Set theory Phil 124

Christian W uthrich Winter 2014

Remark. I claim no originality in the following content, and sometimes not even for the particular formula-
tions used. There are only so many ways in which one can express mathematical denitions grammatically
and with full rigour. In order not to break the ow of the presentation, however, I have not given precise
credit where it is due, but only summarily in the references.
1 What is a set?
It is actually already a very hard question to state exactly how a set diers from the (mereological)
fusion of its elements, and how it diers from other types of collections. We ignore these subtle,
yet important, questions, at least for now and hope that the concept of a set will emerge through
the following work. So lets just dive in.
Denition 1 (Set). A set is any collection into a whole of denite well-distinguished objects of
our intuition or our thought. (Georg Cantor, 1895 in Lexikon der Mathematik)
Denition 2 (Set, alternative denition). A set is a many, which can be thought of as one, i.e.,
a totality of denite elements that can be combined into a whole by a law.
The identity criterion for sets is rather straightforward and demands (and is usually stated, in a
weaker form, as an axiom in axiomatic set theory, e.g., as an axiom of extensionality) that two
sets are identical just in case they have exactly the same elements (or members).
1.1 Specifying a set
There are two ways to specify sets: either we do it extensionally and use curly brackets between
which all its elements are listed, or we do it intensionally by giving some sort of constructive rule,
which allows us to uniquely determine what its elements are, in which case we say that the rule
or condition generates the set.
We can thus extensionally dene a set S by stating S = Mercury, Venus, Earth, Mars, Jupiter,
Saturn, Uranus, Neptune. More abstractly, S = a, b, c says that S is a set whose elements are
a, b, and c.
1
Alternatively, we could specify the set intensionally by stating that S is the set of all
planets in our solar system. Abstractly, if F(x) is a determinate property or condition, we can say
that the set S consists of all objects x which exemplify property F(x) or satisfy condition F(x).
In our notation:
S = x : F(x).
The planetary example can thus be written as S = x : x is a planet in our solar system.
The naive comprehension or existence axiom states that every genuine condition or property F(x)
does in fact generate a set via above procedure. As we will see shortly, this axiom spells trouble.
1
Important note: I often distinguish between a set and its elements, as is common in introductory texts, by
using uppercase Latin letters for the former and lowercase Latin letters for the latter. But this is unprincipled. I
should, as advanced texts usually do, just use one or the other.
1
1.2 Elementary relations
The membership relation, denoted by the inx , obtains between two sets or between an atom
and a set just in case one is an element of the other. Apart from identity, this is the only non-
logical symbol which enters set theory primitively. In the previous case, we thus write, for example,
a S. Since d is not an element of S, we write d , S or (d S).
Another key relation is the subset relation, denoted by the inx .
Denition 3 (Subset). A set S is a subset of another set T just in case every element of S is
also an element of T. Symbolically, S T : a(a S a T).
Denition 4 (Proper subset). S is a proper subset of T, symbolically S T, just in case that
S T and T , S, i.e., every element of S is an element of T, but not every element of T is an
element of S.
In order to give an exemplary list of all subsets of a given set, we need another denition:
Denition 5 (Empty set). The empty set is the (unique) set which contains no elements. Sym-
bolically, the empty set is denoted by .
Why should we think that there is such a thing as an empty set? Apart from its great practical
utility in mathematics, it helps in formalizations of the intuitive dierence between, as Raymond
Smullyan (1992) says, an empty theatre and no theatre at all. Nota bene: no curly brackets are
used for the empty set, except in the alternative notation of .
With this in place, we can now state the following example:
Example 1. For a set S = a, b, c, the following sets are subsets: a, b, c, a, b, a, c, b, c,
a, b, c and .
It is important to note that for any set S, S and are always among its subsets.
Let us dene another concept while we are at it (not a relation):
Denition 6 (Power set). The power set P(S) of the set S is a set whose elements are exactly
the subsets of S. Symbolically, P(S) := T : T S.
Example 2. The power set of S = a, b, c is P(S) = a, b, c, a, b, a, c, b, c, a, b, c, .
For a nite set S (more on those below) of cardinality n, the cardinality of P(S) is 2
n
, hence the
name.
Here are a few more important relations:
Denition 7 (Relative complement). The relative complement ST of T in S, or the set-theoretic
dierence of S and T, is the set of all elements in S, but not in T. Symbolically, ST := x : x S
and x , T.
Denition 8 (Intersection). The intersection of two sets S and T is the set consisting of those
elements which are contained in both S and T. Symbolically, S T := x : x S and x T.
Denition 9 (Union). The union of two sets S and T is the set consisting of those elements
which are contained in either S or T. Symbolically, S T := x : x S or x T.
It is straightforward to generalize the denitions of intersection and union to an arbitrary number
of sets. In this case, the relation is symbolized not by an inx, but by a prex, like so:
2
Denition 10.

S := x : T S(x T).
Denition 11.

S := x : T S(x T).
A bit more terminology.
Denition 12. Two sets are disjoint just in case their intersection is the empty set. A set of sets
is pairwise disjoint if every pair of sets is disjoint, i.e., if no object belongs to more than one of
the sets.
1.3 Some special and specially important sets
Arguably the most important one of thosethe empty set, we have already encountered. An-
other special class of sets is that of singleton sets, i.e. of sets containing just one element such
as the singleton set containing Aristotle, Aristotle. Importantly, Aristotle is dierent from
Aristotle. Some sets are important because they have additional algebraic structure which makes
them very important in mathematics. Here are some famous examples of sets of numbers with
such additional algebraic structure (from http://en.wikipedia.org/wiki/Set_(mathematics)
#Special_sets):
P, the set of all primes: P = 2, 3, 5, 7, 11, 13, 17, ....
N, the set of all natural numbers: N = 0, 1, 2, 3, ....
Z, the set of all integers: Z = ..., 2, 1, 0, 1, 2, ....
Q, the set of all rational numbers: Q = a/b : a, b Z, b ,= 0. Since all integers z can be
expressed by the fraction z/1, it follows that Z Q.
R, the set of all real numbers. Apart from all the rational numbers in Q, R contains irrational
numbers such as , e, and

2 (and others that cannot be dened).
C, the set of all complex numbers: C = a +bi : a, b R.
H, the set of all quaternions: H = a +bi +cj +dk : a, b, c, d R.
It is an interesting fact that P N Z Q R C H. Since these are proper subset relations,
does this mean that the sets get larger and larger as we move from left to right? The answer to
this question is not evident, since these sets are all innite, in a sense yet to be dened. As it
turns out, this is a rather deep question in the foundations of mathematics. Interestingly, it only
turns on a seemingly simple activity that you have all learned a long time ago: counting.
2 Counting all the way to innity
In order to answer the question in the previous paragraph, and in order to conceptualize innity
in a mathematically rigorous sense, we need to understand what it means to say that two sets S
and T are of the same size, or are equinumerous, or have the same cardinality, or have the
same power. Usually, we mean that the elements of the two sets can be matched in a one-one
fashion [see blackboard]. This means that there exists a map f : S T which is one-to-one and
onto, or bijective as mathematicians say. This map (or function) f is a non-ambiguous rule
which associates with each element x S an element f(x) T.
Bijectivity or, equivalently, the relationship of one-to-one and onto conjoined is dened as follows
[see also blackboard]:
3
Denition 13 (Bijective map). A map f : S T is bijective just in case it satises the following
two conditions:
1. For any elements x
1
, x
2
S, if x
1
,= x
2
, then f(x
1
) ,= f(x
2
) (one-to-one).
2. For all y T, there exists an x S such that y = f(x) (onto).
If there exists a bijective map between two sets, we say that its elements can be brought into a
one-one correspondence. Thus, two sets S and T are the same size, or have the same cardinality,
or are equinumerous i there exists a bijective map f : S T. By setting T = N, this denition
allows us to count the elements of a set and to say exactly how many elements a given set has.
Denition 14. For a given n N, a set has exactly n elements if its elements can be brought
into one-one correspondence with the set of natural numbers from 1 to n.
It immediately follows that the above dened set of all planets in our solar system has eight
elements and is thus equinumerous to the set of chemical elements up to oxygen, the set of all
days of Hanukkah, the set of apparitions appearing to Macbeth in Act 4, Scene 1 of Macbeth, and
the set of women appearing in Fran cois Ozons 8 femmes.
2.1 Finite and innite sets
From these denitions, one can then easily articulate a rigorous denition of when sets are nite
or innite.
Denition 15 (Finite set). A set S is nite i there exists an n N such that S has n elements.
Equivalently, S is nite just in case if it cannot be brought into one-one correspondence with a
proper subset of itself.
Denition 16 (Innite set). A set S is innite i it is not nite, i.e., if there exists no n N
such that S has n elements. Equivalently, S is innite just in case it can be brought into a one-one
correspondence with a proper subset of itself.
Exercise 1. Suppose that S is an innite set. Prove that by removing one single element, the set
remains innite. Hint: use the rst part of the denitions of nite and innite.
Exercise 2. Show that the set N is innite by stating a proper subset S N and a bijective map
f : N S.
2.2 Grasping the innite: Hilberts Hotel
Innity is a strange animal. In order to appreciate this, let us engage in a little thought experiment
called Hilberts Hotel, after David Hilbert, which beautifully illustrates this strangeness. It will
also serve as a propaedeutic for Georg Cantors diagonalization proof. What follows in this section
and the next is closely based on Smullyan (1992, Ch. 18).
Suppose you have a hotel of one hundred rooms. There is exactly one guest in each of the rooms.
Late at night, a traveller arrives and requests a room for the night. Can she be accommodated?
The problem is that neither the traveller nor any of the guests are prepared to share rooms,
which renders her accommodation impossible since one cannot put 101 people into a one-one
correspondence with 100 rooms.
But this is very dierent in Hilberts Hotel, which has an innite number of rooms. If it is equally
fully booked by an innite number of lonely, non-sharing guests, the manager can accommodate
the late arrival, as long as everyone is prepared to move rooms.
4
Exercise 3. How can this be achieved?
But just as everyone went back to sleep after having accommodated the traveller, a bus full of
new guests arrives, all seeking a bed for the night. The problem is that its a very large bus; in
fact, the bus contains innitely many new guests! The surprising fact is that all of them can be
accommodated as long every original guest is prepared to move to a new room once.
Exercise 4. How?
Hilberts Hotel, as it turns out, it just one hotel of Hilberts Chain of innitely many hotels with
innitely many rooms each. Suppose that all hotels of Hilberts Chain are fully booked with
exactly one guest in each room of each hotel. One day, the management decides to remodel all but
one of the chains hotels. In order to loose no business, they hatch a plan of how to accommodate
all the guests in all the hotels which close down in the one hotel which will remain open during
the remodelling of the others.
Exercise 5. Is this possible? If so, how? If not, why not?
Exercise 6. Which of the two innitiesthe Chains total number of rooms prior to and during
the remodelling, if any, is larger?
These examples show some of the strange properties of innite sets. For instance, Hilberts Hotel
and Chain show that (and how) there exist bijective maps between innite sets and some of their
proper subsets. I have stated this above as part of the denition of what it is for a set to be
innite. In fact, however, this can be shown as a theorem given just the rst part of the denitions
of nite and innite sets. As strange as this seems, it is not really paradoxical. And it has certainly
been known, at least for special cases, for quite a while. E.g., Galileo has pointed out that the
natural numbers and their squares can be brought into one-one correspondence by the bijective
map f(n) = n
2
.
Despite his advocacy of actual innitiesas opposed to the Aristotelian doctrine that there only
exist potential innities, particularly as concerning the number of individuals in our world,
Gottfried Leibniz did not countenance actual innities of numbers, because, as he claimed, it is
inconsistent with the axiom that the whole is greater than any proper subpart. But cases such as
Hilberts Hotel show that this is not true at least in a very general class of theories. This can also
be seen from another simple thought experiment [see blackboard].
Thus, even though there is a sense in which Hilberts Chain contains more rooms than the original
Hilberts Hotelin the sense that it contains all the rooms of the latter and some morethe
number of its rooms is not numerically larger than the number of rooms in the original hotel.
Above, we stated that two sets are equinumerous, or are numerically the same, if on can nd a
one-one correspondence between them. This is still the case for innite sets, but we have to tread
carefully here. It is not the case for innite sets that a set S is numerically smaller than T (or
S has fewer elements than T) if one can bring S into a one-one correspondence with a proper
subset of T. The problem with this characterization is that it can be the case that S stands in
such a correspondence with a proper subset of T and T stands in such a correspondence with a
proper subset of S.
In order to see this, consider the set O ( N) of odd numbers and the set E ( N) of even numbers.
It is easy to nd a bijective map between O and E, e.g. (where n N):
1, 3, 5, 7, 9, ... 2n + 1...

0, 2, 4, 6, 8, ...2n...
5
This means that O and E are equinumerous. However, it is also rather straightforward to nd a
bijection between O and a proper subset of E, as follows:
1, 3, 5, 7, 9, ... 2n + 1...

2, 4, 6, 8, 10, ...2n + 2...
At the same time, however, we can do the same for E and a proper subset of O:
0, 2, 4, 6, 8, ... 2n...

3, 5, 7, 9, 11, ...2n + 3...
So this would suggest then that on this denition, O would be numerically smaller, equal, and
larger than E! This shows that the denition must be rened for innite sets:
Denition 17. A set S is smaller than another set T, or T is larger than S i the following two
conditions are met:
1. There exists a bijective map between S and a proper subset of T;
2. T cannot be brought into a one-one correspondence with a proper subset of S.
This denition works also for innite sets and reduces to the intuitive characterization given above
for nite sets, since condition (2) is automatically satised for nite sets if condition (1) is.
In order to formulate a theory of the innite, one needs to answer the following question: are
any two arbitrary innite sets necessarily equinumerous, or can they have dierent sizes? Georg
Cantor (1845-1918), who made seminal contributions to set theory, answered this question in his
theory of transnite numbers. This theory met with erce resistance even from some of the greatest
minds of the time, such as Leopold Kronecker and Henri Poincare, and later from Hermann Weyl,
L E J Brouwer, and Ludwig Wittgenstein. Before I give you the answer, or you read on, pause
and do the following
Exercise 7. What do you think, are all innite sets the same size or dierent?
2.3 Cantors heaven
I obviously dont know what your answer to Exercise 7 was, but Cantor rst suspected that any
two innite sets would be of the same size. He spent twelve years trying to prove this when he
found a counterexample. This means that there exist dierent kinds of innities. In fact, as we
will see, there exist innitely many!
Finite sets are all countable in the sense that we can count how many elements they contain.
But this concept can be extended so as to include some innite sets as countable:
Denition 18 (Countable set). A set is called countable just in case it has the same cardinality
as a subset of N. Equivalently, a set S is countable i there exists an injective map from S to N.
6
Exercise 8. Convince yourself that there exist innite sets which are countable. By the way, such
as set is called countably innite or denumerableor simply countable of course. An innite set,
which is not countable, is called nondenumerable.
It is clear from Denition 18 that N is a countably innite set. Thus, Cantors question becomes
whether every innite set is countable. Cantors strategy was to study innite sets which seemed
too large to be denumerable; yet, he found a trick to count them.
Lets start with another thought experiment. Suppose I write a natural number on a piece of
paper and your task is to guess it. It is straightforward to devise a strategy which will succeed in
doing this in a nite number of steps. Lets make your task a little bit harder: this time I write
an integer number on a piece of paper.
Exercise 9. Can you formulate a strategy which guarantees your success in a nite number of
steps? This is essentially the same task as solving Exercise 4 on Hilberts Hotel.
Because this is also possible, it means that N and Z have the same cardinality (which makes Z
denumerable), despite initial appearances that Z might be twice as large as N (modulo 0).
Now the next task is clearly more dicult. Suppose I write two natural numbers on a piece of
paper and your task is to guess both of them at the same step. Only if your answer is the pair of
numbers identical to the one I wrote will the task be solved.
Exercise 10. Do you think there is a strategy which guarantees success in a nite number of
steps? In other words, the question is whether pairs of natural numbers are denumerable, just as
natural or integer numbers are. If so, provide the strategy. If not, why not?
Now lets make it a bit harder yet. Suppose that now you dont only have to guess the two
numbers, but also the order in which they are written down.
Exercise 11. Devise a strategy to solve this in a nite number of steps.
As you get a knack on devising counting strategies, you should be able to answer the following
question:
Exercise 12. Is the set Q
+
of positive rational numbers denumerable?
The answer to this question is part of what incited the strong reactions against Cantors theory
of transnite numbers mentioned above.
Now to another, yet harder, problem. Suppose I write down a nite set of natural numbers. You
do not know either how many numbers I wrote down or which is the largest number among those
on the paper. Do you think there is a strategy to solve this problem in nitely many steps?
Exercise 13. There is, as every nite set of natural numbers is denumerable. But how can we
denumerate the set of nite sets of natural numbers?
This leaves open the question regarding the set of all sets of natural numbersnite and innite.
It was Cantors great discovery to show that this set was nondenumerable. Before we prove this,
let us consider the cardinality of the real numbers R andas a warm-up exerciseprove that R
is nondenumerable.
Exercise 14. Prove that R is nondenumerable. If you havent seen this before, you will probably
not nd the answer. But dont despairit took Cantor a long time to realize this himself and
we will go over this carefully in class! Proving this will introduce you to Cantors ingenious
diagonalization technique.
7
Lets return to the main item of business, to prove Cantors Theorem. First, consider the following
fact:
Fact 1. The set of all sets of natural numbers is nondenumerable. In other words, the power set
P(N) of N is nondenumerable.
Smullyan sketches the proof as follows. Suppose you have a book with countably innitely many
pages, numbered consecutively. In order to show that the above set is not denumerable it suces
to show that there exists a set of natural numbers which cannot be listed on any page of this book.
Exercise 15. Your task now is to describe this set which cannot be given on any page of the book.
Please note that this essentially makes use of the diagonalization technique. [blackboard]
Exercise 16. The set of nite sets of natural numbers, as we have seen, is denumerable. What
about the set of innite sets of natural numbersis it denumerable?
We have thus shown that there is no one-one correspondence between P(N) and N (or, a fortiori,
any subset of N). But can we conclude that P(N) is thus of larger cardinality than N? According
to Denition 17, we also need to show that there exists a bijective map between N and a proper
subset of P(N).
Exercise 17. Show this. Hint: we have actually already found such a map.
The cardinality P(N) is thus larger than that of N. This means that the innity of the number
of sets of natural numbers is larger than the innity of the natural numbers themselves. In other
words, there are innities of dierent sizes, or dierent cardinalities. This immediately leads to
the question of whether there are sets even larger than P(N). As Cantor famously found, the
answer to this question is armative. That P(N) is larger than N is only a special case of his
famous theorem:
Fact 2 (Cantors Theorem). For every set A, the set P(A) of all subsets of A is larger than A.
The proof of the theorem is essentially similar to our answer to Exercise 15. The proof idea is
illustrated by Smullyan as follows. Imagine a universe in which each set of inhabitants forms a
club. The inhabitants decide that each club gets names after an inhabitant of the universe such
that no two clubs bear the name of the same inhabitant and such that each inhabitant lends its
name to exactly one club. To this end, it is not necessary that the inhabitant is a member of the
club which is named after him. It is clear that this project fails in a universe with a nite number
of inhabitants, for n inhabitants, the number of clubs is 2
n
. Fortunately, it turns out that this
universe has innitely many inhabitants, which is why no one sees a reason why the plans couldnt
be realized. However, every scheme that is attempted fails.
Exercise 18. Why is it impossible to nd such a naming scheme, and how does this relate to
Cantors Theorem?
[Proof of Cantors Theorem on the blackboard]. As a corollary, this immediately entails that there
are sets larger than P(N), viz. P(P(N)). In fact, it entails that there are innities of innitely
many diering sizes, as this construction can always be repeated. In other words, there are larger
sets for every set, and hence there are innitely many sizes or cardinalities of innite sets. This
hierarchy of innite cardinal numbers is often referred to as Cantors heaven.
3 Paradoxes and the foundations of set theory
Exercise 19. A warmup exercise for the set-theoretic paradoxes: the quarter and the penny (from
Smullyan, p. 234).
8
3.1 The paradox of the universal set and Russells paradox
Continuing to follow Smullyans presentation, let us see how all of this concerns the foundations
of set theory. In 1897, Cantor discovered a paradox, now often called the paradox of the universal
set. Suppose U is the set of all sets (the universal set), i.e. U := x : x is a set. According to
Cantors Theorem (Fact 2), there is always a larger set, viz. P(U). But could it be that there is a
set larger than the set of all sets? Surely, P(U) must be a subset of U, since U contains all sets.
How can a subset of a set be larger than the set itself? It cant, and thats the paradox.
When Bertrand Russell studied Cantors proof, searching for errors, in 1901, he found what is
now called Russells paradox and is similar to Cantors (and was also found by Zermelo in 1900).
Informally, the idea behind it is as follows. Call those sets which are not elements of themselves
normal. The set of all books surely is normal, since it is itself not a book. A set is abnormal if it
does contain itself. An example would be the set of all things which are not books. Clearly, the
set of all non-books is a not a book and hence an element of itself.
Exercise 20. Now consider the set R of all normal sets. Is R normal or not? Show that it cannot
be determined whether R is normal or abnormal. Hint: Make sure to derive a contraction for all
possibilities.
This is Russells paradox. A more formal version is obtained if we remind ourselves that the
following is an axiom of naive set theory as formulated by Cantor
2
Axiom 1 (Naive comprehension). yx(x y F(x)) for any predicate F in a free variable x.
In other words, for any predicate there exists a set which contains all and only those objects which
exemplify the property denoted by the predicate. Consider the property of not containing itself.
According to the naive comprehension (or existence) Axiom 1, there exists a set containing all and
only objects not containing themselves. If we thus substitute x , x for F(x) and apply existential
and universal instantiation, we get
z z z , z. (1)
This is a contradiction and we have Russells paradox again. Russells paradox is a simplication
of Cantors paradox in that it does not rely on the concept of cardinality.
In 1919, Russell proposed a simple analogue of this paradox in the form of a barber who lives
in a small village, allegedly in Sicily. He says of himself: I shave all and only the men in my
village who do not shave themselves. Thus, the barber doesnt shave any village man who shaves
himself and every man in the village who does not shave himself is shaved by the barber. Does he
shave himself? If he does, he shaves somebody (namely himself) who shaves himself, in violation
of the rule that he shaves nobody who shaves himself. If he doesnt, then he is a village man who
doesnt shave himself and is thus shaved by him, in accordance to the rule that he shaves all men
in the village who dont shave themselves. So we have a contradiction again.
Exercise 21. So, does he shave himself or not? How would you resolve the paradox?
Similarly to the resolution of the barber paradox, one can see that the assumption of Axiom 1
leads to inconsistency, as evidenced by Russells paradox, and must thus be discarded. Axiom 1
also leads to Cantors version of the paradox if we assume that the relevant property is that of
being a set. If we do, the set of all sets also exists. On the one hand, this set can be arbitrarily
large; on the other hand, by Cantors Theorem, for each set there exists a larger set. But this is
the inconsistency again. The error consists in assuming that there is such a set in the rst place.
2
I say remind because we encountered this above, stated informally. I mentioned that it will get us in trouble;
Russells paradox is that trouble.
9
Any solution to these paradoxes must block trouble-making sets such as U and R, while at the
same time allow for suciently rich mathematics, e.g., so as to still allow Cantors heaven. The
rejection of Axiom 1, however, challenged the very foundations of mathematics at the time and
nourished the suspicion that mathematics may not be reconcilable with logic.
3.2 Consequences for the foundations of mathematics
At the time, the most worked-out framework for mathematics was the one by Gottlob Frege.
Freges system was designed to realize his vision of deriving all of mathematics from logic and
set theory. We will have a closer look at the philosophy of his logicism later in the term. For
now, it suces to note that apart from a few axioms of logic, he used only one axiom concerning
set, viz. essentially the (naive) comprehension axiom, Axiom 1. Frege managed to derive all sets
required to do mathematics at the time from this single assumption (plus the logic). The rst
order of business was to get the empty set. One can obtain that by stating a property, which is
not exemplied, such as being dissimilar from oneself or being non-identical with itself. Since
this gives us an admissible predicate, we know by Axiom 1 that there exists the set of all objects
with the property it denotes. But since there are no such objects, the set is empty. Next, take
two arbitrary objects x and y. There is the property of being either identical to x or to y, which
gives us, again via Axiom 1, the set x, y with exactly the elements x and y and none else. This
is still the case if x and y happen to be identical. In this case, we simply have the singleton set
x.
Thus we have the empty set and thus an objectand a constructive rule to produce more
sets. To get the construction going, build the set of exactly those objects which are identical to
. There is only one such object, and we thus have the singleton set , consisting only of the
empty set. Note that and are dierent setsthe latter is a set of one element, the former
has no elements. We can reapply that same step to obtain , the set whose only element is
the set whose only element is the empty set. We continue doing this to get , , ....
We thus obtain an innitude of sets which can serve as natural numbers. In fact, this is how
Ernst Zermelo introduced the natural numbers: we named the empty set 0, 1, 1which
is another name for 2, etc. This procedure gives us all the natural numbers, which
according to Axiom 1 also form a setthe set of natural numbers N.
Given a set S, one can further say that there is the property of being a subset of S. According
to Axiom 1, there consequently exists a set of all subsets of S, i.e., what we called the power set
P(S) of S. Consider the property of being an element of at least one element of S. Hence there
exists a set of all elements of all elements of S, called the set union U(S) of S. For instance, if S
is the set of all clubs, then U(S) is the set of all club members.
So far so good, as it seems as if we can get all the sets which matter in mathematics in this
manner. But there is one problem with all of this: the theory is inconsistent! Sanctioned by
Axiom 1, we can assume the existence of the set of all normal sets, which leads us directly to the
contradiction known as Russells paradox. Similarly for the set of all sets, which leads to Cantors
version of the contradiction. Despite this inconsistency, Freges work was soon recognized (e.g. by
Russell) to contain the seeds of a revolution in the foundations of logic and mathematics. One of
the reasons for this was because the inconsistency could actually be eradicated. One of the ways
to do that (but not the only one) was to ditch Axiom 1 and to replace it with what became known
as the Zermelo-Fraenkel axiomatization of set theory. Before we delve into that, lets briey look
at Russells own solution.
10
3.3 Russells theory of types
The general idea of Russells theory of types was the recognition that sets are something funda-
mentally dierent from elements of sets. Russell constructed a hierarchy of types:
Level 0. individuals a, b, c, ... (dont have set-theoretic elements)
Level 1. sets whose elements live on level 0: a, a, b, ...
Level 2. sets whose elements live on level 0 or 1: a, a, b, c, ...
Level 3. etc.
It is important to insist that only those sets exist, which exist at a level. How does this solve the
paradoxes?
(a) At what level is the universal set U? U U cannot exist on any level and thus doesnt
exist.
(b) At which level do we nd R := x : x is a set x , x? R cannot be an element of itself and
thus doesnt exist either.
While this perfectly solves the paradoxes, the overall system proposed by Russell and Alfred
North Whitehead in their monumental Principia Mathematica (1910, 1912, 1913) was generally
too complicated for many to adopt it, even though many notational innovations remain with
us today. And there was another problem: the theory of types required an axiom of innity,
postulating innitely many individuals at level 0, for otherwise each level consists only of nitely
many objects.
Exercise 22. Why is this a problem for Russells logicistic program? We may have to get back
to this one when we look at logicism.
4 Zermelo-Fraenkel set theory
Before I write down all the axioms in their full glory, lets approach the axiomatization due to
Ernst Zermelo (1871-1953), with later improvements by Adolf (later: Abraham) Fraenkel (1891-
1965), more informally, again following Smullyan. The Zermelo-Fraenkel axiomatization is now
the most widely used axiomatic theory of sets.
4.1 Zermelo set theory
Zermelo started out by replacing the axiom of unrestricted comprehensionour Axiom 1by
what is called the axiom schema of specication (Aussonderungsaxiom) or of restricted com-
prehension, because it does not permit every class as set.
3
As compared to Russells theory of
types, which achieves the elimination of the contradictions by restricting the syntax of admissible
predicates quite signicantly, Zermelos path didnt restrict the syntax, but instead restricted com-
prehension such that it was no longer the case that all classes qualied as sets. Zermelo showed
a quarter century later that his axiom (scheme) of restricted comprehension was a theorem of set
theory as improved by Fraenkel by the introduction of the axiom schema of replacement (Erset-
zungsaxiom) and was thus obsolete as an independent assumption. But we are getting ahead of
ourselves.
Informally stated, Zermelos axiom schema of specication is as follows:
3
This axiom, or axiom schema, actually has a few more names in the literature.
11
Axiom 2 (Axiom schema of specication). Given an arbitrary property as well as an arbitrary
set S, then the set of all elements of the set S which exemplify the property exists.
Because of the need for a prior set S, we can no longer speak of a set of all x with a certain
property. Instead, we can speak of all x S with this property. No contradiction can be derived
from assuming Axiom 2 as far as we know. And it suces for the purposes of mathematical
practice: when mathematicians speak of the set of all numbers, or the set of all points in a
place, etc, they talk about objects and sets whose existence was ordained before.
It should also be clear that this dissolves Russells paradox, as the set of all normal sets cannot
be constructed anymore on the basis of Axiom 2. However, for a given set S, we can create the
set T of all normal elements of the set S (i.e., the set T of all those sets in S which do not contain
themselves). This does not lead to a paradox, since T is notcannot bean element of S, even
though it is a subset of it.
To say that a set S is an element of a set T means that T is a collection of objects, one of which is
S. But to say that S is a subset of T means that all elements of S are also elements of T, but this
does not entail that S itself is one of these elements. Suppose that T is the set of all humans on
Earth and S is the set of human Earthlings who are left-handed. Surely, all left-handed humans
are also humans. Hence S T. But the set of all left-handed humans is not itself a human (even
though all its elements are), and hence S , T.
Exercise 23. Why is the set B of all normal elements of A not itself an element of A, even
though it is certainly a subset of A?
Exercise 24. Show that the universal set U cannot exist under the assumption of Axiom 2. This
resolves Cantors paradox of the universal set.
As a result of giving up Axiom 1, Zermelo could no longer construct the sets needed for math-
ematics as we did in 3.2 and had to introduce specic axioms to guarantee the existence of
, x, y, P(S), and U(S). Similarly, the existence of the natural numbers N had to be underwrit-
ten by a separate axiom, the Axiom of Innity. Zermelo (1908) postulates the following axioms
for set theory:
4
Axiom 3 (Extensionality (Axiom der Bestimmtheit)). If every element of a set M is also an
element of N and vice versa... then M N. Briey, every set is determined by its elements.
Axiom 4 (Elementary sets (Axiom der Elementarmengen)). There exists a set, the null set, ,
that contains no element at all. If a is any object of the domain, there exists a set a containing
a and only a as element. If a and b are any two objects of the domain, there always exists a set
a, b containing as elements a and b but no object x distinct from them both.
Axiom 5 (Separation (Axiom der Aussonderung)). Whenever the propositional function F(x)
is denite for all elements of a set M, M possesses a subset M
F
containing as elements precisely
those elements x of M for which F(x) is true.
This axiom is essentially Axiom 2 above. It is ultimately responsible for the elimination of the
paradoxes as it suces to prove the following important theorem:
Fact 3. Every set M possesses at least one subset M
0
that is not an element of M.
4
Except for a few typographical corrections, all English translations are from Wikipedia (http://en.wikipedia.
org/wiki/Zermelo_set_theory). This translation is not very faithful to the original, but commits no gross mistake
regarding content. The typographical changes have been made both to improve textual accuracy and continuity
with the present text.
12
Proof (as given by Zermelo 1908, 265). For each element x of M, it is determinate whether x x
or not; this possibility x x is in itself not excluded by our axioms. If M
0
is the subset of M
which contains, by [Axiom 5], all those elements of M for which x , x, then M
0
cannot be an
element of M. For either M
0
M
0
or not. In the rst case M
0
contained an element x = M
0
, for
which x x, against the denition of M
0
. Hence, it is certainly not the case that M
0
M
0
, and if
M
0
were an element of M, it would have to also be an element of M
0
, which was just excluded.
This means that the reductio assumption M
0
M is false, proving the theorem.
Zermelo concludes from Fact 3 that not all objects x in the universal domain B can be elements
of one and the same set, which means, in other words, that the domain B itself is not a set. This
precludes the universal set and disposes the Russellian antinomy as far as we are concerned.
(ibid.)
Axiom 6 (Power set (Axiom der Potenzmenge)). To every set T there corresponds a set P(T),
the power set of T, that contains as elements precisely all subsets of T.
Axiom 7 (Union (Axiom der Vereinigung)). To every set T there corresponds a set U(T), the
union of T, that contains as elements precisely all elements of the elements of T.
Axiom 8 (Choice (Axiom der Auswahl)). If T is a set whose elements all are sets that are
dierent from and mutually disjoint, its union U(T) includes at least one subset S
1
having one
and only one element in common with each element of T.
This is the famous an controversial Axiom of Choice, which Zermelo considers an unobjectionable
logical principle. Zermelo states that one can also express this axiom by saying that it is always
possible to choose a particular element m, n, r, ... from each element M, N, R, ... of T and to collect
them to a set S
1
.
Axiom 9 (Innity (Axiom des Unendlichen)). There exists in the domain at least one set Z
that contains the null set as an element and is so constituted that to each of its elements a there
corresponds a further element of the form a, in other words, that with each of its elements a it
also contains the corresponding set a as element.
Axioms 3-9 constitute the axiomatic basis of Zermelo set theory, or the theory ZC.
4.2 The well-ordering theorem and the Banach-Tarski paradox
Assuming the Axiom of Choice (Axiom 8), Zermelo was able to prove the so-called well-ordering
theorem (also called Zermelos theorem), which asserts that every set can be well-ordered.
Denition 19. A set S is well-ordered by a strict total order just in case every non-empty subset
of S has a least element under the ordering.
Orders (total, strict total, partial, etc) are binary relations dened on some set, the domain.
Denition 20 (Total order). A set S is totally ordered under a binary relation, here denoted by
the inx , just in case for all a, b, c S, the following three conditions hold:
1. If a b and b a, then a = b (antisymmetry);
2. If a b and b c, then a c (transitivity);
3. a b or b a (totality).
Exercise 25. Show that totality in Denition 20 entails reexivity, i.e., a S, a a.
13
Denition 21 (Partial order). A set S is partially ordered under a binary relation, denoted by
the inx , just in case for all a, b, c S, the following three conditions hold:
1. If a b and b a, then a = b (antisymmetry);
2. If a b and b c, then a c (transitivity);
3. a a (reexivity).
Note that because of the result established in Exercise 25 any totally ordered set is also partially
ordered, but not vice versa. We can now dene the remaining two undened concepts used in
Denition 19:
Denition 22. The least element of a subset S of a partially ordered set T is an element of S
which is smaller than or equal to any other element of S, where smaller than and equal to is
dened in terms of the partial order of T. The greatest element is dened dually.
Denition 23 (Strict total order). For every total order , there exists an associated asymmetric
relation, denoted by the inx < and called a strict total order, which can be dened in two equivalent
ways:
Either a < b i a b and a ,= b;
or a < b i not b a (making < the inverse of the complement of ).
Exercise 26. An asymmetric relation R is a binary relation dened on a domain S such that
(a, b S, Rab Rba). Show that the asymmetry of a relation entails that it is irreexive, i.e.
a S, Raa.
It turns out that the well-ordering theorem is equivalent to the Axiom of Choice, as is Zorns
lemma, in the sense that in rst-order logic either one of the two conjoined with the (remaining)
axioms of Zermelo-Fraenkel set theory (see 4.3 below) entails the other.
5
Fact 4 (Well-ordering theorem (Zermelos theorem)). For every set S, there exists a well-ordering
with domain S.
For the proof (which is not very hard), cf. e.g. Halmos 1960.
One of the most intriguing paradoxes of modern mathematics arises as a consequence of the well-
ordering theorem: the Banach-Tarski paradox. This paradox was rst stated by Stefan Banach
and Alfred Tarski (1924), who showed how it is possible to cut a solid 3-dimensional ball into
nitely many pieces and to reassemble the pieces into two solid balls of the size as the original
ball. The proof of this crucially depends on assuming the Axiom of Choice (Axiom 8) or the
well-ordering theorem. Please read the assigned article by Robert French if youre interested. You
are also invited to watch this video for an illustration of how to use the Banach-Tarski paradox to
multiply oranges: http://www.youtube.com/watch?v=uFvokQUHh08.
4.3 Zermelo-Fraenkel set theory
For most purposes, the theory ZC introduced in 4.1 suces to satisfy the demands on a set theory.
However, it does not allow the construction of ordinal numbers (cf. Moore 1990) and Axiom 5
(Separation) insists on denite functions (or formulae) without specifying what is meant by
5
This is not true in second-order logic, where the well-ordering theorem is strictly stronger than the Axiom of
Choice.
14
that. Fraenkel (and Thoralf Skolem) thus proposed a slightly modied theory, adding an axiom
scheme of replacement and an axiom of regularity, while freeing some of the other axioms from
redundancy. This resulted in what is today the most commonly used set of axioms for a set theory,
the Zermelo-Fraenkel axioms.
Starting out from an informal notion of classes, which look just as sets naively construed prior to
the advent of axiomatic set theories. All sets as characterized by the axioms to follow are classes,
but there are some classes which are not sets (so-called proper classes), most notoriously the
universal class V = x : x = x. So like most authors, lets presuppose a non-empty domain
of discourse as part of our semantics of the rst-order logic in which the following set theory is
axiomatized. This assumption is harmless, and certainly underwritten by Axiom 15 (Innity)
below, which entails the existence of a set.
In the formal statements of the axioms (ignore them if you want), we will use the notion of formu-
lae, which are constructed from the atomic formulae a b and a = b by means of the usual logical
connectives (conjunction), (disjunction), (negation), (implication),
(equivalence) and quantiers x (universal) and x (existential). This means that fundamen-
tally, we introduce and = as the only non-logical symbols. If a formula has free variables,
we often write (x
1
, ..., x
n
) and mean that the free variables are among the x
i
s. Anyway, if you
want to ignore the formal statement of the axioms below and focus on the informal statements
which always precede the formal ones, I think you can safely forget about everything I said in this
paragraph.
Here are the Zermelo-Fraenkel axioms, given both informally and formally:
6
Axiom 10 (Extensionality). If S and T have the same elements, then S = T. In other words,
ST[x(x S x T) S = T].
It should be noted that the converse, viz. if S = T, then x(x S x T), is an axiom of
predicate calculus. The axiom simply expressed the idea that a set is determined by its extension,
i.e. by its elements.
Axiom 11 (Pairing). For any a and b, there exists a set a, b that contains exactly a and b. In
other words,
abzx(x z x = a x = b),
or, more simply, there exists a set z whose only members are a and b (which themselves may be
sets).
Given Axiom 10 (Extensionality), the set z is guaranteed to be unique, which justies our notation
a, b. You may be surprised that we start by stating an axiom about sets of two members, rather
than just one. But really, we have just included those sets: the singleton a is dened as the set
a := a, a. At this point, we can also dene an ordered pair a, b) of elements or sets a and b
by
a, b) := a, a, b.
Exercise 27. Show that this denition satises the intuitive requirement on ordered pairs that
a, b) = c, d) i a = c b = d.
Axiom 12 (Schema of Separation). If P is a property (with parameter p), then for any S and p,
there exists a set T = x S : P(x, p) that contains all those x S which exemplify property P.
In other words, for a formula (x, p), and for any S and p, there exists a set T = x S; (x, p):
SpTx(x T x S (x, p)).
6
I consulted various sources for this, in an attempt to get a list as standard as possible. Among them are Hajnal
and Hamburger 1999, Jech 2003, Tiles 1989, and http://en.wikipedia.org/wiki/Zermelo-Fraenkel_set_theory.
15
This is called an axiom schema rather than just an axiom because it is strictly speaking a
distinct axiom for each formula (x, p). Again by Axiom 10 (Extensionality), the set T is unique.
As a consequence of Axiom 12 (Schema of Separation), we can introduce set intersection and
dierence, as follows:
S T := x S : x T and S T := x S : x , T,
respectively. Compare this with Denitions 7 and 8 above. Moreover, it follows that the empty
class = x : x ,= x is a setour beloved empty set as originally introduced in Denition 5. The
empty set exists and is unique just in case at least one set exists. But existence of at least one set
is guaranteed by Axiom 15 (Innity).
An important consequence of Axiom 12 (Schema of Separation) is that the universal class V as
dened above is a proper class (and hence not a set), for otherwise, S = x V : x , x would
also be a set.
Axiom 13 (Union). For any S, there exists a set T = S, the union of all elements of S. In
other words, for any S, there exists a set T = S:
STx[x T z(z S x z)].
As a result of this axiom, for every S, there exists a unique set
T = x : z S(x z) =: z : z S = S,
interpreted as the unique set T whose elements are exactly the elements of the elements of a
given set S. Jointly with Axiom 11, this axiom entails that for any two sets S, T, there is a
unique set S T := S, T (and generalize this to any number of sets). A corresponding axiom
of intersection is quite unnecessary, as we were able to dene it above, simply using Axiom 12
(Schema of Separation).
Axiom 14 (Power Set). For any S, there exists a set T = P(S), the set of all subsets of S. In
other words, for any S, there exists a set T = P(S):
STx(x T x S).
At this point of setting up the axioms, we can now introduce quite a bit of useful stu. For
instance, one can use the concept of power sets to dene the Cartesian product of two sets, dene
n-ary relations, elds, functions, etc. I am not going to do this, except to say how to dene
the Cartesian product and if you are interested to see how this can be used to dene relations
and functions, please see my handout on structure, available here: http://philosophyfaculty.
ucsd.edu/faculty/wuthrich/teaching/2010_246/246HandoutStructureMath_2010.pdf.
Denition 24 (Cartesian product). The Cartesian product S T of two sets S and T is the set
of all ordered pairs such that the rst element is an element of S, and the second one of T:
S T := x, y) : x S y T.
Strictly speaking, we have not yet established that S T is a set and that it exists. Noting that
x, y S T and x, x, y P(S T), we recognize that
x, y) = x, x, y P(P(S T)).
This now allows us to recognize that the Cartesian product of two sets is a set itself:
S T P(P(S T)).
16
Axiom 15 (Innity). There exists an innite set.
We will not state this more formally. Although possible, it would involve introducing new notions
we will not otherwise need. Note that this is the only axiom making an existential assertion;
courtesy of this axiom, we are guaranteed that there are sets.
Axiom 16 (Schema of Replacement). If a class F is a function, then for any S, there exists a
set T = F(S) = F(x) : x S. In other words, for each formula (x, y, p), the following formula
is an Axiom (of Replacement):
xyz[(x, y, p) (x, z, p) y = z] STy[y T (x S)(x, y, p)].
Dont worry if the formal statement becomes too unwieldy for your taste. But just so you know,
there are more unwieldy formulations out there! Generally, the exact formulation of the axiom
statements varies from source to source. This is not normally a problem, since dierent for-
mulations can be equivalent. One dierent but equivalent formulation of Axiom 16 (Schema of
Replacement) is that if a class F is a function and dom(F) is a set, then ran(F) is also a set.
7
Axiom 17 (Regularity). Every non-empty set S contains an element T which is disjoint from S.
In other words,
S[T(T S) T(T S X(X S X T))].
There is an important consequence of this axiom: no set is an element of itself. Suppose we have
a set S and apply Axiom 17 (Regularity) to S, which is the singleton set of S whose existence
is guaranteed by Axiom 11 (Pairing). By Axiom 17 (Regularity), there must be an element of
S which is disjoint from S. But the only element of S is S, of course, and thus it must be
that S is disjoint from S. Since S S, it follows that S , S, by the denition of disjoint
(Denition 12).
Axiom 17 (Regularity) plays a lesser role in mathematics, as most results still obtain without it.
It does play a role in establishing well-ordering results and results concerning ordinal numbers.
Axiom 18 (Choice). Every family of non-empty sets has a choice function. In other words, for
a set S of non-empty elements, there exists a function f from S to the union of all elements of
Sthe choice functionsuch that for all T S, f(T) T.
This statement of the axiom of choice (AC) is equivalent to the one given in Axiom 8 (Choice),
where each choice function is replaced by its range. For nite sets, Axiom 18 (Choice) is quite
unnecessary, as it follows as a theorem from Axioms 1017. This is not the case for innite sets.
There are a dozen or so statements used across mathematics which are demonstrably equivalent
to the AC. The rst one to note is Fact 4, the well-ordering or Zermelos theorem. There are other
statements in set theory, such as Tarskis and Konigs theorems, which are equivalent to the AC.
Apart from statements in abstract algebra, functional analysis, topology, and mathematical logic,
which are equivalent to the AC, a famous lemma from order theory is as well:
Fact 5 (Zorns lemma). Every non-empty partially ordered set in which every totally ordered
subset has an upper bound contains at least one maximal element, i.e. an element that is not
smaller (as dened by the ordering relation) than any other element of the subset.
Note that I have labelled Zorns lemma a fact, which of course it is given the AC. For a list of
statements equivalent to the AC, consult http://en.wikipedia.org/wiki/Axiom_of_choice.
7
I am sure most of you have encountered the terms domain and range of a function. What you may not know is
that these concepts can be generalized to relations, such that for instance for a binary relation R, the domain of R is
dened as the set dom(R) := {x : yx, y R} and the range of R is dened as the set ran(R) := {y : xx, y R}.
17
The theory resulting from Axioms 1018 is denoted ZFC or ZF, depending on whether Axiom
18 (Choice) is also assumed or not. For some time, it was an open question not only whether ZFC
was consistent, but also whether the AC was indeed independent of Axioms 1017 in the sense
that neither it nor its negation could be derived from the other axioms. Assuming the consistency
of ZF, Kurt Godel and Paul Cohen settled this question, Godel proving that the negation of the
AC is not a theorem of Axioms 1017 and that hence ZFC is consistent, and Cohen showing that
the AC is not theorem of Axioms 1017 by establishing that ZFC, i.e. ZF conjoined with the
negation of the AC, is consistent.
5 The continuum hypothesis
One important question remains: given that the set P(N) of all subsets of natural numbers is
larger than the set N of natural numbers itself, does there exist a set S such that it is larger than
N, but smaller than P(N)? To ask dierently, is there a set whose cardinality lies between that of
N and P(N), or is P(N) the next largest set after N?
Cantor asked the same question, but the answer is not known to this day. He conjectured that P(N)
is indeed the next largest set after N, and this conjecture is known as the continuum hypothesis
(CH). The reason for calling this conjecture the continuum hypothesis is because one can bring the
set P(N) into a one-one correspondence to the set of points on a innite line, which is sometimes
called the continuum. Thus, P(N) has the same size as the continuum. The question then is
whether there exist sets whose size is larger than N but smaller than the continuum. More
generally, Cantor also hypothesized that there does not exist, for any innite set S, a set whose
size lies between that of S and that of P(S). This is called the generalized continuum hypothesis,
or simply GCH. Some people, including Smullyan (who again served as basis for the informal part
of this section), believe that to proof the GCH or its negation is the biggest outstanding problem
in all of mathematics.
What the status of the CH is depends on your larger point of view about the status and function
of mathematical theories. Cantor himself believed that the CH was true and expended signicant
intellectual energies to prove it. Formalists do not consider it to be true or false simpliciter, but
make it a question of which axiomatic system is used. In some of these, the CH is true, while
in others, it is false. Both the conjunction ZFCGCH of ZFC and the GCH and the conjunction
ZFCGCH of ZFC and the negation of the GCH are consistentassuming that ZFC itself is
consistent, which is what most mathematicians believe. This is because the GCH can be shown
to be independent of ZFC. This independence was established, again, by Godel and Cohen. Godel
showed in 1940 that the GCH cannot be disproved in ZFC, and Cohen showed in 1963 that the
GCH cannot be proved in ZFC either. Together, these results mean that the GCH is independent
of ZFC, similarly to the infamous Fifth Axioms independence of the rest of Euclids axioms. One
thing to note, however, is that ZFGCH entails the AC. This means that, while both independent
of ZF, the GCH is a strictly stronger claim than the AC because, to repeat, the entailment does
not hold in the other direction.
On the other hand, realists/platonists insist that the CH is either true or falsei.e., there is a
fact of the matter whether or not the CH is trueand that we simply dont know which. The
reason for our ignorance, they continue, is not because the question is in principle unanswerable,
but simply because we dont yet know enough about sets. The platonist Godel predicted that the
correct or true axiomatic system for set theory, once found, will show that the CH (and even the
GCH) is false.
In order to just make a connection to what we could go on to study in more detail, let me state the
CH more concisely and more formally. The cardinality of N is usually denoted
0
(aleph-naught).
Since it can be shown that the cardinality of R is the same as that of P(N) (although I dont
18
think we did this here, cf. Jech (2003, 37) or the appendix Basic set theory to Jech (2002)), it is
2
0
. The AC guarantees that there exists a smallest cardinal number
1
larger than
0
, the CH
can be expressed very compactly:
Hypothesis 1 (Continuum).
2
0
=
1
.
This can be generalized for any ordinal :
Hypothesis 2 (Generalized continuum).
2

=
+1
.
At this point, I could go on about transnite mathematics, ordinals and cardinals, and the won-
derful world of the alephs, the Lowenheim-Skolem theorem and a theorem due to Godel, but I will
do so only if there is interest. In this case, we will read material from Chapters 10-12 of Moore
1990.
References
[1] Stefan Banach and Alfred Tarski, Sur la decomposition des ensembles de points en parties
respectivement congruentes, Fundamenta Mathematicae 6 (1924): 244-277.
[2] John Earman, Handout Sets for his class Paradoxes, Fall 2001, University of Pittsburgh.
[3] Herbert B Enderton, Elements of Set Theory, Academic Press (1977).
[4] Robert M French, The Banach-Tarski Theorem, The Mathematical Intelligencer 10 (1988):
21-28.
[5] Kurt Godel, The Consistency of the Axiom of Choice and of the Generalized Continuum
Hypothesis with the Axioms of Set Theory, Princeton University Press (1940).
[6] Andras Hajnal and Peter Hamburger, Set Theory, Cambridge University Press (1999).
[7] Paul Halmos, Naive Set Theory, Litton Educational (1960).
[8] Karel Hrbacek and Thomas Jech, Introduction to Set Theory, Marcel Dekker (
2
1984).
[9] Thomas Jech, Set Theory, Springer (2003).
[10] Thomas Jech, Set Theory, in Edward N. Zalta (ed.), The Stanford Online Encyclopedia of
Philosophy, URL http://plato.stanford.edu/entries/set-theory/ (2002).
[11] David Lewis, Parts of Classes, Blackwell (1991).
[12] A W Moore, The Innite, Routledge (1990).
[13] Michael Potter, Set Theory and Its Philosophy, Oxford University Press (2004).
[14] Raymond Smullyan. Satan, Cantor, and Innity and Other Mind-Boggling Puzzles, Knopf
(1992).
[15] Robert R Stoll, Set Theory and Logic, W H Freeman and Company (1963).
[16] Patrick Suppes, Axiomatic Set Theory, Dover (1972).
19
[17] Mary Tiles, The Philosophy of Set Theory: An Historical Introduction to Cantors Paradise,
Dover (1989).
[18] Ernst Zermelo, Untersuchungen uber die Grundlagen der Mengenlehre I, Mathematische
Annalen 65 (1908): 261-281.
Additional resources used:
Planet math entries (http://planetmath.org/):
http://planetmath.org/encyclopedia/SetTheory.html
http://planetmath.org/encyclopedia/ZermeloFraenkelAxioms.html
Wikipedia entries:
http://en.wikipedia.org/wiki/Set_(mathematics)
http://en.wikipedia.org/wiki/Russells_paradox
http://en.wikipedia.org/wiki/Zermelo_set_theory
http://en.wikipedia.org/wiki/Well-ordering_theorem
http://en.wikipedia.org/wiki/Banach-Tarski_paradox
http://en.wikipedia.org/wiki/Zermelo-Fraenkel_set_theory
http://en.wikipedia.org/wiki/Axiom_of_choice
Wolfram MathWorld entries (http://mathworld.wolfram.com/):
http://mathworld.wolfram.com/Zermelo-FraenkelAxioms.html
20