Frank Roumen WB 2010 PDF
Frank Roumen WB 2010 PDF
Frank Roumen WB 2010 PDF
Natuurwetenschappen
Bacheloronderzoek Wiskunde
Juli 2010
In this thesis, we will consider two models of set theory and look at consequences
of these models in linear algebra. The first model satisfies the Axiom of Choice;
we will show that this is equivalent to existence of bases for all vector spaces.
We will also prove that countability of a vector space is sufficient for proving
existence of bases without the Axiom of Choice. The second model will be
constructed using the forcing technique. It contains an infinite-dimensional
vector space having only finite-dimensional subspaces, which implies that this
vector space has no basis.
iii
Contents
Introduction vii
3 Forcing 13
3.1 Relativization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Idea of the construction . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 Generic extensions . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4 Symmetric extensions . . . . . . . . . . . . . . . . . . . . . . . . 17
3.5 The forcing relation . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.6 Choice of parameters . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.7 Linear algebra in N . . . . . . . . . . . . . . . . . . . . . . . . . 22
Conclusion 25
A Axioms of ZF 27
v
Introduction
vii
Chapter 1
This chapter contains some preliminaries for the rest of this thesis. In Sec-
tions 1.1 and 1.2 we discuss some set theory, especially the concept of models
and equivalents of the Axiom of Choice. Section 1.3 contains an introduction
to order theory. The main reference for Section 1.1 is [2]. Section 1.2 is based
on [5] and the order theory in Section 1.3 comes from [4, 8].
Axiom of Extensionality. Two sets are equal if they contain the same ele-
ments. Formally,
∀x∀y (∀z (z ∈ x ↔ z ∈ y) → x = y) .
The reverse implication is always true, so this axiom could also be stated as
“Two sets are equal if and only if they contain the same elements”.
Axiom of Pairing. If x and y are sets, then there exists a set containing x
and y and nothing else. This set is denoted {x, y}. The formal version of this
axiom is
∀x∀y∃z∀w (w ∈ z ↔ (w = x ∨ w = y)) .
1
2 1. SETS AND ORDER
There exist non-transitive models of ZF, but as transitive models are easier
to work with, we will require our models to be transitive.
The function f is called a choice function. Intuitively the axiom states that,
given a collection of non-empty sets, it is possible to choose one element of every
set in the collection. The function f assigns to every set an element of that set.
To give a very simple example, let
f: {1} 7→ 1
{1, 2} 7 → 2
{1, 2, 3} 7 → 1
Of course, there are also other possible choice functions. The Axiom of Choice
is actually not necessary in this special case, as the existence of f can be derived
from the other axioms. More generally, if X is finite, then there exists a choice
function for X which can be constructed without using Choice. The Axiom of
Choice is only indispensable for some infinite sets.
The Axiom of Choice seems intuitively obvious. Nonetheless some math-
ematicians have objections against the axiom because of its non-constructive
nature and bizarre consequences. The axiom is non-constructive, because it
asserts the existence of a choice function without providing explicit means to
construct such a function. This is a reason why the Axiom of Choice differs in
1.3. PARTIAL ORDERS 3
nature from the other axioms of ZF. Besides being non-constructive, the con-
sequences of the Axiom of Choice can be quite strange. For instance, Choice
implies that it is possible to decompose a sphere into four pieces, which can be
translated and rotated to form two copies of the original sphere. This decom-
position is known as the Banach–Tarski paradox.
Accepting the Axiom of Choice yields paradoxical results, but it has many
desirable consequences as well. In this thesis, we will first look at consequences of
accepting Choice, and thereafter we look at what might happen upon rejection.
In other words, we will look at models of ZF with and without the Axiom
of Choice. The theory consisting of all axioms of ZF together with Choice is
denoted as ZFC.
In its original formulation, the Axiom of Choice can be difficult to apply.
We will therefore state some equivalent formulations of the Axiom of Choice.
The proofs can be found in [5]. The first equivalent statement is the Axiom of
Multiple Choice. The standard Axiom of Choice asserts that it is possible to
select a single element of each set in a collection, or, equivalently, a one-element
subset of each set. In the Axiom of Multiple Choice, this one-element subset is
replaced by a finite subset.
Trivially the Axiom of Choice implies the Axiom of Multiple Choice, but
both axioms are in fact equivalent.
Many equivalents of the Axiom of Choice are phrased in the language of
partial orders. Furthermore, the technique of forcing discussed in Chapter 3 also
heavily relies on the notion of order. Therefore we will give a short introduction
to partially ordered sets in the next section, before discussing other equivalents
of the Axiom of Choice.
So ↑S is the set of all elements in P above some element in S, and is called the
filter generated by S. For y ∈ P we define
↑y = ↑{y} = {x ∈ P | y ≤ x}.
Examples.
1. Let X be an arbitrary set and let PX = {S | S ⊆ X} be the power set of
X, i.e. the set of all subsets of X. Then X can be ordered by inclusion.
PX is totally ordered if and only if X is empty or X is a singleton. The
only maximal element is X. As an example of a partial order with multiple
maximal elements, consider PX \ {X} ordered by inclusion. For this set,
X \ {x} is maximal for each x ∈ X.
or, equivalently,
f ≤ g ⇔ graph(f ) ⊇ graph(g).
The maximal element of this P is the empty function: the function whose
domain is the empty set. The minimal elements are the total functions.
A slight variation on this example will be important later on. Consider
the set P of finite partial functions, which are partial functions with a
finite domain. This set contains no minimal elements. If x ∈ X, then the
set Dx = {f ∈ P | x ∈ dom(f )} is dense in P , since every function can be
extended to a function with x in its domain.
Chapter 2
The set of true facts depends on the particular model in which one is working.
Throughout this chapter, we will work in a model of set theory that satisfies the
Axiom of Choice and look at the consequences in linear algebra. In particular,
we will define the concept of vector space basis for infinite-dimensional vector
spaces and prove that every vector space has a basis assuming the Axiom of
Choice. Furthermore it can be useful to know when this axiom can be avoided
in the proof. If this is the case, then the theorem might be true in other models
of set theory as well. We will show that Choice cannot be avoided in general,
but also provide a condition under which vector spaces still have bases.
The material in Section 2.1 is well-known, see for example [3]. Section 2.2 is
taken from [5].
v = v 1 b1 + · · · + v n bn
for certain n ∈ N, vi ∈ K, bi ∈ B.
7
8 2. LINEAR ALGEBRA IN ZFC
f1 + f2 + · · · + fn
g1 + g2 + · · · + gm
for certain monomials f1 , f2 , . . . , fn and g1 , g2 , . . . , gm . In the proof that ex-
istence of bases implies the Axiom of Choice, we will use a field consisting of
rational functions of this form where all fk and gk have the same i-degree for
all i. The next definition will simplify the terminology a little.
Definition 2.7. A rational function
f1 + f2 + · · · + fn
∈ k(X)
g1 + g2 + · · · + gm
is called i-homogeneous of degree 0 if all fk and gk have the same i-degree.
Example 2.8. Let X1 and X2 be as in Example 2.6. The function
Then K is a subfield of k(X), which implies that k(X) can be viewed as a vector
space over K. By hypothesis, the vector space k(X) over K has a basis B.
Fix i ∈ I. Since B is a basis, any monomial x ∈ Xi can be expressed as a
linear combination
x = x 1 b1 + x 2 b2 + · · · + x n bn
where xj ∈ K, bj ∈ B. Now we will show that xj /x does not depend on the
choice of the monomial x, i.e. x, y ∈ Xi implies xj /x = yj /y.
Let x, y ∈ Xi and write
x = x1 b1 + x2 b2 + · · · + xn bn ;
2.2. NECESSITY OF THE AXIOM OF CHOICE 11
y = y1 c1 + y2 c2 + · · · + ym cm .
We can rewrite y as
y y
y= x = (x1 b1 + x2 b2 + · · · + xn bn ) .
x x
Since expressions in terms of basis vectors are unique and xy ∈ K, it follows that
n = m, bj = cj , and yj = xy xj for each j ∈ {1, . . . , n}. Thus xj /x = yj /y. Since
xj /x depends only on i and j, we can call it αij .
Since xj ∈ K, the rational function xj is by definition i-homogeneous of
degree 0 for all i. Therefore, for all j ∈ {1, . . . , n} the denominator of αij
must contain at least one variable in Xi . Define Fi to be the set of variables
in the denominator of αij for some j ∈ {1, . . . , n}. Then each Fi is finite and
∅ 6= Fi ⊆ Xi , so the Axiom of Multiple Choice holds, whence the Axiom of
Choice is true.
Chapter 3
Forcing
3.1 Relativization
Distinct models of set theory may have different properties. If M is any model
of ZF, then of course all axioms of ZF are true in M , as well as all consequences
of these axioms. But as there are propositions not implied by the axioms, the
choice of M influences the set of true formulas in M . First we will take a closer
look at what it means for a formula to be true in a model. For each formula in
terms of the symbols of set theory, there is a corresponding formula expressing
properties of the model.
13
14 3. FORCING
∃x∃y∃z (y ∈ x ∧ z ∈ x ∧ y 6= z) .
This is a formal way of expressing that a set containing at least two elements
exists, so ϕ is true. ϕM can be written as
This means, informally, that M contains a set consisting of at least two elements.
But since all sets in M contain at most one element, ϕM is false.
Now let ψ be the statement
∃x∀y (y ∈
/ x) .
Then both ψ, asserting the existence of an empty set, and ψ M , claiming that
M contains a set possessing no elements of M , are true.
If M is a model of ZF, then determining the truth of a relativized proposition
is less troublesome: the truth value of ϕM equals the truth value of ϕ whenever
ϕ has a truth value in ZF. If ϕ is independent from the axioms, then it might
be either true or false in M .
Definition 3.3. Let M be a set and ϕ a sentence. We write M |= ϕ, and say
“ϕ is true in M ” or “M is a model for ϕ”, if ϕM is true.
This definition can be extended to sets S of sentences: if M |= ϕ for all
sentences ϕ ∈ S, then this will be written as M |= S. As an example, note that
the phrase “M is a model for ZF” can now be expressed formally as M |= ZF.
of R and hence M is not countable. What in fact happens here is that the set
R in M is different from the “real” R. To avoid ambiguity and confusion, we
will employ the symbols R and RM . Because M is countable, RM must also
be countable, i.e. there exists a bijection f : N → RM . But this bijection f
is not an element of M , since M |= (RM is uncountable). In other words, RM
is countable when considered as a set in the entire universe, but uncountable
when considered as a set in M .
After the resolution of this apparent paradox, the actual construction can
begin. If M already satisfies the desired properties, there is no need to continue,
so suppose that this is not the case. Then we seek for a set G ∈ / M and
extend M to a larger model M [G] containing G. Unfortunately, this model
will satisfy the Axiom of Choice and consequently cannot contain V . We will
proceed by removing some sets from M [G], obtaining a new model N for which
M ⊆ N ⊆ M [G]. Among the removed sets will be many choice functions, well
orders and infinite-dimensional subspaces of vector spaces, so when carried out
correctly N will be the right model.
The construction involves a number of parameters, among which is the set
G. The choice of parameters will influence the true propositions in the models
M [G] and N . We will first describe how to construct M [G] and N in general.
Thereafter a strategic choice of parameters will yield the desired model.
{< {< ∅, p >}, q >, < {< ∅, p >, < ∅, q >}, p >}
et cetera.
Definition 3.7. Given a P -name τ and a subset G ⊆ P , define
This is again a recursive definition. Note that G need not be a generic filter
in this definition.
16 3. FORCING
π P ({< {< ∅, p >}, q >, < {< ∅, p >, < ∅, q >}, p >})
= {< {< ∅, π(p) >}, π(q) >, < {< ∅, π(p) >, < ∅, π(q) >}, π(p)}.
π P (x̌) = π P ({< y̌, 1 > | y ∈ x}) = {< π P (y̌), π(1) >} = {< y̌, 1 > | y ∈ x} = x̌.
• symG (τ ) = {π ∈ G | π P (τ ) = τ }
Theorem 3.20. Let a partially ordered set P with maximal element 1, a generic
filter G ⊆ P , a group G of automorphisms of P and a normal filter F on G be
given. Then N is a c.t.m. of ZF for which M ⊆ N ⊆ M [G].
Proof. Again we will only sketch the proof and refer to [7] for the details.
For the inclusion M ⊆ N , if x ∈ M , then val(x̌, G) = x. From Lemma 3.16
it follows that x̌ is heriditarily symmetric, thus x ∈ N . The inclusion N ⊆ M [G]
follows from the fact that every heriditarily symmetric name is a P -name.
Countability of N follows from the inclusions M ⊆ N ⊆ M [G] and Theo-
rem 3.12.
If x ∈ y ∈ N , then y = val(τ, G) for some heriditarily symmetric P -name
τ . It follows that there is a P -name σ for which x = val(σ, G). σ is the first
component of some ordered pair in τ , so σ is heriditarily symmetric, hence
x ∈ N . This establishes transitivity.
Transitivity implies that N satisfies the Axiom of Extensionality. To show
that Pairing holds, let τ and σ be heriditarily symmetric P -names, so {π ∈
G | π P (τ ) = τ } ∈ F and {π ∈ G | π P (σ) = σ} ∈ F. Since F is a normal filter,
{π ∈ G | π P (τ ) = τ and π P (σ) = σ} ∈ F.
Hence {< τ, 1 >, < σ, 1 >} is a heriditarily symmetric P -name. We omit the
proofs of the other axioms.
The next result is one of the main theorems of the theory of forcing. It
expresses an easy relationship between the forcing relation and truth and is
often used to switch back and forth between |= and . The proof can be found
in [8].
3. M [π −1 (G)] = M [G]
Proof.
4. Follows from 3, since equal generic models yield equal symmetric submod-
els.
By 2 and 4,
Therefore
π(p) ϕ(val(π P (τ1 ), G), . . . , val(π P (τn ), G))
Partially ordered set. Let P be the set of all finite partial functions from
N × PN × PN to {0, 1}. That is, a function f is in P if its domain dom(f ) is
finite, dom(f ) ⊆ N × PN × PN and im(f ) ⊆ {0, 1}. Order this set by reverse
inclusion, i.e.
The maximal element of this P is the empty function: the function whose
domain is the empty set.
Furthermore, let
A = {{xnS | S ∈ PN} | n ∈ N} .
These sets have names
Let
S0 = {n ∈ N | p {ξnS | S ∈ PN} ∈ β}
and choose n, m ∈ N such that
1. (n, S), (m, S) ∈
/ E for all S ∈ PN;
2. (n, S, T ), (m, S, T ) ∈
/ dom(p) for all S, T ∈ PN;
3. n ∈ S0 ;
4. m ∈ N \ S0 .
3.7. LINEAR ALGEBRA IN N 23
It is always possible to find such m and n, because E and dom(p) are finite
whilst S0 and N \ S0 are infinite.
Define a permutation π : N × PN → N × PN by π(n, S) = (m, S), π(m, S) =
(n, S) and π(k, S) = (k, S) for all S ∈ PN, k 6= n, m. This permutation is an
element of fixG (E) ⊆ symG (β), so π P (β) = β. We claim that π P (ξnS ) = ξmS .
This follows from the next calculation.
p {ξnS | S ∈ PN} ∈ β.
Therefore
π∗ p {ξmS | S ∈ PN} ∈ β.
By definition of π, we have dom(p) = dom(π ∗ p), so 2 implies p = π ∗ p. Hence
N |= {xmS | S ∈ PN} ∈ B,
contradicting 4.
1. X is infinite-dimensional.
3. X has no basis.
4. The only subspaces of X that have a linear complement are {0} and X
itself.
24 3. FORCING
Proof.
In this thesis, we have defined the concept of a model of set theory and did
linear algebra in several models. The Axiom of Choice played an important role
in deciding which statements are true. If this axiom is satisfied by the model,
every vector space has a Hamel basis. Furthermore, if we assume that all vector
spaces have a basis, then this implies the Axiom of Choice, which shows that
Choice is necessary in the proof that every vector space has a basis. In the
special case of a countable vector space one can prove the existence of bases in
ZF.
It is possible to construct a model of set theory that fails the Axiom of
Choice by choosing a suitable partially ordered set P , a generic filter G on P ,
a group G of automorphisms of P and a normal filter F on G and forming the
corresponding symmetric extension N . This model contains an amorphous set
and hence an infinite-dimensional vector space whose proper subspaces are all
finite-dimensional. Using this fact it can easily be shown that this vector space
has no basis and that it enjoys several counter-intuitive properties.
We saw a problem in the construction of the amorphous vector space: al-
though it is possible to define the operations of addition and multiplication, we
did not show that these are in the model N . It is unclear to us how to solve
this problem, nor whether it is essential for proving existence of a vector space
without a basis.
25
Appendix A
Axioms of ZF
Here we will present an overview of the axioms of ZF, mainly taken from [8]. In
Section 1.1 we already saw an axiom about equality of sets.
Axiom of Extensionality. Two sets are equal if they contain the same ele-
ments. Formally,
∀x∀y (∀z (z ∈ x ↔ z ∈ y) → x = y) .
The following axioms can be applied to construct new sets from old ones.
Axiom of Pairing. If x and y are sets, then there exists a set containing x
and y and nothing else. This set is denoted {x, y}. The formal version of this
axiom is
∀x∀y∃z∀w (w ∈ z ↔ (w = x ∨ w = y)) .
Axiom of Union. The set-theoretical universe is closed under arbitrary unions.
∀x∃y∀z(z ∈ y ↔ ∃w(z ∈ w ∧ w ∈ x))
The set y in this axiom is called the union of x and is unique by Extensionality.
Axiom of Power Sets. Each set x has a power set denoted by Px. This set
consists of all subsets of x. Uniqueness follows from Extentionality.
∀x∃y∀z(z ∈ y ↔ z ⊆ x)
Next we have two axioms that are actually axiom schemata. This means
that they represent an infinite number of axioms, one axiom for each possible
formula ϕ.
Axiom of Comprehension. Given a set x, it is possible to form the subset of
x consisting of all z satisfying ϕ(z).
∀x∃y∀z(z ∈ y ↔ (z ∈ x ∧ ϕ(z)))
The set y can be written as {z ∈ y | ϕ(z)}.
Axiom of Replacement. The image of a set under a function is again a set.
ϕ(w, z) should be a formula with two free variables, where the first one represents
points in the domain and the second one points in the range.
(∀w∃!zϕ(w, z)) → ∀x∃y∀z(z ∈ y ↔ ∃w(w ∈ x ∧ ϕ(w, z)))
27
28 A. AXIOMS OF ZF
Until now, we have provided a number of ways to construct new sets given
certain sets, but we have not yet established the existence of even a single set.
The next axiom postulates the existence of an infinite set, thus showing that
the set-theoretical universe is non-empty.
Axiom of Infinity. There exists a set containing the empty set and containing
x ∪ {x} whenever it contains x.
∃x((∀y(∀z(z ∈
/ y)) → y ∈ x) ∧ ∀y(y ∈ x → y ∪ {y} ∈ x))
This is a rather difficult existence axiom. Together with the Axiom of Com-
prehension, it can be used to prove the existence of an empty set, which is an
easier statement. Let ω be a set satisfying the condition of the Axiom of Infinity
and construct, by Comprehension, the subset ∅ = {x ∈ ω | x 6= x}. This set
satisfies ∀x(x ∈
/ ∅). Moreover, it is unique by Extensionality.
The last axiom is rarely used in the development ordinary mathematics, but
it proves for example that there is no set x for which x ∈ x.
Axiom of Regularity. Every set x contains a set disjoint from x.
∀x∃y(y ∈ x ∧ x ∩ y = ∅)
Appendix B
29
30 B. THE BASIC FRAENKEL MODEL
x ∈ y ⇒ F (x) ∈ F (y).
This theorem states that a subset of V has the same structure as a subset of
N . It is not possible to embed the full structure of V in N , because ZF without
and with atoms are different theories. However, it is always possible to get V 0
and N 0 “sufficiently large” such that N contains an amorphous set. For details
and a proof of the embedding theorem see [7].
Bibliography
[3] S.H. Friedberg, A.J. Insel, and L.E. Spence. Linear Algebra. Prentice Hall,
2003.
[4] P.R. Halmos. Naive Set Theory. D. Van Nostrand Company, 1960.
[5] H. Herrlich. Axiom of Choice. Springer, 2006.
[6] J.L. Hickman. The construction of groups in models of set theory that fail
the axiom of choice. Bull. Austral. Math. Soc., 14:199–232, 1976.
[7] T.J. Jech. The Axiom of Choice. North-Holland Publishing Company, 1973.
[8] K. Kunen. Set Theory, An Introduction to Independence Proofs. North-
Holland Publishing Company, 1980.
[9] H. Laeuchli. Auswahlaxiom in der algebra. Commentarii Mathematici Hel-
vetici, 37:1–18, 1963.
31