Published by Routledge, London, 2016.

The Logical Foundations

of Scientific Theories:
Languages, Structures,
and Models

Décio Krause

Jonas R. B. Arenhart

Florianópolis
2015
My
 iii

My

To my girls Mercedes,
Heloísa, and Olívia (DK)

To mom, Dai, and Carol

(JRBA)
My
 v

(Blanck)

“Science is not just a collection of laws, a cata-

logue of unrelated facts. It is a creation of human
mind, with its freely invented ideas and concepts.
Physical theories try to form a picture of reality
and to establish its connection with the wide world
of sense impressions. Thus the only justification
for our mental structures is whether and in what
way our theories form such a link.”

A. Einstein & L. Infeld

“No, no, you’re not thinking; you’re just being

logical."

N. Bohr
My
Preface

t is often said that scientific knowledge is, in a huge part, con-

I ceptual knowledge. We approach knowledge (in science) with

the help of concepts we elabore in order to study a certain do-
main we are interested in. But we would like to say that scien-
tific knowledge involves a little bit more: we approach knowledge through
structured concepts. Structures are the way we use to approach scientific
knowledge. It is with the help of structures that we organize the concepts
we use in our scientific activity. With structures, the scientist may have
a list of domains of objects she is interested in, a list of relations, oper-
ations, and distinguished elements that are relevant for her approach; the
interrelations between the distinct concepts becomes explicit. So, a struc-
tured account of scientific knowledge seems to capture much of what is
generally said about scientific knowledge being conceptual.
We remark that the notion of structure we are using here is very gen-
eral. Even in psychoanalysis the specialist uses structures of a kind, say
when she organizes her knowledge with concepts like (for a Lacanian)
the unconscious, repetition, transference, and drive. In this book we shall
be concerned with mathematical structures, that is, those involving con-
cepts of mathematized sciences, mainly physical sciences. We hope that
our investigation may throw some light on the nature of scientific theories,
mainly on the relation between structures and the linguistic apparatuses
employed to describe such structures.
In more precise terms, what is a structure in the sense we use this term
here, and how does it enter in the discussion and in the elaboration of a sci-

i
ii

entific theory? We shall give precise definitions of these ideas and, more
importantly, we shall pay attention to the mathematical framework where
the notion of structure is constructed. It is important to notice that there
are in the literature several approaches to structures, ranging from their
mathematical definitions (Bourbaki) to their use in mathematics (Bourbaki,
again) and in more general scientific contexts (Carnap, Suppes). But no
discussion was provided till now on the kind of mathematical apparatuses
a structure depends on and how it influences the idea of a scientific theory
that depends on such structures. If a structure is, as we usually learn in our
first-order logic courses, a tuple of the form A = hD, {ai }, {R j }, { fk }i, with
i, j, k ranging over certain sets of indices, then we need to use set theory.
But which set theory? We rarely discuss questions like this and others re-
lated. Does the domain D really may be considered as a set (a collection of
distinct elements)?1 And what about the individual constants ai ? Do they
really name some of the individuals of D? Are all of them nameable? What
about the relations R j ? Are they just relations relating elements of the do-
main? But what if the structure is a topological space, a well-ordered set or
something else, as almost all structures that model scientific theories are,
which are not first-order structures like that one exemplified above? Even
if we say that we are reasoning within a specific set theory, what about
the structures that model the set theory itself? Well, one could say, we
can always appeal to informal set theory, where (almost) everything can be
done. This is true, mainly because informal set theory is inconsistent.2 So,
should we regard our best scientific theories to be grounded on an inconsis-
tent mathematical basis? From the point of view of the logical foundations
of science, this sounds undesirable. As philosophers, we should look for
suitable frameworks for grounding our best theories. Except, of course, if
the theory is to be inconsistent. But in this case we need to change logic
accordingly.
In addition, we may ask: why to use set theory and not higher-order
1
A pioneering questioning in this sense appeared in [DalTor.93]; see also [FreKra.06].
2
Inconsistent in an intuitive sense.
 iii

logics or category theory? Are these alternatives not good enough to pro-
vide a better ground to act as pillars of scientific theories? Of course this
also needs to be answered. This book will not address these questions in
details. We simply assume the standard framework most philosophers of
science usually make use of, when they say something about the subject,
namely, set theory. In fact, the semantic approach to scientific theories,
a paradigm we are still living in, considers scientific theories in terms of
their models, which, pushing the issue a little, are set-theoretical entities.
But the semanticists (van Fraassen, Giere, Suppes and many others) do not
speak about the set theory they are working in. Suppes explicitly regard
the metamathematical framework as being informal (not axiomatized) set
theory, as is well known, but if asked to be more precise (as one of us has
required him to be), he would say (as he did) ‘choose one you find suitable
for expressing the mathematical concepts you will be in need of’ (this in-
formal conversation took place in Florianópolis in 2002). Okay, we could
be happy with this answer, but we are not. Ever since the 1990s one of us,
and now both of us, are investigating standard set theories on what regards
the notion of an individual. Since there is a reasonable interpretation of
quantum objects in terms of non-individuals (see [FreKra.06]), we think
that the best way to express a quantum semantics is not in terms of sets,
but by using collections that may comprise indiscernible elements, called
quasi-sets. The theory of these entities is not the subject of this book, but
has motivated us to look to the ‘standard’ set theories used in science. In-
deed, physics uses concepts like vectors, matrices, derivatives, differential
equations, tensors, topological spaces, and so on, all of them described in
terms of sets. Thus, as part of a major project, we have made the present
study we now offer to the general reader interested in the philosophical and
logical foundations of science, mainly of physical theories.

We owe much to the approach of Bourbaki’s notion of species of struc-

tures, mainly in the formulation given by Newton da Costa. In due time,
we shall make all the relevant references. We had planned to write a book
iv

discussing the nature of the metamathematical framework which underlies

the most relevant physical theories, but the project became so wide that we
needed to shorten it and keep with a more restrict goal. This general plan
was not put aside anyway. Thus, this book may be seen as a first contri-
bution to the analyzes of the underlying mathematics of scientific theories.
Indeed, it would be necessary a carefully analysis of all physical concepts
and see which kind of mathematics they seem to demand; for instance,
it is usually said in informal conversations that the denumerable version
of the axiom of choice is enough for physics, but as far as we know, no
proof of this fact was provided, and for sure perhaps it cannot be done
since we don’t know the physics that will be developed in a near future.
The study could of course be restricted to known physical theories. But
this is not what we intend to do here, for we are not concerned with his-
tory of science. Our aim is to offer to the reader an overview of the notion
of mathematical structure that underlies the usual physical theories. Thus,
this book is organized as follows.

The first chapter rehearses most of the classical opposition between the
so-called Syntactical and Semantic approaches to scientific theories. We
briefly characterize the two approaches and their main features, pressing
on the alleged virtues of the semantic approach over the deficiencies of the
syntactic approach. As it is widely known, and we discuss in the chapter,
language is the one of the central points of concern: the syntactic approach
is said to fail due to its heavy reliance on language, while the semantic ap-
proach succeeds because it may characterize theories as language-free. We
presente how current debate has shifted from such a concern on language
and how both approaches may be seen as somehow complementary. In par-
ticular, we advance the thesis present throughout the book, that the charac-
terization of theories most appropriate depends on our purposes. That is,
how we present a theory depends on what we need. In most cases, both a
linguistic approach as well as use of models is useful for distinct purposes,
and both are needed for metamathematical studies of scientific theories.
 v

With the main objection that the use of a specific language may cause
trouble in the characterization of a scientific theory out of the way, in chap-
ter 2 we discuss our first tool: the axiomatic method. In fact, a theory may
be characterized by the linguistic formulation of its axioms or else by a
class of models, models of the axioms. So, axioms are the first fundamen-
tal tool for the relevant study of foundations. We keep the chapter much
self-contained, presenting briefly the history of the method and its evolu-
tion. In the last case, the evolution comes in two senses: in one direction,
there is a growing abstraction in the method from the XIXth century on-
wards. In another sense, there is, also from the XIXth century onwards,
the recognition that this is a useful tool for conceptual clarification also
for empirical sciences such as physics (obviously, the method was a model
and an ideal for philosophers ever since its inception in the antiquities).

We continue our investigation on the axiomatic method in chapter 3.

Now, we exemplify the use of the axiomatic method to describe the theory
that will serve as a mathematical basis for our discussion of structures and
also will work as a framework inside which axiomatization of most scien-
tific theories takes place. The theory in question is the famous Zermelo-
Fraenkel set theory. Set theory is the mathematical basis in which models
as set theoretical structures are developed, so, it is useful, to keep to book
self-contained, to have a good look at how it is developed. Also, it is a case
of a formal theory, so continuing the discussion of chapter 2. Given that
set theory is a sui generis theory, due to the fact that its models cannot be
developed inside set theory itself (unless it is a stronger set theory, as we
shall discuss), we also present a brief discussion on the nature of models
of set theory. Models of set theory play a much greater role on the study
of scientific theories than most people think: the nature of the model of
set theory being employed influences the nature of the models of scientific
theories that are being developed inside that set theory. We discuss such
issues, and even though we cannot develop the whole point in this book,
this is a first approach to the subject.
vi

In chapter 4 we present some criticisms advanced against the axiomatic

method. We argue for its usefulness in the development of science and
try to show that the main criticisms are not effective. In particular, criti-
cisms are presented against the Suppesian approach to axiomatization, an
approach we shall present in detail in the next chapter. We argue that the
criticism in this case are in general misguided, confusing a general pre-
sentation of the axiomatic method with a particular use made of it: that
is, the critics attack the fact that the method was employed only to deal
with Classical Particle Mechanics, and that this restriction renders it use-
less. However, as a presentation of a method, it is perfectly legitimate, and
extensions may still be provided for other areas of physics. That is, the
method is not restrictive and limited because use was made of it to deal
with a limited field of Physics.

Chapter 5 addresses the issue of axiomatization of empirical theories.

We are mainly concerned with two approaches to axiomatization: one by
da Costa and Chuaqui, and another by Suppes himself. Here, our previ-
ous discussion on the use of language and on the nature of models begin
to enter the discussion. As we argue, da Costa and Chuaqui axiomatize a
theory by presenting its postulates in a formal language. The models of the
theory thus framed are in fact models in the style of Tarski. We keep the
text self-contained and present details about the construction of structures
and formal languages. We also present as an example the axiomatization
of Classical Particle Mechanics in the style of da Costa and Chuaqui; this
requires that every mathematical step theory be axiomatized too. In oppo-
sition to that approach, Suppes’ axiomatization in set theory assumes not a
specific formal language for a theory, but rather set theory itself as the lan-
guage in which the theory will be axiomatized. This is in straight contrast
with the da Costa-Chuaqui approach. We exemplify how this restriction to
the language of set theory increases the power of the axiomatization and
how it facilitates axiomatization. Examples are also provided. So, while da
Costa-Chuaqui axiomatization is more apt for metamathematical studies,
 vii

Suppes’ approach focuses on the easiness of application.

Finally, in chapter 6 we develop a discussion on the nature of models
for each of the kinds of axiomatization developed in the previous chapters.
the
xxx As we mentioned, da Costa-Chuaqui axiomatization is a typical logical ax-
iomatization, with models playing the role of providing an interpretation
for the axioms. In this sense, a class of models is a class of models in
the Tarskian sense, with an interpretation of the language of the theory.
This provides for nice opportunities to apply the resources of higher-order
model theory to scientific theories, in particular, the theory of definition
presented briefly in chapter 5 may contribute with a possible solution to
the problem of equivalence between theories as formulated in chapter 1.
We address these issues in this chapter. Suppes’ approach, on the other
hand, is not a legitimate model approach, as it were. The models for an
axiomatic system in the style of Suppes are not taken in the same sense
as Tarski’s models; the notions of interpretation and truth are absent. We
discuss such issues in this chapter and how one can account for some set
theoretical structures being the ‘models’ of a theory. We also discuss how
the models of the set theory being employed have relevance on the nature
of the models that characterize a theory. Even though our discussions are
not fully developed, they set the stage for further investigations and devel-
opments, which we hope to conduce in our investigations to follow.
The reader may take this book as a first step, a re-opening of the field,
an attempt to bring metamathematical issues back to the study of scientific
theories in particular, and of philosophy of science in general. We hope
this rather initial and inconclusive investigation may contribute to inspire
others to follow along similar lines.

Florianópolis, December of 2015

The authors
viii
Acknowledgments

e would like to acknowledge the influence and help of

W several people. With the risk of leaving someone out, we

would like to thank Newton da Costa for his influence,
Otávio Bueno for constant discussion, Steven French,
Francisco Antonio Doria, Fernando T. F. Moraes. We must not forget to
mention our students Conrado Emerick and Lauro de Mattos Nunes Filho,
who had the patience to take our courses on scientific theories while we
were still working on the manuscript. Special thanks to Dean Rickles and
Elaine Landry, the editors of the series Philosophy of Mathematics and
Physics for the kind invitation for us to write this book. We also would
like to give special thanks to Sophie Rudland for the kind correspondence,
attention to us and for the guidelines in preparing this manuscript.

ix
x
Contents

Preface x

Acknowledgments x

1 Characterization 1
1.1 The Received View and Syntactical Approaches . . . . . 5
1.2 The Semantic Approach . . . . . . . . . . . . . . . . . . . 14
1.3 Scientific theories and philosophical tools . . . . . . . . 25

2 Axiomatization 29
2.1 Concrete axiomatics . . . . . . . . . . . . . . . . . . . . 35
2.2 Abstract axiomatics . . . . . . . . . . . . . . . . . . . . 41
2.3 Formal axiomatics . . . . . . . . . . . . . . . . . . . . . 48
2.3.1 First-order classical logic with identity . . . . . 49
2.3.2 Peano Arithmetics . . . . . . . . . . . . . . . . . 54

3 A mathematical background 59
3.1 The Principle of Constructivity . . . . . . . . . . . . . . 60
3.2 The ZF(C) set theories . . . . . . . . . . . . . . . . . . . 64
3.2.1 The postulates of ZF(C) . . . . . . . . . . . . . . 68
3.2.2 A matter of terminology . . . . . . . . . . . . . 73
3.2.3 Inaccessible cardinals . . . . . . . . . . . . . . . 74
3.2.4 Informal semantics of L2 . . . . . . . . . . . . . 75
3.3 ‘Models’ of ZF(C), again . . . . . . . . . . . . . . . . . . 76

xi
xii CONTENTS

3.4 Urelemente . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.4.1 Some comments . . . . . . . . . . . . . . . . . . . 83
3.5 More on the ‘models’ of ZF(C) . . . . . . . . . . . . . . 83

4 Criticism to the axiomatic method and its defense 85

4.1 Does the AM encompass heuristics? . . . . . . . . . . . . 86
4.1.1 The axiomatic method and heuristics . . . . . . . 88
4.2 Truesdell against the ‘Suppesians’ . . . . . . . . . . . . . 89
4.3 Arnol’d and the Bourbaki program . . . . . . . . . . . . 94
4.4 Worth axiomatize? . . . . . . . . . . . . . . . . . . . . . 95
4.5 Does axiomatization really axiomatizes? . . . . . . . . . 98

5 Axiomatization, and Scientific Theories 105

5.1 External axiomatizations . . . . . . . . . . . . . . . . . . 106
5.2 Structures . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.3 Languages and algebras . . . . . . . . . . . . . . . . . . 115
5.4 Languages for speaking of structures . . . . . . . . . . . 120
5.4.1 The language of a structure . . . . . . . . . . . . 122
5.5 Definability and expressive elements . . . . . . . . . . . 124
5.6 On the new symbols . . . . . . . . . . . . . . . . . . . . 127
5.7 da Costa and Chuaqui — Suppes predicates . . . . . . . . 131
5.7.1 A Suppes’ Predicate for Group Theory . . . . . . 133
5.7.2 A Suppes’ Predicate for classical particle mechan-
ics . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.8 The approach of Suppes . . . . . . . . . . . . . . . . . . . 138
5.8.1 An axiomatization of non-relativistic quantum me-
chanics . . . . . . . . . . . . . . . . . . . . . . . 146

6 Models and scientific theories 155

6.1 Again on the order of languages and structures . . . . . 156
6.2 Further remarks on Suppes’ approach . . . . . . . . . . . 158
6.3 da Costa and Chuaqui, further remarks . . . . . . . . . . 162
CONTENTS xiii

6.4 The metamathematical framework . . . . . . . . . . . . . 165

Bibliography 175
xiv CONTENTS
Chapter 1

The Quandary on the Characterization

of Scientific Theories

t is perhaps a common place to say that the debate on the nature

I of scientific theories in contemporary philosophy of science has

mainly revolved around the syntax-semantics debate. During the
first half of the twentieth century, when the Logical Empiricist
program flourished, a version of the syntactical approach ruled. During the
second half of the twentieth century, after the development of model the-
ory by Tarski and the decline of the Vienna Circle, the semantic approach
was developed, and still forms the present day’s paradigm. However, a
closer look would reveal that apart from the story of a heated dispute, there
was not really a debate: after the decline of the Received View — as the
approach by the Logical Empiricists to scientific theories was called — al-
most no one else was really promoting that kind of view, so it is strange
that philosophers of science adhering to the semantic approach took as part
of their tasks to criticize what they took to be the syntactic approach.
F. Suppe’s Preface and Afterword in [Sup.77] illustrate the situation
quite clearly; Suppe’s volume is a collection of papers presented at a Sym-
posium on the Structure of Scientific Theories, held at Illinois in 1969.
The call for the Symposium urged participants to promote new views on
the structure of scientific theories, given that the Received View was con-
2 CHAPTER 1. CHARACTERIZATION

sidered as largely inadequate (again, see the Preface to [Sup.77]). Facing

the ruin of the only unitary framework available for philosophical work on
scientific theories, philosophers of science were looking for a substitute.
As Suppe [Sup.77, p.618] has put it, after such a successful debunking of
the Received View:
. . . the result was widespread confusion and disagreement among
philosophers as to what the main problems in philosophy of sci-
ence were, how they should be approached, and what would con-
stitute acceptable solutions to them. The 1969 Illinois Sympo-
sium on the Structure of Scientific Theories [. . . ] occurred in
the midst of this disarray, and thus provides a particularly vivid
account of a discipline in search of a new direction.
So, it seems, a framework for scientific theories is required in order for
the main problems in philosophy of science to be clearly stated and investi-
gated. The Received View furnished such a framework, and in its absence
no consensus could be achieved about even what were the main problems
in philosophy of science. A replacement was being sought, and was found
on the new semantic approach. The later did not seem to fall prey to the
difficulties that plagued the Received View. However, the dialectics of the
debate seems to put much emphasis on that contrast; a cursory look in the
literature will give the reader the impression that there is still a debate; ad-
herents of the semantic approach still seem to feel the need to constantly
review criticisms of the Received View and promote the advantages of the
semantic approach on the treatment of those difficulties.
Given that still prevailing agonistic spirit, the central issue concerns
whether scientific theories (specifically, empirical theories) are more prop-
erly characterized as syntactical entities, in terms of formal languages and
sets of axioms and inference rules expressed in such formal language, or
as semantic entities, in terms of classes of models and/or structures (there
is disagreement over whether structures and models are the same kind of
thing and about which is more appropriate for the semantic approach, as
 3

we shall see soon). It is generally agreed that the semanticists have won
the battle, and the semantic approach is now considered as the new ortho-
doxy [Con.06]. Now, as we have remarked, it should come as no surprise
that the semantic view established itself, given that the syntactic view was
seen as unable to deal with its criticisms even before the semantic approach
clearly emerged. Along with the rise of the new orthodoxy, a less rigorous
and less formal-friendly mood has dominated the philosophical studies of
scientific theories.
However, apart from how the story is usually told, new studies on the
Received View are emerging and along with them, a more faithful under-
standing of the characterization of scientific theories by the Logical Em-
piricists is being achieved. With the publication of such works, we start
to understand that the debate is presented in such lights that are not favor-
able to the syntactical approach; in fact, the Received View is generally
presented as a caricature of a highly naive and implausible view (see, in
particular [Lut.12], [Hal.15] and the references in those works). The syn-
tactic approach, mainly identified with the approach advanced by some
members of the Vienna Circle such as Carnap and Hempel, was criticized
in almost every aspect. As it was characterized by its opponents, it really
did suffer from profound difficulties as an approach to scientific theories.
Perhaps the heavier criticisms seem to be those accusing it of too radical
deviance of actual scientific practice, mainly due to its heavy reliance on
first-order logic and axiomatization. As an account of scientific theories,
the Received View failed badly by distancing itself from real science and
by being relying so heavily on formal tools and techniques (or, at least, so
the argument goes).
The semantic view, on the other hand, seemed to be completely di-
verse from the syntactical view in those aspects, keeping close to actual
scientific practice and not requiring that scientific theories be formulated
in any specific language. In particular, the last feature was erected as a
great virtue of the view and defended by van Fraassen and his followers
4 CHAPTER 1. CHARACTERIZATION

(see for instance [vanF.89, pp.221,pp.225-6]). The so-called ‘model revo-

lution’ initiated by Patrick Suppes in the 60s would be reduced to nothing
if language were allowed to play a substantial role in the formulation of a
theory (the claim in not from Suppes himself, but see [Sups.60, Sups.67]
and [Mul.11, sec.6]). However, as we shall discuss in what follows, it
is perhaps this sole requirement of being ‘language free’ that makes the
semantic approach almost senseless, while at the same time it is this re-
quirement that allegedly marks a radical divide between both approaches
nowadays. Leave that requirement out and we have a position that can,
perhaps, be made compatible with a syntactical approach too.
We shall not attempt to present here a revision of the literature about the
whole dispute. However, given that the following chapters will deal with
issues that are related to both the semantic and the syntactic approaches,
and references shall be made to those approaches, we shall give here to the
reader a brief summary of the debate and present reasons for its seemingly
going out of the tracks. Our aim is to promote not one of the approaches
as superior, but rather to argue that scientific theories may be profitably
studied by the philosopher and by those interested in foundations from
many distinct points of view. Instead of a competition between distinct
approaches, we propose that they do complement each other. This kind
of claim will involve another revision of the attitude towards the relation
between theories in real scientific practice and our rational reconstruction
of them for philosophical purposes. As we shall argue, our constructs may
employ distinct technical resources, and it is not clear that they should re-
produce in every detail their informal counterparts. Perhaps formalized
theories (be it in a formal language or some set theory) gain a life of their
own, helping us to understand their informal counterparts; that is their pur-
pose.
So, for the moment we shall briefly review the main features of the
Received View (and mention syntactic approaches) and of the semantic
approach. This, we hope, will show how features of both approaches are
1.1. THE RECEIVED VIEW AND SYNTACTICAL APPROACHES 5

employed in the next chapters with the aim of logical analysis of scientific
theories. It should be clear that an emphasis on the identification of theories
with something seems responsible for much of the problems.

1.1 The Received View and Syntactical Approaches

The traditional presentation and criticisms addressed to the syntactic view

of theories should more properly be confined only to what we have called
the Received View on scientific theories. Recall that the Received View is
the view on scientific theories developed by members of the Vienna Cir-
cle, mainly Carnap and Hempel (the classical exposition may be found in
the introduction by F. Suppe to the volume [Sup.77]). While the Received
View is clearly a syntactical view, it is not the only incarnation a syntac-
tical view may have. The Received View is clearly a syntactical view in
its advocacy of formalizing theories by employing a formal language and
presenting it as an axiomatic calculus, but the fact that a theory is charac-
terized as a formal system does not by itself imply that the characterization
follows the Received View; as we shall argue briefly, there is much more
to the Received View than to a syntactical approach in general.
Traditionally, the Received View is characterized by requiring that the-
ories be framed according to the following features:

Language: There is a formal language whose non-logical vocabulary is

divided in two parts: VT and VO . VT is the set of theoretical terms, and
VO is the set of observational terms. Notice that there is no restriction
on the order of the language or kind of language to be used (it does
not prohibit higher-order languages, modal languages, and so on; see
for instance [Car.58]).

Logic: The language has a set of logical axioms, giving rise to its under-
lying logic (in general, type theory; see [Lut.12, sec.2]).
6 CHAPTER 1. CHARACTERIZATION

Theoretical postulates: A set of sentences written exclusively in the the-

oretical vocabulary is selected as the set of theoretical postulates,
which we shall denote by T .

Semantics for VO : Observational terms receive an informal semantics re-

lating them to observable objects and events. This is related with the
theory of meaning as verification adopted by logical empiricists.

Correspondence rules: A set of sentences relating theoretical terms with

observational terms, called correspondence rules. These provide for a
partial semantics for theoretical terms, relating them with immediate
experience. We shall call C the set of correspondence rules.

So, characterized by those requirements, a theory is individuated by

its theoretical postulates and its set of correspondence rules. Let us call,
accordingly, TC the resulting theory. It is of particular relevance that by
framing a theory with the specific choice of a language, one already com-
mits the theory with being characterized not only by its postulates and
correspondence rules, but also with a specific vocabulary and language.
Now, each of those features the Received View demanded of a theory
received its share of criticism (see again the introduction to [Sup.77] for the
classical exposition and criticisms of the Received View; Muller [Mul.11]
also presents some of the same points). It is now almost a widespread
consensus that some of the demands of the Received View could not be
achieved. In particular, there are many aspects that were very closely tied
to philosophical theses of the Vienna Circle, and which ended up by lead-
ing the general program of the logical empiricists to its collapse in its orig-
inal form; those aspects inevitably lead to problems in the Received View
too.
Perhaps the most dramatic example of the mix between philosophical
tenets of the logical empiricists and their characterization of a theory comes
from the peculiar kind of relation between a theory and experience. To
mention some of the most prominent ones, there should be a match be-
1.1. THE RECEIVED VIEW AND SYNTACTICAL APPROACHES 7

tween observational vocabulary and observable entities. The same should

also happen to theoretical vocabulary and non-observable entities. How-
ever, it is really difficult to confine the observational vocabulary to refer
only to observable entities; depending on the use of quantifiers along with
the observational vocabulary it is easy to produce sentences in the obser-
vational vocabulary that do not relate directly with any experience. Also,
depending on how one understands the relation of verification by experi-
ence, it would be simple to produce examples of experiencing theoretical
entities: for instance, by putting our fingers in the electric plug (don’t do
that) we have an experience of electric current, which should be a non-
observable entity.
So, the division in the vocabulary and its corresponding division be-
tween kinds of entities is very problematic, to say the least (Kuhn, Feyer-
abend and Hanson, for instance, made much of those difficulties, but we
shall not enter such a debate here; see the Introduction to [Sup.77, sec.5]).
Other very difficult problem concerns the precise role of correspondence
rules. Carnap, for instance, had many difficulties in explaining how those
rules attribute meaning to theoretical terms. At first, it was required that
the correspondence rules had the form of an explicit definition, such as

8x(T x $ Φx),

where T is a theoretical term and Φ is a sentence in the observational vo-

cabulary. That approach had many problems; in particular, it could not
be employed to dispositional terms, such as “soluble” and “fragile”. For
instance, in the case of fragility, it would lead to the implausible

x is fragile $ (if x is severely beaten, then x breaks).

The problem with that approach is that anything that is never beaten
ends up being fragile (due to traditional truth conditions associated with the
material conditional). So, the requirement of explicit definition of theoret-
ical terms had to be relaxed. Less stringent demands were soon provided
8 CHAPTER 1. CHARACTERIZATION

for, but had their own difficulties. One of the options consisted in cor-
respondence rules providing for conditionals, only partially defining the
meaning of theoretical terms. For instance, consider ‘fragile’ again. Then,
the proposal by Carnap was the following:

If x is severely beaten, then (x breaks $ x is fragile).

It is difficult to specify what does it mean to provide only partial mean-

ing to theoretical terms, and related problems with analyticity helped to
frustrate the program.
For a related problem associated with correspondence rules, it was ar-
gued that demanding that the methods of application of a theory be part of
the formulation of the theory leads to inconvenient situations. For instance,
whenever a new method of application of a theory is discovered (new appa-
ratuses due to new technology, for instance), the formulation of that theory
must be changed accordingly, and we would literally have a distinct the-
ory. Also, requiring that correspondence rules should be included in the
formulation of the theory seems to put the causal mechanisms of theory
test inside the theory itself, which is counterintuitive, to say the least.
In general, the main problem with the Received View concerns its sim-
plified account of the relation between theory and experience. It relied
heavily on the empiricists dogma of reduction (sentences must be verified
or corroborated one by one) and were meant to introduce in the account of
theories the problematic verifiability criterion of meaning. However, no-
tice that one could in fact dispense with those peculiarities of the Received
View in alternative forms of the syntactic approach: one needs not adhere
to the tenets of the logical empiricists to adopt a syntactical view of theo-
ries, with a change on the mechanisms of the relation between theory and
experience being a nice starting point.
But for our purposes here, much more important than those well-known
difficulties in the relation between theory and experience, which most peo-
ple consider as unsurmountable, is the kind of attack leveled against the
1.1. THE RECEIVED VIEW AND SYNTACTICAL APPROACHES 9

use of formal languages made by the Received View. This kind of attack is
also considered as unsurmountable, but the reasons given for it are not that
convincing. Typically, authors complaining about syntactical approaches
(and about the Received View, which as we mentioned is generally iden-
tified with syntactical approaches) claim that the Received View adopts
the thesis that the correct language for formalizing empirical theories is
a first-order language, with an accompanying first-order logic (again, see
[Sup.77] and also [Mul.11] for that claim). That kind of language is clearly
inadequate for most scientific theories, and even for those theories which
can be written by using this kind of formalism, the resulting axiomatiza-
tion is awkward and impractical. In particular, the required mathematics
of a scientific theory would have to be wholly axiomatized from scratch,
leaving us with so much preliminary work to be done before getting to
the theory itself; that requirement not only distances the Received View
from actual scientific practice, but also would render the whole enterprise
useless for any scientific purpose.
As an instance, consider the theory of complete ordered fields, which is
part of a good amount of scientific theories employing mathematics. First-
order languages are not adequate to axiomatize the theory; in particular,
one cannot provide an axiom for the least upper bound principle, — which
characterizes the field of real numbers — as a single formula. In first-order
languages we employ an axiom scheme:

(9x↵ ^ 9y8x(↵ ! x  y)) ! 9z(8x(↵ ! x  z)^

8y(8x(↵ ! x  y) ! z  y)).

Here, ↵ is any formula of the first-order language for ordered fields,

and x, y and z are variables. The need to use a scheme of axioms and
the existence of unintended countable models makes for the theory to be
uncategorical. Second-order versions of the principle employ predicate
variables in the place of ↵, and then the axiom can be expressed as a single
10 CHAPTER 1. CHARACTERIZATION

formula. As it is well known, the second-order theory of complete fields is

categorical. The same kind of reasoning may be employed to argue that the
theory of well-orderings or cyclic groups are not elementary theories. One
cannot obtain the class of well-orderings with a set of first-order sentences;
well-ordering is not a first-order class. That is obviously a problem for
those limiting themselves to theories with first-order languages (see also
[Kun.09, p.89]).
Against that line of criticism, it was argued by Lutz [Lut.12], with ap-
peal to a large textual evidence, that there is no reason to think that Carnap
and other adherents of the Received View ever demanded formalization in
first-order logics, with its devastatingly impractical consequences. In fact,
most of Carnap’s works explicitly employ the theory of types, which from
a logical point of view is enough to ground not only the required math-
ematics, but also the higher-order concepts involved in science (see for
instance [Car.58]). Furthermore, formalization in general, as seen by these
authors, did not demand an exhaustive axiomatization of all the mathe-
matics required: it is enough to know that the relevant mathematics can be
developed inside type theory. Once it is known how to develop the required
mathematics, one may proceed to deal exclusively with the empirical sci-
ence one is concerned with.
So, the objection that the use of first-order languages leads to the fail-
ure of the Received View is based on a widespread misunderstanding. In
fact, due to the large amount of evidence that the proponents of the Re-
ceived View (mainly Carnap) did not require exclusive use of first-order
languages, it is strange, to say the least, that opponents of the view have
focused for so long a time on that aspect.
The objections against languages did not stop at the supposed inade-
quacies of first-order languages. As argued generally, and made famous
by van Fraassen and Suppe, one of the main problems with the Received
View and syntactic approaches in general concerns the use of a specific
vocabulary and of axiomatization in such vocabulary. Given that a theory
1.1. THE RECEIVED VIEW AND SYNTACTICAL APPROACHES 11

is identified with its linguistic formulation (in axiomatic terms), it would

result impossible to have formulations of the same theory in alternative vo-
cabularies. Now, that claim involves two kinds of presuppositions. One of
them is that the Received View did not have means to account for the use
of distinct vocabularies in the formulation of a theory. The second con-
cerns the very idea, widespread among critics of the Received View, that
a theory is identified with its formulation. Do those criticisms really hit
the Received View? What are the consequences for a syntactical approach
that does not necessarily adopt all the features of a theory as required by
the Received View?
As to the first problem, viz., the use of a specific formulation, it seems
clear that the Received View needs not be that specific. As Halvorson
[Hal.15, p.5] and Lutz [Lut.15, p.7] argued, the Received View, and in
general any syntactical view, needs not be committed to the individuation
of a theory as formulated in a particular language: one may clearly ac-
cept equivalent formulations of a theory. That requires certainly a notion
of theory equivalence, but it is clear that such equivalence may be cashed
in distinct terms, such as definitional equivalence or equivalence by trans-
lations (see [Hal.12, p.191] for definitional equivalence and also [Mul.11,
sec6] for translations between distinct formulations). In particular, there is
no evidence that the Received View was committed with the individuation
of a theory by its vocabulary.
So, syntactical approaches in general are not tied to the use of a spe-
cific language forever. This should be clear from Carnap’s writings, for
instance, but was constantly repeated and pointed out as a definitive failure
of the syntactical approaches in general. Closely related with this claim,
there is another very widespread view about the aims of the Received View
(and of the semantic view and syntactic approaches in general): it provides
for an explication of the concept of ‘theory’. Suppe [Sup.77, pp.57-61], in
his famous introduction, identified the whole project of providing a char-
acterization of theories by the Received View as an attempt to provide an
12 CHAPTER 1. CHARACTERIZATION

explication of the concept of theory in general. Halvorson [Hal.12, Hal.15]

also adopts that view, even though his sympathies lie with syntactic ap-
proaches.
The main problem with that approach concerns how an explication is
to be understood according to Carnap, for instance. An explication of a
certain concept C, the explicandum, consists in the development of a new
concept C ⇤ , the explicatum. The relation between explicandum and ex-
plicatum in the Received View, according to Suppe, is such that the ex-
plicatum is a precisification of the explicandum. In other words: i) every
positive instance of a C should also count as a positive instance of C ⇤ ; ii)
every negative instance of C should also count as a negative instance of C ⇤ ;
iii) C ⇤ must be more precise than C whenever possible; iv) C ⇤ should be
fruitful. We can then use C ⇤ to decide about cases in which we ain’t sure
whether something is a C or not. That view on explication, then, applied
to theories, means that the Received View should be seen as advancing an
explicatum for the intuitive informal notion of scientific theory, the expli-
candum.
Now, by understanding explication as a precisification, it is clear that
the Received View should fare rather badly. Indeed, almost any approach
making use of formal methods will have its problems with such an ap-
proach. Obviously, not every scientific theory in the intuitive sense can be
axiomatized as a formal system. Also, not every formal system in accor-
dance to the rules of the Received View counts as a theory (it is easy to
produce simple counterexamples). So, Suppe concludes that the Received
View failed in this aspect too, while the semantic approach (which we shall
present soon) does much better.
However, it is clear that Carnap and other adherents of the Received
View did not adopt a precisificationist view of explication. As argued in
[Lut.12, sec.5], Carnap did allow some deviation from common use in ex-
plications. The proper rigorous explication of a concept in scientific terms
in general advances deviations from common use; in fact, as Carnap him-
1.1. THE RECEIVED VIEW AND SYNTACTICAL APPROACHES 13

self advances an example, the scientific explication of the term ‘Fish’ re-
places it by another concept, let us call it ‘Piscis’. Those concepts do
not coincide: while Fish counts whales and seal as its instances, Piscis is
much narrower and excludes those cases (see [Lut.12, p.101] for the full
quotation and discussion; it is curious that Halvorson [Hal.15] mentions
Suppe’s account [Sup.77] of explication as illuminating, while at the same
time making reference to Lutz’s [Lut.12] paper, which has a much more
convincing analysis).
As suggested by Lutz [Lut.12, sec.5], perhaps it would more useful to
see the Received View as a rational reconstruction of particular theories,
and not as a general reconstruction of the concept of a theory. As Carnap
was aware of, not every theory is so developed that can be axiomatized and
presented in the canonical form proposed by the Received View. So, that
kind of presentation should be seen as an ideal, a model in which theories
mature enough can be put in order for us to study the empirical meaning
of them (or whatever else we are interested in about the theory). So, some
of the most telling problems for the Received View, say those concern-
ing its adequacy as an explication of the term theory, may be overcome
by recognizing that such an explication of the general concept of theory is
not what is really at stake: rather, what is presented is a method or formal
tool for the study of specific theories; it may be useful for certain purposes
in certain circumstances, but it may not work for other purposes (see also
[DutRec.15] for further relations between explications and the use of for-
mal methods).
Let us now put together some of what the previous discussion has en-
lightened: a syntactic approach to theories needs not employ every aspect
of the Received View. In particular, those aspects that were already found
as lacking, such as the specific relation between theories and experience,
the specific division of the vocabulary into theoretical and observational,
the use of rules of correspondence, for instance, need not to be present.
Also, the use of some particular language need not commit us with iden-
14 CHAPTER 1. CHARACTERIZATION

tification of a theory with its particular formulation: i) on the one hand,

the use of formal languages can be made compatible with distinct formu-
lations, being required only a notion of equivalence between theories; ii)
there is no need of an identification of a theory with anything; in fact, the
use of some formal method is not meant to provide for a definition or ex-
plication of the concept of a theory. The use of some syntactical method is
a resource that philosophers of science and those interested in the founda-
tions of science may deploy for the purposes of their work.
In general, then, we agree with Lutz [Lut.15, p.7] when he character-
izes an approach to theories as syntactical if it comprises the following, in
general lines:

Formal language: the theory is formulated in some formal language (pos-

sibly a higher-order language).

Theoretical equivalence: a relation of equivalence between formulations

is provided (so that distinct formulations count as formulation of the
same theory).

Interpretation: some sentences are interpreted (to grant that we are deal-
ing with empirical theories).

Notice that the most troublesome aspects of the Received View are ab-
sent or need not to be included. In particular, the requirement of interpre-
tation need not be framed in terms of correspondence rules or any other
restriction due to the Received View.

1.2 The Semantic Approach

Traditionally, the semantic approach is said to have been born along with
the development of model theory by Tarski, in the 50’s (see also [CosFre.03,
chap.2-3]). Particularly influential in developing the approach are the works
of Suppes from the fifties onwards (see for instance [Sups.60, Sups.67]),
1.2. THE SEMANTIC APPROACH 15

which we shall follow closely here. Although Suppes may be seen as lead-
ing the semantic approach, there is not really a unified view which could
be called the semantic approach, in opposition to the Received View; in
particular, depending on how ‘models’ are understood the view takes on
distinct versions, which vary also as to how models are related to reality
and what role language plays in characterizing a scientific theory.
Now, given the existence of disagreement on what precisely models
are and how they relate to reality, we shall here adhere to what may be
called the hegemonic version of the semantic approach. While this may
be controversial, we shall restrict the scope of what we call the semantic
view here and focus on the set theoretical development of models, leaving
aside other kinds of models that also feature in those discussions. This
should not strike the reader as a big restriction, given that some of the most
recent debates on the syntax-semantics debate were drawn from this kind
of approach to model theory.
Following this particular kind of approach to models, most authors take
that the semantic approach involves the following main features:

Models: A theory is seen as a class of models;

Set theoretical structures: Models are set theoretical structures;

Language independence: A theory is independent of language.

As we mentioned before, “models” here are taken as set theoretical

entities; that is, entities built inside some set theory (typically Zermelo-
Fraenkel set theory, as we shall see in the chapters that follow). Following
model theory and the work of Tarski, one could think that a model, prop-
erly speaking, is an ordered pair

hD, ⇢i,

where D is a non-empty set and ⇢ is an interpretation function, relating the

non-logical vocabulary of a formal language with the corresponding set
16 CHAPTER 1. CHARACTERIZATION

theoretical entities: individual constant symbols are mapped to elements

of D, n-ary function symbols are mapped to n-ary functions over D, and
n-ary predicate symbols are mapped to n-ary relations over D.
That is a typical kind of description of a model found in the context of
the semantic approach. However, notice that some simple shortcomings
seem to be forgotten by this specific use of ‘model’ here. First of all, char-
acterized like that, models are what is usually called ‘first-order models’:
models for languages of first-order, comprising only what we call order-1
(first-order) structures.1 By sticking with those models one cannot solve
the problem of keeping with first-order languages. Really, there are rele-
vant theories, both in mathematics and in the empirical sciences that cannot
be modeled by order-1 structures, requiring structures of higher-order. For
instance, well-orderings are not classes of models according to that char-
acterization, that is, a well-ordering is not a order-1 structure, as is easy to
note. Order-1 structures can be models only for first-order languages and
theories written in those languages, which makes the difficulties appearing
for first-order languages to reflect in the models that may be characterized
by those languages. As a second problem, related with the first one, we
have that it only makes sense to call a structure a model provided we have
a set of axioms that are modeled by the structures in the mentioned set. In
fact, models are models of something, and in the present case study, they
are models of some axiomatics. In that sense, the semantic approach would
predate over a form of syntactic approach: only once we have the axioms
of the theory (that is, characterize it syntactically) can we characterize the
theory semantically. But, as we shall see, it is not necessary to follow all
the steps suggested by the Received View.
Now, even though those problems and limitations are generally over-
looked in the literature on the semantic approach, the greatest challenge to
the above characterization of models would be its violation of the language
independence requirement; that requirement seems to prohibit any charac-
1
As we shall see later, we distinguish between the order of a language from the order of a structure,
hence the distinction in the notation.
1.2. THE SEMANTIC APPROACH 17

terization of models that involve interpreting a language. As van Fraassen

[vanF.89, p.366] has put, the advantages of the semantic approach would
be lost if language were allowed to play that role. Here, it should be under-
stood that those advantages include being free from any specific language;
being a structure related with a specific language would render the theory
formulation-dependent (or, at least, so the argument goes). In this sense,
if that advantage of the semantic approach is to be preserved, it seems the
that the interpretation function must be left out of the account of a model
adequate for the semantic approach. Instead of models in the Tarskian
sense, perhaps the semanticist really means another kind of set theoretical
construct when she speaks about models (see also [Mul.11, sec.6] for this
kind of discussion).
The most common alternative for the characterization of models in this
case is to allow that the relevant ‘models’ are in fact set theoretical struc-
tures: entities of the kind
hD, Ri ii2I ,
where again D is a non-empty set and Ri is a family of relations on the
elements of D (perhaps comprising also operations and distinguished ele-
ments, all of which may be seen as special cases of relations). Here it is
important to note that the Ri need not relate only members of D, giving rise
to higher-order structures or, as we prefer to call them, order-n structures,
with n > 1 (we shall discuss such structures in chapter 5). For instance, a
group is a set theoretical structure which may be written as
hG, ◦, e, −i,
where G is a non-empty set, ◦ is the composition function, e is the identity
element and − is the inverse operation. Here we have only operations over
elements of D, so it is a order-1 structure. A topological space, on the other
hand, is a structure hD, ⌧i, where ⌧ 2 P(P(D)). That is, ⌧ is a family of
subsets of D, so we have a structure or order greater than 1.
Now, if that suggestion is the case (and it really seems to be), then it
is difficult to see what the structures thus gathered are models of. One
18 CHAPTER 1. CHARACTERIZATION

option, as we shall see later (see chapters 5 and 6 on Suppes’ predicates) is

to say that they are models of a Suppes predicate, a set theoretical formula
axiomatizing a theory inside set theory. But then the notion of ‘model’
is rather different from Tarski’s sense of model, and there is no proper
model theory, given that they are higher-order structures. In the end, it
seems, there is a dilemma to be faced: either the relevant models for the
semantic approach are models in the Tarskian sense or else they are not
models in that desired sense, they are not ‘semantic’ in the sense of making
anything true. In the first case, they are inconvenient because they involve
language; in the second sense it is difficult to say what is really semantics
in the semantic approach. Anyway, one must put constraints on how the
structures are gathered. For instance, in the case of topological spaces, it
is not any structure of the form hD, ⌧i where ⌧ must satisfy some standard
requirements! Language seems important here, as we have mentioned.
Later on we shall argue that the mentioned dilemma is not really a
problem if we abandon, as we think we should, the idea that theories are
formulation-independent and/or language-free. This requirement, as we
shall put it, comes from a desire to identify theories with something; our
purpose, as we shall argue, is to provide distinct representations of theo-
ries, and in that case it is not a problem that a representation is framed in a
specific language (as most representations do).
Before proceeding, let us already fix then the first divide between two
notions of ‘model’ that will be relevant for our further investigations. Both
are set theoretical entities, but they operate in distinct ways. It seems that
those adhering to the ‘language-free’ ideal should submit that the ‘models’
the semantic approach speaks about are merely set theoretical structures;
there is no language in which those structures are interpreted, that is, sets
and relations over those sets (in a general characterization). The best we
may have is a ‘satisfaction’ of a Suppes predicate, an approach to axiom-
atization that proceeds in the language of set theory itself, without requir-
ing a separate formal language for a structure. As we shall discuss later,
1.2. THE SEMANTIC APPROACH 19

however, ‘satisfaction’ here has a distinct sense than the traditional model
theoretic one, as we shall see. Those not adhering to the ‘language-free’
ideal may still think that the models work just as models in the sense of
model theory developed by Tarski (although they may also be models of
higher-order languages; see also [KraAreMor.11]). As we shall argue in
the next chapters, adherents of the semantic approach themselves were not
always clear about this. This distinction was even the topic of a heated
debate in the philosophy of science (see [Hal.12] for the first sparkle, and
[Lut.15] for the general overview; see also our [KraAreMor.11], where the
distinction was already clearly drawn).
Those disputes about the proper meaning of models notwithstanding,
what most adherents of the semantic approach agree about is that identify-
ing theories with classes of models allows us to avoid most of the problems
faced by the Received View. Recall that it is this comparison between the
two approaches that sets in general the particular dialectics of the debate,
putting the Received View (and syntactic approaches in general) in rather
unfavorable lights. Due to the alleged language independence of models,
theories (taken as class of models) are independent of their formulations.
The same theory can be presented as associated with distinct languages,
and none of them is essential to it. Also, there is no need to be worried
with formal details of axiomatization in first-order languages; it is enough
to specify directly the class of models inside some set theory. In this case,
the expressive capability of the semantic approach is said to go much be-
yond the Received View. Finally, to present a theory as a class of mod-
els seems to bring us closer to scientific practice, given that scientists are
model builders, not theorem provers ([Hal.15] for a summary of the typical
advantages claimed in the name of the semantic approach).
As we have already mentioned, the semantic approach has been devel-
oped in a much more informal way than the Received View (even though
it is most of the time said to be a descendent of Tarski!). The idea that one
can identify a theory with a class of models and that one can even collect
20 CHAPTER 1. CHARACTERIZATION

a class of models without essential use of a language went largely un-

analyzed through many years, until recently Muller [Mul.11] and Halvor-
son [Hal.12] have taken it seriously and found the idea lacking. The main
difficulty, as we shall see, derives from a tendency to identify a theory with
a class of models, while providing for no relation of equivalence between
classes of models other than identity of classes.
First of all, informal theories have distinct informal presentations, which
may seems to lead to distinct classes of models. As an example, Hamil-
tonian mechanics and Lagrangean mechanics seem to be the same the-
ory, but are certainly not represented by the same class of models; they
even employ distinct mathematical apparatuses. Something similar holds
to Schrödinger’s and Heisenberg’s versions of quantum mechanics. In the
absence of a criterion of theoretical equivalence, there is nothing in the se-
mantic approach that allows us to grant that intuitive equivalence. In fact,
some philosophers adhering to the semantic view have employed such for-
mal dissimilarities to argue that Hamiltonian and Lagrangean versions of
classical mechanics are not the same theories (see [Hal.12, p.187] for the
discussions and references). So, from an intuitive point of view, the seman-
tic approach seems to fail in individuating theories. This, as we mentioned,
seems to be a result from bare identification of an intuitive theory with its
class of models.
For another main difficulty, it is clear that even some simple theories
such as group theory, may be presented as distinct kinds of structures. For
instance, one may present it as a pair hG, ◦i, with a non-empty set and a
binary operation satisfying the axioms for groups. Alternatively, groups
may be presented as hG, ◦, e, −i as we did before (also satisfying the ax-
ioms for groups, now formulated involving also the concepts of neutral
element and opposite). Both kinds of structures are certainly distinct, and
classes of the first kind are not identical to classes of the second. As Muller
[Mul.11, sec.6] and Halvorson [Hal.12] argued, if we had an accompany-
ing language, then we would be able to employ notions such as translata-
1.2. THE SEMANTIC APPROACH 21

bility or mutual definability to show both theories equivalent. But, given

an implicit prohibition on the use of language due to the desire of being
‘language-free’, it seems that any such use would be troublesome. The
general impression is that any appeal to language would throw us back to
the Received View, which clearly is not the case! On the other hand, it
is also argued that only language could help us in avoiding the problems,
which is also clearly not the case. This illustrates how some cherished
features of the semantic approach may also easily distract us from what is
relevant.
Language was also recently involved in another controversy related
with the concepts of morphisms, in particular certain kinds of embeddings
and isomorphisms. Isomorphisms, for instance, are important in the se-
mantic approach, in particular for the relation between models of distinct
theories, when we wish to claim that a theory is an extension of the other
(horizontal relations between theories). They are also important to account
for the relations between a model of the theory and data models (verti-
cal relations), which are then responsible for the relation between a theory
and experience. This is a main feature distinguishing the kind of semantic
approach we are taking into account here from the Received View.
According to Suppes [Sups.60, Sups.67], for instance, and van Fraassen
[vanF.80, p.56], a theory is applied to reality not directly, as in the case of
the Received View, but through a hierarchy of structures. Roughly speak-
ing (see [Sups.60] for details), we begin with simple experiments and con-
struct a structure to model the phenomena. That is in general a qualitative
model, which by itself does not take into account every feature of expe-
rience; as Suppes says, experience must be passed through a ‘conceptual
grinder’ in order to be ready to be taken into account and related with a
theory [Sups.67, p.62]. Later, it is necessary to find a numerical model
isomorphic to the model of the phenomena. Only then can numbers be
applied to things. This structure than may be embedded in still some larger
structures before being embedded in a model of the theory. Models of
22 CHAPTER 1. CHARACTERIZATION

the theory in general use mathematical concepts that have no analogous in

experience, and so cannot be directly applied to nature. So, even though
this was a very rough description of the relation between theories and data,
isomorphisms are involved throughout the whole process.
However, Halvorson [Hal.12] have argued that the idea of isomorphism
(and embeddings in general) do not make sense in the absence of a lan-
guage. It is language that would be responsible for granting that each rela-
tion in a structure is mapped in a correspondent in another structure which
is the interpretation of the same symbol. In other words, without language,
according to Halvorson, an isomorphism between structures hD, Ri i and
hE, Ki i must be composed by both a bijection f between D and E and also
by a function g sending each Ri into some K j . An isomorphism then is
such that for any n-ary Ri , we have
hd1 , . . . dn i 2 Ri iff h f (d1 ) . . . f (dn )i 2 g(Ri ).
So, for instance, given a structure hD, A1 , A2 . . . An i and another structure
hE, B1 , B2 . . . Bn i there is no way to grant that A1 will correspond to B1 ;
that will happen only when g(A1 ) = B1 , but that is not mandatory. The
claim by Halvorson is that by having a language those possible mixing of
properties would be forbidden.
This definition of isomorphism is enough to allow for some tragedies,
such as allowing that structures which are distinct in the truth values they
attribute to sentences being isomorphic (see [Hal.12, p.192] and [Lut.15]
for a discussion). For instance, consider two theories in a first-order lan-
guage with identity containing a denumerable family of unary predicate
symbols P1 , P2 , . . . as its non-logical vocabulary. Theory T 1 has as ax-
ioms only the formula 9!x(x = x), where 9! is the uniqueness quantifier;
the axiom then states that there is only one thing in the domain of quan-
tification. The interpretation of each Pi is then arbitrary on the domain
containing only one element. Let T 2 be comprised by the axiom of T 1
along with the infinite collection of formulas of the form P1 ! Pn , for
every n. Any model of T 2 will also have only one element in the domain
1.2. THE SEMANTIC APPROACH 23

of quantification, but will have to make sentences of the form P1 ! Pn

true. Both theories are intuitively not equivalent, but, with the above no-
tion of isomorphism, it is easy to show that for any model of T 1 there is an
isomorphic model of T 2 and vice-versa.
The main problem comes from the desire to make theories language
independent. The fact that we cannot appeal to language to grant that,
for instance, given a bijection f between the domains of interpretation, f
must preserve the interpretation of P1 , allowing us to mix the interpreta-
tions of the predicate symbols (the role of the function g in the definition
above). So, if the theories are isomorphic they should be elementarily
equivalent, which is clearly not the case.2 Then, being language-free poses
some problems, according to Halvorson, which point to an inadequacy of
the semantic approach on the individuation of theories.
As we shall see later, the notion of isomorphism employed by Halvor-
son is not the correct one. There is a standard notion of isomorphism
between structures that employs the labels of the relations in a structure
to grant that structures have a similarity kind. So, only structures of the
same similarity kind will allow for the definition of isomorphism, without
the need of a language. What is relevant is that structures have the form
hD, Ri ii2I , where the index set I orders the relations; isomorphisms must
somehow observe that order: relations with the same indexes must be pre-
served. So, isomorphisms may be defined both in the presence as well as
in the absence of a language in which the structures are interpreted; both
approaches to models as we presented them above are allowed to use the
concept meaningfully.
So, perhaps, the conflict between structures that are interpreted in a
formal language and structures that are not interpreted is not a substantial
debate. In fact, that is precisely the conclusion arrived at by Lutz [Lut.15]:
the debate between allowing a language and not allowing a language may
be easily dispelled when we notice that a structure such as hD, Ri ii2I , called
2
We recall that elementary equivalent theories are those first-order theories which satisfy the same first-
order sentences. A precise definition can be seen, for instance, in [Men.97].
24 CHAPTER 1. CHARACTERIZATION

indexed structure may be converted into a Tarski-style structure when we

allow the set I of labels to be the non-logical vocabulary of a language. If
that is the case, the indexing is precisely an interpretation, and we have a
labeled structure. There is no reason for accepting one of those kinds of
structures and denying the other.
Our approach in this book will be to accept both kinds of structures
(given that they are convertible into one another). This is in fact a lesson
to be learned from Suppes’ [Sups.67, p.60] presentation of the first fea-
tures of the semantic approach. According to Suppes, a theory may be
presented in two complementary ways: either following an intrinsic ap-
proach or else following an extrinsic approach. The intrinsic approach
comprises the standard axiomatization using linguistic resources. When
no such axiomatization is available, or is just too cumbersome, we may
adopt the extrinsic approach, of characterizing the class of models directly
in set theory. For that, as we shall argue, we may employ what is called a
‘Suppes predicate’, which helps us collecting the relevant models.
In the end, the distinction comes not from the kind of structure em-
ployed, but rather on styles of axiomatization. While employing formal
languages we think immediately about formal theories, with their axioms
framed in the formal language, when we talk about structures as purely
set theoretical constructs we are employing a Suppes predicate to gather
the structures. Both approaches are interesting and each have its own ad-
vantages, as we shall see. So, Suppes had argued that the model approach
is superior when it comes to study the relation between theory and data
and when axiomatization issues appear. In particular, he argued that using
set theory as the language in which to develop the class of models allows
us to avoid axiomatization of the required mathematics of a theory: all of
the mathematics may be assumed as already developed inside set theory.
So, we may proceed directly to the empirical part of the theory that con-
cerns us. Anyway, what is more relevant is that Suppes himself did not
require that we abandon language in any sense (see in particular the re-
1.3. SCIENTIFIC THEORIES AND PHILOSOPHICAL TOOLS 25

marks in [Sups.11, sec.2]). Philosophers of science may benefit from both

approaches: they are representations of theories, and the most adequate
approach will depend on what we need.
This is a lesson that should be already learned with Suppes’ distinc-
tion between intrinsic and extrinsic characterization. That distinction was
employed by da Costa and French [CosFre.03] for many purposes. In par-
ticular, the extrinsic approach (along with the partial structures approach)
was employed to represent theories and their relation with experience. The
intrinsic approach, on the other hand, was employed to discuss proposi-
tional attitudes such as what does it mean to believe in a theory. Belief
is directed towards propositions, so, it requires a linguistic approach. It
is curious that the benefits of such a distinction went largely unnoticed.
What is relevant for us is: theories are merely represented in philosophy
of science for philosophical purposes. What they really are goes beyond
such representations, possibly. Our focus will be on the fact that seman-
tic and syntactical approaches are representational tools for philosophical
purposes.
The adherents of the semantic approach, then, went wrong in simply
claiming that a theory may be identified with a class of models. That left
them with no option but to employ identity of classes as identity of theories,
which lead them astray. Also, that identification left the impression that a
theory is independent of language. Again, much trouble appeared for the
identity of theories. Difficult questions as to whether theories are the same
or not troubled the view. We shall propose, then, in the next section, a more
flexible consideration of both semantic as well as syntactic approaches,
following the lines of Lutz [Lut.15], as well as [KraAreMor.11].

1.3 Scientific theories and philosophical tools

Perhaps the greatest lesson we have learned from those discussions is sim-
ple: deep problems arise when we attempt to identify a theory with some-
26 CHAPTER 1. CHARACTERIZATION

thing (i.e., to reify theories, to use van Fraassen’s [vanF.89, p.222] famous
terms). The Received View faces many obvious inadequacies, for instance,
if we identify a theory with a formulation comprising a specific language
and a specific set of correspondence rules. In the same way, the seman-
tic approach faces many difficulties if a theory is identified with a class of
models: the corresponding criterion of theory identity will identify theories
that are distinct and differentiate theories that are intuitively the same.
Facing all those difficulties, it seems that a distinct approach should be
taken. Perhaps a good try would be to consider that both the syntactic
approach as well as the semantic approach are not philosophical theories
about the nature of something, viz., scientific theories. Or even better,
they should not be treated like that, even though they have been treated
like that. As Lutz [Lut.15] have remarked, the whole debate was based
on misunderstandings, and whether one decides to focus on languages or
on set theoretical structures depends on what are the main interests in the
moment, and on what approach affords more convenience. Philosophical
approaches to scientific theories should not be censored for not capturing
the essence of scientific theorizing; they strive at capturing aspects of such
a diverse theorizing (see also the conciliatory note by Halvorson [Hal.15,
p.15], according to whom “these approaches need not be seen as competi-
tors”).
This idea may be traced back to some of the first proponents of each
view. Recall our brief discussion on Carnap on explication: as we men-
tioned, the goal of the Received View was not to provide for a general
explication of the concept ‘theory’, but to provide general guidelines to
study philosophical problems of scientific theories. Obviously, the prob-
lems and the guidelines were seen through the lenses of logical empiricism,
but that is not the problem. What is relevant is that for many purposes of
philosophical investigation an approach is available that helps philosophers
shape their questions and look for answers.
Again, as Lutz [Lut.12, sec.5] remarked, an explication could work as
1.3. SCIENTIFIC THEORIES AND PHILOSOPHICAL TOOLS 27

a rational reconstruction of an intuitive concept. In the case of particular

theories, the Received View worked as an ideal form the scientific theory
should be put in and from which philosophers could then study important
relations between theoretical terms and their meaning, and whatever other
philosophical problem should interest them. Obviously, the Received View
failed in providing a convincing framework in its overall form, but that
happened for reasons that are not in general those usually pointed out by
the critics.
The same may be said of the semantic approach. In discussing the re-
lation between theory and experiment, Suppes [Sups.67, p.63] noted that
“there is no simple answer to be given” to the question “What is a scientific
theory?”. He further remarked that a precise answer to that question is not
important: what is relevant is
to recognize that the existence of a hierarchy of theories arising
form the methodology of experimentation for testing the fun-
damental theory is an essential ingredient of any sophisticated
scientific discipline. [Sups.67, p.64]
That is, for the point on which he was concerned, which was to understand
the relation between theory and experiment, it matters less that we answer
the question ‘what is a scientific theory?’ than that we understand the
proper relations of experience of the theory with testing. An for those
concerns, he argued, an extrinsic approach is more appropriate. This is
also clear from Suppes’ [Sups.11, sec.2] observation that his approach is
not the same as van Fraassen’s: the latter attempts to formulate a theory as
free of language, and has left it very far from experiment. This is clearly
not adequate for the purposes Suppes had in the moment.
So, taken seriously, those views by Carnap and Suppes seem to lead us
to a distinct direction, other than those that the whole debate has taken.
They seem to concede that the purpose of offering an approach to theories
is not to capture the essence of a general concept, but rather to provide
the formal tools for us to study and develop what for us, as philosophers,
28 CHAPTER 1. CHARACTERIZATION

matters the most (which in general includes topics as truth, belief in theo-
ries, acceptance, empirical adequacy and so on). That is also the general
line we shall take in the following chapters. They should not be read as an
attempt to provide the tools for a definition of theories, but rather as fur-
nishing frameworks in which certain kinds of studies may be taken. Our
focus shall be on logical and metamathematical issues with foundational
purposes, so that this goal should also be kept in mind.
Perhaps it would be appropriate to insist again on a topic recently raised
by French [Fre.15, p.14]: the semantic approach, the syntactic approach
and any other approach to theories are just representations of theories de-
veloped for philosophical purposes. They are not to be confused with the
thing itself. What are those theories in and of themselves? Do theories even
exist? Those are surely interesting issues, but they may be developed inde-
pendently from the semantic and syntactical approaches. They are really
to be treated as rational reconstructions, and may be judged accordingly.
Having that point of view stated clearly, we shall now proceed to the
topic of axiomatization of theories, perhaps the first step for foundational
and metamathematical studies (which is what concerns us here).
Chapter 2

Axiomatization

hy to axiomatize? When we think about the logical foun-

W dations of science, the use of the axiomatic method arises

as a rather natural idea. The axiomatic method is unique
in allowing us to organize a field of knowledge and put
it in a logically coherent structure; it clarifies and provides a sharp un-
derstanding of how distinct parts of the body of knowledge axiomatized
connect themselves, giving us the edifice of the whole discipline. Besides,
the possibility of metatheoretical studies is another great advantage of rig-
orous formulation delivered by the axiomatic method, an advantage that
must be explored for foundational studies. So, in the face of it, we take
it that the axiomatic method is the preferred tool for the kind of study
we wish to approach here. As we discussed rather briefly in the previous
chapter, the main differences between approaches to theories may also be
expressed as differences in styles of axiomatization. So, a first look at the
general features of the method is not only convenient, but also required.
Even though the use of the axiomatic method (AM) is nowadays com-
mon practice in mathematics, the historical roots of its use go back the
early Greeks and their study of philosophy. Of course, a great deal of
applied mathematical knowledge had already been achieved by the Baby-
lonians and Egyptians, but it stood more for ‘collections of prescriptions’
for doing practical calculations than for an organized field of knowledge
30 CHAPTER 2. AXIOMATIZATION

[Sza.64]. These people did not seem to think of their mathematical knowl-
edge as an edifice, resting on a set of basic accepted (i.e. not proved)
propositions whose truth would be enough to derive the truth of all other
known mathematical propositions. As Szabó suggests, that was achieved
by the Greeks, and not in mathematics properly, but in Philosophy: “[they]
[the Greeks] seem to have come to this [that is, to the idea that the starting
principles should not be proved] from the practice of dialectic. They were
accustomed to the fact that, when one of the partners in a debate wanted
to prove something to the other, he was bound to start from an assertion
accepted as true by both of them” [Sza.65].1 An influential example sug-
gested by this author was the use of indirect reasoning, now known as
reductio ad absurdum by the Eleactics, mainly by Zeno of Elea.
Still in ancient times, Aristotle in his Analytica Posteriora declared that
the axiomatic method was the method for the presentation of a deductive
science (see [JonBet.10]). The basic idea behind the use of the axiomatic
method is to present a certain field of knowledge in such a way that once
certain assumptions are made, every other proposition holding in the field
can be deduced from these basic assumptions, today called postulates or
axioms.2 In this sense, the AM is to be applied to an already sufficiently
known field of knowledge, and its main role would be ‘hygienic’, order-
ing propositions in the following sense: to avoid a regress to the infinite,
we select some set of concepts which we suppose are the basic notions
to start with; they must be sufficient to allow us to define every other re-
quired concept. These basic notions are the primitive notions (or primitive
concepts) of the system. The remaining concepts are defined from the
primitive ones (also called derived concepts). The second step is to se-

1
Árpád Szabó attempts to understand why and how the transformation of mathematics into a discipline
grounded in axioms and proofs takes place; see his mentioned works plus [Sza.78]. Patrick Suppes, on the
other hand, in speaking about the origins of the axiomatic method, attributes it to Eudoxus in the 4th century
BC [Sups.02, p.35], but we prefer to follow the (apparently) historically more detailed analysis developed
by Szabó.
2
Today we do not distinguish anymore between axioms (anciently called common assumptions, valid
for all sciences) and postulates (basic propositions of the specific science under investigation).
 31

lect some propositions, which are taken as primitive, being the statements
upon which the system is built. These propositions are written exclusively
in terms of primitive notions and the defined concepts. All other state-
ments of the theory must be obtained from the primitive ones by deduc-
tion, and are then called theorems of the system. This kind of construction
would make explicit the whole development of a field of knowledge: each
concept employed is either primitive or explicitly defined, and each ac-
cepted proposition is either primitive or explicitly derived from previously
accepted propositions.
As is well known, the paradigmatic example of an application of this
method in ancient times is Euclid’s Elements (4 B.C.), which attempted
to systematize the mathematics known by the Greeks up to the time. Eu-
clid’s book has been a major influence in Western thought since then, an
ideal of rigor to be pursued by mathematicians and philosophers until the
nineteenth century. Its influence notwithstanding, is now recognized that
Euclid’s presentation does not conform to the standards of rigor demanded
by the modern use of this technique. On what concerns the presentation of
primitive concepts, Euclid’s book unexpectedly attempts to provide those
terms with definitions. Some historians think those definitions could have
been introduced not by Euclid himself, but by later commentators, in order
to ‘explain’ the meaning of the basic notions [Eucl.08]. As a result, it is
hard to find in the Elements explicit primitive notions. On what concerns
the theorems, which should be proved solely in terms of the axioms and
previously proved theorems, the Elements also have some failures. For
instance, the proof of Proposition 1 (Theorem 1) employs resources not
covered by the previous assumptions (axioms or previously proved theo-
rems).3 Anyway, despite of the problems found in it, Euclid’s book pro-
vided for a wonderful illustration of the general idea: every proposition
to be accepted must be either postulated or else derived (deduced) from
3
The proposition proposes to construct an equilateral triangle on a given segment. But the ‘proof’ uses
a fact neither assumed nor proven before, namely, that two circles that appear in the proof intersect one
another [Eucl.08].
32 CHAPTER 2. AXIOMATIZATION

propositions which were postulated or already proven before.4

A new attitude towards the axiomatic method was deeply involved with
the development of mathematics itself. During the XIXth century, math-
ematics went through another great transformation with the raise of ab-
stract mathematical structures. Groups, rings, fields, geometries, algebras
in general, topological structures and all such rich population of abstract
structures came to light putting this discipline in another paradigmatic level
from which it will never return (but see the criticisms in chapter 4). The
AM was essential for such developments. The XIXth century has also seen
the birth of modern logic, which soon could also be studied as an abstract
mathematical theory. The development of these two fields originated the
mathematics of the XXth century and perhaps still (in a sense to be dis-
cussed later in this chapter) of the XXIst century too.
Recall that in Aristotle’s and Euclid’s time, the choice of primitive con-
cepts and primitive propositions was subject to a condition of intuitive clar-
ity and intuitive truth, respectively. Given that primitive concepts were in-
tuitively clear, the defined concepts could be intuitively understood as well.
Given that the primitive propositions were necessarily true and intuitively
evident, the propositions derived from them were all true as well.5 As a
result of the revolution began in the nineteenth century, the choice of the
primitive notions and postulates are no longer subject to such demands.
The postulates are assumed because they are useful for the purpose in
mind: they serve for the intended finalities, namely, to derive the theorems.
David Hilbert’s 1899 axiomatization of Euclidean geometry makes use of
three primitive notions, namely, those of point, straight line and plane, and
three relations holding among them, incidence, betweenness and congru-
ence. Defined notions are for instance right angle, parallel lines and so
on (the definitions are not relevant for our purposes). But the Italian math-
4
A wonderful analysis of the origins of Greek mathematics is the book The Beginnings of Greek Math-
ematics, by Árpád Szabó [Sza.78], as well as his mentioned papers.
5
According to Szabó, Aristotle did not believe that the fundamental principles of mathematics —that
is, geometry— could be chosen arbitrarily, but that there would be “simple and natural starting point for
mathematics” [Sza.78, p.227], [JonBet.10].
 33

ematician Mario Pieri presented an axiomatic for Euclidean and Bolyai-

Lobachewskian geometry based on only two primitive notions: point and
motion;6 in this axiomatics, straight line and plane are defined concepts.
From these axioms, all geometric theorems can be obtained. This shows
that, in principle, there is no mathematical concept (and we could say the
same also for the empirical theories) that cannot be defined; definability is
relative to the language employed (see chapter 5 on definability).
Other assumptions from ancient times that are also relaxed include the
demand that the postulates must be all independent, that is, a postulate
should not be deducible from the remaining ones. This is taken now as a
matter of convenience. Sometimes a redundant axiomatics is preferable for
ease of treatment. A typical example is the Zermelo-Fraenkel set theory to
be seen later, where the Pair Axiom may be derived from the Substitution
Axiom and the Power Set Axiom. Historically, as the axiomatic method
evolved, some of the ancient demands of intuitiveness were being left be-
hind, with a crescent search for rigor taking its place.
So, the axiomatic method not only benefited from the development of
mathematics, but it also contributed to it. A new understanding of the
method lies at the roots of the rise in abstraction in mathematics from the
nineteenth century on. This ‘paradigm shift’ gave even place to a famous
dispute between Frege and Hilbert on the nature of the axiomatic method
(see [Bla.14]).
But leaving disputes apart, the success of the axiomatic method was
something to be explored further, and this was precisely what Hilbert pro-
posed in the turn of the XXth century: the axiomatization of the theories
of physics, which had been carried out more or less by chance, as in the
case of Newton’s physics — which can be regarded as a kind of axiomatics
—,7 was seen to be the next natural step of application of the method go-
6
His axiomatics preceded Hilbert’s by several months. The set of motions was characterized as a group
of transformations acting on the set of points. See [Mar.93].
7
In fact, Newton’s three laws can be taken as axioms of his system. But no explicit mention to axioma-
tization was made in his work.
34 CHAPTER 2. AXIOMATIZATION

ing beyond mathematics. This proposal was famously advanced by David

Hilbert as the 6th of his celebrated list of 23 Problems of Mathematics, pre-
sented in the II International Congress of Mathematicians in 1900. There,
Hilbert stressed that
. . . whenever, from the side of the theory of knowledge or in
geometry, or from the theories of natural or physical science,
mathematical ideas come up, the problem arises for mathemat-
ical science to investigate the principles underlying these ideas
and so to establish them upon a simple and complete system of
axioms, that the exactness of the new ideas and their applicabil-
ity to deduction shall be in no respect inferior to those of the old
arithmetic concepts. [Hilb.76, p.5]
And, later,
[t]he investigation on the foundations of geometry suggest the
problem: To treat in the same manner, by means of axioms, those
physical sciences in which mathematics plays an important part;
in the first rank are the theory of probability and mechanics.
(ibid., p.14)
His ideas on this respect were already expressed some years before; in
1894, after discussing the role of the AM in geometry, he stressed that
[n]ow also all other sciences are to be treated following the model
of geometry, first of all mechanics, but then also optics and elec-
tricity theory. (quoted from [Sau.98]).
We will give examples of axiomatics in the sciences in chapter 5, but
here we just mention that during the 20th century, much was done in the
directions pointed by Hilbert not only in physics, but also in other disci-
plines, such as biology, initiated by John Woodger (see [All.38] but with
several developments after him — see [Will.70], [Jong.85], [MagKra.01],
[Esan.13], where further references can be found. For a general setting
2.1. CONCRETE AXIOMATICS 35

in the beginnings of much of these applications, see the papers presented

in [Hen.et al.59]). As for the application of the AM in psychology, see
[Bog.79] for an account of Suppes’ works on the subject. But the most
investigated (from the axiomatic point of view) disciplines were those of
physics, such as classical particle mechanics [Sups.02], continuum me-
chanics (see [Ign.96]), thermodynamics, etc.
Now, for a sharper illustration of the advance in the direction of cres-
cent rigor and abstraction on the development of the AM, we shall distin-
guish among three levels of axiomatization. We notice in advance that this
classification is not the only one possible; Hilbert distinguished between
concrete axiomatics and formal axiomatics [Hil.96]; Jean Ladrière went
deeper by distinguishing among intuitive axiomatics, abstract axiomatics,
formal axiomatics, and pure formal systems [Lad.57, pp.36ff], but we think
that our three levels capture the distinctions we need. Our distinction will
be as follows: intuitive (or concrete) axiomatics, abstract axiomatics, and
formal axiomatics. Their meanings and examples are given in the next
sections.

2.1 Concrete axiomatics

In concrete axiomatic systems, which illustrates the way the Greeks used
the AM, the scientist has a well defined field of knowledge or domain of
application in mind: the axiomatics is meant to reflect the logical structure
of a (supposedly) well-known domain. The basic concepts of the axiomat-
ics are already interpreted in that domain; the basic propositions are true of
that domain. Then, the axiomatization aims at the organization of the field,
by presenting in an organized fashion its basic concepts, basic assumptions
and all the possible results that can be derived from this basis. In his paper
Axiomatic Thought, Hilbert put things this way:
When we assemble the facts of a definite, more-or-less compre-
hensive field of knowledge, we soon notice that these facts are
36 CHAPTER 2. AXIOMATIZATION

capable of being ordered. This ordering always comes about

with the help of a certain framework of concepts [Fachwerk von
Begriffen] in the following way: a concept of this framework
corresponds to each individual object of the field of knowledge,
and a logical relation between concepts corresponds to every fact
within the field of knowledge. The framework of concepts is
nothing other than the theory of the field of knowledge. [Hil.96,
pp.1107-8]

That is, by organizing a certain field of knowledge known in advance,

we enlighten its ‘frame of concepts’, which is of course a free choice of
ours (recall that Pieri had chosen a non usual class of primitive notions
than those used in most formulations of Euclidean geometry). In this
sense, axiomatization would not have much to do with the development
of the framework strictly speaking, except for the new theorems that can
be proved (but for some critics, as we shall see later, these theorem are
‘already implicit’ in the postulates, so nothing new would be being intro-
duced in the theory); it just cleans the house; as we shall see later, this is a
misconception.
Besides Euclid’s axiomatics of geometry, typical examples of concrete
axiomatization include arithmetics (in the sense we will discuss in a few
moments), which deals with (natural) numbers and their operations, and
Zermelo’s set theory, which axiomatizes the concept of set (collections of
objects) and objects which can be members of the sets. But let us begin
with a simple case, that of Patrick Suppes’ theory of human paternity just
to emphasize the technique [Sup.77]. Suppes had in mind a specific do-
main of application, namely, human beings. His theory does not apply to
neither worms nor plants, and can be summarized as follows. The primi-
tive notions are alive human being, male human being and father of, which
apply to two human beings (other choices are possible, but here we follow
Suppes).
The postulates are:
2.1. CONCRETE AXIOMATICS 37

(P1) If the human being a is father of the human being b, then it is not
the case that b is father of a.

(P2) All living human being have only one father who is a male human
being.

(P3) All living human beings have only one father who is not a male
human being.

The human being introduced by (P3) can be defined as mother of the

given human being. Other defined concepts are grand father, brother, sis-
ter, etc. Some theorems can be easily stated and proved: the father and the
mother of a given human being are different human beings; no human be-
ing is the father of itself, and so on. It is important to notice that, although
the axiomatics can have other models, in a sense to be made clear later, the
intended domain is already given in advance. In elaborating this theory, we
already know (by hypothesis) the main traits of human paternity theory.
Intuitive Peano’s arithmetic is our second example. The domain of ap-
plication is composed by the natural numbers, which for us includes zero
(Peano started with 1). Supposedly, we know already what to do with nat-
ural numbers, their main characteristics and operations. Then we may take
as primitive notions the following: zero and sucessor (of a natural number).
Defined concepts are for instance one, two, three, etc., prime number, even
number and so on. The typical postulates are:

(PA1) Zero is not a successor. In other words, there is no natural number

from which zero is a sucessor.

(PA2) If two natural numbers have the same successor, they are the same
natural number.

(PA3) If a certain property applies to natural numbers, then if zero has

this property and, whenever a natural number has the property it follows
that its successor has it also, then all natural numbers have that property.
38 CHAPTER 2. AXIOMATIZATION

This theory may have also other models (realizations of the axioms,
a concept here used informally, but which will be considered later); for
instance, think of the same sequence of the (intuitive) natural numbers, but
now call ‘zero’ the number 100, and as a sucessor of a certain number n,
the number n + 100. Then it is clear that the axioms are satisfied, and all
other concepts and operations can be adapted to this case. This would be
a strange arithmetics to be used in a drugstore, but from the mathematical
point of view that doesn’t matter. The system, as an interpreted calculus,
works! As we have seen, the nature of the models of a theory is something
to be carefully analyzed.
A third example of a concrete axiomatics is Zermelo’s set theory pre-
sented in 1908 [Zer.67]. Zermelo provided an axiomatic system for Can-
tor’s theory of sets, seen as collections of objects. The very notion of set
was put by Cantor himself as follows:

by an ‘aggregate’ (Menge) we are to understand any collection

into a whole (Zusammenfassung zu einein Ganzen) M of defi-
nite and separate objects m of our intuition or our thought. (our
emphasis [Low.14, p.85])

In other words, a set is a collection of objects of whatever sort with the

proviso that their elements are distinct from one another.8 We can form
sets of human beings, ants, angels, hurricanes, gods, prime numbers. But
of course Cantor was thinking (again!) of mathematics.9 However, as is
well known, the superb theory he created was inconsistent. The so-called
paradoxes of set theory clearly show this in a unequivocal way (a very clear
discussion can be found in [FraBarLev.73]).
To axiomatize set theory, Zermelo said that “[s]et theory is concerned
with a domain B of individuals, which we shall call simply objects, and
among which are the sets” [Zer.67]. Those individuals which are not sets
8
There is a theory of multisets which enables ‘sets’ to have repeated elements [Bli.88]. There are also
the quasi-sets whose elements may be indiscernible [FreKra.06].
9
For a good history of set theory, see [Dau.90].
2.1. CONCRETE AXIOMATICS 39

are the atoms (Urelemente in the German terminology). The fundamen-

tal relation is that which says that a certain individual a is an element or
belongs to a set b; here, we shall write a 2 b to use an updated notation.
This is the membership relation. As we shall see from the axiomatics, the
atoms may be elements of sets; hence, sets may have as elements either
atoms or other sets. Atoms do not have elements. Zermelo still introduces
the notion of subset: a set x is a subset of a set y if every element of x is an
element of y. In our updated terminology, we write x ✓ y to express that.
Needless to say that we are not using Zermelo’s original terminology.
The aim of Zermelo’s axiomatization was to avoid some inconsistencies
resulting from the far too liberal notion of set used in Cantor’s informal
theory. However, in stating his postulates, Zermelo also did not fulfill all
the logical details. For instance, in his Separation Axiom (see below),
the basis of his development, he used the rather vague notion of a definit
property in order to form sets from already given sets by ‘separating’ in the
given set those elements that fulfill the given property. It took some time
until Skolem and Fraenkel proposed to replace the notion of a ‘definite
property’ with a precise characterization, as we shall see below.
The postulates are as follows:
[Axiom of Extensionality] If two sets have the same elements, then they
are the same set. In other words, given two sets x and y, if every element
of x is an element of y and vice-versa, then x = y.
This axiom makes it useless that a set have repeated elements. If we
write (see below) {1, 2, 3} for the set comprising 1, 2, and 3 as its ele-
ments, then by force of this axiom it is identical to the set represented by
{1, 1, 2, 3, 3, 3} for they have the same elements.
[Axiom of the Elementary Sets] There exists a (‘fictitious’, according to
Zermelo) set, the null set that contains no element at all. Given an object
a, either a set or an ur-element, there exists a set {a} containing just a as
its element. Given objects a and b, there exists a set {a, b} whose only
elements are a and b and nothing else.
40 CHAPTER 2. AXIOMATIZATION

The set {a} is called the unitary of a, and can be derived from {a, b}
when a = b. Hence, it suffices to postulate the existence of the unordered
pair {a, b}. Uniqueness of the null set (or ‘empty set’ in modern language)
can be proved and it is denoted today by ;.
[Axiom (Scheme) of Separation] Whenever the formula F(x) (Zermelo
spoke in terms of ‘propositional function’) is definite for all elements of
a set x, then there is a subset y of x formed by precisely those elements of
x that fulfill the condition F(x).
As already remarked, the notion of ‘definite property’ was not satisfac-
torily clear. Today, it is common to axiomatize set theory as a first-order
system, as we shall see in the next chapter. Then, F(x) is a formula of
a first-order language with just one free variable, x. But, intuitively, the
result is the same: given a set x and a condition F, we can ‘separate’ from
x those elements that fulfill the condition and with them form another set
y. This trick is essential for avoiding the existence of enormous sets which
originate the paradoxes in Cantor’s theory, for instance, the set of all sets or
the set of all those sets not belonging to themselves. So, the whole domain
B of sets and the Urelemente is also not a set.10 Zermelo’s axiom is really
a scheme, for it potentially originates an infinity of axioms, one for each
formula F we use.
The notion of subset was mentioned already: a set x is a subset of a set y
if all elements of x are also elements of y. In today’s terminology, we write
x ✓ y. Some basic properties are immediately obtained: (a) the empty set
is subset of any set, (b) every set is a subset of itself, (c) if x ✓ y and y ✓ x,
then x = y, and so on.
[Axiom of the Power Set] To every set x there corresponds a set P(x), the
power set of x, whose elements are the subsets of x.
[Axiom of Union] To every set x there corresponds a set termed (x)
S
10
This should be qualified, for it will be important for us latter. The domain (so as the universe and many
other collections) are not Zermelo-sets, that is, sets in Zermelo’s theory — supposed consistent —, but
nothing excludes that they can be sets of another set theory. The notion of set is not absolute, and depends
on the theory we use.
2.1. CONCRETE AXIOMATICS 41

whose elements are the elements of the elements of x.

If the set x has just two sets a and b as elements, then it is common
today to write a [ b to indicate the union of x.
[Axiom of Choice] If x is a set whose elements are also non empty sets and
pairwise disjoint, then there exists a set y having one and only one element
in common with each element of x.
[Axiom of Infinity] There exists a set that contains the empty set and, hav-
ing a as an element, has also {a} as an element.
The natural numbers (Zermelo’s natural numbers) are defined (defined
concepts) as follows: 0 := ;, 1 := {0}, 2 := {{0}} = {1}, 3 := {{{0}}} =
{{1}} = {2}, etc. All set theory, including Cantor’s theories of transfinite
ordinals and cardinals, functions, orders and so on can be developed from
this axiomatic basis.
Again, we have here an axiomatics of a certain field known in advance,
namely, set theory. The theory just selects some collections to be sets, so
that the known paradoxes do not arise.11
Axiomatizations in the empirical sciences are in general concrete ax-
iomatics, for we are always thinking of an already known field of knowl-
edge, for instance Darwinian selection theory (see [Will.70]) or classical
particle mechanics [Sups.02], yet the resulting theories may have other
‘models’, as we shall see later.
Important to realize that different axiomatizations of the same domain
are possible. Depending on the chosen primitive notions and axioms we
use, different versions of the theories can be achieved. The discussion
of whether two different axiomatizations lead us to different theories was
central in the traditional debate about the nature of scientific theories (see
again our chapter 1); here, we assume that two different versions of group
theory, say by choosing different primitive concepts or axioms are versions
11
For these ‘known paradoxes’, the reader may have a look at the Introduction of [Men.97] and also in
[FraBarLev.73].
42 CHAPTER 2. AXIOMATIZATION

of the same theory. In the next section we will comment on this point a little
bit more.

2.2 Abstract axiomatics

A further step in the process of rigor in axiomatization appears with the
rise of abstract axiomatics. The reader should note that the three levels of
axiomatic theories we are considering are different levels of abstraction.
Abstract axiomatic systems have not a fixed domain of discourse, although
in general there is a domain which motivates the development of the ax-
iomatics. So, abstraction from a domain of application does not entail that
it has no preferred or intended domain. On the contrary, in general abstract
axiomatics arise from the study of a certain field, in the sense of a concrete
axiomatics, when the scientist realizes that it can have other models, or do-
mains of application, so she develops her theory in such a way that she may
cope with all these domains at once. In this sense, the axiomatic method is
said to provide for an economy in thought.
In the paper The architecture of mathematics [Bou.50], Bourbaki claims
that the rise of abstract axiomatics characterizes modern mathematics, for
the AM has the capacity of unifying fields which could in principle be
seen as distinct, so avoiding a ‘tower of Babel’ of disciplines. The in-
finitely many groups, for instance, so as the infinitely many linear (or vec-
tor) spaces, the infinitely many topological spaces, and so on, can now be
studied within a unified framework, one for each category of mathematical
object (groups, linear spaces, topological spaces, etc.).
Typical abstract axiomatic theories are those mentioned above as well
as algebraic rings, fields, and so on. Here the structural aspect of mathe-
matics appears for the first time in a clear way. For instance, group theory
can be characterized as the mathematical study of structures of the follow-
ing kind:12 there is a non empty domain G and a binary operation ◦ on G
12
The notion of structure shall be introduced precisely later. For now, let us take it intuitively as compris-
2.2. ABSTRACT AXIOMATICS 43

such that this operation is associative, admits a neutral element e and every
element a 2 G has an inverse −a 2 G relative to the operation, as we shall
see with more details below. Historical details about groups can be easily
found in the literature, going back to the studies of zeros of certain polino-
mial equations. But the very concept of group soon became abstract, in the
sense that neither the domain G nor the binary operation are specified, ex-
cept for a particular application. Groups are abstract mathematical struc-
tures of the form G = hG, ◦i. We may find infinitely many structures that
fulfill the definition, that is, cases of groups. The theory of groups studies
the mathematical properties of these entities also in an abstract way. Also,
group theory can be applied to several domains beyond mathematics, such
as physics; Hermann Weyl has pioneered the use of groups in quantum
mechanics.
Just to make justice to what was said in the end of the last section about
different ways of axiomatizing a same field, let us mention that we can
formulate the abstract notion of a group as follows: we take a structure
G = hG, ◦, e, −i comprising a non empty set G and a binary operation ◦
defined on G as before, but now we add other primitive elements to the
structure, say a designated element e of G and an unary function − from G
to G to form the inverse elements of the elements of G. Now it is enough
to adequately modify the axioms given. But the theory, here understood
as comprising all the results that can be obtained from these axiomatics,
is the same as before. Recall the thorny problems faced by the semantic
and syntactic approaches to theories when no such equivalence between
formulations is allowed.
Important to emphasize that in this kind of axiomatization, we do not
have a particular a priori intended domain of application. Linear spaces,
despite their motivations, have also infinitely many instantiations, or mod-
els. For instance, Hilbert spaces, which are important in quantum mechan-
ics, are linear (vector) spaces of a kind.
ing a set or some sets and operations and relations over the elements of these sets, plus some distinguished
elements — not all these objects may be present in every structure, of course.
44 CHAPTER 2. AXIOMATIZATION

Concrete axiomatic systems can be transformed in abstract systems

once we ‘abstract’ the primitive notions of their intuitive meaning. This is a
natural move to abstraction, and is precisely what the history of the method
tells us about its evolution. The rise of abstract mathematical structures
was achieved precisely when mathematicians noticed that their concrete
axiomatics could be applied to other domains beyond the intended one.
In abstract axiomatic systems, however, something is still kept in the
shadows, implicitly assumed: the underlying logic. By logic, except when
explicitly mentioned, we are being liberal, so that it involves either higher-
order logics or set theory (we shall leave category theory out of this dis-
cussion). Thus, in presenting group theory, we are neither making explicit
the mechanisms we use to make deductions nor explaining that the binary
operation ◦ is a mapping from the cartesian product G ⇥ G in G, that is,
a set. The different ways of axiomatizing and their relations to traditional
philosophical approaches to scientific theories will be discussed in a later
chapter.
Our second example of abstract axiomatics is that of a Hilbert space,
the kind of theory we will need when considering quantum mechanics.13 In
short, a Hilbert space is a linear vector space with an inner product which is
complete in the norm defined by the inner product. Let us be more specific.
A linear (or vector) space is a structure of the form E = hV, K, +, ·i, where:

(1) V is a non-empty set whose elements are called vectors. This name
has its origins in the intended model of linear spaces, namely, that where
the vectors are geometric vectors (sometimes also called Euclidean vec-
tors, that is, quantities having a length and a direction, frequently used
in physics and in vector algebra). We shall use small Greek letters like
↵, β, . . . for denoting vectors.

(2) K is a field, which by itself comprises other elements; K is by it-

self a mathematical structure K = hK, +K , ·K i obeying the postulates
13
Truly, there are formulations that do not use Hilbert spaces; see [Sty.02].
2.2. ABSTRACT AXIOMATICS 45

of fields; in most applications to physics, K is taken to be either the

field of the real numbers or the field of complex numbers, being this last
one the most relevant for quantum mechanics. The elements of the do-
main K of the field are called scalars and denoted by small Latin letters
a, b, x, y, . . ., sometimes with indexes.
(3) + is an application14 from V ⇥ V to V, called vector addition. The
image of the pair h↵, βi is written ↵+β and called the sum of the vectors ↵
and β. It is postulated that hV, +i is a commutative group whose neutral
element is called the null vector, denoted by O. The inverse of ↵ in this
group is written −↵.
(4) · is an application from K ⇥ V in V, called product of a vector by a
scalar. The image of the pair hx, ↵i is written x · ↵ or just x↵. In physics,
usually this is also written ↵ · x or just ↵x by an abuse of language. It
is postulated that this operation satisfies the following axioms, for any
↵, β 2 V, and x, y 2 K:
(a) x(↵ + β) = x↵ + xβ
(b) (x +K y)↵ = x↵ + y↵
(c) (x ·K y)↵ = x(y↵)
(d) 1↵ = ↵, being 1 the neutral multiplicative element of the field K.

Later we shall speak of particular linear spaces (or of ‘models’ of this

structure). We shall proceed here as does the standard mathematician. By
an abuse of language, we follow the standard terminology and say that W
is a linear (or a vector) space over the field K. When K = R, that is, the
field is the field of real numbers, we speak of real linear spaces and in
complex linear spaces when K = C.

Definition 2.2.1 (Restriction of an operation) Let X be a set and ◦ a bi-

nary operation on X. Denoting as usual by a ◦ b the image of the pair
14
Here we are taking the words ‘application’, ‘function’ and ‘mapping’ as synonymous.
46 CHAPTER 2. AXIOMATIZATION

ha, bi 2 X ⇥ X by the mapping ◦, let Y ✓ X. We call the restriction of the

operation ◦ to Y, denoted ◦Y the operation defined by a ◦Y b iff a ◦ b and
a, b 2 Y.

In other words, it is ‘the same’ operation but now considered only in

relation to the elements of the subset Y. It is easy to extend this definition
to most general cases as those used in the next definition.

Definition 2.2.2 (Subspace) A structure E0 = hW, K, +W , ·W i is a sub-

space of the linear space E = hV, K, +, ·i if W ✓ V and +W and ·W are
the restrictions of the operations of E to the sets in E0 is also a linear space
over the same field K.

The usual parlance in mathematics does not distinguish the notation, so

identifying the symbols for the operations and for their restrictions, leaving
distinction to the context.
In order to obtain a Hilbert space in which we may develop quantum
mechanics, we fix a complex linear space (hence henceforth K = C) and
enlarge the structure with an additional operation called an inner product,
which is a mapping now from V⇥V to C. The image of the pair h↵, βi will
be denoted by h↵|βi. We call attention to the differences of notation: h↵, βi
is an ordered pair of vectors, while h↵|βi is a scalar (a complex number).
The postulates to be satisfied by this new notion are the following ones,
holding for all vectors and scalars:

(i) h↵|β + γi = h↵|βi +K h↵|γi

(b) h↵|xβi = x ·K h↵|βi
(c) h↵|βi = hβ|↵i⇤ , where the exponentiation ⇤ indicates the complex
conjugation (remembering that is z = a + bi, then z⇤ = a − bi).
(d) h↵|↵i ≥ 0 and h↵|↵i = 0 iff ↵ = O.

Then we introduce the following definition:

2.2. ABSTRACT AXIOMATICS 47

Definition 2.2.3 (Norm) If ↵ 2 V, the norm induced by the inner product

of the vector ↵, denoted k↵k, as follows:
p
k↵k := h↵|↵i.

It results that the norm of a vector is a real number. This norm is said to
be the norm induced by the inner product.15 With this notion, the concept
of distance between two vectors ↵ and β can be defined as follows:

Definition 2.2.4 (Distance) The distance between ↵ and β, denoted d(↵, β)

is given by
d(↵, β) := k↵ + −βk = k↵ − βk.

We just remark that, concerning the vector −β, the following identity
can be easily checked: −β = (−1)β. Let ↵1 , ↵2 , . . . be a sequence of vectors
in V (denoted by (↵i )), and let β a vector in V. Then

Definition 2.2.5 (Convergence) We say that the sequence (↵i ) converges

to β iff for any real number " > 0, there exists a natural number n such
that for any j > n, d(↵ j , β) < ".

Definition 2.2.6 (Cauchy sequence) A sequence (↵i ) of vectors is a Cauchy

sequence if for any real number " > 0 there exists a natural number n such
that for any j, k > n, we have that k↵ j − ↵k k < ".

Every Cauchy sequence converges, as it can be proved. The problem is

that if we suppose a collection of vectors of V and assume that all vectors
of the sequence are in this set, the vector to which it converges may not be
in this set. In particular, this set of vectors, endowed with the restrictions
15
Generally speaking a norm on a linear space is a mapping k k from V in C which associates to any
vector ↵ a scalar k↵k, such that, for any vectors ↵ and β and scalar x, we have: (a) k↵k ≥ 0 and k↵k = 0
iff ↵ = O, (b) kx↵k = |x|k↵k, where |x| is the modulus of x, and (c) k↵ + βk  k↵k + kβk. There are several
different norms over a linear space, but we are interested in the
p one given by the above definition. A further
remark: being x a complex number x = a + bi, then |x| = a2 + b2 . The reader should notice that in the
definitions given in this note we are not making explicit differences between the operations among vectors
and among scalars; by the way, this is the standard procedure we shall pursue from now on.
48 CHAPTER 2. AXIOMATIZATION

of the linear space operations may be also a linear space. In this case, we
say that the corresponding structure is a subspace of the given linear space.
Then we have the main definition:

Definition 2.2.7 (Hilbert space) A Hilbert space is a linear space with an

inner product such that all Cauchy sequences of vectors converge to a vec-
tor still in the space.

A subspace of V which obeys this condition is termed a closed sub-

space. The Hilbert spaces are the mathematical frameworks used in the
most usual formulations of non-relativistic quantum mechanics (see the
formulation of quantum mechanics in chapter 5).
Beyond the particular examples developed, which are useful for the dis-
cussions to come, the important thing at this point is to note that the math-
ematical structures being obtained (groups, linear spaces, Hilbert spaces)
are abstract in the sense of our characterization. They don’t have a par-
ticular domain being supposed in advance. But the formulations are still
supposing the use of concepts like mappings, sets and so on, and the proofs
(if we would provide some) assume an implicit underlying logic. So, we
need to go further.

2.3 Formal axiomatics

Formal axiomatic theories go deeper in abstraction and rigor by making the
underlying logic totally explicit and, in general, by using formal languages
to express the propositions of the theory. Being explicit about the logic
is really necessary mainly because the logic employed could not be stan-
dard (classical) logic. The relevance of making logic explicit became clear
during the last century when several deep metamathematical results where
achieved. Gödel’s incompleteness theorems, Tarski’s theorem on the un-
definability of the notion of truth, the existence of non-standard models
of Peano’s arithmetics, the independence proofs in set theory and much
2.3. FORMAL AXIOMATICS 49

more were possible only after the full understanding of the logic underly-
ing classical mathematical theories was achieved. In particular, the rise of
non-classical logics was possible only after logic achieved total formaliza-
tion, so that several other systems could be developed either by extending it
(‘complimentary logics’) or by modifying some of its basic laws (‘hetero-
dox logics’). On what concerns empirical sciences, it is still a challenge to
convincingly argue that physics, for instance, really needs a non-classical
logic. We shall return to this point later.
For the moment, let us present here the main systems we shall be deal-
ing with in this book as examples of formal axiomatics. The systems below
are all ‘classical’, that is, grounded on classical first-order logic, so let us
present it first. By doing so, we keep the text self-contained.

2.3.1 First-order classical logic with identity

This logic, let us call it L1 , will be presented through the ‘linguistic’ ap-
proach,16 by describing a basic vocabulary, rules of formation (the gram-
matical part of L1 ), and its postulates. Let us call the language L. The basic
(primitive) vocabulary of L comprises the following categories of primitive
symbols (as is common in these presentations, we do not distinguish be-
tween use and mention):

1. The propositional connectives: ¬ and !

2. A denumerable collection of individual variables: x0 , x1 , x2 , . . .. We

shall use x, y, z, . . . as metavariables for individual variables.

3. The universal quantifier: 8

4. Auxiliary symbols: left and right parentheses and the comma.

16
An alternative approach would be ‘algebraic’, seeing a logic as an ordered pair (an algebra) L = hF, `i,
where F is a set whose objects are called formulas and ` is a function from P(F) to F, the deduction
operation, satisfying well-known postulates.
50 CHAPTER 2. AXIOMATIZATION

5. A collection (possibly empty) of individual constants: a1 , a2 , . . .. We

use a, b, c . . . as metavariables for individual constants.
6. For any natural number n > 0, a collection of predicate symbols of
rank n: Pn1 , Pn2 , . . .. We use F, P, Q, . . . as metavariables for predicate
symbols, and the context will indicate their rank.
7. For any natural number n > 0, a collection of functional symbols
(symbols for operations) of rank n: f1n , f2n , . . .. We use f, g, h, . . . as
metavariables for operations.
8. The identity symbol: =
The symbols in 5, 6, and 7 are called specific or non-logical symbols,
and depend on the theory having L1 as underlying logic, as we shall see
below with some examples. The other symbols are the logical symbols and
can be taken as being the same for all theories.17
The grammar is described as follows: firstly, we define the expressions
of L as sequences of finite symbols of L written horizontally from left to
right. For instance, ¬((x1 x2 ! A34 (8 is an expression.
Definition 2.3.1 (Terms of L) The terms of the language are the individ-
ual variables, the individual constants and the expressions of the form
f (t1 , . . . , tn ), where f is a functional symbol of rank n and t1 , . . . , tn are
n terms. These are the only terms.
Definition 2.3.2 (Formulas) The formulas of L (well-formed expressions)
are defined by the following clauses: (1) If t1 and t2 are terms, then t1 = t2
is a formula, called ‘atomic’. (2) If F is a predicate symbol of rank n and
t1 , . . . , tn are terms, then F(t1 , . . . , tn ) is a formula (also ‘atomic’). (3) If ↵
and β are formulas, then expressions like (¬↵) and (↵ ! β) are formulas.
(4) If x is an individual variable and ↵ is a formula, then 8x↵ is a formula.
These are the only formulas of L.
17
Of course there are here also different ways of formulating L1 , say by choosing a distinct set of logical
symbols, but we shall omit the details.
2.3. FORMAL AXIOMATICS 51

In writing formulas, we make a standard convention for elimination

of parentheses: for instance, we can eliminate the external parentheses in
writing formulas; so, ¬↵ abbreviates (¬↵) and ↵ ! β abbreviates (↵ !
β). Parentheses are useful for avoiding ambiguities in writing formulas.
Notice that ↵ and β are also metavariables for formulas. Symbols like
these (Greek lowercase letters) will appear below; thus, the expressions we
shall call ‘axioms’ are in reality axiom schemata, which enable us to obtain
axioms when uniform substitutions of formulas for these metavariables are
performed.
Some additional syntactical definitions and conventions are in order to
facilitate the exposition. An occurrence of a variable x in a formula is
bound if x is the variable that appears just after the universal quantifier
or if it occurs in a formula which is affected by (is in the scope of) the
quantifier. Otherwise, it has a free occurrence in the formula. For instance,
in the formula 8x(x = y) ! 8y(x = y), the two first occurrences of x
are bound and the third one is free, while the first occurrence of y is free
and the last two are bound. If a formula ↵ has xi1 , . . . , xik among its free
variables, we write ↵(xi1 , . . . , xik ). So, ↵(x) stands for a formula which has
free occurrences of x.
A term t is free for the variable x in the formula ↵ if no free occurrence
of x in the formula ↵ lies within the scope of a quantifier 8y, where y is a
variable occurring in t. For instance, being f a functional symbol of rank
2, then f (x, y) is free for x in the formula 8z(x = z) ! (x = w), but is not
free for x in ¬8y¬(x = z) ! (x = w).18
The postulates of L1 are

(L1) ↵ ! (β ! ↵)

(L2) (↵ ! (β ! γ)) ! ((↵ ! β) ! (↵ ! γ))

(L3) (¬↵ ! ¬β) ! ((¬↵ ! β) ! ↵)

18
This example is adapted from [Men.97, p.69], where further details and examples can be found.
52 CHAPTER 2. AXIOMATIZATION

(L4) 8x↵(x) ! ↵(t), where t is a term free for x in ↵(x).

(L5) 8x(↵ ! β) ! (↵ ! 8xβ), provided that ↵ does not contain free

occurrences of x.

(L6) 8x(x = x)

(L7) x = y ! (↵(x) ! ↵(y)), where ↵(y) arises from ↵(x) by replacing

some free occurrences of x by y.

Furthermore, our system comprises the following inference rules, where

` is the symbol of deduction:

(MP) (Modus Ponens) ↵ ! β, ↵ ` β

(Gen) (Generalization) ↵ ` 8x↵

The notion of deduction is crucial in characterizing a logic. If we mod-

ify it, we change the logic. The standard definition is the following one,
where Γ denotes a set of formulas:

Definition 2.3.3 (Deduction) We say that a formula ↵ is deduced, derived,

inferred, from a set Γ of formulas (the premises of the deduction) if the
following clauses are obeyed:

(i) There is a sequence of formulas β1 , β2 , . . . , βn such that:

(ii) βn is ↵

(iii) Each βi (i < n) of the sequence is either

(a) an axiom, or
(b) an element of Γ, or
(c) immediate consequence, by one of the rules of inference, of pre-
ceding formulas in the sequence.
2.3. FORMAL AXIOMATICS 53

If these clauses are fulfilled, we write

Γ`↵

There are situations in which Γ = ;, that is, there are no premises in the
deduction other than the axioms of the logic. In this case, we write

`↵

and say that ↵ is a thesis or a formal theorem of the logic.

Further useful definitions are:

Definition 2.3.4 (Other connectives and the existential quantifier) Taking

into account our conventions for the use of parentheses, we put:
(i) ↵ ^ β := ¬(↵ ! ¬β)
(ii) ↵ _ β := ¬↵ ! β
(iii) ↵ $ β := (↵ ! β) ^ (β ! ↵)
(iv) 9x↵ := ¬8x¬↵

We get a first-order theory when we specify a particular choice of the

so-called non-logical symbols, namely, individual constants, predicate sym-
bols, and functional symbols and add further specific axioms (or proper
axioms) for the particular theory. Thus, the axioms (or postulates) of a
first-order theory will be those of L1 plus those of the specific theory. In
this sense, L1 is by itself a first-order theory, with no additional proper
postulates.
For instance, we can formulate Suppes’s theory of human paternity as
a first-order theory as follows. We assume all the logical vocabulary given
above of a first-order logic, and consider the following specific primitive
symbols: a binary predicate P and two unary predicates L and M. The
only terms are the individual variables x, y, z, . . .. The atomic formulas are
54 CHAPTER 2. AXIOMATIZATION

x = y and the expressions of the form P(x, y), L(x) and M(x). The intended
interpretation of these formulas is obvious: x is a father of y, x is a living
human being, and x is a male human being. The postulates, in addition to
those given above, may be written as follows:

(P1) 8x8y(P(x, y) ! ¬P(y, x))

(P2) 8x(L(x) ! 9!y(M(y) ^ P(y, x)))
(P3) 8x(L(x) ! 9!y(¬M(y) ^ P(y, x)))

Here, 9!x means ‘there exists just one x’.

The next example of a first-order theory is Peano’s Arithmetics. We
abbreviate ¬(t1 = t2 ) by t1 , t2 .

2.3.2 Peano Arithmetics

The underlying logical apparatus of the theory PA is the one presented
above. We maintain the same logical symbols, but the specific symbols are
the following: (1) an individual constant ‘0’, (2) a unary functional symbol
‘σ’, and (3) two binary functional symbols ⊕ and ⌦. Hence, besides the
individual variables, the terms of the language of PA are obtained from
expressions like 0, σt where t is a term, and t1 ⊕ t2 and t1 ⌦ t2 , for t1 and t2
any terms.
The postulates of PA are (L1)–(L7) above plus the following ones:

(PA1) 8x(0 , σx)

(PA2) 8x8y(σx = σy ! x = y)
(PA3) If ↵(x) is a formula with just x free, then

↵(0) ^ 8x(↵(x) ! ↵(σx)) ! 8x↵(x).

(PA4) 8x(0 ⊕ x = x)
2.3. FORMAL AXIOMATICS 55

(PA5) 8x8y(x ⊕ σy = σ(x ⊕ y))

(PA6) 8x(0 ⌦ x = 0)
(PA7) 8x8y(x ⌦ σy = σ(x ⌦ y) ⊕ x)

The reader may easily note that (PA1)–(PA3) are possible formaliza-
tions of those axioms informally stated before. The additional postulates
are necessary because we don’t have in the theory the resources to prove
that the functions representing the operations ⊕ and ⌦ can be defined (which
requires the so-called Theorem of Recursion — see [End.77]). In short:
with the help of this axiom, we can prove that the functions that define
addition and multiplication exist and are unique. But, in first-order logic
this proof is not possible. According to Edmund Landau, this fact was
observed for the first time by Grandjot [Lan.66, p.ix].
Let us give an example of a proof which will show the differences be-
tween the ways we usually proceed in abstract and in formal axiomatics. In
group theory we can prove the following theorem, known as the cut rule,19
which can be stated in group theory as a theorem:

Theorem 2.3.1 (Cut Rule) In every group G = hG, ◦i, for all elements
a, b, c 2 G, we have that c ◦ a = c ◦ b entails a = b.
Proof: (Informal) Let us suppose c ◦ a = c ◦ b. Since every element of
G has an inverse still in G, then c has an inverse c0 . So we can write
c0 ◦ (c ◦ a) = c0 ◦ (c ◦ b) once we can operate any two elements of G (in
fact, c0 , c ◦ a and c ◦ b are elements of G, so they can be operated — but
note the lack of something here: how do we know that in writing c0 before
both terms in the equality, the equality still remains? This can be seen only
with the resources of logic, given below). By the associative axiom, this
last expression is equivalent to (c0 ◦ c) ◦ a = (c0 ◦ c) ◦ b). But, since c0 is
the inverse of c, it results from the last equality that e ◦ a = e ◦ b. So, using
the axiom of the neutral element, we get a = b.
19
Don’t confuse with Gentzen’s cut rule!
56 CHAPTER 2. AXIOMATIZATION

Now we present a detailed (and also boring) proof with all logical de-
tails. The main logical results to be used are the reflexive law of iden-
tity L6, namely 8x(x = x), the substitutive law of identity L7, that is
x = y ! (↵(x) ! ↵(y)), where ↵(x) is a formula with a free occurrence
of x, and ↵(y) is obtained from ↵(x) by the substitution of y in some free
occurrences of x, and the inference rule Modus Ponens (MP). The proof
goes as follows:

Theorem 2.3.2 (Cut Rule) In every group G, for all elements a, b, c 2 G,

we have that c ◦ a = c ◦ b entails a = b.
Proof: (logic; in the right, we mention the used postulates)
1. c ◦ a = c ◦ b (Assumption)
2. (c ◦ a = c ◦ b) ! ((c0 ◦ (c ◦ a) = c0 ◦ (c ◦ a)) ! (c0 ◦ (c ◦ a) = c0 ◦ (c ◦ b)))
(1, L7), with ↵(a) being (c0 ◦ (c ◦ a) = c0 ◦ (c ◦ a))
3. c0 ◦ (c ◦ a) = c0 ◦ (c ◦ a) (L6)
4. (c ◦ a = c ◦ b) ! (c0 ◦ (c ◦ a) = c0 ◦ (c ◦ b))) (2, 3, Propositional logic
and MP, that is, the tautology (↵ ! (↵ ! β) ! (↵ ! β))).
5. c0 ◦ (c ◦ a) = c0 ◦ (c ◦ b) (1, 4, MP)
6. (c0 ◦ c) ◦ a = (c0 ◦ c) ◦ b (5, G1)
7. c0 ◦ c = e (G3)
8. (c0 ◦ c = e) ! (((c0 ◦ c) ◦ a = (c0 ◦ c) ◦ b) ! (e ◦ a = e ◦ b)) (7, L7)
9. (((c0 ◦ c) ◦ a = (c0 ◦ c) ◦ b) ! (e ◦ a = e ◦ b) (7,8, MP)
10. e ◦ a = e ◦ b (6,9, MP)
11. e ◦ a = a (G2)
12. e ◦ b = b (G2)
2.3. FORMAL AXIOMATICS 57

13. (e ◦ a = a) ! ((e ◦ a = e ◦ b) ! (a = ◦b)) (11, L7)

14. (e ◦ a = e ◦ b) ! (a = e ◦ b) (12, 13, MP)

15. a = e ◦ b (10,14, MP)

16. (e ◦ b = b) ! ((a = e ◦ b) ! (a = b)) (15, L7)

17. a = e ◦ b ! a = b (12, 15, MP)

18. a = b (15, 17, MP)

The differences between informal (that is, not making the logical steps
explicit) and formal proofs was discussed by Patrick Suppes in a very in-
teresting and important chapter of his book [Sups.57, Chap.7]. Important
to note that it would be practically impossible to present all mathematical
deductions this way; for instance, in their masterpiece Principia Mathe-
matica, Whitehead and Russell present the proof that 1 + 1 = 2 in the
page 379! [WhiRus.10]. The amount of previously required definitions
and proofs is overwhelmingly big. Anyway, it is beyond doubt that the
logical analysis of mathematical proofs is important, so that Proof Theory,
the part of logic that deals precisely with this kind of analysis is one of
the fundamental subjects of modern logic. What is relevant here is that
the above theorem does not hold only in a particular group, but it is valid
for every group. This generalization is one of the main traits of AM, that
is, the possibility of treating different fields of knowledge at once. What
holds as a theorem for group theory holds for all groups, and the same can
be said of all other abstract axiomatic systems.
Furthermore, without the development of modern logic and the full
knowledge of the abstract and formal axiomatic systems, we could not ar-
rive at strong results such as Gödel’s, Tarski’s and others, like the indepen-
dence of axioms in set theory which, as an example, gave us the possibil-
ity of finding alternative (‘non-cantorian’, in Cohen’s words) mathematics.
Thus, the axiomatic method does involve heuristics. But this is a challenge
58 CHAPTER 2. AXIOMATIZATION

for chapter 4. Before doing that, let us discuss a specific formal theory
which shall have great relevance for our discussion on axiomatization and
models: elementary set theory.
Chapter 3

A mathematical background

e shall now outline a mathematical framework where the

W formal counterpart of scientific theories can be discussed,

namely, the Zermelo-Fraenkel set theory, ZF, perhaps
with the Axiom of Choice, ZFC. Since the relevance of
this subject is in general not well known by the general philosopher of sci-
ence, we provide some further explanations about the details of the theory,
hoping that they will help the reader to agree with us on the importance of
this kind of study to our subject.
We write ZF(C) to refer to both of the mentioned theories; when a dis-
tinction between these theories is required, we shall make explicit refer-
ence. Of course, other frameworks could be used instead; there are many
options for expressing at least part of present day mathematics, such as
higher-order logics, as in Carnap [Car.58] (with some examples of empiri-
cal theories), or category theory [Ger.85].
For the sake of simplicity, and to follow common usage, we have chosen
to treat ZF(C) as a first-order theory, and its language, presented below
(section 3.2), will be called L2 .1 But before describing ZF(C), it will be
1
In fact, we could have used a different language, for instance, by adopting distinct primitive symbols
and perhaps other postulates (which would be in most cases equivalent to the ones presented here). There
is also the more relevant possibility of constructing a theory grounded on second-order logic, as Zermelo
seems to have preferred—see [Moo.82, pp.267ff]—(or still higher-order) logic, and in this case in fact we
should have different theories, for their properties would certainly be distinct.
60 CHAPTER 3. A MATHEMATICAL BACKGROUND

profitable to discuss a little bit about the construction of axiomatic/formal

systems.

3.1 The Principle of Constructivity

In his book Ensaio sobre os Fundamentos da Lógica (‘Essay on the Foun-
dations of Logic’),2 da Costa calls the following methodological rule ‘the
Principle of Constructivity’:

The whole exercise of reason presupposes that it has a certain

intuitive ability of constructive idealization, whose regularities
are systematized by intuitionistic arithmetics, with the inclusion
of its underlying logic. [Cos.80, p.57]

What does he mean by that? The explanations appeared earlier in his

book, and we base our explanation on his ideas, with the risk of not being
able to do justice to much of the rich and clear contents of the book. In this
section, when we make reference to a certain page only (say, by writing
‘page 51’), it refers to the mentioned book by da Costa. To speak a little
bit of such matters is of course quite important, for we intend to present
a formal system (ZF) using expressions such as ‘infinite set of individual
variables’, for instance, which presuppose a previous notion of the con-
cept of ‘infinity’; hence, the construction of set theory as a formal theory
apparently presupposes at least part of the very mathematics it intends to
ground.
But let us pay close attention to da Costa:

Formal disciplines are essentially discursive. But the discourse

develops itself in different levels, and each one of them must be
understood, or intuited, as already noticed by Descartes. Even if
2
There is a French version of this book, translated by Jean-Yves Béziau, called Logique Classique et
Non-Classique, Paris, Flammarion, 1996.
3.1. THE PRINCIPLE OF CONSTRUCTIVITY 61

one reasons symbolically and formally, the different elementary

steps of the evolution of the discourse need to be clear and ev-
ident, for in the contrary there would be no reasoning, and one
would not know what she is doing. (p.50).

In trying to systematize the empirical reality, we make use of our capac-

ity of reasoning. We use intuition and other resources, such as previously
learned theories, personal experiences, memory, imagination, expertise,
insights. In a certain sense, all of these previous knowledge depend on
the present day state of the evolution of science, as well as of our own
biological and cultural characteristics. But even intuition is not enough.
We need to systematize our intuitions, and in mathematics, the axiomatic
method became the (apparently) best methodological tool, being extended
to empirical science, as we have already seen. Well developed sciences,
or disciplines, even though they are not completely axiomatized or formal-
ized yet, can be treated, from a mathematical perspective, in the same way
as the formal sciences, being also essentially discursive. In other words,
scientific knowledge, being essentially conceptual knowledge, needs dis-
course, and our ‘discourse’ needs language and symbolism (p.35) and, we
can add, our inferences depend on logic.3
In a certain sense, we may say that our knowledge of a certain scientific
domain is given by means of the elaboration of structures; we ‘structure’
the domain by gathering concepts, which may be relations and operations
over a certain domain (or domains) of objects we are interested in. Even
if we do it only informally, in reality we are elaborating structures which
concentrate what we think are the basic concepts we intend to use in our
approach (we have already said something on this respect in the Preface).
Certainly, different scientists can choose different concepts. Then we can
work ‘informally’, by providing the relevant definitions and theorems, or
more formally, by providing to these concepts suitable (even if informally
3
In a broad sense, we use not only deductive inferences when elaborating a theory, but also (at least)
inductive and abductive, to use Peirce’s teminology [Dou.11].
62 CHAPTER 3. A MATHEMATICAL BACKGROUND

stated) postulates we think may govern them with respect to our intended
domain(s). Thus, we may say that our scientific knowledge is concep-
tual and structural. Sometimes these ‘postulates’ are not even formulated
explicitly. There are examples in mathematics (and much more in the em-
pirical sciences) where a certain field is not presented axiomatically in the
sense described above. For instance, consider analytic geometry as pre-
sented in a first undergraduate course. It has no axioms properly speaking.
The same happens to differential and integral calculus. These disciplines
are developed from definitions to theorems. But, as Alonso Church has
made clear, every axiomatic theory can be transformed into a ‘definitional’
theory, comprising just definitions and theorems. Thus, in a certain sense,
both analytic geometry and the calculus can be seen as ‘axiomatic’ too.
Going back to the point in discussion, we recall, as is well known, that
in intuitionistic mathematics we have an intuitive ‘visualization’ of the en-
tities that interest us (p.50). This is essentially an intellectual intuition, an
expression that intends to capture the idea that
there cannot be immediate and evident knowledge without con-
templation, without a look to the objects that interest us or, at
least, of the conceptual relations which define them; in an analo-
gous way, there is no intellectual contemplation which does not
enable us to formulate direct judgments, linked to different lev-
els of evidence. (p.51).
The reference to intuitionistic mathematics is grounded on the fact that,
according to da Costa, it provides an intuitive ‘visualization’ of the entities
that interest us (p.52), contrarily to standard mathematics, where such an
intuition needs not be available (p.54). He illustrates the issues with exam-
ples of the following kind: we have no clear ‘vision’ neither of transfinite
cardinals, nor of the totality of the real numbers, but only an intuition of
the system of relations that implicitly define these concepts by means of
axiomatic systems (p.54). This ‘intuitive visualization’ provides us with
an intuitive pragmatic nucleus and we may say that, based on this nucleus,
3.1. THE PRINCIPLE OF CONSTRUCTIVITY 63

we articulate a kind of algebra (da Costa doesn’t use this word in this con-
text) that enables us to compose them, operate with them and so on, going
to more sophisticated and sometimes not intuitively evident conceptualiza-
tions. This way, we go out of the intuitive nucleus, and in flying so high,
the axiomatic method is our ‘autopilot’, which enables us to fly in domains
where our intuitions do not help us much.4
In this sense, we begin by describing a formal system using this intuitive
nucleus of finitist and constructive nature; as da Costa says, “it is today
universally accepted that there cannot be formalized arithmetics without
intuitive arithmetics” (p.57). It is this informal and intuitive handling of
symbols and concepts that, at first glance, enable us to make reference to
the tools we need to characterize our axiomatic/formalized theories. Thus,
it is in this sense that we formulate the language of ZF(C) described be-
low; in saying, for instance, that we have a denumerable infinite collec-
tion of individual variables, we could say that our language encompasses
two symbols, x and 0 , and the variables would be expressions of the form
x, x0 , x00 , . . ., and the same for the individual constants. This way, we avoid
speaking of idealized concepts (in Hilbert’s sense, such as ‘denumerable
many’) and keep ourselves in a constructive setting. Interesting as it is, we
had enough for this kind of discussion.
It is interesting that similar ideas are also advanced by Kenneth Kunen
in his book The Foundations of Mathematics [Kun.09], referring specifi-
cally to logic. Speaking about how set theory is developed, Kunen says
that

[y]ou don’t need any knowledge about infinite sets; you could
learn about these as the axioms are being developed; but you
do need to have some basic understanding of finite combina-
torics even to understand what statements are and are not ax-
ioms. [. . . ] This basic finitistic reasoning, which we do not
4
For instance, usually we reason more or less well in the domains where we suppose classical logic
works, but if we need to go to other domains, logic needs to be made explicit.
64 CHAPTER 3. A MATHEMATICAL BACKGROUND

analyze formally, is called the metatheory. In this metatheory,

we explain various notions such as what a formula is and which
formulas are axioms of our formal theory, which here is ZFC.
[Kun.09, pp.28-9]

We can interpret these words as indicating that we start with intuitive

notions, necessary, say, to distinguish between two different symbols (like
a and b). So we are presupposing things like the intuitive meaning of the
number two, so as the idea of ‘different’ things. Then we start to combine
these symbols in a way that enables us to elaborate sophisticated mathe-
matical languages, such as that of ZF(C). Then, in a second stage, we work
inside ZF(C) and, with its tools, we can reconstruct the steps we have gone
through previously only at an informal level and, in particular, explain rig-
orously what two may be taken to be. In this sense, as says Kunen, “formal
logic must be developed twice” (ibid., p.191).
The metatheory of course can also be treated formally, but in order to
do that we would need to have available once again, as a prerequisite, an
informal metametatheory, and so on. But let us turn now to ZF and ZFC
set theories.

3.2 The ZF(C) set theories

Here we shall present two theories, ZF and ZFC. Some redundancies can
be found by the reader with things already exposed before. But our aim
is to keep this chapter independent, as it were, so we shall not bother with
these repetitions. The reader must also be not confused with the terminol-
ogy: ZF is the first-order Zermelo-Fraenkel set theory without the Axiom
of Choice, while ZFC is the theory obtained from ZF by adding to it the
axiom of choice. The underlying formal language is the same for both
theories, and they differ just by the mentioned axiom, so most of the de-
velopments may be carried on for both theories alike, with the relevant
differences being mentioned when necessary. Important to note, as stated
3.2. THE ZF(C) SET THEORIES 65

by Gödel and Cohen, the axiom of choice cannot be neither proved nor
disproved from the axioms of ZF (supposed consistent).
The language L2 of ‘pure’ ZF(C) (that is, not comprising Urelemente
— see below) has the following categories of primitive symbols:

(i) The propositional connectives: ¬ and !

(ii) The universal quantifier: 8

(iii) Two binary predicates: = and 2

(iv) Auxiliary symbols: left and right parentheses: ( and )

(v) A denumerable infinite set of individual variables: x1 , x2 , . . ., which

in the metalanguage we will denote (provided that there is no risk of
ambiguity) by x, y, z, w, u, v, etc.

The language L2 is a denumerable language (its primitive symbols,

given by the above list, can be enumerated) and its syntax, to be intro-
duced now, is recursive, which intuitively means that we can program a
computer to recognize whether a finite sequence of primitive symbols is a
formula in the sense defined below. First, let us say that the only terms of
this language are the individual variables.

Definition 3.2.1 (Formulas) The class of formulas is defined recursively

as follows:

(i) If x and y are individual variables, then expressions of the form x 2 y

and x = y are atomic formulas.

(ii) If ↵ and β are formulas, so are all expressions of the form ¬↵, (↵ !
β) and 8x↵, where x is an individual variable.

(iii) There are no other formulas.

66 CHAPTER 3. A MATHEMATICAL BACKGROUND

Frequently we omit parentheses when the reading does not lead to am-
biguity, and we reintroduce them in order to avoid ambiguity, according to
standard conventions (see [Men.97]). We remark that ↵, β and other sym-
bols not listed above do not make part of the basic language; they may be
seen as metavariables. In the case, small Greek letters stand for formulas,
and big Greek letters (as Γ, ∆) stand for collections of formulas. A further
remark: in general, we do not work with L2 properly speaking, but with
definitional extensions of it, that is, with languages comprising more sym-
bols than the primitive ones, for instance those given by the next definition,
so as with other common symbols, like ✓, @0 , R, ;, and so on.5 Anyway,
we shall continue to speak about L2 as the language posed above or one of
its definitional extensions.

Definition 3.2.2 (Abbreviations) Some abbreviations, given in the meta-

language, are in order;6
(i) ↵ ^ β := ¬(↵ ! ¬β)
(ii) ↵ _ β := ¬↵ ! β
(iii) ↵ $ β := (↵ ! β) ^ (β ! ↵)
(iv) 9x↵ := ¬8x¬↵
(v) (8x 2 y)↵ := 8x(x 2 y ! ↵)
(vi) (9x 2 y)↵ := 9x(x 2 y ^ ↵)

In the expression 8x↵, the formula ↵ is called the scope of the quantifier
(the same holds for 9, since it is just an abbreviation). A variable x in
a formula ↵ in the scope of a quantifier 8x or 9x is said to be a bound
variable. A free occurrence of a variable x in a formula ↵ is an occurrence
of x which lies outside the scope of a quantifier 8x (idem for 9x). In this
5
We could add these new symbols for relations, operations or individual constants to the language L2 ,
but with some care. The details about how this can be done is presented in [Sups.57, Chap.7].
6
The symbol “:=” means ‘equal by definition’.
3.2. THE ZF(C) SET THEORIES 67

case, x is said to be free in ↵. We frequently write ↵(x1 , . . . , xn ) to indicate

that all free variables of ↵ occur in the list x1 , . . . , xn . A sentence is a
formula without free variables.
Let ↵(x) be a formula of L2 in which x is the only free variable. We shall
call this formula a condition. In the intuitive theory of sets (which admits
as ‘sets’ whatever collection of objects you wish), we have the following
principle, called the Axiom of Comprehension for Classes, namely, given
a condition ↵(x), there exists a collection (called a class) which consists
exactly of those elements that satisfy the condition. We denote this class
by
{x : ↵(x)}, (3.1)
which intuitively means ‘the class of those x that obey the condition ↵(x)’.
As is well known, it is this principle that originates the paradoxes in
the intuitive set theory, such as Russell’s. Just take as the condition the
assertion that a set does not belong to itself (that is, ↵(x) $ x < x); then, if
we take the formed collection to be a set, there exists the set of all sets that
do not belong to themselves, which belongs to itself if and only if it does
not belong to itself.7 So, it is better to distinguish between sets and classes
from now on; sets will always be relative to a set theory, so, Russell’s just
mentioned class {x : x < x} is not a set of ZF(C) (supposed consistent). The
same happens with some ‘large’ classes, such as the class of all groups, of
all vector spaces, of all models of a scientific theory, which are also not
sets in ZF(C). These collections are very large to be sets of these theories
(but they can ‘exist’ for instance in some stronger set theories, as we shall
see later). So, what is a set? It depends on the axioms we use. Below
we shall see the postulates that determine which are the sets of ZF(C).8
Thus, to summarize: we should be careful in saying ZF-sets, ZFC-sets
and so on. In principle, nothing prohibits Russell’s class from being a
7
For more details, see the Introduction of [Men.97].
8
Please, be careful here: a group, a scientific theory like classical particle mechanics, are described
mathematically by structures which are sets. But the collections of all groups, of all particle mechanics are
not sets of ZF(C), supposed consistent.
68 CHAPTER 3. A MATHEMATICAL BACKGROUND

set of some particular set theory; for instance, it is a PZF-set, where PZF
stands for paraconsistent Zermelo-Fraenkel set theory (see da Costa et al.
[CosKraBue.06]). To summarize: given a set theory T , we may have T -
sets and other collections. The postulates of T provide the grounds for the
characterization (implicit definition) of the sets of T .

3.2.1 The postulates of ZF(C)

Let ↵(x) be a formula of L2 where x is free. As we have seen, the collection

of the objects satisfying ↵(x) is written {x : ↵(x)} and is called a class. The
question is: what classes are sets? In other words: what conditions ↵(x)
determine a set? The answer depends on the axioms we use, as we have
already commented. In ZF(C) we shall have some collections that will
deserve to be named sets, while others do not. Thus, the postulates below
determine what are the sets of ZF(C)9 . There are different set theories (in
fact, potentially infinitely many). One of the most well known is the system
NBG, the von Neumann, Bernays and Gödel set theory. In this theory, the
basic objects are classes (see [Men.97, chap.4]). But some of them are
sets, namely, those classes that belong to other classes. Those classes that
do not belong to other classes were termed proper classes by Gödel. The
sets of NBG and the sets of ZF(C) are essentially the same objects (NBG is
a conservative extension of ZF(C)). As a main difference, however, notice
that in ZF(C) we do not officially quantify over classes, while in NBG we
do.
The logical postulates of ZF(C) are those of the first-order logic with
identity, that is, being ↵, β, and γ formulas, the following schema provide
the postulates:10
9
However, notice that the axioms themselves do not contain the word ‘set’. So, in this sense, the axioms
only put some constraints in possible classes of objects satisfying them, with no warrant that every one of
such classes will be close enough to our ‘intended’ model.
10
The meaning of the word ‘schema’ here is the following one: for instance in the first one, we get a
postulate (synonymous of ‘axiom’) by exchanging ↵ and β by formulas of L2 , and in every case the same
Greek letters must be exchanged by the same formulas.
3.2. THE ZF(C) SET THEORIES 69

(1) ↵ ! (β ! ↵)

(2) (↵ ! (β ! γ) ! ((↵ ! β) ! (↵ ! γ))

(3) (¬↵ ! ¬β) ! ((¬↵ ! β) ! ↵))

(4) ↵, ↵ ! β/β (Modus Ponens)

(5) 8x↵(x) ! ↵(t), where t is a term free for x in ↵(x)

(6) 8x(↵ ! β(x)) ! (↵ ! 8xβ(x)), if x does not appear free in ↵

(7) ↵/8x↵ (Generalization)

(8) 8x(x = x)

(9) u = v ! (↵(u) ! ↵(v)), where u and v are distinct terms of L2 .

These are the axioms of the underlying logic, which is the classical
first-order predicate logic with identity. The specific postulates are the
following ones:

(ZF1) [Extensionality] 8x8y(8z(z 2 x $ z 2 y) ! x = y)

This is just stating that sets with the same elements are identical, the
same set. The converse of the postulate (obtained by reverting the arrow)
is a consequence of the logic axioms, in special of (9).

(ZF2) [Pair] 8x8y9z(8w(w 2 z $ w = x _ w = y))

In the metalanguage, this set will be written z = {x, y}. Obviously, the in-
troduction of such names (new symbols, as mentioned above) is allowed
because we can, in each case, prove existence (by the pair axiom ZF2,
in this case), and unicity (by the extensionality axiom ZF1). The same
remarks hold for the new symbols defined in what follows. A particular
case of ZF2 obtains when x = y. In this case, we write z = {x} and call
it the unitary of x, which is unique by ZF1.
70 CHAPTER 3. A MATHEMATICAL BACKGROUND

(ZF3) [Separation Axioms] If ↵(y) is a formula of L2 with y and z1 , . . . zn

as a free variables in which x does not appear, then for each ↵, the fol-
lowing is an axiom: 8z1 . . . 8zn 8w9x8y(y 2 x $ y 2 w ^ ↵(y)). In
particular, when ↵ has only y as its free variable, then 8w9x8y(y 2 x $
y 2 w ^ ↵(y)) is an axiom.

Taking ↵(x) as x , x, and applying the separation schema on a set w

whatever, we get a set with no elements, which using the axiom ZF1 we
can show is unique. This set is called ‘empty set’ and denoted by ;. The set
x of postulate ZF3 is written (in the metalanguage) as x = {y 2 w : ↵(y)}.
This poses a fundamental difference between this set and a class as given by
(3.1). Here, the elements that belong to x are ‘separated’ from an already
given set w, while in (3.1) they are not coming from a previous set. Thus,
axiom (ZF3) is called the postulate of the limitation of size (of sets), and it
is due to Zermelo.

(ZF4) [Union] 8x9w8z(z 2 w $ 9y(z 2 y ^ y 2 x)). The set w is written

x. In particular, we write u [ v for {u, v}, that is, when x has just two
S S
elements, u and v.

Definition 3.2.3 (Subsets) For any sets u and v, we pose u ✓ v :=

8w(w 2 u ! w 2 v) and say that u is a subset of v. Furthermore,
we write u ⇢ v to mean u ✓ v ^ u , v. In this case, we say that u is a
proper subset of v.

(ZF5) [Power set] 8x9y8z(z 2 y $ z ✓ x). The set y is written P(x), the
power set of x.
Let ↵(x, y) be a formula. We say that it is x-functional if for any x
there exists just one y such that ↵(x, y) holds.11 We say this by writing
8x9!y↵(x, y). Then we have:
11
Intuitively speaking, ↵(x, y) is a formula that defines a function or mapping in the variable x.
3.2. THE ZF(C) SET THEORIES 71

(ZF6) [Substitution Axioms]

8x9!y↵(x, y) ! 8u9v8y(y 2 v $ 9x(x 2 u ^ ↵(x, y)).

Intuitively speaking, the axiom (in reality, an axiom schema, for it gives
us an axiom for each functional formula ↵(x, y) we consider) says that
the image of a set by a function (as defined by a functional formula) is
also a set.

Definition 3.2.4 (Sucessor) The sucessor of a set x is termed x+ and

defined as follows:
x+ := x [ {x}.

(ZF7) [Infinity]
9x(; 2 x ^ 8y(y 2 x ! y+ 2 x).

By paying attention to the fact that ;+ = ; [ {;} = {;}, {;}+ = {;, {;}},
etc., we realize that the set postulated to exist by the infinity axiom con-
tains the empty set, its sucessor and the sucessor of all its elements. Such
a set is called inductive. The intersection of all inductive sets is defined
to be the set of natural numbers.

Definition 3.2.5 (The set of natural numbers) The set of natural num-
bers is the least inductive sets, denoted by !.

Such a set, the least inductive set, is formed by taking the intersection of
all inductive sets; in informal mathematics, it is denoted by ‘N’. Now,
following von Neumann, we put

Definition 3.2.6 (The natural numbers) The (von Neumann) natural num-
bers are defined as follows:
0 := ;
72 CHAPTER 3. A MATHEMATICAL BACKGROUND

1 := 0+ = {;} = {0}
2 := 1+ = {0, 1}
3 := 2+ = {0, 1, 2}
...
n := {0, 1, . . . , n − 1}
...

Thus, due to ZF7, we can write

! := {0, 1, 2, 3, . . . }.

(ZF8) [Regularity, or Foundation]

8x(x , ; ! (9y 2 x)(8z 2 x)(z < y)).

This axiom says that any non-empty set x has an element whose inter-
section with x is the empty set. That is, such an element has as elements
no element of x. Thus, in particular x cannot have itself as an element.
Let us prove this result with some care:

Theorem 3.2.1 Due to the axiom of regularity, there exists no set having
itself as an element.

Proof: Let us suppose that x 2 x. Then x , ;. Since x 2 {x}, we have

x 2 x \ {x} (let us call this result (?)). By (ZF8), there exists y 2 {x} such
that y \ {x} = ;. But, since {x} is unitary, the only element to be chosen
is x itself. Hence we must have x \ {x} = ;, which is contradictory with
(?). So, our hypothesis that x 2 x must be rejected.

Important to say that the axiom of regularity is essential in this proof. If

the theory does not assume it — it can be proven that ZF8 is independent
3.2. THE ZF(C) SET THEORIES 73

of the remaining axioms of ZF(C) —,12 then this possibility remains

open. In the same vein, without regularity there may exist extraordinary
sets (a term coined by Mirimanov in 1919), that is, sets involving chains
of the following kind: x 2 xn 2 . . . 2 x2 2 x1 2 x.
By adding the next axiom, we get the theory ZFC,

(ZF9) [Axiom of Choice]

8x((8y 2 x) ! y , ;) ^ (8t 2 x)(8z 2 x)(t , z

! t \ z = ;) ! ((9t ✓ x)(8u 2 x) ! 9v(t \ u = {v}))).

The intuitive explanation of this axiom is: given a non-empty set whose
elements are also non-empty sets, then there exists a set (the choice set)
formed by just one element of each of the elements of the given set. There
are many formulations equivalent to this one.
We could formulate these axioms without employing any of the defined
symbols such as ✓, [ and others. But we have used them here to keep the
text more easily readable. The axiom of choice is considered as the second
most famous axiom of all mathematics, losing only to Euclid’s parallel
postulate. There are excellent books on its history and importance (see for
instance [Moo.82], [FraBarLev.73, ch.2]).

3.2.2 A matter of terminology

We can consider several different theories from the above axiomatics. Let
us fix some terminology:

1. ZF is the above theory without the axiom of choice

2. ZFC is the theory encompassing all axioms above.

12
A sentence S is independent of a theory T if the postulates of T do not prove neither S nor the negation
of S . Remember that this what happens with the famous parallel axiom in Euclidean geometry. The same
will happen with the axiom of choice in ZF, as we have mentioned already.
74 CHAPTER 3. A MATHEMATICAL BACKGROUND

3. ZF− is ZF minus the axiom of regularity, and the same for ZFC− .
4. Z p is the theory got by dropping the axioms of regularity and re-
placement. This is essentially the ‘pure’ Zermelo’s set theory (without
ur-elements – or Urelemente).
5. ZF⇤ is the theory got from the above axiomatics (without choice)
by adding the Axiom of Inaccessible Cardinals to be explained below;
similarly, we get ZFC⇤ .
We remark that all these theories have the same language and underly-
ing logic. Their differences reside in the specific axioms.

3.2.3 Inaccessible cardinals

There are some collections whose existence cannot be proved in ZF(C),
supposing this theory consistent.13 Among them, we mention just the state-
ment to the effect that there exists a strong inaccessible cardinal. We don’t
need to consider those statements here (but see the footnote below); we
mention them just to acknowledge their importance, for by adding the men-
tioned axiom, we get a theory ZF(C)⇤ where we can prove the consistency
of ZF(C).
A cardinal λ is strongly inaccessible if it satisfies the following condi-
tions: (i) for every cardinal β, if β < λ, then 2β < λ; (ii) if {βi }i2I is a
family of cardinals such that card(I) < λ and βi < λ for every i 2 I, then
sup{βi : i 2 I} < λ. It is enough to acknowledge that formulas express-
ing these facts are available and can be added to ZF(C) in order to get a
stronger theory in which a model of ZF(C) can be built; we shall return
to this problem soon. Other postulates could be introduced with the same
finality, such as those postulating the existence of a universe, but we shall
not consider them here (but see the next chapters and [BrigCos.71]).
13
Sometimes logicians bore the reader with details. But it is necessary to take into account that when
we say that something (some set) ‘exists’ in ZF(C), we are assuming that this theory is consistent. If it is
not, since its logic is classical logic, we can prove that any set (collection, class) belongs to it, something of
course we would not like to admit. Hence the remark.
3.2. THE ZF(C) SET THEORIES 75

3.2.4 Informal semantics of L2

From the formal point of view, the symbols of L2 have no meaning. But,
intentionally, we usually accept that they make reference to the objects of a
certain intuitive non-empty domain of objects we conceive as representing
sets, but in principle they could be of any ‘nature’ we wish. Let us be a
little bit more precise.

Definition 3.2.7 A structure for interpreting L2 is an ordered pair A =

hD, ⇠i, where D is a non-empty set and ⇠ is a binary relation on D.

Due to Gödel’s second incompleteness theorem, the consistence of ZF(C)

cannot be proved with its own resources. We need a stronger theory where
we can build a model of ZF(C), such as ZF(C)⇤ mentioned above. Thus,
the structure A of the previous definition cannot be a set of ZF(C). The
construction must be done, say, in ZF(C)⇤ or in informal metamathematics
(informal set theory).
The intended interpretation ascribes the diagonal of D, namely, the set
∆D = {ha, ai : a 2 D} to the identity symbol of L2 and the binary relation
⇠ to the membership relation. The expressions 8x↵(x) and 9x↵(x), where
↵(x) is a formula of L2 in which x is the only free variable, mean ‘for
all elements of D’ and ‘there exists an element of D’ such that ↵(x) holds
respectively.
One of the outstanding problems in the study of models of set theory
is to find an interpretation that makes the postulates of ZF ‘true’ (in the
intuitive sense). To begin with, let us reproduce here some examples of
possible interpretations for L2 . Later we shall mention some more suitable
ones. The first interpretation takes D as the set Z of the integers, and ⇠ as
the usual ordering relation < on the integers.Thus, the integers are now our
‘sets’, and x 2 y means x < y. Of course this is an interpretation as we
have described informally; to the predicate = of identity we associate the
set of all couples hx, xi with x 2 Z, and the connectives and quantifiers are
interpreted as usual (9x means ‘there exists an integer x so that . . .’ and so
76 CHAPTER 3. A MATHEMATICAL BACKGROUND

on). It is easy to see that this structure models the axiom of extensionality;
just change < for 2 in the axiom to see that its translation to the language
of the integers becomes a theorem of this theory. In the same way, we can
show that the pair axiom also holds in this structure. But the theorem that
asserts that there exists an empty set does not hold. In fact, let us consider
the sentence (a formula without free variables)
8x0 9x1 (x1 2 x0 ). (3.2)
The reader should recognize that (3.2) stands for a formula of L2 which
expresses the mentioned theorem. In the metalanguage, we could write
8x9y(y 2 x)). According to our interpretation, it means that for any integer
there exists an integer which is less than it. It is easy to see that the trans-
lation of this sentence to the language of our interpretation (there exists an
integer which is less than any integer) is false. Thus, the given structure is
not a model for ZF(C), for it does not model one of its postulates.
Now let us take another interpretation, just taking D = N (the set of
the intuitive natural numbers) and ⇠ being the order relation < on such a
domain. This is also an interpretation. Now not only the pair axiom and the
extensionality axiom are true according to this interpretation, but also the
theorem asserting that there exists an empty set. But the new structure does
not model all axioms of ZF(C), which can be checked by the reader. In fact,
these interpretations are far from suitable for modeling all the axioms of
ZF(C).

3.3 ‘Models’ of ZF(C), again

A model of a formal theory, as we already know, is an interpretation that
makes its postulates true. Can we think of models of ZF(C) in this sense?
As we have mentioned before, we can, but the corresponding structures
will be ‘sets’ which are not ZF(C)-sets.
The consistency of ZF(C) can be proven only relative to another stronger
theory, whose consistency is then put also into question, and to answer
3.3. ‘MODELS’ OF ZF(C), AGAIN 77

whether this stronger theory is by its turn consistent, we will need a still
stronger theory, and so on. An absolute proof of consistency requires that
we prove that the theory does not entail two contradictory formulas. The
two notions of consistency involved here, namely, relative and absolute are
also called syntactical (when absolute) and another ‘semantic’ (when rela-
tive). More precisely, the syntactic (absolute) definition says that a theory
T , whose language contains a negation symbol ¬, is consistent iff there is
no formula ↵ such that both ↵ and ¬↵ are theorems of T . The semantic
(relative) definition says that a theory is consistent if and only if it has a
model (an interpretation that satisfies the axioms of T ). Semantic consis-
tency implies syntactic consistency and Henkin’s Completeness Theorem
(which holds in a restricted form also for higher-order languages) estab-
lishes that syntactic consistency is enough for a theory to have a model.
The reason we cannot prove the relative consistency of ZF(C) within
the theory itself was mentioned above, and is due to Gödel’s second in-
completeness theorem. But, being consistent, ZF(C) has ‘models’. The
model which interests us here is called the well-founded cumulative hier-
archy of sets, or the von Neumann universe, termed V. The construction of
V is made on transfinite recursion on the class On of the ordinals (which
is also not a ZF(C)-set),
On = {0, 1, 2, . . . , !, ! + 1, . . . , !2, !2 + 1, . . . , !2 , . . .}. (3.3)
The construction goes by defining a hierarchy V↵ (↵ 2 On) as follows:
V0 = ;
V1 = P(V0 )
..
.
Vn+1 = P(Vn )
Vλ = β<λ P(Vβ ), for λ a limit ordinal,14
S

14
An ordinal ↵ is a limit ordinal if there is no ordinal β such that ↵ = β + 1. Typical limit ordinals are !,
!2, !2 , etc.
78 CHAPTER 3. A MATHEMATICAL BACKGROUND

..
.

↵2On V↵ .
S
V=

In terms of structures, we may write WF = hV, 2i, the ‘WF’ stand-

ing for ‘well-founded’, that is, the sets in the universe obey the axiom of
foundation.
Now it is a mathematical exercise to show that the axioms of ZF(C) are
‘true’ (in the intuitive sense) in this structure, which is represented in the
Figure (3.1).
The remarks that follow
will make reference to both On
A "
universes. First of all, it A "
A "
should be noticed that any A "
A "
V↵ is a transitive class. Now A "
let us consider the follow- A "
A " V = hV, 2i
ing question: given a certain A "
A "
ordinal ↵, which axioms of A "
A "
ZFC are satisfied in hV↵ , 2i, A "
A "
where 2 the membership re- A "
lation relativized to V↵ (that A "
A"
is, x 2 y means x 2 y and ;
x, y 2 V↵ )? Interesting is Figure 3.1: The well-founded ‘model’ of pure ZF(C)
that we can prove the fol-
lowing results:15

(i) Any V↵ is a model of the following axioms: extensionality, separa-

tion, union, power set, choice, regularity.
15
Let M be a class (it may be a set) and F a formula of L2 . The formula F M is called the relativization
of F to M if it is obtained by substituting 9x 2 M for 9x, and 8x 2 M for 8x. Thus, it says exactly what
F says but concerning elements of M only. The results mentioned below are proven in [Fra.82, pp.298ff],
[End.77, pp.249ff]; [Fra.82, p.289].
3.3. ‘MODELS’ OF ZF(C), AGAIN 79

(ii) For the pair axiom, we need ↵ as a limit ordinal. This can intuitively
be understood for the set of two objects is of a higher level that the levels
of the objects.

(iii) For the infinity axiom, we need that ↵ be a limite ordinal greater
that !, for instance, !2.

(iv) For the replacement axioms, we need more, for instance, an inac-
cessible cardinal.

(v) The whole V is a model of ZF(C).

Remember that we call Z p the Zermelo set theory, which is ZF without

the substitution (replacement) axioms. All the levels V↵ can be constructed
in Z p as well. But, by the above results, we see that V!2 is a ‘model’ of
Z2 , since !2 is a limit ordinal. Then, if consistent, Z2 cannot admit V!2
as one of its sets, for in this case we would be against Gödel’s second in-
completeness theorem. Reasoning in the same vein, we cannot prove the
existence of inaccessible cardinals within ZF(C). Thus, we can’t prove the
existence of V, the whole universe of sets, within ZF(C). Of course we
could think that the universe can be constructed within a stronger theory,
say ZF ⇤ which is ZF plus an axiom that says that there exists an inacces-
sible cardinal (according to the terminology introduced above). This is
true, but we would just transfer the problem to this stronger theory, for we
also don’t know what its notion of set means, and a model for it would
be needed. This problem brings interesting philosophical questions, due
to the fact that, if consistent, a set theory such as ZF(C) (formulated as a
first-order theory) is not categorical.16
We have made such a digression on set theory and on its ‘models’ just
to show to the reader that there are differences in speaking of certain struc-
tures as models of certain theories and ‘models’ of the set theories them-
selves. The models of both mathematical and scientific theories, such as
16
A theory is categorical if all of its models are isomorphic.
80 CHAPTER 3. A MATHEMATICAL BACKGROUND

groups, vectors spaces, Euclidean geometry, metric spaces, differentiable

manifolds, classical mechanics, relativity theory, etc. can be constructed
within Z p , ZF or ZFC, depending on the particular theory. But the ‘mod-
els’ of these theories themselves cannot be constructed within themselves
(supposing them consistent). Below we shall discuss a little bit a particular
case involving quantum mechanics.

3.4 Urelemente

Abraham Fraenkel has shown that the ur-elements are not necessary for the
foundations of mathematics, and his ideas (so as Skolem’s) have led to a
‘pure’ set theory, whose stages begin from the empty set seen above. But,
for applications in the empirical theories, it seems that we need entities that
are not sets, the ur-elements, or simply atoms. Here we are supposing, of
course, as it seems reasonable, that physical objects like chairs, molecules
of a gas, elementary particles, and so on, are not sets, so they can be rea-
sonably represented by atoms.
A typical set theory with
atoms is termed ZFU (Zermelo-
On
Fraenkel with Urelemente), A A "
A A "
perhaps encompassing the A A "
A A "
axiom of choice. Here, we A
(sets with pure sets
Urelemente) A (no Urelemente) "
A A "
will not write ZFC(U). The A A "
universe of ZFU (drawn A A "
A A "
at Figure (3.4) can also A A "
A A "
be constructed by transfi- A U A "
A A";
nite recursion on the ordi-
nals as shown below, by as- Figure 3.2: The universe of sets with Urelemente. On
suming the existence of a is the class of ordinals, and U is the set of atoms (Ure-
lemente).
non-empty set U of Urele-
mente, as follows:
3.4. URELEMENTE 81

V0 = U
V1 = V0 [ P(V0 )
..
.
Vn+1 = Vn [ P(Vn )
Vλ = β<λ P(Vβ ), for λ a limit ordinal,
S

..
.

↵2On V↵ .
S
V=
The axioms of ZFU are similar to those above, but must be modified
accordingly in order to admit the existence of distinct Urelemente. Let us
present these axioms rather briefly. The language of ZFU, which we call
LU has, in addition to the primitive vocabulary of L2 , an additional unary
predicate symbol U so that U(x) says that x is an Urelement. So, we have
Definition 3.4.1 C (x) := ¬U(x)
Intuitively speaking, C (x) says that x is a set. We use restricted quan-
tifiers in the following sense: 8C x↵(x) means 8x(C (x) ! ↵(x)), and
9C x↵(x) means 9x(C (x) ^ ↵(x)). The same can be done for other de-
fined predicates (see axiom ZFU4 below). The logical postulates are those
of the first-order classical logic with identity, and the specific postulates
may be stated as follows:
(ZFU1)[Extensionality] 8C x8C y(8z(z 2 x $ z 2 y) ! x = y)
Two sets are equal if they have the same elements (either sets or Urele-
mente).
In set theory, Urelemente are synonymous of atoms (objects that are
not sets but which may be elements of sets). Thus, sometimes the words
‘atom’ and ‘atoms’ apear there in this sense. This axiom enables the theory
to be compatible with the existence of different atoms, other objects than
the empty set (which is a set). Below we shall speak a little bit about atoms.
82 CHAPTER 3. A MATHEMATICAL BACKGROUND

(ZFU2)[Empty set] 9C x8y(y < x)

The existence of an empty set, termed ; and proved that it is unique by
extensionality.
(ZFU3)[Pair] 8x8y9C z8w(w 2 z $ w = x _ w = y)
As it is usual, this set is written z = {x, y}.
Definition 3.4.2 E(x) := C (x) ^ 8y(y 2 x ! C (y))
Which says that x is a set whose elements are also sets.
(ZFU4)[Union] 8E x9C y8z(z 2 y $ 9w(w 2 z ^ z 2 x))

In the next two axioms, the definitions are as in the case of ZF(C).
(ZFU5)[Power set] 8C x9C y8z(z 2 y $ C (z) ^ z ✓ x)
(ZFU6)[Infinity] 9C x(; 2 x ^ 8C y(y 2 x ! y+ 2 x))

(ZFC7)[Foundation/Regularity] 8C x(E(x) ! 9y(y 2 x ^ y \ x = ;)

Given that the elements of a set x are also sets, then x has an element
that has no common element with x.
(ZFU8)[Replacement] Let ↵(x, y) be a formula where x and y are free
and z is not free. Then,
8x9!y↵(x, y) ! 8x9C y8z(z 2 y $ 9w(w 2 x ^ ↵(w, z))).

That is, if ↵(x, y) ascribe just one y for each x, then there exists the so-
called ↵-image of any element of the domain.
(ZFU9)[Choice]
8C x(E(x) ^ (8C y 2 x)(y < ;) ^ (8z 2 x)(8w 2 x)(z \ w = ;) !
9C t(8z 2 t)(9s(z \ t = {s})))
3.5. MORE ON THE ‘MODELS’ OF ZF(C) 83

3.4.1 Some comments

We could add to this axiomatic some further axioms, such as one of the
following ones:

(ZFUa)[Non existence of atoms] 8xC (x)

(ZFCb)[Atoms are void] 8x8y(y 2 x ! C (x))

The first one makes the theory essentially ZFC. The second one (which
is implied by the first) says that atoms have no elements. This is interesting,
for since we think of atoms as, say, physical objects, perhaps it would be
better to think of them as composed of parts in a mereological sense.17
Thus, this axiomatics should be increased by a relation symbol for “part-
whole” and the corresponding axioms for a mereology. But this is not our
topic here.
An important result is that ZFU is equiconsistent with ZF(C), that is,
one of then is consistent iff the another one is consistent too.

3.5 More on the ‘models’ of ZF(C)

We have remarked above that ZF (Z p , ZFC, ZFU), formulated as as first-
order theory, is not categorical. Let us comment a little bit on this claim in
order to see its consequences, even to physical theories.
We shall mention just a case, among several possible others which could
be obtained by using techniques such as Cohen’s forcing, but which are out
of the scope of this text. The importance of the ‘models’ of set theory for
the philosophy of science may be explained as follows. A formalized the-
ory (even of the empirical sciences), say by a Suppes predicate is generally
elaborated to cope with a certain informal given theory, as we have seen
17
Mereology, or the logic of parts and wholes, was developed initially by Stanislaw Lesniewicz, and has
been studied from different perspectives ever since; see Simons [Sim.87].
84 CHAPTER 3. A MATHEMATICAL BACKGROUND

before. But in order to achieve this aim, we need to provide an interpre-

tation to our formalism, generally by associating to it a structure built in
set theory or, what it is the same, in a ‘model’ of the set theory we use as
our metatheory. But a set theory such as ZF (ZF(C), ZFU) has different
models. If axiomatized as a first-order theory, it has even a denumerable
‘model’ due to the Löwenheim-Skolem theorem, and since in general we
never know what model we are working in, we really will have difficul-
ties to state that the interpretation we provide fits the intended concepts.
We remark once more that despite the fact that we can think in sets and
related concepts as if we are inside the ‘standard model’ of ZF(C), we can-
not grant that this is really so, for we have no grounds for proving that
such a ‘model’ exists. Let us consider a particular example concerning
denumerable ‘models’ only.

Real numbers can be used in physics for representing time, probabil-

ity measures, eigenvalues of self-adjoint operators, to parametrize certain
functions, and so on. The physicist with no logical background has an in-
tuitive idea of what real numbers are, so as what are sets. If pressed, a
physicist may mention even some construction of the set of real numbers
from the rationals, say by convergent sequences (Cauchy sequences) or
by Dedekind cuts (see [End.77, pp.111ff], [Fra.82, pp.87ff]). These con-
structions are done within ZF(C), say. But suppose ZF(C) is consistent.
Then it has ‘models’. The first we can imagine may be called the standard
model, which is more or less identified with an ‘universe of sets’ such as
the universe V seen before. But, as we have seen, we cannot prove in
ZF(C) that such a ‘model’ exists, so we don’t have any evidence that our
intuitive account of sets, that is, our intuitive idea of a set, is captured by
the elements of such a universe. But there is more. As a first-order theory,
ZF(C) is subjected to the Löwenheim-Skolem theorem, which implies two
things: firstly, since any model of ZF(C) must satisfy the axiom of infinite,
ZF(C) will have an infinite ‘model’ (recall that ‘model’ is being written
in quotation marks to emphasize that is is not a set of ZF(C) but a proper
3.5. MORE ON THE ‘MODELS’ OF ZF(C) 85

class). Then, by the upward Löwenheim-Skolem theorem, ZF(C) has non

isomorphic models of any infinite cardinality. But let us fix in the other
result entailed by this theorem: by the downward Löwenheim-Skolem the-
orem, having models, ZF(C) has a denumerable model.18 What does that
mean for empirical science?
Let us suppose that our theory T demands the real number system, as
most physical theories do. Then, it seems ‘natural’ to suppose that the set
R of the real numbers is not denumerable, as Cantor showed in the XIX
century with his famous diagonal argument. But, in such a denumerable
‘model’, the set that (in the ‘model’) corresponds to the set of real numbers
must be denumerable. This is of course puzzling, but there is not a contra-
diction here. As remarked by Skolem, this result, known as the ‘Skolem
paradox’ is not a formal ‘paradox’, but just a result that goes against our
intuition, showing that within ZF(C) (that is, a ZF(C)-set) there is no bi-
jection between R and !; but this does not entail that such a bijection does
not exists outside ZF(C), that is, as something which is not a ZF(C)-set.
Thus, if a physical theory demands a mathematical concept which, if stan-
dardly defined, presupposes the set of real numbers, how can we grant that
it is non-denumerable? In other words, how can we ensure that, working
within ZF(C) we are making reference to some ‘model’ that fits our intu-
itive claims? Unfortunately, this is not possible, and the best we can do is
to fix a particular ‘model’ of ZF(C) when we need to give an interpretation
of concepts, but we will be always subject to questionings. Indeed, as re-
marked, we can’t show that even the so-called ‘standard’ (intended) model
of ZF(C) does exist!
On what concerns scientific theories, as remarked by da Costa, “we
should never forget that set theories, supposed consistent, have non-standard
‘models’, thus any theory founded on them will have non-standard models

18
Roughly speaking, the downward version of the theorem says that if a consistent theory T in a countable
language has a model, then it has a denumerable model. The upward theorem says that such a theory, having
infinite models, has models of any infinite cardinality.
86 CHAPTER 3. A MATHEMATICAL BACKGROUND

too.” 19 This entails that if we try to provide an understanding for the con-
cepts we use (that is, by given them an interpretation), we need to consider
‘models’ of ZF(C) and this poses us a challenge: we never know in which
model we are working in, so we really never know what our concepts re-
ally mean. All we can do is to suppose we are working in the standard
‘model’, where (apparently) sets are just as we imagine, that the real num-
bers cannot be enumerated, that there are basis for our vector spaces, and
so on. The investigation of non-standard models for physical theories is
still a novel domain of research, and it is most likely to be a rich one.

19
Personal communication.
Chapter 4

Criticism to the axiomatic method and its

defense

here seems to be no doubt that the axiomatic method (AM) has

T achieved a tremendous success in mathematics. Nowadays it

would be difficult to find a field of pure mathematics not pre-
sented by employing this method (generally, as an abstract
axiomatics). Real numbers, natural numbers, topological spaces, vector
(linear) spaces, groups, rings, fields, etc. are presented axiomatically.1 For
instance, many mathematicians think that for some finalities it is easier to
present, say, the real numbers as a complete ordered field than to construct
the real numbers system as Dedekind cuts or as equivalence classes of
Cauchy sequences (among other possibilities). Of course this depends on
the aims and purposes of the mathematician. But it seems clear that the ax-
iomatization of a certain field enables us to determine from a certain point
of view those concepts and assumptions which could be taken as basic in
the field, acknowledging also the resources available to the deductive appa-
ratus employed (we mean the underlying logic — mathematicians usually
1
Textbooks in Differential and Integral Calculus, Analytic Geometry, and many others, are generally
presented in an informal language, without axioms but only with definitions and theorems. But it is easy
to note that these fields make part of a more detailed mathematical theories presented in an axiomatic way.
Furthermore, it is known from logic that every system given by postulates can be transformed in another
system described by nominal definitions only. The mentioned areas can be seen as described this way.
86 CHAPTER 4. CRITICISM TO THE AXIOMATIC METHOD AND ITS DEFENSE

assume something along the lines of classical logic in their deductions, so

the use of classical reduction to absurd, for instance, is available).
When a field of mathematics is presented axiomatically, it is in general
as an abstract axiomatics; the field of real numbers, for instance, when
presented as a complete ordered field, is nothing more than a particular
structure of a certain general kind (the kind of complete orderer fields), or
species, to use Bourbaki’s terminology [Bou.68]. But we can easily realize
that no standard mathematical book of Analysis, Algebra, etc. speaks of the
underlying logic.2 But despite the achieved success, the method was not
taken for granted by all mathematicians and philosophers, not to speak of
physicists. Several criticisms were advanced against axiomatization. In
this chapter, we shall see some of them and in the final part we present a
defense of the method.

4.1 Does the AM encompass heuristics?

One of the most interesting criticisms to the AM was advanced by Imre
Lakatos in his Proofs and Refutations [Lak.76]. It should be remembered
that in this book, as well as in other places, Lakatos tries to bring to math-
ematics the Popperian scheme of knowledge growing: namely, by ‘proofs
and refutations’. Lakatos holds that informal (inhaltliche) mathematics
proceeds exactly as Popper proposed on what regards the empirical disci-
plines. According to Popper, the scientist starts with a problem (P1 ), pro-
vides a tentative theory (T T ), which attempts to give us a tentative solution.
Then, tests are made and if something goes wrong she tries to eliminate the
errors (EE), which leads her to another problem, P2 .3 Then the scheme is
back to its starting point. The same happens in informal mathematics, says
Lakatos. The mathematician starts with a problem, say Euler’s conjecture
that the relation between the number of vertices V, the number of edges E,
2
Although that would be useful. In fact, how can a mathematician speak of proofs — and usually make
several proofs in her training and teaching — without knowing what a proof is?
3
Thus his celebrated formula P1 ! T T ! EE ! P2 [Pop.72, p.119/243].
4.1. DOES THE AM ENCOMPASS HEURISTICS? 87

and the number of faces F of polyhedra, in particular the regular ones, is

given by Euler’s formula V − E + F = 2. There is also a ‘proof’ of this
conjecture. The ‘proof’ then faces counter-examples, showing that the for-
mula does not apply to any polyhedra, say for that one formed by a cube
with a hole whose intersection with the two opposite faces is a square (in
this case, it was later noticed that V − E + F = 0).4 The scheme goes as
follows: given some evidence, the scientist makes a conjecture by means
of a hypothesis (inside a certain scientific theory). The cases that agree
with the conjecture simply corroborate it. But the conjecture may admit
counter examples. In this case, some change is required in the conjecture
(and in the theory). To Lakatos, the same happens in informal mathemat-
ics, as his example shows us. So, he says, (informal) mathematics proceeds
by ‘proofs’ and refutations.
Axiomatic theories, on the other hand, says Lakatos, do not allow such
revisions. What we derive from the axioms are the theorems of the the-
ory, and they do not admit counter examples (in the theory, supposed con-
sistent). According to him, axiomatized mathematical theories, which he
identifies with the formalist school headed by Hilbert (which as we see
is not completely right, for we could make use of either concrete or ab-
stract axiomatics without recurring to strict formalization), are tautologi-
cal, just providing results (the theorems) which in a certain sense are al-
ready implicit in the axioms. A proof, in axiomatized mathematics, is noth-
ing more than an unbreakable chain of mechanical procedures which go
from premises to conclusions. Informal mathematics, on the contrary, as
in Popper’s account to natural sciences, proceeds by presenting criticisms
to the conjectures, explanations, discussion, which make the conjecture
more plausible and convincing due to the existence of counter examples.
So, axiomatization does not involve heuristics, creation of new hypotheses
and explanations other than those already implicit in the axioms.

4
Today we know that the constant k in V − E + F = k is a topological invariant of the polyhedra of a
kind. For instance, those where k = 0 are topological equivalent to a torus, and to a sphere if k = 2.
88 CHAPTER 4. CRITICISM TO THE AXIOMATIC METHOD AND ITS DEFENSE

4.1.1 The axiomatic method and heuristics

There is a reasonable sense according to which we can say that the AM

involves heuristics. Following Suppes, we can distinguish between two
kinds of axiomatics: heuristic and unheuristic. By heuristic axioms, says
Suppes, “I mean that the analysis yields axioms that seem intuitively to
organize and facilitate our thinking about the subject, and in particular to
formulate, in an ordinarily self-contained way, problems concerned with
the phenomena governed by the theory and their solutions.” [Sups.83].
That is, the heuristic axiomatics provides us a clear understanding of the
field, this being understood as linked to our intuitions concerning the field
under analysis. Peano’s original axiomatics and Zermelo’s set theory are
typical examples, for they are in agreement with our intuitions. But Sup-
pes also speaks of unheuristic axiomatics. He does not say that these types
of axiomatization are devoid of value, but simply they are artificial, not
intuitive, not representing “the kind of transparent and conceptually sat-
isfactory solution we should aim at whenever possible” (ibid.). Suppes
doesn’t mention neither Peano nor Zermelo, but the axioms of the field of
real numbers. The construction of the real numbers by Dedekind cuts or
by equivalence classes of Cauchy sequences, he says, are unnatural to deal
with, contrarily to the axioms of a complete ordered field, which he regards
as “very natural and intuitive”. The same happens with Kolmogorov’s ax-
ioms for probabilities. The unheuristic example is Mackey’s axioms for
non-relativistic quantum mechanics. In fact, despite Mackey’s formulation
being very useful, there is no discussion about the motivations and it is
quite artificial, for “the axioms taken literally present a wrong picture of
how to think about physical problems in quantum mechanics.”
Suppes links heuristics with intuitiveness, a vague concept. We think
we have other arguments for defending the heuristic role of the axiomatic
method by pointing, say, that only by axiomatizing arithmetics within a
first-order logic were we able to discover the existence of non-standard
models. The same can be said about first-order real number theory. Thus,
4.2. TRUESDELL AGAINST THE ‘SUPPESIANS’ 89

we have discovered things which were inaccessible to us without the strict

application of AM; that is, without axiomatizing arithmetics, we would
never be aware of non standard natural numbers. The same can be said of
many other fields; without axiomatics, we probably would never have ar-
rived at non-euclidean geometries, to the study of distinct (non-equivalent)
formulations of set theory, versions of set theory where the axiom of choice
does not hold, to non-commutative and non-associative algebras, and so
on. Even logic was made clear only after its formulation in axiomatic
terms, starting with Aristotle. Thus, we regard AM as heuristic, it may
lead to the discovery of many new fields of inquiry, although sometimes
it may be quite artificial. In this sense, heuristic power needs not always
coincide with intuitiveness.

4.2 Truesdell against the ‘Suppesians’

As we shall see with more details in the next chapters, Patrick Suppes and
collaborators have developed a way of axiomatizing a scientific theory by
assuming set theory (some set theory, but since it is not made explicit,
we should take it in terms of his heuristics, that is, the most intuitive one
— he speaks of informal set theory).5 Set theory is a general framework
where all usual concepts needed in mathematics (and in the usual physi-
cal theories) can be developed: functions, vectors, differential equations,
manifolds, geometries.6
Recall that it was traditionally thought that the Received View of sci-
entific theories required that every scientific theory should be axiomatized
in a first-order language. Suppes [Sups.67], in particular, also advanced
5
Even though we should acknowledge that the notion of ‘most intuitive’ is vague and relative.
6
Of course, in present day mathematics not all mathematical concepts can be developed within standard
set theories; category theory is one such example. But such a claim depends on the set theory being con-
sidered; even category theory can be developed within strong set theories, for instance those comprising
universes. By adding to ZFC a postulate saying that a Grothendieck universe exists, we get a new the-
ory so strong that category theory can be developed within it [Low.14]. Universes entail the existence of
inaccessible cardinals.
90 CHAPTER 4. CRITICISM TO THE AXIOMATIC METHOD AND ITS DEFENSE

such a reading (which as we have seen, is misguided; again see our dis-
cussion in chapter 1). As an alternative to such a restrictive framework,
Suppes thought that logic and the basic mathematics should be taken for
granted if we assume a set theory right from the start, as we shall see in
chapters to come. Suppes didn’t mention which set theory he was thinking
about, but this is just what he intended to presuppose: any set theory able
to express the necessary mathematical tools demanded by the theory being
dealt with will do. Without the need of specifying a particular set theory,
naïve set theory can be taken for granted. Hence, by presupposing logic
and the relevant mathematics, he presents a case of what we have termed
above ‘abstract axiomatization’.
As we have already discussed briefly in chapter 1 and we will present
with details in chapters 5 and 6, Suppes’ technique allows us to represent a
theory as a given class of models of a certain set theoretical predicate (here,
recall, models are understood in terms of mathematical structures). One of
the most celebrated examples is classical particle mechanics (CPM), which
McKinsey, Suppes and Sugar have presented in an axiomatized form in
1953 (see [Sups.02] for an updated version, and also the next chapters).
CPM deals with ‘particles’ and the forces acting on them. But, what
are particles? In the spirit of abstract axiomatics, we should not pose this
question. It is a primitive concept. CPM serves to deal with several distinct
models of particle systems, such as the solar system, the molecules of a
certain gas, and in general, with whatever collection of objects that fulfill
the conditions of the theory and make its postulates true. In each of these
models the objects of the domain are the particles. But CPM does not
deal with interaction among particles, with collisions and other phenomena
typical of what in physics is called continuum mechanics. And here lies
the core of one of the most famous attacks on the axiomatic method as
employed by Suppes: Truesdell’s criticism.
Clifford Truesdell was an influent American mathematician who con-
tributed to the foundations of several areas in the physical sciences, in par-
4.2. TRUESDELL AGAINST THE ‘SUPPESIANS’ 91

ticular in continuum rational mechanics. He was also the editor of the

Journal of Rational Mechanics and Analysis, to which McKinsey, Sugar
and Suppes submitted their papers on the axiomatization of classical par-
ticle mechanics [McSS.53], so Truesdell’s opinion on the issue should be
of greatest relevance. One of the papers was published with a note by
Truesdell, stating that

[t]he communicator is in complete disagreement with the view

of classical mechanics expressed in this article. He agrees, how-
ever, that strict axiomatization of general mechanics — not merely
the degenerate and conceptually insignificant special case of par-
ticle mechanics — is urgently required. While he does not be-
lieve the present work achieves any progress whatever toward
the precision of the concept of force, which always has been
and remains still the central conceptual problem, and indeed the
only one not essentially trivial, in the foundations of classical
mechanics, he hopes that publication of this paper may arouse
the interest of students of mechanics and logic alike, thus per-
haps leading eventually to a proper solution of this outstanding
but neglected problem.

Later, in Chap.39 of his book An Idiot’s Fugitive Essay on Science:

Methods, Criticism, Training, Circumstances, titled ‘Suppesian Stews’ [True.84],
he considered once again Suppes et al. methods, providing a long criticism,
saying in particular that

McKinsey, Sugar, & Suppes gave no example of how to use their

paradigm [that of presenting a set theoretical predicate, as it be-
came clear in Suppes’ works, as we shall see later] for any pur-
pose in the practice or development of mechanics, and so far
as I can learn, no-one has done so. While the Suppesians con-
tent themselves with admiring it for its perfection, specialists in
mechanics have given in no heed, perhaps because it leaves out
92 CHAPTER 4. CRITICISM TO THE AXIOMATIC METHOD AND ITS DEFENSE

of account too much of the meat of the the mechanics of New-

ton, Lagrange, and Cauchy — or, for that matter, of the content
of any current textbook of mechanics for engineers. [True.84,
p.539]

It is clear that Truesdell is criticizing the paper by McKinsey et al. for

not covering what the (correctly) considers the main part of mechanics,
namely, continuum mechanics, which deals with collisions and other con-
cepts not covered by particle mechanics. Part of his criticism is taken from
a previous book of Wolfgang Stegmüller [Steg.79], who has distinguished
two versions of the ‘structuralist view’ of theories in the version defended
by W. Balzer and others, which he calls ‘the Carnap approach’, and the
version by Suppes et al., termed Suppes approach. Of course there are
strong differences among these views which do not concern us here. What
is relevant is that Truesdell is complaining about the lack of adequate con-
tent of the proposal of McKinsey et al. on what concerns the ‘real mechan-
ics’.
The question behind this criticism can be posed as follows: should an
axiomatic version of a theory necessarily cover every aspect of the theory?
The answer will lead us to a future section termed ‘Does axiomatics really
axiomatizes?’, so we will not discuss it in full here. We only advance our
opinion that no, an axiomatic theory does not necessarily need to cover
all the theory’s subject due fundamentally to two reasons: i) the scope of
an informal theory (which supposedly is being axiomatized) is not well
defined, and ii) this may not be of immediate interest to those who propose
the axiomatics. The first item is obvious. The second can be enlightened
by McKinsey’s response to Truesdell, who had send him a letter by Georg
Hamel evaluating the papers by the three authors given on the following
grounds:

Thank you [Truesdell] so much for your letter of October 21st,

with the enclosed report of Professor Hamel on the two papers
4.2. TRUESDELL AGAINST THE ‘SUPPESIANS’ 93

by Sugar, Suppes, and me.7 I should like to tell you that we have
not been very impressed by Hamel’s criticisms, and that we still
want to publish these papers in their present form.
In the first place, with regard to Hamel’s criticism that our treat-
ment is restricted to ‘the most meager mechanics of points’, it
does not appear reasonable to us to object to a scientific paper
on the ground that it has not accomplished something which the
authors were not trying to accomplish: one does not criticize a
paper on linear differential equations for not also covering non
linear differential equations. We are of the opinion, moreover,
that as a preliminary to any adequate treatment of the mechanics
of extended bodies,8 it is desirable (or perhaps even necessary) to
present classical particle mechanics in a clear and precise form.
In addition, such a presentation would be useful for an analysis
of relativistic particle mechanics — and of continuum particle
mechanics, both in classical and relativistic form. (quoted from
[True.84, p.525])
They are correct. If that were not the case, no axiomatic presentation
of classical propositional logic would be possible, on the grounds that it
does not cover classical quantified logic, for instance. Furthermore, an ax-
iomatic treatment of continuum mechanics, a topic addressed by Truesdell
as lacking in the approach by the three authors, was developed by Wal-
ter Noll, a former student of Truesdell. It is interesting enough that Noll
presented three versions of such a theory, and the reasons would be ques-
tionable by Truesdell lights: should the first two not be enough to cover all
continuum mechanics? In this case, why Truesdell did not address similar
restrictions to Noll’s work? Well, this is a question to historians, and we
wish just to point that these kind of criticisms to the axiomatic method are
7
Hamel had said that the papers should not be published since the papers dealt only with “meager
mechanics” of points, leaving aside for instance the problem of the continua, that is the mechanics of the
continuum, where deformations are considered.
8
Recall that in particle mechanics, the particles are taken to be pontual, without dimensions.
94 CHAPTER 4. CRITICISM TO THE AXIOMATIC METHOD AND ITS DEFENSE

certainly misguided: Truesdell attacks the method by complaining about

restrictions on the scope of content of the mechanics being axiomatized.

4.3 Arnol’d and the Bourbaki program

The great Russian mathematician Vladimir Arnol’d famously made clear
his opinion about the axiomatic method in general and about Bourbaki’s
program of seeing mathematics as the science of structure, grounded in
that method [Arn.97]. Arnol’d was sanguine in criticizing AM and Bour-
baki, calling him “the devil” of a tendency, which had flourished in the
beginnings of the XXth century (mainly due to Hilbert, he recalls us), of
considering that AM should be propagated in mathematics.9 Arnol’d con-
sidered this “a self-destructive democratic principle”, for it led mathemat-
ics to depart from physics and the other sciences. The strict application of
this method, and surely he is referring here to what we have called abstract
axiomatics in the previous chapter, is that it turns any calculus in a blind
calculation, for the students are invited to acknowledge that the multiplica-
tion of standard numbers is commutative even if they know nothing about
numbers.
Arnol’d suggests that blind calculations should not substitute content,
and that the intuition of the mathematician is something quite important
to be dismissed in favor of purely logical deductions. In this sense, he
is at least partially in agreement with Imre Lakatos and his criticism of
formalization, claiming for intuitive mathematics. In our opinion, his un-
derstanding of the aims of the use of the axiomatic method are completely
misguided. We believe that no one, Hilbert and Bourbaki included, aimed
at substituting mathematics by blind calculations. Hilbert was right in
proposing that the use of formal methods has a finality: mainly, to avoid
that ‘intuitive’ elements not part of the theory play some decisive role in
9
It is well known the claim made by Hilbert in his conference on the 23 Mathematical Problems that
“[t]he mathematician will have also to take into account not only of those theories coming near to reality,
but also, as in geometry, of all logically possible theories.” [Hilb.76, p.15]
4.4. WORTH AXIOMATIZE? 95

our deductions. When such formalization efforts were launched it was

still very fresh in everyone’s mind how geometric proofs were performed
with the aid of geometrical figures, introducing unallowed elements into
the proof (a typical well-known example is Euclid’s Proposition 1).10 The
advantages of using the AM is precisely this: to provide the mathemati-
cian with solid grounds in the proofs: she knows what is being accepted,
the methods of deduction that are allowable by the underlying logic, and
then the kind of theorems she could obtain. Notice it was precisely this
kind of recognition of the importance of making explicit our underlying
assumptions that made possible intuitionistic mathematics, which does not
accept indirect proofs, or paraconsistent mathematics, where some kind of
‘contradictions’ are to be accepted.

4.4 Worth axiomatize?

The advantages of the AM, as posed in the last paragraph, seems to be

clear. Once fully axiomatized, a theory becomes (at least in principle)
a clear and objective object of study. We can know which are its basic
(primitive) concepts, which are its fundamental laws (postulates), and if
complete formalization is also considered, we can also know about its un-
derlying logic. The logical structure of the theory becomes completely
clear, for good or for bad. But perhaps one of the greatest possibilities
opened up by formalization and full employment of the axiomatic method
was that of metamathematical studies. During the XXth century, as is well
known, due to the process of formalization (say of arithmetics, set theory
and many other mathematical theories), we have gained a fine knowledge
of the theories and their limitations. Without such a process there would
be no Gödel’s incompleteness theorems, no proof of the independence of
the continuum hypothesis and of the axiom of choice (hence, no ‘non-
10
This proposition invite us to construct an equilateral triangle over a segment of line. The ‘proof’
assumes that two circles intercept, something that is quite ‘obvious’, but does not follow from the axioms.
96 CHAPTER 4. CRITICISM TO THE AXIOMATIC METHOD AND ITS DEFENSE

Cantorian mathematics’ would be possible), no adequate distinction be-

tween the theory’s language and its metalanguage, with all the metalogical
consequences (for instance, about the concept of truth and all discussions
around it), and so on. But there is still a question to be answered: should
this technique be applied to all domains, to all sciences? That is, is it
desirable, or should we take it as an ideal that every theory be formally
axiomatized? Here we make a digression on this point and put a challenge,
suggesting that perhaps the axiomatic technique is not something to be
pursued in all domains. Let us see why by focusing on physics (we leave
out human sciences in general, where the answer seems to be NO most of
the times, despite the advantages the AM may bring to these fields, as the
works of Suppes have shown).

We shall focus on present day physics of matter. Although we are not

specialists in this field, we think that we can say something on this respect,
at least from a philosophical and foundational point of views. What is rel-
evant for us is that such physics poses important concerns on the axiomatic
study of physics. By ‘present day’ physics of matter we mean the Standard
Model (SM) of particle physics. Roughly speaking, SM is composed by
some sub-theories: quantum electrodynamics (QED), the theory of weak
forces, and quantum chromodynamics (QCD). This theory unifies three of
the four fundamental forces of nature, namely, the electromagnetic force,
the weak force, and the strong force. Only gravitation is left out of the
picture. The mathematical description is in terms of group theory; they
are gauge theories, and we need to know their gauge groups. The first two
sub-theories are unified as what is known as the electroweak theory, whose
gauge group is termed S U(2) ⇥ U(1). It is not relevant for us here to ex-
plain the details, but only to be aware of the differences. The gauge group
of QCD is S U(3). The problem is to find a gauge group unifying all these
groups. Some authors try to define the group S U(3)⇥S U(2)⇥U(1), whose
combination is the symmetry group S U(5) [GlaGio.00]. The problem is
that until now there is no agreement about what S U(5) is, as Glashow tells
 4.4. WORTH AXIOMATIZE? 97

us. In other words, until now there is no consistent way of joining the
electroweak theory with QCD. They work as isolated pieces of a greater
framework, and are used depending on the subject the scientist is interested
in. But things are not so easy even for these sub-theories. Let us follow
some masters on the subject, Arthur Jaffe and David Gross (Nobel Prize
winner 2004).
Gross suggests that QCD does not exist as a mathematical sound theory
[Cao.99, p.164]. Jaffe suggests that this is a question of time, an com-
s
XXX pare the issue with some results in mathematics, where there are results
not known in a time, but that were put on well established basis some time
later. Gross doesn’t disagree, and claims that the search for a mathemati-
cally adequate basis is something that should be looked for. His point was
concerning the present day physics, and both of them agree that these fields
are not yet well developed on what concerns foundational aspects.
Thus, we may conclude that, at least in the present moment, physi-
cal theories form a mosaic composed of different sub-theories, which are
used depending on the field and on the particular problem being dealt with.
Furthermore, sometimes these sub-theories may be even incompatible one
XXX with another, and even so they constitute the mosaics of physics. Typi-
cal examples, besides those mentioned in the previous paragraph, are the
quantum field theories (QFT — QED, QCD and the weak theory are quan-
tum field theories) and general relativity, which comprises the fourth force,
gravity. The attempts of joining the three forces with gravity conduce to
Quantum Gravity, the Holy Grail of present day physics. But Gross em-
phasizes also in his own paper in the book, that an attempt to make sense
to QFT (in general) is still required [Gro.99]. The problem is that we have
not this general theory today.
Well, we may say: should we search for such a unification in an ax-
iomatic fashion?11 From the foundational point of view, of course the
answer is yes. The fact that we still do not have a fully developed uni-
11
This question does not ask for Quantum Gravity, but for an axiomatic quantum gravity.
98 CHAPTER 4. CRITICISM TO THE AXIOMATIC METHOD AND ITS DEFENSE

fied theory may make attempts at axiomatization premature, but that is

not a conclusive argument. Recall that one of the objections to the ax-
iomatic method (by Hempel and others, see a discussion in Suppe [Sup.77,
pp.110-5]) is that it really only predates a well-developed theory; the ax-
iomatic method cannot work on a vacuum. In fact, it can only order a field
of knowledge where there is something reasonably well-developed in or-
der to be ordered. So, the objection would go, it is unreasonable to try to
apply the axiomatic method to those areas of physics yet.
Our answer to those concerns comes from a balance between two ex-
tremes. We should obviously not force every field of knowledge into the
form of the axiomatic method; they may well not be ready for that. Ax-
iomatization is not a mark of scientific status per se. Not even the Logical
Positivists were so rigid in their demands for formalization (see again Lutz
[Lut.12]). On the other hand, to claim that no benefits result from the ap-
plication of the axiomatic method would require a stubborn blindness to
the evidences we have already pointed out of how useful it may be. Per-
haps the best thing is to follow Halvorson [Hal.15, pp.2-3] and allow that
science itself determines which are the fields of scientific knowledge that
could profitably be axiomatized at a time. Scientific development itself
seems to indicate that a field is ready to axiomatization and that it may be
useful trying to axiomatize a theory. Obviously, this is as vague an answer
as it gets, but we would like not to put our philosophical preferences to
cause trouble to science.

4.5 Does axiomatization really axiomatizes?

The next question to be considered regarding the AM is about its results,

something that was already advanced in the Introduction. More specifi-
cally, the question is: suppose we have an informal theory T , say Darwin’s
evolution theory, classical physics before Newton (Galileo, Kepler, and
others), but let us fix Darwin. Suppose now that we aim at axiomatizing
4.5. DOES AXIOMATIZATION REALLY AXIOMATIZES? 99

his theory. This was done by some people, like Mary Williams [Will.70]12
and many others since then. The question is: does Williams’ (or any other)
system captures axiomatically Darwin’s theory? It will all depend on how
deep the idea of an axiomatic system capturing some informal theory goes.
If such a capturing is understood as complete identification between ax-
iomatic and informal theory, then, perhaps the answer is ‘No’, or, at least,
we cannot prove that. In our opinion, axiomatization and formalization
create something that certainly resembles the original theory, but cannot
be said to be that theory in the sense of identification. By axiomatizing
T , we create T 0 and work as if T 0 were T ; we get results in T 0 which are
interpreted in terms of T and so on. This remark is to be taken into account
in all that follows in this book.
Even informal axiomatics, as we have seen, which is most of the times
advanced in order to organize a certain field of knowledge known in ad-
vance, and it is applied, as Lakatos and Hempel have suggested, only to
‘mature’ domains, surely leaves out much of the original theory, which has
imprecise and wide borderlines. Thus, in elaborating an axiomatic version
of a certain theory, the scientist has no way to assure us that the axiomatic
system she has obtained really corresponds to the informal domain she
had before. Let us make an analogy with a famous case. In recursion
theory, there is an important result known as Church thesis (or Church-
Turing thesis, referring to the American mathematician Alonso Church and
the British Alan Turing) which says that all computable functions are re-
cursive. In short, those functions defined on the natural numbers which
are computable in the informal sense (we can effectively get the outcomes
given some inputs) are formally computable, or recursive. Recursive func-
tions have a precise definition which does not concern us here. It is gen-
erally accepted that recursive functions are computable, for they were de-
fined precisely to play this role. The thesis says the converse, and cannot
be formally proven, although some have tried. The reason is that we cannot

12
The use of the axiomatic method in biology was initiated by John Woodger in the 1930s.
100CHAPTER 4. CRITICISM TO THE AXIOMATIC METHOD AND ITS DEFENSE

use the resources of logic to compare a formal concept (that of recursive

function) with an informal one (that of computable function). The point
is that the informal counterpart lacks adequate precision to be compared
with the concept analyzed on logical grounds. By the way, this is just one
of the roles of the axiomatic method, namely, to produce a framework that
can be analyzed in logical terms. The analogy is pertinent. In axiomatizing
an informal field, like Darwin’s natural selection, we get an axiomatic or a
formal theory which may resemble the original one, but due to the impre-
cision of the first, which in general has not its basic concepts clearly stated
or defined, the equivalence of the two must be accepted as a thesis, similar
to Church’s one. In other words, we need to postulate, in general grounded
on the success of the axiomatic version in explaining the informal field,
that the two theories fit. But there cannot be a proof of this result.

As for another example, we present the usually mentioned fact that we

can formulate Aristotle’s theory of categorical syllogism in the language
of first-order logic. It suffices to write the ‘A’ propositions (universal af-
firmatives) as 8x(S (x) ! P(x)), where S and P are the subject and the
predicate terms respectively, the ‘E’ propositions (universal negations) as
8x(S (x) ! ¬P(x)), the particular affirmatives (‘I’) as 9x(S (x) ^ P(x))
and the particular negations (‘O’) as 9x(S (x) ^ ¬P(x)). As postulates, we
can use the inference rules derived from the first figure Barbara, Celari,
Darii and Ferio (or, more economically, it is enough to take Barbara and
Celarent, as is well known), adequately translated to the logical notation.
Have we succeed in ‘reducing’ syllogistic to first-order monadic predicate
calculus? That is clearly doubtful. This syllogistic theory, inspired in Aris-
totle and in further developers, cannot be said to capture all the richness
and complexity of the informal discussion given in the Prior Analytics,
for instance, in distinguishing ‘demonstrative premise’ from ‘dialectical
premise’. The first one is either true or false and has been obtained from
initial assumptions, while the dialectical one just poses a contradiction. In
other words, the demonstrative premises are the parts of a contradiction
4.5. DOES AXIOMATIZATION REALLY AXIOMATIZES? 101

while the dialectical ones are just asking for the contradiction [Aris.89,
24a10, 24b10]. There are huge difficulties in translating this to our present
day logical language. If by a contradiction we understand the conjunction
between a proposition and its negation (what depends on the meaning of
these connectives, that is, of the logic involved), something like ↵ ^ ¬↵,
then both ↵ and ¬↵ would be demonstrative premisses, while the move to
ask if this expression either is or not a contradiction would be considered
as dialectic. Of course this discussion cannot be developed within such a
poor formal theory.
But these remarks do not entail that the axiomatization of a certain field
is not relevant, and let us insist on this point a little bit more. As a first
point, we would like to recall our discussion made before. Recall that we
employ the axiomatic method either as a formal method or else as accord-
ing to Suppes’ suggestion, by devising a Suppes’ predicate. In both cases
we are axiomatizing a field of knowledge, characterizing a theory for cer-
tain purposes. The fact that an axiomatized theory is a way of representing
an informal theory (as French [Fre.15] insists) for philosophical purposes
does not diminish its relevance. Philosophers have many intricate ques-
tions about science that require some way to represent theories. We would
like to discuss what does it mean to say that a theory is true, or quasi-true,
or that it is empirically adequate, and so on. How can that be done with an
informal approach to theories? The axiomatic method helps us discussing
such issues, not only for specific cases where axiomatization is available,
but also in abstracto, by furnishing a general guideline of what a theory
would look like when an axiomatization eventually is developed. So, in
this sense, a ‘rational reconstruction’ effected by the axiomatic method is
certainly something to be sought by philosophers too.
As a second point, it is important once again to point to how useful
the AM may be for scientific purposes. Suppose again that we are deal-
ing with arithmetics. Without formalization, non-standard models would
not become available for investigation. The same can be said of several
102CHAPTER 4. CRITICISM TO THE AXIOMATIC METHOD AND ITS DEFENSE

other domains. Patrick Suppes emphasized the heuristic role of the ax-
iomatic method. As an example, he mentions the axiomatization of the
field of real numbers. As we have already mentioned, Suppes [Sups.83]
says that Dedekind’s and Cauchy’s definitions of real numbers respectively
by means of cuts and by equivalence classes of Cauchy sequences, are
not easy to comprehend and deal with. On the other hand, the axioms of
a complete ordered real field provides a useful mathematical device, and
even the least-upper bound axiom can be understood without difficulty.13
Kolmogorov’s axioms for probability are the second example. According
to Suppes, they provide a clear and intuitive foundation for the concept of
probability. Thus, a system of axioms is heuristic, according to Suppes,
if they provide an intuitive and organized way of looking to the involved
concepts, organizing and facilitating “our thinking about the subject, and
in particular our ability to formulate, in an ordinary self-contained way,
problems concerned with the phenomena governed by the theory and their
solutions” (op.cit.). Non heuristic axioms provide a “sophisticated mathe-
matical foundation of a discipline, but [they are] formulated in such a way
they prohibit natural and intuitive ways of thinking about problems, spe-
cially new problems in the discipline” (ibid.). As an example, we recall
that he mentions Mackey’s axiomatics for orthodox quantum mechanics.
We have no space to discuss the details here, to which we recommend
Suppes’ paper.
The important point to be emphasized is, as we have suggested already,
that without axiomatization, and more, without formalization, we would
not be in a condition to notice that our room has some closed windows
which, once opened, show us new landscapes and even new universes.
This is true of the non-standard models of arithmetics, as we suggested,
but also with non standard models of the real numbers, the rise of hyper
sets, obtained by dropping the axiom of foundation in set theory,14 the ex-
13
This axiom is essential to characterize the structure for it provides its topological completeness.
14
Okay, you may say, these sets were already noted for instance by Mirimanov in 1917, but the impor-
tance of von Neumann’s axiom of foundation cannot be ignored for the formulation of the intuitive notion
4.5. DOES AXIOMATIZATION REALLY AXIOMATIZES? 103

istence of denumerable models of the first-order ZFC set theory (supposed

consistent), Gödel’s incompleteness theorems, Tarski’s theorem about the
undefinability of the notion of truth, and so on. It is not an absurd for sure
to suppose that the same new vistas may be opened with the axiomatic
study of empirical sciences, once adequately formalized. In this sense, ax-
iomatization is indeed welcome. We leave this question for the reader to
think, as a challenge for the philosophical reflection, and advance now to
the topic of axiomatization of scientific theories properly speaking.

of a set.
104CHAPTER 4. CRITICISM TO THE AXIOMATIC METHOD AND ITS DEFENSE
Chapter 5

Axiomatization, and Scientific Theories

ow that we have discussed axiomatization in general and the

N main features of the axiomatic method, it is time to put that

method to work in the case of scientific theories. As dis-
cussed in a previous chapter, the fact that we are using for-
mal tools and axiomatization does not collapse our approach directly in
the traditional Received View. Rather, as we shall discuss later, we follow
mainly the approach by Suppes and employ formal tools in the analysis of
scientific theories wherever possible.
In this chapter we shall proceed as follows. We begin by outlining what
we call an ‘external approach’ to axiomatization; that is, an approach in
which a theory is axiomatized along with all the mathematics required,
in particular, set theory. We then evaluate rather briefly its inadequacies
in the case of a treatment of the models of such theories. Basically, the
main problem comes to the fact that by axiomatizing huge portions of set
theory, an axiomatic system like that would have to have as models large
parts of models of set theory as well! As we shall discuss very briefly, that
is an impractical task for those willing to address philosophical problems
concerning scientific theories (see our previous discussion).
The next step then consists in presenting two distinct approaches work-
ing inside set theory (recall that we have already mentioned that set theory
would be our framework), so that we may call those approaches ‘internal
106 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

approaches’ to axiomatization. The first approach we shall present will

be called ‘da Costa-Chuaqui’s approach. It is based on the work of da
Costa and Chuaqui [CosChu.88] and da Costa and Rodrigues [Cos.07],
and consists of a formalization of the concept of a Suppes’ predicate, by
employing specific formal languages (the languages of the theory, or of
their structures, as we shall see) build inside the language of set theory to
frame the axioms. The second approach is the one by Suppes himself. It
employs the language of set theory itself to develop the axiomatization of
a theory (see [Sups.57] and [Sups.02]).
Both approaches differ in many respects. In particular, both are attempts
to axiomatize a theory by selecting a class of models. However, their re-
lations to models are very different, as we shall see, and the purposes for
which each analysis is used may vary. We shall discuss the benefits and
peculiarities of each approach, arguing in the end for a kind of pluralism
in this aspect: the philosopher of science is free to employ the most appro-
priate analysis according to the needs of the day.

5.1 External axiomatizations

In the traditional debate between adherents of the so-called semantic ap-

proach against the Received View, a very common argument presented
against syntactical analyzes of scientific theories concerns the nature of
axiomatization required by the Received View. As we have already dis-
cussed, the folklore goes to claim that the axiomatization must proceed in
first-order language with identity, by distinguishing a theoretical and an
observational vocabulary, and by providing an interpretation by the Corre-
spondence Rules (cf. the standard exposition of the Received View in the
Introduction of [Sup.77]). Of course, that view has by now been somewhat
demystified (see for instance [Lut.12]), even though this kind of reading is
still not widely known.
A typical argument against the plausibility of the Received View ap-
5.1. EXTERNAL AXIOMATIZATIONS 107

proach concerns the axiomatization of theories requiring large portions of

mathematics, or even of portions of set theory (this is a point advanced by
Suppes [Sups.67, p.58]). In an axiomatization of geometry, for instance,
one may need to use portions of set theory, to deal for instance with lines
as sets of points. In more mathematical theories, like probability theory,
one may need the theory of real numbers. So, axiomatization of those the-
ories would require that we axiomatize portions of set theory and all the
mathematics needed. Of course, one option is to use the set theory to de-
velop the mathematics required, like the theory of real numbers. Anyway,
this kind of approach is said to be too laborious, requiring the unpractical
axiomatization of all the relevant mathematics.
As we mentioned, it is now recognized that it would be unfair to at-
tribute this view to the adherents of the Received View ([Lut.12] is very
convincing about that). However, that is not enough to prevent one from
axiomatizing theories according to that approach. In fact, Worrall [Wor.07,
p.148] and Zahar [Zah.04, p.7] seem to promote this kind of axiomatization
through their Ramsey approach to scientific theories. That is, their idea is
that set theory is involved in the axiomatization of a scientific theory in the
sense that one must either use set theoretical notions in the axiomatization
(Worrall) or else axiomatize set theory as part of the work of axiomatiz-
ing the theory itself (Zahar). So, this is a live option for those adopting a
syntactical approach to theories and a version of epistemic structural real-
ism. Given that we are interested in the syntactical as well as in the se-
mantic aspects of theories, we shall present this approach with some care
and argue that it has serious shortcomings when it comes to deal with its
models. Later we then move on to what we believe are more appropriate
approaches.
According to the view being discussed, we can say that the axiomatic
basis of a theory T encompasses three levels of postulates:

(i) The logical postulates — in general it is said that these encompass

only those postulates of classical first-order logic with identity. But in
108 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

the general setting, we could use also higher-order logics, infinitary lan-
guages or logics, or still other non-classical logics.
(ii) A group of ‘mathematical’ postulates — in general, postulates for
first-order ZFC or, as Zahar prefers, NBG (von Neuman-Bernays-Gödel
set theory, see [Men.97, chap.4]). Perhaps only a small part of ZFC is
to be used, say Z p , the ‘pure’ Zermelo set theory (but without Urele-
mente, although these entities may be interesting in empirical sciences
to represent things that are not sets) plus the axiom of choice. The ques-
tion of inquiring how much of set theory physical theories demand for
is still open and of course depends on the theory. But some mathemati-
cians think that replacement axioms are not necessary, and that only the
denumerable form of the axiom of choice is enough.
(iii) The specific postulates, which depend on the particular field being
axiomatized. According to this scheme, these postulates would be sen-
tences of the language of ZFC, perhaps enriched by additional concepts
and terms referring to the supposed domain of investigation. In the case
of the Received View, the specific vocabulary would still need to be
separated into theoretical terms and observational terms.

The notion of deduction is that one of the underlying logic, in general,

in accordance (when no other logical symbols are introduced) with ele-
mentary (first-order) logic, but now in the deductions we may have axioms
of ZFC as premises, so the logic is not ‘exactly’ elementary, as is neces-
sary for the consideration of higher-order theories and languages, typical
of scientific theories. Indeed, in first-order set theory the intended inter-
pretation regards the individual variables as representing sets, viewed as
collections of objects, which by their turn can be also sets. Since in exten-
sional contexts we can read a set as being the extension of a property (the
property shared just by the elements of the set), ‘to belong to a set’ stands
for ‘to have a property’, so in quantifying over sets, we are quantifying
both over individuals and properties. It is in this sense that set theory acts
5.1. EXTERNAL AXIOMATIZATIONS 109

as a higher-order language, despite being (in most cases) formulated as a

first-order theory.

For example, consider an axiomatization of classical particle mechan-

ics. According to the approach here proposed, it would be comprised by
an axiomatization for first-order logic, axioms for set theory, and finally,
in the third stage, axioms for classical particle mechanics. Notice that in
formulating such axioms we may employ concepts of mathematics, such
as vector spaces, real numbers and functions of real numbers. All those
concepts may be developed inside set theory, in the second level of the
axiomatization. Surely this provides for a powerful ‘mathematical basis’
for a scientific theory, by allowing all of classical mathematics available.
However, this approach to scientific theories also has its drawbacks.

What are the main difficulties with this kind of approach? Well, as we
have seen earlier, ZFC (if consistent) cannot provide for its own models.
So, besides being an awkward approach for axiomatization, this form of
presentation of theories is also unduly laborious for those willing to take
into account the models of the theory: we will have to employ a stronger
metatheory in which the models of ZFC (or whichever set theory we hap-
pen to be employing) are developed. So, for instance, whenever we are
interested in particle mechanics or quantum mechanics, say, we will have
to take into account not only the models of the specific axioms for those
theories, but also huge structures modeling the axioms of set theory. That
goes beyond what any philosopher of science would be willing to take into
account.

Perhaps some simpler approach that employ set theory but does not
require a model of it every time it must be employed should be more plau-
sible. Let us check the alternatives. Before presenting the approaches, let
us introduce the terminology that will guide us; it is common to both the
Suppes’ approach as well as the da Costa-Chuaqui approach.
110 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

5.2 Structures
We shall be working within first-order ZFC set theory. In this section,
we follow the approach presented in [Cos.07], but, differently from the
exposition there, we allow that the domain of the structures be composed of
more than one set, and allow other simplifications which will be explained
later.
Within ZFC we can, at least in principle, construct particular structures
for mathematics and for empirical sciences —sometimes we may need to
strengthen ZFC, say with universes (which enable us to deal also with cat-
egory theory, although we shall be restricted here to set-theoretical struc-
tures). As we shall see, the language of ZFC, termed L2 , taken here to
be a first-order language, is so powerful that using it we construct even
languages which are not first-order and structures which are not order–1
structures (these definitions shall be introduced in what follows). To go to
some details, we need to introduce a few basic definitions we present in the
next sections. A remark is in order here: we usually work with extensions
by definitions of the language of set theory, comprising defined symbols
like ✓ (for set inclusion), @0 , and so on. But we shall continue to speak of
the language of set theory also for these extensions.
The general idea can be seen with an analogy with group theory. Start-
ing with a non-empty set G, by using the set-theoretical operations we ob-
tain P(G ⇥G ⇥G), and then we may choose an element of this set (which is
a set whose elements are collections of 3-uples of elements of G) satisfying
the required properties (the group axioms of course).1 The corresponding
structures (namely, the groups) are order–1 structures.
Summing up, what we are going to see is that a mathematical structure,
such as those used in mathematics and in the empirical sciences, can be
constructed within a set theory such as ZFC (with or without ur-elements).
The figure below provides a schema of this situation, using ‘pure’ set the-
1
The binary group operation ◦ is a certain function from G ⇥ G in G or, what is the same, it is a certain
ternary relation on G, that is, a collection of triples ha, b, ci of elements of G, with c = a ◦ b.
5.2. STRUCTURES 111

On
A "
A " V
A "
A "(D) "
A AA "" "
A A A = hD, r ı i " "
A A " "
A A " "
A "
A D "
A "
A "
A "
A "
A "
A "
A"
;
Figure 5.1: The well-founded ‘model’ V = hV, 2i of ZFC, a structure A built within ZFC,
and the scale "(D) based on the domain D. For details, see the text.

ory (without ur-elements) with an axiom of foundation, as in standard first-

order ZFC. It should be remarked that while the structure A (to be defined
below) is a set, the whole universe V is not, as we have seen in the last
chapter.
Our first definition is of the set T of types. These types are not to be
confused with the types of Russell’s Simple Type Theory. This set will be
important also for the development of formal languages we shall need to
make explicit the da Costa-Chuaqui approach in the next sections. When
we speak of the set of types, we always mean the set T given by the fol-
lowing definition:

Definition 5.2.1 (Types) The set T of types is the least set satisfying the
following conditions:
(a) 0, 1, . . . , n, with 0  n < !, are types. These are the types of the
individuals, , where ! is the first transfinite ordinal.
(b) if a1 , . . . , an 2 T, then ha1 , . . . , an i 2 T, 1  n < !.

Thus, for instance 0, . . . , n−1, h0i, h0, 0i, h0, 1i, hh3i, 7i, hh0ii are types.
112 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

Intuitively speaking, in this list we have types for the individuals of the ba-
sic sets, that is, the sets that form the domains of the structures, as we shall
see below. Thus, the above types stand for sets (or properties) of individu-
als of the set whose elements have type 0, for binary relations on individ-
uals of the same set, for relations between the sets whose elements have
types 0 and 1 respectively, for binary relations between sub-collections of
the set whose individuals have type 3 and the set whose individuals have
type 7, and finally for sets whose elements are subsets of the set whose
elements have type 0.

Definition 5.2.2 (Order of a type) The order of a type, Ord(t), is defined

as follows:
(a) Ord(k) = 0, for k = 0, . . . , n, with 0  n < !.
(b) Ord(ha1 , . . . , an i) = max{Ord(a1 ), . . . , Ord(an )} + 1.

Thus, for instance, Ord(h0i) = Ord(hk1 , k2 i) = 1, with k1 , k2 2 {0, . . . , n},

while Ord(hk1 , hk2 ii) = 2. Relations will be understood here extensionally
(collections of n-tuples) and being of finite rank (that is, having finite ele-
ments only). Unary relations are sets.

Definition 5.2.3 (Order of a relation) The order of a relation is the order

of its type.

Thus, binary relations of individuals are order-1 relations, and so on.

For instance, a binary relation on D is an element of t(hk1 , k2 i) = P(t(k1 ) ⇥
t(k2 )) = P(Dk1 ⇥ Dk2 ). According to definition 5.2.3, the type of a binary
relation on Dk is hk, ki, as intuitively expected.
These definitions will be useful in our discussions to come, when we
shall need to talk about the order of a structure and the order of a lan-
guage. The following definitions help us to construct relations and proper-
ties based on {Dn }, a non-empty family of sets, each one with elements of
a type, which will soon be called the domains of the structure. We write
5.2. STRUCTURES 113

Dn for the nth member of that family. In the next definition, the usual set-
theoretical operations of power set and Cartesian product are being used,
denoted respectively by “⇥” and “P”.
We shall introduce a function ⇡ as follows:

Definition 5.2.4 (Scale based on {Dn }) Let {Dn } = {D1 , D2 , . . . , Dn } a fam-

ily of non-empty sets whose elements have types t1 , . . . , tn respectively.
Then, we define a function ⇡, called a scale based on {Dn } and having
T as its domain, recursively as follows

(a) ⇡(tk ) = Dk (k = 1, . . . , n)

(b) If a1 , . . . , an 2 T, then ⇡(ha1 , . . . , an i) = P(⇡(a1 ) ⇥ . . . ⇥ ⇡(an )).

(c) The scale based on {Dn } is the union of the range of ⇡(⌧), with ⌧ 2 T
and it is denoted by "({Dn }).

Let us give an example. Since a binary operation on a set G (as the

group operation) can be seen as a ternary relation on G, we can deal with
it as follows. Having the scale "(G) (here the family of sets has just
one element, the set G, and let t be its type), we just take an element of
⇡(ht, t, ti) = P(D ⇥ D ⇥ D) satisfying well-known conditions (the group
postulates, written in this ‘relational’ notation). In this sense, it can be
shown that we can map Bourbaki’s echelon construction schema [Bou.68,
chap.4] within this schema.

Definition 5.2.5 The cardinal KDn associated to "({Dn }) is defined as

n
[ n
[ n
[
2
KDn = sup{| Dk |, |P( Dk )|, |P ( Dk )|, . . .}
k=1 k=1 k=1
.
Sn Sn
Here, | k=1 Dk | denotes the cardinal of the set k=1 Dk .
114 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

Definition 5.2.6 (Structure) A structure E based on a family {Dn } is a

n + 1-tuple
E = hD1 , . . . , Dn , Rι i (5.1)
where Di , ; and Rι represents a sequence of relations of degree n be-
longing to ε({Dn }). These relations are called the primitive elements of the
structure.
Here, Rι is a sequence of elements of ε({Dn }), and we suppose that the
domain of this sequence is strictly less than KDn . We say that KDn is the
cardinal associated with E, and that ε({Dn }) is the scale associated with E.
As we said before, each element of ε({Dn }) has a certain type, for it
belongs to π(t) for some t 2 T. Now, as we have seen, the order of a relation
is defined as the order of its type. The order of E, denoted Ord(E), is the
order of the greatest of the types of the relations of the family Rι , if there
is one, and if there is no such relation, we put Ord(E) = ω. Usually, the
relations of Rι have as relata the elements of the basic sets, but more general
structures enable that the relata can also be higher-order individuals, that
is, sets of elements of Di , sets of sets of these elements, and so on.
In the beginning of this section we remarked that our presentation makes
some simplifications on [Cos.07]. Here we depart from da Costa’s original
work in that we allow individuals and operations to occur in the structure,
whereas da Costa reduced operations to relations and identified individu-
als with their unit sets. The main point of this change is to simplify the
exposition, and from a mathematical point of view the difference is purely
a matter of convention.2 So, in the definition of structures, the objects in
the family Rι may be not only relations, but operations as well, that is, rela-
tions satisfying the well-known functional condition, or even distinguished
elements from the domain, which we take to be 0−ary operations. In these
cases we employ as usual the common notation for functions and objects.
2
In fact, to allow individuals as we are doing or to identify them with their unit sets are both commit-
ments to individuals anyway. If one is wishing to avoid individuals for structuralist reasons, then maybe
ZFC is not the best framework to work with, since individuals are always given in the domain of the struc-
ture. Interesting as it is, we shall not concern ourselves with structuralism in this book.
5.3. LANGUAGES AND ALGEBRAS 115

Let us give some examples with due simplifications in the notation. For
instance, a structure such as
K = hK, +, ·, 0, 1i (5.2)
can be used for representing fields, where K is a non-empty set, + and ·
are two binary operations on K, and 0, 1 2 K. Postulates for fields are
added to this description. For vector spaces over a field K, we may write
V = hV, K, +V , ·V , +K , ·K i, where V is the set of vectors, K is a set, the
domain of a field, +V is a binary operation on V, called addition of vectors,
·V is a function from K ⇥V in V, the multiplication of vector by scalar, +K is
a binary operation on K, the addition of scalars, and ·K a binary operation
on K, the multiplication of scalars. All these concepts are subjected to
well-known postulates. For simplicity, we usually omit the reference to
field operations, leaving them presupposed in saying that K is a field, and
write simply
V = hV, K, +, ·i. (5.3)
More sophisticated structures suitable for representing theories in the
empirical sciences, as we shall see below, follow the same schema. For
instance, as we shall see, a structure for classical particle mechanics is a of
the form
M = hP, T, m, s, f, gi,
where P is a non-empty set, T is an interval of the real number line, m is
a function from P to the set of non negative reals, and the remaining con-
cepts are suitable functions we shall see again later in this chapter. Surely
the reader acknowledges that with some effort this structure can be put
according to the above definitions.

5.3 Languages and algebras

We now present the approach to axiomatization based on the development
of da Costa and Chuaqui [CosChu.88]. This work is an attempt to formal-
116 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

ize the idea of Suppes predicate, employing formal languages and char-
acterizing a class of models by employing formal axioms. By a Suppes
predicate we understand a set theoretical predicate in the lines of Suppes,
but as we shall see there are some differences from the definition and the
way Suppes uses it. Now, we shall discuss an approach to formal languages
which will make use of the notion of structure just developed. Here, lan-
guages will be seen as a special kind of structure, a free-algebra. Our aim
is to develop some languages, as the language of simple type theory, but
before doing that, first we recall some useful concepts of Universal Alge-
bra.

Definition 5.3.1 Let Rι be a family of n-ary relations over sets whose el-
ements have types t1 , . . . , tn , and let D1 , . . . , Dn be a family of sets whose
elements have types t1 , . . . , tn respectively. Thus, the relations are rela-
tions over these sets by hypothesis. Let us consider the structure E =
hD1 , . . . , Dn , Rι i. The similarity type, or signature, of this structure is a n-
tuple ha1 , . . . , an i where the ai are the types of the relations in the structure.

For instance, a structure for group theory can be taken as G = hG, ◦, −, ei

(as we shall see below). Here, ◦ is a binary operation over G (hence a
ternary relation on G, − is a mapping from G to G and e 2 G. Hence, the
type of similarity of the structure is hh0, 0, 0i, h0, 0i, 0i.
We say that two structures have the same similarity type when the n-
tuple ha1 , . . . , an i of types is equal for both, that is, their relations have the
same type and the family in the domain is composed of the same number
of elements. Since a family is always ordered, the same type of relations
occur always in the same order when the structures have the same similar-
ity type. A structure for fields is of the similarity hh0, 0, 0i, h0, 0, 0i, 0, 0i,
and so on.
Here we shall use a notion of algebra to mean mainly a structure com-
posed by a set (the domain of the algebra) and a collection of operations
and/or distinguished elements of the domain, according to the standard use
5.3. LANGUAGES AND ALGEBRAS 117

in Universal Algebra [BarMac.75, p.2]. Thus, being s = ha1 , . . . , an i as

above, a n-tuple of types, we have:
Definition 5.3.2 A s-algebra A is a structure hD1 , . . . , Dn , Rι i such that
Ord(Ri )  1 for each i and A has similarity type s.

The restriction to order 1 or less serves to make each s-algebra an al-

gebra in the usual sense, that is, a set with a family of operations and/or
distinguished elements defined over this set and distinguished elements
taken from the domain.
The definitions given below will concentrate in structures having a sole
domain for simplicity. The general case can be obtained by extending the
mappings to all the sets of the domain, if there are many, by a procedure
similar to that one given by Bourbaki [Bou.68, Chap.4]. More general
structures, as those we have been mentioning, will appear again soon.
Definition 5.3.3 (Homomorphism between algebras) Let A, B be s-algebras
with A = hA, Rι i and B = hB, R0ι i. A homomorphism from A into B is a
function ϕ : A 7! B such that, for all Ri of the s-algebra A whose type is
ha1 , . . . , ak i,

ϕ(Ri (x1 , . . . , xk )) = R0i (ϕ(x1 ), . . . , ϕ(xk )).

Let us give an example using groups. Suppose two groups G = hG, ◦i
and G0 = hG0 , ◦0 i. A homomorphism from G into G0 is a function ϕ : G 7!
G0 such that, for every a, b 2 G, we have
ϕ(a ◦ b) = ϕ(a) ◦0 ϕ(b).
A homomorphism between two vector spaces W and W0 over the same
field is a linear transformation from W to W0 .
Definition 5.3.4 (Isomorphism) Let A, B be s-algebras and ϕ : A 7! B a
homomorphism. If ϕ is a bijection we say that it is an isomorphism between
the algebras.
118 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

Definition 5.3.5 (Free Algebra) 3 Let X be any set, let F be a s-algebra

with domain F and let σ : X ! F be a function. We say that hF, σi
(also, for simplicity, denoted by F) is a free s-algebra on the set X of free
generators if, for every s-algebra A and function α : X ! A, there exists a
unique homomorphism ϕ : F ! A such that ϕσ = α.

The following diagram intends to clarify the definition using a simpler

notation.
σ
X - B
H
i HH σ(i) = xi
H
HH
H ϕ(xi ) = yi
HH
H ϕ
HH
α H
HH
j ?
H
α(i) = yi A

Figure 5.2: Free algebras: ϕ ◦ σ = α, that is, the diagram commute.

Theorem 5.3.1 For any set X and any similarity type s, there exists a free
s-algebra on X. This free s-algebra on X is unique up to isomorphism.

We shall not prove this theorem here (but see [BarMac.75]). Now, we
pass to the development of languages properly, which will be considered
as specific free s-algebras as presented above.
Let us suppose a simple case to exemplify how can we see what a lan-
guage is from an algebraic point of view. In order to make the ideas clear,
we shall consider in parallel an example, taking the language of classical
propositional logic, LCPL .
Thus, suppose we have a set X = {xi }i2I (this set may be infinite) and
another set fλ whose elements will be denoted generically by f . We shall
use for this set the same notation we used above to define an algebra. This
3
Most texts on Universal Algebra speak of absolutely free algebras. We shall omit the term ‘absolutely’
from now on.
5.3. LANGUAGES AND ALGEBRAS 119

second set plays the role of the operations of the algebra we shall define. In
our sample case, X = {p0 , p1 , . . .} is the set of propositional variables, and
fλ = {¬, !} are the propositional connectives (we could use any adequate
set of connectives of course). We still suppose that there is a mapping
ar : fλ 7! ω, the arity function, which assigns a natural number to each
element of fλ , called its arity. For instance, ar(¬) = 1 and ar(!) = 2.
Now we define by induction a set F as follows:
(1) F0 := X [ {h f, x1 , x2 , . . . , xar( f ) i : f 2 fλ ^ xi 2 X}. This is the
set of the atomic formulas. The (ar( f ) + 1)-tuple h f, x1 , x2 , . . . , xar( f ) i
is written f x1 x2 . . . xar( f ) . In our example, the atomic formulas are the
propositional variables and the expressions of both forms ¬pi and !
pi p j . This last one may be abbreviated by pi ! p j .
(2) Now let αi , i = 1, . . . denote atomic formulas. Then, the set of
complex formulas is F1 = {h f, α1 , α2 , . . . , αar( f ) i : f 2 fλ ^ αi 2 F1 },
and again the tuples are abbreviated by f α1 α2 . . . αar( f ) . For instance,
in our example, we may have as complex formulas expressions such
as ¬¬pi , ! ¬p1 ! p2 ¬p3 , this last one being abbreviated by ¬p1 !
(p2 ! ¬p3 ), according to a standard notation. In our case study, we
can introduce other operational symbols by definition, such as the other
standard connectives ^, _ and $.
(4) In a general way, let Fn := {h f, α1 , α2 , . . . , αar( f ) i : f 2 fλ ^ αi 2
Fn−1 }, then
(5) F = n2ω Fn .
S

We can prove that A = hF , fλ i is a free algebra by adapting the proof

given in [BarMac.75, p.5]. Thus, the language of the classical proposi-
tional calculus is a free algebra with @0 generators.
We call a free algebra constructed this way a language. Thus we can
see how a language can be constructed in ZFC. The cases of first-order
usual languages, higher-order languages and even infinitary languages can
120 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

be treated the same way, although they will demand more details. Anyway,
they can be treated as being certain free algebras. Thus, when we mention
certain languages to speak of structures in the next section, they can be
considered as constructed within ZFC.

5.4 Languages for speaking of structures

Our goal is to speak of structures and of all the objects of a scale. To do
that in an adequate way, we consider two basic infinitary languages, termed
Lωωω (R) (or simply Lω (R)) and Lωωκ (R)[CosRod.07].
In general, an infinitary language Lηµκ , with κ < µ being infinite cardi-
nals and 1  η  ω, enables us to consider conjunctions and disjunctions of
n  µ formulas and blocks of quantifiers with m < κ many quantifiers. The
superscript η indicates the order of the language (first-order, second-order,
etc.). In both cases R is the set of the constants of the language. Thus, in
Lωµω (R) (ω < µ) we may have infinitely many conjunctions and disjunc-
tions of formulae, but blocks of quantifiers with finitely many quantifiers
only. Lωωω (R) is a higher-order language, suitable for type theory (higher-
order logic). Standard first-order languages are of the kind L1ωω , so is L2 .
Put in a more precise way,
Definition 5.4.1 (Order of a language) A language Lnµκ , with 1  n < ω,
is called a language of order n. A language Lωµκ is said to be or order ω.
A language of order n contains only types of order t  n and quantifi-
cation of variables of types having order  n − 1 [CosRod.07].
In order to exemplify how we can define a higher-order language using
a first-order language (such as L2 ), let us sketch the language Lωω1 ω (R), but
we could consider whatever infinitary language Lωµκ , provided that the in-
volved cardinals exist in ZFC (for instance, we couldn’t use an inaccessible
cardinal).4
4
As we have seen in the previous chapter, there are cardinals whose existence cannot be proved in ZFC,
provided this theory is consistent. Inaccessible cardinals belong to this class.
5.4. LANGUAGES FOR SPEAKING OF STRUCTURES 121

The primitive symbols of Lωω1 ω (R) are the following ones:5

(i) Sentential connectives: ¬, ^, _, !, , and .
V W

(ii) Quantifiers: 8 and 9

(iii) For each type t, a family of variables of type t whose cardinal is ω.
(iv) Primitive relations: for any type t, a collection of constants of that
type (possibly some of them may be empty). The collection of these
constants forms the set R.
(v) Parentheses: left and right parentheses (‘(’ and ‘)’), and comma (‘,’).
(vi) Equality: =t of type t = ht1 , t2 i, with t1 and t2 of the same type.

Variables and constants of type t are terms of that type. If T is a term

of type ht1 , . . . , tn i and T 1 , . . . , T n are terms of types t1 , . . . , tn respectively,
then T (T 1 , . . . , T n ) is an atomic formula. If T 1 and T 2 are terms of the same
type t, then T 1 =ht1 ,t2 i T 2 is an atomic formula. We shall write T 1 = T 2 for
this last formula, leaving the type of the identity relation implicit. If α, β,
αi are formulas (i = 1, . . . ), then ¬α, α ^ β, α _ β, α ! β, αi , and αi
V W
are formulas. Then, we are able to write formulas with denumerably many
conjunctions and disjunctions. Furthermore, if X is a variable of type t,
then 8Xα and 9Xα are also formulas (and only finite blocks of quantifiers
are allowed). These are the only formulas of the language. The concepts
of free and bound variables and other syntactic concepts can be introduced
as usual.
Now let E = hD, rι i be a structure, where Rι 2 R, that is, the primitive
relations of the structure are chosen among the constants of our language
Lωω1 ω (R). Then, Lωω1 ω (R) can be taken as a language for E = hD, rι i, pro-
vided that κD = ω (recall that κD is the cardinal associated to E). Still
5
Of course we could use the above schema of free algebras to characterize this language, but this would
demand a lot of artificiality and will not conduce to nothing really relevant. The important thing is to
acknowledge that the languages we will consider can be treated as free algebras.
122 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

working within (say) ZFC, we can define an interpretation of Lωω1 ω (R) in

E = hD, rι i in an obvious way, so as what we mean by a sentence S of
Lωω1 ω (R) (a formula without free variables) being true in such a structure in
the Tarskian sense, that is,
E |= S . (5.4)

In the same vein, we can define the notion of validity. A sentence S is

valid, and we write
|= S ,
if E |= S for every structure E.
Important to emphasize that we are describing the language Lωω1 ω (R)
using the resources of some set theory such as ZFC. This way, we can
speak of denumerable many variables, for instance, in a precise way. In
this sense, any symbol of Lωωκ (R), as we have remarked already, can be
seen as a name for a set. Thus, ‘(’ (the left parenthesis), for example,
names a set, so as all other symbols of the language.

5.4.1 The language of a structure

In this section, we relate both proposals made above: the theory of struc-
tures and the language of types seen as a free algebra. An interesting dis-
cussion in the philosophy of mathematics and foundations is whether there
is a most natural language to be used associated with a given structure.
According to our approach, associated with every structure E there will be
a language which is the one in which we will talk about the elements of the
structure, having as constant symbols exactly one symbol for each relation
occurring in the structure, with both the symbol and the corresponding re-
lation being of the same type. As we said before, given a structure E, to
build one language for this structure we consider Ord(E) and restrict our-
selves to terms of this order or less. Also, the set X of generators of the
free algebra will be restricted accordingly, that is, in building the language
5.4. LANGUAGES FOR SPEAKING OF STRUCTURES 123

as a free algebra, the generators will be the atomic formulas built from the
symbols available to us.
As a consequence of this discussion, the languages in which we can
treat more adequately the elements of a structure are the languages £ such
that Ord(E)  Ord(£).6 So, for example, in first-order structures, we must
use at least first-order languages. To consider a simple example, let’s take
group theory, which deals with groups (first-order structures). These struc-
tures, according to the approach adopted in this work, are most naturally
treated by second-order languages, for we must talk about subgroups and
quantify over subsets of the domain. That does not mean that it is im-
possible or not fruitful to use first-order languages; in fact, the work is
commonly done in first-order language, as when we use set theory to de-
velop group theory, or other mathematical theories such as well-ordered
structures or Dedekind-complete fields, theories which are not first-order
(see discussion in [Kun.09, pp. 89, 90]).
There is a heated debate on this topic in the philosophy and founda-
tions of mathematics. Should we restrict ourselves to first-order languages
or should we adopt higher-order languages? We shall not enter into the
dispute here (for a defense of second-order logic in these contexts, see,
for example, Shapiro in [Sha.91]). The general approach followed in our
work suggests that a pluralist view can be fruitfully pursued: it is fruit-
ful to explore higher-order languages as well as first-order languages; one
must recognize that some kinds of structures are more naturally dealt with
higher-order languages. In fact, the reader may even wonder whether the
problem of a ‘better’ language for certain mathematical theories is a legit-
imate one (for agnosticism about this problem, see Hodges [Hod.01, pp.
71-73]).
Having a language with which we can talk about the elements of E, it is
now simple to define for this language the notions of the structure modeling
a sentence α of the language, that is, E |= α, as well as other semantical
6
We are following here also some suggestions made by Newton da Costa through personal communica-
tions in seminars.
124 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

notions, but we will not do that here.

So, taking the terminology introduced before this sub-section, let E =
hD, Rι i be a structure, while rng(Rι ) denote the range of Rι . Remember
that Rι stands for a sequence of relations of the scale ε(D), that is, it is
a mapping from a finite ordinal into a collection of relations in the scale.
Thus rng(Rι ) stands for just the set of these relations. So, Lωωω (rng(Rι ))
(= Lω (rng(Rι ))) the basic language of the structure (it is not the only one,
for other stronger languages encompassing it could be used instead). In
this case, we can interpret a sentence containing constants in rng(Rι ) in
E = hD, Rι i, and to define the notion of truth for sentences of this language
according to this structure in an obvious way.

Digression — For certain applications in science, sometimes it is better

to consider partial relations as the primitive relations of a certain structure.
A relation (say, a binary one) R on a set A is partial if there are situations
where we cannot assert neither that aRb not that ¬(aRb) (see [CosFre.03]
for all the philosophical discussion on this topic). In this case, the notion
of truth is changed to partial truth, a concept that generalizes Tarski’s ap-
proach and seems to be more adequate for empirical sciences. But we shall
not touch this point here (but see da [CosFre.03]).

5.5 Definability and expressive elements

Now we wish to understand when an object of a scale ε(D) is definable in
a structure E = hD, rι i by a formula of Lω (rng(rι )) so as when an element
of the scale is expressible in the structure with respect to a sequence of
objects of the scale.

Definition 5.5.1 (Definability of a relation) Let R be a relation of type

t = ht1 , . . . , tn i and E = hD, rι i a structure. We say that R is definable
in E if there exists a formula F(x1 , x2 , . . . , xn ) of Lω (rng(rι )) whose only
5.5. DEFINABILITY AND EXPRESSIVE ELEMENTS 125

free variables are x1 , . . . , xn of types t1 , . . . , tn respectively, such that in

Lω (rng(rι )) [{R}, the formula

8x1 . . . 8xn (R(x1 , . . . , xn ) $ F(x1 , x2 , . . . , xn ))

is true in ε(D).

For instance, for each type t we can define an identity relation =t as

follows. Let Z be a variable of type hti, then we can easily see that for
suitable structures and scales, the following is true:

9!It 8x8y(It (x, y) $ 8Z(Z(x) $ Z(y))).

We may call It the identity of type t, and write x =t y for It (x, y), what
intuitively means that identity is defined by Leibniz Law, as usual: just
re-write the above definition as follows:

x =t y := 8Z(Z(x) $ Z(y)).

Usually, we suppress the index t and write just x = y, leaving the type
implicit (= is of type ht, ti, while x and y are both of type t). This kind of
definability, which involves structures and scales is called semantic defin-
ability, and goes back to Tarski.
Here is another example. Suppose the language L2 of ‘pure’ set theory
ZF. As is well known, this language has 2 as its only non-logical constant.
If we attempt to define the subset relation ✓, we can do it in the extended
language L2 [{✓} by showing that the formula

8x8y(x ✓ y $ 8z(z 2 x ! z 2 y))

is true in any structure built in ZF. Below we shall give some further ex-
amples.
Another important case is the next one, also involving a semantic defin-
ability.
126 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

Definition 5.5.2 (Definability of an object) Let us take E = hD, rι i, ε(D),

and Lω (rng(rι )) as above. Given an object a 2 ε(D) of type t, we say that
it is Lω (rng(rι ))-definable or definable in the strict sense in E = hD, rι i if
there is a formula F(x) in the only free variable x of type t such that

E |= 8x(x =t a $ F(x)). (5.5)

The case of the well-order on the reals, mentioned above, shows that,
taking into account the last definition, the least element of (0,1), that cannot
be definable by a formula. Let us consider a ‘positive’ example. Let N =
hω, +, ·, s, 0i be an of order–1 structure for first-order arithmetics. In order
to define a natural number (any one) we need just a finitary language, say
Lωωω (R) with R = {+, ·, 0, s}. Then, it is easy to see that (in an obvious
abbreviated notation)

N |= 8x(x = n $ x = ss . . . s(0)),

If we consider a suitable infinitary language, say Lω1 ω (where we can admit

denumerable infinite conjunctions and disjunctions), we can insert inside
the parentheses the formula we abbreviate by

x 2 ω $ x = 0 _ x = 1 _ ..., (5.6)

which permits us to define not a particular natural number, but the notion of
‘being a natural number’. An important remark: the expression (5.6) is not
a formula strictly speaking (for the dots do not make part of the language),
but abbreviates a formula of Lω1 ω .
An illustrative case is the following one. We know that within ZFC
(supposed consistent), the set R of the reals is not denumerable. This
means that we cannot find a mapping (a set in ZFC) that maps the reals
onto the natural numbers. Thus, using standard denumerable languages,
we do not have sufficient names for the reals — for instance, we can use
each of the reals to name itself. But if we use a suitable infinitary language
Lµκ (for suitable ordinals µ and κ) we can find a name for each real, so
5.5. DEFINABILITY AND EXPRESSIVE ELEMENTS 127

we can define all of them by the condition given in definition (5.5.2). This
shows that definability and other related concepts depend on the employed
language.
Here is another interesting case related to the above definitions. Using
the axiom of choice we can show that every set is well ordered (by the
way, this statement is equivalent to the axiom of choice). For instance, the
set ω of natural numbers is well ordered by the usual less-than relation .
We can define the usual order  as follows:7 a  b := 9c(b = a + c).
The usual order relation, however, does not well order the set Z of the
whole integers, for the subset {. . . , −2, −1, 0}, for instance, has not a least
element. But if we order Z by writing it as {0, −1, 1, −2, 2, . . .}, then it is
well-ordered by this order, termed 1 , which can be defined by a 1 b :=
(|a| < |b|) _ (|a| = |b| ^ a  b), where  is the usual order relation and < is
defined as a < b := a  b ^ a , b.
Obviously the ‘usual’ relation (defined on R) does not well order the set
of reals (for instance, an open set (a, b), with a < b, has not a least element).
But, can we find a well order on such a set? According to the axiom
of choice, we can assume that this order does exist, since its existence
is consistent with ZFC (supposed consistent). The problem is that it can
be proven that the ZFC axioms (plus the so-called generalized continuum
hypothesis) are not sufficient to show that this order can be definable by
a formula of its language.8 In the same vein, we cannot define the least
element of a certain subset of reals (say, the open interval (0, 1)) in the
sense of definition (5.5.2). All of this shows that the notions of definability
and expressibility, among others, depend on the language and on the theory
we are assuming.

7
Of course this definition can be conformed to definition 5.5.1.
8
The interested reader can check theorem 4.11 of S. Feferman’s paper and the remark at p.342, just after
the proof of the theorem; see http://matwbn.icm.edu.pl/ksiazki/fm/fm56/fm56129.pdf.
128 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

5.6 On the new symbols

In our formalization of ZFC, we have used individual constants to facilitate
our mentioning of certain particular sets, for in doing that we would have
a denumerable set of ‘names’ to them. This is a matter of choice. Most
authors don’t use individual constants, but just individual variables. Since
it is supposed that this move doesn’t change the set of theorems, we regard
the resulting theories as being the same, yet their languages differ. In this
section, we shall make some few remarks we regard are of philosophical
importance which, although they are well known by the logician, may be
not so well understood by the general philosopher interested in founda-
tional issues. So, let us suppose for a moment that our language L2 has no
individual constants. To avoid any confusion, we shall term it L2− .
Due to our new convention, the only non-logical symbol of L2− is 2.
So, how can we refer to a particular relation R? (the same holds for a
particular object whatever, such as a structure named A for instance). This
is a common practice in the mathematical discourse. Really, in geometry
we usually say “Let A be a point in a line r”, making use of individual
letters for naming particular entities, the point and the line. How can we
explain this move of naming objects with constants that do not appear as
primitive concepts of the employed languages? There are two answers to
this challenge.
The first is that we simply extend the language L2− with additional con-
stants to name the objects we intend to make reference to. Thus, we can ex-
tend L2− with symbols of three kinds: (1) additional individual constants,
(2) new predicate symbols, and (3) new operation symbols. In whatever
situation, we must make sure that the so-called Leśniewski’s criteria are
being respected [Sups.57, Chap.8], namely, the Criterion of Eliminability
and the Criterion of Non-Criativity. The first says, in short, that the new
symbols can be eliminated. That is, the formula S introducing the new
symbol must be so that whenever a formula S 1 with the new symbol occurs,
there is another formula S 2 without this symbol such that S ! (S 1 $ S 2 )
5.6. ON THE NEW SYMBOLS 129

is a theorem of the preceding theory (without the new symbol). The second
criterion says that there is no formula T in which the new symbol does not
occur such that S ! T is derivable from the preceding theory, but T is not
so derivable. In other words, no new theorem previously unproved, and
stated in terms of the primitive symbols and already defined symbols only
can be derived. In our case, we can add the desired symbols, say ‘R0 for
the relation in the above definition of definability of a relation (see defini-
tion 5.5.1), once we grant that the Leśniewski’s conditions hold (which of
course we suppose here).
The other alternative is to work with L2− proper, and regard all other
symbols as metalinguistic abbreviations. This is ‘more economic’, and we
usually do it for instance when defining (in L2− ) the concept of subset,
posing for instance that A ✓ B := 8x(x 2 A ! x 2 B). The new symbol ✓
does not make part of the language L2− , but belongs to its metalanguage,
and the expression A ✓ B simply abbreviates a sentence of L2− , namely,
8x(x 2 A ! x 2 B). We can understand this move as enabling us to
use an auxiliary constant (say, ‘R’), provided that the object which it will
name exists (the proof of its existence is called theorem of legitimation by
Bourkaki [Bou.68, p.32].9 In the example, given two sets A and B, we
realize that all elements of A are also elements of B, hence we are justified
to write A ✓ B for expressing that (as a metalinguistic abbreviation). In
logic, we usually express that by the so called method of the auxiliary
constant, which may be formulated as follows. Let c be a constant that
does not appear in the formulae A or B. Assume that we have proven that
9xA (the theorem of legitimation). If we have also proven that A[cx ] ` B,
where A[cx ] stands for the formula obtained by the substitution of c in any
free occurrence of x, then ` B as well.
But we need some care even here. Suppose we wish to refer to real num-
bers. We cannot name all of them in the standard denumerable language
L2− . But in general we can name a particular real number, say by calling it
9
This is essentially what Mendelson calls ‘Rule C’; see [Men.97, p.81].
130 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

zero. But there are real numbers that cannot be named even this way, as we
have seen above when we have discussed the well-order of the reals. This
poses an interesting question regarding empirical sciences. In constructing
a mathematical model of a physical theory, we suppose we represent phys-
ical entities, say quantum objects and the properties and relations holding
among them, in the mathematical framework we have chosen, say ZFC set
theory. How can we ensure that this really makes sense? For instance, in
some common interpretations of quantum mechanics (the Copenhagen in-
terpretation), we really cannot (with any sense) make reference to quantum
entities out of measurement. Out of measurement, quantum entities are no
more than sets of potentialities or possible outcomes of measurement, to
use Paul Davies’ words in his Introduction to Heisenberg’s book [Hei.89,
p.8]. Let us give an example we shall discuss also below. Suppose we are
considering the two electrons of an Helium atom in the fundamental state.
The anti-symmetric wave function of the joint system can be written as
1
|ψ12 i = p (|ψ1 i|ψ2 i − |ψ2 i|ψ1 i). (5.7)
2
where |ψi i (i = 1, 2) are the wave functions of the individual electrons.
Notice that we need to label them by ‘1’ and ‘2’ for our languages are ob-
jectual —we speak of objects.10 But we need to make this labeling to be
not compelled with individuation, so we use (in this case) anti-symmetric
functions, for the if |ψ21 i stands for he wave-function of the system after
a permutation of the electrons, then ||ψ12 i|2 = ||ψ21 i|2 , that is, the relevant
probabilities are the same. For now, what is relevant is that this function
cannot be factored, giving a particular description of the electrons sep-
arately. Only after a measurement, say of the component of their spin
in a given direction, the wave function collapses either in |ψ1 i|ψ2 i or in
|ψ2 i|ψ1 i, then indicating, say, that electron 1 has spin up in the chosen di-
rection, while electron 2 has spin down in the same direction (or the other
way around). But, before the measurement, nothing can be said of them in
10
Toraldo di Francia says that objectuation is a primitive act of our mind; see [Tor.81, p.222].
5.7. DA COSTA AND CHUAQUI — SUPPES PREDICATES 131

isolation. This implies that when we say, for instance, that there are two
electrons, one here and another there, we are already supposing a measure-
ment, thus begging the question concerning their individuation.
Situations such as this one are puzzling if we consider the underlying
mathematics as the classical one.11 For suppose attempt to define the elec-
tron of an Helium atom that has spin up in a given direction. According to
definition (5.5.2), we need to find a formula F(x) so that, if we denote that
electron by a, we can prove that the following formula
8x(x = a $ F(x)) (5.8)
is true in an adequate ‘quantum structure’. But, what would be taken to be
F(x)?
Really, according to standard set theory (ZF), when we say that there
is an electron here and another there, they are already distinct entities,
and we are presupposing a kind of realism concerning these entities. In
other words, in supposing two entities within ZF, we are really begging the
question concerning their individuation.12
These remarks, we think, point to an interesting philosophical prob-
lem of studying the definability of physical ‘objects’ and relations, but this
point will be not discussed here too. Next, let us have a closer look at
Suppes’ and da Costa and Chuaqui’s approaches.

5.7 da Costa and Chuaqui — Suppes predicates

In this section we will return to the general concept of structure having as
domain a family of sets. Given a structure and some language adequate
for this structure, we now discuss how to formulate a Suppes’ predicate for
11
We shall be back to this point in the end of this chapter.
12
This is, in our view, the main difficulty involved in attempts made by Muller and Saunders [MulSau.08]
and by Muller and Seevinck [MulSee.09] to discern quantum entities; by assuming ZF as their mathematical
framework, they are already assuming that the represented entities are either identical (that is, they are the
very same entity) or that they are distinguishable, and this is not a characteristic of quantum objects proper,
but of the mathematics they have employed.
132 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

that structure using that language, according to the directions suggested by

N. da Costa and R. Chuaqui [CosChu.88]. First of all, we recall the defini-
tion of the similarity type of a structure E, which is a family of types that
determines the kinds of the relations present in the structure. According to
this definition, two structures have the same similarity type if the types of
their relations form the same family and they have the same number of sets
in their domain.
Now, given structures E = hD1 , . . . , Dn , Rι i and G = hE1 , . . . , En , Lι i
of the same similarity type, we consider how to extend a given function
f : Dk 7! Ek for k < n to a function mapping from ε({Dn }) to ε({En }).

Definition 5.7.1 Given the function f as described above, we define:

(i) For the objects of type t, with 0  t < n, let f (Dk ) = { f (x) : x 2 Dk )}
(ii) For t 2 T such that t = ha0 , . . . , an−1 i, and R the set of objects of type
t, we have f (R) = P( f (Da1 ) ⇥ f (Da1 ) ⇥ . . . ⇥ f (Dan ))

This function maps objects of the type a in ε(Dn ) to objects of type

a in ε(En ). The interesting case occurs when the following definition is
verified:

Definition 5.7.2 Given structures E = hD1 , . . . Dn , Rι i and G = hE1 , . . . , En ,

Lι i of the same similarity type s, and f a bijection from Dk to Ek with
0  k < n, we say that the family f 0 = f s is an isomorphism between E and
G when ft (Rt ) = Lt , where Rt and Lt means that R and L have type t.

Definition 5.7.3 A sentence Φ of the language appropriate for the struc-

ture E is called transportable if for any structure G isomorphic to E, that
is,
E |= Φ , G |= Φ.

Intuitively speaking, a transportable relation does not depend upon spe-

cific properties of the involved sets, but refers only to the way they enter in
5.7. DA COSTA AND CHUAQUI — SUPPES PREDICATES 133

the relation, something that is given by the axioms. So, a transportable

relation cannot involve specific sets, but must speak of sets arbitrarily.
So, when in a relation it appears something like the empty set ;, this is
just a way of speaking, for this symbol can be substituted by the sentence
9y8x(x < y). We shall be back to this remark later.

Definition 5.7.4 A Suppes’ predicate is a formula P(E) of set theory which

says that E is a structure of similarity type s satisfying Γ, a set of trans-
portable sentences Φ of the language adequate for E.

When P(E), that is, when E satisfies P, we say that E is a P-structure.

According to da Costa and Chuaqui ([CosChu.88, p.104]), this definition
captures the sense in which we can say that a theory is a class of models,
precisely, the class of models that are P-structures for some adequate P.
We now consider some examples of Suppes’ predicates for some theo-
ries.

5.7.1 A Suppes’ Predicate for Group Theory

Let G be a set; as defined above, we introduce the function tG , or simply t,

whose domain is the set T of types, to create the scale ε(G). We suppose
that the type of the elements of G is 0. Then, we choose: (1) a relation ◦ of
type h0, 0, 0i, that is, ◦ 2 P(G ⇥ G ⇥ G), (2) a relation − of the type h0, 0i,
that is, − 2 P(G ⇥ G), and (3) an element e of type 0, that is, e 2 G. As
one can check from the definitions, the order of each of the relations is 1
and the order of e is 0. Remembering that a n-ary function is a n + 1-ary
relation, we have that the usual composition operation becomes a ternary
relation, and the opposite relation becomes a binary relation.
The structure of groups is G = hG, ◦, −, ei and the order of this structure
is the greatest order of its relations, so ord(G) is 1, that is, G is a order-
1 structure. The theory is not done yet, for we need to write down the
postulates in order to give the set-theoretical predicate.
134 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

As defined above, the language for G is a second-order language (note

that the order of the language does not necessarily coincide with the order
of the structure). The set T of terms is formed by a set of variables of type
t = 0 or t = 1 and the set {◦h0,0,0i , −h0,0i , e0 } of constant symbols. So, the set
X of free generators is X = {T t (t0 , . . . , tn−1 ) : T t 2 T t ^ t = ha0 , . . . , an−1 i 2
T} (tk 2 T k ). By the above theorem, there is a free algebra upon the set of
free generators X, which it is the language for the structure G.
With the language so developed we can write the usual axioms for group
theory, let’s call them A1, A2 and A3, respectively as follows:
A1 8x8y8z(((x ◦ y) ◦ z) = (x ◦ (y ◦ z)))
A2 8x9y(x − y = e)
A3 8x(x ◦ e = x)
Then, a Suppes predicate for group theory can be written as follows:

G(X) () 9G9 ◦ 9 − 9e(X = hG, ◦, −, ei ^ A1 ^ A2 ^ A3)

The structures that satisfy this predicate are the models of the theory,
namely, the groups. For more complex theories, of course we do not
present the axiomatics of a theory this way but, as in standard mathematics,
by describing the structure and listing the axioms. In the next example, we
shall proceed this way.

5.7.2 A Suppes’ Predicate for classical particle mechanics

First, we need to present some mathematical ‘step’ structures: the first
one is the complete ordered field of real numbers R = hR, +, ·, 0, 1, <i.
There is just one base set, the set R of real numbers. The objects of this
set are of type 0. The operations are +, ·, 0, 1, <, which are of types
h0, 0, 0i, h0, 0, 0i, 0, 0 and h0, 0i respectively (one must not confuse the
symbol 0 of types with the element 0 of the field). As usual, these are
5.7. DA COSTA AND CHUAQUI — SUPPES PREDICATES 135

the operations of addition, multiplication, the identity element of addition,

the identity element of multiplication and the relation of less than between
real numbers. The language of the field of real numbers according to our
approach is a second-order language and the axioms for the field are well
known and we can easily see that they are transportable.
The next structure is the vector space E = hV, R, +, ·, Oi over the field
of real numbers R. In this case there are two base sets, the set V of vectors
and the set R of real numbers. The objects of V are of type 1, while the real
numbers are of type 0. Besides the field’s operation we have +, ·, O which
are respectively of types h1, 1, 1i, h0, 1, 1i, 1. A Euclidean vector space
is a vector space with addition of inner product and a vector product; we
denote by hx|yi and by [x, y] the inner product and the vectorial product of
vectors x and y respectively. The first product gives a real number, and the
second one another vector, which are objects of types h1, 1, 0i and h1, 1, 1i
respectively. The order of the language of vector spaces and that of the
Euclidean vector space are 2 according to our approach, that is, a second-
order language. The axioms for these structures are well known and clearly
transportable.
Next, we present the real affine space which is a vector real space with
an addition domain A, whose elements are of type 2 and are called points.
The new operation is the difference of points which, for p, q 2 A, is a
vector denoted by q− p, whose type is h2, 2, 1i, that is, the difference of two
points gives us a vector. The new axiom for this concept is the statement
that difference of points obeys the law of addition of points, which says
that for p, q, r 2 A, q − p + r − q = r − p. If the vector space is Euclidean,
the affine space is a Euclidean space. In Euclidean space we can define the
distance
p between points p, q 2 A in the following way: d(p, q) := ||q−p|| :=
hq − p|p − qi.
Now, we present a Galilean space-time system. We add to the four di-
mensional affine space presented earlier a new domain V1 which is a sub-
set of V, and operation t from A into R of type h2, 0i and relations of type
136 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

h3, 3, 1i and h3, 3, 3i denoted by h·|·i and [·, ·] respectively, and t represents
the measure of time. The two axioms following must be satisfied:
(a) V1 is a three dimensional vector subspace of V and h·|·i and [·, ·] are
its scalar product and vectorial product, respectively;
(b) t is a function from A to R such that for each P 2 A the set {Q :
t(Q) = t(P)} is a three dimensional euclidian space with vector space
V1 . The affine space for t(P) = r is denoted by A(r).
For a classical mechanical system we need to add a new universe P, the
set of particles13 and the set N of natural numbers to index the external
forces. So the family of universes can be given by the sequence R, V, A,
V1 , P and N. The operations on these sets are those necessary to make R,
V, A, V1 a Galilean space-time system plus the following new relations:
(i) A function a of type h0, 2i which gives the origin of a systemof coor-
dinates in each instant of time;
(ii) A function s of type h4, 0, 2i for the position of a particle in each
instant of time. We write s p (t) for this function;
(iii) A mass function m of type h4, 0i;
(iv) A force function f of type h4, 4, 0, 1i which represents the internal
forces;
(v) A force function g of type h4, 0, 5, 1i which represents the external
forces.
For the specific axioms of mechanics we need notions of mathemati-
cal analysis such as derivatives and convergence of series. The field of
real numbers must be completed with the corresponding operations of dif-
ferentiation, integration and addition of series. Since differentiation, for
13
Important to realize for what follows that in this case P can be taken as a set of a set theory like ZFC.
The quantum mechanical case will me mention later, and in this situation sometimes the domain of entities
would be not a set, as we shall see.
5.7. DA COSTA AND CHUAQUI — SUPPES PREDICATES 137

example, takes functions of real numbers to functions of real numbers, the

order of this operation will be 2. So, the language needed to talk about this
is structure is at least third-order language.
The kinematical axioms are:

(1) The range of t is an interval I of real numbers;

(2) P is a finite and non-empty set;

(3) a is a function from I to A such that for each i 2 I a(i) 2 A(i);

(4) s is a function from P ⇥ I into A such that for each p 2 P and i 2 I

we have that s p (i) 2 A(t);

(5) m is a function from P into R;

(6) f is a function from P ⇥ P ⇥ I into V1;

(7) g is a function from P ⇥ I ⇥ N into V1 ;

(8) For every p 2 P and i 2 I the vector function s p (i) − a(i) is twice
differentiable at i.

Dynamical Axioms:

(1) For p 2 P m(p) is a positive real number;

(2) For p, q 2 P and i 2 I f(p, q, i) = −(q, p, i);

(3) For p, q 2 P and i 2 I [s(p, i) − s(q, i), f(p, q, i) − f(q, p, i)] = O;

(4) For p 2 P and i 2 I the series Σn (g(p, i, n) is absolutely convergent;

(5) For p 2 P and i 2 I m(p)D2 (s p (i)) = Σq2P f(p, q, i) + Σn (g(p, i, n),

where D2 is the second derivative with respect to i.
138 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

These formulas are transportable, in the sense defined previously.14 The

motivations for these formulas can be found in the works of Suppes cited
in the bibliography. Other examples could (and perhaps should) be pur-
sued, but these ones are enough to emphasize that in da Costa-Chuaqui’s
approach, we need to consider the language of the structure, and that we
need also to axiomatize all the step structures necessary for the main the-
ory. It leads to an Herculean effort of course, practically similar to what
was required by the approached suggested by the Received View. Now we
shall turn to Suppes’ own way of axiomatizing theories, and we shall see
that it has some advantages by its simplicity and power.

5.8 The approach of Suppes

The approach developed by Patrick Suppes in many of his works contrasts

with the previous work by da Costa and Chuaqui. Suppes’ technique of
axiomatizing a class of structures inside set theory properly is described
for instance in his [Sups.57]. He doesn’t use a formal language specific to
theory in study. In his [Sups.67, pp.60-62] these both kinds of approaches
are distinguished as intrinsic characterization of a theory (when the axioms
are formulated in a first-order language) and the extrinsic characterization
of a theory (when the theory is axiomatized using the resources of informal
set theory). Intrinsically, a theory is characterized when no appeal to any of
its models is required, that is, by appealing just to that which is intrinsic to
each model and, consequently, captured by a formal axiomatization. From
an extrinsic point of view, a theory is characterized by describing directly
the class of intended models, as structures build in set theory.
Of course, in the context of the later mentioned work, Suppes distin-
guishes between presenting a class of structures through set-theoretical
predicates, in contrast to characterizing it as a formal elementary theory.
14
In fact, even in the case of the set R of real numbers, the general situation can be described taking any
completed ordered field, so the axiomatics does not speak of any particular set.
5.8. THE APPROACH OF SUPPES 139

He goes on and even calls our attention to the fact that it is not always clear
whether a class of structures characterized extrinsically has an elementary
axiomatization.15 To provide an illustration, consider the class of well-
orderings, that is, of structures W = hD, Ri where R is a well-ordering on
D. As is well known, there is no elementary axiomatization of this class.16
However, we can give an extrinsic characterization of this class of struc-
tures (and, by leaving the restriction to elementary languages behind the
da Costa-Chuaqui approach presented above can also do it).
The distinction between intrinsic and extrinsic characterization and the
benefits of exploring a scientific theory by looking at its models, and not
only at its syntactical characterization is a feature of Suppes’ approach to
the so-called Semantic Approach. In particular, his emphasis on models
is closely related to his own approach to the relation between theory and
experiment, which does not seem to be correctly described only in syn-
tactical terms (as we have already discussed in our chapter 1). Anyway,
Suppes still need a way to extrinsically characterize a class of models, and
he does that by axiomatizing the theories inside set theory.
Suppes says he assumes intuitive set-theory, but we can continue to sup-
pose that our discussion is conduced inside ZFC at an informal level, that
is, without justifying every step of the required constructions. The first im-
mediate advantage of starting with ZFC (or with some suitable set theory)
is that one can employ all of the mathematics needed without explicitly
having to axiomatize or listing its suppositions, for ‘everything’ we need is
by hypothesis already done inside ZFC. Also, one leaves metamathemat-
ical considerations of usual axiomatizations, which proceed through the
elaboration of formal systems, and works in the known mathematical en-
15
Note that in the da Costa–Chuaqui approach we just presented, no restriction to elementary languages
is required.
16
Perhaps it would be interesting to review some concepts from logic here. Let L be a first-order language
and T a theory in L. A model M of T is an L-structure that satisfies the postulates of T . Let us call M(T )
the class (in general not a set) of all l-structures that are models of T . Let K be a class of structures. We
say that K is elementary, or finitely axiomatizable, if there is finite (a finite number of axioms) T such that
K = M(T ).
140 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

vironment, simply by making use of the mathematics at disposal. This last

point was one of the attractions of the semantic approach over the Received
View (see a full discussion in [Sup.77], but look also at [Lut.12] for a more
realistic reading of the demands of the Received View), which proceeded
through formal systems.
In general terms, a class of structures is axiomatized in set theory by
providing for a list of axioms that a theory should satisfy. Those axioms
are stated directly inside set theory, using the vocabulary of set theory and
further symbols that may be introduced by definition and that are specific
to the theory. Let us see how we can define a set-theoretical predicate
for groups. Important to remark that now we do not need to described
all the step theories to be used, since we may assume that all of them
can be given in the set theory being presupposed. Suppes gives various
alternative formulations of this theory (mainly due to his concerns with the
theory of definition and the search for an adequate and perspicuous form
which conforms some rigorous standards of definitions). To concentrate
ourselves in one of the proposed ways of characterizing the theory, we
can define a group through the following set-theoretical predicate, given
informally first: a group is an algebra G = hG, ◦i where G , ; and ◦ is
a binary operation on G, and . . ., the dots being completed by the usual
group axioms, written in the language of ZFC:
A1 (8x, y, z 2 G)(x ◦ y) ◦ z = x ◦ (y ◦ z)
A2 There is an e 2 G such that for all x 2 G, x ◦ e = e ◦ x = x
A3 (8x 2 G)(9x0 2 G)(x ◦ x0 = x0 ◦ x = e)
Now, this is an example of a set theoretical predicate for group theory
and, as Suppes claims, it is entirely in conformity with the usual mathe-
matical practice.
To be more precise, and using some terminology introduced before, we
could specify the nature of the elements of the structure as well. So, a
group G is an ordered pair hG, ◦i with ◦ 2 t(hi, i, ii) satisfying A1, A2 and
5.8. THE APPROACH OF SUPPES 141

A3 stated before. In symbols, the predicate G is (or maybe) the following

one:
P(G) $ 9G9◦(G = hG, ◦i^G , ;^◦ 2 P(G⇥G⇥G)^A1^A2^A3). (5.9)
Suppes calls a group any structure G = hG, ◦i that satisfies this predi-
cate. Nothing specific is said about the satisfaction relation, or about the
language in which the axioms are formulated. But, as we mentioned be-
fore, it is the enlarged language of the set theory (with new symbols such
as G, ◦, etc.) which is being employed, and this is in conformity with the
standard use in mathematics. How does one show that some structure is
in the class characterized by the predicate so defined? For example, how
does one show that hZ, +i is a group? One simply derives, as theorems of
ZFC, that the elements of Z, along with the operation +, have the proper-
ties required for something to be a group. Then, hZ, +i is said to satisfy the
axioms for group theory and, so, the structure is a model of group theory,
or it is a group. We shall discuss precisely the relation of the axioms with
the set theoretical structures later.
Despite its great simplicity, Suppes approach is very powerful indeed,
and here we point to another of its advantages. While the standard first-
order group axioms (to keep with our example) are enough to characterize
all groups as being models of the axiomatics, in using set-theoretical predi-
cates we can axiomatize certain classes of models only. A typical example
would be to axiomatize all groups except a certain specific case; for sim-
plicity, let us suppose that we wish to leave out the additive group of the
integers, hZ, +i. We can do it by adding to the set-theoretical predicate
given above a fourth condition A4 (another axiom), namely, a clause re-
stricting the items that satisfy the predicate, say by requiring that those
structures that satisfy it are groups but different from hZ, +i (this can be
easily expressed for instance by requiring in the set-theoretical predicate
that G , Z). This is most relevant for the semantic approach, which some-
times is said to be able to deal only with classes of intended models of
theories [Sup.00, p.104].
142 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

It is worth noticing that an important difference between Suppes’ and

da Costa-Chuaqui’s approach is that in the former the formula which de-
fined the predicate does not need to be transportable, something considered
crucial in the later, for the authors follow Bourbaki. The explicit reference
to a specific set (in our sample case, Z) shows that the predicate for the
class of structures of groups except the additive group of the integers is not
transportable.

For another example, one can take the Peano structures P = hω, 0, σi,
with the usual Peano axioms in ZFC. Now, the structure 2P = h2ω, 0, 2σi,
where 2ω is the set of even natural numbers and 2σ is the addition of 2,
is also a model of these axioms. One can construct a Suppes predicate
with the Peano postulates plus the requirement that the domain is different
from 2ω, for example, and then this structure is not in the class of models
of the set-theoretical postulate for arithmetics. This is possible because
in the language of ZFC we can make reference to a particular element
we leave outside the class of structures determined by the predicate. This
fact is precisely what distinguishes the Suppes approach from da Costa-
Chuaqui, and even from Bourbaki, let us insist once more. In the last
two cases, the predicate (or the axioms in the case of Bourbaki) has two
main parts: a typification and the axioms themselves. The typification is
a formula, or a conjunction of formulas that specify what are the particu-
lar relations/operations/distinguished elements we are considering (for the
formal definition, see [Bou.68, p.261]). Let us exemplify: in the case of
groups, the typification is ◦ 2 P(G ⇥ G ⇥ G). In the case of vector spaces,
we have + 2 P(V ⇥ V ⇥ V) ^ · 2 P(K ⇥ V ⇥ V). The axioms, in both
cases, must be transportable formulas; roughly speaking, this means that
the definition of the formula does not depend upon any specific property
of the construction made from the basic and auxiliary sets, but only refers
to the way in which they enter in the relation through the axioms [Bou.68,
loc.cit.]. In the Suppes approach, this restriction is not imposed, that is, the
formulas needn’t be transportable, and so we gain a wide range of extra
5.8. THE APPROACH OF SUPPES 143

possibilities.
For another example of these possibilities, let us consider a simple pred-
icate for a structure composed by a set D and an operation σ picked from
t(hi, ii) (that is, a binary operation on D) satisfying the conditions (axioms):
(a) ; 2 D; (b) σ is injective; (c) ; does not belong to the image of σ; (d)
D is the least set satisfying these conditions. In this case, we are mention-
ing a particular item, ;, and saying in the axioms that this chosen element
must belong to the domain of the structure. Since we are making reference
to a well defined object (the set ;), the formula which describe the set-
theoretical predicate for the theory is not transportable. In this particular
example, we still have that it is not possible to axiomatize M = hD, σi in
the standard way, say by fixing a specific vocabulary and making use of
first-order logic (more on this below).
The fact we can do such moves show that set-theoretical predicates dif-
fer from Suppes predicates as defined by da Costa and Chuaqui, so as
from Bourbaki’s species of structures, although some predicates can fit
the claims of transportability. The reason is that an axiomatics (say, in
Bourbaki’s style) should not make reference to specific objects such as the
empty set. In fact, Bourbaki requires that the theory’s postulates must be
transportable formulas with respect to some typification, which teach us
how to deal with the added symbols of the structure. Saying with other
words, transportable formulas must be invariant by isomorphisms, and so
they cannot make reference to specific objects [Bou.68, p.261-2]). These
difficulties can be surpassed with some modifications and adaptations of
our example, but they shall not concern us here. Anyway, in the Suppes’
style axiomatization, we have free access to all mathematics that can be
found in ZFC, without constraints on the kinds of formulas that can be
used to axiomatize a theory.
This previous discussion can be summarized in an interesting result.
Recall that, when we restrict ourselves to first-order languages, in order
to axiomatize a class of structures of order-1 we must give a set Γ of sen-
144 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

tences such that all the structures in the class are models (in the sense of
Tarski) of Γ, and all models of Γ are in the class. A necessary and sufficient
condition for a class of structures to be axiomatized by a set of sentences,
is that it must be closed by elementary equivalence and by formation of
ultraproducts (for the definitions of those notions, see [Men.97]).17
Then for example, in this kind of axiomatization, when we deal with
groups and the usual postulates for this theory written in a first-order lan-
guage, we cannot have hZ, +i out of the class of structures if h2Z, 2+i is
in this class (here, 2Z denotes the set of even integers, and 2+ addition
of 2), for both are elementarily equivalent. The same holds for models
of first-order Peano arithmetics, for the standard model P = hω, 0, si and
2P = h2ω, 0, s0 i, where 2ω is the set of even numbers and s0 is the opera-
tion of adding 2. That is, we cannot have an axiomatization in a first-order
language of one of them without having the other one as well, for they
are elementarily equivalent. But, as we have seen, using the Suppes style
axiomatization, we can define classes of models which are order-1 struc-
tures but which are not closed by elementary equivalence. That is, Suppes’
axiomatization is stronger than the usual one, since it axiomatizes more
classes of structures than it is possible to do with the usual procedure, and
for checking this we need only to consider the examples given above. So,
the examples above show that there are classes of structures of order-1
axiomatizable by a Suppes predicate but not by a set of first-order set of
sentences.
This result can be generalized to higher-order languages by following
the same procedure (just take the class of all well-orderings except a par-
ticular well-ordering, for instance hω, 2i).
For another example, compare now Suppes’ axiomatization of particle
mechanics with the one provided before by the da Costa and Chuaqui ap-
proach.

17
We thank Antonio M. N. Coelho and Newton C. A da Costa for calling our attention to these points
and for discussions on these themes.
5.8. THE APPROACH OF SUPPES 145

A system of particle mechanics is a 5-tuple18

P = hP, T, m, s, f, gi

where P is a non-empty set (the ‘particles’), T is an interval in the set of

real numbers (say expressing an interval of time), m is a function from P to
R+ so that if p 2 P, then m(p) is the mass of p, s is a function from P ⇥ T in
R3 , so that s(p, t) is a vector expressing the position of the particle p at time
t, f is a function with domain P ⇥ T ⇥ I, where I is a set of positive integers,
so that f(p, t, i) is a vector representing the forces acting on p at t, and g is
a vector representing the external forces acting on a particle at a given time
t. All these concepts are subjected to kinematical and dynamical axioms
below:

Definition 5.8.1 A Newtonian particle mechanics is a structure

P = hP, T, m, s, f, gi

satisfying the following axioms:

Kinematical axioms

P1 P is a finite non-empty set;

P2 T is an interval of real numbers;
P3 for p 2 P, s p is twice differentiable in T ;

Dynamical axioms
P4 for p 2 P, m(p) is a positive real number;
P5 for p, q 2 P, t 2 T , f(p, q, t) = −f(q, p, t) ;
P6 for p, q 2 P, t 2 T , s(p, t) ⇥ f(p, q, t) = −s(q, t) ⇥ f(q, p, t);
18
Here we take a simplified version of this structure; see Suppes [Sups.57], chap. 12. In his [Sups.02],
Suppes provides an alternative formulation.
146 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

P7 for p 2 P and t 2 T , m(p)D2 s p (t) =

P
q2P f(p, q, t) + g(p, t).

As we see, it is easier to use this kind of procedure. A possible problem

with this approach is that Suppes does not make reference to the specific
set theory to be used in the axiomatization of a theory. We are free to use
the theory we need to get the concepts we need. But there is also another
kind of situation we shall describe now, which deals not specifically with
the axiomatization of the theory, but with its interpretation, or semantics.
Let us look at quantum-mechanics to have an idea of this problem.

5.8.1 An axiomatization of non-relativistic quantum mechanics

Let us consider a way to axiomatize non-relativistic quantum mechanics by

means of a set-theoretical predicate following Suppes’ approach. We shall
be working informally, hoping that the reader understands that with a great
boring effort, all could be done in terms of a ‘legitimate’ set theoretical
predicate in a set theory like ZFC. We shall just present the relevant struc-
tures and the corresponding postulates, according to the standard mathe-
matical practice. So, we proceed as Suppes in his examples. Furthermore,
we are not claiming that our axiomatics is adequate for all cases, since we
restrict ourselves mainly to the finite dimensional case and to pure states
only (mixtures are mentioned only in brief). So, we shall be working in
informal set theory, but if necessary, we may suppose a system like ZFC.
The axiom of choice seems to be essential here, mainly in the infinite di-
mensional case, for we need to grant that the vector spaces we use (Hilbert
spaces) do have basis, and the general proof that a vector space has a basis
depends on this axiom. The novelty in our approach is that we introduce
a collection of physical objects (or systems) in the formalism, something
that is absent in the standard approaches (really, in these approaches the
reference to physical objects is made only indirectly, when we say — in-
formally — that the vector state represents a quantum physical system;
but this system never appears in the formalism). The definition is the one
5.8. THE APPROACH OF SUPPES 147

given below, where according to Suppes’ approach, we presuppose some

step theories given in set theory, like the theory of Hilbert spaces, proba-
bility, differential equations and all the involved mathematical apparatuses.

Definition 5.8.2 A non relativistic quantum mechanics structure is a tuple

of the form

Q = hS , {Hi }, {Âi j }, {Ûik }, B(R)i, with i 2 I, j 2 J, k 2 K

where:

(i) S is a collection19 whose elements are called physical objects, or

physical systems.

(ii) {Hi } is a collection of mathematical structures, namely, complex sep-

arable Hilbert spaces whose cardinality is defined in the particular ap-
plication of the theory.

(iii) {Âi j } is a collection of self-adjunct (or Hermitian) operators over a

particular Hilbert space Hi .

(iv) {Uik } is a collection of unitary operators over a particular Hilbert

space Hi

(v) B(R) is the collection of Borel sets over the set of real numbers.

The Hilbert space formalism, whose postulates we shall see in a mo-

ment, does not speak of space and time, something essential when we in-
tend to apply it to the ‘real world’. Below we comment on this important
point.

Definition 5.8.3 To each quantum system s 2 S we associate a 4-tuple of

the form
σ = hE4 , ψ(x, t), ∆, Pi.
19
Below we shall question whether this collection can be considered as a set of ZFC.
148 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

Here, E4 is the Galilean spacetime;20 each point is denoted by a 4-tuple

hx, y, z, ti where x = hx, y, zi denote the coordinates of the system and t
is a parameter representing time, ψ(x, t) is a function over E4 called the
wave function of the system, ∆ 2 B(R) is a Borelian, and P is a function
defined, for some i (determined by the physical system s), in Hi ⇥ {Âi j } ⇥
B(R) and assuming values in [0, 1], so that the value P(ψ, Â, ∆) 2 [0, 1]
is the probability that the measurement of the observable A (represented
by the self-adjunct operator Â) for the system in the state ψ(x, t) lies in the
Borelian set ∆. We can see the relationship between the state vector and the
wave function as follows. Let (x, t) denote the location operation at time t.
Then we put ψ(x, t) = h(x, t)|ψi, that is, the wave-function is described by
the coefficients of the expansion of the state vector in the orthonormal basis
of the position operator (a more precise description will be given soon).
These concepts are subjected to the following postulates:

Postulate 1 To each physical system s 2 S we associate a Hilbert space

H 2 {Hi }. Composite quantum systems are associated to complex
Hilbert spaces that are the tensor product of the Hilbert spaces for
each system, as usual.

Postulate 2 The one-dimensional subspaces of H denote the states the

system may be in. These spaces are called rays by the physicists. To
simplify the notation, usually they are represented by unitary vectors
ψ (or by |ψi in Dirac’s notation) that generate these spaces. Hence,
cψ, for c 2 C, represents the same state as ψ, so as does ψ.c (this is
a typical physicists’ abuse of notation).21 These vectors are said to
represent pure states of the system. It is also postulated that linear
combinations of pure states, that is, vectors of the form ψ = n an ψn ,
P
for an 2 C (the linear combination may comprise any finite number
20
For details, which do not interest us here, see [Pen.05, chap.17].
21
The reason for this to be an abuse of notation is easy to explain. The operation of multiplication of a
vector by a scalar (a complex number) is defined to be multiplication to the left, that is, it is an function from
C ⇥ H to H. If we want that this also represents a function from H ⇥ C to H, we need to say it explictly.
5.8. THE APPROACH OF SUPPES 149

of vectors), called superpositions by the physicist, also denote pure

states. This assumption is called the Superposition Postulate.

Postulate 3 To each observable (physical quantity that can be measured)

A we associate a self-adjoint operator  2 {Âi j }.

Postulate 4 The possible values of the measurement of observable A for

the system s in state ψ lie in the spectrum (the set of eigenvalues) of
the associated operator Â. This was called the Quantization Algorithm
by M. Redhead [Red.87, p.5].

Postulate 5 Here we have the Born Rule. Given a system s, which is as-
sociated to a 4-tuple σ according to the above definition, let A be an
observable to be measured on the system in state ψ. First we take the
Hilbert space of the states of the system, H. Now, let {αn } be an or-
thonormal basis for H formed by eigenvectors of  (something that is
possible to assume, since  is diagonalizable), so that there are com-
plex numbers cn such that ψ = n cn αn , with nn |cn |2 = 1. We know
P P
that the cn are the Fourier coefficients cn = hαn |ψi, where h· | ·i is the
inner product. Let us denote the eigenvalues associated to the vectors
αn by an , that is, Âαn = an αn . Then we have the Statistical Algorithm
[Red.87, p.8]: the probability that the measurement of observable A
gives the value an when the system is in the state ψ is

ProbA (an ) = |cn |2 = |hαn |ψi|2

ψ

for the non-degenerate state (that is, all eigenvalues of  are distinct;
when the operator is degenerate, the probability is obtained by sum-
ming the |c j |2 for all α’s associated to the same eigenvalue.22
Other possible states that are not pure are called mixtures. They can
be briefly described as follows by means of statistical operators (or
statistical matrices) [Red.87, pp.15-6]. We assign probabilities wk to
22
For details, see [Red.87, p.8].
150 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

a set of pure states {βk } in which the system may be found, so that
we have a statistical ensemble of several quantum (possible) states.
Let Pβ p denote the projection operator whose range is the unitary sub-
space generated by β p . Then the statistical operator for the system
becomes X
ρ= wk Pβk ,
k

and the expectation value of an observable A is given in terms of its

Hermitean associated operator by

hAi := Tr(ρ · Â),

where Tr is the trace function.

Postulate 6 Let us call this one the Dynamic Postulate. It says that if the
system is in the instant t0 in state ψ(t0 ) — here the notation is adapted
in order to consider the state as depending on time — then in a distinct
time t the system evolves to the state ψ(t) according to the Schrödinger
equation [Pen.05, p.536], [Red.87, p.12]

ψ(t) = Û(t)ψ(t0 ),

where Û is a unitary operator.

Postulate 7 This is the Collapse Postulate. It says that immediately after
the measurement of observable A for the system in state ψ = n cn αn ,
P
gives the value |cn |2 = |hαn |ψi|2 , the system enters in the state de-
scribed by the corresponding eigenvector αn .

We shall give now some hints about how to connect the above structure
with possible applications.
Let is suppose that the state of a physical system is given by the vector
ψ in a suitable Hilbert space. In order to represent observables like position
and momentum, we need infinitely many dimensional Hilbert spaces. So,
 5.8. THE APPROACH OF SUPPES 151

suppose a particle moving in a line, and let X be the position observable,

whose Hermitian corresponding operator is X̂. If {ψ x } is an orthonormal
basis for the Hilbert space formed by eigenvectors of X̂, being λi be cor-
responding eigenvalues, that is, X̂ψi = λi ψi . Thus the state vector can be
expanded in this bases, that is, we may write
Z
ψ = hψ x |ψiψ x dx,

xxx and the Fourier coefficients in this expansions enable us to for the wave-
functions
ψ(x) := hψ x |ψi,
which are the functions ψ(x, t) = h(x, t)|ψi described above with another
notation. If we assume that ψ is normalized, that is kψk = 1, then we have
Z
ψ(x)⇤ ψ(x)dx = 1,

where the star stands for complex conjugation. So, ψ(x) must be square in-
tegrable, that is, an element of the Hilbert space L2 of the square integrable
complex functions.23 But, in general, we need also not-square integrable
functions, so we need to work in an expanded, or rigged Hilbert space, but
we shall not enter in this discussion here [Mad.05].
Of course all non-relativistic quantum mechanics can be developed from
this axiomatic basis.
One of the interesting questions that can be put here is the following
one. Since quantum objects may be indiscernible, sharing all their char-
acteristics (for instance, bosons in the same quantum state), should S be
taken to be a set? We remember that in standard (extensional) mathematics
a set is a collection of distinct objects.24 We will not address this question
in full here, but just to call the attention of the reader to the importance
R1
23
Roughly, this is the space of the complex functions f such that 1 | f (x)|2 dx < 1.
24
Cantor’s ‘definition’ of the concept of set is well known: sets are collections of distinct objects of our
intuition or of our thought [Low.14, p.85].
152 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES

of considering the mathematical apparatus where the structures are devel-

oped. So, let us give two examples.

The first one concerns semantics. The mathematician Yuri Manin said
that quantum mechanics (better, its language) has no semantics. And more:
he suggested that quantum mechanics has not even a language, a proper
language, making use of a fragment of the language of standard functional
analysis (the theory of Hilbert spaces in most presentations) [Man.10]. But,
given the axiomatics presented above, we may say that there is a reason-
able language involved, and we as philosophers or as logicians would like
to know whether it is possible to ascribe to it a reasonable semantics in
the standard sense. In order to to it, first we need a domain of discourse,
let us call it D. But suppose, as it is common in many quantum cases,
that the quantum objects being considered are indiscernible (or identical
in the physicists’ jargon, which is an abuse of language), say a collection
of identical bosons in the same quantum state. In this situation, they have
all properties in common, even spatio-temporal ones, if these notions make
sense at all. In this case, we would be not be able to discern the systems by
any means. But, if we are using ZFC as our metatheory, D is a set, hence its
elements are always distinguishable, even if only in principle; if D is finite,
they can be effectively discerned. This is of course contrary to the idea of
the indiscernibility of the quanta. So, we would be in need of a different
metamathematical framework to elaborate our semantics. Such a theory,
as we have seen already, exists and is called quasi-set theory, where we
may have collections of objects which may be completely indiscernible!
For details, see [AreKra.14], [DomHolKra.08], [FreKra.06], [KraAre.15].

The second example concerns names of quantum entities. By the same

reasons posed above, in certain situations we should have no way to name
certain quantum systems, even if they are fermions, which obey Pauli’s
Exclusion Principle. In fact, suppose that we name Peter and Paul the
two electrons of an Helium atom in its fundamental state (less energy).
5.8. THE APPROACH OF SUPPES 153

Supposing that we can speak of the electrons in this situation,25 they are in
a superposition, and cannot be discerned. But let us suppose we wish to
measure the spin of the electrons in a given direction, say in the z-direction.
Thus, we know from quantum mechanics that one of them will have spin
UP in the chosen direction, while the other one will have spin DOWN in the
same direction. So, they do not have all the same properties in common,
saving Pauli’s principle.26 But, which one has spin UP? Peter, Paul? There
is no way to know the answer! If we admit that something is lacking, we
need to assume that there are hidden variables of some kind and not all
physicists are sympathetic with this idea. Thus, apparently, this is not a
simple epistemological question in which we are not able to identify each
of the two electrons: it is beyond doubt that the names have no meaning
in this world. They are just mock names, which have not significance as
proper names have in standard semantics.27
As we see, there are good reasons for looking for an alternative meta-
mathematical framework, at least in the case of quantum mechanics. For
further details on this point, see the above suggested references.

25
Thus we need to adopt an interpretation which is not ‘instrumentalist’ such as Bohr’s, to whom quantum
mechanics just provides us probabilities of measurements.
26
Electrons are fermions and cannot have all the same quantum properties in common. This is what
Pauli’s principle says, roughly.
27
This ‘mock individuality’ was an idea advanced by Toraldo di Francia; see [FreKra.06, p.361] also for
references.
154 CHAPTER 5. AXIOMATIZATION, AND SCIENTIFIC THEORIES
Chapter 6

Models and scientific theories

n this chapter, we shall return to issues of a foundational-philo-

I sophical nature, where we address important questions about the

nature of scientific theories in the light of the developments made
in the previous chapters. Questions concerning the nature of sci-
entific theories are a constant presence in the philosophy of science, and
as we saw in chapter 1, the debate on the precise nature of the semantic
approach has recently reappeared. Our main aim is to explore the conse-
quences that the specific framing of theories as exposed in the previous
chapter may have when we deal with models, languages, and the general
structure of scientific theories. The investigation will proceed at the met-
alevel.
As we have seen in the previous chapters, models are central to the se-
mantic approach to scientific theories in both Suppes’ and in the da Costa
and Chuaqui’s views. But the word ‘model’ has distinct meanings in these
two approaches. Here, we recall that, just as we have done in previous
chapters, we are following Patrick Suppes, according to whom most of the
distinct senses of the word ‘model’ may be profitably studied as set theo-
retical structures (on the plurality of meanings of the word ‘model’, among
them iconic models, analogy models, logical models, and others, see the
discussion in [Sups.02, §2.1]). As we have remarked in chapter 1, the cur-
rent debate centers around the issue of whether a set theoretical structure
156 CHAPTER 6. MODELS AND SCIENTIFIC THEORIES

must necessarily comprise an interpretation of a language or whether it can

be merely a kind of ‘language-free’ structure. As we have seen, mostly as
a result of the investigations by Lutz [Lut.15], the distinction between both
approaches seems to be practically none. In this chapter, among other
things, we shall see how both da Costa and Chuaqui as well as Suppes’
approaches feature in these debates.

6.1 Again on the order of languages and structures

As a first important remark (which was already highlighted in chapter 1),

it is usually claimed that the semantic or the model-theoretical approach
to scientific theories was inspired somehow in Tarski’s work of the 1950s,
when he started the development of (standard) Model Theory, a branch of
mathematical logic that deals mainly with the relations between first-order
formal languages and set theoretical structures that interpret them. So,
by taking a glimpse in the relevant literature, we find that most philoso-
phers think that the models of scientific theories are models of first-order
languages in the Tarskian sense, which as we shall see soon, is a simpli-
fication of the issue, given that in general the languages of most scientific
theories are not first-order languages and the structures are not (as we shall
call them) order-1 structures. So, let us begin by liberating the whole dis-
cussion from the limitation to first-order languages and structures that are
adequate exclusively to such languages.
In the previous chapter, we have defined rigorously the order of a formal
language. As it is common in most cases, these languages are finitary lan-
guages, comprising only formulas of finite length and in particular encom-
passing finite families of conjunctions and disjunctions, as well as finite
strings of quantifiers. So, we can characterize them with the general nota-
tion Lnωω , where n indicates the order of the language, n 2 {1, 2, 3, . . .}. If n
is not a natural number, we say that the order of the language is ω. When
n = 1, we speak of first-order languages, of second-order languages when
6.1. AGAIN ON THE ORDER OF LANGUAGES AND STRUCTURES 157

n = 2, and so on. The structures where these languages are interpreted are
order-1 structures, and this means that we quantify only over individuals of
the domain(s) of the structure, but neither over sets of such individuals (i.e.
their properties), nor over elements of higher-order, such as sets of sets of
individuals, and so on. Second-order languages admit predicate variables,
which are interpreted (extensionally) as sets of individuals, binary relations
of individuals, and in general in k-ary relations of individuals.1 Third-order
languages are still more general, and so on, until the language of order ω,
which is essentially the language of type theory.
The definition of the order of a structure was given previously, and we
shall not repeat it here; intuitively, the greater the order of the structure,
the greater is the order of the elements which are relata of the involved
relations (sets of sets, sets of sets of sets, and so on). Most scientific the-
ories employ sophisticated mathematical theories, so that they cannot be
characterized as structures of order-1. A simple example is that of a topo-
logical space, which can be written as T = hX, τi, being X a non-empty
set and τ a topology over X, that is, a certain set of subsets of X obeying
well-known postulates.2 But, as we have seen already, more sophisticated
structures, such as those for classical particle mechanics, non-relativistic
quantum mechanics, to keep with our examples, require much more than
simple relations whose relata are elements of the domains. So, in order
to deal with interesting scientific theories, we need to employ higher-order
structures.
In view of that need of higher-order structures, how is the semantic
approach to theories to be framed? As we have seen, our understanding
of the semantic approach is that, following the general lines of French
[Fre.15], it simply furnishes the tools for representing theories. The issue
of whether there is the need to have an underlying language and how it
1
As before, here we concentrate on relations only, since distinguished elements and operations can be
viewed as particular relations.
2
The postulates are: (1) ; and X belong to τ; (2) any finite intersection of elements of τ belongs to τ, and
(3) any reunion of elements of τ belongs to τ, which can be seen in any book Topology (the first postulate
is implied by the other two [?, chap.1]).
158 CHAPTER 6. MODELS AND SCIENTIFIC THEORIES

affects the representational power of the approach is something we shall

also touch on. The main point is that language is a problem when we
identify a theory with some of its formulations. As we shall see, both the
approaches by da Costa and Chuaqui as well as the approach by Suppes
may deal with those issues. Thus, we shall add some more philosophical
remarks on both approaches and the relation between the postulates of their
Suppes’ predicates with their respective models.
We begin with the approach by Suppes to theories.

6.2 Further remarks on Suppes’ approach

Let us recall that Suppes’ approach to scientific theories consists in spec-

ifying a wide class of structures which he calls the ‘models’ of a certain
set theoretical predicate. The predicate itself is understood as axiomatiz-
ing the theory, so that its postulates are formulated in the language of set
theory itself, perhaps extended with the symbols proper of the considered
theory. Our examples in the previous chapter were classical particle me-
chanics, group theory, vector spaces theory, and non-relativistic quantum
mechanics. For continuing the discussion and also for presenting a new
example, let us consider the real vector space Rn = hRn , R, +, ·i, where Rn
is the set of n-tuples of real numbers, and the operations are the addition of
n-tuples (by summing the corresponding coordinates) and the product of
n-tuples by real numbers. This structure is said to be a model of the theory
of vector spaces. But, what does this affirmation mean in precise terms?
Are we really speaking about models in the same sense of Tarski?
The answer to that question will bring to light what is really distinc-
tive of the approach by Suppes to the semantic approach. The main point
is that, by lying the postulates of the theory directly in the language of
set theory, and not in a specific formal language which is interpreted by
the relevant structures, Suppes leaves behind precisely the semantic com-
ponent, which is adopted when we have an interpretation of a language
6.2. FURTHER REMARKS ON SUPPES’ APPROACH 159

[Mul.11, sec.6], [KraAreMor.11], [Lut.15]. In the specific case of Suppes,

there is simply no set Γ of postulates in a formal language so that the struc-
tures A of a given class are such that they not only interpret the non-logical
vocabulary of the language but also are such that

A |= Γ.

In fact, following Suppes’ approach, that semantic component would be

impossible, given that the postulates of Γ involve set theoretical vocabulary
and other elements. Recall: the postulates are formulated directly in the
language of set theory. Any interpretation of that language would require
an interpretation of the language of set theory itself (as we discussed in
chapter 3, this can be done only if a stronger set theory is assumed as the
metalanguage).
But then what is the relevant notion of model that is at stake? Certainly
we have set theoretical structures of any order, and the Suppes’ predicate
serves to gather the relevant structures into classes that represent the theory.
Given those classes, we can then study the relations between distinct the-
ories, search for representation theorems, study the relation of the theory
with data, and so on. But how those structures ‘model’ the set theoretical
predicate, if not in the Tarskian sense of satisfaction and truth?
The whole point is that we can prove within set theory that the struc-
tures in the relevant class have the properties demanded by the set theo-
retical predicate. Let us consider, for instance, the case of the real vector
space Rn = hRn , R, +, ·i. Following the traditional mathematical practice,
we simply prove inside set theory itself that the structures ‘satisfy’ the pos-
tulates of vector space, which we represent by writing

Rn |= (vector space axioms), (6.1)

where |= is to be understood in the usual semantical consequence symbol.

This means that, once obtained the ‘ingredients’ of the structure in set the-
ory, which are just sets (here we are avoiding the Urelemente), all we do
160 CHAPTER 6. MODELS AND SCIENTIFIC THEORIES

is to prove as theorems of set theory those sentences in the language of set

theory (probably expanded by new symbols) which express the postulates
of vector spaces for this particular sample. For instance, in order to show
that Rn |= 9O8α(O + α = α), we take O = h0, . . . , 0i and α = hx1 , . . . , xn i
and then show that

O + α = h0, . . . , 0i + hx1 , . . . , xn i = h0 + x1 , . . . , 0 + xn i = α.

That is, using the terminology with some care but in a very clear way,
we show that

(Set theory) ` ( A |= (vector space axioms) ).

That is, as says Suppes, we are working with mathematics, and not with
metamathematics. This shows that the notion of satisfaction of the postu-
lates by a structure, in Suppes’ approach, is purely syntactical, consisting
in mathematical proofs inside set theory. For an even simpler example, the
set ω of natural numbers, as usually constructed inside set theory (ZFC, for
instance). The typical proof that this set satisfies the Peano Postulates does
not proceed by association of the relevant elements of the set with a formal
language. Rather, what we usually do is to provide for a proof inside set
theory itself that there is a first element (the empty set), a sucessor function
working as expected and that ω is inductive (see chapter 3 and [End.77] for
further details).
Now, with these features in mind, we can characterize a scientific theory
according to Suppes’ ‘semantic’ view: let P the set theoretical predicate
that axiomatizes a certain theory T , and A a structure that satisfies the
predicate. This structure is build with elements of a scale ε(D), where D is
the domain of the structure (in our sample case, Rn ). As a set (in most cases
for scientific theories), each structure belongs to some Vα of the cumulative
hierarchy, so, we can write in a short notation (suggested in [Mul.11]):

T = {A 2 Vα : P(A)}, (6.2)
6.2. FURTHER REMARKS ON SUPPES’ APPROACH 161

which means that a theory is a collection (not a set of standard set theo-
ries, perhaps a proper class)3 of structures that models the set theoretical
predicate.
Recall from the discussion in the previous chapter that a set theoretical
predicate, in the sense considered by Suppes, is able to select very specific
classes of structures. It may provide for a selection of so-called intended
models of a theory, leaving aside models that merely have the desired for-
mal properties but that are of no relevance for the empirical study the the-
ory was developed for. This will contrast with the classes of structures
selected by the da Costa-Chuaqui approach, which we will explain soon,
where one cannot ‘exclude’ some specific unintended structures from be-
ing inside the class comprising (representing) the theory.
So, in a sense, Suppes’ own approach is not really a ‘semantic’ ap-
proach. Muller [Mul.11, sec.6] has noticed that, and Suppes [Sups.11]
agreed. Then, even though Suppes himself mentioned the Tarskian tra-
dition of models as encapsulating precisely the kind of model needed in
science (for instance, in [Sups.62]), in the end Suppes recognized that the
models he needed are of a distinct nature. In fact, this approach is more
likely to be useful when one is interested, as Suppes was, in the relation
between theory and experiments. The simplicity of the approach leads one
more directly to deal with such issues.
That last point is important for a discussion that was already raised in
the first chapter: how can one make sense of the fact that we are ‘axiom-
atizing a theory’, an informal and pre-given theory T , and afterwards, as
it were, to develop a set theoretical surrogate for it? Recall that there is
a simple way out for that. In fact, following French [Fre.15], we are not
claiming that an informal theory must be identified with its set theoretical
counterpart; the set theoretical construction serves specific purposes, and
in the case of Suppes, as he clearly stated in his [Sups.67], two of the main
3
In the von Neumann-Bernays-Gödel (NBG) set theory, the primitive elements are classes, and sets,
which are copies of the ZFC sets, are classes that belong to other classes. The classes that do not belong to
other classes are called proper classes. These entities lie outside ZFC.
162 CHAPTER 6. MODELS AND SCIENTIFIC THEORIES

purposes are the search for a better understanding of the relation between
data and theory and the study of theories even in the absence of standard
axiomatizations. The main problem appears, recall, when one reifies a set
theoretical construction and then must face the battle of proving this to be
the ‘correct’ reification against seemingly equivalent reifications.
Thus we have arrived at a possible characterization of scientific theories.
This is the first step for the discussion of philosophical points of interest,
such as the relation of theory with experience, the characterization of con-
cepts such as empirical adequacy and many others. Bas van Fraassen, for
instance, advocates this approach and complements that we should look for
empirically adequate theories [vanF.80], instead of the traditional search
for ‘true theories’. Here we will not enter this discussion; our aim is to
proceed and present the da Costa-Chuaqui approach and, later, comment
on how metamathematical issues impinge on such characterizations.

6.3 da Costa and Chuaqui, further remarks

Now, let us turn again to the approach by da Costa and Chuaqui to sci-
entific theories. As we have already seen, differently from the approach
of Suppes, this is a semantic approach in the traditional Tarskian style of
semantics. However, differently from what is usually thought in the lit-
erature about the semantic approach, the relevant languages and structures
involved need not be restricted to the elementary case, or order-1 structures
(recall our example of axiomatization of classical particle mechanics in the
previous chapter). Higher-order languages and structures are allowed and,
in fact, they seem to be even required when it comes to most empirical
theories.
What is more relevant is that in this case there is a sensible interpretation
of a formal language (recall our discussion in the previous chapter about
the language of a structure). It is in the language of the structure that
one may write the postulates of the theory. In this sense, a theory is a
6.3. DA COSTA AND CHUAQUI, FURTHER REMARKS 163

class of models of a set of postulates written in a specific language. More

precisely, given the language £ for a species of structures, and given a set
of postulates adequate A£ for that theory, the theory may be characterized
by the class
{A : A |= A£ }.

Here, recall, just as it happened in the case of the Suppes approach, there is
no problem with the fact that we are using a theory, in the informal sense, to
construct a set theoretical representation for philosophical purposes. But,
once we have the approach by Suppes, what are the advantages of follow-
ing this traditional approach with formal languages and axiomatization of
the whole step theories?
This is an important question, and we have already signed to the possi-
ble answers that it may get. Formalization of a theory brings with it much
that may count as advantage. For instance, the discovery of unintended
models (which is deemed as a disadvantage by some adherents of the se-
mantic approach) has proven very helpful in recent times (more on this
soon, with relation to the metalanguage). For another advantage, recall,
formalization opens the door for the use of formal methods that may allow
us to obtain results that, at a purely informal level, are not easily achievable
(see again the chapter 2).
But perhaps the most interesting kind of result that the da Costa-Chuaqui
approach allows us concerns the overcome of some problems related with
theory equivalence. Recall from chapter 1 that a difficult challenge appears
for those identifying a theory either with a class of models or else with a
specific linguistic formulation: in the case of the semantic approach, al-
ternative classes may reasonably claim the rights of representing the same
theory; in the case of the syntactic approach, distinct linguistic formula-
tions may be employed to characterize the same theory. Even those giving
up identification, as we are doing, owe an account of how distinct represen-
tations may be reasonably employed to represent the same theory. As we
have discussed in chapter 1, the main general claim is that theory equiva-
164 CHAPTER 6. MODELS AND SCIENTIFIC THEORIES

lence is achievable in the presence of a language for the structure. That is

precisely what we have in the da Costa-Chuaqui approach.
The approach by da Costa and Chuaqui offers a great opportunity for
the rigorous study of equivalent theories, in the precise sense that we shall
present now, rather briefly. Here we shall rely on the exposition on defin-
ability presented in the previous chapter and on the general development of
those concepts by da Costa and Rodrigues [Cos.07]. In that work, da Costa
and Rodrigues relate definability problems with certain kinds of invariance
by automorphism in a structure. The group of automorphisms of a struc-
ture is the Galois group of that structure. The authors are able to generalize
model theoretic notions and some tools of model theory for higher-order
languages and structures. By considering infinitary higher-order languages
as the language of our structure, following [Cos.07], we shall call a con-
cept definable in this language as being definable in the wide sense. Now,
two structures A and A0 are equivalent, written A ⌘ A0 if and only if the
primitive relations of each of them are definable in terms of the primitive
relations of the other. So, we may think of two structures with distinct sig-
natures as presenting the same theory only when they are equivalent in this
sense.
What is more relevant for us now is that this notion of equivalence
(which is somehow advanced in [Hal.12, Hal.15] and [Lut.15]) has also
a model theoretic counterpart. As a consequence of the work by da Costa
and Rodrigues, da Costa and Bueno [Cos.11, p.157], we can state a theo-
rem with necessary and sufficient conditions for equivalence: a given ob-
ject r is definable in a certain vocabulary on the basis of a certain set of
hypotheses Γ if and only if, for every structure A modeling Γ, r is invariant
by the Galois group of the restriction A0 of A to the language without the
symbol denoting r (for details, see [Cos.07, Cos.11]).
Now, this may be a first step in the study of equivalence of theories.
What is interesting is that the Galois group of structures (theories)4 pro-
4
The Galois group of a theory is the Galois group of the structures that model the set-theoretical predi-
cate that axiomatize the theory.
6.4. THE METAMATHEMATICAL FRAMEWORK 165

vides interesting techniques, relating syntactical aspects which are not al-
ways fully developed with model theoretic tools for the equivalence of
theories. This is, as we mentioned, the first step in such studies, and may,
perhaps, replace the debates on which approach is the ‘correct’ one: the
syntactical or the semantic. They are both required and the correct devel-
opment of their relations may bring useful results for philosophy of science
as a discipline.
Obviously, this is a kind of metamathematical study of theories. This
simply puts the metamathematical resources in the center of the stage in
the philosophy of science again, or at least, in part of it. We shall now
discuss a little more this aspect of the study of theories now, concentrating
in the role of the metamathematical apparatus.

6.4 The metamathematical framework

In order to illustrate why the consideration of the mathematics where the

discussion is being is being developed is so relevant, let us advance some
case-studies, put here without a detailed study, but just as an attempt to
raise the problem. This is what we shall do now.
It is well known that quantum mechanics needs unbounded operators,
such as position and momentum operators [Hea.90]. An unbounded oper-
ator T is a linear operator (here supposed over a suitable Hilbert space H)
such that for any M > 0 there exists a vector α such that ||T α|| > M.||α||.
Otherwise, T is bounded. Now consider the theory ZF+DC, where DC
stands for a weakened form of the axiom of choice, entailing that a ‘count-
able’ form of the axiom of choice can be obtained. (In particular, if {Bn :
n 2 ω} is a countable collection of nonempty sets, then it follows from DC
that there exists a choice function f with domain ω such that f (n) 2 Bn for
each n 2 ω.) It can then be proven, as Solovay showed, that in ZF+DC
(which is supposed to be consistent) the proposition ‘Every subset of R is
166 CHAPTER 6. MODELS AND SCIENTIFIC THEORIES

Lebesgue measurable’ cannot be disproved.5 This proposition is false in

standard ZFC. The same happens with the proposition ‘Each linear oper-
ator on a Hilbert space is bounded’ [Mai.73]. This kind of result poses a
difficulty to the friends of the semantic view: when we speak of the models
of a scientific theory, such as quantum mechanics, which metamathemat-
ics should we use to define its models? Presumably, it cannot be Solovay’s
model in ZF+DC, since we need unbounded operators. So, the choice of
a suitable metamathematics is crucial. The characterization of a scientific
theory according to the semantic approach is sensible to the mathematical
apparatus employed in the very construction of those models!
As a second example, let us acknowledge that in the standard Hilbert
space formalism for quantum mechanics, we deal with bases for the rel-
evant Hilbert spaces. More specifically, we deal with orthonormal bases
formed by eigenvectors of certain Hermitean operators. This is possible
because we can prove, using the axiom of choice (which is part of the
metatheory used here, essentially ZFU, so we are supposing the existence
of Urelemente) that any Hilbert space H has a basis. Moreover, it can also
be shown that each basis has a specific cardinality, which is the same for all
bases of H (this is defined to be the dimension of the space). But in certain
set theories in which the axiom of choice does not hold in full generality,
such as in Laüchi’s permutation models, we obtain: (a) vector spaces with
no basis, and (b) a vector space that has two bases of different cardinali-
ties [Jec.77, p.366]. Now, if a vector space has no basis, it seems that it
cannot be used as part of the standard formalism of quantum mechanics.
The latter formalism presupposes the availability of suitable bases. As a
result, again we see it very clearly that the theory, understood as a class of
models, depends crucially on the metamathematics that is used. Of course
it seems interesting for the philosophy of science to study such models as
well as possible models of a scientific theory.
5
The Lebesgue measure generalizes the usual notion of ‘measure’ of a set, say the length of an interval,
the area of a plane figure, the volume of a tridimensional figure. The precise definition is not necessary
here.
6.4. THE METAMATHEMATICAL FRAMEWORK 167

Our third example also comes from quantum mechanics (for details, see
[KraBue.07]). In order to keep the example self-contained, we recall that
in the 1920’s, Thoralf Skolem noticed that there are concepts that are, as it
were, the same in all models of, say, ZFC (these concepts are termed abso-
lute). For instance, the concept of ‘ordinal’ does not change from model to
model as the concept of ‘cardinal’ does (these are called relative concepts.
For example, if ZFC is formulated as a first-order theory, if consistent, it
has a countable model due to the Lowënheim-Skolem theorem. But, in
this model, the set of real numbers, which can be constructed in ZFC (say
by Dedekind cuts), must be countable — a fact that seems to contradict
‘Cantor’s theorem’, according to which there is no bijection between the
set of real numbers and the set of natural numbers. However, as Skolem
himself noticed, this result does not lead to a ‘real’ paradox, for the bijec-
tion may exist outside the countable model, and not as a set of ZFC proper
[Sko.22]. But, in the standard model of ZFC, the cardinality of the set of
the real numbers cannot be denumerable, as Cantor’s theorem teaches us.
This result shows that the set of real numbers may have different cardinal-
ities depending on the model we consider, and this happens in general for
other sets.
The relativism of the notion of cardinal number can be seen with this
simple example.6 Suppose that our theory makes use of the set A =
{2@0 , @α }, where α is a cardinal. How many elements has this set? Well,
it depends on the model of set theory we are considering. Really, if the
Continuum Hypothesis holds,7 the set has just one element, but has more
than one otherwise (say, when 2@0 = @ω ).
Let us take this result in mind for what follows. After having pre-
6
We owe this example to A. M. N. Coelho.
7
Cantor’s Continuum Hypothesis says that there is no set of real numbers with cardinality between the
cardinality of the natural numbers (@0 ) and the cardinality of the set of real numbers, c (for ‘continuum’).
But Cantor himself proved that 2@0 = c, and postulate that the next cardinal number after @0 , termed @1 ,
is precisely c. The generalized continuum hypothesis (GCH) says that 2@α = @α+1 ; it was known from
Sierpinski that ZF+GCH implies the axiom of choice. It was later shown by Gödel and Cohen that this
hypothesis is independent of the axioms of ZF (without choice).
168 CHAPTER 6. MODELS AND SCIENTIFIC THEORIES

sented his modal interpretation of quantum mechanics, Bas van Fraassen

addresses the problem of identical particles in quantum physics, which he
regards as one of the three main issues in the philosophical discussion on
quantum mechanics [vanF.91, p.133]. And he notes: “identical particles
[. . .] are certainly qualitatively the same, in all the respects represented
in quantum-mechanical models — yet still numerically distinct” (ibid., p.
376). In a previous paper, he was still more explicit, insisting that “if two
particles are of the same kind, and have the same state of motion, noth-
ing in the quantum-mechanical description distinguishes them. Yet this is
possible” [vanF.84].
Van Fraassen’s quotations are intriguing. Particles of the same kind and
in the same state of motion are ‘identical’, in the physicists’ jargon, and
according to their standards, nothing can distinguish them. So, if they can-
not be distinguished within the quantum-mechanical formalism, how can
they still be distinguished at all? The answer, we suggest — following
the parallel case made by Skolem in set theory — is that the particles can
be distinguished outside the framework of quantum mechanics. But what
does this mean? As we have seen, in the foundations of set theory, consid-
erations regarding what holds inside or outside a certain model, are quite
common. But can we make sense of this way of speaking in philosophy of
science as well?
Van Fraassen’s modal interpretation of quantum mechanics takes quan-
tum propositions as modal statements, which give “first and foremost about
what can and what must happen, and only indirectly about what actually
does happen” [vanF.80]. In other words, the modal account, by offering
an interpretation of quantum mechanics, spells out how the world could
be if quantum mechanics were true [vanF.91, p.242]. To motivate his pro-
posal, van Fraassen recalls one of the most intriguing features of quantum
physics, namely, the sense in which quantum mechanics is an indeterminis-
tic theory. Although the dynamics of an isolated system evolves according
to Schrödinger’s equation (hence deterministically), the system as a whole
6.4. THE METAMATHEMATICAL FRAMEWORK 169

cannot be analyzed in terms of its component parts. So, apparently, the

quantum mechanical state of the whole system contains only incomplete
information about the system. Bohr’s proposal, recalls van Fraassen, em-
phasizes that it is still possible to have complete information about the
system, given that the states of the system’s components and the state of
the whole system do not determine each other [vanF.80]. As a result, on
the basis of the state of a complete system X + Y, we can in general ascribe
at most mixed states to X and Y, but from them nothing can be said back
about the state of the whole system. As van Fraassen notes:

“[I]f we can predict the future states of an isolated system on the

basis of its present state [by means of Schrödinger’s equation],
then how can we be ignorant about the future events involving
its components unless the information in those total states is in-
complete? For surely any true description of a component is a
partial but true description of the whole?” (ibid.)

Van Fraassen’s answer is obtained from a distinction between quantum

dynamical states and experimental events. The former are what a vector
or a statistical operator represents. They are things completely embedded
in the theory, whose evolution is governed by dynamical laws. In other
words, we can say that dynamical states are described within the formal-
ism of quantum mechanics. Events, on the contrary, are extra-theoretic
entities that satisfy the probability calculations in the sense physicists do
in quantum theory.
The same conceptual distinction can be drawn by distinguishing be-
tween state attributions and value attributions of a physical system. The
former is a theoretic construct, and part of the challenge involved in the-
ory’s construction depends upon a proper representation of these states.
Value attributions, in turn, express values that an observable actually have.
Since the point is important for our argument, let us consider it in more
detail.
170 CHAPTER 6. MODELS AND SCIENTIFIC THEORIES

A value state is specified by stating which observables have values and

what they are. A value-attributing proposition then states that an observ-
able m actually has a value in a (Borel) set E. (In symbols, hm, Ei.) The
connection between them is that value states are truth-makers of value at-
tributing propositions [vanF.91, pp.275-6]. On the other hand, we have the
dynamic state, which states how the system will evolve, either in isola-
tion or in interaction with another system. A state-attributing proposition
then states that a measurement of an observable m must have a value in
a (Borel) set E. (In symbols, [m, E].) Again, dynamic states and state-
attributing propositions are connected by the fact that the former are what
make the latter true. Now, the crucial feature of the modal account is to dis-
tinguish value- and state-attributing propositions. The motivation for this
distinction comes from difficulties faced by the standard interpretation of
quantum mechanics, as articulated by von Neumann, for not distinguishing
them.
Von Neumann’s interpretation of quantum mechanics identifies these
two concepts. After all, not only does von Neumann consider that a system
can be said to posses a value of a certain variable when it is in an eigenstate
of the corresponding observable, but he also accepts that if the state vector
is not an eigenstate of some observable, then it has no value at all (see also
[Bit.96, p.149]). In this case, the system is supposed to be characterized by
a well-defined value of the observable when the probability is 1. But if this
probability is not 1, then the observable is supposed to have no value at all.
To remove this discontinuity, van Fraassen offers an account according to
which probability ascriptions are not equivalent to value ascriptions (ibid.).
On von Neumann’s interpretation, attributions of values and classifica-
tion of states are closely related: an observable B has value b if and only
if a B-measurement is certain to have outcome b (where b is a real num-
ber). The problem here is that in order to accommodate states for which
measurement has uncertain outcomes, von Neumann made a radical move:
if the outcome of a measurement of B is uncertain, B has no value at all
6.4. THE METAMATHEMATICAL FRAMEWORK 171

[vanF.91, p.274]. To avoid this answer, the modal interpretation introduces

the distinction between values and states. With this distinction in place,
the introduction of ‘unsharp’ values of observables is allowed. And this is
how the possibility of uncertain outcomes in measurement can be accom-
modated.8 As van Fraassen points out [vanF.91, pp.280-1], if a physical
system X has dynamic state (represented by an operator) W at a time t, the
state-attributions [M, E] which are true are those such that Tr(WIEM ) = 1.9
As opposed to state-attributions, value-attributions cannot be deduced from
the dynamic state. But, according to van Fraassen, they are constrained in
three ways:10 (i) If [M, E] is true, so is the value-attribution hm, Ei; that
is, observable M has value in E; (ii) all true value-attributions could have
probability 1 together; and (iii) the set of true value-attributions is maximal
with respect to feature (ii) [vanF.91, p. 281]. So, the assignment of truth-
conditions to state- and value-attributing propositions is crucial to spell out
the difference between them (the former, but not the latter, can be deduced
from the dynamic state).
To sum up, there is an important distinction between state attribution
and value attribution, or between states and events, and this distinction
cannot be reduced to something more basic. States, as already noted, are
described in the scope of (the formalism of) quantum mechanics by vectors
of an appropriate Hilbert space, while events are not. After all, events are
statements such as Observable B pertaining to system X has value b, and
such events are described if they are assigned probabilities, but “they are
not the same thing as the states which assign them probabilities” [vanF.91,
8
Note that the value-state distinction is cashed out in terms of the concept of truth. The relationship
between these ideas and the concept of quasi-truth is developed in [Bue.00].
9
A few comments about the notation: (a) Tr is a linear functional of operators into numbers (the trace
map), which gives us the probability that a measurement of the observable m has a value in the Borel set E;
(b) IEM is an Hermitean operator such that IEM (x) = x if M(x) = ax for some a 2 E, and is the null vector
if M(x) = bx for some value b < E, where M is the Hermitian operator which represents m. (c) That the
trace function Tr provides a probability is due to the fact that Pmx (E) = (x · IEM x) = Tr(I x IEM ), where Pmx (E)
is the probability that a measurement of m has a value in E, (x · IEM x) is the inner product of x and IEM x and
I x is the projection on the subspace [x] spanned by x. For details, see [vanF.91, pp. 147-152, 157-165, and
280-281].
10
Which again are spelled out in terms of truth.
172 CHAPTER 6. MODELS AND SCIENTIFIC THEORIES

p. 279].
The distinction between states and events is similar to the distinction be-
tween absolute and relative notions in set theory discussed above, at least in
the following way: we are contrasting intra-theoretic properties with prop-
erties that hold outside the models under consideration. It’s curious that
when Skolem introduced his ‘paradox’, he intended to use it to show the
inadequacy of set theory as a foundation for mathematics. The outcome,
however, was precisely the opposite. His result was incorporated as part of
the rich conceptual framework offered by set-theoretic notions. Similarly,
van Fraassen developed the modal interpretation of quantum mechanics as
part of a defense of an empiricist view. In the end, however, the modal
interpretation became part the revival of realist interpretations of quantum
theory.
It should now be clear that both in the philosophy of science and in the
foundations of set theory there is room for discussing what holds ‘inside’
a particular model (or formalism) and what holds ‘outside’ the model (for-
malism). If a theory is presented as a class of models, it makes perfect
sense to ask whether there are concepts that remain the same in all models,
and concepts that change from model to model; that is, that have a certain
extension ‘inside’ a model, but a different one in another model or when
considered outside the model.
Finally, let’s examine another case. Consider the concept of indistin-
guishable (or indiscernible) objects. The idea of indiscernibility is of fun-
damental importance in contemporary physics (for a historical account and
further discussion, see [FreKra.06]). Standard mathematics and classical
logic imply that every object is an individual, in the sense that each object
can always be distinguished from any other at least in principle, and these
individuals retain their individuality even when mixed with others of sim-
ilar species.11 As a result, to accommodate indistinguishable objects some
mathematical trick needs to be introduced.
11
Although the concepts of individuality and distinguishability should not be confused; see [FreKra.06].
6.4. THE METAMATHEMATICAL FRAMEWORK 173

In quantum mechanics, this is done by imposing some kind of symmetry

condition. Suppose we are to describe how two identical bosons, 1 and 2,
can be distributed in two possible states, A and B. As is well known, the
vectors in the relevant Hilbert space are |ψ1A i|ψ2A i, which states that both
bosons are in A, while |ψ1Bi|ψ2Bi states that both are in B, and p12 |ψ1A i|ψ2Bi +
p1 |ψA i|ψ B i states that one of them is at A and the other is in B. Thus, the
2 2 1
indistinguishability between 1 and 2 (in the third case) emerges from the
symmetry of the function, which is invariant by permutations of the labels.
Some people claim that the individuality of quantum objects is then lost.
According to our point of view, there is nothing to be lost, for in one of
the possible approaches to the subject, these objects do not have identity
conditions to start with [FreKra.06]. The artificiality of the problem is
that these objects were first assumed to be individuals by their labels 1
and 2. Thus, by an adequate choice of the relevant vectors, we have made
them indiscernible. However, the objects cannot be said to be indiscernible
outside the framework (say, ZFC), since we can distinguish them — e.g.,
by their labels 1 and 2. In this way, the notion of indistinguishable object
seems to be relative.
The mathematical trick we used consists in limiting the discourse to the
scope of a certain set-theoretic structure (as we saw, the models of quan-
tum physics can be taken to be such kind of structure). We then consider as
indiscernible those objects that are invariant by the automorphisms of the
structure. Now, in ZFC any structure can be extended to a rigid structure,
that is, to a structure where the only automorphism is the identity function.
Hence, in the rigid structure (the whole ZFC model V = hV, 2i is rigid),
any object is an individual. In short, there are no truly indiscernible objects
in standard mathematics (and logic). A suitable mathematics for develop-
ing quantum mechanics taking the non-individuality of quantum objects
from the start needs to be taken into account; we have developed quasi-set
theory for that, but this subject extrapolates the contents of this book (the
interested reader can have a look at [FreKra.06], [KraAre.15]).
174 CHAPTER 6. MODELS AND SCIENTIFIC THEORIES
Bibliography

[Acz.98] Aczel, P. 1988, Non-well-founded sets, CSLI Lecture Notes, 14, Stanford, CA: Stanford
University, Center for the Study of Language and Information.

[All.38] Allen, E. S. 1938, Review of The Axiomatic Method in Biology, by J. H. Woodger.

Cambridge, University Press, 1937. 10+174 pp. Bull. Amer. Math. Soc., Vol. 44 (11): 763-
764.

[AreKra.14] Arenhart, J. R. B. and Krause, D. 2014, From primitive identity to the non-
individuality of quantum objects. Studies in History and Philosophy of Modern Physics
46 (B): 273-282.

[Aris.89] Aristotle 1989, Prior Analytics. Trans. Robin Smith. Cambridge: Indianapolis, Hack-
ett Pu.

[Arn.97] Arnol’d W., 1995, Will mathematics survive? The Mathematical Intelligencer 7 (3),
6-10.

[BarMac.75] Barnes, D.W and Mack, J.M. 1975, An Algebraic Introduction to Mathematical
Logic. Springer-Verlag.

[Bell.09] Bell, J. L. 2009, ‘Infinitary Logic’, The Stanford Encyclopedia of Philosophy (Spring
2009 Edition).

[BelCas.81] Beltrametti, E. G. and Casinelli, G. 1981, The Logic of Quantum Mechanics,

Addison-Wesley (Encyclopedia of Mathematics and Its Applications, Vol. 15).

[Bit.96] Bitbol, M. 1996. Schrödinger’s philosophy of quantum mechanics. Dordrecht: Kluwer

Academic Publishers.

[Bla.14] Blanchette, P. 2014, The Frege-Hilbert Controversy, The Stanford Encyclo-

pedia of Philosophy (Spring 2014 Edition), Edward N. Zalta (ed.), URL =
<http://plato.stanford.edu/archives/spr2014/entries/frege-hilbert/>.

[Bli.88] Blizard, W. D. 1988, Multiset theory. Notre Dame Journal of Formal Logic 30 (1): 36-
66.

175
176 BIBLIOGRAPHY

[Bog.79] Bogdan 1979, Patrick Suppes. Dordrecht: D.Reidel.

[Bue.00] Bueno, O. 2000. Quasi-truth in quasi-set theory. Synthese 125: 33-53.

[Bou.50] Bourbaki, N. 1950, The architecture of mathematics. The American Mathematical

Monthly, Vol. 57, No. 4, pp. 221-232.

[Bou.68] Bourbaki, N. 1968, Theory of Sets. Paris: Hermann and Addison-Wesley.

[Bou.90] Bourbaki, N. 1990, Théorie des Ensembles. Paris: Masson.

[BrigCos.71] Brignole, D. and da Costa, N. C. A. 1971, On Supernormal Ehresmann-Dedecker

Universes. Mathematische Zeitschrift (122), pp. 342-350

[Cao.99] Cao, T. Y., 1999, Conceptual Foundations of Quantum Field Theory. Cambridge: Cam-
bridge Un. Press.

[Can.55] Cantor. G. 1955, Contribution to the Founding of the Theory of Transfinite Numbers.
New York: Dover.

[Car.58] Carnap, R. 1958, Introduction to Symbolic Logic and Its Applications, New York, Dover.

[Che.09] Chernoff, P.R. 2009, ‘Andy Gleason and quantum mechanics’, Notices of the American
Mathematical Society, November, pp.1253-1259.

[Con.06] Contessa, G. 2006, Scientific models, partial structures and the new received view of
theories. Studies in History and Philosophy of Science 37: 370-77.

[Cor.92] Corry, L. 1992, Nicolas Bourbaki and the concept of mathematical structure, Synthese
92 (3): 315-48.

[Cor.04] Corry, L. 2004, Modern Algebra and the Rise of Mathematical Structures. 2nd.ed.,
Springer Basel AG.

[Cos.80] da Costa. N. C. A. 1980, Ensaio sobre os fundamentos da lógica (in portuguese). 1 st ed.
São Paulo: Hucitec.

[Cos.07] da Costa,N.C.A. 2007, Abstract Logics, Manuscript, Federal University of Santa Cata-
rina (not published).

[Cos.11] da Costa, N. C. A. and Bueno, O. 2011. Remarks on Abstract Galois Theory. Manuscrito
341: 151-183.

[CosChu.88] da Costa,N.C.A. and Chuaqui,R. 1988, ‘On Suppes’ set theoretical predicates’,
Erkenntnis 29, 95-112.

[CosFre.03] da Costa, N. C. A. and French, S. 2003, Science and Partial Truth: A Unitary Ap-
proach to Models and Scientific Reasoning. Oxford: Oxford Un. Press.
BIBLIOGRAPHY 177

[CosRod.07] da Costa, N. C. A. and Rodrigues, A. M. N. 2007, ‘Definability and invariance’,

Studia Logica 82, 1-30.

[CosDor.08] da Costa, N. C. A. and Doria, F. A. 2008, On the Foundations of Science,

COPPE/UFRJ, Cadernos do Grupo de Altos Estudos, Vol. II.

[CosKraBue.06] da Costa, N. C. A., Krause, D. and Bueno, O. 2006, ‘Paraconsistent logics and
paraconsistency’, in in D. Jacquette, D.M.Gabbay, P.Thagard and J.Woods (eds.), Philoso-
phy of Logic, Elsevier, in the series Handbook of the Philosophy of Science, v. 5, 655-781.

[CovHaw.04] Cover, J. A. and O’leary-Hawthorne, J. 2004, Substance and Individuation in

Leibniz, Cambridge, Cambridge Un. Press.

[DalTor.93] Dalla Chiara, M. L. and Toraldo di Francia, G. 1993, ‘Individuals, kinds and names
in physics’, in Corsi, G., Dalla Chiara, M. L. and Ghirardi, G. C. (eds.), Bridging the Gap:
Philosophy, Mathematics, and Physics, Dordrecht: Kluwer Ac. Press (Boston Studies in
the Philosophy of Science, 140), 261-281.

[DalGiuGre.04] Dalla Chiara, M.L., Giuntini, R., and Greechie, R. 2004, Reasoning in Quan-
tum Theory: Sharp and Unsharp Quantum Logics, Kluwer Ac. Press (Trends in Logic,
vol.22).

[Dau.90] Dauben, J. W. 1990, Georg Cantor: His Mathematics and Philosophy of the Infinite.
Princeton: Princeton Un. Press.

[D’Es.03] D’Espagnat, B. 2003, Veiled Reality: An Analysis of Present-Day Quantum Mechanical

Concepts, Westview Press.

[D’Es.06] D’Espagnat, B. 2006, On Physics and Philosophy. Princeton: Princeton University

Press.

[DomHol.07] Domenech, G. and Holik, F. 2007, ‘A discussion on particle number and quantum
indistinguishability’, Foundations of Physics 37 (6): 855-878.

[DomHolKra.08] Domenech, G., Holik, F. and Krause, D. 2008, ‘Q-spaces and the foundations
of quantum mechanics’, Foundations of Physics 38 (11), 969-994.

[DomHolKniKra.08] Domenech, G., Holik, F., Kniznik, L, and Krause, D. 2010, No Labeling
Quantum Mechanics of Indiscernible Particles, International J. Theoretical Physics 49
(12): 3085-3091.

[Dou.11] Douven, I. 2011, Abduction, The Stanford Encyclopedia of Phi-

losophy (Spring 2011 Edition), Edward N. Zalta (ed.), URL =
<http://plato.stanford.edu/archives/spr2011/entries/abduction/>.
178 BIBLIOGRAPHY

[DutRec.15] Dutilh Novaes, C. and Reck, E. 2015, Carnapian explications, formalisms as cog-
nitive tools, and the paradox of adequate formalization. Forthcoming in Synthese, DOI
10.1007/s11229-015-0816-z, 2015.

[EinInf.38] Einstein, A. and Infeld. L. 1938, The Evolution of Physics, New York, Simon and
Schuster.

[End.77] Enderton, H. B. 1977, Elements of Set Theory. New York: Academic Press.

[Esan.13] Esanu, A. 2013, Evolutionary biology and the axiomatic method revisited, The Roma-
nian Journal of Analytic Philosophy, Vol. VII (1): 19-41.

[Eucl.08] Euclid 2008, Euclid’s Element of Geometry. English translation, by Richard Fitzpatrick
http://farside.ph.utexas.edu/Books/Euclid/Elements.pdf

[Fra.66] Fraenkel, A.A. 1966, Abstract Set Theory, Amsterdam, North-Holland.

[FraBarLev.73] Fraenkel, A. A., Bar-Hillel, Y. and Levy, A. 1973, Foundations of Set Theory.
2nd.ed., Amsterdam and London: North-Holland.

[Fra.82] Franco de Oliveira, A.J. 1982, Teoria de Conjuntos, Intuitiva e Axiomática, Lisboa,
Livraria Escolar Editora.

[Fre.15] French, S. 2015, (Structural) realism and its representational vehicles. Synthese, DOI:
10.1007/s11229-015-0879-x, 2015.

[FreLad.03] French, S. and Ladyman, J. 2003, Remodelling structural realism: quantum physics
and the metaphysics of structure, Synthese 36, 31-56.

[FreKra.06] French, S. and Krause, D. 2006, Identity in Physics: A Historical, Philosophical,

and Formal Analysis, Oxford, Oxford Un. Press.

[FreKra.09] French, S. and Krause, D. 2009, Remarks on the theory of quasi-sets, Studia Logica
95 (1-2): 101-12.

[GlaGio.00] Glashow, S. L. y Georgi, H. 2000, La teoría unificada de las fuerzas de las partículas
elementales. In Glashow, S. L., El Encanto de la Física. Barcelona: Tusquets. Original The
Charm of Physics, 1991.

[Ger.85] Geroch, R. 1985, Mathematical Physics, Chicago, Chicago Univ. Press.

[Gra.08] Gratzer, G.A. 2008, Universal Algebra, Springer.

[Gra.00] Gray, J. J. 2000, The Hilbert Challenge, Oxford, Oxford Un. Press.

[God.38] Gödel, K. 1938, The consistency of the axiom of choice and the generalized continuum
hypothesis. In Gödel, K. 1990, Collected Works, Vol. II. Ed. by S. Feferman et al. New
York and Oxford: Oxford Un. Press.
BIBLIOGRAPHY 179

[Gro.99] Gross, D., 1999, The triumph and limitations of quantum field theory. In [Cao.99], pp.
56-67.

[Hal.12] Halvorson, H. 2012, What scientific theories could not be. Philosophy of science 79 (2),
pp. 183-206, 2012.

[Hal.15] Halvorson, H. 2015, Scientific Theories. Forthcoming in The Oxford Handbook of Phi-
losophy of Science.

[Hea.90] Heathcote, A. 1990, Unbounded operators and the incompleteness of quantum mechan-
ics. Philosophy of Science 57: 523-34.

[Hei.89] Heisenberg, W. 1989, Physics and Philosophy, London, Penguin Books.

[Hen.et al.59] Henkin, L., Suppes, P. & Tarski, A. (eds.) 1959, The Axiomatic Method with Special
Reference to Geometry and Physics. Amsterdam: North-Holland.

[Hil.30] Hilbert, D. 1930, ‘Address to the Society of German Scientists and Physicians’, 8th
September 1930.

[Hil.50] Hilbert, D. 1950, Foundations of Geometry (English translation by e. J. Townsend), La

Salle, Open Court.

[Hilb.76] Hilbert, D. 1976, Mathematical Problems. In Browder, F. E. (ed.), Proceedings of Sym-

posium in Pure and Applied Mathematics: Mathematical Problems Arising from Hilbert
Problems. Providence: American Mathematical Society, Vol.CCVIII, Part. I, pp. 1-34.

[Hil.96] Hilbert, D. 1996, Axiomatic thought. In Ewald, W. B. (ed.), From Kant to Hilbert. A
Source Book in the Foundations of Mathematics, vol. 2. Oxford: Oxford University Press,
pp. 1089-1096.

[HilAck.50] Hilbert, D. and Ackermann, W. 1950, Principles of Mathematical Logic. Provi-

dence: American Mathematical Society.

[Hod.93] Hodges, W. 1993, Model Theory, Cambridge, Cambridge Un. Press (Encyclopedia of
Mathematics and its Applications, 42).

[Hod.01] Hodges, W. 2001, Elementary Predicate Logic. In Gabbay, Dov. M., and Guenthner, F.,
(eds.) Handbook of Philosophical Logic, vol. 1, 2nd. edition, 2001, Springer.

[Hof.79] Hoffstadter, D.R 1979, Gödel, Escher, and Bach: an Eternal Golden Braid, New York,
Basic Books.

[Ign.96] Ignatieff, Y. (ed.) 1996, The Mathematical World of Walter Noll. Berlin and Heidelberg:
Springer-Verlag.

[Jam.62] Jammer, M. 1962, The Concepts of Force, Harper Torchbook.

180 BIBLIOGRAPHY

[Jam.74] Jammer, M. 1974, The Philosophy of Quantum Mechanics, John Wiley & Sons.

[Jau.68] Jauch, J.M 1968, Foundations of Quantum Mechanics, Cambridge, Addison-Wesley.

[Jec.77] Jech, T. 1977, ‘About the axiom of choice’, in Barwise, J. (ed.), Handbook of Mathemat-
ical Logic, Amsterdam, North-Holland, pp. 345-370.

[JonBet.10] Jong, R. de and Betti, A. 2010, The classical model of science: a millennia-old
model of scientific rationality. Synthese 174: 185-203.

[Jong.85] Jongeling, T. B. 1985, On an axiomatization of evolutionary theory, J. Theoretical Bi-

ology 117: 529-43.

[Kei.77] Keisler, H. J. 1977, ‘Fundamentals of model theory’, in Barwise, J. (ed.), Handbook of

Mathematical Logic, Amsterdan, North-Holland, 47-103.

[Kne.63] Kneebone, G.T. 1963, Mathematical Logic and the Foundations of Mathematics, Van
Nostrand.

[KraAre.15] Krause, D. and Arenhart, J. R. B. 2015, Individuality, quantum physics, and a meta-
physics of non- individuals: the role of the formal. In A. Guay and T. Pradeu (eds.) Indi-
viduals Across the Sciences. Oxford Un. Press, pp. 61-70.

[KraAreMor.11] Krause, D., Arenhart, J. R. B. and Moraes, F. T. F. 2011, Axiomatization and

models of scientific theories. Foundations of Science, 16: 363-82.

[KraBue.07] Krause, D. and Bueno, O. 2007. Scientific theories, models, and the semantic ap-
proach. Principia 11 (2): 187-201

[Kun.09] Kunen, K. 2009, The Foundations of Mathematics, College Pu.

[Lad.57] Ladrière, J. 1957, Le Limitations Internes des Formalismes. Louvain: Nauwelaerts;

Paris: Gauthier-Villars.

[Lak.76] Lakatos, I. 1976, Proofs and Refutations: the Logic of Mathematical Discovery. Cam-
bridge: Cambridge Un. Press.

[Lan.66] Landau, E. 1966, Foundations of Analysis. New York: Chelsea Pu. Co.

[Lei.95] Leibniz. G. W. 1995, Philosophical Writings, London and Rutland, Everyman.

[Low.14] Low, Z. I. 2014, Universes for category theory.

http://arxiv.org/pdf/1304.5227.pdf

[Lut.12] Lutz, S. 2012, On a Straw Man in the Philosophy of Science: a Defense of the Received
View. HOPOS: The Journal of the International Society for the History of Philosophy of
Science 2 (1): 77-120.
BIBLIOGRAPHY 181

[Lut.15] Lutz, S. 2015, What Was the Syntax-Semantics Debate in the Philosophy of Science
About? Philosophy and Phenomenological research, doi: 10.1111/phpr.12221.

[Lui.97] Lui, S. H. 1997, An interview with Vladimir Arnol’d. Notices of the AMS 44 (4): 432-8
(April).

[Mac.63] Mackey, G.W. 1963, Mathematical Foundations Of Quantum Mechanics, Addison-

Wesley.

[Mac.98] Mac Lane, S. 1998, Categories for the Working Mathematician (Graduate Texts in
Mathematics), Springer.

[Mad.05] de la Madrid, R. 2005, The role of the rigged space in quantum mechanics. In
arXiv:quant-ph/0502053

[McSS.53] McKinsey, J. C. C., Sugar, A. C. and Suppes, P. 1953, Axiomatic foundations of

classical particle mechanics. Journal of Rational Mechanics and Analysis, Vol.2, No.2
(April): 253-272.

[MagKra.01] Magalhães J. C. and Krause D. 2001, Suppes predicate for genetics and natural
selection, J Theor Biol. 21, 209 (2): 141-53.

[Mai.73] Maitland Wright, J. D. 1973, ‘All operators on a Hilbert space are bounded’, Bulletin
of the Americal Mathematical Society 79 (6), 1247-50.

[Man.10] Manin, Yu. I. 2010, A Course in Mathematical Logic for Mathematicians, 2nd. ed.,
New York, Springer.

[Mar.93] Marchisotto, E. 1993, Mario Pieri and his contributions to geometry and foundations
of mathematics, Historia Mathematica 20: 285-303.

[Mar.07] Marquis, J. -P. 2007, Category theory, Stanford Enc. Philosophy, http://plato.
stanford.edu/entries/category-theory.

[Men.97] Mendelson, E. 1997, Introduction to Mathematical Logic. Cornwall: Chapman & Hall.

[Moo.82] Moore, G. H. 1982, Zermelo’s Axiom of Choice: its Origins, Developments, and Influ-
ence. New York, Heidelberg and Berlin: Springer-Verlag.

[Mul.11] Muller, F. A. 2011, Reflections on the revolution at Stanford. Synthese, 183 (1): 87-
114.

[MulSau.08] Muller, F. A. and Saunders, S. W. 2008, ‘Discerning fermions’, British Journal for
the Philosophy of Science 59, 499Ð548.

[MulSee.09] Muller, F.A. and Seevincki, M.P. 2009, ‘Discerning elementary particles;, forth-
coming in Philosophy of Science.
182 BIBLIOGRAPHY

[Nac.07] Nachotomy, O. 2007, Possibility, Agency, and Individuality in Leibniz’s Metaphysics,

Springer.

[Pop.72] Popper, K. R. 1972, Objective Knowledge: An Evolutionary Approach. Oxford: Claren-

don Press.

[Pen.89] Penrose, R. 1989, The emperor’s new mind, Oxford Un. Press.

[Pen.05] Penrose, R. 2005, The Road to Reality: A Complete Guide to the Laws of the Universe.
New York: Alfred A. Knoff.

[Pos.63] Post, H. 1963, ’Individuality and Physics’, The Listener, 10 october 1963, 534-7,
reprinted in Vedanta for East and West 132, 1973, 14-22.

[Prz.69] Przelecki, M. 1969, The Logic of Empirical Theories, London, Routledge & Kegan-Paul
(Monographs in Modern Logic Series).

[Pug.87] Pugorelov, A. 1987, Geometry. Moscow: Mir.

[RubRub.70] Rubin, H and Rubin, J. E. 1970, Equivalents of the Axiom of Choice. Amsterdam
and London: North-Holland, 2nd. printing.

[Qui.86] Quine, W. V. 1986, Philosophy of Logic, Harvard Un. Press, 2nd.ed.

[Red.87] Redhead, M. 1987, Incompleteness, Nonlocality and Realism: A Prolegomenon to the

Philosophy of Quantum Mechanics, Oxford, Clarendon Press, 1987.

[RedTel.91] Redhead,M. and Teller, P. 1991, ‘Particles, particle labels, and quanta: the toll of
unacknowledged metaphysics’, Foundations of Physics 21, 43-62.

[RosWan.50] Rosser, B. and Wang, H. 1950, ‘Non-standard models of formal logics’, Journal of
Symbolic Logic 15 (2), 113-129.

[Sau.98] Sauer, T. 1998, The relativity of discovery: Hilbert’s first note on the foundations of
physics. In arXiv:physics/9811050v1

[Sch.95] Schrödinger, E. 1995, The Interpretation of Quantum Mechanics: Dublin

Seminars(1949-1955) and Other Unpublished Essays, Ox Bow Press.

[Sha.91] Shapiro, S., 1991, Foundations without foundationalism, a case for second-order logic,
Clarendon Press, Oxford.

[Sho.67] Shoenfield, J. R. 1967, Mathematical Logic, Reading, Addison-Wesley, 1967 (reprinted

by the Association of Symbolic Logic, 2000).

[Sho.77] Shoenfield, J. R. 1977, ‘Axioms of set theory’, in Barwise, J. 1977, Handbook of Math-
ematical Logic, North-Holland, pp. 321-344.
BIBLIOGRAPHY 183

[Sim.87] Simons. P. 1987, Parts: A Study in Ontology. Oxford University Press.

[Sko.22] Skolem, T. 1922. Some remarks on axiomatized set theory. Reproduced in J. van Hei-
jenoort (ed.) 1967. From Frege to Gödel, Harvard University Press: 290-301.

[Smu.91] Smullyan, R. 1991, Gödel’s Incompleteness Theorems, Oxford Univ. Press.

[Steg.79] Stegmüller, W. 1979, The Structuralist View of Theories. A Possible Analogue of Bour-
baki Programme in Physical Sciences. Berlin and Heidelberg: Springer-Verlag.

[Sty.02] Styer, D. F. et al. 2002. Nine formulations of quantum mechanics. Am. J. Physics 70
(3): 288-97.

[Sup.77] Suppe, F. (ed.) 1977, The Structure of Scientific Theories, Urbana & Chicago, Chicago
Un. Press, 2a. ed.

[Sup.00] Suppe, F., 2000, Understanding scientific theories: an assesment of developments, 1969-
1998. Philosophy of science 67, Supplement, pp. S102-S115.

[Sups.57] Suppes, P. 1957, Introduction to Logic, Van Nostrand.

[Sups.60] Suppes, P. 1960, A Comparison of the Meaning and Uses of Models in Mathematics
and the Natural Sciences, Synthese 12: 287-301.

[Sups.62] Suppes, P. 1962, Models of Data. In E. Nagel, P. Suppes, and A. Tarski (eds.),
Logic, Methodology, and Philosophy of Science: Proceedings of the 1960 International
Congress. Stanford: Stanford Un. Press, pp. 252-261.

[Sups.67] Suppes, P. 1967, What is a scientific theory? In Morgenbesser, S. (ed.), Philosophy of

Science Today, New York: Basic Books, pp. 55-67.

[Sups.77] Suppes, P. 1977, The structure of theories and the analysis of data. In Suppe, F. (ed.),
The Structure of Scientific Theories. 2nd. ed. Urbana and Chicago: Un. of Illinois, pp.
266-307.

[Sups.83] Suppes, P. 1983, Heuristics and the axiomatic method. In Groner, I., Groner, M. and
Bischof, W. F. (eds.), Methods of Heuristics. Hillsdale, N.J.: Erlbaum, pp. 79-88.

[Sups.88] Suppes, P. 1988, ‘La estructura de las teorias y el analysis de datos’, in Suppes, P.,
Estudios de Filosofía y Medotologí a de la Ciencia, Madrid, Alianza, pp. 125-145.

[Sups.02] Suppes, P. 2002, Representation and Invariance of Scientific Structures. Stanford: CLI
Pu.

[Sups.11] Suppes, P. 2011, Future developments of scientific structures closer to experiment: Re-
sponse to F. A. Muller. Synthese 183 (1): 115-23.
184 BIBLIOGRAPHY

[SylCos.87] Sylvan, R. and da Costa, N.C.A. 1987, Cause as an implication, Studia Logica 47
(4), 413-428.

[Sza.64] Szabó, Á. [1964], The transformation of mathematics into deductive science and the
beginnings of its foundation on definitions and axioms. Scripta Mathematica 27 (1): 27-
49.

[Sza.65] Szabó, Á. 1965, Greek dialectic and Euclid’s axiomatics. In Lakatos, I. (ed.), Problems
in the Philosophy of Mathematics Proceedings of the International Colloquium in the Phi-
losophy of Science, Vol. 1. London: North-Holland: 1-26.

[Sza.78] Szabó, Á. 1978, The Beginnings of Greek Mathematics. Dordrecht: D. Reidel (Synthese
Historical Library, Vol.17).

[Tar.53] Tarski, A. 1953, Undecidable Theories, Amsterdam, North-Holland.

[TarGiv.87] Tarski, A. and Givant, S. 1987, A Formalization of Set Theory without Variables,
American Mathematical Society.

[Teg.07] Tegmark, M. 2007, Shut up and calculate, arXiv:0709.4024v1

[Tor.81] Toraldo di Francia, G. 1981, The Investigation of the Physical World, Cambridge: Cam-
bridge Un. Press.

[Tor.86] Toraldo di Francia, G. 1986, Le cose e i loro nomi, Bari, Laterza.

[True.84] Truesdell, C. 1984, An Idiot’s Fugitive Essay on Science: Methods, Criticism, Train-
ing, Circumstances. New York: Springer-Verlag.

[vanF.80] van Fraassen, B. 1980, The Scientific Image, Oxford, Clarendon Press.

[vanF.84] van Fraassen, B. 1984. The problem of indistinguishable particles. Reprinted in E.

Castellani (ed.) 1998. Interpreting bodies: classical and quantum objects in modern
physics, Princeton University Press, pp.73-92.

[vanF.89] van Fraassen, B. 1989, Laws and Symmetry. Oxford, Clarendon Press.

[vanF.91] van Fraassen, B. 1991, Quantum Mechanics: An Empiricist View. Oxford, Clarendon
Press.

[vanF.97] van Fraassen, B. 1997, Structure and perspective: philosophical perplexity and para-
dox. In Dalla Chiara, M. L. et al. (eds.), Logic and Scientific Method. Dordrecht, Kluwer.

[vanF.08] van Fraassen, B. 2008, Scientific Representation, Oxford, Oxford Un. Press.

[vanH.67] van Heijenoort, J. 1967, From Frege to Gödel: A Source Book in Mathematical Logic,
1879-1931.
BIBLIOGRAPHY 185

[Wei.03] Weingartner, P. (ed.) 2003, Alternative Logics: Do Sciences Need Them?. Berlin and
Heidelberg: Springer-Verlag.

[Will.70] Williams, M. B. 1970, Deducing the consequences of evolution: a mathematical model.

J. Theoret. Biol. 29: 343-385.

[WhiRus.10] Whitehead, a. N. & Russell, B. 1910, Principia Mathematica, Vol. 1. Merchanf

Books.

[Wor.07] Worral, J 2007, “Miracles and Models: Why Reports on the Death of Structural Real-
ism May be Exaggerated”, in O’Hare, A. (ed.) Philosophy of Science (Royal Institute of
Pilosophy 61), Cambridge: Cambridge University Press pp. 125-154.

[Yan.02] Yandell, B. 2002, The Honors Class: Hilbert’s Problems and Their Solvers. Natick,
MA: A.K.Peters.

[Zah.04] Zahar, E. 2004, Ramseyfication and Structural Realism, Theoria 49, pp. 5-30.

[Zer.67] Zermelo, E. 1967/2000, Investigations in the foundations of set theory. In van Heijenoort
(ed.), FromFrege to Gödel: a Source Book in Mathematical Logic, 1879-1931. S. Jose,
New York, Lincoln, Shangai: toExcel.
Index

KDn , 113 as Urelemente, 38

Rι , 114 attribution of numbers to things, 21
T, 111 auxiliary constant, 129
Lω (rng(Rι )), 124 axiom
Lωωκ (R), 120 of choice, 89, 108, 127, 165
Lωωω (R), 120 of choice, denumerable version, 108
Lωωω (rng(Rι )) , 124 independence of, 73
Lωω1 ω (R), 120, 121 of choice, 127
L2 , 59, 65 in quantum mechanics, 146
Lnωω , 156 in ZFU, 82
L1ωω , 120 independence of, 65
E |= S , 122 of choice, formulation, 40, 73
Ord(t), 112 of elementary sets, 39
rng(Rι ), 124 of extensionality, 39, 69
ε({Dn }), 113 of infinity, 41
{x : α(x)}, 67 of regularity, or foundation, 72
s-algebra, 117 of separation, 39
of the infinity, 71
abbreviations of the power set, 40, 70
in the metalanguage of ZF(C), 66 of union, 40
absolute concepts, 167 of unordered pair, 69
algebras pair, 33
free, 118 model of, 79
homomorphism between, 117 power set, 33
isomorphism between, 117 replacement axioms
AM, see axiomatic method, see axiomatic and the sciences, 108
method separation, 39
analyticity, 8 separation schema, 70
Aristotle, 30, 32, 89, 100 substitution, 33
Arnol’d, V., 94 substitution schema, 71
atoms union, 70
as synonymous of Urelemente, 81 well-order, 127

186
INDEX 187

axiom of comprehension for classes, 67 The Architecture of Mathematics, 42

axiomatic method, v, vi, 85, 88, 90, 93–95, Bueno, O., 164
97–99, 101, 105
as an autopilot, 63 Cantor, G., 85
in biology, 98 definition of set, 38, 151
origins, v, 29 diagonal argument, 85
axiomatic theories, 87 cardinal
axiomatization, 105 strong inaccessible, 74
abstract, v, 35, 41 Carnap, R., ii, 3, 5, 7, 10–13, 26, 27, 59,
and heuristics, 57 178
as the organization of a field, 35 category theory, 59, 89, 110
concrete, 35 Cauchy sequences, 84
external, 105, 106 Cauchy, A.L.
extreinsic, 138 sequences, 47
finitely axiomatizable, 139 Chuaqui, R., vi, vii, 106, 109, 111, 115,
formal, v, 35, 48 131, 133, 138, 139, 144, 155, 164
inside set theory, 139 on the semantic approach, 162
internal, 106 Church thesis, 99
intrinsic, 138 Church, A., 62, 99
intuitive, 35 class, 67
non-relativistic quantum mechanics, 146 class term, 67
of classical particle mechanics, 109 classical particle mechanics, vi, 90
pure formal, 35 Suppes predicate for, 134
role of set theory, 107 classical propositional logic, 118
styles of, 24 Coelho, A.M.N., 143, 167
why?, 29 Cohen, P
axiomatization and heuristics, 87, 88 independence of the axiom of choice,
axioms 65
scheme of, 40 Cohen, P., 57, 167
collapse postulate, 150
Béziau, J.-Y., 60 condition, 67
Balzer, W., 92 consistency
believing a theory, 25 absolute, 77
Bohr, N., 169 of ZF(C), 75
Bolyai-Lobachewski geometry, 32 relative, 76
Borel sets, 147, 170 semantic, 77
Bourbaki, N., ii, 86, 94, 117, 129, 141, 142 syntactical, 77
echelon construction schema, 113 Contessa, G., 176
species of structures, iii, 143 continuum hypothesis, 127, 167
188 INDEX

continuum mechanics, 35, 90 Euclid, 31, 32, 94

convergence, 47 Euclid’s parallel postulate, 73
correspondence rules, 6–8, 25, 106 Euclidean geometry, 32, 36
CPM, see classical particle mechanics Euclidean space, 135
criterion of eliminability, 128 Eudoxus, 30
criterion of non-criativity, 128 explicandum, 11
cut rule explicatum, 12
formal proof, 55 extension by definitions, 110
informal proof, 55 extraordinary sets, 73

da Costa, N.C.A., iii, vi, vii, 24, 60, 62, 63, Feferman, S., 127
85, 106, 109, 111, 114, 115, 123, Feyerabend, P., 7
131, 133, 137–139, 142–144, 155, finitistic reasoning, 63
164 firs-order languages, 10
on the semantic approach, 162 first-order languages, 9, 10
principle of constructivity, 60 first-order logic, 88, 100
Darwin, C., 41, 98, 100 axioms, 51
Davies, P., 130 formulation, 49
DC force
the axiom of dependent choices, 165 electromagnetic, 96
Dedekind cuts, 84 gravitation, 96
Dedekind, R., 123 strong, 96
deduction, 108 weak, 96
definability, vii, 124 formal language, 14
definable in wide sense, 164 formalization
definit property, 39 advantages of, 163
definite property, 39, 40 formulas
derived concepts, 30 of a first-order language, 50
Descartes, 60 of the language of ZF(C), 65
distance, 47 Fraenkel, A.A., 39
dynamic postulate, 150 the elimination of the Urelemente, 80
free algebra, 118
echelon construction schema, 113 Frege, G.
electroweak theory, 96 polemic with Hilbert, 33
elementary class of structures, 139 French, S., 24, 27, 101, 157, 161
elementary equivalence, 143
empirical adequacy, 162 Gödel, K.
empiricists dogma of reduction, 8 proper classes, 68
emptyset, 39 second incompleteness theorem, 79
epistemic structural realism, 107 Galilei, G., 98
INDEX 189

gauge theories, 96 the 6th Problem, 34

general relativity, 97
Gentzen, G., 55 inaccessible cardinal, 79, 120
Giere, R., iii inaccessible cardinals, 74, 89, 120
Grandjot, C., 55 individual
Gross, D., 96 the notion of, iii
Grothendieck, A. inference
universes, 89 abductive, 61
group theory, 43, 96, 110, 116 inductive, 61
basic structure, 42 informal mathematics, 87
different presentations, 20 inner product, 46
language of, 134 intended model of ZF(C), 75
structure for, 133 interpretation, 14
Suppes predicate for, 133, 134 interpretation function, 16
Gödel, K., 57, 95, 102, 167 intuition, 61
consistence of the axiom of choice, 65 intelectual, 62
incompleteness theorems, 48 intuitive arithmetics, 63
second incompleteness theorem, 75, 77 left behind, 33
not enough, 61
Halvorson, H., 11, 12, 19–23, 26, 98, 179 of entities, 62
Hamel, G., 92 of the number two, 64
Hamiltonian mechanics, 20 intuitionistic arithmetics, 60
Hanson, N.R., 7 intuitionistic mathematics, 62
Heisenberg, W., 20, 130 intuitive pragmatic nucleus, 62
Helium atom, 130, 131 isomorphism, 117
Hempel, C.G., 3, 5, 97, 99 definition of, 22
Hermitan operator, 171 isomorphisms, 21
Hermitean operators, 166 between structures, 22
of theories, 23
heuristic axiomatization, 88
hidden variables, 152 Jafee, A., 96
higher-order concepts, 10
higher-order languages, 14, 59 Kepler, J., 98
Hilbert, D., 32, 63, 87, 94 knowledge
23 Problems of Mathematics, 33 conceptual, i
on axiomatization, 35 structural, i
polemic with Frege, 33 Kolmogorov, A.N., 102
rigged Hilbert spaces, 151 Kolmogorov, N., 88
spaces, 43, 44, 46, 146, 166 Kuhn, T., 7
spaces (definition), 48 Kunen, K., 63, 64
190 INDEX

Löwenheim-Skolem logic
downward, 84 complimentary, 49
upwards, 84, 85 deduction symbol, 52
downwards, 85 first-order, 49
Lacan, J., i formal theorem, 53
Ladrière, J., 35 heterodox, 49
Lagrangean mechanics, 20 in abstract axiomatics, 44
Lakatos, I., 86, 87, 94, 99 must be developed twice, 64
language thesis of, 53
independence of, 15, 16 logical consequence, 159
language independence, 25 logical empiricism, 1, 26
language independent, 22 logical positivists, 98
languages, 115 logical-empiricist program, 1
as algebras, 118 Lutz, S., 10, 11, 13, 14, 23, 25, 26, 156,
constructed in ZFC, 119 180, 181
denumerable, 65
finitary, 156 Mackey, G., 88, 102
first-order, 10, 15, 156 Manin, Yu.I., 151
for a structure, 120 mathematics
higher-order, 120 intuitionistic, 95
infinitary, 119, 120, 164 non-cantorian, 95
of a structure, 122, 162 paraconsistent, 95
of group theory, 134 McKinsey, J.C.C., 90
of order ω, 120 answer to Truesdell, 92
of order n, 157 membership relation, 38
of order n, 1  n < ω, 120 mereology, 83
of quantum mechanics, 151 metalanguage, 159
order, 156 metamathematics, 165
order of, 120 metametatheory, 64
order-ω, 156 metatheory, 63
second-order, 156 metavariables, 66
Leśniewicz’s conditions, 128, 129 method of the auxiliary constant, 129
least upper bound, 9 Mirimanov, D., 73, 102
Lebesgue, H. model
measure in the sense of, 166 intended, 37
Lesniewicz, S., 83 of a formal theory, 76
limit ordeinal, 77 of a theory, 139
linear spaces, see vector spaces qualitative, 21
linguistic formulation, 10 the meanings of the word, 155
INDEX 191

two notions of, 18 order of a relation, 112

model revolution, 3 ordinal
model theory, 1, 14, 15, 156 imit, 79
modeling the phenomena, 21
models, 143 paraconsistent set theory, 68
as models of something, 16 partial truth, 124
as set-theoretical structures, 18 Pauli, W.
class of, 106 exclusion principle, 152
classes of, 20 Peano, G,
first-order, 15 postulates, 160
inside and outside of, 172 Peano, G., 88
numerical, 21 arithmetics, 37, 54
of scientific theories, 155 axioms, 37, 54
of theories, 21 axioms in ZFC, 142
of ZF(C), 78 non-standard models, 48
theories as classes of, 19 Peirce, C.S., 61
modern logic, 32 Pieri, M., 32, 36
modern mathematics Popper, K.R., 86
the rise of, 42 postulates
morphisms, 21 independence of, 33
Muller, F.A., 6, 19, 20, 131 of the theory of human paternity, 36
multisets, 38 postulates and axioms, 30
postulates of a theory, 107
NBG set theory, 68, 108, 161 logical, 107
new symbols mathematical, 108
introduction of, 127 specific, 108
Newton, I., 98 precisification, 12
three laws, 33 primitive concepts, 30
Noll, W., 93 primitive notions, 30
nominal definitions, 85 principle of constructivity, 60
non-euclidean geometries, 89 probability theory, 107
non-individuals, iii proof theory, 57
non-standard models, 88 proofs and refutations, 87
norm proper classes, 68, 161
induced by the inner product, 46 proper subset, 70

objects QCD, quantum chromodynamics, 96

definability, 125 QED, quantum electrodynamics, 96
objectuation, 130 quantu mechanics
order of a language, 120 Copenhague interpretation, 130
192 INDEX

quantum field theory, 97 Russell’s set, 67

quantum gravity, 97
quantum mechanics, 20, 43, 46, 48, 165, Saunders, S., 131
167 scales
and Skolem paradox, 168 definition of, 113
as an indeterministic theory, 168 schema of axioms, 68
groups in, 43 Schrödinger, E., 20
postulates of, 148 scientific activity, i
symmetry conditions, 172 scientific practice, 3, 9
wave function, 130 scientific theories, 1, 2, 25, 26, 85, 105, 155
quantum semantics, iii characterization of, 162
quasi-set theory, 152, 173 Seevinck, M.P., 131
quasi-sets, iii semantic approach, iv, 1, 4, 11, 14, 18, 19,
quasi-truth, 171 25–27, 155, 163
advantages of, 16
Ramsey, F. P., 107 and models, 15
real numbers, 101, 107 characteristics, 15
field of, 9 sentence, 67
not denumerable, 126 set
Received View, 1–6, 8–14, 19–21, 26, 89, not a relative notion, 40
105–108, 138, 139 set theory
as an ideal, 26 and mathematical concepts, 109
characterization, 5 as a higher-order theory, 108
failure of, 10 in axiomatization, 107
inadequacies of, 25 informal, iii
Redhead, M., 148 models of, 75, 109
reduction to the absurd, 30 paraconsistent, 68
relations paradoxes, 41
definability, 124 the need of, ii
order of, 112 the well-founded model, 111
partial, 124 set-theoretical predicate, 158
relative concepts, 167 set-theoretical structures, 15
relativization of a formula, 78 sets
restrict quantifiers, 81 in Zermelo’s set theory, 39
restriction of an operation, 45 relative to a theory, 68
Rodrigues, A.A.M., 164 Shapiro, S., 123
Rule C, 129 Sierpinski, W., 167
Russell’s class, 67 simple theory of types, 111
Russell, B., 57 Skolem, T., 39, 80, 167, 172
INDEX 193

paradox of, 85 146

Solovay, R., 165 Suppes, P., ii, iii, vi, vii, 4, 14, 17, 21, 23,
states and events, 171 24, 26, 27, 57, 88–90, 96, 101, 105–
Stegmüller, W., 92 107, 109, 116, 131, 133, 134, 137–
structural knowledge, 62 144, 146, 155
structures set-theoretical predicate, 83
automorphims, 173 and his approach, 138
automorphisms of, 164 and the need of models of another kind,
classical particle mechanics, 115 161
definition of, 114 and the origins of the axiomatic method,
equivalence 30
necessary and sufficient condition for, and the semantic approach, 158
164 mathematics and not metamathematics,
equivalence of, 164 160
fields, 116 model revolution, 4
first-order, 16 on psychology, 34
for empirical sciences, 110 predicate, 17, 18, 24, 116
for the language of ZF(C), v, 75 proofs inside set theory, 160
Galois group, 164 set-theoretical predicate, 159
hierarchy of, 21 the role of axioms, 161
higher-order, 157 theory of human paternity, 36, 53
order, 156 theory of paternity, 36
order–1, 110 syntactic approach, 8, 26, 27, 163
order-1, 156, 157 syntactical approach, iv, 1, 5, 13
rigid, 173 characteristics, 14
set theoretical, 110 syntactical approaches, 5, 8, 10, 11, 19, 28
similarity type syntax-semantic debate, iv, 1, 15
signature, 116 Szabó, A., 31, 32
to organize knowledge, v, 61 the origins of the axiomatic method, 30
truth in, vii, 122
vector spaces, 115 Tarski, A., vi, vii, 1, 23, 57, 102, 122, 124,
subset, 39, 40, 70 125, 143, 156
subspace, 46 and model theory, 14
subspaces as the precursor of the semantic view,
closed, 48 19
Sugar, A.C., 90 model in the sense of, 15
Suppe, F., 1, 2, 5, 10–12, 97 models in the sense of, 17, 18, 156, 158
Suppes predicate on semantics, 162
for non-relativistic quantum mechanics, satisfaction and truth, 159
194 INDEX

undefinability of truth, 48 theory equivalence, 163

terms theory of groups, 17
observational, 5, 6, 108 topological space, 18, 157
of a first-order language, 50 Toraldo di Francia, G., 130
of the language of ZF(C), 65 transfinite cardinals, 62
theoretical, 5, 7, 108 transfinite recursion, 77
the language of ZFU, 81 transitive class, 78
theorems, 31 Truesdell, C., 90, 92, 93
theoretical equivalence, 14, 20 Turing, A., 99
theoretical postulates, 5 type theory, 10, 157
theoretical terms, 26 types
theoretical vocabulary, 5 definition of, 111
theories, 25 order of, 112
as classes of models, 25
ultraproducts, 143
axiomatization, 105
unbounded operators, 165
axiomatization of, 28
unheuristic axiomatization, 88
elementary equivalent, 23
unitary set, 39
equivalence of, 164
universal algebra, 116
extrinsic approach, 24
universes, 74, 89, 110
intrinsic approach, 24
ur-elements, see Urelemente
representation of, 27
Urelemente, 65, 80, 108, 159, 166
theory
and atoms in set theory, 38
as a class of models, 15
elimination of, 80
and experience, 6, 8
and experiment, 27 validity, 122
and reality, 21 van Fraassen, B., iii, 3, 10, 16, 21, 25, 27,
as a class of models, v 168, 170–172
axiomatization, 9, 106 dynamical vs. experimental, 169
categorical, 79 empirical adequacy, 162
characterization of, iv, 11 modal interpretation, 167
extrinsic characterization, 138 state and value attributions, 169
inconsistent, ii vector spaces
individuation of, 6 definition, 44
intrinsic characterization, 138 models of, 158
linguistic formulation, v Vienna Circle, 1, 3, 5, 6
logical postulates, 107 vocabulary
mathematical postulates, 108 theoretical and observational, 106
specific postulates, 108 von Neumann universe, 77
three levels of postulates, 107 von Neumann, J., 170
INDEX 195

wave-function what it is, 64

collapse of, 130 ZFU, 80
wave-functions, 150 equiconsistent with ZF(C), 83
well-founded model of ZF(C), 77
Weyl, H., 43
Whitehead, A.N., 57
Williams, M., 98
Woodger, J., 34, 98
Worrall, J., 107

Z p , 108
Z p 79
Zahar, E., 107, 108
Zeno of Elea, 30
Zermelo’s set theory, 79
Zermelo, E., 38, 59
domain of objects, 38
natural numbers, 41
on the limitation of size, 70
set theory, 88
Zermelo sets, 40
Zermelo-Fraenkel set theory, see ZF, see ZF,
see ZF
Zermelo-Fraenkel set theory with the Ax-
iom of Choice, see ZFC
ZF, 60
what it is, 64
ZF(C), see ZF, ZFC, 59, 63, 64, 79
formulation as a formal theory, 64
logical postulates, 68
models of, 76
standard model, 84
the language of, 65
the postulates of, 68
the theories, 59
ZFC, 64, 108, 139
as a second-order theory, 59
being used, 110
denumerable model, 85