Divisor Theory

Harold M. Edwards

Divisor Theory

Springer Science+Business Media, LLC

Harold M. Edwards
Courant Institute of Mathematical Sciences
New York University
New York, New York 10012
U.S.A.

Library of Congress Cataloging-in-Publication Data

Edwards, Harold M.
Divisor theory I Harold M. Edwards.
p. cm.

ISBN 978-0-8176-4976-0 ISBN 978-0-8176-4977-7 (eBook)

DOI 10.1007/978-0-8176-4977-7
QA242.E33 1990
512'.72---dc20 89-28692

Printed on acid-free paper.

© Springer Science+Business Media New York 1990

Originally published by Birkhauser Boston, in 1990
Softcover reprint of the hardcover 1st edition 1990
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or other-
wise, without prior permission of the copyright owner.

Camera-ready text provided by the author using TEX.

Printed and bound by Edwards Brothers, Inc., Ann Arbor, Michigan.

9 8 7 6 5 4 3 2
 Preface

Man sollte weniger danach streben, die Grenzen der mathe-

matischen Wissenschaften zu erweitern, als vielmehr danach,
den bereits vorhandenen Stoff aus umfassenderen Gesichts-
punkten zu betrachten - E. Study

Today most mathematicians who know about Kronecker's

theory of divisors know about it from having read Hermann
Weyl's lectures on algebraic number theory [We], and regard
it, as Weyl did, as an alternative to Dedekind's theory of ideals.
Weyl's axiomatization of what he calls "Kronecker's" theory is
built-as Dedekind's theory was built-around unique factor-
ization. However, in presenting the theory in this way, Weyl
overlooks one of Kronecker's most valuable ideas, namely, the
idea that the objective of the theory is to define greatest com-
mon divisors, not to achieve factorization into primes.
The reason Kronecker gave greatest common divisors the
primary role is simple: they are independent of the ambient
field while factorization into primes is not. The very notion of
primality depends on the field under consideration-a prime
in one field may factor in a larger field-so if the theory is
founded on factorization into primes, extension of the field
entails a completely new theory. Greatest common divisors,
on the other hand, can be defined in a manner that does not
change at all when the field is extended (see §1.16). Only
after he has laid the foundation of the theory of divisors does
Kronecker consider factorization of divisors into divisors prime
in some specified field.
This book gives a full development of a general theory of
divisors (Part 1), together with applications to algebraic num-
ber theory (Part 2) and the theory of algebraic curves (Part 3).
There is a preliminary section (Part 0) on a theorem of polyno-
mial algebra that is a natural foundation of the theory, and an
Appendix on differentials, which makes possible the statement
VI Preface

and proof of the lliemann-Roch theorem for curves.

The book began more than ten years ago with an effort to
understand Kronecker's theory as it is presented in his trea-
tise Grundzuge einer arithmetischen Theorie der algebraischen
Grossen (§14 et seq.). (For an early version of the result of
these efforts, see the Appendix of [El].) In the intervening
years, as my understanding of the theory has increased, I have
made many simplifications and extensions of it.
The basic idea is simple: For polynomials with rational coef-
ficients, the content of a product is the product of the contents.
(The "content" of a polynomial is the greatest common divi-
sor of its coefficients.) The same should be true of polynomials
with algebraic coefficients. However, the notion of "content"
has no obvious meaning for a polynomial with algebraic co-
efficients. The theory of divisors defines the "content" of a
polynomial with algebraic coefficients in such a way that the
content of a product is the product of the contents.
In fact, this one requirement determines what the content
must be in any particular case; the only problems are to show
that a consistent theory results and to describe in a simple way
what that theory is. As the idea is stated above, and as it will
be sketched in this preface, it applies to polynomials whose
coefficients are algebraic numbers (roots of polynomials in one
indeterminate with coefficients in the ring of integers Z) but
it applies just as well when the integers are replaced by any
natural ring (see §1.2).
Let I and 9 be polynomials (in any number of indetermi-
nates) whose coefficients are algebraic numbers. There is a
polynomial h, whose coefficients are algebraic numbers, such
that Ih has coefficients in Z. (If the coefficients of I are con-
tained in an algebraic number field K, the norm N K I of I
relative to the field extension K :J Q has coefficients in Q,
and one can take h to be (NKJ)/I times a common denomi-
nator of the coefficients of N K I-see §1.17.) The content of I
divides the content of 9 if and only if the content of I h divides
 Preface vii

the content of gh (because the contents of Ih and gh are the

contents of I and g, respectively, times the content of h). But,
since I h has coefficients in Z, its content is a positive integer,
namely, the greatest common divisor d of the coefficients of
I h. Therefore, the statement that the content of I h divides
the content of gh has the natural meaning that each coefficient
of gh is divisible by d (i.e., the coefficient divided by d is an
algebraic integer). Thus, one can test whether the content of
I divides the content of 9 by finding an h, determining d, and
testing whether the coefficients of gh/d are algebraic integers;
the content of I divides the content of 9 if and only if the
answer is yes for all coefficients of gh/d. (See §1.12, Corollary
(12).)
The content of a polynomial is, by definition, a divisor. As
the theory is developed below, the word "content" is not used;
the content of I is called "the divisor represented by I." Divi-
sors are represented by polynomials, divisibility of one divisor
by another can be tested by the method just described, and
two divisors are regarded as equal if each divides the other.
The task is to show that these definitions result in a consistent
theory, and to develop this theory.
The nonzero divisors form a multiplicative group. If one
specifies an algebraic number field K ::J Q and restricts con-
sideration to polynomials with coefficients in K, the multi-
plicative group of divisors coincides with the group of ideals
in K in the sense of Dedekind. The Kroneckerian theory of
divisors has at least three clear advantages over the Dedekin-
dian theory of ideals: (1) It follows from the single, natural
premise that the content of the product of two polynomials is
the product of the contents. (2) It entails an algorithmic test
for divisibility, which, in Dedekindian terms, gives a specific
computation for deciding whether a given element is in the
ideal generated by a finite set of other elements. (For ideologi-
cal reasons that are explained in [E2], Dedekind made a virtue
of the lack of such a test in his theory, whereas Kronecker was
viii Preface

of the opposite opinion.) (3) It is independent of the ambient

field, so that, unlike the Dedekindian theory, all statements
remain true without modification when the ambient field is
extended. (The very definition of an ideal as a certain kind
of subset of the field depends heavily on the ambient field.)
Another advantage of the theory of divisors is its applicability
to integral and nonintegral divisors alike; the theory of ideals
involves both integral and nonintegral ideals, but the integral
ideals are far more natural and are usually the only ones in-
troduced in the early stages of the theory.
As was already remarked, the theorem which states that a
divisor (or an ideal) can be written in one and only one way
as a product of powers of distinct prime divisors (ideals) has
meaning only when an ambient field is specified, because only
then does the word "prime" have meaning. However, many
applications of this theorem require only a decomposition into
a product of powers of relatively prime divisors, a notion inde-
pendent of the ambient field. Thus, factorization into primes
can often be replaced by Theorem 1 of §1.19, which states that
the divisors in any given finite set can be written as products
of powers of relatively prime integral divisors.
A second fundamental theorem of divisor theory is the fol-
lowing: An integral divisor divides at least one algebraic inte-
ger. Therefore, given an integral divisor A, there is an integral
divisor B such that AB is the divisor of an algebraic integer.
Theorem 2 (§1.20) states that, given any integral divisor C,
one can in fact choose B to be relatively prime to C (i.e.,
there is an integral divisor B relatively prime to C for which
AB is the divisor of an algebraic integer). One corollary of this
theorem is that every divisor is the greatest common divisor of
just two algebraic numbers. Another corollary is a divisibility
test of the type used by Kummer in his definition of "ideal
prime factors" of cyclotomic integers, the notion with which
he initiated divisor theory, in a special case, in 1846. (See also
§2.11. )
 Preface ix

The remaining topics treated in the general theory of Part

1 all relate to divisors in some particular ambient field. They
include divisor class groups, the factorization of prime divi-
sors in normal extension fields, ring;s of values of a given field
at a given integral divisor, discriminants, differents, and ram-
ification. A topic not covered in the general theory is the
representation of an arbitrary divisor as a product of powers
of divisors prime in a given field. The problem of giving, in
the general case, an algorithm for fCl.ctoring an integral divisor
into irreducible (prime) factors (see §1.18) appears to be more
difficult than the problems treated here. At any rate, it is a
problem for which I do not have a solution.
However, in the special case of Part 2-the case of algebraic
number fields-it is quite simple to write any given divisor as
a product of powers of prime divisors (§2.1). Much of Part
2 is devoted to proving the validity of the following method
for factoring the divisor of a prime integer p in an algebraic
number field. Let aI, a2, ••. , a v be algebraic integers and
let K = Q( aI, a2, ... , a v ) be the field they generate over
Q. Let X, UI, U2, •.. , U v be indeterminates, let F(X, UI, U2,
... , u v ) be the norm of X - UI al - U2a2 - ••• - uva v relative
to K J Q (F is a polynomial with coefficients in Z), and let
F = 1r 1;2 ... 1!.m mod p be the factorization of F mod pinto
powers of distinct irreducible polynomials with coefficients in
the field Z mod p. Normally the divisor Pi represented by the
polynomial p+ h( al UI + ... +avu v , Ul, U2, •• . , u v ) is prime in
K, Pi =I- Pj for i =I- j, and p;l p;2 ... p~m is the divisor of p.
For a given set of a's, this is true for all but a finite number
of primes p (§2.4). IT the a's have the property that every
algebraic integer in K can be expressed as a polynomial in the
a's with coefficients in Z-and for l~iven K such a set of a's
can always be found-it is true for l~ll primes p.
This method of factoring the divisor of p was first proposed
by Kronecker [Krl, §25], who proposed it in a much more
general case than the case of algebraic number fields. The va-
x Preface

lidity of the method in the number field case was proved by

Hensel, but, although Hensel stated [He, p. 76] that he had a
proof which he would soon publish, I do not know of a proof
of Kronecker's more general case. (IT one could be given, it
would be a large step toward the solution of the problem of
factoring divisors in the general case.)
Two other topics treated in Part 2 are Dedekind's discrimi-
nant theorem (§2.8) and the factorization of primes in subfields
of cyclotomic fields (§2.11).
Application of the theory of divisors to algebraic curves in
Part 3 calls for a slight extension of the notion of "divisor."
The field of functions K on an algebraic curve over the ratio-
nals is an algebraic extension of the natural ring Q[x] and as
such has a divisor theory. This divisor theory depends, how-
ever, on the choice of a parameter x on the curve. A global
divisor in such a field K is the assignment to each parameter
x of a divisor Ax in the divisor theory for this parameter in
such a way that the divisors Ax "agree on overlaps" in a nat-
ural way (§3.4). In Part 3, unless otherwise stated, "divisor"
means "global divisor."
Divisor theory provides the following answer to the perennial
question "What is a point?" in the theory of algebraic curves.
Let K be the field of rational functions on the algebraic curve
defined by the equation F(x, y) = 0, where F is an irreducible
polynomial in two indeterminates with coefficients in Z. Let
a and b be algebraic numbers in K such that F( a, b) = 0.
There is associated to such a pair of algebraic numbers a divisor
in K, namely, the numerator of the divisor represented by
(x - a)U + (y - b)V (where x - a and y - b are elements of
K, and U and V are indeterminates). IT (a, b) is a nonsingular
point of F = 0, that is, if the partial derivatives of F at (a, b)
are not both zero, the divisor in K obtained in this way is
called a place (see §3.13). Places can also be characterized as
divisors in K which are prime in K and in all extensions of
K obtained by adjoining constants to K (§3.22). Yet another
 Preface xi

characterization of places is as divisors for which the ring of

values coincides with the field of constants of K.
Places capture algebraically the notion of a point on a curve.
For example, the origin (0,0) does not give rise to a place on
the folium of Descartes x 3 + y3 - xy = 0, but, rather, gives rise
to a product of two places, namely, the zero and the pole of
the "function" x/yon this curve. That the origin is a product
of two places expresses the geometrical "fact" that the origin
is a double point of the folium of Descartes (§3.13).
A fundamental theorem states that every divisor in the field
of functions K on an algebraic curve, can, when suitable con-
stants are adjoined, be written as a product of powers of places
(§3.18). (This theorem relates divisors as they are treated in
this book to divisors as they are customarily defined in the
theory of algebraic curves as formal sums of places with inte-
ger coefficients, or, equivalently, as formal products of places
with integer exponents.) The degree of a divisor is equal to the
number of places in the numerator of such a representation of
the divisor minus the number of places in the denominator.
The degree of the divisor of an element x of K is always 0, and
the places in its numerator are naturally thought of as "the
points where x is zero" and the places in its denominator as
"the points where x has poles."
Divisor theory also provides the following natural formula-
tion of Abel's theorem: For any given function field (of one
variable, over Q) there is a least integer g, the genus of the
field, such that every divisor of degree 9 is equivalent to an in-
tegral divisor. Otherwise stated, given a set of zeros and a set
of poles, there is a function on the eurve with poles, at most,
at the given poles and zeros at the g:iven zeros (plus, possibly,
other zeros) provided the number of given poles is at least 9
greater than the number of given zeros. This theorem is not
immediately recognizable as Abel's theorem, but the connec-
tion with Abel's own statement is explained in §3.25. (See also
the Corollary of §3.27.)
xii Preface

An element of K can be expanded in a natural way in powers

of a local parameter at a place P in K (§3.16). The principal
part of an element of K at P, relative to a given local parameter
at P, is the terms of this power series expansion which have
negative degree. (In particular, the element of K has a pole
at P if and only if its principal part at P, relative to any
local parameter at P, is nonzero.) Let PI, P2 , ••• , Pk be a
given set of places in the field K of rational functions on an
algebraic curve, and let r be the elements of K which have
poles only at the Pi and whose poles at the Pi have order N
at most. The principal parts of an element of r are described
by Nk constants of K. Abel's theorem easily implies that the
principal parts of elements of r form a subspace of codimension
at most g of Kfk. The Appendix is devoted to proving that
this subspace of Kfk can be described in terms of differentials.
Specifically, differentials are defined in the Appendix, and it
is shown that (1) g is the dimension of the vector space (over
the field of constants Ko of K) of holomorphic differentials,
(2) the sum of the residues of any differential is zero, and
(3) the conditions "any element of K times any holomorphic
differential must have the sum of its residues equal to zero"
give necessary and sufficient conditions for determining which
elements of Kfk are principal parts of elements of r. (For
N large, these conditions are also independent, so that the
codimension is exactly g when N is large.) The lliemann-Roch
theorem is a simple corollary of this method of determining the
principal parts of elements of r.

I wish once again to thank the Vaughn Foundation for more

than a decade of support which made an enormous difference
in my life and work.
 Contents

Preface .............................................................................. v

Part O. A Theorem of Polynomial Algebra ............................... 1

Part 1. The General Theory ........................................................ 13

§1.1 Introduction. §1.2 Natural Rings. §1.3 On Existence.
§1.4 Preliminaries. §1.5 The Basic:Theory. §1.6 Definitions.
§1.7 What is a Divisor? §1.8 §1.91 §1.10 First Propositions.
§1.11 The Main Proposition. §1.12 Concluding Corollaries.
§1.13 Testing for Divisibility. §1.14 The Group of Nonzero
Divisors. §1.15 Greatest Common Divisors. §1.16 Ambient
Fields. §1.17 Norms. §1.18 Factorization of Divisors. §1.19
§1.20 Two Basic Theorems. §1.21 The Divisor Class Group.
§1.22 Prime Factorization and Normal Extensions. §1.23 §1.24
Rings of Values. §1.25 Primitive Elements, Norms, and Traces.
§1.26 §1.27 §1.28 Differents. §1.29 !i1.30 Discriminants. §1.31
§1.32 Ramification.

Part 2. Applications to Algebraic Number Theory ......... 60

§2.1 Factorization into Primes. §2.2 §2.3 §2.4 A Factoriza-
tion Method. §2.5 Examples. §2.6 Integral Bases. §2.7 Proof
of the Theorem of §2.4. §2.8 Dedekind's Discriminant The-
orem. §2.9 Differents and Discriminants. §2.10 §2.11 Cyclo-
tomic Fields. §2.12 Quadratic Reciprocity.

Part 3. Applications to the Theory of Algebraic Curves ... 85

§3.1 Function Fields. §3.2 §3.3 Parameters and Constants.
§3.4 §3.5 §3.6 §3.7 §3.8 §3.9 §3.10 Global Divisors. §3.11 Nu-
merical Extensions. §3.12 §3.13 §3.14 The Idea of a Place.
§3.15 §3.16 Local Parameters at a Place. §3.17 Relative norms.
§3.18 §3.19 A Divisor is a Product of Powers of Places. §3.20
Presentation of Places. §3.21 Degree of a Divisor. §3.22 A
Characteristic of Places. §3.23 Dimension of a Divisor. §3.24
XIV Contents

The Genus of a Function Fiel.d. §3.25 Abel's Theorem. §3.26

The Genus as a Limit. §3.27 A Converse of Abel's Theorem.
§3.28 The Divisor Class Group. §3.29 Examples.

Appendix ...................................................................... 137

§A.l Introduction. §A.2 Definitions and First Propositions.
§A.3 Orders and Residues of Differentials. §A.4 The Sum of
the Residues is Zero. §A.5 Holomorphic Differentials. §A.6
Integral Bases. §A.7 Normal Bases. §A.8 §A.9 The Dual of
a Normal Basis. §A.I0 Construction of Holomorphic Differ-
entials. §A.ll Examples. §A.12 The Riemann-Roch Theorem
for Integral Divisors. §A.13 Riemann-Roch for Reciprocals of
Integral Divisors. §A.14 General Case of the Riemann-Roch
Theorem.

References 165
 Part 0: A Theorem of Polynomial Algebra

What is usually called "Gauss's lemma" today differs from

Gauss's actual lemma (Disquisitione,s Arithmeticae, Art. 42).
GAUSS'S VERSION. Let f and 9 be monic polynomials in one
indeterminate with rational coefficients. If the coefficients of
f and 9 are not all integers, then the coefficients of f 9 cannot
all be integers.
MODERN VERSION. (cf. Bourbaki, Algebre Commutative, VII,
3, 5) Let f and 9 be polynomials in one indeterminate with
integer coefficients. The content of f 9 is the content of f times
the content of g.
(The "content" of a polynomial with integer coefficients is
the greatest common divisor of its eoefficients.) Gauss's ver-
sion has the advantage that it lends itself to the following pro-
found generalization.
GENERALIZATION. Let f and 9 be monic polynomials in one
indeterminate whose coefficients are algebraic numbers. If the
coefficients of f and 9 are not all algebraic integers, then the
coefficients of f 9 cannot all be algebraic integers.
(An algebraic number is an element of an extension of finite
degree of the field of rational numbers Q. The coefficients of f
and 9 are assumed to be elements of a single extension field K
of Q of finite degree. An element of.K is an "algebraic integer"
if some power of it can be written as a linear combination of
lower powers with integer coefficients.)
The modern version has no such generalization because the
notion of "greatest common divisor" is not defined for algebraic
integers. On the other hand, the modern version of Gauss's
lemma does have advantages-for example, it is no harder to
prove, and it easily implies Gauss's version, but Gauss's ver-
sion does not easily imply it-so a formulation of the modern
version which generalizes to the case of algebraic numbers in
2 Divisor Theory

the way that Gauss's does would be desirable. Dedekind rec-

ognized this need and published such a formulation [DI] in
1892.
DEDEKIND'S VERSION. Let 1 and 9 be polynomials in one in-
determinate with rational coefficients. If 19 has integer co-
efficients, then the product of any coefficient of 1 and any
coefficient of 9 is an integer.
This version is as easy to prove as Gauss's, and it implies
both Gauss's version and the modern version. (Since 1 is a
coefficient of a monic polynomial, Gauss's version follows im-
mediately. Let 1 and 9 have integer coefficients and let a and b,
respectively, be their contents. Since ab divides all coefficients
of Ig, it divides the content of Ig. IT it were not equal to the
content of Ig, there would be a prime p such that the coeffi-
cients of (Ig)f(abp) were integers. It would then follow from
Dedekind's version of Gauss's lemma that every coefficient of
1fa times every coefficient of gfb was divisible by p, which
is impossible because both If a and 9 f b have a coefficient not
divisible by p, and p is prime.)
Dedekind's version does generalize, as he proved, to the al-
gebraic case. He published this theorem in the Proceedings of
the German Mathematical Society of Prague and later referred
to it as the Prague theorem.
DEDEKIND'S PRAGUE THEOREM. Let 1 and 9 be polynomials
in one indeterminate whose coefficients are algebraic numbers.
If all coefficients of 19 are algebraic integers, then the product
of any coefficient of 1 and any coefficient of 9 is an algebraic
integer.
Dedekind was understandably not aware that his Prague
theorem was a consequence of a theorem published ten years
earlier by Kronecker [Kr2]. Kronecker's brief paper is exceed-
ingly terse and obscure. However, it contains, couched in Kro-
necker's private terminology, a theorem which implies:
 Part o. A Theorem of Polynomial Algebra 3

THEOREM. Let ao, al, ... , am, bo, bl , ... , bn be indeterminates

and let R be tbe ring of polynomials in tbese indeterminates
witb integer coefficients. Let co, CI, ••• , C m +n E R be defined
by

j+k=i
O:5j:5m,O:5k:5 n

(In other words, Ci is the coefficient of x m+n - i in (aoxm +

alx m- l + ... + am)(box n + blx n - l + ... + bn ), where x is a
new indeterminate.) Tben eacb of tbe (m + 1)( n + 1) elements
ajh (0:::; j :::; m, 0 :::; k :::; n) of R is integral over tbe subring
of R generated by 1, Co, CI, ... , C m + n . In otber words, some
power of eacb ajb k can be written as a linear combination of
lower powers

in wmcb tbe coefficients Pi are elements of R expressible as

polynomials in Co, CI, ... ,C m +n witb integer coefficients.
Examples: When m = n = 1,

as is easily verified by substituting Co = aob o, CI

a l bo,c2 = alb l .
When m = n = 2,

where

PI = 3 CI, P2 = -3ci - 2COC2, P3 = c~ + 4COCIC2,

P4 = -C~CIC3 - 2cocic2 - c~c~ + 4cgC4,
P5 = COC
22 2 243
I C3 + CoCI c 2 - COCI C4, P6 = -COCI
3 42 32
C2 C3 + COC3 + COC 1 C4

(Found with the aid of a computer.)

4 Divisor Theory

COROLLARY 1. Let / and 9 be polynomials in one indeter-

minate with coefficients which are algebraic numbers, and let
/ 9 have integer coefficients. Then any coefficient of / times
any coefficient of 9 is an algebraic integer.
DEDUCTION: Substitute the coefficients of / and 9 in the ob-
vious way into the formula (ajbk)N = EPi . (ajbk)N-i of the
theorem. Since this substitution gives the c's integer values, it
gives the p's integer values and shows that the value it "gives
to ajbk is an algebraic integer.
Corollary 1 is a weakened version of Dedekind's Prague
theorem-it applies only when /g has integer coefficients, not,
as Dedekind's theorem does, when /g has algebraic integer
coefficients. It is easy to deduce Dedekind's theorem itself as
a corollary of the theorem (see Proposition 2 below), but since
Dedekind's theorem will not be needed in the sequel, this de-
duction will be left to the reader.
COROLLARY 2. Dedekind's version of Gauss's lemma.
DEDUCTION: This follows from Corollary 1 and the elemen-
tary fact that a rational number which is an algebraic integer
is an integer.
COROLLARY 3. Corollary 1 is also true for polynomials in any
number of indeterminates. That is, if / and 9 are polynomi-
als (in any number of indeterminates) whose coefficients are
algebraic numbers, and if / 9 has integer coefficients, then the
product of any coefficient of / and any coefficient of 9 is an
algebraic integer.
DEDUCTION: Let Xl,X 2 , ••• ,Xv be the indeterminates which
appear in / and g, and let k be an integer greater than the
degree of /g in any of the indeterminates Xi it contains. Define
a homomorphism if> from polynomials in Xl, X 2 , ••• , Xv to
polynomials in T by if>(Xi ) = T ki - 1 • Because el + e2k + e3k2 +
... = e~ + e~k + e;k2 + ... for 0 ~ ei < k, 0 ~ e~ < k implies
 Part O. A Theorem of Polynomial Algebra 5

ei = e~, <I> applied to different monomials X:l X;2 ... X~v in

which the exponents ei are all less than k gives different powers
of T. Therefore, the coefficients of <I>(J g) = <I>(J)<I>(g) coincide
with the coefficients of Ig. By Corollary 1, any coefficient of
<I>(J) times any coefficient of <I>(g) is an algebraic integer. Since
the coefficients of <I>(J) coincide with those of I, and of <I>(g)
with those of g, Corollary 3 follows.
COROLLARY 4. Let r be an integral domain and let k be its
field of quotients. If 1 and 9 are polynomials with coefficients
in an algebraic extension K of k such that 19 has coefficients
in r, then the product of any coefficient of 1 and any coefficient
of 9 is integral over r (that is, some power of it is equal to a
linear combination of lower powers with coefficients in r).
DEDUCTION: The case where 1 and 9 are polynomials in one
indeterminate follows from the Theorem as in Corollary 1. The
case where they are polynomials in several indeterminants then
follows as in Corollary 3.
It was noted above that the modern version of Gauss's
lemma does not apply to polynomials whose coefficients are
algebraic integers because the concept of "greatest common
divisor" is not defined for sets of algebraic integers. The basic
idea of the theory of divisors is, roughly, to use this identity
"content of product = product of contents" to define the con-
tent of a polynomial whose coefficients are algebraic numbers.
As is shown in Part 1, the key to proving that this idea leads
to a coherent theory with all the important characteristics of
"greatest common divisors" is the above theorem-and more
particularly Corollary 4. Kronecker was unquestionably aware
of the connection, because he opens his paper with a mention
of "the effort to find the simplest foundation of the theory of
forms" (his major treatise [Krl], published one year earlier,
in which he develops the theory of divisors, treats divisors as
an aspect of the theory of "forms" or polynomials). However,
Kronecker appears never to have published anything further
6 Divisor Theory

on the subject of "the simplest foundation of the theory of

forms", and the account in this one paper is, as I have already
said, opaque.
PROOF OF THE THEOREM: (Hurwitz[Hu2]) Let x, to, tt, ... , tn
be indeterminates, and let R = R[x, to, tt, ... , tn] =
Z[ao, ... , am, bo, bl , ... , bn , x, to, tl, ... , tn] be the ring of poly-
nomials in all these indeterminates with coefficients in the ring
of integers Z. Let f(x) = aox m + alX m - 1 + ... + am E R and
g(x) = box n + blx n - l + ... + bn E R.
LEMMA 1. (Lagrange Interpolation) In R,
II (tj - tk) = L Qi(tO, tt, ... ,tn)g(ti) II(x - tj)
n

g(x)
j=# i=O j=j:. i
where the product on the left is over all (n + 1)2 - (n + 1)
factors tj - tk in which 0 ~ j ~ n, 0 ~ k ~ n, j k, where t=
9(ti) = Ej=o bjt~-j E R, where the products on the right are
over all n factors x - tj in which 0 ~ j ~ n, j i, and where t=

a polynomial in the t's of degree (n + 1)2 - (n + 1) - n = n2 •

PROOF: Both sides are elements of R of degree n, at most, in
x, so their difference is also a polynomial of degree n, at most,
in x. This difference is divisible by x - til for v = 0,1, ... , n
by the Remainder Theorem, because division of the left side
by x - til leaves a remainder of g(tll) TIj=j:.k(tj - tk) and di-
vision by x - til of the one term of the sum on the right
side which is not divisible by x - til leaves the same remain-
der QII( to, tl, ... ,tn)g(tll) TI j=j:.II(t1l - t j) = g( til) TIj=j:.k(tj - tk).
Thus, the difference is divisible by the polynomial TIj=o(x-tj)
of degree n + 1 in x, which implies that the difference is zero.
 Part O. A Theorem of Polynomial Algebra 7

LEMMA 2. Let S denote tbe subset of R consisting of tbose

elements wbicb can be expressed as polynomials in tbe a's
and c's tbat are bomogeneous of degree n in tbe a's and
bomogeneous of degree 1 in tbe c's. Tbe coefIicients of
g(x) I1~=o f(ti) I1#k(tj - tk), regarded as a polynomial in x
and tbe t's witb coefIicients in R, are in S.
PROOF: Multiplication of the formula of Lemma 1 by
I1~=o f(ti) shows that the polynomial in question is equal to

i=O j-::f.i j-::f.i

Sincef(ti)g(ti) = L:::oncvtr+n-v andf(tj) = L::=oavtj-V,

while the Qi and I1#i(x-tj) have coefficients in Z, the lemma
follows.
LEMMA 3. Tbe coefIicients of g(x) I1~=o f(ti), regarded as a
polynomial in x and tbe t's witb coefIicients in R, are in S.
PROOF: Let the polynomial of Lemma 2 be successively di-
vided by the (n + 1)2 - (n + 1) polynomials t j - tk' As follows
from the division algorithm, the coefficients of each successive
quotient are in S. Explicitly, if (tj - tk)H = G, and if the
coefficients of G are in S, then the coefficients of H are in
S because, when G = Go + GItj + G 2 t] + ... + GJLtj is di-
vided by tj - tk to find H = Ho + HItj + ... + HJL_It'j-l, the
division algorithm gives H JL - I = G JL , H JL - 2 = G JL - I + tkG JL ,
H JL - 3 = G JL - 2 + tkG JL - I + tiGJL' .. , . Therefore, if all coeffi-
cients of G are in S, so are all coefficients of H. The initial
polynomial is the polynomial of Lemma 2, and the final one is
the polynomial of Lemma 3.
LEMMA 4. Let M I , M 2 , ••• , MT be a list of all mono-
mials a~o a~l ... a~m in tbe a's of degree n. Tbe equations
ajbkMp = L::=I AjkpuMu define linear polynomials Ajkpu in
8 Divisor Theory

Co, CI, ... ,Cm+n with coefficients in Z for all sets of indices in
the ranges 0 ::; j ::; m, 0 ::; k ::; n, 0 ::; p ::; T, 0 ::; u ::; T.
PROOF: Each ajbkMp is a coefficient of g(x) n:=o f(ti).
Therefore, it has the desired form by Lemma 3.
CONCLUSION OF THE PROOF OF THE THEOREM: For fixed
indices j and k, 0 ::; j ::; m, 0 ::; k ::; n, let Ajk be the
TXT matrix whose entries are the >'jkpcr of Lemma 4. Then
(ajbkI - Ajk)M = 0, where I is the TXT identity matrix,
where M is the column matrix with entries M I , M 2 , ••• , M r ,
and where the 0 on the right represents a column matrix of
length T whose entries are all O. Since M is nonzero with
entries in the integral domain Rand ajbkI - Ajk is a square
matrix with entries in R, it follows that det( ajbkI - Ajk) = O.
In other words, ajb k is a root of the polynomial det(XI - A jk )
in the indeterminate X. Since this polynomial has the form
xr +Plxr-l +P2xr-2 + ... +Pr, where the Pi are polynomials
in the >"s, the Theorem follows.
An equivalent theorem somewhat closer to the theorem Kro-
necker states is:
ALTERNATE THEOREM. As before, let R = Z[ao, aI, ... , am,
bo, bl , ... , bnl and let Ci = L:j+k=i ajbk. Let G be (aoxo +
alXI + ... + amxm)(bOYO +.bIYI + ... + bnYn), a polynomial in
xo, Xl, ... , x m , Yo, YI, ... , Yn with coefficients in R. Then,
for some positive integer p, GP can be expressed as a linear
combination of lower powers of G with coefficients which are
polynomials in the c's, x's, and y's with coefficients in Z.
Example: (Kronecker [Kr2, p. 424]) When m = n = 1,
G2 - PG + Q = 0 where, with G' = (boxo + bl XI)( aoYo + al YI),
P = G + G' and Q = GG'. That P and Q are polynomials in
co, CI, C2, Xo, Xl, Yo, YI can be verified as follows. G = coxoYo+
aoblxoYI + alboxlYo + C2XIYl. Thus P = 2coxoYo + CI(XOYI +
xIYO)+2c2XIYI and Q = (coXOYO+C2XIYI?+(COXOYO+C2XIYJ)'
(CIXOYI +CIXI Yo)+( aoblxoYI +al bOXIYO)( al bOXOYI +aoblxIYO)j
 Part o. A Theorem of Polynomial Algebra 9

the last term in this expression of Q is coc2 x5Y; + (a5bi +

aib~)xOXIYoYl + coc2xiY5, which can be expressed in terms of
c's by virtue of ci = a~bi + aib~ + 2COC2.
The previous theorem follows from this one when 1 is sub-
stituted for one x and one y, and 0 is substituted for all the
others. Conversely, this theorem follows from the previous one
when use is made of a few basic facts about integrality:

DEFINITION. An element a of a ring R is integral over a subring

R' of R if some power of a is equal to a linear combination of
lower powers with coefficients in R'.
PROPOSITION 1. An element a of R is integral over a subring
R' of R if and only if there exists, for some positive integer v,
a set of elements WI, W2, ... , Wv of R, not all zero, such that
awi = AilWl + Ai2W2 + ... + AivWv for some v X v matrix of
elements Aij E R'.
PROOF: As in the proof of the theorem above, the equations
awi = 2:= AijWj imply, when the w's are not all zero, that a is
a root of det(X I - A), where I is the v x v identity matrix,
where A = (Aij), and where X is an indeterminate. Therefore,
since det(X I - A) = XV + P1xv-l + P2xv-2 + ... where the
Pi are in R', a is integral over R'. Conversely, if a is integral
over R', say aIL = qla/L-l + q2a/L-2 + ... + q/L with qi E R',
then, with Wi = a i - 1 for i = 1, 2, ... , J-l, awi = Wi+l for i = 1,
2, ... , J-l- 1, and aW/L = qlW/L-l + q2W/L-2 + ... + q/LWl·
COROLLARY. Let R' be a subring of a ring R. The elements
of R integral over R' are a subring of R containing R'.
DEDUCTION: An element a of R' is integral over R' by virture
of a = a . a O. If a and b are elements of R integral over R',
then there exist nonzero elements WI, W2, ... , Wv , W~, W~, ... ,
w~ such that awi = 2:=;=1 AijWj and bwi = 2:=j=l A~jwj, where
the Aij and Ai j are in R'. The J-lV elements wO"w~ (1 ::;; ()" ::;; v,
1 ::;; T ::;; J-l) satisfy awO"w~ = 2:=;=1 AO"jWjW~ and bwO"w~ =
10 Divisor Theory

'L.j=l A~jWuWj, from which it follows easily that both (a +

b)wuw~ and abwuw~ can be expressed as linear combinations
of the J.lV elements wuw~ with coefficients in R'. Therefore, the
elements of R integral over R' are closed under addition and
multiplication.
PROOF OF THE ALTERNATE THEOREM: G= 'L. 'L. ajbkxjYk.
Let R' be the subring of the ring of polynomials with integer
coefficients in the a's, b's, x's, and y's which is generated by
the c's, the x's, and the y's. Then the x's and y's are integral
over R'. By the theorem already proved, each ajb k is integral
over R'. Therefore, G is integral over R' by the Corollary.
Two postscripts: 1) Another basic proposition about inte-
grality will be needed.
PROPOSITION 2. Let R' be a subring of a ring R, and let R"
be tbe ring of elements of R integral over R'. Any element of
R integral over R" is in R".
PROOF: Let b R be integral over R", say bn = a1b n - 1 +
E
a2bn-2 + ... + an where the aj E R". For each i = 1, 2, ... ,
n let ar = 'L.f=l Aija~ - j where Aij E R'. (There is clearly
no loss of generality in using the same exponent N for all
i.) Let the w's be the nNn products bea~la;2 ... a~n in which
o :S e < nand 0 :S ei < N. It is easy to see that b times
anyone of these nNn elements w is a linear combination of
w's with coefficients in R'. Therefore, b is integral over R', as
was to be shown.
2) Kronecker's proof of the theorem, while less slick than the
proof just given, sheds more light on the reason the theorem is
true. Briefly, the argument Kronecker sketches* is as follows:
Let Rl = Z[ao, bo, Ul, U2, ... , u m +n ] be the ring of polyno-
mials in m + n + 2 indeterminates with these names (integer

*Kronecker's proof [Kr2] is not at all clear. I am grateful to Dr. Olaf

Neumann of Jena for explaining it to me. See [E-N-P, p. 69]
 Part O. A Theorem of Polynomial Algebra 11

coefficients). Define aI, a2, ... , am, bl , b2, ... , bn , Co, CI, ... ,
Cm+n E RI by the equations
m m

i=O i=l

n n

i=O i=l
m+n m+n
L Xm n i
Ci + - = aobo (1 II + Ui X).
i=O i=l

(In other words, ai is ao times the ith elementary symmet-

ric function in the first m of the u's, bi is bo times the ith
elementary symmetric function in the last n of the u's, and
Ci is aob o times the ith elementary symmetric function in the
u's.) Let G be L: L: ajhXjYk, a polynomial in new indetermi-
nates Xo, Xl, . .. ,X m , Yo, YI, ... , Yn with coefficients in R I . Let
p = (m + n)!. The p permutations of the u's act on RI and
therefore act on polynomials with coefficients in R I . These
permutations produce p conjugates of G, call them G I , G 2 ,
... , G p, all of them polynomials in the x's and y's with coef-
ficients in R I . Let F(X) = IIf=1 (X - G i ) where X is a new
indeterminate. Then F(G) = 0 (G is among the Gi ) and the
coefficient of X p-r in F is the rth elementary symmetric func-
tion in G I , G 2 , " ' , G p , call it Sr. The main step is to show
that Sr can be expressed as a polynomial in the x's, y's, and
c's with integer coefficients.
Let Sr be regarded as a polynomial in the x's and y's with
coefficients in R I , and let C be one of its coefficients. Since
each coefficient of G is aob o times a polynomial in the u's
which has degree 1 in anyone u, C is a~b~ times a symmetric
polynomial in the u's which has degree r, at most, in anyone
u. By the fundamental theorem on symmetric polynomials,
C is aobo times a polynomial in cdco, C2/CO, ... , cm+n/CO,
12 Divisor Theory

say C = a~b~ . p(cI/co, C2/CO, ... , cm+n/CO), where p has in-

teger coefficients. Since ao bo = co, the factor Co clears the
denominators of p(cI/co, C2/CO, ... , cm+n/co)-so that C is a
polynomial in the c's-by virtue of:
LEMMA. Let h(XI, X 2 , ••• , X,J be a polynomial of total de-
gree v with integer coefficients. Let Zl, Z2, ..• , zp. be indeter-
minates, and let 0"1, 0"2, ••• , 0" P. be the elementary symmetric
functions in the z's. Then h( 0"1, 0"2, ••• , 0" p.) has degree v in
each z.
The lemma can be proved by writing all polynomials using
lexicographic order and examining the leading terms. The
leading term of h( 0"1, 0"2, ••• , 0" p.) has degree v in Zl, so by
symmetry its degree in all z's is v.
Thus F( G) = 0 is a relation of the desired form, except that
the a's and b's it relates are contained in Rl and are not the
indeterminates one begins with. To prove the theorem, one
must show that the mapping R ~ Rl is one-to-one, so that
the relation F(G) = 0 in Rl[xo, Xl, ... , X m , Yo, Yl, ... , Yn]
implies a relation in R[xo, XI, ... , X m , Yo, YI, ... , Yn] of the
same form.
Dedekind gave two other proofs of the theorem [DI], each
containing interesting ideas.
 Part 1: The General Theory

§1.1 Introduction.
The general theory of divisors, as it is developed in this
first part, applies to algebraic extension fields of rings that are
natural in the sense defined in the next article. The case of
number theory in Part 2 is the case of the natural ring Z, and
the case of algebraic curves in Part 3 is the case of the natural
ring Q[x).
The objective of the theory is to define divisors and divisibil-
ity of elements by divisors in such a way that, given a natural
ring r and given a (finite) set of elements, all of which are alge-
braic over r, there is a divisor which is their greatest common
divisor. Divisors form a multiplicative group in which there
are integral elements (a divisor is integral if it is the greatest
common divisor of a set of elements all of which are integral
over the ring) and in which there is consequently a notion of
divisibility (A divides B if BfA is integral). To say that a di-
visor A is the greatest common divisor of the set of elements
ltl, lt2, ... , ltv algebraic over r means simply that A divides
lti for all i and is divisible by any divisor which divides lti for
all i.
The core of the theory is presented, after a few preliminaries,
in §1.5-§1.12. The remainder of Part 1 is devoted to various
developments of the theory that can be carried out in the case
of a general natural ring.

§1.2 Natural rings.

The theory of divisors applies to quantities algebraic over
what I will call "natural" rings.

DEFINITION. A ring r is natural if

(1) r is a commutative integral domain with multiplicative
identity l.
(2) r is a g. c. d. domain. That is, given any two nonzero
14 Divisor Theory

elements a, b E r, one can find a third element c E r which

divides both a and b and is divisible by any d E r which divides
both a and b.
(3) Factorizations of nonzero elements of r contain a finite
number of factors. More precisely, given a E r, a =f 0, one can
find a positive integer n such that any factorization of a into
n factors in r contains at least one factor which is a unit. (An
element of r is a unit if it divides 1, or, what is the same, if it
divides all elements of r.)
(4) There is a factorization method for polynomials in one
variable with coefficients in r. That is, given a polynomial
f(x) = aoxn + alx n - l + ... + an in one indeterminate x with
coefficients ai in r, a finite calculation produces either a fac-
torization f = gh of f into two factors 9 and h, neither of
which is a unit, or a proof that there is no such factorization.
(A polynomial 9 in x with coefficients in r is a unit if, as a
polynomial, it divides 1, which is the case if and only if it is a
constant polynomial whose only term is a unit of r.)
Kronecker applied his theory to rings of the form
r = Z[XI, X2, ••• , x n ] of polynomials in some number of
indeterminates-possibly none-with coefficients in the ring
of integers Z. It is elementary to prove that these rings, which
Kronecker called "natural", are natural in the sense just de-
fined (see, for example, [E3]). In addition, rings of polynomials
r = Q[XI, X2, ••• ,x n ] with coefficients in the rational field Q
or rings of polynomials r = F q [Xl, X2, ••• , X n] with coefficients
in a finite field F q are natural rings.
§1.3 On existence.
In the definition of §1.2, the phrase "one can find" has been
used instead of "there exists" to emphasize the algorithmic
foundation of the theory. However, the entire theory is con-
structed in this book according to Kroneckerian principles,
which means that "one can find" and "there exists" are to
be regarded as synonymous. For example, because Theorem 2
 §1.4 Preliminaries 15

(§1.20) asserts that there exists a divisor with certain prop-

erties, the proof must provide an algorithm (with a simple IL
priori bound on the length of the computation) for constructing
such a divisor. This meaning of "there exists" is the only one
Kronecker recognized. Accordingly, the theorems of the theory
say a great deal more than they appear to say to a reader who
regards "there exists" as meaning "if one could prove there did
not exist, then one could derive a contradiction."
§1.4 Preliminaries.
The development of the theory of divisors will require the
following elementary consequences of the definition of natural
rmgs:
Any finite set aI, a2, . .. , all of nonzero elements of a natural
ring r, has a greatest common divisor c; that is, there is a
c E r which divides all the ai and is divisible by any d E r
which divides all the ai. This is true for v = 2 by axiom (2).
Suppose it is true for v -1, and let b be a g. c. d. of aI, a2, ... ,
all-I. A greatest common divisor c of b and all divides all the
a's, and any common divisor d of all the a's divides both band
all and therefore divides c. Therefore, c is a g. c. d. of the a's,
and the proposition follows by induction.
If a, b, and c are elements of r such that clab but c t a and
c t b, then c = CI C2, where neither CI nor C2 is a unit. To prove
this, let d be a g. c. d. of ab and cb. Then bid, say d = bCI, and
the quotient CI divides both a and c. Set C2 = c/ CI. Since
c divides both ab and cb, cld, that is, CI c2lbcl. Therefore,
c2lb. If cll1 then c = CIC21c2 gives clb, contrary to assumption.
If c211 then CICI, which implies cia, contrary to assumption.
Therefore, neither CI nor C2 is a unit.
Let f and 9 be polynomials with coefficients in r, and let a
and b be greatest common- divisors of the coefficients of 1 and
9 respectively. Then ab is a greatest common divisor of the
coefficients of 19. The coefficients of 19 have a g. c. d., call it c.
Since 19 = ab(f / a )(g / b) and since (f / a ) (g / b) has coefficients
16 Divisor Theory

in r, able, say e = abd. Assume d is not a unit. Let an order

be chosen for the indeterminates in 1 and g, and let the terms
of polynomials be ordered lexicographically, that is, with one
term preceding another if it contains the first indeterminate to
a higher power, or, if they contain the first indeterminate to the
same power, it contains the second indeterminate to a higher
power, etc. Since ad does not divide a, at least one coefficient
of 1 is not divisible by ad. Therefore, 1 can be written in the
form 1 = adII + ah, where the coefficients of II and 12 are in
r, where h =I- 0, and where no coefficient of 12 is divisible by d.
Similarly, let 9 = bdg! + bg2, where g2 i= 0 and no coefficient of
g2 is divisible by d. Since Ig = abd( dfIgl +hgl +fIg2)+abhg2,
and all coefficients of 19 are divisible by abd, all coefficients of
hg2 are divisible by d. The leading term of hg2 is the product
of the leading terms of hand g2' Therefore, the coefficient of
the leading term of hg2 is on the one hand divisible by d
and on the other is a product of two elements of r, neither of
which is divisible by d. As was shown above, it follows that
d = d 1d2, where neither d 1 nor d2 is a unit. But then d 1
and d 2 have the same properties as d: they are not units, and
all coefficients of fg are divisible by abdi . Therefore, by the
same argument, d 1 = d 11 d l2 and d 2 = d21 d 22 , where dij is
not a unit and abdij divides all coefficients of f g. Therefore,
d = d 1d2 = d11d12d21d22 = ... can be written as a product
of arbitrarily many factors, none of them units, contrary to
axiom (3). Therefore, d must be a unit, say 1 = de, which
implies clce = abde = ab and shows that ab is a g. c. d. of the
coefficients of f g.
A polynomial f (in any number of indeterminates) with co-
efficients in a natural ring r is said to be primitive if it is not 0
and if 1 is a greatest common divisor of its coefficients. (Here,
and always, the "coefficients" of f are the coefficients after
like terms have been combined. For example, 2x + 3x is not a
primitive polynomial with coefficients in Z, but 2x + 3x 2 and
2x + 3y are.)
 §1.4 Preliminaries 17

By the above theorem, a product of primitive polynomials is

primitive.
Let f be a nonzero polynomial with coefficients in r, and let
c be a greatest common divisor of the coefficients of I. Then
f / c is primitive. Because, if d is a g. c. d. of the coefficients of
I / c, the above theorem shows cd is a g. c. d. of the coefficients
of I; therefore, cdlc, which implies d11, so any common divisor
of the coefficients of I / c divides l.
If 1 is a greatest common divisor of ~l, a.2, ... , a v , then it
is also a greatest common divisor of aL a~, ... , at for any
integer j 2: 1. .To prove this, it will suffice to prove that 1
is a g. c. d. of ai, a2, a3, ... , a v , because then each a can be
raised to the power j in turn. Let 7r = 2: aiui where the Ui
are indeterminates. Since 7r is primitive by: assumption, 7r j is
primitive. Since the coefficients of 7r j are ai and elements of r
divisible by at least one of a2, a3, ... , a v , any common divisor
of a1, a2, a3, ... , a v divides all coefficients of 7r j and therefore
must be a unit, as was to be shown.
GAUSS'S LEMMA. Let k be the field of quotients of r. If the
product of two monic polynomials in one indeterminate with
coefficients in k has coefficients in r, then both factors must
have coefficients in r.
Otherwise stated, if (aoxm + alx m- 1 + ... + am)(box n +
b1x n - 1 + ... + bn ), where the a's and b's are in r and x is an
indeterminate, has all its coefficients divisible by ao bo , then ao
divides all the a's and bo divides all the b's. This follows, of
course, from the theorem of Part 0 (an element of k is integral
over r if and only if it is in r). It can also be deduced from the
more elementary theorem above as follows.
If c and dare g. c. d. 's of the a's and b's respectively,
then cd is a g. c. d. of the coefficients of the product
(2: ai Xn - i )(2: bix n - i ). By assumption, aob o divides all these
coefficients. Thus aobo divides cd, say cd = eaobo. Since dlb i
for all i, and ciao, cdlaobi, eaobolaobi , ebolb i , bolk In the same
18 Divisor Theory

way, aolai, as was to be shown.

Gauss's lemma for polynomials in several indeterminates can
be deduced from the case of one indeterminate using the tech-
nique of the deduction of Corollary 3, Part O. (In the notation
there, let XII be the indeterminate in which the polynomials
f and g are monic, so that ljJ(f) and ljJ(g) are monic in T.)
Thus, if the product of two polynomials in several indetermi-
nates with coefficients in k has coefficients in r, and if the
polynomials are monic in one of the indeterminates, then both
factors must have coefficients in r.

§1.5 The basic theory.

The main objective of the theory is to define a greatest com-
mon divisor of a finite set of elements al, a2, ... ,all, each of
which is algebraic over r. An algebraic extension of r is a
field containing r, all of whose elements are algebraic over r.
It is natural to assume that the a's are contained in a fixed
algebraic extension K of r.
The notion of greatest common divisor must be preceded by
a notion of divisibility, which must in turn be preceded by a
notion of integrality: an element divides another if the quotient
is integral. An element a algebraic over a natural ring r is said
to be integral over r if some power of it can be written as a
linear combination of lower powers with coefficients in r. As
can easily be shown (see Part 0), the elements of K integral
over r form a ring in K.
The greatest common divisor of a finite subset al, a2, ... ,
all of K will be defined by defining a divisor which depends
on more than the finite subset-specifically, it will depend on
a polynomial whose coefficients are the a's-and then show-
ing that the divisor so defined has all the desired properties.
Among the desired properties is that the divisor represented
by a polynomial with coefficients in K depends only on the
coefficients of the polynomial. (See Corollary (11) of §1.12.)
 §1.7 What is a Divisor? 19

§1.6 Definitions.
Let K be an algebraic extension of r. A polynomial f, in
any number of indeterminates and with coefficients in K, is
said to divide another such polynomial 9 if 9 = f q, where q is
a polynomial whose coefficients are elements of K integral over
r.
We will say that a polynomial f (in any number of indeter-
minates and with coefficients in K) represents a divisor, and
will denote the divisor it represents by [f]. If f and 9 are such
polynomials, we will say that the divisor represented by f di-
vides g-in symbols [fllg-if there is a primitive polynomial7r
with coefficients in r such that f divides g7r. In other words,
[J]lg means that g7r = fq, where 7r is a primitive polynomial
with coefficients in rand q is a polynomial with coefficients
integral over r.
The divisor represented by ft will be said to divide the di-
visor represented by h-in symbols, [ft]l[h]-if [12] Ig implies
[ftllg·
Two polynomials fl and 12 with coefficients in K will be
said to represent the same diviso~in symbols, [ftl = [hl-if
[ftll[hl and [hJl[ftl·
§1.7.
The contemporary style of mathematics trains mathemati-
cians to ask "What is a divisor?" and to want the answer to
be framed in terms of set theory. Those trained in this tradi-
tion will want to think of a divisor as an equivalence class of
polynomials, when equivalence of polynomials is the property
of representing the same divisor. I believe, however, that in-
stead of asking what a divisor is one should ask what it does.
It divides things. Specifically, it divides (or does not divide)
polynomials with coefficients in K. The definition of what a
divisor does involves a given polynomial, and two polynomials
represent the same divisor if the corresponding divisors do the
20 Divisor Theory

same thing.
§1.8 First Propositions.
PROPOSITIONS. (1) If [fllgl and [fllg2, then [fll(gl ± g2).
(2) If [1]1,8 for all coefficients 13 of g, then [fllg.
(3) If [llig and [g]lh, then [fll h.
(4) [f] Ig if and only if [f]l[g 1·
(5) If f has coefficients in r, if f =1= 0, and if d is a greatest
common divisor of the coefficients of f, then [fl = [d].
(6) If [fdl[h]' then [flhll[hh] for all polynomials h with
coefficients in K.
PROOFS: (1) If fql = gl 7rland fq2 = g2 7r2, then f(ql 7r2 ±
q27rJ) = (gl ± g2)7r 1 7r2. The coefficients of ql7r2 ± q27rl are
integral over r because the coefficients of ql, q2, 7rl, 7r2 are, and
because the elements of K integral over r form a ring. The
product 7rl7r2 of two primitive polynomials with coefficients
in r is itself primitive, as was proved above. Therefore, (1)
follows. (2) [fl LB implies fq = ,87r for 7r primitive and for q with
coefficients integral over r. If this equation is multiplied by any
monomial with coefficient 1, it shows that any monomial with
coefficient ,8 is divisible by [fl. Therefore, (2) follows from (1).
(3) If [lllg, say fql = g7rl, and if [gllh, say gq2 = h7r2, then
fqlq2 = g7rlq2 = h7rI7r2, which shows [fllh. (4) Since [gllg (set
q = 7r = 1), [l11[gl implies [f1Ig. Conversely, if [f1lg and if
[gllh then [l11h by (3); thus [f11[g]. (5) f jd is primitive (§1.4).
Therefore, [llid (q = 1, 7r = f jd) and [d]lf (q = f jd, 7r = 1),
so [f] = [d] by (4). (6) If hq = h7r, then hhq = hh7r for
all h, and (6) follows.
§1.9.
PROPOSITION (7). Ifa,f3 E K, and a =1= 0, then [a]l,8 if and
only if ,8 j a is integral over r.
PROO F: If 13 j a is integral over r, then a . (13 j a) = ,8 . 1 shows
that [a] 113. Conversely, suppose [a]lf3, that is, aq = j37r. With
 §1.9 First Propositions 21

'Y = f3 / G, this is simply the statement that there is a primitive

polynomial 7r such that 'Y7r has coefficients integral over r. It
is to be shown that 'Y is then integral over r.
Since 'Y is algebraic over r, h( 'Y) = 0 for some nonzero poly-
nomial h( x) in one indeterminate with coefficients in r. Di-
vision of h( x) by a g. c. d. of its coefficients gives a primitive
h( x) for which h( 'Y) = o. Therefore, one can assume with-
out loss of generality that h( x) is primitive. IT h( x) is not
irreducible, then, by axiom (4), h(x) = hl(X)h2(X), where hI
and h2 have coefficients in r. Since h is primitive, neither
hI nor h2 has degree 0, so both have degree less than deg h.
Since either h l ({) = 0 or h2({) = 0, and both hI and h2 are
primitive, it follows that, if h is not irreducible, there is a poly-
nomial of smaller degree than h with the same properties as h.
Therefore, by the principle of infinite descent, one can assume
without loss of generality that h( x) is a polynomial irreducible
over r for which h({) = 0, say h(x) = boxn+bIXn-1 + .. ·+bn.
Suppose a E r has the property that a'Y is integral over r,
say <fJ( a'Y) = 0 where <fJ is monic with coefficients in r. The Eu-
clidean algorithm applied to <fJ( ax) and h( x), regarded as poly-
nomials with coefficients in the field of quotients k of r, gives
a polynomial d( x) of the form d( x) = s( x )<fJ( ax) + t( x )h( x)
which divides both <fJ( ax) and h( x) as polynomials with coeffi-
cients in k. Clearing denominators then gives equations of the
form cld(x) = SI(x)<fJ(ax) + tl(x)h(x), c2<fJ(ax) = ql(x)d(x),
and C3h(X) = q2(x)d(x), where CI, C2, and C3 are nonzero el-
ements of r and d(x), SI(X), tl(x), ql(X), and q2(X) have co-
efficients in r. There is no loss of generality in assuming that
d(x) is primitive (any common factor of its coefficients could
be absorbed into CI, ql(X), and Q2(X)). Then, since h(x) is
primitive (it is irreducible), C3 divides all coefficients of q2(X)
(1 times a g. c. d. of the coefficients of q2 (x) is a g. c. d. of the
coefficients of C3h(X)). Therefore, h(x) = Q3(x)d(x), where
q3 (x) has coefficients in r. Since h( x) is irreducible over r,
either q3 (x) or d( x) must be a unit of r. Since CI d( 'Y) =
22 Divisor Theory

81 (,,)4>(a,,) + t1(,,)h(,,) = 81(,,)·0 + tl(")· 0 = 0, dCf') = 0, so

degd> o. Therefore g3(X) is a unit of r, say q3(X) = e E r.
Then hex) = ed(x), eC24>(ax) = ql(x)h(x). Since eC2 divides
all coefficients of ql (x )h( x) and h( x) is primitive, eC2 divides
all coefficients of ql (x), and 4>( ax) = q4( x )h( x), where q4( x)
has coefficients in r. Thus 4>(y) = q4(y/a)h(y/a). Let m be the
degree of q4. Then am +n4>(y) = (a m Q4(y/a))(a nh(y/a)). The
factors on the right have coefficients in r, and their product
am +n4>(y) has all its coefficients divisible by its leading coeffi-
cient a m +n . Therefore, by Gauss's Lemma, all coefficients of
anh(y/a) = boyn + ab 1y n-l + a2b2y n-2 + ... + anbn are di-
visible by its leading coefficient bo. Therefore, bo divides all
coefficients of an h( x). Since h( x) is primitive, bo divides an.
Thus, if "a is integral over r for all coefficients a of 71", bo
divides an for all coefficients a of 71". Since 1 is a g. c. d. of the
a's, it is a g. c. d. of their nth powers (§1.4), and bo divides 1,
say boeo = 1. Therefore" is a root of the monic polynomial
eo h( x) with coefficients in r, which means" is integral over r,
as was to be shown.
§1.10.
LEMMA (8). If f is a nonzero polynomial with coefficients in
K, then [f]l[c] for some nonzero c E r.
PROOF: Adjunction of the coefficients of f to the field of quo-
tients k of r gives a subfield K' of K of finite degree over k.
Let WI, W2, ••• , Wp span K' over k. Each fWi for i = 1, 2,
..• , p, can then be written in the form Ej=1 mijWj, where
the mij are polynomials in the same indeterminates as f with
coefficients in k. Let X be a new indeterminate, and let
F(X) = det(XI - M), where I is the p, x p, identity matrix
and M is the p, x p, matrix whose entries are the mij. Then
F(X) = XP+F1 Xp-l + .. ·+Fp, where the Fi are polynomials
in the same indeterminates as f with coefficients in k. Substi-
tution of f for X in X I - M gives a p, X P, matrix f I - M with
entries from the ring of polynomials in the indeterminates of
 §1.11 The Main Proposition 23

f with coefficients in K. In this ring, the f-l x f-l system of

homogeneous linear equations (J I - M)Y = 0 has the non-
trivial solution Y = (w), where (w) denotes the column matrix
whose entries are WI, W2, ... ,ww Therefore, det(J1- M) = 0,
that is, fll + FIf ll - 1 + ... + FIl = o. Thus, FIl = fg,
where FIl is a polynomial with coefficients in k and g =
- f ll - I - F I f ll - 2 - ... - F Il - I is a polynomial with coefficients
in K. Each coefficient f3 of g satisfies an equation of the form
bof3n+blf3n-1 +.. ·+b n = 0, where the bi E r; such an equation
implies that bof3 is integral over r, because multiplication by
b~-I gives (b of3)n + bi (b o f3)n-1 + bob2(b o f3t- 2 + ... + b~-l bn =
o. Thus, for each coefficient f3 of either g or F Il , there is a
bE r such that bf3 is integral over r. Multiplication of FIl = gf
by the product of all these b's gives an equation of the form
h = qf, where h has coefficients in rand q has coefficients in
K integral over r. Thus, [f] divides [h]. Since [h] = [c] for
c E r by (5), the Lemma follows.
§1.11 The Main Proposition.

PROPOSITION (9). [ilia for all coefficients a of f.

This is the final proposition in the basic theory of divisors.
The other propositions have all been deduced from elementary
properties of natural rings and algebraic extensions of them.
This last one depends, however, on a less elementary theo-
rem. It can be deduced, for example, from Corollary 4 of the
theorem of Part 0:
PROOF: If f = 0, there is nothing to prove. Otherwise, by (8),
fq = C1T for some nonzero c E r and for some primitive polyno-
mial 1T with coefficients in r. Thus, the product of the polyno-
mials f and q/ c has coefficients in r. If a is any coefficient of
f and, any coefficient of q, then, by Corollary 4, a, / c is inte-
gral over r. Set ql = aq/c. Then fql = faq/c = aC1T/c = a1T,
where ql has coefficients integral over rand 1T is primitive.
Therefore, [f]ia.
24 Divisor Theory

§1.12 Concluding corollaries.

COROLLARIES. (10) [f] is the greatest common divisor of the
coefficients a of 1 in the sense that [f] la for all a and [I] is
divisible by any divisor [g] which divides all coefficients a of I.
(11) If 1 and 9 have the same coefficients, then [I] = [g].
(12) Given polynomials 1 and 9 with coefficients in K, one
can test whether [J] Ig as follows. If 1 = 0, then [I] Ig if and
only if 9 = o. Otherwise, the construction of Lemma (8) gives
an equation Iq = C7r, where c E r, c =I 0, 7r is primitive
(coefficients in r) and q has coefficients in K integral over r.
Then [1]lg if and only if the coefficients of gq/c are integral
over r.
(13) [l]l[g] if and only if the coefficients of 9 are integral over
r.
DEDUCTIONS: (10) follows from (9), (2), and (4). (11) follows
from (10). If, in (12), all coefficients of ql = gq/c are integral
over r, then the equation 1 ql = 1 qg / c = C7rg / C = g7r shows
[I] Ig directly from the definition. Conversely, if [I] Ig, then,
by (4) and (6), [lq]l[gq]. But [Iq] = [C7r] = [c] by (5). Thus
[c]l[gq]. By (9), it follows that [c] divides all coefficients of gq,
hence, by (7), that all coefficients of gq/c are integral over r.
Finally, (13) is the case 1 = c = 7r = q = 1 of (12).
§1.13 Testing for divisibility.
Application of the divisibility test (12) requires that one be
able to determine, for a given a E K, whether a is integral
over r. As was shown in §1.9, there is an irreducible hex) with
coefficients in r such that h( a) = 0; if aa is integral over r,
then the leading coefficient of h divides an, where n = deg h.
In particular, if a itself is integral over r, then the leading
coefficient of h is a unit of r. Thus, to determine whether
a is integral over r, it suffices to find a relation of the form
h( a) = 0, where h( x) is irreducible over r; a is integral over r
if and only if the leading coefficient of h is a unit of r.
 §1.15 Greatest Common Divisors 25

§1.14 The Group of Nonzero Divisors.

Proposition (6) of §1.8 shows that [!I] = [12] implies [fIg] =
[hg] for all polynomials 9 with coefficients in K. Therefore,
it is valid to define the product of the divisors [f] and [g] to
be [fg]. The product operation defined in this way is obvi-
ously commutative and associative. Moreover, [1] is an iden-
tity element with respect to it. Finally, if [f] 1= [0], then,
by Lemma (8), fq = e7r for some e E r, for some primi-
tive polynomial 7r with coefficients in r, and for some poly-
nomial q with coefficients in K integral over r, which shows
that [f][qje] = [7r] = [1] (by (5)); in particular, every divisor
[I] 1= [0] has a multiplicative inverse. In short, nonzero divi-
sors with the operation of multiplication just defined form a
commutative group.
An integral divisor [g] is a divisor represented by a polyno-
mial g, all of whose coefficients are integral over r. By Corol-
lary (13), [g] is integral if and only if [l]1[g]; therefore, if f is
any polynomial which represents an integral divisor, all coeffi-
cients of f are integral over r. Because elements of K integral
over r form a ring, a product of integral divisors is integral.
Moreover, since it was shown above that every nonzero [f] has
a multiplicative inverse [qje] = [q]j[e], every nonzero [f] can
be written in the form [q]-l[e] in the group of nonzero divi-
sors. Thus every nonzero divisor can be written as a quotient
of integral divisors in the group of nonzero divisors.
Finally, if [f] and [g] are nonzero divisors, then [I]I[g] in
the sense defined in §1.6 if and only if [1]-1 [g] is an inte-
gral divisor. For, if g7r = fq, then [1]-1 [g] = [f]-l [g7r] =
[1]-1 [fq] = [f]-l [f][q] = [q], which is integral, and, conversely,
if [f]-l [g] = [q] is integral, then [g] = [fq] is divisible by [fl
because [I] I[f q] follows directly from the definition.
§1.15 Greatest common divisors.
PROPOSITION (14). Any finite subset [!I], [12]' ... , [fll] of the
group of nonzero divisors has a greatest common divisor [g].
26 Divisor Theory

Thatis, there is a nonzero divisor [g] such that [gJl[fi] for i = 1,

2, ... , 1/, and such that if [h] satisfies [hJl[/i] for i = 1,2, ... ,1/
then [hJl[g]. Of course, by the vezy definition, any two greatest
common divisors of [!I], [12],· .. , [fll] are equal.
PROOF: By Corollary (11), [fi] is not changed if all indetermi-
nates in the polynomial fi are changed. Therefore, there is no
loss of generality in assuming that no indeterminate of fi oc-
curs in any other !;. Then the coefficients of !I +12+···+ fll are
the union of all coefficients of the /i. Let 9 = !I + 12 + ... + f II·
Then [gJl[fi] for all i, and [hJl[/i] for all i implies [hJl[g], both
by Corollary (10). Therefore [g] is the desired g. c. d.
Notation: The greatest common divisor of [!I], [12], ... , [fll]
will be denoted [!I, 12, ... , f II]. In particular, if 0'1, 0'2, ••• ,
0'1' are the coefficients of f, then [f] = [at, 0'2, ••• , 0'1'].

PROPOSITIONS. (15) [f, g, h] is the greatest common divisor of

[f] and [g, h].
(16) [f][g, h] = [fg, fh].
(17) If [f,g] = [1] and [h] is integral, then [f,gh] = [f,h].
(18) If [f,g] = [f,h] = [1], then [f,gh] = [1].
(19) If [f, g] = [1] and [h] is integral, then [f, h][g, h] =
[fg, h].
PROOFS: (15) A divisor which divides [f], [g], and [h] di-
vides [f] and [g, h], and conversely. (16) As in the proof of
(14), change the indeterminates of f, g, and h-which does
not change any of the divisors [f], [g, h], [fg] , [Jh]-so that
no indeterminate occurs in more than one of them. Then
[g + h] = [g, h] and [fg, fh] = [fg + fh] = [f][g + h] = [f][g, h].
(17) By (16), if [J, g] = [1], then [fh, gh] = [1][h] = [h]. Then
[f,h] = [f,fh,gh] by (15), and, since [f] = [f][l] = [f][l,h] =
[I, fh], it follows that [f, fh, gh] = [f, gh], also by (15). (18)
is a special case of (17). (19) Let f, g, and h involve dis-
tinct indeterminates. Then [f, h][g, h] = [f + h][g + h] =
[fg + fh + hg + h2 ] = [fg, fh, gh, h2 ]. Since [f, g, h) di-
 §1.16 Ambient Fields 27

vides [j, g] = [1] and f and 9 are integral, [j, g, h] = [1], [jh,
gh, h2 ] = [h] by (16) and [f, h][g, h] = [fg, h] by (15).
§1.16 Ambient fields.
Up to this point, the field K has remained entirely in the
background. Kronecker deliberately structured his theory (see
the beginning of §17 of [Krl]) in such a way that K plays no
role. Division of polynomials is a rational operation, in that
if F = G H, where F, G, and H are polynomials, in any num-
ber of indeterminates, with coefficients in a field K, then the
coefficients of H are determined as rational functions of the co-
efficients of F and G. (For polynomials in one indeterminate,
this follows from the division algorithm. The proof for poly-
nomials in several indeterminates is then a simple induction.)
Therefore, if f and 9 are polynomials with coefficients in K
and if 7r is a primitive polynomial with coefficients in r such
that flg7r-that is, g7r = fq, where q has coefficients integral
over r-then the same is true (with the same 7r and q) when
K is replaced by any larger or smaller field K', provided K'
contains all coefficients of f and g. Therefore, any statement
about divisors represented by polynomials with coefficients in
K is true for divisors represented by polynomials with coeffi-
cients in any algebraic extension K' of r such that the divisors
involved can all be represented by polynomials with coefficients
in K'.
The above formulation of the theory is even more inde-
pendent of the ambient field than was Kronecker's, because
Kronecker used norms in defining the basic concept [f] Ig.
(Specifically, for a linear polynomial f with coefficients in-
tegral over r, he sets [Krl, §14] N f = P . Fm(f) , where
PEr and where Fm(f) is a primitive polynomial with co-
efficients in r, and defines [f] Ig to mean that (f I Fm(f)) Ig;
since f I Fm(f) = P fiN f, [f] Ig then means that P divides
g. (Nfl!), which is equivalent to the meaning given to [f]lg
in §1. 6, as Corollary (12) shows.)
28 Divisor Theory

Norms are defined in the next section in order to have them

available for the proofs of Theorems 1 and 2 below. The norm
of a polynomial with coefficients in a field K depends on the
ambient field (albeit in a simple way-passing to a larger field
raises the norm to a power), but Theorems 1 and 2 are inde-
pendent of the ambient field that is used, provided only that
it contains all the divisors in question.
§1.17 Norms.
An algebraic extension K of a natural ring r will be said to
be of finite degree if its dimension as a vector space over the
field of quotients k of r is finite.

DEFINITION. Let I be a polynomial with coefficients in an

algebraic extension K of r of finite degree, and let aI, a2, ... ,
a A be a basis of Kover k, the field of quotients of r. Then
the equations I . ai = E~=l <Pijaj define polynomials <Pij in
the same indeterminates as I with coefficients in k. The norm
of I relative to the extension K ::> r, denoted N K I, is the
determinant of the .A x .A matrix whose entries are <Pij. This
definition of N K I is independent of the choice of the basis of
Kover k, because any other basis is transformed to the chosen
one by multiplication by an invertible matrix M with entries
in k, which changes the matrix with entries <Pij by multiplying
it on one side by M and on the other by M- 1 , an operation
which leaves its determinant unchanged. Note that N K I is a
polynomial, in the same indeterminates as I, with coefficients
in k. When the extension K ::> r under consideration does not
need emphasis, N K I will be denoted simply N I.
PROPOSITION. If the coefficients of I are integral over r, then
NI = I q, where q can be expressed in the form

with v a positive integer and Bl, B 2 , ••• , BII polynomials in

 §1.17 Norms 29

the same indeterminates as f with coefficients in r. In partic-

ular, q has coefficients integral over r.
PROOF: There exist elements WI, W2, ••• , WI' of K such that
any coefficient of f times any W is a linear combination of
w's with coefficients in r (Proposition 1 of Part 0). Thus,
fWi = E mijw j, where the mij are polynomials with coef-
ficients in r, and the argument of §1.10 gives a polynomial
F(X) = XI' + FIXI'-I + ... + FI" with coefficients in r,
for which F(f) = O. By the Remainder Theorem, F(X) =
(X - 1)Q(X), where Q(X) is a polynomial with coefficients in
K. The norm of F(X) is the determinant of F(X)·I, where I is
the .A X.A identity matrix, which is F(X)'x. On the other hand,
the norm of a product is the product of the norms (because
the determinant of a product is the product of the determi-
nants), so F(X), = N(X - 1)NQ(X). By Gauss's Lemma,
since F(X), is monic with coefficients in rand N(X - 1) is
monic with coefficients in k, N(X - 1) has coefficients in r, say
N(X - 1) = X),+AIX),-I +.. ·+A)" where the Ai are polyno-
mials in the indeterminates of f with coefficients in r. Substi-
tution of X = 0 in N(X - 1) gives (-l),Nf = A),. Substitu-
tion of X = f in the matrix of which N(X - 1) is the determi-
nant gives a matrix M which satisfies M a = 0, where a is the
column matrix giving the chosen basis of Kover k, and where
ois the column matrix of .A zeros. Thus, det M = 0, which is to
say, f), +A 1f),-1 + ... + A), = 0, that is, N f = (-1), A), = fq,
where q = (-1),-1(1),-1 + A 1 f),-2 + ... + A),_I), as was to
be shown.
COROLLARIES. (1) If [1] is integral, then [f]I[N fl. In partic-
ular, [N f] is integral.
(2) If [f]l[g], then [Nf]I[Ng].
(3) [N f] depends only on [f] and can therefore be defined
to be N[j], the norm of the divisor.
(4) If N[f] = [0], then [f] = [0].
(5) Given a nonzero integral divisor [1] in K, there is an
30 Divisor Theory

integer v such that any representation of [f] as a product of

more than v integral divisors in K contains at least one factor
equal to [1].
DEDUCTIONS: (1) is immediate. (2) IT fq = g7r, then N f .
Nq = Ng·7r A • Since [Nq] is integral by (1) and 7r A is primitive,
[Nf]I[Ng]. (3) follows immediately from (2). (4) IT [f] =I-
[0], then [f] = [g], where 9 is a polynomial with a nonzero
constant term, call it go. Directly from the definition, the
constant term of N 9 is the determinant of the matrix which
represents multiplication by go relative to a basis of Kover k.
Since multiplication by go is invertible, so is the matrix which
represents it; therefore, the constant term of N 9 is nonzero,
and N[f] = [N g] =I- [0], as was to be shown. (5) The norm of a
factorization of [f] is a factorization of [N fl. Since [N J] and
all its factors obtained in this way are of the form [c] for c E r,
Axiom (3) of the definition of a natural ring implies that, for
v sufficiently large, at least one of the factors must divide [1].
By (1), an integral divisor whose norm divides [1] is [1].
§1.18 Factorization of divisors.
Let K be a fixed algebraic extension of the natural ring r.
An integral divisor in K is reducible in K if it can be written
as a product of two integral divisors in K, neither of which
is [1]; otherwise it is irreducible in K. An integral divisor in
K is prime in K if it is not [1] and if it divides a product
of integral divisors only when it divides one of the factors;
otherwise stated, an integral divisor [f] in K is prime if the
integral divisors in K not divisible by [f] form a nonempty set
cloRed llnc.er multiplication.
PROPOSITION. An integral divisor other than [1] is irreducible
in K if and only if it is prime in K.
PROOF: IT [f] is prime in K and if [f] = [g][h] , where [g] and [h]
are integral divisors in K, then [fJl[g)[h), which implies [f]l[g]
 §1.18 Factorization of Divisors 31

or [f]l[h]. If [f]l[g], say [g] = [f][q], where [q] is integral, then

[f] = [g][h] = [f][q][h], [1] = [q][h], [h]l[l] , [h] = [1]. Similarly,
if [f]l[h], then [g] = [1], so [g] or [h] is [1], as was to be shown.
Suppose now that [f] =f [II,is an integral divisor irreducible
in K, and suppose that [f]l[g][h], where [g] and [h] are divisors
in K. If [f, g] = [1] then [f] = [f, gh] = [f, h] by Propo-
sition (17) of §1.15, that is, [f]l[h]. If [f,g] =f [1], then, since
[J] = [f, g][q] for some integral divisor [q], and [J] is irreducible,
[q] must be [1], [f] must be [f, g], and [f] must divide [g]. Thus,
[f] is prime.
To factor an integral divisor in K into divisors irreducible
in K is therefore the same as to factor it into divisors prime
in K. When K ::> r is an extension of finite degree, Corollary
(5) of §1.17 guarantees, in a weak sense, that every integral
divisor in K can be expressed as a product of divisors prime
in K: Given any expression of [f] =f [1] as a product of inte-
gral divisors [f] = [gl][g2] ... [gil]' none of which is equal to [1],
either the factors [gil are all prime or at least one of them is
not prime, that is, at least one of them is not irreducible, and
the factorization can be carried further. The corollary gives
an upper bound on the number of factors in such a factor-
ization, and shows that this factorization process-beginning
with the trivial factorization [f] = [f]-must terminate with
an expression of [J] as a product of divisors prime in K.
These prime factorizations of integral divisors in K "exist"
in a very weak sense, however, because the process is not con-
structive. It lacks an algorithm which, for a given integral
divisor [f] in K, either expresses [f] as a product of two in-
tegral divisors in K, neither of them equal to [1], or proves
that [f] is irreducible in K. Such an algorithm is given in §2.1
in the special case r = Z. Kronecker very briefly indicates an
algorithm [Krl, §18, end of the second paragraph] in the cases
r = Z[X1' X2, ... ,xn]'
The existence of such an algorithm for general r appears to
32 Divisor Theory

be a subtler question than the ones dealt with here. For this
reason, the question of factoring arbitrary integral divisors into
divisors prime (irreducible) in a given field is avoided in what
follows. Although factorization into primes has generally been
regarded since Dedekind's time as the cornerstone of the the-
ory, the lack of such factorizations does not seem to be any
impediment at all. In fact, it forces one to rethink the funda-
mentals of the theory and to base proofs on more elementary
and more constructive arguments.
§1.19 Two Basic Theorems.
THEOREM 1. Let a tinite number of divisors [h]' [12]' ... ,
[fp] be given. There exist relatively prime integral divisors
[gd, [g2], ... , [gv] such that each [Ii] is a product of powers
of [g) 's; that is, [gi, gj] = [1] whenever i =I- j, and there exist
integers (Jij such that [fi] = [gl]U i1 [g2]Ui2 ... [gvr'v for i = 1,
2, ... , p.
PROOF: The [f)'s can be expressed as quotients of integral
divisors, say [fd = (fn/(fI']. Since a set of [g)'s for the 2p
integral divisors [fn, [fI'J will serve as a set of [g)'s for the
given [f)'s, there is no loss of generality in assuming that the
given [f)'s are integral. Consider the integral divisor which
is the product [fl][h]'" [fp] of the [f),s. If, for some pair of
indices, [fi, fj] is a proper divisor of [fi]-that is, [Ii, fj] is
neither [1] nor [fi]-then the factorization [h][h]'" (fp] can
be refined by replacing [fd with two factors [fi, fj][h], where
[h] = [Ii][fi, fjt 1 =I [1]. To find a set of [g]'s for the factors
of [f1][h]'" [fp], it will suffice to find a set of [g)'s for the
factors of the refined factorization. Let this construction be
repeated. It will produce greater and greater refinements of the
factorization [h][h]'" [fp] unless a refinement is reached in
which any two factors are either relatively prime ([fi, Ii] = [1))
or identical ([h, Ii] = [fi] = [Ii))· Let [gIl, [g2], ... , [gv] be the
distinct factors in such a factorization. Then [h ][12]' .. [fp] =
 §1.19 Two Basic Theorems 33

[gIrl [g2r2 [gv r v and the factorization on the right is a re-

•••

finement of the one on the left. Therefore, each [Ii] is a product

of powers of the [g],s, as desired, and the refinement process,
if it terminates, results in a set of [g]'s.
Let J{ :J r be the extension generated by the coefficients
of the polynomials fl, h, ... , f p.. The above construction lies
entirely within the field J{, which is an extension of r of fi-
nite degree. Therefore, by Corollary (5) of §1.17, there is an
upper bound on the number of factors other than [1] in a fac-
torization of [fIHh] ... [Jp.]. Therefore, the above process must
terminate, and Theorem 1 is proved.
PROPOSITION. Let [gl], [g2], ... , [gv] be relatively prime in-
tegral divisors. Tben fHgj]O"j divides I1[gjrj , for 0"1, 0"2, ... ,
O"v, Tl, T2, ... , Tv E Z only if O"j ::; Tj for all j.
PROOF: Because both divisors can be multiplied by high pow-
ers of I1[gj]' there is no loss of generality in assuming that the
O"'s and T'S are all positive. IT O"i > Ti and I1[gj]O"j divides
I1[gjrj , then [gil divides [gi]O";-T; I1#i[gj]O"j which in turn di-
vides IL#i[gjp, contrary to Proposition (18), §1.15.
COROLLARIES. (1) Given two factorizations of an integral di-
visor into integral divisors, tbere is a tbird factorization into
integral divisors wbicb is a refinement of both.
(2) Two representations of a divisor as a product of powers
of distinct divisors prime in a field J{ :J r are identical, except
possibly for tbe order of tbe factors.
(3) An integral divisor in J{ wbicb divides a product of di-
visors prime in J{ is equal to tbe product of a subset of tbose
divisors.
DEDUCTIONS: (1) Let [fIHh]··· [Jp.] = [h 1 Hh 2 ]··· [h m ] be the
given factorizations. By the theorem, there exist [gl], [g2], ... ,
[gv] such that each [Jd and each [hi] is a product of powers of
[g],s. The Proposition implies that these powers are all non-
negative. Thus, the product of the [f]'s is a product of powers
34 Divisor Theory

of [g]'s, and this product of powers of the [g],s is a refinement

of both of the given factorizations. (2) By the theorem, there
exist relatively prime integral divisors [gl], [g2], ... , [g,,] such
that the factors of both factorizations are products of powers of
the [g]'s. But since these factors, call them [Ii], are irreducible,
the factorizations [Ii] = II[gj]lT ij must have Uij > 0 (by the
Proposition) and must have one U ij = 1 and the rest = 0 (by
irreducibility). In short, the [Ii] are included among the [gil.
Both factorizations are therefore of the form II[gj]lTj , and the
uniqueness of such a factorization follows from the Proposition.
(3) A common refinement of [/][q] = [h 1 ][h 2 ]· •• [hIll, where [I]
and [q] are integral and the [hi] are prime in K, gives, in partic-
ular, [I] as a product of integral divisors, each of which divides
one of the [hi]. Since the [hi] are irreducible, this gives, when
factors [1] are ignored, [I] as a product of [h],s. The exponent
of anyone [h] in [I] is of course at most its exponent in [/][q].
§1.20.
THEOREM 2. Let [I] and [g] be nonzero integral divisors.
Then there is an integral divisor [h] such that [g, h] = [1] and
[/][h] = [a] for some a E K.
Otherwise stated, [I] divides a "principal" divisor [a] with
a quotient [h] relatively prime to [g]. (Cf. Gauss, D. A., Art.
228.)
PROOF: Let [g] be written as a product [g] = [gl][g2]··· [g,,]
of integral divisors, none of which is [1]. A process will be
given which produces either an [h] with the required properties
or a proper refinement of the factorization of [g], that is, a
refinement in which at least one [gil is given as a product of at
least two integral divisors i= [1]. Since the process never goes
beyond the field extension K of r, of finite degree, generated
by the coefficients of I and g, Corollary (5) of §1.17 shows that
it must eventually produce an [h].
By Theorem 1 and the Proposition of §1.19, the factorization
of [g] can be assumed to be of the form [g] = [gl]17' 1 [g2] 17'2 •••
 §1.20 Two Basic Theorems 35

[gILt!', where the [gd are relatively prime integral divisors =I [1]
and the O"i are positive integers. Let [g] = [gl][g2]· .. [gIL]. For
each i = 1,2, ... ,p, let hi be a polynomial with coefficients in
J{ such that [hi] is the product of [f] and all the [gj] other than
[gil; in other words, let [hd = [f][g][gd- 1 . Then (Corollary
(10), §1.12) (f][g][gd- 1 is the greatest common divisor of the
coefficients of hi. At least one of these coefficients must not
be divisible by [f][g] since otherwise [f][gJl(f][g][gi]-I, [gdl[l],
[gil = [1], contrary to assumption. Thus, for each i, there is a
(3(i) E J{ integral over r such that [f][g][gi]-l divides (3(i) but
[f] [g] does not.
Again by Theorem 1, there are integral divisors [1h], ['lfJ2],
... , ['lfJr] such that [f]' all the divisors [gi], and all the divisors
[(3(i)] are products of powers of the ['IfJ]'s. If one of the ['IfJ]'s is
a proper divisor of [gd, then a refinement of the factorization
of [g] has been achieved. Otherwise, [gd is equal to one of
the ['IfJ],s. In the expression of the integral divisor [(3(i)] as
[f][g][gd- 1 times a product of powers of the ['IfJ],s, the product
of powers of the ['IfJ]'s cannot include [gil because (f][g] t [(3(i)].
Therefore, [f][gi] does not divide [(3(i)]. Since the same is true
for each i, if this procedure does not produce a refinement of
the factorization of [g], then [f][gi] does not divide [(3(i)] for i
= 1, 2, ... , p.

Assume [f][gd f [(3(i)] for i = 1, 2, ... , p, and set a =

(3(1) + (3(2) + ... + (3(IL). Let [~l], [~2], ... , [~p] be relatively
prime integral divisors such that [f]' all [gi], [a], and all [(3(i)]
are products of powers of the [~]'s (Theorem 1). Again, if the
[~]'s do not refine the factorization of [g] then each [gil must
coincide with one of the [~]'s. Since (f] divides each [(3(i)], it
divides [a], and [a] = [f] times a product of powers of the
[~]'s. Since [f][gd divides (3(j) for i =I j but does not divide
(3(i), (f][gi] does not divide a = L (3(j) (Proposition (1) of
§1.8). Thus, [a] is [f] times a product of powers of [~]'s other
than [gd. Since i was arbitrary and the [~]'s are relatively
36 Divisor Theory

prime, [a] = [f][h], where [h] is relatively prime to [g], as was

to be shown.
COROLLARIES. (1) If [f] is an integral divisor and if a E K
satisfies [f] 1[a], then [f] = [a,,8] for some ,8 E K.
(2) Let [f] be an integral divisor in K and let pEr be an
element of r which [f] divides. Then there is aWE K such
that an element, E K integral over r is divisible by [f]1L if
and only if ,WIL is divisible by plL.
(When K is a number field, when [f] is prime in K, and
when p is prime in r = Z, Corollary (2) gives Kummer's rep-
resentation of "ideal prime factors" and their multiplicities in
terms of the arithmetic of elements of K integral over r = Z.
See also Corollary 2, §2.11.)
DEDUCTIONS: (1) By the theorem, there is an integral divi-
sor [h] relatively prime to [f]-l[a] such that [f][h] = [,8] for
some ,8 E K. With [g] = [f]-l [a], then, [a,,8] = [fg, fh] =
[f][g, h] = [f][l] = [fl. (2) Let [P] = [f][g]. By the theorem,
there is aWE K such that [w] = [g][h] where [f, h] = [1].
Then [f]1L 1(f] if and only if [f]1L 1[,][h]1L ([f, h] = [1] implies
[f1L,h] = [1] and [flL,h lL ] = [1], which implies [flL"h lL ] =
[f1L, ,], by (18) and (17) of §1.15) which is true if and only if
[f]1L [g]1L 1(f][h]lL[g]lL, that is, [P]IL 1(fWIL]. By Proposition (7) of
§1.9, this is true if and only if plL I,WIL.
§1.21 The divisor class group.
A question of considerable importance in the theory of divi-
sors is that of determining whether a given divisor is principal,
where a principal divisor is by definition one of the form [0:']
with a E K. Otherwise stated, given a field K :J r and a
divisor [f] represented by a polynomial with coefficients in K,
the problem is to determine whether [f] = [a] for some a E K.
The answer depends on the field K-a divisor which is not
principal in K may well be principal in some extension of K,
 §1.21 The Divisor Class Group 37

although a divisor which is principal in K is principal in any

extension of K.
This problem is closely related to the problem of factoriza-
tion in the ring R of elements of K integral over r. Indeed,
if , E R can be factored, = a{3, where a, {3 E R are not
units of R, then b] = [a][{3], which shows not only that b] is
reducible but also that it has factors which are principal. If
[,] can be written as a product of factors prime in K, then the
finite number of integral divisors in K which divide [,] can be
listed, and, factors in R if and only if there is a divisor in this
list other than [1] and [,] which is principal. (If a E K, and
if [a] is a proper factor of b]' then [a] is integral and a E R;
moreover, since [a]I" {3 = ,/a E R. Finally, a and (3 are not
units because their divisors are not [1].)
A useful tool in the problem of determining which divisors
are principal is the divisor class group, defined as follows: The
multiplicative group of nonzero divisors represented by poly-
nomials with coefficients in K was defined in §1.14. Let V
denote this group. The nonzero principal divisors are a sub-
group of V, call it Vo. The divisor class group is the quotient
group V/Vo.
A classification of V is (1) a group G and (2) a homomor-
phism V ~ G which is onto and has kernel Vo. The problem
of finding a classification includes the problem of determining
whether a given divisor is principal (a given divisor is princi-
pal if and only if its image under the homomorphism is the
identity of G) but putting the problem in this larger context
can clarify it.
The divisor class group V /Vo can also be defined as the
semigroup of integral divisors modulo the subsemigroup of
principal integral divisors, as can be seen as follows: Any el-
ement of V can be written [Jl][I2]-1, where [11] and [12] are
integral. By Theorem 2, [h][h] = [a] for some integral divi-
sor [h) and for some a E K. Then [ft][h]-l represents the
same element of V/V o as [ft][h]-l[a] = [ft][h] does. Thus,
38 Divisor Theory

any class of V (Do can be represented by an integral divisor,

and the natural map from the semi group of integral divisors
to VIVo is onto with kernel the sub semi group of principal in-
tegral divisors, as was to be shown.
§1.22 Prime Factorization and Normal Extensions.
An interesting and useful topic in arithmetic is the study
of the way in which divisors prime in K factor when K is ex-
tended. The uniqueness of factorization of divisors into primes
(Corollary (2), §1.19) easily implies the following proposition
about the factorization of primes in a normal extension.
PROPOSITION. Let K :J r be an algebraic extension and let
L :J K be a normal field extension of finite degree, that is,
let the group G of automorphisms of Lover r which leave all
elements of K fixed be a finite group with the property that
the only elements of L left fixed by all automorphisms in G are
the elements of K. Let P be a divisor prime in L, and let P
divide a divisor Q prime in K. Then Q is a product of divisors
prime in L of the form

where the Pi are the distinct divisors of the form S( P) for

S E G. Moreover, )"JlV = [L: K] = order of G, where).. is
defined by*
II
S(P) = QA.
SEG

PROOF: Here S(P) means, of course, the divisor [Sf], where

f is any polynomial with coefficients in L which represents P.
Because [g]l[h] implies [Sg]I[Sh], it is easily shown that S(P)
depends only on P and is prime in L. Since PIQ, and SeQ) =
Q, S(P)IQfor all S E G. Therefore, (IIS(p»)IQn, where n is
*The divisor on the left side of this equation is the relative norm from L
to K of P. See §3.17.
 §1.23 Rings of Values 39

n
the order of G. IT P = [I], then S(P) = [n
S(f)] , which is a
divisor represented by a polynomial with coefficientsin K (be-
n
cause the coefficients of S(n are invariant under G). Since
this divisor divides Qn, and since Q is prime in K, Corol-
lary (3) of §1.19 implies that n
S(P) = QA for some integer
A > 0 (n S( P) is divisible by P and is therefore not [1 D. Since
n S(P) is a product of divisors prime in L, the same corollary
implies Q = P:l p;2 ... p;1'
for some integers VI, V2, ••• , VI" all
~ O. Each Pi occurs in n
S(P) exactly nj Jl times. Therefore,
n
Pi divides the left side of S(P) = QA exactly nj Jl times,
and divides the right side exactly AVi times. Therefore, by the
uniqueness of factorization into primes, Vi = nj JlA for all i, as
was to be shown.
Note that PI P2 ••. PI' is invariant under G (because elements
of G permute the P's), but is not a divisor in K when V > 1
(because its vth power is prime in K). Since V > 1 can occurt,
divisors in L invariant under G need not be divisors in K.
However, as is shown in §1.32, V = 1 for all but a finite number
of divisors P prime in L, so most divisors in L invariant under
G are in K.
To study factorization in extensions L :::> K which are not
normal, it is best to find a larger extension L' :::> K which
is normal and to apply the above proposition to the normal
extensions L' :::> L and L' :::> K.
§1.23 Rings of Values.
Given an algebraic extension K of a natural ring r, there is
a way of associating to each nonzero integral divisor A in K a
certain ring, which we will call the ring of values of K at A.
Every nonzero divisor can be written as a quotient of integral
divisors (§1.14). Since the greatest common divisor of numer-

tLet r = Z, K = =k Q, L = Q( v'2). Then [v'2) is invariant under G

(G carries v'2 to ±v'2) but is not a divisor in Q because its square is [2],
which is prime in Q.
40 Divisor Theory

ator and denominator can be cancelled, every nonzero divisor

can be written as a quotient of relatively prime integral divi-
sors. Since Corollary (1) of §1.19 implies that such a represen-
tation is unique (if CD' = C'D, where [C, D] = [C', D'] = [1],
then the factors of a common refinement of CD' = C'D which
divide C must divide C', hence CIC', and, in the same way,
C'IC, so C = C', D = D') a nonzero divisor has a well-defined
numerator and denominator. We will say that an element
a E K, a -:f 0, is finite at A if the denominator of [a] is rela-
tively prime to A. (Let the denominator of [0] be considered
to be [1], so that 0 is finite at A for all A.) Then the elements
of K finite at A form a ring; if a and j3 are both finite at
A, then, by Theorem 2, there exist elements bi and b2 of K
integral over r and relatively prime to A such that bi a and
b2j3 are integral over r, say a = aI/bI, j3 = adb 2, which im-
plies that the denominators of [a + j3] = [(b 2 aI + bIa2)/bIb2]
and [aj3] = [al a2/bi b2] both divide [b i b2] and are therefore
relatively prime to A. We will say that an element a E K,
a -:f 0, is zero at A if A divides the numerator of [a] (which
implies, of course, that a is finite at A). (Let the numerator
of [0] be considered to be [0]' so that 0 is zero at A for all A.)
Then the elements of K which are zero at A are clearly an
ideal in the ring of elements of K finite at A. Therefore, the
quotient-elements finite at A modulo elements zero at A-is
a well-defined ring. This ring is the ring of values of K at A.
§1.24.
PROPOSITIONS. Let A, B, and C be nonzero integral divisors
in K, and let VA, VB, and Vc be the rings of values of K at
A, B, and C, respectively.
(1) If AlB, then there is a natural homomorphism VB -+ VA
which is functorial in the sense that if BIC then the homomor-
phism Vc -+ VA which results from Ale is the composition of
the homomorphisms Vc -+ VB -+ VA resulting from BIC and
AlB.
 §1.24 Rings of Values 41

(2) Tbe bomomorpbism of (1) is onto; it is one-to-one only

if A = B.
(3) (Chinese Remainder Theorem) H A and B are relatively
prime, tben tbe maps VAB ~ VA and VAB ~ VB of (1) com-
bine to give an isomorphism VAB ~ VA ffi VB, wbere VA ffi VB
denotes ordered pairs, of an element of VA and an element of
VB, added and multiplied componentwise.
(4) A is prime if and only if VA is a field. (Here V[l], which
is a ring with one element, is not considered to be afield.)
(5) As an additive group, tbe kernel of tbe bomomorpbism
VB ~ VA of (1) is isomorpbic to VBIA. A specific isomorpbism
is given as follows. Cboose (3 E K such tbat [(3] = AC, wbere
C is an integral divisor relatively prime to B. Let ~ be tbe
element of VB represented by (3. Tbe map VBIA ~ VB wbich
sends ~ element of VBIA to its preimage under VB ~ VBIA
times (3 is independent of tbe cboice of preimage and gives such
an isomorpbism.
PROOFS: (1) If a is finite at B and AlB, then a is finite at A.
If a and a' are both finite at A and if a - a' is zero at B, then
a - a' is zero at A. In other words, if a and a' represent the
same elements of VB, then they represent the same elements
of VA. Thus VB ~ VA can be defined to take the element of
VB represented by a to the element of VA represented by a.
The homomorphism of rings defined in this way clearly has the
functorial property described in (1).
(2) By Theorem 1, if AlB, then A = AIIA2, A21A3, ... ,
AV-IIAv = B, where at each stage Ai = CiD~, Ai+l =
CiDf+l, where Ci and Di are relatively prime integral divi-
sors. Thus, in order to prove (2), it will suffice to prove it
in the special case A = CD k , B = CDkH, [C,D] = [1]. By
Theorem 2, there is an a E K integral over r which is divisible
by CD k but not by CDk+l (provided D =f [1]). Therefore,
a represents the zero class of VA but not of VB, which shows
that VB ~ VA is not one-to-one. Let a represent a class of VA,
where A = CDk. Then the denominator of [a] is relatively
prime to CDk. If k > 0, then a also represents a class of VB,
42 Divisor Theory

where B = CDkH, and the image of this class of VB is the

given class of VA. Thus, VB ~ VA is onto when k > 0. IT
k = 0, VB ~ VA EB VD by (3), which implies VB ~ VA is onto.
(3) IT [A, B] = [1], if a E K is finite at AB, and if the
images of a in both VA and VB are 0, then the numerator of
[a] is divisible both by A and by B, which implies (by (16) of
§1.15, C = [l]C = [A, B]C = [AC, BC], so C is divisible by
AB if it is divisible by both A and B) that the numerator of
[a] is divisible by AB, that is, a represents the zero element
of VAB. Thus, VAB ~ VA EB VB is one-to-one.
Let (3 E K be finite at A and let "Y E K be finite at B. It is
to be shown that there is an a, finite at AB, which represents
the same element of VA as (3 and the same element of VB
as "Y. By Theorem 2, there exist ¢>, tP E K, integral over T,
such that AI¢>, [B, ¢>] = [1], BltP, [A, tP] = [1]. By Theorem
1, [tP] and the denominator of [(3] can be written in the forms
D 1Ul DU22 · . . DUPO
I-' and DTl
1 DT2
2 · . . DTpo
1-" respec t·lveIy, were
h the
Di are relatively prime integral divisors. The denominator of
[(3tP n] then contains Di to the power max(O, Ti - nUi). Thus,
for n large, the denominator does not contain factors Di which
occur with nonzero exponent in [tP]. Since BltP, it follows that
(3tP n is finite at B, and therefore at AB. In the same way, "Y¢>n
is finite at AB for all sufficiently large n. Let

Since ¢>n is divisible by A and tP n is relatively prime to A,

tP n + ¢>n is relatively prime to A. In the same way, tP n + ¢>n
is relatively prime to B. Therefore, a is finite at AB for all
sufficiently large n . .Moreover , a - (3 = (tP n(3 + ¢>n"Y - tPn(3-
¢>n(3)/(tPn + ¢>n) = ¢>.¢>n-l("Y_(3)/(tP n +¢>n), which is ¢> times
an element of K finite at A for large n. Therefore, a - (3 is zero
at A for large n, that is, a and (3 represent the same elements
of VA. In the same way, a and "Y represent the same elements
of VB for all sufficiently large n, so a has the required property
when n is large.
 §1.24 Rings of Values 43

(4) If a represents a nonzero element of VA, then the denom-

inator of [a] is relatively prime to A and its numerator is not
divisible by A. If A is prime, the latter condition implies that
the numerator of a is relatively prime to A; in this case, a-I
represents an element of VA, which shows that the element of
VA represented by a has a multiplicative inverse. Conversely, if
A is not prime, then either A = [1] or A is reducible. If A = [1],
then VA is not a field. If A is reducible, then A = DE, where
neither D nor E is [1]. By Theorem 2, there exist /3, , E K
integral over T' such that DI[/3] with a quotient relatively prime
to E, and Elb] with a quotient relatively prime to D. Then /3
and, represent nonzero elements of VA whose product is the
zero element of VA, and VA cannot be a field.
(5) Given an element of VB/A, let, be an element of K
finite at B which represents it. The element of VB represented
by /3, is independent of the choice of , because if " is finite
at B and represents the same element of VB / A then , - "
is finite at B and zero at B / A, which implies that /3b - ,')
is finite at B and zero at A . (B / A) = B, that is, that /3,
and /3,' represent the same element of VB. This element is
in the kernel of VB -+ VA, because /3 is zero at A. If /3, and
/3,' represent the same element of VB, then the numerator of
[/3(,-,')] is divisible by B = A·(B/A), which, by the choice of
/3, implies the numerator of b - ,'] is divisible by B / A, that is,
, and " represent the same element of VB/A. Thus, the map
VB / A -+ VB defined in this way is one-to-one into the kernel of
VB -+ VA. Finally, any element of the kernel of VB -+ VA can
be represented by an element a of K finite at B and zero at A,
say [a] = AD / E, where D and E are relatively prime integral
divisors and E is relatively prime to B. Then, = a/ /3 has
divisor b) = D / EC with denominator relatively prime to B,
so , represents an element of VB / A whose image is the given
element of the kernel of VB -+ VA, and the map is onto.
44 Divisor Theory

§1.25 Primitive Elements, Norms, and Traces.

The following method of describing extensions of finite de-
gree is often useful.
PROPOSITION 1. Let K be an extension of finite degree of
a natural ring r. There is an irreducible, monic polynomial
F( x) in one indeterminate x with coefficients in r such that
K is isomorphic to the simple algebraic extension k( a) of r
obtained by adjoining a root a of F to the field of quotients k
ofr.
PROOF: By the theorem of the primitive element [E3, p. 46],
there is an a E K which generates Kover k. As was seen
in §1.9, h(a) = 0 for some nonzero, irreducible polynomial
h( x) with coefficients in r, say boa n + bl a n- I + ... + bn = 0,
where bo t o. Then boa generates Kover k and is a root of the
monic polynomial b~-I h(x/bo) = Xn+bIXn-l+bob2Xn-2+ ... +
b~-Ibn = F(x), with coefficients in r. If F(x) were reducible
over k, then h(x) would be reducible over k, which would imply
(clear denominators) that h( x) was reducible over r, contrary
to assumption. Therefore, K is isomorphic to the field ob-
tained by adjoining to k a root of the irreducible polynomial
F(x) [E3, p. 39].
PROPOSITION 2. Let a be a generator of Kover r which is
integral over r, as in the previous proposition, and let F(x) be
a monic, irreducible polynomial with coefficients in r of which
a is a root. Let L be a splitting field of F over k (the field of
quotients ofr), and let F(x) = (x-a)(x-a')(x-a")··· (x-
a(n-I) be the factorization of F over L. For any polynomial
j with coefficients in K, N K j is equal to the product of the n
polynomials j = j(O), j(1), ... , j(n-I) with coefficients in L
obtained by expressing the coefficients of j as polynomials in
a with coefficients in k, and substituting the n elements a, a',
a", ... , a(n-I) of L in place of a in the resulting expression
of j.
 §1.26 Differents 45

PROOF: See [E3, p. 82].

The trace of a polynomial f with coefficients in an extension
f{ of finite degree of a natural ring r is the coefficient of yn-l
in N K(Y + I), where y is a new indeterminate. Otherwise
stated, the trace of f with respect to the extension f{ ::J r,
denoted tr K(!), is the trace of the matrix which represents
multiplication by f relative to a basis of f{ over k, the field
of quotients of r. The first description of trK(!) shows that
if [f] is integral then trK(!) has coefficients in r (§l.17, Cor.
(1)). The second description shows that if a, b E k and f, 9
are polynomials with coefficients in f{ then tr K( af + bg) =
a trK(!) + btrK(g).
PROPOSITION 3. With the notation of Proposition 2, trK(!)
is the sum of the n polynomials f(O), f(1), ... , f(n-l) with
coefficients in L.
PROOF: N K(Y + I) = rr~:ol(y + f(i)) = yn + yn-l L: f(i)+
terms of lower degree in y, by Proposition 2.
§1.26 Differents.
The notion of the discriminant of an algebraic extension
f{ of a natural ring r plays an important role in Kronecker's
theory of divisors, particularly in the study of the factoriza-
tion of divisors of the form [p], where p is a prime in r. Dis-
criminants are also important, in the same way, in Dedekind's
theory of ideals in algebraic number fields. However, Dedekind
simplified and clarified the theory by introducing a more re-
fined concept, a divisor whose norm is the discriminant*; he
called the new divisor the fundamental ideal, but Hilbert re-
named it the different, a name which has become standard.
In Kroneckerian terms, the different (which Kronecker never
used) can be described as follows.

*At least in the case r = Z considered by Dedekind, the discriminant is

the norm of the different (see §2.9).
46 Divisor Theory

Given a subset aI, a2, ... , all of an extension K of finite

degree of a natural ring r, its different with respect to K,
denoted difK(al' a2, ... , all), is the divisor in K represented
by the following polynomial with coefficients in K.
Let UI, U2, ... , UII , and x be indeterminates, let a = al UI +
a2u2 + ... + allu lI , and let F = N K(X - a), a polynomial with
coefficients in the field of quotients k of r, homogeneous of
degree n = [K: k] in x, UI, U2, ... , UII • Let F be regarded
as a polynomial F(x) in x with coefficients in the ring k[UI,
U2, ... , UII] of polynomials in the indeterminates UI, U2, ... ,
U II with coefficients in k. When h is a new indeterminate,
F(x + h) = F(x) + hF'(x) + h2 G(x, h), where F' and G are
uniquely determined polynomials with coefficients in k[UI, U2,
... , u lI ]. Substitution of a for x in F'(x) gives a polynomial
F'( a) in UI, U2, ... , UII with coefficients in K. By definition,
difK(al, a2, ... , all) = [F'(a)].
§1.27.
PROPOSITIONS. (1) A set aI, a2, ... , all E K generates K
over k if and only if dif K( aI, a2, ... , all) =f [0].
(2) Fore E k and aI, a2, ... , all E K, difK(cal' ca2, ... ,
call) = [c]n-l difK(al, a2, ... , all), wbere n = [K: k].
(3) Ifv > 1, difK(al, a2, ... , all) divides dif K (al,a2, ... ,
all-I).
(4) If aI, a2, ... , all are integral over r, tben difK(al, a2,
... , all) depends only on tbe ring r[al, a2, ... , all] in K
generated by tbe a's over r; in otber words, if f31, f32, ... , f3 JL
can all be expressed as polynomials in tbe a's witb coefficients
in r, and all a's can be expressed as polynomials in tbe /3's
witb coefficients in r, tben difK(al, a2, ... , all) = dif K(f3I,
f32, ... , f3 JL ) •
PROOFS: (1) By Proposition 2 of §1.25, F(x) = n~:ol(x -
a(i)), where a(O), a(1), ... , a(n-l) are the n conjugates of a
with coefficients in an extension L of K. Let a = a(O). Thus
n-l
F'(x) = I)x-a(O)) ... (x-a(i-l))(x_a(i+l)) ... (x_a(n-l))
i=O
 §1.27 Differents 47

from which it follows that F' (a) = IT~:II (a - a( i)). If F' (a) =
0, then a = a(i) for some i = 1, 2, ... , n - 1, that is, one of
the conjugations K ~ L other than the identity carries all the
coefficients aI, a2, ... , a v of a to themselves; therefore, * the
a's cannot generate Kover k. Conversely, if the a's do not
generate Kover k, they generate a proper subfield of Kover
k, call it K'. As follows easily from the definition of the norm,
NK(x - a) = NK,(x - a)[K:K'J, that is, F(x) = Fo(x)j, where
Fo(x) = NK'(X - a) and j = [K: K'] > 1. Since F(a) = 0,
Fo(a) = o. Thus, F'(a) = jFo(a)j-1 F~(a) = 0, so [F'(a)] =
[0], as was to be shown.
(Note that [F'(a)] = IT:-{[a-a(i)] gives difK(al, a2, ... ,
a v ) as a product of divisors in L.)
(2) Replacing each ai with cai in each of the n - 1 factors
of F'(a) = IT(a - a(i)) gives cn - I IT(a - a(i)).
(3) If F'(a) is the polynomial which by definition represents
difK(al, a2, ... , a v ), the polynomial which by definition rep-
resents difK(aI, a2, . .. ,av-I) is obtained by setting U v = 0 in
F' (a). Therefore, its coefficients are a subset of the coefficients
of F'(a), so the divisor it represents is divisible by difK(aI,
a2, ... , a v).
(4) Suppose,8 = 'ljJ (aI, a2, ... , a v), where'ljJ is a polynomial
with coefficients in r. If S is an automorphism of Lover k,
then ,8-S,8 = 'ljJ(al, a2,· .. ,av)-'ljJ(Sal, Sa2, ... ,Sav) can
be expressed as a linear combination of al - Sal, a2 - Sa2,
... , a v - Say in which the coefficients are polynomials in the
a's and the Sa's and are therefore integral over r. Thus, for
an indeterminate v, the coefficients of (a + ,8v) - S(a + ,8v) =
(a - Sa) + (,8 - S,8)v are all divisible by [a - Sa] (a divisor in
L, namely, the g. c. d. of the ai - Sai). Therefore, [(a + ,8v)-
S(a+,8v)] = [a-Sa]. Use this identity in [F'(a)] = IT[a-Sa]
to find difK(al, a2, ... , a v ,,8) = difK(al, a2, ... ,av). Thus,

*This argument is valid only when the extension is separable. (Note added
in second printing.)
48 Divisor Theory

if the a's and {3's are as in the statement of the proposition,

dif K (a1, a2, ... , all) = difK (ar, a2, ... , all, {31, {32, ... ,
(3p,) = difK ({3r, (32, ... , (3p,).
Let ar, a2, ... ,all E K be integral over r, and let ROt denote
r[ar, a2, ... , all], the subring of K generated by the a's and
r. Proposition (4) shows that dif K (a1, a2, ... , all) depends
only on ROt, not on the a's used to generate ROt. This divisor
in K is the different of ROt relative to K, denoted difK(ROt).
§1.28.
Let a1, a2, . .. , all be integral over r in an extension K of
r of finite degree, let F'(ii) be defined as in §1.26, and let ROt
be as above.
PROPOSITION. Ifh is a polynomial with coefficients in K, and
if [h] is integral, then F'(ii)h has coefficients in ROt.

PROOF: It will suffice to prove the proposition in the case

where h is constant-that is, h is an element of K-because a
general h is a sum of multiples of such polynomials.
Given h E K integral over r, let trK(hq(x)) = ¢(x), where
q(x) = F(x)/(x-ii). Since F(x) has coefficients in r, the divi-
sion algorithm shows that q(x) has coefficients in the ring ROt.
Therefore, the coefficients of hq(x) are integral over r, which
implies that the coefficients of ¢( x) are in r. On the other
hand, ¢(x) = E,¢(i)(X), where the ,¢(i)(x) are the (not neces-
sarily distinct) conjugates of hq(x) in L (Proposition 3, §1.25).
Therefore, ¢(ii) = E,¢(i)(ii). The summand ,¢(i)(ii) corre-
sponding to the identity automorphism of Lis hq(ii) = hF'(ii).
(q(x) = (F(x) - F(ii))/(x - ii) = (F(ii + x - ii) - F(ii))/(x-
ii) = (F(ii) + (x - ii)F'(ii) + (x - ii)2G(ii, x - ii) - F(ii)) /(x-
ii) = F'(ii) + (x - ii)G(ii,x - ii), so q(ii) = F'(ii).) The
other summands are all zero because they are the values at
x = ii of h times n - 1 factors of the form x - ii(j), one of
 §1.29 Discriminants 49

which is x-a. Therefore, hF'(a) = ¢(a) has coefficients in

R a , as was to be shown.
COROLLARY 1. If the different of Ra is not [0], there is a
divisor A in K such that all elements of K divisible by A are
in Ra.
DEDUCTION: By Proposition (1) of the preceding article,
difK(Ra ) =1= [0] implies that any generator I of Kover k can
be expressed as a polynomial in the a's with coefficients in k.
Multiplication of such an expression by an element of r to clear
denominators shows that a multiple of I is in Ra. A multiple
of I is also a generator of Kover k, so there is a generator I
of Kover k in Ra. Then difK(Ry) =1= [0] and Ry, the subring
of K generated by I and r, is a subring of R a , so it will suffice
to prove the corollary in the case in which there is just one
generator of the ring Ra. In this case, F'(a) is u~-1 times an
element of K, call it F'(a). Let A = [F'(a)]. If AIf3 E K, then
f3/ F' (a) is integral over r, and (f3/ F'(a) )F'(a) = (3u~-1 is a
polynomial with coefficients in R a , as was to be shown.
A divisor A with the property of Corollary 1 is called a con-
ductor of Ra in K. (The* conductor is the greatest common
divisor of the conductors.)
COROLLARY 2. Given a1, a2, ... , all E K integral over r
generating Kover k, there is an element ~ of r such that
every element of K integral over r can be expressed in the
form ¢(a)/ ~ where ¢(a) E Ra = r[a1, a2, ... , all].
DEDUCTION: Let A be as in Corollary 1 and let ~ E r be
divisible by A. If (3 E K is integral over r, then ~(3 is divisible
by A, so ~(3 = ¢( a), as was to be shown.
§1.29 Discriminants.
The discriminant with respect to K of a finite subset 11,
12, ... , I JL of K, denoted disc K (,1, 12, ... , I JL)' is the fol-

*This definition requires that a proof of the g. c. d. of conductors is a

conductor. (Note added in second printing.)
50 Divisor Theory

lowing divisor in K. Let S be the It x It symmetric matrix

whose entry in the ith row of the jth column is trK(fnj), an
element of k. Let n = [K: k1 (finite by assumption), let U
be an n x It matrix whose entries are indeterminates Uij, and
let U t be the transpose of U. By definition*, discK(fl, 1'2,
... , 1'J.t) = [det(USU t )], the divisor represented by the poly-
nomial det(U sut) in the indeterminates Uij with coefficients
in k. The subscript K will sometimes be dropped from disc K
when there is no danger of confusion.
PROPOSITION. If 1'J.t+l = al1'l +
a21'2 + ... + aJ.t1'J.t,
where the
ai E r, then discK(fb 1'2, ... ,1'J.t+r) = discK(fb 1'2, ... ,1'J.t).
PROOF: Let S be the It x It symmetric matrix with entries
(tr(fnj)) for 1 ~ i ~ It, 1 ~ j ~ It, and let 8 be the (It +
1) x (It + 1) symmetric matrix with entries (tr(fnj)) for 1 ~
i ~ It + 1, 1 ~ j ~ It + 1. By the linearity of the trace,
8 = MSM t , where M is the (It + 1) x It matrix whose first
It rows are the It x It identity matrix and whose (It + 1)st row
is aI, a2, ... ,aJ.t' Let V be an n X (It + 1) matrix whose
entries are indeterminates Vij, let U be an n x It matrix whose
entries are indeterminates Uij, let f = det(USU t ), and let
9 = det(V8V t ). The coefficients of f are a subset of the
coefficients of gj more precisely, f is obtained from 9 by the
substitution Vij = Uij for j ~ It, Vi,J.t+l = O. Therefore, [gll[f1·
On the other hand, 9 = det(VMSMtV t ), which is to say that
9 is obtained from f by the substitution U = V M, that is,
Uij = Vij + ajvJ.t+1,j. Since the aj are in r, it follows that any
common divisor of the coefficients of f divides all coefficients
of g, that is, [f11[g]. Therefore, [f1 = [g], as was to be shown.
Corollary. DiscK(fl, 1'2, ... , 1'J.t) depends only on the r-
module generated by the l' 'So In other words, if 1'~, 1'~, ... ,
1'~ E K satisfy 1'~ = l: aij1'j and 1'i = l: bij1'j, where the aij

*It is not difficult to show that this definition is equivalent to Kronecker's

definition [Krl, §8].
 §1.30 Discriminants 51

and bij are in r, then discK(')'I, ,2, ... , ,,,) = discK(,L ,~,
... , ,~).

·
t o dISCK ( ,I, ,2, ... , ,,,,'I, '2, ... , 'v .
DEDUCTION: By the proposition, both discriminants are equal
I I ')

§1.30.
Any ring of the form ROt = r[al, a2, ... , a v ], where the
ai E K are integral over r, is a finitely-generated r-module.
Indeed, if r is an integer such that, for each i, af is a linear
combination of lower powers of ai with coefficients in r, then
the rV monomials, of the form a~l a~2 ... a~" with 0 ~ ei < r
span ROt over r. By the above Corollary, the discriminant of
these ,'s depends only on ROt. It will be denoted discK(ROt).
PROPOSITION. As in §1.25, let K = k(a), where a is integral
over r. Then the discriminant of ROt = r[a] with respect to K
is NK(difK(a)).
PROOF: As was seen in §1.27, difK(a) = Il[a-a(i)], where
a(l), a(2), "', a(n-l) are the roots, other than a, of F(x) =
N( x - a) in a splitting field of F. Then, by Proposition 2
of §1.25, N(difK(a)) = [N(Il(a - a(i»))] = [Il(a(j) - a(i»)],
where the product is over all n( n -1) pairs of roots a(j) , a( i) of
F( x) in a splitting field (a = a(O), n = [K: kD for which a(j) =f
a(i). Therefore*, N(difK(a)) = [Ili<j(a(j) - a(i»)]2. Now
Ili</ a(j) - a(i») is the determinant of the matrix whose ith
column is 1, a(i+1) , (a(i+l)?, ... , (a(i+l»)n-l (a Vandermonde
determinant). Let A denote this matrix. Then N(difK(a)) =
[det(A)]2 = [det(AAt)]. With ,i = ai-I, the entry in the ith
row of the j th column of AAt is tr( ,i,j)' Since the ,'s span
ROt, discK(R Ot ) = [det(U AAtU t )], where U is an n X n matrix

*This equation shows that N(difK(a)) = discK(R a ) is the divisor repre-

sented by the discriminant of the polynomial F (see [E3, §13]), which is
the reason for the name "discriminant"..
52 Divisor Theory

of indeterminates. But det(U AAtUt ) = det(AAt) det(U)2 and

det(U) is primitive (in fact its coefficients are all ±1). Thus
[det(U)] = [1] and discK(Ra) = [det(AAt)] = NK(difK(a)),
as was to be shown.
THEOREM. Let M and M' be finitely generated r-modules in
K, and let M' c M. Then discK(M') = A2 discK(M), where
A is an integral divisor. In particular, if disc K( M) = [0] then
discK(M') = [0]. If discK(M) t- [0], then A can be found as
follows. Let ')'1, ')'2, .•• , ')'Il generate M and ')'~, ')'~, ... , ')'~
generate M'. Choose a basis (1, (2, ... , (n of Kover k and let
T = (tij), T' = (t~j) be the matrices with entries in k defined
bY')'i = Etij(j, ')'~ = Et~j(j. Let U and V benxp, andnxv
matrices, respectively, whose entries are indeterminates. Then
[det(UT)] t- [0] and A = [det(VT')][det(UT)]-I.
PROOF: Let S-y' be the v x v matrix (tr( ')'hj)) and let S-y.-y'
be the analogous (p, + v) x (p, + v) matrix with ')'1, ')'2, ••. ,
" ')'2, ... , ,),,,,.III pace
')'Il' ')'1, I f0 " , By defiill·t·lon,
')'1, ')'2, ... , ,),,,.
disCK(M') = [f]' where f = det(VS-y' vt). By the propo-
sition of §1.29, discK(M) = [g], where 9 = det(WS-y.-y' W t ),
with W an n x (p, + v) matrix of indeterminates. Since
the coefficients of f are a subset of the coefficients of g,
[gJl[J). Thus, discK(M') = B discK(M), where B is an in-
tegral divisor, and it is to be shown that B is a square. If
discK(M) = [0], then discK(M') = [0], and B can be taken
to be [1]2. Assume discK(M) t- [0], and let S, be the n x n
symmetric matrix (tr( (i (j )) . By the linearity of the trace,
S-y' = T' Sc(T,)t. Thus, discK(M') = [det(VT'S,(T,)tVt] =
[det(Sc)][det(VT')]2 (because VT' is n x n). In the same
way, discK(M) = [det(Sc)][det(UT)]2. Thus, [det(UT)] t- [0]
and discK(M') = [det(Sc)][det(VT')]2 is discK(M) times the
square of A = [det(UT)]-1 [det(VT')]. Therefore A2 = B is in-
tegral, which shows that A is integral and satisfies the required
relation.
COROLLARY 1. Let al ')'1 + a2')'2 + ... + an')'n + an+! ')'n+! = 0,
 §1.30 Discriminants 53

where the ai E r, the Ii E K are integral over r, and II, 12,

... , In are a basis of Kover k. Then

[aI, a2,·· . ,an+l]2 discK( It, 12,·· ., In)

= [an+l]2 discK( ,1,,2, ... "n+l)'

In particular, if, as it is natural to assume, [aI, a2, ... , an+l] =

[1], then [an+lF divides discKbb 12, ... , In).
DEDUCTION: If an+l = 0, then ai = 0 for all i and the equa-
tion is trivially true. Otherwise, in the theorem, let J.l = n + 1,
v = n, (i = Ii for i = 1, 2, ... , n, and Ii = Ii for i = 1,
2, ... , n. It is to be shown that A = [an+l]/[al, a2, ... ,
an+l] = [det(VT')]/[det(UT)], that is, [an+d[det(UT)] = [at,
a2, ... , an+!][det(VT')]. Since T' is the n X n identity ma-
trix, [det(VT')] = [det(V)] = [1]. The (n + 1) X n matrix
T has the n X n identity matrix in its first n rows and has
-aI/an+l, -a2/an+l, ... , -an/an+! in its last row. Let the
indeterminates in the first n columns of the n X (n + 1) ma-
trix U be called x ij and let the indeterminates in the last
column be called WI, W2, ... , w n . Then UT is the matrix
which has Xij - (aj/an+!)wi in the ith row of the jth col-
umn. Let M be the matrix obtained by multiplying the ith
row of UT by WI and subtracting from it Wi times the first
row, for i = 2, 3, ... , n. Then det(M) = wr-
l det(UT), and

M has Xlj - (aj/an+l)wl in the first row, WlXij - WiXlj in

rows after the first. It is clear from the definition of the de-
terminant that the coefficients of det(M) are ±1, ±aI/an+l,
... , ±an/an+l' (For example, -aI/an+l is the coefficient of
WiX22X33 ... Xnn in det(M).) Thus, [det(UT)] = [aI/an+l,
adan+I, ... , an/an+l, 1] = [aI, a2, ... , an+l]/[an+l], as was
to be shown.
COROLLARY 2. Let aI, a2, ... , a v E K be integral over r,
and let Ret = r[al, a2, ... , a v ]. Then discK(R et ) =1= [0] if and
only if the a's generate Kover k.
54 Divisor Theory

DEDUCTION: Let (3 be a generator of Kover k which is in-

tegral over r (§1.25), let Rf3 = r[(3], and let RO/,f3 = r[aI, a2,
... , a v ,(3]. By the theorem, discK(Rf3) = A 2 discK(RO/,f3).
By the proposition, diSCK(Rf3) = N(dif K (Rf3)) , and by (1) of
§1.27, dif K (R f3 ) f:. [0]. Therefore, discK(Rf3) f:. [0], which im-
plies diSCK(RO/,f3) f:. [0]. Again by the theorem, discK(RO/) =
Ai discK(RO/,f3), so diSCK(RO/) = [0] if and only if Al = [0],
which, by the theorem, is true if and only if det(VT') = 0,
where V is an n X v matrix of indeterminates and T' is the
v x n matrix whose rows give the coordinates of a set of ele-
ments which span RO/ relative to a basis of Kover k. If the
a's fail to generate Kover k, there is a basis (I, (2, ... , (n of
Kover k whose first s < n elements span the subspace of K
generated over k by the a's; in this case, the last n - s columns
of T' are 0, det(VT') = 0, and diSCK(RO/) = [0]. Conversely,
if the a's do generate Kover k, then the rows of T' span k n ,
the rank of T' is n, a subset of n rows of T' has nonzero deter-
minant, values in k (in fact in the set containing just 0 and 1)
can be assigned to the indeterminates in V to make VT' have
nonzero determinant, det(VT') f:. 0, and discK(RO/) f:. [0].
§1.31 Ramification.
DEFINITION. An element a integral over r ramifies in an al-
gebraic extension K of r containing a if there is an element
(3 E K not divisible by a whose square is divisible by a. (Note
that ramification depends on the ambient field-an element
which does not ramify in K may ramify in an extension of K,
although if a ramifies in K, then it ramifies in any extension
of K. Note also that the definition does not involve divisor
theory.)
PROPOSITION. An element a E K integral over r ramifies in
K if and only if [a] is divisible by the square of an integral
divisor in K other than [1].
PROOF: If A2I[a], where A is integral and A::J [1], then, by
 §1.32 Ramification 55

Theorem 2, there is a f3 E K with [f3] = A-I [a]B, where B is

integral and relatively prime to A. Then [f3 I a] = A-I B is not
integral, so a f f3, but [f32Ia] = A- 2[a]B 2 is integral, so alf32.
Conversely, suppose a f f3 but alf32, where a E K is integral
over r. By Theorem 1, there exist relatively prime integral
divisors Ql, Q2, ... , Qp. such that both [a] and [f3] are products
of powers of the Q i, say [a] = IT Q~; and [f3] = IT Q~;. Since [a]
°
is integral, ai ~ for all i; since [aJl[f32], 2bi ~ ai for all i; but

°
since [a] f [f3]' bj < aj for at least one j (Proposition, §1.19).
The inequalities 2aj > 2bj ~ aj ~ imply first 2aj > 0, then
aj > 0, then 2b j > 0, then bj > 0, then bj ~ 1, then 2aj > 2,
and finally aj > 1. Therefore, Q;I[a].
§1.32.
Let K be an extension of finite degree of a natural ring r.
Let p be a prime element of r. The ring V[Pl of values of K at
(P] contains the ring of values of k (the field of quotients of r)
at (P]. This latter ring is a field ((P] is prime in k), call it kp,
so lipl is a vector space over the field k p • We will say that K
is p-regular if there is a basis of Kover k whose elements are
finite at (P] and whose image in V[Pl is a basis of V[Pl over kp •
As will be shown in §2.7, when r = Z, every K is p-regular for
every p. I do not know of an example r, K, p in which K is
not p-regular.
PROPOSITION 1. Let K be p-regular and let (P] = C;l .
C;2 ... C;p., where the Ci are relatively prime integral divi-
sors. Let tp denote the natural map from elements of K finite
at (P] to V[pl' and let f be a polynomial with coefficients in K
finite at (P]. Then
P.
tp(NKf) = II(Nd)e;
i=l

where Nd is tbe polynomial witb coefficients in kp C lipl

wbicb is tbe determinant of the matrix representing multipli-
56 Divisor Theory

cation by f of elements of VCi relative to a basis of VCi over

kp •
PROO F: Let 131, 132, ... , f3n be a basis of Kover k such that
t p(f3t), tp(f32), ... , tp(f3n) is a basis of V(Pl over kp (K is p-
regular). Since the f3's are a basis of Kover k, the equa-
tions f f3i = L fijf3j define polynomials hj with coefficients
in k. Because k is the field of quotients of r, there is an
a E T such that the polynomials afij all have coefficients in
T. Moreover, one can assume a E r to be chosen with the
additional property that either a is not divisible by p or at
least one coefficient of one afij is not divisible by p (since
otherwise alp would have the same properties as a). If p di-
vided a, then, because all coefficients of f, and all the f3i,
are finite at (P], af f3i would be zero at (P]; then tp applied
to aff3i = L(ahj)f3j would give 0 = Ltp(afij)tp(f3j) and
would imply, contrary to the choice of a, that tp( afij) = 0
for all i and j (because the t p(f3j) are linearly independent
over kp). Therefore, p f a and the coefficients of Iij = ahj/a
are all finite at p. Thus, tp(J)tp(f3i) = Ltp(Jij)t p(f3j) and
tp (N 1) = tp (det(Jij)) = det (tp (hj)) is the determinant of the
matrix which represents multiplication by tp(J) on V(Pl relative
to the basis tp (f3J) , tp(f32), ... , tp(f3n) of V[pl over kp. There-
fore, tp(N 1) can be computed using any basis of V(Pl over kp.
By the Chinese Remainder Theorem (§1.24, (3)), V(Pl is
a direct sum of subspaces VC~i. Relative to a basis of V[pl
which respects this direct sum: decomposition-that is, a ba-
sis whose images under V(Pl -+ VC~i are zero for all but one
i-multiplication by f is represent~d by a block diagonal ma-
trix, and the determinant of such a matrix is the product
of the determinants of the matrices on the diagonal. Thus,
tp(N 1) = rrr=1 D i , where Di is the determinant of the matrix
which represents multiplication by I on VC~i , relative to a basis
of VC~i over kp. It is to be shown that Di = (Nd)e i , where
Ni I i~ as in the proposition.
 §1.32 Ramification 57

e
Let E K be divisible by Ci- I [P] with a quotient relatively
prime to C i (Theorem 2). An element of V[Pl has image 0 in
Vc; if and only if its product with tp(e) is zero. With respect to
the basis tp (f3i) of lipl over kp, this gives a set of homogeneous
linear conditions on k; which describe the kernel of V[Pl -+ Vc;.
Therefore, by linear algebra, one can find a subset of V[Pl whose
image in Vc; is a basis of Vc; over k p • Let (I, (2, ... , (0' E K
represent such a subset of V[pl' so that the ('S are finite at [P]
and their images in Vc; are a basis over kp • Let." E K be
integral over r and have divisor CiQ, where Q is integral and
relatively prime to Ci . Then the images in Vc~; of (s."t, for
•
1 ~ s ~ (J, 0 ~ t < ei, is a basis of Vc~; over kp (by (5)
•
of §1.24 and induction on ei). The matrix which represents
multiplication by f relative to this basis is a (Jei x (Jei matrix
of (J x (J blocks, which is zero in blocks below the main diagonal
and on the main diagonal is ei copies of the (J x (J matrix whose
determinant is Nd. Therefore, the determinant is (Nd)e;, as
was to be shown.
PROPOSITION 2. Let K = k(o:), where 0: is integral over r,
and let F(x) = NK(x - 0:). If [P]2 does not divide N[F'(o:)],
then K is p-regular. In fact, 1, 0:, 0: 2 , ... , o:n-I is a basis of
Kover k whose image in V[Pl is a basis of V[Pl over kp, where
n = [K: k].
PROOF: Let tp denote the natural map from elements of K
finite at [P] to V[pl. The dimension of lipl as a vector space
over kp is at most n, because, given any n + 1 elements of V[pl,
there exist elements 1'1,1'2, ... , I'n+I of K finite at [P] such that
tp(,t), tp(1'2), ... , tp( I'n+t) are the given elements, and, since
n = [K: k] and k is the field of quotients of r, there exist aI,
a2, ... , an+I E r such that aII'I + 0:21'2 + ... + an+II'n+l = 0;
one can assume without loss of generality that [aI, a2, ... ,
a n+l] = [1], in which case t p( at)tp(,I) + t p( a2)t p(,2) + ... +
tp(an+t)tp(I'n+d = 0 and the tp(ai) are not all zero, which
shows that the given classes are linearly dependent over kp •
58 Divisor Theory

The proposition will therefore be proved if it is shown that

lp (a i ) for i = 0, 1, ... , n - 1 are linearly independent over kp.
If the lp( a i ) were linearly dependent over kp, there would
be elements Ci and di of r for i = 0, 1, ... , n - 1 such that
E7:ollp(ci/di)lp(ai) = 0, such that no di was divisible by p,
and such that at least one Ci was not divisible by p, say p f CT'
Multiplication by the product of the lp(di) would then give
lp(to + tla + ... + tn_lan-I) = 0, where the ti were in rand
p f tT' If f3 = to+tla+" ·+tn_la n- l , then f3 would be integral
over rand lp(f3) = 0, which would imply f3 = P1', where l' was
integral over r. By Corollary 1 of §1.30, P1' = to + tla + ... +
tn_la n- l would imply [P]21 discK(l, a, ... , an-I) = N[F'(a)]
(§1.30, Proposition), contrary to assumption. Therefore, the
lp( a i ) are linearly independent.
PROPOSITION 3. Let K :) r be p-regular, and let 0'1, 0'2, ... ,
a v be integral over r and generate Kover k. If[p, difK(Rex)] =
[1] then p does not ramify in K.
PROOF: If p ramifies in K, then [P] = A2 B, where A and
B are integral divisors and A =I [1]. By Theorem 1, [P] =
C~l C;2 ... C;", where the C i are relatively prime integral di-
visors and A is a product of powers of the Ci . In particular,
at least one Ci divides A, say CIIA. Then el ~ 2.
Let F(X,Ul,U2,""U v ) = NK(x - a), where a = alul +
a2u2 + ... + avu ll • By Proposition 1, lp(F) = rrr=l
g;i, where
gi is the determinant of the matrix which represents multi-
plication by x - a relative to a basis of VCi over kp • The
argument of §1.17 generalizes to prove that gi is divisible by
x - a when both are regarded as polynomials with coefficients
in V Ci . Specifically, let (1, (2, ... , (0' be a basis of VCi over
kp, and let M be the q x q matrix of polynomials with coeffi-
cients in kp which represents multiplication of elements of VCi
by x - a relative to this basis. Then gi = det(M). On the
other hand, substitution of a for x in M gives a matrix M of
polynomials with coefficients in K such that M ( = when ( °
 §1.32 Ramification 59

is the column matrix (1, (2, ... , (0' and when the coefficients
of entries of M are regarded as representing elements of VCi .
Therefore, det( M) is zero as a polynomial with coefficients in
Vc ;, or, what is the same, 9i is divisible by x - a when both
are regarded as polynomials with coefficients in Vc;. There-
fore, (x-a? divides F when both are regarded as polynomials
with coefficients in VCl ::) kp • In other words, the remainder
when N(x - a)j(x - a) is divided by x - a has all coefficients
divisible by GI , which is precisely what it means to say that
GIl difK(al' a2, ... , a v ). Since GII[P] and GI =1= [1], it follows
that [p, difK(Ra)] =1= [1], as was to be shown.
COROLLARY 1. In an extension K of r of finite degree, at most
a finite number of primes ramify.
DEDUCTION: Let K = k(a), where a is integral over r. By
. Proposition 2, K is p-regular for all but a finite number of
primes p in r. By Proposition 3, of the primes p for which K
is p-regular, p ramifies for at most a finite number.
COROLLARY 2. In the proposition of §1.22, for fixed L and
K, v is greater than 1 for at most a finite number of P. In
other words, if P is a divisor p:r;ime in L which divides a divisor
prime in K, then, with at most a finite number of exceptions,
the product of the distinct conjugates of P over K is a divisor
in K. (As §1.22 shows, if the product of the distinct conjugates
of P over K is a divisor in K, it is prime in K, because v is
the smallest power of this divisor which is in K.)
DEDUCTION: If v > 1, then P21Q and QI[P] shows that p is in
the finite set of primes which ramify, so P is in the finite set
of divisors which divide such [Pl.
 Part 2: Applications to Algebraic Number Theory

§2.1 Factorization into Primes.

Let Z denote the ring of integers. An algebraic number field
is an extension of Z of finite degree. Since Z is a natural ring,
divisor theory applies to algebraic number fields.
THEOREM. Let K be an algebraic number field and let P E Z
be a prime integer. The factorization of [P] into divisors prime
in K can be accomplished in a finite number of steps.
LEMMA. There is a finite set "YJ, "Y2, ••• ,"Yp. of elements of K
integral over Z such that every element of K integral over Z
is of the form CI"YI + C2"Y2 + ... + cp."Yp. for CI, C2, ... , cp. E Z.
PROOF: Let Q denote the field of quotients of Z, the field
of rational numbers. Because [K: Q] is finite, K = Q(a) for
some a integral over Z (§1.25). By the second corollary of
§1.28, there is a 6 E Z such that every element of K integral
over Z is of the form (ao + ala + a2a2 + ... + a n _Ia n - I )/6.
Since each ai can be written in the form ai = qi6 + ri, where
Iril :::; 161/2, every element of K integral over Z is of the form
qo + ql a + q2a2 + ... + qn-l a n - l + 8, where 8 is one of the
finite set of elements of K of the form (E~:OI Siai)/6, with
integers Si in the range 21sil :::; 161, and is integral over Z. If
"Yi.= a i - l for i = 1,2, ... , n, and "Yn+I, "Yn+2, ... , "Yp. is the list
of elements of the form (E Siai)/ 6 with the Si in this range
which are integral over Z, then every element of K integral
over Z is of the form E~l qiTi + "Yj for som~ j in the range
n < j :::; /-" and the lemma follows.
PROOF OF THE THEOREM: By Corollary (1) of §1.20, every
integral divisor of [P] in K can be written in the form [p,,8]
for some ,8 E K. Since,8 is integral over Z (Cor. (13), §1.12),
,8 = Ef=l CiTi, where the "Yi are as in the lemma and the
Ci E Z. Clearly, [p,,8] = [p,,8 ± P"Yi] for any "Yi, from which
it follows that [p,,8] = [p, Ef=l biTi] where Ef=l biTi is one of
 §2.2 A Factorization Method 61

the pI-' elements* of K of this form in which the bi are integers

satisfying 0 ~ bi < p. Therefore, the pI-' divisors [p, L:f=l bi"(i]
include all integral divisors of [Pl. Given any factorization of
[P] (counting [P] itself as a factorization in which there is one
factor) one can either refine it further (if one of the pI-' divisors
[p, L:f=l bi"(i] is a proper divisor of one of its factors) or prove
there is no refinement (if not). Since the number of integral
factors of [P] other than [1] is bounded (§1.17, Corollary (5))
the theorem follows.
COROLLARY. Any divisor in K can be written as a product of
powers (possibly negative) of distinct divisors prime in K.
DEDUCTION: By the theorem, this is true of divisors of the
form [Pl. Therefore, it is true of any product of divisors of
this form, that is, any divisor of the form [m], where m E Z.
Corollary (3) of §1.19 then implies it is true of any integral
divisor which divides a divisor of the form [m], which is to say,
by Lemma (8) of §1.10, any integral divisor. Since every divisor
is a quotient of integral divisors, the general case follows.
§2.2 A Factorization Method.
The factorization of [P] into divisors prime in a given al-
gebraic number field K can usually be found as follows.
K = Q(a) where a is integral over Z (§1.25). Let F(x)
be N K( x - a), a monic, irreducible polynomial of degree
n = [K: Q] with coefficients in Z. Let the factorization of
F( x) into powers of distinct, irreducible factors mod p be
F(x) = h(xtlh(x)e 2 ••• fm(xym modp. Then

*It is at this point that the assumption r = Z enters. Kronecker claims,

with an indication of a proof [Krl, §6], that the above lemma is true for
r a ring of polynomials with integer coefficients, which implies that every
divisor of [PI is of the form [p, L bi'YiI where the bi E r are reduced mod p.
When r = Z, the set of such divisors is finite.
62 Divisor Theory

is usually the factorization of (P], that is, usually the divisors

(p,fi(a)] are distinct and prime in K, and (1) holds.
(The precise definition of the divisors (p, Ii( a)] requires a
little explanation. Strictly speaking, the coefficients of fi( x)
are in F P' the field of integers mod p-or, bet ter, the ring
of values of Q at the divisor (P] prime in Q-and Ii (a) is
not meaningful. However, it is meaningful to say that (x) h =
h(
Ii( x) mod p, where x) is a polynomial with coefficients in Z,
h( h(
and then a) and (p, a)] are well defined. IT g( x) and h( x)
are polynomials with coefficients in Z, and if g( x) = h( x) mod
p, then g( a) - h( a) is divisible by p-because a is integral over
Z-and (p,g(a)] = (p,g(a),g(a)-h(a)] = (p, h(a)]. Therefore,
h( h(
(p, a)] is independent of the choice of x). This is the
divisor denoted (p, fi( a)] in (1).)
§2.3.
The method of factoring (P] in K sketched in the preceding
article depends on the choice of a generator a of Kover Q,
one which is integral over Z. As examples show (see §2.5),
the method may fail to factor (P] for one choice of a and
succeed for another. However, there are algebraic number
fields K and primes p for which the method fails, no mat-
ter which generator a is used (§2.5). Kronecker strength-
ened the method by replacing a with a linear combination
a = a1 U1 + a2u2 + ... + a"u" of elements ai of K inte-
gral over Z with indeterminate coefficients Ui. Then F( x) =
N(x - a) is replaced by F(X,U1,U2, ... ,U,,) = N(x - a) =
N(x - a1u1 - a2u2 - ... - a"u,,), a homogeneous polyno-
mial of degree [K: Q] in x, u 1 , U2, ... , u" with coefficients in Z.
This F can be factored mod p, that is, written in the form
F = fr f;2 ... f:nm mod p, where the Ii are distinct polyno-
mials in x, UI, U2, ... , Urn (necessarily homogeneous) with co-
efficients in F P' irreducible over F p' The theorem of the next
article gives conditions on a1, a2, .. . ,a" which imply that the
 §2.4 A Factorization Method 63

factorization of (P] is

where (p, h( a)] denotes the following divisor in K: Let h be

a polynomial with coefficients in Z such that h h ;?od p.
Substitution of a = al UI + a2U2 + ... + avu v for x in fi gives
a polynomial, call it h(a), in uI, U2, ... , Uv whose coefficients
are in K. The divisor [p,h(a)] is by definition (p,h(a)], that
i~, the greatest common divisor of p and the coefficients <1
h( a). This definition is valid because different choices of h
h(
give the same divisor (p, a)]. (The ai are integral over Z, so
g( x) = h( x) mod p implies that all coefficients of g( a) - h( a)
are divisible by p.)
§2.4.
DEFINITION. A ring R contained in an algebraic number
field K absorbs division by an integer j E Z if </J E Rand j I</J
imply </J/j E R.
THEOREM. Let K = Q( aI, a2, ... , a v ) be an algebraic
number field generated over Q by a finite number of elements
ai integral over Z, and let Ra = Z[aI' a2, ... , a v ] be the
ring generated in K by 1, aI, a2, ... , avo Let a be aIuI +
a2U2 + ... + avu v , a linear polynomial in the indeterminates
UI, U2, ... , Uv with coefficients in K, and let F be N(x -
a), a homogeneous polynomial of degree [K: Q] in x, UI, U2,
. .. Uv , monic in x, with coefficients in Z. Finally, let F _
fr f;2 ... f!m mod p be the factorization of F over F p as a
product of powers of distinct irreducible monic factors. If Ra
absorbs division by p then

gives (P] as a product of powers of distinct divisors prime in K,

where (p, fi( a)] is the divisor described in the preceding article.
64 Divisor Theory

COROLLARIES. (1) The method of§2.2 gives the factorization

of[P] whenever [PF does not divide N[F'(a)].
(2) The method of§2.2 gives the factorization of [P] whenever
the exponents ei are all 1.
DEDUCTIONS: (1) If RO/ = Z[a] does not absorb division by p,
then some,8 E K oft he form ,8 = (aO+ala+.· ·+an_Ia n- 1)/p,
in which the ai are integers not all divisible by p, is integral
over Z. By Corollary 1 of §l.30, [P]2 then divides disc(l,
a, ... , an-I) = disc(RO/). By the proposition of §l.30,
disc(RO/) = N[F'(a)]. (2) If the method of §2.2 fails, then
[PFIN[F'(a)]. By Proposition 2 of §l.25, N[F'(a)] is the prod-
uct of n = [K: Q] conjugates of [F'(a)] in a normal extension
L of Q containing K. A prime divisor of [P] in L must divide
at least one of these conjugates of [F'(a)], so some conjugate
of it must divide [F' (a)] itself, say P is a divisor of [P] prime
in L which divides F'(a). In the field Vp :J Fp of values of
L at P, the polynomials F( x) and F' (x) have the common
root l( a), the element of Vp represented by a. Two polyno-
mials in one indeterminate x with coefficients in F p have a
common root in an extension of F p if and only if they have
a common factor with coefficients in Fp. Since F = ft;, n
it follows that some h divides F', say fIIF' over Fp. Since
F' = elf~fl1 F + (terms divisible by fI), it follows that !I
divides elfffr- I f;2f;3 .... Since !I f fi for i > 1 (unique
factorization over F p) and fI f ff (fIlii would imply ff = 0
because Ii has lower degree than fI, and f{ = 0 would im-
ply !I was a pth power by Fermat's theorem, contrary to the
assumption that fl is irreducible), it follows that fllelf: 1 - I .
Therefore, el > l.
Corollary (1) shows that the method of §2.2 fails for at most
a finite number of primes p, and justifies the assertion of §2.2
that the factorization method there "usually" works.
Note that, when aI, a2, ... , all have the property of ,1,,2,
... , III in the lemma of §2.1, the ring RO/ is the set of all ele-
 §2.5 Examples 65

ments of K integral over Z, and this ring obviously absorbs

division by p for all p. In this case, therefore, factorization
of N(x - a) mod p gives the factorization of [P] in K for all
primes p E Z.
§2.5 Examples.
A quadratic extension K of Q is one which can be obtained
by adjoining a root of an irreducible polynomial aox 2+alx+a2
with ai E Z. If a is a root of this equation in K, then (2aoa)2 +
2al (2aoa) + 4aOa2 = 0 and (3 = 2aoa + al satisfies ((3 - ad 2 +
2al ((3 - ad + 4aOa2 = 0, which has the form (32 = d for d E Z.
Since K = Q( Jpc) = Q( Jc), one can assume without loss
of generality that d is divisible by no square greater than 1.
Since K =f Q, d is neither 0 nor 1. F(x) = x 2 - d, F'(x) = 2x,
F'(va) = 2va, so N[F'( va)] = [2]2[d]. Since dis squarefree,
Corollary (1) of the preceding article shows that the method
of §2.2 gives the factorization of [P] in Q( va) for all primes
p except, possibly, p = 2. The factorization of [P] in Q( va)
is found by factoring N( x - Vd) = x 2 - d mod p, that is, by
finding the roots, if any, of x 2 = d mod p. If this congruence
has no root, then [P] is prime in Q( Vd), if it has two distinct
roots ±c mod p, then [P] is [p, Jd - c][P, Jd + c], a product
of distinct divisors prime in Q( Jd), and if it has just one
root c mod p (which must be 0 mod p unless p = 2) then
[P] = [p,va - C]2.
Since ~ = d mod 2, this method gives the factorization [2] =
[2, Jd - d]2 of [2] provided Rv'd absorbs division by 2, that is,
provided 21( a + bVd), for a, b E Z, implies 21a and 21b. Now
21(a + bJd) implies 41N(a + bJd) = (a 2 - db 2). If 21a, then
41(a 2 - db 2) implies 41db2, and hence, since 4 f d, implies 21b2
=
and 21b. If 2 f a, then a2 1 mod 4, and 41( a2 - db 2) implies
=
db 2 - 1 mod 4, which implies 2 f band d 1 mod 4. Thus, if
Rv'd does not absorb division by 2, then d = 1 mod 4.
=
It remains to factor [2] in Q( Jd) in the case d 1 mod 4. In
 66 Divisor Theory

this case, N(x-1-yd) = (x-1?-d = x2-2x+(1-d) has the

coefficient of x divisible by 2 and the constant term divisible by
4, so 1 + yd is divisible by 2, that is, w = (1 + yd)/2 is integral
over Z. F(x) = N(x - w) = x 2 - X + l~d, F'(x) = 2x - 1,
F'(w) = 2w - 1 = yd, N[F'(w)] = [d], which shows that the
method of §2.2 factors [2] in Q( yd) = Q(w) when a = w. If
=
d 1 mod 8, then N(x-w) = x 2 -x+ l~d =
x(x+1) mod 2, so
=
[2] = [2,w][2,w+1]; if d 5 mod 8, then N(x-w) x 2 +x+1 =
is irreducible mod 2, so [2] is prime.
A cubic extension K of Q is one which can be obtained by
adjoining a root of an irreducible polynomial aox 3 + al x 2 +
a2x + a3 with ai E Z. If, is a root of this equation in K,
then (3aO,)3 + 3al(3ao,? + 9aOa2(3ao,) + 27a~a3 = 0, and
a = 3ao, + al satisfies (a - at)3 + 3al (a - at)2 + 9aOa2 ( a -
at) + 27a~a3 = 0, which has the form a 3 + aa + b = 0, where
a,b E Z. The norm of F'(a) = 3a 2 +a is (use the basis 1, a,
a 2 of Kover Q)
a 0 3
-3b a - 3a o = 4a 3 + 27b2,
o -3b a-3a

so [P] can be factored in K by factoring x 3 + ax + b mod p,

provided p2 f (4a 3 + 27b2 ). As in the quadratic case, a cubic
polynomial can be factored mod p by finding its roots, if any,
modp.
If p21(4a 3 + 27b 2), it may still be possible to factor [P] in K
by choosing a different generator of Kover Q, just as in the
quadratic case [2] was factored in Q( Vd) for d =
1 mod 4 by
writing Q( Vd) = Q(w), where w = (1 + Vd)/2. The classic
case of a prime [P] which can not be factored in an algebraic
number field K by this method, no matter what generator
of Kover Q is used, is the case p = 2, K = Q(a), where
a 3 = a 2 + 2a + 8 (Dedekind, Werke, vol. 1, p. 225).* With
p = 3a-1, (3a)3 = 3(3a)2+18(3a)+216, (p+l)3 = 3(p+l)2+

*See also Kronecker, II, p. 383.

 §2.5 Examples 67

18(p+l)+216, p3 = 21p+236. Since [4·( -21)3+27·( -236?] =

[2]2 [3J3 [_7 3 + 1182] = [2]2[3]6[503], any [P] other than [2] or [3]
can be factored in K by factoring x 3 - 21x - 236 mod p. For
example, [5] can be factored by factoring x 3 - 21x - 236 =
x 3 - x-I = (x - 2)(x 2 + 2x - 2) mod 5. The second factor
is (x + 1)2 - 3, which is irreducible mod 5 because 3 is not a
square mod 5. Therefore, [5] = [5, p - 2][5, p2 + 2p - 2] factors
[5] into divisors prime in K. (In terms of the original generator
a, these divisors are easily found to be [5, a-I] and [5, a 2 -2].)
Although the generator p does not give the factorization of
=
[3] (x 3 - 21x - 236 == x 3 + 1 (x + 1)3 mod 3), the generator
a does, because x 3 - x 2 - 2x - 8 = x 3 - x 2 + X + 1 mod 3
is irreducible mod 3 (it has no root mod 3). Therefore, [3] is
prime in K by Corollary (2) of §2.4.
Consider, finally, the factorization of [2] in this K. If Ret does
not absorb division by 2, then some polynomial of degree less
than 3 in a with coefficients in Z is divisible by 2 without all
its coefficients being divisible by 2. Since a 3 - a 2 - 2a - 8 = 0,
a 3 - a 2 = a 2(a - 1) is divisible by 2, and a simple candidate
for such a polynomial is a( a - I). Let f3 = a( a - I) /2. Then
f3 . I = -~a + ~a2, f3 . a = _~a2 + ~(a2 + 2a + 8) = 4 + a,
and f3 . a 2 = 4a + a 2 • Thus
1 1
X
2 -2
N(x - (3) = -4 x-I 0
0 -4 x-I

2x 1 -1 x 1 -1
1
- -4 x-I 0 -2 x-I 0
2
0 -4 x-I 0 -4
x-I
which shows that f3 is integral over Z and that Ret does not
absorb division by 2. Since a·l = a, a· a = a 2, and a· a 2 =
8 + 2a + a 2 ,
x -u + Iv
2
_Iv
2
N(x-au-f3v)= -4v x-v -u
-8u - 2u - 4v x - u- v
68 Divisor Theory

x -2u + v -v
-2v x- v -u
-4u -2u - 4v x - u- v

Mod 2, the terms below the main diagonal are 0, so N(X-OlU-

(3v) = x(x - v)(X - U - v) mod 2. H ROt,p = Z[Ol,{3] absorbs
division by 2, the factorization of [2] in K is therefore

[2] = [2, OlU + {3v][2, OlU + {3v - v][2, OlU + {3v - U - v]

= [2, Ol, {3][2, 0l,{3 + 1][2,0l + 1,{3 + 1]

That ROt,p does absorb division by 2-in fact, that ROt,p
includes all elements of K integral over Z-can be proved as
follows. Since 2{3 = 012 - Ol and since 1, Ol, 012 is a basis of K
over Q, the elements 1, Ol, (3 are a basis of Kover Q. Therefore,
any, E K satisfies a relation a, = b + COl + d{3, where a, b, c,
dE Z and [a, b, c, d] = [1]. By Corollary 1 of §1.30, [aJ2 divides
disc(I,Ol,{3). As was shown above, disc(l,p,p2) = [2]2[3]6[503].
On the other hand,

1=1
P = -1 + 30l
p2 = 901 2 - 60l + 1 = 9(2{3 + Ol) - 60l + 1
= 1 + 30l + 18{3.

By the theorem of §1.30, disc(l, p, p2) = [d]2 disc(I,Ol,{3), where

1 0 0
d= -1 3 0 = 2 . 33 ,
1 3 18

so disc(I,Ol,{3) = [503] and [a]21 disc(l, Ol, (3) implies [a] = 1,

a = ±1. Thus, , E ROt,p, as was to be shown.
That the method of §2.2 can not be used to factor [2] in this
case can be seen as follows. Any, E K integral over Z can be
 §2.6 Integral Bases 69

written in the form, = a + bo: + cf3. PI = [2,0:, f3] divides

, if a is even and PI 1(, - 1) if a is odd. Thus, PI 1,(, - 1).
Similarly, P2 = [2,0: + 1, f3] and P 3 = [2,0:, f3 + 1] both divide
,( , - 1). Since [2] = PI P2P 3 and the Pi are distinct and prime
in J{, it follows that [2]I,b -1), that is, ,2 _, mod 2. Thus,
every polynomial f (,) in , with coefficients in Z is congruent
mod 2 to one of the four 0,1"" + 1. There are therefore at
most 4 distinct divisors of the form [2, f( ,)], two of which are
[2] and [1]. Thus, at least one of PI, P2 , and P3 is not of the
form [2, f( ,)].
§2.6 Integral Bases.
THEOREM. Let J{ be an algebraic number field, and let ,1, ,2,
... , 'IL E J{ span J{ as a vector space over Q. Tbere is a basis

as tbe ,'s, tbat is, for wbicb f3i = L:: bij,j and
wbere tbe bij and Cij are integers.
,i
f3I, f32, ... , f3n of J{ over Q wbicb spans tbe same Z-module
= L:: Cijf3j,

PROOF: For each ,i there is an integer ai such that ai/i is

integral over Z. Therefore, there is an integer a such that a,i
is integral over Z for all i. If f31, f32, ... , f3n is a basis of J{
which generates the same Z-module as the a,i, then a-I f31,

module as the ,i, ,i

a-I f32, ... , a-I f3n is a basis of J{ which generates the same Z-
Therefore, it will suffice to prove the theorem
in the case where the are integral over Z, which can be done

,i
using the following algorithm of Kronecker [Krl, §7].
,i
Since the span J{, some subset of n = [J{: Q] of the are a
basis of J{ over Q. The discriminant of a basis is =f [0], because
if it were [0], then, by the theorem of §1.30, the discriminant of
every basis would be [0], contrary to the proposition of §1.30
and (1) of §1.27. Therefore, some subset of n of the ,'s has
a nonzero discriminant. Moreover, if a subset of n of the ,'s
has nonzero discriminant then it is a basis of J{ over Q, by the
theorem of §1.30.
Step 1. Among the (~) discriminants of subsets of n of the
,'s, there is a smallest nonzero one. Therefore, the ,'s can be
70 Divisor Theory

rearranged, if necessary, to make disc('n, 1'2, ... , 1'n) = [d),

where d is a positive integer, and to make the discriminant of
any subset of n of the 1"s have the form [d'], where d' = 0 or
d' ~ d.

,i
Step 2. Since 1'1, 1'2, ... , 1'n is a basis, every,i for i > n
= 2:}=1 bij,j where
(if any) can be written in the form
bij E Q. By subtracting multiples of ,1, ,2, ... , ,n from ,i,
one can put all the coefficients bij in the range 0 ~ bij < 1
without changing the Z-module spanned by the ,'so

,n
If, after step 2, the coefficients bij are all reduced to 0, then
,I, ,2, ... , span the same Z-module as the original ,'s and
the algorithm terminates. Otherwise, return to step 1. At
step 1, d must be decreased, since if, for example, bn+l,l =1= 0,
then disc(,n+l, ,2, ,3, ... , ,n) = [bn+I,I]2disc(,I, ,2, ... ,
,n) by the theorem of §1.30, and 0 < bn+I,1 < 1. Since d can
only be decreased a finite number of times, the algorithm must
terminate.
COROLLARY. Tbere is a basis WI, W2, .•. , Wn of Kover Q
witb tbe property tbat an element (3 of K is integral over Z
if and only if its representation (3 = E biWi in tbe basis bas
integer coefficients bi .
DEDUCTION: Apply the theorem to a set ,I, ,2, ... , ,,, as in
the lemma of §2.1.
Such a basis of Kover Q is called an integral basis of K.
§2.7 Proof of the Theorem of §2.4.
Let the notation be as in §2.4. As was seen in §1.32, V[pj,
the ring of values of K at [P] (see §1.23), is a vector space over
kp, the ring of values of Q at [Pl. The field kp is the field of
quotients of the integral domain Z mod p, which is Z mod p
itself, that is, the field F p with p elements.
LEMMA 1. Tbe image in V[p] of an integral basis of K is a
basis ofV[p] over Fp. In particular, K is p-regular (§1.32).
 §2.7 Proof of the Theorem of §2.4 71

PROOF: As was seen in the proof of Proposition 2, §1.32,

[V[pj= F p] ~ [K: Q], so the lemma will be proved if it is shown
that lp(WJ), lp(W2), ... , lp(W n ) are linearly independent over
F p , where lp is the natural map from elements of K finite at [P]
to V[pj, and WI, W2, ... , Wn is an integral basis of K. Let aI, a2,
... , an E Fp be such that allp(wI)+a2lp(w2)+·· ·+anlp(W n ) =
O. Then'Y = blWI + b2W2 + ... + bnw n is integral over Z and sat-
=
isfies lp(r) = 0, when bl , b2, ... , bn E Z satisfy bi ai modp.
Therefore, 'Y = ph where h is integral over Z. Because the
w's are an integral basis, h = CIWI + C2W2 + ... + CnW n where
Ci E Z. Because 'Y = ph, bi = PCi, so bi = 0 mod p. Since
ai = bi mod p, ai = 0 for all i, so the lp(Wi) are linearly inde-
pendent, as was to be shown.
Let [P] = pr p;2 ... p;p.be the factorization of [P] as a prod-
uct of powers of divisors prime in K. Lemma 1 combines with
Proposition 1 of §1.32 to give
p.
F(x) = II9i(x)e; modp
i=l

where F( x) = F( x, UI, U2, ... , u v ) is the homogeneous polyno-

mial of degree n = [K: Q] = [V[p]: F p] which is the determinant
of the matrix representing the action of multiplication by x - a
on V[p] relative to a basis of V[p] over Fp, and 9i(X) = 9i(X, UI,
U2, ... , u v ) is the homogeneous polynomial of degree [VPi: F p]
which is the determinant of the matrix representing the action
of multiplication by x - a on VPi relative to a basis of VPi over
F p. Since F p is a natural ring which is its own field of quo-
tients, and since VPi :J F p is a field extension (§1.24, (4)) of
finite degree ([Vpi:Fp] ~ [V[prFp] = n), 9i is simply the norm
of x - a when it is regarded as a polynomial with coefficients
in the field extension VPi :J F p (§ 1.1 7).
LEMMA 2. Let P be a divisor of [P] prime in K, let lp denote
the natural map from elements of K finite at P to the field
72 Divisor Theory

Vp ::> F p , let g(x) be the norm of lp(X -a), and let Ra absorb
division by p. Then 9 is irreducible over Fp and P = fp, g(a)].

(As above, fp, g(a)] means fp, g(a)] where 9 is a polynomial

with coefficients in Z congruent to 9 mod p.)
PROOF: Let 6 E K be integral over Z. Since at, a2, ... , a/l
generate Kover Q by assumption, 6 = ¢>(a)jb, where ¢>(a) E
Ra is a polynomial in the a's with coefficients in Z, and bE Z
is nonzero. IT plb, say b = pb', then ¢>(a)/p = b'6 is integral
over Z, and, by the assumption that Ra absorbs division by
p, ¢>(a)/p = 'I/J(a) ERa. Thus 6 = p'I/J(a)/b = 'I/J(a)/b', where
b' is divisible one less time by p than b is. Since this process
can be repeated, if necessary, there is no loss of generality
in assuming that p f b. Then lp(6) = lp(¢>(a))/lp(b), which
shows that the element of Vp represented by 6 is in the subfield
of Vp generated over Fp by lp(aI), lp(a2), ... , lp(a/l)'
Any element of Vp can be represented by an element 6 of K
finite at P, say [6] = B / C, where B and C are integral divisors
and P f C. By Theorem 2, there is a 62 E K, integral over Z,
such that [62 ] = CD, where D is integral and P f D. Then
6 = 61/62 , where 61 and 62 are integral over Z ([61 ] = [6][62 ] =
(BjC)CD = BD) and 62 is not zero at P. Thus lp(61 ) and
lp(62) are in the subfield of Vp generated over Fp by lp(aI),
lp(a2), ... , lp(a/l)' Moreover, lp(62) =f O. Therefore lp(6) =
lp(6I)/lp(62) is in this subfield. Since lp(t5) was arbitrary,
lp(aI), lp(a2), ... , lp(a/l) generate Vp over Fp.
By the theory of finite fields, Vp ::> F p is a normal extension.
(In fact, the Galois group is cyclic of order K, = [Vp : F p] and
is generated by the automorphism ( ~ (P of Vp over F p .)
Therefore, the norm of a polynomial with coefficients in Vp
is the product of its K, conjugates under the Galois group of
Vp ::> Fp. Thus 9 = TIs{x - Slp(a)), where S ranges over
the Galois group of Vp over Fp. Let h be a polynomial with
coefficients in F p which is not relatively prime to g. Then, as
a polynomial with coefficients in Vp ::> F p , h must be divisible
 §2.7 Proof of the Theorem of §2.4 73

by one of the factors x - S t p ( a) of 9 (unique factorization over

Vp). Then x- S' Stp( a) divides h for all S' in the Galois group
of Vp over Fp (h has coefficients in Fp). But these factors of
h are distinct because x - S'Stp(a) = x - S"Stp(a) implies
S'S and S" S have the same effect on the coefficients t p( ai)
of tp(a), and these coefficients generate Vp over F p , so S'S
and S" S can have the same effect on them only if S'S = S" S.
Thus h must be divisible by all '" factors of g, that is, glh,
which shows that 9 is irreducible over Fl!'
Let Q be a divisor of [p, g( a)] which IS prime in f{. Then
QI[P]' and, as was just shown, gQ = N(x-tQ(a)) is irreducible
over F p , where tQ(a) is the polynomial with coefficients in VQ
represented by a. On the other hand, Q divides g( a), which
is the remainder when g( x) is divided by x - a; therefore,
x - tQ(a) divides tQg = 9 as a polynomial with coefficients in
VQ. It follows that gQ = N(x - tQ(a)) divides NvQg, which is
a power of g. Since gQ and 9 are irreducible over F p and monic
in x, gQ = g. In particular, deg gQ = deg 9 = "', which implies
that [VQ: Fp] = '" = [Vp : Fp], and therefore that VQ and Vp
are isomorphic (they are both splitting fields of XP" - X over
Fp). Let B: Vp ---t VQ be an isomorphism. Application of B to
the equation 9 = rr(x - Stp(a)) gives 9 on the left (B(l) = 1,
so B is the identity on Fp) and gives rr(x - BStp(a)) on the
right. Since x - tQ(a) divides N(x - tQ(a)) = g, it follows
that x - lQ(a) = x - BSlp(a) for some automorphism S of
Vp. Thus, for this S, tQ( ai) = BSl p(ai) for all i. Therefore,
if </>( a) E f{ can be represented as a polynomial in the a's
with coefficients in Z, lQ(</>(a)) = BSlp(</>(a)). In particular,
tQ(</>(a)) = 0 if and only if lp(</>(a)) = 0, that is, QI</>(a) if
and only if PI</>( a ). If Q were not equal to P, there would be,
by Theorem 2, an element b of f{ with [b] = PC, where C is
integral and Q f C. As was shown above, b could be written in
the form b = </>(a)jb, where p f b, and [</>(a)] = [b][b] would be
divisible by P and not by Q, contrary to what was just shown.
Therefore, Q must be P, which shows that [p, g(a)] = pj for
74 Divisor Theory

some j 2: o. It remains to show that j = 1.

Since g( a) is the remainder when g( x) is di vi ded by x - a,
tpg(li) is the remainder when tpg(x) = I1(x - Stp(a)) is
divided by x - /,p( a), which is 0; that is, Plg( a). Therefore
j 2: 1, and it remains only to show that p 2 f [p, g( a)).
Suppose p 2 1[P). The lemma will be proved by showing that
in this case p 2 f g(a). Since p2lp, VP 2 is a vector space over
Fp. Since [Vp : Fp) = '" < 00, Vp is generated over Fp by a
single element of Vp (§1.25), say by /,p(, where ( E K is finite
at P. Then the images of 1, (, (2, ... , (,,-1 in Vp are a basis
of Vp over Fp. Let 'r/ E K be integral over Z and divisible
by P but not by p2. As in the proof of Proposition 1, §1.32,
the images of 1, (, (2, ... , (,,-1, 'r/, 'r/(, ... , 'r/(,,-1 in VP 2 are
a basis of VP 2 as a vector space over Fp. In order to have a
convenient description of VP 2 as a ring, it is helpful to adjust
the choice of ( in the following way.
The norm of x - /,p( (a polynomial with coefficients in
Vp :::> Fp) is a monic polynomial ep(x) of degree", in x with
coefficients in Fp of which /'p( is a root. Let (0 be the ele-
ment of VP 2 represented by (. Then ep((o) is a well-defined
element of VP2 :::> Fp , say ep((o) = ao +al(O + ... +a,,_I(;-1 +
bo'r/o + b1'r/o(0 + ... + b,,_I'r/O(;-I, where 'r/O is the element of
V P 2 represented by 'r/, and the a's and b's are in Fp. The image
of ep((o) under VP2 --t Vp is ep(/,pO = 0 on the one hand and
aO+al/'p( + .. ·+a,,_I/,p(,,-1 on the other, which implies the ai
are all zero, that is, ep( (0) = 'r/o(b o + b1(0 + ... + b,,-1 (;-1). For
any 8 E VP2, since 'r/5 = 0, ep((o + 'r/0 8) = ep((o) + 'r/08ep'((0) =
'r/o(b o + b1(0 + ... + b,,_I(;-1 + 8ep'((0)). The image of ep'((o)
in Vp is ep'(/,pO =1= 0 (since ep is irreducible over F p , it has
no multiple roots in Vp :::> Fp). Since Vp is a field, there
is a , in K finite at P such that /,pCl)ep'( /,pO = 1. Let
(1 = (0 - 'r/o,o 2:.7:01 bi(~ where ,0 is the element of VP 2 rep-
resented by,. Then ep((I) = 'r/O(2:.bi(~ -,o2:.bi(~ep'((o)) =
'r/o(l -,oep'((o)) 2:. bi(&. Since the image of 1 -,oep'((o) in
Vp is 1 - tp(,)ep'(tpO = 0, its expansion in the basis (~'r/~
 §2.7 Proof of the Theorem of §2.4 75

(0 S i < K, 0 S j < 2) of VP 2 over F p contains no terms with

j = 0; therefore, since 7]5 = 0, tP( (I) = O. Since the image of
(1 in Vp is lp(, which is a generator of Vp over F p, 1, (1, (f,
... , G- 1 , 7]0, 7]0(1, ... , 7]oG- 1 is a basis of VP2 over Fp. Mul-
tiplication in VP 2 relative to this basis is described by the two
relations tP( (d = 0 (which gives (f as a linear combination of
lower powers of (1) and 7]5 = O.
Let R denote the subring Fp[(1] of VP2 generated by 1 and
(1' An element of Vp 2 can be written in one and only one way
in the form S + 7] ot, where s, t E R. Let lp2CXi = Si + 7]oti
for i = 1, 2, ... , v and let 8 = L: SiUi and t = L: tiui, where
the Ui are indeterminates as before. Then lp2ii = 8 + 7]ot
and g(lp2ii) = g(8 + 7]00 = g(8) + 7]otg'(8) (because 7]5 = 0),
where g(x) = g(x, U1, U2, ... , U/I) is the norm of x - lpii,
a polynomial with coefficients in Fp C VP 2, where g'(x) is
its derivative with respect to x, and where g(8) and g'(8) are
the polynomials with coefficients in R obtained by setting 8 in
place of x in these polynomials. The image of g(8) in Vp is
g(lpii) (because lP2ii = 8 + 7] ot has the same image in Vp as
lPii and 8) which is 0 (by the definition of g). Therefore, since
g(8) is in R, g(lp2ii) = 7] otg'(8). If t were 0, lp20ti would be
Si E R for all i, and l pHp( Ot) would be in R for any ¢>( Ot) E
ROt; but 7] is integral over Z, so 7] = ¢>( Ot )/b where ¢>( Ot) E
ROt, and where b E Z is not divisible by p, which gives 7]0 =
lp27] = lp2(bc7]) = lp2(C¢>(Ot)) = lp2(C)lp2(¢>(Ot)), where C E Z
satisfies bc = 1 mod p. Since 7]0 rf. R, l p2 (c) E R, and R is a
field (isomorphic to Vp under VP2 - t Vp ), this shows that
l p2 ( ¢>( Ot )) rf. R and therefore that t =I- O. The image of g' (8) in
Vp is g'(lpii), which is not zero, because, as was shown above,
x -lpii is a simple factor of g(x). Therefore g'(8) =I- O. Since R
is a field, [g'(8) is a polynomial with coefficients in R which is
not 0, so 7]0 tg' (8) = g( l p2 ii) is a polynomial with coefficients in
VP 2 which is not O. Thus ~ - ii does not divide g( x) when both
are regarded as polynomials with coefficients in VP 2 J Fp. In
other words, p2 does not divide g( ii), as was to be shown.
76 Divisor Theory

PROOF OF THE THEOREM: As was shown above, F(x) =

rrf=19i(xY'modp. Since the 9i(X) are irreducible over Fp
by Lemma 2, and since they are monic in x, they can be found
by factoring F( x) mod p. Then the factorization [P] = rr Pt
can be found using the formula Pi = [p, 9i( 0)] of Lemma 2,
which is the assertion of the theorem.
§2.8 Dedekind's Discriminant Theorem.
As the lemma of §2.1 shows, there exist elements al, a2,
E K integral over Z such that ROt = Z[al, a2, ... , all]
... , all
includes all elements of K integral over Z. The discriminant
and the different of this ring are called the discriminant of the
field K and the different of the field K, denoted disc( K) and
dif(K), respectively.
THEOREM. (Dedekind) Let K be an algebraic number field.
A divisor P of [P] prime in K which divides [P] exactlye times
divides dif(K) at least e - 1 times. It divides dif(K) exactly
e - 1 times except possibly when pie.
PROOF: Since dif(K) = [F'(o)], where F(x) = N(x - 0)
and al, a2, ... , all are such that ROt includes all elements
of K integral over Z, the theorem deals with the number
of times a prime divisor P of [P] divides F'(o). As was
shown in §2.7, F(x) = rrf=l 9i(X)e, modp, where [P] =
rr Pt' is the prime factorization of [P] in K and 9i(X) is the
norm of x - 0 regarded as a polynomial with coefficients
in the field Vp, J Fp (ROt absorbs division by p). There-
fore, F'(x) = L:f=l ei9~(x)9i(X)-1 F(x) modp. Therefore
F'(o) = L:f=l ei9~(0)9i(0)-lF(0) mod P;l. Since P l I91(0),
p;l divides all terms on the right except the first, and p;l-l
divides the first. Therefore, p1el-1IF'(0). It remains to show
that, if p f el, then p1el does not divide F'(o), or, what is the
same, that p1el does not divide e19~ (0)91 (o)el-l rrf=2 9i( o)e,.
Since PI is prime, the multiplicity with which it divides a
product is the sum of the multiplicities with which it di-
 §2.9 Differents and Discriminants 77

vi des the factors. Since Pi = (p, 9i( a)] is not divisible by

PI, Ilf=29i(aYi is not divisible by Pl. Since N(PI) does
not divide N( eI) (p does not divide el), PI does not di-
vide el. Since x - LP1a is a simple factor of 9l(X) over VP1
(§2. 7), 9~ (a) is not divisible by Pl. If el = 1, it follows that
PI f e19~(a)9l(a)el-l ITi>l 9i(a)e i = e19i(a) Ili>l 9i(a)e i , as
desired. If el > 1, then Pllp and the equation PI = (p, 9l(a)]
implies PI divides 91 (a) exactly once and therefore divides
91 (a )e 1 -1 exactly el - 1 times, and the theorem is proved.

COROLLARY. A prime p E Z ramifies in K if and only if (p,

dif(K)] i- [1].
DEDUCTION: If p ramifies in K, then some divisor P of (P]
prime in K divides [p] exactly e times where e > 1. By the
theorem, pe-l divides both [P] and dif(K), so [p, dif(K)] i- [1].
Conversely, if (p, dif(K)] i- [1] then some divisor P of (P] prime
in K also divides dif(K). Let e be the multiplicity with which
P divides p. If pie, then e > 1, so p ramifies in K. If p f e,
then, by the theorem, e - 1 > 0, so p ramifies in K.
§2.9 Differents and Discriminants.

THEOREM (Dedekind*). For algebraic number fields K,

the discriminant is the norm of the different, disc( K) =
N( dif( K)).
PROOF: Let aI, a2, ... , an be an integral basis of K. Then
Ra is the ring of all elements of K integral over Z and dif(K) =
difK(R a ), disc(K) = discK(R a ). Since aI, a2, ... , an span
Ra over Z, discK(Ra) is by definition [det(USU t )], where U is
an n X n matrix of indeterminates and S is the n x n symmetric

*N(dif(K)) = IT..<). [a(j) -a(i)]2 is in essence what Kronecker called "the

discriminant of the fundamental equation". Kronecker conjectured that
it was equal to disc(K), and he proved partial results in this direction
[Krl, §25)' but he admitted he could not prove it for general T. Probably
he could prove the case T = Z stated here, but he never published a proof.
78 Divisor Theory

matrix of integers (tr(aiaj)). Since [det(U)) = [1), disc(K) =

[det(S)).
On the other hand, as was seen in §1.27, difK(Ra) = Il[a-
a(i»), where the product is over all n - 1 conjugates a(i) of
a, other than a itself, in a normal extension of Q containing
K. Therefore, N( difK(Ra )) = Ili<j[a U) - a(i)J2 (Proposition
2, §1.25-see also the proposition of §1.30). This divisor is
the divisor represented by the square of the determinant of
the matrix A whose columns are the n conjugates of 1, a, a2 ,
... , an-I. Thus N( dif(K)) = [det(AAt)) for this A. Now
each a i is of the form <Pil al + <Pi2a2 + ... + <Pin an, where
the <Pij are polynomials in UI, U2, ... , Un (homogeneous of
degree i) with coefficients in Z. Thus, A = if!B, where if! is
the n x n matrix (<Pij) and B is the matrix whose columns
are the n conjugates of aI, a2, ... , an. The entry in the ith
row and the jth column of BBt is E;=l a~p)a;p) = tr(aiaj),
that is, BBt = S. Thus, N(dif(K)) = [det(if!BBtif!t)) =
[det(S)(detif!)2) = disc(K)[detif!)2. It is to be shown that
[detif!) = [1], that is, detif! is primitive.
If det if! were not primitive, there would be a prime p such
=
that det if! 0 mod p. There would then be a row matrix v =
(VI(U), V2(U), ... , vn(u)) of polynomials in UI, U2, ... , Un with
coefficients in Z, such that vif! = 0 mod p, but v 1= 0 mod p.
Then v times the column matrix 1, a, a 2, ... , a n- l would
be vif! times the column matrix aI, a2, ... , an, so VI(U) +
v2(u)a + ... + vn(u)a n- l = 0 modp. Let G(x) = VI(U) +
V2(U)X + ... + vn(u)X n- l . Then G(x) 1= 0 modp but G(a)
would be 0 mod p. That this is impossible, and therefore that
det if! is primitive, follows from:
LEMMA (Hensel). Let G( x) be a polynomial in x, UI, U2, ... ,
Un with coefficients in Z such that G(a) = 0 modp. Then
G(x) is divisible by F(x) = N(x - a) mod p. In particular, if
G 1= 0 mod p then deg x G ~ deg x F = n.
PROOF: Let G == h~l h~2 ... h~tr mod p be the decomposition of
 §2.10 Cyclotomic Fields 79

G as a product of powers of distinct, irreducible polynomials

with coefficients in F p, monic in x. Since G( ii) - 0 mod PI
and PI is prime, hi ( ii) - 0 mod PI for at least one i. As
a polynomial with coefficients in VPl l hi(x) therefore has the
common factor x - lPi ii with gl = N(x - lPi ii). Then, since
hi(x) has coefficients in F p , it is divisible by all deggl distinct
linear factors of gl, so hi ( x) is divisible by gl (x). Since hi ( x)
is irreducible over Fp and monic in x, hi(x) = gl(X). Thus,
exactly one of the hi(x) is equal to gl(X), and PI f hj(ii) for
j =f=. i. Since plei divides [P] and [P] divides G( ii), and since
PI divides hi(ii) once and hj(ii) not at all for j =I i, ti ~ el,
that is, g;i divides G( x). Since the same is true of all prime
power factors of F(x) = gl(X)e i g2 (x)e 2 ···gJL(x)el', it follows
that F(x)IG(x) modp, as was to be shown.
COROLLARY. A prime p ramifies in K if and only if
[P] I disc( K) .
DEDUCTION: See the Corollary of §2.8.
§2.10 Cyclotomic Fields.
For positive integers n, let An be the splitting field of the
polynomial x n - lover Q. These fields are called cyclotomic
fields. A cyclotomic field An contains a primitive nth root of
unity-that is, an element a which satisfies an = 1 but does
not satisfy am = 1 for 0 < m < n-and An is generated over
Q by any such a. An element of the Galois group of An over
Q obviously carries a primitive nth root of unity a to another
primitive nth root of unity. Since the powers of a primitive nth
root of unity a, a 2 , •.• , an = 1 are n distinct roots of xn - 1,
every nth root of unity in An is a power of a. Therefore, an
element of the Galois group of An = Q( a) over Q carries a
to a j for some integer j, and j determines the automorphism.
Since the composition of a 1--+ a j and a 1--+ a i is a 1--+ a ij ,
and since a j depends only on the class of j mod n, the Galois
group of An over Q is isomorphic in this way to a subgroup
80 Divisor Theory

of the multiplicative group of invertible elements in the ring

of integers mod n. By a basic theorem (see, for example, [E3,
§70]) the Galois group of An over Q is the entire group of
invertible elements of the ring of integers mod n.
Since the Galois group of An over Q is commutative,
every subfield K of An is normal, and there is a natural
homomorphism-namely, the restriction mapping-from the
Galois group of An over Q to the Galois group of Kover Q.
THEOREM. Let K be a subfield of a cyclotomic field An. For
any prime integer p which does not divide n, (P] is a product
of distinct divisors PI P2 ••• PIL prime in K. An element of the
Galois group of Kover Q leaves PI fixed (and therefore leaves
Pi nxed for all i) if and only if it is a power of the restriction
to K of the automorphism a ~ a P of An (a a primitive nth
root of unity).

PROOF: Let a be a primitive nth root of unity in An and let

<pn(x) = NAn(x - a). Since an = 1, xn - 1 = <pn(x)Q(x),
where Q is a polynomial with coefficients in Z. Then nx n - I =
<p~(x)Q(x) + <pn(x)Q'(x), from which it follows that na n - I =
<P~ (a )Q( a ), n = <P~ ( a )aQ( a ). Therefore, <P~ (a) divides nand
N[q>~( a)] divides N[n], which is a power of [n]. IT p ramified in
K, then it would ramify in An, and (p, <P~ (a)] would not be [1]
(§2.8). The norm of (p, q>~(a)] would therefore be of the form
(P]j for j > 0, and it would divide a power of [n], contrary to
the assumption p f n. Thus, (P] = PI P2 ••• PIL , where the Pi
are the distinct conjugates under the Galois group of Kover
Q of anyone divisor of (P] prime in K (§1.22).
Let ~ be an element of the Galois group of Kover Q
for which ~(PJ) = Pl. Then there is an element e of
the Galois group of An over Q whose restriction to K is ~.
(The restriction map is a homomorphism from a group with
[An: Q] = [An: K][K: Q] elements to a group with [K: Q] ele-
ments whose kernel has [An: K] elements. Therefore it is onto.)
Let PI = Pi P~ ... P; be the factorization of PI into divisors
 §2.10 Cyclotomic Fields 81

prime in An. By the proposition of §1.22, the Galois group

of An over K acts transitively on the factors PI of Pl. Since
e'(PI) = PI, e(p{) = PI for some i = 1, 2, ... ,0', so there is a
o in the Galois group of An over K such that O(P{) = e'(P{).
Then the element "p = O-Ie' of the Galois group of An over
Q satisfies "p( P{) = P{ and its restriction to K is the given
e
element of the Galois group of Kover Q. It will suffice to
show that an element "p of the Galois group of An over Q leaves
fixed a divisor P' of [P] prime in An if and only if it is a power
of a ~ a P.
If "p(P') = P', where P' is prime in An, then "p induces
an automorphism of the field VP' (the ring of values of An
at P'). An automorphism of the finite field VP' is necessar-
ily a power of the automorphism of raising to the power p.
Therefore, "j;(I,pl(a)) = I, pI (a)pi , for some positive integer i,
where I, pI ( a) is the element of VpI represented by a and "j;
denotes the automorphism of Vp' induced by"p. By defini-
tion, "j; (I, pI (a)) = I, pI ("p( a)). Because I, pI is a homomorphism,
I, pI (a)pi = I,pl(aP \ Therefore, I,pl ("p(a)) = I,pl(api ), which
implies I,pl ("p(a) - a pi ) = 0, which is to say P'\("p(a) _ a pi ).
Since "p (a) = a i for some integer j, P'\ (a i - a P i ). If j were not
congruent to pi mod n, then P' would divide a i - a k , where
a i and a k were distinct roots of <I>n(x); then some conjugate
of P' would divide a - as, where as =I a was a root of <I>n(x),
and consequently would divide <I>~ (a) = nt( a - at) where at
ranges over the roots of <I>n( x) other than a, which in turn
divides n, contrary to [p, n] = [1]. Therefore, j = pi mod n
and "p has the same effect on a as the ith power of a ~ a P •
Since a generates An over Q, "p is therefore the ith power of
a~aP.

It remains only to show that the element, call it 17, of the

Galois group of An over Q which carries a to a P leaves fixed
all divisors P' of [P] prime in An (which implies, of course, that
its restriction to J{ leaves fixed all divisors P of [P] prime in
82 Divisor Theory

K). Let P' be written in the form P' = fp,,8] (Corollary (1),
§1.20). Since,8 is integral over Z and since Ra absorbs division
by p, b,8 = <p( 0;) ERa, where b E Z is not divisible by p. By
Fermat's theorem, <p( o;)P - <p( o;P) is of the form p . h( 0;), where
h( 0;) E Ra. Therefore, P' divides <p( o;)P - <p( o;P) = (b,8)P -
1](b,8) = bP,8P - b1](,8). Since P'I,8, it follows that P'lb1](,8),
and, since P' is prime and does not divide b, that P'I1](,8).
Thus, P' divides fp,1](,8)] = 1](P'). Since 1](P') is prime in
An and therefore irreducible in An, P' = 1](P'), as was to be
shown.
§2.11.
COROLLARY 1. Let K be a subfield of the cyclotomic field An.
The form of the factorization fp] = (PI P2 ••• PIl )1I of fp] in K
depends only on the class of p mod n. If p f n, then v = 1 and
J1 = [K: Q]/ A, where A is the least positive integer for which
0; f---+ o;pA is the identity on K.
DEDUCTION: If pin thenp is the only prime in its class mod n,
and there is nothing to prove. If p f n, then, by the theorem,
fp] is a product of distinct prime factors, and the number J1 of
these factors describes the form of the factorization of fp]. Since
A depends only on the class of p mod n, the second statement
of the corollary implies the first. By the proposition of §1.22,
the Galois group of K acts transitively on the Pi. Since this
group has [K: Q] elements, the second statement says simply
that A is the order of the subgroup leaving PI fixed, which is
immediate from the theorem.
COROLLARY 2. (Kummer's definition of ideal prime factors of
cyclotomic integers [K u, p. 322]) Let p, n be positive integers,
with p prime, n > 2, p f n. Let K be the subfield of An in-
variant under the automorphism 0; f---+ o;p. There is an element
.,p E K whose norm, as an element of K :::> Q, is divisible by
p but not by p2. For any such .,p, let .,pI, .,p2, ... , .,pe, where
e = [K: Q], be its conjugates in K, and let Wi = TIj#i .,pj.
 §2.12 Quadratic Reciprocity 83

Then [P] = IT~=I [p, ¢i], the divisors [p, ¢i] are prime in An,
and [p, ¢d l1 divides j( ex) E An if and only if pl1lj( ex )Wr.
DEDUCTION: By the choice of K, A = 1 in Corollary 1. There-
fore, lP] is a product of [K: Q] = e distinct divisors prime in
K, say lP] = p}pz ··· P e , and the Pi are conjugate under the
Galois group of K ::) Q. By Theorem 2, PI = [p, ¢] for some
¢ E K. When Corollary 1 is applied with An in place of K,
A is the least positive integer, Kummer denotes it by j, for
which pI _ 1 mod n. In other words, j is the order of the sub-
group of the Galois group of An ::) K generated by ex f---t ex p •
This subgroup corresponds to the subfield K of Corollary 2,
so j = [An: K] by basic Galois theory. It follows from Corol-
lary 1 that [p] is a product of [A: Q]/[An: K] = [K: Q] = e
divisors prime in An. Therefore, the factors of lP] = ITi[P, ¢i]
are prime in An and the remaining statements of the corollary
follow easily.
§2.12 Quadratic Reciprocity.
Let q be a prime integer greater than 2, and let Aq be the
splitting field of x q - lover Q. Since the Galois group G of
Aq over Q is cyclic of order q -1 (that is, there ist a primitive
root mod q), Aq contains a unique quadratic extension K of
Q, namely, the subfield of Aq corresponding to the subgroup
of squares in G. By the Corollary of §2.11, the factorization
of [P] in K for p a prime, p ::J q, contains two factors if p is a
square mod q (so A = 1); otherwise, [p] is prime in K.

tBy Fermat's theorem, x q - x has q distinct roots in the field F q • For

any proper divisor d of q - 1, x q- 1 -1 = <l>q_1(X)(X d - I)Q(x), where Q
is a polynomial with coefficients in Z and <I> q -1 = N Aq -1 (x - Q') is the
polynomial with coefficients in Z of which primitive (q -1)st roots of unity
are roots. (x d - 1 divides x q- 1 - 1 and is relatively prime to <l>q_1(X).)
Since x q - x = x<l>q-1 (x )(x d -1)Q(x) has q distinct roots in F q, <l>q-1 (x)
and x d - 1 have no roots in common. Since d was arbitrary, the roots of
<l>q-1 (x) in F q are primitive (q - l)st roots of unity. Therefore, there are
elements of the multiplicative group of F q of order q - 1.
84 Divisor Theory

On the other hand, K has the form K = Q( Vd) for some

squarefree dE Z, and the factorization of (P] in K can be deter-
mined by the method of §2.5. Specificallyt, K = Q( vq) when
q = 1 mod 4 and K = Q(J=fJ) when q =
3 mod 4. Let q*
=
denote q when q 1 mod 4 and -q when q 3 mod 4. Then, =
for P =f 2, (P] is a product of two divisors in K = Q( #) when
q* is a square mod p, and is prime in K otherwise. Therefore:
If p and q are primes distinct from one another and distinct
from 2, then p is a square mod q if and only if q* zs a square
mod p, where q* is defined as above.

tGauss, D. A., §356. Let 00 = L:O'i, where j runs over all nonzero
squares mod q and let 01 = L: O'i , where j runs over nonsquares. Then
1 + 00 + 01 = 1 + 0' + 0'2 + ... + O'q-l, which is 0 because z - 0' divides
=
zq -1 but not z -1, so it divides (zq -l)/(z -1) zq-l + zq-2 + ... + 1.
Both 00 and 01 are elements of K integral over Z, so 0001 is integral
over Z. 00 01 is invariant under the Galois group and is therefore in
Q. Thus, 0001 E Z. Both 00 and 01 are sums of (q - 1)/2 terms, so,
before terms are combined, 0001 is a sum of (q - 1)2/4 terms, each a
power of 0'. If 0'-1 is a term of {;Io-that is, if -1 is a square mod
q-then 1 does not occur among the (q - 1)2/4 terms of BoB!, and, by
symmetry, each of the (q - 1) other powers of 0' occurs (q - 1)/4 times.
Therefore, 0001 = =
(B o+Bt)(q-1)/4 (1-q)/4. In particular, in this case
q == 1 mod 4. If 0'-1 is a term of(lt, then 1 occurs (q -1)/2 times in OOBl,
and the remaining (q-1)2/4)-(q-1)/2 = (q2-4q+3)/4 = (q-1)(q-3)/4
terms of BOOI are evenly distributed among the other q - 1 powers of 0'.
Thus, BOOI = =
(q-1)/2)+(Oo+Ot)(q-3)/4 (2q-2-q+3)/4 (q+1)/4. =
In particular, in this case q == -1 mod 4. Thus, (z - Oo)(z - Bt) =
=
z2 + z + BoB 1 (2z + 1)2 + 400Bl - 1)/4 = (2z + 1)2 - q*)/4, where
= =
q* q if q == 1 mod 4 and q* -q if q == 3 mod 4. Since Q( H) and K
are both splitting fields of (z - Oo)(z - Bd, they are isomorphic.
Part 3: Applications to the Theory of Algebraic Curves

§3.1 Function Fields.

Let Q [x] denote the ring of polynomials in an indeterminate
x with coefficients in the field Q of rational numbers. An
algebraic function field in one variable over Q is an extension
of Q[x] of finite degree. For short, such fields will be called
function fields. Note that function fields are to Q[x] what
algebraic number fields are to Z. Since Q[x] is a natural ring
(§1.2), divisor theory applies* to function fields.
The field of quotients Q(x) of Q[x] is called "the field of
rational functions in x with coefficients in Q." In the 19th
century, it was usual to call the elements of a function field
"functions", but in the 20th century the word ''function'' has
become so firmly fixed in its set-theoretic meaning that to use
it in any other way risks confusion and misunderstanding. For
this reason, the word function will be used here only in the
phrase "function field" and, occasionally, the phraset "rational
function" .
§3.2 Parameters and Constants.
Let u be an element of a function field K :) Q(x). Since u
is algebraic over Q(x), ao(x)un+al(x)u n- l +···+an(x) = 0,
where the ai(x) are in Q[x] and ao(x) I- o. Moreover, as was

*Function fields are also extensions of finite degree of the natural ring
Z[x]; consequently, divisor theory applies in another way to them. These
two divisor theories are different-for example, integers greater than 1
are units in Q[x] but not in Z[x]. Only the theory stemming from Q[x]
will be considered.
tA "rational function" in indeterminates Xl, X2, •.. , Xn is an element of
the quotient field of the natural ring Q[XI, X2, ..• , Xn]. In the case of
one indeterminate, such a quotient of polynomials can be regarded as a
function of a complex variable in the set-theoretic sense, provided 00 is
admitted as an element of the range. This interpretation depends on the
fact that no complex number is simultaneously a root of two relatively
prime polynomials in one indeterminate.
86 Divisor Theory

seen in §1.9, one can assume without loss of generality that

h(X) = ao(x)xn + al(X)xn-1 + ... + an(x) is irreducible as a
polynomial in X with coefficients in Q[x], in which case h(X)
is determined up to multiplication by a unit of Q[x], that is,
up to multiplication by a nonzero rational number. If the ai( x)
are all independent of x, u is called a constant of K. Otherwise,
u is a parameter of K.
IT u is a parameter of K, the homomorphism s from the
natural ring Q[X] to K defined by s(X) = u and s( c) = c for
cEQ is one-to-one. (If g(X) E Q[X] satisfied g(u) = 0, then
h(X) would divide g(X) as a polynomial with coefficients in
Q[x] :) Q. Since 9 can be assumed to be monic, so can hand
9 / h. Then the theorem of Part 0 implies that the coefficients
of h are roots of polynomials with coefficients in Q, which im-
plies that the coefficients of h have degree 0 in x, contrary to
assumption.) Thus, the image of s is a subring of K isomor-
phic to the natural ring Q[X]. Let Q[u] denote the image of
s. The field of quotients Q( u) of Q[u] is a subfield of K iso-
morphic to the field of rational functions in one indeterminate.
The field extension K :) Q( u) is of finite degree, because the
relation ao(x)u n + al(x)u n- l + ... + an(x) = 0 can be rewrit-
ten in the form bo( u )xm + bl (u )x m - l + ... + bm(u) = 0, where
m > 0 and bo =1= 0 (when u is not a constant of K), which
shows that x is algebraic over Q(u); thus, [Q(x,u):Q(u)] <
00, which gives [K:Q(u)] = [K:Q(x,u)][Q(x,u):Q(u)] =
[K:Q(x)][Q(x,u):Q(X)]-l[Q(x,U):Q(u)] ~ [K:Q(x)]. 1 .
[Q(x,u):Q(u)] < 00.
In short, if u is any parameter of K, then Q(u) is a sub-
field of K isomorphic to the field of rational functions in one
indeterminate, and K :) Q( u) is an extension of finite degree.
§3.3.
Let K be a function field, and let u be a parameter of K. As
an extension of finite degree of the natural ring Q[u], K has a
divisor theory. However, this divisor theory in K depends on
 §3.4 Global Divisors 87

the choice of the parameter u. It will be shown in the next

article that there is a natural way to define a global divisor
theory-one not dependent on the choice of the parameter-
using the following lemma.
Notation: Let K be a function field, u a parameter in K,
and f and 9 polynomials with coefficients in K. Let (f]ul[g]u
mean that, in the divisor theory of K as an extension of the
natural ring Q[u], the divisor represented by f divides the
divisor represented by g.
LEMMA. If x and u are parameters of a function field K, and
if x is integral over Q [u], then [f] x I[g] x implies [f] ul [g] u .
PROOF: (f]xl[g]x means fq = g7r, where q has coefficients in
K integral over Q[x] and 7r has coefficients VI, V2, ... , Vn in
Q[x] of which 1 is a g. c. d. By the Euclidean algorithm in Q[x],
there exist aI, a2, ... , an in Q[x] such that aivi +a2v2 + ... +
anV n = 1. In the divisor theory of the extension K :J Q[u],
[7r]ul[l]u because [7r]ul[vi]u ((9) of §1.11) and [vi]ul[aivi]u (from
the definition, because ai is a sum of elements of K integral
over Q[u]) which implies [7r]ul[aivi]u ((3) of §1.8), which in
turn implies [7r]ul[alvl +a2v2+" ·+anvn]u = [l]u ((1) of§1.8).
Since the coefficients Vi of 7r are in Q[x], they are integral over
Q[x], which implies:j: that they are integral over Q[u], that is,
[l]ul[7r]u. Thus, [7r]u = [l]u, and [f]~l[g]u = (f]~I[g7r]u =
(f]~l (fq]u = [q]u. Since the coefficients of q are integral over
Q [x], they are integral over Q [u], [q] u is integral, and [f] ul[g] u,
as was to be shown.
§3.4 Global Divisors.
Let K be a function field, let u and x be parameters of K,
and let 'Du and'Dx denote the groups of nonzero divisors in K
as an extension of the natural rings Q[u] and Q[x] respectively.
The lemma of §3.3 implies that, if x is integral over Q[u],

:j:See Proposition 2 of Part o.

88 Divisor Theory

there is a natural onto homomorphism Vx - t V u, namely, the

homomorphism which sends (f]x E Vx to [flu E Vu. This
homomorphism will be called restriction from x to u.
DEFINITION. A global divisor in a function field K is an as-
signment A, to each parameter u of K, of a nonzero divisor
Au in V u, called the restriction of A to u, in such a way that
Au E Vu is the restriction from x to u of Ax E Vx whenever
this restriction is defined, that is, whenever x is integral over
Q[u].
(Note that, according to this definition, zero is not a global
divisor. One might also define a zero global divisor, but then
the global divisors would not be a group.)
If f is a nonzero polynomial with coefficients in K, then [!]u
represents an element of Vu for each parameter u of K. By the
lemma of §3.3, u ~ [flu is a global divisor in K. This global
divisor will be denoted (f].
It was shown in §1.23 that any Au in Vu can be written in
one and only one way as a quotient Au = Bu/Cu, 'in which
Bu and C u are relatively prime integral divisors in Vu. Let
A be a global divisor in K, and let B and C assign to each
parameter u of K the relatively prime integral divisors Bu
and Cu, respectively, such that Au = Bu/Cu' Then B and
C are global divisors, as can be seen as follows. Let x and u
be parameters of K, with x integral over Q[u], and let t(Bx)
and t( Cx ) denote the restrictions of these divisors from x to
u. Then t(Bx) and t(Cx ) are integral (by the lemma, [l]xIBx
implies [1] u It( B x) and similarly [1] u It( C x)) and relatively prime
([Bx, Cx]xl[l]x implies [t(B x), t(Cx)]ul[l]u) and have quotient
Au (AxC x = Bx implies Aut(Cx) = t(Bx) because A is a
global divisor and t is a homomorphism). In short, Bu = t(Bx)
and Cu = t(Cx ), as was to be shown. The global divisors B
and C defined in this way will be called the numerator and
denominator of A, respectively.
 §3.5 Global Divisors 89

DEFINITION. A global divisor is integral if its denominator is

[1], or, what is the same, if its restriction to u is integral for
all parameters u of K.
Global divisors in K obviously form a commutative group.
Let 'DK denote this group. The integral global divisors are a
subset of'DK closed under multiplication, and every element
of'DK can be written as a quotient of integral elements. One
global divisor will be said to divide another if their quotient
is integral. Clearly, two global divisors are equal if and only if
each divides the other.
If A, B, ... , E are global divisors in K, let [A, B, ... , E] as-
sign to each parameter u of K the divisor [Au, B u, ... , Eu]u in
'Du. Then [A, B, ... , E] is a global divisor in K (the restriction
of a g.c.d. is the g.c.d. of the restrictions) which divides A, B,
... , E and is divisible by any global divisor which divides A, B,
... , E. Thus, [A, B, ... , E] is the greatest common divisor of
A, B, ... , E. In particular, [I] is the greatest common divisor
of the global divisors [a], where a runs over the coefficients of
I. (However, not every global divisor can be expressed in the
form [I] where 1 is a polynomial with coefficients in K-see
§3.23.)
§3.5.
In the remainder of this book, "divisor" will mean global
divisor in a function field.
PROPOSITION. Let A be a divisor in a function field K and
let x be a parameter of K. If the restrictions Ax and A 1 / x of
A to x and l/x are both integral, then A is integral.
COROLLARY. If A and B are divisors in K, if x is a parameter
of K, if Ax = B x, and if A 1 / x = B 1 / x, then A = B.
DEDUCTION: Since AxlBx and AI/xIBI / x , both A;IBx and
A;/xBI/x are integral, so A-I B is integral by the proposition,
that is, A divides B. In the same way, B divides A, so A = B.
90 Divisor Theory

LEMMA. Let x and y be parameters in a function field K.

There is a nonzero polynomial F(X, Y) E Q[X, Y] in two
indeterminates such that F(x, y) = 0 in K. If F(X, Y) =
ao(y)xn + al(y)xn-l + ... + an(Y) is such a polynomial,
and if A is a divisor in K for which Ax is integral, then the
denominator of Ay divides [ao(y)i]y for all sufficiently large
integers j.

PROOF: The existence of F(X, Y) was shown in §3.2 (y is al-

gebraic over Q[x]).
Let z E K be defined by z = y+(l/ao(Y)), that is, yao(y)-
zao(y) + 1 = O. Then y is integral over Q[z], because the
coefficient of the highest power of y in yao(y) - zao(Y) + 1 is
independent of z (it is the coefficient of the leading term of
ao(Y)). Since elements of K integral over Q[z] form a ring,
both ao(Y) and ao(y)-l = z - yare integral over Q[z]. Thus,
both [ao(y)]z and [l/(ao(y))]z are integral (in V z ), which is to
say that [ao(Y)]z = [l]z. On the other hand, multiplication of
ao(y)x n+ ... = 0 by ao(y)n-l shows that ao(Y)x is integral over
Q[y], that is, [1] yI[ao (y )x] y' Since y is integral over Q[z], the
lemma of §3.3 implies that [l]zl[ao(y)x]z = [ao(y)]z[x]z = [x]z,
that is, x is integral over Q[z]. Therefore, Az is the restriction
to z of Ax. Since Ax is integral by assumption, Az is integral.
Let f be a polynomial with coefficients in K such that
Ay = [fly. Since y is integral over Q[z], [f]z is AZl which
was just shown to be integral. Therefore, every coefficient
a of f is integral over Q [z], which is to say that a sat-
isfies a relation of the form all = L::f=l Ci( z )a ll - i , where
Ci(Z) E Q[z]. Because ao(Y)z = ao(Y)Y + 1 can be ex-
pressed as a polynomial in y, ao(y)kci(Z) can be expressed
as a polynomial in y whenever k ~ deg Ci. The relation
(ao(y)ia)1l = L::f=l ci(z)ao(y)ii(ao(y)ia)ll-i then shows that
ao(y)ia is integral over Q[y] for all sufficiently large j. (Specif-
ically, if ij ~ deg Ci for i = 1, 2, ... , /-l, then ao (y)i a is inte-
gral over Q [y]. ) Since this is true for each coefficient a of f 1
 §3.5 Global Divisors 91

[ao(y)i fly is integral for all sufficiently largej. But [fly = A y.

Therefore, [ao(y )]~Ay is integral for all sufficiently large j, that
is, the denominator of Ay divides [ao(y)]~ for large j, as was
to be shown.
PROOF OF THE PROPOSITION: Let A be a divisor in K such
that Ax and A I / x are integral. It is to be shown that Ay is
integral for all parameters y of K.
Consider first the special case in which y = (x - c) - 1, where
cEQ. Then yx - cy - 1 = 0 and, by the lemma, the de-
nominator of Ay divides [yi]y for all sufficiently large j. On
the other hand, (cy + 1 )x- I - Y = 0, and, by the lemma, the
denominator of Ay divides [( cy + 1 )i]y for all sufficiently large
j (because A I / x is integral). But, for any c, [y, cy + I]y = [I]y,
which implies [yi, (cy + 1 )i]y = [I]y for all j > O. Therefore,
the denominator of Ay divides [I]y, that is, Ay is integral, as
was to be shown.
Now let y be an arbitrary parameter of K. Let ao(y)x n +
al(y)x n- 1 + .. ·+an(y) = 0, where ai(y) E Q[y] and ao(Y) =1= O.
Since this relation can be divided by any common factor of the
ai(y), one can assume without loss of generality that [ao(y),
al (y), ... , an(y)]y = [I]y. Let v = (x - c)-I, where cEQ. As
was just shown, Av is integral. A relation satisfied by v and y
can be found by setting x = c + v-I in L ai(y)x n- i = 0 and
multiplying by vn , which gives (ao(y)c n + al(y)cn- 1 + ... +
an(y))v n + (terms of lower degree in v) = o. By the lemma,
the denominator of Ay divides [ao(y)c n + al(y)c n- 1 + ... +
an(Y)]~ for any cEQ, provided j is sufficiently large. Let
bi(y) = ao(y)in+al(y)i n- I + .. ·+an(y) for i = 1, 2, ... , n+l.
Since the (n + 1) x (n + 1) matrix of coefficients (ill) (1 ::; i ::;
n + 1, 0 ::; Ii ::; n) has nonzero determinant (a Vandermonde
determinant), the ai(y) can be written as linear combinations
of the bi(y) with rational coefficients, and [bo(Y), bl(y), ... ,
bn(y)]y = [ao(Y), al(y), ... , an(y)]y = [I]y. Therefore, [bo(y)i,
bl(y)i, ... , bn(y)i]y = [I]y for allj and the denominator of Ay
 92 Divisor Theory

divides [1]y, as was to be shown.

§3.6.
DEFINITIONS. Let B be an integral divisor in a function field
K. A divisor A avoids B if both numerator and denominator
of A are relatively prime to B. A divisor A is concentrated at
B if both numerator and denominator of A divide some power
of B.
PROPOSITION. Let B be an integral divisor in a function field
K. Any divisor in K can be written, in just one way, as a
product of a divisor which avoids B and a divisor concentrated
at B. In otber words, tbe group of divisors is tbe direct product
of tbe subgroup of divisors wbicb avoid B and tbe subgroup of
divisors concentrated at B.
PROOF: Consider first the case in which the given divisor A
is integral. Let eW = A/[A, Bi]. Then e(j) is an integral
divisor, [A, Bi] is an integral divisor concentrated at B, and
e
A is their product. It will be shown that W is relatively
prime to B for j sufficiently large, that is, [e W , B] = [1] for
j sufficiently large. Let x be a parameter of K. By Theo-
rem 1, Ax and Bx can be written as products of (positive)
powers of relatively prime integral divisors E~i) in V x , say
Ax = n(E~i)ri and Bx = n(E~i)ri. The exponent of E~i)
in e~j) is then O"i - min(O"i,j'Ti). Let jl be large enough that
jl'Ti 2: O"i whenever 'Ti > o. Then j > jl implies that the ex-
ponent of E~i) in e~i) is zero whenever 'Ti > o. Thus, e~i) is
a product of divisors E~i) relatively prime to Bx, which im-
plies [e~j), Bx]x = [1]x. In the same way, there is an integer
h such that j > j2 implies [e~~~, Bl/xh/x = [lh/x. By the
proposition of §3.5, j > max(il,j2) implies [e W , B] = [1].
e
Therefore, A = W [A, Bi] is a factorization of the desired
type when j > max(jbh). This factorization is unique, be-
cause if A = A(1) A(2), where A(l) is relatively prime to Band
 §3.7 Global Divisors 93

A(2)IB k for some k, then [A,Bi] = [A(2),Bi] = A(2) when-

ever j ~ k, which shows that the factorization A = A(I)A(2)
coincides with the factorization found above.
To write an arbitrary divisor as a product of a divisor which
avoids B and a divisor which is concentrated at B, apply the
above construction to its numerator and denominator.

§3.7.
PROPOSITION. The restriction map from VK to Vx has as its
kernel the divisors concentrated at the denominator of [x].

PROOF: Let E E VK be the denominator of [x]. Let A = B/C

be a divisor in K, where Band C are relatively prime integral
divisors. If A is concentrated at E, then Band C divide Ei for
large j. Therefore, Bx and C x divide E~ for large j. Since [x]x
is integral, Ex = [1]x. Therefore, Bx and Cx, being integral
divisors which divide [1]t must both be [1]x, that is, A must
be in the kernel of V K --+ V x.
Conversely, if Ax = [1]x, then Bx/Cx = [1]x, where Bx and
C x are integral and relatively prime. Therefore, Bx = C x =
[1]x. Let y = l/x, that is, yx - 1 = O. Since B-1 and C- l
have integral restrictions to x, the denominators of B;;1 and
C;;1 divide [yJi for j sufficiently large (Lemma, §3.5). But By
and C y are the denominators of B;;1 and C;;I, respectively,
and Ey = [Y]y. Thus, By and C y divide Et for large j. Since
B x and C x also divide E~ for large j (all three are [1] x), Band
C are concentrated at E by the proposition of §3.5, as was to
be shown.

DEFINITION. A parameter x in a function field K supports a

divisor A in K if A avoids the denominator of [x].

COROLLARY. The restriction map V K --+ Vx is an isomor-

phism from the subgroup ofVK of divisors supported by x to
V x . In other words, any Ax E Vx is the restriction to x of some
94 Divisor Theory

divisor A E 'DK supported by x, and two divisors supported

by x which have the same restriction to x are equal.
DEDUCTION: As was shown in §3.6, 'DK is the direct product
of divisors concentrated at the denominator of [x] and divisors
which avoid the denominator of [x]. The former are, by the
proposition, the kernel of 'DK -+ 'D x , and the latter are by
definition the divisors supported by x. Every element of Dx
can be written in the form [f]x for some polynomial 1 with
coefficients in K. Since [J]x is the image of [1] under'DK -+
'D x , this homomorphism is onto, and the corollary follows.
§3.8.
PROPOSITION. Given a divisor A in a function field K, and
given a parameter x of K, A is supported by u = (x - a)-l for
all but a finite number of rational numbers a.
PROOF: Since a quotient of divisors supported by u is sup-
ported by u, it will suffice to prove the proposition in the case
in which A is integral. By Lemma (8) of §1.1O, Axl[4>(x)]x for
some 4>(x) E Q[x]. If a E Q is not a root of 4>, then x - a
does not divide 4>. Since Q[x] is a natural ring and x - a is
irreducible (prime) in Q[x], it follows that [x - a]x and [4>(x)]x
are relatively prime. Since Q[x] = Q[x - a], restriction to x is
the same as restriction to x - a, and [x - a]x-a and Ax-a are
relatively prime. Thus, [1/uh/u and A l/u are relatively prime,
that is, the restrictions to 1/u of A and the denominator of [u]
are relatively prime. Since the same is trivially true of their
restrictions to u, A is supported by u = (x - a)-l whenever
4>(a)=fO.
An integral divisor in K is reducible in K if it can be written
as a product of two integral divisors, neither of which is [1];
otherwise it is irreducible in K. An integral divisor A in K
is prime in K if the integral divisors in K not divisible by A
form a nonempty set closed under multiplication. (See §1.18.)
 §3.9 Global Divisors 95

COROLLARY. An integral divisor otber tban [1] is irreducible

in K if and only if it is prime in K.

DEDUCTION: If A is prime, then A ¥ [1]. Moreover, if A =

BC, then A divides B or C, say B = AD; then [1] = DC,
C = [1], which shows that A is irreducible. Suppose now that
A ¥ [1] is integral, that A is irreducible, and that AIBC, where
Band C are integral. By the proposition, there is a parameter
u which supports A, B, and C. Then Au is irreducible (if
Au = DuEu, then Du or Eu are restrictions to u of divisors D
and E supported by u, and A = DE by the corollary of §3.7,
so either D or E is [1] and either Du or Eu is [l]u). Therefore
Au is prime (§1.18), so AulBu or AulCu; by the corollary of
§3.7 it follows that AlB or AIC, so A is prime.
§3.9.

Notation: For a nonzero polynomial f with coefficients in

K, let if} denote the numerator of [fl. If aI, a2, ... , a v E K
are not all zero, let {aI, a2, ... , a v } denote if}, where f is
a polynomial whose coefficients are aI, a2, ... , avo (Since [fl
is independent of the choice of f, so is {f}.) Note that {I j x}
is the denominator of [x]. Thus, x supports A if and only if A
avoids {ljx}.

PROPOSITION. Among integral divisors, {aI, a2, ... , a v } is

tbe greatest common divisor of {ad, {a2}, ... , {a v }; tbat
is, {aI, a2, ... , av}l{ail for eacb i, and any integral divisor
wbicb divides {ad for eacb i divides {aI, a2, ... , a v }.
PROOF: Let A be an integral divisor which divides {ad for i =
1,2, ... , V. It is to be shown that, for each parameter x of K,
Ax divides the numerator of [aI, a2, ... , av]x, which in turn
divides the numerator of [ai]x for all i. By Theorem 1, there
are relatively prime integral divisors B I , B 2, ... , Bp in 1)x such
that Ax and the [adx are all products of powers of the Bi, say
Ax = B~l B;2 ... B~I' and [adx = B~il B~i2 ... B!il'. Let f be a
96 Divisor Theory

polynomial whose coefficients are at, a2, ... , avo Then [f]x =
Bfl B~2 ... B~"', where Cj = min(b1j , b2 j, ... , bvj ) (because
this is the g. c. d. of the [ai]x). Thus, the numerator of [!]x
is BtlB;2 ... B;"', where dj = max(Cj, 0), and the numerator
of [ai]x is B;i1 B~i2 ... B:i'" , where eij = max(bij, 0). Because
AI{ ad, aj :5 eij for all i. For fixed j, if some bij < 0, aj :5
e··
') -
- I) >
0 -< d·) follows·, if all b·· _0 ) -< e··
, then a· I) -
- I) , and
b··
aj :5 Cj = dj follows. Thus, Ax divides the numerator of [f]x.
On the other hand, since Cj :5 bij for all i, dj :5 eij for all i,
that is, the numerator of [f]x divides the numerator of [ai]x
for all i.
§3.10.
PROPOSITION (Analog of Theorem 2). Given two integral di-
visors A and B in a function field K, there is a parameter x
of K, supporting both A and B, such that AI{x} and A-1{x}
is relatively prime to B.
PROOF: Let u be a parameter of K which supports AB. By
Theorem 2, there is an x E K integral over Q[u] such that
Au divides [x]u and A;;l[x]u is relatively prime to Bu. Since
the denominator of [x]u is [1]u, the denominator of [x] is con-
centrated at the denominator of [u] (§3.7), that is, {1/x} di-
vides a power of {1/u}. Since AB avoids {1/u}, it therefore
avoids {1/x}, which implies that x supports both A and B.
The denominator of A-I {x} has restriction [1] u to u (Au I[x] u)
and is supported by u (it divides A), so the denominator of
A- 1{x} is [1] (§3.7, Corollary), that is, AI{x}. In the same
way, [A -1 {X }, B] has restriction [1] u to u and is supported by
u, which implies [A-1{x},B] = [1].
COROLLARIES. (1) If A is an integral divisor in a function
field K, and if x is a parameter of K for which AI {x }, then
A = {x, y} for some y E K. (2) Any divisor in a function field
K can be written in the fonn {x,y}/{u,v}, where x, y, u, v
are nonzero elements of K.
 §3.11 Numerical Extensions 97

DEDUCTIONS: (1) By the proposition, there is a parameter y

of K with AI{y} and with A-l{y} relatively prime to A-l{x}.
Thus, {x, y} = AG for some integral divisor G (A is an integral
divisor which divides both {x} and {y}). Since G divides both
A-l{x} and A- 1 {y}, G = [1]. (2) Any divisor can be written
BIG, where Band G are integral divisors. By the proposition,
there are parameters x and u of K such that B I{x} and G I{u }.
By the first corollary, B = {x, y} and G = {u, v}, where y,
v E K are parameters of K and are therefore nonzero.
Corollary (2) shows that, as in the basic theory of divisors,
the theory of (global) divisors in function fields is independent
of the ambient field in the sense that K C L implies 1)K C 1)L
in a natural way. For, if x, y, u, v are nonzero elements of K,
they are also nonzero elements of L, and {x, y} I {u, v} repre-
sents a divisor in L as well as a divisor in K. Since operations
in 1)K (multiplication, inversion, taking numerators, finding
greatest common divisors, etc.) depend only of the field oper-
ations in K, they are unaffected if K is replaced by a function
field L:J K. Therefore, {x,y}/{u,v} and {x',y'}/{u',v'} are
equal in K if and only if they are equal in L, which gives an
inclusion of 1)K in 1)£. Kronecker believed it was important
to structure the theory of divisors in such a way that divisors
would be independent of the ambient field in this sense.
§3.11 Numerical extensions.
It is usual in algebraic geometry to consider function fields
over an algebraically closed field-the field of complex num-
bers or the field of algebraic numbers-rather than over Q.
In the Kroneckerian approach, the transfinite construction of
algebraically closed fields is avoided by the simple expedient
of adjoining new algebraic numbers to Q as needed.

DEFINITION. A numerical extension of a function field K is

an extension field L :J K which is generated over K by a finite
number of constants of L.
98 Divisor Theory

Clearly, a numerical extension of a function field is a func-

tion field, and a numerical extension of a numerical extension
is a numerical extension. Moreover, given two numerical ex-
tensions L and L' of the same function field K, there is a third
numerical extension L" of K which contains sub extensions iso-
morphic to Land L'.
§3.12 The Idea of a Place.
The notion of primality of divisors depends on the ambient
field-an integral divisor prime in a function field K may not
be prime in an extension of K. However, there may be divisors
prime in K which are prime in all numerical extensions of K.
Such divisors correspond, roughly, to points on an algebraic
curve; more precisely, they correspond to what are called places
on an algebraic curve, as will be shown below. Therefore, such
divisors in K will be called places in K.
Places will not be defined in this way, however. Instead,
they are defined in the next article by an algebraic property,
and only at a later point in the development (see §3.22) is
it shown that places, and only places, are prime divisors in
function fields which remain prime in all numerical extensions.
§3.13.
DEFINITION. A place in a function field K is an integral divisor
in K which can be presented in the form {x, y}, where x, y E K
are such that (1) K is a numerical extension of the subfield
Q(x, y) of K generated by x and y over Q, and (2) x and y
satisfy a polynomial equation F(x, y) = 0, with coefficients
which are constants of K, in which the constant term is zero
but at least one of the two linear terms is nonzero.
Note that the notion of "place" , like the notion of "prime" ,
depends on the ambient field. However, as is seen immediately
from the definition, a place in K is a place in any numerical
extension of K.
In the language of algebraic geometry, a function field K can
be regarded as a field of rational functions on an algebraic curve
 §3.13 The Idea of a Place 99

defined by an irreducible equation of the form F(x, y) = 0,

where x and yare elements of K such that K is a numerical ex-
tension of Q(x, y). The origin (x, y) = (0,0) is a non-singular
point of the curve if it is a point of the curve-that is, if the
constant term of F is zero-and if the differential of Fat (0,0)
is not zero-that is, if the linear terms of F are not both zero.
Thus, a divisor in K is a place if there is a presentation of K as
the field of rational functions on a curve in which the divisor
is {x, y} and in which the origin is a nonsingular point of the
curve.
Example: Let K = Q(x, y) be the extension of Q(x) ob-
tained by adjoining a root y of x 3 + y3 - X y. (Otherwise
stated, K = Vp-see §1.23-where P is the prime* divisor in
Q(X, Y) represented by X3 + y3 - XY E Q[X, Y].) Geo-
metrically, the line x = y intersects the "folium of Descartes"
x 3 + y3 = xy at two points, (0,0) and (~, ~). The latter point
is a nonsingular point of the folium, because d( x 3 + y3 - xy) =
3x 2 dx+3 y 2 dy-xdy-ydx, which at (~,~) is ;jdx+ ;jdy"l- o.
Therefore, {x - ~, y - ~} is a place in K. The origin, however,
is a singular point of the folium, one at which, geometrically,
the curve has two coincident points. This statement about the
geometry of the curve is mirrored by the algebraic fact that the
divisor {x, y} in K is a product of two places, which can be
shown as follows.

Folium of Descartes

*x 3 + y3 - XY is irreducible over Q(X) because it has no root in Q[X].

100 Divisor Theory

The geometry of the curve (see figure) suggests that m =

y I x is a ''function'' on the curve which has two different values,
o and 00, at the origin. Substitution of y = mx in the defining
equation x 3 + y3 - xy = 0 and cancellation of x 2 (which is
nonzero in K) gives x + m 3x - m = 0, x = m/(l + m 3). The
restriction to m of {x} is the numerator of [m]m/[l + m 3 ]m,
which, because numerator and denominator in this quotient
are relatively prime, is [m]m. Similarly, the restriction of
{y} = {mx} = {m 2 /(1 + m 3)} to m is [m]~. Therefore, the
restriction to m of {x,y} is [m]m. Since 11m = xly, and
the defining equation is symmetric in x and y, it follows with-
out further calculation that the restriction of {x, y} to 11m is
[1 1mh/ m • These two restrictions determine {x, y}. The re-
striction of {11m} to m is the numerator of [l/m]m, which
is the denominator of [m]m, which is [l]m. Thus, {x, y} and
{m }{l/m} have the same restriction [m]m to m. Symmetri-
cally, they have the same restriction [l/mh/m to 11m. There-
fore, {x, y} = ,{ m }{I 1m}. The factors on the right are places
in K because K = Q( m) (both x and y were shown above to be
in Q(m» and the equations m-m = 0 and (l/m)-(l/m) = 0
show that {m} = {m, m} and {11m} = {11m, 11m} are places
of Q(m).
§3.14.
The definitions of §1.23 have obvious adaptations to divisors
in function fields: An element z of a function field K is finite at
an integral divisor A in K if z is finite at Ax for all parameters
x of K; similarly, z is zero at A if z is zero at Ax for all
parameters x. An element zero at A is also finite at A, and
o is zero at A for all A. For z =I 0, [z] = {z}/{l/z}, and
z is zero at A if and only if A I{z }, finite at A if and only if
[{II z}, A] = [1].
The elements of K finite at A are a ring, and the elements
of K zero at A are an ideal of this ring (because elements finite
at Ax are a ring, and elements zero at Ax an ideal of this ring,
 §3.15 Local Parameters at a Place 101

for all x). The quotient ring is the ring of values of K at A,

denoted VA. Since a constant of K is zero at A if and only if it
is 0, VA contains the field of constants of K in a natural way.
Therefore, VA is a vector space over the field of constants of K.
The theorem of the next article implies that the constants of K
account for all of VA when A is a place, that is, every element
of K finite at a place can be assigned a constant "value" at
that place.

§3.15 Local Parameters at a Place.

THEOREM. Let P be a place in K. There is a unique homo-

morphism ordp from the multiplicative group K* of nonzero
elements of K onto the additive group Z such that (1) z E K*
is finite at P if and only if ordp(z) ~ 0, and (2) z E K* is zero
at P if and only if ordp(z) > 0. If z E K* and ordp(z) = 0,
there is a unique nonzero constant a of K such that z - a is
zero at P.

PROOF: Let P = {x, y} be a presentation of P as a place, and

°
let n be the degree of the highest order terms of the polynomial
F( X, Y) in the relation F( x, y) = which x and yare assumed
to satisfy. Let a, b, e, d be rational numbers to be determined
later, and let x' = ax + by, y' = ex + dy. Then x' and y'
satisfy G(x',y') = 0, where G(X,Y) = F((dX - bY)/(ad-
be), (-eX + aY)/( ad - be)). The coefficients of xn and yn
in G(X, Y) are nonzero polynomials in a, b, e, d divided by
(ad-bet, and there are no terms in G(X, Y) of degree greater
than n. Choose a, b, e, d to make these coefficients nonzero and
to make ad f. be. Since Q(x, y) = Q(x', y') (by assumption,
ad - be f. 0, so the expression of x' and y' in terms of x and
y is invertible), since the constant term of G is zero, and since
the linear terms of G are not both zero (because the linear
terms of F are not both zero), it follows that one can assume
without loss of generality that the polynomial relation satisfied
by x and y contains xn and yn with nonzero coefficients, and
102 Divisor Theory

contains no terms of degree greater than n. In this case, y is

integral over Q[x] and x is integral over Q[y].
By assumption, K is a numerical extension of Q( x, y), and
the coefficients of F are constants of K. Therefore, there is an
algebraic number field Ko C K which contains all the coeffi-
cients of F and which satisfies K = Ko(x, y). Since x and y
can be interchanged if necessary, one can assume without loss
of generality that the coefficient of y in F is nonzero, and in
fact, since one can divide by this coefficient, that it is 1. Let
all terms of the relation F( x, y) = 0 which contain x be moved
to the right side of the equation, and let a factor of y be taken
out on the left side and a factor of x be taken out on the right,
to put the relation in the form yif>(y) = x'¢(x, y), where if> and
'¢ are polynomials with coefficients in Ko and if>(0) = l.
Every nonzero element z of K can be expressed in the form

8(x,y)
z - -.:......;,...;:.....:...
- 1](x,y)'

where numerator and denominator are nonzero elements of

K expressed as polynomials in x and y with coefficients in
Ko. If 8(0,0) = 0, then 8(x, y) = x(J(x, y) + yr(x, y) and
8(x,y)if>(y) = x8 1 (x,y), where the polynomial 81 is defined by
81 (X, Y) = (J(X, Y)if>(Y) + '¢(X, Y)r(X, Y). If 01 (0,0) = 0,
this process can be repeated to give O(x, y)if>(y? = x202(X, y).
After a finite number of steps, one must reach an equation
of the form O(x, y)if>(y)/L = x/L8/L(x, y) in which O/L(O,O) =I- 0,
because, since [x/L8/L(x, Y)]x is divisible at least fl times by
Px = [x, Y]x and [if>(y)]x is not divisible at all by Px (if>(0) = 1),
fl cannot exceed the number of times Px divides [O(x, Y)]x.
Therefore, multiplication of numerator and denominator of
z = 8/1] by a sufficiently high power of if> ( y) gives
 §3.15 Local Parameters at a Place 103

°°
where 11 (0,0) and 1]m(O, 0) are nonzero. With j = Il- m, and
a = 11 (0, O)/1]m(O, 0), this relation can be written in the fonn

z
-
xi
- a-
0ll(x, y)4>(y)N-1i
- 11
°
(0,0)
-
6(x, y)
- 1]m(x,y)4>(y)N-m 1]m(O,O) - c+c(x,y)'
--'----'-'~

where j E Z, where a, c E Ko are nonzero, and 6(x, y) and

c(x, y) are polynomials in x and y with constant term 0.
The restriction of {zx - i-a} to x is therefore the numera-
tor of the restriction to x of [hex, y)]![c + c(x, y)]. Since y is
integral over Q[x], both [hex, Y)]x and [c + c(x, Y)]x are inte-
gral. Since P x divides the former and is relatively prime to the
latter, Pxl{zx- i -a}x. Since x supports P (PI{x} and {x} is
relatively prime to {l/x}), PI{zx- i - a}, that is, zx- i - a is
zero at P.
Given z, j is characterized by the property that zx- i =
a + (zx- i - a) is finite and nonzero at P, because, if i <
j, then zx- i = xi-izx- i is zero at P, and, if i > j, then
zx- i must not be finite at P, because, if it were, then zx- i =
x i - i zx- i would be zero at P. Therefore, the function z ~ j
is a homomorphism from K* to Z, because if zx- i and wx- i
are finite and nonzero at P the same is true of their product
zwx- i - i . This homomorphism carries x to 1. IT it carries z
to j ~ 0, then z = xi(zx- i - a) + ax i is finite at P, and if
it carries z to j > 0, then z is zero at P (elements of K zero
at P are an ideal in the ring of elements of K finite at P).
Thus, ordp(z) = j defines a homomorphism K* ~ Z with the
required properties.
IT ordp is a homomorphism with the required properties,
then ordp(zx- i ) is nonnegative but not positive; therefore,
ordp(zx- i ) = 0, that is, ordp(z) = j ordp(x). Because ordp is
onto, ordp(x) (which is positive) must be 1. Thus, ordp(z) = j
is the only such homomorphism.
Finally, when j = 0, the above construction of an element
zx- i - a zero at P gives z - a zero at P. IT z - b is also zero
104 Divisor Theory

at P, where b is a constant of K, then b - a = (z - a) - (z - b)

is a constant of K which is zero at P, that is, b = a.

DEFINITION. An element x E K* for which ordp(x)

called a local parameter of K at the place P.
= °is

§3.16.
COROLLARIES. (1) A place in a function field K is prime in
K and in any numerical extension of K. (2) Let x be a local
parameter at a place P in a function field K. If z is any
nonzero element of K and if v = ordp(z), there is an infinite
sequence a", a,,+l, a,,+2, ... of constants of K with a" =1= 0,
such that, for all N ~ v, z = a"x" + ... +aNx N +o(xN), that
is, (z - a"x" - a,,+lx,,+l - ... - aNxN)x- N is zero at P.
DEDUCTIONS: (1) Let A be an integral divisor in K such that
P f A. Then B = P/[P,A] is integral, BIP, and B =1= [1].
Let u, v E K be such that B = {u,v}. Since B =1= [1] and B
divides both {u} and P, [P, { u }] =1= [1], which is to say that
l/u is not finite at P. Thus, ordp (l/u) < 0, ordp(u) > 0, so
u is zero at P, that is, PI{ u}. For the same reason, PI{ v}.
Thus, PI{u,v} = B, which implies P = B = P/[P,A], that is,
[P, A] = [1]. Since P f A implies [P, A] = [1], P is prime.
(2) For N = v, v + 1, v + 2, ... , let qN = (z - a"x" -
a,,+!x,,+l _ ... - aNxN)/x N , where the ai are constants in K
to be determined. The theorem provides the unique nonzero
a" such that q" = zx-" - a" is zero at P. Then ordp(q,,) > 0,
ordp(q,,/x) ~ 0, and the theorem provides a constant a,,+l
(zero if q,,/x is zero at P) such that q,,+! = (q,,/x) - a,,+l is
zero at P. In the same way, once a", a,,+l, ... , aN-l have
been found, aN is determined by the condition that qN =
(qN-dx) - aN be zero at P.
It is a corollary to the proof of the theorem that if a place
in a function field K is given, then the field of all constants in
K can be constructed. All constants in the computation of a,
starting with the coefficients of </J, "p, 6, TJ, are in the algebraic
 §3.17 Relative Norms 105

number field K o, so a is in Ko. In particular, if z E K is

constant, then PI{ z - a} for some a E Ko. Since z - a is
constant, z - a = 0, z = a E K o , and Ko is the field of all
constants of K, as was to be shown.
§3.17 Relative norms.
For the sake of simplicity, only absolute norms of divisors
were defined in §1.17. In the case of function fields, in which
the divisor theory is defined relative to a whole family of nat-
ural rings Q[x] and there is no one field with respect to which
"absolute" norms can be defined, it is natural to introduce rel-
ative norms, which requires only a slight modification of the
proposition of §1.17.
PROPOSITION. Let K and L be extensions of a natural ring r
of finite degree, and let L :J K. Given a polynomial f with
coefficients in L, let N L/ K f be the polynomial with coefficients
in K obtained by taking the determinant of the matrix which
represents multiplication by f relative to a basis of Lover K.
(Clearly N L/ K f is independent of the choice of the basis.) If
the coefficients of f are integral over r, then N L / K f = f q,
where the coefficients of q are elements of L integral over r.
PROOF: There exist elements WI, W2, ... , wJL of L such that
any coefficient of f times any W is a linear combination of w's
with coefficients in T' (Proposition 1 of Part 0). Then W f =
M w, where W is the column vector with entries WI, W2, ... , W JL'
where f denotes the 1 x 1 matrix, and where M is a f-L x f-L matrix
whose entries are polynomials in the indeterminates of f with
coefficients in r. In other words, (J 1- M)w = 0, where I is the
f-L x f-L identity matrix and 0 is the column matrix containing
f-L zeros. Since W is nonzero, it follows that det(J I - M) = 0,
that is, f is a root of det(X 1- M) = F(X) = XJL + FIx JL - I +
F 2 x JL - 2 + ... + F JL , where X is a new indeterminate and the
Fi are polynomials in the indeterminates of f with coefficients
in T'. By the Remainder Theorem, F(X) = (X - J)Q(X),
106 Divisor Theory

where the coefficients of Q are polynomials with coefficients

in L. Taking NL/K on both sides gives F(Xt = NL/K(X -
!)NL/KQ(X) because the norm of a product is clearly the
product of the norms, and because the coefficients of F are in
r C K. (Here n = [L: K).) Since F(x)n has coefficients in r
and NL/KQ(X) is monic, the coefficients of NL/K(X -!) are
roots of polynomials whose coefficients are polynomials in the
coefficients of F(x)n with coefficients in Z (by the theorem of
Part 0 and the method of Corollary 3 of Part 0 for reducing
from the case of several indeterminates to the case of one).
Therefore, the coefficients of N L / K (X - !) are integral over r
(Proposition 2 of Part 0).
Say NL/K(X - J) = xn + A1xn-1 + ... + An, where the
A's are polynomials (in the indeterminates of !) with coef-
ficients in K integral over r. Substitution of X = 0 gives
(-l)nNL/Kf = An. Substitution of X = f in the matrix of
which N L/ K( X -!) is the determinant gives a matrix whose de-
terminant is zero (the chosen basis of Lover K is a column vec-
tor in its kernel). Therefore fn+A1f n- 1+ .. ·+A n = 0, which
gives NL/Kf = ±A n = ':fun + A1f n- 1 + ... + An-I!), and
proves the proposition with q = ':fun-I+Alfn-2+ .. ·+An-I).
COROLLARY. If K and L are function fields with K c L, and
if A is a global divisor in L, then N L/ K A is a global divisor in
K, defined in the obvious way: For any parameter x of K, let f
be a polynomial with coefficients in L such that Ax = [f)x and
set (NL/KA)x = [NL/Kf)x. If A is integral, then AINL/KA.
In particular, N L/ K A is integral if A is.
DEDUCTION: By the proposition, with r = Q[x), if [f)x is
integral, then [f)xl[NL/Kf)x, and, in particular, [NL/Kf)x
is integral. It follows, as in §1.17, that [f)xl[g)x implies
[NL/Kf)xI[NL/K9)x. Therefore, (NL/KA)x is independent of
the choice of f. Moreover, if x is integral over Q[u) and Ax =
[f]x, then Au = [flu, which shows that (NL/KA)u = [NL/Kf]u
is the restriction to u of (NL/KA)x = [NL/Kf]x, so NL/KA is
 §3.18 A Fundamental Theorem 107

a well-defined divisor. IT A is integral, AINL/KA because

AxINL/KAx for all x.
Note: The homomorphism NL/K : 'DL -+ 'DK depend8 on
the ambient field L, as well as the field K to which the norm is
taken. For example, if A is a divisor in L, and if L' :::> L, then
NL'/KA = (NL/KA ) [L'.L]
..
§3.18 A Fundamental theorem.
THEOREM. Given a divisor A in a function field K, there is a
numerical extension of K in which A is a product of powers
(possibly negative) of places.
(Since places are prime, the representation of a divisor as a
product of powers of places in a numerical extension is unique
in a natural way, namely: Let L and L' be numerical extensions
of K in which A is a product of powers of places, say A = p;in
in L and A = nQr i in L'. There is a numerical extension L"

of K in which Land L' can both be embedded isomorphically.

Under such isomorphisms, both the Pi and the Qi correspond
to places in L". By the uniqueness of factorization of divisors
in L", the two factorizations of A must be identical in L".)
LEMMA. Given an integral divisor A in a function field K,
there is an upper bound on the number of nontrivial factors in
a factorization of A in a numerical extension of K. More pre-
cisely, there is an integer v such that if A = B(1) B(2) ... B(I') ,
where the B( i) are integral divisors in a numerical extension
of K and where Jl > v, then B(i) = [1] for at least one i.
PROOF: Let x E K be such that AI{x}. It will be shown that
v = [K: Q ( x)] has the desired property.
Let A = nr=l B(i), where the B(i) are integral divisors in
a numerical extension L of K. Let L = K(cl, C2, ... , cp ),
where the Ci are constants of L. Let L1 = Q( Cl, C2, ••• ,
cp ), an algebraic number field contained in L. Then L 1 (x) is a
function field contained in L. Let N denote the norm N L/L1(x).
108 Divisor Theory

Since AI{x}, NAIN{x} = {x}[L:L1(x»). Since {x} is a place in

the function field L1 (x) and is therefore prime in L1 (x), N A
is thus a power of {x}, say N A = {x} T, where 7 satisfies
o ~ 7 ~ [L:L 1 (x)]. In the same way, NB(i) = {x}°"i, for some
(J'i ~ 0, for each i = 1, 2, ... , f-l. Clearly, (J'1 +(J'2 + ... +(J' I' = 7.
If (J'i = 0, then B(i)l{x}O = [1], and B(i) = [1]. Therefore, if
none of the factors of A = IT B( i) is [1], each (J'i is at least 1,
and 7 = (J'1 + (J'2 + ... + (J' I' ~ f-l. Since 7 ~ [L: L1 (x)], it follows
that [L: L1 (x)] ~ f-l.
A basis of L1 (x) over Q( x) spans L1 over Q( x). Since
L1 spans Lover K :) Q( x), a basis of L1 (x) over Q( x)
spans Lover K. Therefore, [L:K] ::; [L 1(x):Q(x)]. Since
[L:K][K:Q(x)] = [L:Q(x)] = [L:L 1(x)][L 1 (x):Q(x)], it fol-
lows that v = [K:Q(x)] ~ [L:L 1(x)] ~ f-l if no B(i) is [1], as
was to be shown.
PROOF OF THE THEOREM: Since every divisor is a quotient
of integral divisors, it will suffice to prove the theorem for in-
tegral divisors A. IT A = [1], the theorem is true by virtue
of the convention that [1] is a product of no places. Assume,
therefore, that A is integral and not [1]. The method of proof
will be to give a construction which either produces a proper
factorization of A in a numerical extension of K or proves
that A is a place in a numerical extension of K. Since, by the
lemma, the number of times A can be factored is bounded, re-
peated application of this construction will succeed eventually
in writing A as a product of places in a numerical extension of
K.
Let x be a parameter of K such that A -1 { x} is integral and
relatively prime to A (§3.10). By the theorem of the primitive
element (§1.25), the extension K :) Q( x) is generated by a
single element y E K. Let F E Q[X, Y] be an irreducible
polynomial such that F(x, y) = 0 in K. Then K = Q(x, y) is
obtained by adjoining to Q( x) a root y of F( x, Y) = O.
By the proposition of §3.6, A = Be, where B is supported
 §3.18 A Fundamental Theorem 109

by y and e is concentrated at {l/y}. IT neither B nor e

is [1], the construction terminates with a proper factorization
A = Be of A. IT B = [1], A = e is concentrated at {l/y} and
e
is therefore supported by l/y. IT = [1], A = B is supported
by y. In either case, then, there is a parameter of K which
generates Kover Q( x) and which supports A. Let y be such
a parameter.
The restriction to y of N K/Q(y)A is integral, call it [q(y)]y,
where q(y) is a polynomial with rational coefficients. There is
a numerical extension L of K in which q(y) splits into linear
factors, say [q(y)]y = TI[y - ki]y, where the ki are constants in
L. (A splitting field of q over Q can be obtained by adjoining a
single root of a single irreducible polynomial with coefficients in
Q. Such a polynomial has coefficients in K, and adjunction of a
single root of a single factor of it irreducible over K gives a field
L with the desired property.) IT one of the divisors [A, {y - k i }]
is a proper divisor of A, then A is factored and the construction
terminates. Otherwise, every [A, {y-k i }] is [1] or A. IT [A, {y-
kd] were [1] for all roots ki of q, then [Ay, y - ki]y would be
[l]y for all ki, which would imply that Ay was relatively prime
to [q(y)]y = TI[y - ki]y; since Ay divides its norm [q(y)]y,
this would imply Ay = [l]y, contrary to the assumptions that
A =f [1] and y supports A. Therefore, AI{y - k} for some
constant k of L. (In fact, since [{y - k}, {y - k'}] = [1] for
k =f k', this condition determines k.) In summary, if K is a
numerical extension of Q(x, y), and if y supports A, then either
A has a proper factor of the form [A, {y - k }] for some constant
k in a numerical extension of K, or A divides {x, y - k} for
some constant k in a numerical extension of K.
In the second case, if {x,y - k} is a place of L = K(k),
then, because places are prime and A =f [1], A = {x, y - k} is
a place in L and the construction terminates. Otherwise, the
construction continues as follows. Let Yl = (y - k)/x. Then L
is a numerical extension of Q(x, Yl) C L because adjunction of
the constant k to Q( x, yt) gives a field containing y = k + XYl,
110 Divisor Theory

and therefore gives a field containing Q(x, Y), of which L is a

numerical extension. Also, Yl supports A because

{y - k}{l/x}
[Yl] = [y - kJl[x] = {l/(y _ k)}{ x}
and when the common factor A in numerator and denominator
of this quotient is cancelled the denominator becomes {l/(y-
k)}A-1{x}, which is relatively prime to A (because {l/(y-k)}
is relatively prime to {y - k} and therefore. to A, whereas
A-I {x} is relatively prime to A by the choice of x). As before,
either A has a proper divisor of the form [A, {Yl - k'}] where
k' is a constant in a numerical extension of K, or A divides
{x, Yl - k'}, where k' is a constant in a numerical extension of
K. In the first case, the construction terminates. In the second
case, the construction terminates if {x, Yl - k'} is a place, but if
{ x, Yl - k'} is not a place the construction continues by setting
Y2 = (Yl - k')/x and iterating.
If, after m iterations, the construction has not terminated, it
has produced a sequence k, k', k", ... , k(rn-l) of constants in
a numerical extension of K, say in L = K(k, k', ... , k(rn-l»),
and a sequence y, Yl, Y2, •.. , Yrn of parameters of L such
that AI{ x, Yi - k(i)} for i = 0, 1, ... , m - 1, such that Yi+l =
(Yi - k(i»)/x for i = 0, 1, ... , m-1, and such that {x, Yi - k(i)}
is not a place in L. The theorem will be proved if it is shown
that these ~onditions imply an upper bound on m.
As above, let F E Q[X, Y] be an irreducible polynomial
with rational coefficients such that F( x, y) = 0 in K. Since
{x, Y - k} is not [1] and not a place in L, when F is written as
a polynomial in X and Y - k, it contains no terms of degree
less than 2. Substitution of XY for Y - k in this polynomial
therefore gives a polynomial in which all terms are divisible
by X 2 , say F(X,k+XY) = X 2 G1(X,Y). Then G1(x,Yt) =
x- 2 F(x, k + xyt) = x- 2 F(x, y) = 0 in K, and therefore in L.
In the same way, since {x, Yl - k'} is not [1] and not a place,
 §3.18 A Fundamental Theorem 111

G 1 (X, k' + XY) = X 2G2(X, Y), where G 2 is a polynomial

whose coefficients are constants of L and where G 2(x, Y2) = o.
In general, Gi(X, k(i) +XY) = X2Gi+l(X, Y) for i = 1, 2, ... ,
m. Therefore, F(X,k + k'X + k"X2 + ... + k(m-1)xm-l +
xmy) = X 2G1(X, k' +k"X + .. .+k(m-1)xm-2 +Xm-1y) =
X 4 G2(X, kIf + k(3) X + ... + k(m-1) xm-3 + xm- 2y) = ... =
X 2m- 2G m_ 1 (X, k(m-l) + XY) = X 2mG m(X, Y). In particu-
lar, x2m divides F(X, k + k'X + k"X 2 + ... + k(m-1)xm-l +
xmy) (a polynomial in X and Y with coefficients in the alge-
braic number field Q(k, k', ... , k(m-l))).
This condition implies an upper bound on m as follows. Let
Fy(X, Y) denote the partial derivative of F with respect to
Y (that is, the coefficient of h in F(X, Y + h)). When F and
Fy :1re regarded as polynomials in Y with coefficients in the
field Q(X) of rational functions of X, the Euclidean algorithm
can be used to write their greatest common divisor d(Y) in
Q(X)[Yj in the form a(Y)F + b(Y)Fy, where a and b have
coefficients in Q(X). The statement that d(Y) divides F-say
c(Y)d(Y) = F-implies c(X, Y)d(X, Y) = e(X)F(X, Y) when
e(X) E Q[Xj clears the denominators in c and d to make them
polynomials c(X, Y), d(X, Y) E Q[X, Yj. By unique factor-
ization of polynomials in Q[X, Y], and by the irreducibility
of F(X, Y) in Q[X, Yj, F divides either c(X, Y) or d(X, Y)
and the other is a polynomial in X alone. Since d(Y) di-
vides Fy, its degree in Y is at most degy(F) -1, which shows
that F cannot divide d(X, Y). Therefore, d(X, Y) is a poly-
nomial in X alone of the form f(X)a(Y)F + f(X)b(Y)Fy,
say ~(X) = g(X, Y)F(X, Y) + h(X, Y)Fy(X, Y), where g,
hE Q[X, Yj. In this equation, let Y be replaced by k + k'X +
kIf X 2 + ... + k(m-l) xm-l + xmy. The left side does not
change. The first term on the right is divisible by x2m. The
second term on the right is divisible by xm because differen-
tiation of F(X, k + k'X + ... + xmy) = X 2mG m(X, Y) with
respect to Y gives xm Fy(X, k + k' X + ... + xmy) = x 2m
times a polynomial in X and Y. Therefore, xm must divide
112 Divisor Theory

~(X). Since ~(X) can be computed a priori, knowing only

F(X, Y), this condition bounds m and proves the theorem.
§3.19.
COROLLARY. Tbe field of constants of a function field can be
constructed.
DEDUCTION: Given a function field K, let x be a parameter
of K. Since {x} =I- [1], the factorization of {x} into places in a
numerical extension L of K provides a place P in L. As was
noted at the end of §3.16, knowledge of P makes possible the
construction of the field of constants Lo of L. Let 1, Q2, Q3, ... ,
Q v E Lo be a basis of Lover K. Each f3 E Lo can be written in

just one way in the form f3 = 91(f3) + 92(f3)Q2 + ... + 9v(f3)Qv,

and this equation defines functions 9i : Lo ~ K linear over Q.
As a vector space over Q, the field of constants Ko of K is the
subspace of Lo defined by the relations 9i(f3) = 0 for i = 2, 3,
... , v. Thus, a basis of Ko over Q can be constructed.
§3.20 Presentation of Places.
THEOREM. Let x and y be elements of a function field K such
tbat {x, y} is a place in K. Tben x and y present {x, y} as a
place in K, tbat is, (1) K is a numerical extension of Q( x, y)
and (2) x and y satisfy a polynomial relation F( x, y) = 0, witb
coefficients wbicb are constants of K, in wbich tbe constant
term is zero but at least one of tbe two linear terms is nonzero.
LEMMA. Let K be a function field, let Ko be its field of con-
stants, let x, y E K generate Kover K o, and let F(X, Y) E
Ko[X, Y] be a nonzero polynomial, irreducible over K o, such
tbat F(x,y) = 0 in K. Finally, let A = {x,y}. Tben
NK/Ko(x)A = {x},\ wbere A is tbe degree oftbe term or terms
of F(X, Y) of lowest degree.
PROOF: Let F(X, Y) = ao(x)ym+ a1 (x)ym-l+ .. ·+am(X).
Multiplication of the relation F(x, y) = 0 by ao(x)m-l shows
 §3.20 Presentation of Places 113

that ao(x)y is integral over Ko[x]. Therefore,

[ ] _ [ao(x)x,ao(x)y]x
x, y x - [ao(x )]x

presents [x, y]x as a quotient of integral divisors. Thus, Ax,

which is the numerator of [x,Y]x, is [ao(x)x,ao(x)y]x divided
by its g. c. d. with [ao(x)]x, that is, divided by [ao(x)x, ao(x)y,
ao(x)]x = [ao(x)y, ao(x)]x. Therefore,

N A _ N[ao(x)x, ao(x )y]x

x - N[ao(x),ao(x)y]x '

where N is the norm from K to Ko(x).

Now [ao(x)x,ao(x)y]x = [ao(x)]x[xU - Y]x, where U is an
indeterminate. For indeterminate V, an elementary computa-
tion gives

al(x) m-l
m
NKjKo(x)(V-y)=V +-(-)V
ao x
+ ... + am(x)
().
ao x

Therefore,

N[ao( x )x, ao( x )y]x

= [ao(x)m]x[(xu)m + al(x) (xu)m-l + ... + am(x)]
ao ( x ) ao ( x ) x
= [ao(x )m-l]x[ao(x)xmu +al (x)xm-1u - + ... + am(x )]x
m m 1
= [ao(x )m-l]x [ao( x )xm, al (x )x m - 1 , a2(x )x m- 2, . .. ,a m( x )]x.

The calculation of N[ao(x),ao(x)y]x is the same with U -

y in place of xU - y, which gives N[ao(x), ao(x )y]x =
[ao(x)m-l]x[ao(x), al(x), ... , am(x)]x. The second factor on
the right is [l]x because F(X, Y) is irreducible over K o, so
that the ai(x) have no common factor in Ko[x]. In conclusion,
therefore, NAx = [ao(x)x m, al(x)xm-I, ... , am(x)]x.
114 Divisor Theory

Since AI{ x}, N A {x}1' for some J-l ({x} is a place in

Ko(x)). In fact, J-l is the largest integer such that [x]~ di-
vides N Ax, that is, the largest integer such that [x]~ di-
vides [ai(x)xm-i]x for i = 0, 1, ... , m. If F(X,Y) =
E E bUTxoYT, then ai(X) = E bu,m_iXu, and ai(x)x m- i =
"LJu bu,m-i Xu+m-i·IS d·IVISI
. ·ble by x I' if and on1y 1·f bu,m-i = 0
whenever (J' + m - i < J-l. Thus, [x]~ divides N Ax if and only
if F(X, Y) = E E bUTxuYT has bUT = 0 whenever (J' + T < J-l.
In other words, [x]~ divides N Ax if and only if J-l :s; >., as was
to be shown.

PROOF OF THE THEOREM: Let P = {x,y} be a place in K.

Since either x or y must be a local parameter at P, there is
no loss of generality in assuming that x is a local parameter
at P. Let Z E K be a generator of Kover Ko(x) which is
integral over Ko[x], and define a sequence Zo, Zl, Z2, ... of
elements of K finite at P and a sequence ko, kl' k2, ... of
elements of Ko by the conditions Zo = z, Zi - k i is zero at P,
and Zi+l = (Zi - ki)/x.
Let Go(X, Z) be an irreducible polynomial with coefficients
in Ko such that Go(x, zo) = o. (K is an algebraic extension of
K 0 ( x ).) As a polynomial in x and Zo - ko, Go ( x, ko + (Zo - ko ))
has constant term zero (both x and Zo - ko are zero at P, so
the constant term is zero at P, which implies that the constant
term is zero). If either of its linear terms is nonzero, then,
since K = Ko(x, z) = Ko(x, Zo - ko), x and Zo - ko present
P as a place in K. Otherwise, every term of Go(X, ko + XZ)
is divisible by X2 and G1(X,Z) = X- 2Go(X,k o + XZ) is
a polynomial in X and Z with coefficients in Ko. Since
G1(x, Zl) = x- 2Go(x, ko + xzt) = x- 2Go(x, zo) = 0, the
constant term of G 1 (x, kl + (Zl - k 1 ») as a polynomial in
x and Zl - kl is zero. If either linear term is nonzero,
then, because K = Ko(x,zo - ko) = KO(X,Zl - k1 ), x and
Zl - kl present P as a place in K. Otherwise, G 2 (X, Z) =
X- 2 G 1 (X, kl +X Z) is a polynomial with coefficients in Ko for
 §3.21 Degree of a Divisor 115

which G 2(X,Z2) = 0, and the process continues. The process

must reach a point at which x and Zi - k i present P as a place
in K, because, as was shown in §3.18, the condition that there
be a polynomial Gm(X, Z) = X- 2G m_ 1(X, km - l + X Z) =
X- 4 G m_ 2(X, k m- 2+k m- 1X +X2 Z) = ... = X- 2mG o(X, ko+
klX +k2X2+ .. ·+km_1xm-l+xm Z) imposes an upper bound
on m. Therefore, there is awE K of the form w = Zi - ki
such that x and w present P as a place in K.
Let K' = Ko(x,y). Since P = {x,y} is in K',

( ) [K:K'j
NK/Ko(x)P= NK'/Ko(x)P .

By the lemma, NK'/Ko(x)P = {X}A, where A is the degree of

the term or terms of lowest degree in F. Thus, NK/Ko(x)p =
{x }A[K:K'j. On the other hand, since x and w present P =
{x, w} as a place in K, by the lemma, N K / K 0 (x) P = {x}.
Thus, 1 = A[K: K'], which implies A = 1 and [K: K'] = 1, that
is, F contains a nonzero linear term and K = Ko(x, y), which
is a numerical extension of Q( x, y), as was to be shown.
3.21 The Degree of a Divisor.
The degree of a divisor A in a function field K, denoted
degK A, is the sum of the exponents in a representation of A
as a product of powers of places in a numerical extension of
K. (The uniqueness of the representation of A as a product
of places in a numerical extension of K implies that degK A is
well defined. Clearly, when L is a numerical extension of K,
degL A = degK A. Also, degK AB = degK A + degK B.)
THEOREM. If K and L are function fields with K C L, there is
a positive integer m such that, for all divisors A in K, degL A =
m degK A. Specifically, m = [L: K]/[Lo: K o], where Lo and Ko
are the fields of constants of L and K respectively.
(Geometrically, the inclusion K C L corresponds to a mor-
phism from the curve with field of functions L to the curve with
116 Divisor Theory

field of functions K. Each place on the range curve corresponds

to m places on the domain curve, counted with multiplicities.)

LEMMA 1. If L is a numerical extension of K, then [L: K] =

[Lo: Ko].

The proof given in the first printing was erroneous. The

Lemma on page 119 shows that if f3 is constant-that is, if
f3 is integral over both Q[x] and Q[l/x]-then so are the ai.
Therefore, adjoining a constant to K is the same as adjoining
a constant to Ko. (Note added in second printing.)

LEMMA 2. Let A be an integml divisor in a function field K,

and let AI{x} for x E K. Then NK/Ko(x)A = {x}degKA.
PROOF: If A = {x,y} is a place in K, degKA = 1 and the
lemma follows from the theorem and lemma of §3.20.
Given any integral divisor A in K, let L be a numeri-
cal extension of K in which A is a product of places, say
A = p(l) p(2) ... p(v), where the p(i) are places in L and
v = degK A. Let Lo be the field of constants of L. Then
NL/Lo(x)A = I1NL/ Lo (x)p(i) = {xy and what is to be shown
is that NL/Lo(x)A = NK/Ko(x)A. Since both sides of this equa-
tion are powers of {x}, it will suffice to prove that their restric-
 §3.21 Degree of a Divisor 117

tions to x are equal, and for this it will suffice to prove that
NL/Lo(x)! = NK/Ko(x)! for any polynomial f with coefficients
in K. By Lemma 1, [L: K] = [Lo: Ko] = [Lo(x): Ko(x)]. Since
[L: K][K: Ko(x)] = [L: Ko(x)] = [L: Lo(x)][Lo(x): Ko(x)]' it
follows that [K: Ko(x)] = [L: Lo(x)]. A basis of Kover Ko(x)
spans both K and Lo over Lo( x), and therefore spans Lover
Lo(x). Since [K: Ko(x)] = [L: Lo(x)], it follows that a basis
of Kover K 0 ( x) is a basis of Lover Lo ( x). When such a ba-
sis is used to compute NL/Lo(x)! and NK/Ko(x)! for f with
coefficients in K, the results are identical, as was to be shown.
LEMMA 3. If K, L, and M are function fields witb K c L c
M, tben NM/KA = NL/K(NM/LA).
PROOF: This follows immediately from the same formula for
norms of polynomials with coefficients in M, a basic algebraic
identity which can be proved by expressing the norm as a prod-
uct of conjugates under the Galois group (see Proposition 2 of
§1.25).
PROOF OF THE THEOREM: Clearly it will suffice to prove that

[Lo: Ko] degL A = [L: K] degK A

for integral divisors A in K. Given such an A, let x E
K satisfy AI{x}. Then NL/Ko(x)A = NK/Ko(x)(A[L:kl )
{x }[L:KJ degg A while on the other hand

N L/ Ko(x)A = N Lo(x)/ Ko(x)NL/ Lo(x)A

= N Lo(x)/ Ko(xd x} deg L A = {x }[Lo:KoJ degL A,

from which the desired equation follows.

COROLLARY. For any nonzero x E K, degK[x] = O.
DEDUCTION: If x is a constant of K, then [x] = [1] obviously
has degree O. Otherwise, in Q(x) c K, [x] = {x}{l/x}-l gives
[x] as a product of places and shows that its degree in Q(x) is
O. Therefore, by the theorem, its degree in K is O.
118 Divisor Theory

§3.22 A Characteristic of Places.

PROPOSITION. An integral divisor A in a function field K
which is prime in K and in every numerical extension of K
is a place in K.
PROOF: Let A = {x,y} where x, y E K (§3.10), let
Ko be the field of constants of K, and let F(X, Y) E
Ko [X, Y] be a nonzero polynomial, irreducible over K o, such
that F(x, y) = O. By the lemma of §3.20, N K/ Ko(x)A =
(NKo(x,y)/Ko(x)A)[K:Ko(x,y)] = {x}A[K:Ko(x,y)], where A is the
degree of the terms of lowest degree in F. On the other hand,
degK A = 1 by the definition of the degree and the assump-
tion on A. By Lemma 2 of §3.21, N K/ Ko(x)A = {x}. Thus,
A[K: Ko(x, y)] = 1, which implies both that F contains a
nonzero linear term and that K is a numerical extension of
Q(x, y).
§3.23 Dimension of a Divisor.
A divisor A in a function field K will be said to divide an
element z of K, denoted Alz, if, for every parameter x of K,
Axlz (where Ax is the restriction of A to x). Otherwise stated,
Alz if z = 0 or if z f. 0 and Al[z]. IT Alz and z f. 0, then A -1 [z]
is integral, which implies degK(A-l[z]) ~ 0, 0 = degK[z] ~
degK(A). This observation proves the statement of §3.4 that
not every (global) divisor is of the form [1]; in fact, a divisor
of positive degree cannot even divide a divisor of this form
(because it would then divide a divisor of the form [z] for
z f. 0) much less be of this form.
IT Alz then Alaz for any constant a in K, and if Alz1 and
Alz2 then AI(Z1 + Z2). Thus, the set of elements of K divisible
by A form a vector space over the field of constants of K. The
dimension of this vector space will be denoted dimK(A). As
was just observed, degK(A) > 0 implies dimK(A) = o. As will
be proved in the next article, dimK(A) < 00 for all divisors A
inK.
 §3.23 Dimension of a Divisor 119

PROPOSITION. H A is a divisor in K, and if L is a numerical

extension of K, tben dimK(A) = dimL(A).
LEMMA. Let L = K(a) be a numerical extension of K ob-
tained by adjoining a constant a to K. Every element of
L can be written in one and only one way in tbe form
(3 = ao + ala + a2a2 + ... + an_la n- l , wbere n = [L: K]
and wbere ai E K. Given a parameter x of K, (3 is integral
over Q[x] if and only if eacb ai is integral over Q[x].
PROOF: Any constant of L is integral over Q and is therefore
integral over Q [x] ::::) Q. Therefore, (3 is integral over Q [x]
whenever the ai are all integral over Q[x). For the proof of
the converse, let (3 be integral over Q[x). Let F(X) be the
monic, irreducible polynomial with (constant) coefficients in
K of which a is a root (i.e., F(X) = NL/K(X - a)), and let
q(X) = F(X)j(X -a). Let ¢>(X) be the trace of (3q(a)-lq(X)
relative to the extension L ::::) K. Then ¢>(X) = L: 'lj;(i)(X),
where the 'lj;(i)(X) are the n conjugates of (3q(a)-lq(X) in a
splitting field of F(X) over K. Since the coefficients of the
'lj;(i)(X) are integral over Q[x] (both (3 and the coefficients of
q(a)-lq(X) are integral over Q[x)), so are the coefficients of
¢>(X), say ¢>(X) = 'Yo + 'YIX + 'Y2X2 + ... + 'Yn_lxn-l, where
the 'Yi E K are integral over Q[x). Thus, ¢>(a) = 'Yo + 'Yla +
'Y2a2 + ... +'Yn_lan-l. On the other hand, the 'lj;(i)(a) are all
o (because a is a root of the conjugates of q(X)) except for the
one corresponding to the identity conjugation, and that one is
(3q(a)-lq(a) = (3. Therefore, (3 = ¢>(a) = 'Yo + 'Yla + 'Y2 a2 +
... + 'Yn_lan-l where the 'Yi are integral over Q[x). Since the
representation of (3 in this form is unique, the lemma follows.
PROOF OF THE PROPOSITION: There is no loss of generality
in assuming that L = K( a) where a is a constant of K (every
numerical extension can be obtained as the result of a sequence
of such extensions). Then Ko(a) is the field of constants of L
(Lemma 1, §3.21), where Ko is the field of constants of K.
Every element (3 of L can be written in one and only one way
120 Divisor Theory

in the form f3 = ao + ala + ... + an_Ian-I, where the ai are

in K. Let x be a given parameter of K. By Corollary (2) of
§1.20, there exist elements p and \lI in K such that an element
8 of K (or of any extension of K) is divisible by Ax if and
only if 8'I! /p is integral over Q[x]. The lemma then implies,
because f3\l1 / p = "£ ai \lI a i / p, that Ax divides f3 if and only if
Ax divides each of the ai. Since this is true for all parameters
x, A divides f3 if and only if A divides each of the ai. Thus,
if f..L = dimK(A) and if 81, 82 , ••• ,8J.L is a basis over Ko of the
set of elements of K divisible by A, then the nf..L elements ai 8j
are a basis over Ko of the set S of all elements of L divisi-
ble by A, which shows that the dimension of S as a vector
space over Ko is nf..L = ndimK(A). The dimension of S as a
vector space over Lo is by definition dimL(A), which implies,
because [Lo: Ko] = n, that the dimension of S as a vector space
over Ko is ndimL(A). Thus ndimK(A) = ndimL(A), and the
proposition follows.
§3.24 The Genus of a Function Field.
THEOREM. Given a function field K, tbere is an integer 9 such
tbat

for all divisors A in K.

Note that A = [1] gives 1 +0 +9 > 0, so 9 ~ o. The smallest
integer 9 with the property of the theorem is called the genus of
K. The proof of the theorem will include a method of finding
the genus. It will also include a proof of the statement of §3.23
that dimK( A) < 00 for all divisors A.
The formula dimK(A) + degK(A) + 9 > 0 will be familiar to
many readers in the form dimK(A- l ) ~ degK(A) - 9 + 1. If
A is a product of m places, this formula states that the vector
space of ''functions'' with poles at most at the m places of A
(that is, elements z of K such that A[z] is integral or z = 0)
has dimension at least m - 9 + 1.
 §3.24 The Genus of a Function Field 121

PROOF: dimK(A) can be found explicitly in the case of the

divisor A = {1/x} -II, where v is a large positive integer and x
is a parameter of K found in the following way.
Let K be presented in the form K = Ko( u, v), where Ko is
the field of constants of K, and where u and v are parameters of
K which satisfy an irreducible polynomial relation F( u, v) = 0
with coefficients in Ko of degree n = [K: Ko(u)] in v, say

Here ao( u) t 0, and, because, when V is an indeterminate,

V does not divide F( u, V), an ( u) t o. Therefore, there is a
rational number e such that both ao(e) and an(e) are nonzero.
Moreover, since F( u, V) is irreducible, e can be chosen to make
the discriminant of F( e, V) nonzero as well, so that

ao(e)V n + al(e)V n- 1 + ... + an(e)

= ao(e)(V - al)(V - (2)··· (V - an)

where V is an indeterminate, where e is a rational number,

where ao(e) t 0, and where aI, a2, ... , an are distinct,
nonzero elements of an algebraic extension of Ko. (IT F(u, V),
regarded as a polynomial in V with coefficients in the field
Ko( u), had discriminant zero, then it would have a multiple
root in an algebraic extension of Ko(u), and F and its partial
derivative with respect to V would have a nontrivial common
factor over an algebraic extension of Ko(u) and would therefore
have a nontrivial common factor over Ko( u) itself, contrary to
the irreducibility of F and to unique factorization in Ko[u, V].)
With such choices of u, v, e E K, let j be the maximum
of the degrees of the polynomials ai( u) (i = 1, 2, ... , n), let
W = u - e, x = 1/w, t = an(u)/v, and y = txi. Multiplication
of (1) by t n and division by an ( u) gives
122 Divisor Theory

Multiplication by x nj then gives

which can be written

(3)

where

is in Ko[x] by the choice of j. By (3), y is integral over Q[x].

By (2), t = y/x j is integral over Q[u] = Q[u - c] = Q[l/x].
Moreover, y/x j has n distinct, nonzero values where x =
00. More precisely, ai(u)a n(u)n-i-1 = ai(w + c)an(w +
c)n-i-1 = ai(c)a n(c)n-i-1 + W"pi(W), where "pi is a poly-
nomial with coefficients in K o, which gives, by (2), 0
ao( c)a n(c)n-1 + a1 (c)a n(ct-2t + ... + t n + w"p( w, t) =
an(c)-ltn(ao(c)a n(c)nrn + a1(c)a n(ct- 1r (n-1) + ... +
an(c)) +w"p(w, t) = an(C)-ltnao(c)(an(c)r1-a1) (a n(c)r1-
(2)'" (a n(c)r1 - an) + w"p(w,t) = (-aI)(-a2) .. ·(-a n )·
an(c)-lao( c)(t - all an(c)) (t - a;l an( c)) ... (t - a;;-la n(c)) +
w"p( w, t) = (t - f31)(t - 132) ... (t - f3n) + w"p( w, t), where the
f3i = an(c )a;l are distinct, nonzero elements of the numerical
extension K' of K obtained by adjoining aI, a2, ... , an to K,
and where "p is a polynomial in two indeterminates with coef-
ficients in Ko. It follows directly from the definition of a place
that {t - f3i,W} = {~- f3i,~} = Pi is a place in K'. Since
{l/x} is a place in Ko(x), degK{l/x} = [K: K o(x)]·l = n (see
§3.21), and the places PI, P2, ... , Pn account for all n places
dividing {l/x} in a numerical extension of K.
By Corollary 2 of §1.28, there is a ~(x) E Q[x] such that ev-
eryelement of K integral over Q[x] can be written in the form
4>(x,y)/~(x), where 4> is a polynomial with coefficients in Q.
 §3.24 The Genus of a Function Field 123

By (3), it follows that every element z of K integral over Q[x]

can be written in the form z = E~=1 <pi(X)yn-i/~(x), where
the <Pi(X) are in Ko[x]. By division of polynomials, each <pi(X)
can be written in the form <Pi(X) = 8i(X)~(X) + Ei(X), where
deg Ei < deg~, which puts z in the form z = E 8i (x )yn-i +
E(Ei(X)/ ~(x))yn-i.
Let A = {l/x} -v and let S be the set of all elements z
of K divisible by A. Since Ax = [l]x, every z in S is divis-
ible by [l]x and is therefore of the form z = E8 i(x)yn-i +
E(Ei(X)/~(x))yn-i. Moreover, since the first sum on the
right is integral over Q[x] (y is integral over Q[xD, so is the
second, which implies that the n deg ~ coefficients of the e's
satisfy a certain number of linear conditions. Therefore, the
elements of the form E( Ed ~ )yn-i which are integral over
Q[x] are a finite-dimensional vector space over K o, that is,
there is a finite set (1, (2, ... , (k of elements of the form
E(Ei/ ~)yn-i integral over Q[x] such that every element of
this form integral over Q[x] can be written in one and only
one way as a1(1 + a2(2 + ... + ak(k with ai E Ko. Thus, every
z E S can be written in one and only one way in the form
E8 i(x)yn-i + Eai(i with 8i(X) E Ko[x] and ai E Ko. (If
k = 0, the second sum is absent.)
(The conditions that integrality of E(Ei(X)/~(x))yn-i im-
poses on the coefficients of the E's can be described quite ex-
plicitly. Let L be a numerical extension of K in which [~(x)] is
a product of powers of places. For each place P in the numera-
tor of [~(x)], let EEi(X)yn-i be expanded as a power series in
a local parameter at P. The coefficient of each term is a linear
combination of the coefficients of the e's, and the conditions
are that a certain finite subset of these coefficients be zero.)
If z E S, then the restriction to l/x of {l/x y[z] is inte-
gral, which is to say that z/x v is integral over Q[l/x]. When
z is written in the form z = E8 i(x)yn-i + Eai(i, the sec-
ond sum divided by XV is integral over Q[l/x] provided v is
greater than some bound depending only on x. (Since (i is in-
124 Divisor Theory

tegral over Q[x], the restriction of its denominator to x is [l]x,

so, by the proposition of §3.7, the denominator of (i is con-
centrated at {I / x}, that is, {I / x}" (i is integral for v large.)
Therefore, provided v is sufficiently large, ~ 9i( x )yn-i / XV
must be integral over Q[l/x]. Since y/x i is integral over
Q[l/x], ~9i(X)yn-i/xv = ~9i(X)X-v+i(n-i>Cy/xi)n-i is in-
tegral over Q[l/x] whenever deg9 i ::; v- j(n-i), because the
coefficient of (y/xi)n-i is then a polynomial in l/x. That this
sufficient condition is also necessary can be seen as follows.
At each of the n distinct places Pu where x = 00
(i.e., Pul{l/x}), Y /xi = f3u + terms in x-I, while 9i(x) =
xdeg OJ (Si + terms in x-I), where Si is the leading coefficient of
9i(x). Therefore, x- V9i (x)yn-i = x-v+degOj+i(n-i). (Sif3;-i +
terms in x-I). Let /-l be the maximum over i of -v + deg Bi +
j(n - i). If /-l > ° and if X-V~Bi(X)yn-i is finite at Pu ,
then the coefficients of xl! must cancel, which is to say that
I: sif3;-i = 0, where the sum is over all indices i for which
/-l = -v + deg 9i + j (n - i). This is impossible because it states
that n distinct constants f3I, f32, ... , f3n are roots of a nonzero
(by the choice of /-l) polynomial of degree less than n. There-

°
fore, if x-V I: Bi(X)yn-i is integral over Q[l/x], /-l must be::; 0,
that is, -v + degBi + j(n - i) ::; for all i.
In summary, if Alz, then, provided v is large, z =
I:Bi(X)yn-i + I:ai(i, where degBi ::; v - j(n - i). Con-
versely, every element z of K of this form is integral over
Q[x] and has the property that z/x v is integral over Q[l/x];
thus, if z =1= 0, the restrictions of {l/x }"[z] to both x and
l/x are integral, which implies that Alz. Since the expres-
sion of z in this form contains l:?:I(v - j(n - i) + 1) + k =
nv-j G)+n+k constants, this is dimK(A). On the other hand,
degK(A) = -vdegK{l/x} = -vndegQ(x){l/x} = -vn, so
dimK(A) + degK(A) = -j G) + n + k. Let

g=j(;) -n-k+1
 §3.24 The Genus of a Function Field 125

so that dimK(A) + degK(A) + g = 1. To prove the theorem,

and to prove that g is the genus of K, it will suffice to prove
that for any divisor Bin K, dimK(B)+degK(B) ~ dimK(A)+
degK(A).
Let B be a divisor in K, and let C x be the restriction to x
of the denominator of B. There is a g( x) E Q[x] such that
Cxl[g(x)]x (§1.10). For such a g(x), the restriction to x of
[g(x)]B is integral, which implies that {l/xy[g(x)]B is inte-
gral for v sufficiently large (§3.7), say {l/xy[g(x)]B = D is
integral. Since Biz if and only if [g(x)]Blg(x)z, and since mul-
tiplication by g(x) is invertible, dimK(B) = dimK([g(x)]B).
Since degK(B) = degK([g(x)]B), it follows that the desired
inequality dimK(B) + degK(B) ~ dimK(A) + degK(A) will be
proved if it is proved for [g( x )]B = D / {l/x Y = DA. In short,
with A as before (and with v large), it will suffice to prove that
dimK(DA)+degK(DA) ~ dimK(A) + degK(A) for all integral
divisors D in K. This can be done as follows.
Since there is a numerical extension L of K in which D
is a product of places, and since dimK(B) = dimL(B) and
degK(B) = degdB) for all divisors B in K, one can assume
without loss of generality that D is a product of places in K.
Every z divisible by DAis divisible by A. Therefore, the vector
space S' of elements of K divisible by DAis a subspace of the
vector space S of elements divisible by A. In fact, S' C S is
described by degK D linear conditions. (At a place P which
does not divide {I / x}, the expansion of an element z E S in
powers of a local parameter at P has no terms of degree < o.
At a place P which does divide {ljx}, the expansion has no
terms of degree < -v. If P is a place which divides D with
multiplicity e, and if z E S', then the next e terms must also
be absent-that is, there can be no terms of degree < e when
P f {ljx} and no terms of degree < -v + e when PI{l/x}.
The condition that these degK D expansion coefficients be zero
is both necessary and sufficient for z E S'.) Therefore, the
dimension of S is at most degK D more than the dimension
126 Divisor Theory

of S', that is, dimK(A) ~ dimK(DA) + degK D. Addition

of degK(A) on both sides of this inequality gives dimK(A) +
degK(A) ~ dimK(DA) + degK(DA), as was to be shown.
§3.25 Abel's Theorem.
Two divisors A and B in a function field K are equivalent if
A-I B = [z] for some z E K.
ABEL'S THEOREM. In a function field K of genus g, every
divisor of degree 9 or greater is equivalent to an integral divisor.
PROOF: If degKA ~ g, then dimK(A- I ) = dimK(A-l) +
degK(A- I ) + degK(A) ~ dimK(A-I) + degK(A- I ) + 9 > O.
Therefore, A-liz for some nonzero z E K, say [z] = A-I B,
and B is an integral divisor equivalent to A, as desired.
Since Abel's time, the name "Abel's Theorem" has been
applied to a variety of statements, some of them quite re-
mote from Abel. Abel's original statement [A] was roughly
the following: Let C be an algebraic curve in the xy-plane,
let f(x, y) be a rational function on C, and let ¢(P) de-
note the many-valued ''function'' ¢(P) = J~ f(x, y) dx on
C, which depends on the lower limit of integration 0 and
the choice of a path of integration from 0 to P, as well as
on P. Abel stated that when PI, P2 , ••• , Pp. satisfy a cer-
tain set of 9 algebraic conditions (for some 9 ~ 0) the sum
of J-l integrals ¢(Pt) + ¢(P2 ) + ... + ¢(Pp.) can be expressed
as the integral of a rational function. He proved it by con-
structing a family of rational functions (} on C depending
on a set of parameters a, a', a", ... , such that the (}'s all
have the same J-l poles, but their J-l zeros depend on the pa-
rameters a, a', a", .... He then showed that the sum of
f(xi(a),Yi(a))dxi = f(xi(a),Yi(a))xi(a)da over the J-l zeros
(Xi( a), Yi( a)) of the (} corresponding to a given value of the a's is
a rational differential in the a's (because, in essence, this sum
 §3.25 Abel's Theorem 127

is a trace). Integration then gives

(1) t 1~i
,=1 Q.
f (x, y) dx = J
rational differential in the a's

where QI, Q2, ... , Qp. and PI, P2, ... , Pp. are the zeros on
C of the B's corresponding to two different sets of values of
the parameters a, a', a", .... When the Q's are fixed, the
left side of this equation is of the form ~'IjJ(Pi) + const. and
the equation says that ~ 'IjJ(Pi ) is an indefinite integral of a
rational function of the a's, when the P's are the zeros of the
B corresponding to some set of values of a, a' ,a", ... .
For any set of values of the parameters a, a', a", ... , the zero
set PI, P2, ... , Pp. of B is equivalent, as a divisor, to the pole
set, which is the same for all sets of values of the parameters.
Therefore, the condition on the P's implies that the divisors
PI P2 ··· Pp. are all equivalent. When the a's provide enough
degrees of freedom, the only condition on the P's is that these
divisors be equivalent. Thus, when PI P2 ... Pp. is equivalent
to a fixed divisor, ~ 'IjJ( Pi) can be expressed as an indefinite
integral of a rational function of the a's.
From this point of view, the essence of Abel's original the-
orem is the statement that, given a set of places QI, Q2, ... ,
Q p. on an algebraic curve C, the nearby sets of places PI, P2,
... , Pp. such that the divisors PI P2 ··· Pp. and QIQ2··· Qp. are
equivalent can be described by 9 algebraic conditions, for some
g. The minimum value of 9 is the genus of C, and the desired
conditions can be derived from "Abe1's Theorem" above as fol-
lows. Given PI, P2, ... , Pp._g, there is, by "Abel's Theorem",
an integral divisor of degree g, call it R I R 2 ••• R g , which is
equivalent to (QIQ2···Qp.)/(PI P2 ···Pp.-g). The R's depend
on PI, P2, ... , Pp.-g, and the 9 conditions are Pp.-g+i = Ri
for i = 1, 2, ... , g.
128 Divisor Theory

§3.26 The Genus as a Limit.

PROPOSITION. Let A and B be divisors in a function field
K, and let degK A be positive. Then 1 - dimK(A-1L B) -
degK(A -IL B) is the genus of K for all sufficiently large integers
Jl.
PROOF: The case A = {1/x}, B = [1] of this proposition,
where x is a parameter of K of the special type considered in
§3.24, was proved in §3.24. IT A is any divisor with positive de-
gree, and if v is any integer, then degK(AILB- 1{1/x}-1I) > 9
for Jl sufficiently large. By Abel's theorem, AIL B-1 {1/x} -11 is
equivalent to an integral divisor, say AILB-1{1/x}-1I[z] = C,
where C is integral and z is a nonzero element of K. When v
is large enough that 9 = 1-dimK( {1/x} -lI)-degK( {1/x} -11),
1-g = dimK(CB/AIL)+degK(CB/AIL). The space of elements
of K divisible by CB/AIL is a subspace of the space of ele-
ments of K divisible by B / AIL defined by degK( C) linear con-
ditions. Therefore dimK(CB/AIL) ~ dimK(B/AIL)-degK(C),
which combines with the equation above to give 1 - 9 ~
dimK(B/AIL) - degK(C) + degK(C) + degK(B/AIL). Since
dimK(B/AIL) + degK(B/AIL) ~ 1 - 9 by the definition of the
genus, 1 - 9 = dimK(B/AIL) + degK(B/AIL), as was to be
shown.
COROLLARY. The genus of a numerical extension of a function
field K is equal to the genus of K.
DEDUCTION: An integral divisor A in K is also an integral
divisor in any numerical extension L of K. By the proposition,
both the genus of K and the genus of L are equal to 1 -
dimK(A-IL) - degK(A-IL) for all sufficiently large Jl.
§3.27.
PROPOSITION. Let K be a function field of genus g. There is
an integral divisor A of degree 9 in a numerical extension K'
 §3.27 A Converse of Abel's Theorem 129

of K with the property that the only integral divisor in K' or

in a numerical extension of K' equivalent to A is A itself.
PROOF: As before, let Ko be the field of constants of K, let
x be a parameter of K, let v be an integer large enough that
dimK{l/x} -v = nv - 9 + 1 where n = dimK{x}, and let S be
the vector space of elements of K divisible by {l/x} -v.
Let PI, P2, ... , P j be a set of places in K which do not
divide {l/x} and which have the property that the mapping
S - t Kg obtained by evaluating elements of S at these j places
is onto. (The empty set of places is to be regarded as satisfying
this condition.) The main step in the proof is to show that if
j < nv - 9 + 1 then, by passing to a numerical extension of K if
necessary, one can find another place Pj +I such that evaluation
at PI, P2, ... , Pj+I again gives an onto mapping S - t Kt+ I .
For this, note that j < nv - 9 + 1 means dimKt < dimS,
so there is a nonzero z E S whose values at PI, P2, ... , Pj
are all zero. In a numerical extension K' of K, {l/x y[z] =
P I P2 ··· PjQIQ2··· Qnv-j, where the Qi are places. Let c be
a rational number different from the values of x at the places
PI, P2, ... , Pj, QI, Q2, ... Qnv-j. (The value of x at Qi may
be 00.) Let Pj+I be a place in a numerical extension K" of K'
which divides {x - c}. By the choice of c, Pj +I does not divide
{I / x} and z is nonzero at Pj+ 1. Let K be replaced by K".
Then S - t Kt is still onto. Moreover, the map S - t Kt+ I is
also onto, because its kernel is a proper subspace of the kernel
of S - t Kg (because z is not in it), so its image is of dimension
greater than j .
Thus, by induction on j, K can be extended numerically
to find a function field K and places PI, P2, ... , Pnv - g+1
in K such that evaluation at these places gives a map S - t
K;v-g+I which is onto. Let z be a nonzero element of S whose
values at PI, P2, ... , Pnv - g are all zero, and let {l/xy[z] =
PI P2 ... Pnv - gQl Q2 ... Q g. Let B be an integral divisor in a
numerical extension of K which is equivalent to QIQ2··· Qg.
130 Divisor Theory

Then P1 P2 ••• Pnv-gB{l/x} -v is equivalent to [z], which is

equivalent to [1], so there is a w in a numerical extension of
K such that {l/x y[w] = P1 P2 ••• Pnv-gB. Thus, when K is
extended numerically, w is in S and is in the kernel of the map
S --? K;v- g given by evaluation at PI, P2 , ••• , Pnv - g' This
kernel is one-dimensional, so w is z times an element of K o.
Thus, [w] = [z], which implies B = QIQ2'" Qg. Therefore
A = Ql Q2 ... Q 9 is a divisor with the required properties.
COROLLARY. There is a divisor of degree 9 -1 in a numerical
extension of K which is not equivalent to an integral divisor.
(It follows, by Abel's theorem, that 9 can be characterized
as the largest integer such that some divisor of degree 9 - 1 in
some numerical extension of K is not equivalent to an integral
divisor.)
DEDUCTION: Let A be as in the proposition, and let P be a
place in a numerical extension of K which does not divide A.
Then AlPhas the required property.
§3.28 The Divisor Class Group.
Equivalence of divisors (§3.25) is obviously consistent with
multiplication of divisors, and the resulting multiplication of
equivalence classes obviously gives to them the structure of a
commutative group. This is the divisor class group of K.
Equivalent divisors have equal degrees, so equivalence classes
of divisors have degrees. The assignment to divisor classes of
their degrees is a homomorphism from the divisor class group
to Z.
Let a be a divisor in K of degree g. Every divisor of de-
gree zero in K is, by Abel's theorem, of the form Ala where
A is an integral divisor of degree g. Therefore, the mapping
which sends an integral divisor A of degree 9 to the class of
AI a is a map of such divisors onto the subgroup of the di-
visor class group consisting of divisor classes of degree zero.
Thus, elements of the divisor class group of degree zero can be
 §3.29 Examples 131

identified, given 0, with equivalence classes of integral divisors

of degree g. When this identification is made, the group law
(AIO)(B 10) '" C10 takes the form

(1) AB", CO

that is, the product of the elements of the divisor class group
represented by A and B (relative to 0) is represented by any
integral divisor C for which (1) is satisfied. Abel's theorem
implies that for integral divisors A and B of degree g, and for
a divisor 0 of degree g, there is an integral divisor C for which
(1) holds.
§3.29 Examples.
IT K = Q(x) and A = {llx}, then z E K is divisible by
A -v if and only if z is a polynomial of degree v, at most, in x.
The genus of this K is therefore equal to 1- dimK( {llx} -V)_
degK({llx}-V) = 1- (v + 1) - v = 0, as is the genus of any
numerical extension of Q( x).
Let K be a function field of genus O. Let P and Q be distinct
places in a nu~nerical extension L of K. By Abel's theorem,
PI Q is equivalent to an integral divisor, say [z] = PI QR,
where R is integral. Then {z} = P, which gives

1 = degL{z} = [L:Q(z)][Lo:Q]-ldegQ(z){z}
= [L: Lo(z)][Lo(z): Q(z)][L o: Qr l = [L: Lo(z)]

(§3.21), where Lo is the field of constants of L. Therefore,

L = Lo(z) is a numerical extension of Q(z). Thus, a function
field has genus 0 if and only if it has a numerical extension
isomorphic to a numerical extension of Q(x).
Let K be a function field of genus 1, and let P, Q, and
R be distinct places in a numerical extension L of K. By
Abel's theorem, PQ IRis equivalent to an integral divisor, say
[z] = PQIRS, where S is integral. Then {z} = PQ, and,
132 Divisor Theory

by the computation above, 2 = degL{z} = [L:Lo(z)], which

shows that L is a quadratic extension of a numerical extension
Lo(z) of Q(z). Thus, a generator y of Lover Lo(z) satisfies
an equation of the form ao(z)y2 + al(z)y + a2(z) = 0. Then
w = 2ao(z)y + al(z) is also a generator of Lover Lo(z), and
w 2 = al(z)2 - 4aO(z)a2(z). Therefore, a numerical extension
of K is equal to the field obtained by adjoining the square root
of a polynomial to a numerical extension of Q(z).
Let Lo be an algebraic number field, let b(x) E Lo[x], and
let L be the field obtained by adjoining a root y of y2 - b( x )
to Lo (x ). IT b( x) has odd degree 2j - 1, and if e is a rational
number such that bee) ¥- 0, then, with x = c + u- 1 , (uiy? =
u 2i b(c + u- I ) is a polynomial in u of degree 2j. Thus, L can
be obtained by adjoining to Lo(x) = Lo(u) a square root of
a polynomial of degree 2j, and there is no loss of generality
in assuming that the original b( x) is of even degree 2j. By
passing to a numerical extension of L, if necessary, one can
also assume that b( x) is a product of monic, linear factors
b(x) = (x - et)(x - e2)··· (x - e2i). Finally, the factors x - ei
can be assumed to be distinct, because adjoining a square root
of (x - e)2 b( x) is the same as adjoining a square root of b( x).
In summary, every function field K of genus 1 has a numerical
extension of the form Lo(x, .jb(x)), where b(x) = (x - el)(x-
e2) ... (x - e2i) is a product of distinct factors.
The genus of a function field of the form Lo (x, .jb( x)) just
described is j - 1, as can be proved as follows. Because
(y/x i )2 = 1 + terms in x-I, the values of y/xi at x = 00
are ± 1, and 9 = j (~) - 2 - k + 1 = j - 1 - k by the for-
mula of §3.24, where k is the dimension over Lo of the vec-
tor space of elements of L of the form (4)( x) + y?jJ( x )) / 6.( x)
which are integral over Q[x] (deg4> < deg6., deg?jJ < deg6.).
IT z = (4)(x) + y?jJ(x))/6.(x) is integral over Q[x], then
4>(x) + y?jJ(x) - z6.(x) = 0, where 1, yare a basis of L over the
field of quotients Lo(x) of the natural ring Lo[x], where 1, y, z
are integral over Lo[x), and where 4>(x), ?jJ(x), 6.(x) E Lo[x) are
 §3.29 Examples 133

relatively prime. By Corollary 1 of §1.30, ~(x)21 discL(l, y).

By the proposition of §1.30, discL(l, y) is the norm of [F'(y)]
where F(X) = NL/Lo(x)(X - y) = X 2 - b(x), that is,
disq(l, y) = NL/Lo(x) [2y] = [-4b(x)]. Since b(x) has no
square factors, it follows that ~(x) is a unit of Lo[x], and
therefore that </>(x) and ¢(x) are o. Therefore, k = 0 and
9 = j - 1, as claimed.
In conclusion, then, every function field of genus 1 has a
numerical extension of the form Lo( x, .Jb( x )), where Lo is an
algebraic number field and b( x) is a product of 4 distinct factors
(x - eI)(x - e2)(x - e3)(x - e4) with ei E Lo. Conversely, any
function field with a numerical extension of this form has genus
1. Such a function field is called elliptic.
In the case of an elliptic function field K, since 9 = 1, the
construction of §3.28 gives a group structure to the set of places
in K, once a place 0 is chosen to play the role of the identity.
The usual geometrical description of this group of places is the
following. A simple change of variables can be used to put the
defining equation of K in the form y2 = al x 3 +a2x2 + a3x +a4
where the ai are constants of K and al =f o. (Given the
defining equation in the form y2 = (x - el )( x - e2)( x - e3) (x -
e4), set x' = (x - e4)-I and y' = y(x - e4)-2.) With u = l/x
and v = y/x 2, one then finds v2 = alu + a2u2 + a3u3 + a4u\
from which it follows that {u, v} is a place in K at which v is a
local parameter and at which u has a double zero. Therefore,
with 0 = {u, v}, x has a double pole at 0 and y = vx 2 has a
triple pole there. In fact, the denominator of [x] is 0 2 , because
{ x} is a place in K 0 ( x) C K and this extension has degree
[K: Ko(x)] = 2 (the extension is generated by the element
y of degree 2 over Ko(x)), so degK{l/x} = degK{x} = 2
(§3.21). Similarly, {l/y} = 0 3 • Thus, for any bI , b2 , b3 in the
field of constants Ko of K, the denominator of the divisor of
</> = bI + b2 x + b3 y is 0 3 provided b3 =f O. The denominator of
the divisor of </> = bI +x is 0 2 • Given two places P, Q in K (not
necessarily distinct, but neither equal to 0) there exist bI, b2 ,
134 Divisor Theory

b3 E Ko such that ¢ = bI +b2 x+b3 y is not constant and is zero

at P and Q (2 homogeneous linear equations in 3 unknowns of
which 1, 0, 0 is not a solution). Thus [¢] = PQR/0 3 , where
R is a place in K. Geometrically, R is the third point in which
the line ¢ = 0 in the xy-plane through P and Q intersects the
given curve. (If b3 = 0, then R = 0.) Let c be the value of x at
R (unless R = 0) and let the equation [x - c) = R8/0 2 define
a place 8 on the curve. Geometrically, 8 is the other point
in which the vertical line in the xy-plane through R intersects
the curve. Then PQR/0 3 [1] and R8/0 2 [1], which gives
IV IV

PQR 0 3 R80. Therefore, PQ 08, which means that

IV IV IV

8 is the product of P and Q in the group in question. (If

R = 0, then PQO 0 3 , PQ 0 2 , so 0 is the product of P
IV IV

and Q. If either P or Q is 0, their product is the other one.)

A function field of the form Lo(x, ..jb(x)) in which b(x) is
a product of more than 4 distinct linear factors in Lo[x] is
called hyperelliptic. As the above argument shows, there are
hyperelliptic function fields with any genus greater than 1.
A simple example of a function field of higher genus which
is not hyperelliptic is the field K of rational functions on the
so-called Klein curve defined by the equation y3 + x 3 y + X =
o. Clearly P = {x, y} is a place in K at which y is a local
parameter and at which x has a zero of order 3. In fact, since
degK{x} = 3 degQ(x) {x} = 3, {x} = p 3 • Writing the equation
of the Klein curve in the homogeneous form y3 Z + X 3 Y +
Z3 X = 0 with y = Y/Z and x = X/Z brings out a symmetry
which shows that Q = {Z/Y,X/Y} = {l/y, x/y} is a place in
K where x/y is a local parameter and l/y has a zero of order
3, while R = {Y/X, Z/X} = {y/x, I/x} is a place in K where
I/x is a local parameter and y/x has a zero of order 3. Thus
l/x has a simple zero at R and, since I/x = (I/y)/(x/y), it
has a double zero at Q. Therefore, {I/x} = Q2 R. It will be
shown that the genus of K is 3 by applying the Proposition of
§3.26 in the case of A = Q2 R = {I/x} and B = [1] .
 §3.29 Examples 135

Let S be the subspace of elements of K divisible by A -I' for

some large /1. An element z of K is in S if and only if z is
integral over Q[x] and has poles of order 2/1 at most at Q and
of order /1 at most at R. Because 1, y, y2 are a basis of Kover
Q(x), z satisfies a relation of the form ¢(x)z = 6(x)+6(x)y+
6(x)y2, where the coefficients ¢(x) and ei(X) are in Q[x] and
are relatively prime. One can assume, moreover, that ¢( x) is
monic. By Corollary 1 of §1.30, ¢(x)2 divides disc(l, y, y2).
By the proposition of §1.30, this discriminant is N K(3y2 + x 3 ),
which is
o 3
-2x 3 o =4x 9 +27x 2.
-3x -2x 3
Since the only monic square factor of this discriminant is x 2 ,
¢(x) is 1 or x. If it were x, then 6(x) + 6(x)y + 6(x)y2
would be zero at P which is only possible, because y has a
simple zero at P and x has a zero of order 3 there, when the
ei are all divisible by x, contrary to the assumption that ¢(x)
and the ei(X) are relatively prime. Therefore z = 6(x) +
6(x)Y+6(x)y2. Let a, b, c be the degrees of 6,6,6. At the
place Q, x has a pole of order 2 and 1/ y has a zero of order
3, that is, y has a pole of order 3. The individual terms of
z = 6 (x) + e2 (x)y + e3 (x )y2 therefore have poles of order 2a,
2b + 3, 2c + 6 respectively at Q, while z has a pole of order
2/1, at most, at Q. At the place R, x has a pole of order 1 and
y = (y/x )/(l/x) has a zero of order 2, so the individual terms
of z have poles of order a, b-2, c-4, respectively, while z has a
pole of order /1, at most. If a were greater than /1, 6 (x) would
have a pole of order 2a > 2/1 at Q. Therefore, 6 (x )y +6 (X )y2
would have to have a pole of order exactly 2a at Q; since 2b +3
is odd, it would follow that 2c + 6 = 2a and 2c + 6 > 2b + 3.
But then z would have a pole of order a > /1 at R, because
a > b and a > c. Since this is not the case, a ~ /1. Then, since
6 (x )y + 6 (X )y2 has a pole of order 2/1 at most at Q, and since
2b + 3 is odd and 2c + 6 is even, 2b + 3 ~ 2/1 and 2c + 6 ~ 2/1.
136 Divisor Theory

Thus, a ~ J.l, b ~ J.l - 2, c ~ J.l - 3. Conversely, for such a,

b, c, the sum z = 6(x) + 6(x)y + 6(x)y2 is in S because
each term is in S. Thus 9 = 1 - dimK(A-Il) - degK(A-Il) =
1 - ((J.l + 1) + (J.l- 1) + (J.l- 2)) + 3J.l = 3.
 Appendix: Differentials

§A.1.
Let PI, P2, ... , Pk be places in a function field K. The
poles of an element z of K will be said to be concentrated at
Pl P2 ·· ,Pk if {liz} is concentrated at Pl P2 ",Pk (see §3.6).
Let Ui be a local parameter at Pi (i = 1, 2, ... , k). The
condition "z = 7]i + a~i)uil + a~i)ui2 + ... where 7]i is finite
at Pi" determines a polynomial a~i) X + a~i) X 2 + ... = <pi(X)
in X with coefficients in the field of constants Ko of K and
with constant term zero. If the poles of z are concentrated
at Pl P2 ... Pk, the polynomials <Pl(X), <P2(X), ... , <Pk(X) are
called the principal parts of z (relative to the parameters Ui).
The problem dealt with in this appendix is: Given a set of
polynomials <Pl(X), <P2(X), ... , <pk(X) (with coefficients in
the field of constants Ko of K and with constant terms zero)
when are they the principal parts (relative to the Ui) of an
element of K with poles concentrated at Pl P2 ··· Pk? Using
differentials one can find conditions on k-tuples (<PI, <P2, ... ,
<Pk) which are necessary and sufficient for them to be principal
parts. The Riemann-Roch theorem is a simple corollary of this
description of the principal parts.
§A.2 Definitions and First Propositions.
Let K be a function field. Differentials in K are represented
by expressions of the form Xldx2 + X3dX4 + ... + X2k-ldx2k,
where k is a positive integer and Xl, X2, ... , X2k are elements
of K. Such an expression represents the zero differential, de-
noted xldx2+X3dx4+" '+X2k-ldx2k = 0, if, at every place P
of every numerical extension of K, either (1) one of Xl, X2, ... ,
X2k is not finite at P, or (2) they are all finite at P, and, for
a local parameter t at P, when constants ai, bi are defined by
Xi = ai+bit+O(t), the equation al b2+a3b4 +.. ·+a2k-lb2k = 0
holds. (The equation Xi = ai + bit +o( t) means, of course, that
(Xi -ai - bit)lt is zero at P.) If al b2+a3b4 + .. ·+a2k-lb2k = 0
138 Divisor Theory

holds for one local parameter t at P, then it holds for all, be-
cause, if s is another, s = 'Yt+o(t), t = 'Y-ls+o(s) where 'Y is a
constant of K, 'Y =f 0, and Xi = ai+ bi('Y- 1s + 0(s))+0(s) = ai+
'Y-IbiS + o(s), so that the desired relation is Ea2i-l'Y-lb2i =
'Y- 1 Ea2i-lb2i = 0, which holds whenever Ea2i-Ib2i = o.
Two expressions E X2i-l dX2i and E Y2i-1 dY2i represent the
same differential, denoted E X2i-I dx2i = E Y2i-IdY2i, if
E X2i-1 dX2i + E( -Y2i-l )dY2i represents the zero differential.
PROPOSITIONS. (1) EX2i-ldx2i = EY2i-ldY2i is an equiva-
lence relation consistent with addition and with multiplication
by elements of K.
(2) If X E K is constant, then dx = O.
(3) d(x + y) = dx + dy.
(4) d(xy) = xdy + ydx.
(5) Let a parameter x of K be given. There is an integral
divisor C in K such that, for every place P in every numerical
extension of K, either x is not finite at P or x - a is a local
parameter at P for some constant a of K or PIC.
(6) If dx = 0, then x is a constant of K.
(7) Given Xl, X2, ... , X2k and Y in K, where Y is not con-
stant, E X2i-l dX2i = z dy for a unique z in K.
(8) If K and L are function fields with K C L, and if
E X2i-l dX2i = E Y2i-1 dY2i holds in K, then E X2i-l dX2i =
E Y2i-1 dY2i holds in L.
Note that Proposition (7) gives an algorithm for determining
whether a given differential is zero; one need only write it in
the form zdy, where y is not constant, and note that, by (1),
(2), and (6), z dy = 0 if and only if z = O.
PROOFS: (1) These properties are all easy to deduce from the
definition. (2) IT X is constant then x is finite at all places P
in all numerical extensions of K and its expansion in terms of
any local parameter t at P is x = x + 0 . t + 0, so E a2i-1 b2i in
the case of 1· dx is 1· 0 = 0 for all P. (3) and (4) At any place
P in any numerical extension of K at which x and yare both
 §A.2 Definitions and First Propositions 139

finite, x = a + bt + wand y = a' + b't + w', where t is a local

parameter at P, ordp(w) > 1, ordp(w') > 1, and a, a', b, b'
are constants. Then x + y = (a + a') + (b + b')t + (w + w') and
xy = aa' + (ab' + a'b)t + (aw' + a'w), from which d(x + y) -
dx - dy = 0 and d(xy) - x dy - y dx = 0 follow. (5) Since x
is a parameter of K, K is generated over Q( x) by an element
y E K which satisfies an equation of the form F(x, y) = 0,
where F(X, Y) = yn+Cl(x)yn-l+ ... +cn(X) is a polynomial
with coefficients which are constants of K (§3.2, §1.25). Let
Fy(X,Y) = nyn-l+(n-l)cl(x)yn-2+ .. '+C n-l(X), Since
Fy(x, y) =f 0 (F can be assumed to have minimal degree in Y),
{Fy(x, y)} is an integral divisor. Let C = {Fy(x, y)}. Let P
be a place in a numerical extension K' of K at which x is finite.
Then, since y is integral over Q [x], y is fini te at P. Then PI { x-
a, y - a'} where a and a' are the values of x and y, respectively,
at P. Expansion of 0 = F(x, y) = F(a + (x - a),a' + (y - a'))
gives a polynomial relation 0 = Q + (3(x - a) + ,(y - a') + ...
satisfied by x - a and y - a', where Q = F( a, a') is the value
of F(x,y) = 0 at P-that is, Q = D-and, = Fy(a,a') is
the value of Fy(x,y) at P. IT P f C then, =f 0 and {x-
a, y - a'} is, directly from the definition, a place in K'. Since
PI{x - a,y - a'}, it follows that P = {x - a,y - a'}. IT P
divided x - a twice, then all terms of 0 = (3( x - a) + ,(y -
a') + ... would be divisible by p 2 with the possible exception
of ,(y - a'), which would imply P21,(y - a') and therefore
P21{x - a,y - a'} = P, P = [1], which is not the case. Thus,
ord p ( x - a) = 1, and x - a is a local parameter at P whenever
P f C and P f {1/x}, as was to be shown. (6) Assume x
is not constant and let C be as in (5). Let z be a parameter
of K which supports {l/x}C (§3.8) and let P be a place in
a numerical extension K' of K which divides {I/ z} (§3.I8).
Then P is relatively prime to {1/x} and C, so x is finite at P
and P f C. Therefore, by (5), x - a is a local parameter at P
for some constant a of K' (the value of x at P). The equation
x = a + (x - a) then shows that dx = 1 . dx does not represent
140 Divisor Theory

the zero differential. Therefore, x not constant implies dx =f 0,

as was to be shown. (7) Consider first the case k = 1, Xl = l.
Since K is an algebraic extension of Q(y), X2 and y satisfy
a relation of the form G(X2, y) = 0, where G is a polynomial
with coefficients in Z. Since 0 is constant, d(G(X2, y» = 0
by (2). By (2), (3), and (4), d( G(X2' y» represents the same
differential as GX(X2, y)dX2 +GY(X2, y)dy, where Gx and Gy
are polynomials, and the degree of G x (X, Y) in X is less than
that of G(X, Y). Because the degree of G in X can be assumed
to be minimal, G x( X2, y) can be assumed to be nonzero. Thus,
dX2 = zdy, where z = -GY(X2' y)/G X (X2, y). For k ~ 1,
dX2i = z2idy, where Z2, Z4, ... , Z2k E K, and Xldx2 + ... +
X2k-Idx2k = z dy where z = XIZ2 + ... + X2k-IZ2k. That z is
uniquely determined follows from (6). (8) It has been shown
that z dy = 0 if and only if z = 0 or y = constant. Therefore,
z dy = 0 in K if and only if z dy = 0 in L.
§A.3 Orders and Residues of Differentials.
Let L X2i-1 dX2i represent a differential in a function field
K, let P be a place in K, and let t be a local parameter at P.
By Proposition (6) above, L X2i-Idx2i = u dt for some u E K.
The order of the differential L X2i-1 dX2i at P relative to t is
defined to be ord P ( u ).
PROPOSITION (9). The order of a differential at P relative to
t is the same for alllocal parameters t at P.
DEFINITION. The order of a differential LX2i-ldx2i at P,
denoted ordp(L X2i-1 dX2i), is its order at P relative to a local
parameter at P.
PROOF: Let P = {x, y} be a presentation of P as a place
(§3.13). Assume without loss of generality that X is a local
parameter at P, and let z E K be such that z dx represents
the given differential. If t is any local parameter at P, then
ord p( t / x) = ord p t - ord p x = 1 - 1 = 0, so tx -1 = a +
S(x,y)(c+c(x,y»-I, where a and c are nonzero constants of
 §A.3 Orders and Residues of Differentials 141

K and 6 and c are polynomials with no constant term. Multi-

plication of this relation by x( e + c( x, y)) gives t( e + c( x, y )) =
xa(e+c(x,y))+x6(x,y). Thus edt+c(x,y)dt+ iI(x,y,t)dx+
hex, y, t)dy = aedx + 91(X, y)dx + 92(X, y)dy, where iI, 12,
91, 92 are polynomials with no constant term. On the other
hand, differentiation of the relation F( x, y) = 0 which presents
{x, y} as a place gives Fxdx + Fydy = 0, where Fx and Fy
are polynomials in x and y and where Fy has nonzero con-
stant term. Thus, dy = q dx where q = -Fx / Fy is finite at
P. Since (e + c)dt = (ae - iI - qi2 + 91 + q92)dx, it follows
that dt = q1dx where q1 = (ae- iI -qh +91 +q92)/(e+C) is
finite, with value ae/e = a =f 0, at P. Thus, the given differen-
tial L: X2i-1 dX2i is represented by zql 1dt. Since ordp ( zQl1) =
ordp z - ordp Q1 = ordp z, the proposition follows.
Note that if u dy is a differential in K, if z E K, and if P is
a place in K, then ordp(zu dy) = ordp(z) + ordp(u dy).
PROPOSITION (10). Let P be a place in a function field K,
and let z E K have order J.l > 0 at P. Then dz has order J.l - 1
at P.
PROOF: IT x is a local parameter at P, then z = cx"'+w, where
e is a nonzero constant of K and ordp w = v > J.l > O. Let
w' = w/x,,-l. Then z = ex'" + x,,-l w', dz = eJ.lx",-ldx + (v-
I)X,,-2 W 'dx+x,,-ldw'. Since w' and x are local parameters at
P, the last two differentials in the sum on the right have order
v-I at P. The first has order J.l - 1 < v-I, so dz has order
J.l - 1, as was to be shown.
Let u dy represent a differential in a function field K, and let
t be a local parameter at P. The residue of u dy at P relative
to t is defined to be the coefficient of r1 in the expansion of
v in powers of t, where v is defined by u dy = v dt.
PROPOSITION (11). The residue of a differential at P relative
to t is the same for all local parameters t at P.
DEFINITIO N. The residue of a differential u dy at P, denoted
142 Divisor Theory

resp( u dy), is its residue at P relative to a local parameter at

P.
PROOF: Note first that if ordp(udy) ~ 0 then the residue
of u dy at P relative to t is 0 for any local parameter t at
P. Assume, therefore, that ordp ( u dy) = v < 0, say u dy =
(ayt Y + ay-I t y- I + ... + a_I rl + w)dt, where t is a local
parameter at P, the ai are constants of K with a y =1= 0, and
ordp w ~ O. If t' is another local parameter at P, and if
resp( v dz) denotes the residue of v dz at P relative to t', then
resp( u dy) = ay resp(t dt)+ . .. +a-I resp(t-Idt) +resp( w dt).
Y

The last term on the right is zero because ordp( w dt) ~ O.

That resp(tjdt) = 0 for j < -1 can be proved as follows. Let
ti+ I I (j + 1) be expanded in powers of t', say ti+ I I (j + 1) =
Ck(t,)k + Ck+l(t,)k+1 + ... + CN(t,)N + s, where the c's are
constants of K and ordp s > N. Then tjdt = d(t j+1 l(j + 1)) =
(kCk(t,)k-1 +.. ·+NCN(t,)N-I )dt' +ds. Since kCk(t,)k-1 +... +
NCN(t,)N-I has no term in (t')-1, and since resp(ds) = 0
when N 2 0, resp(tjdt) = O. Finally, tit' = a + v, where a is
a nonzero constant of K and ordp v > O. Then t = (a + v)t'
and dtlt = d(a + v)/(a + v) + dt'lt'. Since ordp(a + v) =
0, ordp(d(a + v)/(a + v)) = ordp(d(a + v)) = ordp(dv) =
ordpv-1> -1. Therefore, resp(dtlt) = resp(dt'lt') = 1,
and resp(udy) = a_I = resp(udy), as was to be shown.
§A.4 A Fundamental Theorem.
The sum of the residues of a differential in a function field
is zero. The following proposition makes possible a precise
formulation of this basic fact.
PROPOSITION. Let u dx represent a differential in a function
field K. There is an integral divisor B in K such that every
place P in every numerical extension of K satisfies either PIB
or ordp(u dx) = O.
THEOREM. With K, u dx, and B as in the proposition,
let K' be a numerical extension of K in which B is a product
 §A.4 A Fundamental Theorem 143

of places. Then
L resp(u dx) = 0
PIB

where the sum is over all places of K' which divide B.

PROOF OF THE PROPOSITION: Let C be an integral divisor
as in (5) of §A.2, and let B = {u }{1/u }{1/ x }C. If P is a
place in a numerical extension of K which does not divide
B then u and x are both finite at P and u is nonzero at P.
Moreover, x - a is a local parameter at P, where a is the value
of x at P, so dx = d( x - a) has order 0 at P. Therefore,
ordp( u dx) = ordp( u) + ordp( dx) = 0, as was to be shown.
The proof of the theorem will use three lemmas.
LEMMA 1. Let u and x be elements of a function field K
such that u dx is a differential whose poles are concentrated at
{1/x}. Then the trace ofu relative to the extension K ~ Ko(x)
is in Ko[x]. (Here Ko is the field of constants of K, Ko[x] c K
is the ring of polynomials in x with coefficients in K o, and
Ko(x) C K is the field of quotients of Ko[x].)
PROOF: Let the trace of u relative to K ~ Ko(x) be
</>(x)/'IjJ(x), where </>(x), 'IjJ(x) E Ko[x] are relatively prime. In
a numerical extension K' of K, {'IjJ(x)} is a product of places.
If the poles of u dx are concentrated at {1 / x}, then, at each
place P in K' dividing {'IjJ(x)}, ordp((x-a)u) = ordp(x-a)+
ordp u = ord p (d(x-a))+I+ord p u = ord p (dx)+I+ordp u =
l+ord p (u dx) ;:::: 1, where a is the value of x at P. Thus (x-a)u
is zero at P. Since (x -a)I'IjJ(x) in Ko[x], it follows that 'IjJ(x)u
is zero at P. In particular, 'IjJ( x)u is finite at P. Since this is
true for all places P in K' dividing {'IjJ( x)}, 'IjJ( x)u is finite at
{'IjJ( x)}. If {'IjJ( x)} were not [1], then, since 'IjJ( x)j u j is zero at
pj for allj > 0, there would be an integer j such that 'IjJ(x)ju j
was zero at {'IjJ( x)}. The trace of u can be written in the form
UI +U2 + ... + Un where the Ui are conjugates of u in a function
field L ~ K (see §1.25). Each 'IjJ(X)Ui is finite at {'IjJ(x)} and
144 Divisor Theory

each tP(x)juf would be zero at {tP(x)} if {tP(x)} =f [1]. In this

case, <p( x )nj = (tP( x)( Ul + U2 + ... + un)) nj would be a sum of
terms, each of which was a product of an element of L finite
at {tP(x)} and an element zero at {tP(x)}, so <p(x)n j would be
zero at {tP (x) }, contrary to the assumption that <p( x) and tP (x)
are relatively prime. Therefore, {tP( x)} = [1], which implies
degtP = 0, and tru = <p/tP E Ko[x], as was to be shown.
LEMMA 2. Given any finite set of distinct places PI, P 2 , ••• ,
Pp. in a function field K, there is an element z of a numerical
extension K' of K such that {I / z} is a product of distinct
places in K' including all the given places PI, P 2 , ••• , Pp.-
PROOF: Let x be a parameter of K. For v sufficiently large,
dime {l/x} -II) = 1 - deg( {l/x} -II) - 9 = nv - 9 + 1, where n
is the degree of {x} in K and 9 is the genus of K. There are
finitely many rational numbers c such that {x -c} ramifies (i.e.,
is divisible by the square of a place in a numerical extension
of K) and finitely many such that one of the given places Pi
divides {x - c}. Therefore, there exist constants Cl, C2, ... , cll ,
CII +1 of K such that the {x - cd are unramified and relatively
prime to P 1 P2 ••• Pp.. Let K' be a numerical extension of K
in which the {x - Ci} are all products of places, and let Y =
(x - Cl)(X - C2)··· (x - CII). Then [y] = A{l/x} -II, where A is
a product of nv distinct places in K', say A = QIQ2··· Qnll,
all distinct from the given places Pi. Finally, let R be a place
in K' dividing {x - CII +I}; then R is a place in K' distinct from
all the P's and all the Q's.
Let V be the (nv - 9 + 1)-dimensional vector space (over
the field of constants Kl of K') of elements of K' divisible by
{1/x}-II. The elements of V are finite at Qi for all i, so eval-
uation at these nv places defines a linear function V ~ Ki".
An element of the kernel of this linear function is divisible by
A/ {l/x}" = [y]. (If Yl E V, Yl =f 0, then {l/x }"[Yl] is an inte-

°
gral divisor, and if Yl is zero at Qi then Qi divides {l/x }"[Yl].)
Therefore, if Yl =f is in the kernel, [YdY] is integral, which
 §A.4 A Fundamental Theorem 145

implies [Yl I YJ = [1 J, that is, Yl is a constant times y. In other

words, the kernel of V ---t Kfv is a I-dimensional subspace gen-
erated by y. The image is therefore a subspace of dimension
nv - g. Elementary linear algebra implies that the Q's can be
rearranged so that evaluation at the first nv - 9 of the Q's de-
fines an onto linear function V ---t K;v- g whose kernel is the 1-
dimensional subspace generated by y. Let this be done, and let
B = Ql Q2 ... Qnv-g, C = AlB = Qnv-g+l Qnv-g+2 ... Qnv.
For each i = 1, 2, ... , f-L, PiC IRis equivalent to an integral
divisor by Abel's theorem (its degree is l+g-l = g), say [zd =
RDd PiC, where Zi E K' and Di is an integral divisor in K' of
degree g. The order of Zi at Pi is -1, since otherwise Pi would
divide Di (R is distinct from Pi) and C would be equivalent to
an integral divisor RDd Pi divisible by R and therefore distinct
from C, call it Ei; this is impossible, because C rv Ei would
imply A = BC rv BEi, [lJ rv AI{llxY rv BEi/{llxY, that
is, V would contain an element zero at all places in B but
nonzero at some place in C, contrary to the choice of B.
The poles of each Zi are concentrated at PI P2 ... P/-t C, the
order of Zi at Pi is -1, and the order of Zi at Pj is 2 0 for
j =1= i. Therefore, since C is a product of distinct places, Z =
ZI + Z2 + ... + z/-t has the desired properties.
LEMMA 3. Let x be a parameter of a function field K with the
property that {x} is a product of distinct places, say {x}
P1 P2 ... Pn . Given an element u of K, let

u = a(i) x- N + ... + a(i) X-I + n· (i = 1,2, ... ,n)

-N -1 '/1

where N is a positive integer independent of i, where the a;i)

are constants of K, and where 7Ji has order at least 0 at Pi.
Then
n n
tr u = (2: a~~ )x- N + ... + (2: a~Dx-l + 7]
i=1 i=1
146 Divisor Theory

where tru is the trace ofu relative to the extension K :::> Ko(x)
(Ko is the field of constants of K) and where TJ has order at
least 0 at the place {x} in Ko(x).
PROOF: (Dedekind-Weber [D-W]) By Theorem 2, there ex-
ists, for each i, an element Yi of K which is integral over Q[x)
and divisible by the restriction to x of {x}N Pi- N but not di-
visible by the restriction to x of Pi. Then ordp(Yi) ~ N for
P = Pj with j t- i, while ordp(Yi) = 0 for P = Pi. A set of
elements YI, Y2, ... , Yn E K obtained in this way is a basis of
Kover Ko (x), as can be shown as follows.
By the theorem of §3.21,

n = degK{x} = [K: Ko(x)) degKo(x) x = [K: Ko(x)).

To prove that YI, Y2, ... , Yn give a basis of Kover Ko(x) it will
suffice, therefore, to prove that they are linearly independent
over Ko(x), or, what is the same, to prove that if "pI(X), "p2(X),
... , "pn(x) E Ko[x) satisfy ~ "pi(X)Yi = 0 then "pi(X) = 0 for all
i. If ~"pi(X)Yi is zero at {X}8 for some s > 0 then "pi(X) must
be zero at {X}8 for all i, since otherwise it would be possible to
reorder the y's and "p's and choose a nonnegative integer t < s
such that each "p(x)/x t was in Ko[x) and the constant term
of "pI(X)/X t was nonzero; this would imply a contradiction,
because ~"pi(X)X-tYi would be zero at {x} (s > t), while
at PI the first term of the sum would be nonzero (neither
"pI ( X )x - t nor YI is zero at PI) and the other n - 1 terms would
all be zero at PI ("pi(X)X- t is finite at PI and Yi is zero there
for i > 1). In particular, if ~"pi(X)Yi = 0 then "pi(X) is zero
at {X}B for all i and all s, which shows "pi(X) = 0 for all i.
Therefore, the equations UYi = ~ VijYj define elements
Vij of Ko(x), and tru = ~~=I Vii. Since xNu and Yi are
finite at all the Pi, ~ x NVijYj is finite at all the Pi. If
qi(X) E Ko[x) is a common denominator of ViI, Vi2, ... , Vin, say
Vij = "pij(X)/qi(X) where "pij(X) E Ko[x), then qi(X)XNUYi =
~j "pij( x )x NYj is divisible by {x} at least as many times as
 §A.4 A Fundamental Theorem 147

qi(X) is, which implies that 1/Jij(X)X N is divisible by {x} at

least as many times as qi(X) is (as was just shown), which
implies VijX N = 1/Jij( x )x N / qi( x) is finite at {x} and therefore
that x NUYi is finite at {x}. On the other hand, x- N Yj is finite
at P k whenever k =1= j, so VijYj = VijX N x- N Yj is finite at P k
whenever k =1= j. When k =1= i, UYi = ux N x- NYi is finite at
P k , so every term in the equation UYi = L VijYj is finite at P k
with the possible exception of the term VikYk. Therefore, this
term is finite at Pk, that is, VikYk is finite at Pk when i =1= k.
Since Yk has order 0 at P k , this shows that Vik is finite at P k
when i =1= k. Since Vik E Ko( x), Vik is finite at {x} when i =1= k.
Thus,
(Vii - U)Yi = - L VijYj
j¢i
is finite at Pi because all terms on the right are finite at Pi (Vij
is finite at {x} and Yj is finite-in fact zero-at Pi). Since Yi is
of order 0 at Pi, it follows that Vii - U is finite at Pi. Therefore,

I'
<:.i = ( Vii - U) + ( U- ( i)
a_Nx - N ( i) x -1 )
- ... - a_I = ( Vii - U) + 'fJi
is finite at Pi. Thus Vii = a~~x-N + ... + a~~ x-I + (i where
(i is in Ko(x) (all other terms of the equation are in Ko(x)).
Therefore, (i is in fact finite at {x}. In conclusion, then, tr U =
LVii = (La~~)x-N + ... + (La~Dx-I +(, where (= I:(i
is finite at {x}, as was to be shown.
PROOF OF THE THEOREM: By Lemma 2, there is an x in a
numerical extension Kif of K' such that {I / x} is a product of
distinct places in Kif including all places in Kif which divide B.
Let U E Kif be such that U dx represents the given differential.
For each place Pi dividing {l/x}, x-I is a local parameter at Pi
and the residue of U dx = u( -x 2 )d(x- I ) at Pi is the coefficient
of x = (x- I )-1 in the expansion of -x 2 u in powers of x-I at
Pi. When constants a;i) in Kif are defined by u = a<;}x N +
148 Divisor Theory

a <;}-1 X N -1 + ... + a~i) + a~~ x -1 + 77i, where 77i has order greater
than 1 at Pi, the residue of u dx at Pi is -a~~. By Lemma 3,
tr( -x 2 u) = (- E a<;})x N +2 + ... + (- E a~~)x + 77, where 77
has order at least 0 at {x -1 }, is the expansion of tr( - x 2 u) in
powers of the local parameter x-I at {x-I}. By Lemma 1,
tru is a polynomial in x. Therefore, tr( -x 2 u) is a polynomial
in x with no terms of degree 0 or 1. By the uniqueness of
expansions in the form bN +2 x N +2 + ... + b1 x + 77, it follows
that 77 = 0, Ei a~~ = 0, and - Ei aji) is the coefficient of xi
in tr u for j ~ o. Since a~~ is the residue of u dx at Pi in K'
as well as in Kit, E a~~ = 0 is the equation that was to be
proved.
§A.5 Holomorphic Differentials.
A differential in a function field K is holomorphic if it has
no poles, that is, if its order at any place in any numerical
extension of K is nonnegative.
The holomorphic differentials in K obviously form a vector
space over the field of constants of K.
THEOREM. The dimension of the holomorphic differentials in
a function field K as a vector space over the field of constants
of K is equal to the genus of K.
The proof, which will be given in §A.I0, will include a con-
struction, following [D-W], of the holomorphic differentials in
K.
§A.6 Integral Bases.
PROPOSITION. Let x be a parameter of a function field K.
There is a basis Yl, Y2, ... , Yn of Kover Ko(x), where Ko
is the field of constants of K, such that an element z of K is
integral over Q[x] (or, what is the same, integral over Ko[x])
if and only if its unique representation in the form z = VIYl +
V2Y2 + ... + VnYn with Vi E Ko(x) has Vi E Ko[x] for all i.
 §A.6 Integral Bases 149

DEFINITION. Such a basis of Kover Ko (x) is an integral basis

with respect to x.
PROOF: The proof is a simple adaptation of the proof of the
analogous fact in §2.6. Because [K: Q(x)] < 00, K = Q(x, y)
for some y E K integral over Q[x]. By the second corollary of
§1.28, there is a.6. E Q[x] such that every element of K integral
over Q[x] is of the form (ao + aly+ a2y2 +... +a m _1ym-l)/.6.,
where the ai are in Q[x] and m = [K: Q(x)]. Each ai can
be written in the form ai = qi.6. + ri, where deg ri < deg.6..
Thus, every element of K integral over Q[x] is of the form
qo + QlY + Q2y2 + ... + Qm_ 1ym-l + 0, where 0 is of the form
O:::::~1 siyi)/.6., with Si E Q[x], deg Si < deg.6., and where
o is integral over Q[x]. As was noted in §3.24, integrality
of L: siyi /.6. imposes homogeneous linear conditions on the
m . deg.6. coefficients of the Si, so there is a finite set 01, 02,
... , Ok of such o's with the property that every such 0 is a
linear combination of these with coefficients in Q. Therefore
1, y, y2, ... , ym-l, 01, 02, ... , Ok span the set of all elements
of K integral over Q [x] as a module over Q [x].
Since, a fortiori, it spans this set over Ko[x],the algorithm
of §2.6 can be used to find a basis of Kover Ko( x) whose
elements span the same Ko[x]-module of all elements of K
integral over Q[x]. Specifically, at step one the spanning set
is rearranged to make the discriminant (note that Ko[x] is a
natural ring) of the first n = [K: Ko(x)] elements be a nonzero
element of Ko[x] of minimum degree. At step two, all follow-
ing elements are reduced by subtracting multiples of the first
n to make the coefficients of their representations in this basis
proper rational functions (with a numerator of smaller degree
than the denominator) or zero. Repetition of these steps re-
duces the degree of the discriminant of the first n elements of
the spanning set, unless all elements that follow the first n are
zero.
150 Divisor Theory

§A.7 Normal Bases.

Let x be a parameter of a function field K. The order at
x = 00 of a nonzero element z of K integral over Q [x] is the
least integer e for which z/x e is finite at {1/x}.
PROPOSITION. Given a parameter x of a function field K,
tbere is an integral basis Yl, Y2, ... , Yn of K witb respect
to x such tbat tbe order at x = 00 of any nonzero element
z of K integral over Q[x] is equal to tbe maximum order of
tbe terms in its expansion z = E CijXiyj, wbere tbe Cij are
nonzero elements of Ko.
DEFINITION: Such an integral basis is a normal basis of K
with respect to x.
PROOF: Let Yl, Y2, ... , Yn be an integral basis of K with
respect to x. Let ei be the order of Yi at x = 00. Since
Yi/x~; is finite at {1/x}, it represents an element of the ring
of values of K at {1/x}. Let V denote the ring of values of
K at {1/x}. If the n elements of V represented by yi/x e; for
i = 1, 2, ... , n are not linearly independent over Ko (V is a
ring containing Ko in a natural way, and is therefore a vector
space over K 0) then there exist constants b1 , b2 , ••• , bn E K 0
not all zero such that E7=1 biYiX-e; is zero at {1/x}. Let j
be an integer, 1 ~ j ~ n, such that bj =f 0 and ej is as large
as possible. Since x E~1 biYiX-e; is finite at {1/x}, the order
of yj = E~1 biYixej -e; is at most ej - 1. Thus, replacement
of Yj by yj gives a new integral basis of K with respect to
x (ej - ei ~ 0 and bj =f 0) in which E ei is reduced by at
least 1. Therefore, a finite number of repetitions of these steps
gives an integral basis Yl, Y2, ... , Yn of K with respect to x
for which the elements of V represented by yi/x e; are linearly
independent over Ko. It will be shown that such an integral
basis is a normal basis.
Let z = E CijXiYj where the sum contains at least one term,
where the i's are ~ 0, and where the Cij E Ko are nonzero. Let
 §A.8 The Dual of a Normal Basis 151

e be the maximum over all terms of this sum of i + e j. It

is to be shown that e is the order of z. Since xiYj/x e is
zero at {l/x} when i + ej < e, the element of V repre-
sented by z/x e is the same as the element of V represented by
ECijXiYj/xe = ECijYj/Xej, where the sum is over all terms
(there is at most one term for each j = 1, 2, ... , n) in which
i + e j = e. Thus, since this is a linear combination of linearly
independent elements in which the coefficients are not all zero,
the element of V represented by z / x e is not zero. Therefore,
e is the order of z at x = 00, as was to be shown.

COROLLARY. Tbe genus of K is 1 + E( ei - 1).

DEDUCTION: For v sufficiently large, the dimension of the sub-

space of Kover Ko containing elements divisible by {l/x} -v is
nv-g+1, where n = [K: Ko(x)] and 9 is the genus of K (§3.26).
But an element z of K is divisible by {I / x } -v if and only if it
is integral over Q[x] and of order at most v at x = 00. Thus,
z is in this subspace if and only if its representation relative to
the normal basis Yl, Y2, ... , Yn is of th~ form z = E <pj( x )Yj,
where <pj(x) E Ko[x] is of degree v - ei at most. Therefore,
nv - 9 + 1 = E7=1 (v - ei + 1), which gives the desired formula
for g.

§A.8. The Dual of a Normal Basis.

Let x be a parameter of a function field K. The trace relative
to the field extension K :J Ko(x) (where Ko is the field of
constants of K) is a linear form K -+ Ko(x). Therefore, the
trace of the product is a symmetric bilinear form K x K -+
Ko(x). This bilinear form is nondegenerate, because if f E K
were nonzero and satisfied trK/Ko(x) fg = 0 for all 9 E K, then
trK/Ko(x) 1 would be 0 (set 9 = f- 1 ), contrary to the fact that
trK/Ko(x) 1 = [K: Ko(x)]. Given any basis Yl, Y2, ... , Yn of K
over Ko(x), one can therefore define a dual basis ZI, Z2, ... ,
152 Divisor Theory

Zn of Kover Ko(x) by the equations

if i = j
trK/Ko(x) YiZj = {~ otherwise.
PROPOSITION. Let Yl, Y2, ... , Yn be a nonnal basis of K with
respect to x, and let ei be the order of Yi at x = 00 for i = 1,
2, ... , n. Then Vi = yi/x e ; for i = 1, 2, ... , n defines a
nonnal basis of K with respect to l/x, and the order of Vi at
(l/x) = 00 is ei.
PROOF: The poles of Vi are concentrated at {x}{l/x} (the
poles of Yi are concentrated at {1/x} and the poles of (l/xy;
are concentrated at {x}). By the definition of ei, Vi has no
poles at {1/x}. Therefore, the poles of Vi are concentrated at
{ x }, that is, Vi is integral over Q[1 / x].
Let w E K be nonzero and integral over Q[I/x] and let
e' be its order at (l/x) = 00. Then (as was just shown in
the case of Yi) w/(I/x)e' = x e' w is integral over Q[x]. Thus
xe'w = '2:cijXiYj = '2: cij(l/x)-i-e j Vj. Let e be the order
of xe'w at x = 00. Then e $ e', so the exponents of l/x in
the expansion w = '2:cij(l/x)e'-i-ejVj are all nonnegative.
Therefore, VI, V2, ... , Vn is an integral basis of K with respect
to l/x.
Since vj/(I/xYi = Yj is finite at {x}, the order of Vj at
(l/x) = 00 is at most ej. Since x-1Yj is not finite at {x} (if
it were, x-1Yj would be integral over Q[x], contrary to the
assumption that Yl, Y2, ... , Yn is an integral basis of K with
respect to x), ej is the order of Vj at (l/x) = 00. Therefore,
the order of cij(l/xy'-i-ei Vj at (l/x) = 00 is e' - i. If i
were positive in all tenns of x e' w = '2:cijXiYj, then (the Yi are
an integral basis) xe'-lw would be finite at {x}, contrary to
the definition of e'. Therefore i = 0 in at least one term, so
w = '2:cij(l/x)e'-i-eiVj contains at least one term of order
e' at (l/x) = 00, which shows that VI, V2, ... , Vn is a normal
basis of K with respect to l/x.
 §A.9 The Dual of a Normal Basis 153

§A.9.
PROPOSITION 1. Let Yl, Y2, ... , Yn be an integral basis of a
function field K witb respect to a parameter x of K, and let
Zl, Z2, ... , Zn be its dual basis. Let Dx be tbe different OfYl,
Y2, ... , Yn witb respect to K as an extension of tbe natural
ring Ko[x]. Tben Dx[Zi]x is integral for all i = 1, 2, ... , n.
(Here [Zi]x is the restriction to x of the global divisor [Zi].)
PROOF: In the notation of §1.28, Dx = [F'( a)], where F'( a)
is as defined in §1.26 with r = Ko[x] and (}i = Yi (i = 1, 2,
... , n). Since Zi has the property that trK/Ko(x)YZi E Ko[x]
whenever Y is integral over Ko[x], trK/Ko(x) (Ziq(X)) = <p(X)
has coefficients in Ko[x], where the notation of §1.28 is modi-
fied by setting X in place of x so that q(X) = F(X)/(X - a).
As in §1.28, <p(ii) = ZiF'(ii). Thus, [ZiF'(ii)]x is integral over
Ko [x], as was to be shown. '
PROPOSITION 2. (Cf. §2.8.) Let Dx be as above, let P be a
place in K at wbich x is finite, and let a be tbe value of x at
P. Tben tbe numerator of [x - a]x/Dx is divisible by P x . In
fact, if e is tbe multiplicity witb wbicb P x divides [x -a]x, tben
e - 1 is tbe multiplicity witb wbicb P x divides Dx.
PROOF: Assume, without loss of generality, that a = 0 and
that {x} is a product of places, say {x} = pr p;2 ... pr. Let
V be the ring of values of K at [x]x, and let '- be the natural
map of elements of K finite at [x]x to V. Then Ko is included
in V in a natural way and '-(yd, '-(Y2), ... , t(Yn) are linearly
independent over Ko because alt(Yl)+a2'-(Y2)+·· ·+ant(Yn) =
o for ai E Ko implies (alYl + a2Y2 + ... + anYn)/x is integral
over Q[x], which in turn implies, because the y's are an integral
basis with respect to x, that the ai are aliO. Therefore, as in
Lemma 1 of §2.7, the t(Yi) are a basis of V over K o, K is p-
regular for p = x, and, by Proposition 1 of §1.32, the image in
V[X, Ul, U2, ... , un] of F(X) = NK/Ko(x)(X -UIYI-U2Y2-
... - unYn) is rr7=1 Ni(X - UIYl - ... - unYnYi where Ni is
154 Divisor Theory

the norm with respect to the field extension V(Pi)", :::) Ko of

X - UIYI - ... - UnYn when the Yi are regarded as representing
elements of V(Pi)." (The natural ring Ko[x] mod x is the field
K o, so kp = Ko in this case.) Because Pi is a place, every
element of K finite at {x} has a value in Ko at Pi, so V(P;)", =
Ko. Thus, the image of F(X), which has coefficients in Ko(x)
so that its image in V[X, UI, U2, ... , un] has coefficients in
K 0, IS . Iy Ilki=I (X - (}(i)
. SImp (}(i) (}(i) )e· h
1 UI - 2 U2 - ... - k Uk " were
(})i) E Ko is the value of Yj at Pi.

It follows that F'(X) = E~=I ei(X - (}~i)UI - ... -

(}~i)Un)-I F(X) mod x. Dx is the divisor in K as an extension
of Ko[x] represented by the polynomial obtained by setting
YI UI + Y2U2 + ... + YnUn in place of X in F'(X). Since p;l
divides [x]x and since all terms of F'(YI UI + ... + Ynun) mod x
but the first are obviously divisible by p;l, the statement to
be proved is that the divisor represented by
(1)
el ( (YI - (}I(1) )UI + (Y21- »
(}2 U2 + ... + (Yn - (}n )u n,
) (1) )e1-I .
II((YI - (}I(i) )UI + ... + (Yn - (),;(.) )U n)e.•
i>I
is not divisible by (PI)~l. (It is obviously divisible by
(PI)~l-I.)
Since Pi is a place, (Pi)x = [x, z]x for some z E K integral
over Ko[x]. Since Pi # PI, z is not zero at Pl. Since the y's
are an integral basis, z = E <pj(x)Yj and the value at Pi of
z is E <pj(O)(}ji) = O. If (PI)x divided [E/Yj - (}ji»)Ui]X, the
value of z at PI would be E <pj(O)(})i) = O. Thus, (PI)x does
not divide the factors of (1) other than [E(Yj - (})I»)Uj]~l-i.
If el = 1, the proposition follows. If el > 1, it remains to show
that (PI); does not divide [E(Yj - (})I»)Uj]x.
If (PI); did divide this divisor, then each Yj would be of the
form (})I) +vj, where ordp1(vj) ~ 2. Since ordp1(x) = el ~ 2
 §A.10 Construction of Holomorphic Differentials 155

by assumption, any element of K integral over Ko[x] would

be of the form Z = ~<Pj(x)Yj = f3 + w, where f3 E Ko and
ord p1 (w) 2 2. But then (PI); would divide [x,z]x whenever
Z was integral over Ko[x] and zero at PI, contrary to the fact
that (PI) x can be expressed in the form [x, z] x.
§A.I0. Construction of Holomorphic Differentials.
THEOREM. Let YI, Y2, ... , Yn be a normal basis of K with
respect to x and let ZI, Z2, ... , Zn be its dual basis. A differ-
ential w in K is holomoIphic if and only if the element u E K
for which w = u dx is of the form u = 'l/JIzI +'l/J2Z2 + ... +'l/Jnzn,
where the 'l/Ji are in K 0 [x] and their degrees satisfy deg 'l/Ji <
ei - 2, where ei is the order of Yi at x = 00.
PROOF: If <PI, <P2, ... , <Pn E Ko[x], then
n n n
L <Pi'l/Ji = L L <Pi'l/Jj tr(YiZj) = tr(fu),
i=::1 i=::1 j=::1

where f = ~~=::I <PiYi. If u dx is holomorphic, then, since f

is integral over Q[x], fu dx has poles concentrated at {l/x},
and ~ <Pi'l/Ji E Ko[x] by Lemma 1 of §AA. Therefore, since the
<Pi E Ko[x] are arbitrary, the 'l/Ji must all be in Ko[x].
Let Vi = yi/x e ; and Wi = x e ; Zi for i = 1,2, ... ,n. Then Vi is
a normal basis of K with respect to l/x and Wi is its dual basis.
If w = udx = -x 2ud(1/x) is holomorphic, then fx 2ud(1/x)
has poles only at {x} whenever f = ~ OiVi with Oi E K o[l/x].
Therefore, tr(fx 2u) = ~Oix2'l/Jix-e; is in K o[l/x], which is to
say that, for each i, 'l/Ji(X)/x e;-2 is a polynomial in x-I, or,
what is the same, that the degree of 'l/Jj is at most ej - 2.
Thus, a holomorphic differential is necessarily of the form
(~'l/JiZi)dx, where 'l/Ji E Ko[x] and deg'l/Ji:S ej - 2. It remains
to show that all these differentials are indeed holomorphic, that
is, that xi zjdx is holomorphic whenever 0 :S i :S ej - 2.
At any place P in K or a numerical extension of K at which
x is finite, say x is a at P, ordp(zjdx) = ordp(zj)+ordp(dx) =
156 Divisor Theory

ordp(zj) + ordp(x - a) - 1. Let e = ordp(x - a). Now [x -

a]x[zj]x = (Dx[zj]x)([x - a]x/Dx). The first factor is integral
by Proposition 1 of §A.9 and the second factor has numerator
divisible by Px by Proposition 2. Therefore ord p (( x - a)z j) ~
1 and ordp(zjdx) ~ O. Thus, for any i ~ 0 and for any j = 1,
2, ... , n, xizjdx has poles concentrated at {1/x}.
When x is replaced by l/x and Yj by Vj = Yj/x ej , which
replaces Zj by Wj = x ej Zj and xizjdx by (l/x)i wjd(l/x) =
-ZjX ej -2- i dx, it follows that the poles of -ZjX ej -2- i dx are
concentrated at {x} for ej - 2 - i ~ 0 and j = 1, 2, ... , n.
Thus, if 0 :::; i :::; ej - 2, the poles of xizjdx are concentrated
both at {I / x} and at {x}, so xi Zj dx is holomorphic, as was to
be shown.
COROLLARY. (Theorem of §A.5) The genus of K is the dimen-
sion of the vector space of holomorphic differentials in K.
DEDUCTION: IT ej > 0, the vector space of polynomials of
degree i, 0 :::; i :::; ej - 2, has dimension ej -1. IT ej = 0, it has
dimension 0 = (e j - 1) + 1. A normal basis contains exactly
one constant element, so exactly one ej is 0, and the dimension
of the space of holomorphic differentials is E(ej - 1) + 1 = 9
(§A.7).
§A.l1. Examples.
IT K is the function field of a hyperelliptic curve y2 = f(x),
where f E Q[x] is of degree 2d with distinct roots, then 1,
Y is a normal basis. Since el = 0 and e2 = d, the genus is
(( -1 ) + (d - 1») + 1 = d - 1. Since the 2 x 2 symmetric matrix
which represents the trace of the product relative to this basis
is the diagonal matrix whose entries on the diagonal are 2 and
2f(x), the dual basis is !, 2f~x) = 21y, so the holomorphic
differentials are those of the form g(x)dx/y, where g(x) is a
polynomial of degree:::; d - 2.
IT K is the function field of the Klein curve y3 +x 3 y +X = 0,
 §A.ll Examples 157

then, as is shown in §3.28, 9 = 3. The holomorphic differentials

in ]{ can be found as follows.
At any place P = {x - a, y - b} in ]{ or a numerical extension
of ]{ at which x, and therefore y, is finite, either x - a or y - b
is a local parameter. (Differentiation of y3 + x 3y + X = 0
gives (3y2 + x 3 )dy + (3x 2y + 1 )dx = 0, so x - a is a local
parameter at P if and only if Fy = 3y2 + x 3 is nonzero at P,
and y - b is a local parameter at P if and only if Fx = 3x 2 y + 1
is nonzero at P. If Fx and Fy were both zero at P, then
multiplication of Fy = 0 by 3x 4 would give (3x 2y? + 3x 4X 3 =
0, and therefore, since 3x 2y = -1 at P, 1 + 3x 7 = 0 at P.
But multiplication of y3 + x 3y + X = 0 by 27x 6 would give
(3x2y)3+9x4x3(3x2y)+27x6x = 0, (3x 2y)3+9x 7 (3x 2y+3) = 0,
and therefore -1 + 9x 7 . 2 = 0 at P, which would contradict
1 + 3x 7 = 0 at P.)
Let u dx be a holomorphic differential in ]{. If x - a is a
local parameter at P, then u is finite at P. If x - a is not a
local parameter at P, then y - b is a local parameter at P.
Since u dx = -uFydy j F x :, it follows that uFy j Fx is finite at
P. Since Fx is finite at P, uFy is finite at P. Since Fy is finite
at any place where x is finite, uFy is finite at any place P in
]{ or a numerical extension of ]{ where x is finite. Therefore,
uFy is integral over Q[x].
Conversely, if uFy = v is integral over Q [x], then, at any
place P where x is finite, either x - a is a local parameter, in
which case Fy i=- 0 and u = v j Fy is finite at P so u dx has no
pole at P, or y - b is a local parameter at P, in which case
Fx i=- 0 and u dx = -uFydyj Fx has no pole at P. Thus, the
determination of the holomorphic differentials in ]{ reduces
to the determination of those elements v of ]{ integral over
Q[x] for which (vjFy)dx has no poles at places where x = 00.
(The same is true of the determination of the holomorphic
differentials in the function field ]{ of any curve F( x, y) = 0
monic in y for which Fx and Fy do not vanish simultaneously
for finite x.)
158 Divisor Theory

The two places where x = 00 are Q = {l/y,x/y} and R =

{y/x, l/x}. At Q, s = x/y is a local parameter, t = l/y has a
double zero, and s3 + t 3s + t = 0 holds. Then t = _s3 - st 3 =
_s3 + s10 + ... , x = sit = _S-2 + ... , dx = (2s- 3 + ... )ds,
Fy = 3y2 + x 3 = 3r 2 + (s/t? = t- 3(3t + s3) = (_s3 +
... )-3( -2s 3 + ... ) = 2s-6 + ... , so vdx/Fy = v(s3 + ... )ds,
and v dx / Fy = U dx has a pole at Q if and only if v has a pole at
Q of order greater than 3. At R, z = l/x is a local parameter,
w = y/x has a double zero, and z3 + w 3z + w = o. Then
dx = -z- 2dz and Fy = 3(w/z? + (1/z)3 = z-3(1 + 3zw 2 ) =
z-3 + ... , so v dx / Fy = v( - z + ... )dz has a pole at R if and
only if v has a pole at R of order greater than l.
eo
As was seen in §3.28, v = + 6y + 6y2 where ei E Q[x].
It is easy to check that when v = a + bx + cy for constants a,
b, and c, the pole of v at R has order ~ 1 and its pole at Q
has order ~ 3. Since 9 = 3, it follows that the most general
holomorphic differential in K is

a + bx + cy dx
3y2 + x 3

where a, b, and c are constants.

§A.12. The Riemann-Roch Theorem for Integral
Divisors.
As in §A.1, let PI, P2, ... , Pk be places in a function field K,
let Ui be a local parameter at Pi for i = 1, 2, ... , k, let z E K
have poles concentrated at PI P2 ... Pk , and let (<PI, <P2, ... ,
<Pk) be the principal parts of z relative to the parameters Ui,
that is, let <Pi(X) be the unique polynomial with coefficients in
Ko (the field of constants of K) and with zero constant term
for which z - <Pi( ui 1 ) is finite at Pi (i = 1, 2, ... , k).
If WI, W2, ..• , W 9 are a basis of the vector space (over Ko)
of holomorphic differentials in K, the statement that the sum
of the residues of ZWi is zero gives 9 linear conditions satisfied
 §A.12 The lliemann-Roch Theorem for Integral Divisors 159

by the principal parts (<PI, <P2, ... , <Pk) of z. Specifically, let

N be an integer large enough that N ~ deg <Pi for all i, let
polynomials ~}j) of degree N -1 at most with coefficients in K 0
be defined by the conditions Wj = (dj)(Ui) + o(Uf))dui, and
d
let (<Pi, j )) denote the coefficient of X-I in <Pi(X- l )~}j)(X).
Then (<Pi, ~rj») is the residue of ZWi at Pi and the principal
parts (<PI, <P2, ... , <P k) of Z satisfy the g homogeneous linear
conditions 2:7=1(<Pi,dj)) = 0 (j = 1, 2, ... , g). The main
theorem is that these necessary conditions are sufficient. That
is, given a k-tuple (<PI, <P2, ... , <Pk) of polynomials which satisfy
2:7=1 (<Pi, dj») = 0 (j = 1, 2, ... , g) when the polynomials
d j ) are defined as above, relative to some N greater than the

degrees of all the <Pi, there is an element Z E K with poles

concentrated at Pl P2 ... Pk whose principal parts, relative to
the parameters Ui, are «/>1, <P2, ... , <Pk).
PROOF: Increasing N does not alter the conditions satisfied
by (<PI, <P2, ... , <Pk). Therefore, one can assume without
loss of generality that, with A = Pl P2 ··· Pk, dimK(A- N ) +
degK(A -N) + g = 1 (§3.26). Thus, dimK(A -N) = kN - g + l.
Since the principal parts of an element of K are zero if and only
if the element is constant, the principal parts of elements of K
divisible by A -N are a vector space of dimension kN - g over
Ko. The principal parts of such a function are described by a
k-tuple (<PI, <P2, ... , <Pk) of polynomials of degree N, at most,
with coefficients in Ko and with constant term zero. Since
the space of all such k-tuples is a vector space of dimension
kN over K o, the principal parts of elements of K divisible by
A - N are a subspace of codimension g. Elements of the sub-
space satisfy the g conditions 2:7=1 (<Pi, d j )) = 0 (j = 1, 2,
... , g). These conditions are independent for N sufficiently
large, because, for any fixed i, if alWl + a2W2 + ... + agw g =J 0
then ale?) + a2~?) + ... + agd g) =J 0 for N large. These
conditions therefore define a subspace of co dimension g which
160 Divisor Theory

contains the principal parts of elements of K divisible by A-N

and therefore, because it has the same dimension, coincides
with the principal parts. Therefore, the necessary conditions
are sufficient, as was to be shown.
COROLLARY. (Riemann-Roch Theorem for Integral Divisors)
Let A be an integral divisor in a function field K. Then
dimK(A- 1 )+degK(A-l)+g = l+p where p is the dimension
of the space of holomorphic differentials zero at A. More pre-
cisely, p is the dimension of the vector space of differentials w
which have the property that, for any place P in any numerical
extension of K, ordp(w) is at least as large as the number of
times P divides A.
DEDUCTION: Taking a numerical extension of K does not
change dimK(A-l), degK(A-l), g, or p (see §3.23). There-
fore, one can assume without loss of generality that A is a
product of powers of places, say A = p 1e l p;2 ... P:/e. Let N
be large enough that it is greater than all the ei and that the
theorem holds for N. By the theorem, a k-tuple of polynomi-
als (ifJl, ifJ2, ... , ifJ k) with deg ifJi ~ ei gives the principal parts
of an element of K with poles concentrated at PI P2 ••• Pk if
d
and only if it satisfies the 9 conditions L~=1 (ifJi, j )} = 0 for
j = 1, 2, ... , g. The k-tuples satisfying these 9 conditions
are a vector space over Ko of dimension Lei - tP where tP is
the number of independent conditions imposed by the 9 con-
d
ditions L~=1 (ifJi, j )} = o. Thus tP is 9 minus the dimension
of the space of holomorphic differentials for which the con-
d
ditions L~=1 (ifJi, j )} = 0 impose no conditions on k-tuples
( ifJl' ifJ2, ... , ifJ k) satisfying deg ifJi ~ ei. A holomorphic dif-
ferential w is in this space if and only if in the correspond-
ing (6, 6, ... , ~k) no ~i contains a term of degree ~ ei.
Therefore, tP = 9 - p, where p is as in the statement of the
corollary. Thus, the k-tuples which are principal parts of ele-
ments of K divisible by A-I are a vector space of dimension
E ei - 9 + p = degK A - 9 + p. Two elements of K have the
 §A.13 Riemann-Roch for Reciprocals of Integral Divisors 161

same principal parts if and only if their difference is a con-

stant, so dimK(A -1) is one greater than the dimension of the
space of principal parts of elements divisible by A-I, that is,
dimK( A-I) = degK A - 9 +1 +11, which is the desired formula.
§A.13 The Riemann-Roch Theorem for Reciprocals of
Integral Divisors.
PROPOSITION. Let B be an integral divisor in a function field
K with degK B > O. The differentials w in K divisible by
B- 1 (that is, ordp(w) ~ -e for any place P in any numerical
extension of K which divides B exactly e times) are a vector
space of dimension degK B + 9 - lover the field of constants
Ko of K.
PROOF: Let x be a parameter of K such that B I{x} (see
§3.10), let Yl, Y2, ... , Yn be an integral basis with respect to
x, and let Zl, Z2, ••• , Zn be its dual basis. (Here n = deg{ x} =
[K: Ko(x)].) It was shown in §A.10 that the poles of the differ-
ential xizjdx are concentrated at {x} when i ::; ej - 2 (where
ej is the order of Yj at x = 00). It is holomorphic when, in
addition, i ~ O. It is divisible by {x} -1 when xi+l zjdx has no
pole at {x}, which is true when i + 1 ~ O. Thus, the dimension
of the vector space of differentials divisible by {x} -1 is equal
to the number of indices i for which -1 ::; i ::; ej - 2 (j = 1,
2, ... , n). Therefore, the dimension is 9 + n - 1, because 9
is the number of solutions of 0 ::; i ::; ej - 2 and there is one
additional solution i = -1 for each j = 1, 2, ... , n except for
the one j for which ej = O.
Passage to a numerical extension of K does not change the
dimension of the space of differentials divisible by B- 1 or the
space of differentials divisible by {x} -1. Assume without loss
of generality that {x}, and therefore B, is a product of places,
say {x} = PI P2 •• 'Pn and B = PI P2 •• 'Pv , where 1::; v::; n.
To say that a differential w in K which is divisible by {x} -1 is
also divisible by B- 1 imposes n - v conditions on w, namely,
the conditions that n - v of the coefficients of the expansions
162 Divisor Theory

of W in terms of local parameters at the Pi are zero. (IT the

Pi are distinct, the conditions are simply that the residues of
W at Pi are zero for i > v.) These conditions are independent
because the differentials divisible by {x} -1 have dimension 9 +
n - 1 and the differentials divisible by P1- 1 have dimension 9
(a differential divisible by P1- I has no poles except, possibly,
for a simple pole at PI, but, because the sum of the residues
is zero, it can also have no pole at PI and must therefore be
holomorphic) so the n -1 conditions imposed by divisibility by
Pl- l must be independent. Therefore, the differentials divisible
by B-1 = (PI P2 ... PlI )-1 have dimension g+n-l-(n-v) =
9 + v -1, as was to be shown.

COROLLARY. The Riemann-Roch formula

dimK A-I + degK A-I + 9 = 1 + 1-',

where I-' is the dimension of the space of differentials divisible
by A, holds when A is the reciprocal of an integral divisor.
DEDUCTION: If A = [1], the formula was proved in §A.12.
Otherwise, A = B-1 where B is as in the Proposition. Then
dimK A-I = dimK B = 0 because a nonzero element z of K
divisible by B would satisfy degK[z] ;::: 1. Thus dimK A-I +
degKA- l + 9 = 0 + v + 9 = 1 + 1-', where v = degKB and I-'
is the dimension 9 + v-I of the space of differentials divisible
by A = B- 1 •
§A.14 General Case of the Riemann-Roch Theorem.
Again, let PI, P2 , ••• , Pk be places in a function field K
and let Ui be a local parameter at Pi for each i. Let B be
an integral divisor with degK B = v > 0 which is relatively
prime to PI P2 ••• Pk, and let WI, W2, ••• , W g+1I -1 be a basis
of the vector space of differentials in K divisible by B- 1 • The
principal parts (4)1, 4>2, ... , 4>k) of any element z of K which
has poles concentrated at PI P2 ••• Pk and which is zero at B
 §A.14 General Case of the Riemann-Roch Theorem 163

satisfy l:~1 (4)i, e~j») = 0 for j = 1, 2, ... , 9 + v-I, where

the e;j) are defined by Wj = (e~j)(Ui) + o(Uf))dui for some
N larger than the degrees of the 4>i, and where (4)i, e~j)) is
the coefficient of X-I in 4>i(X- 1 )e~j)(X). These necessary
conditions on (4)1,4>2, ... , 4>k) are also sufficient, that is,
given a k-tuple of polynomials (4)1, 4>2, ... , 4>k) (coefficients in
K o, zero constant terms) satisfying the 9 + v-I conditions
l:(4)i, e~j») = 0, there is an element z of K zero at B and with
poles concentrated at PI P2 ••• Pk whose principal parts are the
4>i.
PROOF: Let A = P1 P2 ··· Pk • For N sufficiently large,
dimK(A- N B) + degK(A- N B) + 9 = 1 (§3.26). That is,
dimK(A-N B) = Nk-v-g+l. Since an element of K divisible
by A - N B is determined by its principal parts (the difference
of two with the same principal parts has no poles, therefore is
constant, therefore is zero because it is zero at B) it follows
that the set of principal parts has codimension v + 9 -1 in the
Nk-dimensional space of k-tuples (4)1, 4>2, ... , 4>k) in which
deg 4>i ~ N. Since, for N sufficiently large, the v + 9 - 1 neces-
sary conditions l:( 4>i, e~j)) = 0 are independent, it follows as
in §A.12 that they suffice to determine the k-tuples that occur
as principal parts.
COROLLARY. (Riemann-Roch Theorem) For any divisor C in
afunctionfieldK, dimK(C- 1 )+degK(C- 1 )+g = 1+11, where
11 is the dimension of the space of differentials in K divisible
by C (as a vector space over the field of constants K 0 of K).
DEDUCTION: One can assume without loss of generality that
C is a product of powers of places (§3.23), say C = AlB
where A = Pt 1p;2 ... p;1c and B = p:~r ...
pte! for distinct
places Pi and for positive integers ei. Also, because the other
cases have already been proved, one can assume degK A > 0
and degK B > O. Then dimK(C- 1 ) is the dimension of the
space of all k-tuples (4)1, 4>2, ... , 4>k) of principal parts of
164 Divisor Theory

elements z of K which are zero at B, have poles concentrated

at PI P2 ••• Pk , and have principal parts satisfying deg ¢>i < ei.
This dimension is el + e2 + ... + ek - t/J = degK A - t/J, where
t/J is the number of independent conditions imposed by the
degK B + 9 - 1 conditions E(¢>i, ~}j») = 0 (j = 1, 2, ... ,
degK B + 9 -1). Since t/J = degK B + 9 -1-11 where 11 is the
dimension of the subspace of the differentials in K divisible
by B- 1 which are zero at A-that is, where 11 is as in the
Corollary-it follows that dimK(C- 1 ) = degK A - degK B -
9 + 1 + 11 = degK C - 9 + 1 + 11, as was to be shown.
 References

[A] Abel, Niels Henrik, Memoire sur une propriete generale

d 'une classe tres-etendue de fonctions transcendantes,
Presente it l' Academie des sciences it Paris Ie 30 Octobre
1826, Memoires presentes par divers savants, vol. VII, Paris,
1841. (Oeuvres, vol. 1, 145-211.)
[D1] Dedekind, Richard, Ueber einen arithmetischen Satz von
Gauss, Mitt. Deut. Math. Gesell. Prague, 1892 1-11.
( Werke, vol. 2, 28-38.)
[D2] Dedekind, Richard, Ueber die Begrundung der Idealtheorie,
Nachr. Kon. Ges. Wiss. Gottingen, 1895, 106-113. (Werke,
vol. 2, 50-58.)
[D-W] Dedekind, Richard and Weber, Heinrich, Theone der alge-
braischen Funktionen einer Veriinderlichen, Jour. fUr Math.,
92 (1882) 181-290. (Dedekind's Werke, 1, 238-349.)
[E1] Edwards, Harold M., The Genesis of Ideal Theory, Arch.
Hist. Ex. Sci., 23 (1980) 321-378.
[E2] Edwards, Harold M., Dedekind's Invention of Ideals, Bull.
Lond. Math. Soc., 15 (1983) 8-17. (Studies in the History of
Mathematics, Esther R. Phillips, ed., MAA, 1987, 8-20.)
[E3] Edwards, Harold M., Galois Theory. Springer-Verlag, New
York Berlin Heidelberg Tokyo, 1984.
E-N-P] Edwards, Harold; Neumann, Olaf; and Purkert, Walter, De-
dekinds "Bunte Bemerkungen" zu Kroneckers "Grundzuge",
Arch. Hist. Ex. Sci., 27 (1982) 49-85.
[He] Hensel, Kurt, Untersuchung der Fundamentalgleichung
einer Gattung fur eine reelle Primzahl als Modul und Bes-
timmung der Theiler ihrer Discriminante, Jour. fiir Math.
113 (1894) 61-83.
[Hu1] Hurwitz, Adolph, Ueber die Theorie der Ideale, Nachr. kon.
Ges. Wiss. Gottingen, 1894, 291-298. (Werke, vol. 2, 191-
197.)
[Hu2] Hurwitz, Adolph, Ueber einen Fundamentalsatz der arith-
metischen Theorie der algebraischen Groflen, Nachr. kon
 166 Divisor Theory

Ges. Wiss. Gottingen, 1895, 230-240. (Werke, vol. 2, 198-

207.)
[Ko] Konig, Julius, Einleitung in die allgemeine Theorie der al-
gebraischen Grossen. Teubner, Leipzig, 1903.
[Kr1] Kronecker, Leopold, Grundzuge einer arithmetischen The-
orie der algebraischen Gross en, Jour. fiir Math. 92 (1882)
1-122. (Also published separately by Reimer, Berlin, 1882.
Also Werke, vol. 2, 237-387.)
[Kr2] Kronecker, Leopold, Zur Theorie der Formen hoherer
Stu/en, Monatsber. Akad. Wiss. Berlin, 1883, 957-960.
(Also Werke, vol. 2,419-424.)
[Ku] Kummer, Ernst Eduard, Zur Theorie der Complexen
Zahlen, Monatsber. Akad Wiss. Berlin, 1846, 87-96. (Also,
Jour. fiir Math., 35 (1847) 319-326 and Collected Papers,
vol. 1, 203-210).
[Wa] Walker, Robert J., Algebraic Curves, Princeton University
Press, Princeton, 1950. (Reprint, Dover, New York, 1962.)
[Wb] Weber, Heinrich, Lehrbuch der Algebra, Vieweg, Braun-
schweig, 1895, 1896; reprint Chelsea, New York, 1962, 1979.
[We] Weyl, Hermann, Algebraic Theory of Numbers. Princeton
University Press, Princeton, 1940, 1951.