. Auxiliary theorems from the theory of modules and b will be a finitely generated module. -/7 /33. are rational integers and the latter satisfies the conditiont 0<yv<a'. . 83/33. ann). f Which is the foundation of the theory of division for rational integers. . 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 + + + . Since members of a module a are changed into other members of a when multiplied b y rational integers tl. n. 82/32... ..7 on ti t2 t3 tn belong to the module a./32. Among these numbers µ'v there will be at least one number µv = alv)/31 + a2v)/32 + + a(v)/3v... for example s/3. Among the numbers in the module [.

obtained by putting v = n.xvav-1 the number PV . n . . . µn themselves belong to m. because µl. . x2i . . it is necessary that yv = 0. p2.").31. 1 in succession.. But since (by definition of the µ") the coefficient of Qv is less than a.)3. xn are rational integers. theorem.02 . §3. . enjoy the properties enunciated in the ...") be divisible by a. .. and so on. Finitely generated modules 69 Then if we put i (v) .x"a2 . . /3n_l]. ... ... that is.(. (v) i (v) yl = yl .. x2.xvµv = Y I31 + y2.(. since p' and µv are in m..J and the module m.Qv belongs to both the module [. /L2. each number 14 belonging to the two modules a and [01] is of the form Al = xuILl where xl is a rational integer. + xn/Ln. and hence also to the module M. Finally. Each number µ'n-1 of this nature is of the form An'-1 = /Ln-2 + xn-l/Ln-1..xval ... /3v_1] and also in m. .(32.. #2. . It follows easily that the n numbers µ". .3v-1 + y. as required. 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.") and at the same time positive. Thus it is proved that each number µ of the module m can be represented in the form µ = xl/Ll + i2µ2 + . .. yv-1 = yv-1 . . since an arbitrar- ily chosen system of rational integers xl.. /n]... where x1. (32i . 2. At the same time AV xvµv = AV-1 becomes a number in [. Conversely..1. µn form a basis of the module m. .. Each number µ in the module in. each number Mn in both a and b = [. or else becomes zero in the case v = 1. the n numbers µl.. . y2 = y2 .31.32 + + yv-1.-. and hence that y" = x"a.. where xn_l is a rational integer and µ'n-2 is a number belonging to the two modules a and Qn-1]. xn certainly produces a number p in the module in.31.

". z2.+znfn of the module b for which the rational integers zi.0'. . We shall show that these numbers .. z. . . form a complete system of representatives of the module b modulo m (or a).<a("). z1 .Zn-11 Z2/ .... z.zn.. . contrary to the definition of the number µ. which satisfy the conditions above. then the numbers z' l'... also produce two numbers 0' in the module b which are incongruent modulo a.. (n) (n-1) zn-1 = zn-1 + an-lxn + an-1 xn-1 ..+znOn in the module b is congruent modulo a (or m) to one of the numbers . satisfy the n conditions 0<z'. + (z' _ zv)QV is evidently a number p' in a and [01.. In the first place.(.) x2ix1 so that the n numbers zn = zn + ann)xn.. z2 .. z2. In the second place. . Auxiliary theorems from the theory of modules To prove the last part of the theorem we have to consider all the numbers x1131+z2132+. . zn.z2... is positive and < a..z1 are not all zero.. + zn/33 = z'.xn-1. z2'.3n] for which the coefficient of 3.. it is clear that we can successively choose n rational integers xn. .' satisfy the same n conditions as the numbers z'1....01 + . zn. Then the difference (zi . z2. Thus any two different systems of n numbers z'1..be the first of them with a nonzero value. . + zn'fn (mod a). the number of which is evidently equal to a' a2 a. . a value which we can assume to be positive by symmetry.(" ).. Then if the n differences zn .zi)al + . it is easy to see that an arbitrary number 0 =x101+x202+ . .. ..0' since. zn are given. . all the numbers Q' in the module b are incongruent modulo m. 70 Chapter 1. let z . If 4f1 + . if z1.02 . n I II zn-1 .. ... and which is also < aU") since both numbers z' and z' are < a( v).

a2..... and its members will be called independent. when there is a system of rational numbers x1i x2. . §4.. an will be called an irreducible system. . . and the numbers themselves will be called dependent on each other. not all zero..1 +x2/ 2 +.. + c.. ... W... + a2x2... In the contrary case. a2i .... z2 = z2 + a2n)xn + a2n-1)xn-1 ... A system of n numbers al..... and hence Q .. a2 = ci al +c2a2 + + cnan. .... . then the n num- bers al = cial + c2a2 + . a single number evidently forms a reducible or irreducible system according as it is zero or not. whose number can be increased enormously.. It then follows that any two different systems of ratio- nal numbers x1i x2.. 2. + cnan.3' (mod m). on the determinants of rational numbers.... .. Xn produce unequal sums a....... satisfy the n conditions 0 < z..(..... If we now put =z1/1+z2/32+.. .E... If the n numbers a1...... Q.... a'n = cin)a1 + c2n)a2 +. an will be called reducible. If one wants to retain this terminol- ogy in the case n = 1.. Irreducible systems 1...... xn which are not all zero.... (n) xn+a1(n-1) i z1 =z1+a1 x1...'). an are independent... . Irreducible systems 71 ....... when the sum a=x1a1+x2a2+"'+xnan is nonzero for any system of rational numbers x1i x2.... a1 we have )3' = 0 +x1/. . a2i ..(nn)an. that is... ... .. xn...D. +xnAn...+z'On.. < a. The definition above easily yields the following theorems... for which the sum a is zero. then the system of numbers a1.

.r quantities xr+l. cannot vanish unless we simultaneously have C11x1 + C1 x2 + . . a'n are dependent. This is seen immediately when all the n2 coeffi- cients c vanish.. x2i . xn satisfying the preceding equa- tions. Cr(r) of maximal rank r < n such that the minor determinants of higher degree vanish.. C22x2 + C2'x2 + ... not all zero. . . .. + C2n)xn.r of the equations above will be consequences of the preceding r.. +pnxn... Thus we have a system of n rational numbers X17 X21 ... . an are independent in that case. X 2 . Hence in this case the n numbers a'. a2i ... a2. xn are arbitrary rational numbers. . among the minor determi- nants of C that do not vanish there will be one. . Now if we give the n . and not all zero.E. Since al. Cnxl+C'n'x2+. cnn) is nonzero or not. . Auxiliary theorems from the theory of modules whose n2 coefficients c are rational numbers. a'2. . + C(1n)xn = 0.. form an irreducible or re- ducible system according as the determinant C = E ±C1 C2 .+xnan =a'.... . where the r(n-r) coefficients p are rational numbers.. as we know. Q.+Cnn)xn=0.. which is impossible when C is nonzero.. xn. for which the sum a' is zero. the sum x1a' +x2a2' +. But if we have C = 0 there is always a system of rational numbers xl. the last n . Hence the numbers al. In this case. 72 Chapter 1... If this is not the case then.. say n f C1iC2 .. . xr will likewise take rational values. the quantities x1. and we can put these r equations in the form x1 = P'+lxr+1 + . . . = 0.. . xn arbitrary rational values then not only can we ensure that they are not all zero. not all zero.. where xl... r xr = pr+lxr+l + +pnr)xn.D. an are independent...

.... en(n) = ±1. Substituting in them the first n equations for the n numbers a'.+bn")/.. . and if n numbers al. it follows that the sum el")c'... and a'. a2. .C. Irreducible systems 73 3.an form one basis of a module a. depend on them via n equations of the form =bl"). .... .D. since the n numbers a are in the module a = [ai.... v' are equal or not. where the coefficients c are rational integers whose determinant = ±1. a. In fact. a2.. . E fe' e2 . Then the product of the determinants C(n) .. 4. a2i . an] there are in any case n equations of the preceding form. it is clear that [ai. Conversely. . + c(")an...(31 i . whose coefficients e are likewise rational integers. . fe1'e2 .. where the n2 coefficients c are rational integers whose determinant is ±1. a2i . + c(")an. forming the basis of a module a. .. and consequently the numbers ai.. .. in which the coefficients c are rational integers. an] there are also n equations of the form a" = e1-)a1 + e2(V)a2 + .3n. an.. then the number of classes is (b. since each factor is a rational integer.. a2.. an] = [al..] when there are n equations of the form a" = cl al + .. since the numbers a'" are in the module a = [al... .31+.. §4. If the n independent numbers . an are also independent. .. + e. Q.. a' form another... + en")an.`) = 1 or 0 according as the indices v. a2.. a2. then we have a" = c1") al + c a2 + . C11 . If the n independent numbers al. ... a) = ±B. 1.. form the basis of a module b.. a" where the coefficients b are rational integers whose determinant B is nonzero. Conversely. and.... . and bearing in mind that the n numbers a form an irreducible system. .E..

By replacing the numbers a1. . a) = a' a2 . of the module [.. . ..+ and consequently ±a' .E. whose coefficients a are rational integers chosen so that (b....a (nn) = E fci .Q... ..... . a)... with rational integer coefficients c whose determinant fc1Jc2 ... b(n) Thus we have (b. a2i ....@ . we then have (by 3) n equations of the form av = clv)al + . . .... on. + . + c(v)b'. 5. that a has (by §4... an.. and each of the two numbers of classes.. a) = ±B.. Q. 3n we see. i3n] can be changed into a member of the module a by multiplying by a nonzero rational number B. c(n) ... a) = ±B(a... If only n among the m numbers al. + a(-))3.D. cnn) = fl. . b). . a(n) Moreover.. Auxiliary theorems from the theory of modules Now each of the numbers. This important theorem can easily be extended (and even more simply by means of the theorem below) to the more general case where the coefficients b are fractional rational numbers..3i. a(n) ±a' a2 . fbi . .. an by their expressions above in terms of the n independent numbers ail... can be deter- mined by a simple rule involving the determinant B and all its minor determinants. . forming a basis .. an likewise form a basis of the module a and since (by 2) each of these two systems of n numbers is irreducible... Qn. + cnv)an.. One then obtains the theorem (b.. since a is divisible by b and hence also equal to the least common multiple of a and b. .32..L) aVv) = C(-)b'. Qn.31. because we assume this of the system Ql. (a. 74 Chapter 1. that cnv)b(... It follows.. b) and (b. since the n numbers a1. and hence each number .3) a basis of n numbers of the form a/ = aiv)/31 + a(-))32 +. by comparison with the preceding expressions for the numbers a'' in terms of the same numbers ..

. I shall therefore assume that the m numbers aµ are represented. n numbers among the m numbers aµ which form an irreducible system. by choosing the latter to be. Conversely. at least one of whose m(m-1). rn. rn. any n of the m num- bers aA would be dependent.µ) (. r2If i . since the m numbers aµ include the n numbers w.. no matter how the numbers w are chosen.. Moreover. + C(-)wv.. The hypothesis of this theorem will evidently be satisfied whenever all the m numbers a1.. wn are dependent there is an equation.... .. a'2.. Otherwise. that the m numbers aµ can always be expressed in terms of n independent numbers wV.. .. Irreducible systems 75 of the module a are independent.. are .. since the n + 1 numbers a. am] consisting of n numbers a' of the form C2")w2 + . an are expressible in terms of n independent numbers W1. as aµ =rlµ)w1 where the system of coefficients rl.. wn are independent. wn. then a has a basis consisting of n independent numbers a'..... W2. there is a basis of the module a = [al. . . . wl. of the form x not all zero. . ate. To do this I remark first that we can evi- dently choose a positive integer k so that the mn products kr. ... and consequently aJ. and I shall show that. .. ... it follows from the hypothesis of the theorem.. Then. can be rep- resented.... in terms of the numbers w. r2 .....(m-n+1) 1 2 n n x n partial determinants R is nonzero.. since w1. in the manner indicated. a2.. .. in terms of the n independent numbers w. .. .... (m) (m) r1 . for example.. (r) rlIf. at least one of the determinants R will be nonzero. W.I I r2.. . for each index p. x0 must be nonzero. in the manner indicated. . Finally. . rn(m) consists of rational numbers. at = clv)w1 + with rational coefficients c. . .

].. b has a basis consisting of n numbers of the form av = alv>/31 + a2v). an. By substi- tuting the first expressions in the second.3n. and hence it is the least common multiple of a. + p. + avv)Nv. Then if P denotes any n x n partial determinant formed from the system of coefficients (p). + q Pv. wn = k)3n.)32 .. be- come numbers of the form x1a1 + x2a2 + ..] is divisible by the module b = [)31.7)= 1 or 0. . + xam. 76 Chapter 1.. 02.. .2) that the least common multiple of the two modules a. and n equations of the form a' where the 2mn coefficients p and q are all rational integers. when multiplied by a nonzero rational. a. according as the m indices v. Since the m numbers a... 6.. and if Q denotes any determinant formed similarly from the system of coefficients (q). it is clear that each number .31. with rational coefficients a for which the product aia2 ann is nonzero.. b. n are equal or not. Auxiliary theorems from the theory of modules integers.. v' from the series 1. . It follows from this (by §3. '2 + . .. then we know that the sum E PQ. since the n numbers f3 become the n numbers w. More- over. + q(m)p(. when multiplied by k and the latter.... + . To the preceding proof I add the following remarks.. . . If we now re-express the numbers 3v in terms of the n numbers w we can conclude that the assertion above is true.. w2 = k. . which at the same time proves the theorem.... when multiplied by a nonzero determinant R. . we deduce that the sum qp' . there are m equations of the form aµ = p1µ>a1 +p2µ) a2 + . with rational coefficients x. and express the numbers aµ in terms of the numbers On it follows that the module a = [al. form a basis for the module a just as much as the n numbers a'. and bearing in mind that the n numbers a' are independent..2. If we now put Wl = k.l3 in the module b.32. itself becomes a number in the module a..

in any case where the coefficients r are given numerically and. a2. The problem of finding all the systems (p) corresponding to a given system (r) can be solved in the most comprehensive and elegant manner by generalising a method applied by Gausst in the special case in which one utilises identities between the partial determinants.. These are based on the evident proposition that a module [al. §4. am] is not altered when we replace al. 279.enn) is nonzero. Consequently. . we get r(A) = piµ)ev +p2µ)eU + . without loss of generality. this would lead us too far away from our present position. Now since the n numbers a can likewise be represented in the form aV = e1V)wl + e2V )w2 + . the two determinants R. . . +pnµ)e. However... .. a°m = am. a3 = a3.. 236.. In practice. A system of coefficients such as (p) is evidently just a special case of the preceding coefficients (r). art... The partial determinants R° correspond- ing to all combinations of n numbers from the new basis a1 = al + xa2. + e(V)wn. One sees immediately (from 3) that one can derive from it all other systems (p) by composition with all possible systems of n2 rational integers with determinant ±1. + pn ) an are to form a basis of the module a.. a2 = a2. is equal to 1. Conversely. . by the number al + xa2i where x is any rational integer. Irreducible systems 77 taken over all combinations of n different upper indices. this property of the determinants P is necessary if the n numbers a. (p) satisfy the relation R = PE. and I am content to have shown the existence of a system such as (p). and consequently the determinants P have no common divisor. for example. and also the m numbers aµ = plµ) al + .. t Disquisitiones Arithmeticae. P corresponding to systems of coefficients (r).... that is. as integers. we arrive most promptly at the goal by a chain of elementary transformations. 234. with n2 rational coefficients e whose determinant Efele2..

. and we keep it fixed in all subsequent transformations of the basis. + a(n)wn..n = s numbers ai ... and this single coefficient is replaced by one of smaller absolute value. coincide in part with the determinants R corresponding to the old basis 0 a1 = a1 0 . treating wn_1 as we previously did wn in working with the m numbers aµ of the original basis. 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. a'n_1... except the first. wn_1. 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 + . 3 . a2 =a20 . 78 Chapter 1. . a9' which are all zero.. rn. and if Ir.1) partial determinant corresponding to an arbitrary combination of n . Auxiliary theorems from the theory of modules and to the new system of coefficients (r°). are nonzero. . . Now if two of these co- efficients... then we can choose a rational integer x such that I rn + xrn I < I rn . + an and m . Since the determinants S cannot all be zero. . or are of the form Sa(nn).. from which we easily deduce that the greatest common divisor E of the determinants R is the same as that of the determinants R°. a2 .. say rn and rn.. where S is an (n -1) x (n . If we continue these transformations then we finally obtain a basis of a consisting of n numbers ai. 0 a3 = a. w2.xa1. 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. By repetition of this procedure we necessarily arrive at a basis in which all but one of the m coefficients of wn are zero. remain the same. and which is formed from the (n . Thus the determinants R° cannot simul- taneously vanish. t The elementary transformation above therefore gives us a new basis in which all the m coefficients rnµ).1 mem- bers of the basis other than an.. since then all the determinants R would be zero..1 of the m .1)2 coefficients corresponding to wl. The partial determinants corresponding to the actual basis either vanish. a'2. am = am 0 They are of the form R° = R1 + xR2. a'' of the form av = a1v)w1 + a2W2 + .'n I > jraI. . and which can therefore be omitted.. we now proceed with these m-1 members of the basis. The n nonzero coefficients a(V) can be taken to be positive.

Q are corresponding determinants then we know that H = Q. since the quantities H are s x s determinants complementary to the determinants P... one obtains the adjoint determinant q1 .. In this way we obtain a second proof of the important theorem (5). At the same time we obtain n equations of the form a = % GA) aA µ and s equations of the form = ow)aµ = 0. In fact.. the m x m determinant p1 pn h'1 . By inversion. k(m q(m) . where K denotes the determinant complementary to Q. aF = (w) P(IA) a v. ao The latter equations are a new expression of the original supposition that only n of the m numbers a..... W. V and since the determinant of each of the substitutions or transformations is equal to 1.. and at the same time it is evident that. we can find the system of coefficients (p) as well as a system of coefficients (q). are independent. since the s numbers a" are zero. K = P. Irreducible systems 79 module. and formed from the system of coefficients (h). and their product a' a' an(n) is evidently the greatest common divisor E of all the partial determinants R. h' =>2PH=1. gnm) k(-) . by composition of the successive transformations and their inversion. and we have been able to base all this study on such a system of s equations. One can generally shorten the calculation itself by carrying out several . p(m) hsm) pim) h(m) . q'n k' ks TQK=1. one first obtains m equations of the form aµ = EP(IL)a/v + V h(µ)a" o 0 V or.. and if P.

33=42w1. 'Y3=-. Suppose for example that m=4. (R) 1 R3. 'Yi =)31 y2= 32=-17W1+w2.4 .4 = 42. /33 = a3 + 2a4 = 25w1 + W2. a2 = 7w1 + 7w2. R1. a4 = 8w1 + 2W2.-3 .. 64=-2y1+y4=0.R These determinants satisfy the identity R1. we form the fourth basis 61 = y1 = 21wi. whence. . r" =7. Auxiliary theorems from the theory of modules elementary transformations simultaneously. whence. 02 = 72. 62 = y2 = -17w1 + w2. giving r' = 21.34 = 2'y2 + 74 At this stage. /34 = a4 = 8W1 + 2W2. a2=#2+304. fir) 2= 0.3 = -63. a3 = 9w1 . 80 Chapter 1. y4=-2. conversely ai = /3i. 2- r2 . in which wi has its least coefficient 21.3 = 0. i r"'=9. R2.3a4 = -17w1 + w2. Now since the smallest nonzero coefficient of w2 is in a4 we form the new basis 01 = a1 = 21w1.33-2. and a1 = 21w1. r""=8 i .32+. r' r" 2 = 7.3w2. whence s = 2.4 = -42. R2.4 = 42.32 for example. . amongst the other three. R1. in . r"" 2 = 2.31. since 72 is the only number in which w2 has a nonzero coefficient. using the abbreviation rl r2µ) . a3=. 63=-2y1+y3=0. and since yi is the number. n=2. $i = yl.32+04=42wi.r1µ )r2/i) . conversely. a4=04- Now since w2 has 1 as its smallest nonzero coefficient. we form the third basis =21wi.3 = -84.2R3.4 + R1. 33 = y2 +'Y3.R1.3R2.4R2. $2 = a2 . We obtain the six partial determinants R1.2 = 147.

µ.3 = -3. a2 661 +762 +364 = 661 +762. q1 f . 74 = 261 + 64 = 0. q1 = 0. (k) f k2. ai.µ' respectively. and Q1. q2 . q2 = 0.4 2.2.1.µ.4 = 0. 2. h2/1 = 1.µ comple- mentary to Pµ. 73 = 261 + 63. k2' -0: k2"=7 we derive the determinants Hµ.2 = 1.3 = 0. pi/ _ -2. 72 = 62.3 = 0 Finally. and conversely 61 a1 21w1.0.4 = -3.. = Q µ and K.. hi = 0. P2. . and successive sub- stitutions give al 61 = 61. hip' = 0. a4 261 +262 +64 = 261 +262. Q1. 1 P3. = Pµ.U. a2.4 = 0.0. (h) l h2 = 0. Q2. §4. k2 =-2.µ. we have pi = 1. AT an. 63.0. (P) P1. P1. = P2. P1. for the determinants proportional to the R. (Q) Q3. q2 . hip = 1. P.1 k111 5. we have denoted by ai. 62 a2 -3a4 -17w. (p) i n-7 p2 . h2I _ -2. 64 are the quantities which. Thus we obtain. g111 . Since 61i 62. p2 m=-3 .3 = -4.2 . . and ki = _2 kl = _1. Q1. and likewise I qi = 1. conversely. in the general theory. +w2. h2 = 3. 63 = -2a1 -a2 +a3 +5a4 0. pi = 6. from the systems of coefficients f hi = 0.2= 7.4 = -2. Irreducible systems 81 whence. k1I .. 7'1 = 61. p2 . and Qµ.4 = 2. Q2. q2 3. Since 63 = 64 = 0 the transformation is completed. 64 = -2a1 -2a2 +7a4 0. a3 -261 -362 +63 -264 = -261 -362. a2.

taking the place of addition.kinikiii K1. but any objects of study. The module a becomes a group of elements. K1 2 = kiiiki" .3 1 ' and the treatment of the example is complete. but it is clear that they do not cease to be true when the Greek letters denote not only numbers. The researches in this first chapter have been expounded in a special form suited to our goal.h'1mi nnh2" .h1nh2nn H1.h1 h2.3 = h'h1111 .4 = h1 h2 . n (H){ i n n i H3.4=k'1"ki 1 2 . The rational integer coefficients indicate how many times an element contributes to the generation of another. under a commutative and uniformly invertible operation (composition). the foundation for the whole divisibility theory of these numbers. 3 produce a determinate third element ry = a + .3 = kmiku .82 Chapter 1.4 = h1 h2m> H1. .h1 h2. Auxiliary theorems from the theory of modules H1ni . H2. m i ' w ' mi nn i H2. the composites of which all belong to the same group.4 = h1h2 .kiik" K14-001-010 1 2 1 2 1 2 1 2> 2 1 21 1 k'1 k2"-kil2o (K) k' kill K34=k'k2 1 2 -k'i12 k K2.3 = h1 n ni > h2 . any two of which a. 1 2 ' K2. To conclude.a of the same type.h1 h2 . I remark that applying this to the case n = 1 leads to the fundamental theorem on the greatest common divisor of an arbitrary number of rational integers.h1inh2.2 = h'1na h2 .

the system of prime numbers occurring as factors in this product is completely determined by giving the number of times a designated prime number occurs as a 83 . where c is also a rational integer. to explain in a particular case the nature of the phenomenon that led Kummer to the creation of ideal numbers. Above all. that is. and that of the multiplication of ideals. the two units ±1 divide all numbers. The theory of divisibility considers the combination of numbers under mul- tiplication. In the latter case we can always express it as a product of prime numbers and . the sum. as indicated in the Introduction. The number a is said to be divisible by the number b when a = bc. then ±a will also be divisible by ±b. Each positive number... difference and product of any two members in this domain also belong to the domain. That is. subtraction and multiplication. The number 0 is divisible by any number. ±2. it should be recalled that these numbers are closed under addition.. § 5.. is either a prime number.which is the most important thing . 2 Germ of the theory of ideals In this chapter I propose.in only one way. and I shall use the same example to explain the concept of ideal introduced by myself. If a is divisible by b. ±1. The rational integers The theory of numbers is at first concerned exclusively with the system of rational integers 0. or else a composite number. and they are the only numbers that enjoy this property.±3. different from unity. a number divisible only by itself and unity. and consequently we can restrict ourselves to the consideration of positive numbers. that is. and it will be worthwhile to re- call in a few words the important laws that govern this domain.

since the definition of congruence in Chapter 1 contains that of Gauss as a special case.t This procedure. 1. This also follows from the studies of the preceding chapter. The simplest way to prove these fundamental propositions of number theory is based on the algorithm taught by Euclid. (Theoria residuorum biquadraticor-am. where q and r are also integers and r is less than m. The notion of congruence of numbers was introduced by Gauss. for example. and likewise the system m of all numbers divisible by m is identical with [m]. m) = ±m. the Vorlesungen fiber Zahlentheorie of Dirichlet. a root of the irreducible quadratic equation i2 + 1 = 0. written z . and consequently we can define divisibility for these numbers in the same way as for rational t See. that is. It is for this reason that the procedure always halts after a finite number of divisions.t Two numbers z. art.z' (mod m).) . 84 Chapter 2. 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. art. In the contrary case z and z' are called incongruent modulo m. as we know. If we arrange the numbers in classes. when he transported them to the domain of complex integers x + yi. This property depends essentially on the theorem that a prime divides a product of two factors only when it divides one of the factors. Thus (by §3. The system o of all rational integers is identical with the finitely generated module [1]. t Disquisitiones Arithmeticae. any number z can be expressed in the form qm + r. The complex integers of Gauss The first and greatest step in the generalisation of these notions was made by Gauss. 42. z' are called congruent modulo the modulus m.4) the number of classes is (o. is based on repeated application of the theorem that.2 or §4.z' is divisible by m. when the difference z . Germ of the theory of ideals factor. The congruence of two numbers modulo m coincides with congruence modulo the system in. for a positive number m. where x and y are any rational integers and i is. in his second memoir on biquadratic residues. which serves to find the greatest common divisor of two numbers. subtraction and multiplication. II. §6. § The word class seems to have been employed by Gauss first a propos of complex numbers. with two numbers in the same class§ only if they are congruent modulo m. The numbers in this domain are closed under addition.

q) < 1/2. four numbers which divide all numbers. There are four units. If now we let z and m be any complex integers. The complex integers of Gauss 85 numbers. exactly as for rational numbers. so called when it is the product of two factors. §6. 24). and it is also clear that for any given w we can choose a complex integer q such that N(w . ±qi are regarded as representatives of the same prime number q. and the proofs of the general laws of divisibility for rational integers can be applied word for word in the domain of complex integers. then the norm of a product will be equal to the product of the norms of the factors. The set of all prime numbers q in the domain of complex integers consists of: 1. z' are called congruent modulo m. and such a number cannot divide a product unless it divides at least one of the factors.¶ that the general propositions on the composition of numbers from primes continue to hold in this new domain. and whose norm is consequently 1.vi. and which likewise implies that theorem. written I Recherches sur les formes quadratiques a coefficients et a indeterminees complexes (Crelle's Journal. it follows by taking w = z/m that we can put z = qm+r where q and r are complex integers such that N(r) < N(m). to be the product u2 + v2 of the two conjugate numbers u + vi and u . which follows immediately from the celebrated theorem of Fermat on the equation p = a2 + b2. where u and v are any rational numbers. 2. with marvellous ease. The existence of the primes a ± bi just mentioned. with m nonzero. . If we define the norm N(w) of a number w = u+vi. 3. The factors a + bi and a . provided of course the four associated primes ±q. One can establish very simply. dividing the rational prime 2 = (1 + i)(1 . or else it is a prime. Every composite number can be expressed uniquely as a product of prime numbers.i) _ -i(1 + i)2. ±i. Numbers z.bi of each rational prime p of the form 4n + 1 with norm a2 + b2 = p. Every other nonzero number is either a composite number. neither of which is a unit. as Dirichlet showed in a very elegant manner. Congruence of complex integers modulo a given number m of the same kind can also be defined in absolutely the same way as in the theory of rational numbers. as a result of the following remark. All the rational prime numbers (taken positively) of the form 4n+3. The number 1 + i. It is a splendid example of the extraordinary power of the principles we have reached through generalisation of the notion of integer. We can then find a greatest common divisor of any two complex integers by a finite number of divisions. ±1. that is. can now be de- rived without the help of the theorem.

If we arrange the numbers into classes. Here we encounter the phenomenon which suggested to Kummer the creation of ideal numbers. and x. -b a §7. i] and likewise the system m of all the numbers m(x + yi) divisible by m forms the module [m. This follows very easily from the researches of the first chapter. 82-3=0. whose basis is related to that of o by two equations of the form Consequently we have (§4. and let x. In each of these domains it is easy to see that one can find the greatest common divisor of two numbers by a finite number of divisions. since the system o of all complex integers x + yi forms a finitely generated module [1. 92+2=0. when z . On the other hand. 92-2=0. this method is not applicable to the domain o of integers w=x+y9 where 0 is a root of the equation 92+5=0. For example. and which we shall now describe in detail by means of examples. then the total number of different classes will be finite and equal to N(m). even though there happen to be an infinite number of units in the last two examples. Then the numbers x + y9 form a corresponding numerical domain. The domain o of numbers x + y/ There are still other numerical domains which can be treated in abso- lutely the same manner. with numbers being in the same class or not according as they are congruent or incongruent modulo m.4) a b (o. y again take all rational integer values. 02+0+2=0.z' is divisible by m. m) = = N(m). 86 Chapter 2. mi]. so that one immediately has general laws of divisibility agreeing with those for rational numbers. . Germ of the theory of ideals z = z' (mod m). y be any rational integers. let 0 be any root of any of the five equations 82+9+1=0.

p must be representable by the binary quadratic form x2 + 5y2. b2=-2-0. c1=2+30. And if p is a particular nonzero number we conclude. The domain o of numbers x + y 87 The numbers w of the domain o we shall now be concerned with are closed under addition. and hence of the form ww'. f2=-1-20. d2=1-0. In the contrary case the number is called indecomposable. if we define the norm N(w) of a number w = x + yO to be the product x2+5y2 of the two conjugate numbers x±yO. A number (different from zero and ±1) is called decomposable when it is the product of two factors. then we must have N(p) = 1 and therefore p = ±1. If p is a unit.t or else by a small number of direct trials. f See Dirichlet's Vorlesungen fiber Zahlentheorie. b2 = b1b2. c2=2-30. subtraction and multiplication. similarly. whose norms are products of two of these three primes. d1=1+0. it is necessary that N(p) = p2 = N(w)N(w'). 3. that is. Also. then the norm of a product will be equal to the product of the norms of the factors. despite the indecomposability of these fifteen numbers. . 7 cannot be represented in this way. c=7. and since w. the same number is susceptible to several. as one sees from the theory of these forms. essentially different. for the other twelve numbers. namely. 91=4+0. e1=3+0. a=2. there are numerous relations between their products. in infinitely many cases an entirely new phenomenon presents itself here. §71. for a rational prime p to be decomposable. that N(µ) expresses how many mutually incongruent numbers there are modulo p. They are therefore indecomposable. It is easy to convince oneself that each of the following fifteen numbers is indecomposable. The simplest examples are the following. It is easy to show the same thing. b1=-2+0. § 7. just as before. representations of this kind. Then it follows from the theorem on the norm that each decomposable number can be expressed as the product of a finite number of indecomposable factors. fi=-1+20. which can all be deduced from the following: (1) ab = d1d2. However. But the three prime numbers 2. b=3. w' are not units we must have p = N(w) = N(w'). 92=4-0. and hence divides all numbers. In fact. neither of which is a unit. abi = di. However. and we therefore define the notions of divisibility and congruence of numbers exactly as before. e2=3-0.

Such an indecomposable number therefore does not possess the property which. a = a' and hence the fifteen numbers can be expressed as follows.88 Chapter 2.al. Then abl = µµ1a2(31 is not a square unless IA = µl. If we imagine for a moment that the fifteen preceding numbers are rational integers then. bh =u. b = µ/31. (Transla- tor's note.abl = p a tl 22 dl =path. bh = µ1t1.Qi. (3) be = fif2 = glg2. Germ of the theory of ideals (2) ac = ele2. aci = ei. in terms of five numbers a. c = 127172. by virtue of the equation µa2 = µ'a'2. agl = d1e2 In each of these ten relations. Thus in fact µ = µl = p2 and hence a=µa2. where µ. dl = µarn. d2 = µa/32. b2 = p202. Note first that abi = d2 and b1 b2 = b2 are both squares. e2 = µ'a'y2. b1b2 is not a square unless Al = µ2.) . the same number is represented in two or three different ways as a product of indecomposable numbers. by the general laws of divisibility. in the theory of rational numbers. we get 2 dl = 2 . is characteristic of a prime number. Similarly. that the four numbers fi.a2. e1 = µ'a'-yl. since this implies p = p' = 1. I include the following proof of the consequences of (1) as an example. /32. b2 102. and from the relations (2) that there are decompositions of the form a =µ'a2. And it follows im- mediately. gi. b2 = bhb2 = p2t1t2 b = ptlt2. which completes the proof of the decompositions claimed by Dedekind. f2. Thus one sees that an indecomposable number may very well divide a product without dividing any of its factors. Al. a. y2: t Since these decompositions do not seem obvious to me. c2 = clc2. 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. d2 = ab2 =µ2a212 d2 = µa(32. 92 are squarefree. yl. c1 = µ''yi. where all the Greek letters denote rational integers. /31. c2 = /U' Y2. b2 = µ(32 Forming products of these. bl = /t. we easily deduce from the relations (1) that there are decompositions of the formt a = µa2.fi = dlel. Suppose that a = pat.

el = aryl. other than unity. one can recognise the essential constitution of a number without effecting its decomposition into prime factors. in the manner indicated above. then we obtain a number w = x + yO not divisible by 2. of the number 2 in our domain o of numbers w = x + y9.32. in all questions of divisibility in the domain o. Role of the number 2 in the domain o Let me begin by remarking that. then we can conclude with certainty that a is the square of a prime number. b=. 'Y2 In a moment I shall explain in detail what these relations between numbers mean. 'Y1. for example. in this sense. whose square is divisible by a. c=yi'Y2. and whose square is divisible by 2. If we know. (4) bl = Ql . Having regard to the preceding remarks on rational numbers. w'2. a prime or the square of a prime. exactly as if they were composed. if the number 2 is to divide the product w2wj2. Role of the number 2 in the domain o 89 a=a2.N(w) (mod 2). Thus if we can ascertain that both these two properties hold for a. we then say that the number 2 behaves in our domain o as though it were the square of a prime number a. Nl. Moreover. N(w'). Now. that a positive number a does not divide a product of two squares unless at least one of the squares is divisible by a. c2 = Ii dl = a/31. and hence also for at least one of the two squares w2. d2 = a. f2 = /3272. We shall now examine the behaviour. /32. . the remarkable thing is that they behave. It is likewise certain that a number a must contain at least one square factor. of five different prime numbers a. when we can prove the existence of a number not divisible by a. y to be any two odd numbers. §8.31/32. and hence also w2w'2 = N(w)N(w') (mod 2). Since any two conjugate numbers are congruent modulo 2 we have w2 . then we can conclude with certainty that a is either 1. observing only how it behaves as a divisor.b2 = /2 i cl = 'Yl . if we take x. to be divisible by 2. e2 = ar'2i A =0171. it is necessary for at least one of these norms. in the theory of rational integers. 91 = /3172. 92 = 327'1 Now even though our fifteen numbers are in reality indecomposable. §8. and hence also the product of the two rational numbers N(w).

trying analogously to determine the role of the number 2 in the domain of numbers x + y/ leads to complete failure. and consequently (a) x = y (mod 2). It follows first. that is. without bringing in a. If both are odd. and how many times. that a number w = x + y9 is divisible by a if and only if N(w) is an even number. because it leads to a mode of expression in perfect harmony with the laws of the theory of rational numbers. since. Now. We now have to find.l+y (mod 2). and consequently x . From this we get the theorem expressing the character of the ideal number a as a prime number: "The product of numbers not divisible by a is also not divisible by a. the highest power a"° of a that divides w. since in fact Kummer managed in similar circumstances with great success by taking such a number a to be an ideal number. Later we shall clearly see the reason for this phenomenon. when a rational number a is known to be the square of a prime a we can easily judge. f Luckily. Germ of the theory of ideals Although such a prime number a does not actually exist in the domain o. The number w is not divisible by a when N(w) is an odd number. . for example. and at least one of the two rational integers x1. and we are allowing ourselves to be guided by analogy with the theory of rational numbers to define the presence of the number a in terms of existing numbers w of the domain o. yl will be odd. It is clear that z is divisible by an if and only if z2 is divisible by a''2." As far as higher powers of a are concerned. for n = 1. We have w = 28w1 = 28(x1 + y19). Thus we extend the criterion to the case we are interested in by saying that a number w of the domain o is divisible by the nth power an of the ideal prime number a when w2 is divisible by 2''2. Experience will show that this definition is very luckilyt chosen. Let s be the exponent of the highest power of 2 that divides w itself. it is by no means necessary to introduce it. whether a is a factor of an arbitrary rational integer z. w1 will be divisible by a and we shall have W1 = xl . we first conclude from the definition that a number w divisible by an is also divisible by all lower powers of a.5y1 + 2x1y19 = 2w2. 90 Chapter 2. when w is nonzero. the highest power of 2 that divides w2. because a number w2 divisible by 2n is also divisible by all lower powers of 2.

then wl and consequently wi will not be divisible by a. We see at the same time that m is also the exponent of the highest power of 2 that divides the norm N(w). if the exponent denoted above by s is < n. §9. We therefore have the theorem: "The exponent of the highest power of a that divides a product is equal to the sum of the exponents of the highest powers of a that divide the factors". . y2 according as it satisfies the corresponding congruences zd2 = 0. Role of the numbers 3 and 7 in the domain o When all the quantities appearing in the equations (4) of §7 are rational integers. It is likewise evident that each number w divisible by a2n is also divisible by 2n since. yl is even. then it is evident that a rational integer z will be divisible by th. yl. These congruences have the peculiarity that they do not involve the numbers . but in both cases w2 = 2''nw' where w' is a number not divisible by a. 2s + 1 and hence also m will be < 2n. Role of the numbers 3 and 7 in the domain o 91 where w2 = x2 + Y20 is not divisible by a. Wed say that a number . 02. and if at the same time a is prime to b and c.Q2. §9. each number divisible by 2n is also divisible by a2n. yl. conversely. so that the other is odd. but not by a2. But if one of the two numbers x1. /32i 'yi. zee = 0.31. then the numbers 2s. zdl = 0 (mod b). we easily see.31. 'Y2 themselves. Since the number 1 + 0 is divisible by a. because x2 is even and y2 is odd. Thus in the first case m = 2s + 1. y2 in the context of numbers in the domain o. It follows immediately from the definition that. contrary to hypothesis. with the help of the preceding theorem. zel = 0 (mod c). and it is for precisely this reason that they are appropriate for introducing the four ideal numbers. which has the advantage of containing the number w only to the first power. that the congruence w2 = 0 (mod 2n) that serves to define divisibility of the number w by an can be replaced by the congruence (an ) w(1 + 9)n = 0 (mod 2n). in the second case m = 2s.

92 Chapter 2.y (mod 3). (/32) x -y (mod 3). w"=ww'=x"+y"9. the number 3 ought to be con- sidered. since each of the ideal prime numbers /31. w'=x'+y'9.132 is also divisible by 3. Moreover. (1 + 9)w 0 (mod 3).3 = (1 + 0) (1 . we ought to regard 3 as the least common multiple of the ideal numbers i31 i 02. so that it is legitimate to say that the four ideal numbers are all different. by analogy with the theory of rational numbers. Multiplication converts these congruences to the following: (/31) x . w'. and since conversely every number divisible by 3 is also divisible by N1 and N2. it does not divide a product ww' unless it divides at least one of the factors w. 3 . (3 .9 by a and /32. Because of this. that is.-y . 1 . and hence x ± y" = (x ± y) (x ± y') (mod 3). since x =. But each of these two ideal numbers also has the character of a prime number.9.3y (mod 7).9)w 0. and the number in question does not satisfy any of the other three. then in view of the fact that 2 behaves like the square of a and 1 + 9 is divisible by a and /31i and 1 .y . /32.9)w 0. 3 + 9. from a certain point of view.0 (mod 3) in that case. then we have x"=xx' -5yy. Each of these conditions can be satisfied by one of the numbers w = 1 + 9.9) that 1 + 0 behaves like . as the product of the two different ideal prime numbers /31. which immediately justifies our assertion.9. concerning which we add the following remarks. In fact if we put w=x+y9. bearing in mind the congru- ences (01). we ought to conclude from the equation 2. (y1) x . y"=xy' + yx'. (72) x . Germ of the theory of ideals w = x + y9 is divisible by one of these four numbers if w is a root of the corresponding congruence (1.-3y (mod 7). (/32) above. since every number w divisible by Q1 and . /32 is different (in the sense indicated above) from the ideal prime number a introduced above. (3 + 9)w 0 (mod 7). Moreover.

Each positive rational prime . because we have y = x + 6y' and consequently x+y0= (1+0)(x+5y'+y'0). Laws of divisibility in the domain o 93 the product of a and . rye in the same way. Conversely. which really exist and have the character of primes. Laws of divisibility in the domain o By similar study of the whole domain o of numbers w = x + y0 we find the following results: 1. §10. sat- isfying the two preceding congruences. that is. 9 (mod 20) can be decomposed into two different factors.y (mod 2).0 (mod 3'' ).0 like the product of a and /32. 13. x =. is also divisible by 1 + 0.11. and this yields a series of theorems which agree perfectly with those of the theory of rational numbers. By analogy with the theory of rational numbers. Q2 ) w(1 + 0)n . In fact this presumption is plainly confirmed: each number w = x + y0 divisible by 1 + 0 is in fact divisible by a and 31. 3. All the positive rational primes .131. as we have done above for powers of the ideal number a. . §10.(31 i and 1.17. We treat the ideal prime numbers y1. The number 2 behaves like the square of an ideal prime number a. . 2. We can now also introduce the powers of the ideal prime 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). because x+y0=(1+0)(x'+y'0) implies x=x'-5y'.y (mod 3).19 (mod 20) behave like actual prime numbers. The number 0 with square -5 has the character of a prime number. and consequently x =. y=x'+y'. each number w = x + y0 divisible by a and '31.1.132.

the greatest circumspection is necessary to avoid being led to premature conclusions.3.t it is necessary. 5. by the introduction of ideal numbers. 02 = -2 . for example (and if I am not mistaken. it has seemed desirable to replace the ideal number of Kummer. to arrive at this result and to become completely certain that the general laws of divisibility governing the domain of rational numbers extend to our domain o with the help of the ideal numbers we have introduced. the notion of product of arbitrary factors. In particular. not from the domain o. . and this can be done in several ways. given above. even supposing knowledge of the theory of quadratic residues and binary quadratic forms (a theory which. Germ of the theory of ideals 4. of the irregular role of the number 2 in the domain of numbers x + y suffices to dispel this illusion. can be derived with great facility from the general theory of algebraic integers). /32 = -2 . as we have remarked in the Introduction. cannot be exactly defined without going into minute detail. or else it behaves in all questions of divisibility as a unique product of actual or ideal prime factors. by a noun for something which actually exists.32. but only as a divisor of actual numbers w in the domain o. Each positive rational prime . however.3. which is never defined in its own right. We can indeed reach the proposed goal with all rigour. Because of these difficulties. 9/31 = -201 . t To some people it seems evident a priori that the establishment of this harmony with the theory of rational numbers can be imposed.0. conversely. 902 = 301 + 202. However. /31/32 = 3. but rather adjoined to this domain in the sense of Galois.9. if we put /31 = -2 + 0. However the example. actual or ideal. whatever happens. this is the way chosen by Kronecker in his researches). replace the ideal numbers by actual algebraic numbers. Each actual number w different from zero and ±1 is either one of the numbers mentioned above as having the character of a prime. Indeed. as we shall soon see when we attempt a rigorous derivation. /31 = -2 + 9. whence it follows that the quadrinomial numbers x+YO +z1/31+z2/32. One can.94 Chapter 2. to make a very deep investigation.7 (mod 20) behaves like the product of two different ideal prime numbers. and if we choose the square roots so that /3102 = 3 then we have 92 = -5.

31 + 2. Likewise. -ti. and consequently a= [2. . c2 denote the systems of all numbers w divis- ible by . because one is forced to pass from the given domain o to a more complicated domain o'. by §8. If we now let m denote any one of these five systems. that is. rye respectively. Thus to obtain the system a of all numbers w divisible by a we put x = y + 2z.3-9]. The system a therefore consists of all numbers of the form 2z+(1+9)y. it does not seem to me as simple as desirable. §11. are closed under addition.32 of the domain o'. For example. sub- traction and multiplication. Ideals in the domain o The condition for a number w = x + y9 to be divisible by the ideal prime number a is that x = y (mod 2). In the Introduction I have explained in detail the train of thought that led me to build this theory on quite a different basis. The domain o' of these numbers contains the domain o.31+/32. where y and z are arbitrary rational integers.31 . a is a finitely generated module with basis consisting of the two independent numbers 2 and 1 + 0. b2. we conclude from the corresponding congruences in §9 that b1=[3. namely on the notion of ideal. the two ideal prime factors of the number 23 in the domain o are replaced by the two actual numbers 2. and it would be superfluous to come back to it here. 7'1 =2. hence I shall confine myself to illustrating the notion by an example. Similarly letting bi. then m enjoys the following properties. z2 are rational integers.1+0].32 all the equations (4) of §7 are satisfied. 7'2 =)31+2. c2 = [7.1-01. ci = [7. It is also easy to see that the choice of the new domain o' is highly arbitrary.1 + 9]. and all the ideal numbers needed for the latter can be replaced by actual numbers of the new domain o'.31i Q2. b2=[3.3+8]. ci. y. and it is the same for all the ideal numbers of the domain o. Although this way is capable of leading to our goal. by putting a=. §11.31+/32./32 and -. Ideals in the domain o 95 where x. z1.

are that m divide k and 1 and that the rational integers a. Each product of a number in the system m by a number in the system o is a number in the system m. m are rational integers. Thus all the numbers in the module o = [1.0 (mod a). The necessary and sufficient conditions for this. as one sees without difficulty.2) that m is a finitely generated module. For the system a this follows from the two equations 20=-1. is evident. 9µ' belong to the same system. µ' it evidently suffices to show that the two products Op. and we begin by posing the problem of finding the general form of all ideals. v are two given numbers in the domain o. II. and it is just the same for the other systems. the norm N(µ) = µµ'. .2+2(1+0). characteristic of each module. when multiplied by the nonzero rational number N(µ). Excluding the singular case where m consists of the single number zero. which is at the same time a multiple of o. 1. But it is a consequence of (§3. The sum and difference of any two numbers in m are also in m. To es- tablish the second property for a system m whose basis consists of the numbers p. Since m already has property I. m(b + B)] also satisfy the congruence b2 -5 (mod a). of the form [k. b appearing in the expression m = [ma. which says that the two products k9 and (l + mG)9 belong to the system m. the question is what follows from property II. But these two properties can also be established without these verifications. where a. If we replace b by any number b (mod a) then the ideal m is unchanged. and hence the product 9N(i) also belongs to the ideal m by virtue of II. Then if u' denotes the conjugate number. become numbers in the module in. 0]. among which k and m can be chosen positive. l +m9] where k.2+(1+0). as a module. by appealing to the fact that each of the five systems m is the set of all numbers w in the domain o satisfying a congruence of the form vw . We now call any system m of numbers in domain o enjoying properties I and II an ideal. we choose an arbitrary nonzero number it in in. Germ of the theory of ideals I. (1+6)6=-3. 96 Chapter 2. The first property.

o(2 ± 39) = [49. or to any of the five domains o considered in §7. since (b + 9) can also be replaced by -(b + 0).1) that two num- bers w.::F2 + 0].M) = m2a of classes into which the domain o is partitioned modulo the module m (§4.20].w' is a number in in. Let p be any number in the domain o.w' (mod p) and we shall have N(m) = N(p).±1+0]. or to the complex integers of Gauss. m(-b + 0)]. o(-2 ± 0) = [9. vb-u(mod a).70]. .w' (mod m) when w . property II is already contained in property I. m1 may be called conjugate ideals. o(-1 ± 20) = [21. The set of all numbers conjugate to the numbers in an ideal m is evidently also an ideal ml = [ma. If m is a principal ideal op then the preceding congruence will be equivalent to w . The norm of any number m(ax+(b+9)y) in the ideal m = [ma. moreover. m(b + 9)]. The system [p. Two such ideals in. b1i b2.±3+9].4). w' are congruent modulo the ideal in. we say (following §2. o(7) = [7. m is the greatest rational integer that divides p = m(u + v9) and we have.±4 + 0]. o(4 ± 0) = [21. Ideals in the domain o 97 The five ideals above. It is also evident that. then one easily sees that every ideal is a principal ideal. o(3±9] = [14. p9] of all numbers divisible by p forms an ideal which we call a principal ideal. m(b+0)] f If we extend the definition of ideal to the domain o of rational integers.t and which we denote by o(p) or op. in the domain of rational integers. for example. o(±1) = o = [1. Since all ideals are also modules. a. c2 are evidently of this form. a= N(2). The norm N(m) of the ideal m = [ma. It is easy to give it the above form [ma. ±17 + 9]. c1.30]. §11. and o(2) = [2. Thus we find. o(3) = [3. 9]. ±10 + 01. and put w . o(l±0) = [6. m(b + 0)] is the number (0.

2). We evidently have om = in. m(b + 9)] are the three congruences m"a-m"a'-m"(b"-b)-0 (mod ma). however. It follows that a principal ideal op" is divisible by a principal ideal op if and only if the number µ" is divisible by the number p. and if we always understand the determinant of the form to be the number B2 . (mm')n = m(m'n). Thus the theory of divisibility of numbers is contained in the theory of divisibility of ideals. according to the definition of Gauss. I shall be obliged to leave certain easy calculations to the reader. For the sake of brevity. m' and denoted by mm'. t The same definition also applies for multiplication of two modules. m' _ [m'a'. One sees immediately that the necessary and sufficient conditions for the ideal m" = [m"a". t See Dirichlet's Vorlesungen fiber Zahlentheorie. m(b + 9)]. m'(b' + 9)]. whence it follows that products of any number of ideals satisfy the same theorems as products of numbers. m"(b" + 8)] to be divisible by the ideal m = [ma.$ Moreover. where B is odd. we derive their product m" = mm' = [mam"(b" + 0)] t The general theory of forms is nevertheless simplified a little when we also admit the forms Ax2 + Bxy + Cy2. Divisibility and multiplication of ideals in o I shall now show how the theory of numbers w = x + y9 in the domain m can be based on the notion of ideal. .4AC. we say that an ideal m" is divisible by an ideal m when all numbers in m" belong to in. 98 Chapter 2. it is clear that the product of two principal ideals op and oµ' is the principal ideal o(µµ'). and µ' through the numbers in the ideal m'. Now given two ideals m = [ma. The definition of multiplication of ideals is the following: if p runs through the numbers in the ideal in. §2. Just as in the theory of modules (§1.$ §12. isb2-ac=-5. mm' = m'm. then all the products µp' and their sums form an ideal m" called the productt of the factors in. Germ of the theory of ideals is equal to the product of N(m) = m2a with the binary quadratic form axe+2bxy+cy2 whose determinant.

8+29]. mm'a(b' + 9). completely analogous to the hypothetical equations (4) of §7: 0(2) = a2. of which only two are independent. o(4 .29) = b2c2. Thus for the ideals considered above..9) = ac2i o(-1 + 29) = b1c1. This module is derived from the one considered at the end of the first chapter (§4. we find the product b1c2 = [21.6). o(-2+0) = bi. m.9) = b2c1.3-0]. mm'a'(b + 9) = p"m"a" + q"m"(b" + 9). 5 and 6). Divisibility and multiplication of ideals in o 99 with the help of the methods indicated in the first chapter (§4. In the same way we obtain all the following results. One arrives at this (by virtue of §4) via four equations of the form mm'aa' = pm"a" + qm"(b" + 0). mm'a(b' + 9) = p'm"a' + q m"(b" + 9). . o(-2 . and by setting w1 = 1. m"(b" + 9). o(4 + 9) = blc2. where p.-17+9] _ [21.4+9] = o(4+9). mm'a' (b + 9). It is clear from the definition that the product mm' is a finitely generated module with basis consisting of the four products mm'aa'. for example b1=[3. o(3 . o(1 + 9) = ab1.9) = b2.. m' in general it is necessary to transform the basis of the four numbers above into one consisting of only two numbers m"a".1+9}. q" denote eight rational integers chosen so that the six . w2 = 9 we conclude b1c2 = [21.9) = ab2. mm'[bb'-5+(b+b')9]=p"m"a"+q"'m"(b"+ 0). o(1 . § 12. . mm'(b + 9)(b' + 9) = mm'[bb' . o(-1 . C2=[7.9-39. o(2+30) = ci. o(3 + 9) = act. p'.7+79. To effect the multiplication of two ideals.o(3) = bib2.5 + (b + b')9]. o(7) = c1c2. o(2-30) = c2.

It follows.-5 (mod a"). c") is composed from the two forms (a. q". c. But.q'b. art. qlb" . P=pq-qp'.a'c' = -5. b + b' = pq"i We have as m" = pmm' a' = 2 = 4q and b" is determined by the congruences qb" = q'b'. that is b"b" . a' = pq". we now conclude without difficulty that these six determinants are respectively proportional to the six numbers a.q p" have no common divisor.b. since q = 0 and q'. T = pig" . U = pogo. c) and (a'. a" yield mj2a" = and hence the theorem N(mm') = N(m)N(m'). b' + b. q"ibii = bb' . Germ of the theory of ideals determinants formed from them. q"' can have no common divisor. m' as follows.qp". that we can determine the product m" = mm' of two given factors m. since these six numbers admit no common divisor. b". c'). The values of m". 242. c'. where c and c' are determined by the equations bb . b' . t This will not always be so in the domain of numbers x + yvl--3.ac = b'b' . . where c" is a rational integer and. S = p'q" . Q = pq" . a'.5 P (mod a"). to use a terminology employed by Gauss.q'p"'. each of which decomposes into two. R=pq"-qp'". At the same time we have b"b" .a"c" = -5. t Disquisitiones Arithmeticae.t they must precisely coincide with the six determinants. .t the binary quadratic form (a". b. 100 Chapter 2. 235.q'p"'. From the four preceding equations. b'. Let p be the greatest common (positive) divisor of the three given numbers a = pq'.

and hence in this case r will be neither o nor p. §12. if the ideal m" _ [m"a". Divisibility and multiplication of ideals in o 101 It is also necessary to note the special case where m' is the ideal m1 conjugate to m. m(b+ 0)] then there is exactly one ideal m' such that mm' = m". then it follows from the three congruences estab- lished at the beginning of this § that m"' is divisible by N(m) = m2 a. different from o and divisible by no ideals other than o and p. At the same time it follows that the equation mm' = mm always implies m' = m"'. actual and ideal. all the products pp' whose factors belong respectively to m. because all numbers in p are also roots of this congruence. Now to arrive at the conclusion of this theory. enumerated in §10. and they correspond exactly to the prime numbers. Thus we have established the following theorem: t "A product rlp of two numbers rl. by property II of ideals. p is not in a prime ideal p unless at least one of the factors is in p. = oN(m). because it has properties I and II. Combining this with the preceding theorem that mm1 = o(m2a). That is. if p is a prime ideal. If the given number rl is not in p then the number 1. Thus. If 77 is a particular number. satisfies the condition mm' = m". if we let ml denote the ideal conjugate to m and form the product mlm" _ IM ///a/. The product mm' is divisible by both m and m' since. will not be a root of the congruence. and it alone. . m"(b" + 0)) is divisible by the ideal m = [m. In fact. then the system r of all roots p of the congruence rlp . where m' is an integer. m'(b'+0)]. This ideal r is a divisor of p. will be called a prime ideal. Conversely." And this immediately yields the theorem: t This theorem leads easily to the determination of all prime ideals contained in o. we easily conclude that the ideal m' = [m'a'. the same then holds for the product ideal itself. then r must be o or p. m' are in both these ideals. The preceding formulas then yield the immediate result mm. m"'(b' + 0)] by the preceding rules. in o. and hence that m"' = m2am'. it only remains to introduce the following notion. The two notions of divisibility and multiplication of ideals are now con- nected in the following manner. all the roots p must be in p. An ideal p.0 (mod p) forms an ideal.

§8. then mm' includes the number pp' not in p. then we shall run into great difficulties. If m = op and m" = op" are principal ideals then the same criterion also decides the divisibility of the number p" by the number A. y are rational integers is not a domain of this nature. Such is the goal I shall pursue in the chapters of this memoir that follow. It is preferable. Germ of the theory of ideals "If neither of the ideals m. as in the modern theory of functions. If we want to treat fields 92 of arbitrary degree in the same way. it forms only a part of the domain o of all the numbers x + yp. All this theory can be applied almost word for word to any domain o consisting of all the integers of a quadratic field St. of numbers x + y/ where x. m' is divisible by the prime ideal p then their product will also not be divisible by p. mentioned above. to seek proofs based immediately on fundamental characteristics. m' respectively include numbers p. t See Dirichlet's Vorlesungen fiber Zahlentheorie. in fact. it is not at all what I propose to carry out." It follows immediately from this theorem that an ideal m" is divisible by an ideal m if and only if all the powers of prime ideals that divide m also divide m". where p is a root of the equation p2 + p + 1 = 0. Even if there were such a theory. . However. One notices. Thus the theory of divisibility of numbers in the domain o is restored to firm and simple laws. by exactly the same reasoningt as in the theory of rational numbers. even though this approach to the theory leaves nothing to be desired in the way of rigour. m(b + B)] and on the effective realisation of multiplication. when the notion of integer is defined as in the Introduction. rather than on calculation. p' not in p. 102 Chapter 2. on a calculus which coincides with the composition of binary quadratic forms given by Gauss. that the proofs of the most important propositions depend upon the representation of an ideal by the expression [ma. based on calculation. in my opinion." Because if m. Combining the theorem just proved with the preceding theorems on the connection between divisibility and multiplication of ideals. at the following theorem: "Each ideal different from o is either a prime ideal or else uniquely expressible as a product of finitely many prime ideals. and bearing in mind that o is the only ideal with norm 1. it still would not be of the highest degree of perfection.4 However. and indeed to construct the theory in such a way that it is able to predict the results of calculation (for example the composition of decomposable forms of all degrees). perhaps insurmountable ones. we arrive. t The domain. that is.

Proof. in which all the coefficients p. 3 General properties of algebraic integers In this chapter we first consider the domain of all algebraic integers. it will also be one of the rational integers 0. numbers from the sequence 0. §13..+an_10+an =0 of finite degree n with rational coefficients al.. Then we introduce the notion of a field SZ of finite degree. there are two equations of the form O(a) = as + piaa-1 + . gl.3b + . difference and product of any two integers a. if a rational number 0 is at the same time an algebraic integer then. the sum.... ... then 0 is called an algebraic integer. that is. By hypothesis. and determine the constitution of the domain o of integers of the field Q.. We now put ab = n 103 . The integers are closed under addition. a2i . conversely. It is clear that the rational integers are also algebraic integers and. If the coefficients of this equation are rational integers. an_1 i an... ±2.. + pa-la + pa.. or simply an integer..ab-1 + V)(3) _ .. subtraction and multipli- cation. 3 are also integers. ±1.. ±2. by virtue of a known theorem. The domain of all algebraic integers A real or complex number 0 is called algebraic when it satisfies an equa- tion en+a1en-1+a2en-2+. that is. + qb-1Q + qb = 0. = 0. ±1.. From the definition of integers we easily derive the following propositions: 1. q are rational integers.

.3..w .. k =0 k(n) k(n) k(2n) . wwn = k(n')w1 + k(n)w2 + . and consequently each of the three numbers a +. .. a/3 then we easily see that each of the n products ww1i ww2. + knwn... wwn can be reduced immediately...../3. (36-1 If w now represents one of the three numbers a +.. Now by elimination of the n numbers w1..a a-1 and one of the b numbers P. Each root w of an equation of the form F(w) =w'n+aw" +Qwm-2+...D. wn. k2.a2 . is an integer. General properties of algebraic integers and let w1. -w which is evidently of the form wn + elwn-1 + . WW2 = ki w1 + k2w2 + .+E=0. all coefficients k of which are rational integers.. . .. . .w k2 .. k'n k1 k. w2. Thus w.. we derive the equation ki . kn are rational integers. + klwn.. where k1. 2. .. where the n coefficients e are formed from the numbers k by addition.. a$. which include the nonzero number 1. and hence are rational integers. ... + k(n)wn. subtraction and multiplication. a . a .. 1..a.+knwn.. with the help of the equations O(a) = 0.....0. Q.. to the form kiwi +k2w2+. 104 Chapter 3.wn denote the n products as /36' formed from one of the a numbers 1.. 0(/3) = 0. We then have n equations of the form wwi = kiwi + k2w2 + . + en-1w + en = 0.w2.E....Q2..3..

. where k1.. WW2... . . .... or that a is a factor or divisor of a. .. Divisibility of integers 105 where the coefficient of the highest degree term is unity and the others a. ....3 when a = '3-y for some integer ^y..3. .. . where the exponents are rational integers satisfying 0<m'<m.. k2. 0(0) = 0. w2. W(Q) _Ob+ql)3b-1+.. §14... is likewise an integer... the coefficients a. + knwn...... . . a . . + se = 0. .. mentioned in the Introduction. where all the coefficients p... ..(3.. ... . 0<a'<a. wn denote all the n products of the form e' wm'aatOb' .+Pa =0. Now if we put n = mab e and let w1. . . By hypothesis. e are roots of equations O(a) =as+pla' 1+. all the products WW1.. . Q... X(C) = Ee + s1Ee-1 + . 1... O(a) = 0... q. If a.. or that . kn are rational integers. Divisibility of integers We say that an integer a is divisible by an integer .. Proof.. e are integers. as in the pre- vious proof. . with the help of the equations F(w) = 0..D. 0<e'<e. It follows from this definition and Theorem 1 of §13 that we have the following two elementary propositions... § 14. 0<b'<b.... that w is an integer. a' are divisible by a then a + a'.. wwn reduce immediately to the form k1w1 + k2w2 + . .a' are also divisible by µ... The same thing is expressed by saying that a is a multiple of .. . It then follows. s are positive rational integers then B a' is also an integer. s are rational integers. It follows from the latter theorem that. x(e) = 0...+qb=0..E. then it is easily seen that.. /3... if a is any integer and r.3 divides a. for example.

disappears again when the integers under consideration are confined to a field of finite degree.32 are the two roots /3 of the equation . that is.32 are integers as well as a. If each of two nonzero integers a. Because. and not a unit. A deeper investigation will enable us to see that two nonzero integers a. because every integer is divisible by 1. if a is a nonzero integer. Fields of finite degree The property of being decomposable in infinitely many ways.32 . In all questions concerning just divisibility. 106 Chapter 3.. we have a = v' and also a = /31. 3 have a greatest common divisor. whether an integer or not.3#' where a' and 3' are integers.32 where )31. to the integers that divide all integers.. A unit a must therefore divide the number 1. associates behave as a single number. but we shall later (§30) be able to derive it very simply from the theory of ideals. However. then we can decompose it in infinitely many ways into factors which are integers but not units. I shall therefore end these preliminary considerations of the domain of all integers with the remark that there are absolutely no numbers in this domain with the character of prime numbers./3 + a = 0. For example. § 15. and it is clear that any associates of a third number are associates of each other. and conversely it is evident that every divisor e of the number 1 is a unit. Indeed. Each algebraic number 0. it is necessary to pay particular attention to units. then every associate of a is divisible by every associate of . a' is divisible by the other then we have a' = ae. a' associates if they are related in this way. General properties of algebraic integers 2. At the same time we see that each product and quotient of two units is itself a unit. This important theorem is not at all easy to prove with the help of the principles developed thus far. First we have to define the extent and nature of such a field. We call two numbers a. Conversely. 31. and hence (by virtue of proposition 2 above) also by E. if a is divisible by . which can be put in the form aa' +. whose pres- ence has just been pointed out in the domain of all integers. if a is a unit then each of the two integers a and a' = ae is divisible by the other. It follows from Theorem 2 of §13 that /a--. where a is a unit. If a' is divisible by a and a is divisible by i then a' is also divisible by u.3. evidently satisfies .3.

4(t) gives. and that f (t) will not be divisible by any polynomial of lower degree and rational coefficients. since ¢(t) has no common divisor with the irreducible function f (t). if we choose any nonzero number a in A.. there is an infinity of polynomial functions F(t) of one variable t which vanish for t = 9 and whose coefficients are rational. xn_1 and degree < n. For this reason.. 9n-1 form an irreducible system (§4. since . It is easy to see that this field is contained in any other field A since. because if we divide fi(t) by f (t) the remainder will be a function 0(t) of the kind above. product and quotient of any two numbers in A also belongs to A. § 15. that is. where 0(t) = xo + xit + x2t2 + . difference. For the first two operations this evidently follows from the common form 0(9) of all the numbers w. satisfying the identity f (t)fi (t) + O(t)01(t) = 1. with rational coefficients. whence the claim follows.02'. . We then see easily that the numbers w are closed under rational operations. Fields of finite degree 107 an infinity of different equations with rational coefficients. When t = 9 this gives the result claimed. addition. We first remark that each such number w = 0(9) is uniquely expressible in this form.. and it follows immediately from the well-known method of polynomial division that each of the functions F(t) will be algebraically divisible by this function f (t). it is necessary that the quotient 1 of the numbers a and a belong to A. and at the same time we have 0(9) = 0(9).. However. and for multiplication it suffices to remark that each number of the form '(9). the method for finding the greatest common divisor of the polynomials f (t). among all these functions F(t) there is necessarily one f (t) whose degree n is as small as possible. x2. that is. as we know. and it is clear at the same time that the n numbers 1.. is likewise a number w. x1.01. the function f (t) and also the equation f (9) = 0 are called irreducible. multiplication and division.. 01(t). The simplest example of a field is the system of rational numbers. We now consider the set Il of all numbers w of the form 0(9). I call a system A of numbers a (not all zero) a field when the sum. subtraction. .1). by virtue of the irreducibility of f (t). + xn_itn-1 is a n y polynomial in t with rational coefficients xo. 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. two polynomials fi (t).. where V)(t) is a polynomial of any degree with rational coefficients. Finally.

. + hnWn. into the new. a/. by the substitution. then a field of finite degree can also be defined as a field with only a finite number of divisors. but this should not result in any confusion. wn a basis of the field Q. a. . Such a field 11 includes n independent numbers. . ... . whereas any n+1 numbers in the field evidently form a reducible system (§4. hn. (Translator's note. 108 Chapter 3. hn will be called the coordinates of the number w with respect to this basis. W2=02(0).) .t and the number n is called its degree. Our system 11 is a field. xn_1 which follow x0. a+)3. The rational numbers come from 0(9) by annulling all the coefficients x1. . . § 16.. B2. Then it is evident that each number w = q5(0) is always expressible.Q t If we understand a divisor of a field A to be any field B whose members all belong to A. for example 1. multiplications and divisions. Wn = 4n (0) of the field Il.3.. t Dedekind calls it a permutation. And conversely. and we say that the old elements are changed. By an isomorphism$ of S2 we mean a substitution which changes each number a. . Now if we arbitrarily choose n numbers W1=01(0).2) form an irreducible system if and only if the determinant of their n2 rational coefficients x is nonzero.. combined with the definition of field. h2. . can also serve as the definition of a field 1 of nth degree. . x2. in the form w = h1w1 + h2w2 + . all numbers w of this form are in Q. The word divisor (and the word multiple) is used here in a sense directly opposite to that used in speaking of modules and ideals. 0. I shall not go into the proof of this assertion. The rational coefficients hl.1).... A field 1 obtained from an irreducible equation f (9) = 0 of degree n in the manner indicated is called a field of finite degree.. 9n-1. uniquely... with rational coefficients hl. . In this case we call the system of n numbers w1i w2.. subtractions. General properties of algebraic integers all the rational numbers are engendered from the number 1 by repeated additions. . . by the results just proved for the numbers w = 0(0)..0. these numbers (by §4. The latter property. a-)3. h2. Now let 11 be any field. However. Conjugate fields We ordinarily understand substitution to be an act by which objects or el- ements being studied are replaced by corresponding objects or elements.

3)'. in view of (3). a' f' are in 1'. Thus the numbers of the system 1' are also closed under subtraction. Since we now have a = it follows from (2) that we also have a' = (a//3)'/3'.Q'. because it satisfies the characteristic conditions (1) and (2).3' = 0 then so too is 3 = 0. (a + Q)'. The correspondence can therefore be reversed in a unique manner.3' is nonzero then. (a -. Thus the numbers of the system S2' are also closed under division. if we let a'. are satisfied and the substitute numbers are not all zero. Q will also be nonzero and hence a/f is a number in the field Q.D.3 and a . and the substitution that changes each number a' in the field 1' into the corresponding number a in the field Sl will be an isomorphism of the field 11'. with the result that a' = (a/. whereas we have agreed that the numbers a' in the system Sl' are not all zero.3Y (3) (a = a' - (4) (a/Q)' = a'/.3'. We shall see that the set S2' of the latter numbers forms a new field.3' Indeed. that an isomorphism changes two different numbers a. if . Q'.3' be any two numbers in the system 0. ./3 are likewise in Q. (a. otherwise every number a in the field S2 could be put in the form (a/0)/3. then there will be two numbers a.3)'/3' = 0.3) +. it follows from (1) and (2) that the numbers a' + /3'. Q. §16.E. which is condition (4). Thus the numbers of the system f' are closed under addition and multiplication.3)'. it follows from (1) that a'=(a-/)'+(3'. and that the isomorphism also satisfies the following conditions: -. and consequently S2' is a field. It evidently follows from this. /3' in the field Sl'. (a/0)' in such a way that the conditions (1) (a+/3)'=a'+O'. . which is condition (3). /3 in the field Il into two different numbers a'. since the numbers a = (a -. a/3 are likewise in S2. Conjugate fields 109 of 1 into a corresponding number a'. Moreover. by virtue of (1). so that each number a' in the field 11' corresponds to a single number a in the field Q. But since the numbers a + 0. Finally. / in the field 11 changed respectively into a'. . (2) (a/3)' = a'. We also note that if .

After these general considerations. 8' will also be an algebraic integer. and any two corresponding numbers a. We also call Il and 1' conjugate fields. and Sl" will be changed into Il by the isomorphism P/-1P-1 (PP')-1 = We have previously remarked that each field includes all the rational numbers. and a' is likewise changed by an isomorphism P into a number a" in a field a". and the rational coefficients of R(t). then it is clear that the substitution changing each a in S l into the corresponding a" in S2" is likewise an isomorphism of the field 1. by rational operations. because when 0 is changed into 0'. our proposition follows immediately from properties (1). and pose the problem of finding all the isomorphisms of Q. and if 8 is changed into the number 0' by an isomorphism of the field. will also be in S2. If we let P-1 denote the inverse isomorphism of P. Thus each number 8' conjugate to an algebraic number 8 is likewise an algebraic number. which replaces each number in St by itself. since each rational number can be engendered from 1 by a series of rational operations. it follows from the preceding results that an isomorphism of S2 will be completely determined by the choice of a root 0' of the equation 0 = f (0'). and then. and hence satisfies an equation of the form 0 = F(8) with rational coefficients. if we choose . 110 Chapter 3. Because if each number a in a field SZ is changed by an isomorphism P into a number a' in a field Il'. then PP-1 will be the identity isomorphism of 1. It is also easy to see that two fields conjugate to a third are conjugate to each other. General properties of algebraic integers Each of the two isomorphisms will be called the inverse of the other. w = 0(8) is changed at the same time into w' = 0(0'). For each field SZ there is evidently an identity isomorphism of SI. if we take a = . Now let 0 be any number in the field St and let R(t) be any rational function of t with rational coefficients. It follows immediately that if 0 is an algebraic number. by §15. since it is formed from the number 8. Thus every field is conjugate to itself. we return to our example of a field l of finite degree n. The number w = R(8). in the case where the denominator of R(t) does not vanish for t = 0. a' will be called conjugate numbers. Since all numbers w in such a field are. (3). and we denote it by PP'. and it is easy to show that each of these is changed into itself by an isomorphism of the field.Q in (4) we get 1' = 1. Conversely. (4). we must also have 0 = F(8'). which apply to all fields. and if 8 is an algebraic integer. (2). Because. of the form 0(e) where 0 is a root of an irreducible nth degree equation 0 = f (8). then the number w will be changed into the number w' = R(8').

Norms and discriminants The norm N(w) of a number w in a field S2 of degree n is the product (1) N(w) = WW.3)' = ()3(9'). . (a. Now if we have a=01(0). (Translator's note.. and consequently a'=01(0% Y=02(0% (a +. we let 01(t). then this substitution will really be an isomorphism of S2. 04(t) = O1(t)-02(t) + 05(t)f (t) Taking t = 9'. and each of them will correspond to an isomorphism P'. p(n) of the field Q. Norms and discriminants 111 9' to be any root of the equation 0 = f (9'). (n) f Nowadays called a Galois or normal extension (of the rationals). 9(n) will be different.. -04(9) = 01(9)Y'2(9) and the irreducibility of the function f (t) that we have identically 03(t)=-O1(t)+02(t). it would distort Dedekind's meaning to use the term "extension" here. denote any particular functions of the form 0(t).(t-9(n)) then the n roots 01.. To avoid misunderstanding. . If we now put f(t) = (t-9')(t-9").. The isomorphism P('') changes each number w = 0(9) of the field SZ into the conjugate number w(') = O(9(r)) of the conjugate field SZ(r). however... 02(t).t The algebraic principles of Galois amount to reducing the study of arbitrary fields of finite degree to the study of normal fields. If they are all the same. lack of space prevents me from going further into this subject. § 17.0 . SZ will be called a Galois or normal field. it follows from the equations 03(0)=01(0)+02(0).. however. it will satisfy conditions (1) and (2)..3=04(0).. 0=02(0). nevertheless may not all be different. To show this. . and replace each number w = 0(9) in the field Sl by the corresponding number w' = 0(9')... P". § 17.3)'=04(0'). although derived from Il by n different isomorphisms.. we point out that the n conju- gate fields S2(''). since the irreducible polynomial f (t) cannot have a common divisor with its derivative f'(t). .) .3=03(0). a. this gives the equations (1) and (2) we have to show. a+. . that is..

112 Chapter 3. General properties of algebraic integers
of the n conjugate numbers w', w", ... , w(') of w under the isomorphisms
P', pt/,..., P(n). It does not vanish unless w = 0. If w is a rational
number, then all the n numbers w(-) equal w, and hence the norm of a
rational number is its nth power. If a, 0 are any two numbers in the
field I we have (a/3)(-) = a(')#(') and consequently
(2) N(a)3) = N(a)N(O).
The discriminant A(al, a2i ... , an) of any n numbers al, a2, ... , an in
the field Il is the square
(n))2
±ala2// ... an
(3) A(al, a2, ... , a,,,)

of the determinant formed by the n2 numbers ai-).
By virtue of a
well-known proposition in the theory of determinants we then have the
relation
(-1)n(n-1)/2N(f'(0)),
(4) 0(1, 0, 02, ... , 0n-1) =

and since f'(9) cannot be zero, by the irreducibility of the polynomial
f (t), it follows that the discriminant (4) has a nonzero value.
Now if then numbers w1i w2, ... , wn form a basis of the field St (§15),
and if
Lo =h1w1+h2w2+"'+hnw,,,
is any number in the field then, since the coordinates hl, h2i .... hn are
rational numbers, the isomorphism P(') will change w into the number
w(-) = h1wi-) + h2w2-) + ... + hnwn-),

from which we conclude that

( `5) 0(a1, a2, ... , a.) = a 20(W1, w2, ... , W,y),
where a is the determinant of the n2 coordinates of the n numbers
al, a2i .,an. This implies, first, that the discriminant of the basis
w1, w2, ... , wn cannot vanish, otherwise every discriminant would vanish,
contrary to the fact from above that 0(1, 0, 02'...' On-1) is nonzero. At
the same time it follows that A(a1 i a2, ...,an) will vanish if and only
if the numbers al, a2, ... , an are dependent on each other (§4,2), and
hence do not form a basis of Q.
Since the numbers in a field are closed under multiplication, for any
number t in St we can put

§ 18. The integers in a field Il of finite degree 113

/2W1 = m1,1W1 + m2,1w2 + + mn,1Wn,
(6) PW2 = m1,2w1 + m2,2W2 + + mn,2Wn,

PWn = m1,nW1 + m2,nW2 +'* * + mn nwn,
where the n2 coordinates mi,i, are rational numbers. Applying the n
isomorphisms P(r) yields n2 new numbers of the form
p wir> = ml iW1T) + m2 iw2*) + ... + Mn iW(T)

and, since their determinant is
N(µ) fWIW2 ...W(n) = tml lm2 2 ... mn,n ±W/1W2 ...W(n)

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)

I ml,n m2,n ... mn,n - z I

which, being valid for every rational value of z, is necessarily an identity
in z. At the same time we see that the n numbers p', p",..., µ(n) con-
jugate to the number p are the set of roots of an nth degree equation
whose coefficients are rational numbers.

§ 18. The integers in a field Il of finite degree
After these preliminaries, we now pass on to our main objective, the
study of the integers in the field Il of degree n, the set of which we

114 Chapter 3. General properties of algebraic integers
denote by o. Since the sum, difference and product of any two integers
are also integers (by §13,1), and in 0 (by §15), the domain o, includ-
ing all the rational integers, is closed under addition, subtraction and
multiplication. However, the first thing is to put all these numbers in a
common, simple form. The following considerations lead us to it.
Since each algebraic number w is a root of an equation of the form
+C1wm-1 +...
cwm +Ctn-lw+Cn = 0,
whose coefficients c, cl, ... , cm_l, cm are rational integers, it follows,
multiplying by cl-1, that each such number yields an integer cw when
multiplied by the nonzero rational integer c. Now if the n numbers
w1, w2,. .., wn form a basis of the field S2, we can take nonzero rational
numbers al, a2i ... , an so that the n numbers
al = alw1, a2 = a2w2, ... , an = anwn
become integers, evidently forming a basis of S2, since they are indepen-
dent (by §4,2). It follows that their discriminant 0(al, a2i ... , an) will
be a rational number (by §17), and in fact a nonzero integer since, by
definition, it is formed from the integers o by addition, subtraction
and multiplication. Moreover, we obtain all numbers w in the field S2 by
letting the coefficients x1i x2, ... , xn in the expression

run through all rational values. If we restrict their values to be rational
integers then we certainly obtain only integers w (§13,1). However, it is
likely that not all integers in the field S2 will be represented in this way.
The situation relates to the following very important theorem:
If there is an integer # of the form
klal + k2a2 + ... + knan
Q
k
where k, ki, k2, ... , kn are rational integers without common divisor, then
there is a basis of the field S2 consisting of n integers ,(31, /32 i ... , Nn which
satisfy the condition
k2A('31,32...... 3n).
0(al,a2,...,an) =
Proof. Since ,Q, al, a2, ... , an are integers, they form a basis of a module
b = [Q, al, a2i ... , an] which includes only integers from the field Q. But
since only n of these n + 1 numbers are independent there are (§4,5)
n independent numbers 01, /32i ... , Qn which form a basis of the same

. and since the same is true of the n + 1 numbers k. 32.. + khan and observing that /31....2132 + . a2.....nQn.r in the rth row of the determinant c by the respective elements Cl. obtained by suppressing one horizontal line.... al = C1.... .. an = C1.6).... . C2. al... k2.2Nn. CO0 .31. an].... ... Substituting the preceding expressions for .2 + ..... The integers in a field f2 of finite degree 115 module b = [.E.. If we put E ±Cl 1C2 2 .. . .. .. that the partial determinant obtained has value Sk ....... + knCl..D.. + Cn...02 + . + knCn n_ If we now replace the elements Cl.. We then have n + 1 equations of the form = cl)3l + C2..... Cn. kc2 = k1c2.. .... .. . C2...1 + k2c1.. we conclude from the preceding equations.. k are rational integers with no common divisor.COn... then we have (§17/....101 + c2....l + k2Cn.2 + .. A(al.(5)) )3n). whose n(n + 1) coefficients are rational integers and whose n + 1 partial n x n determinants. . + en. .. .. have no common divisor (§4.1.l a2 = c1.2 + .Q... . k.r...201 + C2.132.. .32 + .. Cn.. If k > 1.32 + . .n.1 + k2c2. an) = C2A(Ql. k n..n = C.... Qn are independent...n. a2...... Thus the n + 1 quantities ck ck1 ck2 ckn k. . .an] and which are therefore integers in the field. § 18.. a2. k1.. k.. ...... so that the integer ..01 + C2. we necessarily have c = ±k...Q = k1a1 + k2a2 +.... + Cn..3 is not in the module a = [al.. .. 132. using a well-known theorem.. Q... ken = klcn. Cn.. + knc2. we find that kcl = k1C1. . an in the equa- tion k.r.

To investigate the integers w we put t= zx.. Under the nonidentity iso- morphism of the field.changes into -v/d-. Vd. . General properties of algebraic integers then there is a basis of the field consisting of n integers Q1. h2. + hnwn in the field f must have integer coordinates hl. And it follows immediately from the pre- ceding results that. 116 Chapter 3. . .. Moreover. . . wn. a2. the set o of all integers in the field Il is identical with the finitely generated module P1. hn produces an integer w..3n with discriminant 0(/31.. W2. (3n) of absolute value < A(a1. where t. . The discriminant D of such a basis is a fundamentally important in- variant of the field Q.. Thus 11 is a normal field (§16). there must be a basis w1.wn) = D has minimum absolute value.. For this reason we call it the fundamental number or the discriminant of the field 0... and represent it by A(f ).. The numbers a. ... The set of all the numbers 0(0) in the corresponding quadratic field is evidently the set of all numbers of the form w=t+uv/d. an integer w is not divisible by a rational integer k unless all its coordinates are divisible by k. ... and we take its discriminant to be the number +1.. .. and b is nonzero. each integer w = hlwl + h2W2 + . wn] whose basis consists of the n independent integers w1i w2. . . relative to such a basis.. . /32..32. Now since the discriminant of every integer basis of the field Q is a ratio- nal nonzero integer.w2i. since each system of integer coordinates hl. Conversely... h2i .. Each root 0 of an irreducible quadratic equation is of the form 0=a+bV.. To clarify the general case we again consider n = 2. wn whose discriminant A(w1. u= yz . . where d is a nonsquare rational integer which is also not divisible by a square (except 1). that is. SI is the field of rational numbers. an). as shown above. and hence w changes into the conjugate number w'=t-uvd. the case of a quadratic field. . hn. b are rational.. W2. which is also in Q. u run through all rational values. In the singular case n = 1. . .

But if we have d . then z must equal 1 and so we have o = [1. z wwx - z2 must also be integers. By the general definition of divisibility (§14) we then have A = pw. which means that z = 1 or z = 2. . so is w' (§16) and consequently 2 2 dy w+w'=2x. The integers in a field 0 of finite degree 117 where x. if the latter are integers. The system m of all numbers in the field 0 t It follows. and D = 1 1 1 2 d 2 = d. p belongs to S2.1 (mod 4). in the case d = -3 that the integers of the field are not all of the form x + y/ where x. Let A. where invariants additional to the discriminant are necessary for the complete determination of a cubic field. In the second case x is odd. we shall be able to give a complete explanation of this fact only with the help of the theory of ideals. so x2 = 1 (mod 4). §18. Thus z is prime to x while z divides 2x. and since the quotient w of the two numbers A. and therefore we have d . Now if w is an integer. since d is not divisible by any square other than 1. where w is an integer. If the latter condition is not satisfied. Suppose that e is the greatest common divisor of z and x.2 or d . and since we must have x2 = dy2 (mod 4) it is necessary that y also be odd.t and we have O= = 1. z are rational integers with no common divisor and z can be assumed positive.1 (mod 4). In the first case. w = x + yv1d is certainly an integer. This is not so for the next case. that is. then so is w (§13). for example. It is also clear that a quadratic field is completely determined by its discriminant D. if d . However.and 1 2 v"d D= 1 -v =4d. v/4. We now return to considering an arbitrary field Il of degree n. Conversely. hence also dy2 and finally y2.3 (mod 4). namely n = 3. since x. These two cases can be combined into one by noting that o = [1. y are rational integers. and we add the following remarks on divisibility and congruence of numbers in the domain o. w will likewise be a number in the domain o. 12 . p be two such numbers. z can also equal 2.dy2. by definition of a field. Then e must also divide y. It is necessary that e2 divide x2 . z have no common divisor. y. and hence be equal to 1. y. D 2 D ] in each. and suppose that A is divisible by p.

that the number of these classes is (o. Consequently we have m = [pwl./3 (mod p).. it follows from the equation following (7). wn by n equations of the form (6). If we understand a class modulo p to be the set of all numbers in o congruent to a particular number.i. In the contrary case. . W2. through all numbers of the form w=h1w1+h2w2+"'+hnw. the elements of which will be covered in the next chapter. pwn]. that is. o) = 1. in which the coefficients mz.. 118 Chapter 3. The system m is identical with o if and only if p is a unit. as we have already indicated in detail in the Introduction and in Chapter 2./3 is divisible by it and hence in m. together with Theorem 4 of §4. In this conception of congruence. m). whose coordinates hl. that completely new phenomena concerned with the decomposition of numbers into factors arise in this same domain o. as w runs through all numbers of the domain o = [W1. And since the integers pw1i pw2. and hence congruent to each other. the meaning of which was explained in §2. Thus this congruence is completely equivalent to the following: a . wn]. /3 in the domain o are congruent modulo p. However. then. are necessarily rational integers. when the difference a . . h2i . a.. ... .. . . (§17). in which case ±N(p) = (o. . We now say that two integers a. m) = ±N(p). pw2i. pwn forming the basis of m are connected to the numbers w1i w2. General properties of algebraic integers divisible by it therefore consists of all numbers of the form pw. .. there is a complete analogy with the theory of rational numbers. These phenomena are brought back under the rule of simple laws by the theory of ideals. . are rational integers.... /3 are called incongruent modulo p. ham./3 (mod m). the number of these classes will be (o. it is plain. where an actual number p appears as divisor or modulus.. and write a . in the notation of §2...

The theory is founded on the notion of ideal. let f be a field of finite degree n. that is. that is. The sum and difference of any two numbers in a also belong to a. whose origin has been mentioned in the Introduction. §19. and let o be the domain of integers w in Q. Let y be a particular number. It differs mainly in external form. we shall prove the fundamental laws which apply to all fields of finite degree. a is a module. and whose importance has been sufficiently illuminated by the example in Chapter 2 (§§11 and 12). The latter are what we refer to when we speak of "numbers" in what follows. 4 Elements of the theory of ideals In this chapter we shall develop the theory of ideals to the point indicated in the Introduction. Ideals and their divisibility As in the previous chapter. 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). then the system a of all numbers a = pw divisible by p forms an ideal. II. unless the contrary is expressly indicated. The product aw of any number a in a with a number w in o is a number in a. however. and which regulate and explain the phenomena of divisibility in the domain o of all integers in such a field Q. We call such an ideal a principal 119 . In particular. An ideal of this domain o is a system a of numbers a in o with the following two properties: I. the principal difficulty to be surmounted is now thrown more clearly into relief. if the theory has not been shortened it has at least been simplified a little. We begin by mentioning an important special case of the concept of ideal.

120 Chapter 4. because every indecomposable number also has the character of a prime number (see the Introduction and §7). We say that an ideal m is divisible by an ideal a. a number of the form ep. where the resulting ideal consists of the single number zero. It is evident that this ideal will be unchanged when p is replaced by an associate. Elements of the theory of ideals ideal and denote it by o(p). divisible by or7. where every ideal of the field Il is a principal ideal. since all numbers in o are divisible by p. or that it is a multiple of a. It is easy to see that no other ideal can contain a unit. The ideal o plays the same role among the ideals as the number 1 plays among the rational integers. then all the products 77P will again form an ideal. In all these cases. all numbers in the ideal a are likewise in o. if p runs through all the numbers in an ideal r while 77 is a fixed nonzero number. and hence all numbers w in the principal ideal o. However. that is. According to this definition. and hence every number in a. Conversely. or more simply by op or po. If a is a number in the ideal a then the principal ideal oa will be divisible by a (by II). We evidently have (rr7)r7' = r(7777') = (r77')r7. numbers are governed by the same laws that govern the theory of rational integers. because without doubt it will help greatly in understanding our general theory. is divisible by the ideal a. 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. But since. we have a = o. At the same time we say that a is a divisor of in. we shall exclude this case from now on. we immediately carry over to ideals the notion of divisibility of modules (§1). a module of the form [m] where m is a rational integer (§§1 and 5). when all the numbers in m are also in a. The same is true for the special quadratic fields considered in Chapter 2 (§6 and the be- ginning of §7). In the case n = 1. but I mention it now to encourage the reader to make continual comparisons with the special cases. If p is itself a unit we have op = o. This will follow easily from the results below. and 77r' will be divisible by . Because if the unit a is in the ideal a then (by II) all products ew. every ideal is evidently a principal ideal. are in a. each ideal is divisible by the ideal o. Since each ideal is a module (by virtue of I). that is. where our theory becomes the old theory of num- bers. where a is a unit. We shall denote an ideal formed in this way from an ideal r and number 77 by rq or 77r. The notion of principal ideal op also includes the singular case where p = 0. by definition. For this reason we say that the number a. Likewise we say that an ideal a is divisible by the number 77 when a is divisible by the principal ideal o77. and especially with the old theory of rational numbers.

w' in the domain o are congruent or incongruent modulo a according as their difference w . b will always be of the form r7r.a'w will be in a since (by virtue of II) the products aw. If r' is the divisor of r corresponding to the number 77'. we say that two numbers w. and ar7 their greatest common divisor. The notion of a principal ideal op is the special case of rp where r = o. if p = a = 3 is a number in m and hence also in a and b. then the least common multiple m of a. Thus m and a enjoy all the properties of ideals. The laws of divisibility of numbers in o are therefore included in the laws of divisibility of ideals. /3'w are in b. Certainly m and a are modules (§1. This is because 7rr7'r' is the least common multiple of 77r and or777'. §20. This case occurs frequently in what follows. §20. and hence divisible by 77r. b are also ideals. The least common multiple m and the greatest common divisor a of two ideals a. that the ideal r dividing the ideal a corresponds to the number 77. 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 +. Also. 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 . since 77a is a common multiple of a and o77. and they are divisible by o. we must also note that two congruences modulo the . a'w are in a and the products /3w. We express the congruence of w and w' modulo a (§2) by the notation w . Norms 121 r7r if and only if r' is divisible by r. then r' will also be the divisor of a corresponding to the product q?7'. If b is a principal ideal o77. since 77r is the least common multiple of a and o77. Thus the equation 77r' = it entails the equation r' = r. and for that reason we say.w' belongs to a or not. and or777' is divisible by or7.W (mod a). since a and b are divisible by o. where r is another ideal and in fact a divisor of a. 3 and 4). and hence also of a and or777'. Norms Since each ideal a is also a module. At the same time it is clear that mr7 will be the least common multiple of the ideals all. bq. As well as the theorems on congruences established previously for ar- bitrary modules. for brevity.

m). and hence the domain o con- tains only a finite number of mutually incongruent numbers modulo the ideal a (§2. a). if m is a principal ideal op. If p is a particular nonzero number in the ideal a then the principal ideal op will be divisible by a. where p' runs through the same values as p. w" in o are congruent (mod t). this result includes the evident theorem that N(p) is divisible by p (see §17). W=-W. 3 and 4. since a is divisible by o. By §2. let m be their least common multiple. D) (D. 122 Chapter 4. a) will be called the norm of the ideal a and we denote it by N(a). If p runs through a complete system of N(a) incongruent numbers (mod a) then so does (1 + p). and o is evidently the only ideal with norm 1. That is. can be multiplied to give the congruence ww" . m) _ (o. This number (o.w' (mod m) will be identical with the congruence w . A very important consideration is the number of different classes.w'ware numbers in the ideal a. and adding the corresponding congruences 1 + p = p'. which make up the domain o. 'w" -W "' (mod a). a) = (o. (o77. N(m) = (b.w' (mod p). hence N(a) = (b. then (by virtue of §18) the congruence w . since two numbers i'w' and 77w" in the principal ideal 77o are congruent (mod 77r) if and only if the numbers w'.w'w'" (mod a). and since a is divisible by o it follows (§2. As a special case. op). Let a. op) = ±N(p) by §18. Now suppose that r is any ideal and 77 is a nonzero number.2). and let a be their greatest common divisor. N(a) is always divisible by a. r) = N(r). b be any two ideals. since the products (w-w')w" and (w"-and hence also their sum ww" . (o. op) _ (o%a)(a. But the number (o. Always. yields N(a) = 0 (mod a). The norm of the principal ideal op is equal to ±N(p). a)N(b) . mod- ulo the ideal a.4) that (o. Moreover. Elements of the theory of ideals same ideal a. ri7) = (o. a)N(a). we have (b. a) m) a) and. (o. b) (b.

§21. Q. so that m is of the form rrl. This is because the ideals different from o that divide the ideal a include one. then qp will be the least common multiple of the two ideals p and orl. But r cannot be o. Otherwise the ideal 77(op) would be a common multiple of p and oil. consequently r = p. because qo is not divisible by p.D. p is divisible by the prime ideal p. contrary to hypothesis. Q. a) _ (oi'.E.0 (mod a). then we have ('0. and consequently N(a) = N(r)N(D). p. Prime ideals 123 and N(m)N(0) = N(a)N(b). But the divisibility of 77(op) by ilp implies (§19) the divisibility of op by p.D.D. and the latter is certainly a prime ideal. then their product q7P will not be divisible by p.p) > 1. Indeed. different from o and with norm < N(p). Hence N(p) = (a. rrl) = N(r). Thus ?7A is not divisible by p. with the ideal r a divisor of ap and hence equal to o or p. 2. The ideal r can also be defined as the system of all roots p of the con- gruence q7P . whose norm is smallest. we get (b. If neither of the two numbers rl. since the ideal r is the divisor of a corresponding to the number i (§19). as is easy to see.E. Prime ideals An ideal p is called prime when it is different from o and divisible by no ideals except o and p. If we apply these theorems to the case where b is a principal ideal oi'. §21. then. if 0 is an ideal dividing p. Each ideal a different from o is divisible by a prime ideal.E. The least common multiple of p and orl is in any case of the form rlr. p)N(0) > N(0) and a will be a divisor of the ideal a. contrary to hypothesis. This definition yields the following theorems: 1. If the number rl is not divisible by the prime ideal p. Thus p is a prime ideal. Q. but different from p and o. . Consequently it would be divisible by the least common multiple IN of p and orl. 3.

which is always possible because p is not divisible by a. and hence will be different from o. if there is an ideal e different from o which divides a and is not divisible by p. and p cannot be 1 otherwise we should have p = o (§19). and at the same time make r divisible by p.pq is also divisible by p. then we choose q to be a number divisible by e but not divisible by p. if all the ideals (except o) that divide a are divisible by p. where p is a positive rational prime number. for example r7 = 1. form a module [p]. since 77 is not divisible by a. and hence divisible by p. But if a is different from p we first confine ourselves to showing the exis- tence of a number 77 such that the divisor r of the ideal a corresponding to 77 is at the same time divisible by p. satisfies the condition. we see without difficulty that we certainly have r = p if we choose 77 so that N(r) is as small as possible. where a is the greatest common divisor of a and oq (§20). r will also be different from o. b would likewise be divisible by p. To attain this goal. if the ideal r divisible by p is not equal to p. Second. as becomes evident by putting m in the form pq + r. Elements of the theory of ideals It follows immediately from this that all the rational numbers divisible by a prime ideal p. This is because the smallest positive rational p divisible by p cannot be a composite number ab. Then it is clear that a will be divisible by p. 124 Chapter 4. because any number 77 not divisible by p. Moreover. otherwise one of the smaller numbers a. we can proceed with r the . Since we have N(a) = N(r)N(D). and hence N(op) = p"` being divisible by N(p) (§20). since the remainder r = m . This important theorem is evident when we have a = p. and q r is. while having norm less than that of a. 4. we distinguish two cases: First. the latter condition depends on choosing 7) so that N(a) > 1. If the ideal a is divisible by the prime ideal p then there is a number 77 such that r7p is the least common multiple of a and o77. and the exponent f will be called the degree of the prime ideal p. Because. Moreover. r will also be divisible by p. then we choose 77 to be a number divisible by p but not divisible by a. since 77r is divisible by a and hence also by p. amongst which we have the number N(p) (§20). Then a will be divisible by e and hence again different from o. N(p) will be a power pf of p. And each rational integer m divisible by p must be divisible by p. Having established the existence in both cases of at least one number r7 with the required property. which means that a is different from o. because 77 is not divisible by p (by 1). Now with op being divisible by p.

b. we re-establish the following theorems: 1. Thus r = p.. since each product aw again belongs to a. and hence (by I) each sum of such products belongs to a. and b by b'. has no influence on the final result. ab is divisible by a. and /3 runs through those of an ideal b. al = a and in general we have aras = ar+s. 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. .. and we denote it by a. In addition. ab = ba and. where al. are numbers from the respective fac- tors al. since a is in a and hence in a'. . It follows immediately from this definition that oa = a... Multiplication of ideals If a runs through all the numbers in an ideal a. that is. If a is divisible by a'..D. which combine two ideals into a sin- gle product. the order of the multiplications. and we denote it by ab. am. Finally. which can be written simply as al a2 a. (ab)c = a(bc). together with their sums. Finally. a2. (ar)s = ars.E.. whence we conclude by well-known argumentst that in a product of any number of ideals al. . q p is the least common multiple of a and or7. This is because all the numbers Eal in ab are in a'b'. a. we evidently have a(oi) = a77 and (or7)(oi7') = or777'. and . Multiplication of ideals 125 same way as with a.. while likewise being divisible by p. each product of a number Ea. 2. if c is any third ideal.. .Q is in b and hence in W. form an ideal c./3 in c by a number w in o also belongs to c. then ab is divisible by a'b'. The product ab is divisible by the factors a and b.. . then all the products of the form a3. also under subtraction. a. 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. That is. This is because (by virtue of property II) each product aw.. and evidently consists of all numbers of the form Eala2 . hence each product a. We also put ao = o.. If all the m factors equal a their product will be called the mth power of a. §22. This ideal c is called the product of the two factors a.. a2.. t See §2 of Dirichlet's Vorlesungen fiber Zahlentheorie.. a2. These numbers are in o and they are closed under addition. §22...3. because the numbers (-a) are likewise in a. a.. Q.

126 Chapter 4. Elements of the theory of ideals
3. If neither of the ideals a, b is divisible by the prime ideal p, then
the product ab is also not divisible by p.
This is because there are numbers a, 0, in a, b respectively, which are
not divisible by p, and then the number aj3 in ab is also not divisible by
p (§21,3).

§23. The difficulty in the theory
It would be easy to augment considerably the number of theorems con-
necting the notions of divisibility and multiplication of ideals, and we
mention without proof the following propositions, simply to emphasize
the resemblance with the corresponding propositions in the theory of
rational numbers.
If a, b are relatively prime ideals, that is, having greatest common
divisor o, then their least common multiple is ab, and at the same time
N(ab) = N(a)N(b).
If p is a prime ideal, and a is any ideal, then either a is divisible by p
or a and p are relatively prime.
If a is an ideal relatively prime to b and c, then a is also relatively
prime to bc.
If ab is divisible by c and a is relatively prime to c, then b is divisible
by c.
However, all these propositions are insufficient to complete the anal-
ogy with the theory of rational numbers. It is necessary to remember
that divisibility of an ideal c by an ideal a, according to our definition
(§19), means only that all the numbers in the ideal c are also in the
ideal a. It is very easy to see (§22,1) that any product of a by an ideal
b is divisible by a, but it is by no means easy to show the converse,
that each ideal divisible by a is the product of a by an ideal b. This
difficulty, which is the greatest and really the only one presented by the
theory, cannot be surmounted by the methods we have employed thus
far, and it is necessary to examine more closely the reason for this phe-
nomenon, because it is connected with a very important generalisation
of the theory. By attentively considering the theory developed until now,
one notices that all the definitions retain their meaning, and the proofs
of all theorems still hold, when one no longer supposes that the domain
o consists of all integers in the field Q. The only properties of o really
needed are the following:

§23. The difficulty in the theory 127

(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

of the equation 92 + 9 + 1 = 0 the module [1, VI '--3] forms an order o not
containing all the integers of the field. The modules [2,1+v/-3] = p and
[2, 2vl---'-3] = 0(2) must be regarded as ideals in this order o, inasmuch as
they enjoy the properties I and II (§19). However, while o(2) is divisible
by p, there is no ideal q in o such that pq = o(2).
t Disquisitiones Arithmeticae, art. 226.
$ I treat this theory in detail in the recently published memoir: "Ueber die An-
zahl der Ideal-Classen in den verschiedenen Ordnungen eines endlichen Korpers."
(Festschrift zur Sdcularfeier des Geburtstages von C.-F. Gauss. Braunschweig, 30
April 1877).

128 Chapter 4. Elements of the theory of ideals

§24. Auxiliary propositions
To achieve the completion of the theory of ideals in an order o containing
all the integers of a field 1 we need the following lemmas, which are not
true unless we restrict ourselves to such a domain o.
1. Let w, p, v be three nonzero numbers in o such that v is not
divisible by u. Then the terms of the geometric progression
(AV)2, (µ)3
,
w
up to a term
Cvle
W
µ/
at some finite position, will all be in o, and beyond that none of them
will be an integer.
In fact, if the number of integral terms exceeds the absolute value k
of N(w) then there must be (by §18) at least two among them, corre-
sponding to exponents s and r > s, which are congruent modulo w. But
such a congruence,

)(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

where w' is an integer, and hence (§13,2) 7) itself is an integer, contrary
to our hypothesis that v is not divisible by p. Thus at most k terms
of the series can be integers, and hence members of o. Moreover, if the
term

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)

Q. then (§19) the ideal rcvp will be the least .2) the two numbers K. §25. Laws of divisibility 129 will also be an integer. and a number r. Laws of divisibility With the aid of these lemmas it is easy to bring the theory of ideals in the domain o to the desired conclusion.r = wr-sps is an integer (§13. such that rcp is the least common multiple of oA and or.2 is not divisible by A. Since op is divisible by p there is a number v such that vp is the least common multiple of op and ov (§21. v be two nonzero numbers in o. which is found in the following laws: 1. 2. Let p be any nonzero number in the prime ideal p. This completes the proof of the proposition. otherwise the least common multiple of op and ov would be ov and not pv. If are the last two integral terms of the series 2 3 v v Jul /I (A) V . This number v cannot be divisible by p. with v not divisible by A. Then there are two nonzero numbers rc. then we evidently have e > 1 and e+1 r6 V /C2 v A µ A =µ (A Thus n2 is not divisible by A. Now if we choose (§24. Let p. A in o such that K V A p and s. not divisible by p. Proof.E. A t' (A) and hence in o.D. because 0. If p is a prime ideal then there is a number A divisible by p.4). §25.2). A so that rcp = Av and rc2 is not divisible by A..

Each prime ideal p has a multiple. is is not divisible by p. by an ideal a. oA is divisible by the other. and at the same time N(a') < N(a). Since a is different from o there is (§21. Multiplying the equation pb = pa' by a we get Ab = Aa'. Proof. Proof. it follows that a' is different from a. where N(al) < N(a). Each ideal a different from o is either a prime ideal or else express- ible as a product of prime ideals. we have as = Aa' where a' is an ideal (§19). as we now can. w' are two numbers from o. Elements of the theory of ideals common multiple of c(op) = oAv and oicv. Thus each of the two ideals pa.. if K is not divisible by p. since all numbers in the ideal a are of the form b = rcw + Aw'. which is a principal ideal. where w. multiplying by a. We shall show that pa = oA. and hence we can put (by 3) a = plat. 2. that o77A is divisible by Aa'. Suppose.D. because the first factors A. A are in Z.E. It follows (§19) that icp is the least common multiple of oA and oK. otherwise 0c2 would be divisible by rcp and hence also by A.D. Thus pa is divisible by oA.2). Then if a is divisible by p. Then p is the divisor of the principal ideal oA corresponding to the number #c.1) a prime ideal pi dividing a. . and hence that 77 is divisible by a'. However. Thus. and in only one way. In addition. 3. because Kp and hence also Kw is divisible by oA. with w in p and za in p. 130 Chapter 4. if w is any number in p we have wS = icww + Aww' . Conversely. we get Aa = Apa' and hence also a = pa'. otherwise oil and not qp would be the least common multiple of a and or7. and consequently we have N(a') < N(a). whence b = a'. Multiplying by p. zo are in p and the second factors Kcw. there is (§21. We retain K and A from above with the same meaning. We then have A = A . that pa = oA. In fact. and hence as by pa (§22.4) a number 77 such that 77p is the least common multiple of a and oil. and consequently pa = oA. Kw + w A m 0 (mod po). because a' is a divisor of a. Q. However. q is certainly not divisible by a. Proof. Q. and p. If the ideal a is divisible by the prime ideal p then there is exactly one ideal a' such that pa' = a.0 (mod A). since 77 is divisible by a' and not by a. But with 77p being divisible by a = a'p it follows.E. and let a be the greatest common divisor of oA and or. in which case o is the greatest common divisor of or. Now let b be an ideal also satisfying the condition pb = a. we can express the number 1 in o as Kw + w. 4.

. 5... pm by the corresponding ideals 01.... so that a1 is different from o. . We then have p1(p2p3 .. whose norms are smaller and smaller. a = Now if at the same time a = g1g2 .. P2.D.. . becomes a principal ideal.. satisfying the condition ab = c.. Q. Suppose that a = plp2 pm. If we now put m = 002 . If N(a2) > 1 we can continue in the same manner. and ab = ab' implies b=b'.. If the ideal c is divisible by the ideal a then there is an ideal b. . qm. then am = (P101)(p2a2) will be a unique product of principal ideals. am... Laws of divisibility 131 If we have al = o then a = pl will be a prime ideal. ..E. we can likewise put a1 = p2a2 where P2 is a prime ideal and N(a2) < N(al). and exactly the same number of times. Q. and hence itself a principal ideal. If the product ab is divisible by the product ab' then b will be divisible by b'. obtaining ideals al. .. Proof. a2i a3. And since pi is divisible only by the two ideals o and pi.. and only one. just as in the theory of rational numbers.pm. t See Dirichlet's Vorlesungen fiber Zahlentheorie.. q2. qm are also prime ideals. finally arriving at the ideal o = am after a finite number of decompositions. But if N(al) is > 1. . §25.. We can continue in the same manner. pm) = pl (g2q3 " ' qm). when multiplied by a suitable ideal m.t and thus arrive at the result that each prime ideal factor in one product occurs in the other. By 2.D. must be divisible by q1. qm. then q1 will be a divisor of the product P1P2 pm and hence (§22. where q1.. Pm = g2q3.3) at least one of its factors. we necessarily have ql = pi. 6. Each ideal a. §8.. p2a2i .. . am to change them into principal ideals plat.. we can multiply the prime ideals p1. We then have a product of m prime ideals p1p2.E. pi for example. whence (by 3) P2P3 .. since q1 is different from o. 02i .

so that rl is divisible by a... Now let a.2). Now if c is divisible by a. Since a is divisible by p. . it follows that N(a) = N(pi)N(p2p3.. we can. multiplying by m (§22.. and suppose that ab is divisible by ab'.D. If in addition ab = ab'. b. 7. Proof. We first consider the case of a product a = pa'.pm. 132 Chapter 4. We therefore have n = o. and r/p is the least common multiple of a and orl. Since a and of are divisible by a'..E. p. that µb is divisible by pb'. which proves the theorem in the case considered. q2. where a is the greatest common divisor of the ideals a and ord. if we have gig2. qr. a = where pl.. put cm = µb. where the factor p is a prime ideal.pm) and hence N(a) = N(pi)N(p2) . b' be any ideals.. Since each ideal (other than o) is (by 4) of the form p1p2. The first equation is impossible unless we have a = pa' = a.2). Elements of the theory of ideals Proof. The norm of a product of ideals is equal to the product of the norms of the factors: N(ab) = N(a)N(b). and hence (§19) that b is divisible by W. then each of the ideals b. . that is. b' will be divisible by the other. whence a = a' and also N(pa) = N(p)N(a'). pa' is divisible by na'.. a must also be divisible by a' (§1. .. since a is divisible by a. are prime ideals.4).. where b is an ideal. Multiplying by a.. qr again denote prime ideals. N(pm) Moreover. there is (by 3) a number rl divisible by a' but not by a. Again it follows. and this is not the case. and hence there is (by 6) an ideal n satisfying the condition na' = Z. . b = where qi. Moreover.qr. then we get ab = p1p2 . so the prime ideal p must (by 6) be divisible by n. p2. b = W.. and thus we must have n = p or o. pmglg2 . by §19. Q.. whence c = ab.. However. the general theorem is an immediate consequence.pm) = N(pl)N(p2)N(p3. Then we have (§20) N(a) = N(p)N(D). We choose the ideal m so that am is a principal ideal op.. and hence cm by am (§22. we get µc = pab..

then a will evidently be divisible by pm.. r. is verified easily. . The factors of a include first of all the factors of a and.. 8. .. as factors in the decomposition of a. If p is a prime ideal and p7z is a divisor of an ideal a then we have (by 6) a = alp"°. satisfy the conditions 0<a'<a. the total number of different divisors will be (a+ 1)(b + 1)(c + 1). c'... b'.. N(q. 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).. which obviously implies N(ab) = N(a)N(b). . N(qr). c'. If we suppose the latter decomposed into all its prime factors.. where the exponents a'.. . N(ab) = N(pi) ..E. Laws of divisibility 133 and consequently N(b) = N(gl)N(g2) . Q.. §25. where p. . then we have a = aa'. The converse proposition... N(pm)N(ql) . 0<c'<c.D. . Since any two different combinations of exponents a'. O<b'<b. Thus if we suppose that every power of the prime ideal which divides a also divides an ideal a. if we denote the product of the other factors by a'. Q. that if a is a divisor of a then each power of a prime ideal that divides a also divides a. corre- spond (by 4) to different ideals a. this amounts to saying that all the prime factors in the decomposition of a appear. By virtue of the preceding theorems we can show that all the divisors a of a are given by the formula a = pa'gb'rc. Conversely. if the decomposition of Z into prime factors includes the prime ideal p at least m times. Proof. and consequently a is divisible by D. .E. b'.). then a will also be expressed as a product of prime ideals.. q.. at least as often. among which the factor p appears at least m times. are different prime ideals. where al is an ideal.. If we combine all factors of the same prime in the decomposition of an ideal a then we find a = pagbrc .D...

if ae is divisible by b. Moreover. where a'. For the moment we shall be content simply to give indications of the proofs. a to be derived from the decomposi- tion of a.4. w . b is Da'b' = ab' = ba'. 1. and all these roots come under the form w . e will be divisible by W. Q (mod b). and we let b(a) denote the number of mutually incongruent numbers modulo a which are prime to a. b=ab'.or (mod b).T (mod ab). b since r is prime to ab if and only if p is prime to a and or is prime to b. and hence also for numbers in o. 134 Chapter . then (§2. b then we have a=aa'. and the least common multiple m of a. §26. always has roots w. Thus we only need to find the value of . b' are relatively prime ideals. b into prime factors. where r represents a class of numbers modulo ab which is determined by p and o. Congruences Having established the laws of divisibility for ideals. and ab is their least common multiple.p (mod a). If a is the greatest common divisor of the two ideals a. where p.5) the system of two congruences w . or by their corresponding classes modulo a and b respectively. each class r (mod ab) is determined in this way by precisely one combination p (mod a). Elements of the theory of ideals 9. b. Since o is the greatest common divisor of two relatively prime ideals a. We now say that the number p is prime to the ideal a when op and a are relatively prime ideals. One easily derives the theorem that 1'(ab) = V)(a)V)(b) for two relatively prime ideals a. along with the rules that allow m. We leave the task of finding the proof of this proposition to the reader. we shall add some considerations on congruences which are important for the theory of ideals. Conversely. or are two given numbers in o.

The total number of mutually incongruent numbers modulo p7z is. if the greatest common divisor D of a. Since this number is equal to [N(p)]m-1. It is evident that it will also be the same if we replace 77 by a number i' . we arrive.N(p) / whence we conclude immediately. Nl ) (p)/ where the product fl is taken over all the distinct prime ideals p that divide the ideal a. Congruences 135 the function vi(a) in the case where a is a power pm of the prime ideal p. oi' then 7 rlw. if 77a' is the least common multiple of a and oil then we have a = Da'. just as in the theory of rational numbers. (p. by virtue of the preceding theorem. r/ r/'w' (mod a). in the case m > 0.t at the theorem O(d) = N(a). t See Dirichlet's Vorlesungen fiber Zahlentheorie. and r/a' will be the least common multiple of a and oi'. o?7 is also that of a. pm) = (p. p) (p. pm) = (o. pm)N(p) It is necessary to subtract from this the number of numbers not prime to p"°. where D is the greatest common divisor of a and orj. that V(a) = N(a) 1. And conversely.[N(p)]m-1 = N(pm) l 1. and hence divisible by p. that is. It is also clear that the complementary factors D and a' of the ideal a remain the same for all numbers congruent to 77 modulo a. 2. Since we also have 'i(o) = 1. §14.iiw (mod a) where w is a number prime to a'. Conversely. If D is the greatest common divisor of the ideals a and orb then we have a = Da'. pm) = we get '0(pm) = [N(p)]m . where the summation is taken over all the ideals a' that divide a. §26. . a' will be the divisor of a corresponding to the number i (§19). equal to N(pm) = [N(p))m = (o.

corresponding to the divisor a' of a. )) Putting aa' = a. Elements of the theory of ideals whence we deduce 77ww' . by multiplying a by an ideal b' prime to a'. which are N(a) in number. as p remains constant and p' varies.E. . 136 Chapter 4. We must have X(d) = N(a) where the summation is taken over the divisors a' of a. the fact that z'(a') is nonzero means that there is always a number 77. Thus we have proved the following very important theorem: "If a and a' are any two ideals we can always.1 (mod a'). Since each product pp' of numbers p. it follows immediately that X(d) is never zero and always 0(a'). we deduce by the well-known method. pp' runs through a system of /1'(a) mutually incongruent numbers (mod a). To decide this alternative we consider all the incongruent numbers modulo a. corresponding to the divisors a'. 3. Q. the congruence pO(a) = 1 (mod a). However. it is necessary to point out that here we have assumed the existence of at least one such number 77. such that a will be the greatest common divisor of a and o17. which represents the highest generalisation of a celebrated theorem of t See Dirichlet's Vorlesungen fiber Zahlentheorie. §19. and consequently w is a number prime to a'. p' prime to an ideal a is likewise a number prime to a. and since. b' will be an ideal prime to a'. ww' . Thus the number of mu- tually incongruent numbers 77 modulo a which correspond to the same divisor a' of a is '(a'). Thus if we are given an arbitrary divisor a' of the ideal a the most we can affirm so far is that the number X(a') of mutually incongruent numbers 77 mod- ulo a which correspond to the same divisor a' will equal either 'O(d) or zero.t and for each value of the number p.77 (mod a). If we then put o77 = ab'.D. and partition them into groups of X(d) numbers. change it into a principal ideal OW = 077. But since we also have (1) E O(d) = N(a).

where p is the rational prime number divisible by p and f is the degree of the prime ideal p (§21. by well-known reasoning.-1) A(1. §26. Schonemann and Serret. ''2.3). of degree m. the congruence wPf = w (mod p). developed in the works of Gauss.. space not permitting us to pursue the general theory further. . 'y. = k2A(ul) then. are in o. we obtain a complete overview of all the ideals in the domain o by decomposing all ideals of the form op into their prime factors. combined with the preceding theorem.. This theorem is as important for the theory of the domain o as the theorem of Fermat is for the theory of rational numbers. 0. This proposition.. and if the coefficient of the highest degree term is not divisible by the prime ideal p then we deduce. if p is not a divisor of k we decompose op into prime ideals as t See Dirichlet's Vorlesungen fiber Zahlentheorie. leads to a complete theory of binomial congruence modulo p. we have previously ascertained a close connection between the theory of ideals and the theory of higher degree congruences. The theory of congruences provides a procedure capable of doing this in a great number of cases. as we shall at least try to make clear by the following remarks. . lyN(p)-2 are mutually incongruent.t Since all ideals are composed of prime ideals. §26. e2. the existence of primitive roots for the prime ideal p. re- stricted to the case of rational coefficients. meaning numbers ry whose powers 1. . t See my memoir Abriss einer Theorie der hoheren Congruenzen in Bezug auf einen reellen Primzahl-Modulus (Crelle's Journal. among other things. and each prime ideal p divides a determinate rational prime p. that is. If 0 is an integer of the field SZ and 9n. However.. the theory of congruences of higher degree with rational coefficients carries over completely to functions F(x) whose coefficients are numbers in the domain o. Congruences 137 Fermat.t that the congruence F(w) = 0 (mod p) cannot have more than m mutually incon- gruent roots. 54). We de- duce. If the coefficients of the polynomial function F(x). In general. . For a prime ideal p we conclude easily that every number w in the domain o satisfies the congruence wN(p) = w (mod p). . Galois.

Such a case is actually encountered. If we reflect on the infinite variety of these fields 1.t and this is one of the reasons I was determined to build the theory of ideals. . P2 (t)." This theorem is still true. 1490.. consisting of the integers of a field SZ of finite degree.. 138 Chapter 4. it is undoubtedly within the spirit of geometry to ascertain those general laws obeyed by the various theories without exception. where P1 (t). using the example that led Kummer t See the Gottingische gelehrte Anzeigen of 20 September 1871. and to combine it in particular with the algebraic principles of Galois.. not on congruences of higher degree. when the numbers k corresponding to all possible numbers 0 are all divisible by p... f2.. 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. I shall simply try to show. fe. .. . Instead. . the phenomena of divisibility of numbers in each domain o. although the proof is more difficult. If f (t) is the polynomial of degree n in t that vanishes for t = 0 we can put f (t) = Pl(t)a1P2(t)a2 . However. f2i . it is also of practical value. have been reduced to the same fixed laws that govern the old theory of rational numbers. . p2. the fundamentals of which I have devel- oped above. Pe(t)ae (mod p). ... fe. p.. Pe (t) are distinct irreducible polynomials with respective degrees fl. . §27. each of which has its own theory of special numbers. pe where p1. 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.. Pe are distinct prime ideals with respective degrees fl. this is not only of aesthetic or purely the- oretical interest. The knowledge that these general laws exist greatly assists in the discovery and proof of special phenomena in a given field Q. Elements of the theory of ideals follows. To back up this claim to the full we would admittedly need to develop further the general theory of ideals.. Examples borrowed from circle division By the general theory of ideals.. Then we certainly have op = p1 p22 .

)]m-1 However... N(p) must be a power of m. Thus if we make the abbreviation we get m = Eµ.. and since n < m ..1 = 0(m). m" = [N(...f(t) = tt 1 = (t . 0--1 are all roots of this equation we have B. §27. 2.(1 The m . Since the coefficients are rational.. om = op".1.. that is.or 1 + er + 92r + + 1 9T 1 will also be an integer. If we put N(p) = me it follows that n = e(m . Examples borrowed from circle division 139 to introduce his ideal numbers in the first place. then 1-9 1-9''e _ e(s-1)r = . since 0. Let m be a positive rational prime number. by taking norms.1 factors of the right-hand side are integers and associates of each other. . . from a primitive root of the equation 9' = 1. in the manner described above (§15). . the numbers 0. m ..+92+e+1=0.1)..1 (mod m). how the elements of the theory explained above lead to the goal with great facility.0 .1 isomorphisms transforming the normal field 1 into itself.+er. and if s is positive and chosen so that rs .9)(t . since m is a prime number. Moreover.. At the same time we have N(p) = m..92) . we conclude that e = 1 and n = m . because if r is one of the numbers 1. .1 by the remark above.02'... a root of the equation f(o)=om-1+em-2+. . and let Il be the field of degree n resulting. and these numbers correspond to m ..n-1). we have n < m . 1 1-e will be an integer. .-1 where a is a unit of the field Il and hence. and consequently m=(1-9)(1-02).1 then 1-or =1+0+0 +.. 9m-1 are conjugate.. (t - where t is a variable. The preceding equation f (9) = 0 is therefore irreducible.

1) = m it follows from §17 that 0(l. where a and b are ideals different from o. kl.. k1. Because an w divisible by m must also be divisible by µ. Elements of the theory of ideals The principal ideal op is a prime ideal..p it is clear that the two modules [1. 0. it follows that m = N(a)N(b).. . k2. and this makes ko divisible by µ.2'. µ.. 7a-2) _ (-1) m.. . . 9m-2] and [1. At the same time (§21... 2. With the help of this result it is easy to show that the m .. µ. km-2 are divisible by m. each number in the field Q can be put in the form ko + kip + k2p2 + . and therefore k cannot contain prime factors other .. 9m-2 form a basis for the domain o of all integers in the field Q. the numbers 0. and this makes k1 divisible by p.1 form a complete system of mutually incongruent numbers modulo A. 9. and hence also by m. . 140 Chapter 4. whence it follows (§4.. . ... we conclude that the other numbers k1. k1i k2.... for w to be divisible by k. that is. where ko. . if we have op = ab. p. + km-2µm-2. contrary to hypothesis. For this number to be an integer.. N(b) = 1.. and hence also not by µ7a-1.. . N[f'(9)] = and since N(9) = 1 and N(9 . It follows from this that a number of the form w = ko + kip + k2µ2 + .. unless all the numbers ko. km-2 are integers. Continuing in this way. m . 9. km-2 denote rational integers without common divisor. it follows. . Since we have mom_1 = (9 tm . and since m is a prime number we necessarily have. by excluding the uninteresting case m = 2. k2. km-2 are divisible by m. Since the numbers 1. is not divisible by m. µm-2] are identical. .. . and hence also by µ2.9. . . it is necessary (§18) that k2 divide the discriminant of the basis 1.. whence b = o..1 numbers 1.. since p = 1 .3 and §17. 0 = 1 . 1 mm_2.1)f (t). 1.. ... . µm-2 are independent..(5)) that we also have . N(a) = m. say. µ7a-2. µ2. 9m-2) _ (-1) 21 mm-2_ Moreover. .. 92. 02'... In fact. that mm-2. . . .+ km-2µ"Z-2 w k k where k.3) m is the smallest rational number divisible by p..1 = (t . ko. p2.1)f'(9).. .

x. where f is the degree of the prime ideal p. in which case or cannot be congruent to 93 (mod p). L. + 9m. But since all integers in the field Il have the form w = F(9) = x19 + X2 02 + + x7.F(9P'") (mod p). .. 0...3) 9N(p) = 0 (mod p). and we have N(p) =pf. since it has been shown that w is not divisible by m unless the numbers ko. µm_2] = [1.. As we know. that is. are rational integers.. §27. Now let p be any prime ideal different from oµ. whence we have o = [1. true for every prime number p. that wP . .. wP' .. 0 0(Q) = (_1)--r-'Mm-2.n-2 are divisible by m. .. . f must be divisible by a. let a be the least positive exponent for which pa = 1 (mod m)._19"°-1 where x1i x2. This is because r 0 s (mod m) implies 0' . and hence f > a.w (mod p). it follows that pf 1 (mod m). Moreover.. em-2].. and hence wP a .. . + km-2µm-2. and all integers of the field are of the form w = ko + k1µ + k2µ2 + . or again. .. that is. k1. k . The positive rational prime p divisible by p is different from m. Let a be the divisor of O(m) = m .9''-S) = ep where e is a unit. . Now since we have (§26. Thus we must have k = ±1. . Examples borrowed from circle division 141 than the number m. Two powers 0'. 09 are congruent modulo such a prime ideal only if they are equal.... if r s (mod m)..03 = or (1.. since 1 + 0 + 02 + . .1 to which the number p belongs modulo m.F(OP).-2 + Om_ 1 = 0.. k must also not be divisible by m. it follows from well-known theorems.

so w2 and hence also would be divisible by p2g2 = pq.1. m that are prime to m.. the congruence kf . the ideals of a normal field Sl immediately allow us to find the ideals of an arbitrary subfield $ of fl. (Translator's note.pe.. Again let m be a prime number. so ¢(m) = m . and let e be any divisor of m -1 = e f . and the determination of the prime ideals that divide m does not present any extra difficulty. any field H whose members all belong to Q.. then all the numbers F(9) in this field satisfying the t Kummer's researches may be found in Crelle's Journal. as in his footnote in §15. then P1p2. every integer w in the field Sl satisfies the congruence wra . because if op = p2q there would be a number w divisible by pq but wpa not by p. to give a more precise idea of the scope of these researches.. XVI and in the memoirs of the Berlin Academy for the year 1856. The preceding law is proved without any change. thus implying f < a. t Dedekind calls it a "divisor". I mention the following case. op = where pl. contrary to the preceding congruence. If 0 is again a primitive root of the equation 0 = 1 and if Sl is the corresponding field of degree m ." The rest follows easily. Elements of the theory of ideals We conclude first of all that the ideal op is a product of distinct prime ideals. 142 Chapter 4. that is. which therefore has N(p) = pf mutually incongruent roots w. so that O(m) = e f for some e. form a group. And since its degree is pa we must have pf < pa.3. pe are distinct prime ideals of degree f .. in that sense. In the theory of rational numbers..1 (mod m) has precisely f mutually incongruent roots h. .w (mod p). . 2. For example.. But it has already been shown that f > a. The degree of the normal field Sl is always equal to the number O(m) of those numbers among 1. since p is divisible by p. this enables us to know the ideals of any field H resulting from division of the circle and.. p2. so f = a.) . which are closed under multiplication and which. 35. 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. The general case where m is an arbitrary composite number can be treated similarly. and then also by p. in Liouville's Journal. We have therefore arrived at the following result. From the very general researches that I am going to publish shortly. Moreover.1.

according as pmt 1= +1 or . But we can directly study all quadratic fields. 343. . f = 2 In this case the f numbers h are the quadratic residues of m. without recourse to divi- sion of the circle. and ep is a prime ideal. ep is the product of two prime ideals of degree 1. In the case e = m . 0(H) (77 77 / because 77 + 77' = -1. and hence its discriminant will be 2 1/ 11 -771)2.. If we put p = H(1 . 17e are sums of f terms.1 (mod m).1. f = 1. §27. and if pf belongs to the exponent f modulo m.. m is an associate of pe. If we let k denote a quadratic nonresidue. by conclusions like those derived above in the case e = m . Examples borrowed from circle division 143 conditions F(9) = F(9h) form a field H of degree e. Moreover. then f will necessarily be a divisor of e = e' f' and the principal ideal ep will be the product of e' distinct prime ideals of degree f'. using the notation of Legendre. one of which is 9h and they form a basis of the domain a of integers in H. H is identical with Q. and we have already (§18) determined the discriminant D' of such a field H. We now examine more closely the case e = 2. that is. 772. By general results I shall speak of later (or else immediately. then the two conjugate periods 77=1: 9h =1: 9k 7 form a basis of the domain e of all integers in the quadratic field H. according as (m) = +1 or . Moreover. art.1) we now obtain the prime ideals belonging to this subfield H of the normal field Q. then p is an integer of the field H. or else ep is a prime ideal of degree 2. . . if p is a rational prime different from m.9k). From D' we can also derive the prime idealst of § Disquisitiones Arithmeticae.1. § 168. and we again obtain the result proved above. The number m is an associate of the square of the number p = 11(1 . f See Dirichlet's Vorlesungen fiber Zahlentheorie.9h). and the e conjugate periods§ 771. and ep is a prime ideal.

i'. art. Comparing these laws. from entirely general principles and without cal- culation. with the result for the special field H derived using circle division.t Pursuing this comparison further. we see first that D' must be divisible by m. previously demonstrated in the theory of circle division by effective for- mation of the square of i . t Disquisitiones Arithmeticae.. we are led again to the theorem m) = (:. t later reproduced in different forms by Jacobi.:p )' where ±m = 1 (mod 4). If p does not divide D'. or else a prime ideal of degree 2. Elements of the theory of ideals the field H. t Theorematis fundamentalis in doctrina de residuis quadraticis demonstrationes et ampliationes nov&. 1817. or else a prime ideal of degree 2. valid for all quadratic fields.1 or 5 (mod 8). and hence that (§18) 0(H) = D' = (-1)'. and to the theorem Cm) _ (-1) s. I should say that it was by meditating on the essence of that proof and the analogous proofs of cubic and bi- quadratic reciprocity that I was led to the general researches mentioned above and soon to be published. If the rational prime p divides D' then the corresponding principal ideal ep will be the square of a prime ideal. p Finally. then ep will be the product of two prime ideals of degree 1. the result (77 -77')2m. Eisenstein and others. 144 Chapter 4. according as D' . is essentially the same as the celebrated sixth proof of Gauss. in which we also determine the quadratic character of the number -1. . In this way we derive. 356. but not by any other prime number. and if p is odd. according as D' =+1or -1.2 m. ma-1 This proof of the quadratic reciprocity law. if D' is odd and hence .1 (mod 4) then e(2) will be the product of two prime ideals of degree 1.

and since the numbers a. Now if p . In fact. we must have N(a . if ao is a member of the ideal a with minimum norm.1 (mod 4). The discriminant of the field is 2 1 i = -4 1 -i The number 2 = i(i . y are rational integers (§6). of the form w=x+yi. whence it follows immediately that op is the product of two different prime ideals of degree 1.D. ao. This is because (§6) we can choose the integer w so that N(a .3 (mod 4) we have 2 wP=w WP -w (modp). belong to the ideal a. since op is the product of two prime ideals of degree 1 in the case where p is a rational prime =. which is the celebrated theorem of Fermat.wao) = 0. We then have 0 = i = and the integers of the quadratic field 1 are the complex integers.w (mod p). and consequently WP = (x + yi)P .E.1 (mod 4). where w' is the number conjugate to w.1)2 is an associate of the square of the prime number 1 . it follows that p = N(ao) = N(a + bi) = a2 + b2. then each number a in the ideal will be divisible by ao. we consider the case m = 4. and we conclude easily that op is a prime ideal of degree 2. where x.i. first introduced by Gauss. and hence also a-wao. Now. Q. §27. But every ideal a of this field must be a principal ideal. each integer w will satisfy the congruence wP . If p is a positive odd rational prime then we have iP = (-1) P21 i. whence a = wao and consequently a = oao. . Examples borrowed from circle division 145 As a final example. But if p .wao) < N(ao).x + (-1) 2 yi (mod p).

and it follows that (a". Classes of ideals We now return to consideration of an arbitrary field S2 of degree n. am = oµ' by multiplication by the same ideal m.5) that each ideal a can be converted into a principal ideal by multiplication by an ideal m. /3'. a". b" are in B there are four numbers a'. since there are two numbers p. We call this class the principal class and denote it by 0. This is because the hypothesis yields four numbers µ. equivalent to a will be called a class of ideals. it follows that oµ77' = i7'am = 77a'm. And if an ideal a is equivalent to a principal ideal. or the class composed from A and B. Because. We evidently have AB = BA. the system A of all ideals a. a' are certainly equivalent. µ'. We denote the class K to which all the products ab belong by AB and call it the product of A and B.D. and it follows from the equation . a" are equivalent to a third.D. 77. op'r7 = r7a'm. /3"b' = Q'b". then a' and a" are equiv- alent to each other. and b runs through all the ideals in a class B. multiplying a by m to obtain a principal ideal am = oµ. Because if a'.Q")(a'b') = (a'/')(a"b"). in order to establish the distribution of its ideals into classes. If two ideals a'. and hence they are all equivalent. Q.. if such that 77'a = r7a' then the ideals a. 146 Chapter 4. 77"a = rra".. a". It is clear that the system of all principal ideals forms a class by itself. Elements of the theory of ideals §28. Any two ideals in A are equivalent. whence pi7' = µ'r7. a' are called equivalent when they can be converted into principal ideals am = op. and on the following definition: two ideals a. Now if a runs through all the ideals in a class A. and a will be called a representative of the class A.E.E. whence p = µ'µ" and consequently a = op". and hence equivalent to o. and any ideal a' in A can be chosen as a representative in place of a. hence a'm = oµ'. a'b' and a"b" are equivalent ideals. If a is a given ideal. and hence (r7" µ)a' = (p'77)a". then a itself must be a principal ideal. Q. Then we evidently have µ'a = µa' and. It follows that the ideals can be partitioned into classes. Then p77' is divisible by 77. µ' such that µ'a = op and this implies that p is divisible by p'. then all the products ab belong to the same class K. a. that is. Thus the class represented by o includes all principal ideals and no others. conversely.. a'. /3" such that a"a' = a'a". This distribution depends first of all on the theorem (§25.77 " such that µ'a = pa'. a" are in A and b'. since each of them is converted to itself when multiplied by the ideal o. if there are two nonzero numbers q.

B. w2. which we can denote simply by A. Finally. and only one.. .. which are imaginary. Am. (Ar)s = Ars.. wn forming a basis for the field Il. h2i .. and show that. a2i . . . A2 A. (AB)r = ATBr. In addition. am represent the classes A1. The following two cases are particularly important: The equation oa = a yields the theorem that OA = A for any class A. because if the class N is such that AN = 0 it follows that N = NO = N(AM) = M(AN) = MO = M. .. the order in which pairs of classes are combined has no influence on the final result. then their product is called the mth power of A. If all the m factors equal A. s: ArAs = Ar+s. . C. then each number w=hlwl+h2w2+--+hnwn with rational integer coordinates hl. 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. hn will be an integer in the same field. Conversely.. represents the class Al A2 Am.. . . §29.. it is evident that from AB = AC we can always deduce B = C. it is clear that A will be the class inverse to A. are all < rk where r is the sum of the absolute values or moduli of w1. The number of classes of ideals 147 (ab)c = a(bc) that (AB)C = A(BC) for any three classes A. A2. ... and we denote it by A'. The class M is called the class opposite or inverse to A. . in the composition of any number of classes Al. 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. Moreover. §29. w2.. then it is evident that the absolute values of the corresponding numbers w. wn. If we allow the coordinates to take all integer values of absolute value not greater than a particular positive value k. . Then we can apply the same reasoning as for the multiplication of numbers or ideals. which are real. a2 an... or their analytic moduli. multiplying by A-'. Am then the ideal a.. and we denote it by A. If the ideals al. . we put Al = A and A° = O. The number of classes of ideals If we take any n integers W1. A2..

. where m is an ideal in the class M. Q. hn to take all k + 1 values 0.. then we obtain distinct numbers w and. Moreover. . where s is likewise a constant depending only on the basis. a = blwl + . . it follows that N(m) < s. Hence their difference a = (bl . .. and hence ±N(a) = N(a)N(m) < skn. k. 2. . and let k be the positive rational integer determined by the conditions kn < N(a) < (k + 1)n. We deduce from this the following theorem: In each class M of ideals there is at least one ideal m whose norm is bounded by a constant.. Proof. 2. 1.8). c of the numbers .D. h2i ..E. Since a is divisible by a we have oa = am.c of a all have absolute value not greater than k. is divisible by only a finite number of ideals (§25. If we now consider that the norm m of an ideal m is always divisible by m (§20). which are congruent modulo a. since their number is (k+1)n and hence > N(a). + brawn 'Y = C1w1 + . the coordinates b .. . + Cnwn. there cannot be more than a finite number of ideals m with N(m) < s.. we also have ±N(w) < skn.. there are necessarily two different numbers w.Q. If we now allow each of the n coordinates hl. Take any ideal a in the inverse class M-1. 'y come from the sequence 0. 148 Chapter 4... since the norm N(w) is a product of n conjugate numbers w of the form above. because each ideal.. Since there are also only a finite number of rational integers m not exceeding a given constant s. Elements of the theory of ideals and hence a constant independent of k. it is clear that there cannot be more than a finite number of ideals m with given norm m. k..Cl)wl + . 1. which evidently yields the fundamental theorem: . since kn < N(a). and hence ±N(a) < skn. But since the coordinates b. Moreover. and in particular om. + (bra .Cn)wn will be a nonzero number divisible by a.

we know that the problem was first solved by Dirichlet.. art. the product over all prime ideals p. however. § Dirichlet. for infinitely small positive values of the independent variable s . expressed in the terminology of the theory of ideals. t Crelle's Journal. XVI. 21. 305-307. Conclusion We shall derive some further interesting consequences of the fundamental theorem proved above. whose theory essentially coincides with that of binary quadratic forms..1. A. t Crelle's Journal. The h + 1 powers 0. With the aid of these principles.A2. 40.A' t Crelle's Journal. and the identity of the two expressions is an immediate consequence of the laws of divisibility. N(P).§ we are nevertheless far from the complete solution. For example. The sum is taken over all ideals a. 19. . and let A be a particular class. and by Kummer$ for the higher degree fields arising from division of the circle. The exact determination of the number of classes of ideals is incon- testably a very important problem. algebra and other parts of number theory. These researches have excited the liveliest interest because of their astonishing connections with analysis. Liouville's Journal. the problem treated by Kummer is closely related to Dirichlet's proof of the theorem on primes in arithmetic progressions. and for the moment we are confined to studying new special cases. Vorlesungen fiber Zahlentheorie.t His solution. which can be considerably simplified with the aid of these researches. For quadratic fields. 28. (See Disquisitiones Arithmeticae. while part of this research has been successfully completed for an arbitrary field St. Conclusion 149 The number of classes of ideals of the field St is finite..) Let h be the number of classes of ideals of the field S2. but also one of the most difficult in the theory of numbers. rests on the study of the function 1 _7T 1 N(a)s -11 1. §167. the number of classes of forms or ideals has been later determined by Eisensteint for a particular case of a field of degree 3.Ah-i. §30.. §30. There is no doubt that further study of the general problem will lead to important progress in these branches of mathematics.

. and it follows from §13. . which does not in general belong to the field Q. Conversely.. BA 2'. by the general laws of divisibility (§25)... if m is the smallest positive exponent satisfying this condition it is easy to see that the m classes 0. That is. That is. h such that Ar+m = Ar. if a is an integer belonging to S2 and divisible by p. the m classes (B) B. 1. will either be identical or have nothing in common.. that oa is divisible by a. In fact. if B denotes any class.. so that a is a number in the ideal a. the ht' power of any ideal is a principal ideal.A--1 are all different. 150 Chapter 4. at the same time connected with a precise definition of ideal numbers. With this important theorem we come to see the notion of ideal from a new point of view. Moreover. such as the preceding (B) and the following (C) C. BA.A2. CA2. ah will be in ah. It follows that we have the theorem Ah=O for each class A. Thus the whole system of h classes is partitioned into g such complexes and. Then. ..2 that a is divisible by the integer p = a1. A. then ah will be divisible by µh = a1 and consequently (oa)h will be divisible by oat = ah.. if both include the same class BAr = CA8 then we have C = BAr-3. CA. we have h = mg. Elements of the theory of ideals cannot all be different. . whence it follows immediately that the m classes in (C) are the same as those in (B)... and more generally we have Ar = A8 if and only if r . . and any two complexes of m classes. Let a be any ideal and let ah = oat. and consequently Am=0. since each complex includes m different classes. hence there are two different exponents r and r + m > r in the sequence 0. and we say that the class A belongs to the exponent m. Thus the ideal .s (mod m). and hence divisible by the number al. then ah belongs to the class Ah. BA--1 will be all different. h.. CAm-1. the exponent m to which the class A belongs is always a divisor of the number of classes. Now if a is any ideal in any class A. 2. and hence to the principal class. Obviously Al-1 = A-1. Now if a denotes any number in the ideal a. from which we easily conclude.

32 in o satisfying the condition ala2 + /31/32 = 1. and 61 is in o. .130' = 6. an algebraic integer u is said to be an ideal number of the field SZ when there is a power p'. and let h be the number of classes of ideals.0.3' are likewise algebraic integers.3. For this reason we say that the number p.31 are likewise in o. o will be the greatest common divisor of oal.D. 00 = b. and it is a principal ideal if and only if p is an associate of an actual number in the field Q. . and at the same time an ideal a in the field St satisfies the condition aT = op.Q are nonzero. j3' are integers satisfying the condition aa' +. we can put 0. Let o be the domain of integers of this field.Ql6h-1. The latter ideal a is the ideal corresponding to the ideal number p. and that it corresponds to the ideal a. since the number 1 is in o.31 ah = a161. then the integer 6 will be a common divisor of a and . Or. Conclusion 151 a consists of all the integers in SZ divisible by the integer p. Also. Dal = ah. Then it is easy to see that there is a field ci of finite degree including both a. otherwise the theorem is evident. = bh.(3h are divisible by 61i and hence.02/3h-1 =. where a' and .E.) We end our considerations with the proof of the following theorem announced earlier (§14): Any two algebraic integers a. there will be two numbers a2.3h. . where a'. we can put a2ah-1 = a'6h-1. (See the Introduction and §10. Proof. since ah. 0/31 and.32 = 61. ah = 0151. since h > 1. Now put oa = act. 0 have a common divisor 6 which can be expressed in the form 6 = aa' + 30'. o.3. with positive integer exponent r. If we now put 61 = 6h. is an ideal number of the field ci. where al. a little more generally. Q. though not actually a member of Q. §30.13h are divisible by oh. . . equal to an actual rl in St. since a and b are relatively prime ideals. or aha2 +. Since ah. where a is the greatest common divisor of oa. . Nh = )3161. 3. We assume that the two numbers a.

and the converse is equally valid. then the number 6. and two such numbers enjoy the characteristic property that any number p divisible by a and /3 is also divisible by a/3. If 6 is a unit then a. /3 are both nonzero.152 Chapter 4. Elements of the theory of ideals If at least one of the two numbers a. /3 is nonzero. and any of its associates. . This is because the equations µ = aa" = /3/3" and 1 = aa' + /3/3' imply p = a/3(a'/3" +. since a.3'a"). 3. deserves the name greatest common divisor of a. /3 may be called relatively prime.

55 circle division. 20 Arithmetica class number. 140 of incongruent numbers. 38 closure properties. 54. 20 residues. 106. 54. 54. 3. 42. 56 divisibility. 3. 14. 149 associativity of cubic field. 85. 54 modulo an ideal. 41 of fields. 149 automorphism of number field. 20 Brahmagupta Gauss definition. 28 Legendre definition. 61. 41 of quadratic fields. 108 complete system of cyclotomic integers. 20 153 . 20 of cyclotomic field. 84. 19. 4. 46 Dirichlet formula. 53. 39 principal. 39 ideal number. 84 commutativity of. 56 norm. 61 of Diophantus. 144 asociativity of. 4. 26. 20 identity. 150 composite binomial theorem Gaussian integer. 55 of rational integers. 84 definition. 55. 55. 42 associates. 33 congruent. 106 finiteness. 36 rational integer. 41. 20 character adjoined element. 40. 103 cubic. 8 determination of. 35 of Q(/). 11. 83 biquadratic composition reciprocity. 20 coordinates with respect to. 105 modulo a module. 64 greatest common divisor. 118 prime. 111 class group. 84 belonging to an exponent. 55 class definition. 42. 54. 103 of ideals. 31. 18. 146 conjugate. 97. Index abelian group. 65 of module. 53. 42 commutativity basis of composition. 94 biquadratic. 152 number. 61 units. 122 relatively prime. 67 complex integers. 85 mod p. 28. 122 of field. 103 quadratic. 138 decomposable. 38 algebraic integers. 53. 149 of quadratic forms. 149 of composition. 152 modulo an algebraic integer. 108 of representatives. 149 Artin. 56 Baker. 106 of Gaussian integers. 86 algebraic number.

63 coordinates of quadratic integers. 120 continuity. 14. 121 behaviour as a. 3 x4 + y4 = z2. 40. 12. 17. 40 numbers in quadratic field. 60 field. 112 roots of. 84 discriminant. 4 of ideals. 42. 31. 53. 56 ideals. 89 cubic ideal. 6. of quadratic integers. 56 of a form. 139 Dedekind. 58 of modules. 85 87. 7. 110 of Gaussian field. 30. 149 of rational integers. 64 Dirichlet. 18. 145 ideals. 149 of algebraic integers. 7. 116 Disquisitiones number. 5. 32. 4. 124 Euclid. 55 Elements Dedekind of Euclid. 19. 98. 89. 8 definition of algebraic integer. 105 continued fraction. 42 modulo an ideal. 98 formula for Pythagorean triples. 56. 149 elementary transformations. 5 field. 33. 135-137. 149 modulo an algebraic integer. 125. 44. 108 of rational integers. 10 algorithm for gcd. 141 and norm. 31. 8 higher order. 137 determines quadratic field. 31. 30 invention of ideals. 53 and Weber theory. 10 proof of two square theorem. 140 Eisenstein. 83 correspondence divisor. 6. 144 of an ideal. 55 of cyclotomic field. 146 of a prime ideal. 25 y3=x2+2. 45. 58 construction by nth power. 57 identity. 28 theory of ideals. 84 determinant. 139 integers. 41. 54. 23 modulo a module. 117 of a module. 121 theorem on primes. 102 Diophantine equation. 90 straightedge and compass. 21. 144. 63 reciprocity. 5 T+ X y3 = z3. 53 Descartes. 71 Elements. 29 domain. 37 x4 + y4 = z4. 40. 13 degree equivalent of a field. 10 . 61.21 section. 39 hagorean. 84. 97 of quadratic form. 53 of a quadratic form. 46 equation avoidance of symmetric functions. of Gaussian integers. 116 fields. 113 algebraic numbers.9. 54. 28. 6. 87 with respect to a basis. 87 143. 5 equation. 58 equivalence Supplement to Dirichlet. 20 of ideal factors. 110 of Gauss. 108 ideal numbers. 32. 112 of field. 118 class number formula. 119. 41 cyclotomic. 87 divisibility periods. 10 proof of quadratic reciprocity. 7 of numbers and ideals. 98. 97. 154 Index of forms. 8 congruence Arithmetica. 102. 60. 9. 143 by ideal number. 133 cyclotomic domain. 55 Vorlesungen. 77 decomposable numbers. 146 Diophantus. 42 Pell. 32 of algebraic integers. 7. 117 conjugate is rational. 32. 139 class number theorem. 29. 44. 14 in quadratic field. 8 of ideal classes. 100. 37 of quadratic forms.

29 of ideals. 38. 145 for quadratic integers. 96 discriminant of. 56 Galois. 16 norm. 29 of rational integers. 31 group. 110 class representative. 5. 137. 85 criterion. 121. 30 of algebraic integers. 117 classes modulo. 40. 125. 11. 19. 144. 84 last theorem for n = 3. 97. 123 finitely generated module. 116. 16 unique prime factorisation. 29. 116 definition. 10. 84. 31 abelian. 122 cyclotomic. 23 and x2 + 5y2. 95 prime factors. 44. 119 normal. 23 conjecture on x2 + 5y2. 126. 20 use of Pythagorean triples. 107 conjugate. 6 and two square theorem. 9 Euclidean algorithm. 146 cubic. 43. 37 prime. 108 defining properties. 82. 7. 13. 22. 38 Euclidean algorithm. 35 basis of. 56 fundamental number. 60. 120 definition of composition. 125 proofs of quadratic reciprocity. 84 numbers. 60. 56 ideal number of degree n. 111 ideals Galois theory. 116 fundamental properties. 22 class. 59. 134 proved quadratic reciprocity. 120 binary quadratic. 134 composition of forms. Index 155 Euclidean algorithm. 101. 77. 85 Euler. 37 relatively prime. 11 unique prime factorisation. 122 of degree n. 134 Disquisitiones. 121 definition. 61. 54 numbers. 33. 149 of quadratic integers. 108 ideal. 97. 5. 98. 121. 12. 150 quadratic. 120 Gauss. 86 in Gaussian integers. 60. 146 complex integers. 20 least common multiple. 116 divisors of. 100. 23 composite. 24 conjecture on x2 + 5y2. 109 class of. 106. 98. 17 greatest common divisor last theorem. 24. 17. 85 Fermat. 23. 61. 67. 146 conjugate. 142 primes. 98. 10 field Hilbert Zahlbericht. 130 sums. multiplication of. 63 notes on Diophantus. 6. 59 Galois. 13. 97. 139 congruence modulo. 106. 58 closure properties. 28 in Z. 41. 22. 121. 108 power of. 102. 125 127. 94. 9. 137. 77. 150 equivalent. 15. 57 of finite degree. 102. 7. 111. 152 last theorem for n = 4. 84 greatest common divisor. 144 equivalent. 57. 46 automorphism. 116 norm of. 26 congruence. 134 little theorem. 84 Gaussian integers. 22. 45 divisibility of. 17 laws of divisibility. 53. 97 degree of. 10. 11. 85. 85 in Z[V/--2]. 84. 60. 35. 85 conjectured quadratic reciprocity. 97. 90 field. 9 Gaussian primes. 137 of modules. 143 prime. 133 fundamental number of. 85 units. 56 forms principal. 20 two square theorem. 85 proof of two square theorem. 111 in field of degree n. 56 class. 145 class. 5. 138 of quadratic field. 39. 16 product of. 32 identity . 95 existence of primitive roots.

55. 118 irreducible multiple. 94 in Z[V ]. 8. 107 of module. 46 norm Jacobi. 110 greatest common divisor of. 63 mod p. 27 abelian group axioms. 103 linear algebra. 28 of Gaussian integer. 67. 98. 43 congruence modulo. 23 in rational integers. 86 Diophantus. 140 module. 121. 97. 111. 5 Kummer. 40. 142 of principal ideal. 143 integers symbol. 83. 18. 90. theory of fields. 62 Gaussian. 71 of ideals. 6 normal field. 39. 41. 5 divisor of. 71. 97 as a group. 30. 63 infinite descent. 33 algebraic. 21 in Zf / . 84. 94 is rational. 116 identity. 9. 111 ideal factors. 7 equation. 63 substitution. 18. 25 adjoined ideal numbers. 53 Lipschitz. 147 divisibility of. 98 irrational numbers. 94. 13 multiplication of. 86. 134 independent numbers. 32. 88 isomorphism. 22. 63 isomorphism. 26. 113 closure properties. 85. 15 algebraic. 44 notation. 22. 57 modulus. 82 imaginary quadratic. 40. 149 of algebraic number. 112 in field of finite degree. 34 identity. 144 and conjugates. 108 of modules. 44 multiplicative property. 105 polynomial. 107 multiplication irreducible system. 103 reduction process. 83 zero. 110 Newton. 110 least common multiple Lagrange. 41. 44 cubic. 63 inverse modules ideal class. 57 . 110 inverse. 110 multiplicative property fixes rational numbers. 138. 41 Noether. 18 number proof of two squares theorem. 95 quadratic. 54. 54. 53. 84. 63 rational. 122 of quadratic integer. 85 integers. 3. 87 Lagrange. 102. 116 finitely generated. 21 multiple of. 63 of quadratic field. 34 least common multiple of. 110 of quadratic character. 108 of modules. 12. 98 determined by choice of root. 28 in quadratic integers. 6. 4. 113 opposition to infinity. 64 of Q(a). 107 of algebraic integer. 122 main theorem. 67 in field of degree n. 25 basis of. 18 of ideals. 23. 41 Kronecker. 54. 11 Legendre infinity definition of composition. 55. 41 classes. 26. 60. 156 Index Brahmagupta. 35 cyclotomic. 110 in Gaussian integers. 20 horror of. 85 ideal. 40 onto number field. 41 complex. 43 of ideal. 125 isomorphism. 55 laws of divisibility conjugate. 63 system. 46 132 Kummer.

143 x2+y . 116 complete system. 101. 83 congruence. 17. 14 divisor property. 8 x2 + 5yy2. 54 reduction. 6. 97. 15. 16. 45 87 surfaces. 5 . 6. 87 ideal. 22. 17. 87 of quadratic field. 33 representative fields. 60. 144 Euclid's formula. 15. 143 representatives quadratic field. 83 class. 15 Pell equation. 86 has principal ideal multiple. 12. 36. 127 ideal. 21 in 2[(-'2 . 4. 28. 10 of a group element. 143 class number. 143 failure of unique prime factorisation. 7. 8 biquadratic. 12. 125 rational numbers Pythagoras. 144 units. 116 closure properties. 8. 12. 83 of ideals. 19. 117 quadratic. 146 operations. 7 discriminant of. 27 Gaussian. 23. 107 theorem. Riemann. 14 permutation. 6. 53. 60. 14. 89. 107 ideal class. 146 units. 22 prime ideal. 33 algebraic integers. 100. 149 residues conjugate in. 27 periods. 32 rational. 4 integers of. 127 quadratic forms. 146 reciprocity. 102 Plimpton 322. 8 fixed by isomorphism. 110 Pythagorean triples. 107 product prime. 144 primitive. 142 rational principal composite number. 15. 98 prime. 134 character symbol. 38 used by Fermat. 8 field of. 139. 23 equivalent. 33. 65 residues. 14 in algebraic integers. 143 of ideal class. 17. 37. 4. 26 and cyclotomic integers. 10 laws.25 conjugate. 26 x2 + 3y2. 57 prime ideals of. 102. 10 cubic. 137. 14. 56 Euclidean algorithm. 86 order. 83 ideal. 90 ring. 38.17. 15. 144 of number class. Index 157 irrational. 13 Fermat. 12. 144 ideals. 116 biquadratic. 130 norm. 36. 127 x2 + 2y2. 14 quadratic relatively prime character. 41 in o 5 quadratic integers. 15. 95 degree of. 152 character of -1. 16. 144 reduced quadratic forms. 108 composition of. 101 reduced. 85 orders of. 144 primitive root. 6 reciprocity and right-angled triangles. 46 ideal number of. 59. 31 inequivalent. 124 laws of divisibility. 10 quadratic. 16 2x2 + 2xy + 3y2. 65 class number. 56 integers. 16. 6 determinant of. 120 numbers.28 of quadratic forms. 126. 83 of ideal classes. 87 prime factors divisibility. 84 discriminant of. 123 ideals. 37.

28. 38 primitive. 45 and Dedekind. 5. 24 Dedekind proof. 145 Euler proof. 108 inverse. 9 and Gaussian primes. 28 failure in [v/-]. 5. 30 failure in Z[. 13 unimodular. 41. 14 symmetric functions. 137 Stark. 11 Fermat proof. 83 vector space. 54. 15 unimodular substitution. 142 substitution. 33. 27 failure in Z[C23]. 56 in Disquisitiones. 142 Schonemann. 113 Newton theorem. 85 in quadratic field. 42 subfield. 3. 58 Serret. 24. 137 of unity. 137. 32 failure in cyclotomic integers. 30 of ideals. 56. 22. 13 zero module. 25. 87 for ideals. 63 . 26 in Z[C3]. 23. 56 failure in quadratic field. 86 in rational integers. 14 unique prime factorisation and equivalence of forms.]. 46 Weil. 7 in Gaussian integers. 22. 41 Weber./. 137 section. 158 Index root of congruence. 40 two square theorem.]. 44. 11 Lagrange proof . 106 in Gaussian integers. 84 in Z[v/. 102 units in algebraic integers. 130 in complex integers. 85 in rational integers. 139.