Professional Documents
Culture Documents
ETHICAL INTUITIONISM
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
vii
viii Contents
7 Philosophical Preliminaries 93
7.1 Metapreliminaries 93
7.2 Phenomenal conservatism 95
Contents ix
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
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
xi
Preface
xii
Preface xiii
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.
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).
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.
6 Approaching Infinity
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.
1.6 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.
8 Approaching Infinity
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.
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
10 Approaching Infinity
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.
1
Aquinas 1981, I.Q2.A3; Kant 1965, B454; Craig 1991.
2
Russell 1961, 587.
Six Infinite Regresses 11
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.
12 Approaching Infinity
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.
Six Infinite Regresses 13
10
Price 1953; Rodriguez-Pereyra 2002.
14 Approaching Infinity
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):
11
Armstrong 1978, 54–6.
12
Armstrong 1989, 54–7; Price 1953, 23–6.
13
McTaggart 1908.
Six Infinite Regresses 15
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.
16 Approaching Infinity
2.7 Conclusion
15
See, for example, Armstrong 1989, 54; Hochberg 1999, 196; Nolan 2001,
523–4.
3
Seventeen Paradoxes of the Infinite
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:
17
18 Approaching Infinity
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:
(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:
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).
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
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.
20 Approaching Infinity
x
0 r 2r
(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.
Seventeen Paradoxes of the Infinite 21
Second answer: the sum is infinity, because we can group the terms as
follows:
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?
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.
22 Approaching Infinity
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?
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.
Seventeen Paradoxes of the Infinite 23
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.
24 Approaching Infinity
8
Smullyan 2008, 246.
Seventeen Paradoxes of the Infinite 25
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.
26 Approaching Infinity
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.’
Seventeen Paradoxes of the Infinite 27
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.
Stage 1
Stage 2
Stage 3
Stage ∞
the total of all of them must have zero mass and zero volume – so the
material of the stick has been destroyed.12
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.
Seventeen Paradoxes of the Infinite 29
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.
30 Approaching Infinity
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.
Seventeen Paradoxes of the Infinite 31
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.
32 Approaching Infinity
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?
17
This case is from Laraudogoitia 1996.
Seventeen Paradoxes of the Infinite 33
18
Benardete 1964, 149; Moore 1990, 70–1; Oppy 2006, 12.
34 Approaching Infinity
ship could be. Perhaps the ship would go out of existence at that instant,
destroyed by its excessive speed.
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:
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.
Seventeen Paradoxes of the Infinite 35
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.
36 Approaching Infinity
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
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 ∞
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
Seventeen Paradoxes of the Infinite 37
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?
38 Approaching Infinity
3.19 Conclusion
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.
1
Kant 1965, B454; emphasis added.
41
42 Approaching Infinity
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.
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
44 Approaching Infinity
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.
Impossible Infinite Series: Two False Accounts 45
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
Impossible Infinite Series: Two False Accounts 47
‘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
48 Approaching Infinity
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
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
50 Approaching Infinity
Here are some more examples that I think Aristotle would approve of:
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
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)
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.
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.
52 Approaching Infinity
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
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.
Actual and Potential Infinities 53
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.
54 Approaching Infinity
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.
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.
d d
s s
(a) (b)
Figure 5.1 Relationship between the side and the diagonal of a square
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.
Actual and Potential Infinities 57
8
Tooley 1997; Craig 2010.
9
Aristotle 1941, Physics III.6, 206a7–10; VIII.1.
58 Approaching Infinity
10
Aristotle 1941, De Caelo I.9, 278b–279a.
11
Aristotle 1941, Physics IV.6–9.
12
Einstein 1961, chapter 31.
Actual and Potential Infinities 59
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
might still show something of interest. Here they are, along with my
replies:
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.
62 Approaching Infinity
15
See Leibniz’ letters in Leibniz and Clarke 2007.
16
Newton 1846, Book I, 81, from the scholium to the definitions.
Actual and Potential Infinities 63
1 2 3 4
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.
64 Approaching Infinity
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.
Actual and Potential Infinities 65
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:
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.
66 Approaching Infinity
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,
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.
Actual and Potential Infinities 67
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.
68 Approaching Infinity
24
Howell 2014.
25
For discussion of the varieties of multiverse theorized in contemporary
physics and astrophysics, see Greene 2011.
Actual and Potential Infinities 69
5.9 Conclusion
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.
70 Approaching Infinity
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
6.2 Sets
1
Hilbert 1967, 376.
71
72 Approaching Infinity
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
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.
74 Approaching Infinity
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.
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.
The Cantorian Orthodoxy 75
(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.
Many pairs of sets, probably many more than you would have suspected,
turn out to be equinumerous on Cantor’s criterion. Two obvious
76 Approaching Infinity
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 ...
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:
Possible values of n
1 2 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:
Possible values of n
1 2 3 4 5 ...
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 ...
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 ... .
78 Approaching Infinity
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 Cantorian Orthodoxy 79
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
80 Approaching Infinity
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.
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.
The Cantorian Orthodoxy 81
b
b
a c
a e e
d e
d
e c
c d
b
a
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.
12
Russell 1920, 136.
The Cantorian Orthodoxy 83
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
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 Cantorian Orthodoxy 85
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.
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.
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.
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.
88 Approaching Infinity
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.
The Cantorian Orthodoxy 89
6.10 Conclusion
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
93
94 Approaching Infinity
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
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.
96 Approaching Infinity
6
Huemer 2016.
Philosophical Preliminaries 97
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.
98 Approaching Infinity
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
8
See especially BonJour 1998; Bealer 1992; 1999.
9
Incidentally, Quine (1936) showed this last claim to be untenable.
100 Approaching Infinity
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.
102 Approaching Infinity
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.
Philosophical Preliminaries 103
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.
104 Approaching Infinity
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.
Philosophical Preliminaries 105
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
106 Approaching Infinity
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.
Philosophical Preliminaries 107
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?
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.
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:
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
5
Cantor 1932, 204.
112 Approaching Infinity
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
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
114 Approaching Infinity
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.
116 Approaching Infinity
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.
Here are some famous laws of arithmetic: where a, b, and c are any
numbers:
8
Dedekind 1901. Other constructions are possible, but Dedekind cuts are
perhaps the most common way of constructing the real numbers.
Sets 117
9
See Cantor 1915, 85–103; Frege 1980, 81–96.
10
Frege 1980, 81, emphasis mine.
118 Approaching Infinity
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
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
120 Approaching Infinity
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.
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:
3
Frege 1980, 28–9.
Numbers 121
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.
122 Approaching Infinity
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
✈✈ ✈✈✈
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
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
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.
126 Approaching Infinity
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.
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 ( ) .
128 Approaching Infinity
✳1 = 1
✳2 = 2×2 = 4
3
✳3 = 33 = 7,625,597,484,987.
✳4 = 4՛՛4՛՛4՛՛4 = (uncomputable)
⁝
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
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.
(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
132 Approaching Infinity
(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.
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.
134 Approaching Infinity
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.
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.
136 Approaching Infinity
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
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
138 Approaching Infinity
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.
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
〈 〉 + 〈c
〈c , d 〉 = 〈 + c , b + 〉
〈 〉 × 〈〈cc , d 〉 = 〈 − bd , ad + 〉
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.
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.
142 Approaching Infinity
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
1
Cf. Wittgenstein 1975, 157, 162.
143
144 Approaching Infinity
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.
∞ + 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:
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.
146 Approaching Infinity
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.
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.
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
150 Approaching Infinity
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).
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
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
2
Quantum Munich 2014.
3
Giancoli 1988, 425–8.
156 Approaching Infinity
∂E
T= ,
∂S
S ∂S/ ∂E ∂E/∂S
E E E
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’.
158 Approaching Infinity
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
• • •
O5 O4 O3 O2 O1
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
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 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?
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:
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 –
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
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.
168 Approaching Infinity
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
170 Approaching Infinity
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.)
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
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?
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:
6
The argument is based on Arntzenius 2008, Section 10.
174 Approaching Infinity
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
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.
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.
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
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.
Some Paradoxes Mostly Resolved 179
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.
180 Approaching Infinity
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.
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.
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
Some Paradoxes Mostly Resolved 183
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
184 Approaching Infinity
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.
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.
Some Paradoxes Mostly Resolved 185
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.
3
Grünbaum 1968.
Some Paradoxes Mostly Resolved 187
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
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’?
188 Approaching Infinity
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.
5
Oppy 2006, 66–7, 105.
Some Paradoxes Mostly Resolved 189
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.
6
This problem is discussed by Hawthorne and Weatherson (2004).
190 Approaching Infinity
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.
Knife 2 Knife 3
Knife 1
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.
192 Approaching Infinity
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
Some Paradoxes Mostly Resolved 193
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.
194 Approaching Infinity
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,
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
Some Paradoxes Mostly Resolved 195
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
9
Thomson 1954, 5.
10
Benacerraf 1962; Thomson 2001.
196 Approaching Infinity
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).
Some Paradoxes Mostly Resolved 197
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
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.
Some Paradoxes Mostly Resolved 199
Fixed contact
he ht
itc Lig
s
Acme Light
Sw me
Ac
Switches
On Off
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).
200 Approaching Infinity
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.
Some Paradoxes Mostly Resolved 201
1
2
3
4
5
6
8
9
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
204 Approaching Infinity
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.
Some Paradoxes Mostly Resolved 205
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
206 Approaching Infinity
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.
Some Paradoxes Mostly Resolved 207
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.
I I 0e − P x
P( x) = e − μ1x
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?
212 Approaching Infinity
18
Bernoulli [1738] 1954, 24–6, 32–3.
214 Approaching Infinity
∞
⎛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
Some Paradoxes Mostly Resolved 215
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
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
216 Approaching Infinity
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.
⁝
∏ P( O | O
i =1
i 1 & Oi 1 )
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.
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
Some Paradoxes Mostly Resolved 219
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
Now consider the situation three minutes after you started playing.
By this time, you might have attained any of the following possible
outcomes:
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.
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.
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.
Some Paradoxes Mostly Resolved 223
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,
Some Paradoxes Mostly Resolved 225
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:
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
228 Approaching Infinity
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
229
230 Approaching Infinity
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
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).
Assessing Infinite Regress Arguments 233
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.
234 Approaching Infinity
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.
236 Approaching Infinity
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.
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)
Assessing Infinite Regress Arguments 239
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.
240 Approaching Infinity
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.
Assessing Infinite Regress Arguments 241
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.
242 Approaching Infinity
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.
⁝
244 Approaching Infinity
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.
Assessing Infinite Regress Arguments 245
13.5 Conclusion
247
248 Approaching Infinity
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.
250 Approaching Infinity
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.
I am persuaded that the views advanced in this book are by and large
correct, or close enough to correct to advance our understanding.
252 Approaching Infinity
• • •
O5 O4 O3 O2 O1
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
254 Approaching Infinity
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.
256 Approaching Infinity
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’
258 Approaching Infinity
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
260 Approaching Infinity
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
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
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
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