This action might not be possible to undo. Are you sure you want to continue?
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 WellOrderings 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 WellFounded Sets . . . . . . . . . . . . . . . . . . . . . . . 87
4.9 SetTheoretic 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 kfold Cartesian product
N . . . N
. .. .
k
is denoted N
k
. A kplace function is a function f : N
k
→ N and is sometimes
indicated with the lambdanotation,
f = λx
1
x
k
[ f(x
1
, . . . , x
k
) ] .
A numbertheoretic function is a kplace function for some k ≥ 1.
The purpose of this chapter is to deﬁne and study an important class of
numbertheoretic 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 numbertheoretic 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 numbertheoretic 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
kplace functions
λx
1
x
k
[ g
1
(x
1
, . . . , x
k
) ], . . . , λx
1
x
k
[ g
m
(x
1
, . . . , x
k
) ]
and the mplace function λy
1
y
m
[ h(y
1
, . . . , y
m
) ] all belong to C, then
the kplace 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
kplace 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 kplace predicate we mean a subset of N
k
. If
R ⊆ N
k
is a kplace 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 numbertheoretic
predicate is a kplace predicate for some k ≥ 1.
Deﬁnition 1.1.4. A kplace 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 2place 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., 1place 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 variablelength 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 2place numbertheoretic 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 1place numbertheoretic function f(n) given
by
f(n) = 1 +
n−1
¸
k=0
f(k)
n
is primitive recursive.
Solution. Consider the socalled courseofvalues function
¯
f(n) =
n−1
¸
k=0
p
f(k)
k
,
i.e.,
¯
f(n) encodes the variablelength 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 courseofvalues
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 1place numbertheoretic 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 wellknown 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 1place 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 kplace 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 2place function λnx[ A
n
(x) ] is not primitive recursive.
This is known as the Ackermann function.
5. Show that the 3place 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 1place function λx(A
x
(x)) is not primitive recursive.
(e) The 2place 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 1place
function A
x
(x) is not primitive recursive. It follows immediately that the
2place function A
x
(y) is not primitive recursive.
3. Show that the 3place 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 3place predicate A
x
(y) = z can be deﬁned
16
by courseofvalues 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 3place predicate A
x
(y) = z is primitive
recursive.
Applying the least number operator, we see that the 2place function
A
x
(y) is recursive. It follows immediately that the 1place function A
x
(x)
is recursive.
1.3 Computable Functions
In this section we deﬁne the class of computable (i.e., recursive) numbertheoretic
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 kplace numbertheoretic 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 2place numbertheoretic 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 kplace 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 kplace function is just what we have previously called
a kplace 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 kplace (total) func
tion is said to be recursive if and only if it is computable. A kplace predicate
is said to be recursive if its characteristic function is recursive.
Deﬁnition 1.4.2 (Partial Recursive Functions). A kplace 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 kplace 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
toone 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) =
2π
−1
B
(y) if y ∈ B,
2π
−1
N\B
(y) + 1 if y ∈ N ` B.
By courseofvalues 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 kplace 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 1place 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 kplace 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 kplace 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 kplace primitive re
cursive predicates and let f
1
and f
2
be kplace primitive recursive functions.
Assume that P
1
and P
2
are mutually exclusive and exhaustive, i.e., for each
ktuple '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 kplace 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 kplace 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+1place total recursive function
Φ
k
: N
k+1
→N with the following properties:
1. for each e ∈ N, the kplace function λx
1
. . . x
k
[ Φ
k
(e, x
1
, . . . , x
k
) ] is prim
itive recursive;
2. for each kplace 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 kplace partial recursive function ψ(x
1
, . . . , x
k
), show
that there is a 1place 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 kplace 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 3place predicate z = A
x
(y) is prim
itive recursive, even though the 2place 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+1ary 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
hardcoded as the ﬁrst argument
of θ.
Formally, let T
′
be a register machine program which computes the k+1ary
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 kary
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
2place 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 1place 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
(ϕ
(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 onetoone, 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 onetoone.
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 1place 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 numbertheoretic 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 kplace 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+1place 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 kplace predicate P ⊆ N
k
belongs to the class Σ
0
1
if and
only if P = dom(ψ) for some kplace 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 kplace 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 kplace 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 onetoone 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
(π
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 onetoone 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 oneto
one recursive function. Putting f(n) = (π
B
(n))
0
, we see that f is a onetoone
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 onetoone
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 kplace 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 limitrecursive 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 limitrecursive.
Solution. For simplicity, let x be an abbreviation for x
1
, . . . , x
k
.
First assume that χ
P
is limitrecursive, 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 limitrecursive. 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+1place predicate belonging to the class Σ
0
n
; and
2. for any kplace 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 nonreducibility 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 nonnegative integers, +
and denote the 2place operations of addition and multiplication on N, and =
denotes the 2place 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 2place 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 nonnegative 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
nonnegative 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 nonnegative 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 nonnegative 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 kplace 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 1place 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 2place 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 numbertheoretic 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 numbertheoretic 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 kplace 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+1place predicate R(x
1
, . . . , x
k
, y) is arithmetically
deﬁnable, then so is the kplace 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 kplace 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+1place 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
nonnegative integers r
i
< m
i
, 1 ≤ i ≤ k, we can ﬁnd a nonnegative 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 nonnegative 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 nonnegative
integers less than m have distinct remainder sequences. To see this, let r and
s be nonnegative 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
nonnegative 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 nonnegative integers r, a, i, we deﬁne
β(r, a, i) = Remainder(r, a (i + 1) + 1) .
This 3place function is known as G¨odel’s βfunction. Note that the βfunction
is arithmetically deﬁnable.
The signiﬁcance of the βfunction is that it can be used to encode an ar
bitrary sequence of nonnegative 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 nonnegative integers r
0
, r
1
, . . . , r
k
,
we can ﬁnd a pair of nonnegative 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 nonnegative 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
nonnegative integer r such that . . . ,” but it does not allow us to directly say
things like “there exists a sequence of nonnegative integers r
0
, r
1
, . . . , r
x
(of
variable length, x) such that . . . .” The way to overcome this diﬃculty is to
use the βfunction. Instead of saying “there exists a ﬁnite sequence r
0
, r
1
, . . . ,
r
k
,” we can say “there exists a pair of nonnegative integers r and a which
encode the required sequence, via the βfunction.” 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 βfunction. 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 3place 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 numbertheoretic 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 leastnumber 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 leastnumber 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 nonrecursive set. By the Enumeration Theorem,
the 2place 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 kplace 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 wellknown 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 2ary operation symbols, − is a 1ary operation symbol, <
and = are 2place 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 kplace 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 quantiﬁerfree 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 semialgebraic sets.
Thus “eﬀective = semialgebraic 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 1place func
tions. If D(y) is a quantiﬁerfree 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 quantiﬁerfree
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 variablefree 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
noncollinear 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 noncollinear 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 limitationofsize 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 limitationofsize 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 oneelement
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 ntuple '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 onetoone 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 onetoone correspondence between X and Y ,
i.e., a onetoone 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 WellOrdering Theorem. Part 4 is known as the
CantorSchroederBernstein 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 onetoone function f : X → {(X) where
f(a) = ¦a¦. Suppose now that {(X) X holds. Let g : {(X) → X be
onetoone. 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 CantorSchroeder
Bernstein Theorem, prove that
ℵ
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 WellOrderings 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 onetoone 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 wellfounded 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 Rminimal element of X.
Lemma 4.3.5. A relational structure (A, R) is wellfounded if and only if there
is no inﬁnite descending Rsequence. (By an inﬁnite descending Rsequence 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 Rsequence, then X = ¦a
n
[ n ∈ N¦
is a counterexample to wellfoundedness of (A, R). Conversely, suppose that
X ⊆ A is a counterexample to wellfoundedness. 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
Rsequence. 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 wellordering is a linear ordering which is wellfounded.
Note that if (A, R) is a wellordering and X is a nonempty subset of A, then
X has an Rleast 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 wellordering.
For n ∈ N, we identify n with the ordinal number which is the order type of
all nelement wellorderings. 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 WellOrderings). Let (A, R) and (B, S) be
wellorderings. 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 Rleast 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
Rleast element of A ` dom(f), and let b be the Sleast 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 Rleast 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 Sleast 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 wellorderings. 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 wellordering and B ⊆ A, then (B, R∩(B B))
is a wellordering, and
type(B, R ∩ (B B)) ≤ type(A, R) .
Proof. Straightforward, using comparability of wellorderings.
The next theorem implies that any wellordering 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 wellordering of type α.
Proof. Let (A, R) be some ﬁxed wellordering 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 wellordering. Let γ be the type of this wellordering. 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 wellordered under <, hence X has a least element.
Theorem 4.3.21 (BuraliForti 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 wellordering, 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 BuraliForti 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 Rleast 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
Sgreatest 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 (WellOrdering Theorem). Given a set X, we can ﬁnd a rela
tion R ⊆ X X such that (X, R) is a wellordering.
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 onetoone 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 onetoone, 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 WellOrdering Theorem. There is also a converse: the WellOrdering 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 WellOrdering Theorem,
there exists R ⊆ A A such that (A, R) is a wellordering. Deﬁne 'a
i
`
i∈I
by
putting a
i
= the Rleast element of X
i
. Thus 'a
i
`
i∈I
∈
¸
i∈I
X
i
and we have
proved the Axiom of Choice from the WellOrdering 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 WellOrdering 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
wellordering 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
(α
′
, 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 wellordering.
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 wellorderings, 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
ℵ
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
ℵ
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 WellFounded 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 wellfounded if the relational structure
(TC(X), ∈[TC(X)) is wellfounded.
Applying Lemma 4.3.5 to the relational structure (TC(X), ∈[TC(X)), we see
that X is wellfounded 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, wellfounded set.
Proof. It is straightforward to prove by transﬁnite induction on α that all el
ements of R
α
are pure and wellfounded. Conversely, suppose X is pure and
wellfounded. 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, wellfounded sets is denoted V . Thus we have
V =
¸
α∈Ord
R
α
.
If X is a pure, wellfounded 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 wellfounded 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 SetTheoretic 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,
wellfounded 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, wellfounded sets.
For example, the natural numbers 0, 1, 2, . . . are not ordinarily regarded as
being sets, but within the universe of pure, wellfounded 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, wellfounded
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, wellfounded 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, wellfounded 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 singlevalued settheoretic
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 singlevalued,
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 settheoretic reconstruction of the ordinal numbers, due to von
Neumann, is as follows:
Deﬁnition 4.9.4. A von Neumann ordinal is a transitive, pure, wellfounded
set A such that (A, ∈[A) is a wellordering.
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, wellfounded
sets.
89
Finally, we turn to the settheoretic 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, wellfounded sets. In this
sense, pure set theory may be said to encompass virtually all of mathematics,
and one may speak of the settheoretic 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, wellfounded 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, wellfounded sets:
two pure, wellfounded sets are equal if and only if they contain the same pure,
wellfounded 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, wellfounded 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 ZermeloFraenkel 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 selfexplanatory. 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 wellfounded. 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 ZermeloFraenkel set theory.
Actually, these two socalled 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, settheoretic 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 socalled real world,
i.e., the class of pure, wellfounded sets. However, the general notion of pure,
wellfounded 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, wellfounded 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 settheoretic 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 settheoretic 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 kplace 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 settheoretic
notions).
A straightforward generalization of Theorem 5.2.9 is:
Theorem 5.2.11 (Generalized L¨ owenheimSkolem 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, wellfounded 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 wellfounded 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 wellfounded
and extensional. Then there exists a transitive, pure, wellfounded 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, wellfounded 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, wellfounded 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 wellfounded 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, wellfounded set. A kplace
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, wellfounded 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, wellfounded 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, wellfounded 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¨ owenheimSkolem 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, wellfounded, 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, wellfounded 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, wellfounded set,
and L
α
⊆ R
α
.
Proof. For any transitive, pure, wellfounded set T, we have T ⊆ Def(T) ⊆ {(T)
and hence Def(T) is again a transitive, pure, wellfounded 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, wellfounded 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, wellfounded 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, wellfounded 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, wellfounded 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 wellorderable. 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 wellordering 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 wellorderable. 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 Mgeneric if G∩ D = ∅ for all
dense open D ⊆ P such that D ∈ M.
Lemma 5.5.4. Given p ∈ P we can ﬁnd an Mgeneric 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 Mgeneric ﬁlter.
Deﬁnition 5.5.5. Let G be an Mgeneric 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 Mgeneric 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
Mgeneric 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 Mgeneric 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 Mgeneric 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 Mgeneric 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 Mgeneric 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 nhyperinaccessible cardinals. We deﬁne κ to be 0
hyperinaccessible if it is inaccessible. We deﬁne κ to be n+1hyperinaccessible
if it is inaccessible and
¦λ < κ [ λ is nhyperinaccessible¦
is unbounded in κ.
Deﬁnition 6.2.2 (Mahlo cardinals). For each n < ω we deﬁne a class of cardi
nals called the nMahlo cardinals. We deﬁne κ to be 0Mahlo if it is inaccessible.
We deﬁne κ to be n+1Mahlo if it is inaccessible and ¦λ < κ [ λ is nMahlo¦ is
stationary in κ.
Exercise 6.2.3. Show that n+1hyperinaccessible implies nhyperinaccessible.
Show that n+1Mahlo implies nMahlo. Show that 1Mahlo implies nhyperinaccessible
for all n < ω.
Lemma 6.2.4. Let δ be a limit ordinal. Suppose κ < δ and n < ω. Then κ is
nhyperinaccessible if and only if R
δ
satisﬁes “κ is nhyperinaccessible.” Also,
κ is nMahlo if and only if R
δ
satisﬁes “κ is nMahlo.”
Proof. Straightforward.
Theorem 6.2.5.
1. The existence of an n+1hyperinaccessible cardinal is not provable in ZFC
+ “for all α there exists κ > α such that κ is nhyperinaccessible” (as
suming this theory is consistent).
115
2. The existence of an n+1Mahlo cardinal is not provable in ZFC + “for
all α there exists κ > α such that κ is nMahlo” (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 nhyperinaccessible, then L satisﬁes “κ is nhyperinaccessible.”
4. If κ is nMahlo, then L satisﬁes “κ is nMahlo.”
Proof. Straightforward.
Theorem 6.2.7.
1. If ZFC + “there exists an nhyperinaccessible cardinal” is consistent, then
so is ZFC + V = L + “there exists an nhyperinaccessible cardinal.”
2. If ZFC + “there exists an nMahlo cardinal” is consistent, then so is ZFC
+ V = L + “there exists an nMahlo 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
<ω
is the set of ﬁnite sequences of elements of A. Given
f : A
<ω
→ A, put
C
f
= ¦Y ∈ {
c
(A) [ Y is closed under f, i.e., f[Y
<ω
] ⊆ 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
κadditive.
4. For any regular uncountable cardinal κ, the closed unbounded ﬁlter on κ
is κadditive.
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 wellordering 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 welldeﬁned, 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
ℵ
α
, 84
↓, 23
κadditive, 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
courseofvalues 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
limitrecursive, 44
numbertheoretic, 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
limitrecursive 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
courseofvalues, 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
wellfounded, 75
wellordering, 75
wellordering 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 WellOrderings 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 WellFounded Sets . . . . . . . . 4.9 SetTheoretic 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 kfold Cartesian product N× . For example. × N k is denoted Nk . . . The purpose of this chapter is to deﬁne and study an important class of numbertheoretic 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 kplace function is a function f : Nk → N and is sometimes indicated with the lambdanotation. A numbertheoretic function is a kplace 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 numbertheoretic 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 numbertheoretic 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 kplace 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 mplace 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 kplace 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 kplace 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 kplace 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 kplace predicate we mean a subset of Nk . . . z1 . If R ⊆ Nk is a kplace 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 numbertheoretic predicate is a kplace predicate for some k ≥ 1. z1 . the 2place 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. 1place 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 variablelength 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 courseofvalues recursion. 4 3 5 7 9 2n + 1 n=0 12 ∞ n−1 e . We use the wellknown 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 1place numbertheoretic 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 2place numbertheoretic 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 variablelength 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 1place numbertheoretic 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 socalled courseofvalues 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 1place 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 kplace 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 2place 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 3place 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 1place 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 2place 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 1place 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 2place 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 3place 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 3place 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 1place 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 3place 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) numbertheoretic 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 2place 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 courseofvalues 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 kplace numbertheoretic 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 2place numbertheoretic 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 kplace (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 kplace partial function ψ is said to be partial recursive if it is computed by some register machine program P. . for all x1 . . A kplace predicate is said to be recursive if its characteristic function is recursive. . xk . Then the kplace 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 kplace function is just what we have previously called a kplace function.4 Partial Recursive Functions A kplace 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 courseofvalues recursion. then the inverse permutation f −1 is also recursive. although the onetoone 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 kplace 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 1place 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 kplace 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 kplace 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 kplace function f deﬁned by f (x1 . xk [ ϕ(k) (x1 . Proof of the Enumeration Theorem. . . . i. for each ktuple 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 kplace primitive recursive predicates and let f1 and f2 be kplace 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 kplace 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 1place 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 kplace 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+1place 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 kplace 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 kplace 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 3place predicate z = Ax (y) is primitive recursive. show that K0 and K1 are recursively inseparable. Exercise 1. a kplace 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 2place 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+1ary 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+1ary partial recursive function θ. x0 ) ].12 computes the kary 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 hardcoded 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 2place 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 1place 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 kplace 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 1place 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 onetoone. 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 onetoone. . . For n ≥ 1. By the Recursion (1) Theorem. y))1 ≃ z. . y) where R is a k+1place 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 numbertheoretic 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 kplace predicate P ⊆ Nk belongs to the class Σ0 if and 1 P only if P = dom(ψ) for some kplace 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 kplace 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 onetoone function πA : N → N that enumerates the elements of A in increasing order. y) there exists 1 a kplace 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 onetoone 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 onetoone 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 onetoone recursive function. . . A kplace 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 onetoone 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 limitrecursive. z) where R1 and R0 are primitive recursive predicates. . Thus χP is limitrecursive. 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+1place predicate belonging to the class Σ0 . g(n. Theorem 1. y. . for any kplace 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 limitrecursive 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 limitrecursive. . 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 nonreducibility 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 2place 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 nonnegative 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 nonnegative 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 2place 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 2place 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 nonnegative 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 nonnegative 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 1place 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 kplace partial function λx1 · · · xk [ ψ(x1 .) Exercise 2. . mk . xk . x) ] and λxy [ Remainder(y. +. 2. .1 (Arithmetical Deﬁnability). . . . . n ∈ N.. . note that the 2place 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 nonnegative 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 numbertheoretic 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 numbertheoretic 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 kplace 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+1place 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+1place 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 kplace 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 kplace 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 nonnegative 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 nonnegative 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 nonnegative integer r such that Remainder(r. Then for any given nonnegative 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 nonnegative 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 nonnegative 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 nonnegative integer r such that . w) which says. For any nonnegative 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 nonnegative integers r0 .” Namely. . By Lemma 2. w) respectively. . .2. . Remainder(r.” The way to overcome this diﬃculty is to use the βfunction. 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 βfunction o is arithmetically deﬁnable. a · (i + 1) + 1) . .12 we can ﬁnd a nonnegative 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 βfunction is that it can be used to encode an arbitrary sequence of nonnegative 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 nonnegative 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 βfunction. 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 nonnegative 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 3place function is known as G¨del’s βfunction. . . . Given a ﬁnite sequence of nonnegative 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 βfunction. 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 leastnumber 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 leastnumber 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 numbertheoretic 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 3place 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 2place 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 nonrecursive set. Let P be a kplace 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 wellknown 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 2ary 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 2place 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 kplace 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 1ary 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 quantiﬁerfree 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 semialgebraic 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 = semialgebraic 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 1place functions. √ Example 3. If D(y) is a quantiﬁerfree 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 quantiﬁerfree 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 variablefree 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 noncollinear 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 noncollinear 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 limitationofsize 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 limitationofsize 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 oneelement 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 ntuple 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 onetoone 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 onetoone correspondence between X and Y , i.e., a onetoone 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 WellOrdering Theorem. Part 4 is known as the CantorSchroederBernstein 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 onetoone 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 onetoone. 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 onetoone 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 CantorSchroederBernstein 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 WellOrderings 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 wellfoundedness 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 wellfounded if for every nonempty set X ⊆ A. A linear ordering is a relational structure (A. α · β = type(A × B. then X has an Rleast 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 Rminimal element of X.4. R) is wellfounded if and only if there is no inﬁnite descending Rsequence.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 wellordering.3. we can associate to any relational structure (A. R) is a wellordering 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 wellfoundedness. This is an inﬁnite descending Rsequence. 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 Rsequence.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 nelement wellorderings.) Proof. in such a way that (A. S) be ordinal numbers. Straightforward.e. A wellordering is a linear ordering which is wellfounded. (By an inﬁnite descending Rsequence 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 wellorderings. 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 Rleast 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 WellOrderings). 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 Rleast 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 Sleast 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 wellorderings. S). R) onto ({β  β < α}. and β < α holds.3. R ∩ (B × B)) is a wellordering. 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 wellordering 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 Rleast element of A \ dom(f ).12. {(b′′. Proof. The ﬁrst part follows from comparability of wellorderings. then letting b be the Sleast element of B \ rng(f ).3. If rng(f ) = B.16. Let (A. {(b′′. 4. The claim is proved. R) is a wellordering and B ⊆ A. where α = type(A. β)  γ < β < α}) is a wellordering of type α. If = dom(f ) = A. {(γ. R) be some ﬁxed wellordering 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 (BuraliForti 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 wellordering. β) ∈ A × A  α < β}) is a wellordering.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 wellordering. {(γ.. We have seen this in connection with the Russell Paradox and the BuraliForti Paradox. we see that (Ord. The latter set of ordinal numbers is wellordered 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 Sgreatest 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 Rleast 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 onetoone 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 onetoone. R) is a wellordering. 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 (WellOrdering 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 Rleast 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 WellOrdering 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 WellOrdering 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 wellordering. Remark 4.5. Thus ai i∈I ∈ i∈I Xi and we have proved the Axiom of Choice from the WellOrdering theorem. Immediate from Lemma 4. V ∈ C either U ⊆ V or V ⊆ U . and X Y if and only if X ≤ Y  . By the WellOrdering 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 WellOrdering 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 wellordering. From comparability of wellorderings. 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 wellordering 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 wellfounded. 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 wellfounded 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 WellFounded 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 wellfounded. 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 wellfounded if the relational structure (TC(X). ∈TC(X)) is wellfounded.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. wellfounded 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. wellfounded sets is denoted V . real number. If X is a pure. ordinal numbers.2. We begin with ordered pairs and progress to functions. wellfounded sets. function. Note that if a and b are pure. wellfounded 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 wellfounded 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. wellfounded 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 SetTheoretic Foundations In the next chapter we shall begin the study of axiomatic set theory.The class of all pure. wellfounded 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). wellfounded 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. wellfounded set. Thus. let us write (a. Lemma 4. and real numbers. wellfounded 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}. wellfounded 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. wellfounded 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 settheoretic 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 singlevalued settheoretic operations. Moreover.6. the initial segment B = {a  a ∈ b} is again a von Neumann ordinal. which is singlevalued.} = 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 wellordering.
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 settheoretic 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. wellfounded 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 settheoretic 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. wellfounded 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. wellfounded 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. wellfounded 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. wellfounded 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 . wellfounded 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 ZermeloFraenkel 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 wellfounded.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 selfexplanatory.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 socalled 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 ZermeloFraenkel 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, settheoretic 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 socalled real world, i.e., the class of pure, wellfounded sets. However, the general notion of pure, wellfounded 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, wellfounded 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 settheoretic 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 settheoretic 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 kplace 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) = 2A . 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 settheoretic 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. wellfounded 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¨wenheimSkolem 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 wellfounded 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 wellfounded and extensional. wellfounded 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 wellfounded and extensional. It is straightforward to verify that T and f are unique. Fix an object a0 ∈ A.3. wellfounded set. pure. 2. ∈T ) onto (A. ∈T ). wellfounded set and that F ↾T is an isomorphism of (T. ∈T ) are unique. A kplace 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. wellfounded 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. wellfounded 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. wellfounded 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. wellfounded 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. wellfounded. ∈A) is an elementary submodel of (Rλ . For all ordinals α. By the L¨wenheimSkolem 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. wellfounded 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. wellfounded 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. wellfounded 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. wellfounded 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. wellfounded 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 . wellfounded 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 . wellfounded 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 ))} . wellfounded 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 wellorderable. 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 wellordering 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 wellorderable.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 nhyperinaccessible cardinals. We deﬁne κ to be 0hyperinaccessible if it is inaccessible. We deﬁne κ to be n+1hyperinaccessible if it is inaccessible and {λ < κ  λ is nhyperinaccessible} is unbounded in κ. Deﬁnition 6.2.2 (Mahlo cardinals). For each n < ω we deﬁne a class of cardinals called the nMahlo cardinals. We deﬁne κ to be 0Mahlo if it is inaccessible. We deﬁne κ to be n+1Mahlo if it is inaccessible and {λ < κ  λ is nMahlo} is stationary in κ. Exercise 6.2.3. Show that n+1hyperinaccessible implies nhyperinaccessible. Show that n+1Mahlo implies nMahlo. Show that 1Mahlo implies nhyperinaccessible for all n < ω. Lemma 6.2.4. Let δ be a limit ordinal. Suppose κ < δ and n < ω. Then κ is nhyperinaccessible if and only if Rδ satisﬁes “κ is nhyperinaccessible.” Also, κ is nMahlo if and only if Rδ satisﬁes “κ is nMahlo.” Proof. Straightforward. Theorem 6.2.5. 1. The existence of an n+1hyperinaccessible cardinal is not provable in ZFC + “for all α there exists κ > α such that κ is nhyperinaccessible” (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 nMahlo. ∩ Xn ∈ F. Let I be a nonempty set. 5. Such an F is called a principal ﬁlter . . If ZFC + “there exists an nMahlo cardinal” is consistent. then so is ZFC + V = L + “there exists an nMahlo cardinal. then so is ZFC + V = L + “there exists an nhyperinaccessible 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 nhyperinaccessible 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+1Mahlo cardinal is not provable in ZFC + “for all α there exists κ > α such that κ is nMahlo” (assuming this theory is consistent). then L satisﬁes “κ is a cardinal. . If κ is nhyperinaccessible. then L satisﬁes “κ is nhyperinaccessible. then L satisﬁes “κ is nMahlo.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 κadditive. 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 κadditive. 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 wellordering 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 welldeﬁned.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 κadditive. 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 limitrecursive. 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 limitrecursive function.i. 115 minimization. 115 forcing. 89 ordering 121 . 117 courseofvalues 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 numbertheoretic.
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 wellordering. 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 courseofvalues. 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 wellfounded. 92–96 models of. 78 transitive model. 89 additively indecomposable. 36 partial function. 9 principal ﬁlter. 75 well. 75 wellordering theorem. 87.
96 ZFC. 35 Zermelo/Fraenkel set theory. 96 123 . 96 ZF.word problem.