6

Russell’s Logicism
Kevin C. Klement

Published in Russell, Wahl (ed.), The Bloomsbury Companion to Bertrand

Russell. London: Bloomsbury, 2019, pp. 151–178. This is the author’s own
typesetting, with similar style and matching pagination.

1 Introduction

Logicism is typically defined as the thesis that mathematics reduces to, or is an

extension of, logic. Exactly what “reduces” means here is not always made en-
tirely clear. (More definite articulations of logicism are explored in Section 5.)
While something like this thesis had been articulated by others (e.g., Dedekind
1888 and arguably Leibniz 1666), logicism only became a widespread subject of
intellectual study when serious attempts began to be made to provide complete
deductions of the most important principles of mathematics from purely log-
ical foundations. This became possible only with the development of modern
quantifier logic, which went hand in hand with these attempts. Gottlob Frege
announced such a project in his 1884 Grundlagen der Arithmetik (translated as
Frege 1950), and attempted to carry it out in his 1893–1902 Grundgesetze der
Arithmetik (translated as Frege 2013). Frege limited his logicism to arithmetic,
however, and it turned out that his logical foundation was inconsistent.
Working at first in ignorance of Frege, Bertrand Russell’s interests in the
fundamental principles of mathematics date back to the late 1890s. He pub-
lished An Essay on the Foundations of Geometry in 1897 and soon thereafter
began work on the nature and basis of arithmetic.1 He found his work ham-
pered somewhat by a mismatch between his earlier generally Kantian views on
mathematics and the staunch realist metaphysics he had adopted at the end
of the 1890s largely under the influence of G. E. Moore.2 This tension ended
when Russell attended the International Congress of Philosophy in July 1900.
There Russell became acquainted with Giuseppe Peano and his work, which
Russell describes as one of the most significant events of his philosophical ca-
reer (MPD 11). After being exposed to and quickly mastering Peano’s style of
152 The Bloomsbury Companion to Bertrand Russell

symbolic logic, Russell became convinced that it was possible to analyze all the
concepts of pure mathematics in logical terms, and provide deductions of them
from logical axioms.3
The result was Russell’s first major work of lasting significance, The Principles
of Mathematics (POM), published in 1903. In it, Russell laid out his views on
philosophical logic and argued informally for logicism. He described the project
as offering in part:
the proof that all pure mathematics deals exclusively with concepts de-
finable in terms of a very small number of fundamental logical concepts,
and that all its propositions are deducible from a very small number of
fundamental logical principles. (POM v; cf. §434)

His initial plan was to carry out the actual deductions symbolically in a sec-
ond volume, and recruited his Cambridge colleague and former teacher, A. N.
Whitehead, as a co-author. By the time their technical work was ready for pub-
lication, it had grown so large, and their views had changed so significantly in
response to certain paradoxes (see Section 3), that they decided to rename their
work Principia Mathematica (PM). It was published in three large volumes from
1910 to 1913. (A proposed fourth volume, which was to be Whitehead’s re-
sponsibility primarily, was never produced.) Russell later describes the process
as having been exhausting, and seems to have been ready to retire from tech-
nical work in mathematical logic after its publication. Nonetheless, in 1919,
he published Introduction to Mathematical Philosophy, which provides an in-
formal introduction to the theories and results of PM, and in the mid-1920s,
Russell oversaw the publication of a second edition of PM, wherein he added
a new introduction and various appendices exploring certain new theories and
made note of what he saw as important contributions made by others in the in-
termediate decade. While Russell did not write any additional lengthy pieces on
the philosophy or foundations of mathematics after the second edition of PM,
comments on various issues concerning the nature of mathematics are scattered
among his other writings.
Russell’s attempt to establish logicism in PM, and the logical system devel-
oped therein for the purposes of the project, have had a remarkable impact
on the history of logic and philosophy. Many of the most important figures
in the history of analytic philosophy (from Wittgenstein and Carnap to Quine
and Putnam, etc.) describe Russell’s logical work as a significant influence. The
logic of PM is perhaps the single greatest inspiration for the kind of predicate
logic nearly all philosophy students are now required to learn. While it is not
 Russell’s Logicism 153

widely accepted today that PM in fact succeeded in establishing logicism, I think

a case that it did can still be made.

2 The regressive method and its prior successes

According to Russell, progress in mathematics can proceed in either of two di-

rections (IMP, 1–2). Most research proceeds in the “forward” or “constructive”
direction, in which new results are proven from previously known findings. In
this way, one attains the results of newer, more complex, and “higher” math-
ematics. Russell saw his own work, however, as proceeding in a different di-
rection, employing a “regressive” or “analytic” method. Here, one begins with
a body of knowledge already accepted as at least mostly true. The goal is to
work backwards from these known results to a deductive basis for them, a set of
principles more general, less complex, and employing a smaller undefined vo-
cabulary. If it can be shown that the original body of knowledge—or at least all
of it deemed worth preserving—can be deductively recovered from this more
austere basis, we gain insight into the real nature of the truths. Logical con-
nections between the terms employed are revealed. The new basis might also
remove puzzling or unwanted aspects of the original theory. Finally, the pro-
cess organizes the new theory as a deductive system, which invites and facilitates
the discovery of new results.4
Russell portrays the truth of logicism as something emerging from already-
established results in the “regressive” or backwards journey from accepted
mathematical principles to their deductive origins. Russell situates his work as
the next in a series of recent successes in that direction (see Papers 5, 574–78,
IMP, chaps. 1–2).
According to Russell’s story, a number of late nineteenth-century math-
ematicians, including Dedekind, Weierstrass, and the members of Giuseppe
Peano’s Italian school, had succeeded in “arithmetizing” most of pure arith-
metic, and even many areas of pure geometry. That is, they showed how these
areas of mathematics could be deductively recovered starting only with the ba-
sic principles governing natural numbers. Moreover, Peano and his associates
had made further “regressive” progress in number theory by reducing its basis
to five principles, now known as the Peano-Dedekind axioms:

1. Zero is a natural number.

2. Every natural number has a successor, which is also a natural number.
154 The Bloomsbury Companion to Bertrand Russell

3. Zero is not the successor of any natural number.

4. No two natural numbers have the same successor.
5. Whatever holds of zero, and always holds of the successor of any natural
number of which it holds, holds of all natural numbers. (The principle of
mathematical induction.)

At the time, it was widely believed that all of pure arithmetic could be derived
from these principles together with the usual assumptions of logic and set or
class-theory (and even today, it is generally agreed that no axiomatic system for
number theory represents an improvement on this).
However, according to Russell, the work of Peano’s school had not com-
pleted the backwards journey to the basic principles of mathematics, because an
even more minimal basis is possible. Russell portrays Frege as having taken the
next step in this direction, by showing how what amount to Peano’s “axioms” for
number theory could be derived from logical axioms, removing the need for any
basic assumptions beyond the purely logical. Achieving this next step required
first a way of defining the non-logical terminology used in Peano’s postulates:
“zero”, “successor” and “natural number”. There are in fact different ways of
defining these, but let us first focus on natural numbers considered as cardinal
numbers. Taking Frege’s talk of “extensions of concepts” in his Grundgesetze as
interchangeable with talk of “classes”, Frege defined the cardinal number of class
α as the class of all classes whose members could be put in 1–1 correspondence
with the members of α. Zero could then be defined as the cardinal number of
the null class. The relation of successor could be defined as holding between n
and m when there is a class α and member of that class x such that n is the num-
ber of α and m is the number of the class consisting of all members of α except x.
In that case, the members of n are all those containing one more member than
contained in members of m. The natural numbers could be defined as all those
(cardinal) numbers following zero in the series created by the successor rela-
tion: zero, zero’s successor (1), zero’s successor’s successor (2), and so on (Frege
2013, §§41–46; cf. Frege 1950, §§73–83). In particular, n is a natural number
when it has every property possessed by 0 and always passed from something
having it to that something’s successor. Frege also provided an axiomatic sys-
tem for logic, including principles governing classes, in which Peano’s “axioms”,
interpreted using these definitions, could be derived as theorems.
Frege’s proposed basis for arithmetic, although fully logicist (in not employ-
ing any specifically non-logical vocabulary or non-logical principles) and clearly
 Russell’s Logicism 155

more “regressive” or “analyzed” (in Russell’s sense of these words) than Peano’s
basis, was flawed in that its assumptions about classes (or “extensions”, as Frege
called them) led to contradiction. In particular, Russell discovered that Frege’s
Basic Law V led to the antinomy now known as “Russell’s paradox” (discussed in
sec. 3 below). Russell communicated this problem to Frege in June 1902 (Frege
1980, 130–33), and Frege himself admitted the flaw in his system (Frege 2013,
vol. 2, afterword). Russell sees his work as an attempt to improve upon Frege’s
advance by taking yet another “regressive” step towards fundamentality. Rus-
sell sought to recover what was worthwhile in Frege’s proposed basis but at the
same time purge it of inconsistency by making use of a yet even smaller logical
basis, discarding Frege’s problematic assumptions about classes. Here we have
an example of how the regressive method, as Russell sees it, does not necessarily
preserve every aspect of the original body of knowledge being analyzed. It can
reasonably be hoped that sometimes the analytic process might remove prob-
lems with the original theory or show dubious assumptions to be unnecessary.
Changes to the original body of knowledge could be acceptable if what was the-
oretically important was preserved.
Considered as anything like a historical record of the progress in research
in the foundations of mathematics, this story Russell tells must be taken with
several large grains of salt. Frege did not see himself as making further progress
on a basis established by Peano; Frege’s core theses on the nature of arithmetic
had been formed, and at least partially defended, before Peano’s major works
were published. For his own part, Russell had formed the logicist thesis and had
worked out logical definitions of the key concepts of number theory indepen-
dently from Frege, and only discovered Frege’s work later (POM xviii). While
Russell himself discovered Peano before Frege, and no doubt found Frege to be
an improvement over Peano once he had found him, this does not describe very
accurately how anyone else would have surveyed the developments. Moreover,
it has been suggested that Russell’s later depiction of his work on higher math-
ematics as merely exploiting the “arithmetization” already achieved in the late
nineteenth-century does not do adequate justice to the strides towards gener-
ality achieved even by Russell himself (and his co-author Whitehead) in their
technical work.5 Nonetheless, this story, as misleading as it is, tells us something
about how he understood the achievement of his work and what he counted as
“success” when it came to philosophical research in mathematics. It also under-
scores what for him would have been the most important part of mathematics
to be shown to have a logical foundation: the basic principles of number theory.
156 The Bloomsbury Companion to Bertrand Russell

Russell’s regressive method also has implications for how he understood the
epistemological dimensions of his project.6 Unlike Frege (at least as usually
read), Russell was not attempting to argue that simple workaday mathematical
truths such as that 2 + 3 = 5 gained their epistemological justification from
the more general logical axioms that could be used to deduce them. In fact,
Russell saw the issue the other way around: the obviousness and certainty of
such elementary mathematical truths provide “inductive support” for a more
minimalist theory or list of basic assumptions from which they can all be derived
(Papers 5, 572–77).

3 The paradoxes and the development of Russell’s logic

Russell describes the period just after his discovery of Peano’s work in 1900 as a
kind of intellectual honeymoon. Peano’s logic allowed him as he saw it to give
precise definitions for all of the fundamental notions of mathematics and he
“discovered what appeared to be definitive answers” (Auto, 148) to the philo-
sophical doubts he previously had entertained. By the end of this year, he re-
ports (MPD, 73; Auto, 148), he had finished an initial draft of what was to be-
come POM. What ended the honeymoon was the discovery in 1901 that the
assumptions he had been making about the existence of classes, and other en-
tities such as predicates and propositions, were impossible because they led to
contradictions (and that the assumptions made by Peano and Frege were simi-
larly flawed).
The best known difficulty is now known as “Russell’s paradox of classes”. If
we consider it intelligible to ask whether or not a given class x is a member of
itself, x ϵ x, and consider every intelligible condition (every open sentence) to
define a class, then we can consider the class of all such x which are not members
of themselves—which Russell writes as “x̂(x ∼ϵ x)”—and ask whether it is or is
not a member of itself. By the naïve assumptions he made at the time, it would
be a member of itself just in case it met its defining condition, i.e., just in case it
was not a member of itself. This is a contradiction. A similar problem threatens
if one assumes that there is a quality or monadic universal, or “predicate” as
Russell would say at the time, corresponding to every condition defined by an
open sentence. It would seem then that there is a quality q which a predicate
a has just case a does not have quality a. But then we get a contradiction if we
consider whether or not q has quality q (POM, §101). This version has come to
be known as “Russell’s paradox of predication”.
 Russell’s Logicism 157

Russell discovered these contradictions by considering Cantor’s result that

every class has more subclasses than members (POM, §100, MPD, 75). Can-
tor’s reasoning for this result proceeded by showing that a contradiction results
from the assumption that any class has as many members as subclasses as fol-
lows. Suppose there is a mapping from subclasses to members that yields a dis-
tinct member for every subclass. One can then define a “diagonal” subclass s
of all those members which are not in the subclass mapped to them. Since s is
a subclass, it itself should be mapped to some member, a. But then we can ask
whether or not a is in s: since s is defined as the class of all those members not
in the subclass mapped to them, a would be a member of s just in case it isn’t.
Russell arrived at his contradiction more or less by applying Cantor’s argument
to such classes as the universal class or the class of all classes, for which it would
appear impossible that they would have more subclasses than members, since
their subclasses are members.7
Russell also discovered more complicated versions of similar kinds of prob-
lems, such as a paradox of relations (Papers 5, 588), and another involving
propositions (POM, §500). At the time of POM, Russell considered a propo-
sition to be a mind-independent complex object somewhat like a state of af-
fairs. If propositions are genuine objects, by Cantor’s result there ought to be
more subclasses of the class of all propositions than members, i.e., there ought
to be more classes of propositions than propositions. However, it seems pos-
sible, for each class of propositions, m, to generate a distinct proposition, such
as the proposition that every member of the class is true, i.e., (p)(p ϵ m ⊃ p).
Cantor’s reasoning would invite us to consider the class w of all propositions
of the form (p)(p ϵ m ⊃ p) not in the class m they are “about”. Now we con-
sider the proposition (p)(p ϵ w ⊃ p) and ask whether or not it is in w, and a
contradiction arises from either answer.
Russell mentioned these problems in POM, and made certain suggestions
in the direction of solving some of them, but even by his own admission,
he needed to do more work to provide a complete solution before the for-
mal project of PM could proceed. Russell found this task more difficult
than he anticipated, and attempting to solve such paradoxes was his chief
intellectual preoccupation from 1903 to 1907. In his Autobiography (154),
he describes himself as having spent the period largely staring at a blank
page, failing to make progress. In fact, Russell considered a large variety
of different solutions during this period, but usually found that his attempts
were only partly successful, solving some but not all versions of the para-
doxes. For example, the rather elegant theory found in his 1906 withdrawn-
from-publication paper “On the Substitutional Theory of Classes and Relations”
158 The Bloomsbury Companion to Bertrand Russell

(Papers 5, 236–61) solved the paradoxes involving classes and predicates, but fell
prey to complicated versions of the kind of paradox of propositions mentioned
above.8 Most of the manuscripts in which Russell explored various options are
now published in Papers 4 and Papers 5. Many of his attempts are fascinating,
and still worthy of further study, but they cannot be discussed in detail here.
Central to the solution Russell settled on for PM was the suggestion that
many apparent “abstract” entities such as classes and propositions are not gen-
uine entities at all, but only “logical constructions” or “logical fictions”. Appar-
ent terms for them are not genuine terms, but “incomplete symbols”, meaningful
not by naming entities but rather by contributing to the meaning of statements
in which they appear in a more complicated way. PM *20.01 offered the follow-
ing contextual definition of a class abstract “ẑ . . . z . . . ” for a class of individuals:

f {ẑ(ψz)} =df (∃ϕ)[(x)(ϕ!x ≡ ψx) & f (ϕ!ẑ)]

(A similar contextual definition *20.08 was given for class abstracts for classes
of classes.) Roughly, to say something f about the class ẑ(ψz) is really to say
that there is a (predicative–see below) property ϕ coextensive with ψ of which
f holds. It follows from this definition that claims made about classes of indi-
viduals must be reinterpretable as claims about properties of individuals rather
than claims directly about individuals. A formula of the form x ϵ α would only
be meaningful if x and α were of different logical types, and therefore the entire
question of whether or not some thing, x, is a member of itself, x ϵ x, is syntacti-
cally illegitimate and unintelligible. Similarly, would-be discourse about a “class
of all classes not members of themselves” is not possible, and thus the paradox
cannot be formulated. Russell dubbed this method of eliminating commitment
to classes in favor of higher-order quantification, somewhat oxymoronically, as
the “no classes theory of classes.”
Russell’s higher-order logic made use of a hierarchy of variables of
different logical types. Unfortunately, the exposition of the syntax of
the logic of PM does not live up to contemporary standards of rigor,
and as a result, there is a fair amount of disagreement about the pre-
cise nature of its type theory.9 At the bottom of the hierarchy, there
are variables (x, y, z, . . .) for individuals. Russell speaks of the values of
higher-order variables as “propositional functions”, or sometimes, less for-
mally, as “properties” or “relations”. Specific propositional functions could
be represented by open sentences, or formulas with one or more free
 Russell’s Logicism 159

variables, “x is green”, “x is left of y”, or “x = x”. Part of the controversy, how-

ever, is whether Russell regarded propositional functions to be extra-linguistic
entities in some sense denoted by such expressions, or merely talked of “propo-
sitional functions” as a loose way of speaking about such open sentences them-
selves and their functioning in the logic in the “material mode”, to borrow a
phrase from Carnap (1937, §64). At any rate, if the variable ϕ is used to quan-
tify over (predicative) propositional functions of individuals, a formula such as
“(ϕ)ϕ!a” is taken to entail the truth of every one of an entire class of proposi-
tions formed by replacing “ϕ” with an open sentence and letting a be taken as
the value of its variable: “a is green”, “a is left of b”, and so on. There is then a new
class of open sentences containing such variables as ϕ occupying the next high-
est type, and then variables for quantifying over such propositional functions,
and so on. Moreover, variables for propositional functions with two arguments,
or two variables, would be considered as a different logical type from those with
one, and so on. Combined with the “no classes” theory, this meant that one
could introduce eliminable variables for classes of individuals, and variables for
classes of classes of individuals, and so on, forming a hierarchy of types of class
variables and class abstracts, but it would be impossible to have a single logical
type for all classes.
The foregoing is compatible with what is now known as “simple type the-
ory”. However, in addition to the sorts of divisions mentioned above, Russell
also regarded it as unintelligible for a higher-order quantifier to quantify over
properties involving the same kind of quantification as itself or anything of a
higher type or order. Consider our previous example of “(ϕ)ϕ!a”. Loosely, this
can be interpreted as saying that “a has every property”, where properties are
represented by open sentences such as “x is green”, “x is left of b”, and so on.
But what of the open sentence, “(ϕ)ϕ!x” or “x has every property”: is this in-
cluded among the properties a must have if “(ϕ)ϕ!a” were to be true? Arguably,
the suggestion that it is so included leads to a vicious circle in the truth con-
ditions of “(ϕ)ϕ!a”, and Russell took such vicious circles to be what generate
such “semantic” paradoxes as the Liar paradox and Richard’s Paradox (PM, 60–
65). Hence Russell distinguished the “order-type” of propositional functions
not involving quantification over their own or higher types—so called “pred-
icative” propositional functions—from those which do quantify over predica-
tive functions of the same type, and those which quantify over those, and so
on. This is known as Russell’s “ramified” hierarchy. There is some disagree-
ment over how and to what extent this hierarchy is fully enshrined in the logic of
160 The Bloomsbury Companion to Bertrand Russell

PM itself.10 However, Russell found it necessary to add an assumption, the “ax-

iom of reducibility” (*12.1), according to which every monadic propositional
function ϕ is equivalent to a predicative one f, marking predicative ones with
an exclamation point ! (as well as similar principles for higher types and two-
argument functions):
(∃f)(x)(ϕx ≡ f !x)

Notice that the quantified ϕ in Russell’s contextual definition of classes, above,

is restricted to predicative values. If the function ψ were not equivalent to a
predicative ϕ, then according to this definition, all statements f{ẑ(ψz)} about
the class ẑ(ψz) would come out as false. Thus, Russell took assuming the ax-
iom of reducibility to be a necessity for recapturing an adequate class theory
and sometimes also called it “the axiom of classes” (Papers 5 540, 607). If one
interprets the “ϕ” in the axiom as a schematic letter standing in for any complex
open sentence, the axiom takes a form very similar to the standard “principle of
comprehension” in simple type theory, a point to which we shall return later.

4 Logicism in Principia Mathematica

The no classes theory of classes, and PM’s similar treatment of “relations in ex-
tension” (*21), greatly simplified its syntactic primitives and reduced its basic
assumptions. In fact, the only signs taken as primitive in PM are two statement
operators, “∼” (negation), “∨” (disjunction), the universal and existential quan-
tifiers, and variables of the various logical types. Other truth-functional state-
ment operators are defined from negation and disjunction in the usual ways. For
instance, material implication p ⊃ q is defined as ∼p ∨ q. The axioms of the sys-
tem, apart from the various forms of the axiom of reducibility mentioned above,
include only those for standard classical propositional and (higher-order) quan-
tifier logic.11 From this basis alone, Whitehead and Russell sought to define all
of the notions of pure mathematics and derive the laws governing them. (There
are two other principles sometimes misleadingly called “axioms” in the discus-
sions of Russell’s logicism, viz., the “axiom of infinity” and the “multiplicative
axiom”; more on these in what follows.)
Russell’s treatment of cardinal numbers is very similar to Frege’s, with
the substitution of the no classes theory of classes for Frege’s naïve theory of
 Russell’s Logicism 161

extensions the main point of departure. Cardinal numbers are treated as classes
of classes alike in size, i.e., which can be put in 1–1 correspondence with each
other. The number 0 is then the class of all empty classes, of which there is of
course only one (the null class). The number 1 is the class of all single-membered
classes; the number 2 the class of all couples, and so on. One caveat, introduced
by Russell’s type distinctions and its consequences for the no classes theory, is
that numbers are duplicated across various types. For example, the null class
of individuals must be treated differently in the no classes theory from the null
class of classes of individuals, and consequently, the “zero” which contains only
the former is in a different type from the “zero” containing only the latter. In the
summary of the formal treatment below, variables such as x, y and z are used for
individuals, Greek letters such as α, β, etc., as special variables for classes (elim-
inable in favor of propositional function variables on the no classes theory),
variables R, S, etc., for relations in extension (also eliminable); however defini-
tions for higher-type analogues of these concepts would be quite similar.12

identity x = y =df (ϕ)(ϕ!x ⊃ ϕ!y)

non-membership x ∼ϵ α =df ∼(x ϵ α)
universal class V =df x̂(x = x)
complement –α =df x̂(x ∼ϵ α)
null class Λ =df –V
union α ∪ β =df x̂(x ϵ α ∨ x ϵ β)
intersection α ∩ β =df x̂(x ϵ α & x ϵ β)
subset α ⊂ β =df (x)(x ϵ α ⊃ x ϵ β)
singleton y ι‘y =df x̂(x = y)
domain D‘R =df x̂((∃y) xRy)
D
converse domain ‘R =df x̂((∃y) yRx)
class of many-one relations Nc → 1 =df R̂((x)(y)(z)(xRy & xRz ⊃ y = z))
class of one-many relations 1 → Nc =df R̂((x)(y)(z)(yRx & zRx ⊃ y = z))
class of one-one relations 1 → 1 =df Nc → 1 ∩ 1 → Nc
cardinal similarity α sm β =df (∃R)(R ϵ 1 → 1 & α = D‘R &
D
β = ‘R)
cardinal number of α Nc‘α =df β̂(β sm α)
class of cardinal numbers NC =df β̂((∃α)β = Nc‘α)
zero 0 =df Nc‘Λ
successor of α Succ‘α =df β̂((∃x)(x ϵ β & β ∩ –ι‘x ϵ α))
class of natural numbers N =df α̂((β)((0 ϵ β & (γ)(γ ϵ β ⊃
Succ‘γ ϵ β)) ⊃ α ϵ β))
162 The Bloomsbury Companion to Bertrand Russell

Notice here that the successor of a number α is the class of classes which are such
that if a member is removed one gets a member of α: the class of all classes with
one more member than the members of α. The class of natural numbers are
those contained in all classes containing 0 and the successor of anything they
contain, and hence would be the class containing exactly 0, 1, 2, 3, …. Using
these definitions, formal versions of the Peano-Dedekind axioms could be stated
as follows:
1. 0ϵN
2. (α)(α ϵ N ⊃ Succ‘α ϵ N)
3. ∼(∃α)(0 = Succ‘α)
4. (α)(β)((α ϵ N & β ϵ N & Succ‘α = Succ‘β) ⊃ α = β)
5. (α)((0 ϵ α & (γ)(γ ϵ α ⊃ Succ‘γ ϵ α)) ⊃ N ⊂ α)
Using nothing beyond the usual assumptions of higher-order quantifier logic
(which also provide a proxy for class-logic via the no classes theory), four of
these five can be derived as theorems relatively straightforwardly. (An interested
reader may consult PM, parts I–II, for details.)
The problematic one of the bunch is the fourth, which states that no two
natural numbers have the same successor. In order for this to hold, no natural
number can be empty: i.e., for every natural number there must be at least one
class with that many things. Notice further that in order for this to be true for
every natural number, the number of things in total must be infinite. Suppose
instead that there were only finitely many individuals; for convenience, let us
assume there are only three: a, b and c. Then the class consisting of a, b and c
would be the only three membered class, and the only member of the number 3.
The successor of 3 (viz., 4) would be the class of all those classes which are such
that, if you removed a member from them, you’d get a member of 3. But since
a, b and c are the only individuals there are, no class could meet this condition.
Hence, 4 would be empty. The successor of 4 (viz., 5) would also be empty, since
it is clearly impossible to get a member of the null class by removing an element
from a class, as the null class has no members. Hence, 3 and 4 would both
be natural numbers, and both would have the same successor (the null class),
but they would not be identical to each other, and so the fourth Peano “axiom”
would be false.
During earlier periods of his career, when he took propositions and classes to
be more than just “logical fictions”, Russell had regarded it as possible to deduce
the existence of infinitely many individuals purely logically. By the time of PM,
however, Russell rejected such deductions. Such proofs only appear to succeed
 Russell’s Logicism 163

because of a failure to take type distinctions into account. Consider what is now
called the Zermelo sequence: Λ, ι‘Λ, ι‘Λ, ι‘ι‘Λ, . . . (the null class, its singleton,
its singleton’s singleton, and so on). It might appear that this sequence is unend-
ing, and thus infinite, and that there must therefore be infinitely many things.
However, notice that each member of this sequence is of a higher type than the
previous member, and so it cannot be used to establish an infinity of individuals,
or an infinity of things in any one type at all. A similar difficulty makes it impos-
sible for Russell to make use of Frege’s argument (Frege 1950, §82) alleging that
all the numbers preceding a given natural number show that this number must
apply to at least one collection (i.e., the class containing only 0 has one member,
the class of 0 and 1 has two members, the class of 0, 1 and 2 has three, and so
on). Russell also rejects an argument (adapted from Dedekind) that alleges that
the series: Socrates, the idea of Socrates, the idea of the idea of Socrates, the idea
of the idea of the idea of Socrates, …, is infinite, on the grounds that the relevant
sort of ideas can only be assumed to exist when actually thought by some mind,
and hence we have no reason to think that this list goes on beyond some finite
number of steps (IMP, 139).
Russell therefore concludes that one cannot derive something like the fourth
Peano axiom stated above without assuming that every natural number is
nonempty, or what he sometimes calls the axiom of infinity:

(α)(α ϵ N ⊃ (∃β)(β ϵ α)) (Inf)

Despite his use of the word “axiom” here, in PM itself this is not taken as an
axiom in the usual sense, and is simply left as an undischarged antecedent on
many results, including the fourth Peano axiom and all results depending on
it. The question as to whether or not this compromises the success of Russell’s
logicism is revisited in Section 6 below.
For reasons mentioned in Section 2 above, natural number theory is per-
haps the most important test case for Russell’s logicism, but of course, PM cov-
ers far more than simply the theory of natural numbers. In addition to the
treatment of cardinal numbers just sketched, PM (parts IV and V; mostly in
volume 2) also provides a treatment of ordinal numbers, which are taken as a
subclass of a broader notion of “relation numbers”, defined as classes of relations
which are structurally isomorphic to each other. Finite ordinal numbers provide
yet another interpretation of “natural numbers” for the purposes of the Peano-
Dedekind axioms, though a similar problem with infinity arises there. For both
cardinals and ordinals, they discuss infinite numbers (if there are any collections
164 The Bloomsbury Companion to Bertrand Russell

or relations exemplifying such). To capture multiplication for infinite numbers

and achieve the expected results (i.e., those suggested by Cantor’s work), White-
head and Russell discuss an assumption they call the multiplicative axiom, which
asserts that for every class of disjoint (non-overlapping) classes, there is a class
which has among its members exactly one member from each member of the
class of disjoint classes (PM, *88). This assumption is equivalent with Zermelo’s
axiom of choice, which is notoriously independent of the other basic assump-
tions of set theory. Like the so-called “axiom” of infinity, the multiplicative ax-
iom is not formally taken as an axiom or basic assumption in PM, but is simply
abbreviated as “Mult ax” and left as an antecedent on the various results which
require it, such as the principle that all classes can be well-ordered.
In Parts V and VI (mostly in volume III) of PM, Whitehead and Russell pro-
vide treatments of integers, ratios, complex numbers and finally real numbers in
terms of what is in common between certain systems of relations, of which re-
lationships between segments of ratios provide only a single instance.13 Along
the way, they give formal logical definitions of such notions as limit and con-
tinuity. Part VI goes further yet and generalizes on these notions to provide
a generic theory of quantity and measurement. The planned fourth volume of
PM was slated to provide logicist accounts of many aspects of pure geometry
as well, though one can glean certain aspects of how this might have proceeded
from Russell’s chapters on projective, metric and descriptive geometry making
up the later parts of his earlier POM.

5 Gödel’s results and the scope of logicism

We now turn to evaluation of Russell’s logicism. Perhaps the most common

problem one hears cited against it, or logicism in general, especially in casual
conversation, appeals to Gödel’s famous incompleteness results. However, I
think the relevance of these results for assessing logicism, at least on the most
reasonable understandings of the logicist thesis, is greatly exaggerated.
Using PM as an example,14 Gödel showed that any deductive system
for mathematics having certain features contains undecidable sentences, i.e.,
sentences such that neither they nor their negations can be derived as the-
orems. Especially when combined with well-known corollaries shown by
Rosser, Tarski, and others,15 the features can for all intents and purposes
 Russell’s Logicism 165

be reduced to consistency, having the requisite strength to capture the basic ele-
ments of number theory, and having a recursive axiomatization. Since recursive
functions have been shown equivalent with Turing-computable functions and
λ-definable functions, if we accept the Church-Turing thesis that these and only
these functions are effectively computable, having a recursive axiomatization
amounts more or less to there existing an effective decision procedure for de-
termining what does or does not count as a deduction in this system. The fea-
sibility or desirability of a non-recursively axiomatized system is basically nil:
what good could a deductive system be if there was no effective way to tell what
counts as a deduction in it and what doesn’t? Gödel’s “first incompleteness the-
orem” roughly establishes that systems with the desirable features must contain
an undecidable sentence abbreviated G, which, roughly asserts of itself that it is
not a theorem of the system (or really, asserts something analogous of a number
which corresponds to it in an arithmetization of the system’s syntax). Clearly, it
would be undesirable for G to be a theorem, as then one would have a false the-
orem in the system. But if it is not a theorem, it is true, and hence we have what
appears to be a number-theoretic truth not derivable from the axioms. Gödel’s
“second incompleteness theorem” establishes that in such systems one can form
a sentence which roughly asserts that the system is consistent, i.e., that no con-
tradiction is derivable within, but that this sentence is also undecidable, at least
if the system is consistent.
How relevant such results are for evaluating Russell’s (or anyone else’s) form
of logicism depends largely on how strong we take the logicist thesis to be. Sup-
pose we define a “purely logical deductive system” as one whose axioms are all
logical truths and whose inference rules are sound on logical grounds alone,
and so cannot lead from a logical truth or truths to something that isn’t also a
logical truth. A very strong interpretation of the logicist thesis (LT) would be
the following:

There is a practical purely logical deductive system, S, such that for (LT-a)
every mathematical truth, p, p is a theorem of S.

It is likely that Russell believed that (LT-a) was true early on, and that PM was
such a deductive system. However, even if we interpret the “practical” in (LT-a)
to restrict us to recursively axiomatized systems, additional assumptions would
be required to argue that the Gödel results undermine (LT-a). Russell was aware,
even in 1910, that there were undecidable sentences in PM. The (so-called) axiom
166 The Bloomsbury Companion to Bertrand Russell

of infinity, for example is neither derivable nor refutable therein. Russell would
not have taken this to undermine his main thesis, or even (LT-a), since he would
regard neither the axiom of infinity nor its negation as a “mathematical truth”;
if it is true, it is contingently true (IMP: 141). Unless it could be shown that
some of the undecidable sentences proven to exist by Gödel, or their negations,
are mathematically true, there is no problem. Russell even suggested a response
along these lines in 1963 when asked about the importance of Gödel’s results
in a letter.16 Unfortunately, as such, the response seems a bit hasty. Gödel’s G
and the consistency of PM itself, unlike the axiom of infinity (arguably), we do
seem to know to be true, and on a priori grounds. If they are not mathematical
truths, it is very hard to see what other kind of a priori truths they could be.
And if they are mathematical truths, then this would seem to undermine PM’s
claim to justify (LT-a), as they cannot be proven in PM.
However, it is natural to wonder whether or not the heart of Russell’s logi-
cism requires anything quite as strong as (LT-a). Russell saw himself as arguing
against the Kantian thesis that mathematical truths were importantly different
from logical truths, with the former understood as synthetic a priori and the lat-
ter as analytic.17 From this perspective, it seems that Russell’s principal thesis
was that mathematical truths don’t have a special nature or essence distinct from
that possessed by logical truths. Whatever it is that makes logical truths special
or sets them apart from other kinds of truths is also possessed by mathemati-
cal truths: mathematical truths are a species of logical truths. It seems possible
to hold that thesis without insisting on anything as strong as (LT-a). Unfor-
tunately, the issue cannot be formally spelled out without saying exactly what
it is that makes logical truth special, an issue Russell himself struggled with.18
Early on, he seems to have thought that any truth which could be stated using
only logical constants counts as a “logical truth” (PM, §10); later on, however,
he suggests that they must have a special “tautological” form, but neglects to
provide an exact specification of what that amounts to (IMP, 203).
In contemporary research in metalogic, it is often the case that two distinct
criteria are given for what makes formulas logically necessary: a deductive or
proof-theoretic one (derivability from the axioms), and a semantic one (usually
amounting to something like truth in all acceptable models or interpretations).
If a deductive system is both sound and complete, the two characterizations are
equivalent: something is a logical truth in one sense if and only if it is in the
other. However, some deductive systems are incomplete. For example, it is a
well-known corollary of Gödel’s results that second- and higher-order logical
 Russell’s Logicism 167

calculi (including PM), when interpreted according to “standard semantics”,19

cannot be complete when axiomatized recursively. There are formulas that are
logically true according to the semantic criterion that cannot be derived as the-
orems. Obviously, this result should not be interpreted as showing that some
logical truths are not logical truths, and similarly, it is unclear why the fact that
certain apparently “mathematical” truths cannot be derived in a certain deduc-
tive calculus should be taken to establish that those truths are not logical truths.
In other words, if the logicist thesis were interpreted along the lines of (LT-b)
below,20 then it is not clear how or why Gödel’s results should be taken to pose
a problem for it:

Every mathematical truth has the semantic feature which sets apart (LT-b)
logical truths semantically from others.

Of course, PM does not explicitly provide a semantic characterization of logical

truth nor a direct argument for anything like (LT-b). Nonetheless, the deduc-
tions of various mathematical truths provided in PM could be relevant for se-
mantic characterizations of logicism, provided it could be established that the
axioms of PM are logical truths in this sense and that the inference rules pre-
serve this feature.
Moreover, even for those primarily interested in a deduction-centered ac-
count of logical truth, it is not obvious that something as strong as (LT-a) needs
to be maintained by a logicist. Consider the following similar, but slightly
weaker, thesis:

For every mathematical truth, p, there is a practical purely logical (LT-c)

deductive system, S, such that p is a theorem of S.

This is much like (LT-a) except that it does not require there to be a single logi-
cal system in which every mathematical truth is captured: it requires only that
each such truth be captured in some such system or other. To my knowledge,
there is nothing in Gödel’s results to suggest that (LT-c) must be false. Con-
sider again the situation that we seem to know that Gödel’s G is true, despite
its undecidability in PM. So long as our knowledge of this relies only on logical
principles, perhaps there is some purely logical extension of PM in which G is
derivable. (And if our knowledge of G requires something beyond logic, for this
to pose a problem for Russell, it would need to be shown that it doesn’t require
anything beyond mathematics, or his earlier response is sufficient.) If this new
system is recursive, it too will have undecidable (true) sentences, but perhaps
they are derivable in another logical system of the appropriate type. This is more
168 The Bloomsbury Companion to Bertrand Russell

or less in keeping with Russell’s “official” response to Gödel, which suggests that
there is a hierarchy of logical languages or logical systems, and what cannot be
captured in one is captured higher in the hierarchy (Papers, 11 159; MPD, 114).
Indeed, when the issue is posed in terms of the special nature of mathematics
and its relation to logic as a whole—the issue of what makes mathematical truths
different from others, and whether or not this is the same as the special nature
of logical truths—what PM does succeed in establishing seems far more im-
portant than any worries stemming from the few rather recondite undecidable
sentences shown to exist by Gödel. Consider every mathematical truth taught
in primary or even secondary school. Consider all the mathematics of quan-
tity and measurement used in engineering. Consider Kant’s famous example
of the allegedly synthetic 5 + 7 = 12. It appears that all of these can be cap-
tured in any higher-order system interpreting the number theory provided by
Peano arithmetic. If PM or similar work succeeds in showing that this portion
of mathematics consists in nothing other than logical truths, it becomes almost
comical to suggest that the logicist thesis as a whole is mistaken. If nearly all of
the mathematics anyone uses or knows about, the truths about numbers which
most readily come to mind when considering arithmetic, reduce to logic, can it
seriously be maintained that the essence of mathematics is extra-logical, merely
because there are certain highly complex, and application-free truths in certain
formal systems which cannot be shown to be logical in precisely the same way
within a (single) recursive axiom system? If mathematics has a special essence,
a special nature, surely that nature expresses itself already in Peano arithmetic,
so here again it seems right to maintain that capturing Peano arithmetic is the
best test case for Russell’s logicism as a whole.
Of course, doubts may still be raised as to whether or not PM adequately cap-
tures Peano arithmetic, but they are distinct from any stemming from Gödel’s
results. Such doubts are considered in the next section.

6 The controversial “axioms” and if-then-ism

Perhaps the next most commonly heard objection to Russell’s logicism—and

the most common from specialists in the area—points to its employment of
certain basic assumptions which do not seem, or at least do not obviously
seem, like logical truths.21 There are three controversial (so-called) axioms
here: reducibility, infinity and the multiplicative axiom. Of course, it must be
 Russell’s Logicism 169

remembered that the latter two were not treated as axioms in PM, but instead
simply left as undischarged antecedents on results depending on them. This
point, however, simply invites stating the criticism in a different form. The
charge, called if-then-ism by the critics that offer it,22 is that Whitehead and
Russell do not prove the basic principles of mathematics from logic outright,
but merely derive conditionals with certain mathematical assumptions as an-
tecedents and their logical consequences as consequents. For example, rather
than proving that no two natural numbers have the same successor (the fourth
Peano axiom), they only prove that this is true if there are infinitely many indi-
viduals. In the eyes of these critics, this greatly compromises the success of PM’s
logicism. After all, for any theory which can be organized as a deductive system,
even an obviously empirical and non-logical one such as a theory in physics, the
conditionals from the axioms to their logical consequences can take the form of
logical truths. This tells us nothing about the essence of its subject matter.
For what it is worth, Russell himself admitted that the truth of the axiom
of reducibility was less than fully obvious. Indeed, even when first stating it he
admitted that the grounds for accepting it were “inductive”, in keeping with his
general “regressive” method. It could be used as a premise from which many
desirable results could be proven and he knew of no more plausible or no more
general principle from which as many important results followed (PM, 59). Rus-
sell also dedicated much of the new material added to the second edition of PM
to exploring a less than fully successful23 attempt to do away with the axiom by
adopting instead a doctrine of extensionality he took to be motivated by the phi-
losophy of Wittgenstein’s Tractatus (see PM 2, xiv and appendices). However,
it should be remembered that Russell’s main motivation for adopting the axiom
was to recover the ability to speak of classes, which for him amounted simply
to things considered in groups, or plurally,24 without a stronger assumption
of “real” classes taken as genuine things. It is now widely acknowledged that
something like quantification over collections or pluralities, or in general, what
one finds in a “standard” higher-order logic, is necessary for capturing some
logical forms used in every-day reasoning. Perhaps the most famous example
is the Kaplan-Geach sentence, “some critics only admire one another”, which
cannot be interpreted in first-order logic. The history of logic does have its first-
order purists, like Quine, who considered higher-order logic to be “set-theory
in sheep’s clothing” (Quine 1986, 66), and there are even hold-outs among
contemporary philosophers (e.g., Burgess 2005, 201–14). Still, I think it safe
170 The Bloomsbury Companion to Bertrand Russell

to say that these represent the minority. Something like the ability to consider
objects in groups seems to be a necessary central element in reasoning about
nearly every topic, and allowing for this is not to take on a special extra-logical or
specifically mathematical assumption. As noted above, the axiom of reducibility
is roughly equivalent to the usual comprehension principles allowed in simple
type theory or most plural logics (cf. Yi 2013).25 The ideal formulation of a
logical apparatus that allows us to speak of classes, or collections, or plurali-
ties, remains an active area of research, discussion and controversy, but when
compared to the alternatives, Russell’s adoption of something like the axiom of
reducibility seems all-things-considered to be among the most restrained.
The failure to prove the multiplicative axiom or obtain its results outright in
PM also does not seem to pose a serious problem for the success of Russell’s logi-
cism. While the very similar axiom of choice is perhaps widely accepted among
mathematicians, and it is also known to be equivalent to a number of other prin-
ciples with a similar status (the well-ordering principle, Zorn’s lemma, the mul-
tiplicative axiom itself, etc.), no one has been able to provide a proof of it from
more obvious or more fundamental principles, and indeed, it is well-known to
be independent of the other axioms of ZF set theory. The multiplicative ax-
iom is not needed for results in finite arithmetic, but only for certain results in
less-firmly settled infinite arithmetic. The axiom of choice and the multiplica-
tive axiom hardly count as uncontroversial examples of definite mathematical
truths. The only results which are considered definite are their mutual equiv-
alence and their relationships with these other principles, which are captured
in PM. It does not seem reasonable to argue that the logicism of PM is a fail-
ure merely because the multiplicative axiom, or its logical consequences, are not
derived in non-conditional form.
Similar concerns surrounding the axiom of infinity seem to pose a far greater
challenge to a positive evaluation of Russell’s logicism. Without it, as we have
seen, Russell cannot recapture Peano arithmetic in its usual form, as the axiom
of infinity would be needed to obtain the fourth axiom. A defender of Russel-
lian logicism seems to be forced into one of the following positions: (a) argue
that the axiom of infinity can reasonably be taken as a basic logical truth, (b)
provide a demonstration of the axiom of infinity from other plausible purely
logical assumptions (not already included in PM), (c) deny that the fourth
Peano axiom is, without qualification or antecedent, a mathematical truth, or
(d) admit that certain very important mathematical truths cannot be derived
from a logical starting place. Response (d) seems tantamount to abandoning
logicism. Responses (a) and (b) are perhaps worthy of further investigation,
 Russell’s Logicism 171

but (c) seems to have been Russell’s own considered view. Can anything be said
in its favor?
Proponents of the if-then-ist criticism of Russell sometimes point to early
passages such as in POM (§1), where Russell claims that all mathematical propo-
sitions can be seen as taking the form of conditionals:
Pure mathematics is the class of all propositions of the form “p implies
q,” where p and q are propositions containing one or more variables, the
same in the two propositions, and neither p nor q contains any constants
except logical constants.

However, there are a number of reasons for being cautious about reading this in
favor of if-then-ism. For one, in the logic of POM, all other connectives (nega-
tion, disjunction, conjunction, etc.) are defined in terms of material or formal
implication (§§18–19), and so any logically complex proposition would be of the
requisite form, and obviously a logical deduction of a statement relies on its hav-
ing the kind of complexity that might make its derivation from basic principles
possible. Russell’s example in this context (§6) in which he rewords “1 + 1 = 2”
as “If x is one and y is one, and x differs from y, then x and y are two” helps ex-
plain how uncovering the true logical complexity of a proposition is necessary
in order to see its real logical status. As stressed by Griffin (1982, 89), Russell
(even later on) thought many mathematical truths took the form conditionals
not because he thought of them as representing the relationships between the
axioms of a theory and the consequences, but rather because he took them as
general truths with definite application conditions where those conditions were
stated in the antecedents applied to variables. Applying mathematics involves
finding values of the variables satisfying the antecedents, and therefore accept-
ing the relevant instances of the consequents for those values. There is precedent
in Russell’s earlier work for recognizing situations where required antecedents
have been left off of stated theorems by practicing mathematicians because they
took for granted the most likely practical application conditions for their work.
Past geometers who assumed that actual (physical) space was Euclidian took
for granted that they could presume features of Euclidian space when stating
their results. As Russell argues, however, the real “pure” mathematical truth
should be taken as the conditional from their assumptions about the structure
of space to the consequences thereof (POM §§353, 412), rather than the unqual-
ified statements which only are true in certain application conditions. Hence,
it appears Russell would argue that the true form of a given mathematical truth
may reasonably be taken to be conditional in form if the antecedent involves
172 The Bloomsbury Companion to Bertrand Russell

special and restricted conditions for its application, and the consequent itself,
even if true for a given instance, is not true for that instance on mathematical
grounds alone, but only because the antecedent is satisfied. In the geometri-
cal context, if the antecedent is not satisfied, we have a different kind of space,
where a different kind of geometry may be applicable; this does not make it in
any way a logically or mathematically defective space.
Is the case of the relationship between the axiom of infinity and the fourth
Peano axiom at all similar? The suggestion on the table then is that the fourth
Peano axiom (and its consequences) should not be considered mathematical
truths, full stop, just like geometrical statements only true if space is Euclid-
ian. Hence, it is not required for logicism that we derive them, full stop, from a
logical basis. It is enough if we can demonstrate that they hold given the right
conditions. Russell himself does not explicitly argue this way to my knowledge,
and at first blush it seems a fairly radical suggestion. Should we really conclude
that 4 is distinct from 5 only given certain application conditions? Yet, I think
the proposal is not wholly without merit.26 Without presupposing infinity, our
usual assumptions about the behavior of numbers will only start going “awry”
for those greater than the number of individuals there in fact are. If we are ap-
plying the mathematics of PM to a domain of individuals with 3 members, then
all the usual results for numbers up to including 3 would be what we expect.
At higher types, where there are more and more classes formed of our three
things or formed of classes of them, or classes of classes of them, etc., the “well-
behaving” numbers will reach higher and higher (IMP, 134). In short, nothing
will go wrong in our actual conditions of application. Perhaps “non-infinite”
arithmetic, where the series of numbers runs out after a certain stage, or goes
in circles, much like non-Euclidian geometry, should be seen as a new area of
study. Indeed, one could make the case that modular (“clock”) arithmetic al-
ready exemplifies this. Some philosophers take a hard conservative line and
demur from any philosophical proposals which could be seen as suggesting “re-
forms” to mathematics (e.g., Lewis 1986, 109). Russell himself, as we have seen,
thought it an advantage of the process of analysis that it did not necessarily pre-
serve the analyzed body of knowledge as originally conceived.
Whatever may be said for the above proposal, Russell himself seems to have
been disappointed by his failure to establish in any definitive or a priori way
the existence of the complete series of numbers. In fact, while the logic of PM
presupposes there is at least one individual—something Russell eventually came
to regard as “a defect of logical purity” (IMP, 203n)—there is no way in it to
 Russell’s Logicism 173

show that there is any more than one, and therefore, no logical way to arrive
at any numbers beyond one, at least at the lowest type. Ivor Grattan-Guinness
provides us with a report, perhaps apocryphal, from an acquaintance of Russell
who allegedly heard him bemoan this fact after having finished PM:

Bertrand Russell called me aside as a mathematician I suppose and likely

to appreciate the gravity of his statement—‘I have just realised that I have
failed—it is easy to establish the unit one but I have omitted to establish
a second like unit’—(I won’t guarantee the precise wording but it’s not far
off). He went on to say ‘I have finished’. (Quoted in Grattan-Guinness
2000, 401)

Did, however, Russell fail? When we consider the effects of Russell’s pursuit of
logicism on the history of both logic and philosophy, there can be no question
that it has had a lasting and positive impact. It shaped not only Russell’s phi-
losophy but has made modern symbolic logic an irreplaceable part of almost
every analytic philosopher’s toolkit. Even when it comes to the question of the
logicist thesis for mathematics in particular, there is a lot more reason to think
that Russell’s work was a success than is generally realized.

Notes
1 Early drafts of his work on this subject can be found in Papers 2 and Papers 3.
2 See Hylton (1990) and Griffin (1991) for discussion of Russell’s philosophical con-
version to realism.
3 See Proops (2006) for an account of the philosophical motivations behind Russell’s
initial acceptance of logicism.
4 It is worth comparing Russell’s description of philosophical progress in the study
of mathematics, with his overall description of “analysis” as a method employed in
other areas of philosophy; see, e.g., Hager (2003) for discussion.
5 On this point, see especially Gandon (2012, chaps. 5–6) and note 13 below.
6 For further discussion on this point, see Irvine (1989).
7 For further historical details of the context of Russell’s discovery, see Griffin (2004).
8 For discussion, see Landini (1998, chap. 8).
9 For a variety of positions, see Church (1976), Hatcher (1981, chap. 4), Landini (1998,
chap. 10), Linsky (1999, chap. 4) and Klement (2013).
10 In particular, Landini (1998, chap. 10, 2011b, chap. 3) has argued that PM should
be interpreted to make use of no object-language variables except predicative
variables, making the syntactic rules much simpler than what is suggested
by the traditional interpretation found in e.g., Church (1976), where variables
174 The Bloomsbury Companion to Bertrand Russell

of infinitely many orders within the same type are possible. Landini reads the non-
predicative variable ϕ occurring in the axiom of reducibility as a metalinguistic
schematic letter.
11 Strictly speaking, PM offers us two different treatments of the logic of quantifica-
tion, one in *9 and one in *10, with the more austere *9 disallowing quantifiers
subordinate to truth-functional operators except as abbreviation. For the sake of
simplicity, I pass over such complications here.
12 I here try to make use of Whitehead and Russell’s own notation as much as possible,
although note that they used a dot, or concatenation, rather than “&” for conjunc-
tion, and also used dots in place of parentheses for grouping; see PM 9–10. I also
here omit certain intermediately defined signs employed in PM itself, and simply
write the unabbreviated forms where these are used later.
13 In IMP (chap. 7), Russell defines real numbers simply as segments of ratios or ra-
tional numbers according to the “Dedekind cut” method, and does little to explain
the more general application of PM’s theory of quantity; this oversimplification of
the more general treatment provided in PM has been bemoaned by Gandon (2012,
9) among others.
14 It is likely that Gödel’s exposition of PM was not entirely faithful to the original; in
particular, Gödel did not fully take into account that numerals, as signs for classes,
in PM were incomplete symbols and not genuine terms. This issue, and the com-
plications arising from it, cannot be explored in depth without providing a full re-
construction of PM, which cannot be attempted here.
15 Gödel’s original results were published in Gödel (1931), and a summary of them and
the most important corollaries can be found in most textbooks on mathematical
logic, e.g., Mendelson (2010).
16 See Russell’s letter to G. Simons, 1 October 1963; document RA2.710.111928 at the
Bertrand Russell Archives, McMaster University, and compare Papers 11: 65.
17 Interestingly, Russell’s exact attitude of where Kant had erred seems to have changed
over the years; in POM (§434), Russell suggests that Kant was wrong not about
mathematics but about logic, alleging that even logical truths can be understood
as synthetic a priori in Kant’s sense. Later on, Russell claims instead that logicism
shows that mathematics is analytic and not synthetic (e.g., HWP, 740). It is likely,
however, that the different attitudes are as much the result of employing different
definitions of “analytic” and “synthetic” as they are a reflection of a change of mind-
set; see Landini (2011b, 223–25) and Korhonen (2013, chap. 1).
18 For further discussion, see Klement (2015).
19 Standard semantics is often contrasted with “general” or “Henkin” semantics; the
former, unlike the latter, requires that every subset of the domain of the first-
order quantifiers be the extension of some value of the domain of quantification
for second-order variables, and similarly for other types.
 Russell’s Logicism 175

20 Rayo (2007, 240) provides a list of different interpretations of logicism, though cu-
riously he does not bother to make distinctions similar to my (LT-a) and (LT-c) for
forms focused on deductive consequence or derivability.
21 See, e.g., Kneale and Kneale (1962, 669), Roselló (2012, 50), Tennant (2013, §1.3.1)
and Soames (2014, 488).
22 See, e.g., Putnam (1967), Musgrave (1977), Coffa (1981) and Boolos (1998,
chap. 16). For a variety of responses, see Griffin (1982), Landini (2011b), Gandon
(2012), Klement (2012), Galaugher (2013), and Kraal (2014).
23 The difficulties with the attempt were partly appreciated by Russell himself, partly
not. For further discussion, see Landini (1996), Hazen and Davoren (2000), Landini
(2007), and Linsky (2011).
24 For more on Russell’s attitude about the relationship between discourse about
classes and about plurals, see Oliver and Smiley (2005), Bostock (2008) and Kle-
ment (2014).
25 Indeed, if Landini is right that the object-language variables of PM are only pred-
icative variables (see note 10 above), then the system of PM is formally speaking the
same as a simple type theory, and the axiom of reducibility just is its comprehension
principle; see Landini (1998, 264–67).
26 See Landini (2011a) for a sympathetic treatment of such an approach.

References
Boolos, George, 1998. Logic, Logic and Logic. Cambridge, Mass.: Harvard University
Press.
Bostock, David, 2008. “Russell on ‘the’ in the Plural.” In Russell vs. Meinong: The Legacy
of “On Denoting”, edited by Nicholas Griffin and Dale Jacquette, pp. 113–143. New
York: Routledge.
Burgess, John, 2005. Fixing Frege. Princeton: Princeton University Press.
Carnap, Rudolf, 1937. The Logical Syntax of Language, trans. by A. Smeaton. London:
Routledge & Kegal Paul.
Church, Alonzo, 1976. “Comparison of Russell’s Resolution of the Semantical Anti-
nomies with that of Tarski.” Journal of Symbolic Logic 41: 747–760.
Coffa, J. Alberto, 1981. “Russell and Kant.” Synthese 46: 247–263.
Dedekind, Richard, 1888. Was sind und was sollen die Zahlen? Braunschweig: Vieweg.
Frege, Gottlob, 1950. The Foundations of Arithmetic, trans. by J. L. Austin. Evanston:
Northwestern University Press.
, 1980. Philosophical and Mathematical Correspondence, trans. by H. Kaal, edited
by B. McGuinness. Chicago: University of Chicago Press.
, 2013. Basic Laws of Arithmetic, trans. by P. Ebert and M. Rossberg. Oxford:
Oxford University Press.
176 The Bloomsbury Companion to Bertrand Russell

Galaugher, Jolen, 2013. Russell’s Philosophy of Logical Analysis 1897–1905. Basingstoke:

Palgrave Macmillan.
Gandon, Sébastien, 2012. Russell’s Unknown Logicism. Basingstoke: Palgrave Macmil-
lan.
Gödel, Kurt, 1931. “Über formal unentscheidbare Sätze der Principia Mathematica und
verwandter Systeme I.” Monatshefte für Mathematik und Physick 38: 173–98.
Grattan-Guinness, Ivor, 2000. The Search for Mathematical Roots 1870–1940. Princeton:
Princeton University Press.
Griffin, Nicholas, 1982. “New Work on Russell’s Early Philosophy.” Russell 2: 69–83.
, 1991. Russell’s Idealist Apprenticeship. Oxford: Clarendon Press.
, 2004. “The Prehistory of Russell’s Paradox.” In One Hundred Years of Russell’s
Paradox, edited by Godehard Link, pp. 349–371. Berlin: Walter de Gruyter.
Griffin, Nicholas and Bernard Linsky, eds., 2013. The Palgrave Centenary Companion
to Principia Mathematica. Basingstoke: Palgrave Macmillan.
Hager, Paul, 2003. “Russell’s Method of Analysis.” In The Cambridge Companion to
Bertrand Russell, edited by Nicholas Griffin, pp. 310–31. Cambridge: Cambridge
University Press.
Hatcher, William, 1981. The Logical Foundations of Mathematics. Oxford: Pergammon
Press.
Hazen, Allen P. and J. M. Davoren, 2000. “Russell’s 1925 Logic.” Australasian Journal of
Philosophy 78: 534–556.
Hylton, Peter, 1990. Russell, Idealism and the Emergence of Analytic Philosophy. Oxford:
Oxford University Press.
Irvine, A. D., 1989. “Epistemic Logicism and Russell’s Regressive Method.” Philosophical
Studies 55: 303–27.
Klement, Kevin C., 2012. “Neo-logicism and Russell’s Logicism.” Russell n.s. 32: 127–
59.
, 2013. “PM’s Circumflex, Syntax and Philosophy of Types.” In Griffin and Linsky
(2013), pp. 218–47.
, 2014. “Early Russell on Types and Plurals.” Journal for the History of Analytical
Philosophy 2: 1–21.
, 2015. “The Constituents of the Propositions of Logic.” In Acquaintance, Knowl-
edge and Logic: New Essays on Bertrand Russell’s The Problems of Philosophy, edited
by Donovan Wishon and Bernard Linsky, pp. 189–229. Stanford: CSLI Publications.
Kneale, W. and M. Kneale, 1962. The Development of Logic. Oxford: Clarendon Press.
Korhonen, Anssi, 2013. Logic as Universal Science: Russell’s Early Logicism and Its Philo-
sophical Context. Basingstoke: Palgrave Macmillan.
Kraal, Anders, 2014. “The Aim of Russell’s Logicism.” Synthese 191: 1493–1510.
Landini, Gregory, 1996. “The Definability of the Set of Natural Numbers in the 1925
Principia Mathematica.” The Journal of Philosophical Logic 25: 597–615.
, 1998. Russell’s Hidden Substitutional Theory. Oxford: Oxford University Press.
 Russell’s Logicism 177

, 2007. Wittgenstein’s Apprenticeship with Russell. Cambridge: Cambridge Uni-

versity Press.
, 2011a. “Logicism and the Problem of Infinity: The Number of Numbers.”
Philosophia Mathematica 19: 167–212.
, 2011b. Russell. New York: Routledge.
Leibniz, G. W., 1666. Dissertatio de Arte Combinatoria. Leipzig: J. S. Fick and J. P.
Seubold.
Lewis, David, 1986. On the Plurality of Worlds. Oxford: Blakcwell.
Linsky, Bernard, 1999. Russell’s Metaphysical Logic. Stanford: CSLI Publications.
, 2011. The Evolution of Principia Mathematica: Bertrand Russell’s Manuscripts
and Notes for the Second Edition. Cambridge: Cambridge University Press.
Mendelson, Elliot, 2010. Introduction to Mathematical Logic, 5th ed. Boca Raton: CRC
Press.
Musgrave, Alan, 1977. “Logicism Revisited.” British Journal for the Philosophy of Science
28: 99–127.
Oliver, Alex and Timothy Smiley, 2005. “Plural Descriptions and Many-valued Func-
tions.” Mind 114: 1039–68.
Proops, Ian, 2006. “Russell’s Reasons for Logicism.” Journal of the History of Philosophy
44: 267–92.
Putnam, Hilary, 1967. “The Thesis that Mathematics Is Logic.” In Bertrand Russell:
Philosopher of the Century, edited by Ralph Schoenman, pp. 273–303. London: Allen
& Unwin.
Quine, W. V., 1986. Philosophy of Logic, 2nd ed. Cambridge, MA: Harvard University
Press.
Rayo, Agustín, 2007. “Logicism Reconsidered.” In The Oxford Handbook of Philosophy
of Mathematics and Logic, edited by Stewart Shapiro, pp. 203–35. Oxford: Oxford
University Press.
Roselló, Joan, 2012. From Foundations to Philosophy of Mathematics. Newcastle upon
Tyne: Cambridge Scholars Publishing.
Russell, Bertrand, 1897. An Essay on the Foundations of Geometry. Cambridge: Cam-
bridge University Press.
, 1903. The Principles of Mathematics. Cambridge: Cambridge University Press.
, 1919. Introduction to Mathematical Philosophy. London: Allen & Unwin.
, 1946. History of Western Philosophy. London: Allen & Unwin.
, 1958. My Philosophical Development. London: Allen & Unwin.
, 1990. The Collected Papers of Bertrand Russell, vol. 2, Philosophical Papers 1896–
99, edited by N. Griffin and A. C. Lewis. London: Unwin Hyman.
, 1993. The Collected Papers of Bertrand Russell, vol. 3, Toward the “Principles of
Mathematics” 1900–02, edited by G. H. Moore. London: Routledge.
178 The Bloomsbury Companion to Bertrand Russell

, 1994. The Collected Papers of Bertrand Russell, vol. 4, Foundations of Logic 1903–
05, edited by A. Urquhart. London: Routledge.
, 1997. The Collected Papers of Bertrand Russell, vol. 11, Last Philosophical Testa-
ment 1943–68, edited by J. G. Slater. London: Routledge.
, 1998. Autobiography. London: Routledge.
, 2014. The Collected Papers of Bertrand Russell, vol. 5, Towards “Principia Math-
ematica” 1905–08, edited by G. H. Moore. London: Routledge.
Soames, Scott, 2014. The Analytic Tradition in Philosophy, vol. 1, The Founding Giants.
Princeton: Princeton University Press.
Tennant, Neil, 2013. “Logicism and Neologicism.” The Stanford Encyclopedia of Philos-
ophy, http://plato.stanford.edu/entries/logicism/, accessed 19 March 2016.
Whitehead, A. N. and Bertrand Russell, 1925–1927. Principia Mathematica, 2nd ed., 3
vols. Cambridge: Cambridge University Press.
Yi, Byeong-uk, 2013. “The Logic of Classes of the No-Class Theory.” In Griffin and
Linsky (2013), pp. 96–129.