Mathematical Logic

(The Berkeley undergraduate course)

Theodore A. Slaman
W. Hugh Woodin
Contents
1 Propositional logic 1
1.1 The language 1
1.2 Truth assignments 5
1.3 A proof system for L
0
10
1.4 Logical implication and compactness 18
2 First order logicsyntax 21
2.1 Terms 21
2.2 Formulas 23
2.3 Subformulas 24
2.4 Free variables, bound variables 25
3 First order logicsemantics 29
3.1 Formulas and structures 29
3.2 The satisfaction relation 29
3.3 Substitution and the satisfaction relation 33
4 The logic of rst order structures 39
4.1 Isomorphisms between structures 39
4.2 Substructures and elementary substructures 42
4.3 Denable sets and Tarskis Criterion 42
4.4 Dense orders 46
4.5 Countable sets 48
4.6 The Lowenheim-Skolem Theorem 50
4.7 Arbitrary dense total orders 51
5 The Godel Completeness Theorem 55
5.1 The notion of proof 55
5.2 Soundness 57
5.3 Deduction and generalization theorems 59
5.4 The Henkin property 62
5.5 Extensions of consistent sets of L formulas 70
5.6 The Godel Completeness Theorem 72
5.7 The Craig Interpolation Theorem 74
6 The Compactness Theorem 77
6.1 Applications of the Compactness Theorem 77
6.2 Types 79
6.3 Vaughts conjecture 86
iv Contents
7 More on the logic of L
A
structures 87
7.1 Elimination of Quantiers 87
7.2 Elimination of Quantiers for the theory of Dense Orders 89
7.3 Elementary chains 90
7.4 Homogeneous Structures 91
7.5 Model Completeness 92
7.6 Elimination of Quantiers for the theory of N, <, S, 1 94
7.7 Presburger Arithmetic 96
Index 102
1
Propositional logic
Propositional logic governs the way by which propositions are combined in
compound sentences.
Informally, a proposition is a declarative sentence, such as any of the follow-
ing.
Life is nothing but a competition to be the criminal rather than the victim.
(B. Russell)
Life is as tedious as a twice-told tale. (W. Shakespeare)
Life is a dead-end street. (H. L. Menken)
Life is too short to learn German. (R. Porson)
Propositions may be combined by logical connectives to form more complicated
statements, such as If life is a dead-end street, then life is too short to learn
German or If it is not the case that life is too short to learn German, then life
is as tedious as a twice told tale. The truth or falsity of a compound statement
depends solely on that of its parts. Understanding propositional logic is just
understanding that dependence.
In the next section, we will introduce symbols A
n
for propositions, for nega-
tion, and for implication. Using A
1
, . . . , A
4
to denote the above propositions,
our two compound sentences would be denoted by (A
3
A
4
) and ((A
4
) A
2
).
1.1 The language
Our language for propositional logic consists of (certain) nite sequences of sym-
bols.
Denition 1.1 The logical symbols are the following symbols.
( )
The propositional symbols are A
n
, for n in N. (N is the set of non-negative
integers; i.e. the set of natural numbers)
Denition 1.2 If s = s
1
, . . . , s
n
and t = t
1
, . . . , t
m
are nite sequences, we
let s +t denote the nite sequence
s +t = s
1
, . . . , s
n
, t
1
, . . . , t
m
.
2 Propositional logic
Denition 1.3 The propositional language L
0
is the smallest set L of nite
sequences of the above symbols satisfying the following properties.
(1) For each propositional symbol A
n
with n N,
A
n
L.
(2) For each pair of nite sequences s and t, if s and t belong to L, then
(s) L
and
(s t) L.
For the duration of Chapter 1, we will use propositional formula or just
formula to refer to an element of L
0
.
In Denition 1.3, we dened the propositional language L
0
as the smallest
set which is closed under Conditions 1 and 2. In the following, we show that L
0
is well dened.
Theorem 1.4 L
0
is the intersection of all of the sets which satisfy the two
conditions of Denition 1.3.
Proof. Let L
0
be the intersection of all of the sets which satisfy the two conditions
of Denition 1.3. There is at least one such set, since the set of all nite sequences
of symbols does satisfy the two conditions. We claim that L
0
is a set which
satises those two conditions.
For each n N, A
n
is an element of every set which satises Condition 1.
Consequently, A
n
is an element of the intersection of all such sets, and thus it is
an element of L
0
.
Now, suppose that s and t belong to L
0
. Then they belong to every set which
satises Conditions 1 and 2. But then, for every such set, we can apply Condition
2 to conclude that (s) and (s t) also belong to that set. Therefore, (s) and
(s t) belong to the intersection of all such sets, and thus belong to L
0
.
Hence, L
0
satises Conditions 1 and 2. Since it is contained in every set which
also satises those conditions, it must the smallest such set. Consequently, L
0
is
equal to L
0
. .
1.1.1 Subformulas
Denition 1.5 Suppose that s = s
1
, . . . , s
n
is a nite sequence. A nite se-
quence t is a block-subsequence of s if there exist non-negative integers i and j
such that
(1) i +j n,
(2) t = s
i
, s
i+1
, . . . , s
i+j
.
Example 1.6 (1) 3 is a block-subsequence of 1, 2, 3, 4, 5, 6.
The language 3
(2) 3, 4, 5 is a block-subsequence of 1, 2, 3, 4, 5, 6.
(3) 1, 6 is not a block-subsequence of 1, 2, 3, 4, 5, 6.
(4) If s is a nite sequence and s has length n, then there are at most
n

i=1
(n i) + 1 = n
2
+n (1/2)n(n + 1) = 1/2(n + 1)n
block-subsequences of s.
Denition 1.7 (1) A sequence a
1
, . . . , a
k
is an initial segment of another se-
quence b
1
, . . . , b
m
if and only if k is less than or equal to m and for all i k,
a
i
= b
i
. In other words, b
1
, . . . , b
m
is equal to a
1
, . . . , a
k
+b
k+1
, . . . , b
m
,
where b
k+1
, . . . , b
m
could be the empty sequence.
(2) When m is greater than k, we say that a
1
, . . . , a
k
is a proper initial segment
of a
1
, . . . , a
k
, b
k+1
, . . . , b
m
.
Denition 1.8 Suppose that is a formula. A formula is a subformula of
if is a block-subsequence of .
The subformulas of are precisely the formulas which arise in the denition
of .
Denition 1.9 Suppose that
= a
1
, . . . , a
n

is a formula.
Suppose s is a nite sequence. An occurrence of s in is an interval
[j
1
, j
2
]
such that s = a
j1
, . . . , a
j2
.
Remark 1.10 Suppose that is a formula. A formula is a subformula of
if and only if has an occurrence in .
Lemma 1.11 (Readability) Suppose that is a formula in L
0
. Then exactly
one of the following conditions applies.
(1) There is an n such that = A
n
.
(2) There is a L
0
such that = ().
(3) There are
1
and
2
in L
0
such that = (
1

2
).
Proof. Consider the subset L of L
0
which consists of those formulas which sat-
isfy the above three clauses. By the rst clause, if n N, then A
n
L. Con-
sequently, L satises Condition 1 of Denition 1.3. Secondly, if is in L, then
L
0
and so () L
0
. But then () is an element of L
0
which satises the
4 Propositional logic
second of the above clauses, and hence () L. Similarly, if
1
and
2
belong
to L, then so does (
1

2
). Thus, L satises Condition 2 of Denition 1.3. It
follows that L
0
L, and so L
0
= L.
It remains to show that the three possibilities are mutually exclusive.
Clearly, the rst case excludes the other two since both of the formulas in
the latter two cases begin with the symbol (. Now, if = (), then the second
symbol in is . However, if = (
1

2
), then the second symbol in is the
rst symbol in
1
, which by the above is either an A
n
or (. Consequently, these
two cases are mutually exclusive. .
We now prove a technical lemma which we will apply to show that the sub-
formulas of mentioned above are uniquely determined.
Lemma 1.12 If L
0
, then no proper initial segment of is an element of
L
0
.
Proof. We prove Lemma 1.12 by induction on n.
If has length 1 then the only subsequence to be considered is the empty
sequence, which by Lemma 1.11 is not an element of L
0
.
Now suppose that L
0
has length n, n > 1, and Lemma 1.12 holds for all
m less than n. By Lemma 1.11, since has length greater than 1, has one of
two forms: () or (
1

2
).
Suppose that is (). For a contradiction, suppose that L
0
is a proper
initial segment of (). Then the rst symbol in is (, so is not of the formA
i
,
and by Lemma 1.11 the length of is greater than one. Thus, the second symbol
in is , which by Lemma 1.11 is not the rst symbol of any element of L
0
, and
so cannot be of the form (
1

2
). Consequently, there is a
1
L
0
such that
is equal to (
1
). But then () has (
1
) as a proper initial segment, and so
has
1
as a proper initial segment, contradiction to the induction hypothesis.
Finally, suppose that is (
1

2
) and that L
0
is a proper initial
segment of . We can apply Lemma 1.11 and argue as in the previous paragraph
that there are
1
and
2
in L
0
such that = (
1

2
). But then either
1
is
a proper initial segment of
1
(a contradiction),
1
is a proper initial segment
of
1
(a contradiction), or
1
=
1
and
2
is a proper initial segment of
2
(a
contradiction).
In either case, has no proper initial segment in L
0
. .
Theorem 1.13 (Unique Readability) Suppose that is a formula in L
0
.
Then exactly one of the following conditions applies.
(1) There is an n such that = A
n
.
(2) There is a L
0
such that = ().
(3) There are
1
and
2
in L
0
such that = (
1

2
).
Further, in cases (2) and (3), the formulas , and
1
and
2
are unique, respec-
tively.
Truth assignments 5
Proof. By Lemma 1.11, it is enough to check the claim of uniqueness.
First, suppose that = () and = (). Thus, the sequence of symbols
can be read as (, + + ) and as (, + + ). The occurrences of
and within have the same length and the same elements, and therefore are
equal.
Finally, suppose that = (
1

2
) and = (
1

2
). Since both
1
and

1
belong to L
0
, by Lemma 1.12 neither can be a proper initial segment of the
other. Since they are initial segments of each other, they must be equal. As in
the case of negation, it follows that
2
and
2
are also equal. .
1.1.2 Exercises
(1) Give three examples of elements of L
0
with at least 15 symbols each. The
examples should have an interesting structure. For one of these examples,
give a meaningful sentence which has the same structure.
(2) For which natural numbers n are there elements of L
0
of length n?
(3) Show that a sequence is an element of L
0
if and only if there is a nite
sequence of sequences
1
, . . . ,
n
such that
n
= , and for each if i less
than or equal to n either there is an m such that
i
= A
m
, or there is a j
less than i such that
i
is equal to (
j
), or there are j
1
and j
2
less than i
such that
i
is equal to (
j1

j2
).
(4) Give an algorithm (suitable to be programmed for a computer) to determine
whether a given nite sequence belongs to L
0
.
(5) Consider the set of symbols and #. Let L

be the smallest set L of se-

quences of these symbols with the following properties.
a) The length one sequences and # belong to L.
b) If and belong to L, then so do + +# and + + +#.
State Readability and Unique Readability for L

and determine for each

whether it holds.
(6) (Polish Notation) Let T
0
be the smallest set of sequences P such that the
following conditions hold.
a) For each n, A
n
P.
b) If
1
and
2
belong to P, then so do
1
= +
1
and
1

2
= +
1
+
2
.
State and prove the unique readability theorem for T
0
. Note, the Polish
system of notation does away with parentheses.
1.2 Truth assignments
We can now describe the semantics for propositional logic.
Denition 1.14 A truth assignment for L
0
is a function from the set of
propositional symbols A
n
: n N into the set T, F.
Now, () should have the opposite truth value from that of and the truth
value of (
1

2
) should reect whether, if
1
has truth value T, then
2
has
truth value T.
6 Propositional logic
Theorem 1.15 Suppose that is a truth assignment for L
0
. Then there is a
unique function dened on L
0
with the following properties.
(1) For all n, (A
n
) = (A
n
).
(2) For all L
0
,
(()) =

T, if () = F;
F, otherwise.
(3) For all
1
and
2
in L
0
,
((
1

2
)) =

F, if (
1
) = T and (
2
) = F;
T, otherwise.
Proof. We can dene by recursion on the natural numbers greater than or
equal to 1. During the s + 1st step of the recursion, we may assume that is
already dened on all elements of L
0
which have length less than or equal to s.
Base step. For each n N, dene (A
n
) = (A
n
).
Recursion step. Suppose that s 1, that is dened on all sequences from
L
0
of length less than or equal to s, and that is an element of L
0
of
length s + 1.
If = (), we dene () as in (2); if = (
1

2
), we dene ()
as in (3).
It remains to show that is well dened on L
0
, it satises (1), (2), and (3),
and that it is the unique such functionexistence and uniqueness.
We will prove the rst two claims (existence) by induction on the natural
numbers greater than or equal to 1.
Clearly, is well dened on the elements of L
0
of length 1 and satises (1).
Suppose that s 1 and, by induction, that is well dened on the set of
elements of L
0
of length less than or equal to s. Suppose that L
0
and has
length s + 1. By the Unique Readability Theorem 1.13, either there is a L
0
such that = (), or there are
1
and
2
such that = (
1

2
), the
two cases are mutually exclusive, and, in either case, the subformulas of which
appear in the way described are unique. Thus, the condition to dene () is
unambiguous, showing that is well dened at . Further, is dened at so
as to satisfy whichever of (2) or (3) is relevant.
By induction, is well dened on L
0
and satises (1), (2), and (3).
Now, we verify uniqueness. Suppose that

: L
0
T, F and satises (1),
(2), and (3). For the sake of a contradiction, suppose that

is not equal to .
Fix so that

() ,= () and so that there is no L

0
such that is strictly
shorter than and

() ,= ().
Since

satises (1), for every n,

(A
n
) = (A
n
). By denition,

(
A
n
) = (A
n
). Hence, for every n,

(A
n
) = (A
n
).
Consequently, the length of must be greater than 1. By the Unique Read-
ability Theorem 1.13, is either a negation () or an implication (
1

2
)
Truth assignments 7
and uniquely so. In the rst case, since

satises (2),

(()) has the oppo-

site value from

(). Since is shorter than ,

() = (). By denition,
(()) has the opposite value from (). It follows that

(()) and (())

are equal. Thus, cannot be a negation. An analogous argument shows that
cannot be an implication. This is a contradiction to the Readability Lemma 1.11.
Consequently,

is equal to , which completes the proof of the theorem. .

Theorem 1.16 Suppose that L
0
and that and are truth assignments
which agree on the propositional symbols which occur in . Then () = ().
Proof. Proceed just as in the uniqueness part of the proof of Theorem 1.13. Show
that there cannot be a shortest subformula of where and disagree. .
Denition 1.17 (1) A truth assignment satises a formula if and only if
() = T. Similarly, satises a set of formulas if and only if it satises
all of the elements of .
(2) is a tautology if and only if every truth assignment satises .
(3) L
0
or L
0
are satisable if and only if there is a truth assignment
which satises or , respectively.
(4) is a contradiction if and only if there is no truth assignment which satises
.
To give an example, consider the formula (((A
1
) A
2
)) and a truth
assignment such that (A
1
) = (A
2
) = F. By Theorem 1.16, the values of
on A
1
and A
2
determine the value of on (((A
1
) A
2
)). In Figure 1.1, we
show the values of on (((A
1
) A
2
)) and its subformulas.
A
1
A
2
(A
1
) ((A
1
) A
2
) (((A
1
) A
2
))
F F T F T
Fig. 1.1 Extending a truth assignment
We can expand the table to systematically examine all possible truth assign-
ments on (((A
1
) A
2
)), as in Figure 1.2.
A
1
A
2
(A
1
) ((A
1
) A
2
) (((A
1
) A
2
))
T T F T F
T F F T F
F T T T F
F F T F T
Fig. 1.2 The truth table for (((A1) A2))
Truth tables, such as the one in Figure 1.2, provide a systematic method to
examine all the possible truth assignments for a given formula. Given a formula
, we generate a truth table for as follows.
8 Propositional logic
(1) The top row of the table consists of a list
1
,
2
, . . . ,
n
= consisting of
the subformulas of , ordered from left to right as follows.
a) The subformulas of of the form A
m
appear in the list without rep-
etition before any of the other subformulas of .
b) For each i n all of the proper subformulas of
i
appear in the list

1
,
2
, . . . ,
i1
.
c) The last element of the list is .
(2) Letting k be the number of subformulas of of the form A
m
, we consider
all of the 2
k
possible truth assignments for their propositional symbols. We
use a row in the table for each such truth assignment , and we ll in the
cell below A
m
in that row with the value of at A
m
.
(3) Finally, we work our way across each row and ll in the values of at
i
as
determined by the values already lled in for its subformulas.
We give another example in Figure 1.3. This time we have chosen the tau-
tology expressing the principle that if A
1
implies A
2
, then the contrapositive
implication from (A
2
) to (A
1
) also holds.
A
1
A
2
(A
1
A
2
) (A
1
) (A
2
) ((A
2
) (A
1
))
((A
1
A
2
)
((A
2
) (A
1
)))
T T T F F T T
T F F F T F T
F T T T F T T
F F T T T T T
Fig. 1.3 The truth table for ((A1 A2) ((A2) (A1)))
Theorem 1.18 There are algorithms to determine whether a propositional for-
mula is a tautology, satisable, or a contradiction.
Proof. Starting with a formula , we can systematically generate its truth table.
Then is a tautology if and only if every entry in the last column of its truth
table is equal to T. It is satisable if and only if there is an entry in the last
column of its truth table which is equal to T. It is a contradiction if and only if
every entry in the last column of its truth table is equal to F.
Remark 1.19 Roughly speaking, if has n many symbols, then the analysis of
by the method of truth tables involves 2
n
many steps. A question which has
received a considerable amount of attention is whether there is a more ecient
method which when given determines whether is satisable. For more infor-
mation on this problem, known as the P = NP problem, and even a cash prize,
see the following web site.
http://www.claymath.org/millennium/P vs NP/
Truth assignments 9
1.2.1 Truth functions
Denition 1.20 An n-place truth function is a function whose domain is the
set of sequences of Ts and Fs of length n, written T, F
n
and whose range is
contained in T, F.
If is a formula in L
0
and the propositional symbols which occur in
are contained in the set A
0
, . . . , A
n1
, then we can dene the truth function
f

derived from . Given T, F

n
, we let be the truth assignment on
A
0
, . . . , A
n1
such that (A
i1
) is equal to the ith element of , and we dene
f

() to be ().
In the next theorem, we show that L
0
is as expressive as is possible. By this,
we mean that every truth function is represented by a formula in L
0
.
Theorem 1.21 Suppose that f : T, F
n
T, F is a truth function. Then
there is a formula such that f

= f.
Proof. We build up to the formula by a sequence of smaller steps.
For T, F
n
, dene
,i
so that

,i
=

A
i1
, if (i) = T;
(A
i1
), if (i) = F.
Given two formulas
1
and
2
, we dene the conjunction of
1
and
2
to
be the formula ((
1
(
2
))). As is seen in Figure 1.2.1, a truth assignment
satises the conjunction of
1
and
2
if and only if it satises both
1
and
2
.

1

2
(
2
) (
1
(
2
)) ((
1
(
2
)))
T T F F T
T F T T F
F T F T F
F F T T F
Fig. 1.4 The conjunction of 1 and 2.
Given more than two formulas
1
, . . . ,
n
, we use recursion and dene their
conjunction to be the conjunction of
1
with the conjunction of
2
, . . . ,
n
. For
example, the conjunction of
1
,
2
, and
3
is the formula
((
1
(((
2
(
3
)))))).
By induction, if is a truth assignment, then maps the conjunction of

1
, . . . ,
n
to T if and only if maps each of
1
, . . . ,
n
to T.
For T, F
n
, we let

be the conjunction of the formulas

,i
for i less
than or equal to n. The only truth assignments that satisfy

are those which

assign (i) to A
i1
.
Given two formulas
1
and
2
, we dene the disjunction of
1
and
2
to be
the formula ((
1
)
2
). As is seen in Figure 1.2.1, a truth assignment satises
10 Propositional logic

1

2
(
1
) ((
1
)
2
)
T T F T
T F F T
F T T T
F F T F
Fig. 1.5 The disjunction of 1 and 2.
the conjunction of
1
and
2
if and only if it satises at least one of
1
or
2
. As
above, when n is greater than two, we dene the disjunction of
1
, . . . ,
n
to be
the disjunction of
1
with the disjunction of
2
, . . . ,
n
. By another induction,
if is a truth assignment, then maps the disjunction of
1
, . . . ,
n
to T if and
only if it maps at least one of
1
, . . . ,
n
to T.
Now we let
f
be the disjunction of the set of formulas

for which f() = T.

By construction, if is a truth assignment that satises
f
, then there is a
such that f() = T and for all i less than or equal to n, (A
i1
) is equal to the
ith element of . Consequently, f is equal to f

f
, as required. .
Remark 1.22 It is not unusual to include symbols for conjunction, for
disjunction, and for if and only if. By Theorem 1.21, these and all other
logical connectives can be expressed in the language with only and .
Of course, the fewer symbols there are in the language, the fewer the number
of cases there are in proofs by induction, so we decided in favor a small number
of logical symbols. Occasionally, we pay a price for that decision: for example,
with the lengths of the formulas that appeared in the proof of Theorem 1.21.
Remark 1.23 In some applications, it important to the best possible represen-
tative of a truth function f. Best possible could mean having the shortest length
or having the fewest logical connectives of a certain type. When n is large, it is
computationally prohibitive to generate the truth tables for all of the possible
formulas with the desired functionality. Finding the optimal for a specied f
remains an interesting problem.
1.3 A proof system for L
0
Suppose that is a subset of L
0
so that is a set of propositional formulas. We
shall dene a formal notion of proof. Intuitively a proof from will be a nite
sequence,

1
, . . . ,
n

of propositional formulas which satises certain conditions. In order to make the

denition precise we need to rst dene the set of Logical Axioms.
Denition 1.24 Suppose that
1
,
2
and
3
are propositional formulas. Then
each of the following propositional formulas is a logical axiom:
(Group I axioms)
A proof system for L
0
11
(1) ((
1
(
2

3
)) ((
1

2
) (
1

3
)))
(2) (
1

1
)
(3) (
1
(
2

1
))
(Group II axioms)
(1) (
1
((
1
)
2
))
(Group III axioms)
(1) (((
1
)
1
)
1
)
(Group IV axioms)
(1) ((
1
) (
1

2
))
(2) (
1
((
2
) ((
1

2
)))) .
It is easily veried that each logical axiom is a tautology.
Denition 1.25 Suppose that L
0
.
(1) Suppose that
s =
1
, . . . ,
n

is a nite sequence of propositional formulas. The nite sequence s is a

-proof if for each i n at least one of
a)
i
; or
b)
i
is a logical axiom; or
c) there exist j
1
< i and j
2
< i such that

j2
= (
j1

i
).
(2) if and only if there exists a nite sequence
s =
1
, . . . ,
n

such that s is a -proof and such that

n
= . .
Notice that if s =
1
, . . . ,
n
is a -proof and if t =
1
, . . . ,
m
is a
-proof then so is s +t =
1
, . . . ,
n
,
1
, . . . ,
m
.
We shall prove a sequence of simple lemmas about the proof system. For each
lemma we shall note which logical axioms are actually used.
The rst lemma, which concerns inference, requires no logical axioms what-
soever.
Lemma 1.26 (Inference) Suppose that L
0
, is a formula and that is
a formula. Suppose that and that ( ).
Then .
Proof. Let
1
, . . . ,
n
be a -proof of , thus
n
= . Let
1
, . . . ,
m
be a
-proof of ( ). Then,
1
, . . . ,
n
,
1
, . . . ,
m
, is a -proof and . .
12 Propositional logic
The second lemma of our series is the Soundness Lemma, this also is inde-
pendent of the choice of logical axioms, provided that every logical axiom is a
tautology.
Lemma 1.27 (Soundness) Suppose that L
0
, is a formula and that
. Suppose that
: A
1
, . . . , A
n
, . . . T, F
is a truth assignment such that () = T for all .
Then () = T.
Proof. We leave the proof to the reader, but one argues by induction on the
length of the -proof.
The next lemma is the Deduction Lemma. This lemma requires the logical
axioms from Group I.
Lemma 1.28 (Deduction) Suppose that L
0
, is a formula, is a for-
mula and
.
Then ( ).
Proof. Let

1
, . . . ,
n

be a ( )-proof of . We prove by induction on i n that

(
i
).
First we consider the case i = 1. Either
1
or
1
is a logical axiom
(possibly both). So there are three subcases of this case.
Subcase 1.1:
1
. So we must show that (
1
). However
(
1
(
1
))
since (
1
(
1
)) is a logical axiom. Further

1
since
1
. Therefore by the Inference Lemma 1.26, (
1
).
Subcase 1.2:
1
= . Note that ( ) is a logical axiom and so
( ).
Subcase 1.3:
1
is a logical axiom. This is just like subcase 1.1; (
1
(
1
))
is a logical axiom and so
(
1
(
1
)).
Since
1
is a logical axiom,
1
. Therefore by the inference Lemma,
(
1
).
A proof system for L
0
13
We now suppose that k n and assume as an induction hypothesis that for
all i < k,
(
i
).
There are two subcases.
Subcase 2.1:
k
or
k
is a logical axiom. But then exactly as in the
case of
1
, (
k
).
Subcase 2.2: There exist j
1
< k and j
2
< k such that
j2
= (
j1

k
).
By the induction hypothesis; (
j1
) and (
j2
). Now we use
the logical axiom
(( (
j1

k
)) ((
j1
) (
k
))).
By the induction hypothesis,
( (
j1

k
)),
and so by the Inference Lemma,
((
j1
) (
k
)).
Again by the induction hypothesis,
(
j1
),
and so by the Inference Lemma one last time,
(
k
).
This completes the induction and so ( ). Finally we note that only
Group I logical axioms were used. .
Denition 1.29 Suppose that L
0
.
(1) is inconsistent if for some formula , and ().
(2) is consistent if is not inconsistent. .
If is an inconsistent set of formulas then for every formula . This
is the content of the next lemma the proof of which appeals to the Deduc-
tion Lemma and logical axioms in Group II. Therefore only logical axioms from
Groups I and II are needed.
Lemma 1.30 Suppose that L
0
and that is inconsistent. Suppose that
is a formula.
Then .
14 Propositional logic
Proof. Since is inconsistent there exists a formula such that

and ().
But
( (() ))
since ( (() )) is a logical axiom. Therefore by the Inference Lemma,
(() )
and by the Inference Lemma again,
.
This completes the proof. .
Denition 1.31 Suppose that L
0
and that is consistent. Then is
maximally consistent if and only if for each formula if is consistent
then . .
Lemma 1.32 Suppose that L
0
and that is consistent. Suppose that is
a formula.
Then, at least one of or () is consistent, possibly both.
Proof. Suppose that () is inconsistent. Therefore, for each formula ,
()
and in particular, () .
Thus by the Deduction Lemma, (() ). But ((() ) ) is
a logical axiom (Group III), and so by the Inference Lemma, .
Now assume toward a contradiction that is inconsistent. By
Lemma 1.30, for each formula ,
.
By the Deduction Lemma, for each formula ,
( ).
But and so by the Inference Lemma, for each formula , . Thus is
inconsistent, which is a contradiction. Therefore is consistent.
So we have proved, assuming the consistency of , that if () is
inconsistent then is consistent. .
A proof system for L
0
15
Corollary 1.33 Suppose that L
0
and that is maximally consistent. Sup-
pose that is a formula.
Then:
(1) Either or () ;
(2) If then .
Proof. We rst prove (1). By Lemma 1.32, either is consistent or
() is consistent. Therefore since is maximally consistent (1) must
hold.
We nish by proving (2). We are given that . By (1), if / then
() which implies that (). But and so this contradicts the
consistency of . .
We now use the logical axioms in Group IV.
Lemma 1.34 Suppose that L
0
and that is maximally consistent. Suppose
that
1
and
2
are formulas.
Then (
1

2
) if and only if at least one of
1
/ or
2
.
Proof. We rst suppose that
1
/ . We must show that (
1

2
) .
Since
1
/ , by Corollary 1.33, (
1
) .
Thus (
1
). But
((
1
) (
1

2
))
since ((
1
) (
1

2
)) is a logical axiom, and so by the Inference Lemma,
(
1

2
).
Therefore by Corollary 1.33, (
1

2
) .
Next we suppose that
2
. Now
(
2
(
1

2
))
is a logical axiom and so (
2
(
1

2
)). By the Inference Lemma, since

2
,
(
1

2
).
Therefore, again by Corollary 1.33, (
1

2
) .
To nish, we suppose that
1
and
2
/ . Now we must show that
(
1

2
) / .
Since
2
/ , by Corollary 1.33, (
2
) .
16 Propositional logic
Thus
1
and (
2
). But
(
1
((
2
) ((
1

2
)))),
since (
1
((
2
) ((
1

2
)))) is a logical axiom. Therefore by the
Inference Lemma,
((
2
) ((
1

2
))),
and by the Inference Lemma once again,
((
1

2
)).
Finally by Corollary 1.33, ((
1

2
)) and so (
1

2
) / as required.
This completes the proof of the lemma. .
Our goal is to show that if is consistent then is satisable. We rst
consider the special case that is maximally consistent. This case will turn out
to be an easy case for uniquely species the truth assignment which witnesses
that is satisable.
Lemma 1.35 Suppose that L
0
and that is maximally consistent.
Then is satisable.
Proof. Dene a truth assignment as follows. For each i N, let
(A
i
) =

T, if A
i
;
F, if A
i
/ .
We claim that for each formula , () = T if and () = F if / .
We organize our proof of this claim by induction on the length of .
The case that has length 1 is immediate.
Suppose that has length n > 1 and that as induction hypothesis, for all
formulas if has length less than n then () = T if and () = F if
/ .
There are two cases.
Case 1: = (). Since is maximally consistent, if and only if / .
But () = T if and only if () = F. By the induction hypothesis () = T if
and only if .
Thus if then () = T and () = F if / .
Case 2: = (
1

2
). Since is maximally consistent, if and only if
at least one of
1
/ or
2
. This is by Lemma 1.34.
By the denition of , () = T if and only if either (
1
) = F or (
2
) = T.
Therefore by the induction hypothesis, () = T if and only if either
1
/ of

2
.
Thus, () = T if and only if .
This completes the induction. .
A proof system for L
0
17
Theorem 1.36 Suppose that L
0
and that is consistent.
Then there exists a set

L
0
such that

and such that

is maximally consistent.
Proof. Let (
i
: i N) be an enumeration of all of the formulas of L
0
. For
example, we could enumerate the nitely many length 1 formulas which use
only the propositional symbol A
1
; then, we could enumerate the nitely many
formulas of length less than or equal to 2 which use no propositional symbols
other than A
1
and A
2
; and in subsequent steps, enumerate the nitely many n
formulas of length less than or equal to n which use no propositional symbols
other than A
1
, . . . , A
n
.
We construct a sequence of sets (
n
: m N) by recursion on n. To begin,
let
0
equal . Given
n
, let
n+1
be dened as follows.

n+1
=

n

n
, if
n

n
is consistent;

n

n
, otherwise.
We check by induction that each
n
is consistent. Clearly,
0
is consistent, since
we are given that is consistent. Assuming that
n
is consistent, we can apply
Lemma 1.32 to conclude that at least one of
n

n
or
n

n
is also
consistent. But then,
n+1
is also consistent.
Now, dene

so that

=
nN

n
.
For the sake of a contradiction, suppose that

is not consistent, then there

is are

-proofs of some formula and of its negation. These

-proofs are nite,

and so there is nite set of formulas

F
from

which appear in these proofs.

But then

F
must be a subset of some
n
, and so one of the
n
s must be
inconsistent. Since we have already checked that all of the
n
s are consistent,
this is impossible. Thus

is consistent.

is also maximally consistent: For every formula , there is an n such that

is equal to
n
. But we chose
n
so that either
n

n
or (
n
)
n
. Since
=
n
and
n

, either

or ()

, as required for maximality. .

Theorem 1.37 (Completeness; Version I) Suppose that L
0
and that
is consistent.
Then is satisable.
Proof. By Theorem 1.36 extend to a maximally consistent set, and then apply
Lemma 1.35. .
18 Propositional logic
1.4 Logical implication and compactness
Denition 1.38 Let be a subset of L
0
and let be an element of L
0
. Then,
logically implies if and only if, for every truth assignment , if satises
in the sense of Denition 1.17, then satises .
For example, logically implies , and A
1
, (A
1
A
2
) logically implies
A
2
.
If is a set a formulas, we write to indicate that logically implies .
Denition 1.39 A subset of L
0
is nitely satisable if and only if for every
nite subset
0
of , there is a truth assignment such that for all
0
,
() = T.
Theorem 1.40 (Compactness for L
0
) Suppose that L
0
, L
0
, and
logically implies . Then there is a nite set
0
such that
0
and
0
logically
implies .
Proof. Since , () is not satisable. Therefore by the Completeness
Theorem, () is inconsistent. But this implies that there exists a nite
set
0
such that
0
() is inconsistent. Therefore by Lemma 1.27, the
Soundness Lemma,
0
() is not satisable and so
0
. .
We end this chapter by discussing a reformulation of the Completeness The-
orem.
Theorem 1.41 (Completeness; Version II) Suppose that L
0
, is a
formula and that .
Then .
Proof. Since , () is not satisable. Therefore by the Completeness
Theorem, () is inconsistent and so by Lemma 1.30,
() .
By Lemma 1.28, the Deduction Lemma,
(() ).
But ((() ) ) is a logical axiom and so by Lemma 1.26, the Inference
Lemma, . .
1.4.1 Exercises
(1) Which of the formulas
(((A
1
A
1
) A
2
) A
2
)
((((A
1
A
2
) A
2
) A
2
) A
2
)
is a tautology?
Logical implication and compactness 19
(2) For L
0
and and in L
0
, show that logically implies if and
only if logically implies ( ).
(3) Two physicists, A and B, and a logician C, are wearing hats, which they
know are either black or white but not all white. A can see the hats of B
and C; B can see the hats of A and C; C is blind. Each is asked in turn if
they know the color of their own hat. The answers are: A:No. B: No.
C: Yes. What color is Cs hat and how does C know?
(4) For
1
and
2
subsets of L
0
,
1
is logically equivalent to
2
if and only if,
for all L
0
,
1
logically implies if and only if
2
logically implies .
For L
0
, is independent if it is not logically equivalent to any of its
proper subsets. Prove the following.
a) If is nite, then there is a
0
such that
0
, and
0
are logically
equivalent, and
0
is independent.
b) There is an innite set such that has no independent and logically
equivalent subset.
c) For every L
0
, there is a L
0
such that is independent and
logically equivalent to .
(5) Show that the set of logical consequences of
A
i
: i ,= 1 and i N
is consistent but not maximally consistent. Show that the set of logical
consequences of
A
i
: i N
is maximally consistent.
20 Propositional logic