J.S. Milne
Version 3.04
April 16, 2012
An algebraic number eld is a nite extension of ; an algebraic number is an element
of an algebraic number eld. Algebraic number theory studies the arithmetic of algebraic
number elds the ring of integers in the number eld, the ideals and units in the ring of
integers, the extent to which unique factorization holds, and so on.
An abelian extension of a eld is a Galois extension of the eld with abelian Galois
group. Class eld theory describes the abelian extensions of a number eld in terms of the
arithmetic of the eld.
These notes are concerned with algebraic number theory, and the sequel with class eld
theory.
BibTeX information
@misc{milneANT,
author={Milne, James S.},
title={Algebraic Number Theory (v3.04)},
year={2012},
note={Available at www.jmilne.org/math/},
pages={160}
}
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.
v3.03 (May 29, 2011). Minor xes; 167 pages.
v3.04 (April 16, 2012). Minor xes; 160 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, 2011, 2012 J.S. Milne.
Single paper copies for noncommercial personal use may be made without explicit permis
sion from the copyright holder.
Contents
Notations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1 Preliminaries from Commutative Algebra 15
Basic denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Ideals in products of rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Noetherian rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Noetherian modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Local rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Rings of fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
The Chinese remainder theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Review of tensor products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2 Rings of Integers 26
First proof that the integral elements form a ring . . . . . . . . . . . . . . . . . . 26
Dedekinds proof that the integral elements form a ring . . . . . . . . . . . . . . 27
Integral elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Review of bases of modules . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Review of norms and traces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Review of bilinear forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Discriminants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Rings of integers are nitely generated . . . . . . . . . . . . . . . . . . . . . . . 36
Finding the ring of integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Algorithms for nding the ring of integers . . . . . . . . . . . . . . . . . . . . . 41
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3 Dedekind Domains; Factorization 46
Discrete valuation rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Dedekind domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Unique factorization of ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
The ideal class group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Discrete valuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Integral closures of Dedekind domains . . . . . . . . . . . . . . . . . . . . . . . 56
Modules over Dedekind domains (sketch). . . . . . . . . . . . . . . . . . . . . . 57
3
Factorization in extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
The primes that ramify . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Finding factorizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Examples of factorizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Eisenstein extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4 The Finiteness of the Class Number 67
Norms of ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Statement of the main theorem and its consequences . . . . . . . . . . . . . . . . 69
Lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Some calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
Finiteness of the class number . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Binary quadratic forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5 The Unit Theorem 84
Statement of the theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Proof that U
1
is nitely generated . . . . . . . . . . . . . . . . . . . . . . . . . 86
Computation of the rank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Sunits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Example: CM elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Example: real quadratic elds . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Example: cubic elds with negative discriminant . . . . . . . . . . . . . . . . . 91
Finding j(1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Finding a system of fundamental units . . . . . . . . . . . . . . . . . . . . . . . 92
Regulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6 Cyclotomic Extensions; Fermats Last Theorem. 94
The basic results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Class numbers of cyclotomic elds . . . . . . . . . . . . . . . . . . . . . . . . . 100
Units in cyclotomic elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
The rst case of Fermats last theorem for regular primes . . . . . . . . . . . . . 101
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7 Absolute Values; Local Fields 104
Absolute Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Nonarchimedean absolute values . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Equivalent absolute values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Properties of discrete valuations . . . . . . . . . . . . . . . . . . . . . . . . . . 108
Complete list of absolute values for the rational numbers . . . . . . . . . . . . . 108
The primes of a number eld . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
The weak approximation theorem . . . . . . . . . . . . . . . . . . . . . . . . . 112
Completions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Completions in the nonarchimedean case . . . . . . . . . . . . . . . . . . . . . . 114
Newtons lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Extensions of nonarchimedean absolute values . . . . . . . . . . . . . . . . . . . 121
Newtons polygon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Locally compact elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Unramied extensions of a local eld . . . . . . . . . . . . . . . . . . . . . . . 126
Totally ramied extensions of 1 . . . . . . . . . . . . . . . . . . . . . . . . . . 127
Ramication groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
Krasners lemma and applications . . . . . . . . . . . . . . . . . . . . . . . . . 130
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
8 Global Fields 133
Extending absolute values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
The product formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Decomposition groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
The Frobenius element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Computing Galois groups (the hard way) . . . . . . . . . . . . . . . . . . . . . . 142
Computing Galois groups (the easy way) . . . . . . . . . . . . . . . . . . . . . . 142
Applications of the Chebotarev density theorem . . . . . . . . . . . . . . . . . . 147
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
A Solutions to the Exercises 150
B Twohour examination 156
Bibliography 157
Index 159
Notations.
We use the standard (Bourbaki) notations: N={0. 1. 2. . . .]; Z =ring of integers; 1=eld
of real numbers; C = eld of complex numbers; F
]
=Z,Z = eld 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, + 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 nite). Let 1 and be sets; a family of elements of indexed
by 1, denoted (a
i
)
iJ
, is a function i a
i
: 1 .
X Y X is a subset of Y (not necessarily proper);
X
def
=Y X is dened to be Y , or equals Y by denition;
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.
monnnnn question nnnnn in mathoverow.net
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 Q
Prerequisites
The algebra usually covered in a rstyear 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/):
FT Fields and Galois Theory, v4.22, 2011.
GT Group Theory, v3.11, 2011.
CFT Class Field Theory, v4.01, 2011.
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; Edgar Costa, Paul Federbush; Hauwen Huang;
Keenan Kidwell; Roger Lipsett; Loy Jiabao, Jasper; Lee M. Goswick; Samir Hasan; Lars
Kindler; Franz Lemmermeyer; Siddharth Mathur; Bijan Mohebi; Scott Mullane; Safak
Ozden; Wai Yan Pong; Nicol as Sirolli; Thomas Stoll; Bhupendra Nath Tiwari; Vishne Uzi;
and others.
PARI is an open source computer algebra system freely available from http://pari.math.u
bordeaux.fr/.
6
DRAMATIS PERSON
FERMAT (16011665). Stated his last theorem, and proved it for m =4. He also posed
the problem of nding integer solutions to the equation,
X
2
Y
2
=1. Z. (1)
which is essentially the problem
1
of nding the units in Z
_
. The English mathemati
cians found an algorithm for solving the problem, but neglected to prove that the algorithm
always works.
EULER (17071783). He introduced analysis into the study of the prime numbers, and he
discovered an early version of the quadratic reciprocity law.
LAGRANGE (17361813). 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 (17521833). 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 : a quadratic form Q(X. Y. 7) has a nontrivial
zero in if and only if it has one in 1 and the congruence Q0 mod
n
has a nontrivial
solution for all and n.
GAUSS (17771855). He found the rst complete proofs of the quadratic reciprocity law.
He studied the Gaussian integers Zi  in order to nd a quartic reciprocity law. He studied
the classication of binary quadratic forms over Z, which is closely related to the problem
of nding the class numbers of quadratic elds.
DIRICHLET (18051859). 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 coefcients; suppose that (X) has r real roots and 2s complex roots;
then Z
o
n=k
a
n
n
, a
n
{0. 1. . . . . 1]).
HILBERT (18621943). He wrote a very inuential book on algebraic number theory in
1897, which gave the rst systematic account of the theory. Some of his famous problems
were on number theory, and have also been inuential.
TAKAGI (18751960). He proved the fundamental theorems of abelian class eld theory,
as conjectured by Weber and Hilbert.
NOETHER (18821935). Together with Artin, she laid the foundations of modern algebra
in which axioms and conceptual arguments are emphasized, and she contributed to the
classication of central simple algebras over number elds.
HECKE (18871947). Introduced Hecke 1series generalizing both Dirichlets 1series and
Dedekinds zeta functions.
ARTIN (18981962). He found the Artin reciprocity law, which is the main theorem of
class eld theory (improvement of Takagis results). Introduced the Artin 1series.
HASSE (18981979). He gave the rst proof of local class eld theory, proved the Hasse
(localglobal) principle for all quadratic forms over number elds, and contributed to the
classication of central simple algebras over number elds.
BRAUER (19011977). Dened the Brauer group, and contributed to the classication of
central simple algebras over number elds.
WEIL (19061998). Dened the Weil group, which enabled him to give a common gener
alization of Artin 1series and Hecke 1series.
CHEVALLEY (190984). The main statements of class eld 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 eld theory for
innite abelian extensions.
IWASAWA (19171998). He introduced an important new approach into algebraic number
theory which was suggested by the theory of curves over nite elds.
TATE (1925 ). He proved new results in group cohomology, which allowed him to give
an elegant reformulation of class eld theory. With Lubin he found an explicit way of
generating abelian extensions of local elds.
LANGLANDS (1936 ). The Langlands program is a vast series of conjectures that, among
other things, contains a nonabelian class eld theory.
8
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 elds].
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
n
a
1
n1
a
0
=0
with coefcients a
1
. . . . . a
n
in , and is an algebraic integer if it satises such an equation
with coefcients 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 difcult to apply. We shall show (2.11) that is an algebraic
integer if and only if its minimum polynomial over has coefcients in Z.
Consider for example the eld 1 =
_
J, where J is a squarefree integer. The
minimum polynomial of =ab
_
J, b =0, a. b , is
(X (ab
_
J))(X (ab
_
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.
9
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
=Z
_
J =
_
ab
_
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
=Z
1
_
d
2
 =
_
ab
1
_
d
2
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
_
5.
Let
d
be a primitive Jth root of 1, for example,
d
=exp(2i,J), and let 1 =
d
.
Then we shall see (6.2) that
O
1
=Z
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
cant 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 Z
_
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:
_
5 . ab
_
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 =, then Nm() =Nm(1
_
5) =6. Thus Nm() =1. 2. 3, or 6. In the
rst case, is a unit, the second and third cases dont occur, and in the fourth case 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 =(ab
_
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
10
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 Z
_
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 dene 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[ab: 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 =ab 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 sees 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 denitions, 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.
11
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. 22
_
5. 22
_
5. 6). Since every gen
erator is divisible by 2, we see that
(2. 1
_
5)(2. 1
_
5) (2).
Conversely,
2 =64 (4. 22
_
5. 22
_
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 Z
_
5,(3. 1
_
5) is surjective with kernel (3), and so
Z
_
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 nite number: we shall see that there exists a nite
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 nite 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 innitely many units. For example, (1
_
2) is a unit of innite
order in Z
_
2 :
(1
_
2)(1
_
2) =1: (1
_
2)
n
=1 if m=0.
In fact Z
_
2
={(1
_
2)
n
[ m Z], and so
Z
_
2
1
~j(1) Z
i
(as an abelian group);
moreover, we shall nd r.
12
APPLICATIONS
One motivation for the development of algebraic number theory was the attempt to prove
Fermats 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 be proved by the method of innite descent, i.e., from one so
lution, you show that you can construct a smaller solution, which leads to a contradiction
2
.
The proof makes use of the factorization
Y
3
=7
3
X
3
=(7X)(7
2
X7X
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 coefcients 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 Fermats 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 es proof in which he assumed that, in order
to show that each factor (:
i
,) is a th power, it sufces to show that the factors are
relatively prime in pairs and their product is a th power. In fact, Lam e couldnt justify his
step (Z is not always a principal ideal domain), and Fermats last theorem was not proved
for almost 150 years. However, shortly after Lam es embarrassing lecture, Kummer used
his results on the arithmetic of the elds  to prove Fermats last theorem for all regular
primes, i.e., for all primes such that does not divide the class number of
]
.
Another application is to nding Galois groups. The splitting eld of a polynomial
(X) X is a Galois extension of . In a basic Galois theory course, 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 rst assumes the reader knows number theory, whereas
the second develops the whole subject algorithmically. Cohens book is the more useful as
a supplement to this course, but wasnt available when these notes were rst written. While
the books are concerned with moreorless practical algorithms for elds of small degree
and small discriminant, Lenstra (1992) concentrates on nding good general algorithms.
2
The simplest proof by innite descent is that showing that
_
2 is irrational.
13
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 Fermats last theorem. The notes in Narkiewicz 1990
document the origins of most signicant results in algebraic number theory. Lemmermeyer
2009, which explains the origins of ideal numbers, and other writings by the same author,
e.g., Lemmermeyer 2000, 2007.
Exercises
01 Let J be a squarefree integer. Complete the verication that the ring of integers in
_
J is as described.
02 Complete the verication that, in Z
_
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.
14
CHAPTER 1
Preliminaries from Commutative
Algebra
Many results that were rst proved for rings of integers in number elds are true for more
general commutative rings, and it is more natural to prove them in that context.
1
Basic denitions
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
1
. ....
n
of T,
1
. ....
n
 denotes the smallest subring of T containing and
the
i
. It consists of all polynomials in the
i
with coefcients in , i.e., elements of the
form
a
i
1
...i
m
i
1
1
...
i
m
n
. a
i
1
...i
m
.
We also refer to
1
. ....
n
 as the subalgebra of T generated by the
i
, and when
T =
1
. ....
n
 we say that the
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 nite 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 dene
ab ={ab [ 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 ab =(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 aa; then b
1
(b) denes 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.
15
16 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 eld. Every proper ideal a of is contained in a maximal
ideal if is Noetherian (see below) this is obvious; otherwise the proof requires Zorns
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
pT (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. 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) ab.
Recall that an ideal c C is prime if and only if C,c is an integral domain. The map
T ,aT,b. (a. b) (aa. b b)
has kernel ab, and hence induces an isomorphism
(T),(ab) .,aT,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 nite 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 nitely generated.
PROPOSITION 1.3 The following conditions on a ring are equivalent:
(a) is Noetherian.
(b) Every ascending chain of ideals
a
1
a
2
a
n
eventually becomes constant, i.e., for some n, a
n
=a
n1
= .
Noetherian modules 17
(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 nitely generated, say a =
(a
1
. . . . . a
i
). For some n, a
n
will contain all the a
i
, and so a
n
=a
n1
= =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 nitely many steps.
(c)=(a): Let a be an ideal in , and let S be the set of nitely 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 nitely generated ideal in a properly containing a
t
. This contradicts
the denition of a
t
.
2
A famous theorem of Hilbert states that kX
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 kX
1
. . . . . X
n
. . . .
of polynomials in an innite sequence of symbols is not Noetherian because the chain of
ideals
(X
1
) (X
1
. X
2
) (X
1
. X
2
. X
3
)
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) (b) b[a, with equality if and only if b =aunit.
The rst 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 cant 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 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.
Noetherian modules
Let be a ring. An module M is said to be Noetherian if every submodule is nitely
generated.
18 1. Preliminaries from Commutative Algebra
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
M
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 M
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 nitely 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 n1 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 (NAKAYAMAS LEMMA) Let be a local Noetherian ring, and let a be a
proper ideal in . Let M be a nitely generated module, and dene
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.
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
(1a
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
].
Rings of fractions 19
(b) It sufces to show that a(M,N) =M,N for then (a) shows that M,N =0. Con
sider mN, m M. From the assumption, we can write
m=n
a
i
m
i
, with a
i
a, m
i
M.
Then
mN =
(a
i
m
i
N) =
a
i
(m
i
N)
and so mN a(M,N).
2
The hypothesis that M be nitely generated in the lemma is essential. For example, if
is a local integral domain with maximal ideal m = 0, then mM = M for any eld M
containing but M =0.
Rings of fractions
Let be an integral domain; there is a eld 1 , called the eld 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, is the eld of fractions of Z, and k(X) is the eld of fractions of kX.
Let be an integral domain with eld 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 dene
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 whose denominator divides
some power of J:
Z
d
={a,J
n
[ 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 [ n is not divisible by ].
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.
2
This notation conicts with a later notation in which Z
]
denotes the ring of adic integers.
20 1. Preliminaries from Commutative Algebra
PROOF. Let a be an ideal in S
1
. Clearly (a )
e
a because a 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
pS = 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
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 dening ideals a and b
in a ring to be relatively prime if ab =.
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
a
1
a
2
... a
n
.
The Chinese remainder theorem 21
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 rst 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 nd 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 satises the requirements.
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.
22 1. Preliminaries from Commutative Algebra
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 : MN 1 is said to be bilinear if
(mm
t
. n) =(m. n) (m
t
. n)
(m. nn
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 (Q. ) consisting of an module Q and an
bilinear map : M N Q is called the tensor product of M and N if any other 
bilinear map
t
: M N 1 factors uniquely into
t
= with : Q 1 linear.
The tensor product exists, and is unique (up to a unique isomorphism making the obvious
diagram commute). We denote it by M
N 1; thus
Bilin
(MN. 1) =Hom
(M
N. 1).
(b) The module M
i
M
i
_
}
N
}
_
:
i,}
M
i
N
}
.
(
m
i
) (
n
}
)
m
i
n
}
.
It follows that if M and N are free modules
3
with bases (e
i
) and (
}
) respectively, then
M
N
t
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 coefcients 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.
Review of tensor products 23
is bilinear, and therefore factors uniquely through M N M
N. Thus there is a
unique linear map : M
N M
t
N
t
such that
( )(mn) =(m) (n).
REMARK 1.16 The tensor product of two matrices regarded as linear maps is called their
Kronecker product.
4
If is mn (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 : N N
t
are surjective, then so also is
: M
N M
t
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
=(n) for some n N,
and so m
t
n
t
=(m) (n) =()(mn). Therefore the image of contains a
set of generators for M
t
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 M
1 M
tt
1 0.
For example, if we tensor the exact sequence
0 a ,a 0
with M, we obtain an exact sequence
a
M M (,a)
M 0 (2)
The image of a
M in M is
aM
def
={
a
i
m
i
[ a
i
a, m
i
M].
and so we obtain from the exact sequence (2) that
M,aM .(,a)
M (3)
By way of contrast, if M N is injective, then M
1 N
,a
i
with M, we get an isomorphism
M,aM .(,a)
M
:
(,a
i
)
M .
M,a
i
M.
as required.
2
4
Kronecker products of matrices predate tensor products by about 70 years.
24 1. Preliminaries from Commutative Algebra
EXTENSION OF SCALARS
If T is an algebra and M is an module, then T
t
: T
M N such that
t
(1m) =(m). Thus
t
denes an isomorphism
Hom
(M. N) Hom
B
(T
M. N), N a Tmodule.
For example,
M
is a free Tmodule with basis 1e
1
. . . . . 1e
n
.
TENSOR PRODUCTS OF ALGEBRAS
If : T and g: C are algebras, then T
C is a (a) 1 =1g(a).
For example, there is a canonical isomorphism
a a : 1
k
kX
1
. . . . . X
n
 1X
1
. . . . . X
n
 (4)
TENSOR PRODUCTS OF FIELDS
We are now able to compute 1
k
if 1 is a nite separable eld extension of a eld k
and is an arbitrary eld extension of k. According to the primitive element theorem (FT
5.1), 1 =k for some 1. Let (X) be the minimum polynomial of . By denition
this means that the map g(X) g() determines an isomorphism
kX,((X)) 1.
Hence
1
k
.(kX,((X)))
k
.X,((X))
by (3) and (4). Because 1 is separable over k, (X) has distinct roots. Therefore (X)
factors in X 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
X,((X)) =
i
i=1
X,(
i
(X)).
Finally, X,(
i
(X)) is a nite separable eld extension of of degree deg
i
. Thus we
have proved the following result:
Exercise 25
THEOREM 1.18 Let 1 be a nite separable eld extension of k, and let be an arbitrary
eld extension. Then 1
k
is a product of nite separable eld 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 = with algebraic over . Then
C
1 .C
(X,((X))) .CX,(((X)) .
i
i=1
CX,(X
i
) ~C
i
.
Here
1
. . . . .
i
are the conjugates of in C. The composite of 1: 1 C
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
=( 11) =0 such that
]
=
]
11
]
=
]
(11)
]
(11) =0.
Hence 1
k
1 contains a nonzero nilpotent element, and so it cant be a product of elds.
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).
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
.
CHAPTER 2
Rings of Integers
Let be an integral domain, and let 1 be a eld containing . An element of 1 is said
to be integral over if it is a root of a monic polynomial with coefcients in , i.e., if it
satises 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 rst uses Newtons theory of symmetric
polynomials and a result of Eisenstein, and the second is Dedekinds surprisingly modern
proof, which avoids symmetric polynomials.
First proof that the integral elements form a ring
A polynomial 1(X
1
. .... X
i
) X
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 X
1
. .... X
i
 is equal to a polynomial in the symmetric elementary
polynomials with coefcients in , i.e., 1 S
1
. .... S
i
.
PROOF. We dene 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
26
Dedekinds proof that the integral elements form a ring 27
or equality holds and, for some s,
i
1
=
1
. . . . . i
x
=
x
. but i
x1
>
x1
.
Let X
k
1
1
X
k
r
i
be the highest monomial occurring in 1 with a coefcient c =0. Because
1 is symmetric, it contains all monomials obtained fromX
k
1
1
X
k
r
i
by permuting the Xs.
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 nite 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
X, 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 eld
containing . If
1
. . . . .
n
are the roots in of a monic polynomial in X, then any
polynomial g(
1
. . . . .
n
) in the
i
with coefcients in is a root of a monic polynomial in
X.
PROOF. Clearly
h(X)
def
=
cSym
n
(X g(
c(1)
. . . . .
c(n)
))
is a monic polynomial whose coefcients 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 X 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 .
Dedekinds proof that the integral elements form a ring
PROPOSITION 2.4 Let 1 be a eld containing . An element of 1 is integral over A if
and only if there exists a nonzero nitely generated submodule of 1 such that M M
(in fact, we can take M =, the subalgebra generated by ).
28 2. Rings of Integers
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 M.
==: We shall need to apply Cramers 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
}
=
c
11
. . .
c
1}
.
}
. . . c
1n
.
.
.
.
.
.
.
.
.
c
n1
. . .
c
n}
.
}
. . . c
nn
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 coefcients on the lefthand side. Then Cramers rule tells us that
det(C)
i
=0 for all i . Since at least one
i
is nonzero and we are working inside the eld
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 be two elements of 1 integral over , and let
M and N be nitely generated modules in 1 such that M M and N N. Dene
MN =
_
m
i
n
i
[ m
i
M. n
i
N
_
.
Then:
(a) MN is an submodule of 1 (easy);
(b) it is nitely 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 and by .
We can now apply (2.4) to deduce that and are integral over .
Integral elements 29
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 eld 1 is called the ring of
integers O
1
in 1.
Next we want to see that 1 is the eld of fractions of O
1
; in fact we can prove more.
PROPOSITION 2.6 Let 1 be the eld of fractions of , and let 1 be a eld containing 1.
If 1 is algebraic over 1, then there exists a J such that J is integral over .
PROOF. By assumption, satises 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 eld of fractions 1, and let T be the
integral closure of in a eld 1 containing 1. If 1 is algebraic over 1, then it is the eld
of fractions of T.
PROOF. The proposition shows that every 1 can be written = ,J with T,
J .
2
DEFINITION 2.8 A ring is integrally closed if it is its own integral closure in its eld 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 eld 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 satises 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 doesnt divide a, this is a contradiction.
2
30 2. Rings of Integers
The proposition makes it easy to give examples of rings where unique factorization fails
take any ring which is not integrally closed, for example, Z
_
5.
EXAMPLE 2.10 (a) The rings Z and Zi  are integrally closed because both are principal
ideal domains.
(b) Unique factorization fails in Z
_
3 because
4 =22 =(1
_
3)(1
_
3).
and the four factors are all irreducible because they have the minimum norm 4. However,
Z
_
3 Z
3
_
1 which is a principal ideal domain (and hence the integral closure of Z in
_
3 =
3
_
1).
(c) For any eld k, I claim that the integral closure of kS
1
. .... S
n
 in k(X
1
. . . . . X
n
) is
kX
1
. . . . . X
n
 (here the S
i
are the elementary symmetric polynomials).
Let k(X
1
. . . . . X
n
) be integral over kS
1
. . . . . S
n
. Then is integral over kX
1
. . . . . X
n
,
which is a unique factorization domain, and hence is integrally closed in its eld of frac
tions. Thus kX
1
. . . . . X
n
.
Conversely, let kX
1
. . . . . X
n
. Then is a root of the monic polynomial
cSym
m
(T (X
c(1)
. . . . . X
c(n)
)).
The coefcients of this polynomial are symmetric polynomials in the X
i
, and therefore (see
2.2) lie in kS
1
. . . . . S
i
.
PROPOSITION 2.11 Let 1 be the eld of fractions of , and let 1 be an extension of 1 of
nite degree. Assume is integrally closed. An element of 1 is integral over if and
only if its minimum polynomial over 1 has coefcients 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 minimumpolynomial of over 1. For any root
t
of (X), the elds 1
and 1
t
 are both stem elds for (see FT p. 15), and so there exists a 1isomorphism
o: 1 1
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 coefcients 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
_
J is integral over Z if and only if its trace
and norm both lie in Z.
Integral elements 31
PROPOSITION 2.13 If T is integral over and nitely generated as an algebra, then it
is nitely generated as an module.
PROOF. First consider the case that T is generated as an algebra by a single element, say
T =. By assumption
n
a
1
n1
a
n
=0. some a
i
.
Every element of T can be expressed as a nite sum
c
0
c
1
c
2
2
c
1
1
. c
i
,
and we can exploit the preceding equality to replace
n
(successively) with a linear combi
nation of lower powers of . Thus every element of T can be expressed as a nite sum
c
0
c
1
c
2
2
c
n1
n1
, c
i
.
and so 1. .
2
. . . . .
n1
generate T as an module. In order to pass to the general case,
we need a lemma.
2
LEMMA 2.14 Let T C be rings. If T is nitely generated as an module, and C
is nitely generated as a Tmodule, then C is nitely generated as an module.
PROOF. If {
1
. ....
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 {
i
;
}
] is a set of generators for C as an 
module.
2
We now complete the proof of (2.13). Let
1
. . . . .
n
generate T as an algebra, and
consider
1

1
.
2

1
. ....
n
 =T.
We sawabove that
1
 is nitely generated as an module. Since
1
.
2
 =
1

2
,
and
2
is integral over
1
 (because it is over ), the same observation shows that
1
.
2
 is nitely generated as a
1
module. Now the lemma shows that
1
.
2

is nitely generated as an module. Continuing in this fashion, we nd that T is nitely
generated as an module.
PROPOSITION 2.15 Consider integral domains T C; if T is integral over , and C
is integral over T, then C is integral over .
PROOF. Let ; C; it satises an equation
;
n
b
1
;
n1
b
n
=0. b
i
T.
Let T
t
=b
1
. .... b
n
. Then T
t
is nitely 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
; is nitely generated
as an module. Since ;T
t
; T
t
;, Proposition 2.4 shows that ; is integral over .
2
COROLLARY 2.16 The integral closure of in an algebraic extension 1 of its eld of
fractions is integrally closed.
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 T.
2
32 2. Rings of Integers
REMARK 2.17 In particular, the ring of integers in a number eld 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 nite eld, and let 1 be a nite extension of k(X). Let O
1
be
the integral closure of kX in 1. The arithmetic of O
1
is very similar to that of the ring
of integers in a number eld.
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 nn matrices with coefcients 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 T be rings such that T is a free module of rank n. Then any T denes an
linear map
. .: T T.
and the trace and determinant of this map are welldened. We call them the trace Tr
B{
and norm Nm
B{
of in the extension T,. Thus if {e
1
. .... e
n
] is a basis for T over ,
and e
i
=
a
i}
e
}
, then Tr
B{
() =
a
i i
and Nm
B{
() =det(a
i}
). When T is a
nite eld extension, this agrees with the usual denition. The following hold (for a ,
.
t
T):
Tr(
t
) =Tr() Tr(
t
) Nm(
t
) =Nm() Nm(
t
)
Tr(a) =aTr() Nm(a) =a
n
Tr(a) =na
PROPOSITION 2.19 Let 1,1 be an extension of elds of degree n, and let 1. Let
(X) be the minimum polynomial of over 1 and let
1
=,
2
, ... ,
n
be the roots of
(X). Then
Tr
1{1
() =r(
1
n
). Nm
1{1
() =(
1
n
)
i
where r =1 : 1 =n,m.
Review of bilinear forms 33
PROOF. Suppose rst that 1 = 1, and compute the matrix of . . relative to the
basis {1. . . . . .
n1
] one sees easily that it has trace
i
and determinant
i
. For
the general case, use the transitivity of norms and traces (see FT 5.44).
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
() =o
1
o
n
. Nm
1{1
() =o
1
o
n
.
PROOF. Each
i
occurs exactly r times in the family {o
i
].
2
COROLLARY 2.21 Let be an integrally closed integral domain, and let 1 be a nite
extension of the eld of fractions 1 of ; if 1 is integral over , then Tr
1{1
() and
Nm
1{1
() are in .
PROOF. We know that if is integral, then so also is each of its conjugates. Alternatively,
apply 2.11.
2
ASIDE 2.22 Let 1 = 1, and let
1
= .
2
. . . . .
n
be the conjugates of (in some Galois
extension of 1 containing 1). For any =g() in 1.
Nm
1{1
() =
n
i=1
g(
i
). Tr
1{1
() =
n
i=1
g(
i
).
This is a restatement of (2.20), and is Dedekinds original denition (Dedekind 1877, 17).
Review of bilinear forms
Let V be a nitedimensional vector space over a eld 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
) =
i,}
[(a
ki
e
i
. a
I}
e
}
) =
i,}
a
ki
[(e
i
. e
}
) a
I}
.
and so
([(
k
.
I
)) = ([(e
i
. e
}
))
tr
(equality of mm 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 satises 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.
34 2. Rings of Integers
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
}
) =
i}
(Kronecker delta). We can use the isomorphism
V V
}
)).
LEMMA 2.23 If ;
}
=
a
}i
i
, a
i}
, then
D(;
1
. .... ;
n
) =det(a
i}
)
2
D(
1
. ....
n
).
PROOF. See the proof of (5).
2
If the s and ;s each form a basis for T over , then det(a
i}
) is a unit (see p32). Thus
the discriminant D(
1
. ....
n
) of a basis {
1
. ....
n
] of T is welldened 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(
1
. ....
n
) itself regarded as an element of
,
2
, is called the discriminant disc(T,) of T over .
For example, when we have a nite extension of elds 1,1, disc(1,1) is an element
of 1, welldened up to multiplication by a nonzero square in 1.
When =Z, disc(T,) is a welldened integer, because 1 is the only square of a unit
in Z.
Warning: We shall see shortly that, when 1 is a number eld of degree m over ,
the ring of integers O
1
in 1 is free of rank m over Z, and so disc(O
1
,Z) is a well
dened integer. Sometimes this is loosely referred to as the discriminant of 1, strictly
speaking, disc(1,) is the element of
,
2
represented by the integer disc(O
1
,Z).
PROPOSITION 2.24 Let T 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 ).
Discriminants 35
PROOF. Let {
1
. ....
n
] be a basis for T as an module, and let ;
1
. .... ;
n
be any elements
of T. Write ;
}
=
a
}i
i
, a
}i
. Then
D(;
1
. .... ;
n
)
(2.23)
= det(a
i}
)
2
D(
1
. ....
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
nite 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
1
. . . . .
n
for T as a Zmodule, and write ;
}
=
a
}i
i
.
Then both sides equal det(a
i}
)
2
D(
1
. . . . .
n
).
PROPOSITION 2.26 Let 1 be a nite separable extension of the eld 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
1
. ....
n
of 1 over 1.
D(
1
. ....
n
) =det(o
i
}
)
2
=0.
PROOF. By direct calculation, we have
D(
1
. . . . .
n
)
def
=det(Tr(
i
}
))
=det(
k
o
k
(
i
}
)) (by 2.20)
=det(
k
o
k
(
i
) o
k
(
}
))
=det(o
k
(
i
)) det(o
k
(
}
))
=det(o
k
(
i
))
2
.
Suppose that det(o
i
}
) =0. Then there exist c
1
. .... c
n
such that
i
c
i
o
i
(
}
) =0 all .
2
By linearity, it follows that
i
c
i
o
i
() =0 for all 1, but this contradicts Dedekinds
theorem on the independence of characters (apply it with G =1
):
Let G be a group and a eld, 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).
COROLLARY 2.27 Let 1 be the eld of fractions of , and let 1 be a nite 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 {
1
. ....
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
(.
t
) Tr(
t
): 11 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).
36 2. Rings of Integers
Rings of integers are nitely generated
We now show that O
1
is nitely generated as a Zmodule.
PROPOSITION 2.29 Let be an integrally closed integral domain with eld 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 T M
t
. (6)
Therefore T is a nitely generated module if is Noetherian, and it is free of rank m if
is a principal ideal domain.
PROOF. Let {
1
. ....
n
] be a basis for 1 over 1. According to (2.6), there exists a J
such that J
i
T for all i . Clearly {J
1
. . . . . J
n
] is still a basis for 1as a vector space
over 1, and so we may assume to begin with that each
i
T. Because the trace pairing is
nondegenerate, there is a dual basis {
t
1
. ....
t
n
] of 1 over 1 such that Tr(
i
t
}
) =
i}
(see the discussion following (5), p33). We shall show that
2
n
T
t
1
t
2
t
n
.
Only the second inclusion requires proof. Let T. Then can be written uniquely as a
linear combination =
b
}
t
}
of the
t
}
with coefcients b
}
1, and we have to show
that each b
}
. As
i
and are in T, so also is
i
, and so Tr(
i
) (see 2.21).
But
Tr(
i
) =Tr(
}
b
}
t
}
i
) =
}
b
}
Tr(
t
}
i
) =
}
b
}
i}
=b
i
.
Hence b
i
.
If Noetherian, then M
t
is a Noetherian module (see 1.8), and so T is nitely
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
COROLLARY 2.30 The ring of integers in a number eld 1 is the largest subring that is
nitely generated as a Zmodule.
PROOF. We have just seen that O
1
is a nitely generated Zmodule. Let T be another
subring of 1 that is nitely generated as a Zmodule; then every element of T is integral
over Z (by 2.4), and so T O
1
.
2
REMARK 2.31 (a) The hypothesis that 1,1 be separable is necessary to conclude that
T is a nitely generated module (we used that the trace pairing was nondegenerate).
However it is still true that the integral closure of kX in any nite extension of k(X) (not
necessarily separable) is a nitely generated kXmodule.
(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 elds 1,1 such that O
1
is not a free O
1
module.
(c) Here is an example of a nitely generated module that is not free. Let =Z
_
5,
and consider the modules
(2) (2. 1
_
5) Z
_
5.
Rings of integers are nitely generated 37
Both (2) and Z
_
5 are free Z
_
5modules of rank 1, but (2. 1
_
5) is not a free
Z
_
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 eld, 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 =
i
T, with
i
a basis for 1 over 1. Dene
C
+
={ 1 [ Tr(;) for all ; C].
By linearity,
C
+
Tr(
i
) for i =1. .... m.
and it follows that
C
+
=
t
i
.
Thus we have:
C =
i
T
t
i
=C
+
.
(b) Write 1 = with T, and let (X) be the minimum polynomial of . Let
C =Z =Z1Z Z
n1
. We want to nd C
+
.
One can show (Artin 1959, Chapter 7) that
Tr(
i
,
t
()) =
_
0 if 0 _i _m2
1 if i =m1
(these formulas go back to Euler). It follows from this that
det(Tr(
i
}
,
t
()) =(1)
n
(the only term contributing to the determinant is the product of the elements on the other
diagonal). If
t
1
. ....
t
n
is the dual basis to 1. . . . . .
n1
, so that Tr(
i
t
}
) =
i}
, then
det(Tr(
i
t
}
)) =1.
On comparing these formulas, one sees that the matrix relating the family
{1,
t
(). ....
n1
,
t
()]
to the basis
{
t
1
. ....
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
(). . . . .
n1
,
t
()]:
C = T
t
()
1
 =C
+
.
38 2. Rings of Integers
Finding the ring of integers
We now assume 1 to be a eld of characteristic zero.
PROPOSITION 2.34 Let 1 =1 some , and let (X) be the minimum polynomial of
over 1. Suppose that (X) factors into
(X
i
) over the Galois closure of 1. Then
D(1. .
2
. . . . .
n1
) =
1_i~}_n
(
i
}
)
2
=(1)
n(n1){2
Nm
1{1
(
t
()).
PROOF. We have
D(1. .
2
. . . . .
n1
) =det(o
i
(
}
))
2
(2.26)
=det(
}
i
)
2
=(
i~}
(
i
}
))
2
(Vandermonde)
=(1)
n(n1){2
i
(
}yi
(
i
}
))
=(1)
n(n1){2
}
t
(
}
)
=(1)
n(n1){2
Nm(
t
()).
2
The number in (2.34) is called the discriminant of (X). It can also be dened 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
i
with coefcients in 1, and so
(by 2.2) it can be expressed in terms of the coefcients 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 be a root of (X), and let
; =
t
() =n
n1
a.
We compute Nm
1j{1
(;). On multiplying the equation
n
ab =0
by n
1
and rearranging, we obtain the equation
n
n1
=nanb
1
.
Hence
; =n
n1
a =(n1)anb
1
.
Solving for gives
=
nb
; (n1)a
.
from which it is clear that 1 =1;, and so the minimum polynomial of ; over 1 also
has degree n. If we write
_
nb
X (n1)a
_
=1(X),Q(X).
Finding the ring of integers 39
then 1(;),Q(;) =() =0 and so 1(;) =0. Since
1(X) =(X (n1)a)
n
na(X (n1)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 nd that
Nm(;) =n
n
b
n1
(1)
n1
(n1)
n1
a
n
.
Finally we obtain the formula:
disc(X
n
aX b) =(1)
n(n1){2
(n
n
b
n1
(1)
n1
(n1)
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 nding the ring of integers of 1 is to write 1 = 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
: Z)
2
. (7)
If it is not squarefree, {1. . ....
n1
] may still be an integral basis, and sometimes one
can tell this by using Stickelbergers 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).
EXAMPLE 2.36 The polynomial X
3
X1 is irreducible
1
in X, because, if it factored,
it would have a root in , 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  (and Z is
the ring of integers in ).
EXAMPLE 2.37 The polynomial X
3
X 1 is irreducible in X, 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 .
1
In fact, this is the monic irreducible cubic polynomial in ZX with the smallest discriminant.
40 2. Rings of Integers
EXAMPLE 2.38 This example goes back to Dedekind. Let 1 =, 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
=Z, and so
disc(O,Z) =503. In fact Dedekind showed that there is no integral basis of the form 1,
,
2
(see Weiss 1963, p170; for another example of this type, see Exercise 26.)
EXAMPLE 2.39 Consider the eld  where is a root of (X) = X
5
X 1. This
polynomial is irreducible, because it is irreducible in F
3
X. The discriminant of (X) is
2869 =19 151, and so the ring of integers in  is Z.
PROPOSITION 2.40 Let 1 be an algebraic number eld.
(a) The sign of disc(1,) is (1)
x
, where 2s is the number of homomorphisms 1 C
whose image is not contained in 1.
(b) (Stickelbergers theorem) disc(O
1
,Z) 0 or 1 mod 4.
PROOF. (a) Let 1 =, and let
1
=,
2
, ...,
i
be the real conjugates of and
i1
,
i1
. ....
ix
.
ix
the complex conjugates. Then
sign(D(1. ....
n1
)) =sign
_
1_i_x
(
ii
ii
)
_
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 , then either t1 =1
and tN =N, or t1 =N and tN =1. In either case, t xes 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
EXAMPLE 2.41 Consider the eld
_
m, where m is a squarefree integer.
Case m 2. 3 mod 4. Here D(1.
_
m) =disc(X
2
m) =4m, and so Stickelbergers
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 elds. 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 elds with discriminant 4027 (4027
is prime). See (3.48) and (3.49) for two of them.
Algorithms for nding the ring of integers 41
The curious may wonder why we didnt give an example of a eld generated over 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 nding 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 nite 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 eld
of fractions 1, and let T be the integral closure of in a nite separable extension 1 of 1
of degree m.
PROPOSITION 2.43 Let
1
. ....
n
be a basis for 1 over 1 consisting of elements of T,
and let J =disc(
1
. ....
n
). Then
1
...
n
T (
1
,J) ... (
n
,J).
PROOF. Let T, and write
=.
1
1
.
n
n
. .
i
1.
Let o
1
. . . . . o
n
be the distinct 1embeddings of 1 into some large Galois extension of
1. On applying the os to this equation, we obtain a system of linear equations:
o
i
=.
1
o
i
1
.
2
o
i
2
.
n
o
i
n
. i =1. . . . . m.
Hence by Cramers rule
.
i
=;
i
,
where =det(o
i
}
) and ;
i
is the determinant of the same matrix, but with the i th column
replaced with (o
i
). From (2.34), we know that
2
= J. Thus .
i
= ;
i
,J, and ;
i
is
an element of 1 (because it equals J.
i
) and is integral over . Therefore ;
i
, which
completes the proof.
2
Thus there is the following algorithm for nding the ring of integers in a number eld
1. Write 1 = where is integral over . Compute J =disc(1. . ....
n1
). Then
Z O
1
J
1
Z.
Note that (J
1
Z: Z) = J
n
, which is huge but nite. Each coset Z,
J
1
Z, consists entirely of algebraic integers or contains no algebraic integer. Find a
set of representatives
1
. ....
n
for Z in J
1
Z, and test each to see whether it is in
tegral over Z (the coefcients of its minimum polynomial will have denominators bounded
2
In fact, the smallest discriminant is 3, which occurs for
_
3.
42 2. Rings of Integers
by a power of J, and so it is possible to tell whether or not they are integers by computing
them with sufcient accuracy).
3
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 =
, then the index of Z 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
=Z.
Note that PARI can compute the minimum polynomial of an algebraic number. For
example, let a =
3
_
1
_
7. We rst 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 nd a minimum polynomial for a even when a is
transcendental.
I now discuss a practical algorithm for nding 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 (. ) be Euclidean domain, and let M be an mm matrix with coef
cients 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 denition : 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 =qar, with r =0 or (r) <(a).
Apply an operation of type (r2) so that the element of the rst column with the minimum is
in the (1. 1)position. If a
11
divides all elements in the rst column, we can use operations
of type (r1) to make all the remaining elements of the rst column zero. If not, we can use
(r1) to get an element in the rst column that has smaller value than a
11
, and put that
in the (1. 1) position. Repeat eventually, we will have the gcd of the original elements
in the rst column in the (1. 1) position and zeros elsewhere. Then move onto the next
column. . . .
2
REMARK 2.45 (a) The operations (r1) and (r2) are invertible in matrices with coefcients
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 nd 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:
3
If you know the ring of integers of a eld, it is easy to nd the discriminant. Conversely, if you know the
discriminant, this will help in nding the ring of integers; for example, you may get lucky and nd an such
that the discriminant of Z over Z is equal to the discriminant of O
1
.
Algorithms for nding the ring of integers 43
(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 eld 1 = generated over by 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 coefcients). Thus
o
t
1
=a
11
1:
o
t
2
=a
21
1a
22
:
etc.,
where J a
i}
Z, J =[ det(M)[ =[ det(T )[.
EXAMPLE 2.46 Let 1 =
_
m, m squarefree, m1 (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.
In an Appendix, they use it to show that , 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
44 2. Rings of Integers
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 Z in O
1
is 2
56
3
6
5
3
29
9
.
The rst step is to compute D(1. .
2
. . . .) =disc((X)) and to nd 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
difcult. Hundred digit numbers are already difcult. 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 nd 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 eld 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
4X1 has 5 real roots and no nonreal roots and that its stem
eld  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  (cf. p42).
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 eld
_
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
_
3.
For Dedekinds polynomial in (2.38), PARI says that it has one real root and one conju
gate pair of nonreal roots, and that its stem eld has discriminant 503. It nds the integral
basis {1. .
1
2
1
2
]. Note that
Z =Z1. .
2
.
and that
(O
1
: Z) =2 =
_
2012
503
,
as predicted by Equation 7, p39.
Exercises 45
NOTES As noted earlier, it was Dedekind who found the correct denition of the ring of integers
in a number elds. Earlier authors either luckily chose the correct ring, e.g., Kummer chose Z,
n
=1, which is the ring of integers in , or unluckily chose the wrong ring, e.g., Euler gave a
proof of Fermats last theorem for the exponent 3, which becomes correct when the ring Z
_
3 is
replaced in the proof by its integral closure Z,
3
=1.
Exercises
21 Since Z
_
5 is not integrally closed, it can not be a unique factorization domain.
Give an example of an element of Z
_
5 that has two distinct factorizations into irreducible
elements.
22 Let be an integrally closed ring, and let 1 be its eld of fractions. Let (X) X
be a monic polynomial. If (X) is reducible in 1X, show that it is reducible in X.
23 Show that if 1,1 is not separable, then disc(1,1) =0.
24 Let a =(2. 1
_
3) in Z
_
3. Show that a =(2), but a
2
=(2)a. Conclude that
ideals in Z
_
3 do not factor uniquely into prime ideals. (Hence Z
_
3 is the wrong
choice for the ring of integers in
_
3.)
25 Let be a subring of a ring T, and let be a unit in T. Show that every

1
 is integral over . [This has a short solution, but its not obvious.]
26 Let 1 =
_
7.
_
10, and let be an algebraic integer in 1. The following argument
will show that O
1
=Z.
(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 Z; deduce Tr(
n
i
) 1 (mod 3) in Z.]
(b) Assume now that O
1
=Z we shall derive a contradiction. Let (X) be the
minimum polynomial of over . For g(X) ZX, let g(X) denote the image of g in
F
3
X, F
3
=Z,(3). Show that g() is divisible by 3 in Z if and only if g is divisible by
in F
3
X.
(c) For each i , 1 _ i _ 4, let
i
be a polynomial in ZX such that
i
=
i
(). Show
that
[
}
(i = ) in F
3
X, 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?
CHAPTER 3
Dedekind Domains; Factorization
Es steht schon bei Dedekind.
(Its already in Dedekind.)
Emmy Noether
In this Chapter, we dene 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 elds are Dedekind domains,
but rst 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 eld.
A ring satisfying these conditions is called a discrete valuation ring. Later we shall dene
discrete valuations, and so justify the name.
EXAMPLE 3.1 The ring Z
(])
def
={
n
n
[ 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 welldened
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
46
Discrete valuation rings 47
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 rst
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 denition, pb (c), and so p
b
c
, and it is an
ideal in . If p
b
c
p, then
b
c
is integral over (by 2.4, since p is nitely 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 a
1
a
2
.
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 cant be contained in . Let
m be the smallest integer such that a
n
but a
n1
_ . Then a
n
_ p, and so
a
n
=. Hence a =(
n
).
2
48 3. Dedekind Domains; Factorization
Dedekind domains
DEFINITION 3.3 A Dedekind domain is an integral domain , not equal to a eld, 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 eld.
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 (nite) set of generators for a.
(b) Let be an element of the eld of fractions of (=eld 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
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.
Unique factorization of ideals 49
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
ab =1. Consider
1 =(ab)
i
=a
i
_
i
1
_
a
i1
b b
i
.
If r _ mn, 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
denition 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
ap
n
aq
n
: ,p
n
p
,q
n
is an isomorphism.
PROOF. We rst 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
mp
n
=mp), 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(1q), the image of ba
in
p
also has this property and therefore equals
o
x
.
2
50 3. Dedekind Domains; Factorization
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 rst 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 .
REMARK 3.12 Note that
s
i
>0 a
p
i
=
p
i
a p
i
.
COROLLARY 3.13 Let a and b be ideals in ; then
a b a
p
b
p
for all 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 sufciency, 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
b
p
i
r
i
_s
i
.
(recall that
p
i
is a discrete valuation ring) and r
i
_s
i
all i implies a b.
2
Unique factorization of ideals 51
COROLLARY 3.14 Let be an integral domain with only nitely 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 sufces
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 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
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
+
p
some prime ideal, and ac aa
+
=a(c a
+
) ap =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, kX. 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.
52 3. Dedekind Domains; Factorization
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 sufces 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 nitely generated module, because Ja is an integral ideal,
hence nitely generated, for some J =0, and the map . J.: a Ja is an isomorphism
of modules. Conversely, a nonzero nitely 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.
Every nonzero element b of 1 denes 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 dened 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 and eb ,
then Jeab . For principal fractional ideals, (a)(b) =(ab).
EXAMPLE 3.19 Let be a discrete valuation ring with maximal ideal p and eld 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 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.
The ideal class group 53
PROOF. We have noted that the law of composition is welldened. 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).
(b) Let S be a multiplicative subset in a Dedekind domain , and let
S
=S
1
. It is
an integral domain with the same eld of fractions as :
S
1.
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, dene
a
t
={a 1 [ aa ].
This is an module, and if J a, J = 0, then Ja
t
, and so a
t
is a fractional ideal.
From the denition 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
p becomes
bb
t
q, where b, b
t
, and q are the ideals in
p
generated by a, a
t
, and p. Moreover,
b
t
={a 1 [ ab
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 dene 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()
54 3. Dedekind Domains; Factorization
(when nite). In the case that is the ring of integers O
1
in a number eld 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
eld 1 is nite. Understanding how the class numbers of number elds vary remains an
interesting problem. For example, the class number of
_
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 difcult to show that these
elds 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 Heegners proof to be
correct). We have seen that Z
_
5 is not a principal ideal domain, and so cant have class
number 1 in fact it has class number 2. The method we use to prove that the class number
is nite is effective: it provides an algorithm for computing it. There are expected to be an
innite number of real quadratic elds 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 eld
_
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 eld).
1
EXAMPLE 3.22 Consider the afne elliptic curve
Y
2
=X
3
aX b. z=4a
3
27b
2
=0.
The associated ring = CX. 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, for some lattice in C.
2
PROPOSITION 3.23 Let be a Dedekind domain, and let S be a multiplicative set in .
Then a S
1
a denes 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 nite ideal class group. There is then a
nite 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 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:
1
Claborn, Luther. Every abelian group is a class group. Pacic J. Math. 18 1966 219222.
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,
(Milne 2006, I 4.10, III 3.10).
Discrete valuations 55
(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 eld. A discrete valuation on 1 is a nonzero homomorphism : 1
Z such
that (ab) _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)
EXAMPLE 3.26 (a) Let M be the eld 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, dene 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 eld of fractions 1, and let be a prime
element of . Then each element c of 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
nite order because is a homomorphism and Z has no elements of nite order); hence
(a) =(1) (a) =(a). Therefore, if (a) >(b), we have
(b) =(ab a)) _min((ab). (a)) _min((a). (b)) =(b).
and so equality must hold throughout, and this implies (ab) =(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 dened by p in (c).
Example (b) shows that every discrete valuation ring gives rise to a discrete valuation
on its eld 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]
56 3. Dedekind Domains; Factorization
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 denes 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
n1
i
are relatively prime in pairs, and so (1.14)
provides us with an element . such that
. .
i
mod p
n1
i
. i =1. 2. . . . . m.
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 elds are Dedekind
domains, and hence that their ideals factor uniquely into products of prime ideals.
THEOREM 3.29 Let be a Dedekind domain with eld of fractions 1, and let T be the
integral closure of in a nite separable extension 1 of 1. Then T is a Dedekind domain.
PROOF. We have to check the three conditions in the denition of a Dedekind domain (see
3.3). We rst show that T is Noetherian. In (2.29) we showed that T is contained in a
nitely generated module. It follows that every ideal in T is nitely 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 q, =0. Then is integral over
, and so there is an equation
n
a
1
n1
a
n
=0. a
i
.
which we may suppose to have the minimum possible degree. Then a
n
= 0. As a
n
T , 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 eld. We know T,q is an integral domain, and the map
ap aq
identies ,p with a subeld of T,q. s T is integral over , T,q is algebraic over ,p.
The next lemma shows that T,q is a eld, and hence that q is maximal.
2
Modules over Dedekind domains (sketch). 57
LEMMA 3.30 Any integral domain T containing a eld k and algebraic over k is itself a
eld.
PROOF. Let be a nonzero element of T we have to prove that it has an inverse in T.
Because is algebraic over k, the ring k is nitedimensional as a kvector space, and
the map . .: k k is injective (because T is an integral domain). From linear
algebra we deduce that the map is surjective, and so there is an element
t
k such that
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 difculty is that,
without the separability condition, T may fail to be nitely generated as an module, and
so the proof that it is Noetherian is more difcult.
Modules over Dedekind domains (sketch).
The structure theorem for nitely 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 nitely generated torsionfree module M need not be free. For
example, every fractional ideal is nitely 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 nitely generated torsionfree module M is isomorphic to a direct sum of
fractional ideals,
M ~a
1
a
n
.
(b) Two nitely 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 if
they dene 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 eld 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 nitely generated torsion
free modules of rank 1 can be identied 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 nitely generated modules is Cl() Z.
THEOREM 3.32 (INVARIANT FACTOR THEOREM) Let M N be nitely 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
.
58 3. Dedekind Domains; Factorization
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 nitely 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 eld of fractions 1, and let T be the integral closure of
in a nite separable extension 1 of 1.
A prime ideal p of will factor in T,
pT =P
e
1
1
P
e
g
. e
i
_1.
If any of the numbers is > 1, then we say that p is ramied in T (or 1). The number
e
i
is called the ramication 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 ramication index and (P,p) for
the degree of the eld extension T,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) =(1i )
2
in Zi , and so (2) ramies with ramication index 2. On
the other hand, (3) is inert in i  with residue eld Zi ,(3) = F
9
, and (5) splits as the
product of two prime ideals (5) =(2i )(2i ).
LEMMA 3.33 A prime ideal P of T divides p if and only if p =P1.
PROOF. =: Clearly p P1 and P1 = . As p is maximal, this implies that p =
P1.
=: If p P, then pT P, 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
i=1
e
i
i
=m. (9)
If 1 is Galois over 1, then all the ramication numbers are equal, and all the residue class
degrees are equal, and so
eg =m. (10)
PROOF. To prove (9), we shall show that both sides equal T,pT: ,p.
For the equality
i=1
e
i
i
=T,pT: ,p, note that T,pT =T,
P
e
i
i
.
T,P
e
i
i
(Chinese Remainder Theorem), and so it sufces to show that T,P
e
i
i
: ,p =e
i
i
. From
the denition of
i
, we know that T,P
i
is a eld 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
The primes that ramify 59
has dimension
i
over ,p, and so the dimension of T,P
e
i
i
is e
i
i
.
The proof of the equality T,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), p23), which shows
that n =T,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
=T
t
,pT
t
:
t
,p
t
; but
t
is principal, and so T
t
,pT
t
:
t
,p
t
 =m. This
completes the proof (9).
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 nd an
element lies in Q but not in any of the ideals oP. Consider b =Nm()
def
=
o. Then
b , and as Q, it also lies in Q; hence b Q = p. On the other hand, for all
o Gal(1,1), o
1
P, and so o P. The fact that
o p 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 nite extension of a number eld 1, let be a Dedekind
domain in 1 with eld 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 ramies in 1 if and only if p[ disc(T,). In particular, only
nitely 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 nite 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
denitions.
2
60 3. Dedekind Domains; Factorization
LEMMA 3.37 Let be a ring and let T
1
. .... T
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
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 eld, and let T be a kalgebra of nite dimension. Then
T is reduced if and only if disc(T,k) =0.
PROOF. Let =0 be a nilpotent element of T, and choose a basis e
1
. . . . . e
n
for T with
e
1
=. Then e
i
is nilpotent for all i , and so the klinear map
. e
i
.: T T
is nilpotent. Its matrix is also nilpotent, but a nilpotent matrix has trace zeroits minimum
polynomial (and hence its characteristic polynomial) is of the form X
i
and so the rst
row of the matrix (Tr(e
i
e
}
)) is zero. Therefore its determinant is zero.
Conversely, suppose T is reduced. We rst 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
nn
=
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 eld (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
T : k _T,
_
p
i
: k =
T,p
i
: k _r.
Therefore T has only nitely many prime ideals, say p
1
. . . . . p
i=1
T,p
i
.
For each i , T,p
i
is a eld, and it is a nite 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
Finding factorizations 61
We now prove the theorem. From the rst 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 ramies 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 eld 1, and let T be the integral closure
of in a nite extension 1 of 1. It is possible to dene 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 dene disc(S
1
T,S
1
). It is a power (p
p
)
n(p)
of p
p
. Dene
disc(T,) =
p
n(p)
.
The index m(p) is nonzero for only nitely many p, and so this formula does dene an ideal
in . Clearly this denition agrees with the usual one when T is a free module, and the
above proof shows that a prime ideal p ramies 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 elds of meromorphic functions on X and Y . The map is an
inclusion M(X) M(Y ) which makes M(Y ) into a eld of nite degree over M(X);
let m be this degree. Geometrically, the map is m: 1 except at a nite 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) = {Q
1
. .... Q
] and let e
i
be the number
of sheets of Y over X that coincide at Q
i
. Then pT =
q
e
i
i
where q
i
is the prime ideal
{ T [ (Q
i
) =0].
Finding factorizations
The following result often makes it very easy to factor an ideal in an extension eld. Again
is a Dedekind domain with eld of fractions 1, and T is the integral closure of in a
nite separable extension 1 of 1.
THEOREM 3.41 Suppose that T =, 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 X
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
62 3. Dedekind Domains; Factorization
is the factorization of pT into a product of powers of distinct prime ideals. Moreover, the
residue eld T,(p. g
i
()) .(,p)X,( 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 denes an isomorphism
X,((X)) T.
When we divide out by p (better, tensor with ,p), this becomes an isomorphism
kX,(
(X)) T,pT. X .
where k = ,p. The ring kX,(
) 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 kX,(
) 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)X), 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 =
_
m. Recall that disc(1.
_
m) = 4m, and that disc(O
1
,Z) =
disc(1.
_
m) if m2. 3 mod 4, and that disc(O
1
,Z) =disc(1.
_
m),4 if m1 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
: () ramies, 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 () ramies 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:
() is a prime ideal in
_
m m is not a square mod , i.e., (
n
]
) =1.
Examples of factorizations 63
(iii) For the prime 2 when m1 mod 4, we have
() is the product of two distinct ideals m1 mod 8;
() is a prime ideal in
_
m m5 mod 8.
To prove (iii), we must use the integral basis {1. ], = (1
_
m),2. The minimum
polynomial of is X
2
X(1m),4. If m1 mod 8, this factors as X
2
X =X(X1)
mod 2, and so (2) =(2. )(2. 1). If m5 mod 8, then X
2
X (1m),4 X
2
]
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 Zi , 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
Zi . But the only units in Zi  are 1, i , and so obviously a
2
b
2
=. Conversely, if
=a
2
b
2
with a. b Z, then () =(ai b)(ai b) in Zi .
ASIDE 3.46 The fact that every prime of the form 4n1 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 Fermats letter,
found a proof. For some history, and a discussion of algorithms for nding 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 (X2215)
2
(X3624), etc., as in the following
table.
(b) In the next section, we shall show, not only that the class group of a number eld
is nite, but that it is generated by the prime ideals dividing a certain small set of prime
numbers. Finding the class number therefore involves nding 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  is ZZ Z
2
. There are the
following factorizations:
64 3. Dedekind Domains; Factorization
2 (1X)(1X X
2
) (2) = (2. 1)(2. 1
2
)
3 (2X)(2X X
2
) (3) = (3. 2)(3. 2
2
)
5 (1X)(14X X
2
) (5) = (5. 1)(5. 14
2
)
7 (3X)(54X X
2
) (7) = (7. 3)(7. 54
2
)
11 (6X)(25X X
2
) (11) = (11. 6)(11. 25
2
)
13 110X X
3
(13) = (13. 110
3
) =(13)
17 110X X
3
(17) = prime ideal.
4027 (2215X)
2
(3624X) (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  is ZZ Z
2
. There are the
following factorizations:
2 (1X)(1X X
2
) (2) = (2. 1)(2. 1
2
)
3 X(1X
2
) (3) = (3. )(3. 1
2
)
5 X(2X
2
) (5) = (5. )(5. 2
2
)
7 (5X)(32X X
2
) (7) = (7. )(7. 32
2
)
11 (1X)(410X X
2
) (11) = (11. )(11. 410
2
)
13 25X X
3
(13) = (13)
17 (4X)(6X)(7X) (17) = (17. 4)(17. 6)(17. 7)
4027 (509X)(1759X)
2
. (4027) = (4027. 509)(4027. 1759)
2
On comparing the factorizations of (17) in the elds in the last two examples, we see that
the elds are not isomorphic.
REMARK 3.50 When 1 is a number eld, it is interesting to have a description of the set
Spl(1) of prime numbers that split in 1. For 1 =
_
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 , class eld 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 are isomorphic if and only if
Spl(1) =Spl(1
t
). Moreover, this can be made into an effective procedure for determining
when elds are isomorphic. See Theorem 8.38 below.
EXAMPLE 3.51 In (2.39), we saw that (X) =X
5
X1 is irreducible in X, and that
its discriminant is 19 151, which is squarefree, and so, if is a root of (X), then Z is
the ring of integers in . We have the following factorizations:
19
_
(6X)
2
(1013X 17X
2
X
3
)
(19) =(19. 6)
2
(19. 1013 17
2
3
)
151
_
(9X)(39X)
2
(6164X X
2
)
(151) =(151. 9)(151. 39)
2
(151. 6164
2
)
4027
_
(1261X)(2592X)(7903499X 174X
2
X
3
)
(4027) =(4027. 1261)(4027. 2592)(4027. 7903499 174
2
3
.
Thus (19) and (151) are ramied in , and 4027 is not, which is what Theorem 3.35
predicts.
Eisenstein extensions 65
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 Eisensteins 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 X. We will
improve this result, but rst we need to make two observations about discrete valuations.
Let be a Dedekind domain, and let T be its integral closure in a nite extension 1 of
its eld 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 dened
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) X be an Eisenstein polynomial with respect to p. Then
(X) is irreducible, and if is a root of (X), then p is totally ramied in 1; in fact
pT =P
n
with P=(p. ) and m=deg( ).
PROOF. Let 1 be the eld generated by a root of (X); then 1: 1 _m
def
=deg( ). Let
P be a prime ideal dividing p, with ramication 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
) _(mi ) ord
P
() e:
ord
P
(a
n
) =e.
66 3. Dedekind Domains; Factorization
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. .... m1.
From the remark preceding the proposition, we see that m ord
P
() =e. Then
m ord
P
() =e _1 : 1 _m.
and we must have equalities throughout: ord
P
() =1, 1(): 1 =m=e.
2
NOTES Gauss proved the quadratic reciprocity law, and studied the arithmetic of i  in order
to discover the quartic reciprocity law. Kummer made an intense study of the arithmetic of the
elds
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 Dedekinds great achievements
was to realize that, by replacing Kummers ideal numbers with his new notion of ideals, it was
possible to simplify Kummers theory and extend it to the rings of integers in all number elds. A
difcult step for him was showing that if a[b, then there exists an ideal c such that a =bc. Emmy
Noether reexamined Kummers work more abstractly, and named the integral domains for which
his methods applied Dedekind domains.
Exercises
31 Let k be a eld. Is kX. Y  a Dedekind domain? (Explain).
32 Show that Z
_
3 is the ring of integers in
_
3 and Z
_
7 is the ring of integers
in
_
7, but that Z
_
3.
_
7 is not the ring of integers in
_
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
_
 has class number 1 for <5.
34 Let k be a eld, and let be the subring kX
2
. X
3
 of kX.
(a) Show that kX is a nitely generated kX
2
module, and hence is a Noetherian
kX
2
module. Deduce that is Noetherian.
(b) Show that every nonzero prime ideal of is maximal, but that is not a Dedekind
domain.
Hence satises 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, kX. 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 Z
_
3.
CHAPTER 4
The Finiteness of the Class Number
In this section we prove the rst main theorem of the course: the class number of a number
eld is nite. The method of proof is effective: it gives an algorithm for computing the
class group.
Norms of ideals
Let be a Dedekind domain with eld of fractions 1, and let T be the integral closure of
in a nite separable extension 1. We want to dene a homomorphism Nm: Id(T) Id()
which is compatible with taking norms of elements, i.e., such that the following diagram
commutes:
1
Id(T)
1
Id().
b (b)
Nm
Nm
o (o)
(12)
Because Id(T) is the free abelian group on the set of prime ideals, we only have to
dene Nm(p) for p prime.
Let p be a prime ideal of , 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=1: 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 dene Nm(P
i
) =p
(
i
. We take this as our denition:
Nm(P) =p
((P{p)
where p =P and (P,p) =T,P: ,p.
To avoid confusion, I sometimes use N to denote norms of ideals.
If we have a tower of elds M 1 1, then
N
1{1
(N
{1
a) =N
{1
a
because (Q,P) (P,p) = (Q,p), i.e., C,Q : T,P T,P : ,p = C,Q : ,p
where C T are the integral closures of in M, 1, and 1 respectively.
67
68 4. The Finiteness of the Class Number
PROPOSITION 4.1 Let T and 1 1 be as above.
(a) For any nonzero ideal a , N
1{1
(aT) =a
n
, where m=1 : 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
)
e
(cf. 3.34). Then
NP T =(P
1
P
)
e(
=
cGal(1{1)
oP.
(c) For any nonzero element T, Nm() =Nm( T) (i.e., (12) commutes).
PROOF. (a) It sufces 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 rst 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
], and it follows
that each P
i
occurs
n
(o T) =(
o) T =Nm() T
as required.
In the general case, let 1 be a nite Galois extension of 1 containing 1, and let
J = 1: 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
( T)
d
=N
T{1
( C) =Nm
T{1
() =Nm
1{1
()
d
.
As the group of ideals Id() is torsionfree, this implies that N
1{1
( T) =Nm
1{1
()
.
2
Let a be a nonzero ideal in the ring of integers O
1
of a number eld 1. Then a is of
nite 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 eld 1.
(a) For any ideal a in O
1
, N
1{
(a) =(N(a)); therefore N(ab) =N(a)N(b).
(b) Let b a 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
) =Zp
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
has a ltration of
length r
i
whose quotients are vector spaces of dimension
i
over F
]
i
, and so (O
1
: p
i
i
i
) =
(
i
i
i
i
. On taking the product over i , we nd that (O
1
: a) =
(
(
i
i
i
i
) =N
1{
a. When
Statement of the main theorem and its consequences 69
we identify the set of nonzero ideals in Z with the set of positive integers, then N becomes
identied 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 , and let z
1
be the discriminant of
1,. 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) _
n
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
=
n
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 0.9310
42
Here r is the number of real embeddings of 1. We have
1
1 ~1
i
C
x
.
and, if 1 = 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.
THEOREM 4.4 The class number of 1 is nite.
70 4. The Finiteness of the Class Number
PROOF. It sufces to show that there are only nitely 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 nitely 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 nitely many possibilities for
the
i
(and hence for the p
i
), and only nitely 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 nite 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 nding S, and an algorithm for deciding whether
a ~b, and so there is an algorithm for nding Cl(O
1
) (the group, not just its order). To nd
S, nd 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 nding Cl(O
1
), see Cohen
1993, 6.5, and Pohst and Zassenhaus 1989, p424.
EXAMPLE 4.5 Let 1 =i . The condition in Theorem 4.3 is that N(a) _
2
4
4
t
2 < 1.27.
There are no such ideals other than Zi , and so Zi  is a principal ideal domain. (Of course,
the elementary proof of this shows more, namely, that Zi  is a Euclidean domain. Even
for rings of integers in number elds, it is not true that all principal ideal domains are
Euclidean domains. For example,
_
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 =
_
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(Z
_
5). The ideal
p cant be principal because there does not exist an element = mn
_
5 such that
Nm() =m
2
5n
2
=2, and so Cl(Z
_
5) has order 2.
EXAMPLE 4.7 Let 1 be a cubic eld with discriminant <0. Since the sign of z
1
is (1)
x
,
and 1 :  =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 elds with 49 _z
1
<0, the class number h =1.
For example, this is true for the number elds with discriminants 23 and 31 discussed
earlier (see 2.36, 2.37).
1
Lemmermeyer, Franz. The Euclidean algorithm in algebraic number elds. Exposition. Math. 13 (1995),
no. 5, 385416.
Statement of the main theorem and its consequences 71
For the stem eld 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 19151, and so the ring of integers of  is Z.
According to Theorem 4.3, every class of ideals for  contains an integral ideal a
with
N(a) <0.062
_
19151 =3.3 <4.
If p is a prime ideal with N(p) =2, then the residue eld at p must be F
2
, and (X) must
have a root mod 2; however, both (0) and (1) are odd, and so (X) doesnt have a root
in F
2
, which shows that p doesnt 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 eld M of (X) is S
5
(later we shall see how to
nd Galois groups; for the moment type polgalois(X^5X1) in PARI), and hence
M:  =120. It is possible to show that M is unramied over
_
19151.
An extension 1 of a number eld 1 is said to be unramied over 1 if no prime ideal
of O
1
ramies in O
1
.
THEOREM 4.9 There does not exist an unramied extension of .
PROOF. Let 1 be a nite extension of . 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
n
_
4
_
x
_
n
n
n
_
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 ramies.
2
COROLLARY 4.10 There does not exist an irreducible monic polynomial (X) ZX of
degree >1 with discriminant 1.
PROOF. Let (X) be such a polynomial, and let be a root of (X). Then disc(Z,Z) =
1, and so Z is the ring of integers in 1
def
= and disc(O
1
,Z) =1, which contra
dicts the theorem.
2
REMARK 4.11 There may exist unramied extensions of number elds other than . In
fact, class eld theory says that the maximal abelian unramied
2
extension of 1 (called
2
The Hilbert class eld 1 of 1 is required to be unramied even at the innite primes this means that
every real embedding of 1 extends to a real embedding of 1.
72 4. The Finiteness of the Class Number
the Hilbert class eld of 1) has Galois group canonically isomorphic to Cl(O
1
). For
example, the theory says that
_
5 has an unramied extension of degree 2, and one
veries that
_
1.
_
5 is unramied over
_
5. In particular, the discriminant of
_
1.
_
5 over
_
5 is a unit.
3
REMARK 4.12 Let 1
1
be a number eld with class number h
1
1
>1. Its Hilbert class eld
is an abelian unramied extension 1
2
of 1
1
with Gal(1
2
,1
1
) .Cl(1
1
). Let 1
3
be the
Hilbert class eld of 1
2
, and so on. In this way, we obtain a tower of elds,
1
1
1
2
1
3
It was a famous question (class eld tower problem) to decide whether this tower can be
innite, or must always terminate with a eld of class number 1 after a nite number of
steps. It was shown by Golod and Shafarevich in the early 60s that the tower is frequently
innite. See Roquette 1967.
If 1 has class number 1, then it has no abelian unramied extensions, but it may have
nonabelian unramied extensions, even innite (see, for example, D. Brink, Remark on
innite unramied extensions of number elds with class number one, J. Number Theory
130 (2010), 304306; mo53530).
Lattices
Let V be a vector space of dimension n over 1. A lattice in V is a subgroup of the form
=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, = {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 of V such that the map
r
i
.
i
r
i
.
i
: 1
Z
V.
is an isomorphism.
NONEXAMPLE 4.13 The subgroup ZZ
_
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 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 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 is discrete means that every point of has a neighbourhood U in V such that
U ={].
3
The ring of integers in
_
1.
_
5 is Z
_
_
1. (1
_
5),2
_
, which properly contains Z
_
1.
_
5.
Lattices 73
LEMMA 4.14 The following conditions on a subgroup of a nitedimensional real vector
space V are equivalent:
(a) is a discrete subgroup;
(b) there is an open subset U of V such that U ={0];
(c) each compact subset of V intersects in a nite set;
(d) each bounded subset of V intersects in a nite set.
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 ={0], then U is a neighbourhood of such that ( U) ={].
(a)=(c). Condition (a) says that is a discrete space for the induced topology. Hence,
if C is compact, then C is both discrete and compact,
4
and therefore must be nite.
(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 {0] is nite
and hence closed, and so U S is an open neighbourhood of {0] such that (U S) =
{0].
2
PROPOSITION 4.15 A subgroup of V is a lattice if and only if it is discrete.
PROOF. Clearly, a lattice is discrete. For the converse, let be a discrete subgroup of V ,
and choose a maximal 1linearly independent subset {e
1
. . . . . e
i
] of . We shall argue by
induction on r.
If r =0, =0, and there is nothing to prove.
If r =1, then 1e
1
. Because is discrete, for each M >0,
{ae
1
[ [a[ <M]
is nite, and so there is an such that, when we write =ae
1
, a attains its minimum
value > 0. I claim =Z . Any Z will equal (mb) 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
t
= (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,
t
=Z
1
Z
i1
for
some
i
that are linearly independent over 1 (and hence also form a basis for V
t
). Every
can be written uniquely
=a
1
1
a
i1
i1
ae
i
. a
i
. a 1.
Let : 1 be the map a, and let
tt
=Im(). Note that a is also the image of
(a
1
a
1
)
1
(a
i1
a
i1
)
i1
ae
i
. + =integer part,
4
I am implicitly using that a discrete subgroup of a Hausdorff group is closed (note that a discrete subset
need not be closed, e.g., {1,n [ n an integer > 0] is not closed in the real numbers). Here is the proof. Let H
be a discrete subgroup of a Hausdorff group G. There exists a neighbourhood U of 1 such that U H = 1;
choose a neighbourhood V of 1 such that V
1
V is contained in U. For distinct elements a and b of H, Va
and Vb are disjoint. Let g lie in the closure of H, so that H V
1
g is nonempty. If a lies in H V
1
g,
say a =
1
g, then g Va. This shows that H V
1
g ={a]. As g is in the closure of H, this implies that
g =a, and so g lies in H. More generally, every locally compact subgroup of a Hausdorff group is closed.
74 4. The Finiteness of the Class Number
and so each element a
tt
in a bounded set, say with 0 _ [a[ < M, is the image of an
element of in a bounded set,
0 _a
i
<1. i =1. . . . . r 1. [a[ <M.
Thus there are only nitely many such as, and so
tt
is a lattice in 1, say
tt
=Z (
i
),
i
.
Let . Then () = a(
i
) for some a Z, and ( a
i
) = 0. Therefore
a
i
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 =
Z
i
.
2
Let V be a real vector space of dimension n, and let be a full lattice in V , say
=
Ze
i
. For any z
0
, let
D ={z
0
a
i
e
i
[ 0 _a
i
<1].
Such a set is called a fundamental parallelopiped for . The shape of the parallelopiped
depends on the choice of the basis (e
i
), but if we x the basis and vary z
0
, then the
parallelopipeds cover 1
n
without overlaps.
REMARK 4.16 (a) For a fundamental parallelopiped D of a full lattice
=Z
1
Z
n
in 1
n
, the volume of D
j(D) =[ det(
1
. .
n
)[.
(See any good book on calculus.) If also
=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 doesnt depend on the choice of the basis for .
(b) When
t
are two full lattices in 1
n
, we can choose bases {e
i
] and {
i
] for
and
t
such that
i
=m
i
e
i
with m
i
a positive integer. With this choice of bases, the fun
damental parallelopiped D of is a disjoint union of (:
t
) fundamental parallelopipeds
D
t
of
t
. Hence
j(D
t
)
j(D)
=(:
t
) (*).
As we noted above, the choice of a basis for V determines an isomorphism V ~ 1
n
,
and hence a measure j on V . This measure is translation invariant (because the Lebesgue
measure on 1
n
is translation invariant), and welldened up to multiplication by a nonzero
constant (depending on the choice of the basis)
5
. Thus the ratio of the measures of two sets
is welldened, and the equation (*) holds for two full lattices
t
in V .
5
The experts will recognize j as being a Haar measure on V .
Lattices 75
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 j(S) > j(D
0
), then S contains distinct points and
such that .
PROOF. The set S D is measurable for all fundamental parallelopipeds D, and
j(S) =
j(S D)
(sum over translates of D by elements of ). For each D, a (unique) translate of S D by
an element of will be a subset of D
0
. Since j(S) >j(D
0
), at least two of these sets will
overlap, i.e., there exist elements . S such that
z =z
t
. some z. z
t
.
Then .
2
REMARK 4.18 In the language of differential geometry, the theorem can be given a more
geometric statement. Let M = V,; it is an ndimensional torus. The measure j on V
denes a measure on M for which M has measure j(M) =j(D). The theorem says that
if j(S) >j(M), then the restriction of the quotient map V M to S cant be injective.
Let T be a set such that
, T =
1
2
( ) 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 other than the origin whenever
j(D) <j(
1
2
T ) =2
n
j(T ).
i.e., whenever
j(T ) >2
n
j(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 satises (**), and so it will contain a point of \{0] if
its volume is greater than 2
n
j(D).
THEOREM 4.19 (MINKOWSKIS) Let T be a subset of V that is compact, convex, and
symmetric in the origin. If
j(T ) _2
n
j(D)
then T contains a point of the lattice other than the origin.
PROOF. Replace T with (1c)T , c >0. Then
j((1c)T ) =(1c)
n
j(T ) >2
n
j(D).
and so (1 c)T contains a point of other than the origin (see the preceding remark). It
will contain only nitely many such points because is discrete and (1c)T is compact.
Because T is closed
T =
_
t>0
(1c)T.
If none of the (nitely many) points of (1c)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 other than the
originwhich is a contradiction.
2
76 4. The Finiteness of the Class Number
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
) =
(abT cC JD)
2
(aT bcDJC)
2
(aC bDcJT)
2
(aDbC cT J)
2
.
we see that it sufces 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 Z
4
consisting of
(a. b. c. J) such that
c manb. J mb na mod .
Then Z
4
Z
4
and ,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 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
j(T ) >16j(D).
According to Minkowskis theorem, there is a point (a. b. c. J) (\ {0]) T . Because
(a. b. c. J) ,
a
2
b
2
c
2
J
2
a
2
(1m
2
n
2
) b
2
(1m
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 nitedimensional 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;
Some calculus 77
(b) for r 1 and x V , [rx[ =[r[[x[;
(c) (triangle law) for x. y V , [xy[ _[x[[y[.
Let V =1
i
C
x
it is a real vector space of dimension n =r 2s. Dene a norm
on V by
[x[ =
i
i=1
[.
i
[ 2
ix
i=i1
[:
i
[
if x =(.
1
. .... .
i
. :
i1
. .... :
ix
).
LEMMA 4.22 For any real number t >0, let
X(t ) ={x V [ [x[ _t ].
Then
j(X(t )) =2
i
(,2)
x
t
n
,n.
PROOF. Since X(t ) is symmetric with respect to the r real axes, we have
j(X(t )) =2
i
j(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 nd that
j(X(t )) =2
i
4
x
(2)
x
_
7
j
i1
j
ix
J.
1
J.
i
Jj
i1
Jj
ix
where
7 ={(x. j) 1
ix
[ .
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
j(X(t )) =2
i
4
x
(2)
x
t
n
,n
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
m1)
.
78 4. The Finiteness of the Class Number
PROOF. Recall that, by denition, (e.g., Widder, D., Advanced Calculus, 1961, Chapter
11),
1(.) =
_
o
0
e
t
t
x1
Jt.
It takes the value 1(n) =(n1) 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).
Therefore it sufces 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
m1)
.
In the last step, we used the standard formula
_
1
0
.
n1
(1.)
n1
J. =T(m. n) =
1(m)1(n)
1(mn)
.
2
EXAMPLE 4.24 (a) Case r =2, s =0. Then X(t ) is dened by [.[ [,[ _t . It is a square
of side
_
2t , and so j(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 79
Finiteness of the class number
Let 1 be a number eld of degree n over . Suppose that 1 has r real embeddings
{o
1
. . . . . o
i
] and 2s complex embedding {o
i1
. o
i1
. . . . . o
ix
. o
ix
]. Thus n = r 2s.
We have an embedding
o: 1 1
i
C
x
. (o
1
. . . . . o
ix
).
We identify V
def
=1
i
C
x
with 1
n
using the basis {1. i ] for C.
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
i1
i
). `(o
i1
i
). . . .)
has nonzero determinant.
First consider the matrix T whose i th row is
(o
1
(
i
). . . . . o
i
(
i
). o
i1
(
i
). o
i1
(
i
). . . . . o
ix
(
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
i1
(
i
)) in column r 1 and i `(o
i1
(
i
)) in column r 2. Repeat for the other
pairs of columns. These column operations dont 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
j(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
n
n
n
Na[z
1
[
1
2
.
80 4. The Finiteness of the Class Number
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 j(X(t )) _2
n
j(D), Minkowskis Theorem shows that X(t ) contains a point
o() =0 of o(a). For this a,
[ Nm()[ =[o
1
()[ [o
i
()[[o
i1
()[
2
[o
ix
()[
2
_(
[o
i
[
2[o
i
[)
n
,n
n
(by 4.25)
_t
n
,n
n
.
In order to have j(X(t )) _2
n
j(D), we need (see 4.22, 4.26)
2
i
(,2)
x
t
n
,n _2
n
2
x
Na [z
1
[
1
2
.
i.e.,
t
n
_n
2
ni
x
Na [z
1
[
1
2
.
When we take t
n
to equal the expression on the right, we nd that
[ Nm()[ _
n
n
n
2
ni
x
Na [z
1
[
1
2
.
As nr =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
=
n
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 b, =0, with
[ Nm()[ _T
1
Nb.
Now O
1
b =O
1
=ab with a integral, and a ~b
1
~c. Moreover,
Na Nb =[ Nm
1{
[ _T
1
Nb.
On cancelling Nb, we nd 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 dened a product for them. This work
was greatly claried when Kummer and Dedekind dened ideals, and it was realized that
Gausss results were related to the ideal class groups of quadratic number elds. Here I
briey explain the connection.
Binary quadratic forms 81
By a binary quadratic form we mean an expression of the form
Q(X. Y ) =aX
2
bXY cY
2
.
We call the form integral if Q(m. n) is an integer whenever m and n are integers, or, equiv
alently, if a. b. c Z. The discriminant of Q is
J
Q
=b
2
4ac.
A form is said to be nondegenerate if its discriminant is nonzero. Two integral binary
quadratic forms Q and Q
t
are said to be equivalent if there exists a matrix =
_
;
_
SL
2
(Z) such that
Q
t
(X. Y ) =Q(X Y. ;X Y ).
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 xed discriminant.
Let J =1 be a squarefree integer, let 1 =
_
J, and let J
1
=disc(O
1
,Z). Dene
the norm form q
1
by
q
1
(X. Y ) =Nm
1{
(X Y
_
J) =X
2
JY
2
. if J 2. 3 mod 4
or
q
1
(X. Y ) =Nm
1{
(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 Q is an integral binary quadratic form, then J
Q
= J
1
2
, some integer
, where 1 =
_
J
Q
. Moreover, if J
Q
=J
1
, then Q is primitive, i.e., gcd(a. b. c) =1.
Fix a eld 1 =
_
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,) = {1. o]. If J < 0, dene 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
a
1
a
2
oa
1
oa
2
2
=J
1
Na
2
.
After possibly reordering the pair a
1
. a
2
we will have
a
1
a
2
oa
1
oa
2
=
_
J
1
Na.
For such a pair, dene
Q
o
1
,o
2
(X. Y ) =Na
1
Nm
1{
(a
1
X a
2
Y ).
This is an integral binary quadratic form with discriminant J
1
.
82 4. The Finiteness of the Class Number
THEOREM 4.29 The equivalence class of Q
o
1
,o
2
(X. Y ) depends only on the image of a in
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
In particular, the set of equivalence classes is nite, and has the structure of an abelian
group. This was known to Gauss, even though groups had not yet been dened. (Gauss
even knew it was a direct sum of cyclic groups.)
ASIDE 4.30 Write h
d
for the class number of
_
J, J a squarefree integer = 1. In modern
terminology, Gauss conjectured that, for a xed h, there are only nitely 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 Gausss conjecture. Unfortunately, the c in Siegels theorem is not effectively com
putable, and so Siegels theorem gives no way of computing the Js for a given h.
In 1951, Tatuzawa showed that Siegels 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 elds with class number 1, and
possibly 1 more.
In 1952 Heegner proved that there was no 10th such eld, 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 nding 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), 2337.
By way of contrast, it is conjectured that there are innitely many real quadratic elds 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 rst indication that
the arithmetic of the ring of integers in
_
5 is more complicated than it is in the elds with
smaller discriminant. Lagrange found an explanation for Fermats 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
_
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 cant happen if T is integral over see the
paragraph preceding 3.30.)
Exercises 83
42 Let J 1 1 be a sequence of number elds, and let T C 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 = 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.)
44 Show that
_
23 has class number 3, and that
_
47 has class number 5.
45 Let 1 be an algebraic number eld. Prove that there is a nite extension 1 of 1
such that every ideal in O
1
becomes principal in O
1
. [Hint: Use the niteness of the class
number.]
46 Let 1 =  where is a root of X
3
X 2. Show that O
1
= Z 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
(ab c
2
) is an algebraic integer if and only if a, b, and
c are all even, but you may be able to nd a better one.]
47 Let 1 =
_
1.
_
5. Show that O
1
=Z
_
1.
1
_
5
2
. Show that the only primes
(in Z) that ramify in 1 are 2 and 5, and that their ramication indexes are both 2. Deduce
that 1 is unramied over
_
5. Prove that
_
5 has class number 2, and deduce
that 1 is the Hilbert class eld of
_
5. (Cf. 4.11.)
CHAPTER 5
The Unit Theorem
In this section we prove the second main theorem of the course.
Statement of the theorem
Recall that a nitely generated abelian group is isomorphic to
tors
Z
t
for some t where
tors
is the (nite) subgroup of torsion elements of (i.e., of elements of nite 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 eld 1 and 2s
for the number of nonreal complex embeddings. Thus
1
1 ~1
i
C
x
and r 2s =1: . Moreover, if 1 =, 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 eld 1 is nitely generated with rank equal
to r s 1.
For example, for a real quadratic eld, the rank is 201 =1, and for an imaginary
quadratic eld it is 011 =0.
The theorem is usually referred to as the Dirichlet Unit Theorem although Dirichlet
proved it for rings of the form Z 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
j(1) of roots of 1 in 1.
A set of units u
1
. . . . . u
ix1
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
ix1
. j(1). m
i
Z.
The theorem implies that j(1) is nite (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  is a Galois extension of 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
84
Statement of the theorem 85
Z,
i
i
i
Z are exactly the elements divisible by , and there are
i
i
1
i
of these, we see that
(Z,
i
i
i
Z)
=
i
i
1
i
(
i
1), and so
(Z,mZ)
i
i
1
i
(
i
1)
def
=(m).
Since
n
1 =
n
 1 =(m)[1 : .
the eld 1 can contain only nitely many
n
.
LEMMA 5.2 An element 1 is a unit if and only if O
1
and Nm
1{
=1.
PROOF. If is a unit, then there is a O
1
such that =1, and then Nm() and Nm()
lie in Z and 1 =Nm() =Nm() Nm(). Hence Nm Z
={1].
For the converse, x an embedding o
0
of 1 into C, and use it to identify 1 with a
subeld of C. Recall (2.20) that
Nm() =
c:1C
o =
cyc
0
o.
Let =
cyc
0
o. If O
1
, then is an algebraic integer. If Nm() = 1, then
=
1
and so belongs to 1. Therefore, if satises both conditions, it has an inverse
in O
1
, and so is a unit.
2
For all real elds, i.e., elds with an embedding into 1, j(1) = {1]; for most
nonreal elds, this is also true.
EXAMPLE 5.3 Let 1 be a quadratic eld
_
J. Then O
1
={mn
_
J [ m. n Z] or
{mn(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
(2mn)
2
Jn
2
=4.
When J < 0, these equations (obviously) have only nitely many solutions, and so U
1
=
j(1). Note that
n
lies in a quadratic eld if and only if (m) _2. This happens only for
m dividing 4 or 6. Thus j(1) ={1] except for the following elds:
i , j(1) ={1. i ];
_
3, j(1) ={1. j. j
2
], with j =(1
_
3),2).
When J >0, the theoremshows that there are innitely 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.
EXAMPLE 5.4 Let 1 = , 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 innite order, and later we shall show that it is a fundamental unit, and so U
1
=
{
n
[ m Z].
1
Cohn, Harvey. A classical invitation to algebraic numbers and class elds. With two appendices by Olga
Taussky: Artins 1932 G ottingen lectures on class eld theory and Connections between algebraic number
theory and integral matrices. Universitext. SpringerVerlag, New YorkHeidelberg, 1978. xiii+328 pp.
86 5. The Unit Theorem
Proof that U
1
is nitely generated
We rst 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 nite.
PROOF. The rst condition says that is a root of a monic irreducible polynomial of degree
_ m, and the second condition implies that the coefcients of the polynomial are bounded
in terms of M. Since the coefcients are integers, there are only nitely many such poly
nomials, and hence only nitely 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 nite.
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 nite.
Recall that we previously considered the map
o: 1 1
i
C
x
. (o
1
. . . . . o
i
. o
i1
. . . . . o
ix
)
where {o
1
. . . . . o
i
. o
i1
. o
i1
. . . . . o
ix
. o
ix
] 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
ix
. (log[o
1
[. . . . . log[o
i
[. 2log[o
i1
[. . . . . 2log[o
ix
[).
It is a homomorphism. If u is a unit in O
1
, then Nm
1{
u =1, and so
[o
1
u[ [o
i
u[[o
i1
u[
2
[o
ix
u[
2
=1.
On taking logs, we see that 1(u) is contained in the hyperplane
H : .
1
.
i
2.
i1
2.
ix
=0.
Dropping the last coordinate denes an isomorphism H ~1
ix1
.
PROPOSITION 5.8 The image of 1: U H is a lattice in H, and the kernel of 1 is a nite
group (hence is j(1)).
PROOF. Let C be a bounded subset of H containing 0, say
C {x H [ [.
i
[ _M].
If 1(u) C, then [o
}
u[ _ e
for all , and Proposition 5.5 implies that there are only
nitely many such us. Thus 1(U) C is nite, 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 is nite by
Proposition 5.5.
2
Since the kernel of 1 is nite, we have
rank(U) =rank(1(U)) _dimH =r s 1.
Computation of the rank 87
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 nd a way to construct units. We work again
with the embedding
o: 1 1
i
C
x
~1
i2x
.
For x =(.
1
. .... .
i
. .
i1
. ...) 1
i
C
x
, dene
Nm(x) =.
1
.
i
.
i1
.
i1
.
ix
.
ix
.
Then Nm(o()) =Nm(). Note that [ Nm(x)[ =[.
1
[ [.
i
[[.
i1
[
2
[.
ix
[
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
i1
(
i
)). `(o
i1
(
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
i1
(
i
). o
i1
(
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.
Dene
x o(O
1
) ={x o() [ O
1
].
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(.
i1
o
i1
(
i
)). `(.
i1
o
i1
(
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
). . . . . .
i1
o
i1
(
i
). .
i1
o
i1
(
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, Minkowskis 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.
88 5. The Unit Theorem
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 nitely
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
doesnt 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 sufciently 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
ix1
) 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
ix1
(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
ix1
(c
i
) l
ix
(c
i
) =0.
and so the sum of the terms in the i th row
l
1
(c
i
) l
ix1
(c
i
) =l
ix
(c
i
) >0.
The next lemma implies that the matrix is invertible, and so completes the proof of Theorem
5.9.
2
LEMMA 5.10 Let (a
i}
) be a real mm matrix such that
a
i}
<0 for i = :
}
a
i}
>0 for i =1. 2. .... m.
Then (a
i}
) is invertible.
PROOF. If it isnt, 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
}yi
0
a
i
0
}
.
}
_a
i
0
i
0
}yi
0
a
i
0
}
>0.
2
Sunits 89
Sunits
Let S be a nite set of prime ideals of 1, and dene the ring of Sintegers to be
O
1
(S) =
_
pS
O
p
={ 1 [ ord
p
() _0, all p S].
For example, if S =0, then O
1
(S) =O
1
.
Dene 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 j(1).
THEOREM 5.11 The group of Sunits is nitely 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 sufces 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 = 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.
Example: CM elds
Anumber eld 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 = .X,( ) 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 eld is a totally imaginary quadratic extension of a totally real eld. Every such
eld can be obtained from a totally real eld by adjoining the square root of an element all
of whose real conjugates are negative.
Let 1 be a CM eld, which is a quadratic extension of the totally real eld 1
, and let
2n =1 : . Then 1 has 2n complex embeddings and 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 j(1) (by 5.6). Consider the map : U
1
j(1),j(1)
2
determined by a a, a. Clearly is a homomorphism. Suppose u lies in its
kernel, so that u, u =
2
for some j(1). Then u
, u =1, and so u
. It follows
that u j(1) U
1
C. Conversely, if u = u
j(1) U
1
C, then u, u =
2
Ker().
We have shown that Ker() =j(1) U
1
C. As j(1),j(1)
2
has order 2, this completes
the proof.
2
90 5. The Unit Theorem
Example: real quadratic elds
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
a
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 a
0
. a
1
. .... a
n
 is called the nth convergent. Every irrational num
ber can be expressed in just one way as an innite continued fraction, and the continued
fraction is periodic if and only if has degree 2 over . (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
_
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.
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 nd the fundamental unit in
_
94, rst 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 =2143295221064
_
94.
Compute that
(2143295)
2
(221064)
2
94 =1.
which veries that c is a unit.
When one carries out this procedure for
_
9199, the rst coefcient of the funda
mental unit has 88 digits! The computer has no problem nding the fundamental unit
the only problem is counting the length of the period, which is about 180.
Example: cubic elds with negative discriminant 91
Example: cubic elds with negative discriminant
Since the sign of the discriminant is (1)
x
(see 2.40), a cubic eld 1 will have negative
discriminant if and only if r =1 =s. We identify 1 with a subeld 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 nd 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 with negative discriminant, and let c be the
fundamental unit with c >1. Then
[z
1
[ <4c
3
24.
PROOF. Since c , 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.
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 nd that
[z
t
[ _16(
2
1.
2
0
.
4
0
) =4u
6
244(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
92 5. The Unit Theorem
EXAMPLE 5.14 Let 1 = 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 =
1
=
10.00998.... Since is a power of c, we must have = 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 nd the order of p. One veries 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 be that
number. If q is a prime ideal such that O
q
(i.e., such that ord
q
() _ 0), then we can
look at mod q and ask if it is a square.
We rst 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 eld O
1
,q is F
29
=Z,(29), and the map
Z F
29
is 2 (mod 29). Thus
1 1. 2 4. ( 2)
1
22. 1 1.
The numbers 1, 4, and 1 12
2
are squares modulo 29, but 22 is not; hence m must be 0.
Since
1
2
<0 it cant be a square in 1 (since it isnt even in 1), and so the only possibility
for is
1
2
. We eliminate this by looking mod 7.
Take q =(7. 3) (see 3.48). Then in the map Z Z,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 .
Similarly, p
2
=(;) can be shown to be impossible. Thus Cl(O
1
) is a cyclic group of
order 6.
Finding j(1)
As we noted eariler, if
n
 1, where
n
is a primitive mth root of 1, then (m)[1: .
Thus there are only nitely 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 nding units in the general case seems to be to nd lots of solutions to
equations Nm() = m for m a xed small number, and then take quotients of solutions.
Note that there can be only nitely many ideals a with N(a) =m; thus if we have lots of
Regulators 93
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 dene. Let t = r s 1, and let
u
1
. .... u
t
be a system of fundamental units. Then the vectors
1(u
i
)
def
=(log[o
1
u
i
[. .... log[o
i
u
i
[. 2 log[o
i1
u
i
[. . . . . 2log[o
t
u
i
[) 1
t
generate the lattice 1(U) in 1
t
. The regulator is dened 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 dene the regulator of any set {c
1
. .... c
t
] of
independent units, and the index of the group generated by the c
i
and j(1) in the full
group of units is measured by ratio
[Reg(c
1
. . . . . c
t
)[,[Reg(U)[.
There are lower bounds for the regulator (see Pohst and Zassenhaus 1989, p 365) similar
to the one we proved for a cubic eld with one real embedding.
For an algorithm that computes the class group, regulator, and fundamental units of a
general number eld, but which requires the generalized Riemann hypothesis to prove its
correctness, see Cohen 1993, Algorithm 6.5.9.
NOTES To nd the units in
_
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 Z because, at the time, a denition of O
1
was lacking. However, his proof extends
easily to O
1
(and to O
1
(S)).
Exercises
51 Fix an m and M. Is it necessarily true that the set of algebraic integers in C of
degree <m and with [[ <M is nite? [Either prove, or give a counterexample.]
52 Find a fundamental unit for the eld
_
67.
53 Let be an element of a number eld 1. Does Nm
1{
() =1 imply that is unit
in O
1
. [Either prove, or give a counterexample.]
2
Kwangho Choiy writes: in the denition 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 denition in the notes is still correct.
3
Dirichlet, P. G. Lejeune, Sur la th eorie des nombres, C. R. Acad. Sci. Paris 10 (1840), 285288. Dirichlet,
P. G. Lejeune, Zur Theorie der complexen Einheiten. Verhandl. Preuss. Akad. Wiss. (1846), 103107.
CHAPTER 6
Cyclotomic Extensions; Fermats
Last Theorem.
The cyclotomic
1
extensions of are those generated by a root of 1. They provide interest
ing examples of the theory we have developed, but, more signicantly, they have important
applications, for example, to Fermats last theorem and to the existence of reciprocity laws
(more generally, to class eld theory itself).
The basic results
An element of a eld 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
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_n1, 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) X,
and clearly
n
() =0. Therefore,
n
(X) is the minimum polynomial of if and only if it
is irreducible, in which case 1 :  =(n) and the map G (Z,nZ)
is an isomorphism.
Hence the following statements are equivalent:
(a) the map Gal(,) (Z,nZ)
is an isomorphism;
(b)  :  =(n):
(c) Gal(,) 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 =
d[n
d
(X) =(X 1)
n
(X).
To nd 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 rst 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 =.
(a) The eld  is of degree (
i
) =
i1
(1) over .
(b) The ring of integers in  is Z.
(c) The element
def
=1 is a prime element of O
1
, and () =()
e
with e =(
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 .
PROOF. Because is integral over Z, the ring Z 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 Z
t
 =Z and
t
 =. Moreover,
1
t
1
=1
x1
Z.
96 6. Cyclotomic Extensions; Fermats Last Theorem.
Similarly, (1 ),(1
t
) Z, and so (1
t
),(1 ) is a unit in Z (hence also in
O
1
). Note that
]
r (X) =
X
]
r
1
X
]
r1
1
=
t
]
1
t 1
=1t t
]1
. t =X
]
r1
.
and so
]
r (1) =.
For its denition, we see that
]
r (1) =
(1
t
) =
1
t
1
(1) =u (1)
(]
r
)
.
with u a unit in Z. Therefore we have an equality of ideals in O
1
,
() =()
e
.
def
=1. e =(
i
). (14)
and so () has at least (
i
) prime factors in O
1
. Now (3.34) implies that  :  _
(
i
). This proves (a) of the Proposition since we know  :  _(
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
(]
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(Z,Z) is a power of . Since
disc(O
1
,Z) (O
1
: Z)
2
=disc(Z,Z).
this will imply:
(i) disc(O
1
,Z) is a power of ;
(ii) (O
1
: Z) is a power of , and therefore
O
1
Z for some M.
To compute disc(Z,Z), we shall use the formula in (2.34), which in our case reads:
disc(Z,Z) =Nm
1{
(
t
]
r ()).
On differentiating the equation
(X
]
r1
1)
]
r (X) =X
]
r
1
and substituting for X, we nd that
t
]
r
() =
i
]
r
1
,(
]
r1
1). Clearly
Nm
1{
=1. Nm
1{
i
=(
i
)
(]
r
)
=
i(]
r
)
.
We shall show that
Nm
1{
(1
]
s
) =
]
s
. 0 _s <r.
and so
Nm
1{
t
]
r () =
c
. c =r(1)
i1
i1
=
i1
(r r 1).
The basic results 97
First we compute Nm
1{
(1 ). The minimum polynomial of 1 is
]
r (1 X),
which has constant term
]
r (1) =, and so Nm
1{
(1) =.
We next compute Nm
1{
(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
(
p
s
j{
(1
]
s
) =.
Using that
Nm
{1
=Nm
1{1
Nm
{1
and Nm
{1
=
:1j
if 1.
we see that
Nm
1{
(1
]
s
) =
o
where a = :
]
s
 =(
i
),(
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
=ZO
1
.
and so, certainly,
O
1
=Z O
1
.
On multiplying through by , we obtain the equality
O
1
=Z
2
O
1
.
Therefore,
O
1
=Z Z
2
O
1
=Z
2
O
1
.
On repeating this argument, we nd that
O
1
=Z
n
O
1
for all m_1. Since
(]
r
)
=(unit), this implies that
O
1
=Z
n
O
1
for all m _ 1. But for m large enough, we know that
n
O
1
Z, and so Z = O
1
.
This completes the proof of (b).
2
REMARK 6.3 (a) The sign of the disc(,), any root of 1, can be computed most
easily by using (2.40a). Clearly  has no real embeddings unless =1 (and  =),
and so, except for this case,
sign(disc(,)) =(1)
x
. s = : ,2.
If is a primitive
i
th root of 1,
i
>2, then
 : ,2 =(1)
i1
,2
98 6. Cyclotomic Extensions; Fermats 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

t
 =.
because if 1 , then ramies totally in 1 and q does not, and if 1
t
, then q
ramies totally in 1 and does not, and these are contradictory unless 1 =.
THEOREM 6.4 Let be a primitive nth root of 1.
(a) The eld  is of degree (n) over .
(b) The ring of integers in  is Z, and so 1. . . . . .
(n)1
is an integral basis for
O
(j
over Z.
(c) If ramies in  then [n; more precisely, if n =
i
m with m relatively prime
to , then
() =(P
1
P
x
)
(]
r
)
in  with the P
i
distinct primes in .
PROOF. We use induction on the number of primes dividing n. Suppose that [n, and
write n =
i
m with m not 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
 =
]
r 
n
. Consider the elds:
n


]
r 
p
i
(pO)
(]
r
)
=
p
i
O
p
(]
r
)
()
According to Proposition 6.2, () ramies totally in
]
r , say () = p
(]
r
)
, but
is unramied in
n
, say () = p
1
p
x
with the p
i
distinct primes. Because 
is obtained from
n
 by adjoining
]
r , its degree over
n
 is at most (
i
). It
follows from Theorem 3.34 that p
1
p
x
can become a (
i
)th power in  only if
:
n
 = (
i
) and each prime p
i
ramies totally in , say p
i
O
(j
= P
(]
r
)
i
.
Therefore,  :  =(
i
) (m) =(n), and to complete the proof, it remains to show
that O
(j
=Z
]
r .
n
 =Z. This is accomplished by the next lemma, because the only
primes that can divide the discriminant of O
(
m
j
,Z are the divisors of m (induction hy
pothesis and 3.35).
2
LEMMA 6.5 Let 1 and 1 be nite extensions of such that
11:  =1:  1: .
and let J be the greatest common divisor of disc(O
1
,Z) and disc(O
1
,Z)). Then
O
11
J
1
O
1
O
1
.
The basic results 99
PROOF. Let {
1
. ....
n
] and {
1
. ....
n
] be integral bases for 1 and 1 respectively. Then
}
is a basis for 1 1 over . Thus every ; O
11
can be written in the form
; =
i}
a
i}
r
}
. 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 subeld of C, every embedding o of 1 into C will extend
uniquely to an embedding of 1 1 into C xing the elements of 1. To see this, write
1 =; then 1 1=1, and the hypothesis on the degrees implies that the minimum
polynomial of doesnt change when we pass from to 1; there is therefore a unique
1homomorphism 1 C sending to o.
On applying such a o to the above equation, we obtain an equation
o(;) =
i}
a
i}
r
o(
i
)
}
.
Write .
i
=
}
(a
i}
,r)
}
, 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 Cramers 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
}
, and the
}
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 ramies
unless (
i
) = 1. Since (
i
) =
i1
( 1), this happens only if
i
= 2. Thus, if
divides n, then ramies in
n
 except when =2 and n =2 (odd number).
(b) Let m be an integer > 1; then (mn) > (n) except when n is odd and m = 2.
Therefore j(
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,) =disc(1,)
1:j
disc(1,)
1:j
. (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(
n
,) =(1)
(n){2
n
(n)
,
][n
(n){(]1)
.
100 6. Cyclotomic Extensions; Fermats Last Theorem.
The example
i.
_
5 =i 
_
5
shows that the condition on the rings of integers is necessary for (15) to hold, because the
extensions i  and
_
5 have discriminants 4 and 20 respectively, but i.
_
5 has
discriminant 4
2
5
2
=4
2
20
2
,4
2
.
Class numbers of cyclotomic elds
Let be a primitive th root of 1, an odd prime. It is known that the class number of 
grows quite rapidly with , and that in fact the class number is 1 if and only if _19.
Here is how to prove that  has class number > 1 when =23. The Galois group
of  over is cyclic of order 22, and therefore has a unique subgroup of index 2. Hence
 contains a unique quadratic extension 1 of . Since 23 is the only prime ramifying in
, it must also be the only prime ramifying in 1, and this implies that 1 =
_
23.
One checks that (2) splits in
_
23, say (2) =pq, that p is not principal, and that p
3
is
principal. Let P be a prime ideal of Z lying over p. Then NP = p
(
, where is the
residue class degree. Since divides  :
_
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 Fermats last theorem, primes such that does not
divide the class number of  are of particular interest. They are called regular. Kummer
found a simple test for when a prime is regular: dene the Bernoulli numbers T
n
by the
formula
t
e
t
1
=
o
n=0
T
n
t
n
n
. T
n
:
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 innitely many irregular primes, but it is still not
proved that there are innitely many regular primes. It is expected that 61% of primes are
regular and 39% are irregular.
Units in cyclotomic elds
Let be a primitive nth root of 1, n >2. Dene

=
1
.
For example, if =e
2ti{n
, then 
=cos
2t
n
. Under any embedding of  into
C,
1
maps to the complex conjugate of , and therefore the image of 
is xed under
complex conjugation and hence lies in 1. Thus, we see that  is a CMeld with maximal
totally real subeld 
.
The rst case of Fermats last theorem for regular primes 101
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
(j
j,j
2
. j =j().
in the proof of Proposition (5.12) is surjective, and so there exists a unit u of  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,
j ={1] (), and so j
2
=(). Therefore
t
=
n
for some integer m. Let
u =a
0
a
(n)1
(n)1
, a
i
Z.
Then u =a
0
a
(n)1
(n)1
, and modulo the prime ideal p =(1) =(1
) of
O
(j
.
u a
0
a
(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 rst case of Fermats last theorem for regular primes
Kummer proved Fermats last theorem for regular primes. Here we prove a weaker result,
known as the rst case of Fermats last theorem.
THEOREM 6.8 Let be an odd prime. If the class number of  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 rst 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 cant 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
,) =:
]
.
102 6. Cyclotomic Extensions; Fermats Last Theorem.
The idea of the proof is to exploit this factorization and what we know of the arithmetic of
 to obtain a contradiction.
Let p be the unique prime ideal of Z 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 Z 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 . , pZ =(). But :
]
=.
]
,
]
. , 0 mod , and so [:,
which contradicts our hypotheses.
2
LEMMA 6.10 For every Z,
]
ZZ.
PROOF. Write
=a
0
a
1
a
]2
]2
. a
i
Z.
Then
]
a
]
0
a
]
1
a
]
]2
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 nZ, 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 Z. 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 Z. 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 Z. This equation implies that a
i
has order dividing in the class
group, but we are assuming that the class group of Z 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 Z. 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 .
On combining these statements, we nd that
i
(. ,)
i
(.
1
,) mod .
Exercises 103
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 (16) says
,
1
, 0 mod .
and Lemma 6.11 implies [,, which contradicts our original assumption.
(b) 1 =
2i1
; then =
2i
, and (16) 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 (16) 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) sufced to prove Fermats 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 X with three real roots.
Let be one of them, and let 1 =. Compute disc(Z,Z), and deduce that
O
1
Z 3
n
O
1
for some m. Showthat and 2 are units in Z and O
1
, and that (1)
3
=3(2).
Deduce that ( 1) is a prime ideal in O
1
, and show that O
1
=Z ( 1)O
1
. Use
this to show that O
1
=Z. 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 cos
2t
n
 is Z2cos
2t
n
.
CHAPTER 7
Absolute Values; Local Fields
In this section, we dene the notion of an absolute value (or multiplicative valuation) and
study the completions of number elds with respect to absolute values.
Absolute Values
An absolute value or (multiplicative) valuation
1
on a eld 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 absolute value.
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 eld 1, and embedding o: 1 C, we get an absolute
value on 1 by putting [a[ =[oa[.
(b) Let ord: 1
.
i
[ _max{[.
i
[]. (17)
PROPOSITION 7.2 An absolute value [ [ is nonarchimedean if and only if it takes bounded
values on {m1 [ m Z].
PROOF. If [ [ is nonarchimedean, then, for m>0.
[m1[ =[11 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(n1)max{[.[. [,[]
n
.
On taking nth roots we nd that
[. ,[ _N
1{n
(n1)
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 absolute values.
PROOF. In this case, the set {m 1 [ m Z] is nite.
2
ASIDE 7.4 Archimedes stated that for any two line segments, laying the shorter segment endtoend
a sufcient nite number of times will create a segment longer than the other. In other words, for
any two nonzero positive real numbers a and b, there is an n N such that b <na. The proposition
shows that the nonarchimedean absolute values are exactly those that dont have this archimedean
property.
As we noted above, a discrete (additive) valuation ord on 1 determines an absolute
value 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.
106 7. Absolute Values; Local Fields
PROPOSITION 7.5 Let [ [ be a nontrivial nonarchimedean absolute value, and put (.) =
log[.[, . =0 (log to base e for any real e > 1). Then : 1
Z 1.
PROOF. That satises (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
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 wont 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 absolute value. 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 absolute value [ [ is discrete if and only if m is principal, in which case is a discrete
valuation ring.
PROOF. The rst 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
[ 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 absolute values. For example,
let
al
be an algebraic closure of . We shall see later that the adic absolute value
[ [
]
: 1 extends to
al
(in many different ways). Since
al
contains an element
1{n
for all n, we see that [
al
[ (
1
)
1{n
=1,
n
_
for all n, and 1,
n
_
1 as n o. In
fact, one can show that [
al
[ ={
i
[ r ], which is not discrete in 1
>0
.
Equivalent absolute values
Note that an absolute value [ [ denes a metric on 1, with distance function
J(a. b) =[ab[.
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 dened by [ [
]
, a and b are close if their difference
is divisible by a high power of . In particular, the sequence
1. .
2
. . . . .
n
. . . .
Equivalent absolute values 107
converges to 0.
The topology dened by the padic absolute value [ [
p
is called the padic topology on
1.
PROPOSITION 7.8 Let [ [
1
, [ [
2
be absolute values on 1, with [ [
1
nontrivial. The following
conditions are equivalent:
(a) [ [
1
, [ [
2
dene 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 sufces 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 absolute values are said to be equivalent if they satisfy the conditions of the propo
sition.
108 7. Absolute Values; 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(ab) _min{ord(a). ord(b)]
and we checked (p55) that this implies that equality holds if ord(a) = ord(b). For multi
plicative valuations, we are given that
[ab[ _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 dene 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 p65 shows that the maximum value of the summands must be
attained for at least two values of the subscript.
Complete list of absolute values for the rational numbers
We now give a complete list of the absolute values on (up to equivalence). We write [ [
o
for the absolute value on dened by the usual absolute value on 1, and we say that [ [
o
is normalized.
THEOREM 7.12 (OSTROWSKI) Let [ [ be a nontrivial absolute value on .
(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).
Complete list of absolute values for the rational numbers 109
(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 rst inequality, we nd that
[m[ _(1r)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 . Since both [ [ and [ [
o
o
are homomorphisms
1
>0
, the fact that they agree on a set of generators for the group
.
CASE (ii): For some n >1, [n[ _1.
In this case, N =1, and (*) implies [m[ _ 1 for all integers m. Therefore the absolute
value is nonarchimedean. Let be the associated local ring and m its maximal ideal. From
the denition of , we know that Z . Then mZ is a prime ideal in Z (because m is
a prime ideal), and it is nonzero for otherwise the absolute value would be trivial. Hence
mZ = () 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 . .
2
THEOREM 7.13 (PRODUCT FORMULA) For =2. 3. 5. 7. .... o, let [ [
]
be the correspond
ing normalized absolute value on . For any nonzero rational number a
[a[
]
=1 (product over all including o).
110 7. Absolute Values; Local Fields
PROOF. Let =a,b, a. b Z. Then [[
]
=1 unless [a or [b. Therefore [[
=1 for
all but nite many s, and so the product is really nite.
Let (a) =
[a[
. Then is a homomorphism
for an
absolute value 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 [ [
O
1
. For an innite prime, it means that n corresponds to an embedding
o: 1 C that extends the embedding corresponding to (or its complex conjugate).
THEOREM 7.15 (PRODUCT FORMULA) For each prime , let [ [
[[
be the
completion of 1 with respect to the absolute value , and let j be a Haar measure on (1
. ) it is uniquely
determined up to a nonzero constant. For any nonzero a 1
, j
o
(U)
def
= j(aU) is also a Haar measure on
(1
. ), and so j
o
=c(a)j for some constant c(a). In fact, c(a) =[a[, the normalized absolute value of a.
The primes of a number eld 111
PROOF. The product formula for a general number eld follows from the product formula
for and the next result.
2
LEMMA 7.16 Let 1 be a nite extension of a number eld 1.
(a) Each prime on 1 extends to a nite 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 nite extension of k(T ) where k is a nite eld. According to (7.3),
the absolute values of 1 are all discrete, and hence correspond to discrete valuation rings
in 1. As in the number eld case, we can normalize an absolute value by setting [a[
=
(1,N)
ord
v
(o)
where N is the number of elements in the residue eld of the discrete valua
tion ring and ord
: 1
Z. Then
[a[
: 1
Z. Fix an e >1,
and dene [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 eld of meromorphic
functions on X. For each point 1 of X we have a discrete valuation, dened 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 C1, 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: =2i(71)
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
71 =0.
i.e.,
ord
1
( ) =0. 1 M.
112 7. Absolute Values; Local Fields
The weak approximation theorem
Recall that an absolute value on a eld 1 is homomorphism a [a[ : 1
1
>0
such
that [ab[ _[a[ [b[ for all a. b 1
1
i
.
The theorem shows that there can be no nite product formula. More precisely:
COROLLARY 7.21 Let [ [
1
, [ [
2
, . . . , [ [
n
be nontrivial inequivalent absolute values on a
eld 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 sufciently large and the [a[
}
, = i , are
sufciently 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 rst occurs in Artin and Whaples 1945.
3
See also Artin
1959, Our account follows the original.
Completions
Let 1 be a eld with a nontrivial absolute value. 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 eld 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 . 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 .
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 for the adic topology.
3
Axiomatic characterization of elds by the product formula for valuations, Bull. AMS, 51, 1945, pp.
469492.
114 7. Absolute Values; Local Fields
THEOREM 7.23 Let 1 be a eld with an absolute value [ [. Then there exists a complete
valued eld (
1. [ [) and a homomorphism 1
1 preserving the absolute value that is
universal in the following sense: any homomorphism 1 1 from 1 into a complete
valued eld (1. [ [) preserving the absolute value, 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 dening
1 to be the set of equivalence classes of Cauchy sequences in 1.
Dene addition and multiplication of Cauchy sequences in the obvious way, and verify
that
1 is a eld. 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 subeld of
1. We can extend a
homomorphism from 1 into a second complete valued eld 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
, 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 absolute value on
1
also. Let ord: 1
[ is discrete), [a
n
[ =[a[
for all n large enough.
The ring associated with [ [ in
1 is
={a
1 [ [a[ _1].
Completions in the nonarchimedean case 115
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
i=n
a
i
i
. Then
[s
s
1
[ _[[
1
, if M <N.
which shows that the sequence s
is Cauchy. Let
1. Because [
1[ = [1[, we can
write =
n
0
with
0
a unit in
. From the denition 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 rst nonzero coefcient. 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 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 rst to be ignored.
We explain this in more detail. The maps
Z,(
n
) Z
(])
,(
n
) Z
]
,(
n
)
116 7. Absolute Values; Local Fields
are both bijective (see 3.10 for the rst 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
]
but not Z
]
, then
n
Z
]
for a sufciently 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
2
.
12 2
n
converges to 1, because the sum of the rst n terms is
2
n
1
21
=2
n
1
which converges to 1.
EXAMPLE 7.28 I claim that 1 is a square in
5
. We have to nd a series
a
0
a
1
5a
2
5
2
. a
i
=0. 1. 2. 3, or 4
such that
(a
0
a
1
5a
2
5
2
...)
2
1 =0.
We rst 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
(2a
1
5)
2
1 0 mod 5
2
.
and so we want
520a
1
0 (mod 5
2
).
We must take a
1
=1. Suppose we have found
c
n
=a
0
a
1
5a
2
5
2
a
n
5
n
such that
c
2
n
1 0 (mod 5
n1
).
Completions in the nonarchimedean case 117
and consider c
n
a
n1
5
n1
. We want
(c
n
a
n1
5
n1
)
2
1 0 (mod 5
n2
),
for which we need that
c
2
n
12c
n
a
n1
5
n1
0 (mod 5
n2
).
or that
2c
n
a
n1
5
n1
(1c
2
n
) (mod 5
n2
).
or that
2c
n
a
n1
(1c
2
n
),5
n1
(mod 5).
or that
4a
n1
=(1c
2
n
),5
n1
(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
_
15 =
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
2
.
n
n
in
]
. Write
n =a
0
a
1
a
i
i
. 0 _a
i
_1.
Then
ord
]
(n) =
_
n
_
n
2
_
_
n
i
_
.
where here a denotes the oor 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 nd that
ord
]
(n) =a
0
0
1
1
a
1
1
1
1
a
2
2
1
1
a
i
i
1
1
=
n
a
i
1
.
Therefore
ord
]
_
.
n
n
_
=n
_
ord
]
(.)
1
]1
_
o
i
]1
.
As
o
i
]1
_
log(n)
log(])
, we see that
x
n
n
0 if and only if ord(.) >
1
]1
. Therefore (see Exercise
72), the series exp(.) converges for ord(.) >
1
]1
.
118 7. Absolute Values; Local Fields
There is a leisurely, and very detailed, discussion of
]
in the rst chapter of Koblitz
1977
4
.
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
n1
a
n
mod m
n
].
In the case that 1 is a discrete valuation ring, this denition agrees with the above. There is an
injective homomorphism
1 1
t
. a (a
n
). a
n
=a mod
n
.
We can dene 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.
Newtons 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) X, 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
n1
.
Let a
n1
=a
n
h
n1
, h . We want
(a
n
h
n1
) 0 mod
n2
.
Recall (trivial Taylors expansion) that, for any polynomial .
(c t ) =(c) t
t
(c)
where
t
(X) is the formal derivative of (X). Then
(a
n
h
n1
) =(a
n
) h
n1
t
(a
n
) .
which we want 0 mod
n2
. Hence we must take h so that
h =
(a
n
)
n1
t
(a
n
)
1
mod .
This is possible because
n1
[(a
n
) and
t
(a
n
)
t
(a
0
) mod .
which is nonzero, and hence invertible, mod .
2
4
Koblitz, Neal. adic numbers, adic analysis, and zetafunctions. Graduate Texts in Mathematics, Vol.
58. SpringerVerlag, New YorkHeidelberg, 1977.
Newtons lemma 119
There is a stronger form of the proposition. Recall Newtons approximation
5
method
for nding a solution to (.) =0, where is a function of a real variable. Starting from
an a
0
such that (a
0
) is small, dene a sequence a
1
. a
2
. ... by putting
a
n1
=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 (NEWTONS LEMMA) Let (X) X. Let a
0
satisfy
[(a
0
)[ <[
t
(a
0
)[
2
.
Then there is a unique root a of (X) such that
[aa
0
[ _
(a
0
)
t
(a
0
)
2
.
PROOF. Dene a sequence a
0
. a
1
. . . . by setting
a
n1
=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 (HENSELS LEMMA) Let k be the residue eld of ; for (X) X,
write
(X) for the image of in kX. Consider a monic polynomial (X) X. If
(X) factors as
= g
0
h
0
with g
0
and h
0
monic and relatively prime (in kX), 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) =X.
We rst prove that (g. h) = X (such a pair is said to be strictly coprime; in kX
strictly coprime just means coprime, i.e., relatively prime).
LEMMA 7.34 Let be a local ring with residue eld k. If . g X are such that
and g are relatively prime and is monic, then (. g) =X. More precisely, there exist
u. X with degu < degg and deg < deg such that
u g =1. (18)
PROOF. Let M =X,(. g). As is monic, this is a nitely generated module. As
(
. g) =kX, we have that (. g) mX =X and so mM =M. Now Nakayamas
Lemma (1.9) implies that M =0.
This shows that there exist u. X such that (18) holds. If deg _ deg , write
=q r with degr < deg . Then
(uqg) rg =1.
and uqg automatically has degree <degg.
2
5
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, 367371.
120 7. Absolute Values; Local Fields
We next prove uniqueness of g and h.
LEMMA 7.35 Let be a local ring with residue eld 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
) =X, and so there exist r. s
X 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
X such that
g
0
h
0
X.
Suppose we have constructed monic polynomials g
n
, h
n
such that
g
n
h
n
0 mod
n1
X
and g
n
g
0
, h
n
h
0
mod X. We want to nd u, X with degu < degg
0
and
deg < degh
0
such that
(g
n
n1
u)(h
n
n1
) 0 mod
n2
X.
i.e., such that
( g
n
h
n
)
n1
(uh
n
g
n
) 0 mod
n2
X.
Thus we are looking for polynomials u, in X with degu < degg
0
and deg < degh
0
such that
uh
n
g
n
( g
n
h
n
),
n1
mod X.
Because g
0
and h
0
are monic and relatively prime, Lemma 7.34 shows that such polyno
mials exist.
REMARK 7.36 An induction argument extends the theorem to show that a factorization of
into a product of relatively prime polynomials in kX lifts to a factorization in X.
For example, in F
]
X, X
]
X splits into distinct factors, and so it also splits in Z
]
X.
Hence Z
]
contains the (1)st roots of 1. More generally, if 1 has a residue eld 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 eld.
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
, dene
[h[ =max[c
i
[.
Extensions of nonarchimedean absolute values 121
Let
(X) =a
n
X
n
a
n1
X
n1
a
0
X
have [a
n
[ =1 (i.e., a
n
is a unit). Let g
0
(X) and h
0
(X) be polynomials in X 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 X 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
]
X (for example).
Given (X), compute disc( ). If this is zero, then and
t
have a common factor,
which we can nd by the Euclidean algorithm. Otherwise ord(disc( )) =m for some m,
and it is enough to consider factorizations of into polynomials with coefcients in the
nite ring Z,
n
Z. Apparently the fastest algorithms for factoring polynomials in ZX
begin by factoring in Z
]
X 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 ZX of arbitrarily large degree that factor in all the rings Z
]
X
into polynomials of low degree.)
Extensions of nonarchimedean absolute values
We explain how to extend a absolute value to a larger eld.
THEOREM 7.38 Let 1 be complete with respect to a discrete absolute value [ [
1
, and let
1 be a nite separable extension of 1 of degree n. Then [ [
1
extends uniquely to a discrete
absolute value [ [
1
on 1, and 1 is complete for the extended absolute value. For all 1.
[[
1
=[ Nm
1{1
[
1{n
1
.
PROOF. Let 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 absolute values 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 T such that P
1
 =P
2
; for example, choose T such that P
1
,
P
2
. Let (X) be the minimum polynomial of over 1, so that  .X,((X)).
Because (X) is irreducible in X and is complete, Hensels lemma shows that
(X)
(image of (X) in kX, k =,p) must be a power of an irreducible polynomial. Then
,p ~kX,(
(X))
is a local ring, which contradicts the fact that  has two prime ideals containing p.
Hence [ [
p
extends uniquely to an absolute value [ [
1
on 1. Similarly, [ [
p
also extends
uniquely to an absolute value [ [
1
0 on a Galois closure 1
t
of 1.
122 7. Absolute Values; Local Fields
For each o Gal(1,1), consider the map 1C, [o[
1
0 . This is again a absolute
value on 1, and so the uniqueness implies that [[
1
=[o[
1
0 . Now
[ Nm()[
1
=[
o[
1
0 =[[
n
1
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
def
=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 absolute value
can only extend to a nonarchimedean absolute value. It is possible to prove (7.38) without
assuming that the absolute value [ [ on 1 is discrete or even nonarchimedean, but the proof
is then completely different, and much longer we shall in fact need this in the Chapter
8, and so I should have included it. The formula [[
1
=[ Nm
1{1
[
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 innite) separable
algebraic extension of 1. Then [ [
1
extends in a unique way to an absolute value [ [
on
.
PROOF. The theorem shows that [ [
1
extends in a unique way to every nite subextension
of , and hence it extends uniquely to .
2
REMARK 7.41 In the last corollary, the extended absolute value 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 absolute value on the algebraic closure
al
]
of
]
is not discrete, and Exercise 77 shows that
al
]
is not complete. The completion of
al
]
is often denoted C
]
because it plays the same role for the adic absolute value on
that C plays for the real absolute value. (In fact C
]
~C as abstract elds because they are
both algebraically closed, and they both have a transcendence basis with cardinality equal
to that of 1. The isomorphism is as far from being canonical as it is possible to get its
construction requires the axiom of choice.)
COROLLARY 7.42 Let 1 and 1 be as in the theorem; then n =e where n =1 : 1, e
is the ramication index, and is the degree of the residue eld 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 ramied over 1; when =n,
we say that 1 is unramied 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).
Newtons polygon 123
REMARK 7.43 Let 1 be complete with respect to a discrete valuation, and let 1 be a nite
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 ramication 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 ramication index.
Newtons polygon
Let 1 be complete with respect to a discrete absolute value. Let ord be the correspond
ing additive valuation ord: 1
ord(
i
)=x
i
(X
i
) has coefcients in 1.
PROOF. In proving the rst part, we dont have to assume that (X) has coefcients in 1
any nite extension of 1 will do. Thus it sufces 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
.
6
Most people write the polynomial a
0
a
1
X X
n
when they dene Newton polygons. This is
slightly less convenient than the way I do it, but allows you to dene the Newton polygon of a power series.
7
See mo15703 for an explanation of what Newton did.
124 7. Absolute Values; Local Fields
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
n1
b
1
X
n
b
2
X
n1
b
n1
.
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 nds 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)[
i
(X) where s
i
= ord(). If is a root of
i
(X),m
(X)[(
i
,m
). Continuing in this
way, we nd that
i
(X) is a product of polynomials with coefcients in 1.
2
EXAMPLE 7.45 Consider the polynomial
8
(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 . The Newton polygon of relative to ord
2
has slopes 0. 1. 2, each with
.length 1. Therefore splits in
2
X, and it has roots
1
,
2
,
3
with ords 0, 1, 2.
Locally compact elds
We now look at the compactness properties of our elds.
PROPOSITION 7.46 Let 1 be complete with respect to a nonarchimedean discrete absolute
value. 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 nite.
PROOF. Let S be a set of representatives for ,m. We have to show that is compact if
and only if S is nite.
=: Clearly m={. 1 [ [.[ < 1] is open in 1. As is the disjoint union of the open
sets s m, s S, S must be nite if is compact.
8
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 dont 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 coefcients; I havent looked up his paper to
be sure, but I did check in Hensels 1894 Crelle paper on extraordinary prime factors of the discriminant that
he wrote the polynomial as X
3
X
2
2X 8.
Locally compact elds 125
=: 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 nite 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 xed n, there are only nitely many sums
s
0
s
1
s
2
2
s
n
n
. s
i
S.
and every element of is within [
n1
[ of such an element.
2
COROLLARY 7.47 Assume that the residue eld is nite. Then p
n
, 1p
n
, and
are all
compact.
PROOF. They are all closed subsets of .
2
DEFINITION 7.48 A local eld is a eld 1 with a nontrivial absolute value [ [ (as dened
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 classication of local elds.
(a) Let 1 be a eld that is complete with respect to an archimedean absolute value [ [;
then 1 is isomorphic to 1 or C, and the absolute value is equivalent to the usual absolute
value (also a theorem of Ostrowski).
9
Thus for archimedean absolute values, completeness
implies local compactness.
(b) A nonarchimedean local eld 1 of characteristic zero is isomorphic to a nite ex
tension of
]
, and the absolute value is equivalent to the (unique) extension of the adic
absolute value. (To prove this, note that, by assumption, 1 contains . The restriction
of [ [ to cant be the trivial absolute value, because otherwise
wouldnt be compact.
Therefore (see 7.12) [ [ induces a absolute valuen on equivalent to the adic absolute
value for some prime number . The closure of in 1 is therefore
]
. If 1 has innite
degree over
]
, it will not be locally compact.)
(c) A nonarchimedean local eld 1 of characteristic = 0 is isomorphic to the eld
of formal Laurent series k((T )) over a nite eld k. The eld k((T )) is the completion
of k(T ) for the absolute value dened by the ideal (T ) kT ; it consists of nitetailed
formal power series:
o
i_n
a
i
T
i
.
9
Here is a sketch of the proof. The eld 1 contains , and the restriction of [ [ to 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 nd that [.[ _[. :[. Hence [. :[ =[.[ and so (taking . :
in place of .) [. 2:[ =[.[, and thus (repeating the argument) [. n:[ =[.[, contradicting the archimedean
property.
126 7. Absolute Values; Local Fields
Unramied extensions of a local eld
Again 1 is a eld complete with respect to a discrete absolute value [ [. To avoid problems
with separability, we assume that 1 and the residue eld k are both perfect
10
of course
in the case we are particularly interested in, 1 has characteristic zero and k is nite. Let
be the discrete valuation ring in 1 corresponding to [ [.
If 1 is an algebraic (possibly innite) extension of 1, we can still dene
T ={ 1 [ [[ _1]
p ={ T [ [[ <1]
and call T,p the residue eld of 1.
PROPOSITION 7.50 Let 1 be an algebraic extension of 1, and let l be the residue eld of
1. The map 1
t
k
t
sending an unramied extension 1
t
of 1 contained in 1 to its residue
eld k
t
is a onetoone correspondence between the sets
{1
t
1, nite and unramied over 1] {k
t
l, nite over k].
Moreover:
(a) if 1
t
k
t
and 1
tt
k
tt
, then 1
t
1
tt
k
t
k
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 nite extension of k. We can write it k
t
= ka. Let
0
(X) be the
minimum polynomial of a over k, and let (X) be any lifting of
0
(X) to X. As a is
a simple root of
0
(X), Newtons lemma (7.31) shows that there is a (unique) 1 such
that () =0 and a mod p. Now 1
t
def
=1 has residue eld k
t
. Thus 1
t
k
t
is
surjective. Suppose that 1
t
and 1
tt
are unramied extensions of 1 in 1 with the same
residue eld k
t
. Then 1
t
1
tt
is an unramied extension
11
of 1 (see 6.5 and 6.6b) with
residue eld k
t
. Hence
1
t
1
tt
: 1 =k
t
: k =1
t
: 1.
and so 1
tt
=1
t
.
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
= ka,
and let g(X) X be such that g(X) kX 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
X, and this implies that g(X) splits in k
t
X, and so k
t
is Galois over k. Let
=k
t
: k =1
t
: 1, and let
1
. . . . .
(
be the roots of g(X). Then
{
1
. ....
(
] ={o [ o Gal(1,1)].
10
When k is not perfect, we should dene 1,1 to be unramied if (a) the ramication index is 1, and (b)
the residue eld extension is separable. These conditions imply that 1,1 is separable. With this denition,
(7.50) continues to hold without 1 and k being assumed to be perfect
11
The results (6.5) and (6.6b) express the discriminant of the composite of 1
t
and 1
tt
in terms of the
discriminants of 1
t
and 1
tt
, from which it follows that if a prime does not divide the discriminant of 1
t
or of
1
tt
, then it doesnt divide the discriminant of their composite.
Totally ramied extensions of 1 127
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
=ka, and
t
lift a. It follows from Hensels
lemma that
t
contains the conjugates of , and hence that 1
t
is Galois over 1.
2
COROLLARY 7.51 There is a eld 1
0
1 containing all unramied extensions of 1 in
1 (called the largest unramied 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 eld of 1
al
is k
al
; there is a subeld 1
un
of 1
al
such that
a subeld 1 of 1
al
, nite over 1, is unramied if and only if 11
un
. (Recall that we are
assuming k and 1 to be perfect.)
PROOF. Let
0
(X) be any polynomial in kX, and let (X) be any lift of
0
(X) to X.
Then 1
al
contains all the roots of (X), and so the residue eld k
t
of 1
al
contains all the
roots of
0
(X). Hence k
t
is algebraic over k, and every polynomial in kX 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 nite unramied extensions
of 1 and the category of nite (separable) extensions of k.
EXAMPLE 7.54 Let 1 be a local eld of characteristic zero (hence a nite extension of
]
for some ), and let q be the order of the residue eld k of 1.
Recall from (FT 4.19) 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 eld 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 unramied extension 1
n
of 1 of degree n, and
it is unique up to 1isomorphism; it is the splitting eld 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
o
q
(mod p).
all T. (Here T is the discrete valuation ring in 1
n
, and p is the nonzero prime ideal in
B.)
Totally ramied extensions of 1
Let 1 be a complete discretelyvalued nonarchimedean eld, and let be a local uniformiz
ing parameter for 1. A polynomial (X) 1X 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.
128 7. Absolute Values; Local Fields
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 ramied extensions of 1.
PROPOSITION 7.55 Let 1 be a nite extension of 1. Then 1,1 is totally ramied if and
only if 1 =1 with a root of an Eisenstein polynomial.
PROOF. =: Suppose 1=1 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 ramication index of 1,1 is _ n. But it cant be
greater than n, and so it is exactly n 1 is totally ramied over 1. (Compare the proof of
6.2.)
=: Suppose 1 is a totally ramied 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
n1
=0. a
i
1
(because of 7.11). Hence 1=1. 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 nite totally ramied 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 = . The argument is the same as in the proof of 6.2 (see also Exercise
61). Because T and have the same residue eld,
 T =T.
The discriminant of 1. .
2
. . . . is a unit
n
for some m, and so
p
c
T  T
for some c. As before, these two conditions sufce to imply that T =.
Ramication groups
Let 1 be a nite Galois extension of 1, and assume that the residue eld k of 1 is perfect.
As we have noted, G
def
= Gal(1,1) preserves the absolute value on 1. In particular, it
preserves
T ={ 1 [ [[ _1]. p ={ 1 [ [[ <1].
Let be a prime element of 1 (so that p = ()). We dene 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 ramication group,
and the groups G
i
, i >1, are called the higher ramication groups of 1 over 1.
Ramication groups 129
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 eld exten
sion l,k is separable.
(a) The xed eld of G
0
is the largest unramied 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 unramied extension in 1 (see 7.51). Then o1
0
is also
unramied, 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 denition G
0
is the kernel of
G Gal(l,k), and so 1
0
is its xed eld.
(b) Let
0
be the discrete valuation ring in 1
0
. Then T =
0
 (by 7.56). Since G
0
leaves
0
xed, in order to check that o G
i
it sufces to check that [o [ < [[
i
for
the element =.
2
COROLLARY 7.59 We have an exhaustive ltration G G
0
such that
G,G
0
=Gal(l,k):
G
0
,G
1
l
:
G
i
,G
i1
l.
Therefore, if k is nite, then Gal(1,1) is solvable.
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 [ _ [[
i1
, and so o = a
i1
some a T. The
map o a (mod p) is a homomorphism G
i
l with kernel G
i1
.
2
An extension 1,1 is said to be wildly ramied if [e where =char(k). Otherwise
it is said to be tamely ramied. Hence for a Galois extension
1,1 is unramied G
0
={1].
and
1,1 is tamely ramied G
1
={1].
130 7. Absolute Values; Local Fields
Krasners lemma and applications
Again let 1 be complete with respect to a discrete nonarchimedean absolute value [ [,
and extend the absolute value (uniquely) to an absolute value on 1
al
. It is clear from our
discussion of unramied 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 kX. Krasners lemma and its consequences show that the roots of two
polynomials will generate the same extension if they are sufciently close.
PROPOSITION 7.60 (KRASNERS LEMMA) Let . 1
al
, and assume that is separable
over 1. If is closer to than to any conjugate of (over 1), then 1 1.
PROOF. Let o be an embedding of 1.  into 1
al
xing 1. By Galois theory, it
sufces to show that o =. But
[o [ =[o o[ =[ [
because o = and [o +[ =[ +[. Hence
[o [ =[o [ _[ [.
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 dene [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
i
=0.
we must have [
n
[ _[c
}
}
[ for some <n, and so [[
n}
_[c
}
[.
Fix a monic irreducible polynomial (X) in 1X, and let
(X) =
(X
i
).
i
1
al
.
The
i
must be distinct. Let g(X) be a second monic polynomial in 1X, and suppose
that [ g[ is small. For any root of g(X), [()[ = [( g)()[ is small (because
[ g[ small implies that [g[ is bounded, and hence [[ is bounded). But
[()[ =
[
i
[.
In order for this to be small, at least one term [
i
[ must be small. By taking [ g[
small enough, we can force to be closer to one root
i
than
i
is to any other
}
. That is,
we can achieve:
[
i
[ <[
i
}
[, all =i.
In this case, we say that belongs to
i
. Krasners lemma then says that 1
i
 1,
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 1X. Then any
monic polynomial g(X) 1X sufciently close to (X) is also irreducible, and each
root of g(X) belongs to some root of (X). For such a root 1 =1.
Krasners lemma and applications 131
COROLLARY 7.62 Let 1 be a nite extension of
]
. Then there is a nite extension 1 of
contained in 1 such that 1:  =1:
]
 and 1
]
=1.
PROOF. Write 1 =
]
, and let (X) be the minimum polynomial of over
]
.
Choose g(X) X sufciently close to (X), and let 1 =  for a root of g(X)
belonging to .
2
Fix a monic polynomial in 1X, and let
1
.
2
. . . . be its roots in 1
al
. As a second
monic polynomial g in 1X approaches , each root
i
of g approaches some root
}(i)
of
, and the function i (i ) doesnt 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 sufciently close, then each root of g will belong to exactly one root of
, and so
{1 [ a root of ] ={1 [ a root of g].
PROPOSITION 7.64 Assume 1 has characteristic zero and has nite residue eld. Then,
up to isomorphism, there are only nitely many totally ramied extensions of
]
of a given
degree.
PROOF. We x an n and show that there are only nite many extensions of degree _ n.
Each point of
(a
1
. .... a
n
) ppp
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 ppp
i
. Show that 1 =;, and deduce that the minimum
polynomial (X) of ; over has degree 2
n
. Show that (X) factors in Z
]
X into a
product of polynomials of degree _4 ( =2) or of degree _8 ( =2).
77 Fix an algebraic closure
al
]
of
]
, and for each n prime to , let
n
be a primitive
nth root of 1. Show that a nite extension 1 of
]
can contain only nitely many
n
s.
Deduce that the Cauchy sequence
n
does not converge to an element of
al
]
.
78 (a) Find two monic polynomials of degree 3 in
5
X with the same Newton polygon,
but with one irreducible and the other not.
(b) Find a monic irreducible polynomial in ZX of degree 6 which factors in
5
X
into a product of 3 irreducible polynomials of degree 2.
CHAPTER 8
Global Fields
A global eld is an algebraic number eld (nite extension of ) or a function eld in one
variable over a nite eld (nite extension of F
q
(T ) for some q). We are mainly interested
in the number eld case.
Extending absolute values
Let 1 be a eld with a absolute value [ [ (archimedean or discrete nonarchimedean), and let
1 be a nite 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 = 1, 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
Then
1 =
1 because
1 is complete, being nite 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
1X.
Conversely, let g(X) be a monic irreducible factor of (X) in
1X, and let
1. =
to 1; if 1
i
denotes the completion
of 1 with respect to the absolute value [ [
i
, then
1
1
1 .
i=1
1
i
. (19)
PROOF. Since 1 is separable over 1, 1 =1 .1X,((X)) for a primitive element
1 and its minimum polynomial (X). Suppose (X) factors in
1X as
(X) =
1
(X)
2
(X)
(X)
with
i
(X) monic and irreducible. Then (see 1.18)
1
1
1 =1
1
1 ~
1X,(((X)) .
1X,(
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 ab (a
1
b. . . . . a
b).
2
REMARK 8.3 Suppose now that 1 is a number eld, that O
1
=O
1
, and that [ [ =[ [
p
for some prime ideal p in O
1
. Because
i
(X) is irreducible in
1X, Hensels lemma
shows that, modulo p,
i
(X) is a power of an irreducible polynomial, say,
i
(X) =g
i
(X)
e
i
.
Then
(X) =
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 absolute values 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
).
The product formula 135
PROOF. By denition the norm and trace of are the determinant and trace of the 1linear
map . .: 1 1. These dont 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
EXAMPLE 8.5 According to PARI
(X) =X
6
5X
5
5X
3
25X 125
is irreducible in X. Its Newton polygon for ord
5
has three segments of .lengths 3, 2, 1
respectively, and so it has at least three factors in
5
. The discriminant of (X) is
2
4
5
11
(59)(365587).
and so according to (7.37), to nd the number of factors of (X) in
5
X, it sufces to
factor in modulo 5
11
. Better, according to Pari, (X) has exactly 3 irreducible factors in
5
X, namely,
X
_
54 5
2
2 5
3
_
O(5
4
)
X
2
_
3 5
2
_
X (55
2
3 5
3
) O(5
4
)
X
3
(3 5
2
5
3
)X
2
_
4 53 5
2
_
X 5O(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 [[
i
, map =
c
}
}
to
i
=
c
}
}
i
1
i
def
=
5
i
,
i
a root of
i
(X), and use that
[[
i
=[
i
[
i
=[ Nm
1
i
{
5
[
1{deg(
i
i
.
The product formula
Before proving the product formula for a number eld, we need one extra fact for local
elds.
Let 1 be a local eld with normalized absolute value [ [. Recall that this means that [ [
is the usual absolute value if 1 is 1, the square of the usual absolute value if 1 is C, and
[a[ =(1,Np)
ord(o)
if the absolute value is dened by a prime ideal p.
Let 1 be a nite separable extension of 1, and let [ [ be the unique extension of [ [ to
1. Let [ [ be the normalized absolute value on 1 corresponding to [ [. What is the relation
of [ [ to [ [?
LEMMA 8.6 In the above situation, [a[ =[a[
n
, where n =1: 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
.
136 8. Global Fields
as required.
Alternatively, use (7.43). For a 1, we have
[a[
def
=NP
ord
L
o
(T.43)
= (Np
(
)
eord
K
o
=[a[
e(
=[a[
n
.
2
PROPOSITION 8.7 Let 1,1 be a nite extension of number elds. For any prime of 1
and 1.
u[
[[
u
=[Nm
1{1
[
.
Here [ [
u
and [ [
to 1, and let [ [
i
be the normal
ized absolute value corresponding to [ [
i
. Then
[Nm
1{1
[
S.4
= [
i=1
Nm
1
i
{
1
[
i=1
[Nm
1
i
{
1
[
T.3S
=
i=1
[[
n
i
i
S.6
=
i=1
[[
u
.
where n
i
=1
i
:
1.
2
THEOREM 8.8 (PRODUCT FORMULA) Let 1 be an algebraic number eld; for all nonzero
1.
u
[[
u
=1.
where the product is over the primes of 1 and [ [
u
is the normalized absolute value for the
prime n.
PROOF. We have
u
[[
u
=
u[
[[
u
_
(8.7)
=
[Nm
1{
[
where runs through the primes 2. 3. 5. 7. .... oof . The last product is 1 by (7.13).
2
ASIDE 8.9 E. Artin and Whaples (1946)
1
proved that global elds can be characterized axiomati
cally. Let 1 be a eld with a set V of primes (equivalence classes of absolute values) satisfying the
following axioms.
AXIOM I. There is a set of representatives [ [
[a[
is a local eld.
Then 1 is a global eld, and Vconsists of all the primes for 1. They then derived the main theorems
(unit theorem and niteness 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 elds and number elds sufciently well, then one could transfer proofs from
function elds to number elds (e.g. the proof of the Riemann hypothesis!). This hasnt 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 eld theory as much as possible working only from the
axioms.
1
Artin, Emil; Whaples, George. Axiomatic characterization of elds by the product formula for valuations.
Bull. Amer. Math. Soc. 51, (1945). 469492. 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.
Decomposition groups 137
Decomposition groups
Let 1 be a nite Galois extension of a number eld 1, and let G = Gal(1,1). For a
absolute value n of 1, we write on for the absolute value such that [o[
cu
=[[
u
, i.e.,
[[
cu
=[o
1
[
u
. For example, if n is the prime dened by a prime ideal P, then on is
the prime dened 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 xed prime of 1, and we dene
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 dened
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
cG{G
w
1
cu
: 1

(S.2)
_ 1 : 1.
Hence equality holds: (G
u
: 1) =1
u
: 1
2
Alternative proof: If o Gal(1
u
,1
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 denition of D(P).
(b), (c), (d) are similarly straightforward.
2
The diagram, and the proposition, show that we can construct a chain of elds
1 1
J
1
T
1
such that all the ramication of P over p takes place in the top extension, all the residue
eld 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 , with Galois group G. Suppose that
O
1
=Z for some 1. Let (X) be the minimum polynomial of over , and write
, and let P
1
=P1.
(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 elds with Galois group G. Given an ideal P
of 1 that is unramied in 1,1 we dene the Frobenius
3
element o =(P. 1,1) to be the
element of G(P) that acts as the Frobenius automorphism on the residue eld. 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 in the residue
eld O
1
,p, p =P1.
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
.
3
Here is a direct proof of the existence of the Frobenius element. Let 1,1 be a nite Galois extension
of number elds with Galois group G, and let P be a prime ideal of O
1
(not necessarily unramied). 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) =
rG
(Xt). 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
, with P, and so
o; o(
i
)
iq
;
q
mod P.
140 8. Global Fields
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.
8.15 Consider a tower of elds
M Q
1 P
1 p
and assume that Q is unramied 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 elds. Then
(P,p) =k(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 eld , 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 141
Examples
We nd the Frobenius maps for quadratic and cyclotomic elds, and obtain a surprisingly
simple proof of the quadratic reciprocity law.
EXAMPLE 8.18 Let 1 =
n
, where
n
is a primitive nth root of 1. If [n then
ramies in 1, and (. 1,) is not dened. Otherwise o =(. 1,) is the unique element
of Gal(1,) such that
o
]
mod p. for all Z
n
.
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 Z
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,) 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 =
_
J, and let be a prime that is unramied in 1. Identify
Gal(1,) with {1]. Then (. 1,) =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,) =
_
d
]
_
.
APPLICATION: THE QUADRATIC RECIPROCITY LAW
Let 1 =, where is a primitive th root of 1, =2. Because Gal(1,) .(Z,Z)
_
=
_
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[q1, 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.
142 8. Global Fields
constitutes the quadratic reciprocity law. We have already proved the rst 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
]
X
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
]
_
.
Computing Galois groups (the hard way)
Let (X) be a polynomial over a eld 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
), dene
o
t
=(t
c(1)
. . . . . t
c(n)
). Let 0 =
i
t
i
, and dene a polynomial
J(X. t ) =
(X o
t
0) (product over o S
n
).
The coefcients 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 1X. t
1
. . . . . t
n
.
THEOREM 8.20 Let G be the set of o S
n
such that o
t
xes 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) X. We may suppose (X) to be monic with integer
coefcients. First nd the roots of (X) to a high degree of accuracy. Then compute
J(X. t ) exactly, noting that this has coefcients 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
xes J
1
(X. t ). The problem
with this approach is that J(X. t ) has degree n. It will probably work (on a computer) if
n _ 5, but otherwise it is like trying to compute a determinant directly from the denition
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 eld
1, and suppose that the Galois group G of (X) has s orbits (as a group of permutations
Computing Galois groups (the easy way) 143
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 1X of degree n
i
.
PROOF. Write (X) =
(X
i
). For S {1. 2. . . . . n], consider
S
=
iS
(X
i
).
This polynomial divides (X), and it is xed under the action of G (and hence has coef
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
Let o S
n
. In GT 4.26, 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 nite
eld k, and let be the splitting eld 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 kX
(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 eld
1, and let G be the Galois group of . Assume (X) O
1
X 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 kX 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 unramied in the splitting eld (because
it implies that p doesnt divide the discriminant of ).
2
REMARK 8.24 There is a similar statement for real primes, namely, if
(X) =
1
(X)
i
(X)
in 1X 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.
144 8. Global Fields
This suggests the following strategy for factoring a polynomial X: 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).
LEMMA 8.26 Let H be a subgroup of S
n
; if H is transitive (for example, contains an
ncycle) and contains an (n1)cycle and a transposition, then H =S
n
.
PROOF. After possibly renumbering, we may suppose that the (n1)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 (n1)cycle and its powers will transform this into (1 n), (2 n), . . ., (n1 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 coefcients in Z
such that
(a)
1
is irreducible modulo 2;
(b)
2
=(degree 1)(irreducible of degree n1) 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 n1:
(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
]
X. 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 nd
all (X)s by factoring X
q
X.
For example, consider X
125
X F
5
X. Its splitting eld 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
X. 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
Computing Galois groups (the easy way) 145
gives a complete list of monic irreducible polynomials of degree 1 or 5 in F
5
X, 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.
CUBIC POLYNOMIALS
The group S
3
has the following subgroups:
order group group elements
1 1 1
2 C
2
1112
3
3
1123
6 S
3
113223.
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 doesnt 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 in Examples 3.48 and 3.49 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 nite primes in a number eld 1, and let 1 be the set
of all nite primes. We say that S has natural density if
lim
1o
[{p S [ Np _N][
[{p [ Np _N][
=.
THEOREM 8.31 (CHEBOTAREV DENSITY THEOREM) Let 1 be a nite Galois extension
of the number eld 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 =[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 1,1 : 1. In particular, there
exist innitely 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) X (say), there exists a bound T such that, if
146 8. Global Fields
a given cycle type doesnt occur as the Frobenius element of some _ T, then it doesnt
occur at all. For a discussion of the effective version of the Chebotarev density theorem,
see Lagarias and Odlysko, 1977.
4
EXAMPLE 8.34 Let 1 =
n
. Then Gal(1,) =(Z,nZ)
g
i
(X) with g
i
(X) irreducible in 1X. Let be a root of g
i
(X)
in some extension eld. 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 coefcients in (by 2.9).
150
151
23. Consider rst the case that 1 =1,
]
=a 1.
24. Clearly 2 does not divide 1
_
3 in Z
_
3, and so a =(2), but
a
2
=(4. 22
_
3. 22
_
3) =(4. 22
_
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 Z
_
3 doesnt have this property.
25. Let 
1
. By hypothesis, we can write
=a
0
a
1
a
n
n
=b
0
b
1
1
b
n
n
.
Let M be the submodule of T generated by {
n
. . . . . 1. . . . .
n
]. Fromthe rst equation,
we nd that
i
M, 0 _ i _ n, and from the second equation we nd that
}
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 Gausss Lemma, one nds that X : ZX Z denes an isomorphism
ZX,((X)) .Z. Hence
3[g() Jh ZX s.t. [g 3h 3[ g.
(c) OK.
(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) nd that (
_
3
_
7),2 is a root
of the polynomial X
4
5X
2
1.
34. Let =kX
2
. X
3
 kX. As kX =kX
2
 1kX
2
 X, it is a Noetherian kX
2

module. Therefore, an ideal in is nitely generated when regarded as a kX
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 nitedimensional vector space over k.
Since it is an integral domain, it must be a eld (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
T , but
X . Therefore is not integrally closed.
41. For example, take T =kX. Y  kX = and p =(Y ), or T =kX k = and
p =(X).
42. Write pT =
P
e(P
i
{p)
i
and P
i
C =
Q
e(Q
ij
{P
i
)
i}
. Then
pC =
i
(P
i
C)
e(P
i
{p)
=
i,}
Q
e(P
i
{p)e(Q
ij
{P
i
)
i}
.
152 A. Solutions to the Exercises
and Q
i}
=Q
i
0
}
0 unless (i. ) =(i
t
.
t
). For the second part of the problem, see the start of
Chapter 4 of the notes.
43. The possibilities are determined by
e
i
i
=3. Since the discriminant is 31, only 31
ramies, and X
3
X 1 (X 28)(X 17)
2
mod 31. All possibilities except () =p
3
occur.
44. Compute the Minkowski bound to nd 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 sufces to nd a eld
1 such that each a
i
becomes principal in 1. Because the ideal class group is nite, each
of the a
i
is of nite order, say a
n
i
i
=(a
i
), a
i
1. Let 1 be a nite 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 eld of 1 (see 4.9), but this
is very difcult to prove it is the Principal Ideal Theorem (see CFT).]
46. The discriminant of X
3
X 2 is (26)2
2
, and Sticklebergers lemma shows 26 is
not a possible discriminant, and so O
1
= Z. 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
=Zi ;, ; =
1
_
5
2
, observe that D(1. ;) =5, and 5 is not a square
in Zi ; now apply Lemma 2.23. The prime 2 ramies in i , but not in
_
5, and so it
ramies in 1 with ramication index 2 (this follows from the multiplicativity of the es).
Similarly, 5 ramies in 1 with ramication index 2. Since disc(O
1
,Zi ) =(5), only the
divisors of (5) (in Zi ) can ramify in
1
, and hence only 2 and 5 can ramify in 1. The proof
that
_
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 innite sets:
{m
_
2m
_
2 [ m. n Z], + =integer part;
{(
_
21)
n
[ n N];
{
_
n
2
1n [ n N];
{ [ is the smaller root of X
2
mX 1 =0. m Z]
52. The period is 10, and the fundamental unit is
488425967
_
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 , then Nm
1
= =Nm
2
, and
so
1
,
2
has norm 1. For example 5 =(2i )(2i ) in i , and so (2i ),(2i ) 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
, not all a
i
Z, will have norm1, but will not
be an algebraic integer.
61. Let be a root of X
3
3X 1. Then disc(Z,Z) = 81. Since its sign is (1)
x
,
we must have s = 0, r = 3 three real embeddings. From their minimum polynomials,
153
one sees that and 2 are algebraic integers with norms 1 and 1 respectively. From
(1)
3
=3(2) we nd (1)
3
=(3) in O
1
. From the formula
e
i
i
=3, we nd
that there can be no further factorization, and e =3, =1. The second equality implies
that O
1
,(1) =Z,(3), and so O
1
=Z(1)
1
. The proof that O
1
=Z proceeds
as in the proof of 6.2. The Minkowski bound is 2, and 2 O
1
is prime, because X
3
3X1
is irreducible modulo 2.
62. First solution: Let be an algebraic integer in
1
. We can write it
=
a
i
(
1
)
i
. 0 _i <(m),2. a
i
.
Suppose a
n
is the last coefcient 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 nd that
t
=a
n
2n
terms of lower degree in . a
n
Z.
This contradicts the fact that
n
t
is in Z.
Second solution: Clearly, O
((
1
j
=O
(j
1
. It follows that the algebraic
integers in
1
 are those elements that can be expressed
a
i
(
i
i
), a
i
Z.
Now prove inductively that
i
i
Z
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
2n1
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). Newtons lemma shows that this
condition is also sufcient.
For a , 7X
2
=a has a solution in
T
if and only if a =7
2n1
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 Newtons 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 Newtons lemma (7.32) again shows that it
has a root in Z
2
.
(b) Apply the method of proof of (7.31) to nd
a =15 77
3
2 7
4
5 7
5
.
75. If k is a eld of characteristic =2, a quadratic extension of k is of the form k
_
a for
some a k, a k
2
, and two nonsquare elements a and b of k dene the same quadratic
extension if and only if they differ by a square (FT, 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 nd a set of representatives for
2
2
in
2
. Clearly an element u 2
n
of
2
,
u Z
2
, is a square if and only if n is even and u is a square in Z
2
, and Newtons 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))
2
,
2
2
, and so the distinct quadratic
extensions of
2
are the elds
_
a] for a =1. 2. 5. 10.
154 A. Solutions to the Exercises
There is a description of the structure of
]
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
]
contains the group
j
]1
of (1)
xt
roots of 1, and one shows that
]
~Zj
]1
U
1
. U
1
~Z
]
. =2:
2
~ZU
1
. U
1
={1] U
2
. U
2
~Z
2
.
There is a general formula,
(1
: 1
n
) =
m
[m[
(j
n
: 1)
for any nite extension 1 of
]
; here j
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
1
.
2
. ....
i1
 because
i
does
not ramify in
1
.
2
. ....
i1
. Therefore the degree is 2
n
(alternatively, use Kummer
theory). The element ; is moved by every element of Gal(1,), and so it generates 1.
The group Gal(1,) 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
]
=
1
i
,
so that 1
i
~ 1X,(
i
(X)) where
i
(X) is the i
th
irreducible factor of (X) in
]
X
(cf. 8.2). Kummer theory and the description of
]
given above show that 1
i
:
]
 _ 4
if = 2 and 1
i
:
2
 _ 8 (because their Galois groups are abelian of exponent 2). This
implies that (X) factors as stated.
77. The degree of
]
n
, does not divide n, is , where is the smallest integer
such that n[
(
1. As n o, o, and so a nite extension 1 of
]
can contain
only nitely many
n
s. Suppose
n
converges to
al
]
. Then 1 =
]
 is a
nite extension of
]
. Let
t
=
t
n=1
n
. Then
t
is further from its conjugates than
it is from , and so Krasners lemma (7.60) implies that
]
t

]
. It follows (by
induction) that
]
 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
35X
5
35X
4
35
4
X
3
35
4
X
2
35
9
X 35
9
.
It is Eisenstein for 3, and hence is irreducible over . Its Newton polygon for 5 has slopes
1,2, 3,2, and 5,2, each of length 2. Correspondingly, in
5
X it is a product of three
polynomials =
1
3
. Each of the
i
is irreducible because the eld generated by a
root of it is ramied (because the slope isnt 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
2
. Now (8.1) shows that [ [
2
has three distinct extensions to 1. Using
that
e
i
i
=3, we see that 2 doesnt 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).
155
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. 33. 15. 114. 1122. 222.
No element of type 2, 3, 3 2, or 4 2 turns up by factoring modulo any of the rst 400
primes (so Im 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 kX and kX
1
 dene the same valuation of k(X) if and only if
they generate the same prime ideal of kX. X
1
. Thus there is one valuation ord
]
for each
monic irreducible polynomial (X) of kX, and one for the polynomial X
1
in kX
1
.
The normalized absolute value corresponding to (X) is
[g(X),h(X)[ =
_
1
q
deg]
_
ord
p
ord
p
h
where q =#k and ord
]
(g) is the power of (X) dividing g(X), and the normalized absolute
value corresponding to X
1
is
[g(X),h(X)[ =
_
1
q
_
deghdeg
.
Thus the product formula is equivalent to the formula,
](A)
deg(ord
]
g ord
]
h) =degg degh.
which is obvious.
APPENDIX B
Twohour examination
Prove (or, at least, explain) your answers.
1. (a) Is (1i ),
_
2 an algebraic integer?
(b) Is Z
_
29 a principal ideal domain?
2. Let 1 = where is a root of X
n
2, n _2.
(a) Find 1 : .
(b) In how many ways can the usual archimedean absolute value on be extended to
1?
(c) Same question for the 2adic absolute value.
(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  contains exactly 3 subelds of degree
2 over , and they are
_
1,
_
2,
_
2.
4. Let and be nonzero elements of the ring of integers O
1
of a number eld 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 =
3
, where is a primitive 3rd root of 1. Find the Galois group of 1 over
3
and its ramication groups.
6. Let 1 be a nite Galois extension of 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.
156
Bibliography
ARTIN, E. 1959. Theory of algebraic numbers. Notes by Gerhard W urges from lectures
held at the Mathematisches 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 elds, 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 Math
ematics Series. Chapman & Hall, London.
CURTIS, C. W. AND REINER, I. 1962. Representation theory of nite groups and associa
tive algebras. Pure and Applied Mathematics, Vol. XI. Interscience Publishers, a division
of John Wiley & Sons, New YorkLondon.
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. Fermats last theorem, volume 50 of Graduate Texts in Mathe
matics. SpringerVerlag, New York. A genetic introduction to algebraic number theory.
FR OHLICH, A. AND TAYLOR, M. J. 1991. Algebraic number theory, volume 27 of Cam
bridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge.
JANUSZ, G. J. 1996. Algebraic number elds, volume 7 of Graduate Studies in Mathemat
ics. American Mathematical Society, Providence, RI, second edition.
LEMMERMEYER, F. 2000. Reciprocity laws. Springer Monographs in Mathematics.
SpringerVerlag, Berlin. From Euler to Eisenstein.
LEMMERMEYER, F. 2007. The development of the principal genus theorem, pp. 529561.
In The shaping of arithmetic after C. F. Gausss Disquisitiones arithmeticae. Springer,
Berlin.
LEMMERMEYER, F. 2009. Jacobi and Kummers ideal numbers. Abh. Math. Semin. Univ.
Hambg. 79:165187.
157
158 Bibliography
LENSTRA, JR., H. W. 1992. Algorithms in algebraic number theory. Bull. Amer. Math.
Soc. (N.S.) 26:211244.
MILNE, J. S. 2006. Elliptic Curves. BookSurge Publishers.
NARKIEWICZ, W. 1990. Elementary and analytic theory of algebraic numbers. Springer
Verlag, 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, Cam
bridge.
ROQUETTE, P. 1967. On class eld towers, pp. 231249. In Algebraic Number Theory
(Proc. Instructional Conf., Brighton, 1965). Thompson, Washington, D.C.
SERRE, J.P. 1962. Corps locaux. Publications de lInstitut de Math ematique de
lUniversit e de Nancago, VIII. Actualit es Sci. Indust., No. 1296. Hermann, Paris.
WEISS, E. 1963. Algebraic number theory. McGrawHill Book Co., Inc., New York.
Index
absolute
discrete, 106
absolute value, 104
adic, 104
trivial, 105
absolute values
equivalent, 107
algebra, 15
algebraic integer, 9
algorithm, 41
good, 41
practical, 41
basis, 32
binary quadratic form, 81
Cauchy sequence, 113
class eld tower, 72
class number, 12, 53
complete eld, 113
continued fraction, 90
convex set, 75
cyclotomic polynomial, 95
Dedekind domain, 48
discrete subgroup, 72
discrete valuation, 55
discrete valuation ring, 46
discriminant, 33, 34, 38
Eisenstein polynomial, 65, 127
Eisensteins criterion, 65
element
irreducible, 10
prime, 9
eld of fractions, 19
Frobenius element, 139
full lattice, 72
fundamental parallelopiped, 74
fundamental system of units, 84
global eld, 133
group
decomposition, 137
higher ramication, 128
inertia, 128
ramication, 128
splitting, 137
Hermite normal form, 43
Hilbert class eld, 72
ideal
fractional, 52
integral, 52
principal, 52
ideal class group, 53
integral basis, 37
integral closure, 29
integral element, 26
integrally closed ring, 29
lattice, 72
lemma
Hensels, 119
Krasners, 130
Nakayamas, 18
Newtons, 118, 119
local eld, 125
local ring, 18
local uniformizing parameter, 114
maximal ideal, 16
Minkowski bound, 69
Minkowski constant, 69
multiplicative subset, 19
natural density, 145
159
160 Index
Newtons polygon, 123
nilpotent, 60
Noetherian module, 17
Noetherian ring, 16
nondegenerate bilinear form, 33
norm, 32, 76
numerical, 68
norm of an ideal, 67
normalized discrete valuation, 55
PARI, 6, 39, 42, 44, 63, 65, 71, 95, 124,
135, 144, 147
prime ideal, 16
primitive nth root of 1, 94
reduced, 60
regulator, 93
relatively prime, 20
ring of integers, 29
Sinteger, 89
Sunit, 89
symmetric in the origin, 75
symmetric polynomial, 26
elementary, 26
tamely ramied, 129
tensor product, 22
theorem
Chebotarev density, 145
Chinese Remainder, 21
Chinese Remainder (for modules), 21
cyclotomic elds, 98
Dedekinds on computing Galois groups,
143
extending absolute values, 121
factoring primes in an extension, 61
Fermats Last, 101
fractional ideals form group, 52
integral closure of Dedekind domain,
56
invariant factor, 57
Minkowski bound, 69
modules over Dedekind domain, 57
points in lattice, 74
primes of a number eld, 110
primes that ramify, 59
product formula, 109, 110, 136
Stickelbergers, 40
sum of efs is the degree, 58
tensor product of elds, 24
the class number is nite, 69
unique factorization of ideals, 48
unit, 84, 87
topology
padic, 107
trace, 32
unique factorization domain, 10
unit, 9
unramied, 71
valuation
archimedean, 104
multiplicative, 104
wildly ramied, 129
Much more than documents.
Discover everything Scribd has to offer, including books and audiobooks from major publishers.
Cancel anytime.