Knebusch M Scheiderer C Real Algebra A First Course

Universitext
Manfred Knebusch
Claus Scheiderer
Real Algebra
A First Course
Translated and with Contributions by
Thomas Unger
Universitext
Series Editors
Carles Casacuberta, Universitat de Barcelona, Barcelona, Spain
John Greenlees, University of Warwick, Coventry, UK
Angus MacIntyre, Queen Mary University of London, London, UK
Claude Sabbah, École Polytechnique, CNRS, Université Paris-Saclay, Palaiseau,
France
Endre Süli, University of Oxford, Oxford, UK
Universitext is a series of textbooks that presents material from a wide variety of
mathematical disciplines at master’s level and beyond. The books, often well class-
tested by their author, may have an informal, personal even experimental approach
to their subject matter. Some of the most successful and established books in the
series have evolved through several editions, always following the evolution of
teaching curricula, into very polished texts.
Thus as research topics trickle down into graduate-level teaching, first textbooks
written for new, cutting-edge courses may make their way into Universitext.
Manfred Knebusch • Claus Scheiderer
Real Algebra
A First Course
With Contributions by Thomas Unger

Manfred Knebusch Claus Scheiderer
Fakultät für Mathematik FB Mathematik
University of Regensburg University of Konstanz
Regensburg, Germany Konstanz, Germany
Translated by
Thomas Unger
School of Mathematics and Statistics
University College Dublin
Dublin, Ireland
ISSN 0172-5939 ISSN 2191-6675 (electronic)

Universitext
ISBN 978-3-031-09799-7 ISBN 978-3-031-09800-0 (eBook)
https://doi.org/10.1007/978-3-031-09800-0
Translation from the German language edition: “Einführung in die reelle Algebra” by Manfred Knebusch
et al., © Friedr. Vieweg & Sohn Verlagsgesellschaft mbH 1989. Published by Friedr. Vieweg & Sohn,
Braunschweig/Wiesbaden. All Rights Reserved.
Mathematics Subject Classification: 12D15, 14P05, 14P10, 12J15, 12J10, 13J30
© Springer Nature Switzerland AG 2022

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
More than 30 years after its publication, we are pleased and surprised with the
ongoing interest in Einführung in die reelle Algebra. Given the vibrant development
of real algebra and geometry in the previous decades, this seems by no means self-
evident.
Real algebra has grown in many directions, partially from within itself, and
exceedingly also through nudges and stimuli from without. While some of these
developments were already discernible at the end of the 1980s, many modern ones
that currently belong to the core of the area—such as the connections with tropical
geometry, or with semidefinite optimization—would have been difficult to predict.
Hence, it is not obvious that a text that was timely and modern in 1989 still serves
current needs in a reasonable manner.
Nevertheless, we are convinced that also from today’s perspective the choice
and exposition of the material in Einführung in die reelle Algebra still constitute a
solid first course on the basic principles and important techniques of real algebra.
For this reason, we decided to keep the text essentially unchanged in the English
translation—even though it was tempting to change the wording here and there, and
to drop some sections in favour of new ones.
A small number of typos, omissions and errors that were present in the German
original have been corrected, and some notation and terminology have been brought
up to date. Several new examples (some in the form of exercises) have been added.
All definitions and propositions have been numbered in a consistent manner. The
original bibliography has been augmented with a number of newer references.
A feature of this translation is the addition of a short fourth chapter that provides
a succinct overview of the most important developments and advances in those parts
of real algebra that are directly related to material covered in Einführung in die reelle
Algebra.
We are grateful to Springer Nature, and Rémi Lodh in particular, for their
endorsement and friendly and professional assistance.
We received support and encouragement from many quarters, and are pleased to
extend special thanks to Oliver Hien and Marcus Tressl, who made valuable remarks
on the German original.
v
vi Preface
Finally, our very special and heartfelt thanks go to Thomas Unger, who translated
the original text in a conscientious and competent manner, and co-authored the final
chapter. It was always a great pleasure to work with him.
Regensburg, Germany Manfred Knebusch

Konstanz, Germany Claus Scheiderer
May 2022
Preface to Einführung in die reelle
Algebra
Algebra textbooks that are currently in common use present real algebra only in
later chapters and then usually rather tersely. This is so for the influential works
of van der Waerden [113], Jacobson [55] and Lang [73], although Jacobson covers
a bit more than the others. Bourbaki’s substantial multivolume text Eléments de
Mathématique shows a similar picture: the volume Algèbre contains just a short
chapter (Chapter 6, Groupes et corps ordonnés) on real algebra. In contrast, a
complete volume (currently counting nine chapters) is dedicated to commutative
algebra, even though on the whole only the elementary part is covered, and not
more than what is absolutely essential for the foundation of the theory, both by
current standards as well as Bourbaki’s own standards.
That being the case, not too many algebraists today seem to perceive real algebra
as a proper branch of algebra at all. This has not always been so. In the nineteenth
century real algebra flourished. The study of real zeroes of a real polynomial in one
variable was at the centre of algebraic interest during the whole century, and was an
indispensable part of any higher mathematical education.
Heinrich Weber’s large three-volume textbook on algebra [114] provides evi-
dence for this fact. As Weber’s research interests were primarily in number theory,
especially in complex multiplication and class field theory, he focussed his textbook
mostly on these topics. Nevertheless he devoted well over a hundred pages in the
first volume to real zeroes of real polynomials.
The twentieth century witnessed a dramatic decline in interest in real algebra.
This trend only seems to have changed since the late 1970s, which is all the more
astonishing since the essential seeds of a modern real algebra, as we understand it
today, are already present in two works by Artin and Schreier from the 1920s [3, 4].
So, what is real algebra? Instead of giving a formal definition—which would
be difficult—we rather answer the question with an analogy that puts real algebra
in parallel with commutative algebra. Commutative algebra can be seen as that
part of algebra that contains the algebraic foundations that are typically important
for algebraic geometry (in particular in its modern, abstract form); and algebraic
geometry is ultimately the study of solution sets of systems of polynomial equations
F (x1 , . . . , xn ) = 0 and non-equations F (x1 , . . . , xn ) = 0. Correspondingly,
vii
viii Preface to Einführung in die reelle Algebra
real algebra provides algebraic methods that typically function as tools for real
algebraic geometry, and in particular semialgebraic geometry, which is the study
of the solution sets of systems of polynomial inequalities F (x1 , . . . , xn ) > 0 or
F (x1 , . . . , xn ) ≥ 0, where the coefficients classically come from the field of real
numbers and more generally from any ordered field.
This analogy provides an argument for why the interest has been so much
stronger in commutative than in real algebra during our century thus far. Indeed,
algebraic geometry experienced a continuous and ultimately triumphant upswing in
the twentieth century, while real algebraic geometry was only practiced in isolation
and then mostly with transcendental methods. Thus, from the geometric point of
view, there was for many decades no engine available that could have pushed real
algebra forward.
It was not until 1987 that the first textbook [11] on real algebraic geometry
was published. Its authors’ report (in the last section of the introduction to this
commendable work [11, p. 4 ff.]) on the strange sleeping beauty slumber of real
algebraic geometry (insofar as it was practiced with algebraic methods) is worth
considering.
In the meantime this slumber has given way to a lively development. We
consider the introduction—or better: discovery—of the real spectrum Sper A of a
commutative ring A by Michel Coste and Marie-Françoise Roy around the year
1979 as the most important trigger.
One can view Sper A as an analogue of the Zariski spectrum Spec A, intro-
duced by Grothendieck. It is well-known that the Zariski spectrum is the key
to Grothendieck’s abstract algebraic geometry. Likewise, the real spectrum is the
key to an abstract semialgebraic geometry. (There is one difference: in contrast to
Spec A, the real spectrum Sper A seems to carry (at least) two important structure
sheafs, the sheaf of abstract Nash functions, introduced by Coste and Roy [29], as
well as the sheaf of abstract semialgebraic functions, introduced by G. Brumfiel and
N. Schwartz [19, 107, 108].)
Real algebra is necessary to understand the latest developments in real algebraic
geometry and to meet the future intellectual challenges in this area. This brings us
to the objective of this book.
Our book is based on two insights: real algebra is a branch of algebra that is
largely autonomous in its foundations, with methods specific to this branch. These
foundations can be successfully taught with little more preparation than the standard
background from a typical one-semester algebra course on linear algebra, group
theory, field theory and ring theory.
We differentiate more precisely between an elementary and a higher real algebra.
The former can be developed without any special prior knowledge and used with
benefit in real algebraic geometry. The latter makes serious use of resources from
other branches of mathematics, especially real algebraic geometry, commutative
algebra, algebraic geometry, model theory and the theory of quadratic forms, but
occasionally also algebraic topology, real analysis and complex analysis. (This list
can certainly be extended.) An analogous distinction can be made in commutative
algebra. A demarcation between “elementary” and “higher” in either area is not
Preface to Einführung in die reelle Algebra ix
entirely objectively possible, but rather is subject to personal points of view and
taste. Furthermore, the higher one goes, the more fluid and arbitrary the boundary
between the two kinds of algebra and their corresponding geometries becomes.
Our book is dedicated to elementary real algebra in the above sense. Another
book on higher real algebra is planned.1 In the current book we do get by with
previous knowledge of the extent sketched above. By adding another twenty to
thirty pages, we could have reduced the requirements further and, for example,
developed everything that is needed from commutative algebra and the theory of
quadratic forms. We have not done so however, since students who are interested in
real algebra more than likely already mastered almost all of the prerequisites for the
current book, and would easily be able to find the few things they may be missing
elsewhere.
One specific feature of real algebra should be pointed out in particular: the major
role that general (Krull) valuation rings play. In most parts of commutative algebra,
valuation rings that are not discrete are only viewed as an aid that can often be
done without. In real algebra on the other hand, general valuation rings are a natural
and even central concept throughout. The reason is that every convex subring of
an ordered field is a valuation ring, but only in rare cases a discrete valuation
ring. This fact was already observed by Krull in the introduction to his pioneering
work Allgemeine Bewertungstheorie [67]. (It can already be found in embryonic
form, without the concept of valuation ring, in the work of Artin and Schreier,
cf. [4, p. 95].) Krull also identified the theory of ordered fields as an important
application area of general valuation theory, but then devoted only one section—
albeit substantial—of his great work to ordered fields [67, §12].
Our book is divided into three chapters. In addition to the Artin-Schreier theory
of ordered fields and the elementary relationships between orderings and quadratic
forms, the first chapter covers some aspects of real algebra from the nineteenth
century. Various methods for determining the number of real zeroes of a real
polynomial are treated (Sturm’s algorithm, Hermite’s method using quadratic forms,
Hurwitz’ Theorem). Another section is devoted to the relationship between the
Cauchy index and the Hankel form, as well as the Bézoutian of a rational function.
The second chapter deals with real valuation theory. Everything we need from
general valuation theory is developed from scratch. The chapter culminates in a
presentation of Artin’s solution of Hilbert’s 17th Problem.
Finally, the third chapter is devoted to the real spectrum. After a short crash
course on the Zariski spectrum, the real spectrum is examined in detail, but only
as a topological space (i.e., without the introduction of a structure sheaf). In the
geometric setting (affine algebras over real closed fields), the points as well as
certain subsets of the real spectrum are described in terms of filter sets. In the last
five sections of the chapter we come to some parts of real algebra in which the real
spectrum proves to be clarifying and helpful, such as the reduced Witt ring of a field,
1 This book was never written.

x Preface to Einführung in die reelle Algebra
Positivstellensätze, preorderings of rings, convex ideals, and the holomorphy ring of

a field. Many of the previously developed techniques and terms are used again.
Experts should note that the book manages without Tarski’s Principle throughout.
(That so much of real algebra can be developed in a meaningful way without using
this principle should hopefully come as a little surprise.) Although we consider
Tarski’s Principle to be extremely important and by no means particularly difficult,
we still count it as part of higher real algebra.
The book originates in a four-hour course by the first author in Regensburg in
the 1986/87 winter semester. This course was followed by an extensive course on
model theory, in which Tarski’s Principle was proved fairly quickly and painlessly
following Prestel [91], and then by a course on higher real algebra.
To the subject matter of the first course we added individual sections here and
there, so that the book now encompasses more than can be handled in one semester.
However, since some sections can be skipped without consequences for further
understanding, it can still be used as a guide for a one-semester course.
We want to guard against the misunderstanding that our pair of opposites “ele-
mentary–higher” is strongly correlated with the pair “easy–difficult”. We consider
elementary real algebra, as presented here, not to be always easy. Beginners may
need to overcome some pedagogical hurdles. Especially in the third chapter, some
things may seem strange at first glance. Our text is also rather terse and requires
the reader to participate intensively with paper and pencil. Our goal was to delight,
motivate and enable the beginner, but not to put all the obstacles out of the way with
lengthy statements, which after all could be quite tiresome.
We would like to thank the editor of the series “Aufbaukurs Mathematik”,
Gerd Fischer, as well as Vieweg-Verlag, and here in particular Ulrike Schmickler-
Hirzebruch, for her considerable understanding and empathy during the preparation
of the manuscript. Marina Franke TEXed numerous versions of the manuscript
with patience and competence; Uwe Helmke, Roland Huber and Michael Prechtl
proofread the manuscript and helped us a lot with comments and suggestions for
improvement. We would like to take this opportunity to thank all of them.
Regensburg
January 1989
Contents
1 Ordered Fields and Their Real Closures . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1

1.1 Orderings and Preorderings of Fields . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.2 Quadratic Forms, Witt Rings, Signatures . . . . . . . .. . . . . . . . . . . . . . . . . . . . 5
1.3 Extension of Orderings . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 10
1.4 The Prime Ideals of the Witt Ring . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 13
1.5 Real Closed Fields and Their Field Theoretic
Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 15
1.6 Galois Theoretic Characterization of Real Closed Fields .. . . . . . . . . . 17
1.7 Counting Real Zeroes of Polynomials (without
Multiplicities) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 19
1.8 Conceptual Interpretation of the Sylvester Form . . . . . . . . . . . . . . . . . . . . 26
1.9 Cauchy Index of a Rational Function, Bézoutian and
Hankel Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 30
1.10 An Upper Bound for the Number of Real Zeroes
(with Multiplicities) .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 41
1.11 The Real Closure of an Ordered Field . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 46
1.12 Transfer of Quadratic Forms . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 48
2 Convex Valuation Rings and Real Places .. . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 53
2.1 Convex Subrings of Ordered Fields . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 53
2.2 Valuation Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 57
2.3 Integral Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 60
2.4 Valuations, Ideals of Valuation Rings . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 63
2.5 Residue Fields and Subfields of Convex Valuation Rings . . . . . . . . . . 70
2.6 The Topology of Ordered and Valued Fields . . . .. . . . . . . . . . . . . . . . . . . . 73
2.7 The Baer–Krull Theorem . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 75
2.8 Places . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 78
2.9 The Orderings of R(t), R((t)) and Quot R{t} . . . .. . . . . . . . . . . . . . . . . . . . 82
2.10 Composition and Decomposition of Places . . . . . .. . . . . . . . . . . . . . . . . . . . 86
2.11 Existence of Real Places of Function Fields . . . . .. . . . . . . . . . . . . . . . . . . . 89
xi
xii Contents
2.12 Artin’s Solution of Hilbert’s 17th Problem and the Sign

Change Criterion .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 94
3 The Real Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 101
3.1 The Zariski Spectrum. Affine Varieties . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 101
3.2 Reality for Commutative Rings . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 110
3.3 Definition of the Real Spectrum . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 114
3.4 Constructible Subsets and Spectral Spaces . . . . . .. . . . . . . . . . . . . . . . . . . . 120
3.5 The Geometric Setting: Semialgebraic Sets and Filter
Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 127
3.6 The Space of Closed Points . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 135
3.7 Specializations and Convex Ideals . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 140
3.8 The Real Spectrum and the Reduced Witt Ring of a Field .. . . . . . . . . 145
3.9 Preorderings of Rings and Positivstellensätze . . .. . . . . . . . . . . . . . . . . . . . 150
3.10 The Convex Radical Ideals Associated to a Preordering . . . . . . . . . . . . 158
3.11 Boundedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 166
3.12 Prüfer Rings and the Real Holomorphy Ring of a Field . . . . . . . . . . . . 174
4 Recent Developments .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 187
4.1 Counting Real Solutions . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 187
4.2 Quadratic Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 188
4.3 Stellensätze . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 189
4.4 Noncommutative Stellensätze . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 192
References .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 195
Symbol Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 201
Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 203
Chapter 1
Ordered Fields and Their Real Closures
Our starting point is the basic notion for the entire book, the general concept of
orderings of arbitrary fields. Conceived by Artin and Schreier in their foundational
1927 paper [4], it was successfully used by Artin in his solution of Hilbert’s 17th
Problem in the same year [3]. We introduce real closed fields and show that they
have the same algebraic properties as the field of real numbers. Moreover we prove
the existence and uniqueness of a real closure for any ordered field.
The second main topic in this chapter concerns methods for counting real roots
of real polynomials. Instead of trying to calculate or estimate the roots numerically,
we discuss purely algebraic methods that extend from the real numbers to any real
closed base field. These results were mostly found in the nineteenth century or even
earlier, and are connected to famous names such as Descartes, Sturm, Sylvester and
Hermite.
Parallel to these two subjects we give an introduction to the basic notions of
quadratic forms over fields and their algebraic theory. They will find applications in
Chaps. 2 and 3.
1.1 Orderings and Preorderings of Fields
Let K be a field.
Definition 1.1.1 An ordering of K is a subset P of K that satisfies the properties
(O1) P + P ⊆ P , P P ⊆ P ,
(O2) P ∩ (−P ) = {0},
(O3) P ∪ (−P ) = K,
where P + P := {a + b : a, b ∈ P } and P P := {ab : a, b ∈ P }. The pair (K, P )
is called an ordered field.
© Springer Nature Switzerland AG 2022 1

M. Knebusch, C. Scheiderer, Real Algebra, Universitext,
https://doi.org/10.1007/978-3-031-09800-0_1
2 1 Ordered Fields and Their Real Closures
Remark 1.1.2 If we assume (O1) and (O3), then (O2) is equivalent with
(O2’) −1 ∈ P .
Exercise 1.1.3 Let P be an ordering of K. Show that
a ≤P b :⇔ b − a ∈ P (a, b ∈ K)
defines a total order relation ≤P on the set K which satisfies the properties
(i) a ≤P b ⇒ a + c ≤P b + c,
(ii) a ≤P b, c ≥ 0 ⇒ ac ≤P bc
for all a, b, c ∈ K. Conversely, let ≤ be a total order relation on K which satisfies
(i) and (ii). Show that P := {a ∈ K : a ≥ 0} is an ordering of K.
Thus we clearly have a one-one correspondence between orderings of K and total
order relations on K that satisfy (i) and (ii). Usually the latter are called orderings
of K as well. When P is clear we simply write a ≤ b instead of a ≤P b.
Definition 1.1.4 A preordering of K is a subset T of K that satisfies the properties
(P1) T + T ⊆ T , T T ⊆ T ,
(P2) T ∩ (−T ) = {0},
(P3) a 2 ∈ T for all a ∈ K.
Exercise 1.1.5 If T is a preordering of K, show that
a ≤T b :⇔ b − a ∈ T (a, b ∈ K)
defines a partial order relation ≤ on K that is compatible with + and ·.

If T is clear, we simply write a ≤ b instead of a ≤T b.
Remarks 1.1.6
(1) If we assume (P1) and (P3), then (P2) is equivalent with
(P2’) −1 ∈ T .
Note that every ordering is a preordering. Indeed, (P1) is (O1), (P2) is (O2),
and (P3) is a weaker version of (O3) by (P1).
(2) If T = (Tα : α ∈ I ) is a nonempty family of preorderings of K, then α Tα
is also a preordering. If in addition T is upward directed
(i.e., for all α, β ∈ I
there exists γ ∈ I such that Tα ∪ Tβ ⊆ Tγ ), then α Tα is also a preordering.
In particular, if K has a preordering, it will have a smallest one.
We introduce the notation
K 2 := {a12 + · · · + an2 : n ∈ N, a1 , . . . , an ∈ K}.

1.1 Orderings and Preorderings of Fields 3
It is clear that K 2 is contained in every preordering of K, and is itself a preordering

if and only if −1 ∈ K 2 .
Definition 1.1.7 The field K is called real if −1 ∈ K 2 , i.e., when −1 cannot be
written as a sum of squares in K.
The previous discussion shows that a field K has a preordering if and only if it
is real. In this case, K 2 is the smallest preordering on K. Note that all real fields
have characteristic 0.
Preorderings are precisely the intersections of orderings, as we will show next.
Lemma 1.1.8 Let T be a preordering of K and let a ∈ K with a ∈ T . Then
T − aT = {b − ac : b, c ∈ T } is again a preordering of K.
Proof T −aT clearly satisfies (P1) and (P3). We show (P2’): assume on the contrary
that −1 ∈ T − aT . Then −1 = b − ac for some b, c ∈ T . But then c = 0 and so
a = c−2 · c(1 + b) ∈ T , a contradiction.
Lemma 1.1.9 For every preordering T of K there exists an ordering P of K such

that T ⊆ P .
Proof Let M := {T ⊆ K : T is a preordering and T ⊆ T }. Then M is ordered by
inclusion, M = ∅ and Zorn’s Lemma applies (Remark 1.1.6(2)). Thus there exists
a maximal preordering P of K with T ⊆ P . This P is an ordering of K. Indeed,
if a ∈ K, a ∈ P , then P = P − aP by Lemma 1.1.8 and the maximality of P , so
−a ∈ P .
Theorem 1.1.10 Every preordering T of K is an intersection of orderings of K.

Proof The inclusion T ⊆ {P : P is an ordering and T ⊆ P } is trivial. If a ∈ K,
a ∈ T , then T − aT is a preordering and thus contained in an ordering P of K
(Lemma 1.1.9). It follows that a ∈ P since −a ∈ P and a = 0.
Corollary 1.1.11 K has an ordering if and only if K is real.

(This follows already from Lemma 1.1.9.)
Corollary 1.1.12 (E. Artin [3]) Assume char K = 2 and let a ∈ K. Then a ≥ 0
for all orderings of K if and only if a is a sum of squares in K.
Proof If K is real, then K 2 is a preordering and the statement follows from
Theorem 1.1.10. If K is not real and char K = 2, then K 2 = K since
2 2
a+1 a−1
a= + (−1) ·
2 2
for all a ∈ K.
Naturally, we must assume that char K = 2, since otherwise K 2 is a subfield of

K. Next, we consider some examples of orderings. We will see many more examples
later on.
Examples 1.1.13
(1) The field R of real numbers (and thus each of its subfields) has the usual
ordering.
(2) Let Q(t) be the rational function field in one variable over Q and let ϑ ∈ R
be transcendental (over Q). Then f (ϑ) is a well-defined real number for every
f ∈ Q(t). The set
P = {f ∈ Q(t) : f (ϑ) ≥ 0}
is an ordering of Q(t). It is the ordering induced by the field embedding Q(t) →

R, f → f (ϑ).
(3) Let (F, ≤) be an ordered field and F (t) the rational function field in one variable
over F . For f ∈ F (t) the notation f (a) = ∞ signifies that f (t) has no pole
in t = a. Then f (a) ∈ F is well-defined. The notation f (a) = b for a, b ∈ F
implies in particular that f (a) = ∞. The sets

Pa,+ := {0} ∪ (t − a)r f (t) : r ∈ Z, f ∈ F (t) with f (a) = ∞ and f (a) > 0
and

Pa,− := {0} ∪ (a − t)r f (t) : r ∈ Z, f ∈ F (t) with f (a) = ∞ and f (a) > 0
are orderings of F (t) for every a ∈ F . Both orderings extend the ordering of F .
The reason for this notation is that a < t < b with respect to Pa,+ holds for all
b ∈ F with b > a. Thus F (t) is ordered in such a way that the transcendental t
is located “immediately to the right of a” on the “line” F . Similarly, t is located
“immediately to the left of a” with respect to Pa,− .
Notation 1.1.14 If (K, P ) is an ordered field, we denote the sign function with
respect to P by signP : K → {−1, 0, 1}. Thus, signP (0) = 0, and for a ∈ K ∗ we
have signP (a) = 1 if a ∈ P and signP (a) = −1 if a ∈ P .
As usual, | · |P : K → P , |a|P := a · signP (a) denotes the absolute value. If P
is clear, the index P may be omitted.
The meaning of the generalized intervals [a, b]P , [a, b[P , ]a, b]P , ]a, b[P for
a, b ∈ K ∪ {−∞, +∞} is also clear, namely [a, b[P = {x ∈ K : a ≤P x <P b},
]a, ∞[P = {x ∈ K : x >P a}, etc. Here too we will usually drop the index P , and
instead often write [a, b]K , etc. if several fields are considered.
1.2 Quadratic Forms, Witt Rings, Signatures 5
1.2 Quadratic Forms, Witt Rings, Signatures
It is not possible to give an in-depth treatment of the algebraic theory of quadratic

forms in this book. In the next sections we will therefore only develop what we need
from the basics of the theory for later applications in real algebra. We will focus in
particular on the relations between quadratic forms and orderings. For references
and for a detailed treatment of quadratic form theory we recommend the books of
Lam [69, 72] and Scharlau [98]. For Witt rings there is the smaller volume [62].
In this section we assume that K is a field of characteristic char K = 2. All vector
spaces are assumed finite dimensional. Let V be a K-vector space. A symmetric
bilinear form on V (over K) is a K-bilinear map b : V ×V → K such that b(v, w) =
b(w, v) (v, w ∈ V ). A quadratic form on V (over K) is a map q : V → K that
satisfies q(av) = a 2 q(v) (a ∈ K, v ∈ V ) and such that the map bq : V × V → K,
bq (v, w) = q(v + w) − q(v) − q(w) is a (symmetric) K-bilinear form. Every
symmetric bilinear form b on V defines a quadratic form qb on V via qb (v) :=
b(v, v). Since 2 = 0, the maps b → qb and q → 12 bq describe mutually inverse
bijections between the symmetric bilinear forms and the quadratic forms on V , and
as such these concepts are usually identified.
The pair ϕ = (V , b) is called a bilinear space, whereas the pair ϕ = (V , q)
is called a quadratic space. Two quadratic spaces (V , q) and (V , q ) are called
isomorphic (or isometric), denoted (V , q) ∼ = (V , q ), if there exists a vector space
isomorphism φ : V → V such that q = q ◦ φ. Quadratic spaces are most easily
described by symmetric matrices: If B = (bij ) ∈ Mn (K) is symmetric, then B
defines the quadratic space ϕB = (K n , qB ), where

n
qB (x) = bij xi xj x = (x1 , . . . , xn ) ∈ K n .
i,j =1
For symmetric matrices B, B ∈ Mn (K) it is straightforward to check that

ϕB ∼ = ϕB if and only if there exists S ∈ GLn (K) such that B = SBS t . If
B = diag(a1 , . . . , an ) is a diagonal matrix, we simply write ϕB as a1 , . . . , an
Every quadratic space is diagonalizable, i.e., isomorphic to a space a1 , . . . , an .
The hyperbolic plane H is the quadratic space defined by B = 01 10 . In diagonal
form, H ∼ = 1, −1. We write a1 , . . . , an K or HK when we want to emphasize the
base field K.
Given bilinear spaces ϕ = (V , b) and ϕ = (V , b ), we can construct new ones,
namely their orthogonal sum ϕ ⊥ ϕ = (V ⊕ V , b ⊥ b ) and their tensor product
ϕ ⊗ ϕ = (V ⊗ V , b ⊗ b ). Here b ⊥ b and b ⊗ b are defined by (b ⊥ b )(v +
v , w +w ) = b(v, w)+b (v , w ) and (b ⊗b )(v ⊗v , w ⊗w ) = b(v, w) b (v , w ),
respectively. If ϕ and ϕ are represented by matrices B and B , respectively, then
ϕ ⊥ ϕ is represented by B0 B0 and ϕ ⊗ ϕ is represented by the Kronecker product
B ⊗ B . In particular,
a1 , . . . , am ⊥ b1 , . . . , bn ∼
= a1 , . . . , am , b1 , . . . , bn
and
a1 , . . . , am ⊗ b1 , . . . , bn ∼
= ⊥a b .
i,j
i j
The n-fold sum ϕ ⊥ · · · ⊥ ϕ is abbreviated by n × ϕ and the n-fold tensor product

ϕ ⊗ · · · ⊗ ϕ by ϕ ⊗n . Finally, if ϕ = (V , b), we write −ϕ for (V , −b).
Let ϕ = (V , b) be a bilinear space and U ⊆ V a subset. Then U ⊥ := {v ∈
V : b(u, v) = 0 for all u ∈ U } is a linear subspace of V . If V ⊥ = 0, then ϕ (or
b, or q) is called nondegenerate (or regular), otherwise ϕ (or b, or q) is called
degenerate (or singular). For a symmetric matrix B, ϕB is degenerate if and only
if det B = 0. The space V ⊥ is also called the radical of ϕ, denoted Rad(ϕ), and
codimV (V ⊥ ) is called the rank of ϕ, denoted rank(ϕ). Given any quadratic space
ϕ there exists a nondegenerate quadratic space ϕ (unique up to isomorphism) such
that ϕ ∼= ϕ ⊥ Rad(ϕ). For this reason one usually only considers nondegenerate
spaces. Straightforward, yet important, is the following fact: If (V , q) is a quadratic
space and W ⊆ V a subspace such that (W, q|W ) is nondegenerate, then V = W ⊥
W ⊥.
The simplest invariants of nondegenerate quadratic spaces are the dimension and
determinant. We have det(V , q) := (det B)K ∗2 ∈ K ∗ /K ∗2 , where q is represented
by the matrix B. Note that det(ϕ ⊥ ϕ ) = det(ϕ) · det(ϕ ), and det(ϕ ⊗ ϕ ) =
(det ϕ)dim ϕ · (det ϕ )dim ϕ . Of fundamental importance is
Theorem 1.2.1 (Witt’s Cancellation Theorem) Let ϕ1 , ϕ2 and ψ be quadratic
= ϕ2 ⊥ ψ, then ϕ1 ∼
spaces. If ϕ1 ⊥ ψ ∼ = ϕ2 .
The proof can be found in any textbook on quadratic forms (e.g., [62, 69, 72, 98]),
as well as in a number of algebra textbooks (e.g., [54, 55, 73]).
A quadratic form q on V represents an element a ∈ K if there exists 0 = v ∈ V
such that q(v) = a. If q represents all nonzero a ∈ K, then q is called universal.
For example, the hyperbolic plane H is universal. A useful observation: If a1 ∈ K ∗
is represented by q, there exist a2 , . . . , an ∈ K such that q ∼ = a1 , a2 , . . . , an
(n = dim V ). The quadratic space (V , q) is called isotropic if q represents zero, and
anisotropic otherwise. Every nondegenerate quadratic space ϕ has an orthogonal
Witt decomposition ϕ ∼ = ϕ0 ⊥ n × H where n ≥ 0 and ϕ0 is anisotropic. By Witt’s
Cancellation Theorem, ϕ0 (up to isomorphism) and n are uniquely determined. We
call ϕ0 the kernel form and n the Witt index of ϕ. If ϕ0 = 0, i.e., ϕ ∼= n × H , then
ϕ is called hyperbolic. A quadratic space ϕ = (V , q) is hyperbolic if and only if it
is nondegenerate and there exists a subspace U of V such that U = U ⊥ . From Witt
decomposition it follows that every isotropic form is universal.
For the remainder of this section, all quadratic spaces are assumed to be
nondegenerate, unless stated otherwise.
Definition 1.2.2 Two quadratic spaces ϕ, ϕ are called Witt equivalent, denoted
ϕ ∼ ϕ , if their kernel forms are isomorphic. We denote the class of all quadratic
spaces that are Witt equivalent to ϕ by [ϕ] (the Witt class of ϕ), and the set of all
Witt classes of (nondegenerate) quadratic spaces over K by W (K).
Lemma 1.2.3 Let ϕ, ϕ , ψ, ψ be quadratic spaces.

(a) ϕ ∼ ψ ⇔ there exist m, n ≥ 0 such that ϕ ⊥ m · H ∼ = ψ ⊥ n · H.
(b) ϕ ⊗ H ∼= ϕ ⊥ (−ϕ) is a hyperbolic space.
(c) Witt equivalence is compatible with ⊥ and ⊗, i.e., ϕ ∼ ϕ and ψ ∼ ψ imply
ϕ ⊥ ψ ∼ ϕ ⊥ ψ and ϕ ⊗ ψ ∼ ϕ ⊗ ψ .
Proof (a) follows from the uniqueness of Witt decomposition.
(b) Since H ∼ = 1, −1 we have ϕ ⊗ H ∼ = ϕ ⊥ (−ϕ). Since ϕ is diagonalizable,
it suffices to settle the case ϕ = a (a ∈ K ∗ ). Assume that ϕ ⊗ H is represented by
the matrix a0 a0 with respect to a basis (e, f ). Then it is represented by 01 10 with
respect to the basis (e, a −1 f ). This shows ϕ ⊗ H ∼ = H.
(c) follows from (a) and (b).
Theorem 1.2.4 The set W (K) is made into a commutative unitary ring with unit
[1] by the (well-defined) pairings
[ϕ] + [ψ] := [ϕ ⊥ ψ] and [ϕ] · [ψ] := [ϕ ⊗ ψ].
W (K) is called the Witt ring of the field K.

Proof That the pairings are well-defined follows from Lemma 1.2.3(c). That [ϕ] +
[−ϕ] = 0 in W (K) for all ϕ follows from Lemma 1.2.3(b). The other ring axioms
are trivially satisfied.
In order not to overload our notation, we will often write a1 , . . . , an instead of
a1 , . . . , an . Thus, a1 , . . . , an may describe a quadratic form as well as its Witt
class.
Example 1.2.5 With reference to Exercises 1.2.17 and 1.2.18 below we have
W (R) ∼
= Z and W (C) ∼
= Z/2Z.
Exercise 1.2.6 Let K be a finite field of characteristic not 2.

(a) Show that |K ∗ /K ∗2 | = 2.
(b) Show that any 2-dimensional form is universal, and conclude that |W (K)| = 4.
(c) Distinguish the two cases |K| ≡ 1 or − 1 mod 4, and conclude that W (K) ∼ =
Z/2Z[t] ∼ Z/4Z in the second case.
in the first case, and W (K) =
(t 2 )
Example 1.2.7 (Milnor’s Exact Sequence) For every prime number p, we denote
the finite field with p elements by Fp . Then W (Q) can be computed explicitly as an
additive group by the split exact sequence
where i : Z → W (Q) is the unique homomorphism that sends 1 to 1, and ∂2 and
the ∂p for odd primes p are certain group homomorphisms, cf. [72, VI, §4] for the
details. Witt groups of number fields are considerably harder to understand, cf. [72,
VI, §3]. There is a similar split exact sequence for Witt groups of rational function
fields in one variable, cf. [72, IX, §3].
Next we study the relations between the Witt ring of K and orderings of K.
Definition 1.2.8 Let (K, P ) be an ordered field and let ϕ = (V , q) be a quadratic
space over K. Then ϕ (or q) is called positive definite with respect to P if q(v) >P 0
for all 0 = v ∈ V . If −q is positive definite, then q is called negative definite.
The following theorem is well-known:
Theorem 1.2.9 (Sylvester’s Inertia Theorem) Let ϕ be a possibly degenerate
quadratic space over K and P an ordering of K. Then there is an orthogonal
decomposition
ϕ∼
= ϕ+ ⊥ ϕ− ⊥ Rad(ϕ),
where ϕ+ is positive definite with respect to P and ϕ− is negative definite with

respect to P . Furthermore, dim ϕ+ and dim ϕ− only depend on ϕ (but not on ϕ+
and ϕ− ).
Proof We may assume that Rad(ϕ) = 0. The existence of ϕ+ and ϕ− follows from
diagonalizing ϕ. Assume that ϕ ∼ = ϕ+ ⊥ ϕ− ∼ = ψ+ ⊥ ψ− , with ϕ+ , ψ+ positive
definite and ϕ− , ψ− negative definite. Assume that ϕ (resp. ϕ± , resp. ψ± ) is defined
on V (resp. on V± , resp. on W± ). We interpret V± and W± as subspaces of V .
If dim V+ < dim W+ , then V− ∩ W+ = 0 since dim V− + dim W+ > dim V .
This is a contradiction since ϕ is positive and negative definite on V− ∩ W+ . If
dim V+ > dim W+ we get a similar contradiction.
Definition 1.2.10 Let ϕ be a quadratic space over K, P an ordering of K, and

ϕ∼
= ϕ+ ⊥ ϕ− ⊥ Rad(ϕ) a decomposition as in Theorem 1.2.9. Then the integer
signP ϕ := dim(ϕ+ ) − dim(ϕ− )
is called the Sylvester signature of ϕ at P .

Lemma 1.2.11 Let (K, P ) be an ordered field. Then [ϕ] → signP ϕ defines a ring
homomorphism signP from W (K) to Z.
Proof The following equalities are easily verified:
signP (ϕ ⊥ ψ) = signP ϕ + signP ψ,

signP (ϕ ⊗ ψ) = (signP ϕ) · (signP ψ),
signP H = 0.
The statement follows.

Now a critical observation is that, conversely, every ring homomorphism
W (K) → Z arises in this manner. In other words, orderings of K and
homomorphisms W (K) → Z are “the same”:
Theorem 1.2.12 The Sylvester signature defines a bijection sign : P → signP from
the set of orderings of K to the set of ring homomorphisms W (K) → Z.
Proof The map sign is injective since P = {0} ∪ {a ∈ K ∗ : signP [a] = 1}. Let
φ : W (K) → Z be a homomorphism. Define χ : K ∗ → Z by χ(a) := φ[a]. Then
χ : K ∗ → {±1} is a group homomorphism since χ(1) = 1 and χ(ab) = χ(a)χ(b)
(a, b ∈ K ∗ ). We must show that P := {0} ∪ Ker χ is an ordering of K. Then
φ = signP follows immediately. Since [1] + [−1] = 0 in W (K), we have
φ[−1] = χ(−1) = −1, i.e., −1 ∈ P . Then P ∪ (−P ) = K is also clear, and
P P ⊆ P follows since χ is a homomorphism. In order to show P + P ⊆ P
(which completes the proof), we use Lemma 1.2.13 below. Assume that a, b ∈ P .
We may assume a, b, a + b = 0. By Lemma 1.2.13, applying φ to [a, b] gives
χ(a)+χ(b) = χ(a +b)·(1 + χ(a)χ(b)), and χ(a) = χ(b) = 1 gives χ(a +b) = 1.
Lemma 1.2.13 If a, b ∈ K ∗ with a + b = 0, then
a, b ∼
= a + b, (a + b)ab .
Proof Since a + b is represented by a, b, there exists c ∈ K ∗ such that a, b ∼
=
a + b, c. Comparing determinants gives abK ∗2 = (a + b)cK ∗2, i.e., cK ∗2 =
(a + b)abK ∗2. The statement follows.
Theorem 1.2.12 justifies:
Definition 1.2.14 A signature of a field K is a ring homomorphism W (K) → Z.
Remark 1.2.15 We see that a field admits a signature if and only if it is real.
Note that for every field K (with char K = 2) there exists a ring homomorphism
e : W (K) → Z/2Z, defined by e[ϕ] := dim(ϕ) + 2Z. Every signature σ yields a
commutative diagram
Definition 1.2.16 The map e : W (K) → Z/2Z is called the dimension index. Its
kernel is denoted I (K) and is called the fundamental ideal of W (K). (Thus I (K)
consists of the Witt classes of even dimensional quadratic spaces.)
Exercise 1.2.17 Show that if K ∗2 ∪ {0} = {a 2 : a ∈ K} is an ordering of K, then

it is the only ordering of K, and the Sylvester signature gives a ring isomorphism
W (K) → Z.
This is the case for example when K = R or, more generally, when K is any real
closed field, cf. Sect. 1.5.
Exercise 1.2.18 Show that if K is a quadratically closed field (i.e., K ∗2 = K ∗ )
then, up to isomorphism, every nondegenerate quadratic form is determined by its
dimension, and e is a ring isomorphism W (K) → Z/2Z.
1.3 Extension of Orderings
In this section, K denotes a field of characteristic = 2.

Let L/K be a field extension. A bilinear space ϕ = (V , b) over K gives rise to
a bilinear space ϕL = (VL , bL ) over L via extension of scalars: VL = V ⊗K L and
bL is defined by bL (v ⊗ a, v ⊗ a ) = b(v, v )aa for a, a ∈ L, v, v ∈ V . If q
is the quadratic form (over K) associated to ϕ, then qL denotes the quadratic form
(over L) associated to ϕL . If ϕ is represented by a matrix, then ϕL is represented by
the same matrix, but this time considered as a matrix over L.
The form ϕL is nondegenerate if and only if the form ϕ is nondegenerate. If
ϕ is hyperbolic, then ϕL is also hyperbolic, but the converse is in general false
(example?). Consequently, ϕ → ϕL induces a map on Witt classes, iL/K : W (K) →
W (L), which is a ring homomorphism. If M/L is a further field extension, then
iM/K = iM/L ◦ iL/K .
1.3 Extension of Orderings 11
Definition 1.3.1 Let L/K be a field extension.

(a) An ordering P of K extends to an ordering Q of L if P ⊆ Q (and so, P =
K ∩ Q).
(b) A signature σ of K extends to a signature τ of L if σ = τ ◦ iL/K . We also write
τ | σ.
One should convince oneself that for orderings P of K and Q of L, P extends to
Q if and only if signP extends to signQ .
Proposition 1.3.2 Let L/K be a field extension and P an ordering of K. The
following statements are equivalent:
(i) P extends to an ordering Q of L;
(ii) −1 is not contained in the semiring T generated by P and L2 in L (i.e., the
smallest subset T of L such that P ∪ L2 ⊆ T , T + T ⊆ T and T T ⊆ T );
(iii) every quadratic form p1 , . . . , pn with p1 , . . . , pn ∈ P ∗ := P \ {0} is
anisotropic over L.
A further equivalent statement will be listed in Proposition 1.4.4.
Proof (i) ⇒ (iii): If p1 b12 + · · · + pn bn2 = 0, with b1 , . . . , bn ∈ L, then b1 = · · · =
bn = 0 since Q is an ordering.
(iii) ⇒ (ii): We can write T = {p1 b12 + · · · + pm bm 2 : m ≥ 1, p ∈ P ∗ , b ∈ L}.
i i
If −1 ∈ T , then −1 = p1 b12 + · · · + pm bm 2 and the form 1, p , . . . , p is isotropic
1 m
over L.
(ii) ⇒ (i): By assumption T is a preordering of L. Every ordering Q of L such
that Q ⊇ T (cf. Lemma 1.1.9) extends P .
√
Proposition 1.3.3 Let P be an ordering of K, d ∈ K and L = K( d). Then P
extends to L if and only if d ∈ P .
Proof If d ∈ P , then P has no extension to L since d ∈ L∗2 . Assume thus that
d ∈ P and, without loss of generality, that K = L. We verify Proposition 1.3.2(ii).
Assume for the sake of contradiction that there is an equality

m
√
−1 = pi (ai + bi d)2
i=1
in L with m ≥ 1, pi ∈ P and ai , bi ∈ K. It follows in particular (by comparing

√ m
coefficients with respect to the K-basis (1, d) of L) that −1 = pi (ai2 + dbi2 ),
and so −1 ∈ P , a contradiction. i=1
√
We will see later that if√d ∈ P (and d ∈ K), then there are precisely two
extensions of P to L = K( d) (cf. Corollary 1.11.7).
Example 1.3.4 The unique
√ ordering of Q has precisely two extensions,
√ P+ and P− ,
to√the field L = Q( 2). They are distinguished by the fact that 2 ∈ P+ and
− 2 ∈ P− . They are the only orderings of L and correspond to the two possible
embeddings of L into R.
In the following results we consider field extensions for which every ordering
extends:
Proposition 1.3.5 Let L/K be a finite field extension of odd degree. Then every
ordering of K extends to L.
By Proposition 1.3.2, this proposition follows from
Theorem 1.3.6 (T.A. Springer) If L/K is a finite field extension of odd degree and
q is an anisotropic quadratic form over K, then qL is anisotropic over L.
Proof We may assume without loss of generality that L = K(α) is a simple field
extension. Let f ∈ K[t] be the minimal polynomial of α over K. We proceed
by induction on n = [L : K] = deg f . Assume that n > 1 is odd and that the
statement is true for all smaller degrees. Let q = a1 , . . . , am be an anisotropic
form over K. Assume for the sake of contradiction that qL is isotropic. Then there
exist g1 , . . . , gm , h ∈ K[t] with deg gi < n (i = 1, . . . , m) and not all gi = 0, such
that
a1 g1 (t)2 + · · · + am gm (t)2 = f (t)h(t) (1.1)
in K[t]. We may assume that gcd(g1 , . . . , gm ) = 1.

Let d be the maximum of the degrees of the gi . Then n + deg h = 2d since q is
anisotropic. Since d < n we have deg h < n and deg h is odd. In particular, h has an
irreducible factor h1 = h1 (t) of odd degree. For the extension field E = K[t]/(h1 )
of K we thus have [E : K] < n and [E : K] is odd. On the other hand, qE is
isotropic by (1.1). This contradicts the induction hypothesis.
Proposition 1.3.7 If L/K is a purely transcendental field extension (not necessar-

ily finitely generated), then every ordering of K extends to L.
Using Zorn’s Lemma, we may assume that L = K(t) is simple transcendental
and specify explicit extensions (cf. Example 1.1.13(3)). A different proof (again
using Proposition 1.3.2) can be obtained from:
Proposition 1.3.8 If L/K is purely transcendental and q is an anisotropic
quadratic form over K, then qL is also anisotropic.
Proof We may assume without loss of generality that L = K(t) is simple
transcendental. If q = a1 , . . . , am (with ai ∈ K ∗ ) and qL is isotropic, then there
exist polynomials g1 , . . . , gm ∈ K[t], not all 0, with a1 g1 (t)2 + · · · + am gm (t)2 = 0
in K[t]. By considering again the maximal occurring degree, it follows that q is
isotropic.
Remark 1.3.9 From Theorem 1.3.6 and Proposition 1.3.8 it follows in particular
that the homomorphism iL/K : W (K) → W (L) is injective in case L/K is finite of
odd degree or purely transcendental.
1.4 The Prime Ideals of the Witt Ring 13
1.4 The Prime Ideals of the Witt Ring
In this section, K denotes a field of characteristic = 2.

We already know some of the prime ideals of the Witt ring W (K), namely the
fundamental ideal I (K), as well as the kernels of the signatures of K. We will see
that these are essentially the only prime ideals. Thus, Witt rings have a very simple
and clear prime ideal structure, a fact that was observed surprisingly late (J. Leicht,
F. Lorenz in [77] and D.K. Harrison in [44], both in 1970).
Theorem 1.4.1 (The Prime Ideals of W (K))
(a) If K is not real, then I (K) is the only prime ideal of W (K).
(b) If K is real, then the kernels pσ of the signatures σ of K are precisely the
minimal prime ideals of W (K). Every other prime ideal is either equal to I (K),
or is of the form p = pσ + p · W (K) for some signature σ and some prime
number p = 2, both uniquely determined by p.
Thus, pσ + 2 · W (K) = I (K) for every signature σ , as already observed in
Sect. 1.2.
Proof Let p be a prime ideal of W (K), κ(p) = Quot W (K)/p and π : W (K) →
π
W (K)/p the residue map. The composition φ : Z → W (K) − → W (K)/p is
∗
surjective since the Witt classes ξ = [a] (a ∈ K ) generate W (K) as a ring
and ξ 2 − 1 = 0 in W (K), hence ξ ≡ 1 or ξ ≡ −1 mod p.
If char κ(p) = 0, then φ is also injective, and thus an isomorphism, and p is the
kernel of the signature φ −1 ◦ π.
If char κ(p) = p > 0, then φ induces an isomorphism φ : Z/pZ → ∼ W (K)/p.
Assume that p = 2. Then p = I (K) since there is only one ring homomorphism
W (K) → Z/2Z (the generators [a], a ∈ K ∗ , are units in W (K)). Finally, assume
−1
that p > 2. Then +1 = −1 (in Z/pZ), and χ : a → φ ◦ π[a] (a ∈ K ∗ )
defines a group homomorphism χ : K ∗ → {±1}. (Note that χ(K ∗ ) ⊆ {±1} holds
by the argument above.) This map satisfies χ(−1) = −1. Following the proof of
Theorem 1.2.12 verbatim, it follows that P := {0} ∪ Ker χ is an ordering of K. Let
σ := signP . Since φ is an isomorphism, it follows from the commutative diagram
σ

that p = Ker W (K) − → Z → Z/pZ , i.e., that p = pσ + p · W (K). It is clear that
σ and p are uniquely determined by p. This establishes the proof.
Since in every (commutative) ring the nilradical (i.e., the set of all nilpotent
elements) is equal to the intersection of all prime ideals ([55, 68, 73], see also
Proposition 3.1.11), we obtain immediately
Corollary 1.4.2 Let ϕ be a nondegenerate quadratic space over K. The following

statements are equivalent:
(i) ϕ ⊗n is hyperbolic for some n ∈ N;
(ii) dim ϕ is even, and signP ϕ = 0 for every ordering P of K.
If K is ordered, the parity condition on dim ϕ can of of course be omitted
from (ii).
Applying Corollary 1.4.2 to ϕ = 1, 1 = 2 × 1 gives
Corollary 1.4.3 If K is not real, then there exists n ∈ N such that 2n · W (K) = 0.
Having determined the prime ideals of W (K), we can state a further criterion for
when orderings extend:
Proposition 1.4.4 Let L/K be a field extension. A signature of K extends to L if
and only if it vanishes on the kernel of iL/K : W (K) → W (L).
For the proof we require a simple fact from commutative algebra:
Lemma 1.4.5 If φ : A → B is an injective ring homomorphism and p a minimal
prime ideal of A, then there exists a prime ideal q of B such that p = φ −1 (q).
Proof Since S := φ(A\p) does not contain zero, we have that S −1 B = 0 and every
prime ideal q of S −1 B satisfies p = φ −1 j −1 (q ) , where j : B → S −1 B denotes
the canonical homomorphism.
Proof of Proposition 1.4.4 Let I := Ker iL/K and let σ be a signature of K. If σ
has an extension τ , then σ = τ ◦ iL/K , i.e., σ (I ) = 0.
Conversely, assume that σ (I ) = 0, i.e., I ⊆ pσ := Ker σ . Then pσ /I is a
minimal prime ideal of W (K)/I by Theorem 1.4.1. By Lemma 1.4.5 (applied to
−1
W (K)/I → W (L)) there exists a prime ideal q of W (L) such that pσ = iL/K (q).
Since Z ∼= W (K)/pσ → W (L)/q and again by Theorem 1.4.1, it follows that
W (K)/pσ → W (L)/q is an isomorphism, i.e., q determines an extension τ of σ .
In Sect. 1.3 we proved the injectivity of the map iL/K for extensions L/K that are
of odd degree or that are purely transcendental. Next, we will determine the kernel
of iL/K for quadratic extensions (and so with Proposition 1.4.4 give a new proof of
Proposition 1.3.3):
√
Proposition 1.4.6 Let a ∈ K ∗ \ K ∗2 and L = K( a). Then the kernel of iL/K is
the ideal of W (K) generated by 1, −a.
A more precise statement is:
Proposition 1.4.7 Let q be an anisotropic quadratic form over K. There exist
quadratic forms q , q over K with q ∼ = q ⊥ 1, −a ⊗ q such that qL is
anisotropic.
1.5 Real Closed Fields and Their Field Theoretic Characterization 15
Proof By induction on dim q. The case dim q = 1 is clear. Assume thus that
dim q ≥ 2 and without loss of generality that qL is isotropic. If (V , b) denotes
the bilinear space associated to q, then there exist v, w ∈ V , not both zero, such that
√ √
0 = qL (v + a w) = q(v) + a · q(w) + a · b(v, w).
It follows that q(v)+a q(w) = 0 = b(v, w). Since q is anisotropic, q(v), q(w) = 0.
Let W := Kv + Kw ⊆ V . Then dim W = 2 (if v, w were linearly dependent, then
we would have b(v, w) = 0), and q|W ∼= −ac, c = 1, −a⊗c for c := q(w). In
particular, q|W is nondegenerate. Since V = W ⊥ W ⊥ , we can apply the induction
hypothesis to q|W ⊥ and so conclude the proof.
1.5 Real Closed Fields and Their Field Theoretic

Characterization
After our excursion into the theory of quadratic forms we return to real algebra in
the narrower sense. This section is fundamental for everything that follows.
Definition 1.5.1 A field is called real closed if it is real and has no proper algebraic
extension that is real.
The field R of real numbers is a well-known example of a real closed field.
Proposition 1.5.2 Let K be a field. The following statements are equivalent:
(i) K is real closed;
(ii) there exists an ordering P of K that cannot be extended to any proper algebraic
extension of K;
(iii) K ∗2 ∪ {0} = {a 2 : a ∈ K} is an ordering of K, and every polynomial (in one
variable) of odd degree over K has a zero in K.
Moreover, if these conditions are satisfied, then K ∗2 ∪ {0} is the only ordering of K.
Proof The final comment is clear by (iii) (every ordering P satisfies K ∗2 ∪{0} ⊆ P ).
(i) ⇒ (ii) is clear since a field has an ordering if (and only if) it is real.
(ii)√⇒ (iii): If it were true that K ∗2 ∪ √{0} = P , then there would exist d ∈ P
with d ∈ K, and P would extend to K( d) (Proposition 1.3.3), a contradiction.
Furthermore, K has no proper extensions of odd degree by Proposition 1.3.5.
(iii) ⇒ (i): Let L ⊃ K be a finite proper field extension of K. We will show that
L is not real. By (iii), [L : K] is a power of 2. Since K has an ordering by (iii), it
follows that char K = 0 and in particular that L/K is separable. Let L1 be the Galois
L over K, G = Gal(L1 /K) and H = Gal(L1 /L). Then H is a subgroup
closure of
of G, and g∈G H g = {1} by Galois theory, −1
where H g := g H g. By elementary
g
group theory it follows that |G| divides g∈G [G : H ] . By assumption (iii) this
number is a power of 2, and so G is a 2-group. From Sylow’s Theorems (elementary
group theory again) it follows that there exists a subgroup H of G of index 2 with
H ⊂ H . Let F be the fixed field of H . Then K ⊂ F ⊂ L and [F : K] = 2.
√
There exists a ∈ K with F = K( a). Since K ∗2 ∪ {0} is an √ ordering of K and
a∈ / K ∗2 ∪{0}, there exists b ∈ K ∗ with a = −b2 . Hence −1 = ( a/b)2 is a square
in F , and so F is not real. We conclude that L is not real either.
Remark 1.5.3 Condition (iii) in Proposition 1.5.2 is clearly equivalent to

(iii’) K is real, a or −a is a square in K for every a ∈ K, and every polynomial of
odd degree over K has a zero in K.
In the literature it is common to denote real closed fields with capital letters
R, S, . . . . We will usually follow this convention.
Theorem√ 1.5.4 (Fundamental Theorem of Algebra) If R is a real closed field,
then R( −1) is algebraically closed.
Proof (Gauss) We use Proposition 1.5.2(iii). Every
√ finite extension of R (and so
also every finite extension of C := R(i), i := −1) has 2-power degree. Assume
for the sake of contradiction that C = R(i) is not algebraically closed. As in the
proof of Proposition 1.5.2 it follows that C has an extension F with [F : C] = 2.
Thus, there exist a, b ∈ R such that α := a + bi is not a square in C. Hence, b = 0.
Since R 2 is an ordering of R, there exists c ∈ R with a 2 + b 2 = c2 and c > 0, as
well as x, y ∈ R with x > 0, sign y = sign b and x 2 = (c + a)/2, y 2 = (c − a)/2.
(Note that c ± a are positive and that all signs are of course understood to be with
respect to the unique ordering of R.) Now we have (x + iy)2 = (x 2 − y 2 ) + 2xyi,
and also x 2 − y 2 = a and (2xy)2 = b2 . Since sign
√ (xy) = sign b we have 2xy = b,
and it follows that (x + iy)2 = α, contradicting α ∈ C.
A very strong converse to this theorem is given by Theorem 1.6.1.
Corollary 1.5.5 Let R be a real closed field and let K ⊆ R be a subfield. Then K
is real closed if and only if K is (relatively) algebraically closed in R.
√
Proof Assume that K is algebraically closed in R and let i = −1. Then K(i) is
also algebraically closed in R(i). Indeed, if α = a + bi ∈ R(i) is algebraic over
K(i) (a, b ∈ R), then also over K, and the same holds for α = a − bi, thus also for
a = (α + α)/2 and b = i(α − α)/2. It follows that a, b ∈ K, hence α ∈ K(i). Thus
K(i) is algebraically closed by Theorem 1.5.4 and it follows immediately that K is
real closed.
The converse direction is trivial by definition.
1.6 Galois Theoretic Characterization of Real Closed Fields 17
Another characteristic of real closed fields is their relative rigidity:

Proposition 1.5.6 Let R be a real closed field.
(a) Every (ring) endomorphism ϕ of R is order preserving, i.e., a ≤ b ⇒ ϕ(a) ≤
ϕ(b) for all a, b ∈ R.
(b) If K ⊆ R is a subfield and R/K is algebraic, then the identity is the only K-
automorphism of R.
Proof (a) This follows immediately from R 2 = {a ∈ R : a ≥ 0}.
(b) Let ϕ ∈ AutK (R) and a ∈ R. Since R/K is algebraic, {ϕ n (a) : n ∈ N} is a
finite set. If it were true that ϕ(a) = a, thus for instance ϕ(a) > a, it would follow
from (a) by induction that a < ϕ(a) < ϕ 2 (a) < · · · , a contradiction.
If R is real closed, we will henceforth denote the uniquely determined ordering
of R by ≤ without further comment (as we already did several times above).
1.6 Galois Theoretic Characterization of Real Closed Fields
Let K be a field of arbitrary characteristic with algebraic closure K. This section is

dedicated to a proof of the following beautiful result of Artin and Schreier [4]:
Schreier, 1927) If K = K and [K : K] < ∞, then
Theorem 1.6.1 (E. Artin, O.√
K is real closed and K = K( −1).
Proof The elementary proof that we present here goes back to J. Leicht [75]. We
break it down into several steps. Let n = [K : K].
(1) The field K is perfect.
Assume K is not perfect. Then char K = p > 0, and there exists a ∈ K with
α := a 1/p ∈ K. But then also α 1/p ∈ K(α). Indeed, if α = β p with β ∈ K(α), then
p
NK(α)/K (β) = NK(α)/K (α) = (−1)p−1 a = a since (−1)p−1 = 1. Iterating
this step, we obtain field extensions of K of degree p, p2 , p3 , . . . , contradicting
[K : K] < ∞.
It follows that K/K is a finite Galois extension. Let L be a maximal intermediate
field of K/K, different from K, and let q := [K : L]. (Note that q is prime!)
(2) char K = q.
Assume for the sake of contradiction that char K = p = q. By Artin–Schreier
theory for Galois extensions of degree p in characteristic p we have that K is
L-isomorphic to L[t]/(t p − t − a) for some a ∈ L (see, for example, [96,
Theorem 12.2.1], [73, VI, Theorem 6.4] or [55, Vol. II, §8.11]). We only need the
consequence that ψ(L) = L for the map ψ : K → K defined by ψ(b) := bp − b.
Note that ψ is additive, i.e., ψ(b + b ) = ψ(b) + ψ(b ). Consider the trace
tr = trK/L : K → L. Then tr ◦ψ = (ψ|L )◦tr. Indeed, if b ∈ K and if b1 , . . . , bp are
the L-conjugates of b in K, then ψ(b1 ), . . . , ψ(bp ) are the L-conjugates of ψ(b),
and it follows that tr (ψ(b)) = ψ(bi ) = ψ( bi ) = ψ(tr b). Hence the diagram
i i
commutes. Since ψ and tr are surjective (the second map since K/L is separable),
the map ψ|L : L → L is also surjective, a contradiction.
(3) q = 2 (and char K = 2).
L contains the q-th roots of unity since [K : L] = q and the q-th cyclotomic
polynomial is of degree q − 1. Hence, since char K = q, there exists a ∈ L with
2
K = L(a 1/q ) (see for instance [55, Vol. I, §4.7] or [73, VI, §6]). For β := a 1/q we
obtain as in (1) that (NK/L (β))q = (−1)q−1 a. Since a is not a q-th power in L, q
must be even, thus q = 2.
√
(4) K = L( −1).
√
Let i = −1 ∈ K, and let a ∈ L and β ∈ K as in (3), thus with √ β = a
4
and K = L(β ). Since NK/L (β) = −β , we have β /NK/L (β) = −1, and so
2 2 4 2
√
K = L( −1).
(5) L = K, thus K = K(i).
If K = L, then also K(i) = K. Carrying out steps (1) to (4) for K(i) instead of
K leads to a contradiction by (4).
(6) K is real (and thus real closed).
Since i ∈ K it suffices to show that the sum of two squares in K is a square
(Proposition 1.5.2). Thus, let a, b ∈ K ∗ . Since K is algebraically closed, there are
c, d ∈ K with a+bi = (c+di)2. Hence, a−bi = (c−di)2 and a 2 +b 2 = (c2 +d 2 )2 .
This proves the theorem.
Let Ks denote the separable closure of K.
Corollary 1.6.2 If char K = 0 and [Ks : K] < ∞, then K = Ks .
Proof Steps (2)–(6) of the proof can be carried out for Ks just as for K.
We finish this section with a consequence of the theorem of Artin and Schreier
for absolute Galois groups of fields. These are profinite groups (i.e., projective
limits of finite groups). For the basic aspects of infinite Galois theory, [54, Vol. III,
§IV.2], [55, Vol. II, §8.6] or [12, Ch. V, §10], for example, can be consulted, but for
understanding what follows this is not required.
1.7 Counting Real Zeroes of Polynomials (without Multiplicities) 19
Corollary 1.6.3 Let K be a field with separable algebraic closure Ks and absolute
Galois group = Gal (Ks /K). All elements of finite order > 1 in are involutions
(i.e., have order 2), and the map τ → Fix (τ ) (the fixed field of τ in Ks = K) is a
bijection from the set of involutions in to the set of real closed overfields of K in
K. In particular, contains elements of finite order > 1 if and only if K is real.
1.7 Counting Real Zeroes of Polynomials (without

Multiplicities)
One of the oldest problems in real analysis consists of determining the number
and location of the real zeroes of a polynomial f (t) with real coefficients. The
first general answer was formulated by J. C. F. Sturm (1803–1855) in the form of
an algorithm, the so-called Sturm sequence, that determines the number of zeroes
of f in any given interval, counted without multiplicities. Another solution of
the problem, based on the examination of certain quadratic forms, was found by
Ch. Hermite (1822–1901). Both methods will be presented in this section.
It should be remarked that these results used to be part of the mathematics
curriculum at the turn of the nineteenth century (see for example H. Weber’s
Lehrbuch der Algebra [114]).
Sturm’s solution, found in 1829, initiated an historically very interesting and to
some extent stormy period in real algebra. (Sturm published the full proof only
in 1835.) Hermite’s theorem (Theorem 1.7.17) was anticipated by C.G.J. Jacobi
(1804–1851) and his student and friend C.W. Borchardt (1817–1880) (Theo-
rem 1.7.15), and was proved in 1853 by Hermite and—essentially independently—
also by J.J. Sylvester (1814–1897). In addition, Sylvester, A. Cayley (1821–1895)
and Sturm established interesting links between the theorems of Sturm and Hermite.
In addition to the original literature, the reader can find full particulars of this history
in the substantial text [65] of Krein and Naimark, as well as in [39].
We start with a number of classical results, well-known from real analysis, that
stay valid for rational functions over arbitrary real closed fields.
In the remainder of this section R denotes a real closed field.
Proposition 1.7.1 Let f (t) = t n +a1 t n−1 +· · ·+an ∈ R[t] be a monic polynomial.
Then all real zeroes of f in R) are in the interval [−M, M],
of f (i.e., all zeroes
where M := max 1, |a1 | + · · · + |an | .
Proof If 0 = a ∈ R is such that f (a) = 0, then a = −(a1 + a2a −1 + · · ·+ an a 1−n ),
and so |a| ≤ 1 or |a| ≤ |a1 | + · · · + |an | by the triangle inequality.
Proposition 1.7.2 (Intermediate Value Theorem) Let f ∈ R(t) and a, b ∈ R

with a < b, such that f has no pole in [a, b]. If f (a)f (b) < 0, then f has an odd
number of zeroes in ]a, b[, counted with multiplicities.
Proof If f = g/ h with polynomials g, h, where h is nonzero on [a, b], then
we may replace f by f h2 = gh, and thus assume that 0 = f ∈ R[t]. Let
a1 ≤ · · · ≤ ar be the real roots of f , counted with multiplicities (r ≥ 0).
By the Fundamental Theorem of Algebra (Theorem 1.5.4) there exist c ∈ R ∗
and irreducible monic quadratic polynomials p1 , . . . , ps (s ≥ 0) with f (t) =
c · (t − a1 ) · · · (t − ar ) p1 (t) · · · ps (t). Since the pk only take positive values, we
have
f (a)
r
a − aj
−1 = sign = sign .
f (b) b − aj
j =1
The number of j with a < aj < b is therefore odd.
Lemma 1.7.3 Let 0 = f ∈ R(t) be a rational function and a ∈ R a zero of f . The

logarithmic derivative f /f of f changes sign from minus to plus at t = a (it has a
pole at t = a), i.e., there exists ε > 0 such that f /f is negative on ]a − ε, a[ and
positive on ]a, a + ε[.
Proof There exist g, h ∈ R[t] with g(a)h(a) = 0 and f (t) = (t − a)d g(t)/ h(t)
for some d ∈ N. Then
f (t) d g (t) h (t)

= + − ,
f (t) t −a g(t) h(t)
from which the statement follows.
Proposition 1.7.4 Let 0 = f ∈ R(t) and a, b ∈ R with a < b and f (a) = f (b) =
0. If f has neither poles nor zeroes in (a, b), then the number of zeroes of f in
]a, b[, counted with multiplicities, is odd.
Corollary 1.7.5 (Rolle’s Theorem) If f ∈ R(t), f (a) = f (b) ∈ R for a, b ∈

R, a < b, and if f has no poles in [a, b], then f has a zero in ]a, b[.
Proof of Proposition 1.7.4 Let ε > 0 be such that f neither has zeroes in
]a, a + ε], nor in [b − ε, b[, and such that a + ε < b − ε. By Lemma 1.7.3 we
have
f (a + ε) f (b − ε)
· < 0,
f (a + ε) f (b − ε)
and by Proposition 1.7.2 the number of zeroes of f /f in ]a + ε, b − ε[ is odd. This

number equals the number of zeroes of f in ]a, b[.
In what follows we assume that f ∈ R[t] is a fixed non-constant polynomial and

that a, b ∈ R with a < b and f (a)f (b) = 0. Unless stated otherwise, zeroes will
be counted without multiplicities.
We consider
Sturm’s Problem Determine the number of distinct real zeroes of the polynomial
f in the interval [a, b].
Definition 1.7.6 The Sturm sequence of f is the sequence (f0 , f1 , . . . , fr ) of

polynomials, defined recursively as follows: let f0 = f, f1 = f and
f0 = q1 f1 − f2 ,
f1 = q2 f2 − f3 ,
..
.
fr−2 = qr−1 fr−1 − fr ,
fr−1 = qr fr ,
where r ≥ 1, f0 , . . . , fr , q1 , . . . , qr ∈ R[t], fi = 0 and deg fi < deg fi−1 for

i = 2, . . . , r. These conditions determine r, as well as the fi (and qi ) uniquely.
If we change the minus signs on the right hand side into plus signs, we obtain
the usual version of Euclid’s algorithm for the pair f, f . This sign difference is
essential for the determination of the Sturm sequence, but plays no role in the
computation of the greatest common divisor. In particular, fr = gcd(f, f ).
Notation 1.7.7 For (c0 , . . . , cr ) ∈ R r+1 we denote by Var(c0 , . . . , cr ) the num-
ber of sign changes in the given sequence, after cancelling all zeroes. We let
Var(0, . . . , 0) := −1. For example, for the sequence (1, 0, −2, 1, 0, 4, −1, 1, 0)
we obtain Var = 4. For x ∈ R we let V (x) := Var (f0 (x), f1 (x), . . . , fr (x)), where
(f0 , . . . , fr ) is the Sturm sequence of f .
Theorem 1.7.8 (Sturm, 1829) Let f ∈ R[t] be a non-constant polynomial, and

a, b ∈ R with a < b and f (a)f (b) = 0. Then the number of distinct zeroes of f in
the interval ]a, b[ equals V (a) − V (b).
Proof We proceed in two steps.
(1) Let (g0 , . . . , gr ) be a sequence in R[t] \ {0} that satisfies properties (a)–(d)
below:
(a) g0 (a)g0 (b) = 0;
(b) gr (a)gr (b) = 0, and gr is semidefinite on [a, b] (i.e., either gr (x) ≥ 0 or
gr (x) ≤ 0 for all x ∈ [a, b]);
(c) if c ∈ [a, b] and g0 (c) = 0, then g0 (x)g1 (x) changes sign from minus to
plus at x = c;
(d) if c ∈ [a, b] and gi (c) = 0, 0 < i < r, then gi−1 (c)gi+1 (c) < 0.
For x ∈ R let W (x) := Var (g0 (x), . . . , gr (x)). We claim that g0 has exactly W (a)−
W (b) distinct zeroes in [a, b].
To prove the claim, we first note that for every c ∈ R with g0 (c) · · · gr (c) = 0
the function W (x) is constant on a neighbourhood of x = c. Let g := g0 · · · gr , and
let c ∈ [a, b] be a zero of g. We distinguish three cases:
If g0 (c) = 0, then a < c < b, and the function x → Var (g0 (x), g1 (x)) has value
1 on ]c − ε, c[ and value 0 on [c, c + ε[, for some ε > 0. This follows from (c).
If gi (c) = 0, 0 < i < r, then by (d) there exists ε > 0, such that gi−1 (x)gi+1 (x)
is negative on ]c − ε, c + ε[. Hence, x → Var (gi−1 (x), gi (x), gi+1 (x)) is constant
equal to 1 on ]c − ε, c + ε[ (note that for every α ∈ R, Var(−1, α, 1) =
Var(1, α, −1) = 1 !).
If gr (c) = 0, and g0 (c) = 0 or r > 1, then gr−1 (c) = 0 by (d), and x →
Var (gr−1 (x), gr (x)) is constant on {x : 0 < |x − c| < ε} for some ε > 0, since gr
is semidefinite on [a, b].
From these three cases we obtain for all c ∈ [a, b]: If g0 (c) gr (c) = 0, then W (x)
is constant on a neighbourhood of x = c; if g0 (c) = 0, then W (x) is constant on
a punctured neighbourhood of x = c; if g0 (c) = 0, there exists ε > 0 and N ∈ Z
with W (x) = N for c − ε < x < c and W (x) = N − 1 for c < x < c + ε.
The claim then follows since g0 and gr do not vanish at the boundary points a, b.
(2) Now let f be as in the statement of the theorem and let (f0 , . . . , fr ) be the
Sturm sequence of f . We let gi := fi /fr ∈ R[t] (i = 0, . . . , r) and W (x) :=
Var (g0 (x), . . . , gr (x)) (x ∈ R). Then the sequence (g0 , . . . , gr ) satisfies (a)–(d)
from (1). Indeed, since gi−1 = hi gi − gi+1 (i = 1, . . . , r − 1) and gr−1 = hr gr for
suitable hi ∈ R[t], and since gr = 1, it follows that gi−1 and gi have no zeroes in
common (i = 1, . . . , r), and so (d) follows recursively. (a) and (b) are clear, and (c)
follows from Lemma 1.7.3 since g0 g1 = ff /fr2 .
Furthermore, g0 and f0 = f have the same zeroes (if fr (c) = 0, then also
f (c) = 0, and the zero c occurs with larger multiplicity in f than in fr !). It thus
follows from (1) that f has precisely W (a) − W (b) distinct zeroes in [a, b]. The
proof then follows from the observation that V (a) = W (a) and V (b) = W (b) since
fr (a) fr (b) = 0.
We state a more general version of Sturm’s Theorem, also due to Sturm. We make
use of
Definition 1.7.9 Let f ∈ R[t] and a, b ∈ R with a < b and f (a)f (b) = 0. A
generalized Sturm sequence of f on [a, b] is a sequence (f0 , . . . , fr ) in R[t] \ {0}
with r ≥ 1 that satisfies the properties
(1) f0 = f ;
(2) fr (a)fr (b) = 0, and fr is semidefinite on [a, b];
(3) for 0 < i < r and c ∈ [a, b] with fi (c) = 0, we have fi−1 (c)fi+1 (c) < 0.
(Warning: The Sturm sequence of f is usually not a generalized Sturm sequence of
f in the sense of Definition 1.7.9!)
Let f ∈ R[t] and a, b as in Definition 1.7.9. Let (f0 , . . . , fr ) be a generalized

Sturm sequence of f on [a, b] and, once again, V (x) := Var (f0 (x), . . . , fr (x))
(x ∈ R).
Theorem 1.7.10 (Sturm)

V (a) − V (b) = # c ∈ ]a, b[ : f (c) = 0 and ff1 changes sign from − to + at c

− # c ∈ ]a, b[ : f (c) = 0 and ff1 changes sign from + to − at c .
The statement is clear by the proof of Theorem 1.7.8! (We only have to count the
different types of sign changes of ff1 .)
Remark 1.7.11 From the proof we see that simplifications of the original Sturm
algorithm are possible. For example:
(a) The original Sturm sequence f0 = f, f1 = f , . . . may be terminated at fs
provided that fs (a) fs (b) = 0 and fs is semidefinite on [a, b].
(b) For a remainder fi+1 in the modified Euclidean algorithm (Definition 1.7.6) it
is allowed to leave out those factors that are semidefinite on [a, b] and that do
not vanish at a, b.
Exercise 1.7.12 Let (K, ≤) be an ordered field, and let

n
n
f = tn + ai t n−i = (t − ξj )
i=1 j =1
be a monic polynomial in K[t] that splits over K (with ai , ξj ∈ K). Prove that
ξ1 , . . . , ξn ≥ 0 ⇔ (−1)i ai ≥ 0 for i = 1, . . . , n,
and similarly with strict instead of non-strict inequalities.

Next, we discuss Hermite’s method. Let K be an arbitrary field and f (t) =
t n + a1 t n−1 + · · · + an ∈ K[t] a monic polynomial (n ≥ 1). Let α1 , . . . , αn be the
zeroes of f in the algebraic closure of K. For r = 0, 1, 2, . . . , let sr = sr (f ) :=
α1r + · · · + αnr . Then sr (f ) ∈ K since sr (f ) is symmetric in the αi . For instance,
s0 (f ) = n, s1 (f ) = −a1 , s2 (f ) = a12 − 2a2 , s3 (f ) = −a13 + 3a1 a2 − 3a3, and so
on.
Exercise 1.7.13 Prove the Newton identity
sk + sk−1 a1 + sk−2 a2 + · · · + s1 ak−1 + kak = 0
for all k ≥ 0, where we put ak = 0 for k > n. Conclude for r ≥ 0 that sr =

sr (a1 , . . . , an ) is a polynomial in a1 , . . . , an with integer coefficients.
Definition 1.7.14 The Sylvester form S(f ) of f is the n-dimensional quadratic

form over K defined by the n × n-matrix
sj +k−2 (f ) 1≤j,k≤n
.
n
In other words, S(f )(x1 , . . . , xn ) = sj +k−2 (f )xj xk .
j,k=1
Theorem 1.7.15 (Jacobi, Borchardt, Hermite) Let R be a real closed field and
f ∈ R[t] a monic non-constant polynomial. Then the rank of √ S(f ) equals the
number of distinct roots of f in the algebraic closure C = R( −1) of R, and
the signature of S(f ) equals the number of distinct real roots of f .
Note that therefore S(f ) always has nonnegative signature!
Proof Let β1 , . . . , βr be the pairwise distinct real roots of f , and γ1 , γ1 , . . . , γs , γs
the pairwise distinct non-real roots of f , where r, s ≥ 0 and α → α denotes
the nontrivial R-automorphism of√C. Denote the multiplicity of βj by mj and the
multiplicity of γj by nj . Let i = −1 ∈ C, and let Re α, Im α for α ∈ C have their
usual meaning. Then

S(f ) (x1 , . . . , xn ) = αjk+l−2 xk xl
1≤j,k,l≤n
n
n 2
= αjk−1 xk
j =1 k=1

r 2
s 2 2
= mj βjk−1 xk + nj γjk−1 xk + γj k−1 xk
j =1 k j =1 k k

r 2
= mj βjk−1 xk
j =1 k

s 2 2
+2 nj Re(γjk−1 )xk − Im(γjk−1 )xk ,
j =1 k k
and we obtain S(F ) in diagonal form. Indeed, letting

uj (x) := βjk−1 xk (j = 1, . . . , r),
k

vj (x) := Re(γjk−1 )xk , wj (x) := Im(γjk−1 )xk (j = 1, . . . , s),
k k
we have

r
s
S(f ) = mj u2j + 2 nj (vj2 − wj2 ),
j =1 j =1
and the linear forms u1 , . . . , ur , v1 , w1 , . . . , vs , ws are linearly independent.

(The transition matrix arises from the Vandermonde matrix associated to
β1 , . . . , βr , γ1 , γ1 , . . . , γs , γs via two elementary transformations.) It follows that
S(f ) ∼
= r × 1 ⊥ s × H ⊥ (n − r − 2s) × 0,
from which we immediately see that rank S(f ) = r + 2s, sign S(f ) = r.
The following modification makes it possible to also count real zeroes in
intervals:
Definition 1.7.16 Let K be a field, f ∈ K[t] a monic polynomial of degree n ≥ 1,
and λ ∈ K. The Sylvester form of f with parameter λ is the quadratic form

(x1 , . . . , xn ) → λ sj +k−2 (f ) − sj +k−1 (f ) xj xk
1≤j,k≤n
over K, denoted by Sλ (f ).
Theorem 1.7.17 (Hermite, Sylvester, 1853) Let R be real closed and f ∈ R[t] a
monic polynomial of degree n ≥ 1. For λ ∈ R we have
√
rank Sλ (f ) = # a ∈ R( −1) : f (a) = 0 and a = λ ,

sign Sλ (f ) = # a ∈ R : f (a) = 0, a < λ − # a ∈ R : f (a) = 0, a > λ .
Proof We use the notation from the previous proof.

Sλ (f ) (x1 , . . . , xn ) = (λ − αj )αjk+l−2 xk xl
1≤j,k,l≤n

n 2
= (λ − αj ) αjk−1 xk
j =1 k

r
= mj (λ − βj )u2j
j =1

s
+ nj (λ − γj )(vj + iwj )2 + (λ − γj )(vj − iwj )2 .
j =1
We rewrite [· · · ] as follows:
[· · · ] = (λ − γj + λ − γj )(vj2 − wj2 ) + 2i(λ − γj − λ + γj )vj wj
= 2 (λ − Re γj )(vj2 − wj2 ) + 4 Im (γj )vj wj

=: 2ϕj (vj , wj ).
Hence,

r
s
Sλ (f ) = mj (λ − βj )u2j + 2nj ϕj (vj , wj ).
j =1 j =1
The forms ϕj are hyperbolic (j = 1, . . . , s). Indeed, if γj = aj + ibj with aj , bj ∈

R, then ϕj is represented by the matrix

λ − aj bj
bj −(λ − aj )
which has negative determinant since bj = 0. We conclude that
Sλ (f ) ∼
= λ − β1 , . . . , λ − βr ⊥ s × H ⊥ (n − r − 2s) × 0.
1.8 Conceptual Interpretation of the Sylvester Form
Let K be a field of characteristic not 2.

In this section we will present a more conceptual interpretation of the Sylvester
form of a polynomial, namely as the trace form of a certain algebra. We start with
generalities about trace forms of algebras.
By a K-algebra we mean a commutative (and associative) unital finite-
dimensional algebra over K. Let A be a K-algebra. For a ∈ A, let λ(a) be the
K-linear endomorphism b → ab of A, and trA/K (a) the trace of λ(a) (over K).
The K-linear form trA/K : A → K is called the trace of A over K.
The following rules are easily verified:
Lemma 1.8.1
(a) If K /K is a field extension and A = A ⊗K K , then
trA/K (a) = trA /K (a ⊗ 1) for all a ∈ A.

1.8 Conceptual Interpretation of the Sylvester Form 27
(b) If A1 , . . . , Ar are K-algebras and A = A1 × · · · × Ar , then

r
trA/K (a1 , . . . , ar ) = trAi /K (ai ) for all (a1 , . . . , ar ) ∈ A.
i=1
(c) trA/K (Nil A) = 0 (where Nil A = {a ∈ A : a n = 0 for some n ∈ N} is the

nilradical of A).
Assume now that A = K[t]/(f ), where f ∈ K[t] is a monic polynomial of
degree n ≥ 1. We denote the image of t in A by τ . Let α1 , . . . , αn be the roots of f
in the algebraic closure K of K.
n
Proposition 1.8.2 For g ∈ K[t] we have trA/K (g(τ )) = g(αj ).
j =1
Proof By Lemma 1.8.1(a) we may assume K = K. Let β1 , . . . , βr be the distinct

roots of f and e1 , . . . , er their multiplicities, i.e., f (t) = (t − β1 )e1 · · · (t − βr )er .
Let Ai = K[t]/(t −βi )ei (i = 1, . . . , r). Then A → A1 ×· · ·×Ar is an isomorphism
by the Chinese Remainder Theorem, and by Lemma 1.8.1(b) we may assume r = 1,
i.e., f (t) = (t − β)e . There exists h ∈ K[t] such that g(t) = g(β) + (t − β)h(t).
Since (τ − β)h(τ ) ∈ Nil A, and using Lemma 1.8.1(c), we conclude trA/K (g(τ )) =
trA/K (g(β)) = e · g(β).
Notation 1.8.3
(a) Let A be a K-algebra and a ∈ A. Then [A, a]K denotes the quadratic space
over K defined by the quadratic form x → trA/K (ax 2 ) on A.
(b) For polynomials f, g ∈ K[t], f = 0, let
SylK (f ; g) := K[t]/(f ), g(τ ) K
and
SylK (f ) := SylK (f ; 1).
We usually drop the index K in the notation.

Lemma 1.8.1 immediately yields the following computational rules for these
forms:
Lemma 1.8.4 Let A, A , A1 , . . . , Ar be K-algebras.
(a) If ϕ : A → A is a K-algebra isomorphism, then
[A, a] ∼
= A , ϕ(a) for all a ∈ A.
(b) If K /K is a field extension then, for all a ∈ A,
[A, a]K ⊗K K ∼
= [A ⊗K K , a ⊗ 1]K (over K ).
(c) A1 × · · · × Ar , (a1 , . . . , ar ) ∼
= [A1 , a1 ] ⊥ · · · ⊥ [Ar , ar ] (ai ∈ Ai ).
(d) Nil A ⊆ Rad[A, a] for all a ∈ A.
Lemma 1.8.5 Let f, g ∈ K[t] with f = 0.

(a) If K /K is a field extension, then
SylK (f ; g) ⊗K K ∼
= SylK (f ; g) (over K ).
(b) Let f1 , . . . , fr ∈ K[t] be nonzero and pairwise relatively prime, then
Syl(f1 · · · fr ; g) ∼
= Syl(f1 ; g) ⊥ · · · ⊥ Syl(fr ; g).
(c) Syl(f e ; g) ∼
= e ⊗ Syl(f ; g) ⊥ 0, . . . , 0 (e ∈ N).
(d) Syl(af ; g) = Syl(f ; g) and Syl(f ; ag) ∼ = a ⊗ Syl(f ; g) for any a ∈ K ∗ .
(e) Syl (t − a; g(t)) = g(a) for any a ∈ K.
Proof (a) and (b) follow from Lemma 1.8.4, and (d) and (e) are clear. We show (c):
let A = K[t]/(f e ), A = K[t]/(f ), and π : A → A the residue class map. By
Proposition 1.8.2 we have trA/K (a) = e · trA /K (πa) for all a ∈ A, from which the
proof follows.
Corollary 1.8.6 Let f, g ∈ K[t] with f = 0. The following statements are

equivalent:
(i) SylK (f ; g) is nondegenerate;
(ii) f is separable, and f and g are relatively prime.
Proof Since (i) and (ii) are not affected by field extensions, we may assume that
K = K is algebraically closed. Let α1 , . . . , αr be the distinct zeroes of f , with
respective multiplicities e1 , . . . , er ∈ N. By Lemma 1.8.5 we have
r
Syl(f ; g) ∼
= ⊥ Syl (t − α )
j =1
j
ej
; g(t)
r
∼
= ⊥
j =1
ej ⊗ Syl t − αj ; g(t) ⊥ (ej − 1) × 0
r
∼
= ⊥
j =1
ej g(αj ) ⊥ (ej − 1) × 0 ,
from which the statement follows.

1.8 Conceptual Interpretation of the Sylvester Form 29
Proposition 1.8.7 Let f ∈ K[t] be a monic polynomial of degree n ≥ 1. The

Sylvester form S(f ) of f satisifies S(f ) ∼ = Syl(f ), whereas the Sylvester form
Sλ (f ) of f with parameter λ satisfies Sλ (f ) ∼
= Syl(f ; λ − t).
Proof Let α1 , . . . , αn be the zeroes of f in K and let g ∈ K[t] be arbitrary. With
respect to the basis 1, τ, . . . , τ n−1 of A := K[t]/(f ), the quadratic space Syl(f ; g)
is represented by the matrix (bj k )1≤j,k≤n , with bj k = trA/K τ j +k−2 g(τ ) . We have
j +k−2
bj k = ni=1 αi g(αi ) by Proposition 1.8.2, from which the statements follow
with g = 1 and g = λ − t, respectively.
We finish this section by showing that the theorems of Sylvester and Hermite
(Theorems 1.7.15 and 1.7.17) are quite easily obtained by means of trace form
calculus.
To this end, assume that K = R is real closed, and that f ∈ R[t] is monic
of degree n ≥ 1 with distinct real roots β1 , . . . , βr and distinct non-real roots
γ1 , γ1 , · · · , γs , γs . Assume again that the βj and γj have multiplicities mj and nj ,
respectively. We let qj (t) := (t − γj )(t − γj ) (j = 1, . . . , s). Then

r
s
f (t) = (t − βj )mj qj (t)nj ,
j =1 j =1
and by Lemma 1.8.5 it follows that for all g ∈ R[t],

r s
Syl(f ; g) ∼
= ⊥
(b) j =1
Syl (t − βj )mj ; g ⊥ ⊥ Syl(q
j =1
nj
j ; g)
r s
∼
= ⊥ Syl(t − β ; g) ⊥ ⊥ Syl(q ; g) ⊥ 0, . . . , 0
(c) j =1
j
j =1
j
s
∼
= g(β1 ), . . . , g(βr ) ⊥
(e)
⊥ Syl(q ; g) ⊥ 0, . . . , 0.
j =1
j
The forms√ Syl(qj ; g) are all hyperbolic (or zero in case qj | g) since R[t]/(qj ) and
C = R( −1) are isomorphic as R-algebras, and [C; α]R ∼ = 1, −1 = H for all
α ∈ C ∗ . (Choosing β ∈ C with 2αβ 2 = i, the form [C; α]R is represented by the
matrix 01 10 with respect to the basis β, −iβ.)
It follows that
Syl(f ; g) ∼
= g(β1 ), . . . , g(βr ) ⊥ t × H ⊥ (n − r − 2t) × 0
for some 0 ≤ t ≤ s, and in particular that
S(f ) ∼
= Syl(f ) = Syl(f ; 1) ∼
= r × 1 ⊥ s × H ⊥ (n − r − 2s) × 0
and
Syl(f ; λ − t) ∼
= λ − β1 , . . . , λ − βr ⊥ s × H ⊥ (n − r − 2s) × 0,
from which the statements of the theorems of Sylvester and Hermite can be read off
directly.
In addition to the works [65] and [39], mentioned before, the reader may want
to compare the current topic and related themes with the article [10] that contains
interesting historical remarks about Sylvester.
1.9 Cauchy Index of a Rational Function, Bézoutian and

Hankel Forms
This section, and the next one, will not be used in the remainder of this book and can
therefore be skipped during a first reading. Having said that, its content is a classical
part of real algebra, and is essential for understanding the historical development
of mathematics in the nineteenth century. Furthermore, it has become important in
many applications such as the stability of motion or control theory.
We will follow the paper [65] of Krein and Naimark, but can only cover a small
part for lack of space. The interested reader may therefore want to consult the
original source.
R will always denote a real closed field.
Definition 1.9.1 Let ϕ(t) = g(t)/f (t) ∈ R(t) be a rational function with g and f
relatively prime polynomials over R.
(a) Let α ∈ R be a pole of ϕ (i.e., a zero of f ). The Cauchy index indα (ϕ) of ϕ at
α is defined as follows:
⎧
⎪
⎨+1 if the value of ϕ(t) jumps from −∞ to +∞ at t = α
⎪
indα (ϕ) := −1 if the value of ϕ(t) jumps from +∞ to −∞ at t = α .
⎪
⎪
⎩ 0 otherwise
(b) The Cauchy index of ϕ in an open interval ]a, b[ ⊆ R (a and b are allowed to be
±∞), denoted Iab (ϕ), is defined as the sum of the Cauchy indices of ϕ at those
poles of ϕ that are in ]a, b[. If a = −∞ and b = +∞, we speak of the global
Cauchy index of ϕ. (Note that if ϕ has a pole at ∞, this pole is ignored for the
Cauchy index, no matter what a and b are.)
1.9 Cauchy Index of a Rational Function, Bézoutian and Hankel Forms 31
The Cauchy index of ϕ(t) = f (t)/g(t) at the pole t = α is thus +1, −1 or 0,

depending on whether the polynomial f (t)g(t) changes sign from minus to plus,
from plus to minus, or not at all, at t = α.
Examples 1.9.2
(1) Let f ∈ R[t] be a non-constant polynomial with logarithmic derivative
ϕ = f /f . Let α1 , . . . , αr be the distinct real zeroes of f with respective
multiplicities m1 , . . . , mr . Then

r
mi
ϕ(t) = ψ(t) + ,
t − αi
i=1
for some rational function ψ without real poles. Thus ϕ = f /f has precisely
r real poles α1 , . . . , αr and its Cauchy index at each one of them is +1.
(2) More generally: let f, h ∈ R[t] with f = 0. For ϕ = hf /f we have
+∞
I−∞ (ϕ) = # α ∈ R : f (α)= 0 and h(α) > 0 − # α ∈ R : f (α)= 0 and h(α)< 0 ,
as can easily be verified.

How can we determine the integer Iab (ϕ) from the coefficients of ϕ? A solution
was already presented in Sect. 1.7: Carry out Sturm’s algorithm for the polynomials
f, g (instead of f, f ) and apply Theorem 1.7.10 to the generalized Sturm sequence
thus obtained. This procedure was already developed in the nineteenth century as
a practical method for finding Iab (ϕ). For an important special case, the so-called
Routh–Hurwitz Problem, we refer to [40, §16].
Next we will establish a connection between the global Cauchy index and certain
quadratic forms. This will require some preparation. For ϕ = g/f we may always
assume that deg g ≤ deg f since the Cauchy indices do not change after adding a
polynomial.
Consider thus two polynomials
f (t) = a0 t n + a1 t n−1 + · · · + an and g(t) = b0 t n + b1 t n−1 + · · · + bn
with coefficients in a field K (char K = 2) and let a0 = 0. To the pair (f, g) we

associate two quadratic forms over K as follows.
Let x and y be new variables. The polynomial f (x)g(y) − f (y)g(x) is divisible
by x − y. More precisely,
f (x)g(y) − f (y)g(x)
n−1
= cij x i y j ,
x−y
i,j =0
where

i
cij = dk,i+j +1−k with dij := an−j bn−i − an−i bn−j .
k=0
Independently from these explicit formulas (for which we will have no further use),
it is clear that cij = cj i for all i, j = 0, . . . , n − 1.
Definition 1.9.3 The symmetric n × n-matrix (cj i )0≤i,j ≤n−1 is called the Bézout
matrix of f and g, and is denoted B(f, g). The associated quadratic form over K is
called the Bézoutian of f , also denoted B(f, g). Thus,

n−1
B(f, g) (x0 , . . . , xn−1 ) = cij xi xj .
i,j =0
Since deg g ≤ deg f we can write the rational function ϕ(t) = g(t)/f (t) as a
formal power series in t −1 :
ϕ(t) = s−1 + s0 t −1 + s1 t −2 + · · · .
Definition 1.9.4 The symmetric n × n-matrix (si+j )0≤i,j ≤n−1 is called the Hankel
matrix of f and g and is denoted H (f, g). (In general, any quadratic matrix (aij )
whose entry aij only depends on i +j is called a Hankel matrix.) Again we interpret
H (f, g) as a quadratic form,

n−1
H (f, g) (x0 , . . . , xn−1 ) = si+j xi xj ,
i,j =0
and then refer to it as the Hankel form of f and g. For 1 ≤ p ≤ n the truncated
matrix (si+j )0≤i,j ≤p−1 is denoted Hp (f, g).
Remark 1.9.5 H (f, g) clearly only depends on n and the rational function ϕ =
g/f . More generally, a Hankel matrix Hn (ϕ) as in Definition 1.9.4 can be associated
to any formal power series ϕ(t) ∈ K[[t −1 ]] in t −1 and any n ≥ 0. A well-known
(elementary) theorem of Frobenius says that ϕ(t) is rational (i.e., ϕ(t) ∈ K(t)) if
and only if there exists n ≥ 0 such that det Hm (ϕ) = 0 for all m ≥ n. Furthermore,
in this case the smallest such n is the degree of the denominator of ϕ, written as
an irreducible fraction, and every matrix Hm (ϕ) with m > n has rank n. See, for
example, [40, §16.10].
Example 1.9.6 We will compute the Hankel form H (f, f ), where f ∈ K[t] is a
non-constant polynomial. Let f (t) = a(t − α1 ) · · · (t − αn ) be the factorization of
f over the algebraic closure of K, then (geometric series!)
∞
f (t) 1
n n
sj t −j −1
j
= = with sj = αk .
f (t) t − αi
i=1 j =0 k=1
Hence, H (f, f ) is the Sylvester form S(f ) of f , considered in Sect. 1.7.
Remark Concerning the Terminology We will see shortly that H (f, g) is

isometric to B(f, g). Some modern papers and books (e.g., [11]) therefore refer
to the form S(f ) = H (f, f ) as the “Bézoutian of f ”. We feel that this usage is
unfortunate, as it is not in harmony with the terminology developed in the nineteenth
century. Back then, computations were more concrete than is common nowadays
and there was a strict distinction between a quadratic form and its isometry class.
In order to avoid such clashes we called S(f ) the Sylvester form of f , which is
certainly justified given Sylvester’s pioneering work in this area.
Proposition 1.9.7 The Bézoutian B(f, g) is isometric to the Hankel form H (f, g).
Proof As usual, let ϕ(t) = g(t)/f (t). Then

n−1
ϕ(y) − ϕ(x)
cij x i y i = f (x)f (y)
x−y
i,j =0
∞
y −i − x −i
= f (x)f (y) si−1
x−y
i=0
∞
i−1
= f (x)f (y) si−1 x j −i y −j −1
i=1 j =0
∞

= f (x)f (y) sk+l x −(k+1) y −(l+1)
k,l=0
∞

= sk+l (a0 x n−k−1 + · · · + an x −k−1 ) (a0 y n−l−1 + · · · + an y −l−1 ).
k,l=0
Since there are no negative powers of x or y on the left hand side, we may omit all
such terms on the right hand side and obtain

n−1
n−1
cij x i y j = sk+l (a0 x n−k−1 + · · · + an−k−1 ) (a0 y n−l−1 + · · · + an−l−1 ).
i,j =0 k,l=0
Via the transformation of variables
u0 := a0 xn−1 + a1 xn−2 + · · · + an−1 x0

u1 := a0 xn−2 + · · · + an−2 x0
.. ..
. .
un−1 := a0 x0
we then get
B(f, g) (x0 , . . . , xn−1 ) = H (f, g) (u0 , . . . , un−1 ). (1.2)
Since a0 = 0 the theorem is proved.
Corollary 1.9.8 Consider the minors
Bp := det(cij )n−p≤i,j ≤n−1 (1 ≤ p ≤ n)
of the Bézout matrix B(f, g) (the “principal minors from the bottom right up”).
Then
2p
Bp = a0 det Hp (f, g). (1.3)
Proof In the identity (1.2) we let x0 = · · · = xn−p−1 = 0, i.e., up = · · · =

un−1 = 0, and then read the new identity as an equation of symmetric matrices.
Taking determinants finishes the proof.
Theorem 1.9.9 (Hurwitz)

a0 b0 0 0 · · · 0 0

a1 b1 a0 b0 · · · 0 0

Bp = . .. .. .. .. .. , (1.4)
.. . . . . .

a b2p−1 a2p−2 b2p−2 ··· a b
2p−1 p p
where ai = bi = 0 for i > n.

Proof From the identity g(t) = f (t)ϕ(t) we obtain the recursion formulas
bi = a0 si−1 + a1 si−2 + · · · + ai s−1 (1.5)

for all i ≥ 0. These can be used to transform the identity (1.3) as follows:

1 0 · · · 0 0 · · · 0

0 1 · · · 0 0 · · · 0

. . . ..
.. .. . . ... ..
. .

p(p−1) 2p+1 0 0 · · ·

1 0 · · · 0
a0 Bp = (−1) 2 a0
(1.3) 0 s−1 · · · sp−2 sp−1 · · · s2p−2

0 0 · · · sp−3 sp−2 · · · s2p−3

.. .. .. .. ..
. . . . .

0 0 · · · s−1 s0 · · · sp−1

1 0 0 0 ··· 0

0 s ··· s2p−2
−1 s0 s1 a 0 a 1 a 2 · · · a2p
0 1 0 0 ··· 0

0 a0 a1 · · · a2p−1
0 0 s−1 s0 ··· s2p−3
= (−1)p · . . . . .. · 0 0 a0 · · · a2p−2 .
.. .. .. .. . .. .. .. .. .
. . . . ..
0 0 0 0 · · · 1 0 · · · 0

0 0 0 0 · · · 0 s−1 · · · sp−1 0 0 0
· · · a0

0 0 0 0 · · · 0 1 · · · 0
Performing the matrix multiplication gives

a 0 a1 · · · a2p
a2
b 0 b1 · · · b2p−1
0
· · · b2p−1
b0 b1 a a1 · · · a2p−1
0
0 a0 · · · a2p−1
a1

p0
0 b0 · · · b2p−2
a0 Bp = (−1) 0 · · · b2p−2 = (−1)p a0 .
b0 .. .
. ..
(1.5) .. .. .. .. . . .
. . . .
0 0 · · · bp
0 0 0 · · · bp
0 0 · · · ap
0 0 0 · · · ap
Permuting the rows and transposition then yield (1.4).
Definition 1.9.10 The resultant R(f, g) of f and g is defined to be the determinant

(with 2n rows)

a 0 a1 · · · an−1 an 0 0 ··· 0

0 a0 · · · 0 ··· 0
an−2 an−1 an
. .. . . .. .. .. .. ..
..
. . . . . . .

0 0 ··· a0 a1 a2 a3 · · · an
.
b 0 b1 · · · bn−1 bn 0 0 ··· 0

0 b0 · · · bn−2 bn−1 bn 0 ··· 0

.. .. . . .. .. .. .. ..
. . . . . . . .

0 0 · · · b0 b1 b2 b3 · · · bn
If b0 = 0, this is the resultant of f and g in the usual algebra sense (cf. [55, 73, 113]).
In contrast, if deg(g) = m < n = deg(f ), then R(f, g) is obtained from the usual
resultant via multiplication with the (unimportant) factor a0n−m .
In case n = p, Theorem 1.9.9 says
n(n−1)
det B(f, g) = (−1) 2 R(f, g). (1.6)
Since the resultant of two polynomials vanishes if and only if they have a common
divisor (cf. [113], . . . ), we obtain
Corollary 1.9.11 The Bézoutian of f and g is nondegenerate if and only if f and
g have no common divisor.
Denoting the quadratic space (K n , B(f, g)) by V (f, g), we have more generally
Proposition 1.9.12 Let D = t q + d1 t q−1 + · · · + dq be the greatest common divisor
of f and g, and let f = f 0 · D, g = g 0 · D. Then the radical Rad V (f, g) of V (f, g)
(see Sect. 1.2) has dimension q, and the quadratic space V (f, g)/ Rad V (f, g) is
isometric to V (f 0 , g 0 ).
Proof Let (cij0 ) := B(f 0 , g 0 ). Then

n−1
f (x)g(y) − f (y)g(x)
cij x i y j =
x−y
i,j =0
f 0 (x)g 0 (y) − f 0 (y)g 0 (x)

= D(x)D(y)
x−y

n−q−1
= D(x)D(y) cij0 x i y j
i,j =0

n−q−1
= cij0 (d0 x q+i + · · · + dq x i ) (d0 y q+j + · · · + dq y j ),
i,j =0
where d0 := 1. Assume that x0 , . . . , xn−1 are independent variables. We form n − q

new independent variables
u0 = dq x0 + dq−1 x1 + · · · + d0 xq
u1 = dq x1 + · · · + d1 xq + d0 xq+1
..
.
un−q−1 = dq xn−q−1 + ··· + d0 xn−1
(d0 := 1). From the above identity we then read off that
B(f, g) (x0 , . . . , xn−1 ) = B(f 0 , g 0 ) (u0 , . . . , un−q−1 ).
The statement now follows since B(f 0 , g 0 ) is nondegenerate by Corollary 1.9.11.
From now on we assume again that K = R is real closed. As before, let f and g
be polynomials of degree n and degree at most n, respectively, and let ϕ := g/f .
+∞
Theorem 1.9.13 The global Cauchy index I−∞ (ϕ) is equal to the signature of the
Hankel form H (f, g), and thus also equal to the signature of the Bézoutian B(f, g)
of f and g.
Proof By Proposition 1.9.12 we may assume without loss of generality that f and
g are relatively prime and thus that H (f, g) is nondegenerate.
We use complex analysis to prove the case R = R. Specifically, we will compute
the residues of the differential form ω = ϕ(z) θ (z)2 dz, where z is a complex
variable,
θ (z) = x0 + x1 z + · · · + xn−1 zn−1
and x0 , . . . , xn−1 are independent real parameters. In order to compute the residue
of ω at z = ∞, we introduce the local coordinate ζ = z−1 at ∞. Then we have
ω = ϕ(z) θ (z)2 (−ζ −2 ) dζ ,
and so − Res∞ (ω) is the coefficient of ζ in the Laurent expansion of ϕ(z) θ (z)2 in
terms of ζ . We see from

n−1
ϕ(z) θ (z)2 = (s−1 + s0 ζ + s1 ζ 2 + · · · ) · xj xk ζ −j −k
j,k=0
that

n−1
− Res∞ (ω) = sj +k xj xk = H (f, g) (x0 , . . . , xn−1 ) . (1.7)
j,k=0
By the Residue Theorem this expression equals the sum of the residues of ω at the
simple poles of ω, in other words at the zeroes of f .
Assume thus that z = α is a (complex) zero of f (z) of multiplicity m. There is a
Laurent expansion
ϕ(z + α) = c0 z−m + c1 z−m+1 + · · ·
with complex coefficients cj and c0 = 0. If α ∈ R, then the cj are also real.

Furthermore,
θ (z + α) = x0 + x1 (z + α) + · · · + xn−1 (z + α)n−1
= u0 + u1 z + · · · + un−1 zn−1 ,
where u0 , . . . , un−1 are linear combinations of x0 , . . . , xn−1 whose coefficients are

integer (universal) polynomials in α. It follows that
Resα (ω) = cm−1 u20 + 2cm−2 u0 u1 + · · · + c0 (u0 um−1 + · · · + um−1 u0 ).
If m = 2r is even, we can write
Resα (ω) = u0 v0 + u1 v1 + · · · + ur−1 vr−1 (A)
where the vj are appropriate linear combinations of x0 , . . . , xn−1 whose coefficients

are integral polynomials in α, c0 , . . . , cm−1 .
If m = 2r + 1 is odd, on the other hand, we can write
Resα (ω) = u0 v0 + · · · + ur−1 vr−1 + c0 u2r (B)
where the vj are linear combinations as before.

If α ∈ R, then all linear combinations uj , vj have real coefficients. Assume now
that α ∈ R, more precisely Im (α) > 0. Then α is also a zero of f , of the same
multiplicity as α, and precisely the complex conjugate linear combinations uj , v j
appear in the residue Resα (ω). In order to split the uj , vj in real and imaginary parts
we write
√ √
uj = pj + −1 pj , vj = qj − −1 qj ,
for 0 ≤ j ≤ r − 1, where the pj , pj , qj , qj are real linear combinations of

x0 , . . . , xn−1 ; when m = 2r + 1 we write
√
ur = pr + −1 pr
where pr , pr are real linear combinations.

If m = 2r we then obtain

r−1
r−1
Resα (ω) + Resα (ω) = (uj vj + uj v j ) = 2 (pj qj + pj qj ), (C)
j =0 j =0
and if m = 2r + 1 we obtain

r−1
Resα (ω) + Resα (ω) = 2 (pj qj + pj qj ) + 2c0 (pr2 − pr2 ). (D)
j =0
In total precisely n real linear forms occur in the identities (A), (B) (for real
α) and (C), (D) (for nonreal α) since f has precisely n zeroes, counted with
multiplicities. Using the Residue Theorem, (1.7) yields a representation of H (f, g)
as a quadratic form in those n linear forms. These must be linearly independent since
otherwise H (f, g) is degenerate. Furthermore, H (f, g) is the orthogonal sum of the
quadratic forms determined by (A), (B) for α ∈ R and (C), (D) for Im(α) > 0.
The forms coming from (A), (C) and (D) are clearly hyperbolic, while the form
coming from (B) is the orthogonal sum of a hyperbolic form and the form c0 u2r ,
and so its signature is the sign of c0 . Observe now that for α ∈ R the Cauchy index
indα (ϕ) = 0 if m is even, but indα (ϕ) = sgn c0 if m is odd. We conclude that

+∞
sign H (f, g) = indα (ϕ) = I−∞ (ϕ).
α∈R,f (α)=0
There are now two different options for proving Theorem 1.9.13 for an arbitrary
real closed field R. The first option√is to use a purely algebraic residue theorem for
rational differential forms over R( −1). Indeed, for the form ω introduced above
one can determine formal Laurent series at all poles over any algebraically closed
field, use them to define the residues of ω, and then show that their sum is zero. The
proof above then remains valid word for word. See also [109, p. 31].
The other option consists of applying Tarski’s Principle (cf. [90, §5] or [91,
§4.2]). This theorem from model theory states among other things that every
“elementary sentence” in the language of ordered fields that is valid over R is also
valid over every real closed field. Even if one has had only very little exposure to
model theory, one readily recognizes that Theorem 1.9.13 is such an elementary
sentence. Tarski’s Principle plays an important role in higher real algebra (as
mentioned in the Preface), but will not be considered any further in this book.
In this way we declare the theorem proven.
From this result we immediately obtain a method for expressing Cauchy indices
Iab (ϕ) on finite open intervals (a, b) as signatures of “modified” Bézoutians. Without
loss of generality we may assume deg g < deg f and define, for each λ ∈ R, the
quadratic form
Bλ (f, g) := B (f, (λ − t)g) .
Its signature is denoted ρ(λ). If λ is at most a pole of order one of ϕ, then

+∞
ρ(λ) = I−∞ ((λ − t)ϕ) = I−∞
λ
(ϕ) − Iλ∞ (ϕ),
by Theorem 1.9.13. Thus, if a, b are at most poles of order one of ϕ and if a < b,
then
ρ(b) = I−∞
a
(ϕ) + inda (ϕ) + Iab (ϕ) − Ib∞ (ϕ),
ρ(a) = I−∞
a
(ϕ) − Iab (ϕ) − indb (ϕ) − Ib∞ (ϕ).
This proves
Corollary 1.9.14 If ϕ has at most simple poles at a and b and if a < b, then
2Iab (ϕ) = ρ(b) − ρ(a) − inda (ϕ) − indb (ϕ).

In our proof of Theorem 1.9.13 we only used the Hankel form H (f, g). The
Bézoutian B(f, g) was considered only indirectly in the sense that we showed via
the theory of Bézoutians that f and g may be assumed relatively prime and so
H (f, g) may be assumed nondegenerate. One could thus get the impression that the
Hankel form is the central object, and that the Bézoutian only plays an auxiliary
role.
In fact, there are two reasons for casting the Bézoutian in the central role. The first
one is of a formal nature: B(f, g) can be defined for arbitrary polynomials f and
g without restriction on the degrees, as in Definition 1.9.3, whereas for H (f, g) the
natural range of definition is deg g ≤ deg f (or even deg g < deg f ). Furthermore,
the Bézoutian B(f, g), beloved in the nineteenth century, but then mostly forgotten,
has in the meantime been shown to have fascinating properties, not shared with
H (f, g). See [46, 47] and the cited literature.
The second reason is that B(f, g) has practical advantages over H (f, g) since
the coefficients of B(f, g) can be obtained quickly from those of f and g (see the
formulas at the beginning of the section), whereas the Taylor coefficients of g/f that
are needed for H (f, g) depend on f and g in a complicated manner. (On the other
hand, the matrix H (f, g) has a simpler form than B(f, g).) Hence, when f and g
vary it is easier to study the change in B(f, g) than the change in H (f, g), which is
important for applications in control theory.
1.10 An Upper Bound for the Number of Real Zeroes (with Multiplicities) 41
Moreover, Hurwitz’ Theorem 1.9.9 gives us tight control over the signature of
B(f, g). As an illustration we mention
Theorem 1.9.15 (Hurwitz) Let f and g be polynomials over R as above (deg g ≤
deg f = n). The following statements are equivalent:
(i) All the roots of f are real and simple, and between any two such roots there is
a root of g;
(ii) The entries in the sequence of determinants

a0 b0 0 0 · · · 0 0

a b a1 b1 a0 b0 · · · 0 0

1, 0 0 ,..., . .. .. .. .. ..
a1 b1 .. . . . . .

a b2n−1 a2n−2 b2n−2 ··· a b
2n−1 n n
are either all positive, or alternating positive and negative.

+∞
Proof Condition (i) signifies I−∞ (g/f ) = ±n, whereas by Theorem 1.9.9 the
condition on the determinants says that sign B(f, g) = ±n. The claim then follows
from Theorem 1.9.13.
1.10 An Upper Bound for the Number of Real Zeroes

(with Multiplicities)
In this section R denotes a real closed field and, in contrast to previous practice, all
zeroes will be counted with multiplicities.
Let f ∈ R(t) be a non-constant rational function without pole in the interval
[a, b] (a, b ∈ R, a < b). We fix n ≥ 1 such that f (n) (a)f (n) (b) = 0 and denote by
Ni the number of real zeroes of the i-th derivative f (i) in ]a, b] (i = 0, 1, . . . , n); in
particular we let N := N0 denote the number of zeroes of f in ]a, b]. For x ∈ [a, b]
we let V (x) := Var f (x), f (x), . . . , f (n) (x) (cf. Sect. 1.7).
Theorem 1.10.1 (Hurwitz) There exists v ∈ N0 such that
N = Nn + V (a) − V (b) − 2v.
Corollary 1.10.2 (Budan–Fourier) If f ∈ R[t] is a polynomial of degree n ≥ 1,

then there exists v ∈ N0 such that
N = V (a) − V (b) − 2v.

Before proving the theorem, we will derive a number of interesting consequences

from it (more precisely: from Corollary 1.10.2). These show that without any
computations one can already draw remarkable conclusions about the real roots of
a polynomial from the sign sequence of its coefficients alone!
Thus, let f (t) = c0 t n + c1 t n−1 + · · · + cn ∈ R[t], where n ≥ 1 and c0 = 0. We
introduce the following notation:
p := number of strictly postive roots of f,

p := number of strictly negative roots of f,
V := Var(c0 , . . . , cn ),
V := Var c0 , −c1 , c2 , . . . , (−1)n cn .
Corollary 1.10.3
(a) (Rule of Descartes) There exist v, v ∈ N0 such that p = V − 2v and p =
V − 2v .
(b) If cn = 0 and if all roots of f are real, then p = V and p = V .
(c) If cn = 0 and if ci−1 = ci = 0 for some i with 1 < i < n, then f has non-real
roots.
Proof (a) V (0) = Var(cn , 1! cn−1 , . . . , n! c0 ) = V and V (b) = 0 for b 0,
together with Corollary 1.10.2, give the first formula. For the second one, simply
replace f (t) by f (−t).
(b) For x ∈ R, let
W (x) := Var(c0 + x, . . . , cn + x), and

W (x) := Var c0 + x, −(c1 + x), . . . , (−1)n (cn + x) .
Then, for all x ∈ R such that
0 < x < min{|ci | : ci = 0},
we have W (x) + W (x) = n, as well as W (x) ≥ V and W (x) ≥ V . Thus,

V + V ≤ n and it follows from (a) that p + p ≤ V + V . On the other hand
p + p = n by the assumption in (b), and so p = V , p = V .
(c) Omit positions i − 1 and i in the sequences and repeat the argument in the
proof of (b).
For the proof of Hurwitz’ Theorem, we provide two auxiliary results.
Proposition 1.10.4 (Taylor) Let 0 = f ∈ R(t) and let a ∈ R be a zero of f of
multiplicity m ≥ 0. There exists g ∈ R(t) such that g(0) = ∞ and
f (a + t) = 1 (m)
m! f (a) t m + t m+1 g(t).
Proof There exists h ∈ R(t) with the properties
h(0) = ∞, h(0) = 0 and f (a + t) = t m h(t).
The Leibniz rule gives f (m) (a) = m! h(0). There exists g ∈ R(t) such that g(0) =
∞ and h(t) = h(0) + t g(t), from which the statement follows.
Proposition 1.10.5 Let a, b ∈ R, a < b, and f ∈ R(t) without a pole in [a, b].
(a) If f has no zeroes in [a, b[, f (b) = 0 and f (a) = 0, then the number of zeroes
of f in ]a, b[ is

even if f (a)f (a) < 0
.
odd if f (a)f (a) > 0
(b) If f has no zeroes in ]a, b], f (a) = 0 and f (b) = 0, then the number of zeroes
of f in ]a, b[ is

odd if f (b)f (b) < 0
.
even if f (b)f (b) > 0
Proof On grounds of symmetry it suffices to prove (a). By Lemma 1.7.3 we have

f (b − ε) f (b − ε) < 0 for small ε > 0. It follows from the assumptions that
f (a)f (b − ε) > 0. If f (a) f (a) > 0 (resp. < 0), then f (a) and f (b − ε) have
distinct (resp. equal) signs, and so the number of zeroes of f in [a, b − ε] is odd
(resp. even).
We have arrived at the proof of Theorem 1.10.1 and adopt the assumptions and
notation in its formulation.
Proof of Theorem 1.10.1 For all x ∈ R with f (x) = ∞ we define

w(x) := 1
2 1 − sign f (x) f (x) ,

and also (Wf )(x) := 0 if f (x) = 0 and (Wf )(x) := 1
2 1 − sign f (x)f (k) (x) if
f (x) = 0 and k ≥ 1 is minimal such that f (k) (x) = 0.
Lemma 1.10.6 Assume that f (a) f (a) f (b) f (b) = 0. Then there exists v ∈ N0
such that N − N1 = w(a) − w(b) − 2v.
Proof Let c1 , . . . , cr be the distinct zeroes of f in [a, b], with respective multiplic-
ities m1 , . . . , mr , and assume a < c1 < · · · < cr < b. By Proposition 1.7.4 f has
an odd number of zeroes in ]ci−1 , ci [ (i = 2, . . . , r), and in addition an (mi −1)-fold

zero at ci (i = 1, . . . , r). Thus f has precisely
(r − 1) + 2u + (m1 − 1) + · · · + (mr − 1) = N + 2u − 1
zeroes in [c1 , cr ] for some u ∈ N0 . Furthermore, the number of zeroes of f in

]a, c1 [, resp. in ]cr , b[, equals 1−w(a)+2u1 , resp. w(b)+2u2 , for some u1 , u2 ∈ N0
(Proposition 1.10.5). It follows that
N1 = (N + 2u − 1) + (1 − w(a) + 2u1 ) + (w(b) + 2u2 ) ,
i.e.,
N1 = N − w(a) + w(b) + 2v with v := u + u1 + u2 .

(The assumption in Lemma 1.10.6 is now revoked.)
Lemma 1.10.7 For all x ∈ R with f (x) = ∞ we have (Wf )(x) = w(x + h) for
all sufficiently small h > 0.
Proof We distinguish three cases:
1) If f (x) = 0, then (Wf )(x) = 0, and w(x + h) = 0 for a small h > 0 by
Lemma 1.7.3.
2) If f (x) = 0 and f (x) = 0, let k ≥ 1 be minimal such that f (k) (x) = 0. By
Proposition 1.10.4 there exists g ∈ R(t) such that g(0) = ∞ and
f (k) (x)
f (x + h) = · hk−1 + hk g(h).
(k − 1)!
It follows that sign f (x + h) = sign f (k) (x) for a small h > 0, i.e., w(x + h) =
(Wf )(x) for a possibly even smaller h > 0.
3) If f (x) f (x) = 0, then (Wf )(x) = w(x) = w(x + h) for a small h > 0.
Lemma 1.10.8 There exists v0 ∈ N0 such that N −N1 = (Wf )(a)−(Wf )(b)−2v0 .
Proof Let h > 0 be sufficiently small such that f · f has no zeroes nor poles
in ]a, a + h] ∪ ]b, b + h] and thus such that N, resp. N1 , is the number of zeroes
of f , resp. f , in [a + h, b + h]. From Lemma 1.10.6 it follows that N − N1 =
w(a + h) − w(b + h) − 2v0 for some v0 ≥ 0, and the statement follows from
Lemma 1.10.7.
This result can now be iterated. Since the assumptions made about f are also
valid for f , . . . , f (n−1) , we obtain
N − N1 = (Wf )(a) − (Wf )(b) − 2v0 ,

N1 − N2 = (Wf )(a) − (Wf )(b) − 2v1 ,
..
.
Nn−1 − Nn = (Wf (n−1) )(a) − (Wf (n−1) )(b) − 2vn−1
for natural numbers v0 , . . . , vn−1 ≥ 0, and so
N − Nn = W (a) − W (b) − 2v,
where W := Wf + Wf + · · · + Wf (n−1) and v := v0 + v1 + · · · + vn−1 .

However, if f (n) (x) = 0, then W (x) = Var f (x), f (x), . . . , f (n) (x) = V (x)
(and we are done!). Indeed, if k ≥ 1 is minimal such that f (k) (x) = 0 then, by
definition,

0 if f (x) f (k) (x) ≥ 0
(Wf )(x) = .
1 otherwise
Let 0 ≤ k1 < · · · < kr = n be precisely the indices k ∈ {0, . . . , n} such that

f (k) (x) = 0. Among the Wf (i) (x) (i = 0, . . . , n) at most the Wf (kj ) (x) can
be nonzero, and Wf (kj ) (x) = 1 if and only if f (kj ) (x) f (kj+1 ) (x) < 0 (j =
1, . . . , r − 1). Therefore,

(Wf )(x) = Var f (k1 ) (x), . . . , f (kr ) (x) = Var f (x), f (x), . . . , f (n) (x) = V (x),
which concludes the proof of Theorem 1.10.1.

In our proof of the Budan–Fourier Theorem via Hurwitz’ Theorem we followed
the book [85] of N. Obreschkoff [loc.cit. §15], which also contains a substantially
different approach to the theorem of Budan–Fourier [loc.cit. §14] as well as further
investigations on this topic. We recommend in particular a number of interesting
theorems of Laguerre [loc.cit. §16], which generalize the Rule of Descartes in a
way different from the Budan–Fourier Theorem.
1.11 The Real Closure of an Ordered Field
Definition 1.11.1 Let K be a field.

(a) Let P be an ordering of K. A real closure of (K, P ) is a real closed overfield
R of K that is algebraic over K and whose ordering extends P .
(b) A real closure of K is an algebraic extension of K that is real closed.
It follows from Proposition 1.5.2 (the equivalence (i)⇔(ii)) and Zorn’s Lemma
that every ordered field has a real closure. It is uniquely determined (similar to the
algebraic closure of a field), up to unique isomorphism (which is not the case for the
algebraic closure):
Theorem 1.11.2 Let (K, P ) be an ordered field, R a real closure of (K, P ) and
ϕ : K → S an order preserving homomorphism (with respect to P ) into another
real closed field S. Then there exists a unique homomorphism ψ : R → S that
extends ϕ.
Corollary 1.11.3 Let R/K and R /K be real closures of (K, P ). Then there exists
a unique K-isomorphism R → R .
Because of Corollary 1.11.3 we will say the real closure of (K, P ) from now on.
Proposition 1.11.4 Let L/K be a finite field extension such that K ⊆ L ⊆ R. Then
ϕ has an order preserving extension ψL : L → S. (Here “order preserving” is with
respect to the ordering of L induced by R.)
Proof Let α ∈ L be such that L = K(α), and let f ∈ K[t] be the minimal
polynomial of α. By Hermite’s Theorem (Theorem 1.7.15) the Sylvester form S(f )
of f has positive signature over R, and thus also over S. Again by this theorem,
f thus also has a zero in S and so there is at least one extension χ : L → S of ϕ.
Let χ1 , . . . , χr be all such extensions (r ≥ 1). Suppose that none of the χi is order
preserving. Then there exist a1 , . . . , ar ∈ L with ai > 0 (in R), but χi (ai ) < 0
√ √
(in S). Consider L := L( a1 , . . . , ar ) ⊆ R. Then L /K is finite, but there are
actually no extensions χ : L → S of ϕ to L (since χ |L must be one of the χi , i.e.,
√
χ ( ai )2 < 0 for some i), contradicting Hermite’s Theorem.
Proof of Theorem 1.11.2 By Zorn’s Lemma there exists an intermediate field
K /K of R/K and a maximal order preserving extension ψ : K → S of ϕ.
However, it follows from Proposition 1.11.4 that K = R, which establishes the
existence of ψ. Concerning the uniqueness see the next remark. This finishes the
proof of Theorem 1.11.2.
Remark 1.11.5 The extension ψ of ϕ from Theorem 1.11.2 can be “described” as

follows: if α ∈ R is a zero of the polynomial 0 = f ∈ K[t], if α1 < · · · < αr are
the distinct zeroes of f in R, and if β1 < · · · < βs are the distinct zeroes of f ϕ (the
1.11 The Real Closure of an Ordered Field 47
polynomial f , transferred to S by ϕ) in S, then r = s by Hermite, and ψ(αi ) = βi

for i = 1, . . . , r. (A proof of the homomorphy of ψ based on this definition requires
more effort however, cf. [4, p. 93].)
Theorem 1.11.6 Let L/K be an algebraic field extension and ϕ : K → S a

homomorphism into a real closed field S. Let P denote the ordering of K induced
by ϕ. Then there exists a bijection between the extensions ψ : L → S of ϕ to L and
the extensions Q of the ordering P to L; it is given by ψ → Q := ψ −1 (S 2 ).
Proof For every extension ψ of ϕ, ψ −1 (S 2 ) is an extension of P . Conversely, let Q
be an ordering of L such that P ⊆ Q, and let R be the real closure of (L, Q) (thus
also of (K, P )). By Theorem 1.11.2 there exists an extension R → S of ϕ, and its
restriction ψ to L satisfies Q = ψ −1 (S 2 ). Finally, if ψ1 , ψ2 : L ⇒ S are extensions
of ϕ such that ψ1−1 (S 2 ) = Q = ψ2−1 (S 2 ), then there exist extensions χi : R → S
of ψi (i = 1, 2) by Theorem 1.11.2. Since χ1 |K = ϕ = χ2 |K , we have χ1 = χ2 ,
and thus also ψ1 = ψ2 .
Corollary 1.11.7 Let (K, P ) be an ordered field with real closure R, and let L/K
be a finite algebraic extension. If L = K(α) and f ∈ K[t] is the minimal
polynomial of α over K, then the extensions of P to L correspond bijectively to
the zeroes of f in R. In particular, P has at most [L : K] distinct extensions to L.
Example 1.11.8 Let L = Q(α) be a number field and f ∈ Q[t] the minimal
polynomial of α over Q. The zeroes of f correspond to the field embeddings
L → C. The real zeroes of f thus correspond to the embeddings L → R, and
therefore also to the orderings of L.
The following theorem ties in with Corollary 1.6.3:
Theorem 1.11.9 (Orderings and Involutions in the Galois Group) Let K be a
field (of arbitrary characteristic) with absolute Galois group = Gal (Ks /K).
There exists a natural bijection from the set of all conjugacy classes [τ ] of
involutions in to the set of orderings of K. More precisely: ([τ ]) is √
the ordering
√
of K defined by the inclusion K → Fix(τ ), i.e., ([τ ]) = {a ∈ K : τ ( a) = a}.
(Note that Fix(τ ) is real closed by Sect. 1.6.)
Proof (Sketch) is well-defined since Fix(σ τ σ −1 ) = σ (Fix τ ) (σ, τ ∈ ). The
injectivity of ϕ follows from Theorem 1.11.2, and the surjectivity from the existence
of the real closure and the findings in Sect. 1.6. We leave the (easy) details to the
reader.
If τ is an involution in , then the centralizer of τ in is equal to {1, τ }; this
is just a reformulation of Aut(Fix(τ )/K) = {1} (Proposition 1.5.6(b)). Therefore,
the conjugates of τ are in bijective correspondence with the elements of the
homogeneous space /{1, τ }.
To conclude this section, we express several of our previous results in the
language of Witt rings and signatures. Let K be a field of char K = 2. By
Sect. 1.2 the orderings of K correspond bijectively to the signatures of K. Every

homomorphism ϕ : K → R into a real closed field R gives rise to a signature
W (ϕ) := iR/K : W (K) → W (R) = Z
of K (Exercise 1.2.17), and in this way one obtains all signatures in case R/ϕ(K)
is algebraic (existence of the real closure). If ϕ : K → R, ϕ : K → R are
homomorphisms into real closed fields R, R , and if R/ϕ(K) is algebraic, then
W (ϕ) = W (ϕ ) if and only if there exists a K-homomorphism R → R (uniqueness
of the real closure).
Theorem 1.11.6 can be formulated as follows: If L/K is an algebraic field
extension, ϕ : K → R a homomorphism into a real closed field R, and σ = W (ϕ)
the induced signature of K, then the extensions ψ : L → R of ϕ correspond
bijectively to the signatures τ of L that extends σ (i.e., that satisfy σ = τ ◦ iL/K )
via τ = W (ψ).
1.12 Transfer of Quadratic Forms
In this section all field have characteristic = 2.

If L/K is an arbitrary field extension, then the functoriality of the Witt ring yields
a ring homomorphism iL/K : W (K) → W (L). If [L : K] is finite, then every K-
linear form s on L defines a map s∗ in the reverse direction, the transfer via s. This
is only a homomorphism of additive groups. The special case s = trL/K leads to a
trace formula for the extension of signatures.
In the remainder of this section let L/K be a finite field extension and s : L → K
a nonzero linear form on the K-vector space L.
Definition 1.12.1 Let ϕ = (V , b) be a bilinear space over L. The transfer s∗ ϕ of ϕ
(to K, via s) is the bilinear space s∗ ϕ = (V , s ◦ b) over K (where V is considered
as K-vector space).
√
Example 1.12.2 Let L = Q( 2), tr := trL/Q and ϕ = 1, i.e., ϕ(x, y) = xy. The
√
matrix of tr∗ ϕ with respect to the Q-basis (1, 2) of L is 20 04 . Hence, tr∗ ϕ ∼
=
2, 1.
Lemma 1.12.3 Let W ⊆ V be an L-subspace. Then, for all v ∈ V ,
b(v, W ) = 0 ⇔ (s ◦ b)(v, W ) = 0.
Proof One direction is trivial. For the other direction we note that the K-bilinear
form
β : L × L → K, β(a, a ) := s(aa )
1.12 Transfer of Quadratic Forms 49
is nondegenerate. (By assumption there exists a0 ∈ L such that s(a0 ) = 0; thus if

0 = a ∈ L, then β(a, a −1 a0 ) = 0.) Hence, if (s ◦ b)(v, W ) = 0 and w ∈ W , then
0 = s (b(v, aw)) = s (a · b(v, w)) = β (a, b(v, w))
for all a ∈ L, and it follows that b(v, w) = 0.
Lemma 1.12.4 (Elementary Properties of the Transfer) Let ϕ, ϕ be bilinear

spaces over L, and let s be as above.
(a) dimK s∗ ϕ = [L : K] · dimL ϕ;
(b) s∗ (ϕ ⊥ ϕ ) ∼
=K s∗ ϕ ⊥ s∗ ϕ ;
(c) Rad(ϕ) = Rad(s∗ ϕ); in particular, s∗ ϕ is nondegenerate if and only if ϕ is
nondegenerate;
(d) if ϕ is hyperbolic, then s∗ ϕ is also hyperbolic.
Proof (a) and (b) are clear, and (c) follows from Lemma 1.12.3.
(d): If ϕ = (V , b) is hyperbolic, then ϕ is nondegenerate and there exists a
subspace U ⊆ V with U = U ⊥ (cf. Sect. 1.2). By Lemma 1.12.3 both properties
transfer to s∗ ϕ.
From this we immediately obtain:
Proposition 1.12.5 Let L/K be a finite field extension and s : L → K a nonzero
K-linear form. Then s∗ induces a homomorphism s∗ : W (L) → W (K) of additive
groups (which is not multiplicative in general).
Remark 1.12.6 The additive subgroup s∗ W (L) of W (K) does not depend on s.
Indeed, if s : L → K is another K-linear map, s = 0, then there exists a ∈ L∗
with s (b) = s(ab) for all b ∈ L. It follows that s∗ (η) = s∗ (a · η) for η ∈ W (L).
The following theorem shows in particular that s∗ W (L) is an ideal of W (K).
Proposition 1.12.7 (Projection Formula) For all ξ ∈ W (K) and η ∈ W (L) we

have
s∗ iL/K (ξ ) · η = ξ · s∗ (η).
In particular, the composition s∗ ◦ iL/K is multiplication by s∗ (1) in W (K).

Proof Let ξ , resp. η, be represented by bilinear spaces (V , b) over K, resp. (W, c)
over L. Then ξ ·s∗ (η) is represented by (V ⊗K W, b ⊗ (s ◦ c)), and s∗ iL/K (ξ ) · η
is represented by ((V ⊗K L) ⊗L W, s ◦ ((b ⊗ 1) ⊗ c))). For simplicity we write
b := b ⊗ (s ◦ c), c := (b ⊗ 1) ⊗ c; so b is K-bilinear on V ⊗K W and c is
L-bilinear on (V ⊗K L) ⊗L W .
There exists a canonical K-vector space isomorphism ϕ : (V ⊗K L) ⊗L W −→ ∼

V ⊗K W , given by ϕ ((v ⊗ λ) ⊗ w) = v ⊗ λw. This ϕ is an isometry between s ◦ c
and b ; indeed for x := (v ⊗ λ) ⊗ w and x := (v ⊗ λ ) ⊗ w we have
(s ◦ c )(x, x ) = s (b ⊗ 1)(v ⊗ λ, v ⊗ λ ) · c(w, w )

= s b(v, v ) · λλ · c(w, w )
= b(v, v ) · s ◦ c(λw, λ w )
= b (v ⊗ λw, v ⊗ λ w ) = b (ϕx, ϕx ).
The projection formula shows that s∗ is a homomorphism of W (K)-modules,
where the W (K)-module structure of W (L) is defined via iL/K .
Corollary 1.12.8 Let W (L/K) denote the kernel of iL/K : W (K) → W (L). Then
s∗ (W (L)) · W (L/K) = 0.
Thus, if σ is a signature of K and if there exists η ∈ W (L) such that σ (s∗ η) = 0,

then σ has an extension to L! (Proposition 1.4.4).
Theorem 1.12.9 (Trace Formula) Let L/K be a finite field extension, tr := trL/K
the trace of L over K, and σ : W (K) → Z a signature of K. Then, for all η ∈ W (L),

σ (tr∗ η) = τ (η),
τ |σ
where the sum is over all signatures τ of L that extend σ .

Proof Since K is real, char K = 0, hence tr = 0 and tr∗ is defined. Let R be
the real closure of K with respect to σ . By the Primitive Element Theorem ([55,
Vol. I, §4.14], [73, V, Theorem 4.6]) there exists α ∈ L such that L = K(α). Let
f ∈ K[t] be the minimal polynomial of α over K, α1 , . . . , αr the zeroes of f in R,
and ϕi : L → R the embeddings over K defined by ϕi (α) = αi for i = 1, . . . , r.
By Corollary 1.11.7, the τi := W (ϕi ) (i = 1, . . . , r) are precisely the signatures of
L that extend σ .
Since tr∗ is additive, it suffices to prove the statement for η = β, β ∈ L∗ . Let
g ∈ K[t] with β = g(α). We note that for i = 1, . . . , r,
τi (η) = signR g(αi ). ()

1.12 Transfer of Quadratic Forms 51
On the other hand, tr∗ (η) is precisely the Sylvester form SylK (f ; g) (Sect. 1.8).
Then, by Lemma 1.8.5(a),
σ (tr∗ (η)) = σ SylK (f ; g) = signR SylR (f ; g)

r
r
= signR g(αi ) = τi (η) = τ (η),
() ()
i=1 i=1 τ |σ
where the identity () follows from the explicit computation of the Sylvester form
over real closed fields (see the end of Sect. 1.8).
Corollary 1.12.10 Under the assumptions of Theorem 1.12.9, σ (tr∗ (1L )) is

precisely the number of extensions of σ to L.
At this point the circle is complete: Since tr∗ (1L ) = SylK (f ; 1) we have with
Proposition 1.8.7 that
σ (tr∗ (1L )) = signR SylR (f ; 1) = signR S(f )
is the signature of the Sylvester form of f . By Corollary 1.11.7 we thus obtain

Hermite’s Theorem (proved in Sect. 1.7 and again in Sect. 1.8) in this special case.
Chapter 2
Convex Valuation Rings and Real Places
This chapter gives a detailed introduction to valuation theory of fields. Valuations

are presented from three different points of view: valuation maps, valuation rings,
and places. Valuations are naturally associated with orderings, since every convex
subring of an ordered field is a valuation ring. The precise relationship between the
orderings of a field that are compatible with a valuation ring, and the orderings of
the residue field of the latter, is exhibited by the Baer–Krull Theorem. Finally, the
Artin–Lang Theorem is presented in the form of a Stellensatz. Other than the Artin–
Schreier concept of orderings and real closures, this is our essential input for the
solution of Hilbert’s 17th Problem, which we present at the end of the chapter.
2.1 Convex Subrings of Ordered Fields
Definition 2.1.1 Let (M, ≤) be a partially ordered set. A subset X ⊆ M is called

convex in M if for all x, y, z ∈ M,
x ≤ z ≤ y and x, y ∈ X ⇒ z ∈ X.
Arbitrary intersections and upward directed unions of convex subsets are again
convex. In particular, for every subset Y ⊆ M there exists a smallest convex superset
X of Y in M, the convex hull of Y in M.
Definition 2.1.2
(a) An ordered abelian group is a pair (, ≤), where is an abelian group (usually
written additively) and ≤ is a total order relation on the set such that for all
α, β, γ ∈ ,
α ≤ β ⇒ α + γ ≤ β + γ.
If there is no ambiguity, we usually simply write instead of (, ≤).

https://doi.org/10.1007/978-3-031-09800-0_2
54 2 Convex Valuation Rings and Real Places
(b) A homomorphism ϕ : → between ordered abelian groups is called order

preserving if for all α ∈ ,
α ≥ 0 ⇒ ϕ(α) ≥ 0.
Definition 2.1.3 Let be an ordered abelian group and ≤ a subgroup. For

α ∈ we let |α| := α or −α, depending on whether α ≥ 0 or α ≤ 0.
(a) An element γ ∈ is called infinitely small with respect to if nγ < |δ| for all
0 = δ ∈ and all n ∈ Z. Conversely, γ is called infinitely large with respect to
if δ < |γ | for all δ ∈ .
(b) The group is called archimedean over if does not contain any infinitely
large elements with respect to . The group is called archimedean if for all
α, β ∈ with β = 0, there exists n ∈ Z such that α < n|β| (i.e., if is
archimedean over every subgroup different from {0}).
Remarks 2.1.4
(1) An ordered abelian group can also be defined as a pair (, ), where is an
abelian group and ⊆ is a subset such that
+ ⊆ , ∩ (−) = {0}, and ∪ (−) = .
(2) Let (K, ≤) be an ordered field and denote the set of positive elements of K by
K+ ∗ . Then (K, +, ≤) and (K ∗ , ·, ≤) are ordered abelian groups.
+
(3) Ordered abelian groups are torsion free: nα = 0 for all 0 = α ∈ and 0 = n ∈
Z.
(4) Let be an ordered abelian group. The convex hulls of subgroups of are
subgroups again. A subgroup is convex if and only if γ ∈ , δ ∈ and
0 ≤ γ ≤ δ imply that γ ∈ . The convex subgroups of form a chain: every
two convex subgroups are comparable with respect to inclusion.
(5) If ϕ : → is an order preserving homomorphism of ordered abelian groups,
then Ker(ϕ) is convex in . Conversely, for every convex subgroup of
there is precisely one ordering of = / that makes into an ordered
abelian group and π : → order preserving, namely = π(), where
:= {α ∈ : α ≥ 0} (Remark (1)). The quotient group = / will
always be equipped with this ordering. There exists an obvious homomorphism
theorem for ordered abelian groups.
(6) Convex subgroups are sometimes called isolated subgroups in the literature (see
for instance [13, Chap. VI]).
2.1 Convex Subrings of Ordered Fields 55
Examples 2.1.5
(1) Let 1 , . . . , r be ordered abelian groups. We equip := 1 × · · · × r with
the lexicographic order: (γ1 , . . . , γr ) > 0 if and only if
there exists i ∈ {0, . . . , r − 1} such that γ1 = · · · = γi = 0 and γi+1 > 0.
Then (, >) is an ordered abelian group which we denote by (1 × · · · × r )lex .
The projections → (1 × · · · × i )lex where i = 0, . . . , r, are order
preserving, and so their kernels i are convex subgroups of . Clearly we have
= 0 ⊇ 1 ⊇ · · · ⊇ r = 0. If the i are archimedean, then 0 , . . . , r
are the only convex subgroups of .
(2) In the special case 1 = · · · = r = Z we call (e1 , . . . , er ) the lexicographic
basis of Zrlex , where ei denotes i-th unit vector.
Exercise 2.1.6 Let G be a torsion free abelian group and H ⊆ G a subgroup. Let
an ordering ≤ of H be given. Show that ≤ can be extended to an ordering of G, and
that this extension is uniquely determined if G/H is a torsion group.
In the rest of this section (K, P ) will be an ordered field. Whenever it is clear
that convexity, the ≤ sign, etc. refer to P we will not mention this explicitly.
Proposition 2.1.7 Let A be a subring of K.
(a) The convex hull of A in K is a subring of K.
(b) A is convex in K if and only if [0, 1] ⊆ A. In particular, if A is convex in K,
then every overring of A is also convex in K.
Proof (a) This is clear. (b) Obviously [0, 1] ⊆ A is necessary for the convexity of
A. Conversely, if [0, 1] ⊆ A and if a ∈ A, b ∈ K with 0 < b < a, then ba −1 ∈ A
and so b = ba −1 · a ∈ A.
Definition 2.1.8 Let (K, P ) be an ordered field and A ⊆ K a subring.

(a) We denote the convex hull of A in K (with respect to P ) by oP (K/A) or simply
o(K/A). The smallest convex subring of K with respect to P is oP (K/Z) =:
oP (K).
(b) When discussing infinitely small or infinitely large elements of K in relation to
A (and P ), we always refer to the underlying additive groups. In particular,
(K, P ) is called archimedean over A if oP (K/A) = K; the ordered field
(K, P ) is called archimedean if oP (K) = K, i.e., if K does not contain any
proper convex subrings. The last condition is equivalent to the well-known
“Axiom of Archimedes”, which says that for all a ∈ K there exists an n ∈ N
such that a < n.
One should note that a given (real) field in general possesses archimedean as well
as non-archimedean orderings. For the simple transcendental extension Q(t) of Q,
examples of both cases were given in Sect. 1.1.
In a (non-trivial) example we determine the convex hull of a subfield explicitly:

Example 2.1.9 Let (F, P ) be an ordered field, K = F (t) the rational function field
in one variable over F , and Q := P0,+ the ordering of K that extends P (where t is
positive and infinitely small compared to F ) as presented in Example 1.1.13(3). We
will show that oQ (K/F ) = F [t](t ). In (K, Q) we have |at r | < |b| for all r ∈ N and
all a, b ∈ F ∗ . Thus, if g(t) = a0 t d + · · · + ad ∈ F [t] with g(0) = ad = 0, then
2 |g(0)| < |g(t)| < 2|g(0)|

1
in (K, Q). Hence,
4 |f (0)| < |f (t)| < 4|f (0)|

1
for all f ∈ F (t) with f (0) ∈ {0, ∞}. Now let h(t) = t r f (t) ∈ F (t)∗ , where r ∈ Z
and f ∈ F (t) with f (0) ∈ {0, ∞}. By what we have just established, there exist
a, b ∈ F with 0 < a < b and at r < |h(t)| < bt r . It follows that
h(t) ∈ oQ (K/F ) ⇔ r ≥ 0 ⇔ h(t) ∈ F [t](t ).
The following classical theorem of Hölder gives (in principle) a summary of all
archimedean ordered abelian groups and fields:
Theorem 2.1.10 (O. Hölder [52], 1901)
(a) Let be an archimedean ordered abelian group and let 0 < γ ∈ . Then there
exists precisely one order preserving injective group homomorphism ϕ : →
R such that ϕ(γ ) = 1.
(b) Let (K, P ) be an archimedean ordered field with underlying archimedean
ordered abelian group (K, +, P ). The order preserving embedding ϕ : K → R
such that ϕ(1) = 1, furnished by (a), is a ring homomorphism.
Corollary 2.1.11 Up to order compatible isomorphism the archimedean ordered

abelian groups are precisely the subgroups of (R, +), and the archimedean ordered
fields are precisely the subfields of R (each with the ordering induced by R).
Corollary 2.1.12 (von Staudt’s Theorem) The identity is the only endomorphism
of the field R of real numbers.
Corollary 2.1.13 A proper overfield of R does not have any archimedean order-
ings.
Proof of Theorem 2.1.10 With some heuristics one sees quickly how the proof
should proceed. Namely, if there exists a map ϕ as in (a), then for all α ∈ , m ∈ Z,
and n ∈ N we have
m
n ≤ ϕ(α) ⇔ mϕ(γ ) ≤ nϕ(α) ⇔ mγ ≤ nα.
2.2 Valuation Rings 57
Therefore we define for α ∈ :
U (α) := { m
n : m ∈ Z, n ∈ N, mγ ≤ nα}, O(α) := { n : m ∈ Z, n ∈ N, mγ > nα}.
m
The pair U (α), O(α) constitutes a proper Dedekind cut of Q. In other words (cf.
Definition 2.9.3):
(1) U (α) ∪ O(α) = Q;
(2) U (α), O(α) are nonempty;
(3) for all a ∈ U (α) and b ∈ O(α) we have a < b.
Here (1) is trivial, (2) uses the archimedean property, and (3) uses the torsion
freeness of .
As is well-known, the real numbers are axiomatically defined via such cuts. Thus,
for every α ∈ there exists precisely one ϕ(α) ∈ R such that
U (α) = ]−∞, ϕ(α)]R ∩ Q and O(α) = ]ϕ(α), ∞[R ∩ Q.
One verifies that for all α, β ∈ , U (α) + U (β) ⊆ U (α + β) and O(α) + O(β) ⊆
O(α + β), from which it follows that ϕ(α + β) = ϕ(α) + ϕ(β). Then ϕ is order
preserving since γ > 0 and is archimedean. Also, ϕ is injective since Ker(ϕ) =
{α ∈ : − γ ≤ nα ≤ γ for all n ∈ Z}.
For (b), let = K and γ = 1. The map ϕ is then also multiplicative: since
Q ⊆ K we have
U (α) = ]−∞, α]K ∩ Q and O(α) = ]α, ∞[K ∩ Q
for all α ∈ K, from which the homomorphy of ϕ follows.

Warning: By Hölder’s Theorem, Q is dense in every archimedean ordered field
(K, P ) (with respect to the topology induced by P on K, cf. Sect. 2.6). The obvious
guess that this is always so for archimedean extensions K/L of ordered fields is
false, as the following exercise shows.
Exercise 2.1.14 Let L = R(t) and K = R(t 2 ), a subfield of L, and let Q be the
positive cone of L with 0 <Q nt <Q 1 for all n ∈ N. Show that K fails to be dense
in L with respect to the order topology of Q.
We mention the book [93] as a very extensive source on ordered groups, rings
and fields.
2.2 Valuation Rings
The theory of Krull valuations revolves around three fundamental concepts: valua-
tions, valuation rings, and places. These are more or less equivalent to each other,
but it may be more convenient to work with one concept and not the other two,
depending on the situation. Therefore it is important to be familiar with all three of
them and to be able to translate from one to the other, as and when required. In this
section we investigate the first part of this triptych: valuation rings. In Sect. 2.4 we
will consider valuations. Finally, places will be investigated in Sect. 2.8.
Definition 2.2.1 A subring A of a field K is called a valuation ring of K if for
every a ∈ K ∗ we have a ∈ A or a −1 ∈ A. An arbitrary ring A is called a valuation
ring if it has no zero divisors and is a valuation ring of its field of fractions.
Definition 2.2.2 A ring A is called local if A = 0 and A has just one maximal ideal,
which is usually denoted by mA or simply m. We call the field κ(A) := A/mA the
residue field of A.
Proposition 2.2.3 Every valuation ring is a local ring.

Proof Let A be a valuation ring. We must show that m := A \ A∗ is an ideal in A.
It is clear that am ⊆ m for a ∈ A. Let a, b ∈ m \ {0} and assume without loss of
generality that ab−1 ∈ A. Then a − b = (ab −1 − 1)b ∈ m. We conclude that m is
an ideal.
The following fact, although almost trivial, is of great importance for real algebra
and geometry:
Proposition 2.2.4 If (K, P ) is an ordered field, then every subring of K that is
convex with respect to P is a valuation ring of K.
Proof Let A ⊆ K be a convex subring and a ∈ K ∗ . If |a| ≤ 1, then a ∈ A,
otherwise a −1 ∈ A.
Definition 2.2.5 A valuation ring A is called residually real if the field κ(A) =
A/mA is real.
Remarks 2.2.6
(1) Let (K, P ) be an ordered field, A ⊆ K a convex subring of K and π : A →
κ(A) = A/mA the residue homomorphism. We see immediately that
P := π(A ∩ P )
is an ordering of the residue field κ(A). In particular, A is residually real. We

call P the ordering of κ(A), induced by P .
(2) Let (K, P ) be an ordered field and L ⊆ K a subfield. The maximal ideal of
the valuation ring oP (K/L) consists precisely of the elements of K that are
infinitely small with respect to L (i.e., the elements a ∈ K with n|a| < |b| for
all n ∈ N and 0 = b ∈ L).
2.2 Valuation Rings 59
Proposition 2.2.7 Let (K, P ) be an ordered field and A ⊆ K a valuation ring of

K. The following statements are equivalent:
(i) A is convex in K;
(ii) mA is convex in K;
(iii) mA is convex in A;
(iv) mA ⊆ ]−1, 1[ ;
(v) 1 + mA > 0 ;
(vi) oP (K) ⊆ A.
Proof (i) ⇒ (ii): Let a, b ∈ K with 0 < b < a and a ∈ mA . From a −1 ∈ A and
0 < a −1 < b−1 it follows that b−1 ∈ A, hence b ∈ mA .
The implications (ii) ⇒ (iii) ⇒ (iv) ⇒ (v) are trivial.
(v) ⇒ (vi): Let 0 < a ∈ oP (K) and let n ∈ N with a < n. If a ∈ A, then
a −1 ∈ mA and 1 − na −1 < 0, which contradicts (v).
(vi) ⇒ (i): This follows from Proposition 2.1.7.
Further examples of valuation rings can be obtained from:
Proposition 2.2.8 If A is a unique factorization domain and π ∈ A a prime
element, then the localization AπA is a valuation ring (more precisely: a discrete
valuation ring of rank 1, cf. Sect. 2.4).
Proof Every 0 = b ∈ Quot A can be written as b = π e · a/a with e ∈ Z and
a, a ∈ A not divisible by π. If e ≥ 0, then b ∈ AπA ; otherwise b−1 ∈ AπA .
Next we will discuss an important connection between the prime ideals of a
valuation ring A and its overrings in K := Quot A. Note that every overring of
A in K is itself a valuation ring of K (this follows immediately from the definition).
Proposition 2.2.9 Let A be a valuation ring of K.
(a) For every prime ideal p of A we have p = pAp . Thus p is the maximal ideal of
the valuation ring B := Ap .
(b) For every overring B of A in K, p := mB is contained in A, hence is a prime
ideal of A, and B = Ap .
(c) The set of overrings of A in K is totally ordered with respect to inclusion. The
same is true for the set of prime ideals of A.
Corollary 2.2.10 There exists an inclusion reversing bijection between the set of
prime ideals p of A and the set of overrings B of A in K, given by p → Ap =: B.
The inverse is given by B → mB =: p. Both sets are totally ordered (by inclusion).
Corollary 2.2.11 For every two valuation rings A and B of K we have
A ⊆ B ⇔ mA ⊇ mB .
Proof of Proposition 2.2.9 (a) Let a ∈ p and b ∈ A \ p. If ab−1 ∈ A, then

(ab−1)−1 = ba −1 ∈ A, and thus b = (ba −1)a ∈ p, contradiction. Therefore
ab−1 ∈ A, and it follows from a = (ab−1)b and b ∈ p that ab−1 ∈ p.
(b) For all 0 = b ∈ mB =: p we have b−1 ∈ B, thus also b−1 ∈ A, and hence
b ∈ A. This shows p ⊆ A. The inclusion Ap ⊆ B follows from A \ p ⊆ B \ p = B ∗ .
Conversely, assume that b ∈ B. If b ∈ A, then b−1 ∈ A and b−1 ∈ p = mB .
Therefore, b = 1/b−1 ∈ Ap .
(c) Let B and B be overrings of A in K. Assume for the sake of contradiction that
there exist elements b ∈ B \B and b ∈ B \B. Consider a := b−1 b ∈ K ∗ . We have
a ∈ B (for otherwise b = ba ∈ B) and a −1 ∈ B (for otherwise b = b a −1 ∈ B ).
In particular, a ∈ A and a −1 ∈ A, contradiction. Hence, B ⊆ B or B ⊆ B.
We will see in Sect. 2.4 that every two arbitrary ideals of a valuation ring are
comparable.
2.3 Integral Elements
Definition 2.3.1 Let B be a ring and A ⊆ B a subring.

(a) An element b ∈ B is called integral over A if there exist n ∈ N and a1 , . . . , an ∈
A such that
bn + a1 bn−1 + · · · + an = 0.
(It is essential that bn has coefficient 1!)

(b) If every b ∈ B is integral over A, then B is called integral over A and A ⊆ B is
called an integral ring extension.
Proposition 2.3.2 If B is a ring and A ⊆ B a subring, then the elements of B that

are integral over A form a subring of B.
Recall that an A-module M is called faithful if for every 0 = a ∈ A we have
aM = 0. We require the following
Lemma 2.3.3 An element b ∈ B is integral over A if and only if there exists a
faithful A[b]-module M that is finitely generated as an A-module.
Proof If b is integral over A, then A[b] is a finitely generated A-module. Con-
versely, assume that there exists a module M as in the statement of the lemma and
2.3 Integral Elements 61
write M = Au1 + · · · + Aun . There exist elements aij ∈ A (with 1 ≤ i, j ≤ n) such

that bui = j aij uj for i = 1, . . . , n. Thus we have a system of equations

(aij − bδij ) uj = 0 (i = 1, . . . , n)
j
where δij denotes the Kronecker symbol. Let f (t) ∈ A[t] be the characteristic
polynomial of the matrix (aij ). Since for every n × n-matrix S we have S S =

SS = det(S) · 1 (where S denotes the adjugate of S), it follows that f (b)ui = 0 for
i = 1, . . . , n, and so f (b)M = 0. Since M is faithful, we have f (b) = 0, which
shows that b is integral over A by Definition 2.3.1.
Proof of Proposition 2.3.2 If b, b ∈ B are integral over A, then A[b, b ] is a
finitely generated A[b]-module and A[b] is a finitely generated A-module. Hence,
A[b, b ] is also a finitely generated A-module. It is a faithful A[c]-module for every
c ∈ A[b, b ], and the statement follows from Lemma 2.3.3.
Definition 2.3.4
(a) Let B be a ring and A ⊆ B a subring. The subring of elements of B that are
integral over A is called the integral closure of A in B. If the integral closure of
A in B is equal to A, then A is called integrally closed in B.
(b) A ring A is called integrally closed if it has no zero divisors and coincides with
its integral closure in Quot A.
If {A
α } is a family of subrings of a ring B and if all Aα are integrally closed in
B, then Aα is also integrally closed in B.
α
Theorem 2.3.5 (Cohen–Seidenberg) Let A ⊆ B be an integral ring extension.
(a) For every prime ideal p of A there exists a prime ideal q of B such that p = A∩q.
(b) For every prime ideal q of B we have: q is maximal in B ⇔ A ∩ q is maximal
in A.
Proof We first prove (b). It suffices to show that for every integral extension K ⊆ L
of rings K and L without zero divisors, K is a field if and only if L is a field
(consider the integral extension A/A ∩ q ⊆ B/q). If K is a field and 0 = b ∈ L,
then there exists a monic irreducible polynomial f ∈ K[t] such that f (b) = 0,
and it follows that b−1 ∈ L. If L is a field and 0 = a ∈ K, then there exist
a1 , . . . , an ∈ K such that a −n + a1 a 1−n + · · · + an = 0, and multiplying by a n−1
shows that a −1 ∈ K.
Next we prove (a). Let p be a prime ideal of A and let S := A \ p. The natural
homomorphism Ap = S −1 A → S −1 B is injective, and Ap ⊆ S −1 B is also an
integral ring extension. Choosing a maximal ideal q of S −1 B gives q ∩ Ap = pAp
by (b) (Ap is local), and it follows that q ∩ A = p for the preimage q of q in B.
Proposition 2.3.6 Valuation rings are integrally closed.

Proof Let A be a valuation ring of K, b ∈ K, and a1 , . . . , an ∈ A with
bn = a1 bn−1 + · · · + an .
If b ∈ A, then b−1 ∈ mA , and so 1 = a1 b−1 + · · · + an b−n ∈ mA , a contradiction.
It follows that every intersection of valuation rings of a field is integrally closed.

The converse is also true:
the integral
Theorem 2.3.7 (Krull [67]) Let K be a field, A ⊆ K a subring, and A
closure of A in K.
is the intersection of all valuation rings B of K with A ⊆ B.
(a) A
(b) In (a) we may restrict ourselves to those valuation rings B of K with A ⊆ B for
which p := A ∩ mB is a maximal ideal of A and A/p ⊆ κ(B) is an algebraic
field extension.
We consider another definition before turning to the proof of this theorem:
Definition 2.3.8 If A and B are local subrings of a field K, then we say that A is
dominated by B if A ⊆ B and mA ⊆ mB (and so, mA = A ∩ mB ). When this
happens, A ⊆ B induces an embedding κ(A) ⊆ κ(B).
Note that domination is a transitive relation and hence determines a partial
ordering on the set of local subrings of K.

Proof of Theorem 2.3.7 The inclusion A ⊆ B follows from Proposition 2.3.6.
Then x ∈ A[x −1],
Conversely, assume that x ∈ K is not integral over A, i.e., x ∈ A.
for otherwise there would exist an equation x = a1 x
n n−1 + · · · + an with ai ∈ A.
Therefore, A[x −1] has a maximal ideal q such that x −1 ∈ q. Let p := A ∩ q. Then
A/p → A[x −1 ]/q is an isomorphism, and p is in particular a maximal ideal of A.
Let A := A[x −1 ]q . By Zorn’s Lemma there exists a local ring B ⊆ K which is
maximal with respect to the properties
(1) B dominates A , and
(2) κ(B) is algebraic over κ(A ).
Since x −1 ∈ q ⊆ mA ⊆ mB we have x ∈ B. The proof is complete once we have
established that B is a valuation ring of K. Thus, let y ∈ K ∗ . We must show that
y ∈ B or y −1 ∈ B.
Assume first that y is not integral over B. Upon replacing A by B and x by y in
the arguments above, we obtain a local ring C ⊆ K that dominates B, and for which
y −1 ∈ C and κ(C) is algebraic over κ(B). Then C also satisfies (1) and (2), and it
follows from the maximality of B that B = C, and thus y −1 ∈ B.
On the other hand, if y is integral over B, then B[y] contains a maximal ideal n
with mB = B ∩ n by Theorem 2.3.5. Hence, the local ring C := B[y]n dominates
B. Furthermore, the extension κ(B) ⊆ κ(C) is algebraic since κ(C) is generated
2.4 Valuations, Ideals of Valuation Rings 63
over κ(B) by the residue class of y and y satisfies an integrality equation. It follows
again that B = C, and so y ∈ B.
Corollary 2.3.9 The valuation rings of K are precisely those local subrings that
are not properly dominated by any other local subring.
Proof It follows from Theorem 2.3.7 that the integral closure of a local subring
A ⊆ K is equal to the intersection of all valuation rings of K that dominate A. Thus,
if A is not properly dominated by any local subring, then A must be a valuation ring.
The converse direction follows from Proposition 2.2.9.
Corollary 2.3.10 For every subring A of a field K and every prime ideal p of A
there exists a valuation ring B of K such that A ⊆ B, p = A ∩ mB , and κ(B) is
algebraic over κ(p) = Quot A/p.
Proof Choose B such that B dominates the local ring Ap and κ(B) is algebraic over
κ(Ap ) = κ(p) (Theorem 2.3.7). Then A ∩ mB = A ∩ Ap ∩ mB = A ∩ p Ap = p.
Corollary 2.3.11 The fields K that have no proper (i.e., different from K) valuation
rings are precisely the algebraic field extensions of finite fields.
Proof Let K be algebraic over Fp . Since every subring of K contains the prime field
Fp , K is the only subring of K that is integrally closed in K. Conversely, let K be
a field without any proper valuation rings. Then char K =: p > 0, for otherwise K
would be integral over Z by Theorem 2.3.7, contradicting that Z is integrally closed
in Q. If there was a t ∈ K, transcendental over Fp , then t −1 would not be integral
over Fp [t], a contradiction.
2.4 Valuations, Ideals of Valuation Rings
Throughout this section K denotes an arbitrary field.

Let be an ordered abelian group and ∞ an element not in . We write the
disjoint union ∪ {∞} simply as ∪ ∞. Declaring
α<∞ and α+∞=∞+α =∞+∞=∞
for all α ∈ gives ∪ ∞ the structure of a totally ordered semigroup.

Definition 2.4.1 Let K be a field and an ordered abelian group. A valuation of
K with values in is a map v : K → ∪ ∞ such that
(V1) v(a) = ∞ ⇔ a = 0;
(V2) v(ab) = v(a) + v(b);
(V3) v(a + b) ≥ min{v(a), v(b)}
for all a, b ∈ K. If there exists an a ∈ K ∗ with v(a) = 0, then v is called non-trivial.

Otherwise, v is called trivial. The ordered abelian group v(K ∗ ) is called the value
group of v and is denoted v . We will sometimes call valuations Krull valuations in
order to distinguish them from absolute values, cf. Sect. 2.6.
Remarks 2.4.2 Let v : K → ∪ ∞ be a valuation of K.

(1) v is determined by its restriction to K ∗ . Valuations of K can also be defined as
those group homomorphisms K ∗ → that satisfy (V3) for all a, b ∈ K ∗ with
a + b = 0.
(2) For all roots of unity a ∈ K ∗ , we have v(a) = 0 since is torsion free.
Exercise 2.4.3 Let a, b ∈ K with v(a) = v(b). Show that equality holds in (V3),
i.e., that v satisfies the following stronger property:
v(a + b) = min{v(a), v(b)}.
Example 2.4.4 Let A be a unique factorization domain and K := Quot A. Every

prime element π of A defines a valuation vπ : K ∗ → Z of K, the order at π, as
follows: vπ (π e ab−1 ) := e for e ∈ Z and a, b ∈ A not divisible by π. The valuation
ring AπA of K (cf. Proposition 2.2.8) is the “valuation ring belonging to vπ ” in the
sense of:
Proposition 2.4.5 Let v : K → ∪ ∞ be a valuation of K. Then
ov := {a ∈ K : v(a) ≥ 0}
is a valuation ring of K with maximal ideal
mv := {a ∈ K : v(a) > 0}.
We call ov the valuation ring of K associated to v.

Proof If a ∈ K and a ∈ ov , then v(a) < 0 and so a −1 ∈ mv since v(a −1 ) =
−v(a) > 0.
Conversely, we will show that every valuation ring comes from a valuation
(which is essentially unique), hence the name.
Observe that if v : K ∗ → is a valuation, assumed to be surjective, then v and
can be obtained from the valuation ring ov as follows: since Ker v = o∗v , the
valuation v induces a group homomorphism v : K ∗ /o∗v −→ ∼ . For all a, b ∈ K ∗
we have
v(a) ≤ v(b) ⇔ v(a −1 b) ≥ 0 ⇔ a −1 b ∈ ov ,

and so v is an order preserving isomorphism if we order K ∗ /o∗v via
a o∗v ≤ b o∗v :⇔ a −1 b ∈ ov .
(The multiplicative notation for this ordered abelian group should not be a source of
confusion for the reader!)
Now, let A be an arbitrary valuation ring of K. For all a, b ∈ K ∗ we define
analogously,
aA∗ ≤ bA∗ :⇔ a −1 b ∈ A.
It is immediately clear that this definition turns K ∗ /A∗ into an ordered abelian
group.
Definition 2.4.6 Let A be a valuation ring of K. The value group of A is the
ordered abelian group A := K ∗ /A∗ with ordering as above. The canonical
epimorphism vA : K ∗ → A is called the canonical valuation of K associated
to A.
The second definition is justified by:
Proposition 2.4.7 Let A be a valuation ring of K. Then vA : K → A ∪ ∞ is a
valuation of K and the associated valuation ring satisfies ovA = A.
Proof The map vA |K ∗ is a homomorphism, so it remains to show that (V3) holds.
Let a, b ∈ K ∗ with a + b = 0 and vA (a) ≤ vA (b). Then a −1 b ∈ A and it follows
that vA (a + b) = (a + b)A∗ = a(1 + a −1 b)A∗ ≥ aA∗ = vA (a). Clearly, ovA = A.
Conversely, if v : K → ∪ ∞ is a valuation with valuation ring A := ov ,

then there exists precisely one order preserving embedding i : A → such that
v = i ◦ vA . Therefore, a valuation of K can be interpreted as a pair (A, i), where A
is a valuation ring of K and i : A → an order preserving embedding of ordered
abelian groups.
Definition 2.4.8 Let v and v be valuations of K. Then v is a coarsening of v if
ov ⊆ ov . If ov = ov , then v and v are equivalent.
Lemma 2.4.9 Let v : K ∗ → and v : K ∗ → be valuations of K, and assume

that v is surjective. The following statements are equivalent:
(i) v is a coarsening of v;
(ii) there exists an order preserving homomorphism ϕ : → such that v =
ϕ ◦ v.
Proof (i) ⇒ (ii): Let A := ov and A := ov . We have a commutative diagram
where π : K ∗ /A∗ → K ∗ /A ∗ denotes the canonical epimorphism, i and i are order

preserving, and i is an isomorphism. Let ϕ := i ◦ π ◦ i −1 .
(ii) ⇒ (i): This is trivial.
Definition 2.4.10 Let (K, P ) be an ordered field and v : K → ∪ ∞ a valuation

of K. Then v and P are compatible if ov is a convex subring of K with respect to
P . A number of equivalent characterizations is given in Proposition 2.2.7.
Example 2.4.11 Every ordered abelian group occurs as the value group of a
surjective valuation. Indeed, let k be a field and A := k[+ ] the semigroup algebra
of + := {α ∈ : α ≥ 0}. This means that A has a k-vector space basis
{xα : α ∈ + } with xα = xβ for α = β, and multiplication in A is given by
xα xβ = xα+β for all α, β ∈ + .
For 0 = a = α∈+ aα xα ∈ A (with aα ∈ k and almost all aα = 0) let

v (a) := min{α ∈ + : aα = 0}.
Let K := Quot A (note that A has no zero divisors). Then

v extends uniquely to a
surjective homomorphism v : K ∗ → , and v is a valuation of K.
Exercise 2.4.12 Verify that:

(1)
v (ab) =
v (a) +
v (b) for all a, b ∈ A \ {0};
(2)
v (a + b) ≥ min{v (a),v (b)} if a + b = 0.
Example 2.4.13 We continue with the set-up of the previous example and assume
in addition that k has an ordering P . Then P induces an ordering Q of K as follows:
for 0 = a = α aα xα ∈ A we define the “leading coefficient” of a as
L(a) := av(a) (∈ k ∗ ),
and we let Q := {0} ∪ { ab : 0 = a, b ∈ A and L(ab) ∈ P }. Now Q is indeed an

ordering of K since L(ab) = L(a)L(b) for all a, b ∈ A \ {0}, and if a + b = 0, then

L(a) + L(b) if v(a) = v(b) and L(a) + L(b) = 0
L(a + b) = .
L(a) if v(a) < v(b)
We claim that the valuation ring ov is convex in K with respect to Q. To see this,
let 0 = x ∈ mv . There exist a, b ∈ A with v(a) = v(b) = 0, and 0 < α ∈
such that x = xα · a/b. It follows that 1 + x = b−2 (b2 + abxα ) ∈ Q since
L(b2 + abxα ) = L(b)2 ∈ P . This shows that 1 + mv ⊆ Q, from which the claim
follows by Proposition 2.2.7.
Example 2.4.14 Let (K, P ) be an ordered field, F ⊆ K a subfield, and A =

oP (K/F ) its convex hull in K. Then A and vA can be described as follows: we
call a, b ∈ K ∗ archimedean equivalent over F if there exist λ, μ ∈ F ∗ such that
|a| ≤ λ |b| and |b| ≤ μ |a|. Then A is the (multiplicative) group of the resulting
equivalence classes, ordered by the inverse absolute value, and vA is the residue
map.
Recall that an abelian group G is called divisible if nG = G for all n ≥ 1. The
torsion free divisible abelian groups are precisely the Q-vector spaces.
Example 2.4.15 The value group A of any valuation ring A of a real closed or
algebraically closed field is divisible (and thus a Q-vector space), since a real closed
field contains all roots of all positive elements and an algebraically closed field
contains all roots of all elements.
We will now investigate the ideals of valuation rings.
Definition 2.4.16 Let (M, ≤) be an ordered set and X ⊆ M a subset (where X = ∅
is allowed). We call X an upper subset of M if for all x ∈ X and y ∈ M,
x ≤ y ⇒ y ∈ X.
Proposition 2.4.17 (Ideals of Valuation Rings) Let A be a valuation ring of K

with surjective valuation v = vA : K → ∪ ∞. Let + := {α ∈ : α ≥ 0}.
(a) The map
a → v(a \ {0})
defines an inclusion preserving bijection from the ideals of A to the upper

subsets of + . The inverse map is given by
M → v −1 (M ∪ {∞}).
(b) The map
p → v(A∗p ) = v(A \ p) ∪ −v(A \ p)

defines an inclusion reversing bijection from the prime ideals of A to the convex
subgroups of . The inverse map is given by
→ {0} ∪ v −1 (+ \ ).
(c) The ideals of A form a chain (i.e., are totally ordered by inclusion).
(d) A is a Bézout ring, i.e., A has no zero divisors and every finitely generated ideal
is principal. More precisely: for a1 , . . . , an ∈ A and a = Aa1 + · · · + Aan , if
v(a1 ) ≤ v(ai ) for i = 1, . . . , n, then√
a = Aa1 .
(e) If a = A is a radical ideal (i.e., a = a), then a is a prime ideal.
Proof The proofs of (a) and (b) are elementary, (c) follows immediately from (a),
and (d) is clear.
(e) Let a, b ∈ A with ab ∈ a. We may assume that a | b in A, i.e., b = ac for
some c ∈ A. Then b2 = abc ∈ a, and so b ∈ a.
Recall that a sequence
ϕi−1 ϕi
· · · → Gi−1 −−→ Gi −
→ Gi+1 → · · ·
of abelian groups and homomorphisms is called exact if Im(ϕi−1 ) = Ker(ϕi ) for all
i. Sequences of the form
0→G →G→G →0
are called short.

Proposition 2.4.18 Let A and B be valuation rings of K, and assume that A ⊆ B.
Let π : B → κ(B) denote the residue map of B.
(a) C := π(A) = A/mB is a valuation ring of κ(B).
(b) There exists a natural short exact sequence
0 → C → A → B → 0
of order preserving homomorphisms.

Proof (a) If c ∈ κ(B)∗ and b ∈ B ∗ with c = π(b), then b ∈ A or b−1 ∈ A, and so
c ∈ C or c−1 = π(b−1 ) ∈ C.
(b) Since A∗ ⊆ B ∗ , there is an order preserving epimorphism A = K ∗ /A∗
K /B ∗ = B whose kernel is the (convex) subgroup B ∗ /A∗ =: B/A of A .
∗
Furthermore, π|B ∗ : B ∗ → κ(B)∗ is an epimorphism, and π −1 (C ∗ ) = A∗ .

Therefore π induces an isomorphism B/A = B ∗ /A∗ → ∼ κ(B)∗ /C ∗ = , which
C
is clearly order preserving.
Corollary 2.4.19 Let A be a valuation ring of K with natural valuation vA : K ∗ →

A . Then there exists a canonical inclusion preserving bijection from the set of
overrings B of A in K to the set of convex subgroups of A , namely
B → B/A := B ∗ /A∗ = vA (B ∗ ).
The inverse map is

−1
→ {0} ∪ vA ( ∪ + ).
Furthermore, for each overring B of A there is a canonical order preserving

isomorphism
∼ .
A / B/A −→ B
Proof This follows from Corollary 2.2.10 and Propositions 2.4.17 and 2.4.18
Definition 2.4.20
(a) Let be an ordered abelian group. The rank of , denoted rank , is the number
of convex subgroups of that are different from (a natural number or ∞).
(b) Let A be a valuation ring. The rank of A, denoted rank A, is defined as the rank
of the value group A = K ∗ /A∗ , where K = Quot A.
Remarks 2.4.21
(1) Many authors (Bourbaki, for example) refer to the rank of as the height of .
(2) By Corollaries 2.4.19 and 2.2.10, the rank of a valuation ring A is also equal to
the number of prime ideals of A that are different from {0} (the Krull dimension
of A), as well as the number of overrings of A in K that are different from K.
(3) The valuation rings of rank 0 are the fields. The ordered abelian groups of rank 1
are precisely the nontrivial subgroups of (R, +) (Theorem 2.1.10). The only
“discrete” ones among these are the infinite cyclic groups, which justifies the
following definition:
Definition 2.4.22 A valuation ring A is called discrete of rank one if the value
group A is infinite cyclic.
Proposition 2.4.23 Let A be a valuation ring which is not a field. The following
are equivalent:
(i) A is discrete of rank one;
(ii) A is noetherian;
(iii) A is a principal ideal domain.
Proof (i) ⇒ (ii): This follows from Proposition 2.4.17(a). (ii) ⇒ (iii): This follows
from Proposition 2.4.17(d). (iii) ⇒ (i): Let mA = Aπ. In particular, π is a prime
element of A. If π is a further prime element of A, then π | π or π | π (since A

is a valuation ring), and thus π and π are associated, i.e., Aπ = Aπ . Since A is
factorial, every element 0 = a ∈ A is of the form a = uπ e with u ∈ A∗ and e ≥ 0.
Hence A is cyclic with positive generator v(π).
Exercise 2.4.24 Show that a field K has a non-trivial valuation ring with real
residue field if, and only if, K has a non-archimedean ordering.
Exercise 2.4.25 Let k be a field, let t be a variable, and let V be the set of all non-
trivial valuation rings of k(t) that contain k. Show that there is a natural bijection
between V and P ∪ {∞},where P is the set of monic irreducible polynomials in k[t]
and ∞ is an additional symbol. Every B ∈ V is a discrete valuation ring of rank
one.
Exercise 2.4.26 Let k be a field and n ∈ N, put x = (x1 , . . . , xn ). For each

polynomial f = α∈Zn+ cα x α in k[x] let

v(f ) := inf |α| : cα = 0
(multinomial notation, with |α| = i αi as usual). Show the following:

(a) v extends to a valuation v of k(x) = k(x1 , . . . , xn ) which is discrete of rank
one. The valuation ring ov and its maximal ideal mv satisfy k[x] ⊆ ov and
mv ∩ k[x] = (x1 , . . . , xn ).
(b) The residue field of v is isomorphic to the rational function field
k(y1 , . . . , yn−1 ) in n − 1 variables.
2.5 Residue Fields and Subfields of Convex Valuation Rings
Throughout this section K denotes an arbitrary field.

Theorem 2.5.1 Let A be a valuation ring of K and let κ(A) = A/mA .
(a) If K is real closed, then so is κ(A).
(b) If K is real closed, then:
√
κ(A) is real closed ⇔ −1 ∈ κ(A) ⇔ A is convex in K.
Proof Let f ∈ A[t] be a monic polynomial. If f has a root in K, this root is already
in A since A is integrally closed. In particular, the mod mA reduced polynomial has
a root in κ(A), from which (a) follows immediately.
2.5 Residue Fields and Subfields of Convex Valuation Rings 71
Assume now that K is real closed. It follows from the previous argument that
κ(A) has no proper field extensions of odd degree and that P := κ(A)2 satisfies
the ordering axioms (O1) and (O2) (Sect. 1.1). Therefore, κ(A) is real closed if and
only if −1 ∈ κ(A)2 (Proposition 1.5.2).
If A is convex in K, then mA ∩ [1, ∞[ = ∅ (Proposition 2.2.7). Hence, 1 + a 2 ∈
mA for a ∈ A, and so −1 ∈ κ(A)2. Conversely, assume that A is not convex. Then
there exists a ∈ K such that 1 + a 2 ∈ mA (Proposition 2.2.7). It follows that a ∈ A
for if not, we have a −1 ∈ mA and thus 1 = a −2 (1 + a 2 ) − a −2 ∈ mA . We conclude
that −1 ∈ κ(A)2.
Let A be a valuation ring of K. If K is ordered and A is convex in K, then A
contains subfields of K (e.g., Q). More generally, one sees easily that A contains a
subfield if and only if char K = char κ(A) (the so-called “equal characteristic case”;
the unequal characteristic case only occurs when char K = 0 and char κ(A) > 0).
Every subfield F of A is contained in a maximal subfield of A (Zorn’s Lemma).
Also, via F → A → κ(A) we always consider F as a subfield of κ(A).
Proposition 2.5.2 Let A be a valuation ring of K and F a maximal subfield of A.
Then F is algebraically closed in K, and the field extension κ(A)/F is algebraic.
Proof Since A is integrally closed in K, it contains the algebraic closure of F in
K. Let π : A → κ(A) be the residue homomorphism. If there were an a ∈ A
with π(a) transcendental over π(F ), we would have F [a] ∩ mA = {0}, and thus
F F (a) ⊆ A, contradicting the maximality of F .
Proposition 2.5.3 Let R be a real closed field, A a convex subring of R and F a

maximal subfield of A. Then the restriction of π : A → κ(A) to F is an isomorphism
from F to κ(A).
(In the case of algebraically closed fields, there is an analogous statement for
arbitrary valuation rings of the field.)
Proof Since F is algebraically closed in R (Proposition 2.5.2), it is real closed
(Corollary 1.5.5). Since κ(A) is algebraic over π(F ) (Proposition 2.5.2) and κ(A)
is real (Theorem 2.5.1), we have κ(A) = π(F ).
Proposition 2.5.4 Let (K, P ) be an ordered field and A a convex subring of K.

Then A is the convex hull of any of its maximal subfields.
Proof Let F be a maximal subfield of A and let B := oP (K/F ). Consider an
element a ∈ A. By Proposition 2.5.2 there exist n ∈ N and elements b1 , . . . , bn ∈ F
such that a n + b1a n−1 + · · ·+ bn ∈ mA ⊆ mB . Assume that a ∈ B. Then a −1 ∈ mB ,
and multiplication with a −n gives 1 + (b1a −1 + · · · + bn a −n ) ∈ mB , a contradiction.
Definition 2.5.5 Let (K, P ) be an ordered field. We call a subfield F of K

archimedean saturated in K (with respect to P ) if K has no proper overfields F of
F that are archimedean over F .
It follows that the subfields of K that are archimedean saturated in K are

precisely the maximal subfields of convex subrings of K.
Proposition 2.5.6 Let A be a convex subring of an ordered field K and F ⊆ A a
subfield, then tr.deg.(K/F ) ≥ rank A.
Proof Let A = A0 A1 · · · Ar = K be a chain of overrings of A. Let
F0 be a maximal overfield of F in A0 and Fi a maximal overfield of Fi−1 in Ai ,
where i = 1, . . . , r. Since Ai = o(K/Fi ) for i = 0, . . . , r, it follows that all Fi
are different. By Proposition 2.5.2 we have tr.deg.(Fi /Fi−1 ) ≥ 1 for i = 1, . . . , r,
hence tr.deg.(K/F ) ≥ r.
Proposition 2.5.7 Let K be a real closed field and F ⊆ K a subfield. If
F ⊆ F0 F1 · · · Fr = K
and
F ⊆ F0 F1 · · · Fs = K
are two chains of archimedean saturated subfields Fi , Fi of K that cannot be made

any finer, then r = s and Fi ∼
= Fi for i = 0, . . . , r.
Proof Since the o(K/Fi ), resp. the o(K/Fi ), are precisely the distinct convex
overrings of F in K (Proposition 2.5.4), we have r = s = rank o(K/F ) and
o(K/Fi ) = o(K/Fi ) for i = 0, . . . , r. Then, by Proposition 2.5.3, Fi ∼ =
κ o(K/Fi ) ∼= Fi .
Example 2.5.8 (E. Artin) It is essential that K is real closed for the isomorphism
Fi ∼= Fi !
Let R0 be the real closure of Q (i.e., the relative algebraic closure of Q in R) and
F the relative algebraic closure of Q(e) in R (note that e = 2.71828 . . . is transcen-
dental). Consider K = F (t), equipped with the ordering P0,+ (Example 1.1.13(3)).
Then 0 < t < a for all 0 < a ∈ F . Finally, let F := R0 (e + t) ⊆ K. Since F
is not real closed, F and F are not isomorphic. We claim that not only F , but also
F , is archimedean saturated in K. This is clear for F . To see this for F , we need
to cite a result from field theory (cf. e.g., [54, Vol. 3, p. 199]):
If E/k is a field extension such that k is (relatively) algebraically closed in
E, and if E(x)/E is a simple transcendental extension, then k(x) is also
algebraically closed in E(x).
In the above situation, it suffices to verify that F is algebraically closed in K
since tr.deg.(K/F ) = 1 and o(K/F ) = o(K/Z) = K. This, however, is an
immediate application of the cited theorem (R0 is algebraically closed in F and
K = F (e + t)).
2.6 The Topology of Ordered and Valued Fields 73
In this example, one should also observe that R0 (e) is a subfield of K which
is order isomorphic to F , but which is not archimedean saturated in K (F is
archimedean over R0 (e)).
2.6 The Topology of Ordered and Valued Fields
Let (K, P ) be an ordered field. The open intervals
]a, b[P = {x ∈ K : a <P x <P b}
form an open basis of a topology on K that we denote by TP . For n ≥ 1, we

also denote the induced product topology on K n by TP and remark that K n is a
Hausdorff space with respect to this topology. Moreover, (K, TP ) is a topological
field (i.e., the maps (x, y) → x − y and (x, y) → xy from K × K to K are
continuous, as is the map K ∗ → K ∗ , x → x −1 ). It follows that for every rational
function f = g/ h ∈ K(t1 , . . . , tn ) (with g, h polynomials), the evaluation map
x → f (x) from {x ∈ K n : h(x) = 0} to K is continuous. We usually call TP
the strong topology on K n (with respect to P ) in order to distinguish it from the
“weaker” Zariski topology (see Sect. 3.1).
There is however really just one case where the topology TP has satisfying prop-
erties, namely when K = R. This is one of the reasons that makes semialgebraic
topology (see [31–33] and the references in [33]) indispensable when engaging in
geometry over arbitrary real closed fields:
Proposition 2.6.1 Let (K, P ) be an ordered field. If K = R, then the topological
space (K, TP ) is totally disconnected. For a, b ∈ K with a < b, the interval [a, b]
is not compact and (K, TP ) is in particular not locally compact.
Proof If (K, P ) is archimedean, then there exists an order preserving embedding
K → R, and TP is precisely the subspace topology on K induced by R. If K = R,
the statements can be verified directly.
Assume thus that (K, P ) is not archimedean. Then there exists a proper convex
subring A of K, and A is open and closed in K. Let x ∈ K and 0 < ε ∈ K. Choose
a ∈ K ∗ with ε/a ∈ A. Then x + aA ⊆ ]x − ε, x + ε[ and x + aA is a clopen (i.e.,
open and closed) neighbourhood of x. This shows that K is totally disconnected.
Now let a, b ∈ K with a < b. Then there exists an ε ∈ K, ε > 0, such that
nε < b − a for all n ∈ N (if b − a ∈ mA , then one can choose ε = (b − a)2 ,
otherwise ε ∈ mA suffices). The family {]x − ε, x + ε[ : x ∈ K} contains no finite
subcovering of [a, b].
Assume now that K is an arbitrary field and that v : K → ∪ ∞ is a surjective
valuation of K. The sets
Bα (a) := {x ∈ K : v(x − a) > α} (a ∈ K, α ∈ )

form an open basis of a topology Tv on K and (K, Tv ) is again a topological field.

The topology Tv is the discrete topology if and only if v is the trivial valuation (i.e.,
= 0).
As a consequence of the strong triangle inequality (V3) in Definition 2.4.1, all
sets v −1 (α) (where α ∈ ) are open in K (for every a ∈ K with v(a) = α we have
Bα (a) ⊆ v −1 (α)). Since

K \ v −1 (α) = Bα (0) ∪ v −1 (β),
β<α
we have that v −1 (α) is also closed. All sets Bα (a) in the basis that defines Tv are
thus clopen. In particular, (K, Tv ) is totally disconnected. Furthermore, the obvious
assumption that the closure and boundary of Bα (a) are given by {x : v(x − a) ≥ α}
and {x : v(x − a) = α}, respectively, is false.
Proposition 2.6.2 Let (K, P ) be an ordered field and v : K → ∪ ∞ a nontrivial
surjective valuation, compatible with P . Then the order and valuation topologies
on K coincide: TP = Tv .
Proof For every 0 < a ∈ K there exists α ∈ such that Bα (0) ⊆ ]−a, a[, for
instance α = v(a). Indeed, if x ∈ K and v(x) > α = v(a), then x/a ∈ mv ⊆
]−1, 1[, hence |x| < a. Conversely, for every α ∈ there exists 0 < a ∈ K such
that ]−a, a[ ⊆ Bα (0). Indeed, choose a such that v(a) > α. Then if |x| < a, it
follows that |x/a| < 1, hence v (x/a) ≥ 0 and so v(x) ≥ v(a) > α.
The converse of this theorem is false: there may exist valuations that are not
compatible with P , yet induce the same topology as P . This follows easily from the
following:
Exercise 2.6.3 Let v and v be surjective nontrivial valuations of K. Show that if
v is a coarsening of v, then Tv = Tv .
In addition to Krull valuations we introduce absolute values. In number theory
especially they are indispensable.
Definition 2.6.4 Let K be a field. An absolute value on K is a map f : K → R
such that for all a, b ∈ K:
(A1) f (a) ≥ 0, and f (a) = 0 ⇔ a = 0;
(A2) f (ab) = f (a) f (b);
(A3) f (a + b) ≤ f (a) + f (b).
If instead of (A3) the ultrametric inequality
(A4) f (a + b) ≤ max{f (a), f (b)} for all a, b ∈ K,
is satisfied, then f is called an ultrametric absolute value.
In the literature the term “valuation” is often used instead of absolute value.
Depending on whether it is ultrametric or not, this “valuation” is then called “non-
archimedean” or “archimedean”. Clearly this easily causes confusion.
2.7 The Baer–Krull Theorem 75
Remarks 2.6.5
(1) If ϕ : K → C is a field embedding, then a → |ϕ(a)| is a non-ultrametric
absolute value on K. The converse is essentially also true, as the following
classical theorem of A. Ostrowski ([86], 1916) shows:
If f : K → R is a non-ultrametric absolute value on a field K, then there
exists an embedding ϕ : K → C and a real number s with 0 < s ≤ 1,
such that f (a) = |ϕ(a)|s for all a ∈ K. (See for instance also [55, Vol. II,
§9.5].)
(2) If v : K → ∪ ∞ is a surjective valuation with archimedean value group ,
then one obtains an ultrametric absolute value fv on K as follows: choose an
order preserving embedding i : → R (cf. Sect. 2.1) as well as a real number
c with 0 < c < 1 and let fv (a) := ci(v(a)) (a = 0), fv (0) := 0. Conversely,
every ultrametric absolute value can be “logarithmified” to a valuation of rank
≤ 1. Upon defining an equivalence of absolute values in the obvious way,
one then obtains a bijective correspondence between the equivalence classes
of valuations of rank ≤ 1 and the equivalence classes of ultrametric absolute
values. Together with Remark (1) we thus obtain a description of all absolute
values on a field.
Every absolute value f on K defines a metric on K via (a, b) → f (a − b) and
makes K into a metric topological space. We denote the associated topology by Tf .
Proposition 2.6.6 Let (K, P ) be an ordered field and assume that K is either
archimedean, or that K contains a proper convex subring of finite rank. Then
there exists an absolute value f on K such that TP = Tf . (In particular, TP is
metrizable).
Proof If (K, P ) is archimedean with order preserving embedding ϕ : K → R,
then one obtains f as in Remark 2.6.5(1). In the other case, K contains a maximal
proper convex subring A. The valuation vA = v has rank 1, and one obtains f as
in Remark 2.6.5(2). It is clear that Tf = Tv , and Proposition 2.6.2 gives Tf = TP .
2.7 The Baer–Krull Theorem
We have seen that for every convex subring A of an ordered field (K, P ) there is an
induced ordering P of the residue field κ(A) (Remark 2.2.6(1)). Conversely, given
a valuation ring A of a field K and an ordering Q of κ(A), one can wonder if there
exists an ordering P of K that makes A convex and for which P = Q.
The Baer–Krull Theorem provides a (positive) answer to this question. In fact,
the theorem is considerably more precise in that it determines all such orderings Q.
Consider a surjective valuation v : K ∗ → and let A := ov . Then v induces a
surjective homomorphism v2 : K ∗ /K ∗2 /2. Let {γi : i ∈ I } ⊆ be a subset
such that {γi + 2 : i ∈ I } is a basis of the F2 -vector space /2 and γi > 0 for all
i ∈ I . We choose elements πi ∈ K ∗ such that v(πi ) = γi for each i ∈ I . All πi are

in mA .
The system {πi : i ∈ I } will be kept fixed in what follows and is called a
quadratic system of representatives of K with respect to A. Every a ∈ K ∗ has a
representation of the form

a = uc2 πj (2.1)
j ∈J
with J ⊆ I finite, c ∈ K ∗ and u ∈ A∗ , where J is uniquely determined

−1 by a. (The
set J is characterized by v2 (aK ∗2 ) = (γj + 2); since v a πj ∈ 2 one
then obtains the form (2.1).) j ∈J j ∈J
Theorem 2.7.1 (Baer–Krull) Let A be a valuation ring of K, Q an ordering of

κ(A) and {πi : i ∈ I } a quadratic system of representatives as above. Let YQ be the
set of all orderings P of K such that A is convex with respect to P and Q = P .
Then the map P → signP (πi ) i∈I is a bijection from the set YQ to the set {±1}I .
The theorem says in particular that YQ contains precisely one element when I =
∅, i.e., when = 2.
Proof We denote the homomorphism A → κ(A) by a → a. Let ε = (εi )i∈I ∈
{±1}I . Then there exists at most one P ∈ YQ such that signP (πi ) = εi for all i ∈ I
since for every a ∈ K ∗ the sign of a with respect to P is determined by (2.1). In
order to prove the existence of such a P ∈ YQ we may assume εi = 1 for all i
(replace πi by εi πi for all i ∈ I ).
Let a ∈ K ∗ and let

a = u c2 πj = u c 2 πj
j ∈J j ∈J
2
be two representations of the form (2.1). Then cc ∈ A∗ and, since u = cc u, it
follows that signQ (u ) = signQ (u). Hence σ : a → signQ (u) is a well-defined map
σ : K ∗ → {±1}. This is a homomorphism since, if

a = u c2 πj and b = v d 2 π
j ∈J ∈L
are representations of the form (2.1), then there is a representation

ab = w e2 πm
m∈M
of the form (2.1) with w = uv.

We will now show that P := {0}∪Ker(σ ) is an ordering of K. Since σ (−1) = −1
it is clear that P P ⊆ P , P ∪ (−P ) = K and P ∩ (−P ) = {0}. It remains to show
2.7 The Baer–Krull Theorem 77
that P + P ⊆ P . Let 0 = a, b ∈ P with a + b = 0 and with representations

a = u c2 πj , b = v d2 π
j ∈J ∈L
of the form (2.1). Then u, v ∈ Q. We consider two cases.

Case 1: J = L. After exchanging a and b if needed, we obtain x := c
d ∈ A, and
so

a + b = (uc2 + vd 2 ) πj = (ux 2 + v) · d 2 πj .
j ∈J j ∈J
Since u x 2 + v ∈ Q we conclude that σ (a + b) = 1, i.e., a + b ∈ P .

Case 2: J = L. We have v(a) < v(b) after a possible exchange of a and b. Hence
there exists x ∈ mA with b = ax, and it follows that

a + b = (1 + x)a = (1 + x)u · c2 πj ,
j ∈J
i.e., a + b ∈ P .
Thus we have shown that P is an ordering of K. From the definition of P the
convexity of A with respect to P follows immediately (since 1 + mA ⊆ P ), as does
P = Q.
Corollary 2.7.2 Let A be a residually real valuation ring of K (Definition 2.2.5).

Then there exists an ordering of K that makes A convex.
A special case of this result is contained in Theorem 2.5.1.
Remark 2.7.3 Our formulation of the Baer–Krull Theorem in this section exhibits
a certain arbitrariness due to the choice of a system of representatives. We explain
briefly how the statement of the theorem can be presented in a more “invariant”
manner.
Let v : K ∗ be a surjective valuation with valuation ring A. Let k := κ(A), Y
the set of orderings P of K that make A convex, and Xk the set of all orderings of k.
For an abelian group G we denote by G := Hom(G, S 1 ) its character group, where
S = {ζ ∈ C : |ζ | = 1}. Upon identifying an ordering P of K with the character
1
σP : aK ∗2 → signP (a) of K ∗ /K ∗2 , we consider P as an element of K ∗ /K ∗2 . In
particular, Y becomes a subset of K ∗ /K ∗2 in this way.

A quadratic system of representatives = {πi : i ∈ I } is just a homomorphism
s : /2 → K ∗ /K ∗2 such that v2 ◦ s = id/2 ( corresponds to γi + 2 →
πi K ∗2 ). If we keep such a section s fixed, we obtain a map Y → /2 × Xk ,
namely P → (σP ◦ s, P ). The assertion of Baer–Krull is that this map is a bijection.
In order to avoid an arbitrary choice of the section s in the formulation, we

observe that the epimorphism v2 : K ∗ /K ∗2 /2 induces an embedding
→ K
v2 : /2 we have v2 (χ) = χ ◦ v2 ).
∗ /K ∗2 via dualization (for χ ∈ /2
as a subgroup of K /K via v2 (and Y as a subset of K

Interpreting /2
∗ ∗2 ∗ /K ∗2 ,
· Y ⊆ Y , i.e., /2
as explained above), we obtain /2 acts on Y via translation.
This is precisely what we showed in the essential part of the proof of the theorem.
For χ ∈ /2 and P ∈ Y , the ordering Q := χ · P of K is determined by the
following signature:
a → χ v(a) + 2 · signP (a) (for all a ∈ K ∗ ).

As a result, the map P → P from Y to Xk is invariant under this action of /2
and takes different values on different orbits. Thus we obtain:
Theorem 2.7.4 (Baer–Krull Theorem, “Invariant” Formulation) Let v : K ∗
be a surjective valuation with valuation ring A and residue field k = A/mA . Let
Y be the set of all orderings of K compatible with v, and Xk the set of all orderings
on Y and a canonical bijection
of k. Then there is a natural free action of /2

→ Xk .
Y /2
2.8 Places
The operations
Let K and L be fields. We denote the disjoint union K ∪ {∞} by K.

+ and · are partially extended to K as follows:
a + ∞ = ∞ + a = ∞ (for all a ∈ K)
a·∞=∞·a =∞ a = 0).
(for all a ∈ K,
The expressions ∞ + ∞, 0 · ∞ and ∞ · 0 are not defined. The construction of

inverses with respect to + and · is extended via
−∞ = ∞ and 0−1 = ∞, ∞−1 = 0.
As usual we write x − y = x + (−y) and xy = xy −1 whenever the right hand

side is defined. One can verify that the associative and distributive laws are satisfied
whenever “everything” is defined.
→
Definition 2.8.1 A place of K with values in L is a map λ : K L such that for

all a, b ∈ K:
(P1) if λ(a) + λ(b) is defined, then a + b is defined and λ(a + b) = λ(a) + λ(b);
(P2) if λ(a)λ(b) is defined, then ab is defined and λ(ab) = λ(a)λ(b);
(P3) λ(1) = 1.
2.8 Places 79
Let k ⊆ K be a subfield and j : k → L a fixed embedding. If λ(a) = j (a) for all

−
a ∈ k, then λ is called a place over k (with respect to j ) and we write K (there
→L
k
is no confusion about j ).
Exercise 2.8.2 Let A be a valuation ring of K, L := κ(A) and a → a the residue

map A → L. Show that the map λ : K → L defined by λ(a) = a if a ∈ A and

λ(a) = ∞ if a ∈ K \ A is a place. We often write λA instead of λ and call λA the
canonical place of K associated with the valuation ring A.
Exercise 2.8.3 Show that every field homomorphism ϕ : K → L defines a place

→
ϕ: K
L via
ϕ |K = ϕ and
ϕ (∞) = ∞. Such places are called trivial.
We will shortly see that canonical places and trivial places are essentially the
only examples of places.
→L
Proposition 2.8.4 (Properties of Places) Let λ : K be a place. Then:
(a) λ(0) = 0 and λ(∞) = ∞.

(b) λ(−a) = −λ(a) and λ(a −1 ) = λ(a)−1 for all a ∈ K.
→M
(c) If μ : L is a further place, then μ ◦ λ : K
→M is also a place.
Proof (a) If λ(∞) = ∞, then λ(∞) + λ(∞) is defined, and thus also ∞ + ∞,
contradicting (P1). If λ(0) = 0, then λ(0) · λ(∞) is defined, and thus also 0 · ∞,
contradicting (P2).
(b) This is clear for a ∈ {0, ∞} by (a). Thus let a ∈ K ∗ . If λ(a) = λ(−a) = ∞,
then λ(−a) = −λ(a). Otherwise λ(a)+λ(−a) is defined and it follows from (a) that
λ(a) + λ(−a) = λ(0) = 0, hence λ(−a) = −λ(a). Similarly, λ(a −1 ) = λ(a)−1 .
(c) This is trivial.
Proposition 2.8.5 (Relationship Between Places and Valuation Rings)

→
(a) If λ : K L is a place, then
A := λ−1 (L) = {a ∈ K : λ(a) = ∞}
is a valuation ring of K with maximal ideal m = λ−1 (0). There is exactly

one homomorphism ϕ : κ(A) → L such that λ = ϕ ◦ λA (cf. Exercises 2.8.2
and 2.8.3).
(b) Conversely, if A is a valuation ring of K and ϕ : κ(A) → L a homomorphism,
then A = λ−1 (L) for the place λ := →
ϕ ◦ λA : K L.
In other words, a place of K with values in L is “the same” as a valuation ring A
of K together with a homomorphism κ(A) → L.
Proof (a) It is clear that A is a subring of K and that λ|A : A → L is a

homomorphism. If a ∈ K and a ∈ A, then λ(a) = ∞ and so λ(a −1 ) = 0 by
Proposition 2.8.4(b). It follows that A is a valuation ring of K with maximal ideal
λ−1 (0) and that ϕ : κ(A) → L is the homomorphism induced by λ|A .
(b) This is clear.
Notation 2.8.6
(a) A place λ : K→L will from now on be written as λ : K → L ∪ ∞. If ϕ : K →
L is a homomorphism, we denote its associated trivial place K → L ∪ ∞ by ϕ
as well (instead of by
ϕ ).
(b) Let λ : K → L ∪ ∞ be a place. We denote the valuation ring λ−1 (L) of K
by oλ , its maximal ideal by mλ , its residue field oλ /mλ by κλ , and the induced
homomorphism κλ → L by λ. We also denote the value group of oλ by λ :=
K ∗ /o∗λ and the associated canonical valuation by vλ : K ∗ → λ .
Places provide an alternative way of looking at valuations and valuation rings.
We illustrate this in the results below.
Proposition 2.8.7 Let A ⊆ K be a subring. An element a ∈ K is integral over A if
and only if λ(a) = ∞ for every place λ of K that is finite on A.
Proof By Theorem 2.3.7(a).
Theorem 2.8.8 (Chevalley; Extension of Places) Let A ⊆ K be a subring and

assume that L is algebraically closed. Any homomorphism ϕ : A → L can be
extended to an L-valued place of K. In other words, there exists a place λ : K →
L ∪ ∞ such that λ|A = ϕ.
Proof Let p = Ker ϕ and let ϕ : κ(p) = Ap /p Ap → L be the homomorphism
induced by ϕ. By Theorem 2.3.7 there exists a valuation ring B of K which
dominates Ap and such that κ(B) is algebraic over κ(p). It follows that ϕ factors
through a homomorphism ψ : κ(B) → L. Letting λ := ψ ◦ λB finishes the proof.
Next we turn our attention to places of ordered fields.

Definition 2.8.9
(a) A place λ : K → L ∪ ∞ is called real if L is a real field.
(b) Let P and Q be orderings of K and L, respectively. A place λ : K → L ∪ ∞ is
called compatible with P and Q (or order preserving with respect to P and Q)
if λ(P ) ⊆ Q ∪ {∞}.
2.8 Places 81
Proposition 2.8.10 Let (K, P ) and (L, Q) be ordered fields and let λ : K → L∪∞
be a place. The following statements are equivalent:
(i) λ is compatible with P and Q;
(ii) the subring oλ of K is convex with respect to P and the homomorphism
λ : κλ → L is order preserving with respect to P and Q.
Proof (i) ⇒ (ii): It follows from λ(1 + mλ ) = {1} that 1 + mλ ⊆ P , proving the
convexity of oλ with respect to P . It is clear that λ is order preserving.
(ii) ⇒ (i): Let a ∈ P ∩ oλ and let a be the image of a in κλ . Then a ∈ P and so
λ(a) = λ(a) ∈ Q.
Corollary 2.8.11 Let λ : K → L ∪ ∞ be a place. For every ordering Q of L there

exists an ordering P of K such that λ is compatible with P and Q.
Proof By the Baer–Krull Theorem (Sect. 2.7) and Proposition 2.8.10.
Example 2.8.12 If K is a real closed field and (L, Q) an ordered field, then every
place K → L ∪ ∞ is order preserving with respect to Q.
We finish this section with an extension theorem for real places.
Lemma 2.8.13 Let K ⊆ L be a field extension and B ⊆ L a valuation ring of L.
Then A := K ∩ B is a valuation ring of K and mA = K ∩ mB . If L is algebraic
over K, then κ(B) is algebraic over κ(A).
Proof It is clear that A is a valuation ring of K. If a ∈ K ∗ , then
a ∈ mA ⇔ a −1 ∈ A ⇔ a −1 ∈ B ⇔ a ∈ mB .
Assume now that L is algebraic over K and let b ∈ B. There exist a0 , . . . , an ∈ K

such that a0 = 0 and a0 bn + · · · + an = 0. Let m be the smallest index i such that
vA (ai ) ≤ vA (aj ) for j = 0, . . . , n. Division by am gives
b n−m + c1 b n−m−1 + · · · + cn−m = 0
in κ(B), where the cj = aj +m /am are in κ(A). It follows that b is algebraic over
κ(A).
Lemma 2.8.14 Let K ⊆ L be an algebraic field extension, and let B and C be

valuation rings of L such that K ∩ B = K ∩ C and B ⊆ C. Then B = C.
Proof Let A := K ∩ B = K ∩ C. Then there are inclusions κ(A) =
A/mA → B/mC → C/mC = κ(C). Since B/mC is a valuation ring of κ(C) by
Proposition 2.4.18(a) and κ(C) is algebraic over κ(A) by Lemma 2.8.13, it follows
that B/mC = κ(C) and so B = C since valuation rings are integrally closed.
Theorem 2.8.15 (Extension of Real Places) Let (K, P ) be an ordered field, R a

real closed field and λ : K → R ∪ ∞ a P -compatible place. Let L ⊇ K be an
algebraic field extension and Q an extension of the ordering P to L. Then there
exists a unique Q-compatible place μ : L → R ∪ ∞ that extends λ.
Proof The subring A := oλ of K is convex with respect to P and λ : κ(A) → R
is order preserving with respect to P by Proposition 2.8.10. Let B be the Q-convex
hull of A in L. By Lemma 2.8.13 κ(B) is algebraic over κ(A) and Q is an extension
of P . Hence there exists a (unique) order preserving homomorphism ψ : κ(B) → R
such that ψ|κ(A) = λ (by uniqueness of the real closure; Theorem 1.11.2). Therefore
the place μ := ψ ◦ λB : L → R ∪ ∞ satisfies μ|K = λ and is compatible with Q.
To show the uniqueness of μ, assume that μ : L → R∪∞ is also a Q-compatible
extension of λ and let B := oμ . It suffices to show that B = B . Since B is convex
with respect to Q and K ∩ B = K ∩ B = A it follows that B ⊆ B . Hence B = B
by Lemma 2.8.14.
Corollary 2.8.16 Let R be a real closed field and λ : K → R ∪ ∞ a place. There

exists a real closure S of K that admits an extension λS : S → R ∪ ∞ of λ. More
precisely: let S be a real closure of K, then λ has an extension λS : S → R ∪ ∞ if
and only if λ is compatible with the ordering of K induced by S.
Proof By Theorem 2.8.15 and Corollary 2.8.11.
2.9 The Orderings of R(t), R((t)) and Quot R{t}
Let R be a real closed field. In this section we will explicitly determine the orderings
of R(t), R((t)) and Quot R{t}.
We start with the rational function field in one variable R(t). Let c ∈ R. Then c
determines a place λc : R(t) → R ∪ ∞, namely λc (f ) = f (c) (setting f (c) = ∞
if f has a pole at c, as usual). The valuation ring associated to λc is R[t](t −c) . It is
discrete of rank one and t − c is a generator of its maximal ideal. By the Baer–Krull
Theorem, R(t) has precisely two λc -compatible orderings: Pc,+ and Pc,− . These are
characterized by
t − c ∈ Pc,+ and − (t − c) = c − t ∈ Pc,−
and are of course the orderings of Example 1.1.13(3).

There is a further R-valued place of R(t), namely λ∞ : R(t) → R ∪ ∞ which
is defined by λ∞ := λ0 ◦ σ where σ is the R-automorphism of R(t) that satisfies
σ (t) = t −1 . For f ∈ R(t) we again have λ∞ (f ) = f (∞) (correctly interpreted).
The valuation ring of λ∞ is
f
R[t −1 ](t −1) = g : f, g ∈ R[t], g = 0, deg f ≤ deg g .
2.9 The Orderings of R(t), R((t)) and Quot R{t} 83
Its maximal ideal is generated by t −1 , and as above there are precisely two λ∞ -
compatible orderings P∞,+ and P∞,− , characterized by t −1 ∈ P∞,+ and −t −1 ∈
P∞,− . (In a certain sense this notation would be more consistent the other way
around. See the following paragraph however.)
The following geometric interpretation of these orderings may be helpful. If 0 =
f ∈ R(t), then f has only finitely many zeroes and poles. Given any c ∈ R it
therefore makes sense to speak of the sign of f immediately to the right or to the
left of c. Here “f > 0 immediately to the right of c” means that there exists an
ε > 0 such that f (a) > 0 for all a ∈ ]c, c + ε[. Similarly, f (c) has a well-defined
sign for c → +∞ and for c → −∞. Thus we obtain
Pc,+(−) = {0} ∪ {f ∈ R(t)∗ : f > 0 immediately to the right (left) of c},

P∞,+(−) = {0} ∪ {f ∈ R(t)∗ : f (c) > 0 for c 0 (c 0)}.
Proposition 2.9.1 The orderings Pc,+ and Pc,− , for all c ∈ R ∪ {∞}, are precisely
the orderings of R(t) that are non-archimedean over R. If R = R, these are all the
orderings of R(t).
Proof Let P be an ordering of R(t) that is non-archimedean over R and let A :=
oP R(t)/R be the convex hull of R.
Let t ∈ A. Then p := mA ∩ R[t] is a prime ideal of R[t] (the centre of A in
R[t]). Since R[t]p ⊆ A and A = R(t) we have p = (0). Thus p = R[t] · f for some
monic irreducible polynomial f ∈ R[t]. Since κ(A) is a real field and R[t]/(f ) is a
subfield of κ(A) there exists an element c ∈ R such that f = t − c. Since R[t](t −c)
is a valuation ring of R(t) (by Proposition 2.2.8) and is dominated by A, we have
A = R[t](t −c) (by Corollary 2.3.9) and it follows that P = Pc,+ or P = Pc,− .
If t ∈ A, we let u := t −1 and obtain as above that A = R[u](u) . Hence P =
P∞,+ or P = P∞,− .
The statement about R follows from Corollary 2.1.13.
As a consequence of this proof we obtain:
Corollary 2.9.2 Let k be an arbitrary field. Then the nontrivial valuation rings of
k(t) that contain k are precisely the following:
(1) A = k[t](f ) with f ∈ k[t] monic and irreducible;
(2) A = k[t −1 ](t −1) .
In each case the value group is Z and the residue field is a finite simple field
extension of k.
In order to determine the remaining orderings of R(t) we consider for every
ordering P of R(t) the sets
UP := {a ∈ R : t − a ∈ P } and OP := {a ∈ R : a − t ∈ P }.
The pair ηP = (UP , OP ) is a generalized Dedekind cut of R in the following sense:

Definition 2.9.3 Let (M, ≤) be a totally ordered set. A (generalized) Dedekind cut
of M is a pair (U, O) of subsets of M such that U ∪ O = M and u < v for all
u ∈ U and all v ∈ O. If U and O are not empty, the Dedekind cut is called proper.
The orderings of R(t) determined above thus yield
for P = P∞,+ , resp. P∞,− : ηP = (R, ∅), resp. (∅, R);

for P = Pc,+ , resp. Pc,− : ηP = ]−∞, c] , ]c, ∞[ , resp. ]−∞, c[ , [c, ∞[ .
These Dedekind cuts are not free in the following sense:

Definition 2.9.4 A Dedekind cut (U, O) of M is called free if it is proper, U
contains no largest element and O contains no smallest element.
It is clear that the non-free Dedekind cuts of R are precisely those ηP where P
runs through all orderings of R(t) that are non-archimedean over R.
Consider thus a free Dedekind cut ξ = (U, O) of R. We construct an ordering
P = Pξ of R(t) with ξ = ηP that is archimedean over R as follows: Let 0 = f ∈
R(t) and let −∞ < c1 < · · · < cr < ∞ be the distinct zeroes and poles of f in R.
Then f has constant sign ε0 , ε1 , . . . , εr , respectively, on the interval
I0 := ]−∞, c1 [ , Ii := ]ci , ci+1 [ (for i = 1, . . . , r − 1), Ir := ]cr , ∞[ ,
respectively, cf. Proposition 1.7.2. There is precisely one i ∈ {0, . . . , r} such that
Ii ∩ U = ∅ = Ii ∩ O and so we say that f has sign εi at ξ . Thus, if we let
Pξ := {0} ∪ {f ∈ R(t)∗ : f is positive at ξ }

= {f ∈ R(t) : ∃u ∈ U, v ∈ O such that f has no pole in ]u, v[ and f ≥ 0},
then it is immediately clear that P := Pξ is an ordering of R(t) with ξ = ηP .

In fact we have obtained all orderings of R(t) since the converse correspondence
also holds:
Proposition 2.9.5 For every ordering P of R(t) that is archimedean over R, η :=
ηP is a free Dedekind cut of R and P = Pη . Thus there is a canonical bijection
{P : P is an ordering of R(t) that is archimedean over R}

↑↓
{η : η is a free Dedekind cut of R},
given by P → ηP and η → Pη .
Proof Let P be archimedean over R and η := ηP = (UP , OP ). Then UP = ∅ =
OP and for every a ∈ UP there exists 0 < b ∈ R such that b < t − a (otherwise
2.9 The Orderings of R(t), R((t)) and Quot R{t} 85
t − a would be infinitely small compared to R), i.e., such that a + b ∈ UP . Hence

UP does not contain a largest element. In a similar manner we see that OP does not
contain a smallest element. It follows that ηP is free. In order to show that P = Pη
it suffices to prove that P ∩ R[t] = Pη ∩ R[t]. Let 0 = f ∈ R[t] and write

r
s
f (t) = α (t − ai ) (t − bj )2 + cj2
i=1 j =1
with α ∈ R ∗ , ai , bj , cj ∈ R and cj = 0 (for i = 1, . . . , r, j = 1, . . . , s). The

quadratic factors are positive at each ordering of R(t) and for i = 1, . . . , r we have
by definition, t − ai ∈ P ⇔ ai ∈ UP ⇔ t − ai ∈ Pη . We conclude that f ∈ P ⇔
f ∈ Pη .
Corollary 2.9.6 There exists a natural bijection between the orderings of R(t) and
the generalized Dedekind cuts of R.
Next on our list is the field R((t)) of formal Laurent series i≥n ai t i (with n ∈ Z
and ai ∈ R). Recall that R((t)) is the field of fractions of the ring A = R[[t]] of
formal power series in one variable over R. Let 0 = f (t) = i≥n ai t i with an = 0.
Then v(f ) := n defines a valuation v of R((t)) with valuation ring A, maximal ideal
mA = tA, value group Z and residue field A/mA = R. Since for every f ∈ mA the
element
2
1+f = 1/2
i
f i
i≥0
is a square in A (binomial sequence!), A is convex with respect to every ordering

of R((t)). Thus, by the Baer–Krull Theorem (Sect. 2.7), R((t)) has precisely two
orderings:
P+ = {0} ∪ {t n f (t) : n ∈ Z, f ∈ A, f (0) > 0}
and
P− = {0} ∪ {(−t)n f (t) : n ∈ Z, f ∈ A, f (0) > 0}.

Finally, let B := R{t} be the ring of convergent power series in one variable over
R, i.e., the ring of real function germs that are real-analytic at zero. Consider its field
of fractions K = Quot B ⊆ R((t)). Restricting the valuation v of R((t)) to K gives a
valuation w of K with valuation ring B and value group Z. The same argument as
above then shows that K has precisely two orderings, namely the restrictions of the
two orderings of R((t)).
2.10 Composition and Decomposition of Places
Let K, L and M be fields.

Proposition 2.10.1 (Factorization of Places) Let λ : K → L ∪ ∞ and μ : K →
∪∞ be places and let μ be a coarsening of λ (i.e., oμ ⊆ oλ ). If λ is surjective, then
there exists a unique place ν : L → M ∪ ∞ such that μ = ν ◦ λ. Furthermore,
oν = λ(oμ ).
Proof Let A := oλ , B := oμ and C := λ(B). Then C is a valuation ring of L by
∼ C and
Proposition 2.4.18. The restriction λ|B induces an isomorphism B/mA →
∼
thus an isomorphism α : κ(B) → κ(C) such that the diagram
commutes. Consider the embedding μ : κ(B) → M induced by μ, the map ψ :=

μ ◦ α −1 : κ(C) → M and the place ν := ψ ◦ λC : L → M ∪ ∞ induced by ψ.
Then oν = C and μ = ν ◦ λ. The uniqueness of ν is clear (since λ is surjective).
The following result is Proposition 2.4.18 for places:

Proposition 2.10.2 Let λ : K → L ∪ ∞ and ν : L → M ∪ ∞ be places. If λ is
surjective, there exists a canonical exact sequence
0 → ν → ν◦λ → λ → 0 (2.2)
of order preserving homomorphisms.

Recall that a short exact sequence
π
0→G →G−
→G →0
of abelian groups splits if there exists a section of π, i.e., a homomorphism s : G →

G such that π ◦ s = idG . In this case G → G × G , x → (πx, x − sπx) is an
isomorphism.
If the exact sequence (2.2) splits we can compute ν◦λ from ν and λ . Indeed:
Lemma 2.10.3 Let , and be ordered abelian groups and let
π
0→ →−
→ →0
2.10 Composition and Decomposition of Places 87
be a short exact sequence of order preserving homomorphisms. If π has a section s

(as a group homomorphism), then
ϕ : α → (πα, α − sπα)
is an order preserving isomorphism from to ( × )lex (where we identify

with its image in .)
Proof The map ϕ is an isomorphism of abelian groups. Let 0 < α ∈ . If α ∈ ,
then ϕ(α) = (0, α) > 0. If α ∈ , then ϕ(α) > 0 since π(α) > 0.
For example, if is a free abelian group, then π always has a section. The same
is true if is divisible, i.e., a vector space over Q (since is then also a vector
space over Q and π is Q-linear).
Assume again that λ : K → L ∪ ∞ and ν : L → M ∪ ∞ are places, and that λ
is surjective. Assume furthermore that vλ is discrete of rank one, i.e., λ ∼= Z. By
Proposition 2.10.2 and Lemma 2.10.3 there exists an order preserving isomorphism
(in general not unique)
∼ ( × ) = (Z × )
ν◦λ → λ ν lex ν lex
that we will describe explicitly in order to express vν◦λ in terms of vλ and vν .

To this end, let h ∈ mλ be a generator of mλ . Then α0 := vλ (h) is the positive
generator of λ . The choice of h yields a section s : λ → ν◦λ of π : ν◦λ → λ
via s(α0 ) = vν◦λ (h). Let ϕ : ν◦λ → ∼ ( × ) be the isomorphism induced
λ ν lex
by s as in Lemma 2.10.3. Identifying Ker(π) = o∗λ /o∗ν◦λ with ν (via λ, cf.
Proposition 2.4.18) gives
ϕ ◦ vν◦λ (f ) = vλ (f ), vν ◦ λ(f h−vλ (f ) )
for all f ∈ K ∗ . This solves the problem of expressing the valuation vν◦λ in terms
of vλ , vν and λ. (The argument can be repeated without any assumptions on λ , but
then the description of the section s will be more complicated in general.)
Notation 2.10.4 We denote by Zrantilex the group Zr equipped with the antilex-
icographic order, i.e., the positive elements are the tuples (a1 , . . . , as , 0, . . . , 0)
(1 ≤ s ≤ r) with as > 0. Of course, Zrantilex ∼ = Zrlex as ordered abelian groups;
we use the antilexicographic order just to simplify our notation below.
Let us consider the case of rational function fields (in several variables). Let
k be an arbitrary field, t1 , . . . , tr algebraically independent variables over k and
Ki := k(t1 , . . . , ti ) for 0 ≤ i ≤ r. Given a tuple c = (c1 , . . . , cr ) ∈ k r we construct
a place λ : Kr → k ∪ ∞ such that λ(f ) = f (c) for all f ∈ k[t1 , . . . , tr ] as follows:
For every i = 1, . . . , r there is a unique place λi : Ki → Ki−1 ∪ ∞ over Ki−1
such that λi (ti ) = ci (cf. Corollary 2.9.2). Let λ := λ1 ◦ · · · ◦ λr . Since λi = Z
for i = 1, . . . , r, it follows by induction from Proposition 2.10.2 and Lemma 2.10.3
that λ ∼
= Zrlex . We wish to determine vλ explicitly. For α = (α1 , . . . , αr ) ∈ Zr we
write (t − c)α := (t1 − c1 )α1 · · · (tr − cr )αr ∈ Kr , as usual. Let f ∈ k[t1 , . . . , tr ].

Then f can be expressed uniquely in the form

f = aα (t − c)α (with aα ∈ k and almost all aα = 0).
α∈Zr+
After an inductive application of the considerations above we may conclude that

vλ (f ) is the smallest index α in Zrantilex such that aα = 0. In particular we obtain:
Proposition 2.10.5 The valuation v : k(t1 , . . . , tr )∗ → Zrantilex, defined by
v (t1 − c1 )α1 · · · (tr − cr )αr = (α1 , . . . , αr )
is associated to the place λ : k(t1 , . . . , tr ) → k ∪ ∞ with λ(f ) = f (c) for all

f ∈ k[t1 , . . . , tr ], described above. Furthermore, v(tr − cr ), . . . , v(t1 − c1 ) is the
lexicographic basis of λ .
Note that there are of course many other places λ : k(t) → k ∪ ∞ over k
with λ (t) = c. For instance, by choosing a different transcendence basis u =
(u1 , . . . , ur ) such that k[t] = k[u] and replacing c by u(c).
Assume now that k = R is real closed. Let K := R(t1 , . . . , tr ) and λ : K →
R ∪ ∞ the place described above. Let P be an ordering of K, compatible with λ
(by the Baer–Krull Theorem, there are precisely 2r such orderings) and S the real
closure of (K, P ). By Theorem 2.8.15, λ has a unique extension μ : S → R ∪ ∞.
Proposition 2.10.6 μ is isomorphic to Qrlex .
Indeed, more generally we have:
Proposition 2.10.7 Let L ⊇ K be an algebraic field extension with L real closed
or algebraically closed. If μ : L → F ∪ ∞ is a place and λ := μ|K , then there
∼ .
exists an order preserving isomorphism λ ⊗ Q → μ
(Let be an ordered abelian group. Then ⊗Q has a unique ordering that makes
the embedding → ⊗Q order preserving. In the statement of the theorem, λ ⊗Q
is equipped with this ordering.)
Proof The inclusion K ∗ ⊆ L∗ induces an order preserving embedding λ ⊆ μ .
We claim that the group μ / λ is torsion. Indeed, let b ∈ L∗ and let a0 , . . . , an ∈ K
be such that an bn + · · · + a1 b + a0 = 0 and an = 0. Then there exist indices i and
j such that 0 ≤ i < j ≤ n and vμ (ai bi ) = vμ (aj bj ) < ∞ (for otherwise,
vμ ( ak bk ) = min v(ak bk ) < ∞)), and it follows that (j − i)vμ (b) = vμ aaji ∈
k
λ .
Since L is real closed or algebraically closed, μ is divisible (every positive
element, resp. every element of L has arbitrary roots) and so μ = Q·λ = λ ⊗Q.
Let k be a field, K0 := k and Ki := Ki−1 ((ti )) = Quot Ki−1 [[ti ]] for i =

1, 2, . . . , r. (Note that for r ≥ 2, the field k((t1 , . . . , tr )) is a proper subfield of
2.11 Existence of Real Places of Function Fields 89
the field Kr = k((t1 )) · · · ((tr ))!) Let λi : Ki → Ki−1 ∪ ∞ be the place over Ki−1 ,
associated to Ki−1 [[ti ]], and let λ = λ1 ◦· · ·◦λr : Kr → k ∪∞. Then λ ∼ = Zrlex with
lexicographic basis vλ (tr ), . . . , vλ (t1 ), as in Proposition 2.10.5. If k ⊆ K ⊆ Kr is
an intermediate field with t1 , . . . , tr ∈ K, then we also have λ|K = Zrlex .
Using some more commutative algebra we obtain an interesting application
(which will play no further role however):
Let A be a regular local k-algebra with κ(A) = k, A = lim A/mn its completion
←− A
and (f1 , . . . , fr ) a minimal set of generators of mA . It is well-known that there
exists a k-isomorphism A → ∼ k[[t , . . . , t ]] such that f → t for all i = 1, . . . , r,
1 r i i
cf. [81, p. 206] or [13, Ch. IX, § 3, no. 3]. Let K := Quot A. It follows from
K ⊆ Quot A ∼ = k((t1 , . . . , tr )) ⊆ Kr that the place λ|K : K − → k ∪ ∞ also has
k
value group Zrlex , with lexicographic basis v(fr ), . . . , v(f1 ) (where v := vλ|K ).
Furthermore, its valuation ring dominates A.
2.11 Existence of Real Places of Function Fields
We briefly recall some facts from field theory. Let L ⊇ K be a field extension.
Then all maximal families of elements of L that are algebraically independent over
K have the same cardinality. Any such family is called a transcendence basis of L
over K and its cardinality is called the transcendence degree of L over K, denoted
tr.deg.(L/K). The field L is called a d-dimensional function field over K if L is
finitely generated over K and tr.deg.(L/K) = d. (The terminology comes from
the fact that up to K-isomorphism such fields are precisely the fields of rational
functions on irreducible algebraic K-varieties of dimension d, cf. [97, I, §3] or [45].)
The purpose of this section is to provide a proof of:
Theorem 2.11.1 Let R be a real closed field, K an r-dimensional real function
field over R and (t1 , . . . , tr ) a fixed transcendence basis of K over R. Then there
exist elements ai , bi ∈ R with ai < bi for i = 1, . . . , r such that for every real
closed overfield S ⊇ R the following property is satisfied:
(Tr ) For every tuple c = (c1 , . . . , cr ) ∈ S r such that ai < ci < bi for i = 1, . . . , r,
there exists a place λ : K → S ∪ ∞ over R that satisfies λ(ti ) = ci for
i = 1, . . . , r.
Note that such a place λ must be trivial (i.e., a homomorphism) if c1 , . . . , cr
are algebraically independent over R. (Indeed, in this case we have R[t1 , . . . , tr ] ∩
λ−1 (0) = {0}, which implies R(t1 , . . . , tr ) ⊆ oλ . It follows that oλ = K since oλ is
integrally closed.)
Corollary 2.11.2 (Lang’s Embedding Theorem) Let R be a real closed field and
K a real function field over R. For every real closed overfield S of R such that
tr.deg.(S/R) ≥ tr.deg.(K/R) there exists an R-homomorphism K → S.
Proof The statement follows from Theorem 2.11.1 and the remark following it.
Indeed, there are elements ci ∈ S with ai < ci < bi for i = 1, . . . , r such that
c1 , . . . , cr are algebraically independent over R; this is clear for r = 1 and follows
by induction for all r.
We first establish two auxiliary results.
Lemma 2.11.3 Let L ⊇ K be a finite field extension, B a valuation ring of L and
A = K ∩ B. Then the ramification index e = [B : A ] and the residue degree
f = [κ(B) : κ(A)] are finite and ef ≤ [L : K].
Proof Write v = vB . Consider elements b1 , . . . , br ∈ B\{0} such that the images of
the v(bi ) in B / A are distinct, and elements c1 , . . . , cs ∈ B such that the images
cj of the cj in κ(B) are linearly independent over κ(A). We will show that for all
aij ∈ K (1 ≤ i ≤ r, 1 ≤ j ≤ s),

v aij bi cj = min v(aij bi ).
i,j
i,j
This implies in particular that the bi cj are linearly independent over K.

We may assume that for every i ∈ {1, . . . , r} there exists j ∈ {1, . . . , s} such
that aij = 0 (for otherwise bi can be omitted). For every i ∈ {1, . . . , r} we choose
k(i) ∈ {1, . . . , s} such that v(aik(i) ) = min v(aij ). Then aik(i) = 0 and it follows
j
that for all i,

aij bi cj = aik(i) bi aij cj
j j
where aij := aij /aik(i) ∈ A. Since the cj are linearly independent over κ(A) (and
aik(i) = 1) we have aij cj ∈ B ∗ and thus
j

v aij bi cj = v(aik(i) ) + v(bi ) (i = 1, . . . , r).
j
These values are all distinct by assumption and we conclude that

v aij bi cj = min v(aik(i) ) + v(bi ) = min v(aij bi ).
i i,j
i j
Lemma 2.11.4 Let K be a field, L a function field over K and n := tr.deg.(L/K).

Let B = L be a proper valuation ring of L such that K ⊆ B. Then
tr.deg. κ(B)/K ≤ n − 1, and if tr.deg. κ(B)/K = n − 1, then B is discrete
of rank one and κ(B) is a function field (i.e., finitely generated) over K.
Proof Let u1 , . . . , ur ∈ B be such that their images u1 , . . . , ur in κ(B) are

algebraically independent over K. Then u1 , . . . , ur (in L) are also algebraically
independent over K and so r ≤ n. Furthermore, K[u1 , . . . , ur ] ∩ mB = {0}.
It follows that K(u1 , . . . , ur ) ⊆ B. If r = n, then L is algebraic and hence
integral over K(u1 , . . . , ur ) and thus B = L, a contradiction. Assume that r =
tr.deg. κ(B)/K = n − 1. Let F := K(u1 , . . . , un−1 ) and choose t ∈ L such that
(u1 , . . . , un−1 , t) is a transcendence basis of L over K. Then A := B ∩ F (t) is
a proper valuation ring of F (t) and F ⊆ A. It follows from Corollary 2.9.2 that
[κ(A) : F ] < ∞ and A ∼ = Z, and from Lemma 2.11.3 that [κ(B) : κ(A)] and
B / A are finite. We conclude that κ(B) is finitely generated over K and B ∼= Z.
Exercise 2.11.5 Let K be a field, let v be a valuation of K with value group

and residue field κ, and assume char(κ) =
2. Let a ∈ K ∗ with √a ∈/ K, and let
√
L = K( a). Show the following (extension always means “up to equivalence”):
(a) If v(a) ∈ √ v has a unique extension to L, and e = 2, f = 1.
/ 2, then
(b) If v(a) = 0 and √ a∈
/ κ, then v has a unique extension to L, and e = 1, f = 2.
(c) If v(a) = 0 and a ∈ κ, then v has exactly two extensions to L, and both
satisfy e = f = 1.
We will now show that in order to prove Theorem 2.11.1 it suffices to prove the
following weaker property for all r ≥ 1, under the same assumptions:
(
Tr ) There exist ai , bi ∈ R with ai < bi for i = 1, . . . , r such that for every
real closed overfield S ⊇ R and every tuple c = (c1 , . . . , cr ) ∈ S r with
ai < ci < bi for i = 1, . . . , r and c1 , . . . , cr algebraically independent
over R, there exists a homomorphism ϕ : K → S such that ϕ(ti ) = ci for
i = 1, . . . , r.
Proposition 2.11.6 (
Tr ) implies (Tr ) for all r ≥ 1.
Proof Let c ∈ S r with ai < ci < bi and c1 , . . . , cr not necessarily algebraically
independent over R. Let u1 , . . . , ur be independent variables over S and T :=
S(u1 , . . . , ur ) the r-dimensional rational function field over S. Consider the place
μ : T → S ∪ ∞ over S with μ (ui ) = 0 for i = 1, . . . , r, described in Sect. 2.10.
By Corollary 2.8.16 there exists a real closure T of T and an extension μ : T → S ∪
∞ of μ . Since the ui are infinitely small with respect to S we have ai < ci +ui < bi
for i = 1, . . . , r. Since the ci + ui are algebraically independent over R, there exists
a homomorphism ϕ : K → T such that ϕ(ti ) = ci + ui by property ( Tr ). The place
λ := μ ◦ ϕ : K → S ∪ ∞ then satisfies λ(ti ) = ci for i = 1, . . . , r.
Let us now prove ( T1 ). Assume thus that r = 1, that R and K are as in
Theorem 2.11.1 and that F := R(t), where t := t1 . Let tr : K → F be the trace
of the finite extension K ⊇ F and let g1 , . . . , gn ∈ F ∗ be such that tr∗ 1K ∼ =
g1 , . . . , gn . After multiplying the gi with the square of their denominators we may
assume without loss of generality that all gi are in R[t]. Let d1 < · · · < dN be the
distinct zeroes of g1 · · · gn in R, d0 := −∞ and dN+1 := +∞. Every gi has a
fixed sign εij ∈ {±1} on ]dj −1 , dj [ (interpreted as an interval in R or in any ordered

overfield of R) for j = 1, . . . , N + 1, i = 1, . . . , n.
Since K is real there exists an ordering Q of K. Let P := F ∩ Q and let j ∈
{1, . . . , N} be the index for which dj −1 <P t <P dj . Then signP (gi ) = εij for
i = 1, . . . , n. Since

n
εij = signP tr∗ 1K
i=1
is the number of extensions of P to K (cf. Corollary 1.12.10), this number is ≥ 1.

We will show that (T1 ) is satisfied for every choice of a1 , b1 such that ]a1 , b1 [ ⊆
]dj −1 , dj [.
Let S be a real closed overfield of R, c ∈ S \ R with dj −1 < c < dj and
ψ : F → S the R-homomorphism that satisfies ψ(t) = c. By Theorem 1.11.6 the
extensions ϕ : K → S of ψ are in bijective correspondence with the by ψ induced
extensions to F of the orderings Pc of K. Their number is equal to

n
n
signPc tr∗ 1K = signPc (gi ) = εij ≥ 1,
i=1 i=1
which shows in particular that such an extension ϕ exists.

Before proving Theorem 2.11.1 in general we consider some consequences of
(T1 ).
Proposition 2.11.7 Let λ : K → L ∪ ∞ be a real place and z1 , . . . , zn ∈ K. If
λ zi2 = ∞, then λ(zi ) = ∞ for i = 1, . . . , n.
i
Proof By Corollary 2.8.11 there exists a λ-compatible ordering P of K. Since oλ
is convex with respect to P and 0 ≤ zi2 ≤ z12 + · · · + zn2 ∈ oλ , we have zi2 ∈ oλ and
conclude that zi ∈ oλ for i = 1, . . . , n.
Theorem 2.11.8 (Artin–Lang Stellensatz) Let R be a real closed field and K a

real function field over R. For all finitely many z1 , . . . , zn ∈ K there exists a place
λ : K → R ∪ ∞ over R such that λ(zi ) = ∞ for i = 1, . . . , n. Furthermore, if
r = tr.deg.(K/R), then there exists such a place with λ ∼ = Zrlex .
Proof By Proposition 2.11.7 we may assume that n = 1. We proceed by induction
on r. If r = 0, the statement is trivially true since K = R. If r = 1, the existence
of λ follows from (T1 ). Furthermore, λ ∼ = Z for every such λ by Lemma 2.11.4.
Thus, let r > 1 and consider a transcendence basis t1 , . . . , tr of K over R with
z ∈ R(t1 , . . . , tr−1 ) =: F . Let T be a real closure of K, S the algebraic closure
of F in T and L := KS (the subfield of T generated by K and S). Since L is
a one-dimensional real function field over S, there exists a place L − → S∪∞
S
(by the r = 1 case) and via restriction we obtain a place μ : K −
→ S ∪ ∞. Since
F
tr.deg.(K/F ) = 1 (and S/F is algebraic) this place is nontrivial and satisfies μ(z) =
∞. Let A := oμ and λA : K → κ(A) ∪ ∞ the canonical place associated to A. By
Lemma 2.11.4 we have [κ(A) : F ] < ∞ and A ∼ = Z. Furthermore, since κ(A) is
real and tr.deg. κ(A)/R = tr.deg.(F /R) = r − 1 it follows by induction that there
exists a place ν : κ(A) → R ∪ ∞ over R such that ν λA (z) = ∞ and ν ∼ = Zr−1
lex .
Let λ := ν ◦ λA : K − → R ∪ ∞. Then λ(z) = ∞ and λ ∼ = Zlex by Sect. 2.10.
r
R
Theorem 2.11.9 (Artin–Lang Stellensatz, Refined Version) In the situation of

Theorem 2.11.8, consider also an ordering Q of K. Then there exists a place
λ: K −→ R ∪ ∞ as in Theorem 2.11.8 that satisfies the additional property
R
signQ (zi ) = sign λ(zi ) for i = 1, . . . , n.
Proof After multiplying zi with signQ (zi ) we may assume without loss of gen-
erality that zi >Q 0 for i = 1, . . . , n. Applying Theorem 2.11.8 to K :=
√ √ 1
K( z1 , . . . , zn ) and the elements z1 , . . . , zn , z1 ···z ∈ K we obtain a place
n
√
λ:K − → R ∪∞ such that λ (zi ) ∈ {0, ∞}, and thus such that λ (zi ) = λ ( zi )2 >
R
0 for i = 1, . . . , n. Let λ := λ |K . Since λ / λ is finite (by Lemma 2.11.3), we
have rank λ = rank λ = r and it follows easily that λ ∼ = Zrlex .
We will now finish the proof of Theorem 2.11.1 by showing that property ( Tr )
holds for all r ≥ 2. Thus, let t1 , . . . , tr be a transcendence basis of K over R and
let F := R(t1 , . . . , tr ). Assume again that g1 , . . . , gn ∈ R[t1 , . . . , tr ] are such that
tr∗ 1K ∼ = g1 , . . . , gn (where n = [K : F ]). There exists an ordering Q of K.
By Theorem 2.11.9 there exists a place μ : K − → R ∪ ∞ such that μ(ti ) = ∞ for
R
i = 1, . . . , r, and μ(gj ) = ∞ and sign μ(gj ) = signQ (gj ) for j = 1, . . . , n. Let
p := μ(t1 ), . . . , μ(tr ) ∈ R r and P := F ∩ Q. As before, it follows that

n
n
n
1 ≤ signP tr∗ 1K = signP (gj ) = sign μ(gj ) = sign gj (p),
j =1 j =1 j =1
where the final equality follows from the fact that μ(f ) = f (p) for all f ∈
R[t1 , . . . , tr ].
In order to finish the proof of Theorem 2.11.1 we need the following proposition
that we will prove at the end of this section:
Proposition 2.11.10 Let g ∈ R[x1 , . . . , xr ] be a polynomial and p ∈ R r such that
g(p) = 0. Then there exists an element ε ∈ R, ε > 0, such that for every real closed
overfield S of R and every q ∈ S r with pi − ε < qi < pi + ε for i = 1, . . . , r we
have sign g(p) = sign g(q).
For g1 , . . . , gn and p as above, the lemma provides us with an 0 < ε ∈ R. Let
S ⊇ R be a real closed overfield and consider a tuple c = (c1 , . . . , cr ) ∈ S r such
that c1 , . . . , cr are algebraically independent over R and pi − ε < ci < pi + ε for
i = 1, . . . , r. Let ψ : F → S be the embedding determined by ψ(ti ) = ci and Pc
the ordering of F pulled back from S via ψ. As in the proof of property (

T1 ) we
obtain

n
n
n
signPc tr∗ 1K = signPc (gj ) = sign gj (c) = sign gj (p) ≥ 1,
j =1 j =1 j =1
where the third equality follows from Proposition 2.11.10. It follows from
Sects. 1.11 and 1.12 that ψ has an extension ϕ : K → S as required. This concludes
the proof of Theorem 2.11.1
Proof of Proposition 2.11.10 Let x = (x1 , . . . , xr ) and let

g(p + x) − g(p) = cα x α
α
be the Taylor expansion of g at p (where the sum is over all 0 = α ∈ Nr0 , and
almost all cα are 0). If a ∈ S r is such that |ai | < ε ≤ 1 for i = 1, . . . , r, then
|g(p +a)−g(p)| ≤ ε |cα |. Hence, it suffices to choose ε > 0 such that ε |cα | <
|g(p)| and ε ≤ 1. α α
2.12 Artin’s Solution of Hilbert’s 17th Problem and the Sign

Change Criterion
Let k be a field. Throughout this section t1 , . . . , tn denote variables that are

algebraically independent over k. We often write t instead of (t1 , . . . , tn ). If (k, P )
is an ordered field, we endow k n with the strong topology TP (cf. Sect. 2.6).
Proposition 2.12.1 Let (k, P ) be an ordered field, U ⊆ k n a nonempty open subset
and f ∈ k[t1 , . . . , tn ] a polynomial that vanishes identically on U . Then f = 0.
Proof By induction on n. If n = 1 this is clear. Assume the statement holds for
n − 1. Let t := (t1 , . . . , tn−1 ) and
f = f (t , tn ) = f0 (t ) · tnd + · · · + fd (t ),
where d ≥ 0 and f0 , . . . , fd ∈ k[t ]. We may assume without loss of generality that

n
n−1
U is of the form U = ]ai , bi [ with ai , bi ∈ k. Let c ∈ U := ]ai , bi [ be
i=1 i=1
arbitrary. Then f (c , tn ) = f0 (c )tnd + · · · + fd (c ) vanishes identically on ]an , bn [.
Thus f0 (c ) = · · · = fd (c ) = 0. It follows that the fj vanish identically on U and
so the induction hypothesis implies f0 = · · · = fd = 0.
The zero set of a nonzero polynomial in k n is thus thin in the sense that its interior
is empty.
2.12 Artin’s Solution of Hilbert’s 17th Problem and the Sign Change Criterion 95
Definition 2.12.2 Let k be a field and t = (t1 , . . . , tn ).

(a) Consider a polynomial f ∈ k[t]. We denote the zero set of f in k n by Zk (f )
(or simply Z(f )), i.e.,
Zk (f ) = {a ∈ k n : f (a) = 0}.
(b) Consider a rational function f ∈ k(t) and write f = g/ h with g, h ∈ k[t]

relatively prime and h = 0. We denote the domain of f in k n by Dk (f ) (or
simply D(f )), i.e.,
Dk (f ) = {a ∈ k n : h(a) = 0}.
We also let Zk (f ) := Zk (g) ∩ Dk (f ).

If k is ordered, the domain of every rational function is open and dense in k n with
respect to the strong topology (Proposition 2.12.1).
Definition 2.12.3 Let (k, P ) be an ordered field, f ∈ k(t) and M ⊆ Dk (f ).
(a) f is positive definite (resp. positive semidefinite) on M if f (a) > 0 (resp.
f (a) ≥ 0) for all a ∈ M.
(b) f is negative definite (resp. negative semidefinite) on M if −f is positive
definite (resp. positive semidefinite) on M.
(c) f is indefinite on M if f is neither positive nor negative definite on M.
If M = Dk (f ), we say that f is positive (semi)definite, resp. negative (semi)definite,
resp. indefinite over k.
Proposition 2.12.4 Let (k, P ) be an ordered field.

(a) If f ∈ k(t) is positive semidefinite on an open dense subset of Dk (f ), then f is
positive semidefinite over k.
(b) The set of rational functions that are positive semidefinite over k is a preorder-
ing of k(t).
Proof (a) This follows from the continuity of the evaluation map Dk (f ) → k, a →
f (a).
(b) Let f, g ∈ k(t)∗ be positive semidefinite over k. Since f + g and fg are
positive semidefinite on the dense open subset Dk (f )∩Dk (g) of k n , they are positive
semidefinite over k by (a). A similar argument shows that f 2 is positive semidefinite
for every f ∈ k(t).
Remark 2.12.5 Let (k, P ) be an ordered field with real closure R. It follows from
Proposition 2.12.4(b) that every function of the form a1 f12 +· · ·+ar fr2 with ai ∈ P
and fi ∈ k(t) is positive semidefinite over R. A rational function which is positive
semidefinite over k is in general no longer positive semidefinite over R. This is
because k is usually not dense in R, and f (considered over R) may become negative
in the “gaps”. (The reader may want to construct an example, cf. Sect. 2.1.)
In 1900 David Hilbert gave his famous address to the International Congress of
Mathematicians in Paris in which he presented his list of 23 unsolved mathematics
problems. Since then many of these problems have been solved, often involving the
development of completely new methods. This is especially true of the 17th problem
about the representation of forms by sums of squares and its solution by Emil Artin
in 1927, cf. [3, 4]. Artin discovered the notion of ordering of a field, putting it to
elegant and successful use. His ideas sparked new directions in the development of
real algebra in the twentieth century (see the Preface).
Hilbert’s 17th Problem can be formulated as follows:
Hilbert’s 17th Problem Given a positive semidefinite polynomial f =
f (t1 , . . . , tn ) ∈ R[t1 , . . . , tn ], do there exist an integer r ≥ 1 and rational functions
g1 , . . . , gr ∈ R(t1 , . . . , tn ) such that f = g12 + · · · + gr2 ?
Supplementary question: If k ⊆ R is a subfield that contains the coefficients of f ,
can the gi already be found in k(t1 , . . . , tn )?
Remarks 2.12.6
(1) In the formulation of the supplementary question (which is also due to Hilbert)
it would be better to consider representations in the sense of Remark 2.12.5
since otherwise it is immediately clear that there is no solution in general.
(2) An obvious follow-on question is whether there exists a representation f =
g12 +· · ·+gr2 where g1 , . . . , gr are actually polynomials. Hilbert already showed
in 1888 that this is indeed the case if n = 1, but that for n ≥ 2 this is false
in general. (Hilbert’s proof actually provided more precise details.) The first
example of a positive semidefinite polynomial that cannot be written as a sum
of squares of polynomials was not published until 1967. Indeed, in [84] Motzkin
considered the polynomial (here presented in homogeneous form)
f (x, y, z) = x 4 y 2 + x 2 y 4 + z6 − 3x 2 y 2 z2 .
That f is positive semidefinite follows from the arithmetic-geometric inequal-

ity:
1 4 2
(x y + x 2 y 4 + z6 ) ≥
3
x 6 y 6 z6 .
3
A further example is given by
g(x, y, z) = x 4 y 2 + y 4 z2 + z4 x 2 − 3x 2y 2 z2 .
In [26] one can find a simple method for proving that f and g are not sums of
squares of polynomials, as well as further examples.
Let us now consider the solution of Hilbert’s 17th Problem:

Theorem 2.12.7 (E. Artin) Let (k, P ) be an ordered field with real closure R, and
let f ∈ k(t1 , . . . , tn ) be a rational function that is positive semidefinite over R.
Then there exist an integer r ≥ 1, functions g1 , . . . , gr ∈ k(t1 , . . . , tn ) and positive
elements (with respect to P ) a1 , . . . , ar ∈ k such that
f = a1 g12 + · · · + ar gr2 .
Proof The proof is not constructive. Consider the set
T := {a1 g12 + · · · + ar gr2 : r ≥ 1, ai ∈ P , gi ∈ k(t)},
which is a preordering of k(t) and therefore equal to the intersection of all

orderings Q of k(t) such that P ⊆ Q, cf. Theorem 1.1.10. Assume for the sake
of contradiction that the conclusion of the theorem is false. Then there exists
an ordering Q of k(t) such that P ⊆ Q and f ∈ Q. Let S denote the real
closure of k(t) with respect to Q. Then we can think of R as a subfield of S (cf.
Theorem 1.11.2) and in particular consider the subfield R(t) of S with the induced
ordering Q . Since f ∈ Q , there exists a place λ : R(t) − → R ∪ ∞ such that
R
λ(t1 ) = ∞, . . . , λ(tn ) = ∞, λ(f ) = ∞ and λ(f ) < 0 (cf. Theorem 2.11.9).
Letting a := λ(t1 ), . . . , λ(tn ) ∈ R n , we obtain f (a) = λ(f ) < 0, which is
impossible since f is positive semidefinite.
Hilbert’s question is thus fully answered, but there are still many related
questions for which answers remain elusive, at least in part. For example, having
fixed k, P and n, does there exist a positive integer r ≥ 1 with the property that
every rational function f ∈ k(t1 , . . . , tn ) which is positive semidefinite over R has
a representation f = a1 g12 + · · · + ar gr2 of length r, as in the theorem? And if
so, what is the smallest such r? Let us write r(k, n) for this integer. (To be precise
we should write rP (k, n), but k has a unique ordering in the examples that follow.)
When k = R is real closed, it is known that r(R, 1) = 2 (elementary), r(R, 2) = 4
(Cassels, Ellison, Pfister [24], difficult) and n + 2 ≤ r(R, n) ≤ 2n for all n ≥ 2 (this
follows from Pfister [88] and [24]). For the field of rational numbers k = Q one has
r(Q, 1) = 5 (Pourchet [89]) and r(Q, 2) ≤ 8 (Kato, Colliot-Thélène [57]).
As an application of Theorem 2.12.7, we prove the following “sign change
criterion”:
Theorem 2.12.8 (D. Dubois, G. Efroymson) Let (k, P ) be an ordered field with
real closure R, and let f ∈ k[t1 , . . . , tn ] be a non-constant irreducible polynomial.
The following statements are equivalent:
(i) The ordering P extends to the function field K = Quot k[t1 , . . . , tn ]/(f );
(ii) f is indefinite over R.
The use of the word “criterion” is motivated by (ii) which signifies that f changes
sign “along” the hypersurface H := {f = 0}. Algebraic geometry tells us that K is
the field of rational functions on H .
The proof makes use of the following proposition, which we establish first:
Proposition 2.12.9 Let A be a unique factorization domain, E = Quot A and
f ∈ A[u]\A a polynomial in one variable u, irreducible in A[u]. Then A[u]/(f ) →
E[u]/(f ) is injective and E[u]/(f ) is a field (and thus the quotient field of
A[u]/(f )).
Proof The statement follows immediately from the fact that since A is a UFD, A[u]
is also a UFD (a consequence of Gauss’s Lemma, cf. [73, IV, Theorem 2.1] or [55,
Vol. I, §2.16]): if g ∈ A[u] and h ∈ E[u] such that g = f h in E[u], then there exist
0 = a ∈ A and h ∈ A[u] such that ag = f h (in A[u]). Since f is not a divisor
of a, it follows that f | g in A[u]. A similar straightforward argument shows that f
remains irreducible in E[u].
Proof of Theorem 2.12.8 (i) ⇒ (ii): Assume that f is semidefinite. Without loss of
generality we may assume that f ≥ 0 on R n . By Artin’s theorem there exists an
equation of the form

r
f h2 = ai gi2 ,
i=1
where 0 = ai ∈ P , gi , h ∈ k[t] and (without loss of generality)

gcd(g1 , . . . , gr , h) = 1. It follows that f does not divide all gi . In other words,
not all summands on the right hand side disappear modulo (f ). Hence P does not
extend to K by Proposition 1.3.2.
(ii) ⇒ (i): Assume that f is indefinite over R. Then f may no longer be
irreducible over R, but still must have at least one indefinite irreducible divisor f1
in R[t]. Without loss of generality we may assume that the variable tn occurs in f1
(and thus also in f ). Let t := (t1 , . . . , tn−1 ), A := k[t ], E := k(t ), B := R[t ],
F := R(t ), and consider the commutative diagram
(2.3)
where the arrows depict the canonical homomorphisms.

It follows from the lemma that the map ψ in (2.3) is a field embedding from K =
Quot k[t]/(f ) = E[tn ]/(f ) into L := Quot R[t]/(f1 ) = F [tn ]/(f1 ). It suffices to
show that L is a real field since every ordering of L then induces an extension of P
to K. Thus we may assume without loss of generality that k = R and f = f1 and
must show that K is a real field.
If n = 1, then f is linear and hence K = R. Thus we let n ≥ 2 and choose

p, q ∈ R n such that f (p) < 0 < f (q). After a linear change of coordinates we may
assume that p = (a1 , . . . , an−1 , b) and q = (a1 , . . . , an−1 , c). Let us also write t =
(t1 , . . . , tn−1 ) and F := R(t ). Let Q be an ordering of F such that ti −ai is infinitely
small compared to R for i = 1, . . . , n − 1, cf. Sect. 2.10. Then f = f (t , tn ) ∈
F [tn ] is indefinite over F with respect to Q. Indeed, writing o := oQ (F /R), we
get ti ≡ ai mod mo for i = 1, . . . , n − 1. Hence, f (t , b) ≡ f (a1 , . . . , an , b) =
f (p) mod mo . In other words, f (t , b) < 0. Similarly, f (t , c) > 0. Then f is
irreducible in F [tn ] by the lemma. From the n = 1 case it follows that F [tn ]/(f )
possesses an ordering that extends Q. Finally, the embedding K → F [tn ]/(f )
shows that K is a real field.
Chapter 3
The Real Spectrum
The focus of this chapter is on the real spectrum of a commutative ring—discovered

in 1979 by M.F. Roy and M. Coste—and its elementary properties. We will usually
be as general as possible and consider arbitrary commutative rings. However, special
features arise in the “geometric setting” where coordinate rings of affine varieties
over real closed base fields are considered.
The real spectrum and the Zariski spectrum exhibit many similarities (in essence
one considers homomorphisms into real closed fields instead of algebraically closed
fields). Since we will be using the most elementary properties of the Zariski
spectrum, we recall some fundamental facts in the first section of this chapter.
We treat the Zariski spectrum and the real spectrum exclusively as topological
spaces and ignore their structure sheaves. The concept of affine variety will be
treated in the sense of A. Weil since the issue of reducibility will not play any
role and, as such fewer technicalities will need to be considered. Nevertheless
we strongly recommend gaining some familiarity with the modern language of
algebraic geometry (as developed in Hartshorne’s monograph [44], for example).
Throughout this chapter, all rings are assumed commutative and unitary.
3.1 The Zariski Spectrum. Affine Varieties
We start by recalling some definitions. Let A be a ring and a an ideal of A. The set
√
a := {a ∈ A : a n ∈ a for some n ∈ N}
√
√ the radical of a. If a = a, then a is called a radical ideal.
is an ideal of A, called
The ideal Nil A := (0) is the nilradical of A, and A is reduced if Nil A = (0). As
usual, we also let Ared := A/ Nil A.

https://doi.org/10.1007/978-3-031-09800-0_3
102 3 The Real Spectrum
Definition 3.1.1 The (Zariski) spectrum Spec A of A is the set of all prime ideals
of A. Every p ∈ Spec A has an associated residue field
κA (p) = κ(p) = Quot A/p
and evaluation homomorphism
ρA,p = ρp : A → κ(p), f → f mod p.
We also write f (p) instead of ρp (f ) = f mod p.
Remark 3.1.2 If A is the trivial ring, then Spec A = ∅. Otherwise, Spec A = ∅.
Remark 3.1.3 It is helpful to keep the following interpretation of Spec A in mind:

We think of the elements f ∈ A as “functions” on the set Spec A (which is a
topological space, cf. Definition 3.1.4). To each point p ∈ Spec A we associate
its function value under f , f (p) = f mod p ∈ κ(p). (Note that the function value
of each point lives in a different field!) From a formal point of view we are thus
considering the ring homomorphism

ρ: A → κ(p), ρ(f ) = f (p) p∈Spec A
.
p∈Spec A
The kernel of ρ is equal to

!
p = Nil A
p∈Spec A
(cf. Proposition 3.1.11). Hence,

ρ is injective if and only if A is reduced. In this case
we can view A as a subring of κ(p) via ρ.
p
In general the function values of f ∈ A at distinct points p cannot be compared

since they live in different fields. However, there is a meaningful concept of zero set
of f , namely {p ∈ Spec A : f ∈ p}. Taking these zero sets as generators of closed
sets gives rise to the Zariski topology on Spec A:
Definition 3.1.4
(a) For any f ∈ A we let
DA (f ) := {p ∈ Spec A : f (p) = 0} = {p ∈ Spec A : f ∈ p},

VA (f ) := {p ∈ Spec A : f (p) = 0} = {p ∈ Spec A : f ∈ p}.
3.1 The Zariski Spectrum. Affine Varieties 103
More generally, for any subset T of A we let

! !
DA (T ) := DA (f ), VA (T ) := VA (f ).
f ∈T f ∈T
(We usually drop the indices A.)

(b) The Zariski topology on Spec A is the topology that has {DA (f ) : f ∈ A} as a
subbasis of open sets.
We will always think of Spec A as a topological space in this way. Note that the
subbasis {D(f ) : f ∈ A} is actually a basis of the topology since for any f, g ∈ A
we have D(f ) ∩ D(g) = D(fg). By definition, all sets V(T ) (with T ⊆ A) are
closed.
Proposition 3.1.5
√
(a) Let T ⊆ A. Then V(T ) = V( a), where a := t ∈T At is the ideal generated
by T .
(b) Every closed subset of Spec A is of the form V(a) for some (radical) ideal a of
A.
√
Proof (a) The inclusion
√ ⊇ is trivial since T ⊆ a. For the reverse inclusion, let p ∈
V(T ) and f ∈ a. Then there exist r, n ∈ N and ti ∈ T , ai ∈ A for i = 1, . . . , r,
such that f n = a1 t1 + · · · + ar tr . Since ti ∈ p for i = 1, . . . , r, it follows that f ∈ p,
i.e., p ∈ V(f ).
(b) This follows from (a) and the fact that every open subset of Spec A is the
union of subsets D(f ) with f ∈ A.
Definition 3.1.6 Let Y be a subset of Spec A. Then

!
I(Y ) := {f ∈ A : f |Y ≡ 0} = p
p∈Y
denotes the (radical) ideal of functions that vanish on Y .
Proposition 3.1.7 Let Y be a subset of Spec A. Then Y = V I(Y ) .

Proof The inclusion ⊆ follows from Y ⊆ V I(Y ) and the fact that V I(Y ) is
closed. For the reverse inclusion, let a be an ideal of A such that Y = V(a) (cf.
Proposition 3.1.5). Then a ⊆ I(Y ) ⊆ I(Y ), and so Y = V(a) ⊇ V I(Y ) .
Corollary 3.1.8 The closed points of Spec A are precisely the maximal ideals of
A.
Recall that a topological space X is called a T0 space if for every two distinct
points x, y ∈ X, we have x ∈ {y} or y ∈ {x}. If we replace “or” by “and”, then X is
a T1 space.
The Zariski topology is thus very weak as a rule and not even a T1 topology.
However:
Exercise 3.1.9 Show that the Zariski topology is a T0 topology.
Exercise 3.1.10 Let S be a multiplicative subset of A (i.e., SS ⊆ S and 1 ∈ S).

Show that every ideal a of A that is maximal for the property a ∩ S = ∅ is a prime
ideal.
Proposition 3.1.11 (Abstract or Weak Nullstellensatz) If a is an ideal of A, then

√
I V(a) = a.
√
Corollary
√ 3.1.12 For every two ideals a, b of A we have V(a) ⊆ V(b) ⇔ a ⊇
b.
√
Proof of Proposition 3.1.11 We must show that a = {p ∈ Spec A : a√⊆ p}.
The inclusion ⊆ is clear. For the reverse inclusion, let f ∈ A with f ∈ a. By
Zorn’s Lemma, there exists and ideal p that contains a and that is maximal for the
property p ∩ {1, f, f 2 , . . . } = ∅. Then p is a prime ideal by Exercise 3.1.10, and
f ∈ p.
For a = 0, Proposition 3.1.11 says that Nil A is the intersection of all the prime
ideals of A. In particular, A and Ared have “the same” Zariski spectrum (more
precisely: the map A → Ared induces a homeomorphism Spec Ared → Spec A).
Definition 3.1.13 A topological space X is called irreducible if it is not the union of
two proper closed subsets. A point x ∈ X is called a generic point of X if X = {x}.
If X contains a generic point, then X is irreducible. One is often interested in
the converse. Note that in case X is a T0 space (for example a subspace of some
Spec A), then X has at most one generic point.
Proposition 3.1.14 Every nonempty closed irreducible subspace of Spec A con-
tains a (unique) generic point. Hence, the nonempty closed irreducible subsets of
Spec A are precisely the sets V(p) where p ∈ Spec A.
Proof Let ∅ = Y ⊆ Spec A be closed and irreducible, and let a := I(Y ). By
Proposition 3.1.7 we must show that a is a prime ideal. Since Y = ∅, we have
a = A. If f, g ∈ A are such that fg ∈ a, then Y ⊆ V(f ) ∪ V(g). Since Y is
irreducible, we obtain without loss of generality that Y ⊆ V(f ), and in particular
that f ∈ a.
Recall that a topological space is quasi-compact if each of its open coverings has
a finite subcovering, and compact if it is quasi-compact and Hausdorff.
Proposition 3.1.15 Let f ∈ A. Then D(f ) is quasi-compact. In particular,

Spec A = D(1) is quasi-compact.
Proof Let T be a subset of A. Then

D(f ) ⊆ D(g) ⇔ V(f ) ⊇ V(T )
g∈T
"

⇔f ∈ Ag (Corollary 3.1.12)
g∈T
"

⇔f ∈ Ag for some finite T ⊆ T
g∈T

⇔ D(f ) ⊆ D(g) for some finite T ⊆ T .
g∈T
Remark 3.1.16 We use K̊(Spec A) to denote the set of all finite unions D(f1 ) ∪
· · · ∪ D(fr ), where r ≥ 1 and fi ∈ A. By Proposition 3.1.15, the elements of
K̊(Spec A) are precisely the open quasi-compact subsets of Spec A. It follows in
particular that K̊(Spec A) is completely determined by the topological space Spec A.
This is generally not the case for the open basis {D(f ) : f ∈ A} of Spec A, cf.
Sect. 3.4.
Let us look at the functorial behaviour of the Zariski spectrum. Let ϕ : A → B
be a ring homomorphism. If q ∈ Spec B, then ϕ −1 (q) is a prime ideal of A. Hence
ϕ induces a map of spectra in the opposite direction,
Spec B → Spec A, q → ϕ −1 (q),
which we denote by Spec ϕ or ϕ ∗ . This map is continuous since (Spec ϕ)−1 DA (f )

= DB ϕ(f ) for all f ∈ A. We conclude that Spec is a functor from the category
of commutative rings to the category of topological spaces.
The map ϕ also induces embeddings of residue fields: if q ∈ Spec B, then the
embedding A/ϕ −1 (q) → B/q induces a field embedding ϕq : κA (ϕ ∗ q) → κB (q).
In other words, the diagram
is commutative.
Thus, interpreting the rings A and B as rings of functions on their spectra (as
explained in Remark 3.1.3), the map ϕ : A → B just pulls back functions on Spec A
via ϕ ∗ : Spec B → Spec A and via the embeddings ϕq of the residue fields.
Two special cases are worthy of investigation in more detail. For the first case, let
S be a multiplicative subset of A and iS : A → S −1 A the canonical homomorphism
a → a/1.
Proposition 3.1.17 The map
Spec(iS ) = iS∗ : Spec S −1 A → Spec A
is a homeomorphism from Spec S −1 A to the subspace DA (S) of Spec A. The residue

homomorphisms (iS )q with q ∈ Spec S −1 A are all isomorphisms.
Proof The map
DA (S) → Spec S −1 A, p → S −1 p := (S −1 A) · iS (p) = { as : a ∈ p, s ∈ S}
is inverse to iS∗ . If a
s ∈ S −1 A, then
iS∗ DS −1 A ( as ) = DA (a) ∩ DA (S),
which shows that the map iS∗ is open, and thus a homeomorphism. For the second
statement, let q ∈ Spec S −1 A and p := iS∗ (q) = iS−1 (q) (i.e., q = S −1 p).
It is immediately clear that the embedding A/p → (S −1 A)/(S −1 p) induces an
isomorphism of the quotient fields.
By Proposition 3.1.17 we can identify Spec S −1 A with the subspace DA (S) =
{p ∈ Spec A : p ∩ S = ∅} of Spec A. In particular we can interpret every basic open
set DA (f ) as the Zariski spectrum of a ring, namely the ring f −∞ A := S −1 A,
where S := {1, f, f 2 , . . . }.
For the second special case, let a ⊆ A be an ideal and ϕ : A → A/a the residue
homomorphism.
Proposition 3.1.18 The map
ϕ ∗ : Spec A/a → Spec A
is a homeomorphism from Spec A/a to the closed subspace VA (a) of Spec A. The
residue field embeddings ϕq with q ∈ Spec A/a are all isomorphisms.
Proof Analogous to the proof of Proposition 3.1.17, where this time ϕ ∗ DA/a (f +
a) = DA (f ) ∩ VA (a) for all f ∈ A.
The observation we made above for S −1 A goes through, mutatis mutandis, for
the ring A/a. The closed subspaces of Spec A are thus also spectra of rings.
In the final part of this section we will explain what we mean by affine varieties
over a field. Our point of view is in essence Weil’s in the sence that all varieties
are reduced. We will show that by considering the associated function algebras we
obtain a coordinate-free description.
Let k be a field with algebraic closure k. If K ⊇ k is an overfield, T a subset of
k[t1 , . . . , tn ] and V a subset of K n , we write
ZK (T ) := {x ∈ K n : f (x) = 0 for all f ∈ T }
and
Ik (V ) := {f ∈ k[t1 , . . . , tn ] : f (x) = 0 for all x ∈ V }.
n
Definition 3.1.19 An affine k-variety is a subset V of k of the form V = Zk (a),
where a is an ideal of k[t1 , . . . , tn ]. If k ⊆ K ⊆ k is an intermediate field, then the
m
elements of V (K) := V ∩ K n are called the K-rational points of V . If V ⊆ k
n
and W ⊆ k are affine k-varieties, then a map ϕ : V → W is called a k-morphism
from V to W if there exist polynomials f1 , . . . , fn ∈ k[t1 , . . . , tm ] such that ϕ(x) =
f1 (x), . . . , fn (x) for all x ∈ V .
Remark 3.1.20 It is not sufficient to just consider the k-rational points V (k) of V
as these usually do not contain enough information about V . For example, V (k) can
be empty even though V is not the empty variety (concrete example: the R-variety
in the plane defined by 1 + t12 + t22 = 0).
n
Remark 3.1.21 If V ⊆ k is a k-variety, then Ik (V ) is clearly the largest ideal a of
k[t1 , . . . , tn ] such that V = Zk (a).
Remark 3.1.22 The definition of affine k-variety depends on a choice of coordi-

nates, which is unsatisfactory for many reasons. A coordinate-free definition is
possible though, as we will explain next.
n
Let V ⊆ k be an affine k-variety. We call
k[V ] := k[t1 , . . . , tn ]/Ik (V )
the affine (k-)algebra of V . It is clear that k[V ] is an affine (i.e., a finitely generated)
reduced k-algebra, which can be interpreted as the algebra of k-morphisms V →
k. If ϕ : V → W is a k-morphism of affine k-varieties, then ϕ induces a
homomorphism ϕ ∗ : k[W ] → k[V ] of k-algebras by “pulling back”.
The point is now that up to isomorphism, V and k[V ] are the “same”. More
precisely, V is the same as k[V ] plus a choice of coordinates. Let us elaborate this.
Consider any overfield K of k, and denote the set of k-algebra homomorphisms
from k[V ] to K by Homk (k[V ], K). Writing ti := ti + Ik (V ) ∈ k[V ], there is a
canonical bijection
Homk k[V ], K −∼
→ ZK Ik (V ) , x → x(t1 ), . . . , x(tn ) .
In particular, if K = k, there is a bijection Homk k[V ], k → V . This motivates the

following definition:
Definition 3.1.23 Given an affine k-algebra A and an overfield K of k, we write
VA (K) := Homk (A, K). If K = k, then VA := VA (k) = Homk (A, k) is called
the (abstract) variety of A. Every k-algebra homomorphism ϕ : A → B induces
a map ϕK ∗ : V (K) → V (K), namely y → y ◦ ϕ, and so in particular a map
B A
∗
ϕ : VB → VA of varieties.
In analogy to the Zariski spectrum we interpret the elements f ∈ A as functions
(with values in k) on the variety VA by setting f (x) := x(f ) for all x ∈ VA .
If we choose a system of generators u1 , . . . , un of A, then we can identify VA
with an affine k-variety in the sense of Definition 3.1.19: the evaluation map
n
VA → k , x → u1 (x), . . . , un (x)
n
is a bijection from VA to the k-variety Zk (a) ⊆ k , where a is the kernel of
the homomorphism k[t1 , . . . , tn ] → A, ti → ui . The same construction with
a different system of generators results in a k-isomorphic k-variety. In other
words, by considering VA , we have associated a (k-)isomorphism class of affine
k-varieties to the k-algebra A, and choosing a representative of this isomorphism
class corresponds to choosing a system of generators of A.
It is clear that if ϕ : A → B is a homomorphism of affine k-algebras, then
ϕ ∗ : VB → VA is a k-morphism of k-varieties (for every choice of coordinates).
The assignments V → k[V ] and A → VA are inverse to each other, at least if

we consider only reduced k-algebras for the second assignment. Indeed, let V ⊆
n
k be an affine k-variety and let A := k[V ] = k[t1 , . . . , tn ]/Ik (V ). If we choose
ti = ti + Ik (V ) for i = 1, . . . , n as a system of generators of A = k[V ], then
n
the evaluation map from VA = Homk (A, k) to k gives a bijection from VA to
Zk Ik (V ) = V . In this way the k-variety VA can be canonically identified with V .
To establish conversely that k[VA ] is isomorphic to A if A is reduced, we need
the following fundamental result from algebraic geometry, which we state without
proof:
Theorem 3.1.24 (Hilbert’s Nullstellensatz) Let A be an affine k-algebra. Then
Nil A is the intersection of all maximal ideals m of A, and [A/m : k] is finite for
each such m.
To give an idea of the proof, the essential step consists of showing that A/m is
a finite extension field of k for every maximal ideal m of the finitely generated k-
algebra A (this result is often also referred to as Hilbert’s Nullstellensatz). It then
follows easily via localization that the closed points of Spec A are dense in Spec A
and the theorem then follows by the Weak Nullstellensatz (Proposition 3.1.11). For
a complete proof of Hilbert’s Nullstellensatz we refer to [68] or [13, Ch. V, §3.3].
Assume now that A is an affine reduced k-algebra. Then Hilbert’s Nullstellensatz
says that for every 0 = f ∈ A there exists x ∈ VA such that f (x)= 0, which in
turn signifies that the canonical homomorphism A → k[VA ] = A/ Ker(x) is an
x∈VA
isomorphism.
n
Summary We have seen that an affine k-variety V ⊆ k is “the same” as an affine
reduced k-algebra A together with a system of generators of A. Henceforth, when
talking about an affine k-variety we will thus usually mean an affine reduced k-
algebra A together with the set VA = Homk (A, k), which offers the advantage of a
coordinate-free approach.
Remark 3.1.25 Grothendieck already recognized that restricting to reduced affine

algebras is artificial. In order to understand non-reduced k-algebras geometrically,
we need to move away from the point of view that an algebraic function V → k (on
n
a k-variety V ⊆ k ) is already determined by its values at the points of V . The right
tool in this situation, that would also allow us to consider non-affine varieties, is
sheaf theory. However, for the objectives we have in mind the “naive” interpretation
presented in this section suffices.
A final comment we want to make is that the k-rational points of an affine k-
variety can be interpreted as a subset of the Zariski spectrum in a canonical way.
Namely, if A is an affine k-algebra, then
VA (k) → Spec A, x → Ker x
is a bijection from VA (k) = Homk (A, k) to the set of those maximal ideals m of A
for which the map k → A → A/m is an isomorphism (i.e., A = m + k · 1). The
topology induced by Spec A on VA (k) is also called the Zariski topology on VA (k).
However, if k = k, then the variety VA = VA (k) associated to A cannot in general
be squeezed into Spec A: if m ∈ Spec A is a closed point there exist in general
several k-embeddings A/m → k and accordingly several points in VA associated to
m.
On the other hand, if k = k is algebraically closed, then x → Ker x is a bijection
from VA to the space of closed points in Spec A. Therefore, the topology induced
on VA by VA → Spec A is also called the Zariski topology on the variety VA . Of
course, historically speaking the situation was reversed since O. Zariski studied the
n
topology on k-varieties V ⊆ k that was named after him long before the spectrum
was conceived in the 1950s.
3.2 Reality for Commutative Rings
Let A be a ring and A2 := {a12 + · · · + an2 : n ≥ 1, ai ∈ A} the semiring generated

by all squares of A.
When attempting to extend the notion of reality from fields to arbitrary rings, two
distinct concepts emerge from the two equivalent field characterizations:1
Definition 3.2.1
(a) A is called real if −1 ∈ A2 . If A is not real (i.e., if there exist elements
a1 , . . . , an ∈ A such that 1 + a12 + · · · + an2 = 0), then A is called nonreal.
(b) A is called real reduced if for all a1 , . . . , an ∈ A, a12 + · · · + an2 = 0 ⇒ a1 =
· · · = an = 0.
(c) An ideal a ⊆ A is called real (resp. real reduced) if the ring A/a is real (resp.
real reduced).
Remarks 3.2.2
(1) The trivial ring is real reduced, but not real. Every nontrivial real reduced ring is
also real. The ring R[x, y]/(x 2 +y 2 ) is real (cf. (5) below), but not real reduced.
(2) If A is real reduced, then A is reduced, and every subring of A is also real
reduced.
(3) An integral domain is real reduced if and only if its field of fractions is real. A
field K is real (reduced) if and only if it has an ordering (cf. Sect. 1.1).
(4) A valuation ring A is real reduced if and only if it is real. (Proof: Real reduced
implies real is clear from (1). Let A be real and let ai ∈ A, ai = 0 such that
a12 +· · ·+an2 = 0. We may assume that v(a1 ) ≤ · · · ≤ v(an ) < ∞. Letting bi :=
ai /a1 ∈ A for i = 1, . . . , n, we then have 1 + b22 + · · · + bn2 = 0, contradiction.)
For example, if A is a residually real valuation ring, then A is real (reduced).
This follows from Corollary 2.7.2 and (5) below. A real (reduced) valuation ring
is in general not residually real though.
(5) If ϕ : A → B is a homomorphism and B is real, then A is also real.
(6) If ϕ : A → B is a homomorphism and b ⊆ B is a real (resp. real reduced) ideal,
then ϕ −1 (b) is also real (resp. real reduced). If ϕ is surjective, then b is real
(resp. real reduced) if and only if ϕ −1 (b) is real (resp. real reduced).
(7) Let A = A1 × · · · × Ar be a direct product of rings with Ai = 0. Then A is real
reduced if and only if the factors A1 , . . . , Ar are all real reduced, and A is real
if and only if at least one of the factors A1 , . . . , Ar is real.
1 In Einführung in die reelle Algebra we followed a suggestion of T.Y. Lam [71], and used
“semireal” and “real” instead of “real” and “real reduced”, respectively. In the meantime the
terminology we use here has become standard.
3.2 Reality for Commutative Rings 111
(8) If S ⊆ A is a multiplicative subset and 0 ∈ S, then the following implications

hold:
A real reduced ⇒ S −1 A real reduced ⇒ S −1 A real ⇒ A real,
but their converses do not hold in general. (Proof of the first implication: If
−1
i (ai /si ) = 0 in S A, then i (ai ti ) = 0 in A for certain ti ∈ S. If A is
2 2
real reduced, it follows that ai ti = 0 and so ai /si = 0 for all i.)
Notation 3.2.3 We denote the set of all real reduced prime ideals of A by
(Spec A)re , i.e.,
(Spec A)re := {p ∈ Spec A : κ(p) is real}.
Various characterizations of real reduced rings are given in:

Proposition 3.2.4 Let A be a ring. The following statements are equivalent:
(i) A is real reduced;
(ii) A is reduced, and all minimal prime ideals of A are real reduced;
(iii) A is reduced, and (Spec A)re is dense in Spec A;
(iv) the intersection of all p ∈ (Spec A)re is {0}.
Proof (i) ⇒ (ii): Let p be a prime ideal of A. Then Ap is real reduced by
Remark 3.2.2(8), hence reduced by Remark 3.2.2(2). If p is minimal, then pAp
is the only prime ideal of Ap . Hence, pAp = 0 and thus Ap = κ(p), i.e., p is real
reduced.
(ii) ⇒ (iii): This is clear since the set of minimal prime ideals of an arbitrary ring
A is dense in Spec A.
(iii) ⇒ (iv): Since (Spec A)re is dense in Spec A, Proposition 3.1.11 yields
! !
{p : p ∈ (Spec A)re } = {p : p ∈ Spec A} = Nil A
and Nil A = (0) since A is reduced.

(iv) ⇒ (i): Let a1 , . . . , ar ∈ A \ {0}. By assumption (iv) there exists p ∈
(Spec A)re such that a1 ∈ p. Since p is real reduced, we have a12 + · · · + ar2 ∈ p, so
in particular = 0.
Corollary 3.2.5 A ring is real reduced if and only if it is a subring of a direct

product of real fields (with infinitely many factors in general).
For the characterization of real rings, we introduce:
Definition 3.2.6 We call an ideal a ∈ A maximally real reduced if a is real reduced,
a = A, and a is maximal with respect to these two properties (i.e., if a ⊆ a and a
is real reduced, then a = a or a = A). A maximally real ideal is defined similarly.
Proposition 3.2.7 An ideal of A is maximally real reduced if and only if it is

maximally real. The maximally real (reduced) ideals of A are precisely those (prime)
ideals p of A that are maximal for the property p ∩ (1 + A2 ) = ∅.
Proof Let S = 1 + A2 . An ideal a of A is real if and only if a ∩ S = ∅. Without
loss of generality we may assume that 0 ∈ S, i.e., that A is real, since otherwise
there is nothing to prove. Every proper ideal b of S −1 A is real (in S −1 A) since
2 2
1+ as11 +· · ·+ asrr ∈ b (with ai ∈ A, si ∈ S) implies b := s 2 +a12 +· · ·+ar2 ∈ b
for certain s ∈ S and ai ∈ A (common denominator), a contradiction since b ∈ S.
If a is a maximally real ideal of A, then a is a prime ideal by Exercise 3.1.10. Since
S −1 a is a maximal ideal of S −1 A, the field S −1 A/S −1 a is real by the observation
above. The embedding A/a → S −1 A/S −1 a then implies that a is real reduced.
Conversely, if a is a maximally real reduced ideal, then a is also maximally real by
the direction already established.
Corollary 3.2.8 An ideal a of A is real if and only if there exists an ideal p ∈

(Spec A)re such that a ⊆ p. In particular, a ring A is real if and only if it contains a
real reduced prime ideal (i.e., if and only if Sper A = ∅, cf. Sect. 3.3).
Proof If a is real, then a is contained in a maximally real ideal p, and p ∈ (Spec A)re
by Proposition 3.2.7. The converse is clear.
A (surprising) reformulation is given by:
Corollary 3.2.9 If −1 is a sum of squares in every residue field of A, then −1 is
already a sum of squares in A itself.
Section 3.1 featured the weak and a strong Nullstellensatz from commutative
algebra. Real algebra has its own versions of these theorems. The weak real
Nullstellensatz describes the intersection of all real reduced prime ideals in any
ring:
Proposition 3.2.10 (Weak Real Nullstellensatz) For any ideal a ⊆ A and any
f ∈ A, the following statements are equivalent:
(i) f ∈ p for all real reduced prime ideals p ⊇ a;
(ii) there exist N ∈ N and a1 , . . . , ar ∈ A such that f 2N + a12 + · · · + ar2 ∈ a.
Proof It suffices to prove the theorem for the ring A/a. So we may assume that
a = (0).
(i) ⇒ (ii): By assumption we have DA (f ) ∩ (Spec A)re = ∅, thus
(Spec f −∞ A)re = ∅. By Proposition 3.2.7 an equation of the form
a1 2 ar 2
1+ + ···+ =0
fn fn
is satisfied in f −∞ A. Hence, f 2m (f 2n + a12 + · · · + ar2 ) = 0 in A for some m ≥ 0,

which proves (ii).
(ii) ⇒ (i): Let p ∈ (Spec A)re . Then by (ii) we have f 2N ∈ p, and so f ∈ p.
3.2 Reality for Commutative Rings 113
√
Definition 3.2.11 If a ⊆ A is an ideal, its real radical re
a is defined as the
intersection of all real reduced prime ideals p ⊇ a.
Remarks 3.2.12
√
(1) A ring is real reduced if and only if re
(0) = (0), cf. Proposition 3.2.4.
(2) By Proposition 3.2.10 we have
√
re
a = {f ∈ A : ∃ N ∈ N, ∃ a1 , . . . , ar ∈ A such that f 2N + a12 + · · · + ar2 ∈ a}.
When giving a direct proof that the right hand side is an ideal, showing closure
under addition requires some effort: assume f 2M + a12 + · · · + ar2 ∈ a and
g 2N + b12 + · · · + bs2 ∈ a. We may take M = N. Then (f + g)4N + (f − g)4N is
a sum of elements of the form f 2m g 2n with m + n = 2N. Therefore, m ≥ N or
n ≥ N in every such term. It follows that −f 2m g 2n is a sum of squares modulo
a, which yields (f + g)4N + c12 + · · · + ct2 ∈ a.
√
Let R be a real closed field, C = R( −1), A an affine R-algebra, and
VA = HomR (A, C) the variety associated to A. As before, we interpret VA (R) =
HomR (A, R) as a subspace of Spec A (cf. Sect. 3.1), as a matter of fact even of
(Spec A)re and present new formulations of the theorems of Artin and Lang that
were established in Chap. 2:
Theorem 3.2.13 (Stellensatz) An affine R-algebra A is real if and only if its
associated variety VA has real points, i.e., if and only if VA (R) = ∅.
Proof The sufficient direction is clear. For the necessary direction, assume that A is
real. By Corollary 3.2.8 there exists p ∈ (Spec A)re , and K := κ(p) is a real function
field over R. Let f1 , . . . , fr be generators of A, and f1 , . . . , fr their images in K.
By Theorem 2.11.8 there exists a place λ : K → R ∪∞ over R such that λ(fi ) = ∞
for i = 1, . . . , r. The composition λ ◦ ρp is an element of VA (R).
In fact, the following stronger statement is true:
Corollary 3.2.14 Let A be an affine R-algebra. Then VA (R) is Zariski dense in
(Spec A)re .
Proof Let f ∈ A such that DA (f ) ∩ (Spec A)re = ∅. Then f −∞ A is a real affine
R-algebra, and so HomR (f −∞ A, R) = ∅ by Theorem 3.2.13. Every element of
this set yields an element of DA (f ) ∩ VA (R) (via composition with A → f −∞ A).
As a combination of the weak real Nullstellensatz and the Artin–Lang Stellensatz

we obtain the following (strong) real Nullstellensatz:
Theorem 3.2.15 (Real Nullstellensatz, D.W. Dubois, J.-J. Risler, 1969/70) Let A
be an affine R-algebra and f ∈ A. Then f vanishes on VA (R) if and only if there
exist N ∈ N and a1 , . . . , ar ∈ A such that f 2N + a12 + · · · + ar2 = 0.
Proof By definition f vanishes on VA (R) if and only if f ∈ I VA (R) . By

Corollary 3.2.14 we have I VA (R) = I (Spec A)re , and the statement follows
from the description of the second set given in Proposition 3.2.10.
3.3 Definition of the Real Spectrum
Let A be a ring.
Definition 3.3.1 (M. Coste, M.-F. Roy, 1979)
(a) The real spectrum Sper A of A is the set of all pairs α = (p, T ) with p ∈ Spec A
and T an ordering of the field κ(p) = Quot A/p. (Other notations in use are
Specr A and SpecR A. In Sect. 3.6 we will motivate our notation Sper A.) The
prime ideal p is called the support of α, denoted p = supp α.
(b) For α = (p, T ) ∈ Sper A, k(α) denotes the real closure of κ(p) with respect to
T . There is a canonical homomorphism
ρp
rα : A −−→ κ(p) → k(α).
Given f ∈ A, we write f (α) instead of rα (f ). Relations such as “f (α) ≥ 0”

are always with respect to the unique ordering of k(α).
(c) Given any subset T of A, we define
H̊A (T ) := {α ∈ Sper A : f (α) > 0 for all f ∈ T },

H A (T ) := {α ∈ Sper A : f (α) ≥ 0 for all f ∈ T },
ZA (T ) := {α ∈ Sper A : f (α) = 0 for all f ∈ T }.
Usually we drop the index A. If T = {f1 , . . . , fr } is finite, we just write

H̊A (f1 , . . . , fr ) instead of H̊A {f1 , . . . , fr } , etc.
(d) The Harrison topology on Sper A is the topology with H̊A := {H̊A (f ) : f ∈ A}
as subbasis of open sets. We call H̊A the Harrison subbasis of the topological
space Sper A.
Remarks 3.3.2
(1) A difference with the Zariski spectrum is that the Harrison subbasis is usually
not an open basis. An example of an open basis for the Harrison topology is
given by
{H̊ (f1 , . . . , fr ) : r ≥ 1, fi ∈ A}.

3.3 Definition of the Real Spectrum 115
Note that Z(f ) = H (−f 2 ) and Z(f1 , . . . , fr ) = Z(f12 +· · ·+fr2 ) = H (−f12 −

· · · − fr2 ). The sets H (T ) and Z(T ) are closed for any T ⊆ A.
(2) In analogy with the Zariski spectrum (cf. Sect. 3.1), writing f (α) instead of
rα (f ) signifies that we interpret the elements f ∈ A as “functions” on Sper A.
The value fields k(α) vary again from point to point. This interpretation is
“faithful” (in the sense that r = (rα ) : A → k(α) is injective) if and
only if A is real reduced (cf. Proposition 3.2.4).α∈Sper A
(3) Every homomorphism ϕ : A → K into an ordered field (K, P ) defines an
element αϕ in Sper A, namely αϕ := (Ker ϕ, P ), where P is the ordering
of κ(Ker ϕ), induced by P . Therefore Sper A can also be defined as the set
of all homomorphisms of A into ordered (or: real closed) fields, modulo an
appropriate equivalence relation. We will not pursue this approach (cf. [11,
§7.1]).
(4) If A = k is a field, then Sper k is the space of orderings of k. Since H k (f ) =
H̊k (f ) for all f ∈ k ∗ , all the sets of the Harrison subbasis are open and closed.
Sper k is in particular a totally disconnected Hausdorff space. (Since Sper A
is always quasi-compact—as we will show in Sect. 3.4—Sper k is a Boolean
space, i.e., compact and totally disconnected.) This topology was discovered by
Harrison (in the case of fields) around 1970. This motivates our use of the term
“Harrison topology” also in the general case.
(5) A ring A is real if and only if Sper A = ∅. This follows from Corollary 3.2.8.
Proposition 3.3.3 The map supp : Sper A → Spec A is continuous. Its image is the
subspace (Spec A)re of real reduced prime ideals.
Proof We have supp−1 DA (f ) = H̊A (f 2 ). The second statement is clear.
An ordering T of κ(p) is determined by T ∩(A/p), i.e., by ρp−1 (T ). Consequently
the elements of Sper A can also be interpreted as certain subsets of A. Given α =
(p, T ) ∈ Sper A, let Pα := ρp−1 (T ) = {f ∈ A : f (α) ≥ 0}. Then P = Pα satisfies
the following properties:
(RO1) P + P ⊆ P, PP ⊆ P;
(RO2) P ∪ (−P ) = A;
(RO3) supp P := P ∩ (−P ) is a prime ideal of A.
Since (RO1) and (RO2) already imply that supp P is an ideal of A, we may replace
(RO3) by
(RO3’) −1 ∈ P , and: a ∈ P , b ∈ P ⇒ −ab ∈ P (a, b ∈ A).
It is easy to see that every subset P ⊆ A that satisfies properties (RO1) to (RO3)
defines an element αP ∈ Sper A, namely αP = (p, P ), where p := supp P and P
is the ordering on κ(p) induced by P (for a, b ∈ A, b ∈ p we have: ρp (a)/ρp (b) ∈
P ⇔ ab ∈ P ). This motivates:
Definition 3.3.4 An ordering of A is a subset P of A that satisfies properties (RO1)
to (RO3) above. The terms cone and prime cone are also used.
The above argument proves:

Proposition 3.3.5 Sper A can be interpreted as the set of all orderings of A. The
topology is generated by the open subbasis {H̊A (f ) : f ∈ A}, where H̊A (f ) denotes
the set of all orderings P of A for which −f ∈ P .
The points of Sper A can thus be viewed in two possible ways. If P ⊆ A is an
ordering, we denote the real closure of κ(supp P ) with respect to P again by k(P ).
We also write f (P ) for the image of f ∈ A in k(P ), and so on.
Definition 3.3.6 Let X be a topological space and x, y ∈ X. If y ∈ {x}, we call y a
specialization of x and x a generalization of y, denoted x % y. If X is a T0 space,
then x ≤ y :⇔ x % y defines a (partial) order relation on X.
If X = Spec A, then p, q ∈ X satisfy p % q ⇔ p ⊆ q. In other words, the
specialization relation is given by inclusion. A similar result is true for the real
spectrum:
Proposition 3.3.7 Let P , Q ∈ Sper A be orderings of A. Then P % Q (in Sper A)
if and only if P ⊆ Q.
Proof
P % Q ⇔ for any f ∈ A such that Q ∈ H̊ (f ) we have P ∈ H̊ (f )

⇔ for any g ∈ A such that g ∈ P we have g ∈ Q
⇔ P ⊆ Q.
Corollary 3.3.8 Sper A is a T0 space.
Corollary 3.3.9 If α ∈ Sper A and supp (α) is a maximal ideal of A, then α is a

closed point of Sper A.
We will present some examples shortly. In particular we note that Example 3.3.14
below shows that the converse of Corollary 3.3.9 is false. Let us first discuss the
functorial behaviour of Sper, which is similar to the functorial behaviour of Spec
seen in Sect. 3.1.
If ϕ : A → B is a ring homomorphism and Q ⊆ B is an ordering of B, then
ϕ −1 (Q) is an ordering of A. Hence ϕ induces a map
Sper ϕ = ϕ ∗ : Sper B → Sper A, Q → ϕ −1 (Q),
which is continuous since (Sper ϕ)−1 H̊A (f ) = H̊B ϕ(f ) . Thus Sper, like Spec,
is a functor from commutative rings to topological spaces. (Moreover, supp is a
morphism Sper → Spec of functors since suppA ◦(Sper ϕ) = (Spec ϕ) ◦ suppB .)
The map ϕ also yields embeddings of real closed fields as follows: let β ∈ Sper B
and α = ϕ ∗ (β) ∈ Sper A, then ϕ induces an order preserving (with respect to α, β)
embedding ϕsupp β : κ(supp α) → κ(supp β), and hence ϕβ : k(α) → k(β) (cf.
Sect. 1.11). Letting p := supp α and q := supp β we thus obtain the commutative
diagram
Continuing the analogy with the Zariski spectrum we can think of ϕ as pulling back
the “functions” g ∈ B that are defined on Sper B via Sper ϕ (and via the embeddings
ϕβ ).
Proposition 3.3.10 Consider a multiplicative subset S ⊆ A and its associated
canonical homomorphism iS : A → S −1 A. Then Sper(iS ) is a homeomorphism

from Sper S −1 A to the subspace {α ∈ Sper A : S ∩ supp α = ∅} = s∈S H̊A (s 2 )
of Sper A. If β ∈ Sper S −1 A and α = (Sper iS )(β), then (iS )β : k(α) → k(β) is an
isomorphism.
Proposition 3.3.11 Consider an ideal a ⊆ A and its associated residue map

π : A → A/a. Then Sper π is a homeomorphism from Sper A/a to the closed
subspace ZA (a) = {α ∈ Sper A : a ⊆ supp α} of Sper A. If β ∈ Sper A/a and
α = (Sper π)(β), then πβ : k(α) → k(β) is an isomorphism.
The proof of Proposition 3.3.11 is similar to:

Proof of Proposition 3.3.10 Let X := s∈S H̊A (s 2 ). The inverse X → Sper S −1 A
of Sper(iS ) is P → S −2 P := {f/s 2 : f ∈ P , s ∈ S} (P ∈ X). The homeomorphy
follows from
f1 fr
(Sper iS ) H̊S −1 A ,..., = X ∩ H̊A (s1 f1 , . . . , sr fr ).
s1 sr
Finally, (iS )β is an isomorphism since (iS )supp β is an isomorphism (cf. Sect. 3.1)
and order preserving.
Corollary 3.3.12 For any p ∈ Spec A there exists a natural homeomorphism
Sper κ(p) → {α ∈ Sper A : supp α = p},
induced by ρp : A → κ(p).
Example 3.3.13 Let R be a real closed field and A = R[[t]] the ring of formal
power series over R. The only prime ideals of A are (0) and mA = (t), and κ(A) =
A/mA ∼ = R. By Sect. 2.9, Sper A has precisely three elements, namely the orderings
P0 , P− , P+ of A, where supp P0 = mA and supp P− = supp P+ = (0). From
P− ∪ P+ ⊆ P0 it follows that P0 is closed and has P− and P+ as generalizations.
The situation can be illustrated by means of the following symbolic diagram, where
the arrows represent specialization:
Example 3.3.14 Assume again that R is a real closed field and let A = R[t]. The
real reduced prime ideals of R[t] are (0) and all (t − c), where c ∈ R. The orderings
of R(t) were determined in Sect. 2.9. The following list therefore gives a complete
description of all points of Sper R[t]:
1. for every c ∈ R, a closed point Pc and also two generalizations Pc,− and Pc,+ of
Pc (simply written as c, c− , c+ );
2. the closed points P−∞ , P+∞ with support (0) (simply written as −∞, +∞);
3. (only if R = R) for every free Dedekind cut ξ of R, a closed point Pξ with
support (0) (simply written as ξ ).
As in the previous example, we illustrate the situation with a symbolic diagram:
The diagram shows that Sper R[t] possesses a natural linear order relation “≤” that
extends the ordering of R.
An open basis of the topology of Sper R[t] is given by all open intervals
]a, b[ := {α ∈ Sper R[t] : a < α < b}
and all [−∞, b[, ]a, ∞] where a, b ∈ R, defined similarly. Note that ]a, b[ contains
the points a+ and b− , but not the points a− , a, b, b+ , and similarly for the other
intervals.
In this concrete example it is not difficult to see that Sper R[t] is quasi-compact
and that the subspace topology induced on R ⊂ Sper R[t] is precisely the strong
topology. In fact, this observation is true in general, as we will show now.
√
Let R be a real closed field, C = R( −1) and A an affine R-algebra with
associated variety V := VA = HomR (A, C) and set of real points V (R) := VA (R).
In Sect. 3.1 we interpreted V (R) as a subset of Spec A via the identification of x ∈

V (R) with mx := Ker(x). Since κ(mx ) = R possesses a unique ordering we can
view V (R) as a subset of Sper A as well: there is a canonical injection
V (R) → Sper A, x → Px := {f ∈ A : f (x) ≥ 0}

rP
(whose image consists of all P ∈ Sper A for which R → A −→ k(P )
is an isomorphism). Furthermore, V (R) consists of closed points of Sper A by
Corollary 3.3.9.
Let u1 , . . . , un be generators of A and let j : V (R) → R n be the associated
embedding, i.e., j (x) = u1 (x), . . . , un (x) . The strong topology on R n (cf.
Sect. 2.6) induces a topology on V (R) via j , which we also call the strong topology
(and which is independent of the choice of j , as we will see shortly). The following
fact is an indicator of how useful the real spectrum is when studying the geometry
of real varieties:
Proposition 3.3.15 The relative topology of VA (R) in Sper A coincides with the
strong topology on VA (R), which is in particular independent of the choice of
embedding VA (R) → R n .
Proof A basis of the strong topology on V (R) is given by the system of sets {x ∈
V (R) : f1 (x) > 0, . . . , fr (x) > 0} (where r ≥ 1 and fi ∈ A), which can be
rewritten as V (R) ∩ H̊A (f1 , . . . , fr ) upon identification of V (R) with its image in
Sper A.
Remark 3.3.16 The real points of V , equipped with the strong topology, therefore
form a subspace of Sper A. In other words, we do not loose any information about
V (R) after the transition to Sper A. Furthermore, Sper A has a great advantage
compared to V (R) as V (R) has bad topological properties when R = R (totally
disconnected, almost never (locally) compact), whereas Sper A is quasi-compact
and only has finitely many connected components (cf. [11, §7.5]), which facilitates
the definition of a non-trivial concept of connectedness even when R = R. (A
different approach that does not use the real spectrum can be found in [31].) Thus,
the addition of “ideal” points results in a much more geometric space. The example
of the affine line A = R[t] above (Example 3.3.14) already illustrates nicely how
the additional points in Sper A “close the gaps” (if R = R) and make the space
quasi-compact.
An immediate consequence of the Artin–Lang theorem is that V (R) is dense as
a subspace of Sper A!
Theorem 3.3.17 VA (R) is a dense subspace of Sper A.

Proof Let f1 , . . . , fr ∈ A such that H̊ (f1 , . . . , fr ) = ∅ and let α = (p, T ) ∈
H̊ (f1 , . . . , fr ). Then κ(p) is a function field over R and T is an ordering of κ(p)
for which f1 , . . . , fr > 0, where fi := ρp (fi ) ∈ κ(p). By Theorem 2.11.9 there
exists an R-homomorphism ϕ : A/p → R such that ϕ(fi ) > 0 for i = 1, . . . , r.

ϕ
The composition x : A → A/p −
→ R is a point of V (R) ∩ H̊ (f1 , . . . , fr ).
We will present a substantial generalization of this fact in Theorem 3.5.3.
3.4 Constructible Subsets and Spectral Spaces
Let A be a ring. In this section we will introduce a second topology on Sper A,

the constructible topology. We will always think of the Harrison topology as the
“proper” topology of Sper A, while the constructible topology will rather play an
auxiliary role. In situations where we do not specify which topology is used for
Sper A, it will always be the Harrison topology.
Definition 3.4.1
(a) A partially ordered set (L, ≤) is called a lattice (with 0 and 1) if every finite
subset X ⊆ L has a greatest lower bound inf X and a least upper bound sup X
in L. In particular, 0 := sup ∅ = inf L and 1 := inf ∅ = sup L exist. For
x, y ∈ L we use the notation
x ∧ y := inf{x, y} , x ∨ y := sup{x, y}.
(b) A Boolean lattice is a lattice (L, ≤) that satisfies the following properties:
(BL1) Distributivity: x ∧ (y ∨ z) = (x ∧ y) ∨ (x ∧ z) for all x, y, z ∈ L;
(BL2) Complement: for every x ∈ L, there exists x ∈ L such that x ∧ x = 0
and x ∨ x = 1.
Remarks 3.4.2 (cf. [43])

(1) In a Boolean lattice L the complement x of x is uniquely determined and the
dual identity of (BL1),
x ∨ (y ∧ z) = (x ∨ y) ∧ (x ∨ z),
is satisfied for all x, y, z ∈ L.

(2) Boolean lattices can also be defined as those Z/2Z-algebras B that satisfy x 2 =
x for all x ∈ B, the Boolean algebras. Endowing a Boolean lattice with the
operations
x + y := (x ∧ y ) ∨ (x ∧ y) and xy := x ∧ y
3.4 Constructible Subsets and Spectral Spaces 121
turns it into a Boolean algebra, and conversely every Boolean algebra becomes
a Boolean lattice via
x ∧ y := xy , x ∨ y := x + y + xy , x := 1 + x.
Definition 3.4.3
(a) K(Sper A) denotes the Boolean sublattice of the power set 2Sper A generated by
the Harrison subbasis H̊A of Sper A. The elements of K(Sper A) are called the
constructible subsets of Sper A. We let
K̊(Sper A) := {Y ∈ K(Sper A) : Y is open},

K(Sper A) := {Y ∈ K(Sper A) : Y is closed}.
(b) The constructible topology on Sper A is the topology generated by K(Sper A)

as basis of open sets.
Remarks 3.4.4
(1) K(Sper A) is the smallest subset K of 2Sper A such that H̊A ⊆ K and which
is stable under taking finite intersections and complements. The elements of
K(Sper A) are precisely the finite unions of sets of the form
H̊A (f1 , . . . , fr ) ∩ H A (g1 , . . . , gs ) (fi , gj ∈ A),
or those of the form
H̊A (f1 , . . . , fr ) ∩ ZA (g) (fi , g ∈ A),
or those of the form
{α ∈ Sper A : sgn f1 (α) = ε1 , . . . , sgn fr (α) = εr } fi ∈ A, εi ∈ {−1, 0, 1} .
(2) The constructible topology is finer than the Harrison topology. If A = k is a

field, then both topologies coincide since the Harrison subbasis is stable under
taking complements: (Sper k) \ H̊k (f ) = H̊k (−f ) for all f ∈ K ∗ .
The constructible topology is still coarse enough to possess pleasing properties,
as we will see next. We note that the proof of Theorem 3.4.5 belongs to the realm of
model theory rather than real algebra.
Theorem 3.4.5 Sper A, endowed with the constructible topology, is a totally
disconnected compact (Hausdorff) space. The constructible subsets of Sper A are
precisely the clopen sets of the topology.
Proof Let us prove the second statement first. Consider the constructible topology
on Sper A. Then all Y ∈ K(Sper A) are clopen by definition. Conversely, let Y ⊆
Sper A be clopen. Since Sper A is compact (see below), Y is the union of finitely
many constructible sets, and thus constructible
itself.
To prove compactness, consider Z = f ∈A {0, 1}, endowed with the product
topology. Then Z is totally disconnected and also compact by Tykhonov’s theorem.
We identify an element (εf )f ∈A with the subset {f ∈ A : εf = 1} of A, in
other words we identify Z with the power set of A. It follows that j (α) :=
{f ∈ A : f (α) ≥ 0} = Pα defines an injection j : Sper A → Z, and the
topology induced by Z (via j ) on Sper A is precisely the constructible topology, cf.
Remark 3.4.4(1). It remains to show that j (Sper A) is closed in Z, or equivalently
that Z \ j (Sper A) is open. To this end, let S ⊆ A be a subset such that S ∈
j (Sper A), i.e., such that S is not and ordering of A. Then one of the following
axioms (cf. Sect. 3.3) is violated by S:
(1) S + S ⊆ S,
(2) SS ⊆ S,
(3) S ∪ (−S) = A,
(4) −1 ∈ S,
(5) a ∈ S, b ∈ S ⇒ −ab ∈ S (a, b ∈ A).
But then this particular axiom is also violated in a neighbourhood of S in Z! For
example, if S violates axiom (5), then there exist a, b ∈ A \ S such that −ab ∈ S.
Then U := {T ⊆ A : a ∈ T , b ∈ T , −ab ∈ T } is a neighbourhood of S in Z, and
each T ∈ U violates axiom (5) for the same reason that S violates axiom (5).
Since the Harrison topology is coarser than the constructible topology, we obtain:
Corollary 3.4.6 The Harrison topology on Sper A is quasi-compact.
Let ϕ : A → B be a ring homomorphism. Then Sper ϕ : Sper B → Sper A is
continuous in the constructible topology since (Sper ϕ)−1 H̊A (f ) = H̊B ϕ(f )
for all f ∈ A and taking inverse images commutes with the Boolean operations.
Nevertheless, the image of a constructible set is in general not constructible anymore
(consider for example the map R[t] → R(t) and the set Sper R(t)). Consequently
it is often necessary to consider more general subspaces.
Definition 3.4.7 A subset Y of Sper A is called proconstructible if Y is the
intersection of (arbitrarily many) constructible subsets of Sper A or, equivalently,
if Y is closed in the constructible topology.
Remarks 3.4.8
(1) The equivalence of the conditions in the definition follows from the fact that the
complement of a subset Y which is closed in the constructible topology can be
written as the union of constructible subsets of Sper A.
(2) Some examples of proconstructible subspaces of Sper A are given by
Sper S −1 A and Sper A/a. If S, resp. a, is finitely generated as semigroup,
resp. ideal, then it is constructible. A further example is given by the sets

H̊A (T ) and H A (T ) for arbitrary T ⊆ A.
Proposition 3.4.9 Sper ϕ is continuous in the constructible topology for every

ϕ : A → B. Preimages and images of proconstructible subsets are again procon-
structible.
Proof The first statement was remarked earlier. For the second statement, assume
that Y ⊆ Sper B is proconstructible. Then Y is compact in the constructible
topology. The same is then true for (Sper ϕ)(Y ) and so (Sper ϕ)(Y ) is closed in
Sper A for the constructible topology.
Corollary 3.4.10 (of Theorem 3.4.5)

(a) Let Y be a proconstructible subset of Sper A. Then every covering of Y by
constructible subsets of Sper A has a finite subcovering.
(b) If a family of proconstructible subsets of Sper A has empty intersection, this is
already the case for a finite subfamily.
For example, an open subset of Sper A is proconstructible if and only if it is quasi-
compact, and thus constructible (i.e., in K̊(Sper A)). These sets are thus precisely
the finite unions of sets of the form H̊ (f1 , . . . , fr ).
A very important consequence is given by:
Proposition
3.4.11 Let Y be any proconstructible subset of Sper A, then Y =
{y} (with respect to the Harrison topology).
y∈Y
Corollary 3.4.12 A proconstructible subset of Sper A is (Harrison) closed if and
only if it is stable under specialization (in Sper A). A constructible subset is open if
and only if it is stable under generalization.
Proof of Proposition 3.4.11 Only the inclusion “⊆” is nontrivial. Let z ∈ Y and let
{Ui : i ∈ I } be the family of open
constructible neighbourhoods of z in Sper A.
Then
Y ∩ Ui = ∅ for all i and so Y ∩ Ui = ∅ by Corollary 3.4.10. For y ∈ Y ∩ Ui
i∈I i∈I
we have z ∈ {y}.
Whereas the closure of any proconstructible set is trivially proconstructible
again, there are examples of constructible sets whose closure is no longer con-
structible. However, for real spectra of affine algebras over fields (or more generally
excellent rings) this does not happen, cf. [1, Theorem 3.1].
Corollary 3.4.13 If Y is a closed irreducible subset of Sper A and Y = ∅, then Y
contains a (unique) generic point.
Proof Let Z be the intersection of Y with all U ∈ K̊(Sper A) that satisfy Y ∩U = ∅.
Since Y is irreducible, Y meets the intersection of only finitely many such U . Hence,
Z = ∅ by Corollary 3.4.10. For any z ∈ Z we have Y = {z}. Indeed, otherwise
there exists y ∈ Y \ {z} and then y has a neighbourhood V such that z ∈ V , a

contradiction.
Remark 3.4.14 If A is a noetherian ring, for instance an affine algebra over a field,
then X = Spec A has only finitely many irreducible components (i.e., A has only
finitely many minimal prime ideals). These play an important role in algebraic
geometry. In contrast, the spaces X = Sper A almost always have infinitely many
minimal points, also in the “geometric setting”, which is why the concept of
irreducible components is less useful in this context.
We take a temporary leave of real algebra and embark on an excursion into point-
set topology. Specifically we will have a closer look at spectral spaces. These satisfy
all the topological properties that we already observed for the Zariski spectrum and
the real spectrum. Spectral spaces elucidate the boundary between real algebraic
and purely topological arguments. Moreover, in one swoop we obtain the properties
of many new spaces that occur naturally in the study of spectra (for example,
proconstructible subspaces are spectral). The main examples of spectral spaces are
the real spectra and especially the Zariski spectra. In fact, Hochster has shown that
these are actually all the examples, cf. [51].
Definition 3.4.15
(a) A spectral space is a topological space X that satisfies the following
properties:
(1) X is a quasi-compact T0 space;
(2) the intersection of any two open quasi-compact subsets of X is again quasi-
compact;
(3) the open quasi-compact sets form a basis of the topology on X;
(4) every nonempty closed irreducible subspace of X has a (unique) generic
point.
(b) Let X and Y be spectral spaces. A spectral map is a map f : X → Y such that
for all V ⊆ Y , if V is open and quasi-compact, then so is f −1 (V ). (In particular,
f is continuous.)
Definition 3.4.16 Let X be a spectral space.

(a) We denote the Boolean lattice generated by the open quasi-compact subsets of
X (in 2X ) by K(X). Its elements are called the constructible subsets of X. The
open, resp. closed, constructible subsets are denoted K̊(X), resp. K(X).
(b) The topology on X that has K(X) as open basis is called the constructible
topology. The space X endowed with this topology is denoted Xcon . The original
topology on X is then often called the spectral topology on X to make the
distinction.
Examples 3.4.17
(1) The spectral maps are precisely those continuous maps X → Y for which
Xcon → Ycon is also continuous.
(2) Sper A is a spectral space, and the current definition of constructible subsets is
the same as the original one. If ϕ is a ring homomorphism, the map Sper ϕ is
spectral. In a similar fashion, Spec A is a spectral space and Spec ϕ is a spectral
map.
(3) Every Boolean space (i.e., a compact and totally disconnected topological
space) is spectral. Its constructible sets are precisely the clopen sets, which
means that in this case the spectral and constructible topologies coincide. The
Boolean spaces are precisely the spectral spaces that are Hausdorff.
Proposition 3.4.18 Consider a spectral space X. Then Xcon is compact and totally
disconnected. The identity map Xcon → X is spectral, and (Xcon )con = Xcon .
Proof The only nontrivial statement that needs a proof is the quasi-compactness
of Xcon , i.e., the finite intersection property for K(X). We start by showing the
following
Claim: If Y ={Yi : i ∈ I } is a family of either open
quasi-compact or closed sets in
X such that Yj = ∅ for all finite J ⊆ I , then Yi = ∅.
j ∈J i∈I
Proof of the claim: Without loss of generality we may assume that Y is maximal
with respect to the indicated property (by Zorn’s Lemma). Let Z be the intersection
of the closed sets in Y. Then Z = ∅ (since X is quasi-compact), hence also Z ∈ Y.
Assume for the sake of contradiction that Z is reducible. Then Z = Z1 ∪Z2 for some
closed Zi Z. Since Zi ∈ Y, there exist open U1 , U2 ∈ Y such that Ui ∩ Zi = ∅
for i = 1, 2. In particular, (U1 ∩ U2 ) ∩ Z = ∅. But U1 ∩ U2 ∈ Y since Y is maximal,
a contradiction.Hence Z is irreducible. It is now clear that the generic point of Z is
an element of Yi . This proves the claim.
i
The remainder of the proof is a modification of an argument provided by
Marcus Tressl. Consider the set Ultra K(X) of all ultrafilters in the lattice K(X),
cf. Definition 3.5.10. For x ∈ X, let
Fx := {U ∈ K(X) : x ∈ U }
denote the principal ultrafilter

generated by x. If F is an arbitrary ultrafilter in K(X),
then the intersection U ∈F U of all sets in F is nonempty by the claim. Let x be in
this intersection. Then F ⊆ Fx , and so F = Fx . It follows that the map
X → Ultra K(X), x → Fx (3.1)
is surjective, hence also bijective since the injectivity is clear.

If we interpret K(X) in the usual way as a (Boolean) ring, then a subset P of
K(X) is a prime ideal of K(X) if and only if {X \ K : K ∈ P } is an ultrafilter in
K(X). Thus (3.1) yields a bijective map
X → Spec K(X), x → Px := {K ∈ K(X) : x ∈

/ K}. (3.2)
Transferring the Zariski topology from Spec K(X) to the set X via the bijection (3.2)
yields the constructible topology on X, which can be seen by direct verification.
Hence the map (3.2) is a homeomorphism Xcon ≈ Spec K(X) of topological spaces.
Since every Zariski spectrum is quasi-compact (cf. Proposition 3.1.15) we conclude
that Xcon is also quasi-compact.
Corollary 3.4.19 The elements of K̊(X) are precisely the open quasi-compact
subsets of X.
Definition 3.4.20 Let X be a spectral space. A subset Y of X is called procon-

structible if Y is the intersection of constructible subspaces (or equivalently, if Y is
closed in Xcon ).
As before we have:
Corollary 3.4.21 Let X be a spectral space.
(a) Every covering of a proconstructible set by constructible sets has a finite
subcovering.
(b) If a family of proconstructible sets in X has empty intersection, this is already
the case for a finite subfamily.
(c) If Y is a proconstructible subset of X, then Y = {y}.
y∈Y
Suitably adapted, Corollary 3.4.12 also holds in this context.

Proposition 3.4.22 Let X be a spectral space and Y a subspace of X.
(a) Y is proconstructible in X if and only if Y itself is a spectral space and the
inclusion Y → X is spectral.
(b) Assume that Y is proconstructible. Then
K(Y ) = Y ∩ K(X), K̊(Y ) = Y ∩ K̊(X), K(Y ) = Y ∩ K(X),
where Y ∩ K(X) denotes {Y ∩ K : K ∈ K(X)}, etc. The constructible topology

on Y is the subspace topology of the constructible topology on X.
Proof The statements are easy to prove with the help of Proposition 3.4.18. The
details are left to the reader.
We finish this section with another useful concept:
Definition 3.4.23 Let X be a spectral space. The set X, endowed with the topology
generated by K(X) as basis of open sets is denoted X∗ and called the inverse
spectral space of X.
3.5 The Geometric Setting: Semialgebraic Sets and Filter Theorems 127
The terminology is motivated by:

Proposition 3.4.24 Let X be a spectral space. Then X∗ is also a spectral space,
and for all x, y ∈ X we have: x % y in X ⇔ y % x in X∗ . Furthermore, K(X∗ ) =
K(X), K̊(X∗ ) = K(X) and K(X∗ ) = K̊(X), hence also (X∗ )con = Xcon and
X∗∗ = X.
Proof For all x, y ∈ X we have: x % y in X ⇔ y ∈ Y for all Y ∈ K(X) such
that x ∈ Y ⇔ y % x in X∗ . Hence X∗ is a T0 space. Furthermore, K(X) consists
precisely of the sets that are open and quasi-compact in X∗ since every Y ∈ K(X)
is quasi-compact in X∗ by the compactness of Xcon . This establishes the validity
of the spectral space axioms (1) to (3) for X∗ . For axiom (4): Let ∅ = Y ⊆ X
be closed and irreducible in X∗ , and let Z := {Z ∈ K(X) : Y ∩ Z = ∅}. For
every finite collection of Zi ∈ Z we have Y ∩ Z1 ∩ · · · ∩ Zr = ∅ (irreducibility).
Moreover, Y is an intersection of sets from K̊(X), and is thus
proconstructible in
X. By Corollary 3.4.21(b) there is then an element y ∈ Y ∩ Z∈Z Z, and y is the
generic point for the subspace Y of X∗ . We conclude that X∗ is a spectral space,
and the remaining statements follow immediately.
3.5 The Geometric Setting: Semialgebraic Sets and Filter

Theorems
The “geometric setting” refers to the case of an affine variety (or rather an affine
algebra) over a real closed field. In this context there exists quite a concrete
geometric interpretation of Sper A and its constructible subspaces via semialgebraic
subsets of VA (R). √
We assume throughout this section that R is a real closed field, C = R( −1),
A is an affine R-algebra, V = VA = HomR (A, C) is its associated variety and
V (R) = VA (R) = HomR (A, R) is the set of real points of the variety. We view
V (R), endowed with the strong topology, as a subspace of Sper A, as explained in
Sect. 3.3. Given any subset Y of Sper A we may then consider the subset Y ∩ V (R)
of V (R).
Definition 3.5.1
(1) We let

S V (R) := V (R) ∩ K(Sper A) = Y ∩ V (R) : Y ∈ K(Sper A)
and call the sets M ∈ S V (R) the semialgebraic subsets of V (R). More
generally, for M ∈ S V (R) we denote the set of semialgebraic subsets of
M by

S(M) := M ∩ K(Sper A) = M ∩ S V (R) = N ∈ S V (R) : N ⊆ M .
Clearly, S(M) is a Boolean sublattice of 2M .

(2) For any semialgebraic subset M of V (R) we let

S̊(M) := N ∈ S(M) : N is open in M ,

S(M) := N ∈ S(M) : N is closed in M ,
with respect to the strong topology.
Remark 3.5.2 The semialgebraic subsets of V (R) are thus precisely those subsets
that can be described by finitely many (equalities and) inequalities. In other words,
they are the finite unions of sets of the form
{x ∈ V (R) : f (x) = 0, g1 (x) > 0, . . . , gr (x) > 0},
where r ≥ 0 and f, g1 , . . . , gr ∈ A (cf. Sect. 3.4). In the special case A = R[t],

where V (R) = R, they are thus precisely the unions of finitely many (possibly
degenerate) intervals.
A crucial observation is now that in the transition from a constructible subset Y
of Sper A to the semialgebraic subset Y ∩ V (R) of V (R) no information is lost. In
other words, Y can be recovered! This is again a consequence of the Artin–Lang
Theorem:
Theorem 3.5.3 Let Y be a constructible subset of Sper A. Then Y ∩ V (R) is dense
in Y with respect to the constructible topology (and a fortiori with respect to the
Harrison topology).
Proof It suffices to show that if Y = ∅, then also Y ∩ V (R) = ∅. Assume without
loss of generality that Y = ZA (f )∩H̊A (g1 , . . . , gr ) with f, gi ∈ A. Let B := A/Af
and W (R) := VB (R) = HomR (B, R) (the real points of the variety associated to
B). We must show that W (R)∩ H̊B (g1 , . . . , gr ) = ∅, where gi denotes the image of
gi in B. By assumption we have H̊B (g1 , . . . , gr ) = ∅. The statement thus follows
from the density of W (R) in Sper B, cf. Theorem 3.3.17.
Corollary 3.5.4 Let x ∈ Sper A. Then {x} is constructible if and only if x ∈

V (R).
Corollary 3.5.5 The map Y → Y ∩ V (R) defines an isomorphism of Boolean

lattices K(Sper A) → S V (R) . More generally, if Z is a constructible subset
of Sper A and M := Z ∩ V (R), then the map Y → Y ∩ V (R) defines a lattice
isomorphism from K(Z) to S(M).
Definition 3.5.6 We use the tilde to denote the map S V (R) → K(Sper A)
which is the inverse of Y → Y ∩V (R). Thus, for a semialgebraic subset M of V (R),
denotes the (uniquely determined) constructible subset of Sper A that satisfies
M
M = M ∩ V (R). In fact, M
is the closure of M in Sper A with respect to the
constructible topology.
In this context the following fact is very important but beyond the scope of this
book:
Theorem 3.5.7 Let M and N be semialgebraic subsets of V (R) such that N ⊆ M.
is also open in M
If N is open in M, then N (the reverse is trivially true). Hence the
following equalities also hold:
S̊(M) = M ∩ K̊(Sper A) and S(M) = M ∩ K(Sper A).
An equivalent formulation is given by:

Theorem 3.5.8 (Finiteness Theorem) Let U, M ⊆ V (R) be semialgebraic with
U ⊆ M and U open in M. Then U is the union of finitely many sets of the form
{x ∈ M : f1 (x) > 0, . . . , fr (x) > 0},
where r ∈ N and f1 , . . . , fr ∈ A.
Example 3.5.9 Let A = R[t], hence V (R) = R. We refer to Sect. 3.3 for an explicit
description of Sper R[t] and use the natural total order relation ≤ on Sper R[t]. The
images of the intervals M ⊆ R under the map : S V (R) → K Sper R[t] are
as follows:
M = [a, b] = [a, b],

M
M = [a, b[ = [a, b[ = [a, b− ],
M
M = ]a, b[ = ]a, b[ = [a+ , b− ],
M
M = ]−∞, a[ = [−∞, a− ] = [−∞, a[ ,
M
..
.
are described in terms of the order relation ≤ on Sper R[t].

where the intervals M
We should pay attention to the fact that there are other, non-constructible, subsets
Y = M of Sper R[t] such that M = V (R) ∩ Y (for example, Y := [a− , b+ ] is such
a set for M = [a, b], and is actually also closed). Thus, we only obtain a bijection
when restricting ourselves to constructible subspaces of Sper A.
For the remainder of this section we require some additional concepts. Let X
be a set and L a sublattice of 2X (i.e., {∅, X} ⊆ L and L is closed under finite
intersections and unions).
Definition 3.5.10 Let L be a sublattice of 2X .

(a) A filter of L is a nonempty subset F of L with ∅ ∈ F that satisfies the following
properties:
(1) A, B ∈ F ⇒ A ∩ B ∈ F ;
(2) A ∈ F , B ∈ L, A ⊆ B ⇒ B ∈ F .
(b) A filter F is called a prime filter of L if for all A, B ∈ L,
A ∪ B ∈ F ⇒ A ∈ F or B ∈ F.
(c) A maximal filter of L is called an ultrafilter.

(d) The sets of filters, prime filters and ultrafilters of L are denoted Filt(L),
Prime(L) and Ultra(L), respectively.
Remarks 3.5.11

(1) If E is a subset of L such that A∈E A = ∅ for all finite E ⊆ E, then E is
contained in a filter of L. The smallest such filter is
!
B ∈ L : there exists a finite subset E ⊆ E such that A⊆B .
A∈E
Thus, a filter F is ultra if and only if for all B ∈ L \ F there exists an A ∈ F

such that A ∩ B = ∅. Every ultrafilter is a prime filter.
(2) If L is a Boolean sublattice of 2X (i.e., if A ∈ L, then also X \ A ∈ L), then
prime filters and ultrafilters in L coincide. These are precisely the filters F such
that A ∈ F or X \ A ∈ F , for all A ∈ L.
The following terminology is chosen ad hoc:
L of 2X is called well-behaved if for all subsets E
Definition 3.5.12 A sublattice
all B ∈ L such that A∈E A ⊆ B, there exists a finite subset E of E such
of L and
that A ⊆ B. We write
A∈E
!
pro-L := Y ⊆ X : there exists E ⊆ L such that Y = A
A∈E
for the lattice of pro-L sets in X.
Example 3.5.13 Let X be a spectral space. For example, X = Sper A. Then

K(X), K̊(X) and K(X) are well-behaved sublattices of 2X . Furthermore, K(X)
is Boolean. We will shortly see that the filters of these lattices offer a different way
of describing X.
Lemma 3.5.14 Let X be a set and L a well-behaved sublattice of 2X . Then the map
!
Filt(L) → (pro-L) \ {∅}, F → A
A∈F
is an inclusion reversing bijection. The inverse map is given by
Y → FY := {B ∈ L : Y ⊆ B}.
ProofIt is clear that FY is a filter of L for ∅ = Y ⊆ X. If F ∈ Filt(L) and

Y = A∈F A, then we must show that FY = F . The inclusion “⊇” is trivial. The
reverse inclusion follows from the fact that L is well-behaved.
An immediate consequence is:
Corollary 3.5.15 Let L be a well-behaved sublattice of 2X . Under the above
bijection Filt(L) → (pro-L) \ {∅},
(a) the prime filters of L correspond to the pro-L sets ∅ = Y ⊆ X that satisfy
A, B ∈ L, Y ⊆ A ∪ B ⇒ Y ⊆ A or Y ⊆ B;
(b) the ultrafilters of L correspond to the minimal nonempty pro-L sets.
Definition 3.5.16 Let X be a spectral space. We let
X max := {x ∈ X : y ∈ {x} ⇒ y = x} and X min := {x ∈ X : x ∈ {y} ⇒ y = x}.
These are thus the sets of maximal, resp. minimal points with respect to the
specialization relation. We observe that Xmax is precisely the subset of closed points
of X, and that Xmin = (X∗ )max and Xmax = (X∗ )min , where X∗ is the inverse
spectral space of X, cf. Sect. 3.4.
Now we are ready to describe X in terms of filters of lattices of constructible
subsets.
Proposition 3.5.17 Let X be a spectral space.
(a) There exists a canonical inclusion reversing bijection
Filt K(X) → {Y ⊆ X : Y is closed and Y = ∅},

namely F → A∈F A. Its inverse is given by Y → FY = {A ∈ K(X) : Y ⊆
A}.
(b) There exists a canonical bijection
X → Prime K(X), x → Fx = {A ∈ K(X) : x ∈ A}.
(c) The map from (b) induces a bijection
Xmax → Ultra K(X), x → Fx .

Proof (a) is a consequence of Lemma 3.5.14 since pro-K(X) = {Y ⊆

X : Y is closed}. It follows from Corollary 3.5.15 that Prime K(X) corresponds
to the closed irreducible subsets Y = ∅ of X under the bijection (a). Upon
identification of Y with its generic point we obtain (b) and (c).
Interpreting K̊(X) as K(X∗ ) and K(X) as K(Xcon ) yields results such as:
Proposition 3.5.18 Let X be a spectral space. There exists a bijection
X → Prime K̊(X), x → Fx = {A ∈ K̊(X) : x ∈ A},
that induces a bijection Xmin → Ultra K̊(X).
Proposition 3.5.19 Let X be a spectral space. There exists a bijection
Filt K(X) → {Y ⊆ X : Y is proconstructible and Y = ∅},

namely F → A∈F A, with inverse Y → FY = {A ∈ K(X) : Y ⊆ A}, which in
turn induces a bijection
X → Prime K(X) = Ultra K(X), x → Fx = {A ∈ K(X) : x ∈ A}.
This last theorem in particular is often used in the geometric setting:

Corollary 3.5.20 Let R be a real closed field, A an affine R-algebra and V (R) the
set of real points of the variety V = VA associated to A.
(a) (Ultrafilter Theorem, L. Bröcker, 1981) There exists a bijection

Sper A → Ultra S V (R) , x → Fx = {M : x ∈ M}

with inverse F → M∈F M.
(b) There exists a bijection from Filt S V (R) to the set of nonempty procon-

structible subsets of Sper A, namely F → M∈F M.
Let A be an affine algebra over a real closed field R with associated variety
V = VA . We illustrate the correspondence between the real spectrum and ultrafilters
in several examples below. For x ∈ Sper A, we consider the corresponding ultrafilter

Fx = {M ∈ S V (R) : x ∈ M}.
Given a semialgebraic subset M ⊆ V (R) we thus have
= {F : M ∈ F }
M
upon identification of Sper A with Ultra S V (R) .

Example 3.5.21 Let A = R[t], hence V (R) = R (cf. Example 3.3.14). The
correspondence Sper A → Ultra S(R) is as follows:
for x ∈ Sper A, Fx consists of all M such that

c
x=c∈R c∈M
c
x = c− (c ∈ R) ]c − ε, c[ ⊆ M for some ε > 0
c
x = c+ (c ∈ R) ]c, c + ε[ ⊆ M for some ε > 0
x = −∞ ]−∞, a[ ⊆ M for some a ∈ R
x = +∞ ]a, ∞[ ⊆ M for some a ∈ R
x = ξ (free cut) ∃a, b such that a < ξ < b and ]a, b[ ⊆ M
Example 3.5.22 If we view the elements of Sper A as ultrafilters of S V (R) , the

associated orderings of A are easy to describe. For x ∈ Sper A and f ∈ A we have
f (x) ≥ 0 ⇔ there exists M ∈ Fx such that f |M ≥ 0,

f (x) > 0 ⇔ there exists M ∈ Fx such that f |M > 0,
..
.
The support p = supp(x) of x is described as follows: the subvariety W := VA/p of

V associated to p is the smallest closed subvariety of V such that W (R) ∈ Fx ; in
other words,
!
W (R) = W (R),
W
where the intersection is over all closed subvarieties W of V such that W (R) ∈ Fx .
Example 3.5.23 Let x, y ∈ Sper A. Then
x % y (i.e., y ∈ {x}) ⇔ Fx ∩ S V (R) ⊆ Fy ⇔ Fy ∩ S̊ V (R) ⊆ Fx .
then x ∈ M
If M ⊆ V (R) is semialgebraic and x ∈ M, max if and only if for every
N ∈ S(M) with N ∈ Fx there exists an N ∈ Fx which is closed in M and such
that N ⊆ N.
Example 3.5.24 Let A = R[x, y]. Consider the homomorphisms
ϕ0 : A → R, f →
f (0, 0)
ϕ1 : A → R(x), f → f (x, 0)
ϕ2 : A → R(x, y), f → f.
Assume that the ordering of R(x, y) is such that x compared to R and y compared
to R(x) are infinitely small and positive, and assume that R(x) is endowed with the
induced ordering. Let αi ∈ Sper A be the point defined by ϕi (and the indicated
orderings) for i = 0, 1, 2 (cf. Remark 3.3.2(3)). We have
supp(α2 ) = (0) ⊆ supp(α1 ) = (y) ⊆ supp(α0 ) = (x, y),
and it is not difficult to see that α2 % α1 % α0 in Sper A. The associated ultrafilters

are described as follows. For semialgebraic M ⊆ R 2 we have
⇔ α0 = (0, 0) ∈ M;
α0 ∈ M
⇔ ∃ε > 0 such that ]0, ε[ × {0} ⊆ M;
α1 ∈ M
⇔ ∃ε > 0, ∃g ∈ R[t] with g(t) > 0 for t ∈ ]0, ε[ ,
α2 ∈ M
such that {(a, b) ∈ R 2 : 0 < a < ε, 0 < b < g(a)} ⊆ M.
(The verification only requires some effort in the case of α2 . Since α2 ∈ (Sper A)min ,
we have: α2 ∈ M ⇔ there exist f1 , . . . , fn ∈ A such that α2 ∈ H̊A (f1 , . . . , fn ) and

i {x ∈ R : fi (x) > 0} ⊆ M. Then one needs to consider that for every f ∈ A
2
with f (α2 ) > 0, the set {x ∈ R 2 : f (x) > 0} contains a set of the form described
above.)
We can picture α1 as the bud of the positive branch of the x-axis at the origin,
and α2 as the tip of the upper shoreline of the half-plane to the right of the x-axis.
Thus we see that α0 = (0, 0) has a chain of generalizations of (maximal possible)
length 2. It is a straightforward exercise (cf. Example 3.7.7) to show that every
point x ∈ R n has a chain of generalizations of length n in R #n . (In more generality
the analogous statement holds for smooth R-varieties.)
3.6 The Space of Closed Points 135
Example 3.5.25 Assume that A is an integral domain and let K = Quot A. The
orderings of K can be visualized in V (R) as follows.
We call a semialgebraic subset M ⊆ V (R) thin if M is not Zariski dense in
Spec A. (If we define the dimension of a semialgebraic subset as the dimension
of its Zariski closure in Spec A, then M is thin if and only if dim M < dim A =
tr.deg.(K/R) = dim V .)
Let x ∈ Sper A. Then supp(x) = 0 (i.e., x ∈ Sper K ⊆ Sper A) if and only if Fx
contains no thin sets. Expressed in a different way: there is an equivalence relation
on S V (R) defined by
M ∼ N :⇔ M ) N := (M \ N) ∪ (N \ M) is thin,
and the quotient set S V (R) / ∼ carries a lattice structure induced by S V (R) .
The ultrafilters of S V (R) / ∼ then correspond precisely to those ultrafilters of
S V (R) that do not contain any thin sets. This proves:
Proposition 3.5.26 (G.W. Brumfiel) There exists a canonical bijection
Sper K → Ultra S(V (R))/ ∼ .
Remark 3.5.27 We have shown that for every spectral space X there exists a natural
bijection X → Prime K̊(X). In this way it is possible to “recover” the set X
from the lattice K̊(X). In fact this relationship goes much further. Indeed, since
we can determine the topology of X from K̊(X), (the topological space) X and (the
distributive lattice) K̊(X) are essentially the same. We give a short sketch (without
proof) of this duality.
If L is a distributive lattice (with 0 and 1), then the set Prime L carries a natural
topology in which the sets Ux := {F ∈ Prime L : x ∈ F } (x ∈ L) form an open
basis. This topological space is called the Stone space St(L) of L. It is a spectral
space and L is (via x → Ux ) canonically isomorphic to K̊ St(L) . Conversely, if
X is a spectral space, then X is canonically homeomorphic to St K̊(X) (we did
describe the bijection). The following holds (see for instance [43, p. 103] or [56, p.
65 ff.]):
Stone Duality The assignments X → K̊(X) and L → St(L) determine a duality

(i.e., anti-equivalence) between the category of spectral spaces (and spectral maps)
and the category of distributive lattices.
3.6 The Space of Closed Points
This section deals with properties of real spectra that are quite false for general
spectral spaces. As usual, A denotes an arbitrary commutative unitary ring.
Proposition 3.6.1 If X is a quasi-compact T0 space and Xmax is its subspace of

closed points, then Xmax is quasi-compact, and for all x ∈ X we have {x} ∩ Xmax =
∅.
Proof It suffices to show that {x} ∩ Xmax = ∅ for x ∈ X, since this implies that X
is the only neighbourhood of Xmax in X. Thus, let x ∈ X and let Y := {x}. Then
Y is ordered via y ≤ y ⇔ y % y . If M is a totally ordered subset of Y , then

y∈M {y} = ∅ since Y is quasi-compact. By Zorn’s Lemma, Y contains a maximal
element.
Remarks 3.6.2
(1) If X is spectral, the above statement already follows from Proposition 3.5.17.
After passing to the inverse spectral space X∗ , we see that X = {x} (but
Xmin is in general not quasi-compact!). x∈X min
(2) By Corollary 3.4.21(c) the subspace Xmax of a spectral space X is procon-
structible if and only if it is closed. Usually this is not the case however. For
example, if R is a real closed field and A an affine R-algebra with associated
variety V , then we know from Sect. 3.5 that V (R) is already dense in Sper A,
hence a fortiori in (Sper A)max .
(3) Let R be a real closed field and X = Sper R[t]. Then Xmax consists of R ∪
{−∞, +∞} together with all free Dedekind cuts of R. In the special case that
R = R, the space Xmax = R ∪ {±∞} is the natural compactification (with two
end points) of R.
We have reached the point mentioned at the start of the chapter, where the real
spectra of rings exhibit special features that general spectral spaces do not. Despite
its simplicity, the consequences of the next lemma are far reaching!
Lemma 3.6.3 Let P and Q be orderings of A. The following statements are
equivalent:
(i) P ⊆ Q and Q ⊆ P (i.e., P and Q are incomparable with respect to
specialization);
(ii) there exists f ∈ A such that P ∈ H̊A (f ) and Q ∈ H̊A (−f );
(iii) there exist open U, V ⊆ Sper A such that U ∩ V = ∅ and P ∈ U, Q ∈ V .
Proof (i) ⇒ (ii): Choose a ∈ P \ Q and b ∈ Q \ P . Then f := a − b yields
the desired property. Indeed, arguing by contradiction, if −f ∈ P , then also b =
a − f ∈ P , and if f ∈ Q, then also a = b + f ∈ Q. Either conclusion is false.
The implications (ii) ⇒ (iii) ⇒ (i) are trivial.
Theorem 3.6.4 Let X be a proconstructible subspace of Sper A. Then Xmax is

Hausdorff, hence compact.
Proof This follows immediately from Lemma 3.6.3 and Proposition 3.6.1.
Remark 3.6.5 Theorem 3.6.4 is in general completely false for the Zariski spec-
trum. For example, if A is an affine C-algebra with associated variety V =
V (C), then (Spec A)max = V (C) carries the (induced) Zariski topology, which is
profoundly non-Hausdorff whenever dim A > 0.
Remark 3.6.6 Let R be a real closed field and A an affine R-algebra with associated
variety V = VA . It follows from the density of V (R) in (Sper A)max that
(Sper A)max is a natural compactification of V (R).
If x1 , . . . , xn is a system of generators of A and α ∈ (Sper A)max , then k(α)
is archimedean over R if and only if there exists c ∈ R such that x1 (α)2 + · · · +
xn (α)2 < c, as we will see later. (The condition is clearly equivalent to A/ supp(α)
being archimedean over R; we will see in Sect. 3.7 that k(α) is archimedean over
A/ supp(α) for all α ∈ (Sper A)max .)
We call those α ∈ (Sper A)max for which k(α)/R is archimedean the finite points
and all the remaining α ∈ (Sper A)max the infinitely distant points of (Sper A)max .
The set of finitepoints thus forms an open dense subspace of (Sper A)max (equal to
(Sper A)max ∩ c∈R H̊A (c − x12 − · · · − xn2 )). This is interesting in the particular
case R = R, since the set of finite points is already equal to V (R) in this case (cf.
Sect. 2.1). Thus we see that V (R) is contained as an open dense subspace in its
compactification (Sper A)max . This fact has been used by G.W. Brumfiel, who gave
an interpretation of the Thurston compactification of Teichmüller spaces by means
of the real spectrum, cf. [20].
A further and equally fundamental consequence is
Theorem 3.6.7 Let x ∈ Sper A. Then {x} forms a chain with respect to specializa-
tion. In other words, for all y, z ∈ Sper A we have
x % y and x % z ⇒ y % z or z % y.
Corollary 3.6.8 Let X ⊆ Sper A be proconstructible and let x ∈ X. Then {x} ∩

Xmax contains just one element.
Proof of Theorem 3.6.7 For any neighbourhoods U of y and V of z we have x ∈
U ∩ V . Conclude with Lemma 3.6.3.
Remark 3.6.9 For every x ∈ Sper A we can visualize {x} as a “spear”, with the
closed specialization of x as its tip. The notation Sper for real spectrum alludes to
this picture and also to the French term “spectre réel”.
Remark 3.6.10 A spectral space X is said to be completely normal if it has the

property of Theorem 3.6.7, i.e., if for any x, y, z ∈ X with y, z ∈ {x} one
has y ∈ {z} or z ∈ {y}. Real spectra of rings are completely normal spectral
spaces (Theorem 3.6.7). As mentioned before Definition 3.4.15, Hochster [51]
showed that every spectral space is homeomorphic to the Zariski spectrum of
some (commutative) ring. In view of this result it is natural to ask whether
every completely normal spectral space is homeomorphic to the real spectrum of

some ring. However, this turns out not to be true since Delzell and Madden [34]
constructed a counter-example (of combinatorial dimension one). See also Marshall
[80, Sect. 8.8]. As a comprehensive work on all aspects of spectral spaces we
recommend [35]. (Note that in op. cit. the term “spectral root system” is used instead
of “completely normal space”.)
Remark 3.6.11 As mentioned in Sect. 3.5 we will show in Sect. 3.7 that every point
x0 ∈ R n is the tip of a spear
xn % xn−1 % · · · % x1 % x0 (xi = xi−1 , i = 1, . . . , n)
#n = Sper R[t1 , . . . , tn ]. Conversely, any point α ∈ Sper R[t1 , . . . , tn ]

of length n in R
with supp(α) = (0) has at most n distinct specializations. This follows easily from
Theorem 3.6.7 and some results from commutative algebra. (One considers the
support ideals of the specializations; they form a chain.) The spear {α} can indeed be
shorter, for two reasons in fact. First of all, the closed specialization of α need not
be in R n (e.g., α = ±∞ in R). Secondly, the spear can have “gaps”. We illustrate
this for R in the following example.
2
Example 3.6.12 Let
P := {f ∈ R[x, y] : ∃ ε > 0 such that f (t, et − 1) ≥ 0 for 0 < t < ε}.
Then P is an ordering of R[x, y] with supp(P ) = P ∩ (−P ) = (0), and it is clear

that (0, 0) ∈ R2 is the closed specialization of P in Sper R[x, y]. We should now
convince ourselves that R(x, y) contains just one nontrivial convex subring (with
respect to the ordering induced by P ), namely the convex hull o R(x, y)/R of
R. This implies (as we will show in Sect. 3.7) that P actually has just one proper
specialization. Heuristically this is pretty clear: were P to have a one-dimensional
specialization, its support could only be the curve y + 1 = ex , which is not an
algebraic curve.
Proposition 3.6.13 (“Normality” of Sper A) If Y and Z are closed disjoint subsets

of Sper A, there exist open quasi-compact neighbourhoods U of Y and V of Z such
that U ∩ V = ∅.
Proof If X ⊆ Sper A is closed and y ∈ Sper A is such that {y} ∩ X = ∅, then
every x ∈ X has an open quasi-compact neighbourhood Ux such that {y}∩ Ux = ∅.
Since X is quasi-compact, X is covered by finitely many Ux , and so there exists
an open quasi-compact U ⊆ Sper A such that X ⊆ U and {y} ∩ U = ∅. Assume

now that {Ui : i ∈ I }, resp. {Vj : j ∈ J }, is thesystem of open quasi-compact
neighbourhoods of Y , resp. Z. Then by the above, Ui = {x ∈ Sper A : {x} ∩ Y =
i
∅} and similarly for Vj . It follows that
! !
Ui ∩ Vj = ∅.
i∈I j ∈J
Indeed, if the intersection were to contain an element x with closed specialization

x, then we would have x ∈ Y ∩ Z, a contradiction.
By Corollary 3.4.10 there exist
finite I ⊆ I and J ⊆ J such that for U := Ui and V := Vj we already
i∈I j ∈J
have U ∩ V = ∅.
Remark 3.6.14 The proof shows that a spectral

space X satisfies the normality
property from Proposition 3.6.13 only if {x} ∩ Xmax = 1 for all x ∈ X. The
converse is also true, cf. [23]. In the same way Propositions 3.6.16 and 3.6.17
remain valid for these so-called normal spectral spaces. Note that the property from
Theorem 3.6.7 (which is always satisfied for Sper A) is even more specific.
Definition 3.6.15 Let X ⊆ Sper A be proconstructible. The canonical retraction

from X to Xmax is the map ρ = ρX : X → Xmax that is uniquely determined by
ρ(x) ∈ {x} ∩ Xmax for all x ∈ X. Note that ρ 2 = ρ.
Proposition 3.6.16 Let X ⊆ Sper A be proconstructible. The canonical retraction

ρ : X → Xmax is a continuous and closed map.
Proof Let Y ⊆ X be closed. Then ρ(Y ) = Y ∩ Xmax is closed in Xmax . To prove
the continuity, let Y ⊆ Xmax be closed in Xmax , say Y = Y ∩ Xmax for a closed
Y ⊆ X. Then
!
ρ −1 (Y ) = {x ∈ X : {x} ∩ Y = ∅} = Ui
i∈I
is the intersection of all quasi-compact open neighbourhoods of Y (see the proof of

Proposition 3.6.13). Thus ρ −1 (Y ) is in particular proconstructible and hence closed
since ρ −1 (Y ) is stable under specialization.
Proposition 3.6.17 Let X ⊆ Sper A be proconstructible. Then Y → ρ(Y ) is a

bijection from the set of connected components Y of X to the set of connected
components Y of Xmax . The inverse map is given by Y → ρ −1 (Y ).
Proof It suffices to show that every closed Y ⊆ X for which ρ(Y ) is connected is
itself connected. Assume that Y = Y1 ∪Y2 with Y1 , Y2 closed and Y1 ∩Y2 = ∅. Then
also ρ(Y1 ) ∩ ρ(Y2 ) = ∅ and ρ(Y1 ), ρ(Y2 ) are closed in Xmax by Proposition 3.6.16,
from which the assertion follows.
We recognize now that the spaces Xmax (with X ⊆ Sper A proconstructible)
are topologically pleasing, not only because they are compact, but also because they
retain some information about the space X. For example, one can use ρ to reduce the
cohomology of X to the cohomology of Xmax ; both have isomorphic cohomology,
cf. [23]. In the geometric setting it seems important to treat M max as the natural
compactification of M ⊆ V (R). In spite of everything however, Sper A should be
considered as the central object, and (Sper A)max rather as a useful auxiliary space.
3.7 Specializations and Convex Ideals
As usual, we assume that A is an arbitrary (commutative, unitary) ring. Let P ⊆ A

be an ordering of A and p := supp(P ). We denote the natural homomorphism
A → A/p by π and write P for the ordering of κ(p) induced by P . When studying
the ideals of A/p that are convex with respect to P , the following concept presents
itself:
Definition 3.7.1 An ideal a of A is called P -convex if for all f, g ∈ P ,
f + g ∈ a ⇒ f, g ∈ a.
Lemma 3.7.2 The map a → π(a) defines a bijection from the set of P -convex
ideals of A to the set of ideals of A/p that are convex with respect to P . The inverse
is given by a → π −1 (a).
Proof We endow A/p with the total order relation induced by P . Since we have
f ∈ P and −f ∈ P for every f ∈ p, it follows that p is contained in every P -
convex ideal of A. This shows that the two described maps are inverse to each other.
If a ⊆ A is a P -convex ideal, then π(a) is convex in A/p. Indeed, if a ∈ a and
f ∈ A such that 0 ≤ π(f ) ≤ π(a), then f ∈ P and a − f ∈ P , hence f ∈ a by
assumption. Conversely, if a ⊆ A/p is a convex ideal and if f, g ∈ P are such that
f + g ∈ a := π −1 (a), then 0 ≤ π(f ) ≤ π(f + g) ∈ a, and thus also π(f ) ∈ a.
Corollary 3.7.3
(a) The P -convex ideals form a chain, and p = supp(P ) is the smallest ideal in it.
(b) For every P -convex ideal a of A we have P + a = P ∪ a.
Proof (a) See Remark 2.1.4(4).
(b) Let p ∈ P and a ∈ a. If p+a ∈ P , then −(p+a) ∈ P , and −a = p−(p+a) ∈
a implies p + a ∈ a.
3.7 Specializations and Convex Ideals 141
Theorem 3.7.4 Let P ∈ Sper A be an ordering of A and p = supp(P ). Then the

following sets are in canonical bijection:
(1) {P } (the set of specializations Q of P in Sper A);
(2) the set of P -convex prime ideals q of A;
(3) the set of prime ideals q of A/p that are convex with respect to P .
The bijection (1) → (2) is given by Q → supp(Q), its inverse by q → P + q =
P ∪ q. The bijection (2) → (3) is given by q → π(q), its inverse by q → π −1 (q).
Proof The bijection between (2) and (3) was established in Lemma 3.7.2. Let Q ⊇
P be an ordering of A and q := Q ∩ (−Q). Let f, g ∈ P such that f + g ∈ q. Then
f ∈ q since −f = g − (f + g) ∈ Q. Hence q is P -convex and Q = P ∪ q (to see
⊆: if f ∈ Q \ q, then −f ∈ Q and so −f ∈ P which yields f ∈ P ). Conversely,
let q be a P -convex prime ideal. Then Q := P + q = P ∪ q is an ordering of A with
support q since Q ∩ (−Q) = (P ∩ −P ) ∪ (P ∩ q) ∪ (q ∩ −P ) ∪ q = q is a prime
ideal.
Once more we see that {P } forms a chain.
Example 3.7.5 Let λ : K → L∪∞ be a surjective place with valuation ring B = oλ .
Upon identifying Sper K, resp. Sper L, with those points of Sper B whose support
is (0), resp. mB , the following holds for all x ∈ Sper K: λ is compatible with x if
and only if x has a specialization y with support mB ; when this happens, y is the
ordering of L ∼= B/mB induced by x in the sense of Sect. 2.2. As a consequence we
can formulate the Baer–Krull Theorem (cf. Sect. 2.7) as follows:
Let be the value group of λ and let y ∈ Sper B with supp(y) = mB . Then
acts transitively and free on the (nonempty) set of generalizations of y
/2
with support (0).
Thus, if ∼= Znlex and if (0) = p0 ⊂ · · · ⊂ pn = mB are the prime ideals of B,
then every y ∈ Sper B with supp(y) = mB has precisely 2n−i generalizations with
support pi for i = 0, . . . , n.
Definition 3.7.6 Let B be a valuation ring and A a subring of B. The prime ideal
A ∩ mB of A is called the centre of B on A. Similarly, given a place λ : K → L ∪ ∞
and a subring A of K on which λ is finite, we call the kernel of the homomorphism
λ|A : A → L the centre of λ on A.
Example 3.7.7 Let λ : K → L ∪ ∞ be a place and A a subring of K on which

λ
λ is finite and has centre p. The homomorphisms A ⊆ oλ − → L induce maps
Sper L → Sper oλ → Sper A whose composition we denote by λ∗ . Then it follows
from Example 3.7.5 that for every z ∈ Sper L the element λ∗ (z) ∈ Sper A (with
support p) has a generalization with support (0) in Sper A.
In more generality, let λ = λn ◦ · · · ◦ λ1 : K → L ∪ ∞ be a composition of places
which is finite on A ⊆ K and has centre p, and let pi be the centre of λi ◦ · · · ◦ λ1
on A for i = 0, . . . , n. Then (0) = p0 ⊆ p1 ⊆ · · · ⊆ pn = p, and for every z ∈
Sper L, λ∗ (z) has a chain of generalizations with the pi as supports, i.e., there exist
y0 , . . . , yn ∈ Sper A with supp(yi ) = pi such that y0 % y1 % · · · % yn = λ∗ (z).
This shows by way of example that for a real closed field R every point a ∈ R n
has chains of generalizations of length n in Sper R[t1 , . . . , tn ]: namely, in Sect. 2.10
we constructed a place λ = λn ◦ · · · ◦ λ1 : R(t1 , . . . , tn ) → R ∪ ∞ over R for
which λi ◦ · · · ◦ λ1 has centre (tn − an , . . . , tn−i+1 − an−i+1 ) on R[t1 , . . . , tn ] for
i = 0, . . . , n.
Definition 3.7.8 A valuation ring B is called real closed if Quot B is a real closed
field and B is residually real. (The second condition is equivalent to: B is convex in
Quot B, cf. Sect. 2.5.)
Proposition 3.7.9 If B is a real closed valuation ring, then supp : Sper B →

Spec B is a homeomorphism, and κ(p) is real closed for all p ∈ Spec B (thus,
k(α) = κ(supp α) for all α ∈ Sper B).
Proof Let p ∈ Spec B. Then B, hence also Bp , is convex in R := Quot B and
it follows that κ(p) = Bp /pBp = κ(Bp ) is real closed by Theorem 2.5.1. The
bijectivity of supp follows. The map supp is also open: Let f ∈ B. If f < 0 (in R),
then H̊B (f ) = ∅. Otherwise there exists g ∈ B such that f = g 2 and it follows
that H̊B (f ) = {α ∈ Sper B : f (α) = 0}, hence supp H̊B (f ) = DB (f ).
Now consider a real closed field R and a subring A ⊆ R. We denote the convex
hull of a subset M ⊆ R in R by M c . In particular, let B := Ac denote the convex
hull of A. Let P0 := {a ∈ A : a ≥ 0} ∈ Sper A. The P0 -convex ideals of A are
simply the convex ideals of A (with respect to the total order relation induced by
R).
Let a ⊆ A be an ideal. Then ac is an ideal of B and a is P0 -convex if and only if
a = A ∩ ac . Thus b → A ∩ b defines a surjection
{ideals of B} −− {P0 -convex ideals of A}.
If p ⊆ A is a P0 -convex prime ideal, then pc is not necessarily prime. Nevertheless

the following holds:
Lemma 3.7.10 The map
Spec B → {P0 -convex prime ideals of A}, q → A ∩ q
is surjective.
√
Proof If p is a P0 -convex prime ideal √
of A, then pc is a prime ideal of B by
√
Proposition 2.4.17(e), and we have A ∩ pc = p = p.
3.7 Specializations and Convex Ideals 143
Corollary 3.7.11 If C is a convex overring of A in R, then A ∩ mC is a P0 -convex

prime ideal of A and the image of the map Spec C → Spec A consists precisely of
those P0 -convex prime ideals p of A that satisfy p ⊆ A ∩ mC .
Proof This is clear since C = BmC , cf. Sect. 2.2.
Returning to an arbitrary ring A, Theorem 3.7.4 and Proposition 3.7.9 allow us
to reformulate the above facts:
Proposition 3.7.12 Let ϕ : A → R be a homomorphism into a real closed field, B
the convex hull of ϕ(A) in R and ϕ0 : A → B the induced homomorphism. Then the
image of Sper ϕ0 consists precisely of the specializations of αϕ in Sper A, where αϕ
denotes the element in Sper A defined by ϕ (cf. Remark 3.3.2(3)).
Proof Let β0 be the generic point of Sper B. Then αϕ = (Sper ϕ0 )(β0 ) and so
Im(Sper ϕ0 ) ⊆ {αϕ }. Let P = ϕ −1 [0, ∞[R be the ordering of A corresponding
to αϕ . Given any P -convex prime ideal q of A there exists β ∈ Sper B such that
supp (Sper ϕ0 )(β) = q by Lemma 3.7.10 and Proposition 3.7.9. The statement
then follows by Theorem 3.7.4.
Proposition 3.7.13 Let ϕ : A → B be a homomorphism into a real closed

valuation ring B, let β0 be the generic point and β1 the closed point of Sper B,
and let αi = (Sper ϕ)(βi ) for i = 0, 1. Then the image of Sper ϕ is precisely the
“interval” {α ∈ Sper A : α0 % α % α1 }.
Proof Similar to the proof of Proposition 3.7.12, using Corollary 3.7.11.
Let A and B be arbitrary commutative rings. If ϕ : A → B is an integral
homomorphism (i.e., B is integral over ϕ(A)), it is well-known that Spec ϕ is a
closed map. (This is a reformulation of the so-called “Going Up” Theorem of
Cohen–Seidenberg which follows immediately from Theorem 2.3.5, cf. [13, Ch. V,
§2, no. 1] or [68, p. 49].) As an application of the previous results we show that the
same holds for Sper ϕ:
Proposition 3.7.14 If ϕ : A → B is an integral ring homomorphism, then Sper ϕ
is a closed map.
Proof Let ϕ ∗ := Sper ϕ. If Y ⊆ Sper B is closed, then ϕ ∗ (Y ) is proconstructible
and it suffices to show stability under specialization. Thus, let β ∈ Sper B and
α = ϕ ∗ (β). We must show that ϕ ∗ {β} = {α}. Let rβ : B → k(β), b → b(β) denote
the evaluation map associated to β and C the convex hull of rβ ϕ(A) in k(β). Since
B is integral over A, hence also rβ (B) integral over rβ ϕ(A), we have rβ (B) ⊆ C.
We define ψ : A → C and r : B → C via the commutative diagram
Then r ∗ (Sper C) = {β} and ψ ∗ (Sper C) = {α} by Proposition 3.7.12, from which
ϕ ∗ {β} = {α} follows.
We want to make the statements of Lemma 3.7.10 and Proposition 3.7.12 a bit
more precise. Lemma 3.7.10 can also be formulated as follows: for every P0 -convex
prime ideal p of A there exists a convex overring C of A with centre p on A. More
accurately:
Proposition 3.7.15 Let R be a real closed field, A ⊆ R a subring and P = A ∩
[0, ∞[R . If p ⊆√A is a P -convex prime ideal, then C := (Ap )c is the smallest and
Cc , where c := pc , the largest convex overring of A in R with centre p on A.
Proof Let B ⊆ R be a convex overring of A with centre p on A. Then Ap ⊆ B, thus
also C = (Ap )c ⊆ B, and B = Cq for some √ prime ideal q of C. Since p ⊆ mB = q
it must also be the case that pc ⊆ q and c = pc ⊆ q, i.e., B ⊆ Cc (note that c is a
prime ideal of C, cf. Lemma 3.7.10).
Corollary 3.7.16 Let A be an arbitrary commutative ring. Given α, β ∈ Sper A

with α % β, there exists a homomorphism ϕ : A → B into a real closed valuation
ring B such that the image of Sper ϕ is precisely the“interval”
{γ ∈ Sper A : α % γ % β}.
Proof Let q := supp β, A := rα (A) and q := rα (q). It suffices to take for B the
convex hull of Aq in k(α) (and for ϕ the map induced by rα : A → k(α)).
Corollary 3.7.17 (“Convexity” of Sper ϕ) Let ϕ : A → A be a homomorphism

of commutative rings and let α, β ∈ Sper A such that α % β. Then the image of
the interval {γ ∈ Sper A : α % γ % β} in Sper A is again an interval, i.e., given
γ ∈ Sper A such that
(Sper ϕ)(α) % γ % (Sper ϕ)(β),
there exists a preimage γ of γ such that α % γ % β.

Proof This follows immediately from Corollary 3.7.16 and Proposition 3.7.13.
Corollary 3.7.18 Let α ∈ Sper A. Then α is closed in Sper A if and only if k(α) is
archimedean over rα (A).
Proof We may assume without loss of generality that supp α = (0), i.e., that
R := k(α) is the real closure of K := Quot A. Since R is archimedean over K
(cf. Proposition 1.7.1), we have C = R for p = (0) in Proposition 3.7.15, i.e., every
convex overring B of A which is distinct from R induces a proper specialization of
α. Hence, {α} = {α} if and only if there are no such overrings.
3.8 The Real Spectrum and the Reduced Witt Ring of a Field 145
Since convex valuation rings of a real closed field R and real places R → S ∪ ∞
are essentially the same thing, the previous statements can also be formulated in the
language of places. Let us give an example.
Let R and S be real closed fields, ϕ : A → R a homomorphism and λ : R →
S ∪ ∞ a place which is finite on ϕ(A). Then the element αλ◦ϕ is a specialization
of αϕ in Sper A, and for every specialization of αϕ there is such a place λ. More
precisely:
Definition 3.7.19 A homomorphism ϕ : A → R into a real closed field is called
tight if R is archimedean over Quot ϕ(A).
Proposition 3.7.20 Let ϕ : A → R be a homomorphism into a real closed field. For

every specialization β of αϕ there exists a surjective place λ = λβ : R → S ∪∞ to a
real closed field S which is finite on ϕ(A) and such that λ ◦ ϕ is tight and αλ◦ϕ = β.
If γ is another specialization of αϕ , then: β % γ ⇔ λγ factors through λβ :
Proof This is just a reformulation of the results already obtained.
3.8 The Real Spectrum and the Reduced Witt Ring of a Field
In the previous section we made a detailed study of the specialization chains (the
“spears”) in the real spectrum of a ring. Visually speaking these are transversal to
the fibres of the support map supp : Sper A → Spec A since every spear has at most
one point in common with such a fibre. The fibres are precisely the real spectra
of the residue fields of A (cf. Corollary 3.3.12). Thus, in order to gain a better
understanding of the spectral space Sper A, a deeper study of the real spectrum of
fields suggests itself. In this section we will initiate this study.
Throughout, F denotes a field, without loss of generality assumed to be a real.
For the sake of brevity we write XF instead of Sper F . The spectral space XF is
Hausdorff, thus compact and totally disconnected, its (Harrison) topology coincides
with the constructible topology, and the constructible subsets of XF are precisely the
clopen subsets (cf. Remark 3.3.2(4) and Sect. 3.4). If a1 , . . . , an are elements of F ∗ ,
we simply write HF (a1 , . . . , an ) or H (a1, . . . , an ) instead of H̊F (a1 , . . . , an ) =
H F (a1 , . . . , an ).
Thus XF = Sper F is not particularly interesting as a topological space.

However, a stronger structure is provided by the field elements, or rather by the sign
distributions on XF that they determine. In order to utilize this structure we return
to the theory of quadratic forms, which has shown itself to be an invaluable tool
for real algebra and geometry. One should then also consider quadratic forms over
more general (commutative) rings. In this section we will just cover a number of
elementary observations about the interplay of quadratic forms and the real spectrum
of a field.
We remind ourselves that every ordering P of F defines a ring homomorphism
signP : W (F ) → Z, the signature at P , (cf. Sect. 1.2).
Definition 3.8.1 For ϕ ∈ W (F ) the map
sign ϕ : XF → Z, P → signP (ϕ)
is called the total signature of ϕ.

We denote the ring of continuous (= locally constant) functions from XF to Z by
C(XF , Z).
Proposition 3.8.2 Consider any ϕ ∈ W (F ). The total signature of ϕ is a
continuous function from XF to Z. The map
sign : W (F ) → C(XF , Z), ϕ → sign ϕ
(the total signature of F ) is a ring homomorphism with kernel equal to the nilradical
of W (F ).
Proof Since all signP are homomorphisms, the map ϕ → sign ϕ is a homomor-
phism from W (F ) to the ring of all maps XF → Z. Given ϕ = a with a ∈ F ∗ , the
continuity of sign ϕ is clear. It follows that sign ϕ is continuous for all ϕ ∈ W (F ).
The kernel of sign is the intersection of the kernels of all the signP , and these are
precisely the minimal prime ideals of W (F ) (cf. Sect. 1.4). The statement follows.
Definition 3.8.3 We call
W (F ) := W (F )red = W (F )/ Nil W (F )
the reduced Witt ring of F and denote the image of the fundamental ideal I (F ) in
W (F ) by I (F ) = I (F )/ Nil W (F ).
Given ϕ ∈ W (F ), we denote its image in W (F ) by ϕ. By Proposition 3.8.2 we
may identify W (F ) canonically with a subring of C(XF , Z), and usually this is what
we will do. Under this identification we have ϕ = sign ϕ.
We denote the characteristic function of a subset Y ⊆ XF by χY . Given a 2-
dimensional form ϕ = 1, a with a ∈ F ∗ , we then have ϕ = 2 · χH (a) . Since I (F )
is generated as an additive group by the forms 1, a (note: 1, a ⊥ −1, b ∼

a, b), it follows that in C(XF , Z), I (F ) is generated as an additive group by the
functions 2 · χH (a) with a ∈ F ∗ . We observe that W (F ) = Z + I (F ) and so
W (F ) ⊆ Z + 2 · C(XF , Z).
Definition 3.8.4 An n-fold Pfister form over F is a quadratic form that can be
written as 1, a1 ⊗ · · · ⊗ 1, an with ai ∈ F ∗ .
Let ϕ = 1, a1 ⊗ · · · ⊗ 1, an . Then clearly, ϕ = 2n · χH (a1 ,...,an ) . Since I (F ) is
additively generated by the 1-fold Pfister forms, the n-th power I n (F ) := I (F )n
n
is additively generated by the n-fold Pfister forms. Hence, I (F ) := I (F )n is
generated by the functions 2n · χH (a1 ,...,an ) with a1 , . . . , an ∈ F ∗ .
Lemma 3.8.5 Let Y be a constructible subset of XF . Then there exists a finite
sequence a1 , . . . , an of elements in F ∗ such that Y is the (disjoint) union of sets
of the form H (ε1a1 , . . . , εn an ) with εi ∈ {±1}.
Proof There exist finite subsets S1 , . . . , Sr of F ∗ such that Y = H (S1 ) ∪ · · · ∪
H (Sr ). Let {a1 , . . . , an } := S1 ∪ · · · ∪ Sr . Since the elements of Si have constant
sign on each set H (ε1 a1 , . . . , εn an ), every H (Si ) (and thus also Y ) is a union of
such sets.
Proposition 3.8.6 The abelian group C(XF , Z)/W (F ), i.e., the cokernel of the
total signature of F , is a 2-primary torsion group.
Proof Every element of C(XF , Z) can be written in the form m1 χY1 + · · · + mr χYr
with r ≥ 1, mi ∈ Z and Yi ⊆ XF constructible. It follows from Lemma 3.8.5 that
C(XF , Z) is generated as an additive group by the f = χH (a1 ,...,an ) with n ≥ 1 and
ai ∈ F ∗ . For such an f we have 2n f ∈ W (F ). The statement follows.
Definition 3.8.7 The smallest integer s ≥ 0 such that 2s · C(XF , Z) ⊆ W (F ) (or

s = ∞ if there is no such integer) is called the stability index of F and is denoted
st(F ).
In real algebra and geometry the stability index has turned out to be a significant
invariant of real fields. In many important cases st(F ) is finite and often lends itself
to computation.
Since W (F ) ⊆ Z + 2 · C(XF , Z), the condition st(F ) = 0 is clearly equivalent
to the existence of a (unique) ordering of F . We continue with a characterization of
those fields for which st(F ) ≤ 1.
Lemma 3.8.8 If Y ⊆ XF is constructible and if 2 · χY ∈ W (F ), then there exists
an a ∈ F ∗ such that Y = H (a).
Proof Let ϕ = a1 , . . . , a2n be a form such that ϕ = 2 · χY (note that ϕ then
necessarily has even dimension). Then a := a1 · · · a2n has sign (−1)n−1 on Y and
sign (−1)n on XF \ Y . Hence Y = H (a) or Y = H (−a).
Definition 3.8.9 A real field F is called a SAP field if, given any two disjoint closed
subsets Y and Y of XF , there exists an a ∈ F ∗ such that a|Y > 0 and a|Y < 0.
(SAP stands for “strong approximation property”, which is unrelated to the number
theoretic concept of strong approximation.)
Proposition 3.8.10 Let F be a real field. The following statements are

equivalent:
(i) st(F ) ≤ 1;
(ii) every constructible subset of XF is of the form H (a) for some a ∈ F ∗ ;
(iii) W (F ) = Z + 2 · C(X, Z);
(iv) F is a SAP field.
Proof (i) ⇒ (ii): If Y ⊆ XF is constructible, then 2 · χY ∈ W (F ) by (i) and so
Y = H (a) for some a ∈ F ∗ by Lemma 3.8.8.
(ii) ⇒ (iii): For every constructible Y ⊆ XF we have 2 · χY ∈ W (F ). It follows
that 2 · C(XF , Z) ⊆ W (F ).
(iii) ⇒ (i): This is trivial.
(ii) ⇔ (iv): Let Y, Y ⊆ XF be closed and disjoint. Then there exists a
constructible neighbourhood Z of Y such that Z ∩ Y = ∅. Every a ∈ F ∗ such
that Z = H (a) separates Y and Y . Conversely, (ii) follows for a constructible
Y ⊆ XF upon applying the SAP property to the sets Y and Y := XF \ Y .
To convey a better feeling for the stability index we state the following two
theorems without proof:
Theorem 3.8.11 (Special case of [15, Satz 4.8]) Let R be a real closed field and
K a real n-dimensional function field over R. Then st(K) = n.
Theorem 3.8.12 (Special case of [9, Satz 15]) Let F be a real field and n ∈ N
such that n ≥ st(F ). For every (n + 1)-fold Pfister form ϕ over F there exists an
(n-fold) Pfister form ψ over F such that ϕ = 2ψ.
It is not difficult to deduce new characterizations of st(F ) from Theorem 3.8.12.
Since we do not prove the theorem, nor have any use for its corollary below in the
remainder of this book, we leave the proof of the corollary as an exercise to the
reader.
Corollary 3.8.13 Let F be a real field and s ∈ N. The following statements are
equivalent:
(i) st(F ) ≤ s;
(ii) every set H (a1, . . . , aN ) with ai ∈ F ∗ and N ≥ 1 can be written in the form
H (b1, . . . , bs ) with bj ∈ F ∗ ;
(iii) for every (s + 1)-fold Pfister form ϕ over F there exists an (s-fold) Pfister form
ψ over F such that ϕ = 2ψ;
s+1 s
(iv) I (F ) = 2 I (F ).
The term stability index for st(F ) comes from characterization (iv): st(F ) is the
n
smallest index from which point onwards the sequence I (F ) : n = 1, 2, . . . is
stable in the sense of (iv). (This is a K-theoretic notion of stability, cf. [83].)
In the final part of this section we turn our attention to finitely generated field
extensions. If F ⊆ K is a field extension we write rK/F for the restriction map from
XK := Sper K to XF = Sper F . We consider finite extensions first:
Proposition 3.8.14 Let K ⊇ F be a field extension of degree n < ∞. Then:
(a) r := rK/F : XK → XF is an open (and closed) map, i.e., the images of
constructible sets are constructible.
(b) For each x ∈ XF , the fibre r −1 (x) is finite and its cardinality t satisfies t ≤ n
and t ≡ n mod 2.
(c) Let Ut := {x ∈ XF : #r −1 (x) = t} for t ≥ 0. Then every Ut is constructible
and r is topologically trivial over every Ut (i.e., there exists a decomposition
of r −1 (Ut ) into t disjoint open subsets W1 , . . . , Wt such that r|Wi is a
homeomorphism from Wi to Ut for every i = 1, . . . , t).
(d) The map f : XF → Z, x → #r −1 (x) is an element of W (F ).
Proof Consider the trace form ϕ := tr∗ 1K associated to the extension K/F .
Then f = ϕ by Corollary 1.12.10, which proves (d) as well as the constructibility
of the Ut . Furthermore, (b) follows immediately since dim ϕ = n.
Let us prove (a) next: r is closed since it is a continuous map between compact
spaces. In addition, r(XK ) = U1 ∪· · ·∪Un is constructible in XF . Let a1 , . . . , am ∈
√ √
K ∗ . Then HK (a1 , . . . , am ) = rL/K (XL ) for L := K( a1 , . . . , am ) (cf.
Proposition 1.3.3). Consequently, rK/F HK (a1 , . . . , am ) = rK/F rL/K (XL ) =
rL/F (XL ) is also constructible in XF . Since these HK (a1 , . . . , am ) form a basis of
the topology of XK , (a) follows.
It remains to prove the second statement of (c). Let t ≥ 0 and fix an x ∈ Ut . If
r −1 (x) = {y1 , . . . , yt }, there exist pairwise disjoint constructible neighbourhoods
Wi of the yi in r −1 (Ut ) for i = 1, . . . , t. Then V (x) := r(W1 ) ∩ · · · ∩ r(Wt ) is a
constructible neighbourhood of x in Ut and the restriction of r to every Wi (x) :=
Wi ∩ r −1 V (x) is a homeomorphism to V (x). Then we cover Ut with finitely
many V (xi ) (i = 1, . . . , N) and make this covering disjoint by shrinking the V (xi )
if necessary. For every 1 ≤ i ≤ N we have r −1 V (xi ) = tj =1 Wj (xi ) (disjoint
union), and r|Wj (xi ) is a homeomorphism from Wj (xi ) to V (xi ). Finally, letting

Wj := N i=1 Wj (xi ) (j = 1, . . . , t), the sets W1 , . . . , Wt yield the desired result.
With help from the sign change criterion (Theorem 2.12.8), we can generalize
statement (a) from Proposition 3.8.14 to finitely generated field extensions:
Proposition 3.8.15 (R. Elman, T.Y. Lam, A. Wadsworth [38]) Let K ⊇ F be a
finitely generated field extension. Then rK/F : XK → XF is an open (and closed)
map.
Proof The argument used in the proof of Proposition 3.8.14(a) shows that it suffices
to prove that rK/F (XK ) is open in XF . Let α1 , . . . , αd be a transcendence basis of
K over F , and let β ∈ K be such that K = F (α1 , . . . , αd , β) (Primitive Element
Theorem). If p ∈ F [t1 , . . . , td , td+1 ] = F [t] is an irreducible polynomial such that
p(α1 , . . . , αd , β) = 0, then the field of fractions of F [t]/(p) is isomorphic over F
to K (cf. Proposition 2.12.9).
Consider x ∈ rK/F (XK ) and let k(x) be the associated real closure of K.
By the sign change criterion (Theorem 2.12.8) there exist a, b ∈ k(x)d+1 such
that p(a) < 0 < p(b). Let L = F (a, b) be the finite extension of F that is
generated over F by the coordinates of a and b in k(x). Let x be the restriction
of the ordering of k(x) to L. Then x ∈ HL (−p(a), p(b)). By Proposition 3.8.14,
Y := rL/F HL (−p(a), p(b)) is a constructible neighbourhood of x in XF . The
sign change criterion then yields Y ⊆ rK/F (XK ). Indeed, if y ∈ HL (−p(a), p(b))
and y = rL/F (y ) ∈ Y , then k(y) = k(y ) is an overfield of L, and p is indefinite
over k(y) by the choice of y . This implies that y has an extension to K, as required.
Remark 3.8.16 Coste and Roy [28] established a far-reaching generalization of

Proposition 3.8.14(a) by means of model theoretic arguments (quantor elimination
in the theory of real closed fields): if ϕ : A → B is a ring homomorphism of finite
presentation (i.e., B is finitely generated as an A-algebra and Ker ϕ is a finitely
generated ideal of A), then Sper ϕ : Sper B → Sper A is an open map with respect
to the constructible topology. Using this result one can also obtain another proof of
Proposition 3.8.15.
3.9 Preorderings of Rings and Positivstellensätze
Let A be ring. The purpose of this section is to investigate the following question:
let X ⊆ Sper A be an intersection of sets of the form {f > 0}, {f ≥ 0} and {f = 0}
(where f ∈ A), and explicitly represented in this manner. Determine those functions
in A that are positive (resp. are nonnegative, resp. vanish) on X.
Definition 3.9.1 A preordering of A is a subset T ⊆ A that satisfies the following
properties:
(RP1) T + T ⊂ T , T T ⊂ T (i.e., T is a subsemiring of A);
(RP2) a 2 ∈ T for all a ∈ A.
If in addition
(RP3) −1 ∈ T
holds, then T is called a proper preordering of A. Otherwise T is an improper
preordering of A.
3.9 Preorderings of Rings and Positivstellensätze 151
Examples and Remarks 3.9.2

(1) Preorderings of fields in the sense of Definition 1.1.4 are now called proper
preorderings. This slight inconsistency should not lead to confusion.
(2) Preimages of (proper) preorderings under homomorphisms are again (proper)
preorderings; images of preorderings under surjective homomorphisms are
again preorderings.
(3) Every ordering of A is a proper preordering of A. Arbitrary intersections and
upward directed unions of preorderings are again preorderings. The smallest
preordering of A is A2 . Thus A has a proper preordering if and only if A is
real.
(4) If 1/2 ∈ A, then T = A is the only improper preordering of A (since 4a =
(a + 1)2 − (a − 1)2 ).
(5) For a subset F ⊆ A we denote the preordering generated by F (in A) by P [F ].
Then P [F ] consists of all finite sums of elements of the form a ·f1 · · · fn , where
a ∈ A2 , n ≥ 0 and f1 , . . . , fn ∈ F (without loss of generality assumed to be
pairwise distinct). In particular we observe that
H A (F ) = H A (P [F ])
(indeed, the inclusion ⊇ is trivial and the inclusion ⊆ follows from the
description).
(6) For fields k with char k = 2 (in particular real fields), k 2 is the intersection of
all orderings of k (cf. Sect. 1.1). For rings this is false in general. For example,
A = R[t1 , . . . , tn ] is a real reduced ring and if n ≥ 2, then A2 is not
equal to the intersection of all orderings. Indeed, letting K = Quot A, we
can identify (Sper A)min with Sper K (we did not fully prove this, but compare
with Example 3.7.7) and the intersection of all orderings of A is thus equal to
A ∩ K 2 , the positive semidefinite polynomials. But these are not always sums
of squares in A when n ≥ 2 (cf. Sect. 2.12).
As for fields we nevertheless have:
Proposition 3.9.3 Every proper preordering of A is contained in an ordering.
We first prove:
Lemma 3.9.4 Let T ⊆ A be a proper preordering and a ∈ A. Then:
(a) (aT ) ∩ (1 + T ) = ∅ or (−aT ) ∩ (1 + T ) = ∅;
(b) if (aT ) ∩ (1 + T ) = ∅, then T − aT is also a proper preordering of A.
Proof (a) Assume that as = 1 + s and −at = 1 + t for certain s, s , t, t ∈ T .
Then
−a 2 st = 1 + s + t + s t ,
hence −1 = a 2 st + s + t + s t ∈ T , a contradiction.
(b) The statements −1 ∈ T − aT and aT ∩ (1 + T ) = ∅ are equivalent.
Proof of Proposition 3.9.3 By Zorn’s Lemma it suffices to show that maximal

proper preorderings are orderings. Thus, let T be a maximal proper preordering.
Then T ∪ (−T ) = A by Lemma 3.9.4. Hence p := T ∩ (−T ) is an ideal of A.
Assume for the sake of contradiction that p is not a prime ideal. Then there exist
a, b ∈ A \ p such that ab ∈ p. We may take a, b ∈ T , hence a, b ∈ −T . By
Lemma 3.9.4(b) we then have (aT ) ∩ (1 + T ) = ∅ and (bT ) ∩ (1 + T ) = ∅, say
as = 1 + s and bt = 1 + t
for certain s, s , t, t ∈ T . Multiplication gives
abst = 1 + s + t + s t ,
hence −1 = (−ab)st + s + t + s t ∈ T , a contradiction.

The following theorems describe those functions that are positive (semi)definite
on sets of the form H A (F ):
Proposition 3.9.5 Let F be a subset of A and T := P [F ] the preordering it
generates. For any a ∈ A, the following statements are equivalent:
(i) a > 0 on H A (F ) (i.e., a(α) > 0 for all α ∈ H A (F ));
(ii) there exist t, t ∈ T such that at = 1 + t ;
(iii) there exist t, t ∈ T such that a(1 + t) = 1 + t .
Proof (i) ⇒ (ii): If −1 ∈ T , then H A (F ) = H A (T ) = ∅ and we may choose t = 0,
t = −1. Thus, let −1 ∈ T . Assume that (aT ) ∩ (1 + T ) = ∅. By Lemma 3.9.4
and Proposition 3.9.3 there exists an ordering P of A such that P ⊇ T − aT ⊇ T .
Then a(P ) ≤ 0 and P ∈ H A (F ), a contradiction.
(ii) ⇒ (iii): From at = 1 + t we get a(1 + t ) = a 2 t and addition gives a(1 +
t + t ) = 1 + (a 2 t + t ).
(iii) ⇒ (i): For any α ∈ H A (F ) we have t (α) ≥ 0, t (α) ≥ 0, hence a(α) > 0.
Corollary 3.9.6 Let F be a subset of A and T := P [F ]. For any a ∈ A, the

(i) a does not vanish on H A (F );
(ii) a divides an element of 1 + T .
Proof For (i) ⇒ (ii) we may replace a by a 2 and conclude with Proposition 3.9.5,
while (ii) ⇒ (i) is clear.
Corollary 3.9.7 Let R be a real closed field, A = R[t1 , . . . , tn ] and
B := {f/g : f, g ∈ A, g(x) = 0 for all x ∈ R n } ⊆ Quot A

the ring of regular functions on R n . Then B = S −1 A, where S := 1 + A2 . (An

analogous result holds for an arbitrary affine R-algebra A with VA (R) instead of
R n .)
Proof of ⊆ If g ∈ A is such that g(x) = 0 on R n , then g does not vanish on
Sper A since ZA (g) is constructible (cf. Theorem 3.5.3). By Corollary 3.9.6 there
exist s ∈ S and a ∈ A such that ag = s, i.e., such that g1 = as .
Proposition 3.9.8 Let F be a subset of A and T := P [F ]. For any a ∈ A, the

(i) a ≥ 0 on H A (F );
(ii) there exist t, t ∈ T and n ≥ 0 such that at = a 2n + t ;
(iii) there exist t, t ∈ T and n ≥ 0 such that a(a 2n + t) = a 2n + t ;
(iv) there exist t, t ∈ T and n ≥ 0 such that a(a 2n + t) = t .
Proof (i) ⇒ (ii): Let B := a −∞ A and U := {t/a 2n : n ≥ 0, t ∈ T } the preordering
generated by T in B. We identify Sper B with H̊A (a 2 ) ⊆ Sper A (see Sect. 3.3).
Since a is positive on H B (U ), there exist m, n ≥ 0 and t, t ∈ T such that
t t
a· =1+ in B
a 2m a 2n
by Proposition 3.9.5. We may assume without loss of generality that m = n, and it
follows that
at = a 2n + t in B,
i.e.,
a · a 2N t = a 2(N+n) + a 2N t in A
for some N ≥ 0. This is a relation of the form (ii).

(ii) ⇒ (iii): From at = a 2n + t it follows again that a(a 2n + t ) = a 2 t, hence
a(a 2n + t + t ) = a 2n + (a 2 t + t ) by addition. This is a relation of the form (iii).
The implications (iii) ⇒ (iv) and (iv) ⇒ (i) are clear.
Next we consider generalizations of these theorems that describe those functions
that are positive (semi)definite on a class of proconstructible sets that is more general
than sets of the form X = H A (F ).
Let F, G, E ⊆ A be subsets and consider the proconstructible subset
X := H̊A (F ) ∩ H A (G) ∩ ZA (E),
of Sper A, i.e.,

X = α ∈ Sper A : f (α) > 0 ∀f ∈ F, g(α) ≥ 0 ∀g ∈ G, e(α) = 0 ∀e ∈ E .
(Since ZA (a) = H A (−a 2 ) = H A (a) ∩ H A (−a) the contribution of ZA (E) is not

needed for the description of X, but it often does make the representation easier.)
Let S be the semigroup generated by F in A (and note that 1 ∈ S), T := P [F ∪ G]
the preordering generated by F and G, and a := Ae the ideal generated by E.
e∈E
Theorem 3.9.9 (Positivstellensatz) For any a ∈ A, the following statements are
equivalent:
(i) a > 0 on X;
(ii) ∃s ∈ S, ∃t, t ∈ T such that at ≡ s + t mod a;
(iii) ∃s ∈ S, ∃t, t ∈ T such that a(s + t) ≡ s + t mod a.
Theorem 3.9.10 (Nichtnegativstellensatz) For any a ∈ A, the following state-

ments are equivalent:
(i) a ≥ 0 on X;
(ii) ∃s ∈ S, ∃t, t ∈ T , ∃n ≥ 0 such that at ≡ a 2n s + t mod a;
(iii) ∃s ∈ S, ∃t, t ∈ T , ∃n ≥ 0 such that a(a 2n s + t) ≡ a 2n s + t mod a.
Theorem 3.9.11 (Nullstellensatz) For any a ∈ A, the following statements are

equivalent:
(i) a vanishes identically on X;
(ii) ∃s ∈ S, ∃t ∈ T , ∃n ≥ 0 such that a 2n s + t ∈ a.
Theorem 3.9.9 incorporates Proposition 3.9.5, Theorem 3.9.10 incorporates
Proposition 3.9.8 and Theorem 3.9.11 incorporates the weak real Nullstellensatz
Proposition 3.2.10 as a special case.
Exercise 3.9.12 Show that it suffices to consider the case where a = 0 for the
proofs of these Stellensätze.
Proofs By the exercise, we may assume without loss of generality that E = ∅. Let
B := S −1 A and let U = {t/s 2 : t ∈ T , s ∈ S} be the preordering of B generated
by T . Upon canonical identification of Sper B with a subspace of Sper A we have
X ⊆ Sper B, more precisely X = H B (U ). Thus, for all a ∈ A we have
a
a > 0 (≥ 0, = 0) on X ⊆ Sper A ⇔ > 0 (≥ 0, = 0) on X ⊆ Sper B.
1
Note that all s ∈ S are positive on X and that all t ∈ T are nonnegative on X.
Proof of Theorem 3.9.9. For the implication (i) ⇒ (ii) we let a > 0 on X. By
Proposition 3.9.5 (applied to B) there exist s, s ∈ S, t, t ∈ T such that
t t
a· =1+ 2 in B.
s2 s
Rewriting yields the existence of an s ∈ S such that
a(s s )2 t = (ss s )2 + (ss )2 t in A,
in other words, an identity of the form (ii). The implication (ii) ⇒ (iii) follows using
the same trick as in the proof of Proposition 3.9.5 and the implication (iii) ⇒ (i) is
clear.
Proof of Theorem 3.9.10. For the implication (i) ⇒ (ii) we let a ≥ 0 on X. By
Proposition 3.9.8 there exist s, s ∈ S, t, t ∈ T and n ≥ 0 such that
t t
a· 2
= a 2n + 2 in B.
s s
Hence there exists s ∈ S such that
a(s s )2 t = a 2n (ss s )2 + (ss )2 t in A,
an identity of the form (ii). The implication (ii) ⇒ (iii) follows as in the proof of
Proposition 3.9.8 and the implication (iii) ⇒ (i) is clear.
Proof of Theorem 3.9.11. For the implication (i) ⇒ (ii) we let a = 0 on X. By
Theorem 3.9.10 there exist s1 , s2 ∈ S, t1 , t1 , t2 , t2 ∈ T and m, n ≥ 0 such that
at1 = a 2m s12 + t1 and − at2 = a 2n s22 + t2 .
Multiplication then gives
−a 2 t1 t2 = a 2(m+n) (s1 s2 )2 + t
for some t ∈ T , an identity of the form (ii). The converse implication is clear.
Remark 3.9.13 Consider a real closed field R and an affine R-algebra A with
associated variety V . Letting F, G, E ⊆ A and keeping everything else as it was,
we obtain geometric Stellensätze as long as F and G are assumed to be finite. In
this situation X is namely constructible (since A is noetherian, the cardinality of E
does not matter) and in (i) of Theorems 3.9.9, 3.9.10 and 3.9.11 we may replace X
by X(R) := X ∩ V (R).
Example 3.9.14 The solution of Hilbert’s 17th Problem can be deduced from
Theorem 3.9.10: If A = R[t1 , . . . , tn ] and a ∈ A is positive semidefinite on R n ,
there exist f, g ∈ A2 and m ≥ 0 such that
a = (a 2m + f )/(a 2m + g) (and a 2m + g = 0).

This expression can be rewritten as
(a 2m + f )(a 2m + g)
a= ,
(a 2m + g)2
which we recognize as a sum of squares in Quot A. Moreover, it is possible to write

this expression in such a way that the denominators of the rational functions that
occur in it vanish at most at zeroes of a.
We illustrate the results obtained thus far with two applications of Prestel’s
Positivstellensatz (Proposition 3.9.5) and one application of the Nullstellensatz
(Theorem 3.9.11).
First, let A be an arbitrary commutative ring and P an ordering of A. We give
a description of the maximal specialization Q of P in Sper A (cf. Corollary 3.6.8).
Let Q+ denote the complement of q := supp Q in Q. Then A is the disjoint union
of Q+ , −Q+ and q. Thus, Q is in particular determined by Q+ .
Proposition 3.9.15 Q+ is the set of all those elements a ∈ P that divide an element
of 1 + P in A.
Proof Since the spear {P } = H A (P ) consists of generalizations of Q, we have that
Q+ is the set of those a ∈ A that are positive on H A (P ). The statement then follows
from Proposition 3.9.5 (applied to T = P ) since ap = 1 + p with a ∈ A and p,
p ∈ P implies that a ∈ P .
Next we derive a special case of a theorem of Hörmander concerning the growth
of polynomials in several variables (cf. [53], also [112, p. 224]). Refinements of
this theorem (loc. cit., [25, 42], . . . ) are useful in the theory of partial differential
equations (with constant coefficients in particular), which is a fertile ground for
applications of real algebra in general.
Let R be a real closed field and denote the euclidean norm on R n by ,x,. In other
words, ,x,2 = x12 + · · · + xn2 .
Proposition 3.9.16 Let f ∈ R[t1 , . . . , tn ] and let r ≥ 0 in R be such that f (x) = 0
for all x ∈ R n that satisfy ,x, ≥ r. Then there exists a constant C > 0 in R and a
natural number N such that
−N
|f (x)| ≥ C 1 + ,x,2
for all x ∈ R n that satisfy ,x, ≥ r.

Proof We let
g(t1 , . . . , tn ) := t12 + · · · + tn2 − r 2

and M := {x ∈ R n : g(x) ≥ 0}. By assumption, f does not vanish on M. Thus,

by Corollary 3.9.6 (applied to A = R[t1 , . . . , tn ] and F = {g}) there exists a
polynomial h ∈ R[t1 , . . . , tn ] such that
|f (x) h(x)| ≥ 1 for all x ∈ M. (3.3)
Let N denote the total degree of h. Then h is of the form

h(t) = cα t α
|α|≤N
for certain constants cα ∈ R. (In this expression, α = (α1 , . . . , αn ) ∈ Nn0 runs

through those multi-indices that satisfy |α| = α1 + · · · + αn ≤ N.) Since |xi | ≤
1 + ,x,2 , we have

|α| N
|h(x)| ≤ |cα | 1 + ,x,2 ≤ |cα | 1 + ,x,2
|α|≤N |α|≤N
for all x ∈ R n , and the statement follows from (3.3).

Let us return to an arbitrary commutative ring A.
Proposition 3.9.17 Let X be a closed subset of Sper A, and let f, g ∈ A be such
that every x ∈ X that satisfies f (x) = 0 also satisfies g(x) = 0. Then there exist an
a ∈ A and an n ∈ N such that
g(x)2n ≤ 1 + a(x)2 f (x)2
for all x ∈ X.
Conversely, if such an inequality is satisfied for all x ∈ X, then of course g
vanishes at the zeroes of f in X.
Proof By assumption we have X ∩ ZA (f ) ⊆ ZA (g). Since X is an intersection
of closed constructible subsets of Sper A and ZA (g) is constructible, there exists
a constructible closed overset Y of X such that Y ∩ ZA (f ) ⊆ ZA (g) (cf.
Corollary 3.4.10). We may replace X by Y . In other words, we may assume without
loss of generality that X is constructible. Then there exist (finitely generated)
preorderings T1 , . . . , Tr of A such that
X = H A (T1 ) ∪ · · · ∪ H A (Tr ).
(To see this, write the complement of X as a finite union of sets of the form
H̊A (f1 , . . . , fN ).) An application of the Nullstellensatz (Theorem 3.9.11) yields
equations
g 2ni + ti = ai f 2
where ni ∈ N, ti ∈ Ti and ai ∈ A (for i = 1, . . . , r), and after multiplication with

powers of g 2 we obtain n1 = · · · = nr =: n. For x ∈ H A (Ti ) we have
g(x)2n ≤ |ai (x)| f (x)2 ≤ 1 + ai (x)2 f (x)2 .
Hence, for every x ∈ X,
g(x)2n ≤ 1 + a1 (x)2 + · · · + ar (x)2 f (x)2 ≤ 1 + a(x)2 f (x)2 ,
where a := 1 + a12 + · · · + ar2 .
Historical Remarks Our Theorems 3.9.9–3.9.11 are usually attributed to G. Sten-

gle, but this is only partially correct. Stengle ([111], 1974) worked in a geometric
setting. First he proved a “semialgebraic Nullstellensatz” (our Theorem 3.9.11),
from which he then deduced the Nichtnegativstellensatz. However, Theorems 3.9.9
and 3.9.11 were essentially already discovered by J.-L. Krivine in 1964 [66].
After introducing preorderings of rings, Krivine worked with maximal preorderings,
which in our terminology correspond to closed points in the real spectrum. Krivine’s
work was largely unknown in the early years of real algebraic and semialgebraic
geometry, and was “rediscovered” only much later. For more details see [92,
Sect. 4.7]. We also mention the approaches in [17] and [112, §10] that are
historically largely independent of those of Stengle and Prestel.
In 1959 Łojasiewicz [76] proved a statement in the geometric setting about
semialgebraic functions (these are continuous functions with semialgebraic graph),
similar to Proposition 3.9.17. A modern treatment, for arbitrary real closed fields
R, of this famous and important result known as the Łojasiewicz inequality can
be found in [11, §2.6]. The theory of abstract semialgebraic functions (cf. [107]
or [108]) leads to the insight that Proposition 3.9.17 is not only related to the
Łojasiewicz inequality, but is in fact an abstract version of this result. (Łojasiewicz
proved his inequality even for semianalytic functions, over R = R of course. Also
in this case it seems possible to reduce the inequality to Proposition 3.9.17.)
3.10 The Convex Radical Ideals Associated to a Preordering
Let A be a ring. In Sect. 3.7 we studied the ideals of A that are convex with
respect to an ordering P . In this section we will more generally consider the notion
of convexity with respect to a preordering T . Furthermore, the Stellensätze from
Sect. 3.9 will provide a geometric insight in the case of T -convex radical ideals.
Almost all the results concerning T -convex radical ideals proved in this section
go back to G. Brumfiel [17, 18], who also conducted fruitful research on more
general T -convex ideals (“completely convex ideals”) with his student R. Robson,
cf. [18, 95].
3.10 The Convex Radical Ideals Associated to a Preordering 159
Let T be a preordering of A. We define the relation ≤T on A via
a ≤T b :⇔ b − a ∈ T .
Then ≤T is reflexive and transitive, but only antisymmetric (and thus a partial order
relation on A) if and only if T ∩ (−T ) = 0. (If T ∪ (−T ) = A, then ≤T can be
interpreted as an ordering of the abelian group A/T ∩ (−T ).) For every subgroup
G of the additive group (A, +) the following properties are equivalent:
(i) From t, t ∈ T and t + t ∈ G it follows that t ∈ G (and t ∈ G);
(ii) from a, b ∈ G, c ∈ A and a ≤T c ≤T b it follows that c ∈ G;
(iii) from a ∈ G, c ∈ A and 0 ≤T c ≤T a it follows that c ∈ G.
(For (i) ⇒ (ii), take t := c − a and t := b − c. For (iii) ⇒ (i), take a := t + t and
c := t. The multiplicative structure of A does not play a role here. Thus T need not
be a preordering either.)
Definition 3.10.1 Let T be a preordering of A. A subgroup G of A (considered as
additive group) is called T -convex (or convex with respect to T ) if the equivalent
conditions (i)–(iii) are satisfied.
This generalizes Definition 3.7.1, see also Definition 2.1.1. In this section we
only consider T -convex ideals, while in the next section we focus on T -convex
subrings. Since every intersection of T -convex subgroups of A is again T -convex,
every subgroup (resp. every ideal, every subring) of A yields a smallest T -convex
subgroup (resp. a smallest T -convex ideal, a smallest T -convex subring) of A that
contains the given object.
If T = A is the trivial (improper) preordering, then A is the only T -convex
subgroup of A, an uninteresting case.
Consider the smallest preordering T0 = A2 of A and an ideal a of A. If a is
real reduced (i.e., if the ring A/a is real reduced), then a is T0 -convex. The converse
√ take A = k[t]/(t ) and a = (0) for any real
is false in general (counterexample: 2
reduced ring k), but is true if a = a.

Since for preorderings T ⊆ T every T -convex subgroup is also T -convex, it
follows from Proposition 3.9.3 and Sect. 3.7 that for every proper preordering T of
A there exists a T -convex ideal of A, which is proper (i.e., distinct from A).
Let T be any preordering of A. We want to show that if p is any ideal of A that
is maximal with respect to the property p ∩ (1 + T ) = ∅, then p is T -convex.
Exercise 3.10.2 Let a = A be a T -convex ideal. Show that a ∩ (1 + T ) = ∅.
Exercise 3.10.3 Let S = ∅ be a multiplicative subset of A that is contained in

1 + T . Show that
T := {a ∈ A : there exists s ∈ S such that as ∈ T }
is a proper preordering of A that contains T .

Lemma 3.10.4 Assume that every a ∈ A satisfies the property “2a ∈ T ⇒ a ∈ T ”

(e.g., this is the case if 2 is a unit). Then a := T ∩ (−T ) is the smallest T -convex
ideal of A.
Proof a is an additive subgroup of A that satisfies “2c ∈ a ⇒ c ∈ a” and c2 a ⊆ a
for every c ∈ A. All a ∈ a and b ∈ A satisfy
2ba = (1 + b)2 a − b 2a − a.
Since the right hand side is in a, we also have ba ∈ a. Hence a is an ideal. The T -
convexity of a follows immediately (if t, t ∈ T such that t +t ∈ a, then t +t ∈ −T ,
hence t = (t + t ) − t ∈ −T ) and it is clear that a is contained in every T -convex
ideal of A.
Proposition 3.10.5 Let T be a preordering of A and assume that the ideal p of A is

maximal with respect to the property p ∩ (1 + T ) = ∅. Then p is a T -convex prime
ideal.
Proof p is a prime ideal by Exercise 3.1.10. Consider the proper preordering U :=
T + p of A and the multiplicative subset S := {2n : n ≥ 0} ⊆ 1 + U . Applying the
construction from Exercise 3.10.3 to S and U yields a preordering U = {a ∈ A :
2n a ∈ U for some n ≥ 0} that satisfies the property from Lemma 3.10.4. It follows
that p := U ∩ (−U ) is a U -convex, hence also T -convex ideal that contains p.
By Exercise 3.10.2 we have p ∩ (1 + T ) = ∅, and we conclude that p = p.
We deduce from Proposition 3.10.5 and Exercise 3.10.2 that the proper maximal
T -convex ideals of A are precisely those (prime) ideals p of A that are maximal for
the property p∩(1+T ) = ∅. If T0 = A2 we note that the condition a∩(1+T0 ) =
∅ characterizes the real ideals a of A (cf. Sect. 3.2); since the T0 -convex prime ideals
are precisely the real reduced ideals we have thus recovered Proposition 3.10.6 in
this special case.
Proposition 3.10.6 Let a be a T -convex ideal of A. Then every minimal prime ideal
of A that contains a is also T -convex.
Proof Let p be a minimal prime ideal of A that contains a. Then

U := t
s2
: t ∈ T, s ∈ A \ p
is a proper preordering of Ap . Indeed, if it were the case that −1 ∈ U , then there

would exist an s ∈ A \ p such that s 2 ∈ T ∩ (−T ) ⊆ a, contradicting a ⊆ p
and p prime. The (proper) ideal aAp is U -convex in Ap , hence contained in a U -
convex prime ideal of Ap by Zorn’s Lemma and Proposition 3.10.5. But pAp is the
only prime ideal of Ap that contains aAp , implying that pAp is U -convex. Then it
follows easily that p is T -convex in A.
Corollary 3.10.7 Let a be a proper ideal of A.

√
(a) If a is T -convex, then so is a.
(b) If a is a radical ideal, then a is T -convex if and only if all minimal prime ideals
of A that contain a are T -convex.
In general, a preordering is not the intersection of orderings, as remarked in
Sect. 3.9. Nevertheless we may restrict ourselves to intersections of orderings when
studying T -convex radical ideals, as we will show next. By Proposition 3.10.6 it
suffices to consider prime ideals.
Lemma 3.10.8 (A. Klapper, cf. [17, p. 63]) If T and T are preorderings of A and
p is a (T ∩ T )-convex prime ideal, then p is T -convex or T -convex.
Proof We consider the relation ≤T from the beginning of this section. If
a, a , b, b ∈ A, and if 0 ≤T a ≤T b and 0 ≤T a ≤T b , then also
0 ≤T aa ≤T bb . Assume for the sake of contradiction that p is neither T -convex,
nor T -convex. Then there exist a, b ∈ p and u, v ∈ A \ p such that
0 ≤T u ≤T a and 0 ≤T v ≤T b.
It follows that 0 ≤T u2 ≤T a 2 and 0 ≤T v 2 ≤T b2 , hence
0 ≤T u2 v 2 ≤T a 2 v 2 ≤T a 2 v 2 + b2 u2 and 0 ≤T u2 v 2 ≤T u2 b2 ≤T a 2 v 2 + u2 b2 ,
and so 0 ≤T ∩T u2 v 2 ≤T ∩T a 2 v 2 + b 2 u2 . Since a 2 v 2 + b 2 u2 ∈ p and u2 v 2 ∈ p,

this contradicts the (T ∩ T )-convexity of p.
Definition 3.10.9 For a preordering T of A we denote the intersection of all

orderings P of A such that P ⊇ T by T (thus T = A if T = A). If T = T,
we call T saturated.
If T is a proper preordering, then so is T by Proposition 3.9.3. For every
preordering T of A we have
= {a ∈ A : −a 2n ∈ T − aT for some n ≥ 0}
T

= a ∈ A : there exist t, t ∈ T and n ≥ 0 such that a(a 2n + t) = t
by the Nichtnegativstellensatz Proposition 3.9.8.

Proposition 3.10.10 Let T be a preordering of A and a a radical ideal of A. Then
a is T -convex if and only if a is T-convex.
Proof By Corollary 3.10.7(b) we may assume that a = p is prime. We prove the
nontrivial direction. Assume that p is T -convex, but not T-convex. Then there exist
a, b ∈ T such that a + b ∈ p and a, b ∈ p. There exist u, u , v, v ∈ T and m, n ≥ 0
such that au = a 2m + u and bv = b 2n + v . Both elements are in T , but not in p
since p is T -convex. It follows that also c := a · aubv and d := b · aubv are in T ,
but not in p. This contradicts the T -convexity of p since c + d = (a + b)aubv ∈ p.
Corollary 3.9.6 immediately yields that 1 + T consists of divisors of elements

of 1 + T . This result now also follows from Propositions 3.10.5 and 3.10.10. (Let
a ∈ T and assume that 1 + a is not a divisor of an element of 1 + T . Then, by
Proposition 3.10.5, there exists a T -convex prime ideal p such that 1 + a ∈ p. This
ideal is of course not T-convex, a contradiction.)
Definition 3.10.11 We call a closed subset X of Sper A probasic if it is of the form
X = H A (F ) for some subset F of A. Since the intersection of probasic closed sets
is again probasic, every subset X ⊆ Sper A has a smallest probasic closed overset

that we denote by X.
If X is a subset of Sper A we write henceforth

P (X) := a ∈ A : a(x) ≥ 0 ∀x ∈ X .
This is a preordering of A and it is precisely the intersection of all elements of X

when these are viewed as orderings of A. The preorderings of the form P (X) are
thus precisely the saturated preorderings of A. If X = {x}, then P (x) := P {x} is
just x itself, viewed as an ordering of A (cf. Sect. 3.3).
The relationship between preorderings of A and probasic closed subsets of
Sper A can be understood formally as follows: We have defined inclusion reversing
maps
between the power set of A and Sper A. These maps are such that for every subset
F ⊆ A we have
P H (F ) = P
[F ],
the saturation of the preordering of A generated by F , and for every subset X of

Sper A we have

H P (X) = X,
the smallest probasic closed subset of Sper A that contains X.

If we restrict H , resp. P , to the saturated preorderings of A, resp. the probasic
closed subspaces of Sper A, these operators give rise to inclusion reversing bijec-
tions that are inverse to each other. In other words, saturated preorderings of A
can be identified with probasic closed subspaces of Sper A via H and P . In what
follows, we will usually prefer to work with the subspaces of Sper A as these have
the advantage of providing greater geometric clarity.
Therefore, given X ⊆ Sper A we will often say X-convex subgroup (ideal, ring)
instead of P (X)-convex subgroup, etc. (If X = {x} we will also talk of subgroups,
convex in x, etc.) Proposition 3.10.10 thus says that for radical ideals a and arbitrary
preorderings T , a is T -convex if and only if a is H (T )-convex. A subgroup of A is

X-convex if and only if it is X-convex.
Since for subsets Y ⊆ X ⊆ Sper A, every Y -convex ideal is trivially also X-
convex, it follows in particular that for every x ∈ X the ideal supp(x) is X-convex.
We will show next that the converse also holds if X is closed.
Lemma 3.10.12 If X is a proconstructible subset of Sper A and p is a prime ideal
of A which is not convex in any x ∈ X, then there exists a closed constructible set
Y ⊇ X such that p is not Y -convex. In particular, p is not X-convex.
Proof For every x ∈ X there exist elements ax , bx ∈ A \ p such that ax (x) ≥ 0,
bx (x) ≥ 0 and ax + bx ∈ p. The sets Y (x) := H (ax , bx ) (x ∈ X) then constitute
a covering of X by closed constructible sets. Since the constructible topology is
compact (cf. Corollary 3.4.10(a)) there exist finitely many x1 , . . . , xr ∈ X such
that X ⊆ Y1 ∪ · · · ∪ Yr =: Y , where Yi := Y (xi ). There is no i ∈ {1, . . . , r} for
which p is Yi -convex. It follows from Lemma 3.10.8 that p is not Y -convex since
P (Y ) = P (Yi ).
i
Theorem 3.10.13 Let X be a closed subset of Sper A. Then the X-convex prime
ideals of A are precisely the supports supp(x) of all points x ∈ X.
Proof For x ∈ X, the ideal supp(x) is X-convex, as already remarked. Conversely,
consider an X-convex prime ideal p. By Lemma 3.10.12, p is convex in some x ∈ X,
i.e., p = supp(y) for some specialization y of x (cf. Theorem 3.7.4). Then y ∈ X
since X is closed.
From Theorem 3.10.13 we can deduce a very concrete geometric interpretation
of the X-convex radical ideals. If X ⊆ Sper A is a subset, we call a subset Z of X
algebraic in X if there exists and ideal a ⊆ A such that
Z = ZX (a) := X ∩ ZA (a) = {x ∈ X : a(x) = 0 for all a ∈ a}.
Writing
!
I (Y ) := supp (y) = {a ∈ A : a(y) = 0 for all y ∈ Y },
y∈Y
we have:
Theorem 3.10.14 Let X be closed in Sper A. Then the X-convex radical ideals a
of A correspond bijectively to the subsets Z of X that are algebraic in X via
Z = ZX (a) and a = I (Z).
Proof Since ZX ◦ I ◦ ZX = ZX , we have ZX ◦ I (Z) = Z for every subset Z

that is algebraic in X. Conversely, let a be an X-convex radical ideal. It is clear that
a ⊆ I ZX (a) . If f ∈ a, then Proposition 3.10.6 yields an X-convex prime ideal

p ⊇ a such that f ∈ p. By Theorem 3.10.13 we have p = supp(x) for some x ∈ X,
hence x ∈ ZX (p) ⊆ ZX (a). On the other hand, f (x) = 0, hence f ∈ I ZX (a) .
What does Theorem 3.10.14 mean in the “geometric setting”? Let V be an affine
variety over a real closed field R and let A := R[V ] be the associated R-algebra.
Consider a closed semialgebraic subset M ⊆ V (R) of the form

r

M= x ∈ V (R) : fi1 (x) ≥ 0, . . . , fis(i) (x) ≥ 0 (3.4)
i=1
for finitely many fij ∈ A. (By the—as yet unproven—Finiteness Theorem 3.5.8,
every closed semialgebraic M is of the form (3.4). We only require this assumption
in order to ensure that M is closed in V(R).) The (closed) constructible subset
M of Sper A that corresponds to M is the closure of M in Sper A (with respect
to the constructible or Harrison topology). It follows that P (M) = P (M) for the
associated (saturated) preorderings of A.
Let a be a radical ideal of A and W the associated closed subvariety of V . By
Theorem 3.10.14, a is M-convex if and only if a = I ZM (a) = I M ∩ ZA (a) .
However, ZA (a) = W (R), and so M ∩ ZA (a) = M ∩ W (R). Since I (N) = I (N)
for semialgebraic N ⊆ V (R) we can formulate Theorem 3.10.14 in the current
situation as follows:
Theorem 3.10.15 Let V be an affine R-variety, M a closed semialgebraic subset
of V (R) (of the form (3.4)), a a radical ideal of R[V ] and W the subvariety of V
associated to a. Then a is M-convex if and only if M ∩ W (R) is Zariski dense in W .
We return to our arbitrary commutative ring A and deduce two consequences
from Theorems 3.10.13 and 3.10.14. Let T be a preordering and a an ideal of A.
Definition 3.10.16 We denote the intersection of all T -convex ideals b that contain
a (i.e., the smallest T -convex ideal that contains a) by ciT (a). (Here “ci” stands for
“convex ideal”.)
Proposition 3.10.17 For every ideal a of A we have

$
ciT (a) = {a ∈ A : there exists n ≥ 0 and t ∈ T such that a 2n + t ∈ a}.
√
Proof First of all, c := ciT (a) is the smallest T -convex radical ideal ⊇ a by
Corollary 3.10.7(a). Let X := H (T ), i.e., P (X) = T. Since c is also X-convex by
Proposition 3.10.10, c is a fortiori the smallest X-convex radical ideal ⊇ a. Then
c = I ZX (a) by Theorem 3.10.14. By the Nullstellensatz Theorem 3.9.11 (applied
to F = ∅, G = T , E = a), I ZX (a) is exactly the right hand side from the
statement of the theorem.
We note that Proposition 3.10.17 is a generalization of the weak real Nullstellen-

√
satz (Proposition$3.2.10 which deals with the case T = A2 ); the real radical re a
of a is precisely ciA2 (a).
The following direct proof of Proposition 3.10.17 can be found in [18, p.√
59]: Let
b = {a ∈ A : a 2n + t ∈ a for some n ≥ 0 and t ∈ T }. Obviously, b ⊆ ciT (a)
and it suffices to show that b + b ⊆ b. Thus, let b, b ∈ b and b2m + t = a ∈ a,
b 2n + t = a ∈ a with t, t ∈ T . We may assume that m = n. We have
2n
(b + b )2 + (b − b )2 = (2b2 + 2b 2 )2n = b 2n u + b 2n u
with u, u ∈ T . Thus,
2n
(b + b )2 + (b − b )2 + tu + t u = au + a u ∈ a.
Since the left hand side is of the form (b + b )4n + v with v ∈ T , it follows that
b + b ∈ b.
The next theorem concerns the relationship between a subset X of Sper A and
that it generates:
the probasic closed set X
have the same image
Proposition 3.10.18 If X ⊆ Sper A is closed, then X and X
under the support map supp : Sper A → Spec A.

Proof This follows immediately from Theorem 3.10.13 since P (X) = P (X).
We finish this section with the interesting and arguably difficult problem of
determining the relationships between X and X in reasonable situations. For
instance, if

r
X= H (fi1 , . . . , fis(i) )
i=1
is an explicit closed constructible subset of Sper A, how is X characterized in terms

of the fij ? In which cases is X again constructible?
We can give some partial answers to the final question. In the “geometric setting”,
where A is a finitely generated algebra over a real closed field R, X is never

constructible, except in the trivial case where X = X, i.e., where X itself is already
probasic closed. Indeed, X ∩ V (R) = X ∩ V (R) by Proposition 3.10.18.

In other cases however, X = X can certainly be constructible. For an example,
take A = R((x))((y)) and let X consist of three of the four orderings of A.
3.11 Boundedness
In Sect. 2.1 we touched upon the notion of an archimedean field extension with
respect to a fixed ordering. We will now generalize this idea in two respects: instead
of field extensions we allow arbitrary ring extensions A ⊇ , and instead of a
fixed ordering we consider arbitrary subsets X of Sper A. This leads to the notion of
archimedean closure of in A on X, which we will characterize in several ways in
this Section. In fact, the archimedean closure can also be viewed as the convex hull
of a subring with respect to a preordering, as we will explain below.
In this section ϕ : → A will always denote a ring homomorphism.
Definition 3.11.1 Consider a subset X ⊆ Sper A.
(a) An element a ∈ A is called bounded on X over (or with respect to ϕ) if there
exists a λ ∈ such that
|a(x)| ≤ (ϕλ)(x)
for all x ∈ X. Clearly the elements of A that are bounded on X over form
a subring of A that contains ϕ( ). We denote this subring by oX (A/ ) or
oX (A, ϕ). If X = {x}, we simply write ox (A/ ) := o{x} (A/ ).
(b) We call A archimedean on X over (or with respect to ϕ) if oX (A/ ) =
A. If is a subring of A and ϕ the inclusion map, then we call oX (A/ )
the archimedean closure of in A on X. In case = oX (A/ ), we call
archimedean closed in A on X.
If we replace ϕ by the inclusion of the subring ϕ( ) of A, then oX (A/ ) does
not change. Hence, one can always think of as a subring of A (and of ϕ as the
inclusion map).
An element a ∈ A is (already) bounded on X with respect to ϕ if and only if
there exists a μ ∈ such that
|a(x)| ≤ |(ϕμ)(x)|
for all x ∈ X (take for instance λ = 1 + μ2 ). The description of oX (A/ ) as

archimedean closure (of in A on X) is justified since oX (A/ ) is obviously the
largest subring B of A that is archimedean on the image of X in Sper B over .
Note that the condition in Definition 3.11.1 is global with respect to X in the
sense that the elements of A are uniformly bounded on X by elements of . If X is
proconstructible however, we will see in Corollary 3.11.20 that this global condition
is equivalent to suitable local ones.
The archimedean closure of a subring of A is actually a convex hull with respect
to some preordering. Indeed, consider X ⊆ Sper A and let T := P (X) be the
associated preordering as in Sect. 3.10. Again we will say “X-convex” instead of
“T -convex”. The relation ≤T on A takes the form
a ≤T b ⇔ a(x) ≤ b(x) for all x ∈ X.

3.11 Boundedness 167
Now if ⊆ A is a subring, it is immediate that the subring oX (A/ ) is just

the X-convex hull of in A (which is a subring again in the current situation).
Therefore this section essentially deals with subrings that are convex with respect to
a preordering of the large ring. The subring is archimedean closed in A on X if
and only if is X-convex in A.
The construction of the archimedean closure is functorial in the following sense:
if
is a commutative diagram of ring homomorphisms and if X ⊆ Sper A and X ⊆

Sper A are subsets such that (Sper α)(X) ⊆ X , then α oX (A / ) ⊆ oX (A/ ).
The proof is clear.
Examples and Remarks 3.11.2
(1) If A = K is a field, a subring of K and X = {x} just one ordering of K, then
the subring ox (K/ ) was already introduced in Definition 2.1.8: ox (K/ ) is
the convex hull of in K with respect to x.
(2) In the general situation, the following identities hold for every subset X of
Sper A:
oX (A/ ) = oX (A/ ) = oX
(A/ ),
where X denotes the probasic closed subspace generated by X (cf. Sect. 3.10).
Indeed, by definition an element a ∈ A is in oX (A/ ) if and only if there exists
a λ ∈ such that X ⊆ H ϕ(λ)2 − a 2 . (Since the occurring rings are convex
hulls with respect to preorderings, the identities follow from P (X) = P (X) =

P (X).)
Definition 3.11.3 If X is a subset of Sper A, we define oX (A) := oX (A/Z) and call

the elements of oX (A) the absolutely bounded elements (of A) on X (with respect
to the unique homomorphism Z → A of course). In case X = Sper A, the elements
of o(A) := oSper A (A) are called the absolutely bounded elements of A. If A = K
is a field, o(K) is called the (real) holomorphy ring of K. We will study it in more
detail in Sect. 3.12.
From the functoriality of the archimedean closure mentioned above it follows
that α o(A) ⊆ o(A ) for every homomorphism α : A → A. More generally we
have α oX (A) ⊆ oX (A ) for all subsets X ⊆ Sper A and X ⊆ Sper A that satisfy
(Sper α)(X ) ⊆ X.
Example 3.11.4 If a1 , . . . , ar ∈ A and 1 + a12 + · · · + ar2 is a unit in A, then
a1 (1 + a12 + · · · + ar2 )−1 is an element of o(A) and is thus absolutely bounded. In
particular, if F is a real field, then all elements of the form a1 (1 + a12 + · · · + ar2 )−1
with a1 , . . . , ar ∈ F are absolutely bounded. More generally for any preordering T
of A, every element b ∈ A that satisfies an equation of the form b(1 + a 2 + t) = a
with a ∈ A and t ∈ T is absolutely bounded on H (T ).
Examples of archimedean ring extensions can be obtained from
Proposition 3.11.5 Every integral extension of rings A ⊇ is archimedean (on
the whole of Sper A).
Proof Every a ∈ A satisfies an equation of the form
a n + λ1 a n−1 + · · · + λn = 0
for certain λi ∈ . From Proposition 1.7.1 it follows that

|a(x)| ≤ max 1, |λ1 (x)| + · · · + |λn (x)|
for every x ∈ Sper A. Since the inequality |λ| ≤ 1 + λ2 holds in every ordered field
we conclude that
|a(x)| ≤ n + 1 + λ21 + · · · + λ2n (x).
We now prove a result that goes substantially beyond Proposition 3.11.5 and
which characterizes the archimedean closure of a subring on a subset of the
real spectrum. We may assume that char A = 0, i.e., that Z ⊆ A, since otherwise
Sper A = ∅ and the concepts become meaningless.
Definition 3.11.6 Let be a ring with Z ⊆ . We call a polynomial f (t) ∈ [t]
almost monic if its leading coefficient is a positive integer.
Theorem 3.11.7 (G.W. Brumfiel [17, Ch. VI]) Let A be a ring with Z ⊆ A, a
subring of A and X a subset of Sper A. For every a ∈ A the following statements
are equivalent:
(i) a ∈ oX (A/ ), i.e., a is bounded on X over ;
(ii) there exists an almost monic non-constant polynomial f (t) ∈ [t] such that
f (a 2 ) ≤ 0 on X;
(iii) there exists an almost monic non-constant polynomial f (t) ∈ [t] of even
degree such that f (a) ≤ 0 on X.
(For b ∈ A, the expression “b ≤ 0 on X” of course means that b(x) ≤ 0 for all
x ∈ X.)
Proof If a ∈ oX (A/ ) and λ ∈ such that |a(x)| ≤ λ(x) for all x ∈ X, then the
polynomial f (t) = t − λ2 satisfies (ii). The implication (ii) ⇒ (iii) is trivial (just
replace f (t) by f (t 2 )). It remains to show (iii) ⇒ (i). For this implication we appeal
to the remarkable
Lemma 3.11.8 (G.W. Brumfiel) Let n ∈ N and consider independent variables

t, u1 , . . . , u2n over Q. Let u := (u1 , . . . , u2n ). Then there exists k ∈ N and
polynomials b(u) ∈ Q[u] and h1 (u, t), . . . , hk (u, t) ∈ Q[u, t] such that

k
t 2n + u1 t 2n−1 + · · · + u2n = b(u) + hi (u, t)2 .
i=1
Thus, if is a ring such that Z ⊆ and f ∈ [t] is an almost monic polynomial

of even degree 2n, then there exists a positive integer s, an element β(f ) ∈ and
a polynomial h(t) that is a sum of squares in [t] such that β(f ) = sf (t) − h(t).
Returning to the proof of (iii) ⇒ (i), if we apply this observation to the polynomials
f (t) + t and f (t) − t, we obtain a positive integer s and elements β, β ∈ such

that β = s f (t) + t − h(t) and β = s −f (t) + t + h(t), where h, h are sums of
squares in [t]. The substitution t = a then gives

β(x) ≤ sa(x) ≤ β(x)
for all x ∈ X, from which (i) follows immediately.

Proof of Lemma 3.11.8 By induction on n. The case n = 1 follows from
u1 2 u21
t 2 + u1 t + u2 = t + 2 + u2 − 4 .
u1
If n > 1, we apply the “Tschirnhausen transformation” s := t − 2n and obtain
t 2n + u1 t 2n−1 + · · · + u2n = s 2n + v2 s 2n−2 + v3 s 2n−3 + · · · + v2n ,
where vi ∈ Q[u]. Rewriting the right hand side as

2
v2 −1 n−2 v2 −1 2
sn + 2 s + s 2n−2 + v3 s 2n−3 + v4 − 2 s 2n−4 + · · · ,
we can apply the induction hypothesis to the second summand, which proves the
claim for any n.
Now we come to a different description of the archimedean closure of in A on
X, in which a more prominent role is played by the preordering P (X) of elements
of A that are non-negative on X. If T is an arbitrary preordering of A, we will
also write oT (A/ ) instead of oH (T ) (A/ ) (and oT (A) instead of oT (A/Z)). Thus
oT (A/ ) is the T-convex hull of in A. Because of Example 3.11.2(2), every ring
oX (A/ ) is of the form oT (A/ ), for instance with T = P (X).
Proposition 3.11.9 Let be a subring of A and T a preordering of A.

(a) An element a ∈ A is in oT (A/ ) if and only if there exists λ ∈ and t ∈ T
such that
(1 + t)(λ2 + a) ∈ T and (1 + t)(λ2 − a) ∈ T .
(b) An element a ∈ A is in oT (A) if and only if there exists n ∈ N and t ∈ T such

that
(1 + t)(n + a) ∈ T and (1 + t)(n − a) ∈ T .
Proof From the existence of λ (resp. n) and t in (a) (resp. (b)), it follows that
−λ2 ≤ a ≤ λ2 (resp. −n ≤ a ≤ n) on X := H (T ). Hence it suffices to verify
the converse direction in (a). Given a ∈ oT (A/ ), we choose μ ∈ such that
|a(x)| ≤ |μ(x)| for all x ∈ X, and set λ := 1 + μ2. Since |a| ≤ |μ| < λ ≤ λ2 on X,
λ2 ± a are positive on X and the statement follows from Prestel’s Positivstellensatz
(Proposition 3.9.5, the (i) ⇒ (ii) implication).
Corollary 3.11.10 If every element of 1 + T is a unit in A (for instance, if A is a

field and T is a proper preordering of A), then
oT (A/ ) = {a ∈ A : there exists λ ∈ such that λ2 ± a ∈ T }
and
oT (A) = {a ∈ A : there exists n ∈ N such that n ± a ∈ T }.
Corollary 3.11.11 If every element of 1 + A2 is a unit in A, then
o o(A/ )/ = o(A/ ).
Proof Let B := o(A/ ). We need to show that every b ∈ B is bounded on the

whole of Sper B over . Applying Corollary 3.11.10 (with T = A2 ) provides
elements λ ∈ and a1 , . . . , ar ∈ A and an equation
λ4 − b 2 = a12 + · · · + ar2 .
Every |ai | is bounded by λ2 on the whole of Sper A, which shows that the ai are in
B. We conclude that |b| is bounded by λ2 on the whole of Sper B.
In the case of fields, Corollary 3.11.10 provides a parametrization of the elements

of oT (A/ ):
Proposition 3.11.12 Consider a field K, a proper preordering T of K and a
subring ⊆ K. Then oT (K/ ) is the set of all elements
λa
1 + a2 + t
such that λ ∈ , a ∈ K and t ∈ T .

Proof By Example 3.11.4 every such element is in oT (K/ ). Conversely, given
an element z = 0 in oT (K/ ), Corollary 3.11.10 yields an element λ ∈ such
2
that λ2 − z2 ∈ T , i.e., such that λz − 1 =: t ∈ T . Setting a := λz , we have
2 λz a = 2a 2 = 1 + a 2 + t, from which we obtain
2λa
z= ,
1 + a2 + t
as required.
Corollary 3.11.13 If K is a field, X a subset of Sper K and a subring of K, then
oX (K/ ) = · oX (K) = {λz : λ ∈ , z ∈ oX (K)}.
Corollary 3.11.14 If K is a field and T a proper preordering of K, then oT (K) is

generated as an additive group by the elements 1/(1 + t) with t ∈ T .
Proof This follows from Proposition 3.11.12 and the identity
2a (a + 1)2 − (a − 1)2
= .
1+a +t
2 (a + 1)2 + (a − 1)2 + 2t
We return to the case of an arbitrary commutative ring A and want to determine

the dependence of the ring oX (A/ ) on the subset X ⊆ Sper A when the subring
is kept fixed. Since Y ⊆ X implies oY (A/ ) ⊇ oX (A/ ) we always have
!
oX (A/ ) ⊆ ox (A/ ),
x∈X
where the inclusion is strict in general, as the following example shows:

Example 3.11.15 Let R be a real closed field, A = R[t] the polynomial ring in
one variable and = R. Every non-constant polynomial f ∈ R[t] is in precisely
two points of Sper R[t] unbounded over R, namely in ±∞ (cf. Example 3.3.14).
However, on the open subsets X := Sper R[t] \ {±∞} of Sper R[t] only the
constant polynomials are archimedean over R. Hence,

!
oX R[t]/R = R = R[t] = ox R[t]/R .
x∈X
Nevertheless we will shortly see that if X is proconstructible, then the inclusion

is an equality. But first:
Proposition 3.11.16 Let be a subring of A. For every a ∈ A, the “unbounded
locus”
!(a/ ) := {x ∈ Sper A : a ∈ ox (A/ )}
of a over is closed in Sper A and for each point in it, its generalizations are also
in it.
Proof We have
!
!(a/ ) = {x ∈ Sper A : ∀λ ∈ |a(x)| ≥ |λ(x)|} = H (a 2 − λ2 ),
λ∈
and since we may replace “≥” by “>”, we also have

!
!(a/ ) = H̊ (a 2 − λ2 ).
λ∈
The claims then follow from these two representations.
Corollary 3.11.17 If x, y ∈ Sper A are such that y % x, then oy (A/ ) =

ox (A/ ).
Example 3.11.18 Consider again a real closed field R and let V be an affine R-
variety with associated coordinate algebra A = R[V ]. Let u1 , . . . , un be generators
of R[V ] over R and a := u21 + · · · + u2n ∈ A. Then !(a/R) consists of all elements
x ∈ Sper A = V (R) whose “distance to the origin” (with respect to the embedding
of V into the affine space An given by u1 , . . . , un ) is infinitely large compared to R.
By Proposition 3.11.16, !(a/R) consists precisely of all the generalizations of the
points of
!max (a/R) := !(a/R) ∩ (Sper A)max .
Corollary 3.7.18 then gives
!max (a/R) = {x ∈ (Sper A)max : k(x) is not archimedean over R}.

In particular, !(a/R) is independent of the choice of generators of R[V ] and can

thus be referred to as the set of infinitely distant points in V
(R). This fits well with
Remark 3.6.6.
Proposition 3.11.19 Let be a subring of A and X a proconstructible subset of

Sper A. If a ∈ A is bounded in every x ∈ X over , then there exists an open
constructible neighbourhood U of X in Sper A on which a is bounded over .
Proof By Proposition 3.11.16, a is also bounded in every x ∈ X over . Hence
we may assume that X = X. For every x ∈ X there exists λx ∈ such that
|a(x)| < λx (x). Since X is quasi-compact, there are finitely many x1 , . . . , xN ∈ X
such that

N
X⊆ {y ∈ Sper A : |a(y)| < λxi (y)} =: U.
i=1
The set U is an open constructible neighbourhood of X = X, and letting λ :=

1 + λ2x1 + · · · + λ2xN we have |a(y)| < λ(y) for all y ∈ U .
Corollary 3.11.20 Let be a subring of A and X a proconstructible subset of

Sper A. Then
!
oX (A/ ) = ox (A/ ),
x∈X
where we may restrict the intersection on the right hand side to the points x of Xmax
or Xmin .
Proof This follows from Proposition 3.11.19 and Corollary 3.11.17.
Remark 3.11.21 Let V be an affine variety over a real closed field R and A = R[V ]
its coordinate algebra. Given a semialgebraic subset M of V (R) with associated
constructible subset M of Sper A, one might ask if oM (A/ ) is already the
intersection of the rings ox (A/ ) with x ∈ M, for instance for = R. In general
the answer to this question is negative. Trivially, ox (A/R) = A for all x ∈ V (R),
whereas only in special cases the R-valued function x → a(x) is bounded on M for
every a ∈ A. In all other cases, oM (A/R) is a proper subring of A. In connection
with Corollary 3.11.20, the usefulness of “ideal points” in M becomes clear at this
point once more: For every polynomial function f ∈ A that is unbounded on M,
and in this point f does indeed attain an infinitely
there exists such a point in M,
large function value compared to R.
3.12 Prüfer Rings and the Real Holomorphy Ring of a Field
We finish Chap. 3 with a detailed study of the holomorphy ring o(F ) of a (real) field
F . The well-developed theory of the holomorphy ring—and more generally the ring
oT (F )—as well as interesting applications to power sums in fields is mainly due to
E. Becker and his student H.-W. Schülting (cf. [7, 8, 106], . . . ). A key notion in this
context is the concept of a Prüfer ring. We will only explain the rudiments of this
theory, which for the most part already go back to the work of D.W. Dubois [37].
Our presentation (strongly) emphasizes the use of the real spectrum and differs in
this respect from these earlier works. Once again the real spectrum often plays a
helpful and clarifying role.
Definition 3.12.1
(a) An integral domain A is called a Prüfer ring if every localization Ap of A at a
prime ideal p is a valuation ring. If in addition A is a subring of a field F and if
F is the field of fractions of A, then A is called a Prüfer ring of F .
(b) The Prüfer ring A is called residually real if for all p ∈ Spec A the residue field
κ(p) = Ap /pAp is real (i.e., if all valuation rings Ap are residually real in the
sense of Sect. 2.2). The notion of residually real Prüfer ring of F is defined
analogously.
Remarks 3.12.2
(1) Every valuation ring B is a Prüfer ring. If in addition κ(B) = B/mB is real,
then B is convex in F := Quot(B) with respect to an ordering P of F by
Corollary 2.7.2. The same is true for every overring of B in F , and it follows
that Bp /pBp is real for every p ∈ Spec B. This shows that our two definitions
of “residually real” (Definitions 3.12.1(b) and 2.2.5) are equivalent.
(2) If A is a Prüfer ring of F , then every overring B of A in F is also a Prüfer ring
since Bq is an overring of Aq∩A in F for every q ∈ Spec B. Similarly, every
overring (in F ) of a residually real Prüfer ring of F is again a residually real
Prüfer ring (cf. Remark 3.12.2(1)).
(3) Let A be an integral domain. If all localizations Am at maximal ideals m of A
are valuation rings (resp. residually real valuation rings), then A is already a
Prüfer ring (resp. a residually real Prüfer ring) since for every p ∈ Spec A such
that p ⊆ m, Ap is an overring of Am in the field of fractions of A.
(4) If A is an integral domain and F = Quot A, then A = m Am where the
intersectionis taken over all maximal ideals m of A in F . (This is easy to see:
Given λ ∈ m Am , the ideal of all elements a of A such that aλ ∈ A is not
contained in any maximal ideal of A, hence must be the unit ideal.) It follows
that every Prüfer ring of F is the intersection of valuation rings of F and thus
in particular integrally closed.
(5) If A is a Prüfer ring of F and K a further field, then there exists a natural
bijection from the set of places λ : F → K ∪ ∞ that are finite on A to
Hom(A, K), namely λ → λ|A . (Here and later Hom(·, ·) denotes the set of
3.12 Prüfer Rings and the Real Holomorphy Ring of a Field 175
ring homomorphisms.) Indeed, every homomorphism ϕ : A → K has a unique

extension to a homomorphism from a valuation ring that contains A (namely
AKer ϕ ) to K, thus to a place F → K ∪ ∞ (that is integral on A).
Exercise 3.12.3 Let A be a Prüfer ring of F . Show that the map p → Ap is a

bijection from Spec A to the set of all valuation rings B of F that contain A and that
the inverse map is given by B → A ∩ mB .
The importance of Prüfer rings for real algebra is a consequence of the following
theorem of A. Dress [36] (see also [70, p. 86]):
Theorem 3.12.4 Let F be a field of characteristic = 2. Then the subring A of F
that is generated by all elements of the form
1
(a ∈ F, a 2 = −1)
1 + a2
is a Prüfer ring of F .
Proof Let p be a prime ideal of A and B := Ap . Consider x ∈ F ∗ . We must show
that x ∈ B or x −1 ∈ B. First of all, x 2 ∈ B or x −2 ∈ B. This is clear if x 2 = −1.
Otherwise this follows from the identity
1 1
+ = 1,
1 + x2 1 + x −2
since the summands on the left hand side are not both in p, i.e., since one of those
summands is invertible in B. We may assume without loss of generality that x 2 ∈ B.
Let y := 1 + x. If y 2 ∈ B, then 2x = y 2 − 1 − x 2 ∈ B, hence x ∈ B. Otherwise,
y = 0 and y −2 ∈ B. Since (y − 1)2 = x 2 ∈ B we also have
y −2 (y − 1)2 = 1 − 2y −1 + y −2 ∈ B,
hence y −1 ∈ B. From 1 − x = y −1 (1 − x 2 ) we then conclude that x ∈ B, as

required.
We recall that o(F ) denotes the real holomorphy ring of F (cf. Definition 3.11.3).
Theorem 3.12.5 Consider a real field F . The following statements are equivalent
for every subring A of F :
(i) o(F ) ⊆ A;
(ii) A is a residually real Prüfer ring of F ;
(iii) A is archimedean closed in F , i.e., o(F /A) = A;
(iv) A is a F 2 -convex subring of F ;
(v) there exists a subring of F and a subset X of Sper F such that A =
oX (F / ).
In particular, the real holomorphy ring o(F ) of F is the smallest residually real
Prüfer ring of F .
Proof (i) ⇒ (ii): Since o(F ) contains all elements of the form (1 + a 2 )−1 with a ∈
F , o(F ) is a Prüfer ring of F by Theorem 3.12.4. Consider a valuation ring B of F
such that o(F ) ⊆ B. If B is not residually real, there exist b1 , . . . , br ∈ B such that
1+b12 +· · ·+br2 ∈ mB , which is impossible since (1+b12 +· · ·+br2)−1 ∈ o(F ) ⊆ B.
Hence A is a residually real Prüfer ring of F .
(ii) ⇒ (iii): Since intersections of archimedean closed subrings of F are again
archimedean closed in F , it suffices to prove that residually real valuation rings of F
are archimedean closed. If B is a residually real valuation ring of F , then B is convex
in F with respect to some x ∈ Sper F (cf. Sect. 2.7). It follows that ox (F /B) = B,
hence a fortiori o(F /B) = oSper F (F /B) = B.
The implications (iii) ⇒ (iv) ⇒ (v) ⇒ (i) are trivial.
Corollary 3.12.6 The real holomorphy ring o(F ) is the intersection of all residu-
ally real valuation rings of F .
Remark 3.12.7 Corollary 3.12.6 is already contained in earlier results: By Corol-

lary 3.11.20, o(F ) is the intersection of the residually real valuation rings ox (F )
with x ∈ Sper F , and every residually real valuation ring of F contains such
an ox (F ) by Sect. 2.7. More generally we know from Sect. 3.11 that every ring
oX (F / ) (with X ⊆ Sper F and a subring of F ) is the intersection of the
residually real valuation rings ox (F / ) with x ∈ X.
In what follows we fix a field F , assumed to be real without loss of generality,
and investigate the real spectrum of its holomorphy ring o(F ) in some more detail.
For simplicity we write o := o(F ).
Proposition 3.12.8 Let A be a residually real Prüfer ring of F . Viewing Sper F in
the usual way as a subspace of Sper A, we have Sper F = (Sper A)min .
Proof Let x ∈ Sper A, p := supp(x), B := Ap and x the ordering of κ(p) = B/mB
induced by x. By the Baer–Krull Theorem there exists an ordering y of F that makes
B convex and that induces the ordering x of B/mB . Viewing y and x as elements
of Sper B, it is immediately clear that y % x. Hence the inclusion Sper B ⊆ Sper A
shows that y is a generalization of x in Sper A with support (0).
What is happening with the maximal (= closed) points of Sper o? Let x ∈ Sper o,
p := supp(x) and x the induced ordering of κ(p). By Corollary 3.7.18, x is a
closed point if and only if κ(p) is archimedean over o/p with respect to x. Since
o/p is archimedean over Z with respect to x by definition of o, x being closed
is equivalent to x being an archimedean ordering of κ(p), i.e., equivalent to the
existence of an order preserving embedding κ(p) → R. Since this embedding is
uniquely determined by x, we obtain:
Proposition 3.12.9 The canonical map Hom(o, R) → Sper o is a bijection from
Hom(o, R) to (Sper o)max .
Corollary 3.12.10 If F contains R, then the support map supp : Sper o → Spec o
yields a bijection from (Sper o)max to the space (Spec o)max of maximal ideals of o.
Note that this bijection is not a homeomorphism in general.
Proof Since the field R does not possess any proper endomorphisms by Corol-
lary 2.1.12, every homomorphism ϕ : o → R is surjective, and ϕ is determined
by its kernel. Thus supp(x) is a maximal ideal for every x ∈ (Sper o)max
by Proposition 3.12.9. Since all prime ideals of o are real reduced, the map
supp : (Sper o)max → (Spec o)max is surjective. In the commutative diagram
the map Ker is injective. Thus the map supp is also injective by Proposition 3.12.9.
By identifying Hom(o, R) with (Sper o)max we thus obtain a compact (Hausdorff)

topology on Hom(o, R) (cf. Theorem 3.6.4), which can easily be described without
reference to the real spectrum:
Proposition 3.12.11 The topology induced by Hom(o, R) ⊆Sper o on Hom(o, R)
coincides with the subspace topology of Hom(o, R) in Ro = R.
o
Proof Let T, resp. T , denote the subspace topology of Hom(o, R) in Sper o, resp.
in Ro . A subbasis of open sets of T is given by all the sets
S(a) := {ϕ ∈ Hom(o, R) : ϕ(a) > 0},
where a ∈ A, because S(a) = Hom(o, R) ∩ H̊o (a). Since all S(a) are T -open, T
is finer than T. A subbasis of open sets of T is given by all the sets
S(a, p, q) := {ϕ ∈ Hom(o, R) : p < ϕ(a) < q}
where a ∈ A and p, q ∈ Q with p < q. Since S(a, p, q) = S(a − p) ∩ S(q − a)

(note that Q ⊆ o), T is also finer than T .
In the following we denote the compact space Hom(o, R) simply by Y , and
investigate the ring homomorphism
τ : o → C(Y, R), a →
a,
where C(Y, R) denotes the ring of continuous R-valued functions on Y and for all
a ∈ o, a is defined by
a (ϕ) := ϕ(a) for all ϕ ∈ Y .
It follows immediately from the definition that the functions
a separate the points
of Y . Furthermore, Q is contained in o. The Stone–Weierstrass Theorem [14, Ch. 10,
§4, Th. 3] then implies that τ (o) is a dense subring of C(Y, R) with respect to the
uniform convergence topology.
The subset C + (Y, R) of nonnegative functions is a preordering of the ring
C(Y, R) (identical to the set of squares). Its preimage under τ satisfies
Proposition 3.12.12 Let a ∈ o. Then a ∈ C + (Y, R) if and only if a + 1
n is a sum of
squares (in F or o) for every n ∈ N.
Proof First of all, it is clear that o2 = o ∩ F 2 , since if a12 + · · · + ar2 is bounded
a + n1 ∈ C + (Y, R), and
over Z (on Sper F ), then so is each ai . If a + n1 ∈ F 2 , then
if this holds for all n ∈ N it follows that +
a ∈ C (Y, R). Conversely, a ∈ C + (Y, R)
implies that a is nonnegative as a function on (Sper o) max (cf. Proposition 3.12.9).
For all n ∈ N, the intersection of
H o −a − 1
n = {x ∈ Sper o : 1
n + a(x) ≤ 0}
and (Sper o)max is thus empty. Hence, H o −a − n1 = ∅ and so a + n1 > 0 on

Sper o. In particular, a + n1 is positive on Sper F (cf. Proposition 3.12.8) and hence
a sum of squares (cf. Sect. 1.1).
Corollary 3.12.13 The kernel of τ : o → C(Y, R) consists precisely of those a ∈ F

for which n1 ± a is a sum of squares (in o or F ) for all n ∈ N, i.e., those elements of
F that are “infinitesimal” with respect to Q on the whole of Sper F .
Just as for (Sper o)min , there is a purely field theoretic description of (Sper o)max
that makes no reference to o. We use the following notation:
Definition 3.12.14 The set of all R-valued places F → R ∪ ∞ of F is denoted by
M(F ).
Every λ ∈ M(F ) is finite on o by Corollary 3.12.6. There is thus a natural
map M(F ) → Hom(o, R), λ → λ|o , which is a bijection by Remark 3.12.2(5).
Hence we may identify the set M(F ) with Y = Hom(o, R) in what follows. Note
that by Proposition 3.12.9 there is then also a natural bijection between M(F ) and
(Sper o)max .
In order to transfer the (compact) topology from Hom(o, R), resp. (Sper o)max
(Proposition 3.12.11) to M(F ), we view the elements λ ∈ M(F ) as maps from
F to the projective line P1 (R) over R (which is a compact space), by identifying
c ∈ R and ∞ with the points with homogeneous coordinates (c : 1) and (1 : 0),
respectively. For later use we remark that (c : d) → (d : c) defines a continuous
involution ι on P1 (R) that exchanges 0 and ∞ and that maps c ∈ R to 1/c.
For a ∈ F we define the evaluation map a : M(F ) = Y → R ∪ ∞ = P1 (R) by

a (λ) = λ(a). For a ∈ o this definition coincides with the earlier one ( a does not
take the value ∞ in this case), and we have:
Proposition 3.12.15 For every a ∈ F , the map a : M(F ) = Y → P1 (R) is

continuous (with Y carrying the topology defined earlier). Therefore the topology of
M(F ) is also the coarsest topology that makes all
a (a ∈ F ) continuous.
Proof Without loss of generality we may assume that a = 0. Let λ ∈ M(F ). We
prove the continuity of a in λ. Firstly, if λ(a) = ∞, there exist b, c ∈ o such that
a = b/c and λ(c) = 0. The continuity of a in λ follows from the fact that
a coincides
with the continuous function c on the open neighbourhood {μ ∈ M(F ) : μ(c) =
b/
0} of λ. If on the other hand λ(a) = ∞, then where ι denotes the
a = ι ◦ (1/a),
involution on P1 (R) mentioned earlier. The continuity of a in λ thus follows from
the first part of the proof.
Remark 3.12.16 Let ρ : Sper o → (Sper o)max = M(F ) denote the (continuous)
canonical retraction, which maps every x ∈ Sper o to the tip of the spear
{x} in Sper o (i.e., the closed specialization of x in Sper o), cf. Sect. 3.6. Given
x ∈ Sper F = (Sper o)min, it is easy to describe ρ(x) without reference to the
holomorphy ring o. Namely, ρ(x) is the “canonical” place F → R ∪ ∞ associated
to the ordering x of F , in other words the uniquely determined x-compatible place
whose valuation ring is the convex hull ox (F ) of Z in F with respect to x. The
restriction Sper F → M(F ) of ρ then makes M(F ) into a topological quotient
space of Sper F (since this map is closed).
A deeper geometric understanding of the space of places M(F ) (for instance in
the case of function fields over real closed base fields) is beyond our current abilities,
as this would require further knowledge of algebraic geometry beyond what we
can cover in this book (e.g., blow-ups). Nevertheless we present two illustrative
examples. These are quite special since they are “one-dimensional”. Therefore we
must emphasize that the holomorphy ring o(F ) and the space M(F ) are also useful
for higher-dimensional geometry.
Example 3.12.17 (Schülting, [106, p. 11 ff.]) Consider a one-dimensional real-an-
alytic manifold N and let M be a nonempty compact connected subset of N. Let
O(M) := lim O(U ), where U runs through the open neighbourhoods of M and
−→
O(U ) denotes the ring of real-analytic functions U → R. Then O(M) is an integral
domain. Let F denote the field of fractions of O(M), i.e., the field of “meromorphic
functions on M”. We assume that the functions of O(M) separate the points of
M (in fact, this is always the case, cf. [106, p. 13]). Then O(M) = o(F ) and the
canonical evaluation map M → Hom(o(F ), R) is a homeomorphism. Thus we have
rediscovered M with its topology as the space of places M(F ), and the holomorphy
ring o(F ) as the holomorphic functions on it.
Proof Every point x ∈ M gives rise to a discrete valuation vx of F with residue
field R, namely the order (of zeroes) at x. We have O(M) = {f ∈ F : vx (f ) ≥
0 for all x ∈ M}, from which o(F ) ⊆ O(M) follows. Conversely, every f ∈ O(M)
is bounded on M, hence bounded on an open neighbourhood U of √ M since M
is compact. If n ∈ N is chosen such that |f (x)| < n on U , then n ± f are
analytic on U . Hence, n±f are squares in F , from which f ∈ o follows. Therefore,

o(F ) = O(M). Now let o := o(F ) = O(M). The evaluation map M → Hom(o, R)
is trivially continuous and is injective by assumption. Since both spaces are compact,
it suffices to verify the surjectivity of this map. Thus, consider a homomorphism
ϕ : o → R with kernel p. Then p is a maximal ideal of o since R ⊆ o. For f ∈ o, let
Z(f ) := {x ∈ M : f (x) = 0}. If the intersection of all Z(f ) with f ∈ p is empty,
then there exist finitely many f1 , . . . , fr ∈ p such that Z(f1 )∩· · ·∩Z(fr ) = ∅ since
M is compact. But then g := f12 +· · ·+fr2 has no zeroes on a neighbourhood of M,
meaning that 1/g ∈ o which leads to the contradiction 1 = (f12 + · · · + fr2 )/g ∈ p.
Thus, consider x ∈ M such that f (x) = 0 for all f ∈ p. Since p is maximal, it
follows immediately that ϕ(f ) = f (x) for all f ∈ o.
This example illustrates the idea that gave rise to the term “holomorphy ring”: if
F is a field, one views the elements of F as continuous P1 (R)-valued functions
on the compact space M(F ) (“meromorphic functions”). The elements of o(F )
are then precisely those functions that never become infinite, in other words the
“holomorphic functions”.
Example 3.12.18 In this example we use some basic facts from introductory
algebraic geometry, see for instance [97], [45, Ch. I]. Consider a smooth projective
irreducible curve V , which is defined over R (i.e., described by homogeneous
polynomials with coefficients in R). Let F = R(V ) denote its function field. It
is well-known that every point x ∈ V (C) defines a discrete valuation ring Ox of F
(the ring of regular functions in x), where Ox is residually real (with residue field
R) if and only if x ∈ V (R). (Furthermore, Ox = Oy if and only if x and y are
complex conjugate.) Since all the Ox (x ∈ V (C)) are nontrivial valuation rings
of F over R, we have o(F ) = O(V (R)), the ring of functions that are regular
on the whole of V (R), and there is a canonical bijection ε : V (R) → M(F ).
If we endow V (R) with its strong (= classical) topology as usual, then V (R) is
compact. The functions f : V (R) → R that are regular on V (R) are of course
continuous. By Proposition 3.12.11 it then follows that ε is continuous, in other
words a homeomorphism. Thus we may identify V (R) with M(F ).
The next discussion topic focuses on the relationship between preorderings and
(residually real) Prüfer rings with respect to (real) places.
Consider a surjective place λ : F K ∪ ∞ and a preordering T of F . Then
λ(T ∩ oλ ) is a preordering of K (cf. Remark 3.9.2(2); oλ is the valuation ring of λ),
which we—slightly sloppily—denote by λ(T ).
Definition 3.12.19 The place λ is called compatible with T if −1 ∈ λ(T ), i.e., if
λ(T ) is a proper preordering of K.
Remarks 3.12.20
(1) The place λ is compatible with the preordering F 2 of F if and only if λ is real
(i.e., K is a real field). In this case we have λ(F 2 ) = K 2 .
(2) If T = P is an ordering of F , then λ is compatible with P if and only if λ(P )

is an ordering of K. This is equivalent to the existence of an ordering Q of K
such that λ is compatible with P and Q in the sense of Sect. 2.8 (and then we
necessarily have Q = λ(P )).
We will shortly present a whole list of equivalent characterizations of the
compatibility of λ with T , as well as a description of all the places λ of F that
satisfy oT (F ) ⊆ oλ (note that oT (F ) is the intersection of all these oλ ). We start
with some preliminary results.
Lemma 3.12.21 Let T be a proper preordering of F and A an overring of oT (F )
in F . Then every ideal of A is convex in F with respect to T . In particular, A is a
T -convex subring of F .
Proof Consider a ∈ A and b ∈ F such that 0 ≤T b ≤T a. If a = 0, then b = 0 too.
Otherwise we have 0 ≤T ba −1 ≤T 1, i.e., ba −1 ∈ oT (F ) ⊆ A (since oT (F ) is the
T -convex hull of Z in F ), and conclude that b = (ba −1)a ∈ Aa.
This shows in particular that the T -convex subrings of F are precisely the
overrings of the T -convex hull oT (F ) of Z in F .
Lemma 3.12.22 Let T be a proper preordering of F , A a Prüfer ring of F and p a
prime ideal of A. The following statements are equivalent:
(i) Ap is convex in F with respect to T ;
(ii) p is convex in A with respect to T ∩ A.
Proof The T -convexity of pAp in F , thus also the T ∩A-convexity of p = (pAp )∩A
in A, follows from (i) by Lemma 3.12.4. Conversely, assume that (ii) holds and
consider a ∈ Ap and b ∈ F such that 0 ≤T b ≤T a. Assume that b ∈ Ap . Then
b−1 ∈ pAp , hence also ab−1 ∈ pAp , and 0 ≤T 1 ≤T ab−1. There exist c, s ∈ A
such that c ∈ p, s ∈ p and ab−1 = cs −1 . Then (c − s)s = s 2 (ab−1 − 1) is an
element of T ∩ A but not of p, and
s 2 + (c − s)s = cs ∈ p
contradicts the T ∩ A-convexity of p.
Lemma 3.12.23 (G. Brumfiel) Consider an integral domain A, a prime ideal p

of A and a proper preordering T of the field of fractions F of A. The following
(i) p is a T ∩ A-convex prime ideal of A;
(ii) there exists an ordering P of F with P ⊇ T such that p is a P ∩ A-convex
prime ideal of A.
Proof (i) ⇒ (ii): By Zorn’s Lemma there is a preordering U ⊇ T of F for which
p is still U ∩ A-convex and which is maximal with respect to this property. Assume
for the sake of contradiction that U is not an ordering of F . Then there exists an
a ∈ F such that a ∈ U ∪ (−U ). Then U1 := U + aU and U2 := U − aU are

proper preorderings of F by Lemma 1.1.8. Let us show that U1 ∩ U2 = U . Take
b ∈ U1 ∩ U2 and write b = u + av = u − av with u, u , v, v ∈ U . Then
b(v + v ) = uv + u v ∈ U,
hence b ∈ U since if v + v = 0, then also v = v = 0. In particular, we have

U ∩ A = (U1 ∩ A) ∩ (U2 ∩ A). By Klapper’s Lemma (Lemma 3.10.8) p is then
convex with respect to one of the Ui ∩ A, which contradicts the maximality of U .
The implication (ii) ⇒ (i) is trivial.
Now we can present the list of equivalent characterizations of the compatibility
of a place with a preordering, announced earlier. As before, oλ denotes the valuation
ring of a place λ of F and mλ denotes its maximal ideal.
Theorem 3.12.24 Let λ : F K ∪ ∞ be a surjective place of F and T a proper
preordering of F . Denote the holomorphy ring of F by o := o(F ). The following
(i) λ is compatible with T ;
(ii) mλ ∩ (1 + T ) = ∅;
(iii) λ is finite on oT (F ) (i.e., oT (F ) ⊆ oλ );
(iv) oλ is T -convex;
(v) there exists an ordering P ⊇ T of F such that oλ is P -convex;
(vi) there exists an element x ∈ H F (T ) such that λ is compatible with x;
(vii) there exists a coarsening μ of λ in M(F ) (i.e., oμ ⊆ oλ ) which is compatible
with T .
For the remaining statements we assume that λ is real. Then o ⊆ oλ , and the prime
ideal p := o∩mλ of o satisfies oλ = op . The following statements are then equivalent
to those above:
(viii) The prime ideal p of o is T ∩ o-convex;
(ix) there exists an element y ∈ H o (T ∩ o) such that supp(y) = p;
(x) p ∩ (1 + T ∩ o) = ∅.
Proof The hard work has already been done. We just need to put everything
together.
The equivalence of (i) and (ii) follows directly from Definition 3.12.19, while
the equivalence of (iii) and (iv) follows from Lemma 3.12.4. By Corollary 3.11.14,
oT (F ) is generated by all elements of the form (1 + t)−1 with t ∈ T , which
gives the equivalence of (ii) and (iii). The equivalence of (v) and (vi) is clear by
Remark 3.12.20(2) (cf. Proposition 2.8.10), and (v) ⇒ (iv) is trivial. Conversely,
assuming (iv), it follows from Lemma 3.12.5 that mλ is convex with respect to
T ∩ oλ . By Lemma 3.12.23 there is then an ordering P ⊇ T of F such that mλ
is convex with respect to P ∩ oλ . A further application of Lemma 3.12.5 gives (v).
(v) ⇒ (vii): If we choose P as in (v) and μ as a place with valuation ring oP (F ), then
μ is compatible with T , coarsens λ, and satisfies μ ∈ M(F ). Conversely, (vii) ⇒
(iii) follows by applying the proven implication (i) ⇒ (iii) to μ. All this establishes
the equivalence of statements (i) to (vii).
Assume now that λ is real, i.e., o ⊆ oλ (cf. Theorem 3.12.5). The equality
oλ = op was already established in Exercise 3.12.3, and the equivalence of (iv)
and (viii) follows from Lemma 3.12.5. The equivalence of (viii) and (ix) follows
from Proposition 3.10.10 and Theorem 3.10.13, and (ii) ⇒ (x) is trivial. Conversely,
assume (x). If q is an ideal of o that is maximal with respect to the property
q ∩ (1 + T ∩ o) = ∅, then q is prime and convex in o with respect to T ∩ o by
Proposition 3.10.5. Hence, oq is T -convex in F by the proven implication (viii) ⇒
(iv). From oq ⊆ op and Lemma 3.12.4 then follows the T -convexity of op = oλ , in
other words, (iv). This proves the theorem.
Proposition 3.12.25 Consider a surjective place λ : F K∪∞ and a preordering

T of F . If λ is compatible with T , then
λ(oT (F )) = oλ(T ) (K).
In particular, λ(o(F )) = o(K) if λ is real.

Proof By Corollary 3.11.14, oT (F ) is generated by all elements of the form
(1 + t)−1 with t ∈ T . The assumption signifies that λ is finite on oT (F ) (cf.
Theorem 3.12.24). However, for t ∈ T we have

1
if t ∈ T ∩ oλ ,
λ 1
1+t = 1+λ(t )
0 if t ∈ oλ .
The first claim then follows since λ(T ) = λ(T ∩ oλ ), while the second claim is the
special case T = F 2 (i.e., λ(T ) = K 2 ), cf. Remark 3.12.20(1).
Since the prime ideals p of o(F ) correspond to the residually real valuation rings
of F (via p → o(F )p , cf. Exercise 3.12.3 and Theorem 3.12.5), the second claim of
Proposition 3.12.25 can also be formulated as follows:
If p is a prime ideal of o(F ), then o(F )/p is the holomorphy ring of the field
of fractions κ(p) of o(F )/p.
Corollary 3.12.26 If A is a residually real Prüfer ring of F and λ : F K ∪ ∞

a surjective place which is integral on A, then λ(A) is a residually real Prüfer ring
of K.
Proof By Theorem 3.12.5 we have o(F ) ⊆ A, hence o(K) ⊆ λ(A) by Proposi-
tion 3.12.25. Then apply Theorem 3.12.5 again.
Proposition 3.12.27 Consider a preordering T of F , a surjective place λ : F

K ∪ ∞ that is compatible with T and an ordering Q of K. Then Q ⊇ λ(T ) if and
only if there exists an ordering P ⊇ T of F that is compatible with λ and such that
Q = λ(P ).
Proof For the nontrivial direction, let Q ⊇ λ(T ) and write Q∗ := Q\{0}. If P is an
ordering of F such that T ∪ λ−1 (Q∗ ) ⊆ P , then P clearly satisfies the claim. Hence
we must show that T ∪ λ−1 (Q∗ ) generates a proper preordering of F . Assume for
the sake of contradiction that this is not the case. Then there exist si ∈ λ−1 (Q∗ ),
ti ∈ T (i = 1, . . . , N) such that
1 + s1 t1 + · · · + sN tN = 0.
Hence there must be at least one i for which λ(ti ) = ∞. Let v denote the valuation
associated to λ and assume without loss of generality that v(t1 ) ≤ · · · ≤ v(tN ).
Then we obtain
t1−1 + s1 + s2 t2 + · · · + sN tN = 0,
where t1−1 and all ti = ti /t1 are in T ∩ oλ . Applying λ to this identity gives a
contradiction.
Corollary 3.12.28 Let A be a residually real Prüfer ring of F and T a preordering

of F . Then H A (T ∩ A) is the closure of H F (T ) in Sper A (where we view Sper F =
(Sper A)min as a subspace of Sper A).
Proof Since H F (T ) ⊆ H A (T ∩ A), we must show that every y ∈ H A (T ∩ A) has
a generalization in H F (T ). Let p := supp(y) and B := Ap . We view y also as a
point in Sper B (with support mB ). Then y ∈ H B (T ∩ B). The claim follows by
applying Proposition 3.12.27 to the place F → κ(B) ∪ ∞ associated to B.
We conclude this section with a discussion of the relationships between the real
spectra of o := o(F ) and other residually real Prüfer rings of F .
Proposition 3.12.29 Let A be a residually real Prüfer ring of F , thus an overring
of o in F , and let Y (A) be the image of Sper A under the restriction map
rA : Sper A → Sper o.
(a) Y (A) is a proconstructible subspace of Sper o that is stable under generaliza-
tion in Sper o, and rA is a homeomorphism from Sper A to Y (A);
(b) Y (A) consists of all x ∈ Sper o that satisfy osupp(x) ⊇ A, or equivalently, such
that A · supp(x) = A (where A · supp(x) is the usual notation for the ideal of A
generated by supp(x));
(c) Y (A) is the union of all Y (B) where B runs through the valuation rings of F
for which A ⊆ B.
Proof Let p be a prime ideal of A. From Exercise 3.12.3 it follows that op∩o = Ap
(both valuation rings have the same centre o). This implies the commutativity of the
diagram
where the horizontal arrows are the bijections of Exercise 3.12.3 (localization). It
follows that the restriction map Spec A → Spec o is injective and that it preserves
the residue fields, from which already follows that rA is injective and that Y (A) =
{x ∈ Sper o : A ⊆ osupp(x)}. For this reason (c) is clear, as is the stability of Y (A)
under generalization. A priori Y (A) is proconstructible. If a = b/c ∈ A with 0 = b
and c ∈ o, then rA (H̊A (a)) = H̊o (bc)∩Y (A). It follows that rA is a homeomorphism
to Y (A). It remains to show the second claim of (b). If q ∈ Spec o and A ⊆ oq , then
Aq ⊆ A ∩ q oq = A. Conversely, if Aq ⊆ p with p ∈ Spec A, then A ⊆ Ap =
op∩o ⊆ oq .
Corollary 3.12.30 If A and A are residually real Prüfer rings of F with Y (A) =
Y (A ), then A = A .
Proof From Proposition 3.12.29 it follows that given a prime ideal q of o, oq
contains A if and only if oq contains A , and that A resp. A is the intersection
of all such oq .
By Proposition 3.12.29 we may identify Sper A with the subspace Y (A) of
Sper o. This space can be described quite explicitly as follows. Let y ∈ Sper F =
(Sper o)min . The specializations x of y in Sper o correspond uniquely to the subrings
B of F that are convex with respect to y, via the map x → osupp(x). (This
follows from Proposition 3.12.29 and Exercise 3.12.3.) This correspondence is
order reversing. Hence y corresponds to the largest such subring (namely B = F )
and the closed specialization y of y in Sper o corresponds to the smallest such
subring (namely B = oy (F )). Given any specialization x of y, it follows from
Proposition 3.12.29(b) that x is in Y (A) = Sper A if and only if A is contained in
osupp(x). In particular, the support ideal of the closed specialization of y in Sper A is
precisely the centre of A · oy (F ) (the y-convex hull of A) in o.
Results that are similar to Propositions 3.12.9–3.12.15 for the holomorphy ring
o(F ) can also be established for the ring A = oT (F ), where T is any proper
preordering of F . We will just indicate these results and leave the proofs to the
reader.
The canonical map Hom(A, R) → Sper A is a bijection onto the subset
Sper A ∩ (Sper o)max

of (Sper A)max . This subset contains the tips of all the spears {x} in Sper o for x ∈
H F (T ), but is larger in general. There is a natural bijective correspondence between
Hom(A, R) and the closed (hence compact) subspace
M(F /T ) := {λ ∈ M(F ) : λ and T are compatible}
of M(F ). A result similar to Proposition 3.12.11 then also holds for A, and
Proposition 3.12.12 can be generalized as follows:
If a ∈ A and
a : Hom (A, R) → R denotes the evaluation map, then
a ≥ 0 if
and only if
a+ 1
n ∈T for all n ∈ N.
Chapter 4
Recent Developments
In this short chapter we briefly discuss a number of important developments and

advances that mostly occurred after the 1989 publication of Einführung in die reelle
Algebra, and that are directly related to topics covered in Chapters 1, 2 and 3.
4.1 Counting Real Solutions
The results from Sects. 1.7–1.10 on counting real zeroes of univariate polynomials
are just the classical tip of a modern iceberg, that has emerged in the past decades.
We can merely sketch the general idea, for more details we have to refer to the
literature. The support supp(f ) of a polynomial f in n variables is the finite set
of n-tuples (α1 , . . . , αn ) ∈ Zn for which the monomial x1α1 · · · xnαn appears in f
with nonzero coefficient. Let a system of n real polynomials fi (x) (i = 1, . . . , n)
in n variables x = (x1 , . . . , xn ) be given. The number of nondegenerate
complex
solutions of the system f1 (x) = · · · = fn (x) = 0 is at most ni=1 deg(fi ) (Bézout),
and for the number of solutions in (C∗ )n there is the Bernstein-Khovanskiı̆-
Kushnirenko (BKK) upper bound in terms of the mixed volume of the convex
hulls of the supports supp(fi ). When it comes to real zeroes of real polynomials,
Khovanskiı̆ discovered around 1980 that what matters for their count is just the
number of different monomials in the fi , not their degrees or the volume of their
convex hull. More precisely, he gave an explicit number ϕ(n, N) with the following
property: Given any real polynomials f1 , . . . , fn in n variables that together have
at most N monomials, the system f1 (x) = · · · = fn (x) has at most ϕ(n, N)
nondegenerate real solutions in the open positive orthant. Note how in the univariate
case such a statement follows from Descartes’ rule of sign (cf. Corollary 1.10.3(a)).
In more than one variable, proofs for results of this kind become a lot more difficult.
This chapter is co-authored by Thomas Unger.

https://doi.org/10.1007/978-3-031-09800-0_4
188 4 Recent Developments
This discovery gave rise to the term fewnomials, and it has meanwhile grown
into a large theory with many ramifications. The bound ϕ(n, N) itself is huge and
is generally believed to be far from best possible. For small values of N − n, much
better or even sharp bounds have been proved. For example, it has been shown for
n = 2 that two real trinomials f1 (x, y), f2 (x, y) can have at most 5 nondegenerate
positive common zeroes, and that this bound is sharp.
For more details we refer to the books by Khovanskiı̆ [58] and Sottile [110].
4.2 Quadratic Forms
Related to Sect. 3.8, some remarkable developments and applications are to be

reported. The Witt ring of quadratic forms, as defined for fields of characteristic
= 2 in Sect. 1.2, can be defined (and is an important object) for arbitrary rings
(or even schemes) on which 2 is invertible [61]. Let A be a ring with 12 ∈ A. A
regular quadratic space over A is a pair ϕ = (M, b) consisting of a finitely generated
projective A-module M and a regular symmetric bilinear form b : M ×M → A. The
sum and product of quadratic spaces are defined as in the field case, using orthogonal
sum and tensor product, respectively. Let W (A) be the Witt-Grothendieck ring of
isomorphism classes of regular quadratic spaces over A. For any finitely generated
projective module M, the hyperbolic space H(M) of M is hM = (M ⊕ M ∨ , bM )
where bM is the canonical symmetric bilinear form on M ⊕ M ∨ . By definition,
the Witt ring W (A) is the quotient ring W (A)/H where H is the ideal in W (A)
consisting of (the classes of) all hyperbolic spaces H(M). Every element of W (A)
is represented by a regular quadratic space ϕ = (M, b).
Over general rings A, Witt’s Cancellation Theorem (see Theorem 1.2.1 in the
field case) usually fails, and Witt decomposition (see loc. cit.) has no analogue. For
semilocal rings A however, Witt cancellation holds, and much of the theory passes
from the field case to this more general situation. See [63] for general background.
We define a signature of A to be any ring homomorphism σ : W (A) → Z, cf.
Definition 1.2.14. By functoriality of the Witt ring, every element α ∈ Sper(A)
gives a signature signα of A. It maps the Witt class of (M, b) to the signature of the
regular quadratic form (M, b) ⊗A k(α) over the real closed field k(α).
Fixing ϕ ∈ W (A), the map sign(ϕ) : Sper(A) → Z is clearly continuous, and
hence constant on connected components of Sper(A). As in Proposition 3.8.2 over
fields, we get a ring homomorphism sign : W (A) → C(Sper(A), Z), the global
signature of A. It turns out that Proposition 3.8.6 remains true in general as well,
but this becomes much harder to prove. In fact, Mahé [78] showed for arbitrary
A that the cokernel of the global signature is a 2-primary torsion group. As a
consequence it follows that every signature σ : W (A) → Z comes from an ordering,
i.e., σ = signα for some α ∈ Sper(A), and that sign : Sper(A) → Hom(W (A), Z)
induces a bijection from the set of connected components of Sper(A) to the set of
all signatures.
Calculating or estimating the exponent of the cokernel of the global signature
sign : W (A) → C(Sper(A), Z) is an important problem. For general rings it is
4.3 Stellensätze 189
very hard. For the coordinate rings A = R[V ] of affine algebraic varieties V over
R, Mahé [79] proved an explicit (although very large) upper bound: There exists
a function w : N → N such that the cokernel of the global signature of R[V ] is
annihilated by 2w(d) whenever dim(V ) ≤ d.
In the case when A is a field, or more generally a semilocal ring, the integer s ≥ 0
(or s = ∞) for which 2s is the exponent of the cokernel of sign is called the stability
index st(A) of A (see Definition 3.8.7). Via the theory of fans, the stability index is
closely related to real valuations of A, and can be understood via powerful local–
global principles. Marshall’s abstract framework of spaces of orderings provides
strong tools for analyzing the stability index and related invariants.
The stability index has found attractive applications in geometry, and Corol-
lary 3.8.13 (the equivalence of (i) and (ii)) already points in this direction. Let V be
an affine R-variety. If M ⊂ V (R) is a basic open semialgebraic set, let s(M) denote
the minimal number r such that M = {x ∈ V (R) : f1 (x) > 0, . . . , fr (x) > 0} for
suitable fi ∈ R[V ], and put s(V ) = sup{s(M) : M ⊂ V (R) basic open}. For basic
closed sets M, define s(M) similarly using nonstrict inequalities fi (x) ≥ 0, and let
s(V ) = sup{s(M) : M ⊂ V (R) basic closed}.
Theorem 4.2.1 Let V be an affine R-variety of dimension d ≥ 1.
(a) (Bröcker–Scheiderer) s(V ) ≤ d,
(b) s(V ) ≤ 12 d(d + 1).
In fact, equality holds in (a) and (b) if V (R) is Zariski-dense in V .
See [16, 99] where much more general versions are proved. Basic closed sets
that actually require d2 (d + 1) inequalities are somewhat pathological; Averkov–
Bröcker [6] proved later that s(P ) ≤ d holds for any d-dimensional polyhedron P
in Rn . Note however that it remains very difficult to produce an explicit description
of minimal (or just small) length, for a general basic (open or closed) set M. Not
surprisingly, the price one has to pay for a short description is that the inequalities
tend to have very large degrees. See e.g., Burési–Mahé [21] for some explicit degree
bounds.
To allow for more flexible applications to geometrical questions, Marshall’s
spaces of orderings were later generalized to abstract real spectra [80] and to spaces
of signs [2]. The book [2] presents an extensive account of such applications,
at varying levels of abstractness (description of semialgebraic, semianalytic and
constructible sets by few inequalities).
4.3 Stellensätze
In Sect. 3.9 we discussed various Stellensätze. This is another topic where remark-
able progress was made since the German edition appeared. The results in Propo-
sitions 3.9.5 and 3.9.8 and Theorem 3.9.11 are of a general nature and hold for
arbitrary preorderings in arbitrary rings. Their common feature is that if a ring
element f takes a particular sign on Sper(A) (e.g., is positive, or nonnegative, or

zero throughout), they guarantee the existence of a “certificate” for this sign.
The Positivstellensatz is a certificate “with denominator”, in the sense that
it has the form sf = 1 + t, where s and t are sums of squares, and thus
expresses f as a quotient of two elements that are strictly positive. Similarly, the
Nichtnegativstellensatz is a certificate with denominator sf = t , where s and t are
sums of squares, and expresses f as a quotient of two elements that are nonnegative.
Making more specific assumptions, one gets much more powerful statements. A
door-opener and breakthrough in this direction was Schmüdgen’s Positivstellensatz:
Theorem 4.3.1 (Schmüdgen) Let polynomials g1 , . . . , gr ∈ R[x] = R[x1 , . . . , xn ]
be given such that the basic closed set K = {x ∈ Rn : gi (x) ≥ 0 (i = 1, . . . , r)} is
compact. Then every polynomial f with f |K > 0 is contained in the preordering
P = P [g1 , . . . , gr ].
(See Examples and Remarks 3.9.2(5) for the notation; nowadays people usually
write P O(g1 , . . . , gr ) instead of P [g1 , . . . , gr ].)
Schmüdgen’s primary interest came from analysis rather than real algebra, more
precisely from the (multivariate) K-moment problem. Given a closed set K ⊂ Rn ,
the question is to characterize all linear forms L : R[x1 , . . . , xn ] → R that are
integration with respect to a suitable (Borel) measure supported on K. Clearly, such
L has to satisfy L(f ) ≥ 0 whenever f |K ≥ 0. Conversely, Theorem 4.3.1 implies
that this condition is already sufficient for K as in the theorem.
In his original paper [103], Schmüdgen deduced Theorem 4.3.1 from his solution
to the K-moment problem. For this he combined operator-theoretic arguments with
Stengle’s Positivstellensatz. Later it was realized that there exists a purely algebraic
approach to Theorem 4.3.1. The key for this is the archimedean property. A convex
subcone M of an R-algebra A is called archimedean if {0, 1} ⊂ M and Z + M = A.
Thinking of the elements of M as being nonnegative in a suitably general sense,
this property requires that every element of A is exceeded by some positive
integer, whence the terminology archimedean. The hypotheses of the theorem imply
that the preordering P [g1 , . . . , gr ] in A = R[x] is archimedean, which makes
Theorem 4.3.1 a consequence of the general Archimedean Positivstellensatz. For
the precise statement of this latter result we refer to the literature. It allows a variety
of substantially different variations on the question of finding denominator-free
certificates for strict positivity. The archimedean Positivstellensatz is essentially
equivalent to the so-called Representation Theorem, usually attributed to Stone,
Kadison, Dubois and Krivine. See [92, Theorem 5.2.6] or [101, 1.5.9], and compare
also [22] for a different approach.
Seeing Theorem 4.3.1 in the archimedean context allows for useful generaliza-
tions. In the situation of the theorem, the quadratic module M := QM(g1 , . . . , gr )
generated by the gi consists of all polynomials f = s0 + ri=1 si gi where the si
are sums of squares of polynomials. Note that M is a subset (usually proper) of the
preordering P . In contrast to P , the cone M need not be archimedean when K is
compact. But when it is, M contains every polynomial f strictly positive on K. The
smaller complexity of a representation in M, compared with one in P (only r + 1
4.3 Stellensätze 191
generators instead of 2r ) is an important point for applications. From Theorem 4.3.1

it follows that M is archimedean if and only if M contains a polynomial g for
which {x ∈ Rn : g(x) ≥ 0} is compact. This fact is often referred to as Putinar’s
Positivstellensatz.
Jacobi and Prestel used the local–global principle for weakly isotropic quadratic
forms (due to Bröcker and Prestel, around 1973) to prove various refinements and
variations of these results. For example, if K = {x ∈ Rn : g1 (x) ≥ 0, g2 (x) ≥ 0} is
compact (the case r = 2), it is always true that M = QM(g1 , g2 ) is archimedean.
For a detailed account of these results we refer to [92], in particular to [92, Chapters
6–8].
Archimedean Nichtnegativstellensätze (as opposed to Positivstellensätze) were
studied by Scheiderer, see e.g., [101, Sect. 3] for an overview. An important tool
in this context is the archimedean local–global principle. It is used, for example,
to show for nonsingular affine R-varieties V with dim(V ) ≤ 2 and V (R) compact,
that every nonnegative polynomial on V is a sum of squares in R[V ], cf. [100].
In contrast, as soon as the dimension is ≥ 3, there always exist nonnegative
polynomials that are not sums of squares. Similar results hold more generally for
basic closed sets under suitable regularity assumptions.
Such theoretical results have important applications of various kinds. For conse-
quences on multivariate moment problems, see e.g. the overview in [101, Sect. 5].
An area where they are omnipresent is polynomial optimization [74]. Roughly since
the year 2000, semidefinite programming techniques started to be used to optimize
polynomial functions under polynomial side conditions. The moment relaxation
method, suggested by Lasserre and Parrilo, allows optimization of a polynomial
over a compact (basic) semialgebraic set K ⊂ Rn up to arbitrary precision, at least
in theory. The basic idea is to relax the positivity condition f |K ≥ c by the condition
f − c ∈ QM(g1 , . . . , gr ), if K = {x : g1 (x) ≥ 0, . . . , gr (x) ≥ 0}. In other words,
being nonnegative is replaced by being a weighted sum of squares. If we bound the
degrees deg(si ) of the sums of squares in representations

f − c = s0 + si gi (∗)
i
as deg(si ) ≤ 2d, then finding the supremum f(d) of all numbers c for which
(∗) exists becomes a standard semidefinite program. Semidefinite programs (SDP)
are solved very efficiently by interior path methods, due to ground-breaking ideas
of Karmarkar, Nesterov and Nemirenko in the 1980s and 1990s. Observe how
the Schmüdgen and Putinar Positivstellensätze imply that the restricted optima
converge to the true optimum when the degree d is taken to infinity. In other words,
f∗ = min f (K) = limd→∞ f(d) . In practice the semidefinite programs become
more and more complex when d is increased. Still, these ideas have become very
important and are now used in a large variety of situations.
4.4 Noncommutative Stellensätze
We should also mention the appearance of various kinds of noncommutative

Stellensätze. To explain the setting, let A be a not necessarily commutative algebra
with unit over a real field F (in many instances F = R) and with an involution
x → x ∗ . In other words, F is contained in the centre of A and x → x ∗ is an anti-
automorphism of order two, i.e., satisfies (x + y)∗ = x ∗ + y ∗ , (x · y)∗ = y ∗ · x ∗
and (x ∗ )∗ = x for all x, y ∈ A. Such an algebra is often called a ∗-algebra. An
element a ∈ A is called symmetric (or hermitian) if a ∗ = a. The analogue of sums
of squares in this context are sums of hermitian squares, i.e., finite sums i xi∗ xi
with xi ∈ A. We denote the set of sums of hermitian squares by A2 . There exist
several approaches to defining what being positive means for elements of A. As this
subject is witnessing an explosive growth (we refer to Schmüdgen’s survey [105]
for an excellent overview of results in this area up to 2009), we can only highlight a
number of representative examples.
The prototypical example of a ∗-algebra is a full matrix algebra A = Mn (F )
with transpose involution ∗ = T . More generally one can consider central simple
algebras with involution. For example, in their 1976 paper [94] Procesi and Schacher
investigated central simple algebras A with involution ∗, centre a field K and fixed
field F := K ∩ {a ∈ A : a ∗ = a} a real field. Given an ordering P of F , the
involution ∗ is defined to be positive at P if the quadratic form x → TrdA (x ∗ x)
is positive semidefinite at P , where TrdA denotes the reduced trace of A, cf. [64,
§1.A]. In this situation P is called a ∗-ordering of F . Consider such a ∗-ordering
P . A symmetric element a ∈ A is defined to be positive at P if the quadratic
form x → TrdA (x ∗ ax) is positive semidefinite at P . A symmetric element a ∈ A is
defined to be positive if it is positive at all ∗-orderings P of F . Procesi and Schacher
proved a noncommutative version of Artin’s Theorem [94, Thm. 5.4]: Let a ∈ A be
any symmetric element. Then a is positive if and only if it is a weighted sum of
hermitian squares,

a= α1ε1 · · · αnεn xi2 ,
ε∈{0,1}n i
where the (finitely many) xi ∈ A are symmetric and α1 , . . . , αn are the elements of
F appearing in a diagonalization of the form x → TrdA (x ∗ x).
They showed that if deg A = 2, i.e., dimK A = 4, i.e., A is a quaternion algebra,
the weights are superfluous ([94, Cor. 5.5]). In general they are not, cf. [59], [60].
See [5] for an approach via signatures of hermitian forms. The weights are also
superfluous if every ordering of F is a ∗-ordering, cf. [94, Prop. 5.3]. This happens
for example in the “split” case where A = Mn (F ) and ∗ = T , as already observed
in 1974 by Gondard and Ribenboim, cf. [41].
In the context of real multivariate matrix polynomials (or polynomial matrices)
A = Mn (R[x1, . . . , xm ]), analogues have been obtained of the Krivine strict
Positivstellensatz by Cimprič [27] in 2011 and the Schmüdgen and Putinar strict
4.4 Noncommutative Stellensätze 193
Positivstellensätze by Scherer and Hol [102] in 2006. These are the equivalences
(i) ⇔ (ii), (i) ⇔ (ii’) and (i) ⇔ (ii”), respectively, in [27, Thm. 2]: Consider a
finite set of polynomials S = {g1 , . . . , gd } in R[x] = R[x1 , . . . , xm ] and let

S = {g1α1 · · · gdαd : α ∈ {0, 1}d }. Consider the basic closed set KS = {x ∈
Rm : g1 (x) ≥ 0, . . . , gd (x) ≥ 0} and the quadratic modules

d
MSn = C0 + Ci gi : C0 , . . . , Cd ∈ Mn (R[x])2 , TSn = M
n
S
, and TS = M
1
S
.
i=1
Let M ∈ Mn (R[x]) be any symmetric matrix. Then: (i) M(x) is strictly positive
definite for every x ∈ KS if and only if (ii) there exist t ∈ TS and V ∈ TSn such that
(1 + t)M = In + V .
Furthermore, if KS is compact, then (i) and (ii) are equivalent to: (ii’) there exists
ε > 0 such that M − εIn ∈ TSn . Finally, if one of the sets K{gi } = {x ∈ Rm : gi (x) ≥
0}, i = 1, . . . , d, is compact, then (i), (ii) and (ii’) are equivalent to: (ii”) there exists
ε > 0 such that M − εIn ∈ MSn .
A generalization is obtained by dropping the requirements that the variables
x1 , . . . , xm commute and that the matrix size is fixed. For example, in his seminal
2002 paper [48], Helton considered the free ∗-algebra A = Rx, x ∗ generated
by m noncommuting variables x = (x1 , . . . , xm ) and their “transposes” x ∗ =
(x1∗ , . . . , xm
∗ ). It is common practice to refer to the elements of A as noncommutative
(or NC) polynomials (or simply polynomials). When substituting matrices for the
variables of a noncommutative polynomial, the involution ∗ is replaced by matrix
transposition T . A symmetric noncommutative polynomial p ∈ A is matrix-
positive provided the matrix p(X, XT ) is positive semidefinite for all matrix tuples
X ∈ Mr (R)m and all r ∈ N. With this definition of positivity, Helton proved the
following noncommutative analogue of Artin’s Theorem [48, Thm. 1.1]: Let p be
any symmetric noncommutative polynomial. Then p is matrix-positive if and only
if it is a sum of hermitian squares.
In [82] McCullough and Putinar pointed out that Helton’s Theorem remains valid
for Cx, x ∗ and that Helton’s proof can easily be adapted to show this. They also
presented a new proof, based upon Carathéodory’s Theorem from convex geometry
and a Hahn–Banach separation argument.
In 2004, Helton and McCullough [50] established a strict Positivstellensatz
for three classes of matrix-valued noncommutative polynomials. To simplify the
exposition we only describe the case of symmetric noncommutative polynomials.
Given a collection P of symmetric polynomials in noncommuting variables x =
(x1 , . . . , xm ), its positivity domain DP is defined as follows: let H be a separable
real Hilbert space and let DP (H ) denote the collection of tuples X = (X1 , . . . , Xm )
such that each Xj is a symmetric operator on H and p(X) is positive semidefinite
for each p ∈ P. Then DP is defined to be the collection of tuples X such that
X ∈ DP (H ) for some H . The positivity domain DP is bounded if there exists a
constant C > 0 such that C 2 − XjT Xj is positive semidefinite whenever X ∈ DP .

Helton and McCullough proved [50, Thm. 1.2]: Assume that DP is bounded with
constant C > 0 and let q be any symmetric polynomial that is strictly positive
definite on DP . Then there exist polynomials pi ∈ P and polynomials si , rj , tk,
such that q can be represented as a finite weighted sum of hermitian squares,

q= siT pi si + rjT rj + T
tk, (C 2 − xk2 )tk, .
i j k,
By changing some assumptions, this result was turned into a Nichtnegativstel-

lensatz by Helton, Klep and McCullough in 2012. They also improved the weighted
sum of squares representation by giving optimal degree bounds, cf. [49]. Pascoe
derived some Positivstellensätze for noncommutative rational expressions from the
Helton–McCullough theorem in 2018, cf. [87].
An exciting aspect of Positivstellensätze for free ∗-algebras is their applications
to engineering problems, cf. [30].
Our final example is Schmüdgen’s 2005 strict Positivstellensatz for the Weyl
algebra W(d) using methods from operator theory and functional analysis, cf. [104].
The set-up is as follows: for d ∈ N, the Weyl algebra W(d) is the complex ∗-algebra
with unit element 1, generators a1 , . . . , ad , a−1 , . . . , a−d , defining relations
ak a−k − a−k ak = 1 for k = 1, . . . , d,

ak a − a ak = 0 for k, = −d, . . . , −1, 1, . . . , d, k = −,
and involution given by ak∗ = a−k for k = 1, . . . , d. The Weyl algebra W(d)
has a natural filtration with corresponding graded algebra the polynomial algebra
C[z, z] in 2d complex variables z = (z1 , . . . , zd ), z = (z1 , . . . , zd ), where zj and
zj correspond to aj and aj∗ , respectively. If c ∈ W(d) is an element of degree n,
the polynomial associated to the n-th component of c is denoted cn (z, z). Let α be
a fixed positive number which is not an integer, and let N denote the set of all finite
products of the elements a1∗ a1 + · · · + ad∗ ad + (α + n)1, where n ∈ Z. Finally,
consider the positive cone
W(d)+ := {x ∈ W(d) : π0 (x)ϕ, ϕ ≥ 0 for all ϕ ∈ S(Rd )},
where π0 denotes the Schrödinger representation, acting on the Schwartz space

S(Rd ), considered as a dense domain of the Hilbert space L2 (Rd ) with inner product
·, ·. Schmüdgen proved [104, Thm. 1.1]: Let c ∈ W(d) be any symmetric element
of even degree 2m. Assume that there exists ε > 0 such that c − ε · 1 ∈ W(d)+ ,
and that c2m (z, z) > 0 for all z ∈ Cd , z = 0. If m is even, then there exists
b ∈ N such that bcb ∈ W(d)2 . If m is odd, then there exists b ∈ N such that
d ∗
j =1 baj caj b ∈ W(d) .
2
References
1. Andradas, C., Bröcker, L., Ruiz, J.M.: Minimal generation of basic open semianalytic sets.
Invent. Math. 92(2), 409–430 (1988). https://doi.org/10.1007/BF01404461
2. Andradas, C., Bröcker, L., Ruiz, J.M.: Constructible sets in real geometry, Ergebnisse der
Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)],
vol. 33. Springer, Berlin (1996). https://doi.org/10.1007/978-3-642-80024-5
3. Artin, E.: Über die Zerlegung definiter Funktionen in Quadrate. Abh. Math. Sem. Univ.
Hamburg 5(1), 100–115 (1927). https://doi.org/10.1007/BF02952513
4. Artin, E., Schreier, O.: Algebraische Konstruktion reeller Körper. Abh. Math. Sem. Univ.
Hamburg 5(1), 85–99 (1927). https://doi.org/10.1007/BF02952512
5. Astier, V., Unger, T.: Signatures of Hermitian forms, positivity, and an answer to a question
of Procesi and Schacher. J. Algebra 508, 339–363 (2018). https://doi.org/10.1016/j.jalgebra.
2018.05.004
6. Averkov, G., Bröcker, L.: Minimal polynomial descriptions of polyhedra and special semi-
algebraic sets. Adv. Geom. 12(3), 447–459 (2012). https://doi.org/10.1515/advgeom-2011-
059
7. Becker, E.: The real holomorphy ring and sums of 2nth powers. In: Real Algebraic Geometry
and Quadratic Forms (Rennes, 1981). Lecture Notes in Math., vol. 959, pp. 139–181.
Springer, Berlin-New York (1982)
8. Becker, E.: Valuations and real places in the theory of formally real fields. In: Real Algebraic
Geometry and Quadratic Forms (Rennes, 1981). Lecture Notes in Math., vol. 959, pp. 1–40.
Springer, Berlin-New York (1982)
9. Becker, E., Köpping, E.: Reduzierte quadratische Formen und Semiordnungen reeller Körper.
Abh. Math. Sem. Univ. Hamburg 46, 143–177 (1977). https://doi.org/10.1007/BF02993018
10. Becker, E., Spitzlay, K.J.: Zum Satz von Artin-Schreier über die Eindeutigkeit des reellen
Abschlusses eines angeordneten Körpers. Comment. Math. Helv. 50, 81–87 (1975). https://
doi.org/10.1007/BF02565735
11. Bochnak, J., Coste, M., Roy, M.F.: Géométrie algébrique réelle. Ergebnisse der Mathematik
und ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)], vol. 12. Springer,
Berlin (1987)
12. Bourbaki, N.: Algèbre. Masson, Paris
13. Bourbaki, N.: Algèbre Commutative. Masson, Paris
14. Bourbaki, N.: Topologie Générale. Masson, Paris
15. Bröcker, L.: Zur Theorie der quadratischen Formen über formal reellen Körpern. Math. Ann.
210, 233–256 (1974). https://doi.org/10.1007/BF01350587

https://doi.org/10.1007/978-3-031-09800-0
196 References
16. Bröcker, L.: On the stability index of Noetherian rings. In: Real Analytic and Algebraic
Geometry (Trento, 1988). Lecture Notes in Math., vol. 1420, pp. 72–80. Springer, Berlin
(1990). https://doi.org/10.1007/BFb0083912
17. Brumfiel, G.W.: Partially ordered rings and semi-algebraic geometry. London Mathematical
Society Lecture Note Series, vol. 37. Cambridge University Press, Cambridge-New York
(1979)
18. Brumfiel, G.W.: Real valuation rings and ideals. In: Real Algebraic Geometry and Quadratic
Forms (Rennes, 1981). Lecture Notes in Math., vol. 959, pp. 55–97. Springer, Berlin-New
York (1982)
19. Brumfiel, G.W.: Witt Rings and K-Theory. pp. 733–765 (1984). https://doi.org/10.1216/RMJ-
1984-14-4-733. Ordered fields and real algebraic geometry (Boulder, Colo., 1983)
20. Brumfiel, G.W.: The real spectrum compactification of Teichmüller space. In: Geometry of
Group Representations (Boulder, CO, 1987). Contemp. Math., vol. 74, pp. 51–75. Amer.
Math. Soc., Providence (1988). https://doi.org/10.1090/conm/074/957511
21. Burési, J., Mahé, L.: Reducing inequalities with bounds. Math. Z. 227(2), 231–243 (1998).
https://doi.org/10.1007/PL00004371
22. Burgdorf, S., Scheiderer, C., Schweighofer, M.: Pure states, nonnegative polynomials and
sums of squares. Comment. Math. Helv. 87(1), 113–140 (2012). https://doi.org/10.4171/
CMH/250
23. Carral, M., Coste, M.: Normal spectral spaces and their dimensions. J. Pure Appl. Algebra
30(3), 227–235 (1983). https://doi.org/10.1016/0022-4049(83)90058-0
24. Cassels, J.W.S., Ellison, W.J., Pfister, A.: On sums of squares and on elliptic curves
over function fields. J. Number Theory 3, 125–149 (1971).https://doi.org/10.1016/0022-
314X(71)90030-8
25. Chaillou, J.: Hyperbolic differential polynomials and their singular perturbations. Mathe-
matics and its Applications, vol. 3. D. Reidel Publishing Co., Dordrecht-Boston (1979).
Translated from the French by J. W. Nienhuys
26. Choi, M.D., Lam, T.Y.: An old question of Hilbert. In: Conference on Quadratic Forms—
1976 (Proc. Conf., Queen’s Univ., Kingston, Ont., 1976), pp. 385–405. Queen’s Papers in
Pure and Appl. Math., No. 46 (1977)
27. Cimprič, J.: Strict Positivstellensätze for matrix polynomials with scalar constraints. Linear
Algebra Appl. 434(8), 1879–1883 (2011). https://doi.org/10.1016/j.laa.2010.11.046
28. Coste, M., Roy, M.F.: La topologie du spectre réel. In: Ordered Fields and Real Algebraic
Geometry (San Francisco, Calif., 1981). Contemp. Math., vol. 8, pp. 27–59. Amer. Math.
Soc., Providence (1982)
29. Coste-Roy, M.F.: Faisceau structural sur le spectre réel et fonctions de Nash. In: Real
Algebraic Geometry and Quadratic Forms (Rennes, 1981). Lecture Notes in Math., vol. 959,
pp. 406–432. Springer, Berlin-New York (1982)
30. de Oliveira, M.C., Helton, J.W., McCullough, S.A., Putinar, M.: Engineering systems and free
semi-algebraic geometry. In: Emerging Applications of Algebraic Geometry. IMA Vol. Math.
Appl., vol. 149, pp. 17–61. Springer, New York (2009). https://doi.org/10.1007/978-0-387-
09686-5_2
31. Delfs, H., Knebusch, M.: Semialgebraic topology over a real closed field. I. Paths and
components in the set of rational points of an algebraic variety. Math. Z. 177(1), 107–129
(1981). https://doi.org/10.1007/BF01214342
32. Delfs, H., Knebusch, M.: Semialgebraic topology over a real closed field. II. Basic theory of
semialgebraic spaces. Math. Z. 178(2), 175–213 (1981). https://doi.org/10.1007/BF01262039
33. Delfs, H., Knebusch, M.: Locally semialgebraic spaces. Lecture Notes in Mathematics,
vol. 1173. Springer-Verlag, Berlin (1985). https://doi.org/10.1007/BFb0074551
34. Delzell, C.N., Madden, J.J.: A completely normal spectral space that is not a real spectrum. J.
Algebra 169(1), 71–77 (1994). https://doi.org/10.1006/jabr.1994.1272
35. Dickmann, M., Schwartz, N., Tressl, M.: Spectral spaces. New Mathematical Mono-
graphs, vol. 35. Cambridge University Press, Cambridge (2019). https://doi.org/10.1017/
9781316543870
References 197
36. Dress, A.: Lotschnittebenen mit halbierbarem rechtem Winkel. Arch. Math. (Basel) 16, 388–
392 (1965). https://doi.org/10.1007/BF01220047
37. Dubois, D.W.: Infinite primes and ordered fields. Dissertationes Math. (Rozprawy Mat.) 69,
43 (1970)
38. Elman, R., Lam, T.Y., Wadsworth, A.R.: Orderings under field extensions. J. Reine Angew.
Math. 306, 7–27 (1979)
39. Fuller, A.T.: Aperiodicity determinants expressed in terms of roots. Int. J. Control 47(6),
1571–1593 (1988). https://doi.org/10.1080/00207178808906122
40. Gantmacher, F.R.: Matrizentheorie. Springer-Verlag, Berlin (1986). https://doi.org/10.1007/
978-3-642-71243-2. With an appendix by V. B. Lidskij, With a preface by D. P. Želobenko,
Translated from the second Russian edition by Helmut Boseck, Dietmar Soyka and Klaus
Stengert
41. Gondard, D., Ribenboim, P.: Le 17e problème de Hilbert pour les matrices. Bull. Sci. Math.
(2) 98(1), 49–56 (1974)
42. Gorin, E.A.: Asymptotic properties of polynomials and algebraic functions of several
variables. Uspehi Mat. Nauk 16(1 (97)), 91–118 (1961)
43. Grätzer, G.: General Lattice Theory. Birkhäuser Verlag, Basel-Stuttgart (1978). Lehrbücher
und Monographien aus dem Gebiete der Exakten Wissenschaften, Mathematische Reihe,
Band 52
44. Harrison, D.K.: Witt Rings. Lecture Notes. Univ. of Kentucky, Lexington (1970)
45. Hartshorne, R.: Algebraic Geometry. Springer-Verlag, New York-Heidelberg (1977). Gradu-
ate Texts in Mathematics, No. 52
46. Helmke, U.: Rational functions and Bezout forms: a functorial correspondence. Linear
Algebra Appl. 122/123/124, 623–640 (1989). https://doi.org/10.1016/0024-3795(89)90669-
1
47. Helmke, U., Fuhrmann, P.A.: Bezoutians. Linear Algebra Appl. 122/123/124, 1039–1097
(1989). https://doi.org/10.1016/0024-3795(89)90684-8
48. Helton, J.W.: “Positive” noncommutative polynomials are sums of squares. Ann. Math. (2)
156(2), 675–694 (2002). https://doi.org/10.2307/3597203
49. Helton, J.W., Klep, I., McCullough, S.: The convex Positivstellensatz in a free algebra. Adv.
Math. 231(1), 516–534 (2012). https://doi.org/10.1016/j.aim.2012.04.028
50. Helton, J.W., McCullough, S.A.: A Positivstellensatz for non-commutative polynomials.
Trans. Am. Math. Soc. 356(9), 3721–3737 (2004). https://doi.org/10.1090/S0002-9947-04-
03433-6
51. Hochster, M.: Prime ideal structure in commutative rings. Trans. Am. Math. Soc. 142, 43–60
(1969). https://doi.org/10.2307/1995344
52. Hölder, O.: Die Axiome der Quantität und die Lehre vom Maß. Leipz. Ber. 53, 1–64 (1901)
53. Hörmander, L.: On the division of distributions by polynomials. Ark. Mat. 3, 555–568 (1958).
https://doi.org/10.1007/BF02589517
54. Jacobson, N.: Lectures in Abstract Algebra (3 volumes). D. Van Nostrand Co., Inc., Toronto,
New York, London (1951–1954)
55. Jacobson, N.: Basic Algebra (2 volumes). W. H. Freeman and Co., San Francisco, Calif. (1974
and 1980)
56. Johnstone, P.T.: Stone Spaces. Cambridge Studies in Advanced Mathematics, vol. 3. Cam-
bridge University Press, Cambridge (1982)
57. Kato, K.: A Hasse principle for two-dimensional global fields. J. Reine Angew. Math. 366,
142–183 (1986). https://doi.org/10.1515/crll.1986.366.142. With an appendix by Jean-Louis
Colliot-Thélène
58. Khovanskiı̆, A.G.: Fewnomials. Translations of Mathematical Monographs, vol. 88. American
Mathematical Society, Providence (1991). https://doi.org/10.1090/mmono/088. Translated
from the Russian by Smilka Zdravkovska
59. Klep, I., Unger, T.: The Procesi-Schacher conjecture and Hilbert’s 17th problem for algebras
with involution. J. Algebra 324(2), 256–268 (2010). https://doi.org/10.1016/j.jalgebra.2010.
03.022
198 References
60. Klep, I., Špenko, v., Volčič, J.: Positive trace polynomials and the universal Procesi-Schacher
conjecture. Proc. Lond. Math. Soc. (3) 117(6), 1101–1134 (2018). https://doi.org/10.1112/
plms.12156
61. Knebusch, M.: Symmetric bilinear forms over algebraic varieties. In: Conference on
Quadratic Forms—1976 (Proc. Conf., Queen’s Univ., Kingston, Ont., 1976), pp. 103–283.
Queen’s Papers in Pure and Appl. Math., No. 46 (1977)
62. Knebusch, M., Kolster, M.: Wittrings. Aspects of Mathematics, vol. 2. Friedr. Vieweg & Sohn,
Braunschweig (1982)
63. Knus, M.A.: Quadratic and Hermitian Forms Over Rings. Grundlehren der mathematischen
Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 294. Springer-
Verlag, Berlin (1991). https://doi.org/10.1007/978-3-642-75401-2. With a foreword by I.
Bertuccioni
64. Knus, M.A., Merkurjev, A., Rost, M., Tignol, J.P.: The Book of Involutions. American
Mathematical Society Colloquium Publications, vol. 44. American Mathematical Society,
Providence (1998). https://doi.org/10.1090/coll/044. With a preface in French by J. Tits
65. Kreı̆n, M.G., Naı̆mark, M.A.: The method of symmetric and Hermitian forms in the theory
of the separation of the roots of algebraic equations. Linear and Multilinear Algebra 10(4),
265–308 (1981). https://doi.org/10.1080/03081088108817420. Translated from the Russian
by O. Boshko and J. L. Howland
66. Krivine, J.L.: Anneaux préordonnés. J. Anal. Math. 12, 307–326 (1964). https://doi.org/10.
1007/BF02807438
67. Krull, W.: Allgemeine Bewertungstheorie. J. Reine Angew. Math. 167, 160–196 (1932).
https://doi.org/10.1515/crll.1932.167.160
68. Kunz, E.: Einführung in die kommutative Algebra und algebraische Geometrie. Vieweg
Studium: Aufbaukurs Mathematik [Vieweg Studies: Mathematics Course], vol. 46. Friedr.
Vieweg & Sohn, Braunschweig (1980). With an English preface by David Mumford
69. Lam, T.Y.: The algebraic theory of quadratic forms. W. A. Benjamin, Inc., Reading (1973).
Mathematics Lecture Note Series
70. Lam, T.Y.: Orderings, Valuations and Quadratic Forms. CBMS Regional Conference Series
in Mathematics, vol. 52. Published for the Conference Board of the Mathematical Sciences,
Washington, DC; by the American Mathematical Society, Providence (1983). https://doi.org/
10.1090/cbms/052
71. Lam, T.Y.: An introduction to real algebra. Rocky Mount. J. Math. 14(4), 767–814 (1984).
https://doi.org/10.1216/RMJ-1984-14-4-767. Ordered fields and real algebraic geometry
(Boulder, Colo., 1983)
72. Lam, T.Y.: Introduction to Quadratic Forms Over Fields. Graduate Studies in Mathematics,
vol. 67. American Mathematical Society, Providence (2005)
73. Lang, S.: Algebra. Graduate Texts in Mathematics, vol. 211, 3rd edn. Springer, New York
(2002). https://doi.org/10.1007/978-1-4613-0041-0
74. Lasserre, J.B.: An introduction to polynomial and semi-algebraic optimization. Cambridge
Texts in Applied Mathematics. Cambridge University Press, Cambridge (2015). https://doi.
org/10.1017/CBO9781107447226
75. Leicht, J.B.: Zur Charakterisierung reell abgeschlossener Körper. Monatsh. Math. 70, 452–
453 (1966). https://doi.org/10.1007/BF01300449
76. Łojasiewicz, S.: Sur le problème de la division. Stud. Math. 18, 87–136 (1959). https://doi.
org/10.4064/sm-18-1-87-136
77. Lorenz, F., Leicht, J.: Die Primideale des Wittschen Ringes. Invent. Math. 10, 82–88 (1970).
https://doi.org/10.1007/BF01402972
78. Mahé, L.: Signatures et composantes connexes. Math. Ann. 260(2), 191–210 (1982). https://
doi.org/10.1007/BF01457236
79. Mahé, L.: Théorème de Pfister pour les variétés et anneaux de Witt réduits. Invent. Math.
85(1), 53–72 (1986). https://doi.org/10.1007/BF01388792
80. Marshall, M.A.: Spaces of Orderings and Abstract Real Spectra. Lecture Notes in Mathemat-
ics, vol. 1636. Springer, Berlin (1996). https://doi.org/10.1007/BFb0092696
References 199
81. Matsumura, H.: Commutative Algebra. Mathematics Lecture Note Series, vol. 56, 2nd edn.
Benjamin/Cummings Publishing Co., Inc., Reading (1980)
82. McCullough, S., Putinar, M.: Noncommutative sums of squares. Pac. J. Math. 218(1), 167–
171 (2005). https://doi.org/10.2140/pjm.2005.218.167
83. Milnor, J.: Algebraic K-theory and quadratic forms. Invent. Math. 9, 318–344 (1969/1970).
https://doi.org/10.1007/BF01425486
84. Motzkin, T.S.: The arithmetic-geometric inequality. In: Inequalities (Proc. Sympos. Wright-
Patterson Air Force Base, Ohio, 1965), pp. 205–224. Academic Press, New York (1967)
85. Obreschkoff, N.: Verteilung und Berechnung der Nullstellen reeller Polynome. VEB
Deutscher Verlag der Wissenschaften, Berlin (1963)
86. Ostrowski, A.: Über einige Lösungen der Funktionalgleichung ϕ(x) · ϕ(x) = ϕ(xy). Acta
Math. 41(1), 271–284 (1916). https://doi.org/10.1007/BF02422947
87. Pascoe, J.E.: Positivstellensätze for noncommutative rational expressions. Proc. Am. Math.
Soc. 146(3), 933–937 (2018). https://doi.org/10.1090/proc/13773
88. Pfister, A.: Zur Darstellung definiter Funktionen als Summe von Quadraten. Invent. Math. 4,
229–237 (1967). https://doi.org/10.1007/BF01425382
89. Pourchet, Y.: Sur la représentation en somme de carrés des polynômes à une indéterminée sur
un corps de nombres algébriques. Acta Arith. 19, 89–104 (1971). https://doi.org/10.4064/aa-
19-1-89-104
90. Prestel, A.: Lectures on Formally Real Fields. Lecture Notes in Mathematics, vol. 1093.
Springer, Berlin (1984). https://doi.org/10.1007/BFb0101548
91. Prestel, A.: Einführung in die Mathematische Logik und Modelltheorie. Vieweg Studium:
Aufbaukurs Mathematik [Vieweg Studies: Mathematics Course], vol. 60. Friedr. Vieweg &
Sohn, Braunschweig (1986). https://doi.org/10.1007/978-3-663-07641-4
92. Prestel, A., Delzell, C.N.: Positive Polynomials. Springer Monographs in Mathematics.
Springer, Berlin (2001). https://doi.org/10.1007/978-3-662-04648-7. From Hilbert’s 17th
problem to real algebra
93. Prieß-Crampe, S.: Angeordnete Strukturen. Ergebnisse der Mathematik und ihrer Grenzge-
biete [Results in Mathematics and Related Areas], vol. 98. Springer, Berlin (1983). https://
doi.org/10.1007/978-3-642-68628-3. Gruppen, Körper, projektive Ebenen. [Groups, fields,
projective planes]
94. Procesi, C., Schacher, M.: A non-commutative real Nullstellensatz and Hilbert’s 17th
problem. Ann. Math. (2) 104(3), 395–406 (1976). https://doi.org/10.2307/1970962
95. Robson, R.O.: The Ideal Theory of Real Algebraic Curves and Affine Embeddings of Semi-
Algebraic Spaces and Manifolds. Ph.D. Thesis, Stanford University (1981)
96. Roman, S.: Field Theory. Graduate Texts in Mathematics, vol. 158, 2nd edn. Springer, New
York (2006)
97. Schafarewitsch, I.R.: Grundzüge der algebraischen Geometrie. Friedr. Vieweg+Sohn, Braun-
schweig (1972). Übersetzung aus dem Russischen von Rudolf Fragel, Logik und Grundlagen
der Mathematik, Band 12
98. Scharlau, W.: Quadratic and Hermitian Forms. Grundlehren der Mathematischen Wis-
senschaften [Fundamental Principles of Mathematical Sciences], vol. 270. Springer, Berlin
(1985). https://doi.org/10.1007/978-3-642-69971-9
99. Scheiderer, C.: Stability index of real varieties. Invent. Math. 97(3), 467–483 (1989). https://
doi.org/10.1007/BF01388887
100. Scheiderer, C.: Sums of squares on real algebraic surfaces. Manuscr. Math. 119(4), 395–410
(2006). https://doi.org/10.1007/s00229-006-0630-5
101. Scheiderer, C.: Positivity and sums of squares: a guide to recent results. In: Emerging
Applications of Algebraic Geometry. IMA Vol. Math. Appl., vol. 149, pp. 271–324. Springer,
New York (2009). https://doi.org/10.1007/978-0-387-09686-5_8
102. Scherer, C.W., Hol, C.W.J.: Matrix sum-of-squares relaxations for robust semi-definite
programs. Math. Program. 107(1-2, Ser. B), 189–211 (2006). https://doi.org/10.1007/s10107-
005-0684-2
200 References
103. Schmüdgen, K.: The K-moment problem for compact semi-algebraic sets. Math. Ann. 289(2),
203–206 (1991). https://doi.org/10.1007/BF01446568
104. Schmüdgen, K.: A strict Positivstellensatz for the Weyl algebra. Math. Ann. 331(4), 779–794
(2005). https://doi.org/10.1007/s00208-004-0604-4
105. Schmüdgen, K.: Noncommutative real algebraic geometry—some basic concepts and first
ideas. In: Emerging Applications of Algebraic Geometry. IMA Vol. Math. Appl., vol. 149,
pp. 325–350. Springer, New York (2009). https://doi.org/10.1007/978-0-387-09686-5_9
106. Schülting, H.W.: Über reelle Stellen eines Körpers und ihren Holomorphiering. Ph.D. Thesis,
Univ. Dortmund (1979)
107. Schwartz, N.: The Basic Theory of Real Closed Spaces. Regensburger Mathematische
Schriften [Regensburg Mathematical Publications], vol. 15. Universität Regensburg, Fach-
bereich Mathematik, Regensburg (1987)
108. Schwartz, N.: The basic theory of real closed spaces. Mem. Am. Math. Soc. 77(397), viii+122
(1989). https://doi.org/10.1090/memo/0397
109. Serre, J.P.: Groupes algébriques et corps de classes. Hermann, Paris (1975). Deuxième édition,
Publication de l’Institut de Mathématique de l’Université de Nancago, No. VII, Actualités
Scientifiques et Industrielles, No. 1264
110. Sottile, F.: Real Solutions to Equations from Geometry. University Lecture Series, vol. 57.
American Mathematical Society, Providence (2011). https://doi.org/10.1090/ulect/057
111. Stengle, G.: A nullstellensatz and a positivstellensatz in semialgebraic geometry. Math. Ann.
207, 87–97 (1974). https://doi.org/10.1007/BF01362149
112. Swan, R.G.: Topological examples of projective modules. Trans. Am. Math. Soc. 230, 201–
234 (1977). https://doi.org/10.2307/1997717
113. van der Waerden, B.L.: Algebra (2 volumes). Heidelberger Taschenbücher. Springer, Berlin-
Heidelberg-New York (1971 and 1967)
114. Weber, H.: Lehrbuch der Algebra (3 Volumes). Reprint: Chelsea Publ. Comp., New York
(1963)
Symbol Index
The set of natural numbers {1, 2, 3, . . .} is denoted N, while N0 := N ∪ {0}. As

usual, Z, Q, R and C denote the sets of integers, rational numbers, real numbers and
complex numbers, respectively. For a prime power q, the field with q elements is
denoted Fq .
Given a set X, the notation #X is used for the number of elements of X in a naive
sense: #X ∈ N0 ∪ {∞}. The power set of X is denoted 2X . If Y and Z are subsets
of X, we let Y \ Z := {y ∈ Y : y ∈ Z}.
Rings are always assumed commutative and unitary. Subrings contain the identity
and all ring homomorphisms preserve the identity. Given a ring A, the group of units
is denoted A∗ and the subset {a 2 : a ∈ A} of A is denoted A2 . If A is an integral
domain, we write Quot A for the field of fractions of A. The transcendence degree
of a field extension L/K is denoted tr.deg.(L/K) ∈ N0 ∪ {∞}, see also Sect. 2.11.
K 2 , 2 Var(c0 , . . . , cr ), 21
signP a, 4 V (x), 21
a1 , . . . , an , 5 sr (f ), 23
H, 5 S(f ), 24
ϕ ⊥ϕ ,ϕ⊗ϕ ,5 Sλ (f ), 25
n · ϕ, ϕ ⊗n , 6 trA/K , 26
U ⊥, 6 [A, a]K , 27
Rad (ϕ), 6 SylK (f ; g), 27
rank (ϕ), 6 indα (ϕ), 30
det ϕ, 6 Iab (ϕ), 30
dim ϕ, 6 B(f, g), 32
ϕ ∼ϕ,7 H (f, g), 32
[ϕ], 7 R(f, g), 35
W (K), 7 s∗ ϕ, 48
signP ϕ, signP , sign, 8 (1 × · · · × r )lex , 55
e : W (K) → Z/2Z, 10 oP (K/A), o(K/A), 55
I (K), 10 oP (K), o(K), 55
ϕL , 10 mA , 58
iL/K , 10 κ(A), 58
τ | σ , 11 ov , mv , 64

https://doi.org/10.1007/978-3-031-09800-0
202 Symbol Index
A , vA , 65 x % y, 116
rank , rank A, 69 Sper ϕ, ϕ ∗ , 116
TP , Tv , 74 ϕβ , 117
λA , 79 K(Sper A), 121
oλ , mλ , κλ , 80 K̊(Sper A), K(Sper A), 121
λ , vλ , 80 K(X), K̊(X), K(X), 124
R{t},
√ 85 Xcon , 124
a, 101 X ∗ , 126
Ared , 101 S(M), S̊(M), S(M), 128
Spec A, 102 128
M,
κA (p), κ(p), 102 Filt(L), Prime(L), Ultra(L), 130
ρp : A → κ(p), 102 pro-L, 130
f (p), 102 X max , X min , 131
DA (f ), DA (T ), 103 ρX , ρ, 139
VA (f ), VA (T ), 103 M c , Ac , ac , 142
I(Y ), 103 HF (a1 , . . . , an ), 145
K̊(Spec A), 105 sign ϕ, sign, 146
Spec ϕ, ϕ ∗ , 105 W (F ), I (F ), 146
ϕq , 105 ϕ, 146
f −∞ A, 106 χY , 146
n
V (K), 107 I n (F ), I (F ), 147
k[V ], 107 st(F ), 147
VA , VA (K), 108 rK/F , 149
ϕ ∗ , 108 P [F ], 151
A2 , 110 a ≤T b, 159
(Spec
√ A)re , 111 T, 161
re
a, 113 P (X), 162
Sper A, 114 162
X,
supp α, 114 ciT (a), 164
k(α), 114 oX (A/ ), ox (A/ ), 166
f (α), 114 oX (A), o(A), 167
k(P ), 116 oT (A/ ), oT (A), 169
f (P ), 116 !(a/ ), 172
H̊A (T ), H̊A (f1 , . . . , fr ), 114 M(F ), 178
H A (T ), H A (f1 , . . . , fr ), 114
a , 178
ZA (T ), ZA (f1 , . . . , fr ), 114 λ(T ), 180
H̊A , 114 M(F /T ), 186
Index
A in Sper A, 135
Absolutely bounded, 167 Constructible
Absolute value sets, 121, 124
on a field, 74 topology, 121, 124
ultrametric, 74 Convex
Affine algebra of an affine variety, 107 hull, 53
Affine k-variety, 107 subgroup, 54
Almost monic polynomial, 168 subring, 58
Anisotropic quadratic form, 6 subset, 53
Archimedean
closed subring, 166
over a subgroup, subring, . . . , 54, 55, 166 D
saturated subfield, 71 Dedekind cut (generalized), 84
Archimedean closure of a subring, 166 free, 84
Archimedes, axiom of, 55 proper, 57, 84
Artin–Lang Stellensatz, 92, 93, 113 Definite (positive, negative), 8, 95
Descartes, rule of, 42
Determinant of a quadratic form, 6
B Dimension index, 10
Bézout matrix, 32 Discrete valuation ring of rank one, 69
Bézoutian, 32 Distributive lattice, 135
Bilinear space, 5 Domination of local rings, 62
Boolean algebra, 120
Boolean lattice, 120
Boolean space, 115, 125 E
Bounded (over a subring), 166 Extension of
orderings, 11
places (theorem of Chevalley), 80
C real places, 82
Cauchy index, 30 signatures, 11
Centre
of a place, 141
of a valuation ring, 141 F
Closed points Field
in Spec A, 103 ordered, 1

https://doi.org/10.1007/978-3-031-09800-0
204 Index
real closed, 15, 17 Kernel form, 6

Filter, 130
Finiteness theorem, 129
Finite points, 137 L
Function field, 89 Lang’s embedding theorem, 89
Fundamental ideal, 10 Lattice, 120
Fundamental theorem of algebra, 16 Boolean, 120
well-behaved, 130
Lexicographic order, 55
G Local ring, 58
Generalization, 116 Łojasiewicz inequality, 158
Generic point, 104
M
H Maximally
Hankel form, 32 real ideal, 111
Hankel matrix, 32 real reduced ideal, 111
Harrison Maximal points, 131, 136
subbasis, 114 Minimal points, 131
topology, 114
Hilbert
Nullstellensatz, 108 N
17th problem, 96 Nichtnegativstellensatz, 154
solution of 17th problem, 97 Nilradical, 101
Holomorphy ring of a field, 167 Nonreal ring, 110
Hyperbolic Nullstellensatz
form, 6 Hilbert’s, 108
plane, 5 real, 113, 154
weak, 104
weak real, 112
I
Ideal
P -convex (P an ordering), 140 O
T -convex (T a preordering), 159 Ordered abelian group, 53
X-convex (X a subspace of Sper A), 162 rank, 69
Indefinite, 95 Ordered field, 1
inequality of Łojasiewicz, 158 Ordering
Inertia theorem of Sylvester, 8 of a field, 1
Infinitely compatible with a valuation, 66
distant points, 137, 173 of a ring, 115
large, 54, 55 Order preserving
small, 54, 55 homomorphism, 54
Integral closure, 61 place, 80
Integral element, 60 Order toplogy, 73
Integrally closed ring, 61
Integral ring extension, 60
Intermediate value theorem, 20 P
Inverse spectral space, 126 P -convex ideal, 140
Isotropic quadratic form, 6 Pfister form, 147
Place (of a field), 78
associated with a valuation ring, 79
K canonical, 79
k-morphism (of affine k-varieties), 107 compatible with a preordering, 180
K-rational points, 107, 109 compatible with an ordering, 80
Index 205
real, 80 ring, 110

trivial, 79 Reduced ring, 101
Positivstellensatz, 154 Residually real valuation ring, 58
Preordering Resultant, 35
generated, 151 Retraction, canonical, 139
improper, 150
of a field, 2
proper, 150 S
of a ring, 150 SAP field, 148
saturated, 161 Semialgebraic sets, 127
Prime filter, 130 Semidefinite, 95
Probasic subset, 162 Sign change criterion, 97
Proconstructible subset, 122, 126 Signature
Projection formula, 49 of a field, 9
Pro-L-sets, 130 of a quadratic form, 8
Prüfer ring, 174 total (of a field), 146
residually real, 174 total (of a form), 146
Space of places of a field, 178
Spear, 137
Q Specialization, 116
Quadratic form, 5 Spectral map, 124
anisotropic, 6 Spectral space, 124
definite (positive, negative), 8 inverse, 126
degenerate, 6 normal, 139
hyperbolic, 6 Spectrum of a ring, 102
isotropic, 6 real, 114
nondegenerate, 6 Stability index, 147
universal, 6 Stellensatz of Artin–Lang, 92, 93, 113
Quadratic space, 5 Stone duality, 135
Stone space of a distributive lattice, 135
Strong topology, 73, 119
R Sturm algorithm, 21
Radical Sturm sequence, 21
of an ideal, 101 generalized, 22
of a quadratic space, 6 Subgroup
real, 113 convex, 54
Rank T -convex (T a preordering), 159
of an ordered abelian group, 69 X-convex (X a subspace of Sper A), 162
of a valuation ring, 69 Subring
Real convex, 55
closure of a field, 46 T -convex (T a preordering), 159
closure, uniqueness, 46 X-convex (X a subspace of Sper A), 162
field, 3 Support (of a point in Sper A), 114
holomorphy ring of a field, 167 Sylvester form, 24
ideal, 110 with parameter λ, 25
place, 80 Sylvester’s inertia theorem, 8
radical, 113
ring, 110
spectrum, 114 T
Real closed Tarski’s principle, x, 39
field, 15, 17 T -convex (T a preordering), 159
valuation ring, 142 Theorem of
Real reduced Artin–Lang, 92, 93, 113
ideal, 110 Baer–Krull, 76, 78
206 Index
Budan–Fourier, 41 coarsening, 65
Chevalley (extension of places), 80 compatible with an ordering, 66
Cohen–Seidenberg, 61 equivalent, 65
Dress, 175 non-trivial, 64
Hölder, 56 trivial, 64
Hörmander, 156 value group, 64
Hermite, 25, 29 Valuation ring, 58
Hurwitz (different ones), 34, 41 associated to a place, 79
Rolle, 20 associated to a valuation, 64
Sturm, 21 discrete of rank 1, 59
Sylvester, 25, 29 maximal ideal, 58
Sylvester (inertia theorem), 8 prime ideals, 59
von Staudt, 56 rank, 69
Thin semialgebraic set, 135 real closed, 142
Tilde (), 128 residually real, 58
Topological space value group, 65
Boolean, 125 Variety (abstract) associated to an affine
irreducible, 104 algebra, 108
quasi-compact, 104
spectral, 124
Topology W
associated to a valuation, 74 Well-behaved sublattice, 130
associated to an ordering, 73 Witt
strong, 73, 119 decomposition, 6
Zariski, 103, 109 equivalence, 6
Trace (linear form), 26 index, 6
Trace formula, 50 ring, 7
Transfer of a bilinear space, 48 prime ideals, 13
reduced, 146
Witt’s cancellation theorem, 6
U
Ultrafilter, 130
Ultrafilter theorem, 132 X
Ultrametric, 74 X-convex (X a subspace of Sper A), 162
Unbounded locus, 172
Universal quadratic form, 6
Z
Zariski, O.
V spectrum, 102
Valuation, 63 topology, 103, 109
canonically associated to a valuation ring,
65

Knebusch M Scheiderer C Real Algebra A First Course

Uploaded by

Copyright:

Available Formats

Knebusch M Scheiderer C Real Algebra A First Course

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Knebusch M Scheiderer C Real Algebra A First Course

Uploaded by

Copyright:

Available Formats

Universitext

With Contributions by Thomas Unger

ISSN 0172-5939 ISSN 2191-6675 (electronic)

Mathematics Subject Classification: 12D15, 14P05, 14P10, 12J15, 12J10, 13J30

© Springer Nature Switzerland AG 2022

Regensburg, Germany Manfred Knebusch

1 This book was never written.

Positivstellensätze, preorderings of rings, convex ideals, and the holomorphy ring of

1 Ordered Fields and Their Real Closures . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1

2.12 Artin’s Solution of Hilbert’s 17th Problem and the Sign

1.1 Orderings and Preorderings of Fields

© Springer Nature Switzerland AG 2022 1

Exercise 1.1.3 Let P be an ordering of K. Show that

Exercise 1.1.5 If T is a preordering of K, show that

defines a partial order relation ≤ on K that is compatible with + and ·.

K 2 := {a12 + · · · + an2 : n ∈ N, a1 , . . . , an ∈ K}.

It is clear that K 2 is contained in every preordering of K, and is itself a preordering

Lemma 1.1.9 For every preordering T of K there exists an ordering P of K such

Theorem 1.1.10 Every preordering T of K is an intersection of orderings of K.

Corollary 1.1.11 K has an ordering if and only if K is real.

Naturally, we must assume that char K = 2, since otherwise K 2 is a subfield of

is an ordering of Q(t). It is the ordering induced by the field embedding Q(t) →

1.2 Quadratic Forms, Witt Rings, Signatures

It is not possible to give an in-depth treatment of the algebraic theory of quadratic

For symmetric matrices B, B ∈ Mn (K) it is straightforward to check that

The n-fold sum ϕ ⊥ · · · ⊥ ϕ is abbreviated by n × ϕ and the n-fold tensor product

Lemma 1.2.3 Let ϕ, ϕ , ψ, ψ be quadratic spaces.

[ϕ] + [ψ] := [ϕ ⊥ ψ] and [ϕ] · [ψ] := [ϕ ⊗ ψ].

W (K) is called the Witt ring of the field K.

Exercise 1.2.6 Let K be a finite field of characteristic not 2.

where ϕ+ is positive definite with respect to P and ϕ− is negative definite with

Definition 1.2.10 Let ϕ be a quadratic space over K, P an ordering of K, and

signP ϕ := dim(ϕ+ ) − dim(ϕ− )

is called the Sylvester signature of ϕ at P .

signP (ϕ ⊥ ψ) = signP ϕ + signP ψ,

The statement follows.

Lemma 1.2.13 If a, b ∈ K ∗ with a + b = 0, then

Exercise 1.2.17 Show that if K ∗2 ∪ {0} = {a 2 : a ∈ K} is an ordering of K, then

1.3 Extension of Orderings

In this section, K denotes a field of characteristic = 2.

Definition 1.3.1 Let L/K be a field extension.

in L with m ≥ 1, pi ∈ P and ai , bi ∈ K. It follows in particular (by comparing

a1 g1 (t)2 + · · · + am gm (t)2 = f (t)h(t) (1.1)

in K[t]. We may assume that gcd(g1 , . . . , gm ) = 1.

Proposition 1.3.7 If L/K is a purely transcendental field extension (not necessar-

1.4 The Prime Ideals of the Witt Ring

In this section, K denotes a field of characteristic = 2.

Corollary 1.4.2 Let ϕ be a nondegenerate quadratic space over K. The following

1.5 Real Closed Fields and Their Field Theoretic

Remark 1.5.3 Condition (iii) in Proposition 1.5.2 is clearly equivalent to

Another characteristic of real closed fields is their relative rigidity:

1.6 Galois Theoretic Characterization of Real Closed Fields

Let K be a field of arbitrary characteristic with algebraic closure K. This section is

1.7 Counting Real Zeroes of Polynomials (without

Proposition 1.7.2 (Intermediate Value Theorem) Let f ∈ R(t) and a, b ∈ R

The number of j with a < aj < b is therefore odd.

Lemma 1.7.3 Let 0 = f ∈ R(t) be a rational function and a ∈ R a zero of f . The

f (t) d g (t) h (t)

from which the statement follows.

K 2 := {a12 + · · · + an2 : n ∈ N, a1 , . . . , an ∈ K}.

It is clear that K 2 is contained in every preordering of K, and is itself a preordering

Naturally, we must assume that char K = 2, since otherwise K 2 is a subfield of