How Real are the Real Numbers?

Jeremy D. Sher
Abstract. Five theories of mathematical ontology are investigated, with the goal of
demonstrating that mathematical items, such as numbers, exist as objects, not merely as
ideas. It is maintained that numbers are as much a part of the world as physical objects
such as a writing table. We reject four of the theories considered; then we give a positive
argument for accepting the theory we endorse.

1. Introduction. Mathematics stands in peculiar contrast to other fields of inquiry,

in that mathematicians accept a platonic ontology of their field with a striking degree of
unanimity. Mathematicians have often insisted that they work with the pure forms of
existence, the basic objects of whose behavior we can always be certain. But we cannot
observe, even in principle, several concepts that have come to be accepted as essential
to mathematical inquiry. One may therefore be tempted to ask whether the accepted
reasoning of mathematics is sound. In particular, infinities, uncountable sets, infinite
series, and some methodologies of analysis and the calculus are quintessentially difficult
to verify from observation, and are therefore in some sense suspicious. We wish to be
convinced that such mathematical objects exist.
When we say in this paper that some entity E exists as an object, we shall mean
to say more than simply that it could be the object of a sentence. We shall mean that
E exists on its own as part of the objective world, regardless of whether any thinking
being is observing or thinking about it. To be part of the objective world, E must also
be publicly accessible, in the sense that anyone can (in principle) become aware of its
existence. Thus, although many mathematical entities are defined in an obscure way,
which is in fact accessible only to relatively few people who have special training, they
may still qualify as publicly accessible as long as they remain accessible to anyone in
principle.
What could mathematical entities be if they are not objects? Some people maintain
that mathematical entities are ideas that exist only in the mind of a thinking being. If
that being were to disappear, then the ideas in its mind would no longer exist either.
We might say intuitively that objects are real,
whereas ideas are imaginary.
In
one sense, it would be easier for philosophy if mathematical entities were ideas rather
than objects: All other commonly accepted objects exist in the physical world, and if
mathematical objects existed, we would have to accept that some objects exist outside
of the physical world.
On the other hand, it would be a worry for mathematics if mathematical entities were
nothing but ideas. Two mathematicians could not converse rigorously about some entity
E, unless E were an object. For instance, consider the set R of real numbers. Since
no one can perceive the existence of R by using the five physical senses, philosophical
empiricism leads us initially to doubt the existence of R as an object. But if R were not
179

180

MIT Undergraduate Journal of Mathematics

an object, no one mathematician could know whether his or her idea of R corresponds
to any other mathematicians idea of R.
We shall refer to the general belief that mathematical entities exist as objects
outside of the material world, but just as real as material objects as mathematical
platonism. The ontologies of full platonism and quasi-combinatorialism fall under this
rubric. We can thus restate our central aim in this paper: to argue for some form of
mathematical platonism.
Our suggested solution to the quandary of mathematical ontology is that mathematical entities exist as objects, even though they are not physical objects. We suggest
that they can be perceived just as surely as I can perceive my writing table through
the faculty of thought, rather than through the five physical senses. The opposing view
is that we can observe such basic concepts as the positive integers through everyday experience, but that we must construct all other mathematical entities from the integers.
This view, known as constructivism or intuitionism, was popular in the early twentieth century. Constructive reasoning excludes such concepts as infinite series, which
cannot be constructed from Z. In intuitionist theory, such concepts are merely ideas;
therefore, they cannot be considered rigorous.
Intuitionist theory deals with infinities as enumerative processes rather than as freestanding objects. In other words, an intuitionist does not imagine an infinite series per
se; rather, he or she imagines an itemized enumeration of terms, beginning with term
one. To intuitionists, every item discussed must actually be constructed. In a proof, a
series may be taken to an arbitrarily large number of terms, but one cannot appeal to
the totality of the series, because one can never construct the full series. Uncountable
sets, by definition, are a lost cause.
In particular, and even more troubling, if we conclude that infinities and uncountable
sets are not objects, we would be unable to reason soundly about them, and we would
then be forced to reject the familiar law of the excluded middle
insofar as our inquiries
about infinity and uncountability take us. By a standard of proof that requires us
to construct entities in full, starting with the integers, before those entities can be
subjected to reasoning, we shall no longer be able to say that either a predicate holds
for all members of an infinite set, or there is at least one member of the set for which
the predicate is false. For infinite sets S under intuitionism, the sense of it is not true
that for all x in S, the predicate P applies to x
shall no longer correspond to there
exists an x in S such that the predicate P does not apply to x
; instead, these two
sentences will make rather different claims. The former will be interpreted as, It is not
the case that, for every x in S, we can show constructively that P applies to all x in
S,
whereas the latter will be read as, We can construct an x in S such that P does
not apply to x.
Unfortunately, the former could be true when the latter is false, so
intuitionism would not allow us to use indirect proofs for infinite sets.
Unlike modern mathematical theory, intuitionism also does not allow us to use the
modern concept of a limit in the traditional sense. Under intuitionism we could write
down the concept of a limit as a statement quantified over all of the sequences terms.
For example, for Cauchy sequences S, which we shall revisit later in the paper, we say
that  > 0, N Z such that |Sn Sm | <  if m, n N . However, under intuitionism
the senses of the quantifiers ,

change, as they changed above. Again 
is
interpreted to mean we can construct,
and 
is interpreted to mean for each ,
we can show constructively.
The intuitionist concept of limit is thus not very useful
to mathematics. In particular, it cannot allow us to define a limiting function g from a

How Real are the Real Numbers?

181

sequence of functions {fn } we could not say that g(x) = limn fn (x).
The notions of uncountability, infinite sets, and infinite series have proven so fruitful
for mathematics that we do not want their use to cast doubt on the soundness of an
argument. However, as every mathematician knows, reasoning from instinct or strong
feeling is a good way to generate concepts, but too error-prone to be useful in proving
theorems about those concepts. How can we be convinced that the mathematical inquiry
of today, which accepts the existence of objects of which we have not a clear physical
intuition, has not fallen prey to this familiar error?
In this paper, we shall investigate five proposed ontologies of mathematics: full intuitionism, semi-platonism, quasi-combinatorialism, full platonism, and formalism. The
terms semi-platonism and quasi-combinatorialism are borrowed from Paul Bernays
[1]. In Section 2, we define these theories and motivate them with a few examples; we
also reject full platonism and formalism because they are inconsistent. Our choice then
becomes whether to accept or reject mathematical platonism. In Section 3, we state
that countable sets are adaptable to intuitionism, and that a satisfactory intuitionistic
account of discrete algebra is known. In Section 4, on the other hand, we find that intuitionism cannot address uncountability. In Section 5, we revisit intuitionism from the
direction of algebraic topology, and we decide that intuitionism is fitted not to algebra
as a field of study, but to the particular set Z of integers. Finally, in Section 6, we
adopt a theory of mathematical platonism, and give a positive argument for believing
it.
2. Intuitionism, Platonism, and Restricted Platonism. Paul Bernays [1] suggests four philosophies of mathematical ontology, ranging from what we shall call full
intuitionism at one extreme to what we shall call full platonism at the other. A full
intuitionist ontology maintains that we can be sure that a mathematical item can be
treated as an object only when it can be constructed from the integers using finitary
methods alone. At the other extreme, a full platonist ontology holds that every mathematical expression denotes an object existing independently of observation or thought,
solely by virtue of being expressed as a formula.
In between, on the side closer to full intuitionism, we have the doctrine of semiplatonism or nonconstructive reasoning, which holds that nonconstructive existence
proofs are admissible as long as they are based strictly on finitary reasoning. (A nonconstructive existence proof argues for the existence of an object with certain properties
without actually presenting the object.) It is important to note that semiplatonism does
not absolve us of the intuitionistic obligation to reason finitarily that is, to reason in
a step-by-step process that terminates and does not appeal to infinities. It differs from
full intuitionism only in those cases where we want to prove the existence of an object
without giving a full construction of that object.
Finally, between nonconstructive reasoning and full platonism lies the philosophy of
quasi-combinatorialism, which allows us to reason about infinite sets and sequences
as long as the reasoning is motivated in some way by a finitary case. For instance,
Bernays gives the example of a function f : Z Z as quasi-combinatorial; we assume
its behavior is in some sense analogous to that of g : Z/nZ Z/nZ for n N. Quasicombinatorialism allows us to reason with uncountable sets as well as with countable
ones. Unlike full platonism, quasi-combinatorialism does not maintain that entities
exist as objects solely by virtue of being expressible in formulas. The criterion that
an item be motivated by a finitary case
is weak; all it accomplishes is to disallow

182

MIT Undergraduate Journal of Mathematics

self-contradictory concoctions like the Russell set
(discussed below) that cannot have
any resemblance to a finitary case. Intuitively, quasi-combinatorialism maintains that
mathematical objects exist as long as they make sense.

A fifth ontological theory is formalism, which argues that there are no mathematical
objects at all, but that mathematics is a sort of symbols game
that exists only on
paper. Formalists intend not to degrade mathematics indeed, this particular symbols
game
is very important to society but they hold that the symbols do not denote any
particular object or idea. Formalism differs from full platonism in its nonrecognition
of mathematical objects, but it echoes full platonism in granting the same status to
all mathematical formulas, regardless of whether they are finitarily constructable from
integers.
As Bernays notes, paradoxes have been found that show that full platonism is an
incorrect ontology of mathematics. One such paradox is the case of the Russell set

[4], which is expressed by R = {S|S 6 S}. If R R, then, by definition, R would have
to be a set such that R 6 R, which is a contradiction. On the other hand, if R 6 R,
then R must not fulfill the criterion that R 6 R, which is also a contradiction. The
Russell set
cannot exist, for its very existence would lead to a contradiction. But full
platonism would maintain that it exists as an object, since it is expressible in terms of
mathematical symbols. Therefore, argues Bernays, we cannot accept full platonism and
must decide among the other ontologies. Full platonism really is an extreme position; it
is possible to hold platonic beliefs about the existence of mathematical objects without
accepting that self-contradictory expressions such as the Russell set
represent objects.
Therefore, we accept Bernayss rejection of full platonism.
The paradox of the Russell set
also leads us to reject formalism, because formalism fails to distinguish between paradoxical and nonparadoxical formulas. The Russell
paradox forces us to ask metaphysical questions that formalism does not address. In
particular, by rejecting the Russell set
as meaningless, we necessarily concede that
nonparadoxical formulas have some meaning beyond their printed forms.
Bernays suggests that intuitionism fits number theory well [1, p. 269], but that we
should accept quasi-combinatorialism for fields involving the geometric idea of the continuum,
by which he apparently means analysis, topology, and other fields that appeal
to infinities and uncountabilities [1, pp. 26869]. Semiplatonism, he argues, is well suited
for algebra. However, because the subfields of mathematics are richly interrelated, it
makes little sense to accept different ontologies for different subfields. Therefore, while
we recognize that the three candidate ontologies full intuitionism, semiplatonism, and
quasi-combinatorialism are each perhaps more immediately applicable to a particular
subfield of mathematics, we want to adopt a single mathematical ontology. In particular,
we want to render judgment on mathematical platonism.
3. The Adaptability of Countable Sets to Full Intuitionism. As Bernays points
out, the integers Z are well adapted to a full intuitionist ontology. Although traditional
algebra and number theory have accepted Z as simply an infinite set, Bernays states that
Brouwer and Kronecker developed a new system of number theory that does not rely
on infinite sets. Rather than regard Z as infinite, their method relies on ongoing, but
perpetually unfinished, enumerations of integers, so that an enumeration will at some
point consider a set that is arbitrarily large, but not infinite. Under this interpretation,
a theorem stating that for all integers n in Z, the predicate P applies to n, is understood
to mean that if we put any individual number in place of n, we find that P applies to n,

How Real are the Real Numbers?

183

as opposed to understanding it as making a statement about the infinite set Z. Also, a

theorem stating that there exists some integer n such that P applies to n is understood
as a single statement about a particular, but unspecified, number n; as before, it is not
understood as a statement about Z. Brouwer and Kronecker, says Bernays, have shown
that all the expected theorems of number theory still hold under this full intuitionist
interpretation [1, pp. 26263].
Since the integers are adaptable to a full intuitionist interpretation, other countable
sets must be similarly adaptable. Some intuitionists might worry about whether
sets

like the rationals Q, which are dense, or the Gaussian integers G = Z + ( 1)Z, are
adaptable to full intuitionism, since their elements do not correspond to the physical
reality that we can sense. But consider, for example, Q. Because Q is countable, we can
set up an injective function : Q , Z and translate all our theorems about elements
of Q straightforwardly into theorems about elements of Z, which are all adaptable to
full intuitionism. We can do this intuitionistically as long as the function itself is
constructive. In the case of : Q , Z we can give the familiar algorithm whereby one
associates each lattice point (m, n) with a rational number m/n Q and enumerates
the lattice points outward from the origin (ignoring duplicates and rational expressions
with 0 in the denominator).
In this way, we can use intuitionism to speak about the elements of any countable set
that can be enumerated with a constructive algorithm. For such sets, therefore, we can
immediately satisfy any concerns over whether the traditional platonic assumptions are
justified; we may simply refrain from using those assumptions to speak about countable
sets.
4. Uncountability: Beyond the Reach of Intuitionism. There is no intuitionistically satisfactory account of the real numbers R. The two familiar constructions of the
reals from the rationals the method of Cauchy sequences and the method of Dedekind
cuts both appeal to concepts that are intuitionistically undefinable.
The former method begins by constructing Cauchy sequences of rational numbers.
A sequence {Sn } of rationals is a Cauchy sequence if, for every  > 0, there exists an
integer N such that |Sn Sm | <  if m, n N . A decimal expansion, for example, is a
Cauchy sequence of rationals: for any , any N such that 10N <  will do. We define
the real numbers R as the set of equivalence classes of Cauchy sequences of rationals,
where two sequences Sn and Tn are said to be equivalent if limn (Sn Tn ) = 0. (It is
necessary to define equivalence classes to ensure that pairs of decimal expansions such
as 1.000000 and 0.999999 refer to the same real number.)
We have remarked earlier that intuitionism does not allow the concept of a limit.
Indeed, the idea of limit entails a conception of infinite sequences as objects in themselves. If the enumeration of the sequence is never finished, how can we speak of the
sequence as a fully existing object, capable of having a value? Most mathematicians
would reply that even though the full computation of the sequence cannot even in principle be finished, we can still manipulate, compare, and equate infinite sequences by
making statements (such as the criterion for a Cauchy sequence) that are quantified
over all of the sequences terms. But this is exactly the quasi-combinatorial assumption.
The method of Dedekind cuts defines a real number as a particular kind of set of
rational numbers, called a cut. A cut of rationals is a set Q such that has no
largest element, is neither nor Q, and if p , then every q Q with q < p is also
in . We define R as the set of all Dedekind cuts of rational numbers.

184

MIT Undergraduate Journal of Mathematics

We can define all of the familiar operations and relations on R. Let us give a few
for illustrative purposes; we omit the proofs, which can be found, for example, in Rudin
[3]. We define an ordering relation < by putting < whenever Q. Addition
is defined by + = {p + q | p , q }. Multiplication can also be defined, but its
definition is cumbersome and not important for our purposes, so we omit it. The reals
also have the least-upper-bound property; that is, every
S nonempty subset S of R that
is bounded above has a supremum R, where =
{ | S}. In fact, the set
R of Dedekind cuts has all of the properties we would expect in the real numbers.
The intuitionistic trouble with the Dedekind method involves the fact that each cut
is an infinite set. As such, it cannot be manipulated as a self-contained object because it
must forever be stuck in the process of enumeration. To speak of the least upper bound
of a set of elements whose constructive enumeration is never finished also requires the
quasi-combinatorial assumption.
The trouble we have in finding an intuitionistic account of the reals is an example of a
larger problem with full intuitionism. Because of its insistence on constructive methods
and finitary reasoning, full intuitionism is inextricably wedded to the countable. We
cannot compare uncountable sets intuitionistically. We cannot even give intuitionistic
meaning to AB for uncountable sets A and B because we cannot always give a process
for constructing an element that is in A, but not in B. We can never, through the
step-by-step methods that full intuitionism requires, find all elements of A that are not
in B, because constructive methods allow us only a countable number of steps.
As an example, we can construct irrational real numbers in addition to the rationals, but we can only ever construct countably many of them. We can construct the
square-root of every integer geometrically by using Pythagorean arguments; many of
these numbers are irrational. However, we cannot obtain more than countably many irrationals in this way. In particular, this method will not give us transcendental numbers
like or e.
Through each constructive process, we can generate countably many elements, but
even if we carried out several countable processes simultaneously, any union of countable
sets is countable, so we will never generate more than countably many elements with
any finitary process. The constructive methods of intuitionism, which are by definition
countable, can never describe an uncountable set. Intuitionism therefore proves to be
of little use as far as uncountability is concerned.
Let us note as well that insofar as full intuitionism cannot describe uncountable
sets, neither is the doctrine of semi-platonism useful. Semi-platonism requires us to
reason finitarily, just as full intuitionism does. Therefore, a semi-platonist conception
of a Dedekind cut or a Cauchy sequence would encounter the same problems that we
observed with full intuitionism above. Semi-platonism, like full intuitionism, is wedded
to countability, and cannot describe uncountable sets.
5. Finitary Intuitionism Revisited. We have remarked earlier that a central value
of mathematics lies in the applicability of its subfields to one another. We have mentioned that intuitionism offers a satisfactory explanation of algebra on the integers. Let
us therefore turn to algebraic topology, in which the normal algebraic relationships are
applied to groups defined in terms of topological spaces of uncountable cardinality.
Algebraic topology associates to every topological space a fundamental group, a
group not of integers, but of paths (more precisely, of equivalence classes of paths) in
the topological space. Topology very often uses uncountable sets as its spaces. Nev-

How Real are the Real Numbers?

185

ertheless, we can consider spaces whose fundamental groups are isomorphic to groups
with elements such as those used in the intuitionistic account of algebra. We would like
for our theorems about such groups to hold regardless of whether their subject matter
is topological spaces or integers.
We therefore question Bernayss contention that full intuitionism is adequate for
algebra. Bernays states that algebra is in some way a natural area for full intuitionism,
while analysis is not. However, algebraic topology presents us with groups whose subject
matter cannot be described by the constructive methods of intuitionism. Therefore, it
appears to be the set Z of integers, rather than algebra itself as a field of inquiry, that
intuitionism is particularly suited to describe.
This finding does not surprise us, because intuitionism is characterized by its insistence on step-by-step constructive proofs, whose steps and subject matter can be
enumerated straightforwardly with integers. It may be a fact about algebra that many
groups are isomorphic to groups that are products of Z/nZ, and that intuitionism can
offer insight about these groups. However, it must be remembered that intuitionism is
wedded not to algebra, but to the particular set Z.
It might be objected that intuitionistic reasoning still is a useful way to think about
algebra. We agree, and affirm that very often we do want a constructive proof when
possible. There is something intellectually satisfying about solid constructive reasoning.
Nonetheless, this has little bearing on the philosophical question at hand. Certainly we
may continue to use intuitionistic reasoning where it is appropriate, but we must accept
only one ontology of mathematics.
Hence, neither full intuitionism nor semiplatonism is especially useful as a theory of
mathematical ontology. The set Z is only one of many sets and objects that comprise the
subject matter of mathematics. An acceptable ontology of mathematics must account
in some way for, or at least be able to describe, uncountable sets.
6. In Defense of Mathematical Platonism. Of the four ontologies of mathematics
that Bernays describes, the only one that remains acceptable is quasi-combinatorialism.
Quasi-combinatorialism rejects self-contradictory notions such as the Russell set,
but
allows us to speak about whatever infinite sets we want, countable and uncountable.
Bernays may be correct that intuitionism is an alternative way of thinking about certain countable sets, but this observation is ultimately tangential to the question of
mathematical ontology. To ensure consistency and interapplicability, mathematics must
have only one ontology. Now, therefore, we must decide whether to accept or reject
mathematical platonism in the form of Bernayss quasi-combinatorialism.
If we were to reject platonism in mathematics, then we would not be left with very
much of interest. Much fruitful and applicable mathematics has come of uncountability,
including mathematics appliable to science and engineering in the physical world. The
real numbers have proved particularly indispensable, both to mathematicians and to
nonmathematicians.
Attacks on platonism are common, mainly because platonism is not clearly justified
as a theory about the physical world. The platonic statement that there exists an ideal
object called horseness,
and that every actual horse is to some extent an example
of horseness
seems more an arbitrary postulation than a theory arrived at through
inquiry. Insofar as we consider physical objects, platonism does not seem to coexist well
with scientific thought. After all, the science of the physical world is essentially about
observation, and we do not observe horseness
or any other platonic ideal at all.

186

MIT Undergraduate Journal of Mathematics

However, in the realm of mathematics many of the familiar arguments against platonism do not apply, since mathematical entities do not admit direct observation. Mathematics is exactly that discipline that deals with ideal entities; mathematical laws are
not affected by human observation with the five senses. Physical observation is not a
relevant epistemological criterion for mathematics. This observation does not establish
that platonism is the correct ontology of mathematics, but it does show that we have
significantly less reason to attack mathematical platonism than we have to attack platonism in other fields. Mathematical platonism should not be dismissed outright, nor
treated with the same suspicion that would be appropriate for platonic ontologies of the
physical world.
Let us now give a positive argument for accepting quasi-combinatorialism. As we
have shown above, since no finitary method can construct a transcendental number, any
ontology that rejects quasi-combinatorialism must treat transcendentals as nothing more
than ideas, and thus as unfit for rigorous reasoning. We intend to show that at least one
transcendental is fit for rigorous reasoning by showing that people may gain information
about through the empirical process. We will then show that empiricism renders any
doubts about quasi-combinatorialism specious. It follows that modern empirical theory
leads us to accept mathematical platonism in the form of quasi-combinatorialism.
People can and do gain an empirical understanding of the transcendental number
. Empirical observation cannot give us knowledge of the exact value of , but the
empirical process leads us to strong belief that there is such a constant, whose value
is approximately 3.14, which is the ratio of the circumference to the diameter of every
circle. The idea of has so many practical applications, is so universally shared, and fits
so well with peoples observations that we find ourselves believing strongly that it corresponds to something more than an idea. Because of its indispensability, it is reasonable
to believe that exists as an object, just as it is reasonable to believe that a writing
table exists. Although we cannot evaluate empirically, we can fulfill the intuitionists
criterion for rigorous reasoning: We can be sure that when two mathematicians talk
about , they are both referring to the same numerical value; that is, the ratio of the
circumference to the diameter of a circle.
Thus, with a grounding in empiricism, we accept that has a constant value. But
we may also use Leibnizs formula [2]
= 4(1 1/3 + 1/5 1/7 + . . . )
to express in terms of a Cauchy sequence:
S1 = 4(1)
S2 = 4(1 1/3)
S3 = 4(1 1/3 + 1/5)
..
.
Sn = 4 1 1/3 + + (1)n+1 /(2n 1)

The sum of all the elements of this series exists as an object, since it has a value;
that value is itself. Since limn Sn exists as an object, we may add and multiply it
with other numbers without worrying about forever evaluating the terms of an infinite
series.

How Real are the Real Numbers?

187

We now have three options for accepting or not accepting the existence as objects
of limits of Cauchy sequences of rational numbers. We could echo intuitionism and
maintain that a number given only as a sum of an infinite series can never be an object.
We could maintain that = limn Sn is the only limit of a Cauchy sequence that
exists as an object, and that all other limits of Cauchy sequences are merely ideas. Or,
we could maintain that all limits of Cauchy sequences exist as objects.
We cannot accept the first option because we already have empirical grounds for
believing that exists as an object. Meanwhile, the second option seems quite specious.
If we did not believe were an object, we might regard the third option as specious,
but since we do, we find ourselves asking whether anything is really so special about
the sequence S1 , S2 , S3 , S4 , . . . that could distinguish it ontologically from, say, the
sequence S1 + 1, S2 , S3 , S4 , . . . . We should not have some limits of Cauchy sequences
existing and others not existing. To maintain that a limit of a Cauchy sequence pops
out of existence if we change a few of the numbers is like maintaining that a table pops
out of existence if we paint it a different color. The position that the limits of some
Cauchy sequences are objects and the limits of others are ideas makes little sense, and
we reject it as specious.
In this way, we establish empirically that every real number exists as an object, for if
ever the limit of Cauchy sequence in Q exists, then the real numbers follow, as discussed
above. Thus we are able to answer the question posed in the title of this paper: The real
numbers are as real as anything else in the world, along with all of the marvelous things
they bring with them. Uncountability is a reality, not a dream. Nowhere-differentiable
curves exist, although we can never hope to see them. The weird twists of topology,
the stately citizens of algebra, and the bewilderingly complex maps of chaos, all exist.
Each of these strange and wonderful marvels is a part of our world.
References
[1] Bernays, Paul, On Platonism in Mathematics,
Lecture, June 18, 1934, Conferences internationales des Sciences mathematiques, University of Geneva,
published in Benacerraf, Paul and Hilary Putnam, ed., Philosophy of Mathematics: Selected Readings (2nd edition), Cambridge University Press, 1997, pp. 258
71.
[2] Leibniz, Gottfried, Letter to Isaac Newton, 1673.
[3] Rudin, Principles of Mathematical Analysis (third edition), McGraw-Hill, 1976.
[4] Russell, Bertrand, Letter to Gottlob Frege, 16 June 1902.

188

This page will be blank.