You are on page 1of 121

Representing Spacetime Points as Classes of

Mereotopologically Structured Basic Entities.

DIPLOMARBEIT
zur Erlangung des Magistergrades
an der Kultur- und Gesellschaftswissenschaftlichen Fakultät
der Universität Salzburg
Fachbereich Philosophie
Gutacher: Ao.Univ.-Prof. Mag. Dr. Alexander Hieke

eingereicht von
LAURENZ HUDETZ

Salzburg 2014
CONTENTS

1 Introduction 1
1.1 Problem & Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Proposed Definitions and Their Problems 11


2.1 Overview and Classification of the Definitions
Proposed until Now . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Adequacy Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Exclusion and Selection of Definitions . . . . . . . . . . . . . . . . . . . . . . . 19

3 A Unified Framework for Analysing Different Representation Methods 20


3.1 Preliminaries on Topology and Mereology . . . . . . . . . . . . . . . . . . . . . 20
3.1.1 Preliminaries on Relational Structures . . . . . . . . . . . . . . . . . . . 20
3.1.2 Preliminaries on Topological Spaces . . . . . . . . . . . . . . . . . . . . 22
3.1.3 Preliminaries on Mereological Structures . . . . . . . . . . . . . . . . . . 25
3.1.4 Preliminaries on Mereotopological Structures . . . . . . . . . . . . . . . 30
3.2 A General Theory of Point Representations . . . . . . . . . . . . . . . . . . . . 39

4 Analysing the Most Important Families of


Point Representations 45
4.1 Representing Points by Ultrafilters . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.1.1 Stone’s Theorem for Classical Mereological Structures . . . . . . . . . . 48
4.1.2 Extending Stone’s Theorem for
Mereotopological Structures . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.1.3 The Problem of Free Ultrafilters . . . . . . . . . . . . . . . . . . . . . . 56
4.2 Representing Points by Completely Prime Filters . . . . . . . . . . . . . . . . . 60
4.2.1 Advantages of Completely Prime Filters . . . . . . . . . . . . . . . . . . 61
4.2.2 The Problem of Prime Elements . . . . . . . . . . . . . . . . . . . . . . 65
4.3 Representing Points by Maximal Round Filters . . . . . . . . . . . . . . . . . . 70
4.3.1 Problems with Atomistic Structures . . . . . . . . . . . . . . . . . . . . 82
4.3.2 General Adequacy of the Round Filter Representation . . . . . . . . . . 84
4.3.3 The Open Problem of Axiomatisation . . . . . . . . . . . . . . . . . . . 98

5 Metaphysical Questions 100


5.1 The Question of Realism About Abstract Entities . . . . . . . . . . . . . . . . . 100
5.2 The Question of Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.3 Other Questions Concerning Ontological Status . . . . . . . . . . . . . . . . . . 108

6 Summary 113
6.1 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.2 Open Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

Bibliography 114
CHAPTER 1

INTRODUCTION

1.1 Problem & Motivation


This thesis deals with the question what spacetime points are. The concept1 of spacetime
point plays an important theoretical role in our reasoning about the spatial and temporal
structure of the world, both in physics and philosophy. Yet, it is far from clear what exactly
spacetime points are.
An answer to the question what spacetime points are can be given by stating a definition.
But why should the predicate ‘spacetime point’ be defined at all? Should it not better be left
undefined? There are arguments for both sides. First we examine which case can be made
against defining the predicate ‘spacetime point’. After that we turn to arguments in favour of
defining ‘spacetime point’.

Objection 1 (a definition is useless for common scientific practice). A possible objection


against defining the predicate ‘spacetime point’ that is likely to be offered by scientific anti-
realists or working physicists and mathematicians is the following. One should not define the
predicate ‘spacetime point’ for the reason that (a) doing so is neither required nor useful for
obtaining empirically adequate results (e.g. predictions and explanations for observable phe-
nomena) by means of the mathematical apparatus of modern physics, and (b) since obtaining
empirically adequate results is all we need, we should not define what need not be defined to
obtain empirically adequate results (for the sake of generality and parsimony).2
However, even if we concede that a definition of ‘spacetime point’ is not required for
1
In this thesis, I use the words ‘concept’ and ‘notion’ in the sense of ‘predicate’ or ‘meaningful predicate’.
Moreover, I use expressions such as ‘the concept of spacetime point’, ‘the notion of spacetime point’ and ‘the
predicate ‘spacetime point’ in the same sense.
2
Here is another version of this argument, in which the philosophical term ‘empirically adequate’ does not
occur and which is therefore closer to the diction of working physicists and mathematicians: defining ‘spacetime
point’ is of no use for common scientific practice. If it is of no use for common scientific practice, then we
should not waste time with it. So we should not waste time with defining ‘spacetime point’.

1
CHAPTER 1. INTRODUCTION 2

common (experimental) scientific practice, the above objection does not succeed in showing
that the predicate ‘spacetime point’ should not be defined. Objections of the kind in question
rest (more or less explicitly) on an anti-realist understanding of scientific theories. They
presuppose that the only purpose of scientific theories consists in yielding empirically adequate
results. This view in turn is connected with the idea that we do not have to take theoretical
concepts semantically and ontologically seriously (cf. Chakravartty, 2011).
A main problem of the objection in question is simply that it is far from clear that anti-
realism is the right view. If one adopts scientific realism, then assumption (b) is not tenable.
According to a realist understanding of scientific theories, it is not the case that obtaining em-
pirically adequate results is the only purpose of scientific theories. From a realist standpoint,
there is a mind-independent reality that has observable as well as unobservable aspects, and
scientific theories aim at describing what this reality is like (cf. van Fraassen, 1980, p. 8). So
according to scientific realism theoretical concepts should be taken semantically and ontolog-
ically seriously. In this case, the claim that we should not define what need not be defined to
obtain empirically adequate results is no longer tenable.
There may be other reasons for defining theoretical concepts (in terms of other theoretical
concepts). For example there may be philosophical reasons for explicating theoretical concepts
such as the predicate ‘spacetime point’. So this objection can only be successful if anti-realism
is assumed beforehand. However, it is not even clear that all anti-realists should refrain from
defining the predicate ‘spacetime point’. The assumption of anti-realism alone is not sufficient
to render the argument convincing. It is also necessary to accepted the following premise: if
obtaining empirically adequate results is all we need from scientific theories, then we should
not define what need not be defined to obtain empirically adequate results from scientific
theories. But even anti-realists can have reservations about that claim. It cannot be ruled out
that there are other reasons for defining the predicate ‘spacetime point’, which are acceptable
or even attractive for anti-realists. Reducing the theoretical concept of spacetime point in
terms of concepts that apply to (classes of) empirically accessible entities could be such a
reason (see motivation 1 below).
We can conclude that the first objection does not show in general that the predicate
‘spacetime point’ should not be defined.

Objection 2 (a definition is useless, because structure is all that matters). A second possible
objection is based on the view that scientific theories describe only structural aspects of reality,
but say nothing about the individual constituents of reality. Views of that kind are subsumed
under the label ‘structural realism’. There are different kinds of structural realism ranging from
radical (as advocated by French and Ladyman) to moderate (as advocated by Esfeld and Lam).
For an overview see Ladyman (2014). The case against defining the notion of spacetime point
CHAPTER 1. INTRODUCTION 3

requires only a moderate version of structural realism. Moderate structural realism accepts
that there are spacetime points, but holds that nothing can be said about spacetime points
apart from how they are related to each other or, in other words, which structure they form:
“all there is to them are the relations in which they stand.” (Esfeld & Lam, 2008, p. 34) So if all
we can say about spacetime points is how they relate to each other, then we can express all that
can be said about spacetime points by taking the predicate ‘spactime point’ (alongside with
topological and geometrical relations) as a primitive and by formulating suitable structural
principles which describe how spacetime points relate to each other. Then all we can say
about spacetime points is a consequence of these principles. Defining ‘spacetime point’ in
a way that more than the consequences of these purely structural principles can be derived
would go too far. We would go beyond what we can actually say about spacetime points.
So the predicate ‘spacetime point’ needs not and should not be defined, just as the predicate
‘point’ in Euclidean geometry needs not and should not be defined.
The view is indeed closely analogous to the predominant view in the foundations of Eu-
clidean geometry. In modern axiomatisations of Euclidean geometry since Hilbert’s, the pred-
icate ‘point’ is alway left undefined. It does not matter for the mathematical theory what kind
of entities points are. All that Euclidean geometry can say about points (lines and planes)
is how they relate to each other.3 The value of this approach lies in the fact that it allows a
high level of abstraction: the theory is not restricted to a particular domain of application.4
In contrast to that, Euclid himself defined the notion of point. According to the very first
definition in Euclid’s Elements, a point is something which has no parts. This definition is
considered as superfluous nowadays, because it is never used in Euclid’s proofs. Hence, the
definition has no use from a mathematical point of view. Since geometry is only concerned
with the study of certain structures given by transformation groups (cf. Marguis, 2009, p. 26),
defining the predicate ‘point’ is regarded as an unnecessary restriction. It artificially limits
the theory’s domain of application.5 Along these lines Suppes (2002, p. 36) remarks:

“He [Euclid] does not seem to see clearly that the axiomatic development of ge-
ometry must begin with some primitive ideas which are not themselves defined in
terms of others. Perhaps it is fair to say that in this respect he confuses formal
or axiomatic questions with problems concerning the application of geometry. For
example, Definition 1 of Book I asserts that a point is that which has no parts.”
3
Think of Hilbert’s famous remark that it should be possible to say ‘table’, ‘chair’ and ‘beer mug’ instead
of ‘point’, ‘line’ and ‘plane’.
4
Example: Since the days of Descartes it is well known that a certain numerical structure with R3 as domain
satisfies the axioms of Euclidean geometry. Although this is not the interpretation originally intended by the
Greeks, it is of great benefit, because it allows to solve geometrical problems by algebraic and analytic means.
5
The underlying idea is that the domain of a mathematical theory should not be unnecessarily restricted,
in order to allow as many applications as possible.
CHAPTER 1. INTRODUCTION 4

These considerations give rise to the following argument. Euclidean geometry is only about
structure and physics is only about structure (they are analogous in this respect). The predi-
cate ‘point’, which is contained in the vocabulary of Euclidean geometry, should not be defined,
because geometry is only about structure. Therefore, the predicate ‘spacetime point’, which
is contained in the vocabulary of physics, should also not be defined, because physics is only
about structure too.
A problem of the argument above is that the analogy between spacetime physics and
mathematical theories is not perfect. Even if moderate structural realism is taken for granted,
there is a difference between the way in which mathematical theories are only about structure
and the way in which physical theories are only about structure. Mathematical theories are
only about (abstract) structures in a way that it would be problematic if their domains of
application were limited to some paricular kinds of objects. But physical theories are primarily
about reality or, more specifically, about the structures of the physical world (according to
structural realism). Physical theories always have an intended domain of application consisting
of objects in the physical world that surrounds us. It is not problematic if their domains of
application are limited to their intended domain. Now, if a definition of a previously undefined
notion can be found that singles out the right objects belonging to the intended domain, then
it is not really problematic to introduce such a definition.
Moreover, one can also doubt the premise that the predicate ‘point’, which is contained
in the vocabulary of Euclidean geometry, should not be defined, because geometry is only
about structure. If it is possible to give an equivalent axiomatisation of geometry taking other
notions as primitive and defining the notion of point on that basis, then the above mentioned
premise is wrong, because equivalent axiom systems characterise the same structures. Why
should it not be possible to characterise the kind of structures that a certain geometrical theory
is about by means of a different choice of primitives? In fact, it has been proved that this
is possible for specific topological theories, see for example Grzegorczyk (1960). Analogously,
it is possible to use different primitives in physical theories. One possibility is for instance
Benda’s axiomatisation using world-lines as primitive: “Primitive objects of the present theory
are world-lines rather than spacetime points. [. . .] Sets of intersecting worldlines will later
define spacetime points.” (Benda, 2008, p. 448). Regardless of whether such a definition is
plausible, this shows that the same class of structures can be characterised by means of different
primitives and that the predicate ‘spacetime point’ indeed can be defined on the basis of these
primitives. This refutes not only the argument from the analogy between physical theories and
Euclidean geometry, it annuls all structuralist objections against defining ‘spacetime point’.
So we can conclude that objection 2 is not convincing either. It does not show that the
notion of spacetime point should not be defined.
CHAPTER 1. INTRODUCTION 5

Let us now examine the most important arguments that can be given in favour of defining the
notion of spacetime point.

Motivation 1 (a definition is needed if one seeks to dispense with points). A motivation for
defining the predicate ‘spacetime point’ arises from reductionist programs in the philosophy
of spacetime. In the last century, different philosophers argued (for different reasons) that we
should dispense with the assumption that there “really” are spacetime points, i.e. spacetime
points as independently existing entities. Among the first proponents of such programs were
Russell (1914, 1927) and Whitehead (1920, 1929). As Butterfield (1984) points out, programs
of dispensing with points can only be successful if it is possible to rewrite theories which involve
quantification over points in a way that avoids reference to points. Moreover, Butterfield (1984)
explains that if one seeks to drop the assumption that there “really” are spacetime points, one
has to find suitable ersatz points, that can be abstracted by logical or mathematical means
from an acceptable class of basic entities, rather than dispensing with points altogether.

“The reason for this is familiar. Often we can only rewrite a theory while avoid-
ing some of its objects by identifying these objects with constructions from our
favoured objects; compare the reduction of arithmetic to set-theory by identifying
numbers with certain sets.” (Butterfield, 1984, p. 103)

Indeed this is exactly what reductionists with respect to points do: they abstract ersatz points
by some method that seems appropriate to them and then argue that the right topological
and geometrical relations obtain among those ersatz points. For example, in his first approach
Russell proposed to reduce sentences about spatial and temporal points to sentences about
sets of sense data.

“It is customary to think of points as simple and infinitely small, but geometry in
no way demands that we should think of them in this way. All that is necessary for
geometry is that they should have mutual relations possessing certain enumerated
abstract properties, and it may be that an assemblage of data of sensation will
serve this purpose.” (Russell, 1914, p. 119)

Later he proposed to reduce sentences about spacetime points to sentences about sets of phys-
ical events or processes 6 (cf. Russell, 1927). Whitehead’s (1920, 1929) approach is similar.
Both Whitehead and Russell had more or less empiricist motives for their reductionist pro-
grams. The main idea is this: because spacetime points cannot be seen, touched or perceived
6
The terms ‘process’ and ‘event’ have different meanings in various philosophical and scientific contexts. In
the following, I use both words ‘process’ and ‘event’ for all sorts of things that occur or happen, for example
planetary motions, concerts, explosions, propagations of action potentials within neural networks, collisions of
subatomic particles.
CHAPTER 1. INTRODUCTION 6

in any other way, they should be abstracted from empirically accessible entities (e.g. events
and processes).
There is also another kind of reductionists with respect to spacetime points, who are not
necessarily driven by empiricist sentiments. Their motive for abstracting spacetime points
rests on the view that the structure of spacetime regions is atomless, i.e. that there are no
indivisible regions of spacetime. For instance, Arntzenius (2003) holds such a view. If one
thinks that all spacetime regions are infinitely divisible and one nonetheless needs to refer to
points, then points have to be abstracted from regions by some sort of limit process.
In any case, if we seek to dispense with the assumption that there “really” are spacetime
points for whatever reason, then we should define the notion of spacetime point (i.e. what
counts as an ersatz point has to be given by a definition). This can be a motivation for defining
‘spacetime point’.
However, the problem of motivations of the described kind is that it cannot be taken for
granted that we have to accept a reductionist program about spacetime points. Not everyone
accepts Russell’s famous doctrine:

“The supreme maxim in scientific philosophising is this: wherever possible, logical


constructions are to be substituted for inferred entities.” (Russell, 1918, p. 148)

So since reductionist programs about spacetime points are controversial, the motivation is in
general not sufficient to show that we should define the notion of spacetime point.

Motivation 2 (a definition is needed to settle ontological questions). This motivation rests on


the fact that it is not clear what spacetime points are. In particular, there are a number
of open questions concerning their ontological status. For example: To which ontological
category do spacetime points belong? Which identity conditions do spacetime points have?
How are spacetime points related to entities of other kinds such as events and processes?
(Think of the ongoing debate between substantivalism and relationism.) In order to settle
ontological questions such as these, it is necessary to make the informal concept of spacetime
point more precise. As long as one uses an obscure concept these questions cannot be treated
in a rigorous way. Moreover, the concept of spacetime point should best be explicated by
formulating an explicit definition, since structural axioms alone (e.g. axioms along the lines of
synthetic geometry, such as those given by Robb (1936)) only determine which structure the
class of spacetime points has, but such axioms tell us nothing about their ontological status.
So, if we seek to answer questions about the ontological status of spacetime points in a rigorous
manner, we should attempt at formulating an adequate definition of the notion of spacetime
point.
CHAPTER 1. INTRODUCTION 7

It seems to me that this motivation is essentially correct. Defining the notion of spacetime
point is indeed necessary to answer questions about the ontological status of spacetime points.
We can conclude that the notion of spacetime point should be defined given the assumption
that ontological issues about spacetime points are to be treated in a rigorous way. But of
course, this motivation for defining the notion of spacetime point is not convincing if ontological
questions about spacetime points are regarded as meaningless pseudo-questions.
Let us now summarise the upshot of the preceding considerations. There are no generally
acceptable reasons to think that the predicate ‘spacetime point’ should be left undefined.
Likewise, there are no generally acceptable reasons which show that it should be defined.7
But given that ontological questions about spacetime points are to be treated in a rigorous
way, the notion of spacetime point should be defined. Throughout this thesis it is assumed
that ontological questions about spacetime points are not pseudo-questions and that they are
to be treated in a rigorous way. So for the purpose of this thesis it is justified to attempt at
defining the notion of spacetime point.
Up to now, various definitions of ‘spacetime point’ have been proposed and there are
different ontological presuppositions underlying these definitions. Occurrences of the predicate
‘spacetime point’ may have different intended meanings in different contexts or when used by
different scholars or scientists. So there is no point in trying to find the “one true” definition of
‘spacetime point’. Rather one should distinguish several intuitive concepts of spacetime point
and explicate them by means of different definitions. Then ontological questions containing
occurrences of ‘spacetime point’ can be reformulated in terms of the respective notion of
spacetime point.
In particular, it is desirable to have a definition that is ontologically as neutral as possible
(in the sense that the existence of entities falling under the definition has minimal ontological
presuppositions). Call such a definition of the notion of spacetime point a ‘thin definition’.
Finding a thin definition is particularly important, because it can serve as common ground
in ontological disputes where different concepts of spacetime point are involved. For example,
when there is dispute over the question whether spacetime points exist, it is not always clear
whether the disputing parties use the word ‘spacetime point’ in the same sense. In cases like
this, it is useful to have a thin notion of spacetime point such that both parties can agree
that there are spacetime points in this thin sense. Hence, a thin definition might serve as a
common conceptual basis and establish minimal agreement, although the original questions
need not be dismissed: even if there is agreement that there are spacetime points in a thin
sense, the question whether there are also spacetime points in a thick sense (e.g. atomic
spacetime regions as independently existing entities) is left open for discussion.
7
I assume here that we have discussed all relevant objections and motivations.
CHAPTER 1. INTRODUCTION 8

The main aim of this thesis is to find a definition of ‘spacetime point’ that is ontologically
as neutral as possible. The definition should be formally correct (i.e. it should satisfy the
usual formal rules of definition) and materially adequate (i.e. it should be close enough to an
intuitive concept of spacetime point and it should be ontologically as neutral as possible).

1.2 Overview
In order to deal with the question whether there is an adequate definition of the predicate
‘spacetime point’, I proceed as follows. At the beginning, I give an overview and classification
of the various definitions that have been proposed so far (chapter 2). Then I state necessary
conditions that an adequate definition of ‘spacetime point’ should satisfy and rule out some
proposed definitions on the back of those adequacy conditions. The most important remaining
families of definitions have in common that they identify points with classes of mereologically
and topologically structured basic entities such as processes or spacetime regions. I treat
the pre-theoretic philosophical problem whether a given definition is adequate by turning it
into the mathematical problem whether the corresponding point representation method is ad-
equate. Loosely speaking, a point representation method associates a corresponding atomistic
structure with every (possibly atomless) structure of a certain kind by means of some un-
derlying conception of representing points. I propose a general mathematical definition that
specifies what exactly point representation methods are and provide a common framework
for analysing different representation methods in this sense. In particular, I formalise general
adequacy conditions for representation methods within the chosen framework (section 3). In
the main part of the thesis, I scrutinise the most important families of representation methods,
which identify points with classes of basic entities (chapter 4).
Firstly, I discuss the approach of identifying points with ultrafilters (section 4.1). Call
this method of representing points ‘the ultrafilter method’. With minor modifications it is a
method along the lines of Russell’s proposal in Analysis of Matter (1927). Support for the
ultrafilter method comes from Stone’s representation theorem for Boolean algebras (1936).
Every Boolean algebra can be embedded into a complete and atomistic Boolean algebra,
namely the power set algebra over the original algebra’s ultrafilters. (The same result holds
for classical mereological structures, since they are just complete Boolean algebras with the
0-element removed.) One objection against the ultrafilter method is that it takes into account
only mereological, but not topological relations, since “[m]embership conditions for ultrafilters
are formulated exclusively in terms of Boolean relations and operations, so that the ultrafilters
are independent of the topology imposed on the Boolean algebra.” (Roeper, 1997, p. 258)
This is indeed problematic in many ways (particularly concerning the number and identity
CHAPTER 1. INTRODUCTION 9

of points). However, it is not problematic in the following respect: the ultrafilter method
can at least preserve additional topological relations imposed on the underlying mereological
structure. I have been able to show that Stone’s theorem can be extended in a way such that
additional topological structure is preserved by the representation as well. My theorem says
that every classical mereological structure endowed with an additional topological relation of
interior parthood (satisfying certain axioms) can be embedded into an atomistic structure of
the same kind. So Stone’s theorem does not only work for Boolean algebras and mereological
structures, but also for mereotopological structures. This result contributes to an elaboration
of the ultrafilter method. Yet, this method is inadequate. The reason is that it yields too many
points (in a specific sense of ‘too many’), as has been discussed by Gerla (1995) and Mormann
(2008). I prove that the ultrafilter method indeed fails to satisfy an adequacy condition for
that reason.
Secondly, I examine the approach of identifying points with completely prime filters (section
4.2). This method is common in point-free topology and it avoids some of the ultrafilter
method’s disadvantages: I demonstrate that it does not yield too many points. The method of
completely prime filters allows to prove idempotence theorems even for strong mereotopological
structures such as Varzi & Casati’s GEMTC (1999) and the structures defined by Barry Smith
(1996) without having to assume compactness. Yet it has serious problems. Drawing on a
theorem from point-free topology (Gerla, 1995), I show that this approach has to be rejected
as well, because it either presupposes that the mereological structure of the basic entities
already contains atoms or that the weak supplementation principle is violated. Both options
are untenable.
Thirdly, I present the approach of identifying points with maximal limited round filters
(section 4.3). The basic idea is that not all maximal filters represent points, but only maximal
filters that contain at least one limited (bounded) element and whose elements have interior
parts being members of the filter as well. The most comprehensive and detailed elaboration of
this basic idea has been worked out by Peter Roeper (1997). Roeper proved that this method
allows to construct an (up to homeomorphism) unique locally compact Hausdorff space for
every mereotopological structure of a certain kind. Call structures of that kind ‘Roeper-
structures’. The elements of a given Roeper-structure correspond to the regular open sets in
the associated topological space. It can also be shown that every locally compact Hausdorff
space gives rise to a unique Roeper-structure (up to isomorphism). As a consequence of
Roeper’s theorems, I prove that identifying points with maximal limited round filters is an
adequate representation method for a large class of Roeper-structures. Nonetheless, there
is a problem with Roeper’s approach. It establishes only that the maximal limited round
filter method works for atomless and discrete atomistic structures. Roeper’s approach does
CHAPTER 1. INTRODUCTION 10

not work for continuous atomistic structures. But the problem does not lie in the method
of identifying points with maximal limited round filters. Rather it lies in Roeper’s strong
mereotopological axioms, which rule out continuous atomistic structures in order to permit
proving the theorem that Roeper-structures correspond uniquely to locally compact Hausdorff
spaces (up to isomorphism and homeomorphism, respectively). So what needs to be done is
checking whether the method of using maximal limited round filters as points also works for
a broader class of mereotopological structures, which contains all relevant kinds of structures
spacetime might have (in particular continuous atomistic structures). If this task can be
accomplished, then the representation method in question is generally adequate. And as
we will see, the task can indeed be accomplished. Finally, another problem arises: namely
finding mereotopological axioms that are on one hand strong enough to allow the abstraction of
points and on the other hand weak enough to be satisfied by continuous and discrete atomistic
structures as well as by atomless structures. This is not an easy task. It requires a reduction of
the concept of limitedness to the relation of interior parthood in a general setting. It remains
an open problem.
Finally, I discuss metaphysical questions about the approach of identifying points with
certain classes of basic entities (chapter 5). In particular, I address the questions (1) whether
identifying points with classes presupposes realism about abstract objects, (2) which kind
of basic entities could be chosen and (3) which ontological status spacetime points have.
Concerning the first question, I argue that points can be regarded as pluralities rather than
sets or – in Russell’s words – as classes as many rather than classes as one. Identifying points
with classes as many does not presuppose realism about abstract objects. Concerning the
second question, I especially discuss the problem of empty spacetime regions. I propose a
possible solution to this problem, namely choosing the parts of the gravitational field (instead
of processes) as basic entities. This choice of basis ties in with the following ontological view:

“There is no such thing as an empty space, i.e. a space without field. Space-time
does not claim existence on its own, but only as a structural quality of the field[.]”
(Einstein, 1961, pp. 155–156)

What this amounts to is a reduction of sentences about spacetime points, spacetime regions
and spacetime to sentences about the mereotopological structure of the gravitational field.
Concerning the third question, I discuss which identity conditions spacetime points have, to
which ontological category they belong to and how they are related to entities of other kinds.
CHAPTER 2

PROPOSED DEFINITIONS AND THEIR PROBLEMS

In this chapter, I give an overview of the definitions proposed or discussed until now and I
classify these definitions. Then I propose conditions which an adequate thin definition has to
satisfy. Using these adequacy conditions I reject some definitions.

2.1 Overview and Classification of the Definitions


Proposed until Now
The definitions of ‘spacetime point’ that have been proposed until now can be divided into
two broad groups: (1) definitions that identify spacetime points with certain parts of basic
entities, and (2) definitions that identify spacetime points with certain classes of basic entities
or, more precisely, with classes that are grounded in basic entities by the transitive closure of
the membership relation. In the following these two groups of definitions are examined more
closely.

Points as Parts

The following definition is probably the most natural candidate for explicating the notion of
spacetime point.

(D1.1) x is a spacetime point iff x is a spacetime region without proper parts.

This definition basically follows Euclid’s conception of point. It is, for example, used by Field
(1984). Moreover, it is often used implicitly. The basic idea underlying D1.1 is that spacetime
is a substance which has parts (let us call them ‘regions’) some of whom are indivisible, and
that all regions are composed of indivisible regions. Given these assumptions, it suggests itself
to use the predicate ‘spacetime point’ excactly for the indivisible regions of spacetime. The
idea is clear, simple and straighforward.

11
CHAPTER 2. DEFINITIONS PROPOSED UNTIL NOW 12

There are also variants of definition D1.1 that do not use spacetime regions as basic entities,
but rather some other kind of mereologically structured entities. Here is a generalisation of
definition D1.1 for basic entities of some favoured kind K.

(D1.2) x is a spacetime point iff x is a K-entity without proper parts.

Smith (1993) uses a definition of this kind. The case that K comprises spherical regions is
along the lines of Huntington (1913), who developed an approach to Euclidean geometry by
taking the predicates ‘sphere’ and ‘part’ as basic and characterising points as spheres without
proper parts. The case that K comprises (possible) processes is discussed and criticised by
Butterfield (1984). Taking K as the category of processes is linked to the view that processes
are identical to spacetime regions. This view is called ‘supersubstantivalism’.1 It has been
defended by Wheeler (1962), Wheeler and Taylor (1963) and more recently by Skow (2005)
and Schaffer (2009).
Note that D1.1 and D1.2 use mereological structure alone to characterise points. There are
also attempts to give a mereotopological characterisation of points, for example Eschenbach
(1994). Eschenbach introduces a topological notion of region apart from the mereological
notion of parthood. A region in Eschenbach’s sense can be conceived of as an open set plus
an arbitrary (possibly empty) subset as its boundary (cf. 1994, p. 66). Regions in this sense
are taken as basic entities and points are considered as specific parts of regions. Eschenbach
defines the notion of point in the following way.2

(D1.3) x is a spacetime point iff x is part of some spacetime region and

(1) x is part of every spacetime region it overlaps, and


(2) for all y, if y is part of every region it overlaps, then y is part of x.

The important difference to D1.1 and D1.2 is that points in this sense can have proper parts
and need not to be regions themselves. This approach is highly non-standard.

Points as Classes

If one tends to the idea that points are not parts of basic entities, then it is still possible to
think of points as abstractions from mereologically and topologically structured basic entities.
This is particularly appealing if one does not want to rule out the possibility that the basic
1
If supersubstantivalism is combined with perdurantism (which it usually is), then “all concrete physical
objects (tables, chairs, electrons and quarks) are regions of spacetime” (Skow, 2005, p. v), according to this
view.
2
Of course, this definition can also be generalised for basic entities of a given kind K, as long as K comprises
of mereotopologically suitably structured entities.
CHAPTER 2. DEFINITIONS PROPOSED UNTIL NOW 13

entities in question have an atomless mereological sturcture. There are different versions of
carrying out the abstraction of points. In particular, there are different technical methods of
establishing the abstraction. But all methods have in common that they identify points with
certain classes that are grounded in basic entities (by the transitive closure of the membership
relation). There are three main approaches of doing so.
The simplest approach is to characterise points as classes of basic entities that satisfy some
appropriate condition. For example, this is done by Russell (1927), Tarski (1956), Grzegor-
czyk (1960), point-free topology, e.g. Johnstone (1982, 1983), Clarke (1985), Skyrms (1993),
Gerla (1995), Roeper (1997), Mormann (1998, 2009), Düntsch (2001,2002,2003), Vakarelov
et al. (2002), Düntsch & Winter (2005), Leitgeb (2007), Arntzenius (2008), Benda (2008),
Gruszczyński & Pietruszczak (2009), Forrest (2010). Definitions of that kind have the follow-
ing general form.

(D2.1) x is a spacetime point iff x is a class of K-entities such that . . . x . . .

The most important definitions of that kind are the definition of points as ultrafilters of basic
entities, the definition of points as completely prime filters of basic entities and the definition of
points as maximal limited round filters of basic entities. We will come back to these definitions
in chapter 4.
The second approach consists in characterising points as classes of classes of basic entities,
for example as equivalence classes of filters of basic entities as in Roeper (1997) or as equiva-
lence classes of pairs of basic entities as proposed by Bostock (2010). Whitehead (1920, 1929)
identifies points with equivalence classes of so-called abstractive sets. All these definitions
have the general form:

(D2.2) x is a spacetime point iff x is a class of classes of K-entities such that . . . x . . .

The third approach is to characterise points as classes of sequences of basic entites. Historically,
this is the oldest approach. Variants of this approach have been used by DeLaguna (1922),
Nicod (1930) and Menger (1941). The general form is:

(D2.3) x is a spacetime point iff x is a class of sequences of K-entities such that . . . x . . .

All three kinds of definitions have in common that K is typically either the category of space-
time regions or a kind of mereologically structured spatiotemporal entities (e.g. processes and
events). The only exception is Benda (2008), who identifies points with classes of worldlines.
Moreover, they have in common that the condition indicated by ‘. . . x . . .’ expresses that x
or the elements of x are somehow converging. At this stage it is not possible to clarify what
‘converging’ means exactly. But it will become clear in chapter 4, and the following example
may also help to get at least an intuitive grasp of what is meant.
CHAPTER 2. DEFINITIONS PROPOSED UNTIL NOW 14

Suppose our basic entities have the same mereological and topological structure as the
unions of open balls in an n-dimensional Euclidean space. So, every basic entity corresponds
to a union of open balls in the Euclidean space and vice versa. Then the structure is atomless,
i.e. every basic entity has another basic entity as proper part. In this case, defining points
as indivisible basic entities yields that there are no points at all, which is unsatisfactory if we
still want to refer to points. However, we can abstract points in this structure by a simple
method. Call basic entities which correspond to open balls ‘spherical’, and call two spherical
basic entities ‘concentric’ just in case their corresponding open balls are concentric in the
Euclidean plane. Now consider a maximal class of spherical basic entities that are concentric
to each other.

Such a class of basic entities converges in the intuitive sense meant above (although it does not
contain a basic entity that is part of all other members of the class). As Tarski (1956) showed,
the maximal classes of spherical basic entities that are concentric to each other can indeed be
regarded as abstracted points. It can also be shown that the class of these abstracted points
has the same properties as an n-dimensional Euclidean space.
Note that if our basic entities had the same structure as arbitrary non-empty subsets of the
n-dimensional Euclidean space, then the structure of our basic entities would be atomistic.
But nonetheless, the method of identifying points with maximal classes of spherical basic
entities that are concentric to each other would still work and, moreover, there would be an
exact correspondence between the atomic basic entities and the points as classes.
This toy example illustrates a general requirement for methods of defining points as classes
of basic entities. If the underlying structure of basic entities is atomless, then the points as
classes should somehow make up for the missing atoms; and if the underlying structure is
atomistic, then there should be an exact correspondence between the atomic basic entities
(the “real” points) and the points as classes (the “abstracted” points).
CHAPTER 2. DEFINITIONS PROPOSED UNTIL NOW 15

Underlying Structure

The definitions of ‘spacetime point’ which have been proposed so far rest on different assump-
tions about how the respective basic entities are structured. We can classify them by the kind
of structure which they presuppose. Every definition that I know of makes use of one of the
following kinds of structure:

(S1) mereological structure (e.g. parthood)

(S2) mereological and topological structure (e.g. parthood and connection, or parthood
and interior parthood),3

(S3) mereological and metric structure (e.g. parthood and being spherical, or parthood
and a notion of distance),

(S4) mereological and convex structure (e.g. parthood and being convex),

(S5) mereological and measure-theoretic structure (e.g. parthood and a notion of size),

(S6) intersection structure of lines (e.g. being a worldline and a notion of intersection).

Taking together the above distinction with the distinction between points as parts and points
as classes we get the following classification of all relevant definitions proposed until now.

3
This category comprises also theories that have only topological predicates as primitive notions and de-
fine mereological notions on their basis (e.g. Clarke, 1985), as well as theories that have only non-classical
“mereological” primitive predicates and define topological notions on their basis (e.g. Mormann, 1998; Forrest,
2010).
CHAPTER 2. DEFINITIONS PROPOSED UNTIL NOW 16

points as parts of points as classes grounded in basic entities


basic entities
mereological Field (1984), Whitehead (1920), Russell (1927), point-free
structure Smith (1993), topology, e.g. Johnstone (1982, 1983), Skyrms
Ridder (2002, (1993)
p. 117),
implicit in many
others.
mereological + Eschenbach (1994) DeLaguna (1922), Whitehead (1929), Nicod
topological (1930), Menger (1941), Grzegorczyk (1960),
structure Clarke (1985), Roeper (1997), Mormann
(1998, 2009), Düntsch (2001, 2002, 2003),
Vakarelov et al. (2002), Düntsch & Winter
(2005), Gruszczyński & Pietruszczak (2009),
Bostock (2010), Forrest (2010)
mereological + Huntington (1913) Tarski (1956), Gruszczyński & Pietruszczak
metric structure (2008), Gerla (1995)
mereological + Leitgeb’s (2007) reconstruction of Russell’s
convex structure approach (1927)
mereological + Arntzenius (2008)
measure-theoretic
structure
intersection Benda (2008)
structure of lines

Basic Entities

The proposed definitions are usually also accompanied by different choices of basic entities.
The most important kinds of entities in question are spacetime regions and processes. The
choice of basis is in most cases philosophically motivated and does not concern the formal
aspects of the respective definition. For the moment we will stay neutral concerning the
choice of basis. It will be discussed later on in a separate section (5.1).
CHAPTER 2. DEFINITIONS PROPOSED UNTIL NOW 17

2.2 Adequacy Conditions


In order to count as adequate, a definition of ‘spacetime point’ has to satisfy certain neces-
sary conditions. We are looking for a definition that is ontologically as neutral as possible.
Therefore, an adequate definition should not impose too strong restrictions on the structure of
basic entities. In the following we examine some metaphysical assumptions which an adequate
definition should not presuppose.
On the one hand, it should not be presupposed that the structure of basic entities is atom-
istic. Even if one accepts that there are spacetime regions (as independently existing entities),
one can doubt that there are any indivisible regions at all. This is not an inconsistent view.
Some metaphysicians (e.g. Zimmerman, 1996) and philosophers of physics (e.g. Arntzenius,
2003, 2008) think that the mereological structure of spacetime regions is atomless. However,
in physics, spacetime is standardly modelled by a collection of points. Thus, spacetime is
standardly presupposed to be atomistic. Apparently, this is in conflict with the possibility
that spacetime is atomless. Since it is not likely that physicists will give up their familiar
atomistic mathematical tools, the question arises whether it is possible to define the predicate
‘spacetime point’ even if there are no atomic spacetime regions. The same holds also for other
kinds of basic entities that are worthy of consideration.
On the other hand, it should not be presupposed that the mereological structure of basic
entities is atomless. Consider the following example: Whitehead (1920, 1929) developed a
general method for abstracting points under the assumption that the underlying mereological
structure of basic entities is atomless.4 Russell objected to Whitehead’s approach that we
could never know whether the structure of basic entities is atomless.

“As a piece of logic, this method is faultless. But as a method which aims at
starting with the actual constituents of the world it seems to me to have certain
defects. Dr Whitehead assumes that every event encloses and is enclosed by other
events. There is, therefore, for him, no lower limit or minimum, and no upper
limit or maximum, to the size of events. [. . .] I conclude that there is at present
no means of knowing whether events have a minimum or not; that there never
can be conclusive evidence against their having a minimum; but that conceivably
evidence may hereafter be found in favour of a minimum. [. . .] For the above
reasons, I am unable to accept Dr Whitehead’s construction of points by means
of enclosure-series as an adeqaute solution of the problem which it is designed to
solve.” (1927, pp. 292–294)
4
It is called ‘the method of extensive abstraction’.
CHAPTER 2. DEFINITIONS PROPOSED UNTIL NOW 18

Moreover, Butterfield objects that the assumption of an atomless structure of the basic entities
is not more plausible than assuming atomicity:

“One can of course construct extensionless objects mereologically in various ways,


e.g. in terms of sequences of extended objects whose ‘widths’ tend to zero (White-
head [1920], pp. 84–93; Tarski [1956], pp. 24–29). But I think the cure is as bad as
the disease; that is, positing a set of extended objects with the required structure
seems as questionable as positing extensionless objects simpliciter.” (1984, p. 105)

So we should not presuppose either of these principles.


For similar reasons it should also be neither presupposed that the structure of spacetime
is discrete nor that it is continuous. Classically, spacetime is modelled as a continuous struc-
ture, namely as a 4-dimensional differentiable manifold. But there are also many physicists
who think that spactime is discrete5 and therefore should be modelled, for example, as a lo-
cally finite partially ordered set of points as in the causal set approach to quantum gravity
(cf. Bombelli et al., 1987). Analogously, it should neither be presupposed that spacetime has
finitely many parts nor that it has infinitely many parts. Both conjectures cannot be ruled
out by our current empirical evidences.
These considerations can be carried over to other kinds of basic entities that are worthy of
consideration. So the upshot of this section is that an adequate definition of ‘spacetime point’
must satisfy the following conditions:

(A1) It does not presuppose that the mereological structure of the chosen basic entities
is atomistic or atomless, respectively.

(A2) It does not presuppose that the topologcial structure of the chosen basic is discrete
or continuous, respectively.

(A3) It does not presuppose that the totality of the chosen basic entities is finite or
infinite, respectively.

So an adequate definition of ‘spacetime point’ must work for atomistic structures as well as
for atomless structures, for discrete as well as for for continuous structures, and for finite as
well as for infinite structures.
Of course, the central expressions used in the adequacy conditions A1, A2 and A3 do
not have a precise meaning. But in chapter 3, we will explicate what ‘atomistic’, ‘atomless’,
‘discrete’ and ‘continuous’ mean in a formal framework. Then the informal adequacy conditions
stated here can be incorporated into a formal definition of the notion of general adequacy.
5
As Maudlin (2010, p. 65) points out: “there is speculation that, due to quantum effects, space-time may
have a discrete structure at very fine scales.”
CHAPTER 2. DEFINITIONS PROPOSED UNTIL NOW 19

2.3 Exclusion and Selection of Definitions


Based on the adequacy conditions from the preceding section we can immediately exclude
some kinds of definitions.
(1) Definitions that identify spacetime points with indivisible basic entities are inadequate,
because these definitions presuppose atomism. This rules out D1.1 and D1.2.
(2) Definitions that make use of metric (S3) or convex (S4) structure are inadequate.
The reason is that according to the general theory of relativity, convexity is not a property
of spacetime regions and there is no metric (in the strict sense of the word) available. As
Arntzenius (2008, p. 10) points out: “This approach is prima facie ill-suited for the purposes of
modern physics since in general relativity the notion of distance is local and path-dependent
rather, it is not a non-local path-independent relation between regions.” This rules out all
definitions using S3 and S4 that have been proposed so far.
(3) Arntzenius’ (2008) measure theoretic approach is inadequate, because it presupposes
that the structure of basic entities is atomless. This rules out the only definition using measure
theoretic structure (S5).
(4) Moreover, we can rule out Benda’s (2008) definition, which identifies spacetime points
with classes of mutually intersecting worldlines (S6). It presupposes that the structure of basic
entities is infinite and continuous. But even if the method of abstracting points from lines
could be generalised so that it works also for finite and discrete structures, it still would not
be worthy of consideration. Taking lines as basic entities is by no means better than taking
points as basic entities straightaway. So, definitions using intersetion structures of lines (S6)
can also be ruled out.
Eschenbach’s (1994) definition can be ruled out as well, because it is too far off the intu-
itive notion of point. Eschenbach regards points are parts of basic entities. But on the one
hand, points can have proper parts, according to Eschenbach’s definition, i.e. points can be
mereologically complex; and on the other hand, atomic basic entities may not be points. In
general, it is neither the case that all points are atomic nor that all atoms are points. This
does not go well with the intuitive notion of spacetime point. So Eschenbach’s definition is
inadequate for an explication of the notion of spacetime point (although her conception of
point may be useful for the purposes of computer science she mainly pursues).
A glance at the table in section 2.1 shows that the only approach that remains for further
investigation is representing points as classes of mereologically (S1) or mereotopologically (S2)
structured basic entities.
CHAPTER 3

A UNIFIED FRAMEWORK FOR ANALYSING DIFFERENT


REPRESENTATION METHODS

In this chapter, I introduce the formal framework that will be used in the rest of the thesis.
After briefly presenting some necessary preliminaries about topological spaces, mereological
structures and mereotopological structures, I develop a mathematical definition of the notion
of point representation and I propose formal adequacy conditions for point representations in
the defined sense. This is necesary because I transform the pre-theoretic philosophical problem
of finding an adequate definition of ‘spacetime point’ into the mathematical problem whether
there is an adequate point representation.

3.1 Preliminaries on Topology and Mereology

3.1.1 Preliminaries on Relational Structures


First we introduce the general notion of relational structure1 . All structures that are dealt with
in the following sections are relational structures in this sense: topological spaces, mereological
structures, mereotopological structures.

Definition 3.1.1. (X, A1 , . . . , An ) is a relational structure iff X 6= ∅ and for every i ∈


{1, . . . , n}, Ai ⊆ X m for some m ∈ N.

If (X, A1 , . . . , An ) is a relational structure, then X is the structure’s domain and A1 , . . . , An


can be considered as set theoretic attributes (properties or relations) on that domain. It is
1
Sometimes, the word ‘relational system’ is used for relational structure. I use the word ‘structure’ rather
than ‘system’, because ‘system’ is in my opinion better used to denote classes of structures (that are closed
under isomorphism), because classes of structures and not single structures correspond to syntactic axiom
systems. This establishes a conotational link between ‘system of structures’ and ‘axiom system’. Single
structures can only be models of axioms systems.

20
CHAPTER 3. UNIFIED FRAMEWORK 21

useful to have a notion for the case that two relational structures are of the same type in the
sense that their attributes have the same arity in the same order.

Definition 3.1.2. Let S and S0 be relational structures.


S is of the same type as S0 iff for some n ∈ N, there are X, X 0 and A1 , . . . , An as well as
A01 , . . . , A0n such that

(1) S = (X, A1 , . . . , An ) and S0 = (X 0 , A01 , . . . , A0n ), and

(2) for all i ∈ {1, . . . , n}, Ai and A0i have the same arity.

If relational structures of some kind Σ (say Boolean algebras) are given, it is sometimes
convenient or necessary to add further attributes (say closure operators) to these structures.
Then we say that the resulting structures are enriched Σ-structures (e.g. enriched Boolean
algebras). The following definition captures this idea.

Definition 3.1.3. Let Σ be a class of relational structures of the same type.


S is an enriched Σ-structure iff S is a Σ-structure or there is a Σ-structure S0 and there
are attributes A1 , . . . , An on the domain of S0 such that S = (S0 , A1 , . . . , An ).

Observe that the notion of enriched Σ-structure is here used in a weak sense. We can distin-
guish between properly enriched Σ-structures (i.e. Σ-structures that are endowed with addi-
tional attributes) and improperly enriched Σ-structures (which are just normal Σ-structures).
For example, every Boolean algebra which is endowed with additional topological operators is
a properly enriched Boolean algebra, whereas every Boolean algebra is trivially an improperly
enriched Boolean algebra.
Finally, we introduce the concepts of embedding and isomorphism in their most general
form for relational structures of the same type. Intuitively, a relational structure is isomorphic
to another relational structure just in case they have exactly the same form or, in other words,
they differ at most in their individual elements, the pattern of relations being the same.2 A
relational structure can be embedded into another relational structure just in case the first is
isomorphic to a substructure of the second.

Definition 3.1.4. Let S = (X, A1 , . . . , An ) and S0 = (X 0 , A01 , . . . , A0n ) be relational structures


of the same type.

– f is an embedding of S into S0 iff f is an injective function from X to X 0 and for all


i ∈ {1, . . . , n} and all x1 , . . . , xri ∈ X, (x1 , . . . , xri ) ∈ Ai iff (f (x1 ), . . . , f (xri )) ∈ A0i .
2
Since I use the term ‘relational structure’ and not ‘relational system’ the word ‘structure’ cannot as usually
be used to explain the informal notion of being isomorphic (having the same structure).
CHAPTER 3. UNIFIED FRAMEWORK 22

– S can be embedded into S0 (short: S  S0 ) iff there is an embedding of S into S0 .

– f is an isomorphism between S and S0 iff f is an embedding of S into S0 and f is a


bijection between X and X 0 .

– S is isomorphic to S0 (short: S ∼
= S0 ) iff there is a isomorphism between S and S0 .

3.1.2 Preliminaries on Topological Spaces


We need some elementary notions from standard topology. We introduce these notions without
detailed explanations, because this would lead too far.
A topological space is a non-empty set together with a collection of non-empty subsets
that are called ‘open sets’.

Definition 3.1.5. (X, O) is a topological space (short: TS) iff X 6= ∅ & O ⊆ P(X) and
the following conditions are satisfied:

(1) ∅ ∈ O & X ∈ O.

(2) For all U, V ∈ O, U ∩ V ∈ O.


S
(3) For all F ⊆ O, F ∈ O.

A subset of the set of points of a given topological space can be endowed with a topology
inherited from the whole space, namely the subset topology. The following definition makes
this idea more precise.

Definition 3.1.6. Let (X, O) be a topological space and let U ⊆ X.


O|U = {U ∩ V : V ∈ O}.

We use the term ‘the subset topology on U ’ to refer to O|U . It follows that (U, O|U ) is a
topological space if (X, O) is a topological space and U ⊆ X.

Example. Let X be a non-empty set. Then:

– (X, P(X)) is trivially a topological space.

– (X, {∅, X}) is trivially a topological space.

– Rn endowed with the Euclidean topology ORn (containing just the open sets given by
the Euclidean metric) is a topological space.

The next definitions introduces some standard notions from topology that will be used over
and over again in the following chapter.
CHAPTER 3. UNIFIED FRAMEWORK 23

Definition 3.1.7. Let T = (X, O) be a TS and U ⊆ X.

– U is open in T iff U ∈ O.

– U is closed in T iff U c is open in T .


S
– intT (U ) = {V ∈ O : V ⊆ U }.

– clT (U ) = [intT (U c )]c .

– U is regular open in T iff U = intT (clT (U )).

– U is regular closed in T iff U = clT (intT (U )).

– RO(T ) = {U : U is regular open in T }.

– RC(T ) = {U : U is regular closed in T }.

A subset of a topological space is a neighbourhood of a point iff the point is in the interior of
the set, i.e. if it is an interior point of it.

Definition 3.1.8. U is a neighbourhood of x in T iff T is a topological space and x ∈


intT (U ).

Another important notion is the notion of compactness. Intuitively, a subset of a topological


space is compact iff it is not too large. The concept of being compact is a generalisation of
the concept of being closed and bounded.

Definition 3.1.9. Let T = (X, O) be a TS and U ⊆ X.


S
– C is an open cover of U in T iff C ⊆ O and U ⊆ C.

– U is compact in T iff every open cover of U in T has a finite subset that is still an open
cover of U in T .

In the following steps, we introduce several kinds of topological spaces that will play an
important role later on.

Definition 3.1.10. (X, O) is a discrete space iff O = P(X) and X 6= ∅.

In a discrete space there are only isolated points. Everything is disconnected from everything
else.

Definition 3.1.11. (X, O) is a Hausdorff space iff (X, O) is a topological space and for all
x, y ∈ X there are U, V ∈ O such that x ∈ U , y ∈ V and U ∩ V = ∅ .
CHAPTER 3. UNIFIED FRAMEWORK 24

For example, every discrete space is a Hausdorff space. Moreover, Rn endowed with the
Euclidean topology is a Hausdorff space, every topological manifold3 is a Hausdorff space and
every metric on a set of points gives rise to a Hausdorff space. Nearly all topological spaces
that are studied in analysis or used as models in the sciences are Hausdorff spaces. The reason
is that Hausdorff spaces have nice properties. The following lemmas introduce some properties
of Hausdorff spaces that will be relevant for proving theorems in chapter 4. The first thing is
that in Hausdorff spaces all one-point sets are closed. In topological jargon: every Hausdorff
space (T2-space) is also a T1-space.

Lemma 3.1.1. Let T = (X, O) be a Hausdorff space. Then for all x ∈ X: {x} is closed in T .

In Hausdorff spaces, every closed subset of a compact set is compact as well.

Lemma 3.1.2. Let T be a Hausdorff space. If U is compact in T and V is a closed set in T


such that V ⊆ U , V is also compact in T .

Moreover, all compact sets are closed in Hausdorff spaces.

Lemma 3.1.3. Let T be a Hausdorff space. If U is compact in T and U is closed in T .

Proofs for these lemmas can be found in standard textbooks on topology (such as Munkres
(2000) or Bourbaki (1966)).
A topological space is locally compact just in case every point of the space has a compact
neighbourhood.

Definition 3.1.12. T = (X, O) is a locally compact space iff T is a topological space and
for every x ∈ X, there is a U ∈ O such that clT (U ) is compact in T and x ∈ U .

Sometimes it is possible reconstruct the open sets of a space by using only a designated subclass
of open sets. Such a subclass is called ‘basis’ of the respective topological space.

Definition 3.1.13. B is a basis of a topological space (X, O) iff for every U ∈ O, there is
S
an A ⊆ B such that U = A.

Topological spaces with countable bases are often of special interest.

Definition 3.1.14. T is a second countable space iff T has a countable basis.

Isomorphisms between topological spaces are called ‘homeomorphisms’ and topological spaces
of the same form are called ‘homeomorphic’. The next definition clarifies these notions.

3
The concept of topological manifold is explicated in definition 3.1.16.
CHAPTER 3. UNIFIED FRAMEWORK 25

Definition 3.1.15. Let T = (X, O) and T 0 = (X 0 , O0 ) be TS.

– f is a homeomorphism between T and T 0 iff f is a bijection from X to X 0 such that


for all U ⊆ X, U ∈ O iff f (U ) ∈ O0 . (cf. Munkres, 2000, p. 105)

– T is homeomorphic to T 0 (in short: T ∼


=T S T 0 ) iff there is a homeomorphism between
T and T 0 .

Finally, we can define the concept of topological manifold. Roughly speaking, a topological
manifold is a Hausdorff space that looks locally like Rn and has a countable basis.

Definition 3.1.16. Let T = (X, O) be a topological space.

– T is locally n-Euclidean iff for every x ∈ X, there is a neighbourhood U of x in T


such that (U, O|U ) is homeomorphic to Rn endowed with the Euclidean topology.

– T is a topological n-manifold iff T is a second-countable Hausdorff space that is locally


n-Euclidean.

Topological manifolds are particularly imporatant spaces, especially when it comes to applica-
tions of topology in physics. In the special as well as the general theory of relativity spacetime
is modelled as a topological manifold (endowed with additional structure).

3.1.3 Preliminaries on Mereological Structures


Before we can introduce mereological structures we need some general order theoretic concepts.

Definition 3.1.17. (lower and upper bounds, suprema and infima)

– x is an upper [a lower] bound of Y in (X, v) iff

(1) (X, v) is a partially ordered set (poset),


(2) x ∈ X and Y is a non-empty subset of X, and
(2) for all y ∈ Y : y v x [or x v y respectively].

– x is a supremum [an infimum] of Y in (X, v) iff

(1) x is an upper [a lower] bound of Y in (X, v), and


(2) for all upper [lower] bounds y of Y in (X, v), x v y [or y v x respectively].

It can easily be proven that suprema and infima are uniquely determined. This allows us to
introduce the following
CHAPTER 3. UNIFIED FRAMEWORK 26

F
Notation 1. The supremum of a subset Y in a partial order is denoted by ‘ Y ’, the supremum
of {x, y} is denoted by ‘x t y’ and the infimum of {x, y} is denoted by ‘x u y’. The supremum
of {y ∈ X : there is no zsuch that z v xand z v y} is denoted by ‘−x’ and is called the
complement of x.
We treat mereological systems not as sets of sentences but as sets of relational structures.
(This is slightly more convenient for our presentation, but nothing important hinges on that
point.) Every mereological structure consists of a domain of objects and a relation of parthood
on that domain (denoted by ‘v’). We read ‘x v y’ as ‘x is part of y’ and we use the following
convention:
Notation 2. Given a partial order v, let the relations @, ◦ and o be determined as follows: (a)
x @ y (spelled: ‘x is proper part of y’) iff x v y and x 6= y. (b) x ◦ y (spelled: ‘x overlaps y’)
iff there is a z such that z v x and z v y. (c) x o y (spelled: ‘x is disjoint from y’) iff it is not
the case that x ◦ y.

Definition 3.1.18. (mereological structures)

– (X, v) is a complete proto-mereological structure (short: PMS) iff (X, v) is a


poset that satisfies the following axiom of general suprema:

(SUP) For all non-empty Y ⊆ X there is a supremum of Y in (X, v).

– (X, v) is a complete mereological structure (short: COMS) iff (X, v) is a PMS


that satisfies the weak supplementation principle:

(WSP) For all x, y ∈ X, if x @ y, then there is a z such that z @ y and z o x.

– (X, v) is a Heyting-mereological structure (short: HMS) iff (X, v) is a PMS that


satisfies the general principle of distributivity:

(DIS) For all x ∈ X and all non-empty Y ⊆ X, if there is a y ∈ Y such that x ◦ y,


F F
then x u Y = {x u y : y ∈ Y and x ◦ y}.

– (X, v) is a classical mereological structure (short: CMS) iff (X, v) is a COMS that
satisfies the general sum principle:

(SUM) For all non-empty Y ⊆ X there is an x ∈ X such that


(1) for all y, if y ∈ Y , then y v x.
(2) for all z, if z v x, then there is a y ∈ Y such that z ◦ y.
CHAPTER 3. UNIFIED FRAMEWORK 27

Let cms be the set of all classical mereological structures and let hms, coms and pms be the sets
of all Heyting-mereological structures, complete mereological structures and complete proto-
mereological structures, respectively. We call such sets of mereological structures ‘systems of
mereology’. Before we can say more about the interrelation of these systems, we prove two
results that will be useful later on.

Corollary 3.1.1. Let S = (X, v) be a COMS. Then S satisfies the strong supplementation
principle:

(SSP) For all x, y ∈ X, if not x v y, then there is a z ∈ X such that z ◦ x and z o y.

Proof. Let S = (X, v) be a COMS. We show beforehand that S satisfies the following principle
of binary infima:

(BINF) For all x, y ∈ X, if x ◦ y, then there is an infimum z of {x, y} in S.


F
Let x ◦ y and let z = {u ∈ X : u v x and u v y}. (The supremum exists because x ◦ y
and so the set of common parts is not empty.) So we have: (a) z is an upper bound of
{u ∈ X : u v x and u v y}, i.e. for all u ∈ X, if u v x and u v y, then u v z. (b) For all
v ∈ X, if v is an upper bound of {u ∈ X : u v x and u v y}, then z v v. Clearly, x and y
are upper bounds of {u ∈ X : u v x and u v y}, therefore z v x and z v y. Thus, z a lower
bound of {x, y} and every lower bound of {x, y} is part of z. Thus, z is an infimum of {x, y}.
Now we show that BIN and WSP together entail SSP. This part of the proof is due to
Simons (cf. 1987, p. 31): Suppose it is not the case that x v y. For logical reasons, either
x ◦ y or x o y. If x o y, we are home. If x ◦ y, then there is an infimum of {x, y}, namely x u y.
Since it is not the case that x v y, we have x u y @ x. Then by WSP there is a z ∈ X such
that z @ x and z o x u y. Hence, z o y and so there is a z such that z v x and z o y in this case
as well.

Remark. We use SSP usually in its contraposed form, i.e. we show that for every z, if z ◦ x,
then z ◦ y, and then we conclude that x v y.

Corollary 3.1.2. Let (X, v) be a CMS and Y ⊆ X, Y 6= ∅. Then the general sum principle
takes the following form:
F
(GSP) For all x ∈ X: x ◦ Y iff there is a y ∈ Y with x ◦ y.
F F
Proof. (⇒) Let x ◦ Y . Then there is a z ∈ X such that z v x and z v Y . From the latter
we get by SUM (b) that there is a y ∈ Y such that z ◦ x. Hence, x ◦ y. (⇐) Suppose that
there is a y ∈ Y with x ◦ y. Then there is a z ∈ X such that z v x and z v y. From SUM (a)
F F
it follows that y v Y . So x ◦ Y as well.
CHAPTER 3. UNIFIED FRAMEWORK 28

Note that cms ⊂ coms ⊂ pms. Moreover, it can be shown that cms ⊂ hms.

Corollary 3.1.3. Every CMS is an HMS.

Proof. Let y ∈ Y such that x ◦ y. We only have to show that the principle of general dis-
F F F
tributivity holds: x u Y = {x u y : y ∈ Y and x ◦ y}. We use SSP. Let z ◦ x u Y . Let
F F
u = z u (x u Y ). Then u v z, u v x and u v Y . Therefore by GSP there is a y ∈ Y such
that u ◦ y. Note that u u y v x, y, z. Therefore, x ◦ y and z ◦ x u y for some y ∈ Y . Hence,
F F F
{x u y : y ∈ Y and x ◦ y}. Therefore (by SSP), x u Y v {x u y : y ∈ Y and x ◦ y}.
F
Conversely, let z ◦ {x u y : y ∈ Y and x ◦ y}. Then by GSP there is a y ∈ Y such that
F
x ◦ y and x u y ◦ z. Then y u (y u z) v x, y, z and therefore as well y u (y u z) v Y . Thus,
F F F
z ◦ x u Y and by SSP we get: {x u y : y ∈ Y and x ◦ y} v x u Y .

Remark. Note that not every HMS is a COMS. For example, the set of non-empty open subsets
of a Euclidean space with standard topology is a HMS but does not satisfy WSP and, thus, is
no COMS. For a proof see Mormann (2000, p. 471).
There is a verly close connection between complete Boolean algebras and classical mereological
structures. In fact, CMS are just complete Boolean algebras where the 0-element is removed.
This insight is due to Tarski (1935, footnote on pp. 190–191).

Theorem 3.1.1 (Tarski, 1935). Every CMS S can be turned into a complete Boolean algebra
A (by adding a 0-element) and every complete Boolean algebra A can be turned into a CMS S
(by deleting the 0-element).
F
The theorem shows that the mereological operations (−, u, t, ) obey those laws of complete
Boolean algebra where the 0-element is not involved. This allows us, for example, to use the
laws of associativity, distributivity, idempotence and De Morgan’s laws. For a proof of this
and related results see Ridder (2002, pp. 172–179 and pp. 190–193).
So far nothing has been said about atomicity. In order to examine atomistic structures as
well we introduce the notion of an atom in terms of our chosen framework.

Definition 3.1.19. Let S = (X, v) be a PMS.

– x is an atom in S iff x ∈ X and there is no y ∈ X such that y @ x.

– at(S) = {x ∈ X : x is an atom in S}

– atS (x) = {y ∈ at(S) : y v x}

Note that the axioms of classical mereology are neutral concerning the question whether atoms
exist. Atomistic and atomless structures can be specified as follows (cf. Simons, 1987, pp. 41–
43):
CHAPTER 3. UNIFIED FRAMEWORK 29

Definition 3.1.20. Let S = (X, v) be a PMS.

– S is atomistic iff for all x ∈ X there is a y ∈ X such that y v x and y is an atom in


(X, v).

– S is atomless iff for all x ∈ X there is a y ∈ X such that y @ x.

The next corollary gives prime examples of atomistic and atomless CMS.

Corollary 3.1.4. Let X be a non-empty set and let T = (Rn , ORn ), where ORn is the Euclidean
topology of Rn . Then:

(1) (P(X)\{∅}, ⊆) is an atomistic CMS.

(2) (RO(T )\{∅}, ⊆) is an atomless CMS.

(3) (RC(T )\{∅}, ⊆) is an atomless CMS.

Proof. Ad (1): It is well-known that P(X) together with the usual set theoretic opera-
tions forms a complete Boolean algebra, where ∅ is the 0-element and X is the 1-element
(cf. Givant & Halmos, 2009, p. 8). Then we know from theorem 3.1.1 that (P(X)\{∅}, ⊆) is a
CMS. Moreover, the singletons {x} are atoms of this CMS for all x ∈ X. Since every element
S
U of P(X)\{∅} is identical to {{x} : x ∈ U }, the CMS is atomistic. For a proof of (2) and
(3) see Givant and Halmos (2009, chapter 10) and lemma 6.1 in Roeper (1997, p. 283).

Atomistic CMS have particularly nice properties. Each atomistic CMS is isomorphic to the
CMS that arises from the power set algebra over the set of its atoms by deleting the empty set.
F
The function atS establishes the isomorphism, its inverse is the sum operator ( ) restricted
to sets of atoms.

Corollary 3.1.5. Let S = (X, v) be an atomistic CMS. Then:


F F
(1) For every x ∈ X: x = atS (x). For every non-empty U ⊆ at(S): U = atS ( U ).

(2) atS is an isomorphism between (X, v) and (P(at(S))\{∅}, ⊆).


F S
(3) atS ( {x : A[x]}) = {atS (x) : A[x]}.

Proof. Ad (1) and (2): Since (X, v) is an atomistic CMS, atS (x) is non-empty and atS (x) ∈
F
P(at(S))\{∅} for each x ∈ X. Furthermore, U ∈ X for every U ∈ P(at(S))\{∅}. It
F F
follows that x = atS (x) for each x ∈ X, and U = atS ( U ) for each non-empty U ⊆ at(S).
So atS (x) is a bijection between X and P(at(S))\{∅}. Moreover, it is clear that x v y iff
atS (x) ⊆ atS (y) (this is a well-known result in classical atomistic mereology).
CHAPTER 3. UNIFIED FRAMEWORK 30

F F
Ad (3): u ∈ atS ( {x : A[x]}) iff u is an atom in S and u v {x : A[x]} iff u is an atom
in S and for some y ∈ X, A[y] and u ◦ y iff u is an atom in S and for some y ∈ X, A[y] and
S
u v y iff for some y ∈ X, A[y] and u ∈ atS (y) iff u ∈ {atS (x) : A[x]}.

3.1.4 Preliminaries on Mereotopological Structures


A given mereological structure can be endowed with an additional topological relation. It is
well-established to use either a relation of connection or a relation of interior parthood for this
purpose. I choose interior parthood () as an additional relation. The concept of interior
parthood can be understood in different ways. But what all notions of interior parthood have
in common can be explicated by the collection of axioms which is contained in the following
definition.

Definition 3.1.21. (X, v, ) is a weak mereotopological structure (short: WMTS) iff


(X, v) is a CMS and  is a relation on X such that:

(MTS1) For all x, y ∈ X, if x  y, then x v y.

(MTS2) For all x, y, u, v ∈ X, if x v u and u  v and v v y, then x  y.

(MTS3) For all x, y, z ∈ X, if x  y and x  z, then x  y u z.

(MTS4) There is an x ∈ X such that for all y ∈ X: y  x.

(MTS5) For all x, y ∈ X, if x  y, then there is a z ∈ X such that x  z and z  y.

Let wmts be the set of all weak mereotopological structures. For WMTS we can define interior
and closure operators as well as the notions of openness and closedness.

Definition 3.1.22. Let S = (X, v, ) be a WMTS and x ∈ X. Then:

– x has an interior in S iff there is some y ∈ X such that y  x.


F
– intS (x) = {y : y  x} (provided that x has an interior in S).

−int (−x) : x 6= F X and − x has an interior in S.
S
– clS (x) = F
 X : else.

– x is open in S iff intS (x) = x.


F
– x is closed in S iff either −x is open in S or x = X.

– OS = {x ∈ X : x is open in S}.
CHAPTER 3. UNIFIED FRAMEWORK 31

In the case of atomistic WMTS, these notions and operators behave in almost the same manner
as the well-known corresponding notions and operators for topological spaces.

Theorem 3.1.2. Let S = (X, v, ) be an atomistic WMTS. Let x, y ∈ X and let x have an
interior in S. Then:

(1) intS (x) v x.

(2) intS (intS (x)) = intS (x).

(3) intS (x u y) = intS (x) u intS (y), provided that x u y has an interior in S
F F
(4) intS ( X) = X.

Proof. Ad (1): Let u ∈ atS (intS (x)). So z v y for some y with y  x. By MTS2 we get:
u  x. Hence, it follows from MTS1 that u v x. Thus, atS (intS (x)) ⊆ atS (x). So corollary
3.1.5 yields: intS (x) v x.
Ad (2): In view of (1) we only have to show that intS (x) v intS (intS (x)). Let u ∈
atS (intS (x)). Then as before, u  x. Using MTS5 we can conclude that there is some v with
u  v  x. Then by definition of the mereological interior operator: v v intS (x). Now MTS2
yields u  intS (x). Consequently, u v intS (intS (x)). Therefore, atS (intS (intS (x))) ⊆
atS (intS (x)). Hence, intS (intS (x)) v intS (x), by corollary 3.1.5.
Ad (3) Let x u y have an interior in S. Then ‘intS (x u y)’, ‘intS (x)’ and ‘intS (y)’ are
well-defined. First, let u ∈ atS (intS (x) u intS (y)). Then u v intS (x) and u v intS (y). Again
using MTS1 and MTS2 it follows that u  x and u  y. With the help of MTS3 we get
z  x u y and, thus, z v intS (x u y). So intS (x) u intS (y) v intS (x u y). Conversely, let
u ∈ atS (intS (x u y)). Then u  x u y. Since, x u y is a common part of x and y, it follows
by MTS1 and MTS2 that u  x and u  y. Thus, u v intS (x) and intS (y). Therefore,
u v intS (x) u intS (y). Hence also intS (x u y) v intS (x) u intS (y).
F F F F
Ad (4): From MTS5 we get: X  X. Then X v intS ( X). Together with (1)
this yields the required result.

As a consequence of the previous theorem we get the result that the closure operator in
an atomistic WMTS is also very similar to the closure operators in standard topology. The
only difference is that there is no 0-individual in WMTS. Hence there is no condition that
corresponds to the principle that the empty set is closed.
CHAPTER 3. UNIFIED FRAMEWORK 32

Corollary 3.1.6. Let S = (X, v, ) be an atomistic WMTS. Then for all x, y ∈ X:

(1) x v clS (x).

(2) clS (clS (x)) = clS (x).

(3) clS (x t y) = clS (x) t clS (y).

Proof. The Boolean properties of the mereological operations (cf. theorem 3.1.1) lead directly
to the corresponding propositions about the closure operator.
F
Ad (1): Case x 6= X and −x has an interior: x = − − x v −intS (−x) = clS (x). Case
F F
x = X or −x has no interior: Then x v X = clS (x), by definition.
F
Ad (2): Case x 6= X and −x has an interior: clS (clS (x)) = −intS (− − intS (−x)) =
F
−intS (intS (−x)) = −intS (−x) = clS (x). Case x = X or −x has no interior: Then by
F F
definition clS (x) = X and for the same reason clS (clS (x)) = X.
F
Ad (3): Case xty 6= X and −(xty) has an interior: Then clS (xty) = −intS (−(xty)) =
−intS (−x u −y) = −[intS (−x) u intS (−y)] = −intS (−x) t −intS (−y) = clS (x) t clS (y).
F F
Case x t y = X or −(x t y) has no interior: Then again clS (x t y) = X. The first
F
statement of this corollary implies x t y v cl(x) t cl(y). Therefore: cl(x) t cl(y) = X.

In view of these results it suggests itself that the interior parthood relation of an atomistic
WMTS determines a topology on the set of atoms – and this is indeed the case. We show that
it is possible to prove that every atomistic WMTS gives rise to a canonical topological space
with the atoms of the WMTS as points.

Definition 3.1.23. Let S be an enriched atomistic WMTS.


Top(S) = (at(S), Oat (S)) with Oat (S) = {atS (x) : x is open in S} ∪ {∅}.

Theorem 3.1.3. If S = (X, v, ) is an atomistic WMTS, then for all x ∈ X:

(1) x is open in S iff atS (x) is open in Top(S).

(2) x is closed in S iff atS (x) is closed in Top(S).

(3) atS (intS (x)) = intTop(S) (atS (x)) (provided that x has an interior in S).4

(4) atS (clS (x)) = clTop(S) (atS (x)) (provided that −x has an interior in S).
4
Remark for purists: normally, the interior operator intT is defined only for the case that T = (X, O) is a
topological space. But in this case (when using the functor ‘intTop(S) ’ ) we do not want to presuppose that
Top(S) is a topological space. However, observe that the definition of ‘intT ’ does not depend on T being
a topological space. The definiens does not make use of any specific properties of O. It allows O to be an
arbitrary subset of P(X). Therefore, the restriction to the case that T = (X, O) is not necessary and it is
justified to use ‘intTop(S) ’ without having already proved that Top(S) is a topological space.
CHAPTER 3. UNIFIED FRAMEWORK 33

Proof. Note that (1) is trivial. See definition 3.1.23. It is convenient to prove the remaining
statements in the following order: (3), (2), (4).
Ad (3): Let x have an interior in S. Then ‘intS (x)’ is well-defined. Observe the fol-
F S
lowing facts. On the one hand we have: atS (intS (x)) = atS ( {y : y  x}) = {at(y) :
y  x}, which is by definition 3.1.22 and corollary 3.1.5. On the other hand we have:
S
intTop(S) (atS (x)) = {U : U ∈ Oat (S) and U ⊆ atS (x)}. Now we show that the right hand
S
sides of these equations are identical. Let u ∈ {at(y) : y  x}. Hence, u is an atom in S such
that u v y for some y with y  x. Thus, by MTS2, u  x. So u v intS (x). By statement
(1) in theorem 3.1.2 we have: intS (x) v x. Of course, intS (x) is open in S. So atS (intS (x))
is open in Top(S) and atS (intS (x)) ⊆ atS (x). Note that u ∈ atS (intS (x)). Therefore, u ∈ U
S
for some U ∈ Oat (S) with U ⊆ atS (x). Hence, u ∈ {U : U ∈ Oat (S) and U ⊆ atS (x)}.
S
Conversely, let u ∈ {U : U ∈ Oat (S) and U ⊆ atS (x)}. So u ∈ U for some U ∈ Oat (S)
F F F F
with U ⊆ atS (x). It follows that u v U , U = intS ( U ) and U v x. Because u is an
F
atom, we get: u  U (using MTS1 and MTS2) and, furthermore, by MTS2: u  x. Thus,
S
u ∈ {at(y) : y  x}.
F
Ad (2): Consider the case that x 6= X. Then x is closed in S iff −x is open in S iff
−x = intS (−x) iff atS (−x) = atS (intS (−x)) iff atS (x)c = intTop(S) (atS (x)c ) iff atS (x)c is
open in Top(S) iff atS (x) is closed in Top(S). (This sequence of equivalences is justified by
F
corollary 3.1.5 and statement (3) of this theorem.) Now consider the case that x = X. Then
x is closed in S. Moreover, atS (x)c = at(S)c = ∅ ∈ Oat (S). So atS (x) is open in Top(S).
Thus, also in this case x is closed in S iff atS (x) is open in Top(S).
Ad (4): Let −x have an interior in S. We have two cases again. In the case that x 6=
X we have: atS (clS (x)) = atS (−intS (−x)) = atS (intS (−x))c = intTop(S) (atS (−x))c =
F

intTop(S) (atS (x)c )c = clTop(S) (atS (x)).


(These equivalences are again justified by corol-
F
lary 3.1.5 as well as statement (3) of this theorem.) In the case that x = X it follows
that atS (x) = at(S). Therefore, atS (clS (x)) = atS ( X) = at(S) and atS (x)c = ∅.
F

So intTop(S) (atS (x)c ) = ∅. Hence, at(S) = intTop(S) (atS (x)c )c = clTop(S) (atS (x)). Thus,
atS (clS (x)) = at(S) = clTop(S) (atS (x)).

Theorem 3.1.4. If S = (X, v, ) is an atomistic WMTS, then Top(S) is a topological space.


F
Proof. (1) Trivially, ∅ ∈ Oat (S). It follows from MTS4 that for all x ∈ X, x  X. So
F F F F F
X = {x ∈ X : x  X}. Thus, X = {x ∈ X : x  X}. Moreover, at(S) = at( X).
Therefore, at(S) ∈ Oat (S).
(2) Let U, V ∈ Oat (S). In the case that U ∩V = ∅, it follows from (1) that U ∩V ∈ Oat (S).
F F F F
So consider the case that U ∩ V 6= ∅. Then U = intS ( U ) and V = intS ( V ). So it
F F F F F F
follows: U ∩ V = atS ( U ) ∩ atS ( V ) = atS ( U u V ) = atS (intS ( U ) u intS ( V )) =
CHAPTER 3. UNIFIED FRAMEWORK 34

F F F
atS (intS ( Uu V )) = atS (intS ( (U ∩ V ))) = intTop(S) (atS (U ∩ V )).
This sequence of equations is justified by corollary 3.1.5, the assumption that U, V ∈
Oat (S), statement (3) of theorem 3.1.2 and finally statement (3) of the preceding theorem
3.1.3.
S S
(3) Let F ⊆ Oat (S). We have to show that F ∈ Oat (S). If F = ∅ this follows
S S F
from (1). So suppose that F 6= ∅. Observe that F = {U : U ∈ F} = {atS ( U ) : U ∈
F F F F F
F} = atS ( { U : U ∈ F}). Now let u ∈ atS ( { U : U ∈ F}). Then u v U for some
F F
U ∈ F. Thus, U ∈ Oat (S). So U = intS ( U ). Because u is an atom in S, it follows that
F F
u v v for some v  U . Thus by MTS2: u  U . For the same reason it follows from
F F F F F F F
U v { U : U ∈ F} that u  { U : U ∈ F}. Therefore, u v intS ( { U : U ∈ F}).
F F F F
Again because u is an atom in S we get: u ∈ atS (intS ( { U : U ∈ F})). So atS ( { U :
F F
U ∈ F}) ⊆ atS (intS ( { U : U ∈ F})). Moreover, corollary 3.1.5 and theorem 3.1.2 imply
F F F F F F
that atS (intS ( { U : U ∈ F})) ⊆ atS ( { U : U ∈ F }). Consequently, { U : U ∈
F F FS F F
F} = intS ( { U : U ∈ F }). The observation above yields: F = { U : U ∈ F }.
FS FS S
Thus, we finally have F = intS ( F). Hence, F ∈ Oat (S).

As in the case of topological spaces we can define a notion of compactness.

Definition 3.1.24. Let S be an enriched WMTS.


F
– U is an open cover of x in S iff U ⊆ OS and x v U.

– x is compact in S iff for every open cover of x in S has a finite subset that is still an
open cover of x in S.

– S is compact iff for every x in the domain of S, x is compact in S.

Lemma 3.1.4. Let S be an enriched atomistic WMTS. Then for all x in the domain of S, x
is compact in S iff atS (x) is compact in Top(S).

Proof. It follows from theorem 3.1.3 that C is an open cover of x in S iff {atS (u) : u ∈ C} is
an open cover of atS (x) in Top(S). Hence, for every x in the domain of S, every open cover
of x in S has a finite subset that is still an open cover of x in S just in case every open cover
of atS (x) in Top(S) has a finite subset that is still an open cover of atS (x) in Top(S). Thus,
for all x in the domain of S, x is compact in S iff atS (x) is compact in Top(S).

For the moment, this is all we need to say about WMTS in general. Let us now consider some
special kinds of WMTS. Atomistic WMTS can be discrete or continuous. The next definition
says what this means precisely.
CHAPTER 3. UNIFIED FRAMEWORK 35

Definition 3.1.25. Let Σ be a class of enriched WMTS of the same type.


S is a discrete Σ-structure iff S ∈ Σ, S is atomistic and Top(S) is a discrete space.
S is an atomistic continuous Σ-structure iff S ∈ Σ, S is atomistic and Top(S) is a
topological n-manifold for some n ∈ N

As mentioned at the beginning of this section, there are different ways how the notion of interior
parthood can be understood. We will now distinguish between the two most important ways
of doing so by adding further axioms to the axioms that characterise WMTS. The first way
of understanding interior parthood is that x is an interior part of y just in case the closure of
x is part of the interior of y. A notion of interior parthood understood in this way is often
called ‘non-tangential parthood’ (cf. Whitehead, 1929, p. 297). This sense of interior parthood
is captured by the axiom defining normal mereotopological structures.

Definition 3.1.26. S = (X, v, ) is a normal mereotopological structure (short:


F
NMTS) iff (X, v, ) is a WMTS such that for all x, y ∈ X with y 6= X, if x  y,
then −y  −x.

These structures are very similar to normal contact algebras and proximity spaces (cf. Vakarelov,
2007, p. 281). In atomistic NMTS, where mereological closure and interior operators are al-
ways defined, it can be shown that the intended sense of  is at least partially captured by
the axioms of NTMS.

Corollary 3.1.7. Let S = (X, v, ) be an atomistic NMTS. Then for all x, y ∈ X with
F
y 6= X, if x  y, then clS (x) v intS (y).
F
Proof. Let x  y and y 6= X. Then by MTS5, there is some u with x  u  y. Therefore,
x v intS (u) and u v intS (y). Since S is an NMTS, it also follows that −u  −x. Thus,
−u v intS (−x). So −intS (−x) v u. Therefore, clS (x) v u v intS (y).

Normal mereotopological structures and normal Hausdorff spaces are related to each other.
The next theorem shows that NMTS arise from normal Hausdorff spaces in two ways.

Theorem 3.1.5. Let T = (X, O) be a normal Hausdorff space, and for all non-empty subsets
A, B of X, let A T B iff clT (A) ⊆ intT (B). Then:

(E1) (RO(T )\{∅}, ⊆, T ) is an NMTS.

(E2) (P(X)\{∅}, ⊆, T ) is an NMTS.

Proof. Ad E1: The non-empty regular open subsets of the space form a CMS (RO(T )\{∅}, ⊆),
see Givant and Halmos (2009, chapter 10). We have to show that T satisfies the conditions
CHAPTER 3. UNIFIED FRAMEWORK 36

in definition 3.1.21. In the case of regular open sets A, B we have: A T B iff clT (A) ⊆ B.
Verifying conditions 1., 2. and 5. is easy. That condition 3. is also satisfied can be shown as
follows:
Let A T B. Then clT (A) ⊆ B and hence clT (A) ∩ B c = ∅. Since T is normal and
clT (A) and B c are disjoint, closed sets in T , there are non-empty, disjoint, open sets U, V
in T such that clT (A) ⊆ U and B c ⊆ V . U ∩ V = ∅ is equivalent to U ⊆ V c . Since V is
open in T , V c is closed in T . Therefore, clT (U ) ⊆ clT (V )c = V c . Hence clT (U ) ∩ V = ∅.
Therefore, clT (U ) ∩ B c = ∅, and furthermore, clT (U ) ⊆ B. Because U is open, clT (U ) is a
regular closed set in T , i.e. clT (U ) = clT (intT (clT (U ))). So we get clT (intT (clT (U ))) ⊆ B.
Moreover, U ⊆ intT (clT (U )) because U is open, and hence clT (A) ⊆ intT (clT (U )). Thus
C := intT (clT (U )) is a regular open set in T and clT (A) ⊆ C as well as clT (C) ⊆ B, i.e.
A T C and C T B.
Condition 4. is satisfied because finite intersections of regular open sets are regular open
sets themselves. Therefore, A uT B iff A ∩ B. Thus, A T B uT C iff clT (A) ⊆ B ∩ C iff
clT (A) ⊆ B & clT (A) ⊆ C iff A T B & A T C.
Ad E2: We know that (P(X)\{∅}, ⊆) is a CMS. Conditions 1., 2. and 5. are very easily
verified again. The procedure of showing that condition 3. holds is essentially analogous to
the case in E1. Regarding condition 4., note that A uT B = A ∩ B and that the interior of
the intersection of two sets is the intersection of their interiors. Therefore, A T B uT C iff
clT (A) ⊆ intT (B∩C) iff clT (A) ⊆ intT (B)∩intT (C) iff clT (A) ⊆ intT (B) & clT (A) ⊆ intT (C)
iff A T B & A T C.

The second way of understanding interior parthood is that x is an interior part of y just in
case that x is part of the interior of y. This sense of interior parthood can be fully captured
by adding a single axiom to those that characterise WMTS.
Definition 3.1.27. (X, v, ) is a classical mereotopological structure (short: CMTS)
iff (X, v, ) is a WMTS such that: for all x, y ∈ X, if not x  y, then there is a u ∈ X such
that u v x and u o v for all v  y.
Note that the defining axioms of CMTS can be obtained more or less straightforwardly from
the well-known characterization of topological spaces via interior operators. The axioms are
equivalent to the axioms of Barry Smith’s system of mereotopology (1996). The following
corollary shows that every CMTS indeed satisfies the condition expressing the intuitive mean-
ing of .
Corollary 3.1.8. Let S = (X, v, ) be a CMTS, x, y ∈ X and let y have an interior in S.
Then x  y iff x v intS (y).
CHAPTER 3. UNIFIED FRAMEWORK 37

F
Proof. (⇒) Suppose x  y. Then clearly x v {y : y  x} = intS (x). (⇐) Conversely,
suppose that x v intS (y). The contraposition of the condition in the definiens of definition
3.1.27 yields: if for all u such that u v x there is a v such that v  y and u ◦ v, then x  y.
F
We derive the antecedent. Let u v x. Then u v intS (y) = {y : y  x}. So there is a v such
that v  y and u ◦ v. Therefore x  y.

It can be shown that interior operators on CMTS behave in almost the same manner as interior
operators on topological spaces. Note that it is not necessary to assume atomicity in the case
CMTS, whereas we had to assume atomicity for the corresponding result for WMTS.

Theorem 3.1.6. Let S = (X, v, ) be a CMTS, x, y ∈ X and let x have an interior in S.


Then:

(1) intS (intS (x)) = intS (x).

(2) intS (x u y) = intS (x) u intS (y), provided that x u y has an interior in S
F F
(3) intS ( X) = X.

Proof. Ad 1: Using SSP, we let u ◦ intS (x). Then by GSP there is a y  x such that
u ◦ y. From MTS3 it follows that there is a z such that y  z  x. Then z v intS (x).
Hence by MTS2: y  intS (x). Moreover u ◦ y. So by GSP: u ◦ intS (intS (x)). Thus
intS (x) v intS (intS (x)). Because intS (x) has an interior, it follows by the first statement of
this theorem that intS (intS (x)) v intS (x).
Ad 2: Let x u y have an interior in S. Then ‘intS (x u y)’, ‘intS (x)’ and ‘intS (y)’ are well-
defined. We use SSP and show that for every z ∈ X, z◦intS (xuy) iff z◦intS (x)uintS (y). (⇒)
F
Let z ◦ intS (x u y). Because intS (x u y) = {u : u  x u y}, there is a u such that u  x u y
and z ◦ u. Then by MTS3 u  x and u  y. The previous corollary yields u v intS (x) and
u v intS (y). Since z ◦ u, we have z ◦ intS (x) u intS (y). (⇐) Let z ◦ intS (x) u intS (y). Let
u := intS (x) u intS (y). Then u v intS (x) and u v intS (y). So by the last corollary: u  x
and u  y. Again using MTS3 we get: u  x u y. Moreover, u ◦ z. Therefore, z ◦ intS (x u y).
F F
Ad 3: It follows from MTS5 that: X  X. Then the previous corollary together with
the the first statement of this theorem yield the required result.

As a consequence of the previous theorem we get the result that the closure operator in a
CMTS is very similar to the closure operators in standard topology. Again note that it is not
necessary to assume atomicity for CMTS.
CHAPTER 3. UNIFIED FRAMEWORK 38

Corollary 3.1.9. Let S = (X, v, ) be a CMTS and x, y ∈ X. Then:

(1) x v clS (x).

(2) clS (clS (x)) = clS (x).

(3) clS (x t y) = clS (x) t clS (y).

Proof. The Boolean properties of the mereological operations (cf. theorem 3.1.1) lead directly
to the corresponding propositions about the closure operator.
F
Ad 1: Case x 6= X and −x has an interior: x = − − x v −intS (−x) = clS (x). Case
F F
x = X or −x has no interior: Then x v X = clS (x), by definition.
F
Ad 2: Case x 6= X and −x has an interior: clS (clS (x)) = −intS (− − intS (−x)) =
F
−intS (intS (−x)) = −intS (−x) = clS (x). Case x = X or −x has no interior: Then by
F F
definition clS (x) = X and for the same reason clS (clS (x)) = X.
F
Ad 3: Case xty 6= X and −(xty) has an interior: Then clS (xty) = −intS (−(xty)) =
−intS (−x u −y) = −[intS (−x) u intS (−y)] = −intS (−x) t −intS (−y) = clS (x) t clS (y).
F F
Case x t y = X or −(x t y) has no interior: Then again clS (x t y) = X. The first
F
statement of this corollary implies x t y v cl(x) t cl(y). Therefore cl(x) t cl(y) = X as
well.

Finally, we can derive a corollary that clarifies the notion of interior parthood in CMTS further.
Moreover, this result will be helpful for proving theorems about CMTS in the next chapter.

Corollary 3.1.10. Let S = (X, v, ) be a CMTS and x, y ∈ X. Then:


If x  y, then x  intS (y)  y.

Proof. Let x  y. Then y has an interior and x v intS (y), by corollary 3.1.8. Moreover, by
theorem 3.1.6: intS (y) v intS (intS (y)). So intS (y)  intS (y), again by by corollary 3.1.8.
Hence by MTS2, x  intS (y). We know from theorem 3.1.6 that intS (y) v y. Then again
by (2.) intS (y)  y, because intS (y)  intS (y).
CHAPTER 3. UNIFIED FRAMEWORK 39

3.2 A General Theory of Point Representations


The following definition introduces a formal explication of the informal notion of point repre-
sentation or representation method.

Definition 3.2.1. R is a Σ-point representation iff Σ is a class of enriched PMS, and


R = (pt, corr, [.]pt ) for some pt, corr, [.]pt such that:

(1) pt is a function that maps every S ∈ Σ to a set pt(S).

(2) corr is a function that assigns 0 or 1 to every triple (S, x, P ), where S ∈ Σ, x is


an element of the domain5 of S and P ∈ pt(S).

(3) [.]pt is a function that maps every S ∈ Σ to an enriched PMS [S]pt such that

(a) [S]pt is of the same type as S, and


(b) P(pt(S))\{∅} is the domain of [S]pt .

Notation. From now on we simply write ‘Spt ’ instead of ‘[S]pt ’ when there is no risk of confu-
sion.
Let us survey which purposes the three components pt, corr and [.]pt of a Σ-point represen-
tation serve. (1) The function pt is intended to give us the set of points for every Σ-structure.
Generally, pt will use some particular method for abstracting points from a given structure’s
elements and their relations. For instance, pt may be a function that maps every Σ-structure
to the set of all classes of a particular kind, e.g. to the set of all ultrafilters in the structure.
(2) The function corr is the correspondence indicator. It is intended to show whether a point
corresponds to a given basic entity in the sense that the basic entity contains the point. For
example, if pt delivers classes of a certain kind as points, corr may be defined as follows:
‘corr(S, x, P ) = 1 iff P ∈ pt(S), x is in the domain of S and x ∈ P .’ (3) Finally, [.]pt is a
function that puts together what pt and corr give us. It returns a corresponding atomistic
structure [S]pt for every structure S ∈ Σ . Again there will be a particular method used for
this purpose. The basic idea is that the structure [S]pt contains an isomorphic copy of the
original structure S and, if S is not atomistic already, the missing points are contained in
[S]pt as well. Moreover, [S]pt is endowed with copies of the relations of S that are suitably
adjusted to the presence of atoms in [S]pt . How all this works in detail will become clearer
later on in section 4, where we analyse concrete examples of point representations.
5
If S = (X, v, , A1 , . . . , An ) is an enriched WMTS, then we call X ‘the domain of S’.
CHAPTER 3. UNIFIED FRAMEWORK 40

The above definition is intended to capture those features that all point representations
have in common and, therefore, it is very general. However, in many cases it is useful to
have also a notion of point representation in a narrower sense at hand. Normally, we are con-
cerned with the representation of structures that are at least weak mereotopological structures
(possibly endowed with additional relations). The reason is that these structures have corre-
sponding topological spaces if they are atomistic. Hence, they encode topological information
about their domain (besides mereological information) and are therefore good candidates for
the application of point representations. So if we use the expression ‘point representation’
simpliciter – i.e. without mentioning a specific class of enriched PMS –, then we mean a point
representation for some class of enriched WMTS.

Definition 3.2.2. R is a point representation iff for some class Σ of enriched WMTS of
the same type, R is a Σ-point representation.

The next definition specifies what it means to extend a Σ-point representation so that it also
applies to specific enriched Σ-structures.

Definition 3.2.3. (pt∗ , corr∗ , [.]∗pt ) is an extension of Σ-point representation (pt, corr, [.]pt )
for Σ∗ iff Σ∗ is a class of enriched Σ-structures, (pt, corr, [.]pt ) is a Σ-point representation and
(pt∗ , corr∗ , [.]∗pt ) is a Σ∗ -point representation such that for all S∗ ∈ Σ∗ the following conditions
are satisfied, given that S is the Σ-structure which is a component of S∗ :

(1) pt∗ (S∗ ) = pt(S);

(2) corr∗ (F, x, S∗ ) = corr(F, x, S);

(3) [S]pt is a component of [S∗ ]∗pt .

The notion of being an extension as defined in the definition above is a non-strict notion. That
is, every Σ-point representation is an extension of itself for Σ. The reason is that the notion of
being an enriched Σ-structure is also also used in a non-strict sense (so that every Σ-structure
is an improperly enriched Σ-structure), see definition 3.1.3.
An extension in the narrower sense is defined exactly for those kinds of structures, for
which it is an extension of the original representation. The following definition makes this
more precise.

Definition 3.2.4. (pt∗ , corr∗ , [.]∗pt ) is an extension of Σ-point representation (pt, corr, [.]pt )
iff for every class Σ∗ of enriched Σ-structures, if (pt∗ , corr∗ , [.]∗pt ) is a Σ∗ -representation, then
(pt∗ , corr∗ , [.]∗pt ) is an extension of the Σ-point representation (pt, corr, [.]pt ) for Σ∗ .
CHAPTER 3. UNIFIED FRAMEWORK 41

For every Σ-point representation, we can define a function that maps every basic entity of
every Σ-structure to the set of point representatives that corresponds to that basic entity (as
given by the representation method). This device will be used very often later on.

Definition 3.2.5. Let R = (pt, corr, [.]pt ) be a Σ-point representation and let S ∈ Σ.
ptS (x) = {P ∈ pt(S) : corr(S, x, P ) = 1}, for every x in the domain of S.

We can conceive of the elements of ptS (x) as those points which lie within x. So ptS (x) is the
representative of x in the atomistic structure Spt .
On the basis of the above definitions, it is possible to specify in a precise way what it
means that a Σ∗ -point representation is adequate for a class of structures Σ.

Definition 3.2.6. R is an adequate Σ∗ -point representation for Σ iff Σ∗ is a class of


enriched PMS of the same type, Σ ⊆ Σ∗ , R is a Σ∗ -point representation and the following
conditions are satisfied:

(1) For every S ∈ Σ, ptS is an embedding of S into Spt and Spt is an atomistic
Σ∗ -structure.

(2) For every S ∈ Σ, if S is atomistic, then ptS is an isomorphism between S and


Spt .

Again, we also have a more specific notion of being an adequate point representation for class
of structures, which is restricted to enriched WMTS.

Definition 3.2.7. R is an adequate point representation for Σ iff for some class Σ∗ of
enriched WMTS of the same type, R is a an adequate Σ∗ -point representation for Σ.

Let us examine more closely how the two definitions above can be understood. As we see,
definition 3.2.6 contains two main conditions. This is because the problem of deciding whether
a given point representation is adequate for a class of structures Σ comprises two main sub-
problems, namely the embedding problem and the idempotence problem. The embedding prob-
lem consists in showing that every structure in Σ can indeed be embedded into an atomistic
structure (of sufficiently similar kind) by means of the point representation in question, see
part (1) in the above definition 3.2.6. The idempotence problem consists in showing that
every atomistic structure in Σ is isomorphic to its corresponding structure given by the point
representation, see part (2) in the mentioned definition. In particular, this means that if an
adequate point representation for Σ is applied to a structure S and Spt is in Σ, then applying
the representation anew yields a structure [Spt ]pt which is isomorphic to Spt . So iterating
an adequate representation leaves everything the same (up to isomorphism), given that Σ is
CHAPTER 3. UNIFIED FRAMEWORK 42

closed under pt. Under this condition, adequate point representations behave like idempotent
operators on isomorphism classes of Σ-structures. For this reason the second problem is called
‘idempotence problem’. The following corollary states this more precisely.

Corollary 3.2.1. Let R = (pt, corr, [.]pt ) be an adequate Σ∗ -point representation for Σ and Σ
is closed under pt . Then:

(1) For every S ∈ Σ, S  Spt . (Extensivity/Embedding)

(2) For every S ∈ Σ, Spt ∼


= [Spt ]pt . (Idempotence)

Proof. This follows directly from the definition 3.2.6.

In the case that it also holds for all S, S0 ∈ Σ with S  S0 that Spt  S0pt (monotonicity),
this corollary shows that [.]pt acts as a closure operator on isomorphism classes of Σ-structures.
I hope these consequences illustrate what the notion of being an adequate point representation
for a class of structures means. Here are two lemmas that will be useful later on.

Lemma 3.2.1. If Σ ⊆ Σ0 and R is not an adequate point representation for Σ, then R is not
an adequate point representation for Σ0 .

Proof. Let Σ ⊆ Σ0 and assume that R is not an adequate point representation for Σ. So
for every class Σ∗ of enriched WMTS of the same type, R is not an adequate Σ∗ -point rep-
resentation for Σ. Let Σ∗ be a class of enriched WMTS of the same type. In the case that
Σ0 is not a subset of Σ∗ , it follows that R is not an adequate Σ∗ -point representation for Σ0 ,
and we are home. So consider the case that Σ0 ⊆ Σ∗ . Then R is not an adequate Σ∗ -point
representation for Σ. So either (1) there is a S ∈ Σ such that ptS is not an embedding of
S into Spt or Spt is not an atomistic Σ∗ -structure, or (2) there is a S ∈ Σ such that S is
atomistic and ptS is not an isomorphism between S and Spt . In view of Σ ⊆ Σ0 it follows
that either (1) there is a S ∈ Σ0 such that ptS is not an embedding of S into Spt or Spt is not
an atomistic Σ∗ -structure, or (2) there is a S ∈ Σ0 such that S is atomistic and ptS is not an
isomorphism between S and Spt . Therefore, R is not an adequate Σ∗ -point representation for
Σ0 . So for every class Σ∗ of enriched WMTS of the same type, R is not an adequate Σ∗ -point
representation for Σ0 . Finally, we can conclude that R is not an adequate point representation
for Σ0 .

Lemma 3.2.2. If (pt∗ , corr∗ , [.]∗pt ) is an extension of the Σ-point representation (pt, corr, [.]pt )
for Σ∗ , (pt∗ , corr∗ , [.]∗pt ) is an adequate Σ∗ -point representation for Σ∗ and Σ0 = {S ∈ Σ :
S is a component of S∗ for some S∗ ∈ Σ∗ }, then (pt, corr, [.]pt ) is an adequate Σ-point rep-
resentation for Σ0 .
CHAPTER 3. UNIFIED FRAMEWORK 43

Proof. Assume the antecedent. Then (pt, corr, [.]pt ) is a Σ-point representation. We have to
show that for all S ∈ Σ0 , (1) ptS is an embedding of S into Spt and Spt is an atomistic Σ-
structure, (2) if S is atomistic, then ptS is an isomorphism between S and Spt . Let S ∈ Σ0 .
Then there is a S∗ ∈ Σ∗ such that S is a component of S∗ . Because (pt∗ , corr∗ , [.]∗pt ) is
an adequate Σ∗ -point representation for Σ∗ it follows that (1) pt∗S is an embedding of S∗
into [S∗ ]∗ pt and [S∗ ]∗ pt is an atomistic Σ∗ -structure, (2) if S∗ is atomistic, then pt∗S is an
isomorphism between S∗ and [S∗ ]∗ pt . Note that (pt∗ , corr∗ , [.]∗pt ) is an extension of the Σ-
point representation (pt, corr, [.]pt ) for Σ∗ . Thus, in view of definition 3.2.3 we get that: (1)
ptS is an embedding of S into Spt and Spt is an atomistic Σ-structure, and if S is atomistic,
then ptS is an isomorphism between S and Spt .

It is important to note that being an adequate point representation for some specific class of
structures does not mean that the point representation in question is generally adequate. The
concept of being generally adequate is to be understood in the sense that a point representation
falls under it just in case it works for all relevant kinds of structures which spacetime might
possess. So generally adequate point representations must be neutral concerning the questions
whether spacetime is atomistic or atomless, discrete or continuous, finite or infinite, see the
informal adequacy conditions in section 2.2. This idea is stated more precisely in the next
definition.

Definition 3.2.8. R is a generally adequate Σ-point representation iff there is a class Σ∗


of enriched WMTS of the same type such that Σ∗ ⊆ Σ, R is an adequate Σ-point representation
for Σ∗ and the following conditions are satisfied:

(1) Σ∗ contains finite discrete Σ-structures.

(2) Σ∗ contains infinite discrete Σ-structures.

(3) Σ∗ contains atomistic continuous Σ-structures.

(4) Σ∗ contains atomless Σ-structures.

It is convenient to have also a one-place notion of general adequacy. A point representation is


generally adequate simpliciter just in case it is generally adequate for some class of enriched
WMTS.

Definition 3.2.9. R is a generally adequate point representation iff for some class Σ
of enriched WMTS of the same type, R is a generally adequate Σ-point representation.

These definitions enable us to formulate the principal tasks of representing points more accu-
rately.
CHAPTER 3. UNIFIED FRAMEWORK 44

Task 1: Find a class Σ of enriched WMTS that covers all possible kinds of structures space-
time might have (finite discrete, infinite discrete, atomistic continuous, atomless).

Task 2: Find an adequate point representation for Σ.

If both tasks can be accomplished, then there is a generally adequate point representation and
the technical problem of point representation is solved.
CHAPTER 4

ANALYSING THE MOST IMPORTANT FAMILIES OF


POINT REPRESENTATIONS

In this chapter, I analyse the main families of point representation methods, which identify
points with classes as basic entities. These are (1) methods using ultrafilters, (2) methods using
completely prime filters and (3) methods using maximal round filters. All other representation
methods are either very similar to one of these families or inherently problematic. Analysing
these methods would either lead us too far afield or only to already familiar results. Therefore,
I limit the following analysis to the three mentioned representation methods.
The ultrafilter method is along the lines of Russell’s (1927) proposal for abstracting points
from events (cf. Mormann, 2008; Bostock, 2010), and it receives support from Stone’s repre-
sentation theorem for Boolean algebras. The completely prime filter method comes from the
mathematical discipline of point-free topology, which has been developed in the 1960s. In
point-free topology, the completely prime filter method is used successfully to abstract points
from Heyting algebras. Whereas the ultrafilter method and the completely prime filter method
use only mereological concepts to identify points, the maximal round filter method explicitly
takes into account topological structure for identifying points. The notion of maximal round
filter stems mainly from the theory of proximity spaces (Naimpally & Warrack, 1970), which
can be conceived of as a kind of mereotopology.

45
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 46

4.1 Representing Points by Ultrafilters


Consider a set of mutually overlapping basic entities (e.g. processes or regions). If a further
element that overlaps all others is added to the set, then the common part shrinks (or stays the
same). To be sure that this works well, we also add the common part of every two elements to
the set.1 The more overlapping elements are taken together, the smaller is their common part.
So, if we take a maximal set of overlapping basic entities, which cannot be extended without
losing the intersection property, then we arrive at a limit of mutual intersection and thus have
a good representative of a point, even if there are actually no atoms in the structure. The idea
is that we may thus approximate points by such maximal sets of mutually intersecting basic
entities, which are called ‘ultrafilters’. Considerations of that kind underly the approach of
representing points by ultrafilters and can be found in early works by Russell (1914, 1927).2
Support for this approach comes from Stone’s representation theorem for Boolean algebras,
which says that every Boolean algebra can be embedded into a complete and atomistic Boolean
algebra, namely the power set algebra over the original algebra’s ultrafilters (cf. Stone, 1936).
Because of theorem 3.1.1, the same result holds for classical mereological structures.
There are several attempts at using ultrafilters as ‘ersatz atoms’ in specific mereological
structures or Boolean algebras. For example, one might define possible worlds as ultrafilters
of propositions (cf. Bub, 1997, p. 21). In the same spirit, Skyrms (1993) proposed to con-
struct atomic facts, properties and individuals from corresponding complete Boolean algebras,
drawing explicitly upon Stone’s work. Likewise one can define spacetime points as ultrafilters
of basic entities of some kind (e.g. processes or regions).
Now we examine this approach more closely. In this section, we define what ultrafilters
are, we show that support for this approach comes from Stone’s representation theorem, fur-
thermore we prove that Stone’s theorem can be extended to mereotopological structures and
finally we will see that this approach nonetheless has a problem.
1
This rules out some problems that would arise otherwise. For example, consider the situation illustrated
in the following figure. Here the elements a, b and c t d are pairwise overlapping, yet there is no part that is
common to a, b and c t d. So the common part of all the elements vanishes.

By postulating that for every two elements, their mereological product is also element of the set, we can avoid
this problem. This postulate does not allow situations such as the one illustrated above.
2
A set of events satisfies “Russell’s definition of a point-group iff it corresponds, with a few obvious changes,
to what is called an ultrafilter in a Boolean algebra.” (Bostock, 2010, p. 27)
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 47

Definition 4.1.1. Let S = (X, v) be a poset.

– F is a filter in (S, v) iff F is a non-empty subset of X such that:

(1) for all x, y ∈ F , there is a z ∈ F such that z v x and z v y;


(2) for all x, y ∈ X, if x ∈ F and x v y, then y ∈ F .

– A filter F is a maximal filter in S iff there is no filter F 0 in (X, v) such that F ⊂ F 0 .

– A filter F is an ultrafilter in S iff for all x, y ∈ X, if x t y ∈ F , then x ∈ F or y ∈ F .

For classical mereological structures and Boolean algebras the notions of maximal filter and
ultrafilter coincide.

Lemma 4.1.1. Let S be a CMS. Then F is an ultrafilter in S iff F is a maximal filter in S.

Proof. (cf. Bourbaki, 1966, pp. 60–61)


(⇒) Let F be an ultrafilter in S = (X, v). Hence for all x, y ∈ X, if x t y ∈ F , then
x ∈ F or y ∈ F . Suppose that there is a filter F 0 in (X, v) such that F ⊂ F 0 . Then there is
an x ∈ F 0 \F and x 6= X. Because x t −x = X ∈ F and x ∈
F F
/ F , −x must be in F . But
since F ⊂ F 0 , −x must be in F 0 as well. So x ∈ F 0 and −x ∈ F 0 . Thus there had to be a
z ∈ F 0 such that z v x and z v −x, for F 0 is a filter. But this is impossible, because x o −x.
(⇐) Let F be a maximal filter in S = (X, v). It can be shown that: z is in F iff for all
u ∈ F there is a common part of z and u in F . Now let x t y ∈ F . Suppose x ∈
/ F and y ∈
/ F.
Therefore, there is a u ∈ F such that there is no common part of x and u in F , and that there
is a u0 ∈ F such that there is no common part of y and u0 in F . Since u, u0 ∈ F , also u u u0 is
in F . But there is no common part of u u u0 and x t y in F . Thus x t y ∈
/ F , contradicting
the assumption. Thus, x ∈ F or y ∈ F .

The following lemma is an important result about the existence of ultrafilters. It says that
every filter can be extended to an ultrafilter. This result is vital for theorems that rely on
the existence of ultrafilters. Its proof rests on Zorn’s lemma and, thus, is intuitionistically
unacceptable. But note that the ultrafilter lemma is strictly weaker than Zorn’s lemma (or
the axiom of choice) in ZF. For a proof see Halpern (1964). Thus we could have taken the
ultrafilter lemma as a set theoretic axiom in order to avoid the full axiom of choice and its
equivalents.

Lemma 4.1.2 (Ultrafilter Lemma). For every filter F in a CMS S there is an ultrafilter
F 0 such that F ⊆ F 0 .
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 48

Proof. (cf. Bourbaki, 1966, p. 60) The proof uses Zorn’s lemma. Let F be a filter in S =
(X, v) and let F = {F 0 : F 0 is filter in (X, v) and F ⊆ F 0 }. Then (F, ⊆) is a partial order.
S S
Let C be a chain in (F, ⊆). Then it is easy to show that C is a filter in (X, v) and C
is an upper bound of C. Thus, every chain in (F, ⊆) has an upper bound in (F, ⊆). Zorn’s
lemma yields that (F, ⊆) has a maximal element, which is a maximal filter that contains F .
By the previous corollary we obtain that F is an ultrafilter.

So far we know what ultrafilters are and that, given Zorn’s lemma, there are plenty of them.
But how is the ultrafilter-based point representation to be specified when the expression ‘point
representation’ is used in the sense of definition 3.2.1?

Definition 4.1.2. Let S = (X, v) be a CMS.

– uf(S) = {U : U is an ultrafilter in S}.



1 : F ∈ uf(S), x is in the domain of S and x ∈ F.
– corruf (F, x, S) =
0 : otherwise.

– [S]uf = (Xuf , ⊆), whereXuf = P(uf(S))\{∅}.

(uf, corruf , [.]uf ) is called ‘the ultrafilter representation’.

Corollary 4.1.1. (uf, corruf , [.]uf ) is a cms-point representation.

4.1.1 Stone’s Theorem for Classical Mereological Structures


Every classical mereological structure is isomorphic to an algebra of sets, where the ultrafilters
of the structure are regarded as points.

Theorem 4.1.1 (Stone, 1936). If S is a CMS, then ufS is an embedding from S into Suf
and Suf is an atomistic CMS.

Proof (cf. Stone, 1936). Let S be a CMS. First note that ufS (x) 6= ∅ for every x ∈ X, because
{y ∈ X : x v y} is a filter and is contained in some ultrafilter according to the ultrafilter
lemma. Let U = {ufS (x) : x ∈ X}. Then U ⊆ P(uf(S))\{∅}, because ufS (x) ⊆ uf(S) for all
x ∈ X.
ufS is trivially surjective on U. We proof that ufS is injective as well: Let x 6= y. We have
to show: ufS (x) 6= ufS (y). First note that x 6= y and SSP imply that there is a z such that
z v x and z o y, or there is a z such that z v y and z o x.
Case 1: z v x and z o y. We know that ufS (z) is not empty, i.e. there is some ultrafilter
U ∈ ufS (z). Moreover z ∈ U . It follows that x ∈ U , because z v x and U is a filter. Now
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 49

suppose, y ∈ U as well. Because z, y ∈ U and U is a filter, it could not be the case that
z o y – contradicting the case assumption. So y ∈
/ U . Hence there is an ultrafilter U such that
U ∈ ufS (x) but U ∈
/ ufS (y). Thus ufS (x) 6= ufS (y). Case 2 can be treated analogously.
Now we show that x v y iff ufS (x) ⊆ ufS (y) for all x, y ∈ X. (⇒) Let x v y. Let
U ∈ ufS (x). Then x ∈ U . Therefore, y ∈ U as well, because U is a filter and x v y. So
U ∈ ufS (y). Thus ufS (x) ⊆ ufS (y). (⇐) Let ufS (x) ⊆ ufS (y). We use SSP to show x v y.
Let z ◦ x. Then there is a u ∈ X such that u v z and u v x. Due to the ultrafilter lemma
4.1.2, there is some ultrafilter U ∈ ufS (u). Note that u ∈ U , u v x and U is a filter. Therefore,
x ∈ U and U ∈ ufS (x). Then by the assumption: U ∈ ufS (y), i.e. y ∈ U . So U is a filter
containing both u and y. Then u ◦ y and, moreover, z ◦ y. Hence by SSP: x v y.
Moreover, it is clear that Suf = (P(uf(S))\{∅}, ⊆) is an atomistic CMS (see theorem
3.1.4).

So the ultrafilter representation is a cms-point representation that yields an embedding of


every CMS into an atomistic CMS.
But which topological spaces correspond to atomistic CMS? It can be shown that there
is a canonical correspondence between complete Boolean algebras and compact extremally
disconnected Hausdorff spaces (cf. Johnstone, 1982, p. 102). This result can be transferred to
CMS in virtue of theorem 3.1.1.
But since atomistic CMS correspond to topological spaces that are extremally disconnected,
this result is unsatisfactory from a topological point of view. It is not possible to obtain a
continuous or at least connected space by means of this method. This has been pointed out by
Gerla (1995, p. 1019). The problem lies in the fact that CMS are purely mereological structures.
In contrast to mereotopological structures, they are devoid of additional topological relations.
So it is not surprising that the corresponding topological spaces are trivial and do not contain
significant topological information.
Hence, the question arises whether more satisfactory results can be obtained when the
ultrafilter method is applied to mereotopological structures rather than purely mereological
structures. The ultrafilter representation can easily be extended for mereotopological struc-
tures, namely in the following way.

Definition 4.1.3. Let S∗ be a WMTS and let S = (X, v).

– uf(S∗ ) = uf(S).

– corruf (F, x, S∗ ) = corruf (F, x, S).

– [S∗ ]uf = (Xuf , ⊆, uf ), where


CHAPTER 4. ANALYSING POINT REPRESENTATIONS 50

Xuf = P(uf(S∗ ))\{∅} and


2 : for some u, v ∈ X: u  v & Φ ⊆ uf (u) & uf (v) ⊆ Ψ}.
uf = {(Φ, Ψ) ∈ Xuf S S

So (uf, corruf , [.]uf ) is a wmts-point representation.

Corollary 4.1.2. (uf, corruf , [.]uf ) is a point representation.

Note note that the points of a mereotopological structure are identified with the ultrafilters
of the underlying CMS. So the points of a WMTS are determined by purely mereological
means according to the ultrafilter method. The topological relation of interior parthood of
this atomistic WMTS (uf ) is defined in terms of the relation of interior parthood imposed
on the underlying CMS ().
A general objection against the ultrafilter method is that it takes into account only mere-
ological, but not topological structure, since “[m]embership conditions for ultrafilters are for-
mulated exclusively in terms of Boolean relations and operations, so that the ultrafilters are
independent of the topology imposed on the Boolean algebra.” (Roeper, 1997, p. 258) This
objection must not be confused with the above mentioned problem that arises when the ul-
trafilter method is applied to purely mereological structures. The objection says that even in
the case that there are additional topological relations (e.g. ) imposed on the underlying
CMS, the ultrafilter representation does not take into account these relations to determine the
set of points arising from the structure. As we will see later on in the present section, this
is indeed problematic concerning the number and identity of points. The objection suggests
also to doubt whether the ultrafilter representation is able to preserve additional topological
relations imposed on the underlying structure, and whether mereotopological structures can
be embedded into corresponding atomistic structures using the ultrafilter representation only.
But as I have been able to show, the ultrafilter method has no problem there. Indeed it is
provable that Stone’s theorem can be extended in a way such that additional topological struc-
ture is preserved by the representation as well. More specifically, it can be shown that every
mereological structure endowed with an additional topological relation of interior parthood
(satisfying certain axioms) can be embedded into an atomistic structure of the same kind. So
Stone’s theorem does not only work for Boolean algebras and mereological structures, but also
for mereotopological structures.

4.1.2 Extending Stone’s Theorem for


Mereotopological Structures
Now we show that (under specific constraints) the representation of points by ultrafilters
preserves the topological relation of interior parthood () in addition to the Boolean relation
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 51

of parthood (v). This works for different kinds of mereotopological structures. First of all,
we prove that every WMTS can be embedded into an atomistic WMTS via the ultrafilter
representation.

Theorem 4.1.2 (Embedding Theorem for WMTS). If S = (X, v, ) is a WMTS, then


ufS is an embedding of S into Suf and Suf is an atomistic WMTS.

Proof. Proceed as in Stone’s theorem. As before, it is easy to show that for all x, y ∈ X:
x v y iff ufS (x) ⊆ ufS (y). Moreover, it is clear that (Xuf , ⊆) is an atomistic CMS, see lemma
3.1.4. It is also easy to prove that  is preserved by ufS . Let x, y ∈ X. First let x  y.
Then trivially by definition4.1.3 we get: ufS (x) uf ufS (y). Conversely, let ufS (x) uf ufS (y)
. Then there are u, v ∈ X such that u  v and ufS (x) ⊆ ufS (u) and ufS (v) ⊆ ufS (y). Then
x v u and v v y. Consequently, x  y. Hence, for all x, y ∈ X: x  y iff ufS (x) uf ufS (y).
So far, so good. What is left to show is that Suf indeed is a WMTS. In the following we
assume Φ, Ψ, Γ, ∆ ∈ Xuf .

1. Let Φ uf Ψ. Then there are u, v ∈ X such that u  v, Φ ⊆ ufS (u) and ufS (v) ⊆ Ψ.
From MTS1 we get: u v v. Therefore, ufS (u) ⊆ ufS (v), because every F ∈ ufS (u) is an
ultrafilter such that u ∈ F and, hence, v ∈ F as well. So Φ ⊆ Ψ.

2. Let Φ ⊆ Γ, Γ uf ∆ and ∆ ⊆ Ψ. So there are u, v ∈ X such that u  v, Γ ⊆ ufS (u)


and ufS (v) ⊆ ∆. Then also Φ ⊆ ufS (u) and ufS (v) ⊆ Ψ. Thus, Φ uf Ψ.

3. Let Γ uf Φ∩Ψ. So there are u, v ∈ X such that u  v, Γ ⊆ ufS (u) and ufS (v) ⊆ Φ∩Ψ.
Then clearly, Γ uf Φ and Γ uf Ψ, since Φ ∩ Ψ is a subset of Φ and Ψ. Conversely,
let Γ uf Φ and Γ uf Ψ. Then there are u, v ∈ X such that u  v, Γ ⊆ ufS (u) and
ufS (v) ⊆ Φ and there are x, y ∈ X such that x  y, Γ ⊆ ufS (x) and ufS (y) ⊆ Ψ. As
we know, there is a w ∈ X such that u  w  v and a z ∈ X with x  z  y. Note
that Γ ⊆ ufS (u) ∩ ufS (x) ⊆ ufS (w) ∩ ufS (z) ⊆ Φ ∩ Ψ. Since Γ 6= ∅, there is an ultrafilter
F both in ufS (u) and ufS (x). Thus, u ∈ F and x ∈ F . Because F is a filter, there is a
z ∈ F such that z v u and z v x. This means that u and x overlap and their infimum
is well-defined. So we can use ‘u u x’. It follows that ‘w u z’ is also well-defined. From
u  w and x  z we get that u u x  w and u u x  z. Then by MTS3, u u x  w u z.
Since F is a filter, uuv ∈ F iff u ∈ F and v ∈ F . Therefore, ufS (u)∩ufS (x) = ufS (uux).
In analogy, it follows that ufS (w)∩ufS (z) = ufS (w uz). From what has been established
above we obtain that Γ ⊆ ufS (u u x) and ufS (w u z) ⊆ Φ ∩ Ψ. So finally we arrive at
Γ uf Φ ∩ Ψ.
F
4. Let Φ ∈ Xuf , i.e. Φ ⊆ uf(S) and Φ 6= ∅. Because the top element X is element of
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 52

F F F
every ultrafilter, uf(S) = ufS ( X). Moreover, we have X  X. This follows from
F
MTS4. Thus we obtain Φ uf ufS ( X) for every Φ ∈ Xuf .

5. Let Φ uf Ψ. Again, there are u, v ∈ X such that u  v, Φ ⊆ ufS (u) and ufS (v) ⊆ Ψ.
Then by MTS5 there is a z ∈ X such that u  z  v. Hence, ufS (u) ⊆ ufS (z) ⊆ ufS (v).
Therefore, Φ uf ufS (z) and ufS (z) uf Ψ.

This result is not sufficiently strong on its own, because WMTS do not encode enough topo-
logical information to correspond to topological spaces. For this reason, (uf, corruf , [.]uf ) is not
a wmts-point representation in the strict sense. But theorem 4.1.2 can be extended. It allows
us to prove a stronger embedding theorem for normal mereotopological structures.

Theorem 4.1.3 (Embedding Theorem for NMTS). If S = (X, v, ) is an NMTS, then


ufS is an embedding of S into Suf and Suf is an atomistic NMTS.

Proof. Let S = (X, v, ) be an NMTS. Hence it is a WMTS as well. Therefore we know from
the last theorem that ufS is an MTS-embedding of S into Suf and that Suf is an atomistic
WMTS. We have to show that Suf is an NMTS as well, i.e. that for all Φ, Ψ ∈ Xuf , if Φ uf Ψ
and uf(S)\Ψ has an interior in Suf , then uf(S)\Ψ uf uf(S)\Φ.
Let Φ, Ψ ∈ Xuf such that Φ uf Ψ and let uf(S)\Ψ have an interior in Suf . Then there
are u, v ∈ X such that u  v, Φ ⊆ ufS (u) and ufS (v) ⊆ Ψ. Hence, uf(S)\ufS (u) ⊆ uf(S)\Φ
and uf(S)\Ψ ⊆ uf(S)\ufS (v). Because of the ultrafilter property we obtain ufS (−u) =
uf(S)\ufS (u) and ufS (−v) = uf(S)\ufS (v). Thus, ufS (−u) ⊆ uf(S)\Φ and uf(S)\Ψ ⊆
ufS (−v). Since uf(S)\Ψ has an interior in Suf there is some Γ ∈ Xuf such that Γ uf uf(S)\Ψ.
Hence there are w, z ∈ X such that w  z, Γ ⊆ ufS (w) and ufS (z) ⊆ uf(S)\Ψ ⊆ ufS (−v).
Therefore, ufS (z) ⊆ ufS (−v), so we arrive at w  z v −v. Summing up, we have that u  v
and −v has an interior in S. Since S is an NMTS, it follows from definition 3.1.26 that
−v  −u. Taking into account what we have concluded earlier [namely ufS (−u) ⊆ uf(S)\Φ
and uf(S)\Ψ ⊆ ufS (−v)], we finally arrive at uf(S)\Ψ uf uf(S)\Φ.

This theorem is significant. Since NMTS correspond to topological spaces, we can conclude
that (uf, corruf , [.]uf ) is indeed a nmts-point representation. So the theorem above shows that
the embedding problem has been solved for the nmts-point representation (uf, corruf , [.]uf ).
A similar but weaker result can be obtained for classical mereotopological structures. Some-
what surprisingly I have not been able to prove that every CMTS can be embedded into an
atomistic CMTS via ufS . I only have been able to prove that every compact CMTS can be
embedded in this way. The proof rests on the axiom of choice.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 53

Theorem 4.1.4 (Embedding Theorem for CMTS). If S = (X, v, ) is a compact


CMTS, then ufS is an embedding of S into Suf and Suf is an atomistic CMTS.

Proof. We know from theorem 4.1.2 that Suf is a WMTS and that the embedding ufS preserves
both v and  . The difficulty consists in showing that Suf is a CMTS as well, i.e. that uf
does indeed satisfy the axiom in definition 3.1.27. Here the assumption of compactness and
the axiom of choice is needed. Again we assume Φ, Ψ, Γ, ∆ ∈ Xuf .
We have to show:

(*) If not Φ uf Ψ, then there is an ∆ ∈ Xuf such that ∆ ⊆ Φ and ∆ ∩ Γ = ∅ for all
Γ uf Ψ.

First we show that (*) follows from the statement:


S
(**) If Φ ⊆ {Γ : Γ uf Ψ}, then Φ uf Ψ.

We derive the contraposition of (*) from (**): Assume that for every ∆ ∈ Xuf , if ∆ ⊆ Φ, then
there is a Γ uf Ψ such that ∆ ∩ Γ 6= ∅. We already know that (Xuf , ⊆) is a CMS. Thus we
S
can apply SSP as well as GSP and conclude that Φ ⊆ {Γ : Γ uf Ψ}. Then (**) yields:
Φ uf Ψ. Thus the contraposition of (*) follows form (**) and hence also (*) itself. Now we
still have to prove (**).
S
We assume Φ ⊆ {Γ : Γ uf Ψ}. Then for every F ∈ Φ there is a Γ uf Ψ such that F ∈ Γ.
From Γ uf Ψ we get that there are u, v ∈ X such that u  v, Γ ⊆ ufS (u) and ufS (v) ⊆ Ψ.
So for every F ∈ Φ there are u, v ∈ X such that u  v and F ∈ ufS (u) ⊆ ufS (v) ⊆ Ψ.
Now let g(F ) := {(u, v) ∈ X 2 : u  v and F ∈ ufS (u) ⊆ ufS (v) ⊆ Ψ}. Then g(F ) 6= ∅
for all F ∈ Φ. We now use the axiom of choice. Let h be a choice function from Φ to
g(Φ), that is: h(F ) ∈ g(F ) for all F ∈ Φ. We denote the first component of h(F ) by ‘uF ’
and the second component by ‘vF ’. Note that we have for all F ∈ Φ: uF  vF and F ∈
ufS (uF ) ⊆ ufS (vF ) ⊆ Ψ. Moreover, we know from corollary 3.1.10 that uF  intS (vF )  vF .
Therefore, ufS (uF ) ⊆ ufS (intS (vF )) ⊆ ufS (vF ) ⊆ Ψ, since MTS1 applies and the elements
are ultrafilters. Let U = {uF : F ∈ Φ} and O = {intS (vF ) : F ∈ Φ}. Observe that O is a set
of open regions, i.e. O ⊆ OS . Now we are going to show the following three statements; then
we are finished.
F F
(a) U O.
F
(b) Φ ⊆ ufS ( U ).
F
(c) ufS ( O) ⊆ Ψ.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 54

F F
Ad (a): We know that O ⊆ OS . Therefore, O ∈ OS (by theorem 3.1.6), hence O 
F F F F
O. Now we only have to show that U v O. This is easy using SSP and GSP: Let x◦ U ,
F
then there is an F ∈ Φ such that x◦uF . As has been established earlier, uF  intS (vF ) v O.
F F F
Thus, also x ◦ O. So U v O.
Ad (b): Let F ∈ Φ. Then, as noted before, F ∈ ufS (uF ). Thus uF ∈ F . Since uF ∈ U we
F F F
have uF v U . F is a filter, so U ∈ F as well. Hence, F ∈ ufS ( U ).
F
Ad (c): Observe that O is an open cover of O in S. Since S is compact, there is a finite
subcover O0 of O. Thus O v O0 . Because O0 ⊆ O, we have O0 v O as well. Thus,
F F F F F

O = O0 . Now let F ∈ ufS ( O), i.e. O ∈ F . Then O0 ∈ F . Since O0 is finite and


F F F F F

F is an ultrafilter there is an o ∈ O0 such that o ∈ F . The reason is this: O0 = {o1 , . . . , on }


for some n ∈ N, so since O0 = o1 t . . . t on ∈ F and F is an ultrafilter, o1 ∈ F or . . . or
F

on ∈ F . (A trivial induction proves this.) Hence, supposing that there is no o ∈ O0 such that
o ∈ F yields immediately a contradition. Thus there is an o ∈ O0 such that o ∈ F . Then it
follows that F ∈ ufS (o). Because o ∈ O, there is a G ∈ Φ such that o = intS (vG ). As has
been established above, ufS (intS (vG )) ⊆ ufS (vG ) ⊆ Ψ. Thus, ufS (o) ⊆ Ψ and we arrive at:
F ∈ Ψ.
Finally, we conclude from (a), (b) and (c) that Φ uf Ψ.

Up to now I have not been able to find out whether an embedding theorem for the ultra-
filter representation applied to CMTS can also be obtained without using the assumption
of compactness and the axiom of choice. Such as it is, the result about compact CMTS is
rather unsatisfying. Firstly, a point representation should not presuppose that the structure
of spacetime is compact. But the representation at hand does this for CMTS. Secondly, it is
not clear whether there are any non-trivial compact CMTS that are atomless. My conjecture
is that there are none. However, the embedding theorem 4.1.4 is only relevant for CMTS of
this kind, because a restriction to the atomistic case would render the representation of points
unnecessary. Moreover, it is not even clear whether there are any non-trivial atomless CMTS
at all. There are at least two trivial cases of atomless CMTS (corresponding to the trivial cases
of discrete and indiscrete topology on point sets). (i) Any atomless CMS can be extended to
an atomless CMTS just by identifying the structure’s relation of interior parthood with the
parthood relation. In this case every element of the CMTS is both open and closed. This is
a generalisation of the discrete topology in atomistic spaces. (ii) Any atomless CMS can be
extended to an atomless CMTS by defining the relation of interior parthood such that every
element in the domain is an interior part of the maximal element (universe), but apart from
that no other elements are related by interior parthood. This corresponds to the indiscrete
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 55

topology. The question whether there are atomless CMTS other than the described trivial
ones has not been answered up to now. But even if there were non-trivial atomless CMTS,
they had to be rather odd structures. I think that the notions ‘open’, ‘closed’ and ‘boundary’
loose their usually intended meaning in these non-trivial atomless CMTS. If such structures
exist, borders have to be admitted in their domains, because in every domain there must be
an element being not both open and closed (otherwise the structure would be trivial), and the
difference between its closure and its interior is a border then. But borders have to be atomless
in an atomless CMTS, and thus are in some sense extended entities.3 Then the question arises
how to draw the distinction between elements containing their border (closed elements) and
elements not containing their border (open elements), when borders themselves are ‘extended’
atomless entities and hence on a par with all other elements of the structure. I suspect that
there is no viable way of drawing the distinction.
However, these troubles do not mean that there are no non-trivial atomless CMTS, and
they also do not mean that trying to find and study such structures is a futile effort. It does
only mean that the notions ‘boundary’, ‘open’ and ‘closed’ are likely to loose their usual sense
when applied to elements of such structures. I cannot rule out the possibility that non-trivial
atomless CMTS may play a role in the search for theories about spatial or spatiotemporal
entities having no clear-cut boundaries. A basic idea for theories of this kind can be found in
Menger’s (1940) work on point-free topology:

“But well defined boundaries are themselves results of limiting processes rather
than objects of direct observation. Thus, instead of lumps, we might use at the
start something still more vague – something perhaps which has various degrees of
density or at least admits a gradual transition to its complement. Such a theory
might be of use for wave mechanics.” (1940, p. 107)

Yet, it is unclear how the standard conception of space or spacetime could be recovered from
such a theory involving non-trivial atomless CMTS. In any case, the ultrafilter method cannot
do the job, because it applies only to compact CMTS. Maybe other methods can be found,
but this is beyond the scope of this thesis, since there are no precise elaborations of Menger’s
idea and non-trivial atomless CMTS have not been found or constructed up to now.
Let us now survey the upshot of extending Stone’s theorem to mereotopological structures.
Although membership conditions for ultrafilters are formulated in terms of mereological con-
cepts alone and thus do not take into account additional topological structure, the ultrafilter
method nevertheless can be applied to mereotopological structures just as well as to purely
3
It may be that atoms are extended (especially in discrete structures), but it seems to me unconceivable
that there are atomless entities which have no extension.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 56

mereological structures. Not only CMS or Boolean algebras can be embedded into correspond-
ing atomistic structures using Stone’s ultrafilter representation, but also WMTS and, more
specifically, NMTS can be embedded into respective atomistic structures by the same repre-
sentation method. Additional topological structure imposed on the underlying mereological
structure is preserved by the representation just as well as purely mereological structure. Only
in the case of CMTS I have not been able to establish a general embedding theorem, but only a
restricted result. However, the problem might lie in the properties of CMTS rather than in the
ultrafilter method, because in the atomless case, CMTS are problematic per se. In contrast,
there are good examples of atomless WMTS and NMTS – and for these structures general
embedding theorems have been established. So Stone’s theorem is not limited to mereological
structures or Boolean algebras only, but applies also to some relevant mereotopological struc-
tures. Applied to WMTS and NMTS, the ultrafilter representation satisfies at least one of the
two adequacy conditions given in definition 3.2.6. The embedding problem can be solved for
the ultrafilter representation (w.r.t. WMTS, w.r.t. NMTS and w.r.t. compact CMTS). This
result contributes to an elaboration of the ultrafilter method.
Yet, the approach of representing points by ultrafilter is not free of problems. In the next
section we show that it is not generally adequate, despite its being fruitful in consequences.

4.1.3 The Problem of Free Ultrafilters


Solving the embedding problem is only half of what has do be done in order to decide whether
the ultrafilter representation is adequate. The other problem that has to be solved is the
idempotence problem. Actually, this problem turns out to be fatal for the ultrafilter method.
In the case of atomistic CMS, it is to be expected that there is a natural bijection between
points and atoms, i.e. for atomistic S it should be possible to prove that ufS is not only an
embedding but an isomorphism between S and Suf . In particular, ufS has to be a bijection
between the atoms of the structure and the singletons of its ultrafilters. Only if this is the
cases, the idempotence problem is solvable and the ultrafilter method is adequate (see definition
3.2.6). But it turns out that if S is atomistic and infinite, then there are simply too many
ultrafilters: although every atom corrsponds to an ultrafilter, not every ultrafilter corresponds
to an atom. This has been pointed out by Gerla (1995) and discussed more closely by Mormann
(2009). Both Gerla and Mormann reject the ultrafilter method, because it yields too many
points. We will now prove that the ultrafilter method indeed is inadequate for that reason.

Theorem 4.1.5. Let S = (X, v) be an atomistic CMS. Then for every atom x in S there is
exactly one ultrafilter F in S such that x ∈ F .

Proof. Let x be an atom in S. Let G = {y ∈ X : x v y}. Then G is a filter and x ∈ G.


CHAPTER 4. ANALYSING POINT REPRESENTATIONS 57

By the ultrafilter lemma (lemma 4.1.2) there is an ultrafilter F such that G ⊆ F . So there
is an ultrafilter that contains x. We show that it is uniquely determined. Suppose F 0 is an
ultrafilter that contains x. Since x is an atom it follows for every y ∈ F 0 that x = x u y v y,
because F 0 is a filter. Hence for every y ∈ F 0 , x v y. Conversely, if x v y, then y ∈ F , because
x ∈ F 0 and F 0 is a filter. To sum up: y ∈ F 0 iff x v y. The same holds for F as well, because
F is also an ultrafilter that contains x. So we have y ∈ F iff y ∈ F 0 . Therefore, F = F 0 .

Theorem 4.1.6. Let S = (X, v) be an atomistic CMS and let X be infinite. Then there is
an ultrafilter F in S that has no lower bound in S.

Proof. (cf. Chang & Keisler, 1990, p. 167) First let F = {x ∈ X : atS (−x) is finite}. We
show that X being infinite entails that F is a filter. Let x, y ∈ F . Then atS (−x) and atS (−y)
are finite. Hence atS (−x t −y) and therefore atS (−(x u y)) are finite as well. Thus x u y ∈ F
and is a part of both x and y. Now let x ∈ F and x v y. Then atS (−x) is finite and
atS (−y) ⊆ atS (−x), because −y v −x. Hence atS (−y) is finite and so y ∈ F . Thus F is
a filter. F is called the Frechet-filter on (X, v). By the ultrafilter lemma (4.1.2) there is an
ultrafilter U such that F ⊆ U . Suppose U had a lower bound, i.e. there is an x ∈ X such that
x v y for all y ∈ U . Then x must be an atom. The reason is: Suppose that for some y, y @ x.
Then U ⊂ {z ∈ X : y v z}. And {z ∈ X : y v z} is of course a (principal) filter. Then U is
not a maximal filter. But U is an ultrafilter and hence (by lemma 4.1.1) also a maximal filter.
Contradiction! So x is an atom. But then atS (x) is finite and hence −x ∈ F ⊆ U . Then x
as well as −x had to be in U , which is impossible, because U is a filter. Therefore U has no
lower bound.

Corollary 4.1.3. Let S = (X, v) be an atomistic CMS. If for every ultrafilter F in S there
is an atom x in S such that x ∈ F , then X is finite.

Proof. Assume that for every ultrafilter F in S there is an atom x in S such that x ∈ F .
Suppose that X were infinite. The last theorem tells us that there is a ultrafilter F that has
no lower bound in S. If an atom is element of an ultrafilter, then it is a lower bound of it. So
F contains no atom, contrary to the assumption. Therefore X must be finite.

Corollary 4.1.4. Let S = (X, v) be an atomistic CMS. If X is infinite, then ufS is not a
bijection between at(S) and at(Suf ).

Proof. Let X be infinite. Then by corollary 4.1.3 there is some F ∈ uf(S) such that F ∈
/ ufS (x)
for all x ∈ at(S). Thus, ufS is not a bijection between the atoms in S and all singletons of
ultrafilters in S, where the latter are the atoms of Suf .
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 58

On the basis of these results, we can now actually prove that the approach of identifying points
with ultrafilters is only adequate in case the underlying CMS is finite. Hence, it is inadequate
in general.

Theorem 4.1.7. If Σ is a class of WMTS containing infinite, atomistic WMTS,


then (uf, corruf , [.]uf ) is not an adequate point representation for Σ.

Proof. We have to show that for every class Σ∗ of enriched WMTS of the same type, (uf, corruf , [.]uf )
is not an adequate Σ∗ -point representation for the class of infinite, atomistic WMTS. From
theorem 4.1.6 and its corollaries we know that for any infinite, atomistic CMS S, ufS is not a
bijection between the atoms in S and the atoms in the corresponding structure Suf . It follows
by definition 4.1.3 that for every infinite, atomistic WMTS S∗ , ufS∗ is not a bijection between
the atoms in S∗ and the atoms in the corresponding structure Suf ∗ . Therefore, it is not an
isomorphism between S∗ and Suf ∗ , for every infinite, atomistic WMTS S∗ . Let Σ∗ be a class
of enriched WMTS of the same type. In the case that the class of infinite, atomistic WMTS is
not a subset of Σ∗ , (uf, corruf , [.]uf ) is trivially not an adequate Σ∗ -point representation for the
class of infinite, atomistic WMTS. In the case that the class of infinite, atomistic WMTS is
a subset of Σ∗ , it follows by definition 3.2.6 that (uf, corruf , [.]uf ) is not an adequate Σ∗ -point
representation for the class of infinite, atomistic WMTS, because the class of infinite, atomistic
WMTS is not empty.

Theorem 4.1.8. If (uf∗ , corruf


∗ , [.]∗ ) is an extension of the wmts-point representation
uf
(uf, corruf , [.]uf ), then (uf∗ , corruf
∗ , [.]∗ ) is not a generally adequate point representation.
uf

Proof. Let (uf∗ , corruf


∗ , [.]∗ ) be an extension of the wmts-point representation (uf, corr , [.] ).
uf uf uf
Then for every class Σ of enriched WMTS, if (uf∗ , corruf
∗ , [.]∗ ) is a Σ-point representation,
uf
then (uf∗ , corruf
∗ , [.]∗ ) is an extension of the wmts-point representation (uf, corr , [.] ) for Σ.
uf uf uf
Assume for reductio that (uf∗ , corruf
∗ , [.]∗ ) is a generally adequate point representation. Then
uf
there are classes Σ, Σ∗ of enriched WMTS of the same type such that Σ∗ ⊆ Σ, (uf∗ , corruf
∗ , [.]∗ )
uf
is an adequate Σ-point representation for Σ∗ and the following conditions are satisfied: (1)
Σ∗ contains finite discrete Σ-structures; (2) Σ∗ contains infinite discrete Σ-structures; (3)
Σ∗ contains atomistic continuous Σ-structures; (4) Σ∗ contains atomless Σ-structures. Note
that (uf∗ , corruf
∗ , [.]∗ ) is a Σ-point representation then. Hence, (uf∗ , corr ∗ , [.]∗ ) is an exten-
uf uf uf
sion of the wmts-point representation (uf, corruf , [.]uf ) for Σ. Consequently, (uf∗ , corruf
∗ , [.]∗ )
uf
is also an extension of the wmts-point representation (uf, corruf , [.]uf ) for Σ∗ . Let Σ0 = {S ∈
Σ : S is a component of S∗ for some S∗ ∈ Σ∗ }. Then lemma 3.2.2 yields: (uf, corruf , [.]uf ) is
an adequate wmts-point representation for Σ0 . Observe that Σ0 contains infinite, atomistic
WMTS. Then it follows from theorem 4.1.7 that (uf, corruf , [.]uf ) is not an adequate point
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 59

representation for Σ0 . In particular, (uf, corruf , [.]uf ) is not an adequate wmts-point represen-
tation for Σ0 . This contradicts the previous result. So (uf∗ , corruf
∗ , [.]∗ ) cannot be a generally
uf
adequate point representation.

Corollary 4.1.5. (uf, corruf , [.]uf )is not a generally adequate point representation.

Proof. We only have to show that (uf, corruf , [.]uf ) is an extension of the wmts-point rep-
resentation (uf, corruf , [.]uf ). (Remember that we understand the notion of extension in a
non-strict sense.) Let Σ be an enriched WMTS such that (uf, corruf , [.]uf ) is a Σ-point repre-
sentation. It follows that Σ is a class of WMTS, because the functions in (uf, corruf , [.]uf ) are
only defined for WMTS. Note that it is trivial that for every S ∈ Σ: (1) pt(S) = pt(S); (2)
corr(F, x, S) = corr(F, x, S); (3) [S]pt is a component of [S]pt . Therefore, (uf, corruf , [.]uf ) is
an extension of the wmts-point representation (uf, corruf , [.]uf ) for Σ (see definition 3.2.3). Then
definition 3.2.4 yields that (uf, corruf , [.]uf ) is an extension of the wmts-point representation
(uf, corruf , [.]uf ).

Let us now summarise the upshot of the theorems and corollaries above. Although the
ultrafilter representation is adequate for the class of atomless CMS (because in this case
condition (2) in definition 3.2.6 is trivially satisfied), it is nevertheless inadequate for the class
of infinite, atomistic CMS. Therefore, it is in general an inadequate cms-point representation.
Moreover, note that the problem carries over to all mereotopological structures based on
CMS (e.g. WMTS, NMTS and CMTS). As we saw, it is possible to solve the ultrafilter
representation’s embedding problem for important classes of mereotopological structures. But
the ultrafilter method fails because the idempotence problem cannot be solved.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 60

4.2 Representing Points by Completely Prime Filters


As we have seen above, we cannot use all ultrafilters as representatives of points, because there
are too many of them in infinite structures. So it suggests itself to select only a subclass of
preferable ultrafilters as ersatz points. One of the most important ways of doing so consists
in taking only completely prime filters instead of all ultrafilters as representatives of points.
Completely prime filters are a special kind of ultrafilters, whose defining condition arises quite
naturally by strengthening the definiens in the definition of ‘is an ultrafilter in’. Remember
that ultrafilters are filters satisfying the condition that for every binary sum in the filter, one
of the summands is in the filter as well. This defining condition can be regarded as some sort
of a maximality or limit condition.4 In order to select a privileged subset from the class of all
ultrafilters, an even stricter condition is needed. We strengthen the mentioned condition by
extending it from finite sums to infinite sums. This is how we obtain completely prime filters.
Completely prime filters are filters satisfying the condition that for any arbitrary (possibly
infinite) sum in the filter, one of the summands is in the filter as well.

Definition 4.2.1. Let S = (X, v) be a PMS. F is a completely prime filter in S iff F is


F
a filter in S such that for all Y ⊆ X, if Y ∈ F , then there is some x ∈ Y with x ∈ F .

Representing points as completely prime filters is a common method in point-free topology5 .


Point-free topology is a branch of mathematics that studies topological problems by means
of applying algebraic methods to Heyting algebras – and not by means of directly analysing
topological spaces (i.e. sets of points) using set theoretic methods. This approach rest on
the simple insight that the open sets of any topological space form a Heyting algebra. Since
the open sets of a topological space are the bearers of topological information, it is justified
to ignore the underlying points and analyse the algebraic structure of open sets on its own.
Therefore, one analyses different kinds Heyting algebras (open set structures) instead of topo-
logical spaces in point-free topology. This approach is called ‘point-free’, because structures
of open sets are very likely to be atomless lattices.6 Then, of course, the issue arises whether
it is possible to find a corrsponding (i.e. category theoretic equivalent) class of topological
spaces for a given class of Heyting algebras (and vice versa). In order to deal with this issue
one has to construct topological spaces from Heyting algebras. For this purpose a method
of abstracting points from the underlying (possibly atomless) algebra is needed. The usual
method in point-free topology consists in taking the completely prime filters of a given Heyting
4
The reason is that, in the Boolean case, every filter satisfying the condition can be shown to be a maximal
filter. See lemma 4.1.1.
5
For an introduction see Johnstone (1982, 1983).
6
In fact, all topological spaces with no isolated points have atomless open set structures.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 61

algebra as representatives of points.


This method has been applied successfully. It has been shown that there is a specific class
of Heyting algebras (spatial locales) that correspond exactly to a specific class of topological
spaces (sober spaces) such that any given spatial locale L is isomorphic to the Heyting algebra
of open sets Ω(S(L)) of the sober space S(L) abstracted from L (cf. Johnstone, 1982, chapter
II).

Definition 4.2.2. Let S = (X, v) be an HMS.

– cpf(S) = {P : P is a completely prime filter in S}.



1 : U ∈ cpf(S), x is in the domain of S and x ∈ U.
– corrcpf (U, x, S) =
0 : otherwise.

– [S]cpf = (Xcpf , ⊆), whereXcpf = P(uf(S))\{∅}.

(cpf, corrcpf , [.]cpf ) is called ‘the completely prime filter representation’.

Corollary 4.2.1. (cpf, corrcpf , [.]cpf ) is a pms-point representation.

4.2.1 Advantages of Completely Prime Filters


The completely prime filter representation avoids some of the ultrafilter method’s disadvan-
tages. In particular it does not yield too many points: the idempotence problem can be solved
for CMS and for mereotopological structures based on CMS. Moreover, it can be shown that
for every CMTS S, the corresponding structure Scpf is also a CMTS, given that cpfS is an
embedding of S into Scpf (without having to assume compactness). So the problems that
rendered the ultrafilter representation inadequate do not affect the completely prime filter
representation. The following theorems show this in greater detail.

Lemma 4.2.1. Let S = (X, v) be an atomistic CMS. Then:

(1) For every atom x in S there is exactly one completely prime filter F in S such
that x ∈ F .

(2) For every completely prime filter F in S there is exactly one atom x in S such
that x ∈ F .

Proof. Ad 1.: Let x be an atom in S. Then F = {z ∈ X : x v z} is a completely prime filter.


Moreover, it is an ultrafilter and hence a maximal filter as well (see lemma 4.1.1). Suppose,
there is some other completely prime filter F 0 such that x ∈ F 0 . Then F ⊆ F 0 , because F 0 is
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 62

a filter. Since F is maximal, F = F 0 . Ad 2.: Let F be a completely prime filter in S. For all
F F
x ∈ X, x v at(S). So at(S) ∈ F , because F is a filter. Then it follows that there is some
atom x ∈ at(S) such that x ∈ F . Clearly, there cannot be a second atom x0 in F , because
otherwise both x and x0 are members of F , but do not intersect. But this is impossible since
F is a filter.

On the basis of this lemma we can prove that the completely prime filter representation indeed
yields an isomorphism when applied to an atomistic CMS. This means that the idempotence
problem of the completely prime filter representation for CMS is solved. Remember that
the ultrafilter representation turned out to be inadequate for the very reason that it was not
possible to prove an analogous theorem for it.

Theorem 4.2.1 (Idempotence Theorem for CMS ). If S = (X, v) is an atomistic CMS,


then cpfS is an isomorphism between S and Scpf .

Proof. Let S = (X, v) be an atomistic CMS. First we have to show that cpfS is a bijection
between X and Xcpf = P(cpf(S))\{∅}. Note that cpfS (x) = {F ∈ cpf(S) : x ∈ F } ∈ Xcpf
for all x ∈ X (see part 1 of lemma 4.2.1). So it is a function from X and Xcpf . Now let ∆
be a non-empty subset of cpf(S). Then it follows from part 2 of lemma 4.2.1 that for every
F ∈ ∆ there is exactly one atom a(F ) in S such that a(F ) ∈ F . Let A(∆) = {a(F ) : F ∈ ∆}.
F F
Then A(∆) ∈ X and ∆ = cpfS ( A(∆)). It is left to show that for all ∆, Γ ∈ Xcpf ,
F F F F
if A(∆) = A(Γ), then ∆ = Γ. Let ∆, Γ ∈ Xcpf such that A(∆) = A(Γ). Then
A(∆) = A(Γ), because sums of atoms are identical just in case they are sums of the same
atoms. It follows that for all F ∈ cpf(S), a(F ) ∈ A(∆) iff a(F ) ∈ A(Γ). Note that for all
F ∈ cpf(S), F ∈ ∆ iff a(F ) ∈ A(∆). Therefore, F ∈ ∆ iff F ∈ Γ. So ∆ = Γ. Thus cpfS is a
bijection between X and Xcpf .
It is left to sow that cpfS is structure-preserving. Let x, y ∈ X. Assume, x v y. Let
F ∈ cpfS (x). Then x ∈ F . Since x v y and F is a filter it follows that y ∈ F. Hence,
F ∈ cpfS (y). Therefore, cpfS (x) ⊆ cpfS (y). Conversely, assume cpfS (x) ⊆ cpfS (y). Let
z ∈ atS (x). Then it follows from part 1 of lemma 4.2.1 that there is exactly one completely
prime filter F in S such that z ∈ F . Because z v x and F is a filter, it follows that x ∈ F as
well. So F ∈ cpfS (x) and the assumption yields F ∈ cpfS (y). Thus y ∈ F . Note that z ∈ F as
well. Because z is an atom and F is a filter, it must be the case that z v y. So for all z ∈ X,
if z ∈ atS (x), then z ∈ atS (y). Therefore, x v y.

Another disadvantage of the ultrafilter representation is that we had to assume compactness


in order to show that the structure Suf is an atomistic CMTS (see the embedding theorem
for CMTS in section 4.1.2). But using the completely prime filter representation it is now
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 63

possible to show that the structure Scpf is an atomistic CMTS – and it is not necessary to
assume that S is compact. For the purpose of showing this we first extend the completely
prime filter representation so that it applies to mereotopological structures as well.

Definition 4.2.3. Let S∗ = (X, v, ) be a WMTS and let S = (X, v).

– cpf(S∗ ) = cpf(S).

– corrcpf (U, x, S∗ ) = corrcpf (U, x, S).

– [S∗ ]cpf = (Xcpf , ⊆, cpf ), where


Xcpf = P(cpf(S∗ ))\{∅} and
2 : for some u, v ∈ X: u  v & Φ ⊆ cpf (u) & cpf (v) ⊆ Ψ}.
cpf = {(Φ, Ψ) ∈ Xcpf S S

Corollary 4.2.2. (cpf, corrcpf , [.]cpf ) is a wmts-point representation.

Applying the so defined completely prime filter representation to CMTS, we can now show
that Scpf is an atomistic CMTS, given that cpfS is an embedding of S into Scpf .

Theorem 4.2.2. If S is a CMTS and cpfS is an embedding of S into Scpf , then Scpf is an
atomistic CMTS.

Proof. Let S = (X, v, ) be a CMTS and let cpfS be an embedding of S into Scpf . Then
clearly (Xcpf , ⊆) is an atomistic CMS. As in the case of the ultrafilter representation it also
follows that Scpf is a WMTS. But in contrast to using the ultrafilter representation it is now
possible to show that Scpf is a CMTS as well, i.e. that cpf satisfies the axiom in definition
3.1.27, without presupposing that S is compact. Note that compactness is not needed in step
(c). Let Φ, Ψ, Γ, ∆ ∈ Xcpf .
As in the proof of theorem 4.1.4 it is easy to show that our desideratum

(*) If not Φ cpf Ψ, then there is an ∆ ∈ Xcpf such that ∆ ⊆ Φ and ∆ ∩ Γ = ∅ for all
Γ cpf Ψ.

follows from:
S
(**) If Φ ⊆ {Γ : Γ cpf Ψ}, then Φ cpf Ψ.
S
So again it is sufficient to prove (**). Assume that Φ ⊆ {Γ : Γ cpf Ψ}. Then for every
F ∈ Φ there is a Γ cpf Ψ such that F ∈ Γ. From Γ cpf Ψ we get that there are u, v ∈ X
such that u  v, Γ ⊆ cpfS (u) and cpfS (v) ⊆ Ψ. So for every F ∈ Φ there are u, v ∈ X such
that u  v and F ∈ cpfS (u) ⊆ cpfS (v) ⊆ Ψ. Now let g(F ) := {(u, v) ∈ X 2 : u  v and F ∈
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 64

cpfS (u) ⊆ cpfS (v) ⊆ Ψ}. Then g(F ) 6= ∅ for all F ∈ Φ. We now use the axiom of choice. Let
h be a choice function from Φ to g(Φ), that is: h(F ) ∈ g(F ) for all F ∈ Φ. We denote the first
component of h(F ) by ‘uF ’ and the second component by ‘vF ’. Note that we have for all F ∈ Φ:
uF  vF and F ∈ cpfS (uF ) ⊆ cpfS (vF ) ⊆ Ψ. Moreover, we know from corollary 3.1.10 that
uF  intS (vF )  vF . Therefore, cpfS (uF ) ⊆ cpfS (intS (vF )) ⊆ cpfS (vF ) ⊆ Ψ, since MTS1
applies and the elements are filters. Let U = {uF : F ∈ Φ} and O = {intS (vF ) : F ∈ Φ}.
Observe that O is a set of open regions, i.e. O ⊆ OS . All that is left to show are the following
three statements.
F F
(a) U O.
F
(b) Φ ⊆ cpfS ( U ).
F
(c) cpfS ( O) ⊆ Ψ.
F
Ad (a): We know that O ⊆ OS . Therefore, O ∈ OS (from theorem 3.1.6), hence
F F F F
O  O. Now we only have to show that U v O. This is easy using SSP and
F
GSP: Let x ◦ U , then there is an F ∈ Φ such that x ◦ uF . As has been established earlier,
F F F F
uF  intS (vF ) v O. Thus, also x ◦ O. So U v O.
Ad (b): Let F ∈ Φ. Then, as noted before, F ∈ cpfS (uF ). Thus uF ∈ F . Since uF ∈ U we
F F F
have uF v U . F is a filter, so U ∈ F as well. Hence, F ∈ cpfS ( U ).
F F F
Ad (c): Observe that O is an open cover of O in S. Now let F ∈ cpfS ( O), i.e. O ∈ F .
Because F is completely prime it follows that o ∈ F for some o ∈ O. Then it follows that
F ∈ cpfS (o). Because o ∈ O, there is a G ∈ Φ such that o = intS (vG ). As has been established
above, cpfS (intS (vG )) ⊆ cpfS (vG ) ⊆ Ψ. Thus, cpfS (o) ⊆ Ψ and we arrive at: F ∈ Ψ.
Finally, we conclude from (a), (b) and (c) that Φ cpf Ψ

Similarly, it can also be shown that if S is a WMTS and cpfS is an embedding of S into Scpf ,
then Scpf is an atomistic WMTS. However, note that results of that kind do not solve the
embedding problem. They only say that under the assumption that cpfS is an embedding,
cpfS is an embedding with nice properties (that suffice to satisfy what is required in the first
condition in definition 3.2.6). These results do not imply that cpfS really is an embedding.
So let us now turn to the question whether the embedding problem can be solved for the
completely prime filter representation. In the next section we will encounter serious problems
concerning the embeddability of atomless structure into atomistic structures by means of the
completely prime filter representation.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 65

4.2.2 The Problem of Prime Elements


Despite its advantages, the completely prime filter representation suffers from serious problems
too. Its two major problems are:
(a) If the underlying structure is Boolean, then the existence of completely prime filters pre-
supposes the existence of just as many atoms in the underlying structure.
(b) If the underlying structure is a proper Heyting algebra, then WSP is violated. Then there
are open complements of atoms, but there are no atoms. So there is nothing that makes up the
difference between some specific objects and the universe. This means that there are objects
which are exactly like the universe except that one atom is missing. But the missing atom
does not exist. This is highly counter-intuitive for any part-whole relation.
In the following steps we go through the problems sketched above in more detail. First we
define the notion of prime element.

Definition 4.2.4. Let S = (X, v) be a PMS.


x is a prime element in S iff for all y, z ∈ X, if y u z v x, then y v x or z v x.

Remark. In some sense, prime elements can be regarded as the opposites of atoms. The prime
elements in the Heyting algebra of open subsets of a Euclidean space with standard topology
are exactly the complements of single points and thus open sets (cf. Gerla, 1995, p. 1019).
Next we import a theorem from point-free topology (cf. Gerla, 1995, p. 1019), which
concerns the existence of completely prime filter. It can be shown that completely prime
filters in an HMS (or Heyting algebra) correspond to prime elements.

Lemma 4.2.2. Let S = (X, v) be an HMS. Then F is a completely prime filter in S iff there
is an x ∈ X such that x is a prime element in S and F = {y ∈ X : not y v x}.

This fact gives rise to the dilemma that for every completely prime filter F in an HMS, either
(a) there is an atom corresponding to F , or (b) WSP is violated, because there is an object
corresponding to F where exactly one atom missing, but the missing atom does not exist, so
that nothing makes up the difference between this object and the universe.

Theorem 4.2.3. Let S be an HMS. If F is a completely prime filter in S, then either (a)
there is an atom x in S such that x ∈ F , or (b) S violates WSP.

Proof. Let F be a completely prime filter in S. Then it follows from the previous lemma 4.2.2
that there is an x ∈ X such that x is a prime element in S and F = {y ∈ X : not y v x}. We
distinguish the following cases.

Case 1. There is a y ∈ X such that y o x. If y is an atom, then y o x yields y ∈ F and we are


home. If y is not an atom, then there is a u ∈ X with u @ y.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 66

Case i. There is a v ∈ X such that v @ y and v ou. Because x is a prime element,


it follows that for all w, z ∈ X, if w u z v x, then w v x or z v x. Note
that (x t u) u (x t v) v x, but neither (x t u) v x nor (x t v) v x,
contradicting that x is u-irreducible. So this case cannot arise.
Case ii. There is no v ∈ X such that v @ y and v o u. Then S violates WSP,
because in view of u @ y it follows from WSP that there is a v ∈ X such
that v @ y and v o u.
F
Case 2. There is no y ∈ X such that y o x. Note that x 6= X, because otherwise F = ∅
F
and thus not a filter. So x @ X. Suppose WSP holds. Then it follows that there
is a y ∈ X such that y o x, contradicting the case assumption. Therefore, S violates
WSP in this case.

As a corollary from the theorem above we can derive the dilemma in its full form.

Corollary 4.2.3. Let S be an HMS. Then either (a) at(S) is equinumerous to cpf(S), or (b)
S violates WSP.

Proof. Let S = (X, v) be an HMS. Either S violates WSP or S satisfies WSP. We show
that, if S satisfies WSP, then there is a bijection between cpf(S) and at(S). Assume thatS
satisfies WSP. Then for every F ∈ cpf(S) there is an x ∈ X such that x is a prime element
F
in S and F = {y ∈ X : not y v x}, see lemma 4.2.2. It follows that x @ X. Thus, WSP
yields that there is some y ∈ X such that y o x. Then y is an atom and y ∈ F . This follows by
an analogous argument as in case 1.(i) of the preceding theorem’s proof. Since F is a filter,
it cannot contain more than one atom. Conversely, for every x ∈ at(S) there is exactly one
F ∈ cpf(S) such that x ∈ F (see the proof of statement 1 in lemma 4.2.1). Thus there is a
bijection between cpf(S) and at(S).

Both horns of the dilemma are unacceptable. Ad (a): That at(S) is equinumerous to cpf(S)
means that there are as many atoms in S as there are in Scpf . Hence, the representation of S
by means of the completely prime filter method can only be successful if S is already atomistic.
But a point representation method should not presuppose that. Ad (b): That S violates WSP
means that there are objects x and y in S such that x is a proper part of y, but y has no
proper part that is disjoint from x. So x and y are two different things, but there is nothing
that makes up the difference. This is unacceptable, because the domain of S is intended to be
a class of concrete basic entities (such as spacetime regions of processes) and it can certainly
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 67

be ruled out that one such basic entity is proper part of another, but nothing makes up the
difference.
This also affects the use of the completely prime filter representation in point-free topology.
Even if we assume that our basic entities have an atomless structure like open sets, the
approach still rests on the existence of completely prime filters and, hence, on the existence
of prime elements. But complements of points are just as dubious as points themselves. As
Gerla (1995) remarks, atoms are indirectly present through their open complements anyway.
So, why should we accept open regions where exactly one point is cut out, but reject points?
It seems that there is no good reason.
As consequence of these intuitive problems we can prove that the completely prime filter
method is not a generally adequate point representation and that no extension for richer
structures can be an adequate point representation. This is accomplished in the following
steps.

Corollary 4.2.4. If S is an atomless WMTS, then there are no completely prime filters in
S.

Proof. Let S = (X, v, ) be an atomless WMTS. Then (X, v) is an atomless CMS. So (X, v)
does not violate WSP and at(S) = ∅. Then it follows from corollary 4.2.3 that cpf((X, v)) = ∅.
So there are no completely prime filters in S.

Corollary 4.2.5. If S is an atomless WMTS, then cpfS is not an embedding of S into Scpf .

Proof. Let S = (X, v, ) be an atomless WMTS. Assume for reductio that cpfS is an embed-
ding of S into Scpf . Then cpfS has to be a function from X to Xcpf = P(cpf(S))\{∅}. Hence,
cpfS maps every x ∈ X to a non-empty subset of cpf(S). But we know from the previous
corollary 4.2.4 that cpf(S) = ∅. So there are no non-empty subsets of cpf(S). Thus, cpfS is
not a function from X to Xcpf and, therefore, also not an embedding of S into Scpf .

Theorem 4.2.4. If Σ is class of WMTS containing atomless WMTS, then (cpf, corrcpf , [.]cpf )
is not an adequate point representation for Σ.

Proof. Let Σ be a non-empty class of atomless WMTS. We have to show that for every
class Σ∗ of enriched WMTS of the same type, (cpf, corrcpf , [.]cpf ) is not an adequate Σ∗ -point
representation for Σ. Let Σ∗ be a class of enriched WMTS of the same type. If the Σ is
not a subset of Σ∗ , then (cpf, corrcpf , [.]cpf ) is not an adequate Σ∗ -point representation for Σ.
So assume that Σ ⊆ Σ∗ . It follows that Σ∗ is a class of WMTS, because its elements are of
the same type as those of Σ. Assume for reductio that (cpf, corrcpf , [.]cpf ) is not an adequate
Σ∗ -point representation for Σ. Hence, for every S ∈ Σ, ptS is an embedding of S into Spt .
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 68

Remember that Σ contains atomless WMTS. Hence, there is an atomless WMTS S such that
ptS is an embedding of S into Spt . But this is impossible due to the previous corollary 4.2.5.
Therefore, also in this case, (cpf, corrcpf , [.]cpf ) is not an adequate Σ∗ -point representation for
Σ.

It follows that the completely prime filter representation and all its extensions for enriched
WMTS are not generally adequate point representations.

Theorem 4.2.5. If (cpf∗ , corrcpf


∗ , [.]∗ ) is an extension of the wmts-point representation
cpf
(cpf, corrcpf , [.]cpf ), then (cpf∗ , corrcpf
∗ , [.]∗ ) is not a generally adequate point representation.
cpf

Proof. Let (cpf∗ , corrcpf


∗ , [.]∗ ) be an extension of the wmts-point representation (cpf, corr , [.] ).
cpf cpf cpf
So for every every class Σ of enriched WMTS, if (cpf∗ , corrcpf
∗ , [.]∗ ) is a Σ-point representation,
cpf
then (cpf∗ , corrcpf
∗ , [.]∗ ) is an extension of the wmts-point representation (cpf, corr , [.] )
cpf cpf cpf
for Σ. Suppose for reductio that (cpf∗ , corrcpf
∗ , [.]∗ ) is a generally adequate point repre-
cpf
sentation. Then there are classes Σ, Σ∗ of enriched WMTS of the same type such that
Σ∗ ⊆ Σ, (cpf∗ , corrcpf
∗ , [.]∗ ) is an adequate Σ-point representation for Σ∗ and the following
cpf
conditions are satisfied: (1) Σ∗ contains finite discrete Σ-structures; (2) Σ∗ contains infi-
nite discrete Σ-structures; (3) Σ∗ contains atomistic continuous Σ-structures; (4) Σ∗ con-
tains atomless Σ-structures. So (cpf∗ , corrcpf
∗ , [.]∗ ) is a Σ-point representation. Therefore,
cpf
(cpf∗ , corrcpf
∗ , [.]∗ ) is an extension of the wmts-point representation (cpf, corr , [.] ) for Σ.
cpf cpf cpf
Moreover, (cpf∗ , corrcpf
∗ , [.]∗ ) is an extension of the wmts-point representation (cpf, corr , [.] )
cpf cpf cpf
for Σ∗ as well. Let Σ0 = {S ∈ Σ : S is a component of S∗ for some S∗ ∈ Σ∗ }. Then it fol-
lows by lemma 3.2.2 that (cpf, corrcpf , [.]cpf ) is an adequate wmts-point representation for Σ0 .
Note that Σ0 contains atomless WMTS. Therefore, (cpf, corrcpf , [.]cpf ) is not an adequate point
representation for Σ0 . This follows from theorem 4.2.4. In particular, (cpf, corrcpf , [.]cpf ) is not
an adequate wmts-point representation for Σ0 . Contradiction! So (cpf∗ , corrcpf
∗ , [.]∗ ) cannot
cpf
be a generally adequate point representation.

Corollary 4.2.6. (cpf, corrcpf , [.]cpf ) is not a generally adequate point representation.

Proof. We only have to show that (cpf, corrcpf , [.]cpf ) is an extension of the wmts-point rep-
resentation (cpf, corrcpf , [.]cpf ). Let Σ be an enriched WMTS such that (cpf, corrcpf , [.]cpf ) is
a Σ-point representation. Consequently, Σ is a class of WMTS, because the functions in
(cpf, corrcpf , [.]cpf ) are only defined for WMTS. Clearly, for every S ∈ Σ: (1) pt(S) = pt(S);
(2) corr(F, x, S) = corr(F, x, S); (3) [S]pt is a component of [S]pt . Therefore, definition 3.2.3
yields: (cpf, corrcpf , [.]cpf ) is an extension of the wmts-point representation (cpf, corrcpf , [.]cpf )
for Σ (see ). Thus, we get by definition 3.2.4 that (cpf, corrcpf , [.]cpf ) is an extension of the
wmts-point representation (cpf, corrcpf , [.]cpf ).
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 69

The upshot of this section is that the completely prime filter representation works only for
atomistic CMS and mereotopological structures based on atomistic CMS. The idempotence
problem can be solved. But the completely prime filter method fails, because the embedding
problem cannot be solved. Atomless CMS and mereotopological structures based on atomless
CMS cannot be embedded into corresponding atomistic structures by means of the completely
prime filter method. The reason is that the existence of completely prime filters depends on
the existence of prime elements in the underlying structure. The existence of prime elements
in turn presupposes that there are either atomic basic entities or the underlying structure
violates WSP.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 70

4.3 Representing Points by Maximal Round Filters


As we have seen up to now, neither ultrafilters nor completely prime filters can be used
as adequate representatives of points. The most important explanation of this fact is that
the definientia of the definitions of ‘ultrafilter’ and ‘completely prime filter’ are formulated
exclusively in terms of mereological concepts and do not take into account any topological
relations or properties.
In the case of ultrafilters, this is problematic mainly for two reasons. The first reason is
that different ultrafilters may represent the same point, namely when they are collocated. For
example, this is the case when ultrafilters are built from two infinite chains all of whose elements
touch (but do not contain) a common point. Consider the Euclidean plane R2 endowed with
its standard topology and, furthermore, the atomless CMS (RO(R2 )\{∅}, ⊆) of regular open
subsets of R2 . Let C1 be the set of all open discs dε = {(x, y) ∈ R2 : (x − ε)2 − y 2 < ε2 } with
ε > 0, and let C2 be the set of all open discs d∗ε = {(x, y) ∈ R2 : (x + ε)2 − y 2 < ε2 } with
ε > 0. Then C1 and C2 are chains in (RO(R2 )\{∅}, ⊆). See the following illustration.

It follows from the ultrafilter lemma that there are ultrafilters F1 and F2 in (RO(R2 )\{∅}, ⊆)
containing C1 and C2 , respectively. Intuitively, F1 and F2 represent the same point, namely
(0, 0). However, it is actually the case that F1 6= F2 , because F1 and F2 contain disjoint
elements. So there are different ultrafilters (in fact, uncountably many)7 that correspond to
only one point in R2 , because they are collocated. As Roeper (1997, p. 308) points out:

“Taking ultrafilters as points amounts to assuming that ultrafilters are never col-
located with one another, and hence that non-overlapping regions are never con-
nected with one another; i.e. it amounts to the absence of topological structure.
7
For every point of R2 , there are uncountably many ultrafilters corresponding to it. As above, construe
chains of open discs for every given point and observe that we can rotate these pairs of chains around their
point of symmetry.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 71

Not surprisingly then, the Stone space of a Boolean algebra turns out to be a
totally disconnected space in the sense of point-set topology.”

The second reason is that there may be ultrafilters which do not represent points at all. In
particular, ultrafilters that contain only unbounded or unlimited basic entities do not represent
points. Here is an example in the Euclidean plane R2 . Let C3 be the set of all open regions
rε = {(x, y) ∈ R2 : y 2 + ε = x} for ε > 0, which are enclosed by parabolas of the form
y 2 + ε = x. The following diagram illustrates this chain.

Then, again using the ultrafilter lemma, it follows that there is an ultrafilter F3 containing C3 .
But because all elements of F3 must be limited (unbounded), F3 does not represent a point of
R2 . So there are ultrafilters (again, uncountably many) that do not represent points.
In the case of completely prime filters, the strong and purely mereological condition in the
definiens rules out specific filters which intuitively represent points by virtue of the topological
relations between their elements. In atomless WMTS, there are no completely prime filters,
because the definiens of the definition of ‘completely prime filter’ is so strong (see corollary
4.2.4). The only filters that could possibly represent points in an atomless WMTS are filters of
the aforementioned kind, which are not completely prime filters, but which have nice topologi-
cal properties. For example, consider the Euclidean plane R2 again, and consider the atomless
WMTS (RO(R2 )\{∅}, ⊆, ), where U  V iff clR2 (U ) ⊆ intR2 (V ) for all U, V ∈ RO(R2 )\{∅}.
Let C4 be the set of all concentric open discs dε = {(x, y) ∈ R2 : x2 − y 2 < ε2 } with ε > 0.
Then C4 is a chain in (RO(R2 )\{∅}, ⊆). Moreover, C4 is also a chain in (RO(R2 )\{∅}, ).
The illustration below visualises this.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 72

According to the ultrafilter lemma, there is an ultrafilter F4 in (RO(R2 )\{∅}, ⊆) having C4 as


a subset. From an intuitive point of view, F4 represents the point (0, 0), because F4 contains
an infinitely descending chain with respect to the topological relation  (namely C4 ), whose
elements dε converge8 towards {(0, 0)} for ε → ∞. But there are no completely prime filters
in the atomless CMS (RO(R2 )\{∅}, ⊆) and so F4 is not a completely prime filter. This shows
that in atomless WMTS there are no completely prime filters at all, but there may be filters
that intuitively corrspond to points (due to topological relations among their elements).
These exemplifications of the problems associated with the ultrafilter method and the
completely prime filter method suggest that we have to take into account mereological as well
as topological relations for characterising representatives of points. In particular, we have to
deal with the problem of collocation and the problem of unbounded filters. There are different
ways of doing so. One way is to represent point by equivalence classes of ultrafilters that
contain at least one bounded element (where collocation serves as equivalence relation). This
option has been suggested by Roeper (1997). Another way consists in using only maximal
filters of a certain kind rather than ultrafilters, namely maximal limited round filters, which
contain at least one bounded element and for every one of their elements they also contain
a non-tangential part of it. An advocate of this approach is Mormann (2008). Roeper also
employs the method of representing points by maximal limited round filters. In this section
we will only discuss the second method, because both methods have been analysed extensively
by Roeper (1997) and shown to be equivalent within a particular framework, namely Roeper’s
theory of region-based topology.
Let us first examine Roeper’s framework. He works with a specific kind of mereotopological
structures, which he calls ‘region-based topologies’. These structures consist of a complete
Boolean algebra, a connection relation C and a class of limited or bounded regions Λ. Roeper’s
definition (1997, pp. 255–257) goes as follows:
8
The concept of convergence used here can be based on Hausdorff distance.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 73

Definition 4.3.1. S = (X, v, C, Λ) is a region-based topology (short: RBT) iff (X, v) is


a complete Boolean algebra (with 0 as neutral element w.r.t t), C is a binary relation on X
and Λ ⊆ X such that:

(RS1) For all x, y ∈ X, if xCy, then yCx.

(RS2) For all x, y ∈ X, if x 6= y, then xCx.

(RS3) For all x ∈ X, not 0Cx.

(RS4) For all x, y, z ∈ X, if xCy and y v z, then xCz.

(RS5) For all x, y, z ∈ X, if xCy t z, then xCy or xCz.

(RS6) 0 ∈ Λ.

(RS7) For all x, y ∈ X, if x ∈ Λ and y v x, then y ∈ Λ.

(RS8) For all x, y ∈ X, if x, y ∈ Λ, then x t y ∈ Λ.

(RS9) For all x, y ∈ X, if xCy, then there is a z ∈ Λ such that z v y and xCz.

(RS10) For all x, y ∈ X, if x ∈ Λ, y 6= 0 and x  y, then there is z ∈ Λ such that z 6= 0


and x  z  y.9

Note, in particular, that a class of limited basic entities (Λ) has to be used as additional com-
ponent. The property of being limited cannot be reduced to the mereological and topological
relations (without further assumptions).
Apparently, Roeper’s region-based topologies are quite different from the mereotopological
structures that we have analysed up to now. Firstly, every RBT is based on a Boolean algebra
and thus contains a 0-element that is part of every other entity in its domain. Secondly, the
topological structure of an RBT is given by a relation of connection and not by an interior
parthood relation. However, it is easy to show that it is possible to reformulate Roeper’s
axioms using a relation of interior parthood () instead of connection. It turns out that
Roeper’s axioms RS1 – RS5 are equivalent to the axioms that define NTMS, given the reduction
principle ‘x  y iff not xC − y’. We will use this alternative axiomatic basis, because it ties
in with our other mereotopological systems. Of course, it is also possible to use classical
mereological structures instead of complete Boolean algebras as basis for Roeper’s structures,
given appropriate changes that are required by deleting the 0-element.
9
x  y iff not xC − y.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 74

Let us first define the notion of a mereotopological structure which is enriched by infor-
mation about limitedness. The following is to be understood as a minimal notion, which only
comprises very basic bridge principles for parthood and limitedness.

Definition 4.3.2. S = (X, v, , Λ) is an WMTS with information about limitedness


iff (X, v, ) is an WMTS such that:

(LIM1) For all x, y ∈ X, if x ∈ Λ and y v x, then y ∈ Λ.

(LIM2) For all x, y ∈ X, if x, y ∈ Λ, then x t y ∈ Λ.

Now we can define Roeper-structures very easily as normal mereotopological structures with
information about limitedness that are satisfying some additional bridge principles for interior
parthood and limitedness.

Definition 4.3.3. S = (X, v, , Λ) is a Roeper-structure (short: RMTS) iff


(X, v, , Λ) is an NMTS with information about limitedness such that:

(RMTS1) For all x, y ∈ X, if xCy, then there is a z ∈ Λ with z v y and xCz.10

(RMTS2) For all x ∈ X, there is a y ∈ Λ with y  x.

(RMTS3) For all x, y ∈ X, if x ∈ Λ and x  y, then there is a z ∈ Λ with x  z  y.

So Roeper’s system turns out to be not that different from the mereotopological systems which
we have analysed up to now. The basic difference is that Roeper-stuctures have to satisfy quite
strong principles concerning interior parthood. Especially, RMTS2 is very strong and comes
near to atomlessness, although it does not imply it. There are atomistic RMTS as well. (We
will return to this issue later.)
Every region-based topology can be turned into an RMTS (by deleting the 0-element) and
every RMTS can be turned into a region-based topology (by adding a 0-element). We show this
using (Tarski’s) theorem 3.1.1. For every complete Boolean algebra (X, v) with neutral ele-
ment 0, the corresponding CMS is cms(X, v) = (X\{0}, {(x, y) ∈ (X\{0})2 : x v y}); and for
every CMS (X, v) and every 0 ∈
/ X, the corresponding complete Boolean algebra is cba0 (X, v
) = (X ∪ {0}, {(x, y) ∈ (X ∪ {0})2 : x v y or x = 0}). Moreover, we define the RMTS corre-
sponding to a given region-based topology and the region-based topology corresponding to a
given RMTS as follows. For every region-based topology
S = (X, v, C, Λ) with neutral element 0, let rmts(S) = (cms(X, v), C , Λ\{0}) with C =
{(x, y) ∈ (X\{0})2 : not xC − y}; and for every RMTS S = (X, v, , Λ) an every 0 ∈
/ X,
let rbt(S) = (cba0 (X, v), C , Λ ∪ {0}) with C = {(x, y) ∈ (X ∪ {0})2 : not x  −y}. It
follows that:
10 F
xCy iff not x  −y or y = X.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 75

(1) If S is an RBT, then rmts(S) is an RMTS.

(2) If S is an RMTS, then rbt(S) is an RBT.

Roeper’s theorems and proofs are always about RBT. However, it is easy to get analogous
theorems and proofs for RMTS. The above described way of mapping RBT to RMTS and
RMTS to RBT helps to transfer statements about RBT to RMTS and vice versa.
Let us now come to the charaterisation of a kind of filters that can be used as representatives
for points. Since the main problem with ultrafilters and completely prime filters consists in the
fact that these kinds of filters do not take into account topological relations and limitedness,
it suggests itself to look for a kind of filters that does take into account topological structure
and limitedness in a suitable way. In the following two definitions we introduce the notion of
limited round filter and the method of representing points by maximal limited round filters.

Definition 4.3.4. Let S = (X, v, , Λ) be a WMTS with information about limitedness.

– F is a round filter in S iff F is a filter in S such that for all x ∈ F , there is a y ∈ F


with y  x.

– F is a limited round filter in S iff F is a round filter in S such that F ∩ Λ 6= ∅.

– F is a maximal limited round filter in S iff F is a limited round filter in S such


that there is no limited round filter G in S with F ⊂ G.

So why should maximal limited round filters be adequate representatives of points? Limited-
ness clearly rules out the case that a filter fails to represent a point because of containing only
unlimited elements (in which case a filter intuitively does not represent a point). Roundness
rules out the case that a filter fails to represent a point because of being built from an infinite
chain of open elements all of whom touch, but do not contain a certain point. Moreover,
roundness rules out collocation and hence it rules out the case that different filters represent
the same point because of being collocated.

Definition 4.3.5. Let S = (X, v, , Λ) be a WMTS with information about limitedness.

– mrf(S) = {F : F is a maximal limited round filter in S}.



1 : U ∈ mrf(S); x ∈ U or for some u: u v x and u  v for all v ∈ U.
– corrmrf (U, x, S) =
0 : otherwise.

– [S]mrf = (Xmrf , ⊆, mrf , Λmrf ), where


Xmrf = P(mrf(S))\{∅};
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 76

2 : for some u, v ∈ X: u  v, Φ ⊆ mrf (u), mrf (v) ⊆ Ψ};


mrf = {(Φ, Ψ) ∈ Xmrf S S

Λmrf = {Φ ∈ Xmrf : for some u ∈ Λ : Φ ⊆ mrfS (u)}.

We call the triple (mrf, corrmrf , [.]mrf ) of these functions ‘the round filter representation’.

Corollary 4.3.1. Let Σ be the class of WMTS with information about limitedness. Then
(mrf, corrmrf , [.]mrf ) is a Σ-point representation.

Roeper’s approach essentially consists in applying the round filter representation to RMTS. In
the following steps we will examine how this works. In the first step, we associate a topological
space with every RMTS by means of the round filter representation.

Definition 4.3.6. Let S = (X, v, , Λ) be an RMTS.


TS = (PSmrf , OS
mrf
) with
mrf
PS = mrf(S);
mrf
OS is the family of unions of subfamilies of {mrfS (x) : x ∈ X} ∪ {∅}.

Roeper proves that for every RBT, the associated topological space is a locally compact
Hausdorff space. The same holds also for RMTS.

Lemma 4.3.1. If S is an RMTS, then TS is a locally compact Hausdorff space.

Proof. See the proofs of theorem 4.9 and theorem 4.10 by Roeper (1997).

We need two more lemmas about the topology induced by a given RMTS for proving the main
theorems below.

Lemma 4.3.2. If S = (X, v, , Λ) is an RMTS, then for all U ⊆ PSmrf :


U is regular open in TS iff U = ∅ or U = mrfS (x) for some x ∈ X.

Proof. See the proofs of lemma 4.7 and theorem 4.3 by Roeper (1997).

Lemma 4.3.3. If S = (X, v, , Λ) is an RMTS, then for all x having an interior in S:


mrfS (x) = intTS (mrfS (−x)c ).

Proof. See the proofs of lemma 4.7(c) and definition 4.1 by Roeper (1997).

As we saw, every RMTS corresponds to a locally compact Hausdorff space. But it works
also the other way round. Every locally compact Hausdorff space can be associated with a
canonical RMTS.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 77

Definition 4.3.7. Let T be a locally compact Hausdorff space.


ST = (XT , ⊆, T , ΛT ), where rmts(S)
XT = RO(T )\{∅};
T = {(U, V ) ∈ XT2 : clT (U ) ⊆ intT (V )};
ΛT = {U ∈ XT : U ⊆ V for some V that is compact in T }.

Lemma 4.3.4. If T is a locally compact Hausdorff space, then ST is an RMTS.

Proof. See theorem 5.6 by Roeper (1997) and, in particular, the proofs of theorems 5.2, 5.3,
5.4 and 5.5 (ibid.).

In fact, the method of assigning an RMTS to a given locally compact Hausdorff space specified
in the definition above differs slightly from Roeper’s original approach. Roeper takes the
regular closed sets rather than the regular open sets of locally compact Hausdorff spaces as
the elements of associated mereotopological structures. This choice leads to some technical
simplifications. But, I think that from an intuitive point of view it is better to use regular
open sets. The reason is that it can be shown that every element of an RMTS is identical to
the sum of its interior parts.
F
Lemma 4.3.5. If S = (X, v, , Λ) is an RMTS, then for all x ∈ X, x = {y ∈ X : y  x}.

Analogously, every regular open set of a topological space is identical to the union of its interior
parts — but regular closed sets are in general not identical to the union of their interior parts.
Of course, regular closed sets are identical to the closure of this union, which is the right join
operation in the mereological structure of regular open sets. But nonetheless, the analogy is
more straightforward in the case of regular open sets.
Let us now turn to one of Roeper’s main theorems. It states that any given RMTS is
isomorphic to the RMTS which corresponds to the locally compact Hausdorff space of the
original RMTS.

Lemma 4.3.6. If S is an RMTS, then S ∼


= STS .

Proof. See theorem 5.9 by Roeper (1997).


More specifically, we can make note of the fact that mrfS is an isomorphism between S and
STS .
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 78

Lemma 4.3.7. If S = (X, v, , Λ) is an RMTS, then for all x, y ∈ X:

(1) x v y iff mrfS (x) ⊆ mrfS (y).

(2) x  y iff mrfS (x) mrf mrfS (y).

(3) x ∈ Λ iff clTS (mrfS (x)) is compact in TS .

Proof. For (1) see lemma 4.2 and lemma 4.5 in Roeper (1997). For (3) see Roeper’s theorem
4.8. Statement (2) can be obtained as follows: (⇒) If x  y, then trivially mrfS (x) mrf
mrfS (y) (see the definition of mrf ). (⇐) If mrfS (x) mrf mrfS (y), then there are u, v ∈ X
such that u  v and mrfS (x) ⊆ mrfS (u) and mrfS (v) ⊆ mrfS (y). Using (1) we can conclude
that x v u and v v v. So it follows that x  y (by MTS2).

Now we can prove an embedding theorem for RMTS. By means of mrfS , every RMTS can be
embedded into an atomistic WMTS with information about limitedness.

Theorem 4.3.1 (Embedding Theorem for RMTS ). If S is an RMTS, then mrfS is an


embedding of S into Smrf and Smrf is an atomistic WMTS with information about limitedness.

Proof. Let S =(X, v, , Λ) be an RMTS. It follows immediately that (Xmrf , ⊆) is an atomistic


CMS (lemma 3.1.4). Moreover, it follows from lemma 4.3.7 above that for all x, y ∈ X, x v y
iff mrfS (x) ⊆ mrfS (y); x  y iff mrfS (x) mrf mrfS (y). It also follows easily that for all
x ∈ X: x ∈ Λ iff mrfS (x) ∈ Λmrf . Let x ∈ Λ. Then trivially mrfS (x) ∈ Λmrf (see definition
4.3.5). Conversely, assume that there is a u ∈ Λ with mrfS (x) ⊆ mrfS (u). Then we get x v u.
So LIM1 yields: x ∈ Λ.
It is still left to show that Smrf is a WMTS with information about limitedness. Note that
it is trivial that Smrf satisfies the conditions MTS1 and MTS2.
Ad MTS3: Let Φ, Ψ, Ψ0 ∈ Xmrf such that Φ mrf Ψ and Φ mrf Ψ0 . Then there are
u, v, u0 , v 0 ∈ X with (a) u  v and Φ ⊆ mrfS (u) ⊆ mrfS (v) ⊆ Ψ, (b) u0  v 0 and Φ ⊆
mrfS (u0 ) ⊆ mrfS (v 0 ) ⊆ Ψ0 . Note that Φ is not empty. So there is some maximal limited round
filter F ∈ Φ. Because Φ ⊆ mrfS (u) and Φ ⊆ mrfS (u0 ), it follows that u ∈ F and u0 ∈ F . F
is a filter. Hence, u ◦ u0 . Therefore, ‘u u u0 ’ is defined. Now we show that Φ ⊆ mrfS (u u u0 ).
Let F ∈ Φ. Then F ∈ mrfS (u) and F ∈ mrfS (u0 ). Hence, u ∈ F and u0 ∈ F . Remember that
F is a filter. Thus, u u u0 ∈ F . So F ∈ mrfS (u u u0 ). So far so good. u u u0 is a common
part of u and u0 . Observe that it therefore follows from u  v and u0  v 0 that u u u0  v
and u u u0  v 0 . Then, since S is a WMTS, u u u0  v u v 0 . Now let F ∈ mrfS (v u v 0 ), i.e.
v u v 0 ∈ F . So v, v 0 ∈ F , because v u v 0 v v, v u v 0 v v 0 and F is a filter. So F ∈ mrfS (v)
and F ∈ mrfS (v 0 ). That is, mrfS (v u v 0 ) ⊆ mrfS (v) ∩ mrfS (v 0 ). For set theoretic reasons,
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 79

mrfS (v) ∩ mrfS (v 0 ) ⊆ Ψ ∩ Ψ0 . Therefore, Φ ⊆ mrfS (u u u0 ) ⊆ mrfS (v u v 0 ) ⊆ Ψ ∩ Ψ0 and


u u u0  v u v 0 . Thus, Φ mrf Ψ ∩ Ψ0 .
Ad MTS4: Let Φ ∈ Xmrf . Because S is an WMTS, it follows that Φ ⊆ mrf(S) =
F F F
mrfS ( X) and X  X. Therefore, there is a Ψ ∈ Xmrf such that for every Φ ∈ Xmrf ,
Φ mrf Ψ.
Ad MTS5: Let Φ, Ψ ∈ Xmrf such that Φ mrf Ψ. Then there are u, v ∈ X with u  v
and Φ ⊆ mrfS (u) ⊆ mrfS (v) ⊆ Ψ. Because S is a WMTS, we can conclude that there is an
x ∈ X with u  x  v. Moreover, it follows that Φ ⊆ mrfS (u) ⊆ mrfS (x) ⊆ mrfS (v) ⊆ Ψ.
So Φ mrf mrfS (x) mrf Ψ.
Ad LIM1: Let Φ ∈ Λmrf and Ψ ⊆ Φ. Then there is some u ∈ ∆ with Φ ⊆ mrfS (u). Since
Ψ ⊆ Φ, we get Ψ ∈ Λmrf immediately.
Ad LIM2: Let Φ, Ψ ∈ Λmrf . Then there are u, v ∈ Λ such that Φ ⊆ mrfS (u) and Ψ ⊆
mrfS (v). Note that Φ ∪ Ψ ⊆ mrfS (u) ∪ mrfS (v) ⊆ mrfS (u t v). Because S is a WMTS with
information about limitedness, it follows that u t v ∈ Λ. Therefore, Φ ∪ Ψ ∈ Λmrf .

Note that there are two ways of associating a given RMTS with a topological space. Firstly,
there is the topological space TS that arises from an RMTS by Roeper’s round filter rep-
resentation (see definition 4.3.6). Secondly, the structure Smrf is an atomistic WMTS and,
thus, gives rise to the topological space Top(Smrf ), whose topology is determined by the atom-
istic structures’ interior parthood relation mrf (see definition 3.1.23). So the question arises
whether these topological spaces are homeomorphic. The following theorem shows that the
answer is affirmative.

Lemma 4.3.8. Let T = (X, O) be a locally compact Hausdorff space. Then for every U ∈ O
and every x ∈ U , there is a V ∈ RC(T ) such that

(1) x ∈ intT (V ),

(2) V ⊆ U , and

(3) V is compact in T .

Proof. See lemma 5.4. in Roeper (1997).

Theorem 4.3.2. If S is an RMTS, then TS ∼


= Top(Smrf ).

Proof. Remember that TS = (PSmrf , OS


mrf
), where PSmrf = mrf(S) and OS
mrf
is the family
of unions of subfamilies of {mrfS (x) : x ∈ X} ∪ {∅}. Remember also that Top(Smrf ) =
(at(Smrf ), Oat (Smrf )), where at(Smrf ) = {{F } : F ∈ mrf(S)} and Oat (Smrf ) = {atSmrf (Φ) :
Φ is open in Smrf }∪{∅}. Obviously, the function which maps every point of TS to its singleton
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 80

is a bijection between PSmrf and at(Smrf ). We have to prove for all Φ ⊆ mrf(S), Φ is open
in TS just in case atSmrf (Φ) is open in Top(Smrf ). We first show that for every Φ ⊆ mrf(S),
S
Φ is open in TS iff Φ = {mrfS (x) : mrfS (x) ⊆ Φ}. We know that Φ is open in TS iff
S
Φ = intTS (Φ). We show that intTS (Φ) = {mrfS (x) : mrfS (x) ⊆ Φ}. (⊆) Let F ∈ intTS (Φ).
mrf
Then, F ∈ Ψ for some Ψ ∈ OS with Ψ ⊆ Φ. Lemma 4.3.8 and lemma 4.3.1 together yield
that there is a ∆ ∈ RC(TS ) such that F ∈ intTS (∆) and ∆ ⊆ Ψ. Then intTS (∆) is regular
open in TS , F ∈ intTS (∆) and intTS (∆) ⊆ Φ. It follows from lemma 4.3.2 that there is an x in
S
the domain of S such that mrfS (x) = intTS (∆). Therefore, F ∈ {mrfS (x) : mrfS (x) ⊆ Φ}.
S
(⊇) Let F ∈ {mrfS (x) : mrfS (x) ⊆ Φ}. Then there is an x in the domain of S such that
F ∈ mrfS (x) and mrfS (x) ⊆ Φ. Then F is trivially in intTS (Φ), because mrfS (x) is open in
TS . So far so good.
Now observe that for every non-empty Φ ⊆ mrf(S), atSmrf (Φ) is open in Top(Smrf ) iff Φ is
S
open in Smrf (by theorem 3.1.3). We also know that Φ is open in Smrf iff Φ = {Ψ ∈ Xmrf :
S S
Ψ mrf Φ}. So it is only left to show that {Ψ ∈ Xmrf : Ψ mrf Φ} = {mrfS (x) : mrfS (x) ⊆
S
Φ}. (⊆) Let F ∈ {Ψ ∈ Xmrf : Ψ mrf Φ}. Then for F ∈ Ψ for some Ψ ∈ Xmrf with Ψ mrf Φ.
Then there are x, y in the domain of S such that x  y and Ψ ⊆ mrfS (x) ⊆ mrfS (y) ⊆ Φ.
Thus, there is an x in the domain of S such that F ∈ mrfS (x) and mrfS (x) ⊆ Φ. Hence,
S S
F ∈ {mrfS (x) : mrfS (x) ⊆ Φ}. (⊇) Let F ∈ {mrfS (x) : mrfS (x) ⊆ Φ}. Then for some x
in the domain of S: F ∈ mrfS (x) and mrfS (x) ⊆ Φ. So x ∈ F . Since F is a round filter, there
is a y ∈ F such that y  x. Then F ∈ mrfS (y) ⊆ mrfS (x) ⊆ Φ. It follows that mrfS (y)  Φ.
S
Therefore, F ∈ {Ψ ∈ Xmrf : Ψ mrf Φ}.

With the help of the previous theorem and lemma 4.3.1, it follows immediately that the
topological space given by the atomistic WMTS Smrf is a locally compact Hausdorff space.

Corollary 4.3.2. If S is an RMTS, Top(Smrf ) is a locally compact Hausdorff space.

We can prove an idempotence theorem for those atomistic RMTS S, where the corresponding
topological space TS , which is given by the round filter representation, is a discrete space.

Theorem 4.3.3 (Idempotence Theorem for discrete RMTS ). If S is an atomistic


RMTS such that TS is discrete, then mrfS is an isomorphism between S and Smrf .

Proof. Let S = (X, v, , Λ) be an atomistic RMTS such that TS = (PSmrf , OS


mrf
) is a discrete
space. We show that Smrf = STS .
Let us first examine the implications for STS = (XTS , ⊆, TS , ΛTS ) more closely. First
note that for all Φ, Ψ ∈ XTS : Φ TS Ψ iff clT (Φ) ⊆ intT (Ψ) iff Φ ⊆ Ψ, because in discrete
spaces a set is identical to its interior and closure. Since in discrete spaces, compactness coin-
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 81

cides with finiteness, it furthermore follows that ΛTS = {Φ ∈ XTS : clT (Φ) is compact in T } =
{Φ ∈ XTS : Φ is finite}. Therefore, STS = (XTS , ⊆, ⊆, {Φ ∈ XTS : Φ is finite}).
Now recall that S ∼ = STS . Because mrfS is an isomorphism between S and STS (lemma
4.3.7), it follows for all x, y ∈ X that: x  y iff mrfS (x) TS mrfS (y) iff mrfS (x) ⊆ mrfS (y)
iff x v y. Moreover, for all x ∈ X: x ∈ Λ iff mrfS (x) ∈ ΛTS iff mrfS (x) is finite iff atS (x)
is finite. So S = (X, v, v, {x ∈ X : atS (x) is finite}). This tells us that discrete RMTS are
rather trivial structures.
Now let us examine Smrf more closely. The first conclusion about Smrf that we draw from
what has just been said about S is that for all Φ, Ψ ∈ Xmrf the following chain of equivalences
holds: Φ mrf Ψ iff
iff there are u, v ∈ X such that u  v and Φ ⊆ mrfS (u) ⊆ mrfS (v) ⊆ Ψ
iff there are u, v ∈ X such that u v v and Φ ⊆ mrfS (u) ⊆ mrfS (v) ⊆ Ψ
iff there are u, v ∈ X such that Φ ⊆ mrfS (u) ⊆ mrfS (v) ⊆ Ψ
iff Φ ⊆ Ψ.
The last step is justified by the fact that for every non-empty Φ ⊆ mrf(S), there is an x ∈ X
with mrfS (x) = Φ; which in turn is justified because (a) TS is a discrete space and, hence,
RO(TS ) = P(mrf(S)); (b) RO(TS ) = {mrfS (x) : x ∈ X} ∪ {∅}, see lemma 4.3.2.
The second conclusion about Smrf that we draw from the above insight about S is that
for all Φ ∈ Xmrf : Φ ∈ Λmrf iff
iff there is an x ∈ Λ such that Φ ⊆ mrfS (x)
iff there is an x ∈ X such that atS (x) is finite and Φ ⊆ mrfS (x)
iff there is an x ∈ X such that mrfS (x) is finite and Φ ⊆ mrfS (x)
iff Φ is finite.
The last step is again justified by the reason given for the last step in the preceding chain of
equivalences. We can sum up that Smrf = (Xmrf , ⊆, ⊆, {Φ ∈ Xmrf : Φ is finite}). Note that
Xmrf = P(mrf(S))\{∅} = XTS , because TS is discrete. We finally conclude that Smrf = STS .
∼ ST . Thus, we also have S ∼
We know that S = = Smrf .
S

Example. If (X, v) is an (infinite) atomistic CMS, then S = (X, v, v, {x ∈ X : atS (x) is finite})
is an atomistic RMTS such that TS is a discrete space.

Finally, we can conclude from the preceding theorems that Roeper’s round filter representation
is an adequate point representation for the class of all RMTS that are either discrete or
atomless.

Corollary 4.3.3. (mrf, corrmrf , [.]mrf ) is an adequate point representation for the class of all
RMTS S that are either atomless or atomistic such that TS is discrete.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 82

Proof. This follows immediately from the embedding theorem for RMTS, the idempotence
theorem for discrete RMTS and definition 3.2.6.

This result is important. It shows that the round filter representation is the first representation
method which we have examined up to now that can adequately deal with both atomistic and
atomless mereotopological structures. The ultrafilter representation and the completely prime
filter representation failed either for atomistic or for atomless structures. So, compared to
them, the round filter representation is quite successful.
But one might still wonder why the idempotence theorem above is restricted to those
RMTS S where TS is a discrete space. Is it not possible to prove that the round filter
representation is an adequate point representation for the class of all RMTS? In the light of
Roeper’s theorem which says that mrfS is an isomorphism between S and STS , it might seem
that this should be the case. However, it is not so. In the next section we will see why not
and which other problems concerning atomistic structures there are in Roeper’s theory.

4.3.1 Problems with Atomistic Structures


The class of those RMTS S with discrete TS is a particularly important subclass of the class
of all atomistic RMTS. Nonetheless, there are other – in some sense pathological – cases of
atomistic RMTS as well. This leads to the following problem. It is not possible to prove
an idempotence theorem for all atomistic RMTS, because there are infinite, atomistic RMTS
that do not really have topological structure or information about limitedness, namely RMTS
where the relation of interior parthood is identified with the parthood relation and all elements
in the domain are taken as limited. Such structures satisfy the axioms for RMTS, but are
nothing more than bare CMS.11 Thus, maximal limited round filters in such structures are just
ultrafilters in the underlying mereological structure. Therefore, the problem of free ultrafilters
arises in infinite atomistic RMTS of that kind. For this reason, the round filter representation
is not an adequate point representation for the class of all RMTS.

Corollary 4.3.4. It is not the case that (mrf, corrmrf , [.]mrf ) is an adequate point representation
for the class of all RMTS.

Proof. Let (X, v) be a CMS with infinitely many elements. Then S = (X, v, v, X) is an
RMTS. Observe that for all F ⊆ X, F is a maximal limited round filter in S iff F is an
ultrafilter in (X, v). Hence, it follows from theorem 3.1.6 and its corollaries that mrfS is
11
But note that those atomistic RMTS where the relation of interior parthood is identified with the relation of
parthood and where limitedness is identified with having only finitely many atoms are unproblematic, because
they are endowed with a strong notion of limitedness. And this is sufficient to prove an idempotence theorem
for this kind of discrete RMTS.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 83

not a bijection between the atoms in S and the atoms in the corresponding structure Smrf .
Thus, mrfS not an isomorphism between them either. Since S is an element of the class of
all RMTS, (mrf, corrmrf , [.]mrf ) is not an adequate point representation for the class all RMTS
(see definition 3.2.6).

The problem is that Roeper’s axioms for RMTS do not rule out the degenerated cases of
discrete RMTS described above. Roeper is aware of that fact (cf. 1997, p. 282). To rule out
such degenerated RMTS, Roeper studies and adds further axioms to his mereotopological
system (cf. 1997, sections 6 and 7). But in doing so, he completely rules out atomism. All
RMTS satisfying the additional axioms are atomless. As Bostock remarks, “he [Roeper] also
adds a further postulate about boundedness which affirms the infinite divisibility of our space,
and which therefore goes against Russell’s own intentions.” (2010, p. 28)
Naturally, the question suggests itself whether it is possible to rule out degenerated disc-
trete RMTS without ruling out atomism (and thus throwing out the baby with the bath
water). Unfortunately, this is not the possible within Roeper’s framework, because his axioms
are so strong that they allow only the existence of atomistic structures that are topologically
disconnected. The reason is that Roeper presupposes that the elements of his mereotopo-
logical structures correspond to regular open sets (or regular closed sets) in locally compact
Hausdorff spaces. So, atomistic RMTS correspond to atomistic families of regular open sets in
topological spaces of that kind. But all Hausdorff spaces in which no one-point set is open have
atomless families of regular open sets (cf. Roeper, 1997, p. 283), for example all topological
manifolds. Roughly speaking, atomistic RMTS are on the verge of discreteness. All RMTS
with rich topological structure are atomless. It can be shown that every RMTS is perforce
atomless if (a) every element other than the universe is connected to is complement, i.e. none
of its elements other than the universe is an interior part of itself, and (b) it has more than
one element. Roeper’s discovered and proved this fact, see theorem 6.1 (1997, p. 283). Of
course, condition (a) is not satisfied by discrete structures. But both conditions, (a) and (b),
are satisfied by all reasonable structures that are continuous and where  is intended as a
relation of non-tangential interior parthood. In particular, they are also satisfied by atomistic
continuous structures such as the enriched NMTS SRn = (P(Rn )\{∅}, ⊆, Rn , ΛRn ) given by
the Euclidean topology on Rn , with A Rn B iff clRn (A) ⊆ intRn (B) and ΛRn contains the
non-empty subsets of every compact set in Rn . So Roeper’s mereotopology is unable to deal
with atomistic continuous structures.
The upshot of this section is that there are two serious problems with Roeper’s approach.
(1) It is not possible to prove an idempotence theorem for all atomistic RMTS. (2) Some
particularly important atomistic structures are ruled out by Roeper’s axioms, namely contin-
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 84

uous atomistic structures. The first problem is not an obstacle when it comes to the question
whether the round filter representation is generally adequate. But the second problem shows
that the class of RMTS cannot be used to prove that the round filter representation is generally
adequate.

4.3.2 General Adequacy of the Round Filter Representation


But the problem mentioned above does not imply that the round filter representation is not
generally adequate. If it can be proven that the round filter representation is adequate for
a class of mereotopological structures which covers all relevant kinds of structures spacetime
might have, then the round filter representation will still turn out to be generally adequate.
For this purpose it is not necessary that such a class of structures is determined by a unified
system of axioms.12 Of course, it is preferable to find a system of axioms that is on the one
hand strong enough to allow proving idempotence and embedding theorems and on the other
hand weak enough to include structures of all relevant kinds. However, finding such a set of
axioms is a problem that can be separated from the problem of finding a generally adequate
representation method. We will come back to that point later on.
Let us now turn to a preciser elaboration of the ideas sketched above. Let us call a class
of mereotopological structures with information about limitedness ‘a nice class of enriched
WMTS’ if it is suitable for establishing the general adequacy of the round filter representation
and, for that, covers all relevant kinds of structures space might have. In the following, I
propose conditions that are sufficient for determining a nice class of enriched WMTS. Then,
after providing some lemmas, I show that the existence of a nice class of enriched WMTS
indeed implies that the round filter representation is generally adequate (which is the main
theorem of this section). In the last step, I prove that there is a nice class of WMTS and draw
the obvious conclusions.

Definition 4.3.8. Σ is a nice class of enriched WMTS iff Σ is a class of WMTS with
information about limitedness satisfying the following conditions:

(1) Σ contains finite discrete WMTS S with information about limitedness.

(2) Σ contains infinite discrete WMTS S with information about limitedness.

(3) Σ contains atomistic continuous WMTS S with information about limitedness


such that for all x, y in the domain of S, x  y iff clS (x) v intS (y).
12
As we saw in the preceding section, Roeper’s axioms cover discrete and atomless structures (including
some pathological cases of discrete structures), but rule out atomistic continuous structures. Hence, they do
not serve this purpose.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 85

(4) Σ contains atomless WMTS S with information about limitedness such that for
every x in the domain of S, there is some y ∈ Λ with y  x.

(5) For all S = (X, v, , Λ) ∈ Σ, if S is atomistic, then for every x ∈ X, x ∈ Λ iff


there is some y such that y is compact13 in S and x v y.

The definiens of the above definition does not rule out that nice classes of enriched WMTS
contain other (possibly more exotic) kinds of structures than the relevant kinds explicitly
mentioned in (1)–(4). In order to enable us to refer only to the core of relevant structures
within a nice class, we introduce the following predicate.

Definition 4.3.9. Σ is the class of relevant test cases of Σ0 iff Σ0 is a nice class of enriched
WMTS and Σ is the class of all S ∈ Σ0 such that S is either (1) a finite discrete Σ0 -structure,
(2) an infinite discrete Σ0 -structure, (3) an atomistic continuous Σ0 -structure S such that for
all x, y in the domain of S, x  y iff clS (x) v intS (y), or (4) an atomless Σ0 -structure S such
that for every x in the domain of S, there is some y ∈ Λ with y  x.

Before we can proof the main theorems, we need a couple of auxiliary definitions and lemmas.
The next definition introduces a function that maps every atomistic WMTS with informa-
tion about limitedness to an isomorphic copy whose atoms are the singletons of the original
structure’s atoms. This device will simplify some technical work in the proofs later on.

Definition 4.3.10. Let S = (X, v, , Λ) be an atomistic WMTS with information about


limitedness. Sat = (P(at(S))\{∅}, ⊆, at , Λat ) with
F F
U at V iff U  V and
F
U ∈ Λat iff U ∈ Λ for all non-empty U, V ⊆ at(S).

Lemma 4.3.9. Let S = (X, v, , Λ) be an atomistic WMTS with information about limited-
ness. Then atS is an isomorphism between S and Sat .

Proof. Corollary 3.1.5 says that atS is an isomorphism between (X, v) and (P(at(S))\{∅}, ⊆).
So it is only left to show that atS is structure preserving with respect to , at and Λ, Λat .
F F
Let x, y ∈ X. We know from corollary 3.1.5 that x = atS (x) and y = atS (y). Hence:
F F F
x  y iff atS (x)  atS (y) iff atS (x) at atS (y). Furthermore, x ∈ Λ iff atS (x) ∈ Λ
iff atS (x) ∈ Λat .

Lemma 4.3.10. Let S = (X, v, , Λ) be an atomistic WMTS with information about limited-
ness. Then for every F , F is a maximal limited round filter in S iff atS (F ) = {atS (x) : x ∈ F }
is a maximal limited round filter in Sat .
13
Remember that the notion of compactness is definable for atomistic WMTS, see definition 3.1.24.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 86

Proof. This follows immediately from lemma 4.3.9.

Furthermore, we need a principle that allows us to extend a given limited round filter to a
maximal limited round filter, just as the ultrafilter lemma allows us to extend any given filter
to an ultrafilter.

Lemma 4.3.11 (Round Filter Lemma). Let S = (X, v, , Λ) be an atomistic WMTS with
information about limitedness. Then for every limited round filter F in S there is some
maximal limited round filter F 0 such that F ⊆ F 0 .

Proof. We use Zorn’s lemma and the general proof idea of the ultrafilter lemma 4.1.2. Let F
be a limited round filter in S and let F = {F 0 : F 0 is limited round filter in S and F ⊆ F 0 }.
S
Then (F, ⊆) is a partial order. Let C be a chain in (F, ⊆). It follows that C is a limited
round filter in S. Let x, y ∈ C. Then for some F 0 , F 00 ∈ C: x ∈ F 0 and y ∈ F 00 . Without
S

loss of generality, let F 0 ⊆ F 00 . Then x, y ∈ F 00 . Hence, x u y ∈ F 00 , because F 00 is a filter in


S. Therefore, x u y ∈ C. Let x ∈ C and x v y. Then there is some F 0 ∈ C with x ∈ F 0 .
S S

Since F 0 is a filter, y ∈ F 0 as well. So y ∈ C. Hence, C is indeed a filter in S. Now note


S S

that C 6= ∅. So there is some limited round filter F ∈ C. Because F is limited, there is some
S S S
x ∈ F with x ∈ Λ. Thus, x ∈ C and C is limited too. C is also a round filter. Let
0 0 0
S
x ∈ C. Then there is an F ∈ C with x ∈ F . Because F is a round filter, there is some
y ∈ F 0 such that y  x. So there is a y ∈ C with y  x. Consequently, C is a limited
S S
S
round filter in S. Moreover, C is an upper bound of C. Thus, every chain in (F, ⊆) has an
upper bound which is itself in (F, ⊆). Using Zorn’s lemma we get that (F, ⊆) has a maximal
element, which is a maximal limited round filter F 0 such that F ⊆ F 0 .

The next lemma is due to Roeper (1997) and says that the elements of a maximal round filter
in the RMTS of regular closed sets in a locally compact Hausdorff space all contain a single
point of the underlying topological space. This result will be also very helpful in the proof of
the following theorem 4.3.4.

Lemma 4.3.12. Let T = (X, O) be a locally compact Hausdorff space and let SRC (T ) =
(RC(T )\{∅}, ⊆, RC , ΛRC ) with U RC V iff clT (U ) ⊆ intT (V ) and U ∈ ΛRC iff U is
compact in T , for all non-empty U, V ⊆ RC(T ). Then for every maximal limited round filter
F in SRC (T ), there is exactly one x ∈ X such that F = {U ∈ RC(T ) : x ∈ intT (U )}.

Proof. See lemma 5.6 and lemma 5.7 in Roeper’s paper (1997, p. 277).

Now an important and powerful theorem can be derived. The round filter representation is
adequate for the class of relevant test cases of every nice class of enriched WMTS.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 87

Theorem 4.3.4. For every nice class Σ of enriched WMTS, (mrf, corrmrf , [.]mrf ) is an adequate
point representation for the class of relevant test cases of Σ.

Proof. Let Σ be a nice class of enriched WMTS. Let S be a member of the class of relevant
test cases of Σ. Then S is either a finite discrete Σ-structure, an infinite discrete Σ-structure,
an atomistic continuous Σ-structure such that for all x, y in the domain of S, x  y iff
clS (x) v intS (y), or an atomless Σ-structure such that for every x in the domain of S, there
is some y ∈ Λ with y  x.
Case 1 : Let S = (X, v, , Λ) be a finite or an infinite discrete Σ-structure. Then S is
atomistic and, hence, the notion of compactness is defined for S (see definition 3.1.24).

1. Preliminary observations: Since Σ is a nice class of enriched WMTS and S ∈ Σ, it


follows that for all x ∈ X, x ∈ Λ iff for some y, y is compact in S and x v y.
(a) Then lemma 3.1.4 yields that for all y ∈ X, y is compact in S iff atS (y) is compact
in Top(S). Because S is a discrete Σ-structure, Top(S) is a discrete space. In discrete
spaces, compactness coincides with finiteness. So in view of corollary 3.1.5, it follows
that for all x ∈ X, x ∈ Λ iff atS (x) is finite.
(b) Moreover, Top(S) being a discrete space implies that every singleton subset of at(S)
is open in Top(S). Thus, for every x ∈ at(S), x is open in S (by theorem 3.1.3).
Therefore, for every x ∈ at(S), x  x.
(c) For every x ∈ at(S), let F (x) := {u : x  u}. We show that F (x) is a maximal
limited round filter in S with x ∈ F (x), for every atom x in S. Let x ∈ at(S). First
note that x ∈ F (x), because x  x, as shown in (b). Next, let u, v ∈ F (x). Then x  u
and x  v. MTS3 yields: x  u u v. Therefore, u u v ∈ F (x). Now let u ∈ F (x) and
u v v. Then x  u and, by MTS2, x  v. So v ∈ F (x) as well. Thus, F (x) is a filter in
S. Moreover, F (x) is a round filter, because x ∈ F (x). Hence, if u ∈ F (x), then there
is some v ∈ F (x) with v  u. F (x) is also limited. We know that x ∈ F (x), that x is
an atom in S and, thus, atS (x) is finite. By (a) we therefore get that x ∈ Λ. So F (x)
contains a limited element. Now we prove that F (x) is maximal. Assume for reductio
that there is a maximal limited round filter G in S such that F (x) ⊂ G. Then there is
some u ∈ G with u ∈
/ F (x). Moreover, since x ∈ F (x) ⊂ G, we can conclude that x ∈ G.
Because G is a filter, u u x ∈ G. But since x is an atom in S, it follows that x = u u x.
So x v u. Since x  x (see preliminary observations), we get by MTS2 that x  u.
But then u ∈ F (x). Contradiction! So F (x) is a maximal limited round filter in S such
that x ∈ F (x).
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 88

2. Embedding: We show that mrfS is an embedding of S into Smrf . First note that for
every x ∈ X, mrfS (x) 6= ∅. Let x ∈ X. S is atomistic. So there is an atom y in S with
y v x. Then F (y) ∈ mrf(S) and y ∈ F (y), see preliminarly observation (c). Therefore,
F (y) being a filter and y v x yield: x ∈ F (y). Thus, F (y) ∈ mrfS (x). Therefore, mrfS
is a function from X to Xmrf . In the following steps we prove that mrfS is structure
preserving and an embedding. Let x, y ∈ X.
(a) We show that x v y iff mrfS (x) ⊆ mrfS (y). (⇒) Let x v y. Let F ∈ mrfS (x). Then
x ∈ F , or for some u: u v x and u  v for all v ∈ F . If x ∈ F , then y ∈ F as well, since
F is a filter and x v y. So in this case, F ∈ mrfS (y). If for some u: u v x and u  v
for all v ∈ F , then we get by x v y that for some u: u v y and u  v for all v ∈ F .
And thus, F ∈ mrfS (y). (⇐) Let mrfS (x) ⊆ mrfS (y). Let u ∈ atS (x). Then, as before,
preliminary observation (c) yields: F (u) ∈ mrfS (x). So, according to the assumption,
F (u) ∈ mrfS (y). Therefore, either y ∈ F (u) or for some u0 : u0 v y and u0  v for all
v ∈ F . It follows in both cases that u v y, since it must be the case that u = u0 . So
u ∈ atS (y). Consequently, atS (x) ⊆ atS (y) and, finally, x v y.
(b) Next we show that x  y iff mrfS (x) mrf mrfS (x). (⇒) Let x  y. Then
mrfS (x) ⊆ mrfS (y). So it follows by definition of mrf that mrfS (x) mrf mrfS (y). (⇐)
Conversely, let mrfS (x) mrf mrfS (y). Again, by definition of mrf , we get that there
are u, v ∈ X such that u  v and mrfS (x) ⊆ mrfS (u) ⊆ mrfS (v) ⊆ mrfS (y). Using the
result we have just established in (a), we can conclude that x v u and v v y. Therefore,
x  y (by MTS2).
(c) Now we show that x ∈ Λ iff mrfS (x) ∈ Λmrf . (⇒) Let x ∈ Λ. Then trivially by
the definition of Λmrf : mrfS (x) ∈ Λmrf . (⇐) Let mrfS (x) ∈ Λmrf . Then for some y ∈ Λ,
mrfS (x) ⊆ mrfS (y). Using (a) we get x v y. So it follows with the help of LIM1 that
x ∈ Λ.
(d) Note that mrfS (X) = {mrfS (x) : x ∈ X} ⊆ Xmrf . Then clearly, mrfS is a surjective
function from X onto mrfS (X). Let mrfS (x) = mrfS (y). From (a) we get: x v y and
y v x, i.e. x = y. Hence, mrfS is also injective on mrfS (X). So mrfS is a bijection
between X and mrfS (X). Therefore, mrfS is an embedding of S into Smrf .

3. Idempotence: We show that mrfS is a bijection between at(S) and at(Smrf ). This result
implies that it is a bijection between X and Xmrf and, hence, an isomorphism between
S and Smrf .
Let x ∈ at(S). Then F (x) ∈ mrf(S) and x ∈ F (x), as shown in part (c) of the
preliminary observations. Hence, also F (x) ∈ mrfS (x). Of course, there cannot be
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 89

another maximal limited round filter F 0 with F 0 ∈ mrfS (x). If F 0 ∈ mrfS (x), then
x ∈ F 0 or for every y ∈ F 0 , x  y. In the first case, F (x) ⊆ {u : x v u} ⊆ F 0 , and in
the second case, F 0 ⊆ F (x). So in both cases, F 0 = F (x), because F 0 is maximal and
F (x) is maximal. Finally, we can conclude that for every x ∈ at(S), there is exactly one
F ∈ mrf(S) such that F ∈ mrfS (x).
Conversely, let F ∈ mrf(S). We have to show that there is exactly one x ∈ at(S) such
that F ∈ mrfS (x). Because F is limited there is some y ∈ F with y ∈ Λ. We know from
the preliminary observations (1.) that atS (y) is finite then. Hence F has a lower bound
u in (X, v) with u v y. Since S is atomistic, atS (u) is not empty. So we can take some
x with x ∈ atS (u). We know from above that F (x) is maximal limited round filter in
S. Since u is a lower bound of F in (X, v), it follows that x v u v v for all v ∈ F .
Remember that x is an atom and, hence, x  x. MTS2 yields: x  v for all v ∈ F . So
v ∈ F (x) for all v ∈ F , i.e. F ⊆ F (x). Because F is maximal, F = F (x). We know that
x ∈ F (x) and so x ∈ F . Therefore, F ∈ mrfS (x). Observe that there cannot be more
than one atom with this property. If x0 is an atom in S with F ∈ mrfS (x0 ), then x0 ∈ F
or for all y ∈ F : x0  y. In the first case, F being a filter implies that x ◦ x0 . Thus,
x = x0 , because they are atoms. In the second case, F ⊆ F (x0 ). Since F is maximal,
F = F (x0 ). We know that x0 ∈ F (x0 ) and, thus, x0 ∈ F . So we have reduced the second
case to the first case. Therefore, in every case, x = x0 . Thus, for every F ∈ mrf(S),
there is exactly one x ∈ at(S) with F ∈ mrfS (x).

4. Since S ∼
= Smrf and S is a discrete WMTS with information about limitedness, Smrf is
also a discrete WMTS with information about limitedness.

Case 2 : Let S = (X, v, , Λ) be an atomistic continuous Σ-structure such that for all x, y in
the domain of S, x  y iff clS (x) v intS (y). In this case we first have to show that mrfS is an
embedding from S into [S]mrf . Then we show that mrfS is a bijection as well. Before we start,
note that Top(S) is a topological manifold, because S is an atomistic continuum. Therefore,
it is a locally compact Hausdorff space, for every x ∈ at(S), {x} is closed in Top(S) and,
hence, x is closed in S (see lemma 3.1.1).

1. Preliminary observation: If x ∈ at(S), then F (x) := {y : x  y} is a maximal limited


round filter in S with F (x) ∈ mrfS (x). Moreover, there are no other such filters. That
is, mrfS (x) = {F (x)} for every atom x in S. These statements can be shown as follows.
Let x ∈ at(S). First we show that F (x) is indeed a maximal limited round filter in S.
Assume that u, v ∈ F (x). Then x  u and x  v. So x  u u v (because of MTS3).
Hence, u u v ∈ F (x) as well. Now let u ∈ F (x) and u v v. Then, x  u. So by
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 90

MTS2: x  v. Therefore, v ∈ F (x). Thus, F (x) is a filter in S. Moreover, F (x) is


a round filter, because for every y ∈ F (x), x  y. Because Top(S) is locally compact
and x ∈ at(S), there is some U ∈ Oat (S) such that clTop(S) (U ) is compact in Top(S)
F F
and x ∈ U . Observe that clTop(S) (U ) = atS (clS ( U )). Therefore, clS ( U ) is compact
F F F
in S (by lemma 3.1.4) and x v U = intS ( U ). By MTS2 we get x  U , since
F F
x ∈ at(S). Corollary 3.1.6 yields: x  clS ( U ). So clS ( U ) ∈ F (x). Because Σ is a
F
nice class of enriched WMTS and S ∈ Σ, it follows that clS ( U ) ∈ Λ. Therefore, F (x)
is a limited round filter in S. It is only left to show that F (x) is maximal.
Suppose for reductio that F (x) ⊂ G for some maximal limited round filter G. Then
there is a u ∈ G with u ∈
/ F (x). Because G is a round filter, there is a v with v  u.
Hence, clS (v) v intS (u). Moreover, it is not the case that x  u. So it is not the
case that clS (x) v intS (u). Since x ∈ at(S) it follows that x = clS (x) and, hence,
that x o intS (u). So x o clS (v) as well. Consequently, atS (x) ∩ atS (clS (v)) = ∅. So
{x} ∩ clTop(S) (atS (v)) = ∅. Because Top(S) is a locally compact Hausdorff space, it
is a regular space as well. That is, for every point in Top(S) and every closed set not
containing the point, there are disjoint open sets containing the point and the closed
set, respectively. Therefore, there are disjoint open sets U, V in Top(S) such that x ∈ U
F F
and clTop(S) (atS (v)) ⊆ V . It follows from theorem 3.1.3 that U and V are open
F F F F
in S and x v U and clS (v) v V . Thus, clS (x) = x v U = intS ( U ) and
F F F F F
clS (v) v V = intS ( V ). So x  U and v  V . We can conclude that v o U
F F
and U ∈ F (x) ⊂ G. Remember that v ∈ G. But this is impossible, because U ∈ G
F
and v o U and G is a filter. So F (x) is a maximal limited round filter in S. Because
x  u for all u ∈ F (x), it follows that F (x) ∈ mrfS (x).
It follows that mrfS (x) = {F (x)}, for the following reasons. Assume that G ∈ mrfS (x).
Then x ∈ G or for all y ∈ G: x  y . If x ∈ G, then x  x (because G is a round filter)
and x v y for all y ∈ G. So for all y ∈ G: x  y. Hence, it holds in any case that x  y
for all y ∈ G. That is, G ⊆ F (x). It follows that G = F (x), because G is maximal.

2. Embedding: mrfS is an embedding of S into Smrf . This follows from the preliminary
observation. The proof is analogous to the corresponding proof in case 1 (finite and
infinite discrete structures).

3. Idempotence: We show that mrfS is a bijection between at(S) and at(Smrf ). In view of
(2.) this implies that mrfS is also a bijection between X and X mrf and, hence, mrfS is
an isomorphism between S and Smrf .
We already know from (1.) that for every x ∈ at(S), there is exactly one F (x) ∈ at(Smrf )
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 91

such that F (x) ∈ mrfS (x). It is left to show that for every F ∈ mrf(S), there is exactly
one x ∈ at(S) such that F ∈ mrfS (x). So let F ∈ mrf(S). Then Fat := {atS (x) : x ∈ F }
is a maximal limited round filter in Sat , see lemma 4.3.10. In order to show that there is
exactly one x ∈ at(S) such that F ∈ mrfS (x), we proceed as follows. First we show that
F determines a corresponding maximal limited round filter F ∗ in the RMTS of regular
closed sets in Top(S). Next we use a theorem of Roeper (lemma 4.3.12), which shows
that F ∗ converges to a single point in Top(S). Finally, we prove that F converges to a
single atom of S for these reasons.
(a) We show that F ∗ := {U ∈ RC(Top(S)) : V ⊆ U for some V ∈ Fat } is a maximal
limited round filter in SRC (Top(S)) := (RC(Top(S))\{∅}, ⊆, RC , ΛRC ) with U RC
V iff clTop(S) (U ) ⊆ intTop(S) (V ) and U ∈ ΛRC iff U is compact in Top(S), for all
non-empty U, V ⊆ RC(Top(S)).
We show that F ∗ is a filter in SRC (Top(S)). Let U, V ∈ F ∗ . The meet of U, V in
SRC (Top(S)) is clTop(S) (intTop(S) (U ) ∩ intTop(S) (V )). So we have to show that this set
is in F ∗ . From the definition of F ∗ we get that there are U 0 , V 0 ∈ Fat with U 0 ⊆ U and
V 0 ⊆ V . Hence, U 0 ∩ V 0 ∈ Fat . Since, Fat is a round filter, there is some W ∈ Fat such
that W at U 0 ∩ V 0 . So W  (U 0 ∩ V 0 ). So clS ( W ) v intS ( (U 0 ∩ V 0 )) and,
F F F F

in view of corollary .3.1.5, atS (clS ( W )) v atS (intS ( (U 0 ∩ V 0 ))). Hence, by theorem
F F

3.1.3, clTop(S) (W ) ⊆ intTop(S) (U 0 ∩ V 0 ). Hence, clTop(S) (W ) ⊆ intTop(S) (U 0 ∩ V 0 ) =


intTop(S) (U 0 ) ∩ intTop(S) (V 0 ) ⊆ intTop(S) (U ) ∩ intTop(S) (V ). Using the monotonicity of
closure operators on topological spaces, we get: clTop(S) (W ) ⊆ clTop(S) (intTop(S) (U ) ∩
intTop(S) (V )). So W ⊆ clTop(S) (intTop(S) (U ) ∩ intTop(S) (V )) ∈ RC(Top(S)) and W ∈
Fat . Thus, by definition of F ∗ : clTop(S) (intTop(S) (U ) ∩ intTop(S) (V )) ∈ F ∗ . So far so
good. Now let U ∈ F ∗ and U ⊆ V with V ∈ RC(Top(S)). Then there is a W ∈ Fat
such that W ⊆ U ⊆ V . Hence, V ∈ F ∗ as well. Thus, F ∗ is a filter in SRC (Top(S)).
F ∗ is a round filter in SRC (Top(S)) for the following reason. Let U ∈ F ∗ . Then there
is a V ∈ Fat with V ⊆ U . Because Fat is a round filter in Sat , there is some W ∈ Fat
with W at V . So clTop(S) (W ) ⊆ intTop(S) (V ), which is justified by the same argument
we have just used in the course of showing that F ∗ is a filter. Because W ∈ Fat and
Fat is a round filter, there is some W 0 ∈ Fat such that W 0 at W . Employing the
same argument again, it follows that clTop(S) (W 0 ) ⊆ intTop(S) (W ). The monotonicity
and idempotence of the closure operator yields: clTop(S) (W 0 ) ⊆ clTop(S) (intTop(S) (W )).
Summing up, we have: W 0 ⊆ clTop(S) (intTop(S) (W )) ⊆ clTop(S) (W ) ⊆ intTop(S) (V ).
Because of W 0 ∈ Fat we can conclude that clTop(S) (intTop(S) (W )) ∈ F ∗ . Moreover,
observe that intTop(S) (V ) ⊆ intTop(S) (U ), which is justified by V ⊆ U and monotonicity
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 92

of intTop(S) . Thus, clTop(S) (clTop(S) (intTop(S) (W ))) ⊆ intTop(S) (U ), again using the clo-
sure operator’s idempotence. So we also have: clTop(S) (intTop(S) (W )) RC U . Hence,
F ∗ is a round filter in SRC (Top(S)).
F ∗ is limited as well. This can be shown as follows. Since Fat is limited, there is some U ∈
F
Fat with U ∈ Λat . That is, U ∈ Λ. It follows that there is some y such that y is compact
F
in S and U v y, because Σ is a nice class of enriched WMTS and S ∈ Σ. Using lemma
3.1.4 and corollary 3.1.5, we get that atS (y) is compact in Top(S) and U ⊆ atS (y). Since
Fat is a round filter in Sat , we can conclude that atS (y) ∈ Fat and that there is some
V ∈ Fat with V at U . It follows that clTop(S) (V ) ⊆ intTop(S) (U ) and, thus, also
that V ⊆ clTop(S) (V ) = clTop(S) (clTop(S) (V )) ⊆ clTop(S) (intTop(S) (U )). Remember that
V ∈ Fat . Therefore, clTop(S) (intTop(S) (U )) ∈ F ∗ . Moreover, monotonicity of clTop(S)
yields: clTop(S) (intTop(S) (U )) ⊆ clTop(S) (atS (y)). Because Top(S) is a Hausdorff space
and atS (y) is compact in Top(S), it follows by 3.1.3 that atS (y) is closed in Top(S).
So atS (y) = clTop(S) (atS (y)). Hence, we get clTop(S) (intTop(S) (U )) ⊆ atS (y). In a
Hausdorff space, every closed subset of a compact set is compact as well (lemma 3.1.2).
Therefore, clTop(S) (intTop(S) (U )) is compact in Top(S) and an element of F ∗ . So F ∗
contains a compact element and, hence, is a limited filter in SRC (Top(S)).
Now it is only left to show that F ∗ is also maximal. Assume for reductio that there is
a limited round filter G∗ in SRC (Top(S)) such that F ∗ ⊂ G∗ . Let G+ := {U : U ∗ ⊆
U for some U ∗ ∈ G∗ }. Now we proceed as follows. We first derive that Fat ∪ G+ is a
limited round filter in Sat . Then we show that this implies that Fat is not maximal,
which contradicts our starting assumption.
Let U, V ∈ Fat ∪ G+ . The case that U, V ∈ Fat is trivial. In the case that U, V ∈
G+ , there are U ∗ , V ∗ ∈ G∗ with U ∗ ⊆ U and V ∗ ⊆ V . Because G∗ is a filter in
SRC (Top(S)), it follows that U ∗ , V ∗ are regular closed in Top(S) and their regular
open meet is in G∗ as well, i.e. clTop(S) (intTop(S) (U ∗ ) ∩ intTop(S) (V ∗ )) ∈ G∗ . Further-
more observe that: clTop(S) (intTop(S) (U ∗ ) ∩ intTop(S) (V ∗ )) ⊆ clTop(S) (intTop(S) (U ∗ )) ∩
clTop(S) (intTop(S) (V ∗ )) = U ∗ ∩ V ∗ ⊆ U ∩ V . Hence, U ∩ V ∈ G+ . In the case that
U ∈ G+ and V ∈ Fat , there is some W ∈ Fat such that W at V . Then it follows by
the argument that we have already used several times that clTop(S) (intTop(S) (W )) ∈ F ∗ .
Since F ∗ ⊂ G∗ , we know that clTop(S) (intTop(S) (W )) ∈ G∗ . Now we use that U ∈ G+ .
This implies that U ∗ ⊆ U for some U ∗ ∈ G∗ . Remember that G∗ is a filter in
SRC (Top(S)). So the regular open meet of U ∗ and clTop(S) (intTop(S) (W )) is in G∗ .
That is: clTop(S) (intTop(S) (U ∗ ) ∩ intTop(S) (clTop(S) (intTop(S) (W )))) ∈ G∗ . Note that
for topological reasons: clTop(S) (intTop(S) (U ∗ ) ∩ intTop(S) (clTop(S) (intTop(S) (W )))) ⊆
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 93

U ∗ ∩ clTop(S) (intTop(S) (W )). Moreover, U ∗ ∩ clTop(S) (intTop(S) (W )) ⊆ U ∩ V . There-


fore, U ∩ V has a subset that is in G∗ . Hence, U ∩ V ∈ G+ . The case that U ∈ Fat and
V ∈ G+ is analogous. So in every case, U, V ∈ Fat ∪ G+ implies U ∩ V ∈ Fat ∪ G+ .
Now let U ∈ Fat ∪ G+ and U ⊆ V . If U ∈ Fat , then V ∈ Fat as well (because Fat is a
filter in Sat ). If U ∈ G+ , then there is some U ∗ ∈ G∗ with U ∗ ⊆ U . Hence, U ∗ ⊆ V
and, therefore, V ∈ G+ as well. So in any case, V ∈ Fat ∪ G+ . Thus, we finally can
conclude that Fat ∪ G+ is a filter in Sat . Moreover, Fat ∪ G+ is a limited filter, because
Fat is a limited filter and Fat ⊆ Fat ∪ G+ .
It is only left to prove that Fat ∪ G+ is a round filter in Sat . Let U ∈ Fat ∪ G+ . If
U ∈ Fat , then there is some V ∈ Fat with V at U , and we are home. If U ∈ G+ , then
there is some U ∗ ∈ G∗ with U ∗ ⊆ U . Because G∗ is a round filter in SRC (Top(S)),
there is some V ∗ ∈ G∗ with V ∗ RC U ∗ . It follows immediately that V ∗ at U ∗ and,
thus, V ∗ at U . Clearly, V ∗ ∈ Fat ∪ G+ , because V ∗ ∈ G∗ . So in any case there is a
V ∈ Fat ∪ G+ such that V at U . Thus, Fat ∪ G+ is a round filter in Sat .
So we finally have established that Fat ∪ G+ is a limited round filter in Sat . Now we
show that this entails that Fat is not maximal. As assumed for reductio, F ∗ ⊂ G∗ .
So there is some U ∈ G∗ such that U ∈
/ F ∗ . Hence, U ∈ G+ as well. Suppose,
U ∈ Fat . Observe that U ∈ RC(Top(S)), because U ∈ G∗ . So it follows that U ∈ F ∗ –
/ Fat . Therefore, Fat ⊂ Fat ∪ G+ .
contradiction! Therefore, it must be the case that U ∈
But because Fat ∪ G+ is a limited round filter in Sat , this means that Fat is not a
maximal limited round filter in Sat . This contradicts our starting assumption. This
completes the reductio and we can conclude that F ∗ is a maximal limited round filter
in SRC (Top(S)).
(b) Because Top(S) is a locally compact Hausdorff space and F ∗ is a maximal limited
round filter in SRC (Top(S)), we can employ lemma 4.3.12. Consequently, there is
exactly one x ∈ at(S) such that F ∗ = {U ∈ RC(Top(S)) : x ∈ intTop(S) (U )}. So for
every V ∗ ∈ F ∗ , x ∈ intTop(S) (V ∗ ). This allows us to complete the proof in the following
step.
(c) Let U ∈ Fat . Then there is some V ∗ ∈ F ∗ such that V ∗ at U (again by the argument
from Fat being a round filter, which we have used several times above). Using the result of
step (b), we get that x ∈ intTop(S) (V ∗ ). As we have noted at the beginning of this proof,
every atom in S is closed in S. So in particular, x is is closed in S. Hence, {x} is closed
in Top(S), see theorem 3.1.3. That is, clTop(S) ({x}) = {x}. Therefore, clTop(S) ({x}) =
{x} ⊆ intTop(S) (V ∗ ). Moreover, intTop(S) (V ∗ ) ⊆ clTop(S) (V ∗ ) ⊆ intTop(S) (U ). Hence,
clTop(S) ({x}) ⊆ intTop(S) (U ). Consequently, for all U ∈ Fat , {x} at U . Therefore, for
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 94

all u ∈ F , x  u. So F ∈ mrfS (x). Finally, we can conclude that for all F ∈ mrf(S),
there is exactly one x ∈ at(S) such that F ∈ mrfS (x).

4. Since S ∼= Smrf and S is an atomistic continuous WMTS with information about lim-
itedness, Smrf is an atomistic continuous WMTS with information about limitedness as
well.

Case 3 : S = (X, v, , Λ) is an atomless Σ-structure such that for every x ∈ X, there is some
y ∈ Λ with y  x.

1. Preliminary observations:
(a) First we prove that for every x ∈ X, mrfS (x) 6= ∅. Let x ∈ X. Then there is some
y ∈ Λ with y  x. There is also some z ∈ Λ with z  y. Let F = {u : z  u}. Then
F is a limited round filter in S. Let u, v ∈ F . Then z  u and z  v. It follows that
z  u u v (by MTS3). Thus, u u v ∈ F . Now let u ∈ F and u v v. Then z  u and,
hence, z  v (by MTS2). Therefore, v ∈ F . Thus, F is indeed a filter. F is limited,
because y ∈ F and y ∈ Λ. F is a round filter, because for every u ∈ F , z  u. MTS5
yields that there is some v with z  v  u. Then v ∈ F . So there is some v ∈ F with
v  u. Consequently, F is a limited round filter in S and x ∈ F , because z  x. Using
the round filter lemma 4.3.11, we can conclude that there is a maximal limited round
filter F 0 with x ∈ F ⊆ F 0 and, thus, F 0 ∈ mrfS (x). So for every x ∈ X, there is an F 0
with F 0 ∈ mrfS (x).
(b) For all x ∈ X, if F ∈ mrfS (x), then x ∈ F . Let x ∈ X and F ∈ mrfS (x). Then either
x ∈ F or there is some u such that u v x and for all v ∈ F , u  u. In the first case,
there is nothing to show. The second case cannot arise for the following reason. Assume
for reductio that there is some u such that u v x and for all v ∈ F , u  v. Then there
is a y with y @ u and a z with z  y (because we are considering the atomless case
only). It follows that G = {v : z  v} is a limited round filter in S (see preliminary
observation (a)). But F ⊂ G, which is in contradiction to F being maximal.

2. Embedding: mrfS is an embedding of S into Smrf . It follows immediately from prelimi-


nary observation (a) that for every x ∈ X, mrfS (x) ∈ P(mrf(S))\{∅} = Xmrf . So mrfS
is indeed a function from X to Xmrf . We have to show that it is structure preserving
and injective. Let x, y ∈ X.
(a) We prove: x v y iff mrfS (x) ⊆ mrfS (y). (⇒) Let x v y. Let F ∈ mrfS (x). Then
preliminary observation (b) yields: x ∈ F . Since F is a filter, y ∈ F as well. So
F ∈ mrfS (y). (⇐) Conversely, let mrfS (x) ⊆ mrfS (y). Let u ◦ x. Then there is a v such
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 95

that v v u and v v x. Preliminary observation (a) yields that there is some F with
F ∈ mrfS (v). Hence, by preliminary observation (b): v ∈ F . Thus, u ∈ F and x ∈ F
as well (since F is a filter). So F ∈ mrfS (x). Then by assumption, F ∈ mrfS (y). Again
using preliminary observation (b) we conclude that y ∈ F . Because u, y ∈ F , it follows
that u ◦ y. Therefore, SSP yields that x v y.
(b) That x  y iff mrfS (x) mrf mrfS (y) can be proven analogously to statement (2) in
lemma 4.3.7.
(c) x ∈ Λ iff mrfS (x) ∈ Λmrf . (⇒) Trivial. (⇐) Let mrfS (x) ∈ Λmrf . Then there is a
y ∈ Λ with mrfS (x) ⊆ mrfS (y). It follows from (a) that x v y. Therefore, x ∈ Λ (by
LIM1).
(d) That mrfS is injective follows from (a). Let x, y ∈ X and assume that mrfS (x) =
mrfS (y). Then (a) yields: x v y and y v x. Hence, x = y.

3. Smrf is an atomistic WMTS with information about limitedness. It is clear that (Xmrf , ⊆)
is an atomistic CMS (lemma 3.1.4). Verifying that Smrf satisfies MTS1 and MTS2 is
very easy. That MTS3, MTS4, MTS5, LIM1 and LIM2 are satisfied as well, can be
shown analogously to the proof of the embedding theorem for RMTS (theorem 4.3.1).
Therefore, for all S in the class of relevant test cases of Σ, (1)mrfS is an embedding of
S into Smrf , (2) if S is atomistic, then mrfS is an isomorphism between S and Smrf , (3)
Smrf is an atomistic WMTS with information about limitedness. So by definition 3.2.6,
(mrf, corrmrf , [.]mrf ) is an adequate point representation for the class of relevant test cases
of Σ.

Given the powerful theorem 4.3.4, we can easily show that the existence of a nice class of
enriched WMTS implies that the round filter representation is adequate. This theorem shows
why the nice classes of enriched WMTS deserve their name.
Corollary 4.3.5. If there is a nice class of enriched WMTS, then (mrf, corrmrf , [.]mrf ) is a
generally adequate point representation.
Proof. Suppose that there is a nice class Σ∗ of enriched WMTS. Then, by the preceding theo-
rem, (mrf, corrmrf , [.]mrf ) is an adequate point representation for the class of relevant test cases
of Σ∗ . Let Σ∗∗ be the class of relevant test cases of Σ∗ and let Σ be the class WMTS with
information about limitedness. Then Σ∗∗ ⊆ Σ, (mrf, corrmrf , [.]mrf ) is an adequate Σ-point rep-
resentation for Σ∗∗ and the following conditions are satisfied: (1) Σ∗∗ contains finite discrete
Σ-structures. (2) Σ∗∗ contains infinite discrete Σ-structures. (3) Σ∗∗ contains atomistic contin-
uous Σ-structures. (4) Σ∗∗ contains atomless Σ-structures. So using definition 3.2.8 and 3.2.9,
we can conclude that (mrf, corrmrf , [.]mrf ) is a generally adequate point representation.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 96

All that is left to do now is to show that there is indeed a nice class of enriched WMTS. This
is the content of the next theorem.

Theorem 4.3.5. There is a nice class of enriched WMTS.

Proof. (1) Take any finite CMS (X, v) and let S1 = (X, v, v, Λ1 ) with Λ1 = X. Observe that
S1 satisfies the MTS1–MTS5 and trivially satisfies LIM1 and LIM2. Thus, it is a WMTS.14
Every finite CMS is perforce atomistic. So S1 is an atomistic WMTS. Because interior part-
hood in S1 is the same as parthood, every atom in S1 is an interior part of itself and has no
other interior parts. Thus, for every x ∈ at(S1 ), x = intS1 (x) and, hence, x is open in S1 . So
theorem 3.1.3 yields that for every x ∈ at(S1 ), {x} is open in Top(S1 ). Note that the points
of Top(S1 ) are just the atoms in S1 . Hence, all singletons of points in Top(S1 ) are open.
Therefore, Top(S1 ) is a discrete space. Thus, S1 is a finite discrete WMTS with information
about limitedness. Finally, observe that for every x ∈ X, x ∈ Λ1 iff there is some y such that
y is compact in S1 and x v y. This is justified by lemma 3.1.4 and the fact that Top(S1 ) is a
discrete space and compactness coincides with finiteness in discrete spaces. Therefore, every
x ∈ X is compact in S1 . Thus for all x ∈ X, x ∈ X = Λ1 iff there is some y such that y is
compact in S and x v y.
(2) Take any infinite atomistic CMS (X, v). Then we can specify a discrete mereotopologi-
cal structure as follows. Let S2 = (X, v, v, Λ2 ), where Λ2 = {x ∈ X : at(X,v) (x) is finite}.
As in the last case, Top(S2 ) is a discrete space, because interior parthood coincides with
parthood. Note that if x, y ∈ Λ2 , then x and y have both finitely many atoms. So x t y has
also only finitely many atoms and, hence, x t y ∈ Λ2 . Moreover, if x ∈ Λ2 and y v x, then
y can also have only finitely many atoms. Therefore, y ∈ Λ2 as well. So S2 satisfies LIM1
and LIM2. Furthermore, note that S2 satisfies MTS1–MTS5. So S2 is an infinite discrete
WMTS with information about limitedness. Finally, observe that for every x ∈ X, x ∈ Λ2 iff
there is some y such that y is compact in S2 and x v y. This can be shown as follows. It
follows again by lemma 3.1.4, by the fact that Top(S1 ) is a discrete space and by the fact that
compactness coincides with finiteness in discrete spaces that x is compact in S2 iff at(X,v) (x)
is finite. Therefore, x ∈ Λ2 iff there is some y such that y is compact and x v y.
(3) Now take some topological n-manifold T = (X, O) with n ∈ N. Then a corresponding
mereotopological structure can be specified as follows. Let S3 = (P(X)\{∅}, ⊆, 3 , Λ3 ),
where for all non-empty subsets U, V of X, U 3 V iff clT (U ) ⊆ intT (V ), and U ∈ Λ3
iff for some W , W is compact in T and U ⊆ W . Note that S3 satisfies MTS1–MTS5 and
LIM1, LIM2. So it is an atomistic WMTS with information about limitedness. In order to
show that it is atomistic continuous, we have to prove that for all non-empty subsets U of
14
Actually, S1 is not only a WMTS, but also an RMTS.
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 97

X, U is open in S3 iff U is open in T . Let U ∈ P(X)\{∅}. Note that U is open in S3 iff


S S
U = {V ∈ P(X)\{∅} : V 3 U }, and U is open in T iff U = {V : V ∈ O & V ⊆ U }.
S S
So it is sufficient to show that {V ∈ P(X)\{∅} : V 3 U } = {V : V ∈ O & V ⊆ U }.
S
(⇒) Let x ∈ {V ∈ P(X)\{∅} : V 3 U }. Then for some non-empty V ⊆ X, x ∈ V and
clT (V ) ⊆ intT (U ). Then x ∈ intT (U ) ⊆ U . So there is a W ∈ O such that W ⊆ U and x ∈ W .
S S
Thus, x ∈ {V : V ∈ O & V ⊆ U }. (⇐) Conversely, let x ∈ {V : V ∈ O & V ⊆ U }. Then
for some non-empty V ∈ O, x ∈ V and V ⊆ U . So, by lemma 4.3.8, there is a regular closed set
W in T with W ⊆ V and x ∈ W . Thus, x ∈ clT (W ) = W ⊆ V = intT (V ) ⊆ intT (U ). So there
S
is a W ∈ P(X)\{∅} with W 3 U and x ∈ W . Hence, x ∈ {V ∈ P(X)\{∅} : V 3 U }.
This completes the subproof. Then it follows by theorem 3.1.3 that Oat (S3 ) = {atS3 (U ) :
U open in T }. This means that Top(S3 ) = (at(S3 ), Oat (S3 )) is just a copy of T , where every
point x of T is replaced by the singleton {x}, which is the corresponding point in Top(S3 ).
Therefore, Top(S3 ) is homeomorphic to T . Since T is a topological n-manifold, Top(S3 ) is a
topological n-manifold as well. Thus, S3 is an atomistic continuous WMTS with information
about limitedness. Moreover, for all U, V in the domain of S3 , U 3 V iff clT (U ) ⊆ intT (V )
iff clS3 (U ) ⊆ intS3 (V ). This is due to the above established result that U is open in S3 iff U
is open in T . Finally, note that for all U ∈ P(X)\{∅}, U ∈ Λ3 iff there is a V such that V
is compact in S3 and U ⊆ V . This follows immediately from the characterisation of Λ3 and
lemma 3.1.4.
(4) Moreover, let S4 = (RO(T )\{∅}, ⊆, 4 , Λ4 ), where for all non-empty regular open sets
U, V in T , U 4 V iff clT (U ) ⊆ intT (V ), and U ∈ Λ4 iff for some W , W is compact in T
and U ⊆ W . Since T is a topological manifold, it is also a locally compact Hausdorff space.
It follows that S4 is an RMTS (by lemma 4.3.4). Moreover, S4 is atomless for the following
reason. For every Hausdorff space S in which no one-point set is open, (RO(S)\{∅}) is
atomless, see Roeper’s lemma 6.1 (1997, p. 283). Since T is a topological manifold, it is locally
Euclidean and, thus, no one-point set is open in T . Furthermore, it follows from lemma 4.3.8
that for every U ∈ RO(T )\{∅}, there is a regular closed and compact set V in T with V ⊆ U .
Then, intT (V ) is in RO(T )\{∅} as well as in Λ4 , and clT (intT (V )) = V ⊆ U = intT (U ).
Therefore, for every U ∈ RO(T )\{∅}, there is some W ∈ Λ4 such that W 4 U . Consequently,
S4 is an atomless WMTS with information about limitedness such that for every U in the
domain of S4 , there is some W ∈ Λ4 with W 4 V .
Let Σ = {S1 , S2 , S3 , S4 } and let Σ∗ = {S0 : S0 ∼
= S for some S ∈ Σ}. Then Σ∗ contains
(1) finite discrete WMTS with information about limitedness, (2) infinite discrete WMTS with
information about limitedness, (3) atomistic continuous WMTS S with information about
limitedness such that for all x, y in the domain of S, x  y iff clS (x) v intS (y), and (4)
atomless WMTS S with information about limitedness such that for every x in the domain
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 98

of S, there is some y ∈ Λ with y  x. Furthermore, (5) for all S = (X, v, , Λ) ∈ Σ∗ , if S


is atomistic, then for every x ∈ X, x ∈ Λ iff there is some y such that y is compact in S and
x v y. So it follows that Σ∗ is a nice class of enriched WMTS.

At last, the previous results (coroallary 4.3.5 and theorem 4.3.5) allow us to derive that the
round filter representation is generally adequate.

Corollary 4.3.6. (mrf, corrmrf , [.]mrf ) is a generally adequate point representation.

This corollary finally shows that the problem of finding a generally adequate method of repre-
senting points can be solved. Of course, the result does not rule out that there are other gen-
erally adequate representation methods than the round filter representation. For a discussion
of alternative characterisations of points that are equivalent to the round filter representation,
see Roeper (1997, esp. section 3). What is important for the purpose of this thesis is that
there is at least one generally adequate point representation. And this is hereby established.

4.3.3 The Open Problem of Axiomatisation


But there is still an open problem. In the proof of theorem 4.3.5, we constructed a nice class of
enriched WMTS, establishing the general adequacy of the round filter representation. We con-
structed this class by selecting specific structures with different formal properties and forming
the set containing just these and isomorphic structures. However, it is certainly preferable to
single out a nice class of enriched WMTS by adding axioms to those that determine the class
of WMTS with information about limitedness – rather than by picking single structures. So
the following question arises. Is it possible to formulate axioms for a unified mereotopological
system which can be proven to determine a nice class of enriched WMTS? I have to admit
that I do not know whether this is possible.
Nonetheless, this problem does not affect the general adequacy of the round filter represen-
tation. The problem can be stated without any reference to the round filter representation (or
any other point representation). It can be viewed as concerning exclusively the construction of
mereotopological axioms to single out a specific class of structures. So it is rather a technical
problem in mereotopology than a problem that directly concerns the method of round filter
representation itself.
To solve the problem one has to reduce the concept of limitedness to parthood and interior
parthood or at least establish sufficiently strong bridge principles for these concepts. Different
approaches to this task have been proposed in the literature. (1) Roeper himself developed
a method of reducing limitedness to v and  for a specific class of RBT (cf. 1997, p. 295).
But since this reduction works only for atomless structures, his method is not suitable for a
CHAPTER 4. ANALYSING POINT REPRESENTATIONS 99

more general setting that is neutral concerning atomism. (2) Bostock (2010) also proposes a
reduction of limitedness to purely mereotopological notions. His basic idea is to employ the
Bolzano-Weierstraß property of bounded sets. The resulting reduction is quite complicated
and I have not been able to determine, whether it works for all relevant kinds of structures. (3)
Mormann (1998, 2009) tries to avoid the notion of limitedness by using an alternative interior
parthood relation. The intended sense of ‘x is an interior part of y’ according to Mormann’s
understanding is: ‘the closure of x is compact and the closure of x is part of the interior of y’. So
Mormann’s notion of interior parthood is intended to contain information about compactness
(and hence limitedness). Mormann gives axioms for this notion of interior parthood and tries
to represent points using only his alternative mereotopological notions. Mormann’s approach
has been criticised by Vakarelov et al. (2002). They use a counterexample to show that
Mormann’s main representation theorem (1998) does not hold.15 Moreover, they propose an
additional axiom that allows to prove a desired representation theorem. However, the axioms
are so strong that they again rule out atomistic continuous structures. (4) Forrest (2010) has
developed an approach that is very similar to Mormann’s. Forrest uses only a single primitive
of interior parthood and defines limitedness in term of it. A region is limited (bounded) iff it is
interior part of some other region (cf. 2010, p. 234). An advantage of Forrest’s method is that
it does not require the axiom of choice – in contrast to all classical representation methods.
But again, Forrest’s method rules out atomistic continuous structures and thus is inadequate
for our purpose.
For the reasons mentioned above, I believe that these proposed methods of dealing with
limitedness cannot succeed in establishing a nice class of enriched WMTS (possibly with the
exception of Bostock’s approach). We need a definition of or a strong bridge principle for
limitedness that works both in atomless and atomistic cases and has the consequence that, in
atomistic cases, an element is limited iff it is part of a compact element. Note that compactness
is defined for atomistic WMTS, but not for atomless WMTS. So the task is essentially to find
a generalisation of the notion of compactness that applies not only to atomistic but also to
atomless structures.
I will not go into the discussion whether such a generalisation is possible, because it would
lead too far afield. The most important point I want to make is that finding a nice class of
enriched WMTS that is characterised by structural axioms is not a problem for the round
filter representation per se, but rather a mereotopological problem, whose solution could be
deemed to establish further support for the round filter representation.

15
As Mormann told me in private communication, he is not convinced by the objection due to Vakarelov et
al (2002). Mormann thinks that he has found the missing lemma that establishes his proof.
CHAPTER 5

METAPHYSICAL QUESTIONS

5.1 The Question of Realism About Abstract Entities


The fact that points have been identified with sets of basic entities in chapter 4 raises the
following metaphysical question. Does abstracting spacetime points from mereotopologically
structured basic entities by means of the maximal round filter method (or a similar method)
imply that spacetime points are abstract objects? If it does, then it presupposes realism
about abstract objects. Moreover, it gives rise to the following objection. If spacetime points
are abstracted from mereotopologically structured basic entities, then spacetime points are
abstract entities, because they are identified with classes of basic entities, and classes are
abstract objects. That spacetime points are abstract is problematic for the following reason.
An object is abstract just in case it is not a spatiotemporal1 objects. But this means that
spacetime points are not spatiotemporal objects – which sounds quite absurd.
However, the method of identifying points with classes of basic entities can be defended
against this objection. There are two ways to respond. First, one could deny that the ab-
stractness of spacetime points is really a problem. One could argue that the seemingly coun-
terintuitive result that spacetime points are abstract is after all acceptable, because spacetime
points are not ontologically fundamental but rather the results of our abstractions from the
spatiotemporal relations between concrete entities. However, this defence is only convincing
if one anyway shares the ontological view that spacetime points are not ontologically funda-
mental entities.
Second, one can deny that identifying spacetime points with classes of basic entities implies
that spacetime points are abstract. One can argue that points can be identified with classes
as many rather than with classes as one 2 or with pluralities rather than sets. It is certainly
1
Being a spatiotemporal object either means (1) being located in spacetime, or (2) having spatiotemporal
relations (e.g. earlier, later, contemporary) to other objects.
2
The disctinction between classes of one and classes as many goes back to Russell (1903, §70).

100
CHAPTER 5. METAPHYSICAL QUESTIONS 101

convenient to use set theory as formal framework (especially because it is so well known and
things are easy to proof in it), but in order to define the notion of spacetime point it is not
necessary, far weaker systems will do as well – for example a logic of pluralities or a theory
of classes as many. If points are identified with pluralities or classes as many, then it turns
out that spacetime points are concrete rather than abstract (given that the basic entities are
concrete themselves).3
In this section we will examine the second defence in more detail. Of course, talking about
‘classes as many’ and ‘pluralities’ is quite unclear. So we have to be more precise about the
formal framework we use to define the notion of point so that points turn out to be concrete.
I choose Linnebo’s (2012) plural first-order logic PFO+ as framework. The intuitive idea that
points are pluralities can be captured by using plural terms to refer to spacetime points.

Plural Reference and its Logic

The key idea underlying first-order plural logic is that apart from singular terms, which refer to
exactly one thing, there are also plural terms, which refer to several things at once4 . In the fol-
lowing we specify a language LP F O+ of first-order plural logic as developed by Linnebo (2012).
The vocabulary of LP F O+ comprises:

– countably many singular variables u, v, w, x, y, z; ui , vi , wi , xi , yi , zi (i ∈ N);

– countably many plural variables uu, vv, ww, xx, yy, zz; uui , vvi , wwi , xxi , yyi , zzi (i ∈ N);

– countably many singular constants ci and countably many plural constants cci (i ∈ N);

– countably many n-ary singular predicates P1n , P2n , P3n , . . . (n ∈ N);

– countably many n-ary singular functors f1n , f2n , f3n , . . . (n ∈ N);

– countably many n-ary plural predicates Pn1 , Pn2 , Pn2 , . . . (n ∈ N);

– two binary logical predicates = (identity) and ε (being one of);

– logical connectives (¬, ∧, ∨, →, ↔) and quantifiers (∀, ∃);

– auxiliary symbols such as parantheses.

The formulas of LP F O+ are given by the following rules.

– If Pin is an n-ary singular predicate and t1 , . . . , tn are singular terms, then Pin t1 . . . tn is
a formula.
3
This argument has been suggested to me by Peter Simons in oral communication.
4
The border case of a plural term referring to only one thing is admitted.
CHAPTER 5. METAPHYSICAL QUESTIONS 102

– If Pni is an n-ary plural predicate and tt1 , . . . , ttn are plural terms, then Pni tt1 . . . ttn is
a formula.

– If t is a singular term and tt is a plural term, then tεtt is a formula.

– If t1 , t2 are singular terms, then t1 = t2 is a formula.

– For logically complex formulas the usual rules apply.

The system P F O+ has the usual natural deduction rules for the singular part of the language
and corresponding natural deduction rules for the plural part of the language. Moreover,
P F O+ has the follwing axioms:

– tautologies and the usual axioms for identity (=);

– plural comprehension: ∃uφ[u] → ∃xx∀u(uεxx ↔ φ[u]);

– non-emptyness: ∀xx∃u uεxx;

– coextensional pluralities are indiscernible: ∀xx∀yy(∀u(uεxx ↔ uεyy) → (φ[xx] ↔


φ[yy]));

We also add definitions for two binary plural predicates ≺ and  representing ‘are among’ in
a weak and strict sense.

(D) xx  yy :↔ ∀u(uεxx → uεyy)

(D≺) xx ≺ yy :↔ xx  yy ∧ ¬yy  xx

Points as Pluralities

In the next step we formulate a theory within the framework of the plural first-order logic
P F O+ . First reserve a list of singular predicate of LP F O+ for the theory, namely:

– a binary singular predicate for parthood (metalinguistically denoted ‘v’);

– a binary singular predicate for interior parthood (metalinguistically denoted ‘’);

– a unary singular predicate for limitedness (metalinguistically denoted ‘L’);

– a unary singular predicate for being a basic entity (metalinguistically denoted ‘B’).

Futhermore, we reserve singular predicates and functors for the usual mereological predicates
and functors (e.g. @, ◦, u, t) and define them in the familiar way. Using these new resources,
we can formulate the axioms of a minimal mereotopology for basic entities:
CHAPTER 5. METAPHYSICAL QUESTIONS 103

(CM1) ∀x∀y(Bx ∧ By → (x v y ∧ y v x → x = y))

(CM2) ∀x∀y∀z(Bx ∧ By ∧ Bz → (x v y ∧ y v z → x v z))

(CM3) ∀x∀y(Bx ∧ By → (x @ y → ∃z(z @ y ∧ ¬z ◦ x)))

(CM4) ∃x(Bx ∧ φ[x]) → ∃x(Bx ∧ ∀y(Bx → (y ◦ x ↔ ∃z(Bx ∧ φ[z] ∧ y ◦ z))))

(MT1) ∀x∀y(Bx ∧ By → (x  y → x v y))

(MT2) ∀x∀y∀u∀v(Bx ∧ By ∧ Bu ∧ Bv → (x v u ∧ u  v ∧ v  y → x  y))

(MT3) ∀x∀y∀z(Bx ∧ By ∧ Bz → (x  y ∧ x  z → x  y u z))

(MT4) ∃x(Bx ∧ ∀y(By → y  x))

(MT5) ∀x∀y(Bx ∧ By → (x  y → ∃z(x  z ∧ z  y))

(L1) ∀x∀y(Bx ∧ By → (Lx ∧ y v x → Ly))

(L2) ∀x∀y(Bx ∧ By → (Lx ∧ Ly → L(x t y)))


This axiom system characterises exactly the WMTS with information about limitedness.
CM1–CM4 are the axioms for classical mereology (cf. Simons, 1987, p. 37), MT1–MT5 are
mereotopological axioms characteristic for WMTS (see definition 3.1.21), finally the axioms
L1 and L2 are bridge principles for the notion of limitedness and mereological notions (compare
LIM1 and LIM2).
In order to define the notion of spacetime point within this system we first introduce the
auxiliary plural predicates F (for ‘mereological filter’), R (for ‘round filter’) and L (for ‘limited
plurality’).
(DF) Fxx :↔ ∀u(uεxx → Bu) ∧ ∀u∀v(uεxx ∧ vεxx → ∃w(w v u ∧ w v v ∧ wεxx)) ∧
∀u∀v(uεxx ∧ u v v → vεxx)

(DR) Rxx :↔ Fxx ∧ ∀u(uεxx → ∃v(v  u ∧ vεxx))

(DL) Lxx :↔ ∃u(Lu ∧ uεxx)


Now we introduce and define a predicate P which represents the notion of being a spacetime
point (see definition 4.3.4).
(DP) Pxx :↔ Lxx ∧ Rxx ∧ ¬∃yy(Ryy ∧ xx ≺ yy)
We know from theorem 4.3.4, theorem 4.3.5 and their corollaries that DP is a generally ade-
quate definition of ‘spacetime point’. Moreover, if additional axioms can be found such that
the class of models of the resulting axiomatic theory is a nice class of WMTS, then defintion
DP is an adequate definition of ‘spacetime point’ in the respective axiomatic theory.
CHAPTER 5. METAPHYSICAL QUESTIONS 104

Points as Pluralities are Concrete

Furthermore, we can show that spacetime points as pluralities are concrete if the basic entities
are concrete (which they certainly should be). Let C be a singular predicate representing the
notion of being concrete. Note that the predicate ‘concrete’ is clearly a distributive predicate:
some things are concrete just in case every object that is one of them is concrete. The same
should hold for its formal representative C. But formulating the distributivity thesis for C is
not possible, because we do not have predicates that can take both singular and plural terms
as arguments in the language LP F O+ . So we simply define the notion of being concrete for
plural terms (C) by means of the distributivity thesis.

(DC) ∀xx(Cxx ↔ ∀u(uεxx → Cu))

Moreover, we can assume that the basic entities in question are concrete, because it is not a
good idea to abstract spacetime points from objects that are not spatiotemporal.

(B-C) ∀x(Bx → Cx)

Using DC and B-C as additional axioms, we can easily prove within the system that all
spacetime points are concrete.

(P-C) ∀xx(Pxx → Cxx)

The following slightly abbreviated derivation shows this:


(1) Pxx [assumption for conditional proof]
(2) Pxx :↔ Lxx ∧ Rxx ∧ ¬∃yy(Ryy ∧ xx ≺ yy) [definition DP]
(3) Rxx [consequence of (1) and (2)]
(4) Rxx :↔ Fxx ∧ ∀u(uεxx → ∃v(v  u ∧ vεxx)) [definition DR]
(5) Fxx [consequence of (3) and (4)]
(6) Fxx :↔ ∀u(uεxx → Bu)∧ [definition DF]
∀u∀v(uεxx ∧ vεxx → ∃w(w v u ∧ w v v ∧ wεxx))∧
∀u∀v(uεxx ∧ u v v → vεxx)
(7) ∀u(uεxx → Bu) [consequence of (5) and (6)]
(8) ∀u(Bu → Cu) [axiom B-C]
(9) ∀u(uεxx → Cu) [consequence of (7) and (8)]
(10) ∀xx(Cxx ↔ ∀u(uεxx → Cu)) [definition DC]
(11) Cxx [consequence of (9) and (10)]
(12) Pxx → Cxx [conditional proof: (1)–(11)]
(13) ∀xx(Pxx → Cxx) [universal generalisation: (12)]
CHAPTER 5. METAPHYSICAL QUESTIONS 105

This refutes the objection that spacetime points must be abstract if they are abstracted
from mereotopologically structured basic entities by means of the maximal round filter method.
But note that this does not mean that sets or other abstract objects do not have to be
assumed for philosophical and scientific reasoning about spacetime points. Far from it. First-
order plural logic is appropriate to define spacetime points, but it is by no means sufficient
to express and prove powerful theorems about spacetime points. For example, it is not strong
enough to prove that the above definition DP is an adequate definition of ‘spacetime point’,
nor is it strong enough to allow the formulation of physical theories in which spacetime points
play a role.
In order to formulate rich theories and prove powerful theorems it is necessary to add
some axioms of set theory (or a similarly strong theory of abstract objects) as well as suitable
bridge principles to the framework of P F O+ . For several philosophical and scientific purposes,
abstract objects have to be assumed in addition to concrete basic entities and pluralities of
them. However, the important point is that even if abstract objects such as sets are postulated,
spacetime points are nevertheless concrete, according to DP.

5.2 The Question of Basis


Proposed Bases

The two most important kinds of entities that have been proposed as basis for abstracting
points are (1) processes and events, and (2) spacetime regions. Processes and events were,
for example, used by the early Whitehead (1920) and by Russell (1927). Regions were taken
as basis by the later Whitehead (1929) as well as his followers Clarke (1985), Roeper (1997),
Mormann (1998) and Forrest (2010).
Prima facie, there is also a third kind of entities, which Field (1984) calls ‘aggregates
of matter’. But aggregates of matter in Field’s sense turn out to be processes on a closer
look. Field describes them as “spatio-temporally extended physical objects, spatio-temporal
parts of such objects, and aggregates consisting of spatio-temporal parts of different objects”
(1984, p. 34). If aggregates of matter are spatially as well as temporally extended, then Field
obviously endorses a perdurantistic account of physical objects. Thus, aggregates of matter in
Field’s sense can be counted as processes.
If regions are used as basis, then nothing can go wrong as regards formal issues. If the
structure of regions is atomistic, then the abstracted points correspond exactly to the atomic
regions, and if the structure is atomless, then the abstracted points fill in the missing atoms.
In any case, all points are contained in regions, and all regions are filled up with points. This
CHAPTER 5. METAPHYSICAL QUESTIONS 106

is how it is supposed to be. The only problem of using regions as basic entities is that we are
from the outset commited to the assumption that spacetime regions are fundamental. This
excludes relationism. The problem can be avoided by choosing processes instead of regions as
basic entities. However, this choice of basis is also not free of problems.

The Problem of Empty Points and Regions

Taking processes as basic entities has a counterintuitive consequence. This can be shown as
follows. First, we assume that spacetime points are maximal limited round filters of processes.
Next, we introduce two notions in a natural way. We say that a process occurs at a spacetime
point iff it is a member of the spacetime point; and we say that a spacetime point is empty iff
there is no process that occurs at it. It follows immediately that there are no empty spacetime
points. If we define spacetime regions as classes of spacetime points (with at least one member),
this result can be extended to regions. Let us say that a process occurs in a spacetime region
iff it occurs at some point which is a member of the region. Moreover, a spacetime region is
empty iff there is no process that occurs in it. It follows that there are no empty regions. So
taking processes of basic entities has the consequence that there are no regions of spacetime
where nothing happens – everywhere and at every time some process occurs. But, in fact, we
cannot just rule out by defnition that there are regions of spacetime where nothing happens.
Call this problem ‘the problem of empty points and regions’. In order to solve the problem it
has been suggested that merely possible processes should be counted among the basic entities.
But this is problematic too. For a discussion see Butterfield (1984).
The problem of empty points and regions suggests to choose other basic entities than
processes.

Gravitational Fields as Basis

It suggests itself to take gravitational fields as basic entities, because gravitational fields are
assumed to occur in every part of spacetime5 and this choice of basis ties in with the view
that spacetime is somehow abstracted from the structure of the gravitational field:

“There is no such thing as an empty space, i.e. a space without field. Space-time
does not claim existence on its own, but only as a structural quality of the field[.]”
(Einstein, 1961, pp. 155 – 156)
5
“Furthermore, in general relativity there can be material in the sense of mass-energy in a region where
there is not only no material object of an ordinary sort like a chair, but also no field apart from the metric-
gravitational field.” (Butterfield, 1984, p. 104)
CHAPTER 5. METAPHYSICAL QUESTIONS 107

But do gravitational fields really exist? Why should we think that they are real and not only
useful mathematical fictions?
In order to answer these questions, it is useful to contemplate for which reasons we think
that other fields exist. For example, how do we know that electromagnetic fields exist? A
simple but sufficient answer is that they can be detected empirically with suitable measurement
devices and they can even be experienced (think of the attracting or repelling forces of a magnet
in your hand). Now, it can be argued that gravitational fields can also be detected empirically
with suitable measurement devices (think of the earth’s gravitational field, which is quite well
studied). Moreover, gravitational fields can also be experienced, as we all know from carrying
our bodies in the earth’s gravitational field. In these respects, gravitational fields seem very
much like electromagnetic fields. So there seems to be no reason to deny that gravitational
fields are real.
However, some physicists and philosophers of physics hold the opinion that gravitational
fields are mere fictions or illusions. They think that, in reality, there are no gravitational
forces at all; instead, there is only spacetime curvature. This view is often presented as if this
was obviously the correct opinion about the relationship between gravity and the geometry of
spacetime and that it follows directly from general relativity. (Presentations of that kind can
often be found in popular science texts.)
But actually it is not that easy. There are different views about how general relativity can
be interpreted concerning the relationship between gravity and the geometry of spacetime.
Lehmkuhl (2008) distinguishes the following main kinds of such interpretations.

(I1) The Field Interpretation: Spacetime curvature is a manifestation of the gravita-


tional field.

(I2) The Geometric Interpretation: Gravitation is a manifestation of the spacetime


curvature.

(I3) The Strong Egalitarian Interpretation: Gravitation and spacetime curvature can
be identified.

(I4) The Moderate Egalitarian Interpretation: Gravitation and spacetime curvature


are closely related, but cannot be identified and neither reduces to the other.

So according to field interpretations, spacetime curvature results from the presence of grav-
itation (and hence masses), or as Reichenbach (1957, p. 256) puts it: “It is not the theory
of gravitation that becomes geometry, but it is geometry that becomes an expression of the
gravitational field.” According to geometric interpretations, the presence of gravitation results
from spacetime curvature. According to strong egalitarian interpretations, the presence of
CHAPTER 5. METAPHYSICAL QUESTIONS 108

gravitation is just the same as spacetime curvature. Finally, according to moderate egalitar-
ian interpretations, the presence of gravitation and spacetime curvature are closely related,
but they are not the same, and we cannot say whether one results from the other.
Lehmkuhl (2008) advocates an egalitarian interpretation in combination with supersub-
stantivalism. Bunge (1967) is an advocate of the moderate egalitarian interpretation. His
opinion is that “GR welds gravitation with physical geometry. This does not amount to a
geometrization of gravitation – nor to a gravification of geometry.” (1967, p. 225) Misner,
Thorne & Wheeler (1973) advocated a geometric interpretation combined with Wheeler’s
supersubstantivalism. In contrast, Einstein, Reichenbach and Bergmann advocated a field
interpretation, as Lehmkuhl (2008) shows.
Field interpretations go well with the assumption of gravitational fields as basic entities,
because according to them the properties of gravitational fields account for the geometrical
properties of spacetime regions – and not vice versa. Moreover, taking gravitational fields as
basic entities also goes well with egalitarian interpretations and geometric interpretations if
they are combined with a variant supersubstantivalism (as they often are), which identifies
spacetime with the gravitational field. But because of supersubstantivalism these combinations
of view boil down to the assumption that spacetime regions are basic entities. So this option
is out of question for the present purpose. Taking gravitational fields as basic entities can also
be combined with moderate egalitarian interpretations; this is for example done by Bunge
(1967). Only the geometric interpretation is incompatible with taking gravitational fields as
basis (given the assumption that gravitational fields are not identical to spacetime regions).
We conclude that there are different views about the relationship between gravity and the
geometry of spacetime. Some of them are compatible with assuming gravitational fields as
basic entities. In particular, taking gravitational fields as basis ties in with field interpretations.
Moreover, there are axiomatisations of general relativity which explicitly assume gravitational
fields as basic entities (cf. Bunge, 1967; Covarrubias, 1993). Thus, it is by no means clear that
general relativity shows that gravitational fields are not real.
So if one accepts gravitational fields as concrete entities, then one can use them as basis
for abstracting spacetime points. Because gravitational fields fill all of spacetime the problem
of empty points and regions can be avoided if spacetime points are considered as maximal
limited round filters of gravitational fields.

5.3 Other Questions Concerning Ontological Status


Finally, let us consider some questions about the ontological status of spacetime points, where
‘spacetime point’ is defined along the lines of the maximal round filter method. Which on-
CHAPTER 5. METAPHYSICAL QUESTIONS 109

tological category do spacetime points in this sense belong to? Which identity conditions do
they have? How are they related to concrete entities occuring at them?

The Question of Category

The answer to this question depends on whether spacetime points are defined using set theory
or using a logic of pluralities. In the first case, spacetime points belong to the ontological
category of sets. In particular, they are abstract entities then. In the second case, spacetime
points are pluralities. In this case, a spacetime point is not an individual at all, but rather
several things, just as the books written by Bertrand Russell are not one thing but several
things. Hence, spacetime points are concrete if the basic entities are concrete, as shown in
section 5.1. Since pluralities (analysed in terms of plural reference) are nothing over and above
the things they comprise, spacetime points comprise of things of a certain category, but it is
not clear whether they make up a separate category themselves.

The Question of Identity Conditions

Since spacetime points are maximal limited round filters, which in turn are either sets or
pluralities, we immediately get the following identity condition for spacetime points.

(ID1) Spacetime points are identical iff they have the same members.

To be more precise, we have to distinguish two cases. In the case that spacetime points are
pluralities, a spacetime point xx is identical to a spacetime point yy just in case for all basic
entities u, u is one of xx iff u is one of yy. In the case that spacetime points are sets, a
spactime point F is identical to a spacetime point G just in case for all basic entities u, u ∈ F
iff u ∈ G.
These identity conditions lead to other equivalent, but slightly more intuitive identity
conditions, when the kind of basic entities is taken into account. First we explicate the
intuitive notion of occuring at a spacetime point for processes, the notion of being located at
a spacetime point for gravitational fields and the notion of containing a spacetime point for
spacetime regions. (a) If processes are used as basic entities, we say for all sets F and all
processes u that u occurs at F iff F is a spacetime point such that u ∈ F . In the case that
spacetime points are treated as pluralities, the definition is analogous. For all xx and for all
processes u, u occurs at xx iff xx is a spacetime point and u is one of xx. (b) If gravitational
fields are used as basic entities, we can explicate the notion of occurring at in the following
way. In the set theoretic framework: for all sets F and all basic entities u that u is located
at F iff F is a spacetime point such that u ∈ F . In the framework of plural logic: for all xx
CHAPTER 5. METAPHYSICAL QUESTIONS 110

and all basic entities u, u is located at xx iff xx is a spacetime point iff u is one of xx. (c)
If spacetime regions are used as basic entities, then the notion of being contained in can be
explicated analogously. In the set theoretic framework: for all sets F and all spacetime regions
u, F is contained in u iff F is a spacetime point such that u ∈ F . In the framework of plural
logic: for all xx and for all spacetime regions u, xx is contained in u iff xx is a spacetime point
and u is one of xx. Using the so defined notions we can derive specific identity conditions
for spacetime points depending on the class of chosen basic entities, which can be formulated
neutrally (and therefore with less precision) as follows.

(ID20 ) Spacetime points are identical iff the same processes occur at them.

(ID200 ) Spacetime points are identical iff the same gravitational fields are located at them.

(ID2000 ) Spacetime points are identical iff they are contained in the same spacetime regions.

Note that these identity conditions do not hold for one and the same concept of spacetime
point. ID20 , ID200 and ID2000 rest on different underlying definitions of ‘spacetime point’.

The Question of the Relation to Concrete Entities

This question again depends on the kind of basic entities that are used for defining ‘spacetime
point’. In every case, the existence of spacetime points depends on the existence of the
respective basic entities and (in the set theoretic case) on the existence of sets. But since sets
are not concrete entities, the only concrete entities spacetime points depend on ontologically
are their basic entities.
It may be suspected that abstracting points from the mereotopological structure of concrete
basic entities is intrinsically tied to relationism and rules out substantivalism. However, as
we will see, this is not the case. But before we can present the argument, we need a correct
characterisation of relationism and substantivalism.
First note that the debate between substantivalism and relationism in metaphysics and
philosophy of science is often conceived as a debate about the existence of spacetime points
in the sense of atomic spacetime regions. See for example the following quotes:

“By ‘relationism’ I mean the doctrine that our physical theory does not commit
us to the existence of spacetime points, or spatial points and temporal instants.
I call the opposite doctrine ‘substantivalism’ about spacetime, space or time.”
(Butterfield, 1984, p. 101)

“I see the substantival-relational debate as being a debate about whether or not


the existence of spacetime points should in fact be postulated.” (Belot, 1996, p.9)
CHAPTER 5. METAPHYSICAL QUESTIONS 111

“Substantivalists hold that spacetime points are objects. [...] Relationists deny
that there are any such objects as spacetime points.” (Greaves, 2011, p.192)

However, substantivalism and relationism should not be characterised as a claims about the
existence of points. Hartry Field explains why not:

“Notice that it is quite misleading to describe the relationalist/substantivalist dis-


pute by saying that whereas the relationalist believes in aggregates of matter the
substantivalist believes also in space-time points. For in the first place, the substan-
tivalist may not believe in space-time points: he may not believe that the process
of taking smaller and smaller space-time regions has a limit, i.e., that there are
regions with no proper parts.” (1984, p. 35)

So the matter of dispute cannot be the existence of points in the sense of atomic regions.
Neither can it be the existence of points in the sense of classes of basic entities, because their
existence is even granted by relationists (cf. Butterfield, 1984), given that processes or the like
are taken as basic entities. The matter of dispute consists rather in the question whether
spacetime regions or points are ontologically dependent or independent of the basic entities
occurring at them. As Norton (2011) puts it: “spacetime is a substance, a thing that exists
independently of the processes occurring within spacetime. This is spacetime substantivalism.”
A similar (but again point-centered) fomulation is due to Earman (1989, p. 160): “the essence
of space-time substantivalism [is] the notion that events are happenings at space-time points
construed as ontologically prior to the happenings.” So I propose the following characterisation
of substantivalism and relationism.

(SUB) Spacetime regions are ontologically fundamental and independent of concrete en-
tities and their relations.

(REL) Spacetime regions are ontologically not fundamental and depend on concrete enti-
ties and their relations.

This characterisation and the above mentioned fact that spacetime points (understood as
abstractions from mereotopologically structured basic entities) are ontologically dependent on
their basic entities suggest that considering spacetime points as abstractions from concrete
basic entities implies relationism and rules out substantivalism. Nevertheless, considering
spacetime points as abstractions actually does not rule out substantivalism and does not
imply relationism.
The reason is this. One can assume SUB, but nonetheless regard spacetime points as
abstractions. Just take the class of spacetime regions as basic entities for the abstraction of
CHAPTER 5. METAPHYSICAL QUESTIONS 112

points. This is for example advisable if one cannot rule out that the mereological structure
of spacetime regions is atomless. In this case, substantivalism and considering points as ab-
stractions go well together (and hence are not inconsistent). For the same reason considering
points as abstractions does not imply REL.
However, REL can hardly be held without considering points as abstraction from concrete
basic entities (such as processes or gravitational fields). After all, physical theories presuppose
that there are spacetime points and spacetime regions. Since we can hardly afford to discard
those theories, it suggests itself to regard points as abstractions from acceptable concrete basic
entities, so that sentences about points and regions (regions understood as sets of points) can
at least in principle be reduced to sentences about the relations between concrete entities.
To sum up, considering points as abstractions neither rules out substantivalism nor implies
relationism, but relationism calls for considering points as abstractions.
CHAPTER 6

SUMMARY

6.1 Main Results


I gave an overview and classification of the definitions of ‘spacetime point’ that have been
proposed up to now and excluded some of these proposed definitions, because they violated
intuitive adequacy conditions (chapter 2). I treated the pre-theoretic philosophical problem
of finding out whether one of the remaining definitions (which all identify points with classes
of mereotopologically structured basic entities) is adequate by turning it into the mathemat-
ical problem whether there is a generally adequate point representation. For this purpose,
I developed a mathematical definition of the notion of point representation and I formalised
adequacy conditions for representation methods within the developed framework (section 3).
In the large chapter 4, I examined the point representation methods which correspond
to the most important kinds of remaining definitions. First, I analysed the ultrafilter repre-
sentation and showed that this point representation is not generally adequate because of the
problem of free ultrafilters (section 4.1). Next, I analysed the completely prime filter repre-
sentation and showed that it is also not generally adequate because of the problem of prime
elements (section 4.2). Then, I analysed the maximal round filter method and showed that it
is indeed a generally adequate point representation (section 4.3). Proving this result can be
viewed as the main achievement of this thesis.
Finally, I discussed metaphysical questions concerning the abstraction of points from mere-
topologically structured basic entities. I showed that defining ‘spacetime point’ along the lines
of the maximal round filter method does not imply that spacetime points have to be regarded
as abstract. Moreover, I argued that the problem of empty points, which arises when processes
are taken as basic entities, might be solved by taking gravitational fields as basic entities given
that gravitational fields are assumed to be concrete entities. I also discussed to which cate-
gory spacetime points belong, which identity conditions they have and how the approach of
identifying spacetime points with classes is related to substantivalism and relationism.

113
CHAPTER 6. SUMMARY 114

6.2 Open Problems


A problem that remains open in this thesis is the problem of axiomatisation described in
section 4.3.3. This problem can be formulated as follows. Is it possible to formulate axioms
for a unified mereotopological system which can be proven to determine a nice class of enriched
WMTS (in the sense of definition 4.3.8)? To solve the problem one has to find sufficiently
strong bridge principles connecting the notion of limitedness with the notions of parthood and
interior parhood. These principles should imply that an element in an atomistic structure
is limited just in case it is part of a compact element. Since compactness is defined for
atomistic, but not for atomless enriched WMTS, the problem boils down to the task of finding
a generalisation of the notion of compactness that applies not only to atomistic but also to
atomless structures. I do not know whether this can be done.
Another open question is this. Are there representation methods which are generally
adequate, but not equivalent or very similar to the maximal round filter representation? If
there are any, do they need stronger or weaker mereotopological assumptions than the round
filter representations?
BIBLIOGRAPHY

Arntzenius, F.: “Is Quantum Mechanics Pointless?” In: Philosophy of Science, Vol. 70(5),
2003, pp. 1447–1457.
Arntzenius, F.: “Gunk, Topology and Measure.” In: D. Zimmerman (ed.): Oxford Studies in
Metaphysics. Volume 4. Oxford: OUP, 2008, pp. 225–247.
Benda, T.: “A Formal Construction of the Spacetime Manifold.” In: Journal of Philosophical
Logic, Vol. 37(5), 2008, pp. 441–478.
Biacino, L. & Gerla, G.: “Connection Structures: Grzegorczyk’s and Whiteheads Definition
of Point.” In: Notre Dame Journal of Formal Logic, Vol. 37(3), 1996, pp. 431–439.
Birkhoff, G.: Lattice Theory. 2. (revised) edition. New York: American Mathematical Society,
1948.
Bombelli, L., Lee, J., Meyer, D. & Sorkin, R.D.: “Space-time as a Causal Set.” In: Physical
Review Letters. Vol. 59(5), 1987, pp. 521–524.
Bostock, D.: “Whitehead and Russell on Points.” In: Philosophia Mathematica (III), Vol. 18(1),
2010, pp. 1–52.
Bourbaki, N.: General Topology. Paris: Hermann, 1966.
Bub, J.: Interpreting the Quantum World. Cambridge: Cambridge University Press, 1997.
Bunge, M.: Foundation of Physics. Berlin-Heidelberg-New York: Springer, 1967.
Butterfield, J.: “Relationism and Possible Worlds”. In: British Journal for the Philosophy of
Science, Vol. 35, 1984, pp. 101–113.
Casati, R. & Varzi, A.C.: Parts and Places: The Structures of Spatial Representation. Cam-
bridge: MIT Press, 1999.
Chakravartty, A.: “Scientific Realism.” In: E.N. Zalta (ed.): The Stanford Encyclopedia of
Philosophy (Spring 2014 Edition), 2011,
URL = <http://plato.stanford.edu/archives/spr2014/entries/scientific-realism/>.
Chang, C.C. & Keisler, J.H.: Model Theory. Amsterdam: Elsevier, 1990.
Clarke, B.L.: “Individuals and Points.” In: Notre Dame Journal of Formal Logic, Vol. 26(1),
1985, pp. 61–75.

115
BIBLIOGRAPHY 116

Covarrubias, G.M.: “An Axiomatization of General Relativity.” In: International Journal of


Theoretical Physics, Vol. 32(11), 1993, pp. 2135–2154.
De Laguna, T.: “Point, Line and Surface as Sets of Solids.” In: Journal of Philosophy, Vol. 9,
1922, pp. 449–461.
Düntsch, I. & Winter, M.: “A Representation Theorem for Boolean Contact Algebras.” In:
Theoretical Computer Science, Vol. 347, 2005, pp. 498–512.
Einstein, A.: “Appendix 5: Relativity and the Problem of Space.” In: Relativity: The Special
and the General Theory. New York, Crown Publishers: 1961.
Eschenbach, C.: “A Mereotopological Definition of ‘Point’.” In: C. Eschenbach, Ch. Habel &
B. Smith (eds.): Topological Foundations of Cognitive Science. Papers from the Workshop
at the FISI-CS, Buffalo, NY. July 9-10, 1994. Graduiertenkolleg Kognitionswissenschaft
Hamburg, Bericht 37, 1994, pp. 63–80.
Esfeld, M. & Lam, V.: “Moderate Structural Realism about Space-Time.” In: Synthese,
Vol. 160(1), 2008, pp. 27–46.
Field, H.: “Can We Dispense with Space-Time?” In: PSA: Proceedings of the Biennial Meeting
of the Philosophy of Science Association, Vol. 1984, Volume Two: Symposia and Invited
Papers, 1984, pp. 33–90.
Forrest, P.: “Mereotopology without Mereology.” In: Journal of Philosophical Logic, Vol. 39,
2010, pp. 229–254.
Gerla, G.: “Pointless Geometries.” In: F. Buekenhout (ed.): Handbook of Incidence Geometry.
Amsterdam: Elsevier, 1995, pp. 1015–1031.
Givant, S. & Halmos, P.: Introduction to Boolean Algebras. New York: Springer, 2009.
Gruszczyński, R. & Pietruszczak, A.: “Full Development of Tarski’s Geometry of Solids.” In:
The Bulleting of Symbolic Logic, Vol. 14, 2008, pp. 481–540.
Gruszczyński, R. & Pietruszczak, A.: “Space, Points and Mereology: On Foundations of Point-
free Euclidean Geometry.” In: Logic and Logical Philosophy, Vol. 18, 2009, pp. 145–188.
Grzegorczyk, A.: “Axiomatizability of Geometry Without Points.”, In: Synthese, Vol. 12, 1960,
pp. 228–235.
Halpern, J.D.: “The Independence of the Axiom of Choice from the Boolean Prime Ideal
Theorem.” In: Fundamenta Mathematicae, Vol. 55, 1964, pp. 57–66.
Huntington, E.V.: “Postulates for abstract geometry.” In: Mathematische Annalen, Vol. 73,
1913, pp. 522–559.
Johnstone, P.T.: Stone Spaces. Cambridge: Cambridge University Press, 1982.
Johnstone, P.T.: “The Point of Pointless Topology.” In: Bulletin of the Americal Mathematical
Society, Vol. 8(1), 1983, pp. 41–53.
Ladyman, J.: “Structural Realism.” In: E.N. Zalta (ed.): The Stanford Encyclopedia of Phi-
BIBLIOGRAPHY 117

losophy (Spring 2014 Edition), 2014,


URL = <http://plato.stanford.edu/archives/spr2014/entries/structural-realism/>.
Lehmkuhl, D.: “Is Spacetime a Gravitational Field?” In: D. Dieks (ed.): The Ontology of
Spacetime II. Amsterdam: Elsevier, 2008, pp. 83–110.
Linnebo, Ø.: “Plural Quantification.” In: E.N. Zalta (ed.): The Stanford Encyclopedia of
Philosophy (Fall 2014 Edition), 2012,
URL = <http://plato.stanford.edu/archives/fall2014/entries/plural-quant/>.
Mac Lane, S. & Moerdijk, I.: Sheaves in Geometry and Logic: A First Introduction to Topos
Theory. New York: Springer, 1992.
Marquis, J.-P.: “From a Geometrical Point of View: A Study of the History and Philosophy
of Category Theory.” Dordrecht: Springer, 2009.
Maudlin, T.: “Time, Topology and Physical Geometry.” In: Proceedings of the Aristotelian
Society Supplementary Volume, Vol. 84(1), 2010, pp. 63–78.
Menger, K.: “Topology Without Points.” In: Rice Institute Pamphlets, Vol. 27, 1940, pp. 80–
107.
Misner, C., Thorne, K.S. & Wheeler, J.A.: Gravitation. San Francisco: Freeman, 1973.
Mormann, T.: “Continuous Lattices and Whiteheadian Theory of Space.” In: Logic and
Logical Philosophy, Vol. 6, 1998, pp. 35–54.
Mormann, T.: “Topological Representation of Mereological Systems.” In: Poznan Studies in
the Philosophy of the Sciences and the Humanities, Vol. 76, 2000, pp. 463–486.
Mormann, T.: “Russell’s Many Points.” In: A. Hieke, H. Leitgeb (eds.): Reduction - Abstrac-
tion - Analysis. Proceedings of the 31th International Ludwig Wittgenstein-Symposium in
Kirchberg, 2008. Frankfurt: Ontos, 2009, pp. 239–258.
Munkres, J.: Topology. 2nd edition. Englewood Cliffs: Prentice Hall, 2000.
Naimpally, S.A., & Warrack, B.D.: Proximity Spaces. Cambridge: Cambridge University
Press, 1970.
Nicod, J.: Foundations of Geometry and Induction. New York: Harcourt, Brace and Co.,
1930.
Reichenbach, H.: The Philosophy of Space and Time. New York: Dover, 1957.
Ridder, L.: Mereologie: ein Beitrag zur Ontologie und Erkenntnistheorie. Frankfurt: Kloster-
mann, 2002.
Robb, A.A: Geometry Of Time And Space. Cambridge: Cambridge University Press, 1936.
Roeper, P.: “Region-based topology.” In: Journal of Philosophical Logic. Vol. 26, 1997,
pp. 251–309.
Russell, B.: The Principles of Mathematics. New York: Norton, 1903.
Russell, B.: Our Knowledge of the External World. London: Allen & Unwin, 1914.
BIBLIOGRAPHY 118

Russell, B.: Mysticism and Logic, and Other Essays. London: Longmans Green, 1918.
Russell, B.: The Analysis of Matter. New edition. London: Routledge, 1927.
Simons, P.: Parts. A Study in Ontology. Oxford: Clarendon Press, 1987.
Skyrms, B.: “Atoms and Combinatorial Possibility.” In: The Journal of Philosophy, Vol. 90(5),
1993, pp. 219–232.
Smith, B.: “Ontology and the Logistic Analysis of Reality.” In: G. Haiger & P.M. Simons
(eds.): Analytic Phenomenology. Kluwer: Dordrecht, 1993, pp. 223–245.
Smith, B.: “Mereotopology: A Theory of Parts and Boundaries.” In: Data and Knowledge
Engineering, Vol. 20, 1996, pp. 287–303.
Stone, M.H.: “The Theory of Representation for Boolean Algebras.” In: Transactions of the
American Mathematical Society, Vol. 40(1), 1936, pp. 37–111.
Tarski, A.: “Zur Grundlegung der Boole’schen Algebra. I.” In: Fundamenta Mathematicae,
Vol. 24(1), 1935, pp. 177–198.
Tarski, A.: “Foundations of the Geometry of Solids.” In: Logic, Semantics, Metamathematics.
Oxford: Oxford University Press, 1956, pp. 24–30.
Vakarelov, D., Dimov, G., Düntsch, I. & Bennett, B.: “A Proximity Approach to Some Region-
Based Theories of Space.” In: Journal of Applied Non-Classical Logics, Vol. 12(3/4), 2002,
pp. 527–559.
van Fraassen, B.: The Scientific Image. Oxford: Oxford University Press, 1980.
Wheeler, J.A.: “Curved Empty Space-Time as the Building Material of the Physical World.”
In: E. Nagel, P. Suppes & A. Tarski (eds.): Logic, Methodology and Philosophy of Science.
Proceedings of the 1960 International Congress. Stanford: Stanford University Press, 1962.
Wheeler, J.A. & Taylor, E.F.: Spacetime Physics. San Francisco: Freeman, 1963.
Whitehead, A.N.: The Concept of Nature. Cambridge: Cambridge University Press, 1920.
Whitehead, A.N.: Process and Reality. An Essay in Cosmology. New York: Macmillan, 1929.
Zimmerman, D.W.: “Could Extended Objects Be Made Out of Simple Parts? An Argument for
‘Atomless Gunk’.” In: Philosophy and Phenomenological Research. Vol. 56, 1996, pp. 1–29.

You might also like