Copyright c (1995–2007 by Stephen G.

Simpson
Foundations of Mathematics
Stephen G. Simpson
October 16, 2008
Department of Mathematics
The Pennsylvania State University
University Park, State College PA 16802
simpson@math.psu.edu
This is a set of lecture notes for my course, Foundations of Mathematics I,
offered as Mathematics 558 at the Pennsylvania State University, most recently
in Spring 2007.
1
Contents
1 Computable Functions 6
1.1 Primitive Recursive Functions . . . . . . . . . . . . . . . . . . . . 6
1.2 The Ackermann Function . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Computable Functions . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Partial Recursive Functions . . . . . . . . . . . . . . . . . . . . . 23
1.5 The Enumeration Theorem . . . . . . . . . . . . . . . . . . . . . 25
1.6 Consequences of the Enumeration Theorem . . . . . . . . . . . . 30
1.7 Unsolvable Problems . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.8 The Recursion Theorem . . . . . . . . . . . . . . . . . . . . . . . 39
1.9 The Arithmetical Hierarchy . . . . . . . . . . . . . . . . . . . . . 41
2 Undecidability of Arithmetic 48
2.1 Terms, Formulas, and Sentences . . . . . . . . . . . . . . . . . . . 48
2.2 Arithmetical Definability . . . . . . . . . . . . . . . . . . . . . . . 50
2.3 G¨odel Numbers of Formulas . . . . . . . . . . . . . . . . . . . . . 57
3 The Real Number System 60
3.1 Quantifier Elimination . . . . . . . . . . . . . . . . . . . . . . . . 60
3.2 Decidability of the Real Number System . . . . . . . . . . . . . . 66
4 Informal Set Theory 69
4.1 Operations on Sets . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 Cardinal Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3 Well-Orderings and Ordinal Numbers . . . . . . . . . . . . . . . 74
4.4 Transfinite Recursion . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.5 Cardinal Numbers, Continued . . . . . . . . . . . . . . . . . . . . 81
4.6 Cardinal Arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.7 Some Classes of Cardinals . . . . . . . . . . . . . . . . . . . . . . 85
4.8 Pure Well-Founded Sets . . . . . . . . . . . . . . . . . . . . . . . 87
4.9 Set-Theoretic Foundations . . . . . . . . . . . . . . . . . . . . . . 88
5 Axiomatic Set Theory 92
5.1 The Axioms of Set Theory . . . . . . . . . . . . . . . . . . . . . . 92
5.2 Models of Set Theory . . . . . . . . . . . . . . . . . . . . . . . . 96
2
5.3 Transitive Models and Inaccessible Cardinals . . . . . . . . . . . 99
5.4 Constructible Sets . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.5 Forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.6 Independence of CH . . . . . . . . . . . . . . . . . . . . . . . . . 111
6 Topics in Set Theory 114
6.1 Stationary Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.2 Large Cardinals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.3 Ultrafilters and Ultraproducts . . . . . . . . . . . . . . . . . . . . 116
6.4 Measurable Cardinals . . . . . . . . . . . . . . . . . . . . . . . . 119
6.5 Ramsey’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3
List of Figures
1.1 Register Machine Instructions . . . . . . . . . . . . . . . . . . . . 18
1.2 An Addition Program . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3 The Initial Functions . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4 Generalized Composition . . . . . . . . . . . . . . . . . . . . . . 20
1.5 A Multiplication Program . . . . . . . . . . . . . . . . . . . . . . 21
1.6 Primitive Recursion . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.7 Minimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.8 A Program with Labeled Instructions . . . . . . . . . . . . . . . 27
1.9 Incrementing P
i
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.10 Decrementing P
i
. . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.11 Stopping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.12 Parametrization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4
List of Tables
1.1 The Ackermann branches. . . . . . . . . . . . . . . . . . . . . . . 14
5
Chapter 1
Computable Functions
We use N to denote the set of natural numbers,
N = ¦0, 1, 2, . . .¦ .
For k ≥ 1, the k-fold Cartesian product
N . . . N
. .. .
k
is denoted N
k
. A k-place function is a function f : N
k
→ N and is sometimes
indicated with the lambda-notation,
f = λx
1
x
k
[ f(x
1
, . . . , x
k
) ] .
A number-theoretic function is a k-place function for some k ≥ 1.
The purpose of this chapter is to define and study an important class of
number-theoretic functions, the recursive functions (sometimes called the com-
putable functions). We begin with a certain subclass known as the primitive
recursive functions.
1.1 Primitive Recursive Functions
Loosely speaking, a recursion is any kind of inductive definition, and a primitive
recursion is an especially straightforward kind of recursion, in which the value
of a number-theoretic function at argument x + 1 is defined in terms of the
value at argument x. For example, the factorial function λx[ x! ] is defined by
the primitive recursion equations 0! = 1, (x + 1)! = x!(x + 1). A number-
theoretic function is said to be primitive recursive if it can be built up by means
of primitive recursions. This concept is made precise in the following definition.
Definition 1.1.1 (Primitive Recursive Functions). The class PR of primitive
recursive functions is the smallest class C of number-theoretic functions having
the following closure properties.
6
1. The constant zero function Z = λx[ 0 ] belongs to C.
2. The successor function S = λx[ x + 1 ] belongs to C.
3. For each k ≥ 1 and 1 ≤ i ≤ k, the projection function P
ki
= λx
1
x
k
[ x
i
]
belongs to C.
4. C is closed under generalized composition. This means that whenever the
k-place functions
λx
1
x
k
[ g
1
(x
1
, . . . , x
k
) ], . . . , λx
1
x
k
[ g
m
(x
1
, . . . , x
k
) ]
and the m-place function λy
1
y
m
[ h(y
1
, . . . , y
m
) ] all belong to C, then
the k-place function
f = λx
1
x
k
[ h(g
1
(x
1
, . . . , x
k
), . . . , g
m
(x
1
, . . . , x
k
)) ]
also belongs to C. Here f is defined by
f(x
1
, . . . , x
k
) = h(g
1
(x
1
, . . . , x
k
), . . . , g
m
(x
1
, . . . , x
k
)) .
5. C is closed under primitive recursion. This means that whenever the
k-place function λx
1
x
k
[ g(x
1
, . . . , x
k
) ] and the (k+2)-place function
λyzx
1
x
k
[ h(y, z, x
1
, . . . , x
k
) ]
belong to C, then the (k+1)-place function λyx
1
x
k
[ f(y, x
1
, . . . , x
k
) ]
defined by
f(0, x
1
, . . . , x
k
) = g(x
1
, . . . , x
k
)
f(y + 1, x
1
, . . . , x
k
) = h(y, f(y, x
1
, . . . , x
k
), x
1
, . . . , x
k
)
also belongs to C.
We now list some examples of primitive recursive functions.
Examples 1.1.2.
1. The recursion equations
x + 0 = x
x + (y + 1) = (x +y) + 1
show that the addition function λxy [ x +y ] is primitive recursive.
2. The recursion equations
x 0 = 0
x (y + 1) = (x y) +x
show that the multiplication function λxy [ x y ] is primitive recursive.
7
3. The recursion equations
x
0
= 1
x
y+1
= x
y
x
show that the exponentiation function λxy [ x
y
] is primitive recursive.
4. As already mentioned, the recursion equations
0! = 1
(x + 1)! = x! (x + 1)
show that the factorial function λx[ x! ] is primitive recursive.
We now proceed to further enlarge our library of primitive recursive func-
tions. First, the recursion equations P(0) = 0, P(x + 1) = x show that the
“predecessor” function
P(x) =

x −1 if x > 0 ,
0 if x = 0
is primitive recursive. We can then obtain the truncated subtraction function
x

−y =

x −y if x ≥ y ,
0 if x < y
using primitive recursion equations x

−0 = x, x

−(y+1) = P(x

−y). (Truncated
subraction is useful because ordinary subtraction is not a function from N
2
into
N.) We shall also have use for
[x −y[ = (x

−y) + (y

−x)
and α(x) = 1

−x. Note that
α(x) =

0 if x > 0 ,
1 if x = 0 .
The following exercise will become easy after we have developed a little more
machinery.
Exercise 1.1.3. Show that the Fibonacci function, defined by
fib(0) = 0 ,
fib(1) = 1 ,
fib(x + 2) = fib(x) + fib(x + 1)
is primitive recursive. (The first few values of this function are 0, 1, 1, 2, 3, 5,
8, 13, 21, 34, 55, . . . .)
8
In addition to primitive recursive functions, we shall want to consider prim-
itive recursive predicates. By a k-place predicate we mean a subset of N
k
. If
R ⊆ N
k
is a k-place predicate and x
1
, . . . , x
k
are elements of N, we say that
R(x
1
, . . . , x
k
) is true if 'x
1
, . . . , x
k
` ∈ R, otherwise false. A number-theoretic
predicate is a k-place predicate for some k ≥ 1.
Definition 1.1.4. A k-place predicate R ⊆ N
k
is said to be primitive recursive
if its characteristic function
χ
R
(x
1
, . . . , x
k
) =

1 if R(x
1
, . . . , x
k
) is true
0 if R(x
1
, . . . , x
k
) is false
is primitive recursive.
For example, the 2-place predicates x = y and x < y are primitive recursive,
since χ
=
(x, y) = α([x −y[) and χ
<
(x, y) = α(α(y

−x)).
Lemma 1.1.5 (Boolean Connectives). If P and Q are primitive recursive pred-
icates, then so are P, P ∧ Q, and P ∨ Q. (Here , ∧, and ∨ denote negation,
conjunction, and (nonexclusive) disjunction, respectively.)
Proof. We have χ
¬P
= α(χ
P
) and χ
P∧Q
= χ
P
χ
Q
. Also
χ
P∨Q
= α(α(χ
P
) α(χ
Q
))
since P ∨ Q ≡ ((P) ∧ (Q)) (de Morgan’s law).
Lemma 1.1.6 (Iterated Sums and Products). If f(x, y, z
1
, . . . , z
k
) is a primitive
recursive function, then so are
g(y, z
1
, . . . , z
k
) =
y−1
¸
x=0
f(x, y, z
1
, . . . , z
k
) (= 0 if y = 0)
and
h(y, z
1
, . . . , z
k
) =
y−1
¸
x=0
f(x, y, z
1
, . . . , z
k
) (= 1 if y = 0) .
Proof. We have g(y, z
1
, . . . , z
k
) = g

(y, y, z
1
, . . . , z
k
) where
g

(w, y, z
1
, . . . , z
k
) =
w−1
¸
x=0
f(x, y, z
1
, . . . , z
k
) .
The recursion equations
g

(0, y, z
1
, . . . , z
k
) = 0
g

(w + 1, y, z
1
, . . . , z
k
) = g

(w, y, z
1
, . . . , z
k
) +f(w, y, z
1
, . . . , z
k
)
show that g

is primitive recursive, hence g is primitive recursive. The treatment
of h is similar.
9
Lemma 1.1.7 (Finite Conjuction and Disjunction). If R(x, y, z
1
, . . . , z
k
) is a
primitive recursive predicate, then so are
P(y, z
1
, . . . , z
k
) ≡
y−1

x=0
R(x, y, z
1
, . . . , z
k
)
and
Q(y, z
1
, . . . , z
k
) ≡
y−1
¸
x=0
R(x, y, z
1
, . . . , z
k
) .
Proof. We have
χ
P
(y, z
1
, . . . , z
k
) =
y−1
¸
x=0
χ
R
(x, y, z
1
, . . . , z
k
)
and
χ
Q
(y, z
1
, . . . , z
k
) = α

y−1
¸
x=0
α(χ
R
(x, y, z
1
, . . . , z
k
))

so our result follows from the previous lemma.
Note that

y−1
x=0
and

y−1
x=0
can be paraphrased as “for all x in the range
0 ≤ x < y” and “there exists x in the range 0 ≤ x < y”, respectively. These
operators are sometimes called bounded quantifiers.
The above lemmas make it easy to show that many familiar predicates are
primitive recursive. For example, the set (i.e., 1-place predicate) of prime num-
bers is primitive recursive, since
Prime (x) ≡ x is a prime number
≡ x > 1 ∧

u<x

v<x
(x = u v ∧ u > 1 ∧ v > 1) .
Similarly, the following lemma can be used to show that many familiar func-
tions are primitive recursive.
Lemma 1.1.8 (Bounded Least Number Operator). If R(x, y, z
1
, . . . , z
k
) is a
primitive recursive predicate, then the function
f(y, z
1
, . . . , z
k
) =

least x < y such that R(x, y, z
1
, . . . , z
k
) holds ,
if such x exists ,
y otherwise
is primitive recursive.
Proof. We have
f(y, z
1
, . . . , z
k
) =
¸
y−1
x=0

x χ
R
(x, y, z
1
, . . . , z
k
)
¸
x−1
w=0
α(χ
R
(w, y, z
1
, . . . , z
k
))

+y
¸
y−1
x=0
α(χ
R
(x, y, z
1
, . . . , z
k
)) ,
so f is primitive recursive.
10
For example, consider the function λn[ p
n
] which enumerates the prime
numbers in increasing order. (The first few values of this function are p
0
= 2,
p
1
= 3, p
2
= 5, p
3
= 7, . . . .) We want to use the bounded least number operator
to show that λn[ p
n
] is primitive recursive. First, recall a famous theorem of
Euclid which gives the bound p
n+1
≤ p
n
! +1. We can then write p
0
= 2, p
n+1
=
least x ≤ p
n
! + 1 such that Prime (x) and x > p
n
. Thus λn[ p
n
] is primitive
recursive.
As another application of the bounded least number operator, note that the
functions
Quotient (y, x) = ⌊y/x⌋ = q ,
Remainder (y, x) = (y mod x) = r ,
where y = q x +r, 0 ≤ r < x, 0 ≤ q, are primitive recursive, in view of
Quotient (y, x) = least q ≤ y such that

r<x
(y = q x +r) ,
Remainder (y, x) = least r < x such that

q≤y
(y = q x +r) .
Using the bounded least number operator, we can obtain a primitive recur-
sive method of encoding ordered pairs of natural numbers as single numbers.
For our pairing function we use λuv [ 2
u
3
v
]. The unpairing functions are then
λz [ (z)
0
] and λz [ (z)
1
], where
(z)
n
= least w < z such that Remainder(z, p
w+1
n
) = 0
= the exponent of p
n
in the prime power factorization of z .
Note that (2
u
3
v
)
0
= u and (2
u
3
v
)
1
= v.
More generally, we can encode variable-length finite sequences of natural
numbers as single numbers. The sequence 'a
0
, a
1
, . . . , a
m−1
` is encoded by
a =
¸
x<m
p
ax
x
,
and for decoding we can use the primitive recursive function λzx[ (z)
x
], since
(a)
x
= a
x
. This method of prime power coding will be used extensively in the
proof of the Enumeration Theorem, below.
The pairing and unpairing functions make it easy to show that the Fibonacci
function is primitive recursive (cf. Exercise 1.1.3). Namely, we first note that
the auxiliary function λx[ fibpair(x) ], defined by
fibpair(x) = 2
fib(x)
3
fib(x+1)
,
is primitive recursive in view of
fibpair(0) = 2
0
3
1
,
fibpair(x + 1) = 2
(fibpair(x))
1
3
(fibpair(x))
0
+(fibpair(x))
1
.
Then λx[ fib(x) ] is primitive recursive since fib(x) = (fibpair(x))
0
.
11
Exercise 1.1.9. Show that the 2-place number-theoretic functions GCD(x, y)
and LCM(x, y), the greatest common divisor and least common multiple of x
and y, are primitive recursive.
Solution. GCD(x, y) = least z ≤ max(x, y) such that Remainder(x, z) =
Remainder(y, z) = 0. LCM(x, y) = least z ≤ x y such that Remainder(z, x) =
Remainder(z, y) = 0.
Exercise 1.1.10. Show that the 1-place number-theoretic function f(n) given
by
f(n) = 1 +
n−1
¸
k=0
f(k)
n
is primitive recursive.
Solution. Consider the so-called course-of-values function
¯
f(n) =
n−1
¸
k=0
p
f(k)
k
,
i.e.,
¯
f(n) encodes the variable-length finite sequence 'f(0), f(1), . . . , f(n −1)`
via prime power coding. Then
¯
f(n) is primitive recursive, in view of the recur-
sion equations
¯
f(0) = 1 and
¯
f(n + 1) =
¯
f(n) p
h(n,
e
f(n))
n
,
where h(n, z) = 1 +
¸
n−1
k=0
((z)
k
)
n
. It now follows that f(n) = (
¯
f(n + 1))
n
is primitive recursive. This general technique is known as course-of-values
recursion.
Exercise 1.1.11. Show that the function λn

nth digit of

2

is primitive
recursive.
Solution. The nth digit of

2 is f(n) = Remainder(g(n), 10), where g(n) = the
least x < 4 10
2n
such that (x + 1)
2
> 2 10
2n
.
Exercise 1.1.12. Show that the 1-place number-theoretic function
f(n) = the nth decimal digit of π = 3.141592
is primitive recursive.
Hint: You may want to use the fact that [π − a/b[ > 1/b
42
for all integers
a, b > 1. This result is due to K. Mahler, On the approximation of π, Nederl.
Akad. Wetensch. Proc. Ser. A., 56, Indagationes Math., 15, 30–42, 1953.
Solution. We use the well-known series
π
4
= 1 −
1
3
+
1
5

1
7
+
1
9
− =

¸
n=0
(−1)
n
2n + 1
,
12
i.e.,
π = 4 −
4
3
+
4
5

4
7
+
4
9
− =

¸
n=0
(−1)
n
4
2n + 1
.
Let S
k
be the kth partial sum of this series. We have S
k
= a(k)/b(k) where the
functions a(k) and b(k) are primitive recursive, namely
a(k) =
k
¸
n=0
(−1)
n
4(2k + 1)!
2n + 1
and b(k) = (2k + 1)!. Note also that the functions
g(n, a, b) = (µi < 10
n
b) (10
i
a ≥ 10
n
b)
and
h(n, a, b) = Rem(Quot(10
g(n,a,b)
a, b), 10) = the nth digit of a/b
is primitive recursive. By Mahler’s result, for each n there exists k < 10
50n
such that S
k
and S
k+1
have the same first n digits. Since π lies between S
k
and S
k+1
, it follows that S
k
and π have the same first n digits, so in partic-
ular f(n) = the nth digit of S
k
. Using the bounded least number operator,
we have f(n) = h(n, a(k(n)), b(k(n))) where k(n) = the least k < 10
50n
such
that

n
m=0
g(m, a(k), b(k)) = g(m, a(k + 1), b(k + 1)). Clearly this is primitive
recursive.
1.2 The Ackermann Function
In this section we present an example of a function which is not primitive re-
cursive, yet is clearly computable in some intuitive sense. The precise concept
of computability which we have in mind will be explained in the next section.
Definition 1.2.1. We define a sequence of 1-place functions A
n
, n ∈ N, as
follows:
A
0
(x) = 2x,
A
n+1
(x) = A
n
A
n
A
n
. .. .
x
(1) .
Thus A
0
(x) = 2x, A
1
(x) = 2
x
, A
2
(x) = 2
2
·
·
·
2
(height x), etc.
Exercise 1.2.2 (the Ackermann hierarchy).
1. Show that, for each n, λx[ A
n
(x) ] is primitive recursive.
2. Show that
(a) A
n
(x + 1) > A
n
(x) > x for all x ≥ 1 and all n.
13
(b) A
n+1
(x) ≥ A
n
(x + 1) for all x ≥ 3 and all n.
3. Show that for each k-place primitive recursive function
λx
1
. . . x
k
[ f(x
1
, . . . , x
k
) ]
there exists n such that
f(x
1
, . . . , x
k
) ≤ A
n
(max(3, x
1
, . . . , x
k
))
for all x
1
, . . . , x
k
.
4. Show that the 2-place function λnx[ A
n
(x) ] is not primitive recursive.
This is known as the Ackermann function.
5. Show that the 3-place relation λnxy [ A
n
(x) = y ] is primitive recursive.
Use this to show that the Ackermann function is computable, i.e., recur-
sive, in the sense of Section 1.3.
Solutions.
1. Show that A
n
(1) = 2, A
n
(2) = 4, and A
n+1
(3) = A
n
(4) for all n. Compute
A
n
(x) for all n, x with n +x ≤ 8.
Solution. For all n we have A
n+1
(1) = A
n
(1), hence by induction A
n
(1) =
A
0
(1) = 2. Also A
n+1
(2) = A
n
(A
n
(1)) = A
n
(2), hence by induction
A
n
(2) = A
0
(2) = 4. Also, for all n and x we have A
n+1
(x + 1) =
A
n
(A
n+1
(x)), in particular A
n+1
(3) = A
n
(A
n+1
(2)) = A
n
(4). Table 1.1
shows A
n
(x) for small values of n, x.
Table 1.1: The Ackermann branches.
0 1 2 3 4 5
A
0
0 2 4 6 8 10
A
1
1 2 4 8 16 32
A
2
1 2 4 16 2
16
2
2
16
A
3
1 2 4 2
16
2
2
·
·
·
2
(height 2
16
) 2
2
·
·
·
2
(height 2
2
·
·
·
2
(height 2
16
))
A
4
1 2 4 2
2
·
·
·
2
(height 2
16
) A
3
(2
2
·
·
·
2
(height 2
16
))
A
5
1 2 4 A
3
(2
2
·
·
·
2
(height 2
16
))
A
6
1 2 4
2. Prove the following:
14
(a) A
n
(x + 1) > A
n
(x) > x for all x ≥ 1 and all n.
(b) A
n+1
(x) ≥ A
n
(x + 1) for all x ≥ 3 and all n.
(c) For each primitive recursive function f(x
1
, . . . , x
k
) there exists n such
that A
n
covers f, i.e.,
f(x
1
, . . . , x
k
) ≤ A
n
(max(3, x
1
, . . . , x
k
))
for all x
1
, . . . , x
k
.
(d) The 1-place function λx(A
x
(x)) is not primitive recursive.
(e) The 2-place function λxy (A
x
(y)) is not primitive recursive.
Solution. First we prove A
n
(x + 1) > A
n
(x) > x for x ≥ 1, by induction
on n. For n = 0 we have A
0
(x + 1) = 2x + 2 > 2x = A
0
(x) for all
x, and A
0
(x) = 2x > x for x ≥ 1. For n + 1 and x ≥ 1 we have
A
n+1
(x + 1) = A
n
(A
n+1
(x)) > A
n+1
(x) by inductive hypothesis. Thus
A
n+1
is strictly monotone. Since A
n+1
(0) > 0, it follows that A
n+1
(x) > x
for all x.
Next we prove A
n+1
(x) ≥ A
n
(x+1) for x ≥ 3, by induction on x. For x = 3
we have A
n+1
(3) = A
n
(4) as noted above, and inductively A
n+1
(x +1) =
A
n
(A
n+1
(x)) ≥ A
n
(A
n
(x+1)) ≥ A
n
(x+2), since A
n
is strictly monotone
and A
n
(x + 1) ≥ x + 2 by what has already been proved.
Next we prove that each primitive recursive function is covered by A
n
for
some n. We prove this by induction on the class of primitive recursive
functions. We begin by noting that the initial functions are covered by
A
0
.
Suppose f is obtained by generalized composition, say
f(x
1
, . . . , x
k
) = h(g
1
(x
1
, . . . , x
k
), . . . , g
m
(x
1
, . . . , x
k
)).
Let n be such that A
n
covers h and A
n+1
covers g
1
, . . . , g
m
. We then have
f(x
1
, . . . , x
k
) = h(g
1
(x
1
, . . . , x
k
), . . . , g
m
(x
1
, . . . , x
k
))
≤ A
n
(max(3, g
1
(x
1
, . . . , x
k
), . . . , g
m
(x
1
, . . . , x
k
)))
≤ A
n
(A
n+1
(max(3, x
1
, . . . , x
k
)))
= A
n+1
(max(3, x
1
, . . . , x
k
) + 1)
≤ A
n+2
(max(3, x
1
, . . . , x
k
)),
i.e., A
n+2
covers f.
Suppose f is obtained by primitive recursion, say
f(0, x
1
, . . . , x
k
) = g(x
1
, . . . , x
k
),
f(y + 1, x
1
, . . . , x
k
) = h(y, f(y, x
1
, . . . , x
k
), x
1
, . . . , x
k
).
Let n be such that A
n
covers h and A
n+1
covers g. We first claim that
f(y, x
1
, . . . , x
k
) ≤ A
n+1
(y + max(3, x
1
, . . . , x
k
))
15
for all y, x
1
, . . . , x
k
. We prove this by induction on y. For y = 0 we
have f(0, x
1
, . . . , x
k
) = g(x
1
, . . . , x
k
) ≤ A
n+1
(max(3, x
1
, . . . , x
k
)). For the
inductive step we have
f(y + 1, x
1
, . . . , x
k
) = h(y, f(y, x
1
, . . . , x
k
), x
1
, . . . , x
k
)
≤ A
n
(max(3, y, f(y, x
1
, . . . , x
k
), x
1
, . . . , x
k
)
≤ A
n
(max(3, y, A
n+1
(y + max(3, x
1
, . . . , x
k
)), x
1
, . . . , x
k
)
= A
n
(A
n+1
(y + max(3, x
1
, . . . , x
k
)))
= A
n+1
(y + 1 + max(3, x
1
, . . . , x
k
))
and this proves our claim. We then have
f(y, x
1
, . . . , x
k
) ≤ A
n+1
(y + max(3, x
1
, . . . , x
k
))
≤ A
n+1
(2 max(3, y, x
1
, . . . , x
k
))
≤ A
n+1
(A
n+2
(max(3, y, x
1
, . . . , x
k
)))
= A
n+2
(max(3, y, x
1
, . . . , x
k
) + 1)
≤ A
n+3
(max(3, y, x
1
, . . . , x
k
)),
i.e., A
n+3
covers f. This completes the proof that each primitive recursive
function is covered by A
n
for some n.
Now, if A
x
(x) were primitive recursive, then A
x
(x) +1 would be primitive
recursive, hence covered by A
n
for some n ≥ 3. But then in particular
A
n
(n) + 1 ≤ A
n
(max(3, n)) = A
n
(n), a contradiction. Thus the 1-place
function A
x
(x) is not primitive recursive. It follows immediately that the
2-place function A
x
(y) is not primitive recursive.
3. Show that the 3-place relation
¦'x, y, z` [ A
x
(y) = z¦
is primitive recursive. Use this to prove that λxy (A
x
(y)) is recursive.
Hence λx(A
x
(x)) is recursive.
Solution. For all x, y > 0 we have
0 < y < A
x
(y) = A
x−1
(A
x
(y −1)) = A
x−1
(y

)
where y

= A
x
(y − 1). Since A
x−1
(y

) = A
x
(y) ≥ 2, it follows that
0 < y

< A
x−1
(y

) = A
x
(y). Repeating this step x times, we obtain a
finite sequence y
0
, y
1
, y
2
, . . . , y
x
starting with y such that
A
x
(y) = A
x
(y
0
) = A
x−1
(y
1
) = A
x−2
(y
2
) = = A
0
(y
x
) = 2y
x
,
and each of y
0
, y
1
, . . . , y
x
is > 0 and < A
x
(y). Moreover, if y > 2 then we
also have x < A
x
(y). Thus the 3-place predicate A
x
(y) = z can be defined
16
by course-of-values recursion on z as follows:
A
x
(y) = z if and only if
(x = 0 ∧ z = 2y) ∨
(x > 0 ∧ y = 0 ∧ z = 1) ∨
(x > 0 ∧ y = 1 ∧ z = 2) ∨
(x > 0 ∧ y = 2 ∧ z = 4) ∨
(x > 0 ∧ y > 2 ∧ x < z ∧ ∃y
0
, y
1
, . . . , y
x
< z
(y
0
= y ∧ ∀i < x(y
i+1
= A
x−i
(y
i
−1)) ∧ z = 2y
x
)).
Actually, the function being defined by primitive recursion is
a(w) =
¸
¦p
2
x
3
y
5
z [ A
x
(y) = z ∧ x, y, z < w¦.
In any case, it follows that the 3-place predicate A
x
(y) = z is primitive
recursive.
Applying the least number operator, we see that the 2-place function
A
x
(y) is recursive. It follows immediately that the 1-place function A
x
(x)
is recursive.
1.3 Computable Functions
In this section we define the class of computable (i.e., recursive) number-theoretic
functions. We show that the primitive recursive functions form a proper subclass
of the computable functions.
Our definition will be given in terms of a register machine. We assume the
existence of infinitely many registers R
1
, R
2
, . . . , R
i
, . . . . At any given time,
each register contains a natural number. If the number contained in R
i
is 0, we
say that R
i
is empty. At any given time, all but finitely many of the registers
are empty. The basic actions that the machine can perform are to increment or
decrement a register, i.e., add or subtract 1 from the number contained in it.
A register machine program consists of finitely many instructions linked to-
gether in a flow diagram indicating the order in which the instructions are to
be executed. There are four types of instructions: R
+
i
, R

i
, start, and stop.
Each program contains exactly one start instruction, which is executed first.
An R
+
i
instruction is executed by incrementing R
i
and then proceeding to an-
other, specified, instruction. An R

i
instruction is executed by testing R
i
for
emptiness. If R
i
is empty, we proceed to one of two specified instructions. If R
i
is nonempty, we decrement it and then proceed to the other of the two specified
instructions. A stop instruction causes execution of the program to halt. See
Figure 1.1.
For example, consider the addition program depicted in Figure 1.2. If we run
this program starting with natural numbers x and y in R
1
and R
2
respectively,
the run will eventually halt with x +y in R
3
.
17
start
`

?
?
?
?
?
?
?

R
+
i
`








begin run of program increment register R
i

?
?
?
?
?
?
?

R

i
`
e

?
?
?
?
?
?
?








?
?
?
?
?
?
?

stop
`








if R
i
empty go to e, terminate run of program
otherwise decrement R
i
Figure 1.1: Register Machine Instructions
start
`
R

3
`

e

R

1
`

e

R

2
`

e

stop
`
R
+
3
`

R
+
3
`

Figure 1.2: An Addition Program
18
Let { be a register machine program, and let x
1
, . . . , x
k
be natural numbers,
i.e., elements of N. We write {(x
1
, . . . , x
k
) to denote the unique run of { starting
with x
1
in R
1
, . . . , x
k
in R
k
, and all other registers empty. Uniqueness follows
from the fact that the register machine operates deterministically.
Definition 1.3.1 (Computable Functions). A k-place number-theoretic func-
tion
λx
1
. . . x
k
[ f(x
1
, . . . , x
k
) ]
is said to be computable if there exists a register machine program { which
computes it, i.e. for all x
1
, . . . , x
k
∈ N, {(x
1
, . . . , x
k
) eventually halts with
y = f(x
1
, . . . , x
k
) in R
k+1
.
For example, the addition program of Figure 1.2 shows that λx
1
x
2
[ x
1
+x
2
]
is computable.
We shall now prove that all primitive recursive functions are computable.
Lemma 1.3.2. The initial functions Z, S, and P
ki
, 1 ≤ i ≤ k, are computable.
Proof. The functions Z = λx[ 0 ], S = λx[ x + 1 ], and P
ki
= λx
1
. . . x
k
[ x
i
] are
computed by the register machine programs given in Figure 1.3.
Zero start
`
stop
`
Successor start
`
R

1
`

e

R
+
2
`
stop
`
R
+
2
`

Projection start
`
R

i
`

e

stop
`
R
+
k+1
\

Figure 1.3: The Initial Functions
19
start
`

F

1
`

e
. . .
e

F

k
`

e

(
1

G

1 k+1
'`

e
. . .
e

(
m

G

mk+1
'`

e

H

G
+
11
`

G
+
1k
`

H
+
1
`

H
+
m
`

.
.
.

.
.
.

F
+
k+1
`

H

m+1
'`

e

G
+
m1
`

G
+
mk
`

stop
`
Figure 1.4: Generalized Composition
Lemma 1.3.3. The class of computable functions is closed under generalized
composition.
Proof. Assume that
λx
1
. . . x
k
[ g
1
(x
1
, . . . , x
k
) ] , . . . , λx
1
. . . x
k
[ g
m
(x
1
, . . . , x
k
) ]
and λy
1
. . . y
m
[ h(y
1
, . . . , y
m
) ] are computed by register machine programs (
1
,
. . . , (
m
, H respectively. For convenience we regard these programs as being
executed on pairwise disjoint sets of registers. We use G
j1
, . . . , G
jk
, G
j,k+1
,
. . . to denote the registers on which (
j
is executed, 1 ≤ j ≤ m. We use H
1
, . . . ,
H
m
, H
m+1
, . . . to denote the registers on which H is executed.
To compute λx
1
. . . x
k
[ f(x
1
, . . . , x
k
) ] where
f(x
1
, . . . , x
k
) = h(g
1
(x
1
, . . . , x
k
), . . . , g
m
(x
1
, . . . , x
k
)) .
we shall use a register machine program T which we regard as being executed
on registers F
1
, . . . , F
k
, F
k+1
, . . . . In order to make it easy for T to call (
1
,
. . . , (
m
, H, the registers of (
1
, . . . , (
m
, H will be among the auxiliary registers
of T. (The auxiliary registers of T are the registers F
i
, i ≥ k + 2.) Actually,
the auxiliary registers of T consist precisely of the registers of (
1
, . . . , (
m
, H.
Our program T is given in Figure 1.4.
Lemma 1.3.4. The class of computable functions is closed under primitive
recursion.
20
In proving this lemma, the idea will be to write a program containing a loop
which repeatedly calls the iterator h. Let us first illustrate this idea with a
simple example.
Example 1.3.5. The multiplication function λx
1
x
2
[ x
1
x
2
] is computed by
iterated addition, using the program given in Figure 1.5.
start
`
R

1
`
e

R

2
`

e

R

4
`

e

stop
`
R
+
3
`

R
+
2
`

R
+
4
`

Figure 1.5: A Multiplication Program
Exercise 1.3.6. Write a register machine program which computes the ex-
ponential function, i.e., the 2-place number-theoretic function exp(x, y) = x
y
.
Note that x
0
= 1 for all x.
Proof of Lemma 1.3.4. Assume that λx
1
. . . x
k
[ g(x
1
, . . . , x
k
) ] and
λyzx
1
. . . x
k
[ h(y, z, x
1
, . . . , x
k
) ]
are computed by register machine programs ( and H with registers G
1
, . . . ,
G
k
, G
k+1
, . . . and H
1
, H
2
, H
3
, . . . , H
k+2
, H
k+3
, . . . respectively. To compute
λyx
1
. . . x
k
[ f(y, x
1
, . . . , x
k
) ] where
f(0, x
1
, . . . , x
k
) = g(x
1
, . . . , x
k
) ,
f(y + 1, x
1
, . . . , x
k
) = h(y, f(y, x
1
, . . . , x
k
), x
1
, . . . , x
k
) ,
we use a register machine program T with registers F
1
, F
2
, . . . , F
k+1
, F
k+2
,
. . . . See Figure 1.6. The auxiliary registers F
i
, i ≥ k + 3 of T consist of the
registers of ( and H plus two additional registers, U and V .
Theorem 1.3.7. Every primitive recursive function is computable.
21
start
`

F

2
`

e

V
− `

e
. . .
e

F

k+1
`

e

V
− `

e

(

H
+
2
`

V
+ `

F
+
2
`

V
+ `

F
+
k+1
`

F

1
`
e

G

k+1
\

e

G
+
1
`

G
+
k
`

F
+
k+2
`

G

k+1
\

e

stop
`
H
+
2
`

U
+ `







. . .
H

4
`

e

H

3
`

e

H

k+3
\

e

U
− `

e

V
− `

e

F

2
`

e

V
− `

e
. . .
e

F

k+1
`

e

V
− `

e

H

H

2
`

e

V
+ `

U
+ `

V
+ `

F
+
2
`

V
+ `

F
+
k+1
`

F

1
`
e

H

1
`

e

H
+
1
`

H
+
3
`

H
+
k+2
\

F
+
k+2
`

H

k+3
\

e

stop
`
Figure 1.6: Primitive Recursion
22
Proof. The above lemmas show that the computable functions form a class
which contains the initial functions and is closed under generalized composition
and primitive recursion. Since the primitive recursive functions were defined as
the smallest such class, our theorem follows.
1.4 Partial Recursive Functions
A k-place partial function is a function ψ : dom(ψ) → N where dom(ψ) ⊆ N
k
.
We sometimes abbreviate this as ψ : N
k
P
−→ N. We use dom(ψ) and rng(ψ) to
denote the domain and range of ψ, respectively. If dom(ψ) = N
k
, we say that
ψ is total. Thus a total k-place function is just what we have previously called
a k-place function.
The use of partial functions leads to expressions which may or may not have
a numerical value. (For example, ψ(x
1
, . . . , x
k
) + 3 has a numerical value if
and only if 'x
1
, . . . , x
k
` ∈ dom(ψ)). If E is such an expression, we say that E is
defined or convergent (abbreviated E ↓) if E has a numerical value. We say that
E is undefined or divergent (abbreviated E ↑) if E does not have a numerical
value. We write E
1
≃ E
2
to mean that E
1
and E
2
are both defined and equal,
or both undefined.
Definition 1.4.1 (Recursive Functions and Predicates). A k-place (total) func-
tion is said to be recursive if and only if it is computable. A k-place predicate
is said to be recursive if its characteristic function is recursive.
Definition 1.4.2 (Partial Recursive Functions). A k-place partial function ψ is
said to be partial recursive if it is computed by some register machine program
{. This means that, for all x
1
, . . . , x
k
∈ N,
ψ(x
1
, . . . , x
k
) ≃ the number in R
k+1
if and when {(x
1
, . . . , x
k
) stops .
In particular, ψ(x
1
, . . . , x
k
) is defined if and only if {(x
1
, . . . , x
k
) eventually
stops.
Partial recursive functions arise because a particular run of a register ma-
chine program may or may not eventually stop. One way this can happen is
because of an unbounded search, as in the following lemma.
Lemma 1.4.3 (Unbounded Least Number Operator). Let P(x
1
, . . . , x
k
, y) be
a (k+1)-place recursive predicate. Then the k-place partial function ψ defined
by
ψ(x
1
, . . . , x
k
) ≃ least y such that P(x
1
, . . . , x
k
, y) holds
is partial recursive.
23
start
`

F
+
k+1
`
~
~
~
~
~
~
~
. . .
P

2
`

e

P

1
`

e

F

1
`

e

V
− `

e
. . .
e

F

k+1
`

e

V
− `

e

{

P

k+2
`

e
?
?
?
?
?
?
?
V
+ `

F
+
1
`

V
+ `

F
+
k+1
`

stop
`
P
+
1
`

P
+
k+1
`

Figure 1.7: Minimization
Proof. Assume that χ
P
is computed by a register machine program { with
registers P
1
, . . . , P
k
, P
k+1
, P
k+2
, . . . . To compute ψ we use a register machine
program T with registers F
1
, F
2
, . . . , F
k+1
, . . . . The auxiliary registers F
i
,
i ≥ k + 2, of T are the registers of { plus an additional register V . See
Figure 1.7.
The unbounded least number operator is sometimes called the minimization
operator. As a byproduct of the work in the next section, we shall see that all
partial recursive functions can be obtained from primitive recursive functions
by composition and minimization.
Exercise 1.4.4. Show that the function f(n) = nth digit of π is recursive.
Hint: Use an infinite series such as
π
4
= 1 −
1
3
+
1
5

1
7
+
1
9

plus the fact that π is irrational.
Solution. This follows from Exercise 1.1.12. A solution not using Mahler’s result
is as follows.
Let S
k
be the kth partial sum of the alternating series
π = 4 −
4
3
+
4
5

4
7
+
4
9
− =

¸
n=0
(−1)
n
4
2n + 1
,
24
We have S
k
= a(k)/b(k) where a(k) and b(k) are primitive recursive. Since π is
irrational, it follows that for each n there exists k such that S
k
and S
k+1
have
the same first n digits. Since π lies between S
k
and S
k+1
, it follows that S
k
and
π have the same first n digits, so in particular f(n) = the nth digit of S
k
. Note
also that the function
h(n, a, b) = nth digit of a/b
is primitive recursive. Using the least number operator, we have f(n) = h(n, a(k(n)), b(k(n)))
where k(n) = the least k such that

n
m=0
h(m, a(k), b(k)) = h(m, a(k+1), b(k+
1)). Clearly this is recursive.
Exercise 1.4.5. Show that there exists a computable function which is not
primitive recursive. (By 1.2.2 it suffices to show that the Ackermann function
λnx[ A
n
(x) ] is computable.)
Exercise 1.4.6. Let f : N
1−1 onto
−→ N be a permutation of N, the set of natural
numbers. Show that if f is recursive, then the inverse permutation f
−1
is also
recursive.
Solution. Using the least number operator, we have f
−1
(y) = the least x such
that f(x) = y.
Exercise 1.4.7. Give an example of a primitive recursive permutation of N
whose inverse is not primitive recursive.
Solution. From our study of the Ackermann function in Section 1.2, we know
that the predicate ¦'x, y` [ A
x
(x) = y¦ is primitive recursive, although the one-
to-one increasing function x → A
x
(x) is not. Let B be the range of x → A
x
(x),
i.e., B = ¦y [ ∃x(A
x
(x) = y)¦. Then B is primitive recursive, because y ∈ B ⇔

y−1
x=0
A
x
(x) = y. Note also that B is infinite and coinfinite.
For any infinite set S ⊆ N, let π
S
: N → N be the principal function of S,
i.e.,
S = ¦π
S
(0) < π
S
(1) < < π
S
(n) < π
S
(n + 1) < ¦
where π
S
(n) = the nth element of S. Let f be the permutation of N defined by
f(y) =


−1
B
(y) if y ∈ B,

−1
N\B
(y) + 1 if y ∈ N ` B.
By course-of-values recursion, f is primitive recursive. However, f
−1
is not
primitive recursive, because f
−1
(2x) = π
B
(x) = A
x
(x).
1.5 The Enumeration Theorem
To each register machine program c we shall assign a unique number e = #(c).
This number will be called the G¨ odel number of c and will also be called an
index of the k-place partial recursive function which is computed by c
25
Recall that our register machine is equipped with infinitely many registers
R
i
, i ≥ 1. Initially all of the registers are empty except for R
1
, . . . , R
k
which
contain the arguments x
1
, . . . , x
k
. We assume that our program c is given
as a numbered sequence of instructions I
1
, . . . , I
l
. By convention our machine
starts by executing I
1
and stops when it attempts to execute the nonexistent
instruction I
0
. Each instruction I
m
, 1 ≤ m ≤ l, is of the form
increment R
i
then go to instruction I
n0
, (1.0)
or
if R
i
is empty go to I
n0
, otherwise
decrement R
i
then go to instruction I
n1
.
(1.1)
Here n
0
and n
1
are in the range 0 ≤ n ≤ l. To each instruction I
m
we assign a
G¨odel number #(I
m
), where
#(I
m
) =

3
i
5
n0
for I
m
as in (1.0) ,
2 3
i
5
n0
7
n1
for I
m
as in (1.1) .
The G¨odel number of the entire program c is then defined as
#(c) =
l
¸
m=1
p
#(Im)
m
,
where p
0
, p
1
, p
2
, . . . are the prime numbers 2, 3, 5, . . . in increasing order.
Example 1.5.1. Let c be the program in Figure 1.8, which computes λx[ x+1 ].
Listing the instructions I
1
, I
2
, I
3
as shown in the figure, we have
#(I
1
) = 2 3
1
5
3
7
2
= 2 3 125 49 = 36750 ,
#(I
2
) = 3
2
5
1
= 45 ,
#(I
3
) = 3
2
5
0
= 9 ,
so that
#(c) = 3
#(I1)
5
#(I2)
7
#(I3)
= 3
36750
5
45
7
9
.
Lemma 1.5.2. The 1-place predicate
Program(e) ≡ (e is the G¨odel number of some register machine program)
is primitive recursive.
Proof. We have
Program(e) ≡
e
¸
l=0
Program(e, l)
26
I
1
I
3
start
`
R

1
`

e

R
+
2
`
stop
`
R
+
2
`

I
2
Figure 1.8: A Program with Labeled Instructions
where Program(e, l) says that e is the G¨odel number of a register machine
program consisting of l instructions I
1
, . . . , I
l
. We then have
Program(e, l) ≡ e =
l
¸
m=1
p
(e)m
m

l

m=1
e
¸
i=1
l
¸
n0=0
l
¸
n1=0

(e)
m
= 3
i
5
n0
∨ (e)
m
= 2 3
i
5
n0
7
n1

,
the idea being that (e)
m
= #(I
m
). This proves the lemma.
Definition 1.5.3. We denote by ϕ
(k)
e
the k-place partial computable function
which is computed by the register machine program c whose G¨odel number is
e. In more detail, we define
ϕ
(k)
e
(x
1
, , x
k
) ≃ the number in R
k+1
if and when c(x
1
, , x
k
)
stops, where e = #(c), and undefined otherwise.
Note that if e is not the G¨odel number of a register machine program, then
ϕ
(k)
e
(x
1
, , x
k
) is undefined for all x
1
, . . . , x
k
, so in this case ϕ
(k)
e
is the empty
function.
If ψ is a k-place partial recursive function, an index of ψ is any number e
such that ψ = ϕ
(k)
e
, i.e., e is the G¨odel number of a program which computes ψ.
Clearly ψ has many different indices, since there are many different programs
which compute ψ.
Exercise 1.5.4. Find an index of the function
α(x) =

1 if x = 0,
0 if x > 0.
27
Solution. Clearly the labeled program
I
1
I
2
start
`
R

1
`

?
?
?
?
?
?
?
e

R
+
2
`

stop
`
computes α. We have #(I
1
) = 2 3
1
5
2
7
0
= 150 and #(I
2
) = 3
2
5
0
= 9, so
an index of α is e = 3
#(I1)
5
#(I2)
= 3
150
5
9
. In other words, ϕ
(1)
e
= α.
The main theorem on indices reads as follows.
Theorem 1.5.5 (The Enumeration Theorem). For each k ≥ 1, the (k+1)-place
partial function
λex
1
. . . x
k
[ ϕ
(k)
e
(x
1
, . . . , x
k
) ]
is partial recursive.
Remark 1.5.6. The Enumeration Theorem entails the existence of a “univer-
sal” program, i.e., a register machine program which can emulate the action of
any other register machine program. This concept underlies the stored program
digital computer.
In the proof of the Enumeration Theorem, the following easy lemma will be
useful.
Lemma 1.5.7 (Definition by Cases). Let P
1
and P
2
be k-place primitive re-
cursive predicates and let f
1
and f
2
be k-place primitive recursive functions.
Assume that P
1
and P
2
are mutually exclusive and exhaustive, i.e., for each
k-tuple 'x
1
, . . . , x
k
` ∈ N
k
, either P
1
(x
1
, . . . , x
k
) or P
2
(x
1
, . . . , x
k
) holds but not
both. Then the k-place function f defined by
f(x
1
, . . . , x
k
) =

f
1
(x
1
, . . . , x
k
) if P
1
(x
1
, . . . , x
k
) holds
f
2
(x
2
, . . . , x
k
) if P
2
(x
1
, . . . , x
k
) holds
is primitive recursive.
Proof. This is clear since f = f
1
χ
P1
+ f
2
χ
P2
. The extension to more than
two cases is also easy.
Proof of the Enumeration Theorem.
28
The idea of the proof is to represent the state of c(x
1
, . . . , x
k
) after executing
n instructions by a single number
z = State (e, x
1
, . . . , x
k
, n)
= p
m
0

¸

i=1
p
zi
i
,
where z
i
is the number in register R
i
, and I
m
is the next instruction to be
executed. Note that (z)
0
= m and, for all i ≥ 1, (z)
i
= z
i
.
We first show that the State function is primitive recursive. We have
State (e, x
1
, . . . , x
k
, 0) = p
1
0
p
x1
1
. . . p
x
k
k
(begin by executing I
1
) ,
State (e, x
1
, . . . , x
k
, n + 1) = NextState (e, State (e, x
1
, . . . , x
k
, n)) ,
NextState (e, z) =

z p
i
p
−m+n0
0
if ((e)
m
)
0
= 0 ,
z p
−m+n0
0
if ((e)
m
)
0
= 1 and (z)
i
= 0 ,
z p
−1
i
p
−m+n1
0
if ((e)
m
)
0
= 1 and (z)
i
> 0 ,
z otherwise ,
where
m = (z)
0
, i = ((e)
m
)
1
, n
0
= ((e)
m
)
2
, n
1
= ((e)
m
)
3
.
We are now ready to prove the theorem. We use the least number operator
to obtain
Stop (e, x
1
, . . . , x
k
) ≃ least n such that (State (e, x
1
, . . . , x
k
, n))
0
= 0
∧ Program(e) .
(The idea is that our machine stops if and when it is about to execute I
0
. Note
that Stop (e, x
1
, . . . , x
k
) is undefined if e is not the G¨odel number of a register
machine program.) We then use composition to get
FinalState (e, x
1
, . . . , x
k
) ≃ State (e, x
1
, . . . , x
k
, Stop (e, x
1
, . . . , x
k
))
(the state of our machine if and when it stops)
and
Output (e, x
1
, . . . , x
k
) ≃ (FinalState (e, x
1
, . . . , x
k
))
k+1
(the number that is finally in register R
k+1
) .
Since the Output function was obtained by composition, primitive recursion and
the least number operator, it is partial recursive. Moreover, for all e and x
1
,
. . . , x
k
, we clearly have
ϕ
(k)
e
(x
1
, . . . , x
k
) ≃ Output (e, x
1
, . . . , x
k
) .
This completes the proof of the Enumeration Theorem.
29
Exercise 1.5.8. Let f : N
k
→ N be a k-place total recursive function. Show
that f is primitive recursive if and only if there exists an index e of f such that
λx
1
. . . x
k
[ Stop (e, x
1
, . . . , x
k
) ]
is majorized by some primitive recursive function. (See also Exercise 1.2.2.)
Exercise 1.5.9. Fix k ≥ 1. Construct a k+1-place total recursive function
Φ
k
: N
k+1
→N with the following properties:
1. for each e ∈ N, the k-place function λx
1
. . . x
k
[ Φ
k
(e, x
1
, . . . , x
k
) ] is prim-
itive recursive;
2. for each k-place primitive recursive function f : N
k
→ N, there exists an
e such that f = λx
1
. . . x
k
[ Φ
k
(e, x
1
, . . . , x
k
) ].
Exercise 1.5.10. Given a k-place partial recursive function ψ(x
1
, . . . , x
k
), show
that there is a 1-place partial recursive function ψ

(x) such that
ψ

(p
x1
1
p
x
k
k
) ≃ p
ψ(x1,...,x
k
)
k+1
for all x
1
, . . . , x
k
, and ψ

is computable by a register machine program which
uses only two registers, R
1
and R
2
.
Solution. We begin with a register machine program{ which computes ψ(x
1
, . . . , x
k
).
Let P
1
, . . . , P
k
, P
k+1
, . . . , P
s
be the registers used in {. We may safely assume
that, whenever {(x
1
, . . . , x
k
) halts, it leaves all registers except P
k+1
empty.
We transform { into a program{ which uses only two registers, R
1
and R
2
.
The idea is that, if P
1
, . . . , P
s
contain z
1
, . . . , z
s
respectively, then R
1
contains
z = p
z1
1
p
zs
s
, while R
2
contains 0. Incrementing (decrementing) P
i
corre-
sponds to multiplication (division) by p
i
. Each instruction in { is replaced by
a corresponding set of instructions in {.
We replace

P
+
i
`
in { by Figure 1.9 in {.
We replace

P

i
`
e

?
?
?
?
?
?
?
A
B
in { by Figure 1.10 in {.
We replace

stop
`
in { by Figure 1.11 in {.
1.6 Consequences of the Enumeration Theorem
In this section we present some important consequences of the Enumeration
Theorem.
30

R

1
`

e

R

2
`

e

R
+
2
`

R
+
1
`

.
.
.

R
+
2
`

The number of R
+
2
instructions is p
i
.
Figure 1.9: Incrementing P
i
R
+
1
`

R

1
`

e

R

2
`

e

A R
+
1
`

R
+
2
`








R

1
`

e

R
+
1
`
R

2
`
e








R
+
1
`

.
.
.

.
.
.

B
.
.
.

R

1
`
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
e

R
+
1
`

R
+
1
`
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
The number of R

1
instructions is p
i
.
Figure 1.10: Decrementing P
i
31

R

1
`

e

stop
`
R
+
2
`

Figure 1.11: Stopping
Theorem 1.6.1. All partial recursive functions can be obtained from primitive
recursive functions by composition and minimization.
Proof. This is immediate from the proof of the Enumeration Theorem. The
State function is primitive recursive, the Stop function is obtained from the
State function by minimization, and the FinalState and Output functions are
obtained by composing the Stop and State functions with the primitive recursive
function λz [ (z)
k+1
].
The following characterizations of the class of partial recursive functions do
not involve register machines and are similar to our definition of the class of
primitive recursive functions.
Corollary 1.6.2. The class of partial recursive functions is the smallest class
of functions containing the primitive recursive functions and closed under com-
position and minimization.
Corollary 1.6.3. The class of partial recursive functions is the smallest class of
functions containing the initial functions and closed under composition, primi-
tive recursion, and minimization.
Proof. Both corollaries are immediate from the previous theorem and Lem-
mas 1.3.2, 1.3.3, 1.3.4, 1.4.3.
Next we present an interesting example showing that the consideration of
partial functions is in some sense unavoidable or inherent in recursive function
theory.
Example 1.6.4. We present an example of a partial recursive function ψ :
N
P
−→ N which cannot be extended to a total recursive function f : N → N.
Namely, we define
ψ(x) ≃ ϕ
(1)
x
(x) + 1 .
By the Enumeration Theorem, ψ is a partial recursive function. Suppose ψ
were extendible to a total recursive function f. Let e be an index of f. Then
ψ(e) ≃ ϕ
(1)
e
(e) + 1 ≃ f(e) + 1 is defined, hence f(e) ≃ ψ(e) ≃ f(e) + 1, a
contradiction.
32
Definition 1.6.5. A set A ⊆ N is said to be recursive if its characteristic
function χ
A
: N → N is recursive. More generally, a k-place predicate R ⊆ N
k
is said to be recursive if its characteristic function χ
R
: N
k
→N is recursive.
Example 1.6.6. We present an example of a nonrecursive set. Let K be
the subset of N consisting of all x ∈ N such that ϕ
(1)
x
(x) is defined. Thus
K = dom(ψ) where ψ is as in the previous example. We claim that K is not
recursive. If K were recursive, then the total function f : N →N defined by
f(x) =

ψ(x) if x ∈ K,
0 if x / ∈ K
would be recursive, contradicting the fact that ψ is not extendible to a total
recursive function.
Definition 1.6.7. A pair of sets A, B ⊆ N is said to be recursively inseparable
if there is no recursive set X such that A ⊆ X and X ∩ B = ∅.
Exercise 1.6.8. Letting K
n
= ¦x ∈ N [ ϕ
(1)
x
(x) ≃ n¦, show that K
0
and K
1
are recursively inseparable.
Exercise 1.6.9. Show that there exists a set A ⊆ N which is recursive but not
primitive recursive.
(Caution: It can be shown that the 3-place predicate z = A
x
(y) is prim-
itive recursive, even though the 2-place function λxy [ A
x
(y) ] is not primitive
recursive. See Exercises 1.2.2, 1.4.5, 1.5.8, 1.5.9.)
Remark 1.6.10 (Church’s Thesis). Perhaps the most important consequence
of the proof of the Enumeration Theorem is that it provides strong evidence for
Church’s Thesis. We shall first explain what Church’s Thesis says, and then we
shall present the evidence for it.
The context of Church’s Thesis is that, as mathematicians, we have an in-
tuitive notion of what it means for a function f : N
k
→N to be algorithmically
computable. Since recursive functions are register machine computable, they
are obviously algorithmically computable in the intuitive sense. Church’s The-
sis states the converse: All functions f : N
k
→ N which are algoritmically
computable in the intuitive sense are in fact recursive.
To present our evidence for Church’s thesis, assume that we are given a func-
tion f : N
k
→N which is algorithmically computable in the intuitive sense. We
want to show that f is recursive. Since the given algorithm for f is presumably
deterministic, the execution of the algorithm should be describable as a sequence
of states with deterministic transition from one state to the next. The precise
nature of the states depends on the nature of the algorithm, but no matter what
the states actually consist of, it should be possible to view them as finite strings
of symbols and to assign G¨odel numbers to them. Once this has been done, the
transition from the G¨odel number of one state to the G¨odel number of the next
state should be very simple, in particular primitive recursive. Thus we should
33
be able to carry out an analysis similar to what is in the proof of Theorem 1.5.5
(State, NextState, Stop, etc.). Such an analysis will show that f is obtained
from primitive recursive functions by means of composition and minimization.
It will then follow by Corollary 1.6.2 that f is recursive.
The argument in the previous paragraph is of necessity nonrigorous. How-
ever, it can be specialized to provide rigorous proofs that various models of com-
putation (similar to but differing in details from register machine computability)
give rise to exactly the same class of functions, the recursive functions. Some
of the models of computation that have been analyzed in this way are: Turing
machines, Markov algorithms, Kleene’s equation calculus. There is no reason
to think that the same analysis could not be carried out for any similar model.
This constitutes very strong evidence for Church’s thesis.
Note that the same arguments and evidence apply more generally in case f
is a partial rather than a total function. From now on, we shall take Church’s
Thesis for granted and identify the class of partial recursive functions with the
class of partial functions from N
k
into N that are algorithmically computable in
the intuitive sense.
The fact that the intuitive notion of algorithmic computability is captured
by a rigorous mathematical notion of recursiveness is one of the successes of
modern mathematical logic.
1.7 Unsolvable Problems
The purpose of this section is twofold: (1) to discuss and make precise the con-
cept of an unsolvable mathematical problem, and (2) to present some important
examples of such problems.
We begin with a preliminary clarification. In certain contexts, the word
problem refers to a mathematical statement which has a definite truth value,
True or False, but whose truth value is unknown at the present time. (An ex-
ample of a problem in this sense is the Riemann Hypothesis.) However, we shall
not deal with this type of problem now. Instead, we consider a somewhat dif-
ferent concept. For us in this section, a problem is any mathematical statement
that involves a parameter. A solution of such a problem would be an algorithm
which would enable us to compute the truth value of the problem statement for
any given value of the parameter. The problem is said to be solvable if there
exists such an algorithm, otherwise unsolvable. An instance of a problem is the
specialization of the problem statement to a particular parameter value.
As an example of a solvable problem, we mention:
Example 1.7.1. The statement “n is prime” represents the problem of deciding
whether an arbitrary number n ∈ N is prime or composite. Here the parameter
is the variable n. For any particular n (e.g. n = 123456789), the question of
whether this particular n is prime or composite is an instance (i.e., special case)
of the general “primality problem”. Since the set of prime numbers
¦n ∈ N [ n is prime¦
34
is primitive recursive, the general primality problem is solvable.
On the other hand, we have the following example of an unsolvable problem.
Example 1.7.2. The set K defined in Example 1.6.6 is nonrecursive. Hence by
Church’s Thesis there is no algorithm to decide whether or not a given number
n belongs to K. It is therefore appropriate to describe the membership problem
for K (i.e., the problem of computing the truth value of n ∈ K for any given n)
as an unsolvable problem.
The ability to distinguish solvable problems from unsolvable ones is of basic
importance for the mathematical enterprise. Among the most famous unsolvable
mathematical problems are:
Example 1.7.3 (Hilbert’s Tenth Problem). Hilbert’s Tenth Problem is to de-
termine, for a given polynomial p in several variables with integral coeffients,
p ∈ Z[X
1
, . . . , X
n
], whether or not the equation p(X
1
, . . . , X
n
) = 0 has a solu-
tion in integers X
1
, . . . , X
n
∈ N. This problem encompasses the entire theory
of Diophantine equations. A theorem of Matijaseviˇc shows that Hilbert’s Tenth
Problem is unsolvable. Actually Matijaseviˇc produced a particular polynomial
p(X
0
, X
1
, . . . , X
9
)
with 10 indeterminates, such that
¦n ∈ N [ p(n, a
1
, . . . , a
9
) = 0 for some a
1
, . . . , a
9
∈ Z¦
is nonrecursive. Once again, our notion of unsolvable problem is related to the
existence of a nonrecursive set, namely the set of parameter values n for which
the problem statement holds.
Example 1.7.4 (Word Problems). Let G be a group presented by finitely many
generators and relations. The word problem for G is the problem of determining,
for a given word w in the generators of G and their inverses, whether or not
w = 1 in G. In this case the parameter is w, and the word problem for G is
solvable if and only if there exists an algorithm for determining whether or not a
given word w is equal to 1 in G. It is known that the word problem is solvable for
some groups G and not solvable for others. For example, the word problem for
free groups or groups with one relation is solvable, but Boone and Novikov have
exhibited groups G with finitely many relations such that the word problem for
G is unsolvable.
Example 1.7.5 (The Halting Problem).
Some famous unsolvable problems arise from computability theory itself. One
of these is the Halting Problem: To determine whether or not a given register
machine program { will eventually stop, if started with all registers empty. By
G¨odel numbering, we can identify the Halting Problem with the problem of
deciding whether a given natural number e belongs to the set
H = ¦e ∈ N [ ϕ
(1)
e
(0) is defined¦ .
35
We shall prove below that H is nonrecursive, i.e., the Halting Problem is un-
solvable.
In all of the above examples, the issue of solvability or unsolvability of a
particular problem was rephrased as an issue of whether or not a particular
subset of N is recursive. Such considerations based on Church’s Thesis motivate
the following definition:
Definition 1.7.6 (Unsolvability). Recall that a set A ⊆ N is said to be recursive
if and only if its characteristic function χ
A
: N → ¦0, 1¦ is recursive. A problem
is defined to be a subset of N. If A ⊆ N is a problem in this sense, the problem
A is said to be solvable if A is recursive, and unsolvable if A is nonrecursive.
We shall prove the unsolvability of the Halting Problem, i.e., the nonrecur-
siveness of the set H in Example 1.7.5 above. The proof will be accomplished
by showing that the problem of membership in K is “reducible” to the problem
of membership in H. In this context, reducibility of one problem to another
means that each instance of the former problem can be effectively converted to
an equivalent instance of the latter problem. Our precise notion of reducibility
is given by:
Definition 1.7.7 (Reducibility). Let A and B be subsets of N (i.e., problems,
cf. Definition 1.7.6). We say that A is reducible to B if there exists a recursive
function f : N →N such that, for all n ∈ N, n ∈ A implies f(n) ∈ B, and n / ∈ A
implies f(n) / ∈ B.
Lemma 1.7.8. Suppose that A is reducible to B. If B is recursive, then A is
recursive. If A is nonrecursive, then B is nonrecursive.
Proof. The first statement follows easily from the fact that χ
A
(x) = χ
B
(f(x)).
The second statement follows since it is the contrapositive of the first.
Exercise 1.7.9. Write A ≤
m
B to mean that A is reducible to B. Show that
1. A ≤
m
A for all A ⊆ N.
2. A ≤
m
B and B ≤
m
C imply A ≤
m
C.
3. If ∅ =B= N, then for every recursive set A we have A ≤
m
B.
In order to prove that the set H is nonrecursive, we shall need the following
important technical result:
Theorem 1.7.10 (The Parametrization Theorem). Let θ(x
0
, x
1
, . . . , x
k
) be a
(k+1)-ary partial recursive function. Then we can find a unary primitive recur-
sive function f(x
0
) such that, for all x
0
, x
1
, . . . , x
k
∈ N,
ϕ
(k)
f(x0)
(x
1
, . . . , x
k
) ≃ θ(x
0
, x
1
, . . . , x
k
) .
36
Proof. Let T be a register machine programwhich computes the k+1-ary partial
recursive function θ. The idea of the proof is to let f(x
0
) be the G¨odel number
of a program which is similar to T but has x
0
hard-coded as the first argument
of θ.
Formally, let T

be a register machine program which computes the k+1-ary
partial recursive function
θ

(x
1
, . . . , x
k
, x
0
) ≃ θ(x
0
, x
1
, . . . , x
k
) .
Let I
1
, . . . , I
l
be the instructions of T

, and let T
′′
be the same as T

but modified
so that the instructions are numbered I
5
, . . . , I
l+4
instead of I
1
, . . . , I
l
. Then for
any given x
0
∈ N, the program T
′′′
x0
depicted in Figure 1.12 computes the k-ary
partial recursive function λx
1
. . . x
k
[ θ

(x
1
, . . . , x
k
, x
0
) ], i.e. λx
1
. . . x
k
[ θ(x
0
, x
1
, . . . , x
k
) ].
The instructions of T
′′′
x0
are numbered as I
1
, . . . , I
l+x0+5
. Let f(x
0
) be the G¨odel
number of T
′′′
x0
. Note that
f(x
0
) =
¸
l+x0+5
m=1
p
#(Im)
m
=
¸
l+4
m=1
p
#(Im)
m

¸
l+x0+4
m=l+5
p
3
k+1
·5
m+1
m
p
2·3
k+1
·5
5
·7
5
l+x0+5
where the first factor
¸
l+4
m=1
p
#(Im)
m
does not depend on x
0
. Thus f(x
0
) is a
primitive recursive function of x
0
. This completes the proof.
I
1
I
l+5
. . .
I
l+x0+4
I
l+x0+5
I
5
I
l+4
I
2
I
3
start
`
R
+
k+1
\
R
+
k+1
\ . . .
R
+
k+1
\
R

k+1
\

e

T
′′

R

k+1
\

e

R

k+2
\

e

stop
`
x
0 R
+
k+1
\

I
4
Figure 1.12: Parametrization
We can now prove that the Halting Problem is unsolvable.
Theorem 1.7.11 (Unsolvability of the Halting Problem). The Halting Problem
is unsolvable. In other words, the set
H = ¦x ∈ N [ ϕ
(1)
x
(0) is defined¦
of Example 1.7.5 is nonrecursive.
37
Proof. Let H be as in Example 1.7.5 and let K be as in Example 1.6.6, i.e.,
K = ¦x ∈ N [ ϕ
(1)
x
(x) is defined¦ .
We shall prove that K is reducible to H. Since K is known to be nonrecursive
(Example 1.6.6), it will follow by Lemma 1.7.8 that H is nonrecursive.
Consider the partial recursive function θ(x, y) ≃ ϕ
(1)
x
(x). Note that θ is a
2-place function. By the Enumeration Theorem, θ is partial recursive. By the
Parametrization Theorem applied with k = 1, we can find a primitive recursive
function f : N →N such that
ϕ
(1)
f(x)
(y) ≃ θ(x, y) ,
i.e.,
ϕ
(1)
f(x)
(y) ≃ ϕ
(1)
x
(x)
for all x and y. In particular, if x ∈ K then ϕ
(1)
x
(x) is defined, hence ϕ
(1)
f(x)
(0) is
defined, i.e., f(x) ∈ H. On the other hand, if x / ∈ K then ϕ
(1)
x
(x) is undefined,
hence ϕ
(1)
f(x)
(0) is undefined, i.e., f(x) / ∈ H. Thus K is reducible to H via f.
This completes the proof.
Exercise 1.7.12. Show that the following sets and predicates are nonrecursive:
1. ¦x ∈ N [ ϕ
(1)
x
: N
P
−→N is total¦.
2. ¦x ∈ N [ ϕ
(1)
x
is the empty function¦.
3. ¦'x, y` ∈ N N [ ϕ
(1)
x
= ϕ
(1)
y
¦.
4. ¦'x, y` ∈ N N [ y ∈ rng(ϕ
(1)
x
)¦.
5. ¦x ∈ N [ 0 ∈ rng(ϕ
(1)
x
)¦.
6. ¦x ∈ N [ rng(ϕ
(1)
x
) is infinite¦.
Exercise 1.7.13 (Rice’s Theorem). Let { be the class of 1-place partial recur-
sive functions. For ( ⊆ {, define I
C
to be the set of indices of functions in (,
i.e.,
I
C
= ¦x ∈ N [ ϕ
(1)
x
∈ (¦ .
Show that if ∅ = ( = { then I
C
is nonrecursive.
Solution. Let e
0
be an index of the empty function. Let e
1
be an index such
that ϕ
(1)
e1
∈ ( if and only if ϕ
(1)
e0
/ ∈ (. By the Enumeration and Parametrization
theorems, we can find a primitive recursive function f such that
ϕ
(1)
f(x)
(y) ≃

ϕ
(1)
e1
(y) if ϕ
(1)
x
(x) ↓,
↑ otherwise,
38
for all x and y. Thus x ∈ K implies ϕ
(1)
f(x)
= ϕ
(1)
e1
, while x / ∈ K implies
ϕ
(1)
f(x)
= ϕ
(1)
e0
. Thus f reduces K either to I
C
(if ϕ
(1)
e1
∈ () or to the complement
of I
C
(if ϕ
(1)
e1
/ ∈ (). In either case it follows that I
C
is not recursive.
1.8 The Recursion Theorem
In this section we present an interesting and mysterious theorem known as the
Recursion Theorem.
Theorem 1.8.1 (The Recursion Theorem). Let θ(w, x
1
, . . . , x
k
) be a partial
recursive function. Then we can find an index e such that, for all x
1
, . . . , x
k
,
ϕ
(k)
e
(x
1
, . . . , x
k
) ≃ θ(e, x
1
, . . . , x
k
) .
Proof. Applying the Parametrization Theorem and the Enumeration Theorem,
we can find primitive recursive functions f and d such that, for all w, u, x
1
, . . . , x
k
,
ϕ
(k)
f(w)
(x
1
, . . . , x
k
) ≃ θ(w, x
1
, . . . , x
k
) .
and
ϕ
(k)
d(u)
(x
1
, . . . , x
k
) ≃ ϕ
(k)
ϕ
(1)
u
(u)
(x
1
, . . . , x
k
) .
Let v be an index of f ◦ d, i.e., ϕ
(1)
v
(u) = f(d(u)) for all u. Then
ϕ
(k)
d(v)
(x
1
, . . . , x
k
) ≃ ϕ
(k)
ϕ
(1)
v
(v)
(x
1
, . . . , x
k
)
≃ ϕ
(k)
f(d(v))
(x
1
, . . . , x
k
)
≃ θ(d(v), x
1
, . . . , x
k
)
so we may take e = d(v). This completes the proof.
Example 1.8.2. As an example, if we take θ(w, x) = w+x, then we obtain an
index e such that ϕ
(1)
e
(x) = e +x for all x.
Example 1.8.3. As another illustration of the Recursion Theorem, we now
use it to prove that the Ackermann function λnx[ A
n
(x) ] (see Section 1.2) is
computable. The recursion equations defining the Ackermann function can be
written as
A
0
(x) = 2x
A
n+1
(0) = 1
A
n+1
(x + 1) = A
n
(A
n+1
(x)) .
Writing A(x, y) = A
x
(y), this becomes
A(0, y) = 2y
A(x + 1, 0) = 1
A(x + 1, y + 1) = A(x, A(x + 1, y))
39
or in other words
A(x, y) =

2y if x = 0 ,
1 if x > 0 and y = 0 ,
A(x

−1, A(x, y

−1)) if x > 0 and y > 0 .
By the Enumeration Theorem together with the Recursion Theorem, we can
find an index e such that
ϕ
(2)
e
(x, y) ≃

2y if x = 0 ,
1 if x > 0 and y = 0 ,
ϕ
(2)
e
(x

−1, ϕ
(2)
e
(x, y

−1)) if x > 0 and y > 0 .
It is then straightforward to prove by induction on x that, for all y, ϕ
(2)
e
(x, y) ≃
A(x, y). This completes the proof.
Exercise 1.8.4.
1. Find a primitive recursive function f(x, y) such that, for all x and y,
ϕ
(1)
f(x,y)
= ϕ
(1)
x
◦ ϕ
(1)
y
.
2. Find a primitive recursive function g(x, y) such that, for all x and y,
dom(ϕ
(1)
g(x,y)
) = dom(ϕ
(1)
x
) ∩ dom(ϕ
(1)
y
) .
3. Find a primitive recursive function h(x, y) such that, for all x and y,
dom(ϕ
(1)
h(x,y)
) = dom(ϕ
(1)
x
) ∪ dom(ϕ
(1)
y
) .
Solution.
1. By the Enumeration Theorem and the Parametrization Theorem, find a
primitive recursive function
´
f(w) such that
ϕ
(1)
b
f(w)
(z) ≃ ϕ
(1)
(w)1

(1)
(w)2
(z))
for all w, z. Then f(x, y) =
´
f(3
x
5
y
) has the desired property.
2. By the Enumeration Theorem and the Parametrization Theorem, find a
primitive recursive function ´g(w) such that
ϕ
(1)
bg(w)
(z) ≃ ϕ
(1)
(w)1
(z) +ϕ
(1)
(w)2
(z)
for all w, z. Then g(x, y) = ´ g(3
x
5
y
) has the desired property.
40
3. By the Parametrization Theorem, find a primitive recursive function
´
h(w)
such that
ϕ
(1)
b
h(w)
(z) ≃ least n such that (State((w)
1
, z, n))
0
(State((w)
2
, z, n))
0
= 0
for all w, z. Then h(x, y) =
´
h(3
x
5
y
) has the desired property.
Exercise 1.8.5.
1. Find a primitive recursive function f(x) such that for all x, if ϕ
(1)
x
is a
permutation of N, then ϕ
(1)
f(x)
is the inverse permutation.
2. What happens if ϕ
(1)
x
is assumed only to be partial and one-to-one, and
not necessarily a permutation?
Solution. Consider the partial recursive function θ(x, y) ≃ least w such that
(State(x, (w)
1
, (w)
2
))
0
= 0 and (State(x, (w)
1
, (w)
2
))
2
= y. By construction, if
ϕ
(1)
x
(z) ≃ y then (θ(x, y))
1
≃ z. Therefore, by the Parametrization Theorem,
let f(x) be a primitive recursive function such that ϕ
(1)
f(x)
(y) ≃ (θ(x, y))
1
. This
works even if ϕ
(1)
x
is only assumed to be partial and one-to-one.
Exercise 1.8.6. Find m and n such that m = n and ϕ
(1)
m
(0) = n and ϕ
(1)
n
(0) =
m.
Solution. By the Parametrization Theorem, let f be a 1-place primitive recursive
function such that ϕ
(1)
f(x)
(y) = x for all x, y. The construction of f in the proof of
the Parametrization Theorem shows that f(x) > x for all x. By the Recursion
Theorem, let e be such that ϕ
(1)
e
(y) ≃ f(e) for all y. In particular we have
ϕ
(1)
e
(0) ≃ f(e), ϕ
(1)
f(e)
(0) = e, and f(e) > e. So take m = e and n = f(e).
1.9 The Arithmetical Hierarchy
In this section we study some important classes of number-theoretic predicates:
Σ
0
1
, Π
0
1
, Σ
0
2
, Π
0
2
, . . . . These classes collectively are known as the arithmetical
hierarchy.
Definition 1.9.1 (The Arithmetical Hierarchy). We define Σ
0
0
and Π
0
0
to be
the class of primitive recursive predicates. For n ≥ 1, we define Σ
0
n
to be the
class of k-place predicates P ⊆ N
k
(for any k ≥ 1) such that P can be written
in the form
P(x
1
, . . . , x
k
) ≡ ∃y R(x
1
, . . . , x
k
, y)
where R is a k+1-place predicate belonging to the class Π
0
n−1
. Similarly, we
define Π
0
n
to be the class of predicates that can be written in the form
P(x
1
, . . . , x
k
) ≡ ∀y R(x
1
, . . . , x
k
, y)
where R belongs to the class Σ
0
n−1
.
41
For example, a predicate P(x
1
, . . . , x
k
) belongs to the class Σ
0
3
if and only if
it can be written in the form
∃y
1
∀y
2
∃y
3
R(x
1
, . . . , x
k
, y
1
, y
2
, y
3
)
where R is primitive recursive. Similarly, P belongs to the class Π
0
3
if and only
if it can be written in the form
∀y
1
∃y
2
∀y
3
R(x
1
, . . . , x
k
, y
1
, y
2
, y
3
)
where R is primitive recursive.
Theorem 1.9.2. We have:
1. The classes Σ
0
n
and Π
0
n
are included in the classes Σ
0
n+1
and Π
0
n+1
.
2. The classes Σ
0
n
and Π
0
n
are closed under conjunction and disjunction.
3. The classes Σ
0
n
and Π
0
n
are closed under bounded quantification.
4. For n ≥ 1, the class Σ
0
n
is closed under existential quantification.
5. For n ≥ 1, the class Π
0
n
is closed under universal quantification.
6. A predicate P belongs to Σ
0
n
(respectively Π
0
n
) if and only if P belongs
to Π
0
n
(respectively Σ
0
n
).
Proof. Straightforward.
Theorem 1.9.3. A k-place predicate P ⊆ N
k
belongs to the class Σ
0
1
if and
only if P = dom(ψ) for some k-place partial recursive function ψ : N
k
P
−→N.
Proof. If P is Σ
0
1
, then we have
P(x
1
, . . . , x
k
) ≡ ∃y R(x
1
, . . . , x
k
, y)
where R is primitive recursive, hence P = dom(ψ) where
ψ(x
1
, . . . , x
k
) ≃ least y such that R(x
1
, . . . , x
k
, y) holds ,
and clearly ψ is partial recursive. Conversely, if P = dom(ψ), then letting e be
an index of ψ, we have as in the proof of the Enumeration Theorem
P(x
1
. . . , x
k
) ≡ ∃n(State(e, x
1
, . . . , x
k
, n))
0
= 0 ,
hence P is Σ
0
1
.
Corollary 1.9.4. P ⊆ N
k
is Σ
0
1
if and only if
P(x
1
, . . . , x
k
) ≡ ∃y R(x
1
, . . . , x
k
, y)
where R ⊆ N
k+1
is recursive (not only primitive recursive).
42
Remark 1.9.5. It follows that, in the definition of Σ
0
n
and Π
0
n
for n ≥ 1, we
may replace “primitive recursive” by “recursive”.
Exercises 1.9.6. If ψ is a k-place partial function, the graph of ψ is the (k+1)-
place predicate G
ψ
= ¦'x
1
, . . . , x
k
, y` [ ψ(x
1
, . . . , x
k
) ≃ y¦.
1. Show that ψ is partial recursive if and only if the graph of ψ is Σ
0
1
.
2. Show that, for every (k+1)-place Σ
0
1
predicate P(x
1
, . . . , x
k
, y) there exists
a k-place partial recursive function ψ(x
1
, . . . , x
k
) which uniformizes P, i.e.,
G
ψ
⊆ P and dom(ψ) = ¦'x
1
, . . . , x
k
` [ ∃y P(x
1
, . . . , x
k
, y)¦.
Definition 1.9.7. Let A be an infinite subset of N. The principal function of
A is the one-to-one function π
A
: N →N that enumerates the elements of A in
increasing order.
Lemma 1.9.8. Let A be an infinite subset of N. The set A is recursive if and
only if the function π
A
is recursive.
Proof. If A is recursive then the functions ν
A
and π
A
defined by
ν
A
(x) = least y such that y ≥ x and y ∈ A
π
A
(0) = ν
A
(0)
π
A
(x + 1) = ν
A

A
(x) + 1)
are obviously recursive. Conversely, if π
A
is recursive then we have
y ∈ A if and only if
y
¸
x=0
π
A
(x) = y .
Since the class of recursive predicates is closed under bounded quantification, it
follows that A is recursive.
Theorem 1.9.9. For A ⊆ N, the following are pairwise equivalent:
1. A is Σ
0
1
;
2. A = dom(ψ) for some partial recursive function ψ : N
P
−→N;
3. A = rng(ψ) for some partial recursive function ψ : N
P
−→N;
4. A = ∅ or A = rng(f) for some total recursive function f : N →N;
5. A is finite or A = rng(f) for some one-to-one total recursive function
f : N →N.
43
Proof. The equivalence of 1 and 2 is a special case of Theorem 1.9.3, and the
fact that 3 implies 1 is proved similarly. It is easy to see that 5 implies 4 and 4
implies 3. To see that 1 implies 5, suppose that A ⊆ N is infinite and Σ
0
1
, say
x ∈ A ≡ ∃yR(x, y)
where R is primitive recursive. Put
B =

2
x
3
y

R(x, y) ∧
y−1
¸
z=0
R(x, z)
¸
.
Then B is an infinite primitive recursive set, so by Lemma 1.9.8, π
B
is a one-to-
one recursive function. Putting f(n) = (π
B
(n))
0
, we see that f is a one-to-one
recursive function and A = rng(f). This completes the proof.
Remark 1.9.10. A set A satisfying the conditions of Theorem 1.9.9 is some-
times called a recursively enumerable set.
Exercises 1.9.11.
1. Show that every nonempty recursively enumerable set is the range of a
primitive recursive function.
2. Find an infinite primitive recursive set that is not the range of a one-to-one
primitive recursive function.
Definition 1.9.12. For each n ∈ N, the class ∆
0
n
is defined to be the intersection
of the classes Σ
0
n
and Π
0
n
.
Theorem 1.9.13. A k-place predicate P ⊆ N
k
belongs to the class ∆
0
1
if and
only if P is recursive.
Proof. If P is recursive, it follows easily by Theorem 1.9.3 that P is ∆
0
1
. Con-
versely, if P is ∆
0
1
, then we have
P(x
1
, . . . , x
k
) ≡ ∃y R
1
(x
1
, . . . , x
k
, y) ≡ ∀y R
2
(x
1
, . . . , x
k
, y)
where R
1
and R
2
are primitive recursive. Define a total recursive function f by
f(x
1
, . . . , x
k
) = least y such that R
1
(x
1
, . . . , x
k
, y) ∨ R
2
(x
1
, . . . , x
k
, y) .
Then we have
P(x
1
, . . . , x
k
) ≡ R
1
(x
1
, . . . , x
k
, f(x
1
, . . . , x
k
)) ,
hence P is recursive.
44
Exercise 1.9.14. A function f : N
k
→ N is said to be limit-recursive if there
exists a recursive function g : N
k+1
→N such that
f(x
1
, . . . , x
k
) = lim
y
g(x
1
, . . . , x
k
, y)
for all x
1
, . . . , x
k
∈ N. Show that a predicate P is ∆
0
2
if and only if its charac-
teristic function χ
P
is limit-recursive.
Solution. For simplicity, let x be an abbreviation for x
1
, . . . , x
k
.
First assume that χ
P
is limit-recursive, say
χ
P
(x) = lim
n
f(n, x)
for all x, where f(n, x) is a recursive function. Then we have
P(x) ≡ ∃m∀n(n ≥ m ⇒ f(n, x) = 1)
and
P(x) ≡ ∃m∀n(n ≥ m ⇒ f(n, x) = 0)
so P is ∆
0
2
.
For the converse, assume that P is ∆
0
2
, say
P(x) ≡ ∃y ∀z R
1
(x, y, z)
and
P(x) ≡ ∃y ∀z R
0
(x, y, z)
where R
1
and R
0
are primitive recursive predicates. Using the bounded least
number operator, define g(n, x) = the least y < n such that either ∀z <
nR
1
(x, y, z) or ∀z < nR
0
(x, y, z) or both, if such a y exists, and g(n, x) =
n otherwise. Thus g(n, x) is a primitive recursive function, and it is easy
to see that, for all x, g(x) = lim
n
g(n, x) exists and is equal to the least y
such that ∀z R
1
(x, y, z) or ∀z R
0
(x, y, z). Now define h(n, x) = 1 if ∀z <
nR
1
(x, g(n, x), z), and h(n, x) = 0 otherwise. Thus h(n, x) is again a primi-
tive recursive function, and for all x, h(x) = lim
n
h(n, x) exists. Moreover P(x)
implies h(x) = 1, and P(x) implies h(x) = 0. Thus χ
P
is limit-recursive. This
completes the proof.
Theorem 1.9.15 (Universal Σ
0
n
Predicate). For each n ≥ 1 and k ≥ 1, we can
find a predicate U = U
n,k
with the following properties:
1. U is a k+1-place predicate belonging to the class Σ
0
n
; and
2. for any k-place predicate P belonging to the class Σ
0
n
, there exists an e ∈ N
such that
P(x
1
, . . . , x
k
) ≡ U(e, x
1
, . . . , x
k
)
for all x
1
, . . . , x
k
.
45
Proof. This is a straightforward consequence of the Enumeration Theorem. The
proof is by induction on n. For n = 1 we have
U
1,k
(e, x
1
, . . . , x
k
) ≡ ϕ
(k)
e
(x
1
, . . . , x
k
) ↓ ≡ ∃s (State(e, x
1
, . . . , x
k
, s))
0
= 0
in view of Theorem 1.9.3. For n > 1 we have
U
n,k
(e, x
1
, . . . , x
k
) ≡ ∃y U
n−1,k+1
(e, x
1
, . . . , x
k
, y) .
This completes the proof.
Lemma 1.9.16. Let A, B ⊆ N and assume that A is reducible to B. Assume
n ≥ 1. If B belongs to Σ
0
n
, then so does A. If B belongs to Π
0
n
, then so does A.
Proof. Suppose for example that B belongs to Σ
0
3
. Then we have
x ∈ B if and only if ∃y
1
∀y
2
∃y
3
R(x, y
1
, y
2
, y
3
)
where R ⊆ N
4
is a primitive recursive predicate. If A is reducible to B via the
recursive function f, then we have
x ∈ A if and only if f(x) ∈ B
if and only if ∃y
1
∀y
2
∃y
3
R(f(x), y
1
, y
2
, y
3
) .
Note that the predicate R(f(x), y
1
, y
2
, y
3
) is recursive, hence ∆
0
1
. It follows that
the predicate ∃y
3
R(f(x), y
1
, y
2
, y
3
) is Σ
0
1
. Hence A is Σ
0
3
.
Definition 1.9.17. Given n ≥ 1, a set B ⊆ N is said to be complete Σ
0
n
if
1. B belongs to the class Σ
0
n
; and
2. for any set A ⊆ N belonging to the class Σ
0
n
, A is reducible to B.
The notion of complete Π
0
n
set is defined similarly.
Theorem 1.9.18. For each n ≥ 1 there exists a complete Σ
0
n
set. For each
n ≥ 1 there exists a complete Π
0
n
set. For each n ≥ 1, a complete Σ
0
n
set is not
Π
0
n
, and a complete Π
0
n
set is not Σ
0
n
.
Proof. By Theorem 1.9.15 let U(e, x) be a universal Σ
0
n
predicate. Obviously
the set ¦2
e
3
x
[ U(e, x)¦ is Σ
0
n
complete. To show that a Σ
0
n
complete set can
never be Π
0
n
, it suffices by Lemma 1.9.16 to show that there exists a Σ
0
n
set which
is not Π
0
n
. A simple diagonal argument shows that the Σ
0
n
set ¦x [ U(x, x)¦ is
not Π
0
n
. This completes the proof of the Σ
0
n
part of the theorem. The Π
0
n
part
follows easily by taking complements.
Corollary 1.9.19. For all n ∈ N we have

0
n
⊆ Σ
0
n
, ∆
0
n
⊆ Π
0
n
, Σ
0
n
∪ Π
0
n
⊆ ∆
0
n+1
and, except for n = 0, all of these inclusions are proper.
46
Proof. Given n ≥ 1, let A be a complete Σ
0
n
set. Then the complement B = N`A
is complete Π
0
n
. Clearly A belongs to Σ
0
n
` ∆
0
n
and B belongs to Π
0
n
` ∆
0
n
.
Moreover it follows by Theorem 1.9.2 and Lemma 1.9.16 that the set
A ⊕B = ¦2x [ x ∈ A¦ ∪ ¦2x + 1 [ x ∈ B¦
belongs to ∆
0
n+1
` (Σ
0
n
∪ Π
0
n
).
Exercises 1.9.20.
1. Show that the sets
H = ¦x [ ϕ
(1)
x
(0) is defined¦
and
K = ¦x [ ϕ
(1)
x
(x) is defined¦
are complete Σ
0
1
sets.
2. Show that the set
T = ¦x [ ϕ
(1)
x
is total¦
is a complete Π
0
2
set.
Exercise 1.9.21. What reducibility and non-reducibility relations exist among
the following sets? Note that H, T, E, and S are index sets.
K = ¦x ∈ N [ ϕ
(1)
x
(x) ↓¦,
H = ¦x ∈ N [ ϕ
(1)
x
(0) ↓¦,
T = ¦x ∈ N [ ϕ
(1)
x
is total¦,
E = ¦x ∈ N [ ϕ
(1)
x
is the empty function¦,
S = ¦x ∈ N [ dom(ϕ
(1)
x
) is infinite¦.
Prove your answers.
Hint: Using the Parametrization Theorem as in the proof of Theorem 1.7.11,
show that H and K are Σ
0
1
complete, S and T are Π
0
2
complete, and E is Π
0
1
complete. These completeness facts, together with Lemma 1.9.16 and Theorem
1.9.18, determine the reducibility relations among H, K, S, T, and E.
47
Chapter 2
Undecidability of
Arithmetic
We are going to show that arithmetic is undecidable. This means that there is
no algorithm to decide whether a given sentence in the language of arithmetic
is true or false.
2.1 Terms, Formulas, and Sentences
By arithmetic we mean the set of sentences which are true in the structure
(N, +, , 0, 1, =). Here N = ¦0, 1, 2, . . .¦ is the set of non-negative integers, +
and denote the 2-place operations of addition and multiplication on N, and =
denotes the 2-place relation of equality between elements of N. For the benefit
of the reader who has not previously studied mathematical logic, we shall now
review the concept of a sentence being true in a structure. Since we are only
interested in the particular structure (N, +, , 0, 1, =), we shall concentrate on
that case.
It is first necessary to define a language appropriate for the structure
(N, +, , 0, 1, =).
Our language contains infinitely many variables x
0
, x
1
, . . . , x
n
, . . . which are
usually denoted by letters such as x, y, z, . . . . Each of these variables is also a
term of our language. In addition the symbols 0 and 1 are terms. Other terms
are built up using the 2-place operation symbols + and . Examples of terms
are
1 + 1 + 1 , x + 1 , (x +y) z +x.
When writing terms, we may employ the usual abbreviations. For instance
1 + 1 + 1 = 3 and x x x = x
3
. Thus a term is essentially a polynomial in
several variables with non-negative integer coefficients.
48
An atomic formula is a formula of the form t
1
= t
2
where t
1
and t
2
are
terms. Examples of atomic formulas are
x + 1 = y , x = 3y
2
+ 1 .
Formulas are built from atomic formulas by using the Boolean propositional
connectives ∧, ∨, (and, or, not) and the quantifiers ∀, ∃ (for all, there exists).
An example of a formula is
∃z (x +z + 1 = y) (2.1)
In this formula, x, y and z are to be interpreted as variables ranging over the
set N. Similarly the expression ∃z . . . is to be interpreted as “there exists a
non-negative integer z in N such that . . . .” Thus the formula (2.1) expresses
the assertion that x is less than y. From now on we shall write x < y as
an abbreviation for (2.1). This idea of introducing new relation symbols as
abbreviations for formulas allows us to expand our language indefinitely.
Another example of a formula is
x > 1 ∧ ∃y ∃z (y > 1 ∧ z > 1 ∧ x = y z) . (2.2)
This formula expresses the assertion that x is a prime number. Thus we might
choose to abbreviate (2.2) by some expression such as Prime(x) or “x is prime.”
In any particular formula, a free variable is a variable which is not acted on
by any quantifier in that formula. For example, the free variables of (2.1) are x
and y, while in (2.2) the only free variable is x. A sentence is a formula with
no free variables. Examples of sentences are
∀x(∃y (x = 2y) ∨ ∃y (x = 2y + 1)) (2.3)
and
∀x∃y (x = y + 1) . (2.4)
When interpreting formulas or sentences in the structure (N, +, , 0, 1, =),
please bear in mind that the quantifiers ∀ and ∃ range over the set of non-
negative integers, N. Thus ∀x means “for all x in N,” and ∃x means “for some
x in N” or “there exists x in N such that . . . .” With this understanding, we
know what it means for a sentence of our language to be true or false in the
structure (N, +, , 0, 1, =). For example (2.3) is true (because every element of
N is either even or odd) and (2.4) is false (because not every element of N is the
successor of some other element of N).
Let F(x
1
, . . . , x
k
) be a formula whose free variables are x
1
, . . . , x
k
. Then for
any non-negative integers a
1
, . . . , a
k
∈ N, we can form the sentence F(a
1
, . . . , a
k
)
which is obtained by substituting (“plugging in”) the constants a
1
, . . . , a
k
for
the free occurrences of the variables x
1
, . . . , x
k
. For example, let F(x, y) be the
formula
∃z (x
2
+z = y)
49
with free variables x and y. Then for any particular non-negative integers m
and n, F(m, n) is a sentence which expresses the assertion that m
2
is less than
or equal to n. For instance F(5, 40) is true and F(5, 20) is false.
This completes our review of the concept of a sentence being true or false in
the structure (N, +, , 0, 1, =).
Exercise 2.1.1. Write a sentence expressing Goldbach’s Conjecture: Every
even number is the sum of two prime numbers.
2.2 Arithmetical Definability
We now present a key definition.
Definition 2.2.1 (Arithmetical Definability). A k-place partial function
λx
1
x
k
[ ψ(x
1
, . . . , x
k
) ]
is said to be arithmetically definable if there exists a formula
F(x
1
, . . . , x
k
, x
k+1
)
with free variables x
1
, . . . , x
k
, x
k+1
, such that for all m
1
, . . . , m
k
, n ∈ N,
ψ(m
1
, . . . , m
k
) ≃ n if and only if F(m
1
, ..., m
k
, n) is true.
(Of course it doesn’t matter whether we use the variables x
1
, . . . , x
k
, x
k+1
or
some other set of k + 1 distinct variables.)
For example, the 1-place function λx[ 2x] is arithmetically definable, by the
formula y = 2x. (Here we are using a formula with two free variables x and y.)
As another example, note that the 2-place functions λxy [ Quotient(y, x) ] and
λxy [ Remainder(y, x) ] (the quotient and remainder of y on division by x) are
arithmetically definable, by the formulas
∃u ∃v (y = u x +v ∧ v < x ∧ z = u)
and
∃u ∃v (y = u x +v ∧ v < x ∧ z = v)
respectively. (Here we are using formulas with free variables x, y, and z.)
Exercise 2.2.2. Show that the function
λxy [ least common multiple of x and y ]
is arithmetically definable.
50
Remark 2.2.3. A more “mathematical” characterization of arithmetical de-
finability, not involving formulas, may be given as follows. Let us say that
a predicate P ⊆ N
k
is strongly Diophantine if there exists a polynomial with
integer coefficients, f(x
1
, . . . , x
k
) ∈ Z[x
1
, . . . , x
k
], such that
P = ¦'a
1
, . . . , a
k
` ∈ N
k
[ f(a
1
, . . . , a
k
) = 0¦.
For Q ⊆ N
k+1
, the projection of Q is given as
π(Q) = ¦'a
1
, . . . , a
k
` ∈ N
k
[ 'a
1
, . . . , a
k
, a
k+1
` ∈ Q for some a
k+1
∈ N¦ .
Then, the arithmetically definable predicates may be characterized as the small-
est class of number-theoretic predicates which contains all strongly Diophantine
predicates and is closed under union, complementation, and projection. Clearly
each arithmetically definable predicate is obtained by applying these operations
only a finite number of times.
Exercise 2.2.4. Show that the following number-theoretic predicates are arith-
metically definable, by exhibiting formulas which define them over the structure
(N, +, , 0, 1, =).
1. GCD(x, y) = z.
2. LCM(x, y) = z.
3. Quotient(x, y) = z.
4. Remainder(x, y) = z.
5. x is the largest prime number less than y.
6. x is the product of all the prime numbers less than y.
Remark 2.2.5. The principal result of this section, Theorem 2.2.21 below, is
that all recursive functions and predicates are arithmetically definable. From
this it will follow easily that there exists an arithmetically definable predicate
which is not recursive (Corollary 2.2.22 below). In addition, we shall character-
ize the class of arithmetically definable predicates in terms of the arithmetical
hierarchy (Theorem 2.2.23 below).
Lemma 2.2.6. The initial functions are arithmetically definable.
Proof. The constant zero function λx[ 0 ] is defined by the formula
x = x ∧ y = 0 .
The successor function λx[ x + 1 ] is defined by the formula
y = x + 1 .
For 1 ≤ i ≤ k, the projection function λx
1
. . . x
k
[ x
i
] is defined by the formula
x
1
= x
1
∧ . . . ∧ x
k
= x
k
∧ y = x
i
.
This completes the proof.
51
Lemma 2.2.7. If the functions g
1
(x
1
, . . . , x
k
), . . . , g
m
(x
1
, . . . , x
k
) and h(y
1
, . . . , y
m
)
are arithmetically definable, then so is the function
f(x
1
, . . . , x
k
) = h(g
1
(x
1
, . . . , x
k
), . . . , g
m
(x
1
, . . . , x
k
))
obtained by composition.
Proof. Let g
1
(x
1
, . . . , x
k
), . . . , g
m
(x
1
, . . . , x
k
), h(y
1
, . . . , y
m
) be defined by the
formulas
G
1
(x
1
, . . . , x
k
, y) , . . . , G
m
(x
1
, . . . , x
k
, y) , H(y
1
, . . . , y
m
, z)
respectively. Then f(x
1
, . . . , x
k
) is defined by the formula
∃y
1
∃y
m
(G
1
(x
1
, . . . , x
k
, y
1
) ∧ . . . ∧ G
m
(x
1
, . . . , x
k
, y
m
) ∧ H(y
1
, . . . , y
m
, z))
which we may abbreviate as F(x
1
, . . . , x
k
, z). This proves the lemma.
A k-place predicate P(x
1
, . . . , x
k
) is said to be arithmetically definable if it
is definable over the structure (N, +, , 0, 1, =) by a formula with k free variables
x
1
, . . . , x
k
.
Lemma 2.2.8. If the k+1-place predicate R(x
1
, . . . , x
k
, y) is arithmetically
definable, then so is the k-place partial function
ψ(x
1
, . . . , x
k
) ≃ least y such that R(x
1
, . . . , x
k
, y) holds.
Proof. Our λx
1
. . . x
k
[ψ(x
1
, . . . , x
k
)] is arithmetically definable by the formula
R(x
1
, . . . , x
k
, y) ∧ ∃z (z < y ∧ R(x
1
, . . . , x
k
, z))
which we may abbreviate as F(x
1
, . . . , x
k
, y). This proves the lemma.
It remains to prove:
Lemma 2.2.9. If the total k-place function g(y
1
, . . . , y
k
) and the total k+2-
place function h(x, z, y
1
, . . . , y
k
) are arithmetically definable, then so is the total
k+1-place function
f(x, y
1
, . . . , y
k
)
obtained by primitive recursion:
f(0, y
1
, . . . , y
k
) = g(y
1
, . . . , y
k
) ,
f(x + 1, y
1
, . . . , y
k
) = h(x, f(x, y
1
, . . . , y
k
), y
1
, . . . , y
k
)) .
In order to prove Lemma 2.2.9, we need some definitions and lemmas from
number theory. Two positive integers m and n are said to be relatively prime if
they have no common factor greater than 1. A set of positive integers is said to
be pairwise relatively prime if any two distinct integers in the set are relatively
prime.
52
Lemma 2.2.10. Given a positive integer k, we can find infinitely many positive
integers a such that the k integers in the set
a + 1 , 2a + 1 , . . . , ka + 1
are pairwise relatively prime.
Proof. Let a be any positive integer which is divisible by all of the prime numbers
which are less than k. We claim that a + 1, 2a + 1, . . . , ka + 1 are pairwise
relatively prime. Suppose not. Let i and j be such that 1 ≤ i < j ≤ k and ia+1
and ja+1 are not relatively prime. Let p be a prime number which is a factor of
both ia+1 and ja+1. Then p cannot be a factor of m. Hence p is greater than
or equal to k. On the other hand p is a factor of (ja + 1) −(ia + 1) = (j −i)a.
Hence p is a factor of j − i. But j − i is less than k, hence p is less than k, a
contradiction.
Example 2.2.11. If k = 5, the primes less than k are 2 and 3, so we can take
a to be any multiple of 6. Then the integers a + 1, 2a + 1, 3a + 1, 4a + 1,
5a +1 will be pairwise relatively prime. Taking a = 6 we see that 7, 13, 19, 25,
31 are paiwise relatively prime. Taking a = 12 we see that 13, 25, 37, 49, 61
are pairwise relatively prime. And taking a = 600 we see that 601, 1201, 1801,
2401, 3001 are pairwise relatively prime.
Lemma 2.2.12 (Chinese Remainder Theorem). Let m
1
, . . . , m
k
be a set of k
distinct positive integers which are pairwise relatively prime. Then for any given
non-negative integers r
i
< m
i
, 1 ≤ i ≤ k, we can find a non-negative integer r
such that Remainder(r, m
i
) = r
i
for all i.
Proof. Let m be the product of m
1
, . . . , m
k
. For any non-negative integer r
less than m, define the remainder sequence of r to be the sequence 'r
1
, . . . , r
k
`
where r
i
= Remainder(r, m
i
). We claim that any two distinct non-negative
integers less than m have distinct remainder sequences. To see this, let r and
s be non-negative integers less than m. If r and s have the same remainder
sequence, then r − s is divisible by each of m
1
, . . . , m
k
. Since m
1
, . . . , m
k
are
pairwise relatively prime, it follows that r − s is divisible by m. Since r and
s are both less than m, we must have r = s. This proves the claim. Define a
possible remainder sequence to be any sequence 'r
1
, . . . , r
k
` with 0 ≤ r
i
< m
i
for all i. The number of possible remainder sequences is exactly m. From this
and the claim, we see that any possible remainder sequence must actually occur
as the remainder sequence associated with some r in the range 0 ≤ r < m. The
lemma follows immediately.
Example 2.2.13. Continuing the previous example, we see that for any non-
negative integers r
1
< 7, r
2
< 13, r
3
< 19, r
4
< 25, r
5
< 31, there must be a
non-negative integer r less than 7 13 19 25 31 such that Remainder(r, 7) = r
1
,
. . . , Remainder(r, 31) = r
5
. We may view the integer r as a “code” for the
sequence (r
1
, r
2
, r
3
, r
4
, r
5
). The “decoding” is accomplished by means of the
key integer 6, since r
i
= Remainder(r, 6i + 1).
53
Definition 2.2.14. For any non-negative integers r, a, i, we define
β(r, a, i) = Remainder(r, a (i + 1) + 1) .
This 3-place function is known as G¨odel’s β-function. Note that the β-function
is arithmetically definable.
The significance of the β-function is that it can be used to encode an ar-
bitrary sequence of non-negative integers 'r
0
, r
1
, . . . , r
k
` by means of two non-
negative integers r and a. We make this precise in the following lemma.
Lemma 2.2.15. Given a finite sequence of non-negative integers r
0
, r
1
, . . . , r
k
,
we can find a pair of non-negative integers r and a such that β(r, a, 0) = r
0
,
β(r, a, 1) = r
1
, . . . , β(r, a, k) = r
k
.
Proof. By Lemma 2.2.10 (replacing k by k +1), we can find a positive integer a
such that r
0
< a+1, r
1
< 2a+1, . . . , r
k
< (k+1) a+1, and furthermore a+1,
2a +1, . . . , (k +1) a +1 are pairwise relatively prime. Then by Lemma 2.2.12
we can find a non-negative integer r such that Remainder(r, a + 1) = r
0
,
Remainder(r, 2a + 1) = r
1
, . . . , Remainder(r, (k + 1) a + 1) = r
k
. This proves
the lemma.
Exercise 2.2.16. Find a pair of numbers r, a such that β(r, a, 0) = 11, β(r, a, 1) =
19, β(r, a, 2) = 30, β(r, a, 3) = 37, β(r, a, 4) = 51.
Hint: First find an appropriate a by hand. Then write a small computer program
to find r by brute force.
We are now ready to prove Lemma 2.2.9.
Proof of Lemma 2.2.9. Let the total functions g(y
1
, . . . , y
k
) and h(x, z, y
1
, . . . , y
k
)
be defined by the formulas G(y
1
, . . . , y
k
, w) and H(x, z, y
1
, . . . , y
k
, w) respec-
tively. We wish to write down a formula F(x, y
1
, . . . , y
k
, w) which would say that
there exists a finite sequence r
0
, r
1
, . . . , r
x
such that G(y
1
, . . . , y
k
, r
0
) holds, and
H(i, r
i
, y
1
, . . . , y
k
, r
i+1
) holds for all i < x, and finally r
x
= w. If we could do
this, then clearly F(x, y
1
, . . . , y
k
, w) would define the function f(x, y
1
, . . . , y
k
).
This would prove the lemma. The only difficulty is that our language is not
powerful enough to talk directly about finite sequences of variable length in
the required way. Our language allows us to say things like “there exists a
non-negative integer r such that . . . ,” but it does not allow us to directly say
things like “there exists a sequence of non-negative integers r
0
, r
1
, . . . , r
x
(of
variable length, x) such that . . . .” The way to overcome this difficulty is to
use the β-function. Instead of saying “there exists a finite sequence r
0
, r
1
, . . . ,
r
k
,” we can say “there exists a pair of non-negative integers r and a which
encode the required sequence, via the β-function.” Namely, we write down a
formula F(x, y
1
, . . . , y
k
, w) which says, informally, there exist r and a such that
G(y
1
, . . . , y
k
, β(r, a, 0)) holds, and H(i, β(r, a, i), y
1
, . . . , y
k
, β(r, a, i + 1)) holds
for all i < x, and finally β(r, a, x) = w. By Lemma 2.2.15 it is clear that this
54
formula defines the function f(x, y
1
, . . . , y
k
). Formally, let B(r, a, i, w) be a for-
mula which defines the β-function. We may then take F(x, y
1
, . . . , y
k
, w) to be
the formula
∃r ∃a ( ∃u (B(r, a, 0, u) ∧ G(y
1
, . . . , y
k
, u)) ∧ B(r, a, x, w) ∧
∀i ( i ≥ x ∨ ∃u ∃v (B(r, a, i, u) ∧ B(r, a, i + 1, v) ∧ H(i, u, y
1
, . . . , y
k
, v)) ) ) .
This completes the proof of Lemma 2.2.9.
Example 2.2.17. The exponential function λxy [ y
x
] is arithmetically definable
by the formula
∃r ∃a ( B(r, a, 0, 1) ∧ B(r, a, x, w) ∧
∀i ( i ≥ x ∨ ∃u ∃v (B(r, a, i, u) ∧ B(r, a, i + 1, v) ∧ v = u y) ) )
which we may abbreviate as Exp(x, y, w), meaning that y
x
= w.
Exercise 2.2.18. Consider the function f : N →N defined by
f(n) =

n/2 if n is even,
3n + 1 if n is odd.
For each k ∈ N let
f
k
= f ◦ ◦ f
. .. .
k
,
i.e., f
0
(n) = n and f
k+1
(n) = f(f
k
(n)).
Write a formula F(x, y, z) in the language +, , 0, 1, =, < which, when in-
terpreted over the natural number system N, defines the 3-place predicate
f
x
(y) = z.
Exercise 2.2.19. Show that the function λx[ the xth Fibonacci number ] is
arithmetically definable.
Exercise 2.2.20. Which of the following number-theoretic predicates are arith-
metically definable? Prove your answers.
1. x is the sum of all the prime numbers less than y.
2. x
y
= z.
3. x! = y.
Theorem 2.2.21. Every partial recursive function is arithmetically definable.
Proof. Recall that the partial recursive functions have been characterized as the
smallest class of functions which includes the initial functions and is closed under
composition, primitive recursion, and the least-number operator. Lemma 2.2.6
says that the class of arithmetically definable functions includes the initial func-
tions. Lemmas 2.2.7, 2.2.8, and 2.2.9 say that the class of arithmetically de-
finable functions is closed under composition, the least-number operator, and
primitive recursion, respectively. It follows that the arithmetically definable
functions include the partial recursive functions.
55
Corollary 2.2.22. There exists a set K ⊆ N which is arithmetically definable
but not recursive.
Proof. Recall that
K = ¦x ∈ N [ ϕ
(1)
x
(x) is convergent¦
is our basic example of a non-recursive set. By the Enumeration Theorem,
the 2-place partial function λxy[ϕ
(1)
x
(y)] is partial recursive. Hence, by Theo-
rem 2.2.21 λxy[ϕ
(1)
x
(y)] is arithmetically definable. Let F(x, y, z) be a formula
which defines this function. Then K is defined by the formula ∃zF(x, x, z). This
completes the proof.
More generally, recall our discussion of the arithmetical hierarchy in Chap-
ter 1, Section 1.9. The next theorem characterizes arithmetically definability
(Definition 2.2.1) in terms of the arithmetical hierarchy (Definition 1.9.1).
Theorem 2.2.23. Let P be a k-place predicate, P ⊆ N
k
. The following are
equivalent:
1. P is arithmetically definable;
2. P belongs to the arithmetical hierarchy, i.e., P belongs to the class Σ
0
n
for
some n ∈ N.
Proof. Theorem 2.2.21 implies that every predicate in the class Σ
0
0
(= Π
0
0
) is
arithmetically definable. From this it is straightforward to prove by induction
on n ∈ N that every predicate in the class Σ
0
n
∪ Π
0
n
is arithmetically definable.
For the converse, put
Σ
0

=
¸
n∈N
Σ
0
n
=
¸
n∈N
Π
0
n
.
By Theorem 1.9.2, the class Σ
0

is closed under Boolean connectives ∧, ∨, and
universal and existential quantification ∀ and ∃. From this it follows that every
arithmetically definable predicate P belongs to the class Σ
0

; this is proved by
induction on the number of symbols in a defining formula for P.
Exercise 2.2.24. Let A and B be subsets of N. Prove that if A is reducible to
B and B is arithmetically definable, then A is arithmetically definable.
Exercise 2.2.25. For each n ≥ 1 let C
n
be a set which is Σ
0
n
complete. Consider
the set B = ¦2
n
3
x
[ x ∈ C
n
¦. Prove that B is not arithmetically definable.
Remark 2.2.26 (Hilbert’s Tenth Problem). A refinement of Theorem 2.2.23
due to Matiyasevich 1967 is as follows. Let us say that P ⊆ N
k
is Diophantine
if P = π
l
(Q) for some l ≥ 0 and some strongly Diophantine Q ⊆ N
k+l
. Thus
P = ¦'a
1
, . . . , a
k
` ∈ N
k
[ ∃'b
1
, . . . , b
l
` ∈ N
l
(f(a
1
, . . . , a
k
, b
1
, . . . , b
l
) = 0)¦
56
where f(x
1
, . . . , x
k
, y
1
, . . . , y
l
) is a polynomial with integer coefficients. Matiya-
sevich’s Theorem states that P is Diophantine if and only if P is Σ
0
1
.
A corollary of Matiyasevich’s Theorem is that, for example, the sets K and H
are Diophantine. Thus, we can find a polynomial f(z, x
1
, . . . , x
l
) with integer
coefficients, such that the set of a ∈ N for which f(a, x
1
, . . . , x
l
) = 0 has a
solution in N is nonrecursive. Here it is known that one can take l = 9, but
l = 8 is an open question.
Hilbert’s Tenth Problem, as stated in Hilbert’s famous 1900 problem list,
reads as follows:
To find an algorithm which allows us, given a polynomial equation in
several variables with integer coefficients, to decide in a finite number
of steps whether or not the equation has a solution in integers.
From Matiyasevich’s Theorem plus the unsolvability of the Halting Problem, it
follows that there is no such algorithm. In other words, Hilbert’s Tenth Problem
is unsolvable.
A full exposition of Hilbert’s Tenth Problem and the proof of Matiyasevich’s
Theorem is in my lecture notes for Math 574, Spring 2005, at
http://www.math.psu.edu/simpson/notes/.
Exercise 2.2.27. Recall that N = ¦0, 1, 2, . . .¦ = the natural numbers, while
Z = ¦. . . , −2, −1, 0, 1, 2, . . .¦ = the integers. In our formulation of Matiyase-
vich’s Theorem, we have spoken of solutions in N. What happens if we replace
“solution in N” by “solution in Z”?
Hint: Use the following well-known theorem of Lagrange: For each n ∈ N there
exist a, b, c, d ∈ N such that n = a
2
+b
2
+c
2
+d
2
.
2.3 G¨odel Numbers of Formulas
For each formula F in the language of arithmetic, we shall define a unique
positive integer #(F), which will be called the G¨ odel number of F. The number
#(F) will serve as a “code” for the formula F.
Before assigning G¨odel numbers to formulas, we shall first assign G¨odel
numbers to terms. We define
#(0) = 1
#(1) = 3
#(x
i
) = 3
2
5
i
#(t
1
+t
2
) = 3
3
5
#(t1)
7
#(t2)
#(t
1
t
2
) = 3
4
5
#(t1)
7
#(t2)
For example, the G¨odel number of the term 1 +x
0
is
#(1 +x
0
) = 3
3
5
3
7
9
57
since #(1) = 3 and #(x
0
) = 9.
We now define the G¨odel numbers of formulas:
#(t
1
= t
2
) = 2 5
#(t1)
7
#(t2)
#(F ∧ G) = 2 3 5
#(F)
7
#(G)
#(F ∨ G) = 2 3
2
5
#(F)
7
#(G)
#(F) = 2 3
3
5
#(F)
#(∀x
i
F) = 2 3
4
5
i
7
#(F)
#(∃x
i
F) = 2 3
5
5
i
7
#(F)
For example, the G¨odel number of the formula ∃x
1
(x
1
= 1) is
#(∃x
1
(x
1
= 1)) = 2 3
5
5
1
7
2·5
45
·7
3
since #(x
1
) = 45, #(1) = 3, and #(x
1
= 1) = 2 5
45
7
3
.
If T is any collection of formulas, we denote by #(T) the collection of all
G¨odel numbers of formulas in T. Let
Fml = #(all formulas) ,
Snt = #(all sentences) ,
and
TrueSnt = #(all true sentences) .
Note that Fml, Snt, and TrueSnt are subsets of N.
Theorem 2.3.1. The sets Fml and Snt are primitive recursive.
Proof. The proof is straightforward and we omit it.
We are going to prove that the set TrueSnt is nonrecursive. In other words,
the problem of deciding whether a given sentence of the language of arithmetic
is true or false is unsolvable. This result may be paraphrased as “arithmetical
truth is undecidable,” or simply, “arithmetic is undecidable.”
Theorem 2.3.2. Every arithmetically definable set A ⊆ N is reducible to
TrueSnt.
Proof. Given an arithmetically definable set A, let F(x
1
) be a formula with one
free variable x which defines A, i.e.,
A = ¦m ∈ N [ F(m) is true¦ .
Define
f(m) = #(∃x
1
(x
1
= m∧ F(x
1
))) .
58
Thus for all m ∈ N we have that m ∈ A if and only if f(m) ∈ TrueSnt. We
claim that the function f : N →N is primitive recursive. This is clear since
f(m) = 2 3
3
7
2·3·5
#(x
1
=m)
·7
#(F(x
1
))
where
#(x
1
= m) = 2 5
45
7
#(m)
,
and #(m) = #(1 +. . . + 1
. .. .
m
) can be defined primitive recursively by
#(0) = 1 ,
#(m + 1) = 3
3
5
#(m)
7
3
.
Thus A is reducible to TrueSnt via f. This proves the theorem.
Theorem 2.3.3. TrueSnt is not recursive.
Proof. By Corollary 2.2.22 we have an arithmetically definable set K which is
not recursive. By the previous theorem K is reducible to TrueSnt. Hence by
Lemma 1.7.8 TrueSnt is not recursive.
More generally we have the following theorem, which may be paraphrased
as “arithmetical truth is not arithmetically definable.”
Theorem 2.3.4 (Tarski). TrueSnt is not arithmetically definable.
Proof. Suppose that TrueSnt were arithmetically definable. Then by Theo-
rem 2.2.23 we would have that TrueSnt belongs to the class Σ
0
n
for some n.
By Theorem 1.9.18, let C be a complete Σ
0
n+1
set. By Theorem 2.2.23 C is
arithmetically definable. Hence by Theorem 2.3.2 C is reducible to TrueSnt.
Hence by Lemma 1.9.16 C belongs to the class Σ
0
n
. This contradicts that fact
(Theorem 1.9.18) that a complete Σ
0
n+1
set can never belong to the class Σ
0
n
.
This completes the proof.
Exercise 2.3.5. Prove that the set Fml of all G¨odel numbers of formulas is
primitive recursive.
Exercise 2.3.6. Prove that the set Snt of all G¨odel numbers of sentences is
primitive recursive.
59
Chapter 3
The Real Number System
In Chapter 2 we have shown that TrueSnt
N
is nonrecursive, i.e., the theory of
the natural number system is undecidable. In this Chapter we shall show that,
by contrast, TrueSnt
R
is recursive, i.e., the theory of the real number system is
decidable.
3.1 Quantifier Elimination
Let L
OR
be the language of ordered rings, i.e.,
L
OR
= (+, −, , 0, 1, <, =)
where + and are 2-ary operation symbols, − is a 1-ary operation symbol, <
and = are 2-place predicate symbols, and 0 and 1 are constant symbols. We
consider the L
OR
-structure
{ = (R, +
R
, −
R
,
R
, 0
R
, 1
R
, <
R
, =
R
) ,
i.e., the real number system, the ordered field of real numbers. Formulas of L
OR
are interpreted with reference to {, i.e., ∃x . . . means “there exists x ∈ R such
that . . . ,” etc.
Two L
OR
-formulas F and G are said to be equivalent if they have the same
free variables x
1
, . . . , x
k
and define the same k-place predicate P ⊆ R
k
. An
equivalent condition is that the sentence
∀x
1
∀x
k
(F(x
1
, . . . , x
k
) ⇔ G(x
1
, . . . , x
k
))
is true in {.
We are going to prove the following theorem, due originally to Tarski:
Theorem 3.1.1 (Quantifier Elimination). For any L
OR
-formula F, we can find
an equivalent quantifier free L
OR
-formula F

.
60
Example 3.1.2. The formula
∃x(ax
2
+bx +c = 0)
is equivalent to the quantifier free formula
(a = 0 ∧ b = 0 ∧ c = 0) ∨ (a = 0 ∧ b = 0) ∨ (a = 0 ∧ b
2
−4ac ≥ 0) .
Exercise 3.1.3. Find quantifier-free formulas in the language +, −, , 0, 1, <, =
which are equivalent, over the real number system R, to:
1. ∃x(ax
2
+bx +c > 0).
2. ∃x(ax
3
+bx
2
+cx +d > 0).
3. ∃x(ax
4
+bx
3
+cx
2
+dx +e > 0).
Exercise 3.1.4. Does there exist a constant c such that the following holds?
Given a formula F(x) in the language +, , 0, 1, = with exactly one
free variable x, we can find a formula F

(x) in the same language
which is equivalent to F(x) over the natural number system N, and
which contains at most c quantifiers.
Prove your answer.
Remark 3.1.5. The only properties of { that will be used in the proof of
Theorem 3.1.1 are: (1) { is a commutative ordered field; and (2) { has the
intermediate value property for polynomials, i.e.,
( x < y ∧ p(x) < 0 < p(y) ) ⇒ ∃z ( x < z < y ∧ p(z) = 0 )
for any polynomial p(x) ∈ R[x]. An ordered field with these properties is called
a real closed ordered field. This is related to Hilbert’s 17th Problem.
Remark 3.1.6. The proof of Theorem 3.1.1 which we shall present below is
due to P. J. Cohen. In order to present the proof, we shall define and study a
class of functions called the effective functions. This notion of effectivity has no
importance beyond the proof of Theorem 3.1.1. See also Corollary 3.1.21 below.
Definition 3.1.7. A predicate A on the reals, i.e., A ⊆ R
k
, is said to be ef-
fective if it is definable over { by a quantifier free formula. That is, A is a
Boolean combination of sets in R
k
which are defined by equations and inequa-
tions p(x
1
, . . . , x
k
) = 0, p(x
1
, . . . , x
k
) > 0, where p ∈ Z[x
1
, . . . , x
k
].
Remark 3.1.8. If we allow parameters from R, we get semi-algebraic sets.
Thus “effective = semi-algebraic with parameters from Z”.
Definition 3.1.9. A function f : D →R, D ⊆ R
k
is said to be effective if
61
1. D is effective, and
2. for every effective predicate A(x
1
, . . . , x
k
, y, z
1
, . . . , z
n
), the predicate
B(x
1
, . . . , x
k
, z
1
, . . . , z
n
) ≡ A(x
1
, . . . , x
k
, f(x
1
, . . . , x
k
), z
1
, . . . , z
n
)
is effective.
Example 3.1.10. It can be shown that the function

x is effective. This is
because, first, the domain of

x is the effective set ¦x ∈ R [ x ≥ 0¦, and second,
for instance,

x > 3 ≡ x > 9.
In order to prove Theorem 3.1.1, we shall build up a library of effective
functions. We shall use notations such as x and y to abbreviate sequences of
variables such as x
1
, . . . , x
k
and y
1
, . . . , y
m
.
Lemma 3.1.11. The functions x +y, x y, −x and x/y are effective.
Proof. For x+y, x y and −x there is nothing to prove. For x/y, note first that
the domain is ¦(x, y) [ y = 0¦ which is obviously effective. It remains to show
that if A(z, . . .) is an effective predicate then so is A(x/y, . . .). The latter is a
Boolean combination of predicates of the form
a
n

x
y

n
+. . . +a
1

x
y

+a
0
> 0
and this is equivalent to the atomic formula
a
n
x
n
+a
n−1
x
n−1
y +. . . +a
1
xy
n−1
+a
0
y
n
> 0 ,
assuming as we may that n is even.
Corollary 3.1.12. Any rational function
f(x
1
, . . . , x
k
) ∈ Q(x
1
, . . . , x
k
)
is effective.
Lemma 3.1.13. The composition of effective functions is effective.
Proof. Consider for instance f(g(x)) where f and g are effective 1-place func-
tions. If D(y) is a quantifier-free formula defining the dom(f), then D(g(x))
defines the dom(fg), which is therefore effective. If A(y, z) is any effective pred-
icate, then clearly A(f(y), z) is effective, hence A(f(g(x)), z) is effective.
Lemma 3.1.14. The function
sgn(x) =

1 if x > 0
0 if x = 0
−1 if x < 0.
is effective.
62
Proof. Let A(y, z) be an effective predicate. Then A(sgn(x), z) is equivalent to
equivalent to
(x > 0 ∧ A(1, z)) ∨ (x = 0 ∧ A(0, z)) ∨ (x < 0 ∧ A(−1, z))
which is again effective. This proves the lemma.
More generally we have:
Lemma 3.1.15. If a function f(x) takes only finitely many values, all integers,
then f(x) is effective if and only if for each j ∈ Z, the predicate f(x) = j is
effective.
Proof. If: Let j
1
, . . . , j
n
be the finitely many values. For any effective predicate
A(y, z), the predicate A(f(x), z) is equivalent to
(f(x) = j
1
∧ A(j
1
, z)) ∨ ∨ (f(x) = j
n
∧ A(j
n
, z))
and is therefore effective.
Only if: Trivial.
Lemma 3.1.16 (Definition by Cases). If two functions f
1
(x) and f
2
(x) and a
predicate A(x) are effective, then the function
f(x) =

f
1
(x) if A(x),
f
2
(x) if A(x)
is effective.
Proof. We must show that, for each effective predicate B(y, z), the predicate
C(x, z) ≡ B(f(x), z) is effective. This is so because C(x, z) is equivalent to
(A(x) ∧ B(f
1
(x), z)) ∨ (A(x) ∧ B(f
2
(x), z)) .
Since f
1
(x) and f
2
(x) and B(y, z) are effective, B(f
1
(x), z) and B(f
2
(x), z) are
effective. Since A(x) is effective, it follows that C(x, z) is effective. This proves
the lemma.
The extension of the previous lemma to more than two cases is obvious.
Lemma 3.1.17. A function f(x) is effective if and only if for every positive
integer d ≥ 1 and every polynomial q(y) ∈ R[y] of degree d, sgn(q(f(x))) is an
effective function of x and the d + 1 coefficients of q(y).
Proof. Only if: Trivial, since for instance
sgn(q(f(x))) = 1 ≡ q(f(x)) > 0 ≡ A(f(x)),
where A(y) is the predicate q(y) > 0.
63
If: We must show that if A(y, z) is effective then A(f(x), z) is effective. Note
that A(y, z) can be viewed as a Boolean combination of atomic predicates of the
form p(y, z) > 0, for various polynomials p(y, z) with coefficients in Z. It suffices
to show that the predicates p(f(x), z) > 0 are effective. In order to show this,
write p(y, z) = q(y) ∈ Z[z][y] i.e., q(y) is a polynomial in y whose coefficients
are polynomials in z with integer coefficients. Then
p(f(x), z) > 0 ≡ sgn(q(f(x))) = 1 .
By assumption sgn(q(f(x))) is an effective function of x and the cofficients of
q; hence by composition sgn(q(f(x))) is an effective function of x and z.
The next lemma says that the real roots of a polynomial are effective func-
tions of the coefficients. For example, the quadratic formula
x =
−b ±

b
2
−4ac
2a
shows that the roots of ax
2
+bx +c are effective functions of a, b and c.
Lemma 3.1.18 (Main Lemma). Let p(x) = a
n
x
n
+ + a
1
x + a
0
∈ R[x].
Write a = a
0
, . . . , a
n
. There are n + 1 effective functions ξ
1
(a), . . . , ξ
n
(a), and
k = k(a) such that
ξ
1
(a) < < ξ
k
(a)
are all of the real roots of p(x).
Proof. By induction on n. Let
p

(x) = na
n
x
n−1
+ + 2a
2
x +a
1
be the derivative of p(x). This is of degree ≤ n−1. By inductive hypothesis, the
roots of p

(x) are among t
1
< < t
m
, m = n − 1, where the t
i
’s are effective
function of a. Note that p(x) is monotone on each of the open intervals
(−∞, t
1
) , (t
1
, t
2
) , . . . , (t
m−1
, t
m
) , (t
m
, +∞)
So, in each of these intervals, there is at most one root of p(x). In addition the
t
i
’s could be roots of p(x). Thus the number of roots is determined by sgn(p(t
i
)),
1 ≤ i ≤ n − 1 and sgn(p

(t
1
− 1)) and sgn(p

(t
n
+ 1)). We can therefore use
definition by cases to obtain k(a) as an effective function of a.
It remains to show that the roots themselves are effective functions of a.
Consider for example a root ξ = ξ(a) in the interval (t
1
, t
2
) where p(t
1
) > 0 and
p(t
2
) < 0. By the previous lemma, it suffices to show that, for each d ≥ 1 and
polynomial q(x) of degree d, sgn(q(ξ)) is effective (as a function of a and the
coefficients of q).
Replacing q(x) by its remainder on division by p(x), we may assume deg(q(x)) <
n. [Details: Long division gives q(x) = p(x) f(x) + r(x), where deg(r(x)) < n
and the coefficients of r(x) are effective functions of the coefficients of p(x) and
q(x). We can replace q(x) by r(x).]
64
By induction hypothesis, the roots of q(x) can be found effectively. Let the
roots of q(x) be among u
1
< < u
m
. Then the sign of q(x) on the m + 1
intervals
(−∞, u
1
) , (u
1
, u
2
) , . . . , (u
m−1
, u
m
) , (u
m
, +∞)
is given by m + 1 effective functions
sgn(q(u
1
−1)) , sgn

q

u
1
+u
2
2

, . . . , sgn

q

u
m−1
+u
m
2

, sgn(q(u
m
+1)) .
Moreover the position of ξ relative to the u
i
’s is determined by the positions of t
1
and t
2
relative to the u
i
’s and by the sgn(p(u
i
))’s. Thus we can use definition by
cases to obtain sgn(q(ξ)) as an effective function. In view of the previous lemma
characterizing effective functions, this shows that ξ is an effective function. The
proof of the Main Lemma is now complete.
Lemma 3.1.19. If A(x
1
, x
2
, . . . , x
k
) is an effective predicate, then the predicate
B(x
2
, . . . , x
k
) ≡ ∃x
1
A(x
1
, x
2
, . . . , x
k
)
is effective.
Proof. The predicate A(x
1
, x
2
, . . . , x
k
) may be viewed as a Boolean combination
of polynomial inequalities of the form p
i
(x
1
) > 0, 1 ≤ i ≤ l, where the coeffi-
cients of the p
i
(x)’s are polynomials in x
2
, . . . , x
k
with integer coefficients. By
the Main Lemma, the roots of the p
i
(x)’s are effective function of x
2
, . . . , x
k
. Let
ξ
1
, . . . , ξ
m
be all of these roots. Hence the p
i
(x)’s change sign only at ξ
1
, . . . , ξ
m
.
It follows that
∃x
1
A(x
1
, x
2
, . . . , x
k
)
is equivalent to a finite disjunction
A(η
1
, x
2
, . . . , x
k
) ∨ ∨ A(η
n
, x
2
, . . . , x
k
) ,
where η
1
, . . . , η
n
is a list of all the ξ
i
’s and (ξ
i
+ ξ
j
)/2’s and ξ
i
± 1’s. Since the
η
i
’s are effective functions of x
2
, . . . , x
k
, it follows that the above disjunction is
an effective predicate of x
2
, . . . , x
k
. This proves the lemma.
Theorem 3.1.20. Any predicate A ⊆ R
k
which is definable over { is effective.
Proof. A ⊆ R
k
is defined over { by a formula F(x
1
, . . . , x
k
) of L
OR
. Therefore,
it suffices to show that any formula F of L
OR
is equivalent over { to a quantifier
free formula F

. We shall prove this by induction on the number of symbols in
F.
If F is quantifier free, we may take F

≡ F.
If F ≡ G, then we may take F

≡ G

.
If F ≡ G∧ H, then we may take F

≡ G

∧ H

.
If F ≡ ∃xG, let F(y) ≡ ∃xG(x, y), where y is a list of the free variables
of F. By the inductive hypothesis, G(x, y) is equivalent to a quantifier-free
formula G

(x, y). Then G

(x, y) defines an effective predicate B(x, y). By the
65
previous lemma, the predicate A(y) ≡ ∃xB(x, y) is effective, i.e., is defined by
a quantifier free formula F

(y). Clearly F is equivalent to F

. This completes
the proof.
Corollary 3.1.21. For a predicate A ⊆ R
k
the following three conditions are
equivalent:
1. A is effective.
2. A is definable over {.
3. A is definable over { by a quantifier free formula.
Similarly for functions f : D →R, D ⊆ R
k
.
Proof. For predicates this follows immediately from Theorem 3.1.20. Consider
now a function f. Note first that if f is effective then the predicate y = f(x) is
effective, i.e., definable by a quantifier free formula. Conversely, suppose that f
is definable. Then for any effective predicate A(y, z), the predicate A(f(x), z) is
equivalent to the definable predicate ∃y (y = f(x) ∧ A(y, z)), which is therefore
effective in view of what has already been proved. Thus f is effective.
Theorem 3.1.22 (Quantifier Elimination). If a predicate A ⊆ R
k
is definable
over {, then it is definable over { by a quantifier free formula.
Proof. This is merely a restatement of the previous corollary.
Proof of Theorem 3.1.1. Theorem 3.1.1 is a restatement of the previous theo-
rem.
3.2 Decidability of the Real Number System
Theorem 3.2.1. Given a formula F of L
OR
, there is an algorithm to find an
equivalent quantifier free formula F

.
Proof. The algorithm can be obtained by tracing back through the proof of
Theorem 3.1.22.
Theorem 3.2.2. Given a sentence S of L
OR
, there is an algorithm to determine
whether or not { satisfies S, i.e., whether S is true in the real number system.
Proof. Given a sentence S, a special case of the previous theorem is that we can
algorithmically compute S

, an equivalent quantifier free sentence. But then S

is a Boolean combination of atomic sentences of the form t
1
= t
2
and t
1
< t
2
where t
1
and t
2
are variable-free terms, for example 1+(0+1) < (1+1)−(1+0),
and the truth value of such sentences is easily computed. This completes the
proof.
66
Corollary 3.2.3. The set
TrueSnt
R
= ¦#(S) [ S is a sentence ∧ S is true in {¦
is recursive.
Proof. This follows from the previous Theorem plus Church’s Thesis. Alterna-
tively, we can convert the proof of Theorem 3.1.1 into a rigorous proof that the
function taking #(F) to #(F

) is primitive recursive.
Exercise 3.2.4. Recall that N = ¦0, 1, 2, . . .¦ = the natural numbers, Z =
¦. . . , −2, −1, 0, 1, 2, . . .¦ = the integers, and R = (−∞, ∞) = the real numbers.
Show that TrueSnt
N
and TrueSnt
Z
are not recursive. (This is in contrast to
the fact that TrueSnt
R
is recursive.)
Theorem 3.2.5. The theory of the ordered ring of real numbers is decidable.
Proof. This is a restatement of the previous corollary.
Corollary 3.2.6. Plane and solid geometry are decidable.
Proof. By the methods of Cartesian analytic geometry, the theory of points,
lines and circles in the plane is interpretable into the theory of the real numbers.
For example, a point is an ordered pair (x, y) where x and y are real numbers.
A line is an ordered quadruple (a, b, u, v) where (u, v) = (0, 0). A point (x, y)
lies on a line (a, b, u, v) if and only if
∃t (x = a +tu ∧ y = b +tv) .
Two lines are considered identical if and only if they contain the same points.
A circle is an ordered triple (a, b, r) where r > 0. A point (x, y) lies on a circle
(a, b, r) if and only if (x − a)
2
+ (y − b)
2
= r
2
. A triangle consists of three
non-collinear points, the vertices of the triangle. Etc., etc.
Similarly for the theory of points, lines, planes, circles, and spheres in space.
Exercise 3.2.7. Write sentences of the language +, −, , 0, 1, <, = which, when
interpreted over the real number system, express the following statements of
Euclidean plane geometry.
1. For every two points, there is a unique line passing through them.
2. For every three non-collinear points, there is a unique circle passing through
them.
3. For every line L and circle C, the intersection of L and C consists of at
most two points.
4. Given a line L and a point P, among all points on L there is exactly one
which is at minimum distance from P.
67
5. For every circle C and point P lying on C, there exists one and only one
line L such that L ∩ C = P. (I.e., a tangent line.)
6. Every line segment has a unique midpoint.
7. Every angle can be uniquely bisected.
Exercise 3.2.8. Explain in detail how you would translate the following state-
ments of Euclidean plane geometry into sentences of the language +, −, , 0, 1, <
, = over the real number system.
1. The three angle bisectors of any triangle meet in a single point.
2. Every angle can be uniquely trisected.
Exercise 3.2.9. Recall that N = ¦0, 1, 2, . . .¦ = the natural numbers, Z =
¦. . . , −2, −1, 0, 1, 2, . . .¦ = the integers, and R = (−∞, ∞) = the real numbers.
According to Matiyasevich’s Theorem, we can find a polynomial
f(w, x
1
, . . . , x
k
)
with integer coefficients, such that the set of a ∈ N for which the equation
f(a, x
1
, . . . , x
k
) = 0 has a solution in N is noncomputable.
1. Discuss the analogous question in which “solution in N” is replaced by
“solution in Z”.
2. Discuss analogous questions in which “solution in N” is replaced by “so-
lution in R”.
68
Chapter 4
Informal Set Theory
The purpose of this chapter is to develop set theory in an informal, preaxiomatic
way. A good reference for this material is Na¨ıve Set Theory by P. R. Halmos.
4.1 Operations on Sets
Informally, a set is any collection of objects which may be regarded as a com-
pleted totality. We use capital letters X, Y , . . . to denote sets. If a is any
object and X is any set, we write a ∈ X to mean that a belongs to X, and
a / ∈ X to mean that a does not belong to X. Synonyms for “belongs to” are “is
an element of”, “is a member of”, and “is contained in”.
Since a set is nothing but a collection of elements, the set itself having no
further structure, it follows that two sets are equal if and only if they contain
exactly the same elements. Symbolically,
X = Y ⇔ ∀a (a ∈ X ⇔ a ∈ Y ) .
This is known as the principle of extensionality. It can be taken as a definition
of equality between sets.
If P(a) is any definite property that an object a may or may not have,
we use the notation ¦a [ P(a)¦ to denote the set of all objects a which have
property P, if such a set exists. (If such a set exists, it will be unique in view
of extensionality.)
In order to be a set, a collection of objects must be limited in size and
definite. These requirements are rather vague, but we shall try to give some
explanation of what they entail. Definiteness means that any object a either
belongs or does not belong to the collection, i.e., there is no third possibility.
Limitedness means that the collection is in some sense not too large. The need
for some limitation-of-size requirement will be shown below in connection with
the Russell paradox.
One of the most basic concepts in set theory is that of one set being included
in another. We say that Y is a subset of X, symbolically Y ⊆ X, if every element
69
of Y is an element of X, i.e.,
∀a (a ∈ Y ⇒ a ∈ X) .
If P(a) is any definite property as above, and if X is any set, we can form
a subset Y = ¦a ∈ X [ P(a)¦, consisting of all elements a of X which have
property P. Thus
∀a(a ∈ Y ⇔ (a ∈ X ∧ P(a))) .
Note that any set X is itself a mathematical object and as such can be an
element of another set. Indeed, given a set X, we can form a set
{(X) = ¦Y [ Y ⊆ X¦ ,
called the power set of X, whose elements are all possible subsets of X.
We now prove a theorem which implies that not all definite collections of
mathematical objects are sets. This means that some limitation-of-size principle
is needed.
Theorem 4.1.1 (Russell Paradox). The collection of all sets is not itself a set.
Proof. Suppose to the contrary that there were a set S consisting of all sets.
Form the subset D consisting of all sets which are not members of themselves.
Symbolically,
D = ¦X ∈ S [ X / ∈ X¦ .
Then D ∈ D if and only if D / ∈ D, a contradiction. This completes the proof.
The Russell Paradox shows that we cannot form sets with complete freedom.
Nevertheless, a wide variety of sets can be formed. Some examples of sets are ∅
(the empty set, i.e., the unique set which has no elements), ¦∅¦ (the one-element
set whose unique element is the empty set), ¦∅, ¦∅¦¦, etc. For any objects a,
b, and c, we can form the set ¦a, b, c¦ whose elements are exactly a, b, and c.
The cardinality of this set will be one, two or three depending on which of a,
b, and c are equal to each other. Some examples of infinite sets are N (the
set of natural numbers), R (the set of real numbers), {(N), {(R), {({(R)),
etc. Another source of examples is subsets defined by properties, for example
¦n ∈ N [ n is prime¦ and ¦X ⊆ R [ X is countably infinite¦. Further sets can
be obtained using the set operations discussed below.
Given two objects a and b, there is an object (a, b) called the ordered pair of
a and b. The ordered pair operation is assumed to have the following property:
(a, b) = (a

, b

) ⇔ (a = a

∧ b = b

) .
Given two sets X and Y , the Cartesian product X Y is the set of all ordered
pairs (a, b) such that a ∈ X and b ∈ Y .
For any set X, a function with domain X is a rule f associating to each
object a ∈ X a definite, unique object f(a). In this case we write dom(f) = X
and rng(f) = ¦f(a) [ a ∈ X¦. The latter is again a set, called the range of F.
70
If f is a function and Z is a set, there is a unique function f↾Z, the restriction
of f to Z, whose domain is dom(f) ∩ Z and such that (f↾Z)(a) = f(a) for all
a in its domain.
We write f : X → Y to mean that f is a function, dom(f) = X, and
rng(f) ⊆ Y . The set of all such functions is denoted Y
X
.
Summarizing some of the above points, we have the following binary opera-
tions on sets:
X ∪ Y = ¦a [ a ∈ X ∨ a ∈ Y ¦ (union) ,
X ∩ Y = ¦a [ a ∈ X ∧ a ∈ Y ¦ (intersection) ,
X ` Y = ¦a [ a ∈ X ∧ a / ∈ Y ¦ (difference) ,
X Y = ¦(a, b) [ a ∈ X ∧ b ∈ Y ¦ (product) ,
X
Y
= ¦f [ f : Y → X¦ (exponential) .
By an indexed collection with index set I, we mean simply a function f
whose domain is I. In discussing indexed collections, we use notation such as
f = 'a
i
`
i∈I
where a
i
= f(i). For example, an ordered n-tuple 'a
1
, . . . , a
n
` is
a function whose domain is ¦1, . . . , n¦, and a sequence 'a
n
`
n∈N
is a function
whose domain is N.
Given an indexed collection of sets 'X
i
`
i∈I
, the following operations are
defined.
¸
i∈I
X
i
= ¦a [ ∃i ∈ I (a ∈ X
i
)¦ (union) ,
¸
i∈I
X
i
= ¦a [ ∀i ∈ I (a ∈ X
i
)¦ (intersection) ,
¸
i∈I
X
i
= ¦'a
i
`
i∈I
[ ∀i ∈ I (a
i
∈ X
i
)¦ (product) .
If in the latter operation we take all of the sets X
i
to be the same set X, we get
the Cartesian power
¸
i∈I
X = X
I
which is the same as the previously mentioned exponential.
The Axiom of Choice is the assertion that, for any indexed collection of sets
'X
i
`
i∈I
, if ∀i ∈ I (X
i
= ∅) then
¸
i∈I
X
i
= ∅. This implies that it is possible
to choose one element a
i
∈ X
i
for each i ∈ I. In the early years of set theory,
there was some controversy about the Axiom of Choice. Nowadays the Axiom of
Choice is accepted as being intuitively obvious, but we shall follow the custom
of indicating which proofs use it.
4.2 Cardinal Numbers
A function f is said to be one-to-one if for all a, a

∈ dom(f), a = a

implies
f(a) = f(a

). Note that in this case there is an inverse function f
−1
with
dom(f
−1
) = rng(f) and rng(f
−1
) = dom(f), defined by
f
−1
(b) = a ⇔ f(a) = b .
71
Definition 4.2.1. If X and Y are sets, we say X is equinumerous with Y ,
written X ≈ Y , if there exists a one-to-one correspondence between X and Y ,
i.e., a one-to-one function f with dom(f) = X and rng(f) = Y .
Lemma 4.2.2.
1. X ≈ X.
2. X ≈ Y if and only if Y ≈ X.
3. X ≈ Y and Y ≈ Z imply X ≈ Z.
Proof. Straightforward.
Because of the preceding lemma, we can associate to any set X an object
card(X), the cardinality or cardinal number of X, in such a way that
X ≈ Y ⇔ card(X) = card(Y ) .
If X is finite, we take card(X) to be the number of elements in X. For infinite
sets X, it is not important at this stage what sort of object the cardinal number
card(X) is, so long as the above property holds. We use Greek letters κ, λ, µ,
ν, . . . to denote cardinal numbers.
Definition 4.2.3. We write X Y to mean that X ≈ X
1
for some X
1
⊆ Y .
Lemma 4.2.4.
1. If X ≈ X

and Y ≈ Y

, then X Y if and only if X

Y

.
2. X X.
3. X Y and Y Z imply X Z.
4. X Y and Y X imply X ≈ Y .
5. For all sets X and Y , either X Y or Y X.
Proof. Parts 1, 2, and 3 are straightforward. Parts 4 and 5 will be proved
later, as consequences of the Well-Ordering Theorem. Part 4 is known as the
Cantor-Schroeder-Bernstein Theorem.
We can now make the following definition for cardinal numbers: κ ≤ λ if and
only if X Y where κ = card(X) and λ = card(Y ). This does not depend on
the choice of X and Y , as noted in part 1 of Lemma 4.2.4. The rest of Lemma
4.2.4 implies that ≤ is a linear ordering of the cardinal numbers, i.e., we have:
1. κ ≤ κ.
2. (κ ≤ λ ∧ λ ≤ µ) ⇒ κ ≤ µ.
3. (κ ≤ λ ∧ λ ≤ κ) ⇒ κ = λ.
72
4. ∀κ∀λ(κ ≤ λ ∨ λ ≤ κ).
Definition 4.2.5 (Cardinal Arithmetic). For cardinal numbers κ = card(X)
and λ = card(Y ), we define
1. κ +λ = card(X ∪ Y ).
2. κ λ = card(X Y ).
3. κ
λ
= card(X
Y
).
(In the definition of κ +λ, it is assumed that X ∩ Y = ∅.)
For example, 2
κ
= card({(X)) where κ = card(X).
Theorem 4.2.6. For cardinal numbers κ, λ, and µ, we have
1. κ +λ = λ +κ.
2. (κ +λ) +µ = κ + (λ +µ).
3. κ λ = λ κ.
4. (κ λ) µ = κ (λ µ).
5. κ (λ +µ) = κ λ +κ µ.
6. κ
λ+µ
= κ
λ
κ
µ
.
7. κ
λ·µ
= (κ
λ
)
µ
.
8. κ + 0 = κ, κ 0 = 0, κ 1 = κ.
Proof. Straightforward.
Later we shall prove that, for infinite cardinal numbers κ and λ,
κ +λ = κ λ = max(κ, λ) .
Theorem 4.2.7 (Cantor’s Theorem). For any cardinal number κ, we have
2
κ
> κ. In other words, for any set X, we have X {(X) and {(X) X.
Proof. We have X {(X) via the one-to-one function f : X → {(X) where
f(a) = ¦a¦. Suppose now that {(X) X holds. Let g : {(X) → X be
one-to-one. Put
D = ¦a ∈ X [ a ∈ rng(g) ∧ a / ∈ g
−1
(a)¦ .
Then g(D) ∈ D if and only if g(D) / ∈ D, a contradiction.
73
Exercise 4.2.8. Define ℵ
0
= card(N). Without using the Cantor-Schroeder-
Bernstein Theorem, prove that

0
= card(Z) = card(Q)
and
2
ℵ0
= card(R) = card(C) = card([0, 1]) = card([0, 1] [0, 1]).
Definition 4.2.9. If 'κ
i
`
i∈I
is an indexed set of cardinal numbers, we define
1.
¸
i∈I
κ
i
= card(
¸
i∈I
X
i
)
2.
¸
i∈I
κ
i
= card(
¸
i∈I
X
i
)
where κ
i
= card(X
i
). In the definition of
¸
i∈I
κ
i
, it is assumed that X
i
∩X
j
= ∅
for all i, j ∈ I with i = j.
Exercise 4.2.10 (K¨ onig’s Theorem). Suppose that 'κ
i
`
i∈I
and 'λ
i
`
i∈I
are in-
dexed sets of cardinal numbers with the same index set I. Show that if κ
i
< λ
i
for all i ∈ I, then
¸
i∈I
κ
i
<
¸
i∈I
λ
i
.
Remark 4.2.11. Cantor’s Theorem may be viewed as the special case of
K¨onig’s Theorem with κ
i
= 1 and λ
i
= 2. The Axiom of Choice may be
viewed as the special case of K¨onig’s Theorem with κ
i
= 0 and λ
i
> 0.
4.3 Well-Orderings and Ordinal Numbers
Definition 4.3.1. Given a set A, a relation on A is a set R ⊆ A A. A
relational structure is an ordered pair (A, R) where A is a set and R ⊆ A A.
We sometimes write aRa

instead of (a, a

) ∈ R.
Definition 4.3.2. Given two relational structures (A, R) and (B, S), an isomor-
phism from (A, R) to (B, S) is a one-to-one function f such that dom(f) = A,
rng(f) = B, and
aRa

⇔ f(a)Sf(a

)
for all a, a

∈ A. We say that (A, R) is isomorphic to (B, S), symbolically
(A, R)

= (B, S), if there exists an isomorphism from (A, R) to (B, S).
Lemma 4.3.3.
1. (A, R)

= (A, R).
2. (A, R)

= (B, S) if and only if (B, S)

= (A, R).
3. (A, R)

= (B, S) and (B, S)

= (C, T) imply (A, R)

= (C, T).
74
Proof. Straightforward.
Because of the preceding lemma, we can associate to any relational structure
(A, R) a mathematical object type(A, R), the isomorphism type of (A, R), in
such a way that
(A, R)

= (B, S) ⇔ type(A, R) = type(B, S) .
It is not important at this stage exactly what sort of mathematical object
type(A, R) is, so long as the above property holds.
Definition 4.3.4. A relational structure (A, R) is said to be well-founded if for
every nonempty set X ⊆ A, there exists a ∈ X such that there is no b ∈ X with
bRa. Such an a might be called an R-minimal element of X.
Lemma 4.3.5. A relational structure (A, R) is well-founded if and only if there
is no infinite descending R-sequence. (By an infinite descending R-sequence we
mean a sequence 'a
n
`
n∈N
such that a
n+1
Ra
n
for all n ∈ N.)
Proof. If 'a
n
`
n∈N
is an infinite descending R-sequence, then X = ¦a
n
[ n ∈ N¦
is a counterexample to well-foundedness of (A, R). Conversely, suppose that
X ⊆ A is a counterexample to well-foundedness. Then we have X = ∅ and
∀a ∈ X∃b ∈ X bRa. By the Axiom of Choice, there is a function f : X → X
such that f(a)Ra for all a ∈ X. Pick an element a
0
∈ X and define 'a
n
`
n∈N
recursively by putting a
n+1
= f(a
n
) for all n ∈ N. This is an infinite descending
R-sequence. The lemma is proved.
Definition 4.3.6. A linear ordering is a relational structure (A, R) with the
following properties: aRb and bRc imply aRc; and for all a, b ∈ A exactly one of
aRb, a = b, bRa hold. A well-ordering is a linear ordering which is well-founded.
Note that if (A, R) is a well-ordering and X is a nonempty subset of A, then
X has an R-least element, i.e., there is a unique a ∈ X such that aRb holds for
all b ∈ X, b = a.
Definition 4.3.7. An ordinal number is the isomorphismtype of a well-ordering.
For n ∈ N, we identify n with the ordinal number which is the order type of
all n-element well-orderings. Another important ordinal number is ω, the order
type of N itself (more precisely of (N, <) = (N, ¦(m, n) ∈ N N [ m < n¦)).
The following definition and exercise show how to generate further examples of
ordinal numbers.
Definition 4.3.8. Let α = type(A, R) and β = type(B, S) be ordinal numbers.
We define
1. α +β = type(A ∪ B, R ∪ S ∪ (A B)). Here we assume that A ∩ B = ∅.
2. α β = type(A B, ¦((a, b), (a

, b

)) [ bSb

∨ (b = b

∧ aRa

)¦).
75
Thus we have operations of ordinal addition, α +β, and ordinal multiplication,
α β. (Later we shall define an analogous operation of ordinal exponentiation,
α
β
.) Note that the commutative laws fail, for example 1 + ω = ω = ω + 1 and
2 ω = ω = ω 2 = ω +ω. However, we have the following properties.
Exercise 4.3.9. Prove the following laws of ordinal arithmetic.
1. (α +β) +γ = α + (β +γ).
2. (α β) γ = α (β γ).
3. α (β +γ) = α β +α γ.
4. α + 0 = 0 +α = α, α 0 = 0 α = 0, α 1 = 1 α = α.
Give an example showing the failure of (α +β) γ = α γ +β γ.
Definition 4.3.10. If (A, R) is a linear ordering, an initial segment of (A, R)
is any subset of A of the form ¦b [ bRa¦, where a ∈ A. Note that
(¦b [ bRa¦, ¦(c, b) [ cRb ∧ bRa¦)
is again a linear ordering, and we sometimes identify the initial segment with
this ordering.
Theorem 4.3.11 (Comparability of Well-Orderings). Let (A, R) and (B, S) be
well-orderings. Then exactly one of the following holds.
1. (A, R)

= (B, S);
2. (A, R)

= some initial segment of (B, S);
3. (B, S)

= some initial segment of (A, R).
Moreover, in each case, the isomorphism is unique.
Proof. We first prove that the isomorphism is unique. Suppose for instance that
f
1
and f
2
are two different isomorphisms from (A, R) to (B, S) or to some initial
segment of (B, S). Then ¦a ∈ A [ f
1
(a) = f
2
(a)¦ is a nonempty subset of A, so
let a be its R-least element. Then f
1
(a

) = f
2
(a

) for all a

Ra but f
1
(a) = f
2
(a),
say f
1
(a)Sf
2
(a). Then f
1
(a) / ∈ rng(f
2
), contradicting the fact that rng(f
2
) is B
or an initial segment of B with respect to S.
Now let f be a function with dom(f) ⊆ A and rng(f) ⊆ B, defined by
putting f(a) = the unique b ∈ B such that
(¦a

[ a

Ra¦, ¦(a
′′
, a

) [ a
′′
Ra

∧ a

Ra¦)

= (¦b

[ b

Sb¦, ¦(b
′′
, b

) [ b
′′
Sb

∧ b

Sb¦) ,
provided such a b exists. If for a given a ∈ A no such b exists, f(a) is undefined.
If b exists, its uniqueness follows from what we have already proved. It is clear
that a

Ra and a ∈ dom(f) imply a

∈ dom(f), and b

Rb and b ∈ rng(f) imply
b

∈ rng(f).
76
We claim that dom(f) = A or rng(f) = B or both. If not, let a be the
R-least element of A ` dom(f), and let b be the S-least element of B ` rng(f).
Then f is an isomorphism from (¦a

[ a

Ra¦, ¦(a
′′
, a

) [ a
′′
Ra

∧ a

Ra¦) to (¦b

[
b

Sb¦, ¦(b
′′
, b

) [ b
′′
Sb

∧ b

Sb¦). This implies that a ∈ dom(f), a contradiction.
The claim is proved.
If both dom(f) = A and rng(f) = B, then we have (A, R)

= (B, S). If
dom(f) = A, then letting a be the R-least element of A ` dom(f), we see that
f is an isomorphism from (¦a

[ a

Ra¦, ¦(a
′′
, a

) [ a
′′
Ra

∧ a

Ra¦) to (B, S). If
rng(f) = B, then letting b be the S-least element of B ` rng(f), we see that
f is an isomorphism from (A, R) to (¦b

[ b

Sb¦, ¦(b
′′
, b

) [ b
′′
Sb

∧ b

Sb¦). This
completes the proof of the theorem.
Definition 4.3.12. For ordinal numbers α and β, we define α < β to mean
that (A, R)

= some initial segment of (B, S), where α = type(A, R) and β =
type(B, S). We define α ≤ β to mean α < β ∨ α = β.
Lemma 4.3.13. For all ordinal numbers α and β, exactly one of α < β, α = β,
and β < α holds. Moreover α < β and β < γ imply α < γ.
Proof. The first part follows from comparability of well-orderings. The second
part is straightforward.
Exercise 4.3.14. For ordinal numbers α, β, γ, prove that α < β if and only if
α +γ = β for some γ > 0.
Lemma 4.3.15. If (A, R) is a well-ordering and B ⊆ A, then (B, R∩(B B))
is a well-ordering, and
type(B, R ∩ (B B)) ≤ type(A, R) .
Proof. Straightforward, using comparability of well-orderings.
The next theorem implies that any well-ordering is isomorphic to an initial
segment of the ordinal numbers.
Theorem 4.3.16. For any ordinal number α, the relational structure
(¦β [ β < α¦, ¦(γ, β) [ γ < β < α¦)
is a well-ordering of type α.
Proof. Let (A, R) be some fixed well-ordering of type α. Define f : A → ¦β [
β < α¦ by
f(b) = type(¦c ∈ A [ cRb¦, ¦(d, c) [ dRc ∧ cRb¦) .
It is straightforward to verify that f is an isomorphism from (A, R) onto
(¦β [ β < α¦, ¦(γ, β) [ γ < β < α¦) .
This proves the theorem.
77
Corollary 4.3.17. For any ordinal number α, there is a set ¦β [ β < α¦
consisting of all smaller ordinal numbers.
Proof. This follows from the previous theorem.
Lemma 4.3.18. Let X be a set of ordinal numbers. Then there is an ordinal
number γ = sup X which is the least upper bound of X under <, i.e., ∀α(α ∈
X ⇒ α ≤ γ) and ∀β (β < γ ⇒ ∃α(α ∈ X ∧ β < α)).
Proof. Put
A = ¦α [ ∃β (β ∈ X ∧ α < β)¦ =
¸
β∈X
¦α [ α < β¦ .
It is straightforward to verify that
(A, ¦(α, β) ∈ A A [ α < β¦)
is a well-ordering. Let γ be the type of this well-ordering. It is straightforward
to verify that A = ¦α [ α < γ¦ and that γ = supX.
Exercise 4.3.19. If α is an ordinal number and X is a nonempty set of ordinal
numbers, show that α+sup X = sup¦α+β [ β ∈ X¦ and α sup X = sup¦α β [
β ∈ X¦.
Lemma 4.3.20. Let X be a nonempty set of ordinal numbers. Then X has a
smallest element under <.
Proof. Put α = sup X. Then X is a nonempty subset of ¦β [ β ≤ α¦. The latter
set of ordinal numbers is well-ordered under <, hence X has a least element.
Theorem 4.3.21 (Burali-Forti Paradox). The class Ord of all ordinal numbers
is not a set.
Proof. If Ord were a set, then by the above lemmas, (Ord, <) would be a well-
ordering. Letting α be the type of this well-ordering, we see that (Ord, <) would
be isomorphic to an initial segment of itself, namely (¦β [ β < α¦, ¦(γ, β) [ γ <
β < α¦). This contradiction completes the proof.
4.4 Transfinite Recursion
By a class we mean a collection of objects which is not necessarily a set. Every
set is a class, but not every class is a set. Examples of classes which are not sets
are Set = ¦X [ X is a set¦ and Ord = ¦α [ α is an ordinal¦. These classes are
“too big” to be sets. We have seen this in connection with the Russell Paradox
and the Burali-Forti Paradox.
If C is a class, then by a function with domain C we mean a rule F which
associates to each element a of C a uniquely defined object F(a). For example,
although the class Ord of all ordinal numbers is not a set, we shall be interested
in functions with domain Ord. The next theorem gives us a powerful method
for defining such functions.
78
Theorem 4.4.1 (Transfinite Recursion). Let T be the class of all functions
whose domain is an initial segment of Ord. Suppose that G is a function with
domain T. Then there is a unique function F with domain Ord such that, for
all ordinal numbers α,
F(α) = G(F↾¦β [ β < α¦) .
Proof. Let us say that f ∈ T is good if for all β ∈ dom(f), f(β) = G(f↾¦γ [
γ < β¦). We claim that for all ordinal numbers α, there is at most one good f
with dom(f) = ¦β [ β < α¦. If not, let f
1
= f
2
be two such f’s. Let γ be the
smallest β < α such that f
1
(β) = f
2
(β). Then f
1
↾¦β [ β < γ¦ = f
2
↾¦β [ β < γ¦,
hence
f
1
(γ) = G(f
1
↾¦β [ β < γ¦) = G(f
2
↾¦β [ β < γ¦) = f
2
(γ) ,
a contradiction.
Using the above claim, let f
α
be the unique good f with dom(f) = ¦β [
β < α¦, if it exists. We claim that f
α
exists for all α. If not, let α be the
smallest counterexample. Then f
β
exists for all β < α, and it is easy to check
that ¦(β, G(f
β
)) [ β < α¦ is good. This contradicts the choice of α. Thus f
α
exists for all α. Define F by putting F(α) = G(f
α
) for all α. It is easy to check
that F↾¦β [ β < α¦ = f
α
and that F satisfies the desired conclusions. This
completes the proof.
As an example of transfinite recursion, we define the following operations of
ordinal arithmetic.
Definition 4.4.2.
1. α +β = sup¦α, (α +γ) + 1 [ γ < β¦.
2. α β = sup¦(α γ) +α [ γ < β¦.
3. α
β
= sup¦1, α
γ
α [ γ < β¦ (assuming α > 0).
Exercise 4.4.3. Show that parts 1 and 2 of Definition 4.4.2 agree with parts 1
and 2 of Definition 4.3.8. In the next exercise we show how to extend Definition
4.3.8 to encompass part 3 of Definition 4.4.2.
Exercise 4.4.4. Let α = type(A, R) and β = type(B, S) be ordinal numbers.
Show that α
β
= type(C, T) where C is the set of all f : B → A such that, for
all but finitely many b ∈ B, f(b) = a
0
, where a
0
is the R-least element of A.
Here T is the set of all (f
1
, f
2
) ∈ C C such that f
1
(b

)Rf
2
(b

), where b

is the
S-greatest b ∈ B such that f
1
(b) = f
2
(b).
Exercise 4.4.5. If α is an ordinal number and X is a nonempty set of ordinal
numbers, show that
1. α + sup X = sup¦α +β [ β ∈ X¦,
79
2. α supX = sup¦α β [ β ∈ X¦, and
3. α
sup X
= sup¦α
β
[ β ∈ X¦.
Exercise 4.4.6. For ordinal numbers α, β, and γ, show that
1. (α +β) +γ = α + (β +γ).
2. (α β) γ = α (β γ).
3. α (β +γ) = α β +α γ.
4. α
β+γ
= α
β
α
γ
.
5. α
β·γ
= (α
β
)
γ
.
6. α + 0 = 0 +α = α, α 0 = 0 α = 0, α 1 = 1 α = α.
7. α
0
= 1.
8. 0
α
= 0 provided α > 0.
9. α
1
= α, 1
α
= 1.
Exercise 4.4.7. Show that β < γ implies the following:
1. α +β < α +γ;
2. α β < α γ provided α > 0;
3. α
β
< α
γ
provided α > 1.
Definition 4.4.8. A successor ordinal is an ordinal number of the form α +1.
A limit ordinal is an ordinal number δ such that δ > 0 and α + 1 < δ for all
α < δ. Examples of limit ordinals are ω and ω 2.
Exercise 4.4.9. Show that α + 1 is the smallest ordinal number β > α. Show
that every ordinal number is either 0, a successor ordinal, or a limit ordinal.
Show that δ > 0 is a limit ordinal if and only if δ = sup¦α [ α < δ¦. Show that
δ > 0 is a limit ordinal if and only if δ = ω α for some α > 0.
Exercise 4.4.10. Show that the operations of ordinal arithmetic could have
been defined by transfinite recursion as follows (letting δ denote a limit ordinal):
1. α + 0 = α, α + (β + 1) = (α +β) + 1, α +δ = sup¦α +β [ β < δ¦.
2. α 0 = 0, α (β + 1) = (α β) +α, α δ = sup¦α β [ β < δ¦.
3. α
0
= 1, α
β+1
= (α
β
) α, α
δ
= sup¦α
β
[ β < δ¦ (assuming α > 0).
Exercise 4.4.11. For an ordinal number δ > 0, show that the following are
equivalent.
1. α +δ = δ for all α < δ.
80
2. α +β < δ for all α, β < δ.
3. δ = ω
α
for some α ≤ δ.
An ordinal number with these properties is said to be additively indecomposable.
Exercise 4.4.12. Show that for any ordinal number α, there is one and only
one way to write α in the form α = α
1
+ +α
n
where α
1
≥ ≥ α
n
> 0 are
additively indecomposable, and n ∈ N.
4.5 Cardinal Numbers, Continued
Lemma 4.5.1. Given a set X, there exists a choice function for X, i.e., a
function
c : {(X) ` ¦∅¦ → X
such that c(Y ) ∈ Y for all Y ⊆ X, Y = ∅.
Proof. Consider the indexed set of sets 'X
i
`
i∈I
, where I = {(X) ` ¦∅¦ and
X
i
= i for all i ∈ I. By the Axiom of Choice,
¸
i∈I
X
i
is nonempty, i.e., there
exists 'a
i
`
i∈I
such that a
i
∈ X
i
for all i ∈ I. Putting c(i) = a
i
we obtain our
choice function.
Theorem 4.5.2 (Well-Ordering Theorem). Given a set X, we can find a rela-
tion R ⊆ X X such that (X, R) is a well-ordering.
Proof. We shall prove the theorem in the following equivalent formulation: For
any set X, there exists an ordinal number α such that X ≈ ¦β [ β < α¦.
(Neither α nor the one-to-one function from X onto ¦β [ β < α¦ is asserted to
be unique.)
Fix a choice function c for X. Fix an object a
0
such that a
0
/ ∈ X. By
transfinite recursion, define
F(α) =

c(X ` rng(F↾¦β [ β < α¦)), if X ` rng(F↾¦β [ β < α¦) = ∅
a
0
otherwise.
We claim that F(α) = a
0
for some α. If not, we would have F(α) ∈ X for
all α, and F(α) = F(β) for all α = β. Form the set
Y = rng(F) = ¦a ∈ X [ ∃α(F(α) = a)¦ .
Then F
−1
is a function with domain Y , and we have rng(F
−1
) = Ord, hence
Ord is a set. This contradiction proves the claim.
Let α be the smallest ordinal number such that F(α) = a
0
. Then F↾¦β [
β < α¦ is one-to-one, and rng(F↾¦β [ β < α¦) = X. Thus ¦β [ β < α¦ ≈ X.
Our theorem is proved.
81
Remark 4.5.3. The previous theorem shows that the Axiom of Choice implies
the Well-Ordering Theorem. There is also a converse: the Well-Ordering Theo-
rem implies the Axiom of Choice. To see this, suppose we have an indexed set
of nonempty sets 'X
i
`
i∈I
. Put A =
¸
i∈I
X
i
. By the Well-Ordering Theorem,
there exists R ⊆ A A such that (A, R) is a well-ordering. Define 'a
i
`
i∈I
by
putting a
i
= the R-least element of X
i
. Thus 'a
i
`
i∈I

¸
i∈I
X
i
and we have
proved the Axiom of Choice from the Well-Ordering theorem.
Exercise 4.5.4. Let X be a set of sets. By a chain within X we mean a set
C ⊆ X such that for all U, V ∈ C either U ⊆ V or V ⊆ U. A chain within X is
said to be maximal if it is not properly included in any other chain within X.
Use the Axiom of Choice plus transfinite recursion to prove that there exists
a maximal chain within X.
Note: This is a version of Zorn’s Lemma.
Definition 4.5.5. An initial ordinal is an ordinal number α such that, for all
β < α,
¦γ [ γ < α¦ ≈ ¦γ [ γ < β¦ .
The finite ordinal numbers 0, 1, 2, . . . are initial ordinals, as is the first
infinite ordinal number ω. But it is easy to see that ordinal numbers such as
ω + 1, ω + 2, . . . , ω 2, ω 2 + 1, . . . are not initial ordinals. Another simple
fact worth noting is that every infinite initial ordinal is a limit ordinal.
Definition 4.5.6. For any set X, we define [X[ to be the smallest ordinal
number α such that X ≈ ¦β [ β < α¦. (The existence of such an ordinal is a
consequence of the Well-Ordering Theorem.) Clearly [X[ is an initial ordinal.
In fact, [X[ is the unique initial ordinal such that X ≈ ¦β [ β < α¦.
Lemma 4.5.7. If X ⊆ ¦β [ β < α¦, then [X[ ≤ α.
Proof. Immediate from Lemma 4.3.15.
Theorem 4.5.8. For all sets X and Y we have
X ≈ Y if and only if [X[ = [Y [ ,
and
X Y if and only if [X[ ≤ [Y [ .
Proof. The first equivalence is obvious, as is the fact that [X[ ≤ [Y [ implies
X Y . Suppose now that X Y . Then X ≈ Z for some Z ⊆ ¦β [ β < [Y [¦.
By the previous lemma it follows that [X[ = [Z[ ≤ [Y [. This completes the
proof.
Remark 4.5.9. By the previous theorem, we have card(X) = card(Y ) if and
only if [X[ = [Y [, and card(X) < card(Y ) if and only if [X[ < [Y [. Thus we may
identify cardinal numbers with initial ordinals. From now on we shall make
this identification, writing card(X) = [X[. For instance, the finite cardinal
numbers are now identified with the finite ordinal numbers, and the smallest
infinite cardinal number is the same as the ordinal number ω.
82
Theorem 4.5.10.
1. If X Y and Y X, then X ≈ Y .
2. For all sets X and Y , either X Y or Y X.
Proof. Both parts follow from the previous theorem plus the fact that [X[ and
[Y [ are ordinal numbers, hence exactly one of [X[ < [Y [, [X[ = [Y [, [Y [ < [X[
holds.
4.6 Cardinal Arithmetic
We now present some basic results about the arithmetic of infinite cardinal
numbers. Most of these results are easy consequences of the following lemma.
Lemma 4.6.1. For infinite cardinals κ, we have κ κ = κ.
Proof. If not, let κ be the smallest counterexample. Note that κ is an infinite
initial ordinal, and λ λ < κ for all λ < κ.
Put A = ¦α [ α < κ¦ and R = ¦(α

, α) [ α

< α < κ¦. Thus (A, R) is a
well-ordering of type κ. Note that [A[ = κ but every initial segment I of (A, R)
has [I[ < κ.
Put B = AA and define S ⊆ B B by


, β

)S(α, β) if and only if
max(α

, β

) < max(α, β) ∨
(max(α

, β

) = max(α, β) ∧ α

< α) ∨
(max(α

, β

) = max(α, β) ∧ α

= α ∧ β

< β) .
It is straightforward to verify that (B, S) is a well-ordering.
If J ⊆ B is any initial segment of (B, S), we have J ⊆ I I where I is
an appropriately chosen initial segment of (A, R). Thus every initial segment
of (B, S) has cardinality < κ. Hence (A, R) cannot be isomorphic to an initial
segment of (B, S). From comparability of well-orderings, it follows that (B, S)
is isomorphic to (A, R). In particular B ≈ A. In other words, AA ≈ A, hence
κ κ = κ. This completes the proof.
Theorem 4.6.2. For infinite cardinals κ and λ, we have
κ +λ = κ λ = max(κ, λ) .
Proof. Put µ = max(κ, λ). Then by the previous lemma we have
µ ≤ κ +λ ≤ µ +µ = 2 µ ≤ µ µ = µ
and
µ ≤ κ λ ≤ µ µ = µ.
This proves the theorem.
83
Theorem 4.6.3. For λ infinite and 2 ≤ κ ≤ 2
λ
, we have κ
λ
= 2
λ
.
Proof. 2
λ
≤ κ
λ
≤ (2
λ
)
λ
= 2
λ·λ
= 2
λ
.
For any set X, let X

be the set of finite sequences of elements of X, i.e.,
X

= ¦'a
i
`
i<n
[ n < ω, a
i
∈ X for all i < n¦ .
Theorem 4.6.4. For any infinite set X, we have [X

[ = [X[.
Proof. Putting κ = [X[, we have κ
2
= κ κ = κ and it is easy to prove by
induction on n that κ
n
= κ for all n ≥ 1, n < ω. Hence we have
[X

[ =
¸
n<ω
κ
n
= ω κ = κ.
This proves the theorem.
Definition 4.6.5. For any ordinal β we define β
+
= the smallest initial ordinal
κ > β. To each ordinal number α we associate an infinite initial ordinal ω
α
as
follows, by transfinite recursion:
1. ω
0
= ω,
2. ω
α+1
= ω
+
α
,
3. ω
δ
= sup
α<δ
ω
α
for limit ordinals δ.
Remark 4.6.6. It is easy to see that 'ω
α
`
α∈Ord
is a strictly increasing enumer-
ation of all the infinite initial ordinals, i.e., the infinite cardinals. It follows that
the class Card = ¦κ [ κ is an infinite cardinal¦ is not a set.
Exercise 4.6.7. Prove that there exist arbitrarily large ordinals α such that
α = ω
α
.
Remark 4.6.8. Although cardinals are now the same thing as initial ordinals,
the notation ℵ
α
= ω
α
is sometimes used in order to maintain a notational
distinction between cardinals and initial ordinals. ℵ
α
is taken to be a cardinal,
while ω
α
is an initial ordinal. For example, even though ℵ
0
is the same thing
as ω, ℵ
0
is thought of as the cardinality of the set of natural numbers, while ω
is thought of as the order type of the natural numbers under <.
Theorem 4.6.9. Let N, Q, and R be the set of natural numbers, the rational
numbers, and the real numbers respectively. The cardinalities of these sets are
given by [N[ = [Q[ = ℵ
0
, [R[ = 2
ℵ0
.
Proof. The fact that [N[ = ℵ
0
is obvious. Since N ⊆ Q and each rational number
q ∈ Q is of the form q = ±m/n for some (m, n) ∈ N N, we have
[N[ ≤ [Q[ ≤ 2 [N[ [N[ = [N[
so [Q[ = [N[ = ℵ
0
. For the real numbers, note first that {(N) R via the
function which sends X ⊆ N to
¸
n∈X
2/3
n
. Hence 2
ℵ0
≤ [R[. On the other
hand, R {(Q) ≈ {(N) via the function which sends x ∈ R to ¦q ∈ Q [ q < x¦.
Thus [R[ = 2
ℵ0
.
84
Exercise 4.6.10. Prove that [R
N
[ = 2
ℵ0
and [R
R
[ = 2
2

0
.
The most important problem of infinite cardinal arithmetic is the Continuum
Problem: What is the cardinality of R? Equivalently, what is the ordinal number
β such that 2
ℵ0
= ℵ
β
? By Cantor’s Theorem we have 2
ℵ0
≥ ℵ
1
. The assertion
that 2
ℵ0
= ℵ
1
is known as the Continuum Hypothesis, or CH.
More generally, for any infinite cardinal κ, Cantor’s Theorem tells us that
2
κ
≥ κ
+
, and we can ask whether 2
κ
= κ
+
. The assertion that 2
κ
= κ
+
for
all infinite cardinals κ (equivalently, 2
ℵα
= ℵ
α+1
for all ordinal numbers α) is
known as the Generalized Continuum Hypothesis, or GCH.
Exercise 4.6.11. Prove that 2
ℵ0
= ℵ
ω
. (Hint: Use K¨onig’s Theorem.)
4.7 Some Classes of Cardinals
A cardinal is said to be uncountable if it is > ℵ
0
. In this section we introduce
some important classes of uncountable cardinals.
Definition 4.7.1. Let λ be an uncountable cardinal. We say that λ is a suc-
cessor cardinal if λ = κ
+
for some κ < λ. We say that λ is a limit cardinal if
κ
+
< λ for all κ < λ. We say that λ is a strong limit cardinal if 2
κ
< λ for all
κ < λ.
Note that λ is a successor cardinal if and only if λ = ℵ
α+1
for some successor
ordinal α + 1, and λ is a limit cardinal if and only if λ = ℵ
δ
for some limit
ordinal δ. The first few uncountable successor cardinals are ℵ
1
, ℵ
2
, . . . . The
first uncountable limit cardinal is ℵ
ω
. Clearly every strong limit cardinal is a
limit cardinal, and the GCH implies that every limit cardinal is a strong limit
cardinal.
Lemma 4.7.2. Every uncountable cardinal is either a successor cardinal or a
limit cardinal, and exactly one of these possibilities holds.
Proof. Obvious.
Definition 4.7.3. An infinite cardinal λ is said to be regular if it is not the sum
of fewer than λ cardinals each less than λ. In other words, for any indexed set
of cardinals 'κ
i
`
i∈I
with [I[ < λ and κ
i
< λ for all i ∈ I, we have
¸
i∈I
κ
i
< λ.
Trivially ℵ
0
is regular, since it is not the sum of a finite set of finite cardinals.
Theorem 4.7.4. Every uncountable successor cardinal is regular.
Proof. Let λ be an uncountable successor cardinal. Thus λ = κ
+
where κ is an
infinite cardinal. Given an indexed set of cardinals 'κ
i
`
i∈I
, we see that κ
i
< λ
implies κ
i
≤ κ, also [I[ < λ implies [I[ ≤ κ, hence
¸
i∈I
κ
i

¸
i∈I
κ = [I[ κ ≤ κ κ = κ < λ.
This shows that λ is regular.
85
In particular ℵ
1
, ℵ
2
, . . . are regular. An infinite cardinal is said to be sin-
gular if it is not regular. The first singular cardinal is ℵ
ω
, since ℵ
ω
=
¸
n∈N

n
.
Exercise 4.7.5. Let α be an ordinal number which is > ω. Show that α is
a regular cardinal (i.e., a regular initial ordinal) if and only if, for every set of
ordinal numbers X with sup X = α, we have type(X) = α.
Definition 4.7.6. Let λ be an uncountable cardinal. The cofinality of λ,
written cf(λ), is the smallest cardinal κ such that λ =
¸
i∈I
λ
i
for some indexed
set of cardinals λ
i
< λ, i ∈ I, with [I[ = κ.
Exercise 4.7.7. Prove the following facts.
1. cf(λ) is regular.
2. λ is regular if and only if cf(λ) = λ.
3. λ is singular if and only if cf(λ) < λ.
4. λ
cf(λ)
> λ. (Hint: Use K¨onig’s Theorem.)
5. For any infinite cardinal κ we have cf(µ
κ
) > κ for all µ > 1. In particular,
cf(2
κ
) > κ.
Exercise 4.7.8. Let κ be an infinite regular cardinal. Show that there exist
arbitrarily large strong limit cardinals λ such that cf(λ) = κ. Moreover, for all
such λ we have λ
µ
= λ for all µ < κ.
Exercise 4.7.9. Assuming the GCH, prove that for all infinite cardinals κ and
λ we have
λ
κ
=

λ if κ < cf(λ),
λ
+
if cf(λ) ≤ κ ≤ λ,
κ
+
if κ ≥ λ.
Definition 4.7.10. An inaccessible cardinal is an uncountable, regular, strong
limit cardinal. A weakly inaccessible cardinal is an uncountable, regular, limit
cardinal.
Remark 4.7.11. Clearly every inaccessible cardinal is weakly inaccessible, and
the GCH implies that every weakly inaccessible cardinal is inaccessible. It can
be shown that every weakly inaccessible cardinal is a fixed point of the ℵ
α
’s, i.e.,
such cardinals are of the form λ = ℵ
λ
. Moreover, every strongly inaccessible
cardinal has λ
κ
= λ for all κ < λ.
The existence of inaccessible and/or weakly inaccessible cardinals is not ob-
vious. Indeed, we shall see later that the existence of such cardinals cannot be
established using the accepted axioms of set theory.
86
4.8 Pure Well-Founded Sets
A set X is said to be transitive if, for every set Y such that Y ∈ X, we have
Y ⊆ X.
Lemma 4.8.1. Given a set X, there is a smallest transitive set TC(X) including
X. (TC(X) is called the transitive closure of X.)
Proof. Define U(X) =
¸
¦Y [ Y ∈ X, Y a set¦. By recursion on n < ω define
TC
0
(X) = X
TC
n+1
(X) = U(TC
n
(X)) .
Then TC(X) =
¸
n∈N
TC
n
(X) is easily seen to be the smallest transitive set Y
such that Y ⊇ X.
For any set A, we write
∈[A = ¦(b, a) [ a, b ∈ A, a is a set, b ∈ a¦ .
Definition 4.8.2. A set X is said to be well-founded if the relational structure
(TC(X), ∈[TC(X)) is well-founded.
Applying Lemma 4.3.5 to the relational structure (TC(X), ∈[TC(X)), we see
that X is well-founded if and only if there is no infinite sequence of sets 'X
n
`
n∈N
with
X = X
0
∋ X
1
∋ ∋ X
n
∋ .
Definition 4.8.3. A set X is said to be pure if every element of TC(X) is a
set. In other words, X is a pure set if not only X but also all the elements of
X, elements of elements of X, . . . , are sets.
By transfinite recursion we define transitive sets R
α
, α ∈ Ord, as follows:
R
0
= ∅
R
α+1
= {(R
α
)
R
δ
=
¸
α<δ
R
α
for limit ordinals δ .
By transfinite induction on α ∈ Ord, it is clear that β < α implies R
β
∈ R
α
,
hence β ≤ α implies R
β
⊆ R
α
.
Theorem 4.8.4. X ∈
¸
α∈Ord
R
α
if and only if X is a pure, well-founded set.
Proof. It is straightforward to prove by transfinite induction on α that all el-
ements of R
α
are pure and well-founded. Conversely, suppose X is pure and
well-founded. We claim that, for all Y ∈ TC(¦X¦), there exists an ordinal
number α such that Y ∈ R
α
. If not, let Y ∈ TC(¦X¦) be ∈-minimal such that
no such α exists. For each Z ∈ Y , let f(Z) be the least ordinal number β such
that Z ∈ R
β
. Put γ = sup
Z∈Y
f(Z). Then Y ⊆ R
γ
, hence Y ∈ {(R
γ
) = R
γ+1
,
a contradiction. This proves the claim. In particular, X ∈ R
α
for some α. This
proves the theorem.
87
The class of all pure, well-founded sets is denoted V . Thus we have
V =
¸
α∈Ord
R
α
.
If X is a pure, well-founded set, we define the rank of X to be the least ordinal
number α such that X ⊆ R
α
. Then clearly
rank X = sup¦rankY + 1 [ Y ∈ X¦ .
Moreover, for all ordinals α we have
R
α
= ¦X [ X is a pure well-founded set of rank < α¦ ,
and rankR
α
= α.
Exercise 4.8.5. Assuming the GCH, prove that [R
ω+α
[ = ℵ
α
for all ordinals
α.
4.9 Set-Theoretic Foundations
In the next chapter we shall begin the study of axiomatic set theory. In ax-
iomatic studies of set theory, the set concept is usually restricted to pure,
well-founded sets. This restriction tends to isolate set theory from the rest
of mathematics. Nevertheless, the restriction is partially justified by the fact
that many or most mathematical objects can be reconstructed or redefined as
pure, well-founded sets.
For example, the natural numbers 0, 1, 2, . . . are not ordinarily regarded as
being sets, but within the universe of pure, well-founded sets, it is possible to
define a structural replica of the natural numbers. Thus, from a certain per-
spective, natural numbers can be viewed as certain kinds of pure, well-founded
sets.
A similar remark applies to each of following mathematical concepts: natural
number, real number, ordinal number, cardinal number, ordered pair, function.
For each concept in this list, it is possible to identify mathematical objects of
the given type with certain pure, well-founded sets. The purpose of this section
is to show exactly how these identifications can be made. We begin with ordered
pairs and progress to functions, ordinal numbers, and real numbers.
Definition 4.9.1. For any two objects a and b, let us write
(a, b) = ¦¦a¦, ¦a, b¦¦ .
Thus (a, b) is a set. Note that if a and b are pure, well-founded sets, then so is
(a, b).
Lemma 4.9.2. If (a, b) = (a

, b

) then a = a

and b = b

.
88
Proof. Putting X = (a, b), we see that a is the unique element of
¸
Y ∈X
Y , and
b is the unique element of
¸
Y ∈X
Y ` ¦a¦ if the latter is nonempty, otherwise
b = a. Thus a and b can be recovered from (a, b) by single-valued set-theoretic
operations. The lemma follows.
By the above lemma, we may view (a, b) as the ordered pair formed from a
and b. From now on we shall make this identification, which is customary in
pure set theory.
In an earlier section of these notes, we defined a function with domain X
to be a rule associating to each a ∈ X a unique b. In pure set theory, it is
customary to replace this definition by the following, which we shall use from
now on.
Definition 4.9.3. A function is a set of ordered pairs, f, which is single-valued,
i.e.,
∀a ∀b ∀c (((a, b) ∈ f ∧ (a, c) ∈ f) ⇒ b = c) .
The domain of f is dom(f) = ¦a [ ∃b ((a, b) ∈ f)¦. If f is a function and
a ∈ dom(f) we write f(a) = the unique b such that (a, b) ∈ f.
The pure set-theoretic reconstruction of the ordinal numbers, due to von
Neumann, is as follows:
Definition 4.9.4. A von Neumann ordinal is a transitive, pure, well-founded
set A such that (A, ∈[A) is a well-ordering.
Note that if A is a von Neumann ordinal, then for each b ∈ A, the initial segment
B = ¦a [ a ∈ b¦ is again a von Neumann ordinal.
Lemma 4.9.5. For each ordinal number α, there is a unique von Neumann
ordinal A
α
such that type(A
α
, ∈[A
α
) = α. Moreover, the rank of A
α
is α.
Proof. By transfinite recursion we define A
α
= ¦A
β
[ β < α¦ for all ordinals α.
By transfinite induction on α, it is straightforward to verify that A
α
is the unique
von Neumann ordinal such that type(A
α
, ∈[A
α
) = α, and that rank(A
α
) =
α.
Remark 4.9.6. It is customary in pure set theory to identify the ordinal number
α with the von Neumann ordinal A
α
. From now on we shall make this identifi-
cation. Thus we have 0 = ∅ = ¦¦, 1 = ¦0¦ = ¦¦¦¦, 2 = ¦0, 1¦ = ¦¦¦, ¦¦¦¦¦, . . . ,
ω = ¦0, 1, 2, . . .¦ = N. Moreover, for all ordinals α we have α = ¦β [ β < α¦
and α + 1 = α ∪ ¦α¦. Also, if X is any set of ordinals, then
sup X =
¸
X =
¸
α∈X
α.
As for cardinal numbers, we have already seen how cardinal numbers may be
identified with certain ordinal numbers, namely, the initial ordinals. Thus, we
already know how to identify cardinal numbers with certain pure, well-founded
sets.
89
Finally, we turn to the set-theoretic construction of the real number sys-
tem. The construction emplys the usual factorization of a set by an equivalence
relation, as per the following definitions and remark.
Definition 4.9.7. Let A be a set. An equivalence relation on A is a binary
relation R ⊆ AA with the following properties: aRb and bRc imply aRc; aRb
implies bRa; and aRa for all a ∈ A.
Definition 4.9.8. Let R be an equivalence relation on A. For any a ∈ A we
write [a]
R
= ¦b ∈ A [ aRb¦, the equivalence class of a with respect to R. We
write A/R = ¦[a]
R
[ a ∈ A¦.
Remark 4.9.9. Let R be an equivalence relation on A. Then aRb if and only
if [a]
R
= [b]
R
. Moreover A/R is a partition of A, i.e., a collection of pairwise
disjoint sets whose union is A.
Definition 4.9.10 (the real number system). In order to define the real num-
ber system, we follow Dedekind and begin with the natural number system
(N, +, , 0, 1, =, <).
The integers are defined by putting Z = (N N)/≡
Z
, where (m, n) ≡
Z
(m

, n

) if and only if m+n

= m

+n. The ordered ring structure of Z is given
by
[(m, n)] +
Z
[(m

, n

)] = [(m+m

, n +n

)]
[(m, n)]
Z
[(m

, n

)] = [(mm

+nn

, mn

+m

n)]

Z
[(m, n)] = [(n, m)]
0
Z
= [(0, 0)]
1
Z
= [(1, 0)]
[(m, n)] = [(m

, n

)] ⇔ m+n

= m

+n
[(m, n)] <
Z
[(m

, n

)] ⇔ m+n

< m

+n
The rationals are defined by putting Q = (Z Z
+
)/≡
Q
, where Z
+
= ¦b ∈
Z [ b > 0¦, and (a, b) ≡
Q
(a

, b

) if and only if a b

= a

b. The ordered ring
structure of Q is given by
[(a, b)] +
Q
[(a

, b

)] = [(ab

+a

b, b b

)]
[(a, b)]
Q
[(a

, b

)] = [(aa

, bb

)]

Q
[(a, b)] = [(−a, b)]
0
Q
= [(0, 1)]
1
Q
= [(1, 1)]
[(a, b)] = [(a

, b

)] ⇔ ab

= a

b
[(a, b)] <
Q
[(a

, b

)] ⇔ ab

< a

b
Finally, the reals are defined by putting R = S/≡
R
. Here S is defined to be
the set of Cauchy sequences over Q, i.e., sequences 'q
n
`
n∈N
∈ Q
N
satisying
∀ε > 0 ∃m∀n(n > m ⇒ [q
m
− q
n
[ < ε) .
90
And ≡
R
is the equivalence relation on S defined by putting 'q
n
`
n

R
'q

n
`
n
if
and only if lim
n
[q
n
−q

n
[ = 0, i.e.,
∀ε > 0 ∃m∀n(n > m ⇒ [q
n
−q

n
[ < ε) .
The ordered ring structure of R is given by
['q`
n
] +
R
['q

n
`
n
] = ['q
n
+q

n
`
n
]
[ 'q
n
`
n
]
R
['q

n
`
n
] = ['q
n
q

n
`
n
]

R
['q
n
`
n
] = ['−q
n
`
n
]
0
R
= ['0`
n
]
1
R
= ['1`
n
]
[ 'q
n
`
n
] = ['q

n
`
n
] ⇔ ∀ε > 0 ∃m∀n(n > m ⇒ [q
n
−q

n
[ < ε)
[ 'q
n
`
n
] <
R
['q

n
`
n
] ⇔ ∃ε > 0 ∃m∀n(n > m ⇒ q
n
+ε < q

n
)
Exercise 4.9.11. Show that the real number system is complete, i.e., every
nonempty bounded subset of R has a least upper bound.
Remark 4.9.12. In this section we have shown how many or most mathemati-
cal objects may be redefined or reconstructed as pure, well-founded sets. In this
sense, pure set theory may be said to encompass virtually all of mathematics,
and one may speak of the set-theoretic foundations of mathematics. This is why
set theory is viewed as being of fundamental or foundational importance.
91
Chapter 5
Axiomatic Set Theory
This chapter is an introduction to axiomatic set theory.
5.1 The Axioms of Set Theory
Definition 5.1.1. We define L

, the language of set theory. The language
contains variables x, y, z, . . .. The atomic formulas of the language are x = y
and x ∈ y, where x and y are variables. Formulas are built up as usual from
atomic formulas by means of propositional connectives ∧, ∨, , ⇒, ⇔ and
quantifiers ∀, ∃. The notion of free variable is defined as usual. A sentence is a
formula with no free variables.
Remark 5.1.2. The standard or intended interpretation of L

is that the
variables are to range over the class of pure, well-founded sets. Thus formulas
and sentences are normally interpreted as making assertions about pure, well-
founded sets. This standard interpretation or model of L

is sometimes known
as the real world.
Example 5.1.3. An example of a sentence of L

is
∀x∀y (x = y ⇔ ∀u (u ∈ x ⇔ u ∈ y)) .
This sentence asserts the extensionality principle for pure, well-founded sets:
two pure, well-founded sets are equal if and only if they contain the same pure,
well-founded sets as elements.
Example 5.1.4. An example of a formula of L

is
∀u (u ∈ z ⇔ (u = x ∨ u = y)) .
This formula has free variables x, y, and z. It asserts that z is the unordered
pair ¦x, y¦, i.e., the set whose only elements are x and y.
92
As mentioned above, the standard interpretation of L

is in terms of the
real world, i.e., the class of pure, well-founded sets. In later sections of this
chapter, we shall consider interpretations or models of L

other than the real
world. Such alternative interpretations play an essential role in axiomatic set
theory.
We can expand the language L

indefinitely by introducing abbreviations.
Some important abbreviations are given in the following definition.
Definition 5.1.5.
1. (unordered pair) z = ¦x, y¦ is an abbreviation for
∀u (u ∈ z ⇔ (u = x ∨ u = y)).
2. (singleton) z = ¦x¦ is an abbreviation for z = ¦x, x¦.
3. (ordered pair) z = (x, y) is an abbreviation for z = ¦¦x¦, ¦x, y¦¦, i.e.,
∃u ∃v (u = ¦x¦ ∧ v = ¦x, y¦ ∧ z = ¦u, v¦).
4. (subset) x ⊆ y is an abbreviation for ∀u (u ∈ x ⇒ u ∈ y).
5. (powerset) z = {(x) is an abbreviation for
∀y (y ∈ z ⇔ y ⊆ x).
6. (union of a set of sets) z =
¸
x is an abbreviation for
∀u (u ∈ z ⇔ ∃v (v ∈ x ∧ u ∈ v)).
7. (union of two sets) z = x ∪ y is an abbreviation for z =
¸
¦x, y¦, i.e.,
∃w(w = ¦x, y¦ ∧ z =
¸
w).
8. (intersection of a set of sets) z =
¸
x is an abbreviation for
∀u (u ∈ z ⇔ ∀v (v ∈ x ⇒ u ∈ v)).
9. (intersection of two sets) z = x∩y is an abbreviation for z =
¸
¦x, y¦, i.e.,
∃w(w = ¦x, y¦ ∧ z =
¸
w).
10. (empty set) x = ∅ and x = ¦¦ are abbreviations for ∀u (u / ∈ x).
Using these abbreviations, we can write down sentences expressing some of
the axioms of Zermelo-Fraenkel set theory:
Definition 5.1.6 (Some Axioms of Set Theory).
93
1. Axiom of Extensionality: ∀x∀y (x = y ⇔ ∀u (u ∈ x ⇔ u ∈ y)).
2. Empty Set Axiom: ∃x(x = ∅).
3. Pairing Axiom: ∀x∀y ∃z (z = ¦x, y¦).
4. Union Axiom: ∀x∃z (z =
¸
x).
5. Power Set Axiom: ∀x∃z (z = {(x)).
6. Axiom of Foundation: ∀x(x = ∅ ⇒ ∃u (u ∈ x ∧ u ∩ x = ∅)).
Most of the above axioms are self-explanatory. Only the Axiom of Founda-
tion needs explanation. The Axiom of Foundation is an attempt to express the
idea that all of the sets under consideration are well-founded. This is expressed
by saying that, for all sets x, if x is nonempty then x contains an element u
which is ∈-minimal. Note that, for u ∈ x, u ∩ x = ∅ means that u is ∈-minimal
among elements of x, i.e., there is no element v of x such that v ∈ u.
We now introduce some more abbreviations and axioms.
Definition 5.1.7.
1. (Cartesian product) z = x y is an abbreviation of
∀w(w ∈ z ⇔ ∃u ∃v (u ∈ x ∧ v ∈ y ∧ w = (u, v))).
2. (function) Fcn(f) is an abbreviation for a formula saying that f is a func-
tion, i.e.,
∀w(w ∈ f ⇒ ∃x∃y (w = (x, y)))∧∀x∀y ∀z (((x, y) ∈ f∧(x, z) ∈ f) ⇒ y = z) .
3. (value of a function) y = f(x) is an abbreviation of
Fcn(f) ∧ (x, y) ∈ f.
4. (domain of a function) z = dom(f) is an abbreviation of
Fcn(f) ∧ ∀x(x ∈ z ⇔ ∃y (x, y) ∈ f).
5. (generalized Cartesian product) z =
¸
f is an abbreviation of
Fcn(f)∧∀g (g ∈ z ⇔ (dom(g) = dom(f)∧∀x(x ∈ dom(f) ⇒ g(x) ∈ f(x)))) .
Definition 5.1.8. The Axiom of Choice is the sentence
∀f ((Fcn(f) ∧ ∀x(x ∈ dom(f) ⇒ f(x) = ∅)) ⇒
¸
f = ∅).
Definition 5.1.9. The Axiom of Infinity is the sentence
94
∃z (∅ ∈ z ∧ ∀x(x ∈ z ⇒ x ∪ ¦x¦ ∈ z)).
The purpose of the Axiom of Infinity is to assert the existence of at least
one infinite set. This is accomplished by asserting the existence of a set that
contains all of the sets 0 = ¦¦ = ∅, 1 = ¦0¦, 2 = ¦0, 1¦, 3 = ¦0, 1, 2¦, . . . .
We now introduce the two remaining axioms of Zermelo-Fraenkel set theory.
Actually, these two so-called axioms are not individual axioms, but rather axiom
schemes. An axiom scheme is an infinite set of axioms all of which have a
common form.
Recall that, if F is a formula with free variables x
1
, . . . , x
n
, then the universal
closure of F is the sentence ∀x
1
∀x
n
F.
Definition 5.1.10.
1. The Comprehension Scheme is an infinite set of axioms, consisting the
universal closures of all formulas of the form
∃z ∀u (u ∈ z ⇔ (u ∈ x ∧ F(u))) ,
where F(u) is any formula in which z does not occur freely.
2. For any formula F(u), we write z = ¦u ∈ x [ F(u)¦ as an abbreviation for
∀u (u ∈ z ⇔ (u ∈ x ∧ F(u))) .
The Comprehension Scheme is our attempt to express the principle that,
given a set x and a property P that particular elements of x may or may
not have, there necessarily exists a set z ⊆ x consisting of all elements of x
which have the given property P. Since our language L

does not enable us to
discuss or quantify over arbitrary properties, we restrict attention to properties
that are definable, i.e., expressible by means of a formula F(u). The syntactical
requirement that the variable z does not occur freely in F(u) is imposed in order
to avoid obvious contradictions such as z = ¦u ∈ x [ u / ∈ z¦, the idea being that
the set z should be in some sense logically subordinate to the property P.
The Comprehension Scheme is extremely useful and important. For example,
given a function f, the Comprehension Scheme together with the Union, Pairing,
and Power Set Axioms logically imply the existence of a set z which is the domain
of f,
z = domf = ¦x ∈
¸¸
f [ ∃y ((x, y) ∈ f)¦,
and of the generalized Cartesian product
¸
f = ¦g ∈ {{{(
¸¸
f ∪
¸¸¸
f) [ domg = z ∧ ∀x(x ∈ z ⇒ g(x) ∈ f(x))¦.
Definition 5.1.11.
1. We write ∃ ! x to mean “there exists exactly one x such that”. In other
words, for any formula F(x) in which x occurs as a free variable, ∃ ! xF(x)
is an abbreviation of
∃y ∀x(F(x) ⇔ x = y) .
95
2. The Replacement Scheme is an infinite set of axioms, consisting of the
universal closures of all formulas of the form
∀u (u ∈ x ⇒ ∃ ! v F(u, v)) ⇒ ∃y ∀v (v ∈ y ⇔ ∃u (u ∈ x ∧ F(u, v))) ,
where F(u, v) is any formula in which y does not occur freely.
The Replacement Scheme is our attempt to express the principle that, given
a set x and a rule associating to each element u of x a unique object v, there
exists a set y consisting of all the objects v which are associated to elements of x.
Since our language L

does not enable us to discuss or quantify over arbitrary
rules, we restrict attention to rules that are definable, i.e., expressible by means
of a formula F(u, v). The syntactical requirement that the variable y does not
occur freely in F(u, v) is imposed in order to avoid obvious contradictions.
We have now introduced all of the axioms of Zermelo/Fraenkel set theory.
We have, finally:
Definition 5.1.12 (Zermelo/Fraenkel Set Theory). The axioms of Zermelo/Fraenkel
set theory are as follows: the Axiom of Extensionality, the Empty Set Axiom,
the Pairing Axiom, the Union Axiom, the Power Set Axiom, the Axiom of Foun-
dation, the Axiom of Infinity, the Comprehension Scheme, and the Replacement
Scheme. We use ZF as an abbreviation for “Zermelo/Fraenkel set theory”.
Definition 5.1.13 (ZFC). The axioms of ZFC consist of the axioms of ZF plus
the Axiom of Choice.
The Zermelo/Fraenkel axioms together with the Axiom of Choice constitute
the commonly accepted, rigorous, set-theoretic foundation of mathematics. A
mathematical theorem is regarded as proved if and only if it is clear how to de-
duce it as a theorem of ZFC, i.e., a logical consequence
1
of the ZFC axioms. It
can be shown that all of the theorems of 19th and 20th century rigorous math-
ematics are logical consequences of the ZFC axioms. In particular, essentially
all of the results of Chapter 4 can be stated and proved as theorems of ZFC.
5.2 Models of Set Theory
As mentioned above, the intended interpretation of L

is the so-called real world,
i.e., the class of pure, well-founded sets. However, the general notion of pure,
well-founded set is rather vague. In order to study and delimit this vagueness,
axiomatic set theorists frequently consider alternative interpretations of L

.
One important class of interpretations of L

is given in terms of relational
structures. Recall that a relational structure is an ordered pair (A, E) where A
is a set and E ⊆ A A. Given a relational structure (A, E) and a sentence F
of L

, it makes sense to ask whether F is true in (A, E), i.e., true when the
variables are interpreted as ranging over A and x ∈ y is interpreted as xEy, i.e.,
(x, y) ∈ E.
1
The notions of theorem and logical consequence that we are using here will be explained
in the next section.
96
Examples 5.2.1. The Axiom of Extensionality is true in a particular relational
structure (A, E) if and only if (A, E) is extensional, i.e., for all a, b ∈ A, a =
b implies ¦c [ cEa¦ = ¦c [ cEb¦. Note that some relational structures are
extensional and some are not. An example of an extensional relational structure
is any linear ordering. An example of a nonextensional relational structure is
(A, E) whenever [A[ ≥ 2 and E = ∅.
Thus a relational structure is extensional if and only if it is a model of
(i.e., satisfies) the Axiom of Extensionality. The general point here is that any
collection of sentences of L

defines a property of relational structures, namely
the property of satisfying the given sentences. We formalize this in the following
definition.
Definition 5.2.2. Let S be any set of sentences of L

. A model of S is a
relational structure (A, E) such that all of the sentences of S are true in (A, E).
We say that S is consistent if there exists a model of S. If F is another sentence
of L

, we say that F is a logical consequence of S, written S ⊢ F, if F is true
in every model of S.
Note that if S is inconsistent then all sentences are logical consequences of S,
so the notion of logical consequence is uninteresting in this case. If however S is
consistent, then it is meaningful to ask which sentences are logical consequences
of S, i.e., what conclusions follow when the sentences of S are assumed as
axioms. This is the kind of question which axiomatic set theory seeks to answer.
Naturally the focus is on sets of sentences which make assertions that could
reasonably be true in the intended model, i.e., the real world, i.e., the class of
all pure, well-founded sets.
As an easy example of the notion of logical consequence, note that the sen-
tence ∀x(x / ∈ x) is a logical consequence of the Axiom of Foundation plus the
Pairing Axiom. This is so because x ∈ x would imply that ¦x¦ has no ∈-minimal
element.
Specializing Definition 5.2.2 to S = ZF and S = ZFC, we have:
Definition 5.2.3. A model of ZFC is a relational structure (A, E) such that
(A, E) satisfies all of the Zermelo/Fraenkel axioms plus the Axiom of Choice.
A theorem of ZFC is any sentence which is a logical consequence of the ZFC
axioms. The notions model of ZF and theorem of ZF are defined similarly.
Axiomatic set theory is essentially the study of models of ZF and of ZFC,
with an eye to discovering which set-theoretic propositions follow or do not
follow from these axiom systems. For instance, one of the important results
2
of
axiomatic set theory is that there exists a model of ZF which is not a model of
ZFC. In other words, the Axiom of Choice is not a logical consequence of the
ZF axioms. Another key result is that both the Continuum Hypothesis and its
negation are consistent with ZFC. In other words, CH is independent of ZFC.
Thus the ZFC axioms, although powerful and flexible, do not suffice to answer
basic set-theoretic questions such as the Continuum Problem.
2
This result will not be proved here.
97
We end this section by presenting some general definitions and results con-
cerning relational structures.
Definition 5.2.4. Let (A, E) be a relational structure. A k-place predicate P ⊆
A
k
is said to be definable over (A, E) if there exists a formula F(x
1
, . . . , x
k
, y
1
, . . . , y
m
)
of L

and parameters b
1
, . . . , b
m
∈ A such that
P = ¦'a
1
, . . . , a
k
` ∈ A
k
[ (A, E) satisfies F(a
1
, . . . , a
k
, b
1
, . . . , b
m
)¦ . (5.1)
More generally, given B ⊆ A, we say that P ⊆ A
k
is definable over (A, E)
allowing parameters from B if there exists a formula
F(x
1
, . . . , x
k
, y
1
, . . . , y
m
)
of L

and parameters b
1
, . . . , b
m
∈ B such that (5.1) holds.
Definition 5.2.5. Let (A, E) be a relational structure. Then Def((A, E)) is the
set of all subsets of A that are definable over (A, E). Note that Def((A, E)) ⊆
{(A).
Lemma 5.2.6. Let (A, E) be a relational structure.
1. If A is finite, then Def((A, E)) = {(A) and [Def((A, E)[ = 2
|A|
.
2. If A is infinite, then [Def((A, E))[ = [A[.
Proof. For finite A the result is obvious. Suppose now that A is infinite. By
G¨odel numbering, the set of all formulas of L

is countable. Since any element
of Def ((A, E)) is determined by a formula and a finite sequence of parameters
from A, we have
[Def((A, E))[ ≤ ℵ
0
[A

[ = [A[ .
On the other hand ¦¦a¦ [ a ∈ A¦ ⊆ Def((A, E)), hence [A[ ≤ [Def((A, E))[.
This completes the proof.
Definition 5.2.7. Let (A, E) and (A

, E

) be relational structures. We say that
(A

, E

) is a substructure of (A, E), abbreviated (A

, E

) ⊆ (A, E), if A

⊆ A
and E

= E ∩ (A

A

). We say that (A

, E

) is an elementary substructure
of (A, E), abbreviated (A

, E

) ⊆
elem
(A, E), if (A

, E

) ⊆ (A, E) and, for all
formulas F(x
1
, . . . , x
n
) and a
1
, . . . , a
n
∈ A

, (A, E) satisfies F(a
1
, . . . , a
n
) if and
only if (A

, E

) satisfies F(a
1
, . . . , a
n
).
Lemma 5.2.8. Given (A

, E

) ⊆ (A, E), we have (A

, E

) ⊆
elem
(A, E) if and
only if every nonempty subset of A which is definable over (A, E) allowing
parameters from A

has a nonempty intersection with A

.
Proof. Straightforward.
Recall that a relational structure (A, E) is said to be countable if the under-
lying set A is countable.
98
Theorem 5.2.9 (L¨owenheim/Skolem Theorem). For any relational structure
(A, E), there exists a countable elementary substructure (A

, E

) ⊆
elem
(A, E).
Proof. Let c : {(A) ` ¦∅¦ → A be a choice function for A. We define recursively
a sequence of sets A
n
⊆ A, n ∈ N, by A
0
= ∅, A
n+1
= ¦c(X) [ ∅ = X ⊆ A∧X is
definable over (A, E) allowing parameters from A
n
¦. Note that A
n
⊆ A
n+1
for
all n. By induction on n it is straightforward to show that A
n
is countable for all
n. Hence A

=
¸
¦A
n
[ n ∈ N¦ is countable. Moreover c(X) ∈ A

for all X = ∅
definable over (A, E) allowing parameters from A

. Hence by Lemma 5.2.8 we
have (A

, E

) ⊆
elem
(A, E), where E

= E ∩ (A

A

). This completes the
proof.
Corollary 5.2.10 (Skolem Paradox). If ZFC is consistent, then there exists a
countable model of ZFC.
Proof. Assume that ZFC is consistent. Then there exists a model (A, E) of
ZFC. By Theorem 5.2.9 (A, E) has a countable elementary submodel, (A

, E

).
Then (A

, E

) is a countable model of ZFC.
The Skolem Paradox is called a paradox for the following reason: the exis-
tence of a countable model of ZFC would seem to contradict the fact that the
existence of uncountable sets is a theorem of ZFC. Actually, there is no contra-
diction here, because a set that is uncountable within a particular model (A, E)
may be countable in the real world. In other words, the notion of countability
is relative to the model under consideration (as are many other set-theoretic
notions).
A straightforward generalization of Theorem 5.2.9 is:
Theorem 5.2.11 (Generalized L¨ owenheim-Skolem Theorem). Let κ be an in-
finite cardinal. Let (A, E) be a relational structure such that [A[ ≥ κ, and let
X ⊆ A be a subset of A such that [X[ ≤ κ. Then there exists an elementary
substructure (A

, E

) of (A, E) such that X ⊆ A

and [A

[ = κ.
Proof. Straightforward.
Exercise 5.2.12. Prove Theorem 5.2.11.
5.3 Transitive Models and Inaccessible Cardi-
nals
In this section we study an important class of models. Recall that a set T is
transitive if and only if every element of T is a subset of T.
Definition 5.3.1. Let S be a set of sentences of L

. A transitive model of
S is any transitive, pure, well-founded set T such that the relational structure
(T, ∈[T) satisfies all the sentences of S. In this context it is customary to identify
the transitive set T with the relational structure (T, ∈[T).
99
Note that every transitive model is well-founded and extensional. The fol-
lowing theorem provides a converse and thereby characterizes transitive models
up to isomorphism among arbitrary models.
Theorem 5.3.2. Let (A, E) be a relational structure which is well-founded
and extensional. Then there exists a transitive, pure, well-founded set T such
that the relational structures (A, E) and (T, ∈[T) are isomorphic. Moreover the
transitive set T and the isomorphism f : (A, E)

= (T, ∈[T) are unique.
Proof. Fix an object a
0
/ ∈ A. By transfinite recursion on the rank of an arbitrary
pure, well-founded set x, define F(x) as follows: F(x) = the unique a ∈ A such
that rng(F↾x) = ¦b [ bEa¦ if such an a exists; F(x) = a
0
otherwise. Note
that F(x) = F(y) ∈ A implies x = y. Hence T = rng(F
−1
↾A) is a set. It is
easy to verify that T is a transitive, pure, well-founded set and that F↾T is an
isomorphism of (T, ∈[T) onto (A, E). Hence f = F
−1
↾A is an isomorphism of
(A, E) onto (T, ∈[T). It is straightforward to verify that T and f are unique.
Corollary 5.3.3. If (A, E) is a relational structure, then the following asser-
tions are equivalent:
1. (A, E) is isomorphic to some transitive model (T, ∈[T);
2. (A, E) is well-founded and extensional.
The rest of this section is devoted to the study of transitive models, i.e.,
relational structures of the form (A, ∈[A) where T is a transitive, pure, well-
founded set. The following definition concerning transitive models is of general
interest.
Definition 5.3.4. Let T be any transitive, pure, well-founded set. A k-place
predicate P ⊆ T
k
is said to be definable over T if and only if it is definable over
(T, ∈[T) (allowing parameters from T). We write
Def(T) = Def((T, ∈[T)) = ¦X ⊆ T [ X is definable over T¦ .
Of particular interest are transitive models of ZFC. The following lemma
consists of some simple remarks characterizing which transitive models satisfy
which axioms of ZFC.
Lemma 5.3.5. Let T be a transitive, pure, well-founded set.
1. T always satisfies the Axiom of Extensionality.
2. T always satisfies the Axiom of Foundation.
3. T satisfies the Pairing Axiom if and only if T is closed under pairing, i.e.,
∀a ∀b ((a ∈ T ∧ b ∈ T) ⇒ ¦a, b¦ ∈ T).
4. T satisfies the Union Axiom if and only if T is closed under union, i.e.,
100
∀a (a ∈ T ⇒
¸
a ∈ T).
5. T satisfies the Empty Set Axiom if and only if ∅ ∈ T.
6. T satisfies the Axiom of Infinity if and only if ∃a (a ∈ T ∧ ω ⊆ a).
7. T satisfies the Power Set Axiom if and only if
∀a (a ∈ T ⇒ {(a) ∩ T ∈ T).
8. T satisfies the Axiom of Choice if and only if, for every indexed family of
nonempty sets 'a
i
`
i∈I
∈ T, we have T ∩
¸
i∈I
a
i
= ∅.
9. T satisfies the Comprehension Scheme if and only if, for all X ∈ Def(T),
we have
∀a (a ∈ T ⇒ a ∩ X ∈ T).
10. T satisfies the Replacement Scheme if and only if for all functions F : T →
T such that F ∈ Def(T), we have
∀a (a ∈ T ⇒ rng(F↾a) ∈ T).
Proof. Straightforward.
We shall now show that inaccessible cardinals give rise to transitive models
of ZFC. Recall that an inaccessible cardinal is a regular, uncountable, strong
limit cardinal. Recall also that we have identified cardinals with initial von
Neumann ordinals (cf. Sections 4.4.5 and 4.4.8).
Lemma 5.3.6. Let δ be a limit ordinal > ω. Then R
δ
is a transitive model of
all of the ZFC axioms except possibly the Replacement Scheme.
Proof. We apply Lemma 5.3.5. The Axioms of Extensionality and Foundation
hold in R
δ
because R
δ
is a transitive, pure, well-founded set. The Empty
Set, Power Set, Pairing, and Union Axioms and the Axiom of Choice and the
Comprehension Scheme hold in R
δ
because δ is a limit ordinal. The Axiom of
Infinity holds in R
δ
because ω ∈ R
δ
, since ω < δ.
Lemma 5.3.7. An infinite cardinal λ is regular if and only if, for all X ⊆ λ,
[X[ < λ implies sup X < λ.
Proof. Straightforward.
Lemma 5.3.8. If λ is an inaccessible cardinal, then
1. ∀x((x ⊆ R
λ
∧ [x[ < λ) ⇒ x ∈ R
λ
).
2. ∀x(x ∈ R
λ
⇒ [x[ < λ).
101
3. [R
λ
[ = λ.
Proof. 1. Define ρ : x → λ by ρ(u) = rank(u). Then [rngρ[ ≤ [x[ < λ.
Since λ is regular, it follows by the previous lemma that sup(rngρ) < λ, say
rngρ ⊆ α < λ. Hence x ⊆ R
α
, hence x ∈ R
α+1
⊆ R
λ
since λ is a limit ordinal.
2. By transfinite induction on α < λ we prove [R
α
[ < λ. We have R
0
= ∅.
If [R
α
[ = κ < λ, then [R
α+1
[ = [{(R
α
)[ = 2
κ
< λ since λ is a strong limit
cardinal. For limit ordinals δ < λ, we have inductively [R
δ
[ = [
¸
α<δ
R
α
[ =
sup
α<δ
[R
α
[ < λ, since [R
α
[ < λ and λ is regular.
3. [R
λ
[ = sup
α<λ
[R
α
[ = λ.
Theorem 5.3.9. Let λ be an inaccessible cardinal. Then R
λ
is a transitive
model of ZFC.
Proof. Clearly λ is a limit ordinal > ω, hence by Lemma 5.3.6 we see that R
λ
satisfies all of the ZFC axioms except possibly the Replacement Scheme.
Let F : R
λ
→ R
λ
and a ∈ R
λ
be given. Then rng(F↾a) ⊆ R
λ
and
[rng(F↾a)[ ≤ [a[ < λ, hence by the previous lemma rng(F↾a) ∈ R
λ
. Spe-
cializing this to the case when F is definable over R
λ
, we see by Lemma 5.3.5
that the Replacement Scheme holds in R
λ
. This completes the proof.
Corollary 5.3.10. If there exists an inaccessible cardinal, then ZFC is consis-
tent.
Proof. Immediate from the theorem.
Exercise 5.3.11. A hereditarily finite set is a finite set x such that all elements
of x, elements of elements of x, . . . , are finite sets. Show that R
ω
is the set of
all hereditarily finite, pure, well-founded sets. Show that R
ω
is a model of all
of the axioms of ZFC except the Axiom of Infinity.
Exercise 5.3.12. Define E ⊆ N by putting mEn if and only if 2
m
occurs in
the binary expansion of n, i.e., m = n
i
for some i where n = 2
n1
+ + 2
n
k
.
1. Show that (N, E)

= (R
ω
, ∈[R
ω
).
2. Conclude that P ⊆ ω
k
is definable over R
ω
if and only if P is arithmetical.
Theorem 5.3.13. If there exists an inaccessible cardinal, then the existence of
an inaccessible cardinal is not a theorem of ZFC.
Proof. Assume that there exists an inaccessible cardinal. Let λ be the smallest
inaccessible cardinal. By the previous theorem, R
λ
is a model of ZFC. We claim
that R
λ
also satisfies “inaccessible cardinals do not exist”. To see this, suppose
that R
λ
satisfies “there exists at least one inaccessible cardinal”. Let κ ∈ R
λ
be such that R
λ
satisfies “κ is an inaccessible cardinal”. Then it is easy to see
that κ is also an inaccessible cardinal in the real world. But clearly κ < λ.
This contradicts the choice of λ. Thus R
λ
is a model of ZFC + “inaccessible
cardinals do not exist”.
102
The previous theorem shows that, if we assume only the axioms of ZFC,
then we cannot hope to prove the existence of inaccessible cardinals.
Exercise 5.3.14. Show that, if two or more inaccessible cardinals exist, then
the existence of two or more inaccessible cardinals is not a theorem of ZFC +
“there exists at least one inaccessible cardinal”.
Theorem 5.3.15. If there exists an inaccessible cardinal, then there exists a
countable, transitive model of ZFC.
Proof. Let λ be an inaccessible cardinal. Then (R
λ
, ∈[R
λ
) is a model of ZFC.
By the L¨ owenheim-Skolem Theorem, there exists a countable set A ⊆ R
λ
such
that (A, ∈[A) is an elementary submodel of (R
λ
, ∈[R
λ
). Thus (A, ∈[A) is a
countable, well-founded, extensional model of ZFC. By Theorem 5.3.2, (A, ∈[A)
is isomorphic to a transitive model (T, ∈[T). Thus (T, ∈[T) is a countable,
transitive model of ZFC.
Exercise 5.3.16. Let λ be an inaccessible cardinal. Prove that there ex-
ists a limit ordinal δ < λ such that (R
δ
, ∈[R
δ
) is an elementary submodel of
(R
λ
, ∈[R
λ
).
5.4 Constructible Sets
Recall that, if T is any transitive, pure, well-founded set, Def(T) is the set of all
subsets of T that are definable over T (i.e., over (T, ∈[T)) allowing parameters
from T.
Definition 5.4.1. By transfinite recursion we define L
α
, α ∈ Ord, as follows:
L
0
= ∅
L
α+1
= Def(L
α
)
L
δ
=
¸
α<δ
L
α
for limit ordinals δ .
A set X is said to be constructible if X ∈ L
α
for some ordinal α. The class of
all constructible sets is denoted L.
Lemma 5.4.2. For all ordinals α, L
α
is a transitive, pure, well-founded set,
and L
α
⊆ R
α
.
Proof. For any transitive, pure, well-founded set T, we have T ⊆ Def(T) ⊆ {(T)
and hence Def(T) is again a transitive, pure, well-founded set. With these
observations, the lemma follows easily by transfinite induction on α.
Lemma 5.4.3. For all ordinals α, we have α = L
α
∩ Ord.
Proof. If T is any transitive, pure, well-founded set, then for any a ∈ T we
have that a is an ordinal (i.e., a von Neumann ordinal) if and only if T satisfies
“a is transitive and (a, ∈[a) is a linear ordering”. Thus T ∩ Ord ∈ Def(T).
With this observation, the lemma follows easily by transfinite induction on (von
Neumann) ordinals α.
103
We are going to show that the constructible sets form a model of ZFC.
Some of the axioms of ZFC are straightforwardly verified in L using Lemma 5.3.5.
For instance, the Union Axiom holds in L because x ∈ L
α
implies
¸
x ∈ L
α
.
The Pairing Axiom holds in L because x, y ∈ L
α
implies ¦x, y¦ ∈ L
α+1
. The
Empty Set Axiom holds in L because ∅ ∈ L
1
. The Axiom of Infinity holds in
L because ω ∈ L
ω+1
. The Axioms of Extensionality and Foundation hold in L
because L is transitive and consists of pure, well-founded sets.
To show that the Power Set Axiom holds in L, let X be any constructible
set. For each Y ∈ {(X) ∩ L put ρ(Y ) = the least β such that Y ∈ L
β
. Put
α = sup¦ρ(Y ) [ Y ∈ {(X) ∩ L¦. Thus {(X) ∩ L ⊆ L
α
. Hence {(X) ∩ L is
definable over L
α
; namely, it is the set of all Y ∈ L
α
such that L
α
satisfies
Y ⊆ X. Hence {(X) ∩ L ∈ Def(L
α
) = L
α+1
. We have now shown that for all
X ∈ L, {(X) ∩L ∈ L. From this it follows by Lemma 5.3.5 that the Power Set
Axiom holds in L.
To show that Comprehension and Replacement hold in L, we shall need the
following lemmas.
Lemma 5.4.4. Let f
1
, . . . , f
k
be functions from Ord to Ord. Then there exist
arbitrarily large ordinals α such that α is closed under f
1
, . . . , f
k
, i.e., f
i
(β) < α
for all β < α, 1 ≤ i ≤ k.
Proof. Given an ordinal γ, define an increasing sequence of ordinals α
n
, n ∈ N
inductively by α
0
= γ, α
n+1
= max(α
n
+ 1, sup¦f
i
(β) [ β < α
n
, 1 ≤ i ≤
k¦). Putting α = sup¦α
n
[ n ∈ N¦ we see that α > γ and α is closed under
f
1
, . . . , f
k
.
Lemma 5.4.5 (reflection). Let F(x
1
, . . . , x
n
) be a formula of L

with free vari-
ables x
1
, . . . , x
n
. Then there exist arbitrarily large ordinals α such that, for all
a
1
, . . . , a
n
∈ L
α
, L satisfies F(a
1
, . . . , a
n
) if and only if L
α
satisfies F(a
1
, . . . , a
n
).
Proof. Replacing ∀ by ∃ as necessary, we may safely assume that F contains
no occurrences of ∀. Now let ∃y G
i
, i = 1, . . . , k be a list of the subformulas of F
of the form ∃y G. Write G
i
≡ G
i
(y, x
i1
, . . . , x
ini
) where x
i1
, . . . , x
ini
are the free
variables of ∃y G
i
. For a
1
, . . . , a
ni
∈ L, put g
i
(a
1
, . . . , a
ni
) = the least ordinal
β such that a
1
, . . . , a
ni
∈ L
β
and such that, if L satisfies ∃y G
i
(y, a
1
, . . . , a
ni
),
then L satisfies G
i
(b, a
1
, . . . , a
ni
) for some b ∈ L
β
. Define f
i
: Ord → Ord by
f
i
(β) = sup¦g
i
(a
1
, . . . , a
ni
) [ a
1
, . . . , a
ni
∈ L
β
¦. By the previous lemma, there
exist arbitrarily large ordinals α such that that α is closed under f
1
, . . . , f
k
. It
is straightforward to verify that such an α has the desired property.
Remark 5.4.6. The proof of the previous lemma used only the following prop-
erties of the constructible hierarchy: α ≤ β implies L
α
⊆ L
β
; and L
δ
=
¸
α<δ
L
α
for limit ordinals δ. Since the R
α
hierarchy also has these properties, the same
lemma holds for the R
α
hierarchy as well. This has the following interesting
consequence: If F
1
, . . . , F
k
is a finite set of sentences that are true in the real
world, then there exist arbitrarily large ordinals α such that F
1
, . . . , F
k
are true
in R
α
.
104
Lemma 5.4.7. L satisfies the Comprehension and Replacement Schemes.
Proof. To show that the Replacement Scheme holds in L, let f : L → L be a
function which is definable over L. We must prove that, for all a ∈ L, rng(f↾a) =
¦f(u) [ u ∈ a¦ also belongs to L. Note first that, since f is definable over L,
we have parameters c
1
, . . . , c
n
∈ L and a formula F(u, v, z
1
, . . . , z
n
) with free
variables u, v, z
1
, . . . , z
n
such that, for all u ∈ L, f(u) = the unique v ∈ L such
that L satisfies F(u, v, c
1
, . . . , c
n
). Now given a ∈ L, put b = rng(f↾a). We
must show that b ∈ L. Let β be an ordinal so large that a, c
1
, . . . , c
n
∈ L
β
and b ⊆ L
β
. By Reflection, let α be such that α > β and, for all u, v ∈ L
α
,
L satisfies F(u, v, c
1
, . . . , c
n
) if and only if L
α
satisfies F(u, v, c
1
, . . . , c
n
). We
claim that b is definable over L
α
. This is clear since
b = ¦v ∈ L [ L satisfies ∃u (u ∈ a ∧ F(u, v, c
1
, . . . , c
n
))¦
= ¦v ∈ L
α
[ L
α
satisfies ∃u (u ∈ a ∧ F(u, v, c
1
, . . . , c
n
))¦ .
Thus b ∈ Def(L
α
) = L
α+1
, whence b ∈ L. This shows that the Replacement
Scheme holds in L. The proof that the Comprehension Scheme holds in L is
similar.
We introduce some more abbreviations:
Definition 5.4.8.
1. Const(x) is an abbreviation for a formula asserting that a given pure, well-
founded set x is constructible. In more detail, Const(x) asserts the exis-
tence of a transfinite sequence of sets 'L
β
`
β≤α
such that L
β
=
¸
¦Def(L
γ
) [
γ < β¦ for all β ≤ α, and x ∈ L
α
.
2. Recall that V is the class of all pure, well-founded sets, and L is the
class of all constructible sets. We use V = L to abbreviate ∀xConst(x).
Thus V = L is a sentence asserting that all pure, well-founded sets are
constructible.
Theorem 5.4.9. The class L of constructible sets satisfies the ZF axioms plus
V = L.
Proof. The above lemmas show that L satisfies the ZF axioms. It is tedious but
straightforward to show that L satisfies V = L.
Lemma 5.4.10. For all ordinals α ≥ ω, we have [L
α
[ = [α[.
Proof. By Lemma 5.2.6 we have [L
ω
[ = ℵ
0
and, for α ≥ ω, [L
α+1
[ = [Def(L
α
)[ =
[L
α
[. From this the lemma easily follows by induction on α ≥ ω.
Theorem 5.4.11. The class L of constructible sets satisfies the Axiom of
Choice.
105
Proof. Lemma 5.4.10 implies that each L
α
is well-orderable. Refining the proof
of Lemma 5.4.10, we can explicitly define by transfinite recursion a function
F : Ord → V such that, for all ordinals α, F(α) is a well-ordering of L
α
. Since
the definition of F is explicit, its validity does not depend on the Axiom of
Choice. Hence by Theorem 5.4.9 the definition of F can be carried out within L.
In particular L satisfies that each L
α
is well-orderable. Hence by Remark 4.5.3
L satisfies the Axiom of Choice. This argument actually shows that the Axiom
of Choice follows from ZF plus V = L.
Our remaining goal with respect to constructible sets is to show that L
satisfies the GCH.
Lemma 5.4.12. There is a sentence S of L

with the following property. For
all transitive sets A, A satisfies S if and only if A = L
α
for some limit ordinal
α.
Proof. The construction of the sentence S is straightforward but tedious. Roughly
speaking, S is identical with the sentence V = L of Definition 5.4.8. For de-
tails of the construction of S, see Boolos and Putnam, “Degrees of unsolvability
of constructible sets of integers,” Journal of Symbolic Logic, Volume 33, 1968,
pages 497–513.
Lemma 5.4.13. If a is any constructible subset of ω, then a ∈ L
α
for some
countable ordinal α. More generally, if a ∈ {(κ) ∩ L where κ is an infinite
cardinal, then a ∈ L
α
for some ordinal α such that [L
α
[ = κ.
Proof. Let κ be an infinite cardinal. Suppose that a ⊆ κ and a is constructible.
Let δ > κ be a limit ordinal such that a ∈ L
δ
. By the Generalized L¨ owenheim/-
Skolem Theorem (Theorem 5.2.11), we can find a set A ⊆ L
δ
such that κ ∪
¦a¦ ⊆ A, [A[ = κ, and (A, ∈[A) is an elementary submodel of (L
δ
, ∈[L
δ
). By
Theorem 5.3.2 and Lemma 5.4.12, we have (A, ∈[A)

= (L
α
, ∈[L
α
) for some
limit ordinal α. Since κ ∪ ¦a¦ is a transitive subset of A, it follows by another
application of Theorem 5.3.2 that κ ∪ ¦a¦ ⊆ L
α
. In particular a ∈ L
α
. Clearly
[L
α
[ = κ, and this completes the proof.
Lemma 5.4.14. For any infinite cardinal κ, we have [{(κ) ∩ L[ ≤ κ
+
.
Proof. From the previous lemma we have {(κ) ∩L ⊆ L
κ
+. The desired conclu-
sion is immediate, since [L
κ
+[ = [κ
+
[.
Theorem 5.4.15. The class L of constructible sets satisfies the Generalized
Continuum Hypothesis.
Proof. Since L satisfies the axioms of set theory, the proof of the previous lemma
can be carried out within L. Thus for all infinite cardinals κ of L, we have within
L that [{(κ)[ = κ
+
, hence 2
κ
= κ
+
. This proves the theorem.
Theorem 5.4.16.
106
1. If ZF has a transitive model, then so does ZFC + GCH.
2. If ZF is consistent, then so is ZFC + GCH.
Proof. Let A be a transitive model of ZF. Within A we can carry out the
definition of L to obtain a transitive submodel B (sometimes called an “inner
model”) consisting of the constructible sets of A. (It can be shown that B =
L
α
where α is the least ordinal that is not an element of A.) The proofs of
theorems 5.4.9, 5.4.11, and 5.4.15 then show that B is a model of ZF plus
V = L plus the Axiom of Choice plus the GCH. This proves the first part. The
proof of the second part is similar, starting with a model (A, E) that is not
necessarily transitive.
Remark 5.4.17. The previous theorem, due to G¨odel 1939, is one of the most
significant achievements of axiomatic set theory. The second part is sometimes
described as a relative consistency result: ZFC + GCH is consistent relative to
ZF.
5.5 Forcing
Let M be a countable transitive model of ZFC. Let P = (P, ≤) be a partially
ordered set belonging to M. We say that p, q ∈ P are compatible if there exists
r ∈ P such that r ≤ p and r ≤ q. If p, q ∈ P are incompatible, we write p ⊥ q.
Definition 5.5.1. A filter on P is a set G ⊆ P such that
1. p, q ∈ G implies ∃r ∈ G(r ≤ p, q);
2. p ∈ G, p ≤ q imply q ∈ G.
Definition 5.5.2. D ⊆ P is dense open if
1. ∀p ∈ P ∃q ≤ p (q ∈ D);
2. ∀p ∈ D∀q ≤ p (q ∈ D).
Definition 5.5.3. A filter G ⊆ P is said to be M-generic if G∩ D = ∅ for all
dense open D ⊆ P such that D ∈ M.
Lemma 5.5.4. Given p ∈ P we can find an M-generic filter G ⊆ P such that
p ∈ G.
Proof. Let ¦D
n
[ n ∈ N¦ be an enumeration of ¦D ∈ M [ D dense open in P¦.
Put p
0
= p and, given p
n
, let p
n+1
≤ p
n
be such that p
n+1
∈ D
n
. It is easy to
verify that G = ¦q ∈ P [ ∃n(p
n
≤ q)¦ is an M-generic filter.
Definition 5.5.5. Let G be an M-generic filter. We put
M[G] = ¦a
G
[ a ∈ M¦,
107
where
a
G
= ¦b
G
[ (p, b) ∈ a for some p ∈ G¦.
Remark 5.5.6. Think of each a ∈ M as a “name” for a
G
∈ M[G]. We shall
show that M[G] is a countable transitive model of ZFC containing M.
Lemma 5.5.7. M[G] is a countable transitive set. We have M[G] ⊇ M ∪¦G¦,
and Ord ∩ M[G] = Ord ∩ M.
Proof. It is obvious from the definition that M[G] is a countable transitive set.
For all a ∈ M we have a = ( ˙ a)
G
, where ˙ a = P ¦
˙
b [ b ∈ a¦. We also have
G = (
˙
G)
G
, where
˙
G = ¦(p, ˙ p) [ p ∈ P¦. Thus M ∪ ¦G¦ ⊆ M[G], and from this
it follows that Ord ∩ M ⊆ Ord ∩ M[G]. On the other hand, for each a ∈ M we
clearly have rank(a) ≥ rank(a
G
), hence Ord ∩ M ⊇ Ord ∩ M[G].
A major result is:
Theorem 5.5.8. M[G] is a countable transitive model of ZFC.
Remark 5.5.9. The proof of Theorem 5.5.8 is long and employs a new method
known as forcing. However, some parts of the proof are easy and do not require
forcing.
For example, given a, b ∈ M, put c = P ¦a, b¦, then c
G
= ¦a
G
, b
G
¦. This
shows that M[G] satisfies the Pairing Axiom. Also, M[G] satisfies the Axiom of
Infinity because ω = ( ˙ ω)
G
∈ M[G]. Furthermore, M[G] satisfies Extensionality
and Foundation automatically, because M[G] is a transitive set.
So far we have not used the assumption that G is an M-generic filter.
In order to prove the rest of Theorem 5.5.8, we now introduce the method
of forcing.
Definition 5.5.10. The forcing language consists of binary relation symbols ∈
and = plus constant symbols a for each a ∈ M. Sentences of the forcing language
are of the form F(a
1
, . . . , a
n
), where F(x
1
, . . . , x
n
) is a formula of L

with free
variables x
1
, . . . , x
n
, and a
1
, . . . , a
n
∈ M. We have M[G] [= F(a
1
, . . . , a
n
) if and
only if F(a
1
, . . . , a
n
) is true in M[G], where quantifiers are interpreted as ranging
over M[G], and a
1
, . . . , a
n
are interpreted as (a
1
)
G
, . . . , (a
n
)
G
respectively.
Definition 5.5.11 (forcing). Let p ∈ P and let F be a sentence of the forcing
language. We say that p |− F (read p forces F) if and only if M[G] [= F for all
M-generic filters G such that p ∈ G.
Lemma 5.5.12 (the extension lemma). If p |− F and q ≤ p, then q |− F.
Proof. This is obvious, because q ∈ G, q ≤ p imply p ∈ G.
Lemma 5.5.13 (definability of forcing). For each formula F(x
1
, . . . , x
n
) there
is a formula F

(p, x
1
, . . . , x
n
) such that, for all p ∈ P and a
1
, . . . , a
n
∈ M,
p |− F(a
1
, . . . , a
n
) if and only if M [= F

(p, a
1
, . . . , a
n
).
108
Lemma 5.5.14 (forcing equals truth). M[G] [= F if and only if ∃p ∈ G(p |−F).
Proof. We shall prove Lemmas 5.5.13 and 5.5.14 by simultaneous induction on
the number of connectives and quantifiers in F. We assume that F contains
only ∧, , and ∀ (not ∨, ⇒, ⇔, ∃).
For the base step, we need to find formulas ∈

(p, x, y) and =

(p, x, y) of
L

defining the relations p |− a ∈ b and p |− a = b, respectively, over M. The
formulas ∈

and =

are defined by a rather complicated simultaneous transfinite
recursion within M. The properties
p |− a ∈ b if and only if M [= ∈

(p, a, b)
and
p |− a = b if and only if M [= =

(p, a, b)
are proved by transfinite induction on rank(a) and rank(b). We omit the details.
For the inductive step, note that
p |− F
1
∧ F
2
if and only if p |− F
1
and p |− F
2
,
and
p |− ∀xF(x) if and only if p |− F(a) for all a ∈ M.
Thus we may define (F
1
∧ F
2
)

= F

1
∧ F

2
and (∀xF(x))

= ∀xF

(x). This
takes care of ∧ and ∀. For , we claim that
p |− F if and only if ∃q ≤ p (q |− F).
To see this, assume the right hand side. Let G be generic with p ∈ G. To show
M[G] [= F. Otherwise, M[G] [= F so let q ∈ G be such that q |− F. Let
r ∈ G be such that r ≤ p and r ≤ q. Then r ≤ p and r |− F, contradicting
our assumption. For the converse, assume the left hand side. Suppose q ≤ p,
q |− F. Let G be generic such that q ∈ G. Then M[G] [= F. Also p ∈ G since
q ≤ p. Therefore p does not force F, contradicting our assumption.
Thus, for definability of forcing, we may take
(F)

(p, a
1
, . . . , a
n
) ≡ (∃q ≤ p) F

(q, a
1
, . . . , a
n
).
For focing equals truth, suppose M[G] [= F. To show (∃p ∈ G) p |− F. Put
D = ¦p [ p |− F or p |− F¦. Clearly D is dense open. By definability of
forcing, D ∈ M. Let p ∈ D ∩ G. If p |− F, then M[G] [= F, a contradiction.
Hence p |− F.
We now proceed to the proof of Theorem 5.5.8.
Lemma 5.5.15. M[G] [= the Comprehension Scheme.
Proof. Given a, a
1
, . . . , a
n
∈ M, to find c ∈ M such that
M[G] [= ∀u (u ∈ c ⇔ (u ∈ a ∧ F(u, a
1
, . . . , a
n
))).
Put
109
c = ¦(p, b) [ b ∈
¸¸
a and p |− b ∈ a ∧ F(b, a
1
, . . . , a
n
)¦.
Then c ∈ M, by Definability of Forcing in M. Then, by the Forcing Equals
Truth Lemma, we have
c
G
= ¦b
G
[ b ∈
¸¸
a and M[G] [= F(b, a
1
, . . . , a
n
)¦.
Lemma 5.5.16. M[G] [= the Power Set Axiom.
Proof. Given a ∈ M, put c = P {(P
¸¸
a) ∩ M. We claim that
c
G
⊇ {(a
G
) ∩ M[G].
To see this, given e
G
∈ M[G], let d = ¦(p, b) ∈ P
¸¸
a [ p |− b ∈ e ∩ a¦. By
definability of forcing, d ∈ M, hence d
G
∈ c
G
. Moreover d
G
= e
G
∩ a
G
. This
proves our claim. The Power Set Axiom follows by Comprehension in M[G],
since {(a
G
) ∩ M[G] = ¦d
G
∈ c
G
[ M[G] [= d ⊆ a¦.
Lemma 5.5.17. M[G] [= the Union Axiom.
Proof. This is similar to the Power Set Axiom. Given a ∈ M put
c = P
¸¸¸¸
a.
Then c
G

¸
a
G
, and the Union Axiom follows by Comprehension in M[G].
Lemma 5.5.18. M[G] [= the Replacement Scheme.
Proof. It suffices to prove that M[G] [= the Bounding Scheme:
∀w
1
w
n
[ ∀u ∃ ! v F(u, v, w
1
, . . . , w
n
) ⇒ ∀x∃y ∀u ∈ x∃v ∈
y F(u, v, w
1
, . . . , w
n
) ].
This is because Bounding plus Comprehension implies Replacement.
Given a, a
1
, . . . , a
n
∈ M, let c ∈ M be such that, for all (p, b) ∈ P
¸¸
a,
if there exists d ∈ M such that p |− F(b, d, a
1
, . . . , a
n
), then c contains such a
d. We then have
M[G] [= ∀u ∃ ! v F(u, v, w
1
, . . . , w
n
) ⇒ ∀u ∈ a ∃v ∈ c

F(u, v, a
1
, . . . , a
n
),
where c

∈ M, namely c

= P c. Thus M[G] [= Bounding.
Lemma 5.5.19. M[G] [= the Axiom of Choice.
Proof. Given a
G
∈ M[G], let f ∈ M map an ordinal α onto
¸¸
a. Since
M ⊆ M[G], we have f = (
˙
f)
G
∈ M[G]. Composing f with the function
b → b
G
, we obtain in M[G] a mapping of α = ( ˙ α)
G
onto ¦b
G
[ b ∈
¸¸
a¦ ⊇ a
G
.
Thus a
G
is well orderable in M[G].
The proof of Theorem 5.5.8 is now complete.
As a first application, we prove the independence of V = L.
Theorem 5.5.20. There exists a countable transitive model of ZFC plus V = L.
110
Proof. Let M be a countable transitive model of ZFC. Let P be the set of finite
partial functions from ω into 2 = ¦0, 1¦. Partially order P by putting p ≤ q if
and only if p extends q. Then P ∈ M. Let G be an M-generic filter on P. By
Theorem 5.5.8 we have M[G] [= ZFC.
Note that p, q ∈ P are compatible if and only if p ∪ q ∈ P. Thus g =
¸
G
is a partial function from ω into 2. We claim that dom(g) = ω. To see this,
given n ∈ ω, put D
n
= ¦p ∈ P [ n ∈ dom(g)¦. Clearly D
n
is dense open, and
D
n
∈ M. Letting p ∈ G∩ D
n
, we see that n ∈ dom(p), hence n ∈ dom(g).
Thus g : ω → 2 and g ∈ M[G]. We claim that g / ∈ M. If g ∈ M, then clearly
G = ¦p ∈ P [ p ⊆ g¦ ∈ M, so let D = P ` G = ¦p ∈ P [ p / ∈ G¦. Then D ∈ M,
and clearly D is dense open. But G∩ D = ∅, a contradiction.
We claim that M[G] [= V = L. In fact,
M[G] [= ˙ g / ∈ L ∧ ˙ g : ω → 2
where ( ˙ g)
G
= g.
5.6 Independence of CH
As in the previous section, let M be a countable transitive model of ZFC, let P
be a partially ordered set belonging to M, and let G be an M-generic filter on
P. We begin with a discussion of cardinal collapsing and cardinal preservation
in M[G].
Remark 5.6.1. Clearly every cardinal of M[G] is a cardinal of M. However,
the converse does not always hold. Cardinals of M can be collapsed in M[G].
Example 5.6.2. Let κ be an uncountable cardinal of M. Let P be the set of
finite partial functions from ω into κ, ordered by p ≤ q if and only if p extends
q. Let G be an M-generic filter on P. Put g =
¸
G. As in the proof of Theorem
5.5.20, we see that g : ω → κ.
We claim that rng(g) = κ. To see this, give α < κ, put D
α
= ¦p ∈ P [ α ∈
rng(p)¦. Clearly D
α
∈ M. Because ω is infinite, D
α
is dense open. Letting
p ∈ G∩ D
α
, we see that α ∈ rng(p), hence α ∈ rng(g).
Thus g ∈ M[G] maps ω onto κ. It follows that M[G] [= “ ˙ κ is a countable
ordinal”. In particular κ is not a cardinal of M[G].
On the other hand, cardinals of M are often preserved, i.e., remain cardinals
in M[G].
Lemma 5.6.3. Suppose M [= “κ is a cardinal > [P[”. Then M[G] [= “κ is a
cardinal”. In other words, all cardinals > [P[ in M are preserved in M[G].
Proof. Suppose not, say f
G
: λ → κ, λ < κ, rng(f
G
) = κ, f
G
∈ M[G]. Then in
M we have
∀α < κ ∃β < λ ∃p ∈ P (p |− f [
˙
λ → ˙ κ and p |− f(
˙
β) = ˙ α).
111
By the Pigeonhole Principle, we can find p ∈ P, β < λ, α
1
< α
2
< κ such that
p |− f :
˙
λ → ˙ κ and p |− f(
˙
β) = ˙ α
1
and f(
˙
β) = ˙ α
2
. This is a contradiction.
Definition 5.6.4. An antichain in P is a set A ⊆ P such that the elements
of A are pairwise incompatible. P is said to have the countable chain condition
(c.c.c.) if every antichain of P is countable.
Lemma 5.6.5. Suppose M [= P is c.c.c. Then all cardinals of M are preserved
in M[G].
Proof. Suppose not, say κ > λ, M [= “κ is a cardinal”, M[G] [= “f maps
˙
λ onto
˙ κ”. Fix p ∈ P such that p |− “f maps
˙
λ onto ˙ κ”. Reasoning within M, for α < κ
and β < λ say that α is a possible value of f(β) if ∃q ≤ p (q |− f(
˙
β) = ˙ α). Let
X
β
= ¦α [ α is a possible value of f(β)¦. Note that κ =
¸
β<λ
X
β
. Therefore,
some X
β
is uncountable. Fix such a β. For each α ∈ X
β
let q
α
≤ p be such
that q
α
|− f(
˙
β) = ˙ α. Note that α
1
, α
2
∈ X
β
, α
1
= α
2
implies q
α1
⊥ q
α2
. Thus
A
β
= ¦q
α
[ α ∈ X
β
¦ is an uncountable antichain in P. This contradicts the
assumption that P is c.c.c.
We now proceed to the independence of the Continuum Hypothesis.
Definition 5.6.6. A ∆-system is an indexed family of sets 'X
i
`
i∈I
such that,
for some fixed set D, X
i
∩ X
j
= D for all i, j ∈ I, i = j.
Lemma 5.6.7 (the ∆-system lemma). Any uncountable family of finite sets
contains an uncountable subfamily which is a ∆-system.
Proof. Let 'X
i
`
i∈I
be an uncountable family of finite sets. We may safely assume
that [I[ = ℵ
1
and that
¸
i∈I
X
i
⊆ ω
1
. Passing to an uncountable subfamily, we
may assume that ∃n∀i ∈ I [X
i
[ = n. For each i ∈ I, let X
i
(1) < < X
i
(n) be
the elements of X
i
.
Case 1: For each k = 1, . . . , n, ¦X
i
(k) : i ∈ I¦ is countable. In this case,
¸
i∈I
X
i
is countable. Hence, by passing to an uncountable subfamily, we may
assume X
i
= X
j
for all i, j ∈ I. In particular, we have an uncountable ∆-system.
Case 2: Otherwise. Let k be as small as possible such that ¦X
i
(k) [ i ∈ I¦
is uncountable. Then, for each l < k, ¦X
i
(l) [ i ∈ I¦ is countable. By passing
to an uncountable subfamily, we may assume X
i
(l) = X
j
(l) for all l < k and all
i, j ∈ I. Thus we have a fixed finite set D = ¦X
i
(l) [ 1 ≤ l < k¦ for all i ∈ I.
Since ¦X
i
(k) [ i ∈ I¦ is uncountable, we may use transfinite recursion to define
a function f : ω
1
→ I such that, for each α < ω
1
, X
f(α)
(k) > sup
β<α
X
f(β)
(n).
Then 'X
f(α)
`
α<ω1
is an uncountable ∆-system contained in 'X
i
`
i∈I
.
Lemma 5.6.8. Let X be any set. Let P be the set of finite partial functions
from X into ¦0, 1¦, ordered by putting p ≤ q if and only if p ⊇ q. Then P is
c.c.c.
Proof. Suppose not. Let 'p
i
`
i∈I
be an uncountable antichain in P. By the
∆-system lemma, we may pass to a subfamily such that 'dom(p
i
)`
i∈I
is a ∆-
system. Say dom(p
i
)∩dom(p
j
) = D for all i, j ∈ I, i = j. There are only finitely
112
many functions from D into ¦0, 1¦, so by passing to an uncountable subfamily
we may assume that p
i
↾D = p
j
↾D for all i, j ∈ I. Then for all i, j ∈ I we have
that p
i
∪ p
j
is a function, hence p
i
and p
j
are compatible, a contradiction.
Theorem 5.6.9. Let M be a countable transitive model of ZFC. Let κ be an
uncountable cardinal of M. Then there exists a countable transitive model M

of ZFC extending M such that (1) M

satisfies 2
ℵ0
≥ κ, and (2) M

has the
same ordinals and cardinals as M.
Proof. Let P be the set of finite partial functions from κ ω into ¦0, 1¦. Let
G be an M-generic filter on P. By Lemma 5.5.7 and Theorem 5.5.8, M[G] is a
countable transitive model of ZFC which includes M and has the same ordinals
as M. By Lemma 5.6.8 P is c.c.c. By Lemma 5.6.5 M[G] has the same cardinals
as M.
Put g =
¸
G. As in the proof of Theorem 5.5.20 we see that g ∈ M[G] and
g : κ ω → ¦0, 1¦. For α < κ define g
α
: ω → ¦0, 1¦ by g
α
(n) = g((α, n)). We
claim that g
α
= g
β
for all α < β < κ. To see this, let D
αβ
be the set of p ∈ P
such that p((α, n)) = p((β, n)) for some n ∈ ω such that (α, n), (β, n) ∈ dom(p).
Clearly D
αβ
∈ M and is dense open. Hence G∩ D
αβ
= ∅. Hence g
α
= g
β
.
It is now clear that M[G] [= 2
ℵ0
≥ ˙ κ. Thus we may take M

= M[G].
113
Chapter 6
Topics in Set Theory
6.1 Stationary Sets
Definition 6.1.1. Let S be a set of ordinals. We say that S is unbounded in a
limit ordinal δ if sup(S ∩ δ) = δ. We say that S is closed in κ if S ⊆ κ and, for
all limit ordinals δ < κ, if S is unbounded in δ then δ ∈ S. A closed unbounded
set in κ (sometimes called a club of κ) is any subset of κ which is closed in κ
and unbounded in κ.
Lemma 6.1.2. Let κ be a regular uncountable cardinal.
1. If C
i
, i ∈ I, is a collection of closed unbounded sets in κ, and if [I[ < κ,
then
¸
i∈I
C
i
is again a closed unbounded set in κ.
2. If C
α
, α < κ is a collection of closed unbounded sets in κ indexed by the
ordinals less than κ, then the diagonal intersection

α<κ
C
α
= ¦β < κ [ β ∈ C
α
for all α < β¦
is again a closed unbounded set in κ.
Proof. Straightforward.
Definition 6.1.3. Let κ be a regular uncountable cardinal. A set S ⊆ κ is said
to be stationary in κ if S ∩ C = ∅ for every closed unbounded set C in κ.
Lemma 6.1.4. Let κ be a regular uncountable cardinal, and let S ⊆ κ be
stationary in κ. Suppose S =
¸
i∈I
S
i
where [I[ < κ. Then S
i
is stationary for
some i ∈ I.
Proof. Suppose the conclusion fails. Then for each i ∈ I let C
i
be a closed
unbounded set such that S
i
∩ C
i
= ∅. By Lemma 6.1.2.1, C =
¸
i∈I
C
i
is a
closed unbounded set. Since S is stationary, S ∩C is nonempty, say α ∈ S ∩C.
Then for each i ∈ I we have α / ∈ S
i
, a contradiction.
114
Theorem 6.1.5 (Fodor’s Theorem). Let κ be a regular uncountable cardinal,
and let S ⊆ κ be stationary in κ. Suppose f : S → κ is such that f(α) < α for
all α ∈ S. Then f is constant on a stationary set. In other words, there exist a
stationary S
0
⊆ S and a β
0
< κ such that f(α) = β
0
for all α ∈ S
0
.
Proof. Similar to the proof of the previous lemma, using 6.1.2.2 instead of
6.1.2.1. The details are left as an exercise for the reader.
Theorem 6.1.6. For any regular uncountable cardinal κ, there exists a sta-
tionary set S ⊆ κ such that κ ` S is also stationary.
Proof. . . .
We state without proof the following theorem of Solovay.
Theorem 6.1.7. Let κ be a regular uncountable cardinal. Any stationary set
S ⊆ κ can be decomposed into κ pairwise disjoint stationary sets.
6.2 Large Cardinals
Definition 6.2.1 (hyperinaccessible cardinals). For each n < ω we define a
class of cardinals called the n-hyperinaccessible cardinals. We define κ to be 0-
hyperinaccessible if it is inaccessible. We define κ to be n+1-hyperinaccessible
if it is inaccessible and
¦λ < κ [ λ is n-hyperinaccessible¦
is unbounded in κ.
Definition 6.2.2 (Mahlo cardinals). For each n < ω we define a class of cardi-
nals called the n-Mahlo cardinals. We define κ to be 0-Mahlo if it is inaccessible.
We define κ to be n+1-Mahlo if it is inaccessible and ¦λ < κ [ λ is n-Mahlo¦ is
stationary in κ.
Exercise 6.2.3. Show that n+1-hyperinaccessible implies n-hyperinaccessible.
Show that n+1-Mahlo implies n-Mahlo. Show that 1-Mahlo implies n-hyperinaccessible
for all n < ω.
Lemma 6.2.4. Let δ be a limit ordinal. Suppose κ < δ and n < ω. Then κ is
n-hyperinaccessible if and only if R
δ
satisfies “κ is n-hyperinaccessible.” Also,
κ is n-Mahlo if and only if R
δ
satisfies “κ is n-Mahlo.”
Proof. Straightforward.
Theorem 6.2.5.
1. The existence of an n+1-hyperinaccessible cardinal is not provable in ZFC
+ “for all α there exists κ > α such that κ is n-hyperinaccessible” (as-
suming this theory is consistent).
115
2. The existence of an n+1-Mahlo cardinal is not provable in ZFC + “for
all α there exists κ > α such that κ is n-Mahlo” (assuming this theory is
consistent).
Proof. Straightforward using the previous lemma.
Lemma 6.2.6.
1. If κ is a cardinal, then L satisfies “κ is a cardinal.”
2. If κ is a regular cardinal, then L satisfies “κ is a regular cardinal.”
3. If κ is n-hyperinaccessible, then L satisfies “κ is n-hyperinaccessible.”
4. If κ is n-Mahlo, then L satisfies “κ is n-Mahlo.”
Proof. Straightforward.
Theorem 6.2.7.
1. If ZFC + “there exists an n-hyperinaccessible cardinal” is consistent, then
so is ZFC + V = L + “there exists an n-hyperinaccessible cardinal.”
2. If ZFC + “there exists an n-Mahlo cardinal” is consistent, then so is ZFC
+ V = L + “there exists an n-Mahlo cardinal.”
Proof. Straightforward using the previous lemma.
6.3 Ultrafilters and Ultraproducts
Definition 6.3.1. Let I be a nonempty set. A filter on I is a set T ⊆ {(I)
such that
1. ∅ / ∈ T and I ∈ T;
2. if X
1
, . . . , X
n
∈ T then X
1
∩ . . . ∩ X
n
∈ T;
3. if X ∈ T and X ⊆ Y ∈ {(I) then Y ∈ T.
Examples 6.3.2.
1. T = ¦I¦.
2. T = ¦X ⊆ I [ X ⊇ X
0
¦, where ∅ = X
0
⊆ I. Such an T is called a
principal filter.
3. T = ¦X ⊆ I [ X is cofinite, i.e., I ` X is finite¦ (assuming I is infinite).
4. T = ¦X ⊆ I [ [I ` X[ < κ¦, where κ is any infinite cardinal ≤ [I[.
5. I = R
n
, T = ¦X ⊆ R
n
[ R
n
` X has Lebesgue measure 0¦. Here we could
replace R
n
by any measure space.
116
6. I = R
n
, T = ¦X ⊆ R
n
[ R
n
` X is meager¦. Here we could replace R
n
by
any complete metric space.
7. Let I = κ where κ is a regular uncountable cardinal. Then
T = ¦X ⊆ κ [ X ⊇ C for some closed unbounded set C ⊆ κ¦
is a filter, known as the closed unbounded filter on κ.
8. Let A be an uncountable set. Put
I = {
c
(A) = ¦Y ⊆ A [ Y is countable¦ .
Recall that A

is the set of finite sequences of elements of A. Given
f : A

→ A, put
C
f
= ¦Y ∈ {
c
(A) [ Y is closed under f, i.e., f[Y

] ⊆ Y ¦ .
Then
T
c
(A) = ¦X ⊆ {
c
(A) [ X ⊇ C
f
for some f¦
is a filter known as the closed unbounded filter on {
c
(A).
Definition 6.3.3. Let κ be an infinite cardinal. A filter T is said to be κ-
additive if
¸
i∈I
X
i
∈ T whenever X
i
∈ T for all i ∈ I, [I[ < κ.
Examples 6.3.4.
1. Every filter is finitely additive, i.e., ℵ
0
-additive.
2. The Lebesgue and Baire filters on R
n
are countably additive, i.e., ℵ
1
-
additive.
3. For any infinite cardinal κ ≤ [I[, the filter ¦X ⊆ I [ [I − X[ < κ¦ is
κ-additive.
4. For any regular uncountable cardinal κ, the closed unbounded filter on κ
is κ-additive.
5. For any uncountable set A, the closed unbounded filter on {
c
(A) is count-
ably additive.
Definition 6.3.5. An ultrafilter on I is a filter | on I such that for all X ⊆ I
either X ∈ | or I ` X ∈ |.
Remark 6.3.6. The filters in 6.3.2.3–8 are not ultrafilters. Indeed, it is diffi-
cult to find explicit examples of nonprincipal ultrafilters. However, as we shall
now show, nonprincipal ultrafilters can be constructed by means of transfinite
recursion plus the Axiom of Choice.
Theorem 6.3.7. Any filter T on I can be extended to an ultrafilter | on I.
117
Proof. Say that ( ⊆ {(I) has the finite intersection property (f.i.p.) if Y
1
∩. . . ∩
Y
m
= ∅ for all Y
1
, . . . , Y
m
∈ (.
By the well-ordering theorem, let κ = [{(I)[, say
{(I) = ¦X
α
[ α < κ¦ .
We shall use transfinite recursion to define a sequence of sets T
α
⊆ {(I), α ≤ κ,
each of which has the f.i.p.
Stage 0. Put T
0
= T. Note that T has the f.i.p. since it is a filter.
Stage α + 1. Assume inductively that T
α
has the f.i.p. We claim that
at least one of T
α
∪ ¦X
α
¦, T
α
∪ ¦I ` X
α
¦ has the f.i.p. Otherwise we would
have X
α
∩ Y
1
∩ . . . ∩ Y
m
= ∅, Y
1
, . . . , Y
m
∈ T
α
, (I ` X
α
) ∩ Z
1
∩ . . . ∩ Z
m
= ∅,
Z
1
, . . . , Z
n
∈ T
α
. Then Y
1
∩ . . . ∩ Y
m
∩ Z
1
∩ . . . ∩ Z
n
= ∅ so T
α
does not have
the f.i.p., a contradiction. We therefore set
T
α+1
=

T
α
∪ ¦X
α
¦ if this has the f.i.p.,
T
α
∪ ¦I ` X
α
¦ otherwise.
Then clearly T
α+1
has the f.i.p.
Stage δ, where δ is a limit ordinal. Put T
δ
=
¸
α<δ
T
α
. Clearly this has the
f.i.p.
Finally put | = T
κ
=
¸
α<κ
T
α
. Clearly | has the f.i.p. and for every
X ∈ {(I) either X ∈ | or I ` X ∈ |. It follows easily that | is an ultrafilter.
This completes the proof.
Lemma 6.3.8. Any principal ultrafilter | on I is of the form
| = ¦X ⊆ I [ i
0
∈ X¦
for some fixed i
0
∈ I.
Proof. Let | be a principal ultrafilter on I. By definition we have
| = ¦X ⊆ I [ X ⊇ X
0
¦
where ∅ = X
0
⊆ I. If [X
0
[ ≥ 2, let Y ⊆ I be such that Y ∩ X
0
= ∅ and
(I ` Y ) ∩ X
0
= ∅. Then Y, I ` Y / ∈ |, a contradiction. Thus [X
0
[ = 1, i.e.,
X
0
= ¦i
0
¦ for some i
0
∈ I. This proves the lemma.
Theorem 6.3.9. For every infinite set I there exists a nonprincipal ultrafilter
| on I.
Proof. Consider the filter T = ¦X ⊆ I [ I`X is finite¦. By Theorem 6.3.7, let |
be an ultrafilter on I such that T ⊆ |. For all i
0
∈ I we have I ` ¦i
0
¦ ∈ T ⊆ |,
hence ¦i
0
¦ / ∈ |. Thus | is nonprincipal.
Definition 6.3.10. A structure is a relational structure, i.e., an ordered pair
(A, E) where A is a nonempty set and E ⊆ A A.
118
Definition 6.3.11 (ultraproduct). Suppose we are given an indexed family of
structures '(A
i
, E
i
)`
i∈I
and an ultrafilter | on the index set I. We are going to
define a structure
(A, E) =
¸
U
'(A
i
, E
i
)`
i∈I
known as an ultraproduct . Recall that
¸
i∈I
A
i
=

f

f : I →
¸
i∈I
A
i
, f(i) ∈ A
i
for all i ∈ I
¸
.
For f, g ∈
¸
i∈I
A
i
define
f ≈ g ⇔
def
f ≈
U
g

def
¦i ∈ I [ f(i) = g(i)¦ ∈ | .
This is an equivalence relation. We define
[f] =
def
[f]
U
=
def

g ∈
¸
i∈I
A
i

f ≈
U
g
¸
and
A =
¸
U
'A
i
`
i∈I
=
¸
i∈I
A
i

| =

[f]
U

f ∈
¸
i∈I
A
i
¸
.
Finally, for f, g ∈
¸
i∈I
A
i
, we define
([f], [g]) ∈ E ⇔
def
¦i ∈ I [ (f(i), g(i)) ∈ E
i
¦ ∈ | .
Note that this last definition is independent of representatives, i.e., f ≈ f

,
g ≈ g

, ([f], [g]) ∈ E imply ([f

], [g

]) ∈ E. Thus E ⊆ AA is well-defined, and
so (A, E) is a structure.
Theorem 6.3.12 ( Lo´s’s Theorem). Let (A, E) =
¸
U
'(A
i
, E
i
)`
i∈I
be an ultra-
product. Let ψ(x
1
, . . . , x
k
) be a formula with free variables among x
1
, . . . , x
k
.
Then for all [f
1
], . . . , [f
k
] ∈ A we have
[=
(A,E)
ψ([f
1
], . . . , [f
k
]) ⇔ ¦i ∈ I [ [=
(Ai,Ei)
ψ(f
1
(i), . . . , f
k
(i))¦ ∈ | .
6.4 Measurable Cardinals
6.5 Ramsey’s Theorem
119
Index
R
α
, 87

0
n
predicate, 44
Π
0
n
predicate, 41
Σ
0
n
predicate, 41

α
, 84
↓, 23
κ-additive, 117
ω, 75
ω
α
, 84
≃, 23
↑, 23
Lo´s’s theorem, 119
Ackermann function, 13, 25, 30, 33, 39
additive, 117
additively indecomposable ordinal, 80
arithmetic, 48
cardinal, 73
language of, 48
ordinal, 75, 79, 80
arithmetical definability, 50–56, 59
arithmetical hierarchy, 41–47, 56
arithmetical truth, 58, 59
axioms
of set theory, 92–96
axiom of choice, 94, 97, 101, 106
axiom of infinity, 94
axiom scheme, 95
Boolean connective, 9, 49
bounded
least number operator, 10
quantifier, 10
Cantor’s theorem, 73
cardinal
hyperinaccessible, 115
inaccessible, 86
limit, 85
Mahlo, 115
regular, 85
singular, 86
strong limit, 85
successor, 85
uncountable, 85
weakly inaccessible, 86
cardinal arithmetic, 73
cardinal number, 71–74, 81–86
cases
definition by, 28, 63
Cauchy sequence, 90
CH, 85, 97
characteristic function, 9
Chinese remainder theorem, 53
Church’s thesis, 33
class, 78
closed, 114
closed unbounded filter, 117
club, 114
cofinality, 86
complete, 46
comprehension, 95, 101
computable function, 17–23
connective
Boolean, 9, 49
propositional, 49
consequence
logical, 97
consistency, 97
relative, 107
constructible set, 103–107
Continuum Hypothesis, 85
continuum hypothesis, 85, 97, 106
Continuum Problem, 85
120
continuum problem, 85, 97
convergent, 23
countably additive, 117
course-of-values recursion, 12
decidability, 67
Def, 98
definability
arithmetical, 50–56, 59
over the real number system, 60
over a relational structure, 98
over the real number system, 66
defined, 23
definition by cases, 28, 63
dense open set, 107
diagonal intersection, 114
divergent, 23
effective function, 61
enumeration theorem, 28
equivalence relation, 90
extensionality, 92, 94, 97, 100
f.i.p., 118
falsity, 49
filter, 107, 116
closed unbounded, 117
principal, 116
finitely additive, 117
finite intersection property, 118
Fodor’s theorem, 115
forcing, 108
formula, 48, 60, 92
function, 89
computable, 17–23
effective, 61
limit-recursive, 44
number-theoretic, 6
partial, 23
partial recursive, 23
primitive recursive, 6–13, 30
recursive, 23
total, 23
G¨odel number
of a formula, 57
of a program, 26, 27
GCH, 85, 107
generalized continuum hypothesis, 85,
106
halting problem, 35, 37
Hilbert’s 10th problem, 35
Hilbert’s 17th problem, 61
hyperinaccessible cardinal, 115
inaccessible cardinal, 86, 101–103
index
of a partial recursive function, 26,
27
induction
transfinite, 78
initial ordinal, 82
isomorphism, 74
K¨onig’s Theorem, 74
L, 103, 106
L¨ owenheim/Skolem theorem, 99
language
of arithmetic, 48
of ordered rings, 60
of set theory, 92–96
least number operator, 23
bounded, 10
limit-recursive function, 44
limit cardinal, 85
limit ordinal, 80
linear ordering, 75
logical consequence, 97
Mahlo cardinal, 115
minimization, 24, 30
model, 97
number
cardinal, 71–74, 81–86
ordinal, 75, 89
number systems, 90
ordered field, 60
ordered pair, 70, 89
ordering
121
linear, 75
well, 75
ordinal, 75, 89
additively indecomposable, 80
initial, 82
limit, 80
successor, 80
von Neumann, 89
ordinal arithmetic, 75, 79, 80
ordinal number, 75, 89
parameter, 98
parametrization theorem, 36
partial function, 23
partial recursive function, 23, 32
power set, 93, 101
predicate, 9
primitive recursive
function, 6–13, 30
predicate, 9
principal filter, 116
principal function, 43
program, 17
propositional connective, 49
pure set, 87, 92, 93, 96, 97, 102, 104
quantifier, 49
bounded, 10
quantifier elimination, 60, 66
real number system, 60, 90
real world, 92, 93, 97
recursion
course-of-values, 12
primitive, 6
transfinite, 78
recursion theorem, 39
recursively enumerable set, 44
recursive function, 23
reducible, 36
register machine program, 17
regular cardinal, 85
relational structure, 74, 96
relative consistency, 107
replacement, 95, 101
Rice’s theorem, 38
scheme, 95
sentence, 49
set theory
axioms of, 92–96
language of, 92–96
models of, 93, 96, 97, 99–104
singular cardinal, 86
Skolem paradox, 99
Solovay, 115
state, 29
stationary, 114
strong limit cardinal, 85
structure
arithmetic, 48
real number system, 60
relational, 74, 96
successor cardinal, 85
successor ordinal, 80
term, 48
total function, 23
transfinite induction, 78
transfinite recursion, 78
transitive model, 99–103
transitive set, 87, 99
truth
arithmetical, 49, 58, 59
ultrafilter, 117
ultraproduct, 119
unbounded, 114
uncountable cardinal, 85
undecidability, 58
undefined, 23
universal
Σ
0
n
predicate, 45
register machine program, 28
unsolvable problem, 34–39, 58
V , 88
von Neumann ordinal, 89
weakly inaccessible cardinal, 86
well-founded, 75
well-ordering, 75
well-ordering theorem, 81
122
word problem, 35
Zermelo/Fraenkel set theory, 96
ZF, 96
ZFC, 96
123

Contents
1 Computable Functions 1.1 Primitive Recursive Functions . . . . . . . . 1.2 The Ackermann Function . . . . . . . . . . 1.3 Computable Functions . . . . . . . . . . . . 1.4 Partial Recursive Functions . . . . . . . . . 1.5 The Enumeration Theorem . . . . . . . . . 1.6 Consequences of the Enumeration Theorem 1.7 Unsolvable Problems . . . . . . . . . . . . . 1.8 The Recursion Theorem . . . . . . . . . . . 1.9 The Arithmetical Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 6 13 17 23 25 30 34 39 41

2 Undecidability of Arithmetic 48 2.1 Terms, Formulas, and Sentences . . . . . . . . . . . . . . . . . . . 48 2.2 Arithmetical Definability . . . . . . . . . . . . . . . . . . . . . . . 50 2.3 G¨del Numbers of Formulas . . . . . . . . . . . . . . . . . . . . . 57 o 3 The Real Number System 60 3.1 Quantifier Elimination . . . . . . . . . . . . . . . . . . . . . . . . 60 3.2 Decidability of the Real Number System . . . . . . . . . . . . . . 66 4 Informal Set Theory 4.1 Operations on Sets . . . . . . . . . . . 4.2 Cardinal Numbers . . . . . . . . . . . 4.3 Well-Orderings and Ordinal Numbers 4.4 Transfinite Recursion . . . . . . . . . . 4.5 Cardinal Numbers, Continued . . . . . 4.6 Cardinal Arithmetic . . . . . . . . . . 4.7 Some Classes of Cardinals . . . . . . . 4.8 Pure Well-Founded Sets . . . . . . . . 4.9 Set-Theoretic Foundations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 69 71 74 78 81 83 85 87 88

5 Axiomatic Set Theory 92 5.1 The Axioms of Set Theory . . . . . . . . . . . . . . . . . . . . . . 92 5.2 Models of Set Theory . . . . . . . . . . . . . . . . . . . . . . . . 96

2

5.3 5.4 5.5 5.6

Transitive Models and Inaccessible Cardinals Constructible Sets . . . . . . . . . . . . . . . Forcing . . . . . . . . . . . . . . . . . . . . . Independence of CH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. 99 . 103 . 107 . 111 . . . . . 114 114 115 116 119 119

6 Topics in Set Theory 6.1 Stationary Sets . . . . . . . . 6.2 Large Cardinals . . . . . . . . 6.3 Ultrafilters and Ultraproducts 6.4 Measurable Cardinals . . . . 6.5 Ramsey’s Theorem . . . . . .

3

Stopping . . . . . . . . . . . . . . . . . . . . . . . . .9 1. A Multiplication Program . . . . . .12 Register Machine Instructions . . . . . . . . . . . . . .4 1. . . . . . . . . . . . . . Parametrization . . . . . .10 1. . . .1 1. .List of Figures 1. . . . . . . . . . . . . . . . . . .7 1. . . . . . . . . . . . . . . . . .8 1. . . . . . . . . . . . . . Minimization . . . . . . . . . . . . . . . . . .11 1. . . . . . . . . . . . . . . .6 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Initial Functions .2 1. . . . . Primitive Recursion . . . . 18 18 19 20 21 22 24 27 31 31 32 37 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A Program with Labeled Instructions Incrementing Pi .3 1. . . . . . Decrementing Pi . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 1. . . . . . . . Generalized Composition . . . . . . . . . . . An Addition Program . . . . . . . . . . . . . . . . . .

. . . . . . . 14 5 . . . . . . . . . . .List of Tables 1. .1 The Ackermann branches. . . . .

. the factorial function λx [ x! ] is defined by the primitive recursion equations 0! = 1. . N = {0. A numbertheoretic function is said to be primitive recursive if it can be built up by means of primitive recursions. a recursion is any kind of inductive definition.Chapter 1 Computable Functions We use N to denote the set of natural numbers. .} . the k-fold Cartesian product N× . For example. × N k is denoted Nk . . . The purpose of this chapter is to define and study an important class of number-theoretic functions.. This concept is made precise in the following definition.1 (Primitive Recursive Functions). (x + 1)! = x!(x + 1). the recursive functions (sometimes called the computable functions).1. A k-place function is a function f : Nk → N and is sometimes indicated with the lambda-notation. A number-theoretic function is a k-place function for some k ≥ 1. 6 . 1. For k ≥ 1. and a primitive recursion is an especially straightforward kind of recursion. in which the value of a number-theoretic function at argument x + 1 is defined in terms of the value at argument x. f = λx1 · · · xk [ f (x1 . We begin with a certain subclass known as the primitive recursive functions. 1.1 Primitive Recursive Functions Loosely speaking. . Definition 1. xk ) ] . . . The class PR of primitive recursive functions is the smallest class C of number-theoretic functions having the following closure properties. 2.

1. . then the (k+1)-place function λyx1 · · · xk [ f (y. . . xk ) ] and the (k+2)-place function λyzx1 · · · xk [ h(y. This means that whenever the k-place function λx1 · · · xk [ g(x1 . . . . . . . . The successor function S = λx [ x + 1 ] belongs to C. . xk ) ] defined by f (0. xk ) = h(g1 (x1 . For each k ≥ 1 and 1 ≤ i ≤ k. . x1 . . . xk ) ]. xk ) = f (y + 1. . . xk ) ] and the m-place function λy1 · · · ym [ h(y1 . . . . . 7 . . . . Examples 1. . .2. . . . x1 . gm (x1 . Here f is defined by f (x1 . C is closed under generalized composition. . then the k-place function f = λx1 · · · xk [ h(g1 (x1 . . . 4. . The recursion equations x·0 = 0 (x · y) + x x · (y + 1) = show that the multiplication function λxy [ x · y ] is primitive recursive. . x1 . . The constant zero function Z = λx [ 0 ] belongs to C. x1 . . x1 . gm (x1 . . . C is closed under primitive recursion. . . . . xk ) h(y. xk )) . . . This means that whenever the k-place functions λx1 · · · xk [ g1 (x1 . . x1 . 3. We now list some examples of primitive recursive functions. . . . . f (y.1. xk ). . . . . the projection function Pki = λx1 · · · xk [ xi ] belongs to C. xk ) show that the addition function λxy [ x + y ] is primitive recursive. . . . . . ym ) ] all belong to C. . xk )) ] also belongs to C. . . The recursion equations x+0 = x + (y + 1) = x (x + y) + 1 g(x1 . 2. xk ). . . . . xk ) ] belong to C. . xk ). 2. . . λx1 · · · xk [ gm (x1 . . . 5. z. . .1. . . . . . xk ) = also belongs to C.

We can then obtain the truncated subtraction function · x−y = x−y 0 if x ≥ y . As already mentioned. The following exercise will become easy after we have developed a little more machinery.3. . if x < y · · · using primitive recursion equations x −0 = x. . First. We now proceed to further enlarge our library of primitive recursive functions. Show that the Fibonacci function. x −(y +1) = P (x −y). Exercise 1. 5. 1. 2. defined by fib(0) = 0 . the recursion equations 0! = 1 (x + 1)! = x! · (x + 1) show that the factorial function λx [ x! ] is primitive recursive. fib(1) = 1 . 34. 4.) We shall also have use for · · |x − y| = (x − y) + (y − x) · and α(x) = 1 − x. 0 if x = 0 is primitive recursive. if x = 0 . fib(x + 2) = fib(x) + fib(x + 1) is primitive recursive. 3. 1. the recursion equations P (0) = 0. 13. (Truncated subraction is useful because ordinary subtraction is not a function from N2 into N. Note that α(x) = 0 1 if x > 0 . 21.3.1. 8. . P (x + 1) = x show that the “predecessor” function P (x) = x − 1 if x > 0 . . (The first few values of this function are 0.) 8 . 55. The recursion equations x0 x y+1 = 1 = xy · x show that the exponentiation function λxy [ xy ] is primitive recursive.

y. respectively. y. . . z1 . z1 . . . ∧. zk ) .) Proof. . . z1 . . xk ∈ R. y. . .1. .6 (Iterated Sums and Products). . The treatment of h is similar. . . . . . zk ) show that g ∗ is primitive recursive. . . xk ) is true if x1 . Also χP ∨Q = α(α(χP ) · α(χQ )) since P ∨ Q ≡ ¬ ((¬ P ) ∧ (¬ Q)) (de Morgan’s law). The recursion equations g ∗ (0. z1 . zk ) = x=0 f (x. . z1 . xk ) is false g(y. . A k-place predicate R ⊆ Nk is said to be primitive recursive if its characteristic function χR (x1 . and P ∨ Q. . . . zk ) is a primitive recursive function. . . we say that R(x1 . and (nonexclusive) disjunction. then so are y−1 1 0 if R(x1 . .1. . . .5 (Boolean Connectives). and ∨ denote negation. z1 . By a k-place predicate we mean a subset of Nk . . . z1 . If R ⊆ Nk is a k-place predicate and x1 . zk ) (= 1 if y = 0) . . zk ) = x=0 f (x. 9 . y. . . xk ) = is primitive recursive. . . we shall want to consider primitive recursive predicates. y. Definition 1. zk ) = g ∗ (y. y) = α(|x − y|) and χ< (x. (Here ¬ . zk ) = y−1 f (x. . We have g(y. .1. hence g is primitive recursive. z1 . z1 . . . . y. . zk ) ∗ = 0 = g ∗ (w. . y. . . . . y) = α(α(y − x)). . . . . · since χ= (x.In addition to primitive recursive functions. . xk are elements of N. y. z1 . . z1 . For example. xk ) is true if R(x1 . . . P ∧ Q. . Lemma 1. We have χ¬ P = α(χP ) and χP ∧Q = χP · χQ . .4. . . . zk ) (= 0 if y = 0) and h(y. . then so are ¬ P . . zk ) where w−1 g ∗ (w. z1 . If P and Q are primitive recursive predicates. . zk ) g (w + 1. . y. A number-theoretic predicate is a k-place predicate for some k ≥ 1. z1 . the 2-place predicates x = y and x < y are primitive recursive. . . Lemma 1. otherwise false. . . . . . y. . If f (x. x=0 Proof. . . zk ) + f (w. conjunction. . .

.e. zk )) x=0 so our result follows from the previous lemma. f (y. . . . y. . For example. zk ) =  y otherwise is primitive recursive. y.8 (Bounded Least Number Operator). z1 . . y. We have R(x. . z1 . zk ) · +y· y−1 x=0 α(χR (x. . . . . zk ) is a primitive recursive predicate. respectively. z1 . . zk ) = y−1 x=0 x−1 w=0 α(χR (w. zk ) and χQ (y. These operators are sometimes called bounded quantifiers. z1 . then the function   least x < y such that R(x. . If R(x. . zk ) = α y−1 α(χR (x. the set (i. . . zk )) x · χR (x. . . . y. . y. z1 . . . . 10 . z1 . z1 . z1 . . . . . zk ) is a primitive recursive predicate. . . since Prime (x) ≡ x is a prime number ≡ x>1 ∧ ¬ u<x v<x y−1 y−1 (x = u · v ∧ u > 1 ∧ v > 1) . . zk ) holds . Note that x=0 and x=0 can be paraphrased as “for all x in the range 0 ≤ x < y” and “there exists x in the range 0 ≤ x < y”. y. . z1 . z1 . . . zk ) = x=0 χR (x. zk ) ≡ and Q(y. Proof. the following lemma can be used to show that many familiar functions are primitive recursive. . . . z1 . . The above lemmas make it easy to show that many familiar predicates are primitive recursive. x=0 y−1 χP (y. . Lemma 1. y. .1.1. zk ) ≡ Proof. . z1 .7 (Finite Conjuction and Disjunction). .Lemma 1. We have f (y. . . z1 . . zk ) x=0 y−1 R(x. . . . . . zk )) . . . If R(x. . so f is primitive recursive. . z1 . .. z1 . then so are y−1 P (y. . . y. y. . . . . . y. 1-place predicate) of prime numbers is primitive recursive. . . zk ) . if such x exists . z1 . Similarly. .

since (a)x = ax . pw+1 ) = 0 n the exponent of pn in the prime power factorization of z .1. x and for decoding we can use the primitive recursive function λzx [ (z)x ]. 2(fibpair(x))1 3(fibpair(x))0 +(fibpair(x))1 . . Using the bounded least number operator. x) = Remainder (y. For our pairing function we use λuv [ 2u 3v ]. p2 = 5. defined by fibpair(x) = 2fib(x) 3fib(x+1) . 0 ≤ q. Exercise 1. First. Remainder (y. in view of Quotient (y. More generally. The pairing and unpairing functions make it easy to show that the Fibonacci function is primitive recursive (cf. p3 = 7. . p1 = 3. . am−1 is encoded by a= x<m pax . 0 ≤ r < x. we can encode variable-length finite sequences of natural numbers as single numbers. recall a famous theorem of Euclid which gives the bound pn+1 ≤ pn ! + 1. are primitive recursive.) We want to use the bounded least number operator to show that λn [ pn ] is primitive recursive. where y = q · x + r. Namely. x) = least q ≤ y such that r<x (y q≤y (y least r < x such that = q · x + r) . where (z)n = = least w < z such that Remainder(z. (The first few values of this function are p0 = 2. Note that (2u 3v )0 = u and (2u 3v )1 = v. 11 . a1 . consider the function λn [ pn ] which enumerates the prime numbers in increasing order. . The unpairing functions are then λz [ (z)0 ] and λz [ (z)1 ]. Then λx [ fib(x) ] is primitive recursive since fib(x) = (fibpair(x))0 . Thus λn [ pn ] is primitive recursive. = q · x + r) .For example. we first note that the auxiliary function λx [ fibpair(x) ]. . we can obtain a primitive recursive method of encoding ordered pairs of natural numbers as single numbers. .3). . below. is primitive recursive in view of fibpair(0) = fibpair(x + 1) = 20 3 1 . x) = (y mod x) = r. We can then write p0 = 2. x) = ⌊y/x⌋ = q. pn+1 = least x ≤ pn ! + 1 such that Prime (x) and x > pn . . As another application of the bounded least number operator. The sequence a0 . note that the functions Quotient (y. This method of prime power coding will be used extensively in the proof of the Enumeration Theorem.

y) = least z ≤ x · y such that Remainder(z. Hint: You may want to use the fact that |π − a/b| > 1/b42 for all integers a. f (n − 1) via prime power coding. in view of the recursion equations f (0) = 1 and f (n + 1) = f (n) · ph(n. . .. This general technique is known as course-of-values recursion. 4 3 5 7 9 2n + 1 n=0 12 ∞ n−1 e . We use the well-known series π (−1)n 1 1 1 1 = 1 − + − + − ··· = . Solution. the greatest common divisor and least common multiple of x and y. 30–42. On the approximation of π. where g(n) = the least x < 4 · 102n such that (x + 1)2 > 2 · 102n . Mahler. Solution. This result is due to K. LCM(x. y). Show that the 1-place number-theoretic function f (n) = the nth decimal digit of π = 3. √ Solution. Proc. It now follows that f (n) = (f (n + 1))n is primitive recursive.141592 · · · is primitive recursive.Exercise 1.. Then f (n) is primitive recursive. A. z) = 0. Ser.1. Show that the 2-place number-theoretic functions GCD(x..e. 10).f(n)) . Exercise 1.1. y) = least z ≤ max(x. n where h(n. i. y) = 0.12. b > 1.9. Show that the function λn nth digit of 2 is primitive recursive. f (n) encodes the variable-length finite sequence f (0). x) = Remainder(z. 15. . Indagationes Math. z) = Remainder(y. GCD(x. Akad. y) and LCM(x.1. The nth digit of 2 is f (n) = Remainder(g(n). y) such that Remainder(x. √ Exercise 1. 1953. Wetensch. Exercise 1.10. Nederl.11. Show that the 1-place number-theoretic function f (n) given by n−1 f (n) = 1 + k=0 f (k)n is primitive recursive. . f (1).1. are primitive recursive. 56. z) = 1 + k=0 ((z)k )n . Consider the so-called course-of-values function n−1 f (n) = k=0 pk f (k) . Solution.

A1 (x) = 2x . 1. b(k)) = g(m. a(k(n)). yet is clearly computable in some intuitive sense. Clearly this is primitive recursive.e. The precise concept of computability which we have in mind will be explained in the next section. so in particular f (n) = the nth digit of Sk . it follows that Sk and π have the same first n digits. Note also that the functions g(n. Since π lies between Sk and Sk+1 .1. for each n there exists k < 1050n such that Sk and Sk+1 have the same first n digits. b(k(n))) where k(n) = the least k < 1050n such n that m=0 g(m. Show that. x Thus A0 (x) = 2x. b) = Rem(Quot(10g(n.b) a.a. 13 ·2 ·· ..2 (the Ackermann hierarchy).2 The Ackermann Function In this section we present an example of a function which is not primitive recursive. Using the bounded least number operator. 10) = the nth digit of a/b is primitive recursive. An+1 (x) = An An · · · An (1) . By Mahler’s result. Definition 1. a(k). 1. 2.i. Show that (a) An (x + 1) > An (x) > x for all x ≥ 1 and all n. namely a(k) = (−1)n 4(2k + 1)! 2n + 1 n=0 k ∞ and b(k) = (2k + 1)!. we have f (n) = h(n. Exercise 1.2. b(k + 1)).2. b) = (µi < 10n b) (10i a ≥ 10n b) and h(n. (−1)n 4 4 4 4 4 . λx [ An (x) ] is primitive recursive. a. etc. π = 4 − + − + − ··· = 3 5 7 9 2n + 1 n=0 Let Sk be the kth partial sum of this series. We define a sequence of 1-place functions An . n ∈ N. for each n. a. We have Sk = a(k)/b(k) where the functions a(k) and b(k) are primitive recursive. as follows: A0 (x) = 2x. b). A2 (x) = 22 (height x). a(k + 1).

.3. xk ) ≤ An (max(3. Show that for each k-place primitive recursive function λx1 . Compute An (x) for all n. xk )) for all x1 . Also An+1 (2) = An (An (1)) = An (2). . . xk ) ] there exists n such that f (x1 . Use this to show that the Ackermann function is computable.1: The Ackermann branches.1 shows An (x) for small values of n. Show that the 2-place function λnx [ An (x) ] is not primitive recursive. 3. Solutions. 4.. Also. recursive. . This is known as the Ackermann function. . xk . x1 .e. . . x with n + x ≤ 8. hence by induction An (2) = A0 (2) = 4. . 1. hence by induction An (1) = A0 (1) = 2. for all n and x we have An+1 (x + 1) = An (An+1 (x)). . . x. 0 1 2 4 4 4 4 4 ·2 ·· 3 6 8 16 216 22 (height 216 ) 2· 2 ·· ·2 ·· 4 8 16 2 16 ·2 ·· 5 10 32 22 16 ·2 ·· A0 A1 A2 A3 A4 A5 A6 0 2 1 2 1 2 1 2 1 2 1 2 1 2 22 (height 216 ) A3 (22 (height 216 )) ·2 ·· 22 (height 22 (height 216 )) 4 A3 (2 4 (height 216 )) 2. Table 1. . i. Show that the 3-place relation λnxy [ An (x) = y ] is primitive recursive. Solution. For all n we have An+1 (1) = An (1). Table 1. . and An+1 (3) = An (4) for all n. An (2) = 4. .(b) An+1 (x) ≥ An (x + 1) for all x ≥ 3 and all n. xk [ f (x1 . . Prove the following: 14 . in particular An+1 (3) = An (An+1 (2)) = An (4). Show that An (1) = 2. . in the sense of Section 1. . 5. . .

Suppose f is obtained by generalized composition. . f (y. . . . . gm (x1 . . xk )). . x1 . For n = 0 we have A0 (x + 1) = 2x + 2 > 2x = A0 (x) for all x. (d) The 1-place function λx (Ax (x)) is not primitive recursive. . For n + 1 and x ≥ 1 we have An+1 (x + 1) = An (An+1 (x)) > An+1 (x) by inductive hypothesis. x1 . . .(b) An+1 (x) ≥ An (x + 1) for all x ≥ 3 and all n. . . xk ) = g(x1 . . Next we prove An+1 (x) ≥ An (x+1) for x ≥ 3. xk ). . An+2 covers f . . x1 . . xk ). . . We first claim that f (y. We then have f (x1 . = h(y. .e. . xk )). . . . . xk ) ≤ An (max(3. xk ) = h(g1 (x1 . . x1 . . . . xk ) ≤ An+1 (y + max(3. i. x1 . xk ) there exists n such that An covers f . . xk ). . . . xk ) + 1) An+2 (max(3. . g1 (x1 . xk )) 15 . . . . xk ))) An (An+1 (max(3. . . . . . . Next we prove that each primitive recursive function is covered by An for some n. . gm . say f (x1 . it follows that An+1 (x) > x for all x. by induction on n. . . x1 . . xk ) = ≤ ≤ = ≤ i. . . . . . . . . since An is strictly monotone and An (x + 1) ≥ x + 2 by what has already been proved. . xk ).. . . x1 . say f (0. xk . . . . . . . We begin by noting that the initial functions are covered by A0 . . . . . . and A0 (x) = 2x > x for x ≥ 1. . . For x = 3 we have An+1 (3) = An (4) as noted above. . We prove this by induction on the class of primitive recursive functions. . Let n be such that An covers h and An+1 covers g. .e. First we prove An (x + 1) > An (x) > x for x ≥ 1. . . x1 . . . . . h(g1 (x1 . . xk )) for all x1 . . xk ). . . . . . xk ) f (y + 1. xk )) An (max(3. Let n be such that An covers h and An+1 covers g1 . . . by induction on x. Solution. . x1 . f (x1 . . (a) An (x + 1) > An (x) > x for all x ≥ 1 and all n. Since An+1 (0) > 0. . xk ). . . . . . gm (x1 . Thus An+1 is strictly monotone. xk ))) An+1 (max(3. . . . (e) The 2-place function λxy (Ax (y)) is not primitive recursive.. gm (x1 . . and inductively An+1 (x + 1) = An (An+1 (x)) ≥ An (An (x + 1)) ≥ An (x + 2). Suppose f is obtained by primitive recursion. . (c) For each primitive recursive function f (x1 . x1 . . . .

xk )). . y. . . y. . We prove this by induction on y. Solution. x1 . y. . . . xk ))) An+2 (max(3. An+3 covers f . . . . z | Ax (y) = z} is primitive recursive. hence covered by An for some n ≥ 3. x1 . x1 . y1 . x1 . Thus the 1-place function Ax (x) is not primitive recursive. . For all x. if Ax (x) were primitive recursive. . it follows that 0 < y ′ < Ax−1 (y ′ ) = Ax (y). y1 . a contradiction. . . . y. x1 . . 3. It follows immediately that the 2-place function Ax (y) is not primitive recursive. . . . y > 0 we have 0 < y < Ax (y) = Ax−1 (Ax (y − 1)) = Ax−1 (y ′ ) where y ′ = Ax (y − 1). . . . . xk )). Repeating this step x times. xk ). yx is > 0 and < Ax (y).for all y. We then have f (y. Thus the 3-place predicate Ax (y) = z can be defined 16 . . . . xk ) = ≤ ≤ = = h(y. xk )) An+1 (An+2 (max(3. yx starting with y such that Ax (y) = Ax (y0 ) = Ax−1 (y1 ) = Ax−2 (y2 ) = · · · = A0 (yx ) = 2yx . . . . . . . xk ) An (An+1 (y + max(3. . . . Now. For the inductive step we have f (y + 1. . y2 . if y > 2 then we also have x < Ax (y). y. and each of y0 . An+1 (y + max(3. xk )) and this proves our claim. . Moreover. we obtain a finite sequence y0 . xk . Hence λx (Ax (x)) is recursive. x1 . xk ) + 1) An+3 (max(3. x1 . xk ). x1 . x1 . Since Ax−1 (y ′ ) = Ax (y) ≥ 2. then Ax (x) + 1 would be primitive recursive. . . y. . . . .. . . x1 . x1 . xk ) An (max(3. . . . . . . . xk ) An (max(3. x1 . x1 . . . . . xk ) = g(x1 . xk )). . . i. xk )) An+1 (2 max(3. . But then in particular An (n) + 1 ≤ An (max(3. . x1 . . y. . xk ) ≤ ≤ ≤ = ≤ An+1 (y + max(3. . . f (y. Show that the 3-place relation { x. . . Use this to prove that λxy (Ax (y)) is recursive. x1 . . . . . This completes the proof that each primitive recursive function is covered by An for some n. xk ))) An+1 (y + 1 + max(3. . x1 . . xk ) ≤ An+1 (max(3. . For y = 0 we have f (0. . . x1 .e. . n)) = An (n). x1 . . . . . . . f (y. . .

. An Ri instruction is executed by testing Ri for emptiness. Each program contains exactly one start instruction. . specified.. y. 17 . yx < z (y0 = y ∧ ∀i < x (yi+1 = Ax−i (yi − 1)) ∧ z = 2yx )). It follows immediately that the 1-place function Ax (x) is recursive. The basic actions that the machine can perform are to increment or decrement a register. instruction. start. the run will eventually halt with x + y in R3 . i. add or subtract 1 from the number contained in it. it follows that the 3-place predicate Ax (y) = z is primitive recursive. At any given time. Actually. we decrement it and then proceed to the other of the two specified instructions. For example. . which is executed first. We show that the primitive recursive functions form a proper subclass of the computable functions. consider the addition program depicted in Figure 1. See Figure 1. recursive) number-theoretic functions. y1 . A register machine program consists of finitely many instructions linked together in a flow diagram indicating the order in which the instructions are to + − be executed. . we see that the 2-place function Ax (y) is recursive.1. If we run this program starting with natural numbers x and y in R1 and R2 respectively. If the number contained in Ri is 0. and stop.. we say that Ri is empty.3 Computable Functions In this section we define the class of computable (i. . + An Ri instruction is executed by incrementing Ri and then proceeding to an− other. . A stop instruction causes execution of the program to halt.by course-of-values recursion on z as follows: Ax (y) = z if and only if (x = 0 ∧ z = 2y) ∨ (x > 0 ∧ y = 0 ∧ z = 1) ∨ (x > 0 ∧ y = 1 ∧ z = 2) ∨ (x > 0 ∧ y = 2 ∧ z = 4) ∨ (x > 0 ∧ y > 2 ∧ x < z ∧ ∃y0 . z < w}. Ri . 1.e. There are four types of instructions: Ri . R2 . the function being defined by primitive recursion is In any case. . . If Ri is nonempty. Our definition will be given in terms of a register machine. Ri . each register contains a natural number.2. . If Ri is empty. . we proceed to one of two specified instructions. At any given time. . . all but finitely many of the registers are empty. We assume the existence of infinitely many registers R1 . Applying the least number operator. a(w) = {p2x 3y 5z | Ax (y) = z ∧ x.e.

otherwise decrement Ri ?? ?? ?? ? HIJK / ONML stop ?     terminate run of program Figure 1.1: Register Machine Instructions ONML HIJK start `abc g d = HIJK / ONML R− 3 e / ONML e / ONML e / ONML − − HIJK HIJK HIJK stop R1 R2 O O  + HIJK ONML R3  + HIJK ONML R3 Figure 1.2: An Addition Program 18 .HIJK ONML start / begin run of program ?? ?? ?? ? + HIJK / ONML / Ri ?     increment register Ri ?? ?? ?? ? − HIJK / ONML e / Ri ?? ? ??  ??  ?   if Ri empty go to e.

. the addition program of Figure 1. . . .e.3. . xk ) ] is said to be computable if there exists a register machine program P which computes it. . xk ) in Rk+1 . . . P(x1 . . . xk in Rk .1 (Computable Functions). . . .3. elements of N. . i. . S = λx [ x + 1 ]. xk ∈ N. xk be natural numbers. . Zero HIJK ONML start HIJK ONML start HIJK / ONML stop − HIJK / ONML R1 O Successor e / ONML + HIJK R2 HIJK / ONML stop  + HIJK ONML R2 HIJK ONML start − HIJK / ONML Ri O Projection e / ONML HIJK stop  + PQRS WVUT Rk+1 Figure 1. . . . . are computable. and all other registers empty. . .e. . . and Pki = λx1 . Proof. Uniqueness follows from the fact that the register machine operates deterministically. .3.3: The Initial Functions 19 . S.2 shows that λx1 x2 [ x1 + x2 ] is computable. and Pki .. xk ) to denote the unique run of P starting with x1 in R1 . A k-place number-theoretic function λx1 . . 1 ≤ i ≤ k. The initial functions Z. We write P(x1 . . The functions Z = λx [ 0 ]. Lemma 1. .Let P be a register machine program. xk [ xi ] are computed by the register machine programs given in Figure 1.2. and let x1 . for all x1 . . xk [ f (x1 . We shall now prove that all primitive recursive functions are computable. xk ) eventually halts with y = f (x1 . For example. i. . . . Definition 1.

To compute λx1 . e / G m XYZ[ / _^]\ G− k+1 m O  + HIJK ONML Hm + HIJK ONML o Fk+1 e / H  /_^]\ − XYZ[ Hm+1 e  HIJK ONML stop Figure 1. We use Gj1 . . . We use H1 . . Gm . . xk [ g1 (x1 . . . . . . .. .3. . Gjk . . e / ONML e / G − HIJK Fk 1 G  + HIJK ONML G1k  . Fk . to denote the registers on which H is executed. 20 . ym ) ] are computed by register machine programs G1 . . . i ≥ k + 2. . . H will be among the auxiliary registers of F.3. . Lemma 1. In order to make it easy for F to call G1 . (The auxiliary registers of F are the registers Fi . For convenience we regard these programs as being executed on pairwise disjoint sets of registers. Our program F is given in Figure 1. . xk ) = h(g1 (x1 . . . . xk ) ] where f (x1 . . to denote the registers on which Gj is executed. . . . xk ) ] . . . . Hm+1 .4: Generalized Composition Lemma 1. . . .3.  + HIJK ONML G m1 e / . The class of computable functions is closed under generalized composition. . . Gm . . Assume that λx1 . Fk+1 . . . . 1 ≤ j ≤ m. H respectively. .4. . . . . .  + HIJK ONML G mk XYZ[ / _^]\ G−k+1 1 O  + HIJK ONML H1 e / . the registers of G1 .4. Proof. . .) Actually. xk ) ] and λy1 . Gm . . . Hm . . we shall use a register machine program F which we regard as being executed on registers F1 . the auxiliary registers of F consist precisely of the registers of G1 . H. xk [ f (x1 . . ym [ h(y1 . . . gm (x1 . . xk [ gm (x1 . . . . . . .k+1 . . . . The class of computable functions is closed under primitive recursion. . . . . . .. H. . . . . . Gj.HIJK ONML start  − HIJK ONML F1 G  + HIJK ONML G11  . Gm . . . .. . . . . λx1 . xk ). . . . xk )) .. . . .

F2 . . e HIJK ONML start HIJK / ONML R− z 1 − HIJK / ONML R2 D e / ONML − HIJK R4 O  + HIJK ONML R2 e  HIJK ONML stop  + HIJK ONML R3  + HIJK ONML R4 Figure 1.4. .6.e. x1 . . . . . f (y. Fk+2 . Hk+2 . xk ) . Fk+1 . xk ) ] where f (0. xk ) ] and λyzx1 . . xk [ h(y. Gk . .3. . x1 .In proving this lemma. Example 1. U and V . Proof of Lemma 1. . Note that x0 = 1 for all x. xk ) = f (y + 1. Every primitive recursive function is computable. . z. . The multiplication function λx1 x2 [ x1 · x2 ] is computed by iterated addition.5: A Multiplication Program Exercise 1. Let us first illustrate this idea with a simple example. . . .5. .3. . See Figure 1. . using the program given in Figure 1. . . Hk+3 . Gk+1 . . xk [ f (y.. . xk ) . .6. xk ). . xk ) = g(x1 . . .5. . . the idea will be to write a program containing a loop which repeatedly calls the iterator h. . . . . x1 . H2 . . Theorem 1. i. . . . To compute λyx1 . we use a register machine program F with registers F1 . . . . h(y. x1 . x1 . . Assume that λx1 . . . xk [ g(x1 .3. the 2-place number-theoretic function exp(x. xk ) ] are computed by register machine programs G and H with registers G1 . .7. y) = xy . . . . . . . H3 . . 21 . and H1 . The auxiliary registers Fi . x1 . Write a register machine program which computes the exponential function. .3. . . . . . . respectively. i ≥ k + 3 of F consist of the registers of G and H plus two additional registers. . . .

o e − HIJK ONML o H4 } `fec g d + HIJK ONML H2 O − PQRS WVUT Hk+3 e − HIJK ONML o H3 }  e O e e / ONML e / ONML e − HIJK HIJK F2 V− D O  HIJK ONML V+  HIJK ONML H+ 3 / ...6: Primitive Recursion 22 . e / ONML e / ONML − HIJK HIJK Fk+1 V− D O  HIJK ONML V+  HIJK ONML G+ k e / G  − HIJK ONML F1 + HIJK ONML H2 O  + HIJK ONML Fk+1 + HIJK ONML o Fk+2  − PQRS / WVUT G k+1 e  / WVUT PQRS G− k+1 e e  HIJK ONML stop z{ ~ HIJK ONML o U+      HIJK ONML e / ONML HIJK U− V− D O  HIJK ONML V+  HIJK ONML H+ 1 `fec g d . e / ONML − HIJK Fk+1 D  HIJK ONML V+  + PQRS WVUT H k+2 e / ONML e HIJK V− O  + HIJK ONML F k+1 + HIJK ONML o Fk+2 / H  HIJK ONML F− 1 − HIJK fec ONML abd H2 O ] e  HIJK ONML U+  + HIJK ONML F 2 − HIJK fec / ONML abd H1 ] e  / WVUT − PQRS H k+3 e  HIJK ONML stop Figure 1.HIJK ONML start  − HIJK ONML F2 D  HIJK ONML V+  HIJK ONML G+ 1 e / ONML e HIJK V− O  + HIJK ONML F2 / .....

.1 (Recursive Functions and Predicates). . In particular. A k-place (total) function is said to be recursive if and only if it is computable. . . ψ(x1 .4. Definition 1. . . xk ∈ N.3 (Unbounded Least Number Operator). Definition 1. .2 (Partial Recursive Functions). y) holds is partial recursive. xk ) ≃ least y such that P (x1 . . xk . P We sometimes abbreviate this as ψ : Nk −→ N.Proof. we say that ψ is total. We use dom(ψ) and rng(ψ) to denote the domain and range of ψ.4. y) be a (k+1)-place recursive predicate. If dom(ψ) = Nk . . . . . 23 . xk ) is defined if and only if P(x1 . . Since the primitive recursive functions were defined as the smallest such class. The above lemmas show that the computable functions form a class which contains the initial functions and is closed under generalized composition and primitive recursion. . . . . . . A k-place partial function ψ is said to be partial recursive if it is computed by some register machine program P. . for all x1 . . A k-place predicate is said to be recursive if its characteristic function is recursive. . xk . Then the k-place partial function ψ defined by ψ(x1 . xk ) eventually stops. If E is such an expression. . This means that. . ψ(x1 . or both undefined. . We say that E is undefined or divergent (abbreviated E ↑) if E does not have a numerical value.4. xk ) + 3 has a numerical value if and only if x1 . . . . Thus a total k-place function is just what we have previously called a k-place function.4 Partial Recursive Functions A k-place partial function is a function ψ : dom(ψ) → N where dom(ψ) ⊆ Nk . . The use of partial functions leads to expressions which may or may not have a numerical value. . . ψ(x1 . . . We write E1 ≃ E2 to mean that E1 and E2 are both defined and equal. . as in the following lemma. (For example. Lemma 1. . our theorem follows. One way this can happen is because of an unbounded search. 1. . Let P (x1 . . . we say that E is defined or convergent (abbreviated E ↓) if E has a numerical value. . xk ) stops . Partial recursive functions arise because a particular run of a register machine program may or may not eventually stop. xk ) ≃ the number in Rk+1 if and when P(x1 . . respectively. xk ∈ dom(ψ)).

.7: Minimization Proof.4. Assume that χP is computed by a register machine program P with registers P1 . i ≥ k + 2. . . Exercise 1. . . .1. The unbounded least number operator is sometimes called the minimization operator. Hint: Use an infinite series such as 1 1 1 1 π = 1 − + − + − ··· 4 3 5 7 9 plus the fact that π is irrational.. . . of F are the registers of P plus an additional register V . A solution not using Mahler’s result is as follows. . e / ONML e / ONML e − HIJK HIJK Fk+1 V− D O  HIJK ONML V+  + HIJK ONML Pk+1  + HIJK ONML Fk+1 _?? ??e ?? ? / P HIJK / ONML P− k+2  HIJK ONML V+  + ONML HIJK P1  + HIJK ONML F1  HIJK ONML stop Figure 1. .12. .4. we shall see that all partial recursive functions can be obtained from primitive recursive functions by composition and minimization. . Solution. See Figure 1. Fk+1 . Pk+1 .7. o e } − HIJK ONML o P2 `fec g d e − HIJK ONML P1 } e / . This follows from Exercise 1. . To compute ψ we use a register machine program F with registers F1 . Show that the function f (n) = nth digit of π is recursive. As a byproduct of the work in the next section.. Pk+2 . The auxiliary registers Fi . . . Let Sk be the kth partial sum of the alternating series π =4− (−1)n 4 4 4 4 4 + − + − ··· = . F2 . 3 5 7 9 2n + 1 n=0 24 ∞ . Pk .. .HIJK ONML start + HIJK ONML o Fk+1 ~ ~~ ~~  ~~ − HIJK HIJK ONML e / ONML F1 V− D O `fec g d . .

5.6. By course-of-values recursion. then the inverse permutation f −1 is also recursive. although the oneto-one increasing function x → Ax (x) is not. Let f : N −→ N be a permutation of N.. This number will be called the G¨del number of E and will also be called an o index of the k-place partial recursive function which is computed by E 25 . because f −1 (2x) = πB (x) = Ax (x). Since π lies between Sk and Sk+1 .4.e. the set of natural numbers. Give an example of a primitive recursive permutation of N whose inverse is not primitive recursive. Clearly this is recursive. However. S = {πS (0) < πS (1) < · · · < πS (n) < πS (n + 1) < · · ·} where πS (n) = the nth element of S. Note also that the function h(n. For any infinite set S ⊆ N. Exercise 1.2. it follows that for each n there exists k such that Sk and Sk+1 have the same first n digits. b(k)) = h(m.7. y | Ax (x) = y} is primitive recursive. Then B is primitive recursive.We have Sk = a(k)/b(k) where a(k) and b(k) are primitive recursive. b(k + m=0 1)). 1. b) = nth digit of a/b is primitive recursive. Let B be the range of x → Ax (x). Note also that B is infinite and coinfinite.e. Since π is irrational. a(k(n)). Using the least number operator. so in particular f (n) = the nth digit of Sk . Using the least number operator. a(k). let πS : N → N be the principal function of S. i. Solution.2. (By 1. Exercise 1. Solution.2 it suffices to show that the Ackermann function λnx [ An (x) ] is computable. i. we have f (n) = h(n. Let f be the permutation of N defined by f (y) = −1 2πB (y) −1 2πN\B (y) 1−1 onto if y ∈ B. From our study of the Ackermann function in Section 1. +1 if y ∈ N \ B. Show that there exists a computable function which is not primitive recursive.. Show that if f is recursive. it follows that Sk and π have the same first n digits.4. we have f −1 (y) = the least x such that f (x) = y. b(k(n))) where k(n) = the least k such that n h(m. f is primitive recursive. a. B = {y | ∃x (Ax (x) = y)}.4.5 The Enumeration Theorem To each register machine program E we shall assign a unique number e = #(E). we know that the predicate { x. f −1 is not primitive recursive.) Exercise 1. a(k + 1). because y ∈ B ⇔ y−1 x=0 Ax (x) = y.

Rk which contain the arguments x1 . 3.1. . . . or if Ri is empty go to In0 .8. Program (e. . We assume that our program E is given as a numbered sequence of instructions I1 .2. where o #(Im ) = 3i · 5n0 i for Im as in (1. Proof. . I2 . . To each instruction Im we assign a G¨del number #(Im ). The G¨del number of the entire program E is then defined as o l #(E) = m=1 p#(Im ) . l) l=0 26 . Il . (1. p2 . are the prime numbers 2. 1 ≤ m ≤ l. . . n0 2·3 ·5 ·7 n1 for Im as in (1.0) (1. . which computes λx [ x+1 ]. . .Recall that our register machine is equipped with infinitely many registers Ri . is of the form increment Ri then go to instruction In0 .1) Here n0 and n1 are in the range 0 ≤ n ≤ l. xk . in increasing order. m where p0 . We have Program (e) ≡ e 32 · 50 = 9 . Listing the instructions I1 .0) . Lemma 1.5. i ≥ 1. 5. . By convention our machine starts by executing I1 and stops when it attempts to execute the nonexistent instruction I0 . . The 1-place predicate Program (e) ≡ (e is the G¨del number of some register machine program) o is primitive recursive. 2 · 31 · 53 · 72 = 2 · 3 · 125 · 49 = 36750 . otherwise decrement Ri then go to instruction In1 .5. 32 · 51 = 45 . we have #(I1 ) = #(I2 ) = #(I3 ) = so that #(E) = 3#(I1 ) · 5#(I2 ) · 7#(I3 ) = 336750 · 545 · 79 . .1) . . Each instruction Im . p1 . Initially all of the registers are empty except for R1 . Example 1. . . . I3 as shown in the figure. Let E be the program in Figure 1.

l) l e l l ≡ e= m=1 (e) pm m ∧ . Il . and undefined otherwise. m=1 i=1 n0 =0 n1 =0 (e)m = 3i · 5n0 ∨ (e)m = 2 · 3i · 5n0 · 7n1 the idea being that (e)m = #(Im ). .4. we define ϕe (x1 . if x > 0. Note that if e is not the G¨del number of a register machine program. .e. xk . o Clearly ψ has many different indices.3. We denote by ϕe the k-place partial computable function which is computed by the register machine program E whose G¨del number is o e. . e is the G¨del number of a program which computes ψ. .5. where e = #(E). so in this case ϕe is the empty function. · · · . an index of ψ is any number e (k) such that ψ = ϕe . since there are many different programs which compute ψ. Definition 1. xk ) stops. . If ψ is a k-place partial recursive function. . l) says that e is the G¨del number of a register machine o program consisting of l instructions I1 . . · · · . xk ) is undefined for all x1 .5. Find an index of the function α(x) = 1 0 27 if x = 0. In more detail. .I1 HIJK ONML start − HIJK / ONML R1 O I3 e / ONML + HIJK R2 HIJK / ONML stop  + HIJK ONML R2 I2 Figure 1. This proves the lemma. We then have l Program (e. · · · . then o (k) (k) ϕe (x1 . Exercise 1.. . xk ) ≃ (k) (k) the number in Rk+1 if and when E(x1 . i.8: A Program with Labeled Instructions where Program (e.

In other words. . This concept underlies the stored program digital computer. For each k ≥ 1. .5. . the following easy lemma will be useful. In the proof of the Enumeration Theorem. a register machine program which can emulate the action of any other register machine program. . . the (k+1)-place partial function λex1 . . Then the k-place function f defined by f (x1 . xk [ ϕ(k) (x1 . Proof of the Enumeration Theorem. . . . i. for each k-tuple x1 .5 (The Enumeration Theorem). so (1) an index of α is e = 3#(I1 ) · 5#(I2 ) = 3150 · 59 . This is clear since f = f1 · χP1 + f2 · χP2 . Remark 1. . . xk ∈ Nk . ϕe = α. either P1 (x1 . i. . xk ) = is primitive recursive. . . . . . .. . We have #(I1 ) = 2 · 31 · 52 · 70 = 150 and #(I2 ) = 32 · 50 = 9.e. . Let P1 and P2 be k-place primitive recursive predicates and let f1 and f2 be k-place primitive recursive functions.5. xk ) holds 28 . The main theorem on indices reads as follows. .7 (Definition by Cases). . xk ) if P2 (x1 . . Theorem 1. xk ) holds f2 (x2 . . . xk ) or P2 (x1 . . Proof. . .. . xk ) holds but not both. xk ) if P1 (x1 .6. The Enumeration Theorem entails the existence of a “universal” program. . . f1 (x1 .5. Lemma 1. . The extension to more than two cases is also easy.Solution.e. xk ) ] e is partial recursive. . Assume that P1 and P2 are mutually exclusive and exhaustive. Clearly the labeled program I1 HIJK ONML start HIJK / ONML R− 1 I2 ?? ?? ?? ?  HIJK ONML stop e / ONML + HIJK R2 computes α. . . . . .

29 State (e. . Stop (e. · px k 1 k (begin by executing I1 ) . (z)i = zi . . x1 . . Note that Stop (e. . xk . xk ) ≃ least n such that (State (e. . . .) We then use composition to get FinalState (e. . n0 = ((e)m )2 . z) =  z · p−1 · p−m+n1  0 i    z m = (z)0 . x1 . We use the least number operator to obtain Stop (e. . xk ) ≃ and Output (e. . x1 . We have State (e. x1 . xk ) ≃ (FinalState (e. . x1 . i = ((e)m )1 . 0) = 1 p0 · px 1 · . . n + 1) = where  −m+n0  z · pi · p0     z · p−m+n0 0 NextState (e. xk )) (the state of our machine if and when it stops) . for all i ≥ 1. . xk . . xk . . . . it is partial recursive.The idea of the proof is to represent the state of E(x1 . if ((e)m )0 = 1 and (z)i > 0 . . xk . . . e This completes the proof of the Enumeration Theorem. . x1 . i where zi is the number in register Ri . . . . x1 . xk ) after executing n instructions by a single number z = = State (e. . . we clearly have ϕ(k) (x1 . . . . xk . . State (e. (The idea is that our machine stops if and when it is about to execute I0 . primitive recursion and the least number operator. x1 . xk . . State (e. n))0 = 0 ∧ Program (e) . . . . x1 . . . otherwise . . . . x1 . xk ) ≃ Output (e. . . . . . . . . . NextState (e. x1 . Since the Output function was obtained by composition. and Im is the next instruction to be executed. for all e and x1 . . . . . We are now ready to prove the theorem. Note that (z)0 = m and. . if ((e)m )0 = 1 and (z)i = 0 . x1 . n)) . Moreover. . . . . xk ))k+1 (the number that is finally in register Rk+1 ) . n) pm · 0 ∞ i=1 pzi . . xk . . x1 . xk ) . . . if ((e)m )0 = 0 . . . n1 = ((e)m )3 . xk ) is undefined if e is not the G¨del number of a register o machine program. . . We first show that the State function is primitive recursive. .

8.9 in R. We replace We replace HIJK / ONML P− i HIJK / ONML P+ i / in P by Figure 1. . . there exists an e such that f = λx1 . . Pk . . zs respectively..5. . .6 Consequences of the Enumeration Theorem In this section we present some important consequences of the Enumeration Theorem. (See also Exercise 1. Given a k-place partial recursive function ψ(x1 . . then R1 contains z z = p11 · · · pzs . 2.Exercise 1. xk ) ] is primitive recursive. . . .10. We may safely assume that. Pk+1 .. R1 and R2 . .5.5. .2. xk ) ] is majorized by some primitive recursive function. . . . Let P1 . . . . . show that there is a 1-place partial recursive function ψ ∗ (x) such that ψ ∗ (px1 · · · pxk ) ≃ pk+11 1 k ψ(x . ?? e ?? ?? ? B 1. We begin with a register machine program P which computes ψ(x1 . for each k-place primitive recursive function f : Nk → N. . . xk ). . R1 and R2 . x1 . Incrementing (decrementing) Pi corres sponds to multiplication (division) by pi . while R2 contains 0. Construct a k+1-place total recursive function Φk : Nk+1 → N with the following properties: 1. x1 . We replace HIJK / ONML in P by Figure 1. The idea is that. . xk [ Φk (e. .9. . . Exercise 1. Let f : Nk → N be a k-place total recursive function. .. xk . stop / A in P by Figure 1. . .xk ) for all x1 . . . 30 . it leaves all registers except Pk+1 empty. .. . . . . xk ). Solution. x1 . . .10 in R. . and ψ ∗ is computable by a register machine program which uses only two registers. Ps be the registers used in P. . whenever P(x1 . We transform P into a program R which uses only two registers. xk ) halts. Show that f is primitive recursive if and only if there exists an index e of f such that λx1 . Ps contain z1 . .2. Fix k ≥ 1. . . . xk ) ].) Exercise 1. xk [ Φk (e. for each e ∈ N. .11 in R. . . xk [ Stop (e. if P1 . Each instruction in P is replaced by a corresponding set of instructions in R. . the k-place function λx1 . . .

Figure 1. 00 00  HIJK / ONML R− 1 e / ONML e − HIJK R2 e / ONML + HIJK R1 O . . 00 . 00 . 00 . .O  − HIJK ONML R1 e / ONML + HIJK R1 + ONML HIJK R1 ?      + − HIJK ONML HIJK / ONML R1 R2 W00 e 000 00   00 . Figure 1.− HIJK / ONML R1 F e / ONML e − HIJK R2 / O  + HIJK ONML R1  + HIJK ONML R2  .  + HIJK ONML R2 + The number of R2 instructions is pi .9: Incrementing Pi + HIJK ONML R1 O ?      + − HIJK ONML HIJK ONML R2 R1 W00 00 00  00 . 00 00 0  /A + HIJK ONML R1 − The number of R1 instructions is pi . .10: Decrementing Pi 31 . B . .

6. and the FinalState and Output functions are obtained by composing the Stop and State functions with the primitive recursive function λz [ (z)k+1 ]. Namely.4. the Stop function is obtained from the State function by minimization.1. Let e be an index of f .3. hence f (e) ≃ ψ(e) ≃ f (e) + 1. Suppose ψ were extendible to a total recursive function f . Example 1. Corollary 1. The following characterizations of the class of partial recursive functions do not involve register machines and are similar to our definition of the class of primitive recursive functions.2. Next we present an interesting example showing that the consideration of partial functions is in some sense unavoidable or inherent in recursive function theory. Then (1) ψ(e) ≃ ϕe (e) + 1 ≃ f (e) + 1 is defined.4.6. Both corollaries are immediate from the previous theorem and Lemmas 1. We present an example of a partial recursive function ψ : P N −→ N which cannot be extended to a total recursive function f : N → N.3. This is immediate from the proof of the Enumeration Theorem.− HIJK / ONML R1 O + HIJK ONML R2 e / ONML HIJK stop  Figure 1. ψ is a partial recursive function. 1. 32 .6. Corollary 1. Proof. and minimization.4. primitive recursion. a contradiction. All partial recursive functions can be obtained from primitive recursive functions by composition and minimization.3. The class of partial recursive functions is the smallest class of functions containing the primitive recursive functions and closed under composition and minimization.2. we define ψ(x) ≃ ϕ(1) (x) + 1 . 1.3. 1. The State function is primitive recursive. The class of partial recursive functions is the smallest class of functions containing the initial functions and closed under composition.3.6.3.11: Stopping Theorem 1. x By the Enumeration Theorem. Proof.

assume that we are given a function f : Nk → N which is algorithmically computable in the intuitive sense. Example 1. More generally. 1.4. The precise nature of the states depends on the nature of the algorithm.8. the execution of the algorithm should be describable as a sequence of states with deterministic transition from one state to the next. and then we shall present the evidence for it. Church’s Thesis states the converse: All functions f : Nk → N which are algoritmically computable in the intuitive sense are in fact recursive. A set A ⊆ N is said to be recursive if its characteristic function χA : N → N is recursive.6. Exercise 1.5. Thus we should 33 (1) . Once this has been done. then the total function f : N → N defined by f (x) = ψ(x) 0 if x ∈ K.5. We present an example of a nonrecursive set.6. A pair of sets A. as mathematicians. the o transition from the G¨del number of one state to the G¨del number of the next o o state should be very simple. To present our evidence for Church’s thesis. it should be possible to view them as finite strings of symbols and to assign G¨del numbers to them.6.10 (Church’s Thesis).2. 1.6. but no matter what the states actually consist of. See Exercises 1.9.6.9. Let K be (1) the subset of N consisting of all x ∈ N such that ϕx (x) is defined.5.6. 1. (Caution: It can be shown that the 3-place predicate z = Ax (y) is primitive recursive. show that K0 and K1 are recursively inseparable. Exercise 1. a k-place predicate R ⊆ Nk is said to be recursive if its characteristic function χR : Nk → N is recursive. if x ∈ K / would be recursive. B ⊆ N is said to be recursively inseparable if there is no recursive set X such that A ⊆ X and X ∩ B = ∅. Letting Kn = {x ∈ N | ϕx (x) ≃ n}.7. in particular primitive recursive. The context of Church’s Thesis is that. Thus K = dom(ψ) where ψ is as in the previous example. If K were recursive. Since recursive functions are register machine computable. even though the 2-place function λxy [ Ax (y) ] is not primitive recursive.8. Show that there exists a set A ⊆ N which is recursive but not primitive recursive. we have an intuitive notion of what it means for a function f : Nk → N to be algorithmically computable.Definition 1. Definition 1.5. they are obviously algorithmically computable in the intuitive sense.2. Perhaps the most important consequence of the proof of the Enumeration Theorem is that it provides strong evidence for Church’s Thesis.6.) Remark 1. We shall first explain what Church’s Thesis says. contradicting the fact that ψ is not extendible to a total recursive function. We want to show that f is recursive. We claim that K is not recursive. Since the given algorithm for f is presumably deterministic.

n = 123456789). we shall take Church’s Thesis for granted and identify the class of partial recursive functions with the class of partial functions from Nk into N that are algorithmically computable in the intuitive sense. The fact that the intuitive notion of algorithmic computability is captured by a rigorous mathematical notion of recursiveness is one of the successes of modern mathematical logic.2 that f is recursive. For any particular n (e.7. Stop. Kleene’s equation calculus.5 (State.e. we shall not deal with this type of problem now. we mention: Example 1. but whose truth value is unknown at the present time. and (2) to present some important examples of such problems. 1. (An example of a problem in this sense is the Riemann Hypothesis.. However.) However. Some of the models of computation that have been analyzed in this way are: Turing machines. The statement “n is prime” represents the problem of deciding whether an arbitrary number n ∈ N is prime or composite. we consider a somewhat different concept. An instance of a problem is the specialization of the problem statement to a particular parameter value. etc. the word problem refers to a mathematical statement which has a definite truth value.). True or False. Since the set of prime numbers {n ∈ N | n is prime} 34 .6. the question of whether this particular n is prime or composite is an instance (i. it can be specialized to provide rigorous proofs that various models of computation (similar to but differing in details from register machine computability) give rise to exactly the same class of functions. special case) of the general “primality problem”.g. a problem is any mathematical statement that involves a parameter. Markov algorithms. For us in this section. NextState. Such an analysis will show that f is obtained from primitive recursive functions by means of composition and minimization. In certain contexts.1. It will then follow by Corollary 1.7 Unsolvable Problems The purpose of this section is twofold: (1) to discuss and make precise the concept of an unsolvable mathematical problem. As an example of a solvable problem. Here the parameter is the variable n. Instead. We begin with a preliminary clarification. This constitutes very strong evidence for Church’s thesis. otherwise unsolvable. The problem is said to be solvable if there exists such an algorithm. the recursive functions.5. The argument in the previous paragraph is of necessity nonrigorous. A solution of such a problem would be an algorithm which would enable us to compute the truth value of the problem statement for any given value of the parameter. From now on. There is no reason to think that the same analysis could not be carried out for any similar model.be able to carry out an analysis similar to what is in the proof of Theorem 1. Note that the same arguments and evidence apply more generally in case f is a partial rather than a total function.

.6.7. . . Xn ) = 0 has a solution in integers X1 . . Example 1. The ability to distinguish solvable problems from unsolvable ones is of basic importance for the mathematical enterprise. a9 ∈ Z} is nonrecursive. we have the following example of an unsolvable problem. a1 . . . Actually Matijaseviˇ produced a particular polynomial c p(X0 . Among the most famous unsolvable mathematical problems are: Example 1. p ∈ Z[X1 .e.is primitive recursive.6 is nonrecursive. The set K defined in Example 1. . Xn ∈ N. namely the set of parameter values n for which the problem statement holds.5 (The Halting Problem). In this case the parameter is w. X1 . One of these is the Halting Problem: To determine whether or not a given register machine program P will eventually stop. For example. whether or not the equation p(X1 . if started with all registers empty. our notion of unsolvable problem is related to the existence of a nonrecursive set. . and the word problem for G is solvable if and only if there exists an algorithm for determining whether or not a given word w is equal to 1 in G. for a given word w in the generators of G and their inverses. . A theorem of Matijaseviˇ shows that Hilbert’s Tenth c Problem is unsolvable. . for a given polynomial p in several variables with integral coeffients. .2. the general primality problem is solvable. By G¨del numbering. a9 ) = 0 for some a1 . .. Some famous unsolvable problems arise from computability theory itself.3 (Hilbert’s Tenth Problem).4 (Word Problems). such that {n ∈ N | p(n. Example 1. e 35 . Example 1. . .7. This problem encompasses the entire theory of Diophantine equations. . It is known that the word problem is solvable for some groups G and not solvable for others. Let G be a group presented by finitely many generators and relations. we can identify the Halting Problem with the problem of o deciding whether a given natural number e belongs to the set H = {e ∈ N | ϕ(1) (0) is defined} . . The word problem for G is the problem of determining. . Hilbert’s Tenth Problem is to determine. whether or not w = 1 in G. .7. . the word problem for free groups or groups with one relation is solvable. but Boone and Novikov have exhibited groups G with finitely many relations such that the word problem for G is unsolvable.7. Once again. . . Hence by Church’s Thesis there is no algorithm to decide whether or not a given number n belongs to K. Xn ]. X9 ) with 10 indeterminates. . the problem of computing the truth value of n ∈ K for any given n) as an unsolvable problem. On the other hand. It is therefore appropriate to describe the membership problem for K (i. . .

7. .e.7. . Then we can find a unary primitive recursive function f (x0 ) such that. then A is recursive.. The second statement follows since it is the contrapositive of the first. . Proof.e.6 (Unsolvability). The proof will be accomplished by showing that the problem of membership in K is “reducible” to the problem of membership in H. Let θ(x0 . .6). In order to prove that the set H is nonrecursive.We shall prove below that H is nonrecursive. the Halting Problem is unsolvable. . xk ∈ N.7 (Reducibility). . ϕf (x0 ) (x1 .5 above. . (k) 36 . n ∈ A implies f (n) ∈ B. . the issue of solvability or unsolvability of a particular problem was rephrased as an issue of whether or not a particular subset of N is recursive. Suppose that A is reducible to B. Our precise notion of reducibility is given by: Definition 1. Such considerations based on Church’s Thesis motivate the following definition: Definition 1. In this context. Definition 1. xk ) be a (k+1)-ary partial recursive function. for all x0 . . If ∅ =B= N. then for every recursive set A we have A ≤m B.7.10 (The Parametrization Theorem). We shall prove the unsolvability of the Halting Problem. 3. . . A ≤m B and B ≤m C imply A ≤m C. xk ) .7. Let A and B be subsets of N (i. . and n ∈ A / implies f (n) ∈ B. If A is nonrecursive. 2. i. x1 . x1 . In all of the above examples. and unsolvable if A is nonrecursive. 1} is recursive. we shall need the following important technical result: Theorem 1. cf. for all n ∈ N. i. the nonrecursiveness of the set H in Example 1.7.9. A ≤m A for all A ⊆ N. problems. xk ) ≃ θ(x0 ... . If B is recursive.8. .e. x1 .7. Write A ≤m B to mean that A is reducible to B. Show that 1. . . A problem is defined to be a subset of N.7. Exercise 1. If A ⊆ N is a problem in this sense. / Lemma 1. reducibility of one problem to another means that each instance of the former problem can be effectively converted to an equivalent instance of the latter problem. The first statement follows easily from the fact that χA (x) = χB (f (x)). then B is nonrecursive. We say that A is reducible to B if there exists a recursive function f : N → N such that. Recall that a set A ⊆ N is said to be recursive if and only if its characteristic function χA : N → {0. the problem A is said to be solvable if A is recursive.

. λx1 . The Halting Problem is unsolvable. . .7.5 is nonrecursive. and let T ′′ be the same as T ′ but modified so that the instructions are numbered I5 . . .11 (Unsolvability of the Halting Problem). xk . . Let f (x0 ) be the G¨del ′′′ number of Tx0 . . Il . .12: Parametrization We can now prove that the Halting Problem is unsolvable.7. xk [ θ′ (x1 .e. . . Il be the instructions of T ′ . . . Thus f (x0 ) is a primitive recursive function of x0 . . . let T ′ be a register machine program which computes the k+1-ary partial recursive function θ′ (x1 .. . Note that f (x0 ) = = l+x0 +5 m=1 l+4 m=1 l+4 pm #(Im ) pm #(Im ) · l+x0 +4 m=l+5 3 pm k+1 ·5m+1 ·5 · p2·3 0 +5 l+x k+1 5 ·75 where the first factor m=1 pm m does not depend on x0 . . . . xk . 37 . x1 . . xk ) . . Let T be a register machine program which computes the k+1-ary partial recursive function θ. x0 ) ].12 computes the k-ary partial recursive function λx1 . . e / / T ′′ = `abc g d e /WVUT e / ONML − PQRS HIJK stop Rk+2 O + PQRS WVUT Rk+1  x0 I4 Figure 1. . Il+x0 +5 . . the set H = {x ∈ N | ϕ(1) (0) is defined} x of Example 1. x0 ) ≃ θ(x0 . . ′′′ o The instructions of Tx0 are numbered as I1 . In other words. The idea of the proof is to let f (x0 ) be the G¨del number o of a program which is similar to T but has x0 hard-coded as the first argument of θ. . x1 . xk ) ]. I1 HIJK ONML start PQRS / WVUT R+ k+1 #(I ) Il+5 PQRS / WVUT R+ k+1 . Il+4 instead of I1 . Let I1 . Il+x0 +4 PQRS /WVUT R+ k+1 Il+x0 +5 PQRS /WVUT R− k+1 I5 · · · Il+4 I2 PQRS /WVUT R− k+1 I3 / . This completes the proof. . . the program Tx0 depicted in Figure 1. . . . Then for ′′′ any given x0 ∈ N.. Theorem 1. . .Proof.. . i. . . . Formally. xk [ θ(x0 ..

.7.7. 3. 2. Exercise 1.7. y) ≃ ϕx (x).6). On the other hand. i. we can find a primitive recursive function f : N → N such that ϕf (x) (y) ≃ θ(x. Note that θ is a 2-place function.e. { x. K = {x ∈ N | ϕ(1) (x) is defined} . 6. Since K is known to be nonrecursive (Example 1. i.Proof. IC = {x ∈ N | ϕ(1) ∈ C} . define IC to be the set of indices of functions in C. {x ∈ N | ϕx : N −→ N is total}. In particular.6.e.e. Exercise 1.6.13 (Rice’s Theorem).6. otherwise. (1) (1) 38 . y) . we can find a primitive recursive function f such that ϕf (x) (y) ≃ (1) (1) (1) (1) (1) (1) (1) (1) P ↑ ϕe1 (y) if ϕx (x) ↓. Show that the following sets and predicates are nonrecursive: 1. Let e1 be an index such (1) (1) / that ϕe1 ∈ C if and only if ϕe0 ∈ C.e. x Show that if ∅ = C = P then IC is nonrecursive. if x ∈ K then ϕx (x) is defined. if x ∈ K then ϕx (x) is undefined. 4. Solution. { x.. y ∈ N × N | y ∈ rng(ϕx )}. θ is partial recursive. By the Enumeration and Parametrization theorems. f (x) ∈ H. By the Enumeration Theorem. i. Let e0 be an index of the empty function. Let P be the class of 1-place partial recursive functions. By the Parametrization Theorem applied with k = 1.. hence ϕf (x) (0) is (1) (1) (1) (1) (1) defined. (1) Consider the partial recursive function θ(x. Thus K is reducible to H via f . f (x) ∈ H. ϕf (x) (y) ≃ ϕ(1) (x) x for all x and y. Let H be as in Example 1.e. 5.5 and let K be as in Example 1. This completes the proof. / (1) / hence ϕf (x) (0) is undefined. it will follow by Lemma 1. {x ∈ N | rng(ϕx ) is infinite}. i..7.8 that H is nonrecursive.12. y ∈ N × N | ϕx = ϕy }. {x ∈ N | 0 ∈ rng(ϕx )}.. For C ⊆ P. {x ∈ N | ϕx is the empty function}. x We shall prove that K is reducible to H. i.

while x ∈ K implies (1) (1) (1) (1) (1) / of IC (if ϕe1 ∈ C). . . . .8. Thus x ∈ K implies ϕf (x) = ϕe1 . . . . xk . . . Proof. xk . .8. .8 The Recursion Theorem In this section we present an interesting and mysterious theorem known as the Recursion Theorem. xk ) ≃ ≃ ≃ ϕ (k) ϕv (v) (1) (x1 . . y + 1) = 39 2y 1 A(x. for all x1 . . . u. As an example. . . Writing A(x. if we take θ(w. we can find primitive recursive functions f and d such that. . . . Then ϕd(v) (x1 .2. . Example 1. this becomes A(0. . Theorem 1. . . . xk ) .8. . Applying the Parametrization Theorem and the Enumeration Theorem. .e. x1 . Thus f reduces K either to IC (if ϕe1 ∈ C) or to the complement (1) 1. y) = Ax (y). . Then we can find an index e such that. ϕf (x) = ϕe0 . A(x + 1. .2) is computable. . . .. xk ) . x) = w + x. . . ϕf (w) (x1 . x1 . (k) ϕu (u) (1) ϕd(u) (x1 . This completes the proof.3. The recursion equations defining the Ackermann function can be written as A0 (x) = 2x An+1 (0) = 1 An+1 (x + 1) = An (An+1 (x)) . As another illustration of the Recursion Theorem. . then we obtain an (1) index e such that ϕe (x) = e + x for all x. Let v be an index of f ◦ d. y) = A(x + 1. xk ) (k) so we may take e = d(v). i. x1 . x1 . Example 1. . for all w. xk ) and (k) (k) ≃ θ(w. xk ) ≃ (k) θ(e./ for all x and y. . Let θ(w. x1 . . . . xk ) ≃ ϕ (1) (k) (x1 . xk ) ϕf (d(v)) (x1 . xk ) be a partial recursive function. ϕe (x1 . . . . In either case it follows that IC is not recursive. 0) = A(x + 1. . .1 (The Recursion Theorem). we now use it to prove that the Ackermann function λnx [ An (x) ] (see Section 1. y)) . . . . xk ) . . ϕv (u) = f (d(u)) for all u. . xk ) θ(d(v). . . .

y − 1)) if x > 0 and y > 0 .y) ) = dom(ϕ(1) ) ∪ dom(ϕ(1) ) .8. for all y. y).y) ) = dom(ϕ(1) ) ∩ dom(ϕy ) . y) such that. x (1) By the Enumeration Theorem together with the Recursion Theorem. · A(x − A(x. ϕe (x. y) such that. for all x and y. Find a primitive recursive function f (x. z. ϕe (x. 1. x y Solution. z. (2) 2. y) ≃ A(x. for all x and y. Then g(x. for all x and y. ϕe (x. 40 (1) (1) (1) . (1) ϕf (x. Find a primitive recursive function g(x. Exercise 1. y) such that.   · 1. dom(ϕh(x. y) = f (3x 5y ) has the desired property. 1. Find a primitive recursive function h(x. x (1) 3. y − 1)) if x > 0 and y > 0 . (1) dom(ϕg(x. y) = It is then straightforward to prove by induction on x that. This completes the proof. y) = g(3x 5y ) has the desired property. By the Enumeration Theorem and the Parametrization Theorem. A(x.  2y  1 if x > 0 and y = 0 . y) ≃   (2)  (2) · · ϕe (x − 1.y) = ϕ(1) ◦ ϕy . 2.or in other words  if x = 0 . we can find an index e such that  if x = 0 . find a primitive recursive function g(w) such that ϕb(w) (z) ≃ ϕ(w)1 (z) + ϕ(w)2 (z) g for all w. By the Enumeration Theorem and the Parametrization Theorem. Then f (x.4.  2y   (2) 1 if x > 0 and y = 0 . find a primitive recursive function f (w) such that ϕb (1) (z) f (w) (1) ≃ ϕ(w)1 (ϕ(w)2 (z)) (1) (1) for all w.

(w)1 . .8. and not necessarily a permutation? Solution. Π0 . . We define Σ0 and Π0 to be 0 0 the class of primitive recursive predicates. In particular we have (1) (1) ϕe (0) ≃ f (e). we n−1 define Π0 to be the class of predicates that can be written in the form n P (x1 . . Exercise 1. Find a primitive recursive function f (x) such that for all x. Σ0 . . xk . z. . . . by the Parametrization Theorem.6. Therefore. xk ) ≡ ∃y R(x1 . (1) 1. . Definition 1. y) ≃ least w such that (State(x. if ϕx (1) permutation of N.8.3. n−1 41 . Π0 . z. 1. (1) (1) is a 2. and f (e) > e. y) = h(3x 5y ) has the desired property. we define Σ0 to be the n class of k-place predicates P ⊆ Nk (for any k ≥ 1) such that P can be written in the form P (x1 . n))0 · (State((w)2 . z. y. . let f be a 1-place primitive recursive (1) function such that ϕf (x) (y) = x for all x. .5. . . xk . By construction. What happens if ϕx is assumed only to be partial and one-to-one. ϕf (e) (0) = e. let e be such that ϕe (y) ≃ f (e) for all y.1 (The Arithmetical Hierarchy). This works even if ϕx is only assumed to be partial and one-to-one. . . For n ≥ 1. By the Recursion (1) Theorem. y))1 ≃ z. . y) where R is a k+1-place predicate belonging to the class Π0 . . y) where R belongs to the class Σ0 . y))1 . . (w)2 ))2 = y. (1) (1) Exercise 1. n))0 = 0 for all w. (w)2 ))0 = 0 and (State(x. Consider the partial recursive function θ(x. Find m and n such that m = n and ϕm (0) = n and ϕn (0) = m. Solution. . . . (w)1 .9 The Arithmetical Hierarchy In this section we study some important classes of number-theoretic predicates: Σ0 . xk ) ≡ ∀y R(x1 . By the Parametrization Theorem. Similarly. Then h(x. So take m = e and n = f (e). By the Parametrization Theorem. The construction of f in the proof of the Parametrization Theorem shows that f (x) > x for all x.9. These classes collectively are known as the arithmetical 1 1 2 2 hierarchy. if (1) ϕx (z) ≃ y then (θ(x. find a primitive recursive function h(w) such that ϕb (1) (z) h(w) ≃ least n such that (State((w)1 . (1) let f (x) be a primitive recursive function such that ϕf (x) (y) ≃ (θ(x. then ϕf (x) is the inverse permutation.

. . xk . y1 . x1 .For example. xk . Conversely. n n 3.9. . xk ) ≡ ∃y R(x1 . y3 ) where R is primitive recursive. Theorem 1. . n 6. .9. .3. Proof. . 42 . . . xk ) ≃ least y such that R(x1 . y2 . . xk . xk . P ⊆ Nk is Σ1 if and only if ≡ ∃n (State(e.2. . A predicate P belongs to Σ0 (respectively Π0 ) if and only if ¬ P belongs n n to Π0 (respectively Σ0 ). y1 . . Straightforward. . . xk . The classes Σ0 and Π0 are closed under bounded quantification. . then we have 1 P (x1 . n))0 = 0 . . . y) where R is primitive recursive. then letting e be an index of ψ. . Theorem 1. A k-place predicate P ⊆ Nk belongs to the class Σ0 if and 1 P only if P = dom(ψ) for some k-place partial recursive function ψ : Nk −→ N. n+1 n n 2. P (x1 . We have: 0 1. . y) where R ⊆ Nk+1 is recursive (not only primitive recursive).4. . . if P = dom(ψ). . . . . . we have as in the proof of the Enumeration Theorem P (x1 . . the class Π0 is closed under universal quantification. . n n Proof. For n ≥ 1. . 1 0 Corollary 1. . Similarly.9. y) holds . The classes Σ0 and Π0 are included in the classes Σ0 and Πn+1 . . . xk ) hence P is Σ0 . n 5. . If P is Σ0 . xk . P belongs to the class Π0 if and only 3 if it can be written in the form ∀y1 ∃y2 ∀y3 R(x1 . . . the class Σ0 is closed under existential quantification. n n 4. . . . a predicate P (x1 . The classes Σ0 and Π0 are closed under conjunction and disjunction. and clearly ψ is partial recursive. For n ≥ 1. y3 ) where R is primitive recursive. . . xk ) belongs to the class Σ0 if and only if 3 it can be written in the form ∃y1 ∀y2 ∃y3 R(x1 . . hence P = dom(ψ) where ψ(x1 . y2 . xk ) ≡ ∃y R(x1 . . .

3.6. . . The set A is recursive if and only if the function πA is recursive. .8. . Let A be an infinite subset of N. the following are pairwise equivalent: 0 1. . 1 2. P 43 . Show that ψ is partial recursive if and only if the graph of ψ is Σ0 . xk .e. 1. .9. y)}. Lemma 1. . .9. .9. Show that. the graph of ψ is the (k+1)place predicate Gψ = { x1 . . . we n n may replace “primitive recursive” by “recursive”. Theorem 1. .. It follows that. For A ⊆ N. A = ∅ or A = rng(f ) for some total recursive function f : N → N. .Remark 1. .9. . i. If ψ is a k-place partial function. Since the class of recursive predicates is closed under bounded quantification.5. . A = rng(ψ) for some partial recursive function ψ : N −→ N. Proof.7. . . A = dom(ψ) for some partial recursive function ψ : N −→ N. Conversely. in the definition of Σ0 and Π0 for n ≥ 1.9. xk ) which uniformizes P . for every (k+1)-place Σ0 predicate P (x1 . The principal function of A is the one-to-one function πA : N → N that enumerates the elements of A in increasing order. y) there exists 1 a k-place partial recursive function ψ(x1 .9. . A is Σ1 . it follows that A is recursive. . xk . xk . Let A be an infinite subset of N. Definition 1. y | ψ(x1 . Exercises 1. . . 5. . Gψ ⊆ P and dom(ψ) = { x1 . P 2. xk | ∃y P (x1 . A is finite or A = rng(f ) for some one-to-one total recursive function f : N → N. 4. xk ) ≃ y}. if πA is recursive then we have y y∈A if and only if x=0 πA (x) = y . If A is recursive then the functions νA and πA defined by νA (x) = least y such that y ≥ x and y ∈ A πA (0) = πA (x + 1) = νA (0) νA (πA (x) + 1) are obviously recursive. .

and the fact that 3 implies 1 is proved similarly. .3 that P is ∆0 . .9. y) ∨ ¬ R2 (x1 . . .13. For each n ∈ N. Then B is an infinite primitive recursive set. . . . . y) ∧ ¬ R(x. = least y such that R1 (x1 . . .12. . Exercises 1. xk . suppose that A ⊆ N is infinite and Σ1 . . . . .9. . so by Lemma 1. If P is recursive. . .9. xk ) ≡ ∃y R1 (x1 . xk .Proof. say x ∈ A ≡ ∃yR(x. Proof. . it follows easily by Theorem 1. 1.8. .9. y) ≡ ∀y R2 (x1 . Show that every nonempty recursively enumerable set is the range of a primitive recursive function. . y) where R is primitive recursive. . z) z=0 . then we have 1 P (x1 . Define a total recursive function f by f (x1 . y) . The equivalence of 1 and 2 is a special case of Theorem 1. . xk ) Then we have P (x1 . . xk . .9.10. It is easy to see that 5 implies 4 and 4 0 implies 3. Con1 versely.3. This completes the proof. Definition 1. the class ∆0 is defined to be the intersection n of the classes Σ0 and Π0 . . xk ) hence P is recursive. . xk .9. . . we see that f is a one-to-one recursive function and A = rng(f ). Put y−1 B = 2x 3y R(x. Putting f (n) = (πB (n))0 . . Remark 1. . 44 .9.9. y) where R1 and R2 are primitive recursive. . xk . To see that 1 implies 5. . πB is a one-toone recursive function. . . A k-place predicate P ⊆ Nk belongs to the class ∆0 if and 1 only if P is recursive.11. Find an infinite primitive recursive set that is not the range of a one-to-one primitive recursive function. . A set A satisfying the conditions of Theorem 1. 2. n n Theorem 1.9 is sometimes called a recursively enumerable set. ≡ R1 (x1 . if P is ∆0 . xk )) . f (x1 .

First assume that χP is limit-recursive. z) where R1 and R0 are primitive recursive predicates. . Thus χP is limit-recursive. xk ) = lim g(x1 . . x) is again a primitive recursive function. x) = the least y < n such that either ∀z < n R1 (x. . U is a k+1-place predicate belonging to the class Σ0 . g(n. Theorem 1. y. . for any k-place predicate P belonging to the class Σ0 . x) is a primitive recursive function. x) = 0 otherwise. y. . 2 ¬ P (x) ≡ ∃m ∀n (n ≥ m ⇒ f (n. where f (n. if such a y exists. . . xk ∈ N. . . and h(n. there exists an e ∈ N n such that P (x1 . . y.9.15 (Universal Σ0 Predicate). . . y.k with the following properties: 1. x) exists. and n 2. . x) = 0) . xk . x) = 1) and so P is For the converse. . x) n for all x. xk ) for all x1 . g(x) = limn g(n. y. Now define h(n. z) or both. For each n ≥ 1 and k ≥ 1. Then we have P (x) ≡ ∃m ∀n (n ≥ m ⇒ f (n. . y. x1 . and for all x. say 2 P (x) ≡ ∃y ∀z R1 (x.14. we can n find a predicate U = Un. Thus h(n. x) is a recursive function. Using the bounded least number operator. x) = n otherwise. and ¬ P (x) implies h(x) = 0. x) exists and is equal to the least y such that ∀z R1 (x. A function f : Nk → N is said to be limit-recursive if there exists a recursive function g : Nk+1 → N such that f (x1 . z). . . . . say χP (x) = lim f (n. z) or ∀z < n R0 (x. xk . xk . and it is easy to see that.9. h(x) = limn h(n. x) = 1 if ∀z < n R1 (x. y) y for all x1 . z) or ∀z R0 (x. 45 ∆0 . . . xk ) ≡ U (e. x). z) and ¬ P (x) ≡ ∃y ∀z R0 (x. Show that a predicate P is ∆0 if and only if its charac2 teristic function χP is limit-recursive. . For simplicity. for all x. . Thus g(n. assume that P is ∆0 . . Solution. and g(n. Moreover P (x) implies h(x) = 1. .Exercise 1. . define g(n. . z). This completes the proof. . let x be an abbreviation for x1 .

x1 . If B belongs to Π0 . The Π0 part n n n follows easily by taking complements. . . For each n ≥ 1 there exists a complete Σ0 set. y1 . x)} is n n not Π0 .16. y) . n n 0 ∆0 ⊆ Πn . xk . y3 ) is Σ0 .Proof. . y3 ) is recursive. y3 ) where R ⊆ N4 is a primitive recursive predicate. then so does A. y2 . a complete Σ0 set is not n n 0 Π0 . y3 ) . x)} is Σ0 complete. .k+1 (e. n n Proof. y2 . .k (e. . except for n = 0.9. . xk ) ≡ ∃y ¬ Un−1. B belongs to the class Σ0 . n Theorem 1. Lemma 1. By Theorem 1. y1 . xk ) ↓ ≡ ∃s (State(e. This completes the proof. A is reducible to B. Corollary 1. Given n ≥ 1. It follows that 1 the predicate ∃y3 R(f (x).3. The proof is by induction on n. . If A is reducible to B via the recursive function f .9. This is a straightforward consequence of the Enumeration Theorem. . . and n 2. 1 3 Definition 1. .9. x) be a universal Σ0 predicate. n The notion of complete Π0 set is defined similarly. .16 to show that there exists a Σ0 set which n n is not Π0 . then so does A. .15 let U (e. Assume n ≥ 1. B ⊆ N and assume that A is reducible to B. A simple diagonal argument shows that the Σ0 set {x | U (x. To show that a Σ0 complete set can n n never be Π0 . Note that the predicate R(f (x). For n > 1 we have Un. . For all n ∈ N we have ∆0 ⊆ Σ0 . Then we have 3 x∈B if and only if ∃y1 ∀y2 ∃y3 R(x. y1 . For n = 1 we have U1. a set B ⊆ N is said to be complete Σ0 if n 1. 46 .19. for any set A ⊆ N belonging to the class Σ0 . Obviously n the set {2e 3x | U (e. hence ∆0 .18. x1 . n 0 Σ0 ∪ Πn ⊆ ∆0 n n+1 and. . Suppose for example that B belongs to Σ0 . xk . . This completes the proof of the Σ0 part of the theorem. For each n n ≥ 1 there exists a complete Π0 set. all of these inclusions are proper. If B belongs to Σ0 . it suffices by Lemma 1.9. .9. y1 . y2 . then we have x∈A if and only if if and only if f (x) ∈ B ∃y1 ∀y2 ∃y3 R(f (x). and a complete Π0 set is not Σn . . xk ) ≡ ϕ(k) (x1 . Hence A is Σ0 . y2 . Let A.k (e. . s))0 = 0 e in view of Theorem 1.9.9. For each n ≥ 1.17. . x1 . x1 . n n Proof.

7. n+1 n n Exercises 1. K = {x ∈ N | ϕx (x) ↓}. Then the complement B = N\A n is complete Π0 . T = {x ∈ N | ϕx is total}. determine the reducibility relations among H.16 and Theorem 1. E. S = {x ∈ N | dom(ϕx ) is infinite}. What reducibility and non-reducibility relations exist among the following sets? Note that H. Show that the set is a complete Π0 set.18. let A be a complete Σ0 set. together with Lemma 1. and E is Π0 1 2 1 complete.9.Proof. S and T are Π0 complete.9. Show that the sets (1) H = {x | ϕx (0) is defined} and are complete Σ0 sets.21.9. H = {x ∈ N | ϕx (0) ↓}. T . and S are index sets. These completeness facts.9. Given n ≥ 1.11.2 and Lemma 1. K.16 that the set A⊕B = {2x | x ∈ A} ∪ {2x + 1 | x ∈ B} belongs to ∆0 \ (Σ0 ∪ Π0 ). Prove your answers.9. T . and E. Clearly A belongs to Σ0 \ ∆0 and B belongs to Π0 \ ∆0 . 2 T = {x | ϕ(1) is total} x K = {x | ϕ(1) (x) is defined} x Exercise 1. E = {x ∈ N | ϕx is the empty function}. S. 1 2.9. Hint: Using the Parametrization Theorem as in the proof of Theorem 1. show that H and K are Σ0 complete.20. n n n n n Moreover it follows by Theorem 1. 1. (1) (1) (1) (1) (1) 47 .

For the benefit of the reader who has not previously studied mathematical logic. Each of these variables is also a term of our language. + and · denote the 2-place operations of addition and multiplication on N. =). . . . we may employ the usual abbreviations. we shall concentrate on that case. +. +. (x + y) · z + x . . 0. +. 0. . . 1. Our language contains infinitely many variables x0 . In addition the symbols 0 and 1 are terms. xn . Formulas. . ·. . 0.} is the set of non-negative integers.Chapter 2 Undecidability of Arithmetic We are going to show that arithmetic is undecidable. . . z. which are usually denoted by letters such as x. When writing terms. This means that there is no algorithm to decide whether a given sentence in the language of arithmetic is true or false. For instance 1 + 1 + 1 = 3 and x · x · x = x3 . 1. Thus a term is essentially a polynomial in several variables with non-negative integer coefficients. 48 . we shall now review the concept of a sentence being true in a structure. 1. y. . . x+1. 1. and = denotes the 2-place relation of equality between elements of N. Examples of terms are 1+ 1+1.1 Terms. 2. . Since we are only interested in the particular structure (N. and Sentences By arithmetic we mean the set of sentences which are true in the structure (N. . Other terms are built up using the 2-place operation symbols + and ·. ·. =). x1 . =). ·. It is first necessary to define a language appropriate for the structure (N. Here N = {0. 2.

is to be interpreted as “there exists a non-negative integer z in N such that . . ∃ (for all. . Formulas are built from atomic formulas by using the Boolean propositional connectives ∧. . not) and the quantifiers ∀. An example of a formula is ∃z (x + z + 1 = y) (2. . there exists). ak for the free occurrences of the variables x1 . . x. . y and z are to be interpreted as variables ranging over the set N. .” and ∃x means “for some x in N” or “there exists x in N such that . ak ∈ N. .An atomic formula is a formula of the form t1 = t2 where t1 and t2 are terms.1) In this formula. ¬ (and. . This idea of introducing new relation symbols as abbreviations for formulas allows us to expand our language indefinitely.1) are x and y. . 1. . =). 0. Let F (x1 . ·. . . Then for any non-negative integers a1 . let F (x. . . For example. xk .2) This formula expresses the assertion that x is a prime number. Thus ∀x means “for all x in N.1). . (2. .2) the only free variable is x. . Examples of atomic formulas are x+1 = y. +. Examples of sentences are ∀x (∃y (x = 2y) ∨ ∃y (x = 2y + 1)) and ∀x ∃y (x = y + 1) . 0. . a free variable is a variable which is not acted on by any quantifier in that formula. . . we know what it means for a sentence of our language to be true or false in the structure (N. while in (2. the free variables of (2. ∨. or. Another example of a formula is x > 1 ∧ ¬ ∃y ∃z (y > 1 ∧ z > 1 ∧ x = y · z) . . .3) is true (because every element of N is either even or odd) and (2. (2. Thus we might choose to abbreviate (2. . .” Thus the formula (2. . xk ) be a formula whose free variables are x1 . 1.2) by some expression such as Prime(x) or “x is prime. .1) expresses the assertion that x is less than y. we can form the sentence F (a1 . ·. please bear in mind that the quantifiers ∀ and ∃ range over the set of nonnegative integers. .4) is false (because not every element of N is the successor of some other element of N). For example. . x = 3y 2 + 1 . From now on we shall write x < y as an abbreviation for (2. ak ) which is obtained by substituting (“plugging in”) the constants a1 . A sentence is a formula with no free variables. +.” In any particular formula. N. xk . y) be the formula ∃z (x2 + z = y) 49 (2. . .” With this understanding. =).4) When interpreting formulas or sentences in the structure (N. For example (2. Similarly the expression ∃z . .3) .

.2. the 1-place function λx [ 2x ] is arithmetically definable. Write a sentence expressing Goldbach’s Conjecture: Every even number is the sum of two prime numbers. .1. n) is true.2 Arithmetical Definability We now present a key definition. and z. xk+1 or some other set of k + 1 distinct variables. xk ) ] is said to be arithmetically definable if there exists a formula F (x1 . This completes our review of the concept of a sentence being true or false in the structure (N. . xk . Show that the function λxy [ least common multiple of x and y ] is arithmetically definable. . xk+1 ) with free variables x1 . ·. . 0.) For example. A k-place partial function λx1 · · · xk [ ψ(x1 .) Exercise 2. . mk . xk . x) ] and λxy [ Remainder(y. +. 2. .1 (Arithmetical Definability). . . . . n ∈ N.. . note that the 2-place functions λxy [ Quotient(y.2. 20) is false.with free variables x and y.) As another example. . mk ) ≃ n if and only if F (m1 . . . ψ(m1 . . by the formulas ∃u ∃v (y = u · x + v ∧ v < x ∧ z = u) and ∃u ∃v (y = u · x + v ∧ v < x ∧ z = v) respectively. xk . For instance F (5. mk . 1.2. =). F (m. y. . (Here we are using a formula with two free variables x and y. x) ] (the quotient and remainder of y on division by x) are arithmetically definable. . 50 . . such that for all m1 . n) is a sentence which expresses the assertion that m2 is less than or equal to n.. . Then for any particular non-negative integers m and n. (Here we are using formulas with free variables x.1. . (Of course it doesn’t matter whether we use the variables x1 . 40) is true and F (5. by the formula y = 2x. xk+1 . Definition 2. Exercise 2. . . . . .

x is the largest prime number less than y.22 below). .2. ak . the arithmetically definable predicates may be characterized as the smallest class of number-theoretic predicates which contains all strongly Diophantine predicates and is closed under union. .5. xk ]. . . the projection of Q is given as π(Q) = { a1 . y) = z. ak ) = 0}. 4. 1. Lemma 2. we shall characterize the class of arithmetically definable predicates in terms of the arithmetical hierarchy (Theorem 2. 2.2. .2. For 1 ≤ i ≤ k. .23 below). Then. The successor function λx [ x + 1 ] is defined by the formula y = x+1. .3. y) = z. . GCD(x. . ·.Remark 2. This completes the proof. ∧ xk = xk ∧ y = xi . Remainder(x. . . xk [ xi ] is defined by the formula x1 = x1 ∧ . 5. . +. The constant zero function λx [ 0 ] is defined by the formula x = x ∧ y = 0. 3. xk ) ∈ Z[x1 . ak ∈ Nk | a1 . . f (x1 . ak+1 ∈ Q for some ak+1 ∈ N} . . .2. 6. A more “mathematical” characterization of arithmetical definability.21 below. . . Exercise 2. . Proof. Clearly each arithmetically definable predicate is obtained by applying these operations only a finite number of times. . Quotient(x. by exhibiting formulas which define them over the structure (N. such that P = { a1 . In addition. is that all recursive functions and predicates are arithmetically definable. Let us say that a predicate P ⊆ Nk is strongly Diophantine if there exists a polynomial with integer coefficients. . 1. From this it will follow easily that there exists an arithmetically definable predicate which is not recursive (Corollary 2. Theorem 2.4. . The principal result of this section. x is the product of all the prime numbers less than y. 51 . y) = z.6. the projection function λx1 . . y) = z. =). . For Q ⊆ Nk+1 . Show that the following number-theoretic predicates are arithmetically definable.2. and projection. Remark 2. . LCM(x. . 0. . . not involving formulas. may be given as follows.2. complementation. . ak ∈ Nk | f (a1 .2. The initial functions are arithmetically definable.

. . . . y1 ) ∧ . . . . A k-place predicate P (x1 . we need some definitions and lemmas from number theory. . . y) ∧ ¬ ∃z (z < y ∧ R(x1 . . . . Gm (x1 . . . . . xk ). y1 . y) holds. . . . . . = h(g1 (x1 . .7. xk )] is arithmetically definable by the formula R(x1 . . . yk ) . yk ) = g(y1 . . . y1 . . In order to prove Lemma 2. .2. . . . . = h(x. . . . . Lemma 2. . . . y) . .2. xk ) is said to be arithmetically definable if it is definable over the structure (N. . +. . then so is the function f (x1 . y) . . xk . . . ym . xk ). . z. . . . . Our λx1 . This proves the lemma. . . . . . . . 0. . . If the k+1-place predicate R(x1 . . . . . . . . . . xk )) Proof. . . . z)) which we may abbreviate as F (x1 . xk . . . xk . . . gm (x1 . ym ) be defined by the formulas G1 (x1 . ym ) are arithmetically definable. y) is arithmetically definable. z)) which we may abbreviate as F (x1 . . ym . . yk ) obtained by primitive recursion: f (0. A set of positive integers is said to be pairwise relatively prime if any two distinct integers in the set are relatively prime. xk ) is defined by the formula ∃y1 · · · ∃ym (G1 (x1 . y). . . =) by a formula with k free variables x1 . gm (x1 . yk ) f (x + 1. xk ) and h(y1 . . . . . . h(y1 . xk . . It remains to prove: Lemma 2. . xk ) obtained by composition. .8. . . . . xk ) ≃ least y such that R(x1 . . z) respectively. then so is the total k+1-place function f (x. xk . . 1. This proves the lemma. .2. . . Let g1 (x1 . . . . .2. yk ). . . z). . . . . . xk . . xk . . . xk [ψ(x1 . . . y1 . If the total k-place function g(y1 . xk ). . . Proof. . . xk .Lemma 2. . . . yk ) are arithmetically definable. . . . . Then f (x1 . . . ym ) ∧ H(y1 . . If the functions g1 (x1 . y1 .9. . . . . ∧ Gm (x1 . . . . . . xk . . . . . f (x. then so is the k-place partial function ψ(x1 . y1 . . xk . . . .9. gm (x1 . . yk )) . . . . Two positive integers m and n are said to be relatively prime if they have no common factor greater than 1. . . xk ). 52 . . H(y1 . . . ·. . . xk . yk ) and the total k+2place function h(x. . . y1 .

Let m1 .Lemma 2. let r and s be non-negative integers less than m. a contradiction. . Hence p is a factor of j − i. define the remainder sequence of r to be the sequence r1 . 4a + 1. 25. . Taking a = 6 we see that 7. On the other hand p is a factor of (ja + 1) − (ia + 1) = (j − i)a.2. 2a + 1 . we see that for any nonnegative integers r1 < 7. 2a + 1. 37. . Proof. . r3 . But j − i is less than k. . 53 . . . . We claim that a + 1. Let p be a prime number which is a factor of both ia + 1 and ja + 1. To see this. 19. we must have r = s. .10. mk . . since ri = Remainder(r. . rk where ri = Remainder(r. r2 < 13. mk be a set of k distinct positive integers which are pairwise relatively prime. 1 ≤ i ≤ k. hence p is less than k. Let a be any positive integer which is divisible by all of the prime numbers which are less than k. so we can take a to be any multiple of 6. Given a positive integer k. r5 ). Then the integers a + 1. there must be a non-negative integer r less than 7 · 13 · 19 · 25 · 31 such that Remainder(r. 3001 are pairwise relatively prime. r5 < 31.2. 61 are pairwise relatively prime. We may view the integer r as a “code” for the sequence (r1 . .12 (Chinese Remainder Theorem). Continuing the previous example. Suppose not. r2 . Let i and j be such that 1 ≤ i < j ≤ k and ia + 1 and ja+ 1 are not relatively prime. Define a possible remainder sequence to be any sequence r1 . 49. .2. Taking a = 12 we see that 13. we see that any possible remainder sequence must actually occur as the remainder sequence associated with some r in the range 0 ≤ r < m. . Example 2. 31 are paiwise relatively prime. we can find a non-negative integer r such that Remainder(r. Then for any given non-negative integers ri < mi . . we can find infinitely many positive integers a such that the k integers in the set a + 1 . We claim that any two distinct non-negative integers less than m have distinct remainder sequences. the primes less than k are 2 and 3. mk . . 1801. The lemma follows immediately. mi ) = ri for all i. And taking a = 600 we see that 601. . . . Let m be the product of m1 . 2a + 1. Hence p is greater than or equal to k. The “decoding” is accomplished by means of the key integer 6. . it follows that r − s is divisible by m. 5a + 1 will be pairwise relatively prime. Then p cannot be a factor of m. . r4 . rk with 0 ≤ ri < mi for all i. ka + 1 are pairwise relatively prime. . . If k = 5. r3 < 19. The number of possible remainder sequences is exactly m. ka + 1 are pairwise relatively prime. 1201. 2401. . For any non-negative integer r less than m. . mk are pairwise relatively prime. Lemma 2. . . . 6i + 1).11. Remainder(r. .13. Since r and s are both less than m. 3a + 1. Proof. then r − s is divisible by each of m1 . 31) = r5 . 25. . . . 13. Since m1 . Example 2. . . . This proves the claim. If r and s have the same remainder sequence. .2. r4 < 25. 7) = r1 . mi ). From this and the claim.

z. and furthermore a + 1. Our language allows us to say things like “there exists a non-negative integer r such that . w) which says. For any non-negative integers r. 0)) holds. a. If we could do this. . . . .” but it does not allow us to directly say things like “there exists a sequence of non-negative integers r0 .” Namely. . By Lemma 2. w) respectively. . .2. . Remainder(r.” The way to overcome this difficulty is to use the β-function. The only difficulty is that our language is not powerful enough to talk directly about finite sequences of variable length in the required way. r1 . Lemma 2. . yk .2. a. β(r. . . . a.2. . .14.10 (replacing k by k + 1). 1) = 19. . (k + 1) · a + 1) = rk . Let the total functions g(y1 . y1 . and finally rx = w. Proof of Lemma 2. . . w) would define the function f (x. rx such that G(y1 . . . r1 . x) = w. . y1 . Note that the β-function o is arithmetically definable. a · (i + 1) + 1) . .12 we can find a non-negative integer r such that Remainder(r. and H(i. . y1 . a. β(r. . . Proof. . . . . . i + 1)) holds for all i < x. 2) = 30. . We make this precise in the following lemma. β(r. . . i) = Remainder(r. . . rk . β(r. . r0 ) holds.2. yk .2. r1 < 2a + 1. . ri . yk . . . 4) = 51. . rx (of variable length. . z. . . y1 . . a. The significance of the β-function is that it can be used to encode an arbitrary sequence of non-negative integers r0 . 3) = 37. . yk ). i). . (k + 1) · a + 1 are pairwise relatively prime. . . r1 .15. β(r. . informally. . Hint: First find an appropriate a by hand.9. . rk < (k + 1) · a + 1. This would prove the lemma. . . k) = rk . β(r. β(r. y1 . . . . We are now ready to prove Lemma 2. yk . a. . . Remainder(r. Find a pair of numbers r.Definition 2. a. a. we can find a positive integer a such that r0 < a + 1. r1 .2. r1 . a such that β(r. we can find a pair of non-negative integers r and a such that β(r.2. Then write a small computer program to find r by brute force. . By Lemma 2. 2a + 1. . there exist r and a such that G(y1 . yk . We wish to write down a formula F (x. y1 .16. . . . Instead of saying “there exists a finite sequence r0 . y1 . . we write down a formula F (x. . . .2. β(r. . w) and H(x. via the β-function. yk . w) which would say that there exists a finite sequence r0 . . a. i. then clearly F (x. 0) = r0 . . . yk . x) such that . . . Exercise 2. . 2a + 1) = r1 .” we can say “there exists a pair of non-negative integers r and a which encode the required sequence. . 0) = 11. . y1 . yk ) and h(x. 1) = r1 . rk . . and finally β(r. . yk ) be defined by the formulas G(y1 .9. .15 it is clear that this 54 . . β(r. . a. . Then by Lemma 2. yk . This proves the lemma. a. and H(i. a + 1) = r0 . a. . we define β(r. . rk by means of two nonnegative integers r and a. a. a. ri+1 ) holds for all i < x. . yk . This 3-place function is known as G¨del’s β-function. . . . Given a finite sequence of non-negative integers r0 . . . .

2. Proof. i. .formula defines the function f (x.e. w) to be the formula ∃r ∃a ( ∃u (B(r. . v) ∧ H(i.2.2. . Theorem 2. let B(r. 0. 1. v)) ) ) . . y1 .2. . w) be a formula which defines the β-function. y1 . y. when interpreted over the natural number system N.7.9 say that the class of arithmetically definable functions is closed under composition. and the least-number operator. 1.2.2.18.2.17. x is the sum of all the prime numbers less than y. The exponential function λxy [ y x ] is arithmetically definable by the formula ∀i ( i ≥ x ∨ ∃u ∃v (B(r. Lemmas 2. k k (n) = f (f (n)). f (n) = n and f Write a formula F (x. w) ∧ This completes the proof of Lemma 2.20. y1 . primitive recursion. i. ∃r ∃a ( B(r. u) ∧ B(r. a. Consider the function f : N → N defined by f (n) = For each k ∈ N let 0 k+1 ∀i ( i ≥ x ∨ ∃u ∃v (B(r. Show that the function λx [ the xth Fibonacci number ] is arithmetically definable. ·. a. z) in the language +. respectively. w) ∧ n/2 3n + 1 if n is even. Recall that the partial recursive functions have been characterized as the smallest class of functions which includes the initial functions and is closed under composition.. Every partial recursive function is arithmetically definable. yk . the least-number operator. . a.6 says that the class of arithmetically definable functions includes the initial functions. u) ∧ B(r. < which. i. xy = z. Exercise 2. i + 1.21. . =. meaning that y x = w. . Lemma 2. . u. u)) ∧ B(r. u) ∧ G(y1 . a. Exercise 2. a. i + 1. . a. v) ∧ v = u · y) ) ) which we may abbreviate as Exp(x. Which of the following number-theoretic predicates are arithmetically definable? Prove your answers. a. y. x! = y. 3.2.19. x. fk = f ◦ · · · ◦ f . a.8.2. . 0. We may then take F (x. . and 2. Example 2. yk . yk . . w). Exercise 2. a.9. if n is odd. 2. i. 1) ∧ B(r.2. . and primitive recursion. defines the 3-place predicate f x (y) = z. Formally. It follows that the arithmetically definable functions include the partial recursive functions. . . 0. 55 . yk ).2. x.

Then K is defined by the formula ∃zF (x. . then A is arithmetically definable.2.1) in terms of the arithmetical hierarchy (Definition 1. n n For the converse. Theorem 2. . P belongs to the class Σn for some n ∈ N. Proof. For each n ≥ 1 let Cn be a set which is Σ0 complete. Exercise 2. By the Enumeration Theorem. P belongs to the arithmetical hierarchy. z) be a formula which defines this function. . put Σ0 ∞ = n∈N 0 Σn = n∈N Π0 .Corollary 2. Remark 2. Let A and B be subsets of N. .24.2.1). b1 .26 (Hilbert’s Tenth Problem). There exists a set K ⊆ N which is arithmetically definable but not recursive. Let us say that P ⊆ Nk is Diophantine if P = π l (Q) for some l ≥ 0 and some strongly Diophantine Q ⊆ Nk+l . Prove that B is not arithmetically definable.2.2.22. Let F (x. ak . z).9. y. . . P ⊆ Nk . ∨.2. n By Theorem 1. (1) the 2-place partial function λxy[ϕx (y)] is partial recursive. 0 2. . More generally. by Theo(1) rem 2. recall our discussion of the arithmetical hierarchy in Chapter 1.2. . Recall that K = {x ∈ N | ϕ(1) (x) is convergent} x is our basic example of a non-recursive set. Let P be a k-place predicate.2. . .25. . A refinement of Theorem 2. This completes the proof. Proof.2.9. . . . Section 1.21 implies that every predicate in the class Σ0 (= Π0 ) is 0 0 arithmetically definable. From this it follows that every arithmetically definable predicate P belongs to the class Σ0 . Consider n the set B = {2n 3x | x ∈ Cn }. Hence. The next theorem characterizes arithmetically definability (Definition 2. ak ∈ Nk | ∃ b1 . ¬ and ∞ universal and existential quantification ∀ and ∃.e. . Exercise 2. this is proved by ∞ induction on the number of symbols in a defining formula for P . the class Σ0 is closed under Boolean connectives ∧. Thus P = { a1 . From this it is straightforward to prove by induction on n ∈ N that every predicate in the class Σ0 ∪ Π0 is arithmetically definable.2.21 λxy[ϕx (y)] is arithmetically definable. i. P is arithmetically definable. bl ∈ Nl (f (a1 . bl ) = 0)} 56 . .2.. The following are equivalent: 1. Theorem 2.23 due to Matiyasevich 1967 is as follows.9. x.23. Prove that if A is reducible to B and B is arithmetically definable.

1 A corollary of Matiyasevich’s Theorem is that. . Hilbert’s Tenth Problem is unsolvable. Spring 2005. . . but l = 8 is an open question. b. for example. . 1. . A full exposition of Hilbert’s Tenth Problem and the proof of Matiyasevich’s Theorem is in my lecture notes for Math 574. . we shall first assign G¨del o o numbers to terms. yl ) is a polynomial with integer coefficients. In our formulation of Matiyasevich’s Theorem. . . 2. it follows that there is no such algorithm. 2. the sets K and H are Diophantine. . . Hilbert’s Tenth Problem. . . . y1 . while Z = {. reads as follows: To find an algorithm which allows us.edu/simpson/notes/. 1.math. xk . In other words. Thus.psu. . we have spoken of solutions in N. at http://www. What happens if we replace “solution in N” by “solution in Z”? Hint: Use the following well-known theorem of Lagrange: For each n ∈ N there exist a.27. −2.2. 2. . d ∈ N such that n = a2 + b2 + c2 + d2 . Here it is known that one can take l = 9. . . . .where f (x1 .} = the integers. we shall define a unique positive integer #(F ). . Exercise 2. as stated in Hilbert’s famous 1900 problem list. −1.} = the natural numbers. From Matiyasevich’s Theorem plus the unsolvability of the Halting Problem. The number o #(F ) will serve as a “code” for the formula F . . given a polynomial equation in several variables with integer coefficients.3 G¨del Numbers of Formulas o For each formula F in the language of arithmetic. Matiyasevich’s Theorem states that P is Diophantine if and only if P is Σ0 . x1 . xl ) = 0 has a solution in N is nonrecursive. Before assigning G¨del numbers to formulas. such that the set of a ∈ N for which f (a. we can find a polynomial f (z. Recall that N = {0. which will be called the G¨del number of F . the G¨del number of the term 1 + x0 is o #(1 + x0 ) = 33 · 53 · 79 57 . 0. . c. x1 . . . . xl ) with integer coefficients. We define #(0) #(1) #(xi ) #(t1 + t2 ) #(t1 · t2 ) = 1 = 3 = 33 · 5#(t1 ) · 7#(t2 ) = 32 · 5i = 34 · 5#(t1 ) · 7#(t2 ) For example. to decide in a finite number of steps whether or not the equation has a solution in integers.

“arithmetic is undecidable.” or simply. Snt = #(all sentences) . and TrueSnt are subsets of N. We now define the G¨del numbers of formulas: o #(t1 = t2 ) #(F ∧ G) #(F ∨ G) #(¬ F ) #(∀xi F ) #(∃xi F ) = 2 · 5#(t1 ) · 7#(t2 ) = 2 · 33 · 5#(F ) = 2 · 32 · 5#(F ) · 7#(G) = 2 · 34 · 5i · 7#(F ) = 2 · 3 · 5#(F ) · 7#(G) = 2 · 35 · 5i · 7#(F ) For example. the problem of deciding whether a given sentence of the language of arithmetic is true or false is unsolvable. Proof. Proof. A Define f (m) = #(∃x1 (x1 = m ∧ F (x1 ))) . 58 .e.” Theorem 2. let F (x1 ) be a formula with one free variable x which defines A.. = {m ∈ N | F (m) is true} .since #(1) = 3 and #(x0 ) = 9. The sets Fml and Snt are primitive recursive. Given an arithmetically definable set A. Snt. Note that Fml.3.1. and TrueSnt = #(all true sentences) . If F is any collection of formulas.2. #(1) = 3.3. and #(x1 = 1) = 2 · 545 · 73 . we denote by #(F) the collection of all G¨del numbers of formulas in F. Let o Fml = #(all formulas) . This result may be paraphrased as “arithmetical truth is undecidable. The proof is straightforward and we omit it. We are going to prove that the set TrueSnt is nonrecursive. Theorem 2. In other words. the G¨del number of the formula ∃x1 (x1 = 1) is o #(∃x1 (x1 = 1)) = 2 · 35 · 51 · 72·5 45 ·73 since #(x1 ) = 45. Every arithmetically definable set A ⊆ N is reducible to TrueSnt. i.

3. This completes the proof. This is clear since f (m) = 2 · 33 · 72·3·5 where #(x1 = m) = 2 · 545 · 7#(m) . By Corollary 2.2 C is reducible to TrueSnt.2.Thus for all m ∈ N we have that m ∈ A if and only if f (m) ∈ TrueSnt. 59 . Hence by Lemma 1. TrueSnt is not arithmetically definable. #(m + 1) = 33 · 5#(m) · 73 . Proof.18.23 C is arithmetically definable.9. + 1) can be defined primitive recursively by m #(x1 =m) ·7#(F (x1 )) #(0) = 1.3. By the previous theorem K is reducible to TrueSnt.5. This proves the theorem. Suppose that TrueSnt were arithmetically definable. .9. n By Theorem 1. Prove that the set Snt of all G¨del numbers of sentences is o primitive recursive.2. This contradicts that fact n 0 (Theorem 1.4 (Tarski). which may be paraphrased as “arithmetical truth is not arithmetically definable. Then by Theorem 2.9. More generally we have the following theorem.7.3.” Theorem 2. We claim that the function f : N → N is primitive recursive.18) that a complete Σ0 n+1 set can never belong to the class Σn .2. Exercise 2.3. By Theorem 2.23 we would have that TrueSnt belongs to the class Σ0 for some n. and #(m) = #(1 + . Proof. Theorem 2.22 we have an arithmetically definable set K which is not recursive.6. Exercise 2. Thus A is reducible to TrueSnt via f . let C be a complete Σ0 n+1 set. . Hence by Theorem 2.16 C belongs to the class Σ0 .3. Prove that the set Fml of all G¨del numbers of formulas is o primitive recursive. Hence by Lemma 1.3.8 TrueSnt is not recursive. TrueSnt is not recursive.

i. . by contrast. . xk ) ⇔ G(x1 .1. we can find an equivalent quantifier free LOR -formula F ∗ . . where + and · are 2-ary operation symbols. . the theory of the natural number system is undecidable. . .” etc. . . ·R . <R . . the ordered field of real numbers... 1. i. i. the theory of the real number system is decidable. < and = are 2-place predicate symbols.e. Formulas of LOR are interpreted with reference to R. +R . −R . We are going to prove the following theorem. xk and define the same k-place predicate P ⊆ Rk . . .Chapter 3 The Real Number System In Chapter 2 we have shown that TrueSntN is nonrecursive. ·.e. means “there exists x ∈ R such that . xk )) is true in R. 3. ∃x . We consider the LOR -structure R = (R. =) Let LOR be the language of ordered rings. . Two LOR -formulas F and G are said to be equivalent if they have the same free variables x1 . =R ) . An equivalent condition is that the sentence ∀x1 · · · ∀xk (F (x1 . TrueSntR is recursive. 0R . 60 .. 0. . i.. .. i.e. .1 (Quantifier Elimination).1 Quantifier Elimination LOR = (+. − is a 1-ary operation symbol. In this Chapter we shall show that.e. For any LOR -formula F . 1R . and 0 and 1 are constant symbols. . due originally to Tarski: Theorem 3.e. −. . the real number system. <.

Find quantifier-free formulas in the language +. xk ) > 0. Exercise 3. 0.1.1. ∃x (ax2 + bx + c > 0).3. we can find a formula F ∗ (x) in the same language which is equivalent to F (x) over the natural number system N. The only properties of R that will be used in the proof of Theorem 3.1 which we shall present below is due to P. A predicate A on the reals.. D ⊆ Rk is said to be effective if 61 . is said to be effective if it is definable over R by a quantifier free formula. we get semi-algebraic sets. Exercise 3. . . . ·. This is related to Hilbert’s 17th Problem.1.e. .9. −.1. <. . Remark 3. Remark 3.4. That is. In order to present the proof. Definition 3.Example 3.1. = with exactly one free variable x.1.7. 2. = which are equivalent.21 below. The proof of Theorem 3.1. . 3.8.1.1. ∃x (ax4 + bx3 + cx2 + dx + e > 0).1 are: (1) R is a commutative ordered field. .1..1.1. i.2. . Prove your answer.1. If we allow parameters from R. J. . 1. Remark 3. Cohen. ·. The formula ∃x (ax2 + bx + c = 0) is equivalent to the quantifier free formula (a = 0 ∧ b = 0 ∧ c = 0) ∨ (a = 0 ∧ b = 0) ∨ (a = 0 ∧ b2 − 4ac ≥ 0) .5. 1. This notion of effectivity has no importance beyond the proof of Theorem 3. An ordered field with these properties is called a real closed ordered field. where p ∈ Z[x1 . Does there exist a constant c such that the following holds? Given a formula F (x) in the language +. . ∃x (ax3 + bx2 + cx + d > 0). and which contains at most c quantifiers. ( x < y ∧ p(x) < 0 < p(y) ) ⇒ ∃z ( x < z < y ∧ p(z) = 0 ) for any polynomial p(x) ∈ R[x].e. xk ) = 0. A ⊆ Rk . we shall define and study a class of functions called the effective functions. A is a Boolean combination of sets in Rk which are defined by equations and inequations p(x1 . to: 1. Thus “effective = semi-algebraic with parameters from Z”. .6. xk ]. . Definition 3. 0. and (2) R has the intermediate value property for polynomials. i. over the real number system R. See also Corollary 3. A function f : D → R. p(x1 .

. . . If A(y.11. . . is effective. . The composition of effective functions is effective. .1. . we shall build up a library of effective functions. The latter is a Boolean combination of predicates of the form an x y n + . D is effective. xk ). z) is effective.1. . for every effective predicate A(x1 . then clearly A(f (y). z1 . Lemma 3. . . .1. It can be√ shown that the function x is effective. zn ). . . . . . .12. . x · y and −x there is nothing to prove. Lemma 3. y. . .).10. z1 . . Corollary 3.13. .1. The functions x + y. z1 . In order to prove Theorem 3. . . . . . and 2. . For x + y. . The function   1  0 sgn(x) =   −1 62 if x > 0 if x = 0 if x < 0. −x and x/y are effective. Any rational function f (x1 . This is because. . We shall use notations such as x and y to abbreviate sequences of variables such as x1 . xk . Proof. the domain of x is the effective set {x ∈ R | x ≥ 0}.1.1. . then D(g(x)) defines the dom(f g). zn ) ≡ A(x1 . . For x/y. assuming as we may that n is even. note first that the domain is {(x. . z) is effective. ym . . . Consider for instance f (g(x)) where f and g are effective 1-place functions. √ Example 3. If D(y) is a quantifier-free formula defining the dom(f ). . . xk .1.14. . . x > 3 ≡ x > 9. It remains to show that if A(z. + a1 xy n−1 + a0 y n > 0 . xk . Proof. . . .1. zn ) is effective. + a1 x y + a0 > 0 and this is equivalent to the atomic formula an xn + an−1 xn−1 y + . . . hence A(f (g(x)). x · y. . z) is any effective predicate. . xk and y1 . y) | y = 0} which is obviously effective.) is an effective predicate then so is A(x/y. first. Lemma 3. . the predicate B(x1 . √ for instance. f (x1 . and second. . . xk ) ∈ Q(x1 . . . which is therefore effective. . . . xk ) is effective.

1. B(f1 (x). z).16 (Definition by Cases). z)) . . all integers. z)) which is again effective. . sgn(q(f (x))) is an effective function of x and the d + 1 coefficients of q(y).Proof. the predicate C(x. z) be an effective predicate. z)) ∨ · · · ∨ (f (x) = jn ∧ A(jn . Only if: Trivial. z)) ∨ (¬ A(x) ∧ B(f2 (x). jn be the finitely many values. 63 . Proof. z) are effective. If a function f (x) takes only finitely many values. Since A(x) is effective. z) is equivalent to (A(x) ∧ B(f1 (x). For any effective predicate A(y. z) is effective. This is so because C(x. Then A(sgn(x). This proves the lemma. it follows that C(x. Since f1 (x) and f2 (x) and B(y. z) is equivalent to equivalent to (x > 0 ∧ A(1. Proof. We must show that. z) is effective. f1 (x) if A(x). the predicate A(f (x).1. Lemma 3. since for instance sgn(q(f (x))) = 1 ≡ q(f (x)) > 0 ≡ A(f (x)). This proves the lemma. Lemma 3. z) and B(f2 (x).1.17. z) ≡ B(f (x). z)) ∨ (x < 0 ∧ A(−1. . Proof. Let A(y.15. If: Let j1 . for each effective predicate B(y. z) are effective. then f (x) is effective if and only if for each j ∈ Z. . z)) and is therefore effective. If two functions f1 (x) and f2 (x) and a predicate A(x) are effective. More generally we have: Lemma 3. then the function f (x) = is effective. A function f (x) is effective if and only if for every positive integer d ≥ 1 and every polynomial q(y) ∈ R[y] of degree d. f2 (x) if ¬ A(x) where A(y) is the predicate q(y) > 0. z)) ∨ (x = 0 ∧ A(0. Only if: Trivial. z). z) is equivalent to (f (x) = j1 ∧ A(j1 . the predicate f (x) = j is effective. The extension of the previous lemma to more than two cases is obvious.

tm ) . In addition the ti ’s could be roots of p(x).. t1 ) . . . By inductive hypothesis. ξn (a). z) = q(y) ∈ Z[z][y] i.18 (Main Lemma). +∞) So. Thus the number of roots is determined by sgn(p(ti )). For example. we may assume deg(q(x)) < n. . z) with coefficients in Z. Then p(f (x). 1 ≤ i ≤ n − 1 and sgn(p′ (t1 − 1)) and sgn(p′ (tn + 1)). z) can be viewed as a Boolean combination of atomic predicates of the form p(y. This is of degree ≤ n−1. It suffices to show that the predicates p(f (x). the quadratic formula √ −b ± b2 − 4ac x = 2a shows that the roots of ax2 + bx + c are effective functions of a. . sgn(q(ξ)) is effective (as a function of a and the coefficients of q). . m = n − 1. and k = k(a) such that ξ1 (a) < · · · < ξk (a) are all of the real roots of p(x). Replacing q(x) by its remainder on division by p(x). in each of these intervals. for each d ≥ 1 and polynomial q(x) of degree d. . Proof. z) > 0. Note that p(x) is monotone on each of the open intervals (−∞. By assumption sgn(q(f (x))) is an effective function of x and the cofficients of q. In order to show this. (tm . . where deg(r(x)) < n and the coefficients of r(x) are effective functions of the coefficients of p(x) and q(x). (t1 . it suffices to show that. an . Note that A(y. We can therefore use definition by cases to obtain k(a) as an effective function of a. where the ti ’s are effective function of a. t2 ) . It remains to show that the roots themselves are effective functions of a. (tm−1 . z) is effective then A(f (x). for various polynomials p(y. Consider for example a root ξ = ξ(a) in the interval (t1 . Write a = a0 . hence by composition sgn(q(f (x))) is an effective function of x and z. [Details: Long division gives q(x) = p(x) · f (x) + r(x). Let p′ (x) = nan xn−1 + · · · + 2a2 x + a1 be the derivative of p(x). There are n + 1 effective functions ξ1 (a). By the previous lemma. Lemma 3.] 64 . the roots of p′ (x) are among t1 < · · · < tm .If: We must show that if A(y. The next lemma says that the real roots of a polynomial are effective functions of the coefficients. there is at most one root of p(x). z) > 0 are effective. z) > 0 ≡ sgn(q(f (x))) = 1 . . q(y) is a polynomial in y whose coefficients are polynomials in z with integer coefficients. We can replace q(x) by r(x). .e. z) is effective. . .1. write p(y. b and c. . t2 ) where p(t1 ) > 0 and p(t2 ) < 0. By induction on n. Let p(x) = an xn + · · · + a1 x + a0 ∈ R[x].

20. . . xk . . This proves the lemma. . If F ≡ ∃x G. the roots of q(x) can be found effectively. . . . Proof. . then we may take F ∗ ≡ ¬ G∗ . . . where η1 . . this shows that ξ is an effective function. . . . Therefore. . . . If F is quantifier free. . Theorem 3. G(x. xk ) . .1. y) is equivalent to a quantifier-free formula G∗ (x. . Then the sign of q(x) on the m + 1 intervals (−∞. . 1 ≤ i ≤ l. . . sgn(q(um +1)) . . it follows that the above disjunction is an effective predicate of x2 . . . . xk ) of LOR . xk . . then the predicate B(x2 . . In view of the previous lemma characterizing effective functions. By the Main Lemma. xk ) is equivalent to a finite disjunction A(η1 . . . xk ) ∨ · · · ∨ A(ηn . (um . y). x2 . . If F ≡ ¬ G. . Proof. . (um−1 . x2 . .1. The predicate A(x1 . . sgn q um−1 + um 2 . sgn q u1 + u2 2 . let F (y) ≡ ∃x G(x. By the 65 ≡ ∃x1 A(x1 . . Hence the pi (x)’s change sign only at ξ1 . xk ) . Then G∗ (x. where y is a list of the free variables of F . Lemma 3. Moreover the position of ξ relative to the ui ’s is determined by the positions of t1 and t2 relative to the ui ’s and by the sgn(p(ui ))’s. . x2 . xk . . then we may take F ∗ ≡ G∗ ∧ H ∗ . Let ξ1 . Let the roots of q(x) be among u1 < · · · < um . y). . . . . . ηn is a list of all the ξi ’s and (ξi + ξj )/2’s and ξi ± 1’s. . It follows that ∃x1 A(x1 . . it suffices to show that any formula F of LOR is equivalent over R to a quantifier free formula F ∗ .19.By induction hypothesis. xk with integer coefficients. ξm be all of these roots. um ) . . x2 . We shall prove this by induction on the number of symbols in F. . +∞) is given by m + 1 effective functions sgn(q(u1 −1)) . . A ⊆ Rk is defined over R by a formula F (x1 . u1 ) . . (u1 . y) defines an effective predicate B(x. x2 . u2 ) . x2 . . If A(x1 . . . y). . xk ) is an effective predicate. Any predicate A ⊆ Rk which is definable over R is effective. If F ≡ G ∧ H. we may take F ∗ ≡ F . . . ξm . . . Since the ηi ’s are effective functions of x2 . Thus we can use definition by cases to obtain sgn(q(ξ)) as an effective function. . . . The proof of the Main Lemma is now complete. xk ) may be viewed as a Boolean combination of polynomial inequalities of the form pi (x1 ) > 0. where the coefficients of the pi (x)’s are polynomials in x2 . . . . the roots of the pi (x)’s are effective function of x2 . . . . . By the inductive hypothesis. xk ) is effective.

. suppose that f is definable. 3. there is an algorithm to determine whether or not R satisfies S. Theorem 3. 2.1.e. 66 . Given a sentence S of LOR . This completes the proof. i. a special case of the previous theorem is that we can algorithmically compute S ∗ . there is an algorithm to find an equivalent quantifier free formula F ∗ .2. Theorem 3. A is definable over R by a quantifier free formula.21. If a predicate A ⊆ Rk is definable over R.e. y) is effective.20.. definable by a quantifier free formula. D ⊆ Rk . z)).1. for example 1+(0+1) < (1+1)·−(1+0). For predicates this follows immediately from Theorem 3. which is therefore effective in view of what has already been proved. A is definable over R.22 (Quantifier Elimination).previous lemma. 3. i.1. Given a sentence S. Consider now a function f .1. Note first that if f is effective then the predicate y = f (x) is effective. Similarly for functions f : D → R.1.2 Decidability of the Real Number System Theorem 3. z). For a predicate A ⊆ Rk the following three conditions are equivalent: 1.2. Given a formula F of LOR . Proof. Conversely. is defined by a quantifier free formula F ∗ (y). Proof of Theorem 3. and the truth value of such sentences is easily computed. Proof. This completes the proof. The algorithm can be obtained by tracing back through the proof of Theorem 3. the predicate A(f (x).22. whether S is true in the real number system. i.1. Proof. Corollary 3. Then for any effective predicate A(y. then it is definable over R by a quantifier free formula. z) is equivalent to the definable predicate ∃y (y = f (x) ∧ A(y. But then S ∗ is a Boolean combination of atomic sentences of the form t1 = t2 and t1 < t2 where t1 and t2 are variable-free terms. the predicate A(y) ≡ ∃x B(x. This is merely a restatement of the previous corollary.2. Theorem 3. Proof. an equivalent quantifier free sentence. Thus f is effective.e.1.1 is a restatement of the previous theorem. A is effective.. Clearly F is equivalent to F ∗ .1.

u. circles. 4. ·. Etc. 1. . Given a line L and a point P . 3. The theory of the ordered ring of real numbers is decidable. Recall that N = {0.} = the integers. b. ∞) = the real numbers.2.2. 2. . −2. By the methods of Cartesian analytic geometry. . 2. . (This is in contrast to the fact that TrueSntR is recursive. 0. 67 . 2. <. b. . A circle is an ordered triple (a. there is a unique line passing through them. Proof. express the following statements of Euclidean plane geometry. y) where x and y are real numbers. −.. 0.3. Similarly for the theory of points. 1. y) lies on a circle (a.} = the natural numbers. Proof. etc. Z = {.2.7. Corollary 3. . we can convert the proof of Theorem 3. A triangle consists of three non-collinear points. the intersection of L and C consists of at most two points.2. Write sentences of the language +. lines and circles in the plane is interpretable into the theory of the real numbers. v) = (0. Plane and solid geometry are decidable. A point (x. Alternatively. . This is a restatement of the previous corollary. y) lies on a line (a.1 into a rigorous proof that the function taking #(F ) to #(F ∗ ) is primitive recursive. and spheres in space. . the theory of points.1. among all points on L there is exactly one which is at minimum distance from P .Corollary 3. For every two points. when interpreted over the real number system.6. planes. = which. The set TrueSntR = {#(S) | S is a sentence ∧ S is true in R} is recursive. Exercise 3. For every line L and circle C.4. Show that TrueSntN and TrueSntZ are not recursive. Proof. a point is an ordered pair (x. and R = (−∞. For example. v) where (u. Exercise 3. lines.5. r) if and only if (x − a)2 + (y − b)2 = r2 . v) if and only if ∃t (x = a + tu ∧ y = b + tv) . For every three non-collinear points. A line is an ordered quadruple (a. Two lines are considered identical if and only if they contain the same points. b. .2. b. −1. 1. 1. This follows from the previous Theorem plus Church’s Thesis. r) where r > 0.) Theorem 3. 0). u. A point (x. the vertices of the triangle. there is a unique circle passing through them.

1. 1. 1.5. . there exists one and only one line L such that L ∩ C = P . 1. Explain in detail how you would translate the following statements of Euclidean plane geometry into sentences of the language +. Recall that N = {0. x1 .9. 7. . . . . 0. . Discuss analogous questions in which “solution in N” is replaced by “solution in R”. −2. . 2. x1 . 1..} = the integers. a tangent line. . Every angle can be uniquely bisected. < . . 0. we can find a polynomial f (w. xk ) = 0 has a solution in N is noncomputable. 2. Exercise 3. For every circle C and point P lying on C. ∞) = the real numbers. Exercise 3. . The three angle bisectors of any triangle meet in a single point.) 6. = over the real number system. . such that the set of a ∈ N for which the equation f (a. .} = the natural numbers. Discuss the analogous question in which “solution in N” is replaced by “solution in Z”. (I. xk ) with integer coefficients.2. . Z = {. and R = (−∞. . 2.2. . ·. 2. Every line segment has a unique midpoint. 68 . −.8. . According to Matiyasevich’s Theorem. . Every angle can be uniquely trisected. −1.e.

If a is any object and X is any set. Synonyms for “belongs to” are “is / an element of”. These requirements are rather vague. symbolically Y ⊆ X. Since a set is nothing but a collection of elements. the set itself having no further structure. Symbolically. X =Y ⇔ ∀a (a ∈ X ⇔ a ∈ Y ) . The need for some limitation-of-size requirement will be shown below in connection with the Russell paradox. .Chapter 4 Informal Set Theory The purpose of this chapter is to develop set theory in an informal. but we shall try to give some explanation of what they entail. and a ∈ X to mean that a does not belong to X. “is a member of”. and “is contained in”. a set is any collection of objects which may be regarded as a completed totality.. If P (a) is any definite property that an object a may or may not have. We use capital letters X. a collection of objects must be limited in size and definite. ı 4. R. (If such a set exists. This is known as the principle of extensionality. One of the most basic concepts in set theory is that of one set being included in another. . there is no third possibility. to denote sets. It can be taken as a definition of equality between sets.) In order to be a set. Definiteness means that any object a either belongs or does not belong to the collection. Y .e. preaxiomatic way. if every element 69 . i. A good reference for this material is Na¨ve Set Theory by P.1 Operations on Sets Informally. if such a set exists. Halmos. we use the notation {a | P (a)} to denote the set of all objects a which have property P . it will be unique in view of extensionality. . Limitedness means that the collection is in some sense not too large. we write a ∈ X to mean that a belongs to X. We say that Y is a subset of X. it follows that two sets are equal if and only if they contain exactly the same elements.

of Y is an element of X, i.e., ∀a (a ∈ Y ⇒ a ∈ X) . If P (a) is any definite property as above, and if X is any set, we can form a subset Y = {a ∈ X | P (a)}, consisting of all elements a of X which have property P . Thus ∀a(a ∈ Y ⇔ (a ∈ X ∧ P (a))) . Note that any set X is itself a mathematical object and as such can be an element of another set. Indeed, given a set X, we can form a set P(X) = {Y | Y ⊆ X} , called the power set of X, whose elements are all possible subsets of X. We now prove a theorem which implies that not all definite collections of mathematical objects are sets. This means that some limitation-of-size principle is needed. Theorem 4.1.1 (Russell Paradox). The collection of all sets is not itself a set. Proof. Suppose to the contrary that there were a set S consisting of all sets. Form the subset D consisting of all sets which are not members of themselves. Symbolically, D = {X ∈ S | X ∈ X} . / Then D ∈ D if and only if D ∈ D, a contradiction. This completes the proof. / The Russell Paradox shows that we cannot form sets with complete freedom. Nevertheless, a wide variety of sets can be formed. Some examples of sets are ∅ (the empty set, i.e., the unique set which has no elements), {∅} (the one-element set whose unique element is the empty set), {∅, {∅}}, etc. For any objects a, b, and c, we can form the set {a, b, c} whose elements are exactly a, b, and c. The cardinality of this set will be one, two or three depending on which of a, b, and c are equal to each other. Some examples of infinite sets are N (the set of natural numbers), R (the set of real numbers), P(N), P(R), P(P(R)), etc. Another source of examples is subsets defined by properties, for example {n ∈ N | n is prime} and {X ⊆ R | X is countably infinite}. Further sets can be obtained using the set operations discussed below. Given two objects a and b, there is an object (a, b) called the ordered pair of a and b. The ordered pair operation is assumed to have the following property: (a, b) = (a′ , b′ ) ⇔ (a = a′ ∧ b = b′ ) .

Given two sets X and Y , the Cartesian product X × Y is the set of all ordered pairs (a, b) such that a ∈ X and b ∈ Y . For any set X, a function with domain X is a rule f associating to each object a ∈ X a definite, unique object f (a). In this case we write dom(f ) = X and rng(f ) = {f (a) | a ∈ X}. The latter is again a set, called the range of F . 70

If f is a function and Z is a set, there is a unique function f ↾Z, the restriction of f to Z, whose domain is dom(f ) ∩ Z and such that (f ↾Z)(a) = f (a) for all a in its domain. We write f : X → Y to mean that f is a function, dom(f ) = X, and rng(f ) ⊆ Y . The set of all such functions is denoted Y X . Summarizing some of the above points, we have the following binary operations on sets: X ∪Y X ∩Y X \Y X ×Y XY = {a | a ∈ X ∨ a ∈ Y } = {a | a ∈ X ∧ a ∈ Y } = {a | a ∈ X ∧ a ∈ Y } / = {(a, b) | a ∈ X ∧ b ∈ Y } = {f | f : Y → X} (union) , (intersection) , (difference) , (product) , (exponential) .

By an indexed collection with index set I, we mean simply a function f whose domain is I. In discussing indexed collections, we use notation such as f = ai i∈I where ai = f (i). For example, an ordered n-tuple a1 , . . . , an is a function whose domain is {1, . . . , n}, and a sequence an n∈N is a function whose domain is N. Given an indexed collection of sets Xi i∈I , the following operations are defined.
i∈I i∈I i∈I

Xi Xi Xi

= = =

{a | ∃i ∈ I (a ∈ Xi )} {a | ∀i ∈ I (a ∈ Xi )} { ai
i∈I

(union) , (intersection) , (product) .

| ∀i ∈ I (ai ∈ Xi )}

If in the latter operation we take all of the sets Xi to be the same set X, we get the Cartesian power
i∈I

X = XI

which is the same as the previously mentioned exponential. The Axiom of Choice is the assertion that, for any indexed collection of sets Xi i∈I , if ∀i ∈ I (Xi = ∅) then i∈I Xi = ∅. This implies that it is possible to choose one element ai ∈ Xi for each i ∈ I. In the early years of set theory, there was some controversy about the Axiom of Choice. Nowadays the Axiom of Choice is accepted as being intuitively obvious, but we shall follow the custom of indicating which proofs use it.

4.2

Cardinal Numbers

A function f is said to be one-to-one if for all a, a′ ∈ dom(f ), a = a′ implies f (a) = f (a′ ). Note that in this case there is an inverse function f −1 with dom(f −1 ) = rng(f ) and rng(f −1 ) = dom(f ), defined by f −1 (b) = a ⇔ 71 f (a) = b .

Definition 4.2.1. If X and Y are sets, we say X is equinumerous with Y , written X ≈ Y , if there exists a one-to-one correspondence between X and Y , i.e., a one-to-one function f with dom(f ) = X and rng(f ) = Y . Lemma 4.2.2. 1. X ≈ X. 2. X ≈ Y if and only if Y ≈ X. 3. X ≈ Y and Y ≈ Z imply X ≈ Z. Proof. Straightforward. Because of the preceding lemma, we can associate to any set X an object card(X), the cardinality or cardinal number of X, in such a way that X≈Y ⇔ card(X) = card(Y ) .

If X is finite, we take card(X) to be the number of elements in X. For infinite sets X, it is not important at this stage what sort of object the cardinal number card(X) is, so long as the above property holds. We use Greek letters κ, λ, µ, ν, . . . to denote cardinal numbers. Definition 4.2.3. We write X Lemma 4.2.4. 1. If X ≈ X ′ and Y ≈ Y ′ , then X 2. X 3. X 4. X X. Y and Y Y and Y Z imply X Z. Y if and only if X ′ Y ′. Y to mean that X ≈ X1 for some X1 ⊆ Y .

X imply X ≈ Y . Y or Y X.

5. For all sets X and Y , either X

Proof. Parts 1, 2, and 3 are straightforward. Parts 4 and 5 will be proved later, as consequences of the Well-Ordering Theorem. Part 4 is known as the Cantor-Schroeder-Bernstein Theorem. We can now make the following definition for cardinal numbers: κ ≤ λ if and only if X Y where κ = card(X) and λ = card(Y ). This does not depend on the choice of X and Y , as noted in part 1 of Lemma 4.2.4. The rest of Lemma 4.2.4 implies that ≤ is a linear ordering of the cardinal numbers, i.e., we have: 1. κ ≤ κ. 2. (κ ≤ λ ∧ λ ≤ µ) ⇒ κ ≤ µ. 3. (κ ≤ λ ∧ λ ≤ κ) ⇒ κ = λ. 72

Suppose now that P(X) X holds. for infinite cardinal numbers κ and λ.2. / Then g(D) ∈ D if and only if g(D) ∈ D. κ + 0 = κ. Put D = {a ∈ X | a ∈ rng(g) ∧ a ∈ g −1 (a)} . a contradiction. κλ·µ = (κλ )µ . Theorem 4. 5.5 (Cardinal Arithmetic). / 73 . κ · (λ + µ) = κ · λ + κ · µ. 2. we have 2κ > κ.) For example. (κ + λ) + µ = κ + (λ + µ). For any cardinal number κ. For cardinal numbers κ = card(X) and λ = card(Y ). κ+λ = κ·λ = max(κ. In other words. ∀κ ∀λ (κ ≤ λ ∨ λ ≤ κ). κ · λ = λ · κ. (κ · λ) · µ = κ · (λ · µ). Proof.4. 6.7 (Cantor’s Theorem).2. 3. For cardinal numbers κ. Later we shall prove that. κ · λ = card(X × Y ). κλ+µ = κλ · κµ . Straightforward. it is assumed that X ∩ Y = ∅. 2κ = card(P(X)) where κ = card(X). λ) . 3.6. 8. κ · 1 = κ. κ · 0 = 0. Definition 4. 4. We have X P(X) via the one-to-one function f : X → P(X) where f (a) = {a}. and µ. we define 1. Proof. 7. κ + λ = card(X ∪ Y ). κ + λ = λ + κ. (In the definition of κ + λ. λ. we have 1. for any set X. Let g : P(X) → X be one-to-one. 2. κλ = card(X Y ). Theorem 4. we have X P(X) and P(X) X.2.

T ). S) and (B. The Axiom of Choice may be o viewed as the special case of K¨nig’s Theorem with κi = 0 and λi > 0. Definition 4. = Lemma 4. Given a set A. prove that ℵ0 = card(Z) = card(Q) and 2ℵ0 = card(R) = card(C) = card([0. an isomorphism from (A. S). = = = 74 . S) ∼ (C. S). if there exists an isomorphism from (A. 1]). Cantor’s Theorem may be viewed as the special case of K¨nig’s Theorem with κi = 1 and λi = 2. rng(f ) = B. symbolically (A. a′ ) ∈ R.2. R) to (B.2. R) ∼ (C.10 (K¨nig’s Theorem).Exercise 4. R) ∼ (B. R). a relation on A is a set R ⊆ A × A. R) ∼ (B. We say that (A.8. 2. If κi 1. (A. S).9. 1] × [0. S) is a one-to-one function f such that dom(f ) = A. we define κi = card( κi = card( i∈I i∈I Xi ) Xi ) i∈I where κi = card(Xi ). R) ∼ (A. Without using the Cantor-SchroederBernstein Theorem. Given two relational structures (A. R).3. then κi < λi . i∈I i∈I i∈I is an indexed set of cardinal numbers. We sometimes write aRa′ instead of (a. R) and (B. S) if and only if (B. S). it is assumed that Xi ∩Xj = ∅ Exercise 4. 1]) = card([0. In the definition of for all i. = = 3. κi . Define ℵ0 = card(N). A relational structure is an ordered pair (A.3. i∈I i∈I Remark 4.3 Well-Orderings and Ordinal Numbers Definition 4. and aRa′ ⇔ f (a)Sf (a′ ) for all a. R) to (B. Definition 4.2. T ) imply (A. Show that if κi < λi for all i ∈ I. = 2. R) ∼ (B.3.3. Suppose that κi i∈I and λi i∈I are ino dexed sets of cardinal numbers with the same index set I. a′ ∈ A.11. 1. (A. R) where A is a set and R ⊆ A × A. R) is isomorphic to (B.2. j ∈ I with i = j. o 4. S) ∼ (A.2.1. (A.

R). R ∪ S ∪ (A × B)). The following definition and exercise show how to generate further examples of ordinal numbers.7. then X = {an | n ∈ N} is a counterexample to well-foundedness of (A.. there exists a ∈ X such that there is no b ∈ X with bRa. Then we have X = ∅ and ∀a ∈ X ∃b ∈ X bRa. Definition 4. there is a unique a ∈ X such that aRb holds for all b ∈ X. R) is said to be well-founded if for every nonempty set X ⊆ A. A linear ordering is a relational structure (A. α · β = type(A × B. then X has an R-least element. Definition 4. R) with the following properties: aRb and bRc imply aRc. {(m. Note that if (A. Another important ordinal number is ω. Such an a might be called an R-minimal element of X.4. R) is well-founded if and only if there is no infinite descending R-sequence.3. 2.Proof. the order type of N itself (more precisely of (N. R) a mathematical object type(A. A relational structure (A. (a′ . i. Conversely. so long as the above property holds. Because of the preceding lemma. R) is. S) . An ordinal number is the isomorphism type of a well-ordering.3. we can associate to any relational structure (A. R) is a well-ordering and X is a nonempty subset of A. and for all a. R) = type(B. α + β = type(A ∪ B. R) ∼ (B. Lemma 4. Here we assume that A ∩ B = ∅. bRa hold. b). R) and β = type(B. A relational structure (A. Let α = type(A.3.6.5. a = b. b = a. By the Axiom of Choice. suppose that X ⊆ A is a counterexample to well-foundedness. This is an infinite descending R-sequence. the isomorphism type of (A. b ∈ A exactly one of aRb. 75 . R). b′ )) | bSb′ ∨ (b = b′ ∧ aRa′ )}). We define 1. For n ∈ N. It is not important at this stage exactly what sort of mathematical object type(A.8. If an n∈N is an infinite descending R-sequence.3. there is a function f : X → X such that f (a)Ra for all a ∈ X. R). {((a. S) = ⇔ type(A. Definition 4. Definition 4. n) ∈ N × N | m < n})). The lemma is proved.3. <) = (N. we identify n with the ordinal number which is the order type of all n-element well-orderings.) Proof. in such a way that (A. S) be ordinal numbers. Straightforward.e. A well-ordering is a linear ordering which is well-founded. (By an infinite descending R-sequence we mean a sequence an n∈N such that an+1 Ran for all n ∈ N. Pick an element a0 ∈ X and define an n∈N recursively by putting an+1 = f (an ) for all n ∈ N.

Exercise 4. an initial segment of (A. 76 . we have the following properties. R) ∼ (B. R). (A.3. Now let f be a function with dom(f ) ⊆ A and rng(f ) ⊆ B. its uniqueness follows from what we have already proved.) Note that the commutative laws fail. 1. It is clear that a′ Ra and a ∈ dom(f ) imply a′ ∈ dom(f ). Then {a ∈ A | f1 (a) = f2 (a)} is a nonempty subset of A. We first prove that the isomorphism is unique. Theorem 4. R) to (B. Then exactly one of the following holds. S). (α · β) · γ = α · (β · γ). {(a′′ . f (a) is undefined. R) is a linear ordering. 2. Let (A. 1. S) be well-orderings. If (A. and ordinal multiplication. (B. Definition 4. Then f1 (a′ ) = f2 (a′ ) for all a′ Ra but f1 (a) = f2 (a). Proof. so let a be its R-least element.3. If b exists.3. = provided such a b exists. Then f1 (a) ∈ rng(f2 ). α + 0 = 0 + α = α. Suppose for instance that f1 and f2 are two different isomorphisms from (A. However. contradicting the fact that rng(f2 ) is B / or an initial segment of B with respect to S. (Later we shall define an analogous operation of ordinal exponentiation. α · β. R) is any subset of A of the form {b | bRa}. S) ∼ some initial segment of (A. Give an example showing the failure of (α + β) · γ = α · γ + β · γ. in each case. a′ ) | a′′ Ra′ ∧ a′ Ra}) ∼ ({b′ | b′ Sb}. and b′ Rb and b ∈ rng(f ) imply b′ ∈ rng(f ). α · 1 = 1 · α = α. {(c. and we sometimes identify the initial segment with this ordering. S). = 2. Note that ({b | bRa}. α · (β + γ) = α · β + α · γ. α · 0 = 0 · α = 0. for example 1 + ω = ω = ω + 1 and 2 · ω = ω = ω · 2 = ω + ω. = 3. = Moreover. the isomorphism is unique. b) | cRb ∧ bRa}) is again a linear ordering. (A. α + β. defined by putting f (a) = the unique b ∈ B such that ({a′ | a′ Ra}. (α + β) + γ = α + (β + γ). S). 4. where a ∈ A. R) ∼ some initial segment of (B.Thus we have operations of ordinal addition.9. say f1 (a)Sf2 (a). R) and (B. {(b′′.11 (Comparability of Well-Orderings). If for a given a ∈ A no such b exists.10. b′ ) | b′′ Sb′ ∧ b′ Sb}) . S) or to some initial segment of (B. Prove the following laws of ordinal arithmetic. 3. αβ .

c) | dRc ∧ cRb}) . S). R) . prove that α < β if and only if α + γ = β for some γ > 0. Definition that (A. we see that f is an isomorphism from ({a′ | a′ Ra}. The second part is straightforward. R) type(B. Proof. then letting a be the R-least element of A \ dom(f ). S). For ordinal numbers α and β. If both dom(f ) = A and rng(f ) = B. and let b be the S-least element of B \ rng(f ). Define f : A → {β | β < α} by f (b) = type({c ∈ A | cRb}. This proves the theorem.3. {(γ. b′ ) | b′′ Sb′ ∧ b′ Sb}). 77 . and type(B. {(a′′ . γ. R) to ({b′ | b′ Sb}. If not. using comparability of well-orderings. S). R) onto ({β | β < α}. and β < α holds.3. R ∩ (B × B)) is a well-ordering. Then f is an isomorphism from ({a′ | a′ Ra}. S). β) | γ < β < α}) . then we have (A. Theorem 4. a′ ) | a′′ Ra′ ∧ a′ Ra}) to ({b′ | b′ Sb}. {(a′′ . R) and β = = We define α ≤ β to mean α < β ∨ α = β. Straightforward. the relational structure ({β | β < α}. For all ordinal numbers α and β. exactly one of α < β. Lemma 4. Lemma 4. This implies that a ∈ dom(f ). then (B. This completes the proof of the theorem.14. a contradiction.3. b′ ) | b′′ Sb′ ∧ b′ Sb}). α = β.3. we see that f is an isomorphism from (A.15. {(d. The next theorem implies that any well-ordering is isomorphic to an initial segment of the ordinal numbers. For any ordinal number α. For ordinal numbers α.13. Proof. we define α < β to mean ∼ some initial segment of (B. let a be the R-least element of A \ dom(f ).12. {(b′′. Proof. The first part follows from comparability of well-orderings. then letting b be the S-least element of B \ rng(f ).3. If rng(f ) = B.16. Let (A. {(b′′. 4. The claim is proved. R) is a well-ordering and B ⊆ A. where α = type(A. β) | γ < β < α}) is a well-ordering of type α. If = dom(f ) = A. {(γ. R) be some fixed well-ordering of type α.We claim that dom(f ) = A or rng(f ) = B or both. a′ ) | a′′ Ra′ ∧ a′ Ra}) to (B. It is straightforward to verify that f is an isomorphism from (A. R) ∼ (B. Moreover α < β and β < γ imply α < γ. R ∩ (B × B)) ≤ type(A. Exercise 4. If (A. β.

It is straightforward to verify that A = {α | α < γ} and that γ = sup X. ∀α (α ∈ X ⇒ α ≤ γ) and ∀β (β < γ ⇒ ∃α (α ∈ X ∧ β < α)). there is a set {β | β < α} consisting of all smaller ordinal numbers. Every set is a class. Let X be a nonempty set of ordinal numbers. show that α + sup X = sup{α + β | β ∈ X} and α · sup X = sup{α · β | β ∈ X}. β∈X 4. Lemma 4.3.21 (Burali-Forti Paradox). Let X be a set of ordinal numbers. Then X has a smallest element under <. Proof. Put A = {α | ∃β (β ∈ X ∧ α < β)} = It is straightforward to verify that (A. namely ({β | β < α}. <) would be isomorphic to an initial segment of itself. then by a function with domain C we mean a rule F which associates to each element a of C a uniquely defined object F (a). Let γ be the type of this well-ordering. β) ∈ A × A | α < β}) is a well-ordering.3. Exercise 4. For example. Lemma 4. This contradiction completes the proof.20. 78 . This follows from the previous theorem. β) | γ < β < α}). <) would be a wellordering.e. Letting α be the type of this well-ordering. {(γ.. We have seen this in connection with the Russell Paradox and the Burali-Forti Paradox. we see that (Ord. The latter set of ordinal numbers is well-ordered under <.3. Proof. but not every class is a set. Then there is an ordinal number γ = sup X which is the least upper bound of X under <.3. Proof. hence X has a least element. If C is a class. Proof. Examples of classes which are not sets are Set = {X | X is a set} and Ord = {α | α is an ordinal}. we shall be interested in functions with domain Ord. {(α. i.4 Transfinite Recursion By a class we mean a collection of objects which is not necessarily a set. For any ordinal number α. then by the above lemmas. Put α = sup X. although the class Ord of all ordinal numbers is not a set. The next theorem gives us a powerful method for defining such functions.18. Then X is a nonempty subset of {β | β ≤ α}. Theorem 4.19. These classes are “too big” to be sets. {α | α < β} .17. If Ord were a set. (Ord.Corollary 4. The class Ord of all ordinal numbers is not a set.3. If α is an ordinal number and X is a nonempty set of ordinal numbers.

4. where b′ is the S-greatest b ∈ B such that f1 (b) = f2 (b). if it exists. Here T is the set of all (f1 . α · β = sup{(α · γ) + α | γ < β}. for all but finitely many b ∈ B.4. This completes the proof. α + β = sup{α. F (α) = G(F ↾{β | β < α}) . we define the following operations of ordinal arithmetic. Let γ be the smallest β < α such that f1 (β) = f2 (β).4. let α be the smallest counterexample.4. R) and β = type(B. αβ = sup{1.2.1 (Transfinite Recursion). Then there is a unique function F with domain Ord such that. Definition 4. f2 ) ∈ C × C such that f1 (b′ )Rf2 (b′ ). Thus fα exists for all α.2. Let α = type(A. Define F by putting F (α) = G(fα ) for all α. hence f1 (γ) = G(f1 ↾{β | β < γ}) = G(f2 ↾{β | β < γ}) = f2 (γ) . Using the above claim. Exercise 4. f (β) = G(f ↾{γ | γ < β}). If α is an ordinal number and X is a nonempty set of ordinal numbers. (α + γ) + 1 | γ < β}. and it is easy to check that {(β.3. Then f1 ↾{β | β < γ} = f2 ↾{β | β < γ}.3. αγ · α | γ < β} (assuming α > 0).Theorem 4. a contradiction. G(fβ )) | β < α} is good. Exercise 4. Let F be the class of all functions whose domain is an initial segment of Ord. where a0 is the R-least element of A. let f1 = f2 be two such f ’s. If not. We claim that fα exists for all α.4. for all ordinal numbers α. Let us say that f ∈ F is good if for all β ∈ dom(f ).8 to encompass part 3 of Definition 4.5. Exercise 4. Suppose that G is a function with domain F. Then fβ exists for all β < α.3. 2. Show that parts 1 and 2 of Definition 4. S) be ordinal numbers.4. Show that αβ = type(C. Proof. We claim that for all ordinal numbers α. As an example of transfinite recursion.4. 79 . If not. T ) where C is the set of all f : B → A such that. let fα be the unique good f with dom(f ) = {β | β < α}. It is easy to check that F ↾{β | β < α} = fα and that F satisfies the desired conclusions. there is at most one good f with dom(f ) = {β | β < α}.8.2 agree with parts 1 and 2 of Definition 4. In the next exercise we show how to extend Definition 4. 3.4. f (b) = a0 . This contradicts the choice of α. show that 1. α + sup X = sup{α + β | β ∈ X}. 1.

α · 1 = 1 · α = α.4. Show that δ > 0 is a limit ordinal if and only if δ = sup{α | α < δ}. 3. αβ+γ = αβ · αγ . α + 0 = 0 + α = α. α · 0 = 0. Show that δ > 0 is a limit ordinal if and only if δ = ω · α for some α > 0. α · (β + 1) = (α · β) + α.8. Exercise 4. 1. and 3.4. show that 1. α + δ = sup{α + β | β < δ}.2. αβ+1 = (αβ ) · α.7. αδ = sup{αβ | β < δ} (assuming α > 0). α · 0 = 0 · α = 0.4. 5. A limit ordinal is an ordinal number δ such that δ > 0 and α + 1 < δ for all α < δ. α + 0 = α. For an ordinal number δ > 0. Examples of limit ordinals are ω and ω · 2. A successor ordinal is an ordinal number of the form α + 1. 9.6. α0 = 1. 2. α0 = 1. 3. αβ·γ = (αβ )γ . 6. Show that every ordinal number is either 0. For ordinal numbers α. Definition 4. Show that β < γ implies the following: 1.4. α1 = α. Exercise 4.11. 80 . Exercise 4.4. α · δ = sup{α · β | β < δ}. (α · β) · γ = α · (β · γ). a successor ordinal. 2. α + δ = δ for all α < δ. Exercise 4. 3. αβ < αγ provided α > 1. α · (β + γ) = α · β + α · γ. Show that the operations of ordinal arithmetic could have been defined by transfinite recursion as follows (letting δ denote a limit ordinal): 1. show that the following are equivalent. 0α = 0 provided α > 0. αsup X = sup{αβ | β ∈ X}.4. 4. α · β < α · γ provided α > 0. or a limit ordinal. 7. α + β < α + γ.9. 2. Show that α + 1 is the smallest ordinal number β > α. 1α = 1. β.10. Exercise 4. (α + β) + γ = α + (β + γ). α + (β + 1) = (α + β) + 1. α · sup X = sup{α · β | β ∈ X}. and γ. 8.

We shall prove the theorem in the following equivalent formulation: For any set X. Theorem 4. Proof.5. we can find a relation R ⊆ X × X such that (X. If not. δ = ω α for some α ≤ δ.5. and n ∈ N. there exists an ordinal number α such that X ≈ {β | β < α}. there exists a choice function for X. Form the set Y = rng(F ) = {a ∈ X | ∃α(F (α) = a)} . 3. where I = P(X) \ {∅} and Xi = i for all i ∈ I. Let α be the smallest ordinal number such that F (α) = a0 . Given a set X. we would have F (α) ∈ X for all α.e. We claim that F (α) = a0 for some α. and we have rng(F −1 ) = Ord. 4. By / transfinite recursion. Y = ∅. i.4. Show that for any ordinal number α. Then F −1 is a function with domain Y . Continued Lemma 4. (Neither α nor the one-to-one function from X onto {β | β < α} is asserted to be unique.e. Putting c(i) = ai we obtain our choice function... This contradiction proves the claim.5 Cardinal Numbers.12. a0 if X \ rng(F ↾{β | β < α}) = ∅ otherwise. By the Axiom of Choice. i. Then F ↾{β | β < α} is one-to-one. R) is a well-ordering. Our theorem is proved. Given a set X. define F (α) = c(X \ rng(F ↾{β | β < α})). Thus {β | β < α} ≈ X. β < δ. Consider the indexed set of sets Xi i∈I . i∈I Xi is nonempty.2 (Well-Ordering Theorem). α + β < δ for all α.) Fix a choice function c for X. Proof. a function c : P(X) \ {∅} → X such that c(Y ) ∈ Y for all Y ⊆ X. hence Ord is a set. Fix an object a0 such that a0 ∈ X. and rng(F ↾{β | β < α}) = X. 81 . and F (α) = F (β) for all α = β. Exercise 4. there is one and only one way to write α in the form α = α1 + · · · + αn where α1 ≥ · · · ≥ αn > 0 are additively indecomposable.1. An ordinal number with these properties is said to be additively indecomposable. there exists ai i∈I such that ai ∈ Xi for all i ∈ I.2.

Lemma 4. For all sets X and Y we have X ≈ Y if and only if |X| = |Y | . ω + 2. . From now on we shall make this identification. Thus we may identify cardinal numbers with initial ordinals. If X ⊆ {β | β < α}. .3. The finite ordinal numbers 0. 2.7. the finite cardinal numbers are now identified with the finite ordinal numbers. . By a chain within X we mean a set C ⊆ X such that for all U. Definition 4. Define ai i∈I by putting ai = the R-least element of Xi . An initial ordinal is an ordinal number α such that. . as is the fact that |X| ≤ |Y | implies X Y .5. (The existence of such an ordinal is a consequence of the Well-Ordering Theorem. By the previous theorem. writing card(X) = |X|.5. Exercise 4. .8. For any set X. there exists R ⊆ A × A such that (A. To see this. |X| is the unique initial ordinal such that X ≈ {β | β < α}. Note: This is a version of Zorn’s Lemma. By the previous lemma it follows that |X| = |Z| ≤ |Y |. Then X ≈ Z for some Z ⊆ {β | β < |Y |}. Use the Axiom of Choice plus transfinite recursion to prove that there exists a maximal chain within X. . and the smallest infinite cardinal number is the same as the ordinal number ω. are not initial ordinals.5. {γ | γ < α} ≈ {γ | γ < β} . This completes the proof. There is also a converse: the Well-Ordering Theorem implies the Axiom of Choice. Proof. we define |X| to be the smallest ordinal number α such that X ≈ {β | β < α}. then |X| ≤ α. Theorem 4. suppose we have an indexed set of nonempty sets Xi i∈I . for all β < α. Definition 4. R) is a well-ordering. Remark 4.5. Thus ai i∈I ∈ i∈I Xi and we have proved the Axiom of Choice from the Well-Ordering theorem. Immediate from Lemma 4. V ∈ C either U ⊆ V or V ⊆ U . and X Y if and only if |X| ≤ |Y | . By the Well-Ordering Theorem. 82 .Remark 4.5. Proof. and card(X) < card(Y ) if and only if |X| < |Y |. Suppose now that X Y . The first equivalence is obvious. . .5. are initial ordinals. A chain within X is said to be maximal if it is not properly included in any other chain within X.9.15.5. Let X be a set of sets.6. 1. . Put A = i∈I Xi .) Clearly |X| is an initial ordinal. Another simple fact worth noting is that every infinite initial ordinal is a limit ordinal. For instance. .5. we have card(X) = card(Y ) if and only if |X| = |Y |.4. But it is easy to see that ordinal numbers such as ω + 1. In fact. as is the first infinite ordinal number ω. The previous theorem shows that the Axiom of Choice implies the Well-Ordering Theorem.3. ω · 2. ω · 2 + 1.

|Y | < |X| holds.10. Most of these results are easy consequences of the following lemma. In particular B ≈ A. we have J ⊆ I × I where I is an appropriately chosen initial segment of (A. S) is isomorphic to (A. Theorem 4. we have κ · κ = κ. If not. S). β ′ ) = max(α. R). This proves the theorem. Proof. In other words. 4. Proof. If X Y and Y X. R) cannot be isomorphic to an initial segment of (B. 2. Thus every initial segment of (B. β ′ )S(α. Note that |A| = κ but every initial segment I of (A. Put B = A × A and define S ⊆ B × B by (α′ . Note that κ is an infinite initial ordinal. R). S) has cardinality < κ. β) ∧ α′ < α) ∨ . Y or Y X. β ) < max(α. Put µ = max(κ. R) has |I| < κ. it follows that (B. S).6. then X ≈ Y .1.6 Cardinal Arithmetic We now present some basic results about the arithmetic of infinite cardinal numbers. let κ be the smallest counterexample. For infinite cardinals κ and λ.2. This completes the proof.5. λ) . Thus (A. If J ⊆ B is any initial segment of (B. For all sets X and Y . and λ · λ < κ for all λ < κ.Theorem 4. β ′ ) = max(α. hence exactly one of |X| < |Y |. λ). S) is a well-ordering. From comparability of well-orderings. A × A ≈ A. |X| = |Y |. It is straightforward to verify that (B. β) ′ ′ if and only if max(α . β) ∨ (max(α′ . Then by the previous lemma we have µ ≤ κ+λ ≤ µ+µ = 2·µ ≤ µ·µ=µ and µ ≤ κ · λ ≤ µ· µ = µ. For infinite cardinals κ.6. Put A = {α | α < κ} and R = {(α′ . Both parts follow from the previous theorem plus the fact that |X| and |Y | are ordinal numbers. Lemma 4. either X Proof. 1. R) is a well-ordering of type κ. Hence (A. α) | α′ < α < κ}. hence κ · κ = κ. we have κ + λ = κ · λ = max(κ. β) ∧ α′ = α ∧ β ′ < β) . 83 (max(α′ .

ℵα is taken to be a cardinal. n) ∈ N × N. This proves the theorem. R P(Q) ≈ P(N) via the function which sends x ∈ R to {q ∈ Q | q < x}.6. i. and the real numbers respectively. n < ω.5. Thus |R| = 2ℵ0 . It is easy to see that ωα α∈Ord is a strictly increasing enumeration of all the infinite initial ordinals.6. Theorem 4.4. |R| = 2ℵ0 . Theorem 4. i. | n < ω. we have |X ∗ | = |X|.6. Exercise 4. Hence we have |X ∗ | = n<ω κn = ω · κ = κ . by transfinite recursion: 1. The fact that |N| = ℵ0 is obvious.3. X ∗ = { ai i<n For any set X.6. Proof. let X ∗ be the set of finite sequences of elements of X. Putting κ = |X|. we have |N| ≤ |Q| ≤ 2 · |N| · |N| = |N| so |Q| = |N| = ℵ0 . It follows that the class Card = {κ | κ is an infinite cardinal} is not a set. the infinite cardinals. For the real numbers.8. Definition 4. the notation ℵα = ωα is sometimes used in order to maintain a notational distinction between cardinals and initial ordinals. Since N ⊆ Q and each rational number q ∈ Q is of the form q = ±m/n for some (m. note first that P(N) R via the function which sends X ⊆ N to n∈X 2/3n . Q. Let N..9. ωδ = supα<δ ωα for limit ordinals δ. The cardinalities of these sets are given by |N| = |Q| = ℵ0 . Proof. To each ordinal number α we associate an infinite initial ordinal ωα as follows. On the other hand. Although cardinals are now the same thing as initial ordinals.6. while ω is thought of as the order type of the natural numbers under <. For λ infinite and 2 ≤ κ ≤ 2λ . Proof.Theorem 4. we have κ2 = κ · κ = κ and it is easy to prove by induction on n that κn = κ for all n ≥ 1.6. we have κλ = 2λ . ℵ0 is thought of as the cardinality of the set of natural numbers. the rational numbers. For any ordinal β we define β + = the smallest initial ordinal κ > β. For any infinite set X. while ωα is an initial ordinal. 3. even though ℵ0 is the same thing as ω.e. ai ∈ X for all i < n} . For example. 2λ ≤ κλ ≤ (2λ )λ = 2λ·λ = 2λ . and R be the set of natural numbers. Prove that there exist arbitrarily large ordinals α such that α = ωα . Remark 4.e. ωα+1 = ωα . Remark 4. ω0 = ω. + 2.. Hence 2ℵ0 ≤ |R|. 84 .7.6.6.

Exercise 4. and the GCH implies that every limit cardinal is a strong limit cardinal.7 Some Classes of Cardinals A cardinal is said to be uncountable if it is > ℵ0 . In this section we introduce some important classes of uncountable cardinals. Cantor’s Theorem tells us that 2κ ≥ κ+ . and we can ask whether 2κ = κ+ .3. 85 . Lemma 4. Definition 4. and exactly one of these possibilities holds. Prove that 2ℵ0 = ℵω . for any infinite cardinal κ.2. Clearly every strong limit cardinal is a limit cardinal. Prove that |RN | = 2ℵ0 and |RR | = 22 . The most important problem of infinite cardinal arithmetic is the Continuum Problem: What is the cardinality of R? Equivalently. Proof. we see that κi < λ implies κi ≤ κ. Every uncountable successor cardinal is regular. Thus λ = κ+ where κ is an infinite cardinal.7. Trivially ℵ0 is regular. An infinite cardinal λ is said to be regular if it is not the sum of fewer than λ cardinals each less than λ. or GCH.4. what is the ordinal number β such that 2ℵ0 = ℵβ ? By Cantor’s Theorem we have 2ℵ0 ≥ ℵ1 . Theorem 4.6.) ℵ0 4.10.1.11. for any indexed set of cardinals κi i∈I with |I| < λ and κi < λ for all i ∈ I. . Every uncountable cardinal is either a successor cardinal or a limit cardinal. The first uncountable limit cardinal is ℵω . 2ℵα = ℵα+1 for all ordinal numbers α) is known as the Generalized Continuum Hypothesis.7. or CH.7. The assertion that 2ℵ0 = ℵ1 is known as the Continuum Hypothesis. We say that λ is a strong limit cardinal if 2κ < λ for all κ < λ. and λ is a limit cardinal if and only if λ = ℵδ for some limit ordinal δ. also |I| < λ implies |I| ≤ κ. . Note that λ is a successor cardinal if and only if λ = ℵα+1 for some successor ordinal α + 1. o Exercise 4. . Let λ be an uncountable cardinal. . In other words. Definition 4. since it is not the sum of a finite set of finite cardinals. More generally. we have i∈I κi < λ. ℵ2 . hence i∈I κi ≤ i∈I κ = |I| · κ ≤ κ · κ = κ < λ. Proof. Given an indexed set of cardinals κi i∈I . We say that λ is a successor cardinal if λ = κ+ for some κ < λ. The first few uncountable successor cardinals are ℵ1 . We say that λ is a limit cardinal if κ+ < λ for all κ < λ. This shows that λ is regular. Let λ be an uncountable successor cardinal. (Hint: Use K¨nig’s Theorem.6. The assertion that 2κ = κ+ for all infinite cardinals κ (equivalently.7. Obvious.

are regular. since ℵω = n∈N ℵn . cf(2κ ) > κ. is the smallest cardinal κ such that λ = i∈I λi for some indexed set of cardinals λi < λ. regular. such cardinals are of the form λ = ℵλ . Indeed. Exercise 4. cf(λ) is regular.5.11.) o 5.  λ   κ + λ = λ if cf(λ) ≤ κ ≤ λ. prove that for all infinite cardinals κ and λ we have  if κ < cf(λ). Prove the following facts. . Assuming the GCH.e. and the GCH implies that every weakly inaccessible cardinal is inaccessible. strong limit cardinal.10. A weakly inaccessible cardinal is an uncountable. 1.7.In particular ℵ1 . a regular initial ordinal) if and only if.7. 2. It can be shown that every weakly inaccessible cardinal is a fixed point of the ℵα ’s. Show that α is a regular cardinal (i.7. Remark 4.7. Moreover. Let λ be an uncountable cardinal. Exercise 4. ℵ2 . Definition 4. (Hint: Use K¨nig’s Theorem. λ is regular if and only if cf(λ) = λ. regular. i ∈ I. The existence of inaccessible and/or weakly inaccessible cardinals is not obvious.   +  κ if κ ≥ λ. limit cardinal. For any infinite cardinal κ we have cf(µκ ) > κ for all µ > 1. Let α be an ordinal number which is > ω. Definition 4. In particular.. The cofinality of λ. Exercise 4. written cf(λ).7. 3. with |I| = κ..9. λcf(λ) > λ. for all such λ we have λµ = λ for all µ < κ.7. Let κ be an infinite regular cardinal. . we have type(X) = α. An infinite cardinal is said to be singular if it is not regular. Show that there exist arbitrarily large strong limit cardinals λ such that cf(λ) = κ. i. An inaccessible cardinal is an uncountable. 86 . The first singular cardinal is ℵω . λ is singular if and only if cf(λ) < λ. Clearly every inaccessible cardinal is weakly inaccessible.7. for every set of ordinal numbers X with sup X = α. Moreover.7.e.6. . Exercise 4. every strongly inaccessible cardinal has λκ = λ for all κ < λ. 4.8. we shall see later that the existence of such cardinals cannot be established using the accepted axioms of set theory.

4. we have Y ⊆ X. . A set X is said to be pure if every element of TC(X) is a set. This proves the theorem. It is straightforward to prove by transfinite induction on α that all elements of Rα are pure and well-founded. let f (Z) be the least ordinal number β such that Z ∈ Rβ .4. X is a pure set if not only X but also all the elements of X. we see that X is well-founded if and only if there is no infinite sequence of sets Xn n∈N with X = X0 ∋ X1 ∋ · · · ∋ Xn ∋ · · · . X ∈ Rα for some α. Conversely. By recursion on n < ω define TC0 (X) = X TCn+1 (X) = U(TCn (X)) .8 Pure Well-Founded Sets A set X is said to be transitive if. Y a set}. for every set Y such that Y ∈ X. Definition 4.) Proof.1. there is a smallest transitive set TC(X) including X. 87 . Then TC(X) = n∈N TCn (X) is easily seen to be the smallest transitive set Y such that Y ⊇ X. For each Z ∈ Y . elements of elements of X. suppose X is pure and well-founded. Then Y ⊆ Rγ . we write ∈|A = {(b. Applying Lemma 4. Put γ = supZ∈Y f (Z). there exists an ordinal number α such that Y ∈ Rα . If not. . Given a set X. In particular. a contradiction. α ∈ Ord. hence Y ∈ P(Rγ ) = Rγ+1 .2. In other words. as follows: R0 Rα+1 Rδ = = = P(Rα ) α<δ ∅ Rα for limit ordinals δ . . We claim that. Define U(X) = {Y | Y ∈ X.8. (TC(X) is called the transitive closure of X. ∈|TC(X)). are sets. hence β ≤ α implies Rβ ⊆ Rα . a) | a.5 to the relational structure (TC(X). let Y ∈ TC({X}) be ∈-minimal such that no such α exists. A set X is said to be well-founded if the relational structure (TC(X). ∈|TC(X)) is well-founded.3. for all Y ∈ TC({X}). Proof. Definition 4. a is a set. b ∈ A. b ∈ a} .8. . This proves the claim.8. By transfinite recursion we define transitive sets Rα .3. well-founded set. it is clear that β < α implies Rβ ∈ Rα . For any set A. X ∈ α∈Ord Rα if and only if X is a pure.8. Lemma 4. Theorem 4. By transfinite induction on α ∈ Ord.

8. Nevertheless. b}} . and rank Rα = α. well-founded sets is denoted V . real number. If X is a pure. ordinal numbers.2. We begin with ordered pairs and progress to functions. well-founded sets. function. Note that if a and b are pure. well-founded sets. Thus (a. If (a. Thus we have V = α∈Ord Rα . 88 . . b′ ) then a = a′ and b = b′ . for all ordinals α we have Rα = {X | X is a pure well-founded set of rank < α} . Definition 4. Assuming the GCH. 4.9. For each concept in this list. ordered pair. it is possible to identify mathematical objects of the given type with certain pure. well-founded sets. ordinal number. {a. For example. the restriction is partially justified by the fact that many or most mathematical objects can be reconstructed or redefined as pure. it is possible to define a structural replica of the natural numbers.5. the natural numbers 0. we define the rank of X to be the least ordinal number α such that X ⊆ Rα . from a certain perspective.1. b) is a set.9 Set-Theoretic Foundations In the next chapter we shall begin the study of axiomatic set theory.The class of all pure. well-founded sets. 1. A similar remark applies to each of following mathematical concepts: natural number. prove that |Rω+α | = ℵα for all ordinals α. b) = {{a}. . the set concept is usually restricted to pure. 2. then so is (a. b). well-founded sets. Moreover. In axiomatic studies of set theory.9. are not ordinarily regarded as being sets. Then clearly rank X = sup{rank Y + 1 | Y ∈ X} . Exercise 4. The purpose of this section is to show exactly how these identifications can be made. . but within the universe of pure. cardinal number. b) = (a′ . natural numbers can be viewed as certain kinds of pure. well-founded set. Thus. let us write (a. Lemma 4. and real numbers. well-founded sets. This restriction tends to isolate set theory from the rest of mathematics. For any two objects a and b.

and b is the unique element of Y ∈X Y \ {a} if the latter is nonempty. b) ∈ f ∧ (a. otherwise b = a. and that rank(Aα ) = α. The lemma follows. . By transfinite recursion we define Aα = {Aβ | β < α} for all ordinals α. ω = {0. . 2 = {0. From now on we shall make this identification. the rank of Aα is α. It is customary in pure set theory to identify the ordinal number α with the von Neumann ordinal Aα . . Thus. we have already seen how cardinal numbers may be identified with certain ordinal numbers. for all ordinals α we have α = {β | β < α} and α + 1 = α ∪ {α}. well-founded sets. A function is a set of ordered pairs. . If f is a function and a ∈ dom(f ) we write f (a) = the unique b such that (a.9.9. 2.5. Also.3. As for cardinal numbers. ∀a ∀b ∀c (((a. the initial ordinals.. f . . due to von Neumann. we already know how to identify cardinal numbers with certain pure. it is customary to replace this definition by the following. which we shall use from now on. 89 . Thus we have 0 = ∅ = {}. In pure set theory. Thus a and b can be recovered from (a. Putting X = (a.9.9. we see that a is the unique element of Y ∈X Y . then for each b ∈ A. i. namely. pure. Moreover. 1. well-founded set A such that (A. By transfinite induction on α. then sup X = X= α∈X α. we defined a function with domain X to be a rule associating to each a ∈ X a unique b. . The pure set-theoretic reconstruction of the ordinal numbers. From now on we shall make this identification. Proof. we may view (a. Lemma 4. 1} = {{}. ∈|Aα ) = α. The domain of f is dom(f ) = {a | ∃b ((a. ∈|Aα ) = α. Definition 4. b) by single-valued set-theoretic operations. Moreover.6. the initial segment B = {a | a ∈ b} is again a von Neumann ordinal. which is single-valued.} = N. . Remark 4. For each ordinal number α.Proof. c) ∈ f ) ⇒ b = c) . which is customary in pure set theory. {{}}}. 1 = {0} = {{}}. b) as the ordered pair formed from a and b. Note that if A is a von Neumann ordinal.e. b). A von Neumann ordinal is a transitive. is as follows: Definition 4. if X is any set of ordinals. b) ∈ f )}. By the above lemma. there is a unique von Neumann ordinal Aα such that type(Aα . b) ∈ f .4. it is straightforward to verify that Aα is the unique von Neumann ordinal such that type(Aα . In an earlier section of these notes. ∈|A) is a well-ordering.

and aRa for all a ∈ A. The ordered ring structure of Z is given by [(m..Finally. 90 .9. and (a. In order to define the real number system. b)] +Q [(a′ . b · b′ )] [(aa′ . The construction emplys the usual factorization of a set by an equivalence relation. Then aRb if and only if [a]R = [b]R . n)] 0Z ′ = = = = [(mm′ + nn′ .10 (the real number system). b′ )] ⇔ ab′ < a′ b [(a.9. n )] ⇔ = = = = = [(ab′ + a′ b. b)] 0Q 1Q ′ ′ [(m.9. m)] [(0.9.e. n + n′ )] [(m. n′ )] −Z [(m. 1)] [(a. For any a ∈ A we write [a]R = {b ∈ A | aRb}. n)] = [(m . b′ )] [(a. the equivalence class of a with respect to R. we follow Dedekind and begin with the natural number system (N.7. Moreover A/R is a partition of A.9. Remark 4. ·. where (m. b) ≡Q (a′ . b′ )] −Q [(a. as per the following definitions and remark. i. b)] <Q [(a′ . n′ ) if and only if m + n′ = m′ + n. Definition 4. mn′ + m′ n)] [(n. 1. +. 1)] [(1. b)] = [(a . Let R be an equivalence relation on A. i. An equivalence relation on A is a binary relation R ⊆ A × A with the following properties: aRb and bRc imply aRc. =.e. b)] [(0. <). sequences qn n∈N ∈ QN satisying ∀ε > 0 ∃m ∀n (n > m ⇒ |qm − qn | < ε) . 0)] m + n ′ = m′ + n m + n ′ < m′ + n 1Z ′ The rationals are defined by putting Q = (Z × Z+ )/≡Q . n′ )] ⇔ [(m. 0. We write A/R = {[a]R | a ∈ A}. 0)] [(1. Definition 4. The ordered ring structure of Q is given by [(a. n)] +Z [(m′ . Let A be a set.8. bb′ )] [(−a. Here S is defined to be the set of Cauchy sequences over Q. we turn to the set-theoretic construction of the real number system. a collection of pairwise disjoint sets whose union is A. Definition 4. b′ ) if and only if a · b′ = a′ · b. aRb implies bRa. where Z+ = {b ∈ Z | b > 0}. n′ )] = [(m + m′ .. n)] <Z [(m′ . b)] ·Q [(a′ . n)] ·Z [(m′ . Let R be an equivalence relation on A. b )] ⇔ ab′ = a′ b Finally. the reals are defined by putting R = S/≡R . n) ≡Z (m′ . The integers are defined by putting Z = (N × N)/≡Z .

.12. i.9. n ′ ≡R qn n if The ordered ring structure of R is given by ′ [ q n ] +R [ qn n ] ′ [ qn n ] ·R [ qn n ] = = = = = ′ [ qn + qn n ] ′ [ qn · qn n ] −R [ qn n ] 0R 1R [ −qn n ] [ 0 n] [ 1 n] [ qn n ] = [ [ qn n ] <R [ ′ qn n ] ′ qn n ] ′ ⇔ ∃ε > 0 ∃m ∀n (n > m ⇒ qn + ε < qn ) ′ ⇔ ∀ε > 0 ∃m ∀n (n > m ⇒ |qn − qn | < ε) Exercise 4.11. ′ ∀ε > 0 ∃m ∀n (n > m ⇒ |qn − qn | < ε) . In this section we have shown how many or most mathematical objects may be redefined or reconstructed as pure. i. Show that the real number system is complete. well-founded sets.9. every nonempty bounded subset of R has a least upper bound. pure set theory may be said to encompass virtually all of mathematics. Remark 4. 91 .And ≡R is the equivalence relation on S defined by putting qn ′ and only if limn |qn − qn | = 0.e. and one may speak of the set-theoretic foundations of mathematics.. This is why set theory is viewed as being of fundamental or foundational importance.e. In this sense.

1. ∨. . The standard or intended interpretation of L∈ is that the variables are to range over the class of pure. well-founded sets: two pure.1 The Axioms of Set Theory Definition 5.4.2. . An example of a sentence of L∈ is ∀x ∀y (x = y ⇔ ∀u (u ∈ x ⇔ u ∈ y)) . y}. ⇒. ⇔ and quantifiers ∀. and z. The notion of free variable is defined as usual. The language contains variables x.1. well-founded sets. ¬ . We define L∈ ..1. ∃. y. This standard interpretation or model of L∈ is sometimes known as the real world. It asserts that z is the unordered pair {x. Example 5. . This formula has free variables x. the set whose only elements are x and y. Formulas are built up as usual from atomic formulas by means of propositional connectives ∧. This sentence asserts the extensionality principle for pure..e. i. the language of set theory. The atomic formulas of the language are x = y and x ∈ y. Thus formulas and sentences are normally interpreted as making assertions about pure. Example 5. well-founded sets as elements. 92 .Chapter 5 Axiomatic Set Theory This chapter is an introduction to axiomatic set theory. y. wellfounded sets.1. z. 5. An example of a formula of L∈ is ∀u (u ∈ z ⇔ (u = x ∨ u = y)) .3. A sentence is a formula with no free variables. where x and y are variables. Remark 5. well-founded sets are equal if and only if they contain the same pure.1.

(union of a set of sets) z = x is an abbreviation for ∀u (u ∈ z ⇔ ∃v (v ∈ x ∧ u ∈ v)). (subset) x ⊆ y is an abbreviation for ∀u (u ∈ x ⇒ u ∈ y). (unordered pair) z = {x.e. (singleton) z = {x} is an abbreviation for z = {x. (ordered pair) z = (x.1. 3. ∃u ∃v (u = {x} ∧ v = {x. the standard interpretation of L∈ is in terms of the real world. 7. (intersection of a set of sets) z = w). Some important abbreviations are given in the following definition. y) is an abbreviation for z = {{x}. In later sections of this chapter. x is an abbreviation for ∀u (u ∈ z ⇔ ∀v (v ∈ x ⇒ u ∈ v)). {x. We can expand the language L∈ indefinitely by introducing abbreviations. v}).5.As mentioned above. the class of pure. Definition 5. 4. {x. (empty set) x = ∅ and x = {} are abbreviations for ∀u (u ∈ x).1. i. {x.. i.. 1. (intersection of two sets) z = x ∩ y is an abbreviation for z = ∃w (w = {x.. Such alternative interpretations play an essential role in axiomatic set theory.. y}. i. 9.e. (powerset) z = P(x) is an abbreviation for ∀y (y ∈ z ⇔ y ⊆ x). y} ∧ z = w).e. 93 . well-founded sets. / Using these abbreviations. y} ∧ z = 8. y} ∧ z = {u. i. 10.e.6 (Some Axioms of Set Theory). x}. 2. y} is an abbreviation for ∀u (u ∈ z ⇔ (u = x ∨ u = y)). y}}. 6. we shall consider interpretations or models of L∈ other than the real world. (union of two sets) z = x ∪ y is an abbreviation for z = ∃w (w = {x. 5. we can write down sentences expressing some of the axioms of Zermelo-Fraenkel set theory: Definition 5. y}.

1. We now introduce some more abbreviations and axioms. for u ∈ x. 4. ∀w (w ∈ f ⇒ ∃x ∃y (w = (x. 5. z) ∈ f ) ⇒ y = z) . . y}). y) ∈ f . Definition 5. The Axiom of Choice is the sentence ∀f ((Fcn(f ) ∧ ∀x (x ∈ dom(f ) ⇒ f (x) = ∅)) ⇒ Definition 5. Power Set Axiom: ∀x ∃z (z = P(x)).1. The Axiom of Infinity is the sentence 94 f = ∅). Union Axiom: ∀x ∃z (z = x). 1. 5. Definition 5. for all sets x. This is expressed by saying that. 2. 4. Note that. 6. 3. i. (value of a function) y = f (x) is an abbreviation of Fcn(f ) ∧ (x. Axiom of Foundation: ∀x (x = ∅ ⇒ ∃u (u ∈ x ∧ u ∩ x = ∅)). (domain of a function) z = dom(f ) is an abbreviation of Fcn(f ) ∧ ∀x (x ∈ z ⇔ ∃y (x. y) ∈ f ∧(x. Axiom of Extensionality: ∀x ∀y (x = y ⇔ ∀u (u ∈ x ⇔ u ∈ y)). y) ∈ f ). y)))∧∀x ∀y ∀z (((x.8. Pairing Axiom: ∀x ∀y ∃z (z = {x. The Axiom of Foundation is an attempt to express the idea that all of the sets under consideration are well-founded.e.. v))). (generalized Cartesian product) z = f is an abbreviation of Fcn(f )∧∀g (g ∈ z ⇔ (dom(g) = dom(f )∧∀x (x ∈ dom(f ) ⇒ g(x) ∈ f (x)))) . u ∩ x = ∅ means that u is ∈-minimal among elements of x. if x is nonempty then x contains an element u which is ∈-minimal.9.e. (Cartesian product) z = x × y is an abbreviation of ∀w (w ∈ z ⇔ ∃u ∃v (u ∈ x ∧ v ∈ y ∧ w = (u. i. 2. Only the Axiom of Foundation needs explanation. (function) Fcn(f ) is an abbreviation for a formula saying that f is a function. there is no element v of x such that v ∈ u. 3.1.1. Empty Set Axiom: ∃x (x = ∅).. Most of the above axioms are self-explanatory.7.

1. the idea being that / the set z should be in some sense logically subordinate to the property P . We write ∃ ! x to mean “there exists exactly one x such that”. if F is a formula with free variables x1 . consisting the universal closures of all formulas of the form ∃z ∀u (u ∈ z ⇔ (u ∈ x ∧ F (u))) . . 2. given a function f . This is accomplished by asserting the existence of a set that contains all of the sets 0 = {} = ∅. The syntactical requirement that the variable z does not occur freely in F (u) is imposed in order to avoid obvious contradictions such as z = {u ∈ x | u ∈ z}. ∃ ! x F (x) is an abbreviation of ∃y ∀x (F (x) ⇔ x = y) . 1 = {0}. Definition 5. there necessarily exists a set z ⊆ x consisting of all elements of x which have the given property P . Pairing. The Comprehension Scheme is extremely useful and important. z = domf = {x ∈ and of the generalized Cartesian product f = {g ∈ PPP( Definition 5. the Comprehension Scheme together with the Union. Actually. these two so-called axioms are not individual axioms. 95 f∪ f ) | domg = z ∧ ∀x (x ∈ z ⇒ g(x) ∈ f (x))}. . . Since our language L∈ does not enable us to discuss or quantify over arbitrary properties. 1. The Comprehension Scheme is an infinite set of axioms. given a set x and a property P that particular elements of x may or may not have.∃z (∅ ∈ z ∧ ∀x (x ∈ z ⇒ x ∪ {x} ∈ z)). The purpose of the Axiom of Infinity is to assert the existence of at least one infinite set. 2}. 1. . . 1}. . xn . i. .. for any formula F (x) in which x occurs as a free variable. 1. then the universal closure of F is the sentence ∀x1 · · · ∀xn F . y) ∈ f )}. We now introduce the two remaining axioms of Zermelo-Fraenkel set theory. we restrict attention to properties that are definable. For any formula F (u). f | ∃y ((x. 3 = {0. where F (u) is any formula in which z does not occur freely. expressible by means of a formula F (u). . . we write z = {u ∈ x | F (u)} as an abbreviation for ∀u (u ∈ z ⇔ (u ∈ x ∧ F (u))) .1. The Comprehension Scheme is our attempt to express the principle that. In other words.e. 2 = {0. Recall that. and Power Set Axioms logically imply the existence of a set z which is the domain of f .10. An axiom scheme is an infinite set of axioms all of which have a common form.11. For example. but rather axiom schemes.

2. The Replacement Scheme is an infinite set of axioms, consisting of the universal closures of all formulas of the form ∀u (u ∈ x ⇒ ∃ ! v F (u, v)) ⇒ ∃y ∀v (v ∈ y ⇔ ∃u (u ∈ x ∧ F (u, v))) , where F (u, v) is any formula in which y does not occur freely. The Replacement Scheme is our attempt to express the principle that, given a set x and a rule associating to each element u of x a unique object v, there exists a set y consisting of all the objects v which are associated to elements of x. Since our language L∈ does not enable us to discuss or quantify over arbitrary rules, we restrict attention to rules that are definable, i.e., expressible by means of a formula F (u, v). The syntactical requirement that the variable y does not occur freely in F (u, v) is imposed in order to avoid obvious contradictions. We have now introduced all of the axioms of Zermelo/Fraenkel set theory. We have, finally: Definition 5.1.12 (Zermelo/Fraenkel Set Theory). The axioms of Zermelo/Fraenkel set theory are as follows: the Axiom of Extensionality, the Empty Set Axiom, the Pairing Axiom, the Union Axiom, the Power Set Axiom, the Axiom of Foundation, the Axiom of Infinity, the Comprehension Scheme, and the Replacement Scheme. We use ZF as an abbreviation for “Zermelo/Fraenkel set theory”. Definition 5.1.13 (ZFC). The axioms of ZFC consist of the axioms of ZF plus the Axiom of Choice. The Zermelo/Fraenkel axioms together with the Axiom of Choice constitute the commonly accepted, rigorous, set-theoretic foundation of mathematics. A mathematical theorem is regarded as proved if and only if it is clear how to deduce it as a theorem of ZFC, i.e., a logical consequence1 of the ZFC axioms. It can be shown that all of the theorems of 19th and 20th century rigorous mathematics are logical consequences of the ZFC axioms. In particular, essentially all of the results of Chapter 4 can be stated and proved as theorems of ZFC.

5.2

Models of Set Theory

As mentioned above, the intended interpretation of L∈ is the so-called real world, i.e., the class of pure, well-founded sets. However, the general notion of pure, well-founded set is rather vague. In order to study and delimit this vagueness, axiomatic set theorists frequently consider alternative interpretations of L∈ . One important class of interpretations of L∈ is given in terms of relational structures. Recall that a relational structure is an ordered pair (A, E) where A is a set and E ⊆ A × A. Given a relational structure (A, E) and a sentence F of L∈ , it makes sense to ask whether F is true in (A, E), i.e., true when the variables are interpreted as ranging over A and x ∈ y is interpreted as xEy, i.e., (x, y) ∈ E.
1 The notions of theorem and logical consequence that we are using here will be explained in the next section.

96

Examples 5.2.1. The Axiom of Extensionality is true in a particular relational structure (A, E) if and only if (A, E) is extensional, i.e., for all a, b ∈ A, a = b implies {c | cEa} = {c | cEb}. Note that some relational structures are extensional and some are not. An example of an extensional relational structure is any linear ordering. An example of a nonextensional relational structure is (A, E) whenever |A| ≥ 2 and E = ∅. Thus a relational structure is extensional if and only if it is a model of (i.e., satisfies) the Axiom of Extensionality. The general point here is that any collection of sentences of L∈ defines a property of relational structures, namely the property of satisfying the given sentences. We formalize this in the following definition. Definition 5.2.2. Let S be any set of sentences of L∈ . A model of S is a relational structure (A, E) such that all of the sentences of S are true in (A, E). We say that S is consistent if there exists a model of S. If F is another sentence of L∈ , we say that F is a logical consequence of S, written S ⊢ F , if F is true in every model of S. Note that if S is inconsistent then all sentences are logical consequences of S, so the notion of logical consequence is uninteresting in this case. If however S is consistent, then it is meaningful to ask which sentences are logical consequences of S, i.e., what conclusions follow when the sentences of S are assumed as axioms. This is the kind of question which axiomatic set theory seeks to answer. Naturally the focus is on sets of sentences which make assertions that could reasonably be true in the intended model, i.e., the real world, i.e., the class of all pure, well-founded sets. As an easy example of the notion of logical consequence, note that the sentence ∀x (x ∈ x) is a logical consequence of the Axiom of Foundation plus the / Pairing Axiom. This is so because x ∈ x would imply that {x} has no ∈-minimal element. Specializing Definition 5.2.2 to S = ZF and S = ZFC, we have: Definition 5.2.3. A model of ZFC is a relational structure (A, E) such that (A, E) satisfies all of the Zermelo/Fraenkel axioms plus the Axiom of Choice. A theorem of ZFC is any sentence which is a logical consequence of the ZFC axioms. The notions model of ZF and theorem of ZF are defined similarly. Axiomatic set theory is essentially the study of models of ZF and of ZFC, with an eye to discovering which set-theoretic propositions follow or do not follow from these axiom systems. For instance, one of the important results2 of axiomatic set theory is that there exists a model of ZF which is not a model of ZFC. In other words, the Axiom of Choice is not a logical consequence of the ZF axioms. Another key result is that both the Continuum Hypothesis and its negation are consistent with ZFC. In other words, CH is independent of ZFC. Thus the ZFC axioms, although powerful and flexible, do not suffice to answer basic set-theoretic questions such as the Continuum Problem.
2 This

result will not be proved here.

97

We end this section by presenting some general definitions and results concerning relational structures. Definition 5.2.4. Let (A, E) be a relational structure. A k-place predicate P ⊆ Ak is said to be definable over (A, E) if there exists a formula F (x1 , . . . , xk , y1 , . . . , ym ) of L∈ and parameters b1 , . . . , bm ∈ A such that P = { a1 , . . . , ak ∈ Ak | (A, E) satisfies F (a1 , . . . , ak , b1 , . . . , bm )} . (5.1)

More generally, given B ⊆ A, we say that P ⊆ Ak is definable over (A, E) allowing parameters from B if there exists a formula F (x1 , . . . , xk , y1 , . . . , ym ) of L∈ and parameters b1 , . . . , bm ∈ B such that (5.1) holds. Definition 5.2.5. Let (A, E) be a relational structure. Then Def((A, E)) is the set of all subsets of A that are definable over (A, E). Note that Def((A, E)) ⊆ P(A). Lemma 5.2.6. Let (A, E) be a relational structure. 1. If A is finite, then Def((A, E)) = P(A) and |Def((A, E)| = 2|A| . 2. If A is infinite, then |Def((A, E))| = |A|. Proof. For finite A the result is obvious. Suppose now that A is infinite. By G¨del numbering, the set of all formulas of L∈ is countable. Since any element o of Def((A, E)) is determined by a formula and a finite sequence of parameters from A, we have |Def((A, E))| ≤ ℵ0 · |A∗ | = |A| . On the other hand {{a} | a ∈ A} ⊆ Def((A, E)), hence |A| ≤ |Def((A, E))|. This completes the proof. Definition 5.2.7. Let (A, E) and (A′ , E ′ ) be relational structures. We say that (A′ , E ′ ) is a substructure of (A, E), abbreviated (A′ , E ′ ) ⊆ (A, E), if A′ ⊆ A and E ′ = E ∩ (A′ × A′ ). We say that (A′ , E ′ ) is an elementary substructure of (A, E), abbreviated (A′ , E ′ ) ⊆elem (A, E), if (A′ , E ′ ) ⊆ (A, E) and, for all formulas F (x1 , . . . , xn ) and a1 , . . . , an ∈ A′ , (A, E) satisfies F (a1 , . . . , an ) if and only if (A′ , E ′ ) satisfies F (a1 , . . . , an ). Lemma 5.2.8. Given (A′ , E ′ ) ⊆ (A, E), we have (A′ , E ′ ) ⊆elem (A, E) if and only if every nonempty subset of A which is definable over (A, E) allowing parameters from A′ has a nonempty intersection with A′ . Proof. Straightforward. Recall that a relational structure (A, E) is said to be countable if the underlying set A is countable. 98

By Theorem 5. E) of ZFC. In other words. An+1 = {c(X) | ∅ = X ⊆ A ∧ X is definable over (A.10 (Skolem Paradox). where E ′ = E ∩ (A′ × A′ ). We define recursively a sequence of sets An ⊆ A. the notion of countability is relative to the model under consideration (as are many other set-theoretic notions). In this context it is customary to identify the transitive set T with the relational structure (T. by A0 = ∅. Assume that ZFC is consistent. The Skolem Paradox is called a paradox for the following reason: the existence of a countable model of ZFC would seem to contradict the fact that the existence of uncountable sets is a theorem of ZFC. E) be a relational structure such that |A| ≥ κ. ∈|T ). 5. n ∈ N. Note that An ⊆ An+1 for all n.Theorem 5. ∈|T ) satisfies all the sentences of S. For any relational structure o (A. This completes the proof. there is no contradiction here. E) allowing parameters from A′ . Hence A′ = {An | n ∈ N} is countable. Then (A′ . E) may be countable in the real world. Let (A.2. E ′ ) ⊆elem (A. E). E). Proof. E) such that X ⊆ A′ and |A′ | = κ. By induction on n it is straightforward to show that An is countable for all n. well-founded set T such that the relational structure (T.2. E ′ ) is a countable model of ZFC. and let X ⊆ A be a subset of A such that |X| ≤ κ. there exists a countable elementary substructure (A′ . Let κ be an ino finite cardinal.2. E ′ ) ⊆elem (A. Proof. Prove Theorem 5.11 (Generalized L¨wenheim-Skolem Theorem).2. A straightforward generalization of Theorem 5. Let S be a set of sentences of L∈ . Definition 5.2. then there exists a countable model of ZFC. Corollary 5.8 we have (A′ . Proof. Exercise 5. 99 . E). A transitive model of S is any transitive. Then there exists an elementary substructure (A′ . pure. (A′ . Then there exists a model (A. Recall that a set T is transitive if and only if every element of T is a subset of T .1. because a set that is uncountable within a particular model (A. Straightforward.12. If ZFC is consistent. Let c : P(A) \ {∅} → A be a choice function for A.9 (A. E ′ ). E) allowing parameters from An }. E) has a countable elementary submodel.2.2.2. Moreover c(X) ∈ A′ for all X = ∅ definable over (A.3. Hence by Lemma 5. Actually.11.9 (L¨wenheim/Skolem Theorem).3 Transitive Models and Inaccessible Cardinals In this section we study an important class of models.9 is: Theorem 5. E ′ ) of (A.

3. ∈|T ) are isomorphic. The following definition concerning transitive models is of general interest. ∈|T )) = {X ⊆ T | X is definable over T } . i. 100 . Hence f = F −1 ↾A is an isomorphism of (A. 2.. 3. F (x) = a0 otherwise. i.3. ∈|T ). then the following assertions are equivalent: 1. ∈|A) where T is a transitive.3. relational structures of the form (A. Let T be a transitive.. Lemma 5. = Proof. The following lemma consists of some simple remarks characterizing which transitive models satisfy which axioms of ZFC. Then there exists a transitive. b} ∈ T ). i.2. E) be a relational structure which is well-founded and extensional.e. E) onto (T.5. Theorem 5. It is easy to verify that T is a transitive. pure. Note that F (x) = F (y) ∈ A implies x = y. The following theorem provides a converse and thereby characterizes transitive models up to isomorphism among arbitrary models. define F (x) as follows: F (x) = the unique a ∈ A such that rng(F ↾x) = {b | bEa} if such an a exists. Hence T = rng(F −1 ↾A) is a set. E) is well-founded and extensional. well-founded set. E) ∼ (T. Let T be any transitive. T satisfies the Pairing Axiom if and only if T is closed under pairing.. pure. We write Def(T ) = Def((T. (A. Let (A. E) is isomorphic to some transitive model (T. 1. ∈|T ) (allowing parameters from T ). wellfounded set. ∀a ∀b ((a ∈ T ∧ b ∈ T ) ⇒ {a.Note that every transitive model is well-founded and extensional. It is straightforward to verify that T and f are unique. Fix an object a0 ∈ A.3. well-founded set. pure. 2. ∈|T ) onto (A. ∈|T ). well-founded set and that F ↾T is an isomorphism of (T. ∈|T ) are unique. A k-place predicate P ⊆ T k is said to be definable over T if and only if it is definable over (T.3. Definition 5. T satisfies the Union Axiom if and only if T is closed under union. E) and (T. (A.4. T always satisfies the Axiom of Extensionality. well-founded set x. The rest of this section is devoted to the study of transitive models. T always satisfies the Axiom of Foundation. Of particular interest are transitive models of ZFC.e. Moreover the transitive set T and the isomorphism f : (A. If (A. pure. E) is a relational structure. E). pure. By transfinite recursion on the rank of an arbitrary / pure.e. well-founded set T such that the relational structures (A. Corollary 5. 4.

we have ∀a (a ∈ T ⇒ rng(F ↾a) ∈ T ).3. Proof. pure.7. T satisfies the Empty Set Axiom if and only if ∅ ∈ T . 10. since ω < δ. 2. T satisfies the Axiom of Choice if and only if. An infinite cardinal λ is regular if and only if. Straightforward.8.3. ∀x ((x ⊆ Rλ ∧ |x| < λ) ⇒ x ∈ Rλ ). for all X ∈ Def(T ). Proof. Recall also that we have identified cardinals with initial von Neumann ordinals (cf.4.∀a (a ∈ T ⇒ a ∈ T ).4. and Union Axioms and the Axiom of Choice and the Comprehension Scheme hold in Rδ because δ is a limit ordinal. 9. Sections 4.3.5 and 4. T satisfies the Axiom of Infinity if and only if ∃a (a ∈ T ∧ ω ⊆ a). Let δ be a limit ordinal > ω. Pairing. If λ is an inaccessible cardinal. Straightforward. 5. Power Set. 6. Lemma 5. for every indexed family of nonempty sets ai i∈I ∈ T . |X| < λ implies sup X < λ. T satisfies the Comprehension Scheme if and only if. we have ∀a (a ∈ T ⇒ a ∩ X ∈ T ). Lemma 5. The Axioms of Extensionality and Foundation hold in Rδ because Rδ is a transitive.6. then 1.8). T satisfies the Power Set Axiom if and only if ∀a (a ∈ T ⇒ P(a) ∩ T ∈ T ). ∀x (x ∈ Rλ ⇒ |x| < λ). Recall that an inaccessible cardinal is a regular. we have T ∩ i∈I ai = ∅. for all X ⊆ λ. Then Rδ is a transitive model of all of the ZFC axioms except possibly the Replacement Scheme. strong limit cardinal. well-founded set. T satisfies the Replacement Scheme if and only if for all functions F : T → T such that F ∈ Def(T ). 8. The Empty Set. Proof.5. The Axiom of Infinity holds in Rδ because ω ∈ Rδ . uncountable.3. We shall now show that inaccessible cardinals give rise to transitive models of ZFC. 101 . Lemma 5. 7. We apply Lemma 5.

|Rλ | = supα<λ |Rα | = λ.3. To see this. Proof. Thus Rλ is a model of ZFC + “inaccessible cardinals do not exist”. since |Rα | < λ and λ is regular. If there exists an inaccessible cardinal. ∈|Rω ). then the existence of an inaccessible cardinal is not a theorem of ZFC. it follows by the previous lemma that sup(rngρ) < λ. This contradicts the choice of λ. We claim that Rλ also satisfies “inaccessible cardinals do not exist”. = 2. Specializing this to the case when F is definable over Rλ . This completes the proof. Assume that there exists an inaccessible cardinal. . 2. Then it is easy to see that κ is also an inaccessible cardinal in the real world. Let λ be the smallest inaccessible cardinal.6 we see that Rλ satisfies all of the ZFC axioms except possibly the Replacement Scheme. E) ∼ (Rω . are finite sets. Let F : Rλ → Rλ and a ∈ Rλ be given. elements of elements of x.12. Define E ⊆ N by putting mEn if and only if 2m occurs in the binary expansion of n. then |Rα+1 | = |P(Rα )| = 2κ < λ since λ is a strong limit cardinal.3. 102 .3. suppose that Rλ satisfies “there exists at least one inaccessible cardinal”. For limit ordinals δ < λ.5 that the Replacement Scheme holds in Rλ . Exercise 5.. 1. Let λ be an inaccessible cardinal. Show that Rω is a model of all of the axioms of ZFC except the Axiom of Infinity.3. Clearly λ is a limit ordinal > ω. Exercise 5. Hence x ⊆ Rα . Conclude that P ⊆ ω k is definable over Rω if and only if P is arithmetical. say rngρ ⊆ α < λ. Theorem 5.10. 3. Then Rλ is a transitive model of ZFC. hence by the previous lemma rng(F ↾a) ∈ Rλ . m = ni for some i where n = 2n1 + · · · + 2nk . Then |rngρ| ≤ |x| < λ. 1.3. pure. well-founded sets. then ZFC is consistent. Corollary 5. Then rng(F ↾a) ⊆ Rλ and |rng(F ↾a)| ≤ |a| < λ. . If |Rα | = κ < λ. we have inductively |Rδ | = | α<δ Rα | = supα<δ |Rα | < λ.e.3. hence x ∈ Rα+1 ⊆ Rλ since λ is a limit ordinal. By the previous theorem. |Rλ | = λ. By transfinite induction on α < λ we prove |Rα | < λ.3. But clearly κ < λ. Rλ is a model of ZFC. Proof.11. Proof. we see by Lemma 5. . Since λ is regular. i. Define ρ : x → λ by ρ(u) = rank(u). Show that Rω is the set of all hereditarily finite. Let κ ∈ Rλ be such that Rλ satisfies “κ is an inaccessible cardinal”. Immediate from the theorem. If there exists an inaccessible cardinal.3. . hence by Lemma 5. Theorem 5. We have R0 = ∅.9. A hereditarily finite set is a finite set x such that all elements of x. Show that (N.13. Proof.

The previous theorem shows that. then for any a ∈ T we have that a is an ordinal (i. ∈|A) is a countable. By Theorem 5.4 Constructible Sets Recall that. ∈|Rλ ). For all ordinals α. Proof. pure. ∈|Rδ ) is an elementary submodel of (Rλ . (A.1. if T is any transitive. 5. well-founded. ∈|A) is an elementary submodel of (Rλ . For all ordinals α. By the L¨wenheim-Skolem Theorem. pure. the lemma follows easily by transfinite induction on α. Proof.e.16.e. a von Neumann ordinal) if and only if T satisfies “a is transitive and (a. pure. the lemma follows easily by transfinite induction on (von Neumann) ordinals α. Proof. well-founded set T . if two or more inaccessible cardinals exist. A set X is said to be constructible if X ∈ Lα for some ordinal α.2. Show that. transitive model of ZFC. Exercise 5. then there exists a countable. Prove that there exists a limit ordinal δ < λ such that (Rδ . With these observations. ∈|T ). there exists a countable set A ⊆ Rλ such o that (A. 103 . and Lα ⊆ Rα . over (T. well-founded set. Lα is a transitive.3.3.2. ∈|T ) is a countable. Exercise 5. pure. Then (Rλ . If there exists an inaccessible cardinal. extensional model of ZFC. well-founded set. If T is any transitive. transitive model of ZFC. if we assume only the axioms of ZFC. then the existence of two or more inaccessible cardinals is not a theorem of ZFC + “there exists at least one inaccessible cardinal”. pure. Def(T ) is the set of all subsets of T that are definable over T (i. α ∈ Ord.4.3. Definition 5. well-founded set. ∈|Rλ ) is a model of ZFC.4. Thus T ∩ Ord ∈ Def(T ). Thus (T. Let λ be an inaccessible cardinal. ∈|a) is a linear ordering”. The class of all constructible sets is denoted L. With this observation. Thus (A.3. For any transitive..4. Lemma 5. ∈|A) is isomorphic to a transitive model (T. Let λ be an inaccessible cardinal.14. By transfinite recursion we define Lα . ∈|Rλ ). Theorem 5. well-founded set. we have α = Lα ∩ Ord.3.. ∈|T )) allowing parameters from T . Lemma 5.15. as follows: L0 Lα+1 Lδ = = = ∅ Def(Lα ) α<δ Lα for limit ordinals δ . then we cannot hope to prove the existence of inaccessible cardinals. we have T ⊆ Def(T ) ⊆ P(T ) and hence Def(T ) is again a transitive.

and Lδ = α<δ Lα for limit ordinals δ. The Pairing Axiom holds in L because x. . By the previous lemma. then there exist arbitrarily large ordinals α such that F1 . . fk be functions from Ord to Ord. . This has the following interesting consequence: If F1 . y ∈ Lα implies {x. . sup{fi (β) | β < αn . Now let ∃y Gi . for all a1 . . . 1 ≤ i ≤ k. define an increasing sequence of ordinals αn . L satisfies F (a1 . . we shall need the following lemmas. i = 1. . . xini are the free variables of ∃y Gi .. ani ) for some b ∈ Lβ . . . it is the set of all Y ∈ Lα such that Lα satisfies Y ⊆ X.4. . To show that Comprehension and Replacement hold in L. From this it follows by Lemma 5. . Write Gi ≡ Gi (y. ani ) = the least ordinal β such that a1 . namely. fk . . . . . The Axioms of Extensionality and Foundation hold in L because L is transitive and consists of pure. For each Y ∈ P(X) ∩ L put ρ(Y ) = the least β such that Y ∈ Lβ . . . an ) if and only if Lα satisfies F (a1 . the same lemma holds for the Rα hierarchy as well. fk . . We have now shown that for all X ∈ L. . put gi (a1 . . 104 . It is straightforward to verify that such an α has the desired property. Some of the axioms of ZFC are straightforwardly verified in L using Lemma 5. Hence P(X) ∩ L ∈ Def(Lα ) = Lα+1 . ani ) | a1 . . . . xn ) be a formula of L∈ with free variables x1 . an ). . Hence P(X) ∩ L is definable over Lα . a1 . . . Replacing ∀ by ¬ ∃¬ as necessary. fi (β) < α for all β < α. . The proof of the previous lemma used only the following properties of the constructible hierarchy: α ≤ β implies Lα ⊆ Lβ . well-founded sets. . . . To show that the Power Set Axiom holds in L. The Axiom of Infinity holds in L because ω ∈ Lω+1 .4. For a1 . xn . Given an ordinal γ. . let X be any constructible set. . Let f1 . xi1 . Fk is a finite set of sentences that are true in the real world. . ani ∈ L. if L satisfies ∃y Gi (y. Putting α = sup{αn | n ∈ N} we see that α > γ and α is closed under f1 . then L satisfies Gi (b. . Lemma 5. . the Union Axiom holds in L because x ∈ Lα implies x ∈ Lα . . . . . .3.4. there exist arbitrarily large ordinals α such that that α is closed under f1 . . we may safely assume that F contains no occurrences of ∀. The Empty Set Axiom holds in L because ∅ ∈ L1 . . .4. . i. . ani ). . . . . αn+1 = max(αn + 1. . . . . .6. . 1 ≤ i ≤ k}). y} ∈ Lα+1 . . Then there exist arbitrarily large ordinals α such that. Let F (x1 .5 (reflection). P(X) ∩ L ∈ L.3. . Proof.5. Remark 5.e. . . . . .5 that the Power Set Axiom holds in L. . . Put α = sup{ρ(Y ) | Y ∈ P(X) ∩ L}. . . . . . . n ∈ N inductively by α0 = γ. . . . . For instance. ani ∈ Lβ and such that. . . Fk are true in Rα . xini ) where xi1 . . . an ∈ Lα . . . Since the Rα hierarchy also has these properties. a1 .We are going to show that the constructible sets form a model of ZFC. k be a list of the subformulas of F of the form ∃y G. ani ∈ Lβ }. Then there exist arbitrarily large ordinals α such that α is closed under f1 . Define fi : Ord → Ord by fi (β) = sup{gi (a1 . Lemma 5. Thus P(X) ∩ L ⊆ Lα . fk . . Proof. .

Lemma 5. . For all ordinals α ≥ ω. . We must prove that.2. 2. and x ∈ Lα . This is clear since b = = {v ∈ L | L satisfies ∃u (u ∈ a ∧ F (u. . In more detail. v. v. cn ) if and only if Lα satisfies F (u.6 we have |Lω | = ℵ0 and. c1 . . The class L of constructible sets satisfies the ZF axioms plus V = L. for α ≥ ω.4. . We claim that b is definable over Lα . c1 . . z1 . Const(x) asserts the existence of a transfinite sequence of sets Lβ β≤α such that Lβ = {Def(Lγ ) | γ < β} for all β ≤ α. for all u ∈ L. v. v ∈ Lα . Theorem 5. put b = rng(f ↾a). whence b ∈ L. . . From this the lemma easily follows by induction on α ≥ ω. c1 . . cn ∈ Lβ and b ⊆ Lβ . f (u) = the unique v ∈ L such that L satisfies F (u. .4.8. . Theorem 5. c1 . well-founded sets. c1 . let α be such that α > β and. cn ). We use V = L to abbreviate ∀x Const(x). Proof. zn such that. . . for all a ∈ L. . This shows that the Replacement Scheme holds in L. . . Now given a ∈ L.9. . and L is the class of all constructible sets. . . 1. . Recall that V is the class of all pure.7.4. We must show that b ∈ L. . we have parameters c1 . v. . L satisfies the Comprehension and Replacement Schemes. To show that the Replacement Scheme holds in L. cn ))} . well-founded sets are constructible.11. rng(f ↾a) = {f (u) | u ∈ a} also belongs to L. It is tedious but straightforward to show that L satisfies V = L. cn ). v. . Thus b ∈ Def(Lα ) = Lα+1 . By Lemma 5. . for all u. . wellfounded set x is constructible. . v. v. cn ))} {v ∈ Lα | Lα satisfies ∃u (u ∈ a ∧ F (u.4. We introduce some more abbreviations: Definition 5. cn ∈ L and a formula F (u.10. . we have |Lα | = |α|. Proof. |Lα+1 | = |Def(Lα )| = |Lα |. The above lemmas show that L satisfies the ZF axioms. let f : L → L be a function which is definable over L. . . Const(x) is an abbreviation for a formula asserting that a given pure. z1 . . Note first that. 105 . By Reflection. Let β be an ordinal so large that a.4. . The proof that the Comprehension Scheme holds in L is similar. . The class L of constructible sets satisfies the Axiom of Choice. . L satisfies F (u. zn ) with free variables u. . since f is definable over L. Proof. . . Thus V = L is a sentence asserting that all pure.Lemma 5. c1 .

and this completes the proof. In particular L satisfies that each Lα is well-orderable. Lemma 5. we have within L that |P(κ)| = κ+ . Theorem 5.14. Refining the proof of Lemma 5. For any infinite cardinal κ. we have (A. 1968.4.4. Lemma 5. S is identical with the sentence V = L of Definition 5.5. ∈|Lα ) for some = limit ordinal α. The class L of constructible sets satisfies the Generalized Continuum Hypothesis.12.3 L satisfies the Axiom of Choice. hence 2κ = κ+ .3. Clearly |Lα | = κ. 106 . Proof. Suppose that a ⊆ κ and a is constructible.4. then a ∈ Lα for some ordinal α such that |Lα | = κ. In particular a ∈ Lα .13. its validity does not depend on the Axiom of Choice. we have |P(κ) ∩ L| ≤ κ+ . we can find a set A ⊆ Lδ such that κ ∪ {a} ⊆ A. Proof. Let δ > κ be a limit ordinal such that a ∈ Lδ . Since L satisfies the axioms of set theory. since |Lκ+ | = |κ+ |. Lemma 5.3. ∈|A) ∼ (Lα . This proves the theorem. Theorem 5. it follows by another application of Theorem 5.8. for all ordinals α. Hence by Theorem 5.4. pages 497–513.4.11). |A| = κ. see Boolos and Putnam. From the previous lemma we have P(κ) ∩ L ⊆ Lκ+ . Let κ be an infinite cardinal. ∈|A) is an elementary submodel of (Lδ . Hence by Remark 4.” Journal of Symbolic Logic. F (α) is a well-ordering of Lα . then a ∈ Lα for some countable ordinal α. Since κ ∪ {a} is a transitive subset of A.10 implies that each Lα is well-orderable.4. ∈|Lδ ).15.4. Thus for all infinite cardinals κ of L.2 and Lemma 5.2.12. If a is any constructible subset of ω. By the Generalized L¨wenheim/o Skolem Theorem (Theorem 5. Proof. For all transitive sets A.10. Since the definition of F is explicit. For details of the construction of S.Proof. There is a sentence S of L∈ with the following property.4. if a ∈ P(κ) ∩ L where κ is an infinite cardinal.9 the definition of F can be carried out within L. The construction of the sentence S is straightforward but tedious. Roughly speaking. Volume 33. “Degrees of unsolvability of constructible sets of integers. This argument actually shows that the Axiom of Choice follows from ZF plus V = L. By Theorem 5. the proof of the previous lemma can be carried out within L. and (A. More generally. we can explicitly define by transfinite recursion a function F : Ord → V such that. A satisfies S if and only if A = Lα for some limit ordinal α. The desired conclusion is immediate.2 that κ ∪ {a} ⊆ Lα . Proof. Lemma 5.16.4.4. Our remaining goal with respect to constructible sets is to show that L satisfies the GCH.

Definition 5. A filter on P is a set G ⊆ P such that 1. 2. Let P = (P. 2. and 5.3.4. Proof. due to G¨del 1939.4. we write p ⊥ q. A filter G ⊆ P is said to be M -generic if G ∩ D = ∅ for all dense open D ⊆ P such that D ∈ M . Let G be an M -generic filter. If p. 5. Given p ∈ P we can find an M -generic filter G ⊆ P such that p ∈ G. q ∈ P are incompatible. 107 . Let A be a transitive model of ZF. The proof of the second part is similar. then so does ZFC + GCH. Remark 5. starting with a model (A. It is easy to verify that G = {q ∈ P | ∃n (pn ≤ q)} is an M -generic filter. Definition 5. Let {Dn | n ∈ N} be an enumeration of {D ∈ M | D dense open in P }. Lemma 5. Within A we can carry out the definition of L to obtain a transitive submodel B (sometimes called an “inner model”) consisting of the constructible sets of A.4. We put M [G] = {aG | a ∈ M }. Definition 5.5. is one of the most o significant achievements of axiomatic set theory.5.17. We say that p. given pn . This proves the first part. If ZF is consistent. let pn+1 ≤ pn be such that pn+1 ∈ Dn . ∀p ∈ P ∃q ≤ p (q ∈ D). q).1.15 then show that B is a model of ZF plus V = L plus the Axiom of Choice plus the GCH.) The proofs of theorems 5. p ≤ q imply q ∈ G. (It can be shown that B = Lα where α is the least ordinal that is not an element of A. Put p0 = p and.5.5.5. The second part is sometimes described as a relative consistency result: ZFC + GCH is consistent relative to ZF. If ZF has a transitive model. p.1. Definition 5.5. E) that is not necessarily transitive. ∀p ∈ D ∀q ≤ p (q ∈ D).9. 2.4.11. p ∈ G. Proof. 5. then so is ZFC + GCH. ≤) be a partially ordered set belonging to M .5 Forcing Let M be a countable transitive model of ZFC.4. The previous theorem. q ∈ P are compatible if there exists r ∈ P such that r ≤ p and r ≤ q. D ⊆ P is dense open if 1. q ∈ G implies ∃r ∈ G (r ≤ p.2.

Proof. b ∈ M . . . We have M [G] ⊇ M ∪ {G}.7. Also. . .5. . . . M [G] is a countable transitive set. . Lemma 5. Remark 5. . .11 (forcing). The proof of Theorem 5.5. put c = P × {a. . Let p ∈ P and let F be a sentence of the forcing language. . . Furthermore. A major result is: Theorem 5.5.10. and a1 . bG }.6. .5. we now introduce the method of forcing. M [G] satisfies Extensionality ˙ and Foundation automatically. a1 . . M [G] satisfies the Axiom of Infinity because ω = (ω)G ∈ M [G]. q ≤ p imply p ∈ G. Lemma 5. because M [G] is a transitive set. because q ∈ G. . . We shall show that M [G] is a countable transitive model of ZFC containing M . p) | p ∈ P }. M [G] is a countable transitive model of ZFC. b) ∈ a for some p ∈ G}.5. an ∈ M . and from this ˙ G = (G) it follows that Ord ∩ M ⊆ Ord ∩ M [G]. . . where a = P × {b | b ∈ a}. . .where aG = {bG | (p. . where quantifiers are interpreted as ranging over M [G]. for each a ∈ M we clearly have rank(a) ≥ rank(aG ). an ) if and only if M |= F ∗ (p.12 (the extension lemma). then q − F . On the other hand. given a. . . an ∈ M . . . For example. . some parts of the proof are easy and do not require forcing. b}. xn ) is a formula of L∈ with free variables x1 . and Ord ∩ M [G] = Ord ∩ M . . It is obvious from the definition that M [G] is a countable transitive set. . ˙ For all a ∈ M we have a = (a)G . Proof. We say that p − F (read p forces F ) if and only if M [G] |= F for all M -generic filters G such that p ∈ G. Sentences of the forcing language are of the form F (a1 . . where F (x1 . . . For each formula F (x1 . This is obvious. . an ) is true in M [G].13 (definability of forcing). hence Ord ∩ M ⊇ Ord ∩ M [G]. Definition 5. an ). .5. xn ) there is a formula F ∗ (p.8 is long and employs a new method known as forcing. Think of each a ∈ M as a “name” for aG ∈ M [G]. where G = {(p. This shows that M [G] satisfies the Pairing Axiom. xn ) such that. . then cG = {aG . . However. . .5. 108 . In order to prove the rest of Theorem 5. .5.5. p − F (a1 . for all p ∈ P and a1 . Thus M ∪ {G} ⊆ M [G]. . xn . . (an )G respectively.9. We also have ˙ ˙ ˙ ˙ G . Remark 5. If p − F and q ≤ p. So far we have not used the assumption that G is an M -generic filter. . We have M [G] |= F (a1 . .5. x1 . . . Lemma 5. and a1 .8.8. . . . . Definition 5. . an ) if and only if F (a1 . . an ). The forcing language consists of binary relation symbols ∈ and = plus constant symbols a for each a ∈ M . . an are interpreted as (a1 )G .

13 and 5. y) of L∈ defining the relations p − a ∈ b and p − a = b.15. To show (∃p ∈ G) p − ¬ F . we need to find formulas ∈∗ (p. a1 .5. x. Let G be generic with p ∈ G. y) and =∗ (p. assume the right hand side. to find c ∈ M such that M [G] |= ∀u (u ∈ c ⇔ (u ∈ a ∧ F (u. For focing equals truth. Clearly D is dense open. For the converse. Suppose q ≤ p. Proof. a1 . a. an ). M [G] |= F so let q ∈ G be such that q − F . We assume that F contains only ∧. The formulas ∈∗ and =∗ are defined by a rather complicated simultaneous transfinite recursion within M . ⇔. and ∗ ∗ Thus we may define (F1 ∧ F2 )∗ = F1 ∧ F2 and (∀x F (x))∗ = ∀x F ∗ (x). . . for definability of forcing. an ) ≡ ¬ (∃q ≤ p) F ∗ (q. . over M . note that p − F1 ∧ F2 if and only if p − F1 and p − F2 . and ∀ (not ∨. an ∈ M . . To see this. If p − F . We shall prove Lemmas 5. . a1 . . ¬ . . Put 109 . contradicting our assumption. D ∈ M . Thus. a. This takes care of ∧ and ∀. suppose M [G] |= ¬ F . . . ∃).5. Let r ∈ G be such that r ≤ p and r ≤ q. . . Proof. assume the left hand side. . Given a. M [G] |= F if and only if ∃p ∈ G (p − F ). .14 (forcing equals truth). . We now proceed to the proof of Theorem 5. Also p ∈ G since q ≤ p. For the inductive step. contradicting our assumption.8.Lemma 5.5. Let G be generic such that q ∈ G. q − F . we may take (¬ F )∗ (p.14 by simultaneous induction on the number of connectives and quantifiers in F . We omit the details. M [G] |= the Comprehension Scheme. a1 . an ))). Therefore p does not force ¬ F . Lemma 5. Put D = {p | p − F or p − ¬ F }. b) and p − a = b if and only if M |= =∗ (p. ⇒. Hence p − ¬ F . .5. p − ¬ F if and only if ¬ ∃q ≤ p (q − F ). For the base step. we claim that p − ∀x F (x) if and only if p − F (a) for all a ∈ M . By definability of forcing. Otherwise. x. Let p ∈ D ∩ G. The properties p − a ∈ b if and only if M |= ∈∗ (p. Then M [G] |= F . a contradiction. then M [G] |= F .5. respectively. Then r ≤ p and r − F . b) are proved by transfinite induction on rank(a) and rank(b). To show M [G] |= ¬ F . . For ¬ .

We claim that cG ⊇ P(aG ) ∩ M [G]. if there exists d ∈ M such that p − F (b. .5. w1 . Thus aG is well orderable in M [G]. by Definability of Forcing in M . Given a. b) | b ∈ a and p − b ∈ a ∧ F (b. . we prove the independence of V = L. Lemma 5. Proof. . . a1 . Thus M [G] |= Bounding. M [G] |= the Union Axiom. .c = {(p. an )}. M [G] |= the Replacement Scheme. Proof. There exists a countable transitive model of ZFC plus V = L. Given aG ∈ M [G]. This is similar to the Power Set Axiom. . . v. and the Union Axiom follows by Comprehension in M [G]. wn ) ⇒ ∀u ∈ a ∃v ∈ c′ F (u. w1 . . . . aG . . . Given a ∈ M put c=P× Then cG ⊇ a. a1 . wn ) ]. a1 . . Proof. . an ). . This proves our claim. Lemma 5. Composing f with the function b → bG .5. an ). by the Forcing Equals Truth Lemma. 110 M [G] |= ∀u ∃ ! v F (u. This is because Bounding plus Comprehension implies Replacement. . let d = {(p.18. . Proof. .16. It suffices to prove that M [G] |= the Bounding Scheme: ∀w1 · · · wn [ ∀u ∃ ! v F (u. . we have f = (f˙)G ∈ M [G]. . By definability of forcing. wn ) ⇒ ∀x ∃y ∀u ∈ x ∃v ∈ y F (u. let f ∈ M map an ordinal α onto a. . . .19. hence dG ∈ cG . As a first application. .8 is now complete. let c ∈ M be such that. Given a ∈ M .20. Then c ∈ M .5. for all (p. Moreover dG = eG ∩ aG . v.5.5. Lemma 5. then c contains such a d. Then. a1 . we obtain in M [G] a mapping of α = (α)G onto {bG | b ∈ ˙ a} ⊇ aG . To see this. since P(aG ) ∩ M [G] = {dG ∈ cG | M [G] |= d ⊆ a}. given eG ∈ M [G]. we have cG = {bG | b ∈ a and M [G] |= F (b. v. . b) ∈ P × a | p − b ∈ e ∩ a}. b) ∈ P × a. . We then have where c′ ∈ M . v. Since M ⊆ M [G].5. an ∈ M . Lemma 5. . d ∈ M .17. a1 . . . . The proof of Theorem 5. M [G] |= the Power Set Axiom. . an )}. M [G] |= the Axiom of Choice. d. namely c′ = P × c. Theorem 5. w1 . . . The Power Set Axiom follows by Comprehension in M [G]. put c = P × P(P × a) ∩ M .

Dα is dense open.5. To see this.6. Letting p ∈ G ∩ Dα .5. rng(fG ) = κ. In other words. ˙ ˙ 111 . Partially order P by putting p ≤ q if and only if p extends q. Thus g ∈ M [G] maps ω onto κ. Let M be a countable transitive model of ZFC. Clearly every cardinal of M [G] is a cardinal of M . Because ω is infinite.3. If g ∈ M . Put g = G. We claim that M [G] |= V = L.20. cardinals of M are often preserved.6 Independence of CH As in the previous section. Lemma 5.2. a contradiction. On the other hand. all cardinals > |P | in M are preserved in M [G]. Thus g = G is a partial function from ω into 2. ordered by p ≤ q if and only if p extends q. we see that α ∈ rng(p). remain cardinals in M [G]. fG ∈ M [G].6. and let G be an M -generic filter on P . let M be a countable transitive model of ZFC. Clearly Dα ∈ M . We claim that rng(g) = κ. Proof. Suppose not. Then D ∈ M . say fG : λ → κ. put Dα = {p ∈ P | α ∈ rng(p)}. Cardinals of M can be collapsed in M [G]. In particular κ is not a cardinal of M [G].1. we see that g : ω → κ. Suppose M |= “κ is a cardinal > |P |”. / and clearly D is dense open. given n ∈ ω. we see that n ∈ dom(p).6. Then M [G] |= “κ is a cardinal”. let P be a partially ordered set belonging to M . Note that p. We begin with a discussion of cardinal collapsing and cardinal preservation in M [G]. M [G] |= g ∈ L ∧ g : ω → 2 ˙ / ˙ where (g)G = g. Let G be an M -generic filter on P . Let κ be an uncountable cardinal of M . q ∈ P are compatible if and only if p ∪ q ∈ P . Remark 5. We claim that g ∈ M . To see this. Then P ∈ M . In fact. hence n ∈ dom(g). But G ∩ D = ∅. give α < κ. then clearly / G = {p ∈ P | p ⊆ g} ∈ M . Example 5. Thus g : ω → 2 and g ∈ M [G].8 we have M [G] |= ZFC. λ < κ. It follows that M [G] |= “κ is a countable ˙ ordinal”. Then in M we have ˙ ˙ ∀α < κ ∃β < λ ∃p ∈ P (p − f | λ → κ and p − f (β) = α). ˙ 5. so let D = P \ G = {p ∈ P | p ∈ G}. Let P be the set of finite partial functions from ω into κ.. 1}. put Dn = {p ∈ P | n ∈ dom(g)}. i. the converse does not always hold. Letting p ∈ G ∩ Dn . As in the proof of Theorem 5. hence α ∈ rng(g).Proof.e. We claim that dom(g) = ω. and Dn ∈ M . Clearly Dn is dense open. Let P be the set of finite partial functions from ω into 2 = {0. By Theorem 5. However. Let G be an M -generic filter on P .

Xi ∩ Xj = D for all i. ordered by putting p ≤ q if and only if p ⊇ q. j ∈ I. ˙ Proof. n. Proof.) if every antichain of P is countable. we may assume Xi (l) = Xj (l) for all l < k and all i.c. for each l < k. 1}. Hence. we may assume Xi = Xj for all i. We may safely assume that |I| = ℵ1 and that i∈I Xi ⊆ ω1 . let Xi (1) < · · · < Xi (n) be the elements of Xi . j ∈ I. Definition 5. Then P is c. Say dom(pi )∩dom(pj ) = D for all i. Proof.By the Pigeonhole Principle.c. Then all cardinals of M are preserved in M [G]. by passing to an uncountable subfamily.c.c.8. Let Xi i∈I be an uncountable family of finite sets. Let k be as small as possible such that {Xi (k) | i ∈ I} is uncountable. we have an uncountable ∆-system. Lemma 5. M |= “κ is a cardinal”. Reasoning within M .6. For each i ∈ I. Case 2: Otherwise.6. α1 = α2 implies qα1 ⊥ qα2 . By passing to an uncountable subfamily. This is a contradiction. α2 ∈ Xβ . An antichain in P is a set A ⊆ P such that the elements of A are pairwise incompatible. Thus Aβ = {qα | α ∈ Xβ } is an uncountable antichain in P . By the ∆-system lemma. For each α ∈ Xβ let qα ≤ p be such ˙ ˙ that qα − f (β) = α.4. Note that κ = β<λ Xβ . This contradicts the assumption that P is c. {Xi (k) : i ∈ I} is countable. for α < κ ˙ ˙ ˙ and β < λ say that α is a possible value of f (β) if ∃q ≤ p (q − f (β) = α). A ∆-system is an indexed family of sets Xi for some fixed set D. we may pass to a subfamily such that dom(pi ) i∈I is a ∆system. we may use transfinite recursion to define a function f : ω1 → I such that. Then Xf (α) α<ω1 is an uncountable ∆-system contained in Xi i∈I .5.c. for each α < ω1 . i = j. some Xβ is uncountable. M [G] |= “f maps λ onto ˙ κ”. we may assume that ∃n ∀i ∈ I |Xi | = n. . Suppose not. Suppose not. Note that α1 .6. ˙ ˙ ˙ Definition 5. Lemma 5. We now proceed to the independence of the Continuum Hypothesis. Since {Xi (k) | i ∈ I} is uncountable.6. P is said to have the countable chain condition (c. Xf (α) (k) > supβ<α Xf (β) (n). we can find p ∈ P . β < λ. Therefore. i∈I such that. say κ > λ. Thus we have a fixed finite set D = {Xi (l) | 1 ≤ l < k} for all i ∈ I. j ∈ I. Let pi i∈I be an uncountable antichain in P . Fix p ∈ P such that p − “f maps λ onto κ”. i = j. Suppose M |= P is c. Let X be any set. i∈I Xi is countable. In this case. Fix such a β. . .c. Let ˙ Xβ = {α | α is a possible value of f (β)}. Then. Passing to an uncountable subfamily. j ∈ I.c. In particular.7 (the ∆-system lemma). Any uncountable family of finite sets contains an uncountable subfamily which is a ∆-system. . Let P be the set of finite partial functions from X into {0. {Xi (l) | i ∈ I} is countable.6.c. α1 < α2 < κ such that ˙ ˙ ˙ p − f : λ → κ and p − f (β) = α1 and f (β) = α2 . Case 1: For each k = 1. There are only finitely 112 .6. Lemma 5.

many functions from D into {0, 1}, so by passing to an uncountable subfamily we may assume that pi ↾D = pj ↾D for all i, j ∈ I. Then for all i, j ∈ I we have that pi ∪ pj is a function, hence pi and pj are compatible, a contradiction. Theorem 5.6.9. Let M be a countable transitive model of ZFC. Let κ be an uncountable cardinal of M . Then there exists a countable transitive model M ′ of ZFC extending M such that (1) M ′ satisfies 2ℵ0 ≥ κ, and (2) M ′ has the same ordinals and cardinals as M . Proof. Let P be the set of finite partial functions from κ × ω into {0, 1}. Let G be an M -generic filter on P . By Lemma 5.5.7 and Theorem 5.5.8, M [G] is a countable transitive model of ZFC which includes M and has the same ordinals as M . By Lemma 5.6.8 P is c.c.c. By Lemma 5.6.5 M [G] has the same cardinals as M . Put g = G. As in the proof of Theorem 5.5.20 we see that g ∈ M [G] and g : κ × ω → {0, 1}. For α < κ define gα : ω → {0, 1} by gα (n) = g((α, n)). We claim that gα = gβ for all α < β < κ. To see this, let Dαβ be the set of p ∈ P such that p((α, n)) = p((β, n)) for some n ∈ ω such that (α, n), (β, n) ∈ dom(p). Clearly Dαβ ∈ M and is dense open. Hence G ∩ Dαβ = ∅. Hence gα = gβ . It is now clear that M [G] |= 2ℵ0 ≥ κ. Thus we may take M ′ = M [G]. ˙

113

Chapter 6

Topics in Set Theory
6.1 Stationary Sets

Definition 6.1.1. Let S be a set of ordinals. We say that S is unbounded in a limit ordinal δ if sup(S ∩ δ) = δ. We say that S is closed in κ if S ⊆ κ and, for all limit ordinals δ < κ, if S is unbounded in δ then δ ∈ S. A closed unbounded set in κ (sometimes called a club of κ) is any subset of κ which is closed in κ and unbounded in κ. Lemma 6.1.2. Let κ be a regular uncountable cardinal. 1. If Ci , i ∈ I, is a collection of closed unbounded sets in κ, and if |I| < κ, then i∈I Ci is again a closed unbounded set in κ. 2. If Cα , α < κ is a collection of closed unbounded sets in κ indexed by the ordinals less than κ, then the diagonal intersection △α<κ Cα = {β < κ | β ∈ Cα for all α < β} is again a closed unbounded set in κ. Proof. Straightforward. Definition 6.1.3. Let κ be a regular uncountable cardinal. A set S ⊆ κ is said to be stationary in κ if S ∩ C = ∅ for every closed unbounded set C in κ. Lemma 6.1.4. Let κ be a regular uncountable cardinal, and let S ⊆ κ be stationary in κ. Suppose S = i∈I Si where |I| < κ. Then Si is stationary for some i ∈ I. Proof. Suppose the conclusion fails. Then for each i ∈ I let Ci be a closed unbounded set such that Si ∩ Ci = ∅. By Lemma 6.1.2.1, C = i∈I Ci is a closed unbounded set. Since S is stationary, S ∩ C is nonempty, say α ∈ S ∩ C. Then for each i ∈ I we have α ∈ Si , a contradiction. / 114

Theorem 6.1.5 (Fodor’s Theorem). Let κ be a regular uncountable cardinal, and let S ⊆ κ be stationary in κ. Suppose f : S → κ is such that f (α) < α for all α ∈ S. Then f is constant on a stationary set. In other words, there exist a stationary S0 ⊆ S and a β0 < κ such that f (α) = β0 for all α ∈ S0 . Proof. Similar to the proof of the previous lemma, using 6.1.2.2 instead of 6.1.2.1. The details are left as an exercise for the reader. Theorem 6.1.6. For any regular uncountable cardinal κ, there exists a stationary set S ⊆ κ such that κ \ S is also stationary. Proof. . . . We state without proof the following theorem of Solovay. Theorem 6.1.7. Let κ be a regular uncountable cardinal. Any stationary set S ⊆ κ can be decomposed into κ pairwise disjoint stationary sets.

6.2

Large Cardinals

Definition 6.2.1 (hyperinaccessible cardinals). For each n < ω we define a class of cardinals called the n-hyperinaccessible cardinals. We define κ to be 0hyperinaccessible if it is inaccessible. We define κ to be n+1-hyperinaccessible if it is inaccessible and {λ < κ | λ is n-hyperinaccessible} is unbounded in κ. Definition 6.2.2 (Mahlo cardinals). For each n < ω we define a class of cardinals called the n-Mahlo cardinals. We define κ to be 0-Mahlo if it is inaccessible. We define κ to be n+1-Mahlo if it is inaccessible and {λ < κ | λ is n-Mahlo} is stationary in κ. Exercise 6.2.3. Show that n+1-hyperinaccessible implies n-hyperinaccessible. Show that n+1-Mahlo implies n-Mahlo. Show that 1-Mahlo implies n-hyperinaccessible for all n < ω. Lemma 6.2.4. Let δ be a limit ordinal. Suppose κ < δ and n < ω. Then κ is n-hyperinaccessible if and only if Rδ satisfies “κ is n-hyperinaccessible.” Also, κ is n-Mahlo if and only if Rδ satisfies “κ is n-Mahlo.” Proof. Straightforward. Theorem 6.2.5. 1. The existence of an n+1-hyperinaccessible cardinal is not provable in ZFC + “for all α there exists κ > α such that κ is n-hyperinaccessible” (assuming this theory is consistent).

115

where ∅ = X0 ⊆ I. 116 . if X ∈ F and X ⊆ Y ∈ P(I) then Y ∈ F.” 2. I = Rn . 1. . . If κ is n-Mahlo. ∩ Xn ∈ F. Let I be a nonempty set. 5. Such an F is called a principal filter . . If ZFC + “there exists an n-Mahlo cardinal” is consistent. then so is ZFC + V = L + “there exists an n-Mahlo cardinal. then so is ZFC + V = L + “there exists an n-hyperinaccessible cardinal. ∅ ∈ F and I ∈ F. F = {X ⊆ Rn | Rn \ X has Lebesgue measure 0}.e. If κ is a cardinal. Straightforward using the previous lemma. F = {I}. F = {X ⊆ I | |I \ X| < κ}. If ZFC + “there exists an n-hyperinaccessible cardinal” is consistent.2.. / 2.2. 6. if X1 . 4.6. Straightforward. If κ is a regular cardinal.2. Theorem 6. The existence of an n+1-Mahlo cardinal is not provable in ZFC + “for all α there exists κ > α such that κ is n-Mahlo” (assuming this theory is consistent). then L satisfies “κ is a cardinal. . If κ is n-hyperinaccessible. then L satisfies “κ is n-hyperinaccessible. then L satisfies “κ is n-Mahlo.3 Ultrafilters and Ultraproducts Definition 6.3. Proof.7. then L satisfies “κ is a regular cardinal.” 3. F = {X ⊆ I | X ⊇ X0 }. where κ is any infinite cardinal ≤ |I|. Here we could replace Rn by any measure space.3.2. A filter on I is a set F ⊆ P(I) such that 1. F = {X ⊆ I | X is cofinite. Straightforward using the previous lemma. . . 3. 1.” Proof. i.” 4. 3. Xn ∈ F then X1 ∩ . Examples 6.” 2. I \ X is finite} (assuming I is infinite). Lemma 6. 2. 1.1.” Proof.

3.2. For any infinite cardinal κ ≤ |I|. Put I = Pc (A) = {Y ⊆ A | Y is countable} .4. known as the closed unbounded filter on κ. the closed unbounded filter on κ is κ-additive. 1. F = {X ⊆ Rn | Rn \ X is meager}.6.3–8 are not ultrafilters. Here we could replace Rn by any complete metric space. Definition 6.3.3. it is difficult to find explicit examples of nonprincipal ultrafilters. Examples 6. as we shall now show.7.3. The filters in 6. Let κ be an infinite cardinal. Then Fc (A) = {X ⊆ Pc (A) | X ⊇ Cf for some f } is a filter known as the closed unbounded filter on Pc (A). |I| < κ.3. An ultrafilter on I is a filter U on I such that for all X ⊆ I either X ∈ U or I \ X ∈ U.e. Let I = κ where κ is a regular uncountable cardinal. f [Y <ω ] ⊆ Y } . A filter F is said to be κadditive if i∈I Xi ∈ F whenever Xi ∈ F for all i ∈ I. I = Rn . nonprincipal ultrafilters can be constructed by means of transfinite recursion plus the Axiom of Choice..3. i. Theorem 6. put Cf = {Y ∈ Pc (A) | Y is closed under f. For any regular uncountable cardinal κ. 117 . 2. Indeed. i. the filter {X ⊆ I | |I − X| < κ} is κ-additive. 3. The Lebesgue and Baire filters on Rn are countably additive. Given f : A<ω → A. i. For any uncountable set A.5.3.. ℵ1 additive.e. ℵ0 -additive. 5. Let A be an uncountable set. Then F = {X ⊆ κ | X ⊇ C for some closed unbounded set C ⊆ κ} is a filter. 4. Definition 6. 7.. Any filter F on I can be extended to an ultrafilter U on I. the closed unbounded filter on Pc (A) is countably additive. Remark 6. However.6.e. 8. Recall that A<ω is the set of finite sequences of elements of A. Every filter is finitely additive.

. . It follows easily that U is an ultrafilter. For every infinite set I there exists a nonprincipal ultrafilter U on I.p. Lemma 6. . since it is a filter. .) if Y1 ∩ .i.i. a contradiction.7.i.. i. let U be an ultrafilter on I such that F ⊆ U. By Theorem 6. .i. α ≤ κ. / X0 = {i0 } for some i0 ∈ I. Let U be a principal ultrafilter on I. Any principal ultrafilter U on I is of the form U = {X ⊆ I | i0 ∈ X} for some fixed i0 ∈ I. This proves the lemma.3. Thus U is nonprincipal. Then clearly Fα+1 has the f.. .p. Proof. . Stage α + 1.p.3.e. let Y ⊆ I be such that Y ∩ X0 = ∅ and (I \ Y ) ∩ X0 = ∅. Stage δ. Put Fδ = α<δ Fα . . Say that G ⊆ P(I) has the finite intersection property (f.i. .3. We claim that at least one of Fα ∪ {Xα }. .10. 118 . By the well-ordering theorem. Note that F has the f. Clearly U has the f..i. .p. ∩ Ym = ∅.i. Z1 . . . Thus |X0 | = 1..i. ∩ Zm = ∅.8. hence {i0 } ∈ U.3. say P(I) = {Xα | α < κ} . By definition we have U = {X ⊆ I | X ⊇ X0 } where ∅ = X0 ⊆ I.p. each of which has the f. Proof. . . Y1 . ∩ Ym = ∅ for all Y1 . . We therefore set Fα+1 = Fα ∪ {Xα } if this has the f.i. .9.i. . an ordered pair (A. If |X0 | ≥ 2. and for every X ∈ P(I) either X ∈ U or I \ X ∈ U. A structure is a relational structure. E) where A is a nonempty set and E ⊆ A × A. where δ is a limit ordinal.p. Consider the filter F = {X ⊆ I | I \X is finite}.p.Proof. Otherwise we would have Xα ∩ Y1 ∩ .p.p. / Definition 6. Fα ∪ {I \ Xα } has the f. We shall use transfinite recursion to define a sequence of sets Fα ⊆ P(I). . ∩ Ym ∩ Z1 ∩ . Ym ∈ G. Stage 0. Then Y. Clearly this has the f. Put F0 = F. This completes the proof. . . I \ Y ∈ U.p. Theorem 6. Zn ∈ Fα . ∩ Zn = ∅ so Fα does not have the f. For all i0 ∈ I we have I \ {i0} ∈ F ⊆ U. . let κ = |P(I)|. i.e. a contradiction. Then Y1 ∩ . (I \ Xα ) ∩ Z1 ∩ . Ym ∈ Fα . Assume inductively that Fα has the f. Fα ∪ {I \ Xα } otherwise. Finally put U = Fκ = α<κ Fα .

Finally. Recall that Ai = i∈I f f :I→ i∈I Ai . g(i)) ∈ Ei } ∈ U . g ∈ i∈I Ai define f ≈g ⇔def ⇔def f ≈U g {i ∈ I | f (i) = g(i)} ∈ U .Definition 6. . Theorem 6. for f. . . E) = (Ai . We define [f ] =def [f ]U =def and A = U g∈ i∈I Ai f ≈U g Ai i∈I i∈I = i∈I Ai U = [f ]U f ∈ Ai i∈I . . Then for all [f1 ]. . [fk ]) ⇔ {i ∈ I | |=(Ai . f ≈ f ′ . [g]) ∈ E imply ([f ′ ]. 6.4 6. and so (A. . . [g]) ∈ E ⇔def {i ∈ I | (f (i). . . xk . E) is a structure. Note that this last definition is independent of representatives. . [g ′ ]) ∈ E. g ∈ Ai . f (i) ∈ Ai for all i ∈ I . For f. . . xk ) be a formula with free variables among x1 . Suppose we are given an indexed family of structures (Ai . g ≈ g ′ . [fk ] ∈ A we have |=(A. .3. i. . fk (i))} ∈ U .E) ψ([f1 ]. This is an equivalence relation. . Ei ) i∈I be an ultras product. .. ([f ]. Thus E ⊆ A × A is well-defined.5 Measurable Cardinals Ramsey’s Theorem 119 .e.11 (ultraproduct). . . Let ψ(x1 .Ei ) ψ(f1 (i).3. . we define ([f ]. E) = U (Ai . Ei ) i∈I and an ultrafilter U on the index set I. Ei ) i∈I U known as an ultraproduct . We are going to define a structure (A.12 (Lo´’s Theorem). Let (A. .

85 singular. 13. 85. 97. 94. 97 additively indecomposable ordinal. 85. 106 hyperinaccessible. 49 consequence bounded logical. 75. 85 120 . 87 ∆0 predicate. 81–86 cases definition by. 86 strong limit. 119 s inaccessible. 101 axiom of choice. 114 arithmetical truth. 85 Mahlo. 10 consistency. 30. 85 successor. 97 least number operator. 117 arithmetical hierarchy. 85 weakly inaccessible. 49 propositional. 25. 9. 44 n Π0 predicate. 86 cardinal arithmetic. 84 ≃. 41–47. 94 connective axiom scheme. 84 ↓. 95 Boolean. 73 Church’s thesis. 75 ωα . 9. 115 regular. 101. 85 uncountable. 86 limit. 86 axioms complete. 59 cofinality. 46 of set theory. 48 Chinese remainder theorem. 90 additive. 73 Continuum Hypothesis. 115 Continuum Problem. 53 cardinal. 73 cardinal number. 103–107 Cantor’s theorem. 92–96 comprehension. 97 quantifier. 85 cardinal continuum hypothesis. 80 characteristic function. 39 Cauchy sequence. 71–74. 33 language of. 23 Lo´’s theorem. 23 ↑. 33.Index Rα . 9 arithmetic. 28. 78 ordinal. 41 n ℵα . 114 arithmetical definability. 117 CH. 41 n Σ0 predicate. 63 Ackermann function. 95. 117 ω. 48 class. 10 relative. 17–23 axiom of infinity. 23 κ-additive. 56 club. 49 Boolean connective. 107 constructible set. 106 computable function. 80 closed. 58. 79. 50–56. 97. 59 closed unbounded filter.

59 over the real number system. 44 limit cardinal. 92–96 least number operator. 80 linear ordering. 23 partial recursive. 60 over a relational structure. 114 divergent. 85 limit ordinal. 117 finite intersection property. 49 filter. 78 initial ordinal. 67 Def. 48. 97. 27 GCH. 97 convergent. 89 computable. 23 total. 86. 116 closed unbounded. 17–23 effective. 99 o language of arithmetic. 74 K¨nig’s Theorem. 118 falsity. 98 over the real number system.p. 61 limit-recursive. 75 logical consequence. 50–56. 107. 92. 57 of a program. 81–86 ordinal. 116 finitely additive. 27 induction transfinite. 24.continuum problem. 12 decidability. 70. 118 Fodor’s theorem. 66 defined. 30 recursive. 63 dense open set. 90 ordered field. 35 Hilbert’s 17th problem. 85. 107 generalized continuum hypothesis. 26. 28. 35. 30 model. 89 number systems. 97 Mahlo cardinal. 10 limit-recursive function.i. 115 minimization. 115 forcing. 89 ordering 121 . 117 course-of-values recursion. 101–103 index of a partial recursive function. 85. 23 primitive recursive. 23 G¨del number o of a formula. 82 isomorphism. 100 f. 92 function. 98 definability arithmetical. 23 countably additive. 48 of ordered rings. 37 Hilbert’s 10th problem. 108 formula. 85. 6–13. 26.. 6 partial. 90 extensionality. 28 equivalence relation. 74 o L. 106 halting problem. 115 inaccessible cardinal. 23 effective function. 60 ordered pair. 94. 106 L¨wenheim/Skolem theorem. 71–74. 97 number cardinal. 23 bounded. 117 principal. 107 diagonal intersection. 61 hyperinaccessible cardinal. 23 definition by cases. 60 of set theory. 75. 103. 61 enumeration theorem. 60. 44 number-theoretic.

23 reducible. 6 transfinite. 92. 85 successor ordinal. 92. 115 state. 78 transfinite recursion. 117 ultraproduct. 97. 80 successor. 116 principal function. 97. 43 program. 96. 74. 12 primitive. 78 recursion theorem. 114 uncountable cardinal. 39 recursively enumerable set. 48 total function. 49 bounded. 75. 9 primitive recursive function. 58. 66 real number system. 23. 23 universal Σ0 predicate. 98 parametrization theorem. 90 real world. 85 undecidability. 89 ordinal arithmetic. 74. 79. 93. 60. 99–104 singular cardinal. 58 undefined. 96 relative consistency. 87. 6–13. 93. 36 register machine program. 48 real number system. 95 sentence. 80 ordinal number. 89 parameter. 17 propositional connective. 75. 86 Skolem paradox. 23 partial recursive function. 49. 119 unbounded. 80 initial. 85 structure arithmetic. 17 regular cardinal. 45 n register machine program. 88 von Neumann ordinal. 81 122 . 28 unsolvable problem. 29 stationary. 102. 82 limit. 32 power set. 58 V . 44 recursive function. 89 weakly inaccessible cardinal. 101 predicate. 60. 80 term. 59 ultrafilter. 30 predicate. 75 well-ordering. 23 transfinite induction.linear. 107 replacement. 85 relational structure. 75. 80 von Neumann. 38 scheme. 99 Solovay. 96 successor cardinal. 93. 93. 96. 104 quantifier. 114 strong limit cardinal. 97 recursion course-of-values. 49 pure set. 101 Rice’s theorem. 92–96 language of. 99–103 transitive set. 10 quantifier elimination. 34–39. 95. 49 set theory axioms of. 60 relational. 99 truth arithmetical. 75 ordinal. 86 well-founded. 92–96 models of. 78 transitive model. 89 additively indecomposable. 36 partial function. 9 principal filter. 75 well. 75 well-ordering theorem. 87.

96 ZFC. 35 Zermelo/Fraenkel set theory. 96 123 . 96 ZF.word problem.

Sign up to vote on this title
UsefulNot useful