Copyright c _19982003 by Stephen G.

Simpson
Mathematical Logic
Stephen G. Simpson
January 7, 2004
Department of Mathematics
The Pennsylvania State University
University Park, State College PA 16802
http://www.math.psu.edu/simpson/
This is a set of lecture notes for introductory courses in mathematical logic
oered at the Pennsylvania State University.
Contents
Contents 1
1 Propositional Calculus 3
1.1 Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Assignments and Satisability . . . . . . . . . . . . . . . . . . . . 5
1.3 Logical Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 The Tableau Method . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 The Completeness Theorem . . . . . . . . . . . . . . . . . . . . . 14
1.6 Trees and K onigs Lemma . . . . . . . . . . . . . . . . . . . . . . 15
1.7 The Compactness Theorem . . . . . . . . . . . . . . . . . . . . . 16
1.8 Combinatorial Applications . . . . . . . . . . . . . . . . . . . . . 17
2 Predicate Calculus 19
2.1 Formulas and Sentences . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Structures and Satisability . . . . . . . . . . . . . . . . . . . . . 21
2.3 The Tableau Method . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 Logical Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5 The Completeness Theorem . . . . . . . . . . . . . . . . . . . . . 30
2.6 The Compactness Theorem . . . . . . . . . . . . . . . . . . . . . 35
2.7 Satisability in a Domain . . . . . . . . . . . . . . . . . . . . . . 36
3 Proof Systems for Predicate Calculus 38
3.1 Introduction to Proof Systems . . . . . . . . . . . . . . . . . . . . 38
3.2 The Companion Theorem . . . . . . . . . . . . . . . . . . . . . . 39
3.3 A Hilbert-Style Proof System . . . . . . . . . . . . . . . . . . . . 42
3.4 Gentzen-Style Proof Systems . . . . . . . . . . . . . . . . . . . . 45
3.5 The Interpolation Theorem . . . . . . . . . . . . . . . . . . . . . 49
4 Extensions of Predicate Calculus 53
4.1 Predicate Calculus with Identity . . . . . . . . . . . . . . . . . . 53
4.2 The Spectrum Problem . . . . . . . . . . . . . . . . . . . . . . . 56
4.3 Predicate Calculus With Operations . . . . . . . . . . . . . . . . 58
4.4 Predicate Calculus with Identity and Operations . . . . . . . . . 61
4.5 Many-Sorted Predicate Calculus . . . . . . . . . . . . . . . . . . 63
1
5 Theories, Models, Denability 67
5.1 Theories and Models . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.2 Mathematical Theories . . . . . . . . . . . . . . . . . . . . . . . . 69
5.3 Foundational Theories . . . . . . . . . . . . . . . . . . . . . . . . 74
5.4 Denability over a Structure . . . . . . . . . . . . . . . . . . . . . 74
5.5 Denibility over a Theory . . . . . . . . . . . . . . . . . . . . . . 75
5.6 Interpretability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6 Arithmetization of Predicate Calculus 76
6.1 Primitive Recursive Arithmetic . . . . . . . . . . . . . . . . . . . 76
6.2 Interpretability of PRA in Z
1
. . . . . . . . . . . . . . . . . . . . 76
6.3 G odel Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.4 Undenability of Truth . . . . . . . . . . . . . . . . . . . . . . . . 79
6.5 The Provability Predicate . . . . . . . . . . . . . . . . . . . . . . 80
6.6 The Incompleteness Theorems . . . . . . . . . . . . . . . . . . . . 81
6.7 Proof of Lemma 6.5.3 . . . . . . . . . . . . . . . . . . . . . . . . 82
Bibliography 83
Index 84
2
Chapter 1
Propositional Calculus
1.1 Formulas
Denition 1.1.1. The propositional connectives are negation (), conjunction
( &), disjunction ( ), implication ( ), biimplication ( ). They are read as
not, and, or, if-then, if and only if respectively. The connectives &,
, , are designated as binary, while is designated as unary.
Denition 1.1.2. A propositional language L is a set of propositional atoms
p, q, r, . . .. An atomic L-formula is an atom of L.
Denition 1.1.3. The set of L-formulas is generated inductively according to
the following rules:
1. If p is an atomic L-formula, then p is an L-formula.
2. If A is an L-formula, then (A) is an L-formula.
3. If A and B are L-formulas, then (A&B), (AB), (AB), and (AB)
are L-formulas.
Note that rule 3 can be written as follows:
3

. If A and B are L-formulas and b is a binary connective, then (Ab B) is an

L-formula.
Example 1.1.4. Assume that L contains propositional atoms p, q, r, s. Then
(((p q) &(q r)) (p r)) (q s)
is an L-formula.
Denition 1.1.5. If A is a formula, the degree of A is the number of occurrences
of propositional connectives in A. This is the same as the number of times rules
2 and 3 had to be applied in order to generate A.
3
Example 1.1.6. The degree of the formula of Example 1.1.4 is 8.
Remark 1.1.7. As in the above example, we omit parentheses when this can
be done without ambiguity. In particular, outermost parentheses can always be
omitted, so instead of ((A) B) we may write (A) B. But we may not
write AB, because this would not distinguish the intended formula from
(AB).
Denition 1.1.8. Let L be a propositional language. A formation sequence is
nite sequence A
1
, A
2
, . . . , A
n
such that each term of the sequence is obtained
from previous terms by application of one of the rules in Denition 1.1.3. A
formation sequence for A is a formation sequence whose last term is A. Note
that A is an L-formula if and only if there exists a formation sequence for A.
Example 1.1.9. A formation sequence for the L-formula of Example 1.1.4 is
p, q, p q, r, q r, (p q) &(q r), p r, ((p q) &(q r)) (p r),
s, q s, (q s), (((p q) &(q r)) (p r)) (q s) .
Remark 1.1.10. In contexts where the language L does not need to be speci-
ed, an L-formula may be called a formula.
Denition 1.1.11. A formation tree is a nite rooted dyadic tree where each
node carries a formula and each non-atomic formula branches to its immediate
subformulas (see the example below). If A is a formula, the formation tree for
A is the unique formation tree which carries A at its root.
Example 1.1.12. The formation tree for the formula of Example 1.1.4 is
(((p q) &(q r)) (p r)) (q s)
/
((p q) &(q r)) (p r)
/
(p q) &(q r)
/
p q q r
/ /
p q q r
p r
/
p r
(q s)
[
q s
/
q s
or, in an abbreviated style,

_ _

/
&
/ /
p r
/ /
p q q r

/
q s
4
Remark 1.1.13. Note that, if we identify formulas with formation trees in the
abbreviated style, then there is no need for parentheses.
Remark 1.1.14. Another way to avoid parentheses is to use Polish notation.
In this case the set of L-formulas is generated as follows:
1. If p is an atomic L-formula, then p is an L-formula.
2. If A is an L-formula, then A is an L-formula.
3. If A and B are L-formulas and b is a binary connective, then b AB is an
L-formula.
For example, (p) q becomes p q, and (p q) becomes p q. The
formula of Example 1.1.4 becomes
& p q q r p r q s
and a formation sequence for this is
p, q, p q, r, q r, & p q q r, p r, & p q q r p r,
s, q s, q s, & p q q r p r q s .
Obviously Polish notation is dicult to read, but it has the advantages of being
linear and of not using parentheses.
Remark 1.1.15. In our study of formulas, we shall be indierent to the ques-
tion of which system of notation is actually used. The only point of interest for
us is that each non-atomic formula is uniquely of the form A or Ab B, where
A and B are formulas and b is a binary connective.
1.2 Assignments and Satisability
Denition 1.2.1. There are two truth values, T and F, denoting truth and
falsity.
Denition 1.2.2. Let L be a propositional language. An L-assignment is a
mapping
M : p : p is an atomic L-formula T, F .
Note that if L has exactly n atoms then there are exactly 2
n
dierent L-
assignments.
Lemma 1.2.3. Given an L-assignment M, there is a unique L-valuation
v
M
: A : A is an L-formula T, F
given by the following clauses:
5
1. v
M
(A) =
_
T if v
M
(A) = F,
F if v
M
(A) = T.
2. v
M
(A&B) =
_
T if v
M
(A) = v
M
(B) = T,
F if at least one of v
M
(A), v
M
(B) = F.
3. v
M
(AB) =
_
T if at least one of v
M
(A), v
M
(B) = T,
F if v
M
(A) = v
M
(B) = F.
4. v
M
(AB) = v
M
((A&B)) .
5. v
M
(AB) =
_
T if v
M
(A) = v
M
(B) ,
F if v
M
(A) ,= v
M
(B) .
Proof. The truth value v
M
(A) is dened by recursion on L-formulas, i.e., by
induction on the degree of A where A is an arbitrary L-formula.
Remark 1.2.4. Note that each clause of Lemma 1.2.3 corresponds to the fa-
miliar truth table for the corresponding propositional connective. Thus clause
3 corresponds to the truth table
A B AB
T T T
T F T
F T T
F F F
for , and clause 4 corresponds to the truth table
A B AB
T T T
T F F
F T T
F F T
for .
Remark 1.2.5. Lemma 1.2.3 may be visualized in terms of formation trees.
To dene v
M
(A) for a formula A, one begins with an assignment of truth values
to the atoms, i.e., the end nodes of the formation tree for A, and then proceeds
upward to the root, assigning truth values to the nodes, each step being given
by the appropriate clause.
Example 1.2.6. Consider the formula (p q) (q r) under an assignment
M with M(p) = T, M(q) = F, M(r) = T. In terms of the formation tree, this
6
looks like
(p q) (q r)
/
p q q r
/ /
p q q r
T F F T
and by applying clause 4 three times we get
(p q) (q r)
T
_ _
p q q r
F T
/ /
p q q r
T F F T
and from this we see that v
M
((p q) (q r)) = T.
Remark 1.2.7. The above formation tree with truth values can be compressed
and written linearly as
(p q) (q r)
T F F T F T T .
This illustrates a convenient method for calculating v
M
(A), where M is an
arbitrary L-assignment.
Remark 1.2.8. Lemma 1.2.3 implies that there is an obvious one-to-one cor-
respondence between L-assignments and L-valuations. If the language L is
understood from context, we may speak simply of assignments and valuations.
We now present some key denitions. Fix a propositional language L.
Denition 1.2.9. Let M be an assignment. A formula A is said to be true
under M if v
M
(A) = T, and false under M if v
M
(A) = F.
Denition 1.2.10. A set of formulas S is said to be satisable if there exists
an assignment M which satises S, i.e., v
M
(A) = T for all A S.
Denition 1.2.11. Let S be a set of formulas. A formula B is said to be a
logical consequence of S if it is true under all assignments which satisfy S.
Denition 1.2.12. A formula B is said to be logically valid (or a tautology) if
B is true under all assignments. Equivalently, B is a logical consequence of the
empty set.
Remark 1.2.13. B is a logical consequence of A
1
, . . . , A
n
if and only if
(A
1
& &A
n
) B
is logically valid. B is logically valid if and only if B is not satisable.
7
1.3 Logical Equivalence
Denition 1.3.1. Two formulas A and B are said to be logically equivalent,
written A B, if each is a logical consequence of the other.
Remark 1.3.2. A B holds if and only if AB is logically valid.
Exercise 1.3.3. Assume A
1
A
2
. Show that
1. A
1
A
2
;
2. A
1
&B A
2
&B;
3. B&A
1
B&A
2
;
4. A
1
B A
2
B;
5. BA
1
BA
2
;
6. A
1
B A
2
B;
7. BA
1
BA
2
;
8. A
1
B A
2
B;
9. BA
1
BA
2
.
Exercise 1.3.4. Assume A
1
A
2
. Show that for any formula C containing
A
1
as a part, if we replace one or more occurrences of the part A
1
by A
2
, then
the resulting formula is logically equivalent to C. (Hint: Use the results of the
previous exercise, plus induction on the degree of C.)
Remark 1.3.5. Some useful logical equivalences are:
1. commutative laws:
(a) A&B B&A
(b) AB BA
(c) AB BA
Note however that AB , BA.
2. associative laws:
(a) A&(B&C) (A&B) &C
(b) A(BC) (AB) C
(c) A(BC) , (AB) C
Note however that A(B C) , (AB) C.
3. distributive laws:
8
(a) A&(BC) (A&B) (A&C)
(b) A(B&C) (AB) &(AC)
(c) A(B&C) (AB) &(AC)
(d) (AB) C (AC) &(BC)
Note however that A(BC) , (AB) (AC), and (A&B) C ,
(AC) &(BC).
4. negation laws:
(a) (A&B) (A) (B)
(b) (AB) (A) &(B)
(c) A A
(d) (AB) A&B
(e) (AB) (A) B
(f) (AB) A(B)
5. implication laws:
(a) AB (A&B)
(b) AB (A) B
(c) AB (B) (A)
(d) AB (AB) &(BA)
(e) AB (A) (B)
Denition 1.3.6. A formula is said to be in disjunctive normal form if it is
of the form A
1
A
m
, where each clause A
i
, i = 1, . . . , m, is of the form
B
1
& &B
n
, and each B
j
, j = 1, . . . , n is either an atom or the negation of
an atom.
Example 1.3.7. Writing p as an abbreviation for p, the formula
(p
1
&p
2
&p
3
) (p
1
&p
2
&p
3
) (p
1
&p
2
&p
3
)
is in disjunctive normal form.
Exercise 1.3.8. Show that every propositional formula C is logically equivalent
to a formula in disjunctive normal form.
Remark 1.3.9. There are two ways to do Exercise 1.3.8.
1. One way is to apply the equivalences of Remark 1.3.5 to subformulas of C
via Exercise 1.3.4, much as one applies the commutative and distributive
laws in algebra to reduce every algebraic expression to a polynomial.
9
2. The other way is to use a truth table for C. The disjunctive normal form
of C has a clause for each assignment making C true. The clause species
the assignment.
Example 1.3.10. Consider the formula (p q) r. We wish to put this in
disjunctive normal form.
Method 1. Applying the equivalences of Remark 1.3.5, we obtain
(p q) r r (p q)
r (p &q)
r (p &q)
and this is in disjunctive normal form.
Method 2. Consider the truth table
p q r p q (p q) r
1 T T T T T
2 T T F T F
3 T F T F T
4 T F F F T
5 F T T T T
6 F T F T F
7 F F T T T
8 F F F T F
Each line of this table corresponds to a dierent assignment. From lines 1, 3,
4, 5, 7 we read o the following formula equivalent to (p q) r in disjunctive
normal form:
(p &q &r) (p &q &r) (p &q &r) (p &q &r) (p &q &r) .
1.4 The Tableau Method
Remark 1.4.1. A more descriptive name for tableaux is satisability trees. We
follow the approach of Smullyan [3].
Denition 1.4.2. A signed formula is an expression of the form TA or FA,
where A is a formula. An unsigned formula is simply a formula.
Denition 1.4.3. A signed tableau is a rooted dyadic tree where each node
carries a signed formula. An unsigned tableau is a rooted dyadic tree where
each node carries an unsigned formula. The signed tableau rules are presented
in Table 1.1. The unsigned tableau rules are presented in Table 1.2. If is a
(signed or unsigned) tableau, an immediate extension of is a larger tableau

obtained by applying a tableau rule to a nite path of .

10
.
.
.
TA&B
.
.
.
[
TA
TB
.
.
.
FA&B
.
.
.
/
FA FB
.
.
.
TAB
.
.
.
/
TA TB
.
.
.
FAB
.
.
.
[
FA
FB
.
.
.
TAB
.
.
.
/
FA TB
.
.
.
FAB
.
.
.
[
TA
FB
.
.
.
TAB
.
.
.
/
TA FA
TB FB
.
.
.
FAB
.
.
.
/
TA FA
FB TB
.
.
.
TA
.
.
.
[
FA
.
.
.
FA
.
.
.
[
TA
Table 1.1: Signed tableau rules for propositional connectives.
11
.
.
.
A&B
.
.
.
[
A
B
.
.
.
(A&B)
.
.
.
/
A B
.
.
.
AB
.
.
.
/
A B
.
.
.
(AB)
.
.
.
[
A
B
.
.
.
AB
.
.
.
/
A B
.
.
.
(AB)
.
.
.
[
A
B
.
.
.
AB
.
.
.
/
A A
B B
.
.
.
(AB)
.
.
.
/
A A
B B
.
.
.
A
.
.
.
[
A
Table 1.2: Unsigned tableau rules for propositional connectives.
12
Denition 1.4.4. Let X
1
, . . . , X
k
be a nite set of signed or unsigned formulas.
A tableau starting with X
1
, . . . , X
k
is a tableau obtained from
X
1
.
.
.
X
k
by repeatedly applying tableau rules.
Denition 1.4.5. A path of a tableau is said to be closed if it contains a
conjugate pair of signed or unsigned formulas, i.e., a pair such as TA, FA in
the signed case, or A, A in the unsigned case. A path of a tableau is said to
be open if it is not closed. A tableau is said to be closed if each of its paths is
closed.
The tableau method:
1. To test a formula A for validity, form a signed tableau starting with FA, or
equivalently an unsigned tableau starting with A. If the tableau closes
o, then A is logically valid.
2. To test whether B is a logical consequence of A
1
, . . . , A
k
, form a signed
tableau starting with TA
1
, . . . , TA
k
, FB, or equivalently an unsigned
tableau starting with A
1
, . . . , A
k
, B. If the tableau closes o, then B is
indeed a logical consequence of A
1
, . . . , A
k
.
3. To test A
1
, . . . , A
k
for satisability, form a signed tableau starting with
TA
1
, . . . , TA
k
, or equivalently an unsigned tableau starting with A
1
, . . . , A
k
.
If the tableau closes o, then A
1
, . . . , A
k
is not satisable. If the tableau
does not close o, then A
1
, . . . , A
k
is satisable, and from any open path
we can read o an assignment satisfying A
1
, . . . , A
k
.
The correctness of these tests will be proved in Section 1.5. See Corollaries
1.5.9, 1.5.10, 1.5.11 below.
Example 1.4.6. Using the signed tableau method to test (p &q) (q &p) for
logical validity, we have
F(p &q) (q &p)
Tp &q
Fq &p
Tp
Tq
/
Fq Fp
Since (every path of) the tableau is closed, (p &q) (q &p) is logically valid.
13
1.5 The Completeness Theorem
Let X
1
, . . . , X
k
be a nite set of signed formulas, or a nite set of unsigned
formulas.
Lemma 1.5.1 (the Soundness Theorem). If is a nite closed tableau
starting with X
1
, . . . , X
k
, then X
1
, . . . , X
k
is not satisable.
Proof. Straightforward.
Denition 1.5.2. A path of a tableau is said to be replete if, whenever it
contains the top formula of a tableau rule, it also contains at least one of the
branches. A replete tableau is a tableau in which every path is replete.
Lemma 1.5.3. Any nite tableau can be extended to a nite replete tableau.
Proof. Apply tableau rules until they cannot be applied any more.
Denition 1.5.4. A tableau is said to be open if it is not closed, i.e., it has at
least one open path.
Lemma 1.5.5. Let be a replete tableau starting with X
1
, . . . , X
k
. If is
open, then X
1
, . . . , X
k
is satisable.
In order to prove Lemma 1.5.5, we introduce the following denition.
Denition 1.5.6. Let S be a set of signed or unsigned formulas. We say that
S is a Hintikka set if
1. S obeys the tableau rules, in the sense that if it contains the top formula
of a rule then it contains at least one of the branches;
2. S contains no pair of conjugate atomic formulas, i.e., Tp, Fp in the signed
case, or p, p in the unsigned case.
Lemma 1.5.7 (Hintikkas Lemma). If S is a Hintikka set, then S is satis-
able.
Proof. Dene an assignment M by
M(p) =
_
T if Tp belongs to S
F otherwise
in the signed case, or
M(p) =
_
T if p belongs to S
F otherwise
in the unsigned case. It is not dicult to see that v
M
(X) = T for all X S.
To prove Lemma 1.5.5, it suces to note that a replete open path is a Hin-
tikka set. Thus, if a replete tableau starting with X
1
, . . . , X
k
is open, Hintikkas
Lemma implies that X
1
, . . . , X
k
is satisable.
Combining Lemmas 1.5.1 and 1.5.3 and 1.5.5, we obtain:
14
Theorem 1.5.8 (the Completeness Theorem). X
1
, . . . , X
k
is satisable if
and only if there is no nite closed tableau starting with X
1
, . . . , X
k
.
Corollary 1.5.9. A
1
, . . . , A
k
is not satisable if and only if there exists a nite
closed signed tableau starting with TA
1
, . . . , TA
k
, or equivalently a nite closed
unsigned tableau starting with A
1
, . . . , A
k
.
Corollary 1.5.10. A is logically valid if and only if there exists a nite closed
signed tableau starting with FA, or equivalently a nite closed unsigned tableau
starting with A.
Corollary 1.5.11. B is a logical consequence of A
1
, . . . , A
k
if and only if there
exists a nite closed signed tableau starting with TA
1
, . . . , TA
k
, FB, or equiv-
alently a nite closed unsigned tableau starting with A
1
, . . . , A
k
, B.
1.6 Trees and Konigs Lemma
Up to this point, our discussion of trees has been informal. We now pause to
make our tree terminology precise.
Denition 1.6.1. A tree consists of
1. a set T
2. a function : T N
+
,
3. a binary relation P on T.
The elements of T are called the nodes of the tree. For x T, (x) is the level
of x. The relation xPy is read as x immediately precedes y, or y immediately
succeeds x. We require that there is exactly one node x T such that (x) = 1,
called the root of the tree. We require that each node other than the root has
exactly one immediate predecessor. We require that (y) = (x) + 1 for all
x, y T such that xPy.
Denition 1.6.2. A subtree of T is a nonempty set T

T such that for all

y T

and xPy, x T

. Note that T

is itself a tree, under the restriction of

and P to T

. Moreover, the root of T

is the same as the root of T.

Denition 1.6.3. An end node of T is a node with no (immediate) successors.
A path in T is a set S T such that (1) the root of T belongs to S, (2) for each
x S, either x is an end node of T or there is exactly one y S such that xPy.
Denition 1.6.4. Let P

be the transitive closure of P, i.e., the smallest re-

exive and transitive relation on T containing P. For x, y T, we have xP

y
if and only if x precedes y, i.e., y succeeds x, i.e., there exists a nite sequence
x = x
0
Px
1
Px
2
x
n1
Px
n
= y. Note that the relation P

is reexive (xP

x
for all x T), antisymmetric (xP

y and yP

x imply x = y), and transitive

(xP

y and yP

z imply xP

z). Thus P

is a partial ordering of T.
15
Denition 1.6.5. T is nitely branching if each node of T has only nitely
many immediate successors in T. T is dyadic if each node of T has at most two
immediate successors in T. Note that a dyadic tree is nitely branching.
Theorem 1.6.6 (K onigs Lemma). Let T be an innite, nitely branching
tree. Then T has an innite path.
Proof. Let

T be the set of all x T such that x has innitely many successors
in T. Note that

T is a subtree of T. Since T is nitely branching, it follows by
the pigeonhole principle that each x

T has at least one immediate successor
y

T. Now dene an innite path S = x
1
, x
2
, . . . , x
n
, . . . in

T inductively by
putting x
1
= the root of T, and x
n+1
= one of the immediate successors of x
n
in

T. Clearly S is an innite path of T.
1.7 The Compactness Theorem
Theorem 1.7.1 (the Compactness Theorem, countable case). Let S be
a countable set of propositional formulas. If each nite subset of S is satisable,
then S is satisable.
Proof. In brief outline: Form an innite tableau. Apply K onigs Lemma to get
an innite path. Apply Hintikkas Lemma.
Details: Let S = A
1
, A
2
, . . . , A
i
, . . .. Start with A
1
and generate a nite
replete tableau,
1
. Since A
1
is satisable,
1
has at least one open path. Append
A
2
to each of the open paths of
1
, and generate a nite replete tableau,
2
.
Since A
1
, A
2
is satisable,
2
has at least one open path. Append A
3
to each
of the open paths of
2
, and generate a nite replete tableau,
3
. . . . . Put
=

i=1

i
. Thus is a replete tableau. Note also that is an innite, nitely
branching tree. By K onigs Lemma (Theorem 1.6.6), let S

be an innite path
in . Then S

is a Hintikka set containing S. By Hintikkas Lemma, S

is
satisable. Hence S is satisable.
Theorem 1.7.2 (the Compactness Theorem, uncountable case). Let S
be an uncountable set of propositional formulas. If each nite subset of S is
satisable, then S is satisable.
Proof. We present three proofs. The rst uses Zorns Lemma. The second uses
transnite induction. The third uses Tychonos Theorem.
Let L be the (necessarily uncountable) propositional language consisting of
all atoms occurring in formulas of S. If S is a set of L-formulas, we say that
S is nitely satisable if each nite subset of S is satisable. We are trying to
prove that, if S is nitely satisable, then S is satisable.
First proof. Consider the partial ordering F of all nitely satisable sets of
L-formulas which include S, ordered by inclusion. It is easy to see that any
chain in F has a least upper bound in F. Hence, by Zorns Lemma, F has a
maximal element, S

. Thus S

is a set of L-formulas, S

S, S

is nitely
satisable, and for each L-formula A / S

, S

A is not nitely satisable.

16
From this it is straightforward to verify that S

is a Hintikka set. Hence, by

Hintikkas Lemma, S

is satisable. Hence S is satisable.

Second proof. Let A

, < , be a transnite enumeration of all L-formulas.

By transnite recursion, put S
0
= S, S
+1
= S

if S

is nitely
satisable, S
+1
= S

otherwise, and S

=

<
S

for limit ordinals .

Using transnite induction, it is easy to verify that S

is nitely satisable for

each . In particular, S

is nitely satisable. It is straightforward to

verify that S

is a Hintikka set. Hence, by Hintikkas Lemma, S

is satisable.
Hence S is satisable.
Third proof. Let M = T, F
L
be the space of all L-assignments M : L
T, F. Make M a topological space with the product topology where T, F
has the discrete topology. Since T, F is compact, it follows by Tychonos
Theorem that M is compact. For each L-formula A, put M
A
= M M :
v
M
(A) = T. It is easy to check that each M
A
is a topologically closed set
in M. If S is nitely satisable, then the family of sets M
A
, A S has the
nite intersection property, i.e.,

AS0
M
A
,= for each nite S
0
S. By
compactness of M it follows that

AS
M
A
,= . Thus S is satisable.
1.8 Combinatorial Applications
In this section we present some combinatorial applications of the Compactness
Theorem for propositional calculus.
Denition 1.8.1.
1. A graph consists of a set of vertices together with a specication of certain
pairs of vertices as being adjacent. We require that a vertex may not be
adjacent to itself, and that u is adjacent to v if and only if v is adjacent
to u.
2. Let G be a graph and let k be a positive integer. A k-coloring of G is a
function f : vertices of G c
1
, . . . , c
k
such that f(u) ,= f(v) for all
adjacent pairs of vertices u, v.
3. G is said to be k-colorable if there exists a k-coloring of G. This notion is
much studied in graph theory.
Exercise 1.8.2. Let G be a graph and let k be a positive integer. For each
vertex v and each i = 1, . . . , k, let p
vi
be a propositional atom expressing that
vertex v receives color c
i
. Dene C
k
(G) to be the following set of propositional
formulas: p
v1
p
vk
for each vertex v; (p
vi
&p
vj
) for each vertex v and
1 i < j k; (p
ui
&p
vi
) for each adjacent pair of vertices u, v and 1 i k.
1. Show that there is a one-to-one correspondence between k-colorings of G
and assignments satisfying C
k
(G).
2. Show that G is k-colorable if and only if C
k
(G) is satisable.
17
3. Show that G is k-colorable if and only if each nite subgraph of G is
k-colorable.
Denition 1.8.3. A partial ordering consists of a set P together with a binary
relation
P
such that
1. a
P
a for all a P (reexivity);
2. a
P
b, b
P
c imply a
P
c (transitivity);
3. a
P
b, b
P
a imply a = b (antisymmetry).
Example 1.8.4. Let P = N
+
= 1, 2, 3, . . . , n, . . . = the set of positive inte-
gers.
1. Let
P
be the usual order relation on P, i.e., m
P
n if and only if m n.
2. Let
P
be the divisibility ordering of P, i.e., m
P
n if and only if m is
a divisor of n.
Denition 1.8.5. Let P,
P
be a partial ordering.
1. Two elements a, b P are comparable if either a
P
b or b
P
a. Other-
wise they are incomparable.
2. A chain is a set X P such that any two elements of X are comparable.
3. An antichain is a set X P such that any two distinct elements of X are
incomparable.
Exercise 1.8.6. Let P,
P
be a partial ordering, and let k be a positive integer.
1. Use the Compactness Theorem to show that P is the union of k chains if
and only if each nite subset of P is the union of k chains.
2. Dilworths Theorem says that P is the union of k chains if and only if
every antichain is of size k. Show that Dilworths Theorem for arbi-
trary partial orderings follows from Dilworths Theorem for nite partial
orderings.
18
Chapter 2
Predicate Calculus
2.1 Formulas and Sentences
Denition 2.1.1 (languages). A language L is a set of predicates, each pred-
icate P of L being designated as n-ary for some nonnegative
1
integer n.
Denition 2.1.2 (variables and quantiers). We assume the existence of
a xed, countably innite set of symbols x, y, z, . . . known as variables. We
introduce two new symbols: the universal quantier () and the existential
quantier (). They are read as for all and there exists, respectively.
Denition 2.1.3 (formulas). Let L be a language, and let U be a set. It is
understood that U is disjoint from the set of variables. The set of L-U-formulas
is generated as follows.
1. An atomic L-U-formula is an expression of the form Pe
1
e
n
where P
is an n-ary predicate of L and each of e
1
, . . . , e
n
is either a variable or an
element of U.
2. Each atomic L-U-formula is an L-U-formula.
3. If A is an L-U-formula, then A is an L-U-formula.
4. If A and B are L-U-formulas, then A&B, AB, AB, AB are L-
U-formulas.
5. If x is a variable and A is an L-U-formula, then xA and xA are L-U-
formulas.
Denition 2.1.4 (degree). The degree of a formula is the number of occur-
rences of propositional connectives , &, , , and quantiers , in it.
1
It will be seen that 0-ary predicates behave as propositional atoms. Thus predicate
calculus is an extension of propositional calculus.
19
Denition 2.1.5. An L-formula is an L--formula, i.e., an L-U-formula where
U = , the empty set.
Remark 2.1.6. If U is a subset of U

, then every L-U-formula is automati-

cally an L-U

-formula. In particular, every L-formula is automatically an L-U-

formula, for any set U.
Denition 2.1.7. In situations where the language L is understood from con-
text, an L-U-formula may be called a U-formula, and an L-formula a formula.
Denition 2.1.8 (substitution). If A is an L-U-formula and x is a variable
and a U, we dene an L-U-formula A[x/a] as follows.
1. If A is atomic, then A[x/a] = the result of replacing each occurrence of x
in A by a.
2. (A)[x/a] = A[x/a].
3. (A&B)[x/a] = A[x/a] &B[x/a].
4. (AB)[x/a] = A[x/a] B[x/a].
5. (AB)[x/a] = A[x/a] B[x/a].
6. (AB)[x/a] = A[x/a] B[x/a].
7. (xA)[x/a] = xA.
8. (xA)[x/a] = xA.
9. If y is a variable other than x, then (y A)[x/a] = y A[x/a].
10. If y is a variable other than x, then (y A)[x/a] = y A[x/a].
Denition 2.1.9 (free variables). An occurrence of a variable x in an L-U-
formula A is said to be bound in A if it is within the scope of a quantier x or
x in A. An occurrence of a variable x in an L-U-formula A is said to be free
in A if it is not bound in A. A variable x is said to occur freely in A if there is
at least one occurrence of x in A which is free in A.
Exercise 2.1.10.
1. Show that A[x/a] is the result of substituting a for all free occurrences of
x in A.
2. Show that x occurs freely in A if and only if A[x/a] ,= A.
Denition 2.1.11 (sentences). An L-U-sentence is an L-U-formula in which
no variables occur freely. An L-sentence is an L--sentence, i.e., an L-U-sentence
where U = , the empty set.
20
Remark 2.1.12. If U is a subset of U

, then every L-U-sentence is automat-

ically an L-U

-sentence. In particular, every L-sentence is automatically an

L-U-sentence, for any set U.
Denition 2.1.13. In situations where the language L is understood from
context, an L-U-sentence may be called a U-sentence, and an L-sentence a
sentence.
2.2 Structures and Satisability
Denition 2.2.1. Let U be a nonempty set, and let n be a nonnegative
2
integer.
U
n
is the set of all n-tuples of elements of U, i.e.,
U
n
= a
1
, . . . , a
n
) : a
1
, . . . , a
n
U .
An n-ary relation on U is a subset of U
n
.
Denition 2.2.2. Let L be a language. An L-structure M consists of a non-
empty set U
M
, called the domain or universe of M, together with an n-ary
relation P
M
on U
M
for each n-ary predicate P of L. An L-structure may be
called a structure, in situations where the language L is understood from context.
Denition 2.2.3. Two L-structures M and M

are said to be isomorphic if

there exists an isomorphism of M onto M

, i.e., a one-to-one correspondence

: U
M

= U
M
such that for all n-ary predicates P of L and all n-tuples
a
1
, . . . , a
n
) (U
M
)
n
, a
1
, . . . , a
n
) P
M
if and only if (a
1
), . . . , (a
n
)) P
M
.
As usual in abstract mathematics, we are mainly interested in properties of
structures that are invariant under isomorphism.
Lemma 2.2.4. Given an L-structure M, there is a unique valuation or assign-
ment of truth values
v
M
: A : A is an L-U
M
-sentence T, F
dened as follows:
1. v
M
(Pa
1
a
n
) =
_
T if a
1
, . . . , a
n
) P
M
,
F if a
1
, . . . , a
n
) / P
M
.
2. v
M
(A) =
_
T if v
M
(A) = F,
F if v
M
(A) = T.
3. v
M
(A&B) =
_
T if v
M
(A) = v
M
(B) = T,
F if at least one of v
M
(A), v
M
(B) = F.
2
In the special case n = 0 we obtain the notion of a 0-ary relation, i.e., a subset of {}.
There are only two 0-ary relations, {} and {}, corresponding to T and F respectively. Thus
a 0-ary predicate behaves as a propositional atom.
21
4. v
M
(AB) =
_
T if at least one of v
M
(A), v
M
(B) = T,
F if v
M
(A) = v
M
(B) = F.
5. v
M
(AB) = v
M
((A&B)) .
6. v
M
(AB) =
_
T if v
M
(A) = v
M
(B) ,
F if v
M
(A) ,= v
M
(B) .
7. v
M
(xA) =
_
T if v
M
(A[x/a]) = T for all a U
M
,
F if v
M
(A[x/a]) = F for at least one a U
M
.
8. v
M
(xA) =
_
T if v
M
(A[x/a]) = T for at least one a U
M
,
F if v
M
(A[x/a]) = F for all a U
M
.
Proof. The truth value v
M
(A) is dened by recursion on L-U
M
-sentences, i.e.,
by induction on the degree of A where A is an arbitrary L-U
M
-sentence.
Denition 2.2.5 (truth and satisfaction). Let M be an L-structure.
1. Let A be an L-U
M
-sentence. We say that A is true in M if v
M
(A) = T.
We say that A is false in M if v
M
(A) = F.
2. Let S be a set of L-U
M
-sentences. We say that M satises S, abbreviated
M [= S, if all of the sentences of S are true in M.
Theorem 2.2.6.
1. If M and M

are isomorphic L-structures and : M

= M

is an isomor-
phism of M onto M

, then for all L-U

M
-sentences A we have v
M
(A) =
v
M
(A

) where A

= A[a
1
/(a
1
), . . . , a
k
/(a
k
)].
3
Here a
1
, . . . , a
k
are the
elements of U
M
which occur in A.
2. If M and M

are isomorphic L-structures, then they satisfy the same

L-sentences.
Proof. We omit the proof of part 1. A more general result will be proved later
as Theorem 2.7.3. Part 2 follows immediately from part 1.
Denition 2.2.7 (satisability). Let S be a set of L-sentences. S is said to
be satisable
4
if there exists an L-structure M which satises S.
Remark 2.2.8. Satisability is one of the most important concepts of mathe-
matical logic. A key result known as the Compactness Theorem
5
states that a
set S of L-sentences is satisable if and only every nite subset of S is satisable.
3
We have extended the substitution notation 2.1.8 in an obvious way.
4
Similarly, the notions of logical validity and logical consequence are dened for L-
sentences, in the obvious way, using L-structures. An L-sentence is said to be logically valid
if it is satised by all L-structures. An L-sentence is said to be a logical consequence of S if
it is satised by all L-structures satisfying S.
5
See Theorems 2.6.1 and 2.6.2 below.
22
The following related notion is of technical importance.
Denition 2.2.9 (satisability in a domain). Let U be a nonempty set. A
set S of L-U-sentences is said to be satisable in the domain U if there exists
an L-structure M such that M [= S and U
M
= U.
Remark 2.2.10. Let S be a set of L-sentences. Then S is satisable (according
to Denition 2.2.7) if and only if S is satisable in some domain U.
Theorem 2.2.11. Let S be a set of L-sentences. If S is satisable in a domain
U, then S is satisable in any domain of the same cardinality as U.
Proof. Suppose S is satisable in a domain U. Let M be an L-structure M
satisfying S with U
M
= U. Let U

be any set of the same cardinality as U.

Then there exists a one-to-one correspondence : U U

. Let M

be the
L-structure with U
M
= U

, P
M
= (a
1
), . . . , (a
n
)) : a
1
, . . . , a
n
) P
M
for
all n-ary predicates P of L. Then M is isomorphic to M

. Hence, by Theorem
2.2.6, M

[= S. Thus S is satisable in the domain U

.
Example 2.2.12. We exhibit a sentence A

which is satisable in an innite

domain but not in any nite domain. Our sentence A

is (1) &(2) &(3) with

(1) xy z ((Rxy &Ryz) Ryz)
(2) xy (Rxy Ryx)
(3) xy Rxy
See also Example 2.5.9.
2.3 The Tableau Method
Denition 2.3.1. Fix a countably innite set V = a
1
, a
2
, . . . , a
n
, . . .. The
elements of V will be called parameters. If L is a language, L-V -sentences will
be called sentences with parameters.
Denition 2.3.2. A (signed or unsigned) tableau is a rooted dyadic tree where
each node carries a (signed or unsigned) L-V -sentence. The tableau rules for
predicate calculus are the same as those for propositional calculus, with the
following additional rules.
Signed:
.
.
.
TxA
.
.
.
[
TA[x/a]
.
.
.
FxA
.
.
.
[
FA[x/a]
where a is an arbitrary parameter
23
.
.
.
TxA
.
.
.
[
TA[x/a]
.
.
.
FxA
.
.
.
[
FA[x/a]
where a is a new parameter
Unsigned:
.
.
.
xA
.
.
.
[
A[x/a]
.
.
.
xA
.
.
.
[
A[x/a]
where a is an arbitrary parameter
.
.
.
xA
.
.
.
[
A[x/a]
.
.
.
xA
.
.
.
[
A[x/a]
where a is a new parameter
Remark 2.3.3. In the above tableau rules, a is new means that a does not
occur in the path that is being extended. Or, we can insist that a not occur in
the tableau that is being extended.
Remark 2.3.4. We are going to prove that the tableau method for predicate
calculus is sound (Theorem 2.3.9) and complete (Theorem 2.5.5). In particular,
a sentence A of the predicate calculus is logically valid if and only if there exists
a nite closed signed tableau starting with FA, or equivalently a nite closed
unsigned tableau starting with A.
24
Example 2.3.5. The signed tableau
F(xy Rxy) (y xRxy)
Txy Rxy
Fy xRxy
Ty Ray
FxRxb
TRab
FRab
is closed. Therefore, by the Soundness Theorem, (xy Rxy) (y xRxy) is
logically valid.
Example 2.3.6. The unsigned tableau
((x(PxQx)) ((xPx) (xQx)))
/
x(PxQx)
((xPx) (xQx))
xPx
xQx
Pa Qa
/
Pa Qa
Pa Qa
x(PxQx)
(xPx) (xQx)
/
xPx
Pb
(Pb Qb)
Pb
Qb
xQx
Qc
(Pc Qc)
Pc
Qc
is closed. Therefore, by the Soundness Theorem,
(x(PxQx)) ((xPx) (xQx))
is logically valid.
The rest of this section is devoted to proving the Soundness Theorem 2.3.9.
Denition 2.3.7.
1. An L-V -structure consists of an L-structure M together with a mapping
: V U
M
. If A is an L-V -sentence, we write
A

= A[a
1
/(a
1
), . . . , a
k
/(a
k
)]
where a
1
, . . . , a
k
are the parameters occurring in A. Note that A

is an
L-U
M
-sentence. Note also that, if A is an L-sentence, then A

= A.
2. Let S be a nite or countable set of (signed or unsigned) L-V -sentences.
An L-V -structure M, is said to satisfy S if v
M
(A

) = T for all A S.
S is said to be satisable
6
if there exists an L-V -structure satisfying S.
Note that this denition is compatible with Denition 2.2.7.
6
Similarly, the notions of logical validity and logical consequence are extended to L-V -
sentences, in the obvious way, using L-V -structures. An L-V -sentence is said to be logically
valid if it satised by all L-V -structures. An L-V -sentence is said to be a logical consequence
of S if it is satised by all L-V -structures satisfying S.
25
3. Let be an L-tableau. We say that is satisable if at least one path of
is satisable.
Lemma 2.3.8. Let and

be tableaux such that

is an immediate extension
of , i.e.,

is obtained from by applying a tableau rule to a path of . If

is satisable, then

is satisable.
Proof. The proof consists of one case for each tableau rule. We consider some
representative cases.
Case 1. Suppose that

is obtained from by applying the rule

.
.
.
AB
.
.
.
/
A B
to the path in . Since is satisable, it has at least one satisable path, S. If
S ,= , then S is a path of

, so

is satisable. If S = , then is satisable, so

let M and : V U
M
satisfy . In particular v
M
((AB)

) = T, so we have
at least one of v
M
(A

) = T and v
M
(B

) = T. Thus M and satisfy at least

one of , A and , B. Since these are paths of

, it follows that

is satisable.
Case 2. Suppose that

is obtained from by applying the rule

.
.
.
xA
.
.
.
[
A[x/a]
to the path in , where a is a parameter. Since is satisable, it has at least
one satisable path, S. If S ,= , then S is a path of

, so

is satisable. If
S = , then is satisable, so let M and : V U
M
satisfy . In particular
v
M
(x(A

)) = v
M
((xA)

) = T, so v
M
(A

[x/c]) = T for all c U

M
. In
particular
v
M
(A[x/a]

) = v
M
(A

[x/(a)]) = T.
Thus M and satisfy , A[x/a]. Since this is a path of

, it follows that

is
satisable.
Case 3. Suppose that

is obtained from by applying the rule

.
.
.
xA
.
.
.
[
A[x/a]
26
to the path in , where a is a new parameter. Since is satisable, it has at
least one satisable path, S. If S ,= , then S is a path of

, so

is satisable.
If S = , then is satisable, so let M and : V U
M
satisfy . In particular
v
M
(x(A

)) = v
M
((xA)

) = T, so v
M
(A

[x/c]) = T for at least one c U

M
.
Fix such a c and dene

: V U
M
by putting

(a) = c, and

(b) = (b) for

all b ,= a, b V . Since a is new, we have B

= B

for all B , and A

= A

,
hence A[x/a]

= A

[x/

(a)] = A

[x/c]. Thus v
M
(B

) = v
M
(B

) = T for all
B , and v
M
(A[x/a]

) = v
M
(A

[x/c]) = T. Thus M and

satisfy , A[x/a].
Since this is a path of

, it follows that

is satisable.
Theorem 2.3.9 (the Soundness Theorem). Let X
1
, . . . , X
k
be a nite set
of (signed or unsigned) sentences with parameters. If there exists a nite closed
tableau starting with X
1
, . . . , X
k
, then X
1
, . . . , X
k
is not satisable.
Proof. Let be a closed tableau starting with X
1
, . . . , X
k
. Thus there is a nite
sequence of tableaux
0
,
1
, . . . ,
n
= such that

0
=
X
1
.
.
.
X
k
and each
i+1
is an immediate extension of
i
. Suppose X
1
, . . . , X
k
is satisable.
Then
0
is satisable, and by induction on i using Lemma 2.3.8, it follows that all
of the
i
are satisable. In particular
n
= is satisable, but this is impossible
since is closed.
2.4 Logical Equivalence
Denition 2.4.1. Given a formula A, let A

= A[x
1
/a
1
, . . . , x
k
/a
k
], where
x
1
, . . . , x
k
are the variables which occur freely in A, and a
1
, . . . , a
k
are parame-
ters not occurring in A. Note that A

has no free variables, i.e., it is a sentence.

We dene A to be satisable if and only if A

is satisable, in the sense of De-

nition 2.3.7. We dene A to be logically valid if and only if A

is logically valid,
in the sense of Denition 2.3.7.
Exercises 2.4.2. Let A be a formula.
1. Show that A is logically valid if and only if A is not satisable. Show
that A is satisable if and only if A is not logically valid.
2. Let x be a variable. Show that A is logically valid if and only if xA is
logically valid. Show that A is satisable if and only if xA is satisable.
3. Let x be a variable, and let a be a parameter not occurring in A. Show
that A is logically valid if and only if A[x/a] is logically valid. Show that
A is satisable if and only if A[x/a] is satisable.
27
Denition 2.4.3. Let A and B be formulas. A and B are said to be logically
equivalent, written A B, if AB is logically valid.
Exercise 2.4.4. Assume A B. Show that for any variable x, xA xB
and xA xB. Show that for any variable x and parameter a, A[x/a]
B[x/a].
Exercise 2.4.5. For a formula A, it is not in general true that A A

, where
A

is as in Denition 2.4.1. Also, it is not in general true that A xA, or that

A xA, or that A A[x/a]. Give examples illustrating these remarks.
Exercise 2.4.6. If A and B are formulas, put A

= A[x
1
/a
1
, . . . , x
k
/a
k
] and
B

= B[x
1
/a
1
, . . . , x
k
/a
k
], where x
1
, . . . , x
k
are the variables occurring freely in
A and B, and a
1
, . . . , a
k
are parameters not occurring in A or in B. Show that
A B if only if A

.
Remark 2.4.7. The results of Exercises 1.3.3 and 1.3.4 and Remark 1.3.5 for
formulas of the propositional calculus, also hold for formulas of the predicate
calculus.
In particular, if A
1
A
2
, then for any formula C containing A
1
as a part,
if we replace one or more occurrences of the part A
1
by A
2
, then the resulting
formula is logically equivalent to C.
Remark 2.4.8. Some useful logical equivalences are:
1. (a) xA A, provided x does not occur freely in A
(b) xA A, provided x does not occur freely in A
(c) xA y A[x/y], provided y does not occur in A
(d) xA y A[x/y], provided y does not occur in A
2. (a) x(A&B) (xA) &(xB)
(b) x(AB) (xA) (xB)
(c) x(AB) (xA) (xB)
Note however that, in general, x(A&B) , (xA) &(xB), and
x(AB) , (xA) (xB), and x(AB) , (xA) (xB).
On the other hand, we have:
3. (a) x(A&B) A&(xB), provided x does not occur freely in A
(b) x(A&B) (xA) &B, provided x does not occur freely in B
(c) x(AB) A(xB), provided x does not occur freely in A
(d) x(AB) (xA) B, provided x does not occur freely in B
(e) x(AB) A(xB), provided x does not occur freely in A
(f) x(AB) (xA) B, provided x does not occur freely in B
4. (a) xA xA
28
(b) xA xA
(c) xA xA
(d) xA xA
Denition 2.4.9 (prenex form). A formula is said to be quantier-free if it
contains no quantiers. A formula is said to be in prenex form if it is of the form
Q
1
x
1
Q
n
x
n
B, where each Q
i
is a quantier ( or ), each x
i
is a variable,
and B is quantier-free.
Example 2.4.10. The sentence
xy z w(Rxy (Rxz &Rzy &(Rzw&Rwy)))
is in prenex form.
Exercise 2.4.11. Show that every formula is logically equivalent to a formula
in prenex form.
Example 2.4.12. Consider the sentence (xPx) &(xQx). We wish to put
this into prenex form. Applying the equivalences of Remark 2.4.8, we have
(xPx) &(xQx) (xPx) &(y Qy)
x(Px&(y Qy))
xy (Px&Qy)
and this is in prenex form.
Exercises 2.4.13. Let A and B be quantier-free formulas. Put the following
into prenex form.
1. (xA) &(xB)
2. (xA) (xB)
3. (xA) (xB)
Denition 2.4.14 (universal closure). Let A be a formula. The univer-
sal closure of A is the sentence A

= x
1
x
k
A, where x
1
, . . . , x
k
are the
variables which occur freely in A. Note that A

= A

.
Exercises 2.4.15. Let A be a formula.
1. Show that A is logically valid if and only if A

, the universal closure of A,

is logically valid.
2. It is not true in general that A A

. Give an example illustrating this.

3. It is not true in general that A is satisable if and only if A

is satisable.
Give an example illustrating this.
4. For formulas A and B, it is not true in general that A B if and only if
A

. Give an example illustrating this.

29
For completeness we state the following denition.
Denition 2.4.16. Let A
1
, . . . , A
k
, B be formulas. We say that B is a logi-
cal consequence of A
1
, . . . , A
k
if (A
1
& &A
k
) B is logically valid. This is
equivalent to saying that the universal closure of (A
1
& &A
k
) B is logi-
cally valid.
Remark 2.4.17. A and B are logically equivalent if and only if each is a
logical consequence of the other. A is logically valid if and only if A is a logical
consequence of the empty set. xA is a logical consequence of A[x/a], but the
converse does not hold in general. A[x/a] is a logical consequence of xA, but
the converse does not hold in general.
2.5 The Completeness Theorem
Let U be a nonempty set, and let S be a set of (signed or unsigned) L-U-
sentences.
Denition 2.5.1. S is closed if S contains a conjugate pair of L-U-sentences.
In other words, for some L-U-sentence A, S contains TA, FA in the signed
case, A, A in the unsigned case. S is open if it is not closed.
Denition 2.5.2. S is U-replete if S obeys the tableau rules with respect to
U. We list some representative clauses of the denition.
1. If S contains TA, then S contains FA. If S contains FA, then S
contains TA. If S contains A, then S contains A.
2. If S contains TA&B, then S contains both TA and TB. If S contains
FA&B, then S contains at least one of FA and FB. If S contains A&B,
then S contains both A and B. If S contains (A&B), then S contains
at least one of A and B.
3. If S contains TxA, then S contains TA[x/a] for at least one a U. If
S contains FxA, then S contains FA[x/a] for all a U. If S contains
xA, then S contains A[x/a] for at least one a U. If S contains xA,
then S contains A[x/a] for all a U.
4. If S contains TxA, then S contains TA[x/a] for all a U. If S contains
FxA, then S contains FA[x/a] for at least one a U. If S contains
xA, then S contains A[x/a] for all a U. If S contains xA, then S
contains A[x/a] for at least one a U.
Lemma 2.5.3 (Hintikkas Lemma). If S is U-replete and open
7
, then S is
satisable. In fact, S is satisable in the domain U.
7
See also Exercise 2.5.7.
30
Proof. Assume S is U-replete and open. We dene an L-structure M by putting
U
M
= U and, for each n-ary predicate P of L,
P
M
= (a
1
, . . . , a
n
) U
n
: TPa
1
a
n
S
in the signed case, and
P
M
= (a
1
, . . . , a
n
) U
n
: Pa
1
a
n
S
in the unsigned case.
We claim that for all L-U-sentences A,
(a) if S contains TA, then v
M
(A) = T
(b) if S contains FA, then v
M
(A) = F
in the signed case, and
(c) if S contains A, then v
M
(A) = T
(d) if S contains A, then v
M
(A) = F
in the unsigned case.
In both cases, the claim is easily proved by induction on the degree of A.
We give the proof for some representative cases.
1. deg(A) = 0. In this case A is atomic, say A = Pa
1
a
n
.
(a) If S contains TPa
1
a
n
, then by denition of M we have (a
1
, . . . , a
n
)
P
M
, so v
M
(Pa
1
a
n
) = T.
(b) If S contains FPa
1
a
n
, then S does not contain TPa
1
a
n
since
S is open. Thus by denition of M we have (a
1
, . . . , a
n
) / P
M
, so
v
M
(Pa
1
a
n
) = F.
(c) If S contains Pa
1
a
n
, then by denition of M we have (a
1
, . . . , a
n
)
P
M
, so v
M
(Pa
1
a
n
) = T.
(d) If S contains Pa
1
a
n
, then S does not contain Pa
1
a
n
since
S is open. Thus by denition of M we have (a
1
, . . . , a
n
) / P
M
, so
v
M
(Pa
1
a
n
) = F.
2. deg(A) > 0 and A = B. Note that deg(B) < deg(A) so the inductive
hypothesis applies to B.
3. deg(A) > 0 and A = B&C. Note that deg(B) and deg(C) are < deg(A)
so the inductive hypothesis applies to B and C.
(a) If S contains TB&C, then by repleteness of S we see that S contains
both TB and TC. Hence by inductive hypothesis we have v
M
(B) =
v
M
(C) = T. Hence v
M
(B&C) = T.
31
(b) If S contains FB&C, then by repleteness of S we see that S contains
at least one of FB and FC. Hence by inductive hypothesis we have
at least one of v
M
(B) = F and v
M
(C) = F. Hence v
M
(B &C) = F.
(c) If S contains B&C, then by repleteness of S we see that S contains
both B and C. Hence by inductive hypothesis we have v
M
(B) =
v
M
(C) = T. Hence v
M
(B&C) = T.
(d) If S contains (B &C), then by repleteness of S we see that S con-
tains at least one of B and C. Hence by inductive hypothe-
sis we have at least one of v
M
(B) = F and v
M
(C) = F. Hence
v
M
(B&C) = F.
4. deg(A) > 0 and A = xB. Note that for all a U we have deg(B[x/a]) <
deg(A), so the inductive hypothesis applies to B[x/a].
5. deg(A) > 0 and A = xB. Note that for all a U we have deg(B[x/a]) <
deg(A), so the inductive hypothesis applies to B[x/a].

We shall now use Hintikkas Lemma to prove the completeness of the tableau
method. As in Section 2.3, Let V = a
1
, . . . , a
n
, . . . be the set of parameters.
Recall that a tableau is a tree whose nodes carry L-V -sentences.
Lemma 2.5.4. Let
0
be a nite tableau. By applying tableau rules, we can
extend
0
to a (possibly innite) tableau with the following properties: every
closed path of is nite, and every open path of is V -replete.
Proof. The idea is to start with
0
and use tableau rules to construct a sequence
of nite extensions
0
,
1
, . . . ,
i
, . . .. If some
i
is closed, then the construction
halts, i.e.,
j
=
i
for all j i, and we set =
i
. In any case, we set
=

i=0

i
. In the course of the construction, we apply tableau rules
systematically to ensure that

will have the desired properties, using the fact

that V = a
1
, a
2
, . . . , a
n
, . . . is countably innite.
Here are the details of the construction. Call a node X of
i
quasiuniversal
if it is of the form TxA or FxA or xA or xA. Our construction begins
with
0
. Suppose we have constructed
2i
. For each quasiuniversal node X of

2i
and each n 2i, apply the appropriate tableau rule to extend each open
path of
2i
containing X by TA[x/a
n
] or FA[x/a
n
] or A[x/a
n
] or A[x/a
n
] as
the case may be. Let
2i+1
be the nite tableau so obtained. Next, for each
non-quasiuniversal node X of
2i+1
, extend each open path containing X by
applying the appropriate tableau rule. Again, let
2i+2
be the nite tableau so
obtained.
In this construction, a closed path is never extended, so all closed paths of

are nite. In addition, the construction ensures that each open path of

is V -replete. Thus

has the desired properties. This proves our lemma.

Theorem 2.5.5 (the Completeness Theorem). Let X
1
, . . . , X
k
be a nite
set of (signed or unsigned) sentences with parameters. If X
1
, . . . , X
k
is not
32
satisable, then there exists a nite closed tableau starting with X
1
, . . . , X
k
. If
X
1
, . . . , X
k
is satisable, then X
1
, . . . , X
k
is satisable in the domain V .
Proof. By Lemma 2.5.4 there exists a (possibly innite) tableau starting with
X
1
, . . . , X
k
such that every closed path of is nite, and every open path of
is V -replete. If is closed, then by Konigs Lemma (Theorem 1.6.6), is nite.
If is open, let S be an open path of . Then S is V -replete. By Hintikkas
Lemma 2.5.3, S is satisable in V . Hence X
1
, . . . , X
k
is satisable in V .
Denition 2.5.6. Let L, U, and S be as in Denition 2.5.1. S is said to
be atomically closed if S contains a conjugate pair of atomic L-U-sentences.
In other words, for some n-ary L-predicate P and a
1
, . . . , a
n
U, S contains
TPa
1
a
n
, FPa
1
a
n
in the signed case, and Pa
1
a
n
, Pa
1
a
n
in the
unsigned case. S is atomically open if it is not atomically closed.
Exercise 2.5.7. Show that Lemmas 2.5.3 and 2.5.4 and Theorem 2.5.5 continue
to hold with closed (open) replaced by atomically closed (atomically
open).
Remark 2.5.8. Corollaries 1.5.9, 1.5.10, 1.5.11 carry over from the proposi-
tional calculus to the predicate calculus. In particular, the tableau method
provides a test for logical validity of sentences of the predicate calculus.
Note however that the test is only partially eective. If a sentence A is
logically valid, we will certainly nd a nite closed tableau starting with A.
But if A is not logically valid, we will not necessarily nd a nite tableau which
demonstrates this. See the following example.
Example 2.5.9. In 2.2.12 we have seen an example of a sentence A

which
is satisable in a countably innite domain but not in any nite domain. It is
33
instructive to generate a tableau starting with A

.
A

.
.
.
xy z ((Rxy &Ryz) Ryz)
xy (Rxy Ryx)
xy Rxy
y Ra
1
y
Ra
1
a
2
y (Ra
1
y Rya
1
)
Ra
1
a
2
Ra
2
a
1
/
Ra
1
a
2
Ra
2
a
1
y Ra
2
y
Ra
2
a
3
.
.
.
Ra
3
a
2
y z ((Ra
1
y &Ryz) Ra
1
z)
z ((Ra
1
a
2
&Ra
2
z) Ra
1
z)
(Ra
1
a
2
&Ra
2
a
3
) Ra
1
a
3
)
/
(Ra
1
a
2
&Ra
2
a
3
) Ra
1
a
3
/
.
.
.
Ra
1
a
2
Ra
2
a
3
Ra
3
a
1
y Ra
3
y
Ra
3
a
4
.
.
.
An innite open path gives rise (via the proof of Hintikkas Lemma) to an innite
L-structure M with U
M
= a
1
, a
2
, . . . , a
n
, . . ., R
M
= a
m
, a
n
) : 1 m < n.
Clearly M [= A

.
Remark 2.5.10. In the course of applying a tableau test, we will sometimes nd
a nite open path which is U-replete for some nite set of parameters U V .
In this case, the proof of Hintikkas Lemma provides a nite L-structure with
domain U.
Example 2.5.11. Let A be the sentence (x(PxQx)) ((xPx) (xQx)).
34
Testing A for logical validity, we have:
A
x(PxQx)
((xPx) (xQx))
xPx
xQx
Pa
Qb
Pa Qa
Pb Qb
/
Pa Qa
/
Pb Qb
This tableau has a unique open path, which gives rise (via the proof of Hintikkas
Lemma) to a nite L-structure M with U
M
= a, b, P
M
= b, Q
M
= a.
Clearly M falsies A.
2.6 The Compactness Theorem
Theorem 2.6.1 (the Compactness Theorem, countable case). Let S be
a countably innite set of sentences of the predicate calculus. S is satisable if
and only if each nite subset of S is satisable.
Proof. We combine the ideas of the proofs of the Countable Compactness Theo-
rem for propositional calculus (Theorem 1.7.1) and the Completeness Theorem
for predicate calculus (Theorem 2.5.5).
Details: Let S = A
0
, A
1
, . . . , A
i
, . . .. Start by letting
0
be the empty
tableau. Suppose we have constructed
2i
. Extend
2i
to

2i
by appending A
i
to each open path of
2i
. Since A
0
, A
1
, . . . , A
i
is satisable,

2i
has at least
one open path. Now extend

2i
to
2i+1
and then to
2i+2
as in the proof of
Lemma 2.5.4. Finally put =

i=1

i
. As in Lemma 2.5.4 we have that
every closed path of is nite, and every open path of is V -replete. Note
also that is an innite, nitely branching tree. By K onigs Lemma (Theorem
1.6.6), let S

be an innite path in . Then S

is a Hintikka set containing S. By

Hintikkas Lemma for predicate calculus (Lemma 2.5.3), S

is satisable. Hence
S is satisable.
Theorem 2.6.2 (the Compactness Theorem, uncountable case). Let S
be an uncountable set of sentences of the predicate calculus. S is satisable if
and only if each nite subset of S is satisable.
Proof. Assume that S is nitely satisable. For each sentence A S of the form
xB or xB, introduce a new parameter c
A
. Let U
S
be the set of parameters
so introduced. Let S

be S together with the sentences B[x/c

A
] or B[x/c
A
]
35
as the case may be, for all c
A
U
S
. Then S

is a set of L-U
S
-sentences, and
it is easy to verify that S

is nitely satisable. By Zorns Lemma, let S

be a
maximal nitely satisable set of L-U
S
-sentences extending S

.
Now inductively dene S
0
= S, S
n+1
= S

n
, S

n=0
S
n
, U =

n=0
U
Sn
.
It is straightforward to verify that S

is an L-U-Hintikka set. Hence, by Hin-

tikkas Lemma, S

is satisable in the domain U. Since S S

, it follows
that S is satisable in U.
Exercise 2.6.3. Let L be a language consisting of a binary predicate R and
some additional predicates. Let M = (U
M
, R
M
, . . .) be an L-structure such
that (U
M
, R
M
) is isomorphic to (N, <
N
). Note that M contains no innite
R-descending sequence. Show that there exists an L-structure M

such that:
1. M and M

satisfy the same L-sentences.

2. M

contains an innite R-descending sequence. In other words, there

exist elements a

1
, a

2
, . . . , a

n
, . . . U
M
such that a

n+1
, a

n
) R
M
for all
n = 1, 2, . . ..
Hint: Use the Compactness Theorem.
Exercise 2.6.4. Generalize Exercise 2.6.3 replacing (N, <
N
) by an arbitrary
innite linear ordering with no innite descending sequence. Show that M

can
be obtained such that (U
M
, R
M
) is a linear ordering which contains an innite
descending sequence.
2.7 Satisability in a Domain
The notion of satisability in a domain was introduced in Denition 2.2.9.
Theorem 2.7.1. Let S be a set of L-sentences.
1. Assume that S is nite or countably innite. If S is satisable, then S is
satisable in a countably innite domain.
2. Assume that S is of cardinality

. If S is satisable, then S is satisable

in a domain of cardinality

.
Proof. Parts 1 and 2 follow easily from the proofs of Compactness Theorems
2.6.1 and 2.6.2, respectively. In the countable case we have that S is satisable
in V , which is countably innite. In the uncountable case we have that S is
satisable in U, where U is as in the proof of 2.6.2. By the arithmetic of innite
cardinal numbers, the cardinality of U is

0
=

.
In Example 2.2.12 we have seen a sentence A

which is satisable in a
countably innite domain but not in any nite domain. Regarding satisability
in nite domains, we have:
36
Example 2.7.2. Given a positive integer n, we exhibit a sentence A
n
which
is satisable in a domain of cardinality n but not in any domain of smaller
cardinality. Our sentence A
n
is (1) &(2) &(3) with
(1) xy z ((Rxy &Ryz) Ryz)
(2) xy (Rxy Ryx)
(3) x
1
x
n
(Rx
1
x
2
&Rx
2
x
3
& &Rx
n1
x
n
)
On the other hand, we have:
Theorem 2.7.3. Let M and M

be L-structures. Assume that there ex-

ists an onto mapping : U
M
U
M
such that for all n-ary predicates P
of L and all n-tuples a
1
, . . . , a
n
) (U
M
)
n
, a
1
, . . . , a
n
) P
M
if and only if
(a
1
), . . . , (a
n
)) P
M
. Then as in Theorem 2.2.6 we have v
M
(A) = v
M
(A

)
for all L-U
M
-sentences A, where A

= A[a
1
/(a
1
), . . . , a
k
/(a
k
)]. In particular,
M and M

satisfy the same L-sentences.

Proof. The proof is by induction on the degree of A. Suppose for example that
A = xB. Then by denition of v
M
we have that v
M
(A) = T if and only if
v
M
(B[x/a]) = T for all a U
M
. By inductive hypothesis, this holds if and only
if v
M
(B[x/a]

) = T for all a U
M
. But for all a U
M
we have B[x/a]

=
B

[x/(a)]. Thus our condition is equivalent to v

M
(B

[x/(a)]) = T for all

a U
M
. Since : U
M
U
M
is onto, this is equivalent to v
M
(B

[x/b]) = T
for all b U
M
. By denition of v
M
this is equivalent to v
M
(xB

) = T. But
xB

= A

, so our condition is equivalent to v

M
(A

) = T.
Corollary 2.7.4. Let S be a set of L-sentences. If S is satisable in a domain
U, then S is satisable in any domain of the same or larger cardinality.
Proof. Suppose S is satisable in domain U. Let U

be a set of cardinality
greater than or equal to that of U. Let : U

U be onto. If M is any
L-structure with U
M
= U, we can dene an L-structure M

with U
M
= U

by putting P
M
= a
1
, . . . , a
n
) : (a
1
), . . . , (a
n
)) P
M
for all n-ary pred-
icates P of L. By Theorem 2.7.3, M and M

satisfy the same L-sentences. In

particular, if M [= S, then M

[= S.
Remark 2.7.5. We shall see later
8
that Theorem 2.7.3 and Corollary 2.7.4 fail
for normal satisability.
8
See Section 4.1.
37
Chapter 3
Proof Systems for Predicate
Calculus
3.1 Introduction to Proof Systems
Denition 3.1.1. An abstract proof system consists of a set X together with a
relation R

k=0
X
k+1
. Elements of X are called objects. Elements of R are
called rules of inference. An object X X is said to be derivable, or provable,
if there exists a nite sequence of objects X
1
, . . . , X
n
such that X
n
= X and,
for each i n, there exist j
1
, . . . , j
k
< i such that X
j1
, . . . , X
j
k
, X
i
) R. The
sequence X
1
, . . . , X
n
is called a derivation of X, or a proof of X.
Notation 3.1.2. For k 1 it is customary to write
X
1
X
k
Y
indicating that X
1
, . . . , X
k
, Y ) R. This is to be understood as from the
premises X
1
, . . . , X
k
we may immediately infer the conclusion Y . For k = 0
we may write
Y
or simply Y , indicating that Y ) R. This is to be understood as we may
infer Y from no premises, or we may assume Y .
Denition 3.1.3. Let L be a language. Recall that V is the set of parameters.
A Hilbert-style proof system for L is a proof system with the following properties:
1. The objects are sentences with parameters. In other words,
X = A : A is an L-V -sentence .
2. For each rule of inference
A
1
A
k
B
38
(i.e., A
1
, . . . , A
k
, B) R), we have that B is a logical consequence of
A
1
, . . . , A
k
. This property is known as soundness. It implies that every
L-V -sentence which is derivable is logically valid.
3. For all L-V -sentences A, B, we have a rule of inference A, AB, B) R,
i.e.,
A AB
B
.
In other words, fromA and AB we immediately infer B. This collection
of inference rules is known as modus ponens.
4. An L-V -sentence A is logically valid if and only if A is derivable. This
property is known as completeness.
Remark 3.1.4. In Section 3.3 we shall explicitly exhibit a particular Hilbert-
style proof system, LH. The soundness of LH will be obvious. In order to verify
the completeness of LH, we shall rst prove a result known as the Companion
Theorem, which is also of interest in its own right.
3.2 The Companion Theorem
In this section we shall comment on the notion of logical validity for sentences of
the predicate calculus. We shall analyze logical validity into two components: a
propositional component (quasitautologies), and a quanticational component
(companions).
Denition 3.2.1 (quasitautologies).
1. A tautology is a propositional formula which is logically valid.
2. A quasitautology is an L-V -sentence of the form F[p
1
/A
1
, . . . , p
k
/A
k
],
where F is a tautology, p
1
, . . . , p
k
are the atoms occurring in F, and
A
1
, . . . , A
k
are L-V -sentences.
For example, p (q p) is a tautology. This implies that, for all L-V -sentences
A and B, A(BA) is a quasitautology.
Remarks 3.2.2.
1. Obviously, every quasitautology is logically valid.
2. There is a decision procedure
1
for quasitautologies. One such decision
procedure is based on truth tables. Another is based on propositional
tableaux.
1
In other words, there is a Turing algorithm which, given an L-V -sentence A as input, will
eventually halt with output 1 if A is a quasitautology, 0 if A is not a quasitautology.
39
3. It can be shown that there is no decision procedure for logical validity.
(This result is known as Churchs Theorem.) Therefore, in relation to the
problem of characterizing logical validity, we regard the quasitautologies
as trivial.
Let A be an L-V -sentence.
Denition 3.2.3 (companions). A companion of A is any L-V -sentence of
one of the forms
(1) (xB) B[x/a]
(2) B[x/a] (xB)
(3) (xB) B[x/a]
(4) B[x/a] (xB)
where, in (2) and (3), the parameter a may not occur in A or in B.
Lemma 3.2.4. Let C be a companion of A.
1. A is satisable if and only if C &A is satisable.
2. A is logically valid if and only if C A is logically valid.
Proof. Let C be a companion of A.
For part 1, assume that A is satisable. In accordance with Denition 2.3.7,
let M, be an L-V -structure satisfying A. If C is of the form 3.2.3(1) or
3.2.3(4), then C is logically valid, hence M, satises C &A. Next, consider
the case when C is of the form 3.2.3(2). If M, satises xB, then M,
satises C. Otherwise we have v
M
(xB

) = F, so let c U
M
be such that
v
M
(B

[x/c]) = F. Dene

: V U
M
by putting

(a) = c,

(b) = (b)
for b ,= a. Since a does not occur in A, we have that M,

satises A. Also,
since a does not occur in B, we have B[x/a]

= B

[x/c] = B

[x/c], hence
v
M
(B[x/a]

) = v
M
(B

[x/c]) = F, i.e., M,

satises B[x/a]. Thus M,

satises C &A. The case when C is of the form 3.2.3(3) is handled similarly.
For part 2 note that, since C is a companion of A, C is a companion of A.
Thus we have that A is logically valid if and only if A is not satisable, if
and only if C &A is not satisable (by part 1), if and only if (C &A) is
logically valid, i.e., C A is logically valid.
Denition 3.2.5 (companion sequences). A companion sequence of A is a
nite sequence C
1
, . . . , C
n
such that, for each i < n, C
i+1
is a companion of
(C
1
& &C
i
) A.
Lemma 3.2.6. If C
1
, . . . , C
n
is a companion sequence of A, then A is logically
valid if and only if (C
1
& &C
n
) A is logically valid.
Proof. Note that (C
1
& &C
n
) A is quasitautologically equivalent to
C
n
(C
n1
(C
1
A)).
40
Our lemma follows by n applications of part 2 of Lemma 3.2.4.
Theorem 3.2.7 (the Companion Theorem). A is logically valid if and only
if there exists a companion sequence C
1
, . . . , C
n
of A such that
(C
1
& &C
n
) A
is a quasitautology.
Proof. The if part is immediate from Lemma 3.2.6. For the only if part,
assume that A is logically valid. By Theorem 2.5.5 let be a nite closed
unsigned tableau starting with A. Thus we have a nite sequence of tableaux

0
,
1
, . . . ,
n
where
0
= A,
n
= , and each
i+1
is obtained by applying
a tableau rule R
i
to
i
. If R
i
is a quantier rule, let C
i
be an appropriate
companion. Thus C
1
, . . . , C
n
is a companion sequence for A, and we can easily
transform into a closed tableau

starting with
A
C
1
.
.
.
C
n
in which only propositional tableau rules are applied. Thus (C
1
& &C
n
) A
is a quasitautology. This proves our theorem.
For instance, if R
i
is the tableau rule
()
.
.
.
xB
.
.
.
[
B[x/a] ,
where a is an arbitrary parameter, let C
i
be the companion (xB) B[x/a],
and replace the application of () by
.
.
.
(xB) B[x/a]
.
.
.
xB
.
.
.
/
xB B[x/a]
noting that the left-hand path is closed.
41
Similarly, if R
i
is the tableau rule
()
.
.
.
xB
.
.
.
[
B[x/a]
where a is a new parameter, let C
i
be the companion B[x/a] (xB), and
replace the application of () by
.
.
.
B[x/a] (xB)
.
.
.
xB
.
.
.
/
B[x/a] xB
noting that the right-hand path is closed.
Example 3.2.8. As an example illustrating Theorem 3.2.7 and its proof, let
A be the sentence (x(PxQx)) ((xPx) (xQx)). Let be the closed
tableau
A
x(PxQx)
((xPx) (xQx))
xPx
xQx
Pa Qa
Pa
Qa
/
Pa Qa
which shows that A is logically valid. Examining the applications of quantier
rules in , we obtain the companion sequence C
1
, C
2
, C
3
for A, where C
1
is
(x(PxQx)) (Pa Qa), C
2
is Pa xPx, C
3
is Qa xQx. Clearly
(C
1
&C
2
&C
3
) A is a quasitautology.
3.3 A Hilbert-Style Proof System
Let L be a language. Recall that V is the set of parameters.
Denition 3.3.1 (the system LH). Our Hilbert-style proof system LH for
the predicate calculus is as follows:
42
1. The objects are L-V -sentences.
2. For each quasitautology A, A) is a rule of inference.
3. (xB) B[x/a]) and B[x/a] (xB)) are rules of inference.
4. A, AB, B) is a rule of inference.
5. AB[x/a], A(xB)) and B[x/a] A, (xB) A) are rules of in-
ference, provided the parameter a does not occur in A or in B.
Schematically, LH consists of:
1. A, where A is any quasitautology
2. (a) (xB) B[x/a] (universal instantiation)
(b) B[x/a] (xB) (existential instantiation)
3.
A AB
B
(modus ponens)
4. (a)
AB[x/a]
A(xB)
(universal generalization)
(b)
B[x/a] A
(xB) A
(existential generalization),
where a does not occur in A, B.
Lemma 3.3.2 (soundness of LH). LH is sound. In other words, for all
L-V -sentences A, if A is derivable, then A is logically valid.
Proof. The proof is straightforward by induction on the length of a derivation.
The induction step is similar to the proof of Lemma 3.2.4.
Example 3.3.3. In LH we have the following derivation:
1. (xA) A[x/a] (universal instantiation)
2. A[x/a] (xA) (existential instantiation)
3. ((xA) A[x/a]) ((A[x/a] (xA)) ((xA) (xA)))
(This is a quasitautology, obtained from the tautology
(p q) ((q r) (p r)).)
4. (A[x/a] (xA)) ((xA) (xA)) (1, 3, modus ponens)
5. (xA) (xA) (2, 4, modus ponens)
Thus, by Lemma 3.3.2, (xA) (xA) is logically valid.
Example 3.3.4. In LH we have the following derivation:
1. B[x/a] (xB) (existential instantiation)
43
2. (B[x/a] (xB)) ((A&B)[x/a] (xB)) (quasitautology)
3. (A&B)[x/a] (xB) (1, 2, modus ponens)
4. (x(A&B)) (xB) (3, existential generalization)
Thus, by Lemma 3.3.2, (x(A&B)) (xB) is logically valid.
We now turn to the proof that LH is complete.
Lemma 3.3.5. LH is closed under quasitautological consequence. In other
words, if A
1
, . . . , A
k
are derivable, and if B is a quasitautological consequence
of A
1
, . . . , A
k
, then B is derivable.
Proof. We are assuming that B is a quasitautological consequence of A
1
, . . . , A
k
.
Thus A
1
(A
2
(A
k
B)) is a quasitautology, hence derivable. We are
also assuming that A
1
, . . . , A
k
are derivable. Thus we obtain B by k applications
of modus ponens.
Lemma 3.3.6. If C is a companion of A, and if C A is derivable in LH, then
A is derivable in LH.
Proof. First, suppose C is of the form 3.2.3(1), namely (xB) B[x/a]. By
universal instantiation, C is derivable. In addition, we are assuming that C A
is derivable. Hence, by modus ponens, A is derivable.
Next, suppose C is of the form 3.2.3(2), namely B[x/a] (xB), where
a does not occur in A, B. We are assuming that C A is derivable, i.e.,
(B[x/a] (xB)) A is derivable. It follows by Lemma 3.3.5 that both (i)
(A) B[x/a] and (ii) (A) (xB) are derivable. Applying universal gen-
eralization to (i), we see that (A) (xB) is derivable. From this plus (ii),
it follows by Lemma 3.3.5 that A is derivable.
The other cases, where C is of the form 3.2.3(3) or 3.2.3(4), are handled
similarly.
Theorem 3.3.7 (completeness of LH). LH is sound and complete. In other
words, for all L-V -sentences A, A is derivable if and only if A is logically valid.
Proof. The only if part is Lemma 3.3.2. For the if part, assume that
A is logically valid. By Theorem 3.2.7, there exists a companion sequence
C
1
, . . . , C
n
for A such that (C
1
& &C
n
) A is a quasitautology. Hence
C
n
(C
n1
(C
1
A)) is a quasitautology, hence derivable. From this
and n applications of Lemma 3.3.6, we obtain derivability of A.
Exercise 3.3.8. Consider the following proof system LH

, which is a stripped
down version of LH. The objects of LH

are L-V -sentences containing only ,

, (i.e., not containing , , &, ). The rules of LH

are:
1. quasitautologies
2. (xB) B[x/a]
44
3. (x(AB)) (AxB)
4.
A AB
B
(modus ponens)
5.
B[x/a]
xB
(generalization), where a does not occur in B.
Show that LH

is sound and complete.

Exercise 3.3.9.
1. Let S be a set of L-sentences. Consider a proof system LH(S) consisting
of LH with additional rules of inference A), A S. Show that an L-V -
sentence B is derivable in LH(S) if and only if B is a logical consequence
of S.
2. Indicate the modications needed when S is a set of L-V -sentences.
Notation 3.3.10. We write S B to indicate that B is derivable in LH(S).
3.4 Gentzen-Style Proof Systems
Throughout this section, let L be a language. As usual, V is the set of param-
eters.
Before presenting our Gentzen-style proof system for L, we rst discuss the
block tableau method, a trivial variant of the signed tableau method.
Denition 3.4.1. A block is a nite set of signed L-V -sentences. A block is
said to be closed if it contains TA and FA for some L-V -sentence A.
Notation 3.4.2. If S is a block and X is a signed L-V -sentence, we write S, X
instead of S X, etc.
Denition 3.4.3. A block tableau is a rooted dyadic tree where each node car-
ries a block. A block tableau is said to be closed if each of its end nodes is closed.
Given a block S, a block tableau starting with S is a block tableau generated
from S by means of block tableau rules. The block tableau rules are obtained
from the signed tableau rules (pages 11 and 23) as follows. Corresponding to
signed tableau rules of the form
.
.
.
X
.
.
.
[
Y
.
.
.
X
.
.
.
[
Y
1
Y
2
.
.
.
X
.
.
.
/
Y Z
.
.
.
X
.
.
.
/
Y
1
Z
1
Y
2
Z
2
45
we have block tableau rules
S, X
[
S, X, Y
S, X
[
S, X, Y
1
, Y
2
S, X
/
S, X, Y S, X, Z
S, X
/
S, X, Y
1
, Y
2
S, X, Z
1
, Z
2
respectively.
For example, we have the following block tableau rules:
S, TA&B
[
S, TA&B, TA, TB
S, FA&B
/
S, FA&B, FA S, FA&B, FB
S, TAB
/
S, TAB, FA S, TAB, TB
S, FAB
[
S, FAB, TA, FB
S, TxA
[
S, TxA, TA[x/a]
S, FxA
[
S, FxA, FA[x/a]
where a is new.
Example 3.4.4. We exhibit a closed block tableau demonstrating that xA is
a logical consequence of xA.
TxA, FxA
[
TxA, FxA, TA[x/a]
[
TxA, FxA, TA[x/a], FA[x/a]
This block tableau is of course similar to the signed tableau
TxA
FxA
TA[x/a]
FA[x/a]
which demonstrates the same thing.
We now dene our Gentzen-style system, LG.
Denition 3.4.5. A sequent is an expression of the form where and
are nite sets of L-V -sentences. If S = TA
1
, . . . , TA
m
, FB
1
, . . . , FB
n
is a
block, let [S[ be the sequent A
1
, . . . , A
m
B
1
, . . . , B
n
. This gives a one-to-one
correspondence between blocks and sequents.
46
Denition 3.4.6 (the system LG). Our Gentzen-style proof system LG for
the predicate calculus is as follows.
1. The objects of LG are sequents.
2
2. For each closed block S, we have a rule of inference [S[). In other words,
for all nite sets of L-V -sentences and and all L-V -sentences A, we
assume the sequent , A A, .
3. For each non-branching block tableau rule
S
[
S

we have a rule of inference [S

1
[, [S[), i.e.,
[S

[
[S[
.
4. For each branching block tableau rule
S
/
S

we have a rule of inference [S

[, [S

[, [S[), i.e.,
[S

[ [S

[
[S[
.
Thus LG includes the following rules of inference:
, A A,
, A A,
, A
, A A,
A,
, A&B, A, B
, A&B
A&B, A, A&B, B,
A&B,
, AB A, , AB, B
, AB
, A AB, B,
AB,
, xA, A[x/a]
, xA
xA, A[x/a],
xA,
where a does not occur in the conclusion.
2
For this reason, LG is sometimes called a sequent calculus.
47
Exercise 3.4.7. Explicitly display the remaining inference rules of LG.
Denition 3.4.8. A sequent A
1
, . . . , A
m
B
1
, . . . , B
n
is said to be logically
valid if and only if the L-V -sentence
(A
1
& &A
m
) (B
1
B
n
)
is logically valid.
Theorem 3.4.9 (soundness and completeness of LG). LG is sound and
complete. In other words, a sequent is logically valid if and only if it is
derivable in LG. In particular, an L-V -sentence A is logically valid if and only
if the sequent
A
is derivable in LG.
Proof. Note that the sequent A
1
, . . . , A
m
B
1
, . . . , B
n
is logically valid if
and only if the block TA
1
, . . . , TA
m
, FB
1
, . . . , FB
n
is not satisable. Thus,
soundness and completeness of LG is equivalent to soundness and completeness
of the block tableau method. The latter is in turn easily seen to be equivalent
to soundness and completeness of the signed tableau method, as presented in
Theorems 2.3.9 and 2.5.5.
Exercise 3.4.10. Let LG(atomic) be a variant of LG in which , A A, is
assumed only for atomic L-V -sentences A. Show that LG(atomic) is sound and
complete. (Hint: Use the result of Exercise 2.5.7.)
Exercise 3.4.11.
1. The modied block tableau rules are a variant of the block tableau rules of
Denition 3.4.3, replacing each non-branching rule of the form
S, X
[
S, X, Y
1
, Y
2
by a pair of rules
S, X
[
S, X, Y
1
S, X
[
S, X, Y
2
Show that the modied block tableau rules are sound and complete.
2. Let LG

be the variant of LG corresponding to the modied block tableau

rules. Write out all the rules of LG

explicitly. Show that LG

is sound
and complete.
48
3.5 The Interpolation Theorem
As usual, let L be a language and let V be the set of parameters.
Theorem 3.5.1 (the Interpolation Theorem). Let A and B be L-V -sentences.
If AB is logically valid, we can nd an L-V -sentence I such that:
1. AI and I B are logically valid.
2. Each predicate and parameter occurring in I occurs in both A and B.
Such an I is called an interpolant for AB. We indicate this by writing
A
I
B.
Remark 3.5.2. If A and B have no predicates in common, then obviously the
theorem is incorrect as stated, because all L-V -sentences necessarily contain at
least one predicate. In this case, we modify the conclusion of the theorem to
say that at least one of A and B is logically valid.
3
The conclusion is obvious
in this case.
In order to prove the Interpolation Theorem, we introduce a symmetric
variant of LG, wherein sentences do not move from one side of to the other.
Denition 3.5.3. A signed sequent is an expression of the form M N where
M and N are nite sets of signed L-V -sentences. A variant of M N is a
signed sequent obtained from M N by transferring sentences from one side
of to the other, changing signs. In particular, M, X N and M X, N
are variants of each other, where we use an overline to denote conjugation, i.e.,
TA = FA, FA = TA.
Denition 3.5.4. Let
C
1
, . . . , C
m
D
1
, . . . , D
n
be an unsigned sequent
4
. A signed variant of C
1
, . . . , C
m
D
1
, . . . , D
n
is any
variant of the signed sequent
TC
1
, . . . , TC
m
TD
1
, . . . , TD
n
.
Note that each signed sequent is a signed variant of one and only one unsigned
sequent. We dene a signed sequent to be logically valid if and only if the
corresponding unsigned sequent is logically valid.
Denition 3.5.5. LG(symmetric) is the following proof system.
1. The objects are signed sequents.
3
This amounts to saying that at least one of the truth values T and F is an interpolant for
AB.
4
An unsigned sequent is just what we have previously called a sequent.
49
2. We have
M, X X, N
and
M, X, X N
and
M X, X, N
for all X.
3. For each signed tableau rule of the form
.
.
.
X
.
.
.
[
Y
.
.
.
X
.
.
.
[
Y
1
Y
2
.
.
.
X
.
.
.
/
Y Z
.
.
.
X
.
.
.
/
Y
1
Z
1
Y
2
Z
2
we have a corresponding pair of signed sequent rules
M, X, Y N
M, X N
M X, Y , N
M X, N
M, X, Y
1
, Y
2
N
M, X N
M X, Y
1
, Y
2
, N
M X, N
M, X, Y N M, X, Z N
M, X N
M X, Y , N M X, Z, N
M X, N
M, X, Y
1
, Y
2
N M, X, Z
1
, Z
2
N
M, X N
M X, Y
1
, Y
2
, N M X, Z
1
, Z
2
, N
M X, N
respectively.
Lemma 3.5.6. An unsigned sequent is derivable in LG if and only if all of its
signed variants are derivable in LG(symmetric).
Proof. The proof is by induction on the length of derivations in LG. The
base step consists of noting that all signed variants of , A A, are of
the form M, X X, N or M, X, X N or M X, X, N, hence deriv-
able in LG(symmetric). The inductive step consists of checking that, for each
rule of inference of LG, if all signed variants of the premises are derivable
in LG(symmetric), then so are all signed variants of the conclusion. This is
straightforward.
50
Theorem 3.5.7. LG(symmetric) is sound and complete. In other words, a
signed sequent is logically valid if and only if it is derivable in LG(symmetric).
In particular, an L-V -sentence AB is logically valid if and only if the signed
sequent TA TB is derivable in LG(symmetric).
Proof. Soundness and completeness of LG(symmetric) follows from Theorem
3.4.9, soundness and completeness of LG, using Lemma 3.5.6.
We now prove the Interpolation Theorem.
Denition 3.5.8. Let M N be a signed sequent. An interpolant for M N
is an L-V -sentence I such that the signed sequents M TI and TI N are
logically valid, and all predicates and parameters occurring in I occur in both
M and N.
5
We indicate this by writing M
I
N.
In order to prove the Interpolation Theorem, it suces by Theorem 3.5.7 to
prove that every signed sequent derivable in LG(symmetric) has an interpolant.
We prove this by induction on the length of derivations.
For the base step, we note that X is an interpolant for M, X X, N, and
that M, X, X and X, X, N are logically valid. Thus we have M, X
X
X, N
and M, X, X
F
N and M
T
X, X, N.
For the induction step we show that, for each rule of LG(symmetric), given
interpolants for the premises of the rule, we can nd an interpolant for the
conclusion. We present some representative special cases.
M, TA&B, TA, TB
I
N
M, TA&B
I
N
M
I
FA&B, FA, FB, N
M
I
FA&B, N
M, FA&B, FA
I
N M, FA&B, FB
J
N
M, FA&B
I J
N
M
I
TA&B, TA, N M
J
TA&B, TB, N
M
I &J
TA&B, N
M, TA, FA
I
N
M, TA
I
N
M
I
FA, TA, N
M
I
FA, N
M, FA, TA
I
N
M, FA
I
N
M
I
TA, FA, N
M
I
TA, N
5
In the special case when M and N have no predicates in common, we require instead
that at least one of the signed sequents M and N be logically valid. This amounts to
requiring that at least one of T, F be an interpolant for M N.
51
M, FxA, FA[x/a]
I
N
M, FxA
I
N
where a does not occur in the conclusion.
M
I
TxA, TA[x/a], N
M
I
TxA, N
where a does not occur in the conclusion.
M, TxA, TA[x/a]
I
N
M, TxA
K
N
where K = I if a occurs in M, TxA, otherwise K = z I[a/z] where z is a
new variable.
M
I
FxA, FA[x/a], N
M
K
FxA, N
where K = I if a occurs in FxA, N, otherwise K = z I[a/z] where z is a new
variable.
This completes the proof.
52
Chapter 4
Extensions of Predicate
Calculus
4.1 Predicate Calculus with Identity
Denition 4.1.1. A language with identity consists of a language L with a
particular binary predicate, I, designated as the identity predicate.
Denition 4.1.2. Let L be a language with identity. The identity axioms for
L are the following sentences:
1. xIxx (reexivity)
2. xy (Ixy Iyx) (symmetry)
3. xy z ((Ixy &Iyz) Ixz) (transitivity)
4. For each n-ary predicate P of L, we have an axiom
x
1
x
n
y
1
y
n
((Ix
1
y
1
& &Ix
n
y
n
) (Px
1
x
n
Py
1
y
n
))
(congruence).
Exercise 4.1.3. Show that the identity predicate is unique in the following
sense. If L contains two identity predicates I
1
and I
2
, then xy (I
1
xy I
2
xy)
is a logical consequence of the identity axioms for I
1
and I
2
.
Let L be a language with identity.
Denition 4.1.4. An L-structure M is said to be normal if the identity pred-
icate denotes the identity relation, i.e., I
M
= a, a) : a U
M
.
Note that any normal L-structure automatically satises the identity axioms
for L. Conversely, we have:
53
Theorem 4.1.5. Let M be an L-structure satisfying the identity axioms for
L. For each a U
M
put a = b U
M
: v
M
(Iab) = T. Then we have a normal
L-structure M and an onto mapping : U
M
U
M
as in Theorem 2.7.3, dened
by putting U
M
= a : a U
M
, and P
M
= a
1
, . . . , a
n
) : a
1
, . . . , a
n
) P
M

for all n-ary predicates P.

Proof. This is straightforward, using the fact that I
M
is a congruence with
respect to each of the relations P
M
, P L.
Theorem 4.1.6. If M is an L-structure satisfying the identity axioms for L,
then we have a normal L-structure M satisfying the same sentences as M.
Proof. This is immediate from Theorems 4.1.5 and 2.7.3.
Let S be a set of L-sentences.
Denition 4.1.7. S is normally satisable if there exists a normal L-structure
which satises S.
Corollary 4.1.8. S is normally satisable if and only if
S identity axioms for L
is satisable.
We also have the Compactness Theorem for normal satisability:
Corollary 4.1.9. S is normally satisable if and only if each nite subset of S
is normally satisable.
Proof. This is immediate from Corollary 4.1.8 plus the Compactness Theorem
for predicate calculus without identity (Theorems 2.6.1 and 2.6.2), applied to
the set S identity axioms for L.
Regarding normal satisability in particular domains, we have:
Example 4.1.10. Given a positive integer n, we exhibit a sentence E
n
which
is normally satisable in domains of cardinality n but not in domains of any
other cardinality. The sentence
x
1
x
n
(y (Ix
1
y Ix
n
y) &(Ix
1
x
2
Ix
1
x
3
Ix
n1
x
n
))
has this property. Intuitively, E
n
says that there exist exactly n things.
Exercise 4.1.11. Let L be a nite language with identity, and let M be a
nite normal L-structure. Construct an L-sentence A such that, for all normal
L-structures M

, M

[= A if and only if M

is isomorphic to M.
On the other hand, we have:
Theorem 4.1.12 (Lowenheim/Skolem Theorem).
54
1. If S is normally satisable in arbitrarily large nite domains, then S is
normally satisable in some innite domain.
2. If S is normally satisable in some innite domain, then S is normally
satisable in all innite domains of cardinality the cardinality of S.
Proof. For the rst part, let S

= S H
n
: n = 1, 2, . . . where H
n
is the
sentence
x
1
x
n
(Ix
1
x
2
Ix
1
x
3
Ix
n1
x
n
)
saying that there exist at least n things. Since S is normally satisable in
arbitrarily large nite domains, each nite subset of S

is normally satis-
able. Hence, by Corollary 4.1.9, S

is normally satisable. But any normal

L-structure satisfying S

satises S and has an innite domain.

For the second part, let

be a cardinal number the cardinality of S. Let

L

= LQ
i
: i X, where each Q
i
is a new 1-ary predicate, and X is a set of
cardinality

. Let S

= SxQ
i
x : i Xx(Q
i
x&Q
j
x) : i, j X, i ,=
jidentity axioms for L

. Thus S

is a set of L

-sentences of cardinality

.
Furthermore, any domain in which S

is satisable will contain pairwise distinct

elements a
i
, i X, and will therefore have cardinality

. By assumption, S
is normally satisable in some innite domain. It follows that each nite subset
of S

is satisable. Hence, by the Compactness Theorems 2.6.1 and 2.6.2, S

is satisable. Hence, by part 2 of Theorem 2.7.1, S

is satisable in a domain
of cardinality

. Therefore, by Theorem 4.1.6, S

is normally satisable in a
domain of cardinality

, hence =

. Let M

be a normal L

-structure
with U
M
of cardinality

. Let M be the reduct of M

to L, i.e., M is the
L-structure with U
M
= U
M
and P
M
= P
M
for each predicate P in L. Then
M normally satises S and U
M
is of cardinality

.
Exercise 4.1.13. Let L be the following language:
Ox: x = 1
Pxyz: x +y = z
Qxyz: x y = z
Rxy: x < y
Sxy: x + 1 = y
Ixy: x = y (identity predicate)
For each positive integer n, let M
n
be the normal L-structure
M
n
= (U
n
, O
n
, P
n
, Q
n
, R
n
, S
n
, I
n
)
where
U
n
= 1, . . . , n
55
O
n
= 1
P
n
= i, j, k) (U
n
)
3
: i +j = k
Q
n
= i, j, k) (U
n
)
3
: i j = k
R
n
= i, j) (U
n
)
2
: i < j
S
n
= i, j) (U
n
)
2
: i + 1 = j
I
n
= i, j) (U
n
)
2
: i = j
Exhibit an L-sentence A such that, for all nite normal L-structures M

, M

[=
A if and only if M

is isomorphic to M
n
for some n.
Exercise 4.1.14. Let L and M
n
be as in Exercise 4.1.13. Show that there
exists an innite normal L-structure M = M

with the following property: for

all L-sentences A, if M
p
[= A for all suciently large primes p, then M

[= A.
(Hint: Use the Compactness Theorem.)
4.2 The Spectrum Problem
Denition 4.2.1. Let A be a sentence of the predicate calculus with identity.
The spectrum of A is the set of positive integers n such that A is normally
satisable in a domain of cardinality n. A spectrum is a set X of positive
integers, such that X = spectrum(A) for some A.
Remark 4.2.2. The spectrum problem is the problem of characterizing the
spectra, among all sets of positive integers. This is a famous and apparently
dicult open problem.
1
In particular, it is unknown whether the complement
of a spectrum is necessarily a spectrum.
Example 4.2.3. We show that the set n 1 : n is even is a spectrum.
Let U be a nonempty set. A binary relation R U
2
is said to be an
equivalence relation on U if it is reexive, symmetric, and transitive, i.e., if the
structure (U, R) satises (1) &(2) &(3):
(1) xRxx
(2) xy (Rxy Ryx)
(3) xy z ((Rxy &Ryz) Rxz)
In this situation, the equivalence classes [a]
R
= b U : a, b) R, a U,
form a partition of U, i.e., a decomposition of the set U into pairwise disjoint,
nonempty subsets.
1
Jones/Selman [1] show that X is a spectrum if and only if there exists a nondeterministic
Turing machine which accepts X in time 2
ck
, where k is the length of the input. Since the
input is a positive integer n, we have k = [log
2
n], as usual in computational number theory.
56
Let A be the following sentence of the predicate calculus with identity:
(1) & (2) & (3) & xy ((Ixy) &z (Rxz (Ixz Iyz)))
Intuitively, A says that R is an equivalence relation with the property that each
equivalence class consists of exactly two elements. Obviously, a nite set U
admits an equivalence relation with this property if and only if the cardinality
of U is even. Thus the spectrum of A is the set of even numbers.
Exercises 4.2.4. Prove the following.
1. If X is a nite or conite
2
set of positive integers, then X is a spectrum.
2. The set of even numbers is a spectrum.
3. The set of odd numbers is a spectrum.
4. If r and m are positive integers, n 1 : n r mod m is a spectrum.
5. If X and Y are spectra, X Y and X Y are spectra.
Exercise 4.2.5. Prove that, for any sentence A of the predicate calculus with
identity, at least one of spectrum(A) and spectrum(A) is conite. (Hint: Use
part 1 of Theorem 4.1.12.)
Example 4.2.6. We show that the set of composite numbers
3
is a spectrum.
Let L be a language consisting of two binary predicates, R and S, as well
as the identity predicate, I. Let A be an L-sentence saying that R and S are
equivalence relations, each with more than one equivalence class, and
xy ( exactly one z)(Rxz &Syz).
Thus, for any normal L-structure M = (U
M
, R
M
, S
M
, I
M
) satisfying A, we have
that R
M
and S
M
partition U
M
into rows and columns, respectively, in such
a way that the intersection of any row with any column consists of exactly
one element of U
M
. Thus, if U
M
is nite, the elements of U
M
are arranged
in an m n matrix, where m, n 2. Therefore, the number of elements
in U
M
is mn, a composite number. Conversely, for any m, n 2, there is an
L-structure M as above, which satises A. Thus spectrum(A) is the set of
composite numbers.
Exercise 4.2.7. Use the result of Exercise 4.1.13 to prove the following:
1. The set of prime numbers and its complement are spectra.
2. The set of squares 1, 4, 9, . . . and its complement are spectra.
3. The set of powers of 2, 2
n
: n = 1, 2, 3, . . ., and its complement, are
spectra.
2
A set of positive integers is said to be conite if its complement is nite.
3
A composite number is an integer greater than 1 which is not prime.
57
4. The set of prime powers p
n
: p prime, n = 1, 2, . . . and its complement
are spectra.
Exercise 4.2.8.
1. The Fibonacci numbers are dened recursively by F
1
= 1, F
2
= 2, F
n
=
F
n1
+F
n2
for n 3. Show that the set of Fibonacci numbers
F
n
: n = 1, 2, . . . = 1, 2, 3, 5, 8, 13, 21, 34, 55, . . .
and its complement are spectra.
2. Show that x
y
: x, y 2 and its complement are spectra.
4.3 Predicate Calculus With Operations
In this section we extend the syntax and semantics of the predicate calculus,
to encompass operations. As examples of operations, we may cite the familiar
mathematical operations of addition (+) and multiplication (). Such opera-
tions are considered binary, because they take two arguments. More generally,
we consider n-ary operations.
Denition 4.3.1 (languages). A language is a set of operations f, g, h, . . .
and predicates P, Q, R, . . .. Each operation and each predicate is designated as
n-ary for some nonnegative
4
integer n.
Denition 4.3.2 (terms, formulas, sentences). Let L be a language, and
let U be a nonempty set. The set of L-U-terms is generated as follows.
1. Each variable is an L-U-term.
2. Each element of U is an L-U-term.
3. If f is an n-ary operation of L, and if t
1
, . . . , t
n
are L-U-terms, then
ft
1
t
n
is an L-U-term.
An L-U-term is said to be variable-free if no variables occur in it. An atomic
L-U-formula is an expression of the form
Pt
1
t
n
where P is an n-ary predicate of L, and t
1
, . . . , t
n
are L-U-terms. The set
of L-U-formulas is generated as in clauses 2, 3, 4 and 5 of Denition 2.1.3.
The notions of substitution, free variables, and L-U-sentences are dened as in
Section 2.1. Note that Pt
1
t
n
is a sentence if and only if it is variable-free.
Denition 4.3.3 (structures). An L-structure M consists of a nonempty set
U
M
, an n-ary function f
M
: (U
M
)
n
U
M
for each n-ary operation f of L, and
an n-ary relation P
M
(U
M
)
n
for each n-ary predicate P of L.
4
A 0-ary operation is sometimes known as a constant. Syntactically, constants behave as
parameters.
58
Denition 4.3.4 (isomorphism). Two L-structures M and M

are said to
be isomorphic if there exists an isomorphism of M onto M

, i.e., a one-to-one
correspondence : U
M

= U
M
such that:
1. for all n-ary operations f of L and all n-tuples a
1
, . . . , a
n
) (U
M
)
n
,
(f
M
(a
1
, . . . , a
n
) = f
M
((a
1
), . . . , (a
n
)).
2. for all n-ary predicates P of L and all n-tuples a
1
, . . . , a
n
) (U
M
)
n
,
a
1
, . . . , a
n
) P
M
if and only if (a
1
), . . . , (a
n
)) P
M
.
Lemma 4.3.5 (valuations). Let M be an L-structure.
1. There is a unique valuation
v
M
: t : t is a variable-free L-U
M
-term U
M
dened as follows:
(a) v
M
(a) = a for all a U
M
.
(b) v
M
(ft
1
t
n
) = f
M
(v
M
(t
1
), . . . , v
M
(t
n
)) for all n-ary operations f
of L and all variable-free L-U
M
-terms t
1
, . . . , t
n
.
2. There is a unique valuation
v
M
: A : A is an L-U
M
-sentence T, F
dened as follows. For atomic L-U-sentences, we have
v
M
(Pt
1
t
n
) =
_
T if v
M
(t
1
), . . . , v
M
(t
n
)) P
M
,
F if v
M
(t
1
), . . . , v
M
(t
n
)) / P
M
.
For non-atomic L-U
M
-sentences, v
M
(A) is dened as in clauses 2 through
8 of Lemma 2.2.4.
Proof. The proof is as for Lemma 2.2.4.
Denition 4.3.6 (tableau method). The signed and unsigned tableau meth-
ods carry over to predicate calculus with operations. We modify the tableau
rules as follows.
Signed:
.
.
.
TxA
.
.
.
[
TA[x/t]
.
.
.
FxA
.
.
.
[
FA[x/t]
where t is a variable-free term
59
.
.
.
TxA
.
.
.
[
TA[x/a]
.
.
.
FxA
.
.
.
[
FA[x/a]
where a is a new parameter
Unsigned:
.
.
.
xA
.
.
.
[
A[x/t]
.
.
.
xA
.
.
.
[
A[x/t]
where t is a variable-free term
.
.
.
xA
.
.
.
[
A[x/a]
.
.
.
xA
.
.
.
[
A[x/a]
where a is a new parameter
Remark 4.3.7 (soundness and completeness). With the tableau rules as
above, the Soundness Theorem 2.3.9 carries over unchanged to the context of
predicate calculus with operations. The results of Section 2.4 on logical equiv-
alence also carry over. The notion of U-repleteness (Denition 2.5.2) is modi-
ed to say that, for example, if S contains xA then S contains A[x/t] for all
variable-free L-U-terms t. The conclusion of Hintikkas Lemma 2.5.3 is modied
to say that S is satisable in the domain of variable-free L-U-terms. The con-
clusion of the Completeness Theorem 2.5.5 is modied to say that X
1
, . . . , X
k
is satisable in the domain of variable-free L-V -terms. The Compactness The-
orems 2.6.1 and 2.6.2 carry over unchanged.
60
Remark 4.3.8 (satisability in a domain). The notion of satisability in
a domain carries over unchanged to the context of predicate calculus with op-
erations. Theorems 2.2.6 and 2.2.11 on isomorphism, and Theorem 2.7.1 on
satisability in innite domains, also carry over. Theorem 2.7.3 carries over in
an appropriately modied form. See Theorem 4.3.9 and Exercise 4.3.10 below.
Theorem 4.3.9. Let M and M

be L-structures. Assume that : U

M
U
M

is an onto mapping such that conditions 1 and 2 of Denition 4.3.4 hold. Then
as in Theorem 2.2.6 we have v
M
(A) = v
M
(A

) for all L-U

M
-sentences A, where
A

= A[a
1
/(a
1
), . . . , a
k
/(a
k
)]. In particular, M and M

satisfy the same

L-sentences.
Proof. The proof is similar to that of Theorem 2.7.3.
Exercise 4.3.10. Use Theorem 4.3.9 to show that Corollary 2.7.4 carries over
to the context of predicate calculus with operations.
Remark 4.3.11 (companions and proof systems). In our notion of com-
panion (Denition 3.2.3), clauses (1) and (4) are modied as follows:
(1) (xB) B[x/t]
(4) B[x/t] (xB)
where t is any variable-free term. In our Hilbert-style proof system LH, the
instantiation rules are modied as follows:
(a) (xB) B[x/t] (universal instantiation)
(b) B[x/t] (xB) (existential instantiation)
where t is any variable-free term. Also, our Gentzen-style proof system LG is
modied in accordance with the modied tableau rules. With these changes,
the soundness and completeness of LG and LH carry over.
Exercise 4.3.12 (the Interpolation Theorem). Strengthen the Interpola-
tion Theorem 3.5.1 to say that each operation, predicate and parameter occur-
ring in I occurs in both A and B. (Hint: The version with operations can be
deduced from the version without operations.)
Exercise 4.3.13. Skolemization.
4.4 Predicate Calculus with Identity and Oper-
ations
Remark 4.4.1 (predicate calculus with identity and operations). We
augment the identity axioms (Denition 4.1.2) as follows:
5. For each n-ary operation f of L, we have an axiom
x
1
x
n
y
1
y
n
((Ix
1
y
1
& &Ix
n
y
n
) Ifx
1
x
n
fy
1
y
n
).
61
The notions of normal structure and normal satisability are dened as before.
The results of Section 4.1 on predicate calculus with identity carry over un-
changed to predicate calculus with identity and operations. See also Exercise
4.4.2 below.
Exercise 4.4.2 (elimination of operations). Let L be a language with iden-
tity and operations. Let L

be the language with identity and without opera-

tions, obtained by replacing each n-ary operation f belonging to L by a new
(n+1)-ary predicate P
f
belonging to L

. Each normal L-structure M gives rise

to a normal L

-structure M

where
(P
f
)
M
= a
1
, . . . , a
n
, b) (U
M
)
n+1
: f
M
(a
1
, . . . , a
n
) = b.
For each n-ary operation f of L, let B
f
be the L

-sentence
x
1
x
n
y z (Iyz P
f
x
1
x
n
z).
1. Show that to each L-sentence A we may associate an L

-sentence A

such
that, for all L-structures M, M [= A if and only if M

[= A

.
2. Show that a normal L

-structures satises the sentences B

f
, f in L, if and
only if it is of the form M

for some L-structure M.

Exercise 4.4.3. Show that the spectrum problem for predicate calculus with
identity and operations is equivalent to the spectrum problem for predicate
calculus with identity and without operations, as previously discussed in Section
4.2. In other words, given a sentence A involving some operations, construct a
sentence A

involving no operations, such that spectrum(A) = spectrum(A

).
(Hint: Use the result of Exercise 4.4.2.)
Remark 4.4.4 (predicate calculus with equality). The predicate calculus
with identity and operations is well suited for the study of algebraic structures
such as number systems, groups, rings, etc. In such a context, one often writes
t
1
= t
2
instead of It
1
t
2
, and one refers to predicate calculus with equality rather
than predicate calculus with identity. In this notation, the equality axioms (i.e.,
the identity axioms) read as follows:
x(x = x),
xy (x = y y = x),
xy z ((x = y &y = z) x = z),
x
1
y
1
x
n
y
n
((x
1
= y
1
& &x
n
= y
n
) (Px
1
x
n
Py
1
y
n
)),
for each n-ary predicate P,
x
1
y
1
x
n
y
n
((x
1
= y
1
& &x
n
= y
n
) fx
1
x
n
= fy
1
y
n
),
for each n-ary operation f .
One also uses customary algebraic notation, e.g., t
1
+t
2
instead of +t
1
t
2
, t
1
t
2
or t
1
t
2
instead of t
1
t
2
, etc. To avoid ambiguity, parentheses are used.
62
Examples 4.4.5 (groups and rings). Using predicate calculus with identity
and operations, a group may be viewed as an L-structure
G = (U
G
, f
G
, i
G
, e
G
, I
G
).
Here U
G
is the underlying set of the group, and L is the language f, i, e, I,
where f is the group composition law (a binary operation), i is group inversion
(a unary operation), e is the group identity element (a 0-ary operation, i.e., a
constant), and I is the identity predicate (a binary predicate). We could refer
to L as the language of groups. It is customary to write G instead of U
G
, t
1
t
2
or t
1
t
2
instead of ft
1
t
2
, t
1
instead of it, 1 instead of e, and t
1
= t
2
instead of
It
1
t
2
. Thus
G = (G,
G
,
1
G
, 1
G
, =
G
)
and G is required to satisfy the group axioms, consisting of the identity axioms
for L, plus xy z ((xy)z = x(yz)), x(x
1
x = xx
1
= 1), x(1x = x1 = x).
Similarly, a ring may be viewed as a structure
R = (R, +
R
,
R
,
R
, 0
R
, 1
R
, =
R
)
where + and are binary operations, is a unary operation, 0 and 1 are
constants, and = is the equality predicate. We could refer to the language
+, , , 0, 1, = as the language of rings. R is required to satisfy the ring axioms,
consisting of the identity axioms plus xy y ((x + y) + z = x + (y + z)),
xy (x +y = y +x), x(x + 0 = x), x(x + (x) = 0), xy z ((x y) z =
x (y z)), x(x 1 = 1 x = x), xy z (x (y + z) = (x y) + (x z)),
xy z ((x + y) z = (x z) + (y z)), 0 ,= 1.
Exercise 4.4.6. Let G be a group. For a G write a
n
= a a (n times).
Thus a
1
= a and a
n+1
= a
n
a. We say that G is a torsion group if for all a G
there exists a positive integer n such that a
n
= 1. We say that G is torsion-free
if for all a G, if a ,= 1 then a
n
,= 1 for all positive integers n.
1. Show that the class of torsion-free groups can be characterized by a set of
sentences. I.e., there is a set of sentences S such that, for all groups G, G
is torsion-free if and only if G [= S.
2. Show that the class of torsion-free groups cannot be characterized by a
nite set of sentences.
3. Show that the class of torsion groups cannot be characterized by a set of
sentences. I.e., there is no set of sentences S with the property that, for
all groups G, G is a torsion group if and only if G [= S.
4.5 Many-Sorted Predicate Calculus
Denition 4.5.1 (many-sorted languages). A many-sorted language con-
sists of
63
1. a set of sorts , , . . .,
2. a set of predicates P, Q, . . ., each designated as n-ary of type

1

n
for some nonnegative integer n and sorts
1
, . . . ,
n
,
3. a set of operations f, g, . . ., each designated as n-ary of type

1

n

n+1
for some nonnegative integer n and sorts
1
, . . . ,
n
,
n+1
.
Denition 4.5.2 (terms, formulas, sentences). Let L be a many-sorted
language. For each sort , we assume a xed, countably innite set of variables
of sort , denoted x

, y

, z

, . . .. Let U = (U

, U

, . . .) consist of a set U

for
each sort of L. The L-U-terms are generated as follows.
1. Each variable of sort is a term of sort .
2. Each element of U

is a term of sort .
3. If f is an n-ary operaton of type
1

n

n+1
, and if t
1
, . . . , t
n
are terms of sort
1
, . . . ,
n
respectively, then ft
1
. . . t
n
is a term of sort

n+1
.
An atomic L-U-formula is an expression of the form Pt
1
. . . t
n
, where P is an
n-ary predicate of type
1

n
, and t
1
, . . . , t
n
are terms of sort
1
, . . . ,
n
respectively. The L-U-formulas are generated as in Denition 2.1.3, with clause
5 modied as follows:
5

. If x

is a variable of sort , and if A is an L-U-formula, then x

A and
x

A are L-U-formulas.
Our notions of substitution, free and bound variables, sentences, etc. are ex-
tended in the obvious way to the many-sorted context. Naturally, the substitu-
tion A[x

/t] makes sense only when t is a term of sort . An L-formula is an

L-U-formula where U

= for each sort .

Denition 4.5.3 (many-sorted structures). An L-structure M consists of
1. a nonempty set U

M
for each sort of L,
2. an n-ary relation P
M
U
1
M
U
n
M
for each n-ary predicate P of
type
1

n
belonging to L,
3. an n-ary function f
M
: U
1
M
U
n
M
U
n+1
M
for each n-ary operation
of type
1

n

n+1
belonging to L.
64
Notions such as isomorphism, valuation, truth, satisability, and results such
as Theorem 2.2.6 on isomorphism, and Theorem 4.3.9 on onto mappings, carry
over to the many-sorted context in the obvious way.
Denition 4.5.4 (many-sorted domains). We dene a domain for L to be
an indexed family of nonempty sets U = (U

, U

, . . .), where , , . . . are the

sorts of L. In this way, the notion of satisability in a domain generalizes to
the many-sorted context.
Remark 4.5.5 (tableau method, proof systems). For each sort of L, x
a countably innite set V

= a

, b

, . . ., the set of parameters of sort . Then

the tableau method carries over in the obvious way, generalizing Remark 4.3.7.
In the Completeness Theorem for the tableau method, we obtain satisability
in the domain U = (U

, U

, . . .), where U

is the set of variable-free L-V -terms

of sort , with V = (V

, V

, . . .). The soundness and completeness of our proof

systems LH and LG and the Interpolation Theorem also carry over, just as in
Section 4.3.
Remark 4.5.6 (identity predicates). For each sort of L, L may or may not
contain a binary predicate I

of type designated as the identity predicate

for . As identity axioms we may take the universal closures of all L-formulas
of the form
x

(I

xy (AA[x/y]))
where A is atomic. An L-structure M is said to be normal if I

M
= a, a) : a
U

M
for all such that I

belongs to L. The results of Section 4.1 concerning

normal satisability carry over to the many-sorted context.
Denition 4.5.7 (languages with identity). A many-sorted language with
identity is a many-sorted language which contains an identity predicate for each
sort.
Remark 4.5.8 (many-sorted spectrum problem). Let L be a many-sorted
language with identity. If A is an L-sentence and
1
, . . . ,
k
are the sorts occur-
ring in A, the spectrum of A is the set of k-tuples of positive integers (n
1
, . . . , n
k
)
such that there exists a normal L-structure M with U
i
M
of cardinality n
i
, for
i = 1, . . . , k. In this way, the spectrum problem carries over to many-sorted
predicate calculus. So far as I know, the problem of characterizing many-sorted
spectra has not been investigated thoroughly.
Remark 4.5.9 (many-sorted L owenheim/Skolem theorems). It is natu-
ral to try to generalize the Lowenheim/Skolem Theorem 4.1.12 to many-sorted
predicate calculus. This is straightforward provided we consider only normal
structures M where all of the domains U

M
, U

M
, . . . are of the same innite car-
dinality. However, if we require U

M
, U

M
, . . . to be of specied distinct cardinal-
ities, then this leads to dicult issues. Even for two sorts, the topic of so-called
two-cardinal theorems turns out to be rather delicate and complicated. See for
example the model theory textbook of Marker [2].
65
Remark 4.5.10. Our reasons for including many-sorted predicate calculus in
this course are as follows:
1. it is more useful . . . .
FIXME
66
Chapter 5
Theories, Models,
Denability
5.1 Theories and Models
Denition 5.1.1. A theory T consists of a language L, called the language of
T, together with a set of L-sentences called the axioms of T. Thus T = (L, S),
where L is the language of T, and S is the set of axioms of T.
Denition 5.1.2. Let T = (L, S) be a theory.
1. A model of T is an L-structure M such that M satises S. If T contains
identity predicates, then M is assumed to be normal with respect to these
predicates.
2. A theorem of T is an L-sentence A such that A is true in all models of
T, i.e., A is a logical consequence of the axioms of T. Equivalently, A is
derivable in LH(S).
If A is a theorem of T, we denote this by T A.
3. T is consistent if there exists at least one model of T. Equivalently, T is
consistent if and only if there is no L-sentence A such that both T A
and T A. Equivalently, T is consistent if and only if there exists an
L-sentence A such that T , A.
4. T is nitely axiomatized if S is nite.
5. Two theories are equivalent if they have the same language and the same
theorems. I.e., they have the same language and the same models.
6. T is nitely axiomatizable if it is equivalent to a nitely axiomatized theory.
7. T is categorical if T is consistent and all models of T are isomorphic.
67
8. T is complete if T is consistent and all models of T are elementarily equiv-
alent. Equivalently, T is complete if and only if for all L-sentences A either
T A or T A but not both.
Remark 5.1.3. Our formal notion of theory, as dened above, is intended
as a precise explication of the informal notion of deductive scientic theory.
The language of T is the vocabulary of our theory. The theorems of T are
the assertions of our theory. The axioms of T are the basic assertions, from
which all others are deduced. Consistency means that our theory is free of
internal contradictions. Categoricity means that our theory is successful in
that it captures the structure of the underlying reality described by the theory.
Completeness means that our theory is powerful enough to decide the truth of
all statements expressible in the language of the theory.
Remark 5.1.4. In the next two sections we present several interesting examples
of theories. Some of the theories are mathematical, i.e., intended to describe
particular classes of mathematical structures. We shall see that several of these
mathematical theories are complete. On the other hand, some of the theories
are foundational, i.e., intended to serve as a general axiomatic foundation for
all of mathematics, or at least a large part of it. We shall see in Chapter 6 that
these foundational theories are incomplete.
One way to show that a theory is complete is to show that it is categorical.
Theorem 5.1.5. If T is categorical, then (1) T is complete, and (2) the language
of T is a language with identity.
Proof. Assume that T is categorical. Then any two models of T are isomorphic.
Hence by Theorem 2.2.6 (see also Denition 4.5.3), any two models of T are
elementarily equivalent. Hence T is complete. Also, by Theorem 2.7.3 (see also
Theorem 4.3.9 and Denition 4.5.3), T contains an identity predicate for each
sort in the language of T.
On the other hand, we have:
Exercise 5.1.6. Let T be a complete theory in a language with identity. Let
M be a model of T.
1. In the one-sorted case, show that T is categorical if and only if the universe
U
M
is nite.
2. In the many-sorted case with sorts , , . . ., show that T is categorical if
and only if each of the universes U

M
, U

M
, . . . is nite.
Remark 5.1.7. Theorem 5.1.5 and Exercise 5.1.6 show that we cannot use
categoricity as a test for completeness, except in very special circumstances. A
similar but more useful test is provided by the following theorem.
Denition 5.1.8. Let be an innite cardinal number. A one-sorted theory
T is said to be -categorical if all models of T of cardinality are isomorphic.
68
Theorem 5.1.9 (Vaughts Test). Let T be a one-sorted theory. Assume that
(a) T is consistent, (b) all models of T are innite, and (c) there exists an innite
cardinal the cardinality of the language of T such that T is -categorical.
Then T is complete.
Proof. Suppose T is not complete. Since T is consistent, there exist M
1
and M
2
models of T which are not elementarily equivalent. By assumption (b), M
1
and
M
2
are innite. Let be an innite cardinal the cardinality of the language
of T. By the L owenheim/Skolem Theorem 4.1.12, there exist models M

1
, M

2
of cardinality elementarily equivalent to M
1
, M
2
respectively. Clearly M

1
, M

2
are not elementarily equivalent. Hence, by Theorem 2.2.6, M

1
and M

2
are not
isomorphic. This contradicts assumption (c).
Exercise 5.1.10. Generalize Vaughts Test to many-sorted theories.
5.2 Mathematical Theories
In this section we give several examples of theories suggested by abstract algebra
and other specic mathematical topics. We point out that several of these
mathematical theories are complete.
Example 5.2.1 (groups). The language of groups consists of a binary opera-
tion (multiplication), a unary operation
1
(inverse), a constant 1 (the identity
element), and a binary predicate = (equality). The theory of groups consists of
the equality axioms plus
x, y z (x (y z) = (x y) z) (associativity),
x(x x
1
= x
1
x = 1) (inverses),
x(x 1 = 1 x = x) (identity).
A group is a model of the theory of groups. A group is said to be Abelian if it
satises the additional axiom
xy (x y = y x) (commutativity).
A torsion group is a group G such that for all a G there exists a positive
integer n such that a
n
= 1. A group G is torsion-free if for all a G, if a ,= 1
then a
n
,= 1 for all positive integers n. Note that G is torsion-free if and only if
it satises the axioms x(x
n
= 1 x = 1) for n = 2, 3, . . .. A group is said to
be divisible if it satises the axioms xy (y
n
= x), for n = 2, 3, . . ..
Exercise 5.2.2. Show that the theory of torsion-free Abelian groups is not
nitely axiomatizable. Deduce that the theory of torsion-free groups is not
nitely axiomatizable.
69
Exercise 5.2.3. Show that there exist Abelian groups G
1
and G
2
such that
G
1
is a torsion group, G
1
is elementarily equivalent to G
2
, yet G
2
is not a
torsion group. Deduce that there is no theory in the language of groups whose
models are precisely the Abelian torsion groups. Hence, there is no theory in
the language of groups whose models are precisely the torsion groups.
Remark 5.2.4. Let DAG
0
be the theory of innite torsion-free divisible Abelian
groups. It can be shown that DAG
0
is -categorical for all uncountable cardinals
. (This is because such groups may be viewed as vector spaces over the rational
eld, Q.) It follows by Vaughts Test that DAG
0
is complete.
Example 5.2.5 (linear orderings). The language of linear orderings con-
sists of a binary predicate < plus the equality predicate =. The axioms for
linear orderings are xy z ((x < y &y < z) x < z), and x(x < x), and
xy (x < y x = y x > y). A linear ordering is a model of these axioms.
A linear ordering is said to be nontrivial if it satises xy (x < y). It is
said to be dense if it is nontrivial and satises xy (x < y z (x < z < y)).
It is said to be without endpoints if it satises xy (y < x) and xy (y > x).
It is said to have endpoints if it satises xy (y < x) and xy (y > x).
An example of a dense linear ordering without endpoints is (Q, <), where Q is
the set of rational numbers, and < is the usual ordering of Q.
Remark 5.2.6. It can be shown that, up to isomorphism, (Q, <) is the unique
countable dense linear ordering without endpoints. (This is proved by a back-
and-forth argument.) Hence, if we let DLO be the theory of dense linear ordering
without endpoints, DLO is
0
-categorical. It follows by Vaughts Test that DLO
is complete.
Example 5.2.7 (graphs). The language of graphs consists of a binary pred-
icate, R, plus the equality predicate, =. The theory of graphs consists of the
equality axioms plus x(Rxy Ryx) and x(Rxx). A graph is a model of
the theory of graphs.
Thus a graph is essentially an ordered pair G = (V
G
, R
G
), where V
G
is a
nonempty set and R
G
is a symmetric, irreexive relation on V
G
. The elements
of V
G
are called vertices. Two vertices a, b V
G
are said to be adjacent if
a, b) R
G
. A path from a to b is a nite sequence of pairwise distinct vertices
a = v
0
, v
1
, . . . , v
n
= b such that a = v
0
is adjacent to v
1
, v
1
is adjacent to v
2
,
. . . , v
n1
is adjacent to v
n
= b. G is said to be connected if for all a, b V
G
there exists a path from a to b. Equivalently, G is connected if and only if, for
all partitions of V
G
into two disjoint nonempty sets X and Y , there exist a X
and b Y such that a and b are adjacent.
Exercise 5.2.8. Show that there exist graphs G
1
and G
2
such that G
1
is
connected, G
1
is elementarily equivalent to G
2
, yet G
2
is not connected. Deduce
that there is no theory T in the language of graphs such that the models of T
are exactly the connected graphs.
Exercise 5.2.9. A graph G is said to be random if for all nite sets of vertices
a
1
, . . . , a
m
, b
1
, . . . , b
n
there exists a vertex c such that c is adjacent to a
1
, . . . , a
m
70
and not adjacent to b
1
, . . . , b
n
. Show that the theory of random graphs is
0
-
categorical. It follows by Vaughts Test that this theory is complete.
Example 5.2.10 (rings). The language of rings consists of binary operations
+ and (addition and multiplication), a unary operation (subtraction), con-
stants 0 and 1 (the additive and multiplicative identity elements), and a binary
predicate = (equality). The theory of rings consists of the equality axioms plus
xy z (x + (y +z) = (x +y) +z),
xy (x +y = y +x),
x(x + (x) = 0),
x(x + 0 = x),
xy z (x (y z) = (x y) z),
xy z (x (y +z) = (x y) + (x z)) (left distributivity),
xy z ((x +y) z = (x z) + (y z)) (right distributivity),
x(x 1 = 1 x = x),
x(x 0 = 0 x = 0),
0 ,= 1.
A ring is a model of the theory of rings. A ring is said to be commutative if it
satises the additional axiom
xy (x y = y x).
An example of a commutative ring is the ring of integers,
Z = . . . , 2, 1, 0, 1, 2, . . ..
An example of a non-commutative ring is the ring of 2 2 matrices.
Example 5.2.11 (elds). A eld is a commutative ring satisfying the addi-
tional axiom
x(x ,= 0 y (x y = 1)).
A eld is said to be of characteristic 0 if it satises
1 + + 1
. .
n
,= 0
for all positive integers n. Familiar elds such as the eld of rational numbers,
Q, the eld of real numbers, R, and the eld of complex numbers, C, are of
characteristic 0. It can be shown that if a eld satises
1 + + 1
. .
n
= 0
71
for some positive integer n, then the least such n is a prime number, p. In
this case, our eld is said to be of characteristic p. An example of a eld of
characteristic p is the ring of integers modulo p.
A eld F is said to be algebraically closed if for all nontrivial polynomials
f(z) = a
n
z
n
+ + a
1
z + a
0
, a
n
, . . . , a
1
, a
0
F, a
n
,= 0, n 1, there exists
z F such that f(z) = 0. It is known that the complex eld C is algebraically
closed. (This is the so-called Fundamental Theorem of Algebra.) Note that a
eld is algebraically closed if and only if it satises the axioms
x
0
x
1
x
n
(x
n
,= 0 z (x
n
z
n
+ +x
1
z +x
0
= 0))
for n = 1, 2, 3, . . ..
Exercise 5.2.12. Show that the theory of elds of characteristic 0 is not nitely
axiomatizable. Show that the theory of algebraically closed elds is not nitely
axiomatizable. Show that the theory of algebraically closed elds of character-
istic 0 is not nitely axiomatizable.
Remark 5.2.13. Let ACF
0
be the theory of algebraically closed elds of char-
acteristic 0. For each prime number p, let ACF
p
be the theory of algebraically
closed elds of characteristic p. It can be shown that the theories ACF
0
and
ACF
p
are -categorical for all uncountable cardinals . It follows by Vaughts
Test that these theories are complete.
Example 5.2.14 (vector spaces). The language of vector spaces is a 2-sorted
language with sorts and , denoting scalars and vectors respectively. For the
scalars we have binary operations + and of type , a unary operation
of type , constants 0 and 1 of type , and an equality predicate = of
type . For the vectors we have a binary operation + of type ,
a unary operation of type , a constant 0 of type , and an equality
predicate = of type . In addition we have a mixed binary operation of
type , denoting scalar multiplication.
The theory of vector spaces consists of the eld axioms for scalars, the
Abelian group axioms for vectors, and the axioms
x

(x (v +w) = (x v) + (x w)),
x

(x (v) = (x v)),
x

(x 0

= 0),
x

((x +y) v = (x v) + (y v)),

x

((x) v = (x v),
x

((x y) v = x (y v)),
v

(1 v = v),
v

(0 v = 0)
72
for scalar multiplication. A vector space is a model of these axioms.
If V is a vector space, a set of vectors S in V is said to span V if every
vector v in V can be written as a linear combination of vectors in S, i.e., v =
a
1
v
1
+ +a
n
v
n
for some v
1
, . . . , v
n
S and scalars a
1
, . . . , a
n
. An important
invariant of a vector space is its dimension, i.e., the minimum cardinality of a
spanning set. It can be shown that, up to isomorphism over a eld F, the unique
vector space over F of dimension is the familiar space F

. The vectors of F

are sequences a
i
: i < ), with a
i
F for all i, and a
i
= 0 for all but nitely
many i. Vector addition is given by
a
i
: i < ) +b
i
: i < ) = a
i
+b
i
: i < ).
Scalar multiplication is given by
c a
i
: i < ) = c a
i
: i < ).
Example 5.2.15 (ordered algebraic structures). The language of ordered
rings consists of the language of rings +, , , 0, 1, =, together with <. The
ordered eld axioms consist of the eld axioms, plus the linear ordering axioms,
plus
xy z (x < y x +z < y +z),
xy ((x > 0 & y > 0) x y > 0).
An ordered eld is a model of these axioms. An example of an ordered eld is
the eld of rational numbers (Q, <) with its usual ordering.
An ordered eld (F, <) is said to be real closed if for all polynomials f(x) =
a
n
x
n
+ +a
1
x +a
0
, a
n
, . . . , a
1
, a
0
F, and for all b, c F, if f(b) < 0 < f(c)
then there exists x F between b and c such that f(x) = 0. Clearly the
ordered eld of real numbers (R, <) is real closed. Note that an ordered eld is
real closed if and only if it satises the axioms
u v w
0
w
1
w
n
((u < v & w
n
u
n
+ +w
1
u +w
0
< 0 < w
n
v
n
+ +w
1
v +w
0
)
x(u < x < v & w
n
x
n
+ +w
1
x +w
0
= 0))
for n = 1, 2, 3, . . ..
Remark 5.2.16. Let RCOF be the theory of real closed ordered elds. A
famous and important theorem of Tarski says that RCOF is complete. This
holds despite the fact that RCOF is not -categorical for any .
Exercise 5.2.17. Which of the following theories are complete? Justify your
answers.
1. The theory of dense linear orderings with end points.
2. The theory of elds of characteristic 0.
73
3. The theory of innite, torsion-free, Abelian groups.
4. The theory of nite-dimensional vector spaces over a eld of 5 elements.
5. The theory of innite-dimensional vector spaces over a eld of 5 elements.
5.3 Foundational Theories
Example 5.3.1 (rst order arithmetic).
Example 5.3.2 (second order arithmetic).
Example 5.3.3 (set theory).
Remark 5.3.4 (practical completeness).
5.4 Denability over a Structure
Denition 5.4.1 (explicit denability). Let L be a language, let M be an
L-structure, and let R be an n-ary relation on U
M
. We say that R is explicitly
denable over M if there exists an L-formula A with n free variables x
1
, . . . , x
n
,
such that
R = a
1
, . . . , a
n
) [ M [= A[x
1
/a
1
, . . . , x
n
/a
n
].
Example 5.4.2. Consider the binary relation <= a, b) R
2
[ a < b on
the set R of real numbers. Viewing R as a commutative ring, we see that for
all a, b R, a < b if and only if R [= x(x ,= 0 & a + x
2
= b). Thus < is
explicitly denable over the commutative ring R = (R, +, , , 0, 1, =) by the
formula z (z ,= 0 & x +z
2
= y) with free variables x, y.
Example 5.4.3. Let N = (N, +, , , 0, 1, <, =) be the natural number system.
It can be shown that the class of relations which are explicitly denable over N
includes all relations which are computable in the sense of Turing. In particular,
the 3-ary relation m, n, k) N
3
[ m
n
= k is denable over N. The proof of
this result involves some non-trivial number theory.
Denition 5.4.4 (implicit denability). Let L be a language, let M be an
L-structure, and let R be an n-ary relation on U
M
. We say that R is implicitly
denable over M if there exists a sentence A

in the language L

= L R
with an additional n-ary predicate R, such that for all n-ary relations R

on
U
M
, (M, R

) [= A

if and only if R

= R.
Exercise 5.4.5. Let M be a structure. Show that any relation which is ex-
plicitly denable over M is implicitly denable over M. Show that any relation
which is implicitly denable over M is invariant under all automorphisms of
M. Give counterexamples showing that the converses of these assertions fail in
general.
74
Example 5.4.6. It can be shown that there exists a subset of N which is
implicitly denable over the natural number system N = (N, +, , 0, 1, <, =) but
is not explicitly denable over N. See Remark 6.4.6 and Exercise 6.4.7 below.
Exercise 5.4.7. Let R = (R, +, , , 0, 1, <, =) be the ordered eld of real
numbers. Show that the relations y = sinx and y = e
x
are implicitly denable
over R. (Hint: These relations are dened by dierential equations which can
be expressed as formulas of the predicate calculus, using the --method.) It
can be shown that these relations are not explicitly denable over R.
Exercise 5.4.8. Show that if U
M
is nite, then any relation which is implicitly
denable over M is explicitly denable over M.
Exercise 5.4.9. (In this exercise we assume familiarity with saturated models.)
Show that if a structure M is saturated, then any relation which is implicitly
denable over M is explicitly denable over M.
Remark 5.4.10. In this section we have considered explicit and implicit den-
ability over a structure, M. We have given examples showing that, in general,
implicit denability over M does not imply explicit denability over M.
In the next section we shall consider the related but more restrictive notions
of explicit and implicit denability over a theory, T. It will be obvious that
explicit denability over T implies explicit denability over any structures which
happena to be a model of T, and implicit denability over T implies implicit
denability over any structures which happens to be a model of T.
A pleasant surprise is that explicit denability over T is equivalent to implicit
denability over T. This is the content of Beths Denability Theorem, Theorem
5.5.3 below.
5.5 Denibility over a Theory
Theorem 5.5.1 (adding a new predicate).
Theorem 5.5.2 (adding a new operation).
Theorem 5.5.3 (the Beth Denability Theorem).
Proof. This is an application of the Interpolation Theorem, Section 5.6.
5.6 Interpretability
Denition 5.6.1. We say that T
1
is interpretable in T
2
if T
1
is a subtheory of
a denitional extension of T
2
.
Example 5.6.2. First order arithmetic is interpretable in second order arith-
metic. First and second order arithmetic are interpretable in set theory. It can
be shown that set theory is not interpretable in second order arithmetic, and
second order arithmetic is not interpretable in rst order arithmetic. Results of
this kind follow from G odels Second Incompleteness Theorem.
75
Chapter 6
Arithmetization of
Predicate Calculus
6.1 Primitive Recursive Arithmetic
Denition 6.1.1. To each natural number n we associate a PRA-term n as
follows: 0 = 0, n + 1 = S(n). These terms are known as numerals.
Theorem 6.1.2. Let f be a k-ary primitive recursive function. Then for all
k-tuples of natural numbers m
1
, . . . , m
k
we have
PRA f m
1
m
k
= f(m
1
, . . . , m
k
) .
Proof.
6.2 Interpretability of PRA in Z
1
6.3 G odel Numbers
Let L be a countable language. Assume that to all the sorts , predicates P,
and operations f of L have been assigned distinct positive integers #(), #(P),
#(f) respectively.
Denition 6.3.1 (Godel numbers). To each term t and formula A of L we
assign a positive integer, the G odel number of t or of A, denoted #(t) or #(A),
76
respectively.
#(v

i
) = 2 3
#()
5
i
#(a

i
) = 2
2
3
#()
5
i
#(ft
1
t
n
) = 2
3
3
#(f)
p
#(t1)
2
p
#(tn)
n+1
if f is an n-ary operation
#(Pt
1
t
n
) = 2
4
3
#(P)
p
#(t1)
2
p
#(tn)
n+1
if P is an n-ary predicate
#(A) = 2
5
3
#(A)
#(A&B) = 2
6
3
#(A)
5
#(B)
#(AB) = 2
7
3
#(A)
5
#(B)
#(AB) = 2
8
3
#(A)
5
#(B)
#(AB) = 2
9
3
#(A)
5
#(B)
#(v A) = 2
10
3
#(v)
5
#(A)
#(v A) = 2
11
3
#(v)
5
#(A)
Denition 6.3.2. The language L is said to be primitive recursive if the fol-
lowing items are primitive recursive.
Sort(x) x = #() for some sort
Pred(x) x = #(P) for some predicate P
Op(x) x = #(f) for some operation f
arity(#(P)) = n if P is an n-ary predicate
arity(#(f)) = n if f is an n-ary operation
sort(#(P), i) = #(
i
) if 1 i n and P is an n-ary predicate of
type
1

n
sort(#(f), i) = #(
i
) if 1 i n + 1 and f is an n-ary operation
of type
1

n

n+1
Lemma 6.3.3. If L is primitive recursive, then the following are primitive
recursive.
Var(x) x = #(v) for some variable v
Param(x) x = #(a) for some parameter a
Term(x) x = #(t) for some term t
ClTerm(x) x = #(t) for some closed term t
AtFml(x) x = #(A) for some atomic formula A
Fml(x) x = #(A) for some formula A
sort(#(t)) = #() if t is a term of sort
Proof. We have
Var(x) (x)
0
= 1 & x = 2
(x)0
3
(x)1
5
(x)2
& Sort((x)
1
)
77
and
Param(x) (x)
0
= 2 & x = 2
(x)0
3
(x)1
5
(x)2
& Sort((x)
1
) .
To show that the predicate Term(x) is primitive recursive, we rst show that the
function sort(x) is primitive recursive, where sort(#(t)) = #() if t is a term of
sort , sort(x) = 0 otherwise. Put lh(x) = least w < x such that (x)
w
= 0. We
then have
sort(x) =
_

_
(x)
1
if Var(x) Param(x) ,
sort((x)
1
, lh(x)

1) if (x)
0
= 3 & Op((x)
1
) & (+) ,
0 otherwise ,
where
(+) arity((x)
1
) = lh(x)

2 & x =

lh(x)

1
i=0
p
(x)i
i
& (i < lh(x)

2) (sort((x)
i+2
) = sort((x)
1
, i + 1)) .
Then
Term(x) sort(x) > 0 .
For closed terms, dene clsort(x) like sort(x) replacing Var(x) Param(x) by
Param(x). We then have
ClTerm(x) clsort(x) > 0 .
For formulas we have
AtFml(x) ((x)
0
= 4 & Pred((x)
1
) & (+))
and
Fml(x) AtFml(x) ((x)
0
= 5 & x = 2
(x)0
3
(x)1
& Fml((x)
1
))
(6 (x)
0
9 & x = 2
(x)0
3
(x)1
5
(x)2
& Fml((x)
1
) & Fml((x)
2
))
(10 (x)
0
11 & x = 2
(x)0
3
(x)1
5
(x)2
& Var((x)
1
) & Fml((x)
2
))
and this completes the proof.
Lemma 6.3.4 (substitution). There is a primitive recursive function sub(x, y, z)
such that for any formula A and any variable v and any closed term t,
sub(#(A), #(v), #(t)) = #(A[v/t]) .
Proof.
sub(x, y, z) =
_

_
z if x = y
2
(x)0
3
(x)1

lh(x)

1
i=2
p
sub((x)i,y,z)
i
if 3 (x)
0
4
2
(x)0
3
sub((x)1,y,z)
if (x)
0
= 5
2
(x)0
3
sub((x)1,y,z)
5
sub((x)2,y,z)
if 6 (x)
0
9
2
(x)0
3
(x)1
5
sub((x)2,y,z)
if 10 (x)
0
11 &(x)
1
,= y
x otherwise.
78

Lemma 6.3.5. If L is primitive recursive, then the predicate

Snt(x) x = #(A) for some sentence A
is primitive recursive.
Proof. Recall that, by Exercise 2.1.10, a formula A is a sentence if and only if
A[v/a] = A for all variables v occurring in A. Note also that if y = #(v

i
) then
2y = #(a

i
). Thus we have
Snt(x) Fml(x) & (y < x) (Var(y) x = sub(x, y, 2y)) .

Lemma 6.3.6. There is a primitive recursive function num(x) such that

num(n) = #(n)
for any nonnegative integer n.
Proof. The recursion equations for num(x) are
num(0) = #(0) ,
num(x + 1) = 2
3
3
#(S)
5
num(x)
.

6.4 Undenability of Truth

In this section, let T be a theory which includes PRA. For example, we could
take T to be PRA itself. Or, by Section 6.2, we could take T to be an appropriate
denitional extension of Z
1
or Z
2
or ZFC.
Lemma 6.4.1 (Self-Reference Lemma). Let L be the language of T. Let A
be an L-formula with a free number variable x. Then we can nd an L-formula
B such that
T BA[x/#(B) ] .
The free variables of B are those of A except for x. In particular, if x is the
only free variable of A, then B is an L-sentence.
Proof. Put d(z) = sub(z, #(x), num(z)). Thus d is a 1-ary primitive recursive
function such that, if A is any L-formula containing the number variable x
as a free variable, then d(#(A)) = #(A[x/#(A) ]. Now given A as in the
hypothesis of the lemma, let D be the formula A[x/d x], and let B be the
formula A[x/d #(D) ], i.e., D[x/#(D) ]. Note that d(#(D)) = #(B). It follows
by Theorem 6.1.2 that PRA d #(D) = #(B). Since T includes PRA, it
follows that T d #(D) = #(B). Hence T A[x/d #(D)] A[x/#(B) ]. In
other words, T BA[x/#(B) ]. This completes the proof.
79
Denition 6.4.2. If M is any model of T, let True
M
be the set of Godel
numbers of sentences that are true in M, i.e.,
True
M
= #(B) : B is a sentence and M [= B.
Denition 6.4.3. An -model of T is a model M of T such that the number
domain of M is = 0, 1, 2, . . . and 0
M
= 0 and S
M
(n) = n + 1 for all n .
More generally, if M is any model of T, we may assume that is identied
with a subset of the number domain of M in such a way that 0
M
= 0 and
S
M
(n) = n + 1 for each n . Thus each n is identied with the element
of M that is denoted by n, i.e., n = v
M
(n).
Theorem 6.4.4 (undenability of truth). If M is an -model of T, then
True
M
is not explicitly denable over M. More generally, if M is any model of T,
then the characteristic function of True
M
is is not included in the characteristic
function of any subset of M that is explicitly denable over M.
Proof. Let X be a subset of the number domain of M which is explicitly denable
over M. Let A be an L-formula with a free number variable x and no other free
variables, such that A explicitly denes X over M. Applying Lemma 6.4.1 to
the negation of A, we obtain an L-sentence B such that T BA[x/#(B) ].
Since M is a model of T, it follows that #(B) True
M
if and only if #(B) / X.
Hence the characteristic function of True
M
is not included in the characteristic
function of X, q.e.d.
Corollary 6.4.5. Let M = (, +, , 0, 1, =) be the standard model of rst order
arithmetic, Z
1
. Then True
M
is not explicitly denable over M. This
1
may be
paraphrased by saying that arithmetical truth is not arithmetically denable.
Remark 6.4.6. With M = (, +, , 0, 1, =) as above, it can be shown that
True
M
is implicitly denable over M. (See also Exercise 6.4.7.) Thus the Beth
Denability Theorem does not hold for denability over this particular model.
Exercise 6.4.7. Let M be an -model of Z
1
or Z
2
or ZFC. Let Sat
M
be the
satisfaction relation on M. Show that Sat
M
is implicitly denable over M.
6.5 The Provability Predicate
In this section, let L be a primitive recursive language, and let T be an L-
theory which is primitive recursively axiomatizable. For example, T could be
PRA itself, or T could be any of the mathematical or foundational theories
discussed in Sections 5.2 and 5.3.
Denition 6.5.1. Choose a primitive recursive predicate Ax
T
for the set of
G odel numbers of axioms of T. In terms of Ax
T
show that various predicates
1
This result is due to Tarski [4].
80
associated with T are primitive recursive. Introduce the provability predicate
Pvbl
T
by denition:
Pvbl
T
(x) y Prf
T
(x, y) .
Note that, for all L-sentences B, Pvbl
T
(#(B)) is true if and only if T B.
Lemma 6.5.2 (derivability condition 1). For any L-sentence A, if T A
then
PRA Pvbl
T
(#(A)) .
Proof. Suppose T A. Let p be a proof of A in T. Then we have Prf
T
(#(A), #(p)).
Since Prf
T
(x, y) is a primitive recursive predicate, it follows by Theorem 6.1.2
that PRA Prf
T
(#(A), #(p)). Hence PRA Pvbl
T
(#(A)), q.e.d.
Lemma 6.5.3 (derivability condition 2). For any L-sentence A, we have
PRA Pvbl
T
(#(A)) Pvbl
PRA
(#(Pvbl
T
(#(A)))) .
Proof. This is just Lemma 6.5.2 formalized in PRA. The details of the formal-
ization are in Section 6.7.
Lemma 6.5.4 (derivability condition 3). For any L-sentences A and B, we
have
PRA Pvbl
T
(#(AB)) (Pvbl
T
(#(A)) Pvbl
T
(#(B))) .
Proof. This is a straightforward consequence of the fact that our rules of infer-
ence include modus ponens.
6.6 The Incompleteness Theorems
In this section, let T be a theory which is primitive recursively axiomatizable
and includes PRA. For example, T could be PRA itself, or it could be an
appropriate denitional extension of Z
1
or Z
2
or ZFC. As in Section 6.5, let
Pvbl
T
be a provability predicate for T.
Lemma 6.6.1. Let A(x) be a PRA-formula with one free variable x. Then we
can nd a PRA-sentence B such that PRA BA(#(B)).
Proof. This is the Self-Reference Lemma 6.4.1 specialized to PRA.
Lemma 6.6.2. We can nd a PRA-sentence S such that
() PRA S Pvbl
T
(#(S)) .
Note that S is self-referential and says I am not provable in T.
Proof. This is an instance of Lemma 6.6.1 with A(x) Pvbl
T
(x).
81
Lemma 6.6.3. Let S be as in Lemma 6.6.2. If T is consistent, then T , S.
Proof. Suppose for a contradiction that T S. By Lemma 6.5.2 we have
PRA Pvbl
T
(#(S)). Hence by () it follows that PRA S. Since T includes
PRA, we have T S. Thus T is inconsistent.
Theorem 6.6.4 (the First Incompleteness Theorem). If T is consistent,
then we can nd a sentence S

in the language of rst order arithmetic such

that S

is true yet S

is not a theorem of T.
Proof. Let S be a PRA-sentence as in Lemma 6.6.2. By Lemma 6.6.3, T , S.
It follows by () that S is true. As in Section 6.2, let S

be the translation of S
into the language of rst order arithmetic. Thus S

is also true. By the results

of Section 6.2, PRA S S

. Hence T S S

. Hence T , S

.
Assume now that the primitive recursive predicate Ax
T
has been chosen
in such a way that PRA x(Ax
PRA
(x) Ax
T
(x)). It follows that PRA
x(Pvbl
PRA
(x) Pvbl
T
(x)). In particular we have:
Lemma 6.6.5. For all PRA-sentences A, we have
PRA Pvbl
PRA
(#(A)) Pvbl
T
(#(A)) .
Denition 6.6.6. Con
T
is dened to be the sentence Pvbl
T
(#(F)). Here F is
the identically false sentence. Note that Con
T
is a PRA-sentence which asserts
the consistency of T.
Theorem 6.6.7 (the Second Incompleteness Theorem). If T is consistent,
then T , Con
T
.
Proof. Let S be as in Lemma 6.6.2. By Theorem 6.6.4 we have T , S. Therefore,
to show T , Con
T
, it suces to show T Con
T
S. Since T includes PRA, it
suces to show PRA Con
T
S. By () it suces to show
() PRA Con
T
Pvbl
T
(#(S)) .
But this is just Lemma 6.6.3 formalized in PRA.
Details: We need to prove (). Reasoning in PRA, suppose Pvbl
T
(#(S)).
By Lemma 6.5.3 we have Pvbl
PRA
(#(Pvbl
T
(#(S)))). Moreover, from () and
Lemma 6.5.2 we have Pvbl
PRA
(#(S Pvbl
T
(#(S)))). Hence by Lemmas
6.5.2 and 6.5.4 we have Pvbl
PRA
(#(S)). By Lemma 6.6.5 it follows that
Pvbl
T
(#(S)). Hence by Lemmas 6.5.2 and 6.5.4 it follows that Pvbl
T
(#(F)),
i.e., Con
T
. This completes the proof.
Exercise 6.6.8. Show that PRA S Con
T
.
6.7 Proof of Lemma 6.5.3
FIXME Write this section!
82
Bibliography
[1] Neil D. Jones and Alan L. Selman. Turing machines and the spectra of
rst-order formulas. Journal of Symbolic Logic, 39:139150, 1974.
[2] David Marker. Model Theory: An Introduction. Springer, 2002. 342 pages.
[3] Raymond M. Smullyan. First-Order Logic. Dover Publications, Inc., 1995.
XII + 158 pages.
[4] Alfred Tarski. Introduction to Logic and to the Methodology of Deductive
Sciences. Oxford University Press, New York, 4th edition, 1994. XXII +
229 pages.
83
Index
arity, 19, 58, 64
assignment, 5
atomically closed, 33
atomic formula, 3, 19, 58, 64
binary, 3
block tableau, 45
modied, 48
bound variable, 20
Churchs Theorem, 40
clause, 9
closed, 13, 30
atomically, 33
conite, 57
Compactness Theorem, 54
companion, 39, 40
completeness, 39
composite number, 57
congruence, 53, 54
constant, 58
degree, 3, 19
disjunctive normal form, 9
domain, 65
dyadic, 16
end node, 15
equality, 62
equivalence relation, 56
falsity, 22
Fibonacci numbers, 58
nitely branching, 16
formation sequence, 4
formation tree, 4
formula, 3, 19, 58, 64
free variable, 20
Gentzen-style proof system, 47
G odel number, 74
graph, 17
group, 63
Hilbert-style proof system, 38
Hintikkas Lemma, 14, 30
identity, 65
identity axioms, 53, 61, 65
identity predicate, 53
immediate extension, 10, 26
immediate predecessor, 15
immediate successor, 15
interpolation, 49, 51
interpretable, 73
isomorphism, 21, 59
K onigs Lemma, 16
language, 3, 19, 58
primitive recursive, 75
LG, 47, 61, 65
LG

, 48
LG(atomic), 48
LG(symmetric), 49
LH, 42, 61, 65
LH

, 44
logical consequence, 7, 22, 25, 30
logical equivalence, 8, 28
logical validity, 7, 22, 25, 27, 48, 49
many-sorted, 63
modus ponens, 39
n-ary, 19
n-ary operation, 58
normal satisability, 54, 62
84
normal structure, 53, 62, 65
numeral, 74
open, 13, 14, 30
operation, 58, 64
ordering
partial, 18
parameter, 23
parameters, 65
partial ordering, 18
partition, 56
path, 15
predecessor, 15
predicate, 19, 64
prenex form, 29
prime numbers, 57
primitive recursive language, 75
proof system, 38
Gentzen-style, 47
Hilbert-style, 38
quantier, 19
quantier-free, 29
quasitautology, 39
replete, 14, 30
ring, 63
root, 15
satisfaction, 7, 22, 25
satisability, 7, 22, 23, 25, 27
sentence, 20, 21
sequent, 46
sequent calculus, 47
signed formula, 10
signed sequent, 49
signed variant, 49
sort, 64
soundness, 39
spectrum, 56, 65
spectrum problem, 56, 62, 65
structure, 21, 25
sub, 76
subtree, 15
successor, 15
symmetric, 49
tableau, 10, 13, 23, 59
tautology, 7, 39
term, 58, 64
torsion, 63, 68
tree, 15
truth, 22
truth values, 5
unary, 3
universal closure, 29
unsigned formula, 10
unsigned sequent, 49
valuation, 5, 21, 59
variable, 19
bound, 20
free, 20
variable-free, 58
variables, 64
variant, 49
85