Barton

Many philosophers are aware of the paradoxes of set theory

(e.g. Russell’s paradox). For many people, these were solved by
the iterative conception of set which holds that sets are formed
in stages by collecting sets available at previous stages. This
Element will examine possibilities for articulating this solution.
In particular, the author argues that there are different kinds of The Philosophy of
iterative conception, and it’s open which of them (if any) is the
best. Along the way, the intent of this Element is to make some
Mathematics
of the underlying mathematical and philosophical ideas behind
tricky bits of the philosophy of set theory clear for philosophers
more widely and make their relationships to some other
questions in philosophy perspicuous.

Iterative Conceptions of Set

Iterative
Conceptions
About the Series
This Cambridge Elements series provides
Series Editors
Penelope Rush
of Set
an extensive overview of the philosophy University of
of mathematics in its many and varied Tasmania
forms. Distinguished authors will provide
an up-to-date summary of the results of
current research in their fields and give
Stewart Shapiro
The Ohio State
University
Neil Barton

https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

their own take on what they believe are
the most significant debates influencing
research, drawing original conclusions.

Cover image: CTRd / Getty Images ISSN 2399-2883 (online)

ISSN 2514-3808 (print)
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press
 Elements in the Philosophy of Mathematics
edited by
Penelope Rush
University of Tasmania
Stewart Shapiro
The Ohio State University

ITERATIVE CONCEPTIONS OF SET

Neil Barton
National University of Singapore
and University of Oslo
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press
 Shaftesbury Road, Cambridge CB2 8EA, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre,
New Delhi – 110025, India
103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467

Cambridge University Press is part of Cambridge University Press & Assessment,

a department of the University of Cambridge.
We share the University’s mission to contribute to society through the pursuit of
education, learning and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9781009478526
DOI: 10.1017/9781009227223
© Neil Barton 2024
This publication is in copyright. Subject to statutory exception and to the provisions
of relevant collective licensing agreements, no reproduction of any part may take
place without the written permission of Cambridge University Press & Assessment.
When citing this work, please include a reference to the DOI 10.1017/9781009227223
First published 2024
A catalogue record for this publication is available from the British Library.
ISBN 978-1-009-47852-6 Hardback
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

ISBN 978-1-009-22726-1 Paperback

ISSN 2399-2883 (online)
ISSN 2514-3808 (print)
Cambridge University Press & Assessment has no responsibility for the persistence
or accuracy of URLs for external or third-party internet websites referred to in this
publication and does not guarantee that any content on such websites is, or will
remain, accurate or appropriate.
 Iterative Conceptions of Set

Elements in the Philosophy of Mathematics

DOI: 10.1017/9781009227223
First published online: May 2024

Neil Barton
National University of Singapore and University of Oslo

Author for correspondence: Neil Barton, neilalexanderbarton@gmail.com

Abstract: Many philosophers are aware of the paradoxes of set theory (e.g.
Russell’s paradox). For many people, these were solved by the iterative
conception of set which holds that sets are formed in stages by collecting
sets available at previous stages. This Element will examine possibilities for
articulating this solution. In particular, the author argues that there are
different kinds of iterative conception, and it’s open which of them (if any)
is the best. Along the way, the intent of this Element is to make some of
the underlying mathematical and philosophical ideas behind tricky bits of
the philosophy of set theory clear for philosophers more widely and make
their relationships to some other questions in philosophy perspicuous.

Keywords: set theory, iterative conception, uncountable sets, foundations of

mathematics, infinity

© Neil Barton 2024

ISBNs: 9781009478526 (HB), 9781009227261 (PB), 9781009227223 (OC)
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

ISSNs: 2399-2883 (online), 2514-3808 (print)

 Contents

1 Introduction 1

2 Why Set Theory? 4

3 The Naive Conception of Set and the Classic Paradoxes 16

4 The Logical and Combinatorial Conceptions of Set 21

5 Iterative Conceptions: First Examples 25

6 Forcing as a Construction Method 41

7 A ‘New’ Kind of Paradox? 52

8 Countabilist Conceptions of Iterative Set 56

9 Mathematics and Philosophy under the Different

Conceptions 65

10 Conclusions, Open Questions, and the Future 75

https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

References 87
 Iterative Conceptions of Set 1

1 Introduction
If you’re reading this Element, then I presume that you’re curious about infin-
ity, set theory, and its philosophy. Growing up, I’d always been interested in
philosophy. Mathematics, however, I found to be a necessary but tiresome part
of the curriculum, especially through my teenage years. I had great teachers, but
the focus on exam preparation that inevitably took up the bulk of our time was
just plain boring – solving dreary computational problems using known algo-
rithmic methods (a task that I’m not especially good at to this day). This didn’t
fit so well with what my mother Jeanne (a mathematics teacher) had always
told me – that at a certain point mathematical study can feel like ‘doors open-
ing left and right’. It was at university that I saw Cantor’s theorem and Gödel’s
theorems for the first time. Suddenly I understood what my mum had meant –
mathematics was an area where new ideas and methods could result in a com-
plete shift in one’s perspective on the world, and your ability to solve problems
is only bounded by your creativity and the constraints of logical space. The
doors were very much open, and I became increasingly interested in notions of
infinity in mathematics. To understand infinity, it’s very natural to start by con-
sidering our best mathematical theories of it. Set theory, as a theory of infinite
collections and what we can do with them, was the obvious choice. Understand-
ably, philosophers have shown a lot of interest in set theory since its begin-
nings in the late nineteenth and early twentieth centuries. There was already
plenty of philosophical material to get my teeth into, and I tucked in with
gusto.
What I discovered, however, was that the buffet was far richer than I had
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

anticipated. In particular, several philosophical and mathematical advances

have been made in the philosophy of set theory since the early 2000s. Both
mathematicians and philosophers have closely examined ideas concerning
whether there is an all-encompassing domain for set theory, and how the tools
of contemporary set-theoretic practice might bear on philosophy. This has tied
the study of the philosophy of set theory very closely to issues in metaphysics,
including the nature of possibility and absolute generality. However, I think it’s
fair to say that these developments (with some notable exceptions) have been
passed over for mainstream philosophical consideration. Whilst this is under-
standable – the mathematical barrier to entry is high and our time is finite – the
philosophical issues themselves are (in my opinion) understandable to anyone
with some introductory logic courses under their belt.
Many philosophers are aware of the paradoxes of set theory (e.g. Russell’s
paradox). Often people take these to be solved by the iterative conception of set
which holds that sets are formed in stages by collecting together sets available
 2 The Philosophy of Mathematics

at previous stages. This Element will examine possibilities for articulating this
solution. In particular:

Main Aim I will argue that there are different kinds of iterative concep-
tion, and it’s open which of them (if any) is the best.

Along the way, I hope to make some of the underlying mathematical and
philosophical ideas behind tricky bits of the philosophy of set theory clear for
philosophers more widely, and make their relationship to other questions in
philosophy perspicuous.
Here’s the plan. Section 2 will lay down some reasons as to why we should be
interested in set theory as philosophers and mathematicians. This section serves
a dual purpose; first, as a motivation for the reader less familiar with set theory
to get excited, and second, we’ll see some desiderata that will be employed
later in the Element when we come to assess set-theoretic conceptions.
Sections 3, 4, and 5 set up a way of thinking of set-theoretic progress as trad-
ing off inconsistent principles. Section 3 will go over the naive conception of
set and the paradoxes that brought it down. We’ll also provide a diagnosis of
the problem as involving a conflict of two inconsistent principles. This mate-
rial is well-worn, but I’ll explain a twist on the classic paradoxes that has been
examined by philosophers recently (namely that we can think of these para-
doxes as being about the existence of functions) which will help integrate this
material with what comes later. Section 4 will present the emergence of the
combinatorial conception and logical conception of set, before Section 5 gives
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

the iterative conception as a further sharpening of the combinatorial concep-

tion. We’ll also explain the standard ‘strong’ version of the iterative conception
and how it can be given a modal formulation.
Section 6 will then explain some mathematical ideas that have informed the
development of contemporary set theory under the iterative conception, namely
forcing (a way of adding subsets of sets to models). I’ll do my best to make these
mathematically tricky ideas palatable to philosophers.
Section 7 will explain two particular principles that we might take a very
‘rich’ conception of set to have and show them to be inconsistent. In particular,
we’ll see how the power-set axiom is incompatible with the idea that there
should be saturation under forcing. We’ll note that there’s a similarity here with
the situation we found ourselves in with respect to the naive conception, where
a particular conception of set generates two inconsistent principles.
Sections 8 and 9 will identify a split in how we might move forward. Sec-
tion 8 will explain how there is a genuine choice between powerset and forcing
saturation, and will show how forcing saturation can be viewed as arising from
 Iterative Conceptions of Set 3

kinds of set-construction method. Section 9 will explain how mathematics is

interpreted within each conception and will contrast each in the light of the
theoretical virtues discussed in Section 2.
Finally, Section 10 will provide a concluding summary and identify some
further work that is needed in order to obtain greater clarity on these issues.
In particular, I’ll explain some salient objections that need addressing in order
to move forward. I hope that the reader comes away with a sense of how set
theory is philosophically interesting and the vastness of conceptual space.
Before we get going, however, a few remarks are in order. First, whilst I
hope that this Element is of pedagogical value and can help people new to the
philosophy of set theory gain an understanding of some difficult mathematics,
this is not a textbook. My approach is one of conveying underlying ideas, rather
than giving everything in full rigorous detail. Where sensible I’ve tried to give
formal definitions and references for the interested reader in footnotes rather
than the body of the text.
Relatedly, the pacing of this Element will feel slightly odd. There is a tension
in exposition in that I both want to get the newcomer interested but also accom-
plish a significant research-oriented goal. I therefore run the risk of boring the
reader who has been studying these issues for years whilst outstripping what
can be expected of an early student (however talented). I’ve tried to present the
known material in such a way that it makes recent novel twists on old material
clear, and to keep the harder material as accessible as possible. However, this
Element is hard if you aren’t familiar with the relevant bits of mathematical
logic. My aim is to make things accessible and not, per impossibile, easy. To
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

combat this problem, the Element runs along two tracks. The ‘standard’ track is
intended for those who do not necessarily have years of philosophy of set the-
ory under their belt. The ‘expert’ track is for those who already know a good bit
of philosophy of set theory. I denote sections/paragraphs/footnotes that are on
the expert track with a ‘blackbelt’ emoji (and often inside a box). I encour-
age everyone to read all the Element, after all it’s helpful to peek behind the
curtain and see some of the complicated workings of the machine. But read-
ers should not feel disheartened if -parts are tricky to follow – those are
especially difficult and one shouldn’t expect to get everything on the first try.
I’ll use the following conventions. Bits of language (e.g. syntax/utterances)
will be enclosed within double quotation marks. So “Toffee is a clever cat” can
be a sentence or an utterance, “cat” is a word or term of the English language,
and “Toffee” is a name (in this context), whereas Toffee is a (particular) cat who
is also clever. Single quotation marks will be used as ‘scare quotes’, namely
cases where the enquoted phrase is not to be taken literally (though it may
be illustrative). In cases where such usage occurs in a formal context, single
 4 The Philosophy of Mathematics

quotes often denote an abbreviation for a formal claim (e.g. PA ⊢ ‘There are
infinitely many prime numbers’, even though “There are infinitely many prime
numbers” is a sentence of English, not Peano Arithmetic). Italics are reserved
for emphasis, or where they occur in the scope of a definition, the definiendum.
I allow definitions to be informal and philosophical as well as formal, but I will
clearly separate the informal and formal definitions. With these conventions in
hand, let’s get ready to set out!

2 Why Set Theory?

Before we start getting into the iterative woods, I want to give some motivation
for studying set theory and its philosophy.

Question Why do this, given that there are so many good introductions into
these topics?

Answer As well as providing a survey of some of the literature, this section

will lay down some theoretical virtues that we might think theories/conceptions
of set can have. These virtues will be important later when we come to assessing
our options.

What are sets? Here’s a rough-and-ready definition:1

Definition 1 (Informal) A set is a kind of collection that is:

https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

(i) Extensional: Sets with different members are non-identical, and sets with
the same members are identical.
(ii) Objectual: Sets are objects over and above their elements.

So, for example, I can consider the set of books currently on my table. This
is an object, in addition to the books themselves. If I take a book off my table,
the term “the set of books on my table” now denotes a different set, since this
new set of books has different members.
Just given this bare bones story, it’s natural now to ask: Why be interested
in set theory at all? It’s useful first to consider a bad answer (but one that helps
us see the role of set theory more clearly):

1 It’s plausible that nowadays we think that sets are combinatorial too (in the sense of being
extensionally equivalent to pluralities of objects, irrespective of whether we can provide a
circumscribing definition). Later we’ll set up the difference between the logical conception
and combinatorial conception of set, and so I don’t want to commit to this just yet.
 Iterative Conceptions of Set 5

Theory of Collections Set theory provides our best theory of collections.

This is perhaps encapsulated by George Boolos’ (1998) claim ‘I thought that

set theory was supposed to be a theory about all, ‘absolutely’ all, the collections
that there were and that ‘set’ was synonymous with ‘collection’ (p. 35)2
The idea that the interest of set theory derives from ‘set’ being synonymous
with ‘collection’ or providing our best theory of collections is open to at least
two powerful criticisms. First, there are lots of different ways we talk about col-
lections. To take two simple kinds: (1) Collection-like talk needn’t be objectual.
As the vast literature on plural logic indicates,3 we can talk about and quantify
over objects in the plural without thereby committing to a set of them. So,
instead of talking about the set of books on my table, I could just have talked
about the books on my table in the plural. (2) Collection-like talk needn’t be
extensional. Instead, it can be taken intensionally, where identity is not taken to
be governed by an extensionality criterion. Presumably there’s a sense in which
I don’t destroy my beer coaster collection just by giving one of the (many) beer
coasters to a friend. My collection of beer coasters is just the kind of thing that
can survive a loss (or better yet, gain) of some members.
Second, even if set theory did provide our best theory of collections, there’s
much more to the story. Collections of beer coasters are a perfectly good sub-
ject matter for philosophical study, but this observation fails to explain why set
theory is often regarded as central to many areas (and especially mathematics).
Here’s what I take to be the core point: Objectual and extensional collections,
when augmented with the ‘right’ axioms, are powerful devices of represen-
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

tation. And the ability to represent means that all sorts of problems, both
philosophical and mathematical, can be encoded within set theory.
Let’s look at this idea in a little more detail. This representational power
presents two interlinked aspects of set theory:

Foundation for Mathematics Set theory provides a ‘foundation’ for mathe-

matics (and hence mathematical tools in philosophy).

Philosophical Repository Set theory examines many philosophically inter-

esting subjects (e.g. paradoxes, infinity).

2 Boolos here is discussing the contrast between sets and proper classes, so perhaps the
quotation is intended for a slightly different context. Indeed, Boolos himself was key in the
development and philosophical study of plural logic (see Boolos (1984) and so it’s likely that he
didn’t think all collection-like talk had to be encapsulated by set theory. Whatever the weather,
just putting the idea that set theory provides our best theory of collections out there is enough
to get the ball rolling at this stage.
3 See Florio and Linnebo (2021) for a book-length treatment.
 6 The Philosophy of Mathematics

This division is far from exclusive. Certainly there are cases where we might
think that set theory and philosophy are inextricably intertwined.4 Indeed, this
Element emphasises the fact that mathematics and philosophy can become
fruitfully intermixed, and I do not think it is either necessary or desirable to
keep these considerations separate. Nor do I think that every bit of set theory
will be entangled with philosophy, and there are set theorists who study solely
mathematical questions. Still, the distinction serves as a rough categorisation
for different facets of set theory.
At this stage, we’ll keep things relatively informal, but a little precision will
be helpful. One set theory that’s proved to be of central interest is Zermelo–
Fraenkel set theory with the Axiom of Choice (ZFC), which we’ll examine more
closely later. For now let’s just content ourselves with the following rough char-
acterisation: ZFC tells you that there are lots of sets (both finite and infinite)
and lets you do many of the usual set-theoretic operations you want on those
sets (e.g. take the union of two sets).
Recently, Penelope Maddy has isolated some mathematical goals of set-
theoretic foundations built on ZFC.5 I’ll provide some examination of Maddy’s
ideas, and I’ll suggest some modifications and additions of my own.6 These
goals serve a dual purpose. On the one hand, they motivate the consideration
of set theory for the interested reader. On the other, we will use them later to
evaluate particular conceptions of set.
Earlier I mentioned that set theory is a powerful device of representation.
Many of the desiderata we’ll consider are linked to this idea. For instance:
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Observation We can encode/represent all mathematical objects using sets.7

What do I mean by ‘encode/represent’ here? Let’s take a simple example

from high-school mathematics. We want to consider some geometric object
in two-dimensional (Euclidean) space, let’s say a straight line. By picking an
origin and imposing a coordinate system, we can represent this straight line by
some function f (x) = bx + c and think of the straight line as composed of its

4 See, for example, Rittberg (2020) who argues that set-theoretic mathematical practice can be
metaphysically laden.
5 See Maddy (2017) and Maddy (2019).
6 For clarity’s sake, Generous Arena, Shared Standard, Metamathematical Corral, and Risk
Assessment are all explicitly identified by Maddy, and Theory of Collections, Foundation
for Mathematics, Philosophical Repository, Theory of Infinity, Independence, Limits of
Formalisation, and Testing Ground for Paradox are my own additions (though many are
implicit in much of the literature and Maddy’s work).
7 See Posy (2020), section 2, for a very concise survey of the classical situation (Posy sets up the
classical mathematician as a foil for intuitionism), as well as many set theory textbooks (e.g.
Enderton (1977).
 Iterative Conceptions of Set 7

graph of ordered pairs ⟨x, bx + c⟩. This can help us when, for example, trying
to compute the relative lengths of line segments (e.g. by using the Pythagorean
theorem). But the ordered pairs aren’t (intuitively speaking) the same as the
line, they just encode it.
So with sets, but generalised to any mathematical object you’d care to
consider. Zero can be encoded by the empty set, natural numbers by the
finite von Neumann ordinals,8 rationals as pairs of natural numbers, reals as
Dedekind cuts of rationals,9 ordered pairs as Kuratowski ordered pairs,10 and
functions/relations by sets of ordered pairs (i.e. the function f is encoded by
{⟨x, y⟩| f (x) = y}). Of course, there are lots of choices, and this is just an
illustration of one way you might do things.11
Using similar tactics, any mathematical object we have come up with can
be encoded by sets (putting aside some controversial cases).12 This has some
important consequences. First, set theory provides a:
Generous Arena Find representatives for our usual mathematical structures
(e.g. N, R) using our theory of sets.

I think it is worth pausing for a moment to reflect on just how remarkable

Generous Arena is. Just using the membership relation and suitable axioms,
we can find a representative for almost any object you’d care to discuss – all
the vertiginous diversity we see in mathematics can be captured by that one
little relation of membership.13 Because we can encode mathematical objects
as sets, we have a way of relating them to each other within a single domain.
This, Maddy argues, gives us:
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Shared Standard Provide a standard of correctness for proof in mathematics.

8 These can be defined inductively with 0 =df ∅ and n + 1 =df n ∪ {n}.

9 A Dedekind cut is a partition of the rational numbers into two non-empty sets A and B, where
A is closed downwards and does not contain a greatest element.
10 The Kuratowski ordered pair is given by ⟨a, b⟩ =df { {a}, {a, b} }.
11 See Barton et al. (2022) for some of the formal details and further citations.
12 For example, one controversial objection (e.g. Mac Lane (1986) and Muller (2001)) to set
theory goes something like this: ‘Everything in set theory has to be encoded by a set, and we
know that some categories like the category of all sets are too big to be encoded by sets. So set
theory cannot provide a foundation for category theory.’ I do not find this objection convincing
for the following two reasons. (1) set theorists certainly seem to talk about proper-class-sized
objects – the study of proper classes is in my (controversial) opinion a perfectly legitimate part
of set theory, and (2) I don’t think that category-theoretic study of the sets is really directed at
the study of all the sets, but rather the study of the schematic first-order properties that all the
sets happen to satisfy. A full defence of this idea will have to be left for a different day, but
a more detailed explanation of this point can be found in Barton and Friedman (2019) (esp.
section 10.3).
13 We’ll see how this formally plays out when we come to talk about the language of set theory –
see Definition 12 in Section 3.2.
 8 The Philosophy of Mathematics

The thought here is that because we have Generous Arena and can view
mathematical objects as encoded/represented by sets, a proof about a mathe-
matical object can be regarded as correct if it could be (in principle) translated
into a proof in set theory about properties of the relevant mathematical code(s).
Of course, ‘in principle’ is important here – outside of set-theoretic mathe-
matics it is very clunky to work with these codes, and we shouldn’t expect
mathematicians to actually go about their daily lives solely using the language
of set theory. The relevant language of the discipline in question is probably
more flexible than working with just membership. (A desire for a foundation
that ‘will capture the fundamental character of mathematics as it’s actually
done, that will guide mathematicians toward the truly important concepts and
structures, without getting bogged down in irrelevant details’ Maddy terms
Essential Guidance, and since all set theories we’ll consider here perform
pretty badly in this respect, we’ll set it to one side.)14
The ability to manipulate large infinite collections in ZFC-based set theory
yields the following:

Theory of Infinity Set theory provides our best theory of infinite numbers.

Theory of Infinity will be important later and so I’ve explicitly identified it

as a theoretical virtue in contrast to some of the literature that leaves it implicit
(it does not occur, for example, amongst the virtues identified by Maddy (2017)
and Maddy (2019)). To see its significance, we start by examining the two
main kinds of infinite number in set theory, namely ordinal and cardinal num-
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

bers. An ordinal number can be thought of as an answer to the question of

how long an infinite ordering is. Call a set x (under a linear relation R) well-
ordered by R if and only if every subset of x has an R-least element. If x is
well-ordered by R, then there’s no way of descending infinitely in x along R.
This helps us think of performing actions or operations into the infinite along
a suitable infinite relation. Within ZFC one can represent and develop an arith-
metic for these orders, defining notions of ordinal addition, multiplication, and
exponentiation.15 This provides us with ways of generalising normally finite
operations (e.g. computation) into the infinite.16

14 See Maddy (2017), p. 305.

15 There are lots of ways to do this, but one popular way is to use von Neumann ordinals, where we
∪
let 0 = ∅, α + 1 = α ∪ {α}, and limit λ = β <λ β. Addition is represented by the ordered dis-
joint union, multiplication by the lexicographical ordering on the product, and exponentiation
by iterated multiplication.
16 For example, we can consider infinite time Turing machines. See Hamkins and Lewis
(2000).
 Iterative Conceptions of Set 9

Cardinal numbers, by contrast, can be thought of as answers to the question

of how many objects there are in a set. In particular, we say that two sets X and Y
have the same cardinality if and only if there is a bijection between them, where
a bijection f : X ↣
→ Y is a function that ‘pairs off’ the members of X and Y, that
is, f takes no two elements of X to the same element of Y ( f is injective) and
every element of Y is hit by f applied to some element of X ( f is surjective). By
representing cardinals using particular kinds of sets, ZFC provides a theory in
which the cardinal sizes of any sets can be compared and natural operations like
multiplication, addition, and exponentiation generalised and computed.17 The
success of ZFC is striking; it seemingly gives finite beings (e.g. us) the ability to
reason about large infinite objects. Many surprising facts can be thereby shown.
For example, we can prove:

Theorem 2 There are as many natural numbers as there are squares of nat-
ural numbers (in particular f (x) = x2 is just such a bijection from the natural
numbers to the squares of naturals).

This is somewhat surprising since the squares of n and n + 1 get more and
more spread out as n gets larger. Indeed, similar results were even regarded as
kinds of ‘paradox’ by Thābit ibn Qurra and Galileo. We can even show:

Theorem 3 The set of all rational numbers – the numbers expressible by

fractions – is the same size as the set of all natural numbers.18

This is so even though there are infinitely many rational numbers between
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

any two natural numbers. We can also show:

Theorem 4 There are as many real numbers between 0 and 1 (or any two real
numbers for that matter) as there are in the real line, or in any n-dimensional
plane based on the real line (i.e. Rn ).19

Despite these surprising results on sameness of size, we also discovered that

infinity comes in different cardinal sizes:

Theorem 5 (Cantor’s theorem for the reals) The cardinality of real numbers
is greater in size than the cardinality of the natural numbers, in the sense that

17 Again, there’s a variety of ways one might proceed, but here’s a typical one. The cardinality
of X can be represented as the least von Neumann ordinal bijective with X. Cardinal addition
can be computed as the cardinality of the disjoint union, multiplication as the cardinality of the
product, and exponentiation XY as the cardinality of the set of all functions from Y to X.
18 See, for example, chapter 2 of Giaquinto (2002) for an explanation of this result.
19 Again, see chapter 2 of Giaquinto (2002).
 10 The Philosophy of Mathematics

(i) there is no bijection between the natural numbers and the real numbers, and
(ii) there is an injection from the natural numbers to the real numbers.

This phenomenon appears to be more general than merely comparing the

natural numbers and real numbers. We in fact discovered that:

Theorem 6 (Cantor’s theorem) Let P(x) denote the powerset of x, the set of
all subsets of x (that such a set always exists is one of the central axioms of
ZFC). Then the cardinality of P(x) is greater than that of x.20

Again, Cantor’s theorem is striking. It seems to imply, on the basis of natural

principles about sets, that if there’s one infinite set, then there’s a never-ending
hierarchy of infinite sets, since the powerset of any set x is always bigger than
x. Moreover, it produces much of the interest of cardinal arithmetic – whilst
addition and multiplication are trivial for infinite cardinal numbers (one can
show that both addition and multiplication just result in getting the larger of
the two back), cardinal exponentiation is not – one can show that 2κ > κ for
any cardinal κ.21
The ability to work with infinity plays out in various areas of philoso-
phy, including areas outside the philosophy of mathematics.22 Indeed, these
arguments are often regarded as a refutation of the time-honoured position in
philosophy and mathematics that infinity is completely beyond understanding
and intractable.23

20 We’ll discuss a proof of Cantor’s theorem later, in particular as it relates to the paradoxes in
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Section 3.
21 In particular, you can think of 2 κ as the size of P(κ), since any member of P(κ) can be cor-
related with a unique function from κ to 2 = {0, 1} via characteristic functions (where for
X ⊆ κ, f(α) = 1 if and only if α ∈ X).
22 Here’s an example from infinite ethics showing how infinite assumptions can play out with
utility calculations (the example is due to Cain (1995). Suppose we have people arranged at all
coordinates of the real plane indexed by integers (so there’s a single person at every (m, n) for
integers m and n). A circle slowly grows from the origin. In one scenario (the circle of happi-
ness), everyone starts at utility −1 and moves to utility +1000 (or any large finite amount) when
they fall inside the perimeter of the circle (and remains at this value forevermore). For the circle
of negativity, each agent starts at +1 and goes to −1000 when they get caught by the circle. With
simple cardinality arguments one can argue that the sum of the utility for the expanding sphere
of negativity is positively infinite, whereas the expanding sphere of happiness is negatively
infinite (one needs to define these terms, but the rough idea is that there’s always boundedly
many happy/sad people in the circle of happiness/negativity, whereas infinitely many people of
the opposite disposition). Cain argues that we should nonetheless prefer to be in the expanding
happiness world (since then we just have to wait long enough to be blissfully happy forever-
more). Thanks to Joel David Hamkins for communicating this example to me; see Hamkins
and Montero (2000) for some further discussion.
23 See, for example, the paradoxes of the infinite given in the Introduction to Moore (1990). The
place of Cantor, his results, and other scholars in arriving at a final acceptance of infinity is
actually somewhat more subtle than is often acknowledged (see Ferreirós (2007), especially
 Iterative Conceptions of Set 11

However, this success must be tempered by the following phenomenon that

emerged in the twentieth century:

Independence There are sentences of set theory that can neither be proved
nor refuted using our ‘canonical’ theory of sets ZFC, assuming that ZFC is
consistent. Nor can any ‘reasonable’ expansion of ZFC settle all questions
formalisable in the language of set theory.24

Before we discuss this further, let’s remark that the mere fact of independence
is philosophically important. It shows that there will be limits to what any single
formal theory can capture. There are at least two kinds of independence that will
be relevant for us. To set things up, let’s start with the following:

Definition 7 We let the cardinal numbers be indexed by ordinals using a func-

tion we’ll call the ‘aleph’ function (or ℵ). ℵ0 is the smallest cardinal number
(which happens to be the cardinality of the natural numbers). ℵ1 is the next
smallest, and more generally ℵα is the αth cardinal number. We’ll denote
the ordinal corresponding to ℵα by ωα (we’ll often also let ω0 be denoted
by “ω”).

A routine argument shows that 2ℵ0 > ℵ0 (by Cantor’s theorem). But is
there anything in between? That is, does 2ℵ0 = ℵ1 ? Or are there intermediate
cardinalities, and in fact 2ℵ0 > ℵ1 ?

Definition 8 We will use the following for discussing the spread of cardinali-
ties:
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

• The Continuum Hypothesis (or CH) is the statement that 2ℵ0 = ℵ1 .

• The Generalised Continuum Hypothesis (or GCH) is the statement that ‘For
every ordinal α, 2ℵα = ℵα+1 ’ (i.e. every jump in cardinality obtained by
applying the powerset operation to an infinite set just pushes you up one
cardinal number).
• The continuum function is defined by f(ℵα ) = 2ℵα (i.e. the function that
takes an infinite cardinal to the cardinality of its powerset).

the Introduction). In particular, it is somewhat unclear whether our notion of cardinality

had to be the Cantorian one, or we might have ended up with a version of cardinality that
respects the idea that a proper part should always be smaller than the whole. Paolo Mancosu
has championed this idea; see Mancosu (2009) for analysis and references to its mathematical
development (e.g. in the work of Katz, Benci, Di Nasso, and Forti). Gödel provides an argument
that the right notion of cardinality is Cantorian in the opening to his paper on the continuum
hypothesis (Gödel (1947), with revisions in Gödel (1964)), which has in turn been critically
examined by Matthew Parker (see Parker (2019)).
24 Here ‘reasonable’ means recursively enumerable and consistent.
 12 The Philosophy of Mathematics

As it turns out, CH, ¬CH, GCH, and ¬GCH are all consistent with ZFC
(assuming ZFC itself is consistent). We’ll explain how this works later (Sec-
tion 6).
To discuss the other kind of independence, we first need a brief foray
into consistency strengths. Within arithmetic, and hence within ZFC, one can
(computably) encode syntactic notions like sentence, formula, proof, and con-
sistency. This allows you to formulate a sentence within ZFC expressing the
idea that ZFC is itself consistent (more precisely, you can formalise within
ZFC the sentence that there’s no proof of a contradiction derivable from the
axioms of ZFC). Call this sentence Con(ZFC). But now we can point to:

Theorem 9 (Gödel’s second incompleteness theorem) Assuming that ZFC is

consistent, then neither Con(ZFC) nor ¬Con(ZFC) is provable in ZFC. More-
over, this theorem holds for any (suitably nice)25 theory that can represent
arithmetic.

Within set theory we can study a wide variety of sentences that have different
consistency strengths – one can often prove one extension of ZFC consistent
from another. As it turns out, CH and ¬CH are not like this (ZFC, ZFC + CH,
and ZFC + ¬CH are all equiconsistent in that one can prove each consistent
from the other). Obviously adding Con(ZFC) results in a consistency strength
increase. There are other principles – so-called large cardinal axioms – that
are important here. These serve as the natural indices for consistency strength.
They postulate the existence of sets with a lot of closure properties and if they
exist (or are consistent) we can prove that many theories are consistent by
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

finding models of the relevant kind. Set theory has in fact discovered a whole
hierarchy of these cardinals with stronger and stronger closure properties.

Here’s an example:

Definition 10 A cardinal κ is strongly inaccessible (or just inaccessible)

if and only if:

(i) κ is uncountable (i.e. it’s bigger than the cardinality of natural

numbers).
(ii) Given any set x smaller than κ, the cardinality of P(x) is also smaller
than κ. Such κ are called strong limit cardinals.

25 Namely recursively enumerable and consistent.

 Iterative Conceptions of Set 13

(iii) Given any set x smaller than κ, and any function f : x → κ, the range
of f is bounded by some γ < κ. That is, given such an x and f : x → κ,
we always have some γ < κ such that for every y ∈ x, f(y) < γ. Here,
we say that κ is regular.

It’s instructive to think about what such an axiom says. Such a κ seems
very big – clause (i) ensures it’s bigger than N, (ii) says that you can’t catch
it with something smaller by taking our favourite size-increasing opera-
tion (powerset), and clause (iii) says that you can’t catch it by mapping
a smaller object into it using a function. One can show that an inacces-
sible cardinal κ suffices to produce a model for ZFC (and much more),
and so by Gödel’s second incompleteness theorem you can’t produce an
inaccessible cardinal from ZFC alone. We can strengthen this axiom by
postulating that there is a cardinal κ that is (i) strongly inaccessible, and
(ii) has κ-many strongly inaccessibles beneath it. And these cardinals lie
right at the bottom of the large cardinal hierarchy.a
a See, for example, the diagram on p. 472 of Kanamori (2009) for an idea of the extent of
the space.

Those are the two kinds of independence we’ll consider. One (the CH kind),
results in no increase in consistency strength and often involves the relative
sizes of infinite cardinals. The other (the large cardinal kind) involves increases
in consistency strength, and one way to calibrate this is by considering sets
with ever greater and greater closure properties. These aren’t the only kinds
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

of independence (there are also strong axioms that don’t directly postulate the
existence of large cardinals)26 but these are the ones we’ll focus on.
We should pause for a moment to reflect on what this independence tells us
about our ability to provide formalisations of theories of sets. Whilst ZFC does
give us the resources to prove a great many things about the infinite, it does not
yield information about the values of many cardinal computations nor what
kinds of set exist with certain closure properties. How we might respond to this
situation will be a central theme of this Element, but it should be noted that
Independence is a reason for philosophers – that is, not just mathematicians –
to be interested in set theory. Assessing the impact of independence is central
for understanding how our thought, language, and theories relate to the world
and what we can (and maybe can’t) do. I think it’s important therefore to isolate
the following philosophical aspect of set theory.

26 See, for example, so-called Axioms of Definable Determinacy (Koellner, 2014).

 14 The Philosophy of Mathematics

Limits of Formalisation Set theory provides a natural place to examine the

limits of our formalisation, pushing the boundaries of what might be realisti-
cally expected to be captured, and exploring where formalisations may finally
give out.

It’s a beguiling question to think what the implications of Limits of For-

malisation might be. Does it imply that there are limits on what can be known?
Or that there is some kind of metaphysical indeterminacy in the world? These
are important questions for philosophers, and show that Independence is not
merely a mathematical curio.
From the mathematical perspective, set theory is one of the main theories in
which we study Independence. It provides us with flexible tools with which we
can study models of different theories, how they can be built from one another,
and hence how relative provability works (given the completeness theorem).
We can thus (with Maddy) identify:

Metamathematical Corral Provide a theory in which metamathematical

investigations of relative provability and consistency strengths can be easily
conducted.27

As philosophers, we should be keen to assess whether the theories we work

in are consistent. Metamathematical Corral combined with the fact (as we’ll
see later) that set theory often comes with an attendant conception of what the
sets are like gives us:
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Risk Assessment Provide a degree of confidence in theories commensurate

with their consistency strength.

In particular, suppose that you come up with a wild new theory T (either
philosophical or mathematical). If I can use some set theory S to produce a
model of T, then I know that I can be at least as confident in the consistency of
T as I am in S.
Risk Assessment is especially important, as many theories here are incon-
sistent. As many philosophers know, early set theory was subject to paradoxes

27 As experts will know, there are other theories we might pick. One only really needs a theory
of syntax to study consistency (and weak theories of arithmetic suffice for such a theory).
Another salient field here is proof theory and the study of proof-theoretic ordinals. In a way,
set theory provides more than what is required for examining Metamathematical Corral.
However, it is in the variety of models, and what one can build from them, where set theory
really shines. So it is perhaps better to say that set theory provides a piece of the puzzle for
Metamathematical Corral, rather than the whole picture. Thanks to Marcus Giaquinto and
Daniel Waxman for some further discussion here.
 Iterative Conceptions of Set 15

(e.g. Russell’s Paradox). However set theory can also yield inconsistency and
paradox when combined with other philosophical principles, such as when we
layer mereology on top of the sets (e.g. Uzquiano (2006)). I also want to point
out (in line with Philosophical Repository) that an enormous variety of set-
theoretic ideas can be extended to inconsistency. In particular when we push
ideas to their natural limit, they nearly always explode. Perhaps this constitutes
a kind of ‘paradox’ (maybe in a weak sense of the term). Some of these we’ll
see later, and some others I mention in a footnote for the reader who wants to
look further.28 One might think that this is a negative of the discipline – after
all, isn’t inconsistency an (if not the) unforgivable sin? I disagree. Inconsis-
tency can be informative. Set theory gives us the tools to locate and diagnose
these inconsistencies, helping us to elucidate our Limits of Formalisation and
further giving us a:

Testing Ground for Paradox Set theory is very paradox prone, both in terms
of the principles that can be formulated within set theory and when combined
with certain philosophical ideas (e.g. absolute generality and mereology). In
this way, set theory provides a testing ground for seeing when and how ideas
are inconsistent.

So, there’s some interesting and nice features of set theory – not just a The-
ory of Collections, but a field that provides a Foundation for Mathematics
and Philosophical Repository, in particular by yielding a Generous Arena,
Shared Standard, Theory of Infinity, the example of Independence and its
use as a Testing Ground for Paradox, that help articulate the Limits of For-
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

malisation, give us a Metamathematical Corral, and Risk Assessment for

our theories. Before we move on, I want to identify one last important aspect of
set theory. Although many of the preceding constraints are simply reasons to
be interested in set theory, or are things that set theory has happened to be use-
ful for, there is a sense in which set theory was designed to fit these purposes.
Risk Assessment, for example, can’t go ahead without set theorists deliberately
studying Independence and Metamathematical Corral. In this way, many of
the preceding features – notably Generous Arena, Shared Standard, Theory
of Infinity, Metamathematical Corral, and Risk Assessment – are not just
pleasant features of set theory, but constraints/desiderata on its development

28 For example, the embedding template j : V → M for large cardinals explodes when M = V.
Forcing axioms can pop in various ways, either by admitting too many parameters, allowing
too many kinds of forcing, or not keeping a tight enough control on the sentences allowed (see
Bagaria (2005)). Standard reflection principles blow up at the level of third-order reflection (see
Reinhardt (1974) and Koellner (2009)) and modal reflection principles are pretty flammable
too (see Roberts (2019)).
 16 The Philosophy of Mathematics

too. Indeed this is one of the central points of Maddy (2017) and Maddy (2019),
(though she leaves Theory of Infinity implicit). Thinking about these virtues
in this dual light will help to illuminate some of the issues later, and in particu-
lar whether different conceptions/theories of sets are virtuous. We’ll see some
more virtues in due course, and for ease they are collated in Section 9.3.

3 The Naive Conception of Set and the Classic Paradoxes

We’ve now got some virtues of set theory on the table (Section 2). In this section
I want to explain one role for conceptions of set (namely to motivate theories)
and revisit some well-known material on the naive conception of set and the
‘classic’ set-theoretic paradoxes. In doing so, I’ll present a way of looking at
the paradoxes in terms of functions.

3.1 Conceptions of Set and Motivating Theories

One way into the problem of the paradoxes is by considering the following:

Question What do we want out of a conception of set?

At least in this partly philosophical and partly mathematical context, what

we want out of a conception is a satisfying motivation for a good theory, and
this is what I’ll take the primary purpose of a conception of set to be in this
Element. Let’s now clarify these notions a little.
Regarding the theory motivated: What we really want is a theory that can
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

be made suitably precise. For this, I’ll assume that we want to motivate an
axiomatic theory.29 I’ll presuppose that the reader has some understanding of
formal axiomatic theories (later we’ll use a little bit of first-order predicate
logic, plural logic, modal logic, and set theory). Where possible, I’ll provide
informal paraphrases and reference away the formal details.
Regarding the notions of a good theory and a satisfying motivation: We’ve
seen some constraints on a good theory in Section 2. For example, a good theory
should provide a Generous Arena and enable Metamathematical Corral. As
we proceed, we’ll discuss the virtues from Section 2 in more detail with respect
to specific proposals for the theories we adopt. Motivations, on the other hand,
might take the form of a formalisation (as we’ll see later with various modal
set theories), but equally they could be something more informal. In particular,
we’ll talk of conceptions of set. These can be thought of as informal descrip-
tions of what the sets are like, which might then be formalised in various ways.

29 Here I am following some of the remarks in ch. 1 of Incurvati (2020).

 Iterative Conceptions of Set 17

There’s a lot to say about the nature of conceptions, but we’ll avoid getting
into these tricky issues here (though we will mention some open questions in
Section 10).30
Whether these motivations on the basis of conceptions are satisfying also
presents a rather tricky cluster of problems. But some progress can be made
again by thinking about the goals from Section 2 and examining how the the-
ories proposed enable set theory to fulfil its usual roles. But we’ll also be able
to isolate some further desiderata on conceptions of set as we proceed, aside
from Section 2’s virtues pertaining to set theory more broadly. This latter target
we’ll accomplish in Sections 4 and 5.

3.2 The Naive Conception of Set

Our first conception will be the naive conception of set:

Definition 11 (Informal) The naive conception of set holds that sets are exten-
sions of predicates, where the extension of a predicate is the collection of all
the things to which the predicate applies.31

We now want to consider what axioms the naive conception motivates. For
this, it will be helpful to set up an important language for us:

Definition 12 The language of set theory or L ∈ is the first-order language with

one non-logical binary predicate “∈” and well-formed formulas formed in the
obvious way. (Note: We include identity as part of first-order logic throughout
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

this Element.)

The naive conception clearly motivates adoption of the extensionality axiom

(which says that any two sets with the same members are equal) as it is a
conception of set. Unfortunately, it also motivates:

Definition 13 The Naive Comprehension Schema asserts that for every one-
place formula ϕ(x) in the language of set theory L ∈ , there is a set of all and
only the sets satisfying ϕ(x). Formally:
( )
(∃y)(∀z) z ∈ y ↔ ϕ(z)

Sadly, as we know, the Naive Comprehension Schema is inconsistent. Let’s

see how.

30 See Incurvati (2020, p. 13) for discussion.

31 This formulation is taken directly from Incurvati (2020, p. 24).
 18 The Philosophy of Mathematics

3.3 The Paradoxes

Why go over the paradoxes, when excellent introductions are available in a
wide variety of texts?32 Aren’t we just rehashing old material? Here’s why
we’ll look at them:

(1) Part of what we will see later is a ‘new’ kind of paradox (the Cohen–Scott
Paradox) and we’ll discuss how it’s similar to the classic paradoxes. So
getting them on the table early is a good idea.
(2) There has been a shift of focus in the philosophical literature towards
viewing the paradoxes as concerned with the (non-)existence of particular
functions. Aside from the fact that these presentations are independently
interesting, this way of viewing the paradoxes will help us see the afore-
mentioned similarities a little better.

In this Element, I’ll only really consider Russell’s Paradox and Cantor’s Para-
dox. The Burali–Forti Paradox is also interesting; however, it is complicated by
the fact that one has to use set-theoretic codes for the ordinals (which otherwise
could be thought of as sui generis mathematical objects).33 Here they are:

Russell’s Paradox Consider the condition x < x. By Naive Comprehen-

sion, this determines a set r. We ask: “Is r ∈ r?” If yes, then r < r (since
r is in the set of all x < x), contradiction. So, instead assume r < r.
Then r satisfies the condition x < x, and so r ∈ r, contradiction. But
then r ∈ r ↔ r < r, a contradiction!
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Cantor’s Paradox Consider the condition x = x. Let {x|x = x} be denoted

by u (for “universal set”). Now consider P(u), namely the powerset of u.
By Naive Comprehension, this is also a set. Now we show x = P(u) by
noting: (i) every element of P(u) is an element of u (trivially), and (ii) if
x ∈ u, then x ∈ P(u) (since if x ∈ u, then ∀y ∈ x, y ∈ u (i.e. x ⊆ u) and so
x ∈ P(u)). So, u = P(u).
Clearly then, there is a surjectiona f : u ↠ P(u). Now consider the set
c = {x|x < f(x)}. Since f is surjective, there is a y ∈ u such that f(y) = c.
We now ask “Is y ∈ c?” If yes (i.e. y ∈ c), then y ∈ f(y), but then y violates
c’s defining condition, and so y < c, contradiction. So then we assume

32 See, for example, Giaquinto (2002), Potter (2004), and Incurvati (2020), for philosophical
introductions to the paradoxes, but almost any introductory text on set theory will cover them.
33 For some discussion of these issues, see Menzel (1986), Shapiro and Wright (2006), Menzel
(2014), and Florio and Leach-Krouse (2017).
 Iterative Conceptions of Set 19

y < c. But then y < f(y), and so y meets c’s defining condition, and y ∈ c,
contradiction. So y ∈ c ↔ y < c, a contradiction!
In fact, this proof can be transformed into a proof of Cantor’s theorem,
just by replacing u by any old set x and performing a reductio on the claim
that there is a surjection f : x ↠ P(x).
a As a reminder: A surjection f : x ↠ y is a function such that for every y1 ∈ y, there is an
x1 ∈ x such that f(x1 ) = y1 (i.e. every member of y gets hit by f applied to some element
of x).

So far, so well-known. Many introductory textbooks contain a presentation

of the paradoxes. However, something philosophers have paid more attention
to recently (though it has been known for a long time) is that these paradoxes
are closely related:34

The Cantor–Russell Paradox Define u and P(u) as in Cantor’s Paradox.

Consider the case where our surjection f : u ↠ P(u) is the identity map
f (x) = x. Now the problematic set c = {y|y < f(y)} = {y|y < y} = r. We’ll
also refer to this a the Cantor–Russell reasoning.

The important thing to note is that in this context (where f is the identity map)
the contradictory set r we get out is the problematic set for both the Cantor and
Russell reasoning (since f is the identity map here, the set {y|y < f(y)} just is
{y|y < y}). So the two are not just superficially similar, but in many contexts
come down to definition of exactly the same set, and the core issue is whether
there’s a surjection f : u ↠ P(u).
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

This observation works in the other direction too, where we assume that
we have an injection35 f : P(u) ↣ u. Without loss of generality, again this
can be the identity map (since P(u) = u). Now we can just consider the set
{y|y < f −1 (y)} (this is well defined since f is an injection).
Cantor’s Paradox and Russell’s Paradox might still not be exactly the same
(Cantor’s Paradox uses a bit more machinery than Russell’s, e.g. injections),
but there are clearly strong similarities between the two. I’ll remain neutral on
whether they are really ‘the same’ in any deep sense. Important for later will
just be:

(1) We can view each paradox as starting by postulating the existence of a

particular kind of function (either a surjection or an injection).

34 See, in particular, Bell (2014), Whittle (2015), Meadows (2015), Whittle (2018), Incurvati
(2020), Scambler (2021), and Builes and Wilson (2022).
35 Another reminder: An injection f : x ↣ y is a function such that for x1 , x2 ∈ x, if f(x1 ) = f(x2 ),
then x1 = x2 (i.e. f doesn’t take any two distinct elements of x to the same thing in y).
 20 The Philosophy of Mathematics

(2) We can then identify sets x and y such that x ∈ y ↔ x < y (in the case of
Cantor–Russell, x and y are both r).36

3.4 Universality and Indefinite Extensibility

So, Naive Comprehension leads to contradiction. But why, and what options
are we left with? Many have been considered throughout the literature, surveys
are available in Giaquinto (2002), Priest (2002), and Incurvati (2020). We’ll
follow Incurvati’s presentation here, since it will be instructive for making
comparisons.
Let’s start by noting that the Naive Comprehension Schema encodes the
following principle about the concept of set:

Universality A concept/conception C is universal if and only if there exists a

set of all the things falling under C.37

Universality clearly follows from the naive conception, since the condition
x = x is a perfectly legitimate predicate of set theory and the naive concep-
tion immediately licences the Naive Comprehension Schema. However, the
following is also a consequence:

Indefinite extensibility A concept/conception C is indefinitely extensible if

and only if whenever we succeed in defining a set u of objects falling under C,
there is an operation which, given u, produces an object falling under C but not
belonging to u.38
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Indefinite extensibility also follows from the Naive Comprehension

Schema. This is because any time we have a set x, the Naive Comprehension
Schema gives us the juice required for the Cantor–Russell reasoning, and we
can then diagonalise to find a set not in x (e.g. one of the members of P(x)).39
Clearly, any conception that validates both Universality and Indefinite
Extensibility will be inconsistent, since there both must and can’t be a set of all
objects falling under the conception. So in order to proceed, a natural way to go
is to examine conceptions of set that drop one of these fundamental principles.

36 Of course, strictly speaking, anything follows from the contradiction in classical logic. The
point is just that a natural way of reasoning to the contradiction is to note the contradictory
membership conditions.
37 This is adapted from Incurvati (2020), p. 27.
38 Again, adapted from Incurvati (2020), p. 27.
39 This way of looking at things has clear affinities with Priest’s (2002) characterisation of the
Inclosure Schema and Domain Principle. Since we’re concentrating on set theory here, and
Priest’s framework is more general, I’ve chosen to go the Incurvati route.
 Iterative Conceptions of Set 21

And this is just what iterative set theories do. First though, we’ll look at a more
coarse-grained distinction between the logical and combinatorial conceptions
of set.

4 The Logical and Combinatorial Conceptions of Set

We found ourselves in a tricky situation at the turn of the twentieth century.
The burgeoning field of set theory was clearly useful, but the naive conception
of set was deeply flawed. How to respond to this state of affairs?
In this section I want to make a preliminary distinction between the logical
and combinatorial conceptions of set. This distinction will be pretty rough-and-
ready, but it will help to elucidate the strategy for the rest of the Element when
we come to discuss iterative conceptions. Before we get going, I want to lay
down some further desiderata on conceptions of set, to complement the broader
goals of set theory presented in Section 2. These will help us in comparing
different conceptions moving forward.

4.1 Further Desiderata on Conceptions of Set

Earlier, we remarked that we want a conception of set to motivate a good theory
of sets, and do so in a satisfying way. It’s now time to elucidate a little more
what we’d like out of these notions.
For starters, it’s desirable for a conception to have the following feature:

Naturalness Provide a reasonably natural account of what the sets are like,
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

one which avoids ad hoc restrictions.

For example, if I tweak the naive conception of set to say that wherever an
instance would lead to inconsistency it should be rejected, I have made a purely
ad hoc restriction that is not clearly motivated by the underlying conception.
According to the desideratum of Naturalness, we should avoid these kinds
of move – the needed restrictions should flow naturally from the underlying
informal idea provided.40
We want more than merely the underlying picture to be natural though; we
should want it to motivate a good theory of sets (where ‘good’, as discussed in
Section 2, is likely to need some spelling out). So, we identify:

40 Though see Goldstein (2012) for a view that tries to advocate this position. The view has
roots at least far back as Quine, and something like it may even have been held by Zermelo
(see Maddy (1988a)). There are also mathematical difficulties in actually carrying out this
project due to the existence of mutually incompatible maximally consistent sets of instances
of naive comprehension. See Incurvati and Murzi (2017).
 22 The Philosophy of Mathematics

Interpretation A conception should motivate a good theory of sets.

Given the background of classical logic, inconsistent theories of sets are triv-
ial (everything follows by the principle of explosion). In motivating a good
theory, it’s thus a clear constraint that the resulting theory be consistent. How-
ever, it’s one thing to block an inconsistency, and another to diagnose it. Sam
Roberts (MSb) has recently identified the following challenge:

Some conditions, like the condition of being non-self-membered, fail to

determine sets. Nevertheless, set theory tells us that many conditions do
determine sets. The axiom of pairing, for example, says that the condition
of being a or b determines a set whenever a and b are sets. We are thus faced
with a challenge: to provide an account of the dividing line between the con-
ditions that determine sets and those that don’t which explains why there are
many of the sets there are – enough for the purposes of set theory – but not
problematic sets like the Russell set. (p. 2)41

Roberts refers to this problem of identifying which conditions do/do not

determine sets and explaining why as the explanatory challenge. On this basis
we can identify the following desideratum on a conception of set:

Paradox Diagnosis Respond to the explanatory challenge: Explain why the

paradoxical collections aren’t sets and which conditions do (and do not)
determine sets.42
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

To sum up: We want to find a conception of set of that provides a picture of

the sets exhibiting Naturalness, is rich enough to provide an Interpretation
for a good theory of sets (allowing us also to fulfil the goals of Section 2),
and provides us with a Paradox Diagnosis by responding to the explanatory
challenge and telling us which conditions are problematic, and why they don’t
determine sets. In a moment (Section 5.3) we’ll see a further desideratum –
what I’ll call Capture – that applies specifically to iterative conceptions of set.
Before we get there though, I want to consider a couple of more coarse-grained
approaches. Examining them will help us see better the ways in which iterative
conceptions can make progress.

41 See also Roberts (2016), pp. 9–11.

42 Closely related is the challenge of making a metaphysical distinction between sets and proper
classes (e.g. by identifying different ontological kinds). See Maddy (1983) and Barton (2017),
ch. IV.
 Iterative Conceptions of Set 23

4.2 The Logical and Combinatorial Conceptions of Set

We can begin by considering the the distinction between:

Definition 14 (Informal) The logical conception of set holds that sets corre-
spond to well-defined predicates.

and:

Definition 15 (Informal) The combinatorial conception of set holds that sets

correspond to acceptable pluralities (possibly without there being any non-
trivial defining predicate).43

Clearly, these conceptions are pretty rough-and-ready, certainly they are not
fully precise. As we’ll see shortly, each admits of multiple different sharpen-
ings. Moreover, we might not think that the distinction between the two is
sharp – perhaps there are some conceptions of set that borrow a little from
each.44 We’ll see one such conception – the constructibilist conception – in
just a moment, and there may well be others.45 Still, I think they’re useful to
think about as they highlight the following:

(1) When faced with mutually inconsistent principles about the sets, we can
very naturally move forward by developing a conception that rejects (at
least) one of them.
(2) Conceptions are often imprecise and/or underspecified, and further devel-
opment of the conceptions may be needed for progress.
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Let’s examine these two points in more detail.

4.3 Responding to Paradox and Sharpening Conceptions

The consideration of the distinction between the logical and combinatorial con-
ceptions of set indicates the following germ of a response to the paradoxes.
Many ways of making the logical conception precise will hold that the predi-
cate x = x is ‘well-defined’. Thus, under the logical conception, Universality
is likely to be validated and Indefinite Extensibility violated. Conversely,

43 Incurvati (2020), p. 31, talks about the combinatorial conception as holding that sets are char-
acterised via ‘reference to their members’. I wish to avoid awkward metasemantic issues
surrounding reference, and so I’ve used pluralities instead.
44 See Incurvati (2020) (especially section 1.8) for further discussion and references.
45 For example, there are versions of predicativism that seem to combine a notion of ‘good’
definitions with successive set formation from available pluralities. See Linnebo and Shapiro
(2023) for such a view.
 24 The Philosophy of Mathematics

versions of the combinatorial conception will make the notion of ‘acceptable

plurality’ precise in a variety of ways, and in doing so can make it the case that
the plurality of all sets is not an ‘acceptable plurality’. So it is Universality,
and not Indefinite Extensibility, that is often identified as the culprit. We thus
seem to have the beginnings of a Paradox Diagnosis; explanations of what the
sets are like might be used to explain why it is that one of the two principles fail.
So, when faced with paradox, an attractive move is to modify our conception
of set in response, attacking one of the two conflicting principles.46
Moreover, these two conceptions correspond to reasonably Natural concep-
tions. Returning to the example of my collection of beer coasters: let’s suppose
I want to think about the set of all objects in this collection. I may talk about
the relevant set of beer coasters as the set of everything that is both a beer
coaster and currently in a certain shoebox in my bedroom (thereby using the
logical conception). But there is also a certain plurality of objects – the indi-
viduals b1, ..., bn considered in the plural – each of which just happens to be
a beer coaster in that particular shoebox. And it seems that there’s no obsta-
cle to me considering the set of them (thereby thinking of the relevant set
combinatorially).
Though useful both for beginning to sharpen our conception of set and pro-
viding a preliminary classification, the logical and combinatorial conceptions
are still rather imprecise in a number of ways, and this leads to them being
defective (without further sharpening). For example, let’s consider Interpre-
tation. What formal theory is motivated by either the logical or combinatorial
conception alone? There seems to be little that one can say when thinking about
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

either as stated earlier. We first need explanations of what well-defined pred-

icates are, or what it is to be an acceptable plurality. And without a formal
theory we can’t achieve lots of the nice goals from Section 2.
This brings us on to our second point: When faced with some defective-
ness in our conception of set, it’s attractive to develop the conception in order
to make progress. This could be in the face of paradox, but we also might be
galvanised to do so by given other deficiencies (e.g. a failure to address Inter-
pretation). For example, in order to improve the logical conception, we need
to say what ‘well-defined’ means. There are a number of ways of doing this.

46 It should be noted that this diagnosis isn’t completely neat. One might make the case that there
are logical conceptions of set that violate Universality and combinatorial conceptions that
violate Indefinite Extensibility. Incurvati (2020), for example, characterises the limitation of
size conception as logical, but that doesn’t allow for all the self-identical sets forming a set. It’s
not clear to me that the limitation of size conception is in fact logical, but in any case we can
view those (more precise) conceptions that validate Universality as the relevant contrast cases
for what we’re doing here. We’ll see throughout this Element that the classifications amongst
the conceptions might not be sharp.
 Iterative Conceptions of Set 25

One (the stratified conception) holds that there are certain formulas that are
appropriately stratified, and that comprehension should be restricted to these
formulas.47 Another (the iterative property conception) holds that there is a way
of iteratively individuating those formulas that can be used in comprehension
(this is the approach of the property theories of Fine (2005), Linnebo (2006),
and Roberts (MSa)). On each of these, as it happens, the predicate x = x is well-
defined and individuates an extension, validating acceptance of Universality
over Indefinite Extensibility.
The combinatorial conception, by contrast, needs to make precise what it is
for a plurality to be acceptable. One way is to say that some sets are accept-
able if and only if they can be depicted as part of a particular kind of graph
(the graph conception). This conception also conforms to our earlier diagno-
sis, validating Indefinite Extensibility but rejecting Universality.48 Iterative
conceptions (the foci of this Element) also refute Universality whilst accept-
ing Indefinite Extensibility. They hold that a plurality is acceptable if it can
be formed from other sets using set-construction methods. Let’s now turn to
these conceptions.

5 Iterative Conceptions: First Examples

In this section, I want to present the emergence of several conceptions of set and
the eventual rise of what I’ll call the ‘strong iterative conception’. We’ll see that
this idea can be formalised modally and there’s a close affinity with ZFC. This
conception also performs wonderfully with respect to the desiderata we’ve con-
sidered. But I also want to indicate that the strong iterative conception, though
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

it may be the default, isn’t the only conception on the market.

47 The stratified conception is proposed by Quine (1937) and its history is nicely outlined in
Incurvati (2020). One starts with the following definition:

Definition 16 A formula ϕ in the language of set theory is stratified if and only if there is an
assignment of natural numbers to variables such that:

(i) For any subformula of ϕ the form x = y, the natural number assigned to x is the same as
the number assigned to y.
(ii) For any subformula of ϕ the form x ∈ y, the natural number assigned to y is one greater
than the number assigned to x.

By restricting comprehension to stratified formulas we obtain a system known as NF.

48 Since we won’t discuss this much, we’ll set it aside, but see Incurvati (2020), ch. 7 for details.
The relevant notion is an accessible pointed graph, a kind of directed graph where there’s
a distinguished top node (this is the ‘pointed’ part of the definition, and you can think of this
‘point’ as the set we want to code), the edges code the membership relations, with accessibility
meaning that it’s possible to reach each node of the graph by some finite chain of edges starting
from the point.
 26 The Philosophy of Mathematics

5.1 Some Iterative Conceptions of Set

We’ll consider a kind of combinatorial conception known as the iterative con-
ception. We’ll keep things rough and imprecise to begin with (this imprecision
will be helpful later when we separate out different versions of it):

Definition 17 (Informal) The iterative conception of set holds that sets are
formed in stages, and new sets are formed from old by collecting together
sets formed at previous stages. There are no other sets than those found at the
stages.

The rough idea can be filled out as follows. We (or better yet a suitably
idealised being) start at an initial stage with some initially given collection of
objects. These could be a bunch of non-sets (often called Urelemente), or some
antecedently given sets that we take to be acceptable (e.g. the empty set).49 We
then begin forming new sets out of what we have using some given operations,
and in this way obtain the sets. So long as our operations guarantee that new
sets can always be formed, we have an explanation of why Indefinite Exten-
sibility holds and Universality fails – there will never be a stage at which we
can use an operation to collect all the sets into a set.
The iterative conception of set as I’ve given it can in fact be split into two
conceptions, a strong one and a weak one:50

Definition 18 (Informal) The strong iterative conception of set holds that sets
are obtained in stages. At each additional stage we form every possible subplu-
rality of the current stage as a set. There are no other sets beyond those obtained
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

this way.

Definition 19 (Informal) The weak iterative conception of set holds that sets
are formed in stages. Sets are formed by collecting together sets existing at
previous stages using some set-construction methods. We leave it open whether
or not we get every possible subplurality of what we have at a stage as a set
immediately after the current one. There are no other sets beyond those obtained
this way.

I want to suggest that the weak iterative conception is really prior to the
strong iterative conception (conceptually, if not chronologically). Key to the
weak iterative conception are:

49 Depending on what set-construction methods we allow, we have to be careful that we don’t

start with a proper class. If we do, some modification is needed; see, for example, Menzel
(1986) and Menzel (2014).
50 This distinction emerged in discussion with Chris Scambler, and I’m grateful to him for the
suggestion of separating out the two.
 Iterative Conceptions of Set 27

(i) A description of what counts as a starting domain.

(ii) A description of some construction methods for forming new sets from
old.
The strong iterative conception says (i) can be any set of objects, but the
empty set will do, and (ii) that the operations that form new sets consist solely of
powerset (i.e. taking all possible subsets). (Note: As we’ll see later, we also take
unions at limits. No new sets are formed in the limit, however, it just consists
in ‘bundling together’ everything we constructed previously.) It thus sharpens
the weak iterative conception; there are other methods of set-construction that
we might have chosen. Let’s see an example of the difference by going into
more detail on each.
The strong iterative conception is perhaps the simplest version of the weak
iterative conception, so we’ll explore it first. It is also perhaps the ‘default’ ver-
sion – as of this writing, if you put the terms ‘iterative conception of set’ into
a search engine, you’ll get back results about the strong iterative conception.
Whilst I won’t enter into historical details too much here, an excellent descrip-
tion of its emergence is available in Kanamori (2007), and Button (2021a)
impressively charts its formal (stage-theoretic) development.
Often the idea of the strong iterative conception is formalised within ZFC
with the following definition using ordinal numbers:

Definition 20 The Cumulative Hierarchy of Sets or V is defined as follows:51

(i) V0 = ∅
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

(ii) Vα+1 = P(Vα ), where α + 1 is a successor ordinal.

∪
(iii) Vλ = α<λ Vλ (if λ is a limit ordinal)

The structure of the Vα thus captures the idea that we take all possible sub-
sets at each additional stage (i.e. iterate powerset) and collect them together at
limits (i.e. take a union). The often given visual representation is provided in
Figure 1.
The weak iterative conception is in some ways less well studied than the
strong iterative conception, possibly partly because the latter is seen as the
default. However, since the weak iterative conception is more general and will
be important later, it will be worth getting it on the table.

51 For simplicity, I am giving the version for pure sets; if you want to include Urelemente,
then clause (i) should be replaced with V0 = {x| ‘x is an Urelement’}, and clause (ii) by
Vα+1 = P(Vα ) ∪ Vα . The situation can get tricky depending on what Urelmente one allows,
see Menzel (1986), Menzel (2014), and Button (2021a) for discussion.
 28 The Philosophy of Mathematics

Figure 1 A visual representation of the cumulative hierarchy. Each additional

stage is obtained by applying the powerset operation, and at limit stages we
union together the previous stages. The figure contains only a tiny fraction of
the beginnings of the hierarchy; as we go further and further, each ordinal α
indexes some Vα , and ω + ω is a comparatively small countable ordinal.

We’ll see a few examples of the weak iterative conception in this Element,
but some will have to wait until we have a couple of set-theoretic constructions
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

under our belt. For now, here’s an easier example to get a feel for it. Suppose
we want to build the hereditarily finite sets (i.e. finite sets that are built up out
of only finite sets all the way down – formally we say that the empty set is
hereditarily finite, and any other set is hereditarily finite just in case it is finite
and all its members are hereditarily finite). In standard set theory, we can get
these sets just by taking powersets from the empty set (i.e. moving up through
each Vn for every natural number n). But there are other ways we might build
these sets. Suppose we individuate sets in stages by starting with the empty set
at stage 0 and forming at stage n + 1 all sets of size at most n. As we continue
through all the stages up to ω (the first infinite stage), we’ll eventually get every
hereditarily finite set. But we won’t get every possible subset at a successor
stage. For instance, you can check that stage 4 has eight members, so you’ll
miss out some subsets of stage 4 when moving to stage 5 (you’ll have to wait
until stage 8 before you can form all subsets of stage 4). So, this procedure is
weakly but not strongly iterative – there are possible sets that don’t get formed
at the next stage.
 Iterative Conceptions of Set 29

We can also have processes that are not even linearly ordered, for instance
by having two or more set forming operations. For example, let the operation
Even! form the subsets of a stage with an even number of elements. The other
Odd! forms the odd numbered subsets of a given stage. By interleaving Even!
and Odd! finitely many times we can get any hereditarily finite set. But the pro-
cess is not linearly ordered; for instance, we could choose to do Even! a bunch
of times in a row. One doesn’t even have a guarantee that you get every hered-
itarily finite set using these processes (say if you just head off only iterating
Even! over and over again).
There are more mathematically interesting kinds of weak iterative concep-
tion. Here’s a more difficult (but important) example:

Definition 21 (Informal) The constructibilist conception holds that sets are

formed in stages. At subsequent stages we form into sets those pluralities of
previous stages that are definable (i.e. can be picked out by a formula) over
that stage. There are no other sets beyond those obtained this way.

Is conception weakly or strongly iterative? We can show that there are

versions of it that are only weakly iterative.

Often set theorists will talk about the constructible universe (or L) and
constructible hierarchy. L is formed by taking definable powersets. A sub-
set x of the domain of a structure M is definable over M if and only if x is
the unique set containing all and only the y in the domain of M satisfying
ϕ(y) (in M) for some condition ϕ(y) in the language of M.a For a structure
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

M, let’s call the collection of all such M-definable subsets Def(M). Then
L can be defined as:

Definition 22 The constructible hierarchy (or just L) is defined as

follows:

(i) L0 = ∅
(ii) Lα+1 = Def(Lα ) for successor ordinal α + 1.
∪
(iii) Lλ = α<λ Lα for limit ordinal λ.

The axiom that every set is constructible (i.e. ‘For every x there is an α
such that x ∈ Lα ’) is called the Axiom of Constructibility or V = L.

Now the constructible hierarchy clearly satisfies the weak iterative con-
ception and the constructibilist conception. But it doesn’t satisfy the strong
 30 The Philosophy of Mathematics

iterative conception. This is because often new subsets of previous levels

get formed as we climb. For example, new subsets of ω coding new real
numbers get formed as we move up through the first few stages above
Lω . To see this, note that a satisfaction predicate for Lα is definable over
Lα+1 , and above Vω these will code new subsets of natural numbers.
This phenomenon (the slow growth of L) is quite general. Since there
are only as many formulas as there are parameters available (the usual
formula-building operations are trivial at infinite cardinals) we have that
the cardinality of Lα is the same as the cardinality of Lα+1 for every α
(in stark contrast to the Vα -hierarchy where Vα+1 is always bigger than
Vα ). So, the Lα hierarchy does not satisfy the strong iterative conception;
there are possible subsets that don’t get picked up when we move to a
subsequent stage. I represent this visually in Figure 2.
Moreover, we could make the iteration more fine-grained and non-
linearly ordered. I could take each formula to provide its own set-forming
operation, and think of successively forming subsets for specific formulas,
instead of taking the whole definable powerset. This would still qualify
as weakly iterative.
Note: Sometimes you can recover a version of the strong iterative
conception from the weak one. In the case of our n-sized-set-forming
operation, we could eventually recover the Vn -hierarchy if we wait long
enough. This holds for the Lα -hierarchy too; for example, if L satisfies
ZFC, it can recover its own version of the Vα -hierarchy.
However, one can still see the difference between the two hierarchies,
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

even when we assume that V is equal to L. For, even if ZFC holds, it is

not the case that Vα = Lα for every α. Rather the Lα -hierarchy grows
slower than the Vα -hierarchy, it is just that the Lα -hierarchy can ‘catch
up’ at limit stages.
a This is fiddly to formulate. See chapter 3, section 5 of Drake (1974).

Remark 23 An important but somewhat orthogonal side remark: I mentioned

earlier that the combinatorial and logical conceptions of set were pretty rough-
and-ready. I think that the constructibilist conception provides a good example
to showcase how the two might be neither neatly separable nor incompatible.
On the one hand, the constructibilist conception is clearly combinatorial, we
have an explanation of when a plurality is available; namely when it is defin-
able after iterating the formation of definable powersets. But notice also that
under the constructibilist conception the existence of sets is intimately tied
to defining conditions – given the formation of some set x, it must comprise
 Iterative Conceptions of Set 31

Figure 2 The constructible universe sitting inside the cumulative hierarchy.

In the case where every set is constructible, we still do not have Vα = Lα for
every α; rather, the Lα -hierarchy can catch up at limit stages before
continuing to grow slower than the Vα -hierarchy.

exactly the satisfiers of some condition from a previous stage. So it’s very
unclear whether it’s correct to characterise the constructibilist conception as
either combinatorial or logical; it seems to borrow ideas from each.

To sum up: There are multiple sharpenings of the weak iterative concep-
tion. One is the strong iterative conception. But others (e.g. the constructibilist
conception) are mathematically interesting and not strongly iterative. Clearly,
some of these conceptions can be used to motivate certain axioms (for example,
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

the constructibilist conception motivates the axiom that every set is con-
structible). But can we lend formal precision to this idea of motivating theories?
Let’s see how this can be done with the strong iterative conception.

5.2 Modal Set Theory and the Strong Iterative

Conception
We want our conceptions of set to motivate virtuous theories. Later (Section 8)
we’ll see how versions of the weak iterative conception other than the con-
structibilist conception can be used to do just this. For now, we’ll focus on the
‘default’ strong iterative conception and ZFC. In particular, I’ll:

(1) Explain ZFC set theory.

(2) Show how ZFC can be motivated on the basis of a modal axiomatisation
of the strong iterative conception.

So, let’s start by setting up ZFC:

 32 The Philosophy of Mathematics

Definition 24 Zermelo–Fraenkel Set Theory with the Axiom of Choice (ZFC) is

formulated in the language of set theory L ∈ . It comprises the following axioms
(we just give informal statements; formal definitions are available in many set
theory textbooks):

(i) Axiom of Extensionality. Sets with the same members are identical.
(ii) Axiom of Pairing. For any two sets x and y there is a set containing
just x and y.
(iii) Axiom of Union. For any set x, there is a set of all members of members
of x.
(iv) Powerset Axiom. For any set x, there is a set of all subsets of x.
(v) Axiom of Foundation. Every set contains an element that is disjoint
from it. The axiom both rules out self-membered sets and also the
existence of infinite descending membership chains.
(vi) Axiom of Infinity. There’s a non-empty set x such that for any member
y of x there is another member z of x such that y is a member of z. (This
guarantees that there’s an infinite set.)
(vii) Axiom of Choice. (AC) For any non-empty set of pairwise-disjoint
non-empty sets, there is a set that picks one member from each. (Note:
ZFC without AC is just denoted by “ZF”. In the presence of ZF this
is equivalent to the Well-Ordering Principle that every set can be
well-ordered.)
(viii) Axiom Scheme of Replacement. If a formula ϕ(x, y) is function-like
(i.e. for any x, there is exactly one y such that ϕ(x, y)), then the image
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

of any particular set under ϕ(x, y) is also a set.

The following are provable ZFC, normally using Replacement. They
will be relevant later when we consider dropping the Powerset Axiom
(in that context, they can’t be proved from Replacement) and so we’ll
include them as part of ZFC:
(ix) Axiom Scheme of Separation. If ϕ(x) is a formula in one free variable
x, then if y is a set, then there’s a set of all the x in y such that ϕ(x) (i.e.
{z|z ∈ y ∧ ϕ(z)} exists).
(x) Axiom Scheme of Collection. For any formula ϕ(x, y) in two free vari-
ables, if ϕ(x, y) defines a relation, and for some set a and for every
x ∈ a there is always a y ϕ-related to x, then there is a set z ‘col-
lecting’ together at least one ‘ϕ-witness’ for every x ∈ a. Since this
axiom scheme may be a little less familiar, we’ll include its formal
statement:
( )
(∀a) (∀x ∈ a)(∃y)ϕ(x, y) → (∃b)(∀x ∈ a)(∃y ∈ b)ϕ(x, y)
 Iterative Conceptions of Set 33

As noted earlier (Section 2) ZFC is a very nice theory of sets with many
theoretical virtues. But can it be motivated using the iterative conception?
There are different ways to do this. One way is to axiomatise the notion of
a stage directly.52 Whilst this meshes very well with the ‘stage-theoretic’ ter-
minology employed earlier, I want to take a slightly different approach here.
Instead, we’ll think of the iterative conception as describing how new sets can
be formed from old using set-construction methods, and in this way giving us
a kind of modal framework. There are a few reasons for this choice that I’ll
just briefly mention. First, the modal approach makes some questions regard-
ing Interpretation a little more tractable using existing technology (we’ll see
how shortly). Second, thinking of things modally provides an easy integration
with other areas of philosophy. For example, if we think that sets are speci-
fied as part of a modal framework, then the nature of this framework and the
modalities employed may well be of interest to the modal metaphysician.53
Third, there is an easy way to view these modal theories in stage-theoretic
terms – we can simply think of the corresponding worlds as the stages and our
chosen set-constuction methods as providing the accessibility relation. Fourth,
it’s the modal approach to the weak iterative conception that has been better
developed in the literature thus far (indeed, developing properly stage-theoretic
accounts of the weak iterative conception will be left as an open question in
Section 10).
This said, this choice is controversial.54 We might think that the modal
approach deviates somewhat from the stage-theoretic account of the iterative
conception, suggesting different kinds of philosophical question. Moreover, as
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

we’ll explore later (Section 10), there’s some substantial open questions about
how well the modal approach formalises either the weak or strong iterative
conceptions.55 For now I’ll put these questions to one side, but I don’t want to
overstate my position. Iterative conceptions are informal accounts of the nature
of the sets. There are choices to be made in how to formalise these ideas. Some
may be better than others, and there is significant philosophical work to be done
in contrasting the different approaches.
The idea to view the iterative conception modally is relatively old, going
back to Parsons (1983), but has been fruitfully applied recently. In particular,
Linnebo (2013) shows how one can give modal axioms motivating ZFC. Giving

52 See Button (2021a) for a recent article on the state of the art.
53 A friendly introduction to these issues is available in Menzel (2021).
54 I am grateful to Davide Sutto and Chris Scambler for some discussion of this point.
55 For example, as we’ll see, some modal approaches yield non-well-founded accessibility
relations. But let’s defer a more thorough examination of this issue until later.
 34 The Philosophy of Mathematics

the full details would take up too much space, but a flavour of the approach will
be useful.56
As is clear from the way I’ve presented the iterative conception informally,
we’ll want to talk about reifying pluralities into sets, and for this Linnebo uses
a plural logic. Really though, any extensional second-order variables would
do. Since much of the literature (e.g. Scambler (2021)) follows this convention
of using plurals, we’ll stick with it. Again we’ll leave the plural logic rela-
tively informal; the reader wishing to see a concise presentation of the details
is directed to Linnebo (2014) or Oliver and Smiley (2013) for textbook treat-
ments. Plural logic has new variables xx that range over ‘some things’ (e.g. the
books on my table), a binary relation symbol ≺ (where x ≺ xx is to be read as “x
is one of the xx”), with the expected definition of well-formed formula. We’ll
denote the language obtained by adding these resources to L ∈ by “L ∈, ≺ ”. We’ll
routinely abuse singularisation and speak of “a plurality” (a standard move in
this field).
For our plural axioms (here we’re mostly following the presentation in
Scambler (2021)) we’ll take the following:

Definition 25 Plural logic (over set theory) has as axioms (we’ll give these
axioms informally, see Linnebo (2018), ch. 12 for the formal details):

(i) A principle of extensionality for plurals (that if two pluralities xx and

yy comprise the same things, then anything that holds of the xx also
holds of the yy and vice versa).57
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

(ii) Additionally, impredicative plural logic has the following Impredica-

tive Comprehension Scheme:
( )
∃xx∀y y ≺ xx ↔ ϕ(y)

for any ϕ in L ∈, ≺ not containing xx free.

(iii) Predicative plural logic does not contain the Impredicative Compre-
hension Scheme but rather has the following Predicative Compre-
hension Scheme:
( )
∃xx∀y y ≺ xx ↔ ϕ(y)

for any ϕ in L ∈, ≺ not containing xx free and not containing any plural
quantifiers.

56 Details can be found in Linnebo (2013) and Scambler (2021), and different modal approaches
are given by Studd (2013) and Button (2021b).
57 I’m suppressing some subtleties here about how one formulates the extensionality axiom; see
Roberts (2022) for details.
 Iterative Conceptions of Set 35

We then need a background modal logic to talk about what’s possible after
constructing new sets. For this we’ll add a modal operator ^ to L ∈, ≺ to get a
language L ∈,^ ≺ , with well-formed formulas as normal. We’ll also use the modal
operator □, and in this context □ϕ can be treated as shorthand for ¬^¬ϕ. For
modal axioms we’ll use:

Definition 26 Classical S4 is the modal logic with the operator ^ and the
axioms:

(i) The necessity of identity and distinctness (these are sometimes

optional, but we’ll include them):
• x = y → □(x = y)
• x , y → □(x , y)
(ii) K: □(ϕ → ψ) → (□ϕ → □ψ) (this holds for any normal modal logic).
(iii) T: ϕ → ^ϕ (this holds if the accessibility relation is reflexive).
(iv) 4: ^^ϕ → ^ϕ (holds if the accessibility relation is transitive).
To obtain S4.2 we add:
(v) G (sometimes called .2): ^□ϕ → □^ϕ (holds if the accessibility
relation is directed).
The logic S4.3 is obtained by adding:
( )
(vi) .3: (^ϕ ∧ ^ψ) → ^ (^ϕ ∧ ψ) ∨ (ϕ ∧ ^ψ) (holds if the accessibility
relation is linear).
Throughout, we will also assume:
(vii) The Converse Barcan Formula (CBF): ∃x^ϕ → ^∃xϕ (this can be
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

thought of as capturing the idea that domains only grow).

Because we have S4.2 you can think of the space of worlds as a kind of
branching time structure, but where you can always bring together any two
possibilities (this is the content of the G/.2 axiom). Thus □ϕ can be thought
of as saying “in all future worlds ϕ” and ^ϕ as “there is a future world such
that ϕ”. And S4.3 makes it so that there’s a kind of inevitability to how the
possibilities unfold.58
Before we give our modal axioms, we should clarify how we’ll interpret
non-modal set theory. Mostly mathematicians will just want to work with a non-
modal axiomatisation of sets, without paying attention to finicky modal details
about how the sets are formed. So we can ask: Is there a way of interpreting non-
modal set theories in L ∈ into our modal language L ∈,^ ≺ ? Given the iterative

58 It should be noted that the converses of (v) and (vii) do not hold (i.e. there are frames that
are not directed that satisfy G/.2, and non-linear frames satisfying .3).
 36 The Philosophy of Mathematics

conception, how should we interpret the ‘usual’ quantifiers ∀ and ∃? Well, one
natural thought is that ∀xϕ should hold if, no matter how you form sets, ϕ will
always hold, and ∃xϕ tells you that you can go on to form sets such that ϕ. We
can then provide:

Definition 27 Given a sentence ϕ in L ∈ , the potentialist translation of ϕ

(denoted “ϕ^ ”) is obtained by replacing every universal quantifier “∀” by “□∀”,
and every existential quantifier “∃” by “^∃”.

We can then define a version of the modal axioms that is extracted from
Linnebo (2013):59

Definition 28 Linnebo (2013), Linnebo (2018) (here we follow Scambler’s

(2021) presentation) Lin is the following theory in L ∈,^ ≺ :

(i)Classical first-order predicate logic.

(ii)Impredicative plural logic.
(iii)Classical S4.2 with the Converse Barcan Formula added.60
(iv) The Axiom of Foundation (rendered as normal using solely resources
from L ∈ ).61
(v) Extensionality (again using solely resources from L ∈ ).
(vi) Modal Collapse. The principle that any things (at a stage) could form
a set:
□∀xx^∃y□∀x(z ∈ y ↔ z ≺ xx)
(vii) Stability axioms for ≺ and ∈ (these mirror the necessity of iden-
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

tity/distinctness):62
• x ∈ y → □(x ∈ y)

59 I’m basically following the presentation in Scambler (2021), with a few extra tweaks that will
be useful later. Scambler uses “L” to denote Lin, so I’ve opted for syntax that avoids possible
confusion of Lin with the constructible hierarchy L. Strictly speaking, Linnebo (2013) doesn’t
include the plural version of the Axiom of Choice (he is looking for interpretation with ZF)
but Scambler (2021) does (but he throws it in as part of the plural logic; I include it as a hybrid
plural-cum-set-theoretic axiom). With these systems you get as much Choice out as you’re
willing to throw in, and since we’re primarily interested in ZFC in this Element, I’m happy to
throw it in.
60 Normally the Converse Barcan Formula comes for free (one must take steps to block it),
see Linnebo (2018), p. 207. I’ve added it for the sake of explicitness and making the ‘growing
domains’ conception of potentialism clear. I’ll make no further mention of this complication.
61 Another nice option here is to use ∈-induction. Thanks to Øystein Linnebo for some
discussion of this point.
62 It should be noted that a choice point here concerns whether to use the quantified or free-
variable forms of these axioms, since the free-variable versions seem stronger. Thanks to Chris
Scambler for some discussion of this point.
 Iterative Conceptions of Set 37

• x < y → □(x < y)

• x ≺ yy → □(x ≺ yy)
• x ⊀ yy → □(x ⊀ yy)
(viii) Two principles of plural definiteness:
• Plural Membership Definiteness is given by the following
scheme:
(∀x ≺ yy)□ϕ(x) → □(∀x ≺ yy)ϕ(x)

• Subplurality Definiteness: Say that xx ⪯ yy holds just in case

the xx are a subplurality of the yy, that is, for every x such that
x ≺ xx we have x ≺ yy. Then the Subplurality Definiteness
scheme states that:

(∀xx ⪯ yy)□ϕ(xx) → □(∀xx ⪯ yy)ϕ(xx).

(ix) Modal Infinity. The axiom that there could be some things compris-
ing all and only the possible natural numbers.
(x) Modal Powerclass. The axiom that there could be some things that
are all and only the possible subsets of a given set.
(xi) Modal Replacement. Every potentialist translation of the Replace-
ment Scheme of ZFC.
(xii) Plural Choice. A plural version of the Axiom of Choice ‘For any
pairwise-disjoint non-empty sets xx, there are some things yy that
comprise exactly one element from each member of the xx’.63
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Together, these get us some way to providing a modal axiomatisation of the

strong iterative conception. Let’s discuss the Naturalness of these axioms. Of
course the extent to which a modal set theory is natural and/or not ad hoc will
likely be somewhat imprecise and a matter of degree. Given any set x, there
could be (by Modal Powerclass) a plurality of all possible subsets of x. Using
Modal Collapse, this plurality can then be reified into a set. Thus there’s a clear
picture of how the sets are formed – our set-construction method turns plural-
ities of the domain into sets. I’ll refer to such methods as ‘Reify! commands’,
and talk about using Reify! to turn pluralities into sets.
From this picture, we also get a Paradox Diagnosis. It explains why Indef-
inite Extensibility holds and Universality fails – the universal set never gets
formed because at no stage is there a plurality of all possible sets; we can

63 Strictly speaking, Linnebo does not include AC, but I’m happy to throw it in. Some other
authors (e.g. Studd (2013)) do so. Nothing hangs on it for the results we have here, other than
the fact that if Lin is run without a form of AC, the that gets interpreted will also not include
AC (it will be ZF rather than ZFC).
 38 The Philosophy of Mathematics

always form something new. In particular, the Russell plurality of all non-
self-membered sets at a world will be formed (via Reify!) as a set at a later
world.
Unfortunately, we don’t quite yet get the full strong iterative conception.
Although the modal set theory Lin axiomatises an uncountabilist conception of
set (in that it implies that there could be a set that is necessarily uncountable),
the axiomatisation does not exactly correspond to the formation of the uni-
verse via powersets and union. The problem lies in the fact that though Modal
Powerclass and Modal Collapse entail that the full powerset of a set could
exist, there is no guarantee that it is formed immediately. It might take some
time and there might be some intermediate worlds before the relevant powerset
appears.64
Though we won’t go into too much detail on this, here’s how you can enforce
the strong iterative conception. Whilst I am not aware of any axiom in L ∈,^ ≺
that guarantees that the powerset gets formed immediately, one can use bimodal
operators to enforce the immediate collapse of every possible plurality of a
world into a set. James Studd provides one such axiomatisation (in Studd,
2013), with subsequent examination and a further approach by Tim Button (in
Button (2021b); I’ll largely follow Button). Here we have two modal operators
and that capture the ‘timelike’ feature of forming sets; ϕ is intended
to mean that it was the case that ϕ (or “previously ϕ” if you like), and ϕ is
intended to mean that it will be the case that ϕ (or “eventually ϕ”), with the
usual dual necessity operators. Tweaks to the modal logic are needed to incor-
porate these resources (in particular Button and Studd both work in a free logic)
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

but we won’t dwell on that here. The important point is that one can use these
modal operators to ensure that every plurality is formed into a set at subsequent
worlds via an axiom asserting that if every possible ϕ existed earlier, then there
is a set comprising exactly the ϕs. In the context of plural logic (or a suitable
second-order logic) we immediately get the result that every plurality xx forms
a set at all worlds after each of its members exists.65
For the sake of simplicity (and in particular avoiding the complications of
free logic) I’ll stick with Lin, but we could run the same points using Button
or Studd’s systems too. Moreover, the Vα can be used to obtain a Kripke frame

64 In fact one can (with a little work) obtain a Kripke model for Lin from the Lα of a model
of ZFC, and, as noted before, the Lα don’t get you the strong iterative conception. See Barton
(MSb) for discussion and a sketch.
65 Formally: If E(x) is the relevant existence predicate in free logic, one can formalise the
claim as follows (we use the formulation in second-order logic rather than plural logic, as in
Button (2021b)):
(∀F)(∀x : F) E(x) → (∃a)(∀x)(F(x) ↔ x ∈ a)
 Iterative Conceptions of Set 39

for Lin (as we’ll discuss shortly), so Lin is satisfied under the strong iterative
conception (even if there are other conceptions that satisfy Lin).

5.3 Mirroring and Capture

Let’s now look at Interpretation. The core concept will be the idea of Mir-
roring Theorems. These tell you how you can go between the modal theories
and the non-modal theories favoured by mathematicians. In particular we can
show:

Theorem 29 Linnebo (2010), Linnebo (2013) ZFC proves ϕ if and only if Lin
proves ϕ^ .

This theorem shows that the modal idea of reifying all pluralities into sets at
a stage (and continuing this into the transfinite) motivates ZFC concerning the
sets. And as we noted in Section 2, ZFC is a very nice set theory! Moreover, it
shows how Lin is strongly faithful to ‘normal’ set theory under the potentialist
translation.66
Interestingly, the relationship goes back the other way too. Earlier, we men-
tioned that the strong iterative conception suggests that the universe is formed
via the Vα -hierarchy. But one can also show:

Theorem 30 Linnebo (2013) Over a model of ZFC, the Vα under ⊆ provide a

model for Lin.67
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

This shows that not only does Lin motivate ZFC, but if you accept ZFC then
you can also get a model for the modal set-construction methods axiomatised
by Lin. This shows that although Lin doesn’t perfectly capture the notion of the
strong iterative conception, it certainly fits very well with it.
One final piece of the puzzle ties everything together:

Theorem 31 (ZF) For every set x there is an ordinal α such that x ∈ Vα .

66 Indeed, this result can be strengthened to apply to very weak theories, Tim Button (in Button
(2021a) and Button (2021b)) has shown that one can go back and forth between tremendously
weak (i) theories of sets, (ii) theories of stages, and (iii) modal stage theories (the theories in
question do not even imply there are any sets!).
67 Specifically a Kripke frame validating S4.3. In order to interpret the plural logic, strictly
speaking we should think of worlds as pairs of the form (Vα , P(Vα )), with ≺ just interpreted
by ∈.
 40 The Philosophy of Mathematics

This theorem shows you that not only does ZFC allow you to define the Vα ,
but you can prove that every set is contained therein. I think that this high-
lights the following desideratum on modal axiomatisations of conceptions of
set:

Capture Let T^ be a modal theory of sets interpreting a non-modal theory T

via the potentialist translation. Then we say that T^ satisfies Capture if and
only if, given a model M of T, we have a general way of extracting a Kripke
frame KM T^
|= T^ from M such that for every x ∈ M there is a world W ∈ KM T^
such that x ∈ W.68

The statement of Capture is a little complicated. But the core point is the
following: Not only does our modal theory (in this case Lin) motivate a nice
non-modal theory of sets (namely ZFC), but our non-modal theory (ZFC) also
‘thinks’ that there is a model of the modal theory (Lin) such that every set lives
in said model. Perhaps it would be too strong to say that they are sides of the
same coin, but they certainly pair very well.69

5.4 Preliminary Conclusions

We’re now in a position where (i) the strong iterative conception is Natural
(giving us a clear picture of what the sets are like), (ii) it gives us a Paradox
Diagnosis (Reify! can always produce new sets), (iii) when axiomatised by Lin,
it Interprets a good theory of sets (namely ZFC), and (iv) we have a Capture-
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

theorem; ZFC allows you to find a Kripke model for Lin and ZFC proves that
every set lives in said model. For these reasons, I think it’s fair to say that
the strong iterative conception is a very satisfying conception of set. Modal
set theory, suitably formulated, pushes the idea that ZFC should be true of the
sets, and if ZFC is adopted, we can show that a sensible modal theory is a
mathematical fact of life – if you have ZFC you also have the strong iterative
conception and all the sets live within some stage/world so described.
Whilst I don’t want to deny how good a conception of sets the strong iter-
ative conception of set is, I do want to press the point that it might not be the

68 Of course, as I’ve set things up here, the model in question (gotten from all the Vα ) is
proper-class-sized, so not a ‘model’ in the ordinary sense of the term. I handle this in Barton
(MSb) by restricting to arbitrary set-sized models, but let’s suppress these metamathematical
details here.
69 See Button (2021b) for some examination of the extent to which there are definable for-
mal relationships between modal theories of the strong iterative conception and non-modal set
theory.
 Iterative Conceptions of Set 41

only option out there. Later, we’ll see some examples of the weak iterative con-
ception of set that I think are also in the running. In order to see their appeal
though, we’ll have to learn a little about how one can use forcing to build more
sets.

6 Forcing as a Construction Method

In this section I want to outline forcing in set theory; a way of adding subsets to
models. Thorough presentations are available in a wide variety of mathematical
texts and full detail would just bog down the reader in an Element like this one,
so my focus is on giving the informal ideas.
There’s two main reasons to go into depth on this topic. First, we’ll use forc-
ing to articulate the further versions of the weak iterative conception that we’ll
consider later. Second, forcing is tremendously important for understanding
much of the contemporary literature on the philosophy of set theory and the
intuitions that underlie much work in this field. So, having a good grasp of it is
no bad thing.

6.1 Forcing: The Rough Idea

A helpful way to understand forcing is by analogy with field extensions.
Consider the relationship between the fields of real numbers R and complex
numbers C. One way of thinking of obtaining C from R is via the idea of alge-
√
braic closure. Intuitively speaking, we throw in solutions for −1, and then by
closing under the field operations, obtain C.
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Forcing is very similar. In fact, according to Paul Cohen (a father of the tech-
nique), this analogy was part of his discovery.70 To see this analogy, let’s start
by considering the problem forcing was developed to solve. In particular, we
were trying to prove that the continuum hypothesis is independent from ZFC.
Since we knew that given a model M of ZFC, CH is true in the constructible
universe of M (a fact proved in Gödel (1940)) one way to proceed was to find a
way of making a model of ¬CH from one satisfying CH. (One could then infer
by the Completeness Theorem that neither CH nor its negation followed from
ZFC, assuming ZFC to be consistent.) Since we also knew that (again proved in
Gödel (1940)) L was the smallest inner model (i.e. transitive model containing
all ordinals) under inclusion, the natural idea was to break CH by adding sets –
much like we could find a root for −1 by moving from R to C. And this is just
what Cohen did in Cohen (1963).

70 See Cohen (1963), p. 113, and Cohen (2002), pp. 1091, 1093. Thanks to Carolin Antos for
some discussion of the history here.
 42 The Philosophy of Mathematics

In order to figure out what we need to break CH, it’s helpful to think about
what CH and ¬CH say about sets of reals and functions. CH, recall, says that
every set of reals (i.e. something with cardinality no bigger than 2ℵ0 ) is either
countable or the same size as 2ℵ0 . In this way, it says that there are lots of kinds
of function compared with the kinds of sets of reals – every infinite set of reals
has a function that either bijects it with ℵ0 (the cardinality of N) or 2ℵ0 (the
cardinality of R). By contrast, ¬CH says that there are lots of kinds of sets of
reals as compared with kinds of function – there’s some infinite set of reals x
for which there’s no bijection between x and ℵ0 , but also no bijection between
x and 2ℵ0 .
Let’s suppose then that we’re given a model M of ZFC + CH. What could we
do to break CH? Well, we need to (i) add some set x to M, whilst (ii) making
sure that we preserve the axioms of ZFC when we add x, and (iii) having a set of
reals y in the new model such that there’s no bijection between either y and the
new set of all reals or natural numbers. This is what Cohen showed was possible
with forcing. Assuming ZFC is consistent, there’s a model M satisfying ZFC
(by Completeness). Either (i) M satisfies ¬CH (in which case we’re done) or
(ii) M satisfies CH. If (ii), we can then add a bunch of reals G to M, and close
under definable operations to form an extension M[G] satisfying ZFC. In this
new model M[G], you can show that the old set of reals from M is a set of reals
that is neither bijected with ℵ0 nor 2ℵ0 in M[G].
If you haven’t encountered forcing much before, I want the reader to now
stop and pause to think about how, given the rough idea of forcing, we might be
able to take a model of ¬CH and make CH true again by adding sets. What kind
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

of set could we add to a model of ¬CH in order to restore CH again (and what
would we have to simultaneously avoid adding)? (Bear in mind that you can’t
add natural numbers by forcing – a student once made the ingenious suggestion
to me that we bump up the size of ℵ0 . Alas, this doesn’t work since forcing
keeps models transitive, and the natural numbers are isomorphic in all transitive
models of set theory.)
The answer is that we need to add functions that provide the relevant bijec-
tions between the old sets of reals and either ℵ0 or 2ℵ0 , and do so (i) without
adding reals, and whilst (ii) preserving ZFC. Again, Cohen showed that forc-
ing lets you do this. Given an M satisfying ¬CH, one can collapse the cardinals
between ℵ0 and 2ℵ0 to ℵ0 by adding a set H that allows you to get surjections
from the natural numbers to these cardinals. In the new model M [H], CH is
true, since there are now bijections between ℵ0 and the old ‘cardinals’ between
2ℵ0 and ℵ0 (i.e. things that were cardinals between ℵ0 and 2ℵ0 in the ground
model).
 Iterative Conceptions of Set 43

These two kinds of forcing are sufficient to show the following:

Theorem 32 Given a model M of ZFC, so long as we can do forcing over M,

then M has:

(1) An extension M[G] such that M[G] satisfies ¬CH. This can be done using
forcing that collapses no cardinals – it does not add new bijections that
make any set look smaller than before.
(2) An extension M[H] such that M[H] satisfies CH. This can be done by forcing
whilst adding no new reals – we don’t add any new subsets of the natural
numbers.

In this sense CH is like a set-theoretic light switch as regards forcing – we can

flip it on and off at will by successively forcing to add new sets, and all whilst
preserving ZFC.71 Indeed, forcing is incredibly flexible. An example that will
be important for us is the following:

Theorem 33 Assume that we can always force over M. Then for any set x in
M, there is a forcing extension M[G] in which x is countable.

As discussed earlier, the idea for proving this theorem is just to add a
surjection from ℵ0 to x.
Forcing thus provides us with a very controlled way of adding subsets to
models. We’ll discuss this a little in a -section later on (Section 6.2), but
it will be helpful to indicate the shape of what is to come. Forcing, I want to
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

contend, can be thought of as a method of set construction for adding subsets

to a universe and in particular might be a way of generating sets under the
weak iterative conception. Using this idea, we’ll end up with the motivation of
a concept of set on which every set is countable, since given a set x at some
stage/world, we could always add a surjection from the natural numbers to x
by forcing.

6.2 A Little More Depth on Forcing

In this section I add a little more mathematical detail and provide an intu-
itive characterisation of forcing. This whole section is a -section, so the
reader shouldn’t get bogged down in the details unless they really want to.
Still, the section will help inform the idea that we can think of forcing as a

71 This terminology of ‘switches’ is from Hamkins and Loewe (2008).

 44 The Philosophy of Mathematics

way of constructing new sets from old, so I recommend at least giving it a

go. Good introductions to this material can be found in Kunen (1980) and its
update Kunen (2013) (a wonderful pair of books explaining a range of issues
in detail), Drake and Singh (1996) (a nice concise introduction), and Weaver
(2014) (a much easier-going introduction before the applications starting in ch.
14). Many set theory texts contain an introduction, however, and the reader
should feel free to shop around.
We’ll start with an example that will help us follow what comes later a bit
better. We’ll take the idea of adding a Cohen real. Let’s suppose that you’re
in a model of ZFC. For now, we’ll assume that the model is countable (and
transitive) and so (by Cantor’s theorem for the reals) misses out a whole bunch
of real numbers. For our purposes, you can think of a real number as an infinite
ω-length sequence of 0s and 1s (this, in turn, can be thought of as a function
from the natural numbers into {0, 1}, which says whether there’s a 0 or a 1 in the
nth place). I want to now add in a new real number, and do so in such a way that
ZFC is satisfied. So I slowly go through deciding on what I want in the nth place
of my new real for each n (perhaps not in order). I need to do two things: (i)
make sure I’m avoiding the reals of M (i.e. I don’t get something I already have),
and (ii) make sure that when I’m done I close under new definable operations
to ensure ZFC is true. This is what forcing lets you do. Such an object (a new
ω-length sequence of 0s and 1s) is our new real number (our ‘Cohen real’).
Let’s now take a little peek into the tricky machinery of how we do this. The
way I suggest thinking of forcing is as a way of talking about descriptions of
collections that can change their members as we make certain decisions. In the
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

end, if we make decisions in exactly the right way, we’ll end up defining a new
object that isn’t currently in the universe we start in, and fill in all the needed
sets to make ZFC true. The rough ingredients of forcing are the following (i) a
partial order P = (P, <P ) with certain nice properties that make it sufficiently
‘interesting’. You can think of P as the space of possible ‘decisions’ that we
might take. (ii) P-names: these are descriptions of collections that can change
their membership depending on what decisions we take from P, (iii) dense sets:
these are like advisors; no matter what decisions you’ve taken, they’ll always
recommend at least one more you might go on to take, and (iv) a generic filter:
this you can think of as a complete description of all the decisions that were
taken in the limit, consistent with every recommendation given by an advisor.
Let’s look at these in more detail.72

72 Note: Often authors (e.g. Drake and Singh (1996) and Weaver (2014)) write in information-
theoretic terms, P is a space of information, and we slowly get more and more fine-grained
information as we move through P. The way I’m expressing things is essentially equivalent,
 Iterative Conceptions of Set 45

First, we need the notion of a forcing partial order (P, ≤P ). Before we give
the definition, a couple of notes are in order:

• Note 1: We often refer to elements of the partial order as ‘conditions’.

• Note 2: Here the partial order grows ‘downwards’ – the intuition being that
if p <P q, you’ve got a smaller range of possible decisions after p as com-
pared to q. Some people write p >P q to indicate the same state of affairs,
the intuition being that you get more information from p as compared
to q.73

We now define:

Definition 34 A forcing partial order P = (P, ≤P ) is a partial order P such

that:

(i) P has a maximal condition 1P

(ii) P is atomless – any element of p of P has incompatible extensions (i.e.
there’s q ≤P p and r ≤P p such that there’s no s with s ≤P q and
s ≤P r).

The way I’m going to suggest one thinks about this partial order is as an infor-
mation space of possible decisions for settling membership facts. As we’ll see,
we can define a class of ‘names’ for possible sets (these are called P-names).
These we can think of as having their membership facts settled as we take
decisions through P. The conditions of being atomless one can think of as a
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

condition on P being sufficiently interesting or non-trivial – there are always

incompatible decisions one could make about where to go, and there’s no part
of P that admits of ‘inevitability’.
In the specific case of adding a Cohen real, we can define the following
partial order:

Definition 35 Given some model M, the forcing partial order to add a Cohen
real has as its domain (in M) all finite partial functions from ω into {0, 1} and
p ∈ P extends q (i.e. p ≤P q) if and only if p extends q as a function (i.e.
q’s domain is a proper subset of p’s, and they agree on all arguments from q’s
domain).

but a bit easier to think about philosophically, and brings the ‘variable set’ way of thinking to
the fore a little more.
73 See Drake and Singh (1996), p. 155, Warning 8.8.2 for discussion.
 46 The Philosophy of Mathematics

This order gives us a way of thinking of settling the nth place of a new real –
as we move down through P we settle more and more values for a new real to
be defined. In the limit, we’ll have settled every bit of the real.
How to get a handle on this idea of ‘settling values’? For this we’ll need the
definition of a P-name. The definition looks somewhat complicated, but it can
be given an intuitive backing.

Definition 36 A P-name is a relation τ such that ∀⟨σ, p⟩ ∈ τ(‘σ is a P-name

and p ∈ P’).

The definition looks circular, but in fact is not since the empty set is trivially
a P-name. You can think of the P-names as relations where other P-names are
related to conditions in P.
The intuition to have in mind is that a P-name is the name for a possible set.
Given a bunch of good ‘decisions’ from P (we’ll talk about this idea of ‘a bunch
of good decisions’ in a second; the key notion is that of a generic filter), we’ll
evaluate the P-names to different sets in the extension. The way this works
is given a P-name σ, we’re going to rule in or out other (already evaluated)
P-names in the domain of σ according to whether or not they’re related to a
condition in our new object. So P-names are a kind of ‘variable collections’ –
they can change their mind as to what they contain as we move about in P.74
The next notion we need is:

Definition 37 We say that D ⊆ P is dense if and only if for every p ∈ P, there

https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

is a q ∈ D such that q ≤P p.

The way of thinking about a dense set D is that it’s kind of like an advisor.
No matter where you are in P, and what decisions you’ve taken, D can come
up with at least one decision you could take to continue.
Next we need the notion of a generic filter:

Definition 38 G ⊆ P is a filter on P if and only if:

(i) G is non-empty.
(ii) p ∈ G and q ≥P p implies that q ∈ G (i.e. G is closed upwards).
(iii) p ∈ G and q ∈ G implies that there is an r ≤P p, q with r ∈ G (i.e. G
brings any two elements together).

74 Interestingly the idea of ‘variable collection’ corresponds well to the category-theoretic

approach to forcing. See the Appendix to Bell (2011).
 Iterative Conceptions of Set 47

We furthermore say that G is M-P-generic (for some model M) if and only

if G intersects every dense set of P in M. (We’ll often just abbreviate this to
‘generic’ and let context determine the values of P and M.)

The way to think of such a G is that it is a kind of ‘maximal’ collection of

‘good decisions made’. If you include a decision p ∈ G, then you’ve got to
include any earlier decisions that could have led there, and you’ve also got to
be able to bring together any two decisions together later (there’s no including
incompatible decisions allowed). You’ve also got to be ‘good’ in that you agree
with every advisor (i.e. dense set) in at least one place. Part of what genericity
ensures is that you genuinely add something by avoiding sets already in your
starting model, but you also don’t encode any ‘extra’ information in what you
add.
We can then talk about what happens to a P-name when presented with a
generic G.

Definition 39 We evaluate P-names by letting the value of τ under G (written

‘val(τ, G)’ or ‘τG ’) be {val(σ, G)|∃p ∈ G(⟨σ, p⟩ ∈ τ)}.

Again, this looks complicated, but the intuition is as follows. Remember

that a P-name can be thought of as a kind of ‘variable collection’ or ‘name for
a possible set’. When we give some G to a P-name τ, we evaluate stepwise by
analysing the valuation of all the names in the domain of τ and then we add
them to τG according to whether they’re related to some p ∈ G. In particular, if
σ is a P-name in the domain of τ, then we put σG into τG if there is a ⟨σ, p⟩ ∈ τ
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

for which p ∈ G (and do nothing otherwise). So you can think of us running

through the p ∈ G and throwing in already evaluated P-names according to
whether a name is related to some p ∈ G.
Let’s return to our example of adding a Cohen real. Consider the following
conditions from the poset to add a Cohen real:

• f is defined by:
– f(0) = 1
– f(3) = 0
• g is defined by:
– g(0) = 0
– g(3) = 0
Now consider the following names:

• τ=∅
• σ = {⟨τ, f ⟩}
 48 The Philosophy of Mathematics

• µ = {⟨τ, f ⟩, ⟨σ, g⟩}

• ν = {⟨τ, f ⟩, ⟨τ, g⟩, ⟨σ, f ⟩, ⟨σ, g⟩, ⟨µ, f ⟩, ⟨µ, g⟩}

Let’s suppose that f ∈ G but g < G. So this says that the first bit of our new
real is 1, and the third bit is 0. What happens to our P-names under G? Well, τ is
trivial and so remains unchanged. We now have a value τG for τ, so the values
σ, µ, and ν will contain τG = ∅ (since we have ⟨τ, f ⟩ ∈ σ, µ, ν). The evaluation
of σ is now complete, and we know that σG = {∅}. For µ, since we know g < G,
we don’t throw in the evaluation of σ into µG , and so µG = {τG } = {∅}. For ν,
whilst we do have a bunch of P-names correlated with g (and so the evaluation
of those names doesn’t make it in via any ordered pair of the form ⟨ξ, g⟩), we
also have that ν contains ⟨τ, f ⟩, ⟨σ, f ⟩, and ⟨µ, f ⟩ and so the interpretation of
these names gets thrown in. So νG = {τG, σG, µG } = {∅, {∅}}.
Of course, things are much more complicated when we move to names with
more structure (in particular, once you have big infinite names, things are going
to get more subtle) and start to consider further partial functions in G. The fact
that f ∈ G told us only about the values of bits in the 0th and 3rd spots. But
there will be many more functions; since G has to hit every dense set, you can
think about G as filling in more and more spaces in our real to be defined as we
hit more and more dense sets, whilst preserving filter-hood. So we might also
have the partial function h ∈ G such that:

• h is defined as follows:
– h(0) = 1 (note that h has to agree with f, since we assumed f ∈ G)
– h(1) = 1
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

– h(2) = 1
– h(3) = 0 (again because h has to agree with f)
– For every n such that 3 < n < 9001, h(n) = 0.

This function tells us that our new real will have a 1 in the 0th place, a 1 in
the 1st place, a 1 in the 2nd place, and 0s all the way up to the 9,000th place
(it leaves open what happens over 9, 000). Correspondingly, the evaluation of
names will get quite complex as we have to evaluate across all the relevant P-
names, also bearing in mind that if h ∈ G, there must be a lot more conditions
also in G (e.g. since G is a filter, f is also going to have to be in there if h is,
since h extends f). Even if h ∈ G, there will always be a dense set D which
doesn’t contain any conditions p ≥P h, and so G will have to contain a function
settling more than h does, thereby fixing the values of more P-names, and so
on. Once we’ve hit every dense set whilst preserving filter-hood, we’ll have
our generic G settling every bit of a new real, and this can be used to evaluate
every P-name, giving us our new extension M[G].
 Iterative Conceptions of Set 49

These ideas are very difficult when one first encounters them, and should be
paired with a text that goes through forcing in full mathematical detail. How-
ever, I hope the rough idea is clear. We have a ‘space of possible decisions’ (the
partial order P), a bunch of names that can change their mind about what they
contain when presented with some ‘decisions’ from P (i.e. the P-names), and
a bunch of ‘advisors’ (the dense sets) each of which can always present to you
a way of continuing after some point in P. We’re then given a ‘maximal good
bunch of decisions’ (the generic G) that agrees with every dense set at some
point and lets you find your way through P by giving you conditions from P.
G tells each P-name who they are by ruling in evaluations of P-names based
on whether the names in the domain of a P-name are related to the decisions
in G.
Other partial orders that are especially important are:

Definition 40 Forcing to add κ-many Cohen reals.

• P is the collection of all finite partial functions (in M) from κ × ω to {0, 1}

• p ≤P q if and only if p extends q as a function.

A generic for this partial order doesn’t just add a Cohen real and then close
under definability, it adds κ-many. One can then show that you don’t destroy
any cardinals by adding a generic for P (this is a non-trivial lemma)75 . This
then lets you infer (picking big enough κ) that ¬CH holds in M[G]; even if
M satisfies CH, all the cardinals between ω and κ in M are now cardinalities
between ω and κ (which is now the cardinality of the continuum) of different
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

sets of reals in M[G].

As mentioned earlier, any cardinal can be collapsed to the countable using
forcing. This is done using:

Definition 41 The forcing to collapse κ to ω is defined by:

• P is the collection of all countable partial functions from ω into κ.

• p ≤P q if and only if p extends q as a function.

A generic for this partial order allows us to get a surjection from ω to κ, and
collapses the cardinality of κ (and any sets bijective with κ) to ω.

75 See, for example, Weaver (2014), p. 50, theorem 13.3.

 50 The Philosophy of Mathematics

These represent just a taste of some of the possibilities available using forc-
ing. As Joel-David Hamkins (2012) writes (about model-building methods
including forcing):
Set theorists build models to order. (p. 417)
So forcing is a flexible tool that gives us a way of adding sets to models.
There are two points we should note. First:

Fact 42 If P is a forcing partial order in a model M of ZFC, and G is P-M-

generic, then G < M. In particular, P − G = {p|p ∈ P ∧ p < G} is dense (and
clearly missed by G).

This fact will be a little important later when we relate ‘paradoxes’

related to forcing and the Cantor–Russell reasoning (I relegate a proof to a
footnote.76 )

Fact 43 Let M be a transitive model satisfying ZFC and let M[G] be the model
obtained by evaluating all the P-names for a forcing partial order P and M-P-
generic G. Then M[G] also satisfies ZFC, and in particular M[G] is the smallest
transitive model of ZFC containing both every element of M and G.77

The strategy for proving this is to ‘cook up’ P-names that you know (by the
genericity of G) will ensure that ZFC is satisfied. But the fact that you get the
smallest possible extension is important: It shows that you can think of the addi-
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

tion of a forcing generic G and evaluating the P-names as throwing in G and

closing under definable operations – namely you don’t get any ‘extra’ sets than
what is required to get ZFC by throwing in G so long as G is generic. For exam-
ple, one can show that you don’t add any ordinals when you force. In this way,
the P-names and evaluation procedure conspire to make sure the construction
of M[G] is very tightly controlled. This further reinforces the similarity between
forcing and more mathematically familiar constructions like obtaining the field
of complex numbers from the field of real numbers. There, we take R, throw
in i, and close under the usual field operations to get C. Indeed, C is the small-
est such field. So with forcing, M[G] is the smallest model of ZFC you get by
throwing in G and closing under every operation you can define.

76 Proof. Suppose p ∈ P. We must show that there is q ∈ P − G such that q ≤P p. The only non-
trivial case is where p ∈ G. Because P is non-atomic, there are incompatible r and s extending
p. But then one of r and s isn’t in G – all elements of G are compatible with one another.
77 See Kunen (2013), lemma IV.2.19.
 Iterative Conceptions of Set 51

Figure 3 Forcing: (1) We start with a forcing partial order P ∈ M. (2)

https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Assuming that we can meet every dense set of P in M, we can get a generic
G ⊆ P such that G < M. Finally (3) We add G to M, and then, by evaluating
the P-names, close under definable operations to obtain the forcing extension
M [G].

Moreover, there’s a sense in which finding such a G can be thought of as a

set-construction method in its own way. If we’re given a forcing partial order P
and a family of dense sets D (let’s let each dense set Di in D be indexed by some
i in an index set I), we can think of successively hitting each Di in such a way
that we extend our previous choices. In particular, if we start with some M and
D is the collection of all dense sets available in M, what we obtain in the limit
(hitting each Di ) will be an M-generic filter G. It’s a substantive assumption to
assume that this can always be done, but not one without intuitive pull. And if
we can always perform this action for any given P and D, then we can always
force. We can see a visualisation of this idea in Figure 3.
 52 The Philosophy of Mathematics

6.3 Philosophical Upshot

I hope that the reader finds the preceding discussion helpful, and in particular
it can serve as a road-map if you want to learn forcing in detail (alongside an
introductory text). However, it’s understandable if readers newer to set theory
didn’t pick up everything. I therefore want to quickly identify the following:

Main Philosophical Upshot You can think of moving from M to M[G]

by forcing as a way of constructing new sets.

Before we move on, we should note that this is a somewhat controversial

claim. Brauer (MS), for example, raises some queries about whether one can
really think of forcing as a way of generating new sets. I hope I’ve addressed
these worries here (I argue this case further in Barton (MSb)). For now, I’m
happy to take the Main Philosophical Upshot as an assumption from hereon
out, but we will raise some open questions about it in Section 10.

7 A ‘New’ Kind of Paradox?

In this chapter I want to argue that there’s a tension at the heart of set the-
ory. We’ll then (Section 8) explain how this can be resolved into different
conceptions, much as we saw with the naive conception of set.

7.1 The Forcing-Saturated Strong Iterative Conception of Set

A popular thought in set theory is that richness is an essential part of set theory;
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

there should be as many sets as possible.78 Given this thought, it’s natural to
want the universe to be closed under lots of different kinds of set-construction
method. Since we just saw that forcing is one such method, and we already
know that the strong iterative conception is a good conception of sets, the looks
attractive:

Definition 44 (Informal) The forcing-saturated strong iterative conception of

set holds that new sets are formed from old by either (i) forming each possible
plurality of sets over that stage into a set, or (ii) adding in a generic for a partial
order and a family of dense sets.

So, we have two main set-construction methods (plus union, which allows
us to bundle limit stages together). We can either add in a forcing generic, or
we can form the powerset of a set. Clearly then, we have:

78 See Incurvati (2017) for a survey of this idea in set theory.

 Iterative Conceptions of Set 53

Powerset The Powerset Axiom holds.

We’re going to shortly see some conflicts with Powerset. We therefore

define:

Definition 45 ZFC with the Powerset Axiom removed will be called ZFC− .

Note: Dropping the Powerset Axiom is a slightly subtle business.

When we formulate ZFC solely with Replacement (and treat Separation
and Collection as theorem schemas), then deleting the Powerset Axiom
results in a theory weaker than one would like (the version with Separa-
tion and Collection has more consequences than with Replacement alone).
Since we included Separation and Collection as separate axiom schemas
when formulating ZFC earlier, we avoid this complication. See Zarach
(1996) and Gitman et al. (2016) for discussion.

Since we can also introduce a generic for any forcing partial order and fam-
ily of dense sets under the forcing-saturated strong iterative conception, we’ll
introduce the following axiom:

Definition 46 (ZFC− ) The Forcing Saturation Axiom or FSA is the claim that
for any partial order P and set D consisting solely of dense sets for P, there is
a generic G intersecting every member of D.79
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

The forcing saturated strong iterative conception then motivates:

Forcing Saturation The Forcing Saturation Axiom holds.

Readers familiar with forcing may already see the problem with the forcing-
saturated strong iterative conception. For the reader that doesn’t, I suggest
briefly pausing to think about what Powerset entails (especially in light of Can-
tor’s theorem) and what follows from Forcing Saturation (especially given
collapse forcings).

7.2 The Cohen–Scott Paradox

Here’s the problem: The forcing-saturated strong iterative conception moti-
vates both Powerset and Forcing Saturation, but they’re inconsistent with

79 See Barton and Friedman (MS), Definition 9.

 54 The Philosophy of Mathematics

one another. This mirrors how the naive conception was brought down by Uni-
versality and Indefinite Extensibility. I’ll refer to the paradox I’ll give as the
Cohen–Scott Paradox as it originates with the mathematical work of Cohen,
and Scott was one of the first to propose the tension I’ll identify. The paradox
is thus not really that ‘new’, and the idea that there might be a tension between
having uncountable sets and always being able to force has been around since
at least the 1970s. However, recent work has developed the philosophy and
mathematics of these ideas substantially.80
Letting “Powerset” denote the Powerset Axiom, the current theory moti-
vated by the forcing-saturated strong iterative conception is ZFC− + Powerset
+ FSA. But we can now note that because we can produce a generic for any
forcing partial order and family of dense sets, for any set x we like we can use
collapse forcing to add a generic making x countable. In fact, we can note the
following:

Fact 47 (ZFC− ) The forcing saturation axiom is equivalent (modulo ZFC− ) to

the axiom “Every set is countable”.81

We can now present the Cohen–Scott Paradox:

The Cohen–Scott Paradox Simply put, ZFC− + Powerset + FSA implies

that there are uncountable sets (by Cantor’s theorem and the Powerset
Axiom) but also that every set is countable (by the Forcing Saturation
Axiom). Contradiction!
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Before we continue, I want to emphasise: No reasonable classical set theo-

rist has ever accepted both Forcing Saturation and Powerset in this generality.
Perhaps someone learning forcing might unwittingly fall into the trap of accept-
ing the forcing-saturated strong iterative conception, or perhaps it’s appealing
to theorists of a dialethic persuasion. But set theorists are a clever bunch, and
they are able to see this contradiction coming a mile off. In fact, this tension
has been noticed for a while. Discussing forcing in the introduction to Bell’s
book on the subject, Dana Scott (1977) writes:

I see that there are any number of contradictory set theories, all extending
the Zermelo–Fraenkel axioms: but the models are all just models of the first-
order axioms, and first-order logic is weak. I still feel that it ought to be

80 See, for example, Meadows (2015), Scambler (2021), Builes and Wilson (2022), and Barton
and Friedman (MS). Naming the problem “The Cohen–Scott Paradox” is taken from Barton
and Friedman (MS).
81 This is a well-known folklore result, but see Fact 10 of Barton and Friedman (MS) for details.
 Iterative Conceptions of Set 55

possible to have strong axioms, which would generate these types of models
as submodels of the universe, but where the universe can be thought of as
something absolute. Perhaps we would be pushed in the end to say that all
sets are countable (and that the continuum is not even a set) when at last all
cardinals are absolutely destroyed. (p. xv)

So the Cohen–Scott ‘Paradox’ is certainly not new, and was noticed from
the inception of forcing. One might then ask: If it’s so obviously bad, why even
consider the forcing-saturated strong iterative conception? The reason to do so
is not that individual agents hold it, but that it forces us to face a possible choice.
Much as we saw with the naive conception, there are different ways we could
go. We could adopt a version of the logical conception that validates Univer-
sality. Or we could adopt a version of the combinatorial iterative conception on
which Indefinite Extensibility holds. Similarly, we could now adopt Power-
set (for example by holding the strong iterative conception) or we could adopt a
conception that validates Forcing Saturation. We’ll explore this in more detail
shortly (in Section 8). For now I want to consider the relationship between the
Cohen–Scott Paradox and diagonalisation.

7.3 The Cohen–Scott Paradox and Diagonalisation

To see the link with ‘diagonal’ arguments, we start with the question:

Question. What (if any) is the link between the Cantor–Russell reasoning and
the Cohen–Scott Paradox?
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

We have already seen a tight link between Russell’s Paradox and Cantor’s
Paradox in Section 3 – in the case where we first take the universal set, then
consider the identity surjection/injection, and then run the standard proof of
Cantor’s theorem, we get the Russell set.
There is a superficial similarity here, in that the (un)countability of some set
x can be viewed as a claim about the (non-)existence of a surjection from ω to
x. But is there any deeper similarity?
As mentioned earlier, the assumption that every set is countable (i.e. for any
set x there is a surjection from ω to x) is equivalent (over ZFC− ) to the claim
that for any forcing partial order and any set-sized family of dense sets D, there
is a generic intersecting D (i.e. the Forcing Saturation Axiom).82 We can now
present the following ‘diagonal’ version of the Cohen–Scott Paradox.

82 ( ) See Barton and Friedman (MS), Fact 16.

 56 The Philosophy of Mathematics

The Cohen–Scott Paradox, Diagonal Version. If the Powerset Axiom

is true, then the family D∗ of all dense sets for a forcing partial order P
is a set-sized family. By the Forcing Saturation Axiom, there is a generic
G intersecting every member of D∗ . Now consider E = {p|p < G}. It’s an
exercise to show that E is dense, the interested reader can go back and find
the proof in Section 6. Since G is generic for D∗ and E is dense, we know
that G intersects E at some point p. But then we have p ∈ G ↔ p ∈ E
(by choice of p), but p ∈ E ↔ p < G (by the definition of E), and so
p ∈ G ↔ p < G (putting together the biconditionals), contradiction!a
a See here also Meadows (2015) for emphasis of this diagonal version of the Cohen–Scott
Paradox.

The point to note here is that there is a similarity to the Cantor–Russell rea-
soning. There we had the assumption of the existence of a particular surjection
leading to contradictory claims about (non-)self-membership. Here we have
the existence of an surjection, whilst not leading to contradictory claims about
self -membership, we do have the contradictory p ∈ G ↔ p < G. So whilst
the analogy is not perfect, we have a diagonal-style contradiction obtained
by assuming the existence of a particular surjection. We’ll discuss a possible
significance of this in Section 10.4.

7.4 Summing Up
To sum up, we’ve seen that:
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

(1) There is a tension between Forcing Saturation and Powerset.

(2) This can be put in terms of a diagonal argument, with similarities to the
Cantor–Russell reasoning.

So, what to do about this state of affairs?

8 Countabilist Conceptions of Iterative Set

We’ve identified a tension between Forcing Saturation and Powerset, in anal-
ogy with Universality and Indefinite Extensibility. And just as before, we can
move forward by dropping one of the two. One way is to just hold that Forcing
Saturation should be dropped and Powerset accepted. The result of doing so
is the strong iterative conception, and is perhaps the ‘default’ position. We can
then (as noted in Section 5) provide a modal theory of this conception using
Lin (perhaps with some resources added to guarantee that we really do get all
pluralities formed as sets at every subsequent world). This yields the extreme
form of uncountabilism that is standardly associated with set theory with an
 Iterative Conceptions of Set 57

unending hierarchy of uncountable cardinals. But might there be a way of going

forward with Forcing Saturation instead of Powerset? In this section we’ll
see some modal theories that validate Forcing Saturation. Later (Section 9)
we’ll discuss how these conceptions interpret mathematics, and compare the
two approaches in light of the theoretical virtues adumbrated in Sections 2, 4,
and 5.

8.1 Countabilist Stage Theories

As we’ve seen, if you’re going to have Forcing Saturation, then every set is
going to be countable. For the sake of brevity, it will be helpful to introduce
some terminology:

Definition 48 The countabilist axiom (or Count) is the axiom ‘Every set is
countable’.

Definition 49 (Informal) We will refer to the view that holds Count as count-
abilism (with countabilist the corresponding adjective), and uncountabilism as
the position that there are uncountable sets.

It’s fair to say that countabilist options for the (weak) iterative conception
have been a lot less studied than the ‘standard’ strong iterative conception, and
so we will have to proceed with a little more care in articulating the alterna-
tive. This way of viewing the sets is still somewhat nascent with much work
still to be done, and we will have to be cautious in our conclusions. Certainly
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

it is less solidified than the standard strong iterative conception, and I don’t
want to overstate my case. I do want to identify, however, that it’s an attractive
alternative. This section, then, will have the flavour of explaining a promising
road of inquiry, rather than the more established picture of the strong iterative
conception we saw in Section 5.
Since we have Count for the countabilist, we can’t have uncountable sets.
For this reason, we’re going to have to drop the Powerset Axiom and adopt
ZFC− + Count. Since we don’t have the Powerset Axiom (indeed we have its
negation) we don’t have the Vα , and so we’re going to have to adopt the weak
iterative conception, rather than the strong iterative conception. So the question
then becomes: Given that the Vα are out, what could our stages/worlds be, and
how are they constructed? Recall that for any weak iterative conception we
need:

(i) A description of what counts as a starting domain.

(ii) A description of some methods for forming new sets from old.
 58 The Philosophy of Mathematics

Can we come up with weakly iterative theories for the countabilist, and
thereby give a story along the lines of (i) and (ii)? And, in particular, is there
a modal theory of sets that can function for the countabilist much like Lin did
for the uncountabilist; providing a good modal theory of set construction that
upholds their view?

8.2 Reify! and Generify!

I want to argue that there are proposals in the literature that can be viewed as
providing Natural modal theories for countabilist versions of the weak iterative
conception.
Regarding (i): What might the set-construction be? Well, one possibility is
familiar – given some stage/world we want a notion of forming sets out of
the pluralities available at that stage (i.e. Reify!). This is what the Powerset
Axiom codifies – every possible class at some stage Vα is reified into a set
(if it didn’t already exist) at Vα+1 (this can be partially formalised modally by
Lin, and by using bimodal operators guaranteed). But note, we don’t have to
turn every possible class in a set at a subsequent stage. This is made clear by
the constructibilist conception and the constructible hierarchy, at Lα+1 we reify
those classes definable over Lα into sets. In this sense, taking the definable
powerset is a kind of Reify! set-construction method, but it’s not the maximal
such (that would be powerset). For example, in the Lα -hierarchy we’ll get the
‘universal plurality’ of the previous stage at the next one, since x = x is a
perfectly good formula. To be a Reify! command you only need take some
possible pluralities of the domain and reify them into sets; you don’t have to
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

take all possible pluralities.

However, as I hope I convinced the reader in Section 6, another kind of set-
construction method is forcing. We can thus think of having, in addition to
whatever Reify! commands we employ, a class of Generify! commands which
will take in a partial order P and family D of dense sets and spit out a generic for
P and D. Closely linked is the set-construction method I’ll call Enumerate!;
this adds an enumeration between a set and the natural numbers. There are a
class of Enumerate! commands that can be thought of as special cases of the
Generify! operation, in particular the specific case of the forcing that adds a
surjection from the natural numbers to a set.83 If we think that the stages should
support Generify!, then Enumerate! will always be executable. This idea has

83 There might be other Enumerate! commands. A set being generic entails that you don’t
add in ‘extra information’, and an arbitrary enumeration might add in much more. For example,
0♯ can be thought of as a particular kind of countable set (and hence an enumeration), but can’t
be added by standard forcing techniques. See Barton (MSb) for discussion.
 Iterative Conceptions of Set 59

been advocated recently by a few authors. For example, Chris Scambler (2021)
writes:
The guiding idea . . . is to introduce another way of extending a given uni-
verse of sets as an option at each stage of the process. Specifically, we
will imagine we are capable not only of introducing sets whose members
are among already given things..., but also of introducing new functions
between already given (infinite) sets, and in particular of introducing func-
tions defined on the natural numbers and whose range contains any set as a
subset. (p. 1088)

Jessica Wilson and David Builes (2022) express a similar idea (partly
drawing on Scambler (2021)):
Recall that any set-theoretic universe is ultimately generated by two sorts of
processes: the powerset operation and the length of the ordinals. Proponents
of height potentialism maintain that the length of the ordinals is indefinitely
extensible: necessarily, for any ordinals, there could always be more. The
modal approach to [Cantor’s theorem] simply extends this line of thought
to the powerset operation: necessarily, for any subsets of an infinite set,
there could always be more. This is width potentialism. For any set-theoretic
structure, there is both a taller one and a wider one. (p. 2212)

Recall how we could use Even! and Odd! to obtain the hereditarily finite
sets. Can we think of interleaving Reify! and Generify! to obtain a modal
theory for countabilist set theories? The answer is yes.

8.3 A Reifying and Generifying Modal Theory

https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Scambler (2021) has provided a modal theory of sets that can be used for count-
abilist versions of the weak iterative conception. He starts with the background
^
of L ≺, ∈ but adds two modal operators ⟨v⟩ (for ‘vertical’ modality – reifying the
pluralities of the model into sets) and ⟨h⟩ (for ‘horizontal’ modality – adding
in subsets via forcing). Call this language L ∈,^,≺⟨h⟩, ⟨v⟩ . Boxes [h]ϕ and [v]ϕ are
defined as ¬⟨h⟩¬ϕ and ¬⟨v⟩¬ϕ as usual. In this context, the general ^ can be
thought of as ‘possible through some combination of ⟨v⟩ and ⟨h⟩’.
Scambler then provides the following axioms:84

Definition 50 Sca consists of the following axioms in L ∈,^,≺⟨h⟩, ⟨v⟩ (again I’ll
focus on giving more informal statements, the reader should go to Scambler
(2021) for the formal details):85

84 See Scambler (2021), p. 1091. I’m following the presentation in Barton (MSb).
85 Scambler uses the term “M” (for Meadows) to denote Sca, as he takes inspiration for his view
from Meadows (2015). As we’ll see later, Meadows’ work (drawing on Steel (2014)) is slightly
different (he considers proper class models); therefore I’ve chosen “Sca”.
 60 The Philosophy of Mathematics

(i) Classical first-order logic.

(ii) Impredicative plural logic.
(iii) Classical S4.2 with the Converse Barcan Formula for every
modality.
(iv) Plural Membership Definiteness (which is, recall, the scheme):

(∀x ≺ yy)□ϕ(x) → □(∀x ≺ yy)ϕ(x)

(v) The necessity of distinctness and stability axioms for ≺ and ∈.

(vi) Foundation. The Axiom of Foundation (the standard one from
ZFC).
(vii) Extensionality. Extensionality for sets (again, no different from
ZFC).
(viii) Weakening Schemas. ⟨h⟩ϕ → ^ϕ and ⟨v⟩ϕ → ^ϕ, for every ϕ.
(ix) Vertical Collapse. ⟨v⟩∃y□∀z(z ∈ y ↔ z ≺ xx).
(x) Modal Infinity. The axiom that there could vertically be some
things that necessarily comprise all and only the natural numbers:
⟨v⟩∃xx□∀y(y ≺ xx ↔ ‘y is a natural number’).
(xi) Vertical Modal Powerclass. The axiom that it’s vertically possible
to have some things that are vertically necessarily all the subsets of
a set: ∀z⟨v⟩∃xx[v]∀y(y ≺ xx ↔ y ⊆ z).
(xii) Possible Generics. The axiom ‘If P is a forcing partial order and dd
is some dense sets of P, then it’s horizontally possible that there is a
filter meeting each dense set that is one of the dd’.
(xiii) Choice. The plural version of the Axiom of Choice ‘For any
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

pairwise-disjoint non-empty sets xx, there are some things yy that

comprise exactly one element from each member of the xx’.86
(xiv) Modal Collection, Separation, and Replacement. Namely, poten-
tialist translations of the axiom schemas of Collection, Separation,
and Replacement under each modality.87

Some of these axioms deserve a mention. The Weakening Schemas are

meant to capture the idea that if I could get a set by either reifying pluralities
into sets or forcing, then such a set is possible simpliciter. Vertical Collapse
axiomatises the idea, as with Lin, that I could reify any plurality over a world
into a set. Possible Generics corresponds to the idea that I could always add

86 Scambler throws this in with the plural logic, but we’ll keep it separate.
87 Strictly speaking, Replacement is redundant given Separation and Collection. The reason to
separate these out is that Collection and Separation are strictly stronger than Replacement
when Powerset is removed (see Zarach (1996) and Gitman et al. (2016)). Scambler (2021)
works only with the potentialist translations of Replacement.
 Iterative Conceptions of Set 61

a generic for any partial order. One issue then is Vertical Modal Powerclass:
Notice that the version of powerclass – the axiom asserting that it’s possible
to have a world with all possible subsets of a set – is restricted to the vertical
modality. This will not hold with the broader modality since one can always
add subsets along the horizontal modality, and so there’s no world containing
every possible subset of an infinite set.
I think that there’s a good case to be made that Sca satisfies Naturalness.
We have a picture of how sets are formed; given some sets we can either Reify!
pluralities into sets (the vertical modality) or Generify! to add forcing generics
(the horizontal modality).88
Moreover, we again get a Paradox Diagnosis; since pluralities can always
be reified into sets, exactly the same points about the Russell plurality (the
plurality of all non-self-membered sets over a given world/stage) as in the case
of Lin still apply. Moreover, there are other conditions that can be shown to
not determine sets. For example, since any set can always be collapsed to the
countable using Generify!, more sets can always be made countable, and there
is no set of all countable sets. Not only does this conception reject Universality
and accept Indefinite Extensibility (for resolving the standard set-theoretic
paradoxes), but explains why its proponent should accept Forcing Saturation
and reject Powerset too (for a resolution of the Cohen–Scott Paradox).
Regarding Interpretation, two theorems are especially important. First we
have:

Theorem 51 (Scambler, 2021) Sca interprets ZFC− + Count under the poten-
tialist translation (the potentialist translation, recall, takes a formula ϕ in L ∈
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

to a corresponding one in L ∈,^,≺⟨h⟩, ⟨v⟩ by replacing every occurrence of ∀ with

□∀ and every occurrence of ∃ with ^∃).89

So there’s a sense in which when we have the full modality, thinking of the
sets as constructed in a manner consonant with Sca gets us ZFC− + Count.
This is a pretty nice set theory in which one can do much of the usual construc-
tions. However, it’s at least desirable to have contexts in which ZFC is true (for
interpreting the higher reaches of ‘standard’ set theory). For this we also have:

88 I discuss the Naturalness of Sca (in particular going through the axioms one by one) in Barton
(MSb).
89 Scambler actually shows that Sca interprets ZFC with the Powerset axiom merely removed.
However, a trivial modification (adding the modal versions of Collection and Separation
instead of Replacement) to his system gets you full ZFC− , so we state this stronger form of
the theorem. See Barton (MSb) for details. Recent work by Scambler shows that one can get
1
more, in particular that a regularity property for reals – the Π 1 -Perfect Set Property – holds
˜
under Sca. See Scambler (MS) for details.
 62 The Philosophy of Mathematics

Theorem 52 Scambler (2021) Sca interprets ZFC when we restrict to the ver-
tical modality (i.e. when we do the potentialist translation but replace □ and ^
by [v] and ⟨v⟩ throughout).

So when we restrict to the vertical modality we are able to interpret ZFC

(this is basically just because the vertical modality obeys Lin). However, we
have to ignore the horizontal modality that would allow us to collapse any
given uncountable set (and hence break the Powerset Axiom in the non-modal
theory).
The intuition behind Sca is thus the following. We have the vertical modality
that will allow us, starting with the empty set, to obtain ZFC by successively
reifying classes of worlds. However, we could, at any point, choose to introduce
a generic for a given partial order and family of dense sets. And, by interleaving
Reify! and Generify! we can get ZFC− + Count. Note that, unlike the strong
iterative conception or the constructibilist conception, the stages provided by
Sca need not be well-ordered. Instead, much like Odd! and Even!, we have to
think of applying Generify! and Reify! appropriately. Just as if you spin out
applying one of Odd! or Even! you won’t get all the hereditarily finite sets,
so with Generify! and Reify!. If you head off applying Reify! over and over
again, you’ll just get ZFC. And there are lots of ways of applying Generify!
badly too (e.g. by just adding single Cohen reals over and over again). But, if
we apply Generify! and Reify! just right, we will get ZFC− + Count.
The status of Capture is the one thing left outstanding for Sca. Recall that
Capture requires us to be able to produce a Kripke model for our modal theory
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

using the non-modal interpreted theory. Recent work by Scambler can be used
to show that there is a Capture-theorem available for Sca.90 Since the state-
ment is a little more involved and there are still some questions open, I’ll defer
its consideration until later.

8.4 Doing without Reify!

It’s worth mentioning here that one does not need the vertical modality in order
to get a conception of set that motivates ZFC− + Count. Although his work is
not intended for this purpose (his focus is more linguistic), John Steel has pro-
posed a theory of worlds and sets that will do the job without needing a vertical
modality. He proposes (in Steel, 2014) a two-sorted theory with variables for
sets x0, x1, ... and variables for universes W0, W1, ... with the following axioms
(here I follow the presentation in Maddy and Meadows (2020)):

90 See Scambler (MS) for the result, and Barton (MSb) for the application to Capture.
 Iterative Conceptions of Set 63

Definition 53 Steel’s Multiverse Axioms are as follows:

(i) The axiom scheme stating that if W is a world, and ϕ is an axiom of

ZFC, then ϕ holds at W.
(ii) Every world is a transitive proper class.
(iii) If W is a world and P is a forcing partial order in W, then there is a
universe W′ containing a generic for W.
(iv) If U is a world, and U can be obtained by forcing over some world
W, then W is also a world.
(v) Amalgamation. If U and W are worlds, then there are G and H that
are generic over them such that U[G] = W[H].

A discussion of these axioms, explicitly making the link with countabilism,

is available in Meadows (2015). Steel wants to use his theory to isolate the
‘forcing invariant’ part of set theory, regarding some sentences (like CH) as
indeterminate ‘pseudo-questions’ (Steel (2014), p. 154). Further analysis of
Steel’s project on its own terms is provided by (Maddy and Meadows (2020)).
However, I think we can use Steel’s multiverse axioms as inspiration for a
modal theory of sets for a version of the weak iterative conception. We start
with some proper class model(s) of ZFC, and our method of set-construction is
just Generify!.
Formally, we can provide the following axioms:91

Definition 54 Barton (MSb) SteMMe (for Steel-Maddy-Meadows) comprises

^
the following axioms in L ≺, ∈:
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

(i) Classical first-order logic.

(ii) Predicative plural logic.92
(iii) Classical S4.2 with the Converse Barcan Formula for ^.
(iv) The necessity of distinctness and stability axioms for ∈ and ≺.
(v) Plural Membership Definiteness (which we repeat here for ease):

(∀x ≺ yy)□ϕ(x) → □(∀x ≺ yy)ϕ(x)

(vi) The Ordinal Definiteness Schema: This is the schema of assertions

of the form:
( ) ( )
∀x ‘x is an ordinal’ → □ϕ(x) → □∀y ‘y is an ordinal’ → ϕ(y)

91 For more details about SteMMe, and a comparison with Sca, see Barton (MSb).
92 I adopt predicative plural logic since we will only need to talk about definable classes and it
will make some of the model-theoretic analysis easier. One can modify the approach to make
the underlying plural logic impredicative, if one so desires. See Barton (MSb) for discussion.
 64 The Philosophy of Mathematics

(vii) The necessitation of every axiom of first-order ZFC.

(viii) Possible Set-Generics. The axiom ‘If P is a forcing partial order
and D is a set of dense sets of P, then it’s possible that there is a
filter meeting each dense set that is a member of D’.
(ix) Modal Separation, Replacement, and Collection. The potentialist
translations of every instance of the Separation, Replacement, and
Collection schemas.93

Regarding Naturalness: The idea of the SteMMe is to take some proper-

class-sized model of ZFC as our starting sets and Generify! as our sole way
of forming new sets from old. There is no Reify! operation. .2 axiomatises
the idea that any two possibilities can be brought together, in line with Steel’s
Amalgamation axiom. Stability axioms and Plural Membership Definite-
ness are required again to ensure that neither ∈ nor ≺ (nor subplurality-hood)
can behave badly as new sets come into existence. The Ordinal Definiteness
Schema essentially posits the Barcan Formula for the ordinals, axiomatising
the principle that the ordinals can’t get longer. This captures the idea that our
stages are all proper-class-sized and we add the necessitation of first-order ZFC
to capture the idea that ZFC holds in each of these proper class models. Possi-
ble Set Generics is motivated by the idea that our set-construction method is
forcing, and is the axiom corresponding to Generify!
Regarding Interpretation, we can note:

Fact 55 Barton (MSb) SteMMe interprets ZFC− + Count under the potentialist
translation.94
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

and:

Fact 56 Barton (MSb) SteMMe interprets the potentialist translations of the

scheme asserting that every axiom of ZFC holds for the constructible sets.

So, just as in Sca, we have the nice theory ZFC− + Count, and we have
ZFC holding in some restricted contexts. We’ll critically examine how nice
this theory is (especially with respect to the goals of Section 2) in Section 9.
Concerning Paradox Diagnosis, we should note that, in stark contrast to
both the strong iterative conception and the version of the weak iterative con-
ception axiomatised by Sca, our worlds/stages are proper classes. There is a

93 There are redundancies here, but we separate them out in order to aid philosophical discussion.
94 You in fact get a version of the Perfect Set Property too, but I’ll suppress this detail for
now. See Barton (MSb).
 Iterative Conceptions of Set 65

possible puzzle here – why can’t we collect together the sets from one of
these proper class worlds to form a set? After all, all the members of some
proper classes (e.g. the ordinals) are ‘available’ for collection at every world.
The answer is that the collection forming operation – set forcing – does not
allow them to be collected. So we still have Paradox Diagnosis (though one
that merits some serious philosophical scrutiny).95 Although there are worlds
containing proper classes, we avoid contradiction by having a suitably ‘weak’
operation of set formation. This provides an explanation for the advocate of
SteMMe as to why the problematic classes like the Russell plurality do not
form a set, though in a manner somewhat different from Sca. We are, however,
given an explanation of why Indefinite Extensibility holds (one can always
use Generify! to add more sets), and which of Forcing Saturation and Pow-
erset fails (Forcing Saturation holds because any set can be made countable,
and Powerset fails because one can always use Generify! to add more subsets
of a given set).
Discussion of Capture I will defer until Sections 9 and 10. Like with Sca,
there is a Capture-theorem available, but it is somewhat more complicated and
its philosophical import is still open.

8.5 Summing Up
There are many details to be ironed out with these proposals (I will discuss some
in the next couple of sections). For now it suffices to note that though they are
somewhat nascent, there are theories like Sca and SteMMe that provide a modal
theory for the weak iterative conception that validates Forcing Saturation and
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Count. That’s all well and good, but we might ask at this point: Which is better
out of the strong iterative conception and the forcing-saturated weak iterative
conception when we bear in mind the goals of set theory?

9 Mathematics and Philosophy under the Different

Conceptions
This section will examine whether one of the strong iterative conception or the
countabilist versions of the weak iterative conception is best. We’ll do this by
looking at how mathematics is interpreted under each conception of set, and
examine each with respect to the theoretical virtues we discussed in Section 2.
We’ll first provide an explanation of how each handles mathematics, before
contrasting them side by side with respect to our theoretical virtues.

95 For example, Linnebo (2010) and Studd (2019), would not take this response to be satisfactory.
They think that there are good grounds for asserting that whenever we have a plurality, we can
turn it into a set, and so they won’t find the picture provided by SteMMe appealing.
 66 The Philosophy of Mathematics

9.1 Mathematics under the Strong Iterative Conception

Let’s first recap the situation with the strong iterative conception. As we noted
in Section 2, ZFC and the strong iterative conception does an extremely good
job of interpreting mathematics. A couple of extra things should be mentioned
though at this point.
One core problem for the advocate of the strong iterative conception is to
resolve questions about Theory of Infinity. For, whilst they do have ZFC, this
theory tells us vanishingly little about the behaviour of infinite sets, and in
particular the values of the continuum function f(ℵα+1 ) = 2ℵα or whether large
cardinal axioms hold. More has to be done to substantiate new axioms for set
theory, and there’s a rich literature on the topic.96
One kind of mathematics that the advocate of the strong iterative conception
has to interpret are the modal countabilist theories we’ve discussed here. On her
view, these modal theories can be construed as about the hereditarily countable
sets (i.e. sets built up only from countable sets – formally we say that a set is
hereditarily countable if it is a countable set containing only hereditarily count-
able sets). In this way, the advocate of the strong iterative conception holds that
theorists advocating forcing-saturated weak iterative conceptions can be inter-
preted as talking about structures that miss out a great many large sets (and in
particular all the uncountable ones).

9.2 Mathematics under Our Countabilist Conceptions

https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Things are a little more challenging under the forcing-saturated weak itera-
tive conceptions. Because we don’t have Powerset, we can’t just piggy-back
off the ‘standard’ account of mathematics available under the strong iterative
conception.
We’ve seen two versions of the weak iterative conception (given by SteMMe
and Sca) that validate Forcing Saturation. However, in this context we don’t
have the Powerset Axiom, and hence can’t build many of the usual representa-
tions of structures that we want. So there’s a number of questions we can ask
about the forcing-saturated countabilist interpretation of mathematics:

(1) How should we understand the study of theories based on ZFC?

(2) What does ‘mainstream’ mathematics look like under this conception?
(3) What does our Theory of Infinity look like?

96 See, for example, Maddy (1988a), Maddy (1988b), Koellner (2014), and Incurvati (2017)
among many others.
 Iterative Conceptions of Set 67

9.2.1 How Should We Understand the Study of Theories Based

on ZFC?

The quick answer is that you can still have ZFC in a restricted domain; you just
can’t have all subsets of the sets in those domains (since for any set x, there’s a
collapsing function from x to ω). If you want to have ‘uncountable sets’ you just
have to leave out the subsets that witness bijections with the natural numbers.
(A parenthetical remark that should be included at this point: The idea that
sets might be small but ‘appear’ large in some model appears in the work
of Skolem, especially Skolem (1922). Often, however, Skolem’s position is
cashed out via a scepticism and/or referential indeterminacy by asking the ques-
tion “How do I know I’m not living in/speaking about a countable model?”.
The present family of views does not have this flavour. It is compatible with
the idea that we can refer to the universe without an issue; it is just that the level
of Forcing Saturation is so strong that we can only talk about ‘uncountable’
sets by missing out functions.)
One can have very natural looking models here. For example, as well as
countable transitive models, it’s possible to have transitive models of ZFC con-
taining all ordinals (so-called inner models) within a model of ZFC− + Count.
Recall that, for example, Sca interprets ZFC under the vertical modality and
SteMMe can get ZFC in the constructible universe. So any countabilist theory
based on Sca or SteMMe has inner models of ZFC.
Aside from the use of the modal theories presented here, there are also natural
axioms that get us inner models for ZFC plus large cardinals. Since the axioms
are somewhat complex, I’ll provide them in a -box:
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

I’ll mention some in passing, but I won’t go into details since the math-
ematics starts to get tricky. The interested reader is directed to Scambler
(MS), Barton (MSb), and Barton and Friedman (MS) for further refer-
ences and a fuller discussion of these examples. One way is to assert the
existence of ‘sharps’ – these imply that there are self-embeddings from
many inner models and can be used to get ZFC plus large cardinals in
inner models within ZFC− + Count.a Another (related) kind are regular-
ity properties for sets of reals. In fact, both Sca and SteMMe can interpret
a scheme corresponding to a version of the perfect set property (for the
cognoscenti, the Π11 -Perfect Set Property) which implies that there are
˜
many inner models of ZFC.b Other regularity properties for sets of reals
(e.g. Projective Determinacy) can be (schematically) rendered in ZFC− +
Count, and also imply that there are inner models of ZFC plus many large
cardinals. Finally in Barton and Friedman (MS) we propose an axiom (the
 68 The Philosophy of Mathematics

Ordinal Inner Model Hypothesis), which implies that every set is count-
able but also that ZFC with large cardinals added holds in inner models
(again for the cognoscenti – one can get 0♯ ).
a See Regula Krapf’s PhD thesis (Krapf (2017)) for details of handling sharps in the
countabilist context.
b See Scambler (MS) and Barton (MSb). Defining the Π 11 -Perfect Set Property would take
us too far afield; see, for example, Kanamori (2009)˜(sections 11-12) for details.

There is thus a kind of ‘symmetry’ between the strong iterative concep-

tion and the forcing-saturated weak iterative conception. Under the forcing-
saturated weak iterative conception, the theories motivated by the strong
iterative conception should be understood as holding in transitive models that
miss out subsets (in particular all the collapsing functions). But under the strong
iterative conception, the theories motivated by the forcing-saturated weak iter-
ative conception seem to miss out large sets (in particular all the uncountable
ones).97

9.2.2 Mathematics for the Countabilist

The picture of mainstream mathematics is much different when we have
Forcing Saturation. Whilst arithmetic remains unchanged (one can have Vω
exactly as under the strong iterative conception), there are no uncountable set-
sized structures. Rather, all uncountable collections are proper-class-sized.
The study of all the real numbers thus becomes the study of a large proper
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

class.98 Since there are exactly continuum-many continuous functions between

the reals, we can also think of the study of all continuous functions f : R → R as
examining a proper class. But whilst the real numbers and class of all continu-
ous functions are proper classes, yet higher mathematics for larger uncountable
cardinals cannot be interpreted as about the sets without the use of even higher-
order logic. For example, the classical study of the space of all functions
f : R → R (a key structure for functional analysis) cannot be interpreted even
by a proper class. One might ask oneself at this point whether this is bad or just
merely different. We’ll return to this issue later on (Section 9.3).

97 In Barton (MSa) I’ve argued that this symmetry can be used to claim that uncountabilism is in
fact restrictive.
98 In fact, since you can think of a real number as coding a countable set, the study of set
theory is in a way just the study of real numbers under ZFC− + Count. This is supported by
the fact that second-order arithmetic and ZFC− + Count are bi-interpretable. See section 5.1
of Regula Krapf’s PhD thesis (Krapf (2017)) for a nice presentation of this result.
 Iterative Conceptions of Set 69

9.2.3 What Does Our Theory of Infinity Look Like?

How is Theory of Infinity handled? There are (at least) two different kinds of
question one could ask:

(1) How should we understand the Theory of Infinity provided by ZFC?

(2) What is the Theory of Infinity simpliciter?

The former question is easily handled under the forcing-saturated weak itera-
tive conception. Since ZFC is only true relative to a model that misses out sets,
the behaviour of the continuum function (as well as other independent sen-
tences) should be understood via the diverse world-to-world information we
get out of the different models of ZFC. This has affinities with some so-called
multiverse views in the philosophy of set theory (we’ll discuss these later in
Section 10; for our purposes now one can simply read ‘multiverse’ as the col-
lection of all countabilist worlds). For example, Joel-David Hamkins (2012)
writes:

the continuum hypothesis is a settled question; it is incorrect to describe the

CH as an open problem. The answer to CH consists of the expansive, detailed
knowledge set theorists have gained about the extent to which it holds and
fails in the multiverse, about how to achieve it or its negation in combination
with other diverse set-theoretic properties. (p. 429)

Since there is no maximal ZFC structure for the forcing-saturated weak iter-
ative conception, we have an answer to the question of CH in ZFC-based set
theory. Simply put, it is to be found in how CH behaves across structures that
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

satisfy ZFC. No further answer is needed or possible.

This answer only concerns the impoverished ZFC models for the count-
abilist. So what is their Theory of Infinity simpliciter? This question is
answered for sets – every set is either finite or countably infinite. So, in a sense,
the countabilist has a comprehensive (albeit slightly boring) answer for the rel-
ative sizes of sets. However, there are still some interesting questions to be had.
Since the continuum is a proper class, CH is now a claim about what proper
classes exist coding bijections between classes of sets and the universe. Is every
infinite class of reals either countable or the size of the universe? This is the
open question that the countabilist must address.

One very interesting fact is that in this context CH is equivalent to

the claim that the universe is bijectable with the ordinals. So we have an
immediate link with CH and versions of Global Choice. Moreover, CH is
 70 The Philosophy of Mathematics

equivalent for the countabilist to the ‘limitation of size’ principle that all
proper classes are the same size.a If the advocate of the forcing-saturated
weak iterative conception could motivate this limitation of size principle,
they would then have a complete story for Theory of Infinity. Simply put,
every collection would be either (a) finite, (b) infinite, or (c) proper-class-
sized, and the continuum hypothesis (rendered as a claim about proper
classes) would be true.
a See Holmes et al. (2012), section 3.4.

9.3 Contrasting the Conceptions

Is one of the two conceptions better? Both have different ways of responding to
Theory of Infinity and advocate apparently distinct pictures of the foundations
of mathematics. And both have some open questions that remain outstanding.
This all raises the issue of what will become of the different conceptions,
especially when we bear in mind the criteria outlined in Section 2. I won’t come
down one way or the other here – I think there are many questions that should
be left open for the future. The main point I want to press is the following: Both
are attractive conceptions of set.
I do think it’s pretty clear that the strong iterative conception, with the rich
understanding we have of it and theories motivated on its basis, is well in the
lead in the race. This is to be expected; we’ve only recently started looking
seriously at the forcing-saturated weak iterative conception, and so the strong
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

iterative conception had an enormous head start (a good 50 years or so). Races
that seemed one-sided can get more competitive over time though. For exam-
ple, the logical conception is experiencing something of a resurgence due to its
possible application in formal semantics having previously been regarded as
almost dead in the water (or at least deeply problematic).99 So it’s worth think-
ing of how each responds to the desiderata outlined in Section 2, contrasting the
two, and considering whether the forcing-saturated weak iterative conception
might catch up. For the sake of ease, we repeat our theoretical virtues here:

Generous Arena Find representatives for our usual mathematical structures

(e.g. the natural numbers, the real numbers) using our theory of sets.
Shared Standard Provide a standard of correctness for proof in mathematics.
Limits of Formalisation Set theory provides a natural place to examine
the limits of our formalisation, pushing the boundaries of what might be

99 See Linnebo (2006), Linnebo and Shapiro (2023), and Roberts (MSa).
 Iterative Conceptions of Set 71

realistically expected to be captured, and exploring where formalisations may

finally give out.
Testing Ground for Paradox Set theory is very paradox prone, both in terms
of the principles that can be formulated within set theory and when combined
with certain philosophical ideas (e.g. absolute generality and mereology). In
this way, set theory provides a testing ground for seeing when and how ideas
are inconsistent.
Metamathematical Corral Provide a theory in which metamathematical
investigations of relative provability and consistency strengths can be con-
ducted.
Risk Assessment Provide a degree of confidence in theories commensurate
with their consistency strength.
We also added (on conceptions of set):
Naturalness Provide a reasonably natural account of what the sets are like,
one which avoids ad hoc restrictions.
Interpretation A conception should motivate a good theory of sets.
Paradox Diagnosis Respond to the explanatory challenge: Explain why the
paradoxical collections aren’t sets and which conditions do (and do not)
determine sets.
Capture Let T^ be a modal theory of sets interpreting a non-modal theory T
via the potentialist translation. Then we say that T^ satisfies Capture if and
only if, given a model M of T, we have a general way of extracting a Kripke
frame KM T^
|= T^ from M such that for every x ∈ M there is a world W ∈ KM T^
such that x ∈ W.
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Generous arena is handled very differently by the two approaches. But each
has its own answer. The strong iterative conception can essentially piggy-back
off the standard account of Generous Arena given in Section 2. Little more
needs to be said here.
The case of the forcing saturated weak iterative conception is more contro-
versial. Here the reals are a proper class (at least in the non-modal theory).
Set theory here is directly akin to second-order arithmetic, and analysis can be
thereby interpreted (so long as we allow talk of proper classes). But third-order
arithmetic is out of reach, standardly interpreted. However, since we have ZFC
plus large cardinals in inner models, proofs using resources from third-order
arithmetic and above can be interpreted in restricted contexts. Whether this
constitutes a hobbling of mathematical practice or just a different approach is
a question I leave open for philosophical examination.100

100 See, for example, the debate between Solomon Feferman and John Steel in Feferman et al.
(2000), as well as Tatiana Arrigoni and Sy-David Friedman’s take on the matter in Arrigoni
 72 The Philosophy of Mathematics

This has implications for Shared Standard. Both the strong iterative con-
ception and forcing-saturated weak iterative conception provide their own
Generous Arena, and hence their own account of when a proof is legitimate.
Each standard is very different, though, if we have Forcing Saturation, third-
order resources are not legitimate for reasoning about the reals. So both have an
account of Shared Standard, but the forcing-saturated weak iterative concep-
tion deviates substantially from the currently accepted norm. This said, under
this countabilist approach, proofs in third-order arithmetic and/or ZFC are not
wrong, they just need to be interpreted in restricted contexts. Again, I leave it
open whether or not this should count against the position or it is simply merely
different.
Regarding the Limits of Formalisation, both are able to handle Gödelian
incompleteness in much the same way (claims about relative provability can be
construed as claims about first-order arithmetic, and the first-order arithmetic
provided by the two conceptions are not significantly different)101 . However,
since both provide very different pictures of the role of the continuum and
independence, they provide quite different answers to the question of our
knowledge of the continuum. The strong iterative conception has several ques-
tions to answer about large cardinal independence and the behaviour of the
continuum function. The forcing-saturated version of the weak iterative con-
ception, on the other hand, answers basically all questions about sets. Every
set is countable, and there are no large (or even uncountable) cardinals, even
if there are large cardinals and uncountable cardinals in inner models. Since
all sets are countable, it’s perhaps somewhat unsurprising that we can answer
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

questions about them more easily. Still, the continuum hypothesis is pushed
to a question about class theory, and in particular is connected with global
well-orders for the universe (whether there’s a proper-class-sized bijection
F: V ↣ → Ord) . As we noted earlier, if such a countabilist can motivate the
claim that all proper classes are the same size, then CH is solved too. But
perhaps one can argue that whilst the sets are relatively easily known, the con-
tinuum/proper classes are not, and so we leave this question open. But there are
at least avenues for making philosophical progress on this question.

and Friedman (2013). Relevant here is the aforementioned multiverse view provided by John
Steel (in Steel (2014)) with subsequent development by Penelope Maddy and Toby Meadows
(Maddy and Meadows (2020)). See also Barton and Friedman (MS) (for an argument that
many of the usual foundational roles for large cardinals can be performed in the countabilist
setting).
101 Really, all one gets is that the different theories proposed will yield more/less information
about the natural numbers. But any theory of arithmetic compatible with one conception is
compatible with the other.
 Iterative Conceptions of Set 73

Moreover, both provide interesting perspectives as a Testing Ground for

Paradox. This is in two ways. First, the incompatibility between Powerset and
Forcing Saturation and the two conceptions we’ve discussed provides for an
interesting kind of ‘paradox’ in its own right (this is part of what was at play in
the Cohen–Scott Paradox). Interestingly, although each denies the full gener-
ality of the other’s principles, one can incorporate partial amounts thereof. The
proponent of Powerset can add in limited amounts of Forcing Saturation, for
restricted kinds of partial order and families of dense sets (this yields a class of
axioms known as forcing axioms). Interestingly, the addition of such restricted
Forcing Saturation into the strong iterative conception tends to yield a resolu-
tion of CH in the negative, with 2ℵ0 = ℵ2 .102 It is not known how to generalise
these axioms for higher values of the continuum function. For the proponent
of the forcing-saturated weak iterative conception of set, we can begin by not-
ing that axioms postulating the existence of uncountable cardinals are a bit like
large cardinal axioms – they (incorrectly!) assert the existence of sets closed
under various kinds of operation. For example, the least uncountable cardinal
can be thought of as a set that is closed under the formation of hereditarily
countable cardinals. Over ZFC− , an uncountable cardinal behaves a bit like an
inaccessible cardinal does in ZFC.

For example, let κ be the least inaccessible and ω1 be the least uncount-
able cardinal. Both are regular, and both provide a natural model for the
base theory – Vκ provides a model for ZFC (in fact, second-order ZFC),
and H(ω1 ) provides a model for ZFC− .
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Moreover, one can postulate the existence of sets with closure under count-
abilism (just not enough to get you an uncountable cardinal). Here’s a slightly
tricky example:

Consider the following schematic reflection principle (for any ϕ in the

language of set theory):

∀x∃a(x ∈ a ∧ ‘a is transitive’ ∧ ϕ ↔ ϕa )

that is, for any set x there is a transitive set a such that x ∈ a and ϕ is
absolute between a and the universe. ZFC− with this added is known as
ZFC−Ref . This theory is very weak – still far below the consistency strength
of ZFC (and so is consistent if ZFC is). But it adds in sets with closure;

102 For example, the Proper Forcing Axiom implies that 2ℵ0 = ℵ2 . For a survey of the Proper
Forcing Axiom, see Moore (2010).
 74 The Philosophy of Mathematics

in particular, if ϕ holds in the universe, then ϕ holds restricted to some

transitive set a. And since the universe exhibits various closure properties,
this version of reflection will imply that there are sets with those closure
properties too.

So whilst we know that we’ll have to get rid of one of Forcing Saturation or
Powerset, whichever way we go, we can add back in some restricted versions
of whatever we rejected.
Metamathematical Corral can be dealt with immediately. Both concep-
tions motivate theories that can handle talk of set-theoretic models easily, and
so there is no particular difference here. Similarly for Risk Assessment, whilst
there might be small fluctuations dependent upon which theory is eventually
picked, both conceptions can motivate theories with a good deal of strength
on an independently plausible conception. We also might think that there’s no
need to settle on a single conception for Risk Assessment; so long as the con-
ceptions seem cogent and coherent, we can have confidence in the consistency
of theories that are proved consistent on each picture. In particular, if a the-
ory U is proved consistent by theories motivated under each conception, then
more power to U – its consistency is converged upon by two distinct cogent
conceptions of set.
Each of Naturalness, Interpretation, and Paradox Diagnosis has been dis-
cussed earlier, and so I won’t repeat myself. Suffice it to say, all of Lin, Sca,
and SteMMe perform fine with respect to these desiderata. However, we owe
the reader an explanation of Capture. The strong iterative conception and Lin
performs perfectly here; one can use the Vα to quickly provide a Kripke model
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

for Lin, and then (using the theorem that every set belongs to some Vα ) get the
result that every set lives there. For the countabilist versions of the weak itera-
tive conception, there are Capture-theorems available, but since the details are
a little fiddly, I relegate their discussion to a -box (even then though, giving
all the gory details would get us too deep in the weeds, so I direct the inter-
ested reader to the relevant papers). The core point is that though one can get a
Capture-theorem for these theories, it is still somewhat unsatisfying.

The core observation used to get Capture-theorems for Sca and

SteMMe is that both are able to interpret the schemas corresponding
to a regularity property known as the Π11 -Perfect Set Property (I won’t
˜
define it here; many introductory texts on descriptive set theory contain
the details).a Scambler (MS) provides the result for Sca, and Scambler’s
strategy can be used for SteMMe too (see Barton (MSb)). This, in turn
 Iterative Conceptions of Set 75

implies that there are many inner models of ZFC (in fact, the Π11 -Perfect
˜
Set Property implies that L[x] satisfies ZFC for every real x; see Solovay
(1974) and Taranovsky (2004)). This regularity property can then be used
to get Kripke models for Sca and SteMMe containing every set (basically
using the various L[x]).
The slight dissatisfaction results from the fact that in the case of Lin, the
Kripke model we obtain meshes very nicely with the informal description
of the way sets are formed (namely via powerset and union). Indeed, the
accessibility relation we get out of the Vα exactly matches some applica-
tion of powerset and union-bundling. However, for the cases of Sca and
SteMMe, there may be countable sets that are in the Kripke model that
are not added by forcing. One might find this slightly dissatisfying – after
all, aren’t the modal theories supposed to be telling us how the sets are
formed? Coming up with a modal theory that better meshes with the infor-
mal set-construction methods represents an open question for the advocate
of forcing-saturated versions of the weak iterative conception that we’ll
explicitly identify in Section 10.
a See, for example, Kanamori (2009), esp. sections 11–12.

For these reasons I think that both the strong iterative conception and the
forcing-saturated weak iterative conception are each viable conceptions of set.
The strong iterative conception clearly fits better with current orthodoxy, but
that’s not a good reason to discount the forcing-saturated weak iterative con-
ception out of hand. In the end, I think that a careful analysis is needed, either to
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

choose one of the two or to learn to live with the pluralism they offer. For this to
be done successfully, more development of these two (and other) conceptions
is required, especially on the side of the juvenile weak iterative conception.

10 Conclusions, Open Questions, and the Future

A short summary of what I’ve argued in this Element: I think that set theory
provides an interesting case study and tool for both philosophers and mathe-
maticians. I think that progress in set theory often involves trading off different
principles (e.g. Universality and Indefinite Extensibility, Powerset and Forc-
ing Saturation). I think that this is the situation we find ourselves in now (at
least to some degree).
This said, there’s a lot more research to be done in this direction. Some areas I
have already identified, but some are new and so I want to close with a summary
and consolidation of what I take to be the most interesting questions for mov-
ing forward. Importantly, we’ll also be able to identify further connections with
 76 The Philosophy of Mathematics

philosophy more broadly (e.g. absolute generality and modal metaphysics). It

will also be helpful to present some objections to what I’ve argued, and men-
tion how they could be answered. This will make this ‘conclusion’ longer than
usual, and I hope the reader will indulge me in this.

10.1 Capturing the Sets

We finished the previous section by identifying a major issue for countabilist
versions of the weak iterative conception, namely the status of Capture. Whilst
Capture-theorems are available, there is much work to be done here, in par-
ticular finding a conception, modal theory, and Capture-theorem on which
the informal description of the set-construction methods matches up with the
accessibility relation on the relevant Kripke frame (as is the case with Lin and
the Vα ). So we simply ask:

Question Is there a reasonable presentation of a modal theory T^ that moti-

vates an extension T of ZFC− + Count, but where one can (in T) recover a
Kripke model of T^ and prove that every x is a member of some stage, and
have the relevant accessibility relation conform to the informal description of
how the sets can be constructed?

As remarked earlier (Section 9) there is some progress here, but the Capture-
theorems for Sca and SteMMe are not as good a match as we get with Lin and
the Vα . We can also ask similar questions about non-modal theories of the weak
iterative conception. Positively resolving this question would go some way
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

to cementing countabilist weak iterative conceptions as genuine contenders,

rather than up-and-coming prospects.

10.2 The Weak Iterative Conception Needs Work

Earlier (Section 8) I remarked that the strong iterative conception is further
ahead in the race as compared to other versions of the weak iterative conception
(and in particular countabilist ones). I want to address some concerns one might
have that these countabilist conceptions are more problematic than I’m letting
on (perhaps they should not even be regarded as qualifying entrants).
There are a few reasons one could give to substantiate this claim. The strong
iterative conception, one might contend, is well-developed. We have an account
of what the worlds are (the Vα ). By contrast, the weak iterative conception
seems rather underspecified, and clearly in need of sharpening by a further
conception. But what are the constraints here? What is to count as a legiti-
mate method of set-construction? These are all left unanswered by the weak
 Iterative Conceptions of Set 77

iterative conception and we might worry that the weak iterative conception is
not sufficiently well-formulated to provide enough constraints.
Here’s a somewhat silly example of a description of an iterative process.

Definition 57 (Informal) The trivialising conception of set holds that sets are
formed in stages. There are just two stages. At stage 0 we have nothing. At
stage 1 we perform the following operation: Form all the sets!. There are no
other stages.

What’s wrong with this as a version of the weak iterative conception? I think
it’s important to recall (Section 3) what we want out of a conception of set.
We want a conception that satisfies Naturalness, Interpretation, Paradox
Diagnosis, and Capture.
This trivialising conception does not perform well here. Whilst I’m not quite
sure how to assess it for Naturalness (and Capture will likely be trivially sat-
isfied should one be able to provide a suitable theory) it performs very poorly
with respect to Paradox Diagnosis and Interpretation. It doesn’t explain why
paradoxical collections don’t get into its second stage. We have no explana-
tion – beyond wielding a contradiction as a Dummetian ‘big stick’ – of why the
operation Form all the sets! doesn’t form paradoxical ones. And it is totally
uninformative about the theory we should adopt. So, yes, it is a legitimate ver-
sion of the weak iterative conception. But it is also rubbish. We can thus safely
kick it to the kerb. By contrast, the countabilist versions of the weak iterative
conception, with their attendant axioms and modal theories, look promising,
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

even if slightly less developed than the strong iterative conception.

That’s not to say that there aren’t some questions here that need to be
answered under these countabilist conceptions. An important issue is to work
out the details of the modal theories for the weak iterative conception. One of
the major differences between the strong iterative conception and these is that
the modal theory of the former is pretty much fully worked out (beyond Theory
of Infinity – there will be potentialist translations of sentences independent of
ZFC that are up for grabs). For the latter, things are less settled (though there
are options as discussed in Section 8). I want to make a few points about mov-
ing forward with the project of isolating appropriate modal theories, and the
challenges that need to be overcome.
First, I think that the weak iterative conception is extremely broad. This is
evidenced by the fact that the trivialising weak iterative conception is a legiti-
mate version of it, even if terrible as a conception of set. Moreover, there are
very many disparate conceptions that also fall under this banner (e.g. the con-
structibilist conception and the forcing-saturated conception don’t seem to have
 78 The Philosophy of Mathematics

a whole lot in common beyond their weak iterativity). So I don’t think we are
going to get a lot of informativeness out of the weak iterative conception alone.
However, one thing we do get is the idea that there should be some sort of
description of the universe as unfolding as new sets are built. And I think the
following is true: Legitimate ways of constructing sets should be well-founded.
Here lies the challenge for coming up with a more detailed account of the
stages and/or modal theories for the weak iterative conception: Many of the
possible candidates for modal theories considered in Section 8 are not well-
founded in the sense that the corresponding frames don’t have a well-founded
accessibility relation. The problem concerns forcing: It’s pretty rare – though
not impossible – to have forcing extensions that are minimal in any interesting
sense.103

Here’s an example: Imagine I’m adding a single Cohen real to some

structure M. Given such an M, there are always non-interdefinable Cohen
reals G and H, and so there is a choice to be made about which to add. So
such a way of adding sets is not well-ordered (under inclusion, at least).
Nor is it well-founded under inclusion; one can get denseness in the
ordering. Start by identifying that for any Cohen real G, there is a Cohen
real H from which G can be defined but not vice versa. Moreover, if G is
definable from H but not vice versa, there is also a Cohen real I that (i) G
is definable from I, (ii) I is not definable from G, and (iii) I is definable
from H. Thus, given any two single-Cohen-real forcing extensions M[G]
and M[H] such that M[G] ⊊ M[H], there is also a dense ordering of M[I]
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

between them.

Is this a knock-down? I think not. The point I wish to make is that although
accessibility is non-well-founded in the relevant Kripke model, the notion of
a set-construction method might still be well-founded. One just needs it to be
indeterminate what set gets added.
To see this, here’s a simpler non-set-theoretic example. Let’s suppose that
we’re given a finite line segment l ⊂ R. Suppose further that I have a single
construction method Extend! that allows me to extend l in a single direction.
Now I could extend l to the left, or I could extend l to the right. Moreover,
if I extend l left, I don’t get what I get if I extend l right (let’s assume that l1
and l2 have to comprise exactly the same points to be identical). Moreover, any
time I extend l in one of the two directions to a line l′, there’s a dense ordering

103 A good example of a partial order in which one does have an interestingly ‘minimal’ extension
is Sacks forcing. See Geschke and Quickert (2004) for a survey.
 Iterative Conceptions of Set 79

of smaller lines that I could have extended to (with length greater than l but
smaller than l′). But this seems to me like a perfectly fine way of constructing
new lines from old. Indeed, one could formalise this modally if one so desired,
and a corresponding Kripke frame could have a non-well-founded accessibil-
ity relation. But the method of construction isn’t non-well-founded, it’s just
indeterminate.104 This has been recognised since at least the time of Euclid
and Aristotle (indeed, there is a more-than-superficial resemblance to Zeno’s
dichotomy paradox). What I suggest is that one looks at the well-founded sub-
relations of the accessibility relation. These will correspond to ways we can
legitimately construct new objects from old. For the strong iterative concep-
tion, it is just their luck that their accessibility relation is well-founded and
matches their specification of the processes involved in their version of the
weak iterative conception. But this needn’t be the case. It’s then open to us to
say that, whilst I can force to a world (and thereby see a descending sequence in
the accessibility relation), the way I get to any world has to be doable in a well-
founded way. But this suggestion, though promising, is very far from being
worked out in detail and represents a substantial open question that needs to be
answered for modal theories like Sca and SteMMe. So we ask:

Question Is there an account (possibly formal) of the weak iterative concep-

tion that makes clear the notion of a legitimate method of set construction?
A separate question that should be addressed when developing the weak iter-
ative conception (and one I’ve largely sidelined in this Element) is whether it is
better to cash things out in modal or stage-theoretic terms. It is very natural, as
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

I’ve done throughout this Element, to think of modal theories as implicitly giv-
ing us some notion of stage, where a stage can just be identified with a world.
This move isn’t clearly forced on us, however. For example, we might instead
choose to formulate the notion of stage directly (as in much work in the latter
half of the twentieth century).105 So we ask:

Question How do modal theories of sets and stage theories philosophically

relate to one another? Should we think of them as different ways of talking
about the same subject matter?106

10.3 The Story Is Too Neat, and Ignores Much

Throughout this Element, I’ve presented the idea that we can view different
attractive conceptions of set as arising out of trading off Forcing Saturation

104 See also Barton (MSb) for this argument.

105 See Button (2021a) for a summary of the history.
106 I thank Davide Sutto and Chris Scambler for pressing this point.
 80 The Philosophy of Mathematics

and Powerset. But I want to emphasise that whilst I do think this is a fundamen-
tal tension, there are many more options out there, some of which are weakly
iterative. What about, for instance, inner model theory and the Ultimate-L pro-
gramme (Woodin, 2017)? I won’t go into detail about this here, but the rough
idea is to come up with an ‘L-like’ model that is able to give a good structure
theory for V and still incorporate large cardinals (V = L implies that many
large cardinals don’t exist). There are even iterative-style set theories on which
every set has a complement (see Forster (2008) and Button (2022))! And what
about the cornucopia of proposals for motivating set-theoretic axioms under the
strong iterative conception (e.g. Freiling’s darts, determinacy, forcing axioms,
reflection principles)? Aren’t I presenting an all-too-narrow view of the state
of the art?
Yes! It is absolutely too narrow, and space doesn’t permit me to go into the
full details of every possible direction in set theory. My point here was not
to propose Powerset and Forcing Saturation as the two possibilities for set-
theoretic development (though I do think they might be especially attractive to
philosophers). My focus was rather to articulate the idea that in certain contexts
we can see conceptions as emerging from trading off inconsistent principles,
and thereby highlight some similarities between our own predicament and that
of our intellectual ancestors. In particular, I made simplifying assumptions there
too – there’s far more out there than the conceptions I concentrated on.
Really the space of conceptions should be far broader and the distinctions
not as conceptually neat as these pages might seem to indicate. I’ve said little
about other – possibly non-iterative – conceptions of set. We might have exam-
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

ined the graph conception (e.g. Incurvati (2020)). There are conceptions based
on non-classical logics, such as paraconsistent (e.g. Priest (2002); Jockwich
et al. (2022)) or constructivist/intuitionist approaches (e.g. Feferman (2010);
Bell (2014); Scambler (2020)). Others use an idea of predicativity (e.g. Fefer-
man and Hellman (1995); Linnebo and Shapiro (2023)). Some of these may
fit into the weak iterative conception mould, but there’s no requirement to
do so.
The point is just the following: This Element isn’t meant to be providing a
classification for every conception of set. My point is just that by considering
(i) the interrelations between different conceptions, and (ii) how we trade off
inconsistent principles, we can come to understand better the space of possibil-
ities for articulating the mathematically fertile notion of collection. There is a
huge amount of work to be done in developing many more conceptions of set,
and thinking about their relationships to philosophical questions. We should
start broadening our horizons now.
 Iterative Conceptions of Set 81

10.4 Potentialism, Actualism, and Absolute Generality

Throughout this Element, we’ve been discussing modal theories of sets. An
important question in the philosophy of mathematics concerns how we should
think of these modalities. I want to put on the table three possible answers to
this question:

Actualism/Universism There is a single universe of sets and a definite plural-

ity of all sets.
Potentialism There is a single universe of sets, but it is modally indefinite.
There is no definite plurality of all sets.
Multiversism There is no single universe of all sets, rather many universes.

(Note: It may be that we should relativise these questions to a given

conception, with different questions of how the stages/worlds/modalities are
interpreted for different conceptions.)
These views do not exhaust the logical space (e.g. we could have a universe
that is indefinite, but not modally so; see Feferman (2010), Scambler (2020))
but they are the main ones that are relevant for iterative conceptions. Each view
suggests a different way of philosophically interpreting the relevant modalities.
Let’s start with the modalities involved in the theories we’ve considered. The
actualist regards the use of modality as a mere heuristic for talking about the
stage-theoretic structure of the universe. The potentialist takes the modality
seriously, and thinks that it is somehow indicative of the fundamental nature
of reality. The multiversist also thinks that the modality is a mere heuristic but
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

in a very different way from the actualist; for them it is a way of talking about
interrelationships between the different universes on offer, and ways of moving
between them.
Each view has its own idiosyncrasies and suite of problems to be addressed.
One aspect of each is how we regard the determinacy of truth concerning math-
ematical claims (in particular in the language of set theory). The universist will
likely assert that every sentence of set theory has a definite truth value – assum-
ing we can refer to their universe without issue, the truth or falsity of claims
should just be understood as the truth or falsity of claims there. Likewise the
multiversist will likely assert that there are set-theoretic claims of indetermi-
nate truth value – true in some worlds and false in others. The potentialist
(given mirroring) is likely to fall on the side of determinacy, at least insofar
as ‘normal’ mathematical claims go (which should be understood under the
potentialist translation). Whilst this is perhaps somewhat broad-brush – there
are possibilities for modifying the conception of truth for each view – those at
least seem like the main options.
 82 The Philosophy of Mathematics

For the universist, there is also the old problem of the nature of proper
classes. For example, Øystein Linnebo (2010) writes:

Since a set is completely characterized by its elements, any plurality . . .

seems to provide a complete and precise characterization of a set . . . . What
more could be needed for such a set to exist?107 (p. 147)

The problem is as follows. Given the stages of any version of the weak iter-
ative conception, the universist holds that there is a determinate totality of all
the sets in the stages. This can be cashed out in plural terms; there are some
sets xx such that that could be no set of all the xx (for ease, let’s just assume
that the xx comprise every pure set). But what is it then that stops us forming
these sets into a new set? We have a definite plurality of them, and so could
characterise the relevant membership relation. One response is to say that con-
tradiction would ensue. But this only holds if you assume that the xx contain
every possible pure set. So, the universist has to come up with a meaningful
explanation of proper classes that makes it clear why they’re different from
sets, and why the seeming ability to talk about such collections isn’t an issue.
Similarly, many see the generality and flexibility of forcing as evidence that
a given domain of sets can be expanded. Here’s Hamkins (2012) on the subject:

A stubborn geometer might insist – like an exotic-travelogue writer who

never actually ventures west of seventh avenue – that only Euclidean geom-
etry is real and that all the various non-Euclidean geometries are merely
curious simulations within it. Such a position is self-consistent, although sti-
fling, for it appears to miss out on the geometrical insights that can arise
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

from the other modes of reasoning. Similarly, a set theorist with the uni-
verse view can insist on an absolute background universe . . . , regarding all
forcing extensions and other models as curious complex simulations within
it. (I have personally witnessed the necessary contortions for class forcing.)
Such a perspective may be entirely self-consistent, and I am not arguing that
the universe view is incoherent, but rather, my point is that if one regards
all outer models of the universe as merely simulated inside it via complex
formalisms, one may miss out on insights that could arise from the simpler
philosophical attitude taking them as fully real. (p. 426)

So, an open question for the universist is how we should interpret the use of
forcing over the universe (including how natural these interpretations are).108
As noted earlier, the multiversist faces no such difficulties. However, they
find themselves in hot water concerning the usual problems of generality

107 Linnebo (2010) is especially concerned with the semantics of plural quantification here, and
I’ve suppressed this detail for clarity.
108 This is a literature I’ve contributed to in Barton (2021) and Antos et al. (2021).
 Iterative Conceptions of Set 83

relativism. They assert that there is no absolute universe, but then immediately
seem to make claims about all universes. The immediate question is: “Why
can’t we just understand this domain as the absolute universe?”. Since the lit-
erature here is enormous, I’ll say no more about it, but merely point out that it
remains open.109
The potentialist does not face these problems. If one believes that one can
always Reify! and Generify! over any definite plurality, and talk about these
processes modally, one does not face the same difficulties. Any definite plu-
rality forms a set, and any definite plurality can be forced over.110 Since the
universe is not modally definite, they may contend that there is no definite plu-
rality of all sets that could be Reified! into a set, and no necessarily uncountable
partial orders that could be fed into Generify!. This is the response of both Lin-
nebo (2010) (for proper classes only) and Scambler (2021) (for both). Given
that their modality is legitimate, a response can be made out along these lines.
An important question is thus whether that modality can be given an accept-
able gloss, or is parasitic on other (unavailable) notions.111 The multiversist and
universist can both explain the modality by reducing it to other notions (direct
quantification over universes for the former, restricted quantification over the
stages for the latter). So there is a real question of whether the potentialist has
just exchanged one suite of problems for another, and whether one is especially
worse.
A final question regarding absolute generality concerns the similarity
between the reasoning involved in Cantor–Russell and Cohen–Scott. Some
authors have argued that the similarity between the two suggests that if one
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

is a Reify! potentialist/multiversist, then one should be a Generify! potential-

ist/multiversist too.112 Really substantiating this thesis would require a more
detailed analysis of the similarities between the two pieces of reasoning, and is
an open philosophical problem.
Note: This seems like a difficult issue to address, since any such response
will have to distinguish both Cantor–Russell and Cohen–Scott from other kinds
of ‘diagonal’ argument where an ‘indefinite extensibility’ response is not so
attractive (e.g. the halting problem; see Meadows (2015)). I do not see an easy
way to answer this question, in particular because it is not clear to me if there

109 For further reading see Rayo and Uzquiano (2006), Florio and Linnebo (2021) (esp. chapter
11), and Studd (2019).
110 There is a question of whether the motivations for these different positions are satisfactory, see
Roberts (MSb).
111 See Linnebo (2018), chs. 3 and 12 for some discussion.
112 See Meadows (2015), Scambler (2021), and Builes and Wilson (2022) for discussion.
 84 The Philosophy of Mathematics

if there is a sharp characterisation of the notion of diagonal argument (perhaps

instead it is a more ‘family resemblance’ concept?).113

10.5 Connection to Conceptual Engineering

One salient point to be noted is that what I’ve argued here is closely linked to
conceptual engineering. This field concerns itself with the evaluation, design,
and implementation of our concepts.114 There are affinities between what
we’ve discussed here and this literature. For example, Kevin Scharp has argued
that our naive concept of truth is inconsistent and should be replaced with two
concepts (ascending truth and descending truth) which validate each direction
of the Tarski biconditionals separately, but there is no consistent concept that
validates both (Scharp, 2013). There are clear similarities with the way in which
Universality and Indefinite Extensibility can be traded off, and how Forcing
Saturation conflicts with Powerset. There is a natural project here to view
these moves in the light of conceptual engineering. Indeed Incurvati (2020)
explicitly makes this connection for Indefinite Extensibility and Universality.
So we ask:

Question Should we view the project of trading off features of con-

cepts/conceptions of set as an exercise in conceptual engineering? If so, how
should we construe the details?

10.6 Plato and Friends

https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

The next objection comes from the staunch set-theoretic realist/platonist, who
thinks that there’s just a world of sets ‘out there’ where every set-theoretic
sentence has a definite truth value. Conceptions of set are great and all, but at
the end of the day the theories they motivate are either true or false about this
universe, and this is the only arbiter of correctness we need. All this talk of
theoretical virtues and conceptions of set is a mere red herring.
I don’t find this line of argument very persuasive at all. I think the history of
set theory, with all its twists and turns, false starts, and possible choice points,
indicates that this just isn’t a very fruitful way to look at things. To see this,
let’s grant for the sake of argument that there is such a platonic realm. What
should we think of our talk concerning it? There is a pessimistic probabilistic
argument available here: Do we really think, out of all the possible conceptions

113 I thank Toby Meadows for some discussion of this point. See also Simmons (1990).
114 See Chalmers (2020) for a survey.
 Iterative Conceptions of Set 85

we might have and all the ways we might have gone and continue to go, that
we will really select the ‘right’ one? I think it entirely possible what we’ve
discussed here is probably a very small snapshot of what is quite a large space.
The conceptions we’ve come across may well just constitute a fraction of all
the possible conceptions available to humans and gods. What is the probability
(given our lack of perceptual interaction with this universe) that we happen to
pick the right conception? I would say low.115
One could, as a response, say that we do have some sort of perception of the
universe of sets. I don’t have much to say here, beyond the well-worn point
that this seems like mysticism to me. Another option is simply a fatalistic pes-
simism about our chances. But I see a better way out – to regard the interesting
questions as ones concerning what we do with our conceptions and the theo-
ries they motivate, and how they interact with our knowledge as a whole. This
strikes me as an area where we can learn and make progress, rather than simply
arguing about whose mystical eye sees the farthest.116

10.7 Pluralism?
I’ve argued that we now find ourselves at a fundamental choice point, do we go
with Forcing Saturation, Powerset, or something else entirely? There is, how-
ever, a different option: We might end up in a situation in which the various
conceptions perform better with respect to certain criteria and/or in different
contexts. It’s possible that we might be led to a strong kind of pluralism, where
claims using the term “set” need to be relativised to a particular kind of concep-
tion in order to be assessed for truth. There’s a special challenge for analysing
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

mathematical practice here. Normally (at least within ZFC set theory) the ‘spec-
tre’ of pluralism does not too radically alter the typing of mathematical objects
(e.g. within different theories extending ZFC the reals are always a set). How-
ever, here we do have significantly different types – the continuum might be
a proper class under the countabilist but a tiny accessible set under the strong
iterative conception. To me, it seems philosophically open which route we take,
or even if we need to pick one. So we ask:

Question What are the prospects for a set-theoretic pluralism arising out of the
different conceptions of set discussed here?

115 I also make a version of this argument in more detail in Barton (2022).
116 This way of thinking has some affinities with Penelope Maddy’s naturalism (Maddy, 1997),
second philosophy (Maddy (2007)), and thin realism (Maddy (2011)). I perhaps differ from her
in that I think that an appealing underlying conception is more than a mere “useful heuristic”
(Maddy (2011), p. 136).
 86 The Philosophy of Mathematics

10.8 Not the Final Word

I hope to have convinced the reader that there are a host of interesting
philosophical and mathematical questions to be found within contemporary
philosophy of set theory. I want to close with a word on the methodology of
progress in this field. We can only hope to make serious advances on these
issues by thoughtful and meticulous examination of different conceptions. A
full study of these problems will thus require a massive effort from historians,
philosophers, and sociologists of mathematics, as well as philosophically inter-
ested mathematicians, and so there’s a real opportunity for collaboration from
people working in many fields. Even then though, it’s not clear how much con-
trol we have over our semantic whims.117 It may be that significant set-theoretic
activism is needed in order to get conceptions accepted as legitimate and under
consideration. In this way, though mathematics has its own norms and meth-
ods of reasoning, the present study suggests a radical anti-exceptionalism about
mathematics as contiguous with other human endeavours. The future is open
and exciting, with a good deal of work to be done in understanding the world(s)
of infinite sets.
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

117 The idea that we don’t have much control is advocated by Wilson (2006) and Cappelen (2018).
 References
Antos, C., Barton, N., and Friedman, S.-D. (2021). Universism and extensions
of V. The Review of Symbolic Logic, 14(1):112–54.
Arrigoni, T., and Friedman, S.-D. (2013). The Hyperuniverse Program. Bulletin
of Symbolic Logic, 19(1):77–96.
Bagaria, J. (2005). Natural axioms of set theory and the continuum problem.
In Proceedings of the 12th International Congress of Logic, Methodol-
ogy, and Philosophy of Science, pages 43–64. King’s College London
Publications.
Barton, N. (2017). Executing Gödel’s Programme in Set Theory. PhD thesis,
Birkbeck, University of London.
Barton, N. (2021). Indeterminateness and ‘the’ universe of sets: Multiversism,
potentialism, and pluralism. In Fitting, M., editor, Selected Topics from
Contemporary Logics, Landscapes in Logic 2, pages 105–82. College
Publications.
Barton, N. (2022). Structural relativity and informal rigour. In Oliveri, G.,
Ternullo, C., and Boscolo, S., editors, Objects, Structures, and Logics:
FilMat Studies in the Philosophy of Mathematics, pages 133–74. Springer
International Publishing.
Barton, N. (MSa). Is (un)countabilism restrictive? To appear in the
Journal of Philosophical Logic. Preprint: https://philpapers.org/rec/
BARIUR.
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Barton, N. (MSb). What makes a ‘good’ modal theory of sets? Manuscript

under review. Preprint: https://philpapers.org/rec/BARWMA-9.
Barton, N., and Friedman, S. (2019). Set theory and structures. In Sarikaya,
D., Kant, D., and Centrone, S., editors, Reflections on the Foundations of
Mathematics, pages 223–53. Springer.
Barton, N., and Friedman, S. (MS). Countabilism and maximality prin-
ciples. Manuscript under review. Preprint: https://philpapers.org/rec/
BARCAM-5.
Barton, N., Müller, M., and Prunescu, M. (2022). On representations of
intended structures in foundational theories. Journal of Philosophical
Logic, 51(2):283–96.
Bell, J. (2011). Set Theory: Boolean-Valued Models and Independence Proofs.
Oxford University Press.
Bell, J. (2014). Intuitionistic Set Theory. Studies in Logic 50. College Publica-
tions.
 88 References

Boolos, G. (1984). To be is to be a value of a variable (or to be some values of

some variables). The Journal of Philosophy, 81(8):430–49.
Boolos, G. (1998). Reply to Charles Parsons’ ‘Sets and Classes’. In Richard
Jeffrey, editor, Logic, Logic, and Logic, pages 30–6. Harvard University
Press.
Brauer, E. (MS). What is forcing potentialism? Manuscript under review.
Builes, D., and Wilson, J. M. (2022). In defense of countabilism. Philosophical
Studies, 179(7):2199–236.
Button, T. (2021a). Level Theory Part 1: Axiomatizing the bare idea of a cumu-
lative hierarchy of sets. The Bulletin of Symbolic Logic, 27(4):436–60.
Button, T. (2021b). Level Theory, Part 2: Axiomatizing the bare idea of a
potential hierarchy. Bulletin of Symbolic Logic, 27(4):461–84.
Button, T. (2022). Level Theory, Part 3: A Boolean algebra of sets arranged in
well-ordered levels. Bulletin of Symbolic Logic, 28(1):1–26.
Caicedo, A. E., Cummings, J., Koellner, P., and Larson, P. B., editors (2017).
Foundations of Mathematics: Logic at Harvard Essays in Honor of W.
Hugh Woodin’s 60th Birthday, volume 690 of Contemporary Mathemat-
ics. American Mathematical Society.
Cain, J. (1995). Infinite utility. Australasian Journal of Philosophy, 73(3):401–
4.
Cappelen, H. (2018). Fixing Language: An Essay on Conceptual Engineering.
Oxford University Press.
Chalmers, D. J. (2020). What is conceptual engineering and what should it be?
Inquiry. DOI: https://doi.org/10.1080/0020174X.2020.1817141
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Cohen, P. (1963). The independence of the continuum hypothesis. Proceed-

ings of the National Academy of Sciences of the United States of America,
50(6):1143–8.
Cohen, P. (2002). The discovery of forcing. Rocky Mountain Journal of
Mathematics, 32(4):1071–100.
Drake, F. R. (1974). Set Theory: An Introduction to Large Cardinals. North
Holland Publishing Co.
Drake, F. R. and Singh, D. (1996). Intermediate Set Theory. Wiley.
Enderton, H. (1977). Elements of Set Theory. Academic Press.
Feferman, S. (2010). On the strength of some semi-constructive set theories. In
Berger, U., Diener, H., Schuster, P., and Seisenberger, M., editors, Logic,
Construction, Computation, pp. 201–226. De Gruyter.
Feferman, S., Friedman, H., Maddy, P., and Steel, J. (2000). Does mathematics
need new axioms? Bulletin of Symbolic Logic, 6(4):401–46.
Feferman, S., and Hellman, G. (1995). Predicative foundations of arithmetic.
Journal of Philosophical Logic, 24(1):1–17.
 References 89

Ferreirós, J. (2007). Labyrinth of Thought: A History of Set Theory and Its Role
in Modern Mathematics (2nd Edition). Springer, Birkhäuser.
Fine, K. (2005). Class and membership. Journal of Philosophy, 102(11):547–
72.
Florio, S., and Leach-Krouse, G. (2017). What Russell should have said to
Burali-Forti. Review of Symbolic Logic, 10(4):682–718.
Florio, S., and Linnebo, O. (2021). The Many and the One: A Philosophical
Study of Plural Logic. Oxford University Press.
Forster, T. (2008). The iterative conception of set. The Review of Symbolic
Logic, 1(1):97–110.
Geschke, S., and Quickert, S. (2004). On Sacks forcing and the Sacks prop-
erty. In Löwe, B., Piwinger, B., and Räsch, T., editors, Classical and
New Paradigms of Computation and their Complexity Hierarchies, pages
95–139, Springer Netherlands.
Giaquinto, M. (2002). The Search for Certainty: A Philosophical Account of
Foundations of Mathematics. Oxford University Press.
Gitman, V., Hamkins, J. D., and Johnstone, T. A. (2016). What is the theory
ZFC without power set? Mathematical Logic Quarterly, 62(4–5):391–
406.
Gödel, K. (1940). The Consistency of the Continuum Hypothesis. Princeton
University Press.
Gödel, K. (1947). What is Cantor’s continuum problem? In [Gödel, 1990],
pages 176–87. Oxford University Press.
Gödel, K. (1964). What is Cantor’s continuum problem? In [Gödel, 1990],
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

pages 254–70. Oxford University Press.

Gödel, K. (1990). Collected Works, Volume II: Publications 1938–1974.
Oxford University Press. Edited by: Solomon Feferman (Editor-in-chief),
John W. Dawson, Jr., Stephen C. Kleene, Gregory H. Moore, Robert M.
Solovay, Jean van Heijenoort.
Goldstein, L. (2012). Paradoxical partners: Semantical brides and set-
theoretical grooms. Analysis, 73(1):33–7.
Hamkins, J. D. (2012). The set-theoretic multiverse. The Review of Symbolic
Logic, 5(3):416–49.
Hamkins, J. D. and Lewis, A. (2000). Infinite time Turing machines. Journal
of Symbolic Logic, 65(2):567–604.
Hamkins, J. D. and Loewe, B. (2008). The modal logic of forcing. Transactions
of the American Mathematical Society, 360(4):1793–817.
Hamkins, J. D. and Montero, B. (2000). With infinite utility, more needn’t be
better. Australasian Journal of Philosophy, 78(2):231–40.
 90 References

Holmes, M., Forster, T., and Libert, T. (2012). Alternative set theories.
In Sets and Extensions in the Twentieth Century, volume 6 of Hand-
book of the History of Logic, pages 559–632. Elsevier/North-Holland,
Amsterdam.
Incurvati, L. (2017). Maximality principles in set theory. Philosophia Mathe-
matica, 25(2):159–93.
Incurvati, L. (2020). Conceptions of Set and the Foundations of Mathematics.
Cambridge University Press.
Incurvati, L. and Murzi, J. (2017). Maximally consistent sets of instances of
naive comprehension. Mind, 126(502):371–84.
Jockwich, S., Tarafder, S., and Venturi, G. (2022). Ideal objects for set theory.
Journal of Philosophical Logic, 51(3):583–602.
Kanamori, A. (2007). Gödel and set theory. The Bulletin of Symbolic Logic,
13(2):153–88.
Kanamori, A. (2009). The Higher Infinite: Large Cardinals in Set Theory from
Their Beginnings. Springer, 2nd edition.
Koellner, P. (2009). On reflection principles. Annals of Pure and Applied Logic,
157(2–3):206–19.
Koellner, P. (2014). Large cardinals and determinacy. In Zalta, E. N., editor,
The Stanford Encyclopedia of Philosophy. Metaphysics Research Lab,
Stanford University, spring 2014 edition.
Krapf, R. (2017). Class forcing and second-order arithmetic. PhD thesis, The
University of Bonn. Universitäts- und Landesbibliothek Bonn.
Kunen, K. (1980). Set Theory: An Introduction to Independence Proofs. Else-
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

vier.
Kunen, K. (2013). Set Theory. College Publications.
Linnebo, Ø. (2006). Sets, properties, and unrestricted quantification. In Rayo,
A., and Uzquiano, G., editors, Absolute Generality, pages 149–78, Oxford
University Press.
Linnebo, Ø. (2010). Pluralities and sets. Journal of Philosophy, 107(3):144–64.
Linnebo, Ø. (2013). The potential hierarchy of sets. The Review of Symbolic
Logic, 6(2):205–28.
Linnebo, Ø. (2014). Plural quantification. In Zalta, E. N., editor, The Stanford
Encyclopedia of Philosophy. Fall 2014 edition.
Linnebo, Ø. (2018). Thin Objects: An Abstractionist Account. Oxford Univer-
sity Press.
Linnebo, Ø., and Shapiro, S. (2023). Predicativism as a form of potentialism.
Review of Symbolic Logic, 16(1):1–32.
Mac Lane, S. (1986). Mathematics: Form and Function. Springer.
Maddy, P. (1983). Proper classes. Journal of Symbolic Logic, 48(1):113–19.
 References 91

Maddy, P. (1988a). Believing the axioms I. The Journal of Symbolic Logic,

53(2):481–511.
Maddy, P. (1988b). Believing the axioms II. The Journal of Symbolic Logic,
53(3):736–64.
Maddy, P. (1997). Naturalism in Mathematics. Oxford University Press.
Maddy, P. (2007). Second Philosophy. Oxford University Press.
Maddy, P. (2011). Defending the Axioms. Oxford University Press.
Maddy, P. (2017). Set-theoretic foundations. In [Caicedo et al., 2017], pages
289–322. American Mathematical Society.
Maddy, P. (2019). What Do We Want a Foundation to Do?, pages 293–311, in
Centrone, S., Kant, D., and Sarikaya, D., editors, Reflections on the Foun-
dations of Mathematics. Synthese Library, vol. 407. Springer International
Publishing, Cham.
Maddy, P., and Meadows, T. (2020). A reconstruction of Steel’s multiverse
project. The Bulletin of Symbolic Logic, 26(2):118–69.
Mancosu, P. (2009). Measuring the size of infinite collections of natural
numbers: Was Cantor’s theory of infinite number inevitable? Review of
Symbolic Logic, 2(4):612–46.
Meadows, T. (2015). Naive infinitism. Notre Dame Journal of Formal Logic,
56(1):191–212.
Menzel, C. (1986). On the iterative explanation of the paradoxes. Philosophical
Studies, 49(1):37–61.
Menzel, C. (2014). ZFCU, wide sets, and the iterative conception. Journal of
Philosophy, 111(2):57–83.
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Menzel, C. (2021). Modal set theory. In Bueno, O. and Shalkowski, S. A.,

editors, The Routledge Handbook of Modality, pages 292–307. Routledge.
Moore, A. W. (1990). The Infinite. Routledge.
Moore, J. T. (2010). The proper forcing axiom. In Proceedings of the Interna-
tional Congress of Mathematicians 2010 (ICM 2010), pages 3–29.
Muller, F. A. (2001). Sets, classes and categories. British Journal for the
Philosophy of Science, 52:539–73.
Oliver, A., and Smiley, T. (2013). Plural Logic. Oxford University Press.
Parker, M. W. (2019). Gödel’s argument for Cantorian Cardinality. Noûs,
53(2):375–93.
Parsons, C. (1983). Sets and modality. In Mathematics in Philosophy: Selected
Essays, pages 298–342. Cornell University Press.
Posy, C. J. (2020). Mathematical Intuitionism. Elements in the Philosophy of
Mathematics. Cambridge University Press.
Potter, M. (2004). Set Theory and Is Philosophy: A Critical Introduction.
Oxford University Press.
 92 References

Priest, G. (2002). Beyond the Limits of Thought. Clarendon Press.

Quine, W. (1937). New foundations for mathematical logic. The American
Mathematical Monthly, 44(2):70–80.
Rayo, A., and Uzquiano, G., editors (2006). Absolute Generality. Clarendon
Press.
Reinhardt, W. (1974). Remarks on reflection principles, large cardinals, and
elementary embeddings. Proceedings of Symposia in Pure Mathematics,
13(2):189–205.
Rittberg, C. J. (2020). Mathematical Practices Can Be Metaphysically Laden,
pages 1–26. Springer International Publishing. DOI: https://doi.org/
10.1007/978-3-030-19071-2_22-1
Roberts, S. (2016). Reflection and Potentialism. PhD thesis, Birkbeck, Univer-
sity of London.
Roberts, S. (2019). Modal structuralism and reflection. The Review of Symbolic
Logic, 12(4):823–60.
Roberts, S. (2022). Pluralities as nothing over and above. Journal of Philoso-
phy, 119(8):405–24.
Roberts, S. (MSa). The iterative conception of properties. Unpublished
manuscript.
Roberts, S. (MSb). Ultimate V. Manuscript under review. Preprint: https://
philarchive.org/rec/ROBUVS.
Scambler, C. (2020). An indeterminate universe of sets. Synthese, 197(2):545–
73.
Scambler, C. (2021). Can all things be counted? Journal of Philosophical
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Logic, 50: 1079–106.

Scambler, C. (MS). On the consistency of height and width potentialism.
Manuscript under review.
Scharp, K. (2013). Replacing Truth. Oxford University Press.
Scott, D. (1977). Foreword to Boolean-Valued Models and Independence
Proofs. In [Bell, 2011], pages xiii–xviii. Oxford University Press.
Shapiro, S. and Wright, C. (2006). All things indefinitely extensible. In Rayo,
A. and Uzquiano, G., editors, Absolute Generality, pages 255–304. Oxford
University Press.
Simmons, K. (1990). The diagonal argument and the liar. Journal of Philosoph-
ical Logic, 19(3):277–303.
Skolem, T. (1922). Some remarks on axiomitized set theory. In A Source Book
in Mathematical Logic, 1879–1931, van Heijenoort 1967, pages 290–301.
Harvard University Press.
Solovay, R. M. (1974). On the cardinality of Σ21 sets of reals. Journal of
Symbolic Logic, 39(2):330–330.
 References 93

Steel, J. (2014). Gödel’s program. In Kennedy, J., editor, Interpreting Gödel,

pp. 153–179. Cambridge University Press.
Studd, J. P. (2013). The iterative conception of set: A (bi-) modal axiomatisa-
tion. Journal of Philosophical Logic, 42(5):697–725.
Studd, J. P. (2019). Everything, More or Less: A Defence of Generality
Relativism. Oxford University Press.
Taranovsky, D. (2004). Perfect subset property for co-analytic sets in
ZF\P. Online article: https://web.mit.edu/dmytro/www/other/Perfect
SubsetsAndZFC.htm.
Uzquiano, G. (2006). Unrestricted unrestricted quantification: The cardinal
problem of absolute generality. In Absolute Generality, pages 305–32.
Oxford University Press.
Weaver, N. (2014). Forcing for Mathematicians. World Scientific.
Whittle, B. (2015). On infinite size. Oxford Studies in Metaphysics, 9:3–19.
Whittle, B. (2018). Size and function. Erkenntnis, 83(4):853–73.
Wilson, M. (2006). Wandering Significance: An Essay on Conceptual
Behaviour. Oxford University Press.
Woodin, W. H. (2017). In search of Ultimate-L: The 19th Midrasha Mathemat-
icae lectures. The Bulletin of Symbolic Logic, 23(1):1–109.
Zarach, A. M. (1996). Replacement ↛ collection, volume 6 of Lecture Notes
in Logic, pages 307–22. Springer.
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press
 Acknowledgements
This short Element came to be as a result of a lot of help from a lot of different
people. My intellectual debts are great, and any list of thanks is going to be
very incomplete. Constraints of space prevent me from going on as long as I
would like. But it’s still good to try.
I got excellent comments on drafts from Salvatore Florio, Marcus Giaquinto,
Sigurd Jorem, Julie Lauvsland, Øystein Linnebo, Toby Meadows, Sam Roberts,
Chris Scambler, Zeynep Soysal, Davide Sutto, Sofie Vaas, Mahan Vaz, Pascal
Wagner, and Daniel Waxman, as well as two anonymous reviewers for Cam-
bridge University Press. I thank all of you heartily for your time, patience, and
helpfulness.
In addition, there are many people with whom I have had very useful con-
versations regarding ideas behind the text, and whose words helped shape
the Element. I can remember having especially fruitful interactions with Car-
olin Antos, Joan Bagaria, Robert Black, Tim Button, Laura Crosilla, Raphaël
Carroy, Neil Dewar, Alexander Douglas, Monroe Eskew, Juliet Floyd, Sy-
David Friedman, Victoria Gitman, Olav Gjelsvik, Joel-David Hamkins, Luca
Incurvati, Akihiro Kanamori, Daniel Kuby, Maxwell Levine, Penelope Maddy,
Beau Madison Mount, Rupert McCallum, Colin McLarty, Sandra Müller,
Michael Potter, Agustín Rayo, Ian Rumfitt, Ethan Russo, Jeffrey Schatz, Stew-
art Shapiro, Robin Solberg, James Studd, Fenner Tanswell, Philip Welch, and
Kameryn J. Williams. My apologies to the large number of people whom I have
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

forgotten.
Versions of this work were presented at University of Oslo, University
of Konstanz, University of California, Davis, and the National University of
Singapore. I thank those audiences for their input.
I would like also to thank Penny Rush and Stewart Shapiro for their editorial
work and for bringing together the Elements in the Philosophy of Mathematics
series. Shalini Bisa and Hilary Gaskin were very patient and helpful with the
administrative side of things, and I thank them for their hard work.
I’ve been very lucky to be part of some great institutions. Birkbeck provided
a fantastic place to do my graduate work. The Kurt Gödel Research Center for
Mathematical Logic was a hive of logical activity whilst I was there. Konstanz
was both a very supportive and energetic place to do philosophy. And though I
haven’t been there long, I’m having a great time in Oslo and enjoying myself
immensely.
 Acknowledgements 95

I’m grateful also for the financial support of the UK Arts and Humanities
Research Council whilst doing my PhD, the FWF (Austrian Science Fund)
during my stay in Vienna, the VolswagenStiftung whilst I was in Konstanz,
and Norges Forskningsrådet (Research Council of Norway) who provide my
current funding.
I owe much to my partner Nikki. She’s been a constant source of inspiration,
and if this Element holds up halfway to the standards she sets for herself, I’ll
be very happy.
Lastly, I dedicate this Element to my parents Jeanne and Paul, who have
supported my curiosity for as long as I can remember.
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press
 The Philosophy of Mathematics

Penelope Rush
University of Tasmania
From the time Penny Rush completed her thesis in the philosophy of mathematics
(2005), she has worked continuously on themes around the realism/anti-realism divide
and the nature of mathematics. Her edited collection The Metaphysics of Logic
(Cambridge University Press, 2014), and forthcoming essay ‘Metaphysical Optimism’
(Philosophy Supplement), highlight a particular interest in the idea of reality itself and
curiosity and respect as important philosophical methodologies.

Stewart Shapiro
The Ohio State University
Stewart Shapiro is the O’Donnell Professor of Philosophy at The Ohio State University, a
Distinguished Visiting Professor at the University of Connecticut, and a Professorial
Fellow at the University of Oslo. His major works include Foundations without
Foundationalism (1991), Philosophy of Mathematics: Structure and Ontology (1997),
Vagueness in Context (2006), and Varieties of Logic (2014). He has taught courses in logic,
philosophy of mathematics, metaphysics, epistemology, philosophy of religion, Jewish
philosophy, social and political philosophy, and medical ethics.

About the Series

This Cambridge Elements series provides an extensive overview of the philosophy of
mathematics in its many and varied forms. Distinguished authors will provide an
up-to-date summary of the results of current research in their fields and give their own
take on what they believe are the most significant debates influencing research, drawing
original conclusions.
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press
 The Philosophy of Mathematics

Elements in the Series

Mathematical Anti-Realism and Modal Nothingism
Mark Balaguer
Plato Was Not a Mathematical Platonist
Elaine Landry
Mathematics and Explanation
Christopher Pincock
Indispensability
A. C. Paseau and Alan Baker
Lakatos and the Historical Approach to Philosophy of Mathematics
Donald Gillies
Phenomenology and Mathematics
Michael Roubach
Philosophical Uses of Categoricity Arguments
Penelope Maddy and Jouko Väänänen
Number Concepts
Richard Samuels and Eric Snyder
The Euclidean Programme
A. C. Paseau and Wesley Wrigley
Mathematical Rigour and Informal Proof
Fenner Stanley Tanswell
https://doi.org/10.1017/9781009227223 Published online by Cambridge University Press

Mathematical Pluralism
Graham Priest
Iterative Conceptions of Set
Neil Barton

A full series listing is available at: www.cambridge.org/EPM