Graduate Texts in Mathematics 254

Editorial Board
S. Axler K.A. Ribet
Graduate Texts in Mathematics
1 TAKEUTI/ZARING. Introduction to 34 SPITZER. Principles of Random Walk.
Axiomatic Set Theory. 2nd ed. 2nd ed.
2 OXTOBY. Measure and Category. 2nd ed. 35 ALEXANDER/WERMER. Several Complex
3 SCHAEFER. Topological Vector Spaces. Variables and Banach Algebras. 3rd ed.
2nd ed. 36 KELLEY/NAMIOKA et al. Linear
4 HILTON/STAMMBACH. A Course in Topological Spaces.
Homological Algebra. 2nd ed. 37 MONK. Mathematical Logic.
5 MAC LANE. Categories for the Working 38 GRAUERT/FRITZSCHE. Several Complex
Mathematician. 2nd ed. Variables.
6 HUGHES/PIPER. Projective Planes. 39 ARVESON. An Invitation to C*-Algebras.
7 J.-P. SERRE. A Course in Arithmetic. 40 KEMENY/SNELL/KNAPP. Denumerable
8 TAKEUTI/ZARING. Axiomatic Set Theory. Markov Chains. 2nd ed.
9 HUMPHREYS. Introduction to Lie Algebras 41 APOSTOL. Modular Functions and
and Representation Theory. Dirichlet Series in Number Theory.
10 COHEN. A Course in Simple Homotopy 2nd ed.
Theory. 42 J.-P. SERRE. Linear Representations of
11 CONWAY. Functions of One Complex Finite Groups.
Variable I. 2nd ed. 43 GILLMAN/JERISON. Rings of Continuous
12 BEALS. Advanced Mathematical Analysis. Functions.
13 ANDERSON/FULLER. Rings and Categories of 44 KENDIG. Elementary Algebraic Geometry.
Modules. 2nd ed. 45 LOÈVE. Probability Theory I. 4th ed.
14 GOLUBITSKY/GUILLEMIN. Stable Mappings 46 LOÈVE. Probability Theory II. 4th ed.
and Their Singularities. 47 MOISE. Geometric Topology in
15 BERBERIAN. Lectures in Functional Analysis Dimensions 2 and 3.
and Operator Theory. 48 SACHS/WU. General Relativity for
16 WINTER. The Structure of Fields. Mathematicians.
17 ROSENBLATT. Random Processes. 2nd ed. 49 GRUENBERG/WEIR. Linear Geometry.
18 HALMOS. Measure Theory. 2nd ed.
19 HALMOS. A Hilbert Space Problem Book. 50 EDWARDS. Fermat’s Last Theorem.
2nd ed. 51 KLINGENBERG. A Course in Differential
20 HUSEMOLLER. Fibre Bundles. 3rd ed. Geometry.
21 HUMPHREYS. Linear Algebraic Groups. 52 HARTSHORNE. Algebraic Geometry.
22 BARNES/MACK. An Algebraic Introduction 53 MANIN. A Course in Mathematical Logic.
to Mathematical Logic. 54 GRAVER/WATKINS. Combinatorics with
23 GREUB. Linear Algebra. 4th ed. Emphasis on the Theory of Graphs.
24 HOLMES. Geometric Functional Analysis 55 BROWN/PEARCY. Introduction to Operator
and Its Applications. Theory I: Elements of Functional Analysis.
25 HEWITT/STROMBERG. Real and Abstract 56 MASSEY. Algebraic Topology: An
Analysis. Introduction.
26 MANES. Algebraic Theories. 57 CROWELL/FOX. Introduction to Knot Theory.
27 KELLEY. General Topology. 58 KOBLITZ. p-adic Numbers, p-adic Analysis,
28 ZARISKI/SAMUEL. Commutative Algebra. and Zeta-Functions. 2nd ed.
Vol.I. 59 LANG. Cyclotomic Fields.
29 ZARISKI/SAMUEL. Commutative Algebra. 60 ARNOLD. Mathematical Methods in
Vol.II. Classical Mechanics. 2nd ed.
30 JACOBSON. Lectures in Abstract Algebra I. 61 WHITEHEAD. Elements of Homotopy
Basic Concepts. Theory.
31 JACOBSON. Lectures in Abstract Algebra II. 62 KARGAPOLOV/MERLZJAKOV. Fundamentals
Linear Algebra. of the Theory of Groups.
32 JACOBSON. Lectures in Abstract Algebra III. 63 BOLLOBAS. Graph Theory.
Theory of Fields and Galois Theory. 64 EDWARDS. Fourier Series. Vol. I. 2nd ed.
33 HIRSCH. Differential Topology. 65 WELLS. Differential Analysis on Complex
Manifolds. 2nd ed.
(continued after index)
Henning Stichtenoth
Sabanci University
Faculty of Engineering & Natural Sciences
34956 Istanbul
Orhanli, Tuzla
Turkey
[email protected]

Editorial Board
S. Axler K.A. Ribet
Mathematics Department Mathematics Department
San Francisco State University University of California, Berkeley
San Francisco, CA 94132 Berkeley, CA 94720-3840
USA USA
[email protected] [email protected]

ISSN: 0072-5285
ISBN: 978-3-540-76877-7 e-ISBN: 978-3-540-76878-4

Library of Congress Control Number: 2008938193

Mathematics Subject Classification (2000): 12xx, 94xx, 14xx, 11xx

°c 2009 Springer-Verlag Berlin Heidelberg

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,
reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication
or parts thereof is permitted only under the provisions of the German Copyright Law of September 9,
1965, in its current version, and permission for use must always be obtained from Springer. Violations
are liable for prosecution under the German Copyright Law.

The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant pro-
tective laws and regulations and therefore free for general use.

Cover design: deblik, Berlin

Printed on acid-free paper

987654321
springer.com
Preface to the Second Edition

15 years after the first printing of Algebraic Function Fields and Codes, the
mathematics editors of Springer Verlag encouraged me to revise and extend
the book.
Besides numerous minor corrections and amendments, the second edition
differs from the first one in two respects. Firstly I have included a series
of exercises at the end of each chapter. Some of these exercises are fairly
easy and should help the reader to understand the basic concepts, others are
more advanced and cover additional material. Secondly a new chapter titled
“Asymptotic Bounds for the Number of Rational Places” has been added.
This chapter contains a detailed presentation of the asymptotic theory of
function fields over finite fields, including the explicit construction of some
asymptotically good and optimal towers. Based on these towers, a complete
and self-contained proof of the Tsfasman-Vladut-Zink Theorem is given. This
theorem is perhaps the most beautiful application of function fields to coding
theory.
The codes which are constructed from algebraic function fields were first
introduced by V. D. Goppa. Accordingly I referred to them in the first edition
as geometric Goppa codes. Since this terminology has not generally been ac-
cepted in the literature, I now use the more common term algebraic geometry
codes or AG codes.
I would like to thank Alp Bassa, Arnaldo Garcia, Cem Güneri, Sevan
Harput and Alev Topuzoğlu for their help in preparing the second edition.
Moreover I thank all those whose results I have used in the exercises without
giving references to their work.

İstanbul, September 2008 Henning Stichtenoth

Preface to the First Edition

An algebraic function field over K is an algebraic extension of finite degree

over the rational function field K(x) (the ground field K may be an arbitrary
field). This type of field extension occurs naturally in various branches of
mathematics such as algebraic geometry, number theory and the theory of
compact Riemann surfaces. Hence one can study algebraic function fields from
very different points of view.
In algebraic geometry one is interested in the geometric properties of an
algebraic curve C = {(α, β) ∈ K × K | f (α, β) = 0}, where f (X, Y ) is an
irreducible polynomial in two variables over an algebraically closed field K.
It turns out that the field K(C) of rational functions on C (which is an
algebraic function field over K) contains a great deal of information regarding
the geometry of the curve C. This aspect of the theory of algebraic function
fields is presented in several books on algebraic geometry, for instance [11],
[18], [37] and [38].
One can also approach function fields from the direction of complex anal-
ysis. The meromorphic functions on a compact Riemann surface S form an
algebraic function field M(S) over the field C of complex numbers. Here again,
the function field is a strong tool for studying the corresponding Riemann sur-
face, see [10] or [20].
In this book a self-contained, purely algebraic exposition of the theory
of algebraic functions is given. This approach was initiated by R. Dedekind,
L. Kronecker and H. M. Weber in the nineteenth century (over the field C), cf.
[20]; it was further developed by E. Artin, H. Hasse, F. K. Schmidt and A. Weil
in the first half of the twentieth century. Standard references are Chevalley’s
book ‘Introduction to the Theory of Algebraic Functions of One Variable’ [6],
which appeared in 1951, and [7]. The close relationship with algebraic number
theory is emphasized in [1] and [9].
The algebraic approach to algebraic functions is more elementary than the
approach via algebraic geometry: only some basic knowledge of algebraic field
extensions, including Galois theory, is assumed. A second advantage is that
VIII Preface to the First Edition

some principal results of the theory (such as the Riemann-Roch Theorem) can
be derived very quickly for function fields over an arbitrary constant field K.
This facilitates the presentation of some applications of algebraic functions to
coding theory, which is the second objective of the book.
An error-correcting code is a subspace of IFnq , the n-dimensional standard
vector space over a finite field IFq . Such codes are in widespread use for the
reliable transmission of information. As observed by V. D. Goppa in 1975, one
can use algebraic function fields over IFq to construct a large class of inter-
esting codes. Properties of these codes are closely related to properties of the
corresponding function field, and the Riemann-Roch Theorem provides esti-
mates, sharp in many cases, for their main parameters (dimension, minimum
distance).
While Goppa’s construction is the most important, it is not the only link
between codes and algebraic functions. For instance, the Hasse-Weil Theorem
(which is fundamental to the theory of function fields over a finite constant
field) yields results on the weights of codewords in certain trace codes.

A brief summary of the book follows.

The general theory of algebraic function fields is presented in Chapters 1,
3 and 4. In the first chapter the basic concepts are introduced, and A. Weil’s
proof of the Riemann-Roch Theorem is given. Chapter 3 is perhaps the most
important. It provides the tools necessary for working with concrete func-
tion fields: the decomposition of places in a finite extension, ramification and
different, the Hurwitz Genus Formula, and the theory of constant field ex-
tensions. P -adic completions as well as the relation between differentials and
Weil differentials are treated in Chapter 4.
Chapter 5 deals with function fields over a finite constant field. This chap-
ter contains a version of Bombieri’s proof of the Hasse-Weil Theorem as well as
some improvements of the Hasse-Weil Bound. As an illustration of the general
theory, several explicit examples of function fields are discussed in Chapter 6,
namely elliptic and hyperelliptic function fields, Kummer and Artin-Schreier
extensions of the rational function field.
The Chapters 2, 8 and 9 are devoted to applications of algebraic functions
to coding theory. Following a brief introduction to coding theory, Goppa’s
construction of codes by means of an algebraic function field is described
in Chapter 2. Also included in this chapter is the relation these codes have
with the important classes of BCH and classical Goppa codes. Chapter 8 con-
tains some supplements: the residue representation of geometric Goppa codes,
automorphisms of codes, asymptotic questions and the decoding of geomet-
ric Goppa codes. A detailed exposition of codes associated to the Hermitian
function field is given. In the literature these codes often serve as a test for
the usefulness of geometric Goppa codes.
Chapter 9 contains some results on subfield subcodes and trace codes.
Estimates for their dimension are given, and the Hasse-Weil Bound is used
 Preface to the First Edition IX

to obtain results on the weights, dimension and minimum distance of these

codes.
For the convenience of the reader, two appendices are enclosed. Appendix
A is a summary of results from field theory that are frequently used in the
text. As many papers on geometric Goppa codes are written in the language
of algebraic geometry, Appendix B provides a kind of dictionary between the
theory of algebraic functions and the theory of algebraic curves.

Acknowledgements

First of all I am indebted to P. Roquette from whom I learnt the theory

of algebraic functions. His lectures, given 20 years ago at the University of
Heidelberg, substantially influenced my exposition of this theory.
I thank several colleagues who carefully read the manuscript: D. Ehrhard,
P. V. Kumar, J. P. Pedersen, H.-G. Rück, C. Voss and K. Yang. They sug-
gested many improvements and helped to eliminate numerous misprints and
minor mistakes in earlier versions.

Essen, March 1993 Henning Stichtenoth

Contents

1 Foundations of the Theory of Algebraic Function Fields . . . 1

1.1 Places . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The Rational Function Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Independence of Valuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.5 The Riemann-Roch Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.6 Some Consequences of the Riemann-Roch Theorem . . . . . . . . . . 31
1.7 Local Components of Weil Differentials . . . . . . . . . . . . . . . . . . . . . 37
1.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

2 Algebraic Geometry Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

2.1 Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2 AG Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.3 Rational AG Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3 Extensions of Algebraic Function Fields . . . . . . . . . . . . . . . . . . . 67

3.1 Algebraic Extensions of Function Fields . . . . . . . . . . . . . . . . . . . . 68
3.2 Subrings of Function Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.3 Local Integral Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.4 The Cotrace of Weil Differentials and the Hurwitz Genus
Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.5 The Different . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.6 Constant Field Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.7 Galois Extensions I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.8 Galois Extensions II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
3.9 Ramification and Splitting in the Compositum of Function
Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
3.10 Inseparable Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
3.11 Estimates for the Genus of a Function Field . . . . . . . . . . . . . . . . 145
3.12 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
XII Contents

4 Differentials of Algebraic Function Fields . . . . . . . . . . . . . . . . . . 155

4.1 Derivations and Differentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
4.2 The P -adic Completion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.3 Differentials and Weil Differentials . . . . . . . . . . . . . . . . . . . . . . . . . 170
4.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

5 Algebraic Function Fields over Finite Constant Fields . . . . . 185

5.1 The Zeta Function of a Function Field . . . . . . . . . . . . . . . . . . . . . 185
5.2 The Hasse-Weil Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
5.3 Improvements of the Hasse-Weil Bound . . . . . . . . . . . . . . . . . . . . 208
5.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212

6 Examples of Algebraic Function Fields . . . . . . . . . . . . . . . . . . . . . 217

6.1 Elliptic Function Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
6.2 Hyperelliptic Function Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
6.3 Tame Cyclic Extensions of the Rational Function Field . . . . . . . 227
6.4 Some Elementary Abelian p-Extensions of K(x),
char K = p > 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
6.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

7 Asymptotic Bounds for the Number of Rational Places . . . . 243

7.1 Ihara’s Constant A(q) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
7.2 Towers of Function Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
7.3 Some Tame Towers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
7.4 Some Wild Towers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
7.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285

8 More about Algebraic Geometry Codes . . . . . . . . . . . . . . . . . . . . 289

8.1 The Residue Representation of CΩ (D, G) . . . . . . . . . . . . . . . . . . . 289
8.2 Automorphisms of AG Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
8.3 Hermitian Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
8.4 The Tsfasman-Vladut-Zink Theorem . . . . . . . . . . . . . . . . . . . . . . . 297
8.5 Decoding AG Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
8.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308

9 Subfield Subcodes and Trace Codes . . . . . . . . . . . . . . . . . . . . . . . . 311

9.1 On the Dimension of Subfield Subcodes and Trace Codes . . . . . 311
9.2 Weights of Trace Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
9.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325

Appendix A. Field Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327

Appendix B. Algebraic Curves and Function Fields . . . . . . . . . . . . 335

List of Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

 Contents XIII

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
1
Foundations of the Theory of Algebraic
Function Fields

In this chapter we introduce the basic definitions and results of the theory of
algebraic function fields: valuations, places, divisors, the genus of a function
field, adeles, Weil differentials and the Riemann-Roch Theorem.
Throughout Chapter 1 we denote by K an arbitrary field.
It is only in later chapters that we will assume that K has specific properties
(for example, that K is a finite field – the case which is of particular interest
to coding theory).

1.1 Places
Definition 1.1.1. An algebraic function field F/K of one variable over K is
an extension field F ⊇ K such that F is a finite algebraic extension of K(x)
for some element x ∈ F which is transcendental over K.

For brevity we shall simply refer to F/K as a function field. Obviously the set
K̃ := {z ∈ F | z is algebraic over K} is a subfield of F , since sums, products
and inverses of algebraic elements are also algebraic. K̃ is called the field
of constants of F/K. We have K ⊆ K̃  F , and it is clear that F/K̃ is a
function field over K̃. We say that K is algebraically closed in F (or K is the
full constant field of F ) if K̃ = K.

Remark 1.1.2. The elements of F which are transcendental over K can be

characterized as follows: z ∈ F is transcendental over K if and only if the
extension F/K(z) is of finite degree. The proof is trivial.

Example 1.1.3. The simplest example of an algebraic function field is the ra-
tional function field; F/K is called rational if F = K(x) for some x ∈ F
which is transcendental over K. Each element 0 = z ∈ K(x) has a unique
representation
H. Stichtenoth, Algebraic Function Fields and Codes, 1
Graduate Texts in Mathematics 254,

c Springer-Verlag Berlin Heidelberg 2009
2 1 Foundations of the Theory of Algebraic Function Fields

z =a· pi (x)ni , (1.1)
i
in which 0 = a ∈ K, the polynomials pi (x) ∈ K[x] are monic, pairwise distinct
and irreducible and ni ∈ ZZ.

A function field F/K is often represented as a simple algebraic field extension

of a rational function field K(x); i.e., F = K(x, y) where ϕ(y) = 0 for some ir-
reducible polynomial ϕ(T ) ∈ K(x)[T ]. If F/K is a non-rational function field,
it is not so clear, whether every element 0 = z ∈ F admits a decomposition
into irreducibles analogous to (1.1); indeed, it is not even clear what we mean
by an irreducible element of F . Another problem which is closely related to
the representation (1.1) is the following: given elements α1 , . . . , αn ∈ K, find
all rational functions f (x) ∈ K(x) with a prescribed order of zero (or pole or-
der) at α1 , . . . , αn . In order to formulate these problems for arbitrary function
fields properly, we introduce the notions of valuation rings and places.
Definition 1.1.4. A valuation ring of the function field F/K is a ring O ⊆ F
with the following properties:
(1) K  O  F , and
(2) for every z ∈ F we have that z ∈ O or z −1 ∈ O.

This definition is motivated by the following observation in the case of a

rational function field K(x): given an irreducible monic polynomial p(x) ∈
K[x], we consider the set
 
f (x) 
Op(x) :=  f (x), g(x) ∈ K[x], p(x)  g(x) .
g(x)
It is easily verified that Op(x) is a valuation ring of K(x)/K. Note that if q(x)
is another irreducible monic polynomial, then Op(x) = Oq(x) .
Proposition 1.1.5. Let O be a valuation ring of the function field F/K. Then
the following hold:
(a) O is a local ring; i.e., O has a unique maximal ideal P = O \ O× , where
O× = {z ∈ O | there is an element w ∈ O with zw = 1} is the group of
units of O.
(b) Let 0 = x ∈ F . Then x ∈ P ⇐⇒ x−1 ∈ O.
(c) For the field K̃ of constants of F/K we have K̃ ⊆ O and K̃ ∩ P = {0}.
Proof. (a) We claim that P := O \ O× is an ideal of O (from this it follows
at once that P is the unique maximal ideal since a proper ideal of O cannot
contain a unit).
(1) Let x ∈ P , z ∈ O . Then xz ∈ O× (otherwise x would be a unit),
consequently xz ∈ P .
(2) Let x, y ∈ P . W.l.o.g. we can assume that x/y ∈ O. Then 1 + x/y ∈ O
and x + y = y(1 + x/y) ∈ P by (1). Hence P is an ideal of O.
 1.1 Places 3

(b) is obvious.
(c) Let z ∈ K̃. Assume that z ∈ O . Then z −1 ∈ O as O is a valuation
ring. Since z −1 is algebraic over K, there are elements a1 , . . . , ar ∈ K with
ar (z −1 )r + . . . + a1 z −1 + 1 = 0, hence z −1 (ar (z −1 )r−1 + . . . + a1 ) = −1.
Therefore z = −(ar (z −1 )r−1 + . . . + a1 ) ∈ K[z −1 ] ⊆ O, so z ∈ O. This is a
contradiction to the assumption z ∈ O. Hence we have shown that K̃ ⊆ O.
The assertion K̃ ∩ P = {0} is trivial.

Theorem 1.1.6. Let O be a valuation ring of the function field F/K and let
P be its unique maximal ideal. Then the following hold:
(a) P is a principal ideal.
(b) If P = tO then each 0 = z ∈ F has a unique representation of the form
z = tn u for some n ∈ ZZ and u ∈ O× .
(c) O is a principal ideal domain. More precisely, if P = tO and {0} = I ⊆ O
is an ideal then I = tn O for some n ∈ IN.

A ring having the above properties is called a discrete valuation ring. The
proof of Theorem 1.1.6 depends essentially on the following lemma.

Lemma 1.1.7. Let O be a valuation ring of the algebraic function field F/K,
let P be its maximal ideal and 0 = x ∈ P . Let x1 , . . . , xn ∈ P be such that
x1 = x and xi ∈ xi+1 P for i = 1, . . . , n − 1. Then we have

n ≤ [F : K(x)] < ∞ .

Proof. From Remark 1.1.2 and Proposition 1.1.5(c) follows that F/K(x) is
a finite extension, so it is sufficient to prove that x1 , . . . , xn are linearly
independent
n over K(x). Suppose there is a non-trivial linear combination
i=1 ϕ i (x)xi = 0 with ϕi (x) ∈ K(x). We can assume that all ϕi (x) are
polynomials in x and that x does not divide all of them. Put ai := ϕi (0), the
constant term of ϕi (x), and define j ∈ {1, . . . , n} by the condition aj = 0 but
ai = 0 for all i > j. We obtain

−ϕj (x)xj = ϕi (x)xi (1.2)
i=j

with ϕi (x) ∈ O for i = 1, . . . , n (since x = x1 ∈ P ), xi ∈ xj P for i < j and

ϕi (x) = xgi (x) for i > j, where gi (x) is a polynomial in x. Dividing (1.2) by
xj yields
 xi  x
−ϕj (x) = ϕi (x) + gi (x)xi .
i<j
xj x
i>j j

All summands on the right hand side belong to P , therefore ϕj (x) ∈ P . On

the other hand, ϕj (x) = aj + xgj (x) with gj (x) ∈ K[x] ⊆ O and x ∈ P , so
that aj = ϕj (x) − xgj (x) ∈ P ∩ K. Since aj = 0, this contradicts Proposition
1.1.5(c).
4 1 Foundations of the Theory of Algebraic Function Fields

Proof of Theorem 1.1.6. (a) Assume that P is not principal, and choose an
element 0 = x1 ∈ P . As P = x1 O there is x2 ∈ P \ x1 O. Then x2 x−1 1 ∈ O,
thereby x−1
2 x1 ∈ P by Proposition 1.1.5(b), so x1 ∈ x2 P . By induction one
obtains an infinite sequence x1 , x2 , x3 , . . . in P such that xi ∈ xi+1 P for all
i ≥ 1, a contradiction to Lemma 1.1.7.
(b) The uniqueness of the representation z = tn u with u ∈ O× is trivial, so
we only need to show the existence. As z or z −1 is in O we can assume that
z ∈ O. If z ∈ O× then z = t0 z. It remains to consider the case z ∈ P . There
is a maximal m ≥ 1 with z ∈ tm O, since the length of a sequence
x1 = z, x2 = tm−1 , x3 = tm−2 , . . . , xm = t
is bounded by Lemma 1.1.7. Write z = tm u with u ∈ O. Then u must be a unit
of O (otherwise u ∈ P = tO, so u = tw with w ∈ O and z = tm+1 w ∈ tm+1 O,
a contradiction to the maximality of m).
(c) Let {0} = I ⊆ O be an ideal. The set A := {r ∈ IN | tr ∈ I} is non-empty
(in fact, if 0 = x ∈ I then x = tr u with u ∈ O× and therefore tr = xu−1 ∈ I).
Put n := min (A). We claim that I = tn O. The inclusion I ⊇ tn O is trivial
since tn ∈ I. Conversely let 0 = y ∈ I. We have y = ts w with w ∈ O× and
s ≥ 0, so ts ∈ I and s ≥ n. It follows that y = tn · ts−n w ∈ tn O.

Definition 1.1.8. (a) A place P of the function field F/K is the maximal
ideal of some valuation ring O of F/K. Every element t ∈ P such that P =
tO is called a prime element for P (other notations are local parameter or
uniformizing variable).
(b) IPF := {P | P is a place of F/K}.

If O is a valuation ring of F/K and P is its maximal ideal, then O is

uniquely determined by P , namely O = {z ∈ F | z −1 ∈ P }, cf. Proposition
1.1.5(b). Hence OP := O is called the valuation ring of the place P .
A second useful description of places is given in terms of valuations.
Definition 1.1.9. A discrete valuation of F/K is a function v : F → ZZ∪{∞}
with the following properties :
(1) v(x) = ∞ ⇐⇒ x = 0 .
(2) v(xy) = v(x) + v(y) for all x, y ∈ F .
(3) v(x + y) ≥ min {v(x), v(y)} for all x, y ∈ F .
(4) There exists an element z ∈ F with v(z) = 1.
(5) v(a) = 0 for all 0 = a ∈ K.

In this context the symbol ∞ means some element not in ZZ such that
∞ + ∞ = ∞ + n = n + ∞ = ∞ and ∞ > m for all m, n ∈ ZZ. From (2) and
(4) it follows immediately that v : F → ZZ ∪ {∞} is surjective. Property (3) is
called the Triangle Inequality. The notions valuation and triangle inequality
are justified by the following remark:
 1.1 Places 5

Remark 1.1.10. Let v be a discrete valuation of F/K in the sense of Definition

1.1.9. Fix a real number 0 < c < 1 and define the function | |v : F → IR by

cv(z) if z = 0 ,
|z|v :=
0 if z = 0 .

It is easily verified that this function has the properties of an ordinary absolute
value; the ordinary Triangle Inequality |x + y|v ≤ |x|v + |y|v turns out to be
an immediate consequence of condition (3) of Definition 1.1.9.
A stronger version of the Triangle Inequality can be derived from the
axioms and is often very useful:

Lemma 1.1.11 (Strict Triangle Inequality). Let v be a discrete valuation

of F/K and let x, y ∈ F with v(x) = v(y). Then v(x + y) = min {v(x), v(y)}.

Proof. Observe that v(ay) = v(y) for 0 = a ∈ K (by (2) and (5)), in particular
v(−y) = v(y). Since v(x) = v(y) we can assume v(x) < v(y). Suppose that
v(x + y) = min {v(x), v(y)}, so v(x + y) > v(x) by (3). Then we obtain
v(x) = v((x + y) − y) ≥ min {v(x + y), v(y)} > v(x), a contradiction.

Definition 1.1.12. To a place P ∈ IPF we associate a function vP : F →

ZZ∪{∞} (that will prove to be a discrete valuation of F/K) as follows: Choose
a prime element t for P . Then every 0 = z ∈ F has a unique representation
z = tn u with u ∈ OP× and n ∈ ZZ. Define vP (z) := n and vP (0) := ∞.

Observe that this definition depends only on P , not on the choice of t. In

fact, if t is another prime element for P then P = tO = t O, so t = t w for
some w ∈ OP× . Therefore tn u = (tn wn )u = tn (wn u) with wn u ∈ OP× .

Theorem 1.1.13. Let F/K be a function field.

(a) For a place P ∈ IPF , the function vP defined above is a discrete valuation
of F/K. Moreover we have

OP = {z ∈ F | vP (z) ≥ 0} ,
OP× = {z ∈ F | vP (z) = 0} ,
P = {z ∈ F | vP (z) > 0} .

(b) An element x ∈ F is a prime element for P if and only if vP (x) = 1.

(c) Conversely, suppose that v is a discrete valuation of F/K. Then the set
P := {z ∈ F | v(z) > 0} is a place of F/K, and OP = {z ∈ F | v(z) ≥ 0} is
the corresponding valuation ring.
(d) Every valuation ring O of F/K is a maximal proper subring of F .
6 1 Foundations of the Theory of Algebraic Function Fields

Proof. (a) Obviously vP has the properties (1), (2), (4) and (5) of Definition
1.1.9. In order to prove the Triangle Inequality (3) consider x, y ∈ F with
vP (x) = n, vP (y) = m. We can assume that n ≤ m < ∞, thus x = tn u1
and y = tm u2 with u1 , u2 ∈ OP× . Then x + y = tn (u1 + tm−n u2 ) = tn z with
z ∈ OP . If z = 0 we have vP (x + y) = ∞ > min{n, m}, otherwise z = tk u
with k ≥ 0 and u ∈ OP× . Therefore

vP (x + y) = vP (tn+k u) = n + k ≥ n = min{vP (x), vP (y)} .

We have shown that vP is a discrete valuation of F/K. The remaining asser-

tions of (a) are trivial, likewise (b) and (c).
(d) Let O be a valuation ring of F/K, P its maximal ideal, vP the discrete
valuation associated to P and z ∈ F \ O . We have to show that F = O[z].
To this end consider an arbitrary element y ∈ F ; then vP (yz −k ) ≥ 0 for
sufficiently large k ≥ 0 (note that vP (z −1 ) > 0 since z ∈ O ). Consequently
w := yz −k ∈ O and y = wz k ∈ O[z].

According to Theorem 1.1.13 places, valuation rings and discrete valuations

of a function field essentially amount to the same thing.
Let P be a place of F/K and let OP be its valuation ring. Since P is a
maximal ideal, the residue class ring OP /P is a field. For x ∈ OP we define
x(P ) ∈ OP /P to be the residue class of x modulo P , for x ∈ F \ OP we
put x(P ) := ∞ (note that the symbol ∞ is used here in a different sense
as in Definition 1.1.9). By Proposition 1.1.5 we know that K ⊆ OP and
K ∩ P = {0}, so the residue class map OP → OP /P induces a canonical
embedding of K into OP /P . Henceforth we shall always consider K as a
subfield of OP /P via this embedding. Observe that this argument also applies
to K̃ instead of K; so we can consider K̃ as a subfield of OP /P as well.

Definition 1.1.14. Let P ∈ IPF .

(a) FP := OP /P is the residue class field of P . The map x → x(P ) from F
to FP ∪ {∞} is called the residue class map with respect to P . Sometimes we
shall also use the notation x + P := x(P ) for x ∈ OP .
(b) deg P := [FP : K] is called the degree of P . A place of degree one is also
called a rational place of F/K.

The degree of a place is always finite; more precisely the following holds.

Proposition 1.1.15. If P is a place of F/K and 0 = x ∈ P then

deg P ≤ [F : K(x)] < ∞ .

Proof. First we observe that [F : K(x)] < ∞ by Remark 1.1.2. Thus it suf-
fices to show that any elements z1 , . . . , zn ∈ OP , whose residue classes
z1 (P ), . . . , zn (P ) ∈ FP are linearly independent over K, are linearly indepen-
dent over K(x). Suppose there is a non-trivial linear combination
 1.1 Places 7


n
ϕi (x)zi = 0 (1.3)
i=1

with ϕi (x) ∈ K(x). W.l.o.g. we assume that the ϕi (x) are polynomials in x
and not all of them are divisible by x; i.e., ϕi (x) = ai + xgi (x) with ai ∈ K
and gi (x) ∈ K[x], not all ai = 0. Since x ∈ P and gi (x) ∈ OP , ϕi (x)(P ) =
ai (P ) = ai . Applying the residue class map to (1.3) we obtain

n 
n
0 = 0(P ) = ϕi (x)(P )zi (P ) = ai zi (P ) .
i=1 i=1

This contradicts the linear independence of z1 (P ), . . . , zn (P ) over K.

Corollary 1.1.16. The field K̃ of constants of F/K is a finite field extension

of K.

Proof. We use the fact that IPF = ∅ (which will be proved only in Corollary
1.1.20). Choose some P ∈ IPF . Since K̃ is embedded into FP via the residue
class map OP → FP , it follows that [K̃ : K] ≤ [FP : K] < ∞.

Remark 1.1.17. Let P be a rational place of F/K; i.e., deg P = 1. Then we

have FP = K, and the residue class map maps F to K ∪ {∞} . In particular,
if K is an algebraically closed field, then all places are rational and we can
read an element z ∈ F as a function

IPF −→ K ∪ {∞} ,
z: (1.4)
P −→ z(P ) .

This is why F/K is called a function field. The elements of K, interpreted

as functions in the sense of (1.4), are constant functions. For this reason K
is called the constant field of F . Also the following terminology is justified
by (1.4):

Definition 1.1.18. Let z ∈ F and P ∈ IPF . We say that P is a zero of z if

vP (z) > 0; P is a pole of z if vP (z) < 0. If vP (z) = m > 0, P is a zero of z
of order m; if vP (z) = −m < 0, P is a pole of z of order m.

Next we shall be concerned with the question as to whether there exist

places of F/K.

Theorem 1.1.19. Let F/K be a function field and let R be a subring of F

with K ⊆ R ⊆ F . Suppose that {0} = I  R is a proper ideal of R. Then
there is a place P ∈ IPF such that I ⊆ P and R ⊆ OP .

Proof. Consider the set

F := { S | S is a subring of F with R ⊆ S and IS = S } .

8 1 Foundations of the Theory of Algebraic Function Fields

(IS is by definition the set of all finite sums aν sν with aν ∈ I, sν ∈ S;
it is an ideal of S). F is non-empty as R ∈ F, and F is inductively or-
dered by inclusion. In fact, if H ⊆ F is a totally ordered subset of F then
T := { S | S ∈ H } is a subring of F with nR ⊆ T . We have to verify that
IT = T . Suppose this is false, then 1 = ν=1 aν sν with aν ∈ I, sν ∈ T .
SinceH is totally ordered there is an S0 ∈ H such that s1 , . . . , sn ∈ S0 , so
n
1 = ν=1 aν sν ∈ IS0 , a contradiction.
By Zorn’s lemma F contains a maximal element; i.e., there is a ring O ⊆ F
such that R ⊆ O ⊆ F , IO = O, and O is maximal with respect to these
properties. We want to show that O is a valuation ring of F/K.
As I = {0} and IO = O we have O  F and I ⊆ O \ O× . Suppose there
exists an element z ∈ F with z ∈ O and z −1 ∈ O. Then IO[z] = O[z] and
IO[z −1 ] = O[z −1 ], and we can find a0 , . . . , an , b0 , . . . , bm ∈ IO with

1 = a0 + a1 z + · · · + an z n and (1.5)
1 = b0 + b1 z −1 + · · · + bm z −m . (1.6)

Clearly n ≥ 1 and m ≥ 1. We can assume that m, n in (1.5) and (1.6) are

chosen minimally and m ≤ n. We multiply (1.5) by 1 − b0 and (1.6) by an z n
and obtain

1 − b0 = (1 − b0 )a0 + (1 − b0 )a1 z + · · · + (1 − b0 )an z n and

0 = (b0 − 1)an z n + b1 an z n−1 + · · · + bm an z n−m .

Adding these equations yields 1 = c0 + c1 z + · · · + cn−1 z n−1 with coefficients

ci ∈ IO. This is a contradiction to the minimality of n in (1.5). Thus we have
proved that z ∈ O or z −1 ∈ O for all z ∈ F , hence O is a valuation ring of
F/K.

Corollary 1.1.20. Let F/K be a function field, z ∈ F transcendental over

K. Then z has at least one zero and one pole. In particular IPF = ∅.

Proof. Consider the ring R = K[z] and the ideal I = zK[z]. Theorem 1.1.19
ensures that there is a place P ∈ IPF with z ∈ P , hence P is a zero of z. The
same argument proves that z −1 has a zero Q ∈ IPF . Then Q is a pole of z.

Corollary 1.1.20 can be interpreted as follows: each z ∈ F , which is not in

the constant field K̃ of F/K, yields a non-constant function in the sense of
Remark 1.1.17.

1.2 The Rational Function Field

For a thorough understanding of valuations and places in arbitrary function
fields, a precise idea of these notions in the simplest case is indispensable.
For this reason we investigate what these concepts mean in the case of the
 1.2 The Rational Function Field 9

rational function field F = K(x), where x is transcendental over K. Given an

irreducible monic polynomial p(x) ∈ K[x] we consider the valuation ring
 
f (x) 
Op(x) :=  f (x), g(x) ∈ K[x], p(x)  g(x) (1.7)
g(x)
of K(x)/K with maximal ideal

 
f (x) 
Pp(x) =  f (x), g(x) ∈ K[x], p(x)|f (x), p(x)  g(x) . (1.8)
g(x)

In the particular case when p(x) is linear, i.e. p(x) = x − α with α ∈ K , we

abbreviate and write
Pα := Px−α ∈ IPK(x) . (1.9)
There is another valuation ring of K(x)/K, namely

 
f (x) 
O∞ :=  f (x), g(x) ∈ K[x], deg f (x) ≤ deg g(x) (1.10)
g(x)

with maximal ideal

 
f (x) 
P∞ =  f (x), g(x) ∈ K[x], deg f (x) < deg g(x) . (1.11)
g(x)

This place is called the infinite place of K(x). Observe that these labels
depend on the specific choice of the generating element x of K(x)/K (for
example K(x) = K(1/x), and the infinite place with respect to 1/x is the
place P0 with respect to x).

Proposition 1.2.1. Let F = K(x) be the rational function field.

(a) Let P = Pp(x) ∈ IPK(x) be the place defined by (1.8), where p(x) ∈ K[x]
is an irreducible polynomial. Then p(x) is a prime element for P , and the
corresponding valuation vP can be described as follows: if z ∈ K(x) \ {0} is
written in the form z = p(x)n · (f (x)/g(x)) with n ∈ ZZ, f (x), g(x) ∈ K[x],
p(x)  f (x) and p(x)  g(x), then vP (z) = n. The residue class field K(x)P =
OP /P is isomorphic to K[x]/(p(x)); an isomorphism is given by

K[x]/(p(x)) −→ K(x)P ,
φ:
f (x) mod p(x) −→ f (x)(P ) .

Consequently deg P = deg p(x).

(b) In the special case p(x) = x − α with α ∈ K the degree of P = Pα is one,
and the residue class map is given by

z(P ) = z(α) for z ∈ K(x) ,

10 1 Foundations of the Theory of Algebraic Function Fields

where z(α) is defined as follows: write z = f (x)/g(x) with relatively prime

polynomials f (x), g(x) ∈ K[x]. Then

f (α)/g(α) if g(α) = 0 ,
z(α) =
∞ if g(α) = 0 .
(c) Finally, let P = P∞ be the infinite place of K(x)/K defined by (1.11).
Then deg P∞ = 1. A prime element for P∞ is t = 1/x. The corresponding
discrete valuation v∞ is given by

v∞ (f (x)/g(x)) = deg g(x) − deg f (x) ,

where f (x), g(x) ∈ K[x]. The residue class map corresponding to P∞ is de-
termined by z(P∞ ) = z(∞) for z ∈ K(x), where z(∞) is defined as usual:
if
an xn + · · · + a0
z= with an , bm = 0 ,
bm xm + · · · + b0
then ⎧
⎪
⎨an /bm if n = m ,
z(∞) = 0 if n < m ,
⎪
⎩
∞ if n > m .
(d) K is the full constant field of K(x)/K.

Proof. We prove only some essentials of this proposition; the remaining parts
of the proof are straightforward.
(a) Let P = Pp(x) , p(x) ∈ K[x] irreducible. The ideal Pp(x) ⊆ Op(x) is obvi-
ously generated by p(x), hence p(x) is a prime element for P . In order to prove
the assertion about the residue class field we consider the ring homomorphism

K[x] −→ K(x)P ,
ϕ:
f (x) −→ f (x)(P ) .

Clearly the kernel of ϕ is the ideal generated by p(x). Moreover ϕ is surjective:

if z ∈ Op(x) , we can write z = u(x)/v(x) with u(x), v(x) ∈ K[x] such that
p(x)  v(x). Thus there are a(x), b(x) ∈ K[x] with a(x)p(x) + b(x)v(x) = 1,
therefore
a(x)u(x)
z =1·z = p(x) + b(x)u(x) ,
v(x)
and z(P ) = (b(x)u(x))(P ) is in the image of ϕ. Thus ϕ induces an isomorphism
φ of K[x]/(p(x)) onto K(x)P .
 1.2 The Rational Function Field 11

(b) Now P = Pα with α ∈ K. If f (x) ∈ K[x] then (x − α)|(f (x) − f (α)),

hence f (x)(P ) = (f (x) − f (α))(P ) + f (α)(P ) = f (α). An arbitrary element
z ∈ OP can be written as z = f (x)/g(x) with polynomials f (x), g(x) ∈ K[x]
and (x − α)  g(x) , therefore g(x)(P ) = g(α) = 0 and
f (x)(P ) f (α)
z(P ) = = = z(α) .
g(x)(P ) g(α)
(c) We will only show that 1/x is a prime element for P∞ . Clearly we have
that 1/x ∈ P . Consider some element z = f (x)/g(x) ∈ P∞ ; i.e., deg f < deg g.
Then
1 xf
z= · , with deg(xf ) ≤ deg g .
x g
This proves that z ∈ (1/x)O∞ , hence 1/x generates the ideal P∞ and it is
therefore a P∞ -prime element.
(d) Choose a place P of K(x)/K of degree one (e.g. P = Pα with α ∈ K). The
field K̃ of constants of K(x) is embedded into the residue class field K(x)P ,
hence K ⊆ K̃ ⊆ K(x)P = K.
Theorem 1.2.2. There are no places of the rational function field K(x)/K
other than the places Pp(x) and P∞ , defined by (1.8) and (1.11).
Corollary 1.2.3. The places of K(x)/K of degree one are in 1–1 – corres-
pondence with K ∪ {∞} .
The corollary is obvious by Proposition 1.2.1 and Theorem 1.2.2. In terms
of algebraic geometry (cf. Appendix B) K ∪ {∞} is usually interpreted as
the projective line P1 (K) over K, hence the places of K(x)/K of degree one
correspond in a one-to-one way with the points of P1 (K).

Proof of Theorem 1.2.2. Let P be a place of K(x)/K. We distinguish two cases

as follows:
Case 1. Assume that x ∈ OP . Then K[x] ⊆ OP . Set I := K[x] ∩ P ; this
is an ideal of K[x] , in fact a prime ideal. The residue class map induces an
embedding K[x]/I → K(x)P , consequently I = {0} by Proposition 1.1.15.
It follows that there is a (uniquely determined) irreducible monic polynomial
p(x) ∈ K[x] such that I = K[x] ∩ P = p(x) · K[x]. Every g(x) ∈ K[x] with
p(x)  g(x) is not in I , so g(x) ∈ P and 1/g(x) ∈ OP by Proposition 1.1.5. We
conclude that
 
f (x) 
Op(x) =  f (x), g(x) ∈ K[x], p(x)  g(x) ⊆ OP .
g(x)
As valuation rings are maximal proper subrings of K(x), cf. Theorem 1.1.13,
we see that OP = Op(x) .
Case 2. Now x ∈ OP . We conclude that K[x−1 ] ⊆ OP , x−1 ∈ P ∩ K[x−1 ]
and P ∩ K[x−1 ] = x−1 K[x−1 ]. As in case 1,
12 1 Foundations of the Theory of Algebraic Function Fields
 
f (x−1 )  −1 −1 −1 −1 −1
OP ⊇  f (x ), g(x ) ∈ K[x ], x  g(x )
g(x−1 )
 
a0 + a1 x−1 + · · · + an x−n 
=  b0 = 0
b0 + b1 x−1 + · · · + bm x−m
 
a0 xm+n + · · · + an xm 
=  b0 = 0
b0 xm+n + · · · + bm xn
 
u(x) 
=  u(x), v(x) ∈ K[x], deg u(x) ≤ deg v(x)
v(x)
= O∞ .

Thus OP = O∞ and P = P∞ .

1.3 Independence of Valuations

The main result of this section is the Weak Approximation Theorem 1.3.1
(which is also referred to as the Theorem of Independence). Essentially this
says the following: If v1 , . . . , vn are pairwise distinct discrete valuations of
F/K and z ∈ F , and if we know the values v1 (z), . . . , vn−1 (z), then we cannot
conclude anything about vn (z). A substantial improvement of Theorem 1.3.1
will be given later in Section 1.6.

Theorem 1.3.1 (Weak Approximation Theorem). Let F/K be a func-

tion field, P1 , . . . , Pn ∈ IPF pairwise distinct places of F/K, x1 , . . . , xn ∈ F
and r1 , . . . , rn ∈ ZZ. Then there is some x ∈ F such that

vPi (x − xi ) = ri for i = 1, . . . , n .

Corollary 1.3.2. Every function field has infinitely many places.

Proof of Corollary 1.3.2. Suppose there are only finitely many places, say
P1 , . . . , Pn . By Theorem 1.3.1 we find a non-zero element x ∈ F with vPi (x) >
0 for i = 1, . . . , n. Then x is transcendental over K since it has zeros. But x
has no pole; this is a contradiction to Corollary 1.1.20.
Proof of Theorem 1.3.1. The proof is somewhat technical and therefore divided
into several steps. For simplicity we write vi instead of vPi .
Step 1. There is some u ∈ F with v1 (u) > 0 and vi (u) < 0 for i = 2, . . . , n.
Proof of Step 1. By induction. For n = 2 we observe that OP1 ⊆ OP2 and vice
versa, since valuation rings are maximal proper subrings of F , cf. Theorem
1.1.13. Therefore we can find y1 ∈ OP1 \ OP2 and y2 ∈ OP2 \ OP1 . Then
v1 (y1 ) ≥ 0, v2 (y1 ) < 0, v1 (y2 ) < 0 and v2 (y2 ) ≥ 0. The element u := y1 /y2
has the property v1 (u) > 0, v2 (u) < 0 as desired.
For n > 2 we have by induction hypothesis an element y with v1 (y) > 0,
v2 (y) < 0, . . . , vn−1 (y) < 0. If vn (y) < 0 the proof is finished. In case vn (y) ≥ 0
 1.3 Independence of Valuations 13

we choose z with v1 (z) > 0, vn (z) < 0 and put u := y + z r . Here r ≥ 1 is

chosen in such a manner that r · vi (z) = vi (y) for i = 1, . . . , n − 1 (this
is obviously possible). It follows that v1 (u) ≥ min{v1 (y), r · v1 (z)} > 0 and
vi (u) = min{vi (y), r · vi (z)} < 0 for i = 2, . . . , n (observe that the Strict
Triangle Inequality applies).
Step 2. There is some w ∈ F such that v1 (w − 1) > r1 and vi (w) > ri for
i = 2, . . . , n.
Proof of Step 2. Choose u as in Step 1 and put w := (1 + us )−1 . We have, for
sufficiently large s ∈ IN, v1 (w − 1) = v1 (−us (1 + us )−1 ) = s · v1 (u) > r1 , and
vi (w) = −vi (1 + us ) = −s · vi (u) > ri for i = 2, . . . , n.
Step 3. Given y1 , . . . , yn ∈ F , there is an element z ∈ F with vi (z −yi ) > ri
for i = 1, . . . , n.
Proof of Step 3. Choose s ∈ ZZ such that vi (yj ) ≥ s for all i, j ∈ {1, . . . , n}.
By Step 2 there are w1 , . . . , wn with

vi (wi − 1) > ri − s and vi (wj ) > ri − s for j = i .

n
Then z := j=1 yj wj has the desired properties.
Now we are in a position to finish the proof of Theorem 1.3.1. By Step
3 we can find z ∈ F with vi (z − xi ) > ri , i = 1, . . . , n. Next we choose zi
with vi (zi ) = ri (this is trivially done). Again by Step 3 there is z  with
vi (z  − zi ) > ri for i = 1, . . . , n. It follows that

vi (z  ) = vi ((z  − zi ) + zi ) = min{vi (z  − zi ), vi (zi )} = ri .

Let x := z + z  . Then

vi (x − xi ) = vi ((z − xi ) + z  ) = min{vi (z − xi ), vi (z  )} = ri .

In Section 1.4 we shall show that an element x ∈ F which is transcendental

over K has as many zeros as poles (counted properly). An important step
towards that result is our next proposition which sharpens both of Lemma
1.1.7 and Proposition 1.1.15. The Weak Approximation Theorem will play a
significant role in the proof.

Proposition 1.3.3. Let F/K be a function field and let P1 , . . . , Pr be zeros

of the element x ∈ F . Then

r
vPi (x) · deg Pi ≤ [F : K(x)] .
i=1

Proof. We set vi := vPi , fi := deg Pi and ei := vi (x). For all i there is an

element ti with

vi (ti ) = 1 and vk (ti ) = 0 for k = i .

14 1 Foundations of the Theory of Algebraic Function Fields

Next we choose si1 , . . . , sifi ∈ OPi such that si1 (Pi ), . . . , sifi (Pi ) form a basis
of the residue class field FPi over K. By a weak application of Theorem 1.3.1
we can find zij ∈ F such that the following holds for all i, j :

vi (sij − zij ) > 0 and vk (zij ) ≥ ek for k = i . (1.12)

We claim that the elements

tai · zij , 1 ≤ i ≤ r, 1 ≤ j ≤ fi , 0 ≤ a < ei

r
are
r linearly independent over K(x). Their number is equal to i=1 fi ei =
i=1 vPi (x) · deg Pi , so the proposition will follow from this claim.
Suppose there is a non-trivial linear combination


r  i −1
fi e
ϕija (x)tai zij = 0 (1.13)
i=1 j=1 a=0

over K(x). W.l.o.g. we can assume that ϕija (x) ∈ K[x] and not all ϕija (x) are
divisible by x. Then there are indices k ∈ {1, . . . , r} and c ∈ {0, . . . , ek − 1}
such that

x | ϕkja (x) for all a < c and all j ∈ {1, . . . , fk }, and

x  ϕkjc (x) for some j ∈ {1, . . . , fk } . (1.14)

Multiplying (1.13) by t−c

k we obtain


r  i −1
fi e
ϕija (x)tai t−c
k zij = 0 . (1.15)
i=1 j=1 a=0

For i = k all summands of (1.15) are in Pk , since

vk (ϕija (x)tai t−c

k zij ) = vk (ϕija (x)) + avk (ti ) − cvk (tk ) + vk (zij )
≥ 0 + 0 − c + ek > 0 .

For i = k and a < c we have

k zkj ) ≥ ek + a − c ≥ ek − c > 0 .
vk (ϕkja (x)ta−c

(Note that x|ϕkja (x) and therefore vk (ϕkja (x)) ≥ ek .) For i = k and a > c,

k zkj ) ≥ a − c > 0 .
vk (ϕkja (x)ta−c

Combining the above with (1.15) gives


fk
ϕkjc (x)zkj ∈ Pk . (1.16)
j=1
 1.4 Divisors 15

Observe that ϕkjc (x)(Pk ) ∈ K, and not all ϕkjc (x)(Pk ) = 0 (by (1.14)), so
(1.16) yields a non-trivial linear combination


fk
ϕkjc (x)(Pk ) · zkj (Pk ) = 0
j=1

over K. This is a contradiction, as zk1 (Pk ), . . . , zkfk (Pk ) form a basis of

FPk /K.

Corollary 1.3.4. In a function field F/K every element 0 = x ∈ F has only

finitely many zeros and poles.

Proof. If x is constant, x has neither zeros nor poles. If x is transcendental

over K, the number of zeros is ≤ [F : K(x)] by Proposition 1.3.3. The same
argument shows that x−1 has only a finite number of zeros.

1.4 Divisors
The field K̃ of constants of an algebraic function field F/K is a finite extension
field of K, cf. Corollary 1.1.16, and F can be regarded as a function field
over K̃. Therefore the following assumption (which we maintain throughout
the whole book) is not critical to the theory:
From here on, F/K will always denote an algebraic function field
of one variable such that K is the full constant field of F/K.

Definition 1.4.1. The divisor group of F/K is defined as the (additively writ-
ten) free abelian group which is generated by the places of F/K; it is denoted by
Div(F ). The elements of Div(F ) are called divisors of F/K. In other words,
a divisor is a formal sum

D= nP P with nP ∈ ZZ, almost all nP = 0 .
P ∈IPF

The support of D is defined as

supp D := {P ∈ IPF | nP = 0} .

It will often be found convenient to write


D= nP P ,
P ∈S

where S ⊆ IPF is a finite set with S ⊇ supp D .

A divisor of the
 form D = P with  P ∈ IPF is called a prime divisor. Two
divisors D = nP P and D = nP P are added coefficientwise,
16 1 Foundations of the Theory of Algebraic Function Fields


D + D = (nP + nP )P .
P ∈IPF

The zero element of the divisor group Div(F ) is the divisor


0 := rP P , all rP = 0 .
P ∈IPF

For Q ∈ IPF and D = nP P ∈ Div(F ) we define vQ (D) := nQ , therefore

supp D = {P ∈ IPF | vP (D) = 0} and D= vP (D) · P .
P ∈supp D

A partial ordering on Div(F ) is defined by

D1 ≤ D2 : ⇐⇒ vP (D1 ) ≤ vP (D2 ) for all P ∈ IPF .

If D1 ≤ D2 and D1 = D2 we will also write D1 < D2 . A divisor D ≥ 0 is
called positive (or effective). The degree of a divisor is defined as

deg D := vP (D) · deg P ,
P ∈IPF

and this yields a homomorphism deg : Div(F ) → ZZ.

By Corollary 1.3.4 a nonzero element x ∈ F has only finitely many zeros
and poles in IPF . Thus the following definition makes sense.
Definition 1.4.2. Let 0 = x ∈ F and denote by Z (resp. N ) the set of zeros
(resp. poles) of x in IPF . Then we define

(x)0 := vP (x)P , the zero divisor of x ,
P ∈Z

(x)∞ := (−vP (x))P , the pole divisor of x ,
P ∈N
(x) := (x)0 − (x)∞ , the principal divisor of x .

Clearly (x)0 ≥ 0, (x)∞ ≥ 0 and


(x) = vP (x)P . (1.17)
P ∈IPF

The elements 0 = x ∈ F which are constant are characterized by

x ∈ K ⇐⇒ (x) = 0 .
This follows immediately from Corollary 1.1.20 (note the general assumption
made previously that K is algebraically closed in F ).
 1.4 Divisors 17

Definition 1.4.3. The set of divisors

Princ(F ) := { (x) | 0 = x ∈ F }

is called the group of principal divisors of F/K. This is a subgroup of Div(F ),

since for 0 = x, y ∈ F , (xy) = (x) + (y) by (1.17). The factor group

Cl(F ) := Div(F )/Princ(F )

is called the divisor class group of F/K. For a divisor D ∈ Div(F ), the cor-
responding element in the factor group Cl(F ) is denoted by [D], the divisor
class of D. Two divisors D, D ∈ Div(F ) are said to be equivalent, written

D ∼ D ,

if [D] = [D ]; i.e., D = D + (x) for some x ∈ F \ {0}. This is easily verified
to be an equivalence relation.

Our next definition plays a fundamental role in the theory of algebraic

function fields.

Definition 1.4.4. For a divisor A ∈ Div(F ) we define the Riemann-Roch

space associated to A by

L (A) := { x ∈ F | (x) ≥ −A } ∪ {0} .

This definition has the following interpretation: if


r 
s
A= ni Pi − mj Qj
i=1 j=1

with ni > 0, mj > 0 then L (A) consists of all elements x ∈ F such that
• x has zeros of order ≥ mj at Qj , for j = 1, . . . , s, and
• x may have poles only at the places P1 , . . . , Pr , with the pole order at Pi
being bounded by ni (i = 1, . . . , r).

Remark 1.4.5. Let A ∈ Div(F ). Then

(a) x ∈ L (A) if and only if vP (x) ≥ −vP (A) for all P ∈ IPF .
(b) L (A) = {0} if and only if there is a divisor A ∼ A with A ≥ 0.

The proof of these remarks is trivial; nevertheless they are often very
useful. In particular Remark 1.4.5(b) will be used frequently.

Lemma 1.4.6. Let A ∈ Div(F ). Then we have:

(a) L (A) is a vector space over K.
18 1 Foundations of the Theory of Algebraic Function Fields

(b) If A is a divisor equivalent to A, then L (A)  L (A ) (isomorphic as

vector spaces over K).

Proof. (a) Let x, y ∈ L (A) and a ∈ K. Then for all P ∈ IPF , vP (x + y) ≥

min{vP (x), vP (y)} ≥ −vP (A) and vP (ax) = vP (a) + vP (x) ≥ −vP (A). So
x + y and ax are in L (A) by Remark 1.4.5(a).
(b) By assumption, A = A + (z) with 0 = z ∈ F . Consider the mapping

L (A) −→ F ,
ϕ:
x −→ xz .

This is a K-linear mapping whose image is contained in L (A ). In the same

manner, 
 L (A ) −→ F,
ϕ :
x −→ xz −1
is K-linear from L (A ) to L (A). These mappings are inverse to each other,
hence ϕ is an isomorphism between L (A) and L (A ).

Lemma 1.4.7. (a) L (0) = K .

(b) If A < 0 then L (A) = {0}.

Proof. (a) We have (x) = 0 for 0 = x ∈ K, therefore K ⊆ L (0). Conversely,

if 0 = x ∈ L (0) then (x) ≥ 0. This means that x has no pole, so x ∈ K by
Corollary 1.1.20.
(b) Assume there exists an element 0 = x ∈ L (A). Then (x) ≥ −A > 0,
which implies that x has at least one zero but no pole. This is impossible.

In the sequel we shall consider various K-vector spaces. The dimension of

such a vector space V will be denoted by dim V . Our next objective is to show
that L (A) is finite-dimensional for each divisor A ∈ Div(F ).

Lemma 1.4.8. Let A, B be divisors of F/K with A ≤ B. Then we have

L (A) ⊆ L (B) and

dim(L (B)/L (A)) ≤ deg B − deg A .

Proof. L (A) ⊆ L (B) is trivial. In order to prove the other assertion we can
assume that B = A + P for some P ∈ IPF ; the general case follows then by
induction. Choose an element t ∈ F with vP (t) = vP (B) = vP (A) + 1. For
x ∈ L (B) we have vP (x) ≥ −vP (B) = −vP (t), so xt ∈ OP . Thus we obtain
a K-linear map 
L (B) −→ FP ,
ψ:
x −→ (xt)(P ) .
 1.4 Divisors 19

An element x is in the kernel of ψ if and only if vP (xt) > 0; i.e., vP (x) ≥

−vP (A). Consequently Ker(ψ) = L (A), and ψ induces a K-linear injective
mapping from L (B)/L (A) to FP . It follows that

dim(L (B)/L (A)) ≤ dim FP = deg B − deg A .

Proposition 1.4.9. For each divisor A ∈ Div(F ) the space L (A) is a finite-
dimensional vector space over K. More precisely: if A = A+ −A− with positive
divisors A+ and A− , then

dim L (A) ≤ deg A+ + 1 .

Proof. Since L (A) ⊆ L (A+ ), it is sufficient to show that

dim L (A+ ) ≤ deg A+ + 1 .

We have 0 ≤ A+ , so Lemma 1.4.8 yields dim(L (A+ )/L (0)) ≤ deg A+ . Since
L (0) = K we conclude that dim L (A+ ) = dim(L (A+ )/L (0)) + 1 ≤
deg A+ + 1.

Definition 1.4.10. For A ∈ Div(F ) the integer (A) := dim L (A) is called
the dimension of the divisor A.

One of the most important problems in the theory of algebraic function

fields is to calculate the dimension of a divisor. We shall be concerned with
this question in the subsequent sections; the answer to the problem will be
given by the Riemann-Roch Theorem 1.5.15.
We begin by proving a sharpening of Proposition 1.3.3. Roughly speaking,
the next theorem states that an element 0 = x ∈ F has as many zeros as
poles, provided the zeros and poles are counted properly.

Theorem 1.4.11. All principal divisors have degree zero. More precisely: let
x ∈ F \ K and (x)0 resp. (x)∞ denote the zero resp. pole divisor of x. Then

deg (x)0 = deg (x)∞ = [F : K(x)] .

Corollary 1.4.12. (a) Let A, A be divisors with A ∼ A . Then we have

(A) = (A ) and deg A = deg A .
(b) If deg A < 0 then (A) = 0.
(c) For a divisor A of degree zero the following assertions are equivalent:
(1) A is principal.
(2) (A) ≥ 1.
(3) (A) = 1.
20 1 Foundations of the Theory of Algebraic Function Fields

Proof of Corollary 1.4.12. (a) follows immediately from Lemma 1.4.6 and
Theorem 1.4.11.
(b) Suppose that (A) > 0. By Remark 1.4.5 there is some divisor A ∼ A
with A ≥ 0, hence deg A = deg A ≥ 0.
(c) (1) ⇒ (2) : If A = (x) is principal then x−1 ∈ L (A), so (A) ≥ 1.
(2) ⇒ (3) : Assume now that (A) ≥ 1 and deg A = 0. Then A ∼ A for
some A ≥ 0 (Remark 1.4.5(b)). The conditions A ≥ 0 and deg A = 0 imply
that A = 0, hence (A) = (A ) = (0) = 1, by Lemma 1.4.7.
(3) ⇒ (1) : Suppose that (A) = 1 and deg A = 0. Choose 0 = z ∈ L (A),
then (z) + A ≥ 0. Since deg((z) + A) = 0, it follows that (z) + A = 0, therefore
A = −(z) = (z −1 ) is principal.
Proof of Theorem 1.4.11. Set n := [F : K(x)] and


r
B := (x)∞ = −vPi (x)Pi ,
i=1

where P1 , . . . , Pr are all the poles of x. Then


r
deg B = vPi (x−1 ) · deg Pi ≤ [F : K(x)] = n
i=1

by Proposition 1.3.3, and thus it remains to show that n ≤ deg B as well.

Choose a basis u1 , . . . , un of F/K(x) and a divisor C ≥ 0 such that (ui ) ≥ −C
for i = 1, . . . , n. We have

(lB + C) ≥ n(l + 1) for all l ≥ 0 , (1.18)

which follows immediately from the fact that xi uj ∈ L (lB + C) for 0 ≤ i ≤ l,

1 ≤ j ≤ n (observe that these elements are linearly independent over K since
u1 , . . . , un are linearly independent over K(x)). Setting c := deg C we obtain
n(l + 1) ≤ (lB + C) ≤ l · deg B + c + 1 by Proposition 1.4.9. Thus

l(deg B − n) ≥ n − c − 1 (1.19)

for all l ∈ IN. The right hand side of (1.19) is independent of l, therefore (1.19)
is possible only when deg B ≥ n.
We have thus proved that deg (x)∞ = [F : K(x)]. Since (x)0 = (x−1 )∞ , we
conclude that deg (x)0 = deg (x−1 )∞ = [F : K(x−1 )] = [F : K(x)].

Example 1.4.13. Once again we consider the rational function field F = K(x)
as in Section 1.2. For 0 = z ∈ K(x) we have z = a · f (x)/g(x) with a ∈
K \ {0}, f (x), g(x) ∈ K[x] monic and relatively prime. Let


r 
s
ni
f (x) = pi (x) , g(x) = qj (x)mj
i=1 i=1
 1.4 Divisors 21

with pairwise distinct irreducible monic polynomials pi (x), qj (x) ∈ K[x]. Then
the principal divisor of z in Div(K(x)) appears thus :


r 
s
(z) = ni Pi − mj Qj + (deg g(x) − deg f (x))P∞ , (1.20)
i=1 j=1

where Pi resp. Qj are the places corresponding to pi (x) resp. qj (x), cf. Section
1.2. Therefore in arbitrary function fields, principal divisors can be considered
as a substitute for the decomposition into irreducible polynomials that occurs
in the rational function field.

Again we consider an arbitrary algebraic function field F/K. In Proposition

1.4.9 we have seen that the inequality

(A) ≤ 1 + deg A (1.21)

holds for all divisors A ≥ 0. In fact (1.21) holds for every divisor of degree
≥ 0. In order to verify this, we can assume that (A) > 0. Then A ∼ A for
some A ≥ 0 by Remark 1.4.5, so (A) = (A ) ≤ 1 + deg A = 1 + deg A by
Corollary 1.4.12.

Next we want to prove the existence of a lower bound for (A), similar to
the inequality in (1.21).

Proposition 1.4.14. There is a constant γ ∈ ZZ such that for all divisors

A ∈ Div(F ) the following holds:

deg A − (A) ≤ γ .

The emphasis here lies on the fact that γ is independent of the divisor A;
it depends only on the function field F/K.

Proof. To begin with, observe that

A1 ≤ A2 ⇒ deg A1 − (A1 ) ≤ deg A2 − (A2 ) , (1.22)

22 1 Foundations of the Theory of Algebraic Function Fields

by Lemma 1.4.8. We fix an element x ∈ F \K and consider the specific divisor

B := (x)∞ . As in the proof of Theorem 1.4.11 there exists a divisor C ≥ 0
(depending on x) such that (lB + C) ≥ (l + 1) · deg B for all l ≥ 0, see (1.18).
On the other hand, (lB + C) ≤ (lB) + deg C by Lemma 1.4.8. Combining
these inequalities we find

(lB) ≥ (l + 1)deg B − deg C = deg(lB) + ([F : K(x)] − deg C) .

Therefore
deg(lB) − (lB) ≤ γ for all l > 0 (1.23)
with some γ ∈ ZZ. We want to show that (1.23) holds even when we substitute
for lB any A ∈ Div(F ) (with the above γ).
Claim. Given a divisor A, there exist divisors A1 , D and an integer l ≥ 0
such that A ≤ A1 , A1 ∼ D and D ≤ lB.
Using this claim, Proposition 1.4.14 will follow easily :

deg A − (A) ≤ deg A1 − (A1 ) (by (1.22))

= deg D − (D) (by Corollary 1.4.12)
≤ deg(lB) − (lB) (by (1.22))
≤ γ. (by (1.23))

Proof of the Claim. Choose A1 ≥ A such that A1 ≥ 0. Then

(lB − A1 ) ≥ (lB) − deg A1 (by Lemma 1.4.8)

≥ deg(lB) − γ − deg A1 (by (1.23))
>0

for sufficiently large l. Thus there is some element 0 = z ∈ L (lB − A1 ).

Setting D := A1 − (z) we obtain A1 ∼ D and D ≤ A1 − (A1 − lB) = lB as
desired.

Definition 1.4.15. The genus g of F/K is defined by

g := max{ deg A − (A) + 1 | A ∈ Div(F ) } .

Observe that this definition makes sense by Proposition 1.4.14. It will turn
out that the genus is the most important invariant of a function field.

Corollary 1.4.16. The genus of F/K is a non-negative integer.

Proof. In the definition of g, put A = 0. Then deg(0) − (0) + 1 = 0, hence

g ≥ 0.
 1.4 Divisors 23

Theorem 1.4.17 (Riemann’s Theorem). Let F/K be a function field of

genus g. Then we have:
(a) For all divisors A ∈ Div(F ),

(A) ≥ deg A + 1 − g .

(b) There is an integer c, depending only on the function field F/K, such that

(A) = deg A + 1 − g ,

whenever deg A ≥ c.

Proof. (a) This is just the definition of the genus.

(b) Choose a divisor A0 with g = deg A0 − (A0 ) + 1 and set c := deg A0 + g.
If deg A ≥ c then

(A − A0 ) ≥ deg(A − A0 ) + 1 − g ≥ c − deg A0 + 1 − g = 1 .

So there is an element 0 = z ∈ L (A − A0 ). Consider the divisor A := A + (z)

which is ≥ A0 . We have

deg A − (A) = deg A − (A ) (by Corollary 1.4.12)

≥ deg A0 − (A0 ) (by Lemma 1.4.8)
= g − 1.

Hence (A) ≤ deg A + 1 − g.

Example 1.4.18. We want to show that the rational function field K(x)/K
has genus g = 0. In order to prove this, let P∞ denote the pole divisor of
x (notation as in Proposition 1.2.1). Consider for r ≥ 0 the vector space
L (rP∞ ). Obviously the elements 1, x, . . . , xr are in L (rP∞ ), hence

r + 1 ≤ (rP∞ ) = deg(rP∞ ) + 1 − g = r + 1 − g

for sufficiently large r. Thus g ≤ 0. Since g ≥ 0 holds for every function field,
the assertion follows.

In general it is hard to determine the genus of a function field. Large parts of

Chapter 3 will be devoted to this problem.
24 1 Foundations of the Theory of Algebraic Function Fields

1.5 The Riemann-Roch Theorem

In this section F/K denotes an algebraic function field of genus g.

Definition 1.5.1. For A ∈ Div(F ) the integer

i(A) := (A) − deg A + g − 1

is called the index of specialty of A.

Riemann’s Theorem 1.4.17 states that i(A) is a non-negative integer, and

i(A) = 0 if deg A is sufficiently large. In the present section we will provide
several interpretations for i(A) as the dimension of certain vector spaces. To
this end we introduce the notion of an adele.

Definition 1.5.2. An adele of F/K is a mapping


IPF −→ F ,
α:
P −→ αP ,
such that αP ∈ OP for almost all P ∈ IPF . We regard an adele as an element
of the direct product P ∈IPF F and therefore use the notation α = (αP )P ∈IPF
or, even shorter, α = (αP ). The set

AF := { α | α is an adele of F/K }

is called the adele space of F/K. It is regarded as a vector space over K

in the obvious manner (actually AF can be regarded as a ring, but the ring
structure will never be used).
The principal adele of an element x ∈ F is the adele all of whose compo-
nents are equal to x (note that this definition makes sense since x has only
finitely many poles). This gives an embedding F → AF . The valuations vP
of F/K extend naturally to AF by setting vP (α) := vP (αP ) (where αP is the
P -component of the adele α). By definition we have that vP (α) ≥ 0 for almost
all P ∈ IPF .

We note that the notion of an adele is not consistent in the literature. Some
authors use the name repartition for what we call an adele. Others mean by
an adele (or a repartition) a mapping α such that α(P ) is an element of the
P -adic completion F̂P for all P ∈ IPF (cf. Chapter 4).

Definition 1.5.3. For A ∈ Div(F ) we define

AF (A) := { α ∈ AF | vP (α) ≥ −vP (A) for all P ∈ IPF } .

Obviously this is a K-subspace of AF .

 1.5 The Riemann-Roch Theorem 25

Theorem 1.5.4. For every divisor A the index of specialty is

i(A) = dim(AF /(AF (A) + F )) .

Here, as usual, dim means the dimension as a K-vector space. Note that
although the vector spaces AF , AF (A) and F are infinite-dimensional, the
theorem states that the quotient space AF /(AF (A) + F ) has finite dimension
over K. As a corollary, we obtain another characterization of the genus of
F/K.

Corollary 1.5.5. g = dim(AF /(AF (0) + F )) .

Proof of Corollary 1.5.5. i(0) = (0) − deg(0) + g − 1 = 1 − 0 + g − 1 = g.

Proof of Theorem 1.5.4. We proceed in several steps.
Step 1. Let A1 , A2 ∈ Div(F ) and A1 ≤ A2 . Then AF (A1 ) ⊆ AF (A2 ) and

dim(AF (A2 )/AF (A1 )) = deg A2 − deg A1 . (1.24)

Proof of Step 1. AF (A1 ) ⊆ AF (A2 ) is trivial. It is sufficient to prove (1.24) in

the case A2 = A1 + P with P ∈ IPF (the general case follows by induction).
Choose t ∈ F with vP (t) = vP (A1 ) + 1 and consider the K-linear map

AF (A2 ) −→ FP ,
ϕ:
α −→ (tαP )(P ) .

One checks easily that ϕ is surjective and that the kernel of ϕ is AF (A1 ).
Consequently

deg A2 − deg A1 = deg P = [FP : K] = dim(AF (A2 )/AF (A1 )) .

Step 2. Let A1 , A2 ∈ Div(F ) and A1 ≤ A2 as before. Then

dim((AF (A2 ) + F )/(AF (A1 ) + F )

= (deg A2 − (A2 )) − (deg A1 − (A1 )) . (1.25)

Proof of Step 2. We have an exact sequence of linear mappings

σ
0 −→ L (A2 )/L (A1 ) −→
1
AF (A2 )/AF (A1 )
σ
−→
2
(AF (A2 ) + F )/(AF (A1 ) + F ) −→ 0 (1.26)

where σ1 and σ2 are defined in the obvious manner. In fact, the only non-trivial
assertion is that the kernel of σ2 is contained in the image of σ1 . In order to
prove this, let α ∈ AF (A2 ) with σ2 (α+AF (A1 )) = 0. Then α ∈ AF (A1 )+F , so
there is some x ∈ F with α − x ∈ AF (A1 ). As AF (A1 ) ⊆ AF (A2 ) we conclude
that x ∈ AF (A2 ) ∩ F = L (A2 ). Therefore α + AF (A1 ) = x + AF (A1 ) =
σ1 (x + L (A1 )) lies in the image of σ1 .
26 1 Foundations of the Theory of Algebraic Function Fields

From the exactness of (1.26) we obtain

dim(AF (A2 ) + F )/(AF (A1 ) + F )

= dim(AF (A2 )/AF (A1 )) − dim(L (A2 )/L (A1 ))
= (deg A2 − deg A1 ) − ((A2 ) − (A1 )) ,
using (1.24).
Step 3. If B is a divisor with (B) = deg B + 1 − g, then
AF = AF (B) + F . (1.27)
Proof of Step 3. To begin with, observe that for B1 ≥ B we have (by Lemma
1.4.8)

(B1 ) ≤ deg B1 + (B) − deg B = deg B1 + 1 − g .

On the other hand, (B1 ) ≥ deg B1 + 1 − g by Riemann’s Theorem. Therefore
(B1 ) = deg B1 + 1 − g for each B1 ≥ B . (1.28)
Now we prove (1.27). Let α ∈ AF . Obviously one can find a divisor B1 ≥ B
such that α ∈ AF (B1 ). By (1.25) and (1.28),
dim(AF (B1 ) + F )/(AF (B) + F ) = (deg B1 − (B1 )) − (deg B − (B))
= (g − 1) − (g − 1) = 0 .
This implies AF (B) + F = AF (B1 ) + F . Since α ∈ AF (B1 ) it follows that
α ∈ AF (B) + F , and (1.27) is proved.
End of the proof of Theorem 1.5.4. Now we consider an arbitrary divisor A.
By Riemann’s Theorem 1.4.17(b) there exists some divisor A1 ≥ A such that
(A1 ) = deg A1 + 1 − g. By (1.27), AF = AF (A1 ) + F , and in view of (1.25)
we obtain

dim(AF /(AF (A) + F )) = dim(AF (A1 ) + F )/(AF (A) + F )

= (deg A1 − (A1 )) − (deg A − (A))
= (g − 1) + (A) − deg A = i(A) .

Theorem 1.5.4 can be restated as follows: for all A ∈ Div(F ) holds

(A) = deg A + 1 − g + dim(AF /(AF (A) + F ) . (1.29)

This is a preliminary version of the Riemann-Roch Theorem which we shall
prove later in this section.
Next we introduce the concept of Weil differentials which will lead to a
second interpretation for the index of specialty of a divisor.
 1.5 The Riemann-Roch Theorem 27

Definition 1.5.6. A Weil differential of F/K is a K-linear map ω : AF → K

vanishing on AF (A) + F for some divisor A ∈ Div(F ). We call

ΩF := { ω | ω is a Weil differential of F/K }

the module of Weil differentials of F/K. For A ∈ Div(F ) let

ΩF (A) := { ω ∈ ΩF | ω vanishes on AF (A) + F } .

We regard ΩF as a K-vector space in the obvious manner (in fact, if ω1

vanishes on AF (A1 )+F and ω2 vanishes on AF (A2 )+F then ω1 +ω2 vanishes
on AF (A3 ) + F for every divisor A3 with A3 ≤ A1 and A3 ≤ A2 , and aω1
vanishes on AF (A1 ) + F for a ∈ K). Clearly ΩF (A) is a subspace of ΩF .

Lemma 1.5.7. For A ∈ Div(F ) we have dim ΩF (A) = i(A).

Proof. ΩF (A) is in a natural way isomorphic to the space of linear forms on

AF /(AF (A) + F ). Since AF /(AF (A) + F ) is finite-dimensional of dimension
i(A) by Theorem 1.5.4, our lemma follows immediately.

A simple consequence of Lemma 1.5.7 is that ΩF = 0. To see this, choose

a divisor A of degree ≤ −2. Then

dim ΩF (A) = i(A) = (A) − deg A + g − 1 ≥ 1 ,

hence ΩF (A) = 0.

Definition 1.5.8. For x ∈ F and ω ∈ ΩF we define xω : AF → K by

(xω)(α) := ω(xα) .

It is easily checked that xω is again a Weil differential of F/K. In fact, if

ω vanishes on AF (A) + F then xω vanishes on AF (A + (x)) + F . Clearly our
definition gives ΩF the structure of a vector space over F .

Proposition 1.5.9. ΩF is a one-dimensional vector space over F.

Proof. Choose 0 = ω1 ∈ ΩF (we already know that ΩF = 0). It has to be

shown that for every ω2 ∈ ΩF there is some z ∈ F with ω2 = zω1 . We can
assume that ω2 = 0. Choose A1 , A2 ∈ Div(F ) such that ω1 ∈ ΩF (A1 ) and
ω2 ∈ ΩF (A2 ). For a divisor B (which will be specified later) we consider the
K-linear injective maps

L (Ai + B) −→ ΩF (−B) ,
ϕi : (i = 1, 2)
x −→ xωi .
28 1 Foundations of the Theory of Algebraic Function Fields

Claim. For an appropriate choice of the divisor B holds

ϕ1 (L (A1 + B)) ∩ ϕ2 (L (A2 + B)) = {0} .

Using this claim, the proof of the proposition can be finished very quickly:
we choose x1 ∈ L (A1 + B) and x2 ∈ L (A2 + B) such that x1 ω1 = x2 ω2 = 0.
Then ω2 = (x1 x−1 2 )ω1 as desired.
Proof of the Claim. We start with a simple and well-known fact from linear
algebra: if U1 , U2 are subspaces of a finite-dimensional vector space V then

dim(U1 ∩ U2 ) ≥ dim U1 + dim U2 − dim V . (1.30)

Now let B > 0 be a divisor of sufficiently large degree such that

(Ai + B) = deg(Ai + B) + 1 − g

for i = 1, 2 (this is possible by Riemann’s Theorem 1.4.17). We set Ui :=

ϕi (L (Ai + B)) ⊆ ΩF (−B). Since

dim ΩF (−B) = i(−B) = dim(−B) − deg(−B) + g − 1

= deg B − 1 + g ,
we obtain
dim U1 + dim U2 − dim ΩF (−B)
= deg(A1 + B) + 1 − g + deg(A2 + B) + 1 − g − (deg B + g − 1)
= deg B + (deg A1 + deg A2 + 3(1 − g)) .
The term in brackets is independent of B, so

dim U1 + dim U2 − dim ΩF (−B) > 0

if deg B is sufficiently large. By (1.30) it follows that U1 ∩ U2 = {0} which

proves our claim.

We want to attach a divisor to each Weil differential ω = 0. To this end

we consider (for a fixed ω) the set of divisors

M (ω) := { A ∈ Div(F ) | ω vanishes on AF (A) + F } . (1.31)

Lemma 1.5.10. Let 0 = ω ∈ ΩF . Then there is a uniquely determined divisor

W ∈ M (ω) such that A ≤ W for all A ∈ M (ω).

Proof. By Riemann’s Theorem there exists a constant c, depending only on

the function field F/K, with the property i(A) = 0 for all A ∈ Div(F ) of
degree ≥ c . Since dim(AF /(AF (A) + F )) = i(A) by Theorem 1.5.4, we have
that deg A < c for all A ∈ M (ω). So we can choose a divisor W ∈ M (ω) of
maximal degree.
 1.5 The Riemann-Roch Theorem 29

Suppose W does not have the property of our lemma. Then there exists a
divisor A0 ∈ M (ω) with A0 ≤ W , i.e. vQ (A0 ) > vQ (W ) for some Q ∈ IPF . We
claim that
W + Q ∈ M (ω) , (1.32)
which is a contradiction to the maximality of W . In fact, consider an adele
α = (αP ) ∈ AF (W + Q). We can write α = α + α with
 
 αP f or P = Q,  0 f or P = Q,
αP := and αP :=
0 f or P = Q, αQ f or P = Q.
Then α ∈ AF (W ) and α ∈ AF (A0 ), therefore ω(α) = ω(α ) + ω(α ) = 0.
Hence ω vanishes on AF (W + Q) + F , and (1.32) is proved. The uniqueness
of W is now obvious.

The following definition makes sense by the preceding lemma.

Definition 1.5.11. (a) The divisor (ω) of a Weil differential ω = 0 is the

uniquely determined divisor of F/K satisfying
(1) ω vanishes on AF ((ω)) + F , and
(2) if ω vanishes on AF (A) + F then A ≤ (ω).
(b) For 0 = ω ∈ ΩF and P ∈ IPF we define vP (ω) := vP ((ω)).
(c) A place P is said to be a zero (resp. pole) of ω if vP (ω) > 0 (resp.
vP (ω) < 0). The Weil differential ω is called regular at P if vP (ω) ≥ 0, and
ω is said to be regular (or holomorphic) if it is regular at all places P ∈ IPF .
(d) A divisor W is called a canonical divisor of F/K if W = (ω) for some
ω ∈ ΩF .

Remark 1.5.12. It follows immediately from the definitions that

ΩF (A) = { ω ∈ ΩF | ω = 0 or (ω) ≥ A }

and
ΩF (0) = { ω ∈ ΩF | ω is regular } .
As a consequence of Lemma 1.5.7 and Definition 1.5.1 we obtain

dim ΩF (0) = g .

Proposition 1.5.13. (a) For 0 = x ∈ F and 0 = ω ∈ ΩF we have (xω) =

(x) + (ω).
(b) Any two canonical divisors of F/K are equivalent.
30 1 Foundations of the Theory of Algebraic Function Fields

It follows from this proposition that the canonical divisors of F/K form
a whole class [W ] in the divisor class group Cl(F ); this divisor class is called
the canonical class of F/K.
Proof of Proposition 1.5.13. If ω vanishes on AF (A) + F then xω vanishes on
AF (A + (x)) + F , consequently

(ω) + (x) ≤ (xω) .

Likewise (xω) + (x−1 ) ≤ (x−1 xω) = (ω). Combining these inequalities we

obtain
(ω) + (x) ≤ (xω) ≤ −(x−1 ) + (ω) = (ω) + (x) .
This proves (a). Item (b) follows from (a) and Proposition 1.5.9.

Theorem 1.5.14 (Duality Theorem). Let A be an arbitrary divisor and

W = (ω) be a canonical divisor of F/K. Then the mapping

L (W − A) −→ ΩF (A) ,
μ:
x −→ xω

is an isomorphism of K-vector spaces. In particular,

i(A) = (W − A) .

Proof. For x ∈ L (W − A) we have

(xω) = (x) + (ω) ≥ −(W − A) + W = A ,

hence xω ∈ ΩF (A) by Remark 1.5.12. Therefore μ maps L (W − A) into

ΩF (A). Clearly μ is linear and injective. In order to show that μ is surjective,
we consider a Weil differential ω1 ∈ ΩF (A). By Proposition 1.5.9 we can write
ω1 = xω with some x ∈ F . Since

(x) + W = (x) + (ω) = (xω) = (ω1 ) ≥ A ,

we obtain (x) ≥ −(W − A), so x ∈ L (W − A) and ω1 = μ(x). We have

thus proved that dim ΩF (A) = (W − A). Since dim ΩF (A) = i(A) by Lemma
1.5.7, this implies i(A) = (W − A).

Summing up the results of this section we obtain the Riemann-Roch

Theorem; it is by far the most important theorem in the theory of algebraic
function fields.

Theorem 1.5.15 (Riemann-Roch Theorem). Let W be a canonical divi-

sor of F/K. Then for each divisor A ∈ Div(F ),

(A) = deg A + 1 − g + (W − A) .

 1.6 Some Consequences of the Riemann-Roch Theorem 31

Proof. This is an immediate consequence of Theorem 1.5.14 and the definition

of i(A).
Corollary 1.5.16. For a canonical divisor W we have

deg W = 2g − 2 and (W ) = g .

Proof. For A = 0, the Riemann-Roch Theorem and Lemma 1.4.7 give

1 = (0) = deg 0 + 1 − g + (W − 0) .

Thus (W ) = g. Setting A = W we obtain

g = (W ) = deg W + 1 − g + (W − W ) = deg W + 2 − g .

Therefore deg W = 2g − 2.

From Riemann’s Theorem we already know that there is some constant

c such that i(A) = 0 whenever deg A ≥ c. We can now give a more precise
description of how to choose this constant.
Theorem 1.5.17. If A is a divisor of F/K of degree degA ≥ 2g − 1 then

(A) = deg A + 1 − g .

Proof. We have (A) = deg A+1−g+(W −A), where W is a canonical divisor.

Since deg A ≥ 2g − 1 and deg W = 2g − 2, we conclude that deg(W − A) < 0.
It follows from Corollary 1.4.12 that (W − A) = 0.

Observe that the bound 2g − 1 in this theorem is the best possible, since
for a canonical divisor W

(W ) > deg W + 1 − g

by Corollary 1.5.16.

1.6 Some Consequences of the Riemann-Roch Theorem

As before, F/K denotes an algebraic function field of genus g. We want to
discuss various consequences of the Riemann-Roch Theorem. Our first aim is
to show that the Riemann-Roch Theorem characterizes the genus as well as
the canonical class of F/K.
Proposition 1.6.1. Suppose that g0 ∈ ZZ and W0 ∈ Div(F ) satisfy

(A) = deg A + 1 − g0 + (W0 − A) (1.33)

for all A ∈ Div(F ). Then g0 = g, and W0 is a canonical divisor.

32 1 Foundations of the Theory of Algebraic Function Fields

Proof. Setting A = 0 resp. A = W0 in (1.33) we obtain (W0 ) = g0 and

deg W0 = 2g0 − 2 (cf. the proof of Corollary 1.5.16). Let W be a canonical
divisor of F/K. We choose a divisor A with deg A > max{2g − 2, 2g0 − 2}.
Then (A) = deg A + 1 − g by Theorem 1.5.17 and (A) = deg A + 1 − g0 by
(1.33). Therefore g = g0 . Finally we substitute A = W in (1.33). This yields

g = (2g − 2) + 1 − g + (W0 − W ) ,
hence (W0 − W ) = 1. Since deg(W0 − W ) = 0, this implies that W0 − W is
principal (cf. Corollary 1.4.12), so W0 ∼ W .

Another useful characterization of canonical divisors is the following.

Proposition 1.6.2. A divisor B is canonical if and only if deg B = 2g − 2
and (B) ≥ g.
Proof. Suppose that deg B = 2g − 2 and (B) ≥ g. Choose a canonical divi-
sor W . Then
g ≤ (B) = deg B + 1 − g + (W − B) = g − 1 + (W − B) ,
therefore (W − B) ≥ 1. Since deg(W − B) = 0, it follows from Corollary
1.4.12 that W ∼ B.

Next we come to a characterization of the rational function field.

Proposition 1.6.3. For a function field F/K the following conditions are
equivalent:
(1) F/K is rational; i.e., F = K(x) for some x which is transcendental over
the field K.
(2) F/K has genus 0, and there is some divisor A ∈ Div(F ) with deg A = 1.
Proof. (1) ⇒ (2): See Example 1.4.18.
(2) ⇒ (1): Let g = 0 and deg A = 1. As deg A ≥ 2g − 1 we have that
(A) = deg A + 1 − g = 2 by Theorem 1.5.17. Thus A ∼ A for some positive
divisor A (see Remark 1.4.5 (b)). Since (A ) = 2, there exists an element
x ∈ L (A ) \ K, so (x) = 0 and (x) + A ≥ 0. As A ≥ 0 and deg A = 1, this
is possible only if A = (x)∞ , the pole divisor of x. Now
[F : K(x)] = deg (x)∞ = deg A = 1
by Theorem 1.4.11, so F = K(x).
Remark 1.6.4. There exist non-rational function fields of genus 0 (these can-
not have a divisor of degree 1 by Proposition 1.6.3). However, if K is an
algebraically closed field or a finite field, there exists always a divisor of de-
gree 1 (for an algebraically closed field this is trivial, for a finite constant field
we shall prove it in Chapter 5), hence in these cases we have g = 0 if and only
if F/K is rational.
 1.6 Some Consequences of the Riemann-Roch Theorem 33

It would seem appropriate at this point to give some examples of function

fields of genus > 0. However, we defer such examples to Chapter 6 at which
point we will have better methods at hand for calculating the genus.
Our next application of Section 1.5 is a strengthening of the Weak Ap-
proximation Theorem.
Theorem 1.6.5 (Strong Approximation Theorem). Let S  IPF be
a proper subset of IPF and P1 , . . . , Pr ∈ S. Suppose there are given elements
x1 , . . . , xr ∈ F and integers n1 , . . . , nr ∈ ZZ. Then there exists an element
x ∈ F such that
vPi (x − xi ) = ni (i = 1, . . . , r) , and
vP (x) ≥ 0 f or all P ∈ S \ {P1 , . . . , Pr } .

Proof. Consider the adele α = (αP )P ∈IPF with


xi for P = Pi , i = 1, . . . , r ,
αP :=
0 otherwise .

Choose a place Q ∈ IPF \ S. For sufficiently large m ∈ IN we have

 
 r
AF = AF mQ − (ni + 1)Pi + F
i=1

by Theorem 1.5.4 and Theorem 1.5.17 (observe the definition of the index of
r in Definition 1.5.1). So there is an element z ∈ F with
specialty which is given
z − α ∈ AF (mQ − i=1 (ni + 1)Pi ). This means
vPi (z − xi ) > ni for i = 1, . . . , r , and (1.34)
vP (z) ≥ 0 for P ∈ S \ {P1 , . . . , Pr } . (1.35)
Now we choose y1 , . . . , yr ∈ F with vPi (yi ) = ni . In the same manner as above
we construct y ∈ F with
vPi (y − yi ) > ni for i = 1, . . . , r , and (1.36)
vP (y) ≥ 0 for P ∈ S \ {P1 , . . . , Pr } . (1.37)
Then we have for i = 1, . . . , r,
vPi (y) = vPi ((y − yi ) + yi ) = ni (1.38)
by (1.36) and the Strict Triangle Inequality. Putting x := y + z we obtain
vPi (x − xi ) = vPi (y + (z − xi )) = ni (i = 1, . . . , r)
by (1.38). For P ∈ S \ {P1 , . . . , Pr }, vP (x) = vP (y + z) ≥ 0 holds by (1.35)
and (1.37).
 Weierstrass points
34 1 Foundations of the Theory of Algebraic Function Fields

Next we investigate elements in F which have only one pole.

Proposition 1.6.6. Let P ∈ IPF . Then for each n ≥ 2g there exists an ele-
ment x ∈ F with pole divisor (x)∞ = nP .
Proof. By Theorem 1.5.17 we know that ((n − 1)P ) = (n − 1)deg P + 1 − g
and (nP ) = n · deg P + 1 − g, hence L ((n − 1)P )  L (nP ). Every element
x ∈ L (nP ) \ L ((n − 1)P ) has pole divisor nP .
Definition 1.6.7. Let P ∈ IPF . An integer n ≥ 0 is called a pole number of P
if there is an element x ∈ F with (x)∞ = nP . Otherwise n is called a gap
number of P .

Clearly n is a pole number of P if and only if (nP ) > ((n − 1)P ). Moreover,
the set of pole numbers of P is a sub-semigroup of the additive semigroup IN
(to see this note that, if (x1 )∞ = n1 P and (x2 )∞ = n2 P then x1 x2 has the
pole divisor (x1 x2 )∞ = (n1 + n2 )P ).
Theorem 1.6.8 (Weierstrass Gap Theorem). Suppose that F/K has
genus g > 0 and P is a place of degree one. Then there are exactly g gap
numbers i1 < . . . < ig of P . We have
i1 = 1 and ig ≤ 2g − 1 .
Proof. Each gap number of P is ≤ 2g − 1 by Proposition 1.6.6, and 0 is a pole
number. We have the following obvious characterization of gap numbers:
i is a gap number of P ⇐⇒ L ((i − 1)P ) = L (iP ) .
Consider now the sequence of vector spaces
K = L (0) ⊆ L (P ) ⊆ L (2P ) ⊆ · · · ⊆ L ((2g − 1)P ) , (1.39)
where dim L (0) = 1 and dim L ((2g − 1)P ) = g by Theorem 1.5.17. Observe
that
dim L (iP ) ≤ dim L ((i − 1)P ) + 1
for all i, see Lemma 1.4.8. So we have in (1.39) exactly g − 1 numbers 1 ≤ i ≤
2g − 1 with L ((i − 1)P )  L (iP ). The remaining g numbers are gaps of P .
Finally we must show that 1 is a gap. Suppose the converse, so 1 is a
pole number of P . Since the pole numbers form an additive semigroup, every
n ∈ IN is a pole number, and there are no gaps at all. This is a contradiction
because g > 0.
Remark 1.6.9. Suppose that K is algebraically closed. Then one can show
that almost all places of F/K have the same sequence of gap numbers (which
are therefore called the gap numbers of the function field F/K). Such places
of F/K are said to be ordinary places. Every non-ordinary place is called a
Weierstrass point of F/K. If the genus of F/K is ≥ 2, there exists at least
one Weierstrass point, see [21] or [45].
 1.6 Some Consequences of the Riemann-Roch Theorem 35

For a divisor A of degree < 0 we have L (A) = {0} by Corollary 1.4.12. On

the other hand, if deg A > 2g −2 then (A) = deg A+1−g by Theorem 1.5.17.
So the dimension of A depends only on deg A (and the genus) in these cases.
We want to consider more closely the case where 0 ≤ deg A ≤ 2g − 2; here the
situation is rather complicated, but there are still some general results.

Definition 1.6.10. A divisor A ∈ Div(F ) is called non-special if i(A) = 0;

otherwise A is called special.

We note some immediate consequences of this definition.

Remark 1.6.11. (a) A is non-special ⇐⇒ (A) = deg A + 1 − g .

(b) deg A > 2g − 2 ⇒ A is non-special.
(c) The property of a divisor A being special or non-special depends only on
the class [A] of A in the divisor class group.
(d) Canonical divisors are special.
(e) Every divisor A with (A) > 0 and deg A < g is special.
(f ) If A is non-special and B ≥ A then B is non-special.

Proof. (a) is clear from the definition of i(A), (b) is just Theorem 1.5.17, and
(c) follows from the fact that (A) and deg A depend only on the divisor class
of A.
(d) For a canonical divisor W we have i(W ) = (W − W ) = 1 by Theorem
1.5.14, hence W is special.
(e) 1 ≤ (A) = deg A + 1 − g + i(A) ⇒ i(A) ≥ g − deg A > 0 since deg A < g.
Thus A is special.
(f) A is non-special if and only if AF = AF (A) + F , see Theorem 1.5.4. If
B ≥ A then AF (A) ⊆ AF (B), so (f) follows.

With regard to item (e) of the preceding remark, the following result is
interesting.

Proposition 1.6.12. Suppose that T ⊆ IPF is a set of places of degree one

such that |T | ≥ g. Then there exists a non-special divisor B ≥ 0 with deg B =
g and supp B ⊆ T .

Proof. The crucial step of the proof is the following claim:

Claim. Given g distinct places P1 , . . . , Pg ∈ T and a divisor A ≥ 0 with
(A) = 1 and deg A ≤ g − 1, there is an index j ∈ {1, . . . , g} such that
(A + Pj ) = 1 .
Suppose the claim is false; then (A + Pj ) > 1, and there are elements zj ∈
L (A + Pj )\L (A) for j = 1, . . . , g. Since

vPj (zj ) = −vPj (A) − 1 and vPi (zj ) ≥ −vPi (A) for i = j ,
36 1 Foundations of the Theory of Algebraic Function Fields

the Strict Triangle Inequality implies that the g + 1 elements 1, z1 , . . . , zg are

linearly independent over K. Choose a divisor D ≥ A + P1 + . . . + Pg with
deg D = 2g − 1. Then 1, z1 , . . . , zg ∈ L (D), hence (D) ≥ g + 1. On the
other hand, (D) = deg D + 1 − g = g by the Riemann-Roch Theorem. This
contradiction proves our claim.
Now the proof of Proposition 1.6.12 is very simple. By the above claim we
find divisors 0 < Pi1 < Pi1 + Pi2 < . . . < Pi1 + Pi2 + . . . + Pig =: B (with
iν ∈ {1, . . . , g}, not necessarily distinct) such that (Pi1 + . . . + Pij ) = 1 for
j = 1, . . . , g. In particular (B) = 1. The divisor B is non-special because

deg B + 1 − g = g + 1 − g = 1 = (B) .
(cf. Remark 1.6.11(a).)

We conclude this section with an inequality for the dimension of an arbi-

trary divisor of degree ≤ 2g − 2.

Theorem 1.6.13 (Clifford’s Theorem). For all divisors A with 0 ≤

deg A ≤ 2g − 2 holds
1
(A) ≤ 1 + · deg A .
2

The main step in the proof of Clifford’s Theorem is the following result.

Lemma 1.6.14. Suppose that A and B are divisors such that (A) > 0 and
(B) > 0. Then
(A) + (B) ≤ 1 + (A + B) .

Proof of Lemma 1.6.14. Since (A) > 0 and (B) > 0 we can find A0 , B0 ≥ 0
with A ∼ A0 and B ∼ B0 (cf. Remark 1.4.5). The set

X := {D ∈ Div(F ) | D ≤ A0 and L (D) = L (A0 )}

is non-empty because A0 ∈ X. As deg D ≥ 0 for all D ∈ X, there is some

divisor D0 ∈ X of minimal degree. It follows that

(D0 − P ) < (D0 ) for all P ∈ IPF . (1.40)

We want to show that

(D0 ) + (B0 ) ≤ 1 + (D0 + B0 ) . (1.41)

From (1.41) the lemma will follow immediately:

(A) + (B) = (A0 ) + (B0 ) = (D0 ) + (B0 )

≤ 1 + (D0 + B0 ) ≤ 1 + (A0 + B0 ) = 1 + (A + B) .
 1.7 Local Components of Weil Differentials 37

In order to prove (1.41) we make the additional assumption that K is

an infinite field (in fact, we shall show later that Lemma 1.6.14 also holds
in the case of a finite constant field, see Theorem 3.6.3(d)). Let supp B0 =
{P1 , . . . , Pr }. Then L (D0 −Pi ) is a proper subspace of L (D0 ) for i = 1, . . . , r,
and since a vector space over an infinite field is not the union of finitely many
proper subspaces, we can find an element

r
z ∈ L (D0 ) \ L (D0 − Pi ) . (1.42)
i=1

Consider the K-linear map


L (B0 ) −→ L (D0 + B0 )/L (A0 ) ,
ϕ:
x −→ xz mod L (A0 ) .

From (1.42) follows easily that the kernel of ϕ is K, hence

dim L (B0 ) − 1 ≤ dim L (D0 + B0 ) − dim L (A0 ) ,

which proves (1.41).

Proof of Theorem 1.6.13. The case (A) = 0 is trivial. Likewise, if (W −A) =
0 (where W is canonical), then
1 1 1
(A) = deg A + 1 − g = 1 + deg A + (deg A − 2g) < 1 + deg A ,
2 2 2
since deg A ≤ 2g − 2. It remains to consider the case where (A) > 0 and
(W − A) > 0. We can apply Lemma 1.6.14 to obtain

(A) + (W − A) ≤ 1 + (W ) = 1 + g . (1.43)

On the other hand,

(A) − (W − A) = deg A + 1 − g (1.44)

by the Riemann-Roch Theorem. Adding (1.43) and (1.44) yields the desired
result.

1.7 Local Components of Weil Differentials

In Section 1.5 we considered the diagonal embedding F → AF which maps

x ∈ F to the corresponding principal adele. Now we introduce for each place
P ∈ IPF another local embedding ιP : F → AF .
38 1 Foundations of the Theory of Algebraic Function Fields

Definition 1.7.1. Let P ∈ IPF .

(a) For x ∈ F let ιP (x) ∈ AF be the adele whose P -component is x, and all
other components are 0.
(b) For a Weil differential ω ∈ ΩF we define its local component ωP : F → K
by
ωP (x) := ω(ιP (x)) .

Clearly ωP is a K-linear mapping.

Proposition 1.7.2. Let ω ∈ ΩF and α = (αP ) ∈ AF . Then ωP (αP ) = 0 for
at most finitely many places P , and

ω(α) = ωP (αP ) .
P ∈IPF

In particular 
ωP (1) = 0 . (1.45)
P ∈IPF

Proof. We can assume that ω = 0 and we set W := (ω), the divisor of ω (see
Definition 1.5.11). There is a finite set S ⊆ IPF such that
vP (W ) = 0 and vP (αP ) ≥ 0 for all P ∈ S .
Define β = (βP ) ∈ AF by

αP f or P ∈ S ,
βP :=
0 f or P ∈ S .

Then β ∈ AF (W ) and α = β + P ∈S ιP (αP ), hence ω(β) = 0 and

ω(α) = ωP (αP ) .
P ∈S

For P ∈ S, ιP (αP ) ∈ AF (W ) and therefore ωP (αP ) = 0.

We shall see in Chapter 4 that Equation (1.45) is nothing else but the Residue
Theorem for differentials of F/K.
Next we show that a Weil differential is uniquely determined by each of
its local components.
Proposition 1.7.3. (a) Let ω = 0 be a Weil differential of F/K and P ∈ IPF .
Then
vP (ω) = max{ r ∈ ZZ | ωP (x) = 0 for all x ∈ F with vP (x) ≥ −r } .
In particular ωP is not identically 0.
(b) If ω, ω  ∈ ΩF and ωP = ωP for some P ∈ IPF then ω = ω  .
 1.7 Local Components of Weil Differentials 39

Proof. (a) Recall that, by definition, vP (ω) = vP (W ) where W = (ω) denotes

the divisor of ω. Let s := vP (ω). For x ∈ F with vP (x) ≥ −s we have
ιP (x) ∈ AF (W ), hence ωP (x) = ω(ιP (x)) = 0. Suppose now that ωP (x) = 0
for all x ∈ F with vP (x) ≥ −s − 1. Let α = (αQ )Q∈IPF ∈ AF (W + P ). Then

α = (α − ιP (αP )) + ιP (αP )

with α − ιP (αP ) ∈ AF (W ) and vP (αP ) ≥ −s − 1, hence

ω(α) = ω(α − ιP (αP )) + ωP (αP ) = 0 .

Therefore ω vanishes on AF (W + P ), a contradiction to the definition of W .

(b) If ωP = ωP then (ω − ω  )P = 0, hence ω − ω  = 0 by (a).

Once again we consider the rational function field K(x). We use the no-
tation introduced in Section 1.2; i.e., P∞ denotes the pole divisor of x and
Pa denotes the zero divisor of x − a (for a ∈ K). The following result will be
important in Chapter 4.

Proposition 1.7.4. For the rational function field F = K(x) the following
hold:
(a) The divisor −2P∞ is canonical.
(b) There exists a unique Weil differential η ∈ ΩK(x) with (η) = −2P∞ and
ηP∞ (x−1 ) = −1.
(c) The local components ηP∞ resp. ηPa of the above Weil differential η satisfy


0 for n = −1 ,
ηP∞ ((x − a) ) =
n
−1 for n = −1 ,

0 for n = −1 ,
ηPa ((x − a) ) =
n
1 for n = −1 .

Proof. (a) deg(−2P∞ ) = −2 = 2g − 2 and (−2P∞ ) = 0 = g, hence −2P∞ is

canonical by Proposition 1.6.2.
(b) Choose a Weil differential ω with divisor (ω) = −2P∞ . Then ω vanishes on
the space AK(x) (−2P∞ ), but it does not vanish identically on AK(x) (−P∞ ).
Since
dim AK(x) (−P∞ )/AK(x) (−2P∞ ) = 1
(see Equation (1.24) in the proof of Theorem 1.5.4) and

ιP∞ (x−1 ) ∈ AK(x) (−P∞ ) \ AK(x) (−2P∞ ) ,

we conclude that
40 1 Foundations of the Theory of Algebraic Function Fields

ωP∞ (x−1 ) = ω(ιP∞ (x−1 )) =: c = 0 .

Setting η := −c−1 ω we obtain (η) = −2P∞ and ηP∞ (x−1 ) = −1. The unique-
ness of η is easily proved. If η ∗ has the same properties as η then η − η ∗
vanishes on the space AK(x) (−P∞ ), which implies η − η ∗ = 0.
(c) Since a Weil differential vanishes on principal adeles, we have by
Proposition 1.7.2 that

0 = η((x − a)n ) = ηP ((x − a)n ) . (1.46)
P ∈IPK(x)

For P = P∞ and P = Pa holds vP ((x − a)n ) = 0, therefore ηP ((x − a)n ) = 0

by Proposition 1.7.3, and then (1.46) yields

ηP∞ ((x − a)n ) + ηPa ((x − a)n ) = 0 . (1.47)

In the case n ≤ −2 we have vP∞ ((x − a)n ) ≥ 2, hence ηP∞ ((x − a)n ) = 0
by Proposition 1.7.3. Now (1.47) implies that ηPa ((x − a)n ) = 0 as well. If
n ≥ 0, ηPa ((x − a)n ) = 0 by Proposition 1.7.3, and we obtain the result for
ηP∞ ((x − a)n ) again by (1.47).
Eventually we consider the case n = −1. Since
 
1 a 1 a
= + and ιP∞ ∈ AK(x) (−2P∞ ) ,
x−a x(x − a) x x(x − a)

we see that ηP∞ ((x − a)−1 ) = ηP∞ (x−1 ) = −1 (by definition of η), and from
(1.47) follows that ηPa ((x − a)−1 ) = 1.

1.8 Exercises
1.1. Consider the rational function field K(x)/K and a non-constant element
z = f (x)/g(x) ∈ K(x) \ K, where f (x), g(x) ∈ K[x] are relatively prime. We
call deg(z) := max{deg f (x), deg g(x)} the degree of z.
(i) Show that [K(x) : K(z)] = deg(z), and write down the minimal polynomial
of x over K(z) (in order to avoid calculations, you may use Theorem 1.4.11
and Example 1.4.13).
(ii) Show that K(x) = K(z) if and only if z = (ax+b)/(cx+d) with a, b, c, d ∈
K and ad − bc = 0.
1.2. For a field extension L/M we denote by Aut(L/M ) the group of auto-
morphisms of L/M (i.e., automorphisms of L which are the identity on M ).
Let K(x)/K be the rational function field over K. Show:
(i) For every σ ∈ Aut(K(x)/K) there exist a, b, c, d ∈ K such that ad − bc = 0
and σ(x) = (ax + b)/(cx + d).
 1.8 Exercises 41

(ii) Given a, b, c, d ∈ K with ad − bc = 0, there is a unique automorphism

σ ∈ Aut(K(x)/K) with σ(x) = (ax + b)/(cx + d).
by GL2 (K) the group of invertible 2 × 2 - matrices over K. For
(iii) Denote

ac
A = ∈ GL2 (K) denote by σA the automorphism of K(x)/K with
bd
σA (x) = (ax + b)/(cx + d). Show that the map which sends A to σA , is
a homomorphism from GL2 (K)  onto
 Aut(K(x)/K). Its kernel is the set of
a0
diagonal matrices of the form with a ∈ K × , hence
0a

Aut(K(x)/K)  GL2 (K)/K × .

(This group GL2 (K)/K × is called the projective linear group and is denoted
by PGL2 (K).)
1.3. If L is a field and G is a group of automorphisms of L, we denote by

LG := {w ∈ L | σ(w) = w for all σ ∈ G}

the fixed field of G. It is well-known from algebra that if G is a finite group,

then L/LG is a finite extension of degree [L : LG ] = ord(G).
Now let G ⊆ Aut(K(x)/K) be a finite subgroup of the automorphism group
of the rational function field K(x) over K, and put
 
z := σ(x) , u := σ(x) .
σ∈G σ∈G

Show:
(i) Either z ∈ K, or K(z) = K(x)G .
(ii) Either u ∈ K, or K(u) = K(x)G .
(iii) Find examples of finite subgroups G ⊆ Aut(K(x)/K) for both alterna-
tives in (i) (and also in (ii)).
1.4. Let K(x) be the rational function field over K. Find bases of the following
Riemann-Roch spaces:

L (rP∞ ) , L (rPα ) , L (Pp(x) ) ,

where r ≥ 0, and the places P∞ , Pα and Pp(x) are as in Section 1.2.
1.5. (Representation of rational functions by partial fractions)
(i) Show that every element z ∈ K(x) can be written as

r  ki
cij (x)
z= + h(x) ,
p
i=1 j=1 i
(x)j

where
42 1 Foundations of the Theory of Algebraic Function Fields

(a) p1 (x), ..., pr (x) are distinct monic irreducible polynomials in K[x],
(b) k1 , ..., kr ≥ 1 ,
(c) cij (x) ∈ K[x] and deg(cij (x)) < deg(pi (x)),
(d) ciki (x) = 0 for 1 ≤ i ≤ r ,
(e) h(x) ∈ K[x] .
Note that the case r = 0 is allowed; it just means that z = h(x) ∈ K[x].
(ii) Show that the above representation of z is unique.

1.6. Let F = K(x) be the rational function field over K. Show directly that
AF = AF (0) + F (by Corollary 1.5.5 this provides another proof that the
rational function field has genus 0).

1.7. There are many analogies between algebraic function fields and algebraic
number fields (i.e., finite extensions of the field Q of rational numbers). Here
is a first example.
A valuation ring of a field L is a subring O  L such that for all z ∈ L one
has that z ∈ O or z −1 ∈ O.
(i) Show that every valuation ring is a local ring (i.e., it has a unique maximal
ideal).
(ii) Now we consider the field L = Q. Show that for every prime number
p ∈ ZZ, the set ZZ(p) := {a/b ∈ Q | a, b ∈ ZZ and p  b} is a valuation ring of Q.
What is the maximal ideal of ZZ(p) ?
(iii) Let O be a valuation ring of Q. Show that O = ZZ(p) for some prime
number p.

In the following exercises, F/K always denotes a function field of genus g

with full constant field K.

1.8. Assume that g > 0 and A is a divisor with (A) > 0. Show that (A) =
deg(A) + 1 if and only if A is a principal divisor.

1.9. Show that the following conditions are equivalent:

(a) g = 0.
(b) There is a divisor A with deg(A) = 2 and (A) = 3.
(c) There is a divisor A with deg(A) ≥ 1 and (A) > deg(A).
(d) There is a divisor A with deg(A) ≥ 1 and (A) = deg(A) + 1.
In case of charK = 2, also the following condition is equivalent to the above:
(e) There are elements x, y ∈ F such that F = K(x, y) and y 2 = ax2 + b, with
a, b ∈ K × .
 1.8 Exercises 43

1.10. Let IR(x) be the rational function field over the field of real numbers.
(i) Show that the polynomial f (T ) := T 2 + (x2 + 1) ∈ IR(x)[T ] is irreducible
over IR(x).
Let F := IR(x, y), where y 2 + x2 + 1 = 0. By (i), [F : IR(x)] = 2. Show:
(ii) IR is the full constant field of F , and F/IR has genus g = 0.
(iii) F/IR is not a rational function field.
(iv) All places of F/IR have degree 2.

1.11. Assume that char(K) = 2. Let F = K(x, y) with

y 2 = f (x) ∈ K[x] , deg f (x) = 2m + 1 ≥ 3 .

Show:
(i) K is the full constant field of F .
(ii) There is exactly one place P ∈ IPF which is a pole of x, and this place is
also the only pole of y.
(iii) For every r ≥ 0, the elements 1, x, x2 , . . . , xr , y, xy, . . . , xs y with 0 ≤ s <
r − m are in L (2rP ).
(iv) The genus of F/K satisfies g ≤ m.
Remark. We will prove later that the genus is in fact g = m, if the polynomial
f (x) does not have multiple factors, see Example 3.7.6.

1.12. Let K = IF3 be the field with 3 elements and K(x) the rational function
field over K. Show:
(i) The polynomial f (T ) = T 2 + x4 − x2 + 1 is irreducible over K(x).
(ii) Let F = K(x, y) where y is a zero of the polynomial f (T ) as above, and
let K̃ be the full constant field of F . Then K̃ has 9 elements, and F = K̃(x).

1.13. Assume that F/K has a place P ∈ IPF of degree one. Show that there
exist x, y ∈ F such that [F : K(x)] = [F : K(y)] = 2g + 1 and F = K(x, y).

1.14. Let V, W be vector spaces over K. A non-degenerate pairing of V and

W is a bilinear map s : V × W → K such that the following hold: For every
v ∈ V with v = 0 there is some w ∈ W with s(v, w) = 0, and for every w ∈ W
with w = 0 there is some v ∈ V with s(v, w) = 0.
Now we consider a function field F/K, a divisor A ∈ Div(F ) and a non-zero
Weil differential ω ∈ ΩF . Let W := (ω). Show that the map s : L (W − A) ×
AF /(AF (A) + F ) → K given by s(x, α) := ω(xα) is well-defined, and it is a
non-degenerate pairing.

1.15. Assume that the constant field K is algebraically closed. Show that for
every integer d ≥ g, there exists a divisor A ∈ Div(F ) with deg(A) = d and
(A) = deg(A) + 1 − g.
44 1 Foundations of the Theory of Algebraic Function Fields

1.16. Let i(A) denote the index of specialty of the divisor A ∈ Div(F ). Show:
(i) i(A) ≤ max{0, 2g − 1 − deg(A)}.
(ii) Assume that i(A) > 0. Show that for every divisor B,

(A − B) ≤ i(B) .

Hint. Find a monomorphism μ : L (A − B) → ΩF (B).

(iii) As a special case of (ii), show that

i(A) > 0 ⇒ (A) ≤ g .

1.17. For a divisor C ∈ Div(F ) with (C) > 0 we define

|C| := {A ∈ Div(F ) | A ∼ C and A ≥ 0} .

This set is called the linear system corresponding to C. Obviously it depends

only on the divisor class [C] ∈ Cl(F ). The class [C] is called primitive, if there
is no divisor B > 0 such that B ≤ A for all A ∈ |C|. Show:
(i) Every divisor class of degree ≥ 2g is primitive.
(ii) For g ≥ 1, the canonical class is primitive.
(iii) Let g ≥ 1, W a canonical divisor and P be a place of degree one. Then
the class [W + P ] is not primitive.

1.18. The number γ := min{[F : K(z)]|z ∈ F } is called the gonality of F/K.

We also define for all r ≥ 1,

γr := min{deg(A) | A ∈ Div(F ) and (A) ≥ r} .

The sequence (γ1 , γ2 , γ3 , . . .) is called the gonality sequence of F/K.

(i) Show that γ1 = 0 and γ2 = γ.
In parts (ii) - (viii) we assume that there exists a rational place P ∈ IPF .
Prove:
(ii) For all integers r ≥ 1 there exists a divisor Ar ≥ 0 with deg(Ar ) = γr and
(Ar ) = r.
(iii) γr < γr+1 for all r ≥ 1.
(iv) γr = r + g − 1 for all r > g.
(v) If g ≥ 1, then γg = 2g − 2.
(vi) γr ≥ 2(r − 1) for all r ∈ {1, . . . , g}.
(vii) If g ≥ 2, then γ ≤ g.
(viii) Let Γ := {j ≥ 0 | there is no r with γr = j}. Then
(1) |Γ | = g,
(2) 1 ∈ Γ and 2g − 1 ∈ Γ , if g ≥ 1.
2
Algebraic Geometry Codes

In this chapter we describe V.D.Goppa’s construction of error-correcting codes

using algebraic function fields. We start with a brief survey of the concepts
of coding theory. Then we define algebraic geometry codes (AG codes) and
develop their main properties. The codes constructed by means of a rational
function field are discussed in detail in Section 2.3.

2.1 Codes

We are going to introduce some basic notions of coding theory. The reader
who is not familiar with these concepts is referred to the introductory chapter
of any book on error-correcting codes.
Let IFq denote the finite field with q elements. We consider the n-
dimensional vector space IFnq whose elements are n-tuples a = (a1 , . . . , an )
with ai ∈ IFq .

Definition 2.1.1. For a = (a1 , . . . , an ) and b = (b1 , . . . , bn ) ∈ IFnq let

 
d(a, b) :=  { i ; ai = bi }  .

This function d is called the Hamming distance on IFnq . The weight of an

element a ∈ IFnq is defined as
 
wt(a) := d(a, 0) =  { i ; ai = 0}  .

The Hamming distance is a metric on IFnq as one can verify immediately.

In particular, the Triangle Inequality d(a, c) ≤ d(a, b) + d(b, c) holds for all
a, b, c ∈ IFnq .

Definition 2.1.2. A code C (over the alphabet IFq ) is a linear subspace of

IFnq ; the elements of C are called codewords. We call n the length of C and
H. Stichtenoth, Algebraic Function Fields and Codes, 45
Graduate Texts in Mathematics 254,

c Springer-Verlag Berlin Heidelberg 2009
46 2 Algebraic Geometry Codes

dim C (as IFq -vector space) the dimension of C. An [n, k] code is a code of
length n and dimension k.
The minimum distance d(C) of a code C = 0 is defined as

d(C) := min {d(a, b) | a, b ∈ C and a = b} = min {wt(c) | 0 = c ∈ C} .

An [n, k] code with minimum distance d will be referred to as an [n, k, d] code.

Remark 2.1.3. More generally one can define a code to be an arbitrary non-
empty subset C ⊆ An where A = ∅ is a finite set. If A = IFq and C ⊆ IFnq
is a linear subspace, C is said to be a linear code. Most codes employed in
practice belong to this class, therefore we will only consider linear codes in
this book, without writing the attribute ‘linear’.

For a code C with minimum distance d = d(C) we set t := [(d − 1)/2]

(where [x] denotes the integer part of the real number x; i.e., x = [x] + ε
with [x] ∈ ZZ and 0 ≤ ε < 1). Then C is said to be t-error correcting. The
following is obvious: if u ∈ IFnq and d(u, c) ≤ t for some c ∈ C then c is the
only codeword with d(u, c) ≤ t.
A simple way to describe a specific code C explicitly is to write down a
basis of C (as a vector space over IFq ).

Definition 2.1.4. Let C be an [n, k] code over IFq . A generator matrix of C

is a k × n matrix whose rows are a basis of C.

Definition 2.1.5. The canonical inner product on IFnq is defined by


n
a, b := ai bi ,
i=1

for a = (a1 , . . . , an ) and b = (b1 , . . . , bn ) ∈ IFnq .

Obviously this is a non-degenerate symmetric bilinear form on IFnq .

Definition 2.1.6. If C ⊆ IFnq is a code then

C ⊥ := {u ∈ IFnq | u, c = 0 for all c ∈ C}

is called the dual of C. The code C is called self-dual (resp. self-orthogonal) if

C = C ⊥ (resp. C ⊆ C ⊥ ).

It is well-known from linear algebra that the dual of an [n, k] code is an

[n, n − k] code, and (C ⊥ )⊥ = C. In particular, the dimension of a self-dual
code of length n is n/2.
 2.1 Codes 47

Definition 2.1.7. A generator matrix H of C ⊥ is said to be a parity check

matrix for C.

Clearly a parity check matrix H of an [n, k] code C is an (n−k)×n matrix

of rank n − k, and we have

C = { u ∈ IFnq | H · ut = 0 }

(where ut denotes the transpose of u). Thus a parity check matrix ‘checks’
whether a vector u ∈ IFnq is a codeword or not.
One of the basic problems in algebraic coding theory is to construct - over
a fixed alphabet IFq - codes whose dimension and minimum distance are large
in comparison with their length. However there are some restrictions. Roughly
speaking, if the dimension of a code is large (with respect to its length), then
its minimum distance is small. The simplest bound is the following.

Proposition 2.1.8 (Singleton Bound). For an [n, k, d] code C holds

k + d ≤ n + 1.

Proof. Consider the linear subspace E ⊆ IFnq given by

E := { (a1 , . . . , an ) ∈ IFnq | ai = 0 for all i ≥ d } .

Every a ∈ E has weight ≤ d − 1, hence E ∩ C = 0. As dim E = d − 1 we obtain

k + (d − 1) = dim C + dim E
= dim (C + E) + dim (C ∩ E) = dim (C + E) ≤ n .

Codes with k + d = n + 1 are in a sense optimal; such codes are called

MDS codes (maximum distance separable codes). If n ≤ q +1, there exist MDS
codes over IFq for all dimensions k ≤ n (this will be shown in Section 2.3).
The Singleton Bound does not take into consideration the size of the al-
phabet. Several other upper bounds for the parameters k and d (involving
the length n of the code and the size q of the alphabet) are known. They are
stronger than the Singleton Bound if n is large with respect to q. We refer to
[25],[28], see also Chapter 8, Section 8.4.
It is in general a much harder problem to obtain lower bounds for the
minimum distance of a given code (or a given class of codes). Only few such
classes are known, for instance BCH codes, Goppa codes or quadratic residue
codes (cf. [25],[28]). One of the reasons for the interest in algebraic geometry
codes (to be defined in the next section) is that for this large class of codes a
good lower bound for the minimum distance is available.
48 2 Algebraic Geometry Codes

2.2 AG Codes
Algebraic geometry codes (AG codes) were introduced by V.D. Goppa in
[15]. Therefore they are sometimes also called geometric Goppa codes. As a
motivation for the construction of these codes we first consider Reed-Solomon
codes over IFq . This important class of codes is well-known in coding theory
for a long time. Algebraic geometry codes are a very natural generalization of
Reed-Solomon codes.
Let n = q − 1 and let β ∈ IFq be a primitive element of the multiplicative
group IF× ×
q ; i.e., IFq = {β, β , . . . , β = 1}. For an integer k with 1 ≤ k ≤ n
2 n

we consider the k-dimensional vector space

Lk := {f ∈ IFq [X] | deg f ≤ k − 1} (2.1)

and the evaluation map ev : Lk → IFnq given by

ev(f ) := (f (β), f (β 2 ), . . . , f (β n )) ∈ IFnq . (2.2)

Obviously this map is IFq -linear, and it is injective because a non-zero poly-
nomial f ∈ IFq [X] of degree < n has less than n zeros. Therefore

Ck := {(f (β), f (β 2 ), . . . , f (β n )) | f ∈ Lk } (2.3)

is an [n, k] code over IFq ; it is called an RS code (Reed-Solomon code). The

weight of a codeword 0 = c = ev(f ) ∈ Ck is given by
 
wt(c) = n − {i ∈ {1, . . . , n} ; f (β i ) = 0}
≥ n − deg f ≥ n − (k − 1) .

Hence the minimum distance d of Ck satisfies the inequality d ≥ n + 1 − k. On

the other hand, d ≤ n + 1 − k by the Singleton Bound. Thus Reed-Solomon
codes are MDS codes over IFq . Observe however that RS codes are short in
comparison with the size of the alphabet IFq , since n = q − 1.
Now we introduce the notion of an algebraic geometry code. Let us fix
some notation valid for the entire section.

F/IFq is an algebraic function field of genus g.

P1 , . . . , Pn are pairwise distinct places of F/IFq of degree 1.
D = P1 + . . . + P n .
G is a divisor of F/IFq such that supp G ∩ supp D = ∅ .

Definition 2.2.1. The algebraic geometry code (or AG code) CL (D, G) as-
sociated with the divisors D and G is defined as

CL (D, G) := {(x(P1 ), . . . , x(Pn )) | x ∈ L (G)} ⊆ IFnq .

 2.2 AG Codes 49

Note that this definition makes sense: for x ∈ L (G) we have vPi (x) ≥ 0
(i = 1, . . . , n) because supp G ∩ supp D = ∅. The residue class x(Pi ) of x
modulo Pi is an element of the residue class field of Pi (see Definition 1.1.14).
As deg Pi = 1, this residue class field is IFq , so x(Pi ) ∈ IFq .
As in (2.2) we can consider the evaluation map evD : L (G) → IFnq given by

evD (x) := (x(P1 ), . . . , x(Pn )) ∈ IFnq . (2.4)

The evaluation map is IFq -linear, and CL (D, G) is the image of L (G) under
this map. The analogy with the definition of Reed-Solomon codes (2.3) is
obvious. In fact, choosing the function field F/IFq and the divisors D and G
in an appropriate manner, RS codes are easily seen to be a special case of AG
codes, see Section 2.3.
Definition 2.2.1 looks like a very artificial way to define certain codes over
IFq . The next theorem will show why these codes are interesting: one can
calculate (or at least estimate) their parameters n, k and d by means of the
Riemann-Roch Theorem, and one obtains a non-trivial lower bound for their
minimum distance in a very general setting.

Theorem 2.2.2. CL (D, G) is an [n, k, d] code with parameters

k = (G) − (G − D) and d ≥ n − deg G .

Proof. The evaluation map (2.4) is a surjective linear map from L (G) to
CL (D, G) with kernel

Ker(evD ) = {x ∈ L (G) | vPi (x) > 0 for i = 1, . . . , n} = L (G − D) .

It follows that k = dim CL (D, G) = dim L (G) − dim L (G − D) = (G) −

(G − D). The assertion regarding the minimum distance d makes sense only
if CL (D, G) = 0, so we will assume this. Choose an element x ∈ L (G) with
wt(evD (x)) = d. Then exactly n − d places Pi1 , . . . , Pin−d in the support of D
are zeros of x, so

0 = x ∈ L (G − (Pi1 + . . . + Pin−d )) .

We conclude by Corollary 1.4.12.(b) that

0 ≤ deg (G − (Pi1 + . . . + Pin−d )) = deg G − n + d .

Hence d ≥ n − deg G.

Corollary 2.2.3. Suppose that the degree of G is strictly less than n. Then
the evaluation map evD : L (G) → CL (D, G) is injective, and we have:
50 2 Algebraic Geometry Codes

(a) CL (D, G) is an [n, k, d] code with

d ≥ n − deg G and k = (G) ≥ deg G + 1 − g .

Hence
k+d ≥ n+1−g. (2.5)
(b) If in addition 2g − 2 < deg G < n, then k = deg G + 1 − g.
(c) If {x1 , . . . , xk } is a basis of L (G) then the matrix
⎛ ⎞
x1 (P1 ) x1 (P2 ) . . . x1 (Pn )
⎜ .. .. .. ⎟
M =⎝ . . . ⎠
xk (P1 ) xk (P2 ) . . . xk (Pn )

is a generator matrix for CL (D, G).

Proof. By assumption we have deg(G−D) = deg G−n < 0, so L (G−D) = 0.

Since L (G − D) is the kernel of the evaluation map, this is an injective
mapping. The remaining assertions are trivial consequences of Theorem 2.2.2
and the Riemann-Roch Theorem.

We point out that the lower bound (2.5) for the minimum distance looks
very similar to the upper Singleton Bound. Putting both bounds together we
see that for deg G < n,

n + 1 − g ≤ k + d ≤ n + 1. (2.6)

Note that k + d = n + 1 if F is a function field of genus g = 0. Hence the

AG codes constructed by means of a rational function field IFq (z) are always
MDS codes. For more details see Section 2.3.
In order to obtain a meaningful bound for the minimum distance of
CL (D, G) by Theorem 2.2.2, we often assume that deg G < n.

Definition 2.2.4. The integer d∗ := n − deg G is called the designed distance

of the code CL (D, G).

Theorem 2.2.2 states that the minimum distance d of an AG code cannot

be less than its designed distance. The question whether d∗ = d or d∗ < d is
answered by the following remark.

Remark 2.2.5. Suppose that (G) > 0 and d∗ = n − deg G > 0. Then d∗ = d
if and only if there exists a divisor D with 0 ≤ D ≤ D, deg D = deg G and
(G − D ) > 0.
 2.2 AG Codes 51

Proof. First we assume d∗ = d. Then there is an element 0 = x ∈ L (G) such

that the codeword (x(P1 ), . . . , x(Pn )) ∈ CL (D, G)) has precisely n − d =
n − d∗ = deg G zero components, say x(Pij ) = 0 for j = 1, . . . , deg G. Put

G
deg
D := Pij .
j=1

Then 0 ≤ D ≤ D, deg D = deg G and (G − D ) > 0 (as x ∈ L (G − D )).

Conversely, if D has the above properties then we choose an element 0 =
y ∈ L (G − D ). The weight of the corresponding codeword (y(P1 ), . . . , y(Pn ))
is n − deg G = d∗ , hence d = d∗ .

Another code can be associated with the divisors G and D, by using lo-
cal components of Weil differentials. We recall some notation introduced in
Chapter 1. For a divisor A ∈ Div(F ), ΩF (A) is the space of Weil differentials
ω with (ω) ≥ A. This is a finite-dimensional vector space over IFq of dimen-
sion i(A) (the index of specialty of A). For a Weil differential ω and a place
P ∈ IPF , the map ωP : F → IFq denotes the local component of ω at P .

Definition 2.2.6. Let G and D = P1 + . . . + Pn be divisors as before (i.e.,

the Pi are pairwise distinct places of degree one, and supp G ∩ supp D = ∅).
Then we define the code CΩ (D, G) ⊆ IFnq by

CΩ (D, G) := {(ωP1 (1), . . . , ωPn (1)) | ω ∈ ΩF (G − D)} .

Also the code CΩ (D, G) is called an algebraic geometry code. The relation
between the codes CL (D, G) and CΩ (D, G) will be explained in Theorem
2.2.8 and Proposition 2.2.10. Our first result about CΩ (D, G) is an analogue
to Theorem 2.2.2.

Theorem 2.2.7. CΩ (D, G) is an [n, k  , d ] code with parameters

k  = i(G − D) − i(G) and d ≥ deg G − (2g − 2) .

Under the additional hypothesis deg G > 2g − 2, we have k  = i(G − D) ≥

n + g − 1 − deg G. If moreover 2g − 2 < deg G < n then

k  = n + g − 1 − deg G .

Proof. Let P ∈ IPF be a place of degree one and let ω be a Weil differential
with vP (ω) ≥ −1. We claim that

ωP (1) = 0 ⇐⇒ vP (ω) ≥ 0 . (2.7)

In order to prove this we use Proposition 1.7.3 which states that for an integer
r ∈ ZZ,
52 2 Algebraic Geometry Codes

vP (ω) ≥ r ⇐⇒ ωP (x) = 0 for all x ∈ F with vP (x) ≥ −r . (2.8)

The implication ⇐ of (2.7) is an obvious consequence of (2.8). Conversely,

suppose that ωP (1) = 0. Let x ∈ F with vP (x) ≥ 0. Since deg P = 1, we can
write x = a + y with a ∈ IFq and vP (y) ≥ 1. Then

ωP (x) = ωP (a) + ωP (y) = a · ωP (1) + 0 = 0 .

(Observe that ωP (y) = 0 because vP (ω) ≥ −1 and vP (y) ≥ 1, cf. (2.8).) Hence
(2.7) is proved.
Now we consider the IFq -linear mapping

ΩF (G − D) −→ CΩ (D, G) ,
D :
ω −→ (ωP1 (1), . . . , ωPn (1)) .

D is surjective, and its kernel is ΩF (G) by (2.7). Therefore

k  = dim ΩF (G − D) − dim ΩF (G) = i(G − D) − i(G) . (2.9)

Let D (ω) ∈ CΩ (D, G) be a codeword of weight m > 0. Then ωPi (1) = 0 for
certain indices i = i1 , . . . , in−m , so

  
n−m

ω ∈ ΩF G − D − Pij
j=1

by (2.7). Since ΩF (A) = 0 implies deg A ≤ 2g − 2 (by Theorem 1.5.17), we

obtain  
2g − 2 ≥ deg G − n − (n − m) = deg G − m .
Hence the minimum distance d of CΩ (D, G) satisfies the inequality d ≥
deg G − (2g − 2).
Assume now that deg G > 2g − 2. By Theorem 1.5.17 we obtain i(G) = 0.
Now (2.9) and the Riemann-Roch Theorem yield

k  = i(G − D) = (G − D) − deg (G − D) − 1 + g

= (G − D) + n + g − 1 − deg G .

The remaining assertions of Theorem 2.2.7 follow immediately.

In analogy to Definition 2.2.4, the integer deg G − (2g − 2) is called the

designed distance of CΩ (D, G).
There is a close relation between the codes CL (D, G) and CΩ (D, G):

Theorem 2.2.8. The codes CL (D, G) and CΩ (D, G) are dual to each other;
i.e.,
CΩ (D, G) = CL (D, G)⊥ .
 2.2 AG Codes 53

Proof. First we note the following fact: Consider a place P ∈ IPF of degree one,
a Weil differential ω with vP (ω) ≥ −1 and an element x ∈ F with vP (x) ≥ 0.
Then

ωP (x) = x(P ) · ωP (1) . (2.10)

In order to prove (2.10) we write x = a+y with a = x(P ) ∈ IFq and vP (y) > 0.
Then ωP (x) = ωP (a) + ωP (y) = a · ωP (1) + 0 = x(P ) · ωP (1), by (2.8).
Next we show that CΩ (D, G) ⊆ CL (D, G)⊥ . So let ω ∈ ΩF (G − D) and
x ∈ L (G). We obtain


0 = ω(x) = ωP (x) (2.11)
P ∈IPF

n
= ωPi (x) (2.12)
i=1
n
= x(Pi ) · ωPi (1) (2.13)
i=1
= (ωP1 (1), . . . , ωPn (1)), (x(P1 ), . . . , x(Pn )) ,
where  ,  denotes the canonical inner product on IFnq . We still have to justify
the single steps in the above computation. (2.11) follows from Proposition
1.7.2 and the fact that Weil differentials vanish on principal adeles. For P ∈
IPF \{P1 , . . . , Pn } we have vP (x) ≥ −vP (ω) (as x ∈ L (G) and ω ∈ Ω(G−D)),
so ωP (x) = 0 by (2.8). This proves (2.12). Finally, (2.13) follows from (2.10).
Hence CΩ (D, G) ⊆ CL (D, G)⊥ .
It is now sufficient to show that the codes CΩ (D, G) and CL (D, G)⊥
have the same dimension. Using Theorems 2.2.2, 2.2.7 and the Riemann-Roch
Theorem we find:

dim CΩ (D, G) = i(G − D) − i(G)

= (G − D) − deg (G − D) − 1 + g − ((G) − deg G − 1 + g)
= deg D + (G − D) − (G)
= n − ((G) − (G − D))
= n − dim CL (D, G) = dim CL (D, G)⊥ .

Our next aim is to prove that CΩ (D, G) can be represented as CL (D, H)

with an appropriate divisor H. For this purpose we need the following lemma.

Lemma 2.2.9. There exists a Weil differential η such that

vPi (η) = −1 and ηPi (1) = 1 for i = 1, . . . , n.
54 2 Algebraic Geometry Codes

Proof. Choose an arbitrary Weil differential ω0 = 0. By the Weak Approx-

imation Theorem there is an element z ∈ F with vPi (z) = −vPi (ω0 ) − 1
for i = 1, . . . , n. Setting ω := zω0 we obtain vPi (ω) = −1. Therefore
ai := ωPi (1) = 0 by (2.7). Again by the Approximation Theorem we find
y ∈ F such that vPi (y − ai ) > 0. It follows that vPi (y) = 0 and y(Pi ) = ai .
We put η := y −1 ω and obtain vPi (η) = vPi (ω) = −1, and

ηPi (1) = ωPi (y −1 ) = y −1 (Pi ) · ωPi (1) = a−1

i · ai = 1 .

Proposition 2.2.10. Let η be a Weil differential such that vPi (η) = −1 and
ηPi (1) = 1 for i = 1, . . . , n. Then

CL (D, G)⊥ = CΩ (D, G) = CL (D, H) with H := D − G + (η) .

Proof. The equality CL (D, G)⊥ = CΩ (D, G) was already shown in Theorem
2.2.8. Observe that supp (D − G + (η)) ∩ supp D = ∅ since vPi (η) = −1 for
i = 1, . . . , n. Hence the code CL (D, D − G + (η)) is defined. By Theorem
1.5.14 there is an isomorphism μ : L (D − G + (η)) → ΩF (G − D) given by
μ(x) := xη. For x ∈ L (D − G + (η)) we have

(xη)Pi (1) = ηPi (x) = x(Pi ) · ηPi (1) = x(Pi ),

cf. (2.10). This implies CΩ (D, G) = CL (D, D − G + (η)).

Corollary 2.2.11. Suppose there is a Weil differential η such that

2G − D ≤ (η) and ηPi (1) = 1 for i = 1, . . . , n .

Then the code CL (D, G) is self-orthogonal; i.e., CL (D, G) ⊆ CL (D, G)⊥ . If

2G − D = (η) and ηPi (1) = 1 for i = 1, . . . , n ,

then CL (D, G) is self-dual.

Proof. The assumption 2G − D ≤ (η) is equivalent to G ≤ D − G + (η). Hence

Proposition 2.2.10 implies

CL (D, G)⊥ = CL (D, D − G + (η)) ⊇ CL (D, G) .

This proves the first assertion. If we assume equality 2G − D = (η) then we

have G = D − G + (η) and therefore

CL (D, G)⊥ = CL (D, D − G + (η)) = CL (D, G) .

 2.3 Rational AG Codes 55

Remark 2.2.12. Using Proposition 2.2.10, one can reduce Theorem 2.2.7 to
Theorem 2.2.2. This gives another proof of Theorem 2.2.7.

Definition 2.2.13. Two codes C1 , C2 ⊆ IFnq are said to be equivalent if there

is a vector a = (a1 , . . . , an ) ∈ (IF×
q ) such that C2 = a · C1 ; i.e.,
n

C2 = {(a1 c1 , . . . , an cn ) | (c1 , . . . , cn ) ∈ C1 } .

Evidently equivalent codes have the same dimension and the same minimum
distance. Note however that equivalence does not preserve all interesting prop-
erties of a code. For instance, equivalent codes may have non-isomorphic au-
tomorphism groups. We will consider automorphisms of codes in Chapter 8.
Proposition 2.2.14. (a) Suppose G1 and G2 are divisors with G1 ∼ G2 and
supp G1 ∩ supp D = supp G2 ∩ supp D = ∅. Then the codes CL (D, G1 ) and
CL (D, G2 ) are equivalent. The same holds for CΩ (D, G1 ) and CΩ (D, G2 ).
(b) Conversely, if a code C ⊆ IFnq is equivalent to CL (D, G) (resp. CΩ (D, G))
then there exists a divisor G ∼ G such that supp G ∩ supp D = ∅ and C =
CL (D, G ) (resp. C = CΩ (D, G )).
Proof. (a) By assumption we have that G2 = G1 − (z) with vPi (z) = 0
for i = 1, . . . , n. Hence a := (z(P1 ), . . . , z(Pn )) is in (IF× n
q ) , and the mapping
x → xz from L (G1 ) to L (G2 ) is bijective (cf. Lemma 1.4.6). This implies that
CL (D, G2 ) = a · CL (D, G1 ). The equivalence of CΩ (D, G1 ) and CΩ (D, G2 )
is proved similarly.
(b) Let C = a · CL (D, G) with a = (a1 , . . . , an ) ∈ (IF× q ) . Choose z ∈ F with
n

z(Pi ) = ai (i = 1, . . . , n) and set G := G − (z). Then C = CL (D, G ).



Remark 2.2.15. If G is a divisor whose support is not disjoint from supp D,

we can still define an algebraic geometry code CL (D, G) associated with D
and G as follows: Choose a divisor G ∼ G with supp G ∩ supp D = ∅ (which
is possible by the Approximation Theorem) and set CL (D, G) := CL (D, G ).
The choice of G is not unique and the code CL (D, G) is well-defined only up
to equivalence, by Proposition 2.2.14.

2.3 Rational AG Codes

In this section we investigate AG codes associated with divisors of a rational
function field. We shall describe these codes very explicitly by means of gen-
erator and parity check matrices. In coding theory this class of codes is known
by the name of Generalized Reed-Solomon codes. Some of the most important
codes used in practice (such as BCH codes and Goppa codes; these codes will
be defined later in this section) can be represented as subfield subcodes of
Generalized Reed-Solomon codes in a natural manner.
56 2 Algebraic Geometry Codes

Definition 2.3.1. An algebraic geometry code CL (D, G) associated with di-

visors G and D of a rational function field IFq (z)/IFq is said to be rational
(as in Section 2.2 it is assumed that D = P1 + . . . + Pn with pairwise distinct
places of degree one, and supp G ∩ supp D = ∅).

Observe that the length of a rational AG code is bounded by q + 1 because

IFq (z) has only q + 1 places of degree one: the pole P∞ of z and for each
α ∈ IFq , the zero Pα of z − α (see Proposition 1.2.1). The following results
follow immediately from Section 2.2.
Proposition 2.3.2. Let C = CL (D, G) be a rational AG code over IFq , and
let n, k, d be the parameters of C. Then we have:
(a) n ≤ q + 1.
(b) k = 0 if and only if deg G < 0, and k = n if and only if deg G > n − 2.
(c) For 0 ≤ deg G ≤ n − 2,
k = 1 + deg G and d = n − deg G .
In particular, C is an MDS code.
(d) C ⊥ is also a rational AG code.

Next we determine specific generator matrices for rational AG codes.

Proposition 2.3.3. Let C = CL (D, G) be a rational AG code over IFq with
parameters n, k and d.
(a) If n ≤ q then there exist pairwise distinct elements α1 , . . . , αn ∈ IFq and
v1 , . . . , vn ∈ IF×
q (not necessarily distinct) such that

C = {(v1 f (α1 ), v2 f (α2 ), . . . , vn f (αn ))| f ∈ IFq [z] and deg f ≤ k − 1} .

The matrix ⎛ ⎞
v1 v2 ... vn
⎜ α1 v1 α2 v2 ... αn vn ⎟
⎜ ⎟
⎜ α12 v1 α22 v2 ... αn2 vn ⎟
M =⎜ ⎟ (2.14)
⎜ .. .. .. ⎟
⎝ . . . ⎠
α1k−1 v1 α2k−1 v2 . . . αnk−1 vn
is a generator matrix for C.
(b) If n = q + 1, C has a generator matrix
⎛ ⎞
v1 v2 ... vn−1 0
⎜ α1 v1 α2 v2 . . . αn−1 vn−1 0 ⎟
⎜ ⎟
⎜ 2
α22 v2 2 ⎟
M = ⎜ α1 v1 . . . αn−1 vn−1 0 ⎟ (2.15)
⎜ . . .. .. ⎟
⎝ .. .. . . ⎠
α1k−1 v1 α2k−1 v2 k−1
. . . αn−1 vn−1 1

where IFq = {α1 , . . . , αn−1 } and v1 , . . . , vn−1 ∈ IF×

q .
 2.3 Rational AG Codes 57

Proof. (a) Let D = P1 + . . . + Pn . As n ≤ q, there is a place P of degree one

which is not in the support of D. Choose a place Q = P of degree one (e.g.,
Q = P1 ). By Riemann-Roch, (Q − P ) = 1, hence Q − P is a principal divisor
(Corollary 1.4.12). Let Q − P = (z); then z is a generating element of the
rational function field over IFq and P is the pole divisor of z. As usually we
write P = P∞ . By Proposition 2.3.2 we can assume that deg G = k − 1 ≥ 0
(the case k = 0 being trivial). The divisor (k − 1)P∞ − G has degree zero, so
it is principal (Riemann-Roch and Corollary 1.4.12), say (k − 1)P∞ − G = (u)
with 0 = u ∈ F . The k elements u, zu, . . . , z k−1 u are in L (G) and they are
linearly independent over IFq . Since (G) = k, they constitute a basis of L (G);
i.e.,
L (G) = {uf (z) | f ∈ IFq [z] and deg f ≤ k − 1} .
Setting αi := z(Pi ) and vi := u(Pi ) we obtain

(uf (z))(Pi ) = u(Pi )f (z(Pi )) = vi f (αi )

for i = 1, . . . , n. Therefore

C = CL (D, G) = {(v1 f (α1 ), . . . , vn f (αn )) | deg f ≤ k − 1} .

The codeword in C corresponding to uz j is (v1 α1j , v2 α2j , . . . , vn αnj ), so the

matrix (2.14) is a generator matrix of C.
(b) The proof is essentially the same as in the case n ≤ q. Now we have
n = q + 1 and we can choose z in such a way that Pn = P∞ is the pole of
z. As above, (k − 1)P∞ − G = (u) with 0 = u ∈ F , and {u, zu, . . . , z k−1 u} is
a basis of L (G). For 1 ≤ i ≤ n − 1 = q the elements αi := z(Pi ) ∈ IFq are
pairwise distinct, so IFq = {α1 , . . . , αn−1 }. Moreover, vi := u(Pi ) ∈ IF× q for
i = 1, . . . , n − 1. For 0 ≤ j ≤ k − 2 we obtain

((uz j )(P1 ), . . . , (uz j )(Pn )) = (α1j v1 , . . . , αn−1

j
vn−1 , 0) ,

but for j = k − 1 holds

((uz k−1 )(P1 ), . . . , (uz k−1 )(Pn )) = (α1k−1 v1 , . . . , αn−1

k−1
vn−1 , γ)

with an element 0 = γ ∈ IFq . Substituting u by γ −1 u yields the generator

matrix (2.15).

Definition 2.3.4. Let α = (α1 , . . . , αn ) where the αi are distinct elements

of IFq , and let v = (v1 , . . . , vn ) where the vi are nonzero (not necessarily
distinct) elements of IFq . Then the Generalized Reed-Solomon code, denoted
by GRSk (α, v), consists of all vectors

(v1 f (α1 ), . . . , vn f (αn ))

with f (z) ∈ IFq [z] and deg f ≤ k − 1 (for a fixed k ≤ n).

58 2 Algebraic Geometry Codes

In the case α = (β, β 2 , . . . , β n ) (where n = q − 1 and β is a primitive n-th

root of unity) and v = (1, 1, . . . , 1), GRSk (α, v) is a Reed-Solomon code, cf.
Section 2.2.
Obviously GRSk (α, v) is an [n, k] code, and Proposition 2.3.3.(a) states
that all rational AG codes over IFq of length n ≤ q are Generalized Reed-
Solomon codes. The converse is also true:

Proposition 2.3.5. Every generalized Reed-Solomon code GRSk (α, v) can be

represented as a rational AG code.

Proof. Let α = (α1 , . . . , αn ) with αi ∈ IFq and v = (v1 , . . . , vn ) with vi ∈ IF×

q .
Consider the rational function field F = IFq (z). Denote by Pi the zero of
z − αi (i = 1, . . . , n) and by P∞ the pole of z. Choose u ∈ F such that

u(Pi ) = vi for i = 1, . . . , n . (2.16)

Such an element exists by the Approximation Theorem. (One can also de-
termine a polynomial u = u(z) ∈ IFq [z] satisfying (2.16), by using Lagrange
interpolation.) Now let

D := P1 + . . . + Pn and G := (k − 1)P∞ − (u) .

The proof of Proposition 2.3.3 shows that GRSk (α, v) = CL (D, G).

The same arguments apply to a code of length n = q + 1 over IFq which

has a generator matrix of the specific form (2.15). All such codes can be
represented as rational AG codes.
In order to determine the dual of a rational AG code C = CL (D, G), we
need (by Theorem 2.2.8 and Proposition 2.2.10) a Weil differential ω of IFq (z)
such that

vPi (ω) = −1 and ωPi (1) = 1 for i = 1, . . . , n . (2.17)

Lemma 2.3.6. Consider the rational function field F = IFq (z) and n distinct
elements
n α1 , . . . , αn ∈ IFq . Let Pi ∈ IPF be the zero of z − αi and h(z) :=
i=1 (z −α i ). Suppose y is a element of F such that y(Pi ) = 1 for i = 1, . . . , n.
Then there exists a Weil differential ω of F/IFq with the property (2.17) and
the divisor
(ω) = (y) + (h (z)) − (h(z)) − 2P∞ (2.18)
(where h (z) ∈ IFq [z] is the derivative of the polynomial h(z)).

Proof. There is a Weil differential η of F with (η) = −2P∞ and ηP∞ (z −1 ) =

−1 (see Proposition 1.7.4). We set

ω := y · (h (z)/h(z)) · η .
 2.3 Rational AG Codes 59

The divisor of ω is (ω) = (y) + (h (z)) − (h(z)) − 2P∞ , in particular holds
vPi (ω) = −1 for i = 1, . . . , n. We have to verify that ωPi (1) = 1. Write
h(z) = (z − αi )gi (z); then
  
h (z)   gi (z) 1 1
y· = 1 + (y − 1) · + = +u
h(z) gi (z) z − αi z − αi

with u ∈ F and vPi (u) ≥ 0 (because vPi (y − 1) > 0 and vPi (gi (z)) = 0). Since
ηPi ((z − αi )−1 ) = 1 and ηPi (u) = 0 (by Proposition 1.7.4(c) and Proposition
1.7.3(a)), we obtain
   
h (z) 1
ωPi (1) = ηPi y · = ηPi + u = 1.
h(z) z − αi

Note that Lemma 2.3.6 - combined with Theorem 2.2.8, Proposition

2.2.10 and Proposition 2.3.3 - enables us to specify a parity check matrix
for CL (D, G).
Next we would like to describe BCH codes and Goppa codes by means of
rational AG codes. To this end we need the following concept:

Definition 2.3.7. Consider an extension field IFqm of IFq and a code C over
IFqm of length n. Then
C|IFq := C ∩ IFnq
is called the subfield subcode of C (or the restriction of C to IFq ).

C|IFq is a code over IFq . Its minimum distance cannot be less than the
minimum distance of C, and for the dimension of C|IFq we have the trivial
estimate dim C|IFq ≤ dim C. In general this can be a strict inequality.

Definition 2.3.8. Assume that n|(q m − 1) and let β ∈ IFqm be a primitive

n-th root of unity. Let l ∈ ZZ and δ ≥ 2. Define a code C(n, l, δ) over IFqm by
the generator matrix
⎛ ⎞
1 βl β 2l ... β (n−1)l
⎜ 1 β l+1 β 2(l+1) . . . β (n−1)(l+1) ⎟
⎜ ⎟
H := ⎜ . .. .. .. ⎟ (2.19)
⎝ .. . . . ⎠
1 β l+δ−2 β 2(l+δ−2) . . . β (n−1)(l+δ−2)

The code C := C(n, l, δ)⊥ |IFq is called a BCH code with designed distance δ.
In other words,
C = {c ∈ IFnq | H · ct = 0} , (2.20)
where the matrix H is given by (2.19).
60 2 Algebraic Geometry Codes

We note that BCH codes are usually defined as special cyclic codes. One
can easily show however, that our Definition 2.3.8 coincides with the usual
definition, cf. [25],[28].

Proposition 2.3.9. Let n|(q m − 1) and let β ∈ IFqm be a primitive n-th root
of unity. Let F = IFqm (z) be the rational function field over IFqm and P0 (resp.
P∞ ) be the zero (resp. pole) of z. For i = 1, . . . , n denote by Pi the zero of
z − β i−1 , and set Dβ := P1 + . . . + Pn . Suppose that a, b ∈ ZZ are integers with
0 ≤ a + b ≤ n − 2. Then we have
(a) CL (Dβ , aP0 + bP∞ ) = C(n, l, δ) with l = −a and δ = a + b + 2 (where
C(n, l, δ) is as in Definition 2.3.8).
(b) The dual of CL (Dβ , aP0 + bP∞ ) is given by

CL (Dβ , aP0 + bP∞ )⊥ = CL (Dβ , rP0 + sP∞ )

with r = −(a + 1) and s = n − b − 1. Hence the BCH code C(n, l, δ)⊥ |IFq
is the restriction to IFq of the code CL (Dβ , rP0 + sP∞ ), with r = l − 1 and
s = n + 1 − δ − l.

Proof. (a) We consider the code CL (Dβ , aP0 + bP∞ ) where 0 ≤ a + b ≤ n − 2.

The elements z −a · z j with 0 ≤ j ≤ a + b constitute a basis of L (aP0 + bP∞ ).
Hence the matrix
⎛ ⎞
1 β −a β −2a ... (β n−1 )−a
⎜ 1 β −a+1 β −2a+2 ... (β n−1 )−a+1 ⎟
⎜ ⎟
⎜ .. .. .. .. ⎟
⎝ . . . . ⎠
1 β −a+(a+b) β −2a+2(a+b) . . . (β n−1 )−a+(a+b)

is a generator matrix of CL (Dβ , aP0 + bP∞ ). Substituting l := −a and δ :=

a + b + 2 we obtain the matrix (2.19), so CL (Dβ , aP0 + bP∞ ) = C(n, l, δ).
(b) We use the notation of Lemma 2.3.6 and set


n
y := z −n and h(z) := (z − β i−1 ) = z n − 1 .
i=1

Proposition 2.2.10 yields CL (Dβ , aP0 + bP∞ )⊥ = CL (Dβ , B) with

B = Dβ − (aP0 + bP∞ ) + (z −n ) + (h (z)) − (h(z)) − 2P∞

= Dβ − (aP0 + bP∞ ) + n(P∞ − P0 ) + (n − 1)(P0 − P∞ )
−(Dβ − nP∞ ) − 2P∞
= (−a − 1)P0 + (n − b − 1)P∞ .

Since l = −a and δ = a + b + 2 (by (a)), we find CL (Dβ , aP0 + bP∞ )⊥ =

CL (Dβ , rP0 + sP∞ ) with s = n − b − 1 = n − (δ − a − 2) − 1 = n + 1 − δ − l
and r = −a − 1 = l − 1.
 2.3 Rational AG Codes 61

Next we introduce Goppa codes. As with BCH codes, our definition of

Goppa codes differs from the usual definition given in most books on coding
theory. However both definitions are equivalent.

Definition 2.3.10. Let L = {α1 , . . . , αn } ⊆ IFqm with |L| = n, and let g(z) ∈
IFqm [z] be a polynomial of degree t such that 1 ≤ t ≤ n − 1 and g(αi ) = 0 for
all αi ∈ L.
(a) We define a code C(L, g(z)) ⊆ (IFqm )n by the generator matrix
⎛ ⎞
g(α1 )−1 g(α2 )−1 ... g(αn )−1
⎜ α1 g(α1 )−1 α2 g(α2 )−1 . . . αn g(αn )−1 ⎟
⎜ ⎟
H := ⎜ .. .. .. ⎟ (2.21)
⎝ . . . ⎠
α1t−1 g(α1 )−1 α2t−1 g(α2 )−1 . . . αnt−1 g(αn )−1

(b) The code Γ (L, g(z)) := C(L, g(z))⊥ |IFq is called the Goppa code with Goppa
polynomial g(z). This means

Γ (L, g(z)) = {c ∈ IFnq | H · ct = 0}

with H as in (2.21).

Note that the matrix (2.21) is a special case of (2.14) (with vi = g(αi )−1 ),
hence C(L, g(z)) and C(L, g(z))⊥ are Generalized Reed-Solomon codes. Now
we give an explicit description of these codes as rational AG codes.

Proposition 2.3.11. In addition to the notation of Definition 2.3.10, let Pi

denote the zero of z−αi (for αi ∈ L), P∞ the pole of z and DL := P1 +. . .+Pn .
Let G0 be the zero divisor of g(z) (in the divisor group of the rational function
field F = IFqm (z)). Then we have

C(L, g(z)) = CL (DL , G0 − P∞ ) = CL (DL , A − G0 )⊥ (2.22)

and

Γ (L, g(z)) = CL (DL , G0 − P∞ )⊥ |IFq = CL (DL , A − G0 )|IFq ,

where the divisor A is determined as follows: Set


h(z) := (z − αi ) and A := (h (z)) + (n − 1)P∞ .
αi ∈L

Proof. It is sufficient to prove (2.22). For 0 ≤ j ≤ t − 1, the element z j g(z)−1

is in L (G0 − P∞ ) because

(z j g(z)−1 ) = j(P0 − P∞ ) − (G0 − tP∞ ) ≥ −G0 + P∞ .

62 2 Algebraic Geometry Codes

Since dim L (G0 −P∞ ) = t, the elements g(z)−1 , zg(z)−1 , . . . , z t−1 g(z)−1 form
a basis of L (G0 −P∞ ). Thus (2.21) is a generator matrix of CL (DL , G0 −P∞ );
i.e.,
C(L, g(z)) = CL (DL , G0 − P∞ ) .
From Proposition 2.2.10 and Lemma 2.3.6 we obtain CL (DL , G0 − P∞ )⊥ =
CL (DL , B) with

B = DL − (G0 − P∞ ) + (h (z)) − (h(z)) − 2P∞

= DL − G0 + P∞ + A − (n − 1)P∞ − (DL − nP∞ ) − 2P∞
= A − G0 .

In coding theory the so-called BCH Bound (resp. the Goppa Bound) for
the minimum distance of BCH codes (resp. Goppa codes) is well-known. Both
bounds can easily be derived from the above results.

Corollary 2.3.12. (a) (BCH Bound.) The minimum distance of a BCH code
with designed distance δ is at least δ.
(b) (Goppa Bound.) The minimum distance of a Goppa code Γ (L, g(z)) is at
least 1 + deg g(z).

Proof. (a) Using notation as in Proposition 2.3.9 we represent the BCH code
in the form C = CL (Dβ , rP0 + sP∞ )|IFq . The minimum distance of the code
CL (Dβ , rP0 + sP∞ ) is, by Propositions 2.3.2 and 2.3.9(b),
 
d = n − deg (rP0 + sP∞ ) = n − (l − 1) + (n + 1 − δ − l) = δ .

Since the minimum distance of a subfield subcode is not less than the minimum
distance of the original code, the minimum distance of C is ≥ δ.
(b) In the same manner we represent Γ (L, g(z)) as CL (DL , A − G0 )|IFq (nota-
tion as in Proposition 2.3.11). As CL (DL , A − G0 ) has the minimum distance
 
d = n − deg (A − G0 ) = n − (n − 1) − deg g(z) = 1 + deg g(z) ,

the assertion follows.

Remark 2.3.13. The subfield subcode construction makes possible to construct

codes over IFq of arbitrary length, by considering codes over an appropriate
extension field IFqm and restricting them to IFq . Note however, that a code C
over IFqm may have good parameters (i.e., large dimension and minimum dis-
tance) whereas the restricted code C|IFq can be very poor (since the dimension
of C|IFq may be much less than the dimension of C, cf. Chapter 9).
 2.4 Exercises 63

Remark 2.3.14. Subfield subcodes of Generalized Reed-Solomon codes are

known as alternant codes, cf. [28]. Propositions 2.3.3 and 2.3.5 state that the
class of alternant codes over IFq corresponds to the class of subfield subcodes
of rational AG codes which are defined over extension fields IFqm ⊇ IFq .

Remark 2.3.15. From the point of view of algebra, the rational function field
IFq (z) is the most trivial example of an algebraic function field. Nevertheless
the AG codes associated with divisors of IFq (z) are already interesting codes
as we have seen in this section. So it looks promising to consider algebraic
geometry codes associated with non-rational function fields F/IFq .
Often a function field F is represented in the form

F = IFq (x, y) with ϕ(x, y) = 0 ,

where ϕ is a non-constant irreducible polynomial in two variables with coef-

ficients in IFq . Then F can be regarded as a finite algebraic extension of the
rational function field IFq (x) (or IFq (y)). Several problems arise:
(1) Is IFq the full constant field of F ?
(2) Calculate the genus of F .
(3) Describe the places of F explicitly. In particular, which places are of degree
one?
(4) Construct a basis for the spaces L (G), at least in specific cases.
(5) Give a convenient description of Weil differentials and of their local com-
ponents.
Another interesting question is:
(6) How many places of degree one can a function field F/IFq of genus g have?
This question is important for coding theory since one often wants to construct
long codes over IFq , and the length of an AG code associated with a function
field is bounded by the number of places of degree one.

In order to tackle these problems it is necessary to develop further the the-

ory of algebraic function fields. This will be done in the subsequent chapters.
We will continue the discussion of codes in Chapter 8.

2.4 Exercises

2.1. For a non-empty subset M ⊆ IFnq we define the support of M as supp M =

{ i | 1 ≤ i ≤ n, there is some c = (c1 , . . . , cn ) ∈ M with ci = 0}.
Now let C be an [n, k] code over IFq . For all r with 1 ≤ r ≤ k, the r-th
Hamming weight of C is defined as follows:

dr (C) := min{ |supp W | ; W ⊆ C is an r-dimensional subspace of C} .

64 2 Algebraic Geometry Codes

The sequence (d1 (C), d2 (C), . . . , dk (C)) is called the weight hierarchy of the
code C. Show:
(i) d1 (C) is equal to the minimum distance of C.
(ii) 0 < d1 (C) < d2 (C) < . . . < dk (C) ≤ n.
(iii) (Singleton Bound) For all r with 1 ≤ r ≤ k one has dr (C) ≤ n − k + r.

2.2. In this exercise we study relations between the weight hierarchy of AG

codes and the gonality sequence of function fields, cf. Exercise 1.18. Let F/IFq
be a function field of genus g. As usual, we consider divisors D = P1 + . . . + Pn
with n distinct places Pi of degree one, and G with supp D ∩ supp G = ∅. Let
CL (D, G) be the corresponding AG code and set k := dim CL (D, G). Show:
(i) For all r with 1 ≤ r ≤ k one has

dr (CL (D, G)) ≥ n − deg G + γr .

Observe that this is a generalization of Goppa’s Bound for the minimum

distance of AG codes (Theorem 2.2.2).
(ii) Assume in addition that deg G < n. Then

dr (CL (D, G)) = n − k + r for all r with g + 1 ≤ r ≤ k .

2.3. Let C = CL (D, G) be the AG code associated to the divisors G and D =

P1 + . . . + Pn , with n distinct places Pi of degree one and supp D ∩ supp G = ∅.
The integer a := (G − D) is called the abundance of C. Show that the r-th
minimum distance dr = dr (CL (D, G)) of CL (D, G) satisfies the estimate

dr ≥ n − deg G + γr+a ,

where γj denotes the j-th gonality of the function field F/IFq . Conclude the
estimate
d ≥ n − deg G + γ ,
if the abundance of CL (D, G) is ≥ 1 (and γ = γ2 is the gonality of F/IFq ).

2.4. Let C ⊆ IFnq be a rational AG code of dimension k > 0.

(i) Prove that for every m with n + 1 − k ≤ m ≤ n there is a codeword c ∈ C
with wt(c) = m.
(ii) Determine the weight hierarchy of C.

2.5. Let F = IFq (z) be the rational function field over IFq . For α ∈
IFq let Pα
be the zero of z −α, and denote by P∞ the pole of z in F . Let D := α∈IFq Pα
and G = rP∞ with r ≤ (q − 2)/2.
(i) Show that the code CL (D, G) is self-orthogonal.
(ii) For q = 2s and r = (q − 2)/2 show that CL (D, G) is self-dual.
 2.4 Exercises 65

2.6. (i) Take the definition of a BCH code from any textbook on coding
theory, and show that it is equivalent to our Definition 2.3.8.
(ii) The same problem for Goppa codes, see Definition 2.3.10.

2.7. Let C be an [n, k] code over the field IFqm .

(i) Show that dim C|IFq ≥ n − m(n − k) .
(ii) Find non-trivial examples where equality holds.
(iii) Find examples where the minimum distance of C|IFq is larger than the
minimum distance of C.

2.8. (Generalized AG codes) Let F/IFq be a function field of genus g. Assume

that P1 , . . . , Ps ∈ IPF are distinct places and G is a divisor with Pi ∈
/ supp G.
For i = 1, . . . , s let πi : FPi → Ci be an IFq -linear isomorphism from the
residue class field FPi = OPi /Pi onto a linear s code Ci ⊆ IFq , whose param-
ni

eters are [ni , ki = deg Pi , di ]. Let n := i=1 ni and define the linear map
π : L (G) → IFnq by π(f ) := (π1 (f (P1 )), . . . , πs (f (Ps ))). The image of π is
then called a generalized AG code. Formulate and prove estimates for the
dimension and the minimum distance of such codes, analogous to the Goppa
Bound for AG codes (Theorem 2.2.2).

2.9. Let F/IFq be a function field of genus g, and let D = P1 + . . . + Pn with

distinct places Pi of degree 1. Consider a divisor G of the form G = A + B
with A ≥ 0 and B ≥ 0. Assume that Z ≥ 0 is another divisor such that the
following hold:
(1) supp G ∩ supp D = supp Z ∩ supp D = ∅,
(2) (A − Z) = (A) and (B + Z) = (B).
Show that the minimum distance d of the code CΩ (D, G) satisfies

d ≥ deg G − (2g − 2) + deg Z .

Compare with Theorem 2.2.7.

3
Extensions of Algebraic Function Fields

Every function field over K can be regarded as a finite field extension of a

rational function field K(x). This is one of the reasons why it is of interest to
investigate field extensions F  /F of algebraic function fields. In this chapter
we shall study, among other things, the relationship between places, divisors,
Weil differentials and the genera of F  and F . Let us first fix some notation
to be maintained throughout the entire chapter.
F/K denotes an algebraic function field of one variable with full con-
stant field K. The field K is assumed to be perfect; i.e., there does
not exist a purely inseparable extension L/K with 1 < [L : K] < ∞.
We consider function fields F  /K  (where K  is the full constant field
of F  ) such that F  ⊇ F is an algebraic extension and K  ⊇ K. For
convenience, we fix some algebraically closed field Φ ⊇ F and consider
only extensions F  ⊇ F with F  ⊆ Φ.
Actually the perfectness of K will be essential only in a few places of
Chapter 3, in particular in Section 3.6. The fact that we consider only ex-
tensions of F which are contained in Φ is no restriction at all, since Φ is
algebraically closed and thereby every algebraic extension F  /F can be em-
bedded into Φ.
Since this chapter is rather long, we first give a brief survey. In Section
3.1 we introduce the basic concepts: algebraic extensions of function fields,
extensions of places, ramification
 index and residue class degree as well as the
Fundamental Equality ei fi = n.
Subrings of algebraic function fields, in particular holomorphy rings, are
studied in Section 3.2.
In the next section we investigate the integral closure of a subring of F/K
in a finite separable field extension F  /F , and we prove the existence of local
integral bases. This section also contains Kummer’s Theorem which is useful
in determining the decomposition of a place in a finite extension of function
fields.
H. Stichtenoth, Algebraic Function Fields and Codes, 67
Graduate Texts in Mathematics 254,

c Springer-Verlag Berlin Heidelberg 2009
68 3 Extensions of Algebraic Function Fields

Let F  /F be a finite separable extension of algebraic function fields. The

Hurwitz Genus Formula provides a relation between the genus of F , the genus
of F  and the different of F  /F . Section 3.4 is concerned principally with this
result.
In Section 3.5 we study the relation between ramified places and the dif-
ferent (Dedekind’s Different Theorem), and we show how to calculate the
different in specific cases.
Constant field extensions are considered in Section 3.6. The study of such
extensions reduces many problems to the case where the constant field is
algebraically closed (which is often simpler because all places have degree
one).
Section 3.7 is concerned with Galois extensions of algebraic function fields.
For some particular types of Galois extensions F  /F (Kummer and Artin-
Schreier extensions), we determine the genus of F  .
The topic of Section 3.8 is Hilbert’s theory of higher ramification groups,
including Hilbert’s Different Formula.
In Section 3.9 we discuss ramification and splitting of places in the com-
positum of two extensions; one of the main results here is Abhyankar’s Lemma.
In Section 3.10 we consider purely inseparable extensions of an algebraic
function field.
Finally we give in Section 3.11 some upper bounds for the genus of a
function field: Castelnuovo’s Inequality, Riemann’s Inequality, and an estimate
for the genus of the function field of a plane algebraic curve of degree n.
Many results of this chapter (in particular most of Sections 3.1, 3.3, 3.5,
3.7, 3.8 and 3.9) hold not only in the case of algebraic function fields but,
more generally, for extensions of Dedekind domains. Hence the reader who is
familiar with algebraic number theory can skip the appropriate sections.

3.1 Algebraic Extensions of Function Fields

We begin with some basic definitions.

Definition 3.1.1. (a) An algebraic function field F  /K  is called an algebraic

extension of F/K if F  ⊇ F is an algebraic field extension and K  ⊇ K.
(b) The algebraic extension F  /K  of F/K is called a constant field extension
if F  = F K  , the composite field of F and K  .
(c) The algebraic extension F  /K  of F/K is called a finite extension if
[F  : F ] < ∞.

One can also consider arbitrary (not necessarily algebraic) extensions of

function fields. However, we shall restrict ourselves to algebraic extensions
 3.1 Algebraic Extensions of Function Fields 69

since these are by far the most important ones. Arbitrary extensions of func-
tion fields are studied in detail in [7].
We note some simple consequences of the above definitions.
Lemma 3.1.2. Let F  /K  be an algebraic extension of F/K. Then the fol-
lowing hold:
(a) K  /K is algebraic and F ∩ K  = K.
(b) F  /K  is a finite extension of F/K if and only if [K  : K] < ∞.
(c) Let F1 := F K  . Then F1 /K  is a constant field extension of F/K, and
F  /K  is a finite extension of F1 /K  (having the same constant field).

Proof. (a) and (c) are trivial. As to (b), we assume first that F  /K  is a finite
extension of F/K. Then F  can be considered as an algebraic function field
over K whose full constant field is K  . By Corollary 1.1.16 we conclude that
[K  : K] < ∞.
Conversely, suppose that [K  : K] < ∞. Choose x ∈ F \K; then F  /K  (x)
is a finite field extension (since x is transcendental over K  ), and

[K  (x) : K(x)] ≤ [K  : K] < ∞ .

(Actually it holds that [K  (x) : K(x)] = [K  : K], but we do not need this
here.) Therefore

[F  : K(x)] = [F  : K  (x)] · [K  (x) : K(x)] < ∞ .

Since K(x) ⊆ F ⊆ F  , this implies [F  : F ] < ∞.

Now let us study the relation between the places of F and F  .

Definition 3.1.3. Consider an algebraic extension F  /K  of F/K. A place
P  ∈ IPF  is said to lie over P ∈ IPF if P ⊆ P  . We also say that P  is an
extension of P or that P lies under P  , and we write P  |P .

Proposition 3.1.4. Let F  /K  be an algebraic extension of F/K. Suppose

that P (resp. P  ) is a place of F/K (resp. F  /K  ), and let OP ⊆ F (resp.
OP  ⊆ F  ) denote the corresponding valuation ring, vP (resp. vP  ) the corre-
sponding discrete valuation. Then the following assertions are equivalent:
(1) P  |P .
(2) OP ⊆ OP  .
(3) There exists an integer e ≥ 1 such that vP  (x) = e·vP (x) for all x ∈ F .
Moreover, if P  |P then

P = P ∩ F and OP = OP  ∩ F .

For this reason, P is also called the restriction of P  to F .

70 3 Extensions of Algebraic Function Fields

Proof. (1) ⇒ (2): Suppose that P  |P but OP ⊆ OP  . Then there is some

u ∈ F with vP (u) ≥ 0 and vP  (u) < 0. As P ⊆ P  we conclude vP (u) = 0.
Choose t ∈ F with vP (t) = 1, then t ∈ P  and r := vP  (t) > 0. Consequently
vP (ur t) = r · vP (u) + vP (t) = 1 ,
vP  (ur t) = r · vP  (u) + vP  (t) ≤ −r + r = 0 .
Thus ur t ∈ P \P  , a contradiction to P ⊆ P  .
Before proving (2) ⇒ (1) we show the following:
OP ⊆ OP  ⇒ OP = F ∩ OP  . (3.1)
Clearly F ∩OP  is a subring of F with OP ⊆ F ∩OP  , therefore F ∩OP  = OP
or F ∩ OP  = F by Theorem 1.1.13(c). Assume that F ∩ OP  = F ; i.e.,
F ⊆ OP  . Choose an element z ∈ F  \OP  . Since F  /F is algebraic, there is
an equation
z n + cn−1 z n−1 + · · · + c1 z + c0 = 0 (3.2)
with cν ∈ F . We have vP  (z n ) = n · vP  (z) < 0 as z ∈ OP  , therefore
vP  (z n ) < vP  (cν z ν ) for ν = 0, · · · , n − 1 .
The Strict Triangle Inequality yields
vP  (z n + cn−1 z n−1 + · · · + c1 z + c0 ) = n · vP  (z) = vP  (0) .
This contradiction to (3.2) proves (3.1).
(2) ⇒ (1): Now we suppose OP ⊆ OP  . Let y ∈ P ; then y −1 ∈ OP by
Proposition 1.1.5, therefore y −1 ∈ OP  by (3.1). Applying Proposition 1.1.5
once again, we obtain y = (y −1 )−1 ∈ P  , hence P ⊆ P  .
(2) ⇒ (3): Let u ∈ F be an element with vP (u) = 0. Then u, u−1 ∈ OP 
by (2), and so we have vP  (u) = 0. Now choose t ∈ F with vP (t) = 1 and
set e := vP  (t). Since P ⊆ P  it follows that e ≥ 1. Let 0 = x ∈ F and
vP (x) = : r ∈ ZZ. Then vP (xt−r ) = 0, and we obtain
vP  (x) = vP  (xt−r ) + vP  (tr ) = 0 + r · vP  (t) = e · vP (x) .
(3) ⇒ (2): x ∈ OP ⇒ vP (x) ≥ 0 ⇒ vP  (x) = e · vP (x) ≥ 0 ⇒ x ∈ OP  .
So we have proved the equivalences (1) ⇐⇒ (2) ⇐⇒ (3), and OP =
OP  ∩ F if P  |P . The assertion P = P  ∩ F is now trivial (for example, from
(3)).

A consequence of the preceding proposition is that for P  |P there is a

canonical embedding of the residue class field FP = OP /P into the residue
class field FP  = OP  /P  , given by
x(P ) → x(P  ) for x ∈ OP .
Therefore we can consider FP as a subfield of FP  .
 3.1 Algebraic Extensions of Function Fields 71

Definition 3.1.5. Let F  /K  be an algebraic extension of F/K, and let P  ∈

IPF  be a place of F  /K  lying over P ∈ IPF .
(a) The integer e(P  |P ) := e with

vP  (x) = e · vP (x) for all x ∈ F

is called the ramification index of P  over P . We say that P  |P is ramified if

e(P  |P ) > 1, and P  |P is unramified if e(P  |P ) = 1.
(b) f (P  |P ) := [FP  : FP ] is called the relative degree of P  over P .

Note that f (P  |P ) can be finite or infinite; the ramification index is always

a natural number.

Proposition 3.1.6. Let F  /K  be an algebraic extension of F/K and let P 

be a place of F  /K  lying over P ∈ IPF . Then
(a) f (P  |P ) < ∞ ⇐⇒ [F  : F ] < ∞.
(b) If F  /K  is an algebraic extension of F  /K  and P  ∈ IPF  is an exten-
sion of P  , then
e(P  |P ) = e(P  |P  ) · e(P  |P ) ,
f (P  |P ) = f (P  |P  ) · f (P  |P ) .

Proof. (a) Consider the natural embeddings K ⊆ FP ⊆ FP  and K ⊆ K  ⊆ FP  ,

where [FP : K] < ∞ and [FP  : K  ] < ∞. It follows that

[FP  : FP ] < ∞ ⇐⇒ [K  : K] < ∞ .

The latter condition is equivalent to [F  : F ] < ∞ by Lemma 3.1.2.

(b) The assertion regarding ramification indices follows trivially from the
definitions, and f (P  |P ) = f (P  |P  ) · f (P  |P ) follows from the inclusions
FP ⊆ FP  ⊆ FP .

Next we investigate the existence of extensions of places in extensions of

function fields.

Proposition 3.1.7. Let F  /K  be an algebraic extension of F/K.

(a) For each place P  ∈ IPF  there is exactly one place P ∈ IPF such that P  |P ,
namely P = P  ∩ F .
(b) Conversely, every place P ∈ IPF has at least one, but only finitely many
extensions P  ∈ IPF  .
72 3 Extensions of Algebraic Function Fields

Proof. (a) The main step of the proof is the following.

Claim. There is some z ∈ F, z = 0, with vP  (z) = 0 . (3.3)
Assume this is false. Choose t ∈ F  with vP  (t) > 0. Since F  /F is algebraic,
there is an equation
cn tn + cn−1 tn−1 + · · · + c1 t + c0 = 0
with ci ∈ F, c0 = 0 and cn = 0. By assumption we have vP  (c0 ) = 0 and
vP  (ci ti ) = vP  (ci ) + i · vP  (t) > 0 for i = 1, . . . , n, a contradiction to the Strict
Triangle Inequality. Thus (3.3) is proved.
We set O := OP  ∩ F and P := P  ∩ F . It is obvious by (3.3) that O is a
valuation ring of F/K and that P is the corresponding place. The uniqueness
assertion is trivial.
(b) Now a place P of F/K is given. Choose x ∈ F \K whose only zero is
P (this is possible by Proposition 1.6.6). We claim that for P  ∈ IPF  the
following holds:
P  |P ⇐⇒ vP  (x) > 0 . (3.4)
Since x has at least one but only finitely many zeros in F  /K  , assertion (b)
is an immediate consequence of (3.4).
Now we prove (3.4). If P  |P then vP  (x) = e(P  |P ) · vP (x) > 0. Conversely,
assume that vP  (x) > 0. Let Q denote the place of F/K which lies under P 
(here we use (a)). Then vQ (x) > 0, so Q = P since P is the only zero of x in
F/K.

The preceding proposition enables us to define a homomorphism from the

divisor group Div(F ) into Div(F  ).
Definition 3.1.8. Let F  /K  be an algebraic extension of F/K. For a place
P ∈ IPF we define its conorm (with respect to F  /F ) as

ConF /F (P ) := e(P  |P ) · P  ,
P  |P

where the sum runs over all places P ∈ IPF  lying over P . The conorm map
is extended to a homomorphism from Div(F ) to Div(F  ) by setting
  
ConF /F nP · P := nP · ConF /F (P ) .

The conorm behaves well in towers of function fields F  ⊇ F  ⊇ F ; an

immediate consequence of Proposition 3.1.6(b) is the formula
ConF /F (A) = ConF /F  (ConF /F (A))
for every divisor A ∈ Div(F ).
Another nice property of the conorm is that it sends principal divisors of
F to principal divisors of F  . More precisely we have:
 3.1 Algebraic Extensions of Function Fields 73

Proposition 3.1.9. Let F  /K  be an algebraic extension of the function field


F F
F/K. For 0 = x ∈ F let (x)F F
0 , (x)∞ , (x)
F
resp. (x)F
0 , (x)∞ , (x) denote
the zero, pole, principal divisor of x in Div(F ) resp. in Div(F  ). Then
  
ConF /F ((x)F F F F F F
0 ) = (x)0 , ConF /F ((x)∞ ) = (x)∞ , and ConF /F ((x) ) = (x) .

Proof. From the definition of the principal divisor of x follows that

   
(x)F = vP  (x) · P  = e(P  |P ) · vP (x) · P 
P  ∈IPF  P ∈IPF P  |P
   
= vP (x) · ConF /F (P ) = ConF /F vP (x) · P
P ∈IPF P ∈IPF

= ConF /F ((x)F ) .

Considering only the positive (negative) part of the principal divisor, we ob-
tain the corresponding result for the zero (pole) divisor of x.

By this proposition the conorm induces a homomorphism (again denoted

as conorm) of the divisor class groups

ConF /F : Cl(F ) → Cl(F  ) .

This map is in general neither injective nor surjective (whereas the map
ConF /F : Div(F ) → Div(F  ) is trivially injective).
One of our next goals is to find a relation between the degrees of a divisor
A ∈ Div(F ) and of its conorm in Div(F  ) in the case of a finite extension
F  /F (the general case will be considered in Section 3.6). To this end we first
prove a lemma.

Lemma 3.1.10. Let K  /K be a finite field extension and let x be transcen-

dental over K. Then

[K  (x) : K(x)] = [K  : K] .

Proof. We can assume that K  = K(α) for some element α ∈ K  . Clearly

[K  (x) : K(x)] ≤ [K  : K] since K  (x) = K(x)(α). As for the reverse in-
equality we have to prove that the irreducible polynomial ϕ(T ) ∈ K[T ] of α
over K remains irreducible over the field K(x). Suppose that this is false, so
ϕ(T ) = g(T ) · h(T ) with monic polynomials g(T ), h(T ) ∈ K(x)[T ] of degree
< deg ϕ. Since ϕ(α) = 0 we have w.l.o.g. g(α) = 0. We write

g(T ) = T r + cr−1 (x)T r−1 + · · · + c0 (x)

with ci (x) ∈ K(x) and r < deg ϕ; then

74 3 Extensions of Algebraic Function Fields

αr + cr−1 (x)αr−1 + · · · + c0 (x) = 0 .

Multiplying by a common denominator we obtain

gr (x) · αr + gr−1 (x) · αr−1 + · · · + g0 (x) = 0 (3.5)

for certain polynomials gi (x) ∈ K[x], and we can assume that not all gi (x)
are divisible by x. Setting x = 0 in (3.5) yields a non-trivial equation for α
over K of degree less than deg ϕ, a contradiction.
Theorem 3.1.11 (Fundamental Equality). Let F  /K  be a finite exten-
sion of F/K, let P be a place of F/K and let P1 , . . . , Pm be all the places
of F  /K  lying over P . Let ei := e(Pi |P ) denote the ramification index and
fi := f (Pi |P ) the relative degree of Pi |P . Then

m
ei fi = [F  : F ] .
i=1

Proof. Choose x ∈ F such that P is the only zero of x in F/K, and let
vP (x) =: r > 0. Then the places P1 , . . . , Pm ∈ IPF  are exactly the zeros of
x in F  /K  by (3.4). Now we evaluate the degree [F  : K(x)] in two different
ways:

[F  : K(x)] = [F  : K  (x)] · [K  (x) : K(x)]

 m 
= vPi (x) · deg Pi · [K  : K]
i=1

m
= (ei · vP (x)) · ([FP i : K  ] · [K  : K])
i=1

m
=r· ei · [FP i : FP ] · [FP : K]
i=1

m
= r · deg P · ei fi . (3.6)
i=1

(The second line in the above equations follows from Lemma 3.1.10. The fact
that the degree of the zero divisor of x in F  /K  equals [F  : K  (x)] is a
consequence of Theorem 1.4.11.) On the other hand,

[F  : K(x)] = [F  : F ] · [F : K(x)] = [F  : F ] · r · deg P , (3.7)

since rP is the zero divisor of x in F/K. Comparing (3.6) and (3.7) one obtains
the desired result.

Corollary 3.1.12. Let F  /K  be a finite extension of F/K and P ∈ IPF .

Then we have:
 3.1 Algebraic Extensions of Function Fields 75
 
(a)  {P  ∈ IPF  ; P  lies over P }  ≤ [F  : F ].
(b) If P  ∈ IPF  lies over P then e(P  |P ) ≤ [F  : F ] and f (P  |P ) ≤ [F  : F ].

According to Corollary 3.1.12 the following definition makes sense:

Definition 3.1.13. Let F  /K  be an extension of F/K of degree [F  : F ]= n

and let P ∈ IPF .
(a) P splits completely in F  /F if there are exactly n distinct places P  ∈ IPF 
with P  |P .
(b) P is totally ramified in F  /F if there is a place P  ∈ IPF  with P  |P and
e(P  |P ) = n.

By the Fundamental Equality it is clear that a place P ∈ IPF splits com-

pletely in F  /F if and only if e(P  |P ) = f (P  |P ) = 1 for all places P  |P in
F  . If P is totally ramified in F  /F then there is exactly one place P  ∈ IPF 
with P  |P .
As a consequence of the Fundamental Equality we obtain:

Corollary 3.1.14. Let F  /K  be a finite extension of F/K. Then for each

divisor A ∈ Div(F ),

[F  : F ]
deg ConF /F (A) = · deg A .
[K  : K]

Proof. It is sufficient to consider a prime divisor A = P ∈ IPF . We have

 
deg ConF /F (P ) = deg e(P  |P ) · P 
P  |P

= e(P  |P ) · [FP  : K  ]
P  |P
 [FP  : K]
= e(P  |P ) ·
[K  : K]
P  |P
1 
= · e(P  |P ) · [FP  : FP ] · [FP : K]
[K 
: K] 
P |P
 
1  
= · e(P |P ) · f (P |P ) · deg P
[K  : K]  P |P

[F  : F ]
= · deg P (by Theorem 3.1.11) .
[K  : K]
76 3 Extensions of Algebraic Function Fields

The foregoing results can be used to prove a very useful criterion for ir-
reducibility of certain polynomials over a function field. A special case of the
following proposition is known as Eisenstein’s Irreducibility Criterion.

Proposition 3.1.15. Consider a function field F/K and a polynomial

ϕ(T ) = an T n + an−1 T n−1 + . . . + a1 T + a0

with coefficients ai ∈ F . Assume that there exists a place P ∈ IPF such that
one of the following conditions (1) or (2) holds:
(1) vP (an ) = 0 , vP (ai ) ≥ vP (a0 ) > 0 for i = 1, . . . , n − 1 , and
gcd(n, vP (a0 )) = 1.
(2) vP (an ) = 0 , vP (ai ) ≥ 0 for i = 1, . . . , n − 1 , vP (a0 ) < 0 , and
gcd(n, vP (a0 )) = 1.
Then ϕ(T ) is irreducible in F [T ]. If F  = F (y) where y is a root of ϕ(T ),
then P has a unique extension P  ∈ IPF  , and we have e(P  |P ) = n and
f (P  |P ) = 1 (i.e., P is totally ramified in F (y)/F ).

Proof. We consider an extension field F  = F (y) with ϕ(y) = 0. The degree

of F  /F is [F  : F ] ≤ deg ϕ(T ) = n, with equality if and only if ϕ(T ) is
irreducible in F [T ]. Choose an extension P  ∈ IPF  of P . As ϕ(y) = 0,

−an y n = a0 + a1 y + . . . + an−1 y n−1 . (3.8)

First we assume (1). From vP  (an ) = 0 and vP  (ai ) > 0 for i = 1, . . . , n − 1

it follows easily that vP  (y) > 0. Setting e := e(P  |P ), we have vP  (a0 ) =
e · vP (a0 ) and vP  (ai y i ) = e · vP (ai ) + i · vP  (y) > e · vP (a0 ) for i = 1, . . . , n − 1.
By the Strict Triangle Inequality, (3.8) implies

n · vP  (y) = e · vP (a0 ) .

As gcd(n, vP (a0 )) = 1 by assumption (1), we conclude that n|e and therefore

n ≤ e. On the other hand, n ≥ [F  : F ] ≥ e by Corollary 3.1.12. So we obtain

n = e = [F  : F ] . (3.9)

All assertions of Proposition 3.1.15 follow now immediately from (3.9) and
Theorem 3.1.11.
The proof is similar in the case when one assumes (2) instead of (1).

Before we can proceed further with the theory of extensions of algebraic

function fields, we have to study certain subrings of a function field.
 3.2 Subrings of Function Fields 77

3.2 Subrings of Function Fields

As before, F/K denotes a function field with constant field K.

Definition 3.2.1. A subring of F/K is a ring R such that K ⊆ R ⊆ F , and

R is not a field.

In particular, if R is a subring of F/K then K  R  F . Here are two

typical examples:
(a) R = OP for some P ∈ IPF .
(b) R = K[x1 , . . . , xn ] where x1 , . . . , xn ∈ F \K.
While OP is obviously a subring, to see that K[x1 , . . . , xn ] is also a subring,
we have to show that it is not a field. To this end, choose a place P ∈ IPF
such that vP (x1 ) ≥ 0, . . . , vP (xn ) ≥ 0. Let x = x1 and d := deg P . As the
residue classes 1, x(P ), . . . , xd (P ) ∈ OP /P are linearly dependent over K one
can find α0 , . . . , αd ∈ K such that the element z = α0 + α1 x + . . . + αd xd is
not 0 but vP (z) > 0 (observe that x is transcendental over K since x ∈ K).
Clearly z ∈ K[x1 , . . . , xn ] but z −1 ∈ K[x1 , . . . , xn ] (since vP (y) ≥ 0 for every
y ∈ K[x1 , . . . , xn ]).
A more general class of subrings than (a) is given by the following
definition.

Definition 3.2.2. For ∅ = S  IPF let

OS := {z ∈ F | vP (z) ≥ 0 for all P ∈ S}

be the intersection of all valuation rings OP with P ∈ S. A ring R ⊆ F which

is of the form R = OS for some S  IPF , S = ∅ is called a holomorphy ring
of F/K.

For instance, the ring K[x] is a holomorphy ring of the rational function
field K(x)/K. One checks easily that

K[x] = OP ,
P =P∞

where P∞ denotes the unique pole of x in K(x).

We note some simple consequences of Definition 3.2.2.

Lemma 3.2.3. (a) Every valuation ring OP is a holomorphy ring, namely

OP = OS with S = {P }.
(b) Every holomorphy ring OS is a subring of F/K.
78 3 Extensions of Algebraic Function Fields

(c) For P ∈ IPF and ∅ = S  IPF we have

OS ⊆ OP ⇐⇒ P ∈ S .
Consequently, OS = OT ⇐⇒ S = T .

Proof. (b) Since OS is a ring with K ⊆ OS ⊆ F we have only to show that it

is not a field. Choose a place P1 ∈ S. As S = IPF , the Strong Approximation
Theorem yields an element 0 = x ∈ F such that
vP1 (x) > 0 and vP (x) ≥ 0 for all P ∈ S .
Obviously x ∈ OS but x−1 ∈ OS ; therefore OS is not a field.
(c) Suppose P ∈ S. By the Strong Approximation Theorem we can find z ∈ F
with
vP (z) < 0 and vQ (z) ≥ 0 for all Q ∈ S . (3.10)
(This is clear if S ∪ {P } = IPF . If however S ∪ {P } = IPF , choose z ∈ OS
which has at least one zero in S; since z must have some pole it follows that
vP (z) < 0.) Each element z satisfying (3.10) is in OS but not in OP . Thus
we have proved that P ∈ S implies OS ⊆ OP . The remaining assertions are
trivial.
Definition 3.2.4. Let R be a subring of F/K.
(a) An element z ∈ F is said to be integral over R if f (z) = 0 for some monic
polynomial f (X) ∈ R[X]; i.e., if there are a0 , . . . , an−1 ∈ R such that
z n + an−1 z n−1 + · · · + a1 z + a0 = 0 .
Such an equation is called an integral equation for z over R.
(b) The set
icF (R) := {z ∈ F | z is integral over R}
is called the integral closure of R in F .
(c) Let F0 ⊆ F denote the quotient field of R. The ring R is called integrally
closed if icF0 (R) = R; i.e., every element z ∈ F0 which is integral over R is
already in R.

Proposition 3.2.5. Let OS be a holomorphy ring of F/K. Then

(a) F is the quotient field of OS .
(b) OS is integrally closed.
Proof. (a) Let x ∈ F, x = 0. Choose a place P0 ∈ S. By the Strong Approxi-
mation Theorem there is an element z ∈ F such that
vP0 (z) = max{0, vP0 (x−1 ) and vP (z) ≥ max{0, vP (x−1 )} for all P ∈ S .
Clearly z ∈ OS , z = 0 and y := zx ∈ OS , so x = yz −1 is in the quotient field
of OS .
 3.2 Subrings of Function Fields 79

(b) Let u ∈ F be integral over OS . Choose an integral equation

un + an−1 un−1 + · · · + a0 = 0 (3.11)

with ai ∈ OS . We have to show that vP (u) ≥ 0 for all P ∈ S. Suppose this is

false, so that vP (u) < 0 for some P ∈ S. Since vP (ai ) ≥ 0,

vP (un ) = n · vP (u) < vP (ai ui )

for i = 0, . . . , n − 1. Thus the Strict Triangle Inequality yields a contradiction

to (3.11).
Theorem 3.2.6. Let R be a subring of F/K and

S(R) := {P ∈ IPF | R ⊆ OP } .

Then the following hold:

(a) ∅ = S(R)  IPF .
(b) The integral closure of R in F is icF (R) = OS(R) . In particular, icF (R)
is an integrally closed subring of F/K with quotient field F .
Proof. (a) Since R is not a field we can find a proper ideal I  R, and by
Theorem 1.1.19 there exists a place P ∈ IPF such that I ⊆ P and R ⊆ OP .
Therefore S(R) = ∅. On the other hand, consider an element x ∈ R which
is transcendental over K. Each place Q ∈ IPF which is a pole of x is not in
S(R), so S(R) = IPF .
(b) Since R ⊆ OS(R) and OS(R) is integrally closed (by Proposition 3.2.5) it
follows immediately that icF (R) ⊆ OS(R) . In order to prove the inclusion in
the reverse direction, consider an element z ∈ OS(R) . We claim:

z −1 · R[z −1 ] = R[z −1 ] . (3.12)

Suppose that (3.12) is false; i.e., z −1 R[z −1 ] is a proper ideal in R[z −1 ]. By

Theorem 1.1.19 we find a place Q ∈ IPF such that

R[z −1 ] ⊆ OQ and z −1 ∈ Q .

It follows that Q ∈ S(R) and z ∈ OQ which is a contradiction to z ∈ OS(R) ;

thus we have proved (3.12). From (3.12) we obtain a relation

s
1 = z −1 · ai (z −1 )i (3.13)
i=0

with a0 , · · · , as ∈ R. Multiplying (3.13) by z s+1 yields


s
z s+1 − ai z s−i = 0 .
i=0

This is an integral equation for z over R.

80 3 Extensions of Algebraic Function Fields

Remark 3.2.7. Note that in the proof of Theorem 3.2.6 we did not actually
use the assumption that K is the full constant field of F . Thus the theorem
remains true if we only assume that F/K is a function field – the constant
field K̃ of F/K may be larger than K.

An easy consequence of Proposition 3.2.5 and Theorem 3.2.6 is

Corollary 3.2.8. A subring R of F/K with quotient field F is integrally
closed if and only if R is a holomorphy ring.
Proposition 3.2.9. Let OS be a holomorphy ring of F/K. Then there is a
1–1-correspondence between S and the set of maximal ideals of OS , given by
P −→ MP := P ∩ OS (for P ∈ S) .
Moreover, the map

OS /MP −→ FP = OP /P ,
ϕ :
x + MP −→ x+P

is an isomorphism.
Proof. Consider for P ∈ S the ring homomorphism

OS −→ FP ,
φ :
x −→ x + P .

We claim that φ is surjective. In fact, let z + P ∈ FP with z ∈ OP . By the

Strong Approximation Theorem there is some x ∈ F satisfying
vP (x − z) > 0 and vQ (x) ≥ 0 for all Q ∈ S\{P } .
Then x ∈ OS and φ(x) = z + P . The kernel of φ is MP = P ∩ OS , hence
φ induces an isomorphism ϕ : OS /MP → FP . Since FP is a field, MP is a
maximal ideal of OS . If P = Q, the Strong Approximation Theorem shows
that MP = MQ .
It remains to prove that each maximal ideal of OS can be written as P ∩OS
for some P ∈ S. Let M ⊆ OS be a maximal ideal. By Theorem 1.1.19 there
is a place P ∈ IPF with
M ⊆P and OS ⊆ OP .
Lemma 3.2.3(c) shows that P ∈ S. Since M ⊆ P ∩ OS and M is a maximal
ideal of OS , we obtain M = P ∩ OS .

We know by Theorem 1.1.6 that a valuation ring OP of F/K is a principal

ideal domain (i.e., every ideal of OP is principal). In general holomorphy rings
are no longer principal ideal domains. However, the following generalization
of Theorem 1.1.6 holds:
 3.3 Local Integral Bases 81

Proposition 3.2.10. If S ⊆ IPF is a non-empty finite set of places of F/K,

then OS is a principal ideal domain.

Proof. Let S = {P1 , . . . , Ps } and let {0} = I ⊆ OS be an ideal of OS . For

i = 1, . . . , s choose xi ∈ I such that
vPi (xi ) = : ni ≤ vPi (u) for all u ∈ I .
(This is possible since vPi (u) ≥ 0 for all u ∈ I and I = {0}.) By the Approxi-
mation Theorem we can find zi ∈ F such that
vPi (zi ) = 0 andvPj (zi ) > nj for j = i .
s
Clearly zi ∈ OS , therefore the element x := i=1 xi zi is in I. By the Strict
Triangle Inequality we have vPi (x) = ni for i = 1, . . . , s. Our proposition will
be proved when we can show that I ⊆ xOS . Consider an element z ∈ I. Set
y := x−1 z, then
vPi (y) = vPi (z) − ni ≥ 0 for i = 1, . . . , s .
Consequently y ∈ OS and z = xy ∈ xOS .

3.3 Local Integral Bases

In this section we investigate the integral closure of a subring of F/K in an
extension field of F . We consider the following situation:
F/K is a function field with constant field K, and F  ⊇ F is a finite
field extension (the constant field K  of F  may be larger than K).

Proposition 3.3.1. Let R be an integrally closed subring of F/K such that

F is the quotient field of R (i.e., R is a holomorphy ring of F/K). For z ∈ F 
let ϕ(T ) ∈ F [T ] denote its minimal polynomial over F . Then we have:
z is integral over R ⇐⇒ ϕ(T ) ∈ R[T ] .
Proof. By definition, ϕ(T ) is the unique irreducible monic polynomial with
coefficients in F such that ϕ(z) = 0. If ϕ(T ) ∈ R[T ] then z is clearly integral
over R.
The converse is not so evident. In fact, one has to use the assumption
that R is integrally closed. So consider an element z ∈ F  which is integral
over R. Choose a monic polynomial f (T ) ∈ R[T ] with f (z) = 0. Since ϕ(T )
is the minimal polynomial of z over F there is some ψ(T ) ∈ F [T ] such that
f (T ) = ϕ(T ) · ψ(T ). Let F  ⊇ F  be a finite extension field of F containing
all roots of ϕ, and R = icF  (R) be the integral closure of R in F  . Since all
roots of ϕ are roots of f as well, they are in R . The coefficients of ϕ(T ) are
polynomial expressions of the roots of ϕ, so ϕ(T ) ∈ R [T ]. But ϕ(T ) ∈ F [T ]
and F ∩ R = R since R is integrally closed. Therefore ϕ(T ) ∈ R[T ].
82 3 Extensions of Algebraic Function Fields

Corollary 3.3.2. Notation as in Proposition 3.3.1. Let TrF /F : F  → F de-

note the trace map from F  to F and let x ∈ F  be integral over R. Then
TrF /F (x) ∈ R.

This corollary follows easily from well-known properties of the trace map-
ping. Let us briefly recall some of these properties which will be of use in
the sequel. We consider a finite field extension M/L of degree n. If M/L is
not separable, the trace map TrM/L : M → L is the zero map. Hence we can
assume from now on that M/L is separable. In this case, TrM/L : M → L is
an L-linear map which is not identically zero. It can be described as follows.
Choose an algebraically closed field Ψ ⊇ L. An embedding of M/L into Ψ is
a field homomorphism σ : M → Ψ such that σ(a) = a for all a ∈ L. Since
M/L is separable there are exactly n distinct embeddings σ1 , . . . , σn of M/L
into Ψ , and we have for x ∈ M

n
TrM/L (x) = σi (x) .
i=1

If ϕ(T ) = T r + ar−1 T r−1 + . . . + a◦ ∈ L[T ] is the minimal polynomial of x

over L, then

TrM/L (x) = −sar−1 , where s := [M : L(x)] . (3.14)

The trace behaves well in towers of fields, i.e.

TrH/L (x) = TrM/L (TrH/M (x)) (3.15)

whenever H ⊇ M ⊇ L.
Note that Corollary 3.3.2 is an immediate consequence of (3.14) and
Proposition 3.3.1.

Proposition 3.3.3. Let M/L be a finite separable field extension, and con-
sider a basis {z1 , . . . , zn } of M/L. Then there are uniquely determined ele-
ments z1∗ , . . . , zn∗ ∈ M , such that

TrM/L (zi zj∗ ) = δij .

(δij denotes the Kronecker symbol.) The set {z1∗ , . . . , zn∗ } is a basis of M/L as
well; it is called the dual basis of {z1 , . . . , zn } (with respect to the trace).

Proof. We consider the dual space M ∧ of M over L; i.e., M ∧ is the space

of all L-linear maps λ : M → L. It is well-known from linear algebra that
M ∧ is an n-dimensional vector space over L. For z ∈ M and λ ∈ M ∧ define
z · λ ∈ M ∧ by (z · λ)(w) := λ(zw). This turns M ∧ into a vector space over
M of dimension one (as dimL (M ∧ ) = [M : L] · dimM (M ∧ )). Since TrM/L is
 3.3 Local Integral Bases 83

not the zero map, every λ ∈ M ∧ has a unique representation λ = z · TrM/L .

In particular the linear forms λj ∈ M ∧ given by λj (zi ) := δij (i = 1, . . . , n)
can be written as λj = zj∗ · TrM/L with zj∗ ∈ M . This means that

TrM/L (zi zj∗ ) = (zj∗ · TrM/L )(zi ) = λj (zi ) = δij .

As λ1 , . . . , λn are linearly independent over L, the same holds for z1∗ , . . . , zn∗ ,
and hence they constitute a basis of M/L.

Our next result holds essentially without the assumption of separability.

The proof however is simpler under this additional hypothesis, and later on
we shall need the result only for separable extensions.

Theorem 3.3.4. Let R be an integrally closed subring of F/K with quotient

field F , and F  /F be a finite separable extension of degree n. Let R = icF  (R)
denote the integral closure of R in F  . Then we have:
(a) For every basis {x1 , . . . , xn } of F  /F there are elements ai ∈ R\{0} such
that a1 x1 , . . . , an xn ∈ R . Consequently there exist bases of F  /F which are
contained in R .
(b) If {z1 , . . . , zn } ⊆ R is a basis of F  /F and {z1∗ , . . . , zn∗ } denotes the dual
basis with respect to the trace map then

n 
n
Rzi ⊆ R ⊆ Rzi∗ .
i=1 i=1

(c) If in addition R is a principal ideal domain, then there exists a basis

{u1 , . . . , un } of F  /F with the property


n
R = Rui .
i=1

Proof. (a) It must be shown that for every x ∈ F  there is some element
0 = a ∈ R such that ax satisfies an integral equation over R. Since F  /F is
algebraic and F is the quotient field of R, there are elements ai , bi ∈ R with
ai = 0 and
br−1 r−1 b1 b0
xr + x + ··· + x + = 0.
ar−1 a1 a0
Multiplying this equation by ar , where a := a0 · a1 · . . . · ar−1 , we obtain

(ax)r + cr−1 (ax)r−1 + . . . + c1 (ax) + c0 = 0

with ci ∈ R, so ax ∈ R .
84 3 Extensions of Algebraic Function Fields

(b) Now {z1 , . . . , zn } is a basis of F  /F such that all zi ∈ R , and {z1∗ , . . . , zn∗ }
is the dual basis. In particular each z ∈ F  can be represented in the form
z = e1 z1∗ + . . . + en zn∗ with ei ∈ F .
If z ∈ R then zzj ∈ R for j = 1, . . . , n, consequently TrF /F (zzj ) ∈ R by
Corollary 3.3.2. Since

n  n
TrF /F (zzj ) = TrF /F ei zj zi∗ = ei · TrF /F (zj zi∗ ) = ej ,
i=1 i=1
n
we conclude that ej ∈ R, hence R ⊆ i=1 Rzi∗ .
n
(c) Choose a basis {w1 , . . . , wn } of F  /F with R ⊆ i=1 Rwi (this is possible
by (b)). For 1 ≤ k ≤ n set

k
Rk := R ∩ Rwi . (3.16)
i=1
k
We want to construct recursively u1 , . . . , un such that Rk = i=1 Rui . For
k = 1 (i.e., R1 = R ∩ Rw1 ) consider the set
I1 := {a ∈ F | aw1 ∈ R } .
n
It is contained in R, since R ⊆ i=1 Rwi . Actually I1 is an ideal of R,
hence I1 = a1 R for some a1 ∈ R (as R is a principal ideal domain). Setting
u1 := a1 w1 it is easily verified that R1 = Ru1 .
Suppose now that for k ≥ 2 we have already found u1 , . . . , uk−1 such that
k−1
Rk−1 = i=1 Rui . Let
Ik := {a ∈ F | there are b1 , . . . , bk−1 ∈ R
with b1 w1 + . . . + bk−1 wk−1 + awk ∈ R } .
Again Ik is an ideal of R, say Ik = ak R. Choose uk ∈ R with
uk = c1 w1 + . . . + ck−1 wk−1 + ak wk .
k
Clearly Rk ⊇ i=1 Rui . In order to prove the reverse inclusion, let w ∈ Rk .
Write
w = d1 w1 + . . . + dk wk with di ∈ R .
Then dk ∈ Ik , hence dk = dak with d ∈ R and

k−1 
k−1
w − duk ∈ R ∩ Rwi = Rk−1 = Rui .
i=1 i=1
k
Therefore w ∈ i=1 Rui .
n
We have proved that R = Rn = i=1 Rui . Since R contains some basis
of F  /F by (a), the elements u1 , . . . , un are linearly independent over F and
constitute a basis of F  /F .
 3.3 Local Integral Bases 85

Corollary 3.3.5. Let F  /F be a finite separable extension of the function field

F/K and let P ∈ IPF be a place of F/K. Then the integral closure OP of OP
in F  is 
OP = OP  .
P  |P

There exists a basis {u1 , . . . , un } of F  /F such that


n
OP = OP · ui .
i=1

Every such basis {u1 , . . . , un } is called an integral basis of OP over OP (or
a local integral basis of F  /F for the place P ).

Proof. This is clear by Theorem 3.2.6(b), Remark 3.2.7 and Theorem 3.3.4
(observe that OP is a principal ideal domain).

An important supplement to the existence of local integral bases is given

in the next theorem.

Theorem 3.3.6. Let F/K be a function field, F  /F be a finite separable ex-

tension field. Then each basis {z1 , . . . .zn } of F  /F is an integral basis for
almost all (i.e., all but finitely many) places P ∈ IPF .

Proof. We consider the dual basis {z1∗ , . . . , zn∗ } of {z1 , . . . , zn }. The minimal
polynomials of z1 , . . . , zn , z1∗ , . . . , zn∗ over F involve only finitely many coeffi-
cients. Let S ⊆ IPF be the set of all poles of these coefficients. S is finite, and
for P ∈ S we have
z1 , . . . , zn , z1∗ , . . . , zn∗ ∈ OP , (3.17)
where OP = icF  (OP ). Therefore
  
OP · zi ⊆ OP ⊆ OP · zi∗ ⊆ OP ⊆ OP · zi .

The first and third of these inclusions are obvious by (3.17), the second and
fourth follow immediately from Theorem 3.3.4(b) (note that {z1 , . . . , zn } is
the dual basis of {z1∗ , . . . , zn∗ }). Thus {z1 , . . . , zn } is an integral basis for each
P ∈ S.

Next we want to describe a method which can often be used to determine

all extensions of a place P ∈ IPF in F  . For convenience we introduce some
notation.
F̄ := FP is the residue class field of P .
ā := a(P ) ∈ F̄ is the residue class of a ∈ OP .

If ψ(T ) = ci T i is a polynomial with coefficients ci ∈ OP , we set

ψ̄(T ) := c̄i T i ∈ F̄ [T ] .
86 3 Extensions of Algebraic Function Fields

Obviously every polynomial γ(T ) ∈ F̄ [T ] can be represented as γ(T ) = ψ̄(T )

with ψ(T ) ∈ OP [T ] and deg ψ(T ) = deg γ(T ). With these notations we have
the following theorem.

Theorem 3.3.7 (Kummer). Suppose that F  = F (y) where y is integral

over OP , and consider the minimal polynomial ϕ(T ) ∈ OP [T ] of y over F .
Let

r
ϕ̄(T ) = γi (T )εi
i=1

be the decomposition of ϕ̄(T ) into irreducible factors over F̄ (i.e., the polyno-
mials γ1 (T ), . . . , γr (T ) are irreducible, monic, pairwise distinct in F̄ [T ] and
εi ≥ 1). Choose monic polynomials ϕi (T ) ∈ OP [T ] with

ϕ̄i (T ) = γi (T ) and deg ϕi (T ) = deg γi (T ) .

Then for 1 ≤ i ≤ r, there are places Pi ∈ IPF  satisfying

Pi |P , ϕi (y) ∈ Pi and f (Pi |P ) ≥ deg γi (T ) .

Moreover Pi = Pj for i = j.
Under additional assumptions one can prove more. Suppose that at least
one of the following hypotheses (∗) resp. (∗∗) is satisfied:

εi = 1 for i = 1, . . . , r , or (∗)

{1, y, . . . , y n−1 } is an integral basis for P . (∗∗)

Then there exists, for 1 ≤ i ≤ r, exactly one place Pi ∈ IPF  with Pi |P and
ϕi (y) ∈ Pi . The places P1 , . . . , Pr are all the places of F  lying over P , and
we have
r
ConF /F (P ) = εi P i ;
i=1

i.e., εi = e(Pi |P ). The residue class field FP i = OPi /Pi is isomorphic to
F̄ [T ]/(γi (T )), hence f (Pi |P ) = deg γi (T ).

Proof. We set F̄i := F̄ [T ]/(γi (T )). Since γi (T ) is irreducible, F̄i is an exten-

sion field of F̄ of degree

[F̄i : F̄ ] = deg γi (T ) . (3.18)

n−1
Consider the ring OP [y] = j=0 OP · y j where n = deg ϕ(T ) = [F  : F ].
There are ring homomorphisms

OP [T ] −→ OP [y] ,
ρ :  
cj T j −→ cj y j ,
 3.3 Local Integral Bases 87

and 
OP [T ] −→ F̄i ,
πi :  
cj T j −→ c̄j T j mod γi (T ) .
The kernel of ρ is the ideal generated by ϕ(T ). Since

πi (ϕ(T )) = ϕ̄(T ) mod γi (T ) = 0 ,

we have Ker(ρ) ⊆ Ker(πi ). We conclude that there is a unique homomorphism

σi : OP [y] → F̄i with πi = σi ◦ ρ. It is explicitly given by

OP [y] −→ F̄i ,
σi : n−1 n−1
c
j=0 j y j
−
 → j
j=0 j T mod γi (T ) .
c̄

From this it is obvious that σi is an epimorphism. We claim that its kernel is

Ker(σi ) = P · OP [y] + ϕi (y) · OP [y] . (3.19)

The inclusion Ker(σi ) ⊇ P · OP [y] + ϕi (y) · OP [y] is trivial. In order to

n−1
j=0 cj y ∈ Ker(σi ). Then
j
show the reverse inclusion, consider an element
n−1
j=0 c̄j T = ϕ̄i (T ) · ψ̄(T ) for some ψ(T ) ∈ OP [T ], hence
j


n−1
cj T j − ϕi (T ) · ψ(T ) ∈ P · OP [T ] .
j=0

Substituting T = y we obtain that


n−1
cj y j − ϕi (y) · ψ(y) ∈ P · OP [y] .
j=0

This proves (3.19).

By Theorem 1.1.19 there exists a place Pi ∈ IPF  such that Ker(σi ) ⊆ Pi
and OP [y] ⊆ OPi , hence Pi |P and ϕi (y) ∈ Pi . The residue class field OPi /Pi
contains OP [y]/Ker(σi ) which is isomorphic to F̄i via σi . Therefore (by (3.18))

f (Pi |P ) ≥ [F̄i : F̄ ] = deg γi (T ) .

For i = j the polynomials γi (T ) = ϕ̄i (T ) and γj (T ) = ϕ̄j (T ) are relatively

prime in F̄ [T ], so there is a relation

1 = ϕ̄i (T ) · λ̄i (T ) + ϕ̄j (T ) · λ̄j (T )

with λi (T ), λj (T ) ∈ OP [T ]. This implies

ϕi (y) · λi (y) + ϕj (y) · λj (y) − 1 ∈ P · OP [y] .

88 3 Extensions of Algebraic Function Fields

We conclude that 1 ∈ Ker(σi ) + Ker(σj ), by (3.19). Since Pi ⊇ Ker(σi ) and

Pj ⊇ Ker(σj ), we have shown that Pi = Pj for i = j.
Now we suppose that hypothesis (∗) holds, i.e.

r
ϕ̄(T ) = γi (T ) .
i=1

Then

r
[F  : F ] = deg ϕ(T ) = deg ϕi (T )
i=1

r 
r
≤ f (Pi |P ) ≤ e(Pi |P ) · f (Pi |P )
i=1 i=1

≤ e(P  |P ) · f (P  |P ) = [F  : F ] ,
P  |P

by Theorem 3.1.11. This is possible only if e(Pi |P ) = 1, f (Pi |P ) = deg ϕi (T ),

and there are no places P  ∈ IPF  with P  |P other than P1 , . . . , Pr .
Finally we assume hypothesis (∗∗). As before we choose Pi ∈ IPF  such
that Pi |P and ϕi (y) ∈ Pi .
Claim: P1 , . . . , Pr are the only extensions of P in F  .
In fact, let P  ∈ IPF  with P  |P . Since

r
0 = ϕ(y) ≡ ϕi (y)εi mod P · OP [y] ,
i=1

we obtain

r
ϕi (y)εi ∈ P  . (3.20)
i=1

P  is a prime ideal in OP  , so (3.20) implies ϕi (y) ∈ P  for some i ∈ {1, . . . , r}

and
P · OP [y] + ϕi (y) · OP [y] ⊆ P  ∩ OP [y] . (3.21)
The left hand side is a maximal ideal of OP [y] by (3.19), therefore equality
holds in (3.21). As we also have

P · OP [y] + ϕi (y) · OP [y] ⊆ Pi ∩ OP [y] ,

it follows that

P  ∩ OP [y] = Pi ∩ OP [y] = ϕi (y) · OP [y] + P · OP [y] . (3.22)

Since OP [y] is the integral closure of OP in F  by hypothesis (∗∗), Proposition

3.2.9 shows now that P  = Pi , and the claim is proved.
 3.3 Local Integral Bases 89

As an immediate consequence of the Claim and Corollary 3.3.5 we see that


r
OP [y] = OPi . (3.23)
i=1

By the Approximation Theorem one can find elements t1 , . . . , tr ∈ F  satisfy-

ing
vPi (ti ) = 1 and vPj (ti ) = 0 for j = i .
Choose a P -prime element t ∈ F ; then
ti ∈ OP [y] ∩ Pi = ϕi (y) · OP [y] + t · OP [y]
by (3.23) and (3.22). Thus ti can be written as
ti = ϕi (y) · ai (y) + t · bi (y) with ai (y), bi (y) ∈ OP [y] .
From this we obtain

r 
r
tεi i = a(y) · ϕi (y)εi + t · b(y) (3.24)
i=1 i=1

with a(y), b(y) ∈ OP [y]. As


r
ϕi (y)εi ≡ ϕ(y) mod t · OP [y]
i=1

and ϕ(y) = 0, (3.24) implies that


r
tεi i = t · u(y) for some u(y) ∈ OP [y] . (3.25)
i=1

Thereby ⎛ ⎞

r
εi = vP i ⎝ tj j ⎠ ≥ vPi (t) = e(Pi |P ) .
ε
(3.26)
j=1

On the other hand we have

f (Pi |P ) = deg γi (T ) (3.27)
by (3.18), (3.19), (3.22) and Proposition 3.2.9. It follows, by (3.26), (3.27) and
Theorem 3.1.11, that

r
[F  : F ] = e(Pi |P ) · f (Pi |P )
i=1


r
≤ εi · deg γi (T ) = deg ϕ(T ) = [F  : F ] .
i=1
Hence εi = e(Pi |P ) for i = 1, . . . , r.
90 3 Extensions of Algebraic Function Fields

We emphasize a special case of Kummer’s Theorem which is often partic-

ularly useful.
Corollary 3.3.8. Let ϕ(T ) = T n + fn−1 (x)T n−1 + · · · + f0 (x) ∈ K(x)[T ] be
an irreducible polynomial over the rational function field K(x). We consider
the function field K(x, y)/K where y satsfies the equation ϕ(y) = 0, and an
element α ∈ K such that fj (α) = ∞ for all j, 0 ≤ j ≤ n − 1. Denote by
Pα ∈ IPK(x) the zero of x − α in K(x). Suppose that the polynomial

ϕα (T ) := T n + fn−1 (α)T n−1 + · · · + f0 (α) ∈ K[T ]

has the following decomposition in the polynomial ring K[T ]:


r
ϕα (T ) = ψi (T )
i=1

with irreducible, monic, pairwise distinct polynomials ψi (T ) ∈ K[T ]. Then we

have:
(a) For every i = 1, . . . , r there is a uniquely determined place Pi ∈ IPK(x,y)
such that x − α ∈ Pi and ψi (y) ∈ Pi . The element x − α is a prime element
of Pi (i.e., e(Pi |Pα ) = 1), and the residue class field of Pi is K-isomorphic to
K[T ]/(ψi (T )). Hence f (Pi |Pα ) = deg ψi (T ).
(b) If deg ψi (T ) = 1 for at least one i ∈ {1, . . . , r}, then K is the full constant
field of K(x, y).
(c) If ϕα (T ) has n = deg ϕ(T ) distinct roots β in K, then there is for each β
with ϕα (β) = 0 a unique place Pα,β ∈ IPK(x,y) such that

x − α ∈ Pα,β and y − β ∈ Pα,β .

Pα,β is a place of K(x, y) of degree 1.

Proof. We set F := K(x) and F  := K(x, y). The assumption fj (α) = ∞

implies that y is integral over the valuation ring of Pα , and the polynomial
ϕα (T ) is nothing else but ϕ̄(T ) (with notation as in Kummer’s Theorem).
Thus we are in the situation of hypothesis (∗) of Kummer’s Theorem, and our
corollary follows immediately.

3.4 The Cotrace of Weil Differentials and the Hurwitz

Genus Formula
In this section the following situation is considered:
F/K is an algebraic function field, F  /F a finite separable extension,
K  is the constant field of F  . Clearly, K  /K is a finite separable ex-
tension as well.
 3.4 The Cotrace of Weil Differentials and the Hurwitz Genus Formula 91

Our aim is to associate with each Weil differential of F/K a Weil differ-
ential of F  /K  . This will yield a very useful formula for the genus of F  , the
Hurwitz Genus Formula. For this we need to introduce the notion of different
of an extension F  /F . Note that F  /F is always assumed to be a separable
extension, hence the trace map TrF /F is not identically zero.

Definition 3.4.1. For P ∈ IPF let OP := icF  (OP ) denote the integral closure
of OP in F  . Then the set

CP := {z ∈ F  | TrF /F (z · OP ) ⊆ OP }

is called the complementary module over OP .

Proposition 3.4.2. With notation as in Definition 3.4.1 the following hold:

(a) CP is an OP -module and OP ⊆ CP .
(b) If {z1 , . . . , zn } is an integral basis of OP over OP , then

n
CP = OP · zi∗ ,
i=1

where {z1∗ , . . . , zn∗ } is the dual basis of {z1 , . . . , zn }.

(c) There is an element t ∈ F  (depending on P ) such that CP = t · OP .
Moreover,
vP  (t) ≤ 0 for all P  |P ,
and for every t ∈ F  we have:

CP = t · OP ⇐⇒ vP  (t ) = vP  (t) for all P  |P .

(d) CP = OP for almost all P ∈ IPF .

Proof. (a) The assertion that CP is an OP -module is trivial. Since the trace
of an element y ∈ OP is in OP , by Corollary 3.3.2, we have OP ⊆ CP .
(b) First we consider an element z∈ CP . As {z1∗ , . . . , zn∗ } is a basis of F  /F ,
there are x1 , . . . , xn ∈ F with z = i=1 xi zi∗ . Since z ∈ CP and z1 , . . . zn ∈ OP ,
n

it follows that TrF /F (zzj ) ∈ OP for 1 ≤ j ≤ n. Now


n 
TrF /F (zzj ) = TrF /F xi zi∗ zj
i=1

n
= xi · TrF /F (zi∗ zj ) = xj ,
i=1
n
by the properties of the dual basis. Therefore xj ∈ OP and z ∈ i=1 OP · zi∗ .
n n
Conversely,
n let z ∈ i=1 OP · zi∗ and u ∈ OP , say z = ∗
i=1 xi zi and
u = j=1 yj zj with xi , yj ∈ OP . Then
92 3 Extensions of Algebraic Function Fields

n 
TrF /F (zu) = TrF /F xi yj zi∗ zj
i,j=1

n 
n
= xi yj · TrF /F (zi∗ zj ) = xi yi ∈ OP .
i,j=1 i=1

Hence z ∈ CP .
n
(c) By (b) we know that CP = i=1 OP ·ui with appropriate elements ui ∈ F  .
Choose x ∈ F such that
vP (x) ≥ −vP  (ui )
for all P  |P and i = 1, . . . , n. Then

vP  (xui ) = e(P  |P ) · vP (x) + vP  (ui ) ≥ 0

for all P  |P and i = 1, . . . , n, therefore x · CP ⊆ OP (observe that we have

OP = {u ∈ F  | vP  (u) ≥ 0 for all P  |P }, by Corollary 3.3.5). Obviously x · CP
is an ideal of OP . Hence x · CP = y · OP for some y ∈ OP , because OP is
a principal ideal domain by Proposition 3.2.10. Setting t := x−1 y we obtain
CP = t · OP . Since OP ⊆ CP , it follows immediately that vP  (t) ≤ 0 for all
P  |P . Finally we have for t ∈ F  :

t · OP = t · OP ⇐⇒ tt−1 ∈ OP and t−1 t ∈ OP

⇐⇒ vP  (tt−1 ) ≥ 0 and vP  (t−1 t ) ≥ 0 for all P  |P
⇐⇒ vP  (t) = vP  (t ) for all P  |P .

(d) Choose a basis {z1 , . . . , zn } of F  /F . By Theorem 3.3.6 {z1 , . . . zn } and

{z1∗ , . . . , zn∗ } are integral bases for almost all P ∈ IPF . Using (b) we see that
CP = OP for almost all P .
Definition 3.4.3. Consider a place P ∈ IPF and the integral closure OP of
OP in F  . Let CP = t · OP be the complementary module over OP . Then we
define for P  |P the different exponent of P  over P by

d(P  |P ) := −vP  (t) .

By Proposition 3.4.2, d(P  |P ) is well-defined and d(P  |P ) ≥ 0. Moreover

d(P  |P ) = 0 holds for almost all P ∈ IPF and P  |P , since CP = 1 · OP
for almost all P . Therefore we can define the divisor
 
Diff(F  /F ) := d(P  |P ) · P  .
P ∈IPF P  |P

This divisor is called the different of F  /F .

Observe that Diff(F  /F ) is a divisor of F  , and Diff(F  /F ) ≥ 0. Later on

we will develop several methods for determining the different in many cases,
see for instance Theorem 3.5.1, Theorem 3.5.10 and Theorem 3.8.7.
 3.4 The Cotrace of Weil Differentials and the Hurwitz Genus Formula 93

Remark 3.4.4. We have the following useful characterization of the comple-

mentary module CP , which follows immediately from the definitions. For every
element z ∈ F  ,

z ∈ CP ⇐⇒ vP  (z) ≥ −d(P  |P ) for all P  |P .

Let us recall some notions from Chapter 1 which are used in what follows.
AF is the adele space of F/K. For a divisor A ∈ Div(F ) let

AF (A) = {α ∈ AF | vP (α) ≥ −vP (A) for all P ∈ IPF } .

This is a K-subspace of AF . The field F is embedded into AF diagonally. A

Weil differential ω of F/K is a K-linear mapping ω : AF → K vanishing on
AF (A) + F for some divisor A ∈ Div(F ). If ω = 0 is a Weil differential of
F/K then its divisor (ω) ∈ Div(F ) is defined as

(ω) = max{A ∈ Div(F ) | ω vanishes on AF (A) + F } .

Definition 3.4.5. Let

AF /F := {α ∈ AF  | αP  = αQ whenever P  ∩ F = Q ∩ F } .

This is an F  -subspace of AF  . The trace mapping TrF /F : F  → F can

be extended to an F -linear map (again denoted by TrF /F ) from AF /F to AF
by setting
(TrF /F (α))P := TrF /F (αP  ) for α ∈ AF /F ,
where P  is any place of F  lying over P . Observe that αP  ∈ OP  for almost
all P  ∈ IPF  , therefore TrF /F (αP  ) ∈ OP for almost all P ∈ IPF by Corollary
3.3.2. Hence TrF /F (α) is an adele of F/K. Clearly the trace of a principal
adele z ∈ F  is the principal adele of TrF /F (z).
For a divisor A ∈ Div(F  ) we set

AF /F (A ) := AF  (A ) ∩ AF /F .

Theorem 3.4.6. In the above situation, for every Weil differential ω of F/K
there exists a unique Weil differential ω  of F  /K  such that

TrK /K (ω  (α)) = ω(TrF /F (α)) (3.28)

for all α ∈ AF /F . This Weil differential is called the cotrace of ω in F  /F ,

and it is denoted by CotrF /F (ω). If ω = 0 and (ω) ∈ Div(F ) is the divisor of
ω, then
(CotrF /F (ω)) = ConF /F ((ω)) + Diff(F  /F ) .

An important special case of this theorem is:

94 3 Extensions of Algebraic Function Fields

Corollary 3.4.7. Let F/K be a function field and x ∈ K such that the ex-
tension F/K(x) is separable. Let η be the Weil differential of the rational
function field K(x) whose existence was proved in Proposition 1.7.4. Then the
divisor of its cotrace in F/K(x) is

(CotrF/K(x) (η)) = −2(x)∞ + Diff(F/K(x)) .

Remark 3.4.8. Using the notion of local components of a Weil differential (cf.
Section 1.7), Equation (3.28) can be replaced by the following local conditions:
for each P ∈ IPF and each y ∈ F  ,
 
ωP (TrF /F (y)) = TrK /K ωP  (y) . (3.29)
P  |P

The equivalence of (3.28) and (3.29) follows after a little thought from
Proposition 1.7.2, which says that ω(γ) is the sum of its local components
ωP (γP ) for each adele γ = (γP )P ∈IPF .

In the course of the proof of Theorem 3.4.6 we need two lemmas which
will be proved first.

Lemma 3.4.9. For each C  ∈ Div(F  ) we have AF  = AF /F + AF  (C  ) .

Proof. Let α = (αP  )P  ∈IPF  be an adele of F  . For all P ∈ IPF there exists by
the Approximation Theorem an element xP ∈ F  with

vP  (αP  − xP ) ≥ −vP  (C  ) for all P  |P .

We set β = (βP  )P  ∈IPF  with βP  := xP whenever P  |P . Then β ∈ AF /F and

α − β ∈ AF  (C  ). Since α = β + (α − β), the lemma follows.

Lemma 3.4.10. Let M/L be a finite separable field extension, V a vector

space over M and μ : V → L be an L-linear map. Then there is a unique
M -linear map μ : V → M such that TrM/L ◦ μ = μ.

Proof. As in the proof of Proposition 3.3.3 we consider the space of linear

forms M ∧ = {λ : M → L | λ is L-linear} as a vector space over M by setting
(z · λ)(w) = λ(z · w) for λ ∈ M ∧ and z, w ∈ M . The dimension of M ∧ over M
is one, hence every λ ∈ M ∧ has a unique representation λ = z · TrM/L with
z ∈ M.
For a fixed element v ∈ V define the map λv : M → L by λv (a) := μ(av);
it is clearly L-linear. Therefore λv = zv · TrM/L with a unique element zv ∈ M ,
and we set μ (v) := zv . Thus we have

μ(av) = (μ (v) · TrM/L )(a) = TrM/L (a · μ (v)) (3.30)

 3.4 The Cotrace of Weil Differentials and the Hurwitz Genus Formula 95

for all a ∈ M and v ∈ V , and μ (v) is uniquely determined by (3.30). Using

this it is easily verified that μ : V → M is M -linear. Setting a = 1 in (3.30)
we obtain μ = TrM/L ◦ μ , which proves the existence of μ : V → M with the
desired properties.
Suppose that there is another μ∗ : V → M with TrM/L ◦ μ = TrM/L ◦ μ∗
and μ∗ = μ . Then the image of μ − μ∗ is the whole of M , and we have
TrM/L ◦ (μ − μ∗ ) = 0. This is a contradiction, since TrM/L is not the zero
map.

Proof of Theorem 3.4.6. First we want to show the existence of a Weil differ-
ential ω  such that
TrK /K (ω  (α)) = ω(TrF /F (α))
holds for all α ∈ AF /F . For ω = 0 set ω  := 0, therefore we can assume in the
sequel that ω = 0. For brevity we set

W  := ConF /F ((ω)) + Diff(F  /F ) . (3.31)

The construction of ω  is given in three steps.

Step 1. The K-linear mapping ω1 : AF /F → K which is defined by setting
ω1 := ω ◦ TrF /F has the following properties:
(a1 ) ω1 (α) = 0 for α ∈ AF /F (W  ) + F  .
(b1 ) If B  ∈ Div(F  ) is a divisor with B  ≤ W  , then there is an adele
β ∈ AF /F (B  ) with ω1 (β) = 0.
Proof of Step 1. (a1 ) Obviously ω1 is K-linear, and ω1 vanishes on F  since ω
vanishes on F . Now let α ∈ AF /F (W  ). In order to prove ω1 (α) = 0 we only
have to verify that for all P ∈ IPF and P  |P the following holds:

vP (TrF /F (αP  )) ≥ −vP (ω) . (3.32)

(Observe that ω vanishes on AF ((ω)) by definition of the divisor (ω).) Choose

an element x ∈ F with vP (x) = vP (ω). Then

vP  (xαP  ) = vP  (x) + vP  (αP  ) ≥ e(P  |P ) · vP (ω) − vP  (W  )

= vP  (ConF /F ((ω)) − W  ) = −vP  (Diff(F  /F )) = −d(P  |P ) .

(Recall that d(P  |P ) denotes the different exponent of P  |P .) This implies

by Remark 3.4.4 that xαP  ∈ CP , the complementary module over OP , and
therefore vP (TrF /F (xαP  )) ≥ 0. As TrF /F (xαP  ) = x·TrF /F (αP  ) and vP (x) =
vP (ω), assertion (3.32) follows.
(b1 ) Now there is given a divisor B  ≤ W  ; i.e., there is a place P0 ∈ IPF such
that
vP ∗ (ConF /F ((ω)) − B  ) < −d(P ∗ |P0 ) (3.33)
for some P ∗ |P0 . Let OP 0 (resp. CP0 ) denote the integral closure of OP0 in F 
(resp. the complementary module over OP0 ), and consider the set
96 3 Extensions of Algebraic Function Fields

J := {z ∈ F  | vP ∗ (z) ≥ vP ∗ (ConF /F ((ω)) − B  ) for all P ∗ |P0 } .

By the Approximation Theorem there is an element u ∈ J which satisfies

vP ∗ (u) = vP ∗ (ConF /F ((ω)) − B  ) for all P ∗ |P0 , therefore J ⊆
 CP0 by Remark
3.4.4 and (3.33). Since J · OP 0 ⊆ J it follows that

TrF /F (J) ⊆ OP0 . (3.34)

Choose t ∈ F with vP0 (t) = 1. For some r ≥ 0 we have tr · J ⊆ OP 0 (trivial

by definition of J), so tr · TrF /F (J) = TrF /F (tr J) ⊆ OP0 . It is easily checked
that tr · TrF /F (J) is an ideal of OP0 , consequently tr · TrF /F (J) = ts · OP0
with s ≥ 0, and we obtain TrF /F (J) = tm · OP0 for some m ∈ ZZ. By (3.34),
m ≤ −1 and therefore
t−1 · OP0 ⊆ TrF /F (J) . (3.35)
Recall the notion of the local component of a Weil differential, cf. Section 1.7.
By Proposition 1.7.3(a) we can find an element x ∈ F with

vP0 (x) = −vP0 (ω) − 1 and ωP0 (x) = 0 . (3.36)

We choose y ∈ F with vP0 (y) = vP0 (ω), so xy ∈ t−1 · OP0 . By (3.35) there is
some z ∈ J with TrF /F (z) = xy. Consider the adele β ∈ AF /F given by

0 if P   P0 ,
βP  := −1
y z if P  | P0 .

It follows from the definition of J that for P  |P0

vP  (β) = −vP  (y) + vP  (z)

≥ −vP  (ConF /F ((ω))) + vP  (ConF /F ((ω)) − B  )
= −vP  (B  ) .

So β ∈ AF /F (B  ). Finally, we have ω1 (β) = ω(TrF /F (β)) = ωP0 (x) = 0 by

(3.36). This proves (b1 ).
Step 2. We define ω2 : AF  → K as follows. For α ∈ AF  there are adeles
β ∈ AF /F and γ ∈ AF  (W  ) such that α = β + γ, by Lemma 3.4.9. Set

ω2 (α) := ω1 (β) .

This is well-defined. In fact, if we have two representations α = β +γ = β1 +γ1

with β, β1 ∈ AF /F and γ, γ1 ∈ AF  (W  ) then

β1 − β = γ − γ1 ∈ AF /F ∩ AF  (W  ) = AF /F (W  ) .

Hence ω1 (β1 ) − ω1 (β) = ω1 (β1 − β) = 0 by (a1 ). The mapping ω2 is obviously

K-linear, and by (a1 ) and (b1 ) it has the following properties:
 3.4 The Cotrace of Weil Differentials and the Hurwitz Genus Formula 97

(a2 ) ω2 (α) = 0 for α ∈ AF  (W  ) + F  .

(b2 ) If B  ∈ Div(F  ) is a divisor with B  ≤ W  , then there is an adele
β ∈ AF  (B  ) with ω2 (β) = 0.
Thus far, we have constructed a K-linear mapping ω2 : AF  → K vanish-
ing on AF  (W  ) + F  . However, ω2 is not a Weil differential of F  /K  if K  is
strictly larger than K. Therefore we have to ‘lift’ ω2 to a K  -linear map; this
is done in the next step.
Step 3. By Lemma 3.4.10 there exists a K  -linear map ω  : AF  → K  such
that TrK /K ◦ω  = ω2 . From the definition of ω1 and ω2 we obtain immediately
that for α ∈ AF /F

TrK /K (ω  (α)) = ω2 (α) = ω1 (α) = ω(TrF /F (α)) .

This proves (3.28), and it remains to show:

(a3 ) ω  (α) = 0 for α ∈ AF  (W  ) + F  .
(b3 ) If B  ∈ Div(F  ) is a divisor with B  ≤ W  , then there is some adele
β ∈ AF  (B  ) with ω  (β) = 0.
Proof of (a3 ). Since ω  is K  -linear, the image of AF  (W  ) + F  under ω  is
either 0 or the whole of K  . In the latter case there is some α ∈ AF  (W  ) + F 
such that TrK /K (ω  (α)) = 0, since TrK /K : K  → K is not the zero map.
By construction of ω  we have ω2 = TrK /K ◦ ω  . Hence ω2 (α) = 0, which is a
contradiction to (a2 ).
Proof of (b3 ). By (b2 ) there exists some adele β ∈ AF  (B  ) with the property
ω2 (β) = 0. So TrK /K (ω  (β)) = 0, and the assertion follows immediately.
We have established the existence of a Weil differential ω  of F  /K  satis-
fying (3.28), and we have shown that the divisor of ω  is

(ω  ) = W  = ConF /F ((ω)) + Diff(F  /F ) .

In order to prove uniqueness, suppose that ω ∗ is another Weil differential of

F  /K  with the property (3.28); i.e.,

TrK /K (ω ∗ (α)) = TrK /K (ω  (α)) = ω(TrF /F (α))

for all α ∈ AF /F . Setting η := ω ∗ − ω  we obtain

TrK /K (η(α)) = 0 for all α ∈ AF /F . (3.37)

η is a Weil differential of F  /K  , hence η vanishes on AF  (C  ) for some divisor

C  ∈ Div(F  ). By Lemma 3.4.9 and (3.37) it follows that TrK /K (η(α)) = 0
for all α ∈ AF  . This implies η = 0 and ω ∗ = ω  .
We note some formal properties of the cotrace mapping ω → CotrF /F (ω).
98 3 Extensions of Algebraic Function Fields

Proposition 3.4.11. (a) If ω, ω1 and ω2 are Weil differentials of F/K and

x ∈ F , then

CotrF /F (ω1 + ω2 ) = CotrF /F (ω1 ) + CotrF /F (ω2 )

and
CotrF /F (xω) = x · CotrF /F (ω) .
(b) Let F  /F  be another finite separable extension. Then

CotrF /F (ω) = CotrF /F  (CotrF /F (ω))

for each Weil differential ω of F/K.

Proof. Keeping in mind the uniqueness assertion in Theorem 3.4.6, it is suf-

ficient to show that
  
TrK /K CotrF /F (ω1 ) + CotrF /F (ω2 ) (α) = (ω1 + ω2 )(TrF /F (α)) (3.38)

holds for all α ∈ AF /F . The proof of (3.38) is straightforward, using the
linearity of the trace. In the same manner one proves that CotrF /F (xω) = x ·
CotrF /F (ω), as well as assertion (b).

Corollary 3.4.12 (Transitivity of the Different). If F  ⊇ F  ⊇ F are

finite separable extensions, the following hold:
(a) Diff(F  /F ) = ConF /F  (Diff(F  /F )) + Diff(F  /F  ) .
(b) d(P  |P ) = e(P  |P  ) · d(P  |P ) + d(P  |P  ) , if P  (resp. P  , P ) are places
of F  (resp. F  , F ) with P  ⊇ P  ⊇ P .

Proof. (b) is merely a reformulation of (a), so we only prove (a). Choose a

Weil differential ω = 0 of F/K. Then the divisor of CotrF /F (ω) is

(CotrF /F (ω)) = ConF /F ((ω)) + Diff(F  /F ) (3.39)

by Theorem 3.4.6. On the other hand, Proposition 3.4.11 yields

(CotrF /F (ω)) = (CotrF /F  (CotrF /F (ω)))

= ConF /F  ((CotrF /F (ω))) + Diff(F  /F  )
= ConF /F  (ConF /F ((ω)) + Diff(F  /F )) + Diff(F  /F  )
= ConF /F ((ω)) + ConF /F  (Diff(F  /F )) + Diff(F  /F  ) . (3.40)

(We have used the transitivity of the conorm, cf. Definition 3.1.8.) Comparing
(3.39) and (3.40) we obtain (a).

An important consequence of Theorem 3.4.6 is the following result.

 3.5 The Different 99

Theorem 3.4.13 (Hurwitz Genus Formula). Let F/K be an algebraic

function field of genus g and let F  /F be a finite separable extension. Let K 
denote the constant field of F  and g  the genus of F  /K  . Then we have

[F  : F ]
2g  − 2 = (2g − 2) + deg Diff(F  /F ) .
[K  : K]

Proof. Choose a Weil differential ω = 0 of F/K. It follows from Theorem

3.4.6 that
(CotrF /F (ω)) = ConF /F ((ω)) + Diff(F  /F ) . (3.41)
Recall that the degree of a canonical divisor is 2g − 2 (resp. 2g  − 2). Then we
obtain from (3.41) and Corollary 3.1.14

2g  − 2 = deg ConF /F ((ω)) + deg Diff(F  /F )

[F  : F ]
= (2g − 2) + deg Diff(F  /F ) .
[K  : K]

We emphasize a special case of the Hurwitz Genus Formula:

Corollary 3.4.14. Let F/K be a function field of genus g and let x ∈ F \ K

such that the extension F/K(x) is separable. Then

2g − 2 = −2[F : K(x)] + deg Diff(F/K(x)) .

Every function field F/K can be regarded as a finite extension of a rational

function field K(x) (as we shall prove in Section 3.10, one can choose x in
such a way that F/K(x) is separable). Therefore the Hurwitz Genus Formula
(resp. Corollary 3.4.14) is a powerful tool that allows determination of the
genus of F in terms of the different of F/K(x). Thus far however, we have no
methods at hand how to determine this different. The next section addresses
this problem.

3.5 The Different

We consider a finite separable extension F  /F where F/K resp. F  /K 

are algebraic function fields with constant fields K resp. K  . As always,
the field K (hence also K  ) is assumed to be perfect.

For P ∈ IPF and P  ∈ IPF  with P  |P we have defined the ramification

index e(P  |P ) and the different exponent d(P  |P ) (Definition 3.1.5 and 3.4.3).
There is a close relationship between these two numbers, given by the following
theorem:
100 3 Extensions of Algebraic Function Fields

Theorem 3.5.1 (Dedekind’s Different Theorem). With notation as above

we have for all P  |P
(a) d(P  |P ) ≥ e(P  |P ) − 1.
(b) d(P  |P ) = e(P  |P ) − 1 if and only if e(P  |P ) is not divisible by char K.
In particular, if char K = 0 then d(P  |P ) = e(P  |P ) − 1.

We shall first prove part (a) of Dedekind’s Theorem; the proof will require
an understanding of the action of automorphisms on the places of a function
field. More precisely we need:

Lemma 3.5.2. Let F ∗ /F be an algebraic extension of function fields, P ∈ IPF

and P ∗ ∈ IPF ∗ with P ∗ |P . Consider an automorphism σ of F ∗ /F . Then
σ(P ∗ ) := {σ(z) | z ∈ P ∗ } is a place of F ∗ , and we have
(a) vσ(P ∗ ) (y) = vP ∗ (σ −1 (y)) for all y ∈ F ∗ .
(b) σ(P ∗ )|P .
(c) e(σ(P ∗ )|P ) = e(P ∗ |P ) and f (σ(P ∗ )|P ) = f (P ∗ |P ).

Proof of the Lemma. Clearly σ(OP ∗ ) is a valuation ring of F ∗ and σ(P ∗ ) is

its maximal ideal; therefore σ(P ∗ ) is a place of F ∗ , and the corresponding
valuation ring is Oσ(P ∗ ) = σ(OP ∗ ). If t∗ is a prime element of P ∗ , i.e. P ∗ =
t∗ · OP ∗ , then σ(P ∗ ) = σ(t∗ ) · σ(OP ∗ ), so σ(t∗ ) is a prime element for σ(P ∗ ).
(a) Let 0 = y ∈ F ∗ , say y = σ(z). Writing z = t∗r u with r = vP ∗ (z) and u ∈
OP ∗ \P ∗ , we obtain y = σ(t∗ )r · σ(u) where σ(u) ∈ Oσ(P ∗ ) \σ(P ∗ ) and σ(t∗ ) is
a prime element for σ(P ∗ ). Therefore vσ(P ∗ ) (y) = r = vP ∗ (z) = vP ∗ (σ −1 (y)).
(b) σ(P ∗ ) lies over P since σ(P ∗ ) ⊇ σ(P ) = P .
(c) Choose a P -prime element x ∈ F . Then

e(σ(P ∗ )|P ) = vσ(P ∗ ) (x) = vP ∗ (σ −1 (x)) = vP ∗ (x) = e(P ∗ |P ) .

The automorphism σ of F ∗ /F induces an isomorphism σ̄ of the residue class

field FP∗ ∗ onto Fσ(P
∗
∗ ) given by

σ̄(z + P ∗ ) := σ(z) + σ(P ∗ ) .

σ̄ is the identity on FP , hence f (P ∗ |P ) = f (σ(P ∗ )|P ).

Proof of Theorem 3.5.1(a). As before let OP denote the integral closure of OP
in F  , and CP the complementary module over OP . We want to show that

TrF /F (t · OP ) ⊆ OP (3.42)

for every element t ∈ F  which satisfies

vP  (t) = 1 − e(P  |P ) for all P  |P . (3.43)

 3.5 The Different 101

Note that (3.42) implies t ∈ CP , and the characterization of CP given in

Remark 3.4.4 yields 1 − e(P  |P ) ≥ −d(P  |P ), so that d(P  |P ) ≥ e(P  |P ) − 1.
In order to prove (3.42), consider a finite Galois extension F ∗ /F such that
F ⊆ F  ⊆ F ∗ , and choose n := [F  : F ] automorphisms σ1 , . . . , σn of F ∗ /F
whose restrictions to F  are pairwise distinct. For z ∈ OP we have


n
TrF /F (t · z) = σi (t · z) . (3.44)
i=1

We fix some place P ∗ of F ∗ which lies over P , and set Pi∗ := σi−1 (P ∗ )

and Pi := Pi∗ ∩ F  . Note that σi (z) is integral over OP since z ∈ OP , and
therefore vP ∗ (σi (z)) ≥ 0. Then we obtain

vP ∗ (σi (t · z)) = vP ∗ (σi (t)) + vP ∗ (σi (z))

≥ vP ∗ (σi (t)) = vPi∗ (t) (by Lemma 3.5.2)
= e(Pi∗ |Pi )(1 − e(Pi |P )) (by (3.43))
> −e(Pi∗ |Pi ) · e(Pi |P )
= −e(Pi∗ |P ) = −e(P ∗ |P ) (by Lemma 3.5.2).

Using (3.44) we conclude

−e(P ∗ |P ) < vP ∗ (TrF /F (t · z)) = e(P ∗ |P ) · vP (TrF /F (t · z)) .

This implies vP (TrF /F (t · z)) ≥ 0, hence (3.42).

The essential step in the proof of part (b) of Theorem 3.5.1 is the following
lemma.

Lemma 3.5.3. Let P ∈ IPF and P1 , . . . , Pr ∈ IPF  be all the extensions of P

in F  /F . Consider the residue class fields k := OP /P resp. ki := OPi /Pi ⊇ k
and the corresponding residue class maps π : OP → k resp. πi : OPi → ki
(for i = 1, . . . , r). Then we have for every u ∈ OP (the integral closure of OP
in F  )

r
π(TrF /F (u)) = e(Pi |P ) · Trki /k (πi (u)) .
i=1

Proof of Theorem 3.5.1(b). We maintain the notation of Lemma 3.5.3 and

abbreviate ei := e(Pi |P ). Let P  = P1 and e := e(P  |P ). It must be shown
that
d(P  |P ) = e − 1 ⇐⇒ char K does not divide e . (3.45)
First assume that e is not divisible by char K. Suppose d(P  |P ) ≥ e. Then
there exists some w ∈ F  such that

vP  (w) ≤ −e and TrF /F (w · OP ) ⊆ OP . (3.46)

102 3 Extensions of Algebraic Function Fields

Since K is perfect, the extension k1 /k is separable, and we can find y0 ∈ OP 

with Trk1 /k (π1 (y0 )) = 0. By the Approximation Theorem there is an element
y ∈ F  such that
vP  (y − y0 ) > 0
and
vPi (y) ≥ max{1, ei + vPi (w)} for 2 ≤ i ≤ r . (3.47)
Then y ∈ OP and, by Lemma 3.5.3,


r
π(TrF /F (y)) = e · Trk1 /k (π1 (y)) + ei · Trki /k (πi (y))
i=2
= e · Trk1 /k (π1 (y0 )) = 0 .

(Here we use the fact that char K does not divide e, hence e = 0 in k.) We
conclude that
vP (TrF /F (y)) = 0 .
Now choose x ∈ F with vP (x) = 1. Then

TrF /F (x−1 y) = x−1 · TrF /F (y) ∈ OP . (3.48)

On the other hand we have x−1 yw−1 ∈ OP , since

vP  (x−1 yw−1 ) = −e + vP  (y) − vP  (w) ≥ 0

and
vPi (x−1 yw−1 ) = vPi (y) − (ei + vPi (w)) ≥ 0
for i = 2, . . . , r , by (3.46) and (3.47). It follows that x−1 y ∈ w · OP and
TrF /F (x−1 y) ∈ OP by (3.46); this contradicts (3.48). So we have proved the
implication ⇐ of (3.45).
In order to prove the converse, we assume now that char K divides e, and
we have to show that d(P  |P ) ≥ e. Choose u ∈ F  such that

vP  (u) = −e and vPi (u) ≥ −ei + 1 (i = 2, . . . , r) . (3.49)

As before, x ∈ F denotes a P -prime element. For each z ∈ OP we have

vP  (xuz) ≥ 0 and vPi (xuz) > 0

for i = 2, . . . , r. Therefore xuz ∈ OP , and by Lemma 3.5.3,


r
π(TrF /F (xuz)) = e · Trk1 /k (π1 (xuz)) + ei · Trki /k (πi (xuz))
i=2
= e · Trk1 /k (π1 (xuz)) = 0 .
 3.5 The Different 103

We conclude that x · TrF /F (uz) = TrF /F (xuz) ∈ P = xOP , and thereby
TrF /F (uz) ∈ OP for all z ∈ OP . Thus u ∈ CP and −e = vP  (u) ≥ −d(P  |P )
by (3.49) and Remark 3.4.4.

Proof of Lemma 3.5.3. The trace TrF /F (u) can be evaluated as the trace of
the F -linear map μ : F  → F  which is given by μ(z) = u ·z (see Appendix A).
First we show that π(TrF /F (u)) has an interpretation as the trace of a certain
k-linear map μ̄ : V → V (where V is some k-vector space to be defined below);
decomposing V into invariant subspaces will then yield the final result.
Let t ∈ F be a P -prime element. The quotient V := OP /tOP can be
considered as a vector space over k by setting

(x + P ) · (z + tOP ) := xz + tOP (x ∈ OP , z ∈ OP ) . (3.50)

Note that this scalar multiplication is well-defined. Choose an integral basis

{z1 , . . . , zn } of OP over OP (where n = [F  : F ], cf. Corollary 3.3.5). Then
{z1 + tOP , . . . , zn + tOP } constitutes a basis of V over k (the proof is trivial),
in particular dimk (V ) = n. We define a k-linear map μ̄ : V → V by

μ̄(z + tOP ) := u · z + tOP . (3.51)

Let A = (aij )1≤i,j≤n be the matrix which represents μ with respect to the
basis {z1 , . . . , zn }. Since this is an integral basis and u ∈ OP , the coefficients
aij are in OP . Obviously Ā := (π(aij ))1≤i,j≤n represents μ̄ with respect to
{z1 + tOP , . . . , zn + tOP }, therefore

π(TrF /F (u)) = π(Tr A) = Tr (Ā) = Tr (μ̄) . (3.52)

Next we introduce for 1 ≤ i ≤ r the quotients Vi := OPi /Piei and the map-
pings μi : Vi → Vi , defined by

μi (z + Piei ) := u · z + Piei .

Vi is considered as a vector space over k in the obvious manner (cf. (3.50)),

and μi turns out to be k-linear (all this is easily verified). There is a natural
isomorphism
r
f : V −→ Vi ,
i=1

given by
f (z + tOP ) := (z + P1e1 , . . . , z + Prer ) . (3.53)
In fact, f is surjective by the Approximation Theorem. In order to prove
that f is injective suppose f (z + tOP ) = 0. Then vPi (z) ≥ ei ; i.e., we have
vPi (z · t−1 ) ≥ 0 for i = 1, . . . , r. This implies z · t−1 ∈ OP , hence z ∈ tOP .
There is a commutative diagram of k-linear mappings
104 3 Extensions of Algebraic Function Fields

μ̄
V  V

f f

 

r
(μ1 , . . . , μr ) 
r
Vi  Vi
i=1 i=1

Fig. 3.1.

where (μ1 , . . . , μr )(v1 , . . . , vr ) := (μ1 (v1 ), . . . , μr (vr )), vi ∈ Vi . Since f is an

isomorphism,
r it follows that Tr(μ̄) = Tr((μ1 , . . . , μr )), which is obviously equal
to i=1 Tr(μi ). Combining this with (3.52) we obtain

r
π(TrF /F (u)) = Tr(μi ) . (3.54)
i=1

The proof of Lemma 3.5.3 will be completed once we can show that

Tr(μi ) = ei · Trki /k (πi (u)) . (3.55)

Consider the chain of k-subspaces

(0) (1) (ei )
Vi = Vi ⊇ Vi ⊇ . . . ⊇ Vi = {0} ,
(j)
where Vi := Pij /Piei ⊆ Vi . These spaces are invariant under μi , so μi induces
linear maps

(j) (j+1) (j) (j+1)
Vi /Vi −→ Vi /Vi ,
σij :
[z + Piei ] −→ [u · z + Piei ]
for j = 0, . . . , ei − 1. Here [z + Piei ] denotes the residue class of z + Piei in
(j) (j+1)
Vi /Vi . It is easily seen that
i −1
e
Tr(μi ) = Tr(σij ) . (3.56)
j=0

(Represent μi by a matrix with respect to a basis of Vi which is composed of

(j) (j+1)
bases of Vi modulo Vi , for 0 ≤ j ≤ ei − 1.) We know that

Trki /k (πi (u)) = Tr(γi ) , (3.57)

where γi : ki → ki is the k-linear map defined by γi (z + Pi ) = u · z + Pi .

(j) (j+1)
Now we establish for 0 ≤ j ≤ ei − 1 an isomorphism h : ki → Vi /Vi of
k-vector spaces such that the following diagram is commutative:
 3.5 The Different 105

γi
ki  ki

h h

 
(j) (j+1) σij (j) (j+1)
Vi /Vi  Vi /Vi

Fig. 3.2.

This diagram yields Tr(γi ) = Tr(σij ) (because h is an isomorphism), and then

(3.55) follows immediately from (3.56) and (3.57).
The map h is defined as follows: Choose a Pi -prime element ti ∈ F  and
let
h(z + Pi ) := [tji z + Piei ] .
Then it is easily verified that h is well-defined, k-linear, bijective and the
diagram in Figure 3.2 is commutative. This completes the proof of Lemma
3.5.3 and thereby Theorem 3.5.1.

We would like to draw some conclusions from Dedekind’s Theorem. Recall

that P  |P is said to be ramified if e(P  |P ) > 1; otherwise P  |P is unramified
(cf. Definition 3.1.5).

Definition 3.5.4. Let F  /F be an algebraic extension of function fields and

P ∈ IPF .
(a) An extension P  of P in F  is said to be tamely (resp. wildly) ramified if
e(P  |P ) > 1 and the characteristic of K does not divide e(P  |P ) (resp. char K
divides e(P  |P )).
(b) We say that P is ramified (resp. unramified) in F  /F if there is at least
one P  ∈ IPF over P such that P  |P is ramified (resp. if P  |P is unramified
for all P  |P ). The place P is tamely ramified in F  /F if it is ramified in F  /F
and no extension of P in F  is wildly ramified. If there is at least one wildly
ramified place P  |P we say that P is wildly ramified in F  /F .
(c) P is totally ramified in F  /F if there is only one extension P  ∈ IPF  of P
in F  , and the ramification index is e(P  |P ) = [F  : F ].
(d) F  /F is said to be ramified (resp. unramified) if at least one P ∈ IPF is
ramified in F  /F (resp. if all P ∈ IPF are unramified in F  /F ).
(e) F  /F is said to be tame if no place P ∈ IPF is wildly ramified in F  /F .
106 3 Extensions of Algebraic Function Fields

Corollary 3.5.5. Let F  /F be a finite separable extension of algebraic func-

tion fields.
(a) If P ∈ IPF and P  ∈ IPF  such that P  |P , then P  |P is ramified if and only
if P  ≤ Diff(F  /F ).
If P  |P is ramified, then

d(P  |P ) = e(P  |P ) − 1 ⇐⇒ P  |P is tamely ramified ,

d(P  |P ) ≥ e(P  |P ) ⇐⇒ P  |P is wildly ramified .

(b) Almost all places P ∈ IPF are unramified in F  /F .

This corollary follows immediately from Dedekind’s Theorem. Next we

note an important special case of the Hurwitz Genus Formula.

Corollary 3.5.6. Suppose that F  /F is a finite separable extension of alge-

braic function fields having the same constant field K. Let g (resp. g  ) denote
the genus of F/K (resp. F  /K). Then
 
2g  − 2 ≥ [F  : F ] · (2g − 2) + (e(P  |P ) − 1) · deg P  .
P ∈IPF P  |P

Equality holds if and only if F  /F is tame (for instance if K is a field of

characteristic 0).

Proof. Trivial by Theorems 3.4.13 and 3.5.1.

Corollary 3.5.7. Suppose that F  /F is a finite separable extension of func-

tion fields having the same constant field. Let g (resp. g  ) denote the genus of
F (resp. F  ). Then g ≤ g  .

Corollary 3.5.8. Let F/K(x) be a finite separable extension of the rational

function field of degree [F : K(x)] > 1 such that K is the constant field of F .
Then F/K(x) is ramified.

Proof. The Hurwitz Genus Formula yields

2g − 2 = −2[F : K(x)] + deg Diff(F/K(x)) ,

where g is the genus of F/K. Therefore

deg Diff(F/K(x)) ≥ 2([F : K(x)] − 1) > 0 .

The assertion follows since each place in the support of the different ramifies
by Corollary 3.5.5.

We give another application of the above results.

 3.5 The Different 107

Proposition 3.5.9 (Lüroth’s Theorem). Every subfield of a rational func-

tion field is rational; i.e., if K  F0 ⊆ K(x) then F0 = K(y) for some y ∈ F0 .

Proof. Suppose first that K(x)/F0 is separable. Let g0 denote the genus of
F0 /K. Then

−2 = [K(x) : F0 ] · (2g0 − 2) + deg Diff(K(x)/F0 ) ,

which implies g0 = 0. If P is a place of K(x)/K of degree one then P0 = P ∩F0

is a place of F0 /K of degree one. Therefore F0 /K is rational by Proposition
1.6.3.
Now assume that K(x)/F0 is not separable. There is an intermediate field
F0 ⊆ F1 ⊆ K(x) such that F1 /F0 is separable and K(x)/F1 is purely insepa-
rable. According to what we have proved above, it is sufficient to show that
F1 /K is rational. As K(x)/F1 is purely inseparable, [K(x) : F1 ] = q = pν
where p = char K > 0, and z q ∈ F1 for each z ∈ K(x). In particular,

K(xq ) ⊆ F1 ⊆ K(x) . (3.58)

The degree [K(x) : K(xq )] is equal to the degree of the pole divisor of xq
in K(x)/K by Theorem 1.4.11, therefore [K(x) : K(xq )] = q. By (3.58), it
follows that F1 = K(xq ), hence F1 /K is rational.

Next we prove a theorem that is often very useful for evaluating the dif-
ferent of F  /F .

Theorem 3.5.10. Suppose F  = F (y) is a finite separable extension of a

function field F of degree [F  : F ] = n. Let P ∈ IPF be such that the minimal
polynomial ϕ(T ) of y over F has coefficients in OP (i.e., y is integral over
OP ), and let P1 , . . . , Pr ∈ IPF  be all places of F  lying over P . Then the
following hold:
(a) d(Pi |P ) ≤ vPi (ϕ (y)) for 1 ≤ i ≤ r.
(b) {1, y, . . . , y n−1 } is an integral basis of F  /F at the place P if and only if
d(Pi |P ) = vPi (ϕ (y)) for 1 ≤ i ≤ r.
(Here ϕ (T ) denotes the derivative of ϕ(T ) in the polynomial ring F [T ].)

Proof. The dual basis of {1, y, . . . , y n−1 } is closely related to the different
exponents d(Pi |P ) by Proposition 3.4.2, therefore our first aim is to determine
this dual basis. Since ϕ(y) = 0, the polynomial ϕ(T ) factors in F  [T ] as

ϕ(T ) = (T − y)(cn−1 T n−1 + . . . + c1 T + c0 ) (3.59)

with c0 , . . . , cn−1 ∈ F  and cn−1 = 1. We claim:

 
c0 cn−1
, . . . , is the dual basis of {1, y, . . . , y n−1 } . (3.60)
ϕ (y) ϕ (y)
108 3 Extensions of Algebraic Function Fields

(Note that ϕ (y) = 0 since y is separable over F .) By definition of the dual

basis, (3.60) is equivalent to
ci
TrF /F (  · y l ) = δil for 0 ≤ i, l ≤ n − 1 . (3.61)
ϕ (y)
In order to prove (3.61), consider the n distinct embeddings σ1 , . . . , σn of
F  /F into Φ (which denotes, as usual, an algebraically closed extension of F ).
We set yj := σj (y) and obtain

n
ϕ(T ) = (T − yj ) .
j=1

Differentiating this equation and substituting T = yν yields


ϕ (yν ) = (yν − yi ) . (3.62)
i=ν

For 0 ≤ l ≤ n − 1 we consider the polynomial

n 
ϕ(T ) yjl
ϕl (T ) := ·  − T l ∈ Φ[T ] .
j=1
T − y j ϕ (y j )

Its degree is at most n − 1, and for 1 ≤ ν ≤ n we have

  yl
ϕl (yν ) = (yν − yi ) ·  ν − yνl = 0 ,
ϕ (yν )
i=ν

by (3.62). A polynomial of degree ≤ n − 1 having n distinct zeros is the zero

polynomial. Hence ϕl (T ) = 0; i.e.,
n
ϕ(T ) yjl
Tl = ·  for 0 ≤ l ≤ n − 1 . (3.63)
j=1
T − yj ϕ (yj )

The embeddings σi : F  → Φ extend to embeddings σi : F  (T ) → Φ(T ) by

setting σi (T ) = T , and we obtain from (3.63)


n  
ϕ(T ) yl
l
T = σj · 
j=1
T − y ϕ (y)

n n−1
 
yl
= σj ci T · 
i
(by (3.59))
j=1 i=0
ϕ (y)
 
n−1 n
ci

= σj (  · y ) Ti
l

i=0 j=1
ϕ (y)

n−1  
ci
= Tr F /F · y Ti .
l

i=0
ϕ (y)
 3.5 The Different 109

Comparing coefficients yields (3.61) and thereby (3.60).

Next we want to show:

n−1
cj ∈ OP · y i for j = 0, . . . , n − 1 . (3.64)
i=0

The minimal polynomial ϕ(T ) of y over F has the form

ϕ(T ) = T n + an−1 T n−1 + . . . + a1 T + a0 (3.65)

with ai ∈ OP . Thus we have by (3.59) the following recursion formulas:

cn−1 = 1, c0 y = −a0 , and ci y = ci−1 − ai for 1 ≤ i ≤ n − 1 . (3.66)

(3.64) is clearly true for j = n−1. Suppose it holds for some j ∈ {1, . . . , n−1},
say

n−1
cj = si y i with si ∈ OP .
i=0

Then we obtain, using (3.66) and (3.65),


n−2
cj−1 = aj + cj y = aj + si y i+1 + sn−1 y n
i=0

n−2 
n−1 
n−1
= aj + si y i+1 − sn−1 ai y i ∈ OP · y i .
i=0 i=0 i=0

Therefore (3.64) is proved. In a similar manner we can show


n−1
yj ∈ OP · ci for j = 0, . . . , n − 1 . (3.67)
i=0

In fact, (3.67) holds for j = 0; if


n−1
yj = ri ci with ri ∈ OP
i=0

for some j ≥ 0, then by (3.66)


n−1 
n−1
y j+1
= ri ci y = ri (ci−1 − ai ) − r0 a0
i=0 i=1

n−2 n−1
  
n−1
= ri+1 ci − ri ai · cn−1 ∈ OP · ci .
i=0 i=0 i=0
110 3 Extensions of Algebraic Function Fields

Proof of Theorem 3.5.10(a). As always, CP denotes the complementary module

and OP the integral closure of OP in F  . We have to show that d(Pi |P ) ≤
vPi (ϕ (y)), which is (by definition of the different exponent) equivalent to the
following statement:

z ∈ CP ⇒ vPi (z) ≥ −vPi (ϕ (y)) for i = 1, . . . , r . (3.68)

The element z ∈ CP can be written as


n−1
ci
z= ri · 
with ri ∈ F .
i=0
ϕ (y)

(Note that {c0 , . . . , cn−1 } is a basis of F  /F by (3.60).) Since y l is integral

over OP and z ∈ CP , we have TrF /F (z · y l ) ∈ OP . Now
n−1
 
ci
TrF /F (z · y ) = TrF /F
l
ri ·  · yl = rl
i=0
ϕ (y)

by (3.61), so rl ∈ OP . Observing (3.64) we obtain

1 
n−1
1 
n−1
1
z= 
· r c
i i ∈ 
· OP · y i ⊆  · OP .
ϕ (y) i=0 ϕ (y) i=0 ϕ (y)

This implies (3.68) and finishes the proof of part (a) of our theorem.

Proof of Theorem 3.5.10(b). By (3.64) and (3.67) we know that


n−1 
n−1
OP · y i = OP · ci . (3.69)
i=0 i=0

Suppose now that {1, y, . . . , y n−1 } is an integral basis for P . From (3.60) and
Proposition 3.4.2 it follows that


n−1
ci 1 
n−1
CP = OP · 
= 
· OP · ci
i=0
ϕ (y) ϕ (y) i=0

1 
n−1
1
= 
· OP · y i =  · OP .
ϕ (y) i=0 ϕ (y)

Consequently d(P  |P ) = vPi (ϕ (y)) by Definition 3.4.3. Conversely, we have

to prove that the conditions

d(Pi |P ) = vPi (ϕ (y)) for i = 1, . . . , r (3.70)

imply
 3.5 The Different 111


n−1
OP ⊆ OP · y i . (3.71)
i=0
n−1
(The inclusion OP ⊇ i=0 OP · y i is trivial.) Let z ∈ OP , say


n−1
z= ti y i with ti ∈ F .
i=0

Observe that cj ∈ OP by (3.64) and CP = ϕ1(y) ·OP by (3.70) and Proposition
3.4.2(c); therefore  
1
TrF /F · cj · z ∈ OP .
ϕ (y)
Since   n−1 
1  cj
TrF /F · cj · z = TrF /F ti · · yi = tj ,
ϕ (y) i=0
ϕ (y)

we conclude tj ∈ OP . This proves (3.71).

Corollary 3.5.11. Let F  = F (y) be a finite separable extension of function

fields of degree [F  : F ] = n, and let ϕ(T ) ∈ F [T ] be the minimal polynomial
of y over F . Suppose P ∈ IPF satisfies

ϕ(T ) ∈ OP [T ] and vP  (ϕ (y)) = 0

for all P  ∈ IPF  with P  |P . Then P is unramified in F  /F , and {1, y, . . . , y n−1 }

is an integral basis for F  /F at P .

Proof. We have by Theorem 3.5.10

0 ≤ d(P  |P ) ≤ vP  (ϕ (y)) ≤ 0

for all P  |P , hence vP  (ϕ (y)) = d(P  |P ) = 0. The corollary follows immedi-
ately from Dedekind’s Different Theorem and Theorem 3.5.10(b).

We now give a second simple criterion under which the powers of one
element constitute an integral basis at a place P .

Proposition 3.5.12. Let F  /F be a finite separable extension of function

fields, P ∈ IPF and P  ∈ IPF  with P  |P . Suppose that P  |P is totally ramified;
i.e., e(P  |P ) = [F  : F ] = n. Let t ∈ F  be a P  -prime element, and consider
the minimal polynomial ϕ(T ) ∈ F [T ] of t over F . Then d(P  |P ) = vP  (ϕ (t)),
and {1, t, . . . , tn−1 } is an integral basis for F  /F at P .

Proof. First we claim that 1, t, . . . , tn−1 are linearly independent over F . As-
sume the contrary, so that
112 3 Extensions of Algebraic Function Fields


n−1
ri ti = 0 with ri ∈ F, not all ri = 0 .
i=0

For ri = 0 we have
vP  (ri ti ) = vP  (ti ) + e(P  |P ) · vP (ri ) ≡ i mod n .
Therefore vP  (ri ti ) = vP  (rj tj ) whenever i = j, ri =
 0 and rj = 0. The Strict
Triangle Inequality yields
n−1
 
vP  ri t = min{vP  (ri ti ) | ri =
i
 0} < ∞ ,
i=0

which is a contradiction. Thus {1, t, . . . , tn−1 } is a basis of F  /F .


According to the formula ei fi = n (Theorem 3.1.11), P  is the only
place of F lying over P , hence OP  is the integral closure of OP in F  . So we


have to show that


n−1
OP  = OP · ti . (3.72)
i=0
The inclusion ⊇ is trivial, so consider an element z ∈ OP  . Write

n−1
z= xi ti with xi ∈ F .
i=0

Since 0 ≤ vP  (z) = min{n · vP (xi ) + i | 0 ≤ i ≤ n − 1} by the above argument,

we see that vP (xi ) ≥ 0 holds for all i. This proves (3.72). The assertion
d(P  |P ) = vP  (ϕ (t)) now follows from Theorem 3.5.10.

In later sections we shall give several examples how the above results –
combined with the Hurwitz Genus Formula – can be applied to determine the
genus of a function field.

3.6 Constant Field Extensions

We consider an algebraic function field F/K with constant field K, where K
is assumed to be perfect as always in Chapter 3. This assumption is essential
for the validity of most results in this section. For example there are coun-
terexamples to most assertions of Proposition 3.6.1 and Theorem 3.6.3, when
K is not perfect (see [7]). Recall that Φ ⊇ F denotes a fixed algebraically
closed field.
Let K  ⊇ K be an algebraic extension (with K  ⊆ Φ). The compositum
F := F K  is a function field over K  , and its constant field is thereby a finite


extension of K  (cf. Corollary 1.1.16). However it is not clear a priori, whether

K  is the full constant field of F K  . So we begin this section with the following
result:
 3.6 Constant Field Extensions 113

Proposition 3.6.1. Let F  = F K  be an algebraic constant field extension of

F/K (of finite or infinite degree). Then we have:
(a) K  is the full constant field of F  .
(b) Each subset of F that is linearly independent over K remains so over K  .
(c) [F : K(x)] = [F  : K  (x)] for every element x ∈ F \K.

For the proof of this proposition we require a simple lemma that generalizes
Lemma 3.1.10.

Lemma 3.6.2. Suppose α ∈ Φ is algebraic over K. Then [K(α) : K] =

[F (α) : F ].

Proof of the Lemma. The inequality [F (α) : F ] ≤ [K(α) : K] being trivial,

we only have to prove that the minimal polynomial ϕ(T ) ∈ K[T ] of α over
K remains irreducible in F [T ]. Assume the contrary, so ϕ(T ) = g(T ) · h(T )
with monic polynomials g(T ), h(T ) ∈ F [T ] of degree ≥ 1. Each root of g(T )
and h(T ) in Φ is a root of ϕ(T ) as well, hence algebraic over K. Therefore
all coefficients of g(T ) and h(T ) are algebraic over K (observe that they are
polynomial expressions of the roots). On the other hand, these coefficients are
elements of F . Since K is algebraically closed in F we conclude that g(T ),
h(T ) ∈ K[T ], a contradiction to the irreducibility of ϕ(T ) over K.

Proof of Proposition 3.6.1. (a) Consider an element γ ∈ F  which is algebraic

over K  . Then it is algebraic over K, and there are finitely many elements
α1 , . . . , αr ∈ K  such that γ ∈ F (α1 , . . . , αr ). The extension K(α1 , . . . , αr )/K
is finite and separable, therefore K(α1 , . . . , αr ) = K(α) for some α ∈ K  (here
we use the assumption that K is perfect). Since γ is algebraic over K, we can
find β ∈ F  with K(α, γ) = K(β). It follows that F (β) = F (α, γ) = F (α) (as
γ ∈ F (α1 , . . . , αr ) = F (α)), and we obtain from Lemma 3.6.2

[K(β) : K] = [F (β) : F ] = [F (α) : F ] = [K(α) : K] .

This implies K(α) = K(β), hence γ ∈ K(α) ⊆ K  .

(b) Let y1 , . . . , yr ∈ F be linearly independent over K, and consider a linear
combination
r
γi yi = 0 with γi ∈ K  . (3.73)
i=1

Choose α ∈ K such that γ1 , . . . , γr ∈ K(α), and write


n−1
γi = cij αj with cij ∈ K, n = [K(α) : K] .
j=0

From (3.73) we obtain

114 3 Extensions of Algebraic Function Fields
r n−1
    
n−1 r 
0= cij αj yi = cij yi αj (3.74)
i=1 j=0 j=0 i=1

with cij yi ∈ F . Since [F (α) : F ] = [K(α) : K] by Lemma 3.6.2, the
elements 1, α, . . . , αn−1 are linearly independent over F , and (3.74) implies

r
cij yi = 0 for j = 0, . . . , n − 1 .
i=1

As y1 , . . . , yr are linearly independent over K, it follows that all cij = 0, and

thus (3.73) is the trivial linear combination.
(c) Clearly [F  : K  (x)] ≤ [F : K(x)]. It remains to show that any elements
z1 , . . . , zs ∈ F , which are linearly independent over K(x), are linearly inde-
pendent over K  (x) as well. Suppose not, so that

s
fi (x) · zi = 0 (3.75)
i=1

with fi (x) ∈ K  (x), not all fi (x) = 0. Multiplying by a common denominator

we can assume that all fi (x) ∈ K  [x]. Then (3.75) gives a linear dependence
of the set {xj zi | 1 ≤ i ≤ s, j ≥ 0} over K  . Part (b) of our proposition
implies that this set is then linearly dependent over K as well; so z1 , . . . , zs
are linearly dependent over K(x), a contradiction.
Our next theorem contains a summary of the most important properties
of constant field extensions.

Theorem 3.6.3. In an algebraic constant field extension F  = F K  of F/K

the following hold:
(a) F  /F is unramified (i.e., e(P  |P ) = 1 for all P ∈ IPF and all P  ∈ IPF 
with P  |P ).
(b) F  /K  has the same genus as F/K.
(c) For each divisor A ∈ Div(F ) we have deg ConF /F (A) = deg A.
(d) For each divisor A ∈ Div(F ),

(ConF /F (A)) = (A) .

More precisely: Every basis of L (A) is also a basis of L (ConF /F (A)). (Note
that L (A) ⊆ F is a K-vector space whereas L (ConF /F (A)) is considered as
a vector space over K  .)
(e) If W is a canonical divisor of F/K then ConF /F (W ) is a canonical divisor
of F  /K  .
(f ) The conorm map ConF /F : Cl(F ) → Cl(F  ) from the divisor class group
of F/K into that of F  /K  is injective.
 3.6 Constant Field Extensions 115

(g) The residue class field FP  of each place P  ∈ IPF  is the compositum FP K 
of K  and the residue class field FP , where P = P  ∩ F .
(h) If K  /K is of finite degree, every basis of K  /K is an integral basis of
F  /F for all P ∈ IPF .
Proof. The proof is organized as follows: first we discuss (a) and (b) in the
case of a finite constant field extension, then we prove (h), (a), (c), (b), (d),
(e), (f) and (g) in the general case. To begin with, we assume that
K  = K(α) is a finite extension of K . (3.76)
We shall prove (a) and (b) under this additional hypothesis. In this situation,
F  = F (α) and the minimal polynomial ϕ(T ) of α over K remains irreducible
over F by Lemma 3.6.2. Let P ∈ IPF and P  ∈ IPF  with P  |P . The different
exponent d(P  |P ) satisfies
0 ≤ d(P  |P ) ≤ vP  (ϕ (α))
by Theorem 3.5.10. Now α is separable over K, and therefore ϕ (α) = 0. Since
ϕ (α) ∈ K  , this implies vP  (ϕ (α)) = 0. So we have
d(P  |P ) = vP  (ϕ (α)) = 0 . (3.77)
By Dedekind’s Theorem we conclude that P  |P is unramified. The Hurwitz
Genus Formula yields
[F  : F ]
2g  − 2 = (2g − 2) + deg Diff(F  /F ) , (3.78)
[K  : K]
where g (resp. g  ) denotes the genus of F/K (resp. F  /K  ). In our situation
(3.76) we have [F  : F ] = [K  : K] by Lemma 3.6.2 and Diff(F  /F ) = 0 by
(3.77), therefore 2g  − 2 = 2g − 2 by (3.78). We have shown (a) and (b) in the
case of a finite constant field extension.
(h) We can assume that K  = K(α) and set n := [K  : K]. From (3.77) and
Theorem 3.5.10 we obtain that {1, α, . . . , αn−1 } is an integral basis of F  /F
for all P ∈ IPF . Obviously we have for each other basis {γ1 , . . . , γn } of K  /K

n−1 
n
OP · αi = OP · γj .
i=0 j=1

So {γ1 , . . . , γn } is an integral basis as well.

From now on, K  is an arbitrary algebraic extension of K (of finite or
infinite degree).
(a) Let P  ∈ IPF  be an extension of P . We choose a P  -prime element t ∈ F  .
There exists an intermediate field K ⊆ K1 ⊆ K  such that the degree [K1 : K]
is finite and t ∈ F1 := F K1 . Let P1 := P  ∩ F1 , then 1 = vP  (t) = e(P  |P1 ) ·
vP1 (t) and therefore e(P  |P1 ) = 1. We have already proved that e(P1 |P ) = 1,
consequently e(P  |P ) = e(P  |P1 ) · e(P1 |P ) = 1.
116 3 Extensions of Algebraic Function Fields

(c) It is sufficient to consider a prime divisor P ∈ IPF . Choose x ∈ F such that

P is the only zero of x in IPF (such an element exists by Proposition 1.6.6),
so the zero divisor (x)F F
0 of x in Div(F ) has the form (x)0 = rP with r > 0.
It follows from Proposition 3.1.9 that

0 = ConF /F ((x)0 ) = r · ConF /F (P ) .
(x)F F


Now we use the fact that [F  : K  (x)] = deg ((x)F
0 ) (see Theorem 1.4.11), and
we obtain
r · deg ConF /F (P ) = [F  : K  (x)]
= [F : K(x)] (by Proposition 3.6.1)
= deg ((x)F
0) = r · deg P .
Thus (c) is established.
(b) As a first step, we show that
(A) ≤ (ConF /F (A)) (3.79)
holds for A ∈ Div(F ). Indeed, if {x1 , . . . , xr } is a basis of the space L (A)
then xi ∈ L (ConF /F (A)) by Proposition 3.1.9, and x1 , . . . , xr are linearly
independent over K  by Proposition 3.6.1. This proves (3.79).
Let g (resp. g  ) denote the genus of F/K (resp. F  /K  ). Choose a divisor
C ∈ Div(F ) satisfying
deg C ≥ max{2g − 1, 2g  − 1} . (3.80)
The Riemann-Roch Theorem states that
(C) = deg C + 1 − g (3.81)
and
(ConF /F (C)) = deg C + 1 − g  . (3.82)
Here we have used the fact that deg ConF /F (C) = deg C by (c). Now (3.79),
(3.80) and (3.81) imply g  ≥ g.
In order to prove the reverse inequality g ≤ g  , consider a basis {u1 , . . . , us }
of L (ConF /F (C)). There exists a field K ⊆ K0 ⊆ K  with [K0 : K] < ∞ and
u1 , . . . , us ∈ F0 := F K0 . Obviously u1 , . . . , us ∈ L (ConF0/F (C)), thus
(ConF0/F (C)) ≥ (ConF /F (C)) . (3.83)
We have shown above that F0 /K0 has genus g (since it is a finite constant
field extension of F/K), so the Riemann-Roch Theorem yields
(ConF0/F (C)) = deg C + 1 − g . (3.84)
Combining (3.82), (3.83) and (3.84) we obtain g ≤ g  , and the proof of (b) is
complete.
 3.6 Constant Field Extensions 117

(d) Suppose first that deg A ≥ 2g −1. As g  = g we have by the Riemann-Roch

Theorem and (c)
(ConF /F (A)) = deg ConF /F (A) + 1 − g 
= deg A + 1 − g = (A) .
The argument which we used in the proof of (3.79) shows that every basis of
L (A) is also a basis of L (ConF /F (A)).
Now consider an arbitrary divisor A ∈ Div(F ) and a basis {x1 , . . . , xr }
of L (A). Since x1 , . . . , xr ∈ L (ConF /F (A)), and since they are linearly in-
dependent over K  , it remains to prove that each z ∈ L (ConF /F (A)) is a
K  -linear combination of x1 , . . . , xr . We choose prime divisors P1 = P2 of
F/K and set A1 := A + n1 P1 and A2 := A + n2 P2 with n1 , n2 ≥ 0 such that
deg Ai ≥ 2g − 1 for i = 1, 2. Then
A = min{A1 , A2 } and L (A) = L (A1 ) ∩ L (A2 ) .
We extend {x1 , . . . , xr } to bases {x1 , . . . , xr , y1 , . . . , ym } of L (A1 ) resp.
{x1 , . . . , xr , z1 , . . . , zn } of L (A2 ). The elements
x1 , . . . , xr , y1 , . . . , ym , z1 , . . . , zn (3.85)
are linearly independent over K. Indeed, if

r 
m 
n
ai xi + bj y j + ck zk = 0
i=1 j=1 k=1

with ai , bj , ck ∈ K, then

r 
m 
n
ai xi + bj y j = − ck zk ∈ L (A1 ) ∩ L (A2 ) = L (A) .
i=1 j=1 k=1

Since {x1 , . . . , xr } is a basis of L (A) and x1 , . . . , xr , y1 , . . . , ym are linearly

independent, this implies bj = 0 (j = 1, . . . , m), and then (by the linear inde-
pendence of x1 , . . . , xr , z1 , . . . , zn ) it follows that ai = ck = 0 for 1 ≤ i ≤ r,
1 ≤ k ≤ n. Observing that elements of F which are linearly independent
over K remain linearly independent over K  by Proposition 3.6.1, we have
established that the elements (3.85) are linearly independent over K  .
Now let z ∈ L (ConF /F (A)). Since deg Ai ≥ 2g − 1, assertion (d) holds for
Ai , and we can write

r 
m 
r 
n
z= di xi + ej yj = fi xi + gk zk (3.86)
i=1 j=1 i=1 k=1

with di , ej , fi , gk ∈ K  . Since x1 , . . . , xr , y1 , . . . , ym , z1 , . . . , zn are linearly

independent, the two representations in (3.86) coincide, hence ej = gk = 0 for
1 ≤ j ≤ m, 1 ≤ k ≤ n. Thus z is a linear combination of x1 , . . . , xr over K  .
118 3 Extensions of Algebraic Function Fields

(e) If W is a canonical divisor of F/K then deg W = 2g − 2 and (W ) = g.

By all that we have proved above, we obtain
deg ConF /F (W ) = 2g − 2 and (ConF /F (W )) = g .
These two properties characterize canonical divisors of F  /K  (see Proposition
1.6.2), so ConF /F (W ) is a canonical divisor of F  /K  .
(f) Since ConF /F : Cl(F ) → Cl(F  ) is a homomorphism, we must show that
the kernel is trivial. So we consider a divisor A ∈ Div(F ) whose conorm in F 
is principal. This means that
deg ConF /F (A) = 0 and (ConF /F (A)) = 1 .
Therefore deg A = 0 and (A) = 1 by (c) and (d), and this implies that A is
principal, by Corollary 1.4.12.
(g) Let z(P  ) ∈ FP  where z is an element of OP  . There is an intermediate
field K ⊆ K1 ⊆ K  with z ∈ F1 := F K1 and [K1 : K] < ∞. Set P1 := P  ∩ F1
and let P2 , . . . , Pr be the other places of F1 /K1 lying over P . Choose u ∈ F1
such that
vP1 (z − u) > 0 and vPi (u) ≥ 0 for 2 ≤ i ≤ r .
Then z(P  ) = u(P  ), and u lies in the integral closure of OP in F1 (see
Corollary 3.3.5). By (h),

n
u= γi xi with γi ∈ K1 , xi ∈ OP .
i=1

Consequently

n
z(P  ) = u(P  ) = γi · xi (P ) ∈ FP K  .
i=1

We can combine Theorem 3.6.3(c) and Corollary 3.1.14 to obtain a formula

for the degree of the conorm of a divisor in arbitrary algebraic extensions of
function fields (over a perfect constant field).
Corollary 3.6.4. Let F  /K  be an algebraic extension of F/K (not necessar-
ily a constant field extension). Then we have for each divisor A ∈ Div(F ),
deg ConF /F (A) = [F  : F K  ] · deg A .
Proof. By Lemma 3.1.2 we know that [F  : F K  ] < ∞, and F K  /K  is a
constant field extension of F/K. Since
ConF /F (A) = ConF  /F K  (ConF K  /F (A)) ,
we obtain that
deg ConF /F (A) = [F  : F K  ] · deg ConF K  /F (A) = [F  : F K  ] · deg A ,
by Corollary 3.1.14 and Theorem 3.6.3.
 3.6 Constant Field Extensions 119

The following result is a simple consequence of Theorem 3.6.3(a) and (c).

Corollary 3.6.5. Let P ∈ IPF be a place of F/K of degree r and let F̄ = F K̄

be the constant field extension of F/K with the algebraic closure K̄ of K.
Then
ConF̄ /F (P ) = P̄1 + . . . + P̄r
with pairwise distinct places P̄i ∈ IPF̄ .

We finish this section with a result that often allows us to show that the
constant field of a finite extension F  ⊇ F of a function field F/K is no bigger
than K.

Proposition 3.6.6. Let F/K be a function field with constant field K. Sup-
pose that F  /F is a finite extension field, with constant field K  . Let K̄ ⊆ Φ
denote the algebraic closure of K. Then

[F  : F ] = [F  K̄ : F K̄] · [K  : K] . (3.87)

In the special case F  = F (y) we obtain: if ϕ(T ) ∈ F [T ] is the minimal

polynomial of y over F , the following conditions are equivalent:
(1) K  = K.
(2) ϕ(T ) is irreducible in F K̄[T ].

Proof. Since F ⊆ F K  ⊆ F  , we have

[F  : F ] = [F  : F K  ] · [F K  : F ] . (3.88)

The extension K  /K is separable and of finite degree, hence K  = K(α) for

some α ∈ K  , and we obtain from Lemma 3.6.2 that

[F K  : F ] = [K  : K] . (3.89)

Proposition 3.6.1(c) shows that for each x ∈ F \K,

[F K  : K  (x)] = [F K̄ : K̄(x)] and [F  : K  (x)] = [F  K̄ : K̄(x)] .

This implies
[F  : F K  ] = [F  K̄ : F K̄] . (3.90)
Substituting (3.89) and (3.90) into (3.88) yields (3.87).
Consider now the case F  = F (y). Observe that [F  : F ] = deg ϕ(T ),
and [F  K̄ : F K̄] equals the degree of the minimal polynomial of y over F K̄
(which divides ϕ(T ) in F K̄[T ]). The equivalence of (1) and (2) is therefore an
immediate consequence of (3.87).
120 3 Extensions of Algebraic Function Fields

Corollary 3.6.7. Let F/K be a function field and let F  /F be a finite ex-
tension such that K is the full constant field of F and of F  . Let L/K be an
algebraic extension. Then L is the full constant field of F L and F  L, and we
have
[F  : F ] = [F  L : F L] .

Proof. It was already shown in Proposition 3.6.1 that L is the full constant
field of F L and of F  L. Let K̄ ⊇ L be the algebraic closure of K, then we
obtain from Proposition 3.6.6

[F  : F ] = [F  K̄ : F K̄]

and
[F  L : F L] = [F  LK̄ : F LK̄] = [F  K̄ : F K̄] ,
hence [F  : F ] = [F  L : F L].

A polynomial ϕ(T ) ∈ K(x)[T ] (over the rational function field K(x)) is

said to be absolutely irreducible if ϕ(T ) is irreducible in the polynomial ring
K̄(x)[T ] (where K̄ is the algebraic closure of K). The following corollary is a
special case of the Proposition 3.6.6.

Corollary 3.6.8. Let F = K(x, y) be a function field and ϕ(T ) ∈ K(x)[T ] be

the minimal polynomial of y over K(x). The following conditions are equiva-
lent:
(1) K is the full constant field of F .
(2) ϕ(T ) is absolutely irreducible.

3.7 Galois Extensions I

In Sections 3.7 and 3.8 we investigate Galois extensions of algebraic function
fields. Galois extensions have several useful properties that do not hold in
arbitrary finite extensions. Recall that a finite field extension M/L is said to
be a Galois extension if the automorphism group

Aut(M/L) = {σ : M → M | σ is an isomorphism with

σ (a) = a for all a ∈ L}

has order [M : L]. In that case we call Aut(M/L) the Galois group of M/L
and write Gal(M/L) := Aut(M/L). The main properties of Galois extensions
are collected together in Appendix A.
An extension F  /K  of a function field F/K is said to be Galois if F  /F
is a Galois extension of finite degree.
 3.7 Galois Extensions I 121

Let P be a place of F/K. Then Gal(F  /F ) acts on the set of all extensions
{P  ∈ IPF  | P  lies over P } via σ(P  ) = {σ(x) | x ∈ P  }, and we have seen in
Lemma 3.5.2 that the corresponding valuation vσ(P  ) is given by

vσ(P  ) (y) = vP  (σ −1 (y)) for y ∈ F  .

Theorem 3.7.1. Let F  /K  be a Galois extension of F/K and P1 , P2 ∈ IPF 

be extensions of P ∈ IPF . Then P2 = σ(P1 ) for some σ ∈ Gal(F  /F ). In other
words, the Galois group acts transitively on the set of extensions of P .

Proof. Assume that the assertion is false; i.e., σ(P1 ) = P2 for all σ ∈ G :=
Gal(F  /F ). By the Approximation Theorem there is an element z ∈ F  such
that vP2 (z) > 0 and vQ (z) = 0 for all Q ∈ IPF  with Q|P and Q = P2 . Let
NF /F : F  → F be the norm map (see Appendix A). We obtain
  
vP1 (NF /F (z)) = vP1 σ(z) = vP1 (σ(z))
σ∈G σ∈G
 
= vσ−1 (P1 ) (z) = vσ(P1 ) (z) = 0 , (3.91)
σ∈G σ∈G

since P2 does not occur among the places σ(P1 ), σ ∈ G. On the other hand,

vP2 (NF /F (z)) = vσ(P2 ) (z) > 0 . (3.92)
σ∈G

But NF /F (z) ∈ F , therefore

vP1 (NF /F (z)) = 0 ⇐⇒ vP (NF /F (z)) = 0 ⇐⇒ vP2 (NF /F (z)) = 0 .

This is a contradiction to (3.91) and (3.92).

Corollary 3.7.2. Notation as in Theorem 3.7.1 (in particular F  /F is a

Galois extension). Let P1 , . . . , Pr be all the places of F  lying over P . Then
we have:
(a) e(Pi |P ) = e(Pj |P ) and f (Pi |P ) = f (Pj |P ) for all i, j. Therefore we set

e(P ) := e(Pi |P ) and f (P ) := f (Pi |P ) ,

and we call e(P ) (resp. f (P )) the ramification index (resp. relative degree) of
P in F  /F .
(b) e(P ) · f (P ) · r = [F  : F ]. In particular e(P ), f (P ) and r divide the degree
[F  : F ].
(c) The different exponents d(Pi |P ) and d(Pj |P ) are the same for all i, j.
122 3 Extensions of Algebraic Function Fields

Proof. (a) is obvious by Theorem 3.7.1 and Lemma 3.5.2, and (b) is an imme-
diate consequence of (a) and Theorem 3.1.11. As to (c), we have to consider
the integral closure
r
OP = OPi
i=1

of OP in F , and the complementary module
CP = {z ∈ F  | TrF /F (z · OP ) ⊆ OP } .
Let σ ∈ Gal(F  /F ). It is easily seen that σ(OP ) = OP and σ(CP ) = CP (using
the fact that TrF /F (σ(u)) = TrF /F (u) for u ∈ F  ). Writing CP = t · OP we
obtain σ(t) · OP = σ(CP ) = CP = t · OP , so that
−d(Pi |P ) = vPi (t) = vPi (σ(t))
for 1 ≤ i ≤ r (by Proposition 3.4.2(c) and the definition of the different expo-
nent). Consider now two places Pi , Pj lying over P . Choose σ ∈ Gal(F  /F )
such that σ(Pj ) = Pi . Then
−d(Pi |P ) = vPi (σ(t)) = vσ−1 (Pi ) (t) = vPj (t) = −d(Pj |P ) .

We would like to discuss two special types of Galois extensions of a func-

tion field in more detail, namely Kummer extensions and Artin-Schreier ex-
tensions.
Proposition 3.7.3 (Kummer Extensions). Let F/K be an algebraic func-
tion field where K contains a primitive n-th root of unity (with n > 1 and n
relatively prime to the characteristic of K). Suppose that u ∈ F is an element
satisfying
u = wd for all w ∈ F and d | n, d > 1 . (3.93)
Let
F  = F (y) with y n = u . (3.94)

Such an extension F /F is said to be a Kummer extension of F . We have:
(a) The polynomial Φ(T ) = T n − u is the minimal polynomial of y over F (in
particular, it is irreducible over F ). The extension F  /F is Galois of degree
[F  : F ] = n; its Galois group is cyclic, and the automorphisms of F  /F are
given by σ(y) = ζy, where ζ ∈ K is an n-th root of unity.
(b) Let P ∈ IPF and P  ∈ IPF  be an extension of P . Then
n n
e(P  |P ) = and d(P  |P ) = − 1,
rP rP
where
rP := gcd(n, vP (u)) > 0 (3.95)
is the greatest common divisor of n and vP (u).
 3.7 Galois Extensions I 123

(c) If K  denotes the constant field of F  and g (resp. g  ) the genus of F/K
(resp. F  /K  ), then
 
n 1   rP 
g = 1 + g − 1 + 1 − deg P ,
[K  : K] 2 n
P ∈IPF

where rP is defined by (3.95).

We note that every cyclic field extension F  /F of degree n is a Kummer

extension, provided that n is relatively prime to the characteristic of F and
F contains all n-th roots of unity. This fact is well-known from Galois theory,
cf. Appendix A.
The following special case of Proposition 3.7.3 is worth emphasizing.

Corollary 3.7.4. Let F/K be a function field and F  = F (y) with y n = u ∈

F , where n ≡ 0 mod(char K) and K contains a primitive n-th root of unity.
Assume there is a place Q ∈ IPF such that gcd(vQ (u), n) = 1. Then K is the
full constant field of F  , the extension F  /F is cyclic of degree n, and
1 
g  = 1 + n(g − 1) + (n − rP )deg P .
2
P ∈IPF

Proof of Proposition 3.7.3. (a) cf. Appendix A.

(b) Case 1. rP = 1. From (3.94) we obtain

n · vP  (y) = vP  (y n ) = vP  (u) = e(P  |P ) · vP (u) ,

which implies e(P  |P ) = n, as n and vP (u) are relatively prime. Since n is

not divisible by char K, Dedekind’s Different Theorem yields the different
exponent d(P  |P ) = n − 1.
Case 2. rP = n, say vP (u) = l · n with l ∈ ZZ. We choose t ∈ F with
vP (t) = l and set
y1 := t−1 y, u1 := t−n u .
Then y1n = u1 , vP  (y1 ) = vP (u1 ) = 0, and the irreducible polynomial of y1
over F is
ψ(T ) = T n − u1 ∈ F [T ] .
Thus y1 is integral over OP , and Theorem 3.5.10 yields

0 ≤ d(P  |P ) ≤ vP  (ψ  (y1 )) .

Now ψ  (y1 ) = n · y1n−1 , so vP  (ψ  (y1 )) = (n − 1) · vP  (y1 ) = 0 and d(P  |P ) = 0.

By Dedekind’s Theorem, e(P  |P ) = 1, and (b) is established in Case 2.
Case 3. 1 < rP < n. Consider the intermediate field
124 3 Extensions of Algebraic Function Fields

F0 := F (y0 ) with y0 := y n/rP .

Then [F  : F0 ] = n/rP and [F0 : F ] = rP . The element y0 satisfies the

equation
y0rP = u (3.96)
over F . Let P0 := P  ∩ F0 . Case 2 applies to F0 /F , and therefore e(P0 |P ) = 1.
By (3.96)
vP (u)
vP0 (y0 ) = .
rP
This is relatively prime to n/rP , so Case 1 applies to the extension F  = F0 (y)
(note that y n/rP = y0 ). Consequently e(P  |P0 ) = n/rP and

e(P  |P ) = e(P  |P0 ) · e(P0 |P ) = n/rP .

(c) The degree of the different Diff(F  /F ) is

 
deg Diff(F  /F ) = d(P  |P ) · deg P 
P ∈IPF P  |P
 n  
= −1 · deg P  (by (b)) . (3.97)
rP
P ∈IPF  P |P

Observing that for a fixed place P ∈ IPF the ramification index e(P ) = e(P  |P )
does not depend on the choice of the extension P  , we have
  
 1  
deg P = · deg e(P |P ) · P

e(P ) 
P |P P |P
1 rP n
= · deg ConF /F (P ) = · · deg P
e(P ) n [K  : K]
rP
= · deg P ,
[K  : K]
by (b) and Corollary 3.1.14. Substituting this into (3.97) shows that
 n − rP rP
deg Diff(F  /F ) = · · deg P
rP [K  : K]
P ∈IPF
n   rP 
= · 1 − deg P .
[K  : K] n
P ∈IPF

Finally, the Hurwitz Genus Formula proves (c).

Proof of Corollary 3.7.4. From the assumption gcd(vQ (u), n) = 1 follows eas-
ily that u satisfies condition (3.93). It remains to show that the constant field
K  of F  is no larger than K; then the corollary follows immediately from
Proposition 3.7.3. Choose an extension Q of Q in F  . Part (b) of the
proposition shows that
 3.7 Galois Extensions I 125

e(Q |Q) = [F  : F ] = n . (3.98)

Now suppose that [K  : K] > 1, and consider the intermediate field F1 :=

F K   F and the place Q1 := Q ∩ F1 . By (3.98) we have that e(Q1 |Q) =
[F1 : F ] > 1. On the other hand, e(Q1 |Q) = 1 since F1 /F is a constant field
extension (see Theorem 3.6.3). This contradiction proves that K  = K.

Remark 3.7.5. In the above proofs we never used the assumption that K con-
tains a primitive n-th root of unity. Therefore all assertions of Proposition
3.7.3(b),(c) and Corollary 3.7.4 hold in this more general case, with a single
exception: F (y)/F is no longer Galois if K does not contain all n-th roots of
unity.

Thus far we have not given explicit examples of function fields of genus
g > 0. This is easily done now.

Example 3.7.6. Assume char K = 2. Let F = K(x, y) with

y 2 = f (x) = p1 (x) · . . . · ps (x) ∈ K[x] ,

where p1 (x), . . . , ps (x) are distinct irreducible monic polynomials and s ≥ 1.

Let m = deg f (x). Then K is the constant field of F , and F/K has genus

(m − 1)/2 if m ≡ 1 mod 2 ,
g=
(m − 2)/2 if m ≡ 0 mod 2 .

Proof. We have F = F0 (y) where F0 = K(x) is the rational function field.

Let Pi ∈ IPK(x) denote the zero of pi (x) and P∞ the pole of x in K(x). Then
vPi (f (x)) = 1 and vP∞ (f (x)) = −m. From Corollary 3.7.4 we obtain that
F/F0 is cyclic of degree 2 and that K is the constant field of F . The numbers
rP (for P ∈ IPK(x) ) are easily seen to be

rPi = 1 for i = 1, . . . , s ,
rP∞ = 1 if m ≡ 1 mod 2 ,
rP∞ = 2 if m ≡ 0 mod 2 .

Now the assertion follows from Corollary 3.7.4.

We shall return to the previous example in Chapter 6. As a preparation

for Artin-Schreier extensions we need a lemma.

Lemma 3.7.7. Let F/K be an algebraic function field of characteristic p > 0.

Given an element u ∈ F and a place P ∈ IPF , the following holds:
(a) either there exists an element z ∈ F such that vP (u − (z p − z)) ≥ 0 ,
126 3 Extensions of Algebraic Function Fields

(b) or else, for some z ∈ F ,

vP (u − (z p − z)) = −m < 0 with m ≡ 0 mod p .

In the latter case the integer m is uniquely determined by u and P , namely

−m = max{vP (u − (wp − w)) | w ∈ F } . (3.99)

Proof. We begin by proving the following claim. Assume that x1 , x2 ∈ F \{0}

and vP (x1 ) = vP (x2 ). Then there is some y ∈ F with

vP (y) = 0 and vP (x1 − y p x2 ) > vP (x1 ) . (3.100)

Indeed, the residue class (x1 /x2 )(P ) ∈ OP /P is not zero, hence (x1 /x2 )(P ) =
(y(P ))p for some y ∈ OP \P (here the perfectness of OP /P is essential). This
implies vP (y) = 0 and vP ((x1 /x2 ) − y p ) > 0, thereby vP (x1 − y p x2 ) > vP (x1 ).
Next we show: if vP (u − (z1p − z1 )) = −lp < 0, then there is an element
z2 ∈ F with
vP (u − (z2p − z2 )) > −lp . (3.101)
In order to prove this, we choose t ∈ F with vP (t) = −l; then

vP (u − (z1p − z1 )) = vP (tp ) .

By (3.100) we can find y ∈ F with vP (y) = 0 and

vP (u − (z1p − z1 ) − (yt)p ) > −lp .

Since vP (yt) = vP (t) = −l > −lp,

vP (u − (z1p − z1 ) − ((yt)p − yt)) > −lp .

Setting z2 := z1 + yt, we have established (3.101).

From (3.101), the existence of an element z ∈ F such that (a) (resp. (b))
holds, follows immediately. In case (b) we still have to prove the characteriza-
tion of m given in (3.99). By assumption, we have vP (u − (z p − z)) = −m < 0
with m ≡ 0 mod p. For every w ∈ F , p · vP (w − z) = −m holds, so we may
consider the following cases:
Case 1. p · vP (w − z) > −m. Then vP ((w − z)p − (w − z)) > −m and
vP (u − (wp − w)) = vP (u − (z p − z) − ((w − z)p − (w − z))) = −m (we have
used the Strict Triangle Inequality).
Case 2. p · vP (w − z) < −m. In this case we obtain vP (u − (wp − w)) =
vP (u − (z p − z) − ((w − z)p − (w − z))) < −m .
In either case vP (u − (wp − w)) ≤ −m, which proves (3.99).
 3.7 Galois Extensions I 127

Proposition 3.7.8 (Artin-Schreier Extensions). Let F/K be an alge-

braic function field of characteristic p > 0. Suppose that u ∈ F is an element
which satisfies the following condition:

u = wp − w for all w ∈ F . (3.102)

Let
F  = F (y) with y p − y = u . (3.103)
Such an extension F  /F is called an Artin-Schreier extension of F . For P ∈
IPF we define the integer mP by
⎧
⎪
⎨m if there is an element z ∈ F satisfying
mP := vP (u − (z p − z)) = −m < 0 and m ≡ 0 mod p ,
⎪
⎩
−1 if vP (u − (z p − z)) ≥ 0 for some z ∈ F .

(Observe that mP is well-defined by Lemma 3.7.7.) We then have:

(a) F  /F is a cyclic Galois extension of degree p. The automorphisms of F  /F
are given by σ(y) = y + ν, with ν = 0, 1, . . . , p − 1.
(b) P is unramified in F  /F if and only if mP = −1.
(c) P is totally ramified in F  /F if and only if mP > 0. Denote by P  the
unique place of F  lying over P . Then the different exponent d(P  |P ) is given
by
d(P  |P ) = (p − 1)(mP + 1) .
(d) If at least one place Q ∈ IPF satisfies mQ > 0, then K is algebraically
closed in F  and
  
 p−1
g =p·g+ −2 + (mP + 1) · deg P ,
2
P ∈IPF

where g  (resp. g) is the genus of F  /K (resp. F/K).

Proof. (a) This is well-known from Galois theory, see Appendix A.

(b) and (c) First we consider the case mP = −1; i.e., vP (u − (z p − z)) ≥ 0
for some z ∈ F . Let y1 = y − z and u1 = u − (z p − z); then F  = F (y1 ), and
ϕ1 (T ) = T p −T −u1 is the minimal polynomial of y1 over F . Since vP (u1 ) ≥ 0,
y1 is integral over the valuation ring OP , and the different exponent d(P  |P )
of an extension P  of P in F satisfies

0 ≤ d(P  |P ) ≤ vP  (ϕ1 (y1 )) = 0 ,

since ϕ1 (T ) = −1 (see Theorem 3.5.10). Hence d(P  |P ) = 0, and P  |P is

unramified by Dedekind’s Different Theorem.
Next we assume mP > 0. Choose z ∈ F such that vP (u−(z p −z)) = −mP .
Consider the elements y1 = y − z and u1 = u − (z p − z). As before we have
128 3 Extensions of Algebraic Function Fields

F  = F (y1 ), and ϕ1 (T ) = T p − T − u1 is the minimal polynomial of y1 over

F . Let P  ∈ IPF  be an extension of P in F  . Since y1p − y1 = u1 , we obtain
vP  (u1 ) = e(P  |P ) · vP (u1 ) = −mP · e(P  |P )
and
vP  (u1 ) = vP  (y1p − y1 ) = p · vP  (y1 ) .
As p and mP are relatively prime and e(P  |P ) ≤ [F  : F ] = p, this implies
that
e(P  |P ) = p and vP  (y1 ) = −mP .
In particular P is totally ramified in F  /F .
Let x ∈ F be a P -prime element. Choose integers i, j ≥ 0 such that
1 = ip − jmP (this is possible, as p and mP are relatively prime). Then the
element t = xi y1j is a P  -prime element, since vP  (t) = i · vP  (x) + j · vP  (y1 ) =
ip − jmP = 1. By Proposition 3.5.12, the different exponent d(P  |P ) is
d(P  |P ) = vP  (ψ  (t)) ,
where ψ(T ) ∈ F [T ] is the minimal polynomial of t over F . Let G :=
Gal(F  /F ) be the Galois group of F  /F . Clearly

ψ(T ) = (T − σ(t)) = (T − t) · h(T )
σ∈G

with 
h(T ) = (T − σ(t)) ∈ F  [T ] .
σ=id

So ψ (T ) = h(T ) + (T − t) · h (T ) and ψ  (t) = h(t). We conclude that

 

  
d(P  |P ) = vP  (t − σ(t)) = vP  (t − σ(t)) (3.104)
σ=id σ=id

(the sum runs over all σ ∈ G with σ = id). Each σ ∈ G\{id} has the form
σ(y1 ) = y1 + μ for some μ ∈ {1, . . . , p − 1}, hence
j  
 j
t − σ(t) = xi y1j − xi (y1 + μ)j = −xi · y1j−l μl .
l
l=1

Since vP  (y1j−1 ) < vP  (y1j−l ) for l ≥ 2, the Strict Triangle Inequality yields

vP  (t − σ(t)) = vP  (xi ) + vP  (jμy1j−1 )

= ip + (j − 1) · (−mP ) = ip − jmP + mP = mP + 1 . (3.105)
(We have used that j = 0 in K, which follows from ip−jmP = 1.) Substituting
(3.105) into (3.104) gives d(P  |P ) = (p − 1)(mP + 1). Thus we have shown (b)
and (c).
 3.7 Galois Extensions I 129

(d) We assume now that mQ > 0 for at least one place Q ∈ IPF . From (c)
follows that Q is totally ramified in F  /F . That K is the full constant field of
F  follows exactly as in the proof of Corollary 3.7.4.
The formula
  
p−1
g = p · g + −2 + (mP + 1) · deg P
2
P ∈IPF

is an immediate consequence of (b), (c) and the Hurwitz Genus Formula.

Remark 3.7.9. (a) Notation as in Proposition 3.7.8. Suppose that there exists
a place Q ∈ IPF with

vQ (u) < 0 and vQ (u) ≡ 0 mod p .

Then u satisfies condition (3.102) by the Strict Triangle Inequality. Hence

Proposition 3.7.8 applies in this case.
(b) All cyclic field extensions F  /F of degree [F  : F ] = p = char F > 0 are
Artin-Schreier extensions, see Appendix A.

Most of the arguments given in the proof of Proposition 3.7.8 apply in a

more general situation. We call a polynomial of the specific form
n n−1
a(T ) = an T p + an−1 T p + . . . + a1 T p + a0 T ∈ K[T ] (3.106)

(where p = charK > 0) an additive (or linearized) polynomial over K. Observe

that a(T ) is separable if and only if a(T ) and its derivative a (T ) have no
common factor of degree > 0. Here a (T ) = a0 is constant, so the polynomial
(3.106) is separable if and only if a0 = 0.
An additive polynomial has the following remarkable property:

a(u + v) = a(u) + a(v)

for any u, v in some extension field of K. In particular, if a(T ) is an additive

and separable polynomial over K all of whose roots are in K, then these roots
form a subgroup of the additive group of K of order pn = deg a(T ).

Proposition 3.7.10. Consider an algebraic function field F/K with constant

field K of characteristic p > 0, and an additive separable polynomial a(T ) ∈
K[T ] of degree pn which has all its roots in K. Let u ∈ F . Suppose that for
each P ∈ IPF there is an element z ∈ F (depending on P ) such that

vP (u − a(z)) ≥ 0 (3.107)

or
vP (u − a(z)) = −m with m > 0 and m ≡ 0 mod p . (3.108)
130 3 Extensions of Algebraic Function Fields

Define mP := −1 in case (3.107) and mP := m in case (3.108). Then mP is

a well-defined integer. Consider the extension field F  = F (y) of F where y
satisfies the equation
a(y) = u .
If there exists at least one place Q ∈ IPF with mQ > 0, the following hold:
(a) F  /F is Galois, [F  : F ] = pn , and the Galois group of F  /F is isomorphic
to the additive group {α ∈ K | a(α) = 0}, hence isomorphic to (ZZ/pZZ)n .
(Such a group is said to be elementary abelian of exponent p, hence F  /F is
called an elementary abelian extension of exponent p and degree pn .)
(b) K is algebraically closed in F  .
(c) Each P ∈ IPF with mP = −1 is unramified in F  /F .
(d) Each P ∈ IPF with mP > 0 is totally ramified in F  /F , and the different
exponent d(P  |P ) of the extension P  of P in F  is
d(P  |P ) = (pn − 1)(mP + 1) .
(e) Let g  (resp. g) be the genus of F  (resp. F ). Then
  
 pn − 1
g =p ·g+
n
−2 + (mP + 1) · deg P .
2
P ∈IPF

The proof of Proposition 3.7.10 can be omitted; it is, with minor modifi-
cations, the same as that of Proposition 3.7.8.

3.8 Galois Extensions II

We consider a Galois extension F  /F of algebraic function fields with
Galois group G := Gal(F  /F ). Let P be a place of F and let P  be an
extension of P to F  .
Definition 3.8.1. (a) GZ (P  |P ) := {σ ∈ G | σ(P  ) = P  } is called the
decomposition group of P  over P .
(b) GT (P  |P ) := {σ ∈ G | vP  (σz − z) > 0 for all z ∈ OP  } is called the
inertia group of P  |P .
(c) The fixed field Z := Z(P  |P ) of GZ (P  |P ) is called the decomposition field,
the fixed field T := T (P  |P ) of GT (P  |P ) is called the inertia field of P  over
P.
Clearly GT (P  |P ) ⊆ GZ (P  |P ), and both are subgroups of G.

An immediate consequence of the definitions is: for τ ∈ G the decomposi-

tion group and the inertia group of the place τ (P  ) are given by
GZ (τ (P  )|P ) = τ GZ (P  |P )τ −1 ,
GT (τ (P  )|P ) = τ GT (P  |P )τ −1
 3.8 Galois Extensions II 131

Theorem 3.8.2. With notation as above the following hold:

(a) The decomposition group GZ (P  |P ) has order e(P  |P ) · f (P  |P ).
(b) The inertia group GT (P  |P ) is a normal subgroup of GZ (P  |P ) of order
e(P  |P ).
(c) The residue class extension FP  /FP is a Galois extension. Each automor-
phism σ ∈ GZ (P  |P ) induces an automorphism σ̄ of FP  over FP by setting
σ̄(z(P  )) = σ(z)(P  ) for z ∈ OP  . The mapping

GZ (P  |P ) −→ Gal(FP  /FP ) ,
σ −→ σ̄ ,

is a surjective homomorphism whose kernel is the inertia group GT (P  |P ). In

particular, Gal(FP  /FP ) is isomorphic to GZ (P  |P )/GT (P  |P ).
(d) Let PZ (resp. PT ) denote the restriction of P  to the decomposition field
Z = Z(P  |P ) (resp. to the inertia field T = T (P  |P )). Then the ramification
indices and residue degrees of the places P  |PT , PT |PZ and PZ |P are as shown
in Figure 3.3 below.

F P
e(P  |PT ) = e(P  |P ) = [F  : T ]
and f (P  |PT ) = 1
T PT
f (PT |PZ ) = f (P  |P ) = [T : Z]
and e(PT |PZ ) = 1
Z PZ
e(PZ |P ) = f (PZ |P ) = 1

F P
Fig. 3.3.

Proof. (a) By Theorem 3.7.1 G acts transitively on the set of extensions of P

in F  . So we can choose σ1 , . . . , σr ∈ G such that σ1 (P  ), . . . , σr (P  ) are all
places of F  lying over P and σi (P  ) = σj (P  ) for i = j. Then σ1 , . . . , σr are a
complete set of coset representatives of G modulo GZ (P  |P ), hence [F  : F ] =
ord G = r · ord GZ (P  |P ). On the other hand, [F  : F ] = e(P  |P ) · f (P  |P ) · r
by Corollary 3.7.2(b). This proves (a).
Now we consider the restriction PZ = P  ∩ Z of P  to Z. Obviously the
decomposition group of P  over PZ (with respect to the extension F  /Z) equals
GZ (P  |P ), therefore e(P  |PZ ) · f (P  |PZ ) = ord GZ (P  |P ) = e(P  |P ) · f (P  |P )
132 3 Extensions of Algebraic Function Fields

by (a). Since e(P  |P ) = e(P  |PZ )·e(PZ |P ) and f (P  |P ) = f (P  |PZ )·f (PZ |P ),
this implies
e(PZ |P ) = f (PZ |P ) = 1 . (3.109)
Moreover, P  is the only extension of PZ in F  .
Next we prove (c). For z ∈OP  let z̄ := z(P  ) ∈ FP  denote  its residue
class mod P  , and for ψ(T ) = zi T i ∈ OP  [T ] set ψ̄(T ) := z̄i T i ∈ FP  [T ].
By our general assumption that the constant field of F is perfect, the residue
class extension FP  /FP is separable, hence FP  = FP (ū) for some u ∈ OP  .
We claim that FP  is the splitting field of some polynomial over FP (which
implies that FP  /FP is Galois). The place P  is the only extension of PZ in
F  /Z, so OP  is the integral closure in F  of the valuation ring OPZ of PZ (see
Corollary 3.3.5), and the minimal polynomial ϕ(T ) ∈ Z[T ] of u over Z has all
its coefficients in OPZ . As f (PZ |P ) = 1 by (3.109), we obtain z̄ ∈ FP for all
z ∈ OPZ , thus ϕ̄(T ) ∈ FP [T ]. The extension
 F  /Z being Galois, ϕ(T ) splits
completely into linear factors ϕ(T ) = (T − ui ) with ui ∈ OP  , so

ϕ̄(T ) = (T − ūi ) with ūi ∈ FP  . (3.110)

One of the roots of ϕ̄(T ) is ū, hence FP  is the splitting field of ϕ̄(T ) over FP .
Let σ ∈ GZ (P  |P ) and y, z ∈ OP  with ȳ = z̄. Then y − z ∈ P  , hence
σ(y) − σ(z) = σ(y − z) ∈ σ(P  ) = P  and σ(y)(P  ) = σ(z)(P  ). Therefore
σ̄ : FP  → FP  with σ̄(z(P  )) := σ(z)(P  ) is well-defined, and it is easily
verified that the mapping σ → σ̄ defines a homomorphism from GZ (P  |P )
into the Galois group of FP  over FP . The kernel of this homomorphism is
just GT (P  |P ), by definition of the inertia group.
An automorphism α ∈ Gal(FP  /FP ) is uniquely determined by α(ū), and
α(ū) is a root of the minimal polynomial of ū over FP . As this minimal
polynomial divides ϕ̄(T ), there is some root ui ∈ OP  of ϕ(T ) with α(ū) = ūi
(by (3.110)). Since ϕ(T ) is the minimal polynomial of u over Z and F  /Z is
Galois, there is an element σ ∈ Gal(F  /Z) = GZ (P  |P ) such that σ(u) = ui .
Clearly σ̄ = α, so our homomorphism from GZ (P  |P ) to Gal(FP  /FP ) is
surjective. The proof of (c) is now complete.
(b) GT (P  |P ) is a normal subgroup of GZ (P  |P ) since it is the kernel of the
homomorphism considered in (c). We have, by (c) and (a),

f (P  |P ) = [FP  : FP ] = ord Gal(FP  /FP )

= ord GZ (P  |P )/ord GT (P  |P )
= (e(P  |P ) · f (P  |P ))/ord GT (P  |P ) .

Consequently ord GT (P  |P ) = e(P  |P ).

(d) It follows from the definition that the inertia group of P  over PT is equal
to GT (P  |P ). Applying (b) first to the extension F  /T and then to F  /F we
obtain
 3.8 Galois Extensions II 133

e(P  |PT ) = ord GT (P  |P ) = e(P  |P ) . (3.111)

All assertions of (d) are now immediate consequences of (3.109), (3.111) and
the multiplicativity of ramification indices and relative degrees in towers of
fields (see Proposition 3.1.6(b)).

There are some useful characterizations of the decomposition field and the
inertia field:

Theorem 3.8.3. Consider a Galois extension F  /F of algebraic function

fields, a place P ∈ IPF and an extension P  of F in F  . For an interme-
diate field F ⊆ M ⊆ F  let PM := P  ∩ M denote the restriction of P  to M .
Then we have
(a) M ⊆ Z(P  |P ) ⇐⇒ e(PM |P ) = f (PM |P ) = 1.
(b) M ⊇ Z(P  |P ) ⇐⇒ P  is the only place of F  lying over PM .
(c) M ⊆ T (P  |P ) ⇐⇒ e(PM |P ) = 1.
(d) M ⊇ T (P  |P ) ⇐⇒ PM is totally ramified in F  /M .

Proof. By Theorem 3.8.2(d) all implications ⇒ are obvious. Before proving the
converse, we remark that the decomposition group of P  over PM is contained
in GZ (P  |P ), and the inertia group of P  over PM is contained in GT (P  |P )
(this follows immediately from the definition of these groups).
(a) Suppose that e(PM |P ) = f (PM |P ) = 1. Then e(P  |PM ) · f (P  |PM ) =
e(P  |P ) · f (P  |P ), so the decomposition group of P  over PM has the same
order as GZ (P  |P ) by Theorem 3.8.2(a). The above remark shows that
GZ (P  |P ) is equal to the decomposition group of P  over PM , in particu-
lar GZ (P  |P ) ⊆ Gal(F  /M ). By Galois theory this implies Z(P  |P ) ⊇ M .
(b), (c), (d) The proofs are similar.

In what follows, we shall study the phenomenon of wild ramification (see

Definition 3.5.4) in a Galois extension more closely.

Definition 3.8.4. Let F  /F be a Galois extension of algebraic function fields

with Galois group G = Gal(F  /F ). Consider a place P ∈ IPF and an extension
P  of P in F  . For every i ≥ −1 we define the i-th ramification group of P  |P
by
Gi (P  |P ) := {σ ∈ G | vP  (σz − z) ≥ i + 1 for all z ∈ OP  } .

Clearly Gi (P  |P ) is a subgroup of G. For abbreviation we write Gi :=

Gi (P  |P ).
134 3 Extensions of Algebraic Function Fields

Proposition 3.8.5. With the above notations we have:

(a) G−1 = GZ (P  |P ) and G0 = GT (P  |P ). In particular, ord G0 = e(P  |P ).
(b) G−1 ⊇ G0 ⊇ · · · ⊇ Gi ⊇ Gi+1 ⊇ · · · and Gm = {id} for m sufficiently
large.
(c) Let σ ∈ G0 , i ≥ 0 and let t be a P  -prime element; i.e., vP  (t) = 1. Then

σ ∈ Gi ⇐⇒ vP  (σt − t) ≥ i + 1 .

(d) If char F = 0 then Gi = {id} for all i ≥ 1, and G0 = GT (P  |P ) is cyclic.

(e) If char F = p > 0 then G1 is a normal subgroup of G0 . The order of G1 is
a power of p, and the factor group G0 /G1 is cyclic of order relatively prime
to p.
(f ) If char F = p > 0 then Gi+1 is a normal subgroup of Gi (for all i ≥ 1),
and Gi /Gi+1 is isomorphic to an additive subgroup of the residue class field
FP  . Hence Gi /Gi+1 is an elementary abelian p-group of exponent p.

Proof. (a) and (b) are obvious.

(c) Consider the inertia field T of P  over P , the restriction PT = P  ∩ T and
the corresponding valuation ring OPT = OP  ∩ T . The elements 1, t, . . . , te−1
(where e = e(P  |P )) constitute an integral basis for F  /T at PT , since P  |PT
is totally ramified (see Proposition 3.5.12). Suppose now that σ ∈ G0 =
e−1
Gal(F  /T ) satisfies vP  (σt−t) ≥ i+1, and let z ∈ OP  . Writing z = i=0 xi ti
with xi ∈ OPT we obtain


e−1 
e−1
σz − z = xi ((σt)i − ti ) = (σt − t) · xi ui ,
i=1 i=1

where ui = ((σt)i −ti )/(σt−t) ∈ OP  . This implies vP  (σz −z) ≥ vP  (σt−t) ≥

i + 1, hence σ ∈ Gi , and (c) is proved.
We denote by (FP  )× (resp. FP  ) the multiplicative (resp. additive) group
of the residue class field of F  at P  , and we shall establish the following facts:
There is a homomorphism

χ : G0 → (FP  )× with Ker(χ) = G1 , (3.112)

and for all i ≥ 1 there is a homomorphism

ψi : Gi → FP  with Ker(ψi ) = Gi+1 . (3.113)

The assertions (d), (e) and (f) are easy consequences of (3.112) and (3.113).
Indeed, since a finite subgroup of the multiplicative group of a field is cyclic
of order prime to the characteristic, G0 /G1 is a cyclic group by (3.112). If
char F = 0, no subgroup of the additive group FP  is finite, so Gi = Gi+1 for
 3.8 Galois Extensions II 135

all i ≥ 1 by (3.113), and since Gi = {id} for sufficiently large i, (d) follows.
If char F = p > 0, each additive subgroup of FP  is elementary abelian of
exponent p, hence (3.113) implies the remaining assertions of (e) and (f).
In order to prove (3.112) and (3.113), we choose a P  -prime element t and
set for σ ∈ G0 ,
σ(t)
χ(σ) := + P  ∈ (FP  )× .
t
The definition of χ(σ) does not depend on the specific choice of t: Let t∗ = u·t
be another prime element for P  (i.e., u ∈ F  and vP  (u) = 0). Then

σ(t∗ ) σ(t)
+ P = + P , (3.114)
t∗ t
since
σ(t∗ ) σ(t) σ(t) · σ(u) σ(t) σ(t) −1
∗
− = − = · u · (σ(u) − u) ∈ P  .
t t t·u t t
(Observe that σ(u) − u ∈ P  , as σ ∈ G0 .) For σ, τ ∈ G0 we have

(στ )(t) σ(τ (t)) τ (t)

χ(στ ) = + P = · + P  = χ(σ) · χ(τ ) .
t τ (t) t

Therefore χ is a homomorphism (here we have used the fact that τ (t) is a

prime element at P  , and the definition of χ(σ) is independent of the prime
element by (3.114)). An element σ ∈ G0 is in the kernel of χ if and only if
(σ(t)/t) − 1 ∈ P  ; i.e., vP  (σ(t) − t) ≥ 2. Thus Ker(χ) = G1 by (c).
It remains to prove (3.113). Let i ≥ 1 and σ ∈ Gi , then σ(t) = t + ti+1 · uσ
for some uσ ∈ OP  . We define ψi : Gi → FP  by

ψi (σ) := uσ + P  .

(Actually, this definition depends on the choice of t.) For τ ∈ Gi we have

τ (t) = t + ti+1 · uτ , hence

(στ )(t) = σ(t + ti+1 uτ ) = σ(t) + σ(t)i+1 · σ(uτ )

= t + ti+1 · uσ + (t + ti+1 · uσ )i+1 · (uτ + tx)
(with some x ∈ OP  ; observe that σ(uτ ) − uτ ∈ P  since σ ∈ Gi )
= t + ti+1 · uσ + ti+1 (1 + ti uσ )i+1 · (uτ + tx)
= t + ti+1 (uσ + uτ + ty) with y ∈ OP  .

Therefore uστ = uσ + uτ + ty, which implies that

ψi (στ ) = ψi (σ) + ψi (τ ) .

The kernel of ψi is obviously Gi+1 ; this finishes the proof of (3.113).

136 3 Extensions of Algebraic Function Fields

As a consequence of Proposition 3.8.5 we have the following supplement

to Theorem 3.8.3.

Corollary 3.8.6. Suppose that F  /F is a Galois extension of algebraic func-

tion fields of characteristic p > 0. Let P ∈ IPF and let P  ∈ IPF  be an
extension of P in F  , and consider the fixed field V1 (P  |P ) of the first ramifi-
cation group G1 (P  |P ). Denote by PM the restriction of P  to an intermediate
field M , F ⊆ M ⊆ F  . Then the following hold:
(a) M ⊆ V1 (P  |P ) ⇐⇒ e(PM |P ) is prime to p.
(b) M ⊇ V1 (P  |P ) ⇐⇒ PM is totally ramified in F  /M , and e(P  |PM ) is a
power of p.

We omit the proof of this corollary; it is similar to the proof of Theorem

3.8.3.
There is a close relation between the different exponent d(P  |P ) and the
ramification groups Gi (P  |P ).

Theorem 3.8.7 (Hilbert’s Different Formula). Consider a Galois exten-

sion F  /F of algebraic function fields, a place P ∈ IPF and a place P  ∈ IPF 
lying over P . Then the different exponent d(P  |P ) is
∞

d(P  |P ) = (ord Gi (P  |P ) − 1) .
i=0

(Note that this is a finite sum, since Gi (P  |P ) = {id} for large i.)

Proof. First we assume that P  |P is totally ramified; i.e., G := Gal(F  /F ) =

G0 (P  |P ). Set ei := ord Gi (P  |P ) (for i = 0, 1, . . .) and e := e0 = [F  : F ].
Choose a P  -prime element t; then {1, t, . . . , te−1 } is an integral basis for F  /F
at P by Proposition 3.5.12, and d(P  |P ) = vP  (ϕ (t)) where ϕ(T ) ∈ F [T ] is
the minimal polynomial of t over F . Since F  /F is Galois,

ϕ(T ) = (T − σ(t)) .
σ∈G

Consequently 
ϕ (t) = ± (σ(t) − t) .
σ=id

We obtain
 ∞
 
d(P  |P ) = vP  (σt − t) = vP  (σt − t)
σ=id i=0 σ∈Gi \Gi+1
∞
 ∞

= (ei − ei+1 )(i + 1) = (ei − 1) . (3.115)
i=0 i=0
 3.9 Ramification and Splitting in the Compositum of Function Fields 137

So our theorem is proved in the totally ramified case. Now we consider the
general case. Let T denote the inertia field of P  |P and PT := P  ∩ T . Then
PT |P is unramified and P  |PT is totally ramified. The ramification groups
Gi (P  |P ) are the same as Gi (P  |PT ), for i = 0, 1, . . . , and the different expo-
nent d(P  |P ) is

d(P  |P ) = e(P  |PT ) · d(PT |P ) + d(P  |PT ) = d(P  |PT ) , (3.116)

by Corollary 3.4.12(b) (note that d(PT |P ) = 0 as PT |P is unramified). Now

the theorem follows from (3.115) and (3.116).

3.9 Ramification and Splitting in the Compositum

of Function Fields
Quite often one considers the following situation: F1 /F and F2 /F are finite
extensions of a function field F , and one knows the ramification (resp. split-
ting) behavior of a place P ∈ IPF in both extensions F1 /F and F2 /F . What
can then be said about ramification (resp. splitting) of P in the compositum
F1 F2 /F ? In this section we address this problem.
First we deal with ramification. The main result here is:
Theorem 3.9.1 (Abhyankar’s Lemma). Let F  /F be a finite separable
extension of function fields. Suppose that F  = F1 F2 is the compositum of two
intermediate fields F ⊆ F1 , F2 ⊆ F  . Let P  ∈ IPF  be an extension of P ∈ IPF ,
and set Pi := P  ∩ Fi for i = 1, 2. Assume that at least one of the extensions
P1 |P or P2 |P is tame. Then

e(P  |P ) = lcm{e(P1 |P ), e(P2 |P )} .

For the proof of this theorem we need the following lemma.

Lemma 3.9.2. Let G be a finite group and U ⊆ G be a normal subgroup such

that ord U = pn (with either p = 1 or else, p a prime number) and G/U is
cyclic of order relatively prime to p. Suppose that H1 is a subgroup of G with
pn | ord H1 . Then for every subgroup H2 ⊆ G we have

ord (H1 ∩ H2 ) = gcd(ord H1 , ord H2 ) .

Proof of the Lemma. Clearly the order of H1 ∩ H2 divides the orders of H1

and of H2 , thus

ord (H1 ∩ H2 ) | gcd(ord H1 , ord H2 ) .

We set ord H1 = a1 pn and ord H2 = a2 pm with (a1 , p) = (a2 , p) = 1, and

d := gcd(a1 , a2 ). Then gcd(ord H1 , ord H2 ) = pm d. It is sufficient to prove the
following claims:
138 3 Extensions of Algebraic Function Fields

H1 ∩ H2 contains a subgroup of order pm , and (3.117)

H1 ∩ H2 contains an element whose order is a multiple of d . (3.118)
Let V ⊆ H2 be a p-Sylow subgroup of H2 (i.e., ord V = pm ). Since U is
by assumption a normal subgroup of G, it is the only p-Sylow group of G,
therefore V ⊆ U ⊆ H1 . This proves (3.117).
Now we consider the canonical homomorphism π : G → G/U . The groups
π(Hi ) ⊆ G/U are of order ai (i = 1, 2), and π(H1 ) ∩ π(H2 ) is a cyclic group
of order d = gcd(a1 , a2 ) (here we use the fact that G/U is cyclic). Choose
elements g1 ∈ H1 and g2 ∈ H2 such that π(g1 ) = π(g2 ) is a generator of
π(H1 ) ∩ π(H2 ). Then g1−1 g2 =: u ∈ U ⊆ H1 , so g2 = g1 u ∈ H1 ∩ H2 , and the
order of g2 is a multiple of d.

Proof of Theorem 3.9.1. Choose a Galois extension F ∗ /F with F  ⊆ F ∗ and

an extension P ∗ ∈ IPF ∗ of P  in F ∗ . Then we have the following situation:

F∗ P∗

F  = F1 F 2 P
@ @
@ @
@ @
F1 F2 P1 P2
@ @
@ @
@ @
F P

Fig. 3.4.

Consider the group G := GT (P ∗ |P ) and its subgroups Hi := GT (P ∗ |Pi )

for i = 1, 2. Let p = char F (in the case of characteristic 0 we set p = 1). At
least one of the extensions Pi |P is tame, say gcd(e(P1 |P ), p) = 1. The groups
G, H1 and H2 satisfy the hypotheses of Lemma 3.9.2, thus
ord (H1 ∩ H2 ) = gcd(ord H1 , ord H2 ) .
The condition F  = F1 F2 shows that Gal(F ∗ /F  ) = Gal(F ∗ /F1 )∩Gal(F ∗ /F2 )
and GT (P ∗ |P  ) = GT (P ∗ |P1 ) ∩ GT (P ∗ |P2 ) = H1 ∩ H2 . We obtain
e(P ∗ |P  ) = ord GT (P ∗ |P  ) = ord (H1 ∩ H2 )
= gcd(ord H1 , ord H2 ) = gcd(e(P ∗ |P1 ), e(P ∗ |P2 ))
= gcd(e(P ∗ |P  ) · e(P  |P1 ), e(P ∗ |P  ) · e(P  |P2 ))
= e(P ∗ |P  ) · gcd(e(P  |P1 ), e(P  |P2 )) .
 3.9 Ramification and Splitting in the Compositum of Function Fields 139

Therefore
gcd(e(P  |P1 ), e(P  |P2 )) = 1 . (3.119)
On the other hand we have
e(P  |P ) = e(P  |P1 ) · e(P1 |P ) = e(P  |P2 ) · e(P2 |P ) . (3.120)
Equations (3.119) and (3.120) imply that
e(P  |P ) = lcm (e(P1 |P ), e(P2 |P )) .
(This is a simple fact from elementary number theory: if ax = by with non
zero integers a, b, x, y and gcd(x, y) = 1, then the least common multiple of a
and b is lcm(a, b) = ax = by.)

Recall that a place P ∈ IPF is said to be unramified in a finite extension

E/F if e(Q|P ) = 1 for all places Q ∈ IPE with Q|P . We have as an immediate
consequence of Abhyankar’s Lemma:
Corollary 3.9.3. Let F  /F be a finite separable extension of function fields
and let P be a place of F .
(a) Suppose that F  = F1 F2 is the compositum of two intermediate fields
F ⊆ F1 , F2 ⊆ F  . If P is unramified in F1 /F and in F2 /F , then P is unram-
ified in F  /F .
(b) Assume now that F0 is an intermediate field F ⊆ F0 ⊆ F  such that F  /F
is the Galois closure of F0 /F . If P is unramified in F0 /F then P is unramified
in F  /F .
Proof. (a) This is just a special case of Theorem 3.9.1.
(b) The Galois closure F  of F0 /F is the compositum of the fields σ(F0 ),
where σ runs through all embeddings σ : F0 → F̄ over F (where F̄ ⊇ F is the
algebraic closure of F ). As P is unramified in F0 /F , it is also unramified in
σ(F0 )/F . Assertion (b) follows now from (a).

In the case of wild ramification, Abhyankar’s Lemma does not hold in

general. With regard to an application in Chapter 7 (see Proposition 7.4.13)
we discuss now the simplest case of this situation: we consider a function field
F/K of characteristic p > 0 and two distinct Galois extensions F1 /F and
F2 /F of degree [F1 : F ] = [F2 : F ] = p. Let F  = F1 F2 be the compositum of
F1 and F2 ; then F  /F is Galois of degree [F  : F ] = p2 , and also the extensions
F  /F1 and F  /F2 are Galois of degree p. Let P be a place of F and P  a place
of F  lying above P , and denote by Pi := P  ∩ Fi the restrictions of P  to Fi ,
for i = 1, 2. If P1 |P or P2 |P is unramified then Abhyankar’s Lemma describes
how P  |P1 and P  |P2 are ramified. So we assume now that both places P1 |P
and P2 |P are ramified. We then have e(P1 |P ) = e(P2 |P ) = p. By Hilbert’s
Different Formula the different exponents d(Pi |P ) satisfy d(Pi |P ) = ri (p − 1)
with ri ≥ 2 for i = 1, 2. In the following proposition we consider the special
case r1 = r2 = 2.
140 3 Extensions of Algebraic Function Fields

Proposition 3.9.4. In the situation as above we assume that d(P1 |P ) =

d(P2 |P ) = 2(p − 1). Then one of the following assertions holds:
(1) e(P  |P1 ) = e(P  |P2 ) = 1, or
(2) e(P  |P1 ) = e(P  |P2 ) = p and d(P  |P1 ) = d(P  |P2 ) = 2(p − 1).

F P
@ @
@ @
@ @
F1 F2 P1 P2
@ @
@ e=p @ e=p
@ d = 2(p − 1)@ d = 2(p − 1)
F P

Fig. 3.5.

Proof. It suffices to consider the case e(P  |P1 ) = e(P  |P2 ) = p. Denote by
Gi := Gi (P  |P ) ⊆ Gal(F  /F ) the i-th ramification group of P  |P , then

Gal(F  /F ) = G0 ⊇ G1 ⊇ . . . ⊇ Gs−1  Gs = {id} (3.121)

for some integer s ≥ 2. For an intermediate field F ⊆ H ⊆ F  we set Q :=

P  ∩ H. If U := Gal(F  /H) is the subgroup of Gal(F  /F ) corresponding to H,
the i-th ramification group of P  |Q is (by Definition 3.8.4) Gi (P  |Q) = U ∩Gi .
By Hilbert’s Different Formula the different exponents of P  |P and P  |Q are
given by


s−1
d(P  |P ) = (ord(Gi ) − 1) , (3.122)
i=0

s−1
d(P  |Q) = (ord(U ∩ Gi ) − 1) . (3.123)
i=0

We distinguish two cases.

Case 1. ord(Gs−1 ) = p2 . From Equation (3.122) we obtain d(P  |P ) =
s(p − 1). We choose H := F1 , then (3.123) gives d(P  |P1 ) = s(p − 1). Since
2

d(P  |P ) = e(P  |P1 ) · d(P1 |P ) + d(P  |P1 ) by transitivity of different exponents

(Corollary 3.4.12), we conclude that s(p2 − 1) = p · d(P1 |P ) + s(p − 1), hence
d(P1 |P ) = s(p − 1). As d(P1 |P ) = 2(p − 1) by assumption, we then have s = 2
and d(P  |P1 ) = s(p − 1) = 2(p − 1).
 3.9 Ramification and Splitting in the Compositum of Function Fields 141

Case 2. ord(Gs−1 ) = p. At least one of the fields F1 , F2 is not the fixed field
of the group Gs−1 , so we can assume w.l.o.g. that U := Gal(F  /F1 ) = Gs−1 .
We obtain then from (3.123) that d(P  |P1 ) < s(p − 1), hence

d(P  |P ) = e(P  |P1 ) · d(P1 |P ) + d(P  |P1 )

< p · 2(p − 1) + s(p − 1) = (2p + s)(p − 1) . (3.124)

On the other hand, observing that ord(G0 ) = ord(G1 ) = p2 by Proposition

3.8.5 we get from (3.121) and (3.122)

d(P  |P ) ≥ 2(p2 − 1) + (s − 2)(p − 1) = (2p + s)(p − 1) .

This inequality contradicts (3.124) and therefore Case 2 cannot occur.

Now we proceed to splitting places. We recall that a place P ∈ IPF splits

completely in an extension field E/F of degree [E : F ] = n < ∞, if P has n
distinct extensions Q1 , . . . , Qn in E. By the Fundamental Equality 3.1.11 this
is equivalent to the condition that e(Q|P ) = f (Q|P ) = 1 for all Q ∈ IPE lying
above P .

Lemma 3.9.5. Let F0 /F be a finite separable extension of function fields and

let F  ⊇ F0 be the Galois closure of F0 /F . Assume that a place P ∈ IPF is
completely splitting in F0 /F . Then P splits completely in F  /F .

Proof. Let P  be a place of F  lying above P and consider the decomposition

field Z := Z(P  |P ) ⊆ F  (see Definition 3.8.1). Set P0 := P  ∩ F0 . Since P
splits completely in F0 /F we have e(P0 |P ) = f (P0 |P ) = 1, and it follows
from Theorem 3.8.3(a) that F0 ⊆ Z. For every embedding σ : F0 → F  over
F , the place P splits completely in σ(F0 )/F , hence also the field σ(F0 ) is
contained in Z. The Galois closure F  of F/F0 is the compositum of all these
fields σ(F0 ), so we have that F  ⊆ Z and hence F  = Z; we conclude from
Theorem 3.8.2 that e(P  |P ) = f (P  |P ) = 1.

Proposition 3.9.6. Let F  /F be a finite separable extension of function fields

and let F1 , F2 be intermediate fields of F  /F such that F  = F1 F2 is their
compositum.
(a) Suppose that P is a place of F which splits completely in the extension
F1 /F . Then every place Q of F2 lying above P splits completely in the exten-
sion F  /F2 .
(b) If P ∈ IPF splits completely in F1 /F and in F2 /F , then P splits completely
in F  /F .

Proof. (a) Let E/F be the Galois closure of F1 /F ; then P splits completely in
the extension E/F by Lemma 3.9.5. We consider the compositum E  := EF2 .
By Galois theory we know that the extension E  /F2 is Galois, and the Galois
142 3 Extensions of Algebraic Function Fields

group Gal(E  /F2 ) is isomorphic to a subgroup of Gal(E/F ) under the map

σ → σ|E (the restriction of σ to E).
Suppose that there is a place Q of F2 above P which does not split com-
pletely in F  /F2 , so Q does not split completely in E  /F2 . Choose a place Q
of E  lying above Q and set P  := Q ∩ E. The situation is shown in Figure
3.6.

E  = EF2 Q
@ @
@ @
@ @
F  = F1 F 2 E P
@
@
@
F2 F1 Q
@ @
@ @
@ @
F P

Fig. 3.6.

Since e(Q |Q) · f (Q |Q) > 1, there is an automorphism σ ∈ Gal(E  /F2 ) with
σ(Q ) = Q and σ = id, by Theorem 3.8.2(a). Then the restriction σ  := σ|E ∈
Gal(E/F ) is not the identity on E and σ  (P  ) = P  , hence the decomposition
group GZ (P  |P ) ⊆ Gal(E/F ) is non-trivial. It follows that e(P  |P )·f (P  |P ) =
ord GZ (P  |P ) > 1, a contradiction to the fact that P splits completely in E/F .
(b) is an immediate consequence of (a).

Corollary 3.9.7. Let F/K be a function field whose full constant field is K.
(a) Suppose that F  = F1 F2 is the compositum of two finite separable exten-
sions F1 /F and F2 /F . Assume that there exists a place P ∈ IPF of degree
one which splits completely in F1 /F and in F2 /F . Then P splits completely
in F  /F , and K is the full constant field of F  .
(b) Suppose that F0 /F is a finite separable extension and P ∈ IPF is a place
of degree one which splits completely in F0 /F . Let F̃ /F be the Galois closure
of F0 /F . Then P splits completely in F̃ /F and K is the full constant field of
F̃ .

Proof. (a) We only have to show that K is the full constant field of F  = F1 F2 ;
the remaining assertions follow immediately from Proposition 3.9.6. We choose
 3.10 Inseparable Extensions 143

a place P  of F  lying above P , then f (P  |P ) = 1 and therefore the residue

class field FP  of P  is equal to the residue class field FP = K of P . Since the
full constant field K  of F  satisfies K ⊆ K  ⊆ FP  , we conclude that K  = K.
(b) is obvious.

3.10 Inseparable Extensions

Every algebraic extension F  /F of algebraic function fields can be split into

a separable step Fs /F and a purely inseparable step F  /Fs , see Appendix A.
Thus far we have mostly studied separable extensions. In the present section,
purely inseparable extensions will be investigated. Throughout this section,
K is a perfect field of characteristic p > 0, and F/K is a function field with
constant field K.

Lemma 3.10.1. Suppose F  /F is a purely inseparable field extension of de-

gree p. Then K is the constant field of F  as well. Every place P ∈ IPF has
only one extension P  ∈ IPF  , namely

P  = {z ∈ F  | z p ∈ P } .

The corresponding valuation ring is

OP  = {z ∈ F  | z p ∈ OP } .

We have e(P  |P ) = p and f (P  |P ) = 1.

Proof. Let a ∈ F  be algebraic over K. Since F  /F is purely inseparable of

degree p, we have ap ∈ F and ap is algebraic over K. As K is the constant
field of F this shows that ap ∈ K. But K is perfect, so ap ∈ K implies a ∈ K.
Hence K is the constant field of F  .
Next we consider a place P ∈ IPF . Define

R := {z ∈ F  | z p ∈ OP } and M := {z ∈ F  | z p ∈ P } .

Obviously R is a subring of F  /K with OP ⊆ R, and M is a proper ideal of R

containing P . Let P  ∈ IPF  be an extension of P . For z ∈ OP  (resp. z ∈ P  )
we have z p ∈ OP  ∩F = OP (resp. z p ∈ P  ∩F = P ), hence OP  ⊆ R and P  ⊆
M . Since OP  is a maximal proper subring of F  (see Theorem 1.1.12(c)) and
P  is a maximal ideal of OP  , this implies that OP  = R, P  = M , and P  is the
only place of F  lying over P . The residue class field FP  = OP  /P  is clearly
purely inseparable over FP = OP /P , consequently FP  = FP (observe that
FP is a finite extension of the perfect field K, thus each algebraic extension
 
 This proves f (P |P ) = 1, and e(P |P ) = p follows now
of FP is separable).
from the formula Pi |P e(Pi |P ) · f (Pi |P ) = [F  : F ] = p.
144 3 Extensions of Algebraic Function Fields

An element x ∈ F is called a separating element for F/K if F/K(x) is a

finite separable extension. F/K is said to be separably generated if there exists
a separating element for F/K. Next we show, among other things, that every
function field F/K is separably generated (this is not true in general if K is
not assumed to be perfect).
Proposition 3.10.2. (a) Assume z ∈ F satisfies vP (z) ≡ 0 mod p for some
P ∈ IPF . Then z is a separating element for F/K. In particular F/K is
separably generated.
(b) There exist x, y ∈ F such that F = K(x, y).
n n
(c) For each n ≥ 1 the set F p := {z p | z ∈ F } is a subfield of F . It has the
following properties:
n n
(1) K ⊆ F p ⊆ F , and F/F p is purely inseparable of degree pn .
n
(2) The Frobenius map ϕn : F → F , defined by ϕn (z) := z p , is an
n n
isomorphism of F onto F p . Therefore the function field F p /K has
the same genus as F/K.
(3) Suppose that K ⊆ F0 ⊆ F and F/F0 is purely inseparable of degree
n
[F : F0 ] = pn . Then F0 = F p .
(d) An element z ∈ F is a separating element for F/K if and only if z ∈ F p .

Proof. (a) Suppose that z is not separating. The extension F/K(z) is of finite
degree since z ∈ K, hence there is an intermediate field K(z) ⊆ Fs ⊆ F
such that F/Fs is purely inseparable of degree p. Let Ps := P ∩ Fs . By the
preceding lemma we have e(P |Ps ) = p, so vP (z) = p · vPs (z) ≡ 0 mod p.
(b) Choose a separating element x ∈ F \K. Since F/K(x) is a finite separable
field extension, there is some y ∈ F satisfying F = K(x, y) (see Appendix A).
n n n
(c) It is easily verified that F p is a field, and K = K p ⊆ F p because K is
pn pn n
perfect. The extension F/F is purely inseparable since z ∈ F p for each
z ∈ F . We choose x, y ∈ F such that x is separating and F = K(x, y), and
claim that n
F = K(x, y p ) (3.125)
n n
holds. In fact, F = K(x, y p )(y) is purely inseparable over K(x, y p ) since
n n n
y satisfies the equation T p − y p = 0 over K(x, y p ). On the other hand,
n n
K(x) ⊆ K(x, y p ) ⊆ F , and therefore the extension F/K(x, y p ) is separable.
This proves (3.125).
n n n n n n
Now F p = K p (xp , y p ) = K(xp , y p ), and (3.125) implies that F =
pn n n n
F (x). Because x is a zero of the polynomial T p −xp over F p , we conclude
that n
[F : F p ] ≤ pn . (3.126)
n
In order to prove the reverse inequality, choose a place P0 of F p /K and an
n
element u ∈ F p with vP0 (u) = 1. Let P ∈ IPF be an extension of P0 in F ;
n n
then [F : F p ] ≥ e(P |P0 ). Writing u = z p for some z ∈ F we obtain
 3.11 Estimates for the Genus of a Function Field 145
n
pn · vP (z) = vP (z p ) = vP (u) = e(P |P0 ) · vP0 (u) = e(P |P0 ) .

So n
pn ≤ e(P |P0 ) ≤ [F : F p ] . (3.127)
This finishes the proof of (1).
Assertion (2) is trivial, and it remains to prove (3). By assumption, the
n
extension F/F0 is purely inseparable of degree pn . Then z p ∈ F0 for each
n n
z ∈ F , so F p ⊆ F0 ⊆ F . The degree [F : F p ] is pn by (1), consequently we
n
have F p = F0 .
(d) If z is a separating element, K(z) ⊆ F p since F/F p is purely inseparable of
degree > 1. Conversely, if z ∈ F \K is not separating, there is an intermediate
field K(z) ⊆ F0 ⊆ F such that F/F0 is purely inseparable of degree p. By (c),
F0 = F p , hence z ∈ F p .

In characteristic 0 the situation is of course much simpler: each x ∈ F \K is

a separating element, and therefore we have F = K(x, y) with an appropriate
element y.

3.11 Estimates for the Genus of a Function Field

It is often difficult to determine the genus of a function field precisely. There-
fore we would like to derive some bounds for the genus in specific cases. As
always, F/K is an algebraic function field over the perfect constant field K.

Proposition 3.11.1. Let F1 /K be a subfield of F/K and [F : F1 ] = n. As-

sume that {z1 , . . . , zn } is a basis of F/F1 such that all zi ∈ L (C) for some
divisor C ∈ Div(F ). Then

g ≤ 1 + n(g1 − 1) + deg C ,

where g (resp. g1 ) denotes the genus of F/K (resp. of F1 /K).

Proof. Let A1 be a divisor of F1 /K of sufficiently large degree such that

(A1 ) =: t = deg A1 + 1 − g1 .

Choose a basis {x1 , . . . , xt } ⊆ F1 of L (A1 ). Set A := ConF/F1 (A1 ) ∈ Div(F ).

The elements
xi zj (1 ≤ i ≤ t, 1 ≤ j ≤ n)
are in L (A + C) and they are linearly independent over K. Hence

(A + C) ≥ n · (deg A1 + 1 − g1 ) . (3.128)

We may assume that deg (A+C) is sufficiently large so that the Riemann-Roch
Theorem yields
146 3 Extensions of Algebraic Function Fields

 (A + C) = deg (A + C) + 1 − g
= n · deg A1 + deg C + 1 − g . (3.129)

Substituting (3.129) into (3.128) we obtain g ≤ 1 + n(g1 − 1) + deg C.

In the course of the proof of Theorem 3.11.3 we shall need the following
lemma:
Lemma 3.11.2. Assume that K is algebraically closed, and consider a sub-
field F1 /K of F/K such that F/F1 is separable of degree [F : F1 ] = n > 1.
Let y ∈ F be an element with F = F1 (y). Then almost all P ∈ IPF1 have the
following properties:
(a) P splits completely in F/F1 ; i.e., it has n distinct extensions P1 , . . . , Pn
in F/F1 .
(b) The restrictions P1 ∩ K(y), . . . , Pn ∩ K(y) are pairwise distinct places of
K(y).

Proof. Let ϕ(T ) = T n +zn−1 T n−1 +· · ·+z0 ∈ F1 [T ] be the minimal polynomial

of y over F1 . For almost all P ∈ IPF1 the following hold:

{1, y, . . . , y n−1 } is an integral basis of F/F1 for P, and

P is unramified in F/F1 . (3.130)

From now on we assume that P satisfies (3.130). Since K is algebraically

closed, P splits completely in F . For z ∈ OP let z̄ ∈ OP /P = K denote its
residue class modulo P ; then the decomposition of the polynomial ϕ̄(T ) =
T n + z̄n−1 T n−1 + · · · + z̄0 ∈ K[T ] corresponds to the splitting of P in F (by
Kummer’s Theorem). Hence we have by (3.130) a decomposition

n
ϕ̄(T ) = (T − bi )
i=1

with pairwise distinct elements bi ∈ K. For i = 1, . . . , n there exists (by

Kummer’s Theorem) a unique place Pi ∈ IPF such that Pi |P and vPi (y −
bi ) > 0. Since the elements bi are pairwise distinct, the restrictions Pi ∩K(y) ∈
IPK(y) are distinct for i = 1, . . . , n.
Theorem 3.11.3 (Castelnuovo’s Inequality). Let F/K be a function field
with constant field K. Suppose there are given two subfields F1 /K and F2 /K
of F/K satisfying
(1) F = F1 F2 is the compositum of F1 and F2 , and
(2) [F : Fi ] = ni and Fi /K has genus gi (i = 1, 2).
Then the genus g of F/K is bounded by

g ≤ n1 g1 + n2 g2 + (n1 − 1)(n2 − 1) .
 3.11 Estimates for the Genus of a Function Field 147

Proof. We may assume that K is an algebraically closed field. (Otherwise we

replace F/K by the constant field extension F K̄/K̄ with the algebraic closure
K̄ ⊆ Φ of K, and Fi /K by Fi K̄/K̄. In a constant field extension, the genera
remain unchanged by Theorem 3.6.3, and we also have, by Proposition 3.6.6,
that [F K̄ : Fi K̄] = [F : Fi ].) Moreover we can assume that F/F1 is a separable
extension (if both F/F1 and F/F2 were inseparable, then F1 F2 ⊆ F p  F by
Proposition 3.10.2).
The idea behind the proof of Castelnuovo’s Inequality is to find an appro-
priate divisor C ∈ Div(F ) of small degree and a basis {u1 , . . . , un } ⊆ L (C)
of F/F1 such that Proposition 3.11.1 will yield the desired inequality.
Since F = F1 F2 , there are y1 , . . . , ys ∈ F2 with F = F1 (y1 , . . . , ys ). The
extension F/F1 is separable, hence we can find a1 , . . . , as ∈ K such that the
element
 s
y := aj yj ∈ F2
j=1

is a primitive element of F/F1 ; i.e., F = F1 (y) (see Appendix A). By

Proposition 1.6.12 there is a divisor A0 ∈ Div(F2 /K) with A0 ≥ 0, deg A0 =
g2 and (A0 ) = 1. Let P0 ∈ IPF2 be a place not in the support of A0 , and set
B0 := A0 − P0 . Since L (A0 ) = K, it follows that

deg B0 = g2 − 1 and (B0 ) = 0 . (3.131)

Now we choose a place P ∈ IPF1 that has n1 distinct extensions P1 , . . . , Pn1

in F/F1 such that the restrictions

Qi := Pi ∩ F2 ∈ IPF2

are pairwise distinct and Qi ∈ supp B0 , for i = 1, . . . , n1 . This is possible by

Lemma 3.11.2. The Riemann-Roch Theorem yields

(B0 + Qi ) ≥ deg (B0 + Qi ) + 1 − g2 = 1 . (3.132)

By (3.131) and (3.132) there is an element ui ∈ F2 satisfying

(ui ) ≥ −(B0 + Qi ) and vQi (ui ) = −1 . (3.133)

We claim that {u1 , . . . , un1 } is a basis of F/F1 ; so we must show that these
elements are linearly independent over F1 . Suppose that


n1
xi ui = 0 with xi ∈ F1
i=1

is a non-trivial linear combination. Choose j ∈ {1, . . . , n1 } such that

vP (xj ) ≤ vP (xi ) for i = 1, . . . , n1 . (3.134)

148 3 Extensions of Algebraic Function Fields

Then
vPj (xj uj ) = vPj (xj ) + vPj (uj ) ≤ vP (xj ) − 1 .
(Observe that vPj (xj ) = vP (xj ) since Pj |P is unramified, and vPj (uj ) ≤ −1
by (3.133).) For i = j we have

vPj (xi ui ) = vP (xi ) + vPj (ui ) ≥ vP (xi ) ≥ vP (xj ) ,

by (3.133) and (3.134). Therefore the Strict Triangle Inequality implies


n1 
vPj xi ui = vPj (xj uj ) < ∞ .
i=1

This contradiction proves that {u1 , . . . , un1 } is a basis of F/F1 .

Now we consider the divisor
 
n1 
C := ConF/F2 B0 + Qi ∈ Div(F ) .
i=1

Its degree is
 
n1 
deg C = n2 · deg B0 + Qi = n2 (g2 − 1 + n1 ) .
i=1

By (3.133) the elements u1 , . . . , un1 are in L (C). Therefore we can apply

Proposition 3.11.1 to obtain

g ≤ 1 + n1 (g1 − 1) + n2 (g2 − 1 + n1 )
= n1 g1 + n2 g2 + (n1 − 1)(n2 − 1) .

In the special case F1 = K(x) and F2 = K(y), Castelnuovo’s Inequality

yields:

Corollary 3.11.4 (Riemann’s Inequality). Suppose that F = K(x, y).

Then we have the following estimate for the genus g of F/K:

g ≤ ([F : K(x)] − 1) · ([F : K(y)] − 1) .

Riemann’s Inequality (and therefore also Castelnuovo’s Inequality) is often

sharp, and it cannot be improved in general. In some situations however,
another bound for the genus of K(x, y) is even better.
 3.11 Estimates for the Genus of a Function Field 149

Proposition 3.11.5. Consider an algebraic function field F = K(x, y) over

K, where the irreducible equation of y over K(x) has the form

y n + f1 (x)y n−1 + . . . + fn−1 (x)y + fn (x) = 0 (3.135)

with fj (x) ∈ K[x] and deg fj (x) ≤ j for j = 1, . . . , n. Then the genus g of
F/K satisfies the inequality
1
g≤ (n − 1)(n − 2) . (3.136)
2

Proof. The proof is similar to that of Proposition 3.11.1. Let A := (x)∞ denote
the pole divisor of x in F . It is a positive divisor of degree n. We claim that

vP (y) ≥ −vP (A) for all P ∈ IPF . (3.137)

If P is a place with vP (x) ≥ 0 then vP (y) ≥ 0 by (3.135), and therefore (3.137)

holds for P . Now consider the case vP (x) < 0. Then vP (x) = −vP (A), and
the hypothesis deg fj (x) ≤ j implies that vP (fj (x)) ≥ j · vP (x). Suppose that
vP (y) < −vP (A). For j = 1, . . . , n we obtain

vP (fj (x)y n−j ) ≥ j · vP (x) + (n − j) · vP (y)

> j · vP (y) + (n − j) · vP (y) = vP (y n ) .

So Equation (3.135) contradicts the Strict Triangle Inequality, and (3.137) is

proved. We conclude that

(x) ≥ −A and (y) ≥ −A .

It follows that for all l ≥ n the elements

xi y j with 0 ≤ j ≤ n − 1 and 0 ≤ i ≤ l − j

are in L (lA). They are linearly independent over K as 1, y, . . . , y n−1 are

linearly independent over K(x). Therefore


n−1 
n−1
(lA) ≥ (l − j + 1) = n(l + 1) − j
j=0 j=0
1
= n(l + 1) − n(n − 1) . (3.138)
2
For l sufficiently large, the Riemann-Roch Theorem yields

(lA) = l · deg A + 1 − g = ln + 1 − g .

We substitute this into (3.138) and obtain g ≤ (n − 1)(n − 2)/2.

150 3 Extensions of Algebraic Function Fields

3.12 Exercises
In all exercises below we assume that K is a perfect field and F/K is a function
field with full constant field K.
3.1. Let F  /F be an algebraic extension of F and A ∈ Div(F ). Show that
L (ConF  /F (A)) ∩ F = L (A).
3.2. Let P1 , . . . , Pr be places of F/K (r ≥ 1). Show that there is an element
x ∈ F with the following properties:
(a) P1 , . . . , Pr are poles of x, and there are no other poles of x.
(b) The extension F/K(x) is separable.
3.3. Let R = OS be a holomorphy ring of F/K, and let R1 be a subring of F
with R ⊆ R1  F . Show:
(i) For each x ∈ R1 , the ring R[x] is a holomorphy ring.
Hint. Consider the set T := {P ∈ S | vP (x) ≥ 0} and show that R[x] = OT .
(ii) The ring R1 is a holomorphy ring of F/K.
3.4. Consider an extension field F  = F (y) of degree [F  : F ] = n. Let OS ⊆ F
be a holomorphy ring of F and assume that its integral closure in F  is

n−1
icF  (OS ) = OS · y i .
i=0

Show that {1, y, . . . , y n−1 } is an integral basis of F  /F at all places P ∈ S.

3.5. Let S  IPF such that IPF \ S is finite. Show that there are elements
x1 , . . . , xr ∈ F with OS = K[x1 , . . . , xr ].
3.6. We define the ramification locus of a finite separable extension F  /F as
Ram(F  /F ) := {P ∈ IPF | there is some place P  ∈ IPF  with e(P  |P ) > 1},
and its degree as

deg Ram(F  /F ) := deg P .
P ∈Ram(F  /F )

Now let F/K(x) be a finite separable extension of the rational function field,
having K as its full constant field, and [F : K(x)] = n > 1. Show:
(i) Ram(F/K(x)) = ∅.
(ii) If char K = 0 or char K > n, then deg Ram(F/K(x)) ≥ 2. If moreover
the genus of F is > 0, then deg Ram(F/K(x)) ≥ 3.
3.7. Assume that K is algebraically closed, and char K = 0 or char K > n. Let
F/K(x) be a separable extension of degree n such that deg Ram(F/K(x)) = 2.
Show that there is an element y ∈ F such that F = K(y) and y satisfies the
equation y n = (ax + b)/(cx + d), with a, b, c, d ∈ K and ad = bc.
 3.12 Exercises 151

3.8. Consider the rational function field F = K(x) and a polynomial f (x) ∈
K[x] of degree deg f (x) = n ≥ 2. In case of char K = p > 0 we assume that
f (x) ∈
/ K[xp ]. We set z = f (x) and consider the field extension K(x)/K(z),
which is separable of degree n (cf. Exercise 1.1).
(i) Show that exactly the following places of K(x) are ramified in K(x)/K(z):
the pole P∞ of x and the places which are zeros of the derivative f  (x).
(ii) In the special case z = f (x) = xn where n is not divisible by the charac-
teristic of K, show that the zero and the pole of z are the only ramified places
of K(z) in K(x)/K(z). Calculate the different exponents of the places above
them, without using results of Section 3.7.
s
(iii) Now consider the case z = xp − x, where p = char K > 0. Without using
results from Section 3.7, show that only the pole of z is ramified in K(x)/K(z),
with ramification index e = ps and different exponent d = 2(ps − 1).

3.9. (char K = p > 0) With notation as in the previous exercise, let z =

f (x) = g(x) + h(x) be a polynomial over K of degree n, where
 
g(x) = ai xi , h(x) = ai xi .
pi p|i

Since we assume that f (x) ∈ / K[x ], we have that deg g(x) ≥ 1. Clearly the
p

pole P of z in K(z) is totally ramified in K(x)/K(z), the only place above it

is the pole P∞ of x in K(x).
Show that the different exponent of P∞ /P is given by

d(P∞ /P ) = (n − 1) + (n − deg g(x)) .

3.10. (i) Given an extension F  = F (y) of the function field F/K, where y
satisfies the equation

yn = u ∈ F , with (n, char K) = 1 .

Let P be a place of F/K and P  a place of F  /K lying above P . Consider F  as

the compositum of the fields F and K(y) over the rational function field K(u),
and use Exercise 3.8 (ii) and Abhyankar’s Lemma to obtain the ramification
index of P  |P . This gives another proof of Proposition 3.7.3 (ramification in
Kummer extensions).
(ii) (char K = p > 0) Let F  = F (y), where y satisfies the equation
s
yp − y = u ∈ F .

In a similar way as in (i), consider F  as a compositum of two subfields. Use

Exercise 3.8 (iii) and Abhyankar’s Lemma to show: if there is a pole P ∈ IPF
of u with vP (u) = −m < 0 and (m, p) = 1, then [F  : F ] = ps , P is totally
ramified in F  , and the different exponent of the place P  |P is given by
152 3 Extensions of Algebraic Function Fields

d(P  |P ) = (ps − 1)(m + 1) .

This gives another proof for ramification and different exponents in Artin-
Schreier extensions (Propositions 3.7.8 and 3.7.10).
3.11. (char K = p > 0) Determine the genus of the function field F = K(x, y)
which is given by the equation

y rp + ar−1 y (r−1)p + . . . + a1 y p + a0 y = h(x) ∈ K[x] ,

with ai ∈ K , a0 = 0 , deg h(x) = m and (m, rp) = 1 .
Hint. Use Exercise 3.9.
3.12. (char K = p > 0) Let E/F be a Galois extension of function fields E, F
over K. Let P be a place of F and Q a place of E lying above P . Show:
e(Q|P ) ≡ 0 mod p ⇒ d(Q|P ) ≥ (e(Q|P ) − 1) + (p − 1) .
Compare with Exercise 3.9.
3.13. (i) Assume that g ≥ 2. Let σ : F → F be a homomorphism of F to F
over K (i.e., σ|K is the identity on K), such that F/σ(F ) is separable. Show
that σ is surjective.
(ii) (char K = 2) Let F = K(x, y) with y 2 = x3 − x. We know from Example
3.7.6 that K is the full constant field of F , and F has genus g = 1. Set

(x2 + 1)2 (x2 + 1)(y 4 − 4x4 )

u := 2
and v := .
4y 8x2 y 3
Show that there exists a homomorphism σ : F → F over K with σ(x) = u
and σ(y) = v. The extension F/σ(F ) is separable of degree [F : σ(F )] = 4.
3.14. In this exercise we assume for simplicity that K is algebraically closed
and charK = 0. Let K(x) be a rational function field, F1 = K(x, y), F2 =
K(x, z) and F = F1 F2 = K(x, y, z), where
y m = f (x) ∈ K[x] , z n = g(x) ∈ K[x] ,
f (x) and g(x) are square-free, deg f (x) = r and deg g(x) = s. Assume that
(m, r) = 1 , n|s and (f (x), g(x)) = 1. Determine the genera of F1 , F2 and F ,
and show that Castelnuovo’s Inequality is sharp in this case.
3.15. (i) Consider a Galois extension F/K(x) of degree [F : K(x)] = , with
a prime number . Assume that at least 2 + 1 places of K(x) are ramified in
F/K(x). Show that K(x) is the only rational subfield of F with [F : K(x)] = .
(ii) Assume that  = char K is a prime
 number, and a1 , . . . , a, b1 , . . . , b are
distinct elements of K. Set f (x) = 1≤i≤ (x − ai ) and g(x) = 1≤i≤ (x − bi ),
and consider the function field F = K(x, y) with y = f (x)/g(x). Show that
exactly 2 places of K(x) are ramified in F/K(x), and K(x) is not the only
rational subfield of F with [F : K(x)] = .
 3.12 Exercises 153

3.16. Let σ = id be an automorphism of F/K of finite order, and denote by

F σ the fixed field of σ. Let P ∈ IPF be a place of degree one. Show:
P is totally ramified in F/F σ ⇐⇒ σ(P ) = P .
3.17. Let σ be an automorphism of F/K. Assume that there are 2g+3 distinct
places Pi of degree one with σ(Pi ) = Pi . Show that σ = id.
3.18. For simplicity we assume that K is an algebraically closed field. Let
F/K be a function field of genus g ≥ 2, and let G ⊆ Aut(F/K) be a finite
group of automorphisms of F/K. Assume that gcd (ord G, char K) = 1. Show
that ord G ≤ 84(g − 1) (this estimate is due to Hurwitz).
Hint. Let P1 , . . . , Pr be all places of F G which are ramified in F/F G . Denote
their ramification indices in F/F G by e1 , . . . , er and assume that e1 ≤ e2 ≤
. . . ≤ er . Write down the Hurwitz Genus Formula for F/F G and discuss the
possible cases. The case where the field F G is rational, r = 3, e1 = 2, e2 = 3
and e3 = 7, yields the largest possible value for ord G, namely 84(g − 1).
Remark. One can show that the automorphism group Aut(F/K) is always
finite, for all function fields F/K of genus g ≥ 2 (assuming that K is a perfect
field). However, the estimate ord G ≤ 84(g − 1) does not always hold if ord G
is divisible by the characteristic of K.
3.19. Let E/F be a finite extension of F such that E = F1 F2 is the com-
positum of two intermediate fields F ⊆ Fi ⊆ E , i = 1, 2. Assume that
[F1 : F ] = [E : F2 ]. Let P1 be a place of F1 and P2 a place of F2 with
P1 ∩ F = P2 ∩ F . Show that there exists a place Q of E which satisfies
Q ∩ F1 = P1 and Q ∩ F2 = P2 .
3.20. Assume that F/K has at least one rational place. Show that there exist
x, y ∈ F such that F = K(x, y) and K(x) ∩ K(y) = K.
Hint. If F = K(z) is a rational function field, choose x = z n (z − 1) with n ≥ 2
and y = z(z − 1)h(z) with h(0), h(1) ∈ / {0, ∞}. Why is K(x) ∩ K(y) = K? If
F is not the rational function field, construct x and y in an analogous way.
3.21. (char K = p > 0) Consider a Galois extension F  /F of function fields
over K, a place P ∈ IPF and a place P  ∈ IPF  lying over P . Denote by
Gi = Gi (P  |P ) the i-th ramification group of P  |P . An integer s ≥ 1 is called
a jump of P  |P if Gs  Gs+1 .
Assume that the first ramification group G1 is non-cyclic of order ord G1 = p2 ,
and that P  |P has two jumps s < t, thus

G1 = . . . = Gs  Gs+1 = . . . = Gt  Gt+1 = {id} .

Show that s ≡ t mod p.

Hint. Choose a subgroup H ⊆ G1 with H = Gt and ord H = p. Let E be the
fixed field of H and Q := P  ∩ E. Calculate the different exponent of P  |P in
two ways:
154 3 Extensions of Algebraic Function Fields

(a) by Hilbert’s Different Formula, and

(b) by using transitivity of different exponents for P  ⊃ Q ⊃ P .
Remark. Exercise 3.20 is the simplest special case of the Hasse-Arf Theorem
which states that for every abelian extension F  /F , two consecutive jumps
s < t of P  |P satisfy the congruence s ≡ t mod (G1 : Gt ).
4
Differentials of Algebraic Function Fields

In the previous chapters we have seen that Weil differentials provide a useful
tool for studying algebraic function fields. Now we shall develop the theory
of differentials (beginning with a definition of differentials that is closer to
analysis) and show how these are related to the notion of Weil differentials.
In this chapter we consider an algebraic function field F/K of one
variable. K is the full constant field of F , and K is assumed to be
perfect.

4.1 Derivations and Differentials

We begin with some basic notions.

Definition 4.1.1. Let M be a module (i.e., a vector space) over F . A mapping

δ : F → M is said to be a derivation of F/K, if δ is K-linear and the product
rule
δ(u · v) = u · δ(v) + v · δ(u)
holds for all u, v ∈ F .

Some consequences of this definition are listed in the following lemma.

Lemma 4.1.2. Let δ : F → M be a derivation of F/K into M . Then we

have:
(a) δ(a) = 0 for each a ∈ K.
(b) δ(z n ) = nz n−1 · δ(z) for z ∈ F and n ≥ 0.
(c) If char K = p > 0, then δ(z p ) = 0 for each z ∈ F .
(d) δ(x/y) = (y · δ(x) − x · δ(y))/y 2 for x, y ∈ F and y = 0.

The simple proof of this lemma can be omitted.

H. Stichtenoth, Algebraic Function Fields and Codes, 155

Graduate Texts in Mathematics 254,

c Springer-Verlag Berlin Heidelberg 2009
156 4 Differentials of Algebraic Function Fields

Before we show that some specific derivations exist, we prove a uniqueness

assertion. Recall that an element x ∈ F is called a separating element of F/K
if F/K(x) is a separable algebraic extension, cf. Section 3.10.

Lemma 4.1.3. Suppose that x is a separating element of F/K and that δ1 , δ2 :

F → M are derivations of F/K with δ1 (x) = δ2 (x). Then δ1 = δ2 .

Proof. Lemma  4.1.2(b) implies for a polynomial f (x) = ai xi ∈ K[x] that
δj (f (x)) = ( iai x ) · δj (x) for j = 1, 2, hence δ1 (f (x)) = δ2 (f (x)). For an
i−1

arbitrary element z = f (x)/g(x) ∈ K(x) it follows then, by Lemma 4.1.2(d),

that
g(x) · δ1 (f (x)) − f (x) · δ1 (g(x))
δ1 (z) =
g(x)2
g(x) · δ2 (f (x)) − f (x) · δ2 (g(x))
= = δ2 (z) .
g(x)2

Therefore the restrictions of δ1 and δ2 to  K(x) are equal. Now we consider

an arbitrary element y ∈ F . Let h(T ) = ui T i ∈ K(x)[T ] be its minimal
polynomial over K(x). We apply δj (j = 1, 2) to the equation h(y) = 0 and
obtain
  
 
0 = δj ui y i = ui · δj (y i ) + y i · δj (ui )
  
= iui y i−1 · δj (y) + y i · δj (ui ) .


As y is separable over K(x), the derivative h (y) = iui y i−1 does not vanish,
hence
−1  i
δj (y) =  · y · δj (ui )
h (y)
for j = 1, 2. Since ui ∈ K(x), we know already that δ1 (ui ) = δ2 (ui ), therefore
δ1 (y) = δ2 (y).

Proposition 4.1.4. (a) Suppose that E/F is a finite separable extension of

F and δ0 : F → N is a derivation of F/K into some field N ⊇ E. Then δ0 can
be extended to a derivation δ : E → N . This extension is uniquely determined
by δ0 .
(b) If x ∈ F is a separating element of F/K and N ⊇ F is some field, then
there exists a unique derivation δ : F → N of F/K with the property δ(x) = 1.

Proof. (a) Uniqueness follows from the previous lemma. In order to prove the
existence of an extension of δ0 , we introduce two mappings s and s0 from the
polynomial ring F [T ] into N [T ], namely
 
s(T ) = si T i −→ s (T ) := isi T i−1
 4.1 Derivations and Differentials 157

and  
s(T ) = si T i −→ s0 (T ) := δ0 (si )T i .
Clearly both mappings are K-linear and satisfy the product rule. Now we
choose an element u ∈ E such that E = F (u). Let f (T ) ∈ F [T ] be the
minimal polynomial of u over F and set n := [E : F ] = deg f (T ). Every
element y ∈ E has a unique representation

y = h(u) with h(T ) ∈ F [T ] and deg h(T ) < n .

We define δ : E → N by

f 0 (u) 
δ(y) := h0 (u) − · h (u) (4.1)
f  (u)

and have to verify that δ is a derivation of E which extends δ0 (observe that

f  (u) = 0 since u is separable over F , hence (4.1) makes sense).
First of all, if y ∈ F then h(T ) = y, h (T ) = 0 and h0 (T ) = δ0 (y), hence
(4.1) yields δ(y) = δ0 (y). The K-linearity of δ is obvious, and it remains to
prove the product rule for δ. Consider y, z ∈ E, say y = h(u), z = g(u) with
deg h(T ) < n and deg g(T ) < n. Write g(T ) · h(T ) = c(T ) · f (T ) + r(T ) with
c(T ), r(T ) ∈ F [T ] and deg r(T ) < n, hence y · z = c(u) · f (u) + r(u) = r(u).
Therefore
f0  1
δ(y · z) = (r0 − · r )(u) =  · (r0 f  − f 0 r )(u)
f f (u)
1  
= · (gh − cf )0 · f  − f 0 · (gh − cf ) (u) . (4.2)
f  (u)

We evaluate the terms (gh − cf )0 and (gh − cf ) (using the product rule) and
observe that f (u) = 0. Then (4.2) is reduced to
1
δ(y · z) = · (g 0 hf  + gh0 f  − f 0 g  h − f 0 gh )(u) . (4.3)
f  (u)

On the other hand we obtain from (4.1)

f0  f0 
y · δ(z) + z · δ(y) = h(u) · (g 0 − · g )(u) + g(u) · (h 0
− · h )(u)
f f
1
= · (hg 0 f  − hf 0 g  + gh0 f  − gf 0 h )(u) .
f  (u)

This is in accordance with (4.3).

(b) The uniqueness assertion follows from Lemma 4.1.3. In order to prove the
existence of a derivation δ : F → N with δ(x) = 1 it is sufficient to show that
there is a derivation δ0 : K(x) → N of K(x)/K with δ0 (x) = 1, by (a). We
define δ0 by
158 4 Differentials of Algebraic Function Fields
 
f (x) g(x) · f  (x) − f (x) · g  (x)
δ0 := , (4.4)
g(x) g(x)2
where f (x), g(x) ∈ K[x] and f  (x) denotes the formal derivative of f (x) in
K[x]. Equation (4.4) is well-defined, and it is readily checked that δ0 is a
derivation of K(x)/K with δ0 (x) = 1.

Definition 4.1.5. (a) Let x be a separating element of the function field

F/K. The unique derivation δx : F → F of F/K with the property δx (x) = 1
is called the derivation with respect to x.
(b) Let DerF := {η : F → F | η is a derivation of F/K}. For η1 , η2 ∈ DerF
and z, u ∈ F we define

(η1 + η2 )(z) := η1 (z) + η2 (z) and (u · η1 )(z) := u · η1 (z) .

It is obvious that η1 + η2 and u · η1 are derivations of F/K, and DerF becomes

an F -module in this manner. Hence it is called the module of derivations of
F/K.

Lemma 4.1.6. Let x be a separating element of F/K. Then the following

hold:
(a) For each derivation η ∈ DerF we have η = η(x) · δx . In particular, DerF
is a one-dimensional F -module.
(b) (Chain rule) If y is another separating element of F/K, then

δy = δy (x) · δx . (4.5)

(c) For t ∈ F we have

δx (t) = 0 ⇐⇒ t is a separating element.

Proof. (a) Consider the two derivations η and η(x) · δx of F/K into F . Since
(η(x) · δx )(x) = η(x) · δx (x) = η(x) and x is separating, Lemma 4.1.3 implies
that η(x) · δx = η.
(b) This is a special case of (a).
(c) If t is separating, 1 = δt (t) = δt (x)·δx (t) (here we have used the definition of
δt and the chain rule). Hence δx (t) = 0. Suppose now that t is not separating.
If char K = 0 then t ∈ K and δx (t) = 0, since all derivations of F/K vanish on
K. If char K = p > 0, then t = up for some u ∈ F (see Proposition 3.10.2(d)),
and δx (t) = δx (up ) = 0 by Lemma 4.1.2.

We are now ready to introduce the notion of a differential of F/K.

 4.1 Derivations and Differentials 159

Definition 4.1.7. (a) On the set Z := {(u, x) ∈ F × F | x is separating } we

define a relation ∼ by

(u, x) ∼ (v, y) : ⇐⇒ v = u · δy (x) . (4.6)

Using the chain rule (4.5), it is readily verified that ∼ is an equivalence relation
on Z.
(b) We denote the equivalence class of (u, x) ∈ Z with respect to the above
equivalence relation by u dx and call it a differential of F/K. The equivalence
class of (1, x) is simply denoted by dx. Observe that by (4.6),

u dx = v dy ⇐⇒ v = u · δy (x) . (4.7)

(c) Let
ΔF := { u dx | u ∈ F, and x ∈ F is separating }
be the set of all differentials of F/K. We define the sum of two differentials
u dx , v dy ∈ ΔF as follows: Choose a separating element z; then

u dx = (u · δz (x)) dz and v dy = (v · δz (y)) dz ,

by (4.7), and we set

u dx + v dy := (u · δz (x) + v · δz (y)) dz (4.8)

This definition (4.8) is independent of the choice of z by the chain rule. Like-
wise, we define
w · (u dx) := (wu) dx ∈ ΔF
for w ∈ F and u dx ∈ ΔF . One checks easily that ΔF becomes an F -module
in this manner.
(d) For a non-separating element t ∈ F we define dt := 0 (the zero element
of ΔF ); thus we obtain a mapping

F −→ ΔF ,
d: (4.9)
t −→ dt .

The pair (ΔF , d) is called the differential module of F/K ( for brevity we shall
simply refer to ΔF as the differential module of F/K).

The main properties of the differential module are put together in the
following proposition.

Proposition 4.1.8. (a) Let z ∈ F be separating. Then dz = 0, and every

differential ω ∈ ΔF can uniquely be written in the form ω = u dz with u ∈ F .
Hence ΔF is a one-dimensional F -module.
160 4 Differentials of Algebraic Function Fields

(b) The map d : F → ΔF as defined in (4.9) is a derivation of F/K; i.e.,

d(ax) = a dx, d(x + y) = dx + dy , and d(xy) = x dy + y dx

for all x, y ∈ F and a ∈ K.

(c) For t ∈ F we have

dt = 0 ⇐⇒ t is separating.

(d) Suppose that δ : F → M is a derivation of F/K into some F -module M .

Then there exists a unique F -linear map μ : ΔF → M such that δ = μ ◦ d.

Proof. (a) The differential 0 = 0 dz is the zero element of ΔF . By (4.6) we see

immediately that (0, z) is not equivalent to (1, z), hence dz = 0.
Consider now an arbitrary differential ω ∈ ΔF , say ω = v dy with a sepa-
rating element y. Set u := v · δz (y). Using (4.7) we obtain

u dz = (v · δz (y)) dz = v dy = ω .

The uniqueness of u is evident, since dz = 0 and ΔF is a vector space over

the field F .
(b) Fix a separating element z ∈ F . For all t ∈ F we have

dt = δz (t) dz . (4.10)

(For separating t this follows from (4.7). If t is not separating, dt = 0 by

definition, and δz (t) dz = 0 by Lemma 4.1.6.) Using (4.10), it is easily shown
that d : F → ΔF is a derivation of F/K. We prove only the product formula.
Since δz is a derivation of F/K we get

d(xy) = δz (xy) dz = (x · δz (y) + y · δz (x)) dz

= x · (δz (y) dz) + y · (δz (x) dz) = x dy + y dx .

(c) Clear from the definition of d.

(d) Now there is given a derivation δ : F → M . By (a), each ω ∈ ΔF is
uniquely written as ω = u dz, and we can define μ : ΔF → M by μ(ω) :=
u · δ(z). Obviously μ is F -linear. In order to show that δ = μ ◦ d we have only
to prove that
δ(z) = (μ ◦ d)(z) (4.11)
(by Lemma 4.1.3). Equation (4.11) holds trivially by definition of μ.
It remains to prove the uniqueness of μ. Suppose that ν : ΔF → M is
F -linear and δ = ν ◦ d. Then

ν(u dz) = u · ν(dz) = u · ((ν ◦ d)(z)) = u · δ(z) = μ(u dz) .

Hence ν = μ.
 4.2 The P -adic Completion 161

Remark 4.1.9. (a) A differential of the specific form ω = dx (with x ∈ F ) is

said to be exact; the exact differentials form a K-subspace of ΔF .
(b) Since ΔF is a one-dimensional F -module, one can define the quotient
ω1 /ω2 ∈ F for ω1 , ω2 ∈ ΔF and ω2 = 0 by setting
ω1
u= : ⇐⇒ ω1 = uω2 .
ω2
In particular, if z ∈ F is separating and y ∈ F , the quotient dy/dz is defined,
and we have
dy
δz (y) = ,
dz
by Equation (4.10). Using this notation, some previous formulas can be writ-
ten in a more suggestive manner, e.g.
dx dy
u dx = v dy ⇐⇒ v = u · ⇐⇒ u = v · (4.12)
dy dx
and
dy dy dz
= · , (4.13)
dx dz dx
if x and z are separating. The first one of these formulas corresponds to (4.7),
the second one is the chain rule (4.5).

4.2 The P -adic Completion

The real number field IR is the completion of the rational number field Q with
respect to the ordinary absolute value. That means: (1) the field Q is dense in
IR, and (2) every Cauchy sequence in IR is convergent. In the present section
we shall consider an analogous situation, namely the completion of a function
field F/K with respect to a place P ∈ IPF . This will provide us with a useful
tool for calculating the derivation dz/dt (where t is a P -prime element) and
will also enable us to define the residue of a differential at the place P . But
first we need to generalize slightly some earlier notions.

Definition 4.2.1. A discrete valuation of a field T is a surjective mapping

v : T → ZZ ∪ {∞} which satisfies
(1) v(x) = ∞ ⇐⇒ x = 0.
(2) v(xy) = v(x) + v(y) for all x, y ∈ T .
(3) v(x + y) ≥ min {v(x), v(y)} for all x, y ∈ T (Triangle Inequality).

The field T (more precisely: the pair (T, v)) is called a valued field. As in
Lemma 1.1.11 one can easily prove the Strict Triangle Inequality

v(x + y) = min {v(x), v(y)} if x, y ∈ T and v(x) = v(y) .

162 4 Differentials of Algebraic Function Fields

We say that a sequence (xn )n≥0 in T is convergent if there exists an element

x ∈ T (called the limit of the sequence) which satisfies:

for every c ∈ IR there is an index n0 ∈ IN

such that v(x − xn ) ≥ c whenever n ≥ n0 .

A sequence (xn )n≥0 is called a Cauchy sequence if it has the following property:

for every c ∈ IR there is an index n0 ∈ IN

such that v(xn − xm ) ≥ c whenever n, m ≥ n0 .

As in analysis one can readily verify the following facts:

(a) If a sequence (xn )n≥0 is convergent, then its limit x ∈ T is unique. There-
fore we can write x = limn→∞ xn .
(b) Every convergent sequence is a Cauchy sequence.
In general it is not true that all Cauchy sequences are convergent. Hence
we introduce the following notions.

Definition 4.2.2. (a) A valued field T is said to be complete if every Cauchy

sequence in T is convergent.
(b) Suppose that (T, v) is a valued field. A completion of T is a valued field
(T̂ , v̂) with the following properties:
(1) T ⊆ T̂ , and v is the restriction of v̂ to T .
(2) T̂ is complete with respect to the valuation v̂.
(3) T is dense in T̂ ; i.e., for each z ∈ T̂ there is a sequence (xn )n≥0 in T
with limn→∞ xn = z.

Proposition 4.2.3. For each valued field (T, v) there exists a completion
(T̂ , v̂). It is unique in the following sense: If (T̃ , ṽ) is another completion
of (T, v) then there is a unique isomorphism f : T̂ → T̃ such that v̂ = ṽ ◦ f .
Hence (T̂ , v̂) is called the completion of (T, v).

Proof. We give only a sketch of the proof; the tedious details are left to the
reader. First of all, we consider the set

R := {(xn )n≥0 | (xn )n≥0 is a Cauchy sequence in T } .

This is a ring if addition and multiplication are defined in the obvious manner
via (xn ) + (yn ) := (xn + yn ) and (xn ) · (yn ) := (xn yn ). The set

I := {(xn )n≥0 | (xn )n≥0 converges to 0}

is an ideal in R; actually I is a maximal ideal of R. Therefore the residue class

ring
 4.2 The P -adic Completion 163

T̂ := R/I
is a field. For x ∈ T let (x) := (x, x, . . .) ∈ R be the constant sequence and
ν(x) := (x) + I ∈ T̂ . It is obvious that ν : T → T̂ is an embedding, and we
can consider T as a subfield of T̂ via this embedding.
Now we construct a valuation v̂ on T̂ as follows. If (xn )n≥0 is a Cauchy
sequence in T , either
lim v(xn ) = ∞
n→∞

(in this case, (xn )n≥0 ∈ I), or there is an integer n0 ≥ 0 such that

v(xn ) = v(xm ) for all m, n ≥ n0 .

This follows easily from the Strict Triangle Inequality. In any case the limit
limn→∞ v(xn ) exists in ZZ ∪ {∞}. Moreover, if (xn ) − (yn ) ∈ I then we have
limn→∞ v(xn ) = limn→∞ v(yn ). Hence we can define the function v̂ : T̂ →
ZZ ∪ {∞} by
v̂((xn )n≥0 + I) := lim v(xn ) .
n→∞

Using the corresponding properties of v, it is easily verified that v̂ is a valuation

of T̂ and v̂(x) = v(x) for x ∈ T .
Next we consider a Cauchy sequence (zm )m≥0 in T̂ , say

zm = (xmn )n≥0 + I with (xmn )n≥0 ∈ R .

Then the diagonal sequence (xnn )n≥0 is a Cauchy sequence in T and

lim zn = (xnn )n≥0 + I ∈ T̂ .

n→∞

Thus T̂ is complete with respect to v̂.

Now let z = (xn )n≥0 + I be an element of T̂ . Upon checking, one finds
that z = limn→∞ xn , hence T is dense in T̂ .
Thus far we have shown that a completion (T̂ , v̂) of (T, v) exists. Suppose
that (T̃ , ṽ) is another completion of (T, v). For the moment, we denote by v̂-lim
(resp. ṽ-lim) the limit of a sequence in T̂ (resp. T̃ ). Then we can construct a
mapping f : T̂ → T̃ as follows: if z ∈ T̂ is represented as

z = v̂- lim xn with xn ∈ T ,

n→∞

we define
f (z) := ṽ- lim xn .
n→∞

It turns out that f is a well-defined isomorphism of T̂ onto T̃ with the addi-

tional property v̂ = ṽ ◦ f .
164 4 Differentials of Algebraic Function Fields

It is often more convenient to consider convergent series instead m of se-

quences. Let (zn )n≥0 be a sequence
∞ in a valued field (T, v) and sm := i=0 zi .
We say that the infinite series i=0 zi is convergent if the sequence of its par-
tial sums (sm )m≥0 is convergent; in this case we write, as usual,
∞

zi := lim sm .
m→∞
i=0

In a complete field there is a very simple criterion for convergence of an

infinite series.

Lemma 4.2.4. Let (zn )n≥0 be a sequence

∞ in a complete valued field (T, v).
Then we have: The infinite series i=0 i is convergent if and only if the
z
sequence (zn )n≥0 converges to 0.
Proof. 
Suppose that (zn )n≥0 converges to 0. Consider the m-th partial sum
m
sm := i=0 zi . For n > m we have
 
n 
v(sn − sm ) = v zi ≥ min{v(zi ) | m < i ≤ n} ≥ min{v(zi ) | i > m} .
i=m+1

Since v(zi ) → ∞ for i → ∞, this shows that the sequence (sn )n≥0 is a Cauchy
sequence in T , hence convergent.
The converse statement is easy; its proof is the same as in analysis.

Now we specialize the above results to the case of an algebraic function

field F/K.

Definition 4.2.5. Let P be a place of F/K. The completion of F with respect

to the valuation vP is called the P -adic completion of F . We denote this
completion by F̂P and the valuation of F̂P by vP .

Theorem 4.2.6. Let P ∈ IPF be a place of degree one and let t ∈ F be a

P -prime element. Then every element z ∈ F̂P has a unique representation of
the form
∞
z= ai ti with n ∈ ZZ and ai ∈ K . (4.14)
i=n
This representation is called the P -adic power series expansion of z with re-
spect to t.
∞
Conversely, if (ci )i≥n is a sequence in K, then the series i=n ci ti con-
verges in F̂P , and we have
∞ 
vP ci ti = min{i | ci = 0} .
i=n
 4.2 The P -adic Completion 165

Proof. First we prove the existence of a representation of the form in (4.14).

Given z ∈ F̂P we choose n ∈ ZZ with n ≤ vP (z). There is an element y ∈ F
with vP (z − y) > n (since F is dense in F̂P ). By the Triangle Inequality it
follows that vP (y) ≥ n, hence vP (yt−n ) ≥ 0. As P is a place of degree one,
there is an element an ∈ K with vP (yt−n − an ) > 0, and
 
vP (z − an tn ) = vP (z − y) + (y − an tn ) > n .

In the same manner, we find an+1 ∈ K such that

vP (z − an tn − an+1 tn+1 ) > n + 1 .

Iterating this construction, we obtain an infinite sequence an , an+1 , an+2 , . . .

in K such that  
m
vP z − i
ai t > m
i=n

for all m ≥ n. This shows that

∞

z= ai ti .
i=n

In order to prove uniqueness we consider another sequence (bi )i≥m in K

which satisfies
∞
 ∞

z= ai ti = bi t i .
i=n i=m

We can assume that n = m (otherwise, if n < m, define bi := 0 for n ≤ i < m).

Suppose there is some j with aj = bj . We choose j minimal with this property
and obtain for all k > j

k 
k   
k 
vP ai ti − bi t i = vP (aj − bj )tj + (ai − bi )ti = j (4.15)
i=n i=n i=j+1
 
(since vP (aj − bj )tj = j, the Strict Triangle Inequality applies). On the
other hand,

k 
k  
k 
k 
vP ai ti − bi t i = vP ai ti − z + z − bi t i
i=n i=n i=n i=n
    

k 
k
≥ min vP z − ai t i
, vP z − bi t i
. (4.16)
i=n i=n

For k → ∞, (4.16) tends to infinity. This is a contradiction to (4.15) and

proves that the representation (4.14) is unique.
166 4 Differentials of Algebraic Function Fields

Finally we consider an arbitrary sequence (ci )i≥n in K. As vP (ci ti ) ≥ i for

all i, the sequence (ci ti )i≥n converges to 0. Hence by Lemma 4.2.4 the series
∞ i
i=n ci t is convergent in F̂P , say

∞

ci ti =: y ∈ F̂P .
i=n

Set j0 := min {i | ci = 0}. If j0 = ∞ then all ci = 0, hence y = 0 and

vP (y) = ∞ as well. In the case j0 < ∞ we have for all k ≥ j0

k 
vP ci ti = j0
i=n

by the Strict Triangle Inequality. Since

 
k 
vP y − ci ti > j0
i=n

for all sufficiently large k, this implies

 
k 
k 
vP (y) = vP y − ci ti + ci ti
i=n i=n
    

k k
= min vP y − ci ti , vP ci ti = j0 .
i=n i=n

We continue to consider a place P of F/K of degree one and a P -prime el-

ement t. By Proposition 3.10.2, t is a separating element of F/K, and thus one
can speak of the derivation δt : F → F with respect to t (cf. Definition 4.1.5).
Using the P -adic power series expansion, we can easily calculate dz/dt = δt (z)
for z ∈ F (the notation dz/dt is explained in Remark 4.1.9).

∞let t ∈ F be
Proposition 4.2.7. Let P be a place of F/K of degree one and
a P -prime element. If z ∈ F has the P -adic expansion z = i=n ai ti with
coefficients ai ∈ K, then
∞
dz 
= iai ti−1 .
dt i=n

Proof. We define a mapping δ : F̂P → F̂P by


∞  ∞
i
δ ci t := ici ti−1 .
i=m i=m
 4.2 The P -adic Completion 167

This mapping is obviously K-linear and satisfies the product rule δ(u · v) =
u · δ(v) + v · δ(u) for all u, v ∈ F̂P (the verification of the product rule is a
bit technical but straightforward). Moreover δ(t) = 1 holds. Therefore δ(z) =
δt (z) = dz/dt for each z ∈ F , by Proposition 4.1.4(b) and Definition 4.1.5.

Our next goal is to introduce the residue of a differential ω ∈ ΔF at a

place P . For this we need some background.

Definition 4.2.8. Suppose that P is a place of F/K of degree  one and t ∈ F

∞
is a P -prime element. If z ∈ F has the P -adic expansion z = i=n ai ti with
n ∈ ZZ and ai ∈ K we define its residue with respect to P and t by

resP,t (z) := a−1 .

Clearly, resP,t : F −→ K is a K-linear map and

resP,t (z) = 0 if vP (z) ≥ 0 . (4.17)

The residue satisfies the following transformation formula:

Proposition 4.2.9. Let s, t ∈ F be P -prime elements (where P is a place of

degree one). Then
 ds 
resP,s (z) = resP,t z ·
dt
for all z ∈ F .

Proof. The power series expansion of s with respect to t has the following
form (see Theorem 4.2.6):
∞

s= ci ti with c1 = 0 .
i=1

Proposition 4.2.7 yields

∞
ds
= c1 + ici ti−1 . (4.18)
dt i=2

Now we distinguish several cases.

Case 1. vP (z) ≥ 0. Then vP (z · ds/dt) ≥ 0 as well (by (4.18)), and from
(4.17) it follows that
 ds 
resP,s (z) = resP,t z · = 0.
dt

Case 2. z = s−1 . Clearly we have resP,s (s−1 ) = 1. We determine the power

series expansion of s−1 with respect to t:
168 4 Differentials of Algebraic Function Fields

1 1  c2 c3 −1
s−1 = = · 1+ t + t2 + · · ·
c1 t + c2 t + · · ·
2 c1 t c1 c1
 ∞ c r 
1 2 c3 2
= · 1+ (−1)r t+ t + ···
c1 t r=1
c1 c1
1  f2 (c2 ) f3 (c2 , c3 ) 2 
= 1+ t+ 2 t + ··· (4.19)
c1 t c1 c1
with certain polynomials fj (X2 , . . . , Xj ) ∈ ZZ[X2 , . . . , Xj ]. Therefore

ds 1
s−1 · = +y with vP (y) ≥ 0
dt t
from (4.18) and (4.19), and we obtain
 ds 
resP,t s−1 · = 1 + resP,t (y) = 1
dt
by Case 1.
Case 3. z = s−n with n ≥ 2. Here resP,s (s−n ) = 0. To begin with, we
calculate resP,t (s−n · ds/dt) in the case when char K = 0. Then

ds 1 d(s−n+1 )
s−n · = · .
dt −n + 1 dt
We write
∞

s−n+1 = di ti
i=k

with k = −n + 1 and di ∈ K and obtain

∞
d(s−n+1 ) 
= idi ti−1 .
dt
i=k

Hence  
 ds  1
∞
resP,t s−n · = · resP,t idi ti−1 = 0 . (4.20)
dt −n + 1
i=k

Next we consider Case 3 in arbitrary characteristic. By (4.18) and (4.19) we

find
ds 1    f2 (c2 ) f2 (c2 , c3 ) 2 n
s−n · = n n c1 + 2c2 t + · · · · 1 + t+ t + · · ·
dt c1 t c1 c21

1  g2 (c1 , c2 ) g3 (c1 , c2 , c3 ) 2 
= c1 + t+ t + ···
cn1 tn c1 c12

with polynomials gj (X1 , . . . , Xj ) ∈ ZZ[X1 , . . . , Xj ]. These polynomials are in-

dependent of the characteristic of K, and we have
 4.2 The P -adic Completion 169
 ds  1
resP,t s−n · = 2n−1 · gn (c1 , · · · , cn ) .
dt c1

From (4.20) it follows that gn (c1 , . . . , cn ) = 0 for any elements c1 = 0, c2 , . . . , cn

in a field of characteristic zero, so gn (X1 , . . . , Xn ) must be the zero polynomial
in ZZ[X1 , . . . , Xn ]. Thus the equality
 ds 
resP,t s−n · = 0 = resP,s (s−n )
dt
holds for a field of arbitrary characteristic (for n ≥ 2).
Case 4. Finally let z be an arbitrary element of F with vP (z) < 0, say
∞

z= ai si with n≥1 and ai ∈ K .
i=−n

Then resP,s (z) = a−1 and z = a−n s−n + · · · + a−1 s−1 + y with vP (y) ≥ 0.
Using the results of Cases 1, 2 and 3 we get
 ds  −1
  ds   ds 
resP,t z · = ai · resP,t si · + resP,t y ·
dt i=−n
dt dt

 ds 
= a−1 · resP,t s−1 · = a−1 = resP,s (z) .
dt

Definition 4.2.10. Let ω ∈ ΔF be a differential and let P ∈ IPF be a place of

degree one. Choose a P -prime element t ∈ F and write ω = u dt with u ∈ F .
Then we define the residue of ω at P by

resP (ω) := resP,t (u) .

This definition is independent of the specific choice of the prime element t.

In fact, if s is another P -prime element and ω = u dt = z ds, then u = z ·ds/dt,
and Proposition 4.2.9 yields
 ds 
resP,s (z) = resP,t z · = resP,t (u) .
dt
We will show in the following section that the residue of a differential at
a place P of degree one has an interpretation as the local component of a
specific Weil differential at this place.
170 4 Differentials of Algebraic Function Fields

4.3 Differentials and Weil Differentials

The goal in this section is to establish a relationship between the notions of
differentials and Weil differentials of an algebraic function field. As always, K
is assumed to be perfect.
To begin with, we recall some notations and results from previous chapters
(in particular Sections 1.5, 1.7 and 3.4). AF denotes the adele space of F/K;
its elements are adeles α = (αP )P ∈IPF where the P -component αP of α is an
element of F and vP (α) := vP (αP ) ≥ 0 for almost all P ∈ IPF . The field F is
considered as a subspace of AF via the diagonal embedding F → AF . If A is a
divisor of F/K one considers the space AF (A) = {α ∈ AF | vP (α) ≥ −vP (A)
for all P ∈ IPF }. A Weil differential of F is a K-linear map ω : AF → K that
vanishes on AF (A) + F for some divisor A. The Weil differentials constitute
a one-dimensional F -module ΩF . For 0 = ω ∈ ΩF there exists a uniquely
determined divisor W = (ω) ∈ Div(F ) such that ω vanishes on AF (W ) but
not on AF (B) for each divisor B > W . Such a divisor (ω) is called a canonical
divisor of F/K. For each P ∈ IPF we have another embedding ιP : F → AF
where ιP (z) is the adele whose P -component is z, and all other components
of ιP (z) are 0. The local component of the Weil differential ω at the place P
is the mapping ωP : F → K given by ωP (z) := ω(ιP (z)).
If F  /F is a finite separable extension of function fields we have defined the
cotrace ω  := CotrF /F (ω) of a Weil differential ω ∈ ΩF ; this is a Weil differ-
ential of F  , and if F  and F have the same constant field, ω  is characterized
by the condition 
ωP (TrF /F (y)) = ωP  (y) (4.21)
P  |P

for all P ∈ IPF and y ∈ F , cf. Theorem 3.4.6 and Remark 3.4.8. In Proposition
1.7.4 we have shown the existence of a specific Weil differential η of the ra-
tional function field K(x)/K which is uniquely determined by the following
properties:
the divisor of η is (η) = −2P∞ , and ηP∞ (x−1 ) = −1 . (4.22)
(P∞ is the pole of x in K(x), and ηP∞ is the local component of η at P∞ .)
Definition 4.3.1. Let F/K be an algebraic function field. We define a map-
ping 
F → ΩF ,
δ:
x → δ(x)
as follows: if x ∈ F \K is a separating element of F/K we set
δ(x) := CotrF/K(x) (η) ,
where η ∈ ΩK(x) is the Weil differential of K(x)/K characterized by (4.22).
For a non-separating element x ∈ F we define δ(x) := 0. We call δ(x) the
Weil differential of F/K associated with x.
 4.3 Differentials and Weil Differentials 171

Note that δ(x) = 0 if x is separating, hence each Weil differential ω ∈ ΩF

can be written as ω = z · δ(x) with z ∈ F . Now we state the main results of
this section.
Theorem 4.3.2. Suppose that F/K is an algebraic function field over a per-
fect field K, and let x ∈ F be a separating element.
(a) The map δ : F → ΩF given by Definition 4.3.1 is a derivation of F/K.
(b) For every y ∈ F we have
dy
δ(y) = · δ(x) .
dx
(c) The map 
ΔF → ΩF ,
μ:
z dx → z · δ(x)
is an isomorphism of the differential module ΔF onto ΩF . This isomorphism
is compatible with the derivations d : F → ΔF and δ : F → ΩF , that means
μ ◦ d = δ.
(d) If P ∈ IPF is a place of F/K of degree one and ω = z · δ(x) ∈ ΩF , the
local component of ω at P is given by
(z · δ(x))P (u) = resP (uz dx) .
In particular,
(z · δ(x))P (1) = resP (z dx) .
(e) If ω = z · δ(t) ∈ ΩF and t is a prime element at the place P , then we have
vP (ω) = vP (z).

An immediate consequence of this theorem is

Corollary 4.3.3 (Residue Theorem). Let F/K be an algebraic function
field over an algebraically closed field, and let ω ∈ ΔF be a differential of
F/K. Then resP (ω) = 0 for almost all places P ∈ IPF , and

resP (ω) = 0 .
P ∈IPF

Proof of the Corollary. Write ω = z dx with z ∈ F and a separating element

x ∈ F . By Theorem 4.3.2(d) we have resP (ω) = (z·δ(x))P (1). Now Proposition
1.7.2 yields the desired result.

The proof of Theorem 4.3.2 is rather tedious. First we shall assume that
the constant field is algebraically closed. The case of an arbitrary perfect
constant field K will then be reduced to this special case by considering the
constant field extension F̄ = F K̄ of F with the algebraic closure K̄ of K. We
begin with some preliminaries.
172 4 Differentials of Algebraic Function Fields

Lemma 4.3.4. Suppose that F/F0 is a finite separable extension of algebraic

function fields over an algebraically closed field K. Let ψ ∈ ΩF0 be a Weil
differential of F0 /K and ω := CotrF/F0 (ψ). Consider a place P0 ∈ IPF0 which
is unramified in F/F0 , and a place P ∈ IPF which lies over P0 . Then we have

ωP (z) = ψP0 (z) for all z ∈ F0 .

Proof. We can assume that ψ = 0. Let P1 , . . . , Pn ∈ IPF be the places of

F lying over P0 , say P1 = P . Since P0 is unramified in F/F0 and K is
algebraically closed, n = [F : F0 ] by Theorem 3.1.11. By the Approximation
Theorem we can find an element z  ∈ F satisfying

vP (z  − z) ≥ −vP0 (ψ) ,
vPi (z  ) ≥ −vP0 (ψ) for i = 2, . . . , n . (4.23)

(Recall that the integer vQ (ψ) is defined by vQ (ψ) := vQ (W ) where W = (ψ)

denotes the divisor of ψ.) Since P0 is unramified in F/F0 and ω = CotrF/F0 (ψ),

vPi (ω) = vP0 (ψ) for i = 1, . . . , n (4.24)

by Theorem 3.4.6 and Dedekind’s Different Theorem. Consider the adele α =

(αQ )Q∈IPF with

αP := z  − z ,
αPi := z  for i = 2, . . . , n , and
αQ := 0 for Q = P1 , . . . , Pn .

Then α is AF ((ω)) by (4.23) and (4.24), hence


n
ωP (z) = ωP (z) + ω(α) = ωP (z) + ωP (z  − z) + ωPi (z  )
i=2

n
= ωPi (z  ) = ψP0 (TrF/F0 (z  )) . (4.25)
i=1

(The last equality in (4.25) follows from (4.21).) By (4.25), the proof of our
lemma will be finished when we show that

ψP0 (TrF/F0 (z  )) = ψP0 (z) . (4.26)

The trace TrF/F0 can be evaluated by using the embeddings of F/F0 into an
extension field of F . We proceed as follows: Choose a Galois closure E ⊇ F
of F/F0 (i.e., F0 ⊆ F ⊆ E, the extension E/F0 is Galois, and E is minimal
with this property). Then P0 is unramified in E/F0 by Corollary 3.9.3. Choose
places Q1 = Q, Q2 , . . . , Qn ∈ IPE with Qi | Pi . Since E/F0 is Galois, there
exist automorphisms σ1 , . . . , σn ∈ Gal(E/F0 ) such that
 4.3 Differentials and Weil Differentials 173

σi−1 (Q) = Qi for i = 1, . . . , n , (4.27)

cf. Theorem 3.7.1. We assert that the restrictions σi |F (i = 1, . . . , n) of σi to

F are pairwise distinct. In fact, if σi |F = σj |F , we have for each u ∈ F

vPi (u) = vQi (u) = vσ−1 (Q) (u) = vQ (σi (u)) = vQ (σj (u)) = vPj (u)
i

(we have used (4.27) and Lemma 3.5.2(a)). Therefore i = j, and the embed-
dings σi |F : F → E are pairwise distinct. Hence


n
TrF/F0 (u) = σi (u)
i=1

for u ∈ F . Now we can verify (4.26). By (4.23) and (4.27),

vQ (z  − z) ≥ −vP0 (ψ) and

vQ (σi (z  )) = vQi (z  ) = vPi (z  ) ≥ −vP0 (ψ)

for i = 2, . . . , n. Thus

vP0 (TrF/F0 (z  ) − z) = vQ (TrF/F0 (z  ) − z)

 
n
 
= vQ (z − z) + σi (z ) ≥ −vP0 (ψ) .
i=2

Using Proposition 1.7.3 we obtain

ψP0 (TrF/F0 (z  ) − z) = 0 ,

hence (4.26) follows.

Lemma 4.3.5. Let F be an algebraic function field over an algebraically

closed field K. Suppose that x is a separating element of F/K and P0 ∈ IPK(x)
satisfies the following conditions:
(1) P0 is unramified in F/K(x).
(2) P0 is not the pole of x in K(x).
If δ(x) ∈ ΩF denotes the Weil differential associated with x (as defined in
Definition 4.3.1) and u ∈ F , then

δ(x)P (u) = resP (u dx) (4.28)

holds for all P ∈ IPF with P |P0 .

Proof. By (1) and (2) there exists an element a ∈ K such that t := x−a is a P -
prime element. As always, P∞ denotes the pole of x in K(x). Consider the Weil
differential η ∈ ΩK(x) which is given by (4.22); then δ(x) = CotrF/K(x) (η).
174 4 Differentials of Algebraic Function Fields

First we evaluate the left-hand side of (4.28) for u = tk , k ∈ ZZ. By the

previous lemma and Proposition 1.7.4(c) we obtain

k k 0 for k = −1 ,
δ(x)P (t ) = ηP0 (t ) = (4.29)
1 for k = −1 .

An arbitrary element u ∈ F can be written as


l−1
u= aν tν + u
ν=m

with aν ∈ K, l ≥ max{0, −vP (δ(x))} and vP (u ) ≥ l; this follows easily from
the P -adic power series expansion of u with respect to t (Theorem 4.2.6). We
have δ(x)P (u ) = 0 by Proposition 1.7.3, so (4.29) implies


l−1
δ(x)P (u) = aν · δ(x)P (tν ) = a−1 .
ν=m

On the other hand, dt = d(x − a) = dx, hence

resP (u dx) = resP (u dt) = resP,t (u) = a−1 .

This proves (4.28).

Proof of Theorem 4.3.2 under the additional hypothesis that K is algebraically

closed. We begin with (b). If y ∈ F is not separating, δ(y) = 0 and dy/dx = 0
(see Definition 4.3.1 and Proposition 4.1.8). So δ(y) = (dy/dx) · δ(x) in this
case. Henceforth we can assume that y is separating. As δ(x) = 0 and ΩF
is a one-dimensional F -module (Proposition 1.5.9), δ(y) = z · δ(x) for some
z ∈ F . Only finitely many places of F are ramified over K(x) or K(y) (cf.
Corollary 3.5.5), and we can find a place P ∈ IPF such that the restrictions of
P to K(x) (resp. K(y)) are unramified in F/K(x) (resp. F/K(y)), and P is
neither a pole of x nor a pole of y. For every u ∈ F ,

δ(y)P (u) = (z · δ(x))P (u) = δ(x)P (zu)

holds. On the other hand, Lemma 4.3.5 (applied to the extensions F/K(y)
and F/K(x)) yields
   
dy dy
δ(y)P (u) = resP (u dy) = resP u dx = δ(x)P u .
dx dx
Hence  
dy
δ(x)P u(z − ) = 0 for all u ∈ F .
dx
This implies z = dy/dx (by Proposition 1.7.3), and proves (b).
 4.3 Differentials and Weil Differentials 175

(a) We use (b) and Proposition 4.1.8(b) and obtain for y1 , y2 ∈ F and a ∈ K
 
d(y1 + y2 ) dy1 dy2
δ(y1 + y2 ) = · δ(x) = + · δ(x) = δ(y1 ) + δ(y2 ) ,
dx dx dx
d(ay1 ) dy1
δ(ay1 ) = · δ(x) = a · · δ(x) = a · δ(y1 ) ,
dx dx
and the product rule
 
d(y1 y2 ) dy2 dy1
δ(y1 y2 ) = · δ(x) = y1 + y2 · δ(x) = y1 · δ(y2 ) + y2 · δ(y1 ) .
dx dx dx

(c) By Proposition 4.1.8(d) there exists a uniquely determined F -linear map

μ : ΔF → ΩF such that δ(y) = (μ ◦ d)(y) = μ(dy) for all y ∈ F . Since μ is
F -linear, μ(z dx) = z ·μ(dx) = z ·δ(x). As dx generates ΔF and δ(x) generates
ΩF (as F -modules), μ is bijective.
(d) This is a generalization of Lemma 4.3.5, and we want to reduce the proof
of (d) to this lemma. We choose sufficiently many pairwise distinct places
P1 := P, P2 , . . . , Pr such that L (P1 + P2 + . . . + Pr ) is strictly larger than
L (P2 + . . . + Pr ). Each element

t1 ∈ L (P1 + P2 + . . . + Pr )\L (P2 + . . . + Pr )

has only simple poles in F , and P = P1 is one of them. Setting t := t−1

1 and
P0 := P ∩ K(t) ∈ IPK(t) we have seen that P0 is unramified in the separable
extension F/K(t) and that P0 is not the pole of t in K(t). Now (b) and Lemma
4.3.5 yield
   
dx dx
(z · δ(x))P (u) = z · · δ(t) (u) = δ(t)P uz
dt dt
 P
dx
= resP uz dt = resP (uz dx) .
dt

(e) Since t is a P -prime element, the extension F/K(t) is separable

(Proposition 3.10.2), and P is unramified in F/K(t). The divisor of the Weil
differential δ(t) is given by

(δ(t)) = −2(t)∞ + Diff(F/K(t)) ,

where (t)∞ is the pole divisor of t in F , by (4.22) and Theorem 3.4.6. As P is

not a pole of t and P does not occur in the different of F/K(t) (by Dedekind’s
Different Theorem), we have vP (δ(t)) = 0. Hence

vP (z · δ(t)) = vP (z) + vP (δ(t)) = vP (z) .

176 4 Differentials of Algebraic Function Fields

Now we consider a function field F/K where K is an arbitrary perfect

field. Our aim is to prove Theorem 4.3.2 also in this case. Let K̄ ⊇ K be
an algebraic closure of K and let F̄ := F K̄ be the constant field extension of
F/K with K̄. Each place of F/K has finitely many extensions in F̄ ; conversely,
each place of F̄ /K̄ is an extension of some place of F/K. So we have a natural
embedding AF → AF̄ of the adele spaces and principal adeles of F/K are
mapped to principal adeles of F̄ /K̄ under this embedding. Hence we shall
regard AF as a subspace of AF̄ .
We fix a separating element x ∈ F \K; then x is separating for F̄ /K̄ as well,
and [F : K(x)] = [F̄ : K̄(x)], cf. Proposition 3.6.1. For y ∈ F let δ(y) ∈ ΩF
(resp. δ̄(y) ∈ ΩF̄ ) be the Weil differential of F/K (resp. F̄ /K̄) associated
with y. The restriction of a Weil differential ω̄ ∈ ΩF̄ to the space AF ⊆ AF̄
is denoted by ω̄ |AF .

Proposition 4.3.6. With the above notation we have

δ̄(x) |AF = δ(x) . (4.30)

Proof. Recall that δ(x) = CotrF/K(x) (η) and δ̄(x) = CotrF̄/K̄(x) (η̄) with specific
Weil differentials η ∈ ΩK(x) (resp. η̄ ∈ ΩK̄(x) ), cf. Definition 4.3.1. The first
step in the proof of Proposition 4.3.6 is the following assertion:

η̄ |AK(x) = η . (4.31)

In fact, the map η̄ |AK(x) : AK(x) → K̄ is K-linear, and η̄ vanishes on

AK(x) (−2P∞ ) + K(x) (as usual, P∞ is the pole of x in K(x)). Consider the
adele γ := ιP∞ (x−1 ) ∈ AK(x) which is given by γP∞ := x−1 and γQ := 0 for
Q = P∞ . Then η̄(γ) = η(γ) = −1 by (4.22). Observe that

AK(x) = AK(x) (−2P∞ ) + K(x) + K · γ ,

since γ ∈ AK(x) (−2P∞ )+K(x) and dimK (AK(x) /AK(x) (−2P∞ )+K(x)) = 1,
cf. Theorem 1.5.4. So each adele β ∈ AK(x) can be written as β = β0 + cγ
with c ∈ K and η(β0 ) = η̄(β0 ) = 0, and we obtain η̄(β) = η̄(β0 ) + c · η̄(γ) =
−c = η(β). This proves (4.31).
In the second step we evaluate the local component δ(x)P (u) for a place
P ∈ IPF and u ∈ F . Let Q := P ∩ K(x) and let P1 , . . . , Pr ∈ IPF be the places
of F lying over Q, say P = P1 . Let P̄ij ∈ IPF̄ be the extensions of Pi in F̄
(1 ≤ i ≤ r; 1 ≤ j ≤ si ) and Q̄1 , . . . , Q̄s ∈ IPK̄(x) the extensions of Q in K̄(x).
There exists an element z ∈ F such that

δ(x)P (z − u) = 0 ,
δ̄(x)P̄1j (z − u) = 0 for 1 ≤ j ≤ s1 ,
δ(x)Pi (z) = 0 for 2 ≤ i ≤ r ,
δ̄(x)P̄ij (z) = 0 for 2 ≤ i ≤ r, 1 ≤ j ≤ si . (4.32)
 4.3 Differentials and Weil Differentials 177

This follows from the Approximation Theorem and the fact that the local
component ωR (y) of a Weil differential ω at a place R vanishes if vR (y) is
sufficiently large. We obtain

r
δ(x)P (u) = δ(x)P (z) = δ(x)Pi (z) by (4.32)
i=1
= ηQ (TrF/K(x) (z)) by (4.21)
s
= η̄Q̄l (TrF/K(x) (z)) by (4.31).
l=1

Noting that TrF/K(x) (z) = TrF̄/K̄(x) (z) (since [F : K(x)] = [F̄ : K̄(x)]) and
applying (4.21) once again, we get

s 
r 
si 
s1
η̄Q̄l (TrF/K(x) (z)) = δ̄(x)P̄ij (z) = δ̄(x)P̄1j (u) .
l=1 i=1 j=1 j=1

Hence 
δ(x)P (u) = δ̄(x)P̄ (u) (4.33)
P̄ |P

for all u ∈ F (where P̄ runs over all places of F̄ /K̄ lying over P ).
Finally we consider an arbitrary adele α = (αP )P ∈IPF of F/K. Then

δ(x)(α) = δ(x)P (αP ) by Proposition 1.7.2
P ∈IPF
 
= δ̄(x)P̄ (αP ) by (4.33)
P ∈IPF P̄ |P

= δ̄(x)(α) by Proposition 1.7.2.

As before we consider the constant field extension F̄ = F K̄ of F/K

with the algebraic closure K̄ of K. Let (ΔF̄ , d) ¯ be the differential module
of F̄ /K̄. There is an F -linear map μ : ΔF → ΔF̄ given by μ(z dx) = z dx ¯ (cf.
Proposition 4.1.8(d)) , and we can regard ΔF as a submodule of ΔF̄ via this
embedding μ. Since dy ¯ = dy for y ∈ F , we denote the derivation d¯ : F̄ → ΔF̄
by d as well (i.e., F̄ /K̄ has the differential module (ΔF̄ , d)).

Proof of Theorem 4.3.2 for an arbitrary constant field K. The idea is to reduce
the theorem to the case of an algebraically closed constant field by considering
the constant field extension F̄ = F K̄. For y ∈ F and a separating element
x ∈ F we have
dy dy dy
δ(y) = δ̄(y) |AF = ( · δ̄(x)) |AF = · δ̄(x) |AF = · δ(x)
dx dx dx
178 4 Differentials of Algebraic Function Fields

(here we have used Proposition 4.3.6 and Theorem 4.3.2(b) for F̄ /K̄). Hence
(b) is proved. Parts (a), (c) and (e) follow now precisely as in the case of an
algebraically closed constant field.
(d) A place P ∈ IPF of degree one has exactly one extension P̄ ∈ IPF̄ by
Corollary 3.6.5. Thus

(z · δ(x))P (u) = δ(x)P (zu) = δ̄(x)P̄ (zu) = resP̄ (zu dx) = resP (zu dx) .

(The last equation follows from the observation that the residue of a differ-
ential is defined by means of the power series expansion with respect to a
prime element; since P̄ | P is unramified, every prime t ∈ F is a P̄ -prime
element.)
Remark 4.3.7. (a) As a consequence of Theorem 4.3.2 we identify the dif-
ferential module ΔF with the module ΩF of Weil differentials of F/K. This
means that a differential ω = z dx ∈ ΔF is the same as the Weil differential
ω = z · δ(x) ∈ ΩF (where x ∈ F is separating and z ∈ F ). In other words,

ΔF = Ω F and z dx = z · δ(x) . (4.34)

(b) If 0 = ω ∈ ΔF and t is a prime element at the place P ∈ IPF , we can write

ω = z dt with z ∈ F , and we define

vP (ω) := vP (z) and (ω) := vP (ω)P . (4.35)
P ∈IPF

Theorem 4.3.2(e) shows that the definition of vP (ω) is independent of the

choice of the prime element, and it is compatible with the identification (4.34)
of ΔF and ΩF . Hence (ω) is just the divisor of the corresponding Weil differ-
ential ω as defined in Section 1.5.
(c) As an important special case of Theorem 3.4.6 we obtain the following
formula for the divisor of a differential ω = z dx = 0:

(z dx) = (z) + (dx) = (z) − 2(x)∞ + Diff(F/K(x)) . (4.36)

A particular case of this formula is

(dx) = −2(x)∞ + Diff(F/K(x)) . (4.37)

(d) Once again we consider the constant field extension F̄ = F K̄ of F/K with
the algebraic closure of K. We have identified the differential module ΔF with
a submodule of ΔF̄ , hence we obtain a corresponding embedding of ΩF into
ΩF̄ which is given by

ω = z · δ(x) −→ ω̄ := z · δ̄(x)

(we use the notation of Proposition 4.3.6). By Proposition 4.3.6, ω is the re-
striction of ω̄ to AF . This observation yields a formula for the local component
 4.4 Exercises 179

ωP of ω at a place P of degree f ≥ 1: by Corollary 3.6.5 there are exactly

f = deg P places P̄1 , . . . , P̄f ∈ IPF̄ lying over P , hence Theorem 4.3.2(d)
implies
f
ωP (u) = resP̄i (uω) (4.38)
i=1
for ω ∈ ΩF and u ∈ F . In particular we have

ωP (u) = resP (uω) , (4.39)

if deg P = 1.

4.4 Exercises
In all exercises below we assume that K is a perfect field and F/K is a function
field with full constant field K.
4.1. Let E/F be a finite extension field of F . Let P ∈ IPF and Q ∈ IPE such
that Q is an extension of P .
(i) Show that the completion F̂P of F at P can be considered in a natural
way as a subfield of ÊQ .
(ii) Show that F̂P = ÊQ if and only if e(Q|P ) = f (Q|P ) = 1.
4.2. Let P be a place of F/K of degree one, and let t ∈ F be a P -prime
element. Determine the P -adic power series expansions with respect to t for
the following elements of F :
1 d 1 
and , with 0 = r ∈ ZZ ,
1−t r dt 1 − tr
1+t 1−t
and .
1−t 1+t
4.3. (char K = 2) Consider the function field F = K(x, y) with defining
equation y 2 = x3 + x (see Example 3.7.6). The element x has a unique pole in
F , call it P , and then t := x/y is a P -prime element. Determine the P -adic
power series of the elements x and y with respect to t.
4.4. Let x ∈ F be a separating element of F/K. Show that there exists a
unique derivation δ : F → F of F/K such that δ(x) = x.
4.5. Let x, y ∈ F and c ∈ K. Show that the following conditions are equiva-
lent:
(a) dy = c dx.
(b) y = cx + z, where z ∈ F is a non-separating element of F/K (i.e., z ∈ K
if the characteristic of K is 0, and z = up with some element u ∈ F if
char K = p > 0).
180 4 Differentials of Algebraic Function Fields

4.6. (char K = 2) Consider the function field F = K(x, y), given by the
equation

2m+1
y2 = (x − ai ) ,
i=1

where m ≥ 0 and a1 , . . . , a2m+1 ∈ K are distinct elements of K.

(i) Determine the divisor of the differential dx/y and obtain a new proof that
the genus of F/K is g = m (cf. Example 3.7.6).
(ii) Show that the differentials xi dx/y with 0 ≤ i ≤ m − 1 are a basis of the
space ΩF (0) (the space of regular differentials of F/K).

4.7. Let x be a separating element of F/K. Show that

 dx 
resP = vP (x) ,
x
for every place P of degree one.

As pointed out in Remark 4.3.7, we identify the differential module ΔF and

the module of Weil differentials ΩF . Therefore it makes sense to define for a
divisor A ∈ Div(F ) the K-vector space

ΔF (A) := {ω ∈ ΔF | ω = 0 or (ω) ≥ A} ,
which corresponds to ΩF (A) under the identification of ΔF and ΩF . We define
further:
(1) A differential ω ∈ ΔF is called regular (or holomorphic, or a differential
of the first kind), if ω = 0 or (ω) ≥ 0.
(2) A differential ω ∈ ΔF is called exact, if ω = dx for some element x ∈ F .
(3) A differential ω ∈ ΔF is called residue-free, if resP (ω) = 0 for all rational
places P ∈ IPF .
(4) A differential ω ∈ ΔF is called a differential of the second kind, if for all
places P ∈ IPF there exists an element u ∈ F (depending on P ) such that
vP (ω − du) ≥ 0.
Then we consider the following subsets of ΔF :

(1)
ΔF : = ΔF (0) = {ω ∈ ΔF | ω is regular }
(ex)
ΔF : = {ω ∈ ΔF | ω is exact }
(rf)
ΔF : = {ω ∈ ΔF | ω is residue-free }
(2)
ΔF : = {ω ∈ ΔF | ω is of the second kind }
 4.4 Exercises 181

4.8. Prove:
(1) (ex) (rf) (2)
(i) ΔF , ΔF , ΔF and ΔF are K-subspaces of ΔF .
(ex) (2) (rf)
(ii) ΔF ⊆ ΔF ⊆ ΔF .
(2) (rf)
(iii) If char K = 0 and K is algebraically closed, then ΔF = ΔF .

4.9. (The result of this exercise will be used in Exercise 4.10.) Assume that
F/K has genus g > 0. Then there exists a non-special divisor B of degree g
such that B = P1 + . . . + Pg , with pairwise distinct places P1 , . . . , Pg .

4.10. (K algebraically closed and char K = 0) The aim of this exercise is to

show that
(2) (ex)
dimK (ΔF / ΔF ) = 2g ,
if F/K is a function field of genus g over an algebraically closed field K with
char K = 0.
(i) Prove the claim in the case g = 0.
Hence we will assume now that g > 0, and we fix a non-special divisor B =
P1 + . . . + Pg with distinct places P1 , . . . , Pg .
(ii) Show that for every ω ∈ ΔF there exists a unique differential ω ∗ ∈
(2)

ΩF (−2B) such that ω − ω ∗ is exact.

(iii) Show that the map f : ω → ω ∗ , where ω ∗ is defined as in (ii), is a
(2) (2) (ex)
K-linear surjective map from ΔF onto ΔF ∩ ΔF (−2B), with kernel ΔF .
Conclude that
(2) (ex) (2)
dimK (ΔF / ΔF ) = dimK (ΔF ∩ ΔF (−2B)) .

(iv) For i = 1, . . . , g we fix a Pi -prime element ti ∈ F . Then every ω ∈

(2)
ΔF ∩ ΔF (−2B) has a Pi -adic expansion
  (i) j 
ω = a−2 t−2
(i) (i)
i + a0 + aj ti dti
j≥1

(i)
with aj ∈ K, for 1 ≤ i ≤ g. Show that the map

(i) (i)
ω → (a−2 , a0 )i=1,...,g
(2)
defines an isomorphism of ΔF ∩ ΔF (−2B) onto K 2g .
(2) (ex)
(v) Conclude that dimK (ΔF / ΔF ) = 2g.

4.11. (char K = p > 0) Recall that there is a unique subfield M ⊆ F such

that F/M is purely inseparable of degree [F : M ] = p, namely M = F p . In
what follows, we fix a separating element x of F/K.
(i) Show that every element z ∈ F has a unique representation of the form
182 4 Differentials of Algebraic Function Fields


p−1
z= upi xi , with u0 , . . . , up−1 ∈ F .
i=0

(ii) If z ∈ F is given as above, then

dz  p
p−1
= i ui xi−1 .
dx i=1

(iii) For a derivation η of F/K and n ≥ 1 we define η n := η ◦ η ◦ . . . ◦ η

(n times). Show that δxp = 0.
(iv) Is δ p = 0 for every derivation δ ∈ DerF ?

4.12. (char K = p > 0) Fix a separating element x ∈ F ; then every differen-

tial ω ∈ ΔF has a unique representation of the form
 
ω = up0 + up1 x + . . . + upp−1 xp−1 dx

with u0 , . . . , up−1 ∈ F , cf. Exercise 4.11. We define the map C : ΔF → ΔF by

C(ω) := up−1 dx .

This map C is called the Cartier operator. Its definition seems to depend on
the element x. We will show in Exercise 4.13 that it is actually independent
of the choice of x.
Prove the following properties of C, for all ω, ω1 , ω2 ∈ ΔF and all z ∈ F :
(i) C(ω1 + ω2 ) = C(ω1 ) + C(ω2 ).
(ii) C(z p ω) = z C(ω).
(iii) C : ΔF → ΔF is surjective.
(iv) C(ω) = 0 if and only if ω is exact.
(v) If 0 = z ∈ F , then C(dz/z) = dz/z.
Hint. The proofs of (i) - (iv) are easy. In order to prove (v), one may proceed
as follows. Observe that C(dz/z) = dz/z if and only if C(z p−1 dz) = dz, by
(ii). Therefore consider the set

M := {z ∈ F | C(z p−1 dz) = dz}

and show:
(a) F p ⊆ M and x ∈ M .
(b) If 0 = z ∈ M , then z −1 ∈ M .
(c) If z ∈ M , then z + 1 ∈ M .
(d) If z1 , z2 ∈ M , then z1 z2 ∈ M and z1 + z2 ∈ M .
(e) Conclude that M = F , which finishes the proof of (v).
 4.4 Exercises 183

4.13. (char K = p > 0) Suppose that C ∗ : ΔF → ΔF is a map with the

following properties: for all ω, ω1 , ω2 ∈ ΔF and all z ∈ F ,
(i) C ∗ (ω1 + ω2 ) = C ∗ (ω1 ) + C ∗ (ω2 ).
(ii) C ∗ (z p ω) = z C ∗ (ω).
(iv) C ∗ (dz) = 0.
(v) If 0 = z ∈ F , then C ∗ (dz/z) = dz/z.
Show that C ∗ = C, where C is defined as in the previous exercise. This implies
in particular, that the definition of the Cartier operator does not depend on
the choice of the separating element x.

4.14. (char K = p > 0) Prove the following properties of the Cartier divisor.
For all ω ∈ ΔF and all places P ∈ IPF ,
(i) vP (ω) ≥ 0 =⇒ vP (C(ω)) ≥ 0.
(ii) vp (ω) = −1 =⇒ vP (C(ω)) = −1.
(iii) vP (ω) < −1 =⇒ vP (C(ω)) > vP (ω).
(1) (1)
(iv) C(ΔF ) ⊆ ΔF .
(v) resP (ω) = (resP (C(ω))p , if P is a place of F/K of degree one.

4.15. (K algebraically closed and char K = p > 0)

(2) (ex)
(i) Show that dimK (ΔF /ΔF ) = g.
(rf) (ex)
(ii) Show that dimK (ΔF /ΔF ) = ∞.
Compare these results with Exercise 4.10.
(ex)
Hint. Use the fact that ΔF is the kernel of the Cartier operator.

4.16. (char K = p > 0) A differential ω ∈ ΩF is called logarithmic, if ω =

dx/x for some element 0 = x ∈ F . We define

Λ := {ω ∈ ΩF | (ω) ≥ 0 and ω is logarithmic} .

Show:
(i) Λ is an additive subgroup of ΔF (0), i.e., we can consider Λ as a vector
space over the prime field IFp .
(ii) If ω1 , . . . , ωm ∈ Λ are linearly independent over IFp , then they are linearly
independent over K.
(iii) Λ is a finite group of order ps , with 0 ≤ s ≤ g (as usual, g denotes the
genus of F/K).

4.17. ( char K = p > 0) Show that a differential ω ∈ ΩF is logarithmic if and

only if C(ω) = ω.
184 4 Differentials of Algebraic Function Fields

4.18. ( char K = p > 0) Consider the divisor class group Cl(F ) of F/K and
define the subgroup

Cl(F )(p) := {[A] ∈ Cl(F ) | p [A] = 0} .

We want to define a map f : Cl(F )(p) → ΔF . For [A] ∈ Cl(F )(p) there exists
an element 0 = x ∈ F such that pA = (x), by definition of Cl(F )(p). Then we
set f ([A]) := dx/x.
Prove:
(i) f is a well-defined group homomorphism from Cl(F )(p) to Λ (with Λ as
in the previous exercise).
(ii) f is an isomorphism from Cl(F )(p) onto Λ. Conclude that Cl(F )(p) is a
finite group of order ps with 0 ≤ s ≤ g.
Remark. If K is algebraically closed, the number s above is called the p-rank
(or the Hasse-Witt rank) of F/K. The function field is called regular (resp.
singular) if s = g (resp. s < g); it is called supersingular if s = 0.
5
Algebraic Function Fields over Finite Constant
Fields

In the previous chapters we developed the theory of algebraic function fields

over an arbitrary perfect constant field K. We would now like to consider in
greater detail the case of a finite constant field. Observe that a finite field is
perfect, so that all results from Chapters 3 and 4 apply. We will mainly be
interested in the places of degree one of a function field over a finite field.
Their number is finite and can be estimated by the Hasse-Weil Bound (see
Theorem 5.2.3). This bound has many number-theoretical implications, and
it plays a crucial role in the applications of algebraic function fields to coding
theory, cf. Chapter 8 and 9.

Throughout this chapter, F denotes an algebraic function field of genus

g whose constant field is the finite field IFq .

5.1 The Zeta Function of a Function Field

As in Chapter 1,
Div(F ) denotes the divisor group of the function field F/IFq .
A divisor A = P ∈IPF aP P is positive if all aP ≥ 0; we write A ≥ 0.

Lemma 5.1.1. For every n ≥ 0 there exist only finitely many positive divisors
of degree n.

Proof. A positive divisor is a sum of prime divisors. Hence it is sufficient

to prove that the set S := {P ∈ IPF | deg P ≤ n} is finite. We choose an
element x ∈ F \IFq and consider the set S0 := {P0 ∈ IPIFq (x) | deg P0 ≤ n}.
Obviously P ∩ IFq (x) ∈ S0 for all P ∈ S, and each P0 ∈ S0 has only finitely
many extensions in F . Therefore we have only to show that S0 is finite. Since
the places of IFq (x) (except the pole of x) correspond to irreducible monic
polynomials p(x) ∈ IFq [x] of the same degree (cf. Section 1.2), the finiteness
of S0 follows readily.

H. Stichtenoth, Algebraic Function Fields and Codes, 185

Graduate Texts in Mathematics 254,

c Springer-Verlag Berlin Heidelberg 2009
186 5 Algebraic Function Fields over Finite Constant Fields

Let us recall some notation from previous chapters: Princ(F ) denotes the
subgroup of Div(F ) consisting of all principal divisors (x) = P ∈IPF vP (x) · P
(with 0 = x ∈ F ). The factor group Cl(F ) = Div(F )/Princ(F ) is called
the divisor class group of F/IFq . Two divisors A, B ∈ Div(F ) are equivalent
(written A ∼ B) if B = A + (x) for some principal divisor (x) ∈ Princ(F ).
The class of A in the divisor class group Cl(F ) is denoted by [A], hence
A ∼ B ⇐⇒ A ∈ [B] ⇐⇒ [A] = [B]. Equivalent divisors have the same
degree and the same dimension, so the integers

deg [A] := deg A and ([A]) := (A)

are well-defined for a divisor class [A] ∈ Cl(F ).

Definition 5.1.2. The set

Div0 (F ) := {A ∈ Div(F ) | deg A = 0} ,

which is obviously a subgroup of Div(F ), is called the group of divisors of

degree zero, and

Cl0 (F ) := {[A] ∈ Cl(F ) | deg [A] = 0}

is called the group of divisor classes of degree zero.

Proposition 5.1.3. Cl0 (F ) is a finite group. Its order h = hF is called the

class number of F/IFq ; i.e.,

h := hF := ord Cl0 (F ) .

Proof. Choose a divisor B ∈ Div(F ) of degree ≥ g, say n := deg B, and

consider the set of divisor classes

Cln (F ) := {[C] ∈ Cl(F ) | deg [C] = n} .

The map 
Cl0 (F ) −→ Cln (F ) ,
[A] −→ [A + B]
is bijective (this is trivial), so we only have to verify that Cln (F ) is finite. We
claim:

for each divisor class [C] ∈ Cln (F ) there

exists a divisor A ∈ [C] with A ≥ 0 . (5.1)
 5.1 The Zeta Function of a Function Field 187

In fact, since deg C = n ≥ g, we have

([C]) ≥ n + 1 − g ≥ 1 (5.2)
by the Riemann-Roch Theorem, and (5.1) follows from (5.2) and Remark
1.4.5(b). There are only finitely many divisors A ≥ 0 of degree n (by Lemma
5.1.1), so (5.1) implies the finiteness of Cln (F ).

We define the integer ∂ > 0 by

∂ := min {deg A | A ∈ Div(F ) and deg A > 0} . (5.3)
The image of the degree mapping deg : Div(F ) → ZZ is the subgroup of ZZ
generated by ∂, and the degree of each divisor of F/IFq is a multiple of ∂.
In what follows, we would like to study the numbers
 
An :=  {A ∈ Div(F ) | A ≥ 0 and deg A = n}  . (5.4)
For instance, A0 = 1, and A1 is the number of places P ∈ IPF of degree one.

Lemma 5.1.4. (a) An = 0 if ∂  n.

(b) For a fixed divisor class [C] ∈ Cl(F ) we have
  1  
 {A ∈ [C] | A ≥ 0}  = q ([C])
−1 .
q−1

(c) For each integer n > 2g − 2 with ∂|n we have

h  n+1−g 
An = q −1 .
q−1
Proof. (a) is trivial.
(b) The conditions A ∈ [C] and A ≥ 0 are equivalent to
A = (x) + C for some x ∈ F with (x) ≥ −C ;
i.e., x ∈ L (C)\{0}. There exist exactly q ([C]) − 1 elements x ∈ L (C)\{0},
and two of them yield the same divisor if and only if they differ by a constant
factor 0 = α ∈ IFq . This proves (b).
(c) There are h = hF divisor classes of degree n, say [C1 ], . . . , [Ch ]. By (b)
and the Riemann-Roch Theorem,
     
 {A ∈ [Cj ]|A ≥ 0}  = 1 q (Cj ) − 1 = 1 q n+1−g − 1 .
q−1 q−1
Each divisor of degree n lies in exactly one of the divisor classes [C1 ], . . . , [Ch ],
hence
h
   
An =  {A ∈ [Cj ] ; A ≥ 0}  = h q n+1−g − 1 .
j=1
q−1
188 5 Algebraic Function Fields over Finite Constant Fields

Definition 5.1.5. The power series

∞

Z(t) := ZF (t) := An tn ∈ C[[t]]
n=0

(where An is defined by (5.4)) is called the Zeta function of F/IFq .

Observe that we regard t here as a complex variable, and Z(t) is a power

series over the field of complex numbers (rather than the P -adic power series
that we considered in Chapter 4). We shall show now that this power series
converges in a neighbourhood of 0.
∞
Proposition 5.1.6. The power series Z(t) = n=0 An tn is convergent for
|t| < q −1 . More precisely, we have for |t| < q −1 :
(a) If F/IFq has genus g = 0 then
 
1 q 1
Z(t) = − .
q − 1 1 − (qt)∂ 1 − t∂
(b) If g ≥ 1, then Z(t) = F (t) + G(t) with
1 
F (t) = q ([C])
· tdeg[C] ,
q−1
0≤deg[C]≤2g−2

(where [C] runs over all divisor classes [C] ∈ Cl(F ) with 0 ≤ deg[C] ≤ 2g − 2)
and  
h 1 1
G(t) = q 1−g (qt)2g−2+∂ − .
q−1 1 − (qt)∂ 1 − t∂
Proof. (a) g = 0. To begin with, we show that a function field of genus zero
has class number h = 1; i.e., every divisor A of degree 0 is principal. This
fact follows easily from the Riemann-Roch Theorem: as 0 > 2g − 2 , we have
(A) = deg A + 1 − g = 1, and we can therefore find an element x = 0 with
(x) ≥ −A. Both divisors are of degree 0, hence A = −(x) = (x−1 ) is principal.
Now we apply Lemma 5.1.4 and obtain
∞
 ∞

An tn = A∂n t∂n
n=0 n=0
∞
1
= (q ∂n+1 − 1)t∂n
n=0
q − 1
 ∞ ∞

1  
= q (qt) −
∂n
t ∂n
q−1 n=0 n=0
 
1 q 1
= −
q − 1 1 − (qt)∂ 1 − t∂
for |qt| < 1.
(b) For g ≥ 1 the calculation is quite similar. We obtain
 5.1 The Zeta Function of a Function Field 189
∞
   
 q − 1 deg [C]
([C])
An tn = {A ∈ [C] ; A ≥ 0} · tdeg [C] = ·t
n=0
q−1
deg[C]≥0 deg [C]≥0

1  1 
= q ([C])
· tdeg [C] + q deg [C]+1−g · tdeg [C]
q−1 q−1
0≤deg [C]≤2g−2 deg [C]>2g−2

1 
− tdeg [C] = F (t) + G(t) ,
q−1
deg [C]≥0

with 
1
F (t) = q ([C])
· tdeg [C]
q−1
0≤deg [C]≤2g−2

and
∞
 ∞

(q − 1)G(t) = hq n∂+1−g · tn∂ − htn∂
n=((2g−2)/∂)+1 n=0
1 1
= hq 1−g (qt)2g−2+∂ −h .
1 − (qt)∂ 1 − t∂

Corollary 5.1.7. Z(t) can be extended to a rational function on C; it has a

simple pole at t = 1.

Proof. Obvious, since 1/(1 − t∂ ) has a simple pole at t = 1.

In order to study the behavior of the Zeta function of F/IFq under finite
constant field extensions, it is convenient to have a second representation
∞
of Z(t) as an infinite product. Recall that an infinite product i=1 (1 + ai )
(with complexn numbers ai = −1) is said to be convergent with limit a ∈ C
if lim
∞ n→∞ i=1 (1 + ai ) = a = 0. The product is called absolutely convergent
if |a
i=1 i | < ∞. From analysis it is well-known that absolute convergence
implies convergence of the product, and the limit of an absolutely convergent
product
∞ is independent of the order of the factors.Moreover, if the product
∞
i=1 (1 + ai ) = a isabsolutely convergent, then i=1 (1 + ai )−1 converges
∞ −1 −1
absolutely, too, and i=1 (1 + ai ) = a .

Proposition 5.1.8 (Euler Product). For |t| < q −1 the Zeta function can
be represented as an absolutely convergent product

Z(t) = (1 − tdeg P )−1 . (5.5)
P ∈IPF

In particular Z(t) = 0 for |t| < q −1 .

190 5 Algebraic Function Fields over Finite Constant Fields

−1
Proof.
 ∞ side ofn (5.5) converges absolutely for |t| < q , since
The right hand
P ∈IPF |t|deg P
≤ A
n=0 n |t| < ∞ by Proposition 5.1.6. Each factor of (5.5)
can be written as a geometric series, and we obtain
 ∞
 
(1 − tdeg P )−1 = tdeg (nP )
P ∈IPF P ∈IPF n=0
 ∞

= tdegA = An tn = Z(t) .
A∈Div(F );A≥0 n=0

In the following we fix an algebraic closure IF¯q of IFq and consider the

constant field extension F̄ = F IF¯q of F/IFq . For each r ≥ 1 there exists
exactly one extension IFqr /IFq of degree r with IFqr ⊆ IF¯q , and we set

Fr := F IFqr ⊆ F̄ .

Lemma 5.1.9. (a) Fr /F is a cyclic extension of degree r (i.e., Fr /F is Ga-

lois with a cyclic Galois group of order r). The Galois group Gal(Fr /F ) is
generated by the Frobenius automorphism σ which acts on IFqr by σ(α) = αq .
(b) IFqr is the full constant field of Fr .
(c) Fr /IFqr has the same genus as F/IFq .
(d) Let P ∈ IPF be a place of degree m. Then ConFr /F (P ) = P1 + . . . + Pd
with d := gcd (m, r) pairwise distinct places Pi ∈ IPFr and deg Pi = m/d.

Proof. (a) It is well-known that IFqr /IFq is cyclic of degree r, and its Galois
group is generated by the Frobenius map α → αq . Since [Fr : F ] = [IFqr : IFq ]
by Lemma 3.6.2, assertion (a) follows immediately.
(b) and (c) See Proposition 3.6.1 and Theorem 3.6.3.
(d) P is unramified in Fr /F , cf. Theorem 3.6.3. Consider some place P  ∈ IPFr
lying over P . The residue class field of P  is the compositum of IFqr with
the residue class field FP of P , by Theorem 3.6.3(g). Set l := lcm(m, r). As
FP = IFqm , this compositum is

IFqm · IFqr = IFql .

Therefore
deg P  = [IFql : IFqr ] = m/d .
Since deg (ConFr /F (P )) = deg P = m (cf. Theorem 3.6.3(c)), we conclude
that ConFr /F (P ) = P1 + . . . + Pd with places Pi of degree m/d.
 5.1 The Zeta Function of a Function Field 191

In the proof of our next proposition we need a simple polynomial identity:

if m ≥ 1 and r ≥ 1 are integers and d = gcd (m, r) then

(X r/d − 1)d = (X − ζ m ) , (5.6)
ζ r =1

where ζ runs over all r-th roots of unity in C. In fact, both sides of (5.6) are
monic polynomials of the same degree, and each (r/d)-th root of unity is a
d-fold root of them. Hence the polynomials are equal. We substitute X = t−m
in (5.6), multiply by tmr and obtain

(1 − tmr/d )d = (1 − (ζt)m ) . (5.7)
ζ r =1

Proposition 5.1.10. Let Z(t) (resp. Zr (t)) denote the Zeta function of F
(resp. of Fr = F IFqr ). Then

Zr (tr ) = Z(ζt) (5.8)
ζ r =1

for all t ∈ C (ζ runs over the r-th roots of unity).

Proof. It is sufficient to prove (5.8) for |t| < q −1 . In this region the Euler
product representation yields
  
Zr (tr ) = (1 − tr·deg P )−1 . (5.9)
P ∈IPF P  |P

For a fixed place P ∈ IPF we set m := deg P and d := gcd (r, m); then
 
(1 − tr·deg P ) = (1 − trm/d )d
P  |P
 
= (1 − (ζt)m ) = (1 − (ζt)deg P ) ,
ζ r =1 ζ r =1

by (5.7) and Lemma 5.1.9. Now we obtain from (5.9)

  
Zr (tr ) = (1 − (ζt)deg P )−1 = Z(ζt) .
ζ r =1 P ∈IPF ζ r =1

Corollary 5.1.11 (F.K. Schmidt). ∂=1.

∂
Proof. For ζ = 1 we have
 
Z(ζt) = (1 − (ζt)deg P )−1 = (1 − tdeg P )−1 = Z(t) ,
P ∈IPF P ∈IPF

since ∂ divides the degree of P for every P ∈ IPF . Therefore Z∂ (t∂ ) = Z(t)∂
by Proposition 5.1.10. The rational function Z∂ (t∂ ) has a simple pole at t = 1,
by Corollary 5.1.7, and Z(t)∂ has a pole of order ∂ at t = 1. Hence ∂ = 1.
192 5 Algebraic Function Fields over Finite Constant Fields

Corollary 5.1.12. (a) Every function field F/IFq of genus 0 is rational, and
its Zeta function is
1
Z(t) = .
(1 − t)(1 − qt)
(b) If F/IFq has genus g ≥ 1, its Zeta function can be written in the form
Z(t) = F (t) + G(t) with
1 
F (t) = q ([C])
· tdeg [C]
q−1
0≤deg [C]≤2g−2

and  
h 1 1
G(t) = g 2g−1
q t − .
q−1 1 − qt 1 − t
Proof. A function field of genus 0 having a divisor of degree 1 is rational, cf.
Proposition 1.6.3. The remaining assertions follow from Proposition 5.1.6 and
∂ = 1.

Proposition 5.1.13 (Functional Equation of the Zeta Function). The

Zeta function of F/IFq satisfies the functional equation
1
Z(t) = q g−1 t2g−2 Z .
qt

Proof. For g = 0 this is obvious from Corollary 5.1.12(a). For g ≥ 1 we write

Z(t) = F (t) + G(t) as in Corollary 5.1.12(b). Let W be a canonical divisor of
F ; then

(q − 1)F (t) = q ([C]) · tdeg [C]
0≤deg [C] ≤2g−2

([W −C])
= q deg [C]+1−g+ · tdeg [C]
0≤deg [C]≤2g−2

([W −C])
= q g−1 t2g−2 q deg [C]−(2g−2)+ · tdeg [C]−(2g−2)
0≤deg [C] ≤2g−2
  1 deg [W −C]
([W −C])
= q g−1 t2g−2 q ·
qt
0≤deg [C] ≤2g−2
1
= q g−1 t2g−2 (q − 1)F ( ). (5.10)
qt

We have used that deg [W ] = 2g − 2 and, if [C] runs over all divisor classes
with 0 ≤ deg [C] ≤ 2g − 2, so does [W − C]. For the function G(t) we obtain
 5.1 The Zeta Function of a Function Field 193
⎛ ⎞
1 h g−1 2g−2 ⎜
⎜q g ( 1 )2g−1 1 1 ⎟
⎟
q g−1 t2g−2 G( ) = q t ⎝ qt −
qt q−1 1 1 ⎠
1−q 1−
qt qt
⎛ ⎞
h ⎜⎜1 1 q g t2g−1 ⎟⎟ = G(t) .
= − (5.11)
q −1 ⎝t 1 1 ⎠
1− qt(1 − )
t qt
Adding (5.10) and (5.11) yields the functional equation for Z(t).

Definition 5.1.14. The polynomial L(t) := LF (t) := (1 − t)(1 − qt)Z(t) is

called the L-polynomial of F/IFq .
By Corollary 5.1.12 it is obvious that L(t) is a polynomial of degree ≤ 2g.
Observe that L(t) contains all information about the numbers An (n ≥ 0)
since
∞
L(t) = (1 − t)(1 − qt) An tn . (5.12)
n=0

Theorem 5.1.15. (a) L(t) ∈ ZZ[t] and deg L(t) = 2g.

(b) (Functional Equation) L(t) = q g t2g L(1/qt).
(c) L(1) = h, the class number of F/IFq .
(d) We write L(t) = a0 + a1 t + · · · + a2g t2g . Then the following hold:
(1) a0 = 1 and a2g = q g .
(2) a2g−i = q g−i ai for 0 ≤ i ≤ g.
(3) a1 = N − (q + 1) where N is the number of places P ∈ IPF of degree
one.
(e) L(t) factors in C[t] in the form

2g
L(t) = (1 − αi t) . (5.13)
i=1

The complex numbers α1 , . . . , α2g are algebraic integers, and they can be ar-
ranged in such a way that αi αg+i = q holds for i = 1, . . . , g. (We note that
a complex number α is called an algebraic integer if it satisfies an equation
αm + cm−1 αm−1 + · · · + c1 α + c0 = 0 with coefficients ci ∈ ZZ.)
(f ) If Lr (t) := (1 − t)(1 − q r t)Zr (t) denotes the L-polynomial of the constant
field extension Fr = F IFqr , then

2g
Lr (t) = (1 − αir t) ,
i=1

where the αi are given by (5.13).

194 5 Algebraic Function Fields over Finite Constant Fields

Proof. All assertions are trivial for g = 0, hence we can assume from now on
that g ≥ 1.
(a) We have already remarked that L(t) is a polynomial of degree ≤ 2g. In (d)
we shall prove that its leading coefficient is q g , so deg L(t) = 2g. The assertion
L(t) ∈ ZZ[t] follows from (5.12) by comparing coefficients.
(b) is nothing but the functional equation for the Zeta function, see Proposi-
tion 5.1.13.
(c) With the notation of Corollary 5.1.12(b) we have

h  g 2g−1 
L(t) = (1 − t)(1 − qt)F (t) + q t (1 − t) − (1 − qt) .
q−1

Hence L(1) = h.
(d) Write L(t) = a0 + a1 t + · · · + a2g t2g . The functional equation (b) yields

1 a2g a2g−1
L(t) = q g t2g L( ) = g + g−1 t + · · · + q g a0 t2g .
qt q q

Therefore a2g−i = q g−i ai for i = 0, . . . , g, and (2) is proved. Comparing the

coefficients of t0 and t1 in (5.12) shows that a0 = A0 and a1 = A1 − (q + 1)A0 .
Since A0 = 1 and A1 = N (this is trivial by the definition of An , see (5.4)),
we obtain a0 = 1 and a1 = N − (q + 1). Finally a2g = q g a0 = q g by (2).
(e) We consider the reciprocal polynomial
1
L⊥ (t) := t2g L( ) = a0 t2g +a1 t2g−1 +· · ·+a2g = t2g +a1 t2g−1 +· · ·+q g . (5.14)
t
L⊥ (t) is a monic polynomial with coefficients in ZZ, so its roots are algebraic
integers. We write


2g
L⊥ (t) = (t − αi ) with αi ∈ C,
i=1

therefore
1 2g
⊥
2g
L(t) = t L ( ) = (1 − αi t) .
t i=1

Observe that L(αi−1 ) = 0 for i = 1, ..., 2g, and


2g
αi = q g .
i=1

Substituting t = qu and using the functional equation (b) we obtain

 5.1 The Zeta Function of a Function Field 195


2g
1
(t − αi ) = L⊥ (t) = t2g L( )
i=1
t

1 2g
= q 2g u2g L( ) = q g L(u) = q g · (1 − αj u)
qu j=1


2g
αj 2g
αj 
2g
q
=q · g
(1 − t) = q ·
g
· (t − )
j=1
q j=1
q j=1
αj


2g
q
= (t − ).
j=1
αj

So we can arrange the roots of L⊥ (t) as

q q 1/2
α1 , , . . . , αk , , q , . . . , q 1/2 , −q 1/2 , . . . , −q 1/2 ,
α1 αk
where q 1/2 occurs m times and −q 1/2 occurs n times. By (5.14),
q q
α1 · · . . . · αk · · (q 1/2 )m · (−q 1/2 )n = q g .
α1 αk
Therefore n is even. Since n + m + 2k = 2g, m is also even, and we can
rearrange α1 , . . . , α2g such that αi αg+i = q holds for i = 1, . . . , g.
(f) We use Proposition 5.1.10 and obtain

Lr (tr ) = (1 − tr )(1 − q r tr )Zr (tr ) = (1 − tr )(1 − q r tr ) Z(ζt)
ζ r =1
 L(ζt) 
= (1 − tr )(1 − q r tr ) = L(ζt)
(1 − ζt)(1 − qζt)
ζ r =1 ζ r =1


2g  
2g
= (1 − αi ζt) = (1 − αir tr ) .
i=1 ζ r =1 i=1
2g
Hence Lr (t) = i=1 (1 − αir t).

The above theorem shows that the number

 
N (F ) := N =  {P ∈ IPF ; deg P = 1}  (5.15)

can easily be calculated if the L-polynomial L(t) of F/IFq is known. More

generally we consider for r ≥ 1 the number
 
Nr := N (Fr ) =  {P ∈ IPFr ; deg P = 1}  , (5.16)

where Fr = F IFqr is the constant field extension of F/IFq of degree r. In

Section 5.2 the following result will play an essential role.
196 5 Algebraic Function Fields over Finite Constant Fields

Corollary 5.1.16. For all r ≥ 1,


2g
Nr = q r + 1 − αir ,
i=1

where α1 , . . . , α2g ∈ C are the reciprocals of the roots of L(t). In particular,

since N1 = N (F ), we have


2g
N (F ) = q + 1 − αi .
i=1

Proof. By Theorem 5.1.15(d), Nr − (q r + 1) is the coefficient of t in the L-

2g
polynomial Lr (t). On the other hand, since Lr (t) = i=1 (1 − αir t), this coef-
2g
ficient is − i=1 αir .

Conversely, if the numbers Nr are known for sufficiently many r, one can
calculate the coefficients of L(t) as follows.
2g
Corollary 5.1.17. Let L(t) = i=0 ai ti be the L-polynomial of F/IFq , and
Sr := Nr − (q r + 1). Then we have:
∞
(a) L (t)/L(t) = r=1 Sr tr−1 .
(b) a0 = 1 and
iai = Si a0 + Si−1 a1 + . . . + S1 ai−1 (5.17)
for i = 1, . . . , g.
Hence, given N1 , . . . , Ng we can determine L(t) from (5.17) and the equations
a2g−i = q g−i ai (for i = 0, . . . , g).
2g
Proof. (a) Write L(t) = i=1 (1 − αi t) as in (5.13). Then


2g 2g ∞
−αi
L (t)/L(t) = = (−αi ) · (αi t)r
i=1
(1 − α i t) i=1 r=0
∞
 2g  ∞
  
= −αir tr−1 = Sr tr−1 ,
r=1 i=1 r=1

by Corollary 5.1.16 and the definition of Sr .

(b) We know that a0 = 1, by Theorem 5.1.15. From (a) it follows that
∞

a1 + 2a2 t + . . . + 2ga2g t2g−1 = (a0 + a1 t + . . . + a2g t2g ) · Sr tr−1 .
r=1

Comparing the coefficients of t0 , t1 , . . . , tg−1 yields (5.17).

 5.2 The Hasse-Weil Theorem 197

5.2 The Hasse-Weil Theorem

We retain all notation of Section 5.1. Thus F/IFq is a function field of genus
g(F ) = g over the finite field IFq ,

ZF (t) = LF (t)/(1 − t)(1 − qt) is its Zeta function,

α1 , . . . , α2g are the reciprocals of the roots of LF (t),
 
N (F ) =  {P ∈ IPF ; deg P = 1}  ,
Fr = F IFqr is the constant field extension of degree r, and
Nr = N (Fr ).

The main result of this section is

Theorem 5.2.1 (Hasse-Weil). The reciprocals of the roots of LF (t) satisfy

|αi | = q 1/2 for i = 1, . . . , 2g .

Remark 5.2.2. The Hasse-Weil Theorem is often referred to as the Riemann

Hypothesis for Function Fields. Let us briefly explain this notation. One can
regard the Zeta function ZF (t) of a function field F/IFq as an analogue of the
classical Riemann ζ-function
∞

ζ(s) := n−s (5.18)
n=1

(where s ∈ C and Re(s) > 1) in the following manner. Define the absolute
norm of a divisor A ∈ Div(F ) by

N (A) := q deg A
.

For instance, the absolute norm N (P ) of a prime divisor P ∈ IPF is the

cardinality of its residue class field FP . Then the function

ζF (s) := ZF (q −s )

can be written as
∞
 
ζF (s) = An q −sn = N (A)−s ,
n=0 A∈Div(F ),A≥0

which is the appropriate analogue to (5.18). It is well-known from number

theory that the Riemann ζ-function (5.18) has an analytic continuation as a
meromorphic function on C. The classical Riemann Hypothesis states that –
besides the so-called trivial zeros s = −2, −4, −6, . . . – all zeros of ζ(s) lie on
the line Re(s) = 1/2.
198 5 Algebraic Function Fields over Finite Constant Fields

In the function field case, the Hasse-Weil Theorem states that

ζF (s) = 0 ⇒ ZF (q −s ) = 0 ⇒ |q −s | = q −1/2 .

Since |q −s | = q −Re(s) , this means that

ζF (s) = 0 ⇒ Re(s) = 1/2 .

Therefore Theorem 5.2.1 can be viewed as an analogue of the classical Rie-

mann Hypothesis.
Before proving it, we draw an important conclusion from the Hasse-Weil
Theorem.

Theorem 5.2.3 (Hasse-Weil Bound). The number N = N (F ) of places

of F/IFq of degree one satisfies the inequality

|N − (q + 1)| ≤ 2gq 1/2 .

Proof. Corollary 5.1.16 yields


2g
N − (q + 1) = − αi .
i=1

Hence the Hasse-Weil Bound is an immediate consequence of the Hasse-Weil

Theorem.

Note that Theorem 5.2.3, applied to the function field Fr /IFqr , gives

|Nr − (q r + 1)| ≤ 2gq r/2 (5.19)

for all r ≥ 1.
Our proof of the Hasse-Weil Theorem is due to E. Bombieri. The proof is
divided into several steps. The first step is almost trivial.

Lemma 5.2.4. Let m ≥ 1. Then the Hasse-Weil Theorem holds for F/IFq if
and only if it holds for the constant field extension Fm /IFqm .

Proof. The reciprocals of the roots of LF (t) are α1 , . . . , α2g . By Theorem

5.1.15(f), the reciprocals of the roots of Lm (t) are α1m , . . . , α2g
m
(as in Theorem
5.1.15, we denote by Lm (t) the L-polynomial of Fm ). Our lemma follows
immediately since |αi | = q 1/2 ⇐⇒ |αim | = (q m )1/2 .

The next step reduces the proof of the Hasse-Weil Theorem to an assertion
that is closely related to (5.19).
 5.2 The Hasse-Weil Theorem 199

Lemma 5.2.5. Assume there is a constant c ∈ IR such that for all r ≥ 1,

|Nr − (q r + 1) |≤ cq r/2 . (5.20)

Then the Hasse-Weil Theorem holds for F/IFq .

2g
Proof. Corollary 5.1.16 states that Nr − (q r + 1) = − i=1 αir , hence (5.20)
yields
 2g 
 r 
 α 
i  ≤ cq
r/2
(5.21)

i=1

for all r ≥ 1. Consider the meromorphic function


2g
αi t
H(t) := . (5.22)
i=1
1 − αi t

Let  := min {|αi−1 |; 1 ≤ i ≤ 2g}. The convergence radius of the power series
expansion of H(t) around t = 0 is precisely  (since α1−1 , . . . , α2g
−1
are the only
singularities of H(t)). On the other hand we obtain for |t| < 

 ∞
2g  ∞ 
 2g 
H(t) = (αi t)r = αir tr .
i=1 r=1 r=1 i=1

By (5.21) this series converges for |t| < q −1/2 , hence q −1/2 ≤ . This implies
2g
q 1/2 ≥ |αi | for i = 1, . . . , 2g. Since i=1 αi = q g (by Theorem 5.1.15(e)), we
conclude that |αi | = q 1/2 .

Observe that the inequalities (5.20) are equivalent to an upper bound and
a lower bound for Nr : there exist constants c1 > 0 and c2 > 0 such that

Nr ≤ q r + 1 + c1 q r/2 (5.23)

and
Nr ≥ q r + 1 − c2 q r/2 (5.24)
for all r ≥ 1. By Lemma 5.2.4 the Hasse-Weil Theorem holds for F/IFq if it
holds for some constant field extension of F . Therefore it is sufficient to prove
(5.23) and (5.24) under additional assumptions which can be realized in an
appropriate finite constant field extension.

Proposition 5.2.6. Suppose that F/IFq satisfies the following assumptions:

(1) q is a square, and (2) q > (g + 1)4 .

Then the number N = N (F ) of places of F/IFq of degree one can be estimated

by
N < (q + 1) + (2g + 1)q 1/2 .
200 5 Algebraic Function Fields over Finite Constant Fields

Proof. We can assume that there exists a place Q ∈ IPF of degree one (other-
wise N = 0, and the proposition is trivial). Set

q0 := q 1/2 , m := q0 − 1 and n := 2g + q0 .

One checks easily that

r := q − 1 + (2g + 1)q 1/2 = m + nq0 . (5.25)

Let T := {i | 0 ≤ i ≤ m, and i is a pole number of Q }. (Recall that i is

called a pole number of Q if there exists an element x ∈ F with pole divisor
(x)∞ = iQ, cf. Definition 1.6.7.) For every i ∈ T choose an element ui ∈ F
with pole divisor iQ. Then the set {ui | i ∈ T } is a basis of L (mQ). We
consider the space

L := L (mQ) · L (nQ)q0 ⊆ L (rQ) .


(By definition, L consists of all finite sums xν yνq0 with xν ∈ L (mQ) and
yν ∈ L (nQ); obviously L is a vector space over IFq , and L ⊆ L (rQ) by
(5.25).) Our aim is to construct an element 0 = x ∈ L such that

x(P ) = 0 for all P ∈ IPF with deg P = 1 and P = Q . (5.26)

Suppose for a moment that we have found such an element x. Then all places
of degree one (except Q) are zeros of x, and the zero divisor (x)0 has degree

deg (x)0 ≥ N − 1 .

As x ∈ L ⊆ L (rQ),

deg (x)0 = deg (x)∞ ≤ r = q − 1 + (2g + 1)q 1/2 .

Combining these inequalities, we get N ≤ q + (2g + 1)q 1/2 , which proves the
proposition.
Claim 1. Every element y ∈ L can be written uniquely in the form

y= ui ziq0 with zi ∈ L (nQ) , (5.27)
i∈T

where {ui | i ∈ T } is the above-mentioned basis of L (mQ).

The existence of a representation (5.27) follows almost immediately from
the definition of L . In order to prove uniqueness, we assume that there is an
equation 
ui xqi 0 = 0 (5.28)
i∈T

with xi ∈ L (nQ), not all xi = 0. For each index i ∈ T with xi = 0 we have

vQ (ui xqi 0 ) ≡ vQ (ui ) ≡ −i mod q0 .

 5.2 The Hasse-Weil Theorem 201

Since m = q0 −1, the numbers i ∈ T are pairwise distinct modulo q0 . Therefore

the Strict Triangle Inequality yields
 
vQ ui xqi 0 = min {vQ (ui xqi 0 ) | i ∈ T } = ∞ .
i∈T

This contradiction to (5.28) proves Claim 1.

Next we consider the mapping λ : L → L ((q0 m + n)Q) given by
   q
λ ui ziq0 := u i 0 zi
i∈T i∈T

(with zi ∈ L (nQ)). By Claim 1 this map is well-defined. Observe that λ

is not IFq -linear, but it is a homomorphism of the additive group of L into
L ((q0 m + n)Q).
Claim 2. The kernel of λ is not {0}.
As λ is a homomorphism from L into L ((q0 m + n)Q), it is sufficient to
show that
dim L > dim L ((q0 m + n)Q) (5.29)
(where dim denotes the dimension as a vector space over IFq ). We have, by
Claim 1 and the Riemann-Roch Theorem,

dim L = (mQ) · (nQ) ≥ (m + 1 − g)(n + 1 − g) .

On the other hand, since

q0 m + n = q0 (q0 − 1) + (2g + q0 ) = 2g + q ,

we obtain

dim L ((q0 m + n)Q) = (2g + q) + 1 − g = g + q + 1 .

Hence (5.29) follows if we can prove that

(m + 1 − g)(n + 1 − g) > g + q + 1 . (5.30)

Consider the following equivalences:

(m + 1 − g)(n + 1 − g) > g + q + 1
⇐⇒ (q0 − g)(2g + q0 + 1 − g) > g + q + 1
⇐⇒ q − g 2 + q0 − g > g + q + 1
⇐⇒ q0 > g 2 + 2g + 1 = (g + 1)2
⇐⇒ q > (g + 1)4 .

As we assumed that q > (g + 1)4 (see assumption (2) of Proposition 5.2.6),

(5.30) is established.
202 5 Algebraic Function Fields over Finite Constant Fields

Claim 3. Let 0 = x ∈ L be an element in the kernel of λ and let P = Q

be a place of degree one. Then x(P ) = 0.
Note that y(P ) = ∞ for all y ∈ L because Q is the only pole of y.
)q = y(P ). Now
Moreover, since IFq is the residue class field of P , we have y(P
we consider an element x ∈ L with λ(x) = 0. We write x = i∈T ui ziq0 and
obtain
 q0
x(P )q0 = ui (P ) · zi (P )q0
i∈T

= uqi 0 (P ) · zi (P )q
i∈T
 
= uqi 0 zi (P ) = λ(x)(P ) = 0 .
i∈T

So we have proved Claim 3. As we mentioned above, this implies the assertion

of Proposition 5.2.6.

The previous proposition provides an upper bound (5.23) for the numbers
Nr (after an appropriate constant field extension). Next we would like to
obtain a lower bound. We begin with a lemma from group theory.

Lemma 5.2.7. Let G be a group which is the direct product

G = σ × G (5.31)

of a cyclic subgroup σ and a subgroup G ⊆ G such that ord G = m, ord(σ) =

n and m divides n. Suppose that H ⊆ G is a subgroup of G with

ord H = ne and ord (H ∩ G) = e . (5.32)

Then there exist exactly e subgroups U ⊆ H with the following properties:

U is cyclic of order n, and U ∩ G = {1} . (5.33)

Proof. For τ ∈ G we consider the cyclic subgroup στ  ⊆ G . Since στ = τ σ

(by (5.31)), ord (σ) = n and ord (τ )|m we conclude that ord(στ ) = n. More-
over, στ  ∩ G = {1} and στ  = στ   for τ = τ  (all this follows immediately
from (5.31) because the elements λ ∈ G have a unique representation λ = σ i 
with 0 ≤ i < n and  ∈ G). Thus we have found m = ord G distinct subgroups
U ⊆ G with the properties (5.33).
The subgroup G ⊆ G is a normal subgroup, hence H/H ∩ G  HG/G.
By (5.32) this implies that HG = G , and H/H ∩ G  G /G is cyclic of
order n. Choose an element λ0 ∈ H whose order modulo H ∩ G is n, and
write λ0 = σ a τ  with τ  ∈ G and a ∈ ZZ. The exponent a is relatively prime
to n (otherwise there would be an integer 1 ≤ d < n with σ ad = 1, hence
 5.2 The Hasse-Weil Theorem 203

λd0 = τ d ∈ H ∩ G; so the order of λ0 modulo H ∩ G would be less than n).

Therefore a suitable power λ = λt0 has a representation λ = στ0 with τ0 ∈ G.
Let H ∩ G = {ψ1 , . . . , ψe }. We define

U (j) := στ0 ψj  for j = 1, . . . , e .

The subgroups U (j) ⊆ H are cyclic of order n, they are pairwise distinct, and
U (j) ∩ G = {1}.
It remains to show that H contains no other cyclic subgroup U of order
n with U ∩ G = {1}. In fact, let U ⊆ H be a subgroup satisfying (5.33). As
above we find a generator of U of the specific form στ1 with τ1 ∈ G. Since
στ1 ∈ H and στ0 ∈ H,

τ0−1 τ1 = (στ0 )−1 (στ1 ) ∈ H ∩ G = {ψ1 , . . . , ψe } .

Hence τ1 = τ0 ψj for some j, and U = στ1  = στ0 ψj  = U (j) .

The next proposition is the essential step in the proof of a lower bound
for Nr . We consider the following situation: E/L is a Galois extension of
function fields of degree [E : L] = m, and it is assumed that IFq is the full
constant field of E and L. We choose an integer n > 0 with m|n and let
E  := EIFqn (resp. L := LIFqn ⊆ E  ) be the corresponding constant field
extension of degree n. Then E  /L is Galois with Galois group G = σ × G,
where G := Gal(E  /L )  Gal(E/L) and σ is the Frobenius automorphism of
E  /E (i.e., σ(z) = z for z ∈ E and σ(α) = αq for α ∈ IFqn ). By the previous
lemma, G contains exactly m cyclic subgroups U ⊆ G with ord U = n and
U ∩ G = {1}, say U1 , . . . , Um . We can assume that U1 = σ.
Let Ei be the fixed field of Ui (i = 1, . . . , m). Then E1 = E, and we have
the situation as shown in Figure 5.1:

E
@@
@G
σ @
Ui @
L

E1 = E ... Ei ...
@
@ n
@ m
m@
@
L

Fig. 5.1.
204 5 Algebraic Function Fields over Finite Constant Fields

Denote by g(Ei ) the genus of Ei , and by N (Ei ) (resp. N (L)) the number of
places of degree one of Ei (resp. L).

Proposition 5.2.8. Under the above assumptions the following hold:

(a) IFq is the full constant field of Ei , for 1 ≤ i ≤ m.
(b) E  = Ei IFqn and g(Ei ) = g(E), for i = 1, . . . , m.
m
(c) m · N (L) = i=1 N (Ei ).

Proof. (a), (b) Note that Ui ∩ G = {1}. By Galois theory, E  is then the
compositum of Ei and L , hence E  = Ei L = Ei LIFqn = Ei IFqn is the
constant field extension of Ei with IFqn . Since [E  : Ei ] = ord Ui = n, this
implies that IFq is the full constant field of Ei . The genus is invariant under
constant field extensions, so g(Ei ) = g(E  ) = g(E) for i = 1, . . . , m.
(c) We consider the sets X := {P ∈ IPL | deg P = 1} and, for i = 1, . . . , m,
Xi := {Q ∈ IPEi | deg Q = 1}, and we have to prove the following assertion:
m 
 
 
 Xi  = m · |X| . (5.34)
 
i=1

Let P ∈ X. Choose a place P  ∈ IPE  lying over P , and set P1 := P  ∩ E. The

relative degree f (P1 |P ) divides m, as E/L is Galois. Hence f (P1 |P ) divides
n, and the residue class field of P  is IFqn by Theorem 3.6.3(g). This means
that the relative degree of P  |P is f (P  |P ) = n. Denote by e := e(P  |P ) the
ramification index of P in E  /L and by r the number of places in IPE  lying
over P (as E  /L is Galois, e depends only on P ). We obtain

m · n = [E  : L] = e(P  |P ) · f (P  |P ) · r = e · n · r .

Hence m = e · r, and (5.34) is reduced to the following claims:

Claim 1. For every Q ∈ Xi with Q|P there is exactly one place Q ∈ IPE 
which lies above Q.
Claim 2. For every place Q ∈ IPE  with Q |P there are exactly e distinct
places Q ∈ i=1 Xi such that Q |Q.
m

Proof of Claim 1. If Q ∈ IPE  lies above the place Q ∈ Xi and Q|P , then

f (Q |Q) = f (Q |Q) · f (Q|P ) = f (Q |P ) = n .

(Observe that f (Q|P ) = 1 as Q ∈ Xi .) Thus f (Q |Q) = [E  : Ei ], which

implies that Q is the only extension of Q in E  |Ei .
Proof of Claim 2. Here a place Q ∈ IPE  with Q |P is given. Let H ⊆
Gal(E  /L) be the decomposition group of Q over P , Z ⊆ E  the fixed field
of H and PZ := Q ∩ Z. Then

ord H = e(Q |P ) · f (Q |P ) = e · n

 5.2 The Hasse-Weil Theorem 205

and f (PZ |P ) = 1, by Theorem 3.8.2. It follows in particular that

IFq is the full constant field of Z . (5.35)

By Galois theory, the fixed field of H ∩ G is the compositum of Z and L . We

have ZL = ZLIFqn = ZIFqn and [ZIFqn : Z] = n (by (5.35)), hence

ord (H ∩ G) = [E  : Z]/[ZL : Z] = ne/n = e .

As PZ is unramified in ZL = ZIFqn , it also follows that T := ZL is the

inertia field and H ∩ G is the inertia group of Q |P (cf. Theorem 3.8.3).
Now we apply Lemma 5.2.7 once again: Exactly e of the cyclic groups
U1 , . . . , Um ⊆ Gal(E  /L) of order n with Ui ∩ G = {1} are contained in H,
say Ui1 , . . . , Uie . Let Qij := Q ∩ Eij . Because Eij contains the decomposition
field of Q over P , Q is the only place of E  lying over Qij . On the other
hand, e(Q |Qij ) = 1 since E  is a constant field extension of Eij (by (b)).
This implies f (Q |Qij ) = [E  : Eij ] = n = f (Q |P ), hence deg Qij = 1. In
m
this manner we have constructed e distinct places Qij ∈ i=1 Xi such that

Q |Qij .
Conversely, suppose that Q ∈ Xi for some i ∈ {1, . . . , m} and Q |Q. Then
f (Q |Q) = n. So Ui = Gal(E  |Ei ) is contained in the decomposition group H
of Q over P , i.e. Ui is one of the above groups Uij , and Q is the correspond-
ing place Qij (j ∈ {1, . . . , e}). This proves Claim 2 and finishes the proof of
Proposition 5.2.8.

End of the Proof of the Hasse-Weil Theorem 5.2.1. As we mentioned above,

it remains to establish a lower bound (5.24) for Nr = N (Fr ). We proceed
as follows. Choose a rational subfield F0 = IFq (t) ⊆ F such that F/F0 is
separable, and a finite extension E ⊇ F such that E/F0 is Galois (observe that
there exists a separating element t by Proposition 3.10.2). It is possible that
the constant field of E is a proper extension IFqd of IFq . In this case we consider
the fields F IFqd and F0 IFqd = IFqd (t) instead of F and F0 . The extension
E/F0 IFqd is Galois, and it is sufficient to prove the Hasse-Weil Theorem for
F IFqd /IFqd (by Lemma 5.2.4). So we may change notation and assume from
the beginning that IFq is also the full constant field of E. Moreover we can
assume that
q is a square and q > (g(E) + 1)4 . (5.36)
Let m := [E : F ] and n := [E : F0 ], and consider the constant field exten-
sions E  = EIFqn , F  := F IFqn and F0 := F0 IFqn . By Lemma 5.2.7 there
exist exactly m distinct cyclic subgroups V1 , . . . , Vm ⊆ Gal(E  /F ) of order
n such that Vi ∩ Gal(E  /F  ) = {1}. On the other hand there are n cyclic
subgroups U1 , . . . , Un ⊆ Gal(E  /F0 ) with the property ord(Uj ) = n and
Uj ∩ Gal(E  /F0 ) = {1}. It is easily seen that Vi ∩ Gal(E  /F0 ) = {1} (by
showing that E  is the compositum of F0 with the fixed field of Vi ), hence we
206 5 Algebraic Function Fields over Finite Constant Fields

can assume that Vi = Ui for i = 1, . . . , m. Denote by Ei the fixed field of Ui ,

for i = 1, . . . , n. Proposition 5.2.8 yields

m
m · N (F ) = N (Ei ) (5.37)
i=1

and

n
n · N (F0 ) = N (Ei ) . (5.38)
i=1

As we assumed (5.36), the upper bound

N (Ei ) ≤ q + 1 + (2g(E) + 1)q 1/2

holds for 1 ≤ i ≤ n, by Proposition 5.2.6. The places of F0 = IFq (t) of degree

one are the pole of t and, for each α ∈ IFq , the zero of t−α; thus N (F0 ) = q+1.
We combine this with (5.37) and (5.38) and obtain


m 
n
m · N (F ) = n · N (F0 ) + N (Ei ) − N (Ei )
i=1 i=1
n
= n(q + 1) − N (Ei )
i=m+1

≥ n(q + 1) − (n − m)(q + 1 + (2g(E) + 1)q 1/2 )

= m(q + 1) − (n − m)(2g(E) + 1)q 1/2 .

Therefore
n−m
N (F ) ≥ q + 1 − (2g(E) + 1)q 1/2 .
m
Observe that the numbers m, n and g(E) are invariant under constant field
extensions; so we have established a lower bound

Nr ≥ q r + 1 − c2 q r/2 (5.24)

with a constant c2 > 0. This finishes the proof of the Hasse-Weil Theorem.
Using the Hasse-Weil Bound, one can also give an estimate for the number
of places of a fixed degree r. Given a function field F/IFq of genus g, we define
 
Br := Br (F ) :=  {P ∈ IPF ; deg P = r}  . (5.39)

Observe that B1 = N (F ). There is a close relationship between the numbers

Br and Ns (the number of places of degree one in the constant field extension
Fs = F IFqs ), namely 
Nr = d · Bd (5.40)
d|r
 5.2 The Hasse-Weil Theorem 207

(the sum runs over all integers d ≥ 1 that divide r). This formula follows easily
from Lemma 5.1.9(d): every place P ∈ IPF of degree d|r decomposes into d
places of degree one in IPFr , and the extensions P  of P in Fr /F have degree
deg P  > 1 if deg P  r. The Möbius inversion formula (cf. [24]) converts (5.40)
into  r
r · Br = μ · Nd . (5.41)
d
d|r

Here μ : IN → {0, −1, 1} denotes the Möbius function which is defined by

⎧
⎪
⎨ 1 if n = 1 ,
μ(n) = 0 if there is an integer k > 1 with k 2 |n ,
⎪
⎩
(−1)l if n is the product of l distinct primes .

We set

2g
Sr := − αir , (5.42)
i=1

where α1 , . . . , α2g ∈ C are the reciprocals of the roots of LF (t) (for g = 0 we

set Sr := 0). Then
Nr = q r + 1 + S r ,
by Corollary 5.1.16. Substitute this into (5.41) and observe that
 r
μ( ) = 0 for r > 1
d
d|r

(cf. [24]). The result is:

Proposition 5.2.9. For all r ≥ 2,

1  r d
Br = · μ( )(q + Sd ) .
r d
d|r

Corollary 5.2.10. (a) The estimate

qr  q q 1/2  q r/2 − 1 q r/2

|Br − | ≤ + 2g 1/2 · < (2 + 7g) ·
r q−1 q −1 r r

holds for all r ≥ 1.

(b) If g = 0 then Br > 0 for all r ≥ 1.
(c) For each r such that 2g + 1 ≤ q (r−1)/2 (q 1/2 − 1) there exists at least one
place of degree r. In particular, if r ≥ 4g + 3 then Br ≥ 1.
208 5 Algebraic Function Fields over Finite Constant Fields

Proof. (a) For r = 1 we have B1 = N , and the assertion follows easily from
the Hasse-Weil Bound. For r ≥ 2 Proposition 5.2.9 yields
qr 1  r 1  r
Br − = μ qd + μ Sd .
r r d r d
d|r, d<r d|r

Setting l := [r/2] (the integer part of r/2) and observing that |Sd | =
2g
| i=1 αid | ≤ 2gq d/2 , we obtain

1  d 2g  d/2
l r
qr
|Br − |≤ q + q
r r r
d=1 d=1
q q l − 1 2gq 1/2 q r/2 − 1
= · + · 1/2
r q−1 r q −1
 q q 1/2  q r/2 − 1
≤ + 2g 1/2 ·
q−1 q −1 r
q r/2
< (2 + 7g) · .
r

(b),(c) From (a) it follows that Br > 0 whenever

qr  q q 1/2  q r/2 − 1
> + 2g 1/2 · . (5.43)
r q−1 q −1 r
In the case g = 0, (5.43) holds for all r ≥ 1. This implies (b), and we can
assume g ≥ 1 from now on. A simple calculation shows that (5.43) is equivalent
to
1 q r (q 1/2 − 1)
2g + < . (5.44)
1 + q −1/2 q 1/2 (q r/2 − 1)
The inequalities

1 q r (q 1/2 − 1)
2g + < 2g + 1 and q (r−1)/2 (q 1/2 − 1) <
1 + q −1/2 q 1/2 (q r/2 − 1)

are trivial, hence our assumption 2g + 1 ≤ q (r−1)/2 (q 1/2 − 1) implies (5.44)

and therefore Br > 0. If r ≥ 4g + 3 then

2g + 1 < 22g+1 (21/2 − 1) ≤ 2(r−1)/2 (21/2 − 1) ≤ q (r−1)/2 (q 1/2 − 1) .

This completes the proof of (c).

5.3 Improvements of the Hasse-Weil Bound

In general the Hasse-Weil Bound |N − (q + 1) | ≤ 2gq 1/2 is sharp. There
are examples of function fields F/IFq such that N = q + 1 + 2gq 1/2 (resp.
 5.3 Improvements of the Hasse-Weil Bound 209

N = q + 1 − 2gq 1/2 ). We shall present several examples in Chapter 6. Under

certain assumptions however, the bound can be improved. For instance, if q
is not a square, we have the trivial improvement

|N − (q + 1)| ≤ [2gq 1/2 ] , (5.45)

where [a] denotes the integer part of the real number a. The bound (5.45) can
be ameliorated substantially as follows:

Theorem 5.3.1 (Serre Bound). For a function field F/IFq of genus g the
number of places of degree one is bounded by

|N − (q + 1)| ≤ g [2q 1/2 ] . (5.46)

Proof. Let A ⊆ C be the set of algebraic integers; i.e., a complex number α is

in A if and only if α satisfies an equation αm + bm−1 αm−1 + · · · + b1 α + b0 = 0
with coefficients bi ∈ ZZ. It is an elementary fact from algebraic number theory
that
A is a subring of C, and A ∩ Q = ZZ . (5.47)
For the proof of the
Serre Bound we can assume that g > 0. We consider the L-
2g
polynomial L(t) = i=1 (1 − αi t) of F/IFq . The complex numbers α1 , . . . , α2g
are algebraic integers with |αi | = q 1/2 (Theorem 5.1.15 and Theorem 5.2.1).
They can be ordered such that αi αg+i = q, hence

ᾱi = αg+i = q/αi for 1 ≤ i ≤ g .

(We denote by ᾱ the complex conjugate of α.) Set

γi := αi + ᾱi + [2q 1/2 ] + 1 ,

δi := −(αi + ᾱi ) + [2q 1/2 ] + 1 .

By (5.47), γi and δi are real algebraic integers and, as |αi | = q 1/2 , they satisfy

γi > 0 and δi > 0 . (5.48)

Each embedding σ : Q(α1 , . . . , α2g ) −→ C permutes α1 , . . . , α2g because

2g ⊥
i=1 (t − αi ) = L (t) ∈ Z
Z[t], cf. (5.14). Moreover, if σ(αi ) = αj then

σ(ᾱi ) = σ(q/αi ) = q/σ(αi ) = σ(αi ) = ᾱj .

Therefore σ acts as a permutation on the sets {γ1 , . . . , γg } and {δ1 , . . . , δg }.

Define

g g
γ := γi and δ := δi .
i=1 i=1

γ and δ are algebraic integers which are invariant under all embeddings of
Q(α1 , . . . , α2g ) into C. Hence γ, δ ∈ Q ∩ A = ZZ. By (5.48), γ > 0 and δ > 0,
so we have
210 5 Algebraic Function Fields over Finite Constant Fields


g 
g
γi ≥ 1 and δi ≥ 1 .
i=1 i=1

The well-known inequality between the arithmetic and geometric mean yields
now  1/g
1 
g g
γi ≥ γi ≥ 1.
g i=1 i=1

Hence

g 
g≤ (αi + ᾱi ) + g[2q 1/2 ] + g
i=1

2g
= αi + g[2q 1/2 ] + g .
i=1
2g
Observing that i=1 αi = (q + 1) − N by Corollary 5.1.16, we obtain

N ≤ q + 1 + g[2q 1/2 ] .

In the same manner, the inequality

 1/g
1 
g g
δi ≥ δi ≥ 1
g i=1 i=1

implies that
N ≥ q + 1 − g[2q 1/2 ] .

One is often interested in function fields which have many places of degree
one. So we introduce the following notion:

Definition 5.3.2. A function field F/IFq of genus g is said to be maximal if

N = q + 1 + 2gq 1/2 .

Obviously maximal function fields over IFq can exist only if q is a square.
Our next result is due to Y. Ihara. It shows that F/IFq cannot be maximal if
the genus is large with respect to q.

Proposition 5.3.3 (Ihara). Suppose that F/IFq is a maximal function field.

Then g ≤ (q − q 1/2 )/2.

Proof. Let α1 , . . . , α2g be the reciprocals of the roots of L(t). Since


2g
N =q+1− αi and |αi | = q 1/2
i=1
 5.3 Improvements of the Hasse-Weil Bound 211

(by Corollary 5.1.16 and Theorem 5.2.1), the assumption N = q + 1 + 2gq 1/2
implies
αi = −q 1/2 for i = 1, . . . , 2g . (5.49)
Next we consider the number N2 of places of degree one in the constant field
extension F IFq2 /IFq2 . We have N2 ≥ N and


2g
N2 = q 2 + 1 − αi2 = q 2 + 1 − 2gq ,
i=1

by Corollary 5.1.16 and (5.49). So

q + 1 + 2gq 1/2 ≤ q 2 + 1 − 2gq .

The inequality g ≤ (q − q 1/2 )/2 follows immediately.

Ihara’s estimate for the genus of a maximal function field cannot be im-
proved in general: in Chapter 6 we shall show that there exist maximal func-
tion fields of genus g = (q − q 1/2 )/2 (if q is a square).
One can refine the proof of Proposition 5.3.3 in order to obtain other
bounds for the number of places of degree one. This method was developed
by J.-P. Serre. We proceed as follows: Let
 
Nr = N (Fr ) =  {P ∈ IPFr ; deg P = 1}  ,

where Fr = F IFqr is the constant field extension of F of degree r. We write

for i = 1, . . . , 2g,
ωi := αi q −1/2 , (5.50)
where α1 , . . . , α2g are the reciprocals of the roots of LF (t). Then |ωi | = 1 by
the Hasse-Weil Theorem, and we can assume that

ωg+i = ω̄i = ωi−1 for i = 1, . . . , g . (5.51)

(This follows from Theorem 5.1.15(e).) By Corollary 5.1.16,


g
Nr q −r/2 = q r/2 + q −r/2 − (ωir + ωi−r ) . (5.52)
i=1

Given real numbers c1 , c2 , . . . we multiply (5.52) by cr and obtain


g
N1 cr q −r/2
= cr q r/2
+ cr q −r/2
− cr (ωir + ωi−r ) − (Nr − N1 )cr q −r/2 . (5.53)
i=1

Summing up the equations (5.53) for r = 1, . . . , m gives

212 5 Algebraic Function Fields over Finite Constant Fields


g
N1 · λm (q −1/2 ) = λm (q 1/2 ) + λm ( q −1/2 ) + g − fm (ωi )
i=1

m
− (Nr − N1 )cr q −r/2 , (5.54)
r=1

where

m
λm (t) := cr tr (5.55)
r=1

and
fm (t) := 1 + λm (t) + λm (t−1 ) (5.56)
for t ∈ C, t = 0. Note that fm (t) ∈ IR for |t| = 1. For specific choices of the
constants cr , Equation (5.54) yields good estimates for N . For instance we
show:

Proposition 5.3.4 (Serre’s Explicit Formulas). Suppose that c1 , . . . , cm ∈

IR satisfy the following conditions:
(1) cr ≥ 0 for r = 1, . . . , m, and not all cr = 0.
(2) fm (t) ≥ 0 for all t ∈ C with |t| = 1 (where fm (t) is defined by (5.56)).
Then the number of rational places of F/IFq is bounded by

g λm (q 1/2 )
N ≤ + + 1, (5.57)
λm (q −1/2 ) λm (q −1/2 )
with λm (t) as in (5.55).
Proof. We have N = N1 ≤ Nr for all r ≥ 1. So (5.54) and the assumptions
(1) and (2) imply

N · λm (q −1/2 ) ≤ λm (q 1/2 ) + λm (q −1/2 ) + g .

Dividing this inequality by λm (q −1/2 ) yields (5.57) (note that λm (q −1/2 ) > 0
by assumption (1)).

In Chapter 7 we will use Proposition 5.3.4 to prove an asymptotic bound

for N (F ) when the genus of F tends to infinity.

5.4 Exercises
In all exercises below we assume that F/IFq is a function field with full con-
stant field IFq , and N = N1 denotes the number of rational places of F/IFq .
The constant field extension of F of degree r is denoted by Fr = F IFqr , and
Nr is the number of rational places of Fr /IFqr . Moreover, L(t) denotes the
L-polynomial of F/IFq .
 5.4 Exercises 213

5.1. Let F be a function field of genus 1 over IF5 having N = 10 rational

places.
(i) Determine the L-polynomial L(t) and the class number h = hF .
(ii) Determine N2 and N3 .
(iii) How many divisors A ≥ 0 are there with deg A = 1, 2, 3 and 4?
(iv) Determine the number of places of F/IF5 of degree 2, 3 and 4.

5.2. Show that over every finite field IFq there exists a function field without
rational places.

5.3. Using results of this chapter, find a formula for the number of irreducible
monic poynomials p(x) ∈ IFq [x] of degree deg p(x) = n.

5.4. If F/IFq is a function field over a finite field, show that its automorphism
group Aut (F/IFq ) is finite.

5.5. Assume that q = 2 is a square and the genus of F is g ≥ 1.

(i) Show that F/IFq is maximal if and only if L(t) = (1 + t)2g .
(ii) If F/IFq is maximal, show that Fr /IFqr is maximal if and only if r is odd.

5.6. Assume that the number of rational places of F/IFq attains the upper
Serre Bound N = q + 1 + g[2q 1/2 ]. Determine the L-polynomial L(t).
Hint. When is the arithmetic mean equal to the geometric mean?

5.7. (i) Let P be a rational place of F/IFq . Show that all integers k with
1 ≤ k ≤ (N − 2)/q are gap numbers of P .

(ii) Conclude from (i) the estimate N ≤ q + 1 + qg.

Note. This simple estimate is in general much weaker than the Hasse-Weil
Bound. But note that it is better than Hasse-Weil for q = 2 and q = 3 , where
it coincides with the Serre Bound.

5.8. Assume that F/IFq is a non-rational function field. In this exercise we

will provide a lower bound for the class number h = hF .
(i) Show that the number of positive divisors of F/IFq of degree 2g is equal
to h · (q g+1 − 1)/(q − 1).
(ii) Consider the constant field extension F2g of degree 2g. Let Q be a rational
place of F2g and let P = Q ∩ Fq be the restriction of Q to F . Then a :=
2g/ deg P is an integer, hence we obtain a positive divisor a P of F/IFq . Show
that in this way one constructs at least N2g /2g distinct positive divisors of
F/IFq of degree 2g.
(iii) Using (i), (ii) and the Hasse-Weil Bound for N2g , show that

q − 1 q 2g + 1 − 2gq g
h≥ · .
2 g(q g+1 − 1)
214 5 Algebraic Function Fields over Finite Constant Fields

(iv) Prove that h > (q − 1)/4.

(v) Given a real number M > 0, show that there are - up to isomorphism -
only finitely many non-rational function fields over finite fields having class
number h ≤ M . In particular, there exist only finitely many non-rational
function fields over finite fields with class number one (an analogous result
does not hold for algebraic number fields).

5.9. Let F/IFq be a non-rational function field with class number h = 1.

Prove:
(i) q ≤ 4.
(ii) If q = 4, then g = 1.
(iii) If q = 3, then g ≤ 2.
(iv) If q = 2, then g ≤ 4.
Remark. One can show that there is no function field of class number 1 with
q = 3, g = 2 or q = 2, g = 4.

5.10. For this exercise we first recall some notations and introduce a few
others. Let S, T ⊆ IPF be non-empty
! subsets of IPF such that S ∪ T = IPF
and S ∩ T = ∅. Let OS = P ∈S OP be the corresponding holomorphy ring,
cf. Section 3.2. For 0 = x ∈ F , its S-divisor (x)S is defined as

(x)S := vP (x)P .
P ∈S

We consider the following groups:

Div(F ) , the divisor group of F

Div0 (F ) , the group of divisors of degree 0
Princ(F ) , the group of principal divisors of F
Cl0 (F ) = Div0 (F )/Princ(F ) , the group of divisor classes of degree 0
DivS , the subgroup of Div(F ) which is generated by all P ∈ S
PrincS := {(x)S | 0 = x ∈ F }
ClS = DivS /PrincS , the S-class group of F

Finally we define

h := hF , the class number of F

hS := ord (ClS ) , the S-class number of F
rS := ord (DivT ∩ Div0 (F )) / (DivT ∩ Princ(F )) , the regulator of OS
uS := gcd {deg P | P ∈ T }
 5.4 Exercises 215

(i) Show that ClS is isomorphic to Div(F )/(Princ(F ) + DivT ).

(ii) Show that uS = ord(Div(F )/(Div0 (F ) + DivT )).
(iii) Show that there is an exact sequence

DivT ∩ Div0 (F ) Div0 (F ) + DivT

0 −→ −→ Cl0 (F ) −→ −→ 0 .
DivT ∩ Princ(F ) Princ(F ) + DivT

(iv) Conclude that hS and rS are finite, rS |h and

h
hS = u S · .
rS
Remark. We specialize the situation of this exercise as follows: F is an exten-
sion of the rational function field IFq (x) of degree [F : IFq (x)] = 2. Choose
S := {P ∈ IPF | vP (x) ≥ 0}, then OS is the integral closure of IFq [x] in F (see
Section 3.2). Now we distinguish 3 cases.
Case 1. (x)∞ = P1 +P2 with P1 = P2 . Then uS = 1, and rS is the order of the
divisor class [P2 − P1 ] in the class group Cl(F ), hence hS = h/ord([P2 − P1 ]).
Case 2. (x)∞ = 2P . Now uS = rS = 1 and hS = h.
Case 3. (x)∞ = P with deg P = 2. Now uS = 2, rS = 1 and hS = 2h.
In analogy with algebraic number theory, in case 1 the function field F is
called real-quadratic, and in the cases 2 and 3 it is called imaginary quadratic
(with respect to x).
There are more exercises about function fields over finite fields at the end of
Chapter 6.
6
Examples of Algebraic Function Fields

Thus far we have encountered very few explicit examples of algebraic function
fields, namely the rational function field K(x)/K (cf. Section 1.2) and some
quadratic extensions of the rational function field (Example 3.7.6). Now we
would like to discuss some other examples in detail. These examples serve as
an illustration of the general theory of algebraic function fields developed in
Chapters 1, 3, 4 and 5. Some of the examples will be used in Chapter 8 for
the construction of algebraic geometry codes.

Throughout this chapter K denotes a perfect field.

This assumption is not essential; actually most results of Chapter 6 hold for
an arbitrary constant field K, with minor modifications.

6.1 Elliptic Function Fields

The rational function field K(x) has genus 0. Conversely, if F/K is a function
field of genus 0 which has a divisor A ∈ Div(F ) of degree one then F/K is
rational, see Proposition 1.6.3. Therefore the simplest non-rational function
fields are fields of genus one.
Definition 6.1.1. An algebraic function field F/K (where K is the full con-
stant field of F ) is said to be an elliptic function field if the following conditions
hold:
(1) the genus of F/K is g = 1, and
(2) there exists a divisor A ∈ Div(F ) with deg A = 1.

There are numerous connections between elliptic function fields and other
branches of mathematics (such as number theory and complex analysis), and
there exists an extensive literature on the theory of elliptic function fields, cf.
[38]. Here we present only some basic facts on the subject.
H. Stichtenoth, Algebraic Function Fields and Codes, 217
Graduate Texts in Mathematics 254,

c Springer-Verlag Berlin Heidelberg 2009
218 6 Examples of Algebraic Function Fields

Proposition 6.1.2. Let F/K be an elliptic function field.

(a) If char K = 2, there exist x, y ∈ F such that F = K(x, y) and

y 2 = f (x) ∈ K[x] (6.1)

with a square-free polynomial f (x) ∈ K[x] of degree 3.

(b) If char K = 2, there exist x, y ∈ F such that F = K(x, y) and

y 2 + y = f (x) ∈ K[x] with deg f = 3 , (6.2)

or
1
y2 + y = x + with a, b ∈ K and a = 0 . (6.3)
ax + b

Proof. Choose a divisor A of degree one. By the Riemann-Roch Theorem,

(A) = deg A + 1 − g = 1 (note that deg A > 2g − 2). Hence A is equivalent
to a positive divisor A1 , cf. Remark 1.4.5. As deg A1 = 1 we conclude that A1
is a prime divisor A1 = P ∈ IPF . So we have shown that an elliptic function
field F/K has at least one place P ∈ IPF with deg P = 1.
We consider the spaces K = L (0) ⊆ L (P ) ⊆ . . . ⊆ L (nP ) ⊆ . . .. Because
2g − 2 = 0, the Riemann-Roch Theorem gives dim L (iP ) = i for all i > 0,
hence L (P ) = K and L ((i + 1)P )  L (iP ) for i > 0. We choose elements
x1 ∈ L (2P )\K and y1 ∈ L (3P )\L (2P ). Their pole divisors are

(x1 )∞ = 2P and (y1 )∞ = 3P .

As [F : K(x1 )] = 2 and [F : K(y1 )] = 3 (by Theorem 1.4.11), it follows that

F = K(x1 , y1 ).
The seven elements 1, x1 , y1 , x21 , x1 y1 , x31 , y12 are in the space L (6P ). Since
(6P ) = 6, there is a non-trivial relation

α1 y12 + β1 x1 y1 + γ1 y1 = δ1 x31 + ε1 x21 + λ1 x1 + μ1 (6.4)

with α1 , β1 , . . . ∈ K. The coefficient α1 does not vanish; otherwise (6.4) would

give an equation for y1 over K(x1 ) of degree one (which is impossible as
F = K(x1 , y1 ) and [F : K(x1 )] = 2). In the same manner we see that δ1 = 0.
Multiply (6.4) by α13 δ12 ; then

α14 δ12 y12 + · · · = α13 δ13 x31 + · · · .

Setting y2 := α12 δ1 y1 and x2 := α1 δ1 x1 we obtain F = K(x2 , y2 ) and

y22 + (β2 x2 + γ2 )y2 = x32 + ε2 x22 + λ2 x2 + μ2 (6.5)

with β2 , γ2 , . . . ∈ K. Now we must distinguish the cases char K = 2 and

char K = 2.
 6.1 Elliptic Function Fields 219

(a) char K = 2. We set y := y2 + (β2 x2 + γ2 )/2 and x := x2 ; then F = K(x, y)

and
y 2 = x3 + εx2 + λx + μ = f (x) ∈ K[x] (6.1)
with ε, λ, μ ∈ K. It remains to show that f (x) is square-free. Suppose the
contrary; i.e., f (x) = (x − ζ)2 (x − η) with ζ, η ∈ K. Consider the element
z := y/(x − ζ). Then z 2 = x − η, and F = K(x, y) = K(x, z) = K(z). So F/K
is rational, a contradiction.
(b) char K = 2. We have already shown that F = K(x2 , y2 ) with

y22 + (β2 x2 + γ2 )y2 = x32 + ε2 x22 + λ2 x2 + μ2 . (6.5)

We claim that β2 x2 + γ2 = 0. In fact, if β2 x2 + γ2 = 0 then y22 ∈ K(x2 ); i.e.,

the extension F/K(x2 ) is purely inseparable of degree p = 2. By Proposition
3.10.2 the only intermediate field K ⊆ F0 ⊆ F such that F/F0 is purely
inseparable of degree p is the field F0 = F p , so K(x2 ) = F p . However, the
genus of K(x2 )/K is zero, and the genus of F p /K is one (cf. Proposition
3.10.2(c)). This contradiction proves the claim.
We set y3 := y2 (β2 x2 + γ2 )−1 ; then F = K(x2 , y3 ) and

y32 + y3 = (β2 x2 + γ2 )−2 (x32 + ε2 x22 + λ2 x2 + μ2 ) . (6.6)

If β2 = 0, the right hand side of (6.6) is a polynomial f (x2 ) ∈ K[x2 ] of degree

3, and we are in the situation of (6.2).
If β2 = 0, the right hand side of (6.6) can be written in the form
σ τ
νx2 +  + 2
+
(β2 x2 + γ2 ) β2 x2 + γ2

with ν, , σ, τ ∈ K and ν = 0. As K is perfect, σ = σ12 for some σ1 ∈ K, and

the element y4 := y3 + σ1 (β2 x2 + γ2 )−1 satisfies an equation
τ2
y42 + y4 = ν2 x2 + 2 + (6.7)
β2 x2 + γ2
with ν2 , 2 , τ2 ∈ K and ν2 = ν = 0. Also, the coefficient τ2 does not vanish
(otherwise F = K(x2 , y4 ) would be rational by (6.7)). We set y := y4 and
x := ν2 x2 + 2 and obtain F = K(x, y) with
1
y2 + y = x + (6.8)
ax + b
(a, b ∈ K and a = 0).

Next we show that each of the above equations (6.1), (6.2) and (6.3) defines
an elliptic function field.
220 6 Examples of Algebraic Function Fields

Proposition 6.1.3. (a) char K = 2. Suppose that F = K(x, y) with

y 2 = f (x) ∈ K[x] , (6.1)

where f (x)
ris a square-free polynomial of degree 3. Consider the decomposition
f (x) = c i=1 pi (x) of f (x) into irreducible monic polynomials pi (x) ∈ K[x]
with 0 = c ∈ K. Denote by Pi ∈ IPK(x) the place of K(x) corresponding to
pi (x), and by P∞ ∈ IPK(x) the pole of x. Then the following hold:
(1) K is the full constant field of F , and F/K is an elliptic function field.
(2) The extension F/K(x) is cyclic of degree 2. The places P1 , . . . , Pr and
P∞ are ramified in F/K(x); each of them has exactly one extension in
F , say Q1 , . . . , Qr and Q∞ , and we have e(Qj |Pj ) = e(Q∞ |P∞ ) = 2,
deg Qj = deg Pj and deg Q∞ = 1.
(3) P1 , . . . , Pr and P∞ are the only places of K(x) which are ramified in
F/K(x), and the different of F/K(x) is

Diff(F/K(x)) = Q1 + · · · + Qr + Q∞ .

(b) char K = 2. Suppose that F = K(x, y) with

y 2 + y = f (x) ∈ K[x] and deg f (x) = 3 (6.2)

or
1
y2 + y = x + with a, b ∈ K and a = 0 . (6.3)
ax + b
Denote by P∞ ∈ IPK(x) the pole of x in K(x) and by P  ∈ IPK(x) the zero of
ax + b in K(x) (in Case (6.3)). Then the following hold:
(1) K is the full constant field of F , and F/K is an elliptic function field.
(2) The extension F/K(x) is cyclic of degree 2. The only places of K(x)
which ramify in F/K(x) are

P∞ , in Case (6.2) ,
P∞ and P  , in Case (6.3) .

Let Q∞ (resp. Q in Case (6.3)) be the place of F/K lying over P∞

(resp. P  ). Then deg Q∞ = deg Q = 1 and

4Q∞ in Case (6.2) ,
Diff(F/K(x)) = 
2Q∞ + 2Q in Case (6.3) .

Proof. In case char K = 2, all assertions follow easily from Proposition 3.7.3
(see also Corollary 3.7.4 and Example 3.7.6). For the case of char K = 2 apply
Proposition 3.7.8.
 6.1 Elliptic Function Fields 221

For an elliptic function field F/K the zero divisor is a canonical divisor
since
(0) = 1 = g and deg (0) = 0 = 2g − 2
(see Proposition 1.6.2). In each of the cases (6.1), (6.2) and (6.3) one can
easily write down a differential ω ∈ ΩF with (ω) = 0, namely
⎧
⎪ −1
⎨y dx in Case (6.1) ,
ω = dx in Case (6.2) ,
⎪
⎩
(ax + b)−1 dx in Case (6.3) .

The proof of this assertion is left to the reader. (Hint: Calculate the divisor
of the differential dx by using Remark 4.3.7(c).)
Example 6.1.4. Let us briefly describe the classical example of an elliptic func-
tion field (without giving proofs). Consider a lattice Γ ⊆ C; i.e.,

Γ = ZZγ1 ⊕ ZZγ2

with γ1 , γ2 ∈ C\{0} and γ1 /γ2 ∈ IR. An elliptic function (with respect to Γ )

is a meromorphic function f (z) on C satisfying

f (z + γ) = f (z) for all γ ∈ Γ .

The elliptic functions form a subfield M(Γ ) of the field of all meromorphic
functions on C, and C ⊆ M(Γ ) (we consider a complex number as a constant
function). Two specific non-constant elliptic functions are the Weierstrass ℘-
function which is defined by
1   1 1

℘(z) := 2 + − ,
z (z − γ)2 γ2
0=γ∈Γ

and its derivative ℘ (z). It is not difficult to prove the following facts:
(1) M(Γ ) = C(℘(z), ℘ (z)), and
(2) ℘ (z)2 = 4℘(z)3 − g2 · ℘(z) − g3
with constants g2 , g3 ∈ C, where the polynomial f (T ) = 4T 3 −g2 T −g3 ∈ C[T ]
is square-free. Hence M(Γ )/C is an elliptic function field by Proposition 6.1.3.
For α ∈ C, every function 0 = f ∈ M(Γ ) has a Laurent series expansion
∞

f (z) = ai (z − α)i
i=i0

with ai ∈ C, i0 ∈ ZZ and ai0 = 0. Setting vα (f ) := i0 we define a discrete

valuation vα , hence a place Pα of M(Γ )/C. It is obvious that Pα = Pβ if
and only if α ≡ β mod Γ . In this manner one obtains all places of the elliptic
function field M(Γ ).
222 6 Examples of Algebraic Function Fields

Example 6.1.5. We would like to investigate some elliptic function fields F

over the field IF2 . Let N denote the number of places of F/IF2 of degree one.
The Serre Bound states that
√
N ≤ 2 + 1 + g · [2 2] = 5 .

Let us show that (up to isomorphism) there exists exactly one elliptic function
field F/IF2 with N = 5. By Proposition 6.1.2(b) we can write F = IF2 (x, y)
with
1
y2 + y = x + , b ∈ IF2 (6.8)
x+b
or
y 2 + y = f (x) ∈ IF2 [x] , deg f (x) = 3 , (6.9)
Observe that there are exactly 3 places of IF2 (x) of degree one, and every
place of F of degree one must lie over one of them. In Case (6.8) two places
of IF2 (x) of degree one ramify in F/IF2 (x), so N ≤ 4. In Case (6.9) we can
assume that f (x) = x3 + bx + c with b, c ∈ IF2 (if f (x) = x3 + x2 + bx + c, we
replace y by z := y + x; then z 2 + z = x3 + b1 x + c1 ). It remains to consider
the four cases f (x) = x3 , x3 + x, x3 + 1, x3 + x + 1. Using Kummer’s Theorem
(or its Corollary 3.3.8) one can easily calculate N in each of these cases; the
result is ⎧
⎪ 2 3
⎨1 for y + y = x + x + 1 ,
N = 3 for y + y = x3 or x3 + 1 ,
2
⎪
⎩
5 for y 2 + y = x3 + x .
So the only elliptic function field F/IF2 with N = 5 is

F = IF2 (x, y) with y 2 + y = x3 + x . (6.10)

Now we determine the L-polynomial LF (t) of (6.10). We know by Theorem

5.1.15 that LF (t) = a0 +a1 t+a2 t2 with a0 = 1, a2 = 2 and a1 = N −(2+1) =
2. Hence
LF (t) = 1 + 2t + 2t2 = (1 − αt)(1 − ᾱt) (6.11)
√ √
with α = −1 + i = ω 2, ω = exp (3πi/4) (here, i = −1 ∈ C, and ᾱ is the
complex conjugate of α). Consider the constant field extension Fr := F IF2r of
degree r. The number Nr of places of degree one of the function field Fr /IF2r
is given by
Nr = 2r + 1 − (αr + ᾱr ) (6.12)
(cf. Corollary 5.1.16). We obtain Nr = 2r + 1 − 2 · 2r/2 · Re (ω r ), hence
⎧
⎪
⎪
r
for r ≡ 2, 6 mod 8 ,
⎪2 + 1
⎪
⎪
⎪
⎨2 + 1 + 2 · 2 for r ≡ 4 mod 8 ,
r r/2

Nr = 2r + 1 − 2 · 2r/2 for r ≡ 0 mod 8 ,

⎪
⎪
⎪2 + 1 + 2
⎪ r (r+1)/2
for r ≡ 1, 7 mod 8 ,
⎪
⎪
⎩2r + 1 − 2(r+1)/2 for r ≡ 3, 5 mod 8 .
 6.1 Elliptic Function Fields 223

We see that for r ≡ 4 mod 8 the upper Hasse-Weil Bound q + 1 + 2gq 1/2 is
attained; for r ≡ 0 mod 8 the lower Hasse-Weil Bound q+1−2gq 1/2 is attained;
for r = 1 the upper Serre Bound q + 1 + g · [2q 1/2 ] is attained.

We finish this section with a result that is fundamental to the theory of

elliptic function fields F/K (over an arbitrary field K). Let us recall some
notation: Cl(F ) is the divisor class group of F/K, and Cl0 (F ) ⊆ Cl(F ) is
the subgroup consisting of the divisor classes of degree zero. For a divisor
B ∈ Div(F ), [B] ∈ Cl(F ) denotes the corresponding divisor class. A ∼ B
means that the divisors A, B are equivalent.
Proposition 6.1.6. Let F/K be an elliptic function field. Define
(1)
IPF := {P ∈ IPF | deg P = 1} .
Then the following hold:
(a) For each divisor A ∈ Div(F ) with deg A = 1 there exists a unique place
(1) (1)
P ∈ IPF with A ∼ P . In particular IPF = ∅.
(1)
(b) Fix a place P0 ∈ IPF . Then the mapping

(1)
IPF −→ Cl0 (F ) ,
Φ: (6.13)
P −→ [P − P0 ]
is bijective.
Proof. (a) Let A ∈ Div(F ) and deg A = 1. We show the existence of a place
(1)
P ∈ IPF with A ∼ P as in the proof of Proposition 6.1.2; since (A) =
deg A + 1 − g > 0, there is a divisor A1 ∼ A with A1 > 0, and from deg A = 1
(1)
follows immediately that A1 = P ∈ IPF .
(1)
Next we prove uniqueness. Suppose that A ∼ P and A ∼ Q for P, Q ∈ IPF
and P = Q. Then P ∼ Q; i.e., P − Q = (x) for some x ∈ F . By Theorem
1.4.11 we have [F : K(x)] = deg (x)∞ = deg Q = 1, hence F = K(x). This is
a contradiction as F/K is elliptic.
(b) First we show that Φ is surjective: Let [B] ∈ Cl0 (F ). The divisor B + P0
(1)
has degree one. By (a) we find a place P ∈ IPF with B + P0 ∼ P . Then
[B] = [P − P0 ] = Φ(P ), and Φ is surjective.
(1)
Suppose now that Φ(P ) = Φ(Q) for P, Q ∈ IPF . Then P − P0 ∼ Q − P0 ,
hence P ∼ Q. By the uniqueness assertion of (a) it follows that P = Q.

The bijection Φ of the foregoing proposition can be used to carry over

(1)
the group structure of Cl0 (F ) to the set IPF . That means, we define for
(1)
P, Q ∈ IPF
P ⊕ Q := Φ−1 (Φ(P ) + Φ(Q)) . (6.14)
Some consequences of this definition are put together in the following propo-
sition.
224 6 Examples of Algebraic Function Fields

Proposition 6.1.7. Let F/K be an elliptic function field. Then:

(1)
(a) IPF is an abelian group with respect to the operation ⊕ as defined in
(6.14).
(1)
(b) The place P0 is the zero element of the group IPF .
(1)
(c) For P, Q, R ∈ IPF the following holds:
P ⊕ Q = R ⇐⇒ P + Q ∼ R + P0 .
(1)
(d) The map Φ : IPF → Cl0 (F ) given by (6.13) is a group isomorphism.
Proof. (a), (b) and (d) are obvious.
(c) By (6.14) we have the following equivalences:
P ⊕ Q = R ⇐⇒ Φ(R) = Φ(P ) + Φ(Q)
⇐⇒ R − P0 ∼ (P − P0 ) + (Q − P0 )
⇐⇒ P + Q ∼ R + P0 .

(1)
We note that the group law on IPF depends on the choice of the place
(1)
P0 . However, the group-theoretical structure of IPF is independent of this
(1)
choice, since IPF is isomorphic to Cl0 (F ) in any case. If F is represented in
the form F = K(x, y) as in Proposition 6.1.3, one usually chooses P0 := Q∞ ,
the pole of x.

6.2 Hyperelliptic Function Fields

In this section we discuss another important class of non-rational function
fields over K.
Definition 6.2.1. A hyperelliptic function field over K is an algebraic func-
tion field F/K of genus g ≥ 2 which contains a rational subfield K(x) ⊆ F
with [F : K(x)] = 2.
Lemma 6.2.2. (a) A function field F/K of genus g ≥ 2 is hyperelliptic if
and only if there exists a divisor A ∈ Div(F ) with deg A = 2 and (A) ≥ 2.
(b) Every function field F/K of genus 2 is hyperelliptic.
Proof. (a) Suppose that F/K is hyperelliptic. Choose an element x ∈ F such
that [F : K(x)] = 2, and consider the divisor A := (x)∞ . Then deg A = 2 and
the elements 1, x ∈ L (A) are linearly independent over K, hence (A) ≥ 2.
Conversely, assume that F/K has genus g ≥ 2 and that A is a divisor of
degree 2 with (A) ≥ 2. There is a divisor A1 ≥ 0 with A1 ∼ A, so deg A1 = 2
and (A1 ) ≥ 2, and we can find an element x ∈ L (A1 )\K. Then (x)∞ ≤ A1
and therefore [F : K(x)] = deg (x)∞ ≤ 2. Since F/K is not rational, we
conclude that [F : K(x)] = 2.
 6.2 Hyperelliptic Function Fields 225

(b) Now a function field F/K of genus g = 2 is given. For each canonical
divisor W ∈ Div(F ) we have deg W = 2g − 2 = 2 and (W ) = g = 2 by
Corollary 1.5.16. This implies that F/K is hyperelliptic, by (a).

If F/K is hyperelliptic and K(x) is a subfield of F with [F : K(x)] = 2, the

extension F/K(x) is separable (if F/K(x) were purely inseparable, F itself
would be rational, by Proposition 3.10.2). Hence F/K(x) is a cyclic extension
of degree 2, and we can use Proposition 3.7.3 (resp. 3.7.8) to provide an
explicit description of F/K (analogous to the description of elliptic function
fields given in Section 6.1). For simplicity we restrict ourselves to the case
char K = 2.
Proposition 6.2.3. Assume that char K = 2.
(a) Let F/K be a hyperelliptic function field of genus g. Then there exist
x, y ∈ F such that F = K(x, y) and
y 2 = f (x) ∈ K[x] (6.15)
with a square-free polynomial f (x) of degree 2g + 1 or 2g + 2.
(b) Conversely, if F = K(x, y) and y 2 = f (x) ∈ K[x] with a square-free
polynomial f (x) of degree m > 4, then F/K is hyperelliptic of genus

(m − 1)/2 if m ≡ 1 mod 2 ,
g=
(m − 2)/2 if m ≡ 0 mod 2 .

(c) Let F = K(x, y) with y 2 = f (x) as in (6.15). Then the places P ∈ IPK(x)
which ramify in F/K(x) are the following :
all zeros of f (x) , if deg f (x) ≡ 0 mod 2 ,
all zeros of f (x) and the pole of x , if deg f (x) ≡ 1 mod 2 .
Hence, if f (x) decomposes into linear factors, exactly 2g + 2 places of K(x)
are ramified in F/K(x).

Proof. (b) and (c) are special cases of Proposition 3.7.3 (cf. also Example
3.7.6).
(a) As F/K(x) is cyclic of degree 2 and char K = 2, there exists an element
z ∈ F such that F = K(x, z) and z 2 = u(x) ∈ K(x). Write

u(x) = c · pi (x)ri , 0 = c ∈ K ,

with pairwise distinct irreducible monic polynomials pi (x) ∈ K[x] and ri ∈ ZZ.
Let ri = 2si + εi , si ∈ ZZ and εi ∈ {0, 1}. Set

y := z · p−s
i
i
.

Then F = K(x, y) and y 2 = f (x) with a square-free polynomial f (x) ∈ K[x].

Now Example 3.7.6 implies that deg f = 2g + 1 or 2g + 2.
226 6 Examples of Algebraic Function Fields

In the case of char K = 2, all places of F which are ramified in the

quadratic extension F/K(x) are wildly ramified, so their different exponent in
Diff(F/K(x)) is at least 2 (by Dedekind’s Different Theorem or Proposition
3.7.8(c)). It follows that the number s of ramified places lies in the interval
1 ≤ s ≤ g + 1. One can easily construct examples for each s in this range. For
example, the hyperelliptic field F = K(x, y) with

y 2 + y = f (x) ∈ K[x], deg f (x) = 2g + 1 (6.16)

has genus g and exactly one ramified place in F/K(x) (the pole of x). On the
other hand, if

g+1
2
y +y = (x + ai )−1 (6.17)
i=1

with pairwise distinct ai ∈ K, the genus of F/K is g, and we have exactly

g +1 ramified places in F/K(x). All this follows immediately from Proposition
3.7.8.
Thus far the condition g ≥ 2 was not essential. Our previous results on
hyperelliptic function fields hold for elliptic function fields as well. However,
the next proposition is false in the case of an elliptic function field.
We recall that the space ΩF of differentials of F/K is a one-dimensional
F -module. Hence for ω1 , ω2 ∈ ΩF and ω2 = 0, the quotient ω1 /ω2 ∈ F is
defined. ΩF (0) = {ω ∈ ΩF | (ω) ≥ 0} is the space of regular differentials of
F/K.

Proposition 6.2.4. Consider a hyperelliptic function field F/K of genus g

and a rational subfield K(x) ⊆ F with [F : K(x)] = 2. Then the following
hold:
(a) All rational subfields K(z) ⊆ F with [F : K(z)] ≤ g are contained in K(x).
In particular, K(x) is the only rational subfield of F with [F : K(x)] = 2.
(b) K(x) is the subfield of F which is generated by the quotients of regular
differentials of F/K.

Proof. (a) Suppose that [F : K(z)] ≤ g but z ∈ K(x). Then F = K(x, z),
and Riemann’s Inequality (Theorem 3.11.4) yields the contradiction

g ≤ ([F : K(x)] − 1) · ([F : K(z)] − 1) ≤ g − 1 .

(b) First we claim that the divisor W := (g − 1) · (x)∞ ∈ Div(F ) is a canonical

divisor of F/K. This follows from Proposition 1.6.2 since deg W = 2g − 2
(obvious) and (W ) ≥ g (the elements 1, x, . . . , xg−1 are in L (W )). Choose a
differential ω ∈ ΩF with (ω) = W ; then the differentials xi ω , 0 ≤ i ≤ g −1 are
in ΩF (0). As ΩF (0) is a g-dimensional vector space over K (Remark 1.5.12),
this implies
 6.3 Tame Cyclic Extensions of the Rational Function Field 227

ΩF (0) = {f (x) · ω | f (x) ∈ K[x] and deg f (x) ≤ g − 1 } .

Thus K(x) is the subfield of F generated by the quotients of regular differen-

tials.

We mention without proof that if F/K is a non-hyperelliptic function field

of genus g ≥ 2, the quotients of regular differentials generate F (under a weak
additional assumption, e.g. the existence of a divisor of degree one), cf. [6].

6.3 Tame Cyclic Extensions of the Rational Function

Field
We study function fields F = K(x, y) which are defined by an equation

s
yn = a · pi (x)ni (6.18)
i=1

with s > 0 pairwise distinct irreducible monic polynomials pi (x) ∈ K[x],

0 = a ∈ K and 0 = ni ∈ ZZ. Throughout this section we will assume that the
following conditions hold:

char K  n, and gcd(n, ni ) = 1 for 1 ≤ i ≤ s . (6.19)

Note that hyperelliptic function fields of characteristic = 2 are special cases

of (6.18).

Proposition 6.3.1. Suppose that F = K(x, y) is defined by (6.18) and (6.19).

Then we have:
(a) K is the full constant field of F , and [F : K(x)] = n. If K contains a
primitive n-th root of unity, F/K(x) is a cyclic field extension.
(b) Let Pi (resp. P∞ ) denote the zero of pi (x) (resp. the pole of x) in K(x).
The places P1 , . . . , Ps are totally ramified in F/K(x). All places Q∞ ∈ IPF
with Q∞ | P∞ have ramification index e(Q∞ | P∞ ) = n/d where
  s 
d := gcd n, ni · deg pi (x) . (6.20)
i=1

No places P ∈ IPK(x) other than P1 , . . . , Ps , P∞ ramify in F/K(x).

(c) The genus of F/K is
 s 
n−1 d−1
g= −1 + deg pi (x) − ,
2 i=1
2

with d as in (6.20).
228 6 Examples of Algebraic Function Fields

Proof. All assertions follow immediately from Proposition 3.7.3, Corollary

3.7.4 and Remark 3.7.5.

Now we consider some special cases of Proposition 6.3.1.

Example 6.3.2. Let F = K(x, y) with

y n = (xm − b)/(xm − c) ,

where b, c ∈ K\{0}, b = c and char K  mn. Then (6.18) and (6.19) hold, and
we obtain from Proposition 6.3.1
  
g = (n − 1)(m − 1) = [F : K(x)] − 1 [F : K(y)] − 1 .

Thus Riemann’s Inequality (Corollary 3.11.4) is sharp in this case.

Example 6.3.3. The function field F = K(x, y) with defining equation

axm + by n = c, a, b, c ∈ K\{0} , char K  mn

has genus
1 
g= (n − 1)(m − 1) + 1 − gcd(m, n) .
2
Example 6.3.4. A function field F = K(x, y) with

axn + by n = c, a, b, c ∈ K\{0}, char K  n

is said to be of Fermat type. Its genus is g = (n − 1)(n − 2)/2 by the previous

example. This shows that the estimate for the genus given in Proposition
3.11.5 cannot be improved in general.
Example 6.3.5. Let K = IFq2 be the finite field of cardinality q 2 , where q is a
power of a prime number. Consider the function field F = K(x, y) with

axq+1 + by n = c , a, b, c ∈ IFq \{0} , n | (q + 1) . (6.21)

We want to determine the number of rational places

 
N = N (F/IFq2 ) =  {P ∈ IPF ; deg P = 1}  .

First we substitute x1 := γx, y1 := δy with γ q+1 = a/c and δ n = −b/c, and

we obtain F = K(x1 , y1 ) with y1n = xq+1
1 − 1 (observe that γ, δ ∈ IFq2 since all
elements of IFq are (q + 1)-th powers of elements of IFq2 ). So we can assume
from the beginning that F = K(x, y) with

y n = xq+1 − 1 and n | (q + 1) . (6.22)

Let Pα ∈ IPK(x) (resp. P∞ ) denote the zero of x − α (resp. the pole of x)

in K(x). Each place P ∈ IPF of degree one lies over P∞ or some Pα (with
α ∈ K), hence we have to study the decomposition of Pα and P∞ in F/K(x).
 6.3 Tame Cyclic Extensions of the Rational Function Field 229

Case 1. α ∈ K and αq+1 = 1. In this case α is a simple root of the poly-

nomial T q+1 − 1 ∈ K[T ], and Pα is fully ramified in F/K(x) by Proposition
6.3.1(b). So Pα has a unique extension P ∈ IPF , and deg P = 1.
Case 2. α ∈ K and αq+1 = 1. We use Kummer’s Theorem (resp. Corol-
lary 3.3.8) to determine the decomposition of Pα in F/K(x). The minimal
polynomial of y over K(x) is ϕ(T ) = T n − (xq+1 − 1) ∈ K(x)[T ], and
ϕα (T ) := T n − (αq+1 − 1) ∈ K[T ]
has n distinct roots β ∈ K = IFq2 (here we use that αq+1 − 1 ∈ IFq \ {0} and
n | (q + 1)). For any such β there is a unique place Pα,β ∈ IPF with Pα,β | Pα
and y − β ∈ Pα,β , and Pα,β is of degree one. So Pα has n distinct extensions
P ∈ IPF with deg P = 1.
Case 3. α = ∞. In this case Kummer’s Theorem does not apply directly
as not all coefficients of the minimal polynomial of y over K(x) are in the
valuation ring O∞ of P∞ . So we consider the element z := y/x(q+1)/n which
satisfies the equation
z n = 1 − (1/x)q+1 .
As T n − 1 has n distinct roots in K, we see that P∞ has n distinct extensions
P ∈ IPF , all of degree one, by Kummer’s Theorem.
There are q + 1 elements α ∈ IFq2 belonging to Case 1, and q 2 − (q + 1)
elements α that fall under Case 2. Summing up we find that F/IFq2 has
N = (q + 1) + n(q 2 − (q + 1)) + n = q + 1 + n(q 2 − q)
places of degree one. By Example 6.3.3 the genus of F is g = (n − 1)(q − 1)/2,
hence
q 2 + 1 + 2gq = q 2 + 1 + q(n − 1)(q − 1) = q + 1 + n(q 2 − q) .
We see that the function fields F/IFq2 which are defined by (6.21) are maximal;
i.e., they attain the upper Hasse-Weil Bound
N = q 2 + 1 + 2gq (6.23)
(over the constant field IFq2 ). Now one can easily determine the L-polynomial
LF (t) of F/IFq2 : if α1 , . . . , α2g ∈ C are the reciprocals of the roots of LF (t)
then

2g
N = q2 + 1 − αi , (6.24)
i=1

by Corollary 5.1.16. On the other hand, |αi | = q by the Hasse-Weil Theorem.

By (6.23) and (6.24) this implies αi = −q for i = 1, . . . , 2g, and thus
LF (t) = (1 + qt)2g . (6.25)
The above proof shows that Equation (6.25) holds for all maximal function
fields over IFq2 .
230 6 Examples of Algebraic Function Fields

Example 6.3.6. The special case H := IFq2 (x, y) with

xq+1 + y q+1 = 1 (6.26)

is called the Hermitian function field over IFq2 . It is a maximal function field
by (6.23), so it provides an example of a maximal function field of genus
g = q(q − 1)/2 and shows that Proposition 5.3.3 cannot be improved. The
number of places of degree one is N = 1 + q 3 .

The Hermitian function field has some other remarkable properties. For in-
stance, its automorphism group Aut(H/IFq2 ) is very large, cf. [41] and Exercise
6.10. In Section 6.4 we shall give a second description of the Hermitian func-
tion field.

Remark 6.3.7. Once again we consider a function field F = IFq2 (u, v) as in

Example 6.3.5; i.e., auq+1 + bv n = c with a, b, c ∈ IFq \ {0} and n | (q + 1). By
(6.22) we can replace u, v by elements w, t such that F = IFq2 (t, w) and

wn = tq+1 − 1 . (6.27)

On the other hand, let H = IFq2 (x, y) be the Hermitian function field given
by xq+1 + y q+1 = 1. Write q + 1 = sn, choose ζ ∈ IFq2 with ζ n = −1 and set
z := ζy s ∈ H. Then
z n = ζ n y q+1 = xq+1 − 1 .
Thus F = IFq2 (u, v) is isomorphic to the subfield IFq2 (x, z) ⊆ H. In other
words, all function fields which were considered in Example 6.3.5 can be re-
garded as subfields of the Hermitian function field H. More generally, it is
easily seen that the function fields F = IFq2 (u, v) with

aun + bv m = c , a, b, c ∈ IFq \ {0} , m | (q + 1) and n | (q + 1) (6.28)

can be regarded as subfields of H. One can show by direct computation that

all function fields (6.28) are maximal.
Example 6.3.8. We consider the function field F = K(y, z) defined by

z3 + y3 z + y = 0 . (6.29)

F is called the function field of the Klein Quartic. The polynomial T 3 + y 3 T +

y ∈ K(y)[T ] is absolutely irreducible (by Proposition 3.1.15), so K is the full
constant field of F (see Corollary 3.6.8), and [F : K(y)] = 3.
It is convenient to choose other generators of F/K. We multiply (6.29) by
y 6 , set x := −y 2 z and obtain F = K(x, y) with

y 7 = x3 /(1 − x) . (6.30)

If char K = 7, F/K(x) is purely inseparable; therefore F/K is rational in this

case (Proposition 3.10.2(c)). In case char K = 7 we can apply Proposition
 6.3 Tame Cyclic Extensions of the Rational Function Field 231

6.3.1. Exactly three places of K(x) ramify in F/K(x), namely the pole P∞ of
x, the zero P0 of x and the zero P1 of x − 1. All these places have ramification
index e = 7 in the extension F/K(x), and the genus of F/K is g = 3.
Now we specialize to K = IF2 . As in Chapter 5, Nr denotes the number
of places of degree one in the constant field extension Fr = F IF2r . We claim
that
N1 = 3 , N2 = 5 and N3 = 24 . (6.31)
N1 = 3 is obvious since the three places of IF2 (x) of degree one are fully
ramified in F , so each of them has a unique extension of degree one in IPF .
For r = 2 the constant field is IF4 = {0, 1, α, α + 1} where α2 + α + 1 = 0. Let
Pγ ∈ IPK(x) be the zero of x − γ. We determine the decomposition of Pγ in
IF4 (x, y)/IF4 (x) for γ ∈ {α, α + 1}. In order to apply Kummer’s Theorem we
have to study the polynomials

α3
ϕα (T ) = T 7 + = T 7 + α and
1+α
(α + 1)3
ϕα+1 (T ) = T 7 + = T7 + α + 1.
1 + (1 + α)

Both polynomials have only simple irreducible factors in IF4 [T ] as they are
relatively prime to their derivative, and α (resp. α + 1) is the only root of
ϕα (T ) (resp. ϕα+1 (T )) in IF4 . Hence there is exactly one place of degree one
lying over Pα (resp. Pα+1 ), the other extensions of Pα and Pα+1 are of degree
> 1. Summing up, we have found exactly 5 places of IF4 (x, y)/IF4 of degree
one, so N2 = 5.
Next we consider the constant field IF8 = IF2 (β) where β 3 + β + 1 = 0.
One has to study the decomposition of the polynomials

γ3
ϕγ (T ) = T 7 + ∈ IF8 [T ]
1+γ

for γ ∈ IF8 \ {0, 1}. For γ ∈ {β, β 2 , β 4 } we have ϕγ (T ) = T 7 + 1 which

decomposes into seven distinct linear factors in IF8 [T ]. For γ ∈ {β 3 , β 5 , β 6 },
ϕγ (T ) has no root in IF8 . Hence N3 = 3 + 3 · 7 = 24, and the proof of (6.31)
is finished.

Now it is easy to determine the L-polynomial of the Klein Quartic over

IF2 . With the notations as in Corollary 5.1.17 we find S1 = S2 = 0 and
S3 = 24 − (8 + 1) = 15, hence a0 = 1, a1 = a2 = 0, a3 = 5, a4 = a5 = 0 and
a6 = 8. Thus
LF (t) = 1 + 5t3 + 8t6 .
The Klein Quartic over IF8 provides an example where the upper Serre Bound
N = q + 1 +√g · [2q 1/2 ] is attained (Theorem 5.3.1), since we have N = 24 =
8 + 1 + 3 · [2 8].
232 6 Examples of Algebraic Function Fields

6.4 Some Elementary Abelian p-Extensions of K(x),

char K = p > 0

In this section K is a field of characteristic p > 0.

The function fields to be discussed in this section have interesting applications

in coding theory, cf. Chapter 8 and 9.

Proposition 6.4.1. Consider a function field F = K(x, y) with

y q + μy = f (x) ∈ K[x] , (6.32)

where q = ps > 1 is a power of p and 0 = μ ∈ K. Assume that deg f =: m > 0

is prime to p, and that all roots of T q + μT = 0 are in K. Then the following
hold:
(a) [F : K(x)] = q, and K is the full constant field of F .
(b) F/K(x) is Galois. The set A := {γ ∈ K | γ q + μγ = 0} is a subgroup of
order q of the additive group of K. For all σ ∈ Gal(F/K(x)) there is a unique
γ ∈ A such that σ(y) = y + γ, and the map

Gal(F/K(x)) −→ A ,
σ −→ γ

is an isomorphism of Gal(F/K(x)) onto A.

(c) The pole P∞ ∈ IPK(x) of x in K(x) has a unique extension Q∞ ∈ IPF ,
and Q∞ |P∞ is totally ramified (i.e., e(Q∞ /P∞ ) = q). Hence Q∞ is a place
of F/K of degree one.
(d) P∞ is the only place of K(x) which ramifies in F/K(x).
(e) The genus of F/K is g = (q − 1)(m − 1)/2.
(f ) The divisor of the differential dx is
 
(dx) = (2g − 2)Q∞ = (q − 1)(m − 1) − 2 Q∞ .

(g) The pole divisor of x (resp. y) is (x)∞ = qQ∞ (resp. (y)∞ = mQ∞ ).
(h) Let r ≥ 0. Then the elements xi y j with

0 ≤ i, 0 ≤ j ≤ q − 1, qi + mj ≤ r

form a basis of the space L (rQ∞ ) over K.

(i) For all α ∈ K one of the following cases holds:
Case 1. The equation T q + μT = f (α) has q distinct roots in K.
Case 2. The equation T q + μT = f (α) has no root in K.
 6.4 Some Elementary Abelian p-Extensions of K(x), char K = p > 0 233

In Case 1, for each β with β q + μβ = f (α) there exists a unique place Pα,β ∈
IPF such that Pα,β |Pα and y(Pα,β ) = β. Hence Pα has q distinct extensions
in F/K(x), each of degree one.
In Case 2, all extensions of Pα in F have degree > 1.
Proof. Equation (6.32) is a special case of the situation which was considered
in Proposition 3.7.10, so (a) - (e) hold.
(g) (x)∞ = qQ∞ follows from (c). The elements x and y have the same poles,
hence Q∞ is the only pole of y as well. Since q · vQ∞ (y) = vQ∞ (y q + y) =
vQ∞ (f (x)) = −mq we obtain (y)∞ = mQ∞ .
(f) The different of F/K(x) is Diff(F/K(x)) = (q − 1)(m + 1)Q∞ , by
Proposition 3.7.10(d). So Remark 4.3.7(c) yields
 
(dx) = −2(x)∞ + Diff(F/K(x)) = (q − 1)(m − 1) − 2 Q∞ = (2g − 2)Q∞ .

(h) The elements 1, y, . . . , y q−1 form an integral basis of F/K(x) at all places
P ∈ IPK(x) different from P∞ . This follows from Theorem 3.5.10(b), as the
minimal polynomial ϕ(T ) = T q + μT − f (x) of y over K(x) is in OP [T ] and
for all Q|P ,
vQ (ϕ (y)) = vQ (μ) = 0 = d(Q|P ) .
Let z ∈ L (rQ∞ ). As Q∞ is the only pole of z, z is integral over OP for all
q−1
P ∈ IPK(x) , P = P∞ , thus z = j=0 zj y j with zj ∈ K(x), and zj has no
poles other than P∞ . Hence zj is a polynomial in K[x]; i.e.,

q−1 
z= aij xi y j with aij ∈ K . (6.33)
j=0 i≥0

The elements xi y j with 0 ≤ j ≤ q − 1 have pairwise distinct pole orders

because vQ∞ (x) = −q, vQ∞ (y) = −m and m and q are relatively prime.
Therefore the Strict Triangle Inequality implies
vQ∞ (z) = min {−iq − jm | aij = 0} .
This proves (h).
(i) Suppose there is some β ∈ K such that β q + μβ = f (α). It follows that
(β + γ)q + μ(β + γ) = f (α) for all γ with γ q + μγ = 0, so

q
T + μT − f (α) =
q
(T − βj )
j=1

with pairwise distinct elements βj ∈ K. By Corollary 3.3.8(c) there exists for

j = 1, . . . , q, a unique place Pj ∈ IPF such that Pj |Pα and y − βj ∈ Pj , and
the degree of Pj is one.
In Case 2 the polynomial T q + μT − f (α) ∈ K[T ] splits into pairwise
distinct irreducible factors of degree > 1. By Corollary 3.3.8(a), all places
P ∈ IPF with P |Pα have degree > 1.
234 6 Examples of Algebraic Function Fields

Example 6.4.2. We consider a special case of the previous proposition, namely

F = IFq2 (x, y) with y q + y = xm and m|(q + 1) . (6.34)

The genus of F is g = (q − 1)(m − 1)/2. We claim that F/IFq2 has

 
N = 1 + q 1 + (q − 1)m (6.35)

places of degree one. The pole Q∞ of x is one of them. The other places of
degree one are extensions of some place Pα ∈ IPK(x) . Hence, by Proposition
6.4.1(i), we have to count the elements α ∈ IFq2 such that the equation

T q + T = αm (6.36)

has a root β ∈ IFq2 . The map β → β q + β is the trace mapping from IFq2 to
IFq , and therefore it is surjective (cf. Appendix A). Thus (6.36) has a root in
IFq2 if and only if αm ∈ IFq . Let U ⊆ IF∗q2 be the subgroup of order (q − 1)m
(here we use the assumption m|(q + 1)). Then for α ∈ IFq2 ,

αm ∈ IFq ⇐⇒ α ∈ U ∪ {0} .

Hence N = 1 + q((q − 1)m + 1) by Proposition 6.4.1(i). This proves (6.35).

Because 1 + q((q − 1)m + 1) = 1 + q 2 + 2gq, the fields which are defined
by (6.34) provide other examples of maximal function fields over IFq2 .

Example 6.4.3. The Hermitian function field H which was studied in Example
6.3.6 is given by

H = IFq2 (u, v) with uq+1 + v q+1 = 1 . (6.37)

We choose a, b, c ∈ IFq2 such that

aq+1 = −1 , bq + b = 1 and c = −abq ;

then it follows that

abq + c = 0 ,
aq b + cq = (abq + c)q = 0 ,
acq + aq c = a(−aq b) + aq (−abq ) = −aq+1 (b + bq ) = 1 . (6.38)

We set
1 bu + cv
x= and y= .
u + av u + av
Then H = IFq2 (x, y), and we obtain

(u + av)q+1 · xq+1 = 1 (6.39)

 6.5 Exercises 235

and

(u + av)q+1 · (y q + y)
= (u + av)(bu + cv)q + (u + av)q (bu + cv)
= (bq + b)uq+1 + (bq a + c)uq v + (cq + baq )uv q + (acq + aq c)v q+1
= uq+1 + v q+1 = 1 (6.40)

(we have used (6.38)). Comparing Equations (6.39) and (6.40) we see that

H = IFq2 (x, y) and y q + y = xq+1 .

So the Hermitian function field H can be regarded as a special case of the

function fields which were considered in Example 6.4.2. This representation of
H is particularly useful, because we have a simple explicit description of the
canonical divisor W = (dx), the spaces L (rQ∞ ) and all places of degree one
(by Proposition 6.4.1 and Example 6.4.2). For applications to coding theory
in Section 8.3 we put together these results in a lemma.

Lemma 6.4.4. The Hermitian function field over IFq2 can be defined by

H = IFq2 (x, y) with y q + y = xq+1 . (6.41)

It has the following properties:

(a) The genus of H is g = q(q − 1)/2.
(b) H has q 3 + 1 places of degree one over IFq2 , namely
(1) the common pole Q∞ of x and y, and
(2) for each α ∈ IFq2 there are q elements β ∈ IFq2 such that β q +β = αq+1 ,
and for all such pairs (α, β) there is a unique place Pα,β ∈ IPH of degree
one with x(Pα,β ) = α and y(Pα,β ) = β.

(c) H/IFq2 is a maximal function field.

(d) The divisor of the differential dx is (dx) = (q(q − 1) − 2)Q∞ .
(e) For r ≥ 0, the elements xi y j with 0 ≤ i, 0 ≤ j ≤ q −1 and iq +j(q +1) ≤ r
form a basis of L (rQ∞ ).

Remark 6.4.5. One can show that the Hermitian function field is - up to iso-
morphism - the only maximal function field over IFq2 of genus g = q(q − 1)/2,
see [34].

6.5 Exercises
6.1. Consider the function field F = IF2 (x, y) over IF2 which is defined by
the equation y 2 = f (x). For each of the following choices of f (x) ∈ IF2 (x),
determine the L-polynomial L(t).
236 6 Examples of Algebraic Function Fields

(i) f (x) = x3 + 1,
(ii) f (x) = x3 + x,
(iii) f (x) = x3 + x + 1,
(iv) f (x) = (x2 + x)/(x3 + x + 1).

6.2. Let F = IF3 (x, y) be the elliptic function field over IF3 which is defined
by the equation y 2 = x3 − x. Determine Nr for all r ≥ 1.

6.3. Show that there exists an elliptic function field over F/IF5 having 10
rational places, cf. Exercise 5.1. Is F unique (up to isomorphism)?

6.4. Construct a function field of genus g = 2 over IF3 with N = 8 rational

places.

6.5. Consider a function field E = IF2 (x, y, z) and its subfield F = IF2 (x, y),
where x, y, z satisfy the equations

y 2 x + y + x2 + 1 = 0 and z2 y + z + y2 + 1 = 0 .

(i) Show that [E : F ] = [F : IF2 (x)] = 2, and that the pole of x is totally
ramified in E/IF2 (x).
(ii) Determine the genus and the number of rational places of F/IF2 . Compare
with Example 6.1.5.
(iii) Determine the genus and the number of rational places of E/IF2 .

6.6. (p prime, p ≡ 1 mod 4) Consider the function field F = IFp (x, y) with
the defining equation
y p − y = xp+1 .
It is clear from Proposition 6.4.1 that F/IFp (x) is a Galois extension of degree
[F : IFp (x)] = p, and that x has a unique pole Q∞ in F .
(i) Show that the Riemann-Roch space L (p Q∞ ) has dimension 2 and is
generated by the elements 1 and x.
(ii) Determine all rational places of F/IFp .
(iii) Show that the automorphism group Aut (F/IFp ) acts transitively on the
places of degree one.
Hint. Choose α ∈ IFp with α2 = −1 (only here the assumption p ≡ 1 mod 4 is
used). Show that there is an automorphism σ ∈ Aut (F/IFp ) with σ(y) = 1/y
and σ(x) = αx/y. This automorphism permutes the zero and the pole of y.

6.7. (p prime, p ≡ 1 mod 4) Consider the function fields E = IFp (s, t) with
the defining equation
tp + t = sp+1 ,
and F = IFp (x, y) with
y p − y = xp+1
 6.5 Exercises 237

(as in the previous exercise).

(i) Show that the extension E/IFp (s) is not Galois, and determine all rational
places of E/IFp .
(ii) It follows that E and F have the same number of rational places, namely
N = p + 1. Show that E and F are not isomorphic.
(iii) Consider the constant field extensions E2 := EIFp2 and F2 := F IFp2 of
degree 2. Show that E2 is isomorphic to F2 .

The subsequent exercises are related to the automorphism group Aut (F/K) of
a function field F/K. Recall that we have already shown the following facts:
(1) The automorphism group of a rational function field K(x)/K is iso-
morphic to PGL2 (K) (see Exercise 1.2.)
(2) If K is a finite field, then the automorphism group of F/K is finite
(see Exercise 5.4)
(3) Let K be an algebraically closed field and G a finite subgroup of
Aut (F/K). Assume that the order of G is not divisible by the charac-
teristic of K and that the genus of F/K is g ≥ 2. Then the order of G
satisfies the bound ord G ≤ 84(g − 1) (see Exercise 3.18).
We mention without proof that Aut (F/K) is always a finite group if K is
algebraically closed and the genus of F is ≥ 2. Most proofs of this fact use the
theory of Weierstrass points.
6.8. Let F/K be a function field with exact constant field K, and consider
the constant field extension F L/L with some algebraic extension field L ⊇ K.
(i) Let σ be an automorphism of F/K. Show that there is a unique automor-
phism σ̃ of F L/L whose restriction to F is σ.
(ii) Let Aut (F/K) be the group of automorphisms of F over K. With no-
tation as in (i), show that the map σ → σ̃ is a monomorphism of Aut (F/K)
into Aut (F L/L).
As a consequence of (ii), one can consider the automorphism group Aut (F/K)
as a subgroup of Aut (F̄ /K̄), where F̄ is the constant field extension of F with
the algebraic closure K̄ of K.

6.9. Consider the rational function field F := IFq (x) and its automorphism
group G := Aut (F/IFq ). We know that G  PGL2 (IFq ), see Exercise 1.2.
(i) Determine the order of G.
(ii) Let U := {σ ∈ G | σ(x) = x + c with c ∈ IFq }. Show that U is a p-Sylow
subgroup of G (with p := char IFq ). Find an element z ∈ F such that the
fixed field of U is F U = IFq (z). Describe all ramified places, their ramification
indices and different exponents in F/F U .
(iii) Let V := {σ ∈ G | σ(x) = ax + c with a, c ∈ IFq and a = 0}. Find an
element v ∈ F such that the fixed field of V is F V = IFq (v). Describe all
ramified places, their ramification indices and different exponents in F/F V .
238 6 Examples of Algebraic Function Fields

(iv) Show that exactly 2 places of F G are ramified in F/F G , and both of
them are rational places of F G . Determine their relative degrees, ramification
indices and different exponents in F/F G . Show that all places of F of degree
2 are conjugate under the group G.
(v) Find an element t ∈ F such that the fixed field of G is F G = IFq (t).
6.10. Let K = IFq2 and consider the Hermitian function field H = K(x, y)
with defining equation
y q + y = xq+1 ,
cf. Lemma 6.4.4. The element x has a unique pole in H that will be denoted
by Q∞ .
(i) Show that for each pair (d, e) ∈ K × K with eq + e = dq+1 there is an
automorphism σ ∈ Aut (H/K) with σ(x) = x + d and σ(y) = y + dq x + e.
These automorphisms form a subgroup V ⊆ Aut (H/K) of order q 3 .
(ii) Show that for each element c ∈ K × there is an automorphism τ ∈
Aut (H/K) with τ (x) = cx and τ (y) = cq+1 y. These automorphisms form a
cyclic subgroup W ⊆ Aut (H/K) of order q 2 − 1.
(iii) Let U ⊆ Aut (H/K) be the group which is generated by V and W .
Prove:
(a) ord U = q 3 (q 2 − 1), and V is a normal subgroup of U .
(b) For every ρ ∈ U holds ρ(Q∞ ) = Q∞ .
(c) The group U acts transitively on the set S := {Q | Q is a rational place
of H/K and Q = Q∞ }.
(iv) Show that every automorphism λ ∈ Aut (H/K) with λ(Q∞ ) = Q∞ lies
in U .
Hint. Observe that the elements 1, x, y form a K-basis of L ((q + 1)Q∞ ), by
Lemma 6.4.4.
(v) Show that there is an automorphism μ ∈ Aut (H/K) with μ(x) = x/y
and μ(y) = 1/y. This automorphism maps the place Q∞ to the common zero
of x and y.
(vi) Let G ⊆ Aut (H/K) be the group which is generated by U and μ. Prove:
(a) G acts transitively on the set of all rational places of H/K.
(b) G = Aut (H/K) and ord G = q 3 (q 3 + 1)(q 2 − 1).
(c) If g = g(H) denotes the genus of H/K, then ord G > 16g 4 > 84(g − 1).
6.11. We assume for simplicity that K is an algebraically closed field. Let
F/K be a hyperelliptic function field of genus g ≥ 2, and let K(x) ⊆ F be its
unique rational subfield with [F : K(x)] = 2 (cf. Proposition 6.2.4). We set
S := {P ∈ IPF | P is ramified in F/K(x)}.
(i) Recall that |S| = 2g + 2 if char K = 2, and 1 ≤ |S| ≤ g + 1 if char K = 2.
(ii) Show that Aut (F/K) acts on S (i.e., if σ ∈ Aut (F/K) and P ∈ S, then
σ(P ) ∈ S).
 6.5 Exercises 239

(iii) Fix a place P0 ∈ S and consider U := {σ ∈ Aut (F/K) | σ(P0 ) = P0 }.

Show that U is a subgroup of Aut (F/K) of finite index.
(iv) Show that the subgroup U as above is a finite group.
(v) Conclude that the automorphism group of a hyperelliptic function field
(over an arbitrary perfect constant field) is finite.

6.12. Let F/K be a function field over an algebraically closed field. For a
place P ∈ IPF we consider its Weierstrass semigroup

W (P ) := {r ∈ IN0 | r is a pole number of P } ,

see Definition 1.6.7. Assume now that F/K is hyperelliptic of genus g, and
K(x) ⊆ F is the unique rational subfield of F with [F : K(x)] = 2. Let
S ⊆ IPF be the set of ramified places in F/K(x) (as in the previous exercise).
Prove:
(i) If P ∈ S, then W (P ) = {0, 2, 4, . . . , 2g − 2, 2g, 2g + 1, 2g + 2, . . .} =
IN0 \ {1, 3, . . . , 2g − 1}.
(ii) If P ∈/ S, then W (P ) = {0, g + 1, g + 2, . . .} = IN0 \ {1, 2, . . . , g}.
These results show that the Weierstrass points (see Remark 1.6.9) of a hy-
perelliptic function field F/K are exactly the places which are ramified in the
extension F/K(x), where K(x) ⊆ F is the unique rational subfield of degree
[F : K(x)] = 2.

6.13. Let K be a perfect field of characteristic = 2 and F = K(x, y) a hyper-

elliptic function field of genus g ≥ 2 with the defining equation


2g+1
y2 = (x − ai ) ,
i=1

with pairwise distinct elements a1 , . . . , a2g+1 ∈ K. Let Pi ∈ IPF be the unique

zero of x − ai (i = 1, . . . , 2g + 1), and P∞ ∈ IPF the unique pole of x in F . Let
Ai := Pi −P∞ and [Ai ] ∈ Cl(F ) be the divisor class of Ai (cf. Definition 1.4.3).
We study the subgroup of the divisor class group of F/K which is generated
by the classes [A1 ], [A2 ], . . . , [A2g+1 ]. Show:
(i) [Ai ] = [0] and 2[Ai ] = [0], for i = 1, . . . , 2g + 1.
2g+1
(ii) i=1 [Ai ] = [0].
(iii) Let M ⊆ Cl(F ) be the subgroup of the divisor class group of F/K which
is generated by [A1 ], [A2 ], . . . , [A2g+1 ]. Then M  (ZZ/2ZZ)2g .
(iv) All divisor classes [A] ∈ Cl(F ) with 2[A] = [0] (these are called 2-division
classes of F ) are in M . Hence the number of 2-division classes of F is 22g .

Remark. The previous exercise is a special case of the following much more
general result. Let F/K be a function field of genus g over an algebraically
closed field K. For n ≥ 1 consider the group
240 6 Examples of Algebraic Function Fields

Cl(F )(n) := {[A] ∈ Cl(F ) | n[A] = [0]} .

If n is relatively prime to the characteristic of K, then Cl(F )(n)  (ZZ/nZZ)2g .

6.14. Let E/K be an elliptic function field. For simplicity we assume that K is
algebraically closed. We are going to study the structure of the automorphism
group Aut (E/K).
Fix a place P0 ∈ IPE ; then IPE carries the structure of an abelian group (see
Proposition 6.1.7). The addition ⊕ on IPE is given by

P ⊕ Q = R ⇐⇒ P + Q ∼ R + P0

for P, Q, R ∈ IPE (∼ means equivalence of divisors modulo principal divisors).

For P ∈ IPE it follows from Riemann-Roch that (P + P0 ) = 2, hence there
exists an element x ∈ E whose pole divisor is (x)∞ = P + P0 . We define the
automorphism σP ∈ Aut (E/K) as

σP := the non-trivial automorphism of E/K(x) .

We also define
τP := σP ◦ σP0 .
Observe that the definion of σP and τP depends on the choice of the place
P0 . Prove:
(i) The automorphisms σP and τP are well-defined (i.e., they do not depend
on the specific choice of the element x above).
(ii) For P = Q we have σP = σQ and τP = τQ . In particular, Aut (E/K) is
an infinite group.
(iii) For all P, Q ∈ IPE holds σP (Q) ⊕ Q = P and τP (Q) = P ⊕ Q. Hence τP
is called a translation automorphism.
(iv) The map P → τP is a group monomorphism from IPE into Aut (E/K).
Its image T := {τP | P ∈ IPE } ⊆ Aut (E/K) is isomorphic to the group of
divisor classes Cl0 (E) and hence an infinite abelian subgroup of Aut (E/K).
It is called the translation group of E/K.
(v) The translation group T is independent of the choice of the place P0
(which was used for the definition of the group structure on IPE ).
(vi) T is a normal subgroup of Aut (E/K), and the factor group Aut (E/K)/ T
is finite.

6.15. (char K = p > 0) Consider the rational function field K(x)/K and the
element y := x − x−p . Show:
(i) The extension K(x)/K(y) is separable of degree [K(x) : K(y)] = p + 1.
(ii) The only place of K(y) which is ramified in K(x)/K(y), is the pole P∞
of y. There are exactly 2 places of K(x) lying over P∞ .
 6.5 Exercises 241

6.16. (char K = p > 0) Let F/K be a function field over a perfect constant
field K of characteristic p > 0. Using the previous exercise, show that there
exists an element y ∈ F \ K with the following properties:
(i) The extension F/K(y) is separable.
(ii) The pole of y is the only place of K(y) which is ramified in F/K(y).
Compare this result with Exercise 3.6 (ii).
7
Asymptotic Bounds for the Number
of Rational Places

Let F/IFq be a function field over a finite field IFq . We have seen in Chapter
5 that the number N of rational places of F over IFq satisfies the Hasse-
Weil Bound N ≤ q + 1 + 2gq 1/2 , and that this upper bound can be attained
only if g ≤ (q − q 1/2 )/2. Here our aim is to investigate what happens if the
genus is large with respect to q. The results of this chapter have interesting
applications in coding theory, see Section 8.4.
In this chapter we consider function fields F over the finite field IFq .
The number of rational places of F/IFq is denoted by N = N (F ).

7.1 Ihara’s Constant A(q)

In order to describe how many rational places a function field over IFq can
have, we introduce the following notation.

Definition 7.1.1. (a) For an integer g ≥ 0 let

Nq (g) := max {N (F ) | F is a function field over IFq of genus g}.

(b) The real number

A(q) := lim sup Nq (g)/g
g→∞

is called Ihara’s constant.

Remark 7.1.2. Note that Nq (g) ≤ q + 1 + g[2q 1/2 ] by Serre’s Bound (Theorem
5.3.1). Therefore we have the trivial bound 0 ≤ A(q) ≤ [2q 1/2 ].

Our first aim is to improve the estimate A(q) ≤ [2q 1/2 ]. The bound given
in Theorem 7.1.3 below, is based on Serre’s explicit formulas, cf. Proposition
5.3.4. For the convenience of the reader we recall this method briefly:
H. Stichtenoth, Algebraic Function Fields and Codes, 243
Graduate Texts in Mathematics 254,

c Springer-Verlag Berlin Heidelberg 2009
244 7 Asymptotic Bounds for the Number of Rational Places

Let c1 , . . . , cm ≥ 0 be non-negative real numbers, not all of them equal to 0.

We define the functions

m
λm (t) = cr tr and fm (t) = 1 + λm (t) + λm (t−1 )
r=1

for t ∈ C \ {0} , and we assume that

fm (t) ≥ 0 for all t ∈ C with |t| = 1. (7.1)

(Note that fm (t) ∈ IR holds for all t ∈ C with |t| = 1.) Then the number of
rational places of each function field F/IFq of genus g is bounded by

g λm (q 1/2 )
N ≤ + + 1. (7.2)
λm (q −1/2 ) λm (q −1/2 )

Theorem 7.1.3 (Drinfeld-Vladut Bound). Ihara’s constant A(q) is bounded

above by
A(q) ≤ q 1/2 − 1 .
Proof. With notation as above we set, for a fixed integer m ≥ 1,
r
cr := 1 − for r = 1, . . . , m .
m
Then we have
m 
 
r r
λm (t) = 1− t .
r=1
m
In order to verify property (7.1) for the function fm (t) = 1 + λm (t) + λm (t−1 ),
we consider the function

m
tm+1 − t
u(t) := tr = .
r=1
t−1
m
We have u (t) = r=1 rtr−1 and hence

t · u (t)  r r
m
= t ,
m r=1
m

therefore
m 
 
r r t · u (t)
λm (t) = 1− t = u(t) −
r=1
m m
t t (t − 1)((m + 1)tm − 1) − (tm+1 − t)
= (tm − 1) − ·
t−1 m (t − 1)2
 m 
t t −1
= +1−t . (7.3)
(t − 1)2 m
 7.1 Ihara’s Constant A(q) 245

A straightforward calculation shows now that the function fm (t) = 1+λm (t)+
λm (t−1 ) can be written as

2 − (tm + t−m )
fm (t) = . (7.4)
m(t − 1)(t−1 − 1)

Since t−1 = t̄ for |t| = 1, Equation (7.4) yields fm (t) ≥ 0 for all t ∈ C with
|t| = 1, so the function fm (t) satisfies (7.1). Now we obtain from (7.2) the
inequality  
N 1 1 λm (q 1/2 )
≤ + 1+ , (7.5)
g λm (q −1/2 ) g λm (q −1/2 )
where N is the number of rational places of an arbitrary function field over
IFq of genus g. Equation (7.3) implies, for m → ∞,

q −1/2 1
λm (q −1/2 ) −→ (1 − q −1/2 ) = 1/2 .
(q −1/2 − 1)2 q −1
Hence for each ε > 0 there exists m0 ∈ IN with

λm0 (q −1/2 )−1 < q 1/2 − 1 + ε/2 .

We choose g0 such that

 
1 λm0 (q 1/2 )
1+ < ε/2 ,
g0 λm0 (q −1/2 )
and then, by using (7.5), we obtain the estimate
N
< q 1/2 − 1 + ε ,
g
for all g ≥ g0 and all function fields F over IFq of genus g. This finishes the
proof of Theorem 7.1.3.

Remark 7.1.4. Here are some facts about Ihara’s constant A(q).
(a) A(q) > 0 for all prime powers q = pe (p a prime number, e ≥ 1). More
precisely, there exists a constant c > 0 such that A(q) ≥ c · log q for all q.
This result is due to Serre [36], the proof uses class field theory and cannot be
given within the scope of this book. For refinements of Serre’s approach we
refer to the book [31]. In Section 7.3 we will give a simple proof that A(q) > 0
for all q = pe with e > 1.
(b) If q = 2 is a square then A(q) = q 1/2 − 1; i.e., in this case the Drinfeld-
Vladut Bound is attained. This equality was first proved by Ihara [22] and
Tsfasman, Vladut and Zink [44], using the theory of modular curves. We will
present a more elementary approach due to Garcia and Stichtenoth [12] in
Section 7.4 below.
246 7 Asymptotic Bounds for the Number of Rational Places

(c) If q = 3 is a cube than A(q) ≥ 2(2 − 1)/( + 2). For  being a prime, this
is a result of Zink [47]; for arbitrary  this bound was first shown by Bezerra,
Garcia and Stichtenoth [4]. In Theorem 7.4.17 we will give a simpler proof
following Bassa, Garcia and Stichtenoth [2].
(d) The exact value of A(q) is not known, for any non-square q.

7.2 Towers of Function Fields

Suppose that we have a sequence of function fields Fi /IFq (i = 0, 1, 2, . . .)
with g(Fi ) → ∞ and limi→∞ N (Fi )/g(Fi ) > 0 (where N (Fi ) denotes the
number of rational places and g(Fi ) denotes the genus of Fi ). Then it is clear
that A(q) ≥ limi→∞ N (Fi )/g(Fi ); so we obtain a non-trivial lower bound
for A(q). In this and the subsequent sections we will describe a systematic
approach to study such sequences of function fields.

Definition 7.2.1. A tower over IFq is an infinite sequence F = (F0 , F1 , F2 , . . .)

of function fields Fi /IFq such that the following hold:
(i) F0  F1  F2  . . .  Fn  . . . ;
(ii) each extension Fi+1 /Fi is finite and separable;
(iii) the genera satisfy g(Fi ) → ∞ for i → ∞.

Note that we always assume that IFq is the full constant field of Fi , for all
i ≥ 0.

Remark 7.2.2. The condition (iii) above follows from the conditions (i), (ii)
and the following slightly weaker condition
(iii*) g(Fj ) ≥ 2 for some j ≥ 0.

Proof. By Hurwitz Genus Formula one has

g(Fi+1 ) − 1 ≥ [Fi+1 : Fi ](g(Fi ) − 1) for all i.

Since g(Fj ) ≥ 2 and [Fi+1 : Fi ] ≥ 2, it follows that

g(Fj ) < g(Fj+1 ) < g(Fj+2 ) < . . . ,

hence g(Fi ) → ∞ for i → ∞.

As we pointed out above, one is interested in the behavior of the quotient

N (Fi )/g(Fi ) for i → ∞. It is convenient to consider also the behavior of the
number of rational places and the genus separately.

Lemma 7.2.3. Let F = (F0 , F1 , F2 , . . .) be a tower over IFq . Then the follow-
ing hold:
 7.2 Towers of Function Fields 247

(a) The sequence of rational numbers (N (Fi )/[Fi : F0 ])i≥0 is monotonically

decreasing and hence is convergent in IR≥0 .
(b) The sequence of rational numbers ((g(Fi ) − 1)/[Fi : F0 ])i≥0 is monotoni-
cally increasing and hence is convergent in IR≥0 ∪ {∞}.
(c) Let j ≥ 0 such that g(Fj ) ≥ 2. Then the sequence (N (Fi )/(g(Fi ) − 1))i≥j
is monotonically decreasing and hence is convergent in IR≥0 .
Proof. (a) If Q is a rational place of Fi+1 , then the restriction P := Q ∩ Fi
of Q to Fi is a rational place of Fi . Conversely, at most [Fi+1 : Fi ] rational
places of Fi+1 lie above a rational place of Fi (see Corollary 3.1.12). It follows
that N (Fi+1 ) ≤ [Fi+1 : Fi ] · N (Fi ) and therefore
N (Fi+1 ) [Fi+1 : Fi ] N (Fi )
≤ · N (Fi ) = ·
[Fi+1 : F0 ] [Fi+1 : F0 ] [Fi : F0 ]

(b) Using Hurwitz Genus Formula for the field extension Fi+1 /Fi we obtain
g(Fi+1 ) − 1 ≥ [Fi+1 : Fi ] (g(Fi ) − 1). Dividing by [Fi+1 : F0 ] we get the desired
inequality
g(Fi ) − 1 g(Fi+1 ) − 1
≤ ·
[Fi : F0 ] [Fi+1 : F0 ]
(c) The proof is similar to (a) and (b).

Because of Lemma 7.2.3 the following definitions are meaningful.

Definition 7.2.4. Let F = (F0 , F1 , F2 , . . .) be a tower over IFq .

(a) The splitting rate ν(F/F0 ) of F over F0 is defined as

ν(F/F0 ) := lim N (Fi )/[Fi : F0 ] .

i→∞

(b) The genus γ(F/F0 ) of F over F0 is defined as

γ(F/F0 ) := lim g(Fi )/[Fi : F0 ] .

i→∞

(c) The limit λ(F) of the tower F is defined as

λ(F) := lim N (Fi )/g(Fi ) .

i→∞

It is clear from Lemma 7.2.3 and the definition of A(q) that

0 ≤ ν(F/F0 ) < ∞ ,
0 < γ(F/F0 ) ≤ ∞ ,
0 ≤ λ(F) ≤ A(q) .
248 7 Asymptotic Bounds for the Number of Rational Places

Moreover we have the equation

λ(F) = ν(F/F0 )/γ(F/F0 ) . (7.6)

This means in particular that λ(F) = 0 if γ(F/F0 ) = ∞.

Definition 7.2.5. A tower F over IFq is called

(asymptotically) good, if λ(F) > 0 ,

(asymptotically) bad, if λ(F) = 0 ,
(asymptotically) optimal, if λ(F) = A(q) .

As an immediate consequence of Equation (7.6) we obtain the following char-

acterization of good towers.

Proposition 7.2.6. A tower F = (F0 , F1 , F2 , . . .) over IFq is asymptotically

good if and only if ν(F/F0 ) > 0 and γ(F/F0 ) < ∞.

It is a non-trivial task to find asymptotically good towers. If one attempts

to construct good towers, it turns out in “most” cases that either ν(F/F0 ) = 0
or that γ(F/F0 ) = ∞ (and hence the tower is bad). On the other hand, for
applications in coding theory (and further applications in cryptography and
other areas) it is of great importance to have some good towers explicitly. We
will provide several examples in Sections 7.3 and 7.4.

Definition 7.2.7. Let F = (F0 , F1 , F2 , . . .) and E = (E0 , E1 , E2 , . . .) be tow-

ers over IFq . Then E is said to be a subtower of F if for each i ≥ 0 there exists
an index j = j(i) and an embedding ϕi : Ei → Fj over IFq .

The following result is sometimes useful:

Proposition 7.2.8. Let E be a subtower of F. Then λ(E) ≥ λ(F). In partic-

ular one has:
(a) If F is asymptotically good then E is also asymptotically good.
(b) If E is asymptotically bad then F is also asymptotically bad.

Proof. Let ϕi : Ei → Fj(i) be an embedding of Ei into Fj(i) . Let Hi be the

subfield of Fj(i) which is uniquely determined by the following properties:
• ϕi (Ei ) ⊆ Hi ⊆ Fj(i) .
• Hi /ϕi (Ei ) is separable.
• Fj(i) /Hi is purely inseparable.
 7.2 Towers of Function Fields 249

Then Hi is isomorphic to Fj(i) by Proposition 3.10.2.(c) and therefore

N (Fj(i) ) N (Hi ) N (ϕi (Ei )) N (Ei )

= ≤ = ·
g(Fj(i) ) − 1 g(Hi ) − 1 g(ϕi (Ei )) − 1 g(Ei ) − 1

The inequality above follows from Lemma 7.2.3(c). For i → ∞ we obtain the
desired result, λ(F) ≤ λ(E).

In order to study the splitting rate ν(F/F0 ) and the genus γ(F/F0 ) of a
tower F over IFq , we introduce the notions of splitting locus and ramification
locus. Recall that, given a finite extension E/F of function fields, a place
P ∈ IPF splits completely in E/F if there are exactly n := [E : F ] distinct
places Q1 , . . . , Qn ∈ IPE with Qi |P . Likewise, P is called ramified in E/F if
there is at least one place Q ∈ IPE with Q|P and e(Q|P ) > 1.

Definition 7.2.9. Let F = (F0 , F1 , F2 , . . .) be a tower over IFq .

(a) The set

Split(F/F0 ) := { P ∈ IPF0 | deg P = 1 and P splits completely

in all extensions Fn /F0 }

is called the splitting locus of F over F0 .

(b) The set

Ram(F/F0 ) := { P ∈ IPF0 | P is ramified in Fn /F0 for some n ≥ 1 }

is called the ramification locus of F over F0 .

Clearly the splitting locus Split(F/F0 ) is a finite set (which may be empty);
the ramification locus may be finite or infinite.

Theorem 7.2.10. Let F = (F0 , F1 , F2 , . . .) be a tower over IFq .

(a) Let s := |Split(F/F0 )|. Then the splitting rate ν(F/F0 ) satisfies

ν(F/F0 ) ≥ s .

(b) Assume that the ramification locus Ram(F/F0 ) is finite and that for each
place P ∈ Ram(F/F0 ) there is a constant aP ∈ IR, such that for all n ≥ 0
and for all places Q ∈ IPFn lying above P , the different exponent d(Q|P ) is
bounded by
d(Q|P ) ≤ aP · e(Q|P ) . (7.7)
Then the genus γ(F/F0 ) of the tower is finite, and we have the bound
1 
γ(F/F0 ) ≤ g(F0 ) − 1 + aP · deg P .
2
P ∈Ram(F /F0 )
250 7 Asymptotic Bounds for the Number of Rational Places

(c) Now we assume that the splitting locus of F/F0 is non-empty and that
F/F0 satisfies the conditions in (b). Then the tower F is asymptotically good,
and its limit λ(F) satisfies
2s
λ(F) ≥  > 0,
2g(F0 ) − 2 + P ∈Ram(F /F0 ) aP · deg P

where s = |Split(F/F0 )|, and aP is as in (7.7).

Proof. (a) Above each place P ∈ Split(F/F0 ) there are exactly [Fn : F0 ] places
of Fn , and they are all rational. Hence N (Fn ) ≥ [Fn : F0 ] · |Split(F/F0 )|,
and (a) follows immediately.
(b) For simplity we set gn := g(Fn ) for all n ≥ 0. The Hurwitz Genus Formula
for Fn /F0 gives
 
2gn − 2 = [Fn : F0 ](2g0 − 2) + d(Q|P ) · deg Q
P ∈Ram(F /F0 ) Q∈IPFn , Q|P
 
≤ [Fn : F0 ](2g0 − 2) + aP · e(Q|P ) · f (Q|P ) · deg P
P ∈Ram(F /F0 ) Q∈IPFn , Q|P
 
= [Fn : F0 ](2g0 − 2) + aP · deg P · e(Q|P ) · f (Q|P )
P ∈Ram(F /F0 ) Q∈IPFn , Q|P
⎛ ⎞

= [Fn : F0 ] ⎝2g0 − 2 + aP · deg P ⎠ .
P ∈Ram(F /F0 )

Here we have used (7.7) and the Fundamental Equality


e(Q|P ) · f (Q|P ) = [Fn : F0 ] ,
Q|P

see Theorem 3.1.11. Dividing the inequality above by 2[Fn : F0 ] and letting
n → ∞ we obtain the inequality
1 
γ(F/F0 ) ≤ g(F0 ) − 1 + aP · deg P .
2
P ∈Ram(F /F0 )

(c) follows immediately from (a) and (b), since λ(F) = ν(F/F0 )/γ(F/F0 )
(see Equation (7.6)).

The assumption (7.7) about different exponents holds in particular in the

case of tame ramification. We call the tower F = (F0 , F1 , F2 , . . .) tame if all
ramification indices e(Q|P ) in all extensions Fn /F0 are relatively prime to the
characteristic of IFq ; otherwise we say that the tower F/F0 is wild. Then we
obtain the following corollary to Theorem 7.2.10.
 7.2 Towers of Function Fields 251

Corollary 7.2.11. Let F = (F0 , F1 , F2 , . . .) be a tame tower. Assume that

the splitting locus Split(F/F0 ) is non-empty and that the ramification locus
Ram(F/F0 ) is finite. Then F is asymptotically good. More precisely, setting

s := |Split(F/F0 )| and r := deg P ,
P ∈Ram(F /F0 )

we obtain
2s
λ(F) ≥ .
2g(F0 ) − 2 + r
Proof. For a tamely ramified place Q|P we have d(Q|P ) = e(Q|P ) − 1 ≤
e(Q|P ), by Dedekind’s Different Theorem. Therefore we can choose aP := 1
in (7.7), and Theorem 7.2.10(c) gives the desired result.

The towers that we will construct in Sections 7.3 and 7.4 are often given
in a recursive manner. Let us give a precise definition of what this means.
Recall that a rational function f (T ) over IFq is a quotient of two poly-
nomials f (T ) = f1 (T )/f2 (T ) with f1 (T ), f2 (T ) ∈ IFq [T ] and f2 (T ) = 0.
We can assume that f1 (T ), f2 (T ) are relatively prime, and then we put
deg f (T ) := max{deg f1 (T ), deg f2 (T )}, and call it the degree of f (T ). If
f (T ) ∈ IFq (T )\IFq then f (T ) is said to be non-constant; this is equivalent
to the condition deg f (T ) ≥ 1. Note that every polynomial f (T ) ∈ IFq [T ] can
also be regarded as a rational function.

Definition 7.2.12. Let f (Y ) ∈ IFq (Y ) and h(X) ∈ IFq (X) be non-constant

rational functions, and let F = (F0 , F1 , F2 , . . .) be a sequence of function fields.
Suppose that there exist elements xi ∈ Fi (i = 0, 1, 2, . . .) such that
(i) x0 is transcendental over IFq and F0 = IFq (x0 ); i.e., F0 is a rational
function field.
(ii) Fi = IFq (x0 , x1 , . . . , xi ) for all i ≥ 0.
(iii) For all i ≥ 0 the elements xi , xi+1 satisfy f (xi+1 ) = h(xi ) .
(iv) [F1 : F0 ] = deg f (Y ).
Then we say that the sequence F is recursively defined over IFq by the equation
f (Y ) = h(X) .

Remark 7.2.13. With the notation of Definition 7.2.12 it follows that

[Fi+1 : Fi ] ≤ deg f (Y ) for all i ≥ 0 .
Proof. We write f (Y ) = f1 (Y )/f2 (Y ) with relatively prime polynomials
f1 (Y ), f2 (Y ) ∈ IFq [Y ]. Then Fi+1 = Fi (xi+1 ), and xi+1 is a zero of the
polynomial
ϕi (Y ) := f1 (Y ) − h(xi )f2 (Y ) ∈ Fi [Y ] ,
which has degree deg ϕi (Y ) = deg f (Y ).
252 7 Asymptotic Bounds for the Number of Rational Places

Remark 7.2.14. Condition (iv) in Definition 7.2.12 is not really needed in most
proofs on recursive towers. However this assumption holds in all examples of
recursive towers that will be considered in this book, and it will later make
some statements smoother. Clearly (iv) is equivalent to the condition that the
polynomial
ϕ(Y ) := f1 (Y ) − h(x0 )f2 (Y ) ∈ IFq (x0 )[Y ]
is irreducible over the field IFq (x0 ). We remark that the degree [Fi+1 : Fi ] may
be less than deg f (Y ) for some i ≥ 1.

Here are two typical examples of recursively defined sequences of function

fields. The first one is given by the equation

Y m = a(X + b)m + c , (7.8)

where a, b, c ∈ IF×
q , m > 1 and gcd(m, q) = 1. The second one is defined by
the equation
X
Y −Y = , (7.9)
1 − X −1
where q = 2 is a square. We will study these two sequences in detail in
Sections 7.3 and 7.4.
Now we consider a sequence of fields F = (F0 , F1 , F2 , . . .) which is recur-
sively defined by an equation f (Y ) = h(X) with some non-constant rational
functions f (Y ) ∈ IFq (Y ) and h(X) ∈ IFq (X). In order to decide whether the
sequence F is a tower over IFq , one has first to answer the following questions:
• Is Fn  Fn+1 for all n ≥ 0 ?
• Is IFq the full constant field of Fn for all n ≥ 0 ?
Our next proposition (which does not only apply to recursively defined se-
quences of fields) gives sufficient conditions for an affirmative answer to these
questions.

Proposition 7.2.15. Consider a sequence of fields F0 ⊆ F1 ⊆ F2 ⊆ . . . ,

where F0 is a function field with the exact constant field IFq and [Fn+1 : Fn ]
< ∞ for all n ≥ 0. Suppose that for all n there exist places Pn ∈ IPFn and
Qn ∈ IPFn+1 with Qn |Pn and ramification index e(Qn |Pn ) > 1. Then it follows
that Fn  Fn+1 .
If we assume furthermore that e(Qn |Pn ) = [Fn+1 : Fn ] for all n, then IFq is
the full constant field of Fn for all n ≥ 0.

Proof. By the Fundamental Equality we have [Fn+1 : Fn ] ≥ e(Qn |Pn ) and

therefore Fn  Fn+1 . If we assume the equality e(Qn |Pn ) = [Fn+1 : Fn ], then
Fn and Fn+1 have the same constant field (since constant field extensions are
unramified by Theorem 3.6.3).
 7.2 Towers of Function Fields 253

We illustrate the use of Proposition 7.2.15 with a simple example.

Example 7.2.16. Let q be a power of an odd prime number. We claim that

the sequence F = (F0 , F1 , F2 , . . .) which is recursively defined over IFq by the
equation
X2 + 1
Y2 = (7.10)
2X
is a tower over IFq . Hence we need to prove the following:
(i) Fn  Fn+1 , and Fn+1 /Fn is separable for all n ≥ 0.
(ii) IFq is the full constant field of all Fn .
(iii) g(Fj ) ≥ 2 for some j.
First we observe that Fn+1 = Fn (xn+1 ) and x2n+1 = (x2n + 1)/2xn , hence
[Fn+1 : Fn ] ≤ 2. Since q ≡ 1 mod 2, it follows that Fn+1 /Fn is separable.
Our next aim is to find places Pn ∈ IPFn and Qn ∈ IPFn+1 with Qn |Pn
and e(Qn |Pn ) = 2. Having found such places, items (i) and (ii) will follow
immediately from Proposition 7.2.15. We proceed as follows: Let P0 ∈ IPF0 be
the unique pole of x0 in the rational function field F0 = IFq (x0 ) and choose
some place Q0 ∈ IPF1 lying above P0 . From the equation x21 = (x20 + 1)/2x0
we conclude that
 2 
x0 + 1
2vQ0 (x1 ) = vQ0 (x1 ) = e(Q0 |P0 ) · vP0
2
= e(Q0 |P0 ) · (−1) ,
2x0
therefore e(Q0 |P0 ) = 2 and vQ0 (x1 ) = −1 (observe here that e(Q0 |P0 ) ≤
[F1 : F0 ] ≤ 2).
In the next step we take P1 as Q0 , and choose Q1 as a place of F2 lying
above P1 . Again we see from the equation x22 = (x21 +1)/2x1 that e(Q1 |P1 ) = 2
and vQ1 (x2 ) = −1. By iterating this process we obtain the desired places
Pn ∈ IPFn and Qn ∈ IPFn+1 with Qn |Pn and e(Qn |Pn ) = 2, for all n ≥ 0.
It remains to prove (iii). From the equation x21 = (x20 + 1)/2x0 we see that
exactly the following places of F0 are ramified in F1 /F0 :
• the zero and the pole of x0 ;
• the two zeros of x20 + 1, if x20 + 1 splits into linear factors in IFq [x0 ], or the
place of degree 2 corresponding to x20 + 1, if x20 + 1 is irreducible.
In each case the different degree of F1 /F0 is deg Diff(F1 /F0 ) = 4 by Dedekind’s
Different Theorem, and then the Hurwitz Genus Formula for F1 /F0 gives
g(F1 ) = 1. In the extension F2 /F1 there is at least one ramified place (namely
the place Q1 |P1 as constructed above) and hence deg Diff(F2 /F1 ) ≥ 1. Again
by Hurwitz Genus Formula we obtain that g(F2 ) ≥ 2. This proves (iii).

Remark 7.2.17. In the following, we will study towers of function fields over
IFq which are recursively defined by an equation f (Y ) = h(X). Since each step
in a tower is by definition a separable extension, this implies that the rational
function f (Y ) must be separable (i.e., f (Y ) ∈
/ IFq (Y p ) where p = char IFq ).
254 7 Asymptotic Bounds for the Number of Rational Places

Our main goal is to construct asymptotically good towers F over IFq . Ac-
cording to Theorem 7.2.10 we are therefore interested in formulating criteria
which ensure a non-empty splitting locus Split(F/F0 ) and a finite ramification
locus Ram(F/F0 ). In order to tackle these problems for recursively defined
towers, we introduce the notion of the basic function field.
Definition 7.2.18. Let F = (F0 , F1 , F2 , . . .) be a sequence of function fields
which is recursively defined over IFq by the equation f (Y ) = h(X), with non-
constant rational functions f (Y ), h(X) over IFq . We define the basic function
field F corresponding to the tower F by

F := IFq (x, y) with the relation f (y) = h(x) .

Observe that the extension F/IFq (x) is separable by Remark 7.2.17. As a
consequence of condition (iv) in Definition 7.2.12 we have
[F : IFq (x)] = deg f (Y ) and [F : IFq (y)] = deg h(X) .
Moreover, all subfields of Fn of the form IFq (xi , xi+1 ) (with 0 ≤ i ≤ n − 1)
are IFq - isomorphic to the basic function field F = IFq (x, y) under the map
x → xi , y → xi+1 .
Before giving a first application of the basic function field in Proposition
7.2.20, we introduce a convenient notation for the rational places of a rational
function field.
Definition 7.2.19. Let K(z) be a rational function field over an arbitrary
field K. Then we denote for α ∈ K by (z = α) the unique place of K(z) which
is a zero of z − α. Likewise we denote by (z = ∞) the unique pole of z in
K(z).
Observe that this notation differs from the notation for rational places of
K(z) as used in Section 1.2.
Proposition 7.2.20. Let F = (F0 , F1 , F2 , . . .) be a tower over IFq which is
recursively defined by the equation f (Y ) = h(X), and let F = IFq (x, y) be the
corresponding basic function field with the relation f (y) = h(x). Assume that
Σ ⊆ IFq ∪{∞} is a non-empty set which satisfies the following two conditions:
(1) For all α ∈ Σ, the place (x = α) of IFq (x) splits completely in the
extension F/IFq (x).
(2) If α ∈ Σ and Q is a place of F above the place (x = α), then y(Q) ∈ Σ.
Then for all α ∈ Σ, the place (x0 = α) of F0 = IFq (x0 ) splits completely in
F/F0 . In particular, the splitting locus of F/F0 satisfies
|Split(F/F0 )| ≥ |Σ| ,
and therefore we obtain a lower bound for the splitting rate ν(F/F0 ) of the
tower F,
ν(F/F0 ) ≥ |Σ| .
 7.2 Towers of Function Fields 255

Proof. Let α ∈ Σ. We show by induction that the place (x0 = α) splits

completely in Fn /F0 for all n ≥ 0. This is trivial for n = 0, and we assume
now that the assertion holds for some n. We have to show that every place
Q ∈ IPFn lying above (x0 = α) splits completely in Fn+1 /Fn . By condition
(2) we know that xn (Q) =: β ∈ Σ, and then it follows from condition (1) that
the place (xn = β) splits completely in the extension IFq (xn , xn+1 )/IFq (xn ).
Since Fn+1 is the composite field of Fn and IFq (xn , xn+1 ), the place Q splits
completely in Fn+1 /Fn by Proposition 3.9.6. The inequality ν(F/F0 ) ≥ |Σ|
follows now from Theorem 7.2.10(a).

Corollary 7.2.21. Let F = (F0 , F1 , F2 , . . .) be a tower over IFq which is re-

cursively defined by the equation f (Y ) = h(X). Let m := deg f (Y ). Assume
that Σ ⊆ IFq is a non-empty set such that the following condition holds:
For all α ∈ Σ we have h(α) ∈ IFq (i.e., α is not a pole of the rational
function h(X)), and the equation f (t) = h(α) has m distinct roots
t = β in Σ.
Then all places (x0 = α) with α ∈ Σ are in the splitting locus of F over F0 ,
and the splitting rate satisfies ν(F/F0 ) ≥ |Σ|.

Proof. Consider the basic function field F = IFq (x, y) with defining equation
f (y) = h(x). Let P = (x = α) with α ∈ Σ be the place of IFq (x) which is the
zero of x − α, and OP ⊆ IFq (x) be the corresponding valuation ring. We write
f (Y ) = f1 (Y )/f2 (Y ) with relatively prime polynomials f1 (Y ), f2 (Y ) ∈ IFq [Y ]
and max {deg f1 (Y ), deg f2 (Y )} = m, say f1 (Y ) = am Y m + . . . + a0 and
f2 (Y ) = bm Y m + . . . + b0 . Then y ∈ F is a root of the polynomial

ϕ(Y ) = f1 (Y ) − h(x)f2 (Y ) ∈ IFq (x)[Y ] .

By our assumption, the equation f1 (t) − h(α)f2 (t) = 0 has m distinct roots
β1 , . . . , βm ∈ Σ, so the leading coefficient am − bm h(α) is nonzero. This
means that the function am − bm h(x) ∈ IFq (x) is a unit in OP . Dividing
ϕ(Y ) by am − bm h(x) gives now an integral equation for y over OP , whose
reduction modulo P has the roots β1 , . . . , βm . By Kummer’s Theorem 3.3.7
there exist m distinct places Q1 , . . . , Qm ∈ IPF lying above P , with y(Qi ) = βi
for i = 1, . . . , m. We have thus verified the conditions (1), (2) in Proposition
7.2.20, and hence the result follows.

Example 7.2.16(cont.). We return to the tower F in Example 7.2.16; i.e.,

F is recursively given by the equation f (Y ) = h(X) with f (Y ) = Y 2 and
h(X) = (X 2 + 1)/2X over a field IFq of odd characteristic. One can show
that for all squares q = 2 ( being a power of an odd prime) the splitting
locus of F/F0 is non-empty. Since the proof requires tools, which are not
covered in this book, we shall be content with proving the case q = 9. The
256 7 Asymptotic Bounds for the Number of Rational Places

field IF9 can be represented as IF9 = IF3 (δ) with δ 2 = −1, so we have IF9 =
{0, ±1, ±δ, ±(δ+1), ±(δ−1)}. We claim that the set Σ := {±(δ+1), ±(δ−1)}
satisfies the condition of Corollary 7.2.21. In fact one finds by straightforward
calculation that

h (δ + 1) = h(δ − 1) = δ = f (δ − 1) = f (−δ + 1) ,
h (−δ − 1) = h(−δ + 1) = −δ = f (δ + 1) = f (−δ − 1) .

We conclude from Corollary 7.2.21 that the splitting rate of F over F0 satisfies
ν(F/F0 ) ≥ 4.

Next we turn our attention to ramification in towers.

Remark 7.2.22. Let F = (F0 , F1 , F2 , . . .) be a tower over IFq and let L ⊇ IFq
be an algebraic field extension of IFq . Then we can consider the constant field
extension F  := FL of F by L, which is defined as the sequence of fields

F  = (F0 , F1 , F2 , . . .) with Fi := Fi L .


It follows that [Fi+1 : Fi ] = [Fi+1 : Fi ], and L is the full constant field of Fi ,
for all i ≥ 0. Moreover we have g(Fi ) = g(Fi ). A place P ∈ IPFi is ramified in
Fi+1 /Fi if and only if the places P  ∈ IPFi above P are ramified in Fi+1 
/Fi ,
therefore

Ram(F  /F0 ) = {P  ∈ IPF0 | P  ∩ F0 ∈ Ram(F/F0 )}

and
γ(F  /F0 ) = γ(F/F0 ) .
In case of a finite ramification locus we define the ramification divisor of F/F0
by 
R(F/F0 ) := P ;
P ∈Ram(F /F0 )

then we also have

R(F  /F0 ) = ConF0 /F0 R(F/F0 )
and
deg R(F  /F0 ) = deg R(F/F0 ) .
All statements above follow immediately from Section 3.6.

We can now prove a useful criterion for finiteness of the ramification locus
of recursive towers.

Proposition 7.2.23. Let F = (F0 , F1 , F2 , . . .) be a recursive tower over IFq ,

defined by the equation f (Y ) = h(X), and let F  = FL = (F0 , F1 , F2 , . . .) be
the constant field extension of F by an algebraic extension field L ⊇ IFq . We
denote by F (resp. F  ) the basic function field of F (resp. F  ), so F = IFq (x, y)
 7.2 Towers of Function Fields 257

and F  = F L = L(x, y) with the relation f (y) = h(x). We assume that all
places of L(x), which ramify in the extension F  /L(x), are rational and hence
the set
Λ0 := {x(P ) | P ∈ IPL(x) is ramified in F  /L(x)}
is contained in L ∪ {∞}. Suppose that Λ is a finite subset of L ∪ {∞} such
that the following two conditions hold:
(1) Λ0 ⊆ Λ.
(2) If β ∈ Λ and α ∈ IF¯ q ∪ {∞} satisfy the equation f (β) = h(α), then
α ∈ Λ.
Then the ramification locus Ram(F  /F0 ) (and hence also Ram(F/F0 )) is fi-
nite, and we have

Ram(F  /F0 ) ⊆ {P ∈ IPF0 | x0 (P ) ∈ Λ} .

Proof. By definition, the field F0 is the rational function field F0 = L(x0 ) over
L. Let P ∈ Ram(F  /F0 ). There is some n ≥ 0 and some place Q of Fn above

P such that Q is ramified in the extension Fn+1 /Fn . Setting R := Q ∩ L(xn ),
we have the situation as shown in Figure 7.1 below.


Fn+1
• •
@ @
Fn @ Q @
• @ • @
@ @ L(x , x @ @
@ @ n n+1 ) @ @
@ @• @ @•
@ @
• @@• • @
@•
F0 = L(x0 ) L(xn ) P R

Fig. 7.1.


Since Q is ramified in Fn+1 /Fn , it follows by Abhyankar’s Lemma that the
place R is ramified in the extension L(xn , xn+1 )/L(xn ). So we have βn :=
xn (Q) ∈ Λ0 . Setting βi := xi (Q) we obtain f (βi+1 ) = h(βi ) for i = 0, . . . , n−1,
and from Condition (2) we conclude that β0 = x0 (Q) = x0 (P ) is in Λ.

Example 7.2.16(cont.). We apply Proposition 7.2.23 to the recursive tower

F over IFq , which was considered in Example 7.2.16. Recall that F =
(F0 , F1 , F2 , . . .) is recursively defined by the equation

Y 2 = (X 2 + 1)/2X
258 7 Asymptotic Bounds for the Number of Rational Places

over a field IFq of odd characteristic. We fix a finite field L ⊇ IFq which

contains an element δ with δ 2 = −1. The set Λ0 in Proposition 7.2.23 is given
here by
Λ0 = {∞, 0, ±δ} ,
as a direct consequence of Proposition 3.7.3. We consider now the set

Λ := {0, ∞, ±1, ±δ} ⊆ L ∪ {∞} ,

and we claim that Λ satisfies Condition (2) of Proposition 7.2.23. So we have

to show that for all β ∈ Λ, all solutions α ∈ IF ¯ q ∪ {∞} of the equation
2 2
(α + 1)/2α = β are in Λ. This is easily checked as follows:

if β = ∞ then α = 0 or α = ∞ ;
if β = 0 then α = ±δ ;
if β = ±1 then α = 1 ;
if β = ±δ then α = −1 .

Proposition 7.2.23 implies now that the ramification locus of F  = FL over

F0 = L(x0 ) is contained in the set

{(x0 = 0), (x0 = ∞), (x0 = 1), (x0 = −1), (x0 = δ), (x0 = −δ)} ⊆ IPL(x0 ) .

The tower F is tame, as char(IFq ) = 2 and all extensions Fi+1 /Fi are of degree
[Fi+1 : Fi ] = 2. So we get the bound
1  6
γ(F/F0 ) ≤ −1 + degP ≤ −1 + = 2
2 2
P ∈Ram(F /F0 )

for genus of the tower, by Theorem 7.2.10(b)

In the special case q = 9 we have proved (just before Remark 7.2.22) the
inequality ν(F/F0 ) ≥ 4 for the splitting rate ν(F/F0 ), hence it follows that
λ(F) = ν(F/F0 )/γ(F/F0 )√≥ 4/2 = 2. Note that Ihara’s constant A(9) satis-
fies the inequality A(9) ≤ 9 − 1 = 2, so we have 2 ≤ λ(F) ≤ A(9) ≤ 2.
We summarize the results of Example 7.2.16 as follows:

• The tower F over IF9 , which is recursively defined by the equation Y 2 =

(X 2 + 1)/2X, is asymptotically optimal.
• A(9) = 2.

Observe that the second assertion is a special case of the equality A(q 2 ) = q−1
that will be proved in Section 7.4.
 7.3 Some Tame Towers 259

7.3 Some Tame Towers

In this section we present two examples of asymptotically good tame towers.
Both of them are special cases of the next theorem. As usual, IF¯q denotes the
algebraic closure of IFq .

Theorem 7.3.1. Let m ≥ 2 be an integer with q ≡ 1 mod m. Assume that

the polynomial h(X) ∈ IFq [X] has the following properties:
(1) deg h(X) = m, and the leading coefficient of h(X) is an m-th power
of some element c ∈ IF× q .

(2) h(0) = 0 and h (0) = 0; i.e., 0 is a simple zero of h(X).
(3) There exists a subset Λ ⊆ IFq such that for all β, γ ∈ IF¯q ,
(a) h(γ) = 0 =⇒ γ ∈ Λ , and
(b) α ∈ Λ and β m = h(α) =⇒ β ∈ Λ .
Then the equation
Y m = h(X) (7.11)
defines an asymptotically good tower F over IFq with limit
2
λ(F) ≥ .
|Λ| − 2

Proof. We consider the sequence F = (F0 , F1 , F2 , ...), where F0 = IFq (x0 ) is a

rational function field and for all n ≥ 0,

Fn+1 = Fn (xn+1 ) with xm

n+1 = h(xn ) . (7.12)

Clearly Fn+1 /Fn is a separable extension of degree ≤ m. We show now that

at each step Fn+1 /Fn , there is a place with ramification index e = m (which
implies that [Fn+1 : Fn ] = m). Denote by P0 = (x0 = 0), the zero of x0 in F0 .
For n ≥ 0 we choose recursively a place Pn+1 of Fn+1 with Pn+1 |Pn , and we
claim that
vPn (xn ) = 1 and e(Pn+1 |Pn ) = m (7.13)
holds for all n ≥ 0. For n = 0 we have vP0 (x0 ) = 1. Condition (2) implies
that P0 is a simple zero of the function h(x0 ), therefore vP0 (h(x0 )) = 1. From
(7.12) we get

1 ) = vP1 (h(x0 )) = e(P1 |P0 ) ≤ [F1 : F0 ] ≤ m ,

m · vP1 (x1 ) = vP1 (xm

hence (7.13) holds for n = 0. Using the same argument, one shows by induction
that (7.13) holds for all n ≥ 0. We conclude that [Fn+1 : Fn ] = m and that
IFq is the full constant field of Fn for all n ≥ 0, cf. Proposition 7.2.15.
We claim that the pole (x0 = ∞) of x0 splits completely in all extensions
Fn /F0 . In order to prove this we consider the basic function field

F = IFq (x, y) with y m = h(x) .

260 7 Asymptotic Bounds for the Number of Rational Places

According to Proposition 7.2.20, it is sufficient to show:

(i) the place P∞ := (x = ∞) splits completely in F/IFq (x), and
(ii) if Q ∈ IPF is a place with Q|P∞ , then y(Q) = ∞.
By condition (1) we have y m = cm xm + . . . , so the element z := y/x satisfies
an equation

z m = cm + r(1/x) with vP∞ (r(1/x)) > 0 and c ∈ IF×

q .

Hence z is integral over the valuation ring OP∞ . The equation Z m = cm has m
distinct roots in IFq (here we use the assumption q ≡ 1 mod m), and it follows
from Kummer’s Theorem 3.3.7 that P∞ splits completely in F/IFq (x). This is
assertion (i). Assertion (ii) follows immediately from the equation y m = h(x).
We have thus shown that the splitting locus of F/F0 is non-empty. As
a consequence, the number N (Fn ) of rational places of Fn /IFq satisfies the
inequality N (Fn ) ≥ mn and therefore the genus g(Fn ) tends to infinity as
n → ∞. So F is in fact a tower over IFq .
The only places of IFq (x) which ramify in the extension F/IFq (x) are zeros
of h(x), by Theorem 3.7.3. Hence Proposition 7.2.23 implies that the ramifi-
cation locus of F/F0 is finite and

|Ram(F/F0 )| ≤ |Λ| .

(Note that the assumptions of Proposition 7.2.23 follow from the conditions
3(a),(b).) Now Corollary 7.2.11 gives the desired estimate

λ(F) ≥ 2/(|Λ| − 2)

for the limit of the tower F.

We remark that the cardinality of the set Λ above is certainly larger than
2 since a tame extension of a rational function field with at most 2 ramified
places (of degree 1) is rational by the Hurwitz Genus Formula.

Proposition 7.3.2. Let q = 2 be a square,  > 2. Then the equation

−1 −1
Y = 1 − (X + 1)

defines an asymptotically good tower F over IFq with limit

λ(F) ≥ 2/( − 1).

For  = 3 this tower is optimal over the field IF9 .

Proof. We set h(X) = 1 − (X + 1) −1 and Λ = IF . We need to check that the

assumptions of Theorem 7.3.1 are satisfied:
(1) The leading coefficient of h(X) is −1, which is a square in IFq since q = 2 .
 7.4 Some Wild Towers 261

(2) The condition h(0) = 0 is clear, and h (0) = 0 follows from the equation
h (X) = (X + 1) −2 .
(3)(a) Let h(γ) = 0. Then (γ + 1) −1 = 1, hence γ + 1 ∈ IF× and γ ∈ IF = Λ.
(3)(b) Let α ∈ IF and β −1 = h(α) = 1 − (α + 1) −1 . Then h(α) = 0 (if
α = −1) or h(α) = 1 (if α = −1), and therefore β ∈ IF = Λ.
The use of Proposition 7.3.1 now completes the proof.

Next we give examples of asymptotically good towers over all non prime
fields.

Proposition 7.3.3. Let q = e with e ≥ 2. Then the equation

Y m = 1 − (X + 1)m

with m := (q − 1)/( − 1) defines an asymptotically good tower F over IFq

with limit
λ(F) ≥ 2/(q − 2).
For  = e = 2 this is an optimal tower over IF4 .

Proof. In this case we have h(X) = 1−(X +1)m and we set Λ := IFq . Observe
that the map γ → γ m is the norm map from IFq to IF and hence is surjective.
Moreover, every element β ∈ IF¯q with β m ∈ IF is in IFq . Using these facts,
the proof of this proposition is essentially the same as that of Proposition
7.3.2.

Recall that Ihara’s constant A(q) is positive for all prime powers q = pe ,
see Remark 7.1.4(a). Proposition 7.3.3 provides a simple proof of this fact in
the case e > 1. The lower bound A(q) ≥ 2/(q − 2) is however rather weak for
q = 4.

7.4 Some Wild Towers

Ihara’s constant A(q) attains the Drinfeld-Vladut Bound A(q) = q 1/2 −1 when
q = 2 is a square, cf. Remark 7.1.4. In this section we will prove this result, by
providing a recursive tower G = (G0 , G1 , G2 , . . .) over IFq with limit λ(G) =
 − 1, when q = 2 . We also present a recursive tower H = (H0 , H1 , H2 , . . .)
over a cubic field IFq with q = 3 , whose limit satisfies λ(H) ≥ 2(2 −1)/(+2),
thus obtaining the bound A(3 ) ≥ 2(2 − 1)/( + 2). Both towers G and H are
wild towers; i.e., for some i ≥ 1 there are places which are wildly ramified in
the extensions Gi /G0 (resp. Hi /H0 ).
We begin with the tower G over a quadratic field IFq .
262 7 Asymptotic Bounds for the Number of Rational Places

Definition 7.4.1. Let q = 2 where  is a power of some prime number p. We

define the tower G = (G0 , G1 , G2 , . . .) of function fields Gi over IFq recursively
by the equation
X
Y −Y = ; (7.14)
1 − X −1
i.e., G0 = IFq (x0 ) is the rational function field, and for all i ≥ 0 we have
Gi+1 = Gi (xi+1 ) with

xi
xi+1 − xi+1 = . (7.15)
1 − xi −1

We will show that Equation (7.14) actually defines a tower (see Lemma 7.4.3)
and that this tower attains the Drinfeld-Vladut Bound over IFq .

Remark 7.4.2. With the notations of Definition 7.4.1 we set x̃i := ζxi for
i = 0, 1, 2, . . ., where ζ ∈ IFq satisfies the equation ζ −1 = −1. Then it follows
that

x̃i+1 + x̃i+1 = ζ xi+1 + ζxi+1 = −ζ(xi+1 − xi+1 )

xi ζ xi x̃
= −ζ · = = −1i .
1 − xi −1 1 + ζ −1 xi −1 x̃i + 1
This shows that the tower G can also be defined by the equation

X
Y +Y = −1
. (7.16)
X +1
In fact, the tower G was first introduced by Garcia-Stichtenoth [12] with
Equation (7.16) as its defining equation. The reason why we prefer Equation
(7.14) here, is that then the analogy between the towers H, considered below,
and G becomes more obvious.

Lemma 7.4.3. Equation (7.14) defines a recursive tower G = (G0 , G1 , G2 , . . .)

over IFq . All extensions Gi+1 /Gi are Galois of degree [Gi+1 : Gi ] = , and the
place (x0 = ∞) of G0 is totally ramified in all extensions Gn /G0 .
Proof. It is clear that the equation Y − Y = X /(1 − X −1 ) is separable
and hence all extensions Gi+1 /Gi are separable of degree [Gi+1 : Gi ] ≤ . Let
P0 := (x0 = ∞) be the pole of x0 in the rational function field G0 = IFq (x0 ).
For all i ≥ 0 we choose recursively a place Pi+1 ∈ IPGi+1 with Pi+1 |Pi , and
we claim that e(Pi+1 |Pi ) = . From the equation x1 − x1 = x0 /(1 − x0−1 ) we
see that
vP1 (x1 − x1 ) = e(P1 |P0 ) · (−1) < 0 ,
so P1 is a pole of x1 , and then vP1 (x1 − x1 ) =  · vP1 (x1 ) by the Strict Triangle
Inequality. It follows that
 7.4 Some Wild Towers 263

− · vP1 (x1 ) = e(P1 |P0 ) ≤ [G1 : G0 ] ≤  ,

hence e(P1 |P0 ) =  and vP1 (x1 ) = −1. By induction we obtain e(Pi+1 |Pi ) = 
and vPi+1 (xi+1 ) = −1 for all i ≥ 0. We conclude that [Gi+1 : Gi ] = , and IFq
is the full constant field of Gi for all i ≥ 0. Since the equation of xi+1 over
Gi is an Artin-Schreier equation y − y = xi /(1 − xi −1 ), it also follows that
Gi+1 /Gi is Galois.
It remains to show that g(Gn ) ≥ 2 for some n. Since

G1 = G0 (x1 ) with x1 − x1 = x0 /(1 − x0−1 )

and the right hand side of this equation has  simple poles in G0 (namely the
places (x0 = ∞) and (x0 = α) for α −1 = 1), the genus of G1 is

−1
g(G1 ) = (−2 + 2) = ( − 1)2
2
by Proposition 3.7.8. For all  = 2 we have thus g(G1 ) ≥ 2 as desired. For
 = 2 we have g(G1 ) = 1. In the extension G2 /G1 at least one place (namely
the pole of x0 ) ramifies, and then the Hurwitz Genus Formula for G2 /G1
shows that g(G2 ) ≥ 2.

Our next step is to estimate the splitting rate ν(G/G0 ) over the field IFq .
We are going to apply Corollary 7.2.21.

Lemma 7.4.4. Let G be the tower over IFq (with q = 2 ), which is recursively
defined by Equation (7.14). Then the splitting locus of G/G0 satisfies

Split(G/G0 ) ⊇ {(x0 = α) | α ∈ IFq \ IF } ,

and the splitting rate ν(G/G0 ) satisfies

ν(G/G0 ) ≥ 2 −  .

Proof. We want to show that the set Σ := IFq \ IF satisfies the condition of
Corollary 7.2.21. So let α ∈ Σ; then

α −1
−1
∈ IFq since α = 1 .
1−α
¯ q (the algebraic closure of IFq ) with
Consider an element β ∈ IF

α
β −β = −1
. (7.17)
1−α
Then 2
2 α α
β −β = −1 )
= −1 )
. (7.18)
(1 − α (1 − α
264 7 Asymptotic Bounds for the Number of Rational Places

Adding (7.17) and (7.18) we get

2
2 α α (α − α ) + (α − α )
β −β = −1
+ −1 )
=
1−α (1 − α (1 − α −1 ) +1
2
α−α
= = 0,
(1 − α −1 ) +1

2
since α ∈ IFq and q = 2 . Hence β = β; i.e., β ∈ IFq . Since β − β =
α /(1 − α −1 ) = 0, it follows that β ∈
/ IF and hence β ∈ Σ. It is also clear
that Equation (7.17) has  distinct roots β. Thus we have verified the condition
of Corollary 7.2.21. This finishes the proof.

Using Proposition 7.2.23, the ramification locus Ram(G/G0 ) can be deter-

mined as follows. We denote by
−1
G := IFq (x, y) with the relation y − y = x /(1 − x )

the basic function field of the tower G. By the theory of Artin-Schreier exten-
sions, exactly the places (x = ∞) and (x = γ) with γ −1 = 1 (i.e., γ ∈ IF× )
are ramified in G/IFq (x). So we have

Λ0 := {x(P ) | P ∈ IPIFq (x) is ramified in G/IFq (x)} = IF× ∪ {∞} .

We set Λ := IF ∪ {∞}. In order to verify condition (2) of Proposition 7.2.23,

¯ q ∪ {∞} satisfy the relation
we must show the following: if β ∈ Λ and α ∈ IF

α
β −β = −1
, (7.19)
1−α
then α ∈ Λ. We distinguish two cases:
Case 1. β ∈ IF . Then β − β = 0, and from (7.19) follows α = 0 ∈ Λ.
Case 2. β = ∞. Then it follows from (7.19) that α = ∞ or α −1 = 1, so
we have again α ∈ Λ.
Thus we can apply Proposition 7.2.23 and obtain:

Lemma 7.4.5. The tower G over IFq (with q = 2 ) which is recursively defined
by Equation (7.14) has a finite ramification locus. More precisely one has

Ram(G/G0 ) ⊆ {(x0 = β) | β ∈ IF ∪ {∞}} .

It is easy to show that the tower G is a wild tower. As we have seen in

Lemma 7.4.3, in each step Gi+1 /Gi there are places which are totally (and
hence wildly) ramified. So we do not have the estimate d(Q|P ) ≤ e(Q|P )
for all P ∈ Ram(G/G0 ) and Q lying above P in some extension Gn ⊇ G0 .
However, the following weaker assertion holds.
 7.4 Some Wild Towers 265

Lemma 7.4.6. Let G = (G0 , G1 , G2 , . . .) be the tower over IFq which is defined
by Equation (7.14). Let P ∈ IPG0 be a place in the ramification locus of G/G0
and let Q ∈ IPGn be a place lying above P . Then
d(Q|P ) = 2e(Q|P ) − 2 .

We postpone the proof of this lemma, and we first draw an important

conclusion:

Theorem 7.4.7 (Garcia-Stichtenoth). Let q = 2 . Then the equation

−1
Y − Y = X /(1 − X )
defines a recursive tower G = (G0 , G1 , G2 , . . .) over IFq , whose limit is

λ(G) =  − 1 = q 1/2 − 1 .
Therefore the tower G is optimal.
In fact, the first example of a tower over IFq with q = 2 , attaining the limit
 − 1, was provided by Y. Ihara. He used modular curves for his construction.

Corollary 7.4.8 (Ihara). A(q) = q 1/2 − 1, if q is a square.

Proof of Theorem 7.4.7. We will apply Theorem 7.2.10. The splitting locus
Split(G/G0 ) has cardinality
|Split(G/G0 )| =: s ≥ 2 − 
by Lemma 7.4.4. The ramification locus Ram(G/G0 ) has cardinality
|Ram(G/G0 )| ≤  + 1
by Lemma 7.4.5. All places P ∈ Ram(G/G0 ) have degree one, and for all
n ≥ 0 and all Q ∈ IPGn above P the different exponent is bounded by
d(Q|P ) ≤ 2e(Q|P )
by Lemma 7.4.6. Now the formula in Theorem 7.2.10(c) gives
2(2 − )
λ(G) ≥ =  − 1.
−2 + 2( + 1)
By the Drinfeld-Vladut Bound we have also the opposite inequality λ(G) ≤
 − 1, therefore the equality λ(G) =  − 1 holds.

Remark 7.4.9. One can easily show that in Lemma 7.4.4 and Lemma 7.4.5
equality holds; i.e.,
Split(G/G0 ) = {(x0 = α) | α ∈ IFq \ IF } , and
Ram(G/G0 ) = {(x0 = β) | β ∈ IF or β = ∞} .
266 7 Asymptotic Bounds for the Number of Rational Places

Remark 7.4.10. One can determine exactly the genus and the number of ra-
tional places for each field Gn in the tower G. However, this requires lengthy
and very technical calculations.

We still have to prove Lemma 7.4.6. To this end we introduce some notation
which will be useful also in connection with other wild towers. We recall a
property of different exponents in a Galois extension E/F of function fields
of degree [E : F ] = p = char F . If P ∈ IPF and Q ∈ IPE are places with Q|P ,
then

d(Q|P ) = k · (e(Q|P ) − 1) for some integer k ≥ 2 . (7.20)

This follows immediately from Hilbert’s Different Formula. Note that (7.20)
holds also if e(Q|P ) = 1. In the case of e(Q|P ) = p, the integer k in (7.20) is
determined by the higher ramification groups of Q|P as follows:

ord Gi (Q|P ) = p for 0 ≤ i < k , and ord Gk (Q|P ) = 1 .

If we have k = 2 in (7.20), we say that Q|P is weakly ramified. We want to

generalize this notion to certain extensions of degree pm ≥ p.

Remark 7.4.11. Let E/F be an extension of function fields of degree [E : F ] =

pm with p = char F . Assume that there exists a chain of intermediate fields
F = E0 ⊆ E1 ⊆ . . . ⊆ En = E with the property

Ei+1 /Ei is Galois for all 0 ≤ i < n .

Let P ∈ IPF and Q ∈ IPE with Q|P , and denote the restriction of Q to Ei by
Qi := Q ∩ Ei . Then the following conditions are equivalent:
(1) d(Q|P ) = 2(e(Q|P ) − 1).
(2) d(Qi+1 |Qi ) = 2(e(Qi+1 |Qi ) − 1) for all i = 0, . . . , n − 1.

Proof. (2) ⇒ (1): We assume (2) and show by induction that

d(Qi+1 |P ) = 2(e(Qi+1 |P ) − 1) (7.21)

holds for 0 ≤ i ≤ n − 1. The case i = 0 is trivial since P = Q0 . Assume (7.21)

for some i with 0 ≤ i ≤ n − 2. Then we obtain, by transitivity of different
exponents,

d(Qi+2 |P ) = e(Qi+2 |Qi+1 ) · d(Qi+1 |P ) + d(Qi+2 |Qi+1 )

= e(Qi+2 |Qi+1 ) · 2(e(Qi+1 |P ) − 1) + 2(e(Qi+2 |Qi+1 ) − 1)
= 2(e(Qi+2 |P ) − 1) .

Thus we have established the induction step. Setting i := n − 1 in (7.21), we

get d(Q|P ) = 2(e(Q|P ) − 1).
 7.4 Some Wild Towers 267

(1) ⇒ (2): Now we assume that (1) holds. Since the extensions Ei+1 /Ei are
Galois extensions of degree pni (for some ni ≥ 0), Hilbert’s Different Formula
shows that d(Qi+1 |Qi ) ≥ 2(e(Qi+1 |Qi ) − 1) holds for all i. If this inequality is
strict for some i ∈ {0, . . . , n − 1}, then the transitivity of different exponents
yields d(Q|P ) > 2(e(Q|P ) − 1), as in the proof of (2) ⇒ (1). This contradicts
the assumption (1), so we have d(Qi+1 |Qi ) = 2(e(Qi+1 |Qi ) − 1) for all i.

Definition 7.4.12. Let F be a function field with char F = p > 0. A finite

extension E/F is said to be weakly ramified, if the following conditions hold:
(1) There exist intermediate fields F = E0 ⊆ E1 ⊆ . . . ⊆ En = E such
that all extensions Ei+1 /Ei are Galois p-extensions (i.e., [Ei+1 : Ei ]
is a power of p), for i = 0, 1, . . . , n − 1.
(2) For all places P ∈ IPF and Q ∈ IPE with Q|P , the different exponent
is given by d(Q|P ) = 2(e(Q|P ) − 1).

The following proposition is crucial for the proof of Lemma 7.4.6 (and also
for proving that some other wild towers are asymptotically good).

Proposition 7.4.13. Let E/F be a finite extension of function fields and let
M , N be intermediate fields of E ⊇ F such that E = M N is the compositum
of M and N . Assume that both extensions M/F and N/F are weakly ramified.
Then E/F is weakly ramified.

Proof. The special case [M : F ] = [N : F ] = p has been considered in

Proposition 3.9.4. The idea of proof here is to reduce the general case to this
special case. There is a sequence of intermediate fields

F = M 0 ⊆ M1 ⊆ . . . ⊆ Mk = M (7.22)

such that all extensions Mi+1 /Mi are weakly ramified Galois p-extensions. It
is a well-known fact from group theory that every finite p-group G contains
a chain of subgroups {1} = G0 ⊆ G1 ⊆ . . . ⊆ Gs = G, where Gj is a normal
subgroup of Gj+1 of index (Gj+1 : Gj ) = p for j = 0, . . . , s − 1. By Galois
theory we can therefore refine the extensions Mi ⊆ Mi+1 , to obtain Galois
steps of degree p; i.e.,
(0) (1) (ki )
M i = Mi ⊆ Mi ⊆ . . . ⊆ Mi = Mi+1
(j+1) (j)
with weakly ramified Galois extensions Mi /Mi of degree p. Therefore
we can assume a priori that the extensions Mi+1 /Mi in the chain (7.22) are
all Galois of degree p.
In the same way we split the extension N/F into weakly ramified Galois steps
of degree p. By induction over the degree [N : F ], the proof of Proposition
7.4.13 is thus reduced to the case where N/F is Galois of degree [N : F ] = p.
So we have the following situation: F = M0 ⊆ M1 ⊆ . . . ⊆ Mk = M with
268 7 Asymptotic Bounds for the Number of Rational Places

weakly ramified Galois extensions Mi+1 /Mi of degree p, and E = M N where

N/F is a weakly ramified Galois extension of degree p. The extension E/M
is then Galois of degree 1 or p and we have to show that E/M is also weakly
ramified. If [E : M ] = 1 there is nothing to prove, so we consider now the case
[E : M ] = p. Setting Ni := Mi N for i = 0, . . . , k we obtain a chain of Galois
extensions N = N0 ⊆ N1 ⊆ . . . ⊆ Nk = E with [Ni+1 : Ni ] = p, and also
the extensions Ni /Mi are Galois of degree p. Then it follows from Proposition
3.9.4 by induction over i that all extensions Ni /Mi are weakly ramified, and
hence E/M is weakly ramified.

Now we are in a position to prove Lemma 7.4.6, and thereby to finish the
proof of Theorem 7.4.7.

Proof of Lemma 7.4.6. We consider again the recursive tower G = (G0 , G1 , . . .)

over IFq with q = 2 , which is defined recursively by the equation Y − Y =
X /(1 − X −1 ). The assertion of Lemma 7.4.6 is that all extensions Gn /G0
are weakly ramified. We prove this by induction over n. The fields G0 and G1
are given by G0 = IFq (x0 ) and G1 = IFq (x0 , x1 ) with

x1 − x1 = x0 /(1 − x0−1 ) . (7.23)

As follows from Proposition 3.7.8, exactly the places (x0 = ∞) and (x0 = β)
with β ∈ IF× are ramified in the extensions G1 /G0 , with ramification index
e =  and different exponent d = 2( − 1). So G1 /G0 is weakly ramified.
For the induction step we assume that Gn /G0 is weakly ramified, and we
have to show that also the extension Gn+1 /Gn is weakly ramified. We set
Li := IFq (x1 , . . . , xi ) for 1 ≤ i ≤ n + 1, see Figure 7.2 below. The field Ln+1
is IFq -isomorphic to Gn via the isomorphism ϕ : xj → xj+1 for 0 ≤ j ≤ n;
therefore Ln+1 /L1 is weakly ramified by induction hypothesis.

Now we observe that Equation (7.23) can be rewritten as

 
1 1 1
− = .
x0 x0 x1 − x1

This shows that also the extension G1 /L1 is an Artin-Schreier extension of de-
gree . In the same way as for the extension G1 /G0 , it follows from Proposition
3.7.10 that G1 /L1 is weakly ramified (the ramified places are just the places
(x1 = α) with α ∈ IF ). Since Gn+1 is the compositum of G1 and Ln+1 over
L1 , Proposition 7.4.13 implies that Gn+1 /L1 is weakly ramified. Consequently
the extension Gn+1 /Gn is weakly ramified, by Remark 7.4.11.

Although each step Gi+1 /Gi in the tower G is Galois (by Lemma 7.4.3),
the extensions Gi /G0 are not Galois for i ≥ 2. This follows from the fact that
the place (x0 = 0) ∈ IPG0 has unramified as well as ramified extensions in Gi .
We therefore ask if it is possible to “extend” the tower G = (G0 , G1 , G2 , . . .)
 7.4 Some Wild Towers 269

Gn+1
@
@
@
Gn Ln+1
@
@
@
q Ln
q
q

q
q
q
G1
@
@
@
G0 = IFq (x0 ) L1 = IFq (x1 )

Fig. 7.2.

to a tower G ∗ = (G∗0 , G∗1 , G∗2 , . . .) of function fields G∗i over IFq having the
following properties:
(1) G is a subtower of G ∗ ,
(2) G∗i /G∗0 is Galois for all i ≥ 0,
(3) the tower G ∗ is optimal over IFq ; i.e., λ(G ∗ ) = q 1/2 − 1.
An obvious choice would be to take G∗i as the Galois closure of Gi over G0 .
With regard to an application in coding theory in Chapter 8, we proceed in a
slightly different manner. We will need the following lemma.

Lemma 7.4.14. Let K(x) be the rational function field over a field K ⊇ IF .
Consider the subfields K(u) ⊆ K(t) ⊆ K(x) with
−1 −1
t := x − x and u := (x − x) +1 = t + 1. (7.24)

Then the following hold:

(a) The extensions K(x)/K(u), K(x)/K(t) and K(t)/K(u) are Galois of
degree

[K(x) : K(u)] = ( − 1) , [K(x) : K(t)] =  , and [K(t) : K(u)] =  − 1 .

(b) The place (u = ∞) of K(u) is totally ramified in the extension K(x)/K(u).

The place of K(x) above (u = ∞) is the pole (x = ∞) of x, and the place of
K(t) above (u = ∞) is the pole (t = ∞) of t. The ramification indices and
different exponents are
270 7 Asymptotic Bounds for the Number of Rational Places

e((t = ∞)|(u = ∞)) =  − 1 and d((t = ∞)|(u = ∞)) =  − 2 ,

e((x = ∞)|(t = ∞)) =  and d((x = ∞)|(t = ∞)) = 2( − 1) ,
e((x = ∞)|(u = ∞)) = ( − 1) and d((x = ∞)|(u = ∞)) = 2 − 2 .
Hence (x = ∞)|(t = ∞) is weakly ramified.
(c) The place (u = 1) is totally ramified in K(t)/K(u), the place in K(t) above
(u = 1) is the place (t = 0), with ramification index e((t = 0)|(u = 1)) =  − 1
and different exponent d((t = 0)|(u = 1)) =  − 2. In K(x)/K(t) the place
(t = 0) splits completely. The places of K(x) above (t = 0) are exactly the
places (x = β) with β ∈ IF .
(d) No other places of K(u), except (u = ∞) and (u = 1), are ramified in
K(x)/K(u).
(e) If IF 2 ⊆ K then the place (u = 0) of K(u) splits completely in K(x)/K(u);
the places of K(x) above (u = 0) are exactly the places (x = α) with α ∈
IF 2 \ IF .
Proof. (a) We consider the two subgroups U0 , U1 of the automorphism group
of K(x)/K which are defined by
U0 := {σ : x → ax + b | a ∈ IF× , b ∈ IF },
U1 := {σ : x → x + b | b ∈ IF } ⊆ U0 .
It is clear that ord U0 = ( − 1) and ord U1 = , and U1 is a normal subgroup
of U0 . One checks that u is invariant under all σ ∈ U0 , and t is invariant under
all σ ∈ U1 . Since [K(x) : K(u)] = ( − 1) and [K(x) : K(t)] =  by (7.24),
it follows that K(x)/K(u) is Galois with Galois group U0 , and K(x)/K(t) is
Galois with Galois group U1 .
(b) The assertions concerning ramification indices are obvious, since u and t
are polynomials in x. The only non-trivial statements of (b) are
d((x = ∞)|(t = ∞)) = 2( − 1) and d((x = ∞)|(u = ∞)) = 2 − 2 . (7.25)
In order to prove this, we note that the extension K(x)/K(t) is Galois of order
 and e((x = ∞)|(t = ∞)) = . Therefore
d((x = ∞)|(t = ∞)) ≥ 2( − 1)
by Hilbert’s Different Formula. On the other hand, Hurwitz Genus Formula
for K(x)/K(t) yields
−2 = −2 + deg Diff(K(x)/K(t))
and hence
d((x = ∞)|(t = ∞)) = deg Diff(K(x)/K(t)) = 2( − 1) .
So we have d((x = ∞)|(t = ∞)) = 2( − 1). By using the transitivity of the
different exponent in the extensions K(x) ⊇ K(t) ⊇ K(u), we obtain easily
that d((x = ∞)|(u = ∞)) = 2 − 2.
 7.4 Some Wild Towers 271

(c) Since u − 1 = t −1 , it is clear that (t = 0) is the only place of K(t) above

the place (u = 1) of K(u), with ramification index e((t = 0)|(u = 1)) =  − 1
and different exponent d((t = 0)|(u = 1)) =  − 2. The equation x − x = t
shows that (t = 0) splits completely in K(x)/K(t), and the places of K(x)
above (t = 0) are exactly the places (x = β) with β − β = 0.
(d) This follows by Hurwitz Genus Formula for the extension K(x)/K(u) and
parts (b) and (c).
(e) Note that

u = (x − x) −1 + 1 = (x − x)/(x − x) =
2
(x − α) ,
α∈IF2 \IF

so that assertion (e) follows immediately.

We return to the construction of the tower G ∗ over IFq with q = 2 . We

start with the recursive tower G = (G0 , G1 , G2 , . . .) with the defining equation
Y − Y = X /(1 − X −1 ), so we have

xi
G0 = IFq (x0 ) and Gi+1 = Gi (xi+1 ) with xi+1 − xi+1 =
1 − xi −1

for all i ≥ 0. We set

t0 := x0 − x0 and u0 := t0−1 + 1 = (x0 − x0 ) −1

+1 . (7.26)

Then we define G∗0 := IFq (u0 ) and

G∗i := Galois closure of Gi over IFq (u0 ) , for i ≥ 1 . (7.27)

Thus we obtain a sequence of fields

G∗0 = IFq (u0 ) ⊆ IFq (t0 ) ⊆ IFq (x0 ) ⊆ G∗1 ⊆ G∗2 ⊆ . . . ,

where IFq (t0 )/IFq (u0 ) is a Galois extension of degree  − 1 and IFq (x0 )/IFq (t0 )
is Galois of degree . The extension IFq (x0 )/IFq (u0 ) is also Galois, by Lemma
7.4.14.

Theorem 7.4.15. Let q = 2 be a square. With the above notation, the fol-
lowing hold:
(a) IFq is the full constant field of G∗i for all i ≥ 0. Therefore the sequence

G ∗ := (G∗0 , G∗1 , G∗2 , . . .)

is a tower over IFq , where G∗0 = IFq (u0 ) is rational and all extensions G∗i /G∗0
are Galois.
(b) The place (u0 = 0) of IFq (u0 ) splits completely in all extensions G∗i /IFq (u0 ).
272 7 Asymptotic Bounds for the Number of Rational Places

(c) The places (u0 = ∞) and (u0 = 1) are the only places of IFq (u0 ) which
are ramified in the tower G ∗ over IFq (u0 ). Both of them are totally ramified
in the extension IFq (t0 )/IFq (u0 ) of degree  − 1.
(d) The extensions G∗i /IFq (t0 ) are weakly ramified Galois p-extensions (with
p = char IFq ).
(e) The tower G ∗ attains the Drinfeld-Vladut Bound

λ(G ∗ ) =  − 1 = q 1/2 − 1 .

Proof. We consider the field extensions

IFq (u0 ) ⊆ IFq (t0 ) ⊆ IFq (x0 ) = G0 ⊆ G1 ⊆ G2 ⊆ . . . . (7.28)

The first step IFq (t0 )/IFq (u0 ) is Galois of degree  − 1, and all other steps
in (7.28) are weakly ramified Galois extensions of degree  (by Lemma
7.4.14 and Lemma 7.4.6). The place (u0 = 0) of IFq (u0 ) splits completely in
IFq (x0 )/IFq (u0 ). The places of IFq (x0 ) above (u0 = 0) are exactly the places
(x0 = α) with α ∈ IFq \ IF (Lemma 7.4.14(e)). These places (x0 = α) are in
the splitting locus of the tower G over G0 = IFq (x0 ) (Lemma 7.4.4), and we
conclude that the place (u0 = 0) splits completely in all extensions Gi /IFq (u0 ).
It follows from Corollary 3.9.7, that (u0 = 0) splits completely in the Galois
closure G∗i of Gi over IFq (u0 ) and that IFq is the full constant field of G∗i . We
have thus proved parts (a) and (b).
(c) Let P be a place of IFq (u0 ) which is different from (u0 = 1) and (u0 = ∞).
Then P is unramified in G0 = IFq (x0 ), and the places of G0 lying above P
are different from the places (x0 = β) with β ∈ IF ∪ {∞} (Lemma 7.4.14).
By Lemma 7.4.5, the ramification locus of G over G0 is contained in the set
{(x0 = β) | β ∈ IF ∪ {∞}}. It follows that P is unramified in all extensions
Gi /IFq (u0 ). Thus P is unramified in the Galois closure G∗i of Gi over IFq (u0 ),
by Corollary 3.9.3. The rest of part (c) follow immediately from Lemma 7.4.14.
(d) We denote the Galois group of G∗i /IFq (u0 ) by Γi . Every τ ∈ Γi maps the
field IFq (t0 ) to itself since IFq (t0 )/IFq (u0 ) is Galois. By Lemma 7.4.14(b) and
Lemma 7.4.6, the extension Gi /IFq (t0 ) is weakly ramified, hence τ (Gi )/IFq (t0 )
is weakly ramified for all τ ∈ Γi . The field G∗i is the compositum of the fields
τ (Gi ) with τ ∈ Γi , and we conclude from Proposition 7.4.13 that also the
extension G∗i /IFq (t0 ) is weakly ramified.
(e) We will apply Theorem 7.2.10(c) to the tower G ∗ over IFq (u0 ); to this
end we need an estimate for the different exponents of ramified places in
G∗n /IFq (u0 ) as in (7.7). So we consider a place P of IFq (u0 ) which ramifies in
G∗n (we have P = (u0 = ∞) or P = (u0 = 1) by (c)). Let Q∗ be a place of G∗i
lying above P and set Q := Q∗ ∩ IFq (t0 ). Then Q|P is tamely ramified with
ramification index e(Q|P ) =  − 1 by (c), and Q∗ |Q is weakly ramified by (d).
Using transitivity of different exponents we obtain
 7.4 Some Wild Towers 273

d(Q∗ |P ) = d(Q∗ |Q) + e(Q∗ |Q) · d(Q|P )

= 2(e(Q∗ |Q) − 1) + ( − 2) · e(Q∗ |Q)

<  · e(Q∗ |Q) = · e(Q∗ |P ) . (7.29)
−1
So we have the inequality

d(Q∗ |P ) ≤ aP · e(Q∗ |P ) with aP := .
−1
As we have shown in part (b), the place (u0 = 0) of IFq (u0 ) splits completely
in the tower G ∗ . Now Theorem 7.2.10 gives the estimate
2
λ(G ∗ ) ≥ =  − 1.
−2 + 2 · −1

The inequality λ(G ∗ ) ≤  − 1 follows from the Drinfeld-Vladut Bound, hence

we have λ(G ∗ ) =  − 1.

In connection with an application in coding theory (see Section 8.4) we

note some specific properties of the tower G ∗ :

Corollary 7.4.16. With the notations of Theorem 7.4.15, we set

ni := [G∗i : IFq (u0 )] = ( − 1) · mi (7.30)

for every i ≥ 0, so mi = [G∗i : IFq (t0 )] is a power of p = char IFq . Denote by

the ramification index of the place (t0 = 0) in G∗i /IFq (t0 ) ,

(0)
ei ,
the ramification index of the place (t0 = ∞) in G∗i /IFq (t0 ) .
(∞)
ei ,

Thus the principal divisor of t0 in G∗i is

∗ (0) (∞)
(t0 )Gi = ei Ai − ei Bi (7.31)

with positive divisors Ai , Bi ∈ Div(G∗i ). Then the following hold:

(a) The function field G∗i has the genus
 
∗ 1 1
g(Gi ) = 1 + mi 1 − (0) − (∞) .
ei ei

(b) The zero divisor of u0 in G∗i has the form


ni
(i)
Di = Pj , (7.32)
j=1

(i)
with ni rational places Pj ∈ IPG∗i .
274 7 Asymptotic Bounds for the Number of Rational Places

(c) The differential η (i) := du0 /u0 of G∗i has the divisor
(0) (∞)
(η (i) ) = (ei − 2)Ai + (ei − 2)Bi − Di ,
where the divisors Ai , Bi , Di are defined by (7.31) and (7.32). At all places
(i)
P = Pj ≤ Di , the residue of η (i) is

resP (η (i) ) = 1 .
Proof. The places (t0 = 0) and (t0 = ∞) are the only places of IFq (t0 ) which
ramify in the extension G∗i /IFq (t0 ). Since they are weakly ramified (Theorem
7.4.15(c), (d)), the different of G∗i /IFq (t0 ) is given by

Diff(G∗i /IFq (t0 )) = (2ei

(0) (∞)
− 2)Ai + (2ei − 2)Bi ,
(with the divisors Ai , Bi as in (7.31)). Then the differential dt0 (as a differen-
tial of the field G∗i ) has the divisor

Bi + Diff(G∗i /IFq (t0 ))

(∞)
(dt0 ) = −2ei
(0)
= (2ei − 2)Ai − 2Bi , (7.33)
(0) (∞)
by Remark 4.3.7. Since deg Ai = mi /ei
and deg Bi = we obtainmi /ei ,
m m
2g(G∗i ) − 2 = deg(dt0 ) = (2ei − 2) · (0) − 2 · (∞) .
(0) i i

ei ei
This proves (a). Part (b) is clear from Theorem 7.4.15(b). In order to show
part (c) we note that u0 = 1 + t0−1 by (7.26), hence du0 = −t0−2 dt0 and
(0) (∞)
(du0 ) = ( − 2) · (ei Ai − ei Bi ) + (dt0 )
(0) (∞)
= (ei − 2)Ai − (( − 2)ei + 2)Bi .
The principal divisor of u0 in G∗i is
∗ (∞)
(u0 )Gi = Di − ( − 1)ei Bi ,
so the differential η (i) := du0 /u0 has the divisor (in G∗i )
(0) (∞)
(η (i) ) = (ei − 2)Ai + (ei − 2)Bi − Di .
At all places P in the support of Di , the element u0 is a prime element,
consequently we have resP (du0 /u0 ) = 1 by the definition of the residue of a
differential.

Now we turn to finite fields IFq where q = 3 is a cube. We investigate the

recursive tower H over IFq , which is given by the equation

−1 −X ( −1)
(Y − Y ) +1 = . (7.34)
(X −1 − 1) −1
Our main result is
 7.4 Some Wild Towers 275

Theorem 7.4.17. Equation (7.34) defines a recursive tower H over the field
IFq with q = 3 . Its limit λ(H) satisfies

2(2 − 1)
λ(H) ≥ .
+2
The following lower bound for A(3 ) is due to T. Zink in the case of a
prime number ; for arbitrary  it was first shown by Bezerra, Garcia and
Stichtenoth.

Corollary 7.4.18. For q = 3 , Ihara’s constant A(q) is bounded below by

2(2 − 1)
A(q) ≥ .
+2

We will prove Theorem 7.4.17 in several steps. It is convenient to consider

H not only over the field IFq but also over any field K which contains IF . So
we have H = (H0 , H1 , H2 , . . .) where H0 = K(y0 ) is a rational function field
over K, and
( −1)
−1 −yi
Hi+1 = Hi (yi+1 ) with (yi+1 − yi+1 ) +1 = (7.35)
(yi −1 − 1) −1

for all i ≥ 0. We denote the generators of H by y0 , y1 , y2 , . . . in order to avoid

confusion with the generators x0 , x1 , x2 , . . . of the tower G.
The best way to understand the tower H is to study its second basic
function field H := K(x, y, z) which is defined by the equations
( −1)
−1 −x
(y − y) +1 = −1 −1
=: u (7.36)
(x − 1)

and
( −1)
−1 −y
(z − z) +1 = −1 −1
=: v . (7.37)
(y − 1)
First we determine the degree of H over some of its subfields as shown in
Figure 7.3 below.

Lemma 7.4.19. Assume that IF ⊆ K. With the above notation we have:

(a) The extensions K(x)/K(u), K(y)/K(u), K(y)/K(v) and K(z)/K(v) are
Galois of degree ( − 1).
(b) Also the extensions at the next level, K(x, y)/K(x), K(x, y)/K(y),
K(y, z)/K(y) and K(y, z)/K(z) are Galois of degree ( − 1).
(c) The extensions K(x, y, z)/K(x, y) and K(x, y, z)/K(y, z) are Galois of
degree .
(d) K is the full constant field of H = K(x, y, z).
276 7 Asymptotic Bounds for the Number of Rational Places

H = K(x, y, z)
@
@
@
K(x, y) K(y, z)
@ @
@ @
@ @
K(x) K(y) K(z)
@ @
@ @
@ @
K(u) K(v)

Fig. 7.3. The second basic function field H.

Proof. (a) By Lemma 7.4.14(a), the extension K(y)/K(u) is Galois of degree

[K(y) : K(u)] = ( − 1), since u = (y − y) −1 + 1. Now we observe that the
equation u = −x ( −1) /(x −1 − 1) −1 is equivalent to
   −1
1 1 u−1
− +1 = . (7.38)
x x u

As K(x) = K(1/x) and K(u) = K((u−1)/u), we conclude again from Lemma

7.4.14(a) that K(x)/K(u) is a Galois extension of degree ( − 1). The asser-
tions for the extensions K(y)/K(v) and K(z)/K(v) follow analogously.
(b) We consider the extension K(x, y)/K(x). Since K(y)/K(u) is Galois, it
follows from Galois theory that K(x, y)/K(x) is Galois, and its Galois group
Gal(K(x, y)/K(x)) is isomorphic to a subgroup of Gal(K(y)/K(u)). It re-
mains to show that [K(x, y) : K(x)] ≥ ( − 1). To this end we consider
ramification in some subextensions of the field H as in Figure 7.4.
For a place R ∈ IPH we denote its restrictions to the corresponding subfields
of H according to Figure 7.4 ; this means for instance that P = R ∩ K(x)
and Q̃ = R ∩ K(v). Specifically we choose a place Q of K(x, y) which is a
zero of the element u. We then have P ∗ = (u = 0). From Equation (7.36) we
see that P = (x = 0) and e(P |P ∗ ) = ( − 1). On the other hand it follows
from Lemma 7.4.14(d) that e(Q∗ |P ∗ ) = 1 and hence e(Q|Q∗ ) = ( − 1). We
conclude that [K(x, y) : K(x)] = [K(x, y) : K(y)] ≥ ( − 1).
Thus we have shown that K(x, y)/K(x) is a Galois extension of degree
( − 1). The corresponding claims for the other extensions K(x, y)/K(y),
K(y, z)/K(y) and K(y, z)/K(z) follow immediately.
(c) Next we choose a place R of K(x, y, z) which is a pole of x. Thus P =
(x = ∞), and we obtain from Equations (7.36) and (7.37) that the places P ∗ ,
 7.4 Some Wild Towers 277

R
@
@
@
Q R∗
@ @
@ @
@ @
P Q∗ R̃
@ @
@ @
@ @
P∗ Q̃

Fig. 7.4. Places of subfields of H.

Q∗ , Q̃ and R̃ are as follows:

P ∗ = (u = ∞) with e(P |P ∗ ) =  − 1 ,
Q∗ = (y = ∞) with e(Q∗ |P ∗ ) = ( − 1) ,
Q̃ = (v = ∞) with e(Q∗ |Q̃) =  − 1 ,
R̃ = (z = ∞) with e(R̃|Q̃) = ( − 1) .

Now Abhyankar’s Lemma yields e(Q|Q∗ ) = 1, e(R∗ |Q∗ ) =  and e(R|Q) = 

(observe the notation in Figure 7.4), and it follows that

[K(x, y, z) : K(x, y)] ≥ e(R | Q) =  . (7.39)

On the other hand we get from (7.36) and (7.37), that

( −1)
−1 −y
(z − z) = −1
(y −1 − 1) −1
−((y − y) −1 + 1)
=
(y −1 − 1) −1
x ( −1)
= −1
.
(x − 1) −1 (y −1 − 1) −1

Therefore we have
x
z −z =μ· −1 −1
(7.40)
(x − 1)(y − 1)

for some μ ∈ IF× . This is an equation for z over K(x, y) of degree ,

so [K(x, y, z) : K(x, y)] ≤ . With (7.39) we conclude that the extension
278 7 Asymptotic Bounds for the Number of Rational Places

K(x, y, z)/K(x, y) has degree . From Equation (7.40) it also follows that
K(x, y, z)/K(x, y) is Galois.
(d) We have seen in the proof of (b) and (c) that in both extensions
K(x, y)/K(x) and K(x, y, z)/K(x, y) there are totally ramified places. The-
orem 3.6.3(a) implies that K is the full constant field of K(x, y) and of
K(x, y, z).

Next we determine all places of the field H = K(x, y, z) which are ramified
over the field K(u) (where u is given by Equation (7.36)). For simplicity we
can assume that K contains the field IF 2 , since the ramification behavior
(i.e., ramification index and different degree) does not change under constant
field extensions. The following lemma is of fundamental importance for the
investigation of the tower H.

Lemma 7.4.20. We maintain the notation of Equations (7.36), (7.37) and

Figures 7.3, 7.4, and also assume that IF 2 ⊆ K. Let R be a place of H which
is ramified in the extension H/K(u). Then the restriction P ∗ = R ∩ K(u) of
R to K(u) is one of the places (u = 0), (u = 1) or (u = ∞). More precisely
the following hold:
Case 1. P ∗ = (u = 0). Then P = (x = 0),

e(Q|P ) = e(R|Q) = 1 and e(R|R∗ ) =  .

Case 2. P ∗ = (u = 1). Then P = (x = β) with β ∈ IF 2 \ IF ,

e(Q|P ) =  − 1 , e(R|Q) ∈ {1, } and e(R|R∗ ) = 1 .

Case 3. P ∗ = (u = ∞). Then P = (x = α) with some α ∈ IF× ∪ {∞},

Q∗ = (y = ∞),

e(Q|P ) = e(R|Q) =  and e(Q|Q∗ ) = e(R|R∗ ) = e(R∗ |R̃) = 1 .

In all cases above where the ramification index is e = , the corresponding

different exponent is d = 2( − 1).

Proof. Since R|P ∗ is ramified, at least one of the places P |P ∗ or R∗ |P ∗ is

ramified, see Figure 7.4. We distinguish several cases:
(i) Assume that P |P ∗ is ramified. Since K(x) = K(1/x) and
   −1
1 1 u−1
− +1 = =: u
x x u

(see Equation (7.38)), it follows from Lemma 7.4.14 that P ∗ = (u = ∞) or

P ∗ = (u = 1). Since u = ∞ ⇔ u = 0 and u = 1 ⇔ u = ∞, we conclude that
P ∗ = (u = 0) or P ∗ = (u = ∞).
 7.4 Some Wild Towers 279

(ii) Now we assume that R∗ |P ∗ is ramified. Then one of the places Q∗ |P ∗ or

R∗ |Q∗ is ramified.
(ii1 ) If Q∗ |P ∗ is ramified then P ∗ = (u = ∞) or P ∗ = (u = 1) by Lemma
7.4.14.
(ii2 ) If R∗ |Q∗ is ramified then R̃|Q̃ is ramified (see Figure 7.4), hence
Q̃ = (v = ∞) or Q̃ = (v = 1) by Lemma 7.4.14. First we discuss the case
Q̃ = (v = ∞). By Equation (7.37) we have then Q∗ = (y = ∞) or
Q∗ = (y = γ) with γ ∈ IF× . If Q∗ = (y = ∞) then P ∗ = (u = ∞), and
if Q∗ = (y = γ) then P ∗ = (u = 1) by Equation (7.36).
It remains to consider the case Q̃ = (v = 1) = (v  = 0) where v  := (v − 1)/v.
From the equation
   −1
1 1
− + 1 = v
y y
(see Equation (7.38)) and Lemma 7.4.14(e) we obtain Q∗ = (1/y = α) with
α ∈ IF 2 \ IF , hence Q∗ = (y = δ) with δ = α−1 ∈ IF 2 \ IF . Again it follows
from Lemma 7.4.14(e) that P ∗ = (u = 0).
So far we have shown that if R|P ∗ is ramified then P ∗ = (u = 0) or (u = 1)
or (u = ∞). Now we have to discuss these three cases.
Case 1. P ∗ = (u = 0). Using Equations (7.36), (7.37) and Lemma 7.4.14
in the same way as above we find that

P = (x = 0) and Q∗ = (y = γ) with γ ∈ IF 2 \ IF ,
Q̃ = (v = 1) and R̃ = (z = β) with β ∈ IF ,

with ramification indices (resp. different exponents)

e(P | P ∗ ) = ( − 1) and d(P | P ∗ ) = 2 − 2 ,

e(Q∗ | P ∗ ) = e(Q∗ | Q̃) = 1 ,
e(R̃ | Q̃) =  − 1 .

Going up one level in Figure 7.4 it follows immediately that

e(Q | P ) = 1 and e(R∗ | Q∗ ) =  − 1 ,

e(Q | Q∗ ) = ( − 1) and d(Q | Q∗ ) = 2 − 2 .

We go up another level in Figure 7.4, apply Abhyankar’s Lemma and get

e(R | Q) = 1 ,
e(R | R∗ ) =  and d(R | R∗ ) = 2( − 1) .

This finishes the proof of Lemma 7.4.20 in Case 1. The other two cases are
similar.
280 7 Asymptotic Bounds for the Number of Rational Places

Corollary 7.4.21. Both extensions K(x, y, z)/K(x, y) and K(x, y, z)/K(y, z)

are weakly ramified Galois extensions of degree .

We are now able to prove that Equation (7.34) actually defines a tower.

Proposition 7.4.22. Let K be a field with IF ⊆ K, and consider the sequence

H = (H0 , H1 , H2 , . . .) where H0 = K(y0 ) is a rational function field, and
Hi+1 = Hi (yi+1 ) with
( −1)
−1 −yi
(yi+1 − yi+1 ) +1 =
(yi −1 − 1) −1

for all i ≥ 0. Then H is a tower over K. The extension H1 /H0 is Galois of

degree [H1 : H0 ] = ( − 1), and for all i ≥ 1, the extension Hi+1 /Hi is Galois
of degree [Hi+1 : Hi ] = .

Proof. The field H2 = K(y0 , y1 , y2 ) is isomorphic to the function field H =

K(x, y, z) that we studied in Lemmas 7.4.19 and 7.4.20. Therefore we know
already that H1 /H0 is Galois of degree ( − 1), the extension H2 /H1 is
Galois of degree  and K is the full constant field of H2 . Now let i ≥ 2. Since
Hi+1 = Hi (yi+1 ) and yi+1 satisfies the equation

yi−1
yi+1 − yi+1 = μ · −1
(yi−1 − 1)(yi −1 − 1)

with some element μ ∈ IF× (cf. Equation (7.40)), the extension Hi+1 /Hi is
Galois of degree [Hi+1 : Hi ] ≤ . We show now by induction the following
claim which readily implies the remaining assertions of Proposition 7.4.22.
We fix a place P1 ∈ IPH1 which is a pole of the element y0 in H1 .
Claim. Let i ≥ 2 and let Pi be a place of Hi lying above P1 . Then Pi is
also a pole of yi , and we have

e(Pi | Pi−1 ) =  and e(Pi | (yi = ∞)) = 1 .

Proof of the Claim. The case i = 2 follows from Case 3 of Lemma 7.4.20.
So we assume now that the claim holds for some i ≥ 2. Choose a place
Pi+1 of Hi+1 which lies above Pi . The field Hi+1 is the compositum of
the fields Hi and K(yi , yi+1 ) over K(yi ). By induction hypothesis we have
∗
that Pi ∩ K(yi ) = (yi = ∞) and e(Pi |(yi = ∞)) = 1. We set Pi+1 :=
Pi+1 ∩K(yi , yi+1 ) and P̃i+1 := Pi+1 ∩K(yi+1 ). It follows from Case 3 of Lemma
∗
7.4.20 that P̃i+1 is the pole of yi+1 in K(yi+1 ) and that e(Pi+1 |(yi = ∞)) = 
∗
and e(Pi+1 |P̃i+1 ) = 1. The situation is shown below in Figure 7.5.
From this picture we see that e(Pi+1 |Pi ) =  and e(Pi+1 |(yi+1 = ∞)) = 1.
This completes the proof of Proposition 7.4.22.
 7.4 Some Wild Towers 281

Hi+1 Pi+1
• •
@ @
Hi @ Pi @
• @ • @
@ @ @ @ P∗
@ @ @ @
@ @• @ @• i+1
e=1
@ @ @e= @ e=1
@ @ @ @
@• @• @• @•
K(yi ) K(yi+1 ) (yi = ∞) P̃i+1
= (yi+1 = ∞)
Fig. 7.5.

Thus we have proved that Equation (7.34) defines a tower H = (H0 , H1 , H2 , . . .)

over every constant field K ⊇ IF , and that all steps in the tower are Galois,
of degree [H1 : H0 ] = ( − 1) and [Hi+1 : Hi ] =  for i ≥ 1.

Remark 7.4.23. One of the defining properties of a tower is that it must con-
tain a function field of genus g ≥ 2 (see Definition 7.2.1 and Remark 7.2.2). We
have not verified this property for the tower H yet. One way is to show directly
(using Lemma 7.4.20) that the second basic function field H = K(x, y, z) has
genus g ≥ 2. Another way is to show that the number of rational places N (Hi )
tends to infinity as i → ∞. We will prove this fact in Lemma 7.4.26 below.

Next we consider the ramification locus of the tower H over H0 . The

following lemma is analogous to Lemma 7.4.5.

Lemma 7.4.24. Assume that IF 2 ⊆ K. Then the ramification locus of H

over H0 satisfies

Ram(H/H0 ) ⊆ {(y0 = β) | β ∈ IF 2 ∪ {∞}} .

Before giving the proof of this lemma, we observe that the defining equa-
tion of the tower H can be written in a slightly different way as follows. We
set
f (T ) := (T − T ) −1 + 1 = (T − T )/(T − T ) .
2
(7.41)
Then Equation (7.34) is equivalent to the equation
1
f (Y ) = . (7.42)
1 − f (1/X)

From (7.41) it is obvious that the zeros of the polynomial f (T ) in K̄ are

exactly the elements γ ∈ IF 2 \ IF .
282 7 Asymptotic Bounds for the Number of Rational Places

Proof of Lemma 7.4.24. We want to apply Proposition 7.2.23. We set

Λ0 := {y0 (P ) | P ∈ IPH0 is ramified in H1 /H0 } .

Then Λ0 = IF×2 ∪ {∞} by Lemma 7.4.20, Cases 2 and 3. We claim that the
set
Λ := Λ0 ∪ {0} = IF 2 ∪ {∞}
satisfies condition (2) of Proposition 7.2.23. So we consider β ∈ Λ and
α ∈ K̄ ∪ {∞} where
( −1)
−1 −α 1
(β − β) +1 = −1 −1
= (7.43)
(α − 1) 1 − f (1/α)
(note Equation (7.42)). We must show that α ∈ Λ.
Case 1. β ∈ IF . Then we have
1
1 =
1 − f (1/α)
and hence f (1/α) = 0. It follows that 1/α ∈ IF 2 \ IF and therefore
α ∈ IF 2 \ IF ⊆ Λ.
Case 2. β = ∞. It follows from Equation (7.43) that α = ∞ or α ∈ IF× ,
hence α ∈ Λ.
Case 3. β ∈ IF 2 \ IF . Then (β − β) −1 + 1 = 0 and therefore α = 0 ∈ Λ.
This completes the proof, by Proposition 7.2.23.

Corollary 7.4.25. The genus γ(H/H0 ) of the tower H is finite; it is bounded

by
2 + 2
γ(H/H0 ) ≤ .
2
Proof. Recall that both extensions H2 /K(y0 , y1 ) and H2 /K(y1 , y2 ) are weakly
ramified Galois extensions of degree  by Corollary 7.4.21. For all n ≥ 2 we con-
sider the field Hn as the compositum of the fields Hn−1 and K(yn−2 , yn−1 , yn )
over K(yn−2 , yn−1 ), and then it follows by induction from Proposition 7.4.13
that Hn /H1 is weakly ramified of degree [Hn : H1 ] = n−1 .
Now we estimate the different exponent d(Q|P ) where P ∈ Ram(H/H0 )
and Q is a place of Hn lying above P .
Case 1. P = (y0 = α) with α ∈ IF ∪ {∞}. It follows from Lemma 7.4.20
that P is weakly ramified in H2 /H0 , hence Q|P is weakly ramified in Hn /H0
and we obtain in Case 1, that

d(Q | P ) = 2(e(Q | P ) − 1) ≤ 2e(Q | P ) .

Case 2. P = (y0 = β) with β ∈ IF 2 \ IF . Let P1 := Q ∩ H1 , then

e(P1 |P ) =  − 1 and Q|P1 is weakly ramified by Lemma 7.4.20. It follows that
 7.4 Some Wild Towers 283

d(Q | P ) = e(Q | P1 ) · d(P1 | P ) + d(Q | P1 )

= e(Q | P1 ) · ( − 2) + 2(e(Q | P1 ) − 1)
 
= · e(Q | P ) − 2 ≤ · e(Q | P ) .
−1 −1
There are (l + 1) places in Case 1 and (2 − ) places in Case 2, and we obtain
from Theorem 7.2.10(b) the estimate

1 1  2 + 2
γ(H/H0 ) ≤ −1 + · ( + 1) · 2 + · (2 − ) · = .
2 2 −1 2

Observe that Corollary 7.4.25 holds for every constant field K ⊇ IF . One
does not need the assumption that K ⊇ IF 2 , since the genus γ(H/H0 ) does
not change under a constant field extension.
Eventually we investigate completely splitting places in the tower H. We
show that the splitting behavior of the tower H over the constant field IF 3
is very similar to the splitting behavior of the tower G over IF 2 (see Lemma
7.4.4).

Lemma 7.4.26. Consider the tower H = (H0 , H1 , H2 , . . .), which is defined

by Equation (7.34) over the cubic field K = IF 3 . Then all places (y0 = α)
with α ∈ IF 3 \ IF split completely in H/H0 ; i.e.,

Split(H/H0 ) ⊇ {(y0 = α) | α ∈ IF 3 \ IF } .

Therefore the splitting rate ν(H/H0 ) satisfies

ν(H/H0 ) ≥ 3 −  .

Proof. We will use Corollary 7.2.21. Accordingly we set

−1
f (Y ) := (Y − Y ) + 1, (7.44)
( −1)
−X 1
h(X) := = , (7.45)
(X −1 − 1) −1 1 − f (1/X)
Σ := IF 3 \ IF ,

and we must show that the following condition holds:

Claim: For all α ∈ Σ we have h(α) = ∞, and the equation f (t) = h(α)
has ( − 1) distinct roots t = γ ∈ Σ.
All assertions of Lemma 7.4.26 follow then immediately by Corollary 7.2.21.
In order to prove the claim we introduce the polynomial
3
g(Y ) := (Y − Y )/(Y − Y ) ∈ K[Y ] ,
284 7 Asymptotic Bounds for the Number of Rational Places

whose roots are exactly the elements of Σ. We have the polynomial identity

f (Y ) (f (Y ) − 1) + 1 = g(Y ) , (7.46)

since
2
 2

(Y − Y ) Y −Y
f (Y ) (f (Y ) − 1) + 1 = −1 +1
(Y − Y ) Y −Y
3
Y −Y  2

= Y − Y − (Y − Y ) + 1
(Y − Y ) +1
3 3
Y −Y Y −Y
= +1 = = g(Y ) .
Y −Y Y −Y
Now let α ∈ Σ = IF 3 \ IF , then h(α) = ∞ by (7.45). It also follows that
1/α ∈ Σ and hence g(1/α) = 0. From (7.46) we conclude that
1
f (1/α) = . (7.47)
1 − f (1/α)

Let γ be an element in the algebraic closure K̄ with f (γ) = h(α), so

1
f (γ) = = f (1/α) (7.48)
1 − f (1/α)

by (7.45) and (7.47). Then we obtain

g(γ) = f (γ) (f (γ) − 1) + 1

 
1
= (f (1/α) − 1) + 1 = −1 + 1 = 0 ,
1 − f (1/α)

using (7.46) and (7.48). We conclude that γ ∈ Σ, since the zeros of the
polynomial g(Y ) are exactly the elements of Σ. The polynomal f (t) − h(α) ∈
K[t] has no multiple roots since the derivative f  (t) = (t − t) −2 does not
have a common root with f (t) − h(α) = (t − t) −1 + 1 − h(α). This completes
the proof of the claim and hence of Lemma 7.4.26.

Putting together the results of Corollary 7.4.25 and Lemma 7.4.26, we get
the estimate
ν(H/H0 ) 3 −  2(2 − 1)
λ(H) = ≥ 2 =
γ(H/H0 ) ( + 2)/2 +2

for the limit of the tower H. This completes the proof of Theorem 7.4.17.
 7.5 Exercises 285

7.5 Exercises

7.1. Suppose that F = (F0 , F1 , F2 , . . .) and G = (G0 , G1 , G2 , . . .) are towers

over IFq . Call F and G equivalent if for all i ≥ 0 there exist j, k ≥ 0 such that
Fi ⊆ Ej ⊆ Fk . Show:
(i) This is an equivalence relation.
(ii) Suppose that F and G are equivalent. Then the following hold:
(a) λ(F) = λ(G).
(b) γ(F/F0 ) < ∞ ⇔ γ(G/G0 ) < ∞.
(c) ν(F/F0 ) > 0 ⇔ ν(G/G0 ) > 0.

7.2. Let F = (F0 , F1 , F2 , . . .) be a tame tower over IFq . Assume that its ram-
ification locus Ram(F/F0 ) is finite and that its splitting locus Split(F/F0 ) is
non-empty. Define F ∗ = (F0∗ , F1∗ , F2∗ , . . .) where Fi∗ is the Galois closure of
Fi /F0 . Show that F ∗ = (F0∗ , F1∗ , F2∗ , . . .) is an asymptotically good tower over
IFq .

7.3. Assume that F = (F0 , F1 , F2 , . . .) is an asymptotically good Galois tower

(i.e., all extensions Fi /F0 are Galois). Prove:
(i) The ramification locus Ram(F/F0 ) is finite.
(ii) For some n ≥ 0 there exists a rational place P ∈ IPFn which splits
completely in all extensions Fm /Fn with m ≥ n.

7.4. Let F = (F0 , F1 , F2 , . . .) be a tower of function fields over IFq . For n ≥ 1

we set Dn := deg(Diff(Fn /Fn−1 )).
(i) Assume that there is a real number ! with 0 ≤ ! < 1 and an integer m ≥ 1
such that Dn+1 ≤ ! · [Fn+1 : Fn ] · Dn holds for all n ≥ m. Show that the genus
of F/F0 satisfies γ(F/F0 ) < ∞.
(ii) Now assume that Dm = 0 for some m, and Dn+1 ≥ [Fn+1 : Fn ] · Dn for
all n ≥ m. Show that the genus of F/F0 is γ(F/F0 ) = ∞. In particular, the
tower is asymptotically bad.

7.5. Suppose that the tower F = (F0 , F1 , F2 , . . .) is recursively defined by the

equation f (Y ) = h(X) with f (Y ) ∈ IFq (Y ) and h(X) ∈ IFq (x); i.e.,
(1) F0 = IFq (x0 ) and Fn+1 = Fn (xn+1 ) for all n ≥ 0,
(2) f (xn+1 ) = h(xn ) for all n ≥ 0.
Assume moreover that
(3) both rational functions f (Y ) and h(X) are separable,
(4) The equation f (Y ) = h(xn ) is absolutely irreducible over Fn , for all
n ≥ 0,
(5) deg f (Y ) = deg h(X).
Show that F is asymptotically bad.
286 7 Asymptotic Bounds for the Number of Rational Places

7.6. (char IFq = 2) Suppose that f1 (X), f2 (X) ∈ IFq [X] are polynomials with
deg f1 (X) = 1 + deg f2 (X). Show that the equation

Y 2 = f1 (X)/f2 (X)

defines a recursive tower over IFq . If moreover deg f1 ≥ 3, this tower is asymp-
totically bad.
7.7. (q is a square and char IFq = 2) Consider the tower F = (F0 , F1 , F2 , . . .)
which is recursively defined over IFq by the equation

Y 2 = X(1 − X)/(X + 1) .

(Note that this equation defines a tower over IFq , by the previous exercise.)
Show that the ramification locus of F/F0 is the set of places P ∈ IPF0 with

x0 (P ) ∈ {0, 1, −1, ∞}

or
x0 (P ) ∈ {α ∈ IFq | (α2 + 1)(α2 − 2α − 1)(α2 + 2α − 1) = 0} .
Give an upper bound for γ(F/F0 ).
7.8. Consider the tower of the previous exercise over the field IF81 . Show that
at least 8 rational places of F0 are completely splitting in the tower, and
conclude that the limit of the tower satisfies λ(F) ≥ 2.
7.9. Prove Remark 7.4.9.
7.10. Let c ∈ IF8 be an element with c3 + c + 1 = 0.
(i) Show that the equation

Y 2 + Y = 1/(X 2 + cX)

defines a recursive tower F = (F0 , F1 , F2 , . . .) over IF8 .

(ii) Show that the ramification locus of F/F0 is finite.
(iii) Prove that F/F0 is weakly ramified, and give an upper bound for the
genus γ(F/F0 ).
(iv) Is the tower asymptotically good?
7.11. Let F0 = IF2 (x) be the rational function field over IF2 . Consider for all
n ≥ 1 the field Fn = F0 (y1 , . . . , yn ), where

y12 + y1 = x(x2 + x) ,
y22 + y2 = x3 (x2 + x) ,
...
yn2 + yn = x2n−1 (x2 + x) .
 7.5 Exercises 287

(i) Show that Fn /F0 is Galois of degree [Fn : F0 ] = 2n , and that the pole of
x is totally ramified in Fn /F0 .
(ii) Determine the genus and the number of rational places of Fn /IF2 .
(iii) Conclude that the tower F = (F0 , F1 , F2 , . . .) over IF2 is asymptotically
bad.
(iv) Generalize this example to arbitrary finite fields IFq .

Remark. Part (iii) of the above exercise is in fact a special case of the following
general result: If a tower F = (F0 , F1 , F2 , . . .) over IFq has the property that
all extensions Fn /F0 are Galois with abelian Galois groups, then the genus of
F/F0 is infinite, and hence the tower is asymptotically bad.

7.12. Let F = (F0 , F1 , F2 , . . .) be an asymptotically good tower of function

fields over IFq . Consider a finite extension field E ⊇ F0 having the same
constant field IFq , and define the sequence E := (E0 , E1 , E2 , . . .) by setting
Ej := EFj for all j ≥ 0. Assume that there is at least one rational place of
E/IFq whose restriction to F0 is in the splitting locus of F /F0 . Show that E
is an asymptotically good tower over IFq .

7.13. (q = pa with a ≥ 2) Let F/IFq be a function field having at least

one rational place. Show that there exists an asymptotically good tower F =
(F0 , F1 , F2 , . . .) over IFq with F0 = F .
8
More about Algebraic Geometry Codes

In Chapter 2 we studied algebraic geometry (AG) codes associated with divi-

sors of an algebraic function field over IFq . Here we continue their investigation.
Let us fix some notation for the whole of Chapter 8.

F/IFq is an algebraic function field of genus g and IFq is the full con-
stant field of F .
P1 , . . . , Pn ∈ IPF are pairwise distinct places of degree one.
D = P1 + . . . + Pn .
G is a divisor of F with supp G ∩ supp D = ∅.
CL (D, G) = {(x(P1 ), . . . , x(Pn )) ∈ IFnq | x ∈ L (G)} is the algebraic
geometry code associated with D and G.
CΩ (D, G) = {(ωP1 (1), . . . , ωPn (1)) | ω ∈ ΩF (G − D)} is the dual code
of CL (D, G).

8.1 The Residue Representation of CΩ (D, G)

Let P ∈ IPF be a place of degree one and let ω ∈ ΩF be a Weil differential.
In Chapter 4 we identified ΩF with the differential module ΔF (cf. Remark
4.3.7(a)). Via this identification the local component of ω at the place P can
be evaluated by means of the residue of ω at P , namely ωP (u) = resP (uω) for
all u ∈ F (Theorem 4.3.2(d)). In particular we have ωP (1) = resP (ω). Hence
we have the following alternative description of the code CΩ (D, G).

Proposition 8.1.1.

CΩ (D, G) = {(resP1 (ω), . . . , resPn (ω)) | ω ∈ ΩF (G − D)} .

It is this representation that is most commonly used in the literature to define

the code CΩ (D, G).
H. Stichtenoth, Algebraic Function Fields and Codes, 289
Graduate Texts in Mathematics 254,

c Springer-Verlag Berlin Heidelberg 2009
290 8 More about Algebraic Geometry Codes

By Proposition 2.2.10 the code CΩ (D, G) can also be written as CL (D, H)

where H = D −G+(η) and η is a differential with vPi (η) = −1 and ηPi (1) = 1
for i = 1, . . . , n. Using results from Chapter 4 one can easily construct such a
differential η.

Proposition 8.1.2. Let t be an element of F such that vPi (t) = 1 for i =

1, . . . , n. Then the following hold:
(a) The differential η := dt/t satisfies vPi (η) = −1 and resPi (η) = 1 for
i = 1, . . . , n.
 
(b) CΩ (D, G) = CL D, D − G + (dt) − (t) .

Proof. (a) Since t is a prime element of P := Pi , the P -adic power series of

η = dt/t with respect to t is
1
η= dt .
t
Hence vP (η) = −1 and resP (η) = 1.
(b) Follows immediately from (a) and Proposition 2.2.10.

Corollary 8.1.3. Suppose that t ∈ F is a prime element for all places

P1 , ..., Pn .
(a) If 2G − D ≤ (dt/t) then the code CL (D, G) is self-orthogonal; i.e.,

CL (D, G) ⊆ CL (D, G)⊥ .

(b) If 2G − D = (dt/t) then CL (D, G) is self-dual.

Proof. This is an immediate consequence of Corollary 2.2.11.

8.2 Automorphisms of AG Codes

The symmetric group Sn (whose elements are the permutations of the set
{1, . . . , n}) acts on the vector space IFnq via

π(c1 , . . . , cn ) := (cπ(1) , . . . , cπ(n) )

for π ∈ Sn and c = (c1 , . . . , cn ) ∈ IFnq .

Definition 8.2.1. The automorphism group of a code C ⊆ IFnq is defined by

Aut(C) := {π ∈ Sn | π(C) = C} .
 8.2 Automorphisms of AG Codes 291

Obviously Aut(C) is a subgroup of Sn . Many interesting codes have a non-

trivial automorphism group. In this section we study automorphisms of alge-
braic geometry codes that are induced by automorphisms of the corresponding
function field.
Let F/IFq be a function field and let Aut(F/IFq ) be the group of automor-
phisms of F over IFq (i.e., σ(a) = a for σ ∈ Aut(F/IFq ) and a ∈ IFq ). The
group Aut(F/IFq ) acts on IPF by setting σ(P ) := {σ(x) | x ∈ P }, cf. Lemma
3.5.2. The corresponding valuations vP and vσ(P ) are related as follows:

vσ(P ) (y) = vP (σ −1 (y)) for all y ∈ F . (8.1)

Moreover, deg σ(P ) = deg P since σ induces an isomorphism of the residue

class fields of P and σ(P ) given by σ(z(P )) := σ(z)(σ(P )). The action of
Aut(F/IFq ) on IPF extends to an action on the divisor group by setting
  
σ nP P := nP σ(P ) .

As before we consider divisors D = P1 + . . . + Pn and G of F/IFq where

P1 , . . . , Pn are distinct places of degree one and supp G ∩ supp D = ∅.

Definition 8.2.2. We define

AutD,G (F/IFq ) := {σ ∈ Aut(F/IFq ) | σ(D) = D and σ(G) = G} .

Observe that an automorphism σ ∈ AutD,G (F/IFq ) need not fix the places

P1 , . . . , Pn , but it yields a permutation of P1 , . . . , Pn . From (8.1) it follows
easily that
σ(L (G)) = L (G) (8.2)
for σ ∈ AutD,G (F/IFq ), because σ(G) = G. Now we show that every auto-
morphism σ ∈ AutD,G (F/IFq ) induces an automorphism of the corresponding
code CL (D, G).

Proposition 8.2.3. (a) AutD,G (F/IFq ) acts on the code CL (D, G) by

σ((x(P1 ), . . . , x(Pn ))) := (x(σ(P1 )), . . . , x(σ(Pn )))

(for x ∈ L (G)). This yields a homomorphism from AutD,G (F/IFq ) into

Aut(CL (D, G)).
(b) If n > 2g + 2, the above homomorphism is injective. Hence AutD,G (F/IFq )
can be regarded as a subgroup of Aut(CL (D, G)).

Proof. (a) We begin with the following assertion: given a place P of degree
one and an element y ∈ F with vP (y) ≥ 0, we have

σ(y)(σ(P )) = y(P ) . (8.3)

292 8 More about Algebraic Geometry Codes

In fact, setting a := y(P ) ∈ IFq , we obtain y − a ∈ P . Hence σ(y) − a =

σ(y − a) ∈ σ(P ), and (8.3) follows.
For the proof of (a) we have to show that for every x ∈ L (G) and
σ ∈ AutD,G (F/IFq ) the vector (x(σ(P1 )), . . . , x(σ(Pn ))) is in CL (D, G). As
L (G) = σ(L (G)) by (8.2), we can write x = σ(y) with y ∈ L (G), so

(x(σ(P1 )), . . . , x(σ(Pn ))) = (y(P1 ), . . . , y(Pn )) ∈ CL (D, G) ,

by (8.3).
(b) It is sufficient to prove that the only automorphism of F/IFq fixing
more than 2g + 2 places of degree one is the identity. So we assume that
σ(Q) = Q and σ(Qi ) = Qi for i = 1, . . . , 2g + 2, where σ ∈ Aut(F/IFq )
and Q, Q1 , . . . , Q2g+2 are distinct places of degree one. Choose x, z ∈ F such
that (x)∞ = 2gQ and (z)∞ = (2g + 1)Q (this is possible by the Riemann-
Roch Theorem). Then IFq (x, z) = F since the degrees [F : IFq (x)] = 2g and
[F : IFq (z)] = 2g + 1 are relatively prime. The elements x − σ(x) and z − σ(z)
have at least 2g + 2 zeros (namely Q1 , . . . , Q2g+2 ) but their pole divisor has
degree ≤ 2g + 1 because Q is their only pole. We conclude σ(x) = x and
σ(z) = z, hence σ is the identity.

Example 8.2.4. As an example we consider a BCH code C of length n over IFq .

As shown in Section 2.3, C can be realized as a subfield subcode of a rational
AG code as follows: let n | (q m − 1) and let β ∈ IFqm be a primitive n-th root
of unity. Consider the rational function field F = IFqm (z). For i = 1, . . . , n let
Pi be the zero of z − β i−1 , and set Dβ := P1 + . . . + Pn . Denote by P0 resp.
P∞ the zero resp. the pole of z in F . Then

C = CL (Dβ , rP0 + sP∞ ) |IFq

with r, s ∈ ZZ (see Proposition 2.3.9). The automorphism σ ∈ Aut(F/IFqm )

given by σ(z) = β −1 z leaves the places P0 and P∞ invariant, and we have

σ(Pi ) = Pi+1 (i = 1, . . . , n − 1) and σ(Pn ) = P1 .

Hence, by Proposition 8.2.3, σ induces the following automorphism of the

code CL (Dβ , rP0 + sP∞ ) :

σ(c1 , . . . , cn ) = (c2 , . . . , cn , c1 ) . (8.4)

This means (in the usual terminology of coding theory) that BCH codes are
cyclic codes.

8.3 Hermitian Codes

In Chapter 6 we discussed several examples of algebraic function fields. One
can use all these examples for the explicit construction of algebraic geometry
 8.3 Hermitian Codes 293

codes. In this section we investigate some codes which are constructed by

means of the Hermitian function field. This class of codes provides interesting
and non-trivial examples of AG codes. These codes are codes over IFq2 , they
are not too short compared with the size of the alphabet, and their parameters
k and d are fairly good.
First we recall some properties of the Hermitian function field H (cf.
Lemma 6.4.4). H is a function field over IFq2 ; it can be represented as

H = IFq2 (x, y) with y q + y = xq+1 . (8.5)

The genus of H is g = q(q − 1)/2, and H has N = 1 + q 3 places of degree one,

namely

• the unique common pole Q∞ of x and y, and

• for each pair (α, β) ∈ IFq2 × IFq2 with β q + β = αq+1 there is a unique
place Pα,β ∈ IPH of degree one such that x(Pα,β ) = α and y(Pα,β ) = β.

Observe that for all α ∈ IFq2 there exist q distinct elements β ∈ IFq2 with
β q + β = αq+1 , hence the number of places Pα,β is q 3 .

Definition 8.3.1. For r ∈ ZZ we define the code

Cr := CL (D, rQ∞ ) , (8.6)

where 
D := Pα,β (8.7)
β q +β=αq+1

is the sum of all places of degree one (except Q∞ ) of the Hermitian function
field H/IFq2 . The codes Cr are called Hermitian codes.

Hermitian codes are codes of length n = q 3 over the field IFq2 . For r ≤ s we
obviously have Cr ⊆ Cs . Let us first discuss some trivial cases. For r < 0,
L (rQ∞ ) = 0 and therefore Cr = 0. For r > q 3 + q 2 − q − 2 = q 3 + (2g − 2),
Theorem 2.2.2 and the Riemann-Roch Theorem yield

dim Cr = (rQ∞ ) − (rQ∞ − D)

= (r + 1 − g) − (r − q 3 + 1 − g) = q 3 = n .

Hence Cr = IFnq2 in this case, and it remains to study Hermitian codes with
0 ≤ r ≤ q 3 + q 2 − q − 2.
Proposition 8.3.2. The dual code of Cr is

Cr⊥ = Cq3 +q2 −q−2−r .

Hence Cr is self-orthogonal if 2r ≤ q 3 + q 2 − q − 2, and Cr is self-dual for

r = (q 3 + q 2 − q − 2)/2.
294 8 More about Algebraic Geometry Codes

Proof. Consider the element

 2
t := (x − α) = xq − x .
α∈IFq2

t is a prime element for all places Pα,β ≤ D, and its principal divisor is
2
(t) = D − q 3 Q∞ . Since dt = d(xq − x) = −dx, the differential dt has the
divisor (dt) = (dx) = (q 2 − q − 2)Q∞ (Lemma 6.4.4). Now Theorem 2.2.8 and
Proposition 8.1.2 imply
 
Cr⊥ = CΩ (D, rQ∞ ) = CL D, D − rQ∞ + (dt) − (t)
 
= CL D, (q 3 + q 2 − q − 2 − r)Q∞ = Cq3 +q2 −q−2−r .

Our next aim is to determine the parameters of Cr . We consider the set I

of pole numbers of Q∞ (cf. Definition 1.6.7); i.e.,

I = { n ≥ 0 | there is an element z ∈ H with (z)∞ = nQ∞ } .

For s ≥ 0 let
I(s) := {n ∈ I | n ≤ s} . (8.8)
Then |I(s)| = (sQ∞ ), and the Riemann-Roch Theorem gives

|I(s)| = s + 1 − q(q − 1)/2 for s ≥ 2g − 1 = q(q − 1) − 1 .

From Lemma 6.4.4 we obtain the following description of I(s):

I(s) = {n ≤ s | n = iq + j(q + 1) with i ≥ 0 and 0 ≤ j ≤ q − 1} ,

hence
 
|I(s)| = { (i, j) ∈ IN0 × IN0 ; j ≤ q − 1 and iq + j(q + 1) ≤ s } .

Proposition 8.3.3. Suppose that 0 ≤ r ≤ q 3 + q 2 − q − 2. Then the following

hold:
(a) The dimension of Cr is given by

|I(r)| for 0 ≤ r < q 3 ,
dim Cr =
q − |I(s)| for q 3 ≤ r ≤ q 3 + q 2 − q − 2 ,
3

where s := q 3 + q 2 − q − 2 − r and I(r) is defined by (8.8).

(b) For q 2 − q − 2 < r < q 3 we have

dim Cr = r + 1 − q(q − 1)/2 .

 8.3 Hermitian Codes 295

(c) The minimum distance d of Cr satisfies

d ≥ q3 − r .

If 0 ≤ r < q 3 and both numbers r and q 3 − r are pole numbers of Q∞ , then

d = q3 − r .

Proof. (a) For 0 ≤ r < q 3 Corollary 2.2.3 gives

dim Cr = dim L (rQ∞ ) = |I(r)| .

For q 3 ≤ r ≤ q 3 + q 2 − q − 2 we set s := q 3 + q 2 − q − 2 − r. Then 0 ≤ s ≤

q 2 − q − 2 < q 3 . By Proposition 8.3.2 we obtain

dim Cr = q 3 − dim Cs = q 3 − |I(s)| .

(b) For q 2 − q − 2 = 2g − 2 < r < q 3 , Corollary 2.2.3 gives

dim Cr = r + 1 − g = r + 1 − q(q − 1)/2 .

(c) The inequality d ≥ q 3 − r follows from Theorem 2.2.2. Now let 0 ≤ r < q 3
and assume that both numbers r and q 3 − r are pole numbers of Q∞ . In order
to prove the equality d = q 3 − r we distinguish three cases.
Case 1: r = q 3 − q 2 . Choose i := q 2 − q distinct elements α1 , . . . , αi ∈ IFq2 .
Then the element

i
z := (x − αν ) ∈ L (rQ∞ )
ν=1

has exactly qi = r distinct zeros Pα,β of degree one, and the weight of the
corresponding codeword evD (z) ∈ Cr is q 3 − r. Hence d = q 3 − r.
Case 2: r < q 3 − q 2 . We write r = iq + j(q + 1) with i ≥ 0 and 0 ≤ j ≤
q − 1, so i ≤ q 2 − q − 1. Fix an element 0 = γ ∈ IFq and consider the set
A := {α ∈ IFq2 | αq+1 = γ}. Then |A| = q 2 − (q + 1) ≥ i, and we can choose
distinct elements α1 , . . . , αi ∈ A. The element


i
z1 := (x − αν )
ν=1

has iq distinct zeros Pα,β ≤ D. Next we choose j distinct elements β1 , . . . , βj ∈

IFq2 with βμq + βμ = γ and set


j
z2 := (y − βμ ) .
μ=1
296 8 More about Algebraic Geometry Codes

z2 has j(q + 1) zeros Pα,β ≤ D, and all of them are distinct from the zeros of
z1 because βμq + βμ = γ = ανq+1 for μ = 1, . . . , j and ν = 1, . . . , i. Hence
 
z := z1 z2 ∈ L (iq + j(q + 1))Q∞ = L (rQ∞ )

has r distinct zeros Pα,β ≤ D. The corresponding codeword evD (z) ∈ Cr has
weight q 3 − r.
Case 3: q 3 − q 2 < r < q 3 . By assumption, s := q 3 − r is a pole number and
0 < s < q 2 ≤ q 3 − q 2 . By case 2 there exists an element z ∈ H with principal
divisor (z) = D − sQ∞ where 0 ≤ D ≤ D and deg D = s. The element
2
u := xq − x ∈ H has the divisor (u) = D − q 3 Q∞ , hence

(z −1 u) = (D − D ) − (q 3 − s)Q∞ = (D − D ) − rQ∞ .

The codeword evD (z −1 u) ∈ Cr has weight q 3 − r.

We mention that the minimum distance of Cr is known also in the re-

maining cases (where r ≥ q 3 , or one of the numbers r or q 3 − r is a gap of
Q∞ ).
One can easily specify a generator matrix for the Hermitian codes Cr . We
fix an ordering of the set T := {(α, β) ∈ IFq2 × IFq2 | β q + β = αq+1 }. For
s = iq + j(q + 1) (where i ≥ 0 and 0 ≤ j ≤ q − 1) we define the vector
  3
us := αi β j (α,β)∈T ∈ (IFq2 )q .

Then we have:
Corollary 8.3.4. Suppose that 0 ≤ r < q 3 . Let 0 = s1 < s2 < . . . < sk ≤ r
be all pole numbers ≤ r of Q∞ . Then the k × q 3 matrix Mr whose rows are
us1 , . . . , usk , is a generator matrix of Cr .
Proof. Corollary 2.2.3.

In the same manner we obtain a parity check matrix for Cr (for r > q 2 −q −2),
since the dual of Cr is the code Cs with s = q 3 + q 2 − q − 2 − r.
Finally we study automorphisms of Hermitian codes. Let H = IFq2 (x, y)
as before, cf. (8.5). Let

ε ∈ IFq2 \ {0} , δ ∈ IFq2 and μq + μ = δ q+1 . (8.9)

Then μ ∈ IFq2 , and there exists an automorphism σ ∈ Aut(H/IFq2 ) with

σ(x) = εx + δ and σ(y) = εq+1 y + εδ q x + μ . (8.10)

(The existence of an automorphism σ satisfying (8.10) follows from the fact

that σ(y) and σ(x) satisfy the equation σ(y)q +σ(y) = σ(x)q+1 , which is a con-
sequence of (8.9).) The set of all automorphisms (8.10) of H/IFq2 constitutes
 8.4 The Tsfasman-Vladut-Zink Theorem 297

a group Γ ⊆ Aut(H/IFq2 ) of order q 3 (q 2 − 1) (as ε = 0 and δ are arbitrary,

and for each δ there are q possible values of μ). Clearly σ(Q∞ ) = Q∞ for all
σ ∈ Γ , and σ permutes the places Pα,β of H since they are the only places of
H of degree one other than Q∞ . By Proposition 6.3.3, Γ acts as a group of
automorphisms on the Hermitian codes Cr . We have proved:

Proposition 8.3.5. The automorphism group Aut(Cr ) of the Hermitian code

Cr contains a subgroup of order q 3 (q 2 − 1).

Remark 8.3.6. It is easily seen that Γ acts transitively on the places Pα,β ; i.e.,
given Pα,β and Pα ,β  then there exists some σ ∈ Γ with σ(Pα,β ) = Pα ,β  .

8.4 The Tsfasman-Vladut-Zink Theorem

It is well-known in coding theory that large block lengths (hence large dimen-
sion and large minimum distance) are required to achieve reliable transmission
of information. We introduce some notation that will simplify discussion of
asymptotic performance of codes.

Definition 8.4.1. (a) Given an [n, k, d] code C over IFq , we define its infor-
mation rate
R = R(C) := k/n
and its relative minimum distance

δ = δ(C) := d/n .

(b) Let Vq := {(δ(C), R(C)) ∈ [0, 1]2 | C is a code over IFq } and Uq ⊆ [0, 1]2
be the set of limit points of Vq .

This means: a point (δ, R) ∈ IR2 is in Uq if and only if there are codes C over
IFq of arbitrary large length such that the point (δ(C), R(C)) is arbitrarily
close to (δ, R).

Proposition 8.4.2. There is a continuous function αq : [0, 1] → [0, 1] such

that
Uq = {(δ, R) | 0 ≤ δ ≤ 1 and 0 ≤ R ≤ αq (δ)} .
Moreover the following hold: αq (0) = 1, αq (δ) = 0 for 1 − q −1 ≤ δ ≤ 1, and
αq is decreasing in the interval 0 ≤ δ ≤ 1 − q −1 .

The proof of this proposition requires only elementary techniques of coding

theory; we refer to [29].
For 0 < δ < 1 − q −1 the exact value of αq (δ) is unknown. However, several
upper and lower bounds are available. In the following propositions we state
298 8 More about Algebraic Geometry Codes

some of these bounds. Proofs can be found in most books on coding theory,
e.g. in [28]. The q-ary entropy function Hq : [0, 1 − q −1 ] → IR is defined by
Hq (0) := 0 and

Hq (x) := x logq (q − 1) − x logq (x) − (1 − x) logq (1 − x)

for 0 < x ≤ 1 − q −1 .

Proposition 8.4.3. The following upper bounds for αq (δ) hold:

(a) (Plotkin Bound) For 0 ≤ δ ≤ 1 − q −1 ,
q
αq (δ) ≤ 1 − ·δ.
q−1
(b) (Hamming Bound) For 0 ≤ δ ≤ 1,

αq (δ) ≤ 1 − Hq (δ/2) .

(c) (Bassalygo-Elias Bound) For 0 ≤ δ ≤ θ := 1 − q −1 ,

 " 
αq (δ) ≤ 1 − Hq θ − θ(θ − δ) .

Out of the upper bounds in Proposition 8.4.3, the Bassalygo-Elias Bound

is always the best, see Figure 8.1 below; an even better upper bound (which
is more complicated to state and more difficult to prove) is the McEliece-
Rodemich-Rumsey-Welch Bound, see [28],[32].
Perhaps more important than upper bounds are lower bounds for αq (δ), be-
cause every non-trivial lower bound for αq (δ) guarantees the existence of ar-
bitrary long codes with good parameters (δ(C), R(C)).

Proposition 8.4.4 (Gilbert-Varshamov Bound). For 0 ≤ δ ≤ 1 − q −1 ,

αq (δ) ≥ 1 − Hq (δ) .

The Gilbert-Varshamov bound is the best lower bound for αq (δ) which is
known from elementary coding theory. However, its proof is not constructive
(i.e., it does not provide a simple algebraic algorithm for the construction of
good long codes).
Our aim is to construct algebraic geometry codes of large length in order
to improve the Gilbert-Varshamov Bound. Given an algebraic function field
F/IFq with N = N (F ) places of degree one, the length of any AG code
CL (D, G) (resp. CΩ (D, G)) associated with divisors D and G of F is bounded
by N , since D is a sum of places of degree one. In fact, this is the only
restriction on the length of an AG code which can be constructed by means
of the function field F .
 8.4 The Tsfasman-Vladut-Zink Theorem 299
R
6
1
A
A
A
A
A
A Gilbert-Varshamov
A
A
A Bassalygo-Elias
A
A
A
A Plotkin
A
A
A Hamming
A
A
A
A
A
A
A - δ
0 0.5 1

Fig. 8.1. Bounds for q = 2.

Lemma 8.4.5. Suppose that P1 , . . . , Pn are distinct places of F/IFq of degree

one. Then there exists, for each r ≥ 0, a divisor G such that deg G = r and
Pi ∈ supp G (for i = 1, . . . , n).

Proof. The lemma is trivial if there is another place Q of degree one, different
from P1 , . . . , Pn . In this case we set G := rQ. If P1 , . . . , Pn are all the places of
F/IFq of degree one, we choose a divisor G ∼ rP1 (i.e., G is equivalent to rP1 )
such that vPi (G) = 0 for i = 1, . . . , n. This is possible by the Approximation
Theorem.

According to Lemma 8.4.5 one needs function fields over IFq having many
rational places in order to construct long AG codes. We recall the definition
of Ihara’s constant A(q) given in Chapter 7. For g ≥ 0 let

Nq (g) := max{N (F ) | F is a function field over IFq of genus g } ,

where N (F ) denotes the number of places of F/IFq of degree one. Then A(q)
is defined as
Nq (g)
A(q) = lim sup .
g→∞ g
300 8 More about Algebraic Geometry Codes

Proposition 8.4.6. Suppose that A(q) > 1. Then

αq (δ) ≥ (1 − A(q)−1 ) − δ

in the interval 0 ≤ δ ≤ 1 − A(q)−1 .

Proof. Let δ ∈ [0, 1 − A(q)−1 ]. Choose a sequence of function fields Fi /IFq of

genus gi such that
gi → ∞ and ni /gi → A(q) , (8.11)
where ni := N (Fi ). Choose ri > 0 such that

ri /ni → 1 − δ . (8.12)

This is possible as ni → ∞ for i → ∞. Let Di be the sum of all places of

Fi /IFq of degree one, thus deg Di = ni . By Lemma 8.4.5 there exists a divisor
Gi of Fi /IFq such that deg Gi = ri and supp Gi ∩ supp Di = ∅. Consider the
code Ci := CL (Di , Gi ); this is an [ni , ki , di ] code whose parameters ki and di
satisfy the inequalities

ki ≥ deg Gi + 1 − gi = ri + 1 − gi and di ≥ ni − deg Gi = ni − ri

(cf. Corollary 2.2.3). Hence

ri + 1 gi ri
Ri := R(Ci ) ≥ − and δi := δ(Ci ) ≥ 1 − . (8.13)
ni ni ni
W.l.o.g. we can assume that the sequences (Ri )i≥1 and (δi )i≥1 are convergent
(otherwise we choose an appropriate subsequence), say Ri → R and δi → δ̃.
From (8.11), (8.12) and (8.13) it follows that R ≥ 1 − δ − A(q)−1 and δ̃ ≥ δ.
So αq (δ̃) ≥ R ≥ 1 − δ − A(q)−1 . Since αq is non-increasing, this implies

αq (δ) ≥ αq (δ̃) ≥ 1 − δ − A(q)−1 .

Now we can easily prove the main result of this section.

Theorem 8.4.7 (Tsfasman-Vladut-Zink Bound). Let q = 2 be a

square. Then we have for all δ with 0 ≤ δ ≤ 1 − (q 1/2 − 1)−1 ,
 
1
αq (δ) ≥ 1 − 1/2 −δ .
q −1

Proof. By Corollary 7.4.8 we have A(q) = q 1/2 − 1 if q is a square. Now the

assertion follows immediately from Proposition 8.4.6.

For all q ≥ 49 the Tsfasman-Vladut-Zink Bound improves the Gilbert-

Varshamov Bound in a certain interval, see Figure 8.2 .
 8.4 The Tsfasman-Vladut-Zink Theorem 301
R
6
1

Gilbert-Varshamov
@
@
@
@
@
@
@ Tsfasman-Vladut-Zink
@
@
@
@
@
@
@
@
@
@
@
@
@ - δ
0 1− 1
7 1

Fig. 8.2. Bounds for q = 64.

Remark 8.4.8. Also if q is not a square one can obtain an improvement of the
Gilbert-Varshamov Bound if a good lower bound for Ihara’s constant A(q)
is available. For instance, let q = 3 be a cube. Then we have for all δ with
0 ≤ δ ≤ 1 − ( + 2)/(2(2 − 1)) the following lower bound for αq (δ):
 
+2
αq (δ) ≥ 1 − −δ . (8.14)
2(2 − 1)
The proof of this bound is exactly the same as in Theorem 8.4.7; one just uses
the bound for A(3 ) given in Corollary 7.4.18. We note that (8.14) improves
the Gilbert-Varshamov Bound for all cubes q ≥ 73 .

In the proof of the Tsfasman-Vladut-Zink Theorem we have only used

that there exists a sequence of function fields Fi /IFq (for q = 2 ) with
limn→∞ N (Fi )/g(Fi ) =  − 1. If the function fields Fi have additional nice
properties, one can hope that the corresponding AG codes also have nice
properties. As an example for this idea we shall prove the existence of long
self-dual codes whose parameters attain the Tsfasman-Vladut-Zink Bound.
302 8 More about Algebraic Geometry Codes

Theorem 8.4.9. Let q = 2 be a square. Then there exists a sequence of

self-dual codes (Ci )i≥0 over IFq with parameters [ni , ki , di ] such that ni → ∞
and
di 1 1
lim inf ≥ − . (8.15)
i→∞ ni 2 −1
Note that Inequality (8.6) just says that the sequence (Ci )i≥0 attains the
Tsfasman-Vladut-Zink Bound, because the information rate of a self-dual code
C is R(C) = 1/2. As a consequence we obtain that there are self-dual codes
over IFq (with q = 2 ≥ 49) of arbitrary large length whose performance is
better than the Gilbert-Varshamov bound.
Proof of Theorem 8.4.9. For simplicity we will assume that q is even; i.e.,
char IFq = 2 (the assertion is also true in the case of odd characteristic, but
the proof is then a bit more complicated). We will use the Galois tower G ∗ =
(G∗0 , G∗1 , G∗2 , ...) of function fields over IFq that was studied in Section 7.4, see
Theorem 7.4.15 and Corollary 7.4.16. We recall briefly the properties of this
tower that will be needed below.
The field G∗0 = IFq (u0 ) is a rational function field. For i ≥ 1 we have

ni = [G∗i : G∗0 ] = ( − 1)mi ,

where mi ≥  is a power of p = char IFq . The zero divisor of u0 in G∗i has the
form
G∗

ni
(i)
(u0 )0 i = Di = Pj (8.16)
j=1

(i)
with pairwise distinct places Pj of degree one, and the divisor of the differ-
ential η (i) = du0 /u0 in the function field G∗i is given by
(0) (∞)
(η (i) ) = (ei − 2)Ai + (ei − 2)Bi − Di (8.17)

with positive divisors Ai , Bi and

 
supp Ai ∪ supp Bi ∩ supp Di = ∅ .

Moreover the degrees of the divisors Ai , Bi satisfy

(0) (∞)
ei · deg Ai = ei · deg Bi = ni /( − 1) . (8.18)

. Now we define the divisor Hi ∈ Div(G∗i ) as

(0) (∞)
with certain integers ei , ei

 e(0)
i −2
  e(∞) − 2 
Hi := Ai + i Bi .
2 2
At this point we have used the assumption that q (and hence ) is even. Since
2Hi − Di = (η (i) ) by (8.17), it follows from Corollary 8.1.3 that the code
 8.5 Decoding AG Codes 303

Ci := CL (Di , Hi ) ⊆ IFnq i

is self-dual. By Theorem 2.2.2 its minimum distance di := d(Ci ) can be esti-

mated by

di ≥ deg(Di − Hi ) = deg Di − deg Hi

 e(0) − 2   e(∞) − 2 
= ni − i
deg Ai − i deg Bi
2 2
1  (0) (∞)

≥ ni − ei deg Ai + ei deg Bi
2
1  ni ni 
= ni −  +
2 −1 −1
1 1 
= ni −
2 −1
(here we have used Equations (8.16) and (8.18)). Therefore we obtain

di 1 1
δi := δ(Ci ) = ≥ − .
ni 2 −1

8.5 Decoding AG Codes

For a code to have practical use, it is essential that one has an effective
decoding algorithm. Let us briefly explain what this means. We consider an
[n, k, d] code C ⊆ IFnq . Then C is t-error correcting for all t ≤ (d − 1)/2, cf.
Section 2.1. Suppose a ∈ IFnq is an n-tuple such that

a = c + e, (8.19)

where c ∈ C is a codeword and e ∈ IFnq has weight

wt(e) ≤ (d − 1)/2 . (8.20)

Then c is uniquely determined by a and the conditions (8.19) and (8.20);

it is the unique codeword whose distance to a is minimal. The vector e in
(8.19) is called the error vector of a with respect to C. A decoding algorithm
is an algorithm which calculates for every element a ∈ IFnq satisfying (8.19)
and (8.20) the corresponding codeword c (or, equivalently, the corresponding
error vector e).
For algebraic geometry codes a very general decoding algorithm is avail-
able. We consider the code

CΩ := CΩ (D, G) (8.21)
304 8 More about Algebraic Geometry Codes

with D = P1 + . . . + Pn and supp D ∩ supp G = ∅, as always in this chapter.

For b = (b1 , . . . , bn ) ∈ IFnq and f ∈ L (G) we define the syndrome


n
[b, f ] := bν · f (Pν ) . (8.22)
ν=1

The symbol [ , ] is obviously bilinear. As CΩ is the dual of CL (D, G) (Theorem

2.2.8), we have

CΩ = {b ∈ IFnq | [b, f ] = 0 for all f ∈ L (G) } . (8.23)

Let t ≥ 0 be an integer and let G1 be a divisor of F/IFq satisfying the following

conditions:

supp G1 ∩ supp D = ∅ ,
deg G1 < deg G − (2g − 2) − t , (8.24)
(G1 ) > t .

We will show that under these assumptions all errors of weight ≤ t can be
corrected by means of a simple algorithm.

Remark 8.5.1. Let d∗ = deg G − (2g − 2) be the designed distance of CΩ , cf.

Section 2.2. We know that d∗ ≤ d(CΩ ) where d(CΩ ) denotes the minimum
distance of CΩ (Theorem 2.2.7). The following assertions hold:
(a) If G1 and t satisfy (8.24) then t ≤ (d∗ − 1)/2.
(b) If 0 ≤ t ≤ (d∗ − 1 − g)/2 then there exists a divisor G1 such that (8.24) is
satisfied.

Proof. (a) By (8.24) and Inequality (1.21) of Chapter 1 we have t ≤ (G1 ) −

1 ≤ deg G1 and t ≤ deg G − deg G1 − 2g + 1. Adding these inequalities yields
2t ≤ deg G + 1 − 2g = d∗ − 1, hence t ≤ (d∗ − 1)/2.
(b) Now we assume that t ≤ (d∗ − 1 − g)/2. We choose a divisor G1 such that

deg G1 = g + t and supp G1 ∩ supp D = ∅ . (8.25)

This is possible by the Approximation Theorem. By the Riemann-Roch

Theorem,
(G1 ) ≥ deg G1 + 1 − g = t + 1 > t .
The assumption t ≤ (d∗ − 1 − g)/2 is equivalent to d∗ − 2t − g ≥ 1. So

deg G − (2g − 2) − t − deg G1 = d∗ − 2t − g > 0 .

This proves (8.24).

 8.5 Decoding AG Codes 305

From now on we assume (8.24). Suppose that

a = c + e with c ∈ CΩ and wt(e) ≤ t . (8.26)

Let
e = (e1 , . . . , en ) and I := { ν | 1 ≤ ν ≤ n and eν = 0 } (8.27)
be the set of error positions (so |I| = wt(e) ≤ t). In the first step of the de-
coding algorithm we shall construct an error locator function; i.e., an element
0 = f ∈ L (G1 ) with the property f (Pν ) = 0 for all ν ∈ I. That means that
the error positions are contained in the set

N (f ) := { ν | 1 ≤ ν ≤ n and f (Pν ) = 0 } . (8.28)

In the second step we determine the error values eν for all ν ∈ N (f ). As

eν = 0 for ν ∈ N (f ), this yields the error vector e. We shall see that each of
these steps requires the solution of a certain system of linear equations. We
specify bases

{f1 , . . . , fl } of L (G1 ) ,
{g1 , . . . , gk } of L (G − G1 ) ,
{h1 , . . . , hm } of L (G) . (8.29)

Note that the choice of these bases does not depend on the vector a which is
to be decoded. It is obvious that fλ gρ ∈ L (G) for 1 ≤ λ ≤ l and 1 ≤ ρ ≤ k.
Consider the following system of linear equations which plays an essential
role in the decoding algorithm.

l
[a, fλ gρ ] · xλ = 0 , for ρ = 1, . . . , k . (8.30)
λ=1

Proposition 8.5.2. With the above notations and assumptions (in particular
(8.24) and (8.26) - (8.29)), the system (8.30) has a non-trivial solution. If
(α1 , . . . , αl ) is a non-trivial solution of (8.30) we set


l
f := αλ fλ ∈ L (G1 ) . (8.31)
λ=1

Then f (Pν ) = 0 for all error positions ν ∈ I; i.e., f is an error locator

function.
Proof. I ⊆ {1, . . . , n} is the set of error positions,see (8.27). As |I| ≤ t by
(8.26) and (G1 ) > t by (8.24), we have (G1 − ν∈I Pν ) > 0, cf. Lemma
1.4.8. Choose 0 = z ∈ L (G1 − ν∈I Pν ) and write


l
z= γλ fλ with γλ ∈ IFq .
λ=1
306 8 More about Algebraic Geometry Codes

Then zgρ ∈ L (G) for 1 ≤ ρ ≤ k, and we obtain


l
[a, zgρ ] = [a, fλ gρ ] · γλ . (8.32)
λ=1

On the other hand, since c ∈ CΩ and zgρ ∈ L (G), we have [c, zgρ ] = 0 by
(8.23). Hence


n
[a, zgρ ] = [c + e, zgρ ] = [e, zgρ ] = eν · z(Pν ) · gρ (Pν ) = 0 . (8.33)
ν=1

(Observe that  eν = 0 for ν ∈ I and z(Pν ) = 0 for ν ∈ I, because z is an element

of L (G1 − ν∈I Pν ).) Equations (8.32) and (8.33) show that (γ1 , . . . , γl ) is a
non-trivial solution of (8.30).
Now we take an arbitrary solution (α1 , . . . , αl ) of the system (8.30) and
l
set f := λ=1 αλ fλ . Suppose there is an error position ν0 ∈ I such that
f (Pν0 ) = 0. By (8.24),
  
deg G − G1 − Pν ≥ deg G − deg G1 − t > 2g − 2 .
ν∈I

This implies
     
L G − G1 − Pν  L G − G1 − Pν .
ν∈I ν∈I\{ν0 }

So we find an element h ∈ L (G − G1 ) with h(Pν0 ) = 0 and h(Pν ) = 0 for all

ν ∈ I \ {ν0 }. We obtain


n
[a, f h] = [e, f h] = eν · f (Pν ) · h(Pν ) = eν0 · f (Pν0 ) · h(Pν0 ) = 0 . (8.34)
ν=1

However, h is a linear combination of g1 , . . . , gk and


l
[a, f gρ ] = [a, fλ gρ ] · αλ = 0 ,
λ=1

since (α1 , . . . , αl ) is a solution of (8.30). This contradicts (8.34).

The cardinality of N (f ) = { ν| 1 ≤ ν

≤ n and f (P
 ν ) = 0} does not exceed
the degree of G1 , because f ∈ L G1 − ν∈N (f ) Pν implies

deg G1 − |N (f )| ≥ 0 . (8.35)

Observe that in general not all ν ∈ N (f ) are actually error positions.

 8.5 Decoding AG Codes 307

In order to determine the error values eν , we consider another system of

linear equations:

hμ (Pν ) · zν = [a, hμ ] , for μ = 1, . . . , m . (8.36)
ν∈N (f )

(Recall that {h1 , . . . , hm } is a basis of L (G), see (8.29).)

Proposition 8.5.3. Under the above hypotheses the system (8.36) has a
unique solution, namely the vector (eν )ν∈N (f ) .
Proof. As hμ ∈ L (G), we have

n 
[a, hμ ] = [c + e, hμ ] = [e, hμ ] = eν · hμ (Pν ) = hμ (Pν ) · eν .
ν=1 ν∈N (f )

(Note that eν = 0 for ν ∈ N (f ) by Proposition 8.5.2.) Hence (eν )ν∈N (f ) is a

solution of the system (8.36).
Suppose that (bν )ν∈N (f ) is another solution of (8.36). Define the vector
b := (b1 , . . . , bn ) ∈ IFnq by setting bν := 0 for ν ∈ N (f ). Then

[b, hμ ] = hμ (Pν ) · bν = [a, hμ ] = [e, hμ ]
ν∈N (f )

for μ = 1, . . . , m. As {h1 , . . . , hm } is a basis of L (G), this implies b − e ∈ CΩ ,

by (8.23). The weight of b − e can be estimated as follows:

wt(b − e) ≤ |N (f )| ≤ deg G1 < deg G − (2g − 2) = d∗ .

(We have used (8.35) and (8.24) once again.) Since the minimum distance of
CΩ is ≥ d∗ we conclude that b = e.

Proposition 8.5.2 and 8.5.3 are summarized in the following decoding al-
gorithm for the code CΩ . We maintain all previous notation.
Decoding Algorithm 8.5.4. Let an element a ∈ IFnq be given.
(1) Find a non-trivial solution (α1 , . . . , αl ) of the system (8.30), and set f :=
l
λ=1 αλ fλ . (If (8.30) has only the trivial solution, we cannot decode a.)
(2) Determine N (f ) = { ν | 1 ≤ ν ≤ n and f (Pν ) = 0}. (This can be done by
l
evaluating f (Pν ) = λ=1 αλ fλ (Pν ) for ν = 1, . . . , n.)
(3) If the system (8.36) has a unique solution (eν )ν∈N (f ) , we set e :=
(e1 , . . . , en ) with eν = 0 for ν ∈ N (f ). (If (8.36) is not uniquely solvable,
we cannot decode a.)
(4) Check whether c := a−e is an element of CΩ (by calculating the syndromes
[c, hμ ] for μ = 1, . . . , m) and whether wt(e) ≤ t. If the answer is yes, we decode
a to the codeword c. If the answer is no, we cannot decode a.
308 8 More about Algebraic Geometry Codes

Theorem 8.5.5 (Skorobogatov-Vladut). (a) Provided G1 and t satisfy

(8.24), the algorithm 8.5.4 decodes all errors of weight ≤ t.
(b) One can choose the divisor G1 in such a way that the algorithm 8.5.4
decodes all errors e of weight

wt(e) ≤ (d∗ − 1 − g)/2 ,

where d∗ = deg G − (2g − 2) is the designed distance of CΩ .

Proof. (a) is obvious from Proposition 8.5.2 and 8.5.3, and (b) follows from
Remark 8.5.1(b).

Remark 8.5.6. (a) It can happen that steps (1) - (3) of the decoding algorithm
work but that a − e ∈ CΩ or wt(e) > t. In either case there is no codeword
c ∈ CΩ with the property wt(c) ≤ t.
(b) The decoding algorithm 8.5.4 is due to A.N.Skorobogatov and S.G.Vladut,
following an idea of J.Justesen et al. One deficiency of the algorithm is that it
decodes (in general) only errors of weight ≤ (d∗ − 1 − g)/2 but not all errors of
weight ≤ (d∗ −1)/2. There are several variants of the algorithm which correct,
under additional hypotheses, more than (d∗ − 1 − g)/2 errors.
(c) There are some completely different approaches to decode AG codes: we
mention the method due to G. L. Feng and T. R. N. Rao (see Chapter 10
in [32]), and the decoding algorithm of M. Sudan (see [17]); this algorithm is
known as list decoding.

8.6 Exercises
8.1. Let F be a function field over a finite field IFq . We consider divisors G
and D = P1 + . . . + Pn , where P1 , . . . , Pn are distinct places of degree one and
supp G ∩ supp D = ∅ as usual. Show that CL (D, G) = CL (σ(D), σ(G)) and
CΩ (D, G) = CΩ (σ(D), σ(G)), for every automorphism σ of F/IFq .

8.2. (i) Show that the Tsfasman-Vladut-Zink Bound improves the Gilbert-
Varshamov Bound for all q = 2 ≥ 49, but not for q ≤ 25.
(ii) Let q = 3 be a cube. Show that the bound
 
+2
αq (δ) ≥ 1 − −δ
2(2 − 1)

improves the Gilbert-Varshamov bound for all q = 3 ≥ 343 (see Remark

8.4.8).

8.3. (This exercise will be useful for the following exercises.) Consider the
tower G ∗ = (G∗0 , G∗1 , G∗2 , . . .) over the field IFq with q = 2 as in Theorem
7.4.15. It has amongst others the following properties:
 8.6 Exercises 309

(1) G∗0 = IFq (u0 ) is rational,

(2) all extensions G∗i /G∗0 are Galois,
(3) the only places of IFq (u0 ) which are ramified in some extension G∗n /G∗0 ,
are the places (u0 = 1) and (u0 = ∞),
(4) the place (u0 = 0) splits completely in all extensions G∗n /G∗0 .
Show that the ramification indices of the places (u0 = 1) and (u0 = ∞) in
G∗n /G∗0 tend to infinity, as n → ∞.
8.4. A code C ⊆ IFnq is said to be transitive, if its automorphism group Aut (C)
is a transitive subgroup of the symmetric group Sn (i.e., for any two indices
i, j ∈ {1, 2, . . . , n} there is an automorphism π ∈ Aut (C) such that π(i) = j).
This is obviously a generalization of the notion of cyclic codes.
Show that the class of transitive codes over IFq , where q = 2 is a square,
attains the Tsfasman-Vladut-Zink Bound. More precisely, let R, δ ≥ 0 be real
numbers with R = 1 − δ − 1/( − 1). Show that there exists a sequence (Cj )j≥0
of linear codes over IFq with parameters [nj , kj , dj ] having the following prop-
erties:
(1) all Cj are transitive,
(2) nj → ∞ as j → ∞,
(3) limj→∞ kj /nj ≥ R and limj→∞ dj /nj ≥ δ.
8.5. Assume that q = 2 is a square. Let 0 ≤ R ≤ 1/2 and δ ≥ 0 with
R = 1 − δ − 1/( − 1). Show that there is a sequence (Cj )j≥0 of linear codes
over IFq with parameters [nj , kj , dj ] having the following properties:
(1) all Cj are self-orthogonal; i.e., Cj ⊆ Cj⊥ ,
(2) nj → ∞ as j → ∞,
(3) limj→∞ kj /nj ≥ R and limj→∞ dj /nj ≥ δ,
(4) also the dual codes (Cj⊥ )j≥0 attain the Tsfasman-Vladut-Zink
Bound.
8.6. Let q = 2 be a square. Show that there exists a sequence of self-dual
codes (Ci )i≥0 over IFq with parameters [ni , ki , di ] such that ni → ∞ and
di 1 1
lim inf ≥ − .
i→∞ ni 2 −1
Hint. This exercise is just the assertion of Theorem 8.4.9 which we proved
only in the case where q is even. So you can assume now that q is odd. Again
use the tower (G∗0 , G∗1 , G∗2 , . . .) as in Theorem 7.4.15. Consider the elements u0
and t0 as defined in Equation (7.26); they satisfy the equation u0 = t0−1 + 1.
Choose ! ∈ IFq with ! −1 = −1 and consider the differential ω := ! · dt0 /u0
in G∗n . Show that its divisor has the form (ω) = 2H − D where H ≥ 0 and
D is the sum of rational places of G∗n . Calculate the residues of ω at these
rational places and show that they are squares in IF× q . Conclude that the code
CL (D, H) is equivalent to a self-dual code (see Definition 2.2.13).
9
Subfield Subcodes and Trace Codes

A very useful method of constructing codes over IFq is to restrict codes which
are defined over an extension field IFqm . This means that, given a code C ⊆
(IFqm )n , one considers the subfield subcode C|IFq = C ∩ IFnq . Many well-known
codes can be defined in this way, for instance BCH codes, Goppa codes and,
more generally, alternant codes (cf. Section 2.3).
There is yet another method of defining a code over IFq if a code over
IFqm is given. This construction uses the trace mapping Tr : IFqm → IFq .
An important class of codes which can be represented as trace codes in a
natural manner is the class of cyclic codes (see Example 9.2.4 below). The
subfield subcode construction and the trace construction are closely related
by Delsarte’s Theorem 9.1.2.
In this chapter we present some results on subfield subcodes and trace
codes. It is a surprising fact that the study of trace codes leads to a second
non-trivial relation between coding theory and the theory of algebraic function
fields, see Section 9.2.

9.1 On the Dimension of Subfield Subcodes and Trace

Codes

We consider the field extension IFqm /IFq ; this is a Galois extension of degree
[IFqm : IFq ] = m. Let
Tr : IFqm −→ IFq
denote the trace mapping (cf. Appendix A). For a = (a1 , . . . , an ) ∈ (IFqm )n
we define  
Tr(a) := Tr(a1 ), . . . , Tr(an ) ∈ IFnq .
In this manner we obtain an IFq -linear map Tr : (IFqm )n → IFnq .

H. Stichtenoth, Algebraic Function Fields and Codes, 311

Graduate Texts in Mathematics 254,

c Springer-Verlag Berlin Heidelberg 2009
312 9 Subfield Subcodes and Trace Codes

Definition 9.1.1. Let C ⊆ (IFqm )n be a code over IFqm .

(a) C|IFq := C ∩ IFnq is called the subfield subcode (or the restriction of C to
IFq ).
(b) Tr(C) := {Tr(c) | c ∈ C} ⊆ IFnq is called the trace code of C.
Note that the subfield subcode and the trace code of a code C ⊆ (IFqm )n are
codes over IFq of length n.

Theorem 9.1.2 (Delsarte). For a code C over IFqm ,

(C|IFq )⊥ = Tr(C ⊥ ) .

Proof. Recall that we denote by  ,  the canonical inner product on IFnq (resp.
on (IFqm )n ). In order to prove (C|IFq )⊥ ⊇ Tr(C ⊥ ) we have to show that

c, Tr(a) = 0 for all c ∈ C|IFq and a ∈ C ⊥ . (9.1)

Write c = (c1 , . . . , cn ) and a = (a1 , . . . , an ); then


n
 
c, Tr(a) = ci · Tr(ai ) = Tr ci ai = Tr(a, c) = Tr(0) = 0 .
i=1

We have used here the IFq -linearity of the trace and the fact that a, c = 0
(which follows from c ∈ C and a ∈ C ⊥ ). This proves (9.1).
Next we show that (C|IFq )⊥ ⊆ Tr(C ⊥ ). This assertion is equivalent to

Tr(C ⊥ )⊥ ⊆ C|IFq . (9.2)

Suppose that (9.2) does not hold. Then there exists some u ∈ Tr(C ⊥ )⊥ \ C,
hence an element v ∈ C ⊥ with u, v = 0. As Tr : IFqm → IFq is not the zero
map, there is an element γ ∈ IFqm such that Tr(γ · u, v) = 0. Hence

u, Tr(γv) = Tr(u, γv) = Tr(γ · u, v) = 0 .

But on the other hand we have u, Tr(γv) = 0 because u ∈ Tr(C ⊥ )⊥ and
γv ∈ C ⊥ . This contradiction proves (9.2).

There are obvious upper bounds for the dimension of subfield subcodes
and trace codes, namely

dim C|IFq ≤ dim C (9.3)

and
dim Tr(C) ≤ m · dim C . (9.4)
(9.3) follows from the fact that a basis of C|IFq over IFq is also linearly in-
dependent over IFqm , and (9.4) follows since Tr : C → Tr(C) is a surjective
 9.1 On the Dimension of Subfield Subcodes and Trace Codes 313

IFq -linear mapping and the dimension of C, regarded as a vector space over
the field IFq , is m · dim C.
Using Delsarte’s Theorem we also obtain lower bounds for the dimensions
of subfield subcodes and trace codes:
Lemma 9.1.3. Let C be a code of length n over IFqm . Then
dim C ≤ dim Tr(C) ≤ m · dim C (9.5)
and
dim C − (m − 1)(n − dim C) ≤ dim C|IFq ≤ dim C . (9.6)
Proof. By Delsarte’s Theorem and (9.3) we have
dim Tr(C) = dim(C ⊥ |IFq )⊥ = n − dim C ⊥ |IFq ≥ n − dim C ⊥ = dim C .
This proves (9.5). The lower estimate in (9.6) is proved in an analogous
manner.

The bounds given in Lemma 9.1.3 are general bounds, valid for each code
over IFqm . Our aim is to improve these estimates in specific cases, in particular
for certain algebraic geometry codes.
In the following, a subcode of a code C ⊆ (IFqm )n means an IFqm -subspace
U ⊆ C. By U q we denote the set
U q := {(aq1 , . . . , aqn ) | (a1 , . . . , an ) ∈ U } .
It is obvious that U q is also an IFqm -subspace of (IFqm )n .
Proposition 9.1.4. Let C be a code over IFqm and let U ⊆ C be a subcode
with the additional property U q ⊆ C. Then
dim Tr(C) ≤ m · (dim C − dim U ) + dim U |IFq .
Proof. We consider the IFq -linear map φ : U → C, given by φ(u) := uq − u.
The kernel of φ is easily seen to be
Ker(φ) = U |IFq . (9.7)
Since Tr(aq ) = Tr(a) for a ∈ IFqm , the image of φ is contained in the kernel
of the trace map Tr : C → Tr(C); i.e.,
Im(φ) ⊆ Ker(Tr) . (9.8)
From (9.7) and (9.8) we obtain
dim Tr(C) = dimIFq C − dim Ker(Tr) ≤ m · dim C − dim Im(φ)
= m · dim C − (dimIFq U − dim Ker(φ))
= m · (dim C − dim U ) + dim U |IFq .
314 9 Subfield Subcodes and Trace Codes

Note that Proposition 9.1.4 improves the upper bound (9.5) for the di-
mension of trace codes. We also obtain an improvement of the lower bound
(9.6) for the dimension of subfield subcodes:

Corollary 9.1.5. Let C be a code of length n over IFqm and let V ⊆ C ⊥ be a

subcode of C ⊥ satisfying V q ⊆ C ⊥ . Then

dim C|IFq ≥ dim C − (m − 1)(n − dim C) + m · dim V − dim V |IFq

≥ dim C − (m − 1)(n − dim C − dim V ) .

Proof. We use Proposition 9.1.4 and Delsarte’s Theorem:

dim C|IFq = dim Tr(C ⊥ )⊥ = n − dim Tr(C ⊥ )

 
≥ n − m · (dim C ⊥ − dim V ) + dim V |IFq
= dim C − (m − 1)(n − dim C) + m · dim V − dim V |IFq .

Now we apply the above results to AG codes.

Theorem 9.1.6. Let F be an algebraic function field of genus g over the

constant field IFqm . Consider the AG codes

CL := CL (D, G) and CΩ := CΩ (D, G) , (9.9)

where D = P1 + . . . + Pn (with pairwise distinct places P1 , . . . , Pn of degree

one), supp D ∩ supp G = ∅ and deg G < n. Suppose that G1 is a divisor of F
satisfying

G1 ≤ G and q · G1 ≤ G . (9.10)
Then 
m((G) − (G1 )) + 1 if G1 ≥ 0 ,
dim Tr(CL ) ≤ (9.11)
m((G) − (G1 )) if G1 ≥
 0,
and

n − 1 − m((G) − (G1 )) if G1 ≥ 0 ,
dim CΩ |IFq ≥ (9.12)
n − m((G) − (G1 )) if G1 ≥
 0.

Proof. Let U := CL (D, G1 ). It follows from (9.10) that U q ⊆ CL . We can

apply Proposition 9.1.4 and obtain

dim Tr(CL ) ≤ m((G) − (G1 )) + dim U |IFq . (9.13)

So we have to determine the subfield subcode U |IFq = CL (D, G1 )|IFq . Consider
an element x ∈ L (G1 ) such that x(Pi ) ∈ IFq for i = 1, . . . , n. Then xq − x ∈
 9.2 Weights of Trace Codes 315

L (G) and (xq − x)(Pi ) = 0, hence xq − x ∈ L (G − D). Since we assumed

that deg(G − D) < 0, it follows that xq − x = 0; i.e., x ∈ IFq . Consequently

1 if G1 ≥ 0 ,
dim U |IFq =
0 if G1 ≥ 0 .

Substituting this into (9.13) we get the desired estimate (9.11) for the dimen-
sion of Tr(CL ).
The corresponding estimate (9.12) for the dimension of CΩ |IFq follows from
Corollary 9.1.5.

Remark 9.1.7. In addition to the hypotheses of Theorem 9.1.6 assume that

deg G1 > 2g − 2. Then we can replace the terms (G) and (G1 ) in (9.11) and
(9.12) by deg G and deg G1 . This follows immediately from the Riemann-Roch
Theorem.

Example 9.1.8. As an illustration of Theorem 9.1.6 we consider a Goppa code

Γ (L, g(z)) = CΩ (DL , G0 − P∞ )|IFq

(notation as in Definition 2.3.10 and Proposition 2.3.11). Let g1 (z) ∈ IFqm [z]
be the polynomial of maximal degree such that g1 (z)q divides g(z). We set
G1 := (g1 (z))0 − P∞ where (g1 (z))0 is the zero divisor of g1 (z), and obtain
from (9.12) the estimate

dim Γ (L, g(z)) ≥ n − m(deg g(z) − deg g1 (z)) . (9.14)

In many cases, equality holds in (9.14). This will be shown in Proposition

9.2.13.

9.2 Weights of Trace Codes

In this section we investigate some specific trace codes. The main idea is to
relate the weights of their codewords to the number of rational places in certain
algebraic function fields. The Hasse-Weil-Serre Bound then yields estimates
for the weights and the minimum distance of these codes.
First we introduce the codes to be considered.

Definition 9.2.1. Let F be an algebraic function field over the constant

field IFqm and let V ⊆ F be a finite-dimensional IFqm -subspace of F . Let
P1 , . . . , Pn ∈ IPF be n distinct places of degree one such that vPi (f ) ≥ 0 for
all f ∈ V and i = 1, . . . , n. Set D := P1 + . . . + Pn . Then we define

C(D, V ) := {(f (P1 ), . . . , f (Pn )) | f ∈ V } ⊆ (IFqm )n

316 9 Subfield Subcodes and Trace Codes

and
TrD (V ) := Tr(C(D, V )) ⊆ IFnq ;
i.e., TrD (V ) is the trace code of C(D, V ) with respect to the extension
IFqm /IFq .

Note that C(D, V ) is a code over IFqm , whereas TrD (V ) is a code over IFq .
Our main objective in this section is to study the codes TrD (V ). Let us first
give some examples of such codes.

Example 9.2.2. The codes C(D, V ) are a generalization of algebraic geometry

codes. Choosing V := L (G) where G is a divisor with supp G ∩ supp D = ∅
as usual, we obtain C(D, V ) = CL (D, G).

Example 9.2.3. Every code C ⊆ IFnq over IFq can be represented as C =

TrD (V ) for a suitable choice of V and D. This can be seen as follows. Choose
m ∈ IN sufficiently large such that q m ≥ n. Let F := IFqm (z) be the rational
function field over IFqm . Choose n distinct elements α1 , . . . , αn ∈ IFqm and
denote by Pi ∈ IPF the zero of z − αi . Choose a basis {a(1) , . . . , a(k) } of C
(j) (j)
over IFq . Write a(j) = (a1 , . . . , an ). For j = 1, . . . , k choose a polynomial
(j)
fj = fj (z) ∈ IFqm [z] satisfying fj (αi ) = ai for i = 1, . . . , n. Let V ⊆ F be
the IFqm -vector space generated by f1 , . . . , fk . Then it is easily verified that
C = TrD (V ).

More interesting than the previous example is the fact that specific classes
of codes over IFq can be represented as trace codes in a natural manner. In
the following we give such a representation for cyclic codes.
A code C over IFq of length n is said to be cyclic if its automorphism group
Aut(C) contains the cyclic shift; i.e.,

(c0 , c1 , . . . , cn−1 ) ∈ C =⇒ (c1 , . . . , cn−1 , c0 ) ∈ C .

As is common in coding theory, we identify IFnq with the vector space of

polynomials of degree ≤ n − 1 over IFq via

c = (co , . . . , cn−1 ) ←→ c(x) = co + c1 x + . . . + cn−1 xn−1 ∈ IFq [x] . (9.15)

We shall always assume that

gcd(n, q) = 1 . (9.16)

Let m be the least integer ≥ 1 satisfying q m ≡ 1 mod n. Then the polynomial

xn − 1 factors over the field IFqm ⊇ IFq as


n−1
xn − 1 = (x − β ν ) , (9.17)
ν=0
 9.2 Weights of Trace Codes 317

where β ∈ IFqm is a primitive n-th root of unity. All linear factors in (9.17)
are distinct.
Let us briefly recall some basic facts about cyclic codes, cf. [28]. Given
a cyclic code C = {0} of length n over IFq , there exists a unique monic
polynomial g(x) ∈ C of minimal degree; it is called the generator polynomial
of C. The generator polynomial divides xn − 1, so

g(x) = (x − β ν ) , (9.18)
ν∈I

where β is a primitive n-th root of unity as in (9.17) and I is a certain subset

of {0, . . . , n− 1}. The elements β ν with ν ∈ I are called the zeros of C because
one has the following description of C: for c(x) ∈ IFq [x] with deg c(x) ≤ n − 1,

c(x) ∈ C ⇐⇒ c(β ν ) = 0 for all ν ∈ I . (9.19)

The conditions on the right-hand side of (9.19) can be weakened. To this end
we define the cyclotomic coset C(i) of an integer i ∈ ZZ, 0 ≤ i ≤ n − 1, by

C(i) := {j ∈ ZZ | 0 ≤ j ≤ n − 1 and j ≡ q l i mod n for some l ≥ 0} .

It is easily checked that either C(i) = C(i ) or C(i) ∩ C(i ) = ∅. Hence the
set {0, 1, . . . , n − 1} is partitioned into pairwise disjoint cyclotomic cosets;
s
i.e., {0, 1, . . . , n − 1} = μ=1 Cμ (with s ≤ n and Cμ = C(iμ ) for some iμ ,
0 ≤ iμ ≤ n − 1).
For ν ∈ ZZ denote by ν̃ the unique integer in {0, 1, . . . , n − 1} with ν ≡
ν̃ mod n. Let ∅ = M ⊆ {0, 1, . . . , n − 1}. A subset M0 ⊆ ZZ is called a complete
set of cyclotomic coset representatives of M , if for each ν ∈ M there exists a
unique ν0 ∈ M0 such that ν̃0 ∈ C(ν). It is evident that one can always find a
complete set of cyclotomic coset representatives of M which is contained in
{0, 1, . . . , n − 1}.
Now consider the set I ⊆ {0, 1, . . . , n − 1} given by (9.18); i.e., {β ν | ν ∈ I}
is the set of zeros of the cyclic code C. Let I0 be a complete set of cyclotomic
coset representatives of I. Since for c(x) ∈ IFq [x],
 l 
c(β ν ) = 0 ⇐⇒ c β q ν = 0 ,

we can replace (9.19) by the following condition:

c(x) ∈ C ⇐⇒ c(β ν ) = 0 for all ν ∈ I0 , (9.20)

where c(x) ∈ IFq [x] and deg c(x) ≤ n − 1.

The dual code C ⊥ of a cyclic code C is cyclic as well. Let g(x) ∈ IFq [x] be
the generator polynomial of C and

h(x) := (xn − 1)/g(x) ∈ IFq [x] . (9.21)

318 9 Subfield Subcodes and Trace Codes

The polynomial h(x) is called the check polynomial of C. The reciprocal poly-
nomial h⊥ (x) of h(x),

h⊥ (x) := h(0)−1 · xdeg h(x) · h(x−1 ) , (9.22)

is the generator polynomial of C ⊥ . Write


h⊥ (x) = (x − β ρ ) with J ⊆ {0, 1, . . . , n − 1} . (9.23)
ρ∈J

It follows from (9.21), (9.22) and (9.23) that

ρ ∈ J ⇐⇒ h⊥ (β ρ ) = 0 ⇐⇒ h(β −ρ ) = 0
⇐⇒ g(β −ρ ) = 0 ⇐⇒ ρ ≡ −ν mod n for all ν ∈ I .

Let J0 be a complete set of cyclotomic coset representatives of J. From (9.20)

we obtain for a(x) = a0 + a1 x + . . . + an−1 xn−1 ∈ IFq [x],

a(x) ∈ C ⊥ ⇐⇒ a(β ρ ) = 0 for all ρ ∈ J0 . (9.24)

Using the canonical inner product  ,  on (IFqm )n , the equation a(β ρ ) = 0

can also be written as
# $
(a0 , a1 , . . . , an−1 ), (1, β ρ , β 2ρ , . . . , β (n−1)ρ ) = 0 . (9.25)

Now we are in a position to represent the given cyclic code C as a trace

code. We consider the rational function field F = IFqm (z) and the vector
space V ⊆ F generated by the set {z ρ | ρ ∈ J0 }. Denote by Pi ∈ IPF the zero of
z−β i−1 and set D = P1 +. . .+Pn . By definition, the code C(D, V ) is generated
– as a vector space over IFqm – by the vectors (1, β ρ , β 2ρ , . . . , β (n−1)ρ ) with
ρ ∈ J0 . We conclude from (9.24) and (9.25) that

C ⊥ = C(D, V )⊥ |IFq .

Applying Delsarte’s theorem we finally obtain

 ⊥
C = C(D, V )⊥ |IFq = Tr(C(D, V )) = TrD (V ) .

We summarize:

Proposition 9.2.4. Let C be a cyclic code of length n over IFq with generator
polynomial g(x), and let IFqm = IFq (β) where β is a primitive n-th root of
unity. Let J = {0 ≤ ρ ≤ n − 1 | g(β −ρ ) = 0} and let J0 be a complete set
of cyclotomic coset representatives of J. Consider the rational function field
F = IFqm (z) over IFqm and the IFqm -vector space V ⊆ F which is generated
by {z ρ | ρ ∈ J0 }. Then C = TrD (V ) where D = P1 + . . . + Pn and Pi ∈ IPF is
the zero divisor of z − β i−1 (i = 1, . . . , n).
 9.2 Weights of Trace Codes 319

We return to the general situation as described at the beginning of this

section. Given a finite-dimensional vector space {0} = V ⊆ F , there exists
a unique effective divisor A of smallest degree such that (f )∞ ≤ A for all
0 = f ∈ V . One can describe A as follows: choose a basis {f1 , . . . , fk } of V ,
then
vP (A) = max {vP ((fi )∞ ) | 1 ≤ i ≤ k} (9.26)
for all P ∈ IPF . For instance, if V = L (G) and G = G+ − G− where G+ ≥ 0
and G− ≥ 0, then A ≤ G+ . We associate with the divisor A a second divisor
A0 defined by 
A0 := P. (9.27)
P ∈supp A

To avoid complications, we restrict ourselves to the case where q = p is a

prime number and F is the rational function field over IFpm .
Theorem 9.2.5. Let F = IFpm (z) be the rational function field over IFpm
(where p is a prime number), D = P1 + . . . + Pn with pairwise distinct places
Pi ∈ IPF of degree one, and s := pm +1−n. Let V ⊆ F be a finite-dimensional
IFpm -subspace of F such that vPi (f ) ≥ 0 for all f ∈ V and i = 1, . . . , n. Then
the weight w of a codeword in TrD (V ) satisfies w = 0, w = n or
 p − 1  p − 1 p−1

w − n ≤ (−2 + deg A + deg A0 ) · [2pm/2 ] + s,
p 2p p
where the divisors A and A0 are defined by (9.26) and (9.27).

The proof of this theorem requires some preparation. Let us first introduce
some notation. For f ∈ V let
 
TrD (f ) := Tr(f (P1 )), . . . , Tr(f (Pn )) ∈ IFnp ,
so that TrD (V ) = {TrD (f ) | f ∈ V }.
Definition 9.2.6. An element f ∈ F is said to be degenerate if f can be
written as f = γ + (hp − h), with γ ∈ IFpm and h ∈ F . Otherwise f is called
non-degenerate.
Lemma 9.2.7. Suppose f ∈ V is degenerate. Then
TrD (f ) = (α, α, . . . , α) with α ∈ IFp .
Hence the weight of TrD (f ) is 0 or n.
Proof. Write f = γ + (hp − h) with γ ∈ IFpm and h ∈ F . Since vPi (f ) ≥ 0, the
Triangle Inequality yields vPi (h) ≥ 0 for 1 ≤ i ≤ n. Setting γi := h(Pi ) ∈ IFpm
we obtain
Tr(f (Pi )) = Tr(γ) + Tr(γip − γi ) = Tr(γ) ,
independent of i (note that Tr is the trace mapping to IFp , so Tr(γip ) = Tr(γi )).
320 9 Subfield Subcodes and Trace Codes

The non-degenerate case is clearly more interesting. In the following propo-

sition we retain notation of Theorem 9.2.5.
Proposition 9.2.8. Suppose f ∈ V is non-degenerate. Then the polynomial
ϕ(Y ) := Y p − Y − f ∈ F [Y ] is irreducible over F . Let Ef := F (y) where y is
a root of ϕ(Y ); i.e.,

Ef = F (y) with yp − y = f .

The field extension Ef /F is cyclic of degree p, and IFpm is the full constant
field of Ef . Let S := {P ∈ IPF | deg P = 1 and P ∈ supp D}, so |S| = s =
pm + 1 − n. Let S̄ := {Q ∈ IPEf | deg Q = 1 and Q ∩ F ∈ S}, and denote by s̄
the cardinality of S̄. Then
0 ≤ s̄ ≤ ps , (9.28)
and the weight wf := w(TrD (f )) is given by the formula
N (Ef ) − s̄
wf = n − . (9.29)
p
(As usual, N (Ef ) denotes the number of all places of degree one of Ef /IFpm ).

Proof. We use some facts about Artin-Schreier extensions, cf. Appendix A.

The polynomial ϕ(Y ) = Y p −Y −f is either irreducible in F [Y ] or it has a root
in F ; i.e., f = hp −h with h ∈ F . Since f is assumed to be non-degenerate, the
irreducibility of ϕ(Y ) follows, and the field Ef = F (y) defined by y p − y = f
is a cyclic extension of F of degree p.
Suppose that the constant field L of Ef is strictly larger than IFpm , the
constant field of F . Then L/IFpm is cyclic of degree p, hence L = IFpm (δ) with
δ p − δ = ! ∈ IFpm for some δ ∈ L. As δ ∈ F , it is an Artin-Schreier generator
of Ef /F as well, so f = λ! + (hp − h) with 0 = λ ∈ IFp and h ∈ F (see
Appendix A). This is a contradiction as f is non-degenerate, and proves that
IFpm is the full constant field of Ef .
The support of D consists of two disjoint subsets {P1 , . . . , Pn } = N ∪ Z
where
N := {Pi ∈ supp D | Tr(f (Pi )) = 0}
and
Z := {Pi ∈ supp D | Tr(f (Pi )) = 0} .
We determine the decomposition behavior of the places Pi ∈ N (resp. Z) in
the extension Ef /E.
Hilbert’s Theorem 90 (Appendix A) states that for γ ∈ IFpm ,

Tr(γ) = 0 ⇐⇒ γ = β p − β for some β ∈ IFpm . (9.30)

Let Pi ∈ N and γi := f (Pi ). The Artin-Schreier polynomial Y p − Y − γi has

no root in IFpm by (9.30), hence it is irreducible over IFpm . Now Kummer’s
 9.2 Weights of Trace Codes 321

Theorem shows that Pi has a unique extension Q ∈ IPEf , with relative degree
f (Q|Pi ) = p. Consequently there are no places of Ef of degree one lying over
a place Pi ∈ N .
Next we consider a place Pi ∈ Z. By (9.30), f (Pi ) can be written as
f (Pi ) =: γi = βip − βi with βi ∈ IFpm , hence the polynomial Y p − Y − γi
factors over IFpm into p distinct linear factors. In this case Kummer’s Theorem
implies that Pi decomposes into p distinct places of Ef , all of degree one.
The above considerations imply that

N (Ef ) = |S̄| + p · |Z| = s̄ + p(n − |N |) . (9.31)

Since wf = w(TrD (f )) = |N |, (9.29) is an immediate consequence of (9.31).

The estimate (9.28) is trivial.

The next aim is to determine the genera g(Ef ) of the function fields Ef
(where f ∈ V is non-degenerate). Using Proposition 3.7.8 one can sometimes
calculate g(Ef ) precisely. We shall be satisfied with an upper bound on g(Ef ).

Lemma 9.2.9. Suppose f ∈ V is non-degenerate. Then

p−1
g(Ef ) ≤ (−2 + deg A + deg A0 ) ,
2
where A and A0 are defined by (9.26) and (9.27).

Proof. This lemma is an easy application of Proposition 3.7.8(d). Observe

that all places P ∈ supp A are unramified in Ef /F , and for P ∈ supp A the
integer mP (as defined in Proposition 3.7.8) is obviously bounded by vP (A).

Proof of Theorem 9.2.5. By Lemma 9.2.7 we can assume that w = wf is the

weight of a codeword TrD (f ) where f ∈ V is non-degenerate. We use the
notation of Proposition 9.2.8. Equation (9.29) yields

N (Ef ) = p(n − wf ) + s̄ .

We subtract pm + 1, apply the Serre Bound (Theorem 5.3.1) and obtain

 
p(n − wf ) + s̄ − (pm + 1) ≤ g(Ef ) · [2pm/2 ] . (9.32)

Since pm + 1 = s + n, we have

p(n − wf ) + s̄ − (pm + 1) = (p − 1)n − pwf + (s̄ − s) .

We substitute this into (9.32), divide by p and estimate g(Ef ) by Lemma

9.2.9. The result is
322 9 Subfield Subcodes and Trace Codes
 p−1 s̄ − s  p − 1  

wf − n− ≤ −2 + deg A + deg A0 · [2pm/2 ] . (9.33)
p p 2p

Finally (9.28) yields that −s ≤ s̄ − s ≤ (p − 1)s, and therefore

 s̄ − s  p − 1
 
 ≤ · s.
p p
This finishes the proof of Theorem 9.2.5.

Remark 9.2.10. (a) In some specific cases the number s̄ can be determined
more precisely. In such cases, (9.33) provides a better estimate than does
Theorem 9.2.5, cf. Example 9.2.12 below.
(b) Clearly Theorem 9.2.5 gives non-trivial bounds for the weights of code-
words in TrD (V ) only if the length of the code is large in comparison with m
and the degree of A. It turns out that – under this restriction – the estimates
given in Theorem 9.2.5 are often fairly good.

Corollary 9.2.11. Notation as in Theorem 9.2.5. If V = {0} and V = IFpm ,

the minimum distance d of TrD (V ) is bounded from below by
p−1 s p−1
d≥ n− − (−2 + deg A + deg A0 ) · [2pm/2 ] . (9.34)
p p 2p

Proof. The assumption V = {0} and V = IFpm implies that deg A > 0, hence
the right-hand side of (9.34) is ≤ n. Therefore it is sufficient to estimate the
weight wf = w(TrD (f )) for a non-degenerate element f ∈ V . By (9.33),
p−1 s̄ − s p − 1
wf ≥ n+ − (−2 + deg A + deg A0 ) · [2pm/2 ] .
p p 2p
Since s̄ ≥ 0, the corollary follows.

Example 9.2.12. We consider the dual C ⊥ of the BCH code C over IFp of
length n = pm − 1 and designed distance δ = 2t + 1 > 1. Thus
%  n−1 &

C = (c0 , c1 , . . . , cn−1 ) ∈ IFnp  ci β iλ = 0 for λ = 1, . . . , δ − 1 ,
i=0

where β ∈ IFpm is a primitive (pm − 1)-th root of unity, cf. Definition 2.3.8.
Then the weight w of a codeword in the dual code C ⊥ satisfies w = 0, w = n
or
   ' (
w − pm 1 − 1  ≤ (p − 1)(2t − 1) · 2pm/2 . (9.35)
p 2p
In the case p = 2 this can be improved to
 9.2 Weights of Trace Codes 323
  ' (
w − 2m−1  ≤ t − 1 · 2 m2 +1 . (9.36)
2
This is the so-called Carlitz-Uchiyama Bound, cf. [28]. Note that the bounds
(9.35) and (9.36) are not covered by Theorem 9.2.5.
We show now how (9.35) and (9.36) follow from our previous results. Let
F = IFpm (z). For 1 ≤ i ≤ n let Pi ∈ IPF be the zero of z − β i−1 . Set
D = P1 + . . . + Pn and (z) = P0 − P∞ . By Proposition 2.3.9 and Delsarte’s
Theorem,
  ⊥
C ⊥ = CL (D, aP0 + bP∞ )⊥ IF = TrD (L (aP0 + bP∞ )) ,
p

where a = −1 and δ = a + b + 2 = b + 1. Since δ = 2t + 1 we find that

C ⊥ = TrD (L (−P0 + 2tP∞ )) .

Let f ∈ L (−P0 + 2tP∞ ) be non-degenerate. Consider the corresponding field

extension Ef = F (y), defined by y p − y = f . Since f (P0 ) = 0, the place P0
has p extensions of degree one in Ef . The pole P∞ of z ramifies in Ef /F .
With notation as in Proposition 9.2.8 we obtain s̄ = p + 1. Hence
   
wf − 1 − 1 · pm  ≤ g(Ef ) · [2pm/2 ] ,
p p

by (9.32). Since f ∈ L (−P0 + 2tP∞ ), Proposition 3.7.8 yields

p−1 (p − 1)(2t − 1)
g(Ef ) ≤ · (−2 + (2t + 1)) = .
2 2
This proves (9.35). In case p = 2 the genus is ≤ (2t − 2)/2 because the integer
mP∞ in Proposition 3.7.8 is relatively prime to the characteristic and therefore
mP∞ ≤ 2t − 1 for p = 2, which gives (9.36).

The method of proof of Theorem 9.2.5 can also be used to calculate pre-
cisely the dimension of certain trace codes and of their duals. We illustrate
this idea by an example which also shows that the estimate for the dimension
of trace codes given in Theorem 9.1.6 is often tight.

Proposition 9.2.13. Consider a Goppa code Γ (L, g(z)) over IFp , where L =
IFpm and g(z) ∈ IFpm [z] is a monic polynomial without zeros in IFpm (see
Definition 2.3.10). Write


l
g(z) = hj (z)aj
j=1

with pairwise distinct irreducible monic polynomials hj (z) ∈ IFpm [z] and ex-
ponents aj > 0. Let aj = pbj + cj with 0 ≤ cj ≤ p − 1. Set
324 9 Subfield Subcodes and Trace Codes


l 
l
bj
g1 (z) := hj (z) and g2 (z) := hj (z) .
j=1 j=1

Suppose that

2(pm + 1) > (−2 + deg g(z) + deg g2 (z)) · [2pm/2 ] (9.37)

holds. Then the dimension of Γ (L, g(z)) is

dim Γ (L, g(z)) = pm − m(deg g(z) − deg g1 (z)) . (9.38)

Proof. By Proposition 2.3.11 and Delsarte’s Theorem we can represent the

dual code Γ (L, g(z))⊥ as follows:
  ⊥
Γ (L, g(z))⊥ = CL (D, G0 − P∞ )⊥ IFp = TrD (L (G0 − P∞ )) (9.39)

with the usual notation: P∞ is the pole divisor of z in the rational function
field F = IFpm (z), the divisor G0 is the zero divisor of the Goppa polynomial
g(z), and D is the sum of all places of F/IFpm of degree one except P∞ (note
that L = IFpm ). The IFp -linear map TrD : L (G0 − P∞ ) → TrD (L (G0 − P∞ ))
is surjective, and we wish to determine its kernel.
Claim. If f ∈ L (G0 − P∞ ) and TrD (f ) = 0, then f is degenerate.
Proof of the Claim. Suppose that f ∈ L (G0 − P∞ ) is non-degenerate and
TrD (f ) = 0. We consider the corresponding field extension Ef /F of degree
p (see Proposition 9.2.8). As f (P∞ ) = 0, the place P∞ decomposes into p
distinct places of Ef of degree one (by Kummer’s Theorem). So (9.29) yields
0 = wf = pm − p−1 (N (Ef ) − p); i.e.,

N (Ef ) = p(pm + 1) . (9.40)

The genus g(Ef ) is bounded by

p−1
g(Ef ) ≤ (−2 + deg g(z) + deg g2 (z)) , (9.41)
2
by Proposition 3.7.8(d). Combining (9.40), (9.41) and the Serre Bound we
obtain
p−1
p(pm + 1) ≤ pm + 1 + (−2 + deg g(z) + deg g2 (z)) · [2pm/2 ] ,
2
which contradicts (9.37). This proves the claim.
Let G1 := (g1 (z))0 denote the zero divisor of the polynomial g1 (z). We
have an IFp -linear map

L (G1 − P∞ ) −→ Ker(TrD ) ⊆ L (G0 − P∞ ) ,
φ:
h −→ hp − h .
 9.3 Exercises 325

Since hp − h = 0 if and only if h ∈ IFp , the kernel of φ is IFp ∩ L (G1 − P∞ ) =

{0}. Thus φ is injective.
φ is also surjective. In order to see this, let f ∈ Ker(TrD ). By the claim,
f = hp1 − h1 + γ with h1 ∈ F, γ ∈ IFpm . From TrD (f ) = 0 follows that
Tr(γ) = 0. This implies by Hilbert’s Theorem 90 that γ = αp − α with
α ∈ IFpm . Hence f = hp − h with h := h1 + α ∈ F . The Triangle Inequality
shows that h ∈ L (G1 ) (because f = hp − h ∈ L (G0 )). Moreover it follows
from 
(h − μ) = hp − h = f ∈ L (G0 − P∞ ) ,
μ∈IFp

that P∞ is a zero of one of the factors h − μ. This factor h − μ lies in the space
L (G1 − P∞ ), hence f = (h − μ)p − (h − μ) = φ(h − μ) is in the image of φ.
In what follows, we denote by dim V the dimension of a vector space V
over IFp . We conclude that
dim Ker(TrD ) = dim L (G1 − P∞ ) = m · deg g1 (z) ,
hence
dim Γ (L, g(z))⊥ = dim TrD (L (G0 − P∞ ))
= dim L (G0 − P∞ ) − dim Ker(TrD ) = m(deg g(z) − deg g1 (z)) .
The dimension of the dual code is therefore
dim Γ (L, g(z)) = pm − m(deg g(z) − deg g1 (z)) .

9.3 Exercises
9.1. Find a larger class of trace codes for which the method of Proposition
9.2.13 applies.
9.2. Consider the extension IFqm /IFq of degree m ≥ 1 and an arbitrary IFq -
linear map λ : IFqm → IFq which is not identically zero. Extend this to a map
λ : (IFqm )n → IFnq by setting λ(c1 , . . . , cn ) = (λ(c1 ), . . . , λ(cn )). For a code C
of length n over IFqm define
λ(C) := {λ(c) | c ∈ C} ,
which is obviously a code over IFq of length n. Show that λ(C) = Tr (C).
9.3. Let C be a code of length n over IFqm . Assume there is an r × n matrix
M of rank s with coefficients in IFq such that M · ct = 0 for all c ∈ C. Show
that
dim C|IFq ≥ dim C − (m − 1)(n − s − dim C) .
Give examples (with m > 1) of codes where the above estimate is sharp.
326 9 Subfield Subcodes and Trace Codes

9.4. Let n = pm − 1 where p is prime and m > 1. Consider a cyclic code C

over IFp of length n such that (1, 1, . . . , 1) ∈
/ C.
(i) Show that there exists a complete set of coset representatives for C ⊥ of the
form J0 = {ρ1 , . . . , ρs } ⊆ {1, . . . , n − 1} such that (p, ρi ) = 1 for i = 1, . . . , s.
(ii) Show that the weights w of nonzero codewords c ∈ C satisfy
   ' (
w − pm 1 − p−1  ≤ (p − 1)(ρ − 1) · 2pm/2
2p

with ρ := max{ρ1 , . . . , ρs }.
Appendix A. Field Theory

We put together some facts from field theory that we frequently call upon.
Proofs can be found in all standard textbooks on algebra, e.g. [23]

A.1 Algebraic Field Extensions

Let L be a field that contains K as a subfield. Then L/K is called a field

extension. Considering L as a vector space over K, its dimension is called the
degree of L/K and denoted by [L : K].
L/K is said to be a finite extension if [L : K] = n < ∞. Then there exists
a basisn{α1 , . . . , αn } of L/K; i.e., every γ ∈ L has a unique representation
γ = i=1 ci αi with ci ∈ K. If L/K and M/L are finite extensions, then
M/K is finite as well, and the degree is [M : K] = [M : L] · [L : K].
An element α ∈ L is algebraic over K if there is a non-zero polynomial
f (X) ∈ K[X] (the polynomial ring over K) such that f (α) = 0. Among all
such polynomials there is a unique polynomial of smallest degree that is monic
(i.e., its leading coefficient is 1); this is called the minimal polynomial of α
over K. The minimal polynomial is irreducible in the ring K[X], hence it is
often called the irreducible polynomial of α over K.
The field extension L/K is called an algebraic extension if all elements
α ∈ L are algebraic over K.
Let γ1 , . . . , γr ∈ L. The smallest subfield of L that contains K and all ele-
ments γ1 , . . . , γr is denoted by K(γ1 , . . . , γr ). The extension K(γ1 , . . . , γr )/K
is finite if and only if all γi are algebraic over K.
In particular, α ∈ L is algebraic over K if and only if [K(α) : K] < ∞. Let
p(X) ∈ K[X] be the minimal polynomial of α over K and r = deg p(X). Then
[K(α) : K] = r, and the elements 1, α, α2 , . . . , αr−1 form a basis of K(α)/K.
328 Appendix A. Field Theory

A.2 Embeddings and K-Isomorphisms

Consider field extensions L1 /K and L2 /K. A field homomorphism σ : L1 →

L2 is called an embedding of L1 into L2 over K, if σ(a) = a for all a ∈ K. It
follows that σ is injective and yields an isomorphism of L1 onto the subfield
σ(L1 ) ⊆ L2 . A surjective (hence bijective) embedding of L1 into L2 over K is
a K-isomorphism.

A.3 Adjoining Roots of Polynomials

Given a field K and a non-constant polynomial f (X) ∈ K[X], there exists an

algebraic extension field L = K(α) with f (α) = 0. If f (X) is irreducible, this
extension field is unique up to K-isomorphism. This means: if L = K(α ) is
another extension field with f (α ) = 0, then there exists a K-isomorphism
σ : L → L with σ(α) = α . We say that L = K(α) is obtained by adjoining
a root of p(X) to K.
If f1 (X), . . . , fr (X) ∈ K[X] are monic polynomials of degree di ≥ 1, there
exists an extension field Z ⊇ K such that all fi (X) split into linear factors
di
fi (X) = j=1 (X − αij ) with αij ∈ Z, and Z = K({αij | 1 ≤ i ≤ r and 1 ≤
j ≤ di }). The field Z is unique up to K-isomorphism; it is called the splitting
field of f1 , . . . , fr over K.

A.4 Algebraic Closure

A field M is called algebraically closed if every polynomial f (X) ∈ M [X] of

degree ≥ 1 has a root in M .
For every field K there exists an algebraic extension K̄/K with an alge-
braically closed field K̄. The field K̄ is unique up to K-isomorphism; it is
called the algebraic closure of K.
Given an algebraic field extension L/K, there exists an embedding σ :
L → K̄ over K. If [L : K] < ∞, the number of distinct embeddings of L to
K̄ over K is at most [L : K].

A.5 The Characteristic of a Field

Let K be a field and let 1 ∈ K be the neutral element with respect to multipli-
cation. For each integer m > 0, let m̄ = 1 + 1 + . . . + 1 ∈ K (m summands). If
m̄ = 0 (the zero element of K) for all m > 0, we say that K has characteristic
zero. Otherwise there exists a unique prime number p ∈ IN such that p̄ = 0,
and K is said to have characteristic p. We use the abbreviation char K. It
 Appendix A. Field Theory 329

is convenient to identify an integer m ∈ ZZ with the element m̄ ∈ K ; i.e., we

simply write m = m̄ ∈ K.
If char K = 0, then K contains the field Q of rational numbers (up to
isomorphism). In case char K = p > 0, K contains the field IFp = ZZ/pZZ.
In a field of characteristic p > 0 we have (a + b)q = aq + bq for all a, b ∈ K
and q = pj , j ≥ 0.

A.6 Separable Polynomials

Let f (X) ∈ K[X] be a monic polynomial of degree d ≥ 1. Over some extension

d
field L ⊇ K, f (X) splits into linear factors f (X) = i=1 (X − αi ). The
polynomial f (X) is called separable if αi = αj for all i = j; otherwise, f is an
inseparable polynomial.
If char K = 0, all irreducible polynomials
 are iseparable. In case char K =
p > 0, an irreducible polynomial f (X) = ai X ∈ K[X] is separable if and
only if ai = 0 for some i ≡ 0 mod p.

The derivative
 of f (X) = ai X i ∈ K[X] is defined in the usual manner

by f (X) = iai X i−1
(where i ∈ IN is considered as an element of K as
in A.5). An irreducible polynomial f (X) ∈ K[X] is separable if and only if
f  (X) = 0.

A.7 Separable Field Extensions

Let L/K be an algebraic field extension. An element α ∈ L is called separable

over K if its minimal polynomial p(X) ∈ K[X] is a separable polynomial.
L/K is a separable extension if all α ∈ L are separable over K. If char K = 0,
then all algebraic extensions L/K are separable.
Let Φ be an algebraically closed field, Φ ⊇ K, and suppose that L/K is a
finite extension of degree [L : K] = n. Then L/K is separable if and only if
there exist n distinct embeddings σ1 , . . . , σn : L → Φ over K (cf. A.4). In this
case an element γ ∈ L is in K if and only if σi (γ) = γ for i = 1, . . . , n.
Given a tower M ⊇ L ⊇ K of algebraic field extensions, the extension
M/K is separable if and only if both extensions M/L and L/K are separable.

A.8 Purely Inseparable Extensions

Consider an algebraic extension L/K where char K = p > 0. An element
r
γ ∈ L is called purely inseparable over K if γ p ∈ K for some r ≥ 0. In this
e
case the minimal polynomial of γ over K has the form f (X) = X p − c with
330 Appendix A. Field Theory

c ∈ K (and e ≤ r). The extension L/K is purely inseparable if all elements

γ ∈ L are purely inseparable over K.
Given an arbitrary algebraic extension L/K, there exists a unique inter-
mediate field S, K ⊆ S ⊆ L, such that S/K is separable and L/S is purely
inseparable.

A.9 Perfect Fields

A field K is called perfect if all algebraic extensions L/K are separable. Fields
of characteristic 0 are always perfect. A field K of characteristic p > 0 is
perfect if and only if every α ∈ K can be written as α = β p , for some β ∈ K.
All finite fields are perfect (cf. A.15).

A.10 Simple Algebraic Extensions

An algebraic extension L/K is called simple if L = K(α) for some α ∈ L.

The element α is called a primitive element for L/K. Every finite separable
algebraic field extension is simple.
Suppose that L = K(α1 , . . . , αr ) is a finite separable extension and K0 ⊆
K is an infinite
r subset of K. Then there exists a primitive element α of the
form α = i=1 ci αi with ci ∈ K0 .

A.11 Galois Extensions

For a field extension L/K we denote the group of automorphisms of L over

K by Aut(L/K). That is, an element σ ∈ Aut(L/K) is a K-isomorphism
of L onto L. If [L : K] < ∞, the order of Aut(L/K) is always ≤ [L : K].
The extension L/K is said to be Galois if the order of Aut(L/K) is [L : K].
In this case we call Gal(L/K) := Aut(L/K) the Galois group of L/K. The
following conditions are equivalent, for a field extension L/K of finite degree:
(1) L/K is Galois.
(2) L is the splitting field of separable polynomials f1 (X), . . . , fr (X) ∈
K[X] over K.
(3) L/K is separable, and every irreducible polynomial p(X) ∈ K[X] that
has a root in L, splits into linear factors in L[X].
Given a finite separable extension L/K and an algebraically closed field
Φ ⊇ L, there exists a unique field M , L ⊆ M ⊆ Φ, with the following proper-
ties:
(a) M/K is Galois, and
 Appendix A. Field Theory 331

(b) if L ⊆ N ⊆ Φ and N/K is Galois, then M ⊆ N .

This field M is called the Galois closure of L/K. Another characterization
of M is that it is the compositum of the fields σ(L) where σ runs over all
embeddings of L into Φ over K.

A.12 Galois Theory

We consider a Galois extension L/K with Galois group G = Gal(L/K). Let
U := {U ⊆ G | U is a subgroup of G}
and
F := {N | N is an intermediate field of L/K} .
For each intermediate field N of L/K the extension L/N is Galois, thus we
have a mapping
F −→ U ,
N −→ Gal(L/N ) . (∗)
On the other hand, given a subgroup U ⊆ G we define the fixed field of U
by
LU := {c ∈ L | σ(c) = c for all σ ∈ U } .
In this manner we obtain a mapping
U −→ F ,
U −→ LU . (∗∗)
Now we can formulate the main results of Galois theory:
(1) The mappings (∗) and (∗∗) are inverse to each other. They yield a 1–1
correspondence between U and F (Galois correspondence).
(2) For U ∈ U we have
[L : LU ] = ord U and [LU : K] = (G : U ) .

(3) If U ⊆ G is a subgroup, then U = Gal(L/LU ).

(4) If N is an intermediate field of L/K, then N = LU with U = Gal(L/N ).
(5) For subgroups U, V ⊆ G we have U ⊆ V ⇐⇒ LU ⊇ LV .
(6) Let N1 and N2 be intermediate fields of L/K and N = N1 N2 be the
compositum of N1 and N2 . Then Gal(L/N ) = Gal(L/N1 ) ∩ Gal(L/N2 ).
(7) If N1 and N2 are intermediate fields of L/K, then the Galois group
of L/(N1 ∩ N2 ) is the subgroup of G that is generated by Gal(L/N1 ) and
Gal(L/N2 ).
(8) A subgroup U ⊆ G is a normal subgroup of G if and only if the
extension LU /K is Galois. In this case, Gal(LU /K) is isomorphic to the factor
group G/U .
332 Appendix A. Field Theory

A.13 Cyclic Extensions

A Galois extension L/K is said to be cyclic if Gal(L/K) is a cyclic group. Two
special cases are of particular interest: Kummer extensions and Artin-Schreier
extensions.
(1) (Kummer Extensions.) Suppose that L/K is a cyclic extension of de-
gree [L : K] = n, where n is relatively prime to the characteristic of K, and K
contains all n-th roots of unity (i.e., the polynomial X n − 1 splits into linear
factors in K[X]). Then there exists an element γ ∈ L such that L = K(γ)
with

γn = c ∈ K , and c = wd for all w ∈ K and d | n, d > 1 . (◦)

Such a field extension is called a Kummer Extension. The automorphisms

σ ∈ Gal(L/K) are given by σ(γ) = ζ · γ, where ζ ∈ K is an n-th root of unity.
Conversely, if K contains all n-th roots of unity (n relatively prime to
char K), and L = K(γ) where γ satisfies the conditions (◦), then L/K is a
cyclic extension of degree n.
(2) (Artin-Schreier Extensions.) Let L/K be a cyclic extension of degree
[L : K] = p = char K. Then there exists an element γ ∈ L such that L = K(γ),

γp − γ = c ∈ K , and c = αp − α for all α ∈ K . (◦◦)

Such a field extension is called an Artin-Schreier Extension of degree p. The

automorphisms of L/K are given by σ(γ) = γ + ν with ν ∈ ZZ/pZZ ⊆ K.
An element γ1 ∈ L such that L = K(γ1 ) and γ1p − γ1 ∈ K, is called an
Artin-Schreier generator for L/K. Any two Artin-Schreier generators γ and
γ1 of L/K are related as follows: γ1 = μ · γ + (bp − b) with 0 = μ ∈ ZZ/pZZ and
b ∈ K.
Conversely, if we have a field extension L = K(γ) where γ satisfies (◦◦)
(and char K = p), then L/K is cyclic of degree p.

A.14 Norm and Trace

Let L/K be a field extension of degree [L : K] = n < ∞. Each element α ∈ L
yields a K-linear map μα : L → L, defined by μα (z) := α · z for z ∈ L. We
define the norm (resp. the trace) of α with respect to the extension L/K by

NL/K (α) := det(μα ) resp. TrL/K (α) := Trace(μα ) .

This means: if {α1 , . . . , αn } is a basis of L/K and


n
α · αi = aij αj with aij ∈ K ,
j=1
 Appendix A. Field Theory 333

then
  
n
NL/K (α) = det aij 1≤i,j≤n and TrL/K (α) = aii .
i=1

We note some properties of norms and traces.

(1) The norm map is multiplicative; i.e., NL/K (α·β) = NL/K (α)·NL/K (β)
for all α, β ∈ L. Moreover, NL/K (α) = 0 ⇐⇒ α = 0, and for a ∈ K we have
NL/K (a) = an , with n = [L : K].
(2) For α, β ∈ L and a ∈ K the following hold:

TrL/K (α + β) = TrL/K (α) + TrL/K (β) ,

TrL/K (a · α) = a · TrL/K (α) , and
TrL/K (a) = n · a , with n = [L : K] .

In particular TrL/K is a K-linear map.

(3) If L/K and M/L are finite extensions, then

TrM/K (α) = TrL/K (TrM/L (α)) and

NM/K (α) = NL/K (NM/L (α))

for all α ∈ M .
(4) A finite field extension L/K is separable if and only if there exists an
element γ ∈ L such that TrL/K (γ) = 0 (since the trace map is K-linear, it
follows then that TrL/K : L → K is surjective).
(5) Let f (X) = X r + ar−1 X r−1 + . . . + a0 ∈ K[X] be the minimal poly-
nomial of α over K, and [L : K] = n = rs (with s = [L : K(α)]). Then

NL/K (α) = (−1)n as0 and TrL/K (α) = −sar−1 .

(6) Suppose now that L/K is separable of degree n. Consider the n distinct
embeddings σ1 , . . . , σn : L → Φ of L over K into an algebraically closed field
Φ ⊇ K. Then,

n 
n
NL/K (α) = σi (α) and TrL/K (α) = σi (α)
i=1 i=1

for α ∈ L.
(7) In particular, if L/K is Galois with Galois group G = Gal(L/K), then
 
NL/K (α) = σ(α) and TrL/K (α) = σ(α)
σ∈G σ∈G

for α ∈ L.
334 Appendix A. Field Theory

A.15 Finite Fields

Let p > 0 be a prime number and let q = pn be a power of p. Then there
exists a finite field IFq with |IFq | = q, and IFq is unique up to isomorphism; it
is the splitting field of the polynomial X q − X over the field IFp := ZZ/pZZ. In
this manner, we obtain all finite fields of characteristic p.
The multiplicative group IF× q of IFq is a cyclic group of order q − 1; i.e.,

IFq = {0, β, β 2 , . . . , β q−1 = 1} ,

where β is a generator of IF×

q .
Let m ≥ 1. Then IFq ⊆ IFqm , and the extension IFqm /IFq is a Galois
extension of degree m. The Galois group Gal(IFqm /IFq ) is cyclic; it is generated
by the Frobenius automorphism

IFqm −→ IFqm ,
ϕ:
α −→ αq .

In particular, all finite fields are perfect.

The norm and trace map from IFqm to IFq are given by
2
+...+q m−1
NIFqm /IFq (α) = α1+q+q ,
m−1
TrIFqm /IFq (α) = α + αq + . . . + αq .

Hilbert’s Theorem 90 states, for α ∈ IFqm ,

TrIFqm /IFq (α) = 0 ⇐⇒ α = β q − β for some β ∈ IFqm .

A.16 Transcendental Extensions

Let L/K be a field extension. An element x ∈ L that is not algebraic over
K is called transcendental over K. A finite subset {x1 , . . . , xn } ⊆ L is alge-
braically independent over K if there does not exist a non-zero polynomial
f (X1 , . . . , Xn ) ∈ K[X1 , . . . , Xn ] with f (x1 , . . . , xn ) = 0. An arbitrary sub-
set S ⊆ L is algebraically independent over K if all finite subsets of S are
algebraically independent over K.
A transcendence basis of L/K is a maximal algebraically independent sub-
set of L. Any two transcendence bases of L/K have the same cardinality, the
transcendence degree of L/K.
If L/K has finite transcendence degree n and {x1 , . . . , xn } is a trans-
cendence basis if L/K, the field K(x1 , . . . , xn ) ⊆ L is K-isomorphic to
K(X1 , . . . , Xn ), the quotient field of the polynomial ring K[X1 , . . . , Xn ] in
n variables over K. The extension L/K(x1 , . . . , xn ) is algebraic.
Appendix B. Algebraic Curves and Function
Fields

This appendix contains a brief survey of the relations between algebraic curves
and algebraic function fields. For details and proofs we refer to the literature
on algebraic geometry, for instance [11],[18],[37],[38].

We assume that K is an algebraically closed field.

B.1 Affine Varieties

The n-dimensional affine space An = An (K) is the set of all n-tuples of ele-
ments of K. An element P = (a1 , . . . , an ) ∈ An is a point, and a1 , . . . , an are
the coordinates of P .
Let K[X1 , . . . , Xn ] be the ring of polynomials in n variables over K. A
subset V ⊆ An is an algebraic set if there exists a set M ⊆ K[X1 , . . . , Xn ]
such that
V = {P ∈ An | F (P ) = 0 for all F ∈ M } .
Given an algebraic set V ⊆ An , the set of polynomials

I(V ) = {F ∈ K[X1 , . . . , Xn ] | F (P ) = 0 for all P ∈ V }

is called the ideal of V . I(V ) is obviously an ideal in K[X1 , . . . , Xn ], and it

can be generated by finitely many polynomials F1 , . . . , Fr ∈ K[X1 , . . . , Xn ].
Thus we have

V = {P ∈ An | F1 (P ) = . . . = Fr (P ) = 0} .

An algebraic set V ⊆ An is called irreducible if it cannot be written as

V = V1 ∪ V2 , where V1 and V2 are proper algebraic subsets of V . Equivalently,
V is irreducible if and only if the corresponding ideal I(V ) is a prime ideal.
An affine variety is an irreducible algebraic set V ⊆ An .
336 Appendix B. Algebraic Curves and Function Fields

The coordinate ring of an affine variety V is the residue class ring Γ (V ) =

K[X1 , . . . , Xn ]/I(V ) . As I(V ) is a prime ideal, Γ (V ) is an integral domain.
Every f = F + I(V ) ∈ Γ (V ) induces a function f : V → K by setting
f (P ) := F (P ) for P ∈ V . The quotient field

K(V ) = Quot(Γ (V ))

is called the field of rational functions (or the function field) of V . It contains
K as a subfield. The dimension of V is the transcendence degree of K(V )/K.
For a point P ∈ V let

OP (V ) = {f ∈ K(V ) | f = g/h with g, h ∈ Γ (V ) and h(P ) = 0 } .

This is a local ring with quotient field K(V ), its unique maximal ideal is

MP (V ) = {f ∈ K(V ) | f = g/h with g, h ∈ Γ (V ), h(P ) = 0 and g(P ) = 0}.

OP (V ) is called the local ring of V at P . For f = g/h ∈ OP (V ) with h(P ) = 0,

the value of f at P is defined to be f (P ) := g(P )/h(P ).

B.2 Projective Varieties

On the set An+1 \ {(0, . . . , 0)} an equivalence relation ∼ is given by

(a0 , a1 , . . . , an ) ∼ (b0 , b1 , . . . , bn ) : ⇐⇒
there is an element 0 = λ ∈ K such that bi = λai for 0 ≤ i ≤ n .

The equivalence class of (a0 , a1 , · · · , an ) with respect to ∼ is denoted by

(a0 : a1 : . . . : an ). The n-dimensional projective space Pn = Pn (K) is the set
of all equivalence classes

Pn = {(a0 : . . . : an ) | ai ∈ K, not all ai = 0} .

An element P = (a0 : . . . : an ) ∈ Pn is a point, and a0 , . . . , an are called

homogeneous coordinates of P .
A monomial of degree d is a polynomial G ∈ K[X0 , . . . , Xn ] of the form
n n
G = a · i=0 Xidi with 0 = a ∈ K and i=0 di = d.

A polynomial F is a homogeneous polynomial if F is the sum of monomi-

als of the same degree. An ideal I ⊆ K[X0 , . . . , Xn ] which is generated by
homogeneous polynomials is called a homogeneous ideal.
Let P = (a0 : . . . : an ) ∈ Pn and let F ∈ K[X0 , . . . , Xn ] be a homoge-
neous polynomial. We say that F (P ) = 0 if F (a0 , . . . , an ) = 0. This makes
sense: since F (λa0 , . . . , λan ) = λd · F (a0 , . . . , an ) (with d = deg F ), one has
F (a0 , . . . , an ) = 0 ⇐⇒ F (λa0 , . . . , λan ) = 0.
 Appendix B. Algebraic Curves and Function Fields 337

A subset V ⊆ Pn is a projective algebraic set if there exists a set of

homogeneous polynomials M ⊆ K[X0 , . . . , Xn ] such that

V = {P ∈ Pn | F (P ) = 0 for all F ∈ M } .

The ideal I(V ) ⊆ K[X0 , . . . , Xn ], which is generated by all homogeneous

polynomials F with F (P ) = 0 for all P ∈ V , is called the ideal of V . It is a
homogeneous ideal. Irreducibility of projective algebraic sets is defined as in
the affine case. Again, V ⊆ Pn is irreducible if and only if I(V ) is a homo-
geneous prime ideal in K[X0 , . . . , Xn ]. A projective variety is an irreducible
projective algebraic set.
Given a non-empty variety V ⊆ Pn , we define its homogeneous coordinate
ring by
Γh (V ) = K[X0 , . . . , Xn ]/I(V ) ;
this is an integral domain containing K. An element f ∈ Γh (V ) is said to
be a form of degree d if f = F + I(V ) for some homogeneous polynomial
F ∈ K[X0 , . . . , Xn ] with deg F = d. The function field of V is defined by
%g  &

K(V ) :=  g, h ∈ Γh (V ) are forms of the same degree and h = 0 ,
h
which is a subfield of Quot(Γh (V )), the quotient field of Γh (V ).
The dimension of V is the transcendence degree of K(V ) over K.
Let P = (a0 : . . . : an ) ∈ V and f ∈ K(V ). Write f = g/h where
g = G+I(V ), h = H +I(V ) ∈ Γh (V ) and G, H are homogeneous polynomials
of degree d. Since

G(λa0 , . . . , λan ) λd · G(a0 , . . . , an ) G(a0 , . . . , an )

= d = ,
H(λa0 , . . . , λan ) λ · H(a0 , . . . , an ) H(a0 , . . . , an )

we can set f (P ) := G(a0 , . . . , an )/H(a0 , . . . , an ) ∈ K, if H(P ) = 0. Then we

say that f is defined at P and call f (P ) the value of f at P . The ring

OP (V ) = {f ∈ K(V ) | f is defined at P } ⊆ K(V )

is a local ring with maximal ideal

MP (V ) = {f ∈ OP (V ) | f (P ) = 0} .

B.3 Covering Projective Varieties by Affine Varieties

For 0 ≤ i ≤ n we consider the mapping ϕi : An → Pn given by

ϕi (a0 , . . . , an−1 ) = (a0 : . . . : ai−1 : 1 : ai : . . . : an−1 ) .

ϕi is a bijection from An onto the set

338 Appendix B. Algebraic Curves and Function Fields

Ui = {(c0 : . . . : cn ) ∈ Pn | ci = 0} ,
n
and Pn = i=0 Ui . So Pn is covered by n + 1 copies of the affine space An
(this is not a disjoint union).
n
Let V ⊆ Pn be a projective variety, then V = i=0 (V ∩ Ui ). Suppose that
V ∩ Ui = ∅. Then
Vi := ϕ−1
i (V ∩ Ui ) ⊆ A
n

is an affine variety, and the ideal I(Vi ) (in the sense of B.1) is given by

I(Vi ) = {F (X0 , . . . , Xi−1 , 1, Xi+1 , . . . Xn ) | F ∈ I(V )} .

For convenience we restrict ourselves in the following to the case i = n (and

V ∩ Un = ∅). The complement Hn = Pn \ Un = {(a0 : . . . : an ) ∈ Pn | an = 0}
is called the hyperplane at infinity, and the points P ∈ V ∩ Hn are the points
of V at infinity.
There is a natural K-isomorphism α from K(V ) (the function field of the
projective variety V ) onto K(Vn ) (the function field of the affine variety Vn =
ϕ−1
n (V ∩ Un )). This isomorphism is defined as follows: Let f = g/h ∈ K(V )
where g, h ∈ Γh (V ) are forms of the same degree and h = 0. Choose homo-
geneous polynomials G, H ∈ K[X0 , . . . , Xn ] which represent g resp. h. Let
G∗ = G(X0 , . . . , Xn−1 , 1) and H∗ = H(X0 , . . . , Xn−1 , 1) ∈ K[X0 , . . . , Xn−1 ].
Their residue classes in Γ (Vn ) = K[X0 , . . . , Xn−1 ]/I(Vn ) are g∗ resp. h∗ . Then
α(f ) = g∗ /h∗ . Under this isomorphism, the local ring of a point P ∈ V ∩ Un
is mapped onto the local ring of ϕ−1 n (P ) ∈ Vn , hence these local rings are
isomorphic.

B.4 The Projective Closure of an Affine Variety

For a polynomial F = F (X0 , . . . , Xn−1 ) ∈ K[X0 , . . . , Xn−1 ] of degree d we
set
F ∗ = Xnd · F (X0 /Xn , . . . , Xn−1 /Xn ) ∈ K[X0 , . . . , Xn ] .
F ∗ is a homogeneous polynomial of degree d in n + 1 variables.
Consider now an affine variety V ∈ An and the corresponding ideal I(V ) ⊆
K[X0 , . . . , Xn−1 ]. Define the projective variety V̄ ⊆ Pn as follows:

V̄ = {P ∈ Pn | F ∗ (P ) = 0 for all F ∈ I(V )} .

This variety V̄ is called the projective closure of V . One can recover V from
V̄ by the process described in B.3, namely

V = ϕ−1
n (V̄ ∩ Un ) = (V̄ )n .

Consequently the function fields of V and V̄ are naturally isomorphic, and V

and V̄ have the same dimension.
 Appendix B. Algebraic Curves and Function Fields 339

B.5 Rational Maps and Morphisms

Let V ⊆ Pm and W ⊆ Pn be projective varieties. Suppose that F0 , . . . , Fn ∈

K[X0 , . . . , Xm ] are homogeneous polynomials with the following properties:
(a) F0 , . . . , Fn have the same degree;
(b) not all Fi are in I(V );
(c) for all H ∈ I(W ) holds H(F0 , . . . , Fn ) ∈ I(V ).
Let Q ∈ V and assume that Fi (Q) = 0 for at least one i ∈ {0, . . . , n} (by
(b) such a point exists). Then the point (F0 (Q) : . . . : Fn (Q)) ∈ Pn lies in
W , by (c). Let (G0 , . . . , Gn ) be another n-tuple of homogeneous polynomials
satisfying (a), (b) and (c). We say that (F0 , . . . , Fn ) and (G0 , . . . , Gn ) are
equivalent if
(d) Fi Gj ≡ Fj Gi mod I(V ) for 0 ≤ i, j ≤ n.
The equivalence class of (F0 , . . . , Fn ) with respect to this equivalence relation
is denoted by
φ = (F0 : . . . : Fn ) ,
and φ is called a rational map from V to W .
A rational map φ = (F0 : . . . : Fn ) is regular (or defined) at the point
P ∈ V if there exist homogeneous polynomials G0 , . . . , Gn ∈ K[X0 , . . . , Xm ]
such that φ = (G0 : . . . : Gn ) and Gi (P ) = 0 for at least one i. Then we set

φ(P ) = (G0 (P ) : . . . : Gn (P )) ∈ W ,

which is well-defined by (a) and (d).

Two varieties V1 , V2 are birationally equivalent if there are rational maps
φ1 : V1 → V2 and φ2 : V2 → V1 such that φ1 ◦ φ2 and φ2 ◦ φ1 are the identity
maps on V2 and V1 , respectively. V1 and V2 are birationally equivalent if and
only if their function fields K(V1 ) and K(V2 ) are K-isomorphic.
A rational map φ : V → W which is regular at all points P ∈ V is called
a morphism. It is called an isomorphism if there is a morphism ψ : W → V
such that φ ◦ ψ and ψ ◦ φ are the identity maps on W and V , respectively.
In this case V and W are said to be isomorphic. Clearly, isomorphy implies
birational equivalence, but the converse is not true in general.

B.6 Algebraic Curves

A projective (affine) algebraic curve V is a projective (affine) variety of di-
mension one. This means that the field K(V ) of rational functions on V is an
algebraic function field of one variable.
A point P ∈ V is non-singular (or simple) if the local ring OP (V ) is a
discrete valuation ring (i.e., OP (V ) is a principal ideal domain with exactly
340 Appendix B. Algebraic Curves and Function Fields

one maximal ideal = {0}). There exist only finitely many singular points on
a curve. The curve V is called non-singular (or smooth) if all points P ∈ V
are non-singular.
A plane affine curve is an affine curve V ⊆ A2 . Its ideal I(V ) ⊆ K[X0 , X1 ]
is generated by an irreducible polynomial G ∈ K[X0 , X1 ] (which is unique
up to a constant factor). Conversely, given an irreducible polynomial G ∈
K[X0 , X1 ], the set V = {P ∈ A2 | G(P ) = 0} is a plane affine curve, and G
generates the corresponding ideal I(V ). A point P ∈ V is non-singular if and
only if
GX0 (P ) = 0 or GX1 (P ) = 0 (or both) ,
where GXi ∈ K[X0 , X1 ] is the partial derivative of G with respect to Xi
(Jacobi-Criterion).
Accordingly, the ideal of a plane projective curve V ⊆ P2 is generated by
an irreducible homogeneous polynomial H ∈ K[X0 , X1 , X2 ]. A point P ∈ V
is non-singular if and only if HXi (P ) = 0 for at least one i ∈ {0, 1, 2}.
If V = {P ∈ A2 | G(P ) = 0} is a plane affine curve (with an irreducible
polynomial G ∈ K[X0 , X1 ] of degree d), the projective closure V̄ ⊆ P2 is the
set of zeros of the homogeneous polynomial G∗ = X2d · G(X0 /X2 , X1 /X2 ).

B.7 Maps Between Curves

We consider rational maps φ : V → W where V and W are projective curves.
The following hold:
(a) φ is defined at all non-singular points P ∈ V . Hence, if V is a non-
singular curve, then φ is a morphism.
(b) If V is non-singular and φ is non-constant then φ is surjective.

B.8 The Non-Singular Model of a Curve

Let V be a projective curve. Then there exists a non-singular projective curve

V  and a birational morphism φ : V  → V . The pair (V  , φ ) is unique in
the following sense: given another non-singular curve V  and a birational
morphism φ : V  → V , there exists a unique isomorphism φ : V  → V  such
that φ = φ ◦ φ. Hence V  (more precisely: the pair (V  , φ )) is called the
non-singular model of V .
If φ : V  → V is the non-singular model of V and P ∈ V is non-singular,
there exists a unique P  ∈ V  with φ (P  ) = P ; for a singular point P ∈ V
the number of P  ∈ V  with φ (P  ) = P is finite (it may be one).
 Appendix B. Algebraic Curves and Function Fields 341

B.9 The Curve Associated with an Algebraic Function

Field
Starting from an algebraic function field of one variable F/K, there exists a
non-singular projective curve V (unique up to isomorphism) whose function
field K(V ) is (K-isomorphic to) F . One can construct V as follows: choose
x, y ∈ F such that F = K(x, y) (this is possible by Proposition 3.10.2).
Let G(X, Y ) ∈ K[X, Y ] be the irreducible polynomial with G(x, y) = 0. Let
W = {P ∈ A2 | G(P ) = 0} and W̄ ⊆ P2 be the projective closure of W . Let
V be the non-singular model of W̄ ; then K(V )  F .

B.10 Non-Singular Curves and Algebraic Function

Fields
Let V be a non-singular projective curve and let F = K(V ) be its function
field. There is a 1-1 correspondence between the points P ∈ V and the places
of F/K, given by
P −→ MP (V ) ,
the maximal ideal of the local ring OP (V ). This correspondence makes it
possible to translate definitions and results from algebraic function fields to
algebraic curves (and vice versa). We give some examples:
– The genus of the curve V is the genus of the function field K(V ).

– A divisor of V is a formal sum D = P ∈V nP P where nP ∈ ZZ and
almost all nP = 0. The degree of D is deg D = P ∈V nP . The divisors of V
form an additive group Div(V ), the divisor group of V .
– The order of a rational function at a point P ∈ V is defined to be vP (f ),
where vP is the discrete valuation of K(V ) corresponding to the valuation
ring OP (V ).
 – The principal divisor (f ) of a rational function 0 = f ∈ K(V ) is (f ) =
P ∈V vP (f )P . The degree of a principal divisor is 0.
– The principal divisors form a subgroup Princ(V ) of the divisor group
Div(V ). The factor group Jac(V ) = Div0 (V )/Princ(V ), where Div0 (V ) is the
group of divisors of degree 0, is called the Jacobian of V .
– For D ∈ Div(V ) the space L (D) is defined as in the function field case.
It is a finite-dimensional vector space over K, its dimension is given by the
Riemann-Roch Theorem.

B.11 Varieties over a Non-Algebraically Closed Field

Thus far it was assumed that K is an algebraically closed field. Now we drop
this assumption and suppose only that K is a perfect field. Let K̄ ⊇ K be the
algebraic closure of K.
342 Appendix B. Algebraic Curves and Function Fields

An affine variety V ⊆ An (K̄) is said to be defined over K if its ideal I(V ) ⊆

K̄[X1 , . . . , Xn ] can be generated by polynomials F1 , . . . , Fr ∈ K[X1 , . . . , Xn ].
If V is defined over K, the set

V (K) = V ∩ An (K) = {P = (a1 , . . . , an ) ∈ V | all ai ∈ K}

is called the set of K-rational points of V .

Similarly, a projective variety V ⊆ Pn (K̄) is defined over K if I(V ) is
generated by homogeneous polynomials F1 , . . . , Fr ∈ K[X0 , . . . , Xn ]. A point
P ∈ V is called K-rational if there exist homogeneous coordinates a0 , . . . , an
of P which are in K, and we set

V (K) = {P ∈ V | P is K-rational} .

Let V ⊆ An (K̄) be an affine variety defined over K. Define the ideal

I(V /K) = I(V ) ∩ K[X1 , . . . , Xn ]

and the residue class ring

Γ (V /K) = K[X1 , . . . , Xn ]/I(V /K) .

The quotient field

K(V ) = Quot(Γ (V /K)) ⊆ K̄(V )

is the field of K-rational functions of V . The field extension K(V )/K is finetely
generated, its transcendence degree is the dimension of V . In the same manner,
the field of K-rational functions of a projective variety can be defined.
Consider two varieties V ⊆ Pm (K̄) and W ⊆ Pn (K̄). A rational map φ :
V → W is defined over K if there exist homogeneous polynomials F0 , . . . , Fn ∈
K[X0 , . . . , Xm ] satisfying the conditions (a), (b) and (c) of B.5, such that
φ = (F0 : . . . : Fn ).
Another way to describe K-rational points, K-rational functions etc. on a
variety which is defined over K is the following: Let GK̄/K be the Galois group
of K̄/K. The action of GK̄/K on K̄ extends naturally to an action on the sets
An (K̄), Pn (K̄), K̄[X1 , . . . , Xn ], V, Γ (V ), K̄(V ) etc. For instance, consider a
projective variety V ⊆ Pn (K̄) (defined over K), a point P = (a0 : . . . : an ) ∈
V and an automorphism σ ∈ GK̄/K ; then P σ = (aσ0 : . . . : aσn ). It is easily seen
that
V (K) = {P ∈ V | P σ = P for all σ ∈ GK̄/K } ,
K(V ) = {f ∈ K̄(V ) | f σ = f for all σ ∈ GK̄/K } ,
and so on.
 Appendix B. Algebraic Curves and Function Fields 343

B.12 Curves over a Non-Algebraically Closed Field

Consider a projective curve V ⊆ Pn (K̄) which is defined over K (where K is
perfect and K̄ is the algebraic closure of K as in B.11). Then the field K(V )
of K-rational functions on V is an algebraic function field of one variable over
K, and K̄(V ) is the constant field extension of K(V ) with K̄.

A divisor D = P ∈V nP P ∈ Div(V ) is defined over K if Dσ = D for
all σ ∈ GK̄/K (this means that nP σ = nP for all P ∈ V ). The divisors of V
defined over K form a subgroup Div(V /K) ⊆ Div(V ). For D ∈ Div(V /K) the
space LK (D) is given by

LK (D) = K(V ) ∩ L (D) .

It is a finite-dimensional K-vector space, and its dimension (over K) equals

the dimension of L (D) (over K̄), by Theorem 3.6.3(d).
A divisor Q ∈ Div(V /K) with Q > 0 is called a prime divisor of V /K
if Q cannot be written as Q = Q1 + Q2 with effective divisors Q1 , Q2 ∈
Div(V /K). It is easily seen that the divisor group Div(V /K) is the free abelian
group generated by the prime divisors. Prime divisors of V /K correspond to
the places of the function field K(V )/K; under this correspondence, prime
divisors of degree one (i.e., K-rational points) of V correspond to the places
of K(V )/K of degree one.

B.13 An Example
Let K be a perfect field of characteristic p ≥ 0 and let

G(X, Y ) = aX n + bY n + c with a, b, c ∈ K \ {0} , n ≥ 1 and p  n .

(This is Example 6.3.4.) The polynomial G(X, Y ) is irreducible (which follows

easily from Eisenstein’s Criterion, cf. Proposition 3.1.15). The affine curve
V = {P ∈ A2 (K̄) | G(P ) = 0} is non-singular since for P = (α, β) ∈ V ,

GX (α, β) = naαn−1 = 0 or GY (α, β) = nbβ n−1 = 0 .

Let G∗ (X, Y, Z) = aX n + bY n + cZ n ; then the projective closure of V is

the curve
V̄ = {(α : β : γ) ∈ P2 (K̄) | G∗ (α, β, γ) = 0} .
We consider the points P ∈ V̄ at infinity; i.e., P = (α : β : 0) with (α, β) =
(0, 0) and G∗ (α, β, 0) = 0. From the equation 0 = G∗ (α, β, 0) = aαn + bβ n
follows that β = 0, so we can set β = 1; i.e., P = (α : 1 : 0). The equation
aαn + b = 0 has n distinct roots α ∈ K, so there are n distinct points of V̄ at
infinity. All of them are non-singular since G∗Y (α, 1, 0) = nb = 0.
In the special case K = IFq2 and G(X, Y ) = X q+1 + Y q+1 − 1 (the
Hermitian curve, cf. Example 6.3.6) we want to determine the K-rational
344 Appendix B. Algebraic Curves and Function Fields

points P = (α : β : γ) ∈ V̄ (K). First let γ = 0; i.e., P = (α : β : 1).

For all α ∈ K with αq+1 = 1 there are q + 1 distinct elements β ∈ K with
G∗ (α, β, 1) = 0. If αq+1 = 1, β = 0 is the only root of G∗ (α, β, 1) = 0.
Finally, if γ = 0, there are q + 1 points P = (α : 1 : 0) ∈ V̄ (K). We have
thus constructed all K-rational points on the Hermitian curve over IFq2 . Their
number is |(V̄ (IFq2 )| = q 3 + 1, in accordance with Lemma 6.4.4.
List of Notations

K̃ field of constants of F/K, 1

IPF set of places of F/K, 4
OP valuation ring of the place P , 4
vP valuation corresponding to the place P , 5
FP residue class field of the place P , 6
x(P ) residue class of x ∈ OP in FP , 6
deg P degree of the place P , 6
Pp(x) the place of K(x) corresponding to an irreducible poly-
nomial p(x), 9
P∞ the infinite place of K(x), 9
P1 (K) projective line, 11
Div(F ) divisor group of F , 15
supp D support of D, 15
vQ (D) 16
deg D degree of the divisor D, 16
(x)0 , (x)∞ , (x) zero, pole, principal divisor of x, 16
Princ(F ) group of principal divisors of F , 17
Cl(F ) divisor class group of F , 17
[D] divisor class of the divisor D, 17
D ∼ D equivalence of divisors, 17
L (A) Riemann-Roch space of the divisor A, 17
(A) dimension of the Riemann-Roch space L (A), 19
g the genus of F , 22
i(A) index of specialty of A, 24
AF adele space of F , 24
AF (A) 24
ΩF space of Weil differentials of F , 27
ΩF (A) 27
(ω) divisor of the Weil differential ω, 29
vP (ω) 29
ιP (x) 38
ωP local component of a Weil differential, 38
d(a, b) Hamming distance, 45
346 List of Notations

wt(a) weight of a vector a ∈ IFn q , 45

d(C) minimum distance of the code C, 46
[n, k, d] code code with parameters n, k, d, 46
a, b the canonical inner product on IFn q , 46
C⊥ the dual code of C, 46
CL (D, G) the algebraic geometry code associated to the divisors
D and G, 48
evD (x) 49
CΩ (D, G) differential code associated with divisors D and G, 51
GRSk (α, v) Generalized Reed-Solomon code, 57
C|IFq subfield subcode, 59
P  |P the place P  lies over P , 69
e(P  |P ) ramification index of P  |P , 71
f (P  |P ) relative degree of P  |P , 71
ConF /F (P ) conorm of P in the extension F  /F , 72
OS holomorphy ring corresponding to the set S ⊆ IPF , 77
icF (R) integral closure of R in F , 78
OP integral closure of OP in F  , 85
CP complementary module over OP , 91
d(P  |P ) different exponent of P  |P , 92
Diff(F  /F ) different of F  /F , 92
GZ (P  |P ) decomposition group of P  |P , 130
GT (P  |P ) inertia group of P  |P , 130
Gi (P  |P ) i-th ramification group of P  |P , 133
δx derivation with respect to x, 158
DerF module of derivations of F/K, 158
u dx differential of F/K, 159
ΔF , (ΔF , d) differential module of F/K, 159
(T̂ , v̂) completion of the valued field (T, v), 162
F̂P P -adic completion of F , 164
resP,t (z) residue of z with respect to P and t, 167
resP (ω) residue of ω at the place P , 169
δ(x) the Weil differential associated with x, 170
Div0 (F ) group of divisors of degree 0, 186
Cl0 (F ) group of divisor classes of degree 0, 186
h, hF class number of F/IFq , 186
∂ 187
Z(t), ZF (t) Zeta function of F/IFq , 188
Fr = F IFqr the constant field extension of F/IFq of degree r, 190
L(t), LF (t) the L-polynomial of F/IFq , 193
Nr number of rational places of Fr , 195
Br number of places of degree r, 206
Nq (g) 243
A(q) Ihara’s constant, 243
λm (t) 244
fm (t) 244
F = (F0 , F1 , F2 , . . .) a tower of function fields, 246
ν(F /F0 ) splitting rate of the tower F /F0 , 247
γ(F /F0 ) genus of the tower F /F0 , 247
 List of Notations 347

λ(F ) limit of the tower F , 247

Split(F /F0 ) splitting locus of F over F0 , 249
Ram(F /F0 ) ramification locus of F over F0 , 249
(z = α) the zero of z − α in K(z), 254
(z = ∞) the pole of z in K(z), 254
Aut(C) automorphism group of the code C, 290
AutD,G (F/IFq ) 291
R, R(C) information rate of the code C, 297
δ, δ(C) relative minimum distance of the code C, 297
αq , αq (δ) 297
Hq (x) q-ary entropy function, 298
[b, f ] syndrome, 304
C|IFq subfield subcode, 312
Tr(C) trace code of C, 312
K̄ algebraic closure of K, 328
char K characteristic of K, 328
Aut(L/K) automorphism group of L/K, 330
Gal(L/K) Galois group of L/K, 330
NL/K norm map of L/K, 332
TrL/K trace map of L/K, 332
References

1. E. Artin, Algebraic numbers and algebraic functions, Gordon and Breach, New
York, 1967.
2. A.Bassa, A.Garcia and H.Stichtenoth, A new tower over cubic finite fields,
Moscow Math. J., to appear.
3. E. Berlekamp, Algebraic coding theory, McGraw-Hill, New York, 1968.
4. J. Bezerra, A. Garcia and H. Stichtenoth, An explicit tower of function fields
over cubic finite fields and Zink’s lower bound, J. Reine Angew. Math. 589,
2005, pp. 159-199.
5. E. Bombieri, Counting points on curves over finite fields, Sem. Bourbaki, No.
430, Lecture Notes Math. 383, Springer-Verlag, Berlin-Heidelberg-New York,
1974, pp. 234-241.
6. C. Chevalley, Introduction to the theory of algebraic functions of one variable,
AMS Math. Surveys No. 6, 1951.
7. M. Deuring, Lectures on the theory of algebraic functions of one variable, Lec-
ture Notes in Math. 314, Springer-Verlag, Berlin-Heidelberg-New York, 1973.
8. V. G. Drinfeld and S. G. Vladut, The number of points of an algebraic curve,
Func. Anal. 17, 1983, pp. 53-54.
9. M. Eichler, Introduction to the theory of algebraic numbers and functions, Aca-
demic Press, New York, 1966.
10. H. M. Farkas and I. Kra, Riemann surfaces, 2nd edition, Graduate Texts in
Math. 71, Springer-Verlag, New York, 1991.
11. W. Fulton, Algebraic curves, Benjamin, New York, 1969.
12. A. Garcia and H. Stichtenoth, On the asymptotic behavior of some towers of
functions fields over finite fields, J. Number Theory 61, 1996, pp. 248-273.
13. A. Garcia and H. Stichtenoth (eds.), Topics in geometry, coding theory and
cryptography, Algebr. Appl. 6, Springer-Verlag, Dordrecht, 2007.
14. D. M. Goldschmidt, Algebraic functions and projective curves, Graduate Texts
in Math. 215, Springer-Verlag, New York, 2002.
15. V. D. Goppa, Codes on algebraic curves, Soviet Math. Dokl. 24, No. 1, 1981,
pp. 170-172.
16. V. D. Goppa, Geometry and codes, Kluwer Academic Publ., Dordrecht, 1988.
17. V. Guruswami and M. Sudan, Improved decoding of Reed-Solomon and alge-
braic geometry codes, IEEE Trans. Inform. Th. 45, 1999, pp. 1757-1768.
350 References

18. R. Hartshorne, Algebraic geometry, Graduate Texts in Math. 52, Springer-

Verlag, New York-Heidelberg-Berlin, 1977.
19. H. Hasse, Theorie der relativ-zyklischen algebraischen Funktionenkörper, ins-
besondere bei endlichem Konstantenkörper, J. Reine Angew. Math. 172, 1934,
pp. 37-54.
20. K. Hensel and G. Landsberg, Theorie der algebraischen Funktionen einer Vari-
ablen, Leipzig, 1902. Reprinted by Chelsea Publ. Comp., New York, 1965.
21. J. W. P. Hirschfeld, G. Korchmaros and F. Torres, Algebraic curves over a
finite field, Princeton Ser. in Applied Math., Princeton Univ. Press, Princeton
and Oxford, 2008.
22. Y. Ihara, Some remarks on the number of rational points of algebraic curves
over finite fields, J. Fac. Sci. Univ. Tokyo Sect. IA Math. 28, 1981, pp. 721-724.
23. S. Lang, Algebra, 3rd edition, Graduate Texts in Math. 211, Springer-Verlag,
New York, 2002.
24. R. Lidl and H. Niederreiter, Finite fields, 2nd edition, Cambridge Univ. Press,
Cambridge, 1997.
25. J. H. van Lint, Introduction to coding theory, 2nd edition, Graduate Texts in
Math. 86, Springer-Verlag, Berlin-Heidelberg-New York, 1992.
26. J. H. van Lint and G. van der Geer, Introduction to coding theory and algebraic
geometry, DMV Seminar 12, Birkhäuser-Verlag, Basel-Boston-Berlin, 1988.
27. W. Lütkebohmert, Codierungstheorie, Vieweg-Verlag, Braunschweig, 2003.
28. F. J. MacWilliams and N. J. A. Sloane, The theory of error-correcting codes,
North-Holland, Amsterdam, 1977.
29. Y. I. Manin, What is the maximum number of points on a curve over F2 ?, J.
Fac. Sci. Univ. Tokyo Sect. IA Math. 28, 1981, pp. 715-720.
30. C. Moreno, Algebraic curves over finite fields, Cambridge Tracts in Math. 97,
Cambridge Univ. Press, Cambridge, 1991.
31. H. Niederreiter and C. P. Xing, Rational points on curves over finite
fields, London Math. Soc. Lecture Notes Ser. 285, Cambridge Univ. Press,
Cambridge, 2001.
32. V. Pless and W. C. Huffmann (eds.), Handbook of coding theory, Vol. I and II,
Elsevier, Amsterdam, 1998.
33. O. Pretzel, Codes and algebraic curves, Oxford Lecture Ser. in Math. and its
Applications 8, Oxford Univ. Press, Oxford, 1998.
34. H.-G. Rück and H. Stichtenoth, A characterization of Hermitian function fields
over finite fields, J. Reine Angew. Math. 457, 1994, pp. 185-188.
35. J.-P. Serre, Local fields, Springer-Verlag, New York-Berlin, 1979.
36. J.-P. Serre, Sur le nombre des points rationnels d’une courbe algébrique sur un
corps fini, C. R. Acad. Sci. Paris 296, 1983, pp. 397-402.
37. I. R. Shafarevic, Basic algebraic geometry, Grundlehren der math. Wissensch.
213, Springer-Verlag, Berlin-Heidelberg-New York, 1977.
38. J. H. Silverman, The arithmetic of elliptic curves, Graduate Texts in Math.
106, Springer-Verlag, Berlin-Heidelberg-New York, 1986.
39. A. N. Skorobogatov and S. G. Vladut, On the decoding of algebraic-geometric
codes, IEEE Trans. Inform. Th. 36, 1990, pp. 1461-1463.
40. S. A. Stepanov, Arithmetic of algebraic curves, Monographs in Contemp.
Math., Plenum Publ. Corp., New York, 1994.
41. H. Stichtenoth, Über die Automorphismengruppe eines algebraischen Funktio-
nenkörpers von Primzahlcharakteristik, Arch. Math. 24, 1973, pp. 527-544 and
615-631.
 References 351

42. M. A. Tsfasman and S. G. Vladut, Algebraic-geometric codes, Kluwer Academic

Publ., Dordrecht-Boston-London, 1991.
43. M. A. Tsfasman, S. G. Vladut and D. Nogin, Algebraic-geometric codes: basic
notions, AMS Math. Surveys and Monographs 139, 2007.
44. M. A. Tsfasman, S. G. Vladut and T. Zink, Modular curves, Shimura curves,
and Goppa codes, better than the Varshamov-Gilbert bound, Math. Nachr. 109,
1982, pp. 21-28.
45. G. D. Villa Salvador, Topics in the theory of algebraic function fields,
Birkhäuser-Verlag, Boston, 2006.
46. B. L. van der Waerden, Algebra, Teil 1 und 2, 7th edition, Springer-Verlag,
Berlin-Heidelberg-New York, 1966.
47. T. Zink, Degeneration of Shimura surfaces and a problem in coding theory,
in Fundamentals of Computation Theory (L. Budach, ed.), Lecture Notes in
Computer Science 199, Springer-Verlag, Berlin, 1985, pp. 503-511.

A more comprehensive list of references to recent work on function fields and

algebraic geometry codes can be found in [13],[21],[31],[42],[43] and [45].
Index

Abhyankar’s Lemma, 137 decoding algorithm, 303

adele, 24 designed distance of a BCH code, 59
principal adele, 24 designed distance of an AG code, 50,
algebraic function field, 1 52
algebraic extension of function fields, dimension of a code, 46
68 dual code, 46
constant field, 1 error locator function, 305
constant field extension, 68 Generalized Reed-Solomon code, 57
elliptic function field, 217
generator matrix, 46
hyperelliptic function field, 224
Gilbert-Varshamov Bound, 298
rational function field, 1, 8
Goppa Bound, 62
Approximation Theorem
Strong Approximation Theorem, 33 Goppa code, 61, 315, 323
Weak Approximation Theorem, 12 Hamming Bound, 298
Artin-Schreier extension, 127 Hermitian code, 293
MDS code, 47
Bombieri’s proof of the Hasse-Weil minimum distance, 46
Theorem, 198 parity check matrix, 47
Plotkin Bound, 298
canonical class, 30 rational AG code, 56
Castelnouvo’s Inequality, 146 Reed-Solomon code, 48
Cauchy sequence, 162 self-dual code, 46, 290
chain rule, 158 self-orthogonal code, 46, 290
class number, 186
Skorobogatov-Vladut algorithm, 308
Clifford’s Theorem, 36
subfield subcode, 59, 312
code, 45
trace code, 312
AG code, 48
algebraic geometry code, 48, 51 Tsfasman-Vladut-Zink Bound, 300
automorphism group of a code, 290 complementary module, 91
Bassalygo-Elias Bound, 298 completion, 162
BCH Bound, 62 P-adic completion, 164
BCH code, 59 conorm, 72
Carlitz-Uchiyama Bound, 323 constant field extension, 112
cyclic code, 316 cotrace of a Weil differential, 93
354 Index

decomposition group, decomposition evaluation map, 49

field, 130 Explicit Formulas, 212
Dedekind’s Different Theorem, 100
degenerate, 319 F. K. Schmidt’s Theorem, 191
Delsarte’s Theorem, 312 Functional equation
derivation, 155 of the L-polynomial, 193
derivation with respect to x, 158 of the Zeta function, 192
module of derivations, 158 Fundamental Equality, 74
different, 92, 99 Galois extension, 120, 130
transitivity of the different, 98 gap number, 34
different exponent, 92 genus, 22
differential, 27
cotrace of a Weil differential, 93 Hamming distance, 45
differential module, 159 Hasse-Weil Bound, 198
divisor of a Weil differential, 29 Hasse-Weil Theorem, 197
exact differential, 161 Hermitian function field, 230
local component of a Weil differential, Hilbert’s Different Formula, 136
38 holomorphy ring, 77
regular differential, 29 Hurwitz Genus Formula, 99
residue of a differential, 167, 169
Weil differential, 27 Ihara
Weil differential associated with x, Ihara’s constant, 243, 265
170 Ihara’s Theorem, 210
divisor, 15 index of specialty, 24
canonical divisor, 29 inertia group, inertia field, 130
degree of a divisor, 16 Inseparable extension, 143
dimension of a divisor, 19 integral, 78
divisor class group, 17, 186 integral basis, 85
divisor group, 15 integral closure, 78
divisor of a Weil differential, 29 integrally closed, 78
equivalence of divisors, 17 local integral basis, 85
group of divisors of degree zero, 186 Klein Quartic, 230
index of specialty, 24 Kummer extension, 122
non-special divisor, 35 Kummer’s Theorem, 86
pole divisor, 16
prime divisor, 15 L-polynomial of a function field, 193
principal divisor, 16 Lüroth’s Theorem, 107
special divisor, 35 local parameter, 4
zero divisor, 16
Drinfeld-Vladut Bound, 244 maximal function field, 210
dual basis, 82 order
Duality Theorem, 30 pole order, 7
zero order, 7
Eisenstein’s Irreducibility Criterion, 76
elementary abelian extension, 130 place, 4
elliptic function, 221 completely splitting, 75
elliptic function field, 217 degree of a place, 6
entropy function, 298 extension of a place, 69
Euler Product, 189 infinite place, 9
 Index 355

ramified place, 71 Serre Bound, 209

rational place, 6 Singleton Bound, 47
relative degree of a place, 71
restriction of a place, 69
tower of function fields, 246
totally ramified, 75
asymptotically bad, 248
unramified place, 71
asymptotically good, 248
pole number, 34
asymptotically optimal, 248
polynomial
absolutely irreducible polynomial, basic function field, 254
120 Bassa-Garcia-Stichtenoth tower, 275
additive polynomial, 129 Garcia-Stichtenoth tower, 265
power series expansion, 164 genus of a tower, 247
prime element, 4 limit of a tower, 247
ramification locus, 249
ramification index, 71 recursively defined tower, 251
ramified, 105 splitting locus, 249
higher ramification groups, 133 splitting rate, 247
tamely ramified, 105 Triangle Inequality, 4
totally ramified, 105 Strict Triangle Inequality, 5
weakly ramified, 266, 267
wildly ramified, 105
residue class field, 6 valuation
residue class map, 6 discrete valuation, 4
Residue Theorem, 38, 171 discrete valuation ring, 3
Riemann valuation ring, 2
Riemann Hypothesis for Function
Fields, 197 Weierstrass ℘-function, 221
Riemann’s Inequality, 148 Weierstrass Gap Theorem, 34
Riemann’s Theorem, 23 Weierstrass point, 34
Riemann-Roch weight, 45
Riemann-Roch space, 17
Riemann-Roch Theorem, 30
Zeta function, 185, 188
separably generated, 144 Functional Equation, 192
separating element, 144 Zink Bound, 275
Graduate Texts in Mathematics
(continued from page ii)

66 WATERHOUSE. Introduction to Affine 98 BRÖCKER/TOM DIECK. Representations of

Group Schemes. Compact Lie Groups.
67 SERRE. Local Fields. 99 GROVE/BENSON. Finite Reflection Groups.
68 WEIDMANN. Linear Operators in Hilbert 2nd ed.
Spaces. 100 BERG/CHRISTENSEN/RESSEL. Harmonic
69 LANG. Cyclotomic Fields II. Analysis on Semigroups: Theory of
70 MASSEY. Singular Homology Theory. Positive Definite and Related Functions.
71 FARKAS/KRA. Riemann Surfaces. 2nd ed. 101 EDWARDS. Galois Theory.
72 STILLWELL. Classical Topology and 102 VARADARAJAN. Lie Groups, Lie Algebras
Combinatorial Group Theory. 2nd ed. and Their Representations.
73 HUNGERFORD. Algebra. 103 LANG. Complex Analysis. 3rd ed.
74 DAVENPORT. Multiplicative Number 104 DUBROVIN/FOMENKO/NOVIKOV. Modern
Theory. 3rd ed. Geometry—Methods and Applications.
75 HOCHSCHILD. Basic Theory of Algebraic Part II.
Groups and Lie Algebras. 105 LANG. SL2(R).
76 IITAKA. Algebraic Geometry. 106 SILVERMAN. The Arithmetic of Elliptic
77 HECKE. Lectures on the Theory of Curves.
Algebraic Numbers. 107 OLVER. Applications of Lie Groups to
78 BURRIS/SANKAPPANAVAR. A Course in Differential Equations. 2nd ed.
Universal Algebra. 108 RANGE. Holomorphic Functions and
79 WALTERS. An Introduction to Ergodic Integral Representations in Several
Theory. Complex Variables.
80 ROBINSON. A Course in the Theory of 109 LEHTO. Univalent Functions and
Groups. 2nd ed. Teichmüller Spaces.
81 FORSTER. Lectures on Riemann Surfaces. 110 LANG. Algebraic Number Theory.
82 BOTT/TU. Differential Forms in Algebraic 111 HUSEMÖLLER. Elliptic Curves. 2nd ed.
Topology. 112 LANG. Elliptic Functions.
83 WASHINGTON. Introduction to Cyclotomic 113 KARATZAS/SHREVE. Brownian Motion and
Fields. 2nd ed. Stochastic Calculus. 2nd ed.
84 IRELAND/ROSEN. A Classical Introduction 114 KOBLITZ. A Course in Number Theory and
to Modern Number Theory. 2nd ed. Cryptography. 2nd ed.
85 EDWARDS. Fourier Series. Vol. II. 2nd ed. 115 BERGER/GOSTIAUX. Differential Geometry:
86 VAN LINT. Introduction to Coding Theory. Manifolds, Curves, and Surfaces.
2nd ed. 116 KELLEY/SRINIVASAN. Measure and Integral.
87 BROWN. Cohomology of Groups. Vol. I.
88 PIERCE. Associative Algebras. 117 J.-P. SERRE. Algebraic Groups and Class
89 LANG. Introduction to Algebraic and Fields.
Abelian Functions. 2nd ed. 118 PEDERSEN. Analysis Now.
90 BRØNDSTED. An Introduction to Convex 119 ROTMAN. An Introduction to Algebraic
Polytopes. Topology.
91 BEARDON. On the Geometry of Discrete 120 ZIEMER. Weakly Differentiable Functions:
Groups. Sobolev Spaces and Functions of Bounded
92 DIESTEL. Sequences and Series in Banach Variation.
Spaces. 121 LANG. Cyclotomic Fields I and II.
93 DUBROVIN/FOMENKO/NOVIKOV. Modern Combined 2nd ed.
Geometry—Methods and Applications. 122 REMMERT. Theory of Complex Functions.
Part I. 2nd ed. Readings in Mathematics
94 WARNER. Foundations of Differentiable 123 EBBINGHAUS/HERMES et al. Numbers.
Manifolds and Lie Groups. Readings in Mathematics
95 SHIRYAEV. Probability. 2nd ed. 124 DUBROVIN/FOMENKO/NOVIKOV. Modern
96 CONWAY. A Course in Functional Geometry—Methods and Applications
Analysis. 2nd ed. Part III.
97 KOBLITZ. Introduction to Elliptic Curves 125 BERENSTEIN/GAY. Complex Variables:
and Modular Forms. 2nd ed. An Introduction.
126 BOREL. Linear Algebraic Groups. 2nd ed. 157 MALLIAVIN. Integration and
127 MASSEY. A Basic Course in Algebraic Probability.
Topology. 158 ROMAN. Field Theory.
128 RAUCH. Partial Differential Equations. 159 CONWAY. Functions of One Complex
129 FULTON/HARRIS. Representation Theory: A Variable II.
First Course. 160 LANG. Differential and Riemannian
Readings in Mathematics Manifolds.
130 DODSON/POSTON. Tensor Geometry. 161 BORWEIN/ERDÉLYI. Polynomials and
131 LAM. A First Course in Noncommutative Polynomial Inequalities.
Rings. 2nd ed. 162 ALPERIN/BELL. Groups and
132 BEARDON. Iteration of Rational Functions. Representations.
133 HARRIS. Algebraic Geometry: A First 163 DIXON/MORTIMER. Permutation
Course. Groups.
134 ROMAN. Coding and Information Theory. 164 NATHANSON. Additive Number Theory:
135 ROMAN. Advanced Linear Algebra. The Classical Bases.
2nd ed. 165 NATHANSON. Additive Number Theory:
136 ADKINS/WEINTRAUB. Algebra: An Approach Inverse Problems and the Geometry of
via Module Theory. Sumsets.
137 AXLER/BOURDON/RAMEY. Harmonic 166 SHARPE. Differential Geometry: Cartan’s
Function Theory. 2nd ed. Generalization of Klein’s Erlangen
138 COHEN. A Course in Computational Program.
Algebraic Number Theory. 167 MORANDI. Field and Galois Theory.
139 BREDON. Topology and Geometry. 168 EWALD. Combinatorial Convexity and
140 AUBIN. Optima and Equilibria. An Algebraic Geometry.
Introduction to Nonlinear Analysis. 169 BHATIA. Matrix Analysis.
141 BECKER/WEISPFENNING/KREDEL. Gröbner 170 BREDON. Sheaf Theory. 2nd ed.
Bases. A Computational Approach to 171 PETERSEN. Riemannian Geometry.
Commutative Algebra. 172 REMMERT. Classical Topics in Complex
142 LANG. Real and Functional Analysis. Function Theory.
3rd ed. 173 DIESTEL. Graph Theory. 3rd ed.
143 DOOB. Measure Theory. 174 BRIDGES. Foundations of Real and
144 DENNIS/FARB. Noncommutative Abstract Analysis.
Algebra. 175 LICKORISH. An Introduction to Knot
145 VICK. Homology Theory. An Theory.
Introduction to Algebraic Topology. 176 LEE. Riemannian Manifolds.
2nd ed. 177 NEWMAN. Analytic Number Theory.
146 BRIDGES. Computability: A 178 CLARKE/LEDYAEV/STERN/WOLENSKI.
Mathematical Sketchbook. Nonsmooth Analysis and Control
147 ROSENBERG. Algebraic K-Theory and Theory.
Its Applications. 179 DOUGLAS. Banach Algebra Techniques in
148 ROTMAN. An Introduction to the Operator Theory. 2nd ed.
Theory of Groups. 4th ed. 180 SRIVASTAVA. A Course on Borel Sets.
149 RATCLIFFE. Foundations of 181 KRESS. Numerical Analysis.
Hyperbolic Manifolds. 182 WALTER. Ordinary Differential
150 EISENBUD. Commutative Algebra Equations.
with a View Toward Algebraic 183 MEGGINSON. An Introduction to Banach
Geometry. Space Theory.
151 SILVERMAN. Advanced Topics in 184 BOLLOBAS. Modern Graph Theory.
the Arithmetic of Elliptic Curves. 185 COX/LITTLE/O’SHEA. Using Algebraic
152 ZIEGLER. Lectures on Polytopes. Geometry. 2nd ed.
153 FULTON. Algebraic Topology: A First 186 RAMAKRISHNAN/VALENZA. Fourier
Course. Analysis on Number Fields.
154 BROWN/PEARCY. An Introduction to 187 HARRIS/MORRISON. Moduli of Curves.
Analysis. 188 GOLDBLATT. Lectures on the Hyperreals:
155 KASSEL. Quantum Groups. An Introduction to Nonstandard Analysis.
156 KECHRIS. Classical Descriptive Set 189 LAM. Lectures on Modules and Rings.
Theory.
190 ESMONDE/MURTY. Problems in Algebraic 220 NESTRUEV. Smooth Manifolds and
Number Theory. 2nd ed. Observables.
191 LANG. Fundamentals of Differential 221 GRÜNBAUM. Convex Polytopes.
Geometry. 2nd ed.
192 HIRSCH/LACOMBE. Elements of Functional 222 HALL. Lie Groups, Lie Algebras, and
Analysis. Representations: An Elementary
193 COHEN. Advanced Topics in Introduction.
Computational Number Theory. 223 VRETBLAD. Fourier Analysis and
194 ENGEL/NAGEL. One-Parameter Semigroups Its Applications.
for Linear Evolution Equations. 224 WALSCHAP. Metric Structures in
195 NATHANSON. Elementary Methods in Differential Geometry.
Number Theory. 225 BUMP: Lie Groups.
196 OSBORNE. Basic Homological Algebra. 226 ZHU. Spaces of Holomorphic Functions in
197 EISENBUD/HARRIS. The Geometry of the Unit Ball.
Schemes. 227 MILLER/STURMFELS. Combinatorial
198 ROBERT. A Course in p-adic Analysis. Commutative Algebra.
199 HEDENMALM/KORENBLUM/ZHU. Theory of 228 DIAMOND/SHURMAN. A First Course in
Bergman Spaces. Modular Forms.
200 BAO/CHERN/SHEN. An Introduction to 229 EISENBUD. The Geometry of Syzygies.
Riemann–Finsler Geometry. 230 STROOCK. An Introduction to Markov
201 HINDRY/SILVERMAN. Diophantine Processes.
Geometry: An Introduction. 231 BJÖRNER/BRENTI. Combinatorics of
202 LEE. Introduction to Topological Coxeter Groups.
Manifolds. 232 EVEREST/WARD. An Introduction to
203 SAGAN. The Symmetric Group: Number Theory.
Representations, Combinatorial 233 ALBIAC/KALTON. Topics in Banach Space
Algorithms, and Symmetric Functions. Theory.
204 ESCOFIER. Galois Theory. 234 JORGENSEN. Analysis and Probability.
205 FÉLIX/HALPERIN/THOMAS. Rational 235 SEPANSKI. Compact Lie Groups.
Homotopy Theory. 2nd ed. 236 GARNETT. Bounded Analytic Functions.
206 MURTY. Problems in Analytic Number ´
237 MARTINEZ -AVENDANO˜ /ROSENTHAL. An
Theory. Introduction to Operators on the Hardy-
Readings in Mathematics Hilbert Space.
207 GODSIL/ROYLE. Algebraic Graph Theory. 238 AIGNER. A Course in Enumeration.
208 CHENEY. Analysis for Applied 239 COHEN. Number Theory – Volume I:
Mathematics. Tools and Diophantine Equations.
209 ARVESON. A Short Course on Spectral 240 COHEN. Number Theory – Volume II:
Theory. Analytic and Modern Tools.
210 ROSEN. Number Theory in Function 241 SILVERMAN. The Arithmetic of Dynamical
Fields. Systems.
211 LANG. Algebra. Revised 3rd ed. 242 GRILLET. Abstract Algebra. 2nd ed.
212 MATOUŠEK. Lectures on Discrete 243 GEOGHEGAN. Topological Methods in
Geometry. Group Theory.
213 FRITZSCHE/GRAUERT. From Holomorphic 244 BONDY/MURTY. Graph Theory.
245 GILMAN/KRA/RODRIGUEZ. Complex Analysis.
Functions to Complex Manifolds. 246 KANIUTH. A Course in Commutative Banach
214 JOST. Partial Differential Equations. Algebras.
2nd ed. 247 KASSEL/TURAEV. Braid Groups.
215 GOLDSCHMIDT. Algebraic Functions and 248 ABRAMENKO/BROWN. Buildings: Theory and
Projective Curves. Applications.
216 D. SERRE. Matrices: Theory and 249 GRAFAKOS. Classical Fourier Analysis.
250 GRAFAKOS. Modern Fourier Analysis.
Applications. 251 WILSON. The Final Simple Groups.
217 MARKER. Model Theory: An 252 GRUBB. Distributions and Operators.
Introduction. 253 MACCLUER. Elementary Functional Analysis.
218 LEE. Introduction to Smooth Manifolds.
254 STICHTENOTH. Algebraic Function Fields
219 MACLACHLAN/REID. The Arithmetic of and Codes, 2nd ed.
Hyperbolic 3-Manifolds.