This action might not be possible to undo. Are you sure you want to continue?
J.S. Milne
Version 3.02
April 30, 2009
An algebraic number ﬁeld is a ﬁnite extension of Q; an algebraic number is an element
of an algebraic number ﬁeld. Algebraic number theory studies the arithmetic of algebraic
number ﬁelds — the ring of integers in the number ﬁeld, the ideals and units in the ring of
integers, the extent to which unique factorization holds, and so on.
An abelian extension of a ﬁeld is a Galois extension of the ﬁeld with abelian Galois
group. Class ﬁeld theory describes the abelian extensions of a number ﬁeld in terms of the
arithmetic of the ﬁeld.
These notes are concerned with algebraic number theory, and the sequel with class ﬁeld
theory. The original version was distributed during the teaching of a secondyear graduate
course.
BibTeX information
@misc{milneANT,
author={Milne, James S.},
title={Algebraic Number Theory (v3.02)},
year={2009},
note={Available at www.jmilne.org/math/},
pages={156+viii}
}
v2.01 (August 14, 1996). First version on the web.
v2.10 (August 31, 1998). Fixed many minor errors; added exercises and an index; 138
pages.
v3.00 (February 11, 2008). Corrected; revisions and additions; 163 pages.
v3.01 (September 28, 2008). Fixed problem with hyperlinks; 163 pages.
v3.02 (April 30, 2009). Fixed many minor errors; changed chapter and page styles; 164
pages.
Available at www.jmilne.org/math/
Please send comments and corrections to me at the address on my web page.
The photograph is of the Fork Hut, Huxley Valley, New Zealand.
Copyright c (1996, 1998, 2008, 2009, J.S. Milne.
Single paper copies for noncommercial personal use may be made without explicit permis
sion from the copyright holder.
Contents
Notations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1 Preliminaries from Commutative Algebra 7
Basic deﬁnitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Ideals in products of rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Noetherian rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Noetherian modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Local rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Rings of fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
The Chinese remainder theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Review of tensor products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Rings of Integers 19
First proof that the integral elements form a ring . . . . . . . . . . . . . . . . . . 19
Dedekind’s proof that the integral elements form a ring . . . . . . . . . . . . . . 20
Integral elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Review of bases of ·modules . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Review of norms and traces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Review of bilinear forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Discriminants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Rings of integers are ﬁnitely generated . . . . . . . . . . . . . . . . . . . . . . . 29
Finding the ring of integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Algorithms for ﬁnding the ring of integers . . . . . . . . . . . . . . . . . . . . . 34
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3 Dedekind Domains; Factorization 40
Discrete valuation rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Dedekind domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Unique factorization of ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
The ideal class group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Discrete valuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Integral closures of Dedekind domains . . . . . . . . . . . . . . . . . . . . . . . 51
Modules over Dedekind domains (sketch). . . . . . . . . . . . . . . . . . . . . . 52
2
Factorization in extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
The primes that ramify . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Finding factorizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Examples of factorizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Eisenstein extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4 The Finiteness of the Class Number 62
Norms of ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Statement of the main theorem and its consequences . . . . . . . . . . . . . . . . 64
Lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Some calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Finiteness of the class number . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Binary quadratic forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5 The Unit Theorem 79
Statement of the theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Proof that U
1
is ﬁnitely generated . . . . . . . . . . . . . . . . . . . . . . . . . 81
Computation of the rank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Sunits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Example: CM ﬁelds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Example: real quadratic ﬁelds . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Example: cubic ﬁelds with negative discriminant . . . . . . . . . . . . . . . . . 86
Finding u(1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Finding a system of fundamental units . . . . . . . . . . . . . . . . . . . . . . . 88
Regulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6 Cyclotomic Extensions; Fermat’s Last Theorem. 90
The basic results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Class numbers of cyclotomic ﬁelds . . . . . . . . . . . . . . . . . . . . . . . . . 96
Units in cyclotomic ﬁelds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
The ﬁrst case of Fermat’s last theorem for regular primes . . . . . . . . . . . . . 97
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
7 Valuations; Local Fields 100
Valuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Nonarchimedean valuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Equivalent valuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Properties of discrete valuations . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Complete list of valuations for the rational numbers . . . . . . . . . . . . . . . . 104
The primes of a number ﬁeld . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
The weak approximation theorem . . . . . . . . . . . . . . . . . . . . . . . . . 108
Completions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Completions in the nonarchimedean case . . . . . . . . . . . . . . . . . . . . . . 110
Newton’s lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Extensions of nonarchimedean valuations . . . . . . . . . . . . . . . . . . . . . 117
Newton’s polygon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Locally compact ﬁelds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
Unramiﬁed extensions of a local ﬁeld . . . . . . . . . . . . . . . . . . . . . . . 122
Totally ramiﬁed extensions of 1 . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Ramiﬁcation groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Krasner’s lemma and applications . . . . . . . . . . . . . . . . . . . . . . . . . 126
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8 Global Fields 130
Extending valuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
The product formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
Decomposition groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
The Frobenius element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Computing Galois groups (the hard way) . . . . . . . . . . . . . . . . . . . . . . 139
Computing Galois groups (the easy way) . . . . . . . . . . . . . . . . . . . . . . 139
Applications of the Chebotarev density theorem . . . . . . . . . . . . . . . . . . 144
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
A Solutions to the Exercises 147
B Twohour examination 153
Bibliography 154
Index 155
Notations.
We use the standard (Bourbaki) notations: N={0. 1. 2. . . .]; Z =ring of integers; 1=ﬁeld
of real numbers; C = ﬁeld of complex numbers; F
;
=Z¡]Z = ﬁeld with ] elements, ] a
prime number.
For integers m and n, m[n means that m divides n, i.e., n ÷ mZ. Throughout the notes,
] is a prime number, i.e., ] =2. 3. 5. . . ..
Given an equivalence relation, Œ+ denotes the equivalence class containing +. The
empty set is denoted by 0. The cardinality of a set S is denoted by [S[ (so [S[ is the number
of elements in S when S is ﬁnite). Let 1 and · be sets; a family of elements of · indexed
by 1, denoted (a
i
)
i÷J
, is a function i ÷a
i
: 1 ÷·.
X cY X is a subset of Y (not necessarily proper);
X
def
=Y X is deﬁned to be Y , or equals Y by deﬁnition;
X ~Y X is isomorphic to Y ;
X .Y X and Y are canonically isomorphic (or there is a given or unique isomorphism);
·÷ denotes an injective map;
denotes a surjective map.
It is standard to use Gothic (fraktur) letters for ideals:
a b c m n p q A B C M N P Q
a b c m n ] q · T C M N 1 O
Prerequisites
The algebra usually covered in a ﬁrstyear graduate course, for example, Galois theory,
group theory, and multilinear algebra. An undergraduate number theory course will also be
helpful.
References
In addition to the references listed at the end and in footnotes, I shall refer to the following
of my course notes (available at www.jmilne.org/math/):
GT Group Theory, v3.01, 2008.
FT Fields and Galois Theory, v4.21, 2008.
CFT Class Field Theory, v4.00, 2008.
Acknowledgements
I thank the following for providing corrections and comments for earlier versions of these
notes: Vincenzo Acciaro; Michael Adler; Giedrius Alkauskas; Francesc Castell` a; Kwangho
Choiy; Dustin Clausen; Keith Conrad; Paul Federbush; Roger Lipsett; Loy Jiabao, Jasper;
Lee M. Goswick; Samir Hasan; Lars Kindler; Franz Lemmermeyer; Bijan Mohebi; Wai
Yan Pong; Nicol´ as Sirolli; Thomas Stoll; Vishne Uzi; and others.
PARI is an open source computer algebra system freely available from http://pari.math.u
bordeaux.fr/.
5
DRAMATIS PERSONÆ
FERMAT (1601–1665). Stated his last “theorem”, and proved it for m =4. He also posed
the problem of ﬁnding integer solutions to the equation,
X
2
÷·Y
2
=1. · ÷ Z. (1)
which is essentially the problem
1
of ﬁnding the units in ZŒ
·. The English mathemati
cians found an algorithm for solving the problem, but neglected to prove that the algorithm
always works.
EULER (1707–1783). He introduced analysis into the study of the prime numbers, and he
discovered an early version of the quadratic reciprocity law.
LAGRANGE (1736–1813). He found the complete form of the quadratic reciprocity law:
]
q
q
]
=(÷1)
(;1)(q1)¡4
. ]. q odd primes,
and he proved that the algorithm for solving (1) always leads to a solution,
LEGENDRE (1752–1833). He introduced the “Legendre symbol”
n
;
, and gave an incom
plete proof of the quadratic reciprocity law. He proved the following localglobal principle
for quadratic forms in three variables over Q: a quadratic form O(X. Y. 7) has a nontrivial
zero in Q if and only if it has one in 1 and the congruence O÷0 mod ]
n
has a nontrivial
solution for all ] and n.
GAUSS (1777–1855). He found the ﬁrst complete proofs of the quadratic reciprocity law.
He studied the Gaussian integers ZŒi  in order to ﬁnd a quartic reciprocity law. He studied
the classiﬁcation of binary quadratic forms over Z, which is closely related to the problem
of ﬁnding the class numbers of quadratic ﬁelds.
DIRICHLET (1805–1859). He introduced 1series, and used them to prove an analytic for
mula for the class number and a density theorem for the primes in an arithmetic progression.
He proved the following “unit theorem”: let ˛ be a root of a monic irreducible polynomial
}(X) with integer coefﬁcients; suppose that }(X) has r real roots and 2s complex roots;
then ZŒ˛
×
is a ﬁnitely generated group of rank r ÷s ÷1.
KUMMER (1810–1893). He made a deep study of the arithmetic of cyclotomic ﬁelds, mo
tivated by a search for higher reciprocity laws, and showed that unique factorization could
be recovered by the introduction of “ideal numbers”. He proved that Fermat’s last theorem
holds for regular primes.
HERMITE (1822–1901). He made important contributions to quadratic forms, and he showed
that the roots of a polynomial of degree 5 can be expressed in terms of elliptic functions.
EISENSTEIN (1823–1852). He published the ﬁrst complete proofs for the cubic and quartic
reciprocity laws.
KRONECKER (1823–1891). He developed an alternative to Dedekind’s ideals. He also had
one of the most beautiful ideas in mathematics for generating abelian extensions of number
ﬁelds (the Kronecker liebster Jugendtraum).
RIEMANN (1826–1866). Studied the Riemann zeta function, and made the Riemann hy
pothesis.
1
The Indian mathematician Bhaskara (12th century) knew general rules for ﬁnding solutions to the equa
tion.
DEDEKIND (1831–1916). He laid the modern foundations of algebraic number theory by
ﬁnding the correct deﬁnition of the ring of integers in a number ﬁeld, by proving that ideals
factor uniquely into products of prime ideals in such rings, and by showing that, modulo
principal ideals, they fall into ﬁnitely many classes. Deﬁned the zeta function of a number
ﬁeld.
WEBER (1842–1913). Made important progress in class ﬁeld theory and the Kronecker
Jugendtraum.
HENSEL (1861–1941). He gave the ﬁrst deﬁnition of the ﬁeld of ]adic numbers (as the set
of inﬁnite sums
¸
o
n=k
a
n
]
n
, a
n
÷ {0. 1. . . . . ]÷1]).
HILBERT (1862–1943). He wrote a very inﬂuential book on algebraic number theory in
1897, which gave the ﬁrst systematic account of the theory. Some of his famous problems
were on number theory, and have also been inﬂuential.
TAKAGI (1875–1960). He proved the fundamental theorems of abelian class ﬁeld theory,
as conjectured by Weber and Hilbert.
NOETHER (1882–1935). Together with Artin, she laid the foundations of modern algebra
in which axioms and conceptual arguments are emphasized, and she contributed to the
classiﬁcation of central simple algebras over number ﬁelds.
HECKE (1887–1947). Introduced Hecke 1series generalizing both Dirichlet’s 1series and
Dedekind’s zeta functions.
ARTIN (1898–1962). He found the “Artin reciprocity law”, which is the main theorem of
class ﬁeld theory (improvement of Takagi’s results). Introduced the Artin 1series.
HASSE (1898–1979). He gave the ﬁrst proof of local class ﬁeld theory, proved the Hasse
(localglobal) principle for all quadratic forms over number ﬁelds, and contributed to the
classiﬁcation of central simple algebras over number ﬁelds.
BRAUER (1901–1977). Deﬁned the Brauer group, and contributed to the classiﬁcation of
central simple algebras over number ﬁelds.
WEIL (1906–1998). Deﬁned the Weil group, which enabled him to give a common gener
alization of Artin 1series and Hecke 1series.
CHEVALLEY (1909–84). The main statements of class ﬁeld theory are purely algebraic,
but all the earlier proofs used analysis; Chevalley gave a purely algebraic proof. With his
introduction of id` eles he was able to give a natural formulation of class ﬁeld theory for
inﬁnite abelian extensions.
IWASAWA (1917–1998). He introduced an important new approach into algebraic number
theory which was suggested by the theory of curves over ﬁnite ﬁelds.
TATE (1925– ). He proved new results in group cohomology, which allowed him to give
an elegant reformulation of class ﬁeld theory. With Lubin he found an explicit way of
generating abelian extensions of local ﬁelds.
LANGLANDS (1936– ). The Langlands program
2
is a vast series of conjectures that, among
other things, contains a nonabelian class ﬁeld theory.
2
Not to be confused with its geometric analogue, sometimes referred to as the geometric Langlands pro
gram, which appears to lack arithmetic signiﬁcance.
7
Introduction
It is greatly to be lamented that this virtue of
the [rational integers], to be decomposable into
prime factors, always the same ones for a given
number, does not also belong to the [integers of
cyclotomic ﬁelds].
Kummer 1844 (as translated by Andr´ e Weil)
The fundamental theorem of arithmetic says that every nonzero integer m can be writ
ten in the form,
m=˙]
1
]
n
. ]
i
a prime number,
and that this factorization is essentially unique.
Consider more generally an integral domain ·. An element a ÷ · is said to be a unit if
it has an inverse in · (element b such that ab =1 =ba). I write ·
×
for the multiplicative
group of units in ·. An element ¬ of · is said to prime if it is neither zero nor a unit, and if
¬[ab == ¬[a or ¬[b.
If · is a principal ideal domain, then every nonzero element a of · can be written in the
form,
a =u¬
1
¬
n
. u a unit. ¬
i
a prime element.
and this factorization is unique up to order and replacing each ¬
i
with an associate, i.e.,
with its product with a unit.
Our ﬁrst task will be to discover to what extent unique factorization holds, or fails to
hold, in number ﬁelds. Three problems present themselves. First, factorization in a ﬁeld
only makes sense with respect to a subring, and so we must deﬁne the “ring of integers”
O
1
in our number ﬁeld 1. Secondly, since unique factorization will fail in general, we
shall need to ﬁnd a way of measuring by how much it fails. Finally, since factorization is
only considered up to units, in order to fully understand the arithmetic of 1, we need to
understand the structure of the group of units U
1
in O
1
.
THE RING OF INTEGERS
Let 1 be an algebraic number ﬁeld. Each element ˛ of 1 satisﬁes an equation
˛
n
÷a
1
˛
n1
÷ ÷a
0
=0
with coefﬁcients a
1
. . . . . a
n
in Q, and ˛ is an algebraic integer if it satisﬁes such an equation
with coefﬁcients a
1
. . . . . a
n
in Z. We shall see that the algebraic integers form a subring O
1
of 1.
The criterion as stated is difﬁcult to apply. We shall show (2.11) that ˛ is an algebraic
integer if and only if its minimum polynomial over Q has coefﬁcients in Z.
Consider for example the ﬁeld 1 = QŒ
J, where J is a squarefree integer. The
minimum polynomial of ˛ =a÷b
J, b =0, a. b ÷ Q, is
(X ÷(a÷b
J))(X ÷(a÷b
J)) =X
2
÷2aX ÷(a
2
÷b
2
J).
and so ˛ is an algebraic integer if and only if
2a ÷ Z. a
2
÷b
2
J ÷ Z.
1
From this it follows easily that, when J ÷2. 3 mod 4, ˛ is an algebraic integer if and only
if a and b are integers, i.e.,
O
1
=ZŒ
J =
a÷b
J [ a. b ÷ Z
¸
.
and, when J ÷ 1 mod 4, ˛ is an algebraic integer if and only if a and b are either both
integers or both halfintegers, i.e.,
O
1
=ZŒ
1÷
d
2
 =
a÷b
1÷
d
2
ˇ
ˇ
ˇ a. b ÷ Z
¸
.
For example, the minimum polynomial of 1¡2÷
5¡2 is X
2
÷X ÷1, and so 1¡2÷
5¡2
is an algebraic integer in QŒ
5.
Let ¯
d
be a primitive Jth root of 1, for example, ¯
d
=exp(2¬i¡J), and let 1 =QŒ¯
d
.
Then we shall see (6.2) that
O
1
=ZŒ¯
d
 =
˚¸
m
i
¯
i
d
[ m
i
÷ Z
¸
.
as one would hope.
FACTORIZATION
A nonzero element ¬ of an integral domain · is said to be irreducible if it is not a unit, and
can’t be written as a product of two nonunits. For example, a prime element is (obviously)
irreducible. A ring · is a unique factorization domain if every nonzero element of · can
be expressed as a product of irreducible elements in essentially one way. Is the ring of
integers O
1
a unique factorization domain? No, not in general!
We shall see that each element of O
1
can be written as a product of irreducible elements
(this is true for all Noetherian rings), and so it is the uniqueness that fails.
For example, in ZŒ
÷5 we have
6 =2 3 =(1÷
÷5)(1÷
÷5).
To see that 2, 3, 1÷
÷5, 1÷
÷5 are irreducible, and no two are associates, we use the
norm map
Nm: QŒ
÷5 ÷Q. a÷b
÷5 ÷a
2
÷5b
2
.
This is multiplicative, and it is easy to see that, for ˛ ÷ O
1
,
Nm(˛) =1 ” ˛ ¨ ˛ =1 ” ˛ is a unit. (*)
If 1÷
÷5 =˛ˇ, then Nm(˛ˇ) =Nm(1÷
÷5) =6. Thus Nm(˛) =1. 2. 3, or 6. In the
ﬁrst case, ˛ is a unit, the second and third cases don’t occur, and in the fourth case ˇ is a
unit. A similar argument shows that 2. 3, and 1 ÷
÷5 are irreducible. Next note that (*)
implies that associates have the same norm, and so it remains to show that 1 ÷
÷5 and
1÷
÷5 are not associates, but
1÷
÷5 =(a÷b
÷5)(1÷
÷5)
has no solution with a. b ÷ Z.
Why does unique factorization fail in O
1
? The problem is that irreducible elements in
O
1
need not be prime. In the above example, 1÷
÷5 divides 2 3 but it divides neither 2
2
nor 3. In fact, in an integral domain in which factorizations exist (e.g. a Noetherian ring),
factorization is unique if all irreducible elements are prime.
What can we recover? Consider
210 =6 35 =10 21.
If we were naive, we might say this shows factorization is not unique in Z; instead, we
recognize that there is a unique factorization underlying these two decompositions, namely,
210 =(2 3)(5 7) =(2 5)(3 7).
The idea of Kummer and Dedekind was to enlarge the set of “prime numbers” so that, for
example, in ZŒ
÷5 there is a unique factorization,
6 =(p
1
p
2
)(p
3
p
4
) =(p
1
p
3
)(p
2
p
4
).
underlying the above factorization; here the p
i
are “ideal prime factors”.
How do we deﬁne “ideal factors”? Clearly, an ideal factor should be characterized
by the algebraic integers it divides. Moreover divisibility by a should have the following
properties:
a[0: a[a. a[b =a[a˙b: a[a =a[ab for all b ÷ O
1
.
If in addition division by a has the property that
a[ab =a[a or a[b.
then we call a a “prime ideal factor”. Since all we know about an ideal factor is the set of
elements it divides, we may as well identify it with this set. Thus an ideal factor a is a set
of elements of O
1
such that
0 ÷ a: a. b ÷ a =a˙b ÷ a: a ÷ a =ab ÷ a for all b ÷ O
1
:
it is prime if an addition,
ab ÷ a =a ÷ a or b ÷ a.
Many of you will recognize that an ideal factor is what we now call an ideal, and a prime
ideal factor is a prime ideal.
There is an obvious notion of the product of two ideals:
ab[c ” c =
¸
a
i
b
i
. a[a
i
. b[b
i
.
In other words,
ab =
¸
a
i
b
i
[ a
i
÷ a. b
i
÷ b
¸
.
One see easily that this is again an ideal, and that if
a =(a
1
. .... a
n
) and b =(b
1
. .... b
n
)
then
a b =(a
1
b
1
. .... a
i
b
}
. .... a
n
b
n
).
With these deﬁnitions, one recovers unique factorization: if a = 0, then there is an
essentially unique factorization:
(a) =p
1
p
n
with each p
i
a prime ideal.
3
In the above example,
(6) =(2. 1÷
÷5)(2. 1÷
÷5)(3. 1÷
÷5)(3. 1÷
÷5).
In fact, I claim
(2. 1÷
÷5)(2. 1÷
÷5) =(2)
(3. 1÷
÷5)(3. 1÷
÷5) =(3)
(2. 1÷
÷5)(3. 1÷
÷5) =(1÷
÷5)
(2. 1÷
÷5)(3. 1÷
÷5) =(1÷
÷5).
For example, (2. 1÷
÷5)(2. 1÷
÷5) =(4. 2÷2
÷5. 2÷2
÷5. 6). Since every gen
erator is divisible by 2, we see that
(2. 1÷
÷5)(2. 1÷
÷5) c(2).
Conversely,
2 =6÷4 ÷ (4. 2÷2
÷5. 2÷2
÷5. 6)
and so (2. 1 ÷
÷5)(2. 1 ÷
÷5) = (2), as claimed. I further claim that the four ideals
(2. 1 ÷
÷5), (2. 1 ÷
÷5), (3. 1 ÷
÷5), and (3. 1 ÷
÷5) are all prime. For example,
the obvious map Z ÷ZŒ
÷5¡(3. 1÷
÷5) is surjective with kernel (3), and so
ZŒ
÷5¡(3. 1÷
÷5) .Z¡(3).
which is an integral domain.
How far is this from what we want, namely, unique factorization of elements? In other
words, how many “ideal” elements have we had to add to our “real” elements to get unique
factorization. In a certain sense, only a ﬁnite number: we shall see that there exists a ﬁnite
set S of ideals such that every ideal is of the form a (a) for some a ÷ S and some a ÷ O
1
.
Better, we shall construct a group 1 of “fractional” ideals in which the principal fractional
ideals (a), a ÷ 1
×
, form a subgroup 1 of ﬁnite index. The index is called the class number
h
1
of 1. We shall see that
h
1
=1 ” O
1
is a principal ideal domain
” O
1
is a unique factorization domain.
UNITS
Unlike Z, O
1
can have inﬁnitely many units. For example, (1 ÷
2) is a unit of inﬁnite
order in ZŒ
2 :
(1÷
2)(÷1÷
2) =1: (1÷
2)
n
=1 if m=0.
In fact ZŒ
2
×
={˙(1÷
2)
n
[ m÷ Z], and so
ZŒ
2
×
~{˙1] ×{free abelian group of rank 1].
In general, we shall show (unit theorem) that the roots of 1 in 1 form a ﬁnite group u(1),
and that
O
×
1
~u(1) ×Z
i
(as an abelian group);
moreover, we shall ﬁnd r.
4
APPLICATIONS
One motivation for the development of algebraic number theory was the attempt to prove
Fermat’s last “theorem”, i.e., when m _ 3, there are no integer solutions (.. .. :) to the
equation
X
n
÷Y
n
=7
n
with all of .. .. : nonzero.
When m=3, this can proved by the method of “inﬁnite descent”, i.e., fromone solution,
you show that you can construct a smaller solution, which leads to a contradiction
3
. The
proof makes use of the factorization
Y
3
=7
3
÷X
3
=(7÷X)(7
2
÷X7÷X
2
).
and it was recognized that a stumbling block to proving the theorem for larger m is that no
such factorization exists into polynomials with integer coefﬁcients of degree _ 2. This led
people to look at more general factorizations.
In a famous incident, the French mathematician Lam´ e gave a talk at the Paris Academy
in 1847 in which he claimed to prove Fermat’s last theorem using the following ideas. Let
] >2 be a prime, and suppose ., ., : are nonzero integers such that
.
;
÷.
;
=:
;
.
Write
.
;
=:
;
÷.
;
=
¸
(: ÷¯
i
.). 0 _i _]÷1. ¯ =e
2ti¡;
.
He then showed how to obtain a smaller solution to the equation, and hence a contradiction.
Liouville immediately questioned a step in Lam´ e’s proof in which he assumed that, in order
to show that each factor (: ÷¯
i
.) is a ]th power, it sufﬁces to show that the factors are
relatively prime in pairs and their product is a ]th power. In fact, Lam´ e couldn’t justify his
step (ZŒ¯ is not always a principal ideal domain), and Fermat’s last theorem was not proved
for almost 150 years. However, shortly after Lam´ e’s embarrassing lecture, Kummer used
his results on the arithmetic of the ﬁelds QŒ¯ to prove Fermat’s last theorem for all regular
primes, i.e., for all primes ] such that ] does not divide the class number of QŒ¯
;
.
Another application is to ﬁnding Galois groups. The splitting ﬁeld of a polynomial
}(X) ÷ QŒX is a Galois extension of Q. In the basic graduate algebra course (see FT),
we learn how to compute the Galois group only when the degree is very small. By using
algebraic number theory one can write down an algorithm to do it for any degree.
For applications of algebraic number theory to elliptic curves, see, for example, Milne
2006.
Some comments on the literature
COMPUTATIONAL NUMBER THEORY
Cohen 1993 and Pohst and Zassenhaus 1989 provide algorithms for most of the construc
tions we make in this course. The ﬁrst assumes the reader knows number theory, whereas
the second develops the whole subject algorithmically. Cohen’s book is the more useful as
a supplement to this course, but wasn’t available when these notes were ﬁrst written. While
the books are concerned with moreorless practical algorithms for ﬁelds of small degree
and small discriminant, Lenstra (1992) concentrates on ﬁnding “good” general algorithms.
3
The simplest proof by inﬁnite descent is that showing that
2 is irrational.
5
HISTORY OF ALGEBRAIC NUMBER THEORY
Dedekind 1996, with its introduction by Stillwell, gives an excellent idea of how algebraic
number theory developed. Edwards 1977 is a history of algebraic number theory, con
centrating on the efforts to prove Fermat’s last theorem. The notes in Narkiewicz 1990
document the origins of most signiﬁcant results in algebraic number theory.
Exercises
01 Let J be a squarefree integer. Complete the veriﬁcation that the ring of integers in
QŒ
J is as described.
02 Complete the veriﬁcation that, in ZŒ
÷5,
(6) =(2. 1÷
÷5)(2. 1÷
÷5)(3. 1÷
÷5)(3. 1÷
÷5)
is a factorization of (6) into a product of prime ideals.
6
CHAPTER 1
Preliminaries from Commutative
Algebra
Many results that were ﬁrst proved for rings of integers in number ﬁelds are true for more
general commutative rings, and it is more natural to prove them in that context.
1
Basic deﬁnitions
All rings will be commutative, and have an identity element (i.e., an element 1 such that
1a =a for all a ÷ ·), and a homomorphism of rings will map the identity element to the
identity element.
A ring T together with a homomorphism of rings · ÷ T will be referred to as an
·algebra. We use this terminology mainly when · is a subring of T. In this case, for
elements ˇ
1
. .... ˇ
n
of T, ·Œˇ
1
. .... ˇ
n
 denotes the smallest subring of T containing · and
the ˇ
i
. It consists of all polynomials in the ˇ
i
with coefﬁcients in ·, i.e., elements of the
form
¸
a
i
1
...i
m
ˇ
i
1
1
...ˇ
i
m
n
. a
i
1
...i
m
÷ ·.
We also refer to ·Œˇ
1
. .... ˇ
n
 as the ·subalgebra of T generated by the ˇ
i
, and when
T =·Œˇ
1
. .... ˇ
n
 we say that the ˇ
i
generate T as an ·algebra.
For elements a
1
. a
2
. . . . of ·, we let (a
1
. a
2
. . . .) denote the smallest ideal containing
the a
i
. It consists of ﬁnite sums
¸
c
i
a
i
, c
i
÷ ·, and it is called the ideal generated by
a
1
. a
2
. . . .. When a and b are ideals in ·, we deﬁne
a÷b ={a÷b [ a ÷ a, b ÷ b].
It is again an ideal in · — in fact, it is the smallest ideal containing both a and b. If
a =(a
1
. .... a
n
) and b =(b
1
. .... b
n
), then a÷b =(a
1
. .... a
n
. b
1
. .... b
n
).
Given an ideal a in ·, we can form the quotient ring ·¡a. Let } : · ÷·¡a be the
homomorphisma ÷a÷a; then b ÷}
1
(b) deﬁnes a onetoone correspondence between
the ideals of ·¡a and the ideals of · containing a, and
·¡}
1
(b)
:
÷(·¡a)¡b.
1
See also the notes A Primer of Commutative Algebra available on my website.
7
1. PRELIMINARIES FROM COMMUTATIVE ALGEBRA
A proper ideal a of · is prime if ab ÷ a =a or b ÷ a. An ideal a is prime if and only if
the quotient ring ·¡a is an integral domain. A nonzero element ¬ of · is said to be prime
if (¬) is a prime ideal; equivalently, if ¬[ab =¬[a or ¬[b.
An ideal min · is maximal if it is maximal among the proper ideals of ·, i.e., if m=·
and there does not exist an ideal a = · containing m but distinct from it. An ideal a is
maximal if and only if ·¡a is a ﬁeld. Every proper ideal a of · is contained in a maximal
ideal — if · is Noetherian (see below) this is obvious; otherwise the proof requires Zorn’s
lemma. In particular, every nonunit in · is contained in a maximal ideal.
There are the implications: · is a Euclidean domain =· is a principal ideal domain
=· is a unique factorization domain (see any good graduate algebra course).
Ideals in products of rings
PROPOSITION 1.1 Consider a product of rings ·×T. If a and b are ideals in · and T
respectively, then a ×b is an ideal in ·×T, and every ideal in ·×T is of this form. The
prime ideals of ·×T are the ideals of the form
p×T (p a prime ideal of ·), ·×p (p a prime ideal of T).
PROOF. Let c be an ideal in ·×T, and let
a ={a ÷ · [ (a. 0) ÷ c]. b ={b ÷ T [ (0. b) ÷ c].
Clearly a ×b c c. Conversely, let (a. b) ÷ c. Then (a. 0) = (a. b) (1. 0) ÷ c and (0. b) =
(a. b) (0. 1) ÷ c, and so (a. b) ÷ a×b.
Recall that an ideal c cC is prime if and only if C¡c is an integral domain. The map
·×T ÷·¡a×T¡b. (a. b) ÷(a÷a. b ÷b)
has kernel a×b, and hence induces an isomorphism
(·×T)¡(a×b) .·¡a×T¡b.
Now use that a product of rings is an integral domain if and only if one ring is zero and the
other is an integral domain.
2
REMARK 1.2 The lemma extends in an obvious way to a ﬁnite product of rings: the ideals
in ·
1
× ×·
n
are of the forma
1
× ×a
n
with a
i
an ideal in ·
i
; moreover, a
1
× ×a
n
is prime if and only if there is a ; such that a
}
is a prime ideal in ·
}
and a
i
=·
i
for i =;.
Noetherian rings
A ring · is Noetherian if every ideal in · is ﬁnitely generated.
PROPOSITION 1.3 The following conditions on a ring · are equivalent:
(a) · is Noetherian.
(b) Every ascending chain of ideals
a
1
ca
2
c ca
n
c
eventually becomes constant, i.e., for some n, a
n
=a
n÷1
= .
8
Noetherian rings
(c) Every nonempty set S of ideals in · has a maximal element, i.e., there exists an ideal
in S not properly contained in any other ideal in S.
PROOF. (a)=(b): Let a =
a
i
; it is an ideal, and hence is ﬁnitely generated, say a =
(a
1
. . . . . a
i
). For some n, a
n
will contain all the a
i
, and so a
n
=a
n÷1
= =a.
(b)=(c): Let a
1
÷ S. If a
1
is not a maximal element of S, then there exists an a
2
÷ S such
that a
1
´a
2
. If a
2
is not maximal, then there exists an a
3
etc.. From (b) we know that this
process will lead to a maximal element after only ﬁnitely many steps.
(c)=(a): Let a be an ideal in ·, and let S be the set of ﬁnitely generated ideals contained
in a. Then S is nonempty because it contains the zero ideal, and so it contains a maximal
element, say, a
t
= (a
1
. . . . . a
i
). If a
t
= a, then there exists an element a ÷ a a
t
, and
(a
1
. . . . . a
i
. a) will be a ﬁnitely generated ideal in a properly containing a
t
. This contradicts
the deﬁnition of a
t
.
2
A famous theorem of Hilbert states that kŒX
1
. .... X
n
 is Noetherian. In practice, al
most all the rings that arise naturally in algebraic number theory or algebraic geometry
are Noetherian, but not all rings are Noetherian. For example, the ring kŒX
1
. . . . . X
n
. . . .
of polynomials in an inﬁnite sequence of symbols is not Noetherian because the chain of
ideals
(X
1
) c(X
1
. X
2
) c(X
1
. X
2
. X
3
) c
never becomes constant.
PROPOSITION 1.4 Every nonzero nonunit element of a Noetherian integral domain can be
written as a product of irreducible elements.
PROOF. We shall need to use that, for elements a and b of an integral domain ·,
(a) c(b) ” b[a, with equality if and only if b =a×unit.
The ﬁrst assertion is obvious. For the second, note that if a = bc and b = aJ then a =
bc =aJc, and so Jc =1. Hence both c and J are units.
Suppose the statement of the proposition is false for a Noetherian integral domain ·.
Then there exists an element a ÷ · which contradicts the statement and is such that (a) is
maximal among the ideals generated by such elements (here we use that · is Noetherian).
Since a can not be written as a product of irreducible elements, it is not itself irreducible,
and so a =bc with b and c nonunits. Clearly (b) ¯ (a), and the ideals can’t be equal for
otherwise c would be a unit. From the maximality of (a), we deduce that b can be written
as a product of irreducible elements, and similarly for c. Thus a is a product of irreducible
elements, and we have a contradiction.
2
REMARK 1.5 Note that the proposition fails for the ring O of all algebraic integers in the
algebraic closure of Q in C, because, for example, we can keep in extracting square roots
— an algebraic integer ˛ can not be an irreducible element of O because
˛ will also be
an algebraic integer and ˛ =
˛
˛. Thus O is not Noetherian.
9
1. PRELIMINARIES FROM COMMUTATIVE ALGEBRA
Noetherian modules
Let · be a ring. An ·module M is said to be Noetherian if every submodule is ﬁnitely
generated.
PROPOSITION 1.6 The following conditions on an ·module M are equivalent:
(a) M is Noetherian;
(b) every ascending chain of submodules eventually becomes constant;
(c) every nonempty set of submodules in M has a maximal element.
PROOF. Similar to the proof of Proposition 1.3.
2
PROPOSITION 1.7 Let M be an ·module, and let N be a submodule of M. If N and
M¡N are both Noetherian, then so also is M.
PROOF. I claim that if M
t
cM
tt
are submodules of M such that M
t
¨N =M
tt
¨N and
M
t
and M
tt
have the same image in M¡N, then M
t
=M
tt
. To see this, let . ÷ M
tt
; the
second condition implies that there exists a . ÷ M
t
with the same image as . in M¡N, i.e.,
such that . ÷. ÷ N. Then . ÷. ÷ M
tt
¨N cM
t
, and so . ÷ M
t
.
Now consider an ascending chain of submodules of M. If M¡N is Noetherian, the
image of the chain in M¡N becomes constant, and if N is Noetherian, the intersection of
the chain with N becomes constant. Now the claim shows that the chain itself becomes
constant.
2
PROPOSITION 1.8 Let · be a Noetherian ring. Then every ﬁnitely generated ·module is
Noetherian.
PROOF. If M is generated by a single element, then M ~·¡a for some ideal a in ·, and
the statement is obvious. We argue by induction on the minimum number n of generators
of M. Since M contains a submodule N generated by n÷1 elements such that the quotient
M¡N is generated by a single element, the statement follows from (1.7).
2
Local rings
A ring · is said to local if it has exactly one maximal ideal m. In this case, ·
×
=·m
(complement of m in ·).
LEMMA 1.9 (NAKAYAMA’S LEMMA) Let · be a local Noetherian ring, and let a be a
proper ideal in ·. Let M be a ﬁnitely generated ·module, and deﬁne
aM ={
¸
a
i
m
i
[ a
i
÷ a. m
i
÷ M].
(a) If aM =M, then M =0.
(b) If N is a submodule of M such that N ÷aM =M, then N =M.
10
Rings of fractions
PROOF. (a) Suppose that aM = M but M = 0. Choose a minimal set of generators
{e
1
. . . . . e
n
] for M, n _1, and write
e
1
=a
1
e
1
÷ ÷a
n
e
n
, a
i
÷ a.
Then
(1÷a
1
)e
1
=a
2
e
2
÷ ÷a
n
e
n
.
As 1 ÷a
1
is not in m, it is a unit, and so {e
2
. .... e
n
] generates M, which contradicts our
choice of {e
1
. . . . . e
n
].
(b) It sufﬁces to show that a(M¡N) =M¡N for then (a) shows that M¡N =0. Con
sider m÷N, m÷ M. From the assumption, we can write
m=n÷
¸
a
i
m
i
, with a
i
÷ a, m
i
÷ M.
Then
m÷N =
¸
(a
i
m
i
÷N) =
¸
a
i
(m
i
÷N)
and so m÷N ÷ a(M¡N).
2
The hypothesis that M be ﬁnitely generated in the lemma is essential. For example, if
· is a local integral domain with maximal ideal m = 0, then mM = M for any ﬁeld M
containing · but M =0.
Rings of fractions
Let · be an integral domain; there is a ﬁeld 1 ¯·, called the ﬁeld of fractions of ·, with
the property that every c ÷ 1 can be written in the form c =ab
1
with a. b ÷ · and b =0.
For example, Q is the ﬁeld of fractions of Z, and k(X) is the ﬁeld of fractions of kŒX.
Let · be an integral domain with ﬁeld of fractions 1. · subset S of · is said to be
multiplicative if 0 … S, 1 ÷ S, and S is closed under multiplication. If S is a multiplicative
subset, then we deﬁne
S
1
· ={a¡b ÷ 1 [ b ÷ S].
It is obviously a subring of 1.
EXAMPLE 1.10 (a) Let t be a nonzero element of ·; then
S
t
def
={1,t ,t
2
,...]
is a multiplicative subset of ·, and we (sometimes) write ·
t
for S
1
t
·. For example, if J
is a nonzero integer, then
2
Z
d
consists of those elements of Q whose denominator divides
some power of J:
Z
d
={a¡J
n
÷ Q[ a ÷ Z, n _0].
(b) If p is a prime ideal, then S
p
=·p is a multiplicative set (if neither a nor b belongs
to p, then ab does not belong to p). We write ·
p
for S
1
p
·. For example,
Z
(;)
={m¡n ÷ Q[ n is not divisible by ]].
2
This notation conﬂicts with a later notation in which Z
;
denotes the ring of ]adic integers.
11
1. PRELIMINARIES FROM COMMUTATIVE ALGEBRA
PROPOSITION 1.11 Consider an integral domain ·and a multiplicative subset S of ·. For
an ideal a of ·, write a
e
for the ideal it generates in S
1
·; for an ideal a of S
1
·, write
a
c
for a¨·. Then:
a
ce
=a for all ideals a of S
1
·
a
ec
=a if a is a prime ideal of · disjoint from S.
PROOF. Let a be an ideal in S
1
·. Clearly (a ¨·)
e
c a because a ¨· c a and a is an
ideal in S
1
·. For the reverse inclusion, let b ÷ a. We can write it b = a¡s with a ÷ ·,
s ÷ S. Then a =s (a¡s) ÷ a¨·, and so a¡s =(s (a¡s))¡s ÷ (a¨·)
e
.
Let p be a prime ideal disjoint from S. Clearly (S
1
p) ¨· ¯p. For the reverse inclu
sion, let a¡s ÷ (S
1
p) ¨·, a ÷ p, s ÷ S. Consider the equation
o
x
s = a ÷ p. Both a¡s
and s are in ·, and so at least one of a¡s or s is in p (because it is prime); but s … p (by
assumption), and so a¡s ÷ p.
2
PROPOSITION 1.12 Let · be an integral domain, and let S be a multiplicative subset of
·. The map p ÷p
e
def
=p S
1
· is a bijection from the set of prime ideals in · such that
p¨S =∅ to the set of prime ideals in S
1
·; the inverse map is p ÷p¨·.
PROOF. It is easy to see that
p a prime ideal disjoint from S =p
e
is a prime ideal in S
1
·,
p a prime ideal in S
1
· =p¨· is a prime ideal in · disjoint from S.
and (1.11) shows that the two maps are inverse.
2
EXAMPLE 1.13 (a) If p is a prime ideal in ·, then ·
p
is a local ring (because p contains
every prime ideal disjoint from S
p
).
(b) We list the prime ideals in some rings:
Z: (2). (3). (5). (7). (11). . . . . (0):
Z
2
: (3). (5). (7). (11). . . . . (0):
Z
(2)
: (2). (0):
Z
42
: (5). (11). (13). . . . . (0):
Z¡(42): (2). (3). (7).
Note that in general, for t a nonzero element of an integral domain,
{prime ideals of ·
t
]  {prime ideals of · not containing t ]
{prime ideals of ·¡(t )]  {prime ideals of · containing t ].
The Chinese remainder theorem
Recall the classical form of the theorem: let J
1
. .... J
n
be integers, relatively prime in pairs;
then for any integers .
1
. .... .
n
, the congruences
. ÷.
i
mod J
i
12
The Chinese remainder theorem
have a simultaneous solution . ÷ Z; moreover, if . is one solution, then the other solutions
are the integers of the form . ÷mJ with m÷ Z and J =
¸
J
i
.
We want to translate this in terms of ideals. Integers m and n are relatively prime if and
only if (m. n) =Z, i.e., if and only if (m) ÷(n) =Z. This suggests deﬁning ideals a and b
in a ring · to be relatively prime if a÷b =·.
If m
1
. .... m
k
are integers, then
¸
(m
i
) = (m) where m is the least common multiple
of the m
i
. Thus
¸
(m
i
) ¯ (
¸
m
i
), which equals
¸
(m
i
). If the m
i
are relatively prime in
pairs, then m=
¸
m
i
, and so we have
¸
(m
i
) =
¸
(m
i
). Note that in general,
a
1
a
2
a
n
ca
1
¨a
2
¨... ¨a
n
.
but the two ideals need not be equal.
These remarks suggest the following statement.
THEOREM 1.14 Let a
1
. .... a
n
be ideals in a ring ·, relatively prime in pairs. Then for any
elements .
1
. .... .
n
of ·, the congruences
. ÷.
i
mod a
i
have a simultaneous solution . ÷ ·; moreover, if . is one solution, then the other solutions
are the elements of the form . ÷a with a ÷
¸
a
i
, and
¸
a
i
=
¸
a
i
. In other words, the
natural maps give an exact sequence
0 ÷a ÷· ÷
n
¸
i=1
·¡a
i
÷0
with a =
¸
a
i
=
¸
a
i
.
PROOF. Suppose ﬁrst that n = 2. As a
1
÷a
2
= ·, there are elements a
i
÷ a
i
such that
a
1
÷a
2
=1. The element . =a
1
.
2
÷a
2
.
1
has the required property.
For each i we can ﬁnd elements a
i
÷ a
1
and b
i
÷ a
i
such that
a
i
÷b
i
=1, all i _2.
The product
¸
i_2
(a
i
÷b
i
) =1, and lies in a
1
÷
¸
i_2
a
i
, and so
a
1
÷
¸
i_2
a
i
=·.
We can now apply the theorem in the case n =2 to obtain an element .
1
of · such that
.
1
÷1 mod a
1
. .
1
÷0 mod
¸
i_2
a
i
.
These conditions imply
.
1
÷1 mod a
1
. .
1
÷0 mod a
}
, all ; >1.
Similarly, there exist elements .
2
. .... .
n
such that
.
i
÷1 mod a
i
. .
i
÷0 mod a
}
for ; =i.
The element . =
¸
.
i
.
i
now satisﬁes the requirements.
13
1. PRELIMINARIES FROM COMMUTATIVE ALGEBRA
It remains to prove that
¸
a
i
=
¸
a
i
. We have already noted that
¸
a
i
¯
¸
a
i
. First
suppose that n =2, and let a
1
÷a
2
=1, as before. For c ÷ a
1
¨a
2
, we have
c =a
1
c ÷a
2
c ÷ a
1
a
2
which proves that a
1
¨a
2
= a
1
a
2
. We complete the proof by induction. This allows us
to assume that
¸
i_2
a
i
=
¸
i_2
a
i
. We showed above that a
1
and
¸
i_2
a
i
are relatively
prime, and so
a
1
(
¸
i_2
a
i
) =a
1
¨(
¸
i_2
a
i
) =
¸
a
i
.
2
The theorem extends to ·modules.
THEOREM 1.15 Let a
1
. .... a
n
be ideals in ·, relatively prime in pairs, and let M be an
·module. There is an exact sequence:
0 ÷aM ÷M ÷
¸
i
M¡a
i
M ÷0
with a =
¸
a
i
=
¸
a
i
.
This can be proved in the same way as Theorem 1.14, but I prefer to use tensor products,
which I now review.
Review of tensor products
Let M, N, and 1 be ·modules. A mapping } : M×N ÷1 is said to be ·bilinear if
}(m÷m
t
. n) =}(m. n) ÷}(m
t
. n)
}(m. n÷n
t
) =}(m. n) ÷}(m. n
t
)
}(am. n) =a}(m. n) =}(m. an)
¸
¸
all a ÷ ·. m. m
t
÷ M. n. n
t
÷ N.
i.e., if it is linear in each variable. A pair (O. } ) consisting of an ·module O and an
·bilinear map } : M ×N ÷O is called the tensor product of M and N if any other ·
bilinear map }
t
: M ×N ÷1 factors uniquely into }
t
= ˛ ı } with ˛: O ÷1 ·linear.
The tensor product exists, and is unique (up to a unique isomorphism making the obvious
diagram commute). We denote it by M ˝
¹
N, and we write (m. n) ÷m˝n for } . The
pair (M˝
¹
N. (m. n) ÷m˝n) is characterized by each of the following two conditions:
(a) The map M×N ÷M˝
¹
N is ·bilinear, and any other ·bilinear map M×N ÷
1 is of the form (m. n) ÷˛(m˝n) for a unique ·linear map ˛: M˝
¹
N ÷1; thus
Bilin
¹
(M×N. 1) =Hom
¹
(M˝
¹
N. 1).
(b) The ·module M˝
¹
N has as generators the m˝n, m÷M, n ÷N, and as relations
(m÷m
t
) ˝n =m˝n÷m
t
˝n
m˝(n÷n
t
) =m˝n÷m˝n
t
am˝n =a(m˝n) =m˝an
¸
¸
all a ÷ ·. m. m
t
÷ M. n. n
t
÷ N.
Tensor products commute with direct sums: there is a canonical isomorphism
(
i
M
i
) ˝
¹
(
}
N
}
)
:
÷
i,}
M
i
˝
¹
N
}
.
(
¸
m
i
) ˝(
¸
n
}
) ÷
¸
m
i
˝n
}
.
14
Review of tensor products
It follows that if M and N are free ·modules
3
with bases (e
i
) and (}
}
) respectively, then
M ˝
¹
N is a free ·module with basis (e
i
˝}
}
). In particular, if V and W are vector
spaces over a ﬁeld k of dimensions m and n respectively, then V ˝
k
W is a vector space
over k of dimension mn.
Let ˛: M ÷M
t
and ˇ: N ÷N
t
be ·linear maps. Then
(m. n) ÷˛(m) ˝ˇ(n): M×N ÷M
t
˝
¹
N
t
is ·bilinear, and therefore factors uniquely through M ×N ÷M ˝
¹
N. Thus there is a
unique ·linear map ˛ ˝ˇ: M˝
¹
N ÷M
t
˝
¹
N
t
such that
(˛ ˝ˇ)(m˝n) =˛(m) ˝ˇ(n).
REMARK 1.16 The tensor product of two matrices regarded as linear maps is called their
Kronecker product.
4
If · is m×n (so a linear map k
n
÷k
n
) and T is r ×s (so a linear
map k
x
÷k
i
), then ·˝T is the mr ×ns matrix (linear map k
nx
÷k
ni
) with
·˝T =
¸
¸
a
11
T a
1n
T
.
.
.
.
.
.
.
.
.
a
n1
T a
nn
T
.
LEMMA 1.17 If ˛: M ÷M
t
and ˇ: N ÷N
t
are surjective, then so also is
˛ ˝ˇ: M˝
¹
N ÷M
t
˝
¹
N
t
.
PROOF. Recall that M
t
˝N
t
is generated as an ·module by the elements m
t
˝n
t
, m
t
÷
M
t
, n
t
÷ N
t
. By assumption m
t
=˛(m) for some m÷ M and n
t
=ˇ(n) for some n ÷ N,
and so m
t
˝n
t
=˛(m) ˝ˇ(n) =(˛˝ˇ)(m˝n). Therefore the image of ˛˝ˇ contains a
set of generators for M
t
˝
¹
N
t
and so it is equal to it.
2
One can also show that if
M
t
÷M ÷M
tt
÷0
is exact, then so also is
M
t
˝
¹
1 ÷M˝
¹
1 ÷M
tt
˝
¹
1 ÷0.
For example, if we tensor the exact sequence
0 ÷a ÷· ÷·¡a ÷0
with M, we obtain an exact sequence
a˝
¹
M ÷M ÷(·¡a) ˝
¹
M ÷0 (2)
The image of a˝
¹
M in M is
aM
def
={
¸
a
i
m
i
[ a
i
÷ a, m
i
÷ M].
3
Let M be an ·module. Elements e
1
. . . . . e
n
form a basis for M if every element of M can be expressed
uniquely as a linear combination of the e
i
’s with coefﬁcients in ·. Then ·
n
÷M, (a
1
. . . . . a
n
) ÷
¸
a
i
e
i
, is
an isomorphism of ·modules, and M is said to be a free ·module of rank m.
4
Kronecker products of matrices predate tensor products by about 70 years.
15
1. PRELIMINARIES FROM COMMUTATIVE ALGEBRA
and so we obtain from the exact sequence (2) that
M¡aM .(·¡a) ˝
¹
M (3)
By way of contrast, if M ÷N is injective, then M˝
¹
1 ÷N ˝
¹
1 need not be injective.
For example, take · = Z, and note that (Z
n
÷Z) ˝
Z
(Z¡mZ) equals Z¡mZ
n
÷Z¡mZ,
which is the zero map.
PROOF (OF THEOREM 1.15) Return to the situation of the theorem. When we tensor the
isomorphism
·¡a
:
÷
¸
·¡a
i
with M, we get an isomorphism
M¡aM .(·¡a) ˝
¹
M
:
÷
¸
(·¡a
i
) ˝
¹
M .
¸
M¡a
i
M.
as required.
2
EXTENSION OF SCALARS
If · ÷T is an ·algebra and M is an ·module, then T ˝
¹
M has a natural structure of
a Tmodule for which
b(b
t
˝m) =bb
t
˝m. b. b
t
÷ T. m÷ M.
We say that T ˝
¹
M is the Tmodule obtained from M by extension of scalars. The map
m ÷1 ˝m: M ÷T ˝
¹
M has the following universal property: it is ·linear, and for
any ·linear map ˛: M ÷N from M into a Tmodule N, there is a unique Tlinear map
˛
t
: T ˝
¹
M ÷N such that ˛
t
(1˝m) =˛(m). Thus ˛ ÷˛
t
deﬁnes an isomorphism
Hom
¹
(M. N) ÷Hom
B
(T ˝
¹
M. N), N a Tmodule.
For example, ·˝
¹
M =M. If M is a free ·module with basis e
1
. . . . . e
n
, then T ˝
¹
M
is a free Tmodule with basis 1˝e
1
. . . . . 1˝e
n
.
TENSOR PRODUCTS OF ALGEBRAS
If } : · ÷ T and g: · ÷ C are ·algebras, then T ˝
¹
C has a natural structure of an
·algebra: the product structure is determined by the rule
(b ˝c)(b
t
˝c
t
) =bb
t
˝cc
t
and the map · ÷T ˝
¹
C is a ÷}(a) ˝1 =1˝g(a).
For example, there is a canonical isomorphism
a˝} ÷a} : 1˝
k
kŒX
1
. . . . . X
n
 ÷1ŒX
1
. . . . . X
n
 (4)
16
Review of tensor products
TENSOR PRODUCTS OF FIELDS
We are now able to compute 1˝
k
˝ if 1 is a ﬁnite separable ﬁeld extension of a ﬁeld k
and ˝ is an arbitrary ﬁeld extension of k. According to the primitive element theorem (FT
5.1), 1 =kŒ˛ for some ˛ ÷ 1. Let }(X) be the minimum polynomial of ˛. By deﬁnition
this means that the map g(X) ÷g(˛) determines an isomorphism
kŒX¡(}(X)) ÷1.
Hence
1˝
k
˝ .(kŒX¡(}(X))) ˝
k
˝ .˝ŒX¡(}(X))
by (3) and (4). Because 1 is separable over k, }(X) has distinct roots. Therefore }(X)
factors in ˝ŒX into monic irreducible polynomials
}(X) =}
1
(X) }
i
(X)
that are relatively prime in pairs. We can apply the Chinese Remainder Theorem to deduce
that
˝ŒX¡(}(X)) =
¸
i
i=1
˝ŒX¡(}
i
(X)).
Finally, ˝ŒX¡(}
i
(X)) is a ﬁnite separable ﬁeld extension of ˝ of degree deg}
i
. Thus we
have proved the following result:
THEOREM 1.18 Let 1 be a ﬁnite separable ﬁeld extension of k, and let ˝ be an arbitrary
ﬁeld extension. Then 1˝
k
˝ is a product of ﬁnite separable ﬁeld extensions of ˝,
1˝
k
˝ =
¸
i
i=1
˝
i
.
If ˛ is a primitive element for 1¡k, then the image ˛
i
of ˛ in ˝
i
is a primitive element
for ˝
i
¡˝, and if }(X) and }
i
(X) are the minimum polynomials for ˛ and ˛
i
respectively,
then
}(X) =
¸
i
i=1
}
i
(X).
EXAMPLE 1.19 Let 1 =QŒ˛ with ˛ algebraic over Q. Then
C˝
Q
1 .C˝
Q
(QŒX¡(}(X))) .CŒX¡((}(X)) .
¸
i
i=1
CŒX¡(X ÷˛
i
) ~C
i
.
Here ˛
1
. . . . . ˛
i
are the conjugates of ˛ in C. The composite of ˇ ÷1˝ˇ: 1 ÷C˝
Q
1
with projection onto the i th factor is
¸
a
}
˛
}
÷
¸
a
}
˛
}
i
.
We note that it is essential to assume in (1.18) that 1 is separable over k. If not, there
will be an ˛ ÷ 1 such that ˛
;
÷ k but ˛ … k, and the ring 1˝
k
1 will contain an element
ˇ =(˛ ˝1÷1˝˛) =0 such that
ˇ
;
=˛
;
˝1÷1˝˛
;
=˛
;
(1˝1) ÷˛
;
(1˝1) =0.
Hence 1˝
k
1 contains a nonzero nilpotent element, and so it can’t be a product of ﬁelds.
NOTES Ideals were introduced and studied by Dedekind for rings of algebraic integers, and later by
others in polynomial rings. It was not until the 1920s that the theory was placed in its most natural
setting, that of arbitrary commutative rings (by Emil Artin and Emmy Noether).
17
1. PRELIMINARIES FROM COMMUTATIVE ALGEBRA
Exercise
11 Let · be an integral domain. A multiplicative subset S of · is said to be saturated if
ab ÷ S =a and b ÷ S.
(a) Show that S is saturated ” its complement is a union of prime ideals.
(b) Show that given a multiplicative system S, there is a unique smallest saturated multi
plicative systemS
t
containing S, and that S
t
=·
p, where p runs over the prime
ideals disjoint from S. Show that S
t1
· =S
1
·. Deduce that S
1
· is character
ized by the set of prime ideals of · that remain prime in S
1
·.
18
CHAPTER 2
Rings of Integers
Let · be an integral domain, and let 1 be a ﬁeld containing ·. An element ˛ of 1 is said
to be integral over · if it is a root of a monic polynomial with coefﬁcients in ·, i.e., if it
satisﬁes an equation
˛
n
÷a
1
˛
n1
÷ ÷a
n
=0. a
i
÷ ·.
THEOREM 2.1 The elements of 1 integral over · form a ring.
I shall give two proofs of this theorem. The ﬁrst uses Newton’s theory of symmetric
polynomials and a result of Eisenstein, and the second is Dedekind’s surprisingly modern
proof, which avoids symmetric polynomials.
First proof that the integral elements form a ring
A polynomial 1(X
1
. .... X
i
) ÷ ·ŒX
1
. . . . . X
i
 is said to be symmetric if it is unchanged
when its variables are permuted, i.e., if
1(X
c(1)
. . . . . X
c(i)
) =1(X
1
. . . . . X
i
). all o ÷ Sym
i
.
For example
S
1
=
¸
X
i
. S
2
=
¸
i~}
X
i
X
}
. . . . . S
i
=X
1
X
i
.
are all symmetric. These particular polynomials are called the elementary symmetric poly
nomials.
THEOREM 2.2 (Symmetric function theorem) Let · be a ring. Every symmetric polyno
mial 1(X
1
. .... X
i
) in ·ŒX
1
. .... X
i
 is equal to a polynomial in the symmetric elementary
polynomials with coefﬁcients in ·, i.e., 1 ÷ ·ŒS
1
. .... S
i
.
PROOF. We deﬁne an ordering on the monomials in the X
i
by requiring that
X
i
1
1
X
i
2
2
X
i
r
i
>X
}
1
1
X
}
2
2
X
}
r
i
if either
i
1
÷i
2
÷ ÷i
i
>;
1
÷;
2
÷ ÷;
i
19
2. RINGS OF INTEGERS
or equality holds and, for some s,
i
1
=;
1
. . . . . i
x
=;
x
. but i
x÷1
>;
x÷1
.
Let X
k
1
1
X
k
r
i
be the highest monomial occurring in 1 with a coefﬁcient c =0. Because
1 is symmetric, it contains all monomials obtained fromX
k
1
1
X
k
r
i
by permuting the X’s.
Hence k
1
_k
2
_ _k
i
.
Clearly, the highest monomial in S
i
is X
1
X
i
, and it follows easily that the highest
monomial in S
d
1
1
S
d
r
i
is
X
d
1
÷d
2
÷···÷d
r
1
X
d
2
÷···÷d
r
2
X
d
r
i
.
Therefore
1(X
1
. . . . . X
i
) ÷cS
k
1
k
2
1
S
k
2
k
3
2
S
k
r
i
<1(X
1
. . . . . X
i
).
We can repeat this argument with the polynomial on the left, and after a ﬁnite number of
steps, we will arrive at a representation of 1 as a polynomial in S
1
. . . . . S
i
.
2
Let }(X) =X
n
÷a
1
X
n1
÷ ÷a
n
÷ ·ŒX, and let ˛
1
. . . . . ˛
n
be the roots of }(X)
in some ring containing ·, so that }(X) =
¸
(X ÷˛
i
) in the larger ring. Then
a
1
=÷S
1
(˛
1
. . . . . ˛
n
). a
2
=S
2
(˛
1
. . . . . ˛
n
). . . . . a
n
=˙S
n
(˛
1
. . . . . ˛
n
).
Thus the elementary symmetric polynomials in the roots of }(X) lie in ·, and so the
theorem implies that every symmetric polynomial in the roots of }(X) lies in ·.
PROPOSITION 2.3 Let · be an integral domain, and let ˝ be an algebraically closed ﬁeld
containing ·. If ˛
1
. . . . . ˛
n
are the roots in ˝ of a monic polynomial in ·ŒX, then any
polynomial g(˛
1
. . . . . ˛
n
) in the ˛
i
with coefﬁcients in · is a root of a monic polynomial in
·ŒX.
PROOF. Clearly
h(X)
def
=
¸
c÷Sym
n
(X ÷g(˛
c(1)
. . . . . ˛
c(n)
))
is a monic polynomial whose coefﬁcients are symmetric polynomials in the ˛
i
, and there
fore lie in ·. But g(˛
1
. . . . . ˛
n
) is one of its roots.
2
We now prove Theorem 2.1. Let ˛
1
and ˛
2
be elements of 1 integral over ·. There
exists a monic polynomial in ·ŒX having both ˛
1
and ˛
2
as roots. We can now apply (2.3)
with g(˛
1
. . . .) equal to ˛
1
˙˛
2
or ˛
1
˛
2
to deduce that these elements are integral over ·.
Dedekind’s proof that the integral elements form a ring
PROPOSITION 2.4 Let 1 be a ﬁeld containing ·. An element ˛ of 1 is integral over A if
and only if there exists a nonzero ﬁnitely generated ·submodule of 1 such that ˛M cM
(in fact, we can take M =·Œ˛, the ·subalgebra generated by ˛).
20
Dedekind’s proof that the integral elements form a ring
PROOF. =: Suppose
˛
n
÷a
1
˛
n1
÷ ÷a
n
=0. a
i
÷ ·.
Then the ·submodule M of 1 generated by 1, ˛, ..., ˛
n1
has the property that ˛M cM.
==: We shall need to apply Cramer’s rule. As usually stated (in linear algebra courses)
this says that, if
n
¸
}=1
c
i}
.
}
=J
i
. i =1. . . . . m.
then
.
}
=det(C
}
)¡det(C)
where C =(c
i}
) and C
}
is obtained from C by replacing the elements of the ; th column
with the J
i
s. When one restates the equation as
det(C) .
}
=det(C
}
)
it becomes true over any ring (whether or not det(C) is invertible). The proof is elementary—
essentially it is what you wind up with when you eliminate the other variables (try it for
m=2). Alternatively, expand out
det C
}
=
ˇ
ˇ
ˇ
ˇ
ˇ
ˇ
ˇ
c
11
. . .
¸
c
1}
.
}
. . . c
1n
.
.
.
.
.
.
.
.
.
c
n1
. . .
¸
c
n}
.
}
. . . c
nn
ˇ
ˇ
ˇ
ˇ
ˇ
ˇ
ˇ
using standard properties of determinants.
Now let M be a nonzero ·module in 1 such that ˛M c M, and let ·
1
. . . . . ·
n
be a
ﬁnite set of generators for M. Then, for each i ,
˛·
i
=
¸
a
i}
·
}
, some a
i}
÷ ·.
We can rewrite this system of equations as
(˛ ÷a
11
)·
1
÷a
12
·
2
÷a
13
·
3
÷ =0
÷a
21
·
1
÷(˛ ÷a
22
)·
2
÷a
23
·
3
÷ =0
=0.
Let C be the matrix of coefﬁcients on the lefthand side. Then Cramer’s rule tells us that
det(C) ·
i
=0 for all i . Since at least one ·
i
is nonzero and we are working inside the ﬁeld
1, this implies that det(C) =0. On expanding out the determinant, we obtain an equation
˛
n
÷c
1
˛
n1
÷c
2
˛
n2
÷ ÷c
n
=0. c
i
÷ ·.
2
We now prove Theorem 2.1. Let ˛ and ˇ be two elements of 1 integral over ·, and let
M and N be ﬁnitely generated ·modules in 1 such that ˛M cM and ˇN cN. Deﬁne
MN =
¸
m
i
n
i
[ m
i
÷ M. n
i
÷ N
¸
.
Then:
(a) MN is an ·submodule of 1 (easy);
(b) it is ﬁnitely generated because, if {e
1
. . . . . e
n
] generates M and {}
1
. . . . . }
n
] gener
ates N, then {e
1
}
1
. . . . . e
i
}
}
. . . . . e
n
}
n
] generates MN;
(c) it is stable under multiplication by ˛ˇ and by ˛ ˙ˇ.
We can now apply (2.4) to deduce that ˛ˇ and ˛ ˙ˇ are integral over ·.
21
2. RINGS OF INTEGERS
Integral elements
DEFINITION 2.5 The ring of elements of 1 integral over · is called the integral closure
of · in 1. The integral closure of Z in an algebraic number ﬁeld 1 is called the ring of
integers O
1
in 1.
Next we want to see that 1 is the ﬁeld of fractions of O
1
; in fact we can prove more.
PROPOSITION 2.6 Let 1 be the ﬁeld of fractions of ·, and let 1 be a ﬁeld containing 1.
If ˛ ÷ 1 is algebraic over 1, then there exists a J ÷ · such that J˛ is integral over ·.
PROOF. By assumption, ˛ satisﬁes an equation
˛
n
÷a
1
˛
n1
÷ ÷a
n
=0. a
i
÷ 1.
Let J be a common denominator for the a
i
, so that Ja
i
÷ · for all i , and multiply through
the equation by J
n
:
J
n
˛
n
÷a
1
J
n
˛
n1
÷... ÷a
n
J
n
=0.
We can rewrite this as
(J˛)
n
÷a
1
J(J˛)
n1
÷ ÷a
n
J
n
=0.
As a
1
J, ... , a
n
J
n
÷ ·, this shows that J˛ is integral over ·.
2
COROLLARY 2.7 Let · be an integral domain with ﬁeld of fractions 1, and let T be the
integral closure of · in a ﬁeld 1 containing 1. If 1 is algebraic over 1, then it is the ﬁeld
of fractions of T.
PROOF. The proposition shows that every ˛ ÷ 1 can be written ˛ = ˇ¡J with ˇ ÷ T,
J ÷ ·.
2
DEFINITION 2.8 A ring · is integrally closed if it is its own integral closure in its ﬁeld of
fractions 1, i.e., if
˛ ÷ 1. ˛ integral over · =˛ ÷ ·.
PROPOSITION 2.9 A unique factorization domain, for example, a principal ideal domain,
is integrally closed.
PROOF. Let · be a unique factorization domain, and let a¡b, with a. b ÷ ·, be an element
of the ﬁeld of fractions of · integral over ·. If b is a unit, then a¡b ÷ ·. Otherwise we may
suppose that there is an irreducible element ¬ of · dividing b but not a. As a¡b is integral
over ·, it satisﬁes an equation
(a¡b)
n
÷a
1
(a¡b)
n1
÷ ÷a
n
=0. a
i
÷ ·.
On multiplying through by b
n
, we obtain the equation
a
n
÷a
1
a
n1
b ÷... ÷a
n
b
n
=0.
The element ¬ then divides every term on the left except a
n
, and hence must divide a
n
.
Since it doesn’t divide a, this is a contradiction.
2
22
Integral elements
The proposition makes it easy to give examples of rings where unique factorization fails
— take any ring which is not integrally closed, for example, ZŒ
5.
EXAMPLE 2.10 (a) The rings Z and ZŒi  are integrally closed because both are principal
ideal domains.
(b) Unique factorization fails in ZŒ
÷3 because
4 =2×2 =(1÷
÷3)(1÷
÷3).
and the four factors are all irreducible because they have the minimum norm 4. However,
ZŒ
÷3 cZŒ
3
1 which is a principal ideal domain (and hence the integral closure of Z in
QŒ
÷3 =QŒ
3
1).
(c) For any ﬁeld k, I claim that the integral closure of kŒS
1
. .... S
n
 in k(X
1
. . . . . X
n
) is
kŒX
1
. . . . . X
n
 (here the S
i
are the elementary symmetric polynomials).
Let } ÷k(X
1
. . . . . X
n
) be integral over kŒS
1
. . . . . S
n
. Then } is integral over kŒX
1
. . . . . X
n
,
which is a unique factorization domain, and hence is integrally closed in its ﬁeld of frac
tions. Thus } ÷ kŒX
1
. . . . . X
n
.
Conversely, let } ÷ kŒX
1
. . . . . X
n
. Then } is a root of the monic polynomial
¸
c÷Sym
m
(T ÷}(X
c(1)
. . . . . X
c(n)
)).
The coefﬁcients of this polynomial are symmetric polynomials in the X
i
, and therefore (see
2.2) lie in kŒS
1
. . . . . S
i
.
PROPOSITION 2.11 Let 1 be the ﬁeld of fractions of ·, and let 1 be an extension of 1 of
ﬁnite degree. Assume · is integrally closed. An element ˛ of 1 is integral over · if and
only if its minimum polynomial over 1 has coefﬁcients in ·.
PROOF. Let ˛ be an element of 1 integral over ·, so that
˛
n
÷a
1
˛
n1
÷... ÷a
n
=0. some a
i
÷ ·.
Let }(X) be the minimum polynomial of ˛ over 1. For any root ˛
t
of }(X), the ﬁelds
1Œ˛ and 1Œ˛
t
 are both stem ﬁelds for } (see FT p9), and so there exists a 1isomorphism
o: 1Œ˛ ÷1Œ˛
t
. o(˛) =˛
t
:
On applying o to the above equation we obtain the equation
˛
tn
÷a
1
˛
tn1
÷... ÷a
n
=0.
which shows that ˛
t
is integral over ·. Hence all the roots of }(X) are integral over ·, and
it follows that the coefﬁcients of }(X) are integral over · (by 2.1). They lie in 1, and · is
integrally closed, and so they lie in ·. This proves the “only if” part of the statement, and
the “if” part is obvious.
2
REMARK 2.12 As we noted in the introduction, this makes it easy to compute some rings
of integers. For example, an element ˛ ÷ QŒ
J is integral over Z if and only if its trace
and norm both lie in Z.
23
2. RINGS OF INTEGERS
PROPOSITION 2.13 If T is integral over · and ﬁnitely generated as an ·algebra, then it
is ﬁnitely generated as an ·module.
PROOF. First consider the case that T is generated as an ·algebra by a single element, say
T =·Œˇ. By assumption
ˇ
n
÷a
1
ˇ
n1
÷ ÷a
n
=0. some a
i
÷ ·.
Every element of T can be expressed as a ﬁnite sum
c
0
÷c
1
ˇ÷c
2
ˇ
2
÷ ÷c
1
ˇ
1
. c
i
÷ ·,
and we can exploit the preceding equality to replace ˇ
n
(successively) with a linear combi
nation of lower powers of ˇ. Thus every element of T can be expressed as a ﬁnite sum
c
0
÷c
1
ˇ÷c
2
ˇ
2
÷ ÷c
n1
ˇ
n1
, c
i
÷ ·.
and so 1. ˇ. ˇ
2
. . . . . ˇ
n1
generate T as an ·module. In order to pass to the general case,
we need a lemma.
2
LEMMA 2.14 Let · c T c C be rings. If T is ﬁnitely generated as an ·module, and C
is ﬁnitely generated as a Tmodule, then C is ﬁnitely generated as an ·module.
PROOF. If {ˇ
1
. .... ˇ
n
] is a set of generators for T as an ·module, and {,
1
. .... ,
n
] is a
set of generators for C as a Tmodule, then {ˇ
i
,
}
] is a set of generators for C as an ·
module.
2
We now complete the proof of (2.13). Let ˇ
1
. . . . . ˇ
n
generate T as an ·algebra, and
consider
· c·Œˇ
1
 c·Œˇ
1
. ˇ
2
 c c·Œˇ
1
. .... ˇ
n
 =T.
We sawabove that ·Œˇ
1
 is ﬁnitely generated as an ·module. Since ·Œˇ
1
. ˇ
2
 =·Œˇ
1
Œˇ
2
,
and ˇ
2
is integral over ·Œˇ
1
 (because it is over ·), the same observation shows that
·Œˇ
1
. ˇ
2
 is ﬁnitely generated as a ·Œˇ
1
module. Now the lemma shows that ·Œˇ
1
. ˇ
2

is ﬁnitely generated as an ·module. Continuing in this fashion, we ﬁnd that T is ﬁnitely
generated as an ·module.
PROPOSITION 2.15 Consider integral domains ·cT cC; if T is integral over ·, and C
is integral over T, then C is integral over ·.
PROOF. Let , ÷ C; it satisﬁes an equation
,
n
÷b
1
,
n1
÷ ÷b
n
=0. b
i
÷ T.
Let T
t
=·Œb
1
. .... b
n
. Then T
t
is ﬁnitely generated as an ·module (by the last proposi
tion), and , is integral over T
t
(by our choice of the b
i
), and so T
t
Œ, is ﬁnitely generated
as an ·module. Since ,T
t
Œ, cT
t
Œ,, Proposition 2.4 shows that , is integral over ·.
2
COROLLARY 2.16 The integral closure of · in an algebraic extension 1 of its ﬁeld of
fractions is integrally closed.
24
Review of bases of ·modules
PROOF. Let T be the integral closure of · in 1, and let C be the integral closure of T in
1. Then C is integral over ·, and so C cT.
2
REMARK 2.17 In particular, the ring of integers in a number ﬁeld is integrally closed.
Clearly we want this, since we want our ring of integers to have the best chance of being a
unique factorization domain (see 2.9).
EXAMPLE 2.18 Let k be a ﬁnite ﬁeld, and let 1 be a ﬁnite extension of k(X). Let O
1
be
the integral closure of kŒX in 1. The arithmetic of O
1
is very similar to that of the ring
of integers in a number ﬁeld.
Review of bases of ·modules
Let M be an ·module. Recall that a set of elements e
1
. .... e
n
is a basis for M if
(a)
¸
a
i
e
i
=0, a
i
÷ · =all a
i
=0, and
(b) every element . of M can be expressed in the form . =
¸
a
i
e
i
, a
i
÷ ·.
Let {e
1
. .... e
n
] be a basis for M, and let {}
1
. .... }
n
] be a second set of n elements in
M. Then we can write }
i
=
¸
a
i}
e
}
, a
i}
÷ ·, and }
i
is also a basis if and only if the
matrix (a
i}
) is invertible in the ring M
n
(·) of n×n matrices with coefﬁcients in · (this is
obvious). Moreover (a
i}
) is invertible in M
n
(·) if and only if its determinant is a unit in
·, and in this case, the inverse is given by the usual formula:
(a
i}
)
1
=adj(a
i}
) det(a
i}
)
1
.
In the case that ·=Z, the index of N
def
=Z}
1
÷Z}
2
÷ ÷Z}
n
in M is [ det(a
i}
)[ (assum
ing this is nonzero). To prove this, recall from basic graduate algebra that we can choose
bases {e
t
i
] for M and {}
t
i
] for N such that }
t
i
=m
i
e
t
i
, m
i
÷ Z, m
i
> 0. If (e
t
i
) =U (e
i
)
and (}
t
i
) =V (}
i
), then (}
i
) =V
1
DU(e
i
) where D =diag(m
1
. . . . . m
n
), and
det(V
1
DU) =det(V
1
) det(D) det(U) =
¸
m
i
=(M : N).
Review of norms and traces
Let · c T be rings such that T is a free ·module of rank n. Then any ˇ ÷ T deﬁnes an
·linear map
. ÷ˇ.: T ÷T.
and the trace and determinant of this map are welldeﬁned. We call them the trace Tr
B¡¹
ˇ
and norm Nm
B¡¹
ˇ of ˇ in the extension T¡·. Thus if {e
1
. .... e
n
] is a basis for T over ·,
and ˇe
i
=
¸
a
i}
e
}
, then Tr
B¡¹
(ˇ) =
¸
a
i i
and Nm
B¡¹
(ˇ) =det(a
i}
). When T ¯· is a
ﬁnite ﬁeld extension, this agrees with the usual deﬁnition. The following hold (for a ÷ ·,
ˇ. ˇ
t
÷ T):
Tr(ˇ÷ˇ
t
) =Tr(ˇ) ÷Tr(ˇ
t
) Nm(ˇˇ
t
) =Nm(ˇ) Nm(ˇ
t
)
Tr(aˇ) =aTr(ˇ) Nm(a) =a
n
Tr(a) =na
25
2. RINGS OF INTEGERS
PROPOSITION 2.19 Let 1¡1 be an extension of ﬁelds of degree n, and let ˇ ÷ 1. Let
}(X) be the minimum polynomial of ˇ over 1 and let ˇ
1
=ˇ, ˇ
2
, ... , ˇ
n
be the roots of
}(X). Then
Tr
1¡1
(ˇ) =r(ˇ
1
÷ ÷ˇ
n
). Nm
1¡1
(ˇ) =(ˇ
1
ˇ
n
)
i
where r =Œ1 : 1Œˇ =n¡m.
PROOF. Suppose ﬁrst that 1 = 1Œˇ, and compute the matrix of . ÷ˇ. relative to the
basis {1. ˇ. . . . . ˇ
n1
] — one sees easily that it has trace
¸
ˇ
i
and determinant
¸
ˇ
i
. For
the general case, use the transitivity of norms and traces (see FT 5.38).
2
COROLLARY 2.20 Assume 1 is separable of degree n over 1, and let {o
1
. .... o
n
] be the
set of distinct 1homomorphisms 1 ·÷˝ where ˝ is some big Galois extension of 1
(e.g., the Galois closure of 1 over 1). Then
Tr
1¡1
(ˇ) =o
1
ˇ÷ ÷o
n
ˇ. Nm
1¡1
(ˇ) =o
1
ˇ o
n
ˇ.
PROOF. Each ˇ
i
occurs exactly r times in the family {o
i
ˇ].
2
COROLLARY 2.21 Let · be an integrally closed integral domain, and let 1 be a ﬁnite
extension of the ﬁeld of fractions 1 of ·; if ˇ ÷ 1 is integral over ·, then Tr
1¡1
(ˇ) and
Nm
1¡1
(ˇ) are in ·.
PROOF. We know that if ˇ is integral, then so also is each of its conjugates. Alternatively,
apply 2.11.
2
ASIDE 2.22 Let 1 = 1Œ˛, and let ˛
1
= ˛. ˛
2
. . . . . ˛
n
be the conjugates of ˛ (in some Galois
extension of 1 containing 1). For any ˇ =g(˛) in 1.
Nm
1¡1
(ˇ) =
¸
n
i=1
g(˛
i
). Tr
1¡1
(ˇ) =
¸
n
i=1
g(˛
i
).
This is a restatement of (2.20), and is Dedekind’s original deﬁnition (Dedekind 1877, :17).
Review of bilinear forms
Let V be a ﬁnitedimensional vector space over a ﬁeld 1. Recall that a bilinear form on V
is a 1bilinear map
[: V ×V ÷1.
Such a form is symmetric if [(.. .) = [(.. .) for all .. . ÷ V . The discriminant of a
bilinear form [ relative to a basis {e
1
. .... e
n
] of V is det([(e
i
. e
}
)). If {}
1
. .... }
n
] is a set
of elements of V , and }
}
=
¸
a
}i
e
i
, then
[(}
k
. }
/
) =
¸
i,}
[(a
ki
e
i
. a
/}
e
}
) =
¸
i,}
a
ki
[(e
i
. e
}
) a
/}
.
and so
([(}
k
. }
/
)) =· ([(e
i
. e
}
)) ·
tr
26
Discriminants
(equality of m×m matrices) where · is the invertible matrix (a
i}
). Hence
det([(}
i
. }
}
)) =det(·)
2
det([(e
i
. e
}
)) (5)
The form [ is said to be nondegenerate if it satisﬁes each of the following equivalent
conditions:
(a) [ has a nonzero discriminant relative to one (hence every) basis of V :
(b) the left kernel {· ÷ V [ [(·. .) =0 for all . ÷ V ] is zero;
(c) the right kernel of [ is zero.
Thus if [ is nondegenerate, the map · ÷(. ÷[(·. .)) from V onto the dual vector space
V
·
def
=Hom(V. 1) is an isomorphism. Let {e
1
. .... e
n
] be a basis for V , and let }
1
. .... }
n
be the dual basis in V
·
, i.e., }
i
(e
}
) =ı
i}
(Kronecker delta). We can use the isomorphism
V ÷V
·
given by a nondegenerate form [ to transfer {}
1
. .... }
n
] to a basis {e
t
1
. .... e
t
n
] of
V ; it has the property that
[(e
t
i
. e
}
) =ı
i}
.
For example, suppose {e
1
. .... e
n
] is a basis such that ([(e
i
. e
}
)) is a diagonal matrix — the
GramSchmidt process always allows us to ﬁnd such a basis when the form is symmetric
— then e
t
i
=e
i
¡[(e
i
. e
i
).
Discriminants
If 1 is a ﬁnite extension of 1 (1 and 1 ﬁelds), then
(˛. ˇ) ÷Tr
1¡1
(˛ˇ): 1×1 ÷1
is a symmetric bilinear form on 1 regarded as a vector space over 1, and the discriminant
of this form is called the discriminant of 1¡1.
More generally, let T ¯ · be rings, and assume T is free of rank m as an ·module.
Let ˇ
1
. .... ˇ
n
be elements of T. We deﬁne their discriminant to be
D(ˇ
1
. .... ˇ
n
) =det(Tr
B¡¹
(ˇ
i
ˇ
}
)).
LEMMA 2.23 If ,
}
=
¸
a
}i
ˇ
i
, a
i}
÷ ·, then
D(,
1
. .... ,
n
) =det(a
i}
)
2
D(ˇ
1
. .... ˇ
n
).
PROOF. See the proof of (5).
2
If the ˇs and ,s each form a basis for T over ·, then det(a
i}
) is a unit (see p25). Thus
the discriminant D(ˇ
1
. .... ˇ
n
) of a basis {ˇ
1
. .... ˇ
n
] of T is welldeﬁned up to multiplica
tion by the square of a unit in ·. In particular, the ideal in · that it generates is independent
of the choice of the basis. This ideal, or D(ˇ
1
. .... ˇ
n
) itself regarded as an element of
·¡·
×2
, is called the discriminant disc(T¡·) of T over ·.
For example, when we have a ﬁnite extension of ﬁelds 1¡1, disc(1¡1) is an element
of 1, welldeﬁned up to multiplication by a nonzero square in 1.
When ·=Z, disc(T¡·) is a welldeﬁned integer, because 1 is the only square of a unit
in Z.
27
2. RINGS OF INTEGERS
Warning: We shall see shortly that, when 1 is a number ﬁeld of degree m over Q,
the ring of integers O
1
in 1 is free of rank m over Z, and so disc(O
1
¡Z) is a well
deﬁned integer. Sometimes this is loosely referred to as the discriminant of 1¡Q— strictly
speaking, disc(1¡Q) is the element of Q
×
¡Q
×2
represented by the integer disc(O
1
¡Z).
PROPOSITION 2.24 Let · cT be integral domains and assume that T is a free ·module
of rank m and that disc(T¡·) =0. Elements ,
1
. .... ,
n
form a basis for T as an ·module
if and only if
(D(,
1
. .... ,
n
)) =(disc(T¡·)) (as ideals in ·).
PROOF. Let {ˇ
1
. .... ˇ
n
] be a basis for T as an ·module, and let ,
1
. .... ,
n
be any elements
of T. Write ,
}
=
¸
a
}i
ˇ
i
, a
}i
÷ ·. Then
D(,
1
. .... ,
n
)
(2.23)
= det(a
i}
)
2
D(ˇ
1
. .... ˇ
n
).
and, as we noted earlier, {,
1
. . . . . ,
n
] is a basis if and only if det(a
i}
) is a unit.
2
REMARK 2.25 Take · =Z in (2.24). Elements ,
1
. ,
2
. . . . . ,
n
generate a submodule N of
ﬁnite index in T if and only if D(,
1
. . . . . ,
n
) =0, in which case
D(,
1
. . . . . ,
n
) =(T: N)
2
disc(T¡Z).
To prove this, choose a basis ˇ
1
. . . . . ˇ
n
for T as a Zmodule, and write ,
}
=
¸
a
}i
ˇ
i
.
Then both sides equal det(a
i}
)
2
D(ˇ
1
. . . . . ˇ
n
).
PROPOSITION 2.26 Let 1 be a ﬁnite separable extension of the ﬁeld 1 of degree m, and
let o
1
. .... o
n
be the distinct 1homomorphisms of 1 into some large Galois extension ˝
of 1. Then, for any basis ˇ
1
. .... ˇ
n
of 1 over 1.
D(ˇ
1
. .... ˇ
n
) =det(o
i
ˇ
}
)
2
=0.
PROOF. By direct calculation, we have
D(ˇ
1
. . . . . ˇ
n
)
def
=det(Tr(ˇ
i
ˇ
}
))
=det(
¸
k
o
k
(ˇ
i
ˇ
}
)) (by 2.20)
=det(
¸
k
o
k
(ˇ
i
) o
k
(ˇ
}
))
=det(o
k
(ˇ
i
)) det(o
k
(ˇ
}
))
=det(o
k
(ˇ
i
))
2
.
Suppose that det(o
i
ˇ
}
) =0. Then there exist c
1
. .... c
n
÷ ˝ such that
¸
i
c
i
o
i
(ˇ
}
) =0 all ;.
2
By linearity, it follows that
¸
i
c
i
o
i
(ˇ) =0 for all ˇ ÷ 1, but this contradicts Dedekind’s
theorem on the independence of characters (apply it with G =1
×
):
Let G be a group and ˝ a ﬁeld, and let o
1
. .... o
n
be distinct homomorphisms
G ÷˝
×
; then o
1
. .... o
n
are linearly independent over ˝, i.e., there do not
exist c
i
÷ ˝ such that . ÷
¸
i
c
i
o
i
(.): G ÷˝ is the zero map (FT 5.14).
28
Rings of integers are ﬁnitely generated
COROLLARY 2.27 Let 1 be the ﬁeld of fractions of ·, and let 1 be a ﬁnite separable
extension of 1 of degree m. If the integral closure T of · in 1 is free of rank m over ·,
then disc(T¡·) =0.
PROOF. If {ˇ
1
. .... ˇ
n
] is a basis for T as an ·module, then it follows easily from (2.6)
that it is also a basis for 1 as a 1vector space. Hence disc(T¡·) represents disc(1¡1).
2
REMARK 2.28 (a) The proposition shows that the 1bilinear pairing
(ˇ. ˇ
t
) ÷Tr(ˇ ˇ
t
): 1×1 ÷1
is nondegenerate (its discriminant is disc(1¡1)).
(b) The assumption that 1¡1 is separable is essential; in fact, if 1¡1 is not separable,
then disc(1¡1) =0 (see Exercise 23).
Rings of integers are ﬁnitely generated
We now show that O
1
is ﬁnitely generated as a Zmodule.
PROPOSITION 2.29 Let · be an integrally closed integral domain with ﬁeld of fractions
1, and let T the integral closure of · in a separable extension 1 of 1 of degree m. There
exists free ·submodules M and M
t
of 1 such that
M cT cM
t
. (6)
Therefore T is a ﬁnitely generated ·module if · is Noetherian, and it is free of rank m if
· is a principal ideal domain.
PROOF. Let {ˇ
1
. .... ˇ
n
] be a basis for 1 over 1. According to (2.6), there exists a J ÷ ·
such that J ˇ
i
÷T for all i . Clearly {J ˇ
1
. . . . . J ˇ
n
] is still a basis for 1as a vector space
over 1, and so we may assume to begin with that each ˇ
i
÷ T. Because the trace pairing is
nondegenerate, there is a “dual” basis {ˇ
t
1
. .... ˇ
t
n
] of 1 over 1 such that Tr(ˇ
i
ˇ
t
}
) =ı
i}
(see the discussion following (5), p27). We shall show that
·ˇ
1
÷·ˇ
2
÷ ÷·ˇ
n
cT c·ˇ
t
1
÷·ˇ
t
2
÷ ÷·ˇ
t
n
.
Only the second inclusion requires proof. Let ˇ ÷ T. Then ˇ can be written uniquely as a
linear combination ˇ =
¸
b
}
ˇ
t
}
of the ˇ
t
}
with coefﬁcients b
}
÷ 1, and we have to show
that each b
}
÷ ·. As ˇ
i
and ˇ are in T, so also is ˇ ˇ
i
, and so Tr(ˇ ˇ
i
) ÷ · (see 2.21).
But
Tr(ˇ ˇ
i
) =Tr(
¸
}
b
}
ˇ
t
}
ˇ
i
) =
¸
}
b
}
Tr(ˇ
t
}
ˇ
i
) =
¸
}
b
}
ı
i}
=b
i
.
Hence b
i
÷ ·.
If · Noetherian, then M
t
is a Noetherian ·module (see 1.8), and so T is ﬁnitely
generated as an ·module. If · is a principal ideal domain, then T is free of rank _ m
because it is contained in a free ·module of rank m, and it has rank _mbecause it contains
a free ·module of rank m (see any basic graduate algebra course).
2
29
2. RINGS OF INTEGERS
COROLLARY 2.30 The ring of integers in a number ﬁeld 1 is the largest subring that is
ﬁnitely generated as a Zmodule.
PROOF. We have just seen that O
1
is a ﬁnitely generated Zmodule. Let T be another
subring of 1 that is ﬁnitely generated as a Zmodule; then every element of T is integral
over Z (by 2.4), and so T cO
1
.
2
REMARK 2.31 (a) The hypothesis that 1¡1 be separable is necessary to conclude that
T is a ﬁnitely generated ·module (we used that the trace pairing was nondegenerate).
However it is still true that the integral closure of kŒX in any ﬁnite extension of k(X) (not
necessarily separable) is a ﬁnitely generated kŒXmodule.
(b) The hypothesis that · be a principal ideal domain is necessary to conclude from (6)
that T is a free ·module — there do exist examples of number ﬁelds 1¡1 such that O
1
is not a free O
1
module.
(c) Here is an example of a ﬁnitely generated module that is not free. Let ·=ZŒ
÷5,
and consider the ·modules
(2) c(2. 1÷
÷5) cZŒ
÷5.
Both (2) and ZŒ
÷5 are free ZŒ
÷5modules of rank 1, but (2. 1 ÷
÷5) is not a free
ZŒ
÷5module of rank 1, because it is not a principal ideal (see the Introduction). In fact,
it is not a free module of any rank.
DEFINITION 2.32 When 1 is a number ﬁeld, a basis ˛
1
. .... ˛
n
for O
1
as a Zmodule is
called an integral basis for 1.
REMARK 2.33 We retain the notations of the proposition and its proof.
(a) Let C =
¸
·ˇ
i
cT, with ˇ
i
a basis for 1 over 1. Deﬁne
C
+
={ˇ ÷ 1 [ Tr(ˇ,) ÷ · for all , ÷ C].
By linearity,
ˇ ÷ C
+
” Tr(ˇˇ
i
) ÷ · for i =1. .... m.
and it follows that
C
+
=
¸
·ˇ
t
i
.
Thus we have:
C =
¸
·ˇ
i
cT c
¸
·ˇ
t
i
=C
+
.
(b) Write 1 =QŒˇ with ˇ ÷ T, and let }(X) be the minimum polynomial of ˇ. Let
C =ZŒˇ =Z1÷Zˇ÷ ÷Zˇ
n1
. We want to ﬁnd C
+
.
One can show (Artin 1959, Chapter 7) that
Tr(ˇ
i
¡}
t
(ˇ)) =
0 if 0 _i _m÷2
1 if i =m÷1
(these formulas go back to Euler). It follows from this that
det(Tr(ˇ
i
ˇ
}
¡}
t
(ˇ)) =(÷1)
n
30
Finding the ring of integers
(the only term contributing to the determinant is the product of the elements on the other
diagonal). If ˇ
t
1
. .... ˇ
t
n
is the dual basis to 1. ˇ. . . . . ˇ
n1
, so that Tr(ˇ
i
ˇ
t
}
) =ı
i}
, then
det(Tr(ˇ
i
ˇ
t
}
)) =1.
On comparing these formulas, one sees that the matrix relating the family
{1¡}
t
(ˇ). .... ˇ
n1
¡}
t
(ˇ)]
to the basis
{ˇ
t
1
. .... ˇ
t
n
]
has determinant ˙1, and so it is invertible in M
n
(·). Thus we see that C
+
is a free ·
module with basis {1¡}
t
(ˇ). . . . . ˇ
n1
¡}
t
(ˇ)]:
C =·Œˇ cT c}
t
(ˇ)
1
·Œˇ =C
+
.
Finding the ring of integers
We now assume 1 to be a ﬁeld of characteristic zero.
PROPOSITION 2.34 Let 1 =1Œˇ some ˇ, and let }(X) be the minimum polynomial of
ˇ over 1. Suppose that }(X) factors into
¸
(X ÷ˇ
i
) over the Galois closure of 1. Then
D(1. ˇ. ˇ
2
. . . . . ˇ
n1
) =
¸
1_i~}_n
(ˇ
i
÷ˇ
}
)
2
=(÷1)
n(n1)¡2
Nm
1¡1
(}
t
(ˇ)).
PROOF. We have
D(1. ˇ. ˇ
2
. . . . . ˇ
n1
) =det(o
i
(ˇ
}
))
2
(2.26)
=det(ˇ
}
i
)
2
=(
¸
i~}
(ˇ
i
÷ˇ
}
))
2
(Vandermonde)
=(÷1)
n(n1)¡2
¸
i
(
¸
}=i
(ˇ
i
÷ˇ
}
))
=(÷1)
n(n1)¡2
¸
}
}
t
(ˇ
}
)
=(÷1)
n(n1)¡2
Nm(}
t
(ˇ)).
2
The number in (2.34) is called the discriminant of }(X). It can also be deﬁned as the
resultant of }(X) and }
t
(X). The discriminant of } lies in 1, and it is zero if and only if
} has a repeated root. It is a symmetric polynomial in the ˇ
i
with coefﬁcients in 1, and so
(by 2.2) it can be expressed in terms of the coefﬁcients of }(X), but the formulas are quite
complicated.
EXAMPLE 2.35 We compute the discriminant of
}(X) =X
n
÷aX ÷b, a. b ÷ 1.
assumed to be irreducible and separable. Let ˇ be a root of }(X), and let
, =}
t
(ˇ) =nˇ
n1
÷a.
31
2. RINGS OF INTEGERS
We compute Nm
1Œˇ]¡1
(,). On multiplying the equation
ˇ
n
÷aˇ÷b =0
by nˇ
1
and rearranging, we obtain the equation
nˇ
n1
=÷na÷nbˇ
1
.
Hence
, =nˇ
n1
÷a =÷(n÷1)a÷nbˇ
1
.
Solving for ˇ gives
ˇ =
÷nb
, ÷(n÷1)a
.
from which it is clear that 1Œˇ =1Œ,, and so the minimum polynomial of , over 1 also
has degree n. If we write
}
÷nb
X ÷(n÷1)a
=1(X)¡O(X).
then 1(,)¡O(,) =}(ˇ) =0 and so 1(,) =0. Since
1(X) =(X ÷(n÷1)a)
n
÷na(X ÷(n÷1)a)
n1
÷(÷1)
n
n
n
b
n1
is monic of degree n, it must be the minimum polynomial of ,. Therefore Nm(,) is (÷1)
n
times the constant term of this polynomial, and so we ﬁnd that
Nm(,) =n
n
b
n1
÷(÷1)
n1
(n÷1)
n1
a
n
.
Finally we obtain the formula:
disc(X
n
÷aX ÷b) =(÷1)
n(n1)¡2
(n
n
b
n1
÷(÷1)
n1
(n÷1)
n1
a
n
)
For example:
disc(X
2
÷aX ÷b) =÷4b ÷a
2
.
disc(X
3
÷aX ÷b) =÷27b
2
÷4a
3
.
disc(X
4
÷aX ÷b) =256b
3
÷27a
4
,
disc(X
5
÷aX ÷b) =5
5
b
4
÷4
4
a
5
.
For any polynomials more complicated than the above, use a computer program. For
example, typing
poldisc(X^3+a*X^2+b*X+c)
in PARI returns
4*c*a^3 + b^2*a^2 + 18*c*b*a + (4*b^3  27*c^2)
i.e., ÷4ca
3
÷b
2
a
2
÷18cba÷(÷4b
3
÷27c
2
).
The general strategy for ﬁnding the ring of integers of 1 is to write 1 =QŒ˛ with ˛
an integer in 1, and compute D(1. ˛. .... ˛
n1
). It is an integer, and if it is squarefree, then
{1. ˛. .... ˛
n1
] is automatically an integral basis, because (see 2.25)
D(1. ˛. . . . . ˛
n1
) =disc(O
1
¡Z) (O
1
: ZŒ˛)
2
. (7)
If it is not squarefree, {1. ˛. .... ˛
n1
] may still be an integral basis, and sometimes one
can tell this by using Stickelberger’s theorem (see 2.40 below) or by looking at how primes
ramify (see later). If {1. ˛. .... ˛
n1
] is not an integral basis, one has to look for algebraic
integers not in
¸
Z ˛
i
(we describe an algorithm below).
32
Finding the ring of integers
EXAMPLE 2.36 The polynomial X
3
÷X÷1 is irreducible
1
in QŒX, because, if it factored,
it would have a root in Q, which would be an integer dividing 1. Let ˛ be a root of X
3
÷
X ÷1. We have
D(1. ˛. ˛
2
) =disc(}(X)) =÷23.
which contains no square factor, and so {1. ˛. ˛
2
] is an integral basis for QŒ˛ (and ZŒ˛ is
the ring of integers in QŒ˛).
EXAMPLE 2.37 The polynomial X
3
÷X ÷1 is irreducible in QŒX, and, for any root
˛ of it, D(1. ˛. ˛
2
) = disc(}(X)) = ÷31, which contains no square factor, and so again
{1. ˛. ˛
2
] is an integral basis for QŒ˛.
EXAMPLE 2.38 This example goes back to Dedekind. Let 1 =QŒ˛, where ˛ is a root of
}(X) =X
3
÷X
2
÷2X ÷8.
The discriminant of } is ÷2012 =÷4 503, but Dedekind showed that O
1
=ZŒˇ, and so
disc(O¡Z) =÷503. In fact Dedekind showed that there is no integral basis of the form 1,
ˇ, ˇ
2
(see Weiss 1963, p170; for another example of this type, see Exercise 26.)
EXAMPLE 2.39 Consider the ﬁeld QŒ˛ where ˛ is a root of }(X) = X
5
÷X ÷1. This
polynomial is irreducible, because it is irreducible in F
3
ŒX. The discriminant of }(X) is
2869 =19 151, and so the ring of integers in QŒ˛ is ZŒ˛.
PROPOSITION 2.40 Let 1 be an algebraic number ﬁeld.
(a) The sign of disc(1¡Q) is (÷1)
x
, where 2s is the number of homomorphisms 1 ·÷C
whose image is not contained in 1.
(b) (Stickelberger’s theorem) disc(O
1
¡Z) ÷0 or 1 mod 4.
PROOF. (a) Let 1 =QŒ˛, and let ˛
1
=˛, ˛
2
, ..., ˛
i
be the real conjugates of ˛ and ˛
i÷1
,
¨ ˛
i÷1
. .... ˛
i÷x
. ¨ ˛
i÷x
the complex conjugates. Then
sign(D(1. .... ˛
n1
)) =sign
¸
1_i_x
(˛
i÷i
÷ ¨ ˛
i÷i
)
2
because the other terms are either squares of real numbers or occur in conjugate pairs, and
this equals (÷1)
x
.
(b) Recall that disc(O
1
¡Z) =det(o
i
˛
}
)
2
, where ˛
1
. .... ˛
n
is an integral basis. Let 1
be the sum of the terms in the expansion of det(o
i
˛
}
) corresponding to even permutations,
and ÷N the sum of the terms corresponding to odd permutations. Then
disc(O
1
¡Z) =(1 ÷N)
2
=(1 ÷N)
2
÷41N.
If t is an element of the Galois group of the Galois closure of 1 over Q, then either t1 =1
and tN =N, or t1 =N and tN =1. In either case, t ﬁxes 1 ÷N and 1N, and so they
are rational numbers. As they are integral over Z, they must in fact be integers, from which
it follows that
disc(O
1
¡Z) ÷(1 ÷N)
2
÷0 or 1 mod 4.
2
1
In fact, this is the monic irreducible cubic polynomial in ZŒX with the smallest discriminant.
33
2. RINGS OF INTEGERS
EXAMPLE 2.41 Consider the ﬁeld QŒ
m, where m is a squarefree integer.
Case m ÷2. 3 mod 4. Here D(1.
m) =disc(X
2
÷m) =4m, and so Stickelberger’s
theorem shows that disc(O
1
¡Z) =4m, and hence {1.
m] is an integral basis.
Case m ÷1 mod 4. The element (1 ÷
m)¡2 is integral because it is a root of X
2
÷
X ÷(1 ÷m)¡4. As D(1. (1 ÷
m)¡2) = m, we see that {1. (1 ÷
m)¡2] is an integral
basis.
REMARK 2.42 Let 1 and 1
t
be number ﬁelds. If 1 and 1
t
are isomorphic, then they
have the same degree and the same discriminant, but the converse statement is false. For
example, there are four nonisomorphic cubic number ﬁelds with discriminant ÷4027 (4027
is prime). See (3.48) and (3.49) for two of them.
The curious may wonder why we didn’t give an example of a ﬁeld generated over Q by
an integral element whose minimum polynomial has discriminant ˙1. The reason is that
there is no such polynomial of degree > 1 — see the discussion following Theorem 4.9
below.
2
Algorithms for ﬁnding the ring of integers
By an algorithm I mean a procedure that could (in principle) be put on a computer and is
guaranteed to lead to the answer in a ﬁnite number of steps. Suppose the input requires
N digits to express it. A good algorithm is one whose running time is < N
c
for some
c. For example, there is no known good algorithm for factoring an integer. By a practical
algorithm I mean one that has been (or should have been) put on a computer, and is actually
useful.
The following variant of (2.29) is useful. Let · be a principal ideal domain with ﬁeld
of fractions 1, and let T be the integral closure of · in a ﬁnite separable extension 1 of 1
of degree m.
PROPOSITION 2.43 Let ˇ
1
. .... ˇ
n
be a basis for 1 over 1 consisting of elements of T,
and let J =disc(ˇ
1
. .... ˇ
n
). Then
· ˇ
1
÷... ÷· ˇ
n
cT c· (ˇ
1
¡J) ÷... ÷· (ˇ
n
¡J).
PROOF. Let ˇ ÷ T, and write
ˇ =.
1
ˇ
1
÷ ÷.
n
ˇ
n
. .
i
÷ 1.
Let o
1
. . . . . o
n
be the distinct 1embeddings of 1 into some large Galois extension ˝ of
1. On applying the o’s to this equation, we obtain a system of linear equations:
o
i
ˇ =.
1
o
i
ˇ
1
÷.
2
o
i
ˇ
2
÷ ÷.
n
o
i
ˇ
n
. i =1. . . . . m.
Hence by Cramer’s rule
.
i
=,
i
¡ı
where ı =det(o
i
ˇ
}
) and ,
i
is the determinant of the same matrix, but with the i th column
replaced with (o
i
ˇ). From (2.34), we know that ı
2
= J. Thus .
i
= ,
i
ı¡J, and ,
i
ı is
an element of 1 (because it equals J.
i
) and is integral over ·. Therefore ,
i
ı ÷ ·, which
completes the proof.
2
2
In fact, the smallest discriminant is 3, which occurs for QŒ
÷3.
34
Algorithms for ﬁnding the ring of integers
Thus there is the following algorithm for ﬁnding the ring of integers in a number ﬁeld
1. Write 1 =QŒ˛ where ˛ is integral over Q. Compute J =disc(1. ˛. .... ˛
n1
). Then
ZŒ˛ cO
1
cJ
1
ZŒ˛.
Note that (J
1
ZŒ˛: ZŒ˛) = J
n
, which is huge but ﬁnite. Each coset ˇ ÷ZŒ˛, ˇ ÷
J
1
ZŒ˛, consists entirely of algebraic integers or contains no algebraic integer. Find a
set of representatives ˇ
1
. .... ˇ
n
for ZŒ˛ in J
1
ZŒ˛, and test each to see whether it is in
tegral over Z (the coefﬁcients of its minimum polynomial will have denominators bounded
by a power of J, and so it is possible to tell whether or not they are integers by computing
them with sufﬁcient accuracy).
Unfortunately this method is not practical. For example,
}(X) =X
5
÷17X
4
÷3X
3
÷2X
2
÷X ÷1
is irreducible, and has discriminant 285401001. Hence, if ˛ is a root of }(X) and 1 =
QŒ˛, then the index of ZŒ˛ in Z
1
d
÷Z
˛
d
÷ ÷Z
˛
4
d
is (285401001)
5
. Actually, as luck
would have it, 285401001 =3 179 233 2281 is squarefree, and so O
1
=ZŒ˛.
Note that PARI can compute the minimum polynomial of an algebraic number. For
example, let a =
3
1÷
7. We ﬁrst type “a=sqrtn(1+sqrt(7),3)” in PARI, which
reports that a=1.539084083333266359084139071. Now “algdep(a,6)” asks PARI for a
minimum polynomial for a of degree at most 6, which (correctly) reports it to be
X
6
÷2X
3
÷6 =(X
3
÷1)
2
÷7.
Unfortunately, of course, PARI will ﬁnd a “minimum polynomial” for a even when a is
transcendental.
I now discuss a practical algorithm for ﬁnding O
1
for small degrees and small dis
criminants from Pohst and Zassenhaus 1989 (see the additional references at the end of this
section). The next result will help us get an idea of what should be possible.
LEMMA 2.44 Let (·. ı) be Euclidean domain, and let M be an m×m matrix with coef
ﬁcients in ·. Then it is possible to put M into upper triangular form by elementary row
operations of the following type:
(r1) add a multiple of one row to a second;
(r2) swap two rows.
PROOF. By deﬁnition ı: · ÷ Z is a function with the following property: for any two
elements a. b of · with a =0, there exist elements q and r such that
b =qa÷r, with r =0 or ı(r) <ı(a).
Apply an operation of type (r2) so that the element of the ﬁrst column with the minimumı is
in the (1. 1)position. If a
11
divides all elements in the ﬁrst column, we can use operations
of type (r1) to make all the remaining elements of the ﬁrst column zero. If not, we can use
(r1) to get an element in the ﬁrst column that has smaller ıvalue than a
11
, and put that
in the (1. 1) position. Repeat — eventually, we will have the gcd of the original elements
in the ﬁrst column in the (1. 1) position and zeros elsewhere. Then move onto the next
column. . . .
2
35
2. RINGS OF INTEGERS
REMARK 2.45 (a) The operations (r1) and (r2) are invertible in matrices with coefﬁcients
in ·, and they correspond to multiplying on the left with an invertible matrix in M
n
(·).
Hence we have shown that there exists an invertible matrix U in M
n
(·) such that UM is
upper triangular. On taking transposes, we ﬁnd that for any matrix M ÷ M
n
(·), there is an
invertible matrix U in M
n
(·) such that MU is lower triangular.
(b) Take · =Z (for simplicity), and add the (invertible) operation:
(r3) multiply a row by ÷1.
Using (r1,r2,r3), it is possible to make the triangular matrix T = UM satisfy the fol
lowing conditions (assuming det(M) =0):
a
i i
>0 for all i :
the elements a
i}
of the ; th column satisfy 0 _a
i}
<a
}}
.
Then T is unique. It is called the Hermite normal form of ·.
Consider the ﬁeld 1 =QŒ˛ generated over Qby the algebraic integer ˛ with minimum
polynomial }(X). Let {o
1
. .... o
n
] be a basis for O
1
as a Zmodule, and write
· =M ˝
where · =(1. ˛. .... ˛
n1
)
tr
and ˝ =(o
1
. .... o
n
)
tr
. Choose U so that MU is lower trian
gular (and in Hermite normal form), and write
· =MU U
1
˝ =T ˝
t
.
Here ˝
t
def
= U
1
˝ is again a Zbasis for O
1
, and ˝
t
= T
1
· with T
1
also lower
triangular (but not necessarily with integer coefﬁcients). Thus
o
t
1
=a
11
1:
o
t
2
=a
21
1÷a
22
˛:
etc.,
where J a
i}
÷ Z, J =[ det(M)[ =[ det(T )[.
EXAMPLE 2.46 Let 1 =QŒ
m, m squarefree, m÷1 (mod 4). The integral basis
1.
1÷
m
2
is of the above form.
In Pohst and Zassenhaus 1989, 4.6, there is an algorithm that, starting from a monic
irreducible polynomial
}(X) =X
n
÷a
1
X
n1
÷ ÷a
n
. a
n
÷ Z.
constructs an integral basis o
1
. .... o
n
, such that
o
i
=
i
¸
k=1
a
ik
˛
i
¡N
i
where
˛ is a root of }(X). a
ik
÷ Z. N
i
÷ Z. gcd(a
i1
. .... a
i i
) =1.
36
Algorithms for ﬁnding the ring of integers
In an Appendix, they use it to show that QŒ˛, where ˛ is a root of
}(X) =X
11
÷101X
10
÷4151X
9
÷ ÷332150625.
has an integral basis
o
1
=1.
o
2
=(1¡2)˛ ÷1¡2
o
3
=(1¡4)˛
2
÷1¡4
o
4
=(1¡8)˛
3
÷(1¡8)˛
2
÷(1¡8)˛ ÷1¡8
...........
o
11
=(1¡9103145472000)˛
10
÷ ÷4064571¡49948672.
The discriminant of } is 2
130
×3
12
×5
12
×29
1S
×82231
6
, and the index of ZŒ˛ in O
1
is 2
56
×3
6
×5
3
×29
9
.
The ﬁrst step is to compute D(1. ˛. ˛
2
. . . .) =disc(}(X)) and to ﬁnd its square factors.
Finding the square factors of disc(}(X)) is the most timeconsuming part of the algorithm.
The time taken to factor an Ndigit number is exponential in the number of digits of N.
Every computer can factor a 50 digit number easily, but after that it becomes rapidly more
difﬁcult. Hundred digit numbers are already difﬁcult. Thus this is not a good algorithm in
the above sense. Once one has found the square factors of disc(}(X)) the algorithm for
computing an integral basis of the above form is good.
USING PARI
To determine whether a polynomial } is irreducible, use polisirreducible(f). For
example, polisirreducible(X^5+17*X^4+3*X^3+2*X^2+X+1) returns 1, which means
that X
5
÷17X
4
÷3X
3
÷2X
2
÷X ÷1 is irreducible, and polisirreducible(X^21)
returns 0, which means that X
2
÷1 is reducible.
To ﬁnd the discriminant of a polynomial } , use poldisc(f). For example,
poldisc(X^5+17*X^4+3*X^3+2*X^2+X+1) returns 285401001, and poldisc(X^2+3)
returns 12.
To study the stem ﬁeld of a polynomial } , use nfinit(f). For example,
nfinit(X^55*X^3+4*X1) returns
[X^5  5*X^3 + 4*X  1, [5, 0], 38569, ...]
which means that X
5
÷5X
3
÷4X÷1 has 5 real roots and no nonreal roots and that its stem
ﬁeld QŒ˛ has discriminant 38569. Moreover, typing
nfbasis(X^55*X^3+4*X1) returns
[1, X, X^2, X^3, X^4],
which means that
˚
1. ˛. ˛
2
. ˛
3
. ˛
4
¸
is an integral basis for QŒ˛ (cf. p35).
On the other hand, typing
nfinit(X^2+3) returns
[X^2 + 3, [0, 1], 3, ...]
which means that, X
2
÷3 has no real roots and one conjugate pair of complex roots, and
that the ﬁeld QŒ
÷3 has discriminant ÷3. Moreover, typing
nfbasis(X^2+3) returns
[1, 1/2*X + 1/2],
which means that
1.
1
2
÷3÷
1
2
¸
is an integral basis for QŒ
÷3.
For Dedekind’s polynomial in (2.38), PARI says that it has one real root and one conju
gate pair of nonreal roots, and that its stem ﬁeld has discriminant ÷503. It ﬁnds the integral
37
2. RINGS OF INTEGERS
basis {1. ˛.
1
2
˛
2
÷
1
2
˛]. Note that
ZŒ˛ =ZŒ1. ˛. ˛
2
÷˛.
and that
(O
1
: ZŒ˛) =2 =
÷2012
÷503
,
as predicted by Equation 7, p32.
NOTES As noted earlier, it was Dedekind who found the correct deﬁnition of the ring of integers
in a number ﬁelds. Earlier authors either luckily chose the correct ring, e.g., Kummer chose ZŒ¯,
¯
n
=1, which is the ring of integers in QŒ¯, or unluckily chose the wrong ring, e.g., Euler gave a
proof of Fermat’s last theorem for the exponent 3, which becomes correct when the ring ZŒ
÷3 is
replaced in the proof by its integral closure ZŒ¯, ¯
3
=1.
Exercises
21 Since ZŒ
5 is not integrally closed, it can not be a unique factorization domain.
Give an example of an element of ZŒ
5 that has two distinct factorizations into irreducible
elements.
22 Let ·be an integrally closed ring, and let 1 be its ﬁeld of fractions. Let }(X) ÷ ·ŒX
be a monic polynomial. If }(X) is reducible in 1ŒX, show that it is reducible in ·ŒX.
23 Show that if 1¡1 is not separable, then disc(1¡1) =0.
24 Let a =(2. 1 ÷
÷3) in ZŒ
÷3. Show that a =(2), but a
2
=(2)a. Conclude that
ideals in ZŒ
÷3 do not factor uniquely into prime ideals. (Hence ZŒ
÷3 is the wrong
choice for the ring of integers in QŒ
÷3.)
25 Let · be a subring of a ring T, and let ˇ be a unit in T. Show that every ˛ ÷
·Œˇ ¨·Œˇ
1
 is integral over ·. [This has a short solution, but it’s not obvious.]
26 Let 1 =QŒ
7.
10, and let ˛ be an algebraic integer in 1. The following argument
will show that O
1
=ZŒ˛.
(a) Consider the four algebraic integers:
˛
1
=(1÷
7)(1÷
10):
˛
2
=(1÷
7)(1÷
10):
˛
3
=(1÷
7)(1÷
10):
˛
4
=(1÷
7)(1÷
10).
Show that all the products ˛
i
˛
}
, i =; , are divisible by 3 in O
1
, but that 3 does not divide
any power of any ˛
i
. [Hint: Show that ˛
n
i
¡3 is not an algebraic integer by considering its
trace: show that Tr(˛
n
i
) ÷(
¸
˛
n
}
) ÷4
n
(mod 3) in ZŒ˛; deduce Tr(˛
n
i
) ÷1 (mod 3) in Z.]
(b) Assume now that O
1
=ZŒ˛ — we shall derive a contradiction. Let }(X) be the
minimum polynomial of ˛ over Q. For g(X) ÷ ZŒX, let ¨ g(X) denote the image of g in
38
Exercises
F
3
ŒX, F
3
=Z¡(3). Show that g(˛) is divisible by 3 in ZŒ˛ if and only if ¨ g is divisible by
¨
} in F
3
ŒX.
(c) For each i , 1 _ i _ 4, let }
i
be a polynomial in ZŒX such that ˛
i
=}
i
(˛). Show
that
¨
} [
¨
}
i
¨
}
}
(i =; ) in F
3
ŒX, but that
¨
} does not divide
¨
}
n
i
for any n. Conclude that for
each i ,
¨
} has an irreducible factor which does not divide
¨
}
i
but does divide all
¨
}
}
, ; =i .
(d) This shows that
¨
} has at least four distinct irreducible factors over F
3
. On the other
hand, } has degree at most 4. Why is this a contradiction?
39
CHAPTER 3
Dedekind Domains; Factorization
Es steht schon bei Dedekind.
(It’s already in Dedekind.)
Emmy Noether
In this Chapter, we deﬁne the notion of a Dedekind domain, and prove that
˘ ideals in Dedekind domains factor uniquely into products of prime ideals, and
˘ rings of integers in number ﬁelds are Dedekind domains,
but ﬁrst we consider a local version of a Dedekind domain.
Discrete valuation rings
The following conditions on a principal ideal domain are equivalent:
(a) · has exactly one nonzero prime ideal;
(b) up to associates, · has exactly one prime element;
(c) · is local and is not a ﬁeld.
A ring satisfying these conditions is called a discrete valuation ring. Later we shall deﬁne
discrete valuations, and so justify the name.
EXAMPLE 3.1 The ring Z
(;)
def
={
n
n
÷ Q[ n not divisible by ]] is a discrete valuation ring
with (]) as its unique nonzero prime ideal. The units in Z
(;)
are the nonzero elements m¡n
with neither m nor n divisible by ], and the prime elements are those of the form unit×].
In a discrete valuation ring · with prime element ¬, nonzero elements of · can be
expressed uniquely as u¬
n
with u a unit and m _ 0 (and m > 0 unless the element is a
unit). Every nonzero ideal in · is of the form (¬
n
) for a unique m ÷ N. Thus, if a is an
ideal in · and p denotes the (unique) maximal ideal of ·, then a =p
n
for a welldeﬁned
integer m_0.
Recall that, for an ·module M and an m÷ M, the annihilator of m
Ann(m) ={a ÷ · [ am=0].
It is an ideal in ·, which is proper if m = 0. Suppose that · is a discrete valuation ring,
and let c be a nonzero element of ·. Let M =·¡(c). What is the annihilator of a nonzero
40
Discrete valuation rings
element b ÷(c) of M? Fix a prime element ¬ of ·, and let c =u¬
n
, b =·¬
n
with u and
· units. Then n <m (else b ÷(c) =0 in M), and
Ann(b ÷(c)) =(¬
nn
).
Thus, a b for which Ann(b ÷(c)) is maximal, is of the form ·¬
n1
, and for this choice
Ann(b ÷(c)) is a prime ideal generated by
c
b
. We shall exploit these observations in the
proof of the next proposition, which gives a criterion for a ring to be a discrete valuation
ring.
PROPOSITION 3.2 An integral domain · is a discrete valuation ring if and only if
(a) · is Noetherian,
(b) · is integrally closed, and
(c) · has exactly one nonzero prime ideal.
PROOF. The necessity of the three conditions is obvious, and so let ·be an integral domain
satisfying (a), (b), and (c). We have to show that every ideal in · is principal. As a ﬁrst
step, we prove that the nonzero prime ideal is principal. Note that (c) implies that · is a
local ring.
Choose an element c ÷ ·, c =0, c =unit, and consider the ·module M
def
=·¡(c). For
each nonzero element m of M,
Ann(m) ={a ÷ · [ am=0]
is a proper ideal in ·. Because · is Noetherian, we can choose an m so that Ann(m) is
maximal among these ideals. Write m =b ÷(c) and p =Ann(b ÷(c)). Note that c ÷ p,
and so p =0, and that
p ={a ÷ · [ c[ab].
I claim that p is prime. If not there exist elements ., . ÷ · such that .. ÷ p but neither
. nor . ÷ p. Then .b ÷(c) is a nonzero element of M because . … p. Consider Ann(.b ÷
(c)). Obviously it contains p and it contains ., but this contradicts the maximality of p
among ideals of the form Ann(m). Hence p is prime.
I claim that
b
c
… ·. Otherwise b =c
b
c
÷ (c), and m=0 (in M).
I claim that
c
b
÷ ·, and p =(
c
b
). By deﬁnition, pb c(c), and so p
b
c
c·, and it is an
ideal in ·. If p
b
c
cp, then
b
c
is integral over · (by 2.4, since p is ﬁnitely generated), and
so
b
c
÷ · (because of condition (b)), but we know
b
c
… ·. Thus p
b
c
=· (by (c)), and this
implies that p =(
c
b
).
Let ¬ =
c
b
, so that p =(¬). Let a be a proper ideal of ·, and consider the sequence
a ca¬
1
ca¬
2
c .
If a¬
i
=a¬
i1
for some r, then ¬
1
(a¬
i
) =a¬
i
, and ¬
1
is integral over · (by
2.4), and so lies in · — this is impossible (¬ is not a unit in ·). Therefore the sequence
is strictly increasing, and (again because · is Noetherian) it can’t be contained in ·. Let
m be the smallest integer such that a¬
n
c · but a¬
n1
_ ·. Then a¬
n
_ p, and so
a¬
n
=·. Hence a =(¬
n
).
2
41
3. DEDEKIND DOMAINS; FACTORIZATION
Dedekind domains
DEFINITION 3.3 A Dedekind domain is an integral domain ·, not equal to a ﬁeld, such
that
(a) · is Noetherian,
(b) · is integrally closed, and
(c) every nonzero prime ideal is maximal.
Thus Proposition 3.2 says that a local integral domain is a Dedekind domain if and only
if it is a discrete valuation ring.
PROPOSITION 3.4 Let · be a Dedekind domain, and let S be a multiplicative subset of ·.
Then S
1
· is either a Dedekind domain or a ﬁeld.
PROOF. Condition (c) says that there is no containment relation between nonzero prime
ideals of ·. If this condition holds for ·, then (1.12) shows that it holds for S
1
·. Condi
tions (a) and (b) follow from the next lemma.
2
PROPOSITION 3.5 Let · be an integral domain, and let S be a multiplicative subset of ·.
(a) If · is Noetherian, then so also is S
1
·.
(b) If · is integrally closed, then so also is S
1
·.
PROOF. (a) Let a be an ideal in S
1
·. Then a = S
1
(a ¨·) (see 1.11), and so a is
generated by any (ﬁnite) set of generators for a¨·.
(b) Let ˛ be an element of the ﬁeld of fractions of · (=ﬁeld of fractions of S
1
·) that
is integral over S
1
·. Then
˛
n
÷a
1
˛
n1
÷ ÷a
n
=0, some a
i
÷ S
1
·.
For each i , there exists an s
i
÷ S such that s
i
a
i
÷ ·. Set s = s
1
s
n
÷ S, and multiply
through the equation by s
n
:
(s˛)
n
÷sa
1
(s˛)
n1
÷ ÷s
n
a
n
=0.
This equation shows that s˛ is integral over ·, and so lies in ·. Hence ˛ = (s˛)¡s ÷
S
1
·.
2
COROLLARY 3.6 For any nonzero prime ideal p in a Dedekind domain ·, the localization
·
p
is a discrete valuation ring.
PROOF. We saw in (1.13a) that ·
p
is local, and the proposition implies that it is Dedekind.
2
42
Unique factorization of ideals
Unique factorization of ideals
The main result concerning Dedekind domains is the following.
THEOREM 3.7 Let · be a Dedekind domain. Every proper nonzero ideal a of · can be
written in the form
a =p
i
1
1
p
i
n
n
with the p
i
distinct prime ideals and the r
i
>0; the p
i
and the r
i
are uniquely determined.
The proof will require several lemmas.
LEMMA 3.8 Let · be a Noetherian ring; then every ideal a in · contains a product of
nonzero prime ideals.
PROOF. (Note the similarity to the proof of 1.4.) Suppose not, and choose a maximal
counterexample a. Then a itself can not be prime, and so there exist elements . and . of ·
such that .. ÷ a but neither . nor . ÷ a. The ideals a÷(.) and a÷(.) strictly contain a,
but their product is contained in a. Because a is a maximal counterexample to the statement
of the lemma, each of a÷(.) and a÷(.) contains a product of prime ideals, and it follows
that a contains a product of prime ideals.
2
LEMMA 3.9 Let · be a ring, and let a and b be relatively prime ideals in ·; for any m,
n ÷ N, a
n
and b
n
are relatively prime.
PROOF. If a
n
and b
n
are not relatively prime, then they are both contained in some prime
(even maximal) ideal p. But if a prime ideal contains a power of an element, then it contains
the element, and so p ¯a
n
=p ¯a and p ¯b
n
=p ¯b. Thus a and b are both contained
in p, which contradicts the hypothesis.
Alternative proof: We are given that there exist elements a ÷ · and b ÷ T such that
a÷b =1. Consider
1 =(a÷b)
i
=a
i
÷
i
1
a
i1
b ÷ ÷b
i
.
If r _ m÷n, then the term on the right is the sum of an element of a
n
with an element of
b
n
.
2
If p and p
t
are distinct prime ideals of a Dedekind domain, then condition (c) of the
deﬁnition implies that p and p
t
are relatively prime, and the lemma shows that p
n
and p
tn
are also relatively prime for all m. n _1.
LEMMA 3.10 Let p be a maximal ideal of a ring ·, and let q be the ideal it generates in
·
p
, q =p·
p
. The map
a÷p
n
÷a÷q
n
: ·¡p
n
÷·
p
¡q
n
is an isomorphism.
43
3. DEDEKIND DOMAINS; FACTORIZATION
PROOF. We ﬁrst show that the map is onetoone. For this we have to show that q
n
¨· =
p
n
. But q
n
=S
1
p
n
, S =·÷p, and so we have to show that p
n
=(S
1
p
n
) ¨·. An
element of (S
1
p
n
) ¨· can be written a = b¡s with b ÷ p
n
, s ÷ S, and a ÷ ·. Then
sa ÷ p
n
, and so sa = 0 in ·¡p
n
. The only maximal ideal containing p
n
is p (because
m¯p
n
=m¯p), and so the only maximal ideal in ·¡p
n
is p¡p
n
; in particular, ·¡p
n
is
a local ring. As s ÷p
n
is not in p¡p
n
, it is a unit in ·¡p
n
, and so sa =0 in ·¡p
n
=a =0
in ·¡p
n
, i.e., a ÷ p
n
.
We now prove that the map is surjective. Let
o
x
÷ ·
p
. Because s … p and p is maximal,
we have that (s) ÷p = ·, i.e., (s) and p are relatively prime. Therefore (s) and p
n
are
relatively prime, and so there exist b ÷ · and q ÷ p
n
such that bs ÷q =1. Then b maps to
s
1
in ·
p
¡q
n
and so ba maps to
o
x
. More precisely: because s is invertible in ·
p
¡q
n
,
o
x
is
the unique element of this ring such that s
o
x
=a; since s(ba) =a(1÷q), the image of ba
in ·
p
also has this property and therefore equals
o
x
.
2
REMARK 3.11 With the notations of Proposition 1.11, we have shown in the above proof
that a
ec
=a if a is a power of a maximal ideal p and S =S p.
We now prove that a nonzero ideal a of · can be factored into a product of prime ideals.
According to 3.8 applied to ·, the ideal a contains a product of nonzero prime ideals,
b =p
i
1
1
p
i
m
n
.
We may suppose that the p
i
are distinct. Then
·¡b .·¡p
i
1
1
× ×·¡p
i
m
n
.·
p
1
¡q
i
1
1
× ×·
p
m
¡q
i
m
n
where q
i
= p
i
·
p
i
is the maximal ideal of ·
p
i
. The ﬁrst isomorphism is given by the
Chinese Remainder Theorem (and 3.9), and the second is given by (3.10). Under this
isomorphism, a¡b corresponds to q
x
1
1
¡q
i
1
1
× ×q
x
m
n
¡q
i
m
n
for some s
i
_ r
i
(recall that the
rings ·
p
i
are all discrete valuation rings). Since this ideal is also the image of p
x
1
1
p
x
m
n
under the isomorphism, we see that
a =p
x
1
1
p
x
m
n
in ·¡b.
Both of these ideals contain b, and so this implies that
a =p
x
1
1
p
x
m
n
in ·(because there is a onetoone correspondence between the ideals of ·¡b and the ideals
of · containing b).
To complete the proof of Theorem 3.7, we have to prove that the above factorization is
unique. Suppose that we have two factorizations of the ideal a. After adding factors with
zero exponent, we may suppose that the same primes occur in each factorization, so that
p
x
1
1
p
x
m
n
=a =p
t
1
1
p
t
m
n
say. In the course of the above proof, we showed that
q
x
i
i
=a·
p
i
=q
t
i
i
where q
i
the maximal ideal in ·
p
i
. Therefore s
i
=t
i
for all i .
44
Unique factorization of ideals
REMARK 3.12 Note that
s
i
>0 ” a·
p
i
=·
p
i
” a cp
i
.
COROLLARY 3.13 Let a and b be ideals in ·; then
a cb ” a·
p
cb·
p
for all ideals nonzero prime ideals p of ·. In particular, a =b if and only if a·
p
=b·
p
for
all p.
PROOF. The necessity is obvious. For the sufﬁciency, factor a and b
a =p
i
1
1
p
i
m
n
. b =p
x
1
1
p
x
m
n
. r
i
. s
i
_0.
Then
a·
p
i
cb·
p
i
” r
i
_s
i
.
(recall that ·
p
i
is a discrete valuation ring) and r
i
_s
i
all i implies a cb.
2
COROLLARY 3.14 Let · be an integral domain with only ﬁnitely many prime ideals; then
· is a Dedekind domain if and only if it is a principal ideal domain.
PROOF. Assume ·is a Dedekind domain. After (3.7), to showthat ·is principal, it sufﬁces
to show that the prime ideals are principal. Let p
1
. . . . . p
n
be these ideals. Choose an
element .
1
÷ p
1
÷p
2
1
. According to the Chinese Remainder Theorem (1.14), there is an
element . ÷ · such that
. ÷.
1
mod p
2
1
. . ÷1 mod p
i
. i =1.
Now the ideals p
1
and (.) generate the same ideals in ·
p
i
for all i , and so they are equal in
· (by 3.13).
2
COROLLARY 3.15 Let a ¯ b = 0 be two ideals in a Dedekind domain; then a = b ÷(a)
for some a ÷ ·.
PROOF. Let b =p
i
1
1
p
i
m
n
and a =p
x
1
1
p
x
m
n
with r
i
. s
}
_ 0. Because b c a, s
i
_ r
i
for
all i . For 1 _ i _ m, choose an .
i
÷ · such that .
i
÷ p
x
i
i
, .
i
… p
x
i
÷1
i
. By the Chinese
Remainder Theorem, there is an a ÷ · such that
a ÷.
i
mod p
i
i
i
, for all i.
Now one sees that b÷(a) =a by looking at the ideals they generate in ·
p
for all p.
2
COROLLARY 3.16 Let a be an ideal in a Dedekind domain, and let a be any nonzero
element of a; then there exists a b ÷ a such that a =(a. b).
PROOF. Apply (3.15) to a ¯(a).
2
45
3. DEDEKIND DOMAINS; FACTORIZATION
COROLLARY 3.17 Let a be a nonzero ideal in a Dedekind domain; then there exists a
nonzero ideal a
+
in ·such that aa
+
is principal. Moreover, a
+
can be chosen to be relatively
prime to any particular ideal c, and it can be chosen so that aa
+
=(a) with a any particular
element of a (but not both).
PROOF. Let a ÷ a, a =0; then a ¯(a), and so we have
(a) =p
i
1
1
p
i
m
n
and a =p
x
1
1
p
x
m
n
. s
i
_r
i
.
If a
+
=p
i
1
x
1
1
p
i
m
x
m
n
, then aa
+
=(a).
We now show that a
+
can be chosen to be prime to c. We have a ¯ac, and so (by 3.15)
there exists an a ÷ a such that a =ac ÷(a). As a ¯(a), we have (a) =a a
+
for some ideal
a
+
(by the above argument); now, ac ÷aa
+
=a, and so c ÷a
+
=·. (Otherwise c ÷a
+
cp
some prime ideal, and ac ÷aa
+
=a(c ÷a
+
) cap =a.)
2
In basic graduate algebra courses, it is shown that
· a principal ideal domain =· is a unique factorization domain.
The converse is false because, for example, kŒX. Y  is a unique factorization domain in
which the ideal (X. Y ) is not principal, but it is true for Dedekind domains.
PROPOSITION 3.18 ADedekind domain that is a unique factorization domain is a principal
ideal domain.
PROOF. In a unique factorization domain, an irreducible element ¬ can divide a product
bc only if ¬ divides b or c (write bc =¬q and express each of b, c, and q as a product of
irreducible elements). This means that (¬) is a prime ideal.
Now let ·be a Dedekind domain with unique factorization. It sufﬁces to show that each
nonzero prime ideal p of · is principal. Let a be a nonzero element of p. Then a factors
into a product of irreducible elements (see 1.4) and, because p is prime, it will contain one
of these irreducible factors ¬. Now p ¯ (¬) ¯ (0), and, because (¬) is a nonzero prime
ideal, it is maximal, and so equals p.
2
The ideal class group
Let · be a Dedekind domain. A fractional ideal of · is a nonzero ·submodule a of 1
such that
Ja
def
={Ja [ a ÷ a]
is contained in · for some nonzero J ÷ · (or 1), i.e., it is a nonzero ·submodule of 1
whose elements have a common denominator. Note that a fractional ideal is not an ideal
unless it is contained in · — when necessary to avoid confusion, we refer to the ideals in
· as integral ideals.
A fractional ideal a is a ﬁnitely generated ·module, because Ja is an integral ideal,
hence ﬁnitely generated, for some J =0, and the map . ÷J.: a ÷Ja is an isomorphism
of ·modules. Conversely, a nonzero ﬁnitely generated ·submodule of 1 is a fractional
ideal, because a common denominator for the generators will be a common denominator
for all the elements of the module.
46
The ideal class group
Every nonzero element b of 1 deﬁnes a fractional ideal
(b)
def
=b·
def
={ba [ a ÷ ·].
A fractional ideal of this type is said to be principal.
The product of two fractional ideals is deﬁned in the same way as for (integral) ideals
a b ={
¸
a
i
b
i
[ a
i
÷ a. b
i
÷ b].
This is again a fractional ideal: it is obviously an ·module, and if Ja c · and eb c ·,
then Jeab c·. For principal fractional ideals, (a)(b) =(ab).
EXAMPLE 3.19 Let · be a discrete valuation ring with maximal ideal p and ﬁeld of frac
tions 1. Write ¬ for a generator of p. Every nonzero element of 1 can be written uniquely
in the form a =u¬
n
with u a unit in · and m÷ Z. Let a be a fractional ideal of ·. Then
Ja c· for some J ÷ ·, and we can suppose J =¬
n
. Thus ¬
n
a is an ideal in ·, and so it
is of the form (¬
n
) for some m_ 0. Clearly, a =(¬
nn
). Thus the fractional ideals of ·
are of the form (¬
n
), m÷ Z. They form a free abelian group Id(·) of rank 1, and the map
m÷(¬
n
): Z ÷Id(·)
is an isomorphism.
THEOREM 3.20 Let ·be a Dedekind domain. The set Id(·) of fractional ideals is a group;
in fact, it is the free abelian group on the set of prime ideals.
PROOF. We have noted that the law of composition is welldeﬁned. It is obviously com
mutative. For associativity, one checks that
(ab)c =
¸
a
i
b
i
c
i
[ a
i
÷ a. b
i
÷ b. c
i
÷ c
¸
=a(bc).
The ring · plays the role of an identity element: a· =a. In order to show that Id(·) is a
group, it remains to show that inverses exist.
Let a be a nonzero integral ideal. According to (3.17), there is an ideal a
+
and an a ÷ ·
such that aa
+
= (a). Clearly a (a
1
a
+
) = ·, and so a
1
a
+
is an inverse of a. If a is a
fractional ideal, then Ja is an integral ideal for some J, and J (Ja)
1
will be an inverse
for a.
It remains to show that the group Id(·) is freely generated by the prime ideals, i.e.,
that each fractional ideal can be expressed in a unique way as a product of powers of prime
ideals. Let a be a fractional ideal. Then Ja is an integral ideal for some J ÷ ·, and we can
write
Ja =p
i
1
1
p
i
m
n
. (J) =p
x
1
1
p
x
m
n
.
Thus a =p
i
1
x
1
1
p
i
m
x
m
n
. The uniqueness follows from the uniqueness of the factoriza
tion for integral ideals.
2
REMARK 3.21 (a) Conversely, E. Noether showed that an integral domain whose frac
tional ideals form a group under ideal multiplication is a Dedekind domain (see Cohn 1991,
Theorem 4.6).
47
3. DEDEKIND DOMAINS; FACTORIZATION
(b) Let S be a multiplicative subset in a Dedekind domain ·, and let ·
S
=S
1
·. It is
an integral domain with the same ﬁeld of fractions as ·:
· c·
S
c1.
For any fractional ideal a of ·, S
1
a
def
={
o
x
[ a ÷ a, s ÷ S] is a fractional ideal of ·
S
. It is
the ·
S
module generated by a. The following hold for any fractional ideals a and b,
S
1
(ab) =(S
1
a)(S
1
b). S
1
a
1
=(a·
S
)
1
.
For any fractional ideal a, deﬁne
a
t
={a ÷ 1 [ aa c·].
This is an ·module, and if J ÷ a, J = 0, then Ja
t
c ·, and so a
t
is a fractional ideal.
From the deﬁnition of a
t
, we see that aa
t
is an ideal in ·. If it is not equal to ·, then it
is contained in some prime ideal p. When we pass to ·
p
, the inclusion aa
t
c p becomes
bb
t
cq, where b, b
t
, and q are the ideals in ·
p
generated by a, a
t
, and p. Moreover,
b
t
={a ÷ 1 [ ab c·
p
].
But q = (¬), and b = (¬
n
) = ¬
n
·
p
for some m ÷ Z. Clearly b
t
= ¬
n
·
p
, and so
bb
t
=·
p
— we have a contradiction.
We deﬁne the ideal class group Cl(·) of · to be the quotient Cl(·) =Id(·)¡P(·) of
Id(·) by the subgroup of principal ideals. The class number of · is the order of Cl(·)
(when ﬁnite). In the case that · is the ring of integers O
1
in a number ﬁeld 1, we often
refer to Cl(O
1
) as the ideal class group of 1, and its order as the class number of 1.
One of the main theorems of these notes will be that the class number h
1
of a number
ﬁeld 1 is ﬁnite. Understanding how the class numbers of number ﬁelds vary remains an
interesting problem. For example, the class number of QŒ
÷m for m positive and square
free is 1 if and only if m = 1. 2. 3. 7. 11. 19. 43. 67. 163. It not difﬁcult to show that these
ﬁelds have class number 1, but it was not until 1954 that it was shown (by Heegner) that
there were no more (and for more than 15 years, no one believed Heegner’s proof to be
correct). We have seen that ZŒ
÷5 is not a principal ideal domain, and so can’t have class
number 1— in fact it has class number 2. The method we use to prove that the class number
is ﬁnite is effective: it provides an algorithm for computing it. There are expected to be an
inﬁnite number of real quadratic ﬁelds with class number one, but this has not been proved.
Using the equivalent language of binary quadratic forms (see Chapter 4), Gauss showed
that the class group of a quadratic ﬁeld QŒ
J can have arbitrarily many cyclic factors of
even order.
It is known that every abelian group can be realized as the class group of a Dedekind
domain (not necessarily the ring of integers in a number ﬁeld).
1
EXAMPLE 3.22 Consider the afﬁne elliptic curve
Y
2
=X
3
÷aX ÷b. z=÷4a
3
÷27b
2
=0.
1
Claborn, Luther. Every abelian group is a class group. Paciﬁc J. Math. 18 1966 219–222.
48
Discrete valuations
The associated ring · = CŒX. Y ¡(Y
2
÷X
3
÷aX ÷b) of regular functions on · is a
Dedekind domain, and its class group is uncountable. In fact, it is isomorphic in a nat
ural way to C¡¹ for some lattice ¹ in C.
2
PROPOSITION 3.23 Let · be a Dedekind domain, and let S be a multiplicative set in ·.
Then a ÷S
1
a deﬁnes an isomorphism from the subgroup of Id(·) generated by prime
ideals not meeting S to the group Id(S
1
·).
PROOF. Immediate consequence of 1.12 and 3.20.
2
REMARK 3.24 Let · be a Dedekind domain with ﬁnite ideal class group. There is then a
ﬁnite set of ideals a
1
. .... a
n
which is a set of representatives for the ideal classes. Clearly we
may take the a
i
to be integral. Let b be any element in
¸
a
i
, and let S be the multiplicative
set generated by b, S ={1. b. b
2
. . . .]. I claim that S
1
· is a principal ideal domain.
By assumption, any ideal a c · can be written a = (a) a
i
for some a ÷ 1
×
and i ,
1 _ i _ m. Because the map b ÷S
1
b is a homomorphism we have S
1
a =(a) S
1
a
i
where (a) now denotes the ideal generated by a in S
1
·. Since S
1
a
i
contains a unit, it is
the whole ring. Thus S
1
a =(a), and we see that every ideal in S
1
· of the form S
1
a
is principal. According to (1.11), all ideals of S
1
· are of this form.
REMARK 3.25 The following conditions on an integral domain · are equivalent:
(a) · is a Dedekind domain;
(b) for every prime ideal p of ·, ·
p
is a discrete valuation ring;
(c) the fractional ideals of · form a group;
(d) for every fractional ideal a of ·, there is an ideal b such that ab =·.
We have seen that (a) implies (b) , (c), and (d), and the same arguments show that
(b) implies (c) and (d). The conditions (c) and (d) are obviously equivalent, and we have
already noted in (3.21) that (c) implies (a).
Discrete valuations
Let 1 be a ﬁeld. A discrete valuation on 1 is a nonzero homomorphism ·: 1
×
÷Z such
that ·(a÷b) _min(·(a). ·(b)). As · is not the zero homomorphism, its image is a nonzero
subgroup of Z, and is therefore of the form mZ for some m÷ Z. If m=1, then ·: 1
×
÷Z
is surjective, and · is said to be normalized; otherwise, . ÷m
1
·(.) will be a normalized
discrete valuation.
Note that, for a discrete valuation ord,
ord(a
1
÷ ÷a
n
) _min(ord(a
1
). ord(a
2
÷ ÷a
n
)) _ _ min
1_i_n
(ord(a
i
)). (8)
2
Let 1 be the associated complete curve, and let Div
0
(1) be the group of divisors of degree zero on 1.
There is an obvious isomorphism Div
0
(1) . Id(·) under which principal divisors correspond to principal
ideals, and so
Cl(·) .Pic
0
(1) .1(C) .C¡¹
(Milne 2006, I 4.10, III 3.10).
49
3. DEDEKIND DOMAINS; FACTORIZATION
EXAMPLE 3.26 (a) Let M be the ﬁeld of meromorphic functions on a connected open
subset U of the complex plane (or, better, a compact Riemann surface), and let } ÷ M
×
.
For each 1 ÷ U, deﬁne ord
1
(} ) to be ÷m, m, or 0 according as } has a pole of order m
at 1, a zero of order m at 1, or neither a pole nor a zero at 1. Then ord
1
is a normalized
discrete valuation on M.
(b) Let · be a principal ideal domain with ﬁeld of fractions 1, and let ¬ be a prime
element of ·. Then each element c of 1
×
can be expressed uniquely in the form c =¬
no
b
with m÷ Z and a and b elements of · relatively prime to ¬. Deﬁne ·(c) =m. Then · is a
normalized discrete valuation on 1.
(c) Let · be a Dedekind domain and let p be a prime ideal in ·. For any c ÷ 1
×
, let
p
¡(c)
be the power of p in the factorization of (c). Then · is a normalized discrete valuation
on 1.
In all these examples, we have that ·(a ÷b) = ·(b) if ·(a) > ·(b). This is in fact a
general property of discrete valuations. First note that ·(¯) =0 for any element of 1
×
of
ﬁnite order because · is a homomorphism and Z has no elements of ﬁnite order); hence
·(÷a) =·(÷1) ÷·(a) =·(a). Therefore, if ·(a) >·(b), we have
·(b) =·(a÷b ÷a)) _min(·(a÷b). ·(a)) _min(·(a). ·(b)) =·(b).
and so equality must hold throughout, and this implies ·(a÷b) =·(b).
We often use “ord” rather than “·” to denote a discrete valuation; for example, we often
use ord
p
to denote the normalized discrete valuation deﬁned by p in (c).
Example (b) shows that every discrete valuation ring gives rise to a discrete valuation
on its ﬁeld of fractions. There is a converse to this statement.
PROPOSITION 3.27 Let · be a discrete valuation on 1, then
·
def
={a ÷ 1 [ ·(a) _0]
is a principal ideal domain with maximal ideal
m
def
={a ÷ 1 [ ·(a) >0].
If ·(1
×
) =mZ, then the ideal m is generated by any element ¬ such that ·(¬) =m.
PROOF. Routine.
2
Later we shall see that a discrete valuation ord deﬁnes a topology on 1 for which two
elements . and . are close if ord(. ÷.) is large. The Chinese Remainder Theorem can be
restated as an approximation theorem.
PROPOSITION 3.28 Let .
1
. .... .
n
be elements of a Dedekind domain ·, and let p
1
. .... p
n
be distinct prime ideals of ·. For any integer n, there is an . ÷ · such that
ord
p
i
(. ÷.
i
) >n. i =1. 2. .... m.
PROOF. From (3.9) we know that the ideals p
n÷1
i
are relatively prime in pairs, and so (1.14)
provides us with an element . ÷ · such that
. ÷.
i
mod p
n÷1
i
. i =1. 2. . . . . m.
50
Integral closures of Dedekind domains
i.e., such that
ord
p
i
(. ÷.
i
) >n. i =1. 2. .... m.
2
Integral closures of Dedekind domains
We now prove a result that implies that rings of integers in number ﬁelds are Dedekind
domains, and hence that their ideals factor uniquely into products of prime ideals.
THEOREM 3.29 Let · be a Dedekind domain with ﬁeld of fractions 1, and let T be the
integral closure of · in a ﬁnite separable extension 1 of 1. Then T is a Dedekind domain.
PROOF. We have to check the three conditions in the deﬁnition of a Dedekind domain
(p3.3). We ﬁrst show that T is Noetherian. In (2.29) we showed that T is contained in
a ﬁnitely generated ·module. It follows that every ideal in T is ﬁnitely generated when
regarded as an ·module (being a submodule of a Noetherian ·module) and a fortiori as
an ideal (=Tmodule). Next, T is integrally closed because of (2.16). It remains to prove
that every nonzero prime ideal q of T is maximal. Let ˇ ÷ q, ˇ =0. Then ˇ is integral over
·, and so there is an equation
ˇ
n
÷a
1
ˇ
n1
÷ ÷a
n
=0. a
i
÷ ·.
which we may suppose to have the minimum possible degree. Then a
n
= 0. As a
n
÷
ˇT ¨·, we have that q ¨· = (0). But q ¨· is a prime ideal (obviously), and so it is a
maximal ideal p of ·, and ·¡p is a ﬁeld. We know T¡q is an integral domain, and the map
a÷p ÷a÷q
identiﬁes ·¡p with a subﬁeld of T¡q. ·s T is integral over ·, T¡q is algebraic over ·¡p.
The next lemma shows that T¡q is a ﬁeld, and hence that q is maximal.
2
LEMMA 3.30 Any integral domain T containing a ﬁeld k and algebraic over k is itself a
ﬁeld.
PROOF. Let ˇ be a nonzero element of T — we have to prove that it has an inverse in T.
Because ˇ is algebraic over k, the ring kŒˇ is ﬁnitedimensional as a kvector space, and
the map . ÷ˇ.: kŒˇ ÷kŒˇ is injective (because T is an integral domain). From linear
algebra we deduce that the map is surjective, and so there is an element ˇ
t
÷ kŒˇ such that
ˇˇ
t
=1.
2
In fact, Theorem 3.29 is true without the assumption that 1 be separable over 1 —
see Janusz 1996, I 6.1 for a proof of the more general result. The added difﬁculty is that,
without the separability condition, T may fail to be ﬁnitely generated as an ·module, and
so the proof that it is Noetherian is more difﬁcult.
51
3. DEDEKIND DOMAINS; FACTORIZATION
Modules over Dedekind domains (sketch).
The structure theorem for ﬁnitely generated modules over principal ideal domains has an
interesting extension to modules over Dedekind domains. Throughout this subsection, · is
a Dedekind domain.
First, note that a ﬁnitely generated torsionfree ·module M need not be free. For
example, every fractional ideal is ﬁnitely generated and torsionfree but it is free if and only
if it is principal. Thus the best we can hope for is the following.
THEOREM 3.31 Let · be a Dedekind domain.
(a) Every ﬁnitely generated torsionfree ·module M is isomorphic to a direct sum of
fractional ideals,
M ~a
1
˚ ˚a
n
.
(b) Two ﬁnitely generated torsionfree ·modules M ~a
1
˚ ˚a
n
and N ~b
1
˚ ˚
b
n
are isomorphic if and only if m=n and
¸
a
i
÷
¸
b
i
modulo principal ideals.
Hence,
M ~a
1
˚ ˚a
n
~·˚ ˚·˚a
1
a
n
.
Moreover, two fractional ideals a and b of · are isomorphic as ·modules if and only they
deﬁne the same element of the class group of ·.
The rank of a module M over an integral domain 1 is the dimension of 1 ˝
T
M
as a 1vector space, where 1 is the ﬁeld of fractions of 1. Clearly the rank of M ~
a
1
˚ ˚a
n
is m.
These remarks show that the set of isomorphism classes of ﬁnitely generated torsion
free ·modules of rank 1 can be identiﬁed with the class group of ·. Multiplication
of elements in Cl(·) corresponds to the formation of tensor product of modules. The
Grothendieck group of the category of ﬁnitely generated ·modules is Cl(·) ˚Z.
THEOREM 3.32 (INVARIANT FACTOR THEOREM) Let M ¯N be ﬁnitely generated torsion
free ·modules of the same rank m. Then there exist elements e
1
. .... e
n
of M, fractional
ideals a
1
. .... a
n
, and integral ideals b
1
¯b
2
¯... ¯b
n
such that
M =a
1
e
1
˚ ˚a
n
e
n
. N =a
1
b
1
e
1
˚ ˚a
n
b
n
e
n
.
The ideals b
1
, b
2
, ..., b
n
are uniquely determined by the pair M ¯ N, and are called
the invariant factors of N in M.
The last theorem also yields a description of ﬁnitely generated torsion ·modules.
For proofs of the above results, see Curtis and Reiner 1962, III, 22, Fr¨ ohlich and Taylor
1991, II 4, or Narkiewicz 1990, I 3.
Factorization in extensions
Let · be a Dedekind domain with ﬁeld of fractions 1, and let T be the integral closure of
· in a ﬁnite separable extension 1 of 1.
A prime ideal p of · will factor in T,
pT =P
e
1
1
P
e
g
v
. e
i
_1.
52
Factorization in extensions
If any of the numbers is > 1, then we say that p is ramiﬁed in T (or 1). The number
e
i
is called the ramiﬁcation index. We say P divides p (written P[p) if P occurs in the
factorization of p in T. We then write e(P¡p) for the ramiﬁcation index and }(P¡p) for
the degree of the ﬁeld extension ŒT¡P: ·¡p (called the residue class degree). A prime p
is said to split (or split completely) in 1 if e
i
=}
i
=1 for all i , and it said to be inert in 1
if pT is a prime ideal (so g =1 =e).
For example, (2) =(1÷i )
2
in ZŒi , and so (2) ramiﬁes with ramiﬁcation index 2. On
the other hand, (3) is inert in QŒi  with residue ﬁeld ZŒi ¡(3) = F
9
, and (5) splits as the
product of two prime ideals (5) =(2÷i )(2÷i ).
LEMMA 3.33 A prime ideal P of T divides p if and only if p =P¨1.
PROOF. =: Clearly p c P¨1 and P¨1 = ·. As p is maximal, this implies that p =
P¨1.
=: If p cP, then pT cP, and we have seen (3.12) that this implies that P occurs in
the factorization of pT.
2
THEOREM 3.34 Let m be the degree of 1 over 1, and let P
1
. .... P
v
be the prime ideals
dividing p; then
v
¸
i=1
e
i
}
i
=m. (9)
If 1 is Galois over 1, then all the ramiﬁcation numbers are equal, and all the residue class
degrees are equal, and so
e}g =m. (10)
PROOF. To prove (9), we shall show that both sides equal ŒT¡pT: ·¡p.
For the equality
¸
v
i=1
e
i
}
i
=ŒT¡pT: ·¡p, note that T¡pT =T¡
¸
P
e
i
i
.
¸
T¡P
e
i
i
(Chinese Remainder Theorem), and so it sufﬁces to show that ŒT¡P
e
i
i
: ·¡p =e
i
}
i
. From
the deﬁnition of }
i
, we know that T¡P
i
is a ﬁeld of degree }
i
over ·¡p. For each r
i
,
P
i
i
i
¡P
i
i
÷1
i
is a T¡P
i
module, and because there is no ideal between P
i
i
i
and P
i
i
÷1
i
, it
must have dimension one as a T¡P
i
vector space, and hence dimension }
i
as an ·¡p
vector space. Therefore each quotient in the chain
T ¯P
i
¯P
2
i
¯ ¯P
e
i
i
has dimension }
i
over ·¡p, and so the dimension of T¡P
e
i
i
is e
i
}
i
.
The proof of the equality ŒT¡pT: ·¡p = m is easy when T is a free ·module, for
example, if · is a principal ideal domain, because an isomorphism ·
n
÷T of ·modules,
when tensored with 1, gives an isomorphism 1
n
÷1, which shows that n = m, and,
when tensored ·¡p, gives an isomorphism (·¡p)
n
÷T¡pT (see (3), p16), which shows
that n =ŒT¡pT: ·¡p.
Nowlet S be a multiplicative subset of ·disjoint fromp and such that S
1
·is principal
(e.g., S =·÷p). Write T
t
=S
1
T and ·
t
=S
1
·. Then pT
t
=
¸
(P
i
T
t
)
e
i
(see 3.23),
and so
¸
e
i
}
i
=ŒT
t
¡pT
t
: ·
t
¡p·
t
; but ·
t
is principal, and so ŒT
t
¡pT
t
: ·
t
¡p·
t
 =m. This
completes the proof (9).
53
3. DEDEKIND DOMAINS; FACTORIZATION
Now assume 1 is Galois over 1. An element o of Gal(1¡1) maps T isomorphically
onto itself. In particular, if Pis a prime ideal of T, then oPis also a prime ideal. Moreover,
if P divides p, then it follows from (3.33) that oP divides p. Clearly e(oP¡p) =e(P¡p)
and }(oP¡p) =}(P¡p), and so it remains to show that Gal(1¡1) acts transitively on the
prime ideals of T dividing p.
Suppose P and Q both divide p, and suppose Q is not conjugate to P, i.e., that for all
o ÷ Gal(1¡1), oP=Q. According to the Chinese Remainder Theorem, we can ﬁnd an
element ˇ lies in Q but not in any of the ideals oP. Consider b =Nm(ˇ)
def
=
¸
oˇ. Then
b ÷ ·, and as ˇ ÷ Q, it also lies in Q; hence b ÷ Q¨· = p. On the other hand, for all
o ÷ Gal(1¡1), ˇ … o
1
P, and so oˇ … P. The fact that
¸
oˇ ÷ p c P contradicts the
primality of P.
2
The primes that ramify
In this subsection, we obtain a description of the primes that ramify in an extension.
THEOREM 3.35 Let 1 be a ﬁnite extension of a number ﬁeld 1, let · be a Dedekind
domain in 1 with ﬁeld of fractions 1 (e.g., · = O
1
), and let T be the integral closure
of · in 1. Assume that T is a free ·module (this is true for example if · is principal
ideal domain). Then a prime p ramiﬁes in 1 if and only if p[ disc(T¡·). In particular, only
ﬁnitely many prime ideals ramify.
We obtain this as the consequence of a series of lemmas.
LEMMA 3.36 Let · be a ring and let T be a ring containing · and admitting a ﬁnite basis
{e
1
. .... e
n
] as an ·module. For any ideal a of ·, {¨ e
1
. .... ¨ e
n
] is a basis for the ·¡amodule
T¡aT, and
D( ¨ e
1
. .... ¨ e
n
) ÷D(e
1
. .... e
n
) mod a.
PROOF. As in the proof of (3.34), the isomorphism
(a
1
. . . . . a
n
) ÷
¸
a
i
e
i
: ·
n
÷T
gives, when tensored with ·¡a, an isomorphism
(a
1
. . . . . a
n
) ÷
¸
a
i
¨ e
i
: (·¡a)
n
÷T¡a
which shows that ¨ e
1
. .... ¨ e
n
is a basis for T¡aT. The second assertion is obvious from the
deﬁnitions.
2
LEMMA 3.37 Let · be a ring and let T
1
. .... T
v
be rings containing · and free of ﬁnite
rank as ·modules. Then
disc((
¸
T
i
)¡·) =
¸
disc(T
i
¡·).
PROOF. Choose bases c
i
for each of the T
i
(as ·modules), and compute the discriminant
of T¡· using the basis
i
c
i
.
2
54
The primes that ramify
An element ˛ of a ring is said to be nilpotent if ˛
n
=0 for some m > 1. A ring is said
to be reduced if it has no nonzero nilpotent elements.
LEMMA 3.38 Let k be a perfect ﬁeld, and let T be a kalgebra of ﬁnite dimension. Then
T is reduced if and only if disc(T¡k) =0.
PROOF. Let ˇ =0 be a nilpotent element of T, and choose a basis e
1
. . . . . e
n
for T with
e
1
=ˇ. Then ˇe
i
is nilpotent for all i , and so the klinear map
. ÷ˇe
i
.: T
//
T
is nilpotent. Its matrix is also nilpotent, but a nilpotent matrix has trace zero—its minimum
polynomial (and hence its characteristic polynomial) is of the form X
i
—and so the ﬁrst
row of the matrix (Tr(e
i
e
}
)) is zero. Therefore its determinant is zero.
Conversely, suppose T is reduced. We ﬁrst show that the intersection N of the prime
ideals of T is zero (this, in fact, is true for any reduced Noetherian ring). Let b ÷ T, b =0.
Let ˙ be the set of ideals of T containing no power of b. Because b is not nilpotent, ˙
contains the zero ideal, and hence is nonempty. Because T is Noetherian, ˙ has a maximal
element p. We shall show that p is prime. Since b … p, this will show that b … N.
Let .. . be elements of T not in p. Then p÷(.) and p÷(.) strictly contain p, and so
b
n
÷ p÷(.). b
n
÷ p÷(.)
for some m. n, say,
b
n
=]÷c.. b
n
=]
t
÷c
t
.. ]. ]
t
÷ p. c. c
t
÷ T.
Then b
n÷n
= ]]
t
÷]c
t
. ÷]
t
c. ÷cc
t
.. ÷ p ÷(..), and so p ÷(..) is not in ˙; in
particular, p÷(..) =p, and .. … p. Therefore p is prime ideal, which completes the proof
that N=0.
Let p be a prime ideal of T. Then T¡p is an integral domain, algebraic over k, and
hence is a ﬁeld (by 3.30). Therefore p is maximal. Let p
1
. p
2
. . . . . p
i
be prime ideals of
T. Since they are all maximal, they are relatively prime in pairs. Therefore the Chinese
remainder theorem shows that
T¡
¸
p
i
=
¸
T¡p
i
(*).
Note that
ŒT : k _ŒT¡
¸
p
i
: k =
¸
ŒT¡p
i
: k _r.
Therefore T has only ﬁnitely many prime ideals, say p
1
. . . . . p
v
where g _ ŒT: k, and
¸
p
i
=0. When we take r =g in (*) we ﬁnd that
T =
¸
v
i=1
T¡p
i
.
For each i , T¡p
i
is a ﬁeld, and it is a ﬁnite extension of k. Because k is perfect, it is even
a separable extension of k. Now we can apply (2.26) to deduce that disc((T¡p
i
)¡k) =0,
and we can apply the preceding lemma to deduce that disc(T¡k) =0.
2
55
3. DEDEKIND DOMAINS; FACTORIZATION
We now prove the theorem. From the ﬁrst lemma, we see that
disc(T¡·) mod p =disc((T¡pT)¡(·¡p)).
and from the last lemma that disc((T¡pT)¡(·¡p)) =0 if and only T¡pT is not reduced.
Let pT =
¸
P
e
i
i
. Then T¡pT .
¸
T¡P
e
i
, and
¸
T¡P
e
i
is reduced ” each T¡P
e
i
is reduced ” each e
i
=1.
REMARK 3.39 (a) In fact there is a precise, but complicated, relation between the power
of p dividing disc(T¡·) and the extent to which p ramiﬁes in T. It implies for example
that ord
p
(disc(T¡·)) _
¸
}
i
(e
i
÷1), and that equality holds if no e
i
is divisible by the
characteristic of ·¡p. See Serre 1962, III 6.
(b) Let · be the ring of integers in a number ﬁeld 1, and let T be the integral closure
of · in a ﬁnite extension 1 of 1. It is possible to deﬁne disc(T¡·) as an ideal without
assuming T to be a free ·module. Let p be an ideal in ·, and let S = ·÷p. Then
S
1
· =·
p
is principal, and so we can deﬁne disc(S
1
T¡S
1
·). It is a power (p·
p
)
n(p)
of p·
p
. Deﬁne
disc(T¡·) =
¸
p
n(p)
.
The index m(p) is nonzero for only ﬁnitely many p, and so this formula does deﬁne an ideal
in ·. Clearly this deﬁnition agrees with the usual one when T is a free ·module, and the
above proof shows that a prime ideal p ramiﬁes in T if and only if it divides disc(T¡·).
EXAMPLE 3.40 (For experts on Riemann surfaces.) Let X and Y be compact connected
Riemann surfaces, and let ˛: Y ÷X be a nonconstant holomorphic mapping. Write M(X)
and M(Y ) for the ﬁelds of meromorphic functions on X and Y . The map } ÷} ı˛ is an
inclusion M(X) ·÷M(Y ) which makes M(Y ) into a ﬁeld of ﬁnite degree over M(X);
let m be this degree. Geometrically, the map is m: 1 except at a ﬁnite number of branch
points.
Let 1 ÷ X and let O
1
be the set of meromorphic functions on X that are holomorphic
at 1 — it is the discrete valuation ring attached to the discrete valuation ord
1
, and its
maximal ideal is the set of meromorphic functions on X that are zero at 1. Let T be the
integral closure of O
1
in M(Y ). Let ˛
1
(1) = {O
1
. .... O
v
] and let e
i
be the number
of sheets of Y over X that coincide at O
i
. Then pT =
¸
q
e
i
i
where q
i
is the prime ideal
{} ÷ T [ }(O
i
) =0].
Finding factorizations
The following result often makes it very easy to factor an ideal in an extension ﬁeld. Again
· is a Dedekind domain with ﬁeld of fractions 1, and T is the integral closure of · in a
ﬁnite separable extension 1 of 1.
THEOREM 3.41 Suppose that T =·Œ˛, and let }(X) be the minimum polynomial of ˛
over 1. Let p be a prime ideal in ·. Choose monic polynomials g
1
(X). . . . . g
i
(X) in ·ŒX
that are distinct and irreducible modulo p, and such that }(X) ÷
¸
g
i
(X)
e
i
modulo p.
Then
pT =
¸
(p. g
i
(˛))
e
i
56
Examples of factorizations
is the factorization of pT into a product of powers of distinct prime ideals. Moreover, the
residue ﬁeld T¡(p. g
i
(˛)) .(·¡p)ŒX¡( ¨ g
i
), and so the residue class degree }
i
is equal to
the degree of g
i
.
PROOF. Our assumption is that the map X ÷˛ deﬁnes an isomorphism
·ŒX¡(}(X)) ÷T.
When we divide out by p (better, tensor with ·¡p), this becomes an isomorphism
kŒX¡(
¨
} (X)) ÷T¡pT. X ÷˛.
where k = ·¡p. The ring kŒX¡(
¨
} ) has maximal ideals ( ¨ g
1
). .... ( ¨ g
i
), and
¸
( ¨ g
i
)
e
i
= 0
(but no product with smaller exponents is zero). The ideal ( ¨ g
i
) in kŒX¡(
¨
} ) corresponds
to the ideal (g
i
(˛)) ÷pT in T¡pT, and this corresponds to the ideal P
i
def
= (p. g
i
(˛)) in
T. Thus P
1
. .... P
i
is the complete set of prime ideals containing pT, and hence is the
complete set of prime divisors of p (see 3.12). When we write pT =
¸
P
e
i
i
, then the e
i
are characterized by the fact that pT contains
¸
P
e
i
i
but it does not contain the product
when any e
i
is replaced with a smaller value. Thus it follows from the above (parenthetical)
statement that e
i
is the exponent of ¨ g
i
occurring in the factorization of
¨
} .
2
REMARK 3.42 When it applies the last theorem can be used to prove (3.34) and (3.35). For
example, m =deg(} ), and so the equation m =
¸
e
i
}
i
is simply the equation deg(} ) =
¸
e
i
deg(g
i
). Also, disc(T¡·) = disc(}(X)), and this is divisible by p if and only if
¨
} (X) has multiple factors (when regarded as an element of (·¡p)ŒX), i.e., if and only if
some e
i
>0.
REMARK 3.43 The conclusion of the theorem holds for a particular prime p of · under
the following weaker hypothesis: disc(1. ˛. .... ˛
n1
) =a disc(T¡·) with a an ideal of ·
not divisible by p. To prove this, invert any element of a not in p, and apply the theorem to
the new ring and its integral closure.
Examples of factorizations
We use Theorem 3.41 to obtain some factorizations.
EXAMPLE 3.44 Let m = 1 be a squarefree integer. We consider the factorization of
prime integers in 1 = QŒ
m. Recall that disc(1.
m) = 4m, and that disc(O
1
¡Z) =
disc(1.
m) if m÷2. 3 mod 4, and that disc(O
1
¡Z) =disc(1.
m)¡4 if m÷1 mod 4. In
both cases, we can use the set {1.
m] to compute the factorization of an odd prime ] (see
3.43). Note that (3.34) allows only three possible factorizations of (]) in O
1
, namely,
(]) =p
2
: (]) ramiﬁes, e =2, } =1, g =1:
(]) =p: (]) stays prime, e =1, } =2, g =1:
(]) =p
1
p
2
: (]) splits, e =1, } =1, g =2.
One obtains the following result.
(i) If ][ disc(O
1
¡Z), then (]) ramiﬁes in O
1
.
(ii) For an odd prime ] not dividing the m, we have
(]) is the product of two distinct ideals ” m is a square mod ], i.e., (
n
;
) =1:
57
3. DEDEKIND DOMAINS; FACTORIZATION
(]) is a prime ideal in QŒ
m ” m is not a square mod ], i.e., (
n
;
) =÷1.
(iii) For the prime 2 when m÷1 mod 4, we have
(]) is the product of two distinct ideals ” m÷1 mod 8;
(]) is a prime ideal in QŒ
m ” m÷5 mod 8.
To prove (iii), we must use the integral basis {1. ˛], ˛ = (1 ÷
m)¡2. The minimum
polynomial of ˛ is X
2
÷X÷(1÷m)¡4. If m÷1 mod 8, this factors as X
2
÷X =X(X÷1)
mod 2, and so (2) =(2. ˛)(2. 1÷˛). If m÷5 mod 8, then X
2
÷X ÷(1÷m)¡4 ÷X
2
÷
X ÷1 mod 2, which is irreducible, and so (2) =(2. 1÷˛ ÷˛
2
) =(2).
EXAMPLE 3.45 It is proved in basic graduate algebra courses that ZŒi , the Gaussian inte
gers, is a principal ideal domain. I claim that the following conditions on an odd prime ]
are equivalent:
(a) ] ÷1 mod 4;
(b) (]) splits in ZŒi ;
(c) there exist integers a and b such that ] =a
2
÷b
2
.
We know that (]) splits in ZŒi  if and only if X
2
÷1 splits modulo ], but this is so if
and only if F
;
contains a 4th root of 1, i.e., if and only if the group F
×
;
contains an element
of order 4. As F
×
;
is a cyclic group (FT Exercise 13) of order ]÷1, this is so if and only if
4[]÷1. Thus we have shown that (a) and (b) are equivalent.
Suppose (]) splits in ZŒi , say (]) =p
1
p
2
. Then p
1
and p
2
are principal, and if p
1
=
(a ÷i b) then p
2
= (a ÷i b). Therefore a
2
÷b
2
= ] up to multiplication by a unit in
ZŒi . But the only units in ZŒi  are ˙1, ˙i , and so obviously a
2
÷b
2
=]. Conversely, if
] =a
2
÷b
2
with a. b ÷ Z, then (]) =(a÷i b)(a÷i b) in ZŒi .
ASIDE 3.46 The fact that every prime of the form 4n÷1 is a sum of two squares was stated as a
theorem by Fermat in a letter in 1654. Euler, who was almost certainly unaware of Fermat’s letter,
found a proof. For some history, and a discussion of algorithms for ﬁnding a and b, see Edwards
1977, p. 55.
REMARK 3.47 (a) From(3.41) and (3.43) we see that, for almost all ], factoring (]) in O
1
amounts to factoring a polynomial }(X) modulo ] into a product of powers of irreducible
polynomials. Clearly, this can always be done, but it may require a lot of hard work but not
much intelligence. Hence it can safely be left to the computer. In PARI, factormod(f,p)
factors the polynomial } modulo ]. For example,
factormod(X^3+10*X+1,2) returns (X ÷1)(X
2
÷X ÷1).
factormod(X^3+10*X+1,17) returns X
3
÷10X ÷1.
factormod(X^3+10*X+1,4027) returns (X÷2215)
2
(X÷3624), etc., as in the following
table.
(b) In the next section, we shall show, not only that the class group of a number ﬁeld
is ﬁnite, but that it is generated by the prime ideals dividing a certain small set of prime
numbers. Finding the class number therefore involves ﬁnding the prime ideal factors of
these prime numbers, and the relations among them.
EXAMPLE 3.48 Let ˛ be a root of X
3
÷10X ÷1. Recall that the discriminant of the
polynomial is ÷4027, and so the ring of integers in QŒ˛ is Z÷Z˛ ÷Z˛
2
. There are the
following factorizations:
58
Examples of factorizations
2 (1÷X)(1÷X ÷X
2
) (2) = (2. 1÷˛)(2. 1÷˛ ÷˛
2
)
3 (2÷X)(2÷X ÷X
2
) (3) = (3. 2÷˛)(3. 2÷˛ ÷˛
2
)
5 (1÷X)(1÷4X ÷X
2
) (5) = (5. 1÷˛)(5. 1÷4˛ ÷˛
2
)
7 (3÷X)(5÷4X ÷X
2
) (7) = (7. 3÷˛)(7. 5÷4˛ ÷˛
2
)
11 (6÷X)(2÷5X ÷X
2
) (11) = (11. 6÷˛)(11. 2÷5˛ ÷˛
2
)
13 1÷10X ÷X
3
(13) = (13. 1÷10˛ ÷˛
3
) =(13)
17 1÷10X ÷X
3
(17) = prime ideal.
4027 (2215÷X)
2
(3624÷X) (4027) = (4027. 2215÷˛)
2
(4027. 3624÷˛).
EXAMPLE 3.49 Let ˛ be a root of X
3
÷8X ÷15. Here again, the discriminant of the
polynomial is ÷4027, and so the ring of integers in QŒ˛ is Z÷Z˛ ÷Z˛
2
. There are the
following factorizations:
2 (1÷X)(1÷X ÷X
2
) (2) = (2. 1÷˛)(2. 1÷˛ ÷˛
2
)
3 X(1÷X
2
) (3) = (3. ˛)(3. 1÷˛
2
)
5 X(2÷X
2
) (5) = (5. ˛)(5. 2÷˛
2
)
7 (5÷X)(3÷2X ÷X
2
) (7) = (7. ˛)(7. 3÷2˛ ÷˛
2
)
11 (1÷X)(4÷10X ÷X
2
) (11) = (11. ˛)(11. 4÷10˛ ÷˛
2
)
13 2÷5X ÷X
3
(13) = (13)
17 (4÷X)(6÷X)(7÷X) (17) = (17. 4÷˛)(17. 6÷˛)(17. 7÷˛)
4027 (509÷X)(1759÷X)
2
. (4027) = (4027. 509÷˛)(4027. 1759÷˛)
2
On comparing the factorizations of (17) in the ﬁelds in the last two examples, we see that
the ﬁelds are not isomorphic.
REMARK 3.50 When 1 is a number ﬁeld, it is interesting to have a description of the set
Spl(1) of prime numbers that split in 1. For 1 =QŒ
m with m square free, this is the
set of odd ] not dividing m for which (
n
;
) = 1 together possibly with 2 (see 3.44). We
shall see later that the quadratic reciprocity law gives a good description of the set. For
any abelian Galois extension 1 of Q, class ﬁeld theory gives a similarly good description,
but for an arbitrary extension very little is known about what sets can occur. There is a
theorem that says that two Galois extensions 1 and 1
t
of Q are isomorphic if and only if
Spl(1) =Spl(1
t
). Moreover, this can be made into an effective procedure for determining
when ﬁelds are isomorphic. See Theorem 8.38 below.
EXAMPLE 3.51 In (2.39), we saw that }(X) =X
5
÷X÷1 is irreducible in QŒX, and that
its discriminant is 19 151, which is squarefree, and so, if ˛ is a root of }(X), then ZŒ˛ is
the ring of integers in QŒ˛. We have the following factorizations:
19
} ÷(6÷X)
2
(10÷13X ÷17X
2
÷X
3
)
(19) =(19. 6÷˛)
2
(19. 10÷13˛ ÷17˛
2
÷˛
3
)
151
} ÷(9÷X)(39÷X)
2
(61÷64X ÷X
2
)
(151) =(151. 9÷˛)(151. 39÷˛)
2
(151. 61÷64˛ ÷˛
2
)
4027
} ÷(1261÷X)(2592÷X)(790÷3499X ÷174X
2
÷X
3
)
(4027) =(4027. 1261÷˛)(4027. 2592÷˛)(4027. 790÷3499˛ ÷174˛
2
÷˛
3
.
Thus (19) and (151) are ramiﬁed in QŒ˛, and 4027 is not, which is what Theorem 3.35
predicts.
59
3. DEDEKIND DOMAINS; FACTORIZATION
EXAMPLE 3.52 According to PARI,
X
4
÷X
3
÷X
2
÷X ÷1 ÷(X ÷4)
4
mod 5
Why is this obvious?
Eisenstein extensions
Recall that Eisenstein’s Criterion says that a polynomial
X
n
÷a
1
X
n1
÷ ÷a
n
.
such that a
i
÷ Z, ][a
i
all i , and ]
2
does not divide a
n
, is irreducible in QŒX. We will
improve this result, but ﬁrst we need to make two observations about discrete valuations.
Let · be a Dedekind domain, and let T be its integral closure in a ﬁnite extension 1 of
its ﬁeld of fractions 1. Let p be a prime ideal of · and let P be an ideal of T dividing p,
say pT =P
e
. Write ord
p
and ord
P
for the normalized valuations on 1 and 1 deﬁned
by p and P. Then
ord
P
[1 =e ord
p
(11)
because, if (a) =p
n
in ·, then (a) =P
ne
in T.
Next I claim that if
a
1
÷ ÷a
n
=0.
then the minimum value of ord(a
i
) must be attained for at least two i s. Suppose not, say
ord(a
1
) < ord(a
i
) for all i >1. Then ÷a
1
=
¸
i_2
a
i
implies that
ord(a
1
) =ord(
¸
i_2
a
i
)
(8)
_ min
2_i_n
ord(a
i
).
which is a contradiction.
Let · be a Dedekind domain and let p be a prime ideal in ·. A polynomial
X
n
÷a
1
X
n1
÷ ÷a
n
. a
i
÷ ·.
is said to be Eisenstein relative to p if
ord
p
(a
1
) >0, . . . , ord
p
(a
n1
) >0, ord
p
(a
n
) =1.
PROPOSITION 3.53 Let }(X) ÷ ·ŒX be an Eisenstein polynomial with respect to p. Then
}(X) is irreducible, and if ˛ is a root of }(X), then p is totally ramiﬁed in 1Œ˛; in fact
pT =P
n
with P=(p. ˛) and m=deg(} ).
PROOF. Let 1 be the ﬁeld generated by a root ˛ of }(X); then Œ1: 1 _m
def
=deg(} ). Let
P be a prime ideal dividing p, with ramiﬁcation index e say. Consider the equation
˛
n
÷a
1
˛
n1
÷ ÷a
n
=0.
Because }(X) is Eisenstein,
ord
P
(˛
n
) =m ord
P
(˛):
ord
P
(a
i
˛
ni
) _(m÷i ) ord
P
(˛) ÷e:
ord
P
(a
n
) =e.
60
Exercises
If ord
P
(˛) =0, then the minimum value of ord
P
is taken for a single term, namely ˛
n
. This
is impossible, and so ord
P
(˛) _ 1, and ord
P
(a
i
˛
ni
) > ord
P
(a
n
) = e for i = 1. .... m.
From the remark preceding the proposition, we see that m ord
P
(˛) =e. Then
m ord
P
(˛) =e _Œ1Œ˛ : 1 _m.
and we must have equalities throughout: ord
P
(˛) =1, Œ1(˛): 1 =m=e.
2
NOTES Gauss proved the quadratic reciprocity law, and studied the arithmetic of QŒi  in order
to discover the quartic reciprocity law. Kummer made an intense study of the arithmetic of the
ﬁelds QŒ¯
n
, where ¯
n
is a primitive nth root of 1, in order to prove higher reciprocity laws. A
major problem for him was that unique factorization fails already for n = 23. To restore unique
factorization, he developed his theory of “ideal numbers”. One of Dedekind’s great achievements
was to realize that, by replacing Kummer’s “ideal numbers” with his new notion of “ideals”, it was
possible to simplify Kummer’s theory and extend it to the rings of integers in all number ﬁelds. A
difﬁcult step for him was showing that if a[b, then there exists an ideal c such that a =bc. Emmy
Noether reexamined Kummer’s work more abstractly, and named the integral domains for which
his methods applied “Dedekind domains”.
Exercises
31 Let k be a ﬁeld. Is kŒX. Y  a Dedekind domain? (Explain).
32 Show that ZŒ
3 is the ring of integers in QŒ
3 and ZŒ
7 is the ring of integers
in QŒ
7, but that ZŒ
3.
7 is not the ring of integers in QŒ
3.
7. (Hint: look at
(
3÷
7)¡2.)
33 Complete the proofs of the following statements (cf. 3.45):
(a) .
2
÷.
2
=] has a solution in Z ” ] ÷1 mod 4;
(b) .
2
÷2.
2
=] has a solution in Z ” ] ÷1 or 3 mod 8;
(c) .
2
÷3.
2
=] has a solution in Z ” ] ÷1 mod 3.
3
You may assume that QŒ
÷] has class number 1 for ] <5.
34 Let k be a ﬁeld, and let · be the subring kŒX
2
. X
3
 of kŒX.
(a) Show that kŒX is a ﬁnitely generated kŒX
2
module, and hence is a Noetherian
kŒX
2
module. Deduce that · is Noetherian.
(b) Show that every nonzero prime ideal of · is maximal, but that · is not a Dedekind
domain.
Hence · satisﬁes conditions (a) and (c) to be a Dedekind domain, but not (b). There are
also rings that satisfy (b) and (c) but fail (a), and rings that satisfy (a) and (b) but not (c) (for
example, kŒX. Y ).
3
Kwangho Choiy notes that .
2
÷3.
2
=] can be replaced by .
2
÷.. ÷.
2
=], because the norm is of
the form .
2
÷.. ÷.
2
. However, both are true, because (
3
;
) =(
;
3
). Moreover, we can remark that the prime
ideal lying over ] with (
;
3
) =1 can be generated by an element in ZŒ
÷3.
61
CHAPTER 4
The Finiteness of the Class Number
In this section we prove the ﬁrst main theorem of the course: the class number of a number
ﬁeld is ﬁnite. The method of proof is effective: it gives an algorithm for computing the
class group.
Norms of ideals
Let ·be a Dedekind domain with ﬁeld of fractions 1, and let T be the integral closure of ·
in a ﬁnite separable extension 1. We want to deﬁne a homomorphism Nm: Id(T) ÷Id(·)
which is compatible with taking norms of elements, i.e., such that the following diagram
commutes:
1
×
b(b)
÷÷÷÷÷ Id(T)
Nm
Nm
1
×
o(o)
÷÷÷÷÷ Id(·)
(12)
Because Id(T) is the free abelian group on the set of prime ideals, we only have to deﬁne
Nm(p) for p prime.
Let p be a prime ideal ·, and factor pT =
¸
P
e
i
i
. If p is principal, say p =(¬), then
we should have
Nm(pT) =Nm(¬ T) =Nm(¬) · =(¬
n
) =p
n
. m=Œ1: 1.
Also, because Nm is to be a homomorphism, we should have
Nm(pT) =Nm(
¸
P
e
i
i
) =
¸
Nm(P
i
)
e
i
.
On comparing these two formulas, and recalling (3.34) that m =
¸
e
i
}
i
, we see that we
should deﬁne Nm(P
i
) =p
(
i
. We take this as our deﬁnition:
Nm(P) =p
((P¡p)
where p =P¨· and }(P¡p) =ŒT¡P: ·¡p.
To avoid confusion, I sometimes use N to denote norms of ideals.
If we have a tower of ﬁelds M ¯1 ¯1, then
N
1¡1
(N
Æ¡1
a) =N
Æ¡1
a
because }(Q¡P) }(P¡p) = }(Q¡p), i.e., ŒC¡Q : T¡P ŒT¡P : ·¡p = ŒC¡Q : ·¡p
where C ¯T ¯· are the integral closures of · in M, 1, and 1 respectively.
62
Norms of ideals
PROPOSITION 4.1 Let · cT and 1 c1 be as above.
(a) For any nonzero ideal a c·, N
1¡1
(aT) =a
n
, where m=Œ1 : 1.
(b) Suppose 1 is Galois over 1. Let Pbe a nonzero prime ideal of T and let p =P¨·.
Write p T =(P
1
P
v
)
e
(cf. 3.34). Then
NP T =(P
1
P
v
)
e(
=
¸
c÷Gal(1¡1)
oP.
(c) For any nonzero element ˇ ÷ T, Nm(ˇ) · =Nm(ˇ T) (i.e., (12) commutes).
PROOF. (a) It sufﬁces to prove this for a prime ideal p, and for such an ideal we have that
N(pT) =N(
¸
P
e
i
i
)
def
=p
¸
e
i
(
i
=p
n
(by 3.34).
(b) Since NP
i
=p
(
for each i , the ﬁrst equality is obvious. In the course of the proof
of (3.34), we showed that Gal(1¡1) acts transitively on the set {P
1
. .... P
v
], and it follows
that each P
i
occurs
n
v
=e} times in the family {oP[ o ÷ Gal(1¡1)].
(c) Suppose ﬁrst that 1is Galois over 1, and let ˇ T =b. The map a ÷a T: Id(·) ÷
Id(T) is injective (remember they are the free abelian groups on the sets of nonzero prime
ideals), and so it sufﬁces to show that Nm(ˇ) T =Nm(b) T. But
Nm(b) T
(b)
=
¸
ob =
¸
(oˇ T) =(
¸
oˇ) T =Nm(ˇ) T
as required.
In the general case, let 1 be a ﬁnite Galois extension of 1 containing 1, and let
J = Œ1: 1. Let C be the integral closure of T in 1. From (a), the Galois case, and
the transitivity of N we have that
N
1¡1
(ˇ T)
d
=N
T¡1
(ˇ C) =Nm
T¡1
(ˇ) · =Nm
1¡1
(ˇ)
d
·.
As the group of ideals Id(·) is torsionfree, this implies that N
1¡1
(ˇ T) =Nm
1¡1
(ˇ)
·.
2
Let a be a nonzero ideal in the ring of integers O
1
of a number ﬁeld 1. Then a is of
ﬁnite index in O
1
, and we let Na, the numerical norm of a, be this index:
Na =(O
1
: a).
PROPOSITION 4.2 Let O
1
be the ring of integers in a number ﬁeld 1.
(a) For any ideal a in O
1
, N
1¡Q
(a) =(N(a)); therefore N(ab) =N(a)N(b).
(b) Let b ca be fractional ideals in 1; then
(a : b) =N(a
1
b).
PROOF. (a) Write a =
¸
p
i
i
i
, and let }
i
=}(p
i
¡]
i
) where (]
i
) =Z¨p
i
; then Nm(p
i
) =
(]
i
)
(
i
. From the Chinese remainder theorem, O
1
¡a .
¸
O
1
¡p
i
i
i
, and so (O
1
: a) =
¸
(O
1
: p
i
i
i
). In the course of the proof of (3.34), we showed that O
1
¡p
i
i
i
is a vector space
of dimension }
i
r
i
over F
;
i
, and so (O
1
: p
i
i
i
) =]
(
i
i
i
i
. On taking the product over i , we ﬁnd
63
4. THE FINITENESS OF THE CLASS NUMBER
that (O
1
: a) =
¸
(]
(
i
i
i
i
) =N
1¡Q
a. When we identify the set of nonzero ideals in Z with
the set of positive integers, then N becomes identiﬁed with N, and so the multiplicativity
of N follows from that of N.
(b) For any nonzero J ÷ 1, the map . ÷J.: 1 ÷1 is an additive isomorphism, and
so (Ja : Jb) =(a : b). Since (Ja)(Jb)
1
=ab
1
, we may suppose that a and b are integral
ideals. The required formula then follows from (a) and the formulas
(O
1
: a)(a : b) =(O
1
: b)
and
N(a) N(a
1
b) =N(b).
2
Statement of the main theorem and its consequences
We now state the main theorem of this section and discuss some of its consequences.
THEOREM 4.3 Let 1 be an extension of degree n of Q, and let z
1
be the discriminant of
1¡Q. Let 2s be the number of nonreal complex embeddings of 1. Then there exists a set
of representatives for the ideal class group of 1 consisting of integral ideals a with
N(a) _
nŠ
n
n
4
¬
x
[z
1
[
1
2
.
The number on the right is called the Minkowski bound — we sometimes denote it
by T
1
. The term C
1
=
nŠ
n
n
4
t
x
is called the Minkowski constant. It takes the following
values:
n r s C
2 0 1 0.637
2 2 0 0.500
3 1 1 0.283
3 3 0 0.222
4 0 2 0.152
4 2 1 0.119
4 4 0 0.094
5 1 2 0.062
5 3 1 0.049
5 5 0 0.038
. . . . . . . . . . . .
100 100 0 0.93×10
42
Here r is the number of real embeddings of 1. We have
1˝
Q
1 ~1
i
×C
x
.
and, if 1 =QŒ˛ and }(X) is the minimum polynomial of ˛, then r is the number of real
roots of }(X) and 2s is the number of its nonreal roots. To see that these descriptions of r
and s agree, apply (1.18).
Before proving (4.3), we give some applications and examples.
64
Statement of the main theorem and its consequences
THEOREM 4.4 The class number of 1 is ﬁnite.
PROOF. It sufﬁces to show that there are only ﬁnitely many integral ideals a in O
1
such
that N(a) is less than the Minkowski bound — in fact, we shall show that, for any integer
M, there are only ﬁnitely many integral ideals a with N(a) <M. If a =
¸
p
i
i
i
, then N(a) =
¸
]
i
i
(
i
i
where (]
i
) =p
i
¨Z. As N(a) <M, this allows only ﬁnitely many possibilities for
the ]
i
(and hence for the p
i
), and only ﬁnitely many possibilities for the exponents r
i
.
2
Let S be the set of integral ideals in 1 with norm < T
1
. Then S is a ﬁnite set, and
Cl(O
1
) =S¡ ~. where a ~b if one ideal is the product of the other with a principal (frac
tional) ideal. There is an algorithm for ﬁnding S, and an algorithm for deciding whether
a ~b, and so there is an algorithm for ﬁnding Cl(O
1
) (the group, not just its order). To ﬁnd
S, ﬁnd the prime ideal factors of enough prime numbers, and form some of their products.
To decide whether a ~b, one has to decide whether c =ab
1
is principal. From (4.2b) we
know that, for , ÷ c,
c =(,) ” Nc =[ Nm,[
and so we have to solve the equation:
Nm, =constant.
When we express , in terms of an integral basis, this becomes a (very special) type of
diophantine equation. For a descriptions of algorithms for ﬁnding Cl(O
1
), see Cohen
1993, 6.5, and Pohst and Zassenhaus 1989, p424.
EXAMPLE 4.5 Let 1 =QŒi . The condition in Theorem 4.3 is that N(a) _
2
4
4
t
2 < 1.27.
There are no such ideals other than ZŒi , and so ZŒi  is a principal ideal domain. (Of course,
the elementary proof of this shows more, namely, that ZŒi  is a Euclidean domain. Even
for rings of integers in number ﬁelds, it is not true that all principal ideal domains are
Euclidean domains. For example, QŒ
÷19 has class number 1, but its ring of integers
is not a Euclidean domain. For more on such things, see the survey article Lemmermeyer
1995
1
.)
EXAMPLE 4.6 Let 1 =QŒ
÷5. Here N(a) _ 0.63×
20< 3. Any ideal satisfying this
must divide (2). In fact, (2) =p
2
where p =(2. 1 ÷
÷5), and Np
2
=N(2) =4, and so
Np = 2. The ideals O
1
and p form a set of representatives for Cl(ZŒ
÷5). The ideal
p can’t be principal because there does not exist an element ˛ = m÷n
÷5 such that
Nm(˛) =m
2
÷5n
2
=2, and so Cl(ZŒ
÷5) has order 2.
EXAMPLE 4.7 Let 1 be a cubic ﬁeld with discriminant <0. Since the sign of z
1
is (÷1)
x
,
and Œ1 : Q =r ÷2s, we have s =1, r =1. The Minkowski bound is
T
1
<0.283[z
1
[
1
2
.
For [z
1
[ _49, T
1
<2, and so for cubic ﬁelds with ÷49 _z
1
<0, the class number h =1.
For example, this is true for the number ﬁelds with discriminants ÷23 and ÷31 discussed
earlier (see 2.36, 2.37).
1
Lemmermeyer, Franz. The Euclidean algorithm in algebraic number ﬁelds. Exposition. Math. 13 (1995),
no. 5, 385–416.
65
4. THE FINITENESS OF THE CLASS NUMBER
For the stem ﬁeld of X
3
÷10X ÷1, the discriminant is ÷4027, and the Minkowski
bound is <18. Recall from (3.48) that
(2) =(2. 1÷˛)(2. 1÷˛ ÷˛
2
).
Let p =(2. 1 ÷˛); its norm is 2. One can show that it generates the class group, and that
it has order 6 in the class group, i.e., p
6
but no smaller power is principal. Hence the class
group is cyclic of order 6. (The proof takes quite a bit of hard work if you do it by hand —
see Artin 1959, 12.6, 13.3. Using PARI, you can type “bnfclgp(X^3+10*X+1)”)
EXAMPLE 4.8 Let ˛ be a root of }(X) = X
5
÷X ÷1. We saw in (2.39) that }(X) is
irreducible and its discriminant is 19×151, and so the ring of integers of QŒ˛ is ZŒ˛.
According to Theorem 4.3, every class of ideals for QŒ˛ contains an integral ideal a
with
N(a) <0.062×
19×151 =3.3 <4.
If p is a prime ideal with N(p) =2, then the residue ﬁeld at p must be F
2
, and }(X) must
have a root mod 2; however, both }(0) and }(1) are odd, and so }(X) doesn’t have a root
in F
2
, which shows that p doesn’t exist. Similarly, there is no prime ideal p with N(p) =3,
and so O
1
is a principal ideal domain!
The Galois group of the splitting ﬁeld M of }(X) is S
5
(later we shall see how to
ﬁnd Galois groups; for the moment type “polgalois(X^5X1)” in PARI), and hence
ŒM: Q =120. It is possible to show that M is unramiﬁed over QŒ
19×151.
An extension 1 of a number ﬁeld 1 is said to be unramiﬁed over 1 if no prime ideal
of O
1
ramiﬁes in O
1
.
THEOREM 4.9 There does not exist an unramiﬁed extension of Q.
PROOF. Let 1 be a ﬁnite extension of Q. Since a set of representatives for the class group
must have at least one element, and that element will have numerical norm _ 1, Theorem
4.3 shows that
[z[
1
2
_
n
n
nŠ
¬
4
x
_
n
n
nŠ
¬
4
n¡2
.
Let a
n
= rhs. Then a
2
> 1, and
o
nC1
o
n
=
t
4
1
2
(1 ÷
1
n
)
n
> 1, and so the sequence a
n
is
monotonically increasing. Hence the discriminant of 1 has absolute value > 1, and we
know from (3.35) that any prime dividing the discriminant ramiﬁes.
2
COROLLARY 4.10 There does not exist an irreducible monic polynomial }(X) ÷ ZŒX of
degree >1 with discriminant ˙1.
PROOF. Let }(X) be such a polynomial, and let ˛ be a root of }(X). Then disc(ZŒ˛¡Z) =
˙1, and so ZŒ˛ is the ring of integers in 1
def
=QŒ˛ and disc(O
1
¡Z) =˙1, which contra
dicts the theorem.
2
REMARK 4.11 There may exist unramiﬁed extensions of number ﬁelds other than Q. In
fact, class ﬁeld theory says that the maximal abelian unramiﬁed
2
extension of 1 (called
2
The Hilbert class ﬁeld 1 of 1 is required to be unramiﬁed even at the inﬁnite primes — this means that
every real embedding of 1 extends to a real embedding of 1.
66
Lattices
the Hilbert class ﬁeld of 1) has Galois group canonically isomorphic to Cl(O
1
). For
example, the theory says that QŒ
÷5 has an unramiﬁed extension of degree 2, and one
veriﬁes that QŒ
÷1.
÷5 is unramiﬁed over QŒ
÷5. In particular, the discriminant of
QŒ
÷1.
÷5 over QŒ
÷5 is a unit.
REMARK 4.12 Let 1
1
be a number ﬁeld with class number h
1
1
>1. Its Hilbert class ﬁeld
is an abelian unramiﬁed extension 1
2
of 1
1
with Gal(1
2
¡1
1
) .Cl(1
1
). Let 1
3
be the
Hilbert class ﬁeld of 1
2
, and so on. In this way, we obtain a tower of ﬁelds,
1
1
c1
2
c1
3
c
It was a famous question (class ﬁeld tower problem) to decide whether this tower can be
inﬁnite, or must always terminate with a ﬁeld of class number 1 after a ﬁnite number of
steps. It was shown by Golod and Shafarevich in the early 60s that the tower is frequently
inﬁnite. See Roquette 1967.
Lattices
Let V be a vector space of dimension n over 1. A lattice ¹ in V is a subgroup of the form
¹=Ze
1
÷ ÷Ze
i
with e
1
. .... e
i
linearly independent elements of V . Thus a lattice is the free abelian subgroup
of V generated by elements of V that are linearly independent over 1. When r = n, the
lattice is said to be full. At the opposite extreme, ¹ = {0] is a lattice (generated by the
empty set of elements). In terms of tensor products, one can say that a full lattice in V is a
subgroup ¹ of V such that the map
¸
r
i
˝.
i
÷
¸
r
i
.
i
: 1˝
Z
¹÷V.
is an isomorphism.
NONEXAMPLE 4.13 The subgroup Z÷Z
2 of 1 is a free abelian group of rank 2 (be
cause
2 is not rational), but it is not a lattice in 1.
We shall need another criterion for a subgroup ¹ of V to be a lattice. The choice of
a basis for V determines an isomorphism of V with 1
n
, and hence a topology on V ; the
topology is independent of the basis, because any linear automorphism of 1
n
is a homeo
morphism. A subgroup ¹ of V is said to be discrete if it is discrete in the induced topology.
A topological space is discrete if its points (hence all subsets) are open, and so to say
that ¹ is discrete means that every point ˛ of ¹ has a neighbourhood U in V such that
U ¨¹={˛].
LEMMA 4.14 The following conditions on a subgroup ¹of a ﬁnitedimensional real vector
space V are equivalent:
(a) ¹ is a discrete subgroup;
(b) there is an open subset U of V such that U ¨¹={0];
(c) each compact subset of V intersects ¹ in a ﬁnite set;
(d) each bounded subset of V intersects ¹ in a ﬁnite set.
67
4. THE FINITENESS OF THE CLASS NUMBER
PROOF. (a) ”(b). Obviously (a) implies (b). For the converse, note that the translation
map . ÷˛ ÷.: V ÷V is a homeomorphism, and so, if U is a neighbourhood of 0 such
that U ¨¹={0], then ˛ ÷U is a neighbourhood of ˛ such that (˛ ÷U) ¨¹={˛].
(a)=(c). Condition (a) says that ¹ is a discrete space for the induced topology. Hence,
if C is compact, then C ¨¹ is both discrete and compact, and therefore must be ﬁnite.
(c)=(d). The closure of a bounded set in 1
n
(hence in V ) is compact, and so this is
obvious.
(d)=(b). Let U be a bounded open neighbourhood of 0. Then S =U ¨¹{0] is ﬁnite
and hence closed, and so U S is an open neighbourhood of {0] such that (U S) ¨¹=
{0].
2
PROPOSITION 4.15 A subgroup ¹ of V is a lattice if and only if it is discrete.
PROOF. Clearly, a lattice is discrete. For the converse, let ¹ be a discrete subgroup of V ,
and choose a maximal 1linearly independent subset {e
1
. . . . . e
i
] of ¹. We shall argue by
induction on r.
If r =0, ¹=0, and there is nothing to prove.
If r =1, then ¹c1e
1
. Because ¹ is discrete, for each M >0,
{ae
1
[ [a[ <M] ¨¹
is ﬁnite, and so there is an } ÷ ¹ such that, when we write } =ae
1
, a attains its minimum
value > 0. I claim ¹=Z} . Any ˛ ÷ ¹Z} will equal (m÷b)} for some m÷ Z and b
with 0 < b < 1; but then (˛ ÷m} ) =b} =abe
1
, and 0 < ab < a, which contradicts our
choice of }.
If r > 1, we let ¹
t
= ¹¨(1e
1
÷ ÷1e
i1
). Clearly this is a discrete subgroup of
the vector space V
t
def
=1e
1
÷ ÷1e
i1
and so, by induction, ¹
t
=Z}
1
÷ ÷Z}
i1
for
some }
i
that are linearly independent over 1 (and hence also form a basis for V
t
). Every
˛ ÷ ¹ can be written uniquely
˛ =a
1
}
1
÷ ÷a
i1
}
i1
÷ae
i
. a
i
. a ÷ 1.
Let c: ¹÷1 be the map ˛ ÷a, and let ¹
tt
=Im(c). Note that a is also the image of
(a
1
÷Œa
1
)}
1
÷ ÷(a
i1
÷Œa
i1
)}
i1
÷ae
i
. Œ+ =integer part,
and so each element a ÷ ¹
tt
in a bounded set, say with 0 _ [a[ < M, is the image of an
element of ¹ in a bounded set,
0 _a
i
<1. i =1. . . . . r ÷1. [a[ <M.
Thus there are only ﬁnitely many such as, and so ¹
tt
is a lattice in 1, say ¹
tt
=Z c(}
i
),
}
i
÷ ¹.
Let ˛ ÷ ¹. Then c(˛) = ac(}
i
) for some a ÷ Z, and c(˛ ÷a}
i
) = 0. Therefore
˛ ÷a}
i
÷ ¹
t
, and so it can be written
˛ ÷a}
i
=a
1
}
1
÷ ÷a
i1
}
i1
. a
i
÷ Z.
Hence
˛ =a
1
}
1
÷ ÷a
i1
}
i1
÷a}
i
. a
i
. a ÷ Z.
which proves that ¹=
¸
Z}
i
.
2
68
Lattices
Let V be a real vector space of dimension n, and let ¹ be a full lattice in V , say
¹=
¸
Ze
i
. For any z
0
÷ ¹, let
D ={z
0
÷
¸
a
i
e
i
[ 0 _a
i
<1].
Such a set is called a fundamental parallelopiped for ¹. The shape of the parallelopiped
depends on the choice of the basis (e
i
), but if we ﬁx the basis and vary z
0
÷ ¹, then the
parallelopipeds cover 1
n
without overlaps.
REMARK 4.16 (a) For a fundamental parallelopiped D of a full lattice
¹=Z}
1
÷ ÷Z}
n
in 1
n
, the volume of D
u(D) =[ det(}
1
. . }
n
)[.
(See any good book on calculus.) If also
¹=Z}
t
1
÷Z}
t
2
÷ ÷Z}
t
n
.
then the determinant of the matrix relating {}
i
] and {}
t
i
] has determinant ˙1, and so the
volume of the fundamental parallelopiped doesn’t depend on the choice of the basis for ¹.
(b) When ¹¯¹
t
are two full lattices 1
n
, we can choose bases {e
i
] and {}
i
] for ¹ and
¹
t
such that }
i
=m
i
e
i
with m
i
a positive integer. With this choice of bases, the fundamen
tal parallelopiped D of ¹ is a disjoint union of (¹: ¹
t
) fundamental parallelopipeds D
t
of
¹
t
. Hence
u(D
t
)
u(D)
=(¹: ¹
t
) (*).
As we noted above, the choice of a basis for V determines an isomorphism V ~ 1
n
,
and hence a measure u on V . This measure is translation invariant (because the Lebesgue
measure on 1
n
is translation invariant), and welldeﬁned up to multiplication by a nonzero
constant (depending on the choice of the basis)
3
. Thus the ratio of the measures of two sets
is welldeﬁned, and the equation (*) holds for two full lattices ¹¯¹
t
in V .
THEOREM 4.17 Let D
0
be a fundamental parallelopiped for a full lattice in V , and let S
be a measurable subset in V . If u(S) > u(D
0
), then S contains distinct points ˛ and ˇ
such that ˇ÷˛ ÷ ¹.
PROOF. The set S ¨D is measurable for all fundamental parallelopipeds D, and
u(S) =
¸
u(S ¨D)
(sum over translates of D by elements of ¹). For each D, a (unique) translate of S ¨D by
an element of ¹ will be a subset of D
0
. Since u(S) >u(D
0
), at least two of these sets will
overlap, i.e., there exist elements ˛. ˇ ÷ S such that
˛ ÷z =ˇ÷z
t
. some z. z
t
÷ ¹.
Then ˇ÷˛ ÷ ¹.
2
3
The experts will recognize u as being a Haar measure on V .
69
4. THE FINITENESS OF THE CLASS NUMBER
REMARK 4.18 In the language of differential geometry, the theorem can be given a more
geometric statement. Let M = V¡¹; it is an ndimensional torus. The measure u on V
deﬁnes a measure on M for which M has measure u(M) =u(D). The theorem says that
if u(S) >u(M), then the restriction of the quotient map V ÷M to S can’t be injective.
Let T be a set such that
˛,ˇ ÷ T =
1
2
(˛ ÷ˇ) ÷ T. (**)
and let S =
1
2
T . Then T contains the difference of any two points of S, and so T will
contain a point of ¹ other than the origin whenever
u(D) <u(
1
2
T ) =2
n
u(T ).
i.e., whenever
u(T ) >2
n
u(D).
We say that a set T is convex if, with any two points, it contains the line joining the
two points, and that T is symmetric in the origin if ˛ ÷ T implies ÷˛ ÷ T . A convex set,
symmetric in the origin, obviously satisﬁes (**), and so it will contain a point of ¹\{0] if
its volume is greater than 2
n
u(D).
THEOREM 4.19 (MINKOWSKI’S) Let T be a subset of V that is compact, convex, and
symmetric in the origin. If
u(T ) _2
n
u(D)
then T contains a point of the lattice other than the origin.
PROOF. Replace T with (1÷c)T , c >0. Then
u((1÷c)T ) =(1÷c)
n
u(T ) >2
n
u(D).
and so (1 ÷c)T contains a point of ¹ other than the origin (see the preceding remark). It
will contain only ﬁnitely many such points because ¹ is discrete and (1÷c)T is compact.
Because T is closed
T =
¸
t>0
(1÷c)T.
If none of the (ﬁnitely many) points of ¹¨(1÷c)T other than the origin is in T , we will
be able to shrink (1 ÷c)T (keeping c > 0) so that it contains no point of ¹ other than the
origin—which is a contradiction.
2
REMARK 4.20 Theorem 4.19 was discovered by Minkowski in 1896. Although it is al
most trivial to prove, it has lots of nontrivial consequences, and was the starting point for
the branch of number theory called the “geometry of numbers”. We give one immediate
application of it to prove that every positive integer is a sum of four squares of integers.
From the identity
(a
2
÷b
2
÷c
2
÷J
2
)(·
2
÷T
2
÷C
2
÷D
2
) =
(a·÷bT ÷cC ÷JD)
2
÷(aT ÷b·÷cD÷JC)
2
÷
(aC ÷bD÷c·÷JT)
2
÷(aD÷bC ÷cT ÷J·)
2
.
70
Some calculus
we see that it sufﬁces to prove that a prime ] is a sum of four squares.
Since
2 =1
2
÷1
2
÷0
2
÷0
2
.
we can suppose that ] is odd. I claim that the congruence
m
2
÷n
2
÷1 ÷0 mod ]
has a solution in Z. As m runs through 0. 1. . . . . ]÷1, m
2
takes exactly (]÷1)¡2 distinct
values modulo ], and similarly for ÷1 ÷n
2
. For the congruence to have no solution, all
these values, ]÷1 in total, must be distinct, but this is impossible.
Fix a solution m. n to the congruence, and consider the lattice ¹ c Z
4
consisting of
(a. b. c. J) such that
c ÷ma÷nb. J ÷mb ÷na mod ].
Then Z
4
¯ ¹ ¯ ]Z
4
and ¹¡]Z
4
is a 2dimensional subspace of F
4
;
(the a and b can be
arbitrary mod ], but then c and J are determined). Hence ¹ has index ]
2
in Z
4
, and so the
volume of a fundamental parallelopiped is ]
2
. Let T be a closed ball of radius r centered
at the origin. Then T has volume ¬
2
r
4
¡2, and so if we choose r so that 2] > r
2
> 1.9]
say, then
u(T ) >16u(D).
According to Minkowski’s theorem, there is a point (a. b. c. J) ÷ (¹\ {0]) ¨T . Because
(a. b. c. J) ÷ ¹,
a
2
÷b
2
÷c
2
÷J
2
÷a
2
(1÷m
2
÷n
2
) ÷b
2
(1÷m
2
÷n
2
) ÷0 mod ].
and because (a. b. c. J) ÷ T ,
a
2
÷b
2
÷c
2
÷J
2
<2].
As a
2
÷b
2
÷c
2
÷J
2
is a positive integer, these conditions imply that it equals ].
This result was stated by Fermat. Euler tried to prove it over a period of 40 years, and
Lagrange succeeded in 1770.
Some calculus
4.21 Let V be a ﬁnitedimensional real vector space. A norm on V is a function  : V ÷
1 such that
(a) for all x ÷ V , x _0, and x =0 ” x =0;
(b) for r ÷ 1 and x ÷ V , rx =[r[x;
(c) (triangle law) for x. y ÷ V , x÷y _x÷y.
Let V =1
i
×C
x
— it is a real vector space of dimension n =r ÷2s. Deﬁne a norm
on V by
x =
i
¸
i=1
[.
i
[ ÷2
i÷x
¸
i=i÷1
[:
i
[
if x =(.
1
. .... .
i
. :
i÷1
. .... :
i÷x
).
71
4. THE FINITENESS OF THE CLASS NUMBER
LEMMA 4.22 For any real number t >0, let
X(t ) ={x ÷ V [ x _t ].
Then
u(X(t )) =2
i
(¬¡2)
x
t
n
¡nŠ.
PROOF. Since X(t ) is symmetric with respect to the r real axes, we have
u(X(t )) =2
i
u(Y(t ))
where Y(t ) ={x [ x _t , .
1
. .... .
i
_0]. For the complex variables, we make the change
of variable
:
}
=.
}
÷i.
}
=
1
2
j
}
(cos0
}
÷i sin0
}
).
The Jacobian of this change of variables is j
}
¡4. After integrating over the 0
}
, for 0 _0
}
_
2¬, we ﬁnd that
u(X(t )) =2
i
4
x
(2¬)
x
7
j
i÷1
j
i÷x
J.
1
J.
i
Jj
i÷1
Jj
i÷x
where
7 ={(x. j) ÷ 1
i÷x
[ .
i
. j
i
_0.
¸
.
i
÷
¸
j
i
_t ].
The result now follows from the next lemma by taking: m = r ÷s; a
i
= 0, 1 _ i _ r;
a
i
=1, r ÷1 _i _m; for then
u(X(t )) =2
i
4
x
(2¬)
x
t
n
¡nŠ
as required.
2
LEMMA 4.23 For a
i
>0 ÷ 1, let
1(a
1
. .... a
n
. t ) =
7(t)
.
o
1
1
.
o
m
n
J.
1
J.
n
.
where 7(t ) ={. ÷ 1
n
[ .
i
_0,
¸
.
i
_t ]. Then
1(a
1
. . . . . a
n
: t ) =t
¸
o
i
÷n
1(a
1
÷1) 1(a
n
÷1)
1(a
1
÷ ÷a
n
÷m÷1)
.
PROOF. Recall that, by deﬁnition, (e.g., Widder, D., Advanced Calculus, 1961, Chapter
11),
1(.) =
o
0÷
e
t
t
x1
Jt.
It takes the value 1(n) =(n÷1)Š for n a nonnegative integer.
By making the change of variables .
t
i
=t.
i
in 1, we see that
1(a
1
. . . . . a
n
: t ) =t
¸
o
i
÷n
1(a
1
. . . . . a
n
: 1).
72
Finiteness of the class number
Therefore it sufﬁces to prove the formula for t =1. We prove this case by induction on m.
First, we have
1(a
1
: 1) =
1
0
.
o
1
1
J.
1
=
1
a
1
÷1
=
1(a
1
÷1)
1(a
1
÷2)
.
Let
7(.
n
)
t
={x ÷ 1
n1
[ .
i
_0.
¸
.
i
_1÷.
n
].
Then
1(a
1
. .... a
n
: 1) =
1
0
.
o
m
n
7(x
m
)
0
.
o
1
1
.
o
m1
n1
J.
1
J.
n1
J.
n
.
=
1
0
.
o
m
n
1(a
1
. .... a
n1
: 1÷.
n
)J.
n
=1(a
1
. .... a
n1
: 1)
1
0
.
o
m
n
(1÷.
n
)
¸
o
i
÷n1
J.
n
=1(a
1
. .... a
n1
: 1)
1(a
n
÷1)1(a
1
÷ ÷a
n1
÷m)
1(a
1
÷ ÷a
n
÷m÷1)
.
In the last step, we used the standard formula
1
0
.
n1
(1÷.)
n1
J. =T(m. n) =
1(m)1(n)
1(m÷n)
.
2
EXAMPLE 4.24 (a) Case r =2, s =0. Then X(t ) is deﬁned by [.[ ÷[.[ _t . It is a square
of side
2t , and so u(X(t )) =2t
2
.
(b) Case r =0, s =1. Then X(t ) is the circle of radius t ¡2, which has area ¬t
2
¡4.
LEMMA 4.25 Let a
1
. . . . . a
n
be positive real numbers. Then
(
¸
a
i
)
1¡n
_(
¸
a
i
)¡n:
equivalently,
¸
a
i
_(
¸
a
i
)
n
¡n
n
.
(The geometric mean is less than or equal to the arithmetic mean.)
PROOF. See any good course on advanced calculus.
2
Finiteness of the class number
Let 1 be a number ﬁeld of degree n over Q. Suppose that 1 has r real embeddings
{o
1
. . . . . o
i
] and 2s complex embedding {o
i÷1
. ¨ o
i÷1
. . . . . o
i÷x
. ¨ o
i÷x
]. Thus n = r ÷2s.
We have an embedding
o: 1 ·÷1
i
×C
x
. ˛ ÷(o
1
˛. . . . . o
i÷x
˛).
We identify V
def
=1
i
×C
x
with 1
n
using the basis {1. i ] for C.
73
4. THE FINITENESS OF THE CLASS NUMBER
PROPOSITION 4.26 Let a be an ideal in O
1
; then o(a) is a full lattice in V , and the volume
of a fundamental parallelopiped of o(a) is 2
x
Na [z
1
[
1
2
.
PROOF. Let ˛
1
. . . . . ˛
n
be a basis for a as a Zmodule. To prove that o(a) is a lattice
we show that the vectors o(˛
1
). . . . . o(˛
n
) are linearly independent, and we prove this by
showing that the matrix ·, whose i th row is
(o
1
(˛
i
). . . . . o
i
(˛
i
). m(o
i÷1
˛
i
). `(o
i÷1
˛
i
). . . .)
has nonzero determinant.
First consider the matrix T whose i th row is
(o
1
(˛
i
). . . . . o
i
(˛
i
). o
i÷1
(˛
i
). o
i÷1
(˛
i
). . . . . o
i÷x
(˛
i
)).
We saw in (2.26) that det(T)
2
=disc(˛
1
. . . . . ˛
n
) =0.
What is the relation between the determinants of · and T? Add column r ÷2 in T
to column r ÷1, and then subtract 1¡2 column r ÷1 from column r ÷2. This gives us
2m(o
i÷1
(˛
i
)) in column r ÷1 and ÷i `(o
i÷1
(˛
i
)) in column r ÷2. Repeat for the other
pairs of columns. These column operations don’t change the determinant of T, and so
det(T) =(÷2i )
x
det(·).
or
det(·) =(÷2i )
x
det(T) =˙(÷2i )
x
disc(˛
1
. . . . . ˛
n
)
1¡2
=0.
Thus o(a) is a lattice in V.
Since o(a) =
¸
n
i=1
Zo(˛
i
), the volume of a fundamental parallelopiped D for o(a) is
[ det(·)[, and from (2.25) we know that
[ disc(˛
1
. . . . . ˛
n
)[ =(O
1
: a)
2
[ disc(O
1
¡Z)[.
Hence
u(D) =2
x
[ disc(˛
1
. . . . . ˛
n
)[
1
2
=2
x
Na [z
1
[
1
2
.
2
PROPOSITION 4.27 Let a be an ideal in O
1
. Then a contains a nonzero element ˛ of 1
with
[ Nm(˛)[ _T
1
Na =
4
¬
x
nŠ
n
n
Na[z
1
[
1
2
.
PROOF. Let X(t ) be as in (4.22), and let D be a fundamental domain for the lattice o(a).
The set X(t ) is compact convex and symmetric in the origin, and so, when we choose t
so large that u(X(t )) _2
n
u(D), Minkowski’s Theorem shows that X(t ) contains a point
o(˛) =0 of o(a). For this ˛ ÷ a,
[ Nm(˛)[ =[o
1
(˛)[ [o
i
(˛)[[o
i÷1
(˛)[
2
[o
i÷x
(˛)[
2
_(
¸
[o
i
˛[ ÷
¸
2[o
i
˛[)
n
¡n
n
(by 4.25)
_t
n
¡n
n
.
74
Binary quadratic forms
In order to have u(X(t )) _2
n
u(D), we need
2
i
(¬¡2)
x
t
n
¡nŠ _2
n
2
x
Na [z
1
[
1
2
.
i.e.,
t
n
_nŠ
2
ni
¬
x
Na [z
1
[
1
2
.
When we take t
n
to equal the expression on the right, we ﬁnd that
[ Nm(˛)[ _
nŠ
n
n
2
ni
¬
x
Na [z
1
[
1
2
.
As n÷r =2s, this is the required formula.
2
PROOF (OF THEOREM 4.3) Let c be a fractional ideal in 1 — we have to show that the
class of c in the ideal class group is represented by an integral ideal a with
Na _T
1
def
=
nŠ
n
n
4
¬
x
[z
1
[
1
2
.
For some J ÷ 1
×
, Jc
1
is an integral ideal, say (J) c
1
=b. According to the result just
proved, there is a ˇ ÷ b, ˇ =0, with
[ Nm(ˇ)[ _T
1
Nb.
Now ˇO
1
cb =ˇO
1
=ab with a integral, and a ~b
1
~c. Moreover,
Na Nb =[ Nm
1¡Q
ˇ[ _T
1
Nb.
On cancelling Nb, we ﬁnd that Na _T
1
.
2
REMARK 4.28 Proposition 4.27 can be useful in deciding whether an integral ideal is prin
cipal.
Binary quadratic forms
Gauss studied binary quadratic forms, and even deﬁned a product for them. This work
was greatly clariﬁed when Kummer and Dedekind deﬁned ideals, and it was realized that
Gauss’s results were related to the ideal class groups of quadratic number ﬁelds. Here I
brieﬂy explain the connection.
By a binary quadratic form we mean an expression of the form
O(X. Y ) =aX
2
÷bXY ÷cY
2
.
We call the form integral if O(m. n) is an integer whenever m and n are integers, or, equiv
alently, if a. b. c ÷ Z. The discriminant of O is
J
O
=b
2
÷4ac.
75
4. THE FINITENESS OF THE CLASS NUMBER
A form is said to be nondegenerate if its discriminant is nonzero. Two integral binary
quadratic forms O and O
t
are said to be equivalent if there exists a matrix ·=
˛ ˇ
, ı
÷
SL
2
(Z) such that
O
t
(X. Y ) =O(˛X ÷ˇY. ,X ÷ıY ).
Clearly, equivalent forms have the same discriminant, but there exist inequivalent forms
with the same discriminant. The question considered by Gauss was to try to describe the
set of equivalence classes of forms with a ﬁxed discriminant.
Let J =1 be a squarefree integer, let 1 =QŒ
J, and let J
1
=disc(O
1
¡Z). Deﬁne
the norm form q
1
by
q
1
(X. Y ) =Nm
1¡Q
(X ÷Y
J) =X
2
÷JY
2
. if J ÷2. 3 mod 4
or
q
1
(X. Y ) =Nm
1¡Q
(X ÷Y
1÷
d
2
) =X
2
÷XY ÷
1d
4
Y
2
. if J ÷1 mod 4.
In both cases q
1
has discriminant J
1
(=4J or J).
In general, if O is an integral binary quadratic form, then J
O
= J
1
}
2
, some integer
} , where 1 =QŒ
J
O
. Moreover, if J
O
=J
1
, then O is primitive, i.e., gcd(a. b. c) =1.
Fix a ﬁeld 1 = QŒ
J and an embedding 1 ·÷C. We choose
J to be positive if
J >0, and to have positive imaginary part if J is negative. Set
J
1
=2
J or
J. Write
Gal(1¡Q) = {1. o]. If J < 0, deﬁne Cl
÷
(1) = Cl(1) (usual class group of 1) and if
J >0, deﬁne
Cl
÷
(1) =Id(1)¡P
÷
(1)
where 1
÷
(1) is the group of principal ideals of the form (˛) with ˛ > 0 under every
embedding of 1 into 1.
Let a be a fractional ideal in 1, and let a
1
. a
2
be a basis for a as a Zmodule. From
(2.25) we know that
ˇ
ˇ
ˇ
ˇ
a
1
a
2
oa
1
oa
2
ˇ
ˇ
ˇ
ˇ
2
=J
1
Na
2
.
After possibly reordering the pair a
1
. a
2
we will have
ˇ
ˇ
ˇ
ˇ
a
1
a
2
oa
1
oa
2
ˇ
ˇ
ˇ
ˇ
=
J
1
Na.
For such a pair, deﬁne
O
o
1
,o
2
(X. Y ) =Na
1
Nm
1¡Q
(a
1
X ÷a
2
Y ).
This is an integral binary quadratic form with discriminant J
1
.
THEOREM 4.29 The equivalence class of O
o
1
,o
2
(X. Y ) depends only on the image of a in
Cl
÷
(1); moreover, the map sending a to the equivalence class of O
o
1
,o
2
deﬁnes a bijec
tion from Cl
÷
(1) to the set of equivalence classes of integral binary quadratic forms with
discriminant J
1
.
PROOF. See Fr¨ ohlich and Taylor 1991, VII.2 (and elsewhere).
2
76
Exercises
In particular, the set of equivalence classes is ﬁnite, and has the structure of an abelian
group. This was known to Gauss, even though groups had not yet been deﬁned. (Gauss
even knew it was a direct sum of cyclic groups.)
ASIDE 4.30 Write h
d
for the class number of QŒ
J, J a squarefree integer = 1. In modern
terminology, Gauss conjectured that, for a ﬁxed h, there are only ﬁnitely many negative J such that
h
d
=h. (Actually, because of a difference of terminology, this is not quite what Gauss conjectured.)
In 1935, Siegel showed that, for every c >0, there exists a constant c >0 such that
h
d
>c[J[
1
2
t
. J < 0.
This proves Gauss’s conjecture. Unfortunately, the c in Siegel’s theorem is not effectively com
putable, and so Siegel’s theorem gives no way of computing the Js for a given h.
In 1951, Tatuzawa showed that Siegel’s theorem is true with an effectively computable c except
for at most one exceptional J.
It is easy to show that h
d
= 1 for ÷J = 1. 2. 3. 7. 11. 19. 43. 67. 163 (exercise!). Thus in 1951
it was known that there exist these 9 quadratic imaginary number ﬁelds with class number 1, and
possibly 1 more.
In 1952 Heegner proved that there was no 10th such ﬁeld, but his proof was not recognized to
be correct until 1969 (by Deuring and Stark). In the interim, Baker (1966), Stark (1966), and Siegel
(1968) had found proofs.
More recently (1983), Goldfeld, Gross, and Zagier showed, using completely different methods
from Siegel, that there is an effective procedure for ﬁnding all J <0 with h
d
equal to a given h. For
an expository article on this, see Goldfeld, Bull. Amer. Math. Soc. 13 (1985), 23–37.
By way of contrast, it is conjectured that there are inﬁnitely many real quadratic ﬁelds with class
number 1, but this has not been proved.
NOTES Fermat stated, and probably proved, the three statements in Exercise 33. However, for 5
he could only state the following conjecture:
If two primes are of the form 20k ÷3 or 20k ÷7, then their product is of the form
.
2
÷5.
2
.
The fact that this statement is more complicated than it is for 1, 2, or 3 was the ﬁrst indication that
the arithmetic of the ring of integers in QŒ
÷5 is more complicated than it is in the ﬁelds with
smaller discriminant. Lagrange found an explanation for Fermat’s statements by showing that all
binary quadratic forms of discriminant ÷4 are equivalent, and similarly for discriminants ÷8 and
÷12, but that the forms of discriminant ÷20 fall into two equivalence classes. Dedekind was able to
interprete this as showing that QŒ
÷5 has class number 2.
Exercises
41 Give an example of an integral domain T, a nonzero prime ideal p in T, and a subring
· of T such that p¨· =0. (Note that this can’t happen if T is integral over · — see the
paragraph preceding 3.30.)
42 Let J c1 c1 be a sequence of number ﬁelds, and let · cT cC be their rings of
integers. If Q[P and P[p (prime ideals in C, T, and · respectively), show that
e(Q¡P) e(P¡p) =e(Q¡p). }(Q¡P) }(P¡p) =}(Q¡p).
43 Let 1 =QŒ˛ where ˛ is a root of X
3
÷X ÷1 (see 2.37). According to (3.34), what
are the possible ways that (]) can factor in O
1
as a product of prime ideals. Which of these
possibilities actually occur? (Illustrate by examples.)
77
4. THE FINITENESS OF THE CLASS NUMBER
44 Show that QŒ
÷23 has class number 3, and that QŒ
÷47 has class number 5.
45 Let 1 be an algebraic number ﬁeld. Prove that there is a ﬁnite extension 1 of 1
such that every ideal in O
1
becomes principal in O
1
. [Hint: Use the ﬁniteness of the class
number.]
46 Let 1 = QŒ˛ where ˛ is a root of X
3
÷X ÷2. Show that O
1
= ZŒ˛ and that 1
has class number 1. [One approach is to consider the square factors of the discriminant of
X
3
÷X ÷2, and show that
1
2
(a÷b˛ ÷c˛
2
) is an algebraic integer if and only if a, b, and
c are all even, but you may be able to ﬁnd a better one.]
47 Let 1 =QŒ
÷1.
5. Show that O
1
=ZŒ
÷1.
1÷
5
2
. Show that the only primes
(in Z) that ramify in 1 are 2 and 5, and that their ramiﬁcation indexes are both 2. Deduce
that 1 is unramiﬁed over QŒ
÷5. Prove that QŒ
÷5 has class number 2, and deduce
that 1 is the Hilbert class ﬁeld of QŒ
÷5. (Cf. 4.11.)
78
CHAPTER 5
The Unit Theorem
In this section we prove the second main theorem of the course.
Statement of the theorem
Recall that a ﬁnitely generated abelian group ·is isomorphic to ·
tors
˚Z
t
for some t where
·
tors
is the (ﬁnite) subgroup of torsion elements of · (i.e., of elements of ﬁnite order). The
number t is uniquely determined by ·, and is called the rank of ·.
As before, we write r for the number of real embeddings of a number ﬁeld 1 and 2s
for the number of nonreal complex embeddings. Thus
1˝
Q
1 ~1
i
×C
x
and r ÷2s =Œ1: Q. Moreover, if 1 =QŒ˛, then r is the number of real conjugates of ˛
and 2s is the number of nonreal complex conjugates.
THEOREM 5.1 The group of units in a number ﬁeld 1 is ﬁnitely generated with rank equal
to r ÷s ÷1.
For example, for a real quadratic ﬁeld, the rank is 2÷0÷1 =1, and for an imaginary
quadratic ﬁeld it is 0÷1÷1 =0.
The theorem is usually referred to as the “Dirichlet Unit Theorem” although Dirichlet
proved it for rings of the form ZŒ˛ rather than O
1
.
Write U
1
(=O
×
1
) for the group of units in 1. The torsion subgroup of U
1
is the group
u(1) of roots of 1 in 1.
A set of units u
1
. . . . . u
i÷x1
is called a fundamental system of units if it forms a basis
for U
1
modulo torsion, i.e., if every unit u can be written uniquely in the form
u =¯u
n
1
1
u
n
rCs1
i÷x1
. ¯ ÷ u(1). m
i
÷ Z.
The theorem implies that u(1) is ﬁnite (and hence cyclic). As we now explain, this
can be proved directly. In Chapter 7, we shall see that, if ¯
n
is a primitive mth root of
1, then QŒ¯ is a Galois extension of Q with Galois group isomorphic to (Z¡mZ)
×
. If
m=
¸
]
i
i
i
is the factorization of minto powers of distinct primes, then Z¡mZ.
¸
Z¡]
i
i
i
Z
by the Chinese remainder theorem, and so (Z¡mZ)
×
.
¸
Z¡]
i
i
i
Z
×
. As the nonunits of
79
5. THE UNIT THEOREM
Z¡]
i
i
i
Z are exactly the elements divisible by ], and there are ]
i
i
1
i
of these, we see that
ˇ
ˇ
(Z¡]
i
i
i
Z)
×
ˇ
ˇ
=]
i
i
1
i
(]
i
÷1), and so
ˇ
ˇ
(Z¡mZ)
×
ˇ
ˇ
=
¸
]
i
i
1
i
(]
i
÷1)
def
=c(m).
Since
¯
n
÷ 1 =QŒ¯
n
 c1 =c(m)[Œ1 : Q.
the ﬁeld 1 can contain only ﬁnitely many ¯
n
.
LEMMA 5.2 An element ˛ ÷ 1 is a unit if and only if ˛ ÷ O
1
and Nm
1¡Q
˛ =˙1.
PROOF. If ˛ is a unit, then there is a ˇ ÷O
1
such that ˛ˇ =1, and then Nm(˛) and Nm(ˇ)
lie in Z and 1 =Nm(˛ˇ) =Nm(˛) Nm(ˇ). Hence Nm˛ ÷ Z
×
={˙1].
For the converse, ﬁx an embedding o
0
of 1 into C, and use it to identify 1 with a
subﬁeld of C. Recall (2.20) that
Nm(˛) =
¸
c:1·C
o˛ =˛
¸
c=c
0
o˛.
Let ˇ =
¸
c=c
0
o˛. If ˛ ÷ O
1
, then ˇ is an algebraic integer. If Nm(˛) = ˙1, then
ˇ =˙˛
1
and so belongs to 1. Therefore, if ˛ satisﬁes both conditions, it has an inverse
˙ˇ in O
1
, and so is a unit.
2
For all real ﬁelds, i.e., ﬁelds with an embedding into 1, u(1) = {˙1]; for “most”
nonreal ﬁelds, this is also true.
EXAMPLE 5.3 Let 1 be a quadratic ﬁeld QŒ
J. Then O
1
={m÷n
J [ m. n ÷ Z] or
{m÷n(1 ÷
J)¡2 [ m. n ÷ Z]. In the two cases, the units in O
1
are the solutions to the
equations
m
2
÷n
2
J =˙1, or
(2m÷n)
2
÷Jn
2
=˙4.
When J < 0, these equations (obviously) have only ﬁnitely many solutions, and so U
1
=
u(1). Note that ¯
n
lies in a quadratic ﬁeld if and only if c(m) _2. This happens only for
m dividing 4 or 6. Thus u(1) ={˙1] except for the following ﬁelds:
QŒi , u(1) ={˙1. ˙i ];
QŒ
÷3, u(1) ={˙1. ˙j. ˙j
2
], with j =(1÷
÷3)¡2).
When J >0, the theoremshows that there are inﬁnitely many solutions, and that U
1
=˙u
Z
for some element u (called the fundamental unit). As Cohn (1978
1
) puts it, “the actual
computation of quadratic units lies in the realm of popularized elementary number theory,
including devices such as continued fractions.” The method is surprisingly effective, and
yields some remarkably large numbers — see later.
1
Cohn, Harvey. A classical invitation to algebraic numbers and class ﬁelds. With two appendices by Olga
Taussky: ”Artin’s 1932 G¨ ottingen lectures on class ﬁeld theory” and ”Connections between algebraic number
theory and integral matrices”. Universitext. SpringerVerlag, New YorkHeidelberg, 1978. xiii+328 pp.
80
Proof that U
1
is ﬁnitely generated
EXAMPLE 5.4 Let 1 = QŒ˛, where ˛ is a root of X
3
÷10X ÷1. We know that the
discriminant z
1
= ÷4027. Since sign(z
1
) = (÷1)
x
and r ÷2s = 3, we must have r =
1 =s. From its minimum equation, we see that Nm(˛) =÷1, and so ˛ is a unit. Clearly
˛ is of inﬁnite order, and later we shall show that it is a fundamental unit, and so U
1
=
{˙˛
n
[ m÷ Z].
Proof that U
1
is ﬁnitely generated
We ﬁrst need an elementary result.
PROPOSITION 5.5 For any integers m and M, the set of all algebraic integers ˛ such that
˘ the degree of ˛ is _m, and
˘ [˛
t
[ <M for all conjugates ˛
t
of ˛
is ﬁnite.
PROOF. The ﬁrst condition says that ˛ is a root of a monic irreducible polynomial of degree
_ m, and the second condition implies that the coefﬁcients of the polynomial are bounded
in terms of M. Since the coefﬁcients are integers, there are only ﬁnitely many such poly
nomials, and hence only ﬁnitely many ˛s.
2
COROLLARY 5.6 An algebraic integer ˛, all of whose conjugates in C have absolute value
1, is a root of 1.
PROOF. According to the proposition, the set {1. ˛. ˛
2
. . . .] is ﬁnite.
2
REMARK 5.7 It is essential to require ˛ to be an algebraic integer. For example, ˛ =
(3 ÷4i )¡5 and its conjugate both have absolute value 1, as do their powers, but the set
{1. ˛. ˛
2
. . . .] is not ﬁnite.
Recall that we previously considered the map
o: 1 ÷1
i
×C
x
. ˛ ÷(o
1
˛. . . . . o
i
˛. o
i÷1
˛. . . . . o
i÷x
˛)
where {o
1
. . . . . o
i
. o
i÷1
. ¨ o
i÷1
. . . . . o
i÷x
. ¨ o
i÷x
] is the complete set of embeddings of 1 into
C. It takes sums to sums. Now we want a map that takes products to sums, and so we take
logarithms. Thus we consider the map:
1 : 1
×
÷1
i÷x
. ˛ ÷(log[o
1
˛[. . . . . log[o
i
˛[. 2log[o
i÷1
˛[. . . . . 2log[o
i÷x
˛[).
It is a homomorphism. If u is a unit in O
1
, then Nm
1¡Q
u =˙1, and so
[o
1
u[ [o
i
u[[o
i÷1
u[
2
[o
i÷x
u[
2
=1.
On taking logs, we see that 1(u) is contained in the hyperplane
H : .
1
÷ ÷.
i
÷2.
i÷1
÷ ÷2.
i÷x
=0.
Dropping the last coordinate deﬁnes an isomorphism H ~1
i÷x1
.
81
5. THE UNIT THEOREM
PROPOSITION 5.8 The image of 1: U ÷H is a lattice in H, and the kernel of 1 is a ﬁnite
group (hence is u(1)).
PROOF. Let C be a bounded subset of H containing 0, say
C c{x ÷ H [ [.
i
[ _M].
If 1(u) ÷ C, then [o
}
u[ _ e
Æ
for all ; , and Proposition 5.5 implies that there are only
ﬁnitely many such us. Thus 1(U) ¨C is ﬁnite, and this implies that 1(U) is a lattice in H
(by 4.15). If ˛ is in the kernel of 1, then [o
i
˛[ =1 for all i , and so the kernel if ﬁnite by
Proposition 5.5.
2
Since the kernel of 1 is ﬁnite, we have
rank(U) =rank(1(U)) _dimH =r ÷s ÷1.
Computation of the rank
We now prove the unit theorem.
THEOREM 5.9 The image 1(U) of U in H is a full lattice; thus U has rank r ÷s ÷1.
PROOF. To prove the theorem, we have to ﬁnd a way to construct units. We work again
with the embedding
o: 1 ·÷1
i
×C
x
~1
i÷2x
.
For x =(.
1
. .... .
i
. .
i÷1
. ...) ÷ 1
i
×C
x
, deﬁne
Nm(x) =.
1
.
i
.
i÷1
¨ .
i÷1
.
i÷x
¨ .
i÷x
.
Then Nm(o(˛)) =Nm(˛). Note that [ Nm(x)[ =[.
1
[ [.
i
[[.
i÷1
[
2
[.
i÷x
[
2
.
Recall from (4.26), that o(O
1
) is a full lattice in 1
i
×C
x
, and the volume of its funda
mental parallelopiped is 2
x
[z[
1
2
; in more detail, if ˛
1
. . . . . ˛
n
is a Zbasis for O
1
, then
we showed that the absolute value of the determinant of the matrix whose i th row is
o(˛
i
) =(o
1
(˛
i
). . . . . m(o
i÷1
(˛
i
)). `(o
i÷1
(˛
i
)). . . .)
is 2
x
[z[
1
2
. In fact, we showed that we could get this matrix from the matrix whose i th
row is
(o
1
(˛
i
). . . . . o
i÷1
(˛
i
). ¨ o
i÷1
(˛
i
). . . .)
by some elementary column operations that multiplied the absolute value of the determinant
by 2
x
, and we know that the determinant of the second matrix is ˙[z[
1
2
.
In the rest of the proof, x will be a point of 1
i
×C
x
with
1¡2 _[ Nm(x)[ _1.
Deﬁne
x o(O
1
) ={x o(˛) [ ˛ ÷ O
1
].
82
Computation of the rank
Since 1
i
×C
x
is a ring, this product makes sense. This is again a lattice in 1
i
×C
x
, and the
volume of its fundamental parallelopiped is the determinant of the matrix whose i th row is
(.
1
o
1
(˛
i
). . . . . m(.
i÷1
o
i÷1
(˛
i
)). `(.
i÷1
o
i÷1
(˛
i
)). . . .).
As before, the absolute value of the determinant of this matrix is 2
x
times the absolute
value of the determinant of the matrix whose i th row is
(.
1
o
1
(˛
i
). . . . . .
i÷1
o
i÷1
(˛
i
). ¨ .
i÷1
¨ o
i÷1
(˛
i
). . . .).
which is
[z[
1
2
[ Nm(x)[.
Therefore x o(O
1
) is a lattice with 2
x
[z[
1
2
[ Nm(x)[ as the volume of its fundamental
domain. Note that as x ranges over our set these volumes remain bounded.
Let T be a compact convex subset of 1
i
×C
x
, which is symmetric in the origin, and
whose volume is so large that, for every x in the above set, Minkowski’s theorem (4.19)
implies there is a point , of O
1
, , = 0, such that x o(,) ÷ T . The points of T have
bounded coordinates, and hence bounded norms, and so
x o(,) ÷ T =[ Nm(x o(,))[ _M.
for some M (depending on T ); thus
[ Nm(,)[ _M¡Nm(x) _2M.
Consider the set of ideals , O
1
, where , runs through the ,’s in O
1
for which x o(,) ÷T
for some x in our set. The normNof such an ideal is _2M, and so there can only be ﬁnitely
many such ideals, say ,
1
O
1
. . . . . ,
t
O
1
. Now if , is any element of O
1
with x o(,) ÷T ,
some x, then , O
1
=,
i
O
1
for some i , and so there exists a unit c such that , =,
i
c.
Then x o(c) ÷ o(,
1
i
) T . The set T
t
=o(,
1
1
) T L... Lo(,
1
t
) T is bounded, and so
we have shown that, for each x in our set there exists a unit c such that the coordinates of
x o(c) are bounded uniformly in x (the set T
t
doesn’t depend on x).
We are now ready to prove that 1(U) is a full lattice in H. If r ÷s ÷1 = 0, there is
nothing to prove, and so we assume r ÷s ÷1 _1.
For each i , 1 _ i _ r ÷s, we choose an x in our set such that all the coordinates of x
except .
i
are very large (compared with T
t
), and .
i
is sufﬁciently small that [ Nmx[ =1.
We know that there exists a unit c
i
such that x o(c
i
) has bounded coordinates, and we
deduce that [o
}
c
i
[ <1 for ; =i , and hence that log[o
}
c
i
[ <0.
I claim that 1(c
1
). .... 1(c
i÷x1
) are linearly independent vectors in the lattice 1(U).
For this we have to prove that the matrix whose i th row is
(l
1
(c
i
). .... l
i÷x1
(c
i
)). l
i
(c) =log[o
i
c[.
is invertible. The elements of the matrix except those on the diagonal are negative, but the
sum
l
1
(c
i
) ÷ ÷l
i÷x1
(c
i
) ÷l
i÷x
(c
i
) =0.
and so the sum of the terms in the i th row
l
1
(c
i
) ÷ ÷l
i÷x1
(c
i
) =÷l
i÷x
(c
i
) >0.
The next lemma implies that the matrix is invertible, and so completes the proof of Theorem
5.9.
2
83
5. THE UNIT THEOREM
LEMMA 5.10 Let (a
i}
) be a real m×m matrix such that
˘ a
i}
<0 for i =; :
˘
¸
}
a
i}
>0 for i =1. 2. .... m.
Then (a
i}
) is invertible.
PROOF. If it isn’t, then the system of equations
¸
a
i}
.
}
=0 i =1. . . . . m
has a nontrivial solution. Write .
1
. .... .
n
for such a solution, and suppose i
0
is such that
[.
i
0
[ =max{[.
}
[]. We can scale the solution so that .
i
0
=1. Then [.
}
[ _1 for ; =i
0
, and
the i
0
th equation gives a contradiction:
0 =
¸
}
a
i
0
}
.
}
=a
i
0
i
0
÷
¸
}=i
0
a
i
0
}
.
}
_a
i
0
i
0
÷
¸
}=i
0
a
i
0
}
>0.
2
Sunits
Let S be a ﬁnite set of prime ideals of 1, and deﬁne the ring of Sintegers to be
O
1
(S) =
¸
p…S
O
p
={˛ ÷ 1 [ ord
p
(˛) _0, all p … S].
For example, if S =0, then O
1
(S) =O
1
.
Deﬁne the group of Sunits, to be
U(S) =O
1
(S)
×
={˛ ÷ 1 [ ord
p
(˛) =0, all p … S].
Clearly, the torsion subgroup of U(S) is again u(1).
THEOREM 5.11 The group of Sunits is ﬁnitely generated with rank r ÷s ÷#S ÷1.
PROOF. Let p
1
. p
2
. . . . . p
t
be the elements of S. The homomorphism
u ÷(. . . . ord
p
i
(u). . . .): U(S) ÷Z
t
has kernel U. To complete the proof, it sufﬁces to show that the image of U(S) in Z
t
has
rank t . Let h be the class number of 1. Then p
h
i
is principal, say p
h
i
=(¬
i
), and ¬
i
is an
Sunit with image
(0. . . . . h. . . . . 0) (h in the i t h position).
Clearly these elements generate a subgroup of rank t .
2
For example, if 1 =Q and S ={(2), (3), (5)] then
U(S) ={˙2
k
3
n
5
n
[ k. m. n ÷ Z].
and the statement is obvious in this case.
84
Example: CM ﬁelds
Example: CM ﬁelds
Anumber ﬁeld is totally real if all of its embeddings in Clie in 1, and it is totally imaginary
if none of its embeddings in C lie in 1. For example, 1 =QŒ˛ .QŒX¡(} ) is totally real
if all the roots of } are real, and it is totally imaginary if none of the roots of } are real.
A CM ﬁeld is a totally imaginary quadratic extension of a totally real ﬁeld. Every such
ﬁeld can be obtained from a totally real ﬁeld by adjoining the square root of an element all
of whose real conjugates are negative.
Let 1 be a CM ﬁeld, which is a quadratic extension of the totally real ﬁeld 1
÷
, and let
2n =Œ1 : Q. Then 1 has 2n complex embeddings and 1
÷
has n real embeddings, and so
rank(U
1
) =n÷1 =rank(U
1
C).
Therefore, U
1
C has ﬁnite index in U
1
. In fact, it is possible to prove more.
PROPOSITION 5.12 The index of u(1) U
1
C in U
1
is either 1 or 2.
PROOF. Let a ÷ ¨ a be the nontrivial automorphism of 1 ﬁxing 1
÷
. Then j( ¨ a) = j(a)
for all homomorphisms j: 1 ÷C. In particular, for any a ÷ U
1
, all conjugates of a¡¨ a
in C have absolute value 1, and so a¡¨ a ÷ u(1) (by 5.6). Consider the map ç: U
1
÷
u(1)¡u(1)
2
determined by a ÷a¡¨ a. Clearly ç is a homomorphism. Suppose u lies in its
kernel, so that u¡ ¨ u =¯
2
for some ¯ ÷ u(1). Then u
¨
¯¡ ¨ u¯ =1, and so u
¨
¯ ÷ 1
÷
. It follows
that u ÷ u(1) U
1
C. Conversely, if u = ¯ u
÷
÷ u(1) U
1
C, then u¡ ¨ u = ¯
2
÷ Ker(ç).
We have shown that Ker(ç) =u(1) U
1
C. As u(1)¡u(1)
2
has order 2, this completes
the proof.
2
Example: real quadratic ﬁelds
An expression
a
0
÷
1
a
1
÷
1
a
2
÷
1
a
3
÷
is called a continued fraction. We abbreviate the expression on the right as
Œa
0
. a
1
. a
2
. . . ..
We shall always assume that the a
i
are integers with a
1
>0, a
2
>0. . . .. The integers a
i
are
called the quotients, and Œa
0
. a
1
. .... a
n
 is called the nth convergent. Every irrational num
ber ˛ can be expressed in just one way as an inﬁnite continued fraction, and the continued
fraction is periodic if and only if ˛ has degree 2 over Q. (See any book on elementary num
ber theory, for example, Hardy, G. H., and Wright, E. M., An Introduction to the Theory of
Numbers, Oxford Univ. Press, 1960 (4th edition), Chapter X.)
Now let J be a squarefree positive integer, and let c be the (unique) fundamental unit
for QŒ
J with c > 1. Let s be the period of the continued fraction for
J and let ]¡q be
the (s ÷1)th convergent of it; then
c =]÷q
J if J ÷2. 3 mod 4, or J ÷1 mod 8.
85
5. THE UNIT THEOREM
and
c =]÷q
J or c
3
=]÷q
J otherwise.
Using a computer algebra program, it is very easy to carry this out, and one obtains some
spectacularly large numbers.
For example, to ﬁnd the fundamental unit in QŒ
94, ﬁrst compute
94 =9. 6954. . ..
Then compute the continued fraction of
94. One gets
{9. 1. 2. 3. 1. 1. 5. 1. 8. 1. 5. 1. 1. 3. 2. 1. 18. 1. 2. 3. . . .].
This suggests the period is 16. Now evaluate the 15th convergent. One gets
2143295
221064
.
Hence the fundamental unit >1 is
c =2143295÷221064
94.
Compute that
(2143295)
2
÷(221064)
2
94 =1.
which veriﬁes that c is a unit.
When one carries out this procedure for QŒ
9199, the ﬁrst coefﬁcient of the funda
mental unit has 88 digits! The computer has no problem ﬁnding the fundamental unit —
the only problem is counting the length of the period, which is about 180.
Example: cubic ﬁelds with negative discriminant
Since the sign of the discriminant is (÷1)
x
(see 2.40), a cubic ﬁeld 1 will have negative
discriminant if and only if r =1 =s. We identify 1 with a subﬁeld of 1 using its unique
real embedding. We have z < 0, and the group of units is {˙c
n
] for some c (fundamental
unit). We want to ﬁnd c. Since ÷c, ÷c
1
, and c
1
are also fundamental units, we may
suppose that c >1.
LEMMA 5.13 Let 1 be a cubic extension of Q with negative discriminant, and let c be the
fundamental unit with c >1. Then
[z
1
[ <4c
3
÷24.
PROOF. Since c … Q, it must generate 1. The two conjugates of c (other than c itself) must
be complex conjugates, and so the product of c with its conjugates must be ÷1 (rather than
÷1). Write c =u
2
, u ÷ 1, u >1. Then the remaining conjugates of c can be written
u
1
e
i0
. u
1
e
i0
(0 _0 _¬).
Let z
t
=D(1. c. c
2
) be the discriminant of the minimum equation of c. Then
z
t
1
2
=(u
2
÷u
1
e
i0
)(u
2
÷u
1
e
i0
)(u
1
e
i0
÷u
1
e
i0
) =2i(u
3
÷u
3
÷2cos0)sin0.
If we set 2¯ =u
3
÷u
3
, then
[z
t
[
1
2
=4(¯ ÷cos0)sin0.
86
Example: cubic ﬁelds with negative discriminant
which, for a given u, has a maximum where
¯ cos0 ÷cos
2
0 ÷sin
2
0 =0.
or
÷g(.)
def
=¯. ÷2.
2
÷1 =0. [.[ _1. . =cos0.
We seek a root of g(.) with [.[ < 1. But g(1) = 1 ÷¯ < 0 (because u > 1 implies ¯ =
u
3
u
3
2
>1), and g(÷
1
2u
3
) =
3
4
(u
6
÷1) <0. Since g(.) =2.
2
÷ , it follows g(.) has
one root >1, and that the desired root .
0
, with [.
0
[ _1, is <÷
1
2u
3
. But then
.
2
0
>
1
4u
6
=u
6
÷4.
2
0
<0 =u
6
÷4.
2
0
÷4.
4
0
<0. (13)
This maximum yields
[z
t
[ _16(¯
2
÷2¯.
0
÷.
2
0
)(1÷.
2
0
).
and, on applying the conditions ¯.
0
=2.
2
0
÷1, ¯
2
.
2
0
=4.
4
0
÷4.
2
0
÷1, and the inequality
(13) we ﬁnd that
[z
t
[ _16(¯
2
÷1÷.
2
0
÷.
4
0
) =4u
6
÷24÷4(u
6
÷4.
2
0
÷4.
4
0
) <4u
6
÷24.
Hence
[z
t
[ <4c
3
÷24.
Since z
t
=z
1
(square of an integer), this completes the proof.
2
EXAMPLE 5.14 Let 1 =QŒ˛ where ˛ is a real root of X
3
÷10X ÷1. Here the discrim
inant is ÷4027, and so c >
3
402T24
4
> 10 for c the fundamental unit with c > 1. Note
that Nm(˛) =÷1, and so ˛ is a unit. Moreover, ˛ =÷0.0999003... and so ˇ =÷˛
1
=
10.00998.... Since ˇ is a power of c, we must have ˇ = c; i.e., ÷˛
1
is the fundamental
unit >1. Thus
U
1
={˙˛
n
[ m÷ Z].
Once one knows c, it becomes easier to compute the class group. We know (see 3.48)
that there is a prime ideal p =(2. 1 ÷˛) such that N(p) =2. One shows that p generates
the class group, and it then remains to ﬁnd the order of p. One veriﬁes that p
6
is the ideal
generated by
(˛1)
3
˛÷2
, and so it remains to show that p
2
and p
3
are nonprincipal.
Suppose p
3
=(,). Then ,
2
=˙˛
n
(˛1)
3
˛÷2
for some m and choice of signs. But this
says that at least one of the numbers
˛1
˛÷2
, ÷
˛1
˛÷2
, ˛
˛1
˛÷2
, ÷˛
˛1
˛÷2
is a square. Let ˇ be that
number. If q is a prime ideal such that ˇ ÷ O
q
(i.e., such that ord
q
(ˇ) _ 0), then we can
look at ˇ mod q and ask if it is a square.
We ﬁrst work modulo 29. We have
X
3
÷10X ÷1 ÷(X ÷5)(X ÷3)(X ÷2) mod 29.
Take q to be the ideal (29. ˛ ÷2). The residue ﬁeld O
1
¡q is F
29
=Z¡(29), and the map
ZŒ˛ ÷F
29
is ˛ ÷2 (mod 29). Thus
˛ ÷1 ÷1. ˛ ÷2 ÷4. (˛ ÷2)
1
÷22. ÷1 ÷÷1.
87
5. THE UNIT THEOREM
The numbers 1, 4, and ÷1 ÷12
2
are squares modulo 29, but 2 is not; hence m must be 0.
Since
˛1
˛÷2
<0 it can’t be a square in 1 (since it isn’t even in 1), and so the only possibility
for ˇ is ÷
˛1
˛÷2
. We eliminate this by looking mod 7.
Take q =(7. ˛ ÷3) (see 3.48). Then in the map ZŒ˛ ÷ZŒ˛¡q =F
T
.
˛ ÷÷3 =4. ÷
˛ ÷1
˛ ÷2
÷
÷3
6
÷÷
1
2
÷÷4 ÷3 mod 7.
and 3 is not a square modulo 7. Thus ÷
˛1
˛÷2
is not a square in QŒ˛.
Similarly, p
2
=(,) can be shown to be impossible. Thus Cl(O
1
) is a cyclic group of
order 6.
Finding u(1)
As we noted eariler, if QŒ¯
n
 c1, where ¯
n
is a primitive mth root of 1, then c(m)[Œ1: Q.
Thus there are only ﬁnitely many possibilities for m. For each of them, use the test in the
later section on algorithms to determine whether the minimum polynomial ˚
n
for ¯
n
has
a root in 1.
Finding a system of fundamental units
One strategy for ﬁnding units in the general case seems to be to ﬁnd lots of solutions to
equations Nm(˛) = m for m a ﬁxed small number, and then take quotients of solutions.
Note that there can be only ﬁnitely many ideals a with N(a) =m; thus if we have lots of
elements ˛
i
with Nm(˛
i
) =m, then frequently ˛
i
O
1
=˛
}
O
1
, and this implies that ˛
i
and ˛
}
differ by a unit — note that this was the strategy used to prove the unit theorem. See
Pohst and Zassenhaus 1989, Chapter 5.
Regulators
There is one other important invariant that we should deﬁne. Let t = r ÷s ÷1, and let
u
1
. .... u
t
be a system of fundamental units. Then the vectors
1(u
i
)
df
=(log[o
1
u
i
[. .... log[o
i
u
i
[. 2 log[o
i÷1
u
i
[. . . . . 2log[o
t
u
i
[) ÷ 1
t
generate the lattice 1(U) in 1
t
. The regulator is deﬁned to be determinant of the matrix
whose i th row is 1(u
i
). Thus, up to sign, the regulator is the volume of a fundamental
domain for 1(U) (regarded as a full lattice in 1
t
).
2
The regulator plays the same role for the group of units (mod torsion) that the dis
criminant plays for O
1
. One can similarly deﬁne the regulator of any set {c
1
. .... c
t
] of
independent units, and the index of the group generated by the c
i
and u(1) in the full
group of units is measured by ratio
[Reg(c
1
. . . . . c
t
)[¡[Reg(U)[.
2
Kwangho Choiy writes: in the deﬁnition of regulators, I think that 1(u
i
) may have to be more precise,
i.e., we can make sure about the index of o. But the deﬁnition in the notes is still correct.
88
Exercises
There are lower bounds for the regulator (see Pohst and Zassenhaus 1989, p 365) similar
to the one we proved for a cubic ﬁeld with one real embedding.
For an algorithm that computes the class group, regulator, and fundamental units of a
general number ﬁeld, but which requires the generalized Riemann hypothesis to prove its
correctness, see Cohen 1993, Algorithm 6.5.9.
NOTES To ﬁnd the units in QŒ
J, J >0, one has to solve certain diophantine equations (see 5.3),
whose study has a long history. Theorem 5.1 was proved by Dirichlet (1840, 1846)
3
only for rings
of the form ZŒ˛ because, at the time, a deﬁnition of O
1
was lacking. However, his proof extends
easily to O
1
(and to O
1
(S)).
Exercises
51 Fix an m and and M. Is it necessarily true that the set of algebraic integers ˛ in C of
degree <m and with [˛[ <M is ﬁnite? [Either prove, or give a counterexample.]
52 Find a fundamental unit for the ﬁeld QŒ
67.
53 Let ˛ be an element of a number ﬁeld 1. Does Nm
1¡Q
=˙1 imply that ˛ is unit in
O
1
. [Either prove, or give a counterexample.]
3
Dirichlet, P. G. Lejeune, Sur la th´ eorie des nombres, C. R. Acad. Sci. Paris 10 (1840), 285–288. Dirichlet,
P. G. Lejeune, Zur Theorie der complexen Einheiten. Verhandl. Preuss. Akad. Wiss. (1846), 103–107.
89
CHAPTER 6
Cyclotomic Extensions; Fermat’s
Last Theorem.
The cyclotomic
1
extensions of Q are those generated by a root of 1. They provide interest
ing examples of the theory we have developed, but, more signiﬁcantly, they have important
applications, for example, to Fermat’s last theorem and to the existence of reciprocity laws
(more generally, to class ﬁeld theory itself).
The basic results
An element ¯ of a ﬁeld 1 is said to be a primitive nth root of 1 if ¯
n
=1 but ¯
d
=1 for any
J <n, i.e., if ¯ is an element of order n in 1
×
. For example, the nth roots of 1 in C are the
numbers e
2tin¡n
, 0 _ m _ n÷1, and the next lemma shows that e
2tin¡n
is a primitive
nth root of 1 if and only if m is relatively prime to n.
LEMMA 6.1 Let ¯ be a primitive nth root of 1. Then ¯
n
is again a primitive nth root of 1
if and only if m is relatively prime to n.
PROOF. This is a consequence of a more general fact: if ˛ is an element of order n in a
group, then ˛
n
is also of order n if and only if m is relatively prime to n. Here is the proof.
If J[m. n, then (˛
n
)
n
d
=˛
n
m
d
=1. Conversely, if m and n are relatively prime, then there
are integers a and b such that
am÷bn =1.
Now ˛
on
=˛ and so (˛
n
)
d
=1 =˛
d
=(˛
on
)
d
=1 =n[J.
2
Let 1 = QŒ¯, where ¯ is a primitive nth root of 1. Then 1 is the splitting ﬁeld of
X
n
÷1, and so it is Galois over Q. Let G =Gal(QŒ¯¡Q). It permutes the set of primitive
nth roots of 1 in 1, and so, for any o ÷ G, o¯ =¯
n
for some integer m relatively prime to
n; moreover, m is welldeﬁned modulo n. The map o ÷Œm is an injective homomorphism
G ÷(Z¡nZ)
×
. In FT, Proposition 5.7, it is proved that this map is an isomorphism, and
so Œ1 : Q =c(n)
def
=#(Z¡nZ)
×
. We shall give another proof, and at the same time obtain
many results concerning the arithmetic of QŒ¯.
1
The name cyclotomic (circledividing) derives from the fact that the nth roots of 1 are space evenly around
the unit circle.
90
The basic results
The cyclotomic polynomial ˚
n
is deﬁned to be,
˚
n
(X) =
¸
(X ÷¯
n
)
where the product runs over a set of representatives m for the elements of (Z¡nZ)
×
, for
example, over the integers m, 0 _m_n÷1, relatively prime to n. Equivalently,
˚
n
(X) =
¸
(X ÷¯
t
)
where ¯
t
runs over the primitive nth roots of 1. Because G permutes the ¯
t
, ˚
n
(X) ÷ QŒX,
and clearly ˚
n
(¯) =0. Therefore, ˚
n
(X) is the minimum polynomial of ¯ if and only if it
is irreducible, in which case Œ1 : Q =c(n) and the map G ÷(Z¡nZ)
×
is an isomorphism.
Hence the following statements are equivalent:
(a) the map Gal(QŒ¯¡Q)
//
(Z¡nZ)
×
is an isomorphism;
(b) ŒQŒ¯ : Q =c(n):
(c) Gal(QŒ¯¡Q) acts transitively on the set of primitive nth roots of 1 (i.e., they are
conjugates);
(d) ˚
n
(X) is irreducible (and so ˚
n
(X) is the minimum polynomial of ¯).
We shall see that all these statements are true.
Note that each nth root of 1 is a primitive Jth root of 1 for exactly one J[n, and so
X
n
÷1 =
¸
dn
˚
d
(X) =(X ÷1) ˚
n
(X).
To ﬁnd the nth cyclotomic polynomial, type “polcyclo(n,X)” in PARI. For example,
˚
3
(X) =X
2
÷X ÷1
˚
4
(X) =X
2
÷1
˚
6
(X) =X
2
÷X ÷1
˚
12
(X) =X
4
÷X
2
÷1
and
X
12
÷1 =(X ÷1)(X ÷1)(X
2
÷X ÷1)(X
2
÷1)(X
2
÷X ÷1)(X
4
÷X
2
÷1).
We ﬁrst examine a cyclotomic extension in the case that n is a power ]
i
of a prime.
PROPOSITION 6.2 Let ¯ be a primitive ]
i
th root of 1, and let 1 =QŒ¯.
(a) The ﬁeld QŒ¯ is of degree c(]
i
) =]
i1
(]÷1) over Q.
(b) The ring of integers in QŒ¯ is ZŒ¯.
(c) The element ¬
def
=1÷¯ is a prime element of O
1
, and (]) =(¬)
e
with e =c(]
i
).
(d) The discriminant of O
1
over Z is ˙]
c
, some c (in fact, c = ]
i1
(]r ÷r ÷1));
therefore, ] is the only prime to ramify in QŒ¯.
PROOF. Because ¯ is integral over Z, the ring ZŒ¯ is contained in O
1
.
If ¯
t
is another primitive ]
i
th root of 1, then ¯
t
= ¯
x
and ¯ = ¯
tt
for some integers s
and t not divisible by ], and so ZŒ¯
t
 =ZŒ¯ and QŒ¯
t
 =QŒ¯. Moreover,
1÷¯
t
1÷¯
=1÷¯ ÷ ÷¯
x1
÷ ZŒ¯.
91
6. CYCLOTOMIC EXTENSIONS; FERMAT’S LAST THEOREM.
Similarly, (1 ÷¯)¡(1 ÷¯
t
) ÷ ZŒ¯, and so (1 ÷¯
t
)¡(1 ÷¯) is a unit in ZŒ¯ (hence also in
O
1
). Note that
˚
;
r (X) =
X
;
r
÷1
X
;
r1
÷1
=
t
;
÷1
t ÷1
=1÷t ÷ ÷t
;1
. t =X
;
r1
.
and so
˚
;
r (1) =].
For its deﬁnition, we see that
˚
;
r (1) =
¸
(1÷¯
t
) =
¸
1÷¯
t
1÷¯
(1÷¯) =u (1÷¯)
c(;
r
)
.
with u a unit in ZŒ¯. Therefore we have an equality of ideals in O
1
,
(]) =(¬)
e
. ¬
def
=1÷¯. e =c(]
i
). (14)
and so (]) has at least c(]
i
) prime factors in O
1
. Now (3.34) implies that ŒQŒ¯ : Q _
c(]
i
). This proves (a) of the Proposition since we know ŒQŒ¯ : Q _c(]
i
).
Moreover we see that ¬ must generate a prime ideal in O
1
, otherwise, again, (]) would
have too many primeideal factors. This completes the proof of (c).
For future reference, we note that, in O
1
,
(]) =p
c(;
r
)
. p =(¬). }(p¡]) =1.
The last equality means that the map Z¡(]) ÷O
1
¡(¬) is an isomorphism.
We next show that (up to sign) disc(ZŒ¯¡Z) is a power of ]. Since
disc(O
1
¡Z) (O
1
: ZŒ¯)
2
=disc(ZŒ¯¡Z).
this will imply:
(i) disc(O
1
¡Z) is a power of ];
(ii) (O
1
: ZŒ¯) is a power of ], and therefore ]
Æ
O
1
cZŒ¯ for some M.
To compute disc(ZŒ¯¡Z), we shall use the formula in (2.34), which in our case reads:
disc(ZŒ¯¡Z) =˙Nm
1¡Q
(˚
t
;
r (¯)).
On differentiating the equation
(X
;
r1
÷1) ˚
;
r (X) =X
;
r
÷1
and substituting ¯ for X, we ﬁnd that ˚
t
;
r
(¯) =]
i
¯
;
r
1
¡(¯
;
r1
÷1). Clearly
Nm
1¡Q
¯ =˙1. Nm
1¡Q
]
i
=(]
i
)
c(;
r
)
=]
ic(;
r
)
.
We shall show that
Nm
1¡Q
(1÷¯
;
s
) =˙]
;
s
. 0 _s <r.
and so
Nm
1¡Q
˚
t
;
r (¯) =˙]
c
. c =r(]÷1)]
i1
÷]
i1
=]
i1
(]r ÷r ÷1).
92
The basic results
First we compute Nm
1¡Q
(1 ÷¯). The minimum polynomial of 1 ÷¯ is ˚
;
r (1 ÷X),
which has constant term ˚
;
r (1) =], and so Nm
1¡Q
(1÷¯) =˙].
We next compute Nm
1¡Q
(1÷¯
;
s
) some s <r. Because ¯
;
s
is a primitive ]
ix
th root
of 1, the computation just made (with r replaced by r ÷s) shows that
Nm
QŒc
p
s
]¡Q
(1÷¯
;
s
) =˙].
Using that
Nm
Æ¡1
=Nm
1¡1
ıNm
Æ¡1
and Nm
Æ¡1
˛ =˛
ŒÆ:1]
if ˛ ÷ 1.
we see that
Nm
1¡Q
(1÷¯
;
s
) =˙]
o
where a =ŒQŒ¯ : QŒ¯
;
s
 =c(]
i
)¡c(]
ix
) =]
x
.
This completes the proof of (d).
We are now ready to prove (b). As we observed above the inclusion Z ·÷O
1
induces
an isomorphism Z¡(]) ÷O
1
¡(¬). In other words,
O
1
=Z÷¬O
1
.
and so, certainly,
O
1
=ZŒ¯ ÷¬O
1
.
On multiplying through by ¬, we obtain the equality
¬O
1
=¬ZŒ¯ ÷¬
2
O
1
.
Therefore,
O
1
=ZŒ¯ ÷¬ZŒ¯ ÷¬
2
O
1
=ZŒ¯ ÷¬
2
O
1
.
On repeating this argument, we ﬁnd that
O
1
=ZŒ¯ ÷¬
n
O
1
for all m_1. Since ¬
c(;
r
)
=]×(unit), this implies that
O
1
=ZŒ¯ ÷]
n
O
1
for all m _ 1. But for m large enough, we know that ]
n
O
1
c ZŒ¯, and so ZŒ¯ = O
1
.
This completes the proof of (b).
2
REMARK 6.3 (a) The sign of the disc(QŒ¯¡Q), ¯ any root of 1, can be computed most
easily by using (2.40a). Clearly QŒ¯ has no real embeddings unless ¯ =˙1 (and QŒ¯ =Q),
and so, except for this case,
sign(disc(QŒ¯¡Q)) =(÷1)
x
. s =ŒQŒ¯ : Q¡2.
If ¯ is a primitive ]
i
th root of 1, ]
i
>2, then
ŒQŒ¯ : Q¡2 =(]÷1)]
i1
¡2
93
6. CYCLOTOMIC EXTENSIONS; FERMAT’S LAST THEOREM.
which is odd if and only if ]
i
=4 or ] ÷3 mod 4.
(b) Let ¯ and ¯
t
be primitive ]
i
th and q
x
th roots of 1. If ] and q are distinct primes,
then
QŒ¯ ¨QŒ¯
t
 =Q.
because if 1 cQŒ¯, then ] ramiﬁes totally in 1 and q does not, and if 1 cQŒ¯
t
, then q
ramiﬁes totally in 1 and ] does not, and these are contradictory unless 1 =Q.
THEOREM 6.4 Let ¯ be a primitive nth root of 1.
(a) The ﬁeld QŒ¯ is of degree c(n) over Q.
(b) The ring of integers in QŒ¯ is ZŒ¯, and so 1. ¯. . . . . ¯
c(n)1
is an integral basis for
O
QŒc]
over Z.
(c) If ] ramiﬁes in QŒ¯ then ][n; more precisely, if n =]
i
m with m relatively prime
to ], then
(]) =(P
1
P
x
)
c(;
r
)
in QŒ¯ with the P
i
distinct primes in QŒ¯.
PROOF. We use induction on the number of primes dividing n. Write n =]
i
mwith mnot
divisible by ]. We may assume the theorem for m. Note that ¯
;
r
def
=¯
n
is a primitive ]
i
th
root of 1, ¯
n
=¯
;
r
is a primitive mth root of 1, and that QŒ¯ =QŒ¯
;
r  QŒ¯
n
. Consider
the ﬁelds:
QŒ¯
;
r 
QŒ¯
QŒ¯
QŒ¯
n

?
?
?
?
?
?
?
?
QŒ¯
;
r 
Q
?
?
?
?
?
?
?
?
Q
QŒ¯
n

p
c(;
r
)
(pO)
c(;
r
)
=
¸
p
i
O
(pO)
c(;
r
)
=
¸
p
i
O
¸
p
i
?
?
?
?
?
?
?
?
p
c(;
r
)
(])
?
?
?
?
?
?
?
?
(])
¸
p
i
According to Proposition 6.2, (]) ramiﬁes totally in QŒ¯
;
r , say (]) = p
c(;
r
)
, but is un
ramiﬁed in QŒ¯
n
, say (]) =p
1
p
x
with the p
i
distinct primes. Because QŒ¯ is obtained
from QŒ¯
n
 by adjoining ¯
;
r , its degree over QŒ¯
n
 is at most c(]
i
). It follows from
Theorem 3.34 that p
1
p
x
can become a c(]
i
)th power in QŒ¯ only if ŒQŒ¯: QŒ¯
n
 =
c(]
i
) and each prime p
i
ramiﬁes totally in QŒ¯, say p
i
O
QŒc]
= P
c(;
r
)
i
. Therefore,
ŒQŒ¯ : Q = c(]
i
) c(m) = c(n), and to complete the proof, it remains to show that
O
QŒc]
= ZŒ¯
;
r . ¯
n
 = ZŒ¯. This is accomplished by the next lemma, because the only
primes that can divide the discriminant of O
QŒc
m
]
¡Z are the divisors of m (induction hy
pothesis and 3.35).
2
LEMMA 6.5 Let 1 and 1 be ﬁnite extensions of Q such that
Œ11: Q =Œ1: Q Œ1: Q.
and let J be the greatest common divisor of disc(O
1
¡Z) and disc(O
1
¡Z)). Then
O
1·1
cJ
1
O
1
O
1
.
94
The basic results
PROOF. Let {˛
1
. .... ˛
n
] and {ˇ
1
. .... ˇ
n
] be integral bases for 1 and 1 respectively. Then
˛
i
ˇ
}
is a basis for 1 1 over Q. Thus every , ÷ O
1·1
can be written in the form
, =
¸
i}
a
i}
r
˛
i
ˇ
}
. a
i}
. r ÷ Z.
with
o
ij
i
uniquely determined. After dividing out any common factors from top and bottom,
no prime factor of r will divide all the a
i}
, and we then have to show that r[J.
When we identify 1 with a subﬁeld of C, every embedding o of 1 into C will extend
uniquely to an embedding of 1 1 into C ﬁxing the elements of 1. To see this, write
1 =QŒ˛; then 1 1=1Œ˛, and the hypothesis on the degrees implies that the minimum
polynomial of ˛ doesn’t change when we pass from Q to 1; there is therefore a unique
1homomorphism 1Œ˛ ÷C sending ˛ to o˛.
On applying such a o to the above equation, we obtain an equation
o(,) =
¸
i}
a
i}
r
o(˛
i
)ˇ
}
.
Write .
i
=
¸
}
(a
i}
¡r)ˇ
}
, and let o
1
. o
2
. .... o
n
be the distinct embeddings of 1 into C. We
obtain a system of m linear equations
¸
i
o
k
(˛
i
).
i
=o
k
(,). k =1. 2. .... m.
and Cramer’s rule tells us that
D.
i
=D
i
where D =det(o
}
(˛
i
)) and D
i
is a similar determinant. According to (2.26), D
2
=z
def
=
disc(O
1
¡Z), and so
z .
i
=DD
i
.
By construction, both D and D
i
are algebraic integers, and so z .
i
is an algebraic integer.
But z.
i
=
¸
zo
ij
i
ˇ
}
, and the ˇ
}
s form an integral basis for O
1
, and so
zo
ij
i
÷ Z. Hence
r[za
i}
all i. ; , and, because of our assumption on r and the a
i}
s, this implies that r[z.
Similarly, r[ disc(O
1
¡Z), and so r divides the greatest common divisor of disc(O
1
¡Z)
and disc(O
1
¡Z).
2
REMARK 6.6 (a) Statement (c) of the theorem shows that if ] divides n, then ] ramiﬁes
unless c(]
i
) = 1. Since c(]
i
) = ]
i1
(] ÷1), this happens only if ]
i
= 2. Thus, if ]
divides n, then ] ramiﬁes in QŒ¯
n
 except when ] =2 and n =2 (odd number).
(b) Let m be an integer > 1; then c(mn) > c(n) except when n is odd and m = 2.
Therefore u(QŒ¯
n
) is cyclic of order n (generated by ¯
n
) except when n is odd, in which
case it is cyclic of order 2n (generated by ÷¯
n
).
(c) In the situation of the lemma,
disc(11¡Q) =disc(1¡Q)
Œ1:Q]
disc(1¡Q)
Œ1:Q]
. (15)
provided O
11
=O
1
O
1
. This can be proved by an elementary determinant calculation.
Using this, one can show that, for ¯
n
a primitive nth root of 1,
disc(QŒ¯
n
¡Q) =(÷1)
c(n)¡2
n
c(n)
¡
¸
;n
]
c(n)¡(;1)
.
95
6. CYCLOTOMIC EXTENSIONS; FERMAT’S LAST THEOREM.
The example
QŒi.
5 =QŒi  QŒ
÷5
shows that the condition on the rings of integers is necessary for (15) to hold, because the
extensions QŒi  and QŒ
÷5 have discriminants 4 and 20 respectively, but QŒi.
5 has
discriminant 4
2
5
2
=4
2
20
2
¡4
2
.
Class numbers of cyclotomic ﬁelds
Let ¯ be a primitive ]th root of 1, ] an odd prime. It is known that the class number of QŒ¯
grows quite rapidly with ], and that in fact the class number is 1 if and only if ] _19.
Here is how to prove that QŒ¯ has class number > 1 when ] =23. The Galois group
of QŒ¯ over Qis cyclic of order 22, and therefore has a unique subgroup of index 2. Hence
QŒ¯ contains a unique quadratic extension 1 of Q. Since 23 is the only prime ramifying in
QŒ¯, it must also be the only prime ramifying in 1, and this implies that 1 =QŒ
÷23.
One checks that (2) splits in QŒ
÷23, say (2) =pq, that p is not principal, and that p
3
is
principal. Let P be a prime ideal of ZŒ¯ lying over p. Then NP = p
(
, where } is the
residue class degree. Since } divides ŒQŒ¯ : QŒ
÷23 =11, we see that } =1 or 11 (in
fact, } =11). In either case, p
(
is not principal, and this implies that P is not principal,
because the norm of a principal ideal is principal.
Because of the connection to Fermat’s last theorem, primes ] such that ] does not
divide the class number of QŒ¯ are of particular interest. They are called regular. Kummer
found a simple test for when a prime is regular: deﬁne the Bernoulli numbers T
n
by the
formula
t
e
t
÷1
=
o
¸
n=0
T
n
t
n
nŠ
. T
n
÷ Q:
then ] is not regular if and only if ] divides the numerator of some T
k
with k =2. 4. . . . . ]÷
3. It has long been known that there are inﬁnitely many irregular primes, but it is still not
proved that there are inﬁnitely many regular primes. It is expected that 61% of primes are
regular and 39% are irregular.
Units in cyclotomic ﬁelds
Let ¯ be a primitive nth root of 1, n >2. Deﬁne
QŒ¯
÷
=QŒ¯ ÷¯
1
.
For example, if ¯ =e
2ti¡n
, then QŒ¯
÷
=QŒcos
2t
n
. Under any embedding of QŒ¯ into
C, ¯
1
maps to the complex conjugate of ¯, and therefore the image of QŒ¯
÷
is ﬁxed under
complex conjugation and hence lies in 1. Thus, we see that QŒ¯ is a CMﬁeld with maximal
totally real subﬁeld QŒ¯
÷
. According to Proposition 5.12, the index of u(QŒ¯) U
QŒc]
C in
U
QŒc]
is 1 or 2. In fact, when n is a prime power, it must be 1.
PROPOSITION 6.7 Assume that n is a prime power; then every unit u ÷ QŒ¯ can be written
u =¯ ·
with ¯ a root of unity and · a unit in QŒ¯
÷
.
96
The ﬁrst case of Fermat’s last theorem for regular primes
PROOF. We prove this only for powers of odd primes (which is all we shall need in the next
section). If the statement is false, then the homomorphism
u ÷u¡ ¨ u: U
QŒc]
÷u¡u
2
. u =u(QŒ¯).
in the proof of Proposition (5.12) is surjective, and so there exists a unit u of QŒ¯ such that
¨ u = ¯
t
u where ¯
t
is a root of 1 that is not a square. Recall (6.6b) that, because n is odd,
u ={˙1] '¯), and so u
2
='¯). Therefore ¯
t
=÷¯
n
for some integer m. Let
u =a
0
÷ ÷a
c(n)1
¯
c(n)1
, a
i
÷ Z.
Then ¨ u =a
0
÷ ÷a
c(n)1
¨
¯
c(n)1
, and modulo the prime ideal p =(1÷¯) =(1÷
¨
¯) of
O
QŒc]
.
u ÷a
0
÷ ÷a
c(n)1
÷ ¨ u.
Thus
u ÷÷¯
n
u ÷÷u mod p.
and so 2u ÷ p. This is a contradiction because p is prime, 2 … p, and u … p.
2
The ﬁrst case of Fermat’s last theorem for regular primes
Kummer proved Fermat’s last theorem for regular primes. Here we prove a weaker result,
known as the ﬁrst case of Fermat’s last theorem.
THEOREM 6.8 Let ] be an odd prime. If the class number of QŒ¯ is not divisible by ],
then there does not exist an integer solution (.. .. :) to
X
;
÷Y
;
=7
;
with ] relatively prime to ..:.
We show that existence of integers .. .. : with .
;
÷.
;
= :
;
and ] [ ..: leads to a
contradiction. After removing any common factor, we may suppose that gcd(.. .. :) =1.
We ﬁrst treat the case ] =3. The only cubes modulo 9 are ÷1, 0, 1, and so
.
3
÷.
3
÷÷2. 0, or 2 mod 9.
:
3
÷÷1 or 1 mod 9.
which are contradictory. Similarly we may eliminate the case ] =5 by looking modulo 25.
Henceforth we assume ] >5.
If . ÷. ÷÷: mod ], then ÷2:
;
÷:
;
and ][3:, contradicting our hypotheses. Hence
one of the congruences can’t hold, and after rewriting the equation .
;
÷(÷:)
;
=(÷.)
;
if
necessary, we may assume that ] [ . ÷..
The roots of X
;
÷1 are ÷1. ÷¯. . . . . ÷¯
;1
, and so
X
;
÷1 =
¸
;1
i=0
(X ÷¯
i
).
Hence
¸
;1
i=0
(. ÷¯
i
.) =:
;
.
97
6. CYCLOTOMIC EXTENSIONS; FERMAT’S LAST THEOREM.
The idea of the proof is to exploit this factorization and what we know of the arithmetic of
QŒ¯ to obtain a contradiction.
Let p be the unique prime ideal of ZŒ¯ dividing (]); thus p =(1÷¯
i
) where i can be
any integer such that 1 _i _]÷1 (see 6.2).
LEMMA 6.9 The elements . ÷¯
i
. of ZŒ¯ are relatively prime in pairs.
PROOF. We have to show that there does not exist a prime ideal q dividing . ÷¯
i
. and
.÷¯
}
. for i =; . Suppose there does. Then q[((¯
i
÷¯
}
).) =p., and q[((¯
}
÷¯
i
).) =p..
By assumption, . and . are relatively prime, and therefore q =p. Thus .÷. ÷.÷¯
i
. ÷0
mod p. Hence . ÷. ÷ p¨Z =(]). But :
;
=.
;
÷.
;
÷. ÷. ÷0 mod ], and so ][:,
which contradicts our hypotheses.
2
LEMMA 6.10 For every ˛ ÷ ZŒ¯, ˛
;
÷ Z÷]ZŒ¯.
PROOF. Write
˛ =a
0
÷a
1
¯ ÷ ÷a
;2
¯
;2
. a
i
÷ Z.
Then
˛
;
÷a
;
0
÷a
;
1
÷ ÷a
;
;1
mod ].
which lies in Z.
2
LEMMA 6.11 Let ˛ =a
0
÷a
1
¯ ÷ ÷a
;1
¯
;1
with a
i
÷ Z and at least one a
i
=0. If
˛ is divisible by an integer n, i.e., if ˛ ÷ nZŒ¯, then each a
i
is divisible by n.
PROOF. Since 1 ÷¯ ÷ ÷¯
;1
= 0, any subset of {1. ¯. . . . . ¯
;1
] with ] ÷1 elements
will be a Zbasis for ZŒ¯. The result is now obvious.
2
We can now complete the proof of Theorem 6.8. Regard the equation
¸
;1
i=0
(. ÷¯
i
.) =(:)
;
as an equality of ideals in ZŒ¯. Since the factors on the left are relatively prime in pairs,
each one must be the ]th power of an ideal, say
(. ÷¯
i
.) =a
;
i
for some ideal a
i
in ZŒ¯. This equation implies that a
i
has order dividing ] in the class
group, but we are assuming that the class group of ZŒ¯ is of order prime to ], and so a
i
itself is principal, say a
i
=(˛
i
).
Take i =1, and omit the subscript on ˛
1
. Then we have that . ÷¯. =u˛
;
for some
unit u in ZŒ¯. We apply (6.7) to write u =¯
i
· where ¨ · =·. According to (6.10), there is
an a ÷ Z such that ˛
;
÷a mod ]. Therefore
. ÷¯. =¯
i
·˛
;
÷¯
i
·a mod ].
Also
. ÷
¨
¯. =¯
i
· ¨ ˛
;
÷¯
i
·a mod ].
98
Exercises
On combining these statements, we ﬁnd that
¯
i
(. ÷¯.) ÷¯
i
(. ÷¯
1
.) mod ].
or
. ÷¯. ÷¯
2i
. ÷¯
2i1
. ÷0 mod ]. (16)
If 1. ¯. ¯
2i1
. ¯
2i
are distinct, then, because ] _ 5, Lemma 6.11 implies that ] divides .
and ., which is contrary to our original assumption. The only remaining possibilities are:
(a) 1 =¯
2i
; but then (*) says
¯. ÷¯
1
. ÷0 mod ].
and Lemma 6.11 implies ][., which contradicts our original assumption.
(b) 1 =¯
2i1
; then ¯ =¯
2i
, and (*) says
(. ÷.) ÷(. ÷.)¯ ÷0 mod ].
and Lemma 6.11 implies that ][. ÷., which contradicts the choice of . and . made
at the start of the proof.
(c) ¯ =¯
2i1
; but then (*) says
. ÷¯
2
. ÷0 mod ].
and Lemma 6.11 implies that ][., which contradicts our original assumption.
This completes the proof.
NOTES Everything in this section was known to Kummer, but in terms of “ideal numbers” rather
than ideals. The methods of this section have not (so far) sufﬁced to prove Fermat’s last theorem
but, as the reader may already be aware, other methods have.
Exercises
61 Show that X
3
÷3X ÷1 is an irreducible polynomial in QŒX with three real roots.
Let ˛ be one of them, and let 1 =QŒ˛. Compute disc(ZŒ˛¡Z), and deduce that
O
1
¯ZŒ˛ ¯3
n
O
1
for some m. Showthat ˛ and ˛÷2 are units in ZŒ˛ and O
1
, and that (˛÷1)
3
=3˛(˛÷2).
Deduce that (˛ ÷1) is a prime ideal in O
1
, and show that O
1
=ZŒ˛ ÷(˛ ÷1)O
1
. Use
this to show that O
1
=ZŒ˛. Show that (2) is a prime ideal in O
1
, and deduce that O
1
is
a principal ideal domain.
62 Show that the ring of integers in QŒcos
2t
n
 is ZŒ2cos
2t
n
.
99
CHAPTER 7
Valuations; Local Fields
In this section, we deﬁne the notion of a valuation and study the completions of number
ﬁelds with respect to valuations.
Valuations
A (multiplicative) valuation on a ﬁeld 1 is a function . ÷[.[: 1 ÷1 such that
(a) [.[ >0 except that [0[ =0:
(b) [..[ =[.[[.[
(c) [. ÷.[ _[.[ ÷[.[ (triangle inequality).
If the stronger condition
(c
t
) [. ÷.[ _max{[.[. [.[]
holds, then [ [ is called a nonarchimedean valuation.
Note that (a) and (b) imply that [ [ is a homomorphism1
×
÷1
>0
(multiplicative group
of positive real numbers). Since 1
>0
is torsionfree, [ [ maps all roots of unity in 1
×
to 1.
In particular, [ ÷1[ =1, and [ ÷.[ =[.[ for all ..
EXAMPLE 7.1 (a) For any number ﬁeld 1, and embedding o: 1 ·÷C, we get a valuation
on 1 by putting [a[ =[oa[.
(b) Let ord: 1
×
÷Z be an (additive) discrete valuation, and let e be a real number with
e >1; then
[a[ =(1¡e)
ord(o)
. a =0. [0[ =0
is a nonarchimedean valuation on 1. For example, for any prime number ], we have the
]adic valuation [ [
;
on Q:
[a[
;
=(1¡e)
ord
p
(o)
.
Usually we normalize this by taking e =]; thus
[a[
;
=(1¡])
ord
p
(o)
=1¡]
i
if a =a
0
]
i
with ord
;
(a
0
) =0.
Similarly, for any prime ideal p in a number ﬁeld 1, we have a normalized padic valuation
[a[
p
=(1¡Np)
ord
p
(o)
.
(c) On any ﬁeld we can deﬁne the trivial valuation: [a[ =1 for all a =0. When 1 is
ﬁnite, there is no other (because all nonzero elements of a ﬁnite ﬁeld are roots of 1).
100
Nonarchimedean valuations
Nonarchimedean valuations
Recall that this means that, instead of the triangle inequality, we have
[. ÷.[ _max{[.[. [.[].
By induction, this condition implies that
[
¸
.
i
[ _max{[.
i
[]. (17)
PROPOSITION 7.2 A valuation [ [ is nonarchimedean if and only if it takes bounded values
on {m1 [ m÷ Z].
PROOF. If [ [ is nonarchimedean, then, for m>0.
[m1[ =[1÷1÷ ÷1[ _[1[ =1.
As we noted above, [ ÷1[ =[1[, and so [ ÷m1[ =[m1[ _1.
Conversely, suppose [m1[ _N for all m. Then
[. ÷.[
n
=[
¸
n
i
.
i
.
ni
[ _
¸
i
[
n
i
[ [.[
i
[.[
ni
.
Clearly [.[
i
[.[
ni
_max{[.[
n
. [.[
n
] =max{[.[. [.[]
n
and
n
i
is an integer, and so
[. ÷.[
n
_N(n÷1)max{[.[. [.[]
n
.
On taking nth roots we ﬁnd that
[. ÷.[ _N
1¡n
(n÷1)
1¡n
max{[.[. [.[].
When we let n ÷o, the terms involving n tend to 1 (to see this, take logs).
2
COROLLARY 7.3 If char 1 =0, then 1 has only nonarchimedean valuations.
PROOF. In this case, the set {m 1 [ m÷ Z] is ﬁnite.
2
ASIDE 7.4 Archimedes stated that for any two line segments, laying the shorter segment endtoend
a sufﬁcient ﬁnite number of times will create a segment longer than the other. In other words, for
any two nonzero real numbers a and b, there is an n ÷ N such that [b[ <[na[. The proposition shows
that the nonarchimedean valuations are exactly those that don’t have this “archimedean property”.
As we noted above, a discrete (additive) valuation ord on 1 determines a valuation by
[.[ =e
ord(x)
.
any e >1. Taking logs gives log
e
[.[ =÷ord(.), or ord(.) =÷log
e
[.[. This suggests how
we might pass from multiplicative valuations to additive valuations.
PROPOSITION 7.5 Let [ [ be a nontrivial nonarchimedean valuation, and put ·(.) =÷log[.[,
. =0 (log to base e for any real e >1). Then ·: 1
×
÷1 satisﬁes the following conditions:
101
7. VALUATIONS; LOCAL FIELDS
(a) ·(..) =·(.) ÷·(.):
(b) ·(. ÷.) _min{·(.). ·(.)].
If ·(1
×
) is discrete in 1, then it is a multiple of a discrete valuation ord: 1
×
Z c1.
PROOF. That · satisﬁes (a) and (b) is obvious. For the last statement, note that ·(1
×
)
is a subgroup of 1 (under addition). If it is a discrete subgroup, then it is a lattice (by
4.15), which means that ·(1
×
) =Zc for some c. Now ord
def
=c
1
· is an additive discrete
valuation 1
×
Z.
2
We shall say [ [ is discrete when [1
×
[ is a discrete subgroup of 1
>0
. Note that, even
when [1
×
[ is discrete in 1, [1[ usually won’t be, because 0 will be a limit point for the set
[1
×
[. For example, []
n
[
;
=]
n
, which converges to 0 as n ÷o.
PROPOSITION 7.6 Let [ [ be a nonarchimedean valuation. Then
·
def
={a ÷ 1 [ [a[ _1] is a subring of 1, with
U
def
={a ÷ 1 [ [a[ =1] as its group of units, and
m
def
={a ÷ 1 [ [a[ <1] as its unique maximal ideal.
The valuation [ [ is discrete if and only if m is principal, in which case · is a discrete
valuation ring.
PROOF. The ﬁrst assertion is obvious. If [ [ is discrete, then · and mare the pair associated
(as in 3.27) with the additive valuation ÷log[ [, and so · is a discrete valuation ring and m
is generated by any element ¬ ÷ 1
×
such that [¬[ is the largest element of [1
×
[ less than
one. Conversely, if m=(¬), then [1
×
[ is the subgroup of 1
>0
generated by [¬[.
2
REMARK 7.7 There do exist nondiscrete nonarchimedean valuations. For example, let Q
al
be an algebraic closure of Q. We shall see later that the ]adic valuation [ [
;
: Q÷1extends
to Q
al
(in many different ways). Since Q
al
contains an element ]
1¡n
for all n, we see that
[Q
al×
[ ÷ (]
1
)
1¡n
= 1¡
n
] for all n, and 1¡
n
] ÷1 as n ÷o. In fact, one can show
that [Q
al×
[ ={]
i
[ r ÷ Q], which is not discrete in 1
>0
.
Equivalent valuations
Note that a valuation [ [ deﬁnes a metric on 1, with distance function
J(a. b) =[a÷b[.
and hence a topology on 1: for a ÷ 1, the sets
U(a. c) ={. ÷ 1 [ [. ÷a[ <c]. c > 0.
form a fundamental system of open neighbourhoods of a. A set is open if and only if it is a
union of sets of the form U(a. c).
For example, for the topology on Q deﬁned by [ [
;
, a and b are close if their difference
is divisible by a high power of ]. In particular, the sequence
1. ]. ]
2
. . . . . ]
n
. . . .
converges to 0.
The topology deﬁned by the padic valuation [ [
p
is called the padic topology on 1.
102
Equivalent valuations
PROPOSITION 7.8 Let [ [
1
, [ [
2
be valuations on 1, with [ [
1
nontrivial. The following
conditions are equivalent:
(a) [ [
1
, [ [
2
deﬁne the same topology on 1:
(b) [˛[
1
<1 =[˛[
2
<1:
(c) [ [
2
=[ [
o
1
for some a >0.
PROOF. (a) =(b): Since [˛
n
[ =[˛[
n
, clearly ˛
n
÷0 if and only if [˛[ < 1. Therefore (a)
implies that
[˛[
1
<1 ” [˛[
2
<1.
(b) =(c): Because [ [
1
is nontrivial, there exists a . ÷ 1 such that [.[ >1. Let
a =log[.[
2
¡log[.[
1
.
so that
log[.[
2
=a log[.[
1
.
or
[.[
2
=[.[
o
1
.
Now let . be any nonzero element of 1. There is a real number b such that
[.[
1
=[.[
b
1
.
To prove (c), it sufﬁces to prove that
[.[
2
=[.[
b
2
.
because then
[.[
2
=[.[
b
2
=[.[
ob
1
=[.[
o
1
.
Let m¡n, n >0, be a rational number >b. Then
[.[
1
=[.[
b
1
<[.[
m
n
1
and so
[.
n
¡.
n
[
1
<1.
From our assumption (b), this implies that
[.
n
¡.
n
[
2
<1
and so
[.[
2
<[.[
m
n
2
.
This is true for all rational numbers
n
n
>b, and so
[.[
2
_[.[
b
2
.
A similar argument with rational numbers
n
n
<b shows that
[.[
2
_[.[
b
2
.
and so we have equality, which completes the proof of (a).
2
Two valuations are said to be equivalent if they satisfy the conditions of the proposition.
103
7. VALUATIONS; LOCAL FIELDS
Properties of discrete valuations
We make some easy, but important, observations about discrete valuations.
7.9 For an additive valuation, we are given that
ord(a÷b) _min{ord(a). ord(b)]
and we checked (3.26 et seq.) that this implies that equality holds if ord(a) =ord(b). For
multiplicative valuations, we are given that
[a÷b[ _max{[a[. [b[].
and a similar argument shows that equality holds if [a[ =[b[. This has the following con
sequences.
7.10 Recall that we deﬁne a metric on 1 by setting J(a. b) =[a ÷b[. I claim that if . is
closer to b than it is to a, then J(a. .) =J(a. b). For we are given that
[. ÷b[ <[. ÷a[.
and this implies that
[b ÷a[ =[b ÷. ÷. ÷a[ =[. ÷a[.
7.11 Suppose
a
1
÷a
2
÷ ÷a
n
=0.
Then an argument as on p60 shows that the maximum value of the summands must be
attained for at least two values of the subscript.
Complete list of valuations for the rational numbers
We now give a complete list of the valuations on Q (up to equivalence). We write [ [
o
for the valuation on Q deﬁned by the usual absolute value on 1, and we say that [ [
o
is
normalized.
THEOREM 7.12 (OSTROWSKI) Let [ [ be a nontrivial valuation on Q.
(a) If [ [ is archimedean, then [ [ is equivalent to [ [
o
.
(b) If [ [ is nonarchimedean, then it is equivalent to [ [
;
for exactly one prime ].
PROOF. Let m. n be integers >1. Then we can write
m=a
0
÷a
1
n÷ ÷a
i
n
i
with the a
i
integers, 0 _a
i
<n, n
i
_m. Let N =max{1. [n[]. By the triangle inequality,
[m[ _
¸
[a
i
[[n[
i
_
¸
[a
i
[N
i
.
We know
r _log(m)¡log(n).
104
Complete list of valuations for the rational numbers
(log relative to some e >1) and the triangle inequality shows that
[a
i
[ _[1÷ ÷1[ =a
i
[1[ =a
i
_n.
On putting these into the ﬁrst inequality, we ﬁnd that
[m[ _(1÷r)nN
i
_
1÷
logm
logn
nN
logm
logn
.
In this inequality, replace m with m
t
(t an integer), and take t th roots:
[m[ _
1÷
t logm
logn
1
t
n
1
t
N
logm
logn
.
Now let t ÷o. The terms involving t tend to 1, and so
[m[ _N
logm
logn
. (*)
CASE (i): For all integers n >1, [n[ >1.
In this case N =[n[, and (*) yields:
[m[
1¡logn
_[n[
1¡logn
.
By symmetry, we must have equality, and so there is an c >1 such that
c =[m[
1¡logn
=[n[
1¡logn
for all integers m. n > 1. Hence
[n[ =c
logn
=e
logc logn
=n
logc
, all integers n >1.
Let a =logc, and rewrite this
[n[ =[n[
o
o
, all integers n >1.
where [ [
o
is the usual absolute value on Q. Since both [ [ and [ [
o
o
are homomorphisms
Q
×
//
1
>0
, the fact that they agree on a set of generators for the group Q
×
(the primes
and ÷1) implies that they agree on all of Q
×
.
CASE (ii): For some n >1, [n[ _1.
In this case, N =1, and (*) implies [m[ _ 1 for all integers m. Therefore the valuation
is nonarchimedean. Let · be the associated local ring and m its maximal ideal. From the
deﬁnition of ·, we know that Zc·. Then m¨Z is a prime ideal in Z (because mis a prime
ideal), and it is nonzero for otherwise the valuation would be trivial. Hence m¨Z=(]) for
some ]. This implies that [m[ =1 if m is an integer not divisible by ], and so [n]
i
[ =[][
i
if n is a rational number whose numerator and denominator are not divisible by ]. If a is
such that [][ =(1¡])
o
; then [.[ =[.[
o
;
for all . ÷ Q.
2
THEOREM 7.13 (PRODUCT FORMULA) For ] =2. 3. 5. 7. .... o, let [ [
;
be the correspond
ing normalized valuation on Q. For any nonzero rational number a
¸
[a[
;
=1 (product over all ] including o).
105
7. VALUATIONS; LOCAL FIELDS
PROOF. Let ˛ =a¡b, a. b ÷ Z. Then [˛[
;
=1 unless ][a or ][b. Therefore [˛[
¡
=1 for
all but ﬁnite many ·s, and so the product is really ﬁnite.
Let ¬(a) =
¸
[a[
¡
. Then ¬ is a homomorphism Q
×
÷1
×
, and so it sufﬁces to show
that ¬(÷1) = 1 and ¬(]) = 1 for each prime number ]. The ﬁrst is obvious, because
[ ÷1[ =1 for all valuations [ [. For the second, note that
[][
;
=1¡]. [][
q
=1. q a prime =]. [][
o
=].
The product of these numbers is 1.
2
The primes of a number ﬁeld
Let 1 be an algebraic number ﬁeld. An equivalence class of valuations on 1 is called a
prime or place of 1.
THEOREM 7.14 Let 1 be an algebraic number ﬁeld. There exists exactly one prime of 1
(a) for each prime ideal p;
(b) for each real embedding;
(c) for each conjugate pair of complex embeddings.
PROOF. See Chapter 8.
2
In each equivalence class of valuations of 1 we select a normalized valuation
1
as fol
lows:
for a prime ideal p of O
1
, [a[
p
=(1¡Np)
ord
p
(o)
=(O
p
: (a))
1
;
for a real embedding o: 1 ·÷1, [a[ =[oa[:
for a nonreal complex embedding o: 1 ·÷C, [a[ =[oa[
2
.
Note that this last is not actually a valuation, because it doesn’t satisfy the triangle law.
There are various ways of getting around this problem the best of which is simply to ignore
it.
Notations
We generally write · for a prime. If it corresponds to a prime ideal p of O
1
, then we call it a
ﬁnite prime, and we write p
¡
for the ideal. If it corresponds to a (real or nonreal) embedding
of 1, then we call it an inﬁnite (real or complex) prime. We write [ [
¡
for a valuation in the
equivalence class. If 1 ¯ 1 and n and · are primes of 1 and 1 such that [ [
u
restricted
to 1 is equivalent to [ [
¡
, then we say that n divides ·, or n lies over ·, and we write n[·.
For a ﬁnite prime, this means P
u
¨O
1
=p
¡
or, equivalently, that P
u
divides p
¡
O
1
. For
an inﬁnite prime, it means that n corresponds to an embedding o: 1 ·÷C that extends the
embedding corresponding to · (or its complex conjugate).
1
These are the most natural deﬁnitions for which the product formula hold. Alternatively, let 1
¡
be the
completion of 1 with respect to the valuation ·, and let u be a Haar measure on (1
¡
. ÷) — it is uniquely
determined up to a nonzero constant. For any nonzero a ÷ 1
¡
, u
o
(U)
df
= u(aU) is also a Haar measure on
(1
¡
. ÷), and so u
o
=c(a)u for some constant c(a). In fact, c(a) =[a[, the normalized valuation of a.
106
The primes of a number ﬁeld
THEOREM 7.15 (PRODUCT FORMULA) For each prime ·, let [ [
¡
be the normalized valu
ation. For any nonzero ˛ ÷ 1.
¸
[˛[
¡
=1 (product over all primes of 1).
PROOF. The product formula for a general number ﬁeld follows from the product formula
for Q and the next result.
2
LEMMA 7.16 Let 1 be a ﬁnite extension of a number ﬁeld 1.
(a) Each prime on 1 extends to a ﬁnite number of primes of 1.
(b) For any prime · of 1 and ˛ ÷ 1
×
,
¸
u¡
[˛[
u
=[ Nm
1¡1
˛[
¡
.
PROOF. See Chapter 8.
2
REMARK 7.17 The product formula is true in two other important situations.
(a) Let 1 be a ﬁnite extension of k(T ) where k is a ﬁnite ﬁeld. According to (7.3),
the valuations of 1 are all discrete, and hence correspond to discrete valuation rings in 1.
As in the number ﬁeld case, we can normalize a valuation by setting [a[
¡
=(1¡N·)
ord
v
(o)
where N· is the number of elements in the residue ﬁeld of the discrete valuation ring and
ord
¡
: 1
×
Z. Then
¸
¡
[a[
¡
= 1. The proof of this is easy when 1 = k(T ), and the
general case is obtained by means of a result like (7.16).
(b) Let 1 be a ﬁnite extension of k(T ) where k is an algebraically closed ﬁeld. In
this case we look only at primes that are trivial when restricted to k. All such primes are
nonarchimedean, and hence correspond to discrete valuations ord
¡
: 1
×
Z. Fix an e >1,
and deﬁne [a[
¡
=(1¡e)
ord
v
(o)
for every ·. Then
¸
[a[
¡
=1 for all a ÷ 1
×
. This of course
is equivalent to the statement
¸
ord
¡
(a) =0.
For example, let X be a compact Riemann surface, and let 1 be the ﬁeld of meromorphic
functions on X. For each point 1 of X we have a discrete valuation, deﬁned by ord
1
(} ) =
m or ÷m according as } has a zero or pole of order m at 1. The valuations ord
1
are
precisely the valuations on 1 trivial on Cc1, and so the product formula for 1 is simply
the statement that } has as many zeros as poles.
The proof of this runs as follows: the Cauchy integral formula implies that if } is a
nonconstant meromorphic function on an open set U in C, and 1 is the oriented boundary
of a compact set C contained in U, then
T
}
t
(:)
}(:)
J: =2¬i(7÷1)
where 7 is the number of zeros of } in C and 1 is the number of poles of } , both counted
with multiplicities. This formula also holds for compact subsets of manifolds. If the man
ifold M is itself compact, then we can take C = M, which has no boundary, and so the
formula becomes
7÷1 =0.
i.e.,
¸
ord
1
(} ) =0. 1 ÷ M.
107
7. VALUATIONS; LOCAL FIELDS
The weak approximation theorem
Recall that a valuation on a ﬁeld 1 is homomorphism a ÷ [a[ : 1
×
÷ 1
>0
such that
[a ÷b[ _ [a[ ÷[b[ for all a. b ÷ 1
×
. We extend it to 1 by setting [0[ = 0. A valuation
is trivial if [a[ = 1 for all a = 0. Two nontrivial valuations [ [
1
and [ [
2
are equivalent
if [a[
1
< 1 implies [a[
2
< 1, in which case [ [
2
= [ [
i
1
for some r ÷ 1
>0
(see 7.8). The
statements in this section continue to hold if we replace “valuation” with “positive power
of a valuation” (which, in the archimedean case, may fail to satisfy the triangle rule).
LEMMA 7.18 If [ [
1
, [ [
2
, . . . , [ [
n
are nontrivial inequivalent valuations of 1, then there
is an element a ÷ 1 such that
[a[
1
> 1
[a[
i
< 1. i =1.
PROOF. First let n =2. Because [ [
1
and [ [
2
are inequivalent, there are elements b and c
such that
[b[
1
< 1. [b[
2
_1
[c[
1
_ 1. [c[
2
<1.
Now a =
c
b
has the required properties.
We proceed by induction assuming that the lemma is true for n÷1 valuations. There
exist elements b. c such that
[b[
1
> 1. [b[
i
<1. i =2. 3. . . . . n÷1
[c[
1
< 1. [c[
n
>1
If [b[
n
_1, then a =cb
i
works for sufﬁciently large r. If [b[
n
> 1, then a
i
=
cb
r
1÷b
r
works
for sufﬁciently large r, because
b
r
1÷b
r
converges to 0 or 1 according as [b[ <1 or [b[ >1.
2
LEMMA 7.19 In the situation of the last lemma, there exists an element of 1 that is close
to 1 for [ [
1
and close to 0 for [ [
i
, i =2. . . . n.
PROOF. Choose a as in (7.18), and consider a
i
=
o
r
1÷o
r
. Then
[a
i
÷1[
1
=
1
[1÷a
i
[
1
_
1
[a[
i
1
÷1
÷0
as r ÷o. For i _2,
[a
i
[
i
=
[a[
i
i
[1÷a[
i
i
_
[a[
i
i
1÷[a[
i
i
÷0
as r ÷0.
2
THEOREM 7.20 Let [ [
1
, [ [
2
, . . . , [ [
n
be nontrivial inequivalent valuations of a ﬁeld
1, and let a
1
. . . . . a
n
be elements of 1. For any c > 0, there is an element a ÷ 1 such that
[a÷a
i
[
i
<c for all i .
PROOF. Choose b
i
, i =1. . . . . n, close to 1 for [ [
i
and close to 0 for [ [
}
, ; =i . Then
a =a
1
b
1
÷ ÷a
n
b
n
works.
2
108
Completions
Let 1
i
be the completion of 1 for [ [
i
. The statement of the theorem also holds with
a
i
in 1
i
(rather than 1)—choose a
t
i
÷ 1 very close to a
i
and a ÷ 1 very close to each a
t
i
.
Thus 1 (embedded diagonally) is dense in
¸
1
i
.
The theorem shows that there can be no ﬁnite product formula. More precisely:
COROLLARY 7.21 Let [ [
1
, [ [
2
, . . . , [ [
n
be nontrivial inequivalent valuations on a ﬁeld
1. If
[a[
i
1
1
[a[
i
n
n
=1. r
i
÷ 1.
for all a ÷ 1
×
, then r
i
=0 for all i .
PROOF. If any r
i
= 0, an a for which [a[
i
is sufﬁciently large and the [a[
}
, ; = i , are
sufﬁciently small provides a contradiction.
2
The reader should compare the Weak Approximation Theorem with what the Chinese
Remainder Theorem gives (see Exercise 71).
NOTES The Weak Approximation Theorem ﬁrst occurs in Artin and Whaples 1945.
2
See also Artin
1959, Our account follows the original.
Completions
Let 1 be a ﬁeld with a nontrivial valuation. A sequence (a
n
) of elements in 1 is called a
Cauchy sequence if, for every c >0, there is an N such that
[a
n
÷a
n
[ <c, all m. n > N.
The ﬁeld 1 is said to be complete if every Cauchy sequence has a limit in 1. (The limit is
necessarily unique.)
EXAMPLE 7.22 Consider the sequence in Z
4. 34. 334. 3334. . . . .
As
[a
n
÷a
n
[
5
=5
n
(m>n).
this is a Cauchy sequence for the 5adic topology on Q. Note that
3 4 =12. 3 34 =102. 3 334 =1002. 3 3334 =10002. . . .
and so 3 a
n
÷2 ÷0 as n ÷o. Thus lim
no
a
n
=2¡3 ÷ Q.
There is a similar notion of Cauchy series. For example, any series of the form
a
n
]
n
÷ ÷a
0
÷a
1
]÷ ÷a
n
]
n
÷ . 0 _a
i
<].
is a Cauchy series in Q for the ]adic topology.
2
Axiomatic characterization of ﬁelds by the product formula for valuations, Bull. AMS, 51, 1945, pp.
469–492.
109
7. VALUATIONS; LOCAL FIELDS
THEOREM 7.23 Let 1 be a ﬁeld with a valuation [ [. Then there exists a complete valued
ﬁeld (
´
1. [ [) and a homomorphism 1 ÷
´
1 preserving the valuation that is universal in the
following sense: any homomorphism 1 ÷1 from 1 into a complete valued ﬁeld (1. [ [)
preserving the valuation, extends uniquely to a homomorphism
´
1 ÷1.
PROOF (SKETCH) Every point of
´
1 will be the limit of a sequence of points in 1, and the
sequence will be Cauchy. Two Cauchy sequences will converge to the same point in
´
1 if
and only if they are equivalent in the sense that
lim
no
[a
n
÷b
n
[ =0.
This suggests deﬁning
´
1 to be the set of equivalence classes of Cauchy sequences in 1.
Deﬁne addition and multiplication of Cauchy sequences in the obvious way, and verify
that
´
1 is a ﬁeld. There is a canonical map 1 ÷
´
1 sending a to the constant Cauchy
sequence a. a. a. . . ., which we use to identify 1 with a subﬁeld of
´
1. We can extend a
homomorphism from 1 into a second complete valued ﬁeld 1 to
´
1 by mapping the limit
of a Cauchy sequence in
´
1 to its limit in 1.
2
REMARK 7.24 (a) As usual, the pair (1 ÷
´
1. [ [) is uniquely determined up to a unique
isomorphism by the universal property (cf. GT 2.4).
(b) The image of 1 in
´
1 is dense because the closure
¨
1 of 1 in
´
1 is complete, and
(1 ÷
¨
1. [ [) has the same universal property as (1 ÷
´
1. [ [).
For a prime · of 1, we write 1
¡
for the completion of 1 with respect to ·. When
· corresponds to a prime ideal p, we write 1
p
for the completion, and
´
O
p
for the ring of
integers in 1
p
. For example, Q
;
is the completion of Qwith respect to the ]adic valuation
[ [
;
. We write Z
;
(not
´
Z
;
) for the ring of integers in Q
;
(the ring of ]adic integers).
Completions in the nonarchimedean case
Let [ [ be a discrete nonarchimedean valuation on 1, and let ¬ be an element of 1 with
largest value <1 (therefore ¬ generates the maximal ideal m in the valuation ring ·). Such
a ¬ is called a local uniformizing parameter.
The set of values is
[1[ ={c
n
[ m÷ Z] L{0]. c =[¬[.
Let a ÷
´
1
×
, and let a
n
be a sequence in 1 converging to a. Then [a
n
[ ÷[a[ (because [ [
is a continuous map), and so [a[ is a limit point for the set [1
×
[. But [1
×
[ is closed (being
discrete), and so [a[ ÷ [1
×
[. Thus [
´
1[ =[1[, and so [ [ is a discrete valuation on
´
1 also.
Let ord: 1
×
Z be a normalized discrete additive valuation corresponding to [ [; then ord
extends to a normalized discrete valuation on
´
1.
Note that if a
n
÷a =0, then [a
n
[ ÷[a[ =0, and (because [1
×
[ is discrete), [a
n
[ =[a[
for all n large enough.
The ring associated with [ [ in
´
1 is
´
· ={a ÷
´
1 [ [a[ _1].
110
Completions in the nonarchimedean case
Clearly
´
· is the set of limits of Cauchy sequences in ·, and it is therefore the closure of ·
in
´
1. The maximal ideal in
´
· is
´ m={a ÷
´
1 [ [a[ <1].
Again it is the set of limits of Cauchy sequences in m, and so it is the closure of m. Similarly,
´ m
n
is the closure of m
n
. Let ¬ be an element with ord(¬) = 1; then ¬ generates m in ·
and ´ m in
´
·.
LEMMA 7.25 For any n, the map ·¡m
n
÷
´
·¡ ´ m
n
is an isomorphism.
PROOF. Note that
m
n
={a ÷ · [ [a[ _[¬[
n
] ={a ÷ · [ [a[ <[¬[
n1
]
is both open and closed in ·. Because it is closed, the map is injective; because ´ m
n
is open,
the map is surjective.
2
PROPOSITION 7.26 Choose a set S of representatives for ·¡m, and let ¬ generate m. The
series
a
n
¬
n
÷ ÷a
0
÷a
1
¬ ÷ ÷a
n
¬
n
÷ . a
i
÷ S
is a Cauchy series, and every Cauchy series is equivalent to exactly one of this form. Thus
each element of
´
1 has a unique representative of this form.
PROOF. Let s
Æ
=
¸
Æ
i=n
a
i
¬
i
. Then
[s
Æ
÷s
1
[ _[¬[
Æ÷1
, if M <N.
which shows that the sequence s
Æ
is Cauchy. Let ˛ ÷
´
1. Because [
´
1[ = [1[, we can
write ˛ =¬
n
˛
0
with ˛
0
a unit in
´
·. From the deﬁnition of S, we see that there exists an
a
0
÷ S such that ˛
0
÷a
0
÷ ´ m. Now
˛
0
o
0
t
÷
´
·, and so there exists an a
1
÷ S such that
˛
0
o
0
t
÷a
1
÷ ´ m. Now there exists an a
2
such that
˛
0
o
0
o
1
t
t
2
÷a
2
÷ ´ m, etc. In the limit,
˛
0
=a
0
÷a
1
¬ ÷ . ˛ =¬
n
˛
0
.
Note that
[
¸
a
i
¬
i
[ =[¬
n
[
if a
n
is the ﬁrst nonzero coefﬁcient. Therefore
¸
a
i
¬
i
=0 (if and) only if a
i
=0 for all i .
This proves the uniqueness.
2
Thus, for example, every equivalence class of Cauchy sequences in Q for [ [
;
has a
unique representative of the form
a
n
]
n
÷ ÷a
0
÷a
1
]÷a
2
]
2
÷ . 0 _a
i
<].
Note that the partial sums of such a series are rational numbers. It is as easy to work with
such series as with decimal expansions of real numbers — just remember high powers of ]
are small, and hence the ﬁrst to be ignored.
111
7. VALUATIONS; LOCAL FIELDS
We explain this in more detail. The maps
Z¡(]
n
) ÷Z
(;)
¡(]
n
) ÷Z
;
¡(]
n
)
are both bijective (see 3.10 for the ﬁrst map). Let ˛ ÷ Z
;
. Because the map is bijective, for
all n, there is an a
n
÷ Z such that ˛ ÷a
n
mod ]
n
. Note that, if n <m, a
n
÷a
n
mod ]
n
,
which implies that (a
n
) is a Cauchy sequence. Let
a
n
÷c
0
÷c
1
]÷ ÷c
n1
]
n1
mod ]
n
. 0 _c
i
_]÷1:
then
˛ =
¸
i_0
c
i
]
i
.
Conversely, if ˛ =
¸
c
i
]
i
, 0 _c
i
_]÷1, then c
0
. c
1
. . . . is the unique sequence of integers,
0 _c
i
_]÷1, such that
˛ ÷
n1
¸
i=0
c
i
]
i
mod ]
n
.
If ˛ ÷ Q
;
but not Z
;
, then ]
n
˛ ÷ Z
;
for a sufﬁciently large m, and the above arguments
can be applied to it.
The following examples illustrate how to work with ]adic numbers.
EXAMPLE 7.27 In Q
2
.
1÷2÷ ÷2
n
÷
converges to ÷1, because the sum of the ﬁrst n terms is
2
n
÷1
2÷1
=2
n
÷1
which converges to ÷1.
EXAMPLE 7.28 I claim that ÷1 is a square in Q
5
. We have to ﬁnd a series
a
0
÷a
1
5÷a
2
5
2
÷ . a
i
=0. 1. 2. 3, or 4
such that
(a
0
÷a
1
5÷a
2
5
2
÷...)
2
÷1 =0.
We ﬁrst need that
a
2
0
÷1 ÷0 mod 5.
Thus we must take a
0
= 2 or 3; we choose 2 (choosing 3 would lead to the other root).
Next we need
(2÷a
1
5)
2
÷1 ÷0 mod 5
2
.
and so we want
5÷20a
1
÷0 (mod 5
2
).
We must take a
1
=1. Suppose we have found
c
n
=a
0
÷a
1
5÷a
2
5
2
÷ ÷a
n
5
n
112
Completions in the nonarchimedean case
such that
c
2
n
÷1 ÷0 (mod 5
n÷1
).
and consider c
n
÷a
n÷1
5
n÷1
. We want
(c
n
÷a
n÷1
5
n÷1
)
2
÷1 ÷0 (mod 5
n÷2
),
for which we need that
c
2
n
÷1÷2c
n
a
n÷1
5
n÷1
÷0 (mod 5
n÷2
).
or that
2c
n
a
n÷1
5
n÷1
÷(÷1÷c
2
n
) (mod 5
n÷2
).
or that
2c
n
a
n÷1
÷(÷1÷c
2
n
)¡5
n÷1
(mod 5).
or that
4a
n÷1
=(÷1÷c
2
n
)¡5
n÷1
(mod 5).
Since 4 is invertible modulo 5, we can always achieve this. Hence we obtain a series
converging to ÷1. In fact,
÷1 =
1
2
1÷5 =
1
2
¸
o
n=0
(÷1)
n
1
2
n
5
n
.
EXAMPLE 7.29 We study the convergence of the power series
exp(.) =1÷. ÷
.
2
2Š
÷ ÷
.
n
nŠ
÷
in Q
;
. Write
n =a
0
÷a
1
]÷ ÷a
i
]
i
. 0 _a
i
_]÷1.
Then
ord
;
(nŠ) =
¸
n
]
¸
÷
¸
n
]
2
¸
÷ ÷
¸
n
]
i
¸
.
where here Œa denotes the ﬂoor of a (largest integer less than a), and
¸
n
]
¸
= a
1
÷ a
2
]÷a
3
]
2
÷ ÷a
i
]
i1
¸
n
]
2
¸
= a
2
÷a
3
] ÷ ÷a
i
]
i2
¸
n
]
i
¸
= a
i
On summing these equalities, we ﬁnd that
ord
;
(nŠ) =a
0
]
0
÷1
]÷1
÷a
1
]
1
÷1
]÷1
÷a
2
]
2
÷1
]÷1
÷ ÷a
i
]
i
÷1
]÷1
=
n÷
¸
a
i
]÷1
.
113
7. VALUATIONS; LOCAL FIELDS
Therefore
ord
;
.
n
nŠ
=n
ord
;
(.) ÷
1
;1
÷
¸
o
i
;1
.
As
¸
o
i
;1
_
log(n)
log(;)
, we see that
x
n
nŠ
÷0 if and only if ord(.) >
1
;1
. Therefore (see Exercise
72), the series exp(.) converges for ord(.) >
1
;1
.
There is a leisurely, and very detailed, discussion of Q
;
in the ﬁrst chapter of Koblitz
1977
3
.
ASIDE 7.30 Those who have taken a course in commutative algebra will know another method of
completing a local ring 1, namely
1
t
=lim

1¡m
n
={(a
n
) [ a
n
÷ 1¡m
n
. a
n÷1
÷a
n
mod m
n
].
In the case that 1 is a discrete valuation ring, this deﬁnition agrees with the above. There is an
injective homomorphism
1 ÷1
t
. a ÷(a
n
). a
n
=a mod ¬
n
.
We can deﬁne a homomorphism 1
t
÷
´
1 as follows: let (a
n
) ÷ 1
t
, and choose a representative a
t
n
for a
n
in 1; then (a
t
n
) is an Cauchy sequence whose equivalence class is independent of the choices
of the a
t
n
, and we can map (a
n
) to (a
t
n
). It is easy to see that the map 1
t
÷
´
1 is surjective, and it
follows that it is an isomorphism.
Newton’s lemma
The argument in the above example works much more generally. Let }(X) =X
2
÷1. Then
all we in fact used was that }(X) has a simple root modulo 5.
In the rest of this subsection, · is a complete discrete valuation ring and ¬ generates its
maximal ideal (unless we say otherwise).
PROPOSITION 7.31 Let }(X) ÷ ·ŒX, and let a
0
be a simple root of }(X) mod ¬. Then
there is a unique root a of }(X) with a ÷a
0
mod ¬.
PROOF. Suppose we have found a
n
÷a
0
mod ¬ such that
}(a
n
) ÷0 mod ¬
n÷1
.
Let a
n÷1
=a
n
÷h¬
n÷1
, h ÷ ·. We want
}(a
n
÷h¬
n÷1
) ÷0 mod ¬
n÷2
.
Recall (trivial Taylor’s expansion) that, for any polynomial }.
}(c ÷t ) =}(c) ÷t }
t
(c) ÷
where }
t
(X) is the formal derivative of }(X). Then
}(a
n
÷h¬
n÷1
) =}(a
n
) ÷h¬
n÷1
}
t
(a
n
) ÷ .
3
Koblitz, Neal. ]adic numbers, ]adic analysis, and zetafunctions. Graduate Texts in Mathematics, Vol.
58. SpringerVerlag, New YorkHeidelberg, 1977.
114
Newton’s lemma
which we want ÷0 mod ¬
n÷2
. Hence we must take h so that
h =÷
}(a
n
)
¬
n÷1
}
t
(a
n
)
1
mod ¬.
This is possible because ¬
n÷1
[}(a
n
) and
}
t
(a
n
) ÷}
t
(a
0
) mod ¬.
which is nonzero, and hence invertible, mod ¬.
2
There is a stronger form of the proposition. Recall Newton’s approximation
4
method
for ﬁnding a solution to }(.) =0, where } is a function of a real variable. Starting from
an a
0
such that }(a
0
) is small, deﬁne a sequence a
1
. a
2
. ... by putting
a
n÷1
=a
n
÷}(a
n
)¡}
t
(a
n
).
Often a
n
converges to a root of }(.). In the above proof, this is what we did, but the same
argument can be made to work more generally.
THEOREM 7.32 (NEWTON’S LEMMA) Let }(X) ÷ ·ŒX. Let a
0
÷ · satisfy
[}(a
0
)[ <[}
t
(a
0
)[
2
.
Then there is a unique root a of }(X) such that
[a÷a
0
[ _
ˇ
ˇ
ˇ
ˇ
}(a
0
)
}
t
(a
0
)
2
ˇ
ˇ
ˇ
ˇ
.
PROOF. Deﬁne a sequence a
0
. a
1
. . . . by setting
a
n÷1
=a
n
÷
}(a
n
)
}
t
(a
n
)
and prove that it is a Cauchy sequence converging to a root of }(X). See, for example,
Milne 2006, 2.12.
2
Proposition 7.31 shows that a simple factor of degree 1 of }(X) mod ¬ lifts to a factor
of }(X). This generalizes.
THEOREM 7.33 (HENSEL’S LEMMA) Let k be the residue ﬁeld of ·; for }(X) ÷ ·ŒX,
write
¨
} (X) for the image of } in kŒX. Consider a monic polynomial }(X) ÷ ·ŒX. If
¨
} (X) factors as
¨
} = g
0
h
0
with g
0
and h
0
monic and relatively prime (in kŒX), then }
itself factors as } =gh with g and h monic and such that ¨ g =g
0
and
¨
h =h
0
. Moreover,
g and h are uniquely determined, and (g. h) =·ŒX.
We ﬁrst prove that (g. h) = ·ŒX (such a pair is said to be strictly coprime; in kŒX
strictly coprime just means coprime, i.e., relatively prime).
4
When Newton found his interpolation formula in 1670, ancient Chinese mathematicians had been using
the formula in more sophisticated forms for more than one millennium. He, JiHuan, Appl. Math. Comput.
152 (2004), no. 2, 367–371.
115
7. VALUATIONS; LOCAL FIELDS
LEMMA 7.34 Let · be a local ring with residue ﬁeld k. If }. g ÷ ·ŒX are such that
¨
}
and ¨ g are relatively prime and } is monic, then (}. g) =·ŒX. More precisely, there exist
u. · ÷ ·ŒX with degu <degg and deg· < deg} such that
u} ÷·g =1. (18)
PROOF. Let M =·ŒX¡(}. g). As } is monic, this is a ﬁnitely generated ·module. As
(
¨
} . ¨ g) =kŒX, we have that (}. g) ÷m·ŒX =·ŒX and so mM =M. Now Nakayama’s
Lemma (1.9) implies that M =0.
This shows that there exist u. · ÷ ·ŒX such that (18) holds. If deg· _ deg} , write
· =}q ÷r with degr < deg} . Then
(u÷qg)} ÷rg =1.
and u÷qg automatically has degree <degg.
2
We next prove uniqueness of g and h.
LEMMA 7.35 Let · be a local ring with residue ﬁeld k. Suppose } = gh = g
t
h
t
with
g. h. g
t
. h
t
all monic, and ¨ g = ¨ g
t
,
¨
h =
¨
h
t
with ¨ g and
¨
h relatively prime. Then g =g
t
and
h =h
t
.
PROOF. From the preceding lemma we know that (g. h
t
) =·ŒX, and so there exist r. s ÷
·ŒX such that gr ÷h
t
s =1. Now
g
t
=g
t
gr ÷g
t
h
t
s =g
t
gr ÷ghs.
and so g divides g
t
. As both are monic and have the same degree, they must be equal.
2
Finally, we prove the existence of g and h. We are given that there exist monic polyno
mials g
0
, h
0
÷ ·ŒX such that
} ÷g
0
h
0
÷ ¬ ·ŒX.
Suppose we have constructed monic polynomials g
n
, h
n
such that
} ÷g
n
h
n
÷0 mod ¬
n÷1
·ŒX
and g
n
÷g
0
, h
n
÷h
0
mod ¬·ŒX. We want to ﬁnd u, · ÷ ·ŒX with degu < degg
0
and
deg· < degh
0
such that
} ÷(g
n
÷¬
n÷1
u)(h
n
÷¬
n÷1
·) ÷0 mod ¬
n÷2
·ŒX.
i.e., such that
(} ÷g
n
h
n
) ÷¬
n÷1
(uh
n
÷g
n
·) ÷0 mod ¬
n÷2
·ŒX.
Thus we are looking for polynomials u, · in ·ŒX with degu < degg
0
and deg· < degh
0
such that
uh
n
÷g
n
· ÷(} ÷g
n
h
n
)¡¬
n÷1
mod ¬·ŒX.
Because g
0
and h
0
are monic and relatively prime, Lemma 7.34 shows that such polyno
mials exist.
116
Extensions of nonarchimedean valuations
REMARK 7.36 An induction argument extends the theorem to show that a factorization of
} into a product of relatively prime polynomials in kŒX lifts to a factorization in ·ŒX.
For example, in F
;
ŒX, X
;
÷X splits into ] distinct factors, and so it also splits in Z
;
ŒX.
Hence Z
;
contains the (]÷1)st roots of 1. More generally, if 1 has a residue ﬁeld k with
q elements, then 1 contains q roots of the polynomial X
q
÷X. Let S be the set of these
roots. Then
a ÷ ¨ a: S ÷k.
is a bijection preserving multiplication (but not, of course, addition) – the elements of S are
called the Teichm¨ uller representatives for the elements of the residue ﬁeld.
REMARK 7.37 Theorems 7.32 and 7.33 are both stronger versions of 7.31. There is in fact
a stronger version of 7.32. For a polynomial h =
¸
c
i
X
i
, deﬁne
h =max[c
i
[.
Let
}(X) =a
n
X
n
÷a
n1
X
n1
÷ ÷a
0
÷ ·ŒX
have [a
n
[ =1 (i.e., a
n
is a unit). Let g
0
(X) and h
0
(X) be polynomials in ·ŒX with degrees
r and s respectively, and suppose that
}(X) ÷g
0
(X)h
0
(X) <[ Res(g
0
(X). h
0
(X))[
2
where Res denotes the resultant. Then }(X) factors in ·ŒX as the product of a polynomial
of degree r and a polynomial of degree s. The proof follows the same general lines as the
above proofs. In fact, the hypothesis can be replaced by
}(X) ÷g
0
(X)h
0
(X) <[disc(} )[.
(For this, see Cassels 1986, p107.)
Note that, this gives an algorithm for factoring polynomials in Q
;
ŒX (for example).
Given }(X), compute disc(} ). If this is zero, then } and }
t
have a common factor,
which we can ﬁnd by the Euclidean algorithm. Otherwise ord(disc(} )) =m for some m,
and it is enough to consider factorizations of } into polynomials with coefﬁcients in the
ﬁnite ring Z¡]
n
Z. Apparently the fastest algorithms for factoring polynomials in ZŒX
begin by factoring in Z
;
ŒX for an appropriate prime ] — computers seem to have no
problem handling polynomials of degree 200. (But Exercise 76 shows that there exist
irreducible polynomials in ZŒX of arbitrarily large degree that factor in all the rings Z
;
ŒX
into polynomials of low degree.)
Extensions of nonarchimedean valuations
We explain how to extend a valuation to a larger ﬁeld.
THEOREM 7.38 Let 1 be complete with respect to a discrete valuation [ [
1
, and let 1 be a
ﬁnite separable extension of 1 of degree n. Then [ [ extends uniquely to a discrete valuation
[ [
1
on 1, and 1 is complete for the extended valuation. For all ˇ ÷ 1.
[ˇ[
1
=[ Nm
1¡1
ˇ[
1¡n
1
.
117
7. VALUATIONS; LOCAL FIELDS
PROOF. Let A be the discrete valuation ring in 1, and let T be its integral closure in 1.
Let p be the maximal ideal of ·. We know from (3.29) that T is a Dedekind domain, and
the valuations of 1 extending [ [
p
correspond to the ideals of T lying over p.
Suppose that there are distinct prime ideals P
1
and P
2
in T dividing p. There will be
a ˇ ÷ T such that P
1
¨·Œˇ =P
2
¨·Œˇ; for example, choose ˇ ÷ T such that ˇ ÷ P
1
,
ˇ … P
2
. Let }(X) be the minimum polynomial of ˇ over 1, so that ·Œˇ .·ŒX¡(}(X)).
Because }(X) is irreducible in ·ŒX and · is complete, Hensel’s lemma shows that
¨
} (X)
(image of }(X) in kŒX, k =·¡p) must be a power of an irreducible polynomial. Then
·Œˇ¡p·Œˇ ~kŒX¡(
¨
} (X))
is a local ring, which contradicts the fact that ·Œˇ has two prime ideals containing p.
Hence [ [
p
extends uniquely to a valuation [ [ on 1.
Clearly, [ [
p
also extends uniquely to the Galois closure 1
t
of 1. For each o ÷Gal(1¡1),
consider the map 1 ·÷C, ˇ ÷[oˇ[. This is again a valuation of 1, and so the uniqueness
implies that [ˇ[ =[oˇ[. Now
[ Nm(ˇ)[ =[
¸
oˇ[ =[ˇ[
n
which implies the formula.
Finally, we have to show that 1 is complete. Let e
1
. . . . . e
n
be a basis for T as an
·module, and let (˛(m)) be a Cauchy sequence in 1. Write ˛(m) = a
1
(m)e
1
÷ ÷
a
n
(m)e
n
, with a
i
(m) ÷ 1. For each i , a
i
(m) is a Cauchy sequence, and if a
i
denotes its
limit, then ˛
df
=a
1
e
1
÷ ÷a
n
e
n
is the limit of the sequence ˛(m).
2
REMARK 7.39 It is obvious from the criterion (7.2) that a nonarchimedean valuation can
only extend to a nonarchimedean valuation. It is possible to prove (7.38) without assuming
that the valuation [ [ on 1 is discrete or even nonarchimedean, but the proof is then com
pletely different, and much longer — we shall in fact need this in the Chapter 8, and so I
should have included it. The formula [ˇ[
1
= [ Nm
1¡1
ˇ[
1¡n
1
shows that [ [
1
is discrete if
and only if [ [
1
is discrete.
COROLLARY 7.40 Let 1 be as in the theorem, and let ˝ be a (possibly inﬁnite) separable
algebraic extension of 1. Then [ [ extends in a unique way to a valuation [ [ on ˝.
PROOF. The theorem shows that [ [ extends in a unique way to any ﬁnite subextension of
˝, and hence it extends uniquely to ˝.
2
REMARK 7.41 In the last corollary, the extended valuation is still nonarchimedean, but it
need not be discrete, and ˝ need not be complete. However, the completion of ˝ is again
algebraically closed.
For example as we noted in (7.6), the valuation on the algebraic closure Q
al
;
of Q
;
is not
discrete, and Exercise 77 shows that Q
al
;
is not complete. The completion of Q
al
;
is often
denoted C
;
because it plays the same role for the ]adic valuation on Qthat C plays for the
real valuation. (In fact C
;
~C as abstract ﬁelds because they are both algebraically closed,
and they both have a transcendence basis with cardinality equal to that of 1. The isomor
phism is as far from being canonical as it is possible to get — its construction requires the
axiom of choice.)
118
Newton’s polygon
COROLLARY 7.42 Let 1 and 1 be as in the theorem; then n =e} where n =Œ1 : 1, e
is the ramiﬁcation index, and } is the degree of the residue ﬁeld extension.
PROOF. We know from (3.34) that n =
¸
e
i
}
i
. In this case, there is only one prime divid
ing p and so the formula becomes n =e}.
2
When e =n, so that pT =P
n
, we say that 1 is totally ramiﬁed over 1; when } =n,
we say that 1 is unramiﬁed over 1.
Note that the valuation ring T of 1 is the integral closure of the valuation ring · of 1.
Many of the results proved above for complete discrete valuation rings hold also for
Henselian local rings (see :4 of my notes Lectures on Etale Cohomology).
REMARK 7.43 Let 1 be complete with respect to a discrete valuation, and let 1 be a ﬁnite
extension of 1. Let P and p be the maximal ideals in the rings of integers · and T of
1 and 1. Then pT = P
e
where e is the ramiﬁcation index. Let ¬ and ˘ be generators
of p and P. The normalized valuations ord
1
and ord
1
on 1 and 1 are characterized by
equations:
ord
1
(¬) =1. ord
1
(˘) =1.
Note that ¬ =˘
e
×unit, and so
ord
1
=e
1
ord
1
.
If we denote the extension of ord
1
to 1 by ord, then
ord(1
×
) =e
1
Z.
This characterizes the ramiﬁcation index.
Newton’s polygon
Let 1 be complete with respect to a discrete valuation. Let ord be the corresponding addi
tive valuation ord: 1
×
Z, and extend ord to a valuation ord : 1
al×
÷Q. For a polynomial
}(X) =X
n
÷a
1
X
n1
÷ ÷a
n
. a
i
÷ 1.
deﬁne the Newton polygon
5
of }(X) to be the lower convex hull of the set of points
1
i
def
=(i. ord(a
i
)), i =0. .... n.
In more detail, rotate the negative .axis counterclockwise about 1
0
=(0. 0) until it hits
a 1
i
— the ﬁrst segment of the Newton polygon is the line 1
0
1
i
1
where 1
i
1
is the point
furthest from 1
0
on the rotated .axis. Repeat the process rotating about 1
i
1
, etc.. The
resulting polygon starts at 1
0
and ends at 1
n
; each of its segments begins and ends at a 1
i
;
each 1
i
either lies on the polygon or is above it; any line joining two points of the polygon
has no point that is below the polygon (this is what we mean by the Newton polygon being
lower convex).
5
Most people write the polynomial a
0
÷a
1
X ÷ ÷X
n
when they deﬁne Newton polygons. This is
slightly less convenient than the way I do it, but allows you to deﬁne the Newton polygon of a power series.
119
7. VALUATIONS; LOCAL FIELDS
PROPOSITION 7.44 Suppose that the Newton polygon of }(X) ÷ 1ŒX has segments of
.length n
i
and slope s
i
. Then }(X) has exactly n
i
roots ˛ (in 1
al
) with
ord(˛) =s
i
.
Moreover, the polynomial }
i
(X)
def
=
¸
ord(˛
i
)=x
i
(X ÷˛
i
) has coefﬁcients in 1.
PROOF. In proving the ﬁrst part, we don’t have to assume that }(X) has coefﬁcients in 1
— any ﬁnite extension of 1 will do. Thus it sufﬁces to prove the following statement: let
}(X) =
¸
(X ÷˛
}
); if exactly n
i
of the ˛
}
’s have ord(s
i
), then the Newton polygon of
}(X) has a segment of slope s
i
and .length n
i
.
We prove this by induction on n =deg(} ). If n =1, then it is obvious. Assume it for
n, and put
g(X) =(X ÷˛)}(X) =X
n÷1
÷b
1
X
n
÷b
2
X
n1
÷ ÷b
n÷1
.
Note that b
i
=a
i
÷˛a
i1
.
CASE (i). ord(˛) < s
1
. Recall ord(a ÷b) _ min{ord(a). ord(b)], with equality if
ord(a) =ord(b). Using this, one ﬁnds that
the Newton polygon of g is obtained from that of } by adding a segment of slope ord(˛)
and .length 1, and moving the Newton polygon of } to start at (1. ord(˛)). This is what
the proposition predicts.
CASE (ii). ord(˛) =s
1
. In this case, the initial segment of slope s
1
is lengthened by 1,
and the rest of the polygon is as before. This is what the proposition predicts.
The remaining cases are similar.
We now prove the second statement. Let ˛ be a root of }(X), and let m
˛
(X) be the
minimum polynomial of ˛. As we saw in the proof of (7.38), ord(˛
t
) = ord(˛) for all
conjugates ˛
t
of ˛, i.e., for all roots of m
˛
(X). Because }(˛) =0, m
˛
(X)[}(X), and the
remark just made implies that in fact m
˛
(X)[}
i
(X) where s
i
= ord(˛). If ˇ is a root of
}
i
(X)¡m
˛
(X), then a similar argument shows that m
ˇ
(X)[(}
i
¡m
˛
). Continuing in this
way, we ﬁnd that }
i
(X) is a product of polynomials with coefﬁcients in 1.
2
EXAMPLE 7.45 Consider the polynomial
6
}(X)
def
=X
3
÷X
2
÷2X ÷8.
By testing ˙1, ˙2, ˙4, ˙8 (actually, by asking PARI) one sees that this polynomial is
irreducible over Q. The Newton polygon of } relative to ord
2
has slopes 0. 1. 2, each with
.length 1. Therefore } splits in Q
2
ŒX, and it has roots ˛
1
, ˛
2
, ˛
3
with ords 0, 1, 2.
Locally compact ﬁelds
We now look at the compactness properties of our ﬁelds.
6
Keith Conrad suggests changing the polynomial to X
3
÷X
2
÷2X ÷8. As he writes: The roots of this
are the negatives of the roots of X
3
÷X
2
÷2X ÷8, so you don’t lose anything but you do gain simplicity of
appearance: having all signs past the leading term equal makes it easier to remember what the polynomial is!
Perhaps Dedekind himself even used the choice with all negative coefﬁcients; I haven’t looked up his paper to
be sure, but I did check in Hensel’s 1894 Crelle paper on extraordinary prime factors of the discriminant that
he wrote the polynomial as X
3
÷X
2
÷2X ÷8.
120
Locally compact ﬁelds
PROPOSITION 7.46 Let 1 be complete with respect to a nonarchimedean discrete valua
tion. Let · be the ring of integers in 1 and let m be the maximal ideal in ·. Then · is
compact if and only if ·¡m is ﬁnite.
PROOF. Let S be a set of representatives for ·¡m. We have to show that · is compact if
and only if S is ﬁnite.
=: Clearly m={. ÷ 1 [ [.[ < 1] is open in 1. As · is the disjoint union of the open
sets s ÷m, s ÷ S, S must be ﬁnite if · is compact.
=: Recall that a metric space X is compact if and only if it is complete and totally
bounded (this means that for any r > 0, there is a ﬁnite covering of X by open balls of
radius r). But every element of · can be written
s
0
÷s
1
¬ ÷s
2
¬
2
÷ ÷s
n
¬
n
÷ . s
i
÷ S.
For a ﬁxed n, there are only ﬁnitely many sums
s
0
÷s
1
¬ ÷s
2
¬
2
÷ ÷s
n
¬
n
. s
i
÷ S.
and every element of · is within [¬
n÷1
[ of such an element.
2
COROLLARY 7.47 Assume that the residue ﬁeld is ﬁnite. Then p
n
, 1÷p
n
, and ·
×
are all
compact.
PROOF. They are all closed subsets of ·.
2
DEFINITION 7.48 A local ﬁeld is a ﬁeld 1 with a nontrivial valuation [ [ (as deﬁned at the
start of this section) such that 1 is locally compact (and hence complete).
REMARK 7.49 It is possible to give a complete classiﬁcation of local ﬁelds.
(a) Let 1 be a ﬁeld that is complete with respect to an archimedean valuation [ [; then
1 is isomorphic to 1 or C, and the valuation is equivalent to the usual absolute value (also
a theorem of Ostrowski).
7
Thus for archimedean valuations, completeness implies local
compactness.
(b) A nonarchimedean local ﬁeld 1 of characteristic zero is isomorphic to a ﬁnite exten
sion of Q
;
, and the valuation is equivalent to the (unique) extension of the ]adic valuation.
(To prove this, note that, by assumption, 1 contains Q. The restriction of [ [ to Q can’t be
the trivial valuation, because otherwise ·
×
wouldn’t be compact. Therefore (see 7.12) [ [
induces a valuation on Q equivalent to the ]adic valuation for some prime number ]. The
7
Here is a sketch of the proof. The ﬁeld 1 contains Q, and the restriction of [ [ to Q is the usual absolute
value. Therefore 1 contains 1, and after adjoining a square root of ÷1 (if necessary), we may assume 1 ¯C.
Let . ÷ 1C, and let c be the closest element of C to .. Replace . with . ÷c, so that now [. ÷:[ _ [.[
for all : in C. It follows that
[.
n
÷:
n
[ =[. ÷:[[. ÷¯:[[. ÷¯
2
:[ _[. ÷:[[.[
n1
.
where ¯ is a primitive nth root of 1.
On choosing [:[ <1 and letting n ÷o, we ﬁnd that [.[ _[. ÷:[. Hence [. ÷:[ =[.[ and so (taking . ÷:
in place of .) [. ÷2:[ =[.[, and thus (repeating the argument) [. ÷n:[ =[.[, contradicting the archimedean
property.
121
7. VALUATIONS; LOCAL FIELDS
closure of Q in 1 is therefore Q
;
. If 1 has inﬁnite degree over Q
;
, it will not be locally
compact.)
(c) A nonarchimedean local ﬁeld 1 of characteristic ] = 0 is isomorphic to the ﬁeld
of formal Laurent series k((T )) over a ﬁnite ﬁeld k. The ﬁeld k((T )) is the completion of
k(T ) for the valuation deﬁned by the ideal (T ) c kŒT ; it consists of ﬁnitetailed formal
power series:
o
¸
i_n
a
i
T
i
.
Unramiﬁed extensions of a local ﬁeld
Again 1 is a ﬁeld complete with respect to a discrete valuation [ [. To avoid problems with
separability, we assume that 1 and the residue ﬁeld k are both perfect
8
— of course in the
case we are particularly interested in, 1 has characteristic zero and k is ﬁnite. Let · be the
discrete valuation ring in 1 corresponding to [ [.
If 1 is an algebraic (possibly inﬁnite) extension of 1, we can still deﬁne
T ={˛ ÷ 1 [ [˛[ _1]
p ={˛ ÷ T [ [˛[ <1]
and call T¡p the residue ﬁeld of 1.
PROPOSITION 7.50 Let 1 be an algebraic extension of 1, and let l be the residue ﬁeld of
1. The map 1
t
÷k
t
sending an unramiﬁed extension 1
t
of 1 contained in 1 to its residue
ﬁeld k
t
is a onetoone correspondence between the sets
{1
t
c1, ﬁnite and unramiﬁed over 1] {k
t
cl, ﬁnite over k].
Moreover:
(a) if 1
t
k
t
and 1
tt
k
tt
, then 1
t
c1
tt
” k
t
ck
tt
:
(b) if 1
t
k
t
, then 1
t
is Galois over 1 if and only if k
t
is Galois over k, in which case
there is a canonical isomorphism
Gal(1
t
¡1) ÷Gal(k
t
¡k).
PROOF. Let k
t
be a ﬁnite extension of k. We can write it k
t
= kŒa. Let }
0
(X) be the
minimum polynomial of a over k, and let }(X) be any lifting of }
0
(X) to ·ŒX. As a is
a simple root of }
0
(X), Newton’s lemma (7.31) shows that there is a (unique) ˛ ÷ 1 such
that }(˛) =0 and ˛ ÷a mod p. Now 1
t
def
=1Œ˛ has residue ﬁeld k
t
. Thus 1
t
÷k
t
is
surjective. Suppose that 1
t
and 1
tt
are unramiﬁed extensions of 1 in 1 with the same
residue ﬁeld k
t
. Then 1
t
1
tt
is an unramiﬁed extension of 1 (see 6.5 and 6.6b) with
residue ﬁeld k
t
. Hence
Œ1
t
1
tt
: 1 =Œk
t
: k =Œ1
t
: 1.
and so 1
tt
=1
t
.
8
When k is not perfect, we should deﬁne 1¡1 to be unramiﬁed if (a) the ramiﬁcation index is 1, and (b)
the residue ﬁeld extension is separable. These conditions imply that 1¡1 is separable. With this deﬁnition,
(7.50) continues to hold without 1 and k being assumed to be perfect
122
Unramiﬁed extensions of a local ﬁeld
Statement (a) is obvious.
Assume 1
t
is Galois over 1; then Gal(1
t
¡1) preserves ·
t
(the valuation ring in 1
t
)
and its maximal ideal, and so we get a map Gal(1
t
¡1)
//
Aut(k
t
¡k). Write k
t
=kŒa,
and let g(X) ÷ ·ŒX be such that ¨ g(X) ÷ kŒX is the minimum polynomial of a. Let
˛ ÷ ·
t
be the unique root of g(X) such that ¨ ˛ = a. Because 1
t
is Galois over 1, g(X)
splits in ·
t
ŒX, and this implies that ¨ g(X) splits in k
t
ŒX, and so k
t
is Galois over k. Let
} =Œk
t
: k =Œ1
t
: 1, and let ˛
1
. . . . . ˛
(
be the roots of g(X). Then
{˛
1
. .... ˛
(
] ={o˛ [ o ÷ Gal(1¡1)].
Because ¨ g(X) is separable, the ˛
i
are distinct modulo p, and this shows that the image of the
map Gal(1
t
¡1)
//
Gal(k
t
¡k) has order } , and hence is an isomorphism. Conversely,
suppose k
t
¡k is Galois. Again write k
t
=kŒa, and ˛ ÷ ·
t
lift a. It follows from Hensel’s
lemma that ·
t
contains the conjugates of ˛, and hence that 1
t
is Galois over 1.
2
COROLLARY 7.51 There is a ﬁeld 1
0
c 1 containing all unramiﬁed extensions of 1 in
1 (called the largest unramiﬁed extension of 1 in 1). In fact, it is obtained from 1 by
adjoining all roots of 1 of order prime to the characteristic of k.
PROOF. This is an obvious consequence of the theorem.
2
COROLLARY 7.52 The residue ﬁeld of 1
al
is k
al
; there is a subﬁeld 1
un
of 1
al
such that
a subﬁeld 1 of 1
al
, ﬁnite over 1, is unramiﬁed if and only if 1c1
un
. (Recall that we are
assuming k and 1 to be perfect.)
PROOF. Let }
0
(X) be any polynomial in kŒX, and let }(X) be any lift of }
0
(X) to ·ŒX.
Then 1
al
contains all the roots of }(X), and so the residue ﬁeld k
t
of 1
al
contains all the
roots of }
0
(X). Hence k
t
is algebraic over k, and every polynomial in kŒX splits in k
t
, and
so it must be the algebraic closure of k.
2
REMARK 7.53 For those familiar with the language of category theory, we can be a little
more precise: there is an equivalence between the category of ﬁnite unramiﬁed extensions
of 1 and the category of ﬁnite (separable) extensions of k.
EXAMPLE 7.54 Let 1 be a local ﬁeld of characteristic zero (hence a ﬁnite extension of Q
;
for some ]), and let q be the order of the residue ﬁeld k of 1.
Recall from (FT 4.18) that, for each n, there is an extension k
n
of k of degree n, and
that k
n
is unique up to kisomorphism; it is the splitting ﬁeld of X
q
n
÷X. The Galois
group Gal(k
n
¡k) is a cyclic group of order n, having as canonical generator the Frobenius
element . ÷.
q
.
Therefore, for each n, there is an unramiﬁed extension 1
n
of 1 of degree n, and
it is unique up to 1isomorphism; it is the splitting ﬁeld of X
q
n
÷X; the Galois group
Gal(1
n
¡1) is a cyclic group of order n, having as canonical generator the Frobenius ele
ment o which is determined by the property
oˇ ÷ˇ
q
(mod p).
all ˇ ÷ T. (Here T is the discrete valuation ring in 1
n
, and p is the nonzero prime ideal in
B.)
123
7. VALUATIONS; LOCAL FIELDS
Totally ramiﬁed extensions of 1
Let 1 be a complete discretelyvalued nonarchimedean ﬁeld, and let ¬ be a local uniformiz
ing parameter for 1. A polynomial }(X) ÷ 1ŒX is said to be Eisenstein if it is Eisenstein
for the maximal ideal of the ring of integers in 1, i.e., if
}(X) =a
0
X
n
÷a
1
X
n1
÷ ÷a
n
. with [a
0
[ =1. [a
i
[ <1. [a
n
[ =[¬[.
Equivalently,
ord(a
0
) =0. ord(a
i
) >0. ord(a
n
) =1.
for the normalized additive valuation. Equivalently, the Newton polygon of }(X) has only
one segment, which has slope
1
n
, n = deg} . Eisenstein polynomials allow us to give an
explicit description of all totally ramiﬁed extensions of 1.
PROPOSITION 7.55 Let 1 be a ﬁnite extension of 1. Then 1¡1 is totally ramiﬁed if and
only if 1 =1Œ˛ with ˛ a root of an Eisenstein polynomial.
PROOF. =: Suppose 1=1Œ˛ with ˛ a root of an Eisenstein polynomial }(X) of degree
n. If ord is the extension of the normalized discrete (additive) valuation on 1 to 1, then
ord(˛) = 1¡n. This implies that the ramiﬁcation index of 1¡1 is _ n. But it can’t be
greater than n, and so it is exactly n — 1 is totally ramiﬁed over 1. (Compare the proof of
6.2.)
=: Suppose 1 is a totally ramiﬁed extension of 1 of degree n. Let ˛ be a generator
of the maximal ideal in the ring of integers in 1; thus ord(˛) = 1¡n if ord extends the
normalized discrete valuation on 1. The elements 1. ˛. .... ˛
n1
represent different cosets
of ord(1
×
) in ord(1
×
), and so it is impossible to have a nontrivial relation
a
0
÷a
1
˛ ÷ ÷a
n1
˛
n1
=0. a
i
÷ 1
(because of 7.11). Hence 1=1Œ˛. The elements 1. ˛. . . . . ˛
n1
. ˛
n
are linearly dependent
over 1, and so we have a relation:
˛
n
÷a
1
˛
n1
÷ ÷a
n
=0. a
i
÷ 1.
Applying (7.11) again, we see that the minimum ord of a summand must be attained for two
terms. The only way this can happen is if ord(a
i
) >0 for all i and ord(a
n
) =ord(˛
n
) =1,
i.e., if
¸
a
i
X
i
is an Eisenstein polynomial.
2
REMARK 7.56 Let 1 be a ﬁnite totally ramiﬁed extension of 1. Let · and T be the
discrete valuation rings in 1 and 1, and let ¬ and ˘ be a prime elements in · and T. I
claim that T = ·Œ˘. The argument is the same as in the proof of 6.2 (see also Exercise
61). Because T and · have the same residue ﬁeld,
·Œ˘ ÷˘T =T.
The discriminant of 1. ˘. ˘
2
. . . . is a unit×¬
n
for some m, and so
p
c
T c·Œ˘ cT
for some c. As before, these two conditions sufﬁce to imply that T =·Œ˘.
124
Ramiﬁcation groups
Ramiﬁcation groups
Let 1 be a ﬁnite Galois extension of 1, and assume that the residue ﬁeld k of 1 is perfect.
As we have noted, G
def
=Gal(1¡1) preserves the valuation on 1. In particular, it preserves
T ={˛ ÷ 1 [ [˛[ _1]. p ={˛ ÷ 1 [ [˛[ <1].
Let ˘ be a prime element of 1 (so that p = (˘)). We deﬁne a sequence of subgroups
G ¯G
0
¯G
1
¯ by the condition:
o ÷ G
i
” [o˛ ÷˛[ <[˘[
i
, all ˛ ÷ T.
The group G
0
is called the inertia group, the group G
1
is called the ramiﬁcation group,
and the groups G
i
, i >1, are called the higher ramiﬁcation groups of 1 over 1.
LEMMA 7.57 The G
i
are normal subgroups of G, and G
i
={1] for i large enough.
PROOF. For o. t ÷ G,
[t
1
ot˛ ÷˛[ =[o(t˛) ÷(t˛)[
(because [.[ = [t.[). As ˛ runs through T, so also does t˛, and so t
1
ot ÷ G
i
exactly
when o does. This proves that G
i
is normal.
If o =1, then o˛ =˛ for some ˛ ÷ T. Hence o … G
i
as soon as [o˛ ÷˛[ _[˘[
i
.
2
THEOREM 7.58 Let 1¡1 be a Galois extension, and assume that the residue ﬁeld exten
sion l¡k is separable.
(a) The ﬁxed ﬁeld of G
0
is the largest unramiﬁed extension 1
0
of 1 in 1, and
G¡G
0
=Gal(1
0
¡1) =Gal(l¡k).
(b) For i _1, the group
G
i
={o ÷ G
0
[ [o˘ ÷˘[ <[˘[
i
].
PROOF. (a) Let 1
0
be the largest unramiﬁed extension in 1 (see 7.51). Then o1
0
is also
unramiﬁed, and so it is contained in 1
0
. Thus 1
0
is Galois over 1, and the canonical map
Gal(1
0
¡1) ÷Gal(l¡k) is an isomorphism (see 7.50). By deﬁnition G
0
is the kernel of
G ÷Gal(l¡k), and so 1
0
is its ﬁxed ﬁeld.
(b) Let ·
0
be the discrete valuation ring in 1
0
. Then T =·
0
Œ˘ (by 7.56). Since G
0
leaves ·
0
ﬁxed, in order to check that o ÷ G
i
it sufﬁces to check that [o˛ ÷˛[ < [˘[
i
for
the element ˛ =˘.
2
COROLLARY 7.59 We have an exhaustive ﬁltration G ¯G
0
¯ such that
G¡G
0
=Gal(l¡k):
G
0
¡G
1
·÷l
×
:
G
i
¡G
i÷1
·÷l.
Therefore, if k is ﬁnite, then Gal(1¡1) is solvable.
125
7. VALUATIONS; LOCAL FIELDS
PROOF. Let o ÷ G
0
; then o˘ is also a prime element and so o˘ =u˘ with u a unit in
T. The map o ÷u mod p is a homomorphism G
0
÷l
×
with kernel G
1
.
Let o ÷ G
i
. Then [o˘ ÷˘[ _ [˘[
i÷1
, and so o˘ = ˘ ÷a˘
i÷1
some a ÷ T. The
map o ÷a (mod p) is a homomorphism G
i
÷l with kernel G
i÷1
.
2
An extension 1¡1 is said to be wildly ramiﬁed if ][e where ] =char(k). Otherwise
it is said to be tamely ramiﬁed. Hence for a Galois extension
1¡1 is unramiﬁed ” G
0
={1].
and
1¡1 is tamely ramiﬁed ” G
1
={1].
Krasner’s lemma and applications
Again let 1 be complete with respect to a discrete nonarchimedean valuation [ [, and extend
the valuation (uniquely) to a valuation on 1
al
. It is clear from our discussion of unramiﬁed
extensions of 1 that roots of distinct polynomials }(X) and g(X) will often generate the
same extension of 1; in fact, this will be true if
¨
} = ¨ g and both are irreducible in kŒX.
Krasner’s lemma and its consequences show that the roots of two polynomials will generate
the same extension if they are sufﬁciently close.
PROPOSITION 7.60 (KRASNER’S LEMMA) Let ˛. ˇ ÷1
al
, and assume that ˛ is separable
over 1Œˇ. If ˛ is closer to ˇ than to any conjugate of ˛ (over 1), then 1Œ˛ c1Œˇ.
PROOF. Let o be an embedding of 1Œ˛. ˇ into 1
al
ﬁxing 1Œˇ. By Galois theory, it
sufﬁces to show that o˛ =˛. But
[o˛ ÷ˇ[ =[o˛ ÷oˇ[ =[˛ ÷ˇ[
because oˇ =ˇ and [o +[ =[ +[. Hence
[o˛ ÷˛[ =[o˛ ÷ˇ÷ˇ÷˛[ _[˛ ÷ˇ[.
Since o˛ is a conjugate of ˛ over 1, the hypothesis now implies that o˛ =˛.
2
Now assume 1 has characteristic zero (to avoid complications). As before, for h(X) =
¸
c
i
X
i
, we deﬁne h =max{[c
i
[]. Note that if h(X) varies in a family of monic polyno
mials for which h remains bounded, then the maximum value of a root of h is bounded;
in fact, if
¸
c
i
ˇ
i
=0.
we must have [ˇ
n
[ _[c
}
ˇ
}
[ for some ; <n, and so [ˇ[
n}
_[c
}
[.
Fix a monic irreducible polynomial }(X) in 1ŒX, and let
}(X) =
¸
(X ÷˛
i
). ˛
i
÷ 1
al
.
The ˛
i
must be distinct. Let g(X) be a second monic polynomial in 1ŒX, and suppose
that } ÷g is small. For any root ˇ of g(X), [}(ˇ)[ = [(} ÷g)(ˇ)[ is small (because
} ÷g small implies that g is bounded, and hence [ˇ[ is bounded). But
[}(ˇ)[ =
¸
[ˇ÷˛
i
[.
126
Krasner’s lemma and applications
In order for this to be small, at least one term [ˇ ÷˛
i
[ must be small. By taking } ÷g
small enough, we can force ˇ to be closer to one root ˛
i
than ˛
i
is to any other ˛
}
. That is,
we can achieve:
[ˇ÷˛
i
[ <[˛
i
÷˛
}
[, all ; =i.
In this case, we say that ˇ belongs to ˛
i
. Krasner’s lemma then says that 1Œ˛
i
 c 1Œˇ,
and because } and g have the same degree, they must be equal. We have proved:
PROPOSITION 7.61 Let }(X) be a monic irreducible polynomial of 1ŒX. Then any
monic polynomial g(X) ÷ 1ŒX sufﬁciently close to }(X) is also irreducible, and each
root ˇ of g(X) belongs to some root ˛ of }(X). For such a root 1Œ˛ =1Œˇ.
COROLLARY 7.62 Let 1 be a ﬁnite extension of Q
;
. Then there is a ﬁnite extension 1 of
Q contained in 1 such that Œ1: Q =Œ1: Q
;
 and 1 Q
;
=1.
PROOF. Write 1 = Q
;
Œ˛, and let }(X) be the minimum polynomial of ˛ over Q
;
.
Choose g(X) ÷ QŒX sufﬁciently close to }(X), and let 1 = QŒˇ for ˇ a root of g(X)
belonging to ˛.
2
Fix a monic polynomial } in 1ŒX, and let ˛
1
. ˛
2
. . . . be its roots in 1
al
. As a second
monic polynomial g in 1ŒX approaches } , each root ˇ
i
of g approaches some root ˛
}(i)
of
} , and the function i ÷;(i ) doesn’t change once g is close. Let }
x
(X) be the polynomial
with roots the ˛
}(i)
(possibly with repetitions). Then, when g is close to } , it is close to }
x
because each of its roots is close to the corresponding root of }
x
. But if we choose g to be
closer to } than } is to any possible }
x
, this will be impossible. We have proved:
PROPOSITION 7.63 Assume 1 is of characteristic zero. If two monic irreducible polyno
mials } and g are sufﬁciently close, then each root of g will belong to exactly one root of
} , and so
{1Œ˛ [ ˛ a root of } ] ={1Œˇ [ ˇ a root of g].
PROPOSITION 7.64 Assume 1 has characteristic zero and has ﬁnite residue ﬁeld. Then,
up to isomorphism, there are only ﬁnitely many totally ramiﬁed extensions of Q
;
of a given
degree.
PROOF. We ﬁx an n and show that there are only ﬁnite many extensions of degree _ n.
Each point of
(a
1
. .... a
n
) ÷ p×p×p× ×·
×
¬
deﬁnes an Eisenstein polynomial of degree n, namely,
}(X) =X
n
÷a
1
X
n1
÷ ÷a
n
.
and hence a ﬁnite set of totally ramiﬁed extensions of degree n, namely, those generated by
the roots of }(X). According to the last proposition, each point of p ×p ×p × ×·
×
¬
has a neighbourhood such that the points in the neighbourhood all give the same extensions
of 1. In (7.47) we showed that the factors of p×p×p× ×·
×
¬ are compact, hence the
product is compact, and so a ﬁnite number of these neighbourhoods will cover it.
2
REMARK 7.65 We proved above that
127
7. VALUATIONS; LOCAL FIELDS
(a) every ﬁnite extension 1 of 1 contains a largest unramiﬁed extension of 1;
(b) for each m_1, there is an unramiﬁed extension of degree m of 1, and any two such
extensions are 1isomorphic.
Fix an n; then each extension 1 of 1 of degree n can be realized as a totally ramiﬁed
extension of degree n¡mof the (unique) unramiﬁed extension of degree m, some mdividing
n. Clearly there are only ﬁnitely many such 1’s (up to 1isomorphism).
Exercises
71 Let [ [
1
, . . . , [ [
n
be the valuations on a number ﬁeld 1 corresponding to distinct
prime ideals p
i
, and let a
1
. . . . . a
n
be elements of 1. Let J be a common denominator for
the a
i
(so that Ja
i
÷ O
1
). Show that, for any c > 0, there is an element a ÷ 1 such that
[a ÷a
i
[
i
< c for i =1. . . . . n and [a[ _ 1¡[J[ for all valuations [ [ corresponding to prime
ideals other than the p
i
.
Hint: Apply the Chinese Remainder Theorem to the Ja
i
.
72 Let [ [ be nonarchimedean valuation on a ﬁeld 1.
(a) Deﬁne an open disk with radius r and centre a to be
D(a. r) ={. ÷ 1 [ [. ÷a[ <r].
Prove that D(a. r) =D(b. r) for any b ÷ D(a. r). Deduce that if two disks meet, then the
large disk contains the smaller.
(b) Assume 1 to be complete. Show that the series
¸
a
n
converges if and only if
a
n
÷0.
(This problem illustrates the weirdness of the topology deﬁned by a nonarchimedean valu
ation.)
73 For which a ÷ Z is 7X
2
=a solvable in Z
T
? For which a ÷ Q is it solvable in Q
T
?
74 (a) Show that (X
2
÷2)(X
2
÷17)(X
2
÷34) has a root in Z
;
for every ].
(b) Show that 5X
3
÷7X
2
÷3X÷6 has a root ˛ in Z
T
with [˛÷1[
T
<1. Find an a ÷ Z
such that [˛ ÷a[
T
_7
4
.
75 Find all the quadratic extensions of Q
2
. Hint: there are exactly 7 (up to isomorphism).
76 Let ]
1
. . . . . ]
n
be distinct prime numbers, and let ˛
i
=
]. Let 1 =QŒ˛
1
. . . . . ˛
n
.
Show that Œ1: Q =2
n
. Let , =
¸
˛
i
. Show that 1 =QŒ,, and deduce that the minimum
polynomial }(X) of , over Q has degree 2
n
. Show that }(X) factors in Z
;
ŒX into a
product of polynomials of degree _4 (] =2) or of degree _8 (] =2).
77 Fix an algebraic closure Q
al
;
of Q
;
, and for each n prime to ], let ¯
n
be a primitive
nth root of 1. Show that a ﬁnite extension 1 of Q
;
can contain only ﬁnitely many ¯
n
’s.
Deduce that the Cauchy sequence
¸
¯
n
]
n
does not converge to an element of Q
al
;
.
128
Exercises
78 (a) Find two monic polynomials of degree 3 in Q
5
ŒX with the same Newton polygon,
but with one irreducible and the other not.
(b) Find a monic irreducible polynomial in ZŒX of degree 6 which factors in Q
5
ŒX
into a product of 3 irreducible polynomials of degree 2.
129
CHAPTER 8
Global Fields
A global ﬁeld an algebraic number ﬁeld (ﬁnite extension of Q) or a function ﬁeld in one
variable over a ﬁnite ﬁeld (ﬁnite extension of F
q
(T ) for some q). We are mainly interested
in the number ﬁeld case.
Extending valuations
Let 1 be a ﬁeld with a valuation [ [ (archimedean or discrete nonarchimedean), and let 1
be a ﬁnite separable extension of 1. When 1 is complete, we know that there is a unique
extension of [ [ to 1 (see 7.38, 7.39), and we want to understand the extensions when 1 is
not complete.
Write 1 = 1Œ˛, and let }(X) be the minimum polynomial of ˛ over 1. Let [ [
t
be
an extension of [ [ to 1. Then we can form the completion
´
1 of 1 with respect to [ [
t
, and
obtain a diagram:
1
´
1
//
1
1
1
´
1
// ´
1
´
1
Then
´
1 =
´
1Œ˛ because
´
1Œ˛ is complete, being ﬁnite over
´
1, and contains 1. Let g(X)
be the minimum polynomial of ˛ over
´
1. Since }(˛) =0, g(X)[}(X), and so with each
extension of [ [, we have associated an irreducible factor of }(X) in
´
1ŒX.
Conversely, let g(X) be a monic irreducible factor of }(X) in
´
1ŒX, and let
´
1Œ. =
´
1ŒX¡(g(X)). Then we obtain a diagram:
1
´
1
//
1
1
1
´
1Œ.
˛x
// ´
1Œ.
´
1
According to (7.38, 7.39), the valuation on
´
1 extends uniquely to
´
1Œ., and this induces a
valuation on 1 extending [ [.
These two operations are inverse, and so we have proved the following result:
130
Extending valuations
PROPOSITION 8.1 Let 1 = 1Œ˛ be a ﬁnite separable extension of 1, and let }(X) be
the minimum polynomial of ˛ over 1. Then there is a natural onetoone correspondence
between the extensions of [ [ to 1 and the irreducible factors of }(X) in
´
1ŒX.
There is a more canonical way of obtaining the completions of 1 for the various exten
sions of [ [.
PROPOSITION 8.2 Let [ [ be a valuation on 1 (archimedean or discrete nonarchimedean)
and let 1 be a ﬁnite separable extension of 1. Let
´
1 be the completion of 1 with respect
to [ [. Then [ [ has ﬁnitely many extensions [ [
1
. . . . . [ [
v
to 1; if 1
i
denotes the completion
of 1 with respect to the valuation [ [
i
, then
1˝
1
´
1 .
¸
v
i=1
1
i
. (19)
PROOF. Since 1 is separable over 1, 1 =1Œ˛ .1ŒX¡(}(X)) for a primitive element
˛ ÷ 1 and its minimum polynomial }(X). Suppose }(X) factors in
´
1ŒX as
}(X) =}
1
(X) }
2
(X) }
v
(X)
with }
i
(X) monic and irreducible. Then (see 1.18)
1˝
1
´
1 =1Œ˛ ˝
1
´
1 ~
´
1ŒX¡((}(X)) .
¸
´
1ŒX¡(}
i
(X))
and so the proposition follows from (8.1). Denote the canonical map from 1 into its com
pletion by a ÷a
i
, and denote the canonical extension of 1 ÷1
i
to
´
1 by b ÷b; then the
map (19) is a˝b ÷(a
1
b. . . . . a
v
b).
2
REMARK 8.3 Suppose now that 1 is a number ﬁeld, that O
1
=O
1
Œ˛, and that [ [ =[ [
p
for some prime ideal p in O
1
. Because }
i
(X) is irreducible in
´
1ŒX, Hensel’s lemma
shows that, modulo ´ p, }
i
(X) is a power of an irreducible polynomial, say,
¨
}
i
(X) =g
i
(X)
e
i
.
Then
¨
} (X) =
¸
v
i=1
g
i
(X)
e
i
.
and (3.41) tells us that
pO
1
=
¸
P
e
i
i
. P
i
=(p. g
i
(˛)).
The valuations extending [ [
p
correspond to the primes P
i
, and so the two descriptions of
the extensions agree.
COROLLARY 8.4 In the situation of the Proposition, for any element ˛ ÷ 1.
Nm
1¡1
(˛) =
¸
Nm
1
i
¡
¨
1
(˛). Tr
1¡1
(˛) =
¸
Tr
1
i
¡
¨
1
(˛).
(in the i th factor or summand on the right, ˛ is regarded as an element of 1
i
).
PROOF. By deﬁnition the norm and trace of ˛ are the determinant and trace of the 1linear
map . ÷˛.: 1 ÷1. These don’t change when 1 is tensored with
´
1, and it easy to see
that norms and traces in products break up into products and sums respectively.
2
131
8. GLOBAL FIELDS
EXAMPLE 8.5 According to PARI
}(X) =X
6
÷5X
5
÷5X
3
÷25X ÷125
is irreducible in QŒX. Its Newton polygon for ord
5
has three segments of .lengths 3, 2, 1
respectively, and so it has at least three factors in Q
5
. The discriminant of }(X) is
2
4
5
11
(59)(365587).
and so according to (7.37), to ﬁnd the number of factors of }(X) in Q
5
ŒX, it sufﬁces to
factor in modulo 5
11
. Better, according to Pari, }(X) has exactly 3 irreducible factors in
Q
5
ŒX, namely,
X ÷
5÷4 5
2
÷2 5
3
÷O(5
4
)
X
2
÷
3 5
2
X ÷(5÷5
2
÷3 5
3
) ÷O(5
4
)
X
3
÷(3 5
2
÷5
3
)X
2
÷
4 5÷3 5
2
X ÷5÷O(5
4
)
(Type factorpadic(f,p,r) where r is the precision required.)
Suppose have a factorization
}(X) =}
1
(X)}
2
(X)}
3
(X)
(to whatever degree of accuracy we wish). To compute [ˇ[
i
, map ˇ =
¸
c
}
˛
}
to ˇ
i
=
¸
c
}
˛
}
i
÷ 1
i
def
=Q
5
Œ˛
i
, ˛
i
a root of }
i
(X), and use that
[ˇ[
i
=[ˇ
i
[
i
=[ Nm
1
i
¡Q
5
ˇ[
1¡deg(
i
i
.
The product formula
Before proving the product formula for a number ﬁeld, we need one extra fact for local
ﬁelds.
Let 1 be a local ﬁeld with normalized valuation [ [. Recall that this means that [ [
is the usual absolute value if 1 is 1, the square of the usual valuation if 1 is C, and
[a[ =(1¡Np)
ord(o)
if the valuation is deﬁned by a prime ideal p.
Let 1 be a ﬁnite separable extension of 1, and let [ [ be the unique extension of [ [ to
1. Let   be the normalized valuation on 1 corresponding to [ [. What is the relation of
  to [ [?
LEMMA 8.6 In the above situation, a =[a[
n
, where n =Œ1: 1.
PROOF. When 1 is archimedean, there are only two cases to consider, and both are obvi
ous. Thus, assume 1 is nonarchimedean. Since, by assumption,   =[ [
c
for some c, we
only have to check that the formula holds for a prime element ¬ of 1. Let ˘ be a prime
element of 1, and let P=(˘) and p =(¬); then ¬ =(unit) ×˘
e
, and so
¬ =˘
e
 =(1¡NP)
e
=(1¡Np)
e(
=[¬[
n
.
as required.
Alternatively, use (7.43). For a ÷ 1, we have
a
def
=NP
ord
L
o
(T.43)
= (Np
(
)
e·ord
K
o
=[a[
e(
=[a[
n
.
2
132
The product formula
PROPOSITION 8.7 Let 1¡1 be a ﬁnite extension of number ﬁelds. For any prime · of 1
and ˛ ÷ 1.
¸
u¡
˛
u
=Nm
1¡1
˛
¡
.
Here  
u
and  
¡
denote the normalized valuations for the primes n and ·.
PROOF. Let [ [
i
, i =1. 2. . . . . g, be the extensions of  
¡
to 1, and let  
i
be the normal
ized valuation corresponding to [ [
i
. Then
Nm
1¡1
˛
¡
S.4
= 
¸
v
i=1
Nm
1
i
¡
¨
1
˛
¡
=
¸
v
i=1
Nm
1
i
¡
¨
1
˛
¡
T.3S
=
¸
v
i=1
[˛[
n(i)
i
S.6
=
¸
v
i=1
˛
u
.
where n
i
=Œ1
i
:
´
1.
2
THEOREM 8.8 (PRODUCT FORMULA) Let 1 be an algebraic number ﬁeld; for all nonzero
˛ ÷ 1.
¸
u
˛
u
=1.
where the product is over the primes of 1 and  
u
is the normalized valuation for the
prime n.
PROOF. We have
¸
u
˛
u
=
¸
¡
¸
u¡
˛
u
(8.7)
=
¸
¡
Nm
1¡Q
˛
¡
where · runs through the primes 2. 3. 5. 7. .... oof Q. The last product is 1 by (7.13).
2
ASIDE 8.9 E. Artin and Whaples (1946)
1
proved that global ﬁelds can be characterized axiomat
ically. Let 1 be a ﬁeld with a set V of primes (equivalence classes of valuations) satisfying the
following axioms.
AXIOM I. There is a set of representatives [ [
¡
for the primes such that, for any nonzero a ÷ 1,
[a[
¡
=1 for only ﬁnitely many · and
¸
¡
[a[
¡
=1 (product over all · ÷ V).
AXIOM II. There exists at least one prime · for which 1
¡
is a local ﬁeld.
Then 1 is a global ﬁeld, and Vconsists of all the primes for 1. They then derived the main theorems
(unit theorem and ﬁniteness of the class number) directly from the axioms, thereby avoiding the use
of either ideal theory or the Minkowski theory of lattice points.
Throughout his career, E. Artin promoted the idea that if only one could understand the similar
ities between function ﬁelds and number ﬁelds sufﬁciently well, then one could transfer proofs from
function ﬁelds to number ﬁelds (e.g. the proof of the Riemann hypothesis!). This hasn’t worked
as well as he hoped, but the analogy has still been very fruitful. In the above paper, he suggested
one should develop number theory and class ﬁeld theory as much as possible working only from the
axioms.
1
Artin, Emil; Whaples, George. Axiomatic characterization of ﬁelds by the product formula for valuations.
Bull. Amer. Math. Soc. 51, (1945). 469–492. Reprinted in: Artin, Emil. Exposition by Emil Artin: a selection.
Edited by Michael Rosen. History of Mathematics, 30. American Mathematical Society, Providence, RI;
London Mathematical Society, London, 2007. x+346 pp.
133
8. GLOBAL FIELDS
Decomposition groups
Let 1 be a ﬁnite Galois extension of a number ﬁeld 1, and let G = Gal(1¡1). For a
valuation n of 1, we write on for the valuation such that [o˛[
cu
= [˛[
u
, i.e., [˛[
cu
=
[o
1
˛[
u
. For example, if n is the prime deﬁned by a prime ideal P, then on is the prime
deﬁned by the prime ideal oP, because
[˛[
cu
<1 ” o
1
˛ ÷ P ” ˛ ÷ oP.
The group G acts on the set of primes of 1 lying over a ﬁxed prime · of 1, and we deﬁne
the decomposition (or splitting) group of n to be the stabilizer of n in G; thus
G
u
={o ÷ G [ on =n].
Equivalently, G
u
is the set of elements of G that act continuously for the topology deﬁned
by [ [
u
. Each o ÷ G
u
extends uniquely to a continuous automorphism of 1
u
. Note that
G
ru
=tG
u
t
1
.
PROPOSITION 8.10 The homomorphism G
u
÷Gal(1
u
¡1
¡
) just deﬁned is an isomor
phism.
PROOF. Clearly the map is injective, and so (G
u
: 1) _ Œ1
u
: 1
¡
. The valuation on has
decomposition group oG
u
o
1
, which has the same order as G
u
, and so we also have
(G
u
: 1) _Œ1
cu
: 1
¡
. The number of distinct ns dividing · is (G : G
u
), and so
(G : 1) =(G : G
u
)(G
u
: 1) _
¸
c÷G¡G
w
Œ1
cu
: 1
¡

(S.2)
_ Œ1 : 1.
Hence equality holds: (G
u
: 1) =Œ1
u
: 1
¡
 (and G acts transitively on the primes dividing
·, which we knew already from the proof of 3.34).
2
Let D(P) (or G(P)) be the decomposition group of P, so that D(P) =Gal(1
P
¡1
p
),
and let 1(P) cD(P) be the inertia group. We have the following picture:
p 1
P
T
1
T(P)
P
J
1
J(P)
P 1
1
1
T(P)
v
1
T(P)
1
J(P)
(
1
J(P)
1
e
1
1
p
1
T(P)
1
p
1
J(P)
1
J(P)
P
1 1
P
1
p
1
J(P)
P
(
1
J(P)
P
1
P
e
1
P
l
?
?
?
?
?
?
?
?
?
?
?
?
1
p
k
1
J(P)
P
l
k
l
T(P)¡J(P)
Here:
P
J
=P¨1
J(P)
, P
T
=P¨1
T(P)
, p =P¨1:
the ﬁelds in the second column are the completions of those in the ﬁrst;
the ﬁelds in the third column are the residue ﬁelds of those in the second.
134
Decomposition groups
PROPOSITION 8.11 (a) The only prime ideal of 1 lying over P
T
is P.
(b) The prime ideal P
T
is unramiﬁed in 1
J
, and }(P
J
¡P
T
) =}(P¡p).
(c) The prime ideal P
J
is totally ramiﬁed in 1, and e(P¡P
J
) =e(P¡p).
(d) If D(P) is normal in G, then
pO
1
D =
¸
oP
T
where the product is over a set of representatives for G¡D(P).
PROOF. (a) Because 1is Galois over 1
T(P)
, its Galois group D(P) acts transitively on the
set of prime ideals of 1 lying over P
T
. Thus (a) is obvious from the deﬁnition of D(P).
(b), (c), (d) are similarly straightforward.
2
The diagram, and the proposition, show that we can construct a chain of ﬁelds
1 ¯1
J
¯1
T
¯1
such that all the ramiﬁcation of P over p takes place in the top extension, all the residue
ﬁeld extension takes place in the middle extension, and, when 1
T
is normal over 1, all the
splitting takes place in the bottom extension. One should be a little careful about the last
assertion when D(P) is not normal in G; all we know in general is that
p O
1
D =
¸
P
e
i
i
, P
1
=P
T
with e
1
=1 =}
1
(i.e., in general p will not split completely in 1
T
).
REMARK 8.12 Let 1 be a Galois extension of Q, with Galois group G. Suppose that
O
1
=ZŒ˛ for some ˛ ÷ 1. Let }(X) be the minimum polynomial of ˛ over Q, and write
¨
} (X) for }(X) modulo ]. Choose an irreducible factor g
1
(X) of
¨
} (X), and let g
1
(X)
e
1
be the largest power of g
1
(X) dividing
¨
} (X). According to Hensel’s lemma, g
1
(X)
e
1
lifts
to an irreducible factor }
1
(X) of }(X) in Q
;
ŒX, which can be found to any desired degree
of accuracy by factoring }(X) modulo a high power of ] (essentially using the method of
proof of Hensel’s lemma). Let P
1
=(]. h
1
(˛)) for any lifting h
1
of g
1
to ZŒX. Then
D(P
1
) ={o ÷ G [ oP
1
=P
1
].
which can be computed easily (provided G has been found explicitly as a subgroup of the
symmetric group on the set of roots of }(X)). Let ¨ ˛ be the image of ˛ in O
1
¡P
1
=F
;
Œ ¨ ˛.
Then g
1
(X) is the minimum polynomial of ¨ ˛ over F
;
, and 1(P
1
) is the subgroup of D(P
1
)
ﬁxing ¨ ˛. Finally D(P
1
)¡1(P
1
) =Gal(F
;
Œ ¨ ˛¡F
;
).
Consider a tower of ﬁelds
1
1 1
M
1
1 p
1 P
1
M P
G
Assume M is Galois over 1 with Galois group G, and that H is the subgroup of G ﬁxing
1. (Recall D(P) and G(P) are two notations for the same object.)
135
8. GLOBAL FIELDS
PROPOSITION 8.13 Let P be a prime ideal in O
Æ
, and let P
1
=P¨1.
(a) The decomposition group H(P) of P over 1 is G(P) ¨H.
(b) Suppose further that H is a normal subgroup of G, so that G¡H is the Galois group
of 1¡1. The decomposition group of P
1
over 1 is the image of G(P) in G¡H.
PROOF. (a) Clearly
H(P) ={o ÷ G [ o ÷ H. oP=P] =H ¨G(P).
(b) This is equally obvious.
2
The Frobenius element
Let 1¡1 be a Galois extension of number ﬁelds with Galois group G. Given an ideal P
of 1 that is unramiﬁed in 1¡1 we deﬁne the Frobenius
2
element o =(P. 1¡1) to be the
element of G(P) that acts as the Frobenius automorphism on the residue ﬁeld. Thus o is
uniquely determined by the following two conditions:
(a) o ÷ G(P), i.e., oP=P:
(b) for all ˛ ÷ O
1
, o˛ ÷˛
q
mod P, where q is the number of elements the residue ﬁeld
O
1
¡p, p =P¨1.
We now list the basic properties of (P. 1¡1).
8.14 Let tP be a second prime dividing p, t ÷ G. Then G(tP) =tG(P)t
1
, and
(tP. 1¡1) =t(P. 1¡1)t
1
.
PROOF. Let ˛ ÷ O
1
; then
tot
1
(˛) =t((t
1
˛)
q
÷a), some a ÷ P, and
t((t
1
˛)
q
÷a) =˛
q
÷ta ÷˛
q
mod tP.
2
Thus if Gal(1¡1) is abelian, then (P. 1¡1) =(P
t
. 1¡1) for all primes P, P
t
dividing
p, and we write (p. 1¡1) for this element. If Gal(1¡1) is not abelian, then
{(P. 1¡1) [ P[p]
is a conjugacy class in G, which (by an abuse of notation) we again denote (p. 1¡1).
Thus, for a prime p of 1, (p. 1¡1) is either an element of Gal(1¡1) or a conjugacy class
depending on whether Gal(1¡1) is abelian or nonabelian.
2
Here is a direct proof of the existence of the Frobenius element. Let 1¡1 be a ﬁnite Galois extension
of number ﬁelds with Galois group G, and let P be a prime ideal of O
1
(not necessarily unramiﬁed). By the
Chinese remainder theorem, there exists an element ˛ of O
1
such that ˛ generates the group (O
1
¡P)
×
and
lies in tPfor all t … G(P). Let J(X) =
¸
r÷G
(X÷t˛). Then J(˛) ÷0 mod P, and so J(˛
q
) ÷J(˛)
q
÷0
mod P. Therefore ˛
q
÷o˛ mod P for some o ÷ G. If o … G(P), then o
1
P=P, and so ˛ ÷ o
1
P; but
then ˛
q
÷o˛ ÷0 mod P, which is a contradiction. Thus o ÷ G(P). Every element , of O
1
can be written
, =˛
i
÷ˇ, with ˇ ÷ P, and so
o, ÷o(˛
i
) ÷˛
iq
÷,
q
mod P.
136
Examples
8.15 Consider a tower of ﬁelds
1
11
M
1 p
1 P
M Q
and assume that Q is unramiﬁed over p; then
(Q. M¡1)
((P¡p)
=(Q. M¡1).
PROOF. Let k(Q) ¯ k(P) ¯ k(p) be the corresponding sequence of residue ﬁelds. Then
}(P¡p) =Œk(P) : k(p), and the Frobenius element in Gal(k(Q)¡k(P)) is the }(P¡p)th
power of the Frobenius element in Gal(k(Q)¡k(p)).
2
8.16 In (8.15), assume that 1 is Galois over 1; then
(Q. M¡1)[1 =(P. 1¡1).
PROOF. Obvious.
2
Let 1
1
and 1
2
be Galois extensions of 1 contained in some ﬁeld ˝, and let M =
1
1
1
2
. Then M is Galois over 1, and there is a canonical homomorphism
o ÷(o[1
1
. o[1
2
): Gal(M¡1)
//
Gal(1
1
¡1) ×Gal(1
2
¡1)
which is injective.
8.17 Under the above map,
(Q. M¡1) ÷(P
1
. 1
1
¡1) ×(P
2
. 1
2
¡1).
PROOF. This follows from (8.16).
2
Note that p splits completely in 1if and only if (P. 1¡1) =1 for one (hence all) primes
P lying over it. Hence, in the situation of (8.17), p splits completely in M if and only if it
splits completely in 1
1
and 1
2
.
Examples
We ﬁnd the Frobenius maps for quadratic and cyclotomic ﬁelds, and obtain a surprisingly
simple proof of the quadratic reciprocity law.
EXAMPLE 8.18 Let 1 = QŒ¯
n
, where ¯
n
is a primitive nth root of 1. If ][n then ]
ramiﬁes in 1, and (]. 1¡Q) is not deﬁned. Otherwise o =(]. 1¡Q) is the unique element
of Gal(1¡Q) such that
o˛ ÷˛
;
mod p. for all ˛ ÷ ZŒ¯
n
.
137
8. GLOBAL FIELDS
for any prime ideal p lying over ].
I claim that o is the element of the Galois group such that o(¯
n
) =¯
;
n
: let p be a prime
lying over ] in ZŒ¯
n
; then modulo p, we have,
o(
¸
a
i
¯
i
n
) =
¸
a
i
¯
i;
n
÷
¸
a
;
i
¯
i;
n
÷(
¸
a
i
¯
i
n
)
;
as required.
Note that (]. 1¡Q) has order } where } is the smallest integer such that n[]
(
÷1
(because this is the order of ] in (Z¡(n))
×
).
EXAMPLE 8.19 Let 1 = QŒ
J, and let ] be a prime that is unramiﬁed in 1. Identify
Gal(1¡Q) with {˙1]. Then (]. 1¡Q) =÷1 or ÷1 according as ] does, or does not, split
in 1, i.e., according as J is, or is not, a square modulo ]. Thus (]. 1¡Q) =
d
;
.
APPLICATION: THE QUADRATIC RECIPROCITY LAW
Let 1 =QŒ¯, where ¯ is a primitive ]th root of 1, ] =2. Because Gal(1¡Q) .(Z¡]Z)
×
is cyclic of order ]÷1, it contains a unique subgroup of order (]÷1)¡2 (consisting of the
elements of (Z¡]Z)
×
that are squares), and hence 1 contains a unique quadratic extension
J of Q. If ] ÷ 1 mod 4, then ] is the only prime ramifying in QŒ
], and QŒ
] is
the only quadratic ﬁeld for which this is true. Similarly if ] ÷ 3 mod 4, then ÷] ÷ 1
mod 4, and ÷] is the only prime ramifying in QŒ
÷]. Thus J = QŒ
J where J =
(÷1)
(;1)¡2
].
If q is an odd prime =]; then
(q. 1¡Q)(¯) =¯
q
.
Thus (q. 1¡Q) restricts to the identity element of Gal(QŒ
J¡Q) or not according as q is a
square in (Z¡]Z)
×
or not. Thus (q. 1¡Q)[QŒ
J =(
q
;
). But we know that it is also equal
to (
d
q
). Hence
q
]
=
÷1
]
(;1)¡2
]
q
=(÷1)
(;1)(q1)¡4
]
q
.
Here we have used that ÷1 is square in F
q
if and only if 4[q÷1, so that
1
q
=(÷1)
(q1)¡2
.
The displayed formula, together with the equalities
÷1
]
=(÷1)
(;1)¡2
=
1 if ] ÷1 mod 4
÷1 if ] ÷1 mod 4
2
]
=(÷1)
(;
2
1)¡S
=
1 if ] ÷˙1 mod 8
÷1 if ] ÷˙5 mod 8.
constitutes the quadratic reciprocity law. We have already proved the ﬁrst equality, and the
second can be proved as follows. Let ¯ be a primitive 8th root of 1 in an algebraic closure
of F
;
, and let a =¯ ÷¯
1
. From ¯
4
=÷1, we see that
X
4
÷1 =(X
2
÷¯
2
)(X
2
÷¯
2
) in F
;
ŒX
because the roots of both polynomials are ˙¯, ˙¯
1
. Therefore, ¯
2
÷¯
2
= 0, and so
a
2
= 2. When ] ÷ ˙1 mod 8, ¯
;
÷¯
;
= ¯ ÷¯
1
, i.e., a
;
= a, and so 1 = a
;1
=
2
(;1)¡2
=
2
;
. When ] ÷˙5 mod 8, ¯
;
÷¯
;
=¯
5
÷¯
5
=÷(¯ ÷¯
1
), i.e., a
;
=÷a,
and so ÷1 =a
;1
=2
(;1)¡2
=
2
;
.
138
Computing Galois groups (the hard way)
Computing Galois groups (the hard way)
Let }(X) be a polynomial over a ﬁeld 1, and let ˛
1
. . . . . ˛
n
be the roots of }(X) in 1
al
.
We want to determine the Galois group of } as a subgroup of the group of permutations S
n
of {˛
1
. . . . . ˛
n
].
Introduce variables t
1
. . . . . t
n
. For any o ÷ S
n
and polynomial }(t
1
. . . . . t
n
), deﬁne
o
t
} =}(t
c(1)
. . . . . t
c(n)
). Let 0 =
¸
˛
i
t
i
, and deﬁne a polynomial
J(X. t ) =
¸
(X ÷o
t
0) (product over o ÷ S
n
).
The coefﬁcients of this polynomial are symmetric polynomials in the ˛
i
, and so lie in 1.
Now factor
J(X. t ) =J
1
(X. t ) J
i
(X. t )
in 1ŒX. t
1
. . . . . t
n
.
THEOREM 8.20 Let G be the set of o ÷S
n
such that o
t
ﬁxes J
1
(X. t ); then G is the Galois
group of }.
PROOF. See van der Waerden, Algebra, Vol 1, :61 (Calculation of the Galois group).
2
This theorem gives an algorithm (unfortunately impractical) for computing the Galois
group of a polynomial }(X) ÷ QŒX. We may suppose }(X) to be monic with integer
coefﬁcients. First ﬁnd the roots of }(X) to a high degree of accuracy. Then compute
J(X. t ) exactly, noting that this has coefﬁcients in Z. Factor J(X. t ), and take one of the
factors J
1
(X. t ). Finally list the elements o of S
n
such that o
t
ﬁxes J
1
(X. t ). The problem
with this approach is that J(X. t ) has degree nŠ. It will probably work (on a computer) if
n _ 5, but otherwise it is like trying to compute a determinant directly from the deﬁnition
as a sum of products.
Computing Galois groups (the easy way)
We now give a more practical procedure (also largely in van der Waerden with a more direct
proof).
PROPOSITION 8.21 Let }(X) be a monic separable polynomial of degree n over a ﬁeld
1, and suppose that the Galois group G of }(X) has s orbits (as a group of permutations
of the roots of } ) with n
1
. . . . . n
x
elements respectively (so that n
1
÷n
2
÷ ÷n
x
= n);
then there is a factorization
}(X) =}
1
(X) }
i
(X)
with }
i
(X) an irreducible polynomial in 1ŒX of degree n
i
.
PROOF. Write }(X) =
¸
(X ÷˛
i
). For S c {1. 2. . . . . n], consider }
S
=
¸
i÷S
(X ÷˛
i
).
This polynomial divides }(X), and it is ﬁxed under the action of G (and hence has coefﬁ
cients in 1) if and only if S is stable under G. Therefore the irreducible factors of }(X)
are the polynomials }
S
with S a minimal subset of {1. . . . . n] stable under G, but such sets
S are precisely the orbits of G in {1. 2. . . . . n].
2
139
8. GLOBAL FIELDS
Let o ÷ S
n
. In GT, :4, it is proved that o is a product of disjoint cycles. More precisely,
if
o
1
={m
11
. . . . . m
1n
1
]. o
2
={m
21
. . m
2n
2
]. ...
are the orbits of 'o) acting on {1. 2. .... n], numbered in such a way that om
i}
= m
i }÷1
,
then
o =(m
11
. . . m
1n
1
) (m
21
. . . m
2n
2
) . . . .
This remark, together with (8.21), gives us the following result.
COROLLARY 8.22 Let }(X) be a monic separable polynomial of degree n over a ﬁnite
ﬁeld k, and let ( be the splitting ﬁeld of }(X). Suppose that the Frobenius element o ÷
Gal((¡k) (when regarded as a permutation of the roots of }(X)) is a product of disjoint
cycles o =c
1
c
x
with c
i
of length n
i
(so that
¸
n
i
=n). Then }(X) factors as a product
of irreducible polynomials in kŒX
}(X) =}
1
(X) }
i
(X)
with }
i
of degree n
i
.
In other words, the type of the cycle decomposition of o can be read off from the
factorization of }(X).
THEOREM 8.23 (DEDEKIND) Let }(X) be a polynomial of degree n over a number ﬁeld
1, and let G be the Galois group of } . Assume }(X) ÷ O
1
ŒX and is monic. Let p be a
prime ideal of 1, and suppose that
}(X) ÷}
1
(X) }
i
(X) mod p
with the }
i
distinct irreducible polynomials in kŒX and }
i
of degree n
i
, k =O
1
¡p. Then
G contains a permutation o that is a product of disjoint cycles of length n
i
.
PROOF. Take o to be the Frobenius element of any prime lying over p — the hypothesis on
the factorization of }(X) mod p implies that p is unramiﬁed in the splitting ﬁeld (because
it implies that p doesn’t divide the discriminant of } ).
2
REMARK 8.24 There is a similar statement for real primes, namely, if
}(X) =}
1
(X) }
i
(X)
in 1ŒX with }
1
. . . . . }
}
of degree 2 and the remainder of the degree 1, then G contains a
permutation o that is a product of disjoint ; cycles of length 2.
This suggests the following strategy for factoring a polynomial QŒX: factor }(X)
modulo many primes ]; discard the result if }(X) mod ] has multiple factors; continue
until a sequence of, say n, primes has yielded no new cycle types for the elements. Then
attempt to read off the type of the group from tables. We discuss how effective this is later.
EXAMPLE 8.25 Let }(X) = X
5
÷X ÷1. Modulo 2 this factors as (X
2
÷X ÷1)(X
3
÷
X
2
÷1); modulo 3 it is irreducible. Hence G contains (12345) and (i k)((mn) for some
numbering of the roots. It also contains ((i k)((mn))
3
=(i k), and this implies that G =S
5
(see 8.28 below).
140
Computing Galois groups (the easy way)
LEMMA 8.26 Let H be a subgroup of S
n
; if H is transitive (for example, contains an
ncycle) and contains an (n÷1)cycle and a transposition, then H =S
n
.
PROOF. After possibly renumbering, we may suppose that the (n÷1)cycle is (1 2 3 . . . n÷
1). By virtue of the transitivity, the transposition can be transformed into (i n), some i _n÷
1. Now the (n÷1)cycle and its powers will transform this into (1 n), (2 n), . . ., (n÷1 n),
and these elements obviously generate S
n
(because S
n
is generated by transpositions).
2
EXAMPLE 8.27 Select monic polynomials of degree n, }
1
, }
2
, }
3
with coefﬁcients in Z
such that
(a) }
1
is irreducible modulo 2;
(b) }
2
=(degree 1)(irreducible of degree n÷1) mod 3;
(c) }
3
=(irreducible of degree 2)(product of one or two irreducible polynomials of odd
degree) mod 5. We need to choose }
3
to have distinct roots modulo 5.
Take
} =÷15}
1
÷10}
2
÷6}
3
.
and let G be the Galois group of } . Then
(a
t
) G is transitive (it contains an ncycle because of (a));
(b
t
) G contains a cycle of length n÷1:
(c
t
) G contains a transposition (because it contains the product of a transposition with
a commuting element of odd order).
The above lemma shows that G =S
n
.
REMARK 8.28 There are other criteria for a subgroup H of S
n
to be all of S
n
. For example,
a subgroup H of S
;
, ] prime, that contains an element of order ] and a transposition is
equal to S
;
(FT, Lemma 4.14).
REMARK 8.29 In Pohst and Zassenhaus 1989, p73, there are suggestions for constructing
irreducible polynomials }(X) of degree n in F
;
ŒX. A root of such a polynomial will
generate F
q
, q =]
n
, and so every such }(X) will divide X
q
÷X. One can therefore ﬁnd
all }(X)s by factoring X
q
÷X.
For example, consider X
125
÷X ÷ F
5
ŒX. Its splitting ﬁeld is F
125
, which has de
gree 3 over F
5
. The factors of X
125
÷X are the minimum polynomials of the elements
of F
125
. They therefore have degree 1 or 3. There are 5 linear factors, X, X ÷1, X ÷2,
X ÷3, X ÷4, and 40 cubic factors, which constitute a complete list of all the monic irre
ducible cubic polynomials in F
5
ŒX. PARI has no trouble factoring X
125
÷X modulo 5
(factormod(X^125X,5)) or X
625
÷X modulo 5, but for X
3125
÷X modulo 5, which
gives a complete list of monic irreducible polynomials of degree 1 or 5 in F
5
ŒX, I had to
increase the allocated memory (allocatemem(10000000)).
However, if you only want one irreducible polynomial of degree n, it is easier to write
down a polynomial at random, and check whether it is irreducible.
141
8. GLOBAL FIELDS
CUBIC POLYNOMIALS
The group S
3
has the following subgroups:
order group group elements
1 1 1
2 C
2
1×1÷1×2
3 ·
3
1×1÷2×3
6 S
3
1×1÷3×2÷2×3.
By the last row, I mean S
3
has one 1cycle, three 2cycles, and two 3cycles.
Note that any subgroup of S
3
containing cycles of length 2 and 3 is the whole of S
3
; thus
if } is irreducible modulo some prime and has an irreducible factor of degree 2 modulo a
second prime, then its Galois group is S
3
. On the other hand, if factorizing } modulo many
primes doesn’t turn up a factor of degree 2, but } is irreducible, then expect the Galois group
of } to be ·
3
. This can be checked by seeing whether disc(} ) is a square. For example,
the calculations on p. 61 show that the polynomials X
3
÷10X ÷1 and X
3
÷8X ÷15 both
have Galois group S
3
.
To make this more effective (in the technical sense), we need the Chebotarev density
theorem.
CHEBOTAREV DENSITY THEOREM
DEFINITION 8.30 Let S be a set of ﬁnite primes in a number ﬁeld 1, and let 1 be the set
of all ﬁnite primes. We say that S has natural density ı if
lim
1o
[{p ÷ S [ Np _N][
[{p [ Np _N][
=ı.
THEOREM 8.31 (CHEBOTAREV DENSITY THEOREM) Let 1 be a ﬁnite Galois extension
of the number ﬁeld 1, with Galois group G, and let C be a conjugacy class in G. The set
of prime ideals p of 1 such that (p. 1¡1) =C has density ı =[C[ ¡[G[.
PROOF. See my notes CFT (in fact, normally one proves this result with a slightly weaker
notion of density).
2
For example, if G is abelian, then for each o ÷ G, the set of p such that (p. 1¡1) =o
has density 1¡[G[ .
COROLLARY 8.32 The primes that split in 1 have density 1¡Œ1 : 1. In particular, there
exist inﬁnitely many primes of 1 not splitting in 1.
REMARK 8.33 There is a bound for the error in implicit in (8.31) in terms of the dis
criminant of the polynomial, but it is large. The existence of the bound has the following
consequence: given a polynomial }(X) ÷ QŒX (say), there exists a bound T such that, if
a given cycle type doesn’t occur as the Frobenius element of some ] _ T, then it doesn’t
occur at all. For a discussion of the effective version of the Chebotarev density theorem,
see Lagarias and Odlysko, 1977.
3
3
Lagarias, J. C.; Odlyzko, A. M. Effective versions of the Chebotarev density theorem. Algebraic number
ﬁelds: 1functions and Galois properties (Proc. Sympos., Univ. Durham, Durham, 1975), pp. 409–464.
Academic Press, London, 1977.
142
Computing Galois groups (the easy way)
EXAMPLE 8.34 Let 1 =QŒ¯
n
. Then Gal(1¡Q) =(Z¡nZ)
×
and (]. 1¡Q) =Œ]. The
Chebotarev density theorem says that the primes are equidistributed among the congruence
classes. In other words, each of the arithmetic progressions
k, k ÷n, k ÷2n, k ÷3n. . . . gcd(k. n) =1.
contains 1¡c(n) of the primes. In particular, each of the arithmetic progressions contains
inﬁnitely many primes. This statement was conjectured by Legendre and proved by Dirich
let (using Dirichlet series). The proof of the Chebotarev density theorem is a generalization
of that of Dirichlet.
EXAMPLE 8.35 In a quadratic extension, half the primes split and half the primes remain
prime.
EXAMPLE 8.36 Let } be a cubic polynomial with coefﬁcients in Q. The Chebotarev den
sity theorem implies the following statements (see the above table):
G =1: } splits modulo all primes.
G =C
2
: } splits for 1¡2 of the primes and has an irreducible factor of degree 2 for 1¡2
of the primes.
G =·
3
: } splits for 1¡3 of the primes and } remains irreducible for 2¡3 of the primes.
G =S
3
: } splits for 1¡6 of the primes, has a factor of degree 2 for 1¡2 of the primes,
and remains prime for 1¡3 of the primes.
EXAMPLE 8.37 Let } be a quartic polynomial with no linear factor.
(a) When disc(} ) is a square, the possible Galois groups are:
order group elements
2 C
2
1×1÷1×2
2
4 V
4
1×1÷3×2
2
12 ·
4
1×1÷3×2
2
÷8×3
(b) When disc(} ) is not a square, the possible Galois groups are:
order group elements
4 C
4
1×1÷1×2
2
÷2×4
8 D
S
1×1÷2×2÷3×2
2
÷2×4
24 S
4
1×1÷3×2
2
÷6×2÷8×3÷6×4
See FT :4. Thus if } is a polynomial of degree 4 with Galois group D
S
, then it will
split modulo ] for 1¡8 of the primes, factor as the product of a quadratic and two linear
polynomials for 1¡4 of the primes, factor as the product of two quadratics for 3¡8 of the
primes, and remain irreducible for 1¡4 of the primes.
For a similar table for polynomials of degree 5, see Pohst and Zassenhaus 1989, p132.
One strategy for determining the Galois group of a polynomial is
(a) test whether } is irreducible over Q:
(b) compute the discriminant of } ;
(c) factor } modulo good primes (i.e., those not dividing the discriminant) until you
seem to be getting no new cycle types;
143
8. GLOBAL FIELDS
(d) compute the orbit lengths on the rsets of roots (these are the degrees of the irre
ducible factors in QŒX of the polynomial whose roots are the sums of r roots of
} );
(e) ad hoc methods.
As late as 1984, it had not been proved that the Mathieu group M
11
occurs as a Galois
group over Q (M
11
is subgroup of S
11
of order 11Š¡5040 =7920).
References
Butler, Gregory; McKay, John. The transitive groups of degree up to eleven. Comm.
Algebra 11 (1983), no. 8, 863–911. (This lists all transitive subgroups of S
n
, n _ 11,
and gives the cycle types of their elements and the orbit lengths of the subgroup acting on
the rsets of roots; with a few exceptions, these invariants are sufﬁcient to determine the
subgroup up to isomorphism.)
Cohen 1993, Section 6.3.
Ford, David J.; McKay, John, Computation of Galois groups from polynomials over the
rationals. Computer algebra (New York, 1984), 145–150, Lecture Notes in Pure and Appl.
Math., 113, Dekker, New York, 1989.
Pohst and Zassenhaus 1989. Chapter 2 is entirely concerned with computing Galois
groups; for example, II.10.8 discusses the following problem: given G c H c S
n
, deter
mine whether G is contained in a given smaller subgroup J of H.)
Soicher, L. H. An algorithm for computing Galois groups. Computational group theory
(Durham, 1982), 291–296, Academic Press, London, 1984.
Soicher, Leonard; McKay, John. Computing Galois groups over the rationals. J. Num
ber Theory 20 (1985), no. 3, 273–281.
PARI can ﬁnd the Galois group of a polynomial of degree _11.
Applications of the Chebotarev density theorem
We now discuss some other applications of the Chebotarev density theorem.
For any extension 1¡1 of number ﬁelds, write Spl(1¡1) for the set of primes that
split completely in 1, and write Spl
t
(1¡1) for the set of primes that have at least one split
factor. Then Spl(1¡1) cSpl
t
(1¡1) always, and equality holds if 1¡1 is Galois, in which
case the Chebotarev density theorem shows that Spl(1¡1) has density 1¡Œ1 : 1.
THEOREM 8.38 If 1 and M are Galois over 1, then
1 cM ” Spl(1¡1) ¯Spl(M¡1).
PROOF. =: This is obvious.
=: We have
Spl(1M¡1) =Spl(1¡1) ¨Spl(M¡1).
To see this, note that
p ÷ Spl(1M¡1) ” (p. 1M¡1) =1
” (p. 1M¡1)[1 =1 and (p. 1M¡1)[M =1:
144
Applications of the Chebotarev density theorem
but (p. 1M¡1)[1 =(p. 1¡1) and (p. 1M¡1)[M =(p. M¡1). Now
Spl(M¡1) cSpl(1¡1) =Spl(1M¡1) =Spl(M¡1)
S.31
= Œ1M : 1 =ŒM : 1 =1 cM.
2
COROLLARY 8.39 If 1 and M are Galois over 1, then
1 =M ” Spl(M¡1) =Spl(1¡1).
PROOF. Obvious from the Proposition.
2
REMARK 8.40 (a) In fact, 1 =M if Spl(M¡1) and Spl(1¡1) differ by at worst a ﬁnite
set of primes (or if they differ by at worst a set of primes of density zero).
(b) The effective form of the Chebotarev density theorem shows that (8.38) is effective:
in order to show that 1 cM it sufﬁces to check that
p splits in M =p splits in 1
for all primes p less than some bound.
(c) Proposition 8.39 is not true without the Galois assumptions: there exist noniso
morphic extensions 1 and M of Q such that Spl(1¡1) = Spl(M¡1). In fact there exist
nonisomorphic extensions 1 and M of Q of the same degree such that
˘ 1 and M have the same discriminant;
˘ a prime ] not dividing the common discriminant decomposes in exactly the same
way in the two ﬁelds.
(d) It is clear from (8.39) that if a separable polynomial }(X) ÷ 1ŒX splits into linear
factors mod p for all but ﬁnitely many primes p of 1, then }(X) splits into linear factors in
1ŒX. With a little more work, one can show that an irreducible polynomial }(X) ÷ 1ŒX
can not have a root mod p for all but a ﬁnite number of primes. This last statement is false
for reducible polynomials — consider for example,
(X
2
÷2)(X
2
÷3)(X
2
÷6).
For more on these questions, see Exercise 6, p361, of Algebraic number theory. Proceedings
of an instructional conference organized by the London Mathematical Society. Edited by
J. W. S. Cassels and A. Fr¨ ohlich Academic Press, London; Thompson Book Co., Inc.,
Washington, D.C. 1967.
(e) It is easy to give examples of polynomials }(X) that are irreducible over Q but
become reducible over Q
;
for all ], including ] = o. Since the Galois group of any
extension of local ﬁelds is solvable, one only has to chose } to have nonsolvable Galois
group, for example, S
n
for n _5.
EXAMPLE 8.41 Fix a number ﬁeld 1. According to (8.39), a Galois extension 1 of 1
is determined by the set Spl(1¡1). Thus, in order to classify the Galois extensions of
1, it sufﬁces to classify the sets of primes in 1 that can occur as Spl(1¡1). For abelian
extensions of 1, class ﬁeld theory does this — see CFT (they are determined by congruence
conditions). For nonabelian extensions the sets are still a mystery — it is known that they
are not determined by congruence conditions — but Langlands’s conjectures shed some
light.
145
8. GLOBAL FIELDS
Exercises
81 Let 1 = QŒ˛ where ˛ is a root of X
3
÷X
2
÷2X ÷8. Show that there are three
extensions of the 2adic valuation to 1. Deduce that 2[ disc(ZŒ˛¡Z) but not disc(O
1
¡Z).
82 Let 1 be a ﬁnite Galois extension of the local ﬁeld 1, and let G
i
, i _ 0, be the i th
ramiﬁcation group. Let ˘ generate the maximal ideal in O
1
. For o ÷ G
i
, write o˘ =
˘ ÷a(o)˘
i÷1
, and consider the map G
i
÷l, o ÷a(o) mod (˘), where l =O
1
¡(˘).
Show that this is a homomorphism (additive structure on l) if and only if i >0.
83
+
“It is a thoughtprovoking question that few graduate students would know how to
approach the question of determining the Galois group of, say,
4
X
6
÷2X
5
÷3X
4
÷4X
3
÷5X
2
÷6X ÷7.”
(a) Can you ﬁnd it?
(b) Can you ﬁnd it without using a computer?
84 Let 1 =k(X) where k is a ﬁnite ﬁeld. Assume that every valuation of 1 comes from
a prime ideal of kŒX or kŒX
1
], and prove the product formula.
And after the ﬁrst year [as an undergraduate at G¨ ottingen] I went home with
Hilbert’s Zahlbericht under my arm, and during the summer vacation I worked
my way through it — without any previous knowledge of elementary number
theory or Galois theory. These were the happiest months of my life, whose
shine, across years burdened with our common share of doubt and failure, still
comforts my soul.
Hermann Weyl, Bull. Amer. Math. Soc. 50 (1944), 612–654.
4
I don’t remember where this quote is from.
146
APPENDIX A
Solutions to the Exercises
01. Use that ˛ = m÷n
J is an algebraic integer if and only if Tr(˛) = ÷2m ÷ Z and
Nm(˛) =m
2
÷n
2
J ÷ Z.
02. Similar to Exercise 21 below.
11. (a) =: Let S =·
i
p
i
with the p
i
prime ideals.
.. . ÷ S ”Vi. .. . … p
i
”Vi. .. … p
i
” .. ÷ S.
=: Let a … S. Then (a) ¨S = 0 because S is saturated. Let 1 be maximal among the
ideals of · containing a and disjoint from S — exists by Zorn’s Lemma. I’ll show that 1
is prime. Suppose .. ÷ 1.
If . … 1, then 1 ÷(.) properly contains 1, and so (1 ÷(.)) ¨S is nonempty — let
c ÷a. ÷ S with c ÷ 1 and a ÷ ·.
Similarly, if . … 1, there exists c
t
÷a
t
. ÷ S.
But (c ÷a.)(c
t
÷a
t
.) ÷1, which is not possible because S is multiplicative. Therefore
. or . ÷ 1, and so 1 is prime.
(b) Given S, let S
t
={. ÷·[ J. ÷·such that .. ÷S] — verify that it is multiplicative
and saturated, and is the smallest such set containing S; moreover, it is a union of the prime
ideals not meeting S, and S
1
M ŠS
t1
M for all ·modules. For the ﬁnal statement, use
that p remains prime in S
1
· if and only if S ¨p =0.
[Cf. Bourbaki, Alg. Comm., 1961, II, Ex. :2, no. 1, and Atiyah and MacDonald,
Chapt. 3, no. 7.]
21. By inspection, 4 = 2 2 = (3 ÷
5)(3 ÷
5). We have to show that 2, 3 ÷
5, and
3÷
5 are irreducible, and 2 is not an associate of the other two.
If 2 =˛ˇ then 4 =Nm(2) =Nm(˛) Nm(ˇ), from which it follows that Nm(˛) =˙1,
˙2, or ˙4. If Nm(˛) =˙1, ˛ is unit (with inverse ˙its conjugate); by looking mod 5, one
sees that Nm(˛) =˙2 is impossible; if Nm(˛) =˙4, then ˇ is a unit. Hence 2 can’t be
factored into a product of nonunits. The same argument applies to the other two elements.
If 2 and 3 ÷
5 were associates, then there would be a unit m÷n
5 in ZŒ
5 such
that 3÷
5 =2(m÷n
5), but this is impossible.
22. Suppose }(X) =
¸
g
i
(X) with g
i
(X) irreducible in 1ŒX. Let ˛ be a root of g
i
(X)
in some extension ﬁeld. Then g
i
(X) is the minimum polynomial of ˛ over 1. Because ˛
is a root of }(X), it is integral over A, and so g
i
(X) has coefﬁcients in · (by 2.9).
147
A. SOLUTIONS TO THE EXERCISES
23. Consider ﬁrst the case that 1 =1Œ˛, ˛
;
=a ÷ 1.
24. Clearly 2 does not divide 1÷
÷3 in ZŒ
÷3, and so a =(2), but
a
2
=(4. 2÷2
÷3. ÷2÷2
÷3) =(4. 2÷2
÷3) =(2)(2. 1÷
÷3) =(2)a.
If there were unique factorization into products of prime ideals, then
ab =ac. a =0 =b =c.
We have shown that the ring ZŒ
÷3 doesn’t have this property.
25. Let ˛ ÷ ·Œˇ ¨·Œˇ
1
. By hypothesis, we can write
˛ =a
0
÷a
1
ˇ÷ ÷a
n
ˇ
n
˛ =b
0
÷b
1
ˇ÷ ÷b
n
ˇ
n
.
Let M be the ·submodule of T generated by {ˇ
n
. . . . . 1. . . . . ˇ
n
]. Fromthe ﬁrst equation,
we ﬁnd that ˛ˇ
i
÷ M, 0 _ i _ n, and from the second equation we ﬁnd that ˛ˇ
}
÷ M,
0 _; _m. We can apply (2.4) to deduce that ˛ is integral over ·.
26. (a) Check easily that the products ˛
i
˛
}
, i =; , are divisible by 3, and this implies that
(
¸
˛
i
)
n
÷
¸
˛
n
i
mod 3. The rest is easy.
(b) Using Gauss’s Lemma, one ﬁnds that X ÷˛: ZŒX ÷ZŒ˛ deﬁnes an isomorphism
ZŒX¡(}(X)) .ZŒ˛. Hence
3[g(˛) ” Jh ÷ ZŒX s.t. } [g ÷3h ” 3[ ¨ g.
(c) O.K.
(d) Since F
3
has only 3 elements, there are only 3 monic polynomials of degree 1.
31. It is not a Dedekind domain because it has a chain of prime ideals
(X. Y ) ¯(X) ¯(0).
32. From Galois theory (or playing around, or from PARI) ﬁnd that (
3÷
7)¡2 is a root
of the polynomial X
4
÷5X
2
÷1.
34. Let ·=kŒX
2
. X
3
 ckŒX. As kŒX =kŒX
2
 1÷kŒX
2
 X, it is a Noetherian kŒX
2

module. Therefore, an ideal in · is ﬁnitely generated when regarded as a kŒX
2
module,
and a fortiori as an ·module. Thus · Noetherian. If p is nonzero prime ideal of ·, then
p contains a nonzero polynomial, and so ·¡p is a ﬁnitedimensional vector space over k.
Since it is an integral domain, it must be a ﬁeld (see 3.30), and so p is maximal. The element
X of k(X) is integral over · because it is a root of the polynomial T
2
÷X
2
÷ ·ŒT , but
X … ·. Therefore · is not integrally closed.
41. For example, take T =kŒX. Y  ¯kŒX =· and p =(Y ), or T =kŒX ¯k =· and
p =(X).
42. Write pT =
¸
1
e(1
i
¡;)
i
and P
i
C =
¸
Q
e(O
ij
¡1
i
)
i}
. Then
pC =
¸
i
(P
i
C)
e(P
i
¡p)
=
¸
i,}
Q
e(P
i
¡p)e(Q
ij
¡P
i
)
i}
.
148
and Q
i}
=Q
i
0
}
0 unless (i. ; ) =(i
t
. ;
t
). For the second part of the problem, see the start of
:4 of the notes.
43. The possibilities are determined by
¸
e
i
}
i
=3. Since the discriminant is ÷31, only
31 ramiﬁes, and X
3
÷10X ÷1 ÷ (X ÷28)(X ÷17)
2
mod 31. All possibilities except
(]) =p
3
occur.
44. Compute the Minkowski bound to ﬁnd a small set of generators for the class group. In
order to show that two ideals a and b are equivalent, it is often easiest to verify that a b
n1
is principal, where m is the order of b in the class group.
45. Let a
1
. . . . . a
h
be a set of representatives of the ideal classes. It sufﬁces to ﬁnd a ﬁeld
1 such that each a
i
becomes principal in 1. Because the ideal class group is ﬁnite, each
of the a
i
is of ﬁnite order, say a
n
i
i
=(a
i
), a
i
÷ 1. Let 1 be a ﬁnite extension of 1 such
that each a
i
becomes an m
i
th power in 1, say a
i
=˛
n
i
i
, ˛
i
÷ 1. In the group of fractional
ideals of 1, we have
a
n
i
i
1
=(a
i
) =(˛
n
i
i
) =(˛
i
)
n
i
.
Since the group of fractional ideals is torsionfree, this equation implies that a
i
O
1
=(˛
i
).
[In fact, every ideal of 1 becomes principal in the Hilbert class ﬁeld of 1 (see 4.9), but this
is very difﬁcult to prove — it is the Principal Ideal Theorem (see CFT).]
46. The discriminant of X
3
÷X ÷2 is (÷26)2
2
, and Stickleberger’s lemma shows ÷26 is
not a possible discriminant, and so O
1
= ZŒ˛. To show that the class number is 1, it is
only necessary to show that the ideals dividing (2) are principal.
47. To show that O
1
=ZŒi Œ,, , =
1÷
5
2
, observe that D(1. ,) =5, and 5 is not a square
in ZŒi ; now apply Lemma 2.23. The prime 2 ramiﬁes in QŒi , but not in QŒ
5, and so it
ramiﬁes in 1 with ramiﬁcation index 2 (this follows from the multiplicativity of the e’s).
Similarly, 5 ramiﬁes in 1 with ramiﬁcation index 2. Since disc(O
1
¡ZŒi ) =(5), only the
divisors of (5) (in ZŒi ) can ramify in
1
, and hence only 2 and 5 can ramify in 1. The proof
that QŒ
÷5 has class number 2 is sketched in (4.6). [Of course, this problem becomes
much easier once one has (6.5).]
51. No! Some inﬁnite sets:
{m
2÷Œm
2 [ m. n ÷ Z], Œ+ =integer part;
{(
2÷1)
n
[ n ÷ N];
{
n
2
÷1÷n [ n ÷ N];
{˛ [ ˛ is the smaller root of X
2
÷mX ÷1 =0. m÷ Z]
52. The period is 10, and the fundamental unit is
48842÷5967
67.
53. No! One way to obtain a counterexample is to note that, if a prime ] factors as ] =¬
1
¬
2
(¬
i
nonassociate primes) in a quadratic extension of Q, then Nm¬
1
=˙] =Nm¬
2
, and
so ¬
1
¡¬
2
has norm ˙1. For example 5 =(2÷i )(2÷i ) in QŒi , and so (2÷i )¡(2÷i ) has
norm 1, but it is not an algebraic integer. Alternatively, note that any root of an irreducible
polynomial X
n
÷a
1
X
n1
÷ ÷1, a
i
÷ Q, not all a
i
÷ Z, will have norm˙1, but will not
be an algebraic integer.
61. Let ˛ be a root of X
3
÷3X ÷1. Then disc(ZŒ˛¡Z) = 81. Since its sign is (÷1)
x
,
we must have s = 0, r = 3 — three real embeddings. From their minimum polynomials,
149
A. SOLUTIONS TO THE EXERCISES
one sees that ˛ and ˛ ÷2 are algebraic integers with norms 1 and ÷1 respectively. From
(˛÷1)
3
=3˛(˛÷2) we ﬁnd (˛÷1)
3
=(3) in O
1
. From the formula
¸
e
i
}
i
=3, we ﬁnd
that there can be no further factorization, and e =3, } =1. The second equality implies
that O
1
¡(˛ ÷1) =Z¡(3), and so
1
=Z÷(˛ ÷1)
1
. The proof that O
1
=ZŒ˛ proceeds
as in the proof of 6.2. The Minkowski bound is 2, and 2 O
1
is prime, because X
3
÷3X÷1
is irreducible modulo 2.
62. First solution: Let ˛ be an algebraic integer in QŒ¯ ÷¯
1
. We can write it
˛ =
¸
a
i
(¯ ÷¯
1
)
i
. 0 _i <c(m)¡2. a
i
÷ Q.
Suppose a
n
is the last coefﬁcient not in Z. Then ˛
t
=
¸
n
i=0
a
i
(¯ ÷¯
1
)
i
is also an algebraic
integer. On expanding this out, and multiplying through by ¯
n
, we ﬁnd that
¯
n
˛
t
=a
n
¯
2n
÷ terms of lower degree in ¯. a
n
… Z.
This contradicts the fact that ¯
n
˛
t
is in ZŒ¯.
Second solution: Clearly, O
QŒc÷c
1
]
=O
QŒc]
¨QŒ¯ ÷¯
1
. It follows that the algebraic
integers in QŒ¯ ÷¯
1
 are those elements that can be expressed
¸
a
i
(¯
i
÷¯
i
), a
i
÷ Z.
Now prove inductively that ¯
i
÷¯
i
÷ ZŒ¯ ÷¯
1
.
72. (a) Easy. (b) Show s
n
=
¸
n
i=0
a
i
is Cauchy if and only if a
i
÷0.
73. If a = 0, there is a solution, and so we now take a = 0. To have a solution in Z
T
,
clearly it is necessary that a =7
2n÷1
b, m _ 0, with b an integer that is not divisible by
7 but is a square modulo 7 (hence b ÷ 1. 2. 4 mod 7). Newton’s lemma shows that this
condition is also sufﬁcient.
For a ÷ Q, 7X
2
=a has a solution in Q
T
if and only if a =7
2n÷1
b, m ÷ Z, b ÷ Z,
b ÷1. 2. 4 mod 7.
74. (a) Because the product of two nonsquares in Z¡(]) is a square, and least one of
X
2
÷2, X
2
÷17, X
2
÷34 has a root modulo ], and if ] = 2. 17, the root is simple and
hence lifts to a root in Z
;
(by Newton’s lemma). The polynomial X
2
÷2 has 6 as a simple
root modulo 17, and so it has a root in Z
1T
. Let g(X) = X
2
÷17 and a
0
= 1. Then
[g(a
0
)[
2
=1¡16 and [g
t
(a
0
)
2
[
2
=1¡4 and so Newton’s lemma (7.32) again shows that it
has a root in Z
2
.
(b) Apply the method of proof of (7.31) to ﬁnd
a =1÷5 7÷7
3
÷2 7
4
÷5 7
5
÷ .
75. If k is a ﬁeld of characteristic =2, a quadratic extension of k is of the form kŒ
a for
some a ÷ k, a … k
2
, and two nonsquare elements a and b of k deﬁne the same quadratic
extension if and only if they differ by a square (FT, Theorem 5.27). Thus the quadratic
extensions of k are in onetoone correspondence with the cosets of k
×2
in k
×
other than
k
×2
itself.
We have to ﬁnd a set of representatives for Q
×2
2
in Q
×
2
. Clearly an element u 2
n
of Q
×
2
,
u ÷ Z
×
2
, is a square if and only if n is even and u is a square in Z
2
, and Newton’s lemma
shows that u is a square in Z
2
if (and only if) it is a square in Z
2
¡(8) =Z¡(8). The elements
˙1. ˙5 form a set of representatives for (Z¡(8))
×
, and of these only 1 is a square. Hence
{˙1. ˙5 ˙2. ˙10] is a set of representatives for Q
×
2
¡Q
×2
2
, and so the distinct quadratic
extensions of Q
2
are the ﬁelds QŒ
a] for a =÷1. ˙2. ˙5. ˙10.
150
There is a description of the structure of Q
×
;
in Serre, Course..., II.3. Let U =Z
×
;
and
let U
i
be the subgroup 1 ÷]
i
Z
;
of U; we know from (7.27) that Q
;
contains the group
u
;1
of (]÷1)
xt
roots of 1, and one shows that
Q
×
;
~Z×u
;1
×U
1
. U
1
~Z
;
. ] =2:
Q
×
2
~Z×U
1
. U
1
={˙1] ×U
2
. U
2
~Z
2
.
There is a general formula,
(1
×
: 1
×n
) =
m
m
(u
n
: 1)
for any ﬁnite extension 1 of Q
;
; here u
n
is the group of m
th
roots of 1 in K. See CFT VII.
76. If 2 occurs among the ˛
i
, number it ˛
1
. Then ˛
i
… QŒ˛
1
. ˛
2
. .... ˛
i1
 because ]
i
does
not ramify in QŒ˛
1
. ˛
2
. .... ˛
i1
. Therefore the degree is 2
n
(alternatively, use Kummer
theory). The element , is moved by every element of Gal(1¡Q), and so it generates 1.
The group Gal(1¡Q) is abelian of exponent 2 (i.e., every element has square 1). The same
is true of the decomposition groups of the primes lying over ]. Write 1˝
Q
Q
;
=
¸
1
i
,
so that 1
i
~ 1ŒX¡(}
i
(X)) where }
i
(X) is the i
th
irreducible factor of }(X) in Q
;
ŒX
(cf. 8.2). Kummer theory and the description of Q
×
;
given above show that Œ1
i
: Q
;
 _ 4
if ] = 2 and Œ1
i
: Q
2
 _ 8 (because their Galois groups are abelian of exponent 2). This
implies that }(X) factors as stated.
77. The degree of Q
;
Œ¯
n
, ] does not divide n, is } , where } is the smallest integer
such that n[]
(
÷1. As n ÷o, } ÷o, and so a ﬁnite extension 1 of Q
;
can contain
only ﬁnitely many ¯
n
’s. Suppose
¸
¯
n
]
n
converges to ˇ ÷ Q
al
;
. Then 1 = Q
;
Œˇ is a
ﬁnite extension of Q
;
. Let ˛
t
=
¸
t
n=1
¯
n
]
n
. Then ˛
t
is further from its conjugates than
it is from ˇ, and so Krasner’s lemma (7.60) implies that Q
;
Œ˛
t
 c Q
;
Œˇ. It follows (by
induction) that Q
;
Œˇ contains all the ¯
n
, and this is impossible.
78. (a) The polynomial
X
3
÷X
2
÷X ÷1
has the factor X ÷1, but
X
3
÷X
2
÷X ÷1
is irreducible because it is irreducible modulo 5.
(b) Consider
} =X
6
÷3×5X
5
÷3×5X
4
÷3×5
4
X
3
÷3×5
4
X
2
÷3×5
9
X ÷3×5
9
.
It is Eisenstein for 3, and hence is irreducible over Q. Its Newton polygon for 5 has slopes
1¡2, 3¡2, and 5¡2, each of length 2. Correspondingly, in Q
5
ŒX it is a product of three
polynomials } = }
1
}
2
}
3
. Each of the }
i
is irreducible because the ﬁeld generated by a
root of it is ramiﬁed (because the slope isn’t an integer).
81. The Newton polygon of }(X) = X
3
÷X
2
÷2X ÷8 has three distinct slopes 1. 2. 3,
and so it splits over Q
2
. Now (8.1) shows that [ [
2
has three distinct extensions to 1. Using
that
¸
e
i
}
i
=3, we see that 2 doesn’t ramify in 1, and so 2 does not divide discO
1
¡Z.
On the other hand 2[ disc(}(X)) because }(X) has multiple roots modulo 2 (according to
PARI, its discriminant is ÷2012).
151
A. SOLUTIONS TO THE EXERCISES
82. Straightforward.
83. (a) In PARI, type polgalois(X^6+2*X^5+3*X^4+4*X^3+5*X^2+6*X+7).
(b) There are the following factorizations:
mod 3, irreducible;
mod 5, (deg 3) ×(deg 3).
mod 13, (deg 1) ×(deg 5):
mod 19, (deg 1)
2
×(deg 4):
mod 61, (deg 1)
2
×(deg 2)
2
:
mod 79, (deg 2)
3
.
Thus the Galois group of } has elements of type:
6. 3÷3. 1÷5. 1÷1÷4. 1÷1÷2÷2. 2÷2÷2.
No element of type 2, 3, 3 ÷2, or 4 ÷2 turns up by factoring modulo any of the ﬁrst 400
primes (so I’m told). Thus it is the group T14 in the tables in Butler and McKay (see p141)
of the notes. It has order 120, and is isomorphic to PGL
2
(F
5
) (group of invertible 2 ×2
matrices over F
5
modulo the scalar matrices a1
2
, a ÷ F
×
5
).
84. Prime ideals of kŒX and kŒX
1
 deﬁne the same valuation of k(X) if and only if
they generate the same prime ideal of kŒX. X
1
. Thus there is one valuation ord
;
for each
monic irreducible polynomial ](X) of kŒX, and one for the polynomial X
1
in kŒX
1
.
The normalized valuation corresponding to ](X) is
g(X)¡h(X) =
1
q
deg;
ord
p
vord
p
h
where q =#k and ord
;
(g) is the power of ](X) dividing g(X), and the normalized valua
tion corresponding to X
1
is
g(X)¡h(X) =
1
q
deghdegv
.
Thus the product formula is equivalent to the formula,
¸
;(A)
deg](ord
;
g ÷ord
;
h) =degg ÷degh.
which is obvious.
152
APPENDIX B
Twohour examination
Prove (or, at least, explain) your answers.
1. (a) Is (1÷i )¡
2 an algebraic integer?
(b) Is ZŒ
29 a principal ideal domain?
2. Let 1 =QŒ˛ where ˛ is a root of X
n
÷2, n _2.
(a) Find Œ1 : Q.
(b) In how many ways can the usual archimedean valuation on Q be extended to 1?
(c) Same question for the 2adic valuation.
(d) Find the rank of the group of units in O
1
and the order of its torsion subgroup.
3. Let ¯ be a primitive 8
th
root of 1. Show that QŒ¯ contains exactly 3 subﬁelds of degree
2 over Q, and they are QŒ
÷1, QŒ
2, QŒ
÷2.
4. Let ˛ and ¬ be nonzero elements of the ring of integers O
1
of a number ﬁeld 1 with
¬ irreducible (i.e., ¬ = ab =a or b a unit). If ¬[˛
3
, can you conclude that ¬[˛? What
condition on the class group would allow you to conclude this?
5. Let 1 =Q
3
Œ¯, where ¯ is a primitive 3rd root of 1. Find the Galois group of 1 over Q
3
and its ramiﬁcation groups.
6. Let 1 be a ﬁnite Galois extension of Q with Galois group G. For each prime ideal P of
O
1
, let 1(P) be the inertia group. Show that the groups 1(P) generate G.
153
Bibliography
ARTIN, E. 1959. Theory of algebraic numbers. Notes by Gerhard W¨ urges from lectures held at the Mathema
tisches Institut G¨ ottingen, Germany, 1956/7, in the Winter Semester. Translated and distributed by George
Striker, Schildweg 12, G¨ ottingen. Reprinted in Artin 2007.
CASSELS, J. W. S. 1986. Local ﬁelds, volume 3 of London Mathematical Society Student Texts. Cambridge
University Press, Cambridge.
COHEN, H. 1993. A course in computational algebraic number theory, volume 138 of Graduate Texts in
Mathematics. SpringerVerlag, Berlin.
COHN, P. M. 1991. Algebraic numbers and algebraic functions. Chapman and Hall Mathematics Series.
Chapman & Hall, London.
CURTIS, C. W. AND REINER, I. 1962. Representation theory of ﬁnite groups and associative algebras. Pure
and Applied Mathematics, Vol. XI. Interscience Publishers, a division of John Wiley & Sons, New York
London.
DEDEKIND, R. 1877. Sur la th´ eorie des nombres entiers alg´ ebriques. Bull. des Sc. Math. 1:?–?
DEDEKIND, R. 1996. Theory of algebraic integers. Cambridge Mathematical Library. Cambridge University
Press, Cambridge. Translated from the 1877 French original and with an introduction by John Stillwell.
EDWARDS, H. M. 1977. Fermat’s last theorem, volume 50 of Graduate Texts in Mathematics. SpringerVerlag,
New York. A genetic introduction to algebraic number theory.
FR ¨ OHLICH, A. AND TAYLOR, M. J. 1991. Algebraic number theory, volume 27 of Cambridge Studies in
Advanced Mathematics. Cambridge University Press, Cambridge.
JANUSZ, G. J. 1996. Algebraic number ﬁelds, volume 7 of Graduate Studies in Mathematics. American
Mathematical Society, Providence, RI, second edition.
LENSTRA, JR., H. W. 1992. Algorithms in algebraic number theory. Bull. Amer. Math. Soc. (N.S.) 26:211–244.
MILNE, J. S. 2006. Elliptic Curves. BookSurge Publishers.
NARKIEWICZ, W. 1990. Elementary and analytic theory of algebraic numbers. SpringerVerlag, Berlin, second
edition.
POHST, M. AND ZASSENHAUS, H. 1989. Algorithmic algebraic number theory, volume 30 of Encyclopedia
of Mathematics and its Applications. Cambridge University Press, Cambridge.
ROQUETTE, P. 1967. On class ﬁeld towers, pp. 231–249. In Algebraic Number Theory (Proc. Instructional
Conf., Brighton, 1965). Thompson, Washington, D.C.
SERRE, J.P. 1962. Corps locaux. Publications de l’Institut de Math´ ematique de l’Universit´ e de Nancago,
VIII. Actualit´ es Sci. Indust., No. 1296. Hermann, Paris.
WEISS, E. 1963. Algebraic number theory. McGrawHill Book Co., Inc., New York.
154
Index
algebra, 7
algebraic integer, 1
algorithm, 34
good, 34
practical, 34
basis, 25
binary quadratic form, 75
Cauchy sequence, 109
class ﬁeld tower, 67
class number, 4, 48
complete ﬁeld, 109
continued fraction, 85
convex set, 70
cyclotomic polynomial, 91
Dedekind domain, 42
discrete subgroup, 67
discrete valuation, 49
discrete valuation ring, 40
discriminant, 26, 27, 31
Eisenstein polynomial, 60, 124
Eisenstein’s criterion, 60
element
irreducible, 2
prime, 1
equivalent valuations, 103
ﬁeld of fractions, 11
Frobenius element, 136
full lattice, 67
fundamental parallelopiped, 69
fundamental system of units, 79
global ﬁeld, 130
group
decomposition, 134
higher ramiﬁcation, 125
inertia, 125
ramiﬁcation, 125
splitting, 134
Hermite normal form, 36
Hilbert class ﬁeld, 67
ideal
fractional, 46
integral, 46
principal, 47
ideal class group, 48
integral basis, 30
integral closure, 22
integral element, 19
integrally closed ring, 22
lattice, 67
lemma
Hensel’s, 115
Krasner’s, 126
Nakayama’s, 10
Newton’s, 114, 115
local ﬁeld, 121
local ring, 10
local uniformizing parameter, 110
maximal ideal, 8
Minkowski bound, 64
Minkowski constant, 64
multiplicative subset, 11
natural density, 142
Newton’s polygon, 119
nilpotent, 55
Noetherian module, 10
Noetherian ring, 8
nondegenerate bilinear form, 27
norm, 25, 71
155
INDEX
numerical, 63
norm of an ideal, 62
normalized discrete valuation, 49
PARI, 5, 32, 35, 37, 58, 60, 66, 91, 120,
132, 141, 144
prime ideal, 8
primitive nth root of 1, 90
reduced, 55
regulator, 88
relatively prime, 13
ring of integers, 22
Sinteger, 84
Sunit, 84
symmetric in the origin, 70
symmetric polynomial, 19
elementary, 19
tamely ramiﬁed, 126
tensor product, 14
theorem
Chebotarev density, 142
Chinese Remainder, 13
Chinese Remainder (for modules), 14
cyclotomic ﬁelds, 94
Dedekind’s on computing Galois groups,
140
extending valuations, 117
factoring primes in an extension, 56
Fermat’s Last, 97
fractional ideals form group, 47
integral closure of Dedekind domain,
51
invariant factor, 52
Minkowski bound, 64
modules over Dedekind domain, 52
points in lattice, 69
primes of a number ﬁeld, 106
primes that ramify, 54
product formula, 105, 106, 133
Stickelberger’s, 33
sum of ef’s is the degree, 53
tensor product of ﬁelds, 17
the class number is ﬁnite, 64
unique factorization of ideals, 43
unit, 79, 82
topology
padic, 102
trace, 25
unique factorization domain, 2
unit, 1
unramiﬁed, 66
valuation
archimedean, 100
discrete, 102
multiplicative, 100
trivial, 100
wildly ramiﬁed, 126
156