The development of programs for the

foundations of mathematics in the rst third of

the 20th century
Solomon Feferman
1 Legacy of the 19th century: the reductive pro-
grams.
The most prominent schools or programs for the foundations of mathe-
matics that took shape in the rst third of the 20th century emerged directly
from, or in response to, developments in mathematics and logic in the lat-
ter part of the 19th century. The rst of these programs, so-called logicism,
had as its aim the reduction of mathematics to purely logical principles. In
order to understand properly its achievements and resulting problems, it is
necessary to review the background from that previous period.
Following the remarkable but freewheeling oraison of mathematics in
the 17th and 18th centuries, there was increasing attention to questions of
rigorization, conceptual clarication, conceptual reduction, and systematic
organization and exposition. In particular, problematic uses of innitesimals
and of innite sums and products in analysis were eliminated in favor of rig-
orous use of the limit concept both for real and complex numbers. Moreover,
the use of the latter system, generated by the imaginary number i =

1,
was given a solid foundation by reduction to the real number system. This
was simply accomplished (in one way) by representing x+yi, for x, y real, by
the pair (x, y), with suitable denitions of addition and multiplication
of such pairs.
1
The process of reduction was carried a substantial step further (indepen-
dently), by Georg Cantor (18451918) and Richard Dedekind (18311916),
by means of the construction of the real numbers from the system of ra-
tional numbers. Cantor did this by representing real numbers as limits of
sequences of rational numbers, and Dedekind by representing them as least
upper bounds (or, dually, greatest lower bounds) of subsets of the rationals.
Finally, Leopold Kronecker (1823-1891) explained the rational numbers in
terms of the system of positive integers (1, 2, 3, . . .), or, alternatively, the
natural numbers (0, 1, 2, . . .).
2 The general process of reduction.
Each step in the reductive process consisted in replacing the objects of
a given number system S, by representations of them built from objects of
a more basic system S
0
, where these combinations are identied when they
are intended to represent the same object from S. For example, Kroneckers
representation of positive rational numbers takes (n, m), with n, m positive
integers, to represent n/m, and identies (n, m) with (p, q) just in case we
are to have n/m = p/q, i.e. just in case nq = mp. Cantors representation
of real numbers takes a sequence of rational numbers r = r
0
, . . . , r
n
, . . .) to
represent lim
n
r
n
when r satises the (internal) Cauchy convergence crite-
rion; then r = r
0
, . . . , r
n
, . . .) is identied with s = s
0
, . . . , s
n
. . .) when
lim
n
(r
n
s
n
) = 0, the latter limit being explained entirely in terms of ra-
tional numbers. In general, then, the system S is replaced by a system S

of combinations of objects built out of S

0
, and two members a, b of S

are
identied when they are to represent the same object of S. This relation
of identication is an equivalence relation between members of S

, i.e. we
have: (i) a a, (ii) a b implies b a, and (iii) a b & b c implies a c,
for all a, b, c in S

. This can be turned into the actual relation of identity

by associating with each a in S

its equivalence class [a] which is dened to

consist exactly of all those b with a b. Then (assuming that classes are
determined entirely by their members), we have
[a] = [b] if and only if a b,
2
and the operations of S are mirrored by operations on equivalence classes.
Technically speaking, S is in this way replaced up to isomorphism by the
system of equivalence classes of members of S

.
3 Toward the nal reduction: the contributions of
Dedekind, Peano and Cantor.
Inspection of the successive reductions from the complex numbers through
the reals down to the natural numbers, shows that certain general notions
such as those of pair, set (or class), sequence, relation, operation (or func-
tion), etc. form the principal ingredients in the construction of each system
from a more basic one. These notions are fundamental to all mathematics
and, depending on ones point of view, can be regarded as set-theoretical or
logical in nature. It appeared from work of Cantor, Dedekind, and Peano
(Giuseppe Peano, 18581932) toward the end of the 19th century that a -
nal reduction could be accomplished, by explaining the system of natural
numbers entirely in terms of such more basic notions. The contribution of
Dedekind and Peano toward this end was to give an axiomatic characteriza-
tion of the system of natural numbers (in 1888 and 1889, resp.). In Peanos
hands, this essentially takes the following form: the set N is given (i) with
a distinguished element 0 and (ii) a successor operation sc, where sc(x) is
interpreted as x+1 for any x (though prior to the denition of + and 1, it
is not written in that form). The axioms further state that (iii) 0 is not the
successor of any natural number, (iv) sc is a oneone operation from N into
N, and (v) that N is the smallest set which contains 0 and is closed under
successor. A little more formally, using the symbols for is a member of
and for implies, the Peano Axioms are:
I 0 N
II x N sc(x) N
III x N sc(x) ,= 0
IV x N & y N & sc(x) = sc(y) x = y
V If X is any subset of N with 0 X and (x X sc(x) X) for all
x, then X contains all members of N.
3
The last axiom, V, is the basis of proof by induction on N: to show that
a property P(x) holds for all x in N, it is sucient to show that P(0) holds
and that P(x) P(sc(x)) for all x; this is seen by taking X to be the set
of all x N with P(x). Dedekind applied induction to justify denition by
recursion on N, where a function F is determined on N by prescribing F(0)
and by telling how F(sc(x)) is to be dened in terms of F(x) for any x. He
used this in turn to show that the axioms for N are categorical, i.e. if N, 0, sc
and N

, 0

, sc

are any two realizations of I-V then they are isomorphic: one
obtains the required onetoone correspondence x x

by taking F(0) = 0

and F(sc(x)) = sc

(F(x)), and then x

= F(x) for any xN.

What the (Dedekind or) Peano axioms do not by themselves guarantee
is the existence of some realization (or model) of them. Dedekind argued
for this via the (potential) innity of objects of (his) thought, where the
successor of any object x is the thought that x can be an object of (his)
thought. However, this oddly psychological interpretation was accepted by
no one else. Actually, a more mathematical realization was already available
via Cantors theory of sets, which he developed in a remarkable series of
papers stretching from 1874 to 1897 (cf. the collection 1932). Cantors main
new contribution lay in the coherent extension to arbitrary (nite or innite)
sets of the notions of ordinal number and cardinal number. The rst of these
is used to answer questions of the form: How are the elements of a set ranked
in a given (well)-ordering of it? The latter is used to answer the question:
How many elements does a set X have? which is, in turn, closely related to
the question: When do two sets X and Y have the same number of elements?
The answer given to that by Cantor is in terms of the relation of set-theoretic
equivalence , where X Y is dened to hold just in case there is a one-to-
one correspondence between the elements of X and the elements of Y . He
then held that with each set X is associated its cardinal number, card(X), in
such a way that two sets have the same cardinal number just in case they are
equivalent in this sense, i.e. card(X) = card(Y ) if and only if X Y . Cantor
conceived of card(X) as arising by a double process of abstraction from X,
rst by abstracting away from the nature of its individual elements while in
some given order, and then by abstracting away from any order whatever;
for this reason, he used
=
X to denote card(X). However, these processes of
abstraction were only vaguely explained and he was not more specic about
how card(X) might be represented in terms of some more basic objects or
4
notions.
If one grants the association of a cardinal number card(X) with each X
in some way to satisfy the above condition for equality of cardinals, there are
several ways in which a realization of the Peano axioms can be dened set-
theoretically; for simplicity, we shall give only one of these. First of all, one
denes 0 to be the cardinal number of the empty set. Then, for successor,
given x = card(X), one takes sc(x) = card(X

) where X

is obtained from
X by adjoining exactly one new element. (This denition is problematic if a
universal set X consisting of all objects is admitted; however, in that case,
one can replace X by a set Y with X Y , for which Y is not universal).
Finally, N can be dened as the smallest set which contains 0 and is closed
under successor, i.e.:
x N =
def
X[0 X&y(y X sc(y) X) x X]
This denition immediately veries the induction axiom V above, and the
other axioms IIV for 0 and sc are easily checked.
4 The logicist program in Freges hands.
There remained still one more step in order to achieve the nal reduc-
tion of the system of natural numbers, namely to give a denition of cardi-
nal numbers in terms of more basic set-theoretical or logical notions. Both
Gottlob Frege (18481925) and Bertrand Russell (18721970) advanced this
within a more far-reaching logicist program to reduce mathematics to logi-
cal principles. (For Frege this was to include arithmetic and analysis, but not
geometry; for Russell this was to comprise all of pure mathematics). Freges
work began somewhat earlier but Russell became aware of it only after hav-
ing progressed independently some distance along similar lines. Basically,
the idea for both was to dene card(X) as the equivalence class [X] of X in
the relation of between sets, i.e.
card(X) =
def
the class of all sets Y such that X Y
It should be noted that here the notions of set and class are used
interchangeably (unlike the distinction made in modern axiomatic theories
5
of sets and classes). However, the above supposed denition only gives a very
crude idea of the actual development of Freges and Russells treatments of
arithmetic; these are elaborated somewhat in the following.
In order to provide the proper conceptual framework and carry out his
program rigorously Frege rst began by developing a new system of symbolic
logic which he presented in his monograph Begrisschrift (1879). This went
far beyond the Aristotelian syllogistic and the then current logic of classes
and relations (due to Boole, DeMorgan, Peirce and Schr oder). In modern
technical terms, Freges logic was a second order predicate calculus with nega-
tion, implication and universal quantication as the basic logical operators,
and which included a complete system of the propositional and rst-order
predicate calculus. These became permanent contributions to modern logic,
though in a dierent symbolic form than that given by Frege, who used an
ideographic system of notation that was awkward to handle and contributed
to its negative reception. Modern systems of logical symbolism descend in-
stead from the work of Russell who in turn expanded that of Peano.
The rst-order or individual variables in Freges system are to be thought
of as ranging over all objects in the universe, concrete as well as abstract, and
thus among them, all objects of mathematics. The second-order variables are
supposed to range over arbitrary predicates or properties (or concepts in
Freges terminology) as well as arbitrary relations (predicates of more than
one argument). It would get us into too many technicalities to try to describe
precisely the principles accepted by Frege for the second-order variables, and
so we shall take some liberties with our explanations in the following. (Cf.
Dummett 1991 for a thorough analysis).
Basically, Frege assumed that each propositional function (x) of his sys-
tem determines a property P which holds of any x just in case (x) is true.
This is what is nowadays called the Comprehension Axiom for Properties:
P x[P(x) (x)],
where stands for there exists, for for all, for if and only if,
and where the predicate variable P does not occur in . Secondly, each
propositional function (x) determines a class x(x) (written x [ (x) in
more modern symbolism), which is its extension (or Wertverlauf, in Freges
6
terminology), and two functions (x) and (x) determine the same class
just in case they are co-extensive. That is, the Axiom of Extensionality is
assumed for classes in the form:
x(x) = x(x) x[(x) (x)].
Finally, classes are supposed to be mathematical objects and thus belong
to the universe of individuals; speaking formally, each x(x) is a rst order
term which can instantiate properties holding of all individuals. Then the
theory of classes may be interpreted in Freges system, with the membership
relation between individuals and classes introduced explicitly by the following
denition:
(y z) =
def
(P)[z = xP(x)&P(y)].
It follows from the above principles that we have a Comprehension Axiom
for Classes, in the form:
y x(x) (y).
Frege carried out his program for the logical foundations of arithmetic in the
works Die Grundlagen der Arithmetik (1884) and Grundgesetze der Arith-
metik (Vol. I 1893, Vol. II 1903). His denition of equinumerosity of predi-
cates, P Q, is equivalent to Cantors denition for the associated classes
xP(x) xQ(x). Then the number of xs such that P(x), in symbols N
x
P(x),
is dened as the class of extensions of Qs such that P Q; this serves the
role in Freges system of Card(X) for X = xP(x). With that as basis, one
can proceed to dene 0, sc and N just as for sets at the end of the preceding
section.
Despite the great care which Frege bestowed on his conceptual framework
and the precision with which he carried out his derivations, and despite
the supercial plausibility of his program, Freges system was fundamentally
awed, since it proved to be inconsistent. This was discovered by Russell in
1901 and communicated by him in a letter to Frege in 1902. The Russell
paradox, as it is now called, is simply obtained by forming the class
r =
def
x(x , x).
Thus by the Comprehension Axiom for Classes, we have
x[x r x , x]
7
In particular,
r r r , r,
which immediately results in a contradiction.
Russells letter to Frege arrived while the second volume of the Grundge-
setze was at the printers, and Frege was able to add an appendix reproducing
a derivation of Russells paradox. He also hastily suggested that a restric-
tion of the Axiom of Extensionality might serve to avoid it; however, Freges
way out was shown somewhat later to fail. In any case, Freges plan for
his logicist program was totally undermined; he never wrote the projected
third volume of the Grundgesetze and he eventually abandoned his program.
5 The logicist program: Russells salvage operation.
In his own development of the logicist program, Russell was mainly inu-
enced by the works of Cantor and Peano, the former through his theory of
classes, and the latter through his symbolic logic. Russell had met Peano in
1900 and was so impressed by the precision of his approach that he immedi-
ately set out to master it. Russell then began in earnest to elaborate his own
synthesis of Cantors and Peanos ideas, which was to appear in 1903 in his
book, The Principles of Mathematics. Its aim was to demonstrate . . . that
all pure mathematics deals with concepts denable 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 . . . ,
a program subsequently summarized by the thesis that mathematics and
logic are identical. However, the 1903 work is not presented by means of
logical symbolism, despite the technical character of the concepts, principles
and arguments involved. Rather it is set out in ordinary language for a gen-
eral philosophical and mathematical audience. Russells plan was to follow
up with a second volume in which all details of the program would be car-
ried out; this was to be co-authored by his teacher, the mathematician and
philosopher Alfred North Whitehead (18611947) and addressed exclusively
to mathematicians. Late in the preparation for publication of the Principles
Russell became aware of Freges prior work, which he recognized largely an-
ticipated his own; in fact, he had seen the Grundgesetze der Arithmetik, but
8
said that . . . owing to the great diculty of [Freges] symbolism, I had failed
to grasp its importance or to understand its contents. In recompense, Rus-
sell added an Appendix A to his own volume, under the title, The logical
and arithmetical doctrines of Frege.
It was also late in the writing of the Principles that Russell addressed
the problems raised by paradoxes (also called antinomies) in the theory of
classes. His was by no means the rst such: among others, Cesare Burali-
Forti (18611931) had published the paradox of the greatest ordinal number
in 1897, and Cantor had written Dedekind in 1899 concerning the paradox
of the largest cardinal number. Russell was aware of both, but it was in the
process of analyzing Cantors paradox in order to isolate the crucial diculty
that he was led to his own paradox of the class of all classes which are not
members of themselves. Indeed, in comparison with the previous antinomies,
Russells paradox was by far the simplest and most striking, shorn as it was
of all special notions and arguments. Regarding his own, Russell recognized
in contrast to Fregethat the crux of the problem lay in the unrestricted
assumption of the Comprehension Axiom, according to which each property
(x) determines as an object a class x(x) of all objects x satisfying . In
Appendix B to The Principles of Mathematics, Russell made his rst stab
at a solution to the contradictions, under the title, The doctrine of types.
This was to the eect that every propositional function (x) has a prior range
of signicance, the type of its variable x, only within which are questions of
its truth or falsity meaningful. Under a suitable ordering of types, then,
the type of a class is higher than the type of any argument x, and for this
reason it would not be meaningful to ask whether or not x x is true. Thus,
according to this line of reasoning, neither the property (x) = (x x) nor
its negation (x , x) is meaningful.
Russell later introduced the term predicative to distinguish those prop-
erties (x) which determine classes from those, called impredicative, which
do not, but he was unsettled as to the proper criteria to decide between the
two in each case. Between 1903 and 1908 he wrestled with this problem and
seriously considered a variety of solutions other than the use of type dis-
tinctions, among them the so-called no classes theory about which we will
speak below. In the same period he was also fending o attacks on the entire
logicist program by Henri Poincare (18541912). Ironically, it was by com-
9
bining Poincares vicious-circle principle for the analysis and solution of the
antinomies, with the doctrine of types and the no classes theory, that Russell
eventually settled on his own approach in his landmark article Mathematical
logic as based on the theory of types (1908).
Poincares attention had centered on the so-called Richard paradox of
1905 (due to Jules Richard, 18621956), which concerned not classes, but an
unrestricted notion of denability. In his writings on the subject of the para-
doxes in 1906 and thereafter, Poincare identied the source of the diculty
always to lie in the presence of a vicious circle. According to his analysis, in
each such case there is a purported denition of an object in terms implicitly
involving the object itself; for example, in the case of the Richard paradox
the denition is that of a certain real number in terms of the supposed to-
tality of all denable real numbers, while in the paradox of BuraliForti it
is that of the largest ordinal in terms of the supposed totality of all ordi-
nals. Adopting Russells terminology to this more specic analysis, Poincare
called predicative those denitions which do not involve a vicious circle and
impredicative those which do (implicitly or otherwise) and which are thus to
be banned. Russell, in his 1908 article, adopted Poincares proscription of
impredicative denitions in his own formulation of the vicious-circle princi-
ple: No totality can contain members dened [only] in terms of itself. As
indicated above, this was embodied in a modied version of the doctrine of
types, more complicated than originally envisaged in order also to encompass
the no classes theory. We now turn to an explanation of the ideas behind
the resulting formalism.
First of all, the idea of the no classes theory was that, instead of speaking
of the class of all objects having a given property, one speaks only of that
property itself, as given explicitly by means of a formula (x) for a proposi-
tional function. Under this interpretation, x(x) is read as the property
which holds just of those x satisfying (x). Now, since distinct formulas (x)
and (x) may express dierent properties and yet be satised by exactly the
same objects, the Axiom of Extensionality is no longer to be assumed. For
this reason, the no classes theory is to be regarded as an intensional theory
of properties, as opposed to an extensional theory of classes.
Next, special attention is given to (what Russell called) the apparent
variables of a formula , i.e. those that occur bound in some sub-formula or
10
sub-expression, such as an individual or property variable y in (y)(y) or
(y)(y) or y(y). In the case that y is a property variable of given type,
the totality of those properties is implicitly used in dening the property
x(x). Thus we would have a violation of the vicious-circle principle if we
allowed x(x) also to be of the same type. Indeed, Russell reformulated his
vicious-circle principle in the form: Whatever contains an apparent variable
must not be a possible value of the variable; in a suitable sense, it must of
of higher type than that variable.
In order to give specic meaning to these ideas, one must have a sys-
tematic means of assigning types to variables and to property expressions of
the form x(x), as well as an explanation of the relation of being of higher
type. In modern treatments one distinguishes two such relations, but Russell
conated them in a way that can be confusing. Thus we shall explain instead
the modern approach, which uses independent notions of type and level. The
numerals 0, 1, 2, . . . are used for both type and level assignments and then
the relation of being of higher type or level is just that of being numerically
larger.
The objects of type 0 are taken to be the individuals in the universe of
discourse; these are supposed to be simples, not subject to further analysis.
Then the objects of type 1 are understood to be properties of objects of type
0, and, in general, the objects of type n+1 are conceived of as the properties
of objects of type n. Each variable is assigned a specied type; if (x) is a
well-formed formula with x of type n, then x(x) is assigned type n +1. If
s and t are any terms (either variables or property expressions), then s t
is allowed as a well-formed formula only when s is of some type n and t is of
type n + 1. Formulas in general are built up from such atomic formulas by
propositional operations (such as negation and conjunction &) and by
applying the quantiers ( and ) to variables. If no further distinctions
as to level (to be described next) are made, the resulting symbolism is said to
be that of the simple theory of types (STT). Now the comprehension Axiom
is restricted in STT to well-formed formulas (x), to give:
y x(x) (y), when y is of the same type as x.
Already the simple theory of types excludes Russells paradox: x(x x)
is not a legitimate term in its language, no matter what the type of x is.
11
However, the eect of Russells revision of the vicious-circle principle, that
whatever contains an apparent variable must not be a possible value of that
variable required the further division into levels, which correspond to stages
of denition in any type n greater than 0. Each (predicate) variable of type
n is assigned a natural number k as level; then the level of a term of the form
x(x) must be greater than the levels of all bound variables which occur in .
Under this restriction, if a property (P)(P) holds with bound variable P
of level k and if the level of x(x) is larger than k, we cannot instantiate this
universal statement to conclude that ( x(x)). As will be seen, the inability
to do so creates genuine diculties in the foundations of arithmetic.
The logical system incorporating Russells vicious circle principle in the
ways indicated is called the ramied theory of types (RTT), in which each
type is ramied according to level. The plan for the denition of the natural
numbers in RTT was essentially the same as Freges, namely via the notion of
cardinal numbers as equivalence classes for the relation of equinumerosity.
However, in type theory (simple or ramied), there is no single such relation;
rather, one can only dene P Q for classes (qua properties) of the same
type n(,= 0). Then the equivalence classes under are objects of the next
type m = n + 1. Thus there is no single notion of cardinal number, but
rather one that is duplicated in each such type m. Russell made light of
this multiplicity of notions of cardinal numbers in his theory: he argued that
each class of a given type is equinumerous with the class of singletons of its
members, and that by successive applications of this observation, classes of
any two types can be compared as to cardinality. However, this idea cannot
be stated formally in STT, let alone RTT.
Matters become more complicated with the denition of natural numbers
in the ramied theory of types. Given any type n ,= 0, dene 0
n
and sc
n
to be the zero and successor operation for cardinal numbers of type n + 1.
Then one can only dene the property of being a natural number in the
form: Nat
n,k
(x) =
def
P[0
n
P & y[y P sc
n
(y) P] x P,
where P is a predicate variable of type n + 1 and level k. That is, we
have a double multiplicity of notions of natural number depending on both
type and ramication level. Russell observed that this creates problems for
proofs by induction on the natural numbers of properties which themselves
involve Nat
n,k
; any such property must have level greater than k, and so
12
the variable P of level k in the above denition cannot be instantiated by
x(x). In order to get around such diculties, Russell introduced an ad hoc
assumption called the Axiom of Reducibility; informally, this expresses that
each property x(x) is co-extensive with a property of the lowest level. In
eect, this Axiom obliterates the distinctions according to levels and deeply
compromises the vicious-circle principle in the very specic form stated by
Russell. Though he had initially tried to give some arguments in its favor,
he eventually stated that it is not the sort of axiom with which we can
rest content and that its justication is purely pragmatic: it leads to the
desired result and to no others.
There were other troublesome hypotheses that Russell had to make,
namely the so-called Axiom of Innity and the Axiom of Choice. The rst,
that there exist innitely many individuals, is needed to assure that each -
nite cardinal number has a successor. The second had emerged in the theory
of sets as necessary to establish many results in the arithmetic of transnite
cardinals; it asserts the existence of simultaneous choices from arbitrary col-
lection of non-empty sets, even where no means is available to dene those
choices.
The detailed execution of this greatly modied logicist program sketched
in Russells 1908 article was carried out in the monumental work of White-
head and Russell, Principia Mathematica, published in three volumes in the
period 19101913. Whiteheads name was listed rst, as the senior author,
and he certainly contributed a great deal to its writing, but the over-all plan
and major part of the work was Russells. This is what had originally been
intended as the second volume of The principles of mathematics. Although
the passage from the Principles to the Principia through the Ramied The-
ory of Types had required the explicit introduction of assumptions (such as
the Axiom of Reducibility, the Axiom of Innity and the Axiom of Choice)
whose logical status was questionable, Russell still asserted many years later,
in his introduction to the second edition of the Principles: The fundamental
thesis of the following pages, that mathematics and logic are identical, is one
which I have never seen any reason to modify. But in this respect, Russell
had hardly any followers and his extraordinary eort to salvage the logicist
program is generally deemed a failure.
In 1925, Russells student Frank P. Ramsey (19031930) pointed out (as
13
we have already seen for Russells paradox) that to block the paradoxes of
classes it was not necessary to use the full force of the vicious-circle principle
and that the distinctions according to types already served this purpose. The
choice, then, between STT and RTT would have to be on dierent grounds,
with STT interpreted as a theory of classes (for which Extensionality would
be appropriate) and RTT a theory of predicative properties. In philosophical
terms, the former ts a Platonic realist view of mathematics, while the latter
is allied with nominalistic tendencies. In the transition from the Principles to
Principia, Russell himself had consciously made this philosophical passage.
The problem for those who, following Poincare and the later Russell, would
seek predicative foundations of mathematics, would be how to carry such a
program out in a reasonable way without compromising oneself as Russell
had done with the Axiom of Reducibility. The problem for those who were
ready to accept a Platonic conception of logical and mathematical entities
would be how to set up a formal system in which informal practice could
be more readily mirrored while still avoiding the paradoxes of classes. First
answers to both these problems would be given, respectively, by Hermann
Weyl (18851955) and Ernst Zermelo (18711953), as will be described below.
While most of the actual details of the work in Principia Mathematica and
its answer to the question What is mathematics? did not have lasting
value, the fact of its existence did have enormous impact. What the Principia
showed was that it is possible to dene many mathematical notions in precise
symbolic terms from a very few basic notions and to carry out extended tracts
of mathematical argument in a completely rigorous step by step form, from
a few basic principles and rules of reasoning. Whether or not the thesis that
mathematics is logic was justied, the exercise of Principia Mathematica
certainly penetrated to the very logical roots of mathematics and broadened
to all its parts the ideal of formal rigor whose previous exemplar had been
that of Euclidean geometry, two millenia earlier.
6 Predicativism: Poincare and Weyl.
Henri Poincare, one of the foremost mathematicians of his time, is gen-
erally considered to be a precursor of both the predicativist and the intu-
itionistic programs for the foundations of mathematics. Noted for his prolic
14
contributions to all branches of mathematics, but especially of analysis and
mathematical physics, Poincare also wrote many popular books and essays
on the philosophy of science and mathematics in a lively and vivid style. In
his writings on the philosophy of mathematics, beginning in 1893 and con-
tinuing until his death in 1912, Poincare gave primary concern to the role
of intuition as against that of logic. In particular, he considered the nat-
ural number system to be directly understood and the associated principle
of proof by induction to be thereby sanctioned by intuition and thus not to
require reduction to anything (purporting to be) more basic. Beginning in
1905, Poincare mounted an unremitting attack on set-theoretical and logical
programs for the foundations of mathematics, and especially on the logicist
program. Besides seeing no need for the eorts to reduce the notion of nat-
ural number to logical concepts, he argued that a petitio was involved in
whatever way one would attempt to carry out such a reduction. For exam-
ple, the theory of types presumes an understanding of the natural numbers
in the general description of its syntax. Poincares criticism in this respect
is justied if one views the theory of types from the outside, i.e., in metathe-
oretical terms, rather than the inside, as Russell would presumably have it
(cf., however, Goldfarb 1988).
Poincares attack on Cantorian set theory rested on a fundamentally con-
trary view as to the nature of mathematics. According to him, all mathemat-
ical notions have their source in human conceptions which are either given
directly in intuition or obtained from such by explicit denition. Mathemat-
ical objects do not have an independent Platonic existence, as seems to be
required to justify set-theoretic principles (see the next section), and in par-
ticular there are no completed innite totalities. Through this denitionist
philosophy of mathematics, Poincare was led to his analysis of the paradoxes
and to the proscription of impredicative denitions: for, one must be careful
to distinguish apparent denitions from those which are proper. Impred-
icative denitions characteristically purport to single out an object from a
totality of objects by essential reference (either explicit or implicit) to that
totality. It this is viewed as the creation of such an object by denition,
one violates the requirement that the deniens must, in all respects, be prior
to the deniendum.
15
Poincare did not elaborate his ideas in any systematic way; in particular,
he said nothing about the choice of underlying symbolism for denitions or
of underlying logic for mathematical reasoning. The predicativist program
followed out his denitionist philosophy without restriction on the logic in-
volved, i.e., it took the classical predicate calculus for granted. The intu-
itionist program (to be discussed later in this article) took more seriously his
views as to the subjective source of mathematics and abandoned the logic of
truth and falsity in favor of a logic of what can be established by intuitively
evident, constructive means.
Though Russell was the rst to formulate predicative principles explicitly
in his ramied theory of types RTT, as we saw he compromised these fa-
tally by the introduction of the Axiom of Reducibility. The next substantial
advance in the predicativist program was made by Hermann Weyl in 1918,
in his monograph Das Kontinuum. An extremely broad and deep mathe-
matician of the Hilbert school, Weyl also had long-standing interests in the
philosophy of science and mathematics, but the 1918 work was his single
foray into a detailed foundational development. Apparently he arrived at a
predicativist position independently of Poincare and Russell. Like Poincare
and unlike Russell, Weyl accepted the natural number system and the asso-
ciated principles of proof by induction and denition by recursion as basic.
He credited Russell with formulation of the vicious-circle principle, while
referring to Poincares very uncertain remarks concerning impredicative
denitions. On the other hand, Weyl said that he is separated by a ver-
itable abyss from Russell in his (Russells) attempt to dene the natural
numbers as equivalence classes under the equinumerosity relation and with
his assumption of the Axiom of Reducibility.
For Weyl, the essential rst task was to see how much mathematical
analysis could be carried out on a strictly predicativist basis, given the natural
numbers. Since the rational number system can be reduced to the system of
natural numbers and since the continuum of real numbers can be identied
with sequences or sets of rational numbers (` a la Cantor or Dedekind), the
appropriate principles to consider are those that govern sets or sequences
of natural numbers. Thus Weyl erected in Das Kontinuum a second-order
axiomatic system whose variables of type 0 are interpreted as ranging over the
natural numbers, and those of type 1 as ranging over predicatively denable
16
sets, relations and functions of natural numbers. However, unlike Russell,
he did not further ramify the objects of type 1 into levels, for he saw that
this would lead to diculties of the same sort that Russell had previously
run into for arithmetic (and tried to overcome by resort to the predicatively
unjustied Axiom of Reducibility). Speaking informally, one can say that the
type 1 variables in Weyls system range only over level 0 objects of that type
in the ramied hierarchy, and the principles of denition of type 1 objects
give conditions under which the level 0 objects are closed.
Actually, Weyls axiom system does not meet modern standards of for-
malization and there are certain ambiguities that one nds when attempting
to reconstruct it in those terms. An exegesis carried out by the author (Fe-
ferman 1988) has led to two systems, one of which can be interpreted as just
suggested and a second one (still predicatively justied) that requires the
introduction of objects of type 2 and yields stronger principles at type 1. We
shall content ourselves here with an indication of the weaker system (which
is the one customarily ascribed to Weyl).
Beginning with the constant 0 and the successor operation, one may intro-
duce further specic (type 1) functions by explicit and recursive denition.
A formula built up from equations and membership statements without
any bound second-order (function, set or relation) variables is said to be
arithmetical. Then the Arithmetical Comprehension Axiom (ACA) for sets
of numbers is given by the scheme:
Xx[x X (x)],
where is any arithmetical formula which does not contain X free. (More
generally, this scheme is also formulated for relations of any number of argu-
ments). The set X dened by ACA is denoted x [ (x).
By the usual type restrictions, set variables in formulas occur only to
the right of the membership relation symbol . Then if we substitute for
the set variable in such an occurrence, say t Y , an arithmetical denition
x [ (x) of Y , the result is there equivalent to (t); it follows that arith-
metical formulas are closed under substitution for set (or relation) variables
by arithmetical formulas. (This is the closure condition on objects of type 1
and level 0 indicated above.)
17
Since Weyl assumes classical logic in his system, quantication over the
natural numbers is regarded as being truth-functionally determinate. This
may be considered to constitute implicit acceptance of the set N of natural
numbers as a completed innite totality (contrary to Poincare). But there is
no acceptance in Weyls system, either implicitly or explicitly, of a completed
totality of subsets of N, contrary to Cantorian set theory.
Rational numbers are introduced in Weyls system as 4-tuples of natural
numbers (x, y, z, w), with y ,= 0, w ,= 0, representing (
x
y

z
w
), and called
equal under the expected equivalence relation. Then real numbers are
taken to be lower Dedekind sections in the rational numbers Q, i.e. suitable
sets of such 4-tuples. By the one-one correspondence between Q and N, real
numbers correspond to certain subsets of N. Then a set S of real numbers is
given by a property which corresponds to the property (Y ) of subsets Y of
N. Now the l.u.b. (least upper bound) of S (when it is bounded above) is in
Dedekinds model simply the union of the lower sections which are members
of S. Membership in this union corresponds to membership in the union of
sets Y satisfying (Y ), i.e. it should be a set X satisfying:
x[x X Y ((Y ) & x Y )].
However, the existence of such X is not guaranteed by ACA; in fact, this is
prima facie a denition of the subset X of N from the totality of subsets of
N, hence is impredicative, and its use cannot in general be justied in Weyls
system. Since the l.u.b. axiom for real numbers is the fundamental principle
of analysis, the inability to assert the above would appear to constitute an
insuperable obstacle to Weyls predicativist program for analysis. However,
the l.u.b. principle for sequences of reals can be derived in his system, and
this turns out to be sucient for most applications. For, a sequence of real
numbers corresponds to a sequence X
0
, . . . , X
n
, . . .) of subsets of N, and
that is given by a single binary relation R with x X
n
R(x, n). Then the
union of the sets X
n
in this sequence is dened by
x[x X nR(x, n)];
the union set X thus exists by ACA. On the basis of this restricted l.u.b.
principle, Weyl was able to show in Das Kontinuum that the full standard
theory of continuous functions of real numbers (which are determined entirely
18
by their values at rational numbers) can be developed in a straightforward
way on the basis of his system. What Weyl did not do was show how to
deal predicatively with the more modern theories of integration (such as
that of Lebesgue) applying to much wider classes of functions, as needed
for various applications. This was to be left for later developments of the
predicative approach, which was not taken up again in a systematic way
until the 1950s. (Cf. e.g. Feferman 1964). Weyl himself shifted his views
two years after the publication of Das Kontinuum, in the direction of the
still more radical approach of the intuitionist L. E. J. Brouwer (18871966);
however, in later years he became pessimistic about the prospects for the
Brouwerian revolution, though he remained sympathetic with its underlying
philosophy. On the other hand, Weyl never completely disavowed his own
program for predicative foundations, which he continued to mention in his
articles on the philosophy of mathematics over the years.
7 Zermelos axioms for set-theoretical foun-
dations.
The remainder of this article concerns the three other most signicant
programs for the foundations of mathematics which were developed in the
rst third of the 20th century: set-theoretical foundations via axiomatic set
theory, the intuitionist program of constructive mathematics, and Hilberts
nitist consistency program. Because of their continued importance, these
are all treated at length in separate chapters of the Encyclopedia. Thus
we content ourselves here with a relatively brief introduction to each, with
emphasis on their philosophical dierences.
According to Cantor, any two sets A and B can be compared as to their
cardinality, i.e. card(A) card(B) or card(B) card(A). His argument
for this made use of the so-called Well-ordering Principle (WO), according
to which the elements of any set can be laid out in a transnite sequence
X
0
, X
1
, . . . , X

, X
+1
, . . .) in which each non-empty subset has a rst ele-
ment. Cantor also demonstrated that for each set A there is a set B whose
cardinality is larger than that of A, i.e. card(A) < card(B), namely the
set of all subsets X of A, written X [ X A. This is also called the
power set of A, and denoted T (A), since in the arithmetic of cardinals,
19
card(X [ X A) = 2
card(A)
for arbitrary A. Now by WO, we have a
complete transnite scale of innite cardinals, for which Cantor used the He-
brew letter (aleph) with subscripts 0, 1, . . . , , + 1, . . . ;
0
= card(N) is
the cardinal of any countably innite set (e.g. also the set of rational num-
bers), and
1
is the least uncountable cardinal. Since
0
< 2

0
it follows
that
1
2

0
. The number 2

0
is not only the cardinality of T(N), it is
also the cardinality of the set R of all real numbers (via their correspondence
with Dedekind subsets of the rational numbers); thus 2

0
is also called the
cardinality of the continuum. An immediate question to raise from the above
is whether 2

0
=
1
is true; Cantors conjecture that that is the case is called
the continuum hypothesis (CH):
2

0
=
1
To this day CH remains the basic unsettled question in Cantors theory of
cardinal numbers.
Evidently, Cantor took for granted that for any innite set A, the power
set T(A) = X [ X A is a denite completed totality with a denite cardi-
nal number. Moreover, according to his well-ordering principle, there must be
some way of setting up a well-ordering of T(A) as X
0
, X
1
, . . . , X

, X
+1
, . . .),
by making a transnite sequence of arbitrary choices of subsets of A until
T(A) is exhausted. There is no known procedure for well-ordering T(N), let
alone T(A) for any other innite set (e.g., T(R) or T(T(N)), etc.). Despite
the intuitive appeal to many mathematicians of Cantors notions and argu-
ments, even those sympathetic to his ideas found the well-ordering principle
to be problematic.
The emergence of the set-theoretical paradoxes added fuel to the concerns
about the security of the subject, though one might dismiss the sets involved
in these cases (of all sets, or all cardinals, or all ordinals) as somehow
marginal rather than central like the continuum. In addition, there was a
group of inuential mathematicians, beginning with Kronecker, who were
extremely critical of Cantors theory because of its metaphysical basis,
i.e., because of the underlying Platonistic philosophy of mathematics that it
took for granted: sets are supposed to be entities existing independently of
human constructions and denitions.
20
No wonder, then, that in the list of twenty-three major unsolved prob-
lems presented by David Hilbert (18621943) in his famous lecture for the
1900 International Congress of Mathematicians, the rst two concerned the
foundations of mathematics: namely, the continuum problem and the con-
sistency of an axiom system for the real numbers. The second problem was
presumably motivated by the aim to ensure the security of mathematical
analysis, no matter what inconsistencies one might meet at the fringes of
set theory. In his discussion of the rst problem, Hilbert said he was not
convinced by Cantors argument for the well-ordering principle WO, and he
reformulated CH in a way that did not depend on it, namely: if S R and S
is innite, then S N or S R. Hilbert thought there should also be a di-
rect proof of this remarkable statement [WO] perhaps by actually giving
a well-ordering of R.
At the turn of the century the analyst Ernst Zermelo accepted a position
at the university in G ottingen, where he came under Hilberts inuence.
Before long he had shifted his attention to the foundations of set theory
and in particular to WO. In 1904, Zermelo published an article in which
he introduced a new principle called the Axiom of Selection or Axiom of
Choice (AC), from which he proved the well-ordering principle WO. Zermelos
1904 statement of AC was that for any set A there is a function f on the
collection of all non-empty subsets of A which chooses an element of each
such subset, i.e., for which f(X) X for each non-empty X A. In eect,
AC postulated the existence of simultaneous choices which could be used to
make the required successive choices in setting up a well-ordering of A, a
step that could be deemed progressive only if AC were regarded as evident
in a way that WO is not.
Zermelos 1904 paper provoked intense criticism, both in his assump-
tion of AC and in his use of other set-theoretical concepts and principles.
Some critics mistakenly thought that a form of the Burali-Forti paradox was
involved in Zermelos argument. But the main criticism of AC was that a
function f is asserted to exist without any means available to explicitly dene
or construct it (except for those A which are already well-ordered). Indeed, it
is easily seen that WO implies AC, so they are equivalent principles. Later,
Zermelo found a modied but equivalent form of AC which appears intu-
itively clearer. It asserts that if S is any collection of disjoint non-empty
21
sets, then there is a function f on S which assigns to each X in S a member
of X; this form of AC is more readily visualized than the original one and
is considered by many to be intuitively evident, despite the fact that it is no
more constructive. (Cf. Moore 1982 for a thorough history of the Axiom of
Choice).
In order to respond to the various objections to AC and his proof of WO
from it, as well as to make clear what other set-theoretical principles were
involved, Zermelo published in 1908 the rst axiomatic system of set theory.
Intuitively speaking, the universe of discourse of this theory consists of some
basic underlying objects, called urelements, and sets built up out of these
objects. Sets are supposed to exist independently of any means of dening
them, so the Axiom of Extensionality is assumed, i.e. AB[x(x A x
B) A = B]. Zermelos axioms also assert the existence of various sets,
such as the empty set 0, and the singleton of any x, x. Further axioms
allow one to form the union of any two sets, A B, and the power set of
any set, T(A). The existence of an innite set is postulated by having an
A which contains 0 and which satises x(x A x A); by taking
sc(x) = x one can identify the set N of natural numbers with the smallest
subset of A containing 0 and closed under this successor operation. More
generally, Zermelo posited a restricted form of the comprehension axiom,
called the Axiom of Separation (Aussonderungsaxiom), which says that for
any set A and any denite property P(x) of elements of A we can form the
subset B of A consisting exactly of those x A satisfying P(x); in symbols,
B = x A [ P(x).
An informal model of these axioms was rst explained by Zermelo in 1930.
Let V
0
be the set of urelements and let for each n < , V
n+1
= V
n
T(V
n
);
then take V

=

n<
V
n
and continue again with V
+1
= V

T(V

) for
= + n. The model is given nally as V
+
=

n<
V
+n
. Note that
0 V
1
and if x V
n
, then x V
n+1
. Hence N = 0, 0, 0, . . . V

and so N V
+1
. This informal model is called the cumulative hierarchy
since at each stage we combine everything obtained up to that stage, V

,
with all subsets of everything so far obtained, T(V

). This contrasts with

the informal model for simple type theory where one takes S
0
= V
0
and
S
n+1
= T(S
n
) for each n. There is no natural way to carry that further into
the transnite without taking S

=

n<
S
n
and thus cumulating the objects
22
together; if that is to be done, one might as well do it at each stage.
Zermelos system is type-free in the sense that it is meaningful to consider
whether x y for any objects x and y; in particular, one can ask whether
x x holds. However, there is no (evident) way to derive Russells paradox
in his system because of the restricted nature of his comprehension axiom,
the Axiom of Separation. From it we can only conclude that for any set
A we can form the set B = x A [ x , x; but in this case we merely
draw the perfectly consistent conclusion that B , A (for if B A, then B
B B , B). It follows that there is no set of all sets in Zermelos system.
Similarly there is no set of all ordinals or of all cardinals, so the paradoxes of
Burali-Forti and of Cantor are also blocked. Of course, it is still conceivable
that Zermelos axiom system is inconsistent; what the preceding shows is
that there is no obvious inconsistency to be obtained by reproducing one of
the familiar antinomies. On the other hand, there is no means in Zermelos
system to treat cardinal numbers as equivalence classes under `a la Frege
and Russell, since one cannot establish the existence of the set of all sets
equivalent to a given set. In Zermelo-style axiomatic systems alternative
means must therefore be prescribed for identifying cardinal numbers (and,
similarly, ordinal numbers) with specic sets; that was accomplished in the
simplest way in the late 1920s by John von Neumann (19031957).
Later improvements and extensions of Zermelos set theory were given
by Thoralf Skolem (18871963) and Abraham Fraenkel (18911965), Paul
Bernays (18881977) and Kurt G odel (19061978). Thus one speaks nowa-
days of the axiom systems ZF (ZermeloFraenkel) and BG (BernaysG odel).
With the Axiom of Choice added, these provide extremely exible systems in
which all of Cantorian set theory can be represented without any apparent
source of inconsistency. In that respect, they are far superior to theories of
types and have superseded them as a foundation for that part of modern
mathematical practice which makes essential use of set-theoretical concepts
and principles. (See the chapter on set theory in this Encyclopedia for further
detail on these systems).
Zermelos axiomatization and its descendants brought out more clearly
than ever the grounding of set theory in a Platonistic conception of math-
ematics. The ingredients of the latter include the views that: (i) Sets are
23
entities existing independently of human thoughts and constructions, which,
though abstract, are supposed to be part of an external, objective reality;
(ii) innite sets such as the natural numbers and the real numbers are sup-
posed to exist as actual, complete objects; (iii) for each set, the totality of
arbitrary subsets of that set exists as a denite, completed set; (iv) every
proposition about sets has a denite truth value (true or false), independent
of any means we may have to verify it.
It should be noted that acceptance of the general Platonistic position,
under which mathematical objects are viewed as independently existing (ab-
stract) entities, does not necessarily commit oneself to any or all of (i)(iv).
It is for this reason that acceptance of (i)(iv) is called set-theoretical Platon-
ism. Among the statements readily granted on this position are the Axioms
of Extensionality, Innity, Power Set, Separation and Choice; moreover, in
logic one accepts the Law of Excluded Middle, according to which for any
proposition , either or holds, in symbols . As we have seen in
Section 6, the use of the Power Set Axiom applied to N, together with the Ax-
iom of Separation for arbitrary properties formulated in the language of set
theory, leads to impredicative conclusions. The acceptance of classical logic
embodied in the Law of Excluded Middle leads directly to non-constructive
existence results by the method of proof by contradiction. This was one of
the features seized on in the constructivist critique of set-theoretical Platon-
ism and in the development of an opposing foundational program, which we
take up next.
8 Brouwers intuitionism.
The earliest critic of Cantorian set theory from the constructivist stand-
point was Leopold Kronecker, Cantors former teacher. Beyond what is nite,
he would admit only potentially innite sets to mathematics and, of such, only
those reducible to the natural number sequence 0, 1, 2, . . .; furthermore, he
would countenance only constructive existence proofs. His attacks on Can-
tors work in the latter part of the 19th century were quite severe and had
a deleterious personal eect on Cantor. Henri Poincare was, as we already
mentioned, another critic of set theory and logicism who is considered to be
a precursor of the intuitionistic brand of constructivism in certain respects,
24
though not consistently so, nor in any systematic way. More or less con-
temporary with him were the so-called semi-intuitionists, comprising the
French mathematicians Emile Borel (18711956), Rene Baire (1874-1932),
and Henri Lebesgue (18751941) among others. They accepted countably
innite sets and constructions and the transnite iteration of such up to the
least uncountable ordinal. The semi-intuitionists did not accept the Axiom of
Choice, though some forms of this were unwittingly involved in their mathe-
matics (cf. Moore 1982). And while they thought all existence proofs should
be obtained by explicit construction or denition, they did not argue for the
restriction of the logic employed.
It was the Dutch mathematician L. E. J. Brouwer who rst focused at-
tention on the question of justiability of the Law of Excluded Middle (LEM)
for a constructivist view of mathematics. In his 1908 article The unreliabil-
ity of logical principles, Brouwer argued that LEM for innite sets is based
on an unjustied extension of that principle from nite sets. In his doctoral
dissertation of 1907, he had already insisted on the subjective origin of math-
ematics in human intuition, and on the necessity to restrict questions of truth
in mathematics to those statements which can be veried or disproved. Of
course, for a nite set A and decidable P we can verify xP(x)xP(x) by
testing each x A in turn to see whether or not P(x) holds. But in general
there is no way to carry out such a verication when A is innite, even for
decidable P.
From 19081913 Brouwer turned away from foundational questions to
the subject of topology, to which he made deep and important contributions
that established his credentials as a mathematician of the rst rank. Then
from 1918 until the last decade before his death (in 1966) Brouwer set out to
redevelop mathematics entirely on intuitionistic grounds (the term he used
for his form of constructivism). Not only was the Law of Excluded Middle
to be rejected, but also all forms of the completed innite. To verify xP(x)
for x ranging over an innite set S, one must actually produce some a S
for which P(a) is proved; on the other hand, to verify xP(x) one must have
a proof which shows how, given any a S, one may produce a proof of
P(a). The latter justies proof by induction on the natural numbers: given
P(0) and x(P(x) P(sc(x)) one recognizes that one has a procedure
which, given any n N, establishes P(n), rst from P(0) P(1), then from
25
P(1) P(2), . . ., until we reach P(n 1) P(n).
The rst obstacle to the straightforward redevelopment of mathematics
according to intuitionistic tenets arises not in the theory of arithmetic (i.e.,
of N), but in the theory of real numbers. To be sure, real numbers could be
identied with potentially innite Cauchy sequences of rationals, but what
would it mean to deal with arbitrary such sequences? It turns out that if
one restricts attention to just those sequences which are determined by eec-
tive laws, all sorts of anomalies appear. Brouwer introduced instead a novel
conception, that of (free) choice sequences, such as might be determined in
non-lawlike ways by a series of free choices or random acts, and of which
one would have only nite partial information at any stage. (Brouwer was
anticipated in this conception of the continuum by Borel). With real num-
bers viewed as Cauchy choice sequences of rationals, a function from reals
to reals can be determined using only a nite amount of information about
any argument, in order to determine its value to any given degree of preci-
sion. Following this line of reasoning, Brouwer came to the conclusion that
any such function must be continuous, in direct contradiction to the classi-
cal existence of discontinuous functions. Increasingly, Brouwer found himself
forced to introduce unconventional notions and to reach strange conclusions,
which, despite his stature as a mathematician, were received with general
incomprehension and/or rejection by the mathematical community at large.
Nevertheless, the metamathematical study of intuitionism, initiated by his
student and disciple, Arend Heyting (18981980) sparked later development
of other systematic forms of constructivism. These are detailed in the chap-
ter, History of constructivism in the 20th century, in this Encyclopedia.
(cf. also Heyting 1971, and Troelstra and van Dalen, 1988).
9 Hilberts nitist consistency program.
When Hilbert delivered his address Mathematische Probleme at the
International Congress in 1900 he was already acknowledged to be one of the
worlds greatest mathematicians of his time. At mid-career, he had, by then,
made fundamental contributions to algebra, number theory, geometry and
analysis; in the years to come he would make further important contributions
to analysis, mathematical physics and the foundations of mathematics. The
26
list of twenty-three problems in his 1900 address covered considerable parts of
mathematics and have been the source of a great deal of eort and attention
up to the present day; those who solved or made substantial progress on one
or another of Hilberts problems were guaranteed fame in the mathematical
community (cf. Browder 1976).
As has already been mentioned, the rst two of Hilberts problems directly
concerned the foundations of mathematics. Hilberts program to establish the
consistency of axiomatic systems was rst expressed in more specic terms
in Problem 2 with its call for a proof of the consistency of a system of axioms
for the real numbers. This had its origins in Hilberts work in earlier years on
the axiomatic foundations of geometry. In that project, Hilbert had returned
to Euclids axioms, rst to improve them to meet modern standards of rigor,
and then to raise new, metatheoretical questions, such as their independence
and consistency. He demonstrated the consistency of Euclidean plane ge-
ometry by its interpretation in the Cartesian plane, given by pairs of real
numbers, and for the Euclidean geometry of space, in the three-dimensional
real number coordinate system.
Hilberts call for a proof of the consistency of the real number system thus
went a step beyond that for geometry. His statement of Problem 2 shows why
he thinks this is necessary: According to him, the foundations of any science
must provide an exact and complete system of axioms; moreover, a mathe-
matical concept exists if, and only if, such a system of axioms can be shown
to be consistent. In particular, the proof of the compatibility of the axioms
[for real numbers] is at the same time the proof of the mathematical existence
of the complete system of real numbers. What Hilbert did not provide at
the time was any suggestion as to how such a consistency proof might be car-
ried out. Evidently, he believed that it would be straightforward, and that
one would then go on to demonstrate the existence of Cantors transnite
number classes by a consistency proof, just as that of the continuum. But
in a 1904 article he presented a less optimistic prospect for that extension,
in view of the set-theoretical paradoxes.
During the subsequent years, through World War I, Hilberts attention
was partially drawn aside by work on integral equations and mathematical
physics. However, he returned in print to foundational questions in his 1918
27
address Axiomatic thought, where he began to lay out his mature program
for the foundations of mathematics; this was then elaborated in a succession
of publications through 1931. Hilberts ultimate aim was to justify the use of
Cantorian notions and methods in mathematics. The avenue for doing this
would be provided by a consistency proof of a Zermelo-style axiom system
for set theory. But Hilbert was sensitive to the criticisms of Kronecker and
Brouwer and recognized that a genuine justication of this sort of the actual
innite would have to be based on methods from which all innitary ingre-
dients were drained. Hilbert did not think this could be accomplished all at
once, but would have to be carried out step by step, rst for an axiomatic
theory of arithmetic, then for analysis and eventually for set theory.
In more detail, Hilberts program was conceived of as follows. A given
informal body of mathematics is to be treated as formally represented in
an axiomatic theory T, which is to be specied precisely within a formal
language L. This language is to be given by a stock of basic symbols and by
rules for building up from them the well-formed formulas (or statements) of
L as nite sequences of basic symbols. Certain of those formulas are then
to be specied as axioms, including both those of a general logical character
and those concerning the intended subject matter of T; in addition, there are
to be given rules of inference for constructing formal proofs (or derivations)
from the axioms, as nite sequences of formulas. Under these hypotheses,
it can be eectively decided, for each nite sequence of nite sequences
of basic symbols and each formula , whether or not is a proof of .
Once a formal axiomatic theory T is presented in such a way, the set of
provable formulas of T is determined as the end results of such proofs, and
T is consistent if no formal contradiction & is provable from T. The
statement of the consistency of T is purely nitary in the sense that it refers
only to possible nite congurations given as nite sequences of terms, each
of which is a nite sequence of symbols of L. Hilberts program to produce
uncontrovertible justication of such T required that the consistency proof
of T be carried out by purely nitary methods.
In addition to his general program, Hilbert also proposed some specic
proof-theoretical techniques to carry it out. As he saw the matter, the actual
innite already arises in classical arithmetic in formulas such as xR(x)
xR(x), where R is decidable, since this implicitly involves the possibility
28
of surveying the totality of natural numbers (just as Brouwer had argued).
Purely nitary statements would be those of the form R(x) with R decidable
and x an open free variable, which could be veried for each specic instance
R(a) in a domain of nite objects such as the natural numbers. Hilberts
idea was that a proof of such R(x) should somehow be transformed by the
succession elimination of quantied statements in into a proof / of R(x),
all of whose statements are of this nitary form. Hilberts Ansatz showed
how this might be carried out in relatively simple cases, but he left it to
his assistants and co-workers in G ottingen especially to Paul Bernays,
Wilhelm Ackermann (18961962), and John von Neumann to carry out
the Beweistheorie, or theory of proofs, for his program. The initial target
would be a consistency proof for a rst-order version PA of Peano Arithmetic.
But after some missteps by Ackermann (who thought he had obtained not
only a consistency proof for PA but one for a theory of analysis as well), von
Neumann realized that only the consistency of a fragment of PA had actually
been established by clearly nitary methods.
That this was no temporary failure emerged from the stunning results of
Kurt G odel in 1931 (cf. G odel 1986), whereby if T is any nitarily presented
formal theory T which includes PA, the consistency of T cannot be proved by
methods that can be formalized in T, unless T is already inconsistent. More-
over, it appeared that all nitary methods of the sort that had been employed
in the Hilbert school could readily be formalized in PA. Thus, if one could
not come up with essentially new nitary methods that go beyond PA one
could not hope to give a nitary consistency proof of PA (assuming that it is,
indeed, consistentwhich could hardly be doubted). Just such methods were
later proposed by Gerhard Gentzen (19091945) in 1936, by the use of trans-
nite induction up to certain nitely described countable ordinals, applied to
statements of purely nitary form. However, Gentzens extension of nitary
mathematics is controversial and it is by no means generally agreed whether
his proof of the consistency of PA meets Hilberts criteria. At any rate, it
can hardly be doubted that all nitary methods can be formalized in a (rela-
tively weak) part T of Zermelo set theory, and for such T G odels theorem on
the unprovability of consistency constitutes a denitive obstacle to Hilberts
program. For the further history of this subject, see the chapters on Godels
incompleteness theorems and the theory of proofs (or demonstrations) in this
Encyclopedia.
29
Hilbert himself refused to concede (in his Preface to Hilbert and Bernays
1934) that G odels theorem signaled the breakdown of his program. But he
should already have taken cognizance of another result which undermined
his overall views. Namely, the theorem of L owenheimSkolem, established
by 1920, showed that any theory T couched in rst-order logic which has an
innite model must have a denumerable model. Thus no such T could be a
categorical axiom system for the real numbers, i.e., would uniquely determine
the real number concept up to isomorphism. To obtain a categorical system
of axioms for the reals one would have to use essentially innitary higher-
order axioms such as the statement of Cauchy completeness; but this would
take us out of the realm of nitary formal axiomatic systems, contrary to the
requirements of Hilberts program.
What Hilbert did achieve in his teachings on logic and foundations (see
Hilbert and Ackermann 1928) was the nal transformation of the subject
whose initial aim, in the hands of Frege and Russell, had been to set up a
global, universal system of logic encompassing all of mathematics to one
in which the objects of interest would be various individual local axiomatic
theories T for dierent parts of mathematics, to be treated by mathematically
rigorous methods in the emerging and many-faceted discipline of mathemat-
ical logic or metamathematics. Within this, moreover, the theory of proofs
initiated by Hilbert has developed into a substantial and technically sophis-
ticated part of mathematical logic, in which relativized forms of Hilberts
program have been successfully pursued (cf. Feferman 1988, Sch utte 1977,
and Takeuti 1987).
30
References
Brouwer, L. E. J., Collected Works, vol. 1, edited by A. Heyting, North
Holland, Amsterdam, 1975.
Browder, F. (edited by), Mathematical developments arising from Hilbert
problems, Proc. of Symposia in Pure Mathematics, vol. 28, American Math.
Soc., Providence RI, 1976.
Cantor, G., Gesammelte Abhandlungen mathematischen und philosophischen
Inhalts, edited by E. Zermelo, Springer, Berlin, 1932.
Dedekind, R., Was sind und was sollen die Zahlen? Brunswick, 1888.
Dummett, M., Frege: Philosophy of Mathematics, Duckworth, London, 1991.
Feferman, S., Systems of predicative analysis, J. Symbolic Logic 29, 1964,
130.
Feferman, S. Hilberts program relativized: Proof-theoretical and founda-
tional reductions, J. Symbolic Logic 53, 1988, 364384.
Feferman, S., Weyl vindicated: Das Kontinuum 70 years later, in Temi e
prospettive della logica e della losoa contemporanee I, CLUEB, Bologna,
1988, pp. 5993.
Fraenkel, A. and Y. BarHillel, Foundations of set theory, NorthHolland,
Amsterdam, 1958.
Frege, G., Begrisschrift, eine der arithmetischen nachgebildete Formel-sprache
des reinen Denkens, Nebert, Halle, 1879; English translation by S. Bauer
Mengelberg in (van Heijenoort, 1967), pp 182.
Frege, G., Die Grundlagen der Arithmetik, eine logisch-mathematische Un-
tersuchung uber den Begri der Zahl, Breslau, 1884; English translation (with
the German text) by J. L. Austin in The foundations of arithmetic. A logico-
mathematical inquiry into the notion of number, Blackwell, Oxford, 1950.
31
Frege, G., Grundgesetze der Arithmetik, begrisschriftlich abgeleitet, vol. I,
Olms, Hildesheim, 1893, and vol. II, Pohle, Jena, 1903; partial English
translation by M. Furth in The basic laws of arithmetic. Exposition of the
system, Univ. of California Press, Berkeley, 1964.
Godel, K., Collected Works, vol. I. Publications 19281936, edited by S.
Feferman, J. W. Dawson, Jr., S. C. Kleene, G. H. Moore, R. M. Solovay and
J. van Heijenoort, Oxford Univ. Press, New York, 1986.
Goldfarb, W., Poincare against the logicists, in History and Philosophy of
Modern Mathematics, edited by W. Aspray and P. Kitcher, Univ. of Min-
nesota Press, Minneapolis, 1988, pp. 6181.
Hallett, M., Cantorian set theory and the limitation of size, Oxford Univ.
Press, Oxford, 1984.
Heyting, A., Intuitionism. An Introduction (3d edn.) North-Holland, Ams-
terdam, 1971.
Hilbert, D., Axiomatisches Denken, Mathematische Annalen, vol. 78, 1918,
pp. 405415.
Hilbert, D.,

Uber das Unendliche, Mathematische Annalen, vol. 95, 1926, pp.
161190; English translation by S. Bauer-Mengelberg, in (van Heijenoort,
1967), pp. 464479.
Hilbert, D. and W. Ackermann, Grundz uge der theoretischen Logik, Springer,
Berlin, 1928; English translation by L. M. Hammond et al., Principles of
mathematical logic, Chelsea, New York, 1950.
Hilbert, D. and P. Bernays, Grundlagen der Mathematik, vol. I, Springer,
Berlin, 1934; second edition, 1968.
Moore, G., Zermelos axiom of choice: Its origins, development and inu-
ence, Springer, New York, 1982.
Parsons, C., Mathematics, foundations of, in Encyclopedia of Philosophy, vol.
5, edited by P. Edwards, Macmillan, New York, 1967, pp 188213.
32
Peano, G., Arithmetices principia, nova methodo exposita, Turin, 1889, En-
glish translation in (van Heijenoort, 1967), pp. 83-97.
Poincare, H., La science et lhypoth`ese, Flammarion, Paris, 1902.
Poincare, H., Derni`eres pensees, Flammarion, Paris, 1913.
Ramsey, F. P., The foundations of mathematics, and other logical essays,
edited by R. B. Braithwaite, Paul, Trench,Trubner, London, 1931, reprinted
1950 (Humanities Press, New York).
Richard, J., Les principes des mathematiques et le probl`eme des ensembles,
Revue generale des sciences pure et appliquees, vol. 16, p. 541; English
translation in (van Heijenoort, 1967), pp. 142144.
Russell, B., The principles of mathematics, Cambridge Univ. Press, Cam-
bridge, 1903; second edition, Norton, New York (no date).
Russell, B., Mathematical logic as based on the theory of types, Amer. J. of
Mathematics, vol, 30, 1908, pp 222262, reprinted in (van Heijenoort, 1967),
pp. 150182.
Sch utte, K. Proof Theory, Springer Verlag, Berlin, 1977.
Takeuti, G., Proof Theory (2nd edn.), North-Holland, Amsterdam, 1987.
Troelstra, A. S., and D. Van Dalen, Constructivism in Mathematics. An
introduction, Volume I, North-Holland, Amesterdam, 1988.
Van Heijenoort, J. From Frege to G odel: A source book in mathematical logic,
18791931, Harvard Univ. Press, Cambridge MA, 1967.
Von Neumann, J., Zur Hilbertschen Beweistheorie, Mathematische Zeit-
schrift, vol 26, 1927, pp. 146.
33
Weyl, H., Das Kontinuum. Kritische Untersuchungen uber die Grundlagen
der Analysis, Veit, Leipzig, 1918; Italian translation by B. Veit, Il continuo.
Indagine critiche sui fondamenti dellanalisi, Bibliopolis, Napoli, 1977; En-
glish translation by S. Pollard and T. Bole, The continuum. A critical
examination of the foundations of analysis, Thomas Jeerson Univ. Prss,
Kirksville, 1987.
Weyl, H., Philosophy of mathematics and natural science, Princeton Univ.
Press, Princeton, 1949.
Whitehead, A. N. and B. Russell, Principia Mathematica, vols 1/2/3, Cam-
bridge Univ. Press, Cambridge, 1910/12/13; second edition 1925/27/27.
Zermelo, E., Beweis, dass jede Menge wohlgeordnet werden kann, Mathe-
matische Annalen, vol. 59, 1904, pp. 514516; English translation in (van
Heijenoort, 1967), pp. 139141.
Zermelo, E., Untersuchungen uber die Grundlagen der Mengenlehre I, Math-
ematische Annalen, vol. 65, 1908, pp. 261281; English translation in (van
Heijenoort, 1967), pp. 199215.
Zermelo, E.

Uber Grenzzahlen und Mengenbereiche, Fundamenta Mathemat-
icae, vol. 16, 1930, pp. 2947.
34