Michael Huemer (Auth.) - Approaching Infinity-Palgrave Macmillan UK (2016)

Approaching Infinity
Also by Michael Huemer
ETHICAL INTUITIONISM
THE PROBLEM OF POLITICAL AUTHORITY
SKEPTICISM AND THE VEIL OF PERCEPTION

Approaching Infinity
Michael Huemer
Professor of Philosophy, University of Colorado at Boulder, USA
© Michael Huemer 2016
All rights reserved. No reproduction, copy or transmission of this
publication may be made without written permission.
No portion of this publication may be reproduced, copied or transmitted
save with written permission or in accordance with the provisions of the
Copyright, Designs and Patents Act 1988, or under the terms of any licence
permitting limited copying issued by the Copyright Licensing Agency,
Saffron House, 6–10 Kirby Street, London EC1N 8TS.
Any person who does any unauthorized act in relation to this publication
may be liable to criminal prosecution and civil claims for damages.
The author has asserted his right to be identified as the author of this work
in accordance with the Copyright, Designs and Patents Act 1988.
First published 2016 by
PALGRAVE MACMILLAN
Palgrave Macmillan in the UK is an imprint of Macmillan Publishers Limited,
registered in England, company number 785998, of Houndmills, Basingstoke,
Hampshire RG21 6XS.
Palgrave Macmillan in the US is a division of St Martin’s Press LLC,
175 Fifth Avenue, New York, NY 10010.
Palgrave Macmillan is the global academic imprint of the above companies
and has companies and representatives throughout the world.
Palgrave® and Macmillan® are registered trademarks in the United States,
the United Kingdom, Europe and other countries.
ISBN 978-1-137-56086-5 ISBN 978-1-137-56087-2 (eBook)
DOI 10.1057/9781137560872
This book is printed on paper suitable for recycling and made from fully
managed and sustained forest sources. Logging, pulping and manufacturing
processes are expected to conform to the environmental regulations of the
country of origin.
A catalogue record for this book is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Huemer, Michael, 1969– author.
Title: Approaching infinity / Michael Huemer.
Description: New York : Palgrave Macmillan, 2016. | Includes index.
Identifiers: LCCN 2015040546| ISBN 9781137560865 (pbk.)
Subjects: LCSH: Infinite.
Classification: LCC BD411 .H84 2016 | DDC 111/.6—dc23
LC record available at http://lccn.loc.gov/2015040546
For Iskra
This page intentionally left blank
Contents
List of Figures xi
Preface xii
Part I The Need for a Theory of Infinity
1 The Prevalence of the Infinite 3

1.1 The concept of infinity and the infinite 3
1.2 The infinite in mathematics 4
1.3 The infinite in philosophy 5
1.4 The infinite in the physical world 6
1.5 The infinite in modern physics 7
1.6 Controversies 7
2 Six Infinite Regresses 9

2.1 The regress of causes 9
2.2 The regress of reasons 10
2.3 The regress of forms 11
2.4 The regress of resemblances 12
2.5 The regress of temporal series 14
2.6 The regress of truths 15
2.7 Conclusion 16
3 Seventeen Paradoxes of the Infinite 17

3.1 A word about paradoxes 17
3.2 The arithmetic of infinity 17
3.3 The paradox of geometric points 18
3.4 Infinite sums 20
3.5 Galileo’s paradox 21
3.6 Hilbert’s hotel 22
3.7 Gabriel’s horn 23
3.8 Smullyan’s infinite rod 24
3.9 Zeno’s paradox 25
3.10 The divided stick 27
3.11 Thomson’s lamp 28
3.12 The Littlewood-Ross Banker 29
3.13 Benardete’s paradox 31
vii
viii Contents
3.14 Laraudogoitia’s marbles 32

3.15 The spaceship 33
3.16 The Saint Petersburg paradox 34
3.17 The Martingale betting system 35
3.18 The delayed heaven paradox 36
3.19 Conclusion 38
Part II Old Theories of Infinity
4 Impossible Infinite Series: Two False Accounts 41

4.1 ‘An infinite series cannot be completed by
successive synthesis’ 41
4.2 ‘An infinite series of preconditions cannot be satisfied’ 44
4.3 Conclusion 47
5 Actual and Potential Infinities 49

5.1 The theory of potential infinity 49
5.2 Why not actual infinities? 51
5.3 Infinite divisibility 52
5.4 Infinite time 57
5.5 Infinite space 57
5.6 Infinitely numerous numbers 65
5.7 Infinitely numerous abstract objects 67
5.8 Infinitely numerous physical objects 68
5.9 Conclusion 69
6 The Cantorian Orthodoxy 71

6.1 The importance of Georg Cantor 71
6.2 Sets 71
6.3 Cardinal numbers 73
6.4 ‘Greater’, ‘less’, and ‘equal’ 74
6.5 Many sets are equally numerous 75
6.6 The diagonalization argument 78
6.7 Cantor’s theorem 80
6.8 The paradoxes of set theory 82
6.9 Other paradoxes of infinity 85
6.10 Conclusion 89
Part III A New Theory of Infinity and Related Matters
7 Philosophical Preliminaries 93
7.1 Metapreliminaries 93
7.2 Phenomenal conservatism 95
Contents ix
7.3 Synthetic a priori knowledge 97

7.4 Metaphysical possibility 100
7.5 Possibility and paradox 105
7.6 A realist view of mathematics 106
8 Sets 108
8.1 Sets are not collections 108
8.2 Sets are not defined by the axioms 110
8.3 Many regarded as one: the foundational sin? 111
8.4 The significance of the paradoxes 113
8.5 Are numbers sets? 114
8.6 Set theory and the laws of arithmetic 116
9 Numbers 119
9.1 Cardinal numbers as properties 119
9.2 Frege’s objection 120
9.3 Arithmetical operations 122
9.4 The laws of arithmetic 123
9.5 Zero 124
9.6 A digression on large numbers 127
9.7 Magnitudes and real numbers 129
9.8 Indexing uses of numbers 137
9.9 Other numbers 138
10 Infinity 143
10.1 Infinity is not a number 143
10.2 Infinite cardinalities 148
10.3 Infinite extensive magnitudes 150
10.4 Infinite intensive magnitudes 151
10.5 Some a priori physics 158
11 Space 162
11.1 Pointy space versus gunky space 162
11.2 The unimaginability of points 163
11.3 The zero argument 164
11.4 When zero is not mere absence 165
11.5 The paradox of contact 168
11.6 The problem of division 170
11.7 The dimensionality of space is necessary 171
11.8 The measure-theoretic objection 173
12 Some Paradoxes Mostly Resolved 176
12.1 The arithmetic of infinity 176
x Contents
12.2 The paradox of geometric points 176

12.3 Infinite sums 178
12.4 Galileo’s paradox 180
12.7 Smullyan’s infinite rod 184
12.8 Zeno’s paradox 185
12.10 Thomson’s lamp 194
12.11 The Littlewood-Ross Banker 201
12.12 Benardete’s paradox 207
12.13 Laraudogoitia’s marbles 211
12.14 The spaceship 211
12.15 The Saint Petersburg paradox 213
12.16 The Martingale betting system 218
12.17 The delayed heaven paradox 220
12.18 Comment: shallow and deep impossibilities 223
13 Assessing Infinite Regress Arguments 229

13.1 The problem of identifying vicious regresses 229
13.2 Viciousness through metaphysical impossibility 230
13.3 Viciousness through implausibility 232
13.4 Viciousness through explanatory failure 236
13.5 Conclusion 245
14 Conclusion 247
14.1 Why study infinity? 247
14.2 Troubles with traditional approaches 248
14.3 A new approach to infinity 248
14.4 Some controversial views about sets, numbers,
and points 249
14.5 Solving the paradoxes 250
14.6 For further reflection, or: what is wrong
with this book? 251
References 261
Index 269
List of Figures
3.1 A magically expanding rod 20

3.4 Smullyan’s rod 25
5.1 Relationship between the side and the diagonal of a square 56
5.2 Three lines connecting two points 59
5.3 Newton’s bucket 63
6.1 Cantor’s theorem 81
9.1 Three instances of twoness 119
10.1 Infinite temperature and negative temperature 156
10.2 An impossible Newtonian scenario 159
11.1 A line and a triangle 165
11.2 Two ways of dividing a stick in half 171
11.3 The Cantor lines 174
12.1 A magically expanding rod 177
12.2 Dividing a stick with infinitely many knives 190
12.3 Cutting a material 193
12.4 A rotary switch for Thomson’s lamp 199
12.5 The Banker paradox with diminishing bills 203
12.6 Benardete’s walls 208
14.1 An impossible Newtonian scenario 252
xi
Preface
For many years, I have made a point of confounding my philosophy

students with a variety of paradoxes, including several paradoxes of the
infinite. And for many years, I tried myself to think through these para-
doxes, without success. I list seventeen of these paradoxes in Chapter 3.
But over and above the seventeen paradoxes, there were three philo-
sophical questions about the infinite that exercised me.
Here is the first issue. It seems that there are some infinite series that
can be completed. For instance, for an object to move from point A to
point B, it must first travel half the distance, then half the remaining
distance, then half the remaining distance, and so on. According to one
famous argument, because this is an infinite series, the object can never
reach point B. The proper response seems to be to insist that one can in
fact complete an infinite series.
On the other hand, it seems that there are other infinite series that
cannot be completed. For instance, imagine a lamp that starts out on,
then is switched off after half a minute, then back on after another
quarter minute, then off after another eighth of a minute, and so on.
At the end of one minute, is it on or off? The proper response seems
to be that one could not complete such an infinite series of switch-
ings. We might think this is because an infinite series, by definition,
is endless, and one cannot come to the end of something that is
endless.
In other words, the solution to the first puzzle (‘one can complete an
infinite series’) seems to be the opposite of the solution to the second
puzzle (‘one cannot complete an infinite series’). That’s puzzling. What
we would like to say is that some infinite series are possible, and others
are impossible. But why? What is the difference between the infinite
series of halfway-motions, and the infinite series of lamp-switchings?
That’s the first philosophical question I wanted to answer.
This puzzle is tied up with a popular genre of arguments in philosophy:
it is common to argue that some philosophical theory must be rejected
because it leads to an infinite regress. For instance, if every event has a
cause, and the cause of an event is also an event, then there must be an
infinite regress of causes. To avoid this, some say, we should reject the
idea that everything has a cause; instead, we should posit a ‘first cause’,
something that caused everything else and was itself uncaused.
xii
Preface xiii
But almost as common as infinite regress arguments is a certain type of

response, which claims that there is nothing wrong with the infinite regress.
For instance, some say that we should simply accept that there is an infinite
series of causes stretching into the past forever. Now, there seems to be wide
agreement among philosophers that some but not all infinite regresses are
bad; but there has been no consensus on which regresses are bad (‘vicious’)
and which benign. Hence, the second philosophical question I wanted to
answer: What makes an infinite regress vicious or benign?
Finally, I started reflecting on infinities in science, where there seems
to be an analogous issue. Some infinities are considered ‘bad’ – for
instance, the infinite energy density and infinite spacetime curvature of
a black hole are considered problems in astrophysics. They are thought
to indicate a breakdown of accepted theories (notably, general relativity),
and astrophysicists have been trying to devise new theories that elim-
inate these infinities. But there are other infinities that no one seems
worried about: no one seems to consider the notion of an infinitely large
universe, or the notion of an infinite future, to be problematic. Why is
this? What makes some infinite quantities in a theory problematic while
others are perfectly acceptable?
One can see that these three questions – ‘Why are some infinite series
completeable and others not?’, ‘Why are some infinite regresses vicious
and others not?’, and ‘Why are some infinite quantities problematic and
others not?’ – are quite similar, and so perhaps they have a common
answer. After several years of puzzlement concerning the first two ques-
tions, I finally tried to connect them with the third question. It was then
that I thought of a theory that seems to me to account for which sorts
of infinities are possible and which impossible. After publishing one
paper on the subject (‘Virtue and Vice among the Infinite’), I decided to
expand my ideas into this book.
I would like to thank Stuart Rachels for the conversations about
the infinite that led to this expansion. In addition, I am grateful
for the comments of Adrian Moore and Matt Skene on earlier drafts of
the manuscript, and the questions and comments of Peter Klein, Ted
Poston, Jeanne Peijnenberg, David Atkinson, and the other participants
of the Infinite Regress Workshop organized by Scott Aiken at Vanderbilt
University in October, 2013, where I presented my earlier paper on the
subject. If there are any errors in this book, as surely there must be, they
are the fault of one of these other philosophers – most likely Peter Klein,
because he supervised my PhD and should have trained me better.
Part I
The Need for a Theory of Infinity
1
The Prevalence of the Infinite
Two things are infinite: the universe and human stupidity.

– Albert Einstein/Frederick S. Perls1
1.1 The concept of infinity and the infinite
What is the infinite? The infinite is sometimes described as that which

is unlimited, boundless, or so great that it is impossible for anything
to be greater. For instance, an infinitely heavy stone would be a stone
so heavy that it would be conceptually impossible for anything to be
heavier. (Whether there could be such a stone is another matter.) An
infinite area would be an area such that no area could be larger.
What about the noun ‘infinity’ – what does this refer to? Infinity is
commonly thought of as a very large number – a number larger than all
the other numbers; or, a number so large that if you add one to it, the
result is no larger than you started with. Another way to get at the idea
is to define infinity by reference to some particular thing that is infi-
nite. For instance, we might say: infinity is the number that describes
how many natural numbers there are. Or: infinity is the number that
describes how big all of space is.
1
This quotation comes from the Gestalt psychotherapist Frederick Perls (1947,
111), who attributes it to ‘a great astronomer’ then goes on to say that Einstein has
shown that the universe is not infinite. Later, Perls (1969, 33) modified the quota-
tion and ascribed it to Einstein, who is supposed to have said it in conversation
with Perls (the original version contained ‘as far as we know’ after ‘infinite’).
3
4 Approaching Infinity
Much later in this book, we see why the above characterizations are
inadequate and what a more adequate characterization looks like. But
that requires much more work, so for now, I leave the reader with the
above characterizations, which at least point at the intuitive conception
of infinity and the infinite.
Why is infinity interesting? Three reasons: first, the infinite is an
important part of reality. Second, many important philosophical
arguments invoke the concept of infinity, infinite regresses in partic-
ular. Third, infinity is extremely puzzling, and we have yet to attain a
clear grasp of it. We begin in this chapter with the first of these three
points; the second and third points are elaborated in Chapters 2 and 3,
respectively.
1.2 The infinite in mathematics
Pace Einstein, more than two things are infinite. In mathematics, there
are a variety of infinite classes of abstract objects. The most familiar of
these is the set of counting numbers (the numbers 1, 2, 3, and so on).
There is no last member of this set; no matter how high you count,
there are always more counting numbers. And the point is of course not
one about human counting capacities; the main point of interest is one
that would hold even if there were no conscious beings: for any natural
number n, there is another greater than it.
Similarly, the real numbers (which include whole numbers in addition
to fractions and irrational numbers) are infinitely numerous, since again,
for any real number r, there are real numbers larger than r. Indeed, if you
pick any two real numbers, there are infinitely many others sandwiched
between them. For example, between the numbers 2 and 2.1, there are
the numbers 2.01, 2.011, 2.0111, and so on.
These numbers are the least controversial; anyone who believes in
numbers at all believes in the reals. Negative numbers used to be contro-
versial since it seemed odd that there should be a quantity less than
nothing. Potentially controversial classes of numbers include imaginary
numbers, infinitesimals, hyperreals, and transfinite cardinals. Each of
these kinds of number, if they exist at all, are also infinitely numerous.
If there are imaginary numbers, then there are infinitely many of them
(i, −i, 2i, −2i, ... ); if there are infinitesimals, then there are infinitely
many of them; and so on.
That is just to speak of numbers. Other mathematical objects, if they
exist at all, are also infinitely numerous. For instance, if sets exist, then
there are the sets {1}, {2}, {1,2}, {3}, {1,3}, {2,3}, {1,2,3}, and so on. Or, if
The Prevalence of the Infinite 5
you prefer sets built from concrete elements, there are {the moon}, {{the
moon}}, {{the moon}, the moon}, {{{the moon}}}, and so on. Almost any
other type of mathematical object will also be infinitely numerous, if
that type of object exists at all (for example, vectors, groups, functions,
points, spaces).
1.3 The infinite in philosophy
Philosophers often discuss propositions, which are usually thought of as

a special kind of abstract object. A proposition is the sort of thing that
can be either true or false; however, propositions are to be distinguished
both from statements and from beliefs. When you have a belief, the thing
that you believe to be the case is a proposition. Multiple people can
believe the same thing (the same proposition). A person can also have
multiple attitudes toward the same proposition; for instance, you may
both believe and regret that you have eaten all the coffee ice cream.
The phrase ‘that you have eaten all the coffee ice cream’ denotes a
proposition. There can also be multiple sentences that express the same
proposition; for instance, ‘It is raining’ and ‘Il pleut’ (the French transla-
tion of ‘It is raining’) both express the proposition that it is raining.2 If
propositions exist, then there are infinitely many of them. There are,
for instance, the propositions that 2 is greater than 1, that 3 is greater
than 1, that 4 is greater than 1, and so on.
Then there are universals. A ‘universal’ is something that multiple
things could have in common, in the sense of something that could be
predicated of them. For instance, in my refrigerator I have a tomato and
a chili pepper. Both are red; in other words, redness is one of the things
they have in common. So redness is a universal. Note, incidentally, that
redness would be a universal even if there were fewer than two red things
in the universe: redness is a universal because it is the sort of thing that
multiple things could have in common, even if in fact multiple things
don’t have this in common. (I don’t know why philosophers call these
things ‘universals’. The name ‘universal’ makes it sound as though they
are shared by everything, but in fact all that is required is that they are
capable of being shared by more than one thing.)
2
This may need to be relativized to a conversational context – that is, ‘it is
raining’ expresses the proposition that it is raining at the time and place made
salient by the context of the utterance. That it rains at some other place or time
would be a distinct proposition.
How many universals are there? Infinitely many. Every color is a

universal – not just the familiar colors such as red, green, and blue, but
every shade of color, however specific, including the shades for which
English lacks a name. Each of these colors is something that multiple
things could have in common. Similarly, every shape is a universal – not
just the familiar shapes such as round, square, and triangular, but every
irregular shape, including the ones that have no names in English, and
including those shapes such that no physical object is actually shaped
that way. Each of these shapes is something that multiple objects could
have in common. Similarly, every possible size is a universal; every type
of emotional state is a universal (for example, being happy, being afraid,
being angry); and so on.
Finally, many philosophers have held that there is one particularly
interesting entity that is infinite, namely, God. God has been thought to
be infinite in a variety of respects – to have unlimited power, unlimited
knowledge, to be present everywhere at all times, and so on. According
to Descartes, the word ‘infinite’ strictly applies ‘only to that in which
no limits of any kind can be found; and in this sense God alone is
infinite.’3
1.4 The infinite in the physical world
Physical space – the space that we occupy and move around in – is infi-
nite in two ways. First, space is infinitely extended (it extends infinitely far
in all directions). Imagine traveling away from the Earth. No matter how
far away you went, it would always be possible to go farther. You would
never, so to speak, come to the edge of space, like some giant wall. And
the real point here is not about our traveling capabilities; the point of
interest is that there are places arbitrarily far from here. That is, for any
distance, d, there are places farther from here than d (even if there are no
physical objects in those places).
Second, space is infinitely divisible. If you have a line segment, for
example, you can divide the segment into a left half and a right half.
Then the left half can itself be divided into a left half (the leftmost
quarter of the original segment) and a right half. Then that left half
(the leftmost eighth of the original segment) can be divided into a left
half and a right half. And so on. Again, the point here is not one about
human dividing capabilities; the real point is that for any two points,
there are other points in between them, which entails that there are
3
Descartes 1984, 81. Compare Aquinas 1981, I.Q7.
The Prevalence of the Infinite 7
infinitely many points, and infinitely many line segments, within any
given line segment.
The same observations apply to time. Time appears to extend infi-
nitely in both directions. No matter how far into the past one looks,
there were moments earlier than that. No matter how far into the future
one looks, there will be moments after that. There may one day come an
end of humanity, or an end of the Earth, or even an end of our material
universe, but there is not going to come an end of time.
Like space, time is infinitely divisible. Take a one-minute interval of
time. It can be divided into the first half and the last half. Then the first
half can itself be divided into a first half (the first 15 seconds of the orig-
inal interval) and a second half. Then that first half can be divided into
a first half (the first 7.5 seconds of the original interval) and a second
half. And so on. Every time interval contains infinitely many smaller
intervals, and infinitely many instants.
1.5 The infinite in modern physics
In modern physics, some infinite quantities have been theorized but

remain controversial. A black hole, as described by the general theory of
relativity, is a region of infinite density (a positive amount of mass-en-
ergy is concentrated in zero volume) and infinite spacetime curvature.
The same is true of the Big Bang singularity in which the universe is
thought to have originated. However, the appearance of these infinities
is generally regarded as signaling a breakdown of the theory (general
relativity) under these conditions. To accurately describe a black hole
or the conditions at the beginning of the universe, one would have to
take account of quantum mechanics, and there is as yet no accepted
quantum mechanical account of gravitation.4 Scientists working on
quantum gravity are seeking ways to eliminate the infinities from the
theory.
1.6 Controversies
Some (perhaps all) of the above examples are controversial. Some people
would deny that all of these are genuine examples of existing infinities,
either because they deny that the things I am discussing really exist, or
because they think the things I am discussing are finite. For instance,
some would deny that there are infinitely many sets, because they deny
4
Wald 1984, 211–12; Hamade 1996.
that sets exist at all. Some would deny that the past is infinite, because
they think there was a beginning of time. I address some of these contro-
versies later, when I consider certain philosophical theories that have
led people to deny the existence of actual infinities. For now, I hope
simply to have made a prima facie case that infinity is an important
aspect of reality, worthy of investigation. If you accept at least some of
my examples, this should suffice.
2
Six Infinite Regresses
It is commonly thought that there are some kinds of infinite series that
one should avoid in one’s philosophical theories – infinite series such
that, if a theory requires one of them, then the theory should therefore
be rejected. These are referred to as ‘vicious infinite regresses’. On the
other hand, it is also commonly thought that not all infinite series need
be avoided; sometimes a theory entails the existence of an infinite series,
but the theory should not therefore be rejected. These are referred to as
‘virtuous’ or (less misleadingly) ‘benign infinite regresses’.
Infinite regress arguments are common in philosophy. These are
arguments that claim that a certain philosophical doctrine should be
rejected because the doctrine leads to a vicious infinite regress, or that
a certain philosophical doctrine should be accepted in order to avoid a
vicious regress.
What makes an infinite regress virtuous or vicious? This remains a
matter of controversy, and one purpose of this book is to work toward a
satisfying answer to this question. But we will be in a better position to
address the question after we have looked at some examples of infinite
regresses, as well as a variety of problems involving the infinite. In the
present chapter, I confine myself to listing several infinite regresses that
appear in philosophy, with the aim of evincing the need for an account
of virtue and vice among the infinite.
2.1 The regress of causes
Perhaps the most famous infinite regress argument is the First Cause
Argument for the existence of God (also known as the Cosmological
Argument). We find in the world many events that are linked by
9
relations of cause and effect. Suppose some present event, A, was caused
by B. Then we can ask what caused B. If B was caused by C, then we can
ask what caused C. If C was caused by D, then we can ask what caused
D. And so on.
But, the argument goes, there cannot be an infinite regress of causes. If
there were, this would mean that an infinite series had to be completed
in order for event A to occur; but an infinite series cannot be completed
(compare the absurd idea of counting to infinity); therefore, event A
would not have been possible. To avoid the infinite regress, we must
postulate a ‘first cause’, something that started the chain of causation
and was not itself caused by anything.1
How one gets from the idea of a first cause to the idea of God is less
clear. As far as the above reasoning goes, the first cause need not be
unique, that is, there might be more than one uncaused cause; it need
not be a conscious being, let alone an all-knowing, all-powerful, morally
good being; and it need not even exist in the present (the first cause or
causes might have existed only in the distant past). Nevertheless, the
theist would wish to argue that God is the most plausible candidate for
a first cause.
Our concern here, however, is whether the argument for a first cause
succeeds, not whether the first cause is God or something else. Even
before God is introduced, the first cause argument is controversial.
While some philosophers find the idea of an infinite series of causes
stretching back infinitely into the past paradoxical, others find the idea
unproblematic.2 This is to say that there is no general agreement on
whether the infinite regress of causes is vicious or benign.
2.2 The regress of reasons
Suppose you believe some proposition, A, and assume that this is a

rationally justified belief. We can then ask what reason you have to
believe A. Now suppose that your reason for believing A is B. It seems
that, for B to justify A, you must also be justified in believing B. So
we can ask why you believe B. Now suppose your reason for believing
B is C. Again, it seems that C must also be justifiably believed, so we
can ask for your reason for believing C. But this process cannot go on
forever; you cannot actually have an infinite series of reasons for belief.
Therefore, there must be a point at which you have a justified belief, but
1
Aquinas 1981, I.Q2.A3; Kant 1965, B454; Craig 1991.
2
Russell 1961, 587.
Six Infinite Regresses 11
that belief is not based on any reasons. This is referred to as a ‘founda-

tional’ belief, and the people who believe that there are foundational
beliefs are referred to as ‘foundationalists’.
The preceding paragraph summarizes the leading argument for foun-
dationalism. It is due in part to that argument that foundationalism has
been the dominant position in the history of philosophy. Another way
of stating the basic idea is this: we must posit foundational beliefs in
order to avoid a vicious regress of reasons.3
Recently, however, some philosophers have questioned whether the
regress of reasons is really vicious. This position is known as ‘infinitism’.4
Infinitists agree that there are justified beliefs, they agree that circular
reasoning must be avoided, but they disagree that any beliefs are founda-
tional; instead, they hold that the infinite regress of reasons is a virtuous
regress. Traditionally, foundationalists argue that the regress of reasons
is vicious because no human being could complete an infinitely long
chain of reasoning. In response, infinitists maintain that justified belief
only requires that the believer be capable of providing a reason for a
given belief, if and when the belief is challenged; it does not require that
the believer have actually provided a reason for every belief. Thus, there
is no requirement of completing an infinitely long chain of reasoning.
In addition, infinitists maintain that, while it may be impossible for an
individual to have infinitely many beliefs occurring in their conscious-
ness at one time, it is possible for a person to have infinitely many dispo-
sitional beliefs; these are beliefs that one counts as holding by virtue of
one’s dispositions, rather than what one is thinking at the moment.
2.3 The regress of forms
Plato’s best known idea was his theory of universals, the Theory of the
Forms.5 This theory held that there is a realm of non-physical, abstract
objects, the ‘Forms’. There are Forms for every property – or at least every
important property6 – that objects might have in common. Thus, there
are Forms of Roundness, Largeness, Humanness, Beauty, Goodness, and
so on. Furthermore, Plato appears to have held that the Form of F (where
3
Aristotle 1941, Posterior Analytics I.2; Huemer 2010.
4
Infinitism was pioneered by my old grad school advisor, Peter Klein (1999;
2003). See also Turri and Klein 2014.
5
See the Meno (Plato 2012) and The Republic (Plato 1974), 514a–518d. For
Plato’s trenchant criticisms of his own theory, see the Parmenides in Plato 1931.
6
See the Parmenides (Plato 1931) at 130c–d, where Socrates denies the existence
of Forms for such things as hair, mud, and filth.
F is some property) is a perfect example of F. For instance, the Form of

Roundness is a perfect circle, existing in some non-physical realm. All
circles in the physical, observable world are but imperfect approxima-
tions to the Form of Roundness.
There is a famous argument (at least it is famous among philoso-
phers – which really doesn’t make it famous at all) against the Theory
of Forms, known as the Third Man Argument.7 Suppose there are two
men, say Homer and Carl. According to Plato, to explain what Homer
and Carl have in common, we have to posit the Form of Man. This
Form of Man is a perfect man, just as the Form of Roundness is a perfect
circle. (This is already quite bizarre.) The Form of Man will therefore
have something in common with Homer and Carl – it, too, is a man.
Therefore, to explain what the Form of Man has in common with the
ordinary, physical men, we will have to posit another Form. This second
Form will also presumably be a perfect example of manhood, so it will
have something in common with Homer, Carl, and the first Form of
Man. Thus begins an infinite regress.
Unlike the previous two cases, the viciousness of this regress is not
very controversial – no one has tried to defend the idea that there really
is this infinite series of Forms of Man. Contemporary thinkers avoid the
regress by denying that the property of being F is itself an instance of
F. For example, even if there is a universal of Manhood, that universal
is not itself a man; hence, there is no occasion to introduce a second
universal of Manhood.8
2.4 The regress of resemblances
In contrast to Plato, many philosophers have held that universals do

not exist, or that there are no abstract objects; there are only concrete,
particular things.9 This sort of view is called ‘nominalism’.
Nominalists then owe us an explanation for why it seems as though
there are universals. For example, there might be three blue things – say,
the sky, a blueberry, and a blue parrot. It seems as though there is some-
thing they have in common – or as we might say, there is a respect in
which the three objects are the same – namely, their blueness. Blueness
7
See Plato 1931, Parmenides 132a–b, where the argument is given using the
Form of the Large. Aristotle (1941) appears to have changed the example to the
Form of Man (Metaphysics 990b17, 1039a2; Sophistical Refutations 178b36ff.).
8
Armstrong 1978, 71.
9
Hume 1992, 17–25; Ockham 1994.
is not a particular, concrete thing; it is a universal. So it seems that there

is at least one universal.
Here is one way in which nominalists have responded; the following
view is known as ‘resemblance nominalism’: there are three distinct
bluenesses, (A) the blueness of the sky, (B) the blueness of the berry,
and (C) the blueness of the parrot. Each of these is a property confined
to a specific object at a certain place and time; none of them is repeat-
able. Furthermore, the three bluenesses do not themselves literally have
anything in common; they are not the same in any respect. Rather, they
are merely similar, that is they resemble each other to some degree, and
this is why we apply the word ‘blue’ to all of them.10 Resemblance itself
is taken as a basic, unanalyzed notion. (Obviously, the nominalist could
not explain the ‘resemblance’ between two objects in terms of their
sharing some number of properties; this would undermine their aim of
avoiding the need to posit universals.)
David Armstrong has presented a variety of infinite regress argu-
ments against a variety of nominalist theories. One of these arguments,
directed against resemblance nominalism, begins by pointing out that
resemblance itself seems to be a universal. The blueness of the sky resem-
bles that of the berry, which resembles that of the parrot, which resem-
bles that of the sky. So there are at least three examples of resemblance.
Since resemblance can thus be ascribed to more than one pair, resem-
blance is a universal. So it would seem that, just as the nominalist owed
us an explanation for the fact that three objects may all be blue, the
nominalist will also owe us an explanation for the fact that three pairs
may all resemble.
Now, if the nominalist tries to explain resemblance in the same way
that she explained blueness, she will have to say: it’s not that the three
cases of resemblance have something in common; there isn’t anything
that is the same in the three cases. Rather, the three instances of resem-
blance merely resemble each other. So now we have a new level of resem-
blance, with three more resemblances: the resemblance between (the
resemblance between A and B) and (the resemblance between B and
C), the resemblance between (the resemblance between B and C) and
(the resemblance between A and C), and the resemblance between (the
resemblance between A and B) and (the resemblance between A and C).
But these three resemblances, so the nominalist must maintain, do not
literally have anything in common either. Rather, they merely resemble
each other. And so we are off on an infinite regress. Letting ‘A’ stand for
10
Price 1953; Rodriguez-Pereyra 2002.
the blueness of the sky, ‘B’ for the blueness of the blueberry, and ‘C’ for
the blueness of the parrot, and letting ‘R(x,y)’ stand for the resemblance
between x and y, we have the following series (I list just the first four of
the infinitely many stages):
Stage 1 Stage 2 Stage 3

A R(A,B) R[R(A,B), R(B,C)]
B R(B,C) R[R(B,C), R(A,C)]
C R(A,C) R[R(A,B), R(A,C)]
Stage 4 ...
R{R[R(A,B), R(B,C)], R[R(B,C), R(A,C)]}
R{R[R(B,C), R(A,C)], R[R(A,B), R(A,C)]} ...
R{R[R(A,B), R(B,C)], R[R(A,B), R(A,C)]}
All of these increasingly complex resemblances must exist. Armstrong

initially said that this regress was vicious, so that we must reject resem-
blance nominalism.11 Later, however, he decided, along with the
defenders of resemblance nominalism, that the regress is benign.12
2.5 The regress of temporal series
J. M. E. McTaggart presented a famous argument for ‘the unreality of

time’.13 Events in time may have the properties of being past, present,
or future, where these properties are understood to be mutually exclu-
sive. Now consider some particular event, e. Before e occurs, a person
can truthfully say, ‘e is in the future.’ While it is occurring, a person
can truthfully say, ‘e is in the present.’ And after it occurs, a person can
truthfully say, ‘e is in the past.’ Therefore, e must possess all three proper-
ties, past, present, and future. But this is contradictory, since as we have
said, the attributes of pastness, presentness, and futurity are mutually
exclusive. To resolve the paradox, McTaggart urges, we must deny that
anything exists in time.
Instead of rejecting the reality of time, we might try to resolve the
paradox by saying that there is no contradiction since e does not
have the properties of pastness, presentness, and futurity at the same
time. For example, suppose e is the event of my writing this chapter.
11
Armstrong 1978, 54–6.
12
Armstrong 1989, 54–7; Price 1953, 23–6.
13
McTaggart 1908.
e is in the present at the present; however, in the past, e was in the

future; and in the future, e will be in the past. Of course, nothing can
have incompatible properties simultaneously; however, there is no
contradiction in a thing’s having incompatible properties at different
times.
McTaggart says that this response involves introducing a second
temporal series. The original temporal series involved ordinary events,
which are ordered as past, present, and future. The second temporal
series involves the pastness, presentness, and futurity of events, which
themselves may be past, present, or future. But then McTaggart’s objec-
tion can be raised anew for this second time series: the presentness of e
itself has the properties of being past, present, and future. Again, these
are incompatible properties, so this is a contradiction.
We might try to resolve this second contradiction by claiming that –
just as e itself has the properties of being past, present, and future at
different times – so the presentness of e has the properties of being past,
present, and future at different times. (For example, the presentness of
e is present in the present, it was future in the past, and it will be past
in the future.) But now we are embarked on an infinite regress. To avoid
the regress, says McTaggart, we must deny the reality of time. Other
philosophers, however, embrace the infinite regress as benign – there
just are all of these temporal facts. For example, e is present, and the
presentness of e is present, and the presentness of the presentness of e is
present, and so on.14
2.6 The regress of truths
Our final infinite regress is the truth regress. The truth regress results
from the observation that, for any proposition, P, P entails that it is true
that P. The proposition that it is true that P then entails that it is true
that it is true that P. And so on. Thus, if there is any true proposition, P,
then there will be the following infinite series of truths:
P
It is true that P.
It is true that it is true that P.
⁝
14
Smith 1986.
This is the least controversial of the infinite regresses – it is generally

agreed to be benign.15 No one thinks that we need to avoid this infinite
series by denying that P entails that it is true that P; instead, the standard
response is simply to accept that all these truths exist.
2.7 Conclusion
What can we conclude from the preceding six examples of infinite

regresses? Perhaps just this: philosophers are in need of an account
of when an infinite regress is vicious. There is at present no generally
accepted account of this, though we shall see some candidate accounts
later. In fact, it seems likely that most philosophers, even those who
employ infinite regress arguments, do not have any account – not even
a controversial one – of what makes a regress vicious. This may help
to explain why infinite regress arguments in philosophy often produce
stalemates in which one side claims that the regress is vicious while
the other insists that it is virtuous. These regress arguments concern
some of the most fundamental and important issues in all of human
inquiry, including the beginning (if any) of the universe and the struc-
ture of knowledge. A successful account of virtue and vice among the
infinite would therefore be of great interest. In developing such an
account (much later in this book), we should bear in mind the few rela-
tively uncontroversial data points encountered in this chapter: the truth
regress is benign; the regress of Forms is probably vicious; and in the
regress of reasons for belief, an infinite regress of actual, consciously occur-
ring reasons would be vicious.
15
See, for example, Armstrong 1989, 54; Hochberg 1999, 196; Nolan 2001,
523–4.
3
Seventeen Paradoxes of the Infinite
3.1 A word about paradoxes
In this chapter, I aim to convince you that infinity is puzzling. I start

with some simple mathematical puzzles before moving on to more
exotic paradoxes. I will not attempt to solve any of the paradoxes in this
chapter; solutions will be considered later.
A word about the term ‘paradox’. Some people take the word to refer
to a situation in which a contradiction is true. On this interpretation,
there is of course no such thing as a paradox. None of my ‘paradoxes’
would count as true paradoxes, because in fact a true paradox would just
be impossible by definition. I find this use of the word uninteresting.
I understand paradoxes, roughly, as situations that are highly puzzling,
either because there seem to be compelling arguments for incompat-
ible conclusions about the situation, or because there seems to be a
compelling argument for something that seems absurd. A solution to
the paradox would be an account that removes the puzzlement. This
could be done by explaining why the situation in question cannot arise,
by explaining why the seemingly compelling argument (or one of the
seemingly compelling arguments) is fallacious, or by somehow making
the seemingly absurd conclusion of the argument stop seeming absurd.
3.2 The arithmetic of infinity
Suppose there is a number infinity, denoted by ‘∞’. Most people will agree
that if you add one to it, you get the same number, infinity. That is:
∞+1=∞ (Equation 3.1)
If equation 3.1 is true, then it seems that we should be able to subtract

∞ from both sides of the equation, thus obtaining 1 = 0.
17
Here is another way to derive the result. Most people who believe in
the number ∞ agree that if you divide a finite number by it, the result
is zero:
1/∞ = 0 (Equation 3.2)
(If you disagree with equation 3.2, try to think what else could go on
the right hand side of the equation. No number other than zero is small
enough.) If equation 3.2 is correct, then it seems that we should be able
to multiply both sides of the equation by ∞, thus obtaining:
1 = 0·∞ (Equation 3.3)
But in general, 0·x = 0, so it seems that we can again obtain the result
that 1 = 0.
More simply put, it is paradoxical that there should be a quantity that
is not increased when you add something (greater than zero) to it (as
in equation 3.1), or a number such that when you take that number of
zeros, you get a quantity larger than zero (as in equation 3.3).
3.3 The paradox of geometric points
According to standard geometry, space is composed of geometric points.

Every geometrical object – including circles, triangles, lines, planes,
spheres, and space itself – consists of a collection of points.
And what is a point? Recall the idea (Section 1.4) that any line segment
can be divided into two smaller segments. The same is true of two-dimen-
sional objects (for instance, a square can always be cut in half) and three-
dimensional objects (for instance, a cube can be cut in half). Well, a point
is conceived as a part of space so small that it cannot be divided; there is
no such thing as half of a point.1 Points are zero-dimensional objects, with
zero length, zero area, and zero volume. This has to be so, since if a point
had any size greater than zero, then it could be divided in half.
Now, here is something that is generally true about geometrical
objects: their sizes (lengths, areas, or volumes) are additive. For instance,
if you divide a sphere in half, the volume of the left hemisphere plus
the volume of the right hemisphere equals the volume of the sphere. If
you divide a line segment in thirds, the length of the left third plus the
1
Thus, the first definition from Euclid’s (1998) Elements, the definitive text of
geometry for two millennia, reads, ‘A point is that which has no part.’
Seventeen Paradoxes of the Infinite 19
length of the middle third plus the length of the right third equals the
length of the original segment.
Similarly, if a geometrical object is made up of points, one would
think, the size of any geometrical object should equal the sum of the
sizes of the points that make it up. For instance, the volume of a sphere
should equal the sum of the volumes of all the points that make up
the sphere. The length of a line segment should equal the sum of the
lengths of the points that make it up. And so on.
But how can this be, when each of these points has a size of zero? If
one combines many zeroes, the result should be zero, so it would seem
that a line segment consisting of points (even if there are infinitely many
points) should have a length of zero.2 If you are tempted now to question
whether points really have zero size, consider the alternative: if a point has
any length larger than zero, then any line segment would have to have a
length of infinity, since there are infinitely many points in any segment.
Furthermore, a one-meter line segment and a two-meter line segment
contain the same number of points (infinitely many). Therefore, it
appears that we must say that infinity times zero meters equals one
meter, and infinity times zero meters also equals two meters:
∞·0 = 1
∞·0 = 2
By the transitivity of equality, 1 = 2.

Here is a metaphysical ‘application’.3 Suppose there were a stick made
of continuously distributed matter, so that it completely fills a certain
region of space. (This is unlike actual sticks, which are almost entirely
empty space at the atomic level.) And let’s say we have a coordinate
system in which the stick extends from x-coordinate 0 to x-coordinate
1 (the y and z coordinates don’t matter for the example). It seems that
this stick ought to have a part corresponding to each part of the space
it occupies – so, for example, for each x-value between 0 and 1, there
should be a cross-sectional slice of the stick that occupies just that
x-value. Each of these slices would be a single geometric point thick.
These are unusual objects, but if you accept the existence of geometric
points in general, it seems that you ought to accept that the stick could
have such parts.
2
This paradox goes back to Aristotle (1941, On Generation and Corruption,
316a23–35).
3
Brentano (1988, 146) uses this to argue against the existence of points.
x
0 r 2r
Figure 3.1 A magically expanding rod
Now imagine that we cut the stick up into single-point-thick slices,

and we rearrange the pieces as follows: for each slice, if the slice starts
out with an x-coordinate of r, we move that piece to an x-coordinate of
2r (see Figure 3.1). The result: the new stick completely fills its space,
from x=0 to x=2, just as the old stick completely filled the interval from
x=0 to x=1. No x-coordinate in that interval is left empty, and each is
occupied by qualitatively the same thing that initially occupied the
coordinates in the interval [0,1]. So what we have is a new stick that is
just like the old one, only twice as long. We have created more material
just by rearranging its parts.4
3.4 Infinite sums
Consider the following infinite collection of numbers:
1, −1, 2, −2, 3, −3, ...
(all the positive and negative integers). What is the sum of all these
numbers?
There are many ways to answer this question. First answer: the sum is
zero, because we can group the terms to be added as follows:
(1 − 1) + (2 − 2) + (3 − 3) + ...
= 0 + 0 + 0 + ...
= 0.
4
This is similar to the Banach-Tarski Paradox, in which a unit sphere is
dissected into five pieces, which are then rearranged to form two spheres of the
same size as the original (Banach and Tarski 1924). The Banach-Tarski Paradox is
more amazing, since only finitely many pieces are needed; however, it depends
on the controversial Axiom of Choice, and the pieces are not actually describable
or imaginable by any human being.
Second answer: the sum is infinity, because we can group the terms as
follows:
1 + (−1 + 2) + (−2 + 3) + (−3 + 4) + ...

= 1 + 1 + 1 + 1 + ...
= ∞.
Third answer: the sum is negative infinity, because we can arrange and
group the terms as follows:
−1 + (1 − 2) + (2 − 3) + (3 − 4) + ...
= −1 − 1 − 1 − 1 − ...
= −∞.
Since one can obtain different sums depending on the order in which
one adds the numbers, what is the true sum of these numbers? Is there
a ‘correct’ way of ordering the terms?
3.5 Galileo’s paradox
A perfect square is a number that results from multiplying a natural

number by itself – for instance, 1, 4, 9, and so on. The numbers that
must be multiplied by themselves to obtain the perfect squares are called
roots. Thus, 1 is the root of 1, 2 is the root of 4, 3 is the root of 9, and
so on. Galileo discovered the following paradox in 1638.5 Question: in
comparing the perfect squares to the roots, which are more numerous?
Are there more squares than roots, more roots than squares, or the same
number of each? It seems that there must be at least as many squares as
roots, because every root has a distinct square.
Second question: in comparing the perfect squares to the natural
numbers in general, which are more numerous? Well, of course every
perfect square is a natural number, but most of the natural numbers
are not perfect squares. Indeed, as one goes through the list of natural
numbers, perfect squares become ever rarer. For instance, among the
first 100 natural numbers, ten are squares, for a proportion of 0.1.
Among the first 10,000 natural numbers, only a hundred are squares,
for a proportion of 0.01. Among the first 1,000,000 natural numbers, a
5
Galilei 1914, 31–3. Yes, Galileo discussed this 240 years before Cantor’s devel-
opment of transfinite arithmetic. And no, Cantor did not solve the paradox.
thousand are perfect squares, for a proportion of only 0.001. And so on.
As we go on, the ratio of squares to natural numbers approaches ever
closer to zero. So it seems that not only are the natural numbers more
plentiful than the perfect squares; the perfect squares actually form only
an infinitesimal portion of all the natural numbers.
But wait. The set of roots is identical to the set of natural numbers,
since every natural number can be squared. So it cannot be that there
are fewer squares than there are natural numbers, and also that there are
at least as many squares as there are roots.
Galileo’s conclusion was that infinities cannot be compared in terms
of size. If two things are infinite, we cannot say that one is either greater
than, less than, or equal to another; we can only say that both are infi-
nite. But this is strange. If x is a quantity, and y is a quantity, how can
it be that x and y fail to have any quantitative relation to each other –
how can a definite quantity fail to be less than, equal to, or greater than
another definite quantity?
3.6 Hilbert’s hotel
Suppose you run a hotel with an infinite number of rooms. At the

moment, all the rooms are filled. A new customer arrives, seeking a room
for the night. You tell him that all the rooms are taken. The customer is
just about to leave when you say, ‘Oh, don’t worry. I said all the rooms
were filled, but I didn’t say we couldn’t fit you in!’ You then tell each
guest to move down one room: the guest in room 1 moves to room 2,
the guest from room 2 moves to room 3, and so on. This leaves room 1
vacant, which you use to accommodate the new guest.
The next day, all the rooms are still full. But this time, an infinite
collection of new customers arrives, all wanting separate rooms. ‘Well,
we’re all full, but that doesn’t mean we can’t accommodate you!’ you
say. This time, you tell each guest to move to room number 2n, where n
is his current room number. Thus, the guest in room 1 moves to room 2,
the guest from room 2 moves to room 4, the guest from room 3 moves
to room 6, and so on (Figure 3.2). This leaves free all the odd-numbered
rooms, which you use to accommodate the new guests.6 This seems
paradoxical. If a hotel is full, it ought not to be able to accommodate a
new patron, let alone an infinite number of new patrons, without either
6
Hilbert first described the scenario in a 1924 lecture (2013, 730). It was subse-
quently popularized by Gamow (1947, 17). Kragh (2014) traces the entertaining
history of the scenario.
Rm. 1 Rm. 2 Rm. 3 Rm. 4 Rm. 5 Rm. 6
Figure 3.2 Hilbert’s hotel
expelling any of the existing patrons or forcing anyone to double up in

a room.
Let us try to make the puzzle sharper. Let R1 be the number of rooms
in the hotel at the beginning of the story. Let G1 be the number of guests
staying in the hotel at the beginning of the story. Since these guests
exactly fill the rooms, it seems that R1 = G1. Now, let G2 be the infinite
number of new guests who arrive at the hotel, seeking accommodations,
and let R2 be the number of rooms existing in the hotel at the end of the
story. At the end of the story, when all the new guests have been accom-
modated, these new guests together with the old guests exactly fill up all
the rooms. Therefore, it seems that R2 = G1 + G2. And therefore, we seem
obliged to say one of two things: Either
i. We say that R1 = R2, in which case G1 = G1 + G2, so we can infer that G2 = 0.

But we already said that G2 was infinite; hence, infinity is zero. Or,
ii. We say that R2 > R1, in which case we have increased the number of
rooms in our hotel merely by rearranging the guests. Furthermore,
since no new rooms were built, we magically caused the identical set
of rooms to constitute a greater number.
At least one philosopher has taken the strangeness of this scenario to

prove that actual infinities cannot exist.7
3.7 Gabriel’s horn
Gabriel’s Horn is a geometric figure that has a finite volume but an infi-
nite surface area. To envision the shape, first consider the graph of the
7
Craig 1991.
Figure 3.3 Gabriel’s horn
function y = 1/x, for x ≥ 1 (ignore everything to the left of x = 1). Imagine

that that shape is rotated about the x axis. The surface traced out is
Gabriel’s Horn (see Figure 3.3). It can be proven that the surface area of
the horn is infinite, while the volume inside the horn is finite.
Imagine that you had a physical object with such a shape, and you
wanted to paint it. To cover an infinite surface, it would seem that you
would need an infinite amount of paint. However, since the volume
is finite, you could fill the entire horn with a finite amount of paint,
thereby completely covering the inner surface.
3.8 Smullyan’s infinite rod
Imagine that there is an infinite, flat plane. Everywhere on (and above)

the plane, there is a gravitational force pulling downward. There is a
finite vertical rod sticking out of the plane, perpendicular to the plane,
with a hinge at the top. Attached to the hinge is an infinitely long,
perfectly rigid rod (see Figure 3.4). The infinite rod is initially oriented
perpendicular to the finite rod (parallel to the plane). Assume that the
rods and plane are unbendable, unbreakable, and impenetrable.
What happens to the infinite rod in this example? Even though the
force of gravity is pulling it downward, and even though it is supported
on only one end, the rod cannot tilt downward, because any nonzero
amount of downward rotation would require the rod to intersect and
break through the plane somewhere. Therefore, the rod will remain
suspended parallel to the plane, as if by magic.8 In modern physics
terminology, the rod undergoes an infinite net torque, yet it maintains a
constant angular velocity of zero.
8
Smullyan 2008, 246.
Figure 3.4 Smullyan’s rod
3.9 Zeno’s paradox
Perhaps the most famous paradox of the infinite is due to the ancient
Greek philosopher Zeno of Elea.9 Zeno was a student of Parmenides, the
ancient Greek philosopher who held that change was impossible. Zeno
devised his paradoxes of motion to show that – as a special case of the
thesis that change is impossible – motion is impossible. Here, I describe
two of these paradoxes.10
Suppose you drop a ball. What will happen? You may think that the
ball will fall to the ground, but actually, this is impossible. Call the ball’s
starting point in the air ‘point A’, and call the alleged destination on
the ground ‘point B’. For the ball to move from point A to point B, it
must first travel half the distance. After it does that, it will then have to
travel half the remaining distance (bringing it to three quarters of the
total distance). Then it must travel half the remaining distance (taking
it to seven eighths of the total distance). Then it must travel half the
remaining distance ... and so on. This series continues infinitely. But it is
not possible to complete an infinite series. Therefore, the ball can never
reach the ground.
9
Sadly, no writings of Zeno survive. We know of his arguments through
writers who discussed him, notably Aristotle (1941, Physics 233a13, 239b5,
263a4). Benardete (1964) devotes considerable attention to Zeno’s paradoxes.
10
Zeno had a third paradox, but it is too lame to include in the main text. It
seems to involve inferring, from the premise that an arrow in flight cannot move
any nonzero distance at a single instant of time, that the arrow is at rest at each
instant, and thence that the arrow never moves.
Some students who hear of this puzzle draw the conclusion that the
ball will get closer and closer to the ground but never quite reach it. This
is wrong. If the above reasoning is correct, then it can be applied to
any object and any chosen destination point. Therefore, no object can go
anywhere. For example, the ball could not reach the point one foot off
the ground, any more than it could reach the ground.
Many students who hear of Zeno’s paradox naively assume that the
mathematical theory of infinite series somehow solves the paradox, or
that calculus somehow solves it.11 The line of thinking may be some-
thing like this: ‘In a math class, I learned something about infinite series
like this. So that must solve the problem.’ Generally speaking, what
students are taught is that the series,
∞
1 1 1 1
∑2
n =1
n
= + + + ...
2 4 8
converges to 1; that is, as one adds more and more terms, the sum comes
arbitrarily close to 1. Mathematicians therefore say that the infinite sum
equals 1.
And this would solve the problem, if the problem were to calculate
the distance from point A to point B. In that case, the theory of infinite
sums will tell us that, if we view the total distance as a sum of these infi-
nitely many smaller parts, this sum is equal to 1, meaning 100 per cent
of the original distance. Viewing the distance as a single line segment
of a given length, or as an infinite collection of smaller and smaller
segments added together, makes no difference to the total length.
But that was not the problem. The question was not ‘what is the
distance between A and B?’, and Zeno is not saying that the distance
from A to B is infinite. Zeno is saying that to reach point B, one would
have to complete the infinite series. But (allegedly), one cannot do such a
thing – not because the distance is too long, nor because it will take too
much time, but because it is conceptually impossible to complete a series
that has no end. This assumption is in no way challenged by standard
modern mathematical treatments of infinite series, nor of any concepts
used in calculus. Quite the contrary is in fact the case: standard treat-
ments are designed precisely to avoid the assumption that an infinite
series can actually be completed.
11
If you thought this, don’t feel bad; no less a philosopher than Alfred North
Whitehead (1929, section II.ii.2) commented, ‘Zeno produces an invalid argu-
ment depending on ignorance of the theory of convergent numerical series.’
The version of the paradox that I have just stated uses an endless
series to argue that the ball can never complete its journey. Another vari-
ation uses a beginningless series to argue that the ball can never begin its
journey: in order to travel from point A to point B, the ball must first
travel half the distance. But before it can reach the halfway mark, it must
first go one quarter of the distance. But before it does that, it must first
go one eighth of the distance. And so on. Because there is an infinite
series of preconditions, the ball can never get started. This version of the
paradox also has the virtue of forestalling the confused thoughts (a) that
the ball can get close to the ground even if it can’t reach it, and (b) that
∞
1
somehow calculating the infinite sum, ∑ n , solves the problem.
n =1 2
That was Zeno’s first paradox of motion. The second paradox uses
a similar idea. Achilles decides to have a race with a tortoise. Because
Achilles is the faster runner, they both agree that the tortoise should get
a head start. Achilles starts at point A, while the tortoise starts at point
B, which is ahead of A. To overtake the tortoise, Achilles must first get to
point B, where the tortoise was at the start of the race. But by the time
Achilles reaches B, the tortoise will have advanced to a new place, call it
‘point C’. Now Achilles must run to point C. But by the time he reaches
C, the tortoise will have advanced to point D. Then Achilles must run
to D ... Matters continue in this way forever. To overtake the tortoise,
Achilles must first complete this infinite series of catching-up actions.
But one cannot complete an infinite series. Therefore, no matter how
fast Achilles may be, he can never overtake the tortoise.
3.10 The divided stick
Contra Zeno, let us now suppose that it is possible to complete an infinite

series (it certainly seems that things move!). Now, imagine that there is
a stick composed of an infinitely divisible material (forget about atomic
theory). We spend half a minute cutting the stick in half. Then we spend
the next quarter minute cutting both of those halves in half, so that the
original stick has now been cut into quarters. Then we spend an eighth
of a minute cutting all of the quarters in half. And so on (see Figure 3.5).
At the end of one minute, what is left? It appears that the result must
be an infinite number of slices of the stick. How thick will each slice be?
For any nonzero size you name, there was a point in the series when
the pieces were cut smaller than that size. Therefore, the slices have no
nonzero size; that is, they have a size of zero. How much matter could
each of these slices contain? It seems that the answer must, again, be
zero. But it seems that if each slice has zero mass and zero volume, then
Stage 1
Stage 2
Stage 3
Stage ∞
Figure 3.5 The divided stick
the total of all of them must have zero mass and zero volume – so the
material of the stick has been destroyed.12
3.11 Thomson’s lamp
Suppose there is a lamp with an on/off switch. The lamp starts out on.
After half a minute, it is switched off. After another quarter minute, it
is switched back on. After another eighth of a minute, it is switched off.
And so on. At the end of one minute, an infinite number of switchings
will have occurred. Will the lamp then be on or off?13
This question is puzzling, in the first place, because it seems that the
question should have an answer – if such a lamp were made, it would be
in some state at the end of one minute. But no answer to the question
seems right. It doesn’t seem correct to say simply ‘the lamp will be on’,
nor does it seem correct to say ‘the lamp will be off’. The answer seems
to turn on whether infinity is even or odd, but infinity is neither even
nor odd. But nor does it seem correct to declare that the lamp will be
neither on nor off. (How could this be? If the lamp explodes at the last
instant? But surely this isn’t the solution.) Nor does it seem correct to
say that the lamp’s state will be randomly selected – for example, that
it will just have a 50 per cent probability of being on and a 50 per cent
probability of being off.
12
This scenario is from Benardete 1964, 184–5. Benardete ascribes to the final
pieces an ‘infinitesimal’ width rather than zero width. I stick with zero, since
there are no infinitesimal quantities. Cf. Moore 1990, 5; Oppy 2006, 11–12, 66–8.
Oppy (67) rejects the infinitesimal approach on the grounds that there is no way
of identifying a uniquely correct infinitesimal.
13
This is from James Thomson (1954), who uses it to argue that it is impossible
to complete an infinite set of tasks. Variations appear in Benardete 1964, 23.
We can also construct an argument against each answer: on the one

hand, it seems that the lamp cannot be on at the end of the minute,
because it was switched off at the first step of the series, and thereafter
it was never turned on again without subsequently being turned off.
But by the same token, it cannot be off, because it started out on, and
it was never switched off without subsequently being switched back on
again.
3.12 The Littlewood-Ross Banker
You are to play a game with a person known as the Banker.14 Your only
goal in this game is to increase your wealth. At the start of the game,
there is an infinite pile of one dollar bills. The bills are labeled with the
natural numbers: bill #1, bill #2, bill #3, and so on. You start out owning
only bill #1; the Banker owns all the rest. The game has infinitely many
turns. In the first turn, the Banker offers to let you trade bill #1 for
bills #2–10. If you decline, the Banker keeps bills 2–10. If you accept, as
presumably you should, you gain $9 (bills #2–10) and the Banker gains
$1 (bill #1). In the second turn of the game, the Banker lets you take bills
#11–20, which you may keep for yourself, but you must give him your
lowest-numbered bill (either bill #1 or bill #2) – netting you a $9 profit.
The game continues on in this way, so that at each turn, you are offered
the chance to take the next ten bills from the pile for yourself, and give
the Banker the lowest numbered bill from your pile. If you accept every
time, then after each stage n, you have 9n dollars, and the Banker has
n dollars. Let us assume, finally, that each turn of the game is played in
half the time as the previous one: the first turn takes half a minute, the
second turn takes a quarter of a minute, and so on, so that at the end of
one minute, all the infinitely many turns have been played.
If you always act to maximize your wealth, taking the $9 profit in
each round, how much money will you have at the end of the game?
On the one hand, it seems that you should have an infinite amount
of money, because you took a $9 net gain, infinitely many times:
9 3 ∞ = ∞. But on the other hand, you will have nothing, because at
the end of the game, the Banker has bill #1 (which you gave him in
14
What follows in the text is a variant on the paradox first discovered by
Littlewood (1986, 26 [originally published 1953]) and later elaborated by Ross
(1976, 36–8). For some reason, philosophers tend to falsely credit the paradox to
Ross. Barrett and Arntzenius (1999) discuss the decision-theoretic version of the
paradox. Cf. Oppy 2006, 15–16.
the first round), bill #2 (which you gave him in the second round), bill
#3 (which you gave him in the third round), and so on. For every n,
the Banker has bill #n, because you gave it to him in the nth round.
Therefore, in the end the Banker has every bill, so you have nothing.
This is odd, because there is no stage of the game at which you lose
money; at every moment up until the end of the game, you have some
money, getting more and more as the game continues; but at the last
instant, as the clock reaches one minute and the game completes, your
pile of money vanishes.
Next, imagine that you demand a rematch with the Banker. This
time, when he asks you for bill #1, instead of giving it to him, you
take bill #10, surreptitiously erase the ‘0’ on the end of its label so
that it says ‘1’, and hand it to him. Then you take your bill #1 and
append a ‘0’ on the end of its label so that it becomes labeled ‘10’.
When he asks you for bill #2, you likewise relabel bill #20 as #2 and
hand it to him, then relabel your bill #2 as #20. You continue in
this manner through the entire game. At each stage, you effectively
switch the labels of your highest- and lowest-numbered bills, give
the Banker the new lowest-numbered bill, and keep the rest. This
time, how much money do you have at the end of the game? You
still have the bill that was originally labeled bill #1, although you
have relabeled this bill by adding a ‘0’ to its label infinitely many
times (in rounds 1, 10, 100, and so on). So it is now labeled with a
‘1’ followed by infinitely many ‘0’’s, but it is still yours. Similarly for
bills 2–9, 11–19, and so on: you took those, and you never gave them
away. Each time you were supposed to give one of them away, you
surreptitiously gave the Banker your highest-numbered bill instead.
The bills originally labeled ‘10’, ‘20’, ‘30’, and so on, all went to the
Banker, though they were relabeled so that they now say ‘1’, ‘2’, ‘3’,
and so on. So each of you ends up with an infinite pile of money.
This result is surprising, because in this variant, each of you has
qualitatively the same thing at every stage of the game as you had in
the original variant: at the end of round 1, you have bills labeled
‘2’ through ‘10’, and the Banker has a bill labeled ‘1’; at the end of
round 2, you have bills labeled ‘3’ through ‘20’, and the Banker has
bills labeled ‘1’ and ‘2’; and so on. The two series are qualitatively
identical, the only difference between them lying in the bare numer-
ical identity of the bills that each player has at any given stage. Yet
the one series gives you an infinite amount of money, whereas the
other gives you nothing.
3.13 Benardete’s paradox
A ball is placed at the top of a hill, where, if nothing interferes, it will roll
down to the bottom.15 There is, however, an infinite series of impene-
trable walls placed on the hill, any one of which would suffice to stop
the ball. Call these walls w1, w2, w3, and so on. w1 is located halfway
down the hill from where the ball is, w2 is located a quarter of the way
down the hill, w3 is located one eighth of the way down the hill, and so
on. (Assume that each wall is half the thickness of the previous one, so
that there is room for all of them.) What happens to the ball?
The ball will of course be unable to roll any distance down the hill,
since for any distance it might have rolled, there are infinitely many
walls that would stop the ball before it got that far. But which wall stops
the ball? The ball is not stopped by w1, since the ball would have been
stopped by w2 before it could reach w1. Nor is it stopped by w2, since it
would have hit w3 before it could reach w2. Nor is it stopped by w3, since
it would have first hit w4. And so on. For any n, the ball does not hit wn,
since it would have hit wn+1 first. Therefore, the ball will simply hold still
at the top of the hill, without anything stopping it.
One might propose that there is another object, distinct from all the
wi, which stops the ball. Call this object wω. wω is the composite object
composed of all the walls wi put together. You might think of this as
another impenetrable wall (with a thickness equal to half the distance
down the hill), albeit a wall with many gaps in it. Well, whether you
want to call wω a ‘wall’ or not, it stops the ball from rolling.16
Here is a variation of the paradox. Imagine that, instead of an infinite
series of walls, there is an infinite series of gods, G1, G2, and so on, each
of whom is able to create an impenetrable wall at any chosen location. G1
happens to have the intention of creating an impenetrable wall halfway
down the hill, if and only if the ball makes it more than one quarter of the
way down the hill. G2, for his part, intends to create an impenetrable wall
one quarter of the way down the hill, if and only if the ball makes it more
than one eighth of the way down the hill. And so on. In this case, the
ball cannot move any distance down the hill. But which wall, created by
which god, will stop the ball? For every n, Gn fails to create a wall, since
15
The following paradoxes are from Benardete 1964, 236–8, 259–60. Cf. Moore
1990, 4–5; Oppy 2006, 10–11, 16–18.
16
This is Hawthorne’s view (2000, 626). Benardete (1964, 260) entertains
similar suggestions. Cf. Laraudogoitia 2003, 126.
Gn+1 would have stopped the ball before it reached the point where Gn
found it necessary to act. So none of the gods creates a wall. So the ball
simply hovers at the top of the hill with nothing holding it back.
In another variation, Benardete asks us to imagine a pile of infinitely
many opaque stone slabs. The bottom slab is one inch thick and has the
numeral ‘1’ painted on its upper surface. On top of this is a second slab,
which is half an inch thick and has the numeral ‘2’ painted on its upper
surface. On top of that, a slab one quarter inch thick with a ‘3’ painted
on it. And so on. The whole pile is two inches thick. Now, suppose you
stand above this pile and look down. You can’t see the first slab, since
your view of it is blocked by the second slab. Likewise, your view of the
second slab is blocked by the third, your view of the third is blocked by
the fourth, and so on. Your view of each slab is blocked, so you can’t see
any of them. But surely you cannot see through the pile either. So what
do you see?
3.14 Laraudogoitia’s marbles
Imagine an infinite series of marbles, arranged in a straight line, in a

Newtonian world. The first marble is located at point A. The second
marble is located half a meter to the right of A. The third is located a
quarter meter to the right of the second marble (so, 75 centimeters to
the right of A). The fourth is located an eighth of a meter to the right of
the third. And so on. These marbles get smaller and smaller, with each
having half the diameter of the last, so that all of them can fit into a
one-meter-long space. Assume that every marble is perfectly rigid (so all
collisions are perfectly elastic) and all have the same mass. Now, suppose
someone pushes the first marble, imparting to it a velocity of 1 meter
per second to the right. What will happen?17
After half a second, the first marble strikes the second, whereupon the
first marble stops and the second starts moving at 1 meter per second to
the right. After another quarter second, the second marble runs into the
third, whereupon the second marble imparts its momentum to the third.
After another eighth of a second, the third marble strikes the fourth,
and so on. After one second, every one of the marbles has been struck.
However, nothing emerges on the right-hand side of the series. If there
were a last marble in the series, that last marble would emerge at the
right edge (one meter to the right of point A), moving 1 m/s to the right.
17
This case is from Laraudogoitia 1996.
But because the series of marbles is infinite, there is no last marble, so

there is nothing to emerge at the right. Therefore, the momentum and
kinetic energy of the system will simply vanish: all the marbles will have
been struck, every one will have transferred its momentum to another
marble, and nothing will come out on the right.
Laraudogoitia points out that the laws of Newtonian mechanics
are time-reversal-invariant. What this means is that for any sequence
of events that is consistent with Newton’s laws, the reverse sequence
(in which everything happens the same way, except with the opposite
directions of motion and opposite order of events) is also consistent
with the laws. Thus, if the marble sequence described above is possible,
then the reverse sequence must also be possible, in which a system
containing infinitely many marbles starts out with zero momentum and
zero kinetic energy, and then the marbles spontaneously start moving
to the left, each marble transferring a leftward velocity of 1 m/s to the
marble to its left. In this reverse sequence, the process ends up with the
first marble at point A moving 1 m/s to the left and all the other marbles
stationary.
3.15 The spaceship
Imagine that there is a spaceship that travels in a straight line, starting

at a speed of 1 m/s to the right, and accelerates at an increasing rate, so
that after one second it is moving at 2 m/s; after another half second,
it is moving at 4 m/s; after another quarter of a second it is moving at
8 m/s; and so on.18 At the end of two seconds, the ship would be traveling
at infinite velocity. At this time, where would the ship be?
The total distance traveled would be infinite. In the first stage, the
ship travels at a speed of at least 1 m/s for one second, thus covering a
distance of at least one meter. In the second stage, it travels at a speed
of at least 2 m/s for a half second, thus covering a distance of at least
(2)(½) = 1 meter. And so on. Since there are infinitely many stages, the
total distance is at least 1 + 1 + 1 + ... = ∞.
But there simply is no position infinitely far from the starting point;
for any two points in space, there is some (finite) distance between them.
The ship would have a definite position, getting farther and farther
away, at each instant from the starting time up to but not including the
final time two seconds later. But at t = 2 seconds, there is nowhere the
18
Benardete 1964, 149; Moore 1990, 70–1; Oppy 2006, 12.
ship could be. Perhaps the ship would go out of existence at that instant,
destroyed by its excessive speed.
3.16 The Saint Petersburg paradox
Imagine that you have the chance to play a simple game. A fair coin is
to be flipped some number of times. If it comes up heads on the first
flip, you get $2, and the game ends; otherwise, the coin is flipped again.
If it comes up heads on the second flip, you get $4, and the game ends;
otherwise, the coin is flipped again. If it comes up heads on the third
flip, you get $8. And so on. In brief, the coin is flipped until the first time
it comes up heads, whereupon you get a payoff of $2n, where n is the
number of times the coin was flipped. The question is: how much is it
worth to get a chance to play this game? Or, how much should a rational
person be willing to pay to be allowed to play?19
There is a standard way of addressing this kind of question: we calcu-
late the expected payoff of the game. This is done by adding up, for
every possible outcome of the game, the probability of the outcome
occurring multiplied by the payoff you receive if it occurs. The outcomes
are listed below (where, for example, ‘TH’ indicates tails on the first flip,
followed by heads on the second flip), with their associated probabilities
and payoffs:
Outcome Probability Payoff

H 1/2 $2
TH 1/4 $4
TTH 1/8 $8
⁝ ⁝ ⁝
So the expected value of the game is (1/2)($2) + (1/4)($4) + (1/8)($8)

+ ... = 1 + 1 + 1 + ... = $∞. So the fair price for a chance to play the game
is $∞.
Why is this paradoxical? First, it does not seem that the game is worth
anything close to an infinite amount of money. Indeed, few people would
value it even as high as $20. Is our intuitive valuation mistaken, or is
there something wrong with the reasoning of the preceding paragraph?
19
This problem was originally published by the Swiss mathematician Daniel
Bernoulli in 1738, who attributes it to his cousin Nicolas Bernoulli, also a mathe-
matician (not to be confused with Daniel’s brother Nicolas Bernoulli, who was also
a mathematician). For the modern English translation, see Bernoulli 1954, 31.
Second, it seems almost contradictory to hold that the value of the

game is infinite, since if one plays the game, one will never gain an
infinite payout. All the possible payouts are finite (though they increase
without bound), so there is zero chance that one will gain $∞. Therefore,
if one were to pay an infinite amount to play the game, one would
be guaranteed to lose money (and an infinite amount, at that!). More
generally, it is strange that there should be a variable whose actual value
is known with certainty to be less than its expected value.
3.17 The Martingale betting system
You’re about to play roulette in a European casino. You like the color
red, so you bet that the ball will land on red. The probability of this
happening is 18/37 (the wheel has 18 red squares, 18 black squares, and
1 green square), or about 48.6 per cent. So if you play repeatedly, you
should win about 48.6 per cent of the time, with the house winning the
other 51.4 per cent of the time. In the long run, as everyone knows, the
house comes out ahead.
But wait – despite the house’s advantage, there is a way to guarantee
that you come out ahead (without cheating). Unfortunately, this system
requires that you start with an infinite bankroll, and that there be no
betting limits – if these conditions do not apply to you, then don’t try
this system!
First, bet $1 on red. If you win, walk away with your $1 profit.
If you lose, play again, but this time stake $2 on red. Again, if you
win, walk away with your net profit of $1 (the $2 payout, minus the
$1 loss from the first bet). If you lose the second bet, place a third
bet, this time staking $4. If you win, walk away with your $1 profit
($4 from the third bet, minus the $2 loss from the second bet and the
$1 loss from the first bet). And so on. Simply continue doubling down
until you win, at which point you can walk away with your $1 profit.
Eventually, the ball will land on red, so you are guaranteed to come
out ahead.20
So you have a guaranteed way to win $1, provided that you start with
$∞. This may not sound terribly impressive. But if this works, there is
also a way to extract an infinite amount of money from the casino. First,
apply the betting strategy as described above, until you win your dollar
20
It is logically possible, consistent with the stipulations of the scenario, that
you get stuck with an infinite series of black and green; however, the probability
of this happening is zero.
of profit. Then, instead of walking away, simply start over with the $1 bet
on red again. Follow the strategy again, doubling down every time you
lose, until you win a second dollar. Then start over, applying the strategy
until you win a third dollar. And so on.
Something seems to have gone wrong here. Given the house’s built-in
advantage (the odds in each bet favor the house, 19–18), the house should
come out ahead in the long run. Yet the system we have just described
seems to guarantee, over the long run, unlimited losses for the house.
Now, you may worry about the infinite time it will presumably take
to collect your desired infinite winnings. For the impatient, imagine
the following further stipulations, all of which appear jointly logically
consistent: in your first implementation of the betting strategy, your first
spin of the roulette wheel takes one minute, the second spin (if there is one)
takes half a minute, the third spin (if there is one) takes a quarter minute,
and so on. Thus, the first application of the betting strategy, netting you
your first $1 profit, is guaranteed to be complete within two minutes.
Now, the second time you implement the betting strategy, assume
that the first spin takes a half minute, the second spin (if there is one)
takes a quarter minute, and so on. So you are guaranteed to win your
second dollar within one minute.
In your third application of the betting strategy, the first spin take a
quarter minute, the second takes an eighth of a minute, and so on. So
you win your third dollar within a half minute.
And so on. All told, then, you complete infinitely many iterations of
the betting strategy, netting an infinite profit, within four minutes.21
3.18 The delayed heaven paradox
One final decision-theoretic puzzle. You have just died and gone to meet
your Maker. After death, there are three destinations to which souls may
be assigned:
1. People who lived sufficiently good lives are assigned to Heaven,

where they enjoy a high positive level of wellbeing – let’s say, 100
utils – every day for eternity. (Note: a ‘util’ is a hypothetical unit of
happiness or desire-satisfaction.)
∞ ⎛1 ∞
1⎞ ∞
⎛1 ⎞ ∞
1
21
Here is the math: the total time taken is ∑ i ∑2 j i
2 2 ∑2 i
= 2 2 = 4.
i =0 ⎝2 j 0 ⎠ i ⎝2 ⎠ i 0
2. People who lived sufficiently bad lives are assigned to Hell, where
they suffer a similarly high negative welfare level – say, −100 utils –
every day for eternity. (Negative utils represent unhappiness or desire-
frustration.)
3. People who lived so-so lives are assigned to Limbo, where their exist-
ence is completely neutral (0 utils), every day for eternity.
Assume that the suffering of one day in hell exactly counterbalances the
enjoyment of a day in heaven, so that a day in heaven plus a day in hell
is exactly as good as two days in Limbo.
When you arrive at the Pearly Gates, eager (or apprehensive?) to learn
your fate, God informs you that your life was neither good enough to
earn eternity in Heaven, nor yet bad enough to earn eternity in Hell. In
such circumstances, he says, you would normally be assigned to eter-
nity in Limbo. However, as a special offer just for you, you may instead
choose to take one day in Hell, followed by two days in Heaven, followed
by the rest of eternity in Limbo. If you decline, you go straight to Limbo.
It’s a good deal, so you accept.
You spend your day in Hell – where, we presume, you are forced to
grade stacks of philosophy essays by C students – then return to the
Pearly Gates. ‘Whew’, you say to God. ‘That really sucked. Now I’m
ready for some Heaven!’
‘Of course’, says God. ‘You may now take your two days in Heaven,
followed by an eternity in Limbo. Or ... I have another special offer, just for
you! You may spend another day in Hell, after which I will let you have
four days in Heaven, followed by the rest of eternity in Limbo. What say
you?’
Again, it’s a good deal – the goodness of two more days in heaven
more than makes up for the badness of one extra day in Hell. So you
accept, spend another day grading papers, then come back to see God.
Once again, he offers you a special deal: spend a third day in Hell,
then you can have six days in Heaven, followed by Limbo.
Everyone can see where this is going. Each time you return from a day
in Hell, God offers to let you take another day in Hell, thereby earning
an extra two days in Heaven. And yes, you always have to take your day
in Hell before you can take any days in Heaven. Each time, if you are
rational, you accept. The result: you spend eternity grading papers. How
can it be that, by making the prudent choice every time, you wind up
with the worst possible outcome?
3.19 Conclusion
The conclusion to be drawn from this chapter is similar to that of the

last chapter: we need an account of when infinite series are possible or
impossible. It is natural to try to resolve some of the preceding para-
doxes by declaring that the infinite series postulated in the paradoxes
are impossible. If you are tempted to say this, remember that not all of
the paradoxes depended on the assumption that an infinite series can be
completed. Zeno’s paradoxes start from precisely the opposite premise,
that an infinite series cannot be completed. So if one denies in general
that infinite series can be completed, one avoids some of the paradoxes,
but at the price of joining sides with Zeno.
What we would like to be able to say is that some infinite series cannot
be completed, while others can; that, for example, the Littlewood-Ross
Banker and Thomson Lamp are of the uncompletable kind; and that
the Zeno series are of the completable kind. But we can’t just say that.
We need an account of why that is the case. We need a theory of when
infinities are impossible.
Part II
Old Theories of Infinity
4
Impossible Infinite Series:
Two False Accounts
In this chapter and the next, we consider what earlier thinkers have
said about when an infinite series, or other form of infinity, is impos-
sible. There are three main views to consider: (i) that an infinite series
can never be completed by traversing the steps in succession, (ii) that
a condition cannot obtain if it has an infinite series of preconditions,
and (iii) that there can exist only potential infinities, never an actual
infinity. These theories are not mutually exclusive; one could endorse
two or even all three of these views at once, as I believe thinkers about
the infinite have often done. Nevertheless, I shall discuss the three ideas
separately, the first two in this chapter and the third (which requires the
most extended discussion) in the next chapter.
4.1 ‘An infinite series cannot be completed by

successive synthesis’
4.1.1 A quasi-Kantian theory

In his Critique of Pure Reason, Immanuel Kant offered the following argu-
ment for a beginning of the universe:
If we assume that the world has no beginning in time, then up to

every given moment an eternity has elapsed, and there has passed
away in the world an infinite series of successive states of things. Now
the infinity of a series consists in the fact that it can never be completed
through successive synthesis. It thus follows that it is impossible for an
infinite world-series to have passed away, and that a beginning of the
world is therefore a necessary condition of the world’s existence.1
1
Kant 1965, B454; emphasis added.
41
Kant was not really endorsing this argument (nor, however, was the
argument supposed to be a mere fallacy); his purposes in giving this
argument, however, do not concern us here. The italicized phrase in any
case has a certain intuitive plausibility as an account of (one species of)
the impossible infinite. I take it that what Kant means is that an infinite
series cannot be completed by going through each stage of the series, one
after the other (on the other hand, if one could somehow perform every
stage simultaneously, perhaps matters would be different). A number of
other philosophers have endorsed this view.2
Notice that the impossibility of completing an infinite series is here
represented, not, for example, as a contingent observation about our
world, but as a metaphysically necessary truth, perhaps even a conceptual
truth. Notice, too, that the claim is not that an infinite series cannot be
completed because the time required would be infinite – Kant does not,
for example, commit the mathematical error of assuming that the sum
of an infinite series of nonzero intervals must be infinite.3 It is known in
modern mathematics that an infinite series of numbers, each of which
is greater than zero, may have a finite sum, but this does not address
the problem. The problem is that, allegedly, it is conceptually impossible
to complete an infinite series because an infinite series, by definition, is
endless, and one cannot come to the end of something that is endless.
This view would seem to resolve several of the paradoxes of Chapter 3.
Puzzles about Thomson’s Lamp are immediately avoided: since the lamp
in principle cannot be switched infinitely many times, there is no need to
say whether it would end up on or off after infinitely many switchings.
In general, one need not answer questions of the form ‘If X happened,
would Y be the case?’ when X is conceptually impossible. (Compare the
silly question, ‘If 1 equalled 2, how much would 7×9 be?’) Similarly for
the Divided Stick paradox: since the stick cannot be divided infinitely
many times, there is no need to say what the pieces resulting from such
an infinite division would be like. For the Littlewood-Ross paradox: since
the game with the Banker could not be completed, there is no need to
address how much money one would have at the end. For Laraudogoitia’s
Marbles: since the infinite series of collisions could not be completed,
there is no need to puzzle about the creation or destruction of energy
and momentum that would allegedly result from its completion. For the
Spaceship: since the ship could not complete its process of acceleration
to infinite speed, there is no need to say where it would be once it had
2
Thomson 1954; Craig 1991; Wittgenstein 1975, p. 146.
3
Hume, on the other hand, did commit this error (1992, 29–30).
Impossible Infinite Series: Two False Accounts 43
done so. One version of the Martingale Betting paradox would also be
avoided, namely, the version in which you collect an infinite pile of
winnings within four minutes.
4.1.2 The failure of the theory

There are at least two problems with this attempted set of solutions.
First, the claim that an infinite series could not be completed leaves us
with a puzzle as to what exactly would stop it. Take Thomson’s Lamp:
the supposition that the lamp gets switched after intervals of 1/2n
minutes for n = 1, 2, 3, ... seems logically coherent. It entails that an infi-
nite number of switchings occur within one minute. If one thinks that
the infinite number of switchings cannot occur, one must say at what
point the series would stop. Will it have to stop after the one thousandth
switching? The one millionth? The one billionth? And what exactly will
stop it?
One might say: the physical properties of the switch will determine
how many times it can be switched. Depending on how it is constructed,
the switch will break at some point, sooner or later. This would of course
be true of any real, physical mechanism. But now the theory we’re
trying to evaluate claims that it is conceptually impossible to complete an
infinite series – we are supposed to see this to be true just in virtue of the
meaning of ‘infinite series’. It is hard to believe that the way to vindicate
this claim is to point to the fact that switches in the actual world eventu-
ally wear out. We need to know why it is conceptually impossible, for
example, for there to be a switch that would never wear out.
Second and more decisively, the quasi-Kantian theory supports Zeno’s
‘proof’ that motion is impossible. Here is one formulation of Zeno’s
argument:
1. To reach the ground, the ball must complete the following series of
motions in sequence: ½, ¾, ⅞, ... (where each of these numbers repre-
sents how far the ball would have gone after completing the corre-
sponding step in the series of motions).
2. This series is infinite.
3. It is impossible to complete an infinite series by traversing the steps
in sequence.
4. Therefore, it is impossible for the ball to reach the ground.
This is a pretty simple argument; there are not many places where one
could claim that it goes wrong. The logic appears unassailable, and there
are only three premises. Of these three premises, I cannot see how one
can deny #1 or #2. Therefore, if one affirms premise 3 – as the theory we

are now considering explicitly does – it is hard to see how one can avoid
concluding that motion is impossible. Or, to put the point the other way
around: motion is obviously possible. But every motion can be divided
into an infinite series of sub-motions, all of which are completed, one
after another in sequence, every time something moves. Therefore, an
infinite series can in fact be completed by successive synthesis.
Exactly where Zeno and Kant went wrong – what mistake causes
premise 3 to seem true – we discuss in a later chapter (Section 12.8).
4.2 ‘An infinite series of preconditions cannot be satisfied’
4.2.1 The theory

Here is another account of the impossible infinite: suppose there is an
infinite series of conditions or states of affairs, C1, C2, C3, ... . Suppose further
that each member of the series (except the first) is a precondition for the one
before it: C2 is a precondition for C1, C3 is a precondition for C2, and so on.
Equivalently, each member of the series is dependent upon the next one: C1
depends on C2, which depends on C3, and so on. In that case, none of the
conditions can be satisfied.4 A related view would be this: if each member
of the series would need to be explained by the subsequent member, then
none of the Ci can obtain, or none of them can be explained.
This view has some intuitive plausibility. Note that as we proceed through
the series, we never get any closer to completing it; each stage is always
infinitely far away from the end, just as every natural number is infinitely
less than infinity. One is tempted to say, therefore, that one is always infin-
itely far from satisfying all the conditions required for C1 to obtain; or that
one is always infinitely far from successfully explaining C1.
This view does not appear to help us with the paradoxes of Chapter 3.
It may, however, help us to assess the viciousness or virtuousness of the
regresses canvassed in Chapter 2. A reasonable case could be made that
four of the six regresses are vicious on the present theory:
i. In the regress of causes, each event is explained by and is dependent

upon its cause, and the occurrence of the cause is a precondition for
the occurrence of the effect.5 So, on the present theory, an infinite
regress of causes would be vicious.
4
Cameron 2008; Gillett 2003, 712–13.
5
The cause is of course not a logical, conceptual, or metaphysical precondition
for the effect (unless we individuate events by their causes). It is a causal precon-
dition. This statement is trivial; nevertheless, it suffices for the argument.
ii. In the regress of reasons, each belief’s justification is dependent upon

the justifiedness of the reason supporting it; for example, the justi-
fiedness of the reason for A is a precondition on the justifiedness of
A itself. So an infinite regress of reasons would be vicious.
iii. In the regress of Forms, at least on one interpretation, the Form
of Man is supposed to be required to explain the fact that various
particular men have something in common. It is their ‘participating
in’ the Form of Man that makes them men. Plato seems to have
regarded this participation in a Form as being a matter of similarity
to the Form, so indeed it would appear that an individual’s partici-
pating in the Form would need to be explained by that individual
and the Form both participating in some further Form. There thus
appears to be a vicious regress of needed explanatory conditions.
iv. The regress of resemblances, on the other hand, would apparently be
benign. The individual bluenesses, A, B, and C, are not held to depend
upon the resemblance relations, R(A,B), R(B,C), and R(A,C). Indeed,
things would seem to be the other way around – the existence of the
resemblance between A and B depends upon the existence of A and
B. A and B exist independently; the existence of the resemblance
between them stems from their existence.
v. In the regress of time series, McTaggart would likely claim that each
time series depends upon the next one to rescue it from incoherence.
For example, the time series in which ordinary events are ordered
would, on its own, be impossible because contradictory; it depends
for its very consistency on the existence of the time series in which
the presentness, pastness, and futurity of ordinary events are ordered.
Likewise, the time series in which the presentness (etc.) of ordinary
events are ordered depends upon another time series, in which the
presentness (etc.) of the presentness (etc.) of events are ordered. So
the regress would be vicious.
vi. The regress of truths is benign, for the fact that P is true is not a
precondition on the fact that P; rather, the fact that P gives rise to
the fact that P is true, which gives rise to the fact that it’s true that
P is true, and so on. [P is true] obtains in virtue of P’s obtaining, not
vice versa. So there is only an infinite series of consequences of P, not
an infinite series of preconditions.
4.2.2 The failure of the theory

All of this, however, is of somewhat academic interest, because the
present theory of the impossible infinite – that an infinite series of
preconditions cannot be satisfied – is false. Consider the beginningless
Zeno series, in which the ball makes the series of motions: ... , ⅛, ¼, ½, 1.
This is an infinite series of preconditions: to reach the endpoint, the ball
must first reach the halfway mark; to reach the halfway mark, it must
first reach the one quarter mark; and so on. Nevertheless, all of these
conditions are in fact satisfied, every time an object moves.
Or consider a line segment one meter long. The segment is made up
of a left half and a right half. Typically, when a thing is made up of
parts, we think that the thing depends for its existence on the exist-
ence of those parts, and that the thing exists at least partly in virtue
of those parts existing. So the one-meter line depends for its existence
upon, and exists at least partly in virtue of, the existence of the left half
and the right half. Similarly, the left half depends upon the left quarter
and the second quarter; the leftmost quarter depends upon the leftmost
eighth and the second eighth; and so on. So there is an infinite series of
dependencies; nevertheless, all these objects exist. The same point can
be made using a time interval of some specified length.
Now, what was wrong with the argument in favor of this theory of
the impossible infinite? Earlier, we said that as one goes through the
infinite series of preconditions, one is at every stage still infinitely far
from reaching the end, and this is supposed to support the conclusion
that none of the conditions can obtain. Perhaps the thought is that the
first condition, C1, ‘starts out’ not obtaining, and will only get to obtain
if we can get to the end of the entire infinite series of conditions. And
perhaps the argument borrows from the notion, discussed in Section
4.1, that it is impossible to complete such a series, since the series is
endless.
This thinking would be understandable if the members of the series
were actions occurring in a temporal sequence, and if each action
required at least some minimum duration. Then it would be correct
to say such things as ‘we are always infinitely far from completing the
series.’ But (except in the case of the regress of causes; see below) we are
not generally dealing with a series of temporal events. What is needed
for C1 to obtain is simply for C2 to obtain, which in turn requires C3 to
obtain, and so on; so all told, what C1 requires is that every member of
the series obtains. But the members do not need to obtain at different
times, in a certain temporal order; they may all simply obtain simultan-
eously (or they may all obtain timelessly). Nor is there any need for any
human observer, or anyone else, to recognize or otherwise go through
the members of the series. For example, in the infinite regress of Forms,
there is no reason why anyone would need to identify each of the Forms
in the series, or do anything else involving them that might count as
‘going through’ the series. The Forms themselves, if they exist, would all
simply exist timelessly and independently of observers. Of course, this
is not to say that the notion of an infinite series of Forms is plausible.
The point is simply that it is not at all clear why there could not be an
infinite series of objects standing in dependence relations, each one to
the next.
In the special case of the infinite regress of causes, there is a sequence
of events that occur in a certain temporal order. Here, at least, do we
have a problem? Well, it is really not clear what the problem would be.
In most cases, an infinite series of events occurring one after another
in time would be problematic because an infinite duration would be
required to complete the series. For example, we should not postulate a
belief arrived at by an infinitely long chain of argument, because it would
take a person forever to complete such an argument – and we don’t have
forever! But in the case of the regress of causes in the universe, we in fact
have an infinite time to work with – time stretches back infinitely into
the past, and the infinite series of causes occupies exactly that span of
time. Why may not an infinite series of events transpire in an infinite
amount of time?
A defender of the First Cause Argument might object that, if there was
no first cause, then we can never truly explain any given present-day
event, because to explain it, we would have to cite every cause in the
infinite chain preceding that event. Now, it may be true that one cannot
fully explain an event without citing all the causes in its history – but
then, I think it should come as no surprise that we in fact cannot fully
explain events in this sense.
Perhaps the defender of the First Cause Argument would say that there
must at least exist a complete explanation for any given event, even if
we humans cannot give the explanation. Now, it really is not clear to me
why we should assume that everything must have a complete explana-
tion; why may there not simply be brute facts? In addition, it remains
unclear why a present event should not have a complete explanation,
even in a world with no first cause. Certainly, no finite portion of the
series of causes would explain the event. But what about the entire infinite
series of causes in the event’s history – why shouldn’t we say that that is
the complete explanation?
4.3 Conclusion
Since they are both false, the theories considered in this chapter cannot
help us resolve the paradoxes of infinity. An infinite series of events
can be completed. A series that is endless in the sense of having no last

member can nevertheless be completed, in the sense that there may be
a time such that every member of the series occurs before that time.
Similarly, a series with no first member can nevertheless be begun, in
the sense that there may be a time such that every member occurs after
that time. These points are illustrated by the endless and beginningless
Zeno series.
It also seems that an endless series of preconditions, with each member
dependent on the next, can be satisfied. It is not clear why one should
think this impossible, and the possibility seems to be illustrated by the
beginningless Zeno series, as well as any region of space or interval of
time, where the existence of the whole depends at least in part on the
existence of each of its parts.
5
Actual and Potential Infinities
5.1 The theory of potential infinity
Historically, the most popular theory of the impossible infinite is the

one handed down by Aristotle: that infinity can only be potential,
never actual.1 What does this mean? Here are two helpful remarks from
Aristotle’s discussion of the infinite:
A quantity is infinite if it is such that we can always take a part outside

what has already been taken.
Our account does not rob the mathematicians of their science [ ... ].
In point of fact they do not need the infinite and do not use it. They
postulate only that the finite straight line may be produced as far as
they wish.2
The latter passage is particularly instructive. The idea appears to be

this: there is an infinite potentiality, because there is no limit to how
long a line can be. A line cannot be infinitely long – this would be an
‘actual infinity’ – but a line can be of any finite length. The possible
lengths of lines go on up forever – for any line, there could be one
twice as long.
1
Compare Wittgenstein 1975, 162 (emphasis mine): ‘[I]t makes sense to say
there can be infinitely many objects in a direction, but no sense to say there are
infinitely many.’
2
Aristotle 1941, Physics III, 207a8–9; 207b27–31. The second remark is
misleading, since as it turns out, Aristotle thought space had a finite size, which
implies that a straight line may not in fact be extended as far as one wishes – it
cannot be extended beyond the sphere of the heavens (De Caelo II.7, II.9).
49
Here are some more examples that I think Aristotle would approve of:
i. A one-meter line segment is infinitely divisible, but not infinitely

divided. That is, there is no limit to how many times one can divide
it; however many times a line is divided, it is always possible to
divide it again. But one will never have divided it infinitely many
times. Hence, this is a potential, not an actual infinity.
ii. The series of natural numbers is potentially infinite, because there is no
limit to how high one can count. But however high one has counted,
it will always be to a finite number. There is no number ‘infinity’.
iii. There is no limit to how fast something can move (remember that
Aristotle is writing millennia before the special theory of relativity) –
for any possible speed, there are possible speeds twice as fast. But
there is no such speed as ‘infinity’. (If there were, where would you
wind up after traveling at such a speed?)
In saying that there are potential infinities but not actual infinities,
Aristotle is not saying that there are things that are capable of being
infinite, yet somehow they never exercise this capacity and so never
become infinite. He is saying that there are some potentialities that are
unlimited – sometimes, there is no particular limit to how great some-
thing can be in some respect – but nevertheless, no matter what happens,
every quantity will always have some definite, finite value.
Many other thinkers have held some variant of this view. Indeed,
some sort of finitism has been the dominant position in the history of
the subject. A few more quotations:3
[T]here cannot be any infinite magnitude, since no species of magni-

tude is infinite. [ ... ] [I]t is impossible for an actually infinite multi-
tude to exist, even accidentally. (Saint Thomas Aquinas, ca. 1270)
Whatsoever we imagine, is finite. Therefore there is no idea or

conception of any thing we call infinite. [ ... ] When we say any thing
is infinite, we signify only that we are not able to conceive the ends
and bounds of the thing named. (Thomas Hobbes, 1651)
The infinity of numbers, to the end of whose addition every one

perceives there is no approach, easily appears to any one that reflects
3
Sources: Aquinas 1981, Summa I, Q7; Hobbes 1996, 23, emphasis in original;
Locke 1975, II.xvii.8; Gauss, letter to Schumacher in 1831, quoted in Benardete
1964, 13.
Actual and Potential Infinities 51
on it. But, how clear soever this idea of the infinity of number be,
there is nothing yet more evident than the absurdity of the actual
idea of an infinite number. (John Locke, 1689)
I must protest most vehemently against your use of the infinite as

something consummated, as this is never permitted in mathematics.
The infinite is but a façon de parler; an abridged form for the state-
ment that limits exist which certain ratios may approach as closely as
we desire, while other magnitudes may be permitted to grow beyond
all bounds. (Carl Friedrich Gauss, 1831)
5.2 Why not actual infinities?
Why can’t there be an actual infinity? Why should there not be an infi-
nitely long line, an infinite number, or an object of infinite mass?
The following brief passage from Aristotle’s Physics gives a sense of the
fundamental philosophical motivation for rejecting actual infinities. The
context was a discussion of whether there is any infinitely large object.
Surely what is in a specific location is in space, and what is in space

is in a specific location. Just, then, as the infinite cannot be a quan-
tity – that would imply that it has a particular value, e.g., two or three
cubits; quantity just means these – so a thing’s being in space means
that it is somewhere, and that is either up or down or in some other
of the six directions: but each of these is a limit. It is plain from these
arguments that there is no body which is actually infinite.4
4
Aristotle 1941, Physics III.5.206a2–8. Here, I have boldly/rashly (since I do not
read Greek!) modified the translation of R.P. Hardie and R.K. Gaye in four ways: (i) I
have changed ‘in place’ to ‘in space’, because the former expression does not make
sense in the context, notwithstanding that both Ross’ (Aristotle 1936, 366) and
Coughlin’s (Aristotle 2005, 56) translations agree with Hardie and Gaye. Aristotle
presumably was not talking about what normal English speakers call something’s
being ‘in place’ (as when, for example, one puts procedures in place for preventing
fraud). (ii) I have changed ‘special place’ to ‘specific location’. Aristotle presumably
was not talking about what we would call ‘special places’ – like a nice, secluded spot
in the woods by a babbling creek. (iii) I have changed ‘it has a particular quantity’
to ‘it has a particular value’, again for coherence. The hypothesis that infinity is a
quantity does not imply that infinity has a quantity. In English, quantities don’t
‘have’ other quantities (nor do they ‘have’ themselves). (iv) Finally, I changed ‘the
six differences of position’ to ‘the six directions’, because it does not make sense in
English to speak of a thing being ‘in’ a ‘difference of position’. In this last change, I
follow Ross and Coughlin. This experience raises the question of whether Aristotle’s
translators perhaps know Greek better than they know English.
Saint Thomas Aquinas gives a similar argument:
[E]very natural body has some determined substantial form. Since

therefore the accidents follow upon the substantial form, it is neces-
sary that determinate accidents should follow upon a determinate
form; and among these accidents is quantity. So every natural body
has a greater or smaller determinate quantity. Hence it is impossible
for a natural body to be infinite.5
Here is what I make of these passages: Every actual thing in the world
must be fully determinate in respect of all its properties. For example, if
a thing is in space, then it must have a specific location; there cannot
be, as some interpretations of quantum mechanics posit, an object that
is in a certain box, but neither in the left half nor in the right half. Or,
if a thing has mass, then there must be some definite value of its mass.
(Of course, there could be things to which mass and location do not
apply, such as thoughts or numbers – that is why I included the clauses
‘if a thing is in space’ and ‘if a thing has mass’.) However, any definite
value of a variable constitutes a limit – to be in a particular place, or to
have a particular mass, is to be limited in a certain way. The infinite,
by definition, is unlimited. So to be infinite in respect of some variable
is incompatible with having a determinate value of that variable – for
instance, to say that a thing has infinite mass would be to deny that it
has any specific, determinate mass. Therefore, no thing can actually be
infinite in respect of any variable; for instance, no thing can actually
have infinite mass.
Whether that is a correct paraphrase of Aristotle or not, the above
represents an interesting motivation for rejecting the actual infinite,
so let us take that to be the theory under consideration. I shall later
(Chapter 10) discuss some further arguments that ‘infinity’ does not
name any determinate quantity.
5.3 Infinite divisibility
5.3.1 Aristotle on Zeno’s paradox

Of the paradoxes mentioned in Chapter 3, Aristotle directly addresses
those due to Zeno (the other paradoxes being unknown in his time). He
5
Aquinas 1981, Summa I, Q7. Ignore the bits about accidents and substantial
forms, if you don’t know what those are. The last two sentences are the important
part.
makes a number of remarks about Zeno, but I shall focus on his remarks
apropos of the distinction between actual and potential infinities.
Aristotle concedes that, because there cannot be an actual infinity,
Zeno’s endless series would be uncompletable, if indeed it were an
actual infinity. He maintains, however, that the endless series of halfway
motions is a merely potential infinity. The distance between point A and
point B, he says, is capable of being divided as many times as one likes.
However, it is never actually divided infinitely many times.
On the face of it, this claim is a bit obscure. The spatial intervals that
Zeno identifies – the first half of the distance from A to B, the next
quarter of the distance, and so on – appear to be actual segments; it is
not as though some of them are merely capable of coming into exist-
ence but don’t actually exist. (What would it mean for a part of space
to fail to exist?) The suggestion that the line might be capable of being
divided some number of times yet fail to be ‘actually divided’ so many
times makes it sound as though ‘dividing’ a line is a human activity, and
as though the segments Zeno is talking about will not exist (or the ball
will not have to traverse them?) unless some human being performs this
‘dividing’ activity to bring them into existence.
But that is quite a strange suggestion. For an object to travel from
point A to point B, it is necessary that it travel half the distance before
it may reach the endpoint, and this fact is just a matter of the objective
structure of space (given the assumption that the object must take a
straight, continuous path). It does not depend on any human beings,
and it does not matter whether any human being has recognized the
halfway point, or performed any other activity by way of ‘dividing’ the
distance. Even if there were no human beings in the universe, an object
still could not move from point A to point B without traversing the
first half of that distance. And the object would not merely need the
capacity to travel the first half of the distance; it would have to actually
traverse that distance, before it could actually make it to point B (barring
teleportation).
In fact, I think Aristotle’s point is not the confused one we have just
refuted. What is required for a division to be actual, on Aristotle’s view,
is not some mental act by observers. Rather, for the ball’s motion to be
‘actually’ divided into two halves, it would be necessary that the motion
should be discontinuous; specifically, that the ball should pause at the
halfway mark before continuing on its journey.6 Thus, for the endless
Zeno series to count as an actual infinity, on Aristotle’s reckoning, the
6
Aristotle 1941, Physics VIII.8, 263a23–b10.
ball would have to stop and start infinitely many times – and in that
case, Aristotle would indeed consider the series uncompletable (here I
would agree with Aristotle). If, as is normally the case, the ball’s motion
is continuous (with no pauses), then it is really just a single movement,
not an infinite series of movements.
5.3.2 The failure of Aristotle’s solution

Why think that the ball must pause at the halfway mark in order for the
division to be ‘actual’? Aristotle’s stated reason for this is that if there is
no pause, then the first stage in the series (the ball’s traveling the first
half of the distance) would overlap with the second stage in the series
(the ball’s traveling the next quarter of the distance). This overlap would
consist of a single point: the temporal endpoint of the first stage would
be identical with the beginning point of the second stage. This prevents
the alleged two stages from being genuinely separate motions; instead,
there is really just one motion.
If the single point of overlap is the concern, then there are a number
of ways that Zeno could avoid it. Zeno could redefine the stages so that,
for example, the first stage is defined as including the ball’s motion from
point A up to but not including the halfway mark; that is, it is the motion
over the half-open interval, 0 ≤ d < ½. The second stage of the ball’s
motion would then include the points where ½ ≤ d < ¾, and so on. Here,
however, Aristotle might still say that the division is non-actual; he
might insist that an actual division between motions requires a nonzero
temporal gap between them.
Another option for Zeno would be to say that the infinite series of
events consists in (i) the ball’s reaching the halfway point (an event
which occurs at a single instant of time), (ii) the ball’s reaching the three
quarter mark, and so on. Then each item in the series is separated from
the next by a nonzero temporal gap, and it remains the case that the
ball cannot reach the ground before all infinitely many of these events
have occurred.
In any case, it remains unclear why we should accept the Aristotelian
strictures on what may count as distinct motions. What is wrong with
the idea of two distinct actions that are temporally contiguous, or even
that overlap at some point in time? The ball’s traveling the first half
of the distance is an event that actually occurs (assuming, that is, that
objects really move), regardless of what may happen after that. If the
ball stopped at the halfway mark, then Aristotle would certainly accept
that this event had occurred. But the ball’s going on to traverse the
next quarter of the distance can have no effect on whether the event of
its traveling the first half of the distance occurred. So the first halfway
motion actually occurs, whether or not the ball continues moving
without delay. The same is true of each of the motions in Zeno’s series. So
there is an infinity of actual motions, that is, there is an actual infinity.
I have no doubt that it is valid to regard the ball’s motion from A to
B as a single event. I merely insist that this is not the only valid way to
regard things; it is also legitimate to view the ball’s motion as consisting
of two events, or three events, and so on. The world does not contain a
privileged way of dividing it up such that all other ways are ‘false’. Now,
some ways of dividing (or not dividing) the ball’s motion fail to produce
any paradox – if we view it as a single event, no paradox is evident. But
this hardly solves our problem. If there is any legitimate (not objectively
wrong) way of regarding the ball’s motion that generates a paradox,
then we have a problem.
5.3.3 Minimum distances

Perhaps we can avoid Zeno’s paradoxes by postulating indivisible parts
of space, so that at some point, it becomes impossible to further divide
the distance between A and B, and then the ball can simply traverse the
smallest distance in one step. Of course, in standard geometry, there
are indivisible parts of space – they are called ‘points’ – but they will
not serve the present purpose, because they each have a size of zero,
enabling a line segment to be divided into infinitely many of them. The
minimum parts of space in the present view would have to have some
real, nonzero size, such that an ordinary, macroscopic distance could be
divided into only finitely many of these smallest parts.
I must confess that I have difficulty conceiving this alleged possibility.
I cannot understand what an indivisible – yet nonzero! – distance would
be like. The hypothesis that, for example, a millionth of an inch might
be indivisible seems to me to make as much sense as the hypothesis that
the number 1/2,000,000 might fail to exist, even though 1/1,000,000
does, or the hypothesis that there is a real number that cannot be multi-
plied by two.
Nevertheless, I may be able to conduct some reasoning concerning
this hypothesis. Most people who entertain the ‘minimum distance’
theory agree that the following is an implication of the theory: for
any two points in space, the distance between those points is equal to
some integer multiple of the minimum distance. Now, I suppose that
a square may exist, and that it may have four corners (Figure 5.1a).
Call the length of the side of the square (the distance between adjacent
corners) ‘s’, and call the length of the diagonal (the distance between
d d
s s
(a) (b)
Figure 5.1 Relationship between the side and the diagonal of a square
opposite corners) ‘d’. Let the hypothetical minimum possible distance

be m. On the present theory, s is some integer multiple of m, and d is
another integer multiple of m. Therefore, the ratio of the diagonal to the
side, d/s, is a ratio of two integers (since m cancels out). That is, d/s is a
rational number.
But that is false. Consider the argument given by Plato in the Meno.7
Construct a square whose base is the diagonal of the original square
(Figure 5.1b). The area of the small square is s2, and that of the large
square is d2. The area of the large square is also twice that of the small
square (the small square is made up of two triangles, where the large
square is made up of four triangles of the same size). So d2 = 2s2, and so
d/s = √2. Unless we’re radically revising our arithmetic along with our
geometry, √2 is no rational number.
My claim here is not that the theory of minimum distances could not
be made consistent; a sufficiently dogmatic partisan could, for example,
simply deny that squares have diagonals. My claim is that this view
simply is not plausible.
I conclude that space provides us with an example of an actual infinity,
because every nonzero spatial interval actually contains infinitely many
parts. In fact, there are infinitely many sets of infinitely many parts –
that is, there are infinitely many ways of dividing a line segment into an
infinity of parts. Similarly, every volume of space, every interval of time,
and every continuous motion contains an infinity of parts.
7
Plato 2012, 84d–85b. Plato does not make the point about irrational number;
he just gives the argument that the square constructed on the diagonal is of twice
the area as the original square.
5.4 Infinite time
In addition to its infinite divisibility, time exhibits another kind of

infinity, its infinite extent. In particular, time extends infinitely far into
the future. In contrast to the infinite past, no one seems bothered by an
infinite future. Perhaps an Aristotelian could reconcile this by saying
that the future is not (yet) actual. Although there is an infinite amount
of time ahead of us, the future times are not actual until they occur, and
there will never come a time at which they have all occurred, so the
infinite future will never be actualized. It is unclear whether this defense
of the theory of potential infinity succeeds, mostly because the notions
of ‘actuality’ and ‘potentiality’ are unclear.
Time also extends infinitely far into the past. This should be more
troubling to an Aristotelian, since it is more plausible to say that the past
is actual as of the present time than it is to say that the future is actual
as of the present time.8 Interestingly, Aristotle did not accept the notion
that time had a beginning, as his rejection of actual infinities might
lead us to expect. He argues that time is without beginning, though he
neglects to explain why this does not count as an actual infinity. He
likewise maintains that changes have been occurring throughout the
eternal history of the universe, which would seem to constitute another
actual infinity.9
Perhaps Aristotle would deny that these are actual infinities, on the
grounds that the infinitely many past moments and past events do not
exist all at once. But if this were his view, it would be hard to understand
his treatment of Zeno’s paradox. There, he seems to allow that the motions
in Zeno’s series would be actual if they were separated by temporal gaps
(pauses); this is incompatible with the idea that the actual must all exist
simultaneously. It is in any case quite odd to view events that have already
occurred as mere potentialities, rather than actual events.
5.5 Infinite space
5.5.1 The edge of space

Space, too, appears to be an actual infinite, as it extends infinitely far
in all directions. Aristotle denies this, holding that the cosmos consists
of a sphere of finite circumference, with the Earth at the center and
8
Tooley 1997; Craig 2010.
9
Aristotle 1941, Physics III.6, 206a7–10; VIII.1.
the heavens as the outermost spherical shell. He claims that there is

nothing, not even empty space, outside this outermost shell.10
Pace Aristotle, there is no edge of space. Even if there were some
outer shell of the material universe that the laws of physics somehow
prevented one from escaping, one could still at least refer to the very
beyond into which one was prevented from traveling, and this suffices
to conclude that at least there would be space there.
Someone could insist that no, one could not refer to or think about
the beyond in this case. I just find this position obviously false. I find
it inconceivable that there be a sphere (or any other finite region) such
that one could not even refer to the places outside of it. For those who
claim to conceive of such a possibility, the best I can make of it is that
they are actually thinking of a region surrounded by empty space – but
(like Aristotle11) they may have some philosophical theory that tells them
that there cannot be empty space, so they declare that in this scenario
space itself would be limited.
5.5.2 Round space

In modern times, a different theory is sometimes advanced to explain
how space might be finite. The modern theory supposes that space might
have a finite extent and yet there be no edge of it. On this theory, if you
tried to travel a great distance in one direction, instead of hitting an
edge of space and being forced to stop, you would find yourself coming
back to where you started from, as though you had gone around in an
enormous circle – and yet you were truly traveling in a straight line the
whole time.
People who like this idea usually illustrate it using the analogy of the
surface of a sphere. Suppose we did not know that the Earth was round,
and someone undertook to walk in a straight line on the Earth’s surface
for as long as possible, never to return. After walking long enough (and
supposing that he could walk across the ocean), he would be surprised
to find himself returning to where he started, coming from the opposite
direction. The Earth’s surface is of finite size but it has no edge, so if
you try to walk a great distance on it, you will find yourself, instead of
falling off the edge, just coming back to where you started. Similarly, it is
said, the whole of physical space might be like the surface of the sphere:
maybe it has a finite size, but no boundaries.12
10
Aristotle 1941, De Caelo I.9, 278b–279a.
11
Aristotle 1941, Physics IV.6–9.
12
Einstein 1961, chapter 31.
Many people take this analogy quite seriously, and many students
when first introduced to it seem to find it quite clever. Yet it always
struck me as the silliest sophistry. (Well, perhaps not the silliest – there
is after all quite a lot of competition in that category – but certainly
high in silliness.) There is no difficulty in understanding how the
man who walks around the world winds up back where he started: he
walked in a circle. A very large circle, to be sure, but a circle nonethe-
less. We were promised an explanation of how it is possible that by
traveling in a straight line, one should wind up back where one started.
How is that demonstrated by giving an example of someone walking
in a circle?
The standard answer: ‘The man’s path actually was a straight line on
the surface of the Earth. Among paths on a spherical surface, the shortest
path connecting any two points is an arc of a great circle. Therefore,
within a spherical surface, a great circle is a straight line.’
What this argument does is essentially (i) to restrict our attention to only
a certain set of lines, all of which are curved (every line drawn on a globe is
curved), then (ii) to identify the least curved among these and declare them
to be straight. Here is an analogy: consider points A and B in Figure 5.2,
and consider just the three lines L1, L2, and L3 connecting them:
L1
L2
A B
L3
Figure 5.2 Three lines connecting two points
Remember, I’m stipulating that you’re only allowed to consider those

three lines. With that understood, L2 is the shortest path between A and
B. Therefore, L2 is a straight line.
I hope you agree that that is false. And it’s not that L2 is straight ‘within
that set of lines’; it just is not straight at all. Similarly, the path around
the world is not straight at all.
Even after acknowledging this point, there are at least three ways we
might interpret the point of the spherical surface example so that it
might still show something of interest. Here they are, along with my
replies:
i. Perhaps the example of the spherical surface is merely meant to

demonstrate that the concept of a region that is finite but unbounded
is not contradictory. Since it is not contradictory, the whole of phys-
ical space might be finite but unbounded.
Reply: the last sentence is a non-sequitur. That the predicate ‘finite
but unbounded’ is not a contradiction provides no reason to believe
that space might be finite but unbounded.
ii. Perhaps the example is meant to suggest that our space might be
embedded in some higher-dimensional space, and that our space
might be curved in the higher-dimensional space. If so, what look
like straight lines to us might actually be curved in the larger space,
just as what look (to a two-dimensional being) like straight lines on
the surface of a sphere are really curved. In that case, by traveling
far enough in what looks to us like a straight line, we might wind up
back where we started.
Reply: On this hypothesis, our three-dimensional space would be
finite, but then the larger space would be the relevant example of
an actual infinity – unless of course it too is only a finite portion of
some even higher dimensional space. But then the latter would be
the relevant example of the actual infinite. And so on. The example
fails to motivate the idea that a genuinely straight line can double
back on itself; more importantly, it fails to evade the notion of the
actual infinite.
iii. Perhaps the suggestion is that even if our space is not embedded in a
larger space, and what look to us like straight lines are really straight,
it still might turn out that a straight line, if sufficiently extended,
comes back around on itself. This is because even though our space
is not curved in some higher-dimensional space, it still might have
the same geometry as a space that was so curved.
Reply: But this is exactly what the example fails to show. We can
readily imagine the geometry of a spherical surface, by actually imag-
ining a sphere. But if someone says: ‘Okay, now imagine a surface
with exactly that geometry, including the part about how a straight
line doubles back on itself, but this time don’t let the surface curve’ –
well, this is precisely what we cannot imagine. And we have been
given no reason at all for thinking this possible. Nor have we been
given any reason for thinking that our space can have that geometry
without being embedded in a higher-dimensional space.13
Now suppose it turned out – as, empirically, it might – that when-

ever someone tried to travel in a straight line for some very long
distance, he always wound up back where he started. In that case, I
still would not say that space was finite. I would say that something
was preventing anyone from actually traveling such a distance in a
straight line, that is, that all these travelers were actually moving in
circles (or other closed curves). Because of this, some regions of space
might be inaccessible to us. But this, as a metaphysical possibility, is
hardly surprising, nor does it provide any reason for thinking that
space itself might be finite. For an analogy, imagine a person inside
the event horizon of a black hole: it is physically impossible for this
person ever to escape the black hole, but they would be wrong to infer
that there is nothing beyond. In the same way, it might be physically
impossible for anyone in our physical universe to escape beyond a
certain distance, but we would be wrong to infer that there is nothing
beyond.14
I have not tried to prove that a straight line does not double back
on itself, as that is self-evident. What I have been employed about is
pointing out that no good reason for doubting this proposition has been
given. In particular, the analogy of the great circle on the surface of a
sphere provides no reason at all for doubting it, and accordingly casts no
doubt on the idea of space as an actual infinite.
5.5.3 Relational space

How else might we avoid admitting that space is an actual infinite?
Instead of claiming that space is finite, we could claim that it is a merely
13
What about the empirical evidence for General Relativity, such as the
gravitational bending of light (Dyson et al. 1920)? Properly interpreted, this is
not evidence of curved spacetime but rather evidence that light fails to travel
straight paths in spacetime. See Reichenbach’s (1957, section 3) discussion of the
‘universal force’ theory. Pace Reichenbach (22), I think not only that this sort
of theory should not be ruled out by definition, but that it is in fact the correct
interpretation of the evidence.
14
There is much more to say on this subject, because the theory about finite
but unbounded space is tied in various ways to broad philosophical theories (espe-
cially empiricism and verificationism) that held sway in the twentieth century.
These theories were held so widely and so dogmatically among philosophers,
scientists, and mathematicians, that great portions of the edifice of modern
thought in all three areas were built around them. Unfortunately, these philo-
sophical theories are fundamentally false; however, a full treatment of them
would require a separate book. I will touch on them again in Chapter 7.
potential infinity. Maybe points in space – at least the unoccupied

points – exist as mere potentialities, rather than actualities.
According to the relational theory of space, defended famously by
G.W.F. Leibniz, space has no independent existence.15 What exists are
physical objects and the spatial relationships among them. All sensible
discourse about space is really discourse about (possible and actual)
spatial properties and relationships of bodies. When an object is said to
be moving at 100 kilometers an hour, what this means is that its posi-
tion relative to some salient object, such as the Earth, is changing at that
rate.
What about when we talk about regions of empty space? The rela-
tionist (as I shall call defenders of this theory) might say that talk about
empty space is really talk about unactualized potentialities for objects to
stand in certain spatial relations. For instance, suppose I claim to have
a bottle with a perfect vacuum inside. Maybe what this really means
is that there is the unactualized potential for physical material to be
between the sides of the bottle.
On this view, the fact that space is infinitely extended might just be the
fact that, for any given distance one cares to name, there is the (perhaps
unactualized) potential for a material object to be at that distance from
here. Thus, the infinitude of space is a potential infinity.
This would be a terrific way of rescuing Aristotle from the example of
space as an actual infinite ... if the relational theory of space were plau-
sible. In the beginning of his famous Principia, Newton describes the
following scenario.16 A bucket of water is suspended from a rope. The
bucket is rotated several times, so that the rope is twisted up. At this
time, the bucket and water are at rest relative to one another, and the
surface of the water is level – call this Stage 1 (see Figure 5.3). Then
the bucket is released. As the rope untwists, the bucket rotates. At first,
the bucket is rotating relative to the water, and the surface of the water is
level (Stage 2). Then the water starts to rotate as well, so that bucket and
water are again at rest relative to one another. The surface of the water
is now concave, with the water trying to climb the sides of the bucket
(Stage 3). Subsequently, the bucket will start to rotate back in the other
direction, so that now the bucket is rotating relative to the water, in the
direction opposite from the initial direction, and the surface of the water
is still concave (Stage 4).
15
See Leibniz’ letters in Leibniz and Clarke 2007.
16
Newton 1846, Book I, 81, from the scholium to the definitions.
1 2 3 4
Figure 5.3 Newton’s bucket
Here we have an observable phenomenon to be explained: what

causes the surface of the water to be concave in Stages 3 and 4, but flat
in Stages 1 and 2? I assume that the only physical objects involved in the
explanation are the water, the bucket, the rope, and perhaps the Earth
beneath them. (The phenomenon is not caused by, say, a kangaroo in
Australia, or the planet Neptune, or the Andromeda Galaxy.17 It is logi-
cally possible that these things are causally implicated, but not plau-
sible.) The concavity of the water in Stages 3 and 4 is a manifestation
of ‘centrifugal force’, which is due to the rotation of the water.18 The
details of the correct mechanical explanation do not concern us here.
What concerns us is that the phenomenon cannot be explained in terms
of the rotation of the water relative to the bucket, since we have rela-
tive rotation in Stage 2 while the water is flat; we also have a lack of
relative rotation in Stage 3 while the water is concave. It appears that
the phenomenon must be explained in terms of the water’s absolute
rotation. But this notion of absolute rotation presupposes the existence
of space, as a thing distinct from the physical objects occupying it. In
Stages 1 and 3, the water is in the same spatial relationship to the bucket
(in both cases, the two are rotationally at rest relative to one another);
what is different is the relationship of the water to space itself: in Stage 1,
17
This is a gentle dig at Ernst Mach (1919, 232), who proposed that the
phenomenon was to be explained in terms of rotation relative to the fixed stars –
hence, the Andromeda Galaxy inter alia was implicated in pulling the water up
the sides of the bucket.
18
Centrifugal force is referred to as a ‘fictitious force’ in physics, since it
appears only in non-inertial reference frames; nevertheless, the term describes a
real phenomenon, as illustrated by the observably different shapes of the water in
Newton’s bucket experiment. That phenomenon needs to be explained, whether
it be described in terms of a ‘force’ or not.
the water is rotationally at rest in absolute space; in Stage 3, the water is

rotating in absolute space. That is why there is centrifugal force in Stage
3 but not Stage 1. Similar points apply to Stages 2 and 4.
But wait – didn’t Einstein’s Theory of Relativity refute the whole
notion of absolute space? No, despite the similarity between the words
‘relativity’ and ‘relational’ (the first six letters are the same!) the Theory
of Relativity is not, and does not imply, the Relational Theory of Space.
The Theory of Relativity is best understood as an account of the math-
ematical structure of spacetime – roughly speaking, that we occupy a
Minkowski spacetime.19 On this account, there is no objective distinc-
tion between rest and uniform (non-accelerated) motion – or in other
words, between vertical and slanted lines in spacetime. There is, however,
an objective distinction between uniform and accelerated motion, or
between straight lines and curved lines in spacetime. It is that distinc-
tion that is used to explain the bucket phenomenon: in modern physics,
rotational motion is considered a form of accelerated motion, even if
the speed is constant (this is because rotation involves continual change
in direction of motion), and acceleration is absolute. (Yes, it is generally
held that velocity is relative to a reference frame, and yet the rate of
change of velocity is absolute. This is odd but not contradictory.) It is
because the water is rotating absolutely in Stages 3 and 4 that there is the
centrifugal force phenomenon. There is no centrifugal force in Stages 1
and 2 because the water is objectively not rotating. That was Newton’s
analysis, and that is still the modern analysis. The contemporary view,
post Relativity, disagrees with Newton’s when it comes to absolute veloci-
ties, but the two concur on the question of absolute accelerations.
And absolute acceleration is all we need to reject the relational theory
of space (or spacetime). The relational theory of space holds that space
(or spacetime) does not have independent existence, that it exists solely
as a set of possible spatial (or spatiotemporal) relationships among phys-
ical objects. The notion of absolute acceleration is incompatible with the
19
This is of course speaking just of Special Relativity. For an excellent exposi-
tion of spacetime structure in Special Relativity, see Maudlin 2002, ch. 2. Maudlin
(1993) proposes that a relationist could respond to Newton’s bucket scenario by
positing cross-temporal spatial relations, for example, a particle of water at t1
might be at a distance from itself at t2. However, he nevertheless concludes that
General Relativity supports substantival spacetime. For discussion of how General
Relativity supports substantival spacetime, see also Grünbaum 1957; Dainton
2001, 301–2 (arguing that relationists cannot make sense of gravitational waves).
For additional, philosophical arguments for substantival spacetime, see Tooley
1997, 258–82, 360–71.
theory, since the absolute acceleration of an object cannot be defined in

terms of its relationships to other physical objects. Relationists can only
recognize relative accelerations – that is, one physical object can accel-
erate relative to another. In brief, the relational theory contradicts both
classical and modern physics. We therefore cannot rely on the relational
theory to rescue us from the notion of space (or spacetime) as an actual
infinite.
5.6 Infinitely numerous numbers
The set of natural numbers is proverbially infinite, that is, there are infin-
itely many of them. Aristotle would not wish to say that there are only
finitely many natural numbers and hence that there is a last natural
number in the series.20 His view must be, then, that the infinitude of
the natural numbers is merely a potential, rather than an actual infinity.
How might one motivate this?
One possibility would be to claim that the infinity of the natural
numbers consists in the fact that there is no limit to how high one can
count. That is, for any natural number n, it is possible to count higher
than n; one cannot, however, count to infinity. This interpretation is
supported to some degree by Aristotle’s remark:
[I]n the direction of largeness it is always possible to think of a larger

number: for the number of times a magnitude can be bisected is infi-
nite. Hence this infinite is potential, never actual.21
This approach, however, depends upon an overly anthropocentric

conception of number, such that a number does not exist until someone
counts it off or otherwise recognizes the number. On this view, it might
turn out, for example, that there is no natural number between 45,670
and 45,672, because maybe no one has ever named that number. Note
that it would not suffice to hold that a number may exist in virtue of
our accepting propositions that entail its existence – for instance, the
opponent of actual infinities could not say that the successor of 45,670
exists in virtue of our accepting that 45,670 exists and our accepting the
20
Aristotle 1941, 206a7–11.
21
Aristotle 1941, Physics III.7, 207b10–12. Aristotle is no subjectivist in general,
so I find the interpretation suggested in the text improbable qua Aristotelian
exegesis. Nevertheless, the suggestion bears discussion since later opponents of
actual infinities might endorse it, even if Aristotle would not.
general principle that for every number n, n+1 exists. This view would
not enable one to avoid an actual infinity, since we (at any rate, most of
us) already in fact accept arithmetical propositions that entail the exist-
ence of infinitely many numbers. To avoid an actual infinity, one would
have to claim that something more is needed for a number to actually
exist – for example, we must count off that number, or use it in a calcula-
tion, or at least identify it specifically.22
Second, and more plausibly, one might claim that the mode in which
numbers ‘exist’ in general is a mere potential existence. As Benardete
says,
Mathematical existence is ontological possibility. When we say that

there exists an immediate successor for every natural number, what
is the ontological import of that assertion? This: given any class of
n apples (say) ... it is always possible that another class of n+1 apples
might exist.23
Imagine that there are only 1,000 concrete objects in the world. In this
situation, it would not be crazy to claim that the number 1,001 ‘exists’
only in the sense that there could have been another object, so that
there would have been 1,001 objects. We might then want to describe
the existence of the number 1,001 as a kind of ontological potentiality.
But now, what about the number 1,000; would this number also exist as
a mere potentiality? Presumably not. Presumably a number that actu-
ally applies to some collection of actually existing objects itself actually
exists.
But it should be possible to count anything actual, including not only
concrete objects but abstract objects, including numbers themselves. If
so, we only need one concrete object to generate an infinity of actual
numbers. The first concrete object guarantees that the number 1 actu-
ally exists, because 1 applies to that object. Then the concrete object
together with the number 1 constitute two things. So the number 2 also
actually exists. But then the concrete object together with the number 1
and the number 2 constitute three things. And so on.
22
My above use of the expression ‘the successor of 45,670’ might suffice to
bring into existence the number to which that phrase refers. Nevertheless, the
point in the text should be clear enough. It is implausible to claim that we bring
numbers into existence like that.
23
Benardete 1964, 29.
There exists at least one concrete object, in actual fact. Therefore, in

fact infinitely many natural numbers exist.
5.7 Infinitely numerous abstract objects
Besides numbers, it seems that there are also infinitely many proper-
ties that things might have, and infinitely many relations in which
things might stand; that is, there are infinitely many universals. There
are infinitely many physical properties – for instance, infinitely many
shapes, infinitely many amounts of mass that an object might have,
infinitely many velocities, and so on.
As suggested in Chapter 1, there are also infinitely many propos-
itions – for instance, the proposition that 2 is greater than 1, the propos-
ition that 3 is greater than 1, and so on.
Or, if you prefer an example not involving mathematics, consider the
proposition that rabbits are furry. Now here is a distinct proposition:
rabbits are not furry. And a third proposition: rabbits are not not furry.
And a fourth: rabbits are not not not furry.
Why isn’t this an actual infinity? Perhaps because propositions have
a merely potential sort of existence: maybe a proposition is merely a
potential fact (something that could be a fact)?
But whether or not this is a plausible view to take of propositions
in general, surely if one takes this view one will want to say that
there are some actual facts, which do not have a merely potential
existence. If so, how many facts are there? There must be infinitely
many. Here is one way to arrive at that result: suppose rabbits are
furry (as indeed they are). Then there will be (1) the fact that rabbits
are furry. There will also be (2) the fact that there exists the fact that
rabbits are furry. Then there will be (3) the fact that there exists fact
#2. And so on.
Here is another way to get the result. Suppose that infinitely many
numbers exist. Then there will be the fact that 1 exists, the fact that 2
exists, the fact that 3 exists, and so on. Or, suppose that infinitely many
numbers do not exist. Then there must be some n such that there are no
natural numbers greater than n. There will then be the following facts:
the fact that n+1 does not exist, the fact that n+2 does not exist, and so
on. Suppose you think that no numbers exist. Then you should hold it
to be a fact that 0 does not exist, that 1 does not exist, that 2 does not
exist, and so on. So, whether or not there are infinitely many numbers,
there must be infinitely many facts.
5.8 Infinitely numerous physical objects
The visible universe is estimated to contain on the order of

1029 (100,000,000,000,000,000,000,000,000,000) stars.24 This is a lot of
stars. However, it is not infinity. In fact, this number, incomprehensibly
large as it is, is still infinitely far away from infinity.
But that is only the visible universe. There may be any number of
other galaxies beyond what we can see. Indeed, as far as I can tell, there
might even be infinitely many other stars in other galaxies outside the
visible (to us) universe. The Aristotelian prohibition on actual infin-
ities would enable us to rule this out completely a priori, that is, to
predict, in advance of observation, that there must somewhere be a
farthest star from us, beyond which there would be no more stars. But
this just seems misguided; there seems nothing incoherent or other-
wise impossible in the idea that the stars might simply continue on
forever.
In addition, there might be whole other spacetimes, or parallel
universes, as postulated by some theories in contemporary physics.
According to the many-worlds interpretation of quantum mechanics,
each time there is an event that has more than one possible outcome,
the world splits into a number of separate worlds, one for each
possible outcome, and each possible outcome occurs in one of the
worlds. Since there are many events that have infinitely many possible
outcomes, the world would in fact constantly be splitting into infin-
itely many worlds (each of which itself splits again, and so on). More
recently, some string theorists have hypothesized that our spacetime
is one of many spacetimes (called ‘membranes’ or ‘branes’) coexisting
in a higher-dimensional space. A collision between two branes might
explain the Big Bang. There may be infinitely many branes in the
multiverse.25
My point in raising these hypotheses is not that any of them is in fact
correct. My point is that one of them might be correct, and even if none
of them is in fact correct, one of them could have been true. This makes
these theories counterexamples to Aristotle’s theory, which entails that
an actual infinite isn’t even theoretically possible. An infinite collection
of material objects would seem to be the clearest example of an actual
infinite.
24
Howell 2014.
25
For discussion of the varieties of multiverse theorized in contemporary
physics and astrophysics, see Greene 2011.
5.9 Conclusion
It really would be difficult to overstate Aristotle’s influence on thinking

about infinity throughout Western intellectual history, both in math-
ematics and in philosophy. Virtually every thinker on the subject takes
account of his distinction between actual and potential infinities –
whether it be to affirm or to deny the ‘actual infinite’. Unfortunately,
the distinction is less than completely clear. Are future events that will
in fact happen but have not yet happened ‘actual’ or only ‘potential’?
Are numbers actual or potential existents? What about properties, prop-
ositions, facts, past events, unoccupied regions of space?
If we recall the motivation for rejecting actual infinities, then we
should be prepared to take a very broad view of what counts as ‘actual’.
The argument we ascribed to Aristotle was roughly this:
1. Every actual thing must be fully determinate.

2. To be infinite is to lack a determinate value for some quantity.
Therefore,
3. No actual thing can be infinite.
If this is the argument, then the term ‘actual thing’ must be understood
to include everything of which premise 1 would be true – thus, every-
thing that is fully determinate. And this would seem to include space,
time, the set of natural numbers, the set of physical objects, the set of
physical properties, the set of facts, and the set of steps in the Zeno
series. It would seem that each of these things must be determinate, in
the sense that it cannot lack a specific value for a variable that applies to
it, for example, it cannot have mass without having any specific mass,
or duration without any specific duration.
Perhaps you disagree with this. In the Copenhagen Interpretation of
quantum mechanics, it is said that objects can be in indeterminate states,
for example, an electron that is in a box but in neither the left half nor
the right half, or the famous Schrödinger’s cat that is neither alive nor
dead.26 But if you believe in this sort of thing, then you should reject
Aristotle’s theory at the start, since you would reject premise 1 of the
above argument (assuming here that there is not some other convincing
rationale for the conclusion).
I am not addressing people who reject premise (1). Here is what I think
would be odd. It would be odd if someone accepted (1) (that all actual
26
Albert 1994, 10–16.
things must be fully determinate), and yet also thought that none of the
things I listed above – space, time, the set of natural numbers, the set of
physical objects, the set of physical properties, the set of facts, and the
set of steps in the Zeno series – need be determinate. I assume that, at a
bare minimum, the claim ‘all actual things must be fully determinate’
entails that every individual physical object must be fully determinate.
I think it would be very odd if someone held that to be true but yet
thought, for example, that the set of all physical objects could exhibit
indeterminacy. Nor do I see any reason why the set of numbers would be
any more capable of indeterminacy, or the whole of space, or the whole
of time, and so on.
The reader need not agree with all of my examples of actual infini-
ties. Since Aristotle’s theory holds that actual infinities are in principle
impossible, we only need one example of something that (i) is or could
be infinite and (ii) falls within the scope of the Aristotelian principle
that things must be determinate. I happen to think that there are many
such examples, which are pretty easy to think of. But as long as you
think there is at least one, you should reject the Aristotelian account of
the impossible infinite.
6
The Cantorian Orthodoxy
No one shall be able to drive us from the paradise that Cantor

created for us.
– David Hilbert1
6.1 The importance of Georg Cantor
I have already remarked upon Aristotle’s enormous influence on thinking

about infinity. If I had to name one other figure whose influence on the
field has been prodigious, it would be Georg Cantor, the mathemat-
ician who (along with Bernard Bolzano) founded set theory, founded
the theory of transfinite numbers, and was perhaps history’s greatest
champion for the actual infinite. Most of the standard doctrine now
accepted among mathematicians and philosophers concerning infinity
is taken straight from the pages of Cantor.
Cantor was concerned to combat the Aristotelian view that there cannot
be an actual infinity, mainly because Cantor believed that God was infinite.
Believe it or not, he developed a whole new branch of mathematics to
vindicate his theological views by showing that the idea of an actual
infinity is coherent. In this chapter, I shall explain Cantor’s theory of trans-
finite numbers; I shall not, however, explore the theological implications,
if any, of the theory. Readers already familiar with the standard, Cantorian
theory of transfinite numbers may wish to skip over the exposition.
6.2 Sets
The core notion of Cantor’s theory is that of a set. What is a set?

Some treatments of the subject simply state that the term is basic and
1
Hilbert 1967, 376.
71
undefinable, and leave it at that.2 This is quite unsatisfactory. ‘Set’ is a

technical term; readers therefore cannot simply be assumed to already
know its technical meaning in mathematics. If nothing helpful can be
said to explain the concept to new readers, then one ought to suspect
that the concept itself is unclear.
Other treatments say that the notion of a set is defined by the axioms
of set theory – that is, a set is defined as the type of thing of which those
axioms are true.3 This assumes that we can just see that there must be
such a thing.
Most commonly, however, it is said that a ‘set’ is simply a collection or
group.4 Presumably, ‘collection’ is here intended in the ordinary English
sense, for if ‘collection’ too is being used in a technical sense, then we
have made no progress. One text gives ‘a pack of wolves’ and ‘a bunch
of grapes’ as examples.5 The objects that are to be collected together are
referred to as the ‘members’ or ‘elements’ of the set, and they are said to
‘belong to’ the set. This characterization roughly accords with Cantor’s
own definition:
By a ‘set’ we are to understand any collection into a whole M of defi-

nite and separate objects m of our intuition or our thought.6
Some comments: (a) In standard versions of set theory, there is no

requirement that the members of a set be objects of thought or intuition;
a set can have members that no one has ever thought about or intuited.
(b) The use of ‘definite’ either is redundant or also conflicts with standard
set theory: Are there objects that are not definite? If there are, then there
is no reason why they might not also belong to sets. If there aren’t, then
the use of ‘definite’ is redundant. (c) Similarly for the use of ‘separate’.
2
Kunen 1980, xi; Enderton 1977, 11.
3
Fraenkel 1967. We discuss this theory below in Chapter 8, Section 8.2. For a
list of axioms, see below, Section 6.8.3.
4
Milewski 1989, 7; Hrbacek and Jech 1999, 1. Sometimes, textbooks state that
a set must be defined by some rule that all the members satisfy (for example,
Courant and Robbins 1996, 78). This is erroneous since the widely accepted
Axiom of Choice asserts the existence of sets that are not formed according to
any particular rule (what rule, for example, picks out a unique member from any
set of real numbers?). More simply, one can have sets of objects that, intuitively,
have nothing in common.
5
Halmos 1974, 1.
6
Cantor 1915, 85. Jourdain translates the German ‘Menge’ as ‘aggregate’. I
have substituted ‘set’, since this is the currently accepted term for these math-
ematical objects.
The Cantorian Orthodoxy 73
If this is non-redundant, presumably it rules out such things as over-

lapping objects. But in standard set theory, there is no prohibition on
overlapping objects belonging to sets – for example, there is a set whose
members are my left hand and my left thumb. (d) The phrase ‘into a
whole’ appears to be redundant (is there the possibility of a collection
into a non-whole?). Finally, (e) even the term ‘objects’ is either redun-
dant or conflicts with standard set theory. If the term is non-redundant,
then presumably ‘objects’ would contrast with such things as properties,
relationships, or actions. But in standard set theory, sets can perfectly
well contain properties, relationships, and actions, as well as objects.
Taking all these points into account, then, the definition reduces to this:
a set is a collection.
Cantor elsewhere gave another characterization of sets:
A set is a many which allows itself to be thought of as a one.7
By using ‘many’ as a singular count noun (‘a many’), it appears that

we are already assuming that it is legitimate to regard the many as one
thing. So it seems that this definition could be more perspicuously
written: a set is a multitude.
This is probably about as much as we are going to get in the way
of a definition of sets. I will have more to say about the concept later
(Chapter 8).
By convention, sets are designated with capital letters. Suppose A and
B are sets. Then A is the same set as B (that is, A=B) if and only if they
have exactly the same members, that is, every member of A is a member
of B and vice versa. A is said to be a subset of B when every member of A
is a member of B (this has the consequence that every set is a subset of
itself). A is a proper subset of B when A is a subset of B but A is not iden-
tical with B (so B has some members that are not in A).
6.3 Cardinal numbers
The cardinal numbers are the numbers used for counting, or saying how
many of something there are. Cardinal number terms, such as ‘two’ or
‘eight’, have two interestingly different uses: sometimes, they are used
as adjectives, as in ‘Two octopi entered the room’; other times, they are
used as nouns, as in ‘Two is prime.’ The latter use seems to be intended
to refer to some special, mathematical object. What is this object?
7
Cantor 1932, 204.
Here is Cantor’s description:
We will call by the name ‘power’ or ‘cardinal number’ of M the

general concept which, by means of our active faculty of thought,
arises from the set M when we make abstraction of the nature of its
various elements m and of the order in which they are given.8
This is not extremely clear. Later thinkers were to propose that a cardinal
number is a kind of set: the numeral ‘1’ (in its noun usage) refers to the set
of all single-membered sets (the set whose members are all and only the sets
that have exactly one member). ‘2’ refers to the set of all two-membered
sets. And so on. This explains the natural numbers, which are all of the
finite cardinal numbers. This is the view of numbers that we shall discuss.9
What about infinity – are there any infinite numbers? Yes. Consider
the set of all the natural numbers. It has infinitely many elements. Now
consider the set of all sets with exactly that many elements (that is,
the same number as the set of natural numbers): this, too, is a cardinal
number. It is represented Յ0 (this is read ‘aleph null’). Յ0 is the first
infinite cardinal number.
Are there any more infinite cardinal numbers? We shall see.
6.4 ‘Greater’, ‘less’, and ‘equal’
Suppose two sets, A and B, are the same size, that is, contain equally
many elements. Then it will be possible to pair off the sets’ members
with each other, in such a way that every member of A is paired with
exactly one member of B, and vice versa. For instance, say you have a
set of three horses, {Seabiscuit, Mr Ed, Shadowfax}. You also have a set
of three carrots, {carrot1, carrot2, carrot3} (carrots have less interesting
names than horses do). Then, because the sets are equally numerous, it
is possible to pair the carrots with horses in such a way that each horse
gets one and only one carrot, and each carrot goes to one and only one
horse. For instance, we could pair them up like this:
Seabiscuit ➟ carrot2
Mr Ed ➟ carrot1
Shadowfax ➟ carrot3
8
Cantor 1915, 86 (again reading ‘set’ for ‘menge’).
9
Frege 1980, 79–80, 87–90; Russell 1920, 18–19. Frege portrays a number as a
set of concepts, where a concept is a Platonic object; for simplicity, I use Russell’s
notion of a set of sets. A more recent proposal identifies the number 0 with the
empty set, 1 with {0}, 2 with {0,1}, and so on. Since the Frege/Russell proposal is
much more natural, I prefer to discuss it in the text.
(I have assigned the carrots arbitrarily; there are four other ways of doing
the pairing, but all that matters is that there be at least one way.) This
kind of pairing of elements is known as a one-to-one function (or mapping)
from the first set onto the second set. In this example, we have a one-
to-one function from the set of horses onto the set of carrots.
Cantor uses this notion to define numerical equality: two sets are
equally numerous (have the same number of elements, have exactly as
many elements as each other) if and only if there is a one-to-one func-
tion from one set onto the other.
On the other hand, suppose that there is no one-to-one function from
A onto B, but there is a one-to-one function from A onto a proper subset
of B. In other words, when we try to pair up the elements of the two sets,
some elements of B are always left over. In that case, we can say that A
is smaller than B, or contains fewer elements than B, or is less numerous
than B. For instance, if, when we paired up horses with carrots, assigning
one carrot to each horse, we found that no matter how we did the pair-
ings, we would always have some carrots left over, then we would say
that the set of horses was ‘less numerous’ than the set of carrots.
Similarly, if there is no one-to-one function from A onto B but there is
a one-to-one function from a proper subset of A onto B, then we can say
that A is larger than (contains more elements than, is more numerous
than) B. Thus, if when we try pairing horses with carrots, we find that no
matter how we do the pairings, all the carrots are used up and there are
still horses left without carrots, then we would say that the set of horses
was ‘more numerous’ than the set of carrots.
Importantly, none of these are supposed to be strange, new, technical
uses of terms. The foregoing is intended to explain the existing meanings
of the familiar expressions ‘equal’, ‘less’, and ‘greater’, as they are used
in arithmetic and ordinary life.
Furthermore, Cantor proposes that these same definitions of terms can
be applied equally to infinite collections as to finite collections. Thus, just
as in the case of finite sets, two infinite sets will be said to have ‘equally
many’ elements if and only if there is a one-to-one function from the one
set onto the other. One will be ‘larger’ than the other if there is no one-
to-one function from the one onto the other, but there is a one-to-one
function from a proper subset of the first onto the second.
6.5 Many sets are equally numerous
Many pairs of sets, probably many more than you would have suspected,
turn out to be equinumerous on Cantor’s criterion. Two obvious
examples: the set of natural numbers is the same size as the set of even
numbers, and also the same size as the set of perfect squares. The diagram
below illustrates how these sets’ members can be paired up:
2 4 6 8 ... 2n ...
↕ ↕ ↕ ↕ ↕
1 2 3 4 ... n ...
↕ ↕ ↕ ↕ ↕
1 4 9 16 ... n2 ...
More surprisingly, it turns out that the set of rational numbers is the
same size as the set of natural numbers (this despite the fact that the
rational numbers include all the natural numbers, and that between
any two natural numbers is an infinity of rational numbers). A rational
number is defined as a number that can be written as the ratio of two
integers. That is, if r is rational, then there is a pair of natural numbers,
(n, m), such that r = n/m. This is equivalent to saying that the decimal
expansion of r either terminates or repeats. It is therefore possible to
make a table listing all the rational numbers, as follows:
Possible values of n
1 2 3 4 ...
1 1 2/1 3/1 4/1 ...

Possible
2 1/2 2/2 3/2 4/2 ...
values of m
3 1/3 2/3 3/3 4/3 ...
4 1/4 2/4 3/4 4/4 ...
⁝ ⁝ ⁝ ⁝ ⁝
Some of the entries in that table are redundant. For example, ‘1/1’ and
‘2/2’ both appear, but those expressions denote the same number. So
imagine a similar table, but with the redundant entries deleted, keeping
only fractions that are reduced to lowest terms:
1 2 3 4 ...
1 1 2/1 3/1 4/1 ...

Possible
2 1/2 – 3/2 – ...
values of m
3 1/3 2/3 – 4/3 ...
4 1/4 – 3/4 – ...
⁝ ⁝ ⁝ ⁝ ⁝
Now imagine going through all the entries in the table in a certain
order: start at 1/1, then go to 2/1, then 1/2, then 1/3, then 3/1, and so
on. The order is shown by the dotted line in the table below:
1 2 3 4 5 ...
1 1 2/1 3/1 4/1 5/1 ...

Possible 2 1/2 – 3/2 – 5/2 ...
values of m 3 1/3 2/3 – 4/3 5/3 ...
4 1/4 – 3/4 – 5/4 ...
5 1/5 2/5 3/5 4/5 – ...
⁝ ⁝ ⁝ ⁝ ⁝ ⁝
It should be clear that the dotted line can be continued indefinitely
and that it passes through all the entries in the table, going through
them in a particular order. This indicates how the pairing of rational
numbers with natural numbers works: the first rational number (1/1)
gets paired with ‘1’; the second rational number that we come to (2/1)
gets paired with ‘2’; the third (1/2) gets paired with ‘3’; and so on:
1 2 3 4 5 6 7 8 9 10 ...
↕ ↕ ↕ ↕ ↕ ↕ ↕ ↕ ↕ ↕
1/1 2/1 1/2 1/3 3/1 4/1 3/2 2/3 1/4 1/5 ...
This constitutes a one-to-one function from the natural numbers onto

the rational numbers. So there are exactly as many natural numbers
as rational numbers. This number, Յ0, is referred to as a ‘count-
able infinity’, since it is the infinity of the counting numbers. Thus,
the natural numbers, the perfect squares, the even numbers, and the
rational numbers are all said to be countably infinite. If there are any
larger infinities, they will be called ‘uncountable infinities’.
It turns out, by the way, that the set of real numbers contains exactly
as many members as the set of pairs of real numbers; that the points on
a line are exactly as numerous as the points in a two-dimensional plane;
and indeed, perhaps most amazingly, that the set of points in, say, a
one-inch line segment is exactly as numerous as the set of points in all
of space. (I omit the proofs of these theorems.10)
10
The key point is that one can encode the decimal expansion of two or more
numbers into another decimal expansion. For example, a pair of real numbers,
(.459131 ... , .208817 ... ) can be mapped to a single number constructed from
alternating digits of the first two numbers: .425098183117 ... .
6.6 The diagonalization argument
At this point, you might be wondering if all infinite collections are

equally numerous. The answer is no. The proof of this negative result is
famous enough that everyone who reads a book about infinity should
know it, so I explain it here.
Let us start by comparing the set of all natural numbers with the set of
real numbers between 0 and 1. Each of these real numbers can be written
as an infinitely long decimal expansion, for example, ‘0.19479843 ... ’.
(In some cases, the infinite decimal expansion contains an infinite series
of zeroes, as in the case of ½, which is 0.50000 ... .) Which set, if any, is
the more numerous?
Let’s start with the hypothesis that they are equally numerous, and see
if that works out. That means that there is a one-to-one function from
the natural numbers onto the real numbers between 0 and 1. Another
way of saying this is that it is possible to make a list of all the real
numbers between 0 and 1, where you have a first real number, a second
one, a third one, and so on, and where every real appears somewhere on
the list (that is, as the nth entry, for some natural number n). Suppose we
have such a list. It might look something like this:
n f(n)
1 .545092 ...
2 .436214 ...
3 .197967 ...
4 .849465 ...
5 .465546 ...
6 .654654 ...
⁝ ⁝
We’re supposing that f is a function that maps the natural numbers one-
to-one onto the real numbers between 0 and 1. I have made up the
entries in the ‘f(n)’ column at random (so don’t bother looking for a
pattern). For the following argument, it doesn’t matter what you put
there; all that matters is that we have some list of real numbers between
0 and 1 in the right-hand column.
Now I am going to tell you about a real number that does not appear
in that list. I call this number ‘R’. R has an infinite decimal expansion.
The way you find the digits of R is by going down the list of real numbers
above, and adding one to the first digit of the first number, then adding
one to the second digit of the second number, then adding one to the
third digit of the third number, and so on. (If a digit is ‘9’, then consider
the result of adding one to it to be ‘0’.) You write down the results in
sequence, and that gives you the decimal expansion of R. This is illus-
trated in the following table:
n f(n) Digits of R
1 . 5 4 5 0 9 2 ... .6
2 . 4 3 6 2 1 4 ... .64
3 . 1 9 7 9 6 7 ... .648
4 . 8 4 9 4 6 5 ... .6485
5 . 4 6 5 5 9 6 ... .64850
6 . 6 5 4 6 5 0 ... .648501
⁝ ⁝ ⁝
This is referred to as the ‘diagonalization argument’, because you
construct R by moving diagonally down the digits in the list of values of
f(n), starting at the upper left. In the table above, I have underlined the
digits from the list that are used to construct R. Now, R cannot be one of
the values of f(n). It is not f(1), because it differs from f(1) in (at least) its
first decimal place. It is not f(2), because it differs from f(2) in the second
decimal place. And so on. In general, for any n, R is not equal to f(n),
because R differs from f(n) in the nth decimal place. But R is obviously
a real number between 0 and 1. This proves that the function f, which
was supposed to be a one-to-one function from the natural numbers onto
the real numbers between 0 and 1, is not in fact such a function, because
there is at least one real number between 0 and 1 that is not paired with
any natural number.
The same sort of argument can be applied no matter what the func-
tion is. That is, Cantor’s diagonalization procedure can be applied to
any list of decimal numbers that might have been placed opposite the
numbers 1, 2, 3, and so on in the table. No matter what the list is, you
can prove that there is a real number that isn’t on the list. Therefore,
it is in fact impossible to have a one-to-one function from the natural
numbers onto the real numbers between 0 and 1; if you attempt such a
pairing, then there will always be some real numbers left over.
It is, however, perfectly possible to have a one-to-one function from
the natural numbers onto a proper subset of the real numbers between 0
and 1. For example, the function f(n) = 1/n will do.
Thus, the set of real numbers between 0 and 1 satisfies the criterion
given in Section 6.4 for being ‘more numerous than’ the set of natural
numbers. Of course, the set of all real numbers (which, by the way, has
the same cardinality as the set of real numbers between 0 and 1) is also
more numerous than the set of natural numbers. The cardinal number of
the real numbers is referred to as c, or ‘the continuum’. It is an uncount-
able infinity – but not, as we shall see, the only uncountable infinity.
6.7 Cantor’s theorem
So now we have two infinite numbers: Յ0 and c. How many more infinite
numbers are there? As it turns out, infinitely many. This is guaranteed
by another result known as Cantor’s Theorem, which is another thing
that anyone who reads a book about infinity should know.
First, some more set theory background. In standard set theory, there
is a set called ‘the empty set’, which contains no members. It is desig-
nated ‘{ }’ or ‘∅’. (Frege defined it, roughly, as the set of all things that are
not identical with themselves.11) There are also ‘singleton sets’, each of
which has only one member. Sets are distinct from their members, even
if the set has only one member. For example, King Kong, {King Kong},
and {{King Kong}} are supposed to be three different objects.
Given a set, A, the powerset of A (denoted ‘Ե(A)’) is defined as the set
of all subsets of A. For example, take the empty set. What is its powerset?
Well, the empty set has exactly one subset (namely, itself – note that
every set is a subset of itself). So the powerset of the empty set is {∅}. The
powerset of that, in turn, is {∅,{∅}}.
Take the set containing King Kong and Godzilla, which is represented
{King Kong, Godzilla}. Its powerset is {∅, {King Kong}, {Godzilla}, {King
Kong, Godzilla}}.
Notice that, in all these examples, the powerset of A has more members
than A: the empty set has zero members, but its powerset has one
member, and the powerset of that powerset has two members. The set
containing King Kong and Godzilla has two members, but its powerset
has four members. It can be shown in general that if A has n members,
then the powerset of A has 2n members.
When A is a finite set, it is easy to see that the powerset of A will have
more members than A. What is not so obvious is what happens when A
is an infinite set. Will the powerset of A still have more members than A?
The answer turns out to be yes, and this is Cantor’s theorem. Following
is the proof.
Let A be any set (it could be finite or infinite; doesn’t matter). Let Ե(A)
be the powerset of A. Now suppose f is a function from A to Ե(A). So
(just restating that), f maps each member of A to a specific subset of A.
11
Frege 1980, 87.
Now I am going to tell you about a subset of A that is not included in

that mapping – that is, there is no member of A that is mapped to this
subset. (This is going to be analogous to the diagonalization argument
from the last section.) I’m going to call this special subset S. Here is how
you construct S. Go through the members of A. Each of these members
is mapped to some subset of A. So let’s say a is some member of A. a is
mapped to f(a), which is some subset of A. Now, either f(a) contains a
as one of its members, or it doesn’t. If f(a) contains a, then a does not
go into S; if f(a) does not contain a, then a goes into S. In other words,
S is going to be formed from all the elements of A that are mapped to
subsets that do not contain those elements. Figure 6.1 provides a graph-
ical representation of what’s going on.
Each ellipse represents a set. a, b, c, and so on are the elements of set
A. I use three dots to indicate that there may be many more elements
not listed. The arrows indicate the mapping, given by f, from A to the
powerset of A. Finally, S is represented on the right. Notice that in the
diagram, b was mapped to a set that does not contain b (specifically, b
was mapped to the empty set). That is why b is included in set S on the
right. e was also mapped to a set that does not contain e (specifically, to
the set {d}), so e also goes into S. Similarly for c, since it was mapped to
{a, b}, which does not contain c. However, a was mapped to a set that
contains a (specifically, to {a, e}), so a does not go into S. Similarly for
item d, mapped to {e, d}.
Once you see how set S is constructed, you can see that S is not in the
range of f. That is, there is no element from A that gets mapped to S.
Using the example from Figure 6.1: we know element a is not mapped
to S, because a is mapped to a set that contains a, whereas S does not
contain a. We know b is not mapped to S, because b is mapped to a set
Set A Powerset of A Set S
b
b
a c
a e e
d e
d
e c
c d
b
a
Figure 6.1 Cantor’s theorem

that does not contain b, whereas S does contain b. And so on. For each
member m of A, m is not mapped to S, because either (i) m is mapped
to a subset that contains m, in which case S does not contain m, or
(ii) m is mapped to a subset that does not contain m, in which case S does
contain m. The way S is constructed guarantees that it is not one of the
subsets used in the mapping.
But S is definitely a subset of A. So f does not succeed in mapping the
elements of A one-to-one onto the subsets of A. Since this reasoning can
be carried out no matter what f is, it shows that it is impossible to pair up
the elements of A with all of the subsets of A; no matter how you do the
pairing, there are always subsets of A that are left over. This shows that
there are more subsets of A than there are elements of A. This reasoning,
again, applies regardless of whether A is finite or infinite.
So now there are not just two infinities. For any set, we can construct
a set with a higher cardinality, that is, with ‘more’ elements. Thus, we
have an infinite hierarchy of ever larger infinities, starting with the
set of natural numbers, followed by the powerset of the set of natural
numbers, followed by the powerset of that, and so on.
6.8 The paradoxes of set theory
6.8.1 Russell’s paradox

Mathematicians tend to love Cantor’s mind-boggling hierarchy of infin-
ities (in fact, the hierarchy of infinities that mathematicians posit today
is even larger and more mind-boggling than I have so far indicated) –
hence Hilbert’s famous remark about ‘the paradise that Cantor has
created’. However, set theory has also been beset with paradoxes from
its beginning. Let’s look at a couple of these paradoxes, starting with the
most famous set theoretic paradox, Russell’s Paradox.12
Most sets are not members of themselves. For example, the set of all
left-handed marmots is not a member of itself, since it is not itself a
left-handed marmot. On the other hand, it seems that at least some sets
are members of themselves. For instance, the set of all sets must be a
member of itself, since it is a set.
Now, let’s think about all the sets that are not members of themselves.
We ought to be able to collect them together, right? The result would
be the set of all the sets that are not members of themselves. Call this
set R (for ‘Russell set’). So R is the set that contains x whenever x is not
a member of x.
12
Russell 1920, 136.
Question: is R a member of R? Well, as we said, for any x, R contains

x if and only if x does not contain x. Now you can substitute ‘R’ for ‘x’:
that means that R contains R if and only if R does not contain R. But that
is a contradiction.
Here’s the reasoning more slowly: suppose that R contains R. Then,
since the only sets it contains are sets that don’t contain themselves,
that means R must be a set that does not contain itself. So, if R contains
itself, then R does not contain itself.
Now suppose R does not contain itself. Then, since R includes
all the sets that don’t contain themselves, R must include R. So, if R
does not contain itself, then R contains itself. Either way, we have a
contradiction.
6.8.2 Cantor’s paradox

Here is another paradox. Consider the set of all sets, commonly known
as the Universal Set, or V. Is V larger than, smaller than, or the same size
as its powerset, Ե(V)?
Well, on the one hand, we already proved that every set is smaller than
its own powerset, so V must contain fewer elements than Ե(V).
On the other hand, Ե(V) is a set of sets (every powerset is a set of sets).
And V, by definition, contains every set whatsoever. So V must contain
every element of Ե(V). So V must contain at least as many elements as
Ե(V) does.
Again, we have a contradiction. V can’t be smaller than and at the
same time at least as large as Ե(V).
6.8.3 The standard treatment

How do set theorists and logicians deal with these paradoxes? The
currently accepted approach is simply to banish the offenders from
Cantor’s paradise. That is, we deny that R exists, we deny that V exists,
and the paradoxes go away.
Well, it’s a little more complicated than that – set theorists don’t
merely declare that those two sets don’t exist and leave it at that. Rather,
they try to devise systems of axioms for set theory, where these axioms
entail the existence of the ‘good’ sets (the ones that don’t generate
paradoxes, such as the set of all natural numbers, or the set of all real
numbers), but where the axioms do not entail the existence of any of
the ‘bad’ sets (the sets that generate paradoxes, such as the Russell Set
or the Universal Set). Early versions of set theory unfortunately implied
the existence of such paradoxical sets (essentially because they implied
that for any description you care to formulate, there must exist the set of
the things that satisfy that description). After the paradoxes were discov-
ered, people devised several different systems of axioms that avoid the
paradoxes by rejecting the existence of such things as the Russell Set
and the Universal Set. The currently most popular system of axioms is
known as Zermelo-Fraenkel set theory, or ZF. One axiom, the Axiom of
Choice, is controversial; if it is included, then the result is called ZFC.13 I
list the axioms here, but don’t worry too much about them:
Axioms of ZFC
1. Axiom of Extensionality: If X and Y have the same elements, then

X=Y.
2. Unordered Pair Axiom: For any a and b there exists a set {a, b} that
contains just a and b.
3. Subset Axiom: If F is a property then for any set X, there exists a set
containing exactly those members of X that have F.
4. Union Axiom: For any X there exists the set ‫׫‬X, the union of all
the elements of X.
5. Powerset Axiom: For any X there exists the set Ե(X), the set of all
subsets of X.
6. Axiom of Infinity: There exists an infinite set, namely, a set that
contains the empty set, and for every member that it contains,
it contains the union of that member with that member’s own
singleton.
7. Axiom of Replacement: If F is a function, then for any X there exists
the set, {F(x): x ‫א‬X}, which contains the values that the function
returns for each element of X.
8. Axiom of Foundation: Every nonempty set has a minimal element,
that is, one which does not contain another in the set.
9. Axiom of Choice: Every family of nonempty sets has a choice func-
tion, namely, a function which assigns to each of the sets one of
its own elements.
13
The Axiom of Choice is controversial, among other reasons, because it
enables one to prove the Banach-Tarski Theorem (or Banach-Tarski Paradox,
depending on how you feel about it). This theorem states that it is possible to
dissect a sphere of radius 1 into a finite number of pieces, rearrange the pieces
using only rigid rotations and translations (with pieces allowed to move through
each other), and reassemble them into two spheres of radius 1. No one can actu-
ally describe these pieces (which are infinitely complex), but the Axiom of Choice
allows one to prove that such must exist. For an attempt at an intuitive account
of the theorem, see The Writer 2003.
The important points for our purposes are that ZF implies the existence
of an infinite hierarchy of sets, including sets with (each of infinitely
many) infinite cardinalities, starting from Յ0; and that as far as anyone
knows so far, ZF does not entail the existence of any paradoxical sets,
such as the Russell Set or the Universal Set.
6.9 Other paradoxes of infinity
Maybe we can avoid the paradoxes of set theory, such as Russell’s

Paradox and Cantor’s Paradox. But what about all the paradoxes of the
infinite that we surveyed in Chapter 3? Now that we have laid down a
theory of infinite numbers, can we use the theory to solve the previous
paradoxes?
6.9.1 Identifying infinite cardinalities in the paradoxes

Most of the paradoxes we discussed involve an infinite series of actions
or events. In each case, Cantor’s theory can tell us how many members
the series has: it is Յ0, as opposed to, say, c, or Յ2. But that interesting
bit of information hardly enables us to resolve the paradoxes. For
example, we can now say that the Thomson Lamp scenario posits Յ0
switchings of the lamp. But that hardly tells us whether at the end the
lamp will be on or off. Similarly, Hilbert’s Hotel contains Յ0 rooms and
Յ0 guests; Zeno’s Paradox divides the ball’s motion into Յ0 stages; the
Divided Stick is divided Յ0 times; the game played in the Littlewood-
Ross Paradox contains Յ0 rounds; Benardete’s Paradox posits Յ0 walls (or
gods, or slabs); Laraudogoitia posited Յ0 marbles; the Spaceship doubles
its speed Յ0 times; the St. Petersburg game has Յ0 possible outcomes; the
Martingale Betting System requires Յ0 rounds of gambling; and finally,
the Delayed Heaven Paradox takes place over Յ0 days. And that informa-
tion helps us resolve ... none of them.
In one case, a cardinality other than Յ0 makes an appearance: in the
Paradox of Geometric Points, we have continuum many geometric
points, which is more than Յ0. But this information hardly seems to
help solve the puzzle – knowing that the order of infinity is c rather than
Յ0 hardly renders it perfectly clear how infinitely many zeroes add up to
some nonzero quantity.
6.9.2 Infinite magnitudes

Some of the paradoxes involve a different sort of infinity altogether, an
infinite magnitude rather than an infinite cardinal number (sometimes
there is an infinite magnitude in addition to an infinite number). In the
puzzle of Smullyan’s Infinite Rod, it isn’t that there are infinitely many
of something; the rod is just infinitely long. Gabriel’s Horn doesn’t
involve infinitely many of anything; it just has an infinite surface area.
The Spaceship doubles its speed Յ0 times, which results in its traveling
an infinite distance, so in this case, an infinite cardinality appears
(the number of doublings) in addition to an infinite magnitude (the
distance). In the St. Petersburg Paradox, besides the infinite cardinality
(the number of possible outcomes), there is an infinite magnitude (the
expected monetary value of the game). Cantor’s theory says nothing
about infinite magnitudes; it is solely a theory of cardinal numbers.
6.9.3 Actual infinities and Zeno’s paradoxes

Cantor intended his mathematical work to have philosophical signifi-
cance. Specifically, he thought that his mathematical theory of infinite
cardinalities provided support for the idea that, contra Aristotle, actual
infinities are possible. If we buy this, it would help us resolve one of
the paradoxes: we could perhaps resolve Zeno’s Paradox by declaring
that, since actual infinities are perfectly possible, there is no problem
with the ball’s completing an infinite series of actions en route to its
destination.
However, this view seems to exacerbate almost all of the other para-
doxes (Thomson’s Lamp, Hilbert’s Hotel, Littlewood-Ross, and so on),
since almost all of the paradoxes depend upon postulating actual
infinities.
6.9.4 The paradoxes of one-to-one mapping

There are two paradoxes that Cantor’s theory seems to bear on rather
directly. I have in mind Galileo’s Paradox and Hilbert’s Hotel. In response
to Galileo’s question, Cantor declares that there are exactly as many
natural numbers as perfect squares, because it is possible to place the
natural numbers and the squares in a one-to-one correspondence.
Does this solve Galileo’s puzzle? Hardly. Galileo already thought of
that exact point. The fact that the natural numbers and the squares can
be put in one-to-one correspondence isn’t the solution to the paradox;
it is part of what generates the paradox. (A good rule of thumb: typi-
cally, one does not solve a paradox by reiterating part of its setup.) The
puzzle is that there is a compelling argument both that the two sets are
equally numerous (the one-to-one correspondence argument), and that
one set is larger than the other (for the squares are a proper subset of
the natural numbers). Cantor embraces the first of these arguments and
rejects the second. His only justification for this is fiat: he proposes to
simply define the relations ‘equal to’, ‘greater than’, and ‘less than’ using
the one-to-one correspondence criterion, while ignoring the proper-
subset criterion.
Of course, one could not consistently embrace both of the arguments
Galileo mentions, since they entail contradictory results. But Cantor’s
decision to embrace only the one-to-one correspondence argument is
not the only alternative. One could embrace the proper-subset argu-
ment while rejecting the one-to-one correspondence argument.14 Or
one could, like Galileo, reject both arguments and hold that the ‘greater
than’, ‘less than’, and ‘equal to’ relations do not apply to pairs of infinite
sets. Cantor does not argue for his alternative over the others.
The same odd feature of infinite sets – namely, that an infinite set
can be mapped one-to-one onto a proper subset of itself – is at work in
Hilbert’s Hotel. Here again, we know what Cantor would say. He would
agree that one can make room for the new guests by moving the old
guests around, and he would counsel us to get used to this interesting
feature of the infinite. Noting that Յ0 + Յ0 = Յ0,15 he would say that,
when the infinitely many new guests are added to the infinitely many
old guests, the resulting set of guests is no bigger than it was before the
new guests arrived; hence, there is nothing wrong with the idea that
we can still accommodate everyone in the same set of rooms. Again,
this is not so much an attempt to solve the paradox as it is a decision to
embrace paradox.
6.9.5 The arithmetic of infinity

Some of the paradoxes involve performing arithmetical operations on
infinity, for example, subtracting ∞ from both sides of the equation
‘∞ + 1 = ∞’. With our new knowledge of transfinite numbers, we might
try subtracting Յ0 from both sides of ‘Յ0 +1 = Յ0’, obtaining ‘0 = 1’. The
standard response to these problems is similar to the standard response
to the set theoretic paradoxes: prohibit the troublesome operations. In
the standard view of transfinite arithmetic, even though Յ0 is a number,
one cannot subtract it from itself, that is, ‘Յ0 − Յ0’ is undefined. Similarly,
division by Յ0 is undefined. Followers of Cantor counsel us to simply
get used to the fact that infinite numbers behave differently from finite
numbers in certain respects, and this is one of them.16
14
This would give us only a partial ordering, but what’s wrong with that?
15
This equation actually appears in Cantor 1915, 106.
16
Oppy 1995.
Note that in standard arithmetic, something similar is true of zero:

one can divide any real number by any real number, except zero. So zero
behaves differently from other numbers; however, this is not generally
taken to show that zero isn’t a number. Similarly, Cantor would argue,
we should not deny that Յ0 is a number merely because it behaves differ-
ently from the more familiar numbers.
6.9.6 The divided stick paradox

With Cantor’s help, we can now formulate a new version of the Divided
Stick Paradox from Section 3.10. We assume the stick is in a world where
the matter comprising a physical object is continuously distributed
throughout the space that the object occupies (rather than the object’s
being decomposable into a finite number of elementary particles, as in
our actual world). The stick gets divided into halves, then into quarters,
then eighths, and so on. After the infinitely many stages of division, the
stick has been converted into an infinity of slices, each of which has a
thickness of zero.
How many of these slices will there be? Infinitely many, but which
order of infinity is it? It appears to be Յ0, for consider where cuts
have been made by the end of the process. If we designate one end
of the stick ‘0’ and the other end ‘1’, then the stick gets cut at the
following locations: 1/2, 1/4, 3/4, 1/8, 3/8, 5/8, 7/8, ... (including all
locations of the form n/2m, where n and m are natural numbers and
n < 2m). All of these are rational numbers, and we already know that
the rational numbers have cardinality Յ0. So there are only Յ0 cuts
made in the stick, which seems to imply that the stick is divided into
only Յ0 pieces.17 And as we said, each of these pieces must have a
width (and a mass) of zero.
But now think back to before the stick was divided. The stick filled
a certain space. Space contains continuum many points. Furthermore,
if we consider a line stretching from position 0 (the left end of the
stick) to position 1 (the right end of the stick), that line would have
continuum many points. The stick therefore contained continuum
many cross-sectional slices, each of thickness zero, one for each point
along that line.
17
Oppy (2006, 66–8) seems to agree with this. However, here is a contrary
argument: after n stages in the series of divisions, there are 2n pieces (2 pieces
after the first stage, 4 after the second, and so on). Therefore, after Յ0 stages,
there will be 2Յ0 pieces, and 2Յ0 = c. For the resolution of this puzzle, see below,
Section 12.9.
In other words: we started with a composite of continuum many

pieces, and by cutting it up, we turned it into only Յ0 pieces, where each
of the Յ0 pieces is qualitatively indistinguishable (the same size, mass,
and so on) as each of the continuum many pieces was. So it appears that
continuum many pieces disappeared. Where did they go?
6.10 Conclusion
Cantor has not helped us to resolve the paradoxes of infinity. On most of

the paradoxes, Cantor’s system of transfinite numbers seems to simply
have no bearing, as in the case of Thomson’s Lamp or Littlewood-Ross.
One might think Cantorian acceptance of actual infinities helps resolve
Zeno’s Paradox, but only at the cost of exacerbating all the paradoxes
that rest on the possibility of actual infinities. In some cases (Galileo’s
Paradox, Hilbert’s Hotel), the Cantorian approach is simply to embrace
the paradoxical results. In one case (the Divided Stick), the Cantorian
framework enables us to reformulate the paradox in a way that adds to
our puzzlement.
In other cases, the accepted treatment within modern mathematics is
to rule out paradoxes by fiat: for the set theoretic paradoxes, we deny
the existence of the sets that lead to paradox, tailoring our mathematical
systems ad hoc to rule them out. For the paradoxes of infinite arith-
metic, we disallow the troublesome operations (subtracting or dividing
by infinity).
This, it seems to me, hardly constitutes a satisfying solution to the
paradoxes. We already know at the start, before looking at the details
of any paradoxes, that there cannot be a true contradiction. Therefore,
if we have seemingly compelling premises leading to contradictory
conclusions, we know that at least one of them is wrong. It is not illu-
minating to be told that much, nor does that solve any intellectual
problem. Nor does our commentary become illuminating or intellec-
tually satisfying if we simply pick a premise to reject, with no more
account of why that premise is wrong than to say that it (perhaps
together with other premises that seem more obviously true) leads to
a contradiction – namely, the very contradiction that led us to iden-
tify a paradox in the first place. To solve a paradox and not merely
avoid it, one must say something to explain how the premise one
identifies as false could be false, or at least something that makes the
premise cease to seem like something that ought to be true.
It is for this reason that I say that the set theoretic paradoxes have yet
to be solved.18 It certainly seems as though, if we have a well-formed
description, there ought to be a set of the things satisfying that descrip-

tion. This principle is known as the Comprehension Axiom of naive set
theory. The Comprehension Axiom seems obvious to most people who
are introduced to set theory, and it still seems as though it ought to be
true, even after one has heard about the set theoretic paradoxes. The
standard approach is simply to say: ‘The Comprehension Axiom is false,
since it leads to those set theoretic paradoxes.’ But that does nothing to
explain how it can be false, or to remove the intuition that it ought to
be true. If our standards for solving paradoxes were so generous, then
any paradox would be solved as soon as we identified the premises on
which it rested.
Similar remarks apply to the problems of infinite arithmetic. There
must be something wrong, either with adding infinity or with subtracting
infinity, since addition and subtraction of infinity enables us to derive
that 1 = 0. But again, our standards for solving a paradox should not be
so low that the preceding observation counts as a solution.
Ideally, we would like something like this: an illuminating account
of the nature of a set that reveals why, for example, there could not be
such a thing as the set of all sets (say, because the idea is confused or a
misuse of the concept); followed by an account of the nature of subtrac-
tion, division, and/or infinity that reveals why the idea of subtracting or
dividing by infinity similarly fails to make sense.
18
This criticism applies more fairly to the Zermelo-Fraenkel approach than to,
say, Russell’s theory of types. Russell’s approach attempts a natural explanation
for why there could be no set of all sets that do not contain themselves; however,
it requires that ‘is a member of’ is infinitely ambiguous (having infinitely many
senses). The von Neumann-Bernays-Gödel approach turns on the similarly weird
thesis that there is a special type of class (‘proper classes’) that are incapable of
being members of other classes.
Part III
A New Theory of Infinity and
Related Matters
7
Philosophical Preliminaries
7.1 Metapreliminaries
7.1.1 Why a chapter of preliminaries?

Before explaining my views concerning infinity and related matters,
I want to first set out some philosophical preliminaries. This chapter
explains some general features of my philosophical outlook, including
some broad epistemological and metaphysical points.
Why is such a chapter needed? Briefly, the reason is that there are
certain philosophical assumptions that tend to generate strong resist-
ance to my views, and these assumptions are commonly accepted by
those interested in issues connected with science and mathematics, and
hence by many likely readers for a book on infinity. I have in mind
especially the assumptions of modern (twentieth-century) empiricism.
Readers sympathetic to empiricism will tend to reject the rest of this
book. Indeed, empiricists tend to take their philosophical assumptions
so much for granted that they are likely to think that I have committed
elementary oversights or engaged in obvious nonsense, rather than that
I have systematic reasons for rejecting their underlying philosophical
assumptions. Many individuals have empiricist assumptions but are
unaware that they constitute a substantive and controversial philosoph-
ical theory.
7.1.2 Modern empiricism

Empiricism is, roughly, the doctrine that it is impossible to attain any
substantive knowledge of the world except on the basis of observation.
Thus, if a statement cannot be observationally tested, it is unknowable.
In contrast, rationalism holds that there is some substantive knowledge
about the world that is not based on observation; this knowledge is
usually said to derive from the exercise of pure reason.
93
In the twentieth century, empiricists developed increasingly extreme

anti-rationalist positions. Thus, an extreme form of empiricism, logical
positivism, went on to claim that if a statement cannot be observation-
ally tested, it is not merely unknowable; it is literally meaningless, that
is, it fails to assert anything; it is a mere pseudo-statement.1 These ideas
were used to declare wide swathes of human intellectual endeavor
meaningless, including such fields as metaphysics, theology, and ethics.
Another implication of the view was that if any scientific theory entails
the existence of observationally undetectable facts, that theory must be
either false or meaningless.
Ironically, empiricists have generally treated their philosophy more
like an a priori axiom than like an empirical hypothesis. That is, empiri-
cism is usually assumed as self-evident, whereupon the philosopher
proceeds to unhesitatingly embrace whatever consequences follow.
Perhaps the central motivation for empiricism in the last century
stemmed from the extremely high regard in which natural science
has been held. Peter van Inwagen describes an ideology of ‘scientism’,
consisting in ‘an exaggerated respect for science [ ... ] and a corresponding
disparagement of all other areas of human intellectual endeavour’.2 One
consequence was the idea that all knowledge must be like scientific
knowledge, which empiricists take to be purely empirical. Hence David
Hume’s famous declaration that all books not containing mathematics
or science should be burned, because they ‘contain nothing but soph-
istry and illusion’.3 Hume wrote in the eighteenth century; since then,
the tide of scientism only grew stronger.
7.1.3 The significance of empiricism

The more extreme forms of empiricism, particularly logical positivism,
have dramatically lost favor since the mid-twentieth century, at least
among philosophers. But even these extreme views remain important,
because they have left behind a pervasive influence on modern thought,
even among people who do not generally subscribe to them.
1
For classic statements of this view, see Carnap 1932 and Ayer 1952. See also Quine
(1951), who, despite the misleading title ‘Two Dogmas of Empiricism’, is concerned
to defend an even more extreme empiricism than that of Ayer and Carnap.
2
van Inwagen 1983, 215.
3
From David Hume (1975, III.iii, 165): ‘If we take in our hand any volume;
of divinity or school metaphysics, for instance; let us ask, Does it contain any
abstract reasoning concerning quantity or number? No. Does it contain any
experimental reasoning concerning matter of fact and existence? No. Commit it
then to the flames: for it can contain nothing but sophistry and illusion.’
Philosophical Preliminaries 95
In particular, logical positivism was so widely accepted in the early

twentieth century that it tended to be taken for granted in the devel-
opment of theories in science and mathematics. Positivistic assumptions
were crucial in motivating Einstein’s Theory of Relativity, the Copenhagen
Interpretation of quantum mechanics, and the formalist approach to
mathematics, to name three particularly prominent developments.4
Today, even those who would explicitly disavow positivism tend to accept
positivist-inspired theories without knowing their provenance. We gener-
ally trust that the scientists of the past considered all the relevant alterna-
tives before adopting the theories that they have passed down to us. But
in fact, alternatives that conflicted with positivism or empiricism tended
to be dismissed without fair consideration in the twentieth century.
It is hard to overstate the extent to which the thinking of an indi-
vidual is influenced by the intellectual culture in which the individual
is trained. This is clear in the case of scientists, who inherit a large body
of specific, detailed knowledge. In the case of philosophers, our inher-
itance tends to be more inchoate – rather than a body of well-defined
principles, we inherit certain broad habits of thought; a general sense of
what sorts of ideas are ‘plausible’; as well as some very broad philosoph-
ical viewpoints, such as ‘naturalism’. Vague as these things are, they are
of enormous importance in forming one’s philosophical views. (How
would your philosophical views differ if you had been trained in the
medieval philosophical tradition?)
Empiricism and positivism set the agenda of twentieth-century
analytic philosophy, which remains the leading philosophical tradition
in the English-speaking world. As a result, they have informed contem-
porary theories, philosophical methodology, and intuitions in a variety
of ways, both great and subtle. The influence of empiricism was particu-
larly powerful in the philosophy of science and philosophy of mathe-
matics, fields that have particular bearing on the subject of the infinite.
7.2 Phenomenal conservatism
I have elsewhere defended an epistemological thesis known as phenom-

enal conservatism.5 Because I have discussed it at length elsewhere,
4
See Einstein 1961, 2 (advancing a formalist-like philosophy of mathematics)
and 22–3 (advancing the positivist account of meaning as a prerequisite for
understanding relativity). For discussion of the influence of logical positivism on
the Copenhagen Interpretation, see Kenna 2011.
5
Huemer 2001; 2006; 2007. For discussion of the theory, including a variety of
objections and replies, see the papers in Tucker 2013.
here I will be brief. Phenomenal conservatism holds that undefeated

appearances are a source of justification (perhaps the only source of
justification) for belief. An appearance is ‘undefeated’ when there are
no positive grounds for doubting it. An appearance is a broad type
of mental state, distinct from and normally prior to belief, which is
reported by expressions such as ‘it appears to me that P’ or ‘it seems to
me that P’, where P is some proposition. There are a number of species
of appearances, including sensory experiences, memory experiences,
and rational intuitions. Rational intuitions are appearances that result
from intellectual consideration of a proposition, but not from infer-
ence. There are also appearances that occur during reasoning, known as
inferential appearances, whereby one proposition or set of propositions
seem to support another proposition.6 Examples of appearances include
the following: When I look out the window, it seems to me that there
is a squirrel in the tree (visual appearance); I seem to recall falling off
my motorcycle a few times in college (memory appearance); it seems
to me that the shortest path between any two points must be a straight
line (rational intuition); it seems to me that, given that cows have four
stomachs, it follows that cows have more than one stomach (inferential
appearance).
All beliefs that are initially plausible candidates for being justified
are based on appearances. This includes philosophical beliefs about
when beliefs are justified or not. Those who disagree with phenomenal
conservatism are just those to whom phenomenal conservatism does not
seem correct. But when one holds a belief, one is rationally committed
to holding the source of one’s belief to be an adequate source of justifi-
cation; one is in a self-defeating position if one thinks that one’s belief
is based on something that in no way justifies it. Therefore, once one
recognizes the actual source of our relevant beliefs, one must endorse
appearances as a source of justification. There are a number of other
considerations in favor of phenomenal conservatism, for which see the
previously cited works.
There is no rational basis for discriminating fundamentally among
species of appearances. What makes sensory experiences a source of
justification is the same thing that makes memories and intuitions a
source of justification: namely, all three mental states are appearances,
that is, they are all states in virtue of which the world seems a certain
way to the subject. It is, for example, arbitrary to hold that sensory expe-
riences may be presumed reliable but that intuitions should not be. It
6
Huemer 2016.
is equally arbitrary to hold that we must start by trusting intuition and

reason but not sensory experience.
Note that phenomenal conservatism is not the thesis that all appear-
ances are true. It does not hold that, once we describe how things
initially appear, inquiry is at an end. Rather, how things initially appear
is the rational starting point: things are presumed to be as they appear
unless contrary evidence (which itself would have to come from other
appearances) arises.
7.3 Synthetic a priori knowledge

7.3.1 The question of the synthetic, a priori
More than any other question, the question of synthetic, a priori knowl-
edge has dominated philosophy during the last three hundred years.
Here is some terminology:
● Empirical knowledge is knowledge whose justification depends essen-

tially on observations – that is, if A is known empirically, then there is
some set of observations that constitute the evidence that justifies A.
● A priori knowledge is knowledge that is not empirical. That is, its justifi-
cation does not depend essentially on any set of observations. (Note:
the question is not about how concepts are formed, whether there are
innate concepts, or whether there is innate knowledge. It is solely a
question about the justification for certain beliefs.)
● An analytic statement is, roughly, one whose denial is a contradic-
tion. For example, ‘All bachelors are unmarried’ is analytic, because
its denial – ‘Some bachelors are married’ – is a contradiction. Analytic
statements are sometimes called ‘true by definition’.7
● A synthetic statement is one that is not analytic, that is, its denial is
consistent. For instance, ‘Most bachelors are slobs’ is synthetic; it is
not contradictory to deny that most bachelors are slobs.
So here is the question that, as I say, has dominated the last three hundred
years of philosophy: is there any synthetic, a priori knowledge? In other
words, can there be a sentence whose denial is not contradictory, but
7
The definition of ‘analytic’ and ‘synthetic’ is tricky, and perhaps no defini-
tion is fully satisfactory. For the problem with the ‘true by definition’ or ‘true in
virtue of meaning’ characterization, see Quine 1936. For other (less convincing)
criticisms of the distinction, see Quine 1951. Nevertheless, standard examples of
‘analytic’ sentences seem to be interestingly different from standard examples of
‘synthetic’ sentences, and this shows, pace Quine, that there is some distinction.
nevertheless we can see it to be true in a way that does not depend on

observation for its justification?
7.3.2 The answer

There are a great many examples of synthetic a priori knowledge. Here
are twelve of them:
1. No object is completely purple and also completely orange.

2. For any two moments in time, one is earlier than the other.
3. Other things being equal, it is better to be happy than to be
miserable.
4. If x is inside y, and y is inside z, then x is inside z.
5. For any three outcomes, x, y, and z, if x is better than y, then a
lottery which gives a 10 per cent chance of getting x and a 90 per
cent chance of getting z is better than a lottery which gives a 10 per
cent chance of getting y and a 90 per cent chance of getting z.
6. The appearance of white is more similar to the appearance of yellow
than to that of any other chromatic color.
7. No action can cause an effect to occur before the action itself
occurs.
8. The number three exists.
9. There could have been only seven planets in the solar system.
10. The shortest path between any two points is a straight line.
11. The present is more real than the future.
12. The probability that either it is raining or it isn’t is 1.
I have tried to give a sense of how widespread and significant the

apparent synthetic a priori is. There are examples involving numbers,
spatial properties, colors, ethical value, decision theory, time, causation,
probability, possibility, and probably many more things.
Why describe these things as a priori? Take a famous epistemological
thought experiment: suppose that I am a disembodied brain being kept
alive by scientists in a vat of nutrients; my brain is being artificially
stimulated to create the illusion of a physical world, and so on. In this
case, all of my observations are false. Nevertheless, even if this scenario
is true, I still know that no object is completely purple and completely
orange. I of course don’t know that there are any orange objects or
purple objects in the real world outside my vat. But that doesn’t matter.
If there are no orange objects, then there also aren’t any objects that are
orange and purple. If there are some orange objects, then those objects
are not purple. I don’t need to exit the vat to know this.
But anything that I can still know even if all my observations are false
is something that is not dependent on observation for its justification.
That is, it is a priori. So my belief that nothing can be completely purple
and completely orange is a priori. A similar argument applies to all the
other items in the list above.
Why think that the items on that list are synthetic? To show that
one of those statements is analytic, this is what one would have to do:
produce the definitions of the relevant terms in the statement, substi-
tute the definitions for the terms they define, then, from the denial of
the resulting statement, derive a sentence of the form ‘A and not A’.
One should not merely assert that there is such a derivation; one should
give the derivation. If this cannot be done, then the original sentence
is synthetic.
Generally speaking, it cannot be done. Take the sentence, ‘No object is
completely purple and also completely orange.’ The project of showing
this to be analytic is stymied at the first stage by the fact that there is no
verbal definition of either ‘purple’ or ‘orange’. (There is only an ‘osten-
sive definition’, that is, these terms have to be explained by pointing
to examples of the relevant colors.) The sentence’s denial, ‘Something
is completely purple and also completely orange’, has the same logical
form as ‘Something is completely furry and also completely happy’,
which is not a contradiction.
For now, the above will have to do as a brief indication of some of the
grounds for rejecting empiricism. There is of course quite a bit more to be
said, and others have written much more extensively on the subject.8
7.3.3 How much synthetic a priori knowledge is there?

The central idea of empiricism is that one cannot attain substantive
knowledge of the world without relying on observation. Traditionally,
the distinction between analytic and synthetic statements matters to
empiricists because they think that analytic knowledge is not substan-
tive, that it is not genuine knowledge of the world, that it only reflects
knowledge of the meanings of words or concepts.9 For this reason, they
are willing to accept analytic a priori knowledge, but not synthetic a
priori knowledge. If there is even one example of synthetic, a priori
knowledge, then empiricism must be rejected.
Now suppose we must reject empiricism for that reason (as in fact
we must). Is there nevertheless some reason why we should endeavor
8
See especially BonJour 1998; Bealer 1992; 1999.
9
Incidentally, Quine (1936) showed this last claim to be untenable.
to minimize the extent of synthetic a priori knowledge that we

acknowledge?
I see no such reason. If there is at least one item of synthetic a priori
knowledge, then, whatever the source of that item of knowledge is, it
may also generate other items of knowledge. I see no reason to assume a
priori that the amount of knowledge attributable to this source is small,
rather than large.
What is the source of our synthetic a priori knowledge? The best
answer to this is rational intuition, and the best explanation for our
justification for relying on rational intuition is phenomenal conserva-
tism. If we accept rational intuition as a source of knowledge, there is
no reason why we must attempt to minimize our reliance on it. Rational
intuition is not relevantly different from observation, reasoning, or
memory for present purposes; all of these are simply different species
of appearances, and this is what makes each a presumptive source of
justified belief. To attempt to minimize the use of intuition would
be as sensible as attempting to minimize the use of observation or
reasoning. (On the other hand, we have good reason for seeking
multiple justifications for a given conclusion; if it is possible to support
a conclusion through both observation and intuition, this is preferable
to supporting it through only one of these means. But it is not the case
that it is preferable to ascribe knowledge to sensory observation rather
than intuition.)
7.4 Metaphysical possibility
7.4.1 The meaning of ‘possible’ and related terms

Now some more philosophical terminology: a proposition is possible if
it could be true (or could have been true). A proposition is impossible if
it is not possible. A proposition is necessary if it could not be (and could
not have been) false; that is, its denial is impossible. A proposition is
contingent if it is neither necessary nor impossible; that is, it could be
true and could be false.
All of these words, however, are ambiguous. There are multiple senses
of ‘possible’, and hence also multiple senses of ‘necessary’, ‘contingent’,
and ‘impossible’. In one sense, a thing is said to be possible when one
lacks sufficient evidence to rule it out. (‘I don’t remember where I left
my wallet. It could be in the restaurant.’) In another sense, a thing is
said to be possible when we have an available way of bringing it about.
(‘We could get the report done today, but we’d have to work overtime.’)
Neither of these senses is relevant to our inquiry; I mention them only

to contrast them with the type of possibility we’ll be concerned with.
Philosophers usually focus on one or two very broad senses of possi-
bility, ‘logical possibility’ and ‘metaphysical possibility’. A proposition
is said to be logically possible provided only that it is not contradictory.
This seems like a pretty uninteresting sense of ‘possible’, but more on
this later.
Metaphysical possibility is the broadest sense of ‘possibility’, that is,
the sense in which the greatest number of things count as ‘possible’,
short of the uninteresting ‘logical possibility’. Metaphysical possibility
is undefinable; however, a few things can be said to help point toward
its meaning, as follows.
7.4.2 Metaphysical possibility is broader than

physical possibility
According to the philosophical tradition, even things that conflict
with the laws of nature may nonetheless be in some sense ‘possible’,
because in one sense, the laws of nature could have been different.10
For example, one can say: ‘It could have turned out that we lived in a
perfectly Newtonian world.’ That sense of ‘could’ is metaphysical possi-
bility. However, not just any proposition is metaphysically possible; even
when we are speaking in the broad sense, such that the world could
have been Newtonian, it still is not the case that, for example, two and
two could have been five.
7.4.3 Metaphysical possibility as conceivability?

To say that something is metaphysically possible is something close to
saying that it is coherently conceivable. This characterization helps to
get people into the neighborhood of the concept of metaphysical possi-
bility, but it won’t do as a definition. A small problem: ‘conceivable’
just means ‘possible to conceive’, so we would then need to explain the
meaning of that use of ‘possible’.
A more important problem: a proposition might be impossible for
human beings to conceive solely because of some cognitive defect or
limitation on our part. For example, suppose there is a proposition
so complicated that to express it in English would require a book one
10
Although this is the received view in metaphysics, I think it may be false – I
think some or all of the laws of nature may be metaphysically necessary. However,
I shall not argue the point here.
million pages long. This proposition might be inconceivable to us purely

because of its complexity. This would not show that the proposition is
impossible in the intended, metaphysical sense. We could get closer to
the intended notion of possibility by speaking of conceivability by an
ideal observer, say, one with unlimited intelligence.11
7.4.4 Metaphysical possibility and science fiction

The notion of metaphysical possibility can also be illuminated by consid-
ering the limitations of fictional stories. In some perfectly good stories,
the laws of nature are different from what they are in the real world. For
instance, in Star Trek, people travel faster than the speed of light all the
time. Though this violates the actual laws of physics, viewers rarely find
it objectionable; we accept that the Star Trek universe has an alternative
physics.
By contrast, other aspects of stories are rejected by alert fans as ‘not
making sense’. Time travel stories sometimes evoke this kind of criti-
cism, as when a character is portrayed as going back in time and ‘altering
the past’ in a way that would prevent him from going back in time in
the first place. On the other hand, science fiction enthusiasts sometimes
try to invent alternate interpretations to defend the coherence of such
stories – for example, on the ‘branching timelines’ theory, when a person
‘goes back’ to some earlier time t, what actually happens is that a new
branch of the universe is generated in which things are exactly as they
were in the original universe at time t, except that the ‘time traveler’ is
added to the world; the new branch then evolves normally from there,
while the old branch is unaltered.
Now, don’t worry about whether time travel stories are really coherent,
or whether branching timelines is an admissible interpretation of such
stories. That isn’t the point of this discussion. The point here is to
gesture at a certain sort of possibility. Implicit in the discussion between
the critics and the defenders of time travel stories is the assumption that
there is a certain very broad sort of possibility that ought to constrain
stories, such that if a story is impossible in this broad sense, it is thereby
defective. But mere violation of the laws of physics does not make a
11
Even with this modification, this type of definition fails. It isn’t really the
ability to conceive a proposition that matters, since a person might conceive of P
in some sense (certainly, one might understand P) while having a strong intuition
that it is impossible that P. The type of conceivability we want would really be
something like ‘the ability to perfectly understand P without having the intuition
that it is impossible that P’. But the sort of ability there mentioned would prob-
ably itself need to be explained in terms of metaphysical impossibility.
story impossible in this sense – witness the lack of objection to faster-

than-light travel. This sense of possibility is the metaphysical: critics
are claiming that the time travel stories are metaphysically impossible.
Interpretations such as the branching timelines interpretation are
designed to show that the stories are metaphysically possible.
7.4.5 Metaphysical possibility as a priori knowability?

Metaphysical possibility is closely tied to the notion of a priori knowl-
edge. It appears that all the things that are known a priori are metaphys-
ically necessary; their denial is metaphysically impossible.12 This goes
for the analytic as well as the synthetic a priori. Thus, we know a priori
that pentagons have five sides, and it is metaphysically necessary that
pentagons have five sides; we know a priori that no object is completely
purple and completely orange, and it is metaphysically necessary that
no object is completely purple and completely orange.
Does the reverse implication hold – if a proposition is metaphysically
necessary, must it be knowable a priori?13 Surely not by us. Take the case
of Goldbach’s Conjecture, the hypothesis that every even number is the
sum of two prime numbers. This proposition is not intuitively obvious,
and thus far, it has yet to be proved or disproved. It may be impossible
to prove or disprove. (If it is false, then it could be disproved; one would
only need to exhibit the counterexample. But if it is true, then it might
be unprovable.) But it is not metaphysically contingent. If every even
number is the sum of two prime numbers, then this is a necessary truth;
if some even number is not the sum of two prime numbers, then that is a
necessary truth. So there might be a metaphysically necessary truth that
12
Possible exception: assume there is some variable, X, about which we have
no a priori knowledge except that the variable exists and has a continuum
of possible values. Let x be one of these possible values of X. Claim: we can
know a priori that X≠x. For the a priori probability that X=x is zero (each of
the continuum many possibilities has probability zero). But it is metaphysically
possible that X=x. Hence, we can have a priori knowledge of a metaphysically
contingent truth. However, this sort of case is not particularly relevant to the
issues we’ll be discussing.
13
Why do I not rely on analytic philosophers’ favorite example of a necessary
empirical truth, viz., [Water is H2O] (see, for example, Putnam 1975)? I think the
proposition [Water is H2O] might be identical with either [H2O is H2O] or [The
watery stuff is H2O] (it is odd to think that there should be three propositions
here; of course, there would have to be something like different modes of presen-
tation of a proposition). But neither of the latter two are examples of necessary
truths not knowable a priori. Hence, I am not certain that [Water is H2O] is such
an example either.
we cannot know a priori. As a matter of fact, we have excellent empirical

evidence for Goldbach’s conjecture: computers have checked all the
even numbers up to about 4×1018, and so far, every one has turned out
to be the sum of two primes.14
It may be that every necessary truth could be known a priori by an
ideal intellect, such as God. However, as this wouldn’t do us much good,
I shall not further discuss this possibility.
7.4.6 Logical vs. metaphysical possibility

Return now to the notion of logical possibility. A proposition is supposed
to be logically possible provided only that it is free of self-contradiction.
What is the relationship between metaphysical and logical possibility?
Are they the same?
From what we have said above, it is clear that they are not the same.
The synthetic a priori truths are metaphysically necessary, yet their denial
is not contradictory (this is what we meant by calling them ‘synthetic’).
It is just a paraphrase of this to say that they are not logically necessary.
Why should one care about logical possibility? The notion is of interest
to empiricists, since their view is that all a priori knowledge is of logi-
cally necessary truths, which is just another way of saying there is no
synthetic a priori knowledge.15 The synthetic, necessary, a priori truths
that I have been defending are anathema to empiricists.
Once we reject empiricism, however, logical possibility loses most
of its interest. The mere fact that a sentence fails to contradict itself
really does not constitute any interesting sense in which what it says
is possible. For instance, the statement ‘some object is entirely purple
and entirely orange’ is not a logical contradiction – it does not entail
a sentence of the form ‘A and not A’. However, there is no interesting
sense in which it is possible for something to be completely purple and
completely orange; if we choose to misleadingly label the sentence ‘logi-
cally possible’, then we must remind ourselves that there is no meta-
physical import to this ‘possibility’; it is merely a comment on a certain
formal property of the sentence.
14
e Silva 2014.
15
Some people would define ‘logical necessity’ even more narrowly, such that
only propositions derivable from the laws of logic, without relying on definitions
of any predicates or terms, count as logically necessary. Thus, ‘All bachelors are
bachelors’ would be logically necessary but ‘All bachelors are unmarried’ would
be logically contingent. This sense of ‘necessity’ is even less interesting.
7.4.7 False signs of possibility: Analogies and

mathematical systems
No one should believe that a proposition is metaphysically possible
merely because it is logically possible. Similarly, here are two other types
of bad reasons for ascribing metaphysical possibility:
First, the argument from mathematical systems: in this type of
argument, one cites the existence of a coherent mathematical system
as proof (or at least evidence) that it is possible for something in
reality to satisfy the system. This is fallacious. Mathematical systems
in modern times are constrained by nothing other than mere consist-
ency and mathematicians’ sense of what is interesting – they are not
constrained by metaphysical possibility. Thus, for example, the fact
that one can develop a coherent mathematical system in which one
talks about ‘infinite numbers’ such as Յ0, does not show that any
such numbers exist or could exist. Nor, similarly, does the exist-
ence of a consistent mathematical system for talking about ‘infini-
tesimal numbers’ mean that there are or could be any infinitesimal
quantities.
Second, the argument by analogy: A is possible, and there is an
analogy to be drawn between A and B; therefore, B is possible. This is
usually fallacious, particularly when the analogy is based upon mere
similarity of mathematical structure. For example, it is possible to
travel in either direction along a spatial dimension, and there is an
analogy to be drawn between the temporal dimension and a spatial
dimension (they have the same mathematical structure); therefore, it
is possible to travel forward or backward in time. The problem in this
case is that although time and space are similar in some respects, such
as the mathematical relations of their points, nothing at all follows
about their being similar in the relevant respect, that is, the possibility
of travel in them.
7.5 Possibility and paradox
How does all this bear on the infinite and its paradoxes? In the following
chapters, I argue that most of the paradoxes of the infinite involve
scenarios that are logically possible but metaphysically impossible. That is:
there are no formal contradictions in the descriptions of the scenarios,
but there is no metaphysically serious sense in which such things could
happen. This, I claim, resolves the paradoxes. However, such solutions
would not be available to those who deny the existence of synthetic
a priori knowledge or identify logical possibility with metaphysical

possibility.16
I take a liberal view of metaphysical necessity. That is, I hold that there
are many more necessary truths than are standardly recognized by philoso-
phers, scientists, and mathematicians; equivalently, I think there are many
fewer possibilities than are commonly recognized. I believe that the explana-
tion for the sparse view of necessities (and expansive view of possibilities)
taken by most today is that it is a holdover from empiricism and logical
positivism. For an empiricist, nothing can be ruled out a priori unless it is
actually a contradiction (or, of course, it conflicts with empiricism). Even
though empiricism is less popular than it once was, contemporary philos-
ophy retains a significant resistance to synthetic necessary truths.
Suppose we want to know whether some proposition P is metaphysi-
cally necessary, impossible, or contingent. Many think that the presump-
tion should be that P is contingent, that is, that one who thinks P either
necessary or impossible has a sort of burden of proof. This is understand-
able if one identifies metaphysical possibility with logical possibility: in
that case, to claim that P is impossible would be to claim that P entails a
contradiction, that is, something of the form ‘A and not-A’; likewise, to
claim that P is necessary would be to claim that the denial of P entails a
contradiction. If someone thinks either of these things is the case, it is
reasonable to ask them to show us the contradiction in question.
If, however, metaphysical possibility is distinct from logical possi-
bility, and if there are many things that are logically contingent yet
metaphysically necessary or impossible, then it is unclear why there
should be any such burden of proof. If, when one considers P, P initially
seems possible, then one should presume that it is possible unless and
until reasons to the contrary appear. Similarly, if P seems impossible, one
should presume that P is impossible, and so on. What if P neither seems
possible nor seems impossible – that is, we have no intuition either way?
In this case, one should withhold judgment, pending further argument.
One should not presume that P is possible.
7.6 A realist view of mathematics
A final preliminary topic: the nature of mathematics. Mathematics, in

my view, is a body of a priori knowledge, some of it analytic and some
16
Here, I disagree with Oppy (2006, 48), who contends that, in discussing the
paradoxes of the infinite, ‘the question is whether it is possible to tell consistent
stories’ involving the infinite. This is not the interesting question.
synthetic. This knowledge concerns the properties of and relations

between certain kinds of universals. These universals are properties and
relationships that other things could instantiate. For a mathematical
object or other universal to exist, it is not necessary that anything in
fact exemplify the property. It is, however, necessary that there could be
things exemplifying the property.
Thus, geometry studies shapes – their properties and the relationships
among them. Shapes are properties that physical objects could have.
Arithmetic studies the properties and relationships of numbers, where
numbers are properties that groups of things (whether physical or non-
physical) can exemplify – for example, whenever there are two of anything,
the property of twoness is exemplified. Probability theory studies prob-
abilities, which are degrees of support that propositions can have. Game
theory studies instrumental rationality, which is a property that intentional
actions can exemplify. And transfinite mathematics studies infinitude,
which is a property that certain collections have; for example, the collec-
tion of all the points of space exemplifies a certain kind of infinitude.
This view of mathematics (which is a form of mathematical Platonism)
always seemed to me completely natural, so much so that I have never
seriously entertained any other view. Nevertheless, the view is quite
controversial in contemporary philosophy of mathematics.
Some philosophers and mathematicians believe, instead, that mathe-
matics is essentially a conventional symbol-manipulation game: mathema-
ticians invent a series of symbols, arbitrarily stipulate rules for manipulating
them, and then work out what strings of symbols can be obtained by
playing according to the rules. In this theory, a mathematical system need
not have anything whatever to do with reality – the whole system is pure
invention, so there is no question of justifying the rules or securing ‘correct’
rules. If it should happen that some feature of reality mirrors the relations
among the symbols in the game, then we’ve gotten lucky.
That view of mathematics, known as ‘formalism’, is motivated by
empiricism. The empiricists of the twentieth century were loath to admit
that there was any a priori knowledge of objective reality, yet they had
to contend with the obviously a prioristic approach of mathematics –
mathematicians do not go out and do experiments or observations.
Some empiricists turned to the formalist philosophy of mathematics to
rescue them from what would otherwise be a striking collection of coun-
terexamples to their epistemological theory. If, as I have been urging, we
reject empiricism, there is no reason for believing formalism.
8
Sets
8.1 Sets are not collections
According to the now-standard view developed by the likes of Cantor,

Frege, and Russell, all of mathematics is based on set theory. Numbers,
functions, vectors, and all the other objects studied in other branches of
mathematics are constructed out of sets. This makes the concept of a set
very important! We should make sure that we understand this concept
very clearly.
I must confess, most unfortunately, that I do not. I doubt that you
do either. The most common explanation given is that a set is simply
a collection or group. But it is easy to see that ‘set’ in set theory is not
in fact used in the way that the rest of us (that is, people other than set
theorists) use ‘collection’. Note the following features of the concept of
sets:
i. It is uncontroversial in set theory that there is an ‘empty set’, a set

with no members. What collection is this supposed to be? Do we
really understand the idea that there is a collection when there is
nothing at all that is collected?
ii. It is uncontroversial in set theory that there are singleton sets, and
that they are distinct from their members. That is, for any object a,
{a}≠a. Again, do we really understand the notion that, when only
one object is present to start with, there is then a collection of that
one object – and that this collection is not the object itself but is
some other thing? If I have exactly one car, do I have a collection of
cars, which is a distinct object from my car?
iii. It is again uncontroversial in set theory that there are infinitely
many ‘pure sets’, that is, sets that are constructed, from the ground
108
Sets 109
up, using no objects other than sets. In fact, there is no limit to the
order of infinity one can construct using pure sets – Յ0, Յ1, Յɘ, and
so on. Are we to believe that this, too, is supported on the ordinary
notion of a collection? That in the ordinary English sense of ‘collec-
tion’, there are infinitely many infinite collections built up from
nothing but other collections, all starting from the collection that
you have when you have nothing to collect?
iv. It is commonly accepted in set theory (but this is somewhat more
controversial than the preceding points) that there are non-con-
structible sets – these are sets such that there is no way of describing
their membership. (We get this from the Axiom of Choice, which is
widely used but not uncontroversial.1) Is it true that, in the ordinary
sense of ‘collection’, there are collections such that there is no way
of specifying what is collected together?
v. In set theory, sets are typically understood as abstract, non-phys-
ical objects, even when their members are physical.2 Is a collec-
tion or group of physical objects an abstract, non-physical object?
When introducing the notion of a set in Chapter 6, I mentioned
that according to one textbook, a pack of wolves and a bunch of
grapes would be examples of sets. But these are not abstract objects:
a deer might be killed by a pack of wolves, but no deer is killed by
an abstract object; a person may eat a bunch of grapes, but no one
eats an abstract object. If the set theorists are to be believed, then,
in addition to the individual grapes, and in addition to the bunch
of grapes, there is also a third sort of object in the offing, an abstract,
non-physical, mathematical object, that has each of the grapes as
‘members’. Really, what reason is there for believing this?
Perhaps it will be said that, when we call a set a collection, ‘collection’

is not to be understood in the ordinary sense of the word but in some
technical, mathematical sense. But then we have made no progress at all
toward understanding ‘set’. It is as if I introduced the word ‘flumir’ by
explaining that a flumir is simply a smoople.
My reason for skepticism about set theory is not anything to do with
‘simplicity’ or ‘weirdness’, as are the usual reasons for skepticism on the
part of contemporary philosophers about abstract objects. I am not saying
1
See above, Chapter 6, fn. 13.
2
But see Maddy (1990, 58–67), who claims that sets built from physical
elements have spatiotemporal locations and can be perceived. On this unconven-
tional view, sets might no longer qualify as ‘abstract’ (Maddy 1990, 59); neverthe-
less, they are not material objects.
that we should doubt sets because they make our ontology larger, and I am
not saying that we should resist them because they’re unobservable, non-
spatiotemporal, or the like. I am saying we should be skeptical because no
one has been able to explain what these things are supposed to be.
8.2 Sets are not defined by the axioms
The second most popular explanation of the notion of a set is that the
term ‘set’ is defined by the axioms of set theory. That is, sets are simply
the things that satisfy those axioms.
There are two problems with this view. First, it assumes that we can
just see that there is something satisfying those axioms. I confess that
I don’t find this obvious at all; I simply don’t know what the objects
would be. They’re not collections. They’re not aggregates. I don’t see
that there is in reality anything having the characteristics mentioned in
points (i)–(v) above.
Second, the claim that the axioms of set theory define the term ‘set’
is refuted by a famous result in model theory: the Löwenheim-Skolem
Theorem. The proof of this theorem is beyond the scope of this book;
I will, however, explain what the theorem says.3 It should be noted
that logicians do not throw the word ‘theorem’ around; things that are
called ‘theorems’ (Gödel’s Theorem, the Löwenheim-Skolem Theorem,
the Compactness Theorem) are things that have been proved. Among
the experts in the subject, there is no controversy about them. In the
present case, this is what the theorem states:
Löwenheim-Skolem Theorem: If a countable set of sentences of first

order logic has a model, then it has a countable model.
To explain: a ‘countable’ set is a set with at most Յ0 members, that is, it

has either finitely many members or Յ0 members. ‘First order logic’ is a
type of formal logical system. Of particular interest here, the standard set
of axioms of set theory can be formulated as a countable set of sentences
of first order logic; thus, the Theorem applies to them.4 A ‘model’ for a
set of sentences is a way of assigning interpretations to the sentences
(specifying what the sentences are about, and what each of the predicates
applies to) that makes all of the sentences come out true. Finally, a ‘count-
able model’ is a model in which there are at most Յ0 objects.
3
For discussion of the mathematics, see Ebbinghaus 2007. For discussion of the
theorem’s philosophical implications, see Bays 2014.
4
For a list of the standard axioms of set theory, see above, Section 6.8.3.
Sets 111
So the Löwenheim-Skolem Theorem implies that, if there is any way

of interpreting the axioms of set theory so that they are all true, then
there is a way of doing so using at most Յ0 objects. But it is uncontrover-
sial among set theorists that there are (many, many) more than Յ0 sets.
That was proved in Chapter 6, Sections 6.6–6.7. So an interpretation of
the axioms that uses only Յ0 objects would be a misinterpretation, that
is, it would fail to capture what set theorists intend these axioms to be
about. Löwenheim-Skolem tells us that such a misinterpretation is not
ruled out by the formal structure of the axioms.
Now, according to the proposal under consideration in this section,
the concept of a set is supposed to be defined just by that formal struc-
ture. ‘Set’ and ‘member of’ are supposed to be defined merely as ‘what-
ever would make the following axioms true: ... ’, where you fill in the
ellipsis with a formal statement of the Zermelo-Fraenkel axioms. This
cannot be correct, since the axioms could be made true by an interpreta-
tion that clearly deviates from what set theorists intend. Hence, we have
yet to find an acceptable explanation of the meaning of ‘set’.
But you might object: ‘The preceding argument presupposes that you
actually do know what set theorists mean by ‘set’; otherwise, how could
you say what counts as a misinterpretation of the notion of a set?’ The
answer is that one can know some of the things that are supposed to
be true of X’s (according to the people who believe in X’s), even if one
has no idea what things might satisfy those conditions. In the present
case, we know that, according to the people who believe in sets, there
are supposed to be uncountably many of them. I know that much, even
though I don’t know what a set really is.
Another objection: ‘But the proof of the Löwenheim-Skolem Theorem
uses sets. How can you appeal to the theorem if you don’t believe in
sets?’ Reply: if sets exist, then the Löwenheim-Skolem Theorem is true,
in which case sets are not adequately defined by the ZF axioms. So either
sets don’t exist, or they are not adequately defined by the ZF axioms.
This is enough for present purposes.
8.3 Many regarded as one: the foundational sin?
Recall Cantor’s second characterization of sets: ‘A set is a many which

allows itself to be thought of as a one.’5 This might be the most philo-
sophically illuminating characterization.
5
Cantor 1932, 204.
Notice first, however, that this rules out the empty set, for surely a
mere absence of objects does not constitute ‘a many’. It also rules out
singletons, since a single object isn’t a many either. Even if we somehow
allowed singletons to fit under this definition, a singleton would not be
distinct from its ‘member’, for surely if x is already one object, we do not
convert it into a different object by regarding it as ‘a one’.
Leaving those cases, assume we have two or more objects. Cantor’s
suggestion appears to be that (at least in most cases), it is permissible to
regard these objects as one.
But wait. How can this be legitimate? By hypothesis, the objects under
consideration are many (or at least two). Two does not equal one, nor
does three equal one, nor do any of the other, larger numbers equal one.
Many things are not in fact a single thing. Does set theory then rest on
a contradiction?
Perhaps Cantor was speaking loosely. Suppose there are two objects,
a and b. When we form the set {a, b}, we do not literally treat the two
objects as one – we do not find ourselves saying that a = b. Rather, we
treat {a, b} as a third thing, which is identical with neither a nor b, but
which ‘contains’ each of them.
This is fair enough as a way of avoiding inconsistency. But notice
the split personality of sets. When the theorist initially introduces sets,
he downplays their ontological significance: gathering some objects
together into a set is not introducing some completely different kind
of object; it is just choosing to view the existing objects in a certain
way. Or so Cantor’s remark would lead us to think. We therefore need
not scruple about the addition to our ontology. But when the theory of
sets is developed, sets then are treated as a completely different kind of
object, distinct from their members but standing in some undefinable
relation to them.
I conjecture that the double life of sets is crucial to their widespread
acceptance. It is only in the first stage, when we are informally intro-
duced to sets as ‘collections of objects’ or as ‘multitudes regarded as
ones’ that it seems harmless to admit their existence. By the time we’re
being introduced to the cardinality Յɘ, we’ve forgotten what led us to
regard these objects as ontologically innocent to begin with.
If sets are understood as a fundamentally different kind of thing from
concrete objects (even sets whose members are all concrete objects),
then we need a more serious account of what these things are, and a
more serious reason for believing that such things exist, than we have
been given by Cantor, Frege, Russell, et al. If, on the other hand, sets
are understood in the seemingly ontologically innocent way suggested
Sets 113
by Cantor, as in some sense nothing over and above their members,

then we should worry about two things. First, we should worry about
the idea that one can somehow build up infinite hierarchies of infinite
sets, starting from only finitely many (or even zero!) concrete objects.
Second, we should worry about the seeming inconsistency of regarding
many distinct things as one thing.
8.4 The significance of the paradoxes
For those who believe in sets, the Naive Comprehension Axiom seems
intuitively obvious. This is the principle that for any well-formed predi-
cate, there is a set of the things that satisfy it. For example, if you have
the predicate ‘ ... is red’, then you can say there is a set of all the things
that are red. But this axiom generates both Russell’s Paradox and Cantor’s
Paradox.
This should have tipped us off that there is something wrong with
the notion of a set. The Comprehension Axiom was not, after all,
merely some peripheral, dubious hypothesis about sets. Before the para-
doxes were discovered, a proponent of sets would have said that the
Comprehension Axiom was a central and self-evident principle about the
way sets work. Giving up the Comprehension Axiom while still believing
in sets is comparable to believing in the addition operation but giving
up the idea that x+1 is greater than x.
If a concept generates paradoxes, that is generally a reason for thinking
it an invalid concept. Now, one might say: ‘But the paradoxes have now
been excised! As far as anyone can tell, the newer versions of set theory,
without the Comprehension Axiom, entail no contradictions.’ But the
ad hoc manner in which this result was achieved deprives it of evidential
import. If the foundational intuitions on the basis of which some objects
were initially introduced are proven to be contradictory, this removes
the central reason we had for believing in those objects. The fact that
someone can subsequently tailor the theory to avoid the contradictions,
in a completely ad hoc manner, hardly restores our reason for believing
those objects exist. The reason we have for believing in a set of all cats
is the same reason – in essence, the same intuition – that we had for
believing in a set of all sets.
Furthermore, the modifications made to the theory in this case are
particularly suspicious: we are supposed to believe in all the various indi-
vidual sets, but deny the existence of the set of all sets. Now, one of the
things you are supposed to be able to do with sets is to explain seman-
tics: according to many champions of sets, the meaning of a general
term is to be explained, in part, in terms of the set of things to which

the term applies. For instance, the meaning of ‘red’ is explained in terms
of the set of all red things.6 If this is correct, then if a general term has
no associated set (N.B., it isn’t merely that the set of things it applies to
is the empty set; it’s that there is no set of the things it applies to), the
term must be meaningless. But the term ‘set’, according to modern set
theory, has no associated set – there is no set of all the things it applies
to, since that would have to be the set of all sets. Therefore, it seems, the
term ‘set’ is meaningless.
Of course, one might simply say that the meaning of ‘set’ is not given
by the set of things to which it applies, but is to be understood in some
other way. Suppose there is some such other way. Maybe the meaning of
‘set’ is given by the Platonic Form of Sethood. But if there is such a Form,
then why wouldn’t there also be a Form of Redness, and why wouldn’t
that explain the meaning of ‘red’? A similar question would seem to
apply no matter what we say about the meaning of ‘set’ – which shows
that sets really don’t help us explain semantics after all.
None of this proves that there are no sets. It is not obvious that there
isn’t anything that ‘set’ refers to. But it is also not obvious that there is,
and there is no strong reason to believe that there is.
8.5 Are numbers sets?
Wait, how can I say that? Everyone knows that set theory is the founda-
tion of virtually all of mathematics! If we reject sets, won’t we have to
give up the rest of mathematics?
Let us examine the sort of foundation that set theory provides. In
Chapter 6, we saw the set-theoretic account (or at least one set theoretic
account) of the natural numbers. This was Russell’s view: The number
zero is just the empty set. The number one is the set of all sets that are
equinumerous with {0}, where two sets are defined to be ‘equinumerous’
if and only if there exists a one-to-one function from either set onto the
6
From Sider 2005, 2: ‘In linguistics, for example, one can think of the meaning
of a predicate, “is red” for instance, as a set – the set of all red things. Here’s one
hint of why sets are so useful: we can apply this idea to predicates that apply to
other predicates. For instance, think of “red is a color”. “Is a color” is a predicate,
and it looks like it holds of things like the meaning of “is red”. So we can think
of “is a color” as meaning a set containing all the colors. And these colors in turn
are sets: the set of all red things, the set of all green things, etc.’ Sider’s remarks
should presumably be taken to apply to one sort of ‘meaning’, namely, reference,
as opposed to sense.
Sets 115
other. The number two is the set of all sets that are equinumerous, in
that same sense, with {0,1}. And so on.
In other words, when I claim to have two apples in the cupboard, here
is what I am saying: there is a set of which every apple in my cupboard
is a member, and this set is itself a member of the set of all sets such that
there is a one-to-one mapping between them and the set which contains
only the following members: (i) the set that has no members, (ii) the set
of all sets such that there is a one-to-one mapping between them and
the set that contains only the set that has no members.
Now you understand why I didn’t try to explain the claim that there
are six apples.
Now that you’re comfortable with the natural numbers, let me explain
fractions.7 When I mention the number ⅔, I am referring to an equiva-
lence class of ordered pairs, namely, {‫ۃ‬2,3‫ۄ‬, ‫ۃ‬4,6‫ۄ‬, ‫ۃ‬6,9‫ۄ‬, ‫ۃ‬8,12‫ۄ‬, ... }. What
is an ordered pair? Well, the ordered pair ‫ۃ‬2,3‫ ۄ‬is actually just the set {{2},
{2,3}}. And you remember what 2 and 3 are. So actually the number ⅔
is an infinite set, whose first member is the set containing the following
members:
I. the set containing the set of all sets such that there is a one-to-one
mapping between them and the set containing (A) the set that has
no members and (B) the set of all sets for which there is a one-to-one
mapping between them and the set that contains only the set that
has no members, and
II. the set containing the following members:
A. the set of all sets such that there is a one-to-one mapping between
them and the set containing (1) the set that has no members and
(2) the set of all sets for which there is a one-to-one mapping
between them and the set that contains only the set that has no
members, and
B. the set of all sets such that there is a one-to-one mapping between
them and the set containing just (1) the set that has no members,
(2) the set of all sets for which there is a one-to-one mapping
between them and the set that contains the set that has no
members, and (3) the set of all sets such that there is a one-to-one
mapping between them and the set that contains (a) the set that
has no members, and (b) the set of all sets for which there is a
7
The explanation derives from Wildberger 2006, though Wildberger himself
hilariously ridicules this sort of construction.
one-to-one mapping between them and the set that contains the
set that has no members.
That is just the first member of the infinite set of ordered pairs that is the
number ⅔. I don’t have time to write down what the second member is.
Now you would like to know what a real number is, wouldn’t you?
Well, I don’t have time to tell you. Let’s just say it is a pair of infinite sets
of rational numbers.8 And now you have a sense of how set theory is the
foundation of the rest of mathematics.
But ... really? What is going on with theories like this? Are we supposed
to think that when I say my gas tank is two thirds full, I am discussing a
relation between the gas tank and that set I just started to tell you about
above (where I only had the patience to tell you the first member)? I find
it hard to credit that anyone, even a philosopher, believes that. What
crazypants semantic theory is going to make that set turn out to be the
referent of ‘two thirds’ in ‘The pie is two thirds gone’?
Either the set theoretic constructions are supposed to identify the actual
referents of mathematical terms, in their normal usage, or they are not.
If they are, then the constructions are wildly implausible on their face
and completely unjustified, as no attempt at all is made to argue that the
set theoretic objects are the actual referents. On the other hand, if the
constructions are not supposed to identify the actual referents of math-
ematical terms, then their philosophical significance is obscure. If Cantor,
Russell, et al., are merely saying, ‘Here are some objects that have the same
formal properties as the number system (because I deliberately constructed
them that way)’, I suppose this might be mildly interesting. But it hardly
justifies claims to have identified the foundation of mathematics.
8.6 Set theory and the laws of arithmetic
Here are some famous laws of arithmetic: where a, b, and c are any
numbers:
a+b = b+a (Commutative Law of Addition)

(a + b) + c = a + (b + c) (Associative Law of Addition)
a×b = b×a (Commutative Law of Multiplication)
(a × b) × c = a × (b × c) (Associative Law of Multiplication)
a(b + c) = ab + ac (Distributive Law)
8
Dedekind 1901. Other constructions are possible, but Dedekind cuts are
perhaps the most common way of constructing the real numbers.
Sets 117
How do we know these things?

Here is one answer: because they can be derived using one of the set-
theoretic constructions for the numbers. This answer is crazily implau-
sible. Derivations using set theoretic constructions tend to be complex
and difficult to follow; it is much harder to see that they are correct
than it is to see that the above laws are correct. Furthermore, the only
way anyone knows that a set theoretic construction is adequate is that
it yields the correct results. When someone wants to argue that we ‘can
construe’ the numbers 0, 1, 2, ... as the sets ∅, {∅}, {∅, {∅}}, ... , they do
this by arguing that those sets have the formal properties of the natural
numbers – that is, the properties that we already know the numbers have.
The set theorist will then define addition, multiplication, and other
operations and predicates, and argue that these operations and predi-
cates, on the proposed definitions, have the correct properties – that
is to say, again, the properties that we already know the arithmetical
operations and predicates have.9 Frege, after defining numbers, actually
says, ‘Let us try, therefore, whether we can derive from our definition
[ ... ] any of the well-known properties of numbers.’10 If, for example, a set
theoretic construal of arithmetic had the result that addition was not
commutative, that construal would ipso facto be rejected as inadequate.
This is hardly a startling revelation on my part. There is no one working
on the ‘foundations of mathematics’ who doesn’t know all this. Yet it
directly and obviously entails that what they are working on is not actu-
ally the foundations of mathematics, since the propositions that they
seek to derive are known prior to and (epistemically) independently of
the derivations. This epistemic independence is shown by the fact that,
if a proposed construction were to conflict with one of the ‘well-known’
properties of numbers, of addition, or the like, the well-known property
would not be in the least called into question; the construction would
just be refuted. In fact, the things being derived are used as justification
for the things from which the derivations start – for example, the ability
to derive commutativity is offered as (part of) a reason for construing
numbers as sets.
Now, one might be tempted to say: ‘Perhaps the set theoretic construc-
tions explain, not how we come to know the truths of arithmetic, but
rather what makes them true. The notion of ‘foundations’ of math-
ematics is ontological, rather than epistemological.’ But this, again,
is just implausible, because on the proposed account of what makes
9
See Cantor 1915, 85–103; Frege 1980, 81–96.
10
Frege 1980, 81, emphasis mine.
arithmetical truths true, there is no plausible way in which we could know

those facts independently of knowing anything about set theory. If, for
example, the number ⅔ is the sort of incredibly complicated set that we
discussed in Section 8.5 above, and similarly with all the other numbers,
then there is no plausible way in which people could have known that
2 1 5 without knowing any set theory and without doing any
+ =
3 6 6
derivations using it.
Conclusion: set theory is not the foundation of arithmetic, let alone
all of mathematics. The rejection of sets therefore would not undermine
the foundations of mathematics in any interesting sense.
9
Numbers
9.1 Cardinal numbers as properties
Having rejected the set theoretic foundation for arithmetic, let us return
to the beginning: what is a number?
Begin with the cardinal numbers, that is, the counting numbers. A
cardinal number is a kind of property.1 These properties are best ‘defined
ostensively’, that is, by giving examples. Suppose I want to explain what
the number two is, to someone who has never heard of it. I would show
them something like Figure 9.1.
In that picture, there are two stars, two hexagons, and two lightning
bolts. That is, two is the property that the stars instantiate, and the hexa-
gons also instantiate it, and the lightning bolts also instantiate it.2 It is
no coincidence that this is how children are actually taught number
Figure 9.1 Three instances of twoness
1
My view here is completely unoriginal: it is based on Byeong-Uk Yi’s excellent
(1999) ‘Is Two a Property?’ For discussion of the logic of plural predication, plural
quantifiers, and so on, see his 2002.
2
Abstract objects can also instantiate numbers, for example, redness and square-
ness are two universals. I use examples of concrete particulars for convenience.
119
concepts (not by telling them about the set of all sets with one-to-one
mappings onto the set {0,1}!). The ontological status of the number two
is thus the same as that of other universals, such as redness and square-
ness. If you believe in those other properties, as well you should, then
you should also believe in numbers.
9.2 Frege’s objection
Gottlob Frege thought he had a refutation of the view that numbers are
properties of concrete objects.3 His argument: the same concrete thing
can be said to instantiate different numbers. Suppose you have a deck of
cards. What is the number that it instantiates? Well, it is one deck, but
it is fifty-two cards. But if numbers are properties, presumably different
numbers are incompatible properties, so the same thing couldn’t possess
both oneness and fifty-twoness at the same time. Therefore, number is
not a property of a concrete object, such as the deck of cards. Instead,
Frege proposes, numbers must be properties of ‘concepts’ (but note that
Fregean ‘concepts’ are not mental; they are more like what the rest of
us would call properties). Thus, we resolve the puzzle of the deck of cards
by saying that there are two distinct concepts: the concept ‘deck’ (or
maybe, ‘deck that is on this table now’) and the concept ‘card’ (or, ‘card
that is on this table now’), and the number 1 applies to the former,
while the number 52 applies to the latter. Cantor and Russell, on the
other hand, would ditch the talk of concepts and say that there are two
distinct sets: the singleton set of which the deck is the sole member, and
the fifty-two-membered set which has each card as a member.
Frege’s argument refutes a theory about numbers, but not my theory.
The view Frege refutes is one that holds the following:
i. Each number is a property of a concrete particular (at least, a prop-

erty that may be instantiated by a concrete particular),
ii. In the deck example, the deck has the property 1, and the cards have
the property 52, and
iii. The deck is identical with the cards.
Indeed, that is a confused view; Frege is right about that.

Here is how I would correct that view: First, thesis (i) is false, because
there is only one number that is capable of being a property of a concrete
particular, and that is the number 1. It is logically impossible, for example,
3
Frege 1980, 28–9.
Numbers 121
for the number 2 to be a property of a concrete object; the number 2 can

only be a property of two concrete objects. In general, the number n can
only be a property of n objects. Numbers are different from other proper-
ties in this respect: most predicates are capable of being instantiated by a
single subject, but natural numbers, as predicates, require plural subjects
(except for the number one, which requires exactly one subject). The
fundamental twoness facts are facts of the form T(a,b) (in English: ‘a
and b are two’), where T is the property of twoness, and a and b are the
subjects which instantiate that property. They don’t each instantiate the
property; they jointly instantiate it.4 The plurality of the subjects is an
irreducible aspect of the proposition: it is not that twoness applies to the
set {a,b}; it’s not that there’s a special thing called a ‘pair’ and twoness
applies to the pair; and it’s not that a and b are parts of a larger object,
a+b, where twoness applies to the composite object a+b. All of those
are variants on the idea that there is a thing (that is, one thing) which
instantiates twoness, which is contradictory. The only subjects involved
in the situation are a and b, and they, that is, those two things, together
instantiate twoness.
Second, thesis (iii) above is incoherent. What does it mean to say some-
thing is identical with more than one thing? Identity is a two-place rela-
tion, and a single thing must play the role of each term of the relation.
This is part of the grammar of ‘=’. Thus, ‘the deck is identical with the
cards’ is not merely false; it is malformed. The only way to make sense
of it is to modify it. Thus, we could interpret it as ‘The deck is identical
with each of the cards’ (obviously false), or ‘The deck is identical with the
set of the cards’ (also false), or ‘The deck is identical with the aggregate
of the cards’, where the aggregate of the cards is understood to be the
composite object that has each of the cards as parts and has no other
parts that don’t overlap with those. This last statement is the only one
that isn’t obviously false.5 But that interpretation generates no problem.
If we’re talking about the aggregate of the cards, that instantiates the
number 1. It does not instantiate 52. Remember, the aggregate of cards
is an object, which has each of the cards as parts. So its parts number 52,
but it numbers 1. There is no way of getting one thing instantiating two
different numbers.
4
Note also that the T(a,b) here does not ascribe a relation between a and b. It
ascribes a property, but the subject of the property is irreducibly plural.
5
It’s not obviously true, either. If we replace a single card from the deck,
perhaps the result would count as ‘the same deck’, but as a different aggregate.
But nevermind that.
9.3 Arithmetical operations
What is addition? ‘Two apples plus three apples make five apples’
means something like this: if you have two apples, and you also have
three more apples (that is, three that are each different from either of
the original two), then you have five apples. What of the statement
of pure mathematics, ‘2 + 3 = 5’? This means something like: if there
are two (of anything), and there are three (of anything, but different
from the original two), then the two things and the three things are
five.
What is multiplication? ‘2 × 3’ means something like: the result of
taking two, three times, or: the number of things that there are, when
there are three separate instances of twoness.
What is subtraction? ‘3 − 2’ means something like: the result of excluding
two things from three things, or: the number of things that remain, if,
out of three things, two things are excluded from consideration.
And division? ‘6 ÷ 3’ means something like: the number of instances
of threeness that would constitute six things.
Notice that none of the arithmetical operations is a physical opera-
tion, not even when applied to physical objects. The idea of adding
two apples to three apples does not refer to doing anything physically
with the apples. It is not, for example, a matter of bringing the apples
into spatial proximity. If it were some physical operation, then arith-
metic would be contingent and empirical, since it is possible that upon
moving objects into closer proximity, or performing whatever other
physical operation, we would find that the total number of objects
changes (perhaps apples at a certain distance from each other merge,
or destroy each other, or reproduce). Adding two apples to three apples
is solely a matter of considering two apples and, without making any
changes to any of the apples, considering an additional three apples,
and figuring out how many apples that would make. There is not
even any passage of time assumed: if at t there are two apples and
at the same time there are an additional three apples, then, at that
time, there are five apples; the arithmetical statement does not entail
any empirical prediction that there will continue to be five apples at
any later time.6 This is why arithmetical statements are necessary and
knowable a priori.
6
This is not meant to exclude cross-temporal counting. For instance, if Iskra ate
two apples yesterday and three apples today, then she ate five apples in two days.
Numbers 123
9.4 The laws of arithmetic
Recall the most famous laws of arithmetic:
a+b = b+a (Commutative Law of Addition)

(a+b)+c = a+(b+c) (Associative Law of Addition)
a×b = b×a (Commutative Law of Multiplication)
(a×b)×c = a×(b×c) (Associative Law of Multiplication)
a(b+c) = ab+ac (Distributive Law)
Why do these hold, and how do we know them?

In essence, the reason why (a + b) is equal to (b + a) is that the expres-
sions ‘(a + b)’ and ‘(b + a)’ are synonymous: both simply refer to the result
of combining those two numbers. The order in which the numbers’
names are written on the page has no semantic significance, any more
than the color of the font in which they are written. Here is a simple
illustration. You manage an airline. You have two airplanes parked at
terminal A, and another three parked at terminal B.
✈✈ ✈✈✈
A B
How many airplanes do you have in total? Okay, now what if you
‘instead’ had three airplanes parked at terminal B and two at terminal
A; then how many airplanes would you have in total? Now, it’s not that
these two questions happen to have the same answer. It’s that these are
the same question. ‘Two airplanes at terminal A and three airplanes at
terminal B’ is synonymous with ‘three airplanes at terminal B and two
airplanes at terminal A’. That is why addition is commutative. A similar
observation applies to Associativity.
Why is multiplication commutative? The following diagram illustrates
the meaning of 2 × 3 and 3 × 2:
9 9
9 9 9
9 9
9 9 9
9 9
The left part of the picture illustrates 2 × 3, as it contains two rows of

three circles. The right part of the diagram, in turn, illustrates 3 × 2,
using the same representational conventions. Notice that the right hand
picture is just the same as the left one, rotated ninety degrees. You can
see that a similar diagram could be created for n × m, for any natural
numbers n and m. And you can see that the point applies not just to
circles but to any objects, as the circles can be taken merely as symbols
for any objects (including non-spatial objects). That is how we know
that multiplication is commutative.
A similar diagram (in three dimensions) can be used to illustrate
Associativity; the generalization is obvious.
Why does the Distributive Law hold, that is, a(b + c) = ab + ac? Again,
a diagram:
9 9 9 + 9 9 9 9
9 9 9 + 9 9 9 9
Each row represents 3 + 4, and there are two rows, so that shows
2(3 + 4). Now, here is a representation of 2 × 3 + 2 × 4:
⎡9 9 9 ⎤ ⎡9 9 9 9 ⎤
⎢9 9 9 ⎥ + ⎢9 9 9 9 ⎥
⎣ ⎦ ⎣ ⎦
On the left are two rows of three, and on the right two rows of four, so
we have two times three, plus two times four. But this is the same as the
previous diagram, except with brackets and plus symbols rearranged.
The brackets and plus symbols are extraneous; they are just there to
direct your attention. You can see now that this sort of illustration could
be made for any other three numbers besides 2, 3, and 4. This is why
Distribution holds.
Now, to get to the philosophical point. I am not trying to convince you
of the truth of the laws of arithmetic, as I presume you already believe
them. I am trying to convince you that we don’t need any complex set
theoretic derivations, or any other abstruse reasoning, to explain our
knowledge of these principles of arithmetic. Nor are they arbitrary rules
that mathematicians just made up. The laws of arithmetic are simple,
necessary truths, which when seen in the right light, are self-evident.
9.5 Zero
Here is an objection: ‘Zero is the same kind of thing as the (other) natural
numbers. If the number n is instantiated by n things, then the number
Numbers 125
zero must be instantiated by zero things. But there cannot be a property

that, when it is instantiated, is instantiated by zero things. So zero is not
a property. So the natural numbers are not properties.’
Reply: zero is not the same kind of thing as one, two, three, and so on.
In the original, core sense of the term ‘number’, zero is not a number.
We can christen zero as a number, but this is an extension of the concept
‘number’. Because it is an extension of the concept, we should not be
surprised that the ‘number’ zero differs in important ways from the orig-
inal numbers.
Why is zero not a number in the original sense? Because a number, in
the primary sense, is a property that objects can have, whereas zero is
not a property that objects can have. This remains plausible in ordinary
language: if I have zero cookies, I should not say, ‘I have a number of
cookies’; I should say ‘I do not have any cookies.’ It is no coincidence
that the concept of zero has a quite different history from that of the
positive integers, having started life as a mere placeholder – something
like a punctuation mark used to indicate that a certain position was
unoccupied.7
Again, we find an objection in Frege: A cardinal number is an answer
to a ‘how many?’ question. The answer to ‘How many cookies do you
have?’ might be ‘Zero’. Therefore, zero is a number.8
To borrow from Husserl’s reply: The answer to ‘How many cookies
do you have?’ might be ‘None’. But none isn’t a number.9 Furthermore,
the answer ‘none’ means the same as the answer ‘zero’ in this context;
therefore, zero isn’t a number either.
Compare Frege’s argument to the following: ‘A time is an answer to
a “when” question. Now, the answer to “when are you going to admit
that set theory is a fantasy?” might be “Never”. Therefore, never is a
time.’ The conclusion is false; never is not a time. Not all answers to
‘when’ questions name times; one kind of answer to a ‘when’ ques-
tion is an answer that denies that there is any time such as the one being
asked about. Only a positive answer to a ‘when’ question is a time. By
the same token, we could at most say that any positive answer to a ‘how
many’ question is a number. ‘None’ would be a negative answer. Despite
7
On the history of zero, see O’Connor and Robertson 2000.
8
Frege 1980, section 44, p. 57.
9
Husserl (2003, 137–8) gives this argument, including the diagnosis that only
a positive answer to ‘how many ... ?’ names a number. Benardete (1964, 32–42)
also endorses this line. Frege (1972, 333–4) responds unconvincingly. Husserl and
Benardete also deny that one is a number, since they think a number must be a
multiplicity; on that point, I must differ with them.
the surface grammar, the logical form of ‘I have no cookies’ is not the
same as that of ‘I have two cookies’; the latter ascribes twoness to my
cookies, whereas the former simply denies that I have cookies. What
holds of ‘none’ and ‘no’ holds for ‘zero’, since it is synonymous in these
contexts: all three words, unlike the positive numerical adjectives, are
used to make negative statements.
Objection: ‘But we can do arithmetical operations using zero. How
could that be, if zero is not a true number, or not a number in the same
way that the other numbers are?’
Reply: When we decide to extend the number system by including
zero, we define arithmetical operations for zero in such a way as to keep
the whole system coherent. When you don’t have any A’s, we decide
to say ‘there are zero A’s’. ‘a + b’, in general, means something like: the
number of things that there are, when there are a things, and there
are also b additional things. ‘a + 0’ means: the number of things there
are, when there are a things, and there are not any additional things.
Similarly, a × b, in general, means something like: the number of things
that there are, when there are b occurrences of a things. ‘a × 0’ means:
the number of things that there are, when there aren’t any occurrences
of a things.
Famously, there is one case in which we do not define the arithmetical
operations for zero, namely, the case of division. This is because there is
no natural extension that makes sense. ‘a ÷ b’ means something like: the
number of instances of b that would constitute a things, or: the answer
to the question, ‘If you divide a things into equal-sized groups, such
that each group contains b things, how many groups would there be?’
But then ‘a ÷ 0’ would mean something like: the number of instances
of not-having-anything that would constitute having a things, or: the
answer to the question, ‘If you divide a things into equal-sized groups in
such a way that each group doesn’t contain anything, how many groups
would there be?’ These things make no sense; that is why division by
zero is prohibited.
You might wonder: if the other numerals, in their noun uses, refer
to properties, what is the referent of ‘0’ in its noun usage? The answer
is that the symbol does not require a referent to be meaningful. In the
same way that the noun ‘nothing’ lacks a referent, ‘zero’ lacks a referent.
Expressions that contain ‘0’ are to be understood as described above, for
example, as in the above explanation of ‘a + 0’. When we say ‘there are
a things, and there are not any additional things’, the expression ‘not
any’ does not require a referent.
Numbers 127
The point here is not to expel zero from the number system. After all,
the extension of the number system to include zero has proved fruitful.
The point is simply that ‘zero’ functions differently in some important
ways from ‘one’, ‘two’, and so on, and that to understand the use of
‘zero’ does not require positing a property of zeroness.
9.6 A digression on large numbers
To some, the natural numbers seem too familiar and too effable to be
interesting. But some natural numbers are ineffable.
Let me tell you about my favorite natural number. It is too large to
write its conventional representation in this book. In fact, even in scien-
tific notation, there isn’t enough material in the observable universe to
write this number. To tell you what it is, I have to introduce some new
terminology.
Multiplication is repeated addition: a × b is the result of adding a to
itself repeatedly, with b occurrences of a. Exponentiation, by the same
token, is repeated multiplication. So 33 = 3 × 3 × 3 = (3 + 3 + 3) + (3 + 3 + 3)
+ (3 + 3 + 3). Exponentiation suffices for expressing even the largest
numbers that scientists have occasion to discuss. For example, there are
about 1080 atoms in the observable universe; it was easy to write that
using exponentiation.10
Now we can define another operation known as ‘tetration’, symbol-
ized by ‘՛՛’, to represent repeated exponentiation.11 So 3՛՛2 is 33, which
3
is 27, and 3՛՛3 would be 33 327 = 7, 625, 597, 484, 987. Then 3՛՛4 is
37,625,597,484,987, which I can’t show you in conventional decimal nota-
tion, because my publisher won’t pay for the 1.8 billion pages of type
that would be required to write it down.12 When we get to 3↑↑5, we’re at
a number too large to be written in decimal notation using the resources
of the observable universe.
After that, there is the operation of pentation (symbolized by ‘՛՛՛’),
which is repeated tetration. These operations are just the start of an infi-
nite series of mathematical operations:
10
Villanueva 2009.
11
The notation, known as ‘Knuth up-arrow notation’, derives from Knuth
(1976, 1235–6). Knuth represents exponentiation with a single arrow, ‘↑’.
12
It’s about ten to the 3.6 trillionth power, so the number would have 3.6 tril-
lion digits. Assuming about 2000 characters per page, I’d need 1.8 billion pages.
With extra small type, we might be able to get by with only 200 million pages;
the publisher still wouldn’t go for it. Note that exponentiation, tetration, and the
33 ( 33 ) 3 3
like, are right-associative, so for example, 3 is read as 3 , not ( ) .
First operation: a+b

Second operation: a × b
Third operation: ab
Fourth operation: a՛՛b
Fifth operation: a՛՛՛b
⁝
Now let’s define a new operation, which I will represent ‘✳’. ✳n is

the result of applying the nth operation in that list to the number n
with itself, with n repetitions of n. Thus, here are some numbers we
can refer to using this notation, along with their conventional decimal
representations:
✳1 = 1
✳2 = 2×2 = 4
3
✳3 = 33 = 7,625,597,484,987.
✳4 = 4՛՛4՛՛4՛՛4 = (uncomputable)
⁝
Now, the ✳ operation is iterable: for example, ✳✳8 is the number

that results from applying the ✳ operation to the number ✳8, that is,
applying the ✳8th operation in that list repeatedly to the number ✳8,
with ✳8 repetitions. Let us use the expression ‘✳nk’ to denote the result
of prefixing n occurrences of the ✳ operator to the number k.
Finally, my new favorite number: ✳✳88. I suspect that it is the largest
finite number ever referred to. Of course, it would now be easy to name
much larger numbers, but I don’t like any of them as much, so I will
refrain from identifying any of them.
Now, you may think that ✳✳88 is a freakish number, and indeed I
know of no use for such a number other than making points about
incomprehensibly large numbers. But in a purely statistical sense, ✳✳88
is not freakishly large; it is, rather, freakishly small. For remember, there
are infinitely many natural numbers. So in particular, there are infinitely
many that are larger than ✳✳88, and only finitely many that are smaller.
We might say: to a first approximation, all numbers are larger than ✳✳88.
If God were to pick a natural number at random, he would, with prob-
ability 1, pick one larger than ✳✳88.13 ✳✳88 is also unusual in another
13
This is assuming, falsely, that it makes sense to speak of picking a natural number
at random. But though this doesn’t really make sense, it illustrates the point.
Numbers 129
respect (besides its smallness): namely, that it can be expressed using

an existing human symbol system (albeit one just recently invented).
100
In the same way that a googleplex ( 10 ) is too large to express using
decimal notation (given our physical constraints), and ✳5 is too large to
express using exponential notation, there are infinitely many numbers
that remain too large to express even using the ‘✳’ notation, because the
number of ‘✳’’s that would be required is too large to fit in the observ-
able universe. Even among the numbers less than ✳✳88, almost all of
them remain in practice inexpressible. ✳✳88 is expressible only because
it happens to result neatly from applying certain mathematical opera-
tions that we can independently grasp, a small number of times. But, for
example, between ✳✳87 and ✳✳88, there are very many numbers, almost
all of which no one will ever be able to name, because they cannot
be obtained by applying any defined operations a sufficiently small
number of times. To a first approximation, all the natural numbers are
both incomprehensible and unnameable.
9.7 Magnitudes and real numbers
9.7.1 Magnitudes vs. numbers

Magnitudes are a fundamental and pervasive feature of the world. The
concept of magnitude is probably undefinable, but here are some exam-
ples: the height of the Eiffel Tower, the temperature of a cup of coffee,
the mass of the Milky Way Galaxy. Just as cardinal numbers are proper-
ties, it is even clearer that magnitudes are properties, typically properties
of concrete objects. The height of the Eiffel Tower is a property of the
Eiffel Tower, the mass of the Milky Way is a property of the Milky Way,
and so on.
‘Magnitude’ is ambiguous between a sense that refers to a dimen-
sion of a certain kind (as in ‘Length is a magnitude’) and a sense that
refers to a particular value on such a dimension (as in ‘The height of the
Eiffel Tower is a magnitude’). In most cases, context makes clear which
sense is intended. If clarification is needed, I shall speak of ‘magnitude
dimensions’ or ‘magnitude values’; for instance, height is a magnitude
dimension, while the height of the Eiffel Tower (324 meters) is a magni-
tude value.
Magnitude values are represented using real numbers. The magni-
tudes, however, are not themselves numbers. Here, a Fregean-style
argument against numbers as properties of concrete objects applies,
and this time the argument succeeds. The Eiffel Tower is approximately
324 meters tall, which is to say about 1063 feet tall. If the height of the
Tower is a number, which one is it? Is it around 324, or around 1063?
Different units of height could be used to justify assigning any positive
real number whatsoever to represent the height of this same Tower. But
if numbers are properties, then different numbers would presumably
be incompatible properties; the same object could not instantiate two
of them, let alone all of them (above zero) at the same time. Therefore,
the number that we use to represent the height of the Tower is not a
property of the tower. But the height is a property of the tower. So the
height is not a number.
9.7.2 Real numbers as relationships

What, then, is a real number? Earlier, I briefly mentioned a way of
constructing real numbers from sets; it involves infinite sets of rational
numbers, with the rational numbers themselves being infinite sets of
ordered pairs of natural numbers. That is a very unworldly and, to me,
unnatural approach. Here is a different approach: start with how we use
real numbers to describe the world. We say that the Eiffel Tower is 324
meters tall. What is ‘324’ doing there; what role is it playing in the truth-
conditions of that statement?
The answer is that the ‘324’ records a relationship between the height
of the Tower and a unit, the unit in this case being ‘the meter’. The meter
used to be defined as the distance between a certain pair of marks to be
found on the Prototype Meter, which was a particular platinum-iridium
rod located in Sèvres, France. (Today, the meter is defined in terms of the
speed of light, but let’s stick with the Prototype Meter for simplicity.)
The metaphysical background is that magnitude values come in classes
that are comparable to each other. Every length is comparable to every
other length, in the sense that for any two lengths, the first is greater
than, less than, or equal to the other. Similarly, every temperature is
greater than, less than, or equal to every other temperature. But lengths
are not comparable to temperatures – the Eiffel tower is not taller than it
is hot, nor shorter than it is hot, nor yet equally tall as it is hot.
When a magnitude is greater than another magnitude, this greater-
ness comes in degrees. The Eiffel Tower and the Burj Khalifa are both
taller than my house, but the Burj Khalifa is much more taller than my
house than the Eiffel Tower is. The degree to which the Eiffel Tower’s
height exceeds that of my house is a relationship between the Tower
and the house.
Similarly, the Eiffel Tower’s height stands in a certain relationship to
the length of the Prototype Meter – the ‘324-times-greater’ relationship.
Numbers 131
That relationship is what the number 324 is. The relationship is a

universal, since other pairs of magnitudes can be related in that way.
The relata are also universals, since other objects could have the same
height the Eiffel Tower has, and other objects could have the length
the standard meter has. On the other hand, the Eiffel Tower’s height is
related in a different way to the unit ‘foot’, which is another universal,
from how it is related to the unit ‘meter’. It is related to the foot in the
‘1063-times-greater’ way. That is why both numbers, 324 and 1063, can
be used to represent the Tower’s height.
9.7.3 In defense of the bifurcated account of number

In the real world, nothing’s length is ever a whole number of meters,
unless it is so by definition (for example, the Prototype Meter is precisely
one meter long, but only because this is true by definition). The Eiffel
Tower is not exactly 324 meters tall. Nevertheless, it could have been
exactly that tall, that is, this is metaphysically possible. This possibility
raises the following objection: ‘324 is a real number, since an object
could be exactly 324 times as long as the standard meter. 324 is also a
natural number. But on your account, a natural number is a property,
whereas a real number is a relationship. How can the same thing be both
a property and a relationship?’
Reply: It turns out that there are two numbers 324: the cardinal number
and the real number. The cardinal number 324 is the property that is
instantiated by any 324 things. The real number 324 is the relationship
that almost holds between the Eiffel Tower and the standard meter. In
the possible world where the Eiffel Tower was exactly 324 meters tall, it
would be the real number 324, not the cardinal number, that applied to
the Tower’s height. Note incidentally that, on the set-theoretic construal
of numbers, it is also true that there are two distinct 324’s: the natural
number 324 is supposed to be the set of all 324-membered sets, while
the real number is a pair of infinite sets of rational numbers.
Now you might wonder: why not take a uniform view of cardinal
numbers and real numbers? There are two natural ways of doing this:
(i) First, we could say that natural numbers are relationships, just as
real numbers are. We could address Frege’s deck-of-cards problem by
saying that the same physical aggregate is differently related to two
different units: it bears the ‘52-fold’ relationship to the unit ‘card’,
while bearing the ‘one-fold’ relationship to the unit ‘deck’.
I once held this view. But here is why the ‘relationship’ theory is
inferior to the ‘property’ theory of cardinal numbers. The relationship
theory requires that there be an object to be the first relatum, that

is, the thing that is supposed to stand in a relationship to a unit. In
the deck-of-cards example, this is no problem: the first relatum is
the deck, or the aggregate of the cards. But what about this case: I’m
thinking of two things: Beethoven’s Ninth Symphony, and my cat.
What exactly is related to what in the ‘two-fold’ manner? Perhaps
the unit in question is ‘thing’. What is the object that stands in a
relationship to that unit – the object that is composed of Beethoven’s
Ninth Symphony and the cat? It is quite doubtful that there is any
such object.
Another case: suppose there are seven reasons for being suspicious
of set theory. Then on the relationship account, there is something
that is related in the ‘seven-fold’ way to the unit ‘reason for being
suspicious of set theory’. But is there such an object, an object that
has each of the seven reasons as parts?
We could avoid these problems by invoking the idea of irreducibly
plural subjects of a proposition, but once we do that, it seems that
we don’t need to posit a relationship to explain number; a property
with plural subjects will do.
(ii) Alternately, we could say that real numbers are properties in the
same way that natural numbers are. Perhaps the height of the Eiffel
Tower in meters should be construed as the number of meter-long
segments that make up the tower’s height.
But how would one accommodate this case: the temperature of
my coffee is 325 degrees Kelvin. Are we to say that this means there
are 325 of something in the coffee? Or this case: I put my thumb on
a scale and exert 5 pounds of force. The ‘number as property’ theory
would seem to require there to be five subjects of this proposition,
which collectively instantiate fiveness. Presumably, it would have to
be five acts of exerting one pound of force. But there aren’t any such
things; there is only the one force.
The theory seems adequate to the Eiffel Tower example because
the Eiffel Tower can be divided into separate parts, each with its own
height, where these heights add up. But the five-pound force doesn’t
contain five one-pound parts, nor does the coffee contain 325 one-
degree-Kelvin parts.
9.7.4 Measures and magnitudes

Above, I suggested that the notion of a magnitude is undefinable. There
is a branch of mathematics known as measure theory, which studies
Numbers 133
‘measures’. Because measure theory might be thought to provide insight

into the nature of magnitudes, I briefly discuss it here.
Suppose there is a set S (it doesn’t matter what its members are). And
suppose there is a function, m, from subsets of S to numbers (where
‘infinity’ counts as a number), and that m satisfies the following
conditions:
i. It is nonnegative, that is, the values it assigns to subsets are always

either zero, some positive real number, or infinity.
ii. It assigns to the empty set the value 0.
iii. It is additive, that is: if S1, S2, ... are non-overlapping subsets of S, and
‫׫‬Si is the set formed by combining S1, S2, ... (taking all the members
from all those subsets), then m(‫׫‬Si) = m(S1) + m(S2) + ... .
In that case, m is referred to as a ‘measure’ on the set S.14 This captures

the minimal requirements of a function that could be construed as
telling us the sizes of the subsets of S.
Example: suppose we have an infinitely long line. We can divide it
into various segments. We might say one segment is two meters long,
another 6.7 meters long, and so on. Our concept of length-in-meters
provides a measure in the above defined sense: S would be the set of
points on the line. m is the function that assigns to (the set of points
making up) each portion of the line the length in meters of that portion.
For example, for the two-meter segment, m assigns the number 2 to the
set of points in that segment. m doesn’t assign a negative number to
anything, because no part of the line has a negative length. m assigns
to the empty set the number zero, because intuitively, if you take no
points at all, then you have zero length. Illustrating additivity, if you
have a 2-meter segment and a non-overlapping 6.7-meter segment, m
will assign to the combined segment (that is, the set of all the points in
either segment) the number 6.7 + 2 = 8.7. The set of all the points on the
line will naturally be assigned the ‘number’ infinity, because the line is
infinitely long.
How might the notion of a measure shed light on the concept of a
magnitude? Perhaps we could define a magnitude simply as a value
14
Additional conditions: m need not be defined (that is, it need not assign a
value) for every subset of S. However, m must be defined for at least one of the
subsets of S, and for every pair of subsets for which m is defined, m should also
be defined for their intersection, their union, and their difference. Furthermore,
for any countable collection of subsets for which m is defined, m should also be
defined for their intersection. See Cortzen and Weisstein 2015.
assigned by a measure, in the mathematical sense of ‘measure’. This would

involve rejecting my above account, since this approach assumes that real
numbers can be understood prior to understanding magnitudes, and it
also rejects the distinction between magnitudes and the real numbers used
to represent them.15 Of course, the standard definition of measure also
makes liberal use of sets, about which I have expressed some skepticism.
Two reasons why this approach is wrong: First, the mathematical
notion of a measure is much too weak to capture that of a magnitude.
Consider the function, defined on the set of all sets of points in three-di-
mensional space, which assigns 0 to the empty set, and which assigns to
any other set of points the value ∞. That is a measure in good standing
in the mathematical sense. But there is no corresponding magnitude
dimension. It is not true, for example, that a single geometric point has
an infinite magnitude, just because some mathematical function arbi-
trarily assigns it the value ‘∞’. Similarly, there are infinitely many more
measures that fail to correspond to any magnitude.
Second, only some magnitude dimensions have measures in the
mathematical sense. Consider the temperature of a cup of coffee. The
left half of the cup might be 325 Kelvin, and the right half likewise 325
Kelvin, but the whole is not 650 Kelvin. Temperature thus violates the
additivity condition.
9.7.5 Intensive vs. extensive magnitudes

As the preceding example illustrates, there are different kinds of magni-
tudes. Extensive magnitudes are those that are additive across the parts
of an object. That is, if E is an extensive magnitude dimension, then if E
applies to some object and the object can be decomposed into some non-
overlapping parts, then the magnitude values of all of these parts will
contribute additively to the magnitude value of the whole.16 For example,
length is an extensive magnitude. If you have a line segment, and you
divide it into a left half and a right half, the length of the left half plus
15
The latter problem could doubtless be fixed. Perhaps we could identify a
magnitude with a set of ordered pairs, where the first member of each pair is a
measure, and the second member is the value that that measure assigns to the
thing whose magnitude we are considering.
16
I have deliberately described extensive magnitudes in this way so as to allow
the possibility of treating mass as extensive, despite what is said about it below in
the text. I prefer, however, to avoid having to classify mass. For more discussion
of the nature of extensive magnitudes and the possibility of infinite extensive
magnitudes, see Oppy 2006, pp. 137–41.
Numbers 135
the length of the right half will equal the length of the whole segment.
Similarly for volume: if you add together the volumes of a number of
(non-overlapping) regions of space, you get the volume of the combined
region. Similarly for duration: if you add together the durations of some
non-overlapping time intervals, then you get their combined duration.
Intensive magnitudes, by contrast, are magnitudes that do not arise
from adding together the magnitudes of the parts of an object, even when
the object is decomposable into non-overlapping parts. For example, the
temperature of the left half of my cup of coffee plus the temperature of
the right half does not equal the temperature of the whole cup of coffee.
Other intensive magnitude dimensions include the pitch and loudness
of sounds, pressure, matter density (the mass per unit volume in some
region), and brightness of colors.17
There may be ambiguous cases. Consider the property of mass. We
generally think of the mass of an object as the sum of the masses of its
non-overlapping parts. For instance, my mass is the sum of the masses
of the molecules that compose me. However, in chemistry, it is known
that part of the mass of a molecule is contributed by its potential energy,
and the potential energy does not belong to any particular part of the
molecule but instead arises from the configuration of the parts. So the
mass of the molecule is not exactly equal to the sum of the masses of
its component atoms. However, mass is certainly closer to typical exam-
ples of extensive magnitudes than it is to typical examples of intensive
magnitudes, since the mass of the molecule is the sum of the masses
of its component molecules and the mass contributed by the potential
energy. It would therefore be strange to call mass an intensive magnitude.
Perhaps it were better styled an approximately extensive magnitude.
One other addendum: it may be that the intensive/extensive distinc-
tion applies only to natural magnitudes. I discuss this notion in the
following subsection.
9.7.6 Natural vs. artificial magnitudes

Consider the (alleged) magnitude, ‘schmass’, where the schmass of an object
is defined to be the reciprocal of 3 minus the object’s mass in kilograms:
1
s=
3−m
17
Aside: brightness and saturation of color are magnitudes, but it is doubtful
that hue should be considered a magnitude. One shade may be twice as bright or
saturated as another, but not twice as hued.
For example, when an object has a mass of 5 kg, it has a schmass of

−½ kg−1.
This seems to me, in some sense, not a genuine example of a magni-
tude. It is not really part of the object’s nature; it is just a number that
one can obtain by performing an arbitrarily chosen series of arithmetical
operations on (the number representing) the object’s mass, which is the
only real magnitude involved in the story. However, I will soften that a
bit and simply call schmass an ‘artificial’ magnitude, in contrast to the
‘natural’ magnitudes that objects instantiate, such as their mass, height,
temperature, and so on.
As a first approximation, I will take causal efficacy as the test of natu-
ralness. An object’s schmass does not causally explain any aspect of its
behavior, so we can infer that schmass is an artificial magnitude. But an
object’s mass, temperature, height, and so on are each causally effica-
cious in a variety of circumstances.18
Causal efficacy is not identical with naturalness. There could be a
magnitude that is a genuine aspect of an object’s nature but yet has no
effect on anything else (that is, it is epiphenomenal) – this is neither
conceptually nor metaphysically impossible. However, I know of no
examples of such magnitudes, and I think there is no reason for believing
that there are any. On the other hand, I do not believe that an artificial
magnitude could be causally efficacious. However, I leave open the possi-
bility that naturalness might be relative to a set of causal laws, such that
a magnitude that is artificial in one possible world might be natural in
a world where the causal laws were different. (I do not claim that this is
possible; I merely refrain from taking a stand here, since I don’t want to
argue about whether it is possible.)
The distinction between natural and artificial magnitudes is not to be
confused with that between fundamental and derived magnitudes. For
example, mass density can be calculated in terms of other magnitudes
(density = mass per unit volume) that might be regarded as more funda-
mental. In that sense, it is a derivative magnitude; density, however,
can most certainly be causally efficacious. For example, the density of
a sample of fissile material influences whether the material sustains a
nuclear chain reaction (critical mass depends on density). So there is
18
Note: if a given magnitude value is (or would be) ever causally efficacious,
then the magnitude value is natural. I phrase the criterion in terms of magnitude
values rather than dimensions to avoid the problem of artificially constructed
‘magnitude dimensions’ – for example, there might be an artificial ‘dimen-
sion’ some of whose values are masses and some of whose values are instead
schmasses.
Numbers 137
no difficulty in having a natural magnitude definable in terms of other

magnitudes. What is wrong with schmass is not that it can be calculated
from, or is definable in terms of, mass. What is wrong with schmass is
that it is just a made up magnitude that doesn’t affect anything.
9.8 Indexing uses of numbers
Sometimes, number words are used in a manner that does not reflect
the nature of numbers as I have explained them earlier. Thus, suppose
you are in a hotel, in room 210. What is the number 210 doing there? Is
it functioning as a cardinal number, thus indicating that there are 210
of something; or a real number, indicating that something is 210 times
greater than some unit?
Of course, the answer is neither. There needn’t be 210 rooms in the
hotel, for example, for the hotel to have a room 210, nor need there be
210 of anything else relevant, nor need any relevant magnitude be 210
times greater than anything. In this case, ‘210’ is simply being used as a
name for that room. It would have been just as good to call it ‘room 2H’,
or ‘room Sally’.
That was a pretty clear case of a number term being used in a way that
does not really refer to a number. Some cases are a little more ambig-
uous. Suppose the date is 10 July. What is ‘10’ doing there? One might
say again that ‘10’ is simply being used as (part of) a name for that
particular day, just as in the case of ‘room 210’. Or one could say that it
is actually the cardinal number 10, indicating that ten days have passed
since the end of June. (But then, when fifty days have passed, why may
one not say that the date is the 50th?) Thus, to find out how many days
later the 20th is than the 10th, one can subtract 10 from 20. I won’t here
try to resolve what the best interpretation is. What matters is just to see
that there are cases shading from clearly using numerals as mere names,
to using numerals in somewhat name-like and somewhat number-like
ways, to using numerals to clearly refer to numbers.
There are artificial uses of real number expressions as well. Thus,
consider the Celsius temperature scale. If an object has a tempera-
ture of 10 ºC, this does not mean that it is ten times hotter than an
object that is only 1 ºC. If an object is 0 ºC, this does not indicate a
complete absence of heat (or anything else). Nor, finally, does a nega-
tive temperature indicate the presence of some sort of anti-heat that
would cancel out an equivalent quantity of heat (nor does it indicate an
opposite of any other quantity). Celsius temperatures do, however, indi-
cate some genuine quantitative facts. For example, if three objects have
temperatures of 0 ºC, 10 ºC, and 20 ºC, then one can infer that the first
and second differ in temperature just as much as the second and third.
On the other hand, in the Kelvin temperature scale, ‘0’ really does
indicate a complete absence of heat, an object at 10 K is ten times hotter
than one at 1 K, and there are no negative temperatures because there
is no anti-heat.19
So one way to view the Celsius scale is that it is an artificial way of
measuring temperature, which is a natural magnitude. Another way to
view it is that there are two magnitudes (or ‘magnitudes’): temperature
is a natural magnitude, which can be measured in degrees Kelvin. Then
there is the artificial magnitude, ‘temperature − 273.15’, which is what
the Celsius scale measures (0 ºC = 273.15 K). Note that, when tempera-
ture figures in laws of nature, as in the Ideal Gas Law, it is temperature in
some absolute scale (that is, a scale in which ‘0’ refers to absolute zero, as
in the Kelvin scale) that must be used. This coheres with the observation
that only natural magnitudes have causal efficacy.
A similar case is that of longitude. Differences in longitude numbers
represent genuine quantitative differences. However, ‘0º’ is a mere name
for a certain meridian, rather than a genuine number term; the ‘0’ does
not, for example, indicate that some quantity is completely absent.
This is what I refer to as an indexing use of numbers. In an indexing
use, some number is used as a name for a more or less arbitrary item or
value, but then other numbers are assigned to other items or values,
according to genuine quantitative relationships to the first item or
value. Thus, the Prime Meridian is arbitrarily named ‘0º’, but then other
latitude numbers indicate amounts of angular deviation from the Prime
Meridian.
9.9 Other numbers
This section is a digression. Nevertheless, it may be of interest to discuss

how I would construe certain other kinds of numbers, given my philo-
sophical approach to mathematics, which I describe as a sort of ‘worldly
Platonism’. By that, I mean that although I regard mathematical objects
as perfectly real, non-physical, and independent of human observers,
I also believe that the best way of understanding the nature of a
19
Here I speak of temperature as understood in terms of the mean kinetic
energy of molecules. In the thermodynamic definition of temperature, where
T = ∂E/∂S, negative temperatures become possible (Quantum Munich 2014). For
discussion, see below, section 10.4.3.
Numbers 139
mathematical object is to consider its role in helping us understand the

concrete world. This is because, although a mathematical object need
not be realized in the concrete world in order to exist as a mathemat-
ical object, it must be in principle capable of being concretely realized.
Moreover, we typically form concepts for the purpose of understanding
the concrete world.
9.9.1 Rational vs. irrational numbers

Non-integer rational numbers are the same kind of thing as real numbers:
2/3, for example, is a relationship between a given aggregate and a unit.
The rational numbers differ from the irrational numbers only in that
the rational numbers are more readily expressible in our language (for
example, rational numbers have either finite or repeating decimal expan-
sions, whereas irrational numbers have infinite non-repeating decimal
expansions) – there is no metaphysical difference between these kinds
of numbers. Thus, when I say that my gas tank is two thirds full, I am
reporting a relationship between the volume or mass of the gas in the
tank and the volume/mass of a full tank (thus, I treat the full tank as the
unit of measure). When I say the pie is two thirds gone, I am reporting
the quantitative relationship between the current size of the remaining
pie and the original size of the pie before any of it was eaten. Most such
statements are of course only approximations.
9.9.2 Negative numbers

On the face of it, the notion of a number less than zero appears inco-
herent, since it is impossible to have less than nothing. If, for example, I
have no beer at all in my refrigerator, then no one can have less beer in
their refrigerator than I have.
Negative numbers only make sense in certain contexts. Sometimes,
there are two quantities that are related as opposites, such that the
combination of the two quantities is, at least for some purposes, equiva-
lent to the absence of both. For example, if the bank wants to evaluate
your suitability to take a loan, it may make sense for them to calcu-
late a number called your ‘net worth’ by subtracting your debts from
your assets, thus treating each dollar of debt as canceling out a dollar of
assets. In this case, a debt can be viewed as a negative asset, or an asset
as a negative debt.
Or suppose that an engineer is analyzing an airplane design to assess
the ability of the airplane to fly. In this case, it makes sense to calculate
a number to be called the ‘net lift’ on the airplane in a certain scenario,
by subtracting the downward forces on the airplane from the upward
forces, thus treating each pound of downward force as canceling out a

pound of upward force. The airplane will then climb only if this number
is positive. Alternately, we could treat downward forces as positive and
upward forces as negative; in that case, the airplane climbs if and only if
the net force is negative.
A negative number, then, simply represents a magnitude that is the
opposite of some other magnitude we have chosen to regard as positive.
Negative values only make sense for some quantities and in some contexts,
because there are only some cases in which a quantity has an opposite.
9.9.3 Imaginary numbers

Students are often taught that there is a special number called an ‘imagi-
nary number’, symbolized ‘i’, which is the square root of negative one,
that is, it is the solution to the equation ‘x2 = −1’. A complex number is
then understood as a number of the form a + bi, where a and b are any
two real numbers.
Many find complex numbers puzzling, because it seems that there
simply is no number that when multiplied by itself yields negative
one. One cannot conjure such a number into existence just by fiat, as
the standard approach seems to suppose. Trying to solve the equation
‘x2 = −1’ by inventing a new number is like trying to solve the equation
‘x + 1 = x’ by inventing a new number, q, defined to be the number that
is one greater than itself.
Nevertheless, there is an acceptable substitute for the notion of −1.
Instead of thinking of a complex number as the result of adding a real
number to the product of a real number and the square root of negative
one, we can simply think of a complex number as an ordered pair of real
numbers.20 Now, ordinary addition and multiplication are operations on
numbers, not pairs of numbers, so we will need to define special senses
of ‘addition’ and ‘multiplication’ for pairs. We define them as follows:
〈 〉 + 〈c
〈c , d 〉 = 〈 + c , b + 〉
〈 〉 × 〈〈cc , d 〉 = 〈 − bd , ad + 〉
where all the occurrences of addition, multiplication, and subtraction

on the right hand sides of the above equations refer to the ordinary
20
This was discovered by the mathematician William Hamilton (1837). A
generalization of this approach is known as the Cayley-Dickson Construction
(Baez 2002, 153–6).
Numbers 141
arithmetic operations on numbers, and the ‘+’ and ‘×’ on the left
hand sides refer to the new operations of addition and multiplication
of ordered pairs. This interpretation gives all the same properties of
complex numbers as the more familiar approach, only without positing
that −1 has a square root.
9.9.4 Infinitesimal numbers

An infinitesimal number is a number that is smaller than any positive
real number but yet greater than zero. One can also think of an infinites-
imal as the result of dividing a finite number by an infinite number. The
mathematician Abraham Robinson worked out a consistent theory of
infinitesimals (and infinite numbers), called ‘non-standard analysis’.21
Infinitesimals are sometimes used in explaining concepts of calculus.
For instance, a derivative may be thought of as the ratio of two infini-
tesimal quantities, dy/dx, where dy is the infinitesimal increase in the
variable y that would correlate with an infinitesimal increase in x, dx.
Note that the standard analysis treats talk of infinitesimals as a mere
façon de parler; the true definition of dy/dx is given in terms of the limit
of a certain ratio of (real) numbers, and the point of the standard delta-
epsilon definitions is precisely to eliminate all reference to infinitesimal
(or infinite) quantities. Robinson’s non-standard analysis, however, takes
talk of infinitesimals literally.
Infinitesimals may also be invoked to address the paradox of geometric
points: instead of viewing a line segment as made up of points each with
a size of zero, we can view a line segment as consisting of infinitely
many infinitesimal points, where the product of an infinite number and
an infinitesimal number can be a positive real number.
A third application of infinitesimals is in probability theory. In cases
where an event has a continuum of possible outcomes, the standard
analysis assigns an initial probability of zero to each outcome. For
example, suppose that a dart is to be thrown at a dartboard. Assume
that the dart is guaranteed to hit some single geometric point on the
dartboard, but that where the dart hits will be completely random. For
any given point, the probability of the dart hitting that specific point is,
according to the standard analysis, zero. (It must be zero, because if you
pick any real number greater than zero, then you get a total probability
21
Robinson 1966. Robinson’s theory includes infinite numbers along with the
infinitesimals. Robinson’s infinite numbers are not to be confused with Cantor’s
infinite cardinalities; Robinson’s infinite numbers would represent infinite magni-
tudes, just as real numbers represent finite magnitudes.
greater than one.) Some believe that a more intuitive treatment is to say
that each point has an infinitesimal probability of being hit.
The only problem that I have with all this is that there simply are
no such numbers. Or, to rephrase my point in a manner that sounds
slightly less question-begging: I have no clear concept of a number that
is smaller than every positive real number but more than zero. The real
numbers, as far as I understand them, go all the way down to zero; they
fill all of the space between, for example, zero and one. For instance,
there are the numbers 0.1, 0.01, 0.001, and so on, and one can continue
in that way indefinitely, finding real numbers as small as you like. I have
no notion of a degree of smallness such that the real numbers never
reach it. Indeed, it seems to me self-evident that there is no such degree
of smallness.
The same is true in the direction of largeness: the real numbers get as
large as you like; I have no notion of a size so big that the real numbers
never reach it. I thus think Robinson’s infinite numbers are as much a
fiction as his infinitesimals.
10
Infinity
10.1 Infinity is not a number
10.1.1 The concept of infinite number

Although there are actual instances of infinitude, there are no infinite
numbers. When, for example, there are infinitely many A’s, the A’s do
not instantiate some specially large number; rather, the A’s are literally
numberless, or beyond number.1
The concept of an infinite number is that of a number so large that it
exceeds all the more familiar numbers, the numbers of the sort discussed
in Chapter 9. Thus, an infinite cardinal number would be a cardinal
number larger than every natural number. An infinite magnitude would
be a magnitude too large to be represented by any real number.
This, I claim, is an empty concept. It is not that the concept is contra-
dictory; it is simply that we have no clear conception of anything that
could satisfy it. The real numbers, so to speak, go all the way up. That is,
there are real numbers as large as you like; there is no degree of largeness
that they fail to reach. The same is true of natural numbers. (Think about
the number ✳✳88, and remember that almost all the natural numbers are
larger than it.) Thus, there cannot be any number greater than all the
finite numbers.
Why do the followers of Cantor hold that there are infinite numbers?
Because they have a theory of the nature of numbers, which identifies
natural numbers with a certain sort of set. Once one does this, one can
go on to notice that there are sets of the same kind as those one identi-
fied with the natural numbers, except that these new sets would have
to correspond to infinite numbers. Thus, if we identify the number 2
1
Cf. Wittgenstein 1975, 157, 162.
143
with the set of all sets such that there is a one-to-one mapping between
them and the set {0, 1}, it is then natural to ask about the set of all sets
such that there is a one-to-one mapping between them and the set
{0, 1, 2, ... }. The latter would appear to have an equally good claim to
being considered a number, but it would have to be an infinite number.
This argument collapses once we reject the set theoretic account of the
natural numbers. Natural numbers are not sets; they are plurally instanti-
ated properties. On this latter theory, are there infinite cardinal numbers?
That is, are there properties that are of the same sort as the natural
numbers, except that they would be larger than every natural number?
The property theory alone does not answer this – that is, merely from
the idea of a cardinal number as a property that requires plural subjects,
one can make no inference as to whether there is or is not a property
of that kind that requires infinitely many subjects. The claim that there
isn’t rests on further intuitions. But the property theory of numbers,
unlike the set-theoretic account, at least allows the possibility that there
are no infinite numbers.
Doubtless what I have said so far is insufficiently persuasive. Intuitions
about the natural numbers and the real numbers ‘going all the way up’
are meant to cohere with the arguments that appear in the following
sections, strengthening them by offering a conceptual explanation of
the fact that infinity is not a number.
10.1.2 The arithmetic of infinity

A number is the sort of thing to which arithmetical operations apply. A
number can be added to another number, multiplied by another number,
and so on. This is not an accidental or insignificant feature of numbers; it
is part of the essence of a number. If someone announces, ‘I am thinking
of a certain number. However, it does not make sense to talk about adding
anything to this number’, then I think we should say this person either
fails to understand what a number is, or fails to understand what addition
is. If there is a number of objects, how could it not make sense to speak
of how many objects would result from combining them with another
number of objects? Or to speak of how many objects would result from
excluding some number of objects from the original group?
But as we have seen earlier, the normal arithmetical operations do not
apply to infinity. Sometimes, in informal discourse, infinity is treated as
though arithmetical operations apply to it, in which case the friends of
infinity will assert such things as
∞ + 1 = ∞. (Equation 10.1)
Infinity 145
But this attempted equation makes no sense. It says that a certain number
is one unit larger than itself, or that a certain number is not increased by
adding to it, either of which is absurd. In addition, if we admit the above
equation, there seems to be no clear reason why we should not subtract
∞ from both sides of it, thus obtaining ‘1=0’.
Suppose we ban the subtraction of infinite numbers but permit addition
involving infinity. Then we could allow Equation 10.1 while avoiding
the consequence that 1=0. However, as long as we recognize −∞, we will
have the same problem over again:
−∞ + 1 = −∞. (Equation 10.2)
We ought to be able to add ∞ to both sides of Equation 10.2, again

obtaining ‘1=0’. The best course seems to be to ban addition and subtrac-
tion of infinity.
What about multiplication and division? Most partisans of infinity as
a number will agree with statements such as:
1/∞ = 0. (Equation 10.3)

2/∞ = 0. (Equation 10.4)
Combining Equations 10.3 and 10.4, we then learn that 1/∞ = 2/∞, and
therefore 1=2. Moreover, once we reject addition of infinity, we should
reject multiplication by infinity on conceptual grounds, since multi-
plication is just repeated addition. It seems that the best course is to
prohibit multiplication and division by infinity.
Now it looks as though infinity does not bear any arithmetical opera-
tions, which strongly suggests that it is not a number.
One might say: ‘But there is at least one number for which at least one
arithmetical operation is proscribed: division by zero makes no sense.
But we do not conclude that zero is not a number.’
The fact that one cannot divide by zero reflects the fact that zero is,
in fact, not the same kind of number as the positive integers. The inclu-
sion of zero represents, as we have said (Section 9.5), an extension of
the concept of number. Now, one might ask: why not make a further
extension so as to include infinity in the number system? But infinity
is much more different from the natural numbers than zero is. Infinity
shrugs off all the arithmetical operations. I submit that this makes it too
unlike the standard numbers for a reasonable extension of the concept
‘number’ to include it.
10.1.3 Galileo’s argument

Recall Galileo’s paradox from Section 3.5:
i. It seems that there are many more natural numbers than perfect
squares, because the natural numbers include all the perfect squares,
plus many more numbers besides. But,
ii. it also seems that there must be at least as many perfect squares
as natural numbers, because every natural number has a distinct
square.
Galileo drew the conclusion that one cannot apply the notions of
‘greater’, ‘less’, and ‘equal’ to infinities. One cannot say that there are
more squares than natural numbers, nor fewer, nor yet the same number;
one can only say that each class is infinite.
But how could a number fail to be greater than, less than, or equal to
another number? Again, this is not an incidental feature of numbers.
It is essential to a number that it be a determinate size, which must be
more than, less than, or the same as any other determinate size. The best
explanation of the fact that ‘the number of perfect squares’ is quantita-
tively incomparable to ‘the number of natural numbers’ is that there are
no such numbers.
Nowadays, this argument is commonly rejected, based on the
Cantorian criterion for numerical comparisons in terms of one-to-one
mappings. The followers of Cantor simply accept (ii) above (the argu-
ment for why there must be at least as many squares as natural numbers),
while rejecting or ignoring (i) (the argument for why there must be more
natural numbers than perfect squares). But there is no actual argument
for rejecting (i), other than that (i) conflicts with (ii) (which is in fact the
argument that is always cited). This is unsatisfactory, because both of the
following are self-evident truths about the greater-than relation:
1. A number is made greater when one adds to it. That is, if there are
n things, and more things are added without removing any of the
original things, the result is that there are more things.
2. If the members of some collection can be paired up with the members of
another collection, then neither collection is greater than the other.
(2) is no more obvious or essential to the ordinary notion of ‘greater

than’ than (1) is. Both are essential. Therefore, if both principles cannot
be satisfied when we try to apply the notion of ‘greater than’ to infinite
sets, what that shows is not that one of the above principles is false;
Infinity 147
what it shows is that ‘greater than’, in the ordinary sense, does not apply
to infinite sets. To put the point without reference to sets: if there are
infinitely many A’s and there are infinitely many B’s, then there are not
more A’s than B’s, nor fewer, nor the same number. And the reason for
this is that there is no number of either A’s or B’s; the A’s and the B’s are
beyond number.
10.1.4 Back to Aristotle

Aristotle’s argument against actual infinities was essentially this (Section
5.2): any actual thing must be fully determinate, but infinity is no deter-
minate quantity; therefore, no actual quantity can be infinite.
We have now seen some reasons for agreeing with Aristotle’s second
premise, since in fact infinity is no number. But it is not that infinity is
completely unrelated to number: to say that some collection is infinite
is surely to comment on its numerousness in some manner, and to say
that a magnitude is infinite is surely to comment on its size. Anything
infinite surely exceeds every finite thing of its kind, even if it is incom-
parable to another infinite. So how does infinity fall short of being a
number?
Here is an explanation: although infinitude is surely a quantitative
property, it is not a determinate quantity. To call something infinite
is to say something about its number or size, but it is not to assign a
specific value to its number or size. Rather, an ascription of infinitude is
essentially negative: it says, roughly, that any assigned number or size
is insufficient. Thus, the infinitude of space consists in the fact that any
assigned size is insufficient to capture the extent of space.
Of course, having thus agreed in part with Aristotle, we shall have to
explain how we avoid his seemingly radical proscriptions on the actual
infinite. But of that, more later.
10.1.5 Is infinity more than three?

The thesis that infinity is not a number naturally invites the speculation
that an infinite thing cannot even be said to be greater than a finite
thing, because ‘greater’ is a relation that only makes sense when applied
to numbers.
The above discussion (Section 10.1.4), however, hints at the reply to
this concern. If infinitude simply had nothing to do with number – if an
ascription of infinitude said nothing about size or numerousness – then
indeed an infinite thing would not thereby be larger than a finite thing.
But that is not the view suggested here. The view is that infinity is not
a specific numerical value (nor is Յ0 or any of the other alleged infinite
numbers). An infinite thing may definitely exceed the finite, while

bearing no determinate quantitative relation to another infinite.
More precisely, the infinite may exceed the finite by having finite parts
that are greater than any chosen finite object. For example, the whole of
physical space exceeds any finite region, because for any finite volume
V, there is a part of space with a volume larger than V. The latter part
need not be infinite; it may simply be a part with a finite volume greater
than V. Because there are always larger and larger finite volumes, we can
say that space is infinite. Again, more of this in a moment.
10.2 Infinite cardinalities
I have suggested, following Aristotle, that infinity is not a determinate

quantity (or, more to the point, that no determinate quantity is infi-
nite). Aristotle inferred from this that there can be no actual infini-
ties but only potential infinities. But we have earlier seen that this
conclusion is untenable; there are many actual examples of infinitude
(Chapter 5). How can we accept the Aristotelian premise while rejecting
the Aristotelian conclusion? Should we say that some actual things are
indeterminate?
In fact, we need not say that, because it is possible to describe what
there is, including the legitimate examples of infinitude, without
mentioning any infinite quantities. There are certain kinds of infinity
that can, so to speak, be constructed entirely out of the finite.
Thus, begin with the most famous example of the infinite, the infini-
tude of the natural numbers. What does it mean to say that there are infi-
nitely many natural numbers? Here is one account: it means that the set
of natural numbers has a certain number of members, and this number
(known as Յ0) is greater than every natural number. That account is defi-
nitely incompatible with the views I have taken in this chapter.
Here is another account: it means that for every natural number n,
there exist more than n natural numbers. For example, 100 is a natural
number. There are more than 100 natural numbers, since there are, for
example, the natural numbers from 1 to 101 inclusive. Another natural
number is 1,000. There are also more than 1,000 natural numbers, since
there are, for example, the ones from 1 to 1,001. Another natural number
is 1,000,000. Again, there are more than 1,000,000 natural numbers,
considering those between 1 and 1,000,001. The same holds for every
natural number you pick.
Notice that nowhere in that explanation did I refer to any non-natural
numbers. At no point did I mention any single number that is greater
Infinity 149
than every natural number. For each natural number, there is one greater
than it; but there is no one number that is greater than all of them.
This account, simple though it seems, is an important advance from
the earlier account referring to the alleged number Յ0, because the present
account enables us to reconcile the Aristotelian premises (no determinate
number is infinite, and all actual things are determinate) with the actual
infinite. Note that on my view, it is not that the natural numbers are
merely potentially infinite. There is no suggestion in the above state-
ments about the natural numbers that, for example, some of them don’t
yet exist and merely have the potential to come into existence. As far as I
am concerned, every natural number actually exists, and for each one of
them, there is one larger than it. This makes the natural numbers both
actual and infinitely numerous. Yet at the same time, we have found no
need to ascribe to any particular thing an infinite number.
In answer to the question, ‘What is the number of all the natural
numbers?’, the answer is that there is no such number. But you might
ask: ‘Doesn’t this still mean there is something – namely, the set of all
the natural numbers – that is indeterminate in a certain respect, since it
has no definite number?’
My reply is that there is no one thing that consists of, or otherwise
includes, all the natural numbers. There are only the individual numbers,
each of them a distinct object – and each of them is perfectly determi-
nate. The objection relies on a set to play the role of ‘a many regarded as
a one’ (to paraphrase Cantor) – the set of all natural numbers would be
the single thing that includes all the natural numbers and that exempli-
fies indeterminacy. We have already seen the reasons to be skeptical of
sets, and there is no apparent way to pose the objection without relying
on something like a set.
But suppose you are uncomfortable with my skepticism about set
theory; you would like to keep sets (or at least remain agnostic about
them). In that case, what should you say? You should say that the set
of all natural numbers simply fails to have a number associated with
it. There is no such thing as the number of elements it contains, since
for every number n, it has a subset with greater than n elements. Does
this make the set an example of indeterminacy? It is not obvious that
it does; if one wants to keep sets, one should deny that this constitutes
an example of indeterminacy. Not every object need have a position on
every dimension. For example, what is the mass of Beethoven’s Ninth
Symphony? It does not have a mass, but this is not an example of meta-
physical indeterminacy. It’s not that the Symphony has indeterminate
mass; mass just fails to apply to it. Similarly, one might say, it isn’t that
the set of natural numbers has an indeterminate number of elements;

number just fails to apply to it. How convincing this is I am not sure.
My account of the infinitude of the natural numbers generalizes
straightforwardly to any case of infinite cardinality. Thus, suppose we
want to discuss the hypothesis that there are infinitely many stars. This
does not mean that there is a certain number which is the number of all
the stars and which is greater than every natural number. What it means
is that for any natural number n, there exist more than n stars. There
needn’t be any one number that applies to all the stars; there just needs
to be, for each n, a number n + 1 that applies to some n + 1 stars.
10.3 Infinite extensive magnitudes
Now consider the second most famous example of the infinite: space.
One way in which space is said to exemplify infinitude is that it is infi-
nitely divisible: there is no limit to how many times a volume of space
could be divided in half. This is just an example of infinite cardinality:
there are infinitely many sub-regions in any given extended spatial
region. This means that for any natural number n, and any region r,
there is a way of dividing r into more than n sub-regions. Notice, inci-
dentally, that the regions I am talking about are not points (nor are
they anything else with a volume of zero). They are all regions with a
positive, real volume. There is no limit to how many regions of nonzero
volume can be found within a given spatial region.
The other way in which space is infinite is in its extent: space stretches
infinitely far in all directions. One way of thinking about this infinitude
is to say there is a certain number, ∞, which represents the total volume
of all space (in, say, liters) and that number is greater than every real
number. This is precisely the picture I have been concerned to reject in
this chapter.
So here is the other way of understanding the infinite extent of space:
it means that for any given volume, there exists a region of space larger
than that volume. For example, 100 liters is a possible volume. And there
are regions of space larger than 100 liters – for example, there are 101-liter
regions. Another volume is 1,000 liters. There are also regions of space
larger than that – for example, 1,001-liter volumes. Another volume is
1,000,000 liters. There are volumes larger than that. And so on.
Notice that, just as in the case of cardinal numbers, we have been able
to explain this infinitude without at any point mentioning any infi-
nite number or magnitude. Only real numbers are used in the preceding
paragraph. Nor do I say that there is a region larger than all the other
Infinity 151
regions; I say only that for any given volume, there is a region larger
than it.
Here one might ask: ‘Yes, but what about space itself – the totality of
all regions. What is its volume?’ As in the case of the set of all natural
numbers, we have two possible replies. On the one hand, we might deny
the existence of the object about which the question is posed. We might
say there is no object that is all of space; there are only particular, finite-
sized regions (but infinitely many of them, to be sure). And there are just
larger and larger regions, without end; there is no Master Region that
includes all the others.
On the other hand, we could allow that the totality of space exists, but
hold that it does not have a volume. As in the case of the set of natural
numbers, the claim would not be that Space is of indeterminate volume.
The claim would be that volume fails to apply to the totality of Space;
volume only applies to particular, limited regions of space.
Again, I find the second reply dubious (if there were a region that
includes all the other regions, it seems as though the property of volume
ought to apply to it). So I prefer the first reply: there is no particular
thing that is Space; there are only finite spatial regions, with no upper
bound to their size.
This view of extensive magnitudes generalizes straightforwardly. What
does it mean to say that the past is infinite? It means that for any chosen
temporal interval, there were times further in the past than that. For
instance, there are times more than 100 years ago, there are times more
than 1,000 years ago, and so on. To say that the future is infinite is to say
that for any chosen temporal distance, there will be times further in the
future than that. To say that the universe has infinite mass (considering
this as an extensive magnitude) is to say that for any given value of mass,
there are aggregates with a mass greater than that value (or perhaps: there
are spatial regions containing more than that amount of mass).
10.4 Infinite intensive magnitudes
10.4.1 Why there cannot be infinite intensive magnitudes

All of this may leave the reader, so far, with the impression that a staunch
rejection of the notion of infinite numbers leaves us – quite surprisingly
and contrary to the Aristotelian tradition – perfectly free to embrace all
manner of actual infinities.
Alas, no. The strategy employed with infinite cardinalities and exten-
sive magnitudes cannot be transferred to intensive magnitudes. The
strategy is to capture infinitude by appealing to cardinal numbers or
magnitudes that increase without bound. It is essential to this view

that there is no one thing that is larger than every finite number, that
there merely are arbitrarily large things – no one cardinal number larger
than every natural number, but cardinal numbers larger than any given
number; no one magnitude larger than every finite magnitude, but
magnitudes larger than any given magnitude.
The strategy fails for intensive magnitudes because there is no infi-
nite series of distinct things with distinct magnitudes. With an inten-
sive magnitude, there is just the single object with its magnitude. For
instance, suppose we wanted to say that some object has an infinite
velocity. This is not a matter of there being ever faster finite groups of
some things. Nor is it a matter of there being some objects with increasing
finite velocities. Rather, we have to say that this one object has a velocity
which exceeds all finite velocities. This requires infinity to function as a
specific quantity – which is precisely the view of infinity I am rejecting.
Therefore, it is impossible for something to have an infinite velocity.
To state the view more generally, we need to make use of the distinc-
tion between natural and artificial magnitudes again. Recall that the
‘schmass’ of an object is defined to be the reciprocal of three minus the
object’s mass:
1
s=
3−m
When an object’s mass is 3, its schmass is thus 1/0, which we might say
is infinity. Well, you might object to saying that, but let’s grant that for
the sake of argument. Schmass also appears to be an intensive magni-
tude according to our earlier definition, since schmasses are not addi-
tive (if A has a schmass of 1 and B has a schmass of 1, the aggregate of
A and B has a schmass of −1, not 2). Now suppose we were to say that
no intensive magnitude – whether artificial or natural – can be infinite.
On the present assumptions, we would then have to say that no object
can have infinite schmass, which in turn means that no object can have
a mass of 3 – clearly absurd. Now, we could try to avoid this result by
saying that when m=3, s is merely undefined, rather than being infinite.
But rather than debating that matter, it is simpler to exclude artificial
magnitudes like schmass from consideration. Who knows what some
artificially constructed ‘magnitude’ might do – and who cares? We have
no need to articulate metaphysical principles concerning that.
Thus, our metaphysical principle will be this: Every natural, intensive
magnitude must be finite. This idea bears emphasis, since it will be the key
Infinity 153
to the resolution of most of the paradoxes of the infinite. Aristotle over-

reached in claiming that there can be no actual infinities whatsoever; he
was thus led into the absurd position of asserting the finitude of space. In
fact, it is only infinite intensive magnitudes that pose a problem, because
it is only these that resist explanation in terms of finite quantities along
the lines suggested in Sections 10.2 and 10.3.
The ‘must’ in my principle indicates metaphysical necessity. It is
not, for example, merely that the laws of nature do not permit infinite
intensive magnitudes. It is that in no possible world, with no possible
laws, could there be infinite (natural) intensive magnitudes, because the
idea of an infinite intensive magnitude requires there to be a quantity
larger than all other possible quantities. In saying that there is no such
quantity, I do not mean merely that there is no thing that instantiates
such a quantity; I mean there isn’t even a possible value for a thing to
instantiate. The way in which an infinite magnitude value fails to exist
is like the way in which the largest prime number fails to exist, or (in
all probability) the first even number that is not the sum of two prime
numbers. This sort of existence is not a contingent matter. If there is no
such number, then there could be no such number.
10.4.2 Constructing infinite intensive magnitudes from

finite magnitudes
But one might think that there is after all a way to express a notion of
infinite intensive magnitude using only concepts of finite magnitudes.
We need only make use of the infinite divisibility of space. Suppose
there is a box full of some material which can be divided into infinitely
many parts, starting from the left side of the box and moving to the
right. The whole box has a volume of one liter. Suppose the first half of
the box has a mass of 1 gram; since its volume is half a liter, the density
is 2 g/L. Suppose the next quarter of the box also has a mass of 1 gram,
and hence a density of 4 g/L. The next eighth of the box contains 1 gram
of material again, for a density of 8 g/L, and so on. In saying this, we
do not ascribe an infinite magnitude to any particular part of the box;
each particular density in the series is finite. But now what is the density
of the whole box? It seems that we would have to say it is infinite. This
seems also to be a natural intensive magnitude. Is this possible?
I want to reject this as a possibility. Part of the motivation for rejecting
this alleged possibility is that it seems to fit in with the kinds of infinity
that generated paradoxes in Chapter 3, and not with the kinds of infinity
that we find in reality (per Chapter 5). It seems that the infinities that
we have antecedent reason to believe in are all of cardinal numbers or
extensive magnitudes, while the infinities that generate problems are all
of intensive magnitudes (of this, more in Chapter 12). If we allow infi-
nite intensive magnitudes as described in the preceding paragraph, it is
unclear what restriction we might place on infinities to avoid paradoxes.
My theory, then, is that infinite natural intensive magnitudes are impos-
sible, including the one described in the preceding paragraph.
But now, what is the conceptual difference between the infinite density
of that example, which I reject, and the infinite extensive magnitudes
and cardinal numbers that I have accepted in Sections 10.2 and 10.3?
My claim is, very roughly, that a natural intensive magnitude cannot be
infinite because there is no such number as infinity (compare this idea:
an object cannot have imaginary mass because there is no square root
of −1). Why does this not also mean that the series of natural numbers
must be finite, and that space must have finite extent, since otherwise
an infinite number would apply to these things?
In the case of the natural number series, I said that one can avoid
positing an infinite number in one of two ways: (a) one can deny that
there is such a thing as the set of all natural numbers, or (b) one can
accept that this set exists but deny that any cardinal number applies to
it. In the case of space, I suggested that one can avoid countenancing an
infinite number by either (a) denying that there is such a thing as the
totality of all space, or (b) holding that this totality lacks a volume. In
each case, my preferred approach would be (a).
The density case differs in that these sorts of reply are simply much
less plausible for the density example than for the natural number or
space examples. If there were a series of spatial regions having the densi-
ties described in the example, it would be very implausible to claim that
there did not exist a region containing all of them – what sort of view
would license that claim? The view that even finite regions don’t exist?
It would also be implausible to hold that the box exists but lacks a
density. This would be implausible, because some physical effects seem-
ingly could be attributed to the box’s density. For instance, if there were
such a box, a spacetime singularity with infinite curvature would seem-
ingly be created (according to general relativity) because of the infinite
density. So it seems as though there would be a real, causally efficacious
property, distinct from any finite density property.
No such problem confronts us with the natural number series. There
is no causally efficacious property that would appear to be the series’
overall cardinal number. Of course, there can very well be causally inert
properties; indeed, every property of an abstract object is causally inert.
So the lack of efficacy does not establish the non-existence of the infinite
Infinity 155
cardinal number. It is merely that the lack of efficacy fails to compel us

to posit such a number.
Nor are we compelled to posit an infinite magnitude in the case of
space. No event appears to be causally influenced by the size of the
totality of space, even if there is such a size – unless of course one counts
some ‘event’ that is spread across the whole of space. But we can reason-
ably deny the existence of such events, without greatly revising our
conception of the world. We may say that there are only events with
finite locations (albeit perhaps infinitely many such events). There is no
aggregate of all the events, and there is no particular object or event that
takes up the whole of space.
In one sense, I am denying the existence of infinite extensive magni-
tudes, just as I deny the existence of infinite (natural) intensive magni-
tudes: that is, if by ‘an infinite extensive magnitude’ one means a
particular magnitude value that exceeds all finite values, then I deny
that there could be any such thing. It is just that, for extensive magni-
tudes – but not for intensive magnitudes – I am prepared to interpret
talk of infinitude as talk about arbitrarily large finite magnitudes.
10.4.3 A putative counterexample: Infinite thermodynamic

temperatures
Temperature is a paradigmatic example of an intensive magnitude. Yet
recent work in physics has demonstrated the possibility of what physicists
call negative temperatures, as well as infinite temperatures.2 The phenom-
enon depends (wouldn’t you know it?) on quantum mechanics – it turns
out, however, that the actual physical phenomenon is not as strange as
the way it is described.
In the kinetic theory of gases, the temperature of a sample of matter is
reducible to the average kinetic energy of the molecules in that sample.3 In
this sense, there are of course no negative temperatures, since this would
require negative kinetic energies. Nor are there any infinite temperatures,
since no sample of material possesses infinite kinetic energy.
In thermodynamics, however, there is another, technical definition
of temperature. Background fact: when energy is added to a physical
system, the entropy (the ‘amount of disorder’) of the system typically
increases. Furthermore, as a general rule, the hotter the system already
is, the smaller will be the increase in entropy due to a small increase in
energy. Relying on this fact, the thermodynamic definition states that
2
Quantum Munich 2014.
3
Giancoli 1988, 425–8.
the temperature of a physical system is the reciprocal of the rate at which

entropy increases with increases in energy, thus:
∂E
T= ,
∂S
where E is the system’s internal energy and S is its entropy.

As noted, typically as a system becomes more energetic, entropy
increases. For most physical systems, this relation continues to hold
at all energy levels (there being infinitely many possible energy
levels). However, there are some special physical systems in which
the particles have a maximum possible energy level. In such systems,
as the system’s internal energy approaches its maximum value (the
number of particles times the maximum energy level of a particle),
the entropy diminishes. In other words, as E approaches its maximum,
∂S/∂E is negative; therefore, ∂E/∂S is also negative. If the system
reaches its maximum energy level, all the particles will be in the
same, maximum energy state, which means there will be no entropy
(the system is perfectly ordered). Figure 10.1 shows the situation:
graph (a) shows entropy as a function of energy level, (b) shows the
rate of change of entropy with increasing energy, and (c) shows the
reciprocal of that rate of change, which is the thermodynamic defini-
tion of temperature.
That, briefly, is the concept of negative temperature. It does not
indicate a situation in which something is colder than absolute zero;
it simply indicates that a certain variable diminishes with increases in
another variable, rather than increasing as is the usual case.
S ∂S/ ∂E ∂E/∂S
E E E
(a) (b) (c)
Figure 10.1 Infinite temperature and negative temperature

Infinity 157
To explain infinite temperature: in these same physical systems, there is

a point at which, with a marginal increase in energy, entropy is unchanged
(this happens when the system is exactly midway between its minimum
and its maximum energy state). That is, ∂S/∂E is zero. That means that
∂E/∂S is 1/0, hence infinity. That’s the infinite temperature case.
Now, does this constitute a counterexample to the metaphysical thesis
that no natural intensive magnitude can be infinite? Not really. Plausibly,
both entropy and energy are natural magnitudes. The rate at which
entropy increases as energy increases might be a natural magnitude. But
it is hard to see the reciprocal of that as a natural magnitude. The funda-
mental situation is that depicted in Figure 10.1(a). Because that graph
goes horizontal at the middle, it is possible to define a ‘magnitude’ that
is infinite at that point – the reciprocal of the slope of the graph. This
would be true for any quantity whose graph is horizontal at any point. For
instance, we could define an object’s ‘inverse-velocity’ as 1 divided by the
rate of change of the object’s position. Then we could say that whenever
an object is at rest, it has an infinite property, namely, its inverse velocity!
So far, I see little reason for thinking that this means some genuine prop-
erty of a physical thing is infinite, in any interesting sense.4
You might say: ‘But surely temperature is a natural magnitude – at
least, we would have said so before hearing about the present cases. If
this turns out to be the best account of temperature, then the reciprocal
of the rate of increase in entropy with increasing internal energy must
be a natural magnitude.’
But there is an obvious reason for thinking that the technical defini-
tion of temperature as ∂E/∂S, (a) does not capture what we ordinarily
call temperature, and (b) does not name a natural magnitude: namely,
that on this definition, so-called ‘negative temperatures’ are hotter than
so-called positive temperatures. Thus, for example, if an object with
‘negative temperature’ comes in contact with an object with positive
temperature, energy flows from the negative-temperature object to
the positive-temperature object. In the ordinary sense, having a higher
temperature means being hotter, yet in the technical, thermodynamic
sense, positive temperatures are colder than negative temperatures.
Another expectation we have about temperature, as we ordinarily
understand it, is that if objects with different initial temperatures are
allowed to reach thermal equilibrium, the result (barring energy loss
4
For further discussion of infinite temperature, see Oppy (2006, 138), who
suggests that because we can modify the temperature scale so as to remove the
infinities, there is no ‘genuinely problematic infinite quantity in the world’.
from the system) should be an intermediate temperature. But when an

object with a temperature of 10K reaches thermal equilibrium with an
otherwise similar object with a temperature of −10K, the result will be
an object with a temperature of ‘∞’, not absolute zero.
The existence of ‘infinite temperatures’, then, is just another conse-
quence of an artificial definition of ‘temperature’. The thermodynamic
definition was adopted because it works well for ordinary cases, where
entropy increases with increasing energy; it just breaks down for the
unusual cases where entropy decreases with increasing energy. A better
conception is that temperature is something like internal energy per
particle.
10.5 Some a priori physics
10.5.1 A constraint on causal laws

The view taken in Section 10.4 leads to substantive restrictions on the
sort of laws of nature that might exist. (I use ‘causal laws’ interchange-
ably with ‘laws of nature’.) The laws can be thought of as determining a
function from possible initial conditions of a system to final conditions
of the system. That is, if you plug a set of initial conditions into the laws,
they tell you what the final conditions of the system will be (or might
be) after some specified interval of time. The following seems to me a
valid constraint on the form the laws might have: No correct set of causal
laws can map a physically possible set of initial conditions to a metaphysically
impossible set of final conditions.5
I am going to simply assume that that principle is correct. Now let’s
put that together with the principle that it is metaphysically impos-
sible for any natural, intensive magnitude to be infinite. This means that
the causal laws cannot permit one to generate infinite natural intensive
magnitudes, starting from physically possible initial conditions.
This is an important, substantive constraint. For example, it entails
the falsity of the general theory of relativity, because General Relativity
predicts that, starting from certain initial conditions (which are physi-
cally possible, both in actual fact and according to the theory), a singu-
larity would be generated with infinite energy density and infinite
spatiotemporal curvature. The initial conditions involve having a star
above a certain mass; this, according to the theory, eventually results in
5
More precisely: there is no metaphysically possible world, w, such that the
causal laws that are true in w map a state of affairs that is physically possible in w
to a state of affairs that is metaphysically impossible.
Infinity 159
the star’s collapse into a black hole. Because it is metaphysically impos-

sible for there to be a region of infinite density or infinite curvature,
General Relativity cannot be a fully accurate account of gravitation and
spacetime structure. The real, perfectly general laws must somehow
prevent such infinities from appearing.
This may seem a rather rash declaration; the history of philosophical
arguments against empirical scientific theories is not encouraging. But
in fact, what I have just said is widely accepted in astrophysics. That is, it
is widely held that the infinities appearing in General Relativity signal a
breakdown of the theory, and that some more advanced theory is needed
to explain what goes on inside a black hole. No one now knows what
that theory is (the most popular current alternatives are string theory
and loop quantum gravity).
10.5.2 A Newtonian illustration

Here is a less exciting example that is easier to think about. Let’s say
we’re living in a Newtonian world. There is a series of infinitely many
objects, O1, O2, and so on. O1 is a rod one meter long, with some finite
mass density. O2 is a rod half a meter long, with the same cross-sectional
area as O1 but twice the mass density (hence, O2 has the same mass as
O1, packed into half the space). O3 has the same cross-sectional area
again, but twice the density and half the length as O2. And so on. At
time t0, these objects are arranged along a single line, with one-meter
gaps between adjacent members of the series. The series of objects thus
stretches infinitely off into the distance, away from O1. Finally, all the
objects are initially moving toward O1. O2 is moving at a speed of 1
meter per second, O3 at 2 m/s, O4 at 3 m/s, and so on (see Figure 10.2).
All of this seems physically and metaphysically possible so far. Notice
that there are no infinite intensive magnitudes in these initial conditions.
Although the densities of the objects in the series grow without bound,
none of them is infinite; they are just larger and larger finite densities.
The objects O1, O2, ... collectively have infinite mass, but thanks to the
way they are spread out along the infinite line stretching away from O1,
4 m/s 3 m/s 2 m/s 1 m/s
• • •
O5 O4 O3 O2 O1
Figure 10.2 An impossible Newtonian scenario

no finite region of space contains an infinite amount of mass; hence,

there is no region of infinite density. Nor need there be any infinite
intensive magnitudes in the history of the system – if we retrodict the
state of the system one second in the past, two seconds in the past, and
so on, all we find is the same objects, spread even further apart.
But what state will the system be in one second after the stated
initial conditions obtain? That is the time at which all of the rods, O1,
O2, and so on, simultaneously collide, thus forming (perhaps only for a
moment) a single rod two meters long. At each instant before this, the
assemblage occupies infinite space; only at t = 1 second does the assem-
blage suddenly occupy a finite region. Within this two meter long
region, what is the mass density? Since each of the rods has the same
mass, and infinitely many of them converge within a finite region,
that region must have infinite mass density at the moment of impact.
If the objects cannot interpenetrate or deform, then each object will
have to experience a force at the moment of impact that decelerates
it instantaneously to velocity zero (hence, an infinite force) – but the
possibility of deformation would allow the forces to be finite and the
decelerations smooth. What will happen next? O1, having been, in
effect, struck by an object with infinite momentum (the fusion of O2,
O3, and so on), will have to move off to the right with infinite velocity
after the collision (perhaps not immediately, but some finite time after
the collision).
This situation, on my view, is metaphysically impossible. My conclu-
sion is that, in any possible world, either the stated initial conditions
are physically impossible, or the laws of nature somehow prevent the
objects from coming together in the manner predicted.
How might this be so? One way is that there might be a physical limit
on how tightly mass can be packed into a region of space – that is, the
laws of nature might determine a maximum possible density of matter.
In that case, there would be some n such that the objects after On in
the envisioned series would be physically impossible. Perhaps different
possible laws of nature provide different limits, so the value of n would
depend on the particular laws.
Another way is that there might be a physical limit on the speed at
which objects can travel. In that case, there would be some n such that the
objects after On could not travel at the speeds stipulated in the scenario.
Since the assemblage of objects initially occupies an infinite span of
space, there would be no time at which all the objects had converged
within a finite region – there would always be infinitely many of the
objects that had yet to arrive.
Infinity 161
A third possibility is that there might be repulsive forces that increase

as material objects approach each other, preventing contact, or perhaps
even preventing them from getting closer than some minimum distance,
thus guaranteeing that the assemblage of objects in the above scenario
must always occupy an infinite amount of space.6 Such forces may also
explain why (as many believe) it is impossible for two solid objects to
overlap in space.
There need not be any one way in which the laws must prohibit the
above scenario; it may be that the scenario is physically impossible for
different reasons in different possible worlds. But every metaphysically
possible set of laws makes the scenario impossible in some way or other.
6
This solution, however, could be circumvented by imagining that the objects
decrease in size by more than half at each stage; for example, instead of the
length of On+1 being 1/2 the length of On, we could stipulate that the length of
On+1 is 1/4 the length of On, thus allowing the objects to fit within a finite region
without coming into contact. To prevent this, the repulsive forces would have
to increase in such a way that the sum of the distances between the Oi remains
divergent.
11
Space
11.1 Pointy space versus gunky space
In standard geometry, space is said to be composed of indivisible parts

called ‘points’. There are continuum many of these in any region with
a nonzero size. All geometric objects – lines, planes, triangles, circles,
and so on – are said to be built out of points. It is usually said that these
geometric objects are sets of points; however, it makes more sense to
regard them as fusions of points, so that is how I shall henceforth speak.
The fusion of a and b is understood as an object that has a and b as parts
and has no other parts that don’t overlap with a or b. (The generalization
to cover fusions of any number of objects should be obvious.) Fusions
differ importantly from sets: for example, the fusion of two physical
objects is itself a physical object, whereas a set of physical objects would
be an abstract, mathematical object. Similarly, the fusion of some spatial
regions would plausibly be a spatial region, whereas the set containing
some spatial regions would not itself be a region but would instead be
some abstract object.
In contrast to the standard view, some theorists have proposed that
space may be a kind of ‘gunk’, that is, something with no smallest parts.1
In this view, instead of thinking of spatial regions as needing to be built
up from smaller components, we start (conceptually) with nonzero
regions. Any region can then be divided into parts, and the parts divided
into more parts, forever, and one never reaches a smallest part.
1
Whitehead 1917, 157–78; Arntzenius 2008. There are at least two other ways
of defining gunk: (1) stuff that has no measure-zero parts; (2) stuff each of whose
parts contains a completely interior region.
162
Space 163
The current hegemony of the ‘pointy’ conception of space over the

‘gunky’ or ‘pointless’ conception is undeserved. Here I suggest some
reasons for taking pointless space seriously. These reasons do not defi-
nitely establish that space is gunky; to repeat, they merely give us reason
for taking that possibility seriously.
11.2 The unimaginability of points
Many of us have gotten so used to hearing and talking about points

that we have forgotten the intrinsic strangeness of the concept. Imagine
taking a sphere and progressively shrinking it. Its volume gets smaller
and smaller: one cubic meter, then one cubic millimeter, then one cubic
micron, and so on. As it gets smaller and smaller, it remains a sphere. But
what happens when you get down to size zero? When you actually reach
zero, is it the case that you now have a really small object there? Or is it
that the sphere is simply gone, leaving nothing behind?
If you think there is a really small object there, try to imagine this
object. I predict that anything that you imagine will not be the geometric
point. You might, for example, imagine a tiny, black dot. But that can’t
be the point; however tiny the dot you imagine, it’s not small enough.
In geometry textbooks, points are represented by small, round dots. But
an actual point is not round; it has no shape at all, because to have a
shape, an object must have some extension. (Shape is a matter of how
an object’s parts relate to each other; for example, in a circle, the parts
are equidistant from a single, central point.) So try to imagine a shape-
less, sizeless, but somehow spatial object.
Now you might say: ‘So points are unimaginable. So what? Many
mathematical objects are unimaginable – that is, they cannot be visual-
ized – but are nonetheless real (in whatever sense mathematical objects
are ever real). So this is no proof that points are not real.’
But rather than asking for proof that there are no points, shouldn’t we
first ask what reason we have for believing that there are such things?
Their unimaginability does not preclude their existence, but doesn’t it
at least indicate that we would need some pretty impressive reason to
believe in them?
Certainly, some mathematical objects are difficult or impossible to
visualize. But this would typically be for quite different reasons from the
reason why points cannot be visualized. Some mathematical objects are
impossible to visualize because they are non-spatial, and visualization
only applies to things in space. Other mathematical objects are unvisu-
alizable because they are too complex to visualize. But this could not be
the case for points, which are supposed to be the simplest of all spatial
objects. Points are in fact supposed to be the basic building blocks of all
other geometrical objects. Is it not strange that such a thing should be
impossible to imagine?
11.3 The zero argument
Imagine pouring some water out of a pitcher. The pitcher starts with one
liter of water, which drops to 0.9 liters, then 0.8 liters, and so on. What
happens when the water gets down to exactly 0 liters; what then is left?
A very small portion of water? Or no water at all?
We all know that the answer is no water at all. We do not postulate
that there might be some residual part of the water, which oddly would
have a volume of zero, clinging to the bottom of the pitcher. This is
true even if one does not know (or one abstracts from one’s knowledge)
of atomic theory – that is, even if we imagine that there is some mate-
rial that is continuously distributed through some volume (rather than
consisting of discrete bits as actual materials do), we still have no temp-
tation to say that some of that stuff might be left in the pitcher, when
the quantity of stuff was zero.
The reason for this is a conceptual point about ‘zero’: to have zero of
something is simply to fail to have that thing. This applies as well to the
real number zero as to the cardinal number zero: to have zero turtles is
to have no turtles; to have zero liters of water is to have no water.
In the case of a material substance, its volume is a measure of how
much space it takes up. When we have one liter of water, that means
we have water occupying a certain amount of space. Volume measure-
ments can also be applied directly to regions of space. When we ascribe a
volume to a region of space, the volume is a measure of how much space
the region consists of.
Now, what is the volume of a geometric point? The standard answer is
‘zero’. It cannot be anything else; if it is anything greater than zero, then
you can divide it in half. Furthermore, if a point has a volume greater
than zero, the total volume arising from infinitely many points would
have to be infinite – but any nonzero region of space is supposed to
contain infinitely many points; thus, every such region would have to be
infinite. This is false, since clearly the apartment I am in is not infinite.
These premises lead to the following argument:
1. The volume of x is how much space x occupies or consists of.

2. To have zero of something is to not have any of that thing.
Space 165
3. Therefore, to have a volume of zero is to not occupy or consist of any

space. (From 1, 2.)
4. If points exist, they are parts of space with zero volume.
5. But there cannot be a part of space that does not consist of any
space.
6. Therefore, there are no points. (From 4, 5.)
If this argument is correct, then it supports a more sweeping conclusion:

there also are no lines, planes, triangles (since these are plane figures),
circles, or any other geometrical object with a volume of zero. Of course,
there are still lines, triangles, and so on, in the ordinary senses of those
words – the senses in which we can say, for example, that Figure 11.1
contains a line and a triangle. But in the idealized, mathematical sense,
that figure contains neither a line nor a triangle, because the alleged line
has some nonzero thickness, as do the sides of the alleged triangle.
11.4 When zero is not mere absence
Does the preceding argument prove too much? There are many cases in
which we apply the number zero but in which this does not appear to
indicate a complete absence. Hence, premise (2) above –
2. To have zero of something is to not have any of that thing.
– appears false. Let us consider a few of these cases.

First, an easy case: the Greenwich Observatory has a latitude of 0º.
But this obviously does not indicate that it fails to have latitude, nor
that it lacks any other relevant quantity. In this case, we simply have
to clarify that premise (2) is intended to apply only to quantitative uses
of ‘zero’, as opposed to mere indexing or naming uses (see Section 9.8).
The use of ‘0’ in reporting the location of the Greenwich Observatory is
an indexing use.
Figure 11.1 A line and a triangle

Second example: a certain ice cube has a temperature of 0 ºC. But this
does not indicate that the ice cube lacks temperature, lacks heat, or lacks
anything else relevant. We might consider this another indexing use of
‘0’. Or we might say that ‘0 ºC’ refers to the value of an artificial magni-
tude, namely, ‘temperature minus 273.15’, as suggested in Section 9.8.
Of course you can always construct an artificial magnitude with a value
of zero, even when there is no relevant absence. We thus need to clarify/
qualify premise 2 by stipulating that premise 2 is to apply only in cases
where zero is a cardinal number or the value of a natural magnitude.
Thus, if an object had a temperature of zero Kelvin, this really would
indicate a complete absence of thermal motion.
Third example: in the mathematical sense of ‘measure’, one can speak
of a measure on the set of real numbers. The natural measure is one
in which, for example, the measure of the set of all real numbers is
infinity, the measure of [3,7] (the set of all real numbers between 3 and
7 inclusive) is 4, and the measure of {2} (the set containing just that one
number) is 0. Intuitively, this means something like: the space taken
up on the real number line by just the number 2 is zero. But {2} is not a
complete absence; it is not like {}.
I think the most plausible response to this example is to observe
that the measure on the set of real numbers does not correspond
to any natural magnitude. Notice that to explain what the putative
magnitude is, I had to invoke a metaphor: that the real numbers form
a ‘line’, and the measure tells us ‘how much space’ a set of numbers
takes up on that line. But there is not literally any line there; the
numbers are completely non-spatial objects. Without that metaphor,
all we have is a mathematical function, satisfying certain formal
rules, that assigns numbers to subsets of the set of real numbers. That
does not suffice to have a genuine magnitude. Notice also that this
alleged magnitude has no causal efficacy; nothing can be causally
influenced by some set of numbers’ having some particular measure,
in the way that things certainly can be causally influenced by some
concrete objects’ having some magnitudes. Finally, notice that there
appears to be no way of explaining this alleged magnitude without
reference to sets; it could only be a property of a set. This is problem-
atic not only because there is some reason to doubt the existence of
sets, but also because, even if sets exist, their properties, again, have
no causal powers. This contrasts with the cases that I would accept
as genuine natural magnitudes, such as volume or mass, where there
is some concrete object (or at least a region of space) to which the
magnitude can be ascribed.
Space 167
Fourth example: in probability theory, some events are assigned a prob-

ability of zero, even though they are possible. This normally happens
when there is a continuum of possibilities. For example, suppose a
person could have any of continuum many heights within a certain
range. Any given exact height will have probability zero of being realized
(this must be so, otherwise the total probability will exceed 1).2 But does
this ‘zero’ indicate a complete absence of something?
To address this, we must first ask what probability numbers measure.
First answer: they measure probability. This is unhelpful, since without
further explanation it is hard to say whether the events in question have
a complete absence of probability or not.
Second answer: they measure possibility; therefore, because the events
in question are possible, the zero cannot indicate complete absence. But
this answer is mistaken: if probability measured possibility, then every-
thing would have a probability of either one or zero, since everything is
either possible or impossible and there are no degrees in between.
Third answer: probability numbers measure the strength of evidence
or justification for believing something. On this interpretation, the use
of ‘0’ in probability assessments is an indexing use rather than a numer-
ical use. We arbitrarily assign the bottom of the scale to be the number
zero; we could as easily have a probability scale ranging from −1 to 1,
or from 1 to 100. (I do not mean that these choices would be equally
convenient, of course – only that they would be no less adequate to
accurately representing reality.) Even in the case where A is logically
impossible, the zero probability does not indicate anything like a mere
absence of justification; rather, it indicates (on the present interpreta-
tion of probability) the strongest possible justification for denying A.
Fourth answer: probability numbers are limiting long-range frequen-
cies.3 In other words: suppose we want to know the probability of A
occurring in conditions C. We imagine that C were repeated over and
over. At any point, there would be a ratio of the number of times that
A has occurred to the number of times that C was repeated. If this ratio
would approach some limit as the number of repetitions of C increased
without bound, then that limit is the probability of A in conditions
C. In the case discussed above, the zero probability means something
2
One could assign nonzero probabilities to some countable subset of heights
and then assign zero probability to the rest of the heights, but this violates some
implicit constraint against arbitrarily treating certain possibilities in a radically
different way from the nearby possibilities.
3
von Mises 1957.
like this: if we measured many individuals’ heights, the frequency with

which those heights turned out to be exactly six feet would approach
zero as the number of people measured increased without bound. (I
think this is true; in fact, I think the frequency would be zero at every
stage.) Does this ‘zero’ indicate a complete absence of something? If the
ratio is actually zero, then this indeed indicates a complete absence of
occurrences of A. On the other hand, if the ratio is nonzero but is merely
approaching zero as a limit, this does not seem to indicate a complete
absence; however, in this case zero is also not the actual value of any
natural magnitude. The limit approached by a certain quotient in a
counterfactual situation does not appear to be a property of anything,
nor could it be causally efficacious. The only natural quantities in the
story appear to be the actual number of times C is repeated and the
actual number of times A occurred. If either of these were zero, that
would indicate a complete absence.
11.5 The paradox of contact
The most popular objection to pointy space is not anything I have

mentioned so far. The most popular objection is that pointy space makes
it difficult to understand physical contact, which is assumed to be meta-
physically possible.4
Imagine a world containing extended material objects, each of which
completely fills the (completely connected) region of space that it occu-
pies. (For simplicity, assume there aren’t any objects with holes in them.)
Three assumptions create trouble for understanding physical contact.
The first assumption is that it is metaphysically possible for some such
objects as I have described to exist in a pointy space. Second, no two of
these material objects could occupy any of the same space at the same
time (that is, they could not even partially overlap). Third, two objects
are in contact only if there is no space at all between them.
If space is pointy, then there would be a distinction between ‘open’
objects and ‘closed’ objects. A closed object would be one that includes
an outer surface, whereas an open object occupies all the points inside a
certain surface, but does not occupy the surface itself. (The distinction is
analogous to the distinction between (a) the set of real numbers greater
than zero and less than one and (b) the set of real numbers greater than
4
The argument of this section is based on Zimmerman 1996, though
Zimmerman’s conclusion concerns extended objects rather than space itself. Cf.
Brentano 1988, 146–7.
Space 169
or equal to zero and less than or equal to one. (a) is an open set, whereas
(b) is a closed set. Similarly, a closed object occupies its boundary.)
Now suppose two open objects wanted to come into contact. It would
be impossible for them to do so, because there would always have to be
at least one unoccupied point of space between them. When the objects
were as close as they could be, the two objects’ surfaces (that is, the
surfaces that surround the objects but which the objects themselves fail
to occupy) would coincide at (at least) one point. That would be the
unoccupied point separating the two objects.
Suppose that two closed objects wanted to come into contact. Again,
they could not do so, because if ever there were no empty space between
the two objects, their surfaces would have to coincide at at least one
point, in which case both objects would be occupying that point in
space, and we have already assumed that objects cannot even partially
overlap in space. The only way for the objects to fail to overlap in space
is for them to have a nonzero gap between them. So that is what would
have to happen. If the two objects moved toward each other, some force
would have to appear to stop them before they made contact.
Perhaps the only way for two objects to come in contact would be for
an open object to come into contact with a closed object, or more gener-
ally (since an object could be partially open) for an object that is open
at some point to come into contact with an object that is closed at some
point, where those points might come to coincide.
What seems wrong about this is, first of all, the notion that two objects
of the same kind could not come into contact. Of course, there might
be (as in the actual world) causal laws that prevent two objects from
coming into contact. What seems wrong is that it should be metaphysi-
cally impossible for any pair of qualitatively identical objects to come
into contact, given just the structure of space – for example, that there
might be two spheres that are metaphysically debarred from touching
unless one of them either (very slightly) expands or contracts.
Second, the action at a distance in the case of the closed objects is
suspicious. Assuming that it is metaphysically impossible for two mate-
rial objects to spatially overlap, this is what would have to be true of the
pair of closed objects: as they moved toward each other, it would have to
be that, while there was still a gap between them, some force appeared
to keep them apart. This must happen because if the objects wait until
the gap between them is closed, they would already overlap in space (at
least at one point).
Here, by the way, is how things work in the actual world: material
objects in fact never come into contact at the microscopic level. When
you put your apple on the table, the molecules in the apple merely come
close to the molecules in the tabletop, whereupon repulsive forces hold
them apart, thus preventing the apple from falling through the table.
So there’s nothing wrong with positing force fields that hold physical
objects forever at a distance from each other. The idea of the objection
is merely that this should not be metaphysically necessary. That is, it
is strange that there should be a pair of objects that are metaphysically
required to exert forces on each other at a distance, merely in virtue of
their shapes; and yet, another pair of objects that was in every way iden-
tical to the first pair, save that they were each missing a part with zero
thickness, would not be required to exert forces at a distance.
I find the preceding argument somewhat less persuasive than do most
advocates of gunky space. Why? To begin with, I am not sure that it
is metaphysically impossible for two material objects to interpenetrate.
Either there is some limit to the mass density that may exist in a given
region of space, or there is no such limit. First, suppose there is no limit.
On that assumption, I think it plausible that two material bodies can
interpenetrate. If the two objects each have a uniform mass density
d, interpenetration would merely result in a region of overlap having
mass density 2d; this region would be just like an ordinary region that
contains a single object with mass density 2d, so I don’t see why this
should not be possible.
On the other hand, suppose there is a maximum possible mass density.
In that case, we can readily see why two material objects might be inca-
pable of interpenetration, say, because their mass densities sum to more
than the maximum possible density. On this assumption, however, I
find it not so implausible that material objects must exert forces on each
other at a distance. (Note that this would seem to be required to avoid
infinite decelerations at the moment of impact, in the event of a colli-
sion of the two objects.)
11.6 The problem of division
Finally, the pointy conception of space entails that, if an object

completely fills a connected region of space, then it is metaphysically
impossible to divide that object exactly in half.5 If one cuts a stick in
half, the points at the middle of the stick – the cross sectional slice
5
This objection is colorfully presented by Brentano (1988, 146–7), who, among
other things, objects to Bolzano’s ‘monstrous doctrine that there would exist
bodies with and without surfaces’.
Space 171
Two possibilities
Figure 11.2 Two ways of dividing a stick in half
a single point thick – must either go with the left half of the stick or
go with the right half. Thus, the result will be that one half of the
stick winds up containing a closed end while the other winds up with
an open end (Figure 11.2). The latter (almost-)half of the stick will
be like the former except with an infinitely thin piece of its surface
stripped off. There is nothing formally inconsistent in this picture.
It is just very bizarre to think that this is how physical reality would
work. Doesn’t all this rather have the feel of a series of artifacts of a
human-created conceptual scheme, as opposed to insights into objec-
tive reality?
11.7 The dimensionality of space is necessary
The preceding arguments provide reason to think that space is neces-

sarily gunky, rather than pointy. What they suggest is that the notion
of geometric points is conceptually confused. These arguments do not
definitely establish this conclusion; they merely provide some reason
to suspect that this is the case. One could consistently maintain that
points exist despite being unimaginable, that having zero of something
is different from failing to have any of it, that contact between type-iden-
tical solid objects is impossible, that dividing a solid object into identical
halves is metaphysically impossible. But I think that our ending up with
such views is a sign of some foundational, conceptual error.
There is, incidentally, nothing especially odd about the situation
of having inconclusive reasons to believe that something is necessarily
true. For instance, we have strong but inconclusive grounds to believe
Goldbach’s Conjecture (that every even number is the sum of two prime
numbers), because many numbers have been checked and no counterex-
ample has been found. But if the conjecture is true, it is metaphysically
necessary. For a simpler example, suppose a smart person tells me that
A is necessarily true. Then I have evidence that A is necessary, but the
evidence is inconclusive.
Suppose that the above arguments are correct. They suggest that it is
impossible for there to be a part of space with a measure of zero. Not
only are points impossible, then; lines and planes are impossible. But
then it follows that space is necessarily at least three-dimensional. Since,
for example, a plane (metaphysically) cannot exist, it is metaphysically
impossible that all of space should consist of a plane. Similarly, space
could not have consisted of a single line.
Could space have had four or more dimensions? No ... not unless it
actually does. If there is a possible world in which space has four dimen-
sions, then from the standpoint of that world, it would be true that
three-dimensional regions were impossible, since they would have
measure zero. But, assuming that our space actually has three dimen-
sions, three-dimensional objects are possible. So the putative four-di-
mensional possible world must not exist. In other words, the people
in the four-dimensional world could give arguments analogous to the
above arguments against points, lines, and planes, to show that our
world is impossible. But since we know our world is possible, there must
not be any such possible people.
You might worry that the above argument turns on taking ‘possible
worlds’ too seriously, treating them as if they were things that really
existed. The argument can be rephrased without the possible-worlds
talk:
1. For any propositions A and B, if A is true and B is metaphysically

possible, then if B were the case, A would (still) be metaphysically
possible.
2. It’s true that space is three-dimensional.
3. So if it is metaphysically possible that space be four-dimensional,
then if space were four-dimensional, it would (still) be metaphysi-
cally possible that space be three-dimensional. (From 1, 2.)
4. But it is not the case that if space were four-dimensional, then it
would be possible that space be three-dimensional.
5. So it is metaphysically impossible that space be four-dimensional.
(From 3, 4.)
But what about contemporary physical theories that postulate many

dimensions? For instance, recent versions of string theory postulate
nine or more spatial dimensions. The above argument does not refute
this, since the argument merely assumes that space is actually three-
dimensional. The argument shows only that the dimensionality of
space – whatever it is – is necessary, not contingent. That is, if space
Space 173
actually has three dimensions, then it is metaphysically necessary that it

does. If space actually has nine dimensions, then that is metaphysically
necessary.
This view is not popular among philosophers, who tend to assume
that n-dimensional space is possible, for any n. I think there are two
common but bad reasons for thinking this. The first is that one can
formulate logically consistent descriptions of many-dimensional spaces.
Even though we can’t imagine these spaces, we know how to describe
them without contradicting ourselves. But as discussed earlier, mere
formal consistency is no evidence of metaphysical possibility.
The second reason is a superficial analogy, which starts from the
assumption that there could be two-dimensional beings, and extrapo-
lates from there to the idea that there could be four-dimensional beings
who would be related to us as we are related to the possible two-dimen-
sional beings. The starting assumption is false, and the extrapolation is
uncompelling.
11.8 The measure-theoretic objection
There is at least one interesting technical objection to pointless space.6

It starts with a variation of a construction known as the Cantor Set. The
Cantor Set is constructed like this: start with a line one unit long. Remove
a segment with a length of L1 (where L1 is more than zero but less than
the whole segment) from the middle of the line. This leaves two line
segments, one to the left and one to the right. Next, from the middle of
each of these two remaining segments, remove a segment of length L2
(where L2 is more than zero but less than the whole segment). This leaves
four line segments. From the middle of each of these, remove a segment
of length L3 (more than zero but less than the whole segment). And so
on. Imagine this process repeated infinitely many times. What remains,
in the standard view, is a set of points known as the Cantor Set. (In the
best-known version, Ln is always one third of the size of the segments
existing at the beginning of each stage.) The segments that were removed
each had a positive, real-valued length. I shall call these segments the
Cantor Lines. So the Cantor Set is what is left after removing the Cantor
Lines (the Set is the complement of the Lines within the unit interval).
Now, consider this version of the construction: in the first stage, we
remove one segment with a length of 1/4. In the second stage, we remove
two segments, each with length 1/16. In the third stage, we remove four
6
The argument is based on Arntzenius 2008, Section 10.
Figure 11.3 The Cantor lines
segments, each with length 1/64. And so on. In other words, assume in
general that Ln = 1/4n. Figure 11.3 shows the results of the first three cuts.
What is the total length of all the removed lines? This is the infinite sum
of the lengths removed at each stage:
1 ⎛ 1 ⎞ ⎛ 1 ⎞
+ 2⎜ ⎟ + 4⎜ ⎟ + "
4 ⎝ 16 ⎠ ⎝ 64 ⎠
1 1 1
= + + + "
4 8 16
1
= .
2
So a total length of ½ is removed from the original segment, which itself

had a length of 1 unit. So whatever remains must also have a measure
of ½. Now comes the problem: on the pointless view of space, it looks
as though nothing remains. This is because the Cantor Set contains no
extended segments (no connected parts with nonzero length). In the
standard view, the Cantor Set consists entirely of disconnected points.
(For any two points in the Set, there was some stage at which a segment
in between them was removed. In fact, there were infinitely many such
stages.) But since we’ve rejected points, it seems that we must say the
Cantor Set does not exist, and thus that the Cantor Lines must comprise
the entire original line segment. Yet the original line segment has a
measure of 1, while the Cantor Lines only add up to ½.
Now to clean up the exposition a bit. First point: the talk of ‘removing’
segments is just there to make the exposition easier. Since we’re talking
about regions of space, it is not possible to literally remove any of them.
The point is just to focus your attention on certain parts of a line
segment. You can think of all the parts that I previously described as
being ‘removed’, just sitting there in the segment. We can then think
Space 175
about the fusion of all these parts, and we can ask questions about it, such
as how big it is and whether it is distinct from the original segment.
Second point: if one doesn’t believe in geometric points, then presum-
ably one doesn’t believe in geometric (infinitely thin) lines either. So
one might want to reject the whole construction, seeing as how it starts
out by assuming that we have a line segment. But the use of a line was
also just for expository purposes. There is a three-dimensional analog
of the Cantor Set construction, but it is more complicated to imagine.
It would give us qualitatively the same result. Alternately, instead of a
line in the geometric sense, we could imagine an elongated cylinder,
where we cut the middle quarter out of it, and so on. The volume of the
removed parts will sum to one half of the original volume. So the talk of
a ‘line’ is harmless in this context; it isn’t the source of the problem.
My preferred approach to this problem may not seem persuasive at
this point, but I will tell it to you anyway. It is to reject the notion of
an infinite sum. The problem is generated by taking an infinite series
of line segments, adding together their lengths, and then comparing
the result to the original, one-unit-long segment. This presupposes that
there is such a thing as the sum of an infinite series of numbers. In my
view, there is no such thing. However, I say more about that in the next
chapter (Section 12.3), because this bears directly on the resolution of
one of our earlier paradoxes of the infinite.
12
Some Paradoxes Mostly Resolved
This chapter gives the payoff for our theory of the infinite: the reso-
lution of the paradoxes of Chapter 3 (at last!). I assume here that the
reader has read Chapter 3, and thus that only brief reminders of each of
the paradoxes are required.
12.1 The arithmetic of infinity
Here, the puzzle was that, for example, from ‘∞+1 = ∞’, one can seem-
ingly derive that 1=0; from ‘1/0 = ∞’ or ‘1/∞ = 0’ one can seemingly
derive that 1=0; and from ‘1/∞ = 0’ and ‘2/∞ = 0’ one can seemingly
derive that 1=2.
The resolution of these puzzles is straightforward on my view. The
expression ‘∞+1’ makes no sense because addition is an operation on
numbers, and ‘∞’ does not name any number; ‘∞+1’ is like the expres-
sion ‘green + 7’. Consequently, the pseudo-equation ‘∞+1 = ∞’ is a cate-
gory mistake. Similarly for ‘1/0 = ∞’, ‘1/∞ = 0’, and ‘2/∞ = 0’.
The Cantorian approach, by contrast, holds that there are infinite numbers,
but they simply obey different rules from all the other numbers. This is not
formally inconsistent. However, the theory that infinity violates the rules
that apply to numbers because infinity is in fact not a kind of number, is
more explanatory than the theory that infinity is a kind of number that
simply violates the rules that apply to all other numbers. Intuitively, it
should be possible to add to a quantity, and doing so should increase it.
Intuitively, if a number can be added, then it can also be subtracted.
12.2 The paradox of geometric points
Here, the problem was that an extended region of space with some posi-
tive volume is supposed to be composed entirely out of parts (points)
each of which has zero volume.
176
Some Paradoxes Mostly Resolved 177
The standard treatment of this puzzle: reject additivity of sizes for

uncountable collections. On the standard view, when there are uncount-
ably many parts of something, the size of the whole cannot be computed
from the size and number of its basic parts, even though such parts
exist and even though they (each) have a size and (collectively) have a
number. The size of the whole is just a further property of the object. If
a region is really built up from points, how is it that its magnitude is not
explained by their magnitudes and number? There is no real explanation
for this, other than that this is what we must say to avoid paradox.
Note that the traditional view accepts that there are infinite sums of
Յ0 elements (at least in a manner of speaking – see Section 12.3 below);
there just aren’t any infinite sums of continuum many or more elem-
ents. In the standard treatment, regions of space satisfy countable addi-
tivity (one can add a finite or countably infinite number of parts to find
the size of the whole that they compose), yet there is no analogous prin-
ciple of uncountable additivity. This is metaphysically odd.
We also noted that if an object were to fill a continuous volume of
space, it would seemingly be conceptually possible, on the standard
view, for the point-sized parts to be rearranged to completely fill a space
exactly twice as large (mapping each part with coordinates (x, y, z) to
coordinates (2x, y, z)). A proponent of the standard view would most
likely respond to this by simply ‘biting the bullet’ – that is, just embracing
the conclusion that the rearrangement in question is conceptually and
metaphysically possible (though not physically possible in our world).
One might think: ‘This is not so metaphysically strange. Perhaps
we start with an object occupying a certain region of space with mass
density d, and when we expand it to twice its original size, the result is
simply that the new region has mass density d/2, so that total mass is
conserved.’ However, here is what is odd. Suppose the object is expanded
horizontally, to twice its original width, as in Figure 12.1.
Consider the left half of the stretched object. This half would seem
to be qualitatively identical to the whole of the original object, since it
consists entirely of qualitatively identical parts, identically arranged.
Similarly for the right half. It is therefore unclear why we have not just
Figure 12.1 A magically expanding rod

duplicated the original object, in what should be an impossible manner,

merely by moving its parts around.
The simplest way to avoid these paradoxes is to reject the existence
of size-zero parts of an object. The size of an extended region is not the
sum of the sizes of the points that make it up, because there in fact are
no such things as points; there are only positive-sized regions.
We could say that, even if there are points, there is no such thing as
the sum of infinitely many sizes, because there are no infinite sums in
general. I take up this view in the following section. However, this view
does not avoid the paradox in which an object is duplicated by rear-
ranging its parts. To avoid this result, we should reject the existence of
size-zero parts.1
12.3 Infinite sums
Here, the puzzle was that the same (infinite collection of) numbers can
seemingly have different sums, depending on the order in which the
numbers are added. If we add together all the positive and negative inte-
gers, we can obtain 0 if we add them like this: (1−1) + (2−2) + ... . But we
can obtain ∞ if we add them like this: 1 + (−1+2) + (−2+3) + ... .
The standard view: one can (sort of) talk about a sum of infinitely
many numbers, but the correct sum sometimes depends upon the order
in which the numbers are added. An infinite sum must be a sum of
a sequence of numbers, not of a mere set of numbers. The two above-
described ways of adding the integers represent two different sequences,
and therefore there is no problem in the fact that they yield different
sums, even though they are built from the identical set of numbers.
The sum of an infinite series of numbers is defined to be the limit
(if there is one) of the sequence of finite sums as the number of terms
increases; if there is no limit, then the series has no sum. In other words,
as we add more and more terms, if the sum approaches arbitrarily close
1
One could avoid the rearrangement paradox while accepting the existence of
some size-zero parts, as long as the stick was not completely decomposable into
size-zero parts (for example, what if the stick contained just three geometric points,
but was otherwise gunky?). However, I think there is no well-motivated view with
this consequence. I mention this in a footnote, only for the benefit of readers who
like to look for any annoying exception to any statement found in the text.
One could also avoid the paradox by claiming that, although space consists of
points, an extended object occupying space does not and cannot consist of point-
like parts, that is, it does not have parts corresponding to each part of space that
it occupies. This would be odd and ad hoc.
to some particular value, then that value is regarded as the sum of the
infinite series. Otherwise, the series has no sum. For example, the infinite
1 1 1
sum, + + + ", is said to be 1, because as you add up the terms, you
2 4 8
get ½, then ¾, then ⅞, and so on, getting closer and closer to 1.2 You can
find this sort of definition in any standard calculus textbook.
Notice something crucial about this way of defining an infinite sum:
it has the consequence that ‘sum’ has a different meaning when speaking
of an infinite sum than it does when speaking of a finite sum. When you
have a finite collection of numbers, their sum is the result you end up
with after adding up all the numbers. But the sum of an infinite series
is decidedly not understood as the result of adding up all the terms. The
standard definition is deliberately and explicitly formulated to avoid the
assumption that one can actually add up all the numbers in the series.
That is the whole purpose of the talk of limits.
In fact, the standard definition seems to presuppose that one cannot
add up all the terms of an infinite series. To see why I say this, note
first that the limit of a function as the independent variable approaches
some value need not be the actual output of the function at that value
(even if the function is defined there) – in mathematical symbolism,
Lim f x)x need not = f(a). To illustrate, recall the Littlewood-Ross Banker
x a
Paradox (Section 3.12). In that scenario, you gain $9 in each round of
the game, so the limit of your funds as the number of rounds of the game
increases is $∞. However, if you actually complete the series, the result is
$0, not $∞. To put the point in mathematical symbolism: where f(n) is
the amount of money you have after n rounds of the game, Lim f n ) = ∞,
n →∞
but f(∞) = 0.
Now, if there were such a thing as the result of actually adding up infi-
nitely many numbers, presumably that result should be called the infi-
nite sum. And there is no reason to assume that that result must always
equal the limit of the finite sums. So by defining the infinite sum as the
limit of the finite sums, mathematicians are presupposing that there is
no such thing as the result of adding up all the infinitely many terms.
In this, I agree with the standard view (at last!). Aside: there are many
standard views in mathematics, which sometimes derive from the work of
different mathematicians with very different views of infinity. Thus, the
standard views in set theory and transfinite mathematics, under Cantor’s
influence, are very friendly to actual infinities and infinite numbers on
2
More precisely, where ‘Sn’ denotes the sum of the first n terms: for any ε > 0,
there is a number δ such that |Sn − 1| < ε for all n > δ.
all fours with the natural numbers. But the standard views in calculus,
under the influence of Karl Weierstrass, are very unfriendly to infinity
as a number or as an actual magnitude. (Robinson’s aptly named ‘non-
standard analysis’ offers a much more infinity-friendly approach, which
may fit better with Cantor.)
As I say, I agree in essence with the standard view of infinite sums,
although I find the terminology misleading. It is misleading to say
that there is a sum of an infinite series, if what you mean is actually
that it is impossible to add up infinitely many terms, but that if you
add up increasing finite numbers of terms, you approach ever closer to
some value. Many students wind up confusedly thinking that standard
analysis supports the exact opposite of the point that it is designed to
support – witness the view that the theory of infinite sums resolves
Zeno’s Paradox. This could only be the case if the theory of infinite sums
somehow shows how one can complete the infinite series, which is the
precise opposite of what the standard account assumes. I shall hereafter
use scare quotes when speaking of the sort of ‘infinite sums’ defined in
terms of limits, to avoid giving the impression that ‘infinite sums’ are
really sums in the same sense as finite sums.
Why aren’t there any genuine infinite sums, in the sense of ‘sum’ that
applies to finite cases? So far, I have put the point by saying that one
cannot add up infinitely many numbers. This is true but is not really
the point – after all, a pair of numbers may have a sum even if no one
ever computes that sum. The real thesis is this: as a conceptual matter,
addition is an operation defined on pairs of numbers. From this basic
understanding, one can define a notion of the sum of three numbers
(‘a + b + c’ refers to sum of [the sum of a and b] and c), the sum of four
numbers, and so on. But one cannot thus define the sum of infinitely
many numbers. One can define an ‘infinite sum’ using limits, but then
one is changing the meaning of ‘sum’.
This explains why ‘infinite sums’ may violate the rules of ordinary
addition, specifically, the commutative and associative laws (or rather,
infinitary versions of those laws). If they were genuine sums, then this
would be paradoxical, since the commutative and associative laws are
self-evident truths (see Section 9.4 above). ‘Infinite sums’ violate the
rules for addition because they are not actually sums.
12.4 Galileo’s paradox
My solution to Galileo’s paradox was given in Chapter 10 (Section 10.1.3):

since the greater-than, less-than, and equal-to relations are relations
between numbers, and there are no infinite numbers, these relations do

not apply to the infinite. Thus, in answer to the question ‘Which are more
numerous: the perfect squares or the natural numbers?’, we can only
answer that both are beyond number; hence, neither collection can prop-
erly be said to be greater than the other, nor can they be called equal.
This thesis implies a rejection of the conclusion that is usually said to
be established by Cantor’s famous Diagonalization Argument (Section
6.6): namely, that the set of real numbers is larger than the set of natural
numbers. These sets, too, (if sets exist) are incomparable since both are
infinite. The Diagonalization Argument indeed establishes that there is
no one-to-one function from the natural numbers onto the real numbers,
which is an interesting discovery. But it is not the same as the conclu-
sion that the real numbers are more numerous than (that is, instantiate
a larger number than) the natural numbers; to draw this latter conclu-
sion, we must also make the assumption that the greater-than relation is
identical to the relation that a set bears to another set when there is no
one-to-one function between the two, but there is a one-to-one function
from a proper subset of the former onto the latter.
Now one might think: ‘But the primary objection to relying exclusively
on the one-to-one function criterion is that there is a second criterion
of greaterness – namely, the proper subset criterion (every set is greater
than any proper subset of itself) – which is equally important. But in this
case, the two criteria are in agreement. Why, then, may we not conclude
that in such a case the one set is really greater than the other?’
The answer is that the best explanation for the fact that the two
‘criteria’ for greaterness often diverge in the case of infinite collections
is that infinity is not a kind of number. This explanation in turn implies
that no infinite collection stands in the greater-than relation. I enclose
‘criteria’ in scare quotes in this paragraph because the two principles I
mentioned earlier –
1. A number is increased when one adds to it.

2. If the members of some collection can be paired up with the members
of another collection, then neither collection is larger than the
other.
– do not (either singly or in conjunction) constitute a definition of

‘greater than’ but simply are two self-evident principles about the great-
er-than relation. There are other self-evident principles about the great-
er-than relation, such as that it is transitive and asymmetric, which we
would not be tempted to call ‘criteria’ of greaterness; this is just how we
should think about principles (1) and (2). With that understood, the
mere fact that a particular attribution of the greater-than relation fails
to violate either of these principles does not show that the attribution
is true.
My treatment of Galileo’s Paradox is superior to the Cantorian treat-
ment, because my treatment respects the intuitions on which the
paradox rests, rather than simply ignoring one of these intuitions.
12.5 Hilbert’s hotel
Begin with the first version of the story, in which the completely occu-
pied hotel with infinitely many rooms is able to accommodate one new
guest by moving everyone to the next room down. This could not really
be done. Since the hotel has infinitely many rooms, these rooms must
be located farther and farther away from the front desk, with no limit
to their distance. When the new guest shows up at the front desk, he
may obtain room #1 right away. But the guest from room #1 will then
be temporarily unaccommodated as he is moving to room #2. The guest
from #2 will then be out of a room temporarily, as he relocates to room
#3. And so on. As the infinitely long relocation task proceeds, there will
always be one guest temporarily out of a room. A similar point applies
to the later version in which the hotel tries to accommodate infinitely
many new guests.
But you might think: ‘Why suppose that the hotel guests relocate
in sequence, with the guest from room #2 moving after the guest from
room #1 arrives at room #2, and so on? Why not suppose that the hotel
manager notifies all the guests of the new arrangement, using his PA
system which communicates with all the guests simultaneously? Then
the rearrangement can be completed in a finite time.’
The answer is that there cannot be such a PA system, since simulta-
neous communication with all of the guests would require a signal that
travels at infinite velocity. Velocity is a natural intensive magnitude, and
no such magnitude can be infinite. There is no metaphysical limit to
how fast a signal might travel, but it must have some finite velocity.
Given that the hotel extends an infinite distance from the front desk,
the signal cannot reach all of the infinitely many guests in any finite
time. The signal can continue traveling indefinitely, so that it reaches
each guest at some time, but it will never have reached all the guests.
But now suppose that no communication was sent out. Instead, each
of the guests, completely on his own, just decides to move to the next
room down. Coincidentally, they all decide this at the same time. This
is metaphysically possible, even though the probability of such a coin-

cidence happening is zero. The new guest then moves into room #1. As
far as I can tell, this version of the scenario really does offer a metaphysi-
cally possible way for the hotel to accommodate a new guest. Strange
but true.
Part of the intuitive resistance to this scenario may stem from the sense
that one is getting something for nothing, in a way that shouldn’t be
possible: we get one more person accommodated, but without creating
any new rooms and without the existing guests having to give anything
up. See if this reflection reduces your intuitive resistance: something is
in fact given up to obtain the benefit for the new customer. Each of the
previously existing guests must give up a certain amount of in-room time,
namely, the amount of time required for them to move from one room
to the next. Because there are infinitely many guests, the total amount
of room time given up is infinite. The way is then paved for the new
guest to enjoy as much room time as he wishes, but the benefit gained
in room-hours will never exceed the infinite cost in room-hours.
12.6 Gabriel’s horn
Gabriel’s Horn has a finite volume but an infinite surface area. How
much paint would be required to paint it?
This puzzle does not require any special views about the infinite for its
resolution. The reason why, in the real world, one cannot paint an infin-
itely large surface with a finite volume of paint is that, in the real world,
there is a limit to how thin a coat of paint can be (for example, it cannot
be less than one molecule thick). If, therefore, one divides a given (finite)
volume of paint by this minimum thickness, one obtains the maximum
surface area that that volume of paint can possibly cover, which must be
a finite magnitude. If we assume that this minimum-thickness constraint
applies to the hypothetical painting of Gabriel’s Horn, then the Horn
could not in fact be painted with a finite amount of paint. If one tried
pouring the paint into the Horn (even given a volume of paint no less
than the Horn’s total volume), the paint would only extend so far: the
Horn gets thinner and thinner, approaching a diameter of zero as a limit.
So at the point at which the diameter was less than the minimum thick-
ness for a coat of paint, the paint would be unable to move any further
into the Horn (for example, because the diameter of the Horn at that
point was less than the size of a paint molecule).
Suppose, on the other hand, that we assume there is no limit to how
thin a coat of paint may be. In that case, there is no problem: an infinite
surface can be covered by a finite volume of paint. The coat of paint just
has to get thinner and thinner, approaching a thickness of zero.
12.7 Smullyan’s infinite rod
In this case, the infinitely long, horizontal rod remains parallel to the
plane, despite the fact that it is supported at only one end. The rod
experiences an infinite torque (a force tending to cause it to rotate), and
there is no counterbalancing torque, since nothing is actually pushing
on the rod at any point other than the one endpoint. Yet the rod fails
to rotate.
The stipulations of this scenario are, however, impossible. The finite
vertical rod, the infinite horizontal rod, and the plane are all stipulated
to be absolutely unbendable, unbreakable, and impenetrable, no matter
the force applied to them. That is, all three are composed of material
with infinite yield strength (roughly, resistance to bending), infinite
tensile strength (resistance to stretching), and infinite compressive
strength (resistance to compression). What happens when such a mate-
rial is subjected to an infinite force, such as the weight of the infinite
rod? This question is unanswerable because it is ill-conceived, in the
same manner as the question ‘What happens when an object with infi-
nite inertia is subjected to an infinite force?’ The only answer is that
such a scenario is impossible, since there cannot be such an object or
such a force. Material strength is a natural, intensive magnitude; on
my account, therefore, there cannot be an object with infinite yield,
compressive, or tensile strength.
The postulation of one infinite intensive magnitude usually leads to
others. In this case, a material with infinite strength would have infinite
binding energy, which in turn means that the construction of a sample
of such material would involve liberating an infinite amount of energy
from particles with infinite rest mass.
Given that the materials must have finite strength, at least one of
four things will happen: (a) the vertical rod breaks under the weight of
the horizontal rod, (b) the horizontal rod punctures the plane beneath
it, (c) the horizontal rod breaks in two, leaving a finite part having one
end supported by the vertical rod and the other end on the ground,
and an infinite part lying flat on the ground, or (d) the horizontal rod
bends, leaving an infinite portion of it lying flat on the ground and a
finite portion stretching in a concave curve from the ground up to the
end attached to the vertical rod. In each case, the force that would cause
these deformations would be the weight of the infinite horizontal rod.
12.8 Zeno’s paradox
12.8.1 How to complete an endless series

Zeno’s argument against the possibility of motion goes something like
this:
1. To reach the ground, the ball must complete the series, ½, ¾, ⅞, ... .
2. This series is endless.
3. It is impossible to complete an endless series.
4. Therefore, it is impossible for the ball to reach the ground.
What is wrong with this argument is that it trades on confusion between

two meanings of ‘endless’ and two meanings of ‘complete’. To disam-
biguate, I will distinguish ‘endless1’ and ‘endless2’, ‘complete1’ and
‘complete2’. Assume that S is some series of events or actions. Then we
can give the following definitions:
S is endless1 =df S has no last member.

S is endless2 =df There is no time at which every member of S has
occurred.
S is completed1 =df The last member of S has occurred.
S is completed2 =df Every member of S has occurred.
Plainly, if a series is endless1, then it cannot be completed1. Also, if a

series is endless2, then it is never completed2. This, I suggest, accounts
for the plausibility of premise 3. Furthermore, confusion between
completion1 and completion2 would be understandable, since the two
concepts are coextensive for all finite cases; that is, a finite series is
always completed2 exactly when it is completed1 (where we under-
stand the ‘last member’ of the series as the one that occurs last in time).
Nevertheless, matters stand otherwise for infinite series, and from the
fact that an infinite series is endless1, it does not follow that it cannot
be completed2.
Suppose I am going out of town for the week, and I get you to promise
to feed all of my pets while I’m gone. If I have a turtle, then to keep
your promise, you would have to feed the turtle. On the other hand, if
I don’t have a turtle, then feeding my nonexistent turtle is not required.
Similarly, if series S has a last member, then going through the last
member is required for one to count as having gone through all of the
series. But if S does not have a last member, then going through this
nonexistent member is not required.
The Zeno series has no last member, so reaching this nonexistent

member is not required for completing2 the series. One completes2 the
series simply by going through each member at some time. If we are
given the speed at which the ball falls and the distance it is to fall, we
could say, for any member of the series, precisely when that member
gets completed; furthermore, there would be a time such that every
member of the series would get completed before that time. So the series
is endless1 but not endless2; it is never completed1 but it is completed2.
Analogous observations apply to the variation of Zeno’s Paradox that
uses a beginningless series. In this variation, the ball, to reach the ground,
must first reach the halfway point; but before that, it must reach the
one-quarter mark; but before that, it must reach the one-eighth mark;
and so on. Zeno might argue that because this series has no beginning,
it cannot be begun, and therefore, the ball cannot even start its journey
to the ground. Again, Zeno’s case would rest on equivocation. True,
the series has no first member, so it cannot be ‘begun’ in the sense of
performing the first member of the series; but the series can be begun in
the sense that there can be a time such that every member of the series
occurs after that time.
Neither the beginningless nor the endless Zeno series is ruled out by
our account of the impossible infinite: no natural intensive magnitude
is required to take on an infinite value in either of the Zeno series. There
is only a finite distance, traversed in a finite time at a finite speed. The
only infinity appearing in the story is the infinite cardinal number of
stages of the series, but this in itself poses no problem.
12.8.2 The staccato run

There is a variation on the Zeno series that is sometimes called ‘the stac-
cato run’.3 In this version, a person goes through an endless Zeno series,
but the person pauses after each part of the journey. Thus, one travels a
distance of ½ unit, stops, then travels ¼ unit, stops, then travels ⅛ unit,
stops, and so on. (If the person stops for some nonzero temporal interval
each time, then these intervals must get progressively shorter.)
I agree with Aristotle that this infinite series is impossible – but not
for the reason Aristotle gives. Aristotle’s reason for rejecting the stac-
cato run is that the pauses render the divisions between motions ‘actual’
rather than merely potential, and thus that this series would be an
actual infinity, unlike the original Zeno series, which is a mere potential
infinity. My reason for rejecting the staccato run is that this variation
3
Grünbaum 1968.
of the Zeno series, unlike the original version, requires infinite natural
intensive magnitudes.
To see why, consider what happens to your acceleration as you proceed
through this series. In stage n, you travel a distance dn, in a time tn.
Your average velocity during the stage is thus dn/tn. This average velocity
remains constant throughout the series; in stage n + 1, you travel half
the distance as in stage n, in half the time, so your average velocity in
stage n + 1 is (dn/2)/(tn/2) = dn/tn, the same as the average velocity in stage
n. But in each stage, you have to accelerate from speed zero up to your
peak velocity, then decelerate back to speed zero – and in each stage, you
have half the time available to do this as you had in the preceding stage.
This means that your (average) acceleration must double with each
succeeding stage. This in turn means that the (average) force applied
to the runner doubles in every stage. The runner’s body will thus need
infinite material strength in order to avoid disintegration; the runner
will also need to start the journey with an infinite store of energy, thus
making the runner’s body a region of infinite energy density.4
12.8.3 The short staccato run

The staccato run can be modified to avoid the problems just described.
Instead of letting dn+1 = dn/2, suppose we stipulate that dn+1 = dn/4, so that
in each stage, the runner moves only one quarter of the distance he
moved in the previous stage. (The ratio of times remains ½, as before.)
In that case, the accelerations required in each stage remain constant,
and there is no longer any need for infinite material strength, nor for
infinite energy.
This type of series may be uncompletable nonetheless. Most moving
mechanisms contain surfaces in contact that slide relative to one
another, creating frictional resistance. In such cases, some minimum
force is required to overcome static friction, to set the surfaces in rela-
tive motion starting from a position of relative rest. As a result, some
minimum energy is required each time such a mechanism is stopped
4
The energy expended in moving the runner is ∫ F · dx, where x is the distance
through which the force moves the runner. Distance diminishes by half in each
stage, while (average) force doubles, so a fixed amount of energy is expended in
each of infinitely many stages. Oddly enough, Grünbaum (1968, 401) recognizes
these problems (omitting the point about material strength) yet still seems to
regard the staccato motion as possible in some sense. About these problems, he
says only, ‘Let us disregard these specifically dynamical difficulties’ and proceeds
to describe the staccato motion as ‘kinematically’ physically possible. I am unsure
what to make of this. Is he proposing that there are two kinds of modality, ‘kine-
matic’ and ‘dynamic’?
and restarted; thus, an infinite number of starts and stops would require
infinite energy.
However, I have no general proof that there cannot be a moveable
object that stops and starts an infinite number of times in a finite time
period. The possibility of such an infinite series would depend upon
there being some frictionless mechanism for moving the object, a
mechanism that has no minimum energy requirement for starting it up
from its ‘off’ position. If such a mechanism could exist, then the Short
Staccato Run would be possible.
12.9 The divided stick
12.9.1 The standard view

In the Divided Stick problem, the stick is divided infinitely many times,
into halves, quarters, eighths, and so on. What is left at the end?
One puzzle we raised (Section 6.9.6) concerns how many pieces are
left. It looks as though there are Յ0 cuts made in the stick, resulting in Յ0
pieces.5 Each piece will be a cross sectional slice with a thickness of zero.
But the stick started out composed of continuum many pieces just like
that; so where did all the extra pieces go?
First, let’s answer this puzzle according to the standard view: indeed,
the stick would be divided Յ0 times, and it would thus be cut into slices
a single geometric point thick, that is, slices of zero thickness. But there
would not be Յ0 pieces; there would be continuum many pieces. The
stick, in short, would just wind up resolved into the continuum many
single-point-thick cross-sections of which it was composed to begin
with. This must be true, because for any two of those cross-sectional
parts, there would be a stage (in fact, infinitely many stages) at which
the stick was cut somewhere in between those two parts. So at the end,
no connected regions of nonzero length would remain.
The fact that there are Յ0 cuts does not mean that only Յ0 pieces are
produced. To convince yourself of this, consider this example: start with
the real number line between 0 and 1, and imagine that every rational
number in that interval is removed. What is left is the set of all irrational
numbers between 0 and 1. Try to imagine this as a stick with many holes
in it. There are Յ0 holes, at the locations of the rational numbers. But
the parts that remain – all the irrational numbers – have the cardinality
of the continuum. And there will be no extended, connected parts of
5
Oppy 2006, 66–7, 105.
the line, that is, no extended intervals without holes in them, because
between any two real numbers, there are some (in fact, infinitely many)
rational numbers. So cutting out the rational numbers amounts to
making Յ0 cuts, producing continuum many pieces.
What about the question of the mass and length of the stick: since
each of the slices remaining at the end has a thickness and a mass of
zero, why isn’t their collective size and mass also zero? On the standard
view, the answer is that there is no principle of additivity of sizes (or
masses, or anything else) for uncountable collections. Since there are
continuum many of these size-zero parts, all bets are off as regards the
size of the aggregate composed of them. Furthermore, one could rear-
range these pieces to make a stick of any desired length – just move each
slice with original x-coordinate x0 to location r/x0, where r is the desired
length of the new stick. The standard view boldly embraces such coun-
terintuitive consequences.
12.9.2 The impossibility of infinite division with a single knife

So much for the standard view. On my view of infinity and related
matters, what should one think of the divided stick? If, as suggested
in Chapter 11, we reject the possibility of size-zero regions of space,
and presumably also the existence of size-zero parts of an object, then
we have a problem.6 We cannot say that what is left of the stick is a
collection of slices with a thickness of zero. Since we have rejected infin-
itesimals, it also cannot be a collection of slices with infinitesimal thick-
ness. Perhaps there will be nothing left – the stick will simply have been
destroyed by the cutting process?
A more plausible response is that the infinite cutting process is impos-
sible. Suppose, to begin with, that the stick is to be cut using a single
knife. The knife must be moved faster and faster, without limit, to
make all the cuts in a finite time interval. This will require the knife to
undergo unlimited accelerations and decelerations. A material of any
finite strength would break apart at some point in this process. So the
scenario requires a knife with infinite material strength. (As in the case
of Smullyan’s infinite rod, this also implies an infinite binding energy
for the material, whose construction would liberate infinite energy from
particles with infinite rest mass.) Because material strength is a natural
intensive magnitude, this is metaphysically impossible.
In addition, since the knife (with nonzero mass) would have to move an
infinite total distance in a finite time, the amount of work done in moving
6
This problem is discussed by Hawthorne and Weatherson (2004).
the knife would be infinite. Since the scenario is to occur within a finite time
period within the vicinity of the rod, this spatiotemporal region would have
infinite mass-energy density. The vicinity of the rod would thus be a singu-
larity, with infinite space-time curvature according to general relativity.
12.9.3 The case of multiple knives

What if there is more than one knife? Adding any finite number of
knives makes no difference: some of the knives would still have to make
infinitely many cuts in a finite time, and so infinite material strength
would still be required. So what if there is a countable infinity of knives?
Perhaps each knife is initially poised over one of the division points.
The knives could all move simultaneously, thus completing Յ0 divisions
without any knife being subjected to unlimited forces or accelerations.
Alas, there would be no room for all these knives. Recall that the divi-
sions are to be made arbitrarily close to each other – that is, for any
given division location (place where the rod is to be cut), there are other
division locations arbitrarily close. A knife with any nonzero thickness
would therefore get in the way of infinitely many other knives. Each
knife must therefore have a thickness of zero – but we have already
rejected the possibility of size-zero objects.
Okay, forget about making the cuts simultaneously. Let’s suppose that
the Յ0 cuts are made with Յ0 knives, each knife making a single cut. Knife
1 starts out one centimeter above the first division point (in the middle
of the stick). Meanwhile, knives 2 and 3 stand two centimeters above
the second and third division points; knives 4, 5, 6, and 7 are 3 centim-
eters above the fourth through seventh division points; and so on (see
Figure 12.2). So the collection of knives stretches an infinite distance
above the rod. Furthermore, the knives get thinner and thinner as we go
up: knives 2 and 3 are each half the thickness of knife 1; knives 4, 5, 6,
Knife 4 Knife 5 Knife 6 Knife 7
Knife 2 Knife 3
Knife 1
Figure 12.2 Dividing a stick with infinitely many knives

and 7 are half as thick again; and so on. This prevents any of the knives
from getting in the way of each other. Assume that the material strength
of the knives increases without bound, so that the knives always remain
able to cut through the rod despite their diminishing thicknesses; also,
the initial velocity of knife n toward the rod increases as n increases, such
that all the knives pass through the rod within a finite time period.
What happens when a part of the rod is cut in two – do the two resulting
halves remain where they are, or are they pushed apart? Suppose we say
they remain where they are. In that case, there is no real puzzle about
what is left at the end: what is left will just be the same one-meter rod
we started with. The rod will have had a countable infinity of objects
pass through it, but it will remain as it was. Now, you might think that
merely having a knife pass through it would be enough to divide a rod
in two. To see why this is a mistake, we need to first think about what
makes a normal solid object a single object, and what happens when an
object is cut in two.
First, let’s describe how these things work in the actual world. Let’s
say we have a metal rod. What makes this rod a single object is the
fact that there are electromagnetic forces (or dispositions to exert forces)
holding its molecules in a particular configuration. If someone pulls on
one end of the rod, the whole rod moves, precisely because of these
dispositions. The intermolecular forces are sensitive to small changes in
distances: when one pulls on one end of the rod, the rod slightly deforms
(normally, this deformation is undetectably small); the molecules at that
end move very slightly farther away from the rest of the rod than they
were before, and this slight increase in distance causes attractive forces
between the molecules to arise, pulling them back together. If one grabs
both ends of the rod and tries pulling them apart, the attractive forces
can become very large (depending on the material’s tensile strength),
with very slight stretching of the rod.
On the other hand, if one tries pushing the ends of the rod toward
each other, this will cause a very slight compression of the rod, where-
upon repulsive forces arise between the molecules, tending to push
them apart, back to their original distances. The molecules have a sort of
preferred distance from each other, such that if they are moved slightly
closer than that distance, repulsive forces arise, and if they are moved
slightly farther apart than that distance, attractive forces arise.7
However, it is also important that the attractive forces only behave
this way within a certain narrow range of distances: that is, if some part
7
Giancoli 1988, 412.
of the rod is moved very slightly farther away from the rest of the rod
(that is, the rod is stretched), large attractive forces arise; however, once
a part of the rod is moved a significant distance away from the rest of
the rod, the attractive forces become negligible. This latter situation is
what we call ‘breaking’ or ‘cutting’ the rod in two. A certain minimum
amount of energy is required to move the parts of a solid object apart
to sufficient distances that the attractive forces become negligible; this
quantity is referred to as the object’s binding energy. Liquids and gases
have zero or negligible binding energies; solid objects have significant
binding energies.
Notice that these are not minor details; all of this concerns the very
essence of solidity, and what it means to break or cut a solid object apart.
A solid object is essentially an object with a significant binding energy,
and to break or cut such an object apart is just to cause the parts to
become separated to such a distance that the attractive forces between
the parts become negligible.
Now, for purposes of the divided stick example, we have to set aside
atomic theory and imagine that we have a rod composed of atomless
gunk. It is unclear what else we should assume about the physics of this
alternate world. However, it seems fair to assume that we’re supposed to
understand the objects in this hypothetical world in something like the
way we understand real objects, except for whatever needs to be changed
when we give up atomic theory. So even though we are no longer supposing
that the rod contains smallest parts (atoms or molecules), we should still
assume that the parts of the rod are stuck together (they resist being sepa-
rated); this is what makes the rod count as a solid object. We should also
understand that there is some distance such that, if two adjacent parts of
the rod are separated by at least that distance, then they cease to attract
each other (to any non-negligible degree), and the rod is thus broken in
two. Some nontrivial energy input must be required to achieve this result;
that is, the rod must have some nontrivial binding energy.
When a knife passes through an object, thereby cutting it in two, what
happens is that the knife pushes the two parts of the object apart, suffi-
ciently far that the attractive forces between the parts become negligible.
For simplicity, assume that the knife has a triangular cross-section. As
the knife moves into the material that it is cutting, each of the surfaces
of the knife exerts a force on the part of the material that it is in contact
with, pushing that material apart, as in Figure 12.3.
Importantly, the amount of force with which the knife pushes the
two halves of the object apart is related to the thickness of the knife
(and correspondingly, to the angle of the wedge). The thinner the knife
Figure 12.3 Cutting a material
is, the less force it will exert pushing the pieces apart as the knife moves
through the material. As the thickness of the knife approaches zero, that
force approaches zero.8
Now, whatever material the rod is made of, there will be some tensile
strength of that material, and hence, a minimum energy required to
separate two parts of the rod sufficiently to make them into two separate
objects. But in the scenario as we are now envisioning it, the thickness
of the knives approaches zero as the cutting process proceeds. Hence, at
some point in the series of ‘cuts’, the knives are no longer able to exert
sufficient force to cut apart the pieces of the rod. After this point, the
knives would either pass through the rod without breaking it, or get
stuck at the surface. Note that in the actual world, a sufficiently thin
object can in fact pass through a material without breaking it apart.
But suppose we say that the hypothetical world has a different sort of
physics, in which there is no minimum degree of separation required
to overcome the attractive forces between the parts of an object; any
nonzero separation between two parts of the rod causes the parts to
become two separate objects. What then?
The amount of work done in cutting an object apart is equal to the
magnitude of the force exerted on the parts multiplied by the distance
the parts are moved apart by that force; this is also equal to the energy
expended. On the present proposal, since there is no lower limit to the
distance the parts may be moved, there is no lower limit to how much
energy is required to break the object apart. That is to say, the object
has a binding energy of zero. But this means that it is not, after all, a
8
Why? The force moving the knife is a downward force. As the knife approaches
zero thickness, its shape approaches that of a vertical plane. A downward force
impressed on a vertical plane translates into zero horizontal force, since the
driving force is completely orthogonal to the horizontal dimension.
solid object. If a non-solid object – for instance, a pool of liquid – has an

infinite number of blades pass through it, what is left at the end is simply
the same pool of liquid one started with; there is no paradox there.
12.9.4 Infinite energy density

Here is a simpler consideration applicable to either the single-knife or
the multiple-knife story. Some minimum amount of energy must be
expended each time the rod is cut, because one must at least overcome
the binding energy of the rod, which holds any two parts of the rod
together. If, therefore, infinitely many cuts are to be made within a finite
time interval and within the region occupied by the rod, that spatio-
temporal region must have infinite energy density. But this would be an
infinite natural intensive magnitude. Thus, the whole scenario is impos-
sible; there cannot be enough energy in any region to execute infinitely
many cuts.
12.10 Thomson’s lamp
12.10.1 Puzzles of inconsistency and indeterminacy

After the lamp is switched infinitely many times, we are asked to say
whether it ends up on or off. Call this ‘the lamp question’. Why is this
a puzzle?
There are at least two different kinds of puzzling question. One is a
question in which it seems as though any of the possible answers leads
to a contradiction or other absurdity; we might call this an inconsistency
puzzle. For instance, consider the liar sentence,
(L) Sentence L is false.
Is L true or false? If it’s true, then it’s true that L is false, so L is false. But if
it’s false, then it’s false that L is false, so L is true. So both of the possible
answers, ‘true’ and ‘false’, are ruled out.
Another sort of puzzling question is one in which none of the possible
answers is ruled out; this we may call an indeterminacy puzzle. Thus,
consider the truth-teller sentence,
(T) Sentence T is true.
Is T true or false? We could say that T is true, and all that follows is that
it’s true that T is true, which is perfectly consistent. Or we could say
that T is false, and all that follows is that it’s false that T is true, which is
equally consistent. But surely this doesn’t mean that there is no problem
here. Imagine a philosopher declaring that T is simply true. Granted,
this philosopher’s view would not be contradictory, but it would still be
crazy to hold this view. For what could make this the right view to take,
rather than the equally consistent claim that T is false?
The suggestion that T is either true or false but that we simply don’t
know which it is, seems equally benighted. It’s not that there are some
epistemically inaccessible facts about the sentence or the world out
there. It seems, rather, that there just aren’t any facts of a sort that could
make T count as true rather than false, or vice versa. That is, there is no
truth-maker for either T or its negation.
Of course, not all cases in which a question lacks an answer consti-
tute intellectual problems. For instance, when is Chewbacca’s birthday?
There is no determinate answer to this question, but this is not intellec-
tually puzzling, because we have no independent expectation that the
question should have an answer. An indeterminacy puzzle occurs, then,
when a question seemingly should have an answer, and yet none of the
possible answers has a claim to being the correct one. The question of
the truth-value of T seemingly should have an answer, because we have
a background belief or intuition to the effect that meaningful, declara-
tive sentences should generally be either true or false.
I don’t mean to suggest that this problem is insoluble, nor that the
background assumption is unassailable. A solution to the puzzle might
consist of motivating a theoretical account of the conditions under
which a statement lacks a definite truth value, and explaining how T
satisfies these conditions. My point here is simply that some sort of solu-
tion is called for.
What about Thomson’s Lamp? Here, the question is ‘After the infi-
nitely many switchings, would the lamp be on or off?’ Thomson presents
this as an inconsistency puzzle: he says the lamp cannot be on, because
for every time that it was switched on, it was subsequently switched off;
but it cannot be off, because for every time that it was switched off, it
was later switched back on. Yet the lamp must be either on or off. Quoth
Thomson, ‘This is a contradiction.’9
12.10.2 Benacerraf’s solution

Paul Benacerraf penned a solution to the lamp puzzle, a solution that
Thomson subsequently accepted.10 As Benacerraf notes, the description
9
Thomson 1954, 5.
10
Benacerraf 1962; Thomson 2001.
of what happens during the infinite series of switchings does not logically
entail anything about anything that happens outside of (either before or
after) the series. Thus, there is no contradiction in a story according to
which a lamp is switched infinitely many times, in the way specified in
Thomson’s example, and at the instant after all of these switchings the
lamp is on. Nor is there a contradiction in a story in which the lamp is
switched infinitely many times, and at the instant following the switch-
ings the lamp is off.
Benacerraf is correct to note that Thomson’s Lamp does not pose an
inconsistency puzzle; neither answer to the Lamp Question is incon-
sistent. What he overlooks, however, is the indeterminacy puzzle. Indeed,
the very fact that neither answer to the lamp question is inconsistent is
part of the setup of the indeterminacy puzzle. It initially seems as though
the lamp question should have an answer – there ought to be something
that would happen if we had such a lamp. But neither ‘on’ nor ‘off’
seems to be the correct answer, because neither of these answers is better
than the other. As Benacerraf notes, logic won’t tell us how the lamp
ends up; furthermore, it seems as though, short of actually building such
a lamp, nothing else will either.
Benacerraf would probably deny that the lamp question ‘should’
have a determinate answer. Questions about what would happen in
some hypothetical scenario often lack answers because the hypothet-
ical scenario has not been described in enough detail. Benacerraf might
say that either answer to the lamp question is possible, and it is simply
up to the person who wants us to consider such a scenario to stipulate
which thing happens. Their failure to do so leaves the question without
a determinate answer, but this is no more puzzling than the fact that
Chewbacca’s birthday is indeterminate because George Lucas failed to
specify it. So one might argue.11
What should we think of this proposed dissolution of the problem?
It is true that many questions about counterfactual scenarios lack deter-
minate answers. On the other hand, many such questions have determi-
nate answers. Moreover, it is not the case that such questions only have
determinate answers when the descriptions of the scenarios entail those
answers.
For example: suppose there is an ordinary lamp which starts out on.
At time t, I switch it off, then I leave the country. No one else interferes
11
This position is most clearly defended by Peijnenburg and Atkinson (2010,
50–1).
with the lamp either. End of story. Question: one minute after t, would
the lamp be on or off?
This is not indeterminate. Obviously, the answer is ‘off’. Notice that
this answer is not entailed by the description of the scenario. After all,
isn’t it logically possible that I switch the lamp off at t, and then the lamp
just spontaneously goes back on? Yes, but this is not what would in fact
happen if I enacted this scenario.
As this case illustrates, in hypothetical scenarios involving lamps, one
normally does not need to expressly stipulate the state of the lamp at
each specific instant of time in order for there to be an answer to what
state it would be in. Normally, it is enough to describe what things are
done to the lamp in the time leading up to (and possibly including) t.
As long as that is sufficiently described, there is an answer to what state
the lamp would be in at t. In Thomson’s scenario, we are told everything
that is done to the lamp leading up to time t. We may further assume
that no one interferes with the lamp at t either. On the face of it, there-
fore, there ought to be an answer to the question of what state the lamp
would be in at t. Certainly Benacerraf has given us no reason to think
otherwise – the mere fact that the scenario description does not entail
an answer to the question cuts no ice. But, as we have said, no answer
to the question suggests itself as plausibly the right one. So we have an
indeterminacy puzzle, which so far remains unaddressed.
12.10.3 The impossibility of infinite switching

My solution to the puzzle will be that there is no need for the lamp
question to have an answer, because the question contains a false
presupposition: the question presupposes that a lamp could be switched
infinitely many times in a finite time. This is in fact metaphysically
impossible, because it would require the instantiation of infinite inten-
sive magnitudes.
In Thomson’s original description of the scenario, the lamp is
controlled by a pushbutton switch that, when pushed, switches the lamp
from on to off or vice versa. We are supposed to imagine it being pushed
Յ0 times within one minute, with the switchings being performed faster
and faster, such that infinitely many of them fit within any arbitrarily
small neighborhood of the boundary instant, t = 1 minute.
Each time it is pushed, the button, starting from rest, moves a certain
distance downward, then stops, then moves back up and stops. Each
time, this is accomplished in half the time of the previous button-
pushing. This means that the average speed of the button doubles in
each stage of the series, which in turn means that (the lower bound of)
the magnitude of the button’s acceleration doubles, which means that

the force required to accelerate the button doubles.12 All of this implies
several infinite intensive magnitudes. The speed of the switch within any
temporal neighborhood of the final time (t = 1 minute) will be infinite,
since the switch moves an infinite total distance in a finite time. The
amount of work done, and hence the energy expended, in pushing the
switch these infinitely many times would also be infinite. This implies
that the spatiotemporal neighborhood of the switch would have infinite
mass-energy density; hence, it would be a black hole. The friction created
by the moving parts would also increase without bound, leading to the
generation of infinite heat within the switching mechanism. Finally, the
switch must have infinite material strength, since otherwise it would
be destroyed by the unlimited forces applied to it during the interval.
Needless to say, on my view all of this is quite impossible.
12.10.4 The rotary switch variation

All of these infinite magnitudes are generated because we assume that
the switch must move some minimum distance each time the lamp is
turned on or off. But what if there is a way to design a switch such
that there is no minimum distance it must move in order to change its
state?
Here is a possible mechanism (Figure 12.4). There are two metal
contact points. One of them is fixed; the other is attached to a dial.
When the dial is in a certain position – call this position ‘0°’ – the metal
contact points are in contact, which completes a circuit, turning the
lamp on. When the dial is in any other position, the circuit is broken
and the lamp is off. The dial starts out at 0°. After half a minute, it is
rotated 180°, turning the lamp off. After another quarter minute, the
dial is turned back to 0°. After another eighth of a minute, it is turned
45° (one quarter of the first rotation). Then back to 0°, then 11.25°, then
0°, then 2.8125°, and so on. (Aside: why does the rotation multiply by
one quarter each time, rather than by one half? To avoid divergent accel-
erations, as discussed in Sections 12.8.2–12.8.3.)
This series of turns effectively turns the lamp on and off infinitely
many times. Unlike the original version of the story, this story does
12
The accelerations need not be uniform. However, if an object, starting from
rest, is to move a given distance within a given time, then there will be a certain
lower bound to the average value of the magnitude (this being a scalar) of the
object’s acceleration during that time period. This enables us to predict, with
generality, that the lamp switch must be subjected to accelerations that increase
without bound.
Fixed contact
he ht
itc Lig
s
Acme Light
Sw me
Ac
Switches
On Off
Figure 12.4 A rotary switch for Thomson’s lamp
not require the switch to travel an infinite distance in a finite time. The
‘infinite sum’ of the rotations is 2(180°) + 2(45°) + 2(11.25°) + ... = 480°.
Since the angular distance the dial is turned decreases exponentially,
just as the time allotted for the switchings decreases exponentially,
there is no need for ever-increasing average speeds. The average speed
of the dial diminishes by ½ in each stage, while average acceleration
and average force are constant. Since the dial needn’t move an infinite
distance, there is no longer any need for infinite energy, and since the
force applied to the mechanism remains bounded, there is no need for
infinite material strength.
On my view, then, this scenario is metaphysically possible – or at least,
I don’t see any reason why it should not be possible (though it may be
impossible for reasons I haven’t thought of). But in this version of the
story, there is no indeterminacy puzzle anymore. The question, ‘What
would be the lamp’s state at the end of the switching process?’ has a
definite answer: the answer is that the lamp would be on.13
Why? As the series of switchings proceeds, the angle the dial is turned
in each stage progressively diminishes, approaching zero as a limit. That
is, the metal contact attached to the dial comes closer and closer to
remaining at its 0° position. Therefore, if the spatiotemporal path of the
contact point is to be continuous, there is only one place the moveable
contact point can be at the end of the minute: it has to be at the 0° posi-
tion. Any other position would require it to discontinuously jump some
distance at the last instant.
13
This point is insightfully argued by Grünbaum (1968, 403–4) and Beresford
(1981).
Now, you might say: ‘But it is logically possible that the contact have a
discontinuous spatiotemporal path. It is, for example, logically possible
that it just appears at the 90° position at the last instant.’ Yes, it is. It is
also logically possible that the dial turns into a rabbit at the last instant
and hops away. But these things are not what would in fact happen.
In general, physical objects do not teleport randomly, and there is no
reason to think that this general principle would change if the Thomson
Lamp were to exist. So what would happen is that the dial would end in
the on position.
In fact, the lamp would be on not only at the last instant but
throughout some short interval at the end of the minute. The reason
is that when the contact points were sufficiently close, electric current
would arc across them; thus, turning the dial a very small distance fails
to turn the lamp off. This is necessarily true in any world with anything
like electricity as we understand it. To see why, let’s review a little about
electricity.
The current in an electric circuit, according to Ohm’s Law, equals the
voltage of the circuit divided by its resistance. The resistance of a given
part of a circuit (with uniform material and uniform cross-section) is
equal to the resistivity of the material (an intrinsic property of the mate-
rial, which indicates its propensity to resist the flow of electrons through
it), times its length, divided by its cross-sectional area. Now, the resis-
tivity of air is extremely high, on the order of 1016 Ohm-meters, which
is approximately twenty-four orders of magnitude greater than that of a
good conductor, such as copper.14 This is why in most practical contexts,
it is a reasonable approximation to treat air as being completely non-
conductive, or having infinite resistivity. But nothing (not even empty
space) can have literally infinite resistivity. Given the finite resistivity of
air, the resistance created by the gap between the metal contacts would
be proportional to the width of the gap. As the series of lamp-switch-
ings proceeds, this gap becomes arbitrarily small, which means that the
resistance it adds to the circuit comes arbitrarily close to zero, and thus,
the current flow becomes arbitrarily close to the normal current flow of
the lamp when it is turned on.
In the scenario, we are supposed to imagine that any nonzero gap
between the contacts completely disrupts the current flow. This is an
idealization that would be harmless in most practical contexts, but its
falsity matters crucially in this context. The only way in which it could
be true would be if either the voltage of the lamp’s power source was zero,
14
Helmenstine 2014.
or the resistivity of air was infinite. But resistivity cannot be infinite,

since resistivity is a natural intensive magnitude, and if the voltage is
zero then the lamp is always off.
The principles relied upon in this analysis are not minor details –
Ohm’s Law and the inverse relationship between resistance and the
length of a resistor are central principles about how electrical circuits
work. If one says, ‘Well, suppose those laws were false’, then one is really
postulating some phenomenon completely different from electricity. In
that case, it would be up to the person introducing this postulate to
articulate the nature and principles of this alien phenomenon.
12.11 The Littlewood-Ross Banker
12.11.1 The impossibility of infinite motion

In the Banker paradox, you start with bill #1, and you successively trade
bill 1 for bills 2–10, bill 2 for bills 11–20, and so on. You take a $9 profit,
infinitely many times; yet at the end, you have nothing, because for
every n, you gave away bill n in the nth round of the game. To complete
the game in a finite time, you play the first round in half a minute, the
second round in a quarter of a minute, and so on.
By now, it will come as no surprise that I consider this scenario to be
impossible, since it requires various infinite intensive magnitudes. Let’s
try to describe the scenario in a little more detail. You start out with a
‘pile’ consisting of just one dollar, which is labeled ‘1’. The Banker starts
with an infinitely tall pile of $1 bills, with bill #2 on the bottom. Bill 3 is
on top of bill 2, bill 4 on top of bill 3, and so on. Each time the Banker
is called upon to give you some bills, he physically moves them from his
pile onto the top of your pile; similarly, each time you are called upon
to give him a bill, you physically move it from the bottom of your pile
to the bottom of the Banker’s pile.
In this case, as the game proceeds, the speed with which bills must
be moved increases without bound; likewise, both your hand and the
Banker’s hand must move faster and faster without bound, as you
grab and move bills ever more rapidly. The Banker’s hand will also
have to move ever-increasing distances as the minute elapses and the
height of your pile approaches infinity. The velocities of your hands
(or whatever objects are used to move the bills) would be infinite in
any neighborhood of the boundary instant (t = 1 minute), since an
infinite total distance must be covered in a finite time. The paper of
which the bills are made would need to have infinite material strength,
as would your hand and the Banker’s hand, in order to withstand

the unlimited accelerations. You and the Banker would each need
to possess an infinite store of energy in order to complete the infi-
nite series of motions, so your bodies would need to contain infinite
mass-energy density, and hence would be black holes. In brief, this
is impossible in the same way that the original version of Thomson’s
Lamp is impossible.
12.11.2 Motionless variations

If moving infinitely many dollar bills is the problem, perhaps we can
imagine a version of the paradox in which the bills are never physi-
cally moved. Perhaps the infinite pile of money stays where it is, but we
merely designate different bills as belonging to either you or the Banker
in each round of the game. Thus, in round 1, you agree to stipulate –
without physically doing anything to the bill – that bill 1 becomes the
property of the Banker, and he agrees to stipulate – again, without physi-
cally affecting any of them – that bills 2–10 become yours. Is this version
of the scenario possible?
Alas, no. If this ‘designation’ of bills is to mean anything, then the
owner of each bill must be recorded somewhere, in some manner.
Perhaps you have a notebook in which you and the Banker write down
which bills are yours and his. Then your pencil must move with infinite
speed as the minute elapses, since it will have to write infinitely many
numerals in a finite time period. This just recreates the same sort of
infinite intensive magnitudes as the original version of the story. Or
perhaps you record the bills and their owners in a computer. In that
case, your computer will have to possess infinite memory or storage
capacity, infinite read and write speeds, and infinite processor capacity,
in order to record the changes of ownership of all of the dollar bills in a
finite time period.
Or perhaps you and the Banker merely remember which bills belong
to whom. This requires you to have infinite memory capacity and
infinite speed of thought, since you must think of each of infinitely
many numbers, in sequence, within a finite time period. Incidentally,
as noted in Section 9.6, because human beings actually have limited
mental capacity, almost all natural numbers are in fact unthinkable by
any human being.
12.11.3 The case of diminishing bills

Let’s go back to the assumption of bills that are moved as they change
ownership. Suppose that the bills are initially all lined up along a line,
Your money Banker’s money
1
2
3
4
5
6
8
9
Figure 12.5 The Banker paradox with diminishing bills
as in Figure 12.5.15 Bills to the right of the line are in the Banker’s ‘pile’;
those to the left are in yours. In round n, you push bill n to the right,
across the line, while the Banker pushes the bills numbered (10n−9)
through 10n, inclusive, across the line to the left.
To make it possible to move infinitely many bills, we must suppose
that the bills diminish in size as the series progresses. In particular, to
avoid various illicit infinities (see Sections 12.8.2–12.8.3 above), assume
that for each n, bill n+1 is one quarter the height and one quarter the
width of bill n.16 We might as well give the bills uniform thickness and
mass density, so that bill n+1 has one sixteenth the mass of bill n. All this
15
This is based on the presentations of Allis and Koetsier (1991) and van Bendegem
(1994), with slight modifications. In the illustration, I have given each bill #(n+1)
one half the width and height of bill n, rather than one quarter the width and
height of #n; this is for convenience, to make the illustration easy to make out.
16
As you proceed through the game, your hand engages in two kinds of motion:
(i) horizontal motions in which you push bills to the right then pull your hand
back, and (ii) vertical motions in which you move your hand down to the next bill
in the series. The horizontal motions would constitute a series of motions analo-
gous to the Staccato Run discussed earlier. For the reason given in Section 12.8.2,
the horizontal distances cannot diminish by only one half in each stage; they
may, however, diminish by a factor of four in each stage, as discussed in Section
12.8.3. Furthermore, assuming that the vertical motion of your hand stops in each
appears to make it metaphysically possible that all of the infinitely many

bills should be moved in the required way in a finite time period.
What is the outcome of this scenario? Littlewood and Ross’ reasoning
is correct here: at the end, all the bills will lie to the right of the line. You
might consider this paradoxical; indeed, you may now think that the
original paradox has simply been traded in for a very slightly, and unim-
portantly, different version of the paradox. So let me explain why the
situation is at least a little less paradoxical than it appeared in our original
statement of the problem. There are three reasons why I say this.
First, one of the bizarre things about the Littlewood-Ross scenario
was the way in which a certain quantity (how much money you have)
approaches infinity as the game proceeds, and yet at the end of the game,
that quantity stands at zero. There is a discontinuity not only in a certain
conventional quantity (monetary value) but also in certain objective,
physical quantities: if all the dollar bills have the same size and mass,
then the size and mass of your money pile increases without bound as
the game nears its completion, but then suddenly ends up at zero.
In the new version of the scenario, where the dollar bills become
smaller and smaller, it remains true that the number of bills that belong
to you approaches infinity as the game proceeds yet ends up at zero.
However, if we characterize your holdings in terms of any natural, phys-
ical magnitude, the discontinuity evaporates. For example, consider the
mass of your money pile. In our new version of the scenario, each bill in
the series is more massive than all the succeeding bills combined.17 Thus,
you start with bill #1, which is more massive than all the remaining
bills combined. In the first round, you trade that bill for bills 2–10, thus
stage (while you are pushing the appropriate bill to the right), the vertical motions
also have the structure of the Staccato Run. For the same reason, therefore, the
vertical distances should diminish by a factor of four. Hence, we should let the
bills’ widths and heights both diminish by a factor of four in each stage.
17
To see why, recall that we have stipulated that for each n, the mass of bill
n + 1 is one sixteenth the mass of bill n. Thus, if the first bill has a mass of m1,
( ) ( )
n −1 n
then bill n has a mass of m1 1 and bill (n+1) has a mass of m1 116 . The
16 ∞
a
formula for the sum of a convergent infinite geometric series is ∑ ai = ,
i =1 1− r
where a is the first term of the series and each successive term is equal to r times
the term before it. Thus, the combined mass of bills (n+1) through infinity will be
n −1
⎡ ⎛ 1 ⎞n⎤ ⎛ 1 ⎞ ⎡ ⎛ 1 ⎞ ⎤ ⎛ 16 ⎞ ⎡ ⎛ 1 ⎞ ⎤ ⎛ 1 ⎞
n
⎢m1 ⎝ ⎠ ⎥ ⎝ 1 ⎢ m1 ⎥ = ⎢m1 ⎥
⎣ 16 ⎦ 16 ⎠ ⎣ ⎝ 16 ⎠ ⎦ ⎝ 15 ⎠ ⎣ ⎝ 16 ⎠ ⎦ ⎝ 15 ⎠ . Thus, the combined
mass of bills (n+1) through infinity is one fifteenth of the mass of bill n.
shrinking the mass of your money pile. In the second round, you trade
bill 2 (whose mass is greater than that of bills 3 through ∞ combined) for
bills 11–20, thus shrinking the mass of your money pile again. Matters
continue this way, with the mass of your money pile approaching zero
as the number of rounds played approaches infinity. At the end of the
game, as one should expect, the mass of your money pile stands at zero.
The same would be true of the volume of your money pile, which dwin-
dles as the game proceeds and then ends at zero.
Second, in the original statement of the problem (Section 3.12), we
considered a variation in which, when called upon to trade in bill #1,
you disguise bill #10 as bill #1 and vice versa, hand the Banker bill #10,
and keep bill #1. You behave similarly with respect to bills 2 and 20,
3 and 30, and so on. Call this the ‘swapping variation’. In the swap-
ping variation, both you and the Banker have qualitatively the same
thing at each stage as you had in the original version of the game, yet
in the swapping variation you wind up with an infinite amount of
money instead of none. If you found that puzzling, that puzzle has
now been avoided. In the old version of the Banker game, each of the
bills is qualitatively identical, save for the number printed on it. In our
new version of the Banker game, the bills are no longer so similar to
each other; instead, the bills get smaller and smaller as you proceed
through the series. So swapping bills 1 and 10 (and so on) does not
result in the two players having qualitatively identical things at each
stage of the game as they have in the version in which those bills are
not swapped.
Third, part of the Littlewood-Ross Paradox was a decision theoretic
problem: it seems that you do the prudent thing in each stage of the
game by trading your lowest numbered bill for ten bills from the Banker’s
pile – and yet you wind up with the worst possible outcome (from the
standpoint of self-interest) at the end of the game. But this depends
upon the assumption that each of the bills in the series has equal value,
and that the value of the bills does not diminish as one moves to higher-
numbered bills. This assumption, I claim, is false on any rational utility
function, given our present version of the scenario.
How can that be? Begin with normal assumptions about the value of
money: you value money because you can use it to buy things. But you
could not buy something with a dollar bill if the seller could not detect
your bill. Given the way in which the bills diminish in size and mass
as we go through the series, there will quickly come a point at which
the bills are too small for a human being to see. Not long after that, the
bills will be too small to see with a microscope. Later, they will be too
small to detect even with the most powerful electron microscope in the
world. Now, if it is proposed that you trade a detectable dollar bill for ten
undetectable dollar bills, you would be prudentially rational to reject
the trade. Thus, there is a point in the game at which you should stop
accepting the Banker’s offers.
Now let’s relax the assumptions of the preceding argument. Perhaps
you don’t need to buy things with the dollar bills; let’s just suppose
you have some use for them, whatever it might be. Still, it seems that
any use must require the bills to interact with the world around them
in some way. But any effect that a dollar bill might have will at some
point diminish as the size of the bill diminishes, and therefore, whatever
effect you value, at some point in the series, the bills that the Banker is
offering will become unable, or less able, to produce the desired effect
(and remember that any given bill is larger than the sum of all the bills
that follow it in the series). In the case of buying things, one desires the
bills to produce a psychological effect on people selling goods. But the
point generalizes to any real effect.
But what if the bills have some other property, besides their size and
mass, that produces some desired effect? For instance, maybe each bill
has an electric charge of +1, which enables it to be detected however
small the bill may be. In that case, the charge density of the region of
space occupied by the pile of bills would be infinite. This would consti-
tute an infinite natural intensive magnitude; hence, the initial condi-
tions of the scenario would be impossible.
But now, what if you just value having bills intrinsically, so that
it does not matter to you what effects they have on anything else?
That is, for every natural number n, you intrinsically prefer having
n+1 bills on your side of the line over having n or fewer bills on
your side of the line. In that case, I would say you have an irrational
utility function – it is irrational for a person to value the moving of
bills from one side of a line to another (with non-diminishing value
placed on arbitrarily small changes in position), completely inde-
pendently of whether those motions have any effect on anything,
including the person himself (and thus independently of whether
that person could even detect the motions). This is irrational because
there is no understandable way in which such motions can be viewed
as good. The fact that, given this utility function, the Banker scenario
generates a decision-theoretic paradox (maximizing your utility in
each stage of the game results in the minimum utility at the end of
the game) simply underscores the irrationality of the assumed utility
function.
12.12 Benardete’s paradox
12.12.1 The first impossibility: Infinite binding energy

Benardete asks us to imagine a ball poised at the top of a hill, with
impenetrable walls placed halfway down the hill, a quarter of the way
down the hill, an eighth of the way, and so on. The ball cannot roll any
distance down the hill, yet no particular wall will be the one to stop
it, since for every wall, there are other walls that the ball would have
encountered before getting to that one.
This scenario is metaphysically impossible. To begin with, there cannot
be an impenetrable wall, if this is understood as a wall that would not
be penetrated by any force. Such a wall would require a material with
infinite yield strength and infinite binding energy.
Assume, then, that the walls are not impenetrable in general; rather,
each wall is merely strong enough that it would not be penetrated by
this particular ball under the force of gravity. Since the walls have dimin-
ishing thickness – wall n+1 is half as thick as wall n, for each n – we must
assume that the material strength of the walls increases correspondingly,
so that wall n+1 is made of a material twice as strong as that of wall n.
Call the top of the hill ‘point A’. The series of walls converge on A, so
that within any nonzero distance of A there are infinitely many walls.
Each of these walls must exceed some minimum binding energy, because
each must be capable of stopping the ball on its own. Therefore, the
binding energy of the system including the infinite set of walls would
be infinite.
Is this infinite binding energy enough to render this setup metaphys-
ically impossible? Arguably, yes. Binding energy is a natural magnitude
(it is causally relevant to the conditions under which a system can be
broken apart). Is it intensive or extensive? Consider the simple case of
a hydrogen molecule (H2). The molecule can be divided into two parts,
namely, the two individual hydrogen atoms. But the binding energy of
the molecule of course does not equal the binding energy of the first atom
plus the binding energy of the second atom (breaking the molecule apart
into subatomic particles would require more energy than that required
to break apart two separate hydrogen atoms). So binding energy is non-
additive in the way required for a magnitude to qualify as intensive. Since
binding energy is a natural intensive magnitude, it cannot be infinite.
12.12.2 The second impossibility: Infinite material strength

When an object comes to rest against a wall and the wall stops the object
from moving, it typically appears to casual observation that the wall is
completely unaffected by the object. For example, if I lean against a brick

wall, the brick wall appears to remain completely rigid, not bending or
moving at all. But this is an illusion; on the microscopic level, the wall
deforms – that is, it bends in response to my pushing against it (this
bending is of course too slight for the naked eye to observe). This is true of
any solid object, no matter what it is made of; no object is perfectly rigid.
Assume, therefore, that each wall is such that, if the ball were to lean
against the wall, the wall would bend slightly in the direction that the
ball was pushing it. By how much would the wall bend? (Of course,
according to the story, the ball never actually winds up leaning against
any of the individual walls; this question is just about what would
happen if, hypothetically, the ball could lean against one of the walls.)
In the original version of the story, the walls are said to be ‘impene-
trable’. Let us replace that supposition with the supposition that each
wall n is strong enough that it would prevent the ball from reaching
the wall immediately downhill from it (wall n−1). I assume, then, that
the deformation in wall n would be small enough that wall n would
not come in contact with the downhill wall (wall n−1) – for example, if
the ball were to rest against wall #3, wall #3 would bend less than the
amount necessary for wall #3 to collide with wall #2 (see Figure 12.6).
Now, because the walls in this scenario are placed closer and closer
together, the amount of deformation allowed must decrease; specifically,
wall n+1 must in general be constructed so as to deform half as much (or
less) as wall n. This means that the compressive and yield strengths of the
walls continually double – wall n+1 has double the compressive strength
and double the yield strength as wall n. As we get closer and closer to the
top of the hill, the material strength of the walls approaches infinity.
Wall 3 Wall 2 Wall 1
Figure 12.6 Benardete’s walls

Now consider the composite object composed of all the walls, and
call this object ‘wall ω’. What is the compressive strength of wall ω? At
its downhill edge, it has some finite compressive strength. But at its top
(in the neighborhood of the top of the hill), its compressive strength
is infinite. There is no nonzero distance that this object would deform,
since any nonzero deformation would involve some of the walls in the
series being pushed into and even through other walls. In other words,
at its leading (uphill) edge, wall ω is perfectly rigid. This is metaphysically
impossible. Hence, there simply cannot be a collection of walls such as
we have postulated.
12.12.3 The infinite pile of slabs

In another variation, Benardete asks us to imagine a pile consisting of
infinitely many opaque slabs, where slab n+1 lies on top of, and is half
the thickness of, slab n, for n = 1, 2, 3, and so on. If you stand above the
pile and look down, you will not see any particular slab, since for each n,
your view of slab n is blocked by slab n+1. But nor can you see through
the pile. So what do you see?
It may not be immediately obvious that this scenario is impossible in
essentially the same way as the previous version of Benardete’s Paradox.
In the previous variation, the impossibility lay in the assumption that
as the walls become ever thinner, each wall nevertheless remains
impenetrable to the ball. In the present variation of the paradox, the
impossibility lies in the assumption that as the slabs become ever
thinner, each slab nevertheless remains opaque – that is, impenetrable
to photons.
No physical material is either completely opaque or completely trans-
parent; for any material, a sufficiently thin layer of it is transparent,
and a sufficiently thick layer is opaque. For instance, we normally think
of paper as opaque; however, a very thin piece of paper is transparent.
We normally think of water as transparent; however, a very thick layer
of water – say, the thickness of the ocean – is opaque. The thickness
required to render a layer of material for practical purposes ‘opaque’
depends upon a physical property of the material known as its attenu-
ation coefficient (its resistance to transmission of light). The greater the
attenuation coefficient, the more opaque a substance is (or, the thinner
an opaque layer of the material can be). Paper, for example, has a much
higher attenuation coefficient than water.
If light with a given intensity, I0, strikes a layer of some material with
an attenuation coefficient of μ, the intensity of light, I, that will pass
through the layer is given by a formula known as the Beer-Lambert

Law:
I I 0e − P x
where x is the thickness of the layer of material and e is Euler’s number

(≈2.718). Every material has an attenuation coefficient greater than
zero and less than infinity, so for any given thickness, some light will
pass through, and some light will instead be absorbed or scattered. The
substances that we colloquially call ‘opaque’ are simply those with a
relatively high attenuation coefficient, such that a small proportion of
light passes through even with relatively thin layers. The substances that
we normally call ‘transparent’ are those with a low attenuation coeffi-
cient, such that a high proportion of light passes through even with
relatively thick layers.
Suppose that all the slabs in the Benardete scenario are composed of
a material with an attenuation coefficient of μ1, and thus the pile as a
whole has an attenuation coefficient of μ1. In that case, the observer
standing over the pile will see through the top of the pile to a certain
depth – that is, to a first approximation, there will be some n such that
slabs n+1 through ∞ will be transparent. More precisely, the slabs will
approach perfect transparency as n increases.
What the observer will see will simply be determined by the distribu-
tion of photons that are reflected, absorbed, and scattered by the pile.
The fact that there is no top slab is irrelevant. A given photon will have
a probability of penetrating the pile to a given depth as determined by
(another form of) the Beer-Lambert Law:
P( x) = e − μ1x
where P(x) is the probability of penetrating to (at least) depth x. Thus,

what the observer sees is perfectly determinate. The appearance of
paradox arises only from the false assumption that each slab would be
opaque, which, when combined with the fact that there is no ‘top’ slab
in the pile, would render it indeterminate what one would see.
But now suppose the slabs are each composed of different materials.
Slab 1 is composed of a material with attenuation coefficient μ1, slab 2
has attenuation coefficient μ2, and so on. Suppose further that these
materials are of ever increasing opacity, so that for each n, μn+1 = 2μn,
and assume that μ1 is sufficiently high that slab 1 is ‘opaque’ by normal
standards. In this case, every slab has the same opacity, that is, it is just
as difficult to see through (it is just as unlikely for a photon to pass

through) slab n + 1 as it is for slab n.
This version of the case restores Benardete’s paradox. This version,
however, is metaphysically impossible. Think of the pile of slabs as a
single composite material object. What is the attenuation coefficient
of this object? This will be relative to a direction of approach: from the
bottom its attenuation coefficient is μ1. But from the top, its attenu-
ation coefficient is infinity – that is, no photon would have any nonzero
chance of penetrating the object to any nonzero depth. But attenuation
coefficient is a natural, intensive magnitude; it therefore cannot be
infinite. There cannot exist a material object such as the one described.
12.13 Laraudogoitia’s marbles
Laraudogoitia asks us to imagine an infinite series of marbles arranged in

a straight line, with each marble having half the diameter but the same
mass as the previous one, so that the whole collection occupies a finite
space. If the first marble is pushed directly toward the second one, what
will happen? An infinite series of collisions, one after another, by which
each marble transfers its momentum to the next one in line. Since there
is no last marble in the series, once the series of collisions is complete
there will be no particular marble that carries the momentum. Thus,
momentum will have been destroyed. Furthermore, since Newtonian
mechanics is time reversible, this shows that it is also possible for
momentum to be spontaneously created.
Now consider: what is the mass density of the region in which the
marbles are located? This is a finite volume of space, which contains
infinitely many marbles, each with the same nonzero mass; thus,
the density of this region is infinite. At the beginning of the chain of
marbles is marble #1; at the end of the chain there is no last marble but
simply a series of ever smaller marbles. Any region of space extending
any distance surrounding the end of the chain will contain infinitely
many marbles and thus will have infinite density. Density is a natural,
intensive magnitude. Hence, the setup of the scenario is metaphysi-
cally impossible. The fact that the scenario would result in a violation of
momentum conservation therefore should not trouble us.
12.14 The spaceship
A spaceship travels at a speed of (at least) one meter per second for one
second, then 2 m/s for a half second, then 4 m/s for a quarter second,
and so on. At the end of two seconds, where is the spaceship?
This scenario is of course impossible. It involves a spaceship that

travels an infinite distance in a finite time; thus, the velocity of the
spaceship over the two-second interval would be infinite. Velocity is a
natural intensive magnitude; hence, it cannot be infinite.
But what exactly would prevent the spaceship from accelerating in the
manner described? At what point in the series would the ship be meta-
physically prevented from further acceleration?
Leaving aside relativity (which is metaphysically contingent), there
is no metaphysical speed limit that all objects must obey, nor is there
any metaphysical limit to the acceleration that an object may undergo.
Rather, any given spaceship must have some limits depending upon its
specific design and properties, with different spaceships having different
maximum accelerations.
Bear in mind that the Spaceship Paradox requires the spaceship not only
to have a velocity that increases without bound, but also to have an acceler-
ation that increases without bound as t approaches 2 seconds. That is, not
only is the ship’s velocity increasing, but the rate of increase of its velocity is
increasing: first the ship increases its speed from 1 to 2 in one second (for an
average acceleration of 1 m/s2); then it increases its speed from 2 to 4 in half
a second (for an average acceleration of (2)/(½) = 4 m/s2); then it increases
from 4 to 8 in a quarter second (a = (4)/(¼) = 16 m/s2); and so on.
Suppose for simplicity that the spaceship is propelled by rockets. The
maximum achievable acceleration would depend, among other things,
on how much fuel the spaceship had on board at the start of its trip. The
quantity of fuel might be any finite amount, but it cannot be infinity.
Whatever that amount is, there would be some stage in the acceleration
process at which the rocket would run out of fuel, thus preventing it
from completing the next stage.
But now suppose that, as the spaceship travels through space, it
collects more fuel. Perhaps in the initial conditions of the scenario, there
just happens to be rocket fuel distributed throughout space in the path
of the spaceship, so much fuel that the spaceship need never run out.
Then what will limit the spaceship’s acceleration?
In this case, its maximum acceleration would most likely be deter-
mined by the design of the engines, which determine how much fuel
they are capable of burning per unit time. After a certain point, the ship
will be unable to increase its rate of acceleration further due to such
factors as the inability to fit more fuel into the engine, or the engine’s
inability to sustain higher temperatures without melting, and so on.
Speaking more generally, the ship must have some maximum rate of
acceleration determined by the strength of the materials of which it is
constructed. The acceleration an object undergoes is proportional to

the force impressed upon it. At some point (if the engines didn’t melt
first, or reach their maximum fuel throughput, or the like), the forces
impressed on the spaceship would be so large as to break the ship apart.
To avoid this, we would have to postulate infinite material strength,
which is another infinite natural intensive magnitude.
12.15 The Saint Petersburg paradox
12.15.1 The traditional version: Monetary payouts

A coin is flipped some number of times, until the first time it lands
heads. You then receive a payout equal to $2n, where n is the number of
flips that it took to get heads. The possible payoffs are thus $2, $4, $8,
and so on, which have probabilities of occurrence of ½, ¼, ⅛, and so on.
The expected payoff of the game is thus ($2)(½) + ($4)(¼) + ($8)(⅛) + ‫ڮ‬
= ∞. However, most people find it absurd that a chance to play this game
is infinitely valuable.
Bernoulli’s original solution succeeds for this version of the paradox:
money has diminishing marginal utility; thus, for example, a ¼ prob-
ability of getting $4 is worth less than a ½ probability of getting $2. In
general, a 1/2n−1 probability of getting 2n+1 dollars is worth less than a
1/2n probability of getting 2n dollars.18
You might think: suppose we just increase the monetary amounts to
compensate. While the probabilities diminish by one half with each
successive possible outcome, we could stipulate that the monetary
payouts more than double each time – perhaps they multiply by four
each time, or perhaps they are squared each time, etc. This, however,
will not genuinely get around Bernoulli’s point, for there is a finite limit
to the benefit that one could gain from money, whatever the quan-
tity. Perhaps the utility of money asymptotically approaches this limit,
or perhaps at some point it reaches the limit then remains constant;
either way, one simply cannot produce arbitrarily large utility merely
by adding more money, as this version of the paradox requires us to
assume.
12.15.2 The supernatural version: Days in heaven

To revive the paradox, assume that the payouts are given in something
that does not have diminishing marginal utility. Imagine that you meet
18
Bernoulli [1738] 1954, 24–6, 32–3.
a person claiming to be God who performs a series of miracles to prove

his identity. Once you are convinced, God offers to play a version of the
coin-flipping game with you (for a certain price, to be discussed later).
He assures you that the coin is fair, that it will be flipped until the first
time it lands heads, and that He will then give you a payoff of 2n days
in Heaven, where n is the number of flips. (If you reject the offer, then
when you die, you will simply pass out of existence, with no time in
Heaven.) Assume also that each day in Heaven is just as good as the last,
so there is no diminishing marginal utility.
A chance to play the game now has infinite expected utility, so it
would be rational to accept any finite cost for a chance to play – or so
one could argue. Thus, for example, it would be rational to hand over all
of your worldly possessions, if that were the price of playing the game
once. It would even be rational to agree to spend a trillion years (but not
forever) in Hell, if that were the price of playing one round of the game.
Can this be correct?
Here is one objection to the foregoing reasoning. The expected utility
calculation assumes that you know, for certain, such propositions as
(i) that paying a certain price will result in your being allowed to play the
game, (ii) that the coin to be flipped is fair, (iii) that for each n, you will
really get the payout of 2n days in Heaven if the coin comes up heads on
the nth flip, and (iv) that each day in heaven will really be just as good
as the last. Call the conjunction of propositions (i)–(iv) ‘the Setup’. So
the expected utility calculation assumes that the Setup has a probability
of 1. But this is metaphysically impossible. It is in principle impossible
for any external-world proposition, including any prediction about any
causal consequences of any contemplated action, to have a probability
of 1. (There are well-known arguments for this in epistemology; I will
simply assume that this point is accepted.) Therefore, the calculation
by which we deduced that the expected payoff of playing the game is
infinity is mistaken.
So far, so good. Unfortunately, this objection barely wards off paradox
for a moment. The original expected utility calculation was this:
∞
⎛1⎞ ⎛1⎞ ⎛1⎞
Expected Utility = ⎜ ⎟ (2 ) + ⎜ ⎟ ( 4) + ⎜ ⎟ (8)
⎝2⎠ ⎝4⎠ ⎝8⎠
∑ 1 = ∞.
i =1
Once we take into account the possibility that the Setup is false, we must
modify this calculation. Suppose there is a probability of p that the Setup
is correct. For simplicity, assume that if the Setup is false, then playing
one round of the game will net you zero utility. In that case, the new
expected utility calculation must be as follows:
∞
⎛ p⎞ ⎛ p⎞ ⎛ p⎞
Expected Utility = ⎜ ⎟ (2 ) + ⎜ ⎟ ( 4) + ⎜ ⎟ (8) + = ∑ p.
⎝2⎠ ⎝4⎠ ⎝8⎠ i =1
As long as p is nonzero, this sum is again infinite. Thus, according to

the expected utility calculation, even if you are 99.9% confident (but
not 100%!) that ‘God’ is lying to you and will not actually give you any
days in Heaven no matter what the outcome of the coin tossing exercise,
it would still be rational for you to give up all your worldly possessions
and agree to spend 10100 years in Hell in exchange for a chance to play
this game once (of course, you could never be 100% certain that you
would wind up spending any time in Hell either). But ... really?
12.15.3 The probability of the setup is zero

There seem to be the following possible ways of avoiding this
consequence:
a. Reject expected utility maximization as a rational decision-making

rule, at least for this case. If we want to say that it would be positively
irrational to agree to spend a googol of years in Hell in exchange for
a chance to play this game, then we would have to say that maxi-
mizing expected utility is in this instance not only not required, but
rationally forbidden.
b. Insist that days in Heaven necessarily have diminishing marginal
utility, and that this cannot be stipulated away, nor can we find any
other prize that does not have diminishing marginal utility as its
quantity increases.
c. Maintain that p actually equals zero, even after one has been presented
with what might superficially appear to be strong evidence for the
Setup (whether because the Setup is metaphysically impossible or for
some other reason).
I don’t see a plausible independent motivation for (a) or (b). (c), however,
has some independent motivation and is not obviously wrong.
Recall that the Setup is a conjunction of four conditions. If any one
of these conditions has probability zero, then the Setup has probability
zero. Now focus on condition (iii), the condition that for each n, you
will get a payout of 2n days in Heaven if the coin comes up heads on the
nth flip. This can be viewed as a countably infinite conjunction of the

following propositions:
O1: If the coin comes up heads on the first flip, you will get 2 days in
Heaven.
O2: If the coin comes up heads on the second flip, you will get 4 days
in Heaven.
O3: If the coin comes up heads on the third flip, you will get 8 days
in Heaven.
⁝
The probability of this infinite conjunction is equal to the probability of

O1 being true, times the probability of O2 being true given the truth of
O1, times the probability of O3 being true given the truth of both O1 and
O2, and so on, that is,
∏ P( O | O
i =1
i 1 & Oi 1 )
(stipulate that P(O1) counts as the first factor in that product).

Call the ith factor in that product ‘pi’ (that is, let pi = P(Oi|O1& ... &Oi−1)
for i > 1, p1 = P(O1)). For the infinite product, and hence the probability
of the Setup, to be nonzero, pi must approach 1 as a limit as i increases.
That is, pi must come arbitrarily close to 1 for sufficiently large i. Should
this be true on a rational probability distribution?
Here is one reason why one might think that pi should constantly
increase: (O1& ... &Oi−1) constitutes something like inductive evidence
for Oi. The larger i is, the greater the inductive evidence. Hence, pi should
increase with i.
However, here is a reason why one might think that pi should not
increase with i. The third condition in the Setup, specifying the payouts
you receive for the different possible outcomes of the coin-tossing,
amounts to an infinite collection of promises that God makes to you:
God promises that O1, that O2, that O3, and so on. This is a series of ever-
increasing commitments. The first promise, O1, is modest; it commits
God only to keeping you in heaven for one day should the coin come
up heads on the first toss. The second promise, O2, is in one sense twice
as great a commitment: God commits to your being in heaven for twice
as long, should the coin come up tails first and then heads. The third
commitment, O3, is twice as great again, since it commits God to keeping
you in heaven for twice as long again (should the coin come up heads
on the third toss). One might think that, as the time commitments
increase, it becomes less likely, rather than more likely, that the promise
would actually be kept. If this isn’t obvious in the case of the one-day
and two-day commitments, compare the commitment to give you one
day in heaven (under certain conditions), to a commitment to give you
a trillion years in heaven (under certain conditions): the latter commit-
ment is less likely to actually be kept. There is more time during which
things might go wrong (God might change His mind, Satan might over-
throw God, heaven might fall into disarray, and so on). In general, a
prediction is less likely to be correct as the length of time that it covers
increases. This includes your (conditional) prediction that God would
keep you in heaven for n years.
Arguably, something similar obtains for any sort of prize that might be
offered. The point is obvious in the case of the monetary payouts in the
original version of the paradox: a commitment to pay $2 is much more
likely to actually be honored than is a commitment to pay out a trillion
dollars; the latter is in fact virtually certain not to be honored. Of course,
the latter commitment is much less likely to be triggered, since it would
only be triggered by a series of 39 tails followed by heads, but that is irrele-
vant to the present point. The present point is that even if the commit-
ment were triggered (if the coin actually came up tails 39 times before
heads finally appeared), it is virtually certain that no one would actually
pay you $1 trillion, no matter what they had previously said. The general
point is that a larger payout entails a larger change in the world, which is
less likely to actually be successfully brought about by a given exercise.
12.15.4 A counter-argument: The perfection of God

One might try to counter the foregoing reasoning by arguing that since
God is morally perfect, He would not break a promise, no matter how
much time elapses (perhaps God is a Kantian?). Furthermore, since God
is all-powerful, he has the power to guarantee you any amount of time
in heaven. These things, it might be said, are true by definition. And in
case you are tempted to say that one could never be sure that the person
with whom one was playing the game was really God, remember that to
get the paradox going we only need a nonzero probability that the Setup
is true. So it suffices that there is a nonzero probability that the person
with whom you are playing the game is God, where God, by definition,
is (among other things) an all-powerful Kantian (hence, He never lies or
breaks promises).
My response at this point is that God, so conceived, is metaphysic-
ally impossible on my account of the infinite. This is because God is
supposed to be infinitely powerful, and this appears to constitute an

infinite, natural intensive magnitude.
Objection: perhaps omnipotence is not an infinite natural intensive
magnitude, because it is not a magnitude at all. It is just a property.
Reply: omnipotence in the sense intended in definitions of God at
least includes or entails certain magnitudes, and this will suffice for the
point. To illustrate, imagine three beings, A, B, and C. A can lift up to
100 pounds; B can lift up to 200 pounds; C can lift up to 300 pounds.
There is at least one sense, or one respect, in which B and C are each
more powerful than A. Furthermore, we can say that C is more powerful
than A to a greater degree than B is. That is, C’s power exceeds A’s by
a larger margin than B’s power exceeds A’s. So this sort of power is a
magnitude.
Granted, that sort of power is not what theists are talking about when
they say that God is powerful – they do not merely mean that God
can lift any weight. However, that sort of power is included in God’s
power, else human beings would possess a power that God lacks. Thus,
God’s omnipotence entails or includes the sort of power I have referred
to, which we may call ‘weight-lifting power’. The weight-lifting power
of an omnipotent being would presumably be infinite. Hence, God’s
omnipotence entails his instantiation of an infinite, intensive magni-
tude. By the same sort of reasoning, an omnipotent God would presum-
ably instantiate any number of other infinite intensive magnitudes.
For these reasons, I believe that there could not be an omnipotent
being. There can be very powerful beings, and there is no limit to how
much power a being can possess, but no being can possess unlimited
power. For this reason, too, there cannot be a being who is able to guar-
antee you n days in heaven for every possible value of n.
12.16 The Martingale betting system
You are to play a gambling game in which (as in real life) the odds favor
the house. You start with an infinite bankroll, and there are no betting
limits. Despite the fact that the odds always favor the house, it appears
that you can guarantee that you come out ahead, simply by doubling
your bet every time you lose. Eventually, you will win and the win will
outweigh all your previous losses. Furthermore, if this strategy works, it
seems that you can repeat it infinitely many times, thus guaranteeing
an infinite profit.
Something seems wrong with this. Since the odds on each bet favor
the house, it seems that the house should come out ahead, particularly
in an infinitely long series of plays. There is also a decision-theoretic

puzzle: it appears that each bet is prudentially irrational on your part,
and rational on the part of the house, independent of what other bets
have been made; yet the whole series of bets guarantees a net profit for
you and a net loss for the house.
To begin resolving this paradox, note first that the strategy for winning
an infinite amount of money from the house does not work: it depends
upon the possibility of playing the game infinitely many times in a finite
time period, which depends on the ability to play the game faster and
faster without limit. This is an ability that no one could have, for the
same sort of reasons that rule out Thomson’s Lamp and the Littlewood-
Ross Banker (see Sections 12.10.3 and 12.11).
Let us then focus on the simple strategy, in which you hope to win
a single dollar by first betting $1, and then doubling down with each
loss. Having given up the idea that you could play the game faster and
faster without limit, we assume that each play of the game takes one
minute.
Now, it is true that, if this game can be continued indefinitely, then
you will eventually win. However, keep in mind that there is no specific
time by which you are guaranteed to win. Furthermore, for any given
time, your expected profit at that time, conditional on your having
adopted the Martingale betting strategy, is negative, and it becomes
more negative for later times. This is because, although the probability
of your coming out ahead by $1 approaches 1, the total amount that
you will have lost if you haven’t yet come out ahead also increases – and
the amount you stand to lose increases faster than the probability of a
repeated loss decreases.
For illustration, assume that the game is roulette, so you have an 18/37
probability of winning each time you play. Now consider a time just two
minutes after you started playing. At this time, you could have played
the game either one or two times. There are therefore the following
possible outcomes at this point in time:19
Outcome Probability Net payoff
W 0.486 $1
LW 0.250 $1
LL 0.264 −$3
19
Probability values are rounded to the nearest thousandth. To calculate these
values: the first probability is 18/37, the probability of winning in one round of
roulette. The second is (19/37)×(18/37), which is the probability of losing in the
first play times the probability of winning on the second try given a loss the first
time. The third is (19/37)×(19/37), the probability of losing twice in a row.
The first row of the table represents the possibility that you win in the first
round, in which case you quit. The second row represents the possibility
that you lost, then won, then quit. The third row represents the possibility
that you have so far lost twice in a row, in which case you lost $1, then $2,
for a total loss of $3. Your expected profit at this point in time is therefore
( .486)(1) ( 0. )( ) ( .264)( −3)

3)) .055.
Now consider the situation three minutes after you started playing.
By this time, you might have attained any of the following possible
outcomes:
Outcome Probability Net payoff

W 0.486 $1
LW 0.250 $1
LLW 0.128 $1
LLL 0.135 −$7
Your expected profit at this time is therefore
( .486)(1) ( 0. )( ) ( .128)(1) ( 0. )( ) 0.083.
Notice that this is not only negative; it is less than your expected profit
after only two minutes. This is a general point: the expected profit at
time t is always negative, and it only gets worse, approaching −∞ as t
increases. After four minutes, it is −0.113; after five minutes, −0.143;
and so on. There will never come a point in time at which the expected
profit of a follower of the Martingale betting system turns positive.
Thus, the impression that the Martingale betting system is a rational
strategy, or that it enables one to circumvent the house’s advantage, is
an illusion produced by the failure to consider the time required to play
the game. Since you cannot ever have played infinitely many rounds,
there is never a time at which all logically possible outcomes of playing
the game are available to you. The last available outcome at any given
time is always the one in which you’ve so far lost every time, and this
outcome gets worse and worse.
12.17 The delayed heaven paradox
12.17.1 The multi-choice version

In the Delayed Heaven Paradox, you are repeatedly offered the chance
to accept one day in Hell in exchange for two days in Heaven, where
all the Hell days must be served before any Heaven days. It is always
rational to accept, so you wind up accepting infinitely many days in
Hell – the worst possible outcome. The scenario poses a challenge for the
plausible principle that if each member of a series of decisions is good/
prudent, given all the choices that preceded it, then the series as a whole
is good/prudent.
This paradox can be resolved in a similar manner to the Saint
Petersburg Paradox (see Sections 12.15.3–12.15.4). At the beginning of
each day n, you have 2(n−1) days in heaven already banked, and you are
asked to choose whether to take the days in heaven, or to take another
day in hell in order to bank another two days in heaven. In reasoning
about the choice situation, we have assumed that you always believe,
with certainty, that you will really get all of the banked days in heaven
when you finally choose to take them.
We could relax this assumption – we could suppose that you merely
believe with 90% confidence that you will get the banked days when
you choose to take them. But this is still insufficiently realistic. To take
account of the arguments in Section 12.15.3 above, we should assume
that the probability that you will ever really get all of the banked days
diminishes as the number of banked days increases. That is, if God (or
someone claiming to be God) has promised you n days in heaven, the
probability that He will keep that promise diminishes with n. It is much
less likely that you will get one trillion years in heaven (if promised)
than it is that you will get two days in heaven (if promised).
More to the point, each time you are offered the choice, the cost you
are asked to take on (another day in Hell) is imminent – that is, you
are to go to Hell immediately – but the promised benefit (an extra two
days in Heaven) becomes increasingly remote. Thus, the first time, the
promised benefit is just a day away. The second time you are asked to
choose, the promised benefit is three days away (since the extra two days
in heaven that you’re going to get would be enjoyed only after another
day in Hell followed by the two days in heaven that you already have
banked). The third time, the benefit is five days away. And so on. The
nth time, the benefit is (2n−1) days away. Because the promised benefit
becomes increasingly remote, it becomes progressively less likely that it
would actually be received. Thus, at some point, it becomes rational to
reject the trade.
12.17.2 The single-choice version

Here is a simple variation of the paradox: God asks you to name a
cardinal number, with the promise that whatever number you name,
you will get that many days in Hell, followed by twice as many days
in Heaven. For present purposes, assume that God will accept Յ0 as a
cardinal number.
What number should you name? For every natural number n , it is
better to pick n +1 than to pick n , because the benefit (two more days
in Heaven) outweighs the cost (one more day in Hell). In general,
for every natural number n , you should not pick n, because it would
be better to pick a number larger than n . But nor should you pick
Յ0, since you will then spend an eternity in Hell. (You would also
be entitled to spend twice eternity in Heaven, but only after the
eternity in Hell was over, which would never happen.) This differs
from the previous version of the paradox in that in this version you
are asked to directly decide, in a single choice, how many days you
spend in heaven, whereas in the previous version, you determine
how many days to spend in heaven by a series of distinct choices.
This version raises a challenge for the principle that in any given
choice situation, there is always a best choice (at least a choice tied
for best).
Again, the resolution of the paradox is reminiscent of the resolution of
the Saint Petersburg Paradox. In the single-choice version of the Delayed
Heaven Paradox, we assume that you accept a promise from God, to the
effect that for every n, if you name n, then God will give you n days in
Hell followed by 2n days in Heaven. This can be thought of as an infinite
collection of promises:
P1 If you choose 1, then God will give you 1 day in Hell followed
by 2 days in Heaven.
P2 If you choose 2, then God will give you 2 days in Hell followed
by 4 days in Heaven.
⁝
Pn If you choose n, then God will give you n days in Hell followed
by 2n days in Heaven.
⁝
But for the reasons discussed earlier (Section 12.15.3), the probability
that Pn would be kept diminishes with n. As a result, it is not always
better to pick a larger number rather than a smaller one. On any rational
probability distribution, there will be some optimal number to pick,
such that larger numbers are worse choices due to the lower probability
that God would keep his promise.
12.18 Comment: shallow and deep impossibilities
12.18.1 Solving paradoxes by appeal to impossibility

The paradoxes we have discussed involve scenarios in which some sort
of infinitude makes an appearance, and some seemingly absurd impli-
cation or unanswerable question results. One way of solving such a
paradox is to explain how the scenario is impossible. Of course, it is not
sufficient merely to point to the absurd implication or unanswerable
question as proof of impossibility. Rather, a solution should explain
how the absurd implication or unanswerable question results from
a more fundamental impossibility implicit in the description of the
scenario.
But as noted earlier (Section 7.4), there are stronger and weaker kinds
of impossibility. In what way must a paradox-generating scenario be
found ‘impossible’ in order for the paradox to be resolved?
We can first note that some kinds of impossibility would be irrelevant;
that is, some reasons for counting a scenario impossible would fail to
resolve a paradox. Take the scenario of Thomson’s Lamp, which leads
to the unanswerable question of whether the lamp would be on or off
at the end of the series of switchings. Suppose I could argue, completely
convincingly, that (i) a budget of at least $200 trillion would be required
to construct a lamp such as the one Thomson describes, and (ii) no
one would ever be able to marshal so many resources for such a frivo-
lous purpose. This would show in some sense that the Thomson Lamp
scenario is impossible – no one is in a position to bring it about. But this
obviously would not count as a solution to the paradox. We might say:
this sort of impossibility is too shallow. We need a stronger, deeper sense
in which the Thomson Lamp is impossible.
Even physical impossibility may not be enough. Suppose I had an
argument that the Thomson Lamp was physically impossible, where my
argument turned on precise estimates of the value of Coulomb’s constant
(the proportionality constant that relates electrostatic force to charge
and distance). (Don’t ask how such an argument would go.) Suppose
that if Coulomb’s constant were 10% larger than it actually is, then my
argument would fail and the Thomson Lamp would be constructible; yet
given the constant’s actual value, the Lamp is unconstructible. Again,
we would surely say that I had not solved the paradox in any intellectu-
ally satisfying sense. Plausibly, the reason would be that it would remain
sensible to wonder something like this: ‘If Coulomb’s constant had been
10% larger, and the Thomson Lamp had been constructed, then what
state would the lamp have been in at the end of the infinite series of
switchings?’
By contrast, consider a solution that would be more satisfying:
suppose I argue that the scenario is impossible because it is impos-
sible, in general, to complete an infinite series of actions, and the
reason this is impossible is that an infinite series is a series that has no
end, and it is conceptually impossible to complete a series that has no
end. I take it that this impossibility claim is of the right kind to solve
the paradox. This solution fails, not because it appeals to a shallow
notion of impossibility, but because its impossibility claim is simply
false (see Section 12.8.1).
Why should a paradox be resolved by a ‘deep’ impossibility but not
by a ‘shallow’ impossibility? There are two important points to make
here. First, a deep impossibility renders absurd implications expectable,
whereas a shallow impossibility does not. If we fail to construct a certain
device due to budgetary limitations, we may still be mystified by an argu-
ment showing that if we had constructed the device, some seemingly
absurd consequence would have resulted. But if a certain device fails
to exist due to some deep conceptual, logical, or metaphysical impossi-
bility, then we do not feel mystified by a showing that the device’s exist-
ence would entail (further) absurd consequences. We should expect, in
particular, that a metaphysically impossible scenario would have meta-
physically impossible consequences.
Second, a deep impossibility may render certain counterfactual ques-
tions – questions about what would happen if the scenario obtained –
moot, either in the sense that the questions would be uninteresting or
in the sense that they would lack determinate answers. Consider the
question, ‘If two were equal to three, how much would seven times nine
be?’ This question is rendered uninteresting by the fact that two really
could not equal three. The supposition is sufficiently absurd that there
seems no point to even formulating it. In addition, the question seems
to lack any determinate answer.
Similarly, if the Thomson Lamp is impossible in a deep sense, we may
think that there is no point to asking what would happen if the lamp
were to exist, and that there need be no determinate answer to such a
question.
12.18.2 An objection
All this is by way of introduction to an objection. Some philosophers may
be tempted to complain that my solutions to the paradoxes of infinity
appeal to mere physical impossibilities, not metaphysical impossibilities,
and that a genuine solution must show the problematic scenarios to be

metaphysically impossible.
I have been at pains to emphasize that infinite natural intensive
magnitudes are not ‘merely physically impossible’; they are metaphys-
ically impossible, because they require there to exist a kind of number
that does not exist. Compare the strong sense in which it is impossible
to have more than four cats while also having fewer than three – this is
deeply impossible because there is no number greater than four and less
than three. Similarly, an infinite natural intensive magnitude is impos-
sible because there is no infinite number.
Some will be tempted to press the objection nonetheless. Several of
my solutions to the paradoxes take the form:
1. Scenario S requires magnitude M to take on an infinite value.

2. M is a natural intensive magnitude.
3. No natural intensive magnitude can assume an infinite value.
4. So S is (metaphysically) impossible.
Having found S metaphysically impossible, I conclude that there is no

need to concern ourselves with what would happen if S occurred. The
objector might concede that (2) and (3) are each metaphysically neces-
sary, yet challenge whether (1) is metaphysically necessary.
Take the case of Thomson’s Lamp, which I claim cannot be switched
infinitely many times in a finite period due to the impossibility of infinite
material strength (among other things). One might object that it is only
a contingent truth of our physics that the infinite switching would
require the switch to have infinite material strength. Perhaps there is
some possible world with a different physics, in which a switch can be
switched infinitely many times in a finite period, while possessing only
finite material strength.
Compare the case where an alleged solution depends on the specific
value of Coulomb’s constant. In that case, we judged that the solution
would not succeed, since we could still wonder what would happen if
Coulomb’s constant were larger. Similarly, since my solution to Thomson’s
Lamp Paradox depends upon the laws of physics that happen to obtain in
our world, one could still wonder what would happen if those laws were
different; thus, my solution does not succeed – or so one might argue.
12.18.3 Deep physical impossibilities

I think that the preceding objection presupposes a sharper division
between physical and metaphysical impossibilities than really exists,
and it wrongly treats all physical impossibilities as equally shallow. Some

‘physical impossibilities’ are sufficiently deep to resolve paradoxes.
The specific value of Coulomb’s constant seems like a shallow physical
necessity. One can easily imagine the constant having a different value,
and one can predict what would result.
By contrast, the physical principles used in my resolution of the
Thomson Lamp paradox are much deeper principles; it is much harder to
imagine their being false or to evaluate counterfactuals about what would
happen if they were false. The objection requires us to suppose an alter-
nate physics in which a switch can be flipped faster and faster without
limit, and yet the switch need not have infinite material strength. It
really is extremely unclear how we are supposed to imagine this working.
Are we to suppose a physics in which an object can somehow be acceler-
ated faster and faster without limit, without increasing the force applied
to it? Or in which an object withstands ever-increasing forces despite
having only limited material strength?
There are three related problems here. First, it isn’t clear that such alter-
native physics are genuinely metaphysically possible. There really is not
a clear intuitive division between the principles that are generally recog-
nized as metaphysically necessary – such as that nothing can be both
completely red and completely blue, or that the shortest path between
two points is a straight line – and the very general, deep physical prin-
ciples involved in this example – such as that moving an object with
increasing acceleration requires applying increasing force, or that greater
material strength is required for an object to withstand greater forces.
Second, even if an alternate physics that rejects these principles is
in some sense possible, the injunction to imagine that physics were
somehow altered so as to allow the Thomson Lamp to exist is not suffi-
ciently determinate to generate answers to counterfactual questions. If,
that is, someone wants us to imagine a radically alien physics, and then
say what would happen to a lamp in a world with this alien physics, the
person proposing this scenario would have to tell us much more about
what the physics of this hypothetical world would be like. Otherwise,
there will be no answer to such questions as ‘At the end of one minute,
what state would the lamp be in?’ The absence of a determinate answer
to this question would be no more mystifying than the absence of deter-
minate answers to such questions as ‘If the sky were some other color
than what it is, what color would it be?’ or ‘If someone else were writing
this chapter, what would that person say?’
Third, even if someone were to describe a specific, alternative physics
in which it is possible for a lamp to be switched infinitely many times
within a minute, such that it was plausible that there should be a deter-
minate answer to what state the lamp would be in at the end of that
process, this would simply be to describe a different problem from the
original one. Compare this pair of questions:
a. If I were to add a teaspoon of salt to this recipe, how would it taste?

b. If I were to add a teaspoon of salt to this recipe, in an alternate possible
world in which salt is a compound of plutonium and mercury and
we are sea creatures who evolved living on kelp and plankton, how
would it taste?
Whatever might be said about (b), it is surely a different question from

(a), and one’s answer to question (a) is not impugned by one’s failure to
answer (b). Similarly, the original Thomson Lamp problem is different
from the problem that someone might one day pose in which we have
to work out the implications of an alien physics (in which perhaps
forces are unrelated to accelerations, etc.). It is hardly an indictment of
a theory that it fails to provide answers to hypothetical problems that
have yet to be clearly posed. It is enough that the original, reasonably
clear version of the problem has been addressed.
12.18.4 Generalizing
I have stated the above points in terms of the example of Thomson’s
Lamp, but I believe similar remarks apply to the other solutions that
rely on general physical principles. In no case do I rely upon shallow
or seemingly arbitrary physical principles, such as the specific value of
a physical constant. In each case, the physical principles I employ are
general ones that are essential to the nature of the physical phenomena
of interest – for instance, principles concerning binding energy are
essential to the distinction between solid and non-solid materials
(Section 12.9); the relationship between resistivity and current flow is
essential to the nature of electric circuits (Section 12.10.4); the Beer-
Lambert Law is essential to the nature of transparency and opacity
(Section 12.12.3). To suppose these principles false is to suppose a
fundamentally alien physics – perhaps even one with alien proper-
ties, not really solidity, electricity, and opacity as we understand those
terms.
It is really not so clear that one could describe a coherent, metaphysi-
cally possible physics without these principles. Even if that would be
possible, counterfactual questions about what would happen in an
alternate world in which these fundamental physical principles failed
probably do not have determinate answers – at least, not until one speci-
fies in some detail the alternative physics that would obtain. If someone
did so, this would really be introducing a different set of puzzles, rather
than impugning my solutions to the original paradoxes, as those para-
doxes were posed in Chapter 3.
13
Assessing Infinite Regress
Arguments
13.1 The problem of identifying vicious regresses
A common form of argument in philosophy claims that some philo-

sophical thesis, T, generates an infinite regress, whereupon the reader is
supposed to see that T must therefore be rejected (see Chapter 2). Such
arguments often prove controversial, and the controversy usually centers
on whether the infinite regress in question is vicious or benign. In this
chapter, I offer an account of when an infinite regress is vicious, which I
deploy to assess the six infinite regresses discussed in Chapter 2.
To review, in Chapter 2 we discussed the following six infinite regresses
and/or regress arguments:
1. The Cosmological Argument (concerning the Regress of Causes): This argu-

ment claims that because an infinite regress of causes is impossible,
the universe must have a first cause, that is, something that was not
itself caused by anything but that caused everything else.
2. The Regress Argument for Foundationalism (the Regress of Reasons): This
argument claims that because there cannot be an infinite series of
reasons for any belief, there must be some beliefs that are justified in
a way that does not depend on reasons. These ‘foundational’ beliefs
would be the source of the justification for all other justified beliefs.
3. The Third Man Argument (the Regress of Forms): This argument claims
that Plato’s theory of the Forms should be rejected because it entails
that there is an infinite series of Forms for any property – for example,
that there is an infinite series of Forms of Man.
4. The Regress Argument against Resemblance Nominalism (the Regress of
Resemblances): This argument claims that resemblance nominalism
should be rejected because it entails the existence of an infinite series
of resemblance relations.
229
5. McTaggart’s Argument against the Reality of Time (the Regress of Temporal

Properties): This argument claims that time must be unreal, because
the reality of time requires an infinite series of ever more complex
temporal properties (for example, the presentness of e, the present-
ness of the presentness of e, and so on).
6. The Truth Regress: P entails that it is true that P, which in turn entails
that it is true that it is true that P, and so on. Unlike the previous
cases, no one claims that this regress is vicious, so no infinite regress
argument has been advanced based on this regress.
In the first five of these cases, an important philosophical argument

hinges on the claim that the infinite regress in question is ‘vicious’. But
those who advance these arguments are rarely clear about what would
make a regress vicious.
There seem to be at least three main ways that an infinite regress might
be problematic (perhaps there are more, but these are the three obvious
ways). The three problems are: metaphysical impossibility, extreme
implausibility, and explanatory failure.1 Let us discuss these in turn.
13.2 Viciousness through metaphysical impossibility
Let us begin with the problem of metaphysical impossibility. This is the

alleged problem that motivates the Cosmological Argument. It is said
that an infinite regress of causes is metaphysically impossible, either
because it is in principle impossible to complete an infinite series, or
because it is impossible for there to be an ‘actual infinity’.
My account of metaphysically impossible infinities is by now familiar;
I have nothing substantive to add to it here. On my account, an infinite
regress would be vicious by virtue of a metaphysically impossible form
of infinity only if it somehow implicated an infinite natural intensive
magnitude. No such impossibility is involved in the regress of causes.
The infinite series of causes would be an example of infinite cardinality
(infinitely many causes). If we assume that each cause must have at least
some minimum duration and that the causes are sequentially arranged,
1
This taxonomy draws on the work of Nolan (2001) and Bliss (2013). Nolan
recognizes four problems that infinite regresses may have: contradiction, failure
of intended reduction, conflict with the known finitude of a domain, and quan-
titative extravagance. The first corresponds (but only loosely) to my category
of metaphysical impossibility, the second to my category of explanatory failure,
and the third and fourth to my category of implausibility. Bliss emphasizes the
problem of explanatory failure.
Assessing Infinite Regress Arguments 231
or if we simply assume that there is no beginning of time, then we also

have an example of an infinite extensive magnitude (the duration of the
universe’s history). On my view, neither of these things is problematic,
and there is no need for any intensive magnitude to be infinite. In short,
the Cosmological Argument fails when understood as an appeal to meta-
physical impossibility.
Similar remarks apply to the regresses of Forms, resemblances, temporal
properties, and truths. Each of these, like any infinite series, involves an
infinite cardinality; none involve an infinite intensive magnitude.
The status of the regress of reasons is open to debate. Suppose we accepted
the following principles about reasons for belief: (i) all justification for belief
is provided by reasons; (ii) a reason provides justification for a belief only
if it is itself a justified belief. These principles generate an infinite regress of
reasons, all of which must be justified beliefs. It is plausible to think that,
for a subject to have all of these justified beliefs, the subject must instan-
tiate some infinite intensive magnitude – for instance, the person must
have infinite intelligence, or infinite mental capacity in some sense, where
this mental capacity would be an intensive magnitude.
Here is one way to avoid that conclusion. First we must distinguish
dispositional from occurrent beliefs: a belief is ‘occurrent’ at a given
time iff the subject is actually thinking that belief at the moment.
‘Dispositional’ beliefs are those that one is not thinking at the moment.
For example, most readers, I presume, believe that seven is greater
than four. Before reading that sentence, you probably believed [seven
is greater than four] dispositionally; upon reading the sentence, your
belief became occurrent. Now, to have an infinite collection of occur-
rent beliefs, one might well need to possess some sort of infinite mental
capacity of a sort that would be metaphysically impossible.
However, to have an infinite collection of merely dispositional beliefs
may not require one to have an infinite mental capacity. Note that some
dispositional beliefs seem to be things that a subject has never expli-
citly, occurrently thought; instead, they seem to be merely things that
are obviously entailed by other things the subject believes (where the
subject would immediately accept the entailment upon consideration).
For instance, a normal person might count as believing that 6,387 is
greater than 1, despite never having thought that specific proposition,
simply because the person believes things about the number system that
obviously entail that 6,387 is greater than 1. If we accept this sort of
dispositional belief, it then becomes plausible that one could have infin-
itely many dispositional beliefs without manifesting an infinite mental
capacity in any problematic sense.
However, it is highly implausible that such beliefs could be related in

the right way to provide an infinite series of justifying reasons for any
proposition. Suppose I believe an infinite series of propositions, P1, P2,
P3, ... , where all of the propositions after a certain point are believed
only dispositionally, and only in the sense that they are obvious logical
consequences of earlier propositions in the series that I believe. It just
is not plausible that in such a case, these propositions might be related
in such a way that for every n, Pn+1 provides a non-question-begging
reason for Pn.
Now here is another way to avoid the requirement of an infinite
mental capacity while still embracing an infinite series of reasons in some
sense. We could hold that for a belief to be justified, it is only required
that the subject could come up with a reason (which would itself then
be justifiable), not that the subject presently has such a reason among
his beliefs.2 That is, justifying reasons need only be potential beliefs,
not actual beliefs. On this view, it would be plausible to claim that no
infinite intelligence, or other infinite mental capacity, is required in
order to have justified beliefs. To be sure, the view might be implaus-
ible for other reasons: if a belief needs a reason in order to be justified,
then I think it implausible that the mere potential to come up with a
reason, which the subject does not actually believe, would suffice for
justification. But the view does not suffer from the particular problem of
invoking a metaphysically impossible kind of infinity.
13.3 Viciousness through implausibility
13.3.1 The regress of reasons

Sometimes, we find that a certain infinite series, even if it is not strictly,
metaphysically impossible, is extremely implausible. This is the usual
charge against the infinite regress of reasons. Perhaps there could be a
being with an infinite series of reasons for one of its beliefs; perhaps not.
Be that as it may, it is extremely, empirically implausible that humans
are such beings. Even among those few philosophers who defend
2
It is unclear whether Peter Klein’s well-known infinitist view is to be under-
stood thus, as requiring only the ability to devise a reason, or rather as requiring
an actual (albeit dispositional) belief that serves as a reason. Klein says only that
a reason must be a proposition that is ‘available’ to the subject, and that to be
thus available, a proposition need not be occurrently believed, but that it must be
‘appropriately “hooked up” to S’s beliefs and other mental contents’ – thus leaving
it open whether the proposition must be dispositionally believed (2005, 136).
the possibility of an infinite series of reasons, none has provided any

examples to show how such a series would go.
In fairness, publisher-imposed length limits prevent a philosopher
from stating an infinitely long argument. If, however, someone were
to state even the first fifty steps in the infinitely long chain of reasons
that justifies the proposition [I exist], this would go a long way toward
convincing me that there might be such an infinite chain. No one has
done anything like that; indeed, no one seems able to provide even the
first ten steps.
This sort of extreme implausibility applies to most (perhaps all) other
infinite regresses that involve human beings. For instance, suppose one
held that in order for any choice to be truly free, the agent must freely
choose the motives for which the original choice was made. This leads
to an infinite regress of choices, and whether or not it is metaphysically
possible, it is extremely implausible that any human being ever performs
such an infinite series of choices. Similarly for the idea that all words, to be
understood, must be defined in terms of other words that are understood.
13.3.2 Regresses of abstract objects

Infinite regresses that involve only abstract objects are generally much
less implausible. Witness the truth regress, which almost everyone accepts
as benign. There is an infinite series of true propositions, but no human
(or other) being need entertain all of them, so there is not the same kind
of empirically implausible demand as there is in the case of an infinite
series of reasons or choices. Abstract objects in general have a habit of
being infinitely numerous, as in the cases of numbers, sets (if such exist),
geometric points (if such exist), properties, and propositions.
In my judgment, the infinite regresses of resemblances and temporal
properties are similarly untroubling. In the case of the regress of Forms,
I think it implausible that the series of Forms exists, but this implaus-
ibility is not something about the infinitude of the series. Rather, even
the existence of a second Form of Man strikes me as extremely implaus-
ible and as betokening some confusion in the theory. Of course, some
find the very idea of the Forms implausible from the start. But even for
those who are initially willing to entertain the idea, surely by the time
the second Form of Man arrives on the scene, it is clear that something
has gone off the rails.3
3
Of course, what has gone wrong is (perhaps among other things) the self-
predication assumption: the idea that a property possesses itself – for example,
that the property of Manhood is itself a man – is extremely confused.
13.3.3 The regress of causes

What about the regress of causes addressed by the Cosmological
Argument? I detect no similar implausibility here. The idea that the
chain of causes and effects in the universe’s history stretches back infi-
nitely into the past – once we have overcome the mistaken view that
an ‘actual infinity’ is impossible or that an infinite series can never be
completed – just does not seem to reveal any sort of confusion in the
way that the idea of there being more than one Form of Man reveals
a confusion. It is also not empirically implausible in the way that it is
implausible to ascribe an infinite series of reasons to a human being.
But one might think that the regress of causes is improbable because
of its ‘quantitative extravagance’ – that is, because it so blatantly violates
Occam’s Razor, the principle that other things being equal, simpler
explanations are to be preferred over more complex explanations. To
explain the current state of the universe, or to explain any current event,
we postulate an infinite number of past states or events: this seems
to be the least simple sort of explanation one could give. It would be
much simpler if we could cite some entity, such as God, that ultimately
explains all the events in the universe.
I think this reasoning, though superficially plausible, is incorrect for
at least two reasons. The first reason is that the theory of an infinite
past (that is, a chain of past causes and effects with an infinite duration)
does not actually require a greater number of entities than the theory
of a finite past. This is because on any reasonable view, there are infin-
itely many events in the history of the universe, even if we look at only
a finite duration. This is because time is infinitely divisible, as are the
processes that take place in time. If, for example, an object moves from
point A to point B, this motion can, with equal scientific validity, be
regarded either as a single event or as an infinite series of events (for
instance, moving the first half of the distance, moving the next quarter
of the distance, and so on). If the latter seems to you an arbitrary way of
dividing up the motion, keep in mind that all ways of dividing up the
universe’s history into events are equally arbitrary from the standpoint
of physics. When one speaks of an object moving from point A to point
B, this just signals that one has chosen two points in the history of the
object – the point when the object was at A and the point when it was at
B – as the boundaries of an ‘event’; one could equally well have chosen
an earlier or a later stage of the object’s history. So we have infinitely
many events, whether the universe’s history is finite or infinite. On the
other hand, if we postulate a supernatural entity, God, as the cause of
the rest of the universe, we will be introducing a new kind of entity not
present in a purely naturalistic worldview; thus, in one respect we will
be making our theory more complex.
My second point is more interesting. It is that the argument from
quantitative extravagance rests on a confusion about the theoretical
virtue of parsimony. I have discussed this issue at greater length else-
where; here I just give a very brief, dogmatic statement of my view.4
It is frequently but not always true that the simpler explanation of a
phenomenon is more likely to be true. But most who invoke this prin-
ciple have no account of why this should be; as a result, they also have
no idea of when the principle applies.
Here is why simpler theories tend to be better than more complex
theories. A simpler theory, typically, has fewer adjustable parameters than
a more complex theory. These ‘adjustable parameters’ are places where
assumptions of the theory could be varied, without thereby making it into
a different theory; for example, the gravitational constant is an adjust-
able parameter in Newtonian gravitational theory. Because a complex
theory has more adjustable parameters, it is typically capable in principle
of accommodating a wider range of possible data than a simple theory.
As a result, the complex theory makes weaker predictions than the simple
theory. In probabilistic terms, what this means is that if theories C and
S both accommodate some evidence E, and C is more complex than E,
then typically the probability of E given S is higher than the probability
of E given C (P(E|S) > P(E|C)). Therefore, if and when E is discovered to be
true, it acts as a stronger confirmation of S than of C.
Thus, suppose we are trying to fit a curve to some number of data
points. If the data can be accommodated either by the hypothesis of a
linear relation (an equation of the form y = Ax + B) or by the hypoth-
esis of a cubic relation (an equation of the form y = Ax3 + Bx2 + Cx + D),
the former is more likely to be correct. The reason is that the range of
data that can be accommodated by a linear relation is a very tiny subset
of the range of data that can be accommodated by a cubic relation.
Therefore, intuitively, if the true relationship between x and y is cubic,
then it would be an amazing coincidence that the data happen also to
fit a linear relation. Essentially this sort of reasoning, I maintain, can be
given for all the other clear cases in which a simpler theory is more likely
to be correct than a complex theory.
Notice that on this account, the virtue of parsimony does not mean
that a simple theory has a higher a priori probability than a complex
4
For elaboration, see Huemer 2009.
theory. Rather, the virtue of parsimony consists in the fact that a simple
theory is more easily confirmed than a complex theory – that is, when
both theories accommodate some data, that data will provide more
support to the simple theory than to the complex theory. Thus, the
virtue of parsimony only applies when we have two theories, differing
in complexity, that both accommodate some data, and the question is
which theory is better supported by that data. Notice also that on this
account, the virtue of parsimony only applies in cases where the more
complex theory accommodates a wider range of possible data than the
simpler theory. On this showing, nearly all appeals to simplicity in phil-
osophy are fallacious – almost never can a philosophical theory lay claim
to accommodating a narrower range of possible data than its rivals. Most
philosophical theories in fact accommodate either every metaphysically
possible data set, or none of them.
Be that as it may, the appeal to simplicity fails, in particular, in the case
of the Cosmological Argument. The theory that God created the universe
does not accommodate a narrower range of data than the theory that
the universe always existed. No matter what we had observed (given that
we could not have observed the infinitude of the universe’s history), it
would have been consistent with the hypothesis of a universe created by
God. Proponents of the first-cause theory cannot explain what different
evidence we might have expected to observe if there had been an infi-
nite series of causes rather than a first cause. It thus appears that we have
again failed to find anything vicious about the regress of causes.
13.4 Viciousness through explanatory failure
13.4.1 Two fictitious scientific theories

The third way that an infinite regress can be vicious is that it indicates a
theory’s failure to explain what it is intended to explain. Following are
two examples of this problem.
First, the homuncular theory of perception: this theory holds that for
a person to see an object is for the person to have an internal image,
which is seen by a tiny person inside the first person’s head.
Leave aside the general ridiculousness of a tiny person inside your
head, and focus instead on the infinite regress: the homunculus inside
the first observer’s head will need another homunculus inside its head,
and so on. This regress arises out of the same feature of the theory that
prevents the theory from being genuinely explanatory, namely, that the
thing to be explained (vision) reappears within the proposed explanation.
Notice that the problem is not merely that it is impossible for a person to
contain an infinite series of persons. Even if we accepted that there might

be an infinite series of persons within a given person, the homuncular
theory still would not explain vision; it tells us only that whenever vision
occurs, an infinite series of instances of vision occur. Compare this case:
suppose I tell you that to flub is to have another person inside of you
who flubs. Even if you believe that there could be an infinite series of
persons inside you, this still does not tell you what flubbing is.
My second example concerns a certain chelonian theory of cosmology.
According to a popular story, a philosopher (sometimes identified as
Bertrand Russell, sometimes William James) once gave a lecture on
astronomy, describing how the Earth flies through space in its orbit
around the sun. After the lecture, an old woman told the speaker that
his theory was rubbish; in fact, she maintained, the Earth rests upon
the back of a giant turtle. And what supports the turtle? the speaker
wondered. Another giant turtle. But what holds up the last turtle in the
series? Answer: ‘It’s turtles all the way down.’
What is wrong with such a theory? Leave aside the modern scientific
evidence and the general implausibility of a world turtle. Leave aside
also the questions of why any explanation is needed for why the Earth
does not fall, and how we know that the Earth is not presently falling.
Focus instead on the theory’s explanatory strategy: the theory aims to
explain why the Earth does not fall. It postulates a World Turtle under-
neath the Earth, but the World Turtle itself must be supposed to have
the very same property (that of remaining suspended without falling)
that we wanted explained for the Earth. So the postulation of the World
Turtle does not seem to make any explanatory advance at all.
But what if there were infinitely many turtles – wouldn’t this at last
explain the Earth’s stationariness, if only we could believe in such a
chelonian series? Not really. For even if there were an infinite series of
world turtles, it could just as easily be an infinite series of falling turtles
as an infinite series of stationary turtles. Some explanation is thus needed
for why the whole system, including the Earth together with all the
turtles, remains suspended rather than falling.
No one holds either the homuncular theory of perception or the
chelonian theory of cosmology, so refuting these theories is not particu-
larly important per se. What is of interest is the types of explanatory
failure that these theories exhibit. When we come to considering more
controversial theories, we must be on guard against these types of
explanatory failure, namely, those in which the explanans contains the
explanandum, or in which the infinite series as a whole stands in need
of explanation in the same way as its first member did.
13.4.2 The regress of causes

Proponents of the Cosmological Argument believe that the theory of
an infinite series of causes suffers from the same sort of problems as
the theories of the homunculi and the world turtles. To begin with, we
seek an explanation for the occurrence of some present event. This is
explained by citing its cause, which is another, earlier event. But this
earlier event must also have occurred; thus, the same question arises
with it as arose with the original event. This seems analogous to citing
a World Turtle that stands just as much in need of support as the Earth.
And just as we can ask why the entire series of world turtles does not
fall, we can ask why the entire series of causes and effects in the world’s
history occurred, rather than none of it occurring.5
I think there is something to these observations, but the argument is
not as strong as its proponents make it out to be. The regress of causes is
not as bad as the homuncular theory of perception. For as we described
it, the aim of the homuncular theory was to explain what vision is. The
aim of the theory of the regress of causes, however, is not to explain what
an event is, or what causation is, or what anything else is. In both cases,
there is a failure to explain the nature of something: the homuncular
theory fails to explain what vision is; just so, the causal regress theory
fails to explain what events are and what causation is. But since the
causal regress theory was never intended to explain those things, this is
not a failure of the theory.
The causal regress theory is meant to explain, for each event, why that
event occurred. And the theory succeeds on that score – or at least, if
one succeeds in identifying the cause of a given event, one succeeds in
explaining (causally) why that event occurred. The theory fails, however,
to explain why the entire series occurred rather than none of the events in
the series occurring. This question is a slightly more specific version of
Leibniz’ famous metaphysical question, ‘Why is there something rather
than nothing?’6 Leibniz and other proponents of the Cosmological
Argument are correct so far: the infinite series of causes fails to answer
5
Clarke 1998, section 2, 10.
6
From ‘The Principles of Nature and of Grace, Based on Reason’: ‘[N]ow we
must advance to metaphysics, making use of the great principle [ ... ] that nothing
happens without a sufficient reason; that is to say, that nothing happens without
its being possible for him who should sufficiently understand things, to give a
reason sufficient to determine why it is so and not otherwise. This principle laid
down, the first question which should rightly be asked, will be, Why is there some-
thing rather than nothing?’ ([1714] 1908, 303; emphasis in original)
Leibniz’ question, just as the turtle theory fails to explain what holds up
the entire series of turtles.
But the causal regress theory is in better shape than the turtle theory,
for at least two important reasons. First, the turtle theory must compete
with a rival theory, modern astronomy, that is supported by a great deal
of evidence, and this rival theory avoids the problem facing the turtle
theory (modern astronomy does not have to explain, for example, what
‘holds up’ the solar system). By contrast, the causal regress theory faces
a much weaker rival. The theory of theism is much less well supported
than modern astronomy, or so I would argue. The theistic theory also
fails to offer a coherent explanation of the fact that supposedly needs
explaining: if the question is why there is something rather than
nothing, then to say that God created the rest of the cosmos does not
answer the question, for it does not explain why there was God rather
than nothing.
Leibniz has an answer to this: God, he claims, is a ‘necessary being’,
that is, a being who could not possibly have failed to exist. Thus, unlike
the universe, there is no need to wonder why God exists. Assessing this
claim in detail would take us too far afield. Here, I will baldly assert my
view that the idea of God as a necessary being is incoherent and that it
rests on the infamous Ontological Argument, one of the most patent
sophisms in the history of philosophy.7
Be that as it may, the current version of the Cosmological Argument
does something akin to begging the question. If we already accept the
Ontological Argument for the existence of God, then we don’t need
the Cosmological Argument, because we’ve already established the
conclusion that God exists. If, on the other hand, we do not accept
the Ontological Argument, then we should not accept the Cosmological
Argument either (in the version currently under consideration), because
there is no reason to think that God would be any less in need of expla-
nation than the universe; the theistic theory thus could not claim an
explanatory advantage over the causal regress theory.
The second reason why the causal regress theory is in better shape
than the turtle theory is this: the turtle theory is only motivated to
begin with if one accepts some general principle, roughly to the effect
that all material things have a natural tendency to fall; only thus is
7
The Ontological Argument claims, in essence, that God must exist because
existence is built into the definition of ‘God’. For classic statements, see Anselm
1926, Chapter 2; Descartes 1984, 5th Meditation. For a review of the literature,
including the problems with the argument, see Oppy 2014.
there a need to explain why the Earth is not falling. But if one accepts
that principle, there is no reason why the material object consisting of
the entire collection of turtles should not fall. (One might say: ‘Maybe
the whole system is falling; how would we know?’ Or: ‘Falling can only
be defined relative to other objects.’ Yes, but then there would be no
need to posit any of the turtles in the first place.) By contrast, the causal
regress theory need not be motivated by any analogous principle – one
need not, for example, think that events have a ‘natural tendency
to not happen’, nor need one accept the general claim, advanced by
Leibniz and many other theists, that everything has an explanation.
One can reasonably suppose that some things simply have no explan-
ation – including, perhaps, the fact that there is something rather than
nothing.
But, one might wonder, if one thinks that the universe as a whole
requires no explanation, then why posit any causes in the first place;
why not stop with the present state of the universe and claim that
it needs no explanation? I think the answer is, essentially, that we
have empirical reasons, not a priori reasons, to think that events in
the observable universe have causes. We have experience of many
regularities in nature that lead us to believe that there are causal laws,
and there are no confirmed cases of any events not governed by such
laws. It is reasonable to hypothesize that all events have always been
governed by these laws. Given the law of conservation of energy, it
is also reasonable to suppose that the energy existing in the universe
has always existed. We do not know for certain that these things are
the case, but they are reasonable hypotheses given our evidence. The
fact that we lack an explanation for why in general there is something
rather than nothing does not remove the evidence that we have for
the laws of nature.
My conclusion is that the regress of causes is not vicious by reason
of explanatory failure. It indeed fails to explain why the universe
as a whole exists, but the belief in such a series of causes was never
intended to explain that in the first place, and the belief is moti-
vated independently of any such explanatory claim. Nevertheless, if
a theistic account could explain why there is something rather than
nothing, this would be a definite advantage of theism. I rather doubt,
however, that any theory can explain why there is something rather
than nothing.8
8
For a review of some failed attempts to answer the question, see Kusch 1990.
13.4.3 The regress of reasons

The regress of epistemic reasons for belief seems to suffer from a problem
akin to that of the turtle regress. We start out wanting to know what
makes a particular belief justified. We are told that this belief is justified
by virtue of its being supported by an available reason. But it seems that
this reason could only explain the first belief’s justification if the reason
is itself justified. The reason, in turn, is supported by a second reason,
but this would only render the first reason justified if the second reason
was justified. So far, then, we would seem to have made no progress.9
Ah, but what if there were an infinite series of reasons? Then we could
see why the original belief was justified, couldn’t we?
Not really, no. Given an infinite series of reasons, there are two possi-
bilities: (i) each reason in the chain is justified because the next reason
in the chain is justified; (ii) each reason in the chain is unjustified
because the next reason in the chain is unjustified. This is analogous to
the fact that, in the case of turtle cosmology, we saw two possibilities:
an infinite series of stationary turtles, and an infinite series of falling
turtles. We would still need an explanation for why the whole chain of
reasons was a chain of justified reasons, rather than a chain of unjusti-
fied reasons. The theory of an infinite chain of reasons does not answer
the question of why any of one’s beliefs should be justified, rather than
all being unjustified. This is a troubling explanatory failure, because this
is precisely the sort of question that a theory about reasons for belief
should address.10 And it does not seem at all plausible – as it might have
been plausible to maintain that there is no explanation for the exist-
ence of the universe – to claim that there is no explanation for why our
9
Klein (2003, 722–3) addresses objections in this neighborhood, claiming
that the justification for the original belief is explained in terms of there being an
infinite series of available reasons, where one can give purely descriptive condi-
tions for a reason’s ‘availability’ (though he does not specify these conditions).
My reply: if the conditions for ‘availability’ do not include that a reason must be
justified in order to be ‘available’, then one’s having a series (whether finite or
infinite) of available reasons for a given belief does not explain why the belief
is justified, since one’s having this series is compatible with none of the reasons
being justified, in which case the original belief would not be justified.
10
Klein (2003, 729) seems to suggest that infinitism is only intended to
explain, for each belief, why that belief is justified, not to explain why any of our
beliefs are justified rather than all being unjustified. This is unsatisfying, since
rival theories (foundationalism, coherentism) do explain why any of our beliefs
are justified. If we must adopt some rival theory to account for why any of our
beliefs are justified, we should also use that theory to explain why any given
belief is justified.
belief systems are justified. The regress of reasons therefore appears to be

vicious by reason of explanatory failure.
13.4.4 The regress of universals

Our paradigm of a benign regress, the Truth Regress, does not suffer
from explanatory failure because it is not intended to explain anything
to begin with. More precisely, the thesis that P entails [P is true] indeed
fails to explain anything, but this is not a problem because that thesis
is motivated independently of any explanatory claim. We do not think
that P entails [P is true] because this somehow explains what ‘P’ means
or explains the nature of propositions. We think that P entails [P is true]
because this just follows from the meaning of the word ‘true’.
The regress of Platonic Forms, on the other hand, appears vicious. The
theory of Forms is apparently meant to explain what it is for a number
of objects to share a property or belong to a common kind. The theory
holds that this consists in the objects’ resembling or ‘participating in’
a Form, which itself has the relevant property or belongs to the rele-
vant kind, only in a more perfect manner than ordinary objects. This
explanans contains the explanandum: if we thought it needed explana-
tion how an ordinary object can have a property, or what it is to have a
property, then we should equally well demand an explanation of how
the Form can have a property. The regress of Forms is analogous to the
regress of homunculi: just as we fail to explain what vision is by invoking
an infinite series of acts of seeing, so we fail to explain what having a
property is by invoking an infinite series of objects with properties.
Would the problem go away if we simply dropped the self-predication
assumption? That is, suppose we give up the confused idea that the Form of
Man is itself a man; does this enable the theory of Forms to avoid a vicious
regress? Not really; it avoids the infinite regress of Forms of Man, but the
theory will still be saddled with another infinite regress: to explain what
makes a particular man a man, we postulate that the individual ‘partici-
pates in’ the Form of Man (whatever that means). But this ‘participation’
is itself a universal, since there can be multiple different instances of it.
Therefore, we must explain what it is for an individual to participate in a
Form: presumably, it is for the pair (the individual and the Form) to partici-
pate in the Form of Participation. And for this to occur is for the individual
and the first Form, together with the Form of Participation, to participate
again in the Form of Participation ... . This is not a satisfactory explanation.
What goes for Forms goes for Resemblances. If resemblance nominalism
is meant to explain what it is, in general, for something to have some prop-
erty (where relations count as properties), it fails because it requires a prior
understanding of what it is for a pair of objects to resemble, and this is an

example of the very thing the theory is supposed to explain. The hypoth-
esis of an infinite series of relations, even if we believe such a series exists,
does not explain what it is for a pair of things to stand in a relation.
At this point, one might well wonder: do all solutions to the Problem
of Universals similarly fail? If the problem is to explain predication, or
to explain in general what it is for a thing to be some way, then yes, all
solutions fail. For suppose this is our theory:
Predicate P applies to object x if and only if φ(P,x).
φ can be any condition, it doesn’t matter what. Some possible candi-

dates are:
φ(P,x) = x participates in the Form of P-ness.

φ(P,x) = x resembles the paradigms of P-ness at least as much as they
resemble each other.
φ(P,x) = x is a member of the set of P’s.
Whatever φ is, the statement of this condition will contain at least one
predicate (‘participates in’, ‘resembles’, ‘is a member of’, etc.). So an
understanding of predication will be presupposed in the theory.
There is therefore no solution to the ‘problem’ of universals, on this
interpretation of the problem. It is not possible to explain in general
what it is for something to be a certain way. But this is hardly troubling;
if anything qualifies as a basic notion, the notion of a thing being a
certain way seems to be one.
This does not mean that all theories about universals are false; it only
means that no such theory can explain predication in general, and thus
any such theory must be motivated by considerations other than the
claim to explain predication. For example, Platonism might be true
(as I believe), in the sense that universals exist necessarily. This thesis
would have to be motivated by reasons other than the claim to explain
predication. If it is so motivated, then the theory avoids vicious regress.
Whenever x is F, where x is some object and F is some property, there will
be an infinite series of facts:
x instantiates Fness.
〈x, Fness〉 instantiates Instantiation.
〈〈 x, Fness〉, Instantiation〉 instantiates Instantiation.
⁝
But as long as we don’t claim that any of this explains what it is to be F,

this regress is as benign as the truth regress.
13.4.5 The regress of temporal properties

According to McTaggart, the reality of time requires the existence of an
infinite series of temporal properties. For any given event, e, when e is
present, the presentness of e is also present, and so on. At first glance,
there is no case for explanatory failure here: the presentness of e is not
being used to explain e, nor is the presentness of the presentness of e
used to explain the presentness of e.
McTaggart might claim, however, that there is an explanatory task
that needs doing: the passage of time seems contradictory, he would
claim, because it involves a given event possessing two or more incom-
patible properties. We thus need to explain why the notion of the
passage of time is not contradictory. The proposed explanation is:
because the event possesses these incompatible properties at different
times. Allegedly, this introduces a second temporal series, about which
the same problem can be raised again. McTaggart’s picture seems to
be something like this: we start out with a contradiction (the same
event can be past, present, and future), and we have to introduce a
second time series to resolve the contradiction, but the second time
series introduces its own contradiction (similarly for a third time
series, and so on), so that there is ultimately no way of expunging all
contradictions.
McTaggart’s argument relies on the law of non-contradiction.
Understandably, he does not attempt to argue for the law of non-contra-
diction but instead takes it for granted as a starting premise. But how is
he understanding this law? Here are two readings of it:
The Qualified Law of Non-Contradiction: Nothing can have two or more

incompatible properties at the same time.
The Unqualified Law of Non-Contradiction: Nothing can have two or

more incompatible properties, regardless of whether they be possessed
at the same time or at different times.
If McTaggart is relying on the Qualified Law, then his argument does not
get off the ground. To claim that the idea of the passage of time involves
a contradiction, he must claim that, for time to pass, a given event must
possess incompatible properties at the same time. But there is no basis for
that claim.
If, on the other hand, McTaggart is relying on the Unqualified Law,

then his opponents (all those sane people who believe that time passes)
have a ready response. We should not respond by appealing to an infinite
series of temporal properties. We should simply reject the Unqualified
Law of Non-Contradiction. The Unqualified version of the law seems
obviously false, and McTaggart says nothing to support it.
Of course McTaggart’s argument does not succeed. If McTaggart were
correct in thinking that we start out with a burden of explaining away
a contradiction implicit in the notion of time, then I think he would
also be correct in thinking that the infinite regress of temporal prop-
erties is vicious. However, if there is no such initial burden, then the
regress of temporal properties is benign. And the defender of time in
fact faces no initial burden of explaining away a contradiction, since
either (a) the notion of a ‘contradiction’ includes cases in which a thing
possesses incompatible properties at different times, in which case we
should happily embrace the possibility of true ‘contradictions’, or (b)
‘contradiction’ does not cover such cases, in which case McTaggart in
fact never pointed out any contradiction in the first place.
13.5 Conclusion
What have we learned about assessing infinite regress arguments? When

faced with an infinite regress, there are three problems to look for:
First, metaphysical impossibility. This problem is vastly overdiagnosed,
owing to mistaken Aristotelian and Kantian notions about the impos-
sibility of actual or completed infinities. The metaphysically impossible
infinities are just those that require infinite natural, intensive magni-
tudes, and these hardly ever appear in philosophical arguments.
Second, implausibility. This problem is more likely to afflict theories
about some type of human action, state, or capacity (for example, theo-
ries about the nature of knowledge, intentional action, or free will) than
theories about more abstract matters, because human beings are notori-
ously limited.
Third, explanatory failure. This problem is the most difficult to diag-
nose. Proper diagnosis requires answering the following questions: (a) Is
there a valid explanatory demand? That is, can the theory under discus-
sion fairly be required to explain something, or can the theory reason-
ably be embraced even if it does not explain anything? (b) If there is a
valid explanatory demand, does the theory’s explanation generate an
infinite regress? (c) If so, does the original explanatory demand apply to
the infinite series as a whole? A special case of this is the case in which
a thing to be analyzed appears infinitely many times in the full anal-

ysis. In most cases, avoiding a charge of explanatory failure will depend
upon a judicious rejection of the initial explanatory demand, as when
we reject the demand to explain why anything exists, or what predica-
tion is, or why the passage of time is not contradictory.
14
Conclusion
In this final chapter, I offer a summary restatement of the main ideas of

the book, followed by a discussion of where I think further research and
refinement may be called for.
14.1 Why study infinity?
Infinity has puzzled philosophers and mathematicians for at least two

millennia. It has given rise to numerous paradoxes – by my count, at
least seventeen – from Zeno’s paradoxes in ancient Greek philosophy
to Benardete’s paradoxes in the twentieth century. Paradox is just what
we’ve come to expect whenever infinity enters the parlor. In some cases,
there is no accepted resolution to the paradox; in others, the dominant
view seems to be a bullet-biting response, that is, to simply acquiesce in
whatever crazy conclusion infinity seems to have brought in the door
with her.
This situation might not be so disturbing if we could pass off infinity
as a mere human construct; yet we seem to find infinity intertwined
with some of the most fundamental, pervasive aspects of the world –
from the infinite divisibility and infinite extent of both space and time,
to the infinitude of numbers and other abstract objects.
In addition, infinity appears in numerous philosophical arguments,
in which profound philosophical conclusions are supposed to be
mandated by the need to avoid vicious infinite regresses. Regress argu-
ments have been used to establish theism, foundationalism in episte-
mology, the unreality of time, the falsity of Plato’s theory of universals,
and the falsity of nominalist theories of universals – and that is only a
sampling from the genre of philosophical infinite regress arguments.
A common pattern is that the author of the regress argument insists
247
that the discovered regress is ‘vicious’, while opponents insist that it

is a ‘virtuous’ regress, with neither party having any clear and well-
motivated account of the difference.
14.2 Troubles with traditional approaches
Traditional discussion of the infinite has been dominated by the debate

over whether the infinite may ever be ’actual’ or ‘completed’, or whether
infinity must instead be always relegated to some kind of mere potenti-
ality. The potentiality view is motivated by Aristotle’s observation that
to be infinite is to lack a definite quantity, and that all actual things
must be fully determinate. Around the turn of the twentieth century,
defenders of actual infinities acquired a powerful new ally, when Georg
Cantor invented set theory and proposed to use it to found a theory of
numbers, which turned out to include an infinite hierarchy of infinite
cardinal numbers.
Each side in the traditional debate misses something crucial. The
Aristotelians simply cannot account for the instances of infinity that
we find in reality – the infinite divisibility of space and time, the infi-
nite extension of space and time (especially the infinite past), and the
infinite number of abstract objects of various kinds. Those who claim
that an infinite series can never be completed are skewered by Zeno’s
paradox, for every time an object moves, it must complete an infinite
series of sub-motions.
On the other hand, the Cantorians, with their unhesitating embrace
of the actual infinite, can offer us no resolution of the paradoxes of
the infinite. In some cases (Hilbert’s Hotel, Galileo’s paradox), the
Cantorians happily propound the conclusions that we find paradoxical.
In other cases, we would intuitively like to say that an infinite series
is impossible (Thomson’s Lamp, Benardete’s Paradox, the Littlewood-
Ross Banker, and others) – but the Cantorians have no principled way of
rejecting these infinities.
14.3 A new approach to infinity
Aristotle is correct in thinking that infinity is no determinate quan-

tity. He is also correct to ban the indeterminate from actuality. But the
right conclusion is not that infinity can never be actual or completed.
The right conclusion is, in a sense, that infinity can never be localized;
there cannot be any single individual or limited region that exemplifies
infinitude.
Conclusion 249
Strictly speaking, there are no infinite quantities: there is no cardinal

number greater than all the natural numbers, nor is there any number
larger than all the real numbers. Any ascription of infinitude must
therefore be able to be paraphrased in such a way that the paraphrase
ascribes only finite quantities to any individual. In the case of infinite
cardinality, such a paraphrase can be given: we can say that the F’s are
infinitely numerous, provided that for every natural number n, there
exist more than n F’s.1 This paraphrase only makes reference to natural
numbers. It assumes that for every natural number n, there is a number
larger than n – but it expressly avoids assuming that there is a number
larger than every natural number. When asked what is the number that
applies to all the F’s taken together, we should reply that there is no such
number; the F’s are literally numberless.
So too, an adequate paraphrase can be given for ascriptions of infi-
nite extensive magnitudes. For example, space is infinitely extended,
provided that for every (finite) volume v, there exist regions larger than
v. This paraphrase only makes reference to finite volumes. It assumes
that for every size, there is a larger size; but it does not assume that
there is a size larger than every other size. If asked what is the size of all
of space, we should deny that there is any such size. Similarly, time has
infinite extent, in the sense that for any chosen temporal interval, there
are intervals longer than it.
But no such paraphrase can be given for an ascription of infinite
intensive magnitude. If an object is ascribed an infinite intensive magni-
tude – for instance, an infinite mass density, or an infinite temperature –
this can only mean that the object possesses some one magnitude that
exceeds all other possible magnitudes of its type. This is metaphysically
impossible.
14.4 Some controversial views about sets,

numbers, and points
There are reasons to doubt that sets exist. No one seems to be able to
explain what they are, they do not correspond to the ordinary notion
of a collection, and core intuitions about sets, particularly the naive
comprehension axiom, lead to contradictions. It is therefore prudent
to attempt to phrase whatever claims we want to make so as to avoid
alluding to sets.
1
If one believes in sets, one can say that a set is of infinite cardinality provided
that for every n, it contains a subset with more than n elements.
Numbers are not sets as the followers of Cantor claim. Cardinal

numbers are properties – for instance, two is the property that is instan-
tiated by any two things. Twoness is what my hands have in common
with the sun and the moon, and with the Empire State Building and
the Chrysler building, and with the colors red and yellow, and so on.
Real numbers, on the other hand, are relationships that may obtain
between specific magnitudes and units – for instance, in saying that
the Eiffel Tower is 324 meters tall, one is describing the relationship
between a certain magnitude (the height of the Tower) and a certain
unit (the meter).
On the standard conception of geometry, extended regions are
composed of points, which are understood as infinitely small, indivis-
ible atoms of space. This notion is problematic for a variety of reasons:
First, points seem to be unimaginable, as they are parts of space with no
shape and zero size. Second, if we ask how much space a point consists
of, the answer has to be ‘zero’, which means that it does not consist of
any space. But a spatial region cannot be composed of things that do not
consist of any space. Third, intuitively, the size of a region should be the
sum of the sizes of its parts. Since points each have a size of zero, every
region should also have a size of zero, if regions are composed entirely
of points. Fourth, if there are measure-zero regions, then there would be
a distinction between a ‘closed’ object (which contains its own surface)
and an ‘open’ object (which occupies all the space inside a certain closed
surface but not the surface itself), and it would be metaphysically impos-
sible for two open objects to ever come in contact, or for two closed
objects to come in contact. Fifth, if there were points, then it would
be metaphysically impossible for an object that completely fills a given
region to be divided into two identical parts; instead, after any division,
one object would have to be ‘open’ and the other ‘closed’ at the place
where the two had been joined.
What goes for points goes also for lines, planes, and other alleged
parts of space that have a size of zero.
14.5 Solving the paradoxes
The above views help to resolve the paradoxes of the infinite. For
example, since there are no infinite numbers, it does not make sense
to apply arithmetical operations to infinity. We can thus avoid various
spurious calculations in which one derives that 1=0 and the like. We can
similarly avoid Galileo’s Paradox. In answer to the question ‘which are
more numerous: the natural numbers or the perfect squares?’, we should
Conclusion 251
say, with Galileo, that neither are more numerous, nor are they equally
numerous; both are simply infinite, and hence numberless.
By denying the existence of points, we avoid the paradox in which
it is said that an object composed of pointlike parts could be converted
into two objects, each qualitatively identical to the original, merely by
moving around the object’s parts.
We can resolve Zeno’s Paradox by recognizing, pace Zeno, that an infi-
nite series can be completed. Here we note that the Zeno series involves
an infinite cardinal number of stages, but that its completion requires no
infinite intensive magnitudes; there is thus no objection to the comple-
tion of the Zeno series from within our theory of the infinite.
On the other hand, a variety of other paradoxes do require infinite
intensive magnitudes, including Thomson’s Lamp, Smullyan’s Rod,
Benardete’s Paradox (including both the infinite series of walls and the
infinite pile of slabs), the Littlewood-Ross Banker, the Spaceship, and
Laraudogoitia’s Marbles. These scenarios require such things as infinite
material strengths, infinite energy density, infinite electrical resistivity,
and infinite attenuation coefficient (opacity). Each of these paradoxes
thus posits a metaphysically impossible scenario, on my account. There
is therefore no need to answer what would happen if such scenarios
occurred.
The Saint Petersburg Paradox can be resolved by noting that the para-
doxical reasoning requires ascribing a nonzero probability to a certain
infinite conjunction. But the conjuncts in that conjunction should in
fact be assigned diminishing probabilities – or at least should not be
assigned ever increasing probabilities, approaching probability one. As
a result, the probability of the infinite conjunction is zero. The Delayed
Heaven Paradox similarly depends on ascribing a nonzero probability to
an infinite conjunction that should really be assigned probability zero.
The puzzle about the Martingale betting system is resolved when we
note that at any given time, the expected amount of money the gambler
will have won by that time will be negative, and the expected winnings
only decrease (that is, become more negative) as one considers longer
finite time periods. Since it is not possible to have played infinitely
many times, the strategy never yields an expected profit.
14.6 For further reflection, or: what is wrong

with this book?
I am persuaded that the views advanced in this book are by and large
correct, or close enough to correct to advance our understanding.
Nevertheless, probably some of this book is wrong; of course, I do not

know exactly which parts. In addition, there are some interesting ques-
tions about the infinite that I have not answered because I do not know
the answers. I use this final section of the book to point out some of
the less satisfactory parts of this book, as well as some of the interesting
questions or issues that I have failed to address.
14.6.1 The dense divided stick

In Section 10.5.2, I discussed a hypothetical Newtonian world in which
there exists a series of objects of increasing density, arranged in a line as
depicted in Figure 14.1.
The objects all have the same mass but have ever-decreasing sizes and
hence increasing densities. In addition, they have ever-increasing right-
ward velocities, such that, after one second, all the objects should simul-
taneously collide within a finite region. The result would be that, at the
moment of the collision, there would be a spatial region containing an
infinite amount of mass. But because mass density is a natural, inten-
sive magnitude, on my view no finite region can contain infinite mass.
I thus claimed that the laws of nature must somehow prevent such a
scenario from taking place – perhaps because there is a finite limit to
how fast an object may move or to how densely matter may be packed
within a region.
This treatment of the scenario is not very satisfying. I suppose that
the laws (in fact, any metaphysically possible set of natural laws) must
somehow prohibit the scenario – but I have hardly explained how or
why this is so. Usually, when a scenario contains an impossible form of
infinitude, we can explain its impossibility by saying that some infinite
intensive magnitude must be postulated in order for the scenario to take
place – that is, the infinite intensive magnitude would be a necessary
causal factor, or in some other sense a precondition, for the scenario to
play out as initially imagined. For example, a lamp cannot be switched
off and on infinitely many times as in Thomson’s scenario, unless we
first have a switch with infinite material strength.
4 m/s 3 m/s 2 m/s 1 m/s
• • •
O5 O4 O3 O2 O1
Figure 14.1 An impossible Newtonian scenario

Conclusion 253
What is unsatisfying about my treatment of the Dense Divided Stick

example is that I rule out the initial conditions on the grounds, not that
they contain or presuppose the initial existence of some infinite inten-
sive magnitude, but that they lead to an infinite intensive magnitude as
a final state of the system. The claim, ‘This event cannot occur because
some impossible initial conditions would be required in order for it to
occur’ is more persuasive than the claim ‘These initial conditions are
impossible because if they occurred some impossible final conditions
would ensue.’ It would thus be desirable to have a better account of why
the dense divided stick scenario is impossible.
14.6.2 Gunky time

I have argued that space is not composed of geometric points. At least
one of my arguments can also be used to reject the existence of temporal
points, that is, instants. Adapting the Zero Argument from Section 11.3:
1. The duration of x is how much time x takes up or consists of.

2. To have zero of something is to fail to have any of that thing.
3. Therefore, to have a duration of zero is to fail to take up or consist of
any time.
4. If instants exist, they are parts of time with zero duration.
5. There cannot be a part of time that does not consist of any time.
6. Therefore, instants do not exist.
This suggests that time is gunky, just as space is; there are only
extended intervals, no single instants. This conclusion is at least some-
what less plausible than the gunky conception of space. One reason is
that there is some intuitive plausibility to presentism in the philosophy
of time, that is, the view that only the present is real. Furthermore, even
if one is unsympathetic to presentism, there is still more plausibility to
the weaker view that the present is in some sense at least more real than
either the past or the future. But ‘the present’ is generally taken to refer
to a single instant of time, not an extended interval. So the gunky view
of time would seemingly require us to hold anti-presentism, the view
that the present does not exist.
Of course, if one holds a gunky conception of time, one might wish to
claim that the English expression ‘the present’ actually refers to a small
but nonzero temporal interval. But there is no non-arbitrary answer to
what is the duration of ‘the present’.
All this might make one suspicious of the Zero Argument, even as
applied to space. However, even if that argument is misguided, there
remain other arguments against pointy space that do not transfer to

time.
14.6.3 Points as locations

Some say that points exist, but that they simply are not the parts of
space. Instead, points are merely ‘locations’.2
I do not have any argument against this view. My arguments against
points are only arguments against the notion that space is comprised of
points. For instance, only if points are supposed to be parts of space does
the Zero Argument apply.
However, I also have not endorsed the ‘location’ theory of points,
because I do not understand it. If a location is not a part of space, then I
do not know what it is. Should we think that locations are a completely
separate kind of thing from space and spatial regions? Could one have
existed without the other? Unfortunately, I have found no clear expla-
nation of the view. But this does not mean the view is incorrect.
14.6.4 Counting to and from infinity

A number of variations on the idea of counting to or from infinity have
been discussed, particularly by Wittgenstein. Thus, one may ask:
(a) Is it possible to count to infinity? To do so, one would have to count

ever faster: with each natural number that one names, one doubles
one’s speed of counting, so that in a finite time, one would have
counted every natural number.
This, in my view, is impossible, because as the numbers one
counts become ever larger, a larger memory and a faster informa-
tion processing capacity is required. Counting to infinity would
thus require infinite information processing capability.
(b) Wittgenstein reportedly once asked an audience to imagine meeting
a man who, as one approaches, is just saying: ‘ ... 5, 4, 3, 2, 1, 0.
Finished!’ The person explains that he has just counted backwards
from infinity. He has been engaged in this task for all of past time,
having never begun but having just now finished. Is this possible?3
Leave aside the extreme implausibility of the claim. It seems to
me that there could (metaphysically) be someone who has always
existed; at least, I know of no reason why there could not. And if
2
Angelo 2014; Bogomolny 2015.
3
Oppy (2006, 10, 60) ascribes this example to Wittgenstein. Of course, I mean
to ask whether it is possible that what the man claims is true, not whether it is
possible to meet someone who makes such a claim.
Conclusion 255
there were such a being, I can think of no reason why that being
should not have always been counting, having been employed
about naming ever larger natural numbers the further into the past
one looks.
Why does this scenario not require infinite information processing
ability as in scenario (a) above? Because there is no need to carry out
infinitely many operations in a finite time (nor within a finite space).
Thus, the man could have used an infinitely large piece of paper to
write down the numbers and to perform his calculations (namely,
subtracting one in each stage), and he could have used longer and
longer time periods for counting off numbers, as one looks further
and further into the past.
To make the idea clearer, imagine the symmetrical case in which
a person starts at 0, and writes down all the natural numbers in
sequence, taking more and more time to write down numbers as
the series proceeds. He spends eternity writing. Every number will
be written down at some time, and at every future time he will be
writing some number. This does not seem impossible. Scenario (b) is
just a time-reversed version of this. It is also a version of the hypoth-
esis of an infinite chain of causes: at each stage, the person would be
caused to name the number n by the immediately preceding event
of his naming the number n+1. His naming n would in turn cause
him next to name n−1.
I must say, however, that scenario (b) seems very suspect (as its
temporal mirror image does not), as though it ought to be somehow
impossible; yet I can find no principled grounds for rejecting its
possibility. Perhaps what is suspicious about it is merely its extreme
improbability.
(c) Similarly, imagine meeting a man who says, ‘ ... 5, 1, 4, 1, 3. Finished’,
then explains that he has just recited the decimal expansion of pi
backwards, having been engaged in this task for all of past time. Is
this possible?4
We might wish to argue that this scenario is impossible because
in order to recite the decimal expansion of pi backwards, one would
have to first know the complete decimal expansion, but in order to
have this knowledge, one would first have to complete an infinitely
long calculation. Why is this different from scenario (b)? Because in
scenario (b), the man could always ‘calculate’ the next number he
4
Moore (1990, 44) ascribes this example to Wittgenstein, in an unidentified
lecture.
has to recite on the basis of the number he just finished reciting.

But one cannot calculate the nth digit of pi on the basis of the (n+1)th
digit.
It is possible, however, to directly calculate the nth digit of pi
without first knowing any of the other digits.5 Therefore, we can
imagine that at each stage, when he is to recite the nth digit, the
man directly calculates this digit. He knows that he is to recite the
nth digit, because he just previously recited the (n+1)th digit. As we
look further and further into the past, we see the man taking more
and more time for each digit recital. But since he has always existed,
there is for each digit in the complete decimal expansion some time
at which he recited it. None of this requires the man to have held
the complete expansion in mind at any particular time.
Again, this scenario strikes me as very suspect, but I can think of
no reason why it is not metaphysically possible.
(d) Finally, Wittgenstein asks us to imagine someone who claims to
have counted all the digits of pi forwards. He has been counting for
all of past time, and he has just gotten to the end; the final digit of
pi, he announces, is 2. Is this possible?
Certainly not, but this impossibility does not, as Wittgenstein
claims, constitute a ‘reductio ad absurdum of the concept of an infi-
nite totality’.6 The reason the man’s claim cannot be true is that pi
does not have a last digit. It does not follow from this that there is
nothing infinite, nor even that the infinite fails to be a ‘totality’,
whatever that might mean.
14.6.5 The eternal watch

In the course of motivating his famous Argument from Design, William
Paley asks readers to imagine walking through a wilderness area, where
you happen to hit your foot against a stone.7 You wonder how the
stone got there. Assuming that this scenario is taking place in a time
5
This can be done in base 16 using Bailey, Borwein, and Plouffe’s (1997) formula,
∞
⎛ 1 ⎞⎛ 4 2 1 1 ⎞.
π = ∑ − − −
⎝ k ⎠ ⎝ 8k + 1 8k + 4 8k + 5 8k + 6 ⎠
k = 0 16
6
Wittgenstein 1975, 166.
7
Paley 1802, 1ff. Compare Leibniz’ (1908, p. 106) example of the geometry
book that was copied from an earlier geometry book, which was copied from
a still earlier book, and so on. Leibniz says that even if we postulate an infinite
series of books, we would still be in need of an explanation for why the books
exist at all, rather than none of them existing.
Conclusion 257
when such things as the history of the Earth and the origin of rocks are
unknown, it would not be unreasonable to hypothesize that perhaps the
stone had always been there.
But now suppose that, a little farther along, you come upon what
looks like a watch sitting on the ground. Again you wonder how it came
to be there. But in this case, it would be laughable to hypothesize that
perhaps the watch simply was always there. Instead, Paley notes, you
would assume that the watch must have been made by someone. This
would be true even if you had no background knowledge about watches,
even if, let us say, this was the first time you had ever seen or heard of
such a device. Paley then leads into his argument for the existence of
God (the rest of which does not concern us here).
Paley’s remarks thus far seem right: it would be crazy to explain the
presence of the watch by the hypothesis that it was always there, in a
way in which it would not be crazy to explain the stone’s presence by
the hypothesis that it was always there. What is the difference between
these two hypotheses?
It seems to me that it cannot be that the one hypothesis is metaphysi-
cally possible while the other is not – if it is possible for a rock to have
always existed, then it has to be possible for a watch to have always
existed. Nevertheless, ‘it always existed’ seems to provide a better expla-
nation for the presence of a rock than it would for the presence of a
watch. Why is this?
One might think the two explanations should be equally good or
bad. In both cases, the presence of the object at time t is explained
by that very object’s existence a moment before t, with the previous
moment’s existence explained by the object’s existence at a still earlier
moment, and so on ad infinitum. Or perhaps more precisely (because
a moment of time doesn’t really have one immediately preceding
moment), it is supposed that the object’s existence at any given time
may be explained by its existence at any earlier time. If this sort of
explanation works for rocks, then it ought to work equally for watches,
universes, or anything else.
Perhaps the explanation fails in the case of watches only because of the
existence of a better, competing explanation: that the watch was made
by an intelligent being. The hypothesis of intelligent design is a better
explanation for the existence of an object exhibiting a high degree of
order and complexity than it is for the existence of an object exhibiting
little order or little complexity. Thus, the hypothesis of eternal existence
is not intrinsically worse for the watch than it is for the rock; it is merely
comparatively worse. In the case of the watch, the ‘eternal existence’
hypothesis loses out decisively to the ‘intelligent design’ hypothesis,

whereas in the case of the rock, the ‘eternal existence’ hypothesis is
perhaps on a par with (or at least does not lose so badly to) the ‘intel-
ligent design’ hypothesis. This would presumably be because intelligent
design raises the probability of complex order, compared to the hypoth-
esis of no design.
The next question that seems naturally to arise is whether the universe
is more like a watch or like a rock – and that turns on whether the
universe as a whole exhibits complex order. My inclination would be
to say that the initial conditions of the universe, going back as far as
we can (that is, to the time of the Big Bang) exhibit simple order, not
complex order, and that for this reason intelligent design does not seem
to be required, even if later stages of the universe exhibit complex order.
However, now we are straying beyond the scope of this book.
A further question puzzles me. As we have seen, it appears that the
adequacy of an ‘eternal existence’ explanation (a theory that says some-
thing always existed) depends upon whether the thing to be explained
exhibits complex order or not. Now suppose that the universe has a
cyclical history: it expands from a Big Bang for some billions of years,
then collapses back into a singularity, whereupon another Big Bang
occurs. Suppose that this cycle has happened at least once. This of course
is speculation. But now suppose also that the universe exhibits complex
order in some parts of the cycle, but that in other parts of the cycle the
universe is either simple or disorderly. In that case, would it be a reason-
able explanation to hypothesize that the universe has always been going
through these cycles? I don’t know.
14.6.6 Sets revisited

Perhaps I have been too hard on sets. Perhaps a set is after all just a
collection, and collections exist. Perhaps the notions of the empty set
and of singleton sets are simply mistakes; alternately, perhaps they are
permissible extensions of the notion of a collection, in the same way that
(as I allowed in Section 9.5) the concept of zero is a permissible exten-
sion of the concept of a number.8
I am skeptical of treating the empty set and singletons as acceptable
extensions of the notion of a collection. Extending the number system
to include ‘zero’ is permissible, I believe, because we can explain the
meanings of various statements that contain the symbol ‘0’ without
8
I thank an anonymous referee who I assume to be Adrian Moore for raising
this issue.
Conclusion 259
assuming that that symbol refers to anything – for example, we can say
that ‘4+0 = 4’ means something like: if you have 4 things, and you fail to
have any additional things, then you have a total of four things.
Similarly, some uses of ‘the empty set’ can be given acceptable para-
phrases. For instance, ‘the set of all purple unicorns is the empty set’
really just means ‘there aren’t any purple unicorns.’ But when the empty
set itself is allowed to be a member of other sets (as in the set {{ }, {{ }}}),
I no longer see any ontologically innocent way of interpreting this; I
think this requires ‘{ }’ to have a referent. Similarly, when, in set theory,
we treat {Sue} and {{Sue}} as two distinct sets, I see no ontologically inno-
cent way of understanding the singleton.
What about a version of set theory that simply does away with the
empty set and singleton sets, so that all sets are required to have at
least two members? This strikes me as less objectionable. However,
this sort of theory would still be subject to the standard paradoxes
of set theory. And the fundamental idea of set theory still strikes me
as metaphysically very suspect – that if there are two things, a and b,
then, automatically, there has to be some third thing, which ‘contains’
the first two. But this third thing is not a composite object literally
composed of a and b. If a and b are physical objects, the third thing
is not a physical object but an ‘abstract particular’. Why must this
strange thing exist?
Another idea: perhaps sets are just a façon de parler, a device for
making statements about multiple things at once. Thus, ‘the set of cats
is a subset of the set of furry things’ is perhaps just a fancy way of saying
‘all cats are furry’, with no more genuine ontological import than the
latter statement. This would be fine, except that many statements of set
theory (including the ones that lead to the set theoretic paradoxes) lack
any such innocent paraphrases. Once we start talking about sets that
contain other sets as members, it becomes very hard to see how this talk
could be paraphrased in a way that does not sound as though we are
positing a special sort of object called a ‘set’.
Complete rejection of sets is not entirely satisfactory either, because
it looks as though there are legitimate facts whose description requires
us to speak of sets, or set-like entities. For instance, one might want
to explain the meaning of the statement ‘4×3=12’ by saying: ‘Suppose
there are four non-overlapping groups, each containing three objects.
Then there will be a total of twelve objects.’ The mentioned groups
seem to be functionally equivalent to sets (perhaps without some of the
more bizarre aspects, such as an ‘empty group’ or ‘singleton groups’),
and it is hard to see how to paraphrase the statement without seeming
to posit some setlike entities. Conclusion: it is hard to know what to

say about sets.
This concludes my approach to the subject of infinity. As the fore-
going remarks indicate, I continue to find much in this area unclear.
Nevertheless, I hope the reader will now find the infinite at least some-
what less puzzling than it was at the end of Chapter 3.
References
Albert, David Z. 1994. Quantum Mechanics and Experience. Cambridge, Mass.:

Harvard University Press.
Allis, Victor and Teunis Koetsier. 1991. ‘On Some Paradoxes of the Infinite’, British
Journal for the Philosophy of Science 42: 187–94.
Angelo, Robert Wesley. 2014. ‘The Philosophy of Geometry’, <http://www.
roangelo.net/logwitt/logwit19.html>, accessed June 17, 2015.
Anselm, St. 1926. Anselm’s Proslogium or Discourse on the Existence of God, in
Anselm: Basic Writings, tr. Sidney Norton Deane. Chicago: Open Court. Available
at <http://www.ccel.org/ccel/anselm/basic_works>, accessed June 17, 2015.
Originally published in 1078.
Aquinas, Thomas. 1981. Summa Theologica, tr. by Fathers of the English Dominican
Province. Westminster, Md.: Christian Classics. Available at <http://sacred-
texts.com/chr/aquinas/summa/>.
Aristotle. 1936. Aristotle’s Physics, tr. W.D. Ross. Oxford: Clarendon.
Aristotle. 1941. The Basic Works of Aristotle, ed. Richard McKeon. New York:
Random House.
Aristotle. 2005. Physics, or Natural Hearing, tr. Glen Coughlin. South Bend, Ind.:
St. Augustine’s Press.
Armstrong, David. 1978. Nominalism and Realism. Cambridge: Cambridge
University Press.
Armstrong, David. 1989. Universals: An Opinionated Introduction. Boulder, Colo.:
Westview.
Arntzenius, Frank. 2008. ‘Gunk, Topology, and Measure’, pp. 225–47 in Oxford
Studies in Metaphysics, vol. 4, ed. Dean Zimmerman. Oxford: Oxford University
Press.
Ayer, Alfred Jules. 1952. Language, Truth and Logic. New York: Dover.
Baez, John. 2002. ‘The Octonions’, Bulletin of the American Mathematical Society
39: 145–205. Available at <http://math.ucr.edu/home/baez/octonions/oct.
pdf>, accessed June 16, 2015.
Bailey, David, Peter Borwein, and Simon Plouffe. 1997. ‘On the Rapid Computation
of Various Polylogarithmic Constants’, Mathematics of Computation 66: 903–13.
Banach, Stefan and Alfred Tarski. 1924. ‘Sur la Décomposition des Ensembles de
Points en Parties Respectivement Congruentes’, Fundamenta Mathematicae 6:
244–77.
Barrett, Jeffrey and Frank Arntzenius. 1999. ‘An Infinite Decision Puzzle’, Theory
and Decision 46: 101–3.
Bays, Timothy. 2014. ‘Skolem’s Paradox’, The Stanford Encyclopedia of Philosophy,
ed. Edward N. Zalta, <http://plato.stanford.edu/archives/spr2014/entries/para-
dox-skolem/>, accessed June 30, 2014.
Bealer, George. 1992. ‘The Incoherence of Empiricism’, Proceedings of the
Aristotelian Society 66, supplement: 99–138.
Bealer, George. 1999. ‘A Theory of the A Priori’, pp. 29–55 in Philosophical Perspectives
13: Epistemology, ed. James Tomberlin. Cambridge, Mass.: Blackwell.
261
262 References
Benacerraf, Paul. 1962. ‘Tasks, Super-Tasks, and the Modern Eleatics’, Journal of
Philosophy 59: 765–84.
Benardete, J. 1964. Infinity: An Essay in Metaphysics. Oxford: Clarendon.
Beresford, Geoffrey C. 1981. ‘A Note on Thomson’s Lamp “Paradox”’, Analysis
41: 1–3.
Bernoulli, Daniel. 1954. ‘Exposition of a New Theory on the Measurement of
Risk’, tr. Louise Sommer. Econometrica 22: 23–36. Originally published in 1738.
Bliss, Ricki Leigh. 2013. ‘Viciousness and the Structure of Reality’, Philosophical
Studies 166: 399–418.
Bogomolny, Alexander. 2015. ‘What Is Point?’ from Interactive Mathematics
Miscellany and Puzzles, <http://www.cut-the-knot.org/WhatIs/WhatIsPoint.
shtml>, accessed June 17, 2015.
BonJour, Laurence. 1998. In Defense of Pure Reason. Cambridge: Cambridge
University Press.
Brentano, Franz. 1988. Philosophical Investigations on Space, Time, and the
Continuum, tr. Barry Smith. London: Croom Helm.
Cameron, Ross P. 2008. ‘Turtles All the Way Down: Regress, Priority and
Fundamentality’, Philosophical Quarterly 58: 1–14.
Cantor, Georg. 1915. Contributions to the Founding of the Theory of Transfinite
Numbers, tr. Philip E.B. Jourdain. Chicago: Open Court.
Cantor, Georg. 1932. Gesammelte Abhandlungen, ed. Ernst Zermelo and Adolf
Fraenkel. Berlin: Springer-Verlag.
Carnap, Rudolf. 1932. ‘The Elimination of Metaphysics through Logical Analysis
of Language’, trans. Arthur Pap, pp. 60–81 in Logical Positivism, ed. A.J. Ayer.
New York: Free Press.
Clarke, Samuel. 1998. A Demonstration of the Being and Attributes of God and Other
Writings, ed. Ezio Vailati. Cambridge: Cambridge University Press.
Cortzen, Allan and Weisstein, Eric W. 2015. ‘Measure’, from Wolfram MathWorld.
Available at <http://mathworld.wolfram.com/Measure.html>, accessed June
16, 2015.
Courant, Richard and Herbert Robbins. 1996. What Is Mathematics? An Elementary
Approach to Ideas and Methods, second edition. New York: Oxford University
Press.
Craig, William Lane. 1991. ‘The Existence of God and the Beginning of the
Universe’, Truth: A Journal of Modern Thought 3: 85–96. Available at <http://
www.leaderu.com/truth/3truth11.html>, accessed June 25, 2014.
Craig, William Lane. 2010. ‘Taking Tense Seriously in Differentiating Past and
Future: A Response to Wes Morriston’, Faith and Philosophy 27: 451–6.
Dainton, Barry. 2001. Time and Space. Montreal: McGill-Queen’s University Press.
Dedekind, Richard. 1901. ‘Continuity and Irrational Numbers’, pp. 1–13 in
Essays on the Theory of Numbers, tr. by Wooster Woodruf Beman. Chicago: Open
Court.
Descartes, René. 1984. The Philosophical Writings of Descartes. Edited by John
Cottingham, Robert Stoothoff, and Dugald Murdoch. Cambridge: Cambridge
University Press.
Dyson, Frank, Arthur Eddington, and Charles Davidson. 1920. ‘A Determination
of the Deflection of Light by the Sun’s Gravitational Field, from Observations
Made at the Total Eclipse of May 29, 1919’, Philosophical Transactions of the
Royal Society A: Mathematical, Physical and Engineering Sciences 220: 291–333.
References 263
e Silva, Tomás Oliveira. 2014. ‘Goldbach Conjecture Verification’, <http://sweet.

ua.pt/tos/goldbach.html>, accessed June 28, 2014.
Ebbinghaus, Heinz-Dieter. 2007. ‘Löwenheim-Skolem Theorems’, pp. 587–614 in
Philosophy of Logic, ed. Dale Jacquette. London: Elsevier.
Einstein, Albert. 1961. Relativity: The Special and the General Theory, tr. Robert W.
Lawson. New York: Bonanza Books.
Enderton, Herbert. 1977. Elements of Set Theory. San Diego, Calif.: Academic Press.
Euclid. 1998. Euclid’s Elements, ed. David Joyce, <http://aleph0.clarku.edu/~djoyce/
java/elements/>, accessed June 26, 2014.
Fraenkel, Abraham. 1967. ‘Set Theory’, pp. 420–7 in The Encyclopedia of Philosophy,
vol. 7, ed. Paul Edwards. New York: Macmillan.
Frege, Gottlob. 1972. ‘Review of Dr. E. Husserl’s Philosophy of Arithmetic’, tr.
E.W. Kluge, Mind 81: 321–37.
Frege, Gottlob. 1980. The Foundations of Arithmetic. Evanston, Ill.: Northwestern
University Press.
Galilei, Galileo. 1914. Dialogues Concerning Two New Sciences by Galileo Galilei, tr.
Henry Crew and Alfonso de Salvio. New York: Macmillan. Available at <http://
oll.libertyfund.org/titles/753>.
Gamow, George. 1947. One, Two, Three ... Infinity: Facts and Speculations of Science.
New York: Viking Press.
Giancoli, Douglas C. 1988. Physics for Scientists and Engineers, second edition.
Englewood Cliffs, N.J.: Prentice Hall.
Gillett, Carl. 2003. ‘Infinitism Redux? A Response to Klein’, Philosophy and
Phenomenological Research 66: 709–17.
Greene, Brian. 2011. The Hidden Reality: Parallel Universes and the Deep Laws of the
Cosmos. New York: Alfred A. Knopf.
Grünbaum, Adolf. 1957. ‘The Philosophical Retention of Absolute Space in
Einstein’s General Theory of Relativity’, Philosophical Review 66: 525–34.
Grünbaum, Adolf. 1968. ‘Are “Infinity Machines” Paradoxical?’ Science 159:
396–406.
Halmos, Paul R. 1974. Naive Set Theory. New York: Springer-Verlag.
Hamade, Rufus. 1996. ‘Black Holes and Quantum Gravity’, Cambridge Relativity
and Cosmology, University of Cambridge. Available at <http://www.damtp.
cam.ac.uk/research/gr/public/bh_hawk.html>, accessed June 25, 2014.
Hamilton, William R. 1837 (based on a lecture first given in 1833). ‘Theory of
Conjugate Functions, or Algebraic Couples; With a Preliminary and Elementary
Essay on Algebra as the Science of Pure Time’, Transactions of the Royal Irish
Academy 17, part 1: 293–422. Available at <http://www.maths.tcd.ie/pub/
HistMath/People/Hamilton/PureTime/> (edited by David R. Wilkins), accessed
June 16, 2015.
Hawthorne, John. 2000. ‘Before-Effect and Zeno Causality’, Noûs 34: 622–33.
Hawthorne, John and Brian Weatherson. 2004. ‘Chopping up Gunk’, The Monist
87: 339–50.
Helmenstine, Anne Marie. 2014. ‘Table of Electrical Resistivity and Conductivity:
Flow of Electric Current through Materials’, About.com Chemistry, <http://
chemistry.about.com/od/moleculescompounds/a/Table-Of-Electrical-
Resistivity-And-Conductivity.htm>, accessed July 30, 2014.
Hilbert, David. 1967. ‘On the Infinite’, tr. Stefan Bauer-Mengelberg, pp. 367–92 in
From Frege to Gödel: A Source Book in Mathematical Logic 1879–1931, ed. Jean van
264 References
Heijenoort. Cambridge, Mass.: Harvard University Press. Originally published

as ‘Über das Unendliche’ in Mathematische Annalen 95 (1926): 161–90.
Hilbert, David. 2013. David Hilbert’s Lectures on the Foundations of Arithmetics and
Logic 1917–1933, eds. William Ewald and Wilfried Sieg. Heidelberg: Springer-
Verlag.
Hobbes, Thomas. 1996. Leviathan, ed. Richard Tuck. Cambridge: Cambridge
University Press. Originally published in 1651.
Hochberg, Herbert. 1999. Complexes and Consciousness. Stockholm: Thales.
Howell, Elizabeth. 2014. ‘How Many Stars Are in the Universe?’ Space.com,
http://www.space.com/26078-how-many-stars-are-there.html, accessed June
19, 2014.
Hrbacek, Karel and Thomas Jech. 1999. Introduction to Set Theory, third edition.
New York: Marcel Dekker.
Huemer, Michael. 2001. Skepticism and the Veil of Perception. Lanham, Md.:
Rowman & Littlefield.
Huemer, Michael. 2006. ‘Phenomenal Conservatism and the Internalist Intuition’,
American Philosophical Quarterly 43: 147–58.
Huemer, Michael. 2007. ‘Compassionate Phenomenal Conservatism’, Philosophy
& Phenomenological Research 74: 30–55.
Huemer, Michael. 2009. ‘When Is Parsimony a Virtue?’ Philosophical Quarterly 59:
216–36.
Huemer, Michael. 2010. ‘Foundations and Coherence’, pp. 22–33 in A Companion
to Epistemology, second edition, ed. Jonathan Dancy, Ernest Sosa, and Matthias
Steup. Malden, Mass.: Wiley-Blackwell.
Huemer, Michael. 2016. ‘Inferential Appearances’, pp. 144–60 in Intellectual
Assurance: Essays on Traditional Epistemic Internalism, ed. Brett Coppenger and
Michael Bergmann. Oxford: Oxford University Press.
Hume, David. 1975. Enquiry Concerning Human Understanding. Reprinted in
Enquiries Concerning Human Understanding and Concerning the Principles of Morals,
ed. P. H. Nidditch. Oxford: Clarendon. Originally published in 1758.
Hume, David. 1992. Treatise of Human Nature. Buffalo, N.Y.: Prometheus.
Husserl, Edmund. 2003. Philosophy of Arithmetic: Psychological and Logical
Investigations with Supplementary Texts from 1887–1901, tr. Dallas Willard.
Dordrecht: Kluwer.
Kant, Immanuel. 1965. Critique of Pure Reason, tr. Norman Kemp Smith. New
York: St. Martin’s Press.
Kenna, Aaron. 2011. ‘Logical Positivism and the Copenhagen Interpretation of
Quantum Mechanics’, <http://unfspb.wordpress.com/2011/07/12/logical-posi-
tivism-and-the-copenhagen-interpretation/>, accessed June 29, 2014.
Klein, Peter. 1999. ‘Human Knowledge and the Infinite Regress of Reasons’,
pp. 297–325 in Philosophical Perspectives, vol. 13, ed. James E. Tomberlin.
Oxford: Blackwell.
Klein, Peter. 2003. ‘When Infinite Regresses Are Not Vicious’, Philosophy and
Phenomenological Research 66: 718–29.
Klein, Peter. 2005. ‘Infinitism Is the Solution to the Regress Problem’, pp. 131–40
in Contemporary Debates in Epistemology, ed. Matthias Steup and Ernest Sosa.
Malden, Mass.: Blackwell.
Knuth, Donald E. 1976. ‘Mathematics and Computer Science: Coping with
Finiteness’, Science 194: 1235–42.
References 265
Kragh, Helge. 2014. ‘The True (?) Story of Hilbert’s Infinite Hotel’, <http://arxiv.
org/abs/1403.0059v2> [physics.hist-ph], accessed June 26, 2014.
Kunen, Kenneth. 1980. Set Theory: An Introduction to Independence Proofs.
Amsterdam: Elsevier.
Kusch, Martin. 1990. ‘Why Is There Something Rather than Nothing?’ American
Philosophical Quarterly 27: 253–7.
Laraudogoitia, Jon Perez. 1996. ‘A Beautiful Supertask’, Mind 105: 81–3.
Laraudogoitia, Jon Perez. 2003. ‘A Variant of Benardete’s Paradox’, Analysis 63:
124–31.
Leibniz, G.W.F. 1908. The Philosophical Works of Leibniz, second edition, tr. and ed.
George Martin Duncan. New Haven, Conn.: Tuttle, Morehouse, and Taylor.
Leibniz, G.W.F. and Samuel Clarke. 2007. Exchange of Papers between Leibniz and
Clarke, tr. Jonathan Bennett. Available at <http://www.earlymoderntexts.com/
pdfs/leibniz1715_1.pdf>, accessed June 27, 2014.
Littlewood, J.E. 1986. Littlewood’s Miscellany, ed. Béla Bollobás. Cambridge:
Cambridge University Press.
Locke, John. 1975. An Essay Concerning Human Understanding, ed. Peter H.
Nidditch. Oxford: Clarendon. Originally published in 1689.
Mach, Ernst. 1919. The Science of Mechanics: A Critical and Historical Account of Its
Development, tr. Thomas McCormack. Chicago, Ill.: Open Court. Available at
<http://archive.org/stream/scienceofmechani005860mbp>, accessed June 27,
2014.
Maddy, Penelope. 1990. Realism in Mathematics. Oxford: Clarendon.
Maudlin, Tim. 1993. ‘Buckets of Water and Waves of Space: Why Spacetime Is
Probably a Substance’, Philosophy of Science 60: 183–203.
Maudlin, Tim. 2002. Quantum Nonlocality and Relativity. Malden, Mass.:
Blackwell.
McTaggart, J.M.E. 1908. ‘The Unreality of Time’, Mind: A Quarterly Review of
Psychology and Philosophy 17: 456–73.
Milewski, Emil. 1989. The Essentials of Set Theory. Piscataway, N.J.: Research and
Education Association.
Moore, Adrian. 1990. The Infinite. London: Routledge.
Newton, Isaac. 1846. The Mathematical Principles of Natural Philosophy, tr. Andrew
Motte. New York: Daniel Adee. Available at <http://en.wikisource.org/wiki/The_
Mathematical_Principles_of_Natural_Philosophy_(1846)>, accessed June 24,
2014. Originally published as Philosophiæ Naturalis Principia Mathematica, 1687.
Nolan, Daniel. 2001. ‘What’s Wrong with Infinite Regresses?’ Metaphilosophy 32:
523–38.
O’Connor, John J. and Edmund F. Robertson. 2000. ‘A History of Zero’, The
MacTutor History of Mathematics archive, University of St. Andrews, <http://
www-history.mcs.st-and.ac.uk/HistTopics/Zero.html>, accessed July 3, 2014.
Ockham, William. 1994. ‘Five Questions on Universals’ from his Ordinatio, d.
2, qq. 4–8; pp. 114–231 in Five Texts on the Mediaeval Problem of Universals:
Porphyry, Boethius, Abelard, Duns Scotus, Ockham, tr. and ed. by Paul Vincent
Spade. Indianapolis, Ind.: Hackett.
Oppy, Graham. 1995. ‘Inverse Operations with Transfinite Numbers and the Kalam
Cosmological Argument’, International Philosophical Quarterly 35: 219–21.
Oppy, Graham. 2006. Philosophical Perspectives on Infinity. Cambridge: Cambridge
University Press.
266 References
Oppy, Graham. 2014. ‘Ontological Arguments’, Stanford Encyclopedia of Philosophy,

ed. Edward N. Zalta, <http://plato.stanford.edu/archives/win2014/entries/
ontological-arguments/>.
Paley, William. 1802. Natural Theology: Or, Evidences of the Existence and Attributes
of the Deity Collected from the Appearances of Nature. London: Wilks and Taylor.
Peijnenburg, Jean and David Atkinson. 2010. ‘Lamps, Cubes, Balls and Walls:
Zeno Problems and Solutions’, Philosophical Studies 150: 49–59.
Perls, Frederick S. 1969. Gestalt Therapy Verbatim, ed. John O. Stevens. Lafayette,
Calif.: Real People Press.
Perls, Frederick S. 1947. Ego, Hunger, and Aggression: The Beginning of Gestalt
Psychotherapy. New York: Random House.
Plato. 1931. The Dialogues of Plato, third edition, vol. 4, tr. Benjamin Jowett. London:
Oxford University Press. Available at <http://oll.libertyfund.org/titles/768>.
Plato. 1974. Plato’s Republic, tr. G.M.A. Grube. Indianapolis, Ind.: Hackett.
Plato. 2012. Meno, tr. Cathal Woods. Available at <http://ssrn.com/
abstract=1910945>, accessed June 26, 2014.
Price, H. H. 1953. Thinking and Experience. Cambridge, Mass.: Harvard University
Press.
Putnam, Hilary. 1975. ‘The Meaning of “Meaning”’, pp. 131–93 in Language,
Mind, and Knowledge: Minnesota Studies in the Philosophy of Science, vol. 7, ed.
Keith Gunderson. Minneapolis: University of Minnesota Press.
Quantum Munich. 2014. ‘Negative Absolute Temperature’, <http://www.quantum-
munich.de/research/negative-absolute-temperature/>, accessed August 2, 2014.
Quine, Willard van Orman. 1936. ‘Truth by Convention’, pp. 77–106 in The Ways
of Paradox and Other Essays. Cambridge, Mass.: Harvard University Press.
Quine, Willard van Orman. 1951. ‘Two Dogmas of Empiricism’, Philosophical
Review 60: 20–43.
Reichenbach, Hans. 1957. The Philosophy of Space and Time, tr. Maria Reichenbach
and John Freund. New York: Dover.
Robinson, Abraham. 1966. Non-standard Analysis. Amsterdam: North Holland
Publishing.
Rodriguez-Pereyra, Gonzalo. 2002. Resemblance Nominalism: A Solution to the
Problem of Universals. Oxford: Clarendon.
Ross, Sheldon. 1976. A First Course in Probability. New York: Macmillan.
Russell, Bertrand. 1920. Introduction to Mathematical Philosophy. London: Allen &
Unwin. Available at <http://people.umass.edu/klement/imp/imp.html>.
Russell, Bertrand. 1961. ‘Why I Am Not a Christian’, pp. 585–97 in The Basic
Writings of Bertrand Russell, ed. Robert Egner and Lester Dennon. New York:
Simon & Schuster.
Sider, Ted. 2005. Set Theory (class notes), available at <http://tedsider.org/teaching/
st/st_notes.pdf>, accessed June 30, 2014.
Smith, Quentin. 1986. ‘The Infinite Regress of Temporal Attributions’, Southern
Journal of Philosophy 24: 383–96.
Smullyan, Raymond. 2008. ‘Memories and Inconsistencies’, pp. 245–52 in
A Lifetime of Puzzles, ed. Erik Demaine, Martin Demaine, and Tom Rodgers.
Wellesley, Mass.: A K Peters.
The Writer (screen name). 2003. ‘Layman’s Guide to the Banach-Tarski Paradox’,
Kuro5hin, <http://www.kuro5hin.org/story/2003/5/23/134430/275>, accessed
August 28, 2015.
References 267
Thomson, James. 1954. ‘Tasks and Super-Tasks’, Analysis 15: 1–13.

Thomson, James. 2001. ‘Comments on Professor Benacerraf’s Paper’, pp. 130–8 in
Zeno’s Paradoxes, ed. Wesley C. Salmon. Indianapolis, Ind.: Hackett.
Tooley, Michael. 1997. Time, Tense, and Causation. Oxford: Clarendon.
Tucker, Chris, ed. 2013. Seemings and Justification: New Essays on Dogmatism and
Phenomenal Conservatism. Oxford: Oxford University Press.
Turri, John and Peter Klein, eds. 2014. Ad Infinitum: New Essays on Epistemological
Infinitism. Oxford: Oxford University Press.
van Bendegem, Jean Paul. 1994. ‘Ross’ Paradox Is an Impossible Super-task’,
British Journal for the Philosophy of Science 45: 743–8.
van Inwagen, Peter. 1983. An Essay on Free Will. Oxford: Clarendon.
Villanueva, John Carl. 2009. ‘Atoms in the Universe’, Universe Today, July 30,
<http://www.universetoday.com/36302/atoms-in-the-universe/>, accessed
June 30, 2014.
von Mises Richard. 1957. Probability, Statistics and Truth, revised English edition.
New York: Macmillan.
Wald, Robert M. 1984. General Relativity. Chicago: University of Chicago Press.
Whitehead, Alfred N. 1917. The Organisation of Thought, Educational and Scientific.
London: Williams and Norgate.
Whitehead, Alfred North. 1929. Process and Reality: An Essay in Cosmology. New
York: Macmillan.
Wildberger, Norman J. 2006. ‘Set Theory: Should You Believe?’ <http://web.
maths.unsw.edu.au/~norman/papers/SetTheory.pdf>, accessed June 30, 2014.
Wittgenstein, Ludwig. 1975. Philosophical Remarks, ed. Rush Rhees, tr. Raymond
Hargreaves and Roger White. New York: Harper & Row.
Yi, Byeong-uk. 1999. ‘Is Two a Property?’ Journal of Philosophy 96: 163–90.
Yi, Byeong-Uk. 2002. Understanding the Many. New York: Routledge.
Zimmerman, Dean W. 1996. ‘Could Extended Objects Be Made Out of Simple
Parts? An Argument for “Atomless Gunk”’, Philosophy and Phenomenological
Research 56: 1–29.
Index
a priori knowledge, 68, 94, 98–100, 10–11, 45, 96, 100, 229, 231–2, 241;
103–4, 106, 122, 158, 235, 240 philosophical, 96
abstract objects. See objects: abstract Benacerraf, Paul, 195–7
acceleration, 64–5 Bernadete, José, 66, 209
actual infinity, 8, 23, 41, 49–53, 55–7, Bernadete’s Paradox, 31–2, 85, 207,
60–2, 65–71, 86, 89, 147–53, 179, 209–11, 247–8, 251
186, 230, 234, 245, 248 Bernoulli, Daniel, 213
addition, 11, 90, 113, 117, 122–3, 127, Big Bang, 7, 68, 258
140–1, 144–5, 176, 180 binding energy, 192–4, 207, 227
analogy, argument by, 105 black hole, 7, 61, 159, 198, 202
analytic a priori knowledge, 99 Bolzano, Bernard, 71
analytic philosophy, 95 brain in a vat, 98–9
analytic statements, 97 branching timelines theory, 102–3
appearances: species of, 96, 100;
undefeated, 96 calculus, 26, 141, 179–80
Aquinas, Saint Thomas, 50, 52 Cantor Lines, 173–4
argument by analogy, 105 Cantor Set, 173–5
Argument from Design, 256 Cantor, Georg, 71–5, 79, 82, 86–9, 108,
Aristotle, 49–54, 57–8, 62, 65, 68–71, 111–13, 116, 120, 143, 146, 149, 176,
86, 147–9, 151, 153, 186, 245, 248 179–82, 248, 250
arithmetic, 56, 75, 88, 107, 117–19, Cantor’s Paradox, 83, 85, 113
122, 124, 141, 144; laws of, 116, Cantor’s Theorem, 71, 80–2, 85–6
123–4; transfinite, 87. See also Cantorians, 71, 89, 146, 176, 182, 248
infinity: arithmetic of causation, 10, 63, 98, 136, 138,
Armstrong, David, 13–14 154–6, 166, 168, 214, 238; causal
artificial magnitudes. See magnitudes: explanation, 136, 238; causal laws,
artificial vs. natural 136, 158, 169, 240; causal relevance,
Associative Law, 116, 123–4, 180 207; first cause, 10, 47, 229, 236;
astronomy, 237, 239 regress of causes, 9–10, 44, 46–7,
astrophysics, 159 229–30, 234, 236, 238–40
atomic theory, 27, 164, 192 Celsius temperature scale, 137–8
atoms, 127, 135, 192, 207, 250 centrifugal force, 63–4
attenuation coefficient, 209–11 chelonian cosmology, 237
Axiom of Choice, 84, 109 choices, infinite regress of, 233
axioms: of arithmetic, 116; of set collections, 18, 66, 75, 107, 109, 146,
theory, 84 177, 179, 181, 189–90, 209, 211, 240;
infinite, 20, 22, 26, 68, 75, 78, 147,
Beer-Lambert Law, 210, 227 178, 181, 216, 222, 231; relation to
begging the question, 142, 232, 239 sets, 72–3, 107–8, 110, 112, 249, 258
beliefs, 5, 10, 47, 96, 99, 195, 229, Commutative Law, 116, 123, 124, 180
231–2, 240–1; actual vs. potential, Compactness Theorem, 110
232; dispositional vs. occurrent, 11, completed infinity, 245, 248
231; foundational, 11, 229; justified, Comprehension Axiom 90, 113, 249
269
270 Index
conceivability, 101–2 epistemology, 93, 95, 98, 107, 117,

concept, empty, 143 214, 247
conceptual truth, 42 equality, 75
concrete world, 139 equinumerousness, 75, 114–15
contact, 168 ethics, 94
Copenhagen Interpretation of Euler’s number, 210
quantum mechanics, 69, 95 evidence: empirical, 104; inductive,
Cosmological Argument. See First 216; scientific, 237
Cause Argument expected utility, 214–15
Coulomb’s Constant, 223, 225–6 experience: memory, 96; sensory,
countable additivity, 177 96–7
countable infinity, 77, 190–1 explanatory condition, 45
countable model, 110 exponentiation, 127
Critique of Pure Reason, 41 extension of a concept, 125, 145
extensive magnitudes. See magnitudes:
decision theoretic paradox, 36, 206, extensive
219
Delayed Heaven Paradox, 36, 85, façon de parler, 51, 141, 259
220–2, 251 final conditions, 158, 253
density, 135–6, 153–4, 159–60, 170, finitism, 50
177, 203, 206, 211, 252; infinite, 7, First Cause Argument, 9–10, 47,
154, 159–60, 211 229–31, 234, 236, 238–9
Descartes, René, 6 first order logic, 110
desire-frustration, 37 formalism, 107
desire-satisfaction, 36 Forms, Platonic, 11–12, 229, 242, 247;
determinate answers, 196, 224, regress of, 11–12, 16, 45–6, 229,
226–8 231, 233–4, 242
determinate quantities, 52, 147–8, 248 foundationalism, 11, 247
Diagonalization Argument, 78–9, 81, Frege, Gottlob, 80, 108, 112, 117, 120,
181 125, 129
diminishing marginal utility, 213–15
Distributive Law, 116, 123–4 Gabriel’s Horn, 23–4, 86, 183
Divided Stick Paradox, 27–8, 42, 85, Galileo, 21–2, 86–7, 146, 251
88–9, 188–9, 192, 253 Galileo’s Paradox, 21–2, 86, 89, 146,
divisibility, infinite, 52–4, 57, 153, 180, 182, 248, 250
234, 247–8 game theory, 107
division, 53–4, 87–8, 122, 126, 145, Gauss, Carl Friedrich, 51
170, 190, 225–6, 250; infinite, 42, general relativity. See relativity
90, 145, 188–9 geometric points. See points
geometry, 18, 55–6, 60, 107, 162–3,
efficacy, lack of, 154–5 250
Einstein, Albert, 3–4. See also relativity God, 6, 9–10, 37, 71, 104, 128,
electromagnetic force, 191 214–18, 221–2, 234, 236, 239,
empirical hypothesis, 94 257
empiricism, 93–5, 99, 104, 106–7 Gödel’s Theorem, 110
empty set, 80, 108, 112–14, 133–4, Goldbach’s Conjecture, 103–4, 171
258–9 Greek philosophy, 247
entropy, 155–8 gunk 162–3, 170–5, 253. See also
epiphenomena, 136 points
Index 271
Hilbert, David, 71, 82 213, 225–6, 251–2; momentum,

Hilbert’s Hotel, 22–3, 85–7, 89, 182, 160; memory, 202; money, 30, 32,
248 35–6, 43, 218–19; resistivity, 200–1,
Hobbes, Thomas, 50 251; space, 57–8, 150, 160–1, 247;
homuncular theory of perception, spacetime curvature, 158–9, 190;
236–8 temperature, 155–8, 249; time, 31,
Hume, David, 94 47, 57, 151, 234, 247–8; torque,
Husserl, Edmund, 125 184; value, 186, 225; velocity, 33,
42, 152, 160, 182, 201–2, 212
Ideal Gas Law, 138 infinite regress. See infinite series
identity, 121 infinite regress arguments. See regress
impossibility: conceptual, 3, 26, 42–3, arguments
136, 224; logical, 120, 167, 224; infinite series, 26–7, 38, 41–3, 179–80,
metaphysical, 103, 105–6, 136, 185–6, 188, 196, 211, 219, 224,
158–60, 169–72, 189, 197, 207, 209, 229–37, 242, 245, 247–8, 251; of
211, 214–15, 217, 224–5, 230–2, actions, 224; benign/virtuous, 9,
245, 249, 250–1; physical, 61, 44–45, 229, 233, 242, 244–5, 248; of
160–1, 223, 225; shallow vs. deep, causes, 10, 44, 47, 229–30, 236, 238;
223–4, 226 of choices, 233; of conditions, 45–6;
inconsistency puzzle, 194–6 of events, 47, 54, 234; of facts, 243;
indeterminancy puzzle, 194–9 of Forms, 12, 46–7, 229, 233, 242; of
infinite cardinality. See number: line segments, 175; of mathematical
infinite operations, 127; of motions, 54,
infinite collections. See collections: 202, 248; of numbers, 42, 175, 178;
infinite of objects, 47, 159, 242; of persons,
infinite conjunction, 216, 251 237; of propositions, 232–3; of
infinite divisibility. See divisibility: reasons, 11, 45, 229, 231–4, 241; of
infinite relations, 243; of resemblances, 229,
infinite magnitudes, 50, 85–6, 134, 231, 233; of temporal properties,
143, 147, 153, 155, 198. See also 14, 230–3, 244; of turtles, 237, 241;
infinite quantities; infinitude vicious, 9, 11, 16, 44–5, 229–30,
infinite potentiality, 49 236, 240, 242–3, 245, 247–8
infinite product, 216 infinite sets, 80, 87, 115–16, 131, 146–7
infinite quantities, 7, 141, 147–8, infinite sums, 20, 26, 174–5, 177–80,
249; area, 23, 86, 183; attenuation 199, 215
coefficient, 251; binding energy, infinitesimal points, 141
184, 189, 207; compressive strength, infinitism, 11
184, 209; density, 7, 154, 159–60, infinitude, 62, 65, 107, 143, 147–8,
211, 249; distance, 189–90, 198–9, 150–1, 155, 223, 233, 236, 247–9;
201, 212; energy, 187–9, 199, 202; of abstract objects, 248; impossible
energy density, 158, 187, 190, 194, forms of, 44–6, 49, 70, 186, 203,
198, 202, 251; expected utility, 214; 232, 252; of natural numbers, 65,
extent, 57, 247–9; force, 160, 184; 148, 150; of past states or events,
heat, 198; inertia, 184; information 234; of rational numbers, 76
processing capacity, 202, 254–5; infinity, 3–260; arithmetic of,
intelligence, 231–2; kinetic energy, 17–18, 87, 89, 90, 144–5, 176; in
155; mass, 51–2, 151, 159–60, 184, mathematics, 4–5; as a number, 133,
189, 252; material strength, 184, 145, 180; in philosophy, 5–6; in the
187, 189–90, 198–9, 201, 207, physical world, 6–7
272 Index
initial conditions, 158–60, 206, 212, 153, 155, 198; infinite extensive,
253, 258 150, 154–5, 231, 249; infinite
instrumental rationality, 107 intensive, 151, 153–5, 158–60,
intensive magnitudes. See magnitudes: 184, 187, 194, 197–8, 201–2, 206,
intensive 218, 225, 230–1, 245, 249, 251–3;
intuitions, 96, 100, 113 intensive, 134–5, 151–2, 154–5,
Iskra, v, 122n6 157–8, 182, 184, 186, 189, 201,
207, 211–12, 225, 231, 252; natural,
James, William, 237 physical, 204
many regarded as one, 111–13
Kant, Immanuel, 41–3, 217, 245 mapping. See one-to-one function
Kelvin temperature scale, 138 Martingale betting system, 29, 43, 85,
kinetic energy, 33, 155 218, 220, 251
kinetic theory of gases, 155 mass, 32, 52, 67, 69, 89, 129, 137,
knowledge: a priori, 97, 103–4, 106–7; 149, 151, 153, 158, 160, 166, 177,
analytic, 99; empirical, 97; synthetic 189, 204–6, 211, 252; critical, 136;
a priori, 97–100, 104 extensive or intensive, 135
mass density. See density
Laraudogoitia, Jon Perez, 211 mathematical systems, argument
Laraudogoitia’s marbles, 32–3, 42, 85, from, 105
211, 251 mathematics, 4, 42, 51, 67, 69, 71–2,
law of non–contradiction, 244–5 89, 93–5, 106–8, 114, 116–18, 132,
laws: of arithmetic, 116, 123–4; causal, 138, 179; formalist approach to, 95;
136, 158, 169, 240; of nature, pure, 122; transfinite, 107, 179
101–2, 138, 153, 158, 160, 240, 252; McTaggart, J.M.E., 14–15, 45, 244–5;
of physics, 58, 102, 225 argument against the reality of
Leibniz, G.W.F., 62, 238–40 time, 230
line segment, 6–7, 18–19, 46, 50, 55, measure theory, 132–3
134, 141 Meno, 56
Littlewood-Ross Paradox, 29–30, 38, mental states, 96
42, 85–6, 89, 179, 201, 204–5, 219, metaphysical indeterminacy, 149
248, 251 metaphysical necessity, 103–4, 106,
Littlewood, J.E., 204 153, 170, 173, 225–6
location theory of points, 254 metaphysical contingency, 103, 106,
Locke, John, 51 212, 225
logical positivism, 94–5, 106 metaphysics, 94
loop quantum gravity, 159 minimum distance theory of space,
Löwenheim-Skolem Theorem, 110–11 55–6
Minkowski spacetime, 64
magnitudes, 51, 65, 129–38, 140, 143, models, 110
150, 152, 155, 157, 166, 177, 180, multiplication, 116–17, 122–4, 127,
182–3, 193, 198, 200, 207, 218, 140–1, 145
225, 249–50; artificial vs. natural,
135–6, 138, 152, 157, 166, 168, natural magnitudes. See magnitudes:
207; dimensions vs. values, 129–30, artificial vs. natural
134–5, 153, 155; extensive, 134–5, naturalism, 95
151, 154–5; finite, 152, 153, 155, necessary truths, 103–4, 106, 124,
183; fundamental vs. derived, 136; 153, 170–1, 173, 200, 225–6
infinite, 50, 85–6, 134, 143, 147, Newton, Isaac, 62
Index 273
Newtonian gravitational theory, 235 Occam’s Razor, 234

Newtonian mechanics, 33, 211 Ohm’s Law, 200–1
Newtonian world, 32, 101, 159, 252 omnipotence, 218
nominalism, 12–13, 247; resemblance, one-to-one function, 75, 77–9,
13–14, 229, 242 114–16, 144, 146, 181
non-standard analysis, 141, 180 ontology, 66, 110, 112, 117, 120, 259
nonzero quantities: deformation, Ontological Argument, 239
209; depth, 211; distance, 55, 169, ordered pairs, 115–16, 130, 140–1
193, 200, 209; mass, 189, 211; ostensive definition, 99, 119
probability, 211, 215, 217, 251;
thickness, 165, 190; time interval, Paley, William, 256–7
253; volume, 162, 164 Paradox of Contact, 168
numbers: anthropocentric conception paradox, defined, 17
of, 65; bifurcated account of, 131; parallel universes, 68
cardinal, 73–4, 85–6, 119–20, 125, Parmenides, 25
129, 143–4, 221–2, 230–1, 248–51; parsimony, 235–6
determinate, 149; finite, 18, 50, pentation, 127
87–8, 128, 141, 143, 152, 180, 190; Perls, Frederick, 3
hyperreals, 4; imaginary, 4, 140; Phenomenal Conservatism, 95–7, 100
indexing use of, 137–8; ineffability philosophy of mathematics, 95, 107
of, 127–9; infinite, 4, 51, 71, 74, 80, philosophy of science, 95
87, 89, 104, 141–5, 149–51, 154–5, photons, 210–11
176, 179–80, 225, 250; infinitesimal, physical necessity, 226
4, 105, 141–2; infinitude of, 50, 148, physical principles, 226–7
150; irrational, 139, 188; natural, physics, 7, 24, 58, 64–5, 68, 102, 155,
50, 65, 67, 69–70, 74, 76–9, 82, 158, 192–3, 225–8, 234; Aristotelian,
86, 88, 114–15, 117, 121, 127–32, 51; laws of, 58, 102, 225
143–6, 148–52, 154, 180–1, 202, Plato, 11–12, 45, 56, 229; theory of
206, 222, 249–50, 254–5; negative, universals, 11, 107, 138, 243, 247
4, 133, 139–40; non–natural, 148; pointless space. See gunk
prime, 103, 153, 171; probability, points, 18–19, 55, 85, 162–5, 168–9,
167; as properties, 120, 129–32, 144; 171–2, 174–6, 178, 233, 250–1,
rational, 76–7, 116, 130, 139, 188–9; 253–4; fusions of, 162, 173;
real, 4, 78–9, 88, 116, 129–34, 137, infinitesimal, 141; paradox of, 18,
139, 140–4, 150, 166, 168, 181, 85, 141, 176; sets of, 162, 173
188–9, 250; as sets, 73–4, 114–18 pointy space, 162–3, 168, 170–1, 254
possibility: degrees of, 167; logical,
objective reality, 107, 171 101, 104–6, 197, 200, 220;
objects, 73; abstract, 4–5, 11–12, metaphysical, 61, 100–6, 131, 159,
66–7, 109, 154, 162, 233, 247–8; 161, 168, 172–3, 177, 183, 199, 204,
composite, 31, 121, 209, 259; 226–7, 233, 236, 256–7; ontological,
concrete, 66–7, 112–13, 120–1, 129, 66; physical, 101, 158–9, 177
166; geometric, 162, 164–5; infinite, possible worlds, 131, 136, 153, 160–1,
148; mathematical, 107, 138–9, 172, 225, 227
162–3; non-physical, 109, 138; potential infinity, 49–50, 53, 57, 62,
non-spatial, 124, 163, 166; open vs. 69, 149, 186
closed, 168–9, 250; size-zero, 190; potentiality, ontological, 66
solid, 161, 171, 191–4, 208; spatial, Powerset Axiom, 84
163–4; three-dimensional, 172 powerset, 80, 82–3
274 Index
presentism, 253 set theory, 71–3, 80, 82–5, 108–18,

Principia Mathematica, 62 132, 149, 179, 248, 259
probabilities, 27, 107, 141, 167, 182, sets, 4, 71–4, 108–18, 120–1, 130,
210, 213–16, 219, 221–2, 235, 251, 133, 143, 147–50, 154, 162, 166,
258; infinitesimal, 142; rational 178, 181, 233, 249–50, 258–60;
distribution of, 216, 222 countable, 110; defined, 71–3;
problem of universals, 243 finite, 80; infinite, 80, 87, 115–16,
property theory of numbers, 131, 144 131, 146–7; open vs. closed, 169;
pure sets, 108–9 Russell, 83–5; set of all sets, 74,
82–3, 90, 113–15, 120, 134, 144;
quantum mechanics, 7, 52, 68–9, 95, singleton, 80, 108, 112, 120, 258–9
155 Smullyan’s infinite rod, 24–5, 86, 94,
quasi-Kantian theory, 41–3 184, 189, 251
question-begging, 142, 232, 239 solidity, 191–4
sophistry, 59, 94, 239
rationalism, 93 space: absolute, 64; empty, 19, 58,
reasons: a priori vs. empirical, 240; 62, 169, 200; gunky, 162–3, 170–1,
infinite series of, 11, 45, 229, 231–4, 173–4, 253; higher-dimensional, 60,
241; justified vs. unjustified, 241 68, 172; infinitude of, 3, 6, 150–1,
referent, 116, 126, 259 153–5, 161; minimum distance
regress arguments, 9–15, 229–46; for theory of, 55–6; pointy, 162–3, 168,
foundationalism, 10–11; against 170–1, 254; relational, 61; round,
Platonism, 11–12; against reality of 58–60; structure of as precluding
time, 14–15; against resemblance contact, 169; three-dimensional,
nominalism, 12–14; for theism, 172–3; volume as measure of,
9–10 164–5, 169, 249
regress, infinite. See infinite series spaceship, 33, 42, 85–6, 211–12,
relational theory of space, 62, 64–5 251
relationship theory of numbers, 131–2 staccato run, 186–8
relativity, theory of: general, 7, 64, 95, string theory, 68, 159, 172
158–9, 190; special, 50 Subset Axiom, 84
resemblances. See nominalism; subtraction, 90, 122, 140, 145
infinite series: of resemblances successive synthesis, 41, 43
Robinson, Abraham, 141–2, 180 sums, infinite, 20, 26, 174–5, 177–80,
Ross, Sheldon, 204 199, 215
Russell, Bertrand, 108, 112, 114, 116, synthetic a priori knowledge, 97–100,
120, 237 104
Russell Set, 83–5 synthetic statements, 97, 99, 104
Russell’s Paradox, 82, 85, 113
temperature, 58, 130, 132, 134–8,
Saint Petersburg Paradox, 26, 85–6, 155–8, 166; Celsius scale of,
213, 221–2, 251 137–8; infinite, 155–8, 249; Kelvin
Schrödinger’s Cat, 69 scale of, 138; negative, 155–7;
science, 49, 93–5 thermodynamic definition of, 156,
scientism, 94 158
semantics, 113–14 temporal points (instants), 253
series: beginningless, 27, 45, 48, 55, temporal properties, regress of, 14,
186; endless, 27, 46, 48, 53, 55, 230, 233, 244–5
184–6. See also infinite series tetration, 127
Index 275
theism, 239–40, 247 volume, 18–19, 23–4, 56, 135–6, 139,

theology, 94 148, 150–1, 163–6, 176–7, 183, 211,
thermal equilibrium, 157–8 249
thermal motion, 166
thermodynamics, 155–8 Weierstrass, Karl, 180
Third Man Argument, 12, 229 Whitehead, Alfred North, 26n11
Thomson, James, 195–7 Wittgenstein, Ludwig, 49n1, 254,
Thomson’s Lamp, 28–9, 38, 42–3, 255n4, 256
85–6, 89, 194–6, 200, 202, 219,
223–7, 248, 251–2 Zeno series, 69–70, 186–7, 251;
time, 7, 47, 57, 69–70, 105, 125; beginningless, 27, 45, 48, 55, 186;
gunky, 253; infinite vs. finite, 231, endless, 27, 46, 48, 53, 55, 184–6
234; infinite series of temporal Zeno, 25–7, 38, 43–4, 52–4, 186, 251;
properties, 14, 230–3, 244–5; Aristotle’s response to, 52–5, 57;
reality of, 14–15, 230, 244, 247; and Cantor’s theory, 85, 86, 89; first
reversibility, 33 paradox of motion, 25–27, 52–6,
transfinite numbers. See numbers: 248; second paradox of motion, 27;
infinite solution to paradoxes, 185–8, 251;
truth regress, 15–6, 45, 230–1, 233, and theory of infinite sums, 26, 180;
242, 244 third paradox of motion, 25n10
Zermelo-Fraenkel set theory (ZF), 84,
uncountable infinity, 77, 80 111
Union Axiom, 84 Zero Argument, 164–5, 253–4
Universal Set, 83–5 zero, 18–20, 124–27; as complete
universals, 5–6, 12, 107, 120, 131, absence, 164–8; duration, 253; and
242–3; Plato’s theory of, 11–12, 229, extension of concept of number,
242, 247 258; mass, 27–8, 88, 189; measure
Unordered Pair Axiom, 84 of, 172; probability, 141, 167, 183,
utility functions, 205–6 215, 251; ratio, 168; real number,
utils, 35 164; size, 7, 18–19, 27–8, 133,
150, 163, 165, 176, 178, 189–90,
van Inwagen, Peter, 94 205, 250; temperature, 138, 156,
velocity, 32, 64, 152, 157, 182, 187, 158, 166; thickness, 88, 170, 184,
191, 212 188–90, 193; utility, 215; velocity,
voltage, 200 24, 160; voltage, 201

Michael Huemer (Auth.) - Approaching Infinity-Palgrave Macmillan UK (2016)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Michael Huemer (Auth.) - Approaching Infinity-Palgrave Macmillan UK (2016)

Uploaded by

Copyright:

Available Formats

Approaching Infinity

Also by Michael Huemer

THE PROBLEM OF POLITICAL AUTHORITY

SKEPTICISM AND THE VEIL OF PERCEPTION

Part I The Need for a Theory of Infinity

1 The Prevalence of the Infinite 3

2 Six Infinite Regresses 9

3 Seventeen Paradoxes of the Infinite 17

3.14 Laraudogoitia’s marbles 32

Part II Old Theories of Infinity

4 Impossible Infinite Series: Two False Accounts 41

5 Actual and Potential Infinities 49

6 The Cantorian Orthodoxy 71

Part III A New Theory of Infinity and Related Matters

7.3 Synthetic a priori knowledge 97

12.2 The paradox of geometric points 176

13 Assessing Infinite Regress Arguments 229

3.1 A magically expanding rod 20

For many years, I have made a point of confounding my philosophy

But almost as common as infinite regress arguments is a certain type of

Two things are infinite: the universe and human stupidity.

1.1 The concept of infinity and the infinite

What is the infinite? The infinite is sometimes described as that which

1.2 The infinite in mathematics

1.3 The infinite in philosophy

Philosophers often discuss propositions, which are usually thought of as

How many universals are there? Infinitely many. Every color is a

1.4 The infinite in the physical world

1.5 The infinite in modern physics

In modern physics, some infinite quantities have been theorized but

2.1 The regress of causes

2.2 The regress of reasons

Suppose you believe some proposition, A, and assume that this is a

that belief is not based on any reasons. This is referred to as a ‘founda-

2.3 The regress of forms

F is some property) is a perfect example of F. For instance, the Form of

2.4 The regress of resemblances

In contrast to Plato, many philosophers have held that universals do

is not a particular, concrete thing; it is a universal. So it seems that there

Stage 1 Stage 2 Stage 3

All of these increasingly complex resemblances must exist. Armstrong

2.5 The regress of temporal series

J. M. E. McTaggart presented a famous argument for ‘the unreality of

e is in the present at the present; however, in the past, e was in the

2.6 The regress of truths

This is the least controversial of the infinite regresses – it is generally

What can we conclude from the preceding six examples of infinite

3.1 A word about paradoxes

In this chapter, I aim to convince you that infinity is puzzling. I start

3.2 The arithmetic of infinity

∞+1=∞ (Equation 3.1)

If equation 3.1 is true, then it seems that we should be able to subtract

1/∞ = 0 (Equation 3.2)

1 = 0·∞ (Equation 3.3)

3.3 The paradox of geometric points

According to standard geometry, space is composed of geometric points.

By the transitivity of equality, 1 = 2.

Figure 3.1 A magically expanding rod

Now imagine that we cut the stick up into single-point-thick slices,

3.4 Infinite sums

Consider the following infinite collection of numbers:

1, −1, 2, −2, 3, −3, ...

1 + (−1 + 2) + (−2 + 3) + (−3 + 4) + ...